1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/pagewalk.h>
3 #include <linux/mm_inline.h>
4 #include <linux/hugetlb.h>
5 #include <linux/huge_mm.h>
6 #include <linux/mount.h>
7 #include <linux/seq_file.h>
8 #include <linux/highmem.h>
9 #include <linux/ptrace.h>
10 #include <linux/slab.h>
11 #include <linux/pagemap.h>
12 #include <linux/mempolicy.h>
13 #include <linux/rmap.h>
14 #include <linux/swap.h>
15 #include <linux/sched/mm.h>
16 #include <linux/swapops.h>
17 #include <linux/mmu_notifier.h>
18 #include <linux/page_idle.h>
19 #include <linux/shmem_fs.h>
20 #include <linux/uaccess.h>
21 #include <linux/pkeys.h>
25 #include <asm/tlbflush.h>
28 #define SEQ_PUT_DEC(str, val) \
29 seq_put_decimal_ull_width(m, str, (val) << (PAGE_SHIFT-10), 8)
30 void task_mem(struct seq_file *m, struct mm_struct *mm)
32 unsigned long text, lib, swap, anon, file, shmem;
33 unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
35 anon = get_mm_counter(mm, MM_ANONPAGES);
36 file = get_mm_counter(mm, MM_FILEPAGES);
37 shmem = get_mm_counter(mm, MM_SHMEMPAGES);
40 * Note: to minimize their overhead, mm maintains hiwater_vm and
41 * hiwater_rss only when about to *lower* total_vm or rss. Any
42 * collector of these hiwater stats must therefore get total_vm
43 * and rss too, which will usually be the higher. Barriers? not
44 * worth the effort, such snapshots can always be inconsistent.
46 hiwater_vm = total_vm = mm->total_vm;
47 if (hiwater_vm < mm->hiwater_vm)
48 hiwater_vm = mm->hiwater_vm;
49 hiwater_rss = total_rss = anon + file + shmem;
50 if (hiwater_rss < mm->hiwater_rss)
51 hiwater_rss = mm->hiwater_rss;
53 /* split executable areas between text and lib */
54 text = PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK);
55 text = min(text, mm->exec_vm << PAGE_SHIFT);
56 lib = (mm->exec_vm << PAGE_SHIFT) - text;
58 swap = get_mm_counter(mm, MM_SWAPENTS);
59 SEQ_PUT_DEC("VmPeak:\t", hiwater_vm);
60 SEQ_PUT_DEC(" kB\nVmSize:\t", total_vm);
61 SEQ_PUT_DEC(" kB\nVmLck:\t", mm->locked_vm);
62 SEQ_PUT_DEC(" kB\nVmPin:\t", atomic64_read(&mm->pinned_vm));
63 SEQ_PUT_DEC(" kB\nVmHWM:\t", hiwater_rss);
64 SEQ_PUT_DEC(" kB\nVmRSS:\t", total_rss);
65 SEQ_PUT_DEC(" kB\nRssAnon:\t", anon);
66 SEQ_PUT_DEC(" kB\nRssFile:\t", file);
67 SEQ_PUT_DEC(" kB\nRssShmem:\t", shmem);
68 SEQ_PUT_DEC(" kB\nVmData:\t", mm->data_vm);
69 SEQ_PUT_DEC(" kB\nVmStk:\t", mm->stack_vm);
70 seq_put_decimal_ull_width(m,
71 " kB\nVmExe:\t", text >> 10, 8);
72 seq_put_decimal_ull_width(m,
73 " kB\nVmLib:\t", lib >> 10, 8);
74 seq_put_decimal_ull_width(m,
75 " kB\nVmPTE:\t", mm_pgtables_bytes(mm) >> 10, 8);
76 SEQ_PUT_DEC(" kB\nVmSwap:\t", swap);
78 hugetlb_report_usage(m, mm);
82 unsigned long task_vsize(struct mm_struct *mm)
84 return PAGE_SIZE * mm->total_vm;
87 unsigned long task_statm(struct mm_struct *mm,
88 unsigned long *shared, unsigned long *text,
89 unsigned long *data, unsigned long *resident)
91 *shared = get_mm_counter(mm, MM_FILEPAGES) +
92 get_mm_counter(mm, MM_SHMEMPAGES);
93 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
95 *data = mm->data_vm + mm->stack_vm;
96 *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
102 * Save get_task_policy() for show_numa_map().
104 static void hold_task_mempolicy(struct proc_maps_private *priv)
106 struct task_struct *task = priv->task;
109 priv->task_mempolicy = get_task_policy(task);
110 mpol_get(priv->task_mempolicy);
113 static void release_task_mempolicy(struct proc_maps_private *priv)
115 mpol_put(priv->task_mempolicy);
118 static void hold_task_mempolicy(struct proc_maps_private *priv)
121 static void release_task_mempolicy(struct proc_maps_private *priv)
126 static struct vm_area_struct *proc_get_vma(struct proc_maps_private *priv,
129 struct vm_area_struct *vma = vma_next(&priv->iter);
132 *ppos = vma->vm_start;
135 vma = get_gate_vma(priv->mm);
141 static void *m_start(struct seq_file *m, loff_t *ppos)
143 struct proc_maps_private *priv = m->private;
144 unsigned long last_addr = *ppos;
145 struct mm_struct *mm;
147 /* See m_next(). Zero at the start or after lseek. */
148 if (last_addr == -1UL)
151 priv->task = get_proc_task(priv->inode);
153 return ERR_PTR(-ESRCH);
156 if (!mm || !mmget_not_zero(mm)) {
157 put_task_struct(priv->task);
162 if (mmap_read_lock_killable(mm)) {
164 put_task_struct(priv->task);
166 return ERR_PTR(-EINTR);
169 vma_iter_init(&priv->iter, mm, last_addr);
170 hold_task_mempolicy(priv);
171 if (last_addr == -2UL)
172 return get_gate_vma(mm);
174 return proc_get_vma(priv, ppos);
177 static void *m_next(struct seq_file *m, void *v, loff_t *ppos)
183 return proc_get_vma(m->private, ppos);
186 static void m_stop(struct seq_file *m, void *v)
188 struct proc_maps_private *priv = m->private;
189 struct mm_struct *mm = priv->mm;
194 release_task_mempolicy(priv);
195 mmap_read_unlock(mm);
197 put_task_struct(priv->task);
201 static int proc_maps_open(struct inode *inode, struct file *file,
202 const struct seq_operations *ops, int psize)
204 struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
210 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
211 if (IS_ERR(priv->mm)) {
212 int err = PTR_ERR(priv->mm);
214 seq_release_private(inode, file);
221 static int proc_map_release(struct inode *inode, struct file *file)
223 struct seq_file *seq = file->private_data;
224 struct proc_maps_private *priv = seq->private;
229 return seq_release_private(inode, file);
232 static int do_maps_open(struct inode *inode, struct file *file,
233 const struct seq_operations *ops)
235 return proc_maps_open(inode, file, ops,
236 sizeof(struct proc_maps_private));
240 * Indicate if the VMA is a stack for the given task; for
241 * /proc/PID/maps that is the stack of the main task.
