task_mmu.c revision 8d0920bd
1// SPDX-License-Identifier: GPL-2.0
2#include <linux/pagewalk.h>
3#include <linux/vmacache.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>
22
23#include <asm/elf.h>
24#include <asm/tlb.h>
25#include <asm/tlbflush.h>
26#include "internal.h"
27
28#define SEQ_PUT_DEC(str, val) \
29		seq_put_decimal_ull_width(m, str, (val) << (PAGE_SHIFT-10), 8)
30void task_mem(struct seq_file *m, struct mm_struct *mm)
31{
32	unsigned long text, lib, swap, anon, file, shmem;
33	unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
34
35	anon = get_mm_counter(mm, MM_ANONPAGES);
36	file = get_mm_counter(mm, MM_FILEPAGES);
37	shmem = get_mm_counter(mm, MM_SHMEMPAGES);
38
39	/*
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.
45	 */
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;
52
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;
57
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);
77	seq_puts(m, " kB\n");
78	hugetlb_report_usage(m, mm);
79}
80#undef SEQ_PUT_DEC
81
82unsigned long task_vsize(struct mm_struct *mm)
83{
84	return PAGE_SIZE * mm->total_vm;
85}
86
87unsigned long task_statm(struct mm_struct *mm,
88			 unsigned long *shared, unsigned long *text,
89			 unsigned long *data, unsigned long *resident)
90{
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))
94								>> PAGE_SHIFT;
95	*data = mm->data_vm + mm->stack_vm;
96	*resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
97	return mm->total_vm;
98}
99
100#ifdef CONFIG_NUMA
101/*
102 * Save get_task_policy() for show_numa_map().
103 */
104static void hold_task_mempolicy(struct proc_maps_private *priv)
105{
106	struct task_struct *task = priv->task;
107
108	task_lock(task);
109	priv->task_mempolicy = get_task_policy(task);
110	mpol_get(priv->task_mempolicy);
111	task_unlock(task);
112}
113static void release_task_mempolicy(struct proc_maps_private *priv)
114{
115	mpol_put(priv->task_mempolicy);
116}
117#else
118static void hold_task_mempolicy(struct proc_maps_private *priv)
119{
120}
121static void release_task_mempolicy(struct proc_maps_private *priv)
122{
123}
124#endif
125
126static void *m_start(struct seq_file *m, loff_t *ppos)
127{
128	struct proc_maps_private *priv = m->private;
129	unsigned long last_addr = *ppos;
130	struct mm_struct *mm;
131	struct vm_area_struct *vma;
132
133	/* See m_next(). Zero at the start or after lseek. */
134	if (last_addr == -1UL)
135		return NULL;
136
137	priv->task = get_proc_task(priv->inode);
138	if (!priv->task)
139		return ERR_PTR(-ESRCH);
140
141	mm = priv->mm;
142	if (!mm || !mmget_not_zero(mm)) {
143		put_task_struct(priv->task);
144		priv->task = NULL;
145		return NULL;
146	}
147
148	if (mmap_read_lock_killable(mm)) {
149		mmput(mm);
150		put_task_struct(priv->task);
151		priv->task = NULL;
152		return ERR_PTR(-EINTR);
153	}
154
155	hold_task_mempolicy(priv);
156	priv->tail_vma = get_gate_vma(mm);
157
158	vma = find_vma(mm, last_addr);
159	if (vma)
160		return vma;
161
162	return priv->tail_vma;
163}
164
165static void *m_next(struct seq_file *m, void *v, loff_t *ppos)
166{
167	struct proc_maps_private *priv = m->private;
168	struct vm_area_struct *next, *vma = v;
169
170	if (vma == priv->tail_vma)
171		next = NULL;
172	else if (vma->vm_next)
173		next = vma->vm_next;
174	else
175		next = priv->tail_vma;
176
177	*ppos = next ? next->vm_start : -1UL;
178
179	return next;
180}
181
182static void m_stop(struct seq_file *m, void *v)
183{
184	struct proc_maps_private *priv = m->private;
185	struct mm_struct *mm = priv->mm;
186
187	if (!priv->task)
188		return;
189
190	release_task_mempolicy(priv);
191	mmap_read_unlock(mm);
192	mmput(mm);
193	put_task_struct(priv->task);
194	priv->task = NULL;
195}
196
197static int proc_maps_open(struct inode *inode, struct file *file,
198			const struct seq_operations *ops, int psize)
199{
200	struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
201
202	if (!priv)
203		return -ENOMEM;
204
205	priv->inode = inode;
206	priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
207	if (IS_ERR(priv->mm)) {
208		int err = PTR_ERR(priv->mm);
209
210		seq_release_private(inode, file);
211		return err;
212	}
213
214	return 0;
215}
216
217static int proc_map_release(struct inode *inode, struct file *file)
218{
219	struct seq_file *seq = file->private_data;
220	struct proc_maps_private *priv = seq->private;
221
222	if (priv->mm)
223		mmdrop(priv->mm);
224
225	return seq_release_private(inode, file);
226}
227
228static int do_maps_open(struct inode *inode, struct file *file,
229			const struct seq_operations *ops)
230{
231	return proc_maps_open(inode, file, ops,
232				sizeof(struct proc_maps_private));
233}
234
235/*
236 * Indicate if the VMA is a stack for the given task; for
237 * /proc/PID/maps that is the stack of the main task.
238 */
239static int is_stack(struct vm_area_struct *vma)
240{
241	/*
242	 * We make no effort to guess what a given thread considers to be
243	 * its "stack".  It's not even well-defined for programs written
244	 * languages like Go.
245	 */
246	return vma->vm_start <= vma->vm_mm->start_stack &&
247		vma->vm_end >= vma->vm_mm->start_stack;
248}
249
250static void show_vma_header_prefix(struct seq_file *m,
251				   unsigned long start, unsigned long end,
252				   vm_flags_t flags, unsigned long long pgoff,
253				   dev_t dev, unsigned long ino)
254{
255	seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
256	seq_put_hex_ll(m, NULL, start, 8);
257	seq_put_hex_ll(m, "-", end, 8);
258	seq_putc(m, ' ');
259	seq_putc(m, flags & VM_READ ? 'r' : '-');
260	seq_putc(m, flags & VM_WRITE ? 'w' : '-');
261	seq_putc(m, flags & VM_EXEC ? 'x' : '-');
262	seq_putc(m, flags & VM_MAYSHARE ? 's' : 'p');
263	seq_put_hex_ll(m, " ", pgoff, 8);
264	seq_put_hex_ll(m, " ", MAJOR(dev), 2);
265	seq_put_hex_ll(m, ":", MINOR(dev), 2);
266	seq_put_decimal_ull(m, " ", ino);
267	seq_putc(m, ' ');
268}
269
270static void
271show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
272{
273	struct mm_struct *mm = vma->vm_mm;
274	struct file *file = vma->vm_file;
275	vm_flags_t flags = vma->vm_flags;
276	unsigned long ino = 0;
277	unsigned long long pgoff = 0;
278	unsigned long start, end;
279	dev_t dev = 0;
280	const char *name = NULL;
281
282	if (file) {
283		struct inode *inode = file_inode(vma->vm_file);
284		dev = inode->i_sb->s_dev;
285		ino = inode->i_ino;
286		pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
287	}
288
289	start = vma->vm_start;
290	end = vma->vm_end;
291	show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino);
292
293	/*
294	 * Print the dentry name for named mappings, and a
295	 * special [heap] marker for the heap:
296	 */
297	if (file) {
298		seq_pad(m, ' ');
299		seq_file_path(m, file, "\n");
300		goto done;
301	}
302
303	if (vma->vm_ops && vma->vm_ops->name) {
304		name = vma->vm_ops->name(vma);
305		if (name)
306			goto done;
307	}
308
309	name = arch_vma_name(vma);
310	if (!name) {
311		if (!mm) {
312			name = "[vdso]";
313			goto done;
314		}
315
316		if (vma->vm_start <= mm->brk &&
317		    vma->vm_end >= mm->start_brk) {
318			name = "[heap]";
319			goto done;
320		}
321
322		if (is_stack(vma))
323			name = "[stack]";
324	}
325
326done:
327	if (name) {
328		seq_pad(m, ' ');
329		seq_puts(m, name);
330	}
331	seq_putc(m, '\n');
332}
333
334static int show_map(struct seq_file *m, void *v)
335{
336	show_map_vma(m, v);
337	return 0;
338}
339
340static const struct seq_operations proc_pid_maps_op = {
341	.start	= m_start,
342	.next	= m_next,
343	.stop	= m_stop,
344	.show	= show_map
345};
346
347static int pid_maps_open(struct inode *inode, struct file *file)
348{
349	return do_maps_open(inode, file, &proc_pid_maps_op);
350}
351
352const struct file_operations proc_pid_maps_operations = {
353	.open		= pid_maps_open,
354	.read		= seq_read,
355	.llseek		= seq_lseek,
356	.release	= proc_map_release,
357};
358
359/*
360 * Proportional Set Size(PSS): my share of RSS.
