xref: /openbmc/linux/fs/proc/task_mmu.c (revision 68f436a8)
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>
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)
30 void 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 
82 unsigned long task_vsize(struct mm_struct *mm)
83 {
84 	return PAGE_SIZE * mm->total_vm;
85 }
86 
87 unsigned 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  */
104 static 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 }
113 static void release_task_mempolicy(struct proc_maps_private *priv)
114 {
115 	mpol_put(priv->task_mempolicy);
116 }
117 #else
118 static void hold_task_mempolicy(struct proc_maps_private *priv)
119 {
120 }
121 static void release_task_mempolicy(struct proc_maps_private *priv)
122 {
123 }
124 #endif
125 
126 static struct vm_area_struct *proc_get_vma(struct proc_maps_private *priv,
127 						loff_t *ppos)
128 {
129 	struct vm_area_struct *vma = vma_next(&priv->iter);
130 
131 	if (vma) {
132 		*ppos = vma->vm_start;
133 	} else {
134 		*ppos = -2UL;
135 		vma = get_gate_vma(priv->mm);
136 	}
137 
138 	return vma;
139 }
140 
141 static void *m_start(struct seq_file *m, loff_t *ppos)
142 {
143 	struct proc_maps_private *priv = m->private;
144 	unsigned long last_addr = *ppos;
145 	struct mm_struct *mm;
146 
147 	/* See m_next(). Zero at the start or after lseek. */
148 	if (last_addr == -1UL)
149 		return NULL;
150 
151 	priv->task = get_proc_task(priv->inode);
152 	if (!priv->task)
153 		return ERR_PTR(-ESRCH);
154 
155 	mm = priv->mm;
156 	if (!mm || !mmget_not_zero(mm)) {
157 		put_task_struct(priv->task);
158 		priv->task = NULL;
159 		return NULL;
160 	}
161 
162 	if (mmap_read_lock_killable(mm)) {
163 		mmput(mm);
164 		put_task_struct(priv->task);
165 		priv->task = NULL;
166 		return ERR_PTR(-EINTR);
167 	}
168 
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);
173 
174 	return proc_get_vma(priv, ppos);
175 }
176 
177 static void *m_next(struct seq_file *m, void *v, loff_t *ppos)
178 {
179 	if (*ppos == -2UL) {
180 		*ppos = -1UL;
181 		return NULL;
182 	}
183 	return proc_get_vma(m->private, ppos);
184 }
185 
186 static void m_stop(struct seq_file *m, void *v)
187 {
188 	struct proc_maps_private *priv = m->private;
189 	struct mm_struct *mm = priv->mm;
190 
191 	if (!priv->task)
192 		return;
193 
194 	release_task_mempolicy(priv);
195 	mmap_read_unlock(mm);
196 	mmput(mm);
197 	put_task_struct(priv->task);
198 	priv->task = NULL;
199 }
200 
201 static int proc_maps_open(struct inode *inode, struct file *file,
202 			const struct seq_operations *ops, int psize)
203 {
204 	struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
205 
206 	if (!priv)
207 		return -ENOMEM;
208 
209 	priv->inode = inode;
210 	priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
211 	if (IS_ERR(priv->mm)) {
212 		int err = PTR_ERR(priv->mm);
213 
214 		seq_release_private(inode, file);
215 		return err;
216 	}
217 
218 	return 0;
219 }
220 
221 static int proc_map_release(struct inode *inode, struct file *file)
222 {
223 	struct seq_file *seq = file->private_data;
224 	struct proc_maps_private *priv = seq->private;
225 
226 	if (priv->mm)
227 		mmdrop(priv->mm);
228 
229 	return seq_release_private(inode, file);
230 }
231 
232 static int do_maps_open(struct inode *inode, struct file *file,
233 			const struct seq_operations *ops)
234 {
235 	return proc_maps_open(inode, file, ops,
236 				sizeof(struct proc_maps_private));
237 }
238 
239 /*
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.
242  */
243 static int is_stack(struct vm_area_struct *vma)
244 {
245 	/*
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
248 	 * languages like Go.