243 static int is_stack(struct vm_area_struct *vma)
246 * We make no effort to guess what a given thread considers to be
247 * its "stack". It's not even well-defined for programs written
250 return vma->vm_start <= vma->vm_mm->start_stack &&
251 vma->vm_end >= vma->vm_mm->start_stack;
254 static void show_vma_header_prefix(struct seq_file *m,
255 unsigned long start, unsigned long end,
256 vm_flags_t flags, unsigned long long pgoff,
257 dev_t dev, unsigned long ino)
259 seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
260 seq_put_hex_ll(m, NULL, start, 8);
261 seq_put_hex_ll(m, "-", end, 8);
263 seq_putc(m, flags & VM_READ ? 'r' : '-');
264 seq_putc(m, flags & VM_WRITE ? 'w' : '-');
265 seq_putc(m, flags & VM_EXEC ? 'x' : '-');
266 seq_putc(m, flags & VM_MAYSHARE ? 's' : 'p');
267 seq_put_hex_ll(m, " ", pgoff, 8);
268 seq_put_hex_ll(m, " ", MAJOR(dev), 2);
269 seq_put_hex_ll(m, ":", MINOR(dev), 2);
270 seq_put_decimal_ull(m, " ", ino);
275 show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
277 struct mm_struct *mm = vma->vm_mm;
278 struct file *file = vma->vm_file;
279 vm_flags_t flags = vma->vm_flags;
280 unsigned long ino = 0;
281 unsigned long long pgoff = 0;
282 unsigned long start, end;
284 const char *name = NULL;
287 struct inode *inode = file_inode(vma->vm_file);
288 dev = inode->i_sb->s_dev;
290 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
293 start = vma->vm_start;
295 show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino);
298 * Print the dentry name for named mappings, and a
299 * special [heap] marker for the heap:
303 seq_file_path(m, file, "\n");
307 if (vma->vm_ops && vma->vm_ops->name) {
308 name = vma->vm_ops->name(vma);
313 name = arch_vma_name(vma);
315 struct anon_vma_name *anon_name;
322 if (vma->vm_start <= mm->brk &&
323 vma->vm_end >= mm->start_brk) {
333 anon_name = anon_vma_name(vma);
336 seq_printf(m, "[anon:%s]", anon_name->name);
348 static int show_map(struct seq_file *m, void *v)
354 static const struct seq_operations proc_pid_maps_op = {
361 static int pid_maps_open(struct inode *inode, struct file *file)
363 return do_maps_open(inode, file, &proc_pid_maps_op);
366 const struct file_operations proc_pid_maps_operations = {
367 .open = pid_maps_open,
370 .release = proc_map_release,
374 * Proportional Set Size(PSS): my share of RSS.
376 * PSS of a process is the count of pages it has in memory, where each
377 * page is divided by the number of processes sharing it. So if a
378 * process has 1000 pages all to itself, and 1000 shared with one other
379 * process, its PSS will be 1500.
381 * To keep (accumulated) division errors low, we adopt a 64bit
382 * fixed-point pss counter to minimize division errors. So (pss >>
383 * PSS_SHIFT) would be the real byte count.
385 * A shift of 12 before division means (assuming 4K page size):
386 * - 1M 3-user-pages add up to 8KB errors;
387 * - supports mapcount up to 2^24, or 16M;
388 * - supports PSS up to 2^52 bytes, or 4PB.
392 #ifdef CONFIG_PROC_PAGE_MONITOR
393 struct mem_size_stats {
394 unsigned long resident;
395 unsigned long shared_clean;
396 unsigned long shared_dirty;
397 unsigned long private_clean;
398 unsigned long private_dirty;
399 unsigned long referenced;
400 unsigned long anonymous;
401 unsigned long lazyfree;
402 unsigned long anonymous_thp;
403 unsigned long shmem_thp;
404 unsigned long file_thp;
406 unsigned long shared_hugetlb;
407 unsigned long private_hugetlb;
417 static void smaps_page_accumulate(struct mem_size_stats *mss,
418 struct page *page, unsigned long size, unsigned long pss,
419 bool dirty, bool locked, bool private)
424 mss->pss_anon += pss;
425 else if (PageSwapBacked(page))
426 mss->pss_shmem += pss;
428 mss->pss_file += pss;
431 mss->pss_locked += pss;
433 if (dirty || PageDirty(page)) {
434 mss->pss_dirty += pss;
436 mss->private_dirty += size;
438 mss->shared_dirty += size;
441 mss->private_clean += size;
443 mss->shared_clean += size;
447 static void smaps_account(struct mem_size_stats *mss, struct page *page,
448 bool compound, bool young, bool dirty, bool locked,
451 int i, nr = compound ? compound_nr(page) : 1;
452 unsigned long size = nr * PAGE_SIZE;
455 * First accumulate quantities that depend only on |size| and the type
456 * of the compound page.
458 if (PageAnon(page)) {
459 mss->anonymous += size;
460 if (!PageSwapBacked(page) && !dirty && !PageDirty(page))
461 mss->lazyfree += 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;
470 * Then accumulate quantities that may depend on sharing, or that may
471 * differ page-by-page.
473 * page_count(page) == 1 guarantees the page is mapped exactly once.
474 * If any subpage of the compound page mapped with PTE it would elevate
477 * The page_mapcount() is called to get a snapshot of the mapcount.