361 *
362 * PSS of a process is the count of pages it has in memory, where each
363 * page is divided by the number of processes sharing it.  So if a
364 * process has 1000 pages all to itself, and 1000 shared with one other
365 * process, its PSS will be 1500.
366 *
367 * To keep (accumulated) division errors low, we adopt a 64bit
368 * fixed-point pss counter to minimize division errors. So (pss >>
369 * PSS_SHIFT) would be the real byte count.
370 *
371 * A shift of 12 before division means (assuming 4K page size):
372 * 	- 1M 3-user-pages add up to 8KB errors;
373 * 	- supports mapcount up to 2^24, or 16M;
374 * 	- supports PSS up to 2^52 bytes, or 4PB.
375 */
376#define PSS_SHIFT 12
377
378#ifdef CONFIG_PROC_PAGE_MONITOR
379struct mem_size_stats {
380	unsigned long resident;
381	unsigned long shared_clean;
382	unsigned long shared_dirty;
383	unsigned long private_clean;
384	unsigned long private_dirty;
385	unsigned long referenced;
386	unsigned long anonymous;
387	unsigned long lazyfree;
388	unsigned long anonymous_thp;
389	unsigned long shmem_thp;
390	unsigned long file_thp;
391	unsigned long swap;
392	unsigned long shared_hugetlb;
393	unsigned long private_hugetlb;
394	u64 pss;
395	u64 pss_anon;
396	u64 pss_file;
397	u64 pss_shmem;
398	u64 pss_locked;
399	u64 swap_pss;
400	bool check_shmem_swap;
401};
402
403static void smaps_page_accumulate(struct mem_size_stats *mss,
404		struct page *page, unsigned long size, unsigned long pss,
405		bool dirty, bool locked, bool private)
406{
407	mss->pss += pss;
408
409	if (PageAnon(page))
410		mss->pss_anon += pss;
411	else if (PageSwapBacked(page))
412		mss->pss_shmem += pss;
413	else
414		mss->pss_file += pss;
415
416	if (locked)
417		mss->pss_locked += pss;
418
419	if (dirty || PageDirty(page)) {
420		if (private)
421			mss->private_dirty += size;
422		else
423			mss->shared_dirty += size;
424	} else {
425		if (private)
426			mss->private_clean += size;
427		else
428			mss->shared_clean += size;
429	}
430}
431
432static void smaps_account(struct mem_size_stats *mss, struct page *page,
433		bool compound, bool young, bool dirty, bool locked)
434{
435	int i, nr = compound ? compound_nr(page) : 1;
436	unsigned long size = nr * PAGE_SIZE;
437
438	/*
439	 * First accumulate quantities that depend only on |size| and the type
440	 * of the compound page.
441	 */
442	if (PageAnon(page)) {
443		mss->anonymous += size;
444		if (!PageSwapBacked(page) && !dirty && !PageDirty(page))
445			mss->lazyfree += size;
446	}
447
448	mss->resident += size;
449	/* Accumulate the size in pages that have been accessed. */
450	if (young || page_is_young(page) || PageReferenced(page))
451		mss->referenced += size;
452
453	/*
454	 * Then accumulate quantities that may depend on sharing, or that may
455	 * differ page-by-page.
456	 *
457	 * page_count(page) == 1 guarantees the page is mapped exactly once.
458	 * If any subpage of the compound page mapped with PTE it would elevate
459	 * page_count().
460	 */
461	if (page_count(page) == 1) {
462		smaps_page_accumulate(mss, page, size, size << PSS_SHIFT, dirty,
463			locked, true);
464		return;
465	}
466	for (i = 0; i < nr; i++, page++) {
467		int mapcount = page_mapcount(page);
468		unsigned long pss = PAGE_SIZE << PSS_SHIFT;
469		if (mapcount >= 2)
470			pss /= mapcount;
471		smaps_page_accumulate(mss, page, PAGE_SIZE, pss, dirty, locked,
472				      mapcount < 2);
473	}
474}
475
476#ifdef CONFIG_SHMEM
477static int smaps_pte_hole(unsigned long addr, unsigned long end,
478			  __always_unused int depth, struct mm_walk *walk)
479{
480	struct mem_size_stats *mss = walk->private;
481
482	mss->swap += shmem_partial_swap_usage(
483			walk->vma->vm_file->f_mapping, addr, end);
484
485	return 0;
486}
487#else
488#define smaps_pte_hole		NULL
489#endif /* CONFIG_SHMEM */
490
491static void smaps_pte_entry(pte_t *pte, unsigned long addr,
492		struct mm_walk *walk)
493{
494	struct mem_size_stats *mss = walk->private;
495	struct vm_area_struct *vma = walk->vma;
496	bool locked = !!(vma->vm_flags & VM_LOCKED);
497	struct page *page = NULL;
498
499	if (pte_present(*pte)) {
500		page = vm_normal_page(vma, addr, *pte);
501	} else if (is_swap_pte(*pte)) {
502		swp_entry_t swpent = pte_to_swp_entry(*pte);
503
504		if (!non_swap_entry(swpent)) {
505			int mapcount;
506
507			mss->swap += PAGE_SIZE;
508			mapcount = swp_swapcount(swpent);
509			if (mapcount >= 2) {
510				u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
511
512				do_div(pss_delta, mapcount);
513				mss->swap_pss += pss_delta;
514			} else {
515				mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
516			}
517		} else if (is_pfn_swap_entry(swpent))
518			page = pfn_swap_entry_to_page(swpent);
519	} else if (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap
520							&& pte_none(*pte))) {
521		page = xa_load(&vma->vm_file->f_mapping->i_pages,
522						linear_page_index(vma, addr));
523		if (xa_is_value(page))
524			mss->swap += PAGE_SIZE;
525		return;
526	}
527
528	if (!page)
529		return;
530
531	smaps_account(mss, page, false, pte_young(*pte), pte_dirty(*pte), locked);
532}
533
534#ifdef CONFIG_TRANSPARENT_HUGEPAGE
535static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
536		struct mm_walk *walk)
537{
538	struct mem_size_stats *mss = walk->private;
539	struct vm_area_struct *vma = walk->vma;
540	bool locked = !!(vma->vm_flags & VM_LOCKED);
541	struct page *page = NULL;
542
543	if (pmd_present(*pmd)) {
544		/* FOLL_DUMP will return -EFAULT on huge zero page */
545		page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
546	} else if (unlikely(thp_migration_supported() && is_swap_pmd(*pmd))) {
547		swp_entry_t entry = pmd_to_swp_entry(*pmd);
548
549		if (is_migration_entry(entry))
550			page = pfn_swap_entry_to_page(entry);
551	}
552	if (IS_ERR_OR_NULL(page))
553		return;
554	if (PageAnon(page))
555		mss->anonymous_thp += HPAGE_PMD_SIZE;
556	else if (PageSwapBacked(page))
557		mss->shmem_thp += HPAGE_PMD_SIZE;
558	else if (is_zone_device_page(page))
559		/* pass */;
560	else
561		mss->file_thp += HPAGE_PMD_SIZE;
562	smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd), locked);
563}
564#else
565static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
566		struct mm_walk *walk)
567{
568}
569#endif
570
571static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
572			   struct mm_walk *walk)
573{
574	struct vm_area_struct *vma = walk->vma;
575	pte_t *pte;
576	spinlock_t *ptl;
577
578	ptl = pmd_trans_huge_lock(pmd, vma);
579	if (ptl) {
580		smaps_pmd_entry(pmd, addr, walk);
581		spin_unlock(ptl);
582		goto out;
583	}
584
585	if (pmd_trans_unstable(pmd))
586		goto out;
587	/*
588	 * The mmap_lock held all the way back in m_start() is what
589	 * keeps khugepaged out of here and from collapsing things
590	 * in here.
