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 anon_vma_name *anon_name = NULL;
278 struct mm_struct *mm = vma->vm_mm;
279 struct file *file = vma->vm_file;
280 vm_flags_t flags = vma->vm_flags;
281 unsigned long ino = 0;
282 unsigned long long pgoff = 0;
283 unsigned long start, end;
285 const char *name = NULL;
288 struct inode *inode = file_inode(vma->vm_file);
289 dev = inode->i_sb->s_dev;
291 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
294 start = vma->vm_start;
296 show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino);
298 anon_name = anon_vma_name(vma);
301 * Print the dentry name for named mappings, and a
302 * special [heap] marker for the heap:
307 * If user named this anon shared memory via
308 * prctl(PR_SET_VMA ..., use the provided name.
311 seq_printf(m, "[anon_shmem:%s]", anon_name->name);
313 seq_file_path(m, file, "\n");
317 if (vma->vm_ops && vma->vm_ops->name) {
318 name = vma->vm_ops->name(vma);
323 name = arch_vma_name(vma);
330 if (vma->vm_start <= mm->brk &&
331 vma->vm_end >= mm->start_brk) {
343 seq_printf(m, "[anon:%s]", anon_name->name);
355 static int show_map(struct seq_file *m, void *v)
361 static const struct seq_operations proc_pid_maps_op = {
368 static int pid_maps_open(struct inode *inode, struct file *file)
370 return do_maps_open(inode, file, &proc_pid_maps_op);
373 const struct file_operations proc_pid_maps_operations = {
374 .open = pid_maps_open,
377 .release = proc_map_release,
381 * Proportional Set Size(PSS): my share of RSS.
383 * PSS of a process is the count of pages it has in memory, where each
384 * page is divided by the number of processes sharing it. So if a
385 * process has 1000 pages all to itself, and 1000 shared with one other
386 * process, its PSS will be 1500.
388 * To keep (accumulated) division errors low, we adopt a 64bit
389 * fixed-point pss counter to minimize division errors. So (pss >>
390 * PSS_SHIFT) would be the real byte count.
392 * A shift of 12 before division means (assuming 4K page size):
393 * - 1M 3-user-pages add up to 8KB errors;
394 * - supports mapcount up to 2^24, or 16M;
395 * - supports PSS up to 2^52 bytes, or 4PB.
399 #ifdef CONFIG_PROC_PAGE_MONITOR
400 struct mem_size_stats {
401 unsigned long resident;
402 unsigned long shared_clean;
403 unsigned long shared_dirty;
404 unsigned long private_clean;
405 unsigned long private_dirty;
406 unsigned long referenced;
407 unsigned long anonymous;
408 unsigned long lazyfree;
409 unsigned long anonymous_thp;
410 unsigned long shmem_thp;
411 unsigned long file_thp;
413 unsigned long shared_hugetlb;
414 unsigned long private_hugetlb;
424 static void smaps_page_accumulate(struct mem_size_stats *mss,
425 struct page *page, unsigned long size, unsigned long pss,
426 bool dirty, bool locked, bool private)
431 mss->pss_anon += pss;
432 else if (PageSwapBacked(page))
433 mss->pss_shmem += pss;
435 mss->pss_file += pss;
438 mss->pss_locked += pss;
440 if (dirty || PageDirty(page)) {
441 mss->pss_dirty += pss;
443 mss->private_dirty += size;
445 mss->shared_dirty += size;
448 mss->private_clean += size;
450 mss->shared_clean += size;
454 static void smaps_account(struct mem_size_stats *mss, struct page *page,
455 bool compound, bool young, bool dirty, bool locked,
458 int i, nr = compound ? compound_nr(page) : 1;
459 unsigned long size = nr * PAGE_SIZE;
462 * First accumulate quantities that depend only on |size| and the type
463 * of the compound page.
465 if (PageAnon(page)) {
466 mss->anonymous += size;
467 if (!PageSwapBacked(page) && !dirty && !PageDirty(page))
468 mss->lazyfree += size;
471 mss->resident += size;
472 /* Accumulate the size in pages that have been accessed. */
473 if (young || page_is_young(page) || PageReferenced(page))
474 mss->referenced += size;
477 * Then accumulate quantities that may depend on sharing, or that may
478 * differ page-by-page.
480 * page_count(page) == 1 guarantees the page is mapped exactly once.
481 * If any subpage of the compound page mapped with PTE it would elevate
484 * The page_mapcount() is called to get a snapshot of the mapcount.