478 * Without holding the page lock this snapshot can be slightly wrong as
479 * we cannot always read the mapcount atomically. It is not safe to
480 * call page_mapcount() even with PTL held if the page is not mapped,
481 * especially for migration entries. Treat regular migration entries
484 if ((page_count(page) == 1) || migration) {
485 smaps_page_accumulate(mss, page, size, size << PSS_SHIFT, dirty,
489 for (i = 0; i < nr; i++, page++) {
490 int mapcount = page_mapcount(page);
491 unsigned long pss = PAGE_SIZE << PSS_SHIFT;
494 smaps_page_accumulate(mss, page, PAGE_SIZE, pss, dirty, locked,
500 static int smaps_pte_hole(unsigned long addr, unsigned long end,
501 __always_unused int depth, struct mm_walk *walk)
503 struct mem_size_stats *mss = walk->private;
504 struct vm_area_struct *vma = walk->vma;
506 mss->swap += shmem_partial_swap_usage(walk->vma->vm_file->f_mapping,
507 linear_page_index(vma, addr),
508 linear_page_index(vma, end));
513 #define smaps_pte_hole NULL
514 #endif /* CONFIG_SHMEM */
516 static void smaps_pte_hole_lookup(unsigned long addr, struct mm_walk *walk)
519 if (walk->ops->pte_hole) {
520 /* depth is not used */
521 smaps_pte_hole(addr, addr + PAGE_SIZE, 0, walk);
526 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
527 struct mm_walk *walk)
529 struct mem_size_stats *mss = walk->private;
530 struct vm_area_struct *vma = walk->vma;
531 bool locked = !!(vma->vm_flags & VM_LOCKED);
532 struct page *page = NULL;
533 bool migration = false, young = false, dirty = false;
535 if (pte_present(*pte)) {
536 page = vm_normal_page(vma, addr, *pte);
537 young = pte_young(*pte);
538 dirty = pte_dirty(*pte);
539 } else if (is_swap_pte(*pte)) {
540 swp_entry_t swpent = pte_to_swp_entry(*pte);
542 if (!non_swap_entry(swpent)) {
545 mss->swap += PAGE_SIZE;
546 mapcount = swp_swapcount(swpent);
548 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
550 do_div(pss_delta, mapcount);
551 mss->swap_pss += pss_delta;
553 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
555 } else if (is_pfn_swap_entry(swpent)) {
556 if (is_migration_entry(swpent))
558 page = pfn_swap_entry_to_page(swpent);
561 smaps_pte_hole_lookup(addr, walk);
568 smaps_account(mss, page, false, young, dirty, locked, migration);
571 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
572 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
573 struct mm_walk *walk)
575 struct mem_size_stats *mss = walk->private;
576 struct vm_area_struct *vma = walk->vma;
577 bool locked = !!(vma->vm_flags & VM_LOCKED);
578 struct page *page = NULL;
579 bool migration = false;
581 if (pmd_present(*pmd)) {
582 /* FOLL_DUMP will return -EFAULT on huge zero page */
583 page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
584 } else if (unlikely(thp_migration_supported() && is_swap_pmd(*pmd))) {
585 swp_entry_t entry = pmd_to_swp_entry(*pmd);
587 if (is_migration_entry(entry)) {
589 page = pfn_swap_entry_to_page(entry);
592 if (IS_ERR_OR_NULL(page))
595 mss->anonymous_thp += HPAGE_PMD_SIZE;
596 else if (PageSwapBacked(page))
597 mss->shmem_thp += HPAGE_PMD_SIZE;
598 else if (is_zone_device_page(page))
601 mss->file_thp += HPAGE_PMD_SIZE;
603 smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd),
607 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
608 struct mm_walk *walk)
613 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
614 struct mm_walk *walk)
616 struct vm_area_struct *vma = walk->vma;
620 ptl = pmd_trans_huge_lock(pmd, vma);
622 smaps_pmd_entry(pmd, addr, walk);
627 if (pmd_trans_unstable(pmd))
630 * The mmap_lock held all the way back in m_start() is what
631 * keeps khugepaged out of here and from collapsing things
634 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
635 for (; addr != end; pte++, addr += PAGE_SIZE)
636 smaps_pte_entry(pte, addr, walk);
637 pte_unmap_unlock(pte - 1, ptl);
643 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
646 * Don't forget to update Documentation/ on changes.
648 static const char mnemonics[BITS_PER_LONG][2] = {
650 * In case if we meet a flag we don't know about.
652 [0 ... (BITS_PER_LONG-1)] = "??",
654 [ilog2(VM_READ)] = "rd",
655 [ilog2(VM_WRITE)] = "wr",
656 [ilog2(VM_EXEC)] = "ex",
657 [ilog2(VM_SHARED)] = "sh",
658 [ilog2(VM_MAYREAD)] = "mr",
659 [ilog2(VM_MAYWRITE)] = "mw",
660 [ilog2(VM_MAYEXEC)] = "me",
661 [ilog2(VM_MAYSHARE)] = "ms",
662 [ilog2(VM_GROWSDOWN)] = "gd",
663 [ilog2(VM_PFNMAP)] = "pf",
664 [ilog2(VM_LOCKED)] = "lo",
665 [ilog2(VM_IO)] = "io",
666 [ilog2(VM_SEQ_READ)] = "sr",
667 [ilog2(VM_RAND_READ)] = "rr",
668 [ilog2(VM_DONTCOPY)] = "dc",
669 [ilog2(VM_DONTEXPAND)] = "de",
670 [ilog2(VM_ACCOUNT)] = "ac",
671 [ilog2(VM_NORESERVE)] = "nr",
672 [ilog2(VM_HUGETLB)] = "ht",
673 [ilog2(VM_SYNC)] = "sf",
674 [ilog2(VM_ARCH_1)] = "ar",
675 [ilog2(VM_WIPEONFORK)] = "wf",
676 [ilog2(VM_DONTDUMP)] = "dd",
677 #ifdef CONFIG_ARM64_BTI
678 [ilog2(VM_ARM64_BTI)] = "bt",
680 #ifdef CONFIG_MEM_SOFT_DIRTY
681 [ilog2(VM_SOFTDIRTY)] = "sd",
683 [ilog2(VM_MIXEDMAP)] = "mm",
684 [ilog2(VM_HUGEPAGE)] = "hg",
685 [ilog2(VM_NOHUGEPAGE)] = "nh",
686 [ilog2(VM_MERGEABLE)] = "mg",
687 [ilog2(VM_UFFD_MISSING)]= "um",
688 [ilog2(VM_UFFD_WP)] = "uw",
689 #ifdef CONFIG_ARM64_MTE
690 [ilog2(VM_MTE)] = "mt",
691 [ilog2(VM_MTE_ALLOWED)] = "",
693 #ifdef CONFIG_ARCH_HAS_PKEYS
694 /* These come out via ProtectionKey: */
695 [ilog2(VM_PKEY_BIT0)] = "",
696 [ilog2(VM_PKEY_BIT1)] = "",
697 [ilog2(VM_PKEY_BIT2)] = "",
698 [ilog2(VM_PKEY_BIT3)] = "",
700 [ilog2(VM_PKEY_BIT4)] = "",
702 #endif /* CONFIG_ARCH_HAS_PKEYS */
703 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_MINOR
704 [ilog2(VM_UFFD_MINOR)] = "ui",
705 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_MINOR */
709 seq_puts(m, "VmFlags: ");
710 for (i = 0; i < BITS_PER_LONG; i++) {
711 if (!mnemonics[i][0])
713 if (vma->vm_flags & (1UL << i)) {
714 seq_putc(m, mnemonics[i][0]);
715 seq_putc(m, mnemonics[i][1]);
722 #ifdef CONFIG_HUGETLB_PAGE
723 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
724 unsigned long addr, unsigned long end,
725 struct mm_walk *walk)
727 struct mem_size_stats *mss = walk->private;
728 struct vm_area_struct *vma = walk->vma;
729 struct page *page = NULL;
731 if (pte_present(*pte)) {
732 page = vm_normal_page(vma, addr, *pte);
733 } else if (is_swap_pte(*pte)) {
734 swp_entry_t swpent = pte_to_swp_entry(*pte);
736 if (is_pfn_swap_entry(swpent))
737 page = pfn_swap_entry_to_page(swpent);
740 int mapcount = page_mapcount(page);
743 mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
745 mss->private_hugetlb += huge_page_size(hstate_vma(vma));
750 #define smaps_hugetlb_range NULL
751 #endif /* HUGETLB_PAGE */
753 static const struct mm_walk_ops smaps_walk_ops = {
754 .pmd_entry = smaps_pte_range,
755 .hugetlb_entry = smaps_hugetlb_range,
758 static const struct mm_walk_ops smaps_shmem_walk_ops = {
759 .pmd_entry = smaps_pte_range,
760 .hugetlb_entry = smaps_hugetlb_range,
761 .pte_hole = smaps_pte_hole,
765 * Gather mem stats from @vma with the indicated beginning
766 * address @start, and keep them in @mss.