591	 */
592	pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
593	for (; addr != end; pte++, addr += PAGE_SIZE)
594		smaps_pte_entry(pte, addr, walk);
595	pte_unmap_unlock(pte - 1, ptl);
596out:
597	cond_resched();
598	return 0;
599}
600
601static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
602{
603	/*
604	 * Don't forget to update Documentation/ on changes.
605	 */
606	static const char mnemonics[BITS_PER_LONG][2] = {
607		/*
608		 * In case if we meet a flag we don't know about.
609		 */
610		[0 ... (BITS_PER_LONG-1)] = "??",
611
612		[ilog2(VM_READ)]	= "rd",
613		[ilog2(VM_WRITE)]	= "wr",
614		[ilog2(VM_EXEC)]	= "ex",
615		[ilog2(VM_SHARED)]	= "sh",
616		[ilog2(VM_MAYREAD)]	= "mr",
617		[ilog2(VM_MAYWRITE)]	= "mw",
618		[ilog2(VM_MAYEXEC)]	= "me",
619		[ilog2(VM_MAYSHARE)]	= "ms",
620		[ilog2(VM_GROWSDOWN)]	= "gd",
621		[ilog2(VM_PFNMAP)]	= "pf",
622		[ilog2(VM_LOCKED)]	= "lo",
623		[ilog2(VM_IO)]		= "io",
624		[ilog2(VM_SEQ_READ)]	= "sr",
625		[ilog2(VM_RAND_READ)]	= "rr",
626		[ilog2(VM_DONTCOPY)]	= "dc",
627		[ilog2(VM_DONTEXPAND)]	= "de",
628		[ilog2(VM_ACCOUNT)]	= "ac",
629		[ilog2(VM_NORESERVE)]	= "nr",
630		[ilog2(VM_HUGETLB)]	= "ht",
631		[ilog2(VM_SYNC)]	= "sf",
632		[ilog2(VM_ARCH_1)]	= "ar",
633		[ilog2(VM_WIPEONFORK)]	= "wf",
634		[ilog2(VM_DONTDUMP)]	= "dd",
635#ifdef CONFIG_ARM64_BTI
636		[ilog2(VM_ARM64_BTI)]	= "bt",
637#endif
638#ifdef CONFIG_MEM_SOFT_DIRTY
639		[ilog2(VM_SOFTDIRTY)]	= "sd",
640#endif
641		[ilog2(VM_MIXEDMAP)]	= "mm",
642		[ilog2(VM_HUGEPAGE)]	= "hg",
643		[ilog2(VM_NOHUGEPAGE)]	= "nh",
644		[ilog2(VM_MERGEABLE)]	= "mg",
645		[ilog2(VM_UFFD_MISSING)]= "um",
646		[ilog2(VM_UFFD_WP)]	= "uw",
647#ifdef CONFIG_ARM64_MTE
648		[ilog2(VM_MTE)]		= "mt",
649		[ilog2(VM_MTE_ALLOWED)]	= "",
650#endif
651#ifdef CONFIG_ARCH_HAS_PKEYS
652		/* These come out via ProtectionKey: */
653		[ilog2(VM_PKEY_BIT0)]	= "",
654		[ilog2(VM_PKEY_BIT1)]	= "",
655		[ilog2(VM_PKEY_BIT2)]	= "",
656		[ilog2(VM_PKEY_BIT3)]	= "",
657#if VM_PKEY_BIT4
658		[ilog2(VM_PKEY_BIT4)]	= "",
659#endif
660#endif /* CONFIG_ARCH_HAS_PKEYS */
661#ifdef CONFIG_HAVE_ARCH_USERFAULTFD_MINOR
662		[ilog2(VM_UFFD_MINOR)]	= "ui",
663#endif /* CONFIG_HAVE_ARCH_USERFAULTFD_MINOR */
664	};
665	size_t i;
666
667	seq_puts(m, "VmFlags: ");
668	for (i = 0; i < BITS_PER_LONG; i++) {
669		if (!mnemonics[i][0])
670			continue;
671		if (vma->vm_flags & (1UL << i)) {
672			seq_putc(m, mnemonics[i][0]);
673			seq_putc(m, mnemonics[i][1]);
674			seq_putc(m, ' ');
675		}
676	}
677	seq_putc(m, '\n');
678}
679
680#ifdef CONFIG_HUGETLB_PAGE
681static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
682				 unsigned long addr, unsigned long end,
683				 struct mm_walk *walk)
684{
685	struct mem_size_stats *mss = walk->private;
686	struct vm_area_struct *vma = walk->vma;
687	struct page *page = NULL;
688
689	if (pte_present(*pte)) {
690		page = vm_normal_page(vma, addr, *pte);
691	} else if (is_swap_pte(*pte)) {
692		swp_entry_t swpent = pte_to_swp_entry(*pte);
693
694		if (is_pfn_swap_entry(swpent))
695			page = pfn_swap_entry_to_page(swpent);
696	}
697	if (page) {
698		int mapcount = page_mapcount(page);
699
700		if (mapcount >= 2)
701			mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
702		else
703			mss->private_hugetlb += huge_page_size(hstate_vma(vma));
704	}
705	return 0;
706}
707#else
708#define smaps_hugetlb_range	NULL
709#endif /* HUGETLB_PAGE */
710
711static const struct mm_walk_ops smaps_walk_ops = {
712	.pmd_entry		= smaps_pte_range,
713	.hugetlb_entry		= smaps_hugetlb_range,
714};
715
716static const struct mm_walk_ops smaps_shmem_walk_ops = {
717	.pmd_entry		= smaps_pte_range,
718	.hugetlb_entry		= smaps_hugetlb_range,
719	.pte_hole		= smaps_pte_hole,
720};
721
722/*
723 * Gather mem stats from @vma with the indicated beginning
724 * address @start, and keep them in @mss.
725 *
726 * Use vm_start of @vma as the beginning address if @start is 0.
727 */
728static void smap_gather_stats(struct vm_area_struct *vma,
729		struct mem_size_stats *mss, unsigned long start)
730{
731	const struct mm_walk_ops *ops = &smaps_walk_ops;
732
733	/* Invalid start */
734	if (start >= vma->vm_end)
735		return;
736
737#ifdef CONFIG_SHMEM
738	/* In case of smaps_rollup, reset the value from previous vma */
739	mss->check_shmem_swap = false;
740	if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
741		/*
742		 * For shared or readonly shmem mappings we know that all
743		 * swapped out pages belong to the shmem object, and we can
744		 * obtain the swap value much more efficiently. For private
745		 * writable mappings, we might have COW pages that are
746		 * not affected by the parent swapped out pages of the shmem
747		 * object, so we have to distinguish them during the page walk.
748		 * Unless we know that the shmem object (or the part mapped by
749		 * our VMA) has no swapped out pages at all.
