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>
8 #include <linux/seq_file.h>
9 #include <linux/highmem.h>
10 #include <linux/ptrace.h>
11 #include <linux/slab.h>
12 #include <linux/pagemap.h>
13 #include <linux/mempolicy.h>
14 #include <linux/rmap.h>
15 #include <linux/swap.h>
16 #include <linux/sched/mm.h>
17 #include <linux/swapops.h>
18 #include <linux/mmu_notifier.h>
19 #include <linux/page_idle.h>
20 #include <linux/shmem_fs.h>
21 #include <linux/uaccess.h>
22 #include <linux/pkeys.h>
23 #include <linux/minmax.h>
24 #include <linux/overflow.h>
28 #include <asm/tlbflush.h>
31 #define SEQ_PUT_DEC(str, val) \
32 seq_put_decimal_ull_width(m, str, (val) << (PAGE_SHIFT-10), 8)
33 void task_mem(struct seq_file *m, struct mm_struct *mm)
35 unsigned long text, lib, swap, anon, file, shmem;
36 unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
38 anon = get_mm_counter(mm, MM_ANONPAGES);
39 file = get_mm_counter(mm, MM_FILEPAGES);
40 shmem = get_mm_counter(mm, MM_SHMEMPAGES);
43 * Note: to minimize their overhead, mm maintains hiwater_vm and
44 * hiwater_rss only when about to *lower* total_vm or rss. Any
45 * collector of these hiwater stats must therefore get total_vm
46 * and rss too, which will usually be the higher. Barriers? not
47 * worth the effort, such snapshots can always be inconsistent.
49 hiwater_vm = total_vm = mm->total_vm;
50 if (hiwater_vm < mm->hiwater_vm)
51 hiwater_vm = mm->hiwater_vm;
52 hiwater_rss = total_rss = anon + file + shmem;
53 if (hiwater_rss < mm->hiwater_rss)
54 hiwater_rss = mm->hiwater_rss;
56 /* split executable areas between text and lib */
57 text = PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK);
58 text = min(text, mm->exec_vm << PAGE_SHIFT);
59 lib = (mm->exec_vm << PAGE_SHIFT) - text;
61 swap = get_mm_counter(mm, MM_SWAPENTS);
62 SEQ_PUT_DEC("VmPeak:\t", hiwater_vm);
63 SEQ_PUT_DEC(" kB\nVmSize:\t", total_vm);
64 SEQ_PUT_DEC(" kB\nVmLck:\t", mm->locked_vm);
65 SEQ_PUT_DEC(" kB\nVmPin:\t", atomic64_read(&mm->pinned_vm));
66 SEQ_PUT_DEC(" kB\nVmHWM:\t", hiwater_rss);
67 SEQ_PUT_DEC(" kB\nVmRSS:\t", total_rss);
68 SEQ_PUT_DEC(" kB\nRssAnon:\t", anon);
69 SEQ_PUT_DEC(" kB\nRssFile:\t", file);
70 SEQ_PUT_DEC(" kB\nRssShmem:\t", shmem);
71 SEQ_PUT_DEC(" kB\nVmData:\t", mm->data_vm);
72 SEQ_PUT_DEC(" kB\nVmStk:\t", mm->stack_vm);
73 seq_put_decimal_ull_width(m,
74 " kB\nVmExe:\t", text >> 10, 8);
75 seq_put_decimal_ull_width(m,
76 " kB\nVmLib:\t", lib >> 10, 8);
77 seq_put_decimal_ull_width(m,
78 " kB\nVmPTE:\t", mm_pgtables_bytes(mm) >> 10, 8);
79 SEQ_PUT_DEC(" kB\nVmSwap:\t", swap);
81 hugetlb_report_usage(m, mm);
85 unsigned long task_vsize(struct mm_struct *mm)
87 return PAGE_SIZE * mm->total_vm;
90 unsigned long task_statm(struct mm_struct *mm,
91 unsigned long *shared, unsigned long *text,
92 unsigned long *data, unsigned long *resident)
94 *shared = get_mm_counter(mm, MM_FILEPAGES) +
95 get_mm_counter(mm, MM_SHMEMPAGES);
96 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
98 *data = mm->data_vm + mm->stack_vm;
99 *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
105 * Save get_task_policy() for show_numa_map().
107 static void hold_task_mempolicy(struct proc_maps_private *priv)
109 struct task_struct *task = priv->task;
112 priv->task_mempolicy = get_task_policy(task);
113 mpol_get(priv->task_mempolicy);
116 static void release_task_mempolicy(struct proc_maps_private *priv)
118 mpol_put(priv->task_mempolicy);
121 static void hold_task_mempolicy(struct proc_maps_private *priv)
124 static void release_task_mempolicy(struct proc_maps_private *priv)
129 static struct vm_area_struct *proc_get_vma(struct proc_maps_private *priv,
132 struct vm_area_struct *vma = vma_next(&priv->iter);
135 *ppos = vma->vm_start;
138 vma = get_gate_vma(priv->mm);
144 static void *m_start(struct seq_file *m, loff_t *ppos)
146 struct proc_maps_private *priv = m->private;
147 unsigned long last_addr = *ppos;
148 struct mm_struct *mm;
150 /* See m_next(). 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 || !mmget_not_zero(mm)) {
160 put_task_struct(priv->task);
165 if (mmap_read_lock_killable(mm)) {
167 put_task_struct(priv->task);
169 return ERR_PTR(-EINTR);
172 vma_iter_init(&priv->iter, mm, last_addr);
173 hold_task_mempolicy(priv);
174 if (last_addr == -2UL)
175 return get_gate_vma(mm);
177 return proc_get_vma(priv, ppos);
180 static void *m_next(struct seq_file *m, void *v, loff_t *ppos)
186 return proc_get_vma(m->private, ppos);
189 static void m_stop(struct seq_file *m, void *v)
191 struct proc_maps_private *priv = m->private;
192 struct mm_struct *mm = priv->mm;
197 release_task_mempolicy(priv);
198 mmap_read_unlock(mm);
200 put_task_struct(priv->task);
204 static int proc_maps_open(struct inode *inode, struct file *file,
205 const struct seq_operations *ops, int psize)
207 struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
213 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
214 if (IS_ERR(priv->mm)) {
215 int err = PTR_ERR(priv->mm);
217 seq_release_private(inode, file);
224 static int proc_map_release(struct inode *inode, struct file *file)
226 struct seq_file *seq = file->private_data;
227 struct proc_maps_private *priv = seq->private;
232 return seq_release_private(inode, file);
235 static int do_maps_open(struct inode *inode, struct file *file,
236 const struct seq_operations *ops)
238 return proc_maps_open(inode, file, ops,
239 sizeof(struct proc_maps_private));
242 static void show_vma_header_prefix(struct seq_file *m,
243 unsigned long start, unsigned long end,
244 vm_flags_t flags, unsigned long long pgoff,
245 dev_t dev, unsigned long ino)
247 seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
248 seq_put_hex_ll(m, NULL, start, 8);
249 seq_put_hex_ll(m, "-", end, 8);
251 seq_putc(m, flags & VM_READ ? 'r' : '-');
252 seq_putc(m, flags & VM_WRITE ? 'w' : '-');
253 seq_putc(m, flags & VM_EXEC ? 'x' : '-');
254 seq_putc(m, flags & VM_MAYSHARE ? 's' : 'p');
255 seq_put_hex_ll(m, " ", pgoff, 8);
256 seq_put_hex_ll(m, " ", MAJOR(dev), 2);
257 seq_put_hex_ll(m, ":", MINOR(dev), 2);
258 seq_put_decimal_ull(m, " ", ino);
263 show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
265 struct anon_vma_name *anon_name = NULL;
266 struct mm_struct *mm = vma->vm_mm;
267 struct file *file = vma->vm_file;
268 vm_flags_t flags = vma->vm_flags;
269 unsigned long ino = 0;
270 unsigned long long pgoff = 0;
271 unsigned long start, end;
273 const char *name = NULL;
276 const struct inode *inode = file_user_inode(vma->vm_file);
278 dev = inode->i_sb->s_dev;
280 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
283 start = vma->vm_start;
285 show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino);
287 anon_name = anon_vma_name(vma);
290 * Print the dentry name for named mappings, and a
291 * special [heap] marker for the heap:
296 * If user named this anon shared memory via
297 * prctl(PR_SET_VMA ..., use the provided name.