768 * Use vm_start of @vma as the beginning address if @start is 0.
770 static void smap_gather_stats(struct vm_area_struct *vma,
771 struct mem_size_stats *mss, unsigned long start)
773 const struct mm_walk_ops *ops = &smaps_walk_ops;
776 if (start >= vma->vm_end)
780 if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
782 * For shared or readonly shmem mappings we know that all
783 * swapped out pages belong to the shmem object, and we can
784 * obtain the swap value much more efficiently. For private
785 * writable mappings, we might have COW pages that are
786 * not affected by the parent swapped out pages of the shmem
787 * object, so we have to distinguish them during the page walk.
788 * Unless we know that the shmem object (or the part mapped by
789 * our VMA) has no swapped out pages at all.
791 unsigned long shmem_swapped = shmem_swap_usage(vma);
793 if (!start && (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
794 !(vma->vm_flags & VM_WRITE))) {
795 mss->swap += shmem_swapped;
797 ops = &smaps_shmem_walk_ops;
801 /* mmap_lock is held in m_start */
803 walk_page_vma(vma, ops, mss);
805 walk_page_range(vma->vm_mm, start, vma->vm_end, ops, mss);
808 #define SEQ_PUT_DEC(str, val) \
809 seq_put_decimal_ull_width(m, str, (val) >> 10, 8)
811 /* Show the contents common for smaps and smaps_rollup */
812 static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss,
815 SEQ_PUT_DEC("Rss: ", mss->resident);
816 SEQ_PUT_DEC(" kB\nPss: ", mss->pss >> PSS_SHIFT);
817 SEQ_PUT_DEC(" kB\nPss_Dirty: ", mss->pss_dirty >> PSS_SHIFT);
820 * These are meaningful only for smaps_rollup, otherwise two of
821 * them are zero, and the other one is the same as Pss.
823 SEQ_PUT_DEC(" kB\nPss_Anon: ",
824 mss->pss_anon >> PSS_SHIFT);
825 SEQ_PUT_DEC(" kB\nPss_File: ",
826 mss->pss_file >> PSS_SHIFT);
827 SEQ_PUT_DEC(" kB\nPss_Shmem: ",
828 mss->pss_shmem >> PSS_SHIFT);
830 SEQ_PUT_DEC(" kB\nShared_Clean: ", mss->shared_clean);
831 SEQ_PUT_DEC(" kB\nShared_Dirty: ", mss->shared_dirty);
832 SEQ_PUT_DEC(" kB\nPrivate_Clean: ", mss->private_clean);
833 SEQ_PUT_DEC(" kB\nPrivate_Dirty: ", mss->private_dirty);
834 SEQ_PUT_DEC(" kB\nReferenced: ", mss->referenced);
835 SEQ_PUT_DEC(" kB\nAnonymous: ", mss->anonymous);
836 SEQ_PUT_DEC(" kB\nLazyFree: ", mss->lazyfree);
837 SEQ_PUT_DEC(" kB\nAnonHugePages: ", mss->anonymous_thp);
838 SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp);
839 SEQ_PUT_DEC(" kB\nFilePmdMapped: ", mss->file_thp);
840 SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb);
841 seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ",
842 mss->private_hugetlb >> 10, 7);
843 SEQ_PUT_DEC(" kB\nSwap: ", mss->swap);
844 SEQ_PUT_DEC(" kB\nSwapPss: ",
845 mss->swap_pss >> PSS_SHIFT);
846 SEQ_PUT_DEC(" kB\nLocked: ",
847 mss->pss_locked >> PSS_SHIFT);
848 seq_puts(m, " kB\n");
851 static int show_smap(struct seq_file *m, void *v)
853 struct vm_area_struct *vma = v;
854 struct mem_size_stats mss;
856 memset(&mss, 0, sizeof(mss));
858 smap_gather_stats(vma, &mss, 0);
860 show_map_vma(m, vma);
862 SEQ_PUT_DEC("Size: ", vma->vm_end - vma->vm_start);
863 SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma));
864 SEQ_PUT_DEC(" kB\nMMUPageSize: ", vma_mmu_pagesize(vma));
865 seq_puts(m, " kB\n");
867 __show_smap(m, &mss, false);
869 seq_printf(m, "THPeligible: %d\n",
870 hugepage_vma_check(vma, vma->vm_flags, true, false, true));
872 if (arch_pkeys_enabled())
873 seq_printf(m, "ProtectionKey: %8u\n", vma_pkey(vma));
874 show_smap_vma_flags(m, vma);
879 static int show_smaps_rollup(struct seq_file *m, void *v)
881 struct proc_maps_private *priv = m->private;
882 struct mem_size_stats mss;
883 struct mm_struct *mm = priv->mm;
884 struct vm_area_struct *vma;
885 unsigned long vma_start = 0, last_vma_end = 0;
887 MA_STATE(mas, &mm->mm_mt, 0, 0);
889 priv->task = get_proc_task(priv->inode);
893 if (!mm || !mmget_not_zero(mm)) {
898 memset(&mss, 0, sizeof(mss));
900 ret = mmap_read_lock_killable(mm);
904 hold_task_mempolicy(priv);
905 vma = mas_find(&mas, ULONG_MAX);
910 vma_start = vma->vm_start;
912 smap_gather_stats(vma, &mss, 0);
913 last_vma_end = vma->vm_end;
916 * Release mmap_lock temporarily if someone wants to
917 * access it for write request.