249 	 */
250 	return vma->vm_start <= vma->vm_mm->start_stack &&
251 		vma->vm_end >= vma->vm_mm->start_stack;
252 }
253 
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)
258 {
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);
262 	seq_putc(m, ' ');
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);
271 	seq_putc(m, ' ');
272 }
273 
274 static void
275 show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
276 {
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;
284 	dev_t dev = 0;
285 	const char *name = NULL;
286 
287 	if (file) {
288 		struct inode *inode = file_inode(vma->vm_file);
289 		dev = inode->i_sb->s_dev;
290 		ino = inode->i_ino;
291 		pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
292 	}
293 
294 	start = vma->vm_start;
295 	end = vma->vm_end;
296 	show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino);
297 	if (mm)
298 		anon_name = anon_vma_name(vma);
299 
300 	/*
301 	 * Print the dentry name for named mappings, and a
302 	 * special [heap] marker for the heap:
303 	 */
304 	if (file) {
305 		seq_pad(m, ' ');
306 		/*
307 		 * If user named this anon shared memory via
308 		 * prctl(PR_SET_VMA ..., use the provided name.
309 		 */
310 		if (anon_name)
311 			seq_printf(m, "[anon_shmem:%s]", anon_name->name);
312 		else
313 			seq_file_path(m, file, "\n");
314 		goto done;
315 	}
316 
317 	if (vma->vm_ops && vma->vm_ops->name) {
318 		name = vma->vm_ops->name(vma);
319 		if (name)
320 			goto done;
321 	}
322 
323 	name = arch_vma_name(vma);
324 	if (!name) {
325 		if (!mm) {
326 			name = "[vdso]";
327 			goto done;
328 		}
329 
330 		if (vma->vm_start <= mm->brk &&
331 		    vma->vm_end >= mm->start_brk) {
332 			name = "[heap]";
333 			goto done;
334 		}
335 
336 		if (is_stack(vma)) {
337 			name = "[stack]";
338 			goto done;
339 		}
340 
341 		if (anon_name) {
342 			seq_pad(m, ' ');
343 			seq_printf(m, "[anon:%s]", anon_name->name);
344 		}
345 	}
346 
347 done:
348 	if (name) {
349 		seq_pad(m, ' ');
350 		seq_puts(m, name);
351 	}
352 	seq_putc(m, '\n');
353 }
354 
355 static int show_map(struct seq_file *m, void *v)
356 {
357 	show_map_vma(m, v);
358 	return 0;
359 }
360 
361 static const struct seq_operations proc_pid_maps_op = {
362 	.start	= m_start,
363 	.next	= m_next,
364 	.stop	= m_stop,
365 	.show	= show_map
366 };
367 
368 static int pid_maps_open(struct inode *inode, struct file *file)
369 {
370 	return do_maps_open(inode, file, &proc_pid_maps_op);
371 }
372 
373 const struct file_operations proc_pid_maps_operations = {
374 	.open		= pid_maps_open,
375 	.read		= seq_read,
376 	.llseek		= seq_lseek,
377 	.release	= proc_map_release,
378 };
379 
380 /*
381  * Proportional Set Size(PSS): my share of RSS.
382  *
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.
387  *
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.
391  *
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.
396  */
397 #define PSS_SHIFT 12
398 
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;
412 	unsigned long swap;
413 	unsigned long shared_hugetlb;
414 	unsigned long private_hugetlb;
415 	u64 pss;
416 	u64 pss_anon;
417 	u64 pss_file;
418 	u64 pss_shmem;
419 	u64 pss_dirty;
420 	u64 pss_locked;
421 	u64 swap_pss;
422 };
423 
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)
427 {
428 	mss->pss += pss;
429 
430 	if (PageAnon(page))
431 		mss->pss_anon += pss;
432 	else if (PageSwapBacked(page))
433 		mss->pss_shmem += pss;
434 	else
435 		mss->pss_file += pss;
436 
437 	if (locked)
438 		mss->pss_locked += pss;
439 
440 	if (dirty || PageDirty(page)) {
441 		mss->pss_dirty += pss;
442 		if (private)
443 			mss->private_dirty += size;
444 		else
445 			mss->shared_dirty += size;
446 	} else {
447 		if (private)
448 			mss->private_clean += size;
449 		else
450 			mss->shared_clean += size;
451 	}
452 }
453 
454 static void smaps_account(struct mem_size_stats *mss, struct page *page,
455 		bool compound, bool young, bool dirty, bool locked,
456 		bool migration)
457 {
458 	int i, nr = compound ? compound_nr(page) : 1;
459 	unsigned long size = nr * PAGE_SIZE;
460 
461 	/*
462 	 * First accumulate quantities that depend only on |size| and the type
463 	 * of the compound page.