300 seq_printf(m, "[anon_shmem:%s]", anon_name->name);
302 seq_path(m, file_user_path(file), "\n");
306 if (vma->vm_ops && vma->vm_ops->name) {
307 name = vma->vm_ops->name(vma);
312 name = arch_vma_name(vma);
319 if (vma_is_initial_heap(vma)) {
324 if (vma_is_initial_stack(vma)) {
331 seq_printf(m, "[anon:%s]", anon_name->name);
343 static int show_map(struct seq_file *m, void *v)
349 static const struct seq_operations proc_pid_maps_op = {
356 static int pid_maps_open(struct inode *inode, struct file *file)
358 return do_maps_open(inode, file, &proc_pid_maps_op);
361 const struct file_operations proc_pid_maps_operations = {
362 .open = pid_maps_open,
365 .release = proc_map_release,
369 * Proportional Set Size(PSS): my share of RSS.
371 * PSS of a process is the count of pages it has in memory, where each
372 * page is divided by the number of processes sharing it. So if a
373 * process has 1000 pages all to itself, and 1000 shared with one other
374 * process, its PSS will be 1500.
376 * To keep (accumulated) division errors low, we adopt a 64bit
377 * fixed-point pss counter to minimize division errors. So (pss >>
378 * PSS_SHIFT) would be the real byte count.
380 * A shift of 12 before division means (assuming 4K page size):
381 * - 1M 3-user-pages add up to 8KB errors;
382 * - supports mapcount up to 2^24, or 16M;
383 * - supports PSS up to 2^52 bytes, or 4PB.
387 #ifdef CONFIG_PROC_PAGE_MONITOR
388 struct mem_size_stats {
389 unsigned long resident;
390 unsigned long shared_clean;
391 unsigned long shared_dirty;
392 unsigned long private_clean;
393 unsigned long private_dirty;
394 unsigned long referenced;
395 unsigned long anonymous;
396 unsigned long lazyfree;
397 unsigned long anonymous_thp;
398 unsigned long shmem_thp;
399 unsigned long file_thp;
401 unsigned long shared_hugetlb;
402 unsigned long private_hugetlb;
413 static void smaps_page_accumulate(struct mem_size_stats *mss,
414 struct page *page, unsigned long size, unsigned long pss,
415 bool dirty, bool locked, bool private)
420 mss->pss_anon += pss;
421 else if (PageSwapBacked(page))
422 mss->pss_shmem += pss;
424 mss->pss_file += pss;
427 mss->pss_locked += pss;
429 if (dirty || PageDirty(page)) {
430 mss->pss_dirty += pss;
432 mss->private_dirty += size;
434 mss->shared_dirty += size;
437 mss->private_clean += size;
439 mss->shared_clean += size;
443 static void smaps_account(struct mem_size_stats *mss, struct page *page,
444 bool compound, bool young, bool dirty, bool locked,
447 int i, nr = compound ? compound_nr(page) : 1;
448 unsigned long size = nr * PAGE_SIZE;
451 * First accumulate quantities that depend only on |size| and the type
452 * of the compound page.
454 if (PageAnon(page)) {
455 mss->anonymous += size;
456 if (!PageSwapBacked(page) && !dirty && !PageDirty(page))
457 mss->lazyfree += size;
463 mss->resident += size;
464 /* Accumulate the size in pages that have been accessed. */
465 if (young || page_is_young(page) || PageReferenced(page))
466 mss->referenced += size;
469 * Then accumulate quantities that may depend on sharing, or that may
470 * differ page-by-page.
472 * page_count(page) == 1 guarantees the page is mapped exactly once.
473 * If any subpage of the compound page mapped with PTE it would elevate
476 * The page_mapcount() is called to get a snapshot of the mapcount.
477 * Without holding the page lock this snapshot can be slightly wrong as
478 * we cannot always read the mapcount atomically. It is not safe to
479 * call page_mapcount() even with PTL held if the page is not mapped,
480 * especially for migration entries. Treat regular migration entries
483 if ((page_count(page) == 1) || migration) {
484 smaps_page_accumulate(mss, page, size, size << PSS_SHIFT, dirty,
488 for (i = 0; i < nr; i++, page++) {
489 int mapcount = page_mapcount(page);
490 unsigned long pss = PAGE_SIZE << PSS_SHIFT;
493 smaps_page_accumulate(mss, page, PAGE_SIZE, pss, dirty, locked,
499 static int smaps_pte_hole(unsigned long addr, unsigned long end,
500 __always_unused int depth, struct mm_walk *walk)
502 struct mem_size_stats *mss = walk->private;
503 struct vm_area_struct *vma = walk->vma;
505 mss->swap += shmem_partial_swap_usage(walk->vma->vm_file->f_mapping,
506 linear_page_index(vma, addr),
507 linear_page_index(vma, end));
512 #define smaps_pte_hole NULL
513 #endif /* CONFIG_SHMEM */
515 static void smaps_pte_hole_lookup(unsigned long addr, struct mm_walk *walk)
518 if (walk->ops->pte_hole) {
519 /* depth is not used */
520 smaps_pte_hole(addr, addr + PAGE_SIZE, 0, walk);
525 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
526 struct mm_walk *walk)
528 struct mem_size_stats *mss = walk->private;
529 struct vm_area_struct *vma = walk->vma;
530 bool locked = !!(vma->vm_flags & VM_LOCKED);
531 struct page *page = NULL;
532 bool migration = false, young = false, dirty = false;
533 pte_t ptent = ptep_get(pte);
535 if (pte_present(ptent)) {
536 page = vm_normal_page(vma, addr, ptent);
537 young = pte_young(ptent);
538 dirty = pte_dirty(ptent);
539 } else if (is_swap_pte(ptent)) {
540 swp_entry_t swpent = pte_to_swp_entry(ptent);
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 page = vm_normal_page_pmd(vma, addr, *pmd);
583 } else if (unlikely(thp_migration_supported() && is_swap_pmd(*pmd))) {
584 swp_entry_t entry = pmd_to_swp_entry(*pmd);
586 if (is_migration_entry(entry)) {
588 page = pfn_swap_entry_to_page(entry);
591 if (IS_ERR_OR_NULL(page))
594 mss->anonymous_thp += HPAGE_PMD_SIZE;
595 else if (PageSwapBacked(page))
596 mss->shmem_thp += HPAGE_PMD_SIZE;
597 else if (is_zone_device_page(page))
600 mss->file_thp += HPAGE_PMD_SIZE;
602 smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd),
606 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
607 struct mm_walk *walk)
612 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
613 struct mm_walk *walk)
615 struct vm_area_struct *vma = walk->vma;
619 ptl = pmd_trans_huge_lock(pmd, vma);
621 smaps_pmd_entry(pmd, addr, walk);
626 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
628 walk->action = ACTION_AGAIN;
631 for (; addr != end; pte++, addr += PAGE_SIZE)
632 smaps_pte_entry(pte, addr, walk);
633 pte_unmap_unlock(pte - 1, ptl);
639 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
642 * Don't forget to update Documentation/ on changes.
644 static const char mnemonics[BITS_PER_LONG][2] = {
646 * In case if we meet a flag we don't know about.