919 if (mmap_lock_is_contended(mm)) {
921 mmap_read_unlock(mm);
922 ret = mmap_read_lock_killable(mm);
924 release_task_mempolicy(priv);
929 * After dropping the lock, there are four cases to
930 * consider. See the following example for explanation.
932 * +------+------+-----------+
933 * | VMA1 | VMA2 | VMA3 |
934 * +------+------+-----------+
938 * Suppose we drop the lock after reading VMA2 due to
939 * contention, then we get:
943 * 1) VMA2 is freed, but VMA3 exists:
945 * find_vma(mm, 16k - 1) will return VMA3.
946 * In this case, just continue from VMA3.
948 * 2) VMA2 still exists:
950 * find_vma(mm, 16k - 1) will return VMA2.
951 * Iterate the loop like the original one.
953 * 3) No more VMAs can be found:
955 * find_vma(mm, 16k - 1) will return NULL.
956 * No more things to do, just break.
958 * 4) (last_vma_end - 1) is the middle of a vma (VMA'):
960 * find_vma(mm, 16k - 1) will return VMA' whose range
961 * contains last_vma_end.
962 * Iterate VMA' from last_vma_end.
964 vma = mas_find(&mas, ULONG_MAX);
970 if (vma->vm_start >= last_vma_end)
974 if (vma->vm_end > last_vma_end)
975 smap_gather_stats(vma, &mss, last_vma_end);
978 } while ((vma = mas_find(&mas, ULONG_MAX)) != NULL);
981 show_vma_header_prefix(m, vma_start, last_vma_end, 0, 0, 0, 0);
983 seq_puts(m, "[rollup]\n");
985 __show_smap(m, &mss, true);
987 release_task_mempolicy(priv);
988 mmap_read_unlock(mm);
993 put_task_struct(priv->task);
1000 static const struct seq_operations proc_pid_smaps_op = {
1007 static int pid_smaps_open(struct inode *inode, struct file *file)
1009 return do_maps_open(inode, file, &proc_pid_smaps_op);
1012 static int smaps_rollup_open(struct inode *inode, struct file *file)
1015 struct proc_maps_private *priv;
1017 priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT);
1021 ret = single_open(file, show_smaps_rollup, priv);
1025 priv->inode = inode;
1026 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
1027 if (IS_ERR(priv->mm)) {
1028 ret = PTR_ERR(priv->mm);
1030 single_release(inode, file);
1041 static int smaps_rollup_release(struct inode *inode, struct file *file)
1043 struct seq_file *seq = file->private_data;
1044 struct proc_maps_private *priv = seq->private;
1050 return single_release(inode, file);
1053 const struct file_operations proc_pid_smaps_operations = {
1054 .open = pid_smaps_open,
1056 .llseek = seq_lseek,
1057 .release = proc_map_release,
1060 const struct file_operations proc_pid_smaps_rollup_operations = {
1061 .open = smaps_rollup_open,
1063 .llseek = seq_lseek,
1064 .release = smaps_rollup_release,
1067 enum clear_refs_types {
1071 CLEAR_REFS_SOFT_DIRTY,
1072 CLEAR_REFS_MM_HIWATER_RSS,
1076 struct clear_refs_private {
1077 enum clear_refs_types type;
1080 #ifdef CONFIG_MEM_SOFT_DIRTY
1082 static inline bool pte_is_pinned(struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1086 if (!pte_write(pte))
1088 if (!is_cow_mapping(vma->vm_flags))
1090 if (likely(!test_bit(MMF_HAS_PINNED, &vma->vm_mm->flags)))
1092 page = vm_normal_page(vma, addr, pte);
1095 return page_maybe_dma_pinned(page);
1098 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1099 unsigned long addr, pte_t *pte)
1102 * The soft-dirty tracker uses #PF-s to catch writes
1103 * to pages, so write-protect the pte as well. See the
1104 * Documentation/admin-guide/mm/soft-dirty.rst for full description
1105 * of how soft-dirty works.
1109 if (pte_present(ptent)) {
1112 if (pte_is_pinned(vma, addr, ptent))
1114 old_pte = ptep_modify_prot_start(vma, addr, pte);
1115 ptent = pte_wrprotect(old_pte);
1116 ptent = pte_clear_soft_dirty(ptent);
1117 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent);
1118 } else if (is_swap_pte(ptent)) {
1119 ptent = pte_swp_clear_soft_dirty(ptent);
1120 set_pte_at(vma->vm_mm, addr, pte, ptent);
1124 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1125 unsigned long addr, pte_t *pte)
1130 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
1131 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1132 unsigned long addr, pmd_t *pmdp)
1134 pmd_t old, pmd = *pmdp;
1136 if (pmd_present(pmd)) {
1137 /* See comment in change_huge_pmd() */
1138 old = pmdp_invalidate(vma, addr, pmdp);
1140 pmd = pmd_mkdirty(pmd);
1142 pmd = pmd_mkyoung(pmd);
1144 pmd = pmd_wrprotect(pmd);
1145 pmd = pmd_clear_soft_dirty(pmd);
1147 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1148 } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1149 pmd = pmd_swp_clear_soft_dirty(pmd);
1150 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1154 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1155 unsigned long addr, pmd_t *pmdp)
1160 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
1161 unsigned long end, struct mm_walk *walk)
1163 struct clear_refs_private *cp = walk->private;
1164 struct vm_area_struct *vma = walk->vma;
1169 ptl = pmd_trans_huge_lock(pmd, vma);
1171 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1172 clear_soft_dirty_pmd(vma, addr, pmd);
1176 if (!pmd_present(*pmd))
1179 page = pmd_page(*pmd);
1181 /* Clear accessed and referenced bits. */
1182 pmdp_test_and_clear_young(vma, addr, pmd);
1183 test_and_clear_page_young(page);
1184 ClearPageReferenced(page);
1190 if (pmd_trans_unstable(pmd))
1193 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1194 for (; addr != end; pte++, addr += PAGE_SIZE) {
1197 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1198 clear_soft_dirty(vma, addr, pte);
1202 if (!pte_present(ptent))
1205 page = vm_normal_page(vma, addr, ptent);
1209 /* Clear accessed and referenced bits. */
1210 ptep_test_and_clear_young(vma, addr, pte);
1211 test_and_clear_page_young(page);
1212 ClearPageReferenced(page);
1214 pte_unmap_unlock(pte - 1, ptl);
1219 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1220 struct mm_walk *walk)
1222 struct clear_refs_private *cp = walk->private;
1223 struct vm_area_struct *vma = walk->vma;
1225 if (vma->vm_flags & VM_PFNMAP)
1229 * Writing 1 to /proc/pid/clear_refs affects all pages.