750		 */
751		unsigned long shmem_swapped = shmem_swap_usage(vma);
752
753		if (!start && (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
754					!(vma->vm_flags & VM_WRITE))) {
755			mss->swap += shmem_swapped;
756		} else {
757			mss->check_shmem_swap = true;
758			ops = &smaps_shmem_walk_ops;
759		}
760	}
761#endif
762	/* mmap_lock is held in m_start */
763	if (!start)
764		walk_page_vma(vma, ops, mss);
765	else
766		walk_page_range(vma->vm_mm, start, vma->vm_end, ops, mss);
767}
768
769#define SEQ_PUT_DEC(str, val) \
770		seq_put_decimal_ull_width(m, str, (val) >> 10, 8)
771
772/* Show the contents common for smaps and smaps_rollup */
773static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss,
774	bool rollup_mode)
775{
776	SEQ_PUT_DEC("Rss:            ", mss->resident);
777	SEQ_PUT_DEC(" kB\nPss:            ", mss->pss >> PSS_SHIFT);
778	if (rollup_mode) {
779		/*
780		 * These are meaningful only for smaps_rollup, otherwise two of
781		 * them are zero, and the other one is the same as Pss.
782		 */
783		SEQ_PUT_DEC(" kB\nPss_Anon:       ",
784			mss->pss_anon >> PSS_SHIFT);
785		SEQ_PUT_DEC(" kB\nPss_File:       ",
786			mss->pss_file >> PSS_SHIFT);
787		SEQ_PUT_DEC(" kB\nPss_Shmem:      ",
788			mss->pss_shmem >> PSS_SHIFT);
789	}
790	SEQ_PUT_DEC(" kB\nShared_Clean:   ", mss->shared_clean);
791	SEQ_PUT_DEC(" kB\nShared_Dirty:   ", mss->shared_dirty);
792	SEQ_PUT_DEC(" kB\nPrivate_Clean:  ", mss->private_clean);
793	SEQ_PUT_DEC(" kB\nPrivate_Dirty:  ", mss->private_dirty);
794	SEQ_PUT_DEC(" kB\nReferenced:     ", mss->referenced);
795	SEQ_PUT_DEC(" kB\nAnonymous:      ", mss->anonymous);
796	SEQ_PUT_DEC(" kB\nLazyFree:       ", mss->lazyfree);
797	SEQ_PUT_DEC(" kB\nAnonHugePages:  ", mss->anonymous_thp);
798	SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp);
799	SEQ_PUT_DEC(" kB\nFilePmdMapped:  ", mss->file_thp);
800	SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb);
801	seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ",
802				  mss->private_hugetlb >> 10, 7);
803	SEQ_PUT_DEC(" kB\nSwap:           ", mss->swap);
804	SEQ_PUT_DEC(" kB\nSwapPss:        ",
805					mss->swap_pss >> PSS_SHIFT);
806	SEQ_PUT_DEC(" kB\nLocked:         ",
807					mss->pss_locked >> PSS_SHIFT);
808	seq_puts(m, " kB\n");
809}
810
811static int show_smap(struct seq_file *m, void *v)
812{
813	struct vm_area_struct *vma = v;
814	struct mem_size_stats mss;
815
816	memset(&mss, 0, sizeof(mss));
817
818	smap_gather_stats(vma, &mss, 0);
819
820	show_map_vma(m, vma);
821
822	SEQ_PUT_DEC("Size:           ", vma->vm_end - vma->vm_start);
823	SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma));
824	SEQ_PUT_DEC(" kB\nMMUPageSize:    ", vma_mmu_pagesize(vma));
825	seq_puts(m, " kB\n");
826
827	__show_smap(m, &mss, false);
828
829	seq_printf(m, "THPeligible:    %d\n",
830		   transparent_hugepage_active(vma));
831
832	if (arch_pkeys_enabled())
833		seq_printf(m, "ProtectionKey:  %8u\n", vma_pkey(vma));
834	show_smap_vma_flags(m, vma);
835
836	return 0;
837}
838
839static int show_smaps_rollup(struct seq_file *m, void *v)
840{
841	struct proc_maps_private *priv = m->private;
842	struct mem_size_stats mss;
843	struct mm_struct *mm;
844	struct vm_area_struct *vma;
845	unsigned long last_vma_end = 0;
846	int ret = 0;
847
848	priv->task = get_proc_task(priv->inode);
849	if (!priv->task)
850		return -ESRCH;
851
852	mm = priv->mm;
853	if (!mm || !mmget_not_zero(mm)) {
854		ret = -ESRCH;
855		goto out_put_task;
856	}
857
858	memset(&mss, 0, sizeof(mss));
859
860	ret = mmap_read_lock_killable(mm);
861	if (ret)
862		goto out_put_mm;
863
864	hold_task_mempolicy(priv);
865
866	for (vma = priv->mm->mmap; vma;) {
867		smap_gather_stats(vma, &mss, 0);
868		last_vma_end = vma->vm_end;
869
870		/*
871		 * Release mmap_lock temporarily if someone wants to
872		 * access it for write request.
873		 */
874		if (mmap_lock_is_contended(mm)) {
875			mmap_read_unlock(mm);
876			ret = mmap_read_lock_killable(mm);
877			if (ret) {
878				release_task_mempolicy(priv);
879				goto out_put_mm;
880			}
881
882			/*
883			 * After dropping the lock, there are four cases to
884			 * consider. See the following example for explanation.
885			 *
886			 *   +------+------+-----------+
887			 *   | VMA1 | VMA2 | VMA3      |
888			 *   +------+------+-----------+
889			 *   |      |      |           |
890			 *  4k     8k     16k         400k
891			 *
892			 * Suppose we drop the lock after reading VMA2 due to
893			 * contention, then we get:
894			 *
895			 *	last_vma_end = 16k
896			 *
897			 * 1) VMA2 is freed, but VMA3 exists:
898			 *
899			 *    find_vma(mm, 16k - 1) will return VMA3.
900			 *    In this case, just continue from VMA3.
901			 *
902			 * 2) VMA2 still exists:
903			 *
904			 *    find_vma(mm, 16k - 1) will return VMA2.
905			 *    Iterate the loop like the original one.
906			 *
907			 * 3) No more VMAs can be found:
908			 *
909			 *    find_vma(mm, 16k - 1) will return NULL.
910			 *    No more things to do, just break.
911			 *
912			 * 4) (last_vma_end - 1) is the middle of a vma (VMA'):
913			 *
914			 *    find_vma(mm, 16k - 1) will return VMA' whose range
915			 *    contains last_vma_end.
916			 *    Iterate VMA' from last_vma_end.