648 [0 ... (BITS_PER_LONG-1)] = "??",
650 [ilog2(VM_READ)] = "rd",
651 [ilog2(VM_WRITE)] = "wr",
652 [ilog2(VM_EXEC)] = "ex",
653 [ilog2(VM_SHARED)] = "sh",
654 [ilog2(VM_MAYREAD)] = "mr",
655 [ilog2(VM_MAYWRITE)] = "mw",
656 [ilog2(VM_MAYEXEC)] = "me",
657 [ilog2(VM_MAYSHARE)] = "ms",
658 [ilog2(VM_GROWSDOWN)] = "gd",
659 [ilog2(VM_PFNMAP)] = "pf",
660 [ilog2(VM_LOCKED)] = "lo",
661 [ilog2(VM_IO)] = "io",
662 [ilog2(VM_SEQ_READ)] = "sr",
663 [ilog2(VM_RAND_READ)] = "rr",
664 [ilog2(VM_DONTCOPY)] = "dc",
665 [ilog2(VM_DONTEXPAND)] = "de",
666 [ilog2(VM_LOCKONFAULT)] = "lf",
667 [ilog2(VM_ACCOUNT)] = "ac",
668 [ilog2(VM_NORESERVE)] = "nr",
669 [ilog2(VM_HUGETLB)] = "ht",
670 [ilog2(VM_SYNC)] = "sf",
671 [ilog2(VM_ARCH_1)] = "ar",
672 [ilog2(VM_WIPEONFORK)] = "wf",
673 [ilog2(VM_DONTDUMP)] = "dd",
674 #ifdef CONFIG_ARM64_BTI
675 [ilog2(VM_ARM64_BTI)] = "bt",
677 #ifdef CONFIG_MEM_SOFT_DIRTY
678 [ilog2(VM_SOFTDIRTY)] = "sd",
680 [ilog2(VM_MIXEDMAP)] = "mm",
681 [ilog2(VM_HUGEPAGE)] = "hg",
682 [ilog2(VM_NOHUGEPAGE)] = "nh",
683 [ilog2(VM_MERGEABLE)] = "mg",
684 [ilog2(VM_UFFD_MISSING)]= "um",
685 [ilog2(VM_UFFD_WP)] = "uw",
686 #ifdef CONFIG_ARM64_MTE
687 [ilog2(VM_MTE)] = "mt",
688 [ilog2(VM_MTE_ALLOWED)] = "",
690 #ifdef CONFIG_ARCH_HAS_PKEYS
691 /* These come out via ProtectionKey: */
692 [ilog2(VM_PKEY_BIT0)] = "",
693 [ilog2(VM_PKEY_BIT1)] = "",
694 [ilog2(VM_PKEY_BIT2)] = "",
695 [ilog2(VM_PKEY_BIT3)] = "",
697 [ilog2(VM_PKEY_BIT4)] = "",
699 #endif /* CONFIG_ARCH_HAS_PKEYS */
700 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_MINOR
701 [ilog2(VM_UFFD_MINOR)] = "ui",
702 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_MINOR */
703 #ifdef CONFIG_X86_USER_SHADOW_STACK
704 [ilog2(VM_SHADOW_STACK)] = "ss",
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;
730 pte_t ptent = ptep_get(pte);
732 if (pte_present(ptent)) {
733 page = vm_normal_page(vma, addr, ptent);
734 } else if (is_swap_pte(ptent)) {
735 swp_entry_t swpent = pte_to_swp_entry(ptent);
737 if (is_pfn_swap_entry(swpent))
738 page = pfn_swap_entry_to_page(swpent);
741 if (page_mapcount(page) >= 2 || hugetlb_pmd_shared(pte))
742 mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
744 mss->private_hugetlb += huge_page_size(hstate_vma(vma));
749 #define smaps_hugetlb_range NULL
750 #endif /* HUGETLB_PAGE */
752 static const struct mm_walk_ops smaps_walk_ops = {
753 .pmd_entry = smaps_pte_range,
754 .hugetlb_entry = smaps_hugetlb_range,
755 .walk_lock = PGWALK_RDLOCK,
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,
762 .walk_lock = PGWALK_RDLOCK,
766 * Gather mem stats from @vma with the indicated beginning
767 * address @start, and keep them in @mss.
769 * Use vm_start of @vma as the beginning address if @start is 0.
771 static void smap_gather_stats(struct vm_area_struct *vma,
772 struct mem_size_stats *mss, unsigned long start)
774 const struct mm_walk_ops *ops = &smaps_walk_ops;
777 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\nKSM: ", mss->ksm);
837 SEQ_PUT_DEC(" kB\nLazyFree: ", mss->lazyfree);
838 SEQ_PUT_DEC(" kB\nAnonHugePages: ", mss->anonymous_thp);
839 SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp);
840 SEQ_PUT_DEC(" kB\nFilePmdMapped: ", mss->file_thp);
841 SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb);
842 seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ",
843 mss->private_hugetlb >> 10, 7);
844 SEQ_PUT_DEC(" kB\nSwap: ", mss->swap);
845 SEQ_PUT_DEC(" kB\nSwapPss: ",
846 mss->swap_pss >> PSS_SHIFT);
847 SEQ_PUT_DEC(" kB\nLocked: ",
848 mss->pss_locked >> PSS_SHIFT);
849 seq_puts(m, " kB\n");
852 static int show_smap(struct seq_file *m, void *v)
854 struct vm_area_struct *vma = v;
855 struct mem_size_stats mss = {};
857 smap_gather_stats(vma, &mss, 0);
859 show_map_vma(m, vma);
861 SEQ_PUT_DEC("Size: ", vma->vm_end - vma->vm_start);
862 SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma));
863 SEQ_PUT_DEC(" kB\nMMUPageSize: ", vma_mmu_pagesize(vma));
864 seq_puts(m, " kB\n");
866 __show_smap(m, &mss, false);
868 seq_printf(m, "THPeligible: %8u\n",
869 !!thp_vma_allowable_orders(vma, vma->vm_flags, true, false,
870 true, THP_ORDERS_ALL));
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 VMA_ITERATOR(vmi, mm, 0);
889 priv->task = get_proc_task(priv->inode);
893 if (!mm || !mmget_not_zero(mm)) {
898 ret = mmap_read_lock_killable(mm);
902 hold_task_mempolicy(priv);
903 vma = vma_next(&vmi);
908 vma_start = vma->vm_start;
910 smap_gather_stats(vma, &mss, 0);
911 last_vma_end = vma->vm_end;
914 * Release mmap_lock temporarily if someone wants to
915 * access it for write request.
917 if (mmap_lock_is_contended(mm)) {
918 vma_iter_invalidate(&vmi);
919 mmap_read_unlock(mm);
920 ret = mmap_read_lock_killable(mm);
922 release_task_mempolicy(priv);
927 * After dropping the lock, there are four cases to
928 * consider. See the following example for explanation.
930 * +------+------+-----------+
931 * | VMA1 | VMA2 | VMA3 |
932 * +------+------+-----------+
936 * Suppose we drop the lock after reading VMA2 due to
937 * contention, then we get:
941 * 1) VMA2 is freed, but VMA3 exists:
943 * vma_next(vmi) will return VMA3.
944 * In this case, just continue from VMA3.
946 * 2) VMA2 still exists:
948 * vma_next(vmi) will return VMA3.
949 * In this case, just continue from VMA3.
951 * 3) No more VMAs can be found:
953 * vma_next(vmi) will return NULL.
954 * No more things to do, just break.
956 * 4) (last_vma_end - 1) is the middle of a vma (VMA'):
958 * vma_next(vmi) will return VMA' whose range
959 * contains last_vma_end.
960 * Iterate VMA' from last_vma_end.
962 vma = vma_next(&vmi);
967 /* Case 1 and 2 above */
968 if (vma->vm_start >= last_vma_end)
972 if (vma->vm_end > last_vma_end)
973 smap_gather_stats(vma, &mss, last_vma_end);
975 } for_each_vma(vmi, vma);
978 show_vma_header_prefix(m, vma_start, last_vma_end, 0, 0, 0, 0);
980 seq_puts(m, "[rollup]\n");
982 __show_smap(m, &mss, true);
984 release_task_mempolicy(priv);
985 mmap_read_unlock(mm);
990 put_task_struct(priv->task);
997 static const struct seq_operations proc_pid_smaps_op = {
1004 static int pid_smaps_open(struct inode *inode, struct file *file)
1006 return do_maps_open(inode, file, &proc_pid_smaps_op);
1009 static int smaps_rollup_open(struct inode *inode, struct file *file)
1012 struct proc_maps_private *priv;
1014 priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT);
1018 ret = single_open(file, show_smaps_rollup, priv);
1022 priv->inode = inode;
1023 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
1024 if (IS_ERR(priv->mm)) {
1025 ret = PTR_ERR(priv->mm);
1027 single_release(inode, file);
1038 static int smaps_rollup_release(struct inode *inode, struct file *file)
1040 struct seq_file *seq = file->private_data;
1041 struct proc_maps_private *priv = seq->private;
1047 return single_release(inode, file);
1050 const struct file_operations proc_pid_smaps_operations = {
1051 .open = pid_smaps_open,
1053 .llseek = seq_lseek,
1054 .release = proc_map_release,
1057 const struct file_operations proc_pid_smaps_rollup_operations = {
1058 .open = smaps_rollup_open,
1060 .llseek = seq_lseek,
1061 .release = smaps_rollup_release,
1064 enum clear_refs_types {
1068 CLEAR_REFS_SOFT_DIRTY,
1069 CLEAR_REFS_MM_HIWATER_RSS,
1073 struct clear_refs_private {
1074 enum clear_refs_types type;
1077 #ifdef CONFIG_MEM_SOFT_DIRTY
1079 static inline bool pte_is_pinned(struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1083 if (!pte_write(pte))
1085 if (!is_cow_mapping(vma->vm_flags))
1087 if (likely(!test_bit(MMF_HAS_PINNED, &vma->vm_mm->flags)))
1089 page = vm_normal_page(vma, addr, pte);
1092 return page_maybe_dma_pinned(page);
1095 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1096 unsigned long addr, pte_t *pte)
1099 * The soft-dirty tracker uses #PF-s to catch writes
1100 * to pages, so write-protect the pte as well. See the
1101 * Documentation/admin-guide/mm/soft-dirty.rst for full description
1102 * of how soft-dirty works.