464 	 */
465 	if (PageAnon(page)) {
466 		mss->anonymous += size;
467 		if (!PageSwapBacked(page) && !dirty && !PageDirty(page))
468 			mss->lazyfree += size;
469 	}
470 
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;
475 
476 	/*
477 	 * Then accumulate quantities that may depend on sharing, or that may
478 	 * differ page-by-page.
479 	 *
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
482 	 * page_count().
483 	 *
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
489 	 * as mapcount == 1.
490 	 */
491 	if ((page_count(page) == 1) || migration) {
492 		smaps_page_accumulate(mss, page, size, size << PSS_SHIFT, dirty,
493 			locked, true);
494 		return;
495 	}
496 	for (i = 0; i < nr; i++, page++) {
497 		int mapcount = page_mapcount(page);
498 		unsigned long pss = PAGE_SIZE << PSS_SHIFT;
499 		if (mapcount >= 2)
500 			pss /= mapcount;
501 		smaps_page_accumulate(mss, page, PAGE_SIZE, pss, dirty, locked,
502 				      mapcount < 2);
503 	}
504 }
505 
506 #ifdef CONFIG_SHMEM
507 static int smaps_pte_hole(unsigned long addr, unsigned long end,
508 			  __always_unused int depth, struct mm_walk *walk)
509 {
510 	struct mem_size_stats *mss = walk->private;
511 	struct vm_area_struct *vma = walk->vma;
512 
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));
516 
517 	return 0;
518 }
519 #else
520 #define smaps_pte_hole		NULL
521 #endif /* CONFIG_SHMEM */
522 
523 static void smaps_pte_hole_lookup(unsigned long addr, struct mm_walk *walk)
524 {
525 #ifdef CONFIG_SHMEM
526 	if (walk->ops->pte_hole) {
527 		/* depth is not used */
528 		smaps_pte_hole(addr, addr + PAGE_SIZE, 0, walk);
529 	}
530 #endif
531 }
532 
533 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
534 		struct mm_walk *walk)
535 {
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;
541 	pte_t ptent = ptep_get(pte);
542 
543 	if (pte_present(ptent)) {
544 		page = vm_normal_page(vma, addr, ptent);
545 		young = pte_young(ptent);
546 		dirty = pte_dirty(ptent);
547 	} else if (is_swap_pte(ptent)) {
548 		swp_entry_t swpent = pte_to_swp_entry(ptent);
549 
550 		if (!non_swap_entry(swpent)) {
551 			int mapcount;
552 
553 			mss->swap += PAGE_SIZE;
554 			mapcount = swp_swapcount(swpent);
555 			if (mapcount >= 2) {
556 				u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
557 
558 				do_div(pss_delta, mapcount);
559 				mss->swap_pss += pss_delta;
560 			} else {
561 				mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
562 			}
563 		} else if (is_pfn_swap_entry(swpent)) {
564 			if (is_migration_entry(swpent))
565 				migration = true;
566 			page = pfn_swap_entry_to_page(swpent);
567 		}
568 	} else {
569 		smaps_pte_hole_lookup(addr, walk);
570 		return;
571 	}
572 
573 	if (!page)
574 		return;
575 
576 	smaps_account(mss, page, false, young, dirty, locked, migration);
577 }
578 
579 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
580 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
581 		struct mm_walk *walk)
582 {
583 	struct mem_size_stats *mss = walk->private;
584 	struct vm_area_struct *vma = walk->vma;
585 	bool locked = !!(vma->vm_flags & VM_LOCKED);
586 	struct page *page = NULL;