917			 */
918			vma = find_vma(mm, last_vma_end - 1);
919			/* Case 3 above */
920			if (!vma)
921				break;
922
923			/* Case 1 above */
924			if (vma->vm_start >= last_vma_end)
925				continue;
926
927			/* Case 4 above */
928			if (vma->vm_end > last_vma_end)
929				smap_gather_stats(vma, &mss, last_vma_end);
930		}
931		/* Case 2 above */
932		vma = vma->vm_next;
933	}
934
935	show_vma_header_prefix(m, priv->mm->mmap->vm_start,
936			       last_vma_end, 0, 0, 0, 0);
937	seq_pad(m, ' ');
938	seq_puts(m, "[rollup]\n");
939
940	__show_smap(m, &mss, true);
941
942	release_task_mempolicy(priv);
943	mmap_read_unlock(mm);
944
945out_put_mm:
946	mmput(mm);
947out_put_task:
948	put_task_struct(priv->task);
949	priv->task = NULL;
950
951	return ret;
952}
953#undef SEQ_PUT_DEC
954
955static const struct seq_operations proc_pid_smaps_op = {
956	.start	= m_start,
957	.next	= m_next,
958	.stop	= m_stop,
959	.show	= show_smap
960};
961
962static int pid_smaps_open(struct inode *inode, struct file *file)
963{
964	return do_maps_open(inode, file, &proc_pid_smaps_op);
965}
966
967static int smaps_rollup_open(struct inode *inode, struct file *file)
968{
969	int ret;
970	struct proc_maps_private *priv;
971
972	priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT);
973	if (!priv)
974		return -ENOMEM;
975
976	ret = single_open(file, show_smaps_rollup, priv);
977	if (ret)
978		goto out_free;
979
980	priv->inode = inode;
981	priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
982	if (IS_ERR(priv->mm)) {
983		ret = PTR_ERR(priv->mm);
984
985		single_release(inode, file);
986		goto out_free;
987	}
988
989	return 0;
990
991out_free:
992	kfree(priv);
993	return ret;
994}
995
996static int smaps_rollup_release(struct inode *inode, struct file *file)
997{
998	struct seq_file *seq = file->private_data;
999	struct proc_maps_private *priv = seq->private;
1000
1001	if (priv->mm)
1002		mmdrop(priv->mm);
1003
1004	kfree(priv);
1005	return single_release(inode, file);
1006}
1007
1008const struct file_operations proc_pid_smaps_operations = {
1009	.open		= pid_smaps_open,
1010	.read		= seq_read,
1011	.llseek		= seq_lseek,
1012	.release	= proc_map_release,
1013};
1014
1015const struct file_operations proc_pid_smaps_rollup_operations = {
1016	.open		= smaps_rollup_open,
1017	.read		= seq_read,
1018	.llseek		= seq_lseek,
1019	.release	= smaps_rollup_release,
1020};
1021
1022enum clear_refs_types {
1023	CLEAR_REFS_ALL = 1,
1024	CLEAR_REFS_ANON,
1025	CLEAR_REFS_MAPPED,
1026	CLEAR_REFS_SOFT_DIRTY,
1027	CLEAR_REFS_MM_HIWATER_RSS,
1028	CLEAR_REFS_LAST,
1029};
1030
1031struct clear_refs_private {
1032	enum clear_refs_types type;
1033};
1034
1035#ifdef CONFIG_MEM_SOFT_DIRTY
1036
1037static inline bool pte_is_pinned(struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1038{
1039	struct page *page;
1040
1041	if (!pte_write(pte))
1042		return false;
1043	if (!is_cow_mapping(vma->vm_flags))
1044		return false;
1045	if (likely(!test_bit(MMF_HAS_PINNED, &vma->vm_mm->flags)))
1046		return false;
1047	page = vm_normal_page(vma, addr, pte);
1048	if (!page)
1049		return false;
1050	return page_maybe_dma_pinned(page);
1051}
1052
1053static inline void clear_soft_dirty(struct vm_area_struct *vma,
1054		unsigned long addr, pte_t *pte)
1055{
1056	/*
1057	 * The soft-dirty tracker uses #PF-s to catch writes
1058	 * to pages, so write-protect the pte as well. See the
1059	 * Documentation/admin-guide/mm/soft-dirty.rst for full description
1060	 * of how soft-dirty works.
1061	 */
1062	pte_t ptent = *pte;
1063
1064	if (pte_present(ptent)) {
1065		pte_t old_pte;
1066
1067		if (pte_is_pinned(vma, addr, ptent))
1068			return;
1069		old_pte = ptep_modify_prot_start(vma, addr, pte);
1070		ptent = pte_wrprotect(old_pte);
1071		ptent = pte_clear_soft_dirty(ptent);
1072		ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent);
1073	} else if (is_swap_pte(ptent)) {
1074		ptent = pte_swp_clear_soft_dirty(ptent);
1075		set_pte_at(vma->vm_mm, addr, pte, ptent);
1076	}
1077}
1078#else
1079static inline void clear_soft_dirty(struct vm_area_struct *vma,
1080		unsigned long addr, pte_t *pte)
1081{
1082}
1083#endif
1084
1085#if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
1086static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1087		unsigned long addr, pmd_t *pmdp)
1088{
1089	pmd_t old, pmd = *pmdp;
1090
1091	if (pmd_present(pmd)) {
1092		/* See comment in change_huge_pmd() */
1093		old = pmdp_invalidate(vma, addr, pmdp);
1094		if (pmd_dirty(old))
1095			pmd = pmd_mkdirty(pmd);
1096		if (pmd_young(old))
1097			pmd = pmd_mkyoung(pmd);
1098
1099		pmd = pmd_wrprotect(pmd);
1100		pmd = pmd_clear_soft_dirty(pmd);
1101
1102		set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1103	} else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1104		pmd = pmd_swp_clear_soft_dirty(pmd);
1105		set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1106	}
1107}
1108#else
1109static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1110		unsigned long addr, pmd_t *pmdp)
1111{
1112}
1113#endif
1114
1115static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
1116				unsigned long end, struct mm_walk *walk)
1117{
1118	struct clear_refs_private *cp = walk->private;
1119	struct vm_area_struct *vma = walk->vma;
1120	pte_t *pte, ptent;
1121	spinlock_t *ptl;
1122	struct page *page;
1123
1124	ptl = pmd_trans_huge_lock(pmd, vma);
1125	if (ptl) {
1126		if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1127			clear_soft_dirty_pmd(vma, addr, pmd);
1128			goto out;
1129		}
1130
1131		if (!pmd_present(*pmd))
1132			goto out;
1133
1134		page = pmd_page(*pmd);
1135
1136		/* Clear accessed and referenced bits. */
1137		pmdp_test_and_clear_young(vma, addr, pmd);
1138		test_and_clear_page_young(page);
1139		ClearPageReferenced(page);
1140out:
1141		spin_unlock(ptl);
1142		return 0;
1143	}
1144
1145	if (pmd_trans_unstable(pmd))
1146		return 0;
1147
1148	pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1149	for (; addr != end; pte++, addr += PAGE_SIZE) {
1150		ptent = *pte;
1151
1152		if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1153			clear_soft_dirty(vma, addr, pte);
1154			continue;
1155		}
1156
1157		if (!pte_present(ptent))
1158			continue;
1159
1160		page = vm_normal_page(vma, addr, ptent);
1161		if (!page)
1162			continue;
1163
1164		/* Clear accessed and referenced bits. */
1165		ptep_test_and_clear_young(vma, addr, pte);
1166		test_and_clear_page_young(page);
1167		ClearPageReferenced(page);
1168	}
1169	pte_unmap_unlock(pte - 1, ptl);
1170	cond_resched();
1171	return 0;
1172}
1173
1174static int clear_refs_test_walk(unsigned long start, unsigned long end,
1175				struct mm_walk *walk)
1176{
1177	struct clear_refs_private *cp = walk->private;
1178	struct vm_area_struct *vma = walk->vma;
1179
1180	if (vma->vm_flags & VM_PFNMAP)
1181		return 1;
1182
1183	/*
1184	 * Writing 1 to /proc/pid/clear_refs affects all pages.
1185	 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1186	 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1187	 * Writing 4 to /proc/pid/clear_refs affects all pages.
1188	 */
1189	if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1190		return 1;
1191	if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1192		return 1;
1193	return 0;
1194}
1195
1196static const struct mm_walk_ops clear_refs_walk_ops = {
1197	.pmd_entry		= clear_refs_pte_range,
1198	.test_walk		= clear_refs_test_walk,
1199};
1200
1201static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1202				size_t count, loff_t *ppos)
1203{
1204	struct task_struct *task;
1205	char buffer[PROC_NUMBUF];
1206	struct mm_struct *mm;
1207	struct vm_area_struct *vma;
1208	enum clear_refs_types type;
1209	int itype;
1210	int rv;
1211
1212	memset(buffer, 0, sizeof(buffer));
1213	if (count > sizeof(buffer) - 1)
1214		count = sizeof(buffer) - 1;
1215	if (copy_from_user(buffer, buf, count))
1216		return -EFAULT;
1217	rv = kstrtoint(strstrip(buffer), 10, &itype);
1218	if (rv < 0)
1219		return rv;
1220	type = (enum clear_refs_types)itype;
1221	if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1222		return -EINVAL;
1223
1224	task = get_proc_task(file_inode(file));
1225	if (!task)
1226		return -ESRCH;
1227	mm = get_task_mm(task);
1228	if (mm) {
1229		struct mmu_notifier_range range;
1230		struct clear_refs_private cp = {
1231			.type = type,
1232		};
1233
1234		if (mmap_write_lock_killable(mm)) {
1235			count = -EINTR;
1236			goto out_mm;
1237		}
1238		if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1239			/*
1240			 * Writing 5 to /proc/pid/clear_refs resets the peak
1241			 * resident set size to this mm's current rss value.