485 * Without holding the page lock this snapshot can be slightly wrong as
486 * we cannot always read the mapcount atomically. It is not safe to
487 * call page_mapcount() even with PTL held if the page is not mapped,
488 * especially for migration entries. Treat regular migration entries
491 if ((page_count(page) == 1) || migration) {
492 smaps_page_accumulate(mss, page, size, size << PSS_SHIFT, dirty,
496 for (i = 0; i < nr; i++, page++) {
497 int mapcount = page_mapcount(page);
498 unsigned long pss = PAGE_SIZE << PSS_SHIFT;
501 smaps_page_accumulate(mss, page, PAGE_SIZE, pss, dirty, locked,
507 static int smaps_pte_hole(unsigned long addr, unsigned long end,
508 __always_unused int depth, struct mm_walk *walk)
510 struct mem_size_stats *mss = walk->private;
511 struct vm_area_struct *vma = walk->vma;
513 mss->swap += shmem_partial_swap_usage(walk->vma->vm_file->f_mapping,
514 linear_page_index(vma, addr),
515 linear_page_index(vma, end));
520 #define smaps_pte_hole NULL
521 #endif /* CONFIG_SHMEM */
523 static void smaps_pte_hole_lookup(unsigned long addr, struct mm_walk *walk)
526 if (walk->ops->pte_hole) {
527 /* depth is not used */
528 smaps_pte_hole(addr, addr + PAGE_SIZE, 0, walk);
533 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
534 struct mm_walk *walk)
536 struct mem_size_stats *mss = walk->private;
537 struct vm_area_struct *vma = walk->vma;
538 bool locked = !!(vma->vm_flags & VM_LOCKED);
539 struct page *page = NULL;
540 bool migration = false, young = false, dirty = false;
542 if (pte_present(*pte)) {
543 page = vm_normal_page(vma, addr, *pte);
544 young = pte_young(*pte);
545 dirty = pte_dirty(*pte);
546 } else if (is_swap_pte(*pte)) {
547 swp_entry_t swpent = pte_to_swp_entry(*pte);
549 if (!non_swap_entry(swpent)) {
552 mss->swap += PAGE_SIZE;
553 mapcount = swp_swapcount(swpent);
555 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
557 do_div(pss_delta, mapcount);
558 mss->swap_pss += pss_delta;
560 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
562 } else if (is_pfn_swap_entry(swpent)) {
563 if (is_migration_entry(swpent))
565 page = pfn_swap_entry_to_page(swpent);
568 smaps_pte_hole_lookup(addr, walk);
575 smaps_account(mss, page, false, young, dirty, locked, migration);
578 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
579 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
580 struct mm_walk *walk)
582 struct mem_size_stats *mss = walk->private;
583 struct vm_area_struct *vma = walk->vma;
584 bool locked = !!(vma->vm_flags & VM_LOCKED);
585 struct page *page = NULL;
586 bool migration = false;
588 if (pmd_present(*pmd)) {
589 /* FOLL_DUMP will return -EFAULT on huge zero page */
590 page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
591 } else if (unlikely(thp_migration_supported() && is_swap_pmd(*pmd))) {
592 swp_entry_t entry = pmd_to_swp_entry(*pmd);
594 if (is_migration_entry(entry)) {
596 page = pfn_swap_entry_to_page(entry);
599 if (IS_ERR_OR_NULL(page))
602 mss->anonymous_thp += HPAGE_PMD_SIZE;
603 else if (PageSwapBacked(page))
604 mss->shmem_thp += HPAGE_PMD_SIZE;
605 else if (is_zone_device_page(page))
608 mss->file_thp += HPAGE_PMD_SIZE;
610 smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd),
614 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
615 struct mm_walk *walk)
620 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
621 struct mm_walk *walk)
623 struct vm_area_struct *vma = walk->vma;
627 ptl = pmd_trans_huge_lock(pmd, vma);
629 smaps_pmd_entry(pmd, addr, walk);
634 if (pmd_trans_unstable(pmd))
637 * The mmap_lock held all the way back in m_start() is what
638 * keeps khugepaged out of here and from collapsing things
641 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
642 for (; addr != end; pte++, addr += PAGE_SIZE)
643 smaps_pte_entry(pte, addr, walk);
644 pte_unmap_unlock(pte - 1, ptl);
650 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
653 * Don't forget to update Documentation/ on changes.
655 static const char mnemonics[BITS_PER_LONG][2] = {
657 * In case if we meet a flag we don't know about.
659 [0 ... (BITS_PER_LONG-1)] = "??",
661 [ilog2(VM_READ)] = "rd",
662 [ilog2(VM_WRITE)] = "wr",
663 [ilog2(VM_EXEC)] = "ex",
664 [ilog2(VM_SHARED)] = "sh",
665 [ilog2(VM_MAYREAD)] = "mr",
666 [ilog2(VM_MAYWRITE)] = "mw",
667 [ilog2(VM_MAYEXEC)] = "me",
668 [ilog2(VM_MAYSHARE)] = "ms",
669 [ilog2(VM_GROWSDOWN)] = "gd",