1104 pte_t ptent = ptep_get(pte);
1106 if (pte_present(ptent)) {
1109 if (pte_is_pinned(vma, addr, ptent))
1111 old_pte = ptep_modify_prot_start(vma, addr, pte);
1112 ptent = pte_wrprotect(old_pte);
1113 ptent = pte_clear_soft_dirty(ptent);
1114 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent);
1115 } else if (is_swap_pte(ptent)) {
1116 ptent = pte_swp_clear_soft_dirty(ptent);
1117 set_pte_at(vma->vm_mm, addr, pte, ptent);
1121 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1122 unsigned long addr, pte_t *pte)
1127 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
1128 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1129 unsigned long addr, pmd_t *pmdp)
1131 pmd_t old, pmd = *pmdp;
1133 if (pmd_present(pmd)) {
1134 /* See comment in change_huge_pmd() */
1135 old = pmdp_invalidate(vma, addr, pmdp);
1137 pmd = pmd_mkdirty(pmd);
1139 pmd = pmd_mkyoung(pmd);
1141 pmd = pmd_wrprotect(pmd);
1142 pmd = pmd_clear_soft_dirty(pmd);
1144 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1145 } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1146 pmd = pmd_swp_clear_soft_dirty(pmd);
1147 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1151 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1152 unsigned long addr, pmd_t *pmdp)
1157 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
1158 unsigned long end, struct mm_walk *walk)
1160 struct clear_refs_private *cp = walk->private;
1161 struct vm_area_struct *vma = walk->vma;
1166 ptl = pmd_trans_huge_lock(pmd, vma);
1168 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1169 clear_soft_dirty_pmd(vma, addr, pmd);
1173 if (!pmd_present(*pmd))
1176 page = pmd_page(*pmd);
1178 /* Clear accessed and referenced bits. */
1179 pmdp_test_and_clear_young(vma, addr, pmd);
1180 test_and_clear_page_young(page);
1181 ClearPageReferenced(page);
1187 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1189 walk->action = ACTION_AGAIN;
1192 for (; addr != end; pte++, addr += PAGE_SIZE) {
1193 ptent = ptep_get(pte);
1195 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1196 clear_soft_dirty(vma, addr, pte);
1200 if (!pte_present(ptent))
1203 page = vm_normal_page(vma, addr, ptent);
1207 /* Clear accessed and referenced bits. */
1208 ptep_test_and_clear_young(vma, addr, pte);
1209 test_and_clear_page_young(page);
1210 ClearPageReferenced(page);
1212 pte_unmap_unlock(pte - 1, ptl);
1217 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1218 struct mm_walk *walk)
1220 struct clear_refs_private *cp = walk->private;
1221 struct vm_area_struct *vma = walk->vma;
1223 if (vma->vm_flags & VM_PFNMAP)
1227 * Writing 1 to /proc/pid/clear_refs affects all pages.
1228 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1229 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1230 * Writing 4 to /proc/pid/clear_refs affects all pages.
1232 if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1234 if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1239 static const struct mm_walk_ops clear_refs_walk_ops = {
1240 .pmd_entry = clear_refs_pte_range,
1241 .test_walk = clear_refs_test_walk,
1242 .walk_lock = PGWALK_WRLOCK,
1245 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1246 size_t count, loff_t *ppos)
1248 struct task_struct *task;
1249 char buffer[PROC_NUMBUF] = {};
1250 struct mm_struct *mm;
1251 struct vm_area_struct *vma;
1252 enum clear_refs_types type;
1256 if (count > sizeof(buffer) - 1)
1257 count = sizeof(buffer) - 1;
1258 if (copy_from_user(buffer, buf, count))
1260 rv = kstrtoint(strstrip(buffer), 10, &itype);
1263 type = (enum clear_refs_types)itype;
1264 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1267 task = get_proc_task(file_inode(file));
1270 mm = get_task_mm(task);
1272 VMA_ITERATOR(vmi, mm, 0);
1273 struct mmu_notifier_range range;
1274 struct clear_refs_private cp = {
1278 if (mmap_write_lock_killable(mm)) {
1282 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1284 * Writing 5 to /proc/pid/clear_refs resets the peak
1285 * resident set size to this mm's current rss value.
1287 reset_mm_hiwater_rss(mm);
1291 if (type == CLEAR_REFS_SOFT_DIRTY) {
1292 for_each_vma(vmi, vma) {
1293 if (!(vma->vm_flags & VM_SOFTDIRTY))
1295 vm_flags_clear(vma, VM_SOFTDIRTY);
1296 vma_set_page_prot(vma);
1299 inc_tlb_flush_pending(mm);
1300 mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY,
1302 mmu_notifier_invalidate_range_start(&range);
1304 walk_page_range(mm, 0, -1, &clear_refs_walk_ops, &cp);
1305 if (type == CLEAR_REFS_SOFT_DIRTY) {
1306 mmu_notifier_invalidate_range_end(&range);
1308 dec_tlb_flush_pending(mm);
1311 mmap_write_unlock(mm);
1315 put_task_struct(task);
1320 const struct file_operations proc_clear_refs_operations = {
1321 .write = clear_refs_write,
1322 .llseek = noop_llseek,
1329 struct pagemapread {
1330 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */
1331 pagemap_entry_t *buffer;
1335 #define PAGEMAP_WALK_SIZE (PMD_SIZE)
1336 #define PAGEMAP_WALK_MASK (PMD_MASK)
1338 #define PM_ENTRY_BYTES sizeof(pagemap_entry_t)
1339 #define PM_PFRAME_BITS 55
1340 #define PM_PFRAME_MASK GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1341 #define PM_SOFT_DIRTY BIT_ULL(55)
1342 #define PM_MMAP_EXCLUSIVE BIT_ULL(56)
1343 #define PM_UFFD_WP BIT_ULL(57)
1344 #define PM_FILE BIT_ULL(61)
1345 #define PM_SWAP BIT_ULL(62)
1346 #define PM_PRESENT BIT_ULL(63)
1348 #define PM_END_OF_BUFFER 1
1350 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1352 return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1355 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1356 struct pagemapread *pm)
1358 pm->buffer[pm->pos++] = *pme;
1359 if (pm->pos >= pm->len)
1360 return PM_END_OF_BUFFER;
1364 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1365 __always_unused int depth, struct mm_walk *walk)
1367 struct pagemapread *pm = walk->private;
1368 unsigned long addr = start;
1371 while (addr < end) {
1372 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1373 pagemap_entry_t pme = make_pme(0, 0);
1374 /* End of address space hole, which we mark as non-present. */
1375 unsigned long hole_end;
1378 hole_end = min(end, vma->vm_start);
1382 for (; addr < hole_end; addr += PAGE_SIZE) {
1383 err = add_to_pagemap(addr, &pme, pm);
1391 /* Addresses in the VMA. */
1392 if (vma->vm_flags & VM_SOFTDIRTY)
1393 pme = make_pme(0, PM_SOFT_DIRTY);
1394 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1395 err = add_to_pagemap(addr, &pme, pm);
1404 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1405 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1407 u64 frame = 0, flags = 0;
1408 struct page *page = NULL;
1409 bool migration = false;
1411 if (pte_present(pte)) {
1413 frame = pte_pfn(pte);
1414 flags |= PM_PRESENT;
1415 page = vm_normal_page(vma, addr, pte);
1416 if (pte_soft_dirty(pte))
1417 flags |= PM_SOFT_DIRTY;
1418 if (pte_uffd_wp(pte))
1419 flags |= PM_UFFD_WP;
1420 } else if (is_swap_pte(pte)) {
1422 if (pte_swp_soft_dirty(pte))
1423 flags |= PM_SOFT_DIRTY;
1424 if (pte_swp_uffd_wp(pte))
1425 flags |= PM_UFFD_WP;
1426 entry = pte_to_swp_entry(pte);
1430 * For PFN swap offsets, keeping the offset field
1431 * to be PFN only to be compatible with old smaps.