1242			 */
1243			reset_mm_hiwater_rss(mm);
1244			goto out_unlock;
1245		}
1246
1247		if (type == CLEAR_REFS_SOFT_DIRTY) {
1248			for (vma = mm->mmap; vma; vma = vma->vm_next) {
1249				if (!(vma->vm_flags & VM_SOFTDIRTY))
1250					continue;
1251				vma->vm_flags &= ~VM_SOFTDIRTY;
1252				vma_set_page_prot(vma);
1253			}
1254
1255			inc_tlb_flush_pending(mm);
1256			mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY,
1257						0, NULL, mm, 0, -1UL);
1258			mmu_notifier_invalidate_range_start(&range);
1259		}
1260		walk_page_range(mm, 0, mm->highest_vm_end, &clear_refs_walk_ops,
1261				&cp);
1262		if (type == CLEAR_REFS_SOFT_DIRTY) {
1263			mmu_notifier_invalidate_range_end(&range);
1264			flush_tlb_mm(mm);
1265			dec_tlb_flush_pending(mm);
1266		}
1267out_unlock:
1268		mmap_write_unlock(mm);
1269out_mm:
1270		mmput(mm);
1271	}
1272	put_task_struct(task);
1273
1274	return count;
1275}
1276
1277const struct file_operations proc_clear_refs_operations = {
1278	.write		= clear_refs_write,
1279	.llseek		= noop_llseek,
1280};
1281
1282typedef struct {
1283	u64 pme;
1284} pagemap_entry_t;
1285
1286struct pagemapread {
1287	int pos, len;		/* units: PM_ENTRY_BYTES, not bytes */
1288	pagemap_entry_t *buffer;
1289	bool show_pfn;
1290};
1291
1292#define PAGEMAP_WALK_SIZE	(PMD_SIZE)
1293#define PAGEMAP_WALK_MASK	(PMD_MASK)
1294
1295#define PM_ENTRY_BYTES		sizeof(pagemap_entry_t)
1296#define PM_PFRAME_BITS		55
1297#define PM_PFRAME_MASK		GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1298#define PM_SOFT_DIRTY		BIT_ULL(55)
1299#define PM_MMAP_EXCLUSIVE	BIT_ULL(56)
1300#define PM_UFFD_WP		BIT_ULL(57)
1301#define PM_FILE			BIT_ULL(61)
1302#define PM_SWAP			BIT_ULL(62)
1303#define PM_PRESENT		BIT_ULL(63)
1304
1305#define PM_END_OF_BUFFER    1
1306
1307static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1308{
1309	return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1310}
1311
1312static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1313			  struct pagemapread *pm)
1314{
1315	pm->buffer[pm->pos++] = *pme;
1316	if (pm->pos >= pm->len)
1317		return PM_END_OF_BUFFER;
1318	return 0;
1319}
1320
1321static int pagemap_pte_hole(unsigned long start, unsigned long end,
1322			    __always_unused int depth, struct mm_walk *walk)
1323{
1324	struct pagemapread *pm = walk->private;
1325	unsigned long addr = start;
1326	int err = 0;
1327
1328	while (addr < end) {
1329		struct vm_area_struct *vma = find_vma(walk->mm, addr);
1330		pagemap_entry_t pme = make_pme(0, 0);
1331		/* End of address space hole, which we mark as non-present. */
1332		unsigned long hole_end;
1333
1334		if (vma)
1335			hole_end = min(end, vma->vm_start);
1336		else
1337			hole_end = end;
1338
1339		for (; addr < hole_end; addr += PAGE_SIZE) {
1340			err = add_to_pagemap(addr, &pme, pm);
1341			if (err)
1342				goto out;
1343		}
1344
1345		if (!vma)
1346			break;
1347
1348		/* Addresses in the VMA. */
1349		if (vma->vm_flags & VM_SOFTDIRTY)
1350			pme = make_pme(0, PM_SOFT_DIRTY);
1351		for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1352			err = add_to_pagemap(addr, &pme, pm);
1353			if (err)
1354				goto out;
1355		}
1356	}
1357out:
1358	return err;
1359}
1360
1361static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1362		struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1363{
1364	u64 frame = 0, flags = 0;
1365	struct page *page = NULL;
1366
1367	if (pte_present(pte)) {
1368		if (pm->show_pfn)
1369			frame = pte_pfn(pte);
1370		flags |= PM_PRESENT;
1371		page = vm_normal_page(vma, addr, pte);
1372		if (pte_soft_dirty(pte))
1373			flags |= PM_SOFT_DIRTY;
1374		if (pte_uffd_wp(pte))
1375			flags |= PM_UFFD_WP;
1376	} else if (is_swap_pte(pte)) {
1377		swp_entry_t entry;
1378		if (pte_swp_soft_dirty(pte))
1379			flags |= PM_SOFT_DIRTY;
1380		if (pte_swp_uffd_wp(pte))
1381			flags |= PM_UFFD_WP;
1382		entry = pte_to_swp_entry(pte);
1383		if (pm->show_pfn)
1384			frame = swp_type(entry) |
1385				(swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1386		flags |= PM_SWAP;
1387		if (is_pfn_swap_entry(entry))
1388			page = pfn_swap_entry_to_page(entry);
1389	}
1390
1391	if (page && !PageAnon(page))
1392		flags |= PM_FILE;
1393	if (page && page_mapcount(page) == 1)
1394		flags |= PM_MMAP_EXCLUSIVE;
1395	if (vma->vm_flags & VM_SOFTDIRTY)
1396		flags |= PM_SOFT_DIRTY;
1397
1398	return make_pme(frame, flags);
1399}
1400
1401static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1402			     struct mm_walk *walk)
1403{
1404	struct vm_area_struct *vma = walk->vma;
1405	struct pagemapread *pm = walk->private;
1406	spinlock_t *ptl;
1407	pte_t *pte, *orig_pte;
1408	int err = 0;
1409
1410#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1411	ptl = pmd_trans_huge_lock(pmdp, vma);
1412	if (ptl) {
1413		u64 flags = 0, frame = 0;
1414		pmd_t pmd = *pmdp;
1415		struct page *page = NULL;
1416
1417		if (vma->vm_flags & VM_SOFTDIRTY)
1418			flags |= PM_SOFT_DIRTY;
1419
1420		if (pmd_present(pmd)) {
1421			page = pmd_page(pmd);
1422
1423			flags |= PM_PRESENT;
1424			if (pmd_soft_dirty(pmd))
1425				flags |= PM_SOFT_DIRTY;
1426			if (pmd_uffd_wp(pmd))
1427				flags |= PM_UFFD_WP;
1428			if (pm->show_pfn)
1429				frame = pmd_pfn(pmd) +
1430					((addr & ~PMD_MASK) >> PAGE_SHIFT);
1431		}
1432#ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1433		else if (is_swap_pmd(pmd)) {
1434			swp_entry_t entry = pmd_to_swp_entry(pmd);
1435			unsigned long offset;
1436
1437			if (pm->show_pfn) {
1438				offset = swp_offset(entry) +
1439					((addr & ~PMD_MASK) >> PAGE_SHIFT);
1440				frame = swp_type(entry) |
1441					(offset << MAX_SWAPFILES_SHIFT);
1442			}
1443			flags |= PM_SWAP;
1444			if (pmd_swp_soft_dirty(pmd))
1445				flags |= PM_SOFT_DIRTY;
1446			if (pmd_swp_uffd_wp(pmd))
1447				flags |= PM_UFFD_WP;
1448			VM_BUG_ON(!is_pmd_migration_entry(pmd));
1449			page = pfn_swap_entry_to_page(entry);
1450		}
1451#endif
1452
1453		if (page && page_mapcount(page) == 1)
1454			flags |= PM_MMAP_EXCLUSIVE;
1455
1456		for (; addr != end; addr += PAGE_SIZE) {
1457			pagemap_entry_t pme = make_pme(frame, flags);
1458
1459			err = add_to_pagemap(addr, &pme, pm);
1460			if (err)
1461				break;
1462			if (pm->show_pfn) {
1463				if (flags & PM_PRESENT)
1464					frame++;
1465				else if (flags & PM_SWAP)
1466					frame += (1 << MAX_SWAPFILES_SHIFT);
1467			}
1468		}
1469		spin_unlock(ptl);
1470		return err;
1471	}
1472
1473	if (pmd_trans_unstable(pmdp))
1474		return 0;
1475#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1476
1477	/*
1478	 * We can assume that @vma always points to a valid one and @end never
1479	 * goes beyond vma->vm_end.