670 [ilog2(VM_PFNMAP)] = "pf",
671 [ilog2(VM_LOCKED)] = "lo",
672 [ilog2(VM_IO)] = "io",
673 [ilog2(VM_SEQ_READ)] = "sr",
674 [ilog2(VM_RAND_READ)] = "rr",
675 [ilog2(VM_DONTCOPY)] = "dc",
676 [ilog2(VM_DONTEXPAND)] = "de",
677 [ilog2(VM_LOCKONFAULT)] = "lf",
678 [ilog2(VM_ACCOUNT)] = "ac",
679 [ilog2(VM_NORESERVE)] = "nr",
680 [ilog2(VM_HUGETLB)] = "ht",
681 [ilog2(VM_SYNC)] = "sf",
682 [ilog2(VM_ARCH_1)] = "ar",
683 [ilog2(VM_WIPEONFORK)] = "wf",
684 [ilog2(VM_DONTDUMP)] = "dd",
685 #ifdef CONFIG_ARM64_BTI
686 [ilog2(VM_ARM64_BTI)] = "bt",
688 #ifdef CONFIG_MEM_SOFT_DIRTY
689 [ilog2(VM_SOFTDIRTY)] = "sd",
691 [ilog2(VM_MIXEDMAP)] = "mm",
692 [ilog2(VM_HUGEPAGE)] = "hg",
693 [ilog2(VM_NOHUGEPAGE)] = "nh",
694 [ilog2(VM_MERGEABLE)] = "mg",
695 [ilog2(VM_UFFD_MISSING)]= "um",
696 [ilog2(VM_UFFD_WP)] = "uw",
697 #ifdef CONFIG_ARM64_MTE
698 [ilog2(VM_MTE)] = "mt",
699 [ilog2(VM_MTE_ALLOWED)] = "",
701 #ifdef CONFIG_ARCH_HAS_PKEYS
702 /* These come out via ProtectionKey: */
703 [ilog2(VM_PKEY_BIT0)] = "",
704 [ilog2(VM_PKEY_BIT1)] = "",
705 [ilog2(VM_PKEY_BIT2)] = "",
706 [ilog2(VM_PKEY_BIT3)] = "",
708 [ilog2(VM_PKEY_BIT4)] = "",
710 #endif /* CONFIG_ARCH_HAS_PKEYS */
711 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_MINOR
712 [ilog2(VM_UFFD_MINOR)] = "ui",
713 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_MINOR */
717 seq_puts(m, "VmFlags: ");
718 for (i = 0; i < BITS_PER_LONG; i++) {
719 if (!mnemonics[i][0])
721 if (vma->vm_flags & (1UL << i)) {
722 seq_putc(m, mnemonics[i][0]);
723 seq_putc(m, mnemonics[i][1]);
730 #ifdef CONFIG_HUGETLB_PAGE
731 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
732 unsigned long addr, unsigned long end,
733 struct mm_walk *walk)
735 struct mem_size_stats *mss = walk->private;
736 struct vm_area_struct *vma = walk->vma;
737 struct page *page = NULL;
739 if (pte_present(*pte)) {
740 page = vm_normal_page(vma, addr, *pte);
741 } else if (is_swap_pte(*pte)) {
742 swp_entry_t swpent = pte_to_swp_entry(*pte);
744 if (is_pfn_swap_entry(swpent))
745 page = pfn_swap_entry_to_page(swpent);
748 if (page_mapcount(page) >= 2 || hugetlb_pmd_shared(pte))
749 mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
751 mss->private_hugetlb += huge_page_size(hstate_vma(vma));
756 #define smaps_hugetlb_range NULL
757 #endif /* HUGETLB_PAGE */
759 static const struct mm_walk_ops smaps_walk_ops = {
760 .pmd_entry = smaps_pte_range,
761 .hugetlb_entry = smaps_hugetlb_range,
764 static const struct mm_walk_ops smaps_shmem_walk_ops = {
765 .pmd_entry = smaps_pte_range,
766 .hugetlb_entry = smaps_hugetlb_range,
767 .pte_hole = smaps_pte_hole,
771 * Gather mem stats from @vma with the indicated beginning
772 * address @start, and keep them in @mss.
774 * Use vm_start of @vma as the beginning address if @start is 0.
776 static void smap_gather_stats(struct vm_area_struct *vma,
777 struct mem_size_stats *mss, unsigned long start)
779 const struct mm_walk_ops *ops = &smaps_walk_ops;
782 if (start >= vma->vm_end)
786 if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
788 * For shared or readonly shmem mappings we know that all
789 * swapped out pages belong to the shmem object, and we can
790 * obtain the swap value much more efficiently. For private
791 * writable mappings, we might have COW pages that are
792 * not affected by the parent swapped out pages of the shmem
793 * object, so we have to distinguish them during the page walk.
794 * Unless we know that the shmem object (or the part mapped by
795 * our VMA) has no swapped out pages at all.
797 unsigned long shmem_swapped = shmem_swap_usage(vma);
799 if (!start && (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
800 !(vma->vm_flags & VM_WRITE))) {
801 mss->swap += shmem_swapped;
803 ops = &smaps_shmem_walk_ops;
807 /* mmap_lock is held in m_start */
809 walk_page_vma(vma, ops, mss);
811 walk_page_range(vma->vm_mm, start, vma->vm_end, ops, mss);
814 #define SEQ_PUT_DEC(str, val) \
815 seq_put_decimal_ull_width(m, str, (val) >> 10, 8)
817 /* Show the contents common for smaps and smaps_rollup */
818 static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss,
821 SEQ_PUT_DEC("Rss: ", mss->resident);
822 SEQ_PUT_DEC(" kB\nPss: ", mss->pss >> PSS_SHIFT);
823 SEQ_PUT_DEC(" kB\nPss_Dirty: ", mss->pss_dirty >> PSS_SHIFT);
826 * These are meaningful only for smaps_rollup, otherwise two of
827 * them are zero, and the other one is the same as Pss.