587 	bool migration = false;
588 
589 	if (pmd_present(*pmd)) {
590 		/* FOLL_DUMP will return -EFAULT on huge zero page */
591 		page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
592 	} else if (unlikely(thp_migration_supported() && is_swap_pmd(*pmd))) {
593 		swp_entry_t entry = pmd_to_swp_entry(*pmd);
594 
595 		if (is_migration_entry(entry)) {
596 			migration = true;
597 			page = pfn_swap_entry_to_page(entry);
598 		}
599 	}
600 	if (IS_ERR_OR_NULL(page))
601 		return;
602 	if (PageAnon(page))
603 		mss->anonymous_thp += HPAGE_PMD_SIZE;
604 	else if (PageSwapBacked(page))
605 		mss->shmem_thp += HPAGE_PMD_SIZE;
606 	else if (is_zone_device_page(page))
607 		/* pass */;
608 	else
609 		mss->file_thp += HPAGE_PMD_SIZE;
610 
611 	smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd),
612 		      locked, migration);
613 }
614 #else
615 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
616 		struct mm_walk *walk)
617 {
618 }
619 #endif
620 
621 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
622 			   struct mm_walk *walk)
623 {
624 	struct vm_area_struct *vma = walk->vma;
625 	pte_t *pte;
626 	spinlock_t *ptl;
627 
628 	ptl = pmd_trans_huge_lock(pmd, vma);
629 	if (ptl) {
630 		smaps_pmd_entry(pmd, addr, walk);
631 		spin_unlock(ptl);
632 		goto out;
633 	}
634 
635 	pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
636 	if (!pte) {
637 		walk->action = ACTION_AGAIN;
638 		return 0;
639 	}
640 	for (; addr != end; pte++, addr += PAGE_SIZE)
641 		smaps_pte_entry(pte, addr, walk);
642 	pte_unmap_unlock(pte - 1, ptl);
643 out:
644 	cond_resched();
645 	return 0;
646 }
647 
648 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
649 {
650 	/*
651 	 * Don't forget to update Documentation/ on changes.
652 	 */
653 	static const char mnemonics[BITS_PER_LONG][2] = {
654 		/*
655 		 * In case if we meet a flag we don't know about.
656 		 */
657 		[0 ... (BITS_PER_LONG-1)] = "??",
658 
659 		[ilog2(VM_READ)]	= "rd",
660 		[ilog2(VM_WRITE)]	= "wr",
661 		[ilog2(VM_EXEC)]	= "ex",
662 		[ilog2(VM_SHARED)]	= "sh",
663 		[ilog2(VM_MAYREAD)]	= "mr",
664 		[ilog2(VM_MAYWRITE)]	= "mw",
665 		[ilog2(VM_MAYEXEC)]	= "me",
666 		[ilog2(VM_MAYSHARE)]	= "ms",
667 		[ilog2(VM_GROWSDOWN)]	= "gd",
668 		[ilog2(VM_PFNMAP)]	= "pf",
669 		[ilog2(VM_LOCKED)]	= "lo",
670 		[ilog2(VM_IO)]		= "io",
671 		[ilog2(VM_SEQ_READ)]	= "sr",
672 		[ilog2(VM_RAND_READ)]	= "rr",
673 		[ilog2(VM_DONTCOPY)]	= "dc",
674 		[ilog2(VM_DONTEXPAND)]	= "de",
675 		[ilog2(VM_LOCKONFAULT)]	= "lf",
676 		[ilog2(VM_ACCOUNT)]	= "ac",
677 		[ilog2(VM_NORESERVE)]	= "nr",
678 		[ilog2(VM_HUGETLB)]	= "ht",
679 		[ilog2(VM_SYNC)]	= "sf",
680 		[ilog2(VM_ARCH_1)]	= "ar",
681 		[ilog2(VM_WIPEONFORK)]	= "wf",
682 		[ilog2(VM_DONTDUMP)]	= "dd",
683 #ifdef CONFIG_ARM64_BTI
684 		[ilog2(VM_ARM64_BTI)]	= "bt",
685 #endif
686 #ifdef CONFIG_MEM_SOFT_DIRTY
687 		[ilog2(VM_SOFTDIRTY)]	= "sd",
688 #endif
689 		[ilog2(VM_MIXEDMAP)]	= "mm",
690 		[ilog2(VM_HUGEPAGE)]	= "hg",
691 		[ilog2(VM_NOHUGEPAGE)]	= "nh",
692 		[ilog2(VM_MERGEABLE)]	= "mg",
693 		[ilog2(VM_UFFD_MISSING)]= "um",
694 		[ilog2(VM_UFFD_WP)]	= "uw",
695 #ifdef CONFIG_ARM64_MTE
696 		[ilog2(VM_MTE)]		= "mt",