1480	 */
1481	orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1482	for (; addr < end; pte++, addr += PAGE_SIZE) {
1483		pagemap_entry_t pme;
1484
1485		pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1486		err = add_to_pagemap(addr, &pme, pm);
1487		if (err)
1488			break;
1489	}
1490	pte_unmap_unlock(orig_pte, ptl);
1491
1492	cond_resched();
1493
1494	return err;
1495}
1496
1497#ifdef CONFIG_HUGETLB_PAGE
1498/* This function walks within one hugetlb entry in the single call */
1499static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1500				 unsigned long addr, unsigned long end,
1501				 struct mm_walk *walk)
1502{
1503	struct pagemapread *pm = walk->private;
1504	struct vm_area_struct *vma = walk->vma;
1505	u64 flags = 0, frame = 0;
1506	int err = 0;
1507	pte_t pte;
1508
1509	if (vma->vm_flags & VM_SOFTDIRTY)
1510		flags |= PM_SOFT_DIRTY;
1511
1512	pte = huge_ptep_get(ptep);
1513	if (pte_present(pte)) {
1514		struct page *page = pte_page(pte);
1515
1516		if (!PageAnon(page))
1517			flags |= PM_FILE;
1518
1519		if (page_mapcount(page) == 1)
1520			flags |= PM_MMAP_EXCLUSIVE;
1521
1522		flags |= PM_PRESENT;
1523		if (pm->show_pfn)
1524			frame = pte_pfn(pte) +
1525				((addr & ~hmask) >> PAGE_SHIFT);
1526	}
1527
1528	for (; addr != end; addr += PAGE_SIZE) {
1529		pagemap_entry_t pme = make_pme(frame, flags);
1530
1531		err = add_to_pagemap(addr, &pme, pm);
1532		if (err)
1533			return err;
1534		if (pm->show_pfn && (flags & PM_PRESENT))
1535			frame++;
1536	}
1537
1538	cond_resched();
1539
1540	return err;
1541}
1542#else
1543#define pagemap_hugetlb_range	NULL
1544#endif /* HUGETLB_PAGE */
1545
1546static const struct mm_walk_ops pagemap_ops = {
1547	.pmd_entry	= pagemap_pmd_range,
1548	.pte_hole	= pagemap_pte_hole,
1549	.hugetlb_entry	= pagemap_hugetlb_range,
1550};
1551
1552/*
1553 * /proc/pid/pagemap - an array mapping virtual pages to pfns
1554 *
1555 * For each page in the address space, this file contains one 64-bit entry
1556 * consisting of the following:
1557 *
1558 * Bits 0-54  page frame number (PFN) if present
1559 * Bits 0-4   swap type if swapped
1560 * Bits 5-54  swap offset if swapped
1561 * Bit  55    pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst)
1562 * Bit  56    page exclusively mapped
1563 * Bits 57-60 zero
1564 * Bit  61    page is file-page or shared-anon
1565 * Bit  62    page swapped
1566 * Bit  63    page present
1567 *
1568 * If the page is not present but in swap, then the PFN contains an
1569 * encoding of the swap file number and the page's offset into the
1570 * swap. Unmapped pages return a null PFN. This allows determining
1571 * precisely which pages are mapped (or in swap) and comparing mapped
1572 * pages between processes.
1573 *
1574 * Efficient users of this interface will use /proc/pid/maps to
1575 * determine which areas of memory are actually mapped and llseek to
1576 * skip over unmapped regions.
1577 */
1578static ssize_t pagemap_read(struct file *file, char __user *buf,
1579			    size_t count, loff_t *ppos)
1580{
1581	struct mm_struct *mm = file->private_data;
1582	struct pagemapread pm;
1583	unsigned long src;
1584	unsigned long svpfn;
1585	unsigned long start_vaddr;
1586	unsigned long end_vaddr;
1587	int ret = 0, copied = 0;
1588
1589	if (!mm || !mmget_not_zero(mm))
1590		goto out;
1591
1592	ret = -EINVAL;
1593	/* file position must be aligned */
1594	if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1595		goto out_mm;
1596
1597	ret = 0;
1598	if (!count)
1599		goto out_mm;
1600
1601	/* do not disclose physical addresses: attack vector */
1602	pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1603
1604	pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1605	pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL);
1606	ret = -ENOMEM;
1607	if (!pm.buffer)
1608		goto out_mm;
1609
1610	src = *ppos;
1611	svpfn = src / PM_ENTRY_BYTES;
1612	end_vaddr = mm->task_size;
1613
1614	/* watch out for wraparound */
1615	start_vaddr = end_vaddr;
1616	if (svpfn <= (ULONG_MAX >> PAGE_SHIFT))
1617		start_vaddr = untagged_addr(svpfn << PAGE_SHIFT);
1618
1619	/* Ensure the address is inside the task */
1620	if (start_vaddr > mm->task_size)
1621		start_vaddr = end_vaddr;
1622
1623	/*
1624	 * The odds are that this will stop walking way
1625	 * before end_vaddr, because the length of the
1626	 * user buffer is tracked in "pm", and the walk
1627	 * will stop when we hit the end of the buffer.