829 SEQ_PUT_DEC(" kB\nPss_Anon: ",
830 mss->pss_anon >> PSS_SHIFT);
831 SEQ_PUT_DEC(" kB\nPss_File: ",
832 mss->pss_file >> PSS_SHIFT);
833 SEQ_PUT_DEC(" kB\nPss_Shmem: ",
834 mss->pss_shmem >> PSS_SHIFT);
836 SEQ_PUT_DEC(" kB\nShared_Clean: ", mss->shared_clean);
837 SEQ_PUT_DEC(" kB\nShared_Dirty: ", mss->shared_dirty);
838 SEQ_PUT_DEC(" kB\nPrivate_Clean: ", mss->private_clean);
839 SEQ_PUT_DEC(" kB\nPrivate_Dirty: ", mss->private_dirty);
840 SEQ_PUT_DEC(" kB\nReferenced: ", mss->referenced);
841 SEQ_PUT_DEC(" kB\nAnonymous: ", mss->anonymous);
842 SEQ_PUT_DEC(" kB\nLazyFree: ", mss->lazyfree);
843 SEQ_PUT_DEC(" kB\nAnonHugePages: ", mss->anonymous_thp);
844 SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp);
845 SEQ_PUT_DEC(" kB\nFilePmdMapped: ", mss->file_thp);
846 SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb);
847 seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ",
848 mss->private_hugetlb >> 10, 7);
849 SEQ_PUT_DEC(" kB\nSwap: ", mss->swap);
850 SEQ_PUT_DEC(" kB\nSwapPss: ",
851 mss->swap_pss >> PSS_SHIFT);
852 SEQ_PUT_DEC(" kB\nLocked: ",
853 mss->pss_locked >> PSS_SHIFT);
854 seq_puts(m, " kB\n");
857 static int show_smap(struct seq_file *m, void *v)
859 struct vm_area_struct *vma = v;
860 struct mem_size_stats mss;
862 memset(&mss, 0, sizeof(mss));
864 smap_gather_stats(vma, &mss, 0);
866 show_map_vma(m, vma);
868 SEQ_PUT_DEC("Size: ", vma->vm_end - vma->vm_start);
869 SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma));
870 SEQ_PUT_DEC(" kB\nMMUPageSize: ", vma_mmu_pagesize(vma));
871 seq_puts(m, " kB\n");
873 __show_smap(m, &mss, false);
875 seq_printf(m, "THPeligible: %d\n",
876 hugepage_vma_check(vma, vma->vm_flags, true, false, true));
878 if (arch_pkeys_enabled())
879 seq_printf(m, "ProtectionKey: %8u\n", vma_pkey(vma));
880 show_smap_vma_flags(m, vma);
885 static int show_smaps_rollup(struct seq_file *m, void *v)
887 struct proc_maps_private *priv = m->private;
888 struct mem_size_stats mss;
889 struct mm_struct *mm = priv->mm;
890 struct vm_area_struct *vma;
891 unsigned long vma_start = 0, last_vma_end = 0;
893 VMA_ITERATOR(vmi, mm, 0);
895 priv->task = get_proc_task(priv->inode);
899 if (!mm || !mmget_not_zero(mm)) {
904 memset(&mss, 0, sizeof(mss));
906 ret = mmap_read_lock_killable(mm);
910 hold_task_mempolicy(priv);
911 vma = vma_next(&vmi);
916 vma_start = vma->vm_start;
918 smap_gather_stats(vma, &mss, 0);
919 last_vma_end = vma->vm_end;
922 * Release mmap_lock temporarily if someone wants to
923 * access it for write request.
925 if (mmap_lock_is_contended(mm)) {
926 vma_iter_invalidate(&vmi);
927 mmap_read_unlock(mm);
928 ret = mmap_read_lock_killable(mm);
930 release_task_mempolicy(priv);
935 * After dropping the lock, there are four cases to
936 * consider. See the following example for explanation.
938 * +------+------+-----------+
939 * | VMA1 | VMA2 | VMA3 |
940 * +------+------+-----------+
944 * Suppose we drop the lock after reading VMA2 due to
945 * contention, then we get:
949 * 1) VMA2 is freed, but VMA3 exists:
951 * vma_next(vmi) will return VMA3.
952 * In this case, just continue from VMA3.
954 * 2) VMA2 still exists:
956 * vma_next(vmi) will return VMA3.
957 * In this case, just continue from VMA3.
959 * 3) No more VMAs can be found:
961 * vma_next(vmi) will return NULL.
962 * No more things to do, just break.
964 * 4) (last_vma_end - 1) is the middle of a vma (VMA'):
966 * vma_next(vmi) will return VMA' whose range
967 * contains last_vma_end.
968 * Iterate VMA' from last_vma_end.
970 vma = vma_next(&vmi);
975 /* Case 1 and 2 above */
976 if (vma->vm_start >= last_vma_end)
980 if (vma->vm_end > last_vma_end)
981 smap_gather_stats(vma, &mss, last_vma_end);
983 } for_each_vma(vmi, vma);
986 show_vma_header_prefix(m, vma_start, last_vma_end, 0, 0, 0, 0);
988 seq_puts(m, "[rollup]\n");
990 __show_smap(m, &mss, true);
992 release_task_mempolicy(priv);
993 mmap_read_unlock(mm);
998 put_task_struct(priv->task);
1005 static const struct seq_operations proc_pid_smaps_op = {
1012 static int pid_smaps_open(struct inode *inode, struct file *file)
1014 return do_maps_open(inode, file, &proc_pid_smaps_op);
1017 static int smaps_rollup_open(struct inode *inode, struct file *file)
1020 struct proc_maps_private *priv;
1022 priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT);
1026 ret = single_open(file, show_smaps_rollup, priv);
1030 priv->inode = inode;
1031 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
1032 if (IS_ERR(priv->mm)) {
1033 ret = PTR_ERR(priv->mm);
1035 single_release(inode, file);
1046 static int smaps_rollup_release(struct inode *inode, struct file *file)
1048 struct seq_file *seq = file->private_data;
1049 struct proc_maps_private *priv = seq->private;
1055 return single_release(inode, file);
1058 const struct file_operations proc_pid_smaps_operations = {
1059 .open = pid_smaps_open,
1061 .llseek = seq_lseek,
1062 .release = proc_map_release,
1065 const struct file_operations proc_pid_smaps_rollup_operations = {
1066 .open = smaps_rollup_open,
1068 .llseek = seq_lseek,
1069 .release = smaps_rollup_release,
1072 enum clear_refs_types {
1076 CLEAR_REFS_SOFT_DIRTY,
1077 CLEAR_REFS_MM_HIWATER_RSS,
1081 struct clear_refs_private {
1082 enum clear_refs_types type;
1085 #ifdef CONFIG_MEM_SOFT_DIRTY
1087 static inline bool pte_is_pinned(struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1091 if (!pte_write(pte))
1093 if (!is_cow_mapping(vma->vm_flags))
1095 if (likely(!test_bit(MMF_HAS_PINNED, &vma->vm_mm->flags)))
1097 page = vm_normal_page(vma, addr, pte);
1100 return page_maybe_dma_pinned(page);
1103 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1104 unsigned long addr, pte_t *pte)
1107 * The soft-dirty tracker uses #PF-s to catch writes
1108 * to pages, so write-protect the pte as well. See the
1109 * Documentation/admin-guide/mm/soft-dirty.rst for full description
1110 * of how soft-dirty works.