1230 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1231 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1232 * Writing 4 to /proc/pid/clear_refs affects all pages.
1234 if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1236 if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1241 static const struct mm_walk_ops clear_refs_walk_ops = {
1242 .pmd_entry = clear_refs_pte_range,
1243 .test_walk = clear_refs_test_walk,
1246 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1247 size_t count, loff_t *ppos)
1249 struct task_struct *task;
1250 char buffer[PROC_NUMBUF];
1251 struct mm_struct *mm;
1252 struct vm_area_struct *vma;
1253 enum clear_refs_types type;
1257 memset(buffer, 0, sizeof(buffer));
1258 if (count > sizeof(buffer) - 1)
1259 count = sizeof(buffer) - 1;
1260 if (copy_from_user(buffer, buf, count))
1262 rv = kstrtoint(strstrip(buffer), 10, &itype);
1265 type = (enum clear_refs_types)itype;
1266 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1269 task = get_proc_task(file_inode(file));
1272 mm = get_task_mm(task);
1274 MA_STATE(mas, &mm->mm_mt, 0, 0);
1275 struct mmu_notifier_range range;
1276 struct clear_refs_private cp = {
1280 if (mmap_write_lock_killable(mm)) {
1284 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1286 * Writing 5 to /proc/pid/clear_refs resets the peak
1287 * resident set size to this mm's current rss value.
1289 reset_mm_hiwater_rss(mm);
1293 if (type == CLEAR_REFS_SOFT_DIRTY) {
1294 mas_for_each(&mas, vma, ULONG_MAX) {
1295 if (!(vma->vm_flags & VM_SOFTDIRTY))
1297 vma->vm_flags &= ~VM_SOFTDIRTY;
1298 vma_set_page_prot(vma);
1301 inc_tlb_flush_pending(mm);
1302 mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY,
1303 0, NULL, mm, 0, -1UL);
1304 mmu_notifier_invalidate_range_start(&range);
1306 walk_page_range(mm, 0, -1, &clear_refs_walk_ops, &cp);
1307 if (type == CLEAR_REFS_SOFT_DIRTY) {
1308 mmu_notifier_invalidate_range_end(&range);
1310 dec_tlb_flush_pending(mm);
1313 mmap_write_unlock(mm);
1317 put_task_struct(task);
1322 const struct file_operations proc_clear_refs_operations = {
1323 .write = clear_refs_write,
1324 .llseek = noop_llseek,
1331 struct pagemapread {
1332 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */
1333 pagemap_entry_t *buffer;
1337 #define PAGEMAP_WALK_SIZE (PMD_SIZE)
1338 #define PAGEMAP_WALK_MASK (PMD_MASK)
1340 #define PM_ENTRY_BYTES sizeof(pagemap_entry_t)
1341 #define PM_PFRAME_BITS 55
1342 #define PM_PFRAME_MASK GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1343 #define PM_SOFT_DIRTY BIT_ULL(55)
1344 #define PM_MMAP_EXCLUSIVE BIT_ULL(56)
1345 #define PM_UFFD_WP BIT_ULL(57)
1346 #define PM_FILE BIT_ULL(61)
1347 #define PM_SWAP BIT_ULL(62)
1348 #define PM_PRESENT BIT_ULL(63)
1350 #define PM_END_OF_BUFFER 1
1352 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1354 return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1357 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1358 struct pagemapread *pm)
1360 pm->buffer[pm->pos++] = *pme;
1361 if (pm->pos >= pm->len)
1362 return PM_END_OF_BUFFER;
1366 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1367 __always_unused int depth, struct mm_walk *walk)
1369 struct pagemapread *pm = walk->private;
1370 unsigned long addr = start;
1373 while (addr < end) {
1374 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1375 pagemap_entry_t pme = make_pme(0, 0);
1376 /* End of address space hole, which we mark as non-present. */
1377 unsigned long hole_end;
1380 hole_end = min(end, vma->vm_start);
1384 for (; addr < hole_end; addr += PAGE_SIZE) {
1385 err = add_to_pagemap(addr, &pme, pm);
1393 /* Addresses in the VMA. */
1394 if (vma->vm_flags & VM_SOFTDIRTY)
1395 pme = make_pme(0, PM_SOFT_DIRTY);
1396 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1397 err = add_to_pagemap(addr, &pme, pm);
1406 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1407 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1409 u64 frame = 0, flags = 0;
1410 struct page *page = NULL;
1411 bool migration = false;
1413 if (pte_present(pte)) {
1415 frame = pte_pfn(pte);
1416 flags |= PM_PRESENT;
1417 page = vm_normal_page(vma, addr, pte);
1418 if (pte_soft_dirty(pte))
1419 flags |= PM_SOFT_DIRTY;
1420 if (pte_uffd_wp(pte))
1421 flags |= PM_UFFD_WP;
1422 } else if (is_swap_pte(pte)) {
1424 if (pte_swp_soft_dirty(pte))
1425 flags |= PM_SOFT_DIRTY;
1426 if (pte_swp_uffd_wp(pte))
1427 flags |= PM_UFFD_WP;
1428 entry = pte_to_swp_entry(pte);
1432 * For PFN swap offsets, keeping the offset field
1433 * to be PFN only to be compatible with old smaps.