1433 if (is_pfn_swap_entry(entry))
1434 offset = swp_offset_pfn(entry);
1436 offset = swp_offset(entry);
1437 frame = swp_type(entry) |
1438 (offset << MAX_SWAPFILES_SHIFT);
1441 migration = is_migration_entry(entry);
1442 if (is_pfn_swap_entry(entry))
1443 page = pfn_swap_entry_to_page(entry);
1444 if (pte_marker_entry_uffd_wp(entry))
1445 flags |= PM_UFFD_WP;
1448 if (page && !PageAnon(page))
1450 if (page && !migration && page_mapcount(page) == 1)
1451 flags |= PM_MMAP_EXCLUSIVE;
1452 if (vma->vm_flags & VM_SOFTDIRTY)
1453 flags |= PM_SOFT_DIRTY;
1455 return make_pme(frame, flags);
1458 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1459 struct mm_walk *walk)
1461 struct vm_area_struct *vma = walk->vma;
1462 struct pagemapread *pm = walk->private;
1464 pte_t *pte, *orig_pte;
1466 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1467 bool migration = false;
1469 ptl = pmd_trans_huge_lock(pmdp, vma);
1471 u64 flags = 0, frame = 0;
1473 struct page *page = NULL;
1475 if (vma->vm_flags & VM_SOFTDIRTY)
1476 flags |= PM_SOFT_DIRTY;
1478 if (pmd_present(pmd)) {
1479 page = pmd_page(pmd);
1481 flags |= PM_PRESENT;
1482 if (pmd_soft_dirty(pmd))
1483 flags |= PM_SOFT_DIRTY;
1484 if (pmd_uffd_wp(pmd))
1485 flags |= PM_UFFD_WP;
1487 frame = pmd_pfn(pmd) +
1488 ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1490 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1491 else if (is_swap_pmd(pmd)) {
1492 swp_entry_t entry = pmd_to_swp_entry(pmd);
1493 unsigned long offset;
1496 if (is_pfn_swap_entry(entry))
1497 offset = swp_offset_pfn(entry);
1499 offset = swp_offset(entry);
1501 ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1502 frame = swp_type(entry) |
1503 (offset << MAX_SWAPFILES_SHIFT);
1506 if (pmd_swp_soft_dirty(pmd))
1507 flags |= PM_SOFT_DIRTY;
1508 if (pmd_swp_uffd_wp(pmd))
1509 flags |= PM_UFFD_WP;
1510 VM_BUG_ON(!is_pmd_migration_entry(pmd));
1511 migration = is_migration_entry(entry);
1512 page = pfn_swap_entry_to_page(entry);
1516 if (page && !migration && page_mapcount(page) == 1)
1517 flags |= PM_MMAP_EXCLUSIVE;
1519 for (; addr != end; addr += PAGE_SIZE) {
1520 pagemap_entry_t pme = make_pme(frame, flags);
1522 err = add_to_pagemap(addr, &pme, pm);
1526 if (flags & PM_PRESENT)
1528 else if (flags & PM_SWAP)
1529 frame += (1 << MAX_SWAPFILES_SHIFT);
1535 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1538 * We can assume that @vma always points to a valid one and @end never
1539 * goes beyond vma->vm_end.
1541 orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1543 walk->action = ACTION_AGAIN;
1546 for (; addr < end; pte++, addr += PAGE_SIZE) {
1547 pagemap_entry_t pme;
1549 pme = pte_to_pagemap_entry(pm, vma, addr, ptep_get(pte));
1550 err = add_to_pagemap(addr, &pme, pm);
1554 pte_unmap_unlock(orig_pte, ptl);
1561 #ifdef CONFIG_HUGETLB_PAGE
1562 /* This function walks within one hugetlb entry in the single call */
1563 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1564 unsigned long addr, unsigned long end,
1565 struct mm_walk *walk)
1567 struct pagemapread *pm = walk->private;
1568 struct vm_area_struct *vma = walk->vma;
1569 u64 flags = 0, frame = 0;
1573 if (vma->vm_flags & VM_SOFTDIRTY)
1574 flags |= PM_SOFT_DIRTY;
1576 pte = huge_ptep_get(ptep);
1577 if (pte_present(pte)) {
1578 struct page *page = pte_page(pte);
1580 if (!PageAnon(page))
1583 if (page_mapcount(page) == 1)
1584 flags |= PM_MMAP_EXCLUSIVE;
1586 if (huge_pte_uffd_wp(pte))
1587 flags |= PM_UFFD_WP;
1589 flags |= PM_PRESENT;
1591 frame = pte_pfn(pte) +
1592 ((addr & ~hmask) >> PAGE_SHIFT);
1593 } else if (pte_swp_uffd_wp_any(pte)) {
1594 flags |= PM_UFFD_WP;
1597 for (; addr != end; addr += PAGE_SIZE) {
1598 pagemap_entry_t pme = make_pme(frame, flags);
1600 err = add_to_pagemap(addr, &pme, pm);
1603 if (pm->show_pfn && (flags & PM_PRESENT))
1612 #define pagemap_hugetlb_range NULL
1613 #endif /* HUGETLB_PAGE */
1615 static const struct mm_walk_ops pagemap_ops = {
1616 .pmd_entry = pagemap_pmd_range,
1617 .pte_hole = pagemap_pte_hole,
1618 .hugetlb_entry = pagemap_hugetlb_range,
1619 .walk_lock = PGWALK_RDLOCK,
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)) {
1690 ret = mmap_read_lock_killable(mm);
1693 start_vaddr = untagged_addr_remote(mm, svpfn << PAGE_SHIFT);
1694 mmap_read_unlock(mm);
1696 end = start_vaddr + ((count / PM_ENTRY_BYTES) << PAGE_SHIFT);
1697 if (end >= start_vaddr && end < mm->task_size)
1701 /* Ensure the address is inside the task */
1702 if (start_vaddr > mm->task_size)
1703 start_vaddr = end_vaddr;
1706 while (count && (start_vaddr < end_vaddr)) {
1711 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1713 if (end < start_vaddr || end > end_vaddr)
1715 ret = mmap_read_lock_killable(mm);
1718 ret = walk_page_range(mm, start_vaddr, end, &pagemap_ops, &pm);
1719 mmap_read_unlock(mm);
1722 len = min(count, PM_ENTRY_BYTES * pm.pos);
1723 if (copy_to_user(buf, pm.buffer, len)) {
1732 if (!ret || ret == PM_END_OF_BUFFER)
1743 static int pagemap_open(struct inode *inode, struct file *file)
1745 struct mm_struct *mm;
1747 mm = proc_mem_open(inode, PTRACE_MODE_READ);
1750 file->private_data = mm;
1754 static int pagemap_release(struct inode *inode, struct file *file)
1756 struct mm_struct *mm = file->private_data;
1763 #define PM_SCAN_CATEGORIES (PAGE_IS_WPALLOWED | PAGE_IS_WRITTEN | \
1764 PAGE_IS_FILE | PAGE_IS_PRESENT | \
1765 PAGE_IS_SWAPPED | PAGE_IS_PFNZERO | \
1766 PAGE_IS_HUGE | PAGE_IS_SOFT_DIRTY)
1767 #define PM_SCAN_FLAGS (PM_SCAN_WP_MATCHING | PM_SCAN_CHECK_WPASYNC)
1769 struct pagemap_scan_private {
1770 struct pm_scan_arg arg;
1771 unsigned long masks_of_interest, cur_vma_category;
1772 struct page_region *vec_buf;
1773 unsigned long vec_buf_len, vec_buf_index, found_pages;
1774 struct page_region __user *vec_out;
1777 static unsigned long pagemap_page_category(struct pagemap_scan_private *p,
1778 struct vm_area_struct *vma,
1779 unsigned long addr, pte_t pte)
1781 unsigned long categories = 0;
1783 if (pte_present(pte)) {
1786 categories |= PAGE_IS_PRESENT;
1787 if (!pte_uffd_wp(pte))
1788 categories |= PAGE_IS_WRITTEN;
1790 if (p->masks_of_interest & PAGE_IS_FILE) {
1791 page = vm_normal_page(vma, addr, pte);
1792 if (page && !PageAnon(page))
1793 categories |= PAGE_IS_FILE;
1796 if (is_zero_pfn(pte_pfn(pte)))
1797 categories |= PAGE_IS_PFNZERO;
1798 if (pte_soft_dirty(pte))
1799 categories |= PAGE_IS_SOFT_DIRTY;
1800 } else if (is_swap_pte(pte)) {
1803 categories |= PAGE_IS_SWAPPED;
1804 if (!pte_swp_uffd_wp_any(pte))
1805 categories |= PAGE_IS_WRITTEN;
1807 if (p->masks_of_interest & PAGE_IS_FILE) {
1808 swp = pte_to_swp_entry(pte);
1809 if (is_pfn_swap_entry(swp) &&
1810 !PageAnon(pfn_swap_entry_to_page(swp)))
1811 categories |= PAGE_IS_FILE;
1813 if (pte_swp_soft_dirty(pte))
1814 categories |= PAGE_IS_SOFT_DIRTY;
1820 static void make_uffd_wp_pte(struct vm_area_struct *vma,
1821 unsigned long addr, pte_t *pte)
1823 pte_t ptent = ptep_get(pte);
1825 if (pte_present(ptent)) {
1828 old_pte = ptep_modify_prot_start(vma, addr, pte);
1829 ptent = pte_mkuffd_wp(ptent);
1830 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent);
1831 } else if (is_swap_pte(ptent)) {