1114 if (pte_present(ptent)) {
1117 if (pte_is_pinned(vma, addr, ptent))
1119 old_pte = ptep_modify_prot_start(vma, addr, pte);
1120 ptent = pte_wrprotect(old_pte);
1121 ptent = pte_clear_soft_dirty(ptent);
1122 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent);
1123 } else if (is_swap_pte(ptent)) {
1124 ptent = pte_swp_clear_soft_dirty(ptent);
1125 set_pte_at(vma->vm_mm, addr, pte, ptent);
1129 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1130 unsigned long addr, pte_t *pte)
1135 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
1136 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1137 unsigned long addr, pmd_t *pmdp)
1139 pmd_t old, pmd = *pmdp;
1141 if (pmd_present(pmd)) {
1142 /* See comment in change_huge_pmd() */
1143 old = pmdp_invalidate(vma, addr, pmdp);
1145 pmd = pmd_mkdirty(pmd);
1147 pmd = pmd_mkyoung(pmd);
1149 pmd = pmd_wrprotect(pmd);
1150 pmd = pmd_clear_soft_dirty(pmd);
1152 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1153 } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1154 pmd = pmd_swp_clear_soft_dirty(pmd);
1155 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1159 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1160 unsigned long addr, pmd_t *pmdp)
1165 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
1166 unsigned long end, struct mm_walk *walk)
1168 struct clear_refs_private *cp = walk->private;
1169 struct vm_area_struct *vma = walk->vma;
1174 ptl = pmd_trans_huge_lock(pmd, vma);
1176 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1177 clear_soft_dirty_pmd(vma, addr, pmd);
1181 if (!pmd_present(*pmd))
1184 page = pmd_page(*pmd);
1186 /* Clear accessed and referenced bits. */
1187 pmdp_test_and_clear_young(vma, addr, pmd);
1188 test_and_clear_page_young(page);
1189 ClearPageReferenced(page);
1195 if (pmd_trans_unstable(pmd))
1198 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1199 for (; addr != end; pte++, addr += PAGE_SIZE) {
1202 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1203 clear_soft_dirty(vma, addr, pte);
1207 if (!pte_present(ptent))
1210 page = vm_normal_page(vma, addr, ptent);
1214 /* Clear accessed and referenced bits. */
1215 ptep_test_and_clear_young(vma, addr, pte);
1216 test_and_clear_page_young(page);
1217 ClearPageReferenced(page);
1219 pte_unmap_unlock(pte - 1, ptl);
1224 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1225 struct mm_walk *walk)
1227 struct clear_refs_private *cp = walk->private;
1228 struct vm_area_struct *vma = walk->vma;
1230 if (vma->vm_flags & VM_PFNMAP)
1234 * Writing 1 to /proc/pid/clear_refs affects all pages.
1235 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1236 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1237 * Writing 4 to /proc/pid/clear_refs affects all pages.
1239 if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1241 if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1246 static const struct mm_walk_ops clear_refs_walk_ops = {
1247 .pmd_entry = clear_refs_pte_range,
1248 .test_walk = clear_refs_test_walk,
1251 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1252 size_t count, loff_t *ppos)
1254 struct task_struct *task;
1255 char buffer[PROC_NUMBUF];
1256 struct mm_struct *mm;
1257 struct vm_area_struct *vma;
1258 enum clear_refs_types type;
1262 memset(buffer, 0, sizeof(buffer));
1263 if (count > sizeof(buffer) - 1)
1264 count = sizeof(buffer) - 1;
1265 if (copy_from_user(buffer, buf, count))
1267 rv = kstrtoint(strstrip(buffer), 10, &itype);
1270 type = (enum clear_refs_types)itype;
1271 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1274 task = get_proc_task(file_inode(file));
1277 mm = get_task_mm(task);
1279 VMA_ITERATOR(vmi, mm, 0);
1280 struct mmu_notifier_range range;
1281 struct clear_refs_private cp = {
1285 if (mmap_write_lock_killable(mm)) {
1289 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1291 * Writing 5 to /proc/pid/clear_refs resets the peak
1292 * resident set size to this mm's current rss value.