1435 if (is_pfn_swap_entry(entry))
1436 offset = swp_offset_pfn(entry);
1438 offset = swp_offset(entry);
1439 frame = swp_type(entry) |
1440 (offset << MAX_SWAPFILES_SHIFT);
1443 migration = is_migration_entry(entry);
1444 if (is_pfn_swap_entry(entry))
1445 page = pfn_swap_entry_to_page(entry);
1446 if (pte_marker_entry_uffd_wp(entry))
1447 flags |= PM_UFFD_WP;
1450 if (page && !PageAnon(page))
1452 if (page && !migration && page_mapcount(page) == 1)
1453 flags |= PM_MMAP_EXCLUSIVE;
1454 if (vma->vm_flags & VM_SOFTDIRTY)
1455 flags |= PM_SOFT_DIRTY;
1457 return make_pme(frame, flags);
1460 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1461 struct mm_walk *walk)
1463 struct vm_area_struct *vma = walk->vma;
1464 struct pagemapread *pm = walk->private;
1466 pte_t *pte, *orig_pte;
1468 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1469 bool migration = false;
1471 ptl = pmd_trans_huge_lock(pmdp, vma);
1473 u64 flags = 0, frame = 0;
1475 struct page *page = NULL;
1477 if (vma->vm_flags & VM_SOFTDIRTY)
1478 flags |= PM_SOFT_DIRTY;
1480 if (pmd_present(pmd)) {
1481 page = pmd_page(pmd);
1483 flags |= PM_PRESENT;
1484 if (pmd_soft_dirty(pmd))
1485 flags |= PM_SOFT_DIRTY;
1486 if (pmd_uffd_wp(pmd))
1487 flags |= PM_UFFD_WP;
1489 frame = pmd_pfn(pmd) +
1490 ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1492 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1493 else if (is_swap_pmd(pmd)) {
1494 swp_entry_t entry = pmd_to_swp_entry(pmd);
1495 unsigned long offset;
1498 if (is_pfn_swap_entry(entry))
1499 offset = swp_offset_pfn(entry);
1501 offset = swp_offset(entry);
1503 ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1504 frame = swp_type(entry) |
1505 (offset << MAX_SWAPFILES_SHIFT);
1508 if (pmd_swp_soft_dirty(pmd))
1509 flags |= PM_SOFT_DIRTY;
1510 if (pmd_swp_uffd_wp(pmd))
1511 flags |= PM_UFFD_WP;
1512 VM_BUG_ON(!is_pmd_migration_entry(pmd));
1513 migration = is_migration_entry(entry);
1514 page = pfn_swap_entry_to_page(entry);
1518 if (page && !migration && page_mapcount(page) == 1)
1519 flags |= PM_MMAP_EXCLUSIVE;
1521 for (; addr != end; addr += PAGE_SIZE) {
1522 pagemap_entry_t pme = make_pme(frame, flags);
1524 err = add_to_pagemap(addr, &pme, pm);
1528 if (flags & PM_PRESENT)
1530 else if (flags & PM_SWAP)
1531 frame += (1 << MAX_SWAPFILES_SHIFT);
1538 if (pmd_trans_unstable(pmdp))
1540 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1543 * We can assume that @vma always points to a valid one and @end never
1544 * goes beyond vma->vm_end.
1546 orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1547 for (; addr < end; pte++, addr += PAGE_SIZE) {
1548 pagemap_entry_t pme;
1550 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1551 err = add_to_pagemap(addr, &pme, pm);
1555 pte_unmap_unlock(orig_pte, ptl);
1562 #ifdef CONFIG_HUGETLB_PAGE
1563 /* This function walks within one hugetlb entry in the single call */
1564 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1565 unsigned long addr, unsigned long end,
1566 struct mm_walk *walk)
1568 struct pagemapread *pm = walk->private;
1569 struct vm_area_struct *vma = walk->vma;
1570 u64 flags = 0, frame = 0;
1574 if (vma->vm_flags & VM_SOFTDIRTY)
1575 flags |= PM_SOFT_DIRTY;
1577 pte = huge_ptep_get(ptep);
1578 if (pte_present(pte)) {
1579 struct page *page = pte_page(pte);
1581 if (!PageAnon(page))
1584 if (page_mapcount(page) == 1)
1585 flags |= PM_MMAP_EXCLUSIVE;
1587 if (huge_pte_uffd_wp(pte))
1588 flags |= PM_UFFD_WP;
1590 flags |= PM_PRESENT;
1592 frame = pte_pfn(pte) +
1593 ((addr & ~hmask) >> PAGE_SHIFT);
1594 } else if (pte_swp_uffd_wp_any(pte)) {
1595 flags |= PM_UFFD_WP;
1598 for (; addr != end; addr += PAGE_SIZE) {
1599 pagemap_entry_t pme = make_pme(frame, flags);
1601 err = add_to_pagemap(addr, &pme, pm);
1604 if (pm->show_pfn && (flags & PM_PRESENT))
1613 #define pagemap_hugetlb_range NULL
1614 #endif /* HUGETLB_PAGE */
1616 static const struct mm_walk_ops pagemap_ops = {
1617 .pmd_entry = pagemap_pmd_range,
1618 .pte_hole = pagemap_pte_hole,
1619 .hugetlb_entry = pagemap_hugetlb_range,
1623 * /proc/pid/pagemap - an array mapping virtual pages to pfns
1625 * For each page in the address space, this file contains one 64-bit entry
1626 * consisting of the following:
1628 * Bits 0-54 page frame number (PFN) if present
1629 * Bits 0-4 swap type if swapped
1630 * Bits 5-54 swap offset if swapped
1631 * Bit 55 pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst)
1632 * Bit 56 page exclusively mapped
1633 * Bit 57 pte is uffd-wp write-protected
1635 * Bit 61 page is file-page or shared-anon
1636 * Bit 62 page swapped
1637 * Bit 63 page present
1639 * If the page is not present but in swap, then the PFN contains an
1640 * encoding of the swap file number and the page's offset into the
1641 * swap. Unmapped pages return a null PFN. This allows determining
1642 * precisely which pages are mapped (or in swap) and comparing mapped
1643 * pages between processes.
1645 * Efficient users of this interface will use /proc/pid/maps to
1646 * determine which areas of memory are actually mapped and llseek to
1647 * skip over unmapped regions.
1649 static ssize_t pagemap_read(struct file *file, char __user *buf,
1650 size_t count, loff_t *ppos)
1652 struct mm_struct *mm = file->private_data;
1653 struct pagemapread pm;
1655 unsigned long svpfn;
1656 unsigned long start_vaddr;
1657 unsigned long end_vaddr;
1658 int ret = 0, copied = 0;
1660 if (!mm || !mmget_not_zero(mm))
1664 /* file position must be aligned */
1665 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1672 /* do not disclose physical addresses: attack vector */
1673 pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1675 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1676 pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL);
1682 svpfn = src / PM_ENTRY_BYTES;
1683 end_vaddr = mm->task_size;
1685 /* watch out for wraparound */
1686 start_vaddr = end_vaddr;
1687 if (svpfn <= (ULONG_MAX >> PAGE_SHIFT))
1688 start_vaddr = untagged_addr(svpfn << PAGE_SHIFT);
1690 /* Ensure the address is inside the task */
1691 if (start_vaddr > mm->task_size)
1692 start_vaddr = end_vaddr;
1695 * The odds are that this will stop walking way
1696 * before end_vaddr, because the length of the
1697 * user buffer is tracked in "pm", and the walk
1698 * will stop when we hit the end of the buffer.