697 		[ilog2(VM_MTE_ALLOWED)]	= "",
698 #endif
699 #ifdef CONFIG_ARCH_HAS_PKEYS
700 		/* These come out via ProtectionKey: */
701 		[ilog2(VM_PKEY_BIT0)]	= "",
702 		[ilog2(VM_PKEY_BIT1)]	= "",
703 		[ilog2(VM_PKEY_BIT2)]	= "",
704 		[ilog2(VM_PKEY_BIT3)]	= "",
705 #if VM_PKEY_BIT4
706 		[ilog2(VM_PKEY_BIT4)]	= "",
707 #endif
708 #endif /* CONFIG_ARCH_HAS_PKEYS */
709 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_MINOR
710 		[ilog2(VM_UFFD_MINOR)]	= "ui",
711 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_MINOR */
712 	};
713 	size_t i;
714 
715 	seq_puts(m, "VmFlags: ");
716 	for (i = 0; i < BITS_PER_LONG; i++) {
717 		if (!mnemonics[i][0])
718 			continue;
719 		if (vma->vm_flags & (1UL << i)) {
720 			seq_putc(m, mnemonics[i][0]);
721 			seq_putc(m, mnemonics[i][1]);
722 			seq_putc(m, ' ');
723 		}
724 	}
725 	seq_putc(m, '\n');
726 }
727 
728 #ifdef CONFIG_HUGETLB_PAGE
729 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
730 				 unsigned long addr, unsigned long end,
731 				 struct mm_walk *walk)
732 {
733 	struct mem_size_stats *mss = walk->private;
734 	struct vm_area_struct *vma = walk->vma;
735 	struct page *page = NULL;
736 	pte_t ptent = ptep_get(pte);
737 
738 	if (pte_present(ptent)) {
739 		page = vm_normal_page(vma, addr, ptent);
740 	} else if (is_swap_pte(ptent)) {
741 		swp_entry_t swpent = pte_to_swp_entry(ptent);
742 
743 		if (is_pfn_swap_entry(swpent))
744 			page = pfn_swap_entry_to_page(swpent);
745 	}
746 	if (page) {
747 		if (page_mapcount(page) >= 2 || hugetlb_pmd_shared(pte))
748 			mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
749 		else
750 			mss->private_hugetlb += huge_page_size(hstate_vma(vma));
751 	}
752 	return 0;
753 }
754 #else
755 #define smaps_hugetlb_range	NULL
756 #endif /* HUGETLB_PAGE */
757 
758 static const struct mm_walk_ops smaps_walk_ops = {
759 	.pmd_entry		= smaps_pte_range,
760 	.hugetlb_entry		= smaps_hugetlb_range,
761 };
762 
763 static const struct mm_walk_ops smaps_shmem_walk_ops = {
764 	.pmd_entry		= smaps_pte_range,
765 	.hugetlb_entry		= smaps_hugetlb_range,
766 	.pte_hole		= smaps_pte_hole,
767 };
768 
769 /*
770  * Gather mem stats from @vma with the indicated beginning
771  * address @start, and keep them in @mss.
772  *
773  * Use vm_start of @vma as the beginning address if @start is 0.
774  */
775 static void smap_gather_stats(struct vm_area_struct *vma,
776 		struct mem_size_stats *mss, unsigned long start)
777 {
778 	const struct mm_walk_ops *ops = &smaps_walk_ops;
779 
780 	/* Invalid start */
781 	if (start >= vma->vm_end)
782 		return;
783 
784 	if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
785 		/*
786 		 * For shared or readonly shmem mappings we know that all
787 		 * swapped out pages belong to the shmem object, and we can
788 		 * obtain the swap value much more efficiently. For private
789 		 * writable mappings, we might have COW pages that are
790 		 * not affected by the parent swapped out pages of the shmem
791 		 * object, so we have to distinguish them during the page walk.
792 		 * Unless we know that the shmem object (or the part mapped by
793 		 * our VMA) has no swapped out pages at all.