1294 reset_mm_hiwater_rss(mm);
1298 if (type == CLEAR_REFS_SOFT_DIRTY) {
1299 for_each_vma(vmi, vma) {
1300 if (!(vma->vm_flags & VM_SOFTDIRTY))
1302 vm_flags_clear(vma, VM_SOFTDIRTY);
1303 vma_set_page_prot(vma);
1306 inc_tlb_flush_pending(mm);
1307 mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY,
1309 mmu_notifier_invalidate_range_start(&range);
1311 walk_page_range(mm, 0, -1, &clear_refs_walk_ops, &cp);
1312 if (type == CLEAR_REFS_SOFT_DIRTY) {
1313 mmu_notifier_invalidate_range_end(&range);
1315 dec_tlb_flush_pending(mm);
1318 mmap_write_unlock(mm);
1322 put_task_struct(task);
1327 const struct file_operations proc_clear_refs_operations = {
1328 .write = clear_refs_write,
1329 .llseek = noop_llseek,
1336 struct pagemapread {
1337 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */
1338 pagemap_entry_t *buffer;
1342 #define PAGEMAP_WALK_SIZE (PMD_SIZE)
1343 #define PAGEMAP_WALK_MASK (PMD_MASK)
1345 #define PM_ENTRY_BYTES sizeof(pagemap_entry_t)
1346 #define PM_PFRAME_BITS 55
1347 #define PM_PFRAME_MASK GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1348 #define PM_SOFT_DIRTY BIT_ULL(55)
1349 #define PM_MMAP_EXCLUSIVE BIT_ULL(56)
1350 #define PM_UFFD_WP BIT_ULL(57)
1351 #define PM_FILE BIT_ULL(61)
1352 #define PM_SWAP BIT_ULL(62)
1353 #define PM_PRESENT BIT_ULL(63)
1355 #define PM_END_OF_BUFFER 1
1357 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1359 return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1362 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1363 struct pagemapread *pm)
1365 pm->buffer[pm->pos++] = *pme;
1366 if (pm->pos >= pm->len)
1367 return PM_END_OF_BUFFER;
1371 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1372 __always_unused int depth, struct mm_walk *walk)
1374 struct pagemapread *pm = walk->private;
1375 unsigned long addr = start;
1378 while (addr < end) {
1379 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1380 pagemap_entry_t pme = make_pme(0, 0);
1381 /* End of address space hole, which we mark as non-present. */
1382 unsigned long hole_end;
1385 hole_end = min(end, vma->vm_start);
1389 for (; addr < hole_end; addr += PAGE_SIZE) {
1390 err = add_to_pagemap(addr, &pme, pm);
1398 /* Addresses in the VMA. */
1399 if (vma->vm_flags & VM_SOFTDIRTY)
1400 pme = make_pme(0, PM_SOFT_DIRTY);
1401 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1402 err = add_to_pagemap(addr, &pme, pm);
1411 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1412 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1414 u64 frame = 0, flags = 0;
1415 struct page *page = NULL;
1416 bool migration = false;
1418 if (pte_present(pte)) {
1420 frame = pte_pfn(pte);
1421 flags |= PM_PRESENT;
1422 page = vm_normal_page(vma, addr, pte);
1423 if (pte_soft_dirty(pte))
1424 flags |= PM_SOFT_DIRTY;
1425 if (pte_uffd_wp(pte))
1426 flags |= PM_UFFD_WP;
1427 } else if (is_swap_pte(pte)) {
1429 if (pte_swp_soft_dirty(pte))
1430 flags |= PM_SOFT_DIRTY;
1431 if (pte_swp_uffd_wp(pte))
1432 flags |= PM_UFFD_WP;
1433 entry = pte_to_swp_entry(pte);
1437 * For PFN swap offsets, keeping the offset field
1438 * to be PFN only to be compatible with old smaps.
1440 if (is_pfn_swap_entry(entry))
1441 offset = swp_offset_pfn(entry);
1443 offset = swp_offset(entry);
1444 frame = swp_type(entry) |
1445 (offset << MAX_SWAPFILES_SHIFT);
1448 migration = is_migration_entry(entry);
1449 if (is_pfn_swap_entry(entry))
1450 page = pfn_swap_entry_to_page(entry);
1451 if (pte_marker_entry_uffd_wp(entry))
1452 flags |= PM_UFFD_WP;
1455 if (page && !PageAnon(page))
1457 if (page && !migration && page_mapcount(page) == 1)
1458 flags |= PM_MMAP_EXCLUSIVE;
1459 if (vma->vm_flags & VM_SOFTDIRTY)
1460 flags |= PM_SOFT_DIRTY;
1462 return make_pme(frame, flags);
1465 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1466 struct mm_walk *walk)
1468 struct vm_area_struct *vma = walk->vma;
1469 struct pagemapread *pm = walk->private;
1471 pte_t *pte, *orig_pte;
1473 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1474 bool migration = false;
1476 ptl = pmd_trans_huge_lock(pmdp, vma);
1478 u64 flags = 0, frame = 0;
1480 struct page *page = NULL;
1482 if (vma->vm_flags & VM_SOFTDIRTY)
1483 flags |= PM_SOFT_DIRTY;
1485 if (pmd_present(pmd)) {
1486 page = pmd_page(pmd);
1488 flags |= PM_PRESENT;
1489 if (pmd_soft_dirty(pmd))
1490 flags |= PM_SOFT_DIRTY;
1491 if (pmd_uffd_wp(pmd))
1492 flags |= PM_UFFD_WP;
1494 frame = pmd_pfn(pmd) +
1495 ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1497 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1498 else if (is_swap_pmd(pmd)) {
1499 swp_entry_t entry = pmd_to_swp_entry(pmd);
1500 unsigned long offset;
1503 if (is_pfn_swap_entry(entry))
1504 offset = swp_offset_pfn(entry);
1506 offset = swp_offset(entry);
1508 ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1509 frame = swp_type(entry) |
1510 (offset << MAX_SWAPFILES_SHIFT);
1513 if (pmd_swp_soft_dirty(pmd))
1514 flags |= PM_SOFT_DIRTY;
1515 if (pmd_swp_uffd_wp(pmd))
1516 flags |= PM_UFFD_WP;
1517 VM_BUG_ON(!is_pmd_migration_entry(pmd));
1518 migration = is_migration_entry(entry);
1519 page = pfn_swap_entry_to_page(entry);
1523 if (page && !migration && page_mapcount(page) == 1)
1524 flags |= PM_MMAP_EXCLUSIVE;
1526 for (; addr != end; addr += PAGE_SIZE) {
1527 pagemap_entry_t pme = make_pme(frame, flags);
1529 err = add_to_pagemap(addr, &pme, pm);
1533 if (flags & PM_PRESENT)
1535 else if (flags & PM_SWAP)
1536 frame += (1 << MAX_SWAPFILES_SHIFT);
1543 if (pmd_trans_unstable(pmdp))
1545 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1548 * We can assume that @vma always points to a valid one and @end never
1549 * goes beyond vma->vm_end.