1701 while (count && (start_vaddr < end_vaddr)) {
1706 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1708 if (end < start_vaddr || end > end_vaddr)
1710 ret = mmap_read_lock_killable(mm);
1713 ret = walk_page_range(mm, start_vaddr, end, &pagemap_ops, &pm);
1714 mmap_read_unlock(mm);
1717 len = min(count, PM_ENTRY_BYTES * pm.pos);
1718 if (copy_to_user(buf, pm.buffer, len)) {
1727 if (!ret || ret == PM_END_OF_BUFFER)
1738 static int pagemap_open(struct inode *inode, struct file *file)
1740 struct mm_struct *mm;
1742 mm = proc_mem_open(inode, PTRACE_MODE_READ);
1745 file->private_data = mm;
1749 static int pagemap_release(struct inode *inode, struct file *file)
1751 struct mm_struct *mm = file->private_data;
1758 const struct file_operations proc_pagemap_operations = {
1759 .llseek = mem_lseek, /* borrow this */
1760 .read = pagemap_read,
1761 .open = pagemap_open,
1762 .release = pagemap_release,
1764 #endif /* CONFIG_PROC_PAGE_MONITOR */
1769 unsigned long pages;
1771 unsigned long active;
1772 unsigned long writeback;
1773 unsigned long mapcount_max;
1774 unsigned long dirty;
1775 unsigned long swapcache;
1776 unsigned long node[MAX_NUMNODES];
1779 struct numa_maps_private {
1780 struct proc_maps_private proc_maps;
1781 struct numa_maps md;
1784 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1785 unsigned long nr_pages)
1787 int count = page_mapcount(page);
1789 md->pages += nr_pages;
1790 if (pte_dirty || PageDirty(page))
1791 md->dirty += nr_pages;
1793 if (PageSwapCache(page))
1794 md->swapcache += nr_pages;
1796 if (PageActive(page) || PageUnevictable(page))
1797 md->active += nr_pages;
1799 if (PageWriteback(page))
1800 md->writeback += nr_pages;
1803 md->anon += nr_pages;
1805 if (count > md->mapcount_max)
1806 md->mapcount_max = count;
1808 md->node[page_to_nid(page)] += nr_pages;
1811 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1817 if (!pte_present(pte))
1820 page = vm_normal_page(vma, addr, pte);
1821 if (!page || is_zone_device_page(page))
1824 if (PageReserved(page))
1827 nid = page_to_nid(page);
1828 if (!node_isset(nid, node_states[N_MEMORY]))
1834 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1835 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1836 struct vm_area_struct *vma,
1842 if (!pmd_present(pmd))
1845 page = vm_normal_page_pmd(vma, addr, pmd);
1849 if (PageReserved(page))
1852 nid = page_to_nid(page);
1853 if (!node_isset(nid, node_states[N_MEMORY]))
1860 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1861 unsigned long end, struct mm_walk *walk)
1863 struct numa_maps *md = walk->private;
1864 struct vm_area_struct *vma = walk->vma;
1869 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1870 ptl = pmd_trans_huge_lock(pmd, vma);
1874 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1876 gather_stats(page, md, pmd_dirty(*pmd),
1877 HPAGE_PMD_SIZE/PAGE_SIZE);
1882 if (pmd_trans_unstable(pmd))
1885 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1887 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1890 gather_stats(page, md, pte_dirty(*pte), 1);
1892 } while (pte++, addr += PAGE_SIZE, addr != end);
1893 pte_unmap_unlock(orig_pte, ptl);
1897 #ifdef CONFIG_HUGETLB_PAGE
1898 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1899 unsigned long addr, unsigned long end, struct mm_walk *walk)
1901 pte_t huge_pte = huge_ptep_get(pte);
1902 struct numa_maps *md;
1905 if (!pte_present(huge_pte))
1908 page = pte_page(huge_pte);
1911 gather_stats(page, md, pte_dirty(huge_pte), 1);
1916 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1917 unsigned long addr, unsigned long end, struct mm_walk *walk)
1923 static const struct mm_walk_ops show_numa_ops = {
1924 .hugetlb_entry = gather_hugetlb_stats,
1925 .pmd_entry = gather_pte_stats,
1929 * Display pages allocated per node and memory policy via /proc.
1931 static int show_numa_map(struct seq_file *m, void *v)
1933 struct numa_maps_private *numa_priv = m->private;
1934 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1935 struct vm_area_struct *vma = v;
1936 struct numa_maps *md = &numa_priv->md;
1937 struct file *file = vma->vm_file;
1938 struct mm_struct *mm = vma->vm_mm;
1939 struct mempolicy *pol;
1946 /* Ensure we start with an empty set of numa_maps statistics. */
1947 memset(md, 0, sizeof(*md));
1949 pol = __get_vma_policy(vma, vma->vm_start);
1951 mpol_to_str(buffer, sizeof(buffer), pol);
1954 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1957 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1960 seq_puts(m, " file=");
1961 seq_file_path(m, file, "\n\t= ");
1962 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1963 seq_puts(m, " heap");
1964 } else if (is_stack(vma)) {
1965 seq_puts(m, " stack");
1968 if (is_vm_hugetlb_page(vma))
1969 seq_puts(m, " huge");
1971 /* mmap_lock is held by m_start */
1972 walk_page_vma(vma, &show_numa_ops, md);
1978 seq_printf(m, " anon=%lu", md->anon);
1981 seq_printf(m, " dirty=%lu", md->dirty);
1983 if (md->pages != md->anon && md->pages != md->dirty)
1984 seq_printf(m, " mapped=%lu", md->pages);
1986 if (md->mapcount_max > 1)
1987 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1990 seq_printf(m, " swapcache=%lu", md->swapcache);
1992 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1993 seq_printf(m, " active=%lu", md->active);
1996 seq_printf(m, " writeback=%lu", md->writeback);
1998 for_each_node_state(nid, N_MEMORY)
2000 seq_printf(m, " N%d=%lu", nid, md->node[nid]);
2002 seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
2008 static const struct seq_operations proc_pid_numa_maps_op = {
2012 .show = show_numa_map,
2015 static int pid_numa_maps_open(struct inode *inode, struct file *file)
2017 return proc_maps_open(inode, file, &proc_pid_numa_maps_op,
2018 sizeof(struct numa_maps_private));
2021 const struct file_operations proc_pid_numa_maps_operations = {
2022 .open = pid_numa_maps_open,
2024 .llseek = seq_lseek,
2025 .release = proc_map_release,
2028 #endif /* CONFIG_NUMA */