1832 ptent = pte_swp_mkuffd_wp(ptent);
1833 set_pte_at(vma->vm_mm, addr, pte, ptent);
1835 set_pte_at(vma->vm_mm, addr, pte,
1836 make_pte_marker(PTE_MARKER_UFFD_WP));
1840 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1841 static unsigned long pagemap_thp_category(struct pagemap_scan_private *p,
1842 struct vm_area_struct *vma,
1843 unsigned long addr, pmd_t pmd)
1845 unsigned long categories = PAGE_IS_HUGE;
1847 if (pmd_present(pmd)) {
1850 categories |= PAGE_IS_PRESENT;
1851 if (!pmd_uffd_wp(pmd))
1852 categories |= PAGE_IS_WRITTEN;
1854 if (p->masks_of_interest & PAGE_IS_FILE) {
1855 page = vm_normal_page_pmd(vma, addr, pmd);
1856 if (page && !PageAnon(page))
1857 categories |= PAGE_IS_FILE;
1860 if (is_zero_pfn(pmd_pfn(pmd)))
1861 categories |= PAGE_IS_PFNZERO;
1862 if (pmd_soft_dirty(pmd))
1863 categories |= PAGE_IS_SOFT_DIRTY;
1864 } else if (is_swap_pmd(pmd)) {
1867 categories |= PAGE_IS_SWAPPED;
1868 if (!pmd_swp_uffd_wp(pmd))
1869 categories |= PAGE_IS_WRITTEN;
1870 if (pmd_swp_soft_dirty(pmd))
1871 categories |= PAGE_IS_SOFT_DIRTY;
1873 if (p->masks_of_interest & PAGE_IS_FILE) {
1874 swp = pmd_to_swp_entry(pmd);
1875 if (is_pfn_swap_entry(swp) &&
1876 !PageAnon(pfn_swap_entry_to_page(swp)))
1877 categories |= PAGE_IS_FILE;
1884 static void make_uffd_wp_pmd(struct vm_area_struct *vma,
1885 unsigned long addr, pmd_t *pmdp)
1887 pmd_t old, pmd = *pmdp;
1889 if (pmd_present(pmd)) {
1890 old = pmdp_invalidate_ad(vma, addr, pmdp);
1891 pmd = pmd_mkuffd_wp(old);
1892 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1893 } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1894 pmd = pmd_swp_mkuffd_wp(pmd);
1895 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1898 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1900 #ifdef CONFIG_HUGETLB_PAGE
1901 static unsigned long pagemap_hugetlb_category(pte_t pte)
1903 unsigned long categories = PAGE_IS_HUGE;
1906 * According to pagemap_hugetlb_range(), file-backed HugeTLB
1907 * page cannot be swapped. So PAGE_IS_FILE is not checked for
1910 if (pte_present(pte)) {
1911 categories |= PAGE_IS_PRESENT;
1912 if (!huge_pte_uffd_wp(pte))
1913 categories |= PAGE_IS_WRITTEN;
1914 if (!PageAnon(pte_page(pte)))
1915 categories |= PAGE_IS_FILE;
1916 if (is_zero_pfn(pte_pfn(pte)))
1917 categories |= PAGE_IS_PFNZERO;
1918 if (pte_soft_dirty(pte))
1919 categories |= PAGE_IS_SOFT_DIRTY;
1920 } else if (is_swap_pte(pte)) {
1921 categories |= PAGE_IS_SWAPPED;
1922 if (!pte_swp_uffd_wp_any(pte))
1923 categories |= PAGE_IS_WRITTEN;
1924 if (pte_swp_soft_dirty(pte))
1925 categories |= PAGE_IS_SOFT_DIRTY;
1931 static void make_uffd_wp_huge_pte(struct vm_area_struct *vma,
1932 unsigned long addr, pte_t *ptep,
1935 unsigned long psize;
1937 if (is_hugetlb_entry_hwpoisoned(ptent) || is_pte_marker(ptent))
1940 psize = huge_page_size(hstate_vma(vma));
1942 if (is_hugetlb_entry_migration(ptent))
1943 set_huge_pte_at(vma->vm_mm, addr, ptep,
1944 pte_swp_mkuffd_wp(ptent), psize);
1945 else if (!huge_pte_none(ptent))
1946 huge_ptep_modify_prot_commit(vma, addr, ptep, ptent,
1947 huge_pte_mkuffd_wp(ptent));
1949 set_huge_pte_at(vma->vm_mm, addr, ptep,
1950 make_pte_marker(PTE_MARKER_UFFD_WP), psize);
1952 #endif /* CONFIG_HUGETLB_PAGE */
1954 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLB_PAGE)
1955 static void pagemap_scan_backout_range(struct pagemap_scan_private *p,
1956 unsigned long addr, unsigned long end)
1958 struct page_region *cur_buf = &p->vec_buf[p->vec_buf_index];
1960 if (cur_buf->start != addr)
1961 cur_buf->end = addr;
1963 cur_buf->start = cur_buf->end = 0;
1965 p->found_pages -= (end - addr) / PAGE_SIZE;
1969 static bool pagemap_scan_is_interesting_page(unsigned long categories,
1970 const struct pagemap_scan_private *p)
1972 categories ^= p->arg.category_inverted;
1973 if ((categories & p->arg.category_mask) != p->arg.category_mask)
1975 if (p->arg.category_anyof_mask && !(categories & p->arg.category_anyof_mask))
1981 static bool pagemap_scan_is_interesting_vma(unsigned long categories,
1982 const struct pagemap_scan_private *p)
1984 unsigned long required = p->arg.category_mask & PAGE_IS_WPALLOWED;
1986 categories ^= p->arg.category_inverted;
1987 if ((categories & required) != required)
1993 static int pagemap_scan_test_walk(unsigned long start, unsigned long end,
1994 struct mm_walk *walk)
1996 struct pagemap_scan_private *p = walk->private;
1997 struct vm_area_struct *vma = walk->vma;
1998 unsigned long vma_category = 0;
1999 bool wp_allowed = userfaultfd_wp_async(vma) &&
2000 userfaultfd_wp_use_markers(vma);
2003 /* User requested explicit failure over wp-async capability */
2004 if (p->arg.flags & PM_SCAN_CHECK_WPASYNC)
2007 * User requires wr-protect, and allows silently skipping
2010 if (p->arg.flags & PM_SCAN_WP_MATCHING)
2013 * Then the request doesn't involve wr-protects at all,
2014 * fall through to the rest checks, and allow vma walk.
2018 if (vma->vm_flags & VM_PFNMAP)
2022 vma_category |= PAGE_IS_WPALLOWED;
2024 if (vma->vm_flags & VM_SOFTDIRTY)
2025 vma_category |= PAGE_IS_SOFT_DIRTY;
2027 if (!pagemap_scan_is_interesting_vma(vma_category, p))
2030 p->cur_vma_category = vma_category;
2035 static bool pagemap_scan_push_range(unsigned long categories,
2036 struct pagemap_scan_private *p,
2037 unsigned long addr, unsigned long end)
2039 struct page_region *cur_buf = &p->vec_buf[p->vec_buf_index];
2042 * When there is no output buffer provided at all, the sentinel values
2043 * won't match here. There is no other way for `cur_buf->end` to be
2044 * non-zero other than it being non-empty.
2046 if (addr == cur_buf->end && categories == cur_buf->categories) {
2052 if (p->vec_buf_index >= p->vec_buf_len - 1)
2055 cur_buf = &p->vec_buf[++p->vec_buf_index];
2058 cur_buf->start = addr;
2060 cur_buf->categories = categories;
2065 static int pagemap_scan_output(unsigned long categories,
2066 struct pagemap_scan_private *p,
2067 unsigned long addr, unsigned long *end)
2069 unsigned long n_pages, total_pages;
2075 categories &= p->arg.return_mask;
2077 n_pages = (*end - addr) / PAGE_SIZE;
2078 if (check_add_overflow(p->found_pages, n_pages, &total_pages) ||
2079 total_pages > p->arg.max_pages) {
2080 size_t n_too_much = total_pages - p->arg.max_pages;
2081 *end -= n_too_much * PAGE_SIZE;
2082 n_pages -= n_too_much;
2086 if (!pagemap_scan_push_range(categories, p, addr, *end)) {
2092 p->found_pages += n_pages;
2094 p->arg.walk_end = *end;
2099 static int pagemap_scan_thp_entry(pmd_t *pmd, unsigned long start,
2100 unsigned long end, struct mm_walk *walk)
2102 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2103 struct pagemap_scan_private *p = walk->private;
2104 struct vm_area_struct *vma = walk->vma;
2105 unsigned long categories;
2109 ptl = pmd_trans_huge_lock(pmd, vma);
2113 categories = p->cur_vma_category |
2114 pagemap_thp_category(p, vma, start, *pmd);
2116 if (!pagemap_scan_is_interesting_page(categories, p))
2119 ret = pagemap_scan_output(categories, p, start, &end);
2123 if (~p->arg.flags & PM_SCAN_WP_MATCHING)
2125 if (~categories & PAGE_IS_WRITTEN)
2129 * Break huge page into small pages if the WP operation
2130 * needs to be performed on a portion of the huge page.