1551 orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1552 for (; addr < end; pte++, addr += PAGE_SIZE) {
1553 pagemap_entry_t pme;
1555 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1556 err = add_to_pagemap(addr, &pme, pm);
1560 pte_unmap_unlock(orig_pte, ptl);
1567 #ifdef CONFIG_HUGETLB_PAGE
1568 /* This function walks within one hugetlb entry in the single call */
1569 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1570 unsigned long addr, unsigned long end,
1571 struct mm_walk *walk)
1573 struct pagemapread *pm = walk->private;
1574 struct vm_area_struct *vma = walk->vma;
1575 u64 flags = 0, frame = 0;
1579 if (vma->vm_flags & VM_SOFTDIRTY)
1580 flags |= PM_SOFT_DIRTY;
1582 pte = huge_ptep_get(ptep);
1583 if (pte_present(pte)) {
1584 struct page *page = pte_page(pte);
1586 if (!PageAnon(page))
1589 if (page_mapcount(page) == 1)
1590 flags |= PM_MMAP_EXCLUSIVE;
1592 if (huge_pte_uffd_wp(pte))
1593 flags |= PM_UFFD_WP;
1595 flags |= PM_PRESENT;
1597 frame = pte_pfn(pte) +
1598 ((addr & ~hmask) >> PAGE_SHIFT);
1599 } else if (pte_swp_uffd_wp_any(pte)) {
1600 flags |= PM_UFFD_WP;
1603 for (; addr != end; addr += PAGE_SIZE) {
1604 pagemap_entry_t pme = make_pme(frame, flags);
1606 err = add_to_pagemap(addr, &pme, pm);
1609 if (pm->show_pfn && (flags & PM_PRESENT))
1618 #define pagemap_hugetlb_range NULL
1619 #endif /* HUGETLB_PAGE */
1621 static const struct mm_walk_ops pagemap_ops = {
1622 .pmd_entry = pagemap_pmd_range,
1623 .pte_hole = pagemap_pte_hole,
1624 .hugetlb_entry = pagemap_hugetlb_range,
1628 * /proc/pid/pagemap - an array mapping virtual pages to pfns
1630 * For each page in the address space, this file contains one 64-bit entry
1631 * consisting of the following:
1633 * Bits 0-54 page frame number (PFN) if present
1634 * Bits 0-4 swap type if swapped
1635 * Bits 5-54 swap offset if swapped
1636 * Bit 55 pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst)
1637 * Bit 56 page exclusively mapped
1638 * Bit 57 pte is uffd-wp write-protected
1640 * Bit 61 page is file-page or shared-anon
1641 * Bit 62 page swapped
1642 * Bit 63 page present
1644 * If the page is not present but in swap, then the PFN contains an
1645 * encoding of the swap file number and the page's offset into the
1646 * swap. Unmapped pages return a null PFN. This allows determining
1647 * precisely which pages are mapped (or in swap) and comparing mapped
1648 * pages between processes.
1650 * Efficient users of this interface will use /proc/pid/maps to
1651 * determine which areas of memory are actually mapped and llseek to
1652 * skip over unmapped regions.
1654 static ssize_t pagemap_read(struct file *file, char __user *buf,
1655 size_t count, loff_t *ppos)
1657 struct mm_struct *mm = file->private_data;
1658 struct pagemapread pm;
1660 unsigned long svpfn;
1661 unsigned long start_vaddr;
1662 unsigned long end_vaddr;
1663 int ret = 0, copied = 0;
1665 if (!mm || !mmget_not_zero(mm))
1669 /* file position must be aligned */
1670 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1677 /* do not disclose physical addresses: attack vector */
1678 pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1680 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1681 pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL);
1687 svpfn = src / PM_ENTRY_BYTES;
1688 end_vaddr = mm->task_size;
1690 /* watch out for wraparound */
1691 start_vaddr = end_vaddr;
1692 if (svpfn <= (ULONG_MAX >> PAGE_SHIFT))
1693 start_vaddr = untagged_addr(svpfn << PAGE_SHIFT);
1695 /* Ensure the address is inside the task */
1696 if (start_vaddr > mm->task_size)
1697 start_vaddr = end_vaddr;
1700 * The odds are that this will stop walking way
1701 * before end_vaddr, because the length of the
1702 * user buffer is tracked in "pm", and the walk
1703 * will stop when we hit the end of the buffer.