794 		 */
795 		unsigned long shmem_swapped = shmem_swap_usage(vma);
796 
797 		if (!start && (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
798 					!(vma->vm_flags & VM_WRITE))) {
799 			mss->swap += shmem_swapped;
800 		} else {
801 			ops = &smaps_shmem_walk_ops;
802 		}
803 	}
804 
805 	/* mmap_lock is held in m_start */
806 	if (!start)
807 		walk_page_vma(vma, ops, mss);
808 	else
809 		walk_page_range(vma->vm_mm, start, vma->vm_end, ops, mss);
810 }
811 
812 #define SEQ_PUT_DEC(str, val) \
813 		seq_put_decimal_ull_width(m, str, (val) >> 10, 8)
814 
815 /* Show the contents common for smaps and smaps_rollup */
816 static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss,
817 	bool rollup_mode)
818 {
819 	SEQ_PUT_DEC("Rss:            ", mss->resident);
820 	SEQ_PUT_DEC(" kB\nPss:            ", mss->pss >> PSS_SHIFT);
821 	SEQ_PUT_DEC(" kB\nPss_Dirty:      ", mss->pss_dirty >> PSS_SHIFT);
822 	if (rollup_mode) {
823 		/*
824 		 * These are meaningful only for smaps_rollup, otherwise two of
825 		 * them are zero, and the other one is the same as Pss.
826 		 */
827 		SEQ_PUT_DEC(" kB\nPss_Anon:       ",
828 			mss->pss_anon >> PSS_SHIFT);
829 		SEQ_PUT_DEC(" kB\nPss_File:       ",
830 			mss->pss_file >> PSS_SHIFT);
831 		SEQ_PUT_DEC(" kB\nPss_Shmem:      ",
832 			mss->pss_shmem >> PSS_SHIFT);
833 	}
834 	SEQ_PUT_DEC(" kB\nShared_Clean:   ", mss->shared_clean);
835 	SEQ_PUT_DEC(" kB\nShared_Dirty:   ", mss->shared_dirty);
836 	SEQ_PUT_DEC(" kB\nPrivate_Clean:  ", mss->private_clean);
837 	SEQ_PUT_DEC(" kB\nPrivate_Dirty:  ", mss->private_dirty);
838 	SEQ_PUT_DEC(" kB\nReferenced:     ", mss->referenced);
839 	SEQ_PUT_DEC(" kB\nAnonymous:      ", mss->anonymous);
840 	SEQ_PUT_DEC(" kB\nLazyFree:       ", mss->lazyfree);
841 	SEQ_PUT_DEC(" kB\nAnonHugePages:  ", mss->anonymous_thp);
842 	SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp);
843 	SEQ_PUT_DEC(" kB\nFilePmdMapped:  ", mss->file_thp);
844 	SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb);
845 	seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ",
846 				  mss->private_hugetlb >> 10, 7);
847 	SEQ_PUT_DEC(" kB\nSwap:           ", mss->swap);
848 	SEQ_PUT_DEC(" kB\nSwapPss:        ",
849 					mss->swap_pss >> PSS_SHIFT);
850 	SEQ_PUT_DEC(" kB\nLocked:         ",
851 					mss->pss_locked >> PSS_SHIFT);
852 	seq_puts(m, " kB\n");
853 }
854 
855 static int show_smap(struct seq_file *m, void *v)
856 {
857 	struct vm_area_struct *vma = v;
858 	struct mem_size_stats mss;
859 
860 	memset(&mss, 0, sizeof(mss));
861 
862 	smap_gather_stats(vma, &mss, 0);
863 
864 	show_map_vma(m, vma);
865 
866 	SEQ_PUT_DEC("Size:           ", vma->vm_end - vma->vm_start);
867 	SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma));
868 	SEQ_PUT_DEC(" kB\nMMUPageSize:    ", vma_mmu_pagesize(vma));
869 	seq_puts(m, " kB\n");
870 
871 	__show_smap(m, &mss, false);
872 
873 	seq_printf(m, "THPeligible:    %d\n",
874 		   hugepage_vma_check(vma, vma->vm_flags, true, false, true));
875 
876 	if (arch_pkeys_enabled())
877 		seq_printf(m, "ProtectionKey:  %8u\n", vma_pkey(vma));
878 	show_smap_vma_flags(m, vma);
879 
880 	return 0;
881 }
882 
883 static int show_smaps_rollup(struct seq_file *m, void *v)
884 {