2132 if (end != start + HPAGE_SIZE) {
2134 split_huge_pmd(vma, pmd, start);
2135 pagemap_scan_backout_range(p, start, end);
2136 /* Report as if there was no THP */
2140 make_uffd_wp_pmd(vma, start, pmd);
2141 flush_tlb_range(vma, start, end);
2145 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
2150 static int pagemap_scan_pmd_entry(pmd_t *pmd, unsigned long start,
2151 unsigned long end, struct mm_walk *walk)
2153 struct pagemap_scan_private *p = walk->private;
2154 struct vm_area_struct *vma = walk->vma;
2155 unsigned long addr, flush_end = 0;
2156 pte_t *pte, *start_pte;
2160 arch_enter_lazy_mmu_mode();
2162 ret = pagemap_scan_thp_entry(pmd, start, end, walk);
2163 if (ret != -ENOENT) {
2164 arch_leave_lazy_mmu_mode();
2169 start_pte = pte = pte_offset_map_lock(vma->vm_mm, pmd, start, &ptl);
2171 arch_leave_lazy_mmu_mode();
2172 walk->action = ACTION_AGAIN;
2176 if ((p->arg.flags & PM_SCAN_WP_MATCHING) && !p->vec_out) {
2177 /* Fast path for performing exclusive WP */
2178 for (addr = start; addr != end; pte++, addr += PAGE_SIZE) {
2179 if (pte_uffd_wp(ptep_get(pte)))
2181 make_uffd_wp_pte(vma, addr, pte);
2184 flush_end = addr + PAGE_SIZE;
2186 goto flush_and_return;
2189 if (!p->arg.category_anyof_mask && !p->arg.category_inverted &&
2190 p->arg.category_mask == PAGE_IS_WRITTEN &&
2191 p->arg.return_mask == PAGE_IS_WRITTEN) {
2192 for (addr = start; addr < end; pte++, addr += PAGE_SIZE) {
2193 unsigned long next = addr + PAGE_SIZE;
2195 if (pte_uffd_wp(ptep_get(pte)))
2197 ret = pagemap_scan_output(p->cur_vma_category | PAGE_IS_WRITTEN,
2201 if (~p->arg.flags & PM_SCAN_WP_MATCHING)
2203 make_uffd_wp_pte(vma, addr, pte);
2208 goto flush_and_return;
2211 for (addr = start; addr != end; pte++, addr += PAGE_SIZE) {
2212 unsigned long categories = p->cur_vma_category |
2213 pagemap_page_category(p, vma, addr, ptep_get(pte));
2214 unsigned long next = addr + PAGE_SIZE;
2216 if (!pagemap_scan_is_interesting_page(categories, p))
2219 ret = pagemap_scan_output(categories, p, addr, &next);
2223 if (~p->arg.flags & PM_SCAN_WP_MATCHING)
2225 if (~categories & PAGE_IS_WRITTEN)
2228 make_uffd_wp_pte(vma, addr, pte);
2236 flush_tlb_range(vma, start, addr);
2238 pte_unmap_unlock(start_pte, ptl);
2239 arch_leave_lazy_mmu_mode();
2245 #ifdef CONFIG_HUGETLB_PAGE
2246 static int pagemap_scan_hugetlb_entry(pte_t *ptep, unsigned long hmask,
2247 unsigned long start, unsigned long end,
2248 struct mm_walk *walk)
2250 struct pagemap_scan_private *p = walk->private;
2251 struct vm_area_struct *vma = walk->vma;
2252 unsigned long categories;
2257 if (~p->arg.flags & PM_SCAN_WP_MATCHING) {
2258 /* Go the short route when not write-protecting pages. */
2260 pte = huge_ptep_get(ptep);
2261 categories = p->cur_vma_category | pagemap_hugetlb_category(pte);
2263 if (!pagemap_scan_is_interesting_page(categories, p))
2266 return pagemap_scan_output(categories, p, start, &end);
2269 i_mmap_lock_write(vma->vm_file->f_mapping);
2270 ptl = huge_pte_lock(hstate_vma(vma), vma->vm_mm, ptep);
2272 pte = huge_ptep_get(ptep);
2273 categories = p->cur_vma_category | pagemap_hugetlb_category(pte);
2275 if (!pagemap_scan_is_interesting_page(categories, p))
2278 ret = pagemap_scan_output(categories, p, start, &end);
2282 if (~categories & PAGE_IS_WRITTEN)
2285 if (end != start + HPAGE_SIZE) {
2286 /* Partial HugeTLB page WP isn't possible. */
2287 pagemap_scan_backout_range(p, start, end);
2288 p->arg.walk_end = start;
2293 make_uffd_wp_huge_pte(vma, start, ptep, pte);
2294 flush_hugetlb_tlb_range(vma, start, end);
2298 i_mmap_unlock_write(vma->vm_file->f_mapping);
2303 #define pagemap_scan_hugetlb_entry NULL
2306 static int pagemap_scan_pte_hole(unsigned long addr, unsigned long end,
2307 int depth, struct mm_walk *walk)
2309 struct pagemap_scan_private *p = walk->private;
2310 struct vm_area_struct *vma = walk->vma;
2313 if (!vma || !pagemap_scan_is_interesting_page(p->cur_vma_category, p))
2316 ret = pagemap_scan_output(p->cur_vma_category, p, addr, &end);
2320 if (~p->arg.flags & PM_SCAN_WP_MATCHING)
2323 err = uffd_wp_range(vma, addr, end - addr, true);
2330 static const struct mm_walk_ops pagemap_scan_ops = {
2331 .test_walk = pagemap_scan_test_walk,
2332 .pmd_entry = pagemap_scan_pmd_entry,
2333 .pte_hole = pagemap_scan_pte_hole,
2334 .hugetlb_entry = pagemap_scan_hugetlb_entry,
2337 static int pagemap_scan_get_args(struct pm_scan_arg *arg,
2340 if (copy_from_user(arg, (void __user *)uarg, sizeof(*arg)))
2343 if (arg->size != sizeof(struct pm_scan_arg))
2346 /* Validate requested features */
2347 if (arg->flags & ~PM_SCAN_FLAGS)
2349 if ((arg->category_inverted | arg->category_mask |
2350 arg->category_anyof_mask | arg->return_mask) & ~PM_SCAN_CATEGORIES)
2353 arg->start = untagged_addr((unsigned long)arg->start);
2354 arg->end = untagged_addr((unsigned long)arg->end);
2355 arg->vec = untagged_addr((unsigned long)arg->vec);
2357 /* Validate memory pointers */
2358 if (!IS_ALIGNED(arg->start, PAGE_SIZE))
2360 if (!access_ok((void __user *)(long)arg->start, arg->end - arg->start))
2362 if (!arg->vec && arg->vec_len)
2364 if (arg->vec && !access_ok((void __user *)(long)arg->vec,
2365 arg->vec_len * sizeof(struct page_region)))
2368 /* Fixup default values */
2369 arg->end = ALIGN(arg->end, PAGE_SIZE);
2371 if (!arg->max_pages)
2372 arg->max_pages = ULONG_MAX;
2377 static int pagemap_scan_writeback_args(struct pm_scan_arg *arg,
2378 unsigned long uargl)
2380 struct pm_scan_arg __user *uarg = (void __user *)uargl;
2382 if (copy_to_user(&uarg->walk_end, &arg->walk_end, sizeof(arg->walk_end)))
2388 static int pagemap_scan_init_bounce_buffer(struct pagemap_scan_private *p)
2390 if (!p->arg.vec_len)
2393 p->vec_buf_len = min_t(size_t, PAGEMAP_WALK_SIZE >> PAGE_SHIFT,
2395 p->vec_buf = kmalloc_array(p->vec_buf_len, sizeof(*p->vec_buf),
2400 p->vec_buf->start = p->vec_buf->end = 0;
2401 p->vec_out = (struct page_region __user *)(long)p->arg.vec;
2406 static long pagemap_scan_flush_buffer(struct pagemap_scan_private *p)
2408 const struct page_region *buf = p->vec_buf;
2409 long n = p->vec_buf_index;
2414 if (buf[n].end != buf[n].start)
2420 if (copy_to_user(p->vec_out, buf, n * sizeof(*buf)))
2423 p->arg.vec_len -= n;
2426 p->vec_buf_index = 0;
2427 p->vec_buf_len = min_t(size_t, p->vec_buf_len, p->arg.vec_len);
2428 p->vec_buf->start = p->vec_buf->end = 0;
2433 static long do_pagemap_scan(struct mm_struct *mm, unsigned long uarg)
2435 struct pagemap_scan_private p = {0};
2436 unsigned long walk_start;
2437 size_t n_ranges_out = 0;
2440 ret = pagemap_scan_get_args(&p.arg, uarg);
2444 p.masks_of_interest = p.arg.category_mask | p.arg.category_anyof_mask |
2446 ret = pagemap_scan_init_bounce_buffer(&p);