1628	 */
1629	ret = 0;
1630	while (count && (start_vaddr < end_vaddr)) {
1631		int len;
1632		unsigned long end;
1633
1634		pm.pos = 0;
1635		end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1636		/* overflow ? */
1637		if (end < start_vaddr || end > end_vaddr)
1638			end = end_vaddr;
1639		ret = mmap_read_lock_killable(mm);
1640		if (ret)
1641			goto out_free;
1642		ret = walk_page_range(mm, start_vaddr, end, &pagemap_ops, &pm);
1643		mmap_read_unlock(mm);
1644		start_vaddr = end;
1645
1646		len = min(count, PM_ENTRY_BYTES * pm.pos);
1647		if (copy_to_user(buf, pm.buffer, len)) {
1648			ret = -EFAULT;
1649			goto out_free;
1650		}
1651		copied += len;
1652		buf += len;
1653		count -= len;
1654	}
1655	*ppos += copied;
1656	if (!ret || ret == PM_END_OF_BUFFER)
1657		ret = copied;
1658
1659out_free:
1660	kfree(pm.buffer);
1661out_mm:
1662	mmput(mm);
1663out:
1664	return ret;
1665}
1666
1667static int pagemap_open(struct inode *inode, struct file *file)
1668{
1669	struct mm_struct *mm;
1670
1671	mm = proc_mem_open(inode, PTRACE_MODE_READ);
1672	if (IS_ERR(mm))
1673		return PTR_ERR(mm);
1674	file->private_data = mm;
1675	return 0;
1676}
1677
1678static int pagemap_release(struct inode *inode, struct file *file)
1679{
1680	struct mm_struct *mm = file->private_data;
1681
1682	if (mm)
1683		mmdrop(mm);
1684	return 0;
1685}
1686
1687const struct file_operations proc_pagemap_operations = {
1688	.llseek		= mem_lseek, /* borrow this */
1689	.read		= pagemap_read,
1690	.open		= pagemap_open,
1691	.release	= pagemap_release,
1692};
1693#endif /* CONFIG_PROC_PAGE_MONITOR */
1694
1695#ifdef CONFIG_NUMA
1696
1697struct numa_maps {
1698	unsigned long pages;
1699	unsigned long anon;
1700	unsigned long active;
1701	unsigned long writeback;
1702	unsigned long mapcount_max;
1703	unsigned long dirty;
1704	unsigned long swapcache;
1705	unsigned long node[MAX_NUMNODES];
1706};
1707
1708struct numa_maps_private {
1709	struct proc_maps_private proc_maps;
1710	struct numa_maps md;
1711};
1712
1713static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1714			unsigned long nr_pages)
1715{
1716	int count = page_mapcount(page);
1717
1718	md->pages += nr_pages;
1719	if (pte_dirty || PageDirty(page))
1720		md->dirty += nr_pages;
1721
1722	if (PageSwapCache(page))
1723		md->swapcache += nr_pages;
1724
1725	if (PageActive(page) || PageUnevictable(page))
1726		md->active += nr_pages;
1727
1728	if (PageWriteback(page))
1729		md->writeback += nr_pages;
1730
1731	if (PageAnon(page))
1732		md->anon += nr_pages;
1733
1734	if (count > md->mapcount_max)
1735		md->mapcount_max = count;
1736
1737	md->node[page_to_nid(page)] += nr_pages;
1738}
1739
1740static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1741		unsigned long addr)
1742{
1743	struct page *page;
1744	int nid;
1745
1746	if (!pte_present(pte))
1747		return NULL;
1748
1749	page = vm_normal_page(vma, addr, pte);
1750	if (!page)
1751		return NULL;
1752
1753	if (PageReserved(page))
1754		return NULL;
1755
1756	nid = page_to_nid(page);
1757	if (!node_isset(nid, node_states[N_MEMORY]))
1758		return NULL;
1759
1760	return page;
1761}
1762
1763#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1764static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1765					      struct vm_area_struct *vma,
1766					      unsigned long addr)
1767{
1768	struct page *page;
1769	int nid;
1770
1771	if (!pmd_present(pmd))
1772		return NULL;
1773
1774	page = vm_normal_page_pmd(vma, addr, pmd);
1775	if (!page)
1776		return NULL;
1777
1778	if (PageReserved(page))
1779		return NULL;
1780
1781	nid = page_to_nid(page);
1782	if (!node_isset(nid, node_states[N_MEMORY]))
1783		return NULL;
1784
1785	return page;
1786}
1787#endif
1788
1789static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1790		unsigned long end, struct mm_walk *walk)
1791{
1792	struct numa_maps *md = walk->private;
1793	struct vm_area_struct *vma = walk->vma;
1794	spinlock_t *ptl;
1795	pte_t *orig_pte;
1796	pte_t *pte;
1797
1798#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1799	ptl = pmd_trans_huge_lock(pmd, vma);
1800	if (ptl) {
1801		struct page *page;
1802
1803		page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1804		if (page)
1805			gather_stats(page, md, pmd_dirty(*pmd),
1806				     HPAGE_PMD_SIZE/PAGE_SIZE);
1807		spin_unlock(ptl);
1808		return 0;
1809	}
1810
1811	if (pmd_trans_unstable(pmd))
1812		return 0;
1813#endif
1814	orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1815	do {
1816		struct page *page = can_gather_numa_stats(*pte, vma, addr);
1817		if (!page)
1818			continue;
1819		gather_stats(page, md, pte_dirty(*pte), 1);
1820
1821	} while (pte++, addr += PAGE_SIZE, addr != end);
1822	pte_unmap_unlock(orig_pte, ptl);
1823	cond_resched();
1824	return 0;
1825}
1826#ifdef CONFIG_HUGETLB_PAGE
1827static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1828		unsigned long addr, unsigned long end, struct mm_walk *walk)
1829{
1830	pte_t huge_pte = huge_ptep_get(pte);
1831	struct numa_maps *md;
1832	struct page *page;
1833
1834	if (!pte_present(huge_pte))
1835		return 0;
1836
1837	page = pte_page(huge_pte);
1838	if (!page)
1839		return 0;
1840
1841	md = walk->private;
1842	gather_stats(page, md, pte_dirty(huge_pte), 1);
1843	return 0;
1844}
1845
1846#else
1847static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1848		unsigned long addr, unsigned long end, struct mm_walk *walk)
1849{
1850	return 0;
1851}
1852#endif
1853
1854static const struct mm_walk_ops show_numa_ops = {
1855	.hugetlb_entry = gather_hugetlb_stats,
1856	.pmd_entry = gather_pte_stats,
1857};
1858
1859/*
1860 * Display pages allocated per node and memory policy via /proc.
1861 */
1862static int show_numa_map(struct seq_file *m, void *v)
1863{
1864	struct numa_maps_private *numa_priv = m->private;
1865	struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1866	struct vm_area_struct *vma = v;
1867	struct numa_maps *md = &numa_priv->md;
1868	struct file *file = vma->vm_file;
1869	struct mm_struct *mm = vma->vm_mm;
1870	struct mempolicy *pol;
1871	char buffer[64];
1872	int nid;
1873
1874	if (!mm)
1875		return 0;
1876
1877	/* Ensure we start with an empty set of numa_maps statistics. */
1878	memset(md, 0, sizeof(*md));
1879
1880	pol = __get_vma_policy(vma, vma->vm_start);
1881	if (pol) {
1882		mpol_to_str(buffer, sizeof(buffer), pol);
1883		mpol_cond_put(pol);
1884	} else {
1885		mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1886	}
1887
1888	seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1889
1890	if (file) {
1891		seq_puts(m, " file=");
1892		seq_file_path(m, file, "\n\t= ");
1893	} else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1894		seq_puts(m, " heap");
1895	} else if (is_stack(vma)) {
1896		seq_puts(m, " stack");
1897	}
1898
1899	if (is_vm_hugetlb_page(vma))
1900		seq_puts(m, " huge");
1901
1902	/* mmap_lock is held by m_start */
1903	walk_page_vma(vma, &show_numa_ops, md);
1904
1905	if (!md->pages)
1906		goto out;
1907
1908	if (md->anon)
1909		seq_printf(m, " anon=%lu", md->anon);
1910
1911	if (md->dirty)
1912		seq_printf(m, " dirty=%lu", md->dirty);
1913
1914	if (md->pages != md->anon && md->pages != md->dirty)
1915		seq_printf(m, " mapped=%lu", md->pages);
1916
1917	if (md->mapcount_max > 1)
1918		seq_printf(m, " mapmax=%lu", md->mapcount_max);
1919
1920	if (md->swapcache)
1921		seq_printf(m, " swapcache=%lu", md->swapcache);
1922
1923	if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1924		seq_printf(m, " active=%lu", md->active);
1925
1926	if (md->writeback)
1927		seq_printf(m, " writeback=%lu", md->writeback);
1928
1929	for_each_node_state(nid, N_MEMORY)
1930		if (md->node[nid])
1931			seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1932
1933	seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1934out:
1935	seq_putc(m, '\n');
1936	return 0;
1937}
1938
1939static const struct seq_operations proc_pid_numa_maps_op = {
1940	.start  = m_start,
1941	.next   = m_next,
1942	.stop   = m_stop,
1943	.show   = show_numa_map,
1944};
1945
1946static int pid_numa_maps_open(struct inode *inode, struct file *file)
1947{
1948	return proc_maps_open(inode, file, &proc_pid_numa_maps_op,
1949				sizeof(struct numa_maps_private));
1950}
1951
1952const struct file_operations proc_pid_numa_maps_operations = {
1953	.open		= pid_numa_maps_open,
1954	.read		= seq_read,
1955	.llseek		= seq_lseek,
1956	.release	= proc_map_release,
1957};
1958
1959#endif /* CONFIG_NUMA */
1960