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 const struct file_operations proc_pagemap_operations = {
1764 .llseek = mem_lseek, /* borrow this */
1765 .read = pagemap_read,
1766 .open = pagemap_open,
1767 .release = pagemap_release,
1769 #endif /* CONFIG_PROC_PAGE_MONITOR */
1774 unsigned long pages;
1776 unsigned long active;
1777 unsigned long writeback;
1778 unsigned long mapcount_max;
1779 unsigned long dirty;
1780 unsigned long swapcache;
1781 unsigned long node[MAX_NUMNODES];
1784 struct numa_maps_private {
1785 struct proc_maps_private proc_maps;
1786 struct numa_maps md;
1789 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1790 unsigned long nr_pages)
1792 int count = page_mapcount(page);
1794 md->pages += nr_pages;
1795 if (pte_dirty || PageDirty(page))
1796 md->dirty += nr_pages;
1798 if (PageSwapCache(page))
1799 md->swapcache += nr_pages;
1801 if (PageActive(page) || PageUnevictable(page))
1802 md->active += nr_pages;
1804 if (PageWriteback(page))
1805 md->writeback += nr_pages;
1808 md->anon += nr_pages;
1810 if (count > md->mapcount_max)
1811 md->mapcount_max = count;
1813 md->node[page_to_nid(page)] += nr_pages;
1816 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1822 if (!pte_present(pte))
1825 page = vm_normal_page(vma, addr, pte);
1826 if (!page || is_zone_device_page(page))
1829 if (PageReserved(page))
1832 nid = page_to_nid(page);
1833 if (!node_isset(nid, node_states[N_MEMORY]))
1839 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1840 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1841 struct vm_area_struct *vma,
1847 if (!pmd_present(pmd))
1850 page = vm_normal_page_pmd(vma, addr, pmd);
1854 if (PageReserved(page))
1857 nid = page_to_nid(page);
1858 if (!node_isset(nid, node_states[N_MEMORY]))
1865 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1866 unsigned long end, struct mm_walk *walk)
1868 struct numa_maps *md = walk->private;
1869 struct vm_area_struct *vma = walk->vma;
1874 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1875 ptl = pmd_trans_huge_lock(pmd, vma);
1879 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1881 gather_stats(page, md, pmd_dirty(*pmd),
1882 HPAGE_PMD_SIZE/PAGE_SIZE);
1887 if (pmd_trans_unstable(pmd))
1890 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1892 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1895 gather_stats(page, md, pte_dirty(*pte), 1);
1897 } while (pte++, addr += PAGE_SIZE, addr != end);
1898 pte_unmap_unlock(orig_pte, ptl);
1902 #ifdef CONFIG_HUGETLB_PAGE
1903 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1904 unsigned long addr, unsigned long end, struct mm_walk *walk)
1906 pte_t huge_pte = huge_ptep_get(pte);
1907 struct numa_maps *md;
1910 if (!pte_present(huge_pte))
1913 page = pte_page(huge_pte);
1916 gather_stats(page, md, pte_dirty(huge_pte), 1);
1921 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1922 unsigned long addr, unsigned long end, struct mm_walk *walk)
1928 static const struct mm_walk_ops show_numa_ops = {
1929 .hugetlb_entry = gather_hugetlb_stats,
1930 .pmd_entry = gather_pte_stats,
1934 * Display pages allocated per node and memory policy via /proc.
1936 static int show_numa_map(struct seq_file *m, void *v)
1938 struct numa_maps_private *numa_priv = m->private;
1939 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1940 struct vm_area_struct *vma = v;
1941 struct numa_maps *md = &numa_priv->md;
1942 struct file *file = vma->vm_file;
1943 struct mm_struct *mm = vma->vm_mm;
1944 struct mempolicy *pol;
1951 /* Ensure we start with an empty set of numa_maps statistics. */
1952 memset(md, 0, sizeof(*md));
1954 pol = __get_vma_policy(vma, vma->vm_start);
1956 mpol_to_str(buffer, sizeof(buffer), pol);
1959 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1962 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1965 seq_puts(m, " file=");
1966 seq_file_path(m, file, "\n\t= ");
1967 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1968 seq_puts(m, " heap");
1969 } else if (is_stack(vma)) {
1970 seq_puts(m, " stack");
1973 if (is_vm_hugetlb_page(vma))
1974 seq_puts(m, " huge");
1976 /* mmap_lock is held by m_start */
1977 walk_page_vma(vma, &show_numa_ops, md);
1983 seq_printf(m, " anon=%lu", md->anon);
1986 seq_printf(m, " dirty=%lu", md->dirty);
1988 if (md->pages != md->anon && md->pages != md->dirty)
1989 seq_printf(m, " mapped=%lu", md->pages);
1991 if (md->mapcount_max > 1)
1992 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1995 seq_printf(m, " swapcache=%lu", md->swapcache);
1997 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1998 seq_printf(m, " active=%lu", md->active);
2001 seq_printf(m, " writeback=%lu", md->writeback);
2003 for_each_node_state(nid, N_MEMORY)
2005 seq_printf(m, " N%d=%lu", nid, md->node[nid]);
2007 seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
2013 static const struct seq_operations proc_pid_numa_maps_op = {
2017 .show = show_numa_map,
2020 static int pid_numa_maps_open(struct inode *inode, struct file *file)
2022 return proc_maps_open(inode, file, &proc_pid_numa_maps_op,
2023 sizeof(struct numa_maps_private));
2026 const struct file_operations proc_pid_numa_maps_operations = {
2027 .open = pid_numa_maps_open,
2029 .llseek = seq_lseek,
2030 .release = proc_map_release,
2033 #endif /* CONFIG_NUMA */