2450 for (walk_start = p.arg.start; walk_start < p.arg.end;
2451 walk_start = p.arg.walk_end) {
2452 struct mmu_notifier_range range;
2455 if (fatal_signal_pending(current)) {
2460 ret = mmap_read_lock_killable(mm);
2464 /* Protection change for the range is going to happen. */
2465 if (p.arg.flags & PM_SCAN_WP_MATCHING) {
2466 mmu_notifier_range_init(&range, MMU_NOTIFY_PROTECTION_VMA, 0,
2467 mm, walk_start, p.arg.end);
2468 mmu_notifier_invalidate_range_start(&range);
2471 ret = walk_page_range(mm, walk_start, p.arg.end,
2472 &pagemap_scan_ops, &p);
2474 if (p.arg.flags & PM_SCAN_WP_MATCHING)
2475 mmu_notifier_invalidate_range_end(&range);
2477 mmap_read_unlock(mm);
2479 n_out = pagemap_scan_flush_buffer(&p);
2483 n_ranges_out += n_out;
2488 if (p.arg.vec_len == 0 || p.found_pages == p.arg.max_pages)
2492 /* ENOSPC signifies early stop (buffer full) from the walk. */
2493 if (!ret || ret == -ENOSPC)
2496 /* The walk_end isn't set when ret is zero */
2497 if (!p.arg.walk_end)
2498 p.arg.walk_end = p.arg.end;
2499 if (pagemap_scan_writeback_args(&p.arg, uarg))
2506 static long do_pagemap_cmd(struct file *file, unsigned int cmd,
2509 struct mm_struct *mm = file->private_data;
2513 return do_pagemap_scan(mm, arg);
2520 const struct file_operations proc_pagemap_operations = {
2521 .llseek = mem_lseek, /* borrow this */
2522 .read = pagemap_read,
2523 .open = pagemap_open,
2524 .release = pagemap_release,
2525 .unlocked_ioctl = do_pagemap_cmd,
2526 .compat_ioctl = do_pagemap_cmd,
2528 #endif /* CONFIG_PROC_PAGE_MONITOR */
2533 unsigned long pages;
2535 unsigned long active;
2536 unsigned long writeback;
2537 unsigned long mapcount_max;
2538 unsigned long dirty;
2539 unsigned long swapcache;
2540 unsigned long node[MAX_NUMNODES];
2543 struct numa_maps_private {
2544 struct proc_maps_private proc_maps;
2545 struct numa_maps md;
2548 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
2549 unsigned long nr_pages)
2551 int count = page_mapcount(page);
2553 md->pages += nr_pages;
2554 if (pte_dirty || PageDirty(page))
2555 md->dirty += nr_pages;
2557 if (PageSwapCache(page))
2558 md->swapcache += nr_pages;
2560 if (PageActive(page) || PageUnevictable(page))
2561 md->active += nr_pages;
2563 if (PageWriteback(page))
2564 md->writeback += nr_pages;
2567 md->anon += nr_pages;
2569 if (count > md->mapcount_max)
2570 md->mapcount_max = count;
2572 md->node[page_to_nid(page)] += nr_pages;
2575 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
2581 if (!pte_present(pte))
2584 page = vm_normal_page(vma, addr, pte);
2585 if (!page || is_zone_device_page(page))
2588 if (PageReserved(page))
2591 nid = page_to_nid(page);
2592 if (!node_isset(nid, node_states[N_MEMORY]))
2598 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2599 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
2600 struct vm_area_struct *vma,
2606 if (!pmd_present(pmd))
2609 page = vm_normal_page_pmd(vma, addr, pmd);
2613 if (PageReserved(page))
2616 nid = page_to_nid(page);
2617 if (!node_isset(nid, node_states[N_MEMORY]))
2624 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
2625 unsigned long end, struct mm_walk *walk)
2627 struct numa_maps *md = walk->private;
2628 struct vm_area_struct *vma = walk->vma;
2633 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2634 ptl = pmd_trans_huge_lock(pmd, vma);
2638 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
2640 gather_stats(page, md, pmd_dirty(*pmd),
2641 HPAGE_PMD_SIZE/PAGE_SIZE);
2646 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
2648 walk->action = ACTION_AGAIN;
2652 pte_t ptent = ptep_get(pte);
2653 struct page *page = can_gather_numa_stats(ptent, vma, addr);
2656 gather_stats(page, md, pte_dirty(ptent), 1);
2658 } while (pte++, addr += PAGE_SIZE, addr != end);
2659 pte_unmap_unlock(orig_pte, ptl);
2663 #ifdef CONFIG_HUGETLB_PAGE
2664 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
2665 unsigned long addr, unsigned long end, struct mm_walk *walk)
2667 pte_t huge_pte = huge_ptep_get(pte);
2668 struct numa_maps *md;
2671 if (!pte_present(huge_pte))
2674 page = pte_page(huge_pte);
2677 gather_stats(page, md, pte_dirty(huge_pte), 1);
2682 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
2683 unsigned long addr, unsigned long end, struct mm_walk *walk)
2689 static const struct mm_walk_ops show_numa_ops = {
2690 .hugetlb_entry = gather_hugetlb_stats,
2691 .pmd_entry = gather_pte_stats,
2692 .walk_lock = PGWALK_RDLOCK,
2696 * Display pages allocated per node and memory policy via /proc.
2698 static int show_numa_map(struct seq_file *m, void *v)
2700 struct numa_maps_private *numa_priv = m->private;
2701 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
2702 struct vm_area_struct *vma = v;
2703 struct numa_maps *md = &numa_priv->md;
2704 struct file *file = vma->vm_file;
2705 struct mm_struct *mm = vma->vm_mm;
2707 struct mempolicy *pol;
2714 /* Ensure we start with an empty set of numa_maps statistics. */
2715 memset(md, 0, sizeof(*md));
2717 pol = __get_vma_policy(vma, vma->vm_start, &ilx);
2719 mpol_to_str(buffer, sizeof(buffer), pol);
2722 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
2725 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
2728 seq_puts(m, " file=");
2729 seq_path(m, file_user_path(file), "\n\t= ");
2730 } else if (vma_is_initial_heap(vma)) {
2731 seq_puts(m, " heap");
2732 } else if (vma_is_initial_stack(vma)) {
2733 seq_puts(m, " stack");
2736 if (is_vm_hugetlb_page(vma))
2737 seq_puts(m, " huge");
2739 /* mmap_lock is held by m_start */
2740 walk_page_vma(vma, &show_numa_ops, md);
2746 seq_printf(m, " anon=%lu", md->anon);
2749 seq_printf(m, " dirty=%lu", md->dirty);
2751 if (md->pages != md->anon && md->pages != md->dirty)
2752 seq_printf(m, " mapped=%lu", md->pages);
2754 if (md->mapcount_max > 1)
2755 seq_printf(m, " mapmax=%lu", md->mapcount_max);
2758 seq_printf(m, " swapcache=%lu", md->swapcache);
2760 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
2761 seq_printf(m, " active=%lu", md->active);
2764 seq_printf(m, " writeback=%lu", md->writeback);
2766 for_each_node_state(nid, N_MEMORY)
2768 seq_printf(m, " N%d=%lu", nid, md->node[nid]);
2770 seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
2776 static const struct seq_operations proc_pid_numa_maps_op = {
2780 .show = show_numa_map,
2783 static int pid_numa_maps_open(struct inode *inode, struct file *file)
2785 return proc_maps_open(inode, file, &proc_pid_numa_maps_op,
2786 sizeof(struct numa_maps_private));
2789 const struct file_operations proc_pid_numa_maps_operations = {
2790 .open = pid_numa_maps_open,
2792 .llseek = seq_lseek,
2793 .release = proc_map_release,
2796 #endif /* CONFIG_NUMA */