885 	struct proc_maps_private *priv = m->private;
886 	struct mem_size_stats mss;
887 	struct mm_struct *mm = priv->mm;
888 	struct vm_area_struct *vma;
889 	unsigned long vma_start = 0, last_vma_end = 0;
890 	int ret = 0;
891 	VMA_ITERATOR(vmi, mm, 0);
892 
893 	priv->task = get_proc_task(priv->inode);
894 	if (!priv->task)
895 		return -ESRCH;
896 
897 	if (!mm || !mmget_not_zero(mm)) {
898 		ret = -ESRCH;
899 		goto out_put_task;
900 	}
901 
902 	memset(&mss, 0, sizeof(mss));
903 
904 	ret = mmap_read_lock_killable(mm);
905 	if (ret)
906 		goto out_put_mm;
907 
908 	hold_task_mempolicy(priv);
909 	vma = vma_next(&vmi);
910 
911 	if (unlikely(!vma))
912 		goto empty_set;
913 
914 	vma_start = vma->vm_start;
915 	do {
916 		smap_gather_stats(vma, &mss, 0);
917 		last_vma_end = vma->vm_end;
918 
919 		/*
920 		 * Release mmap_lock temporarily if someone wants to
921 		 * access it for write request.
922 		 */
923 		if (mmap_lock_is_contended(mm)) {
924 			vma_iter_invalidate(&vmi);
925 			mmap_read_unlock(mm);
926 			ret = mmap_read_lock_killable(mm);
927 			if (ret) {
928 				release_task_mempolicy(priv);
929 				goto out_put_mm;
930 			}
931 
932 			/*
933 			 * After dropping the lock, there are four cases to
934 			 * consider. See the following example for explanation.
935 			 *
936 			 *   +------+------+-----------+
937 			 *   | VMA1 | VMA2 | VMA3      |
938 			 *   +------+------+-----------+
939 			 *   |      |      |           |
940 			 *  4k     8k     16k         400k
941 			 *
942 			 * Suppose we drop the lock after reading VMA2 due to
943 			 * contention, then we get:
944 			 *
945 			 *	last_vma_end = 16k
946 			 *
947 			 * 1) VMA2 is freed, but VMA3 exists:
948 			 *
949 			 *    vma_next(vmi) will return VMA3.
950 			 *    In this case, just continue from VMA3.
951 			 *
952 			 * 2) VMA2 still exists:
953 			 *
954 			 *    vma_next(vmi) will return VMA3.
955 			 *    In this case, just continue from VMA3.
956 			 *
957 			 * 3) No more VMAs can be found:
958 			 *
959 			 *    vma_next(vmi) will return NULL.
960 			 *    No more things to do, just break.
961 			 *
962 			 * 4) (last_vma_end - 1) is the middle of a vma (VMA'):
963 			 *
964 			 *    vma_next(vmi) will return VMA' whose range
965 			 *    contains last_vma_end.
966 			 *    Iterate VMA' from last_vma_end.
967 			 */
968 			vma = vma_next(&vmi);
969 			/* Case 3 above */
970 			if (!vma)
971 				break;
972 
973 			/* Case 1 and 2 above */
974 			if (vma->vm_start >= last_vma_end)
975 				continue;
976 
977 			/* Case 4 above */
978 			if (vma->vm_end > last_vma_end)
979 				smap_gather_stats(vma, &mss, last_vma_end);
980 		}
981 	} for_each_vma(vmi, vma);
982 
983 empty_set:
984 	show_vma_header_prefix(m, vma_start, last_vma_end, 0, 0, 0, 0);
985 	seq_pad(m, ' ');
986 	seq_puts(m, "[rollup]\n");
987 
988 	__show_smap(m, &mss, true);
989 
990 	release_task_mempolicy(priv);
991 	mmap_read_unlock(mm);
992 
993 out_put_mm:
994 	mmput(mm);
995 out_put_task:
996 	put_task_struct(priv->task);
997 	priv->task = NULL;
998 
999 	return ret;
1000 }
1001 #undef SEQ_PUT_DEC
1002 
1003 static const struct seq_operations proc_pid_smaps_op = {
1004 	.start	= m_start,
1005 	.next	= m_next,
1006 	.stop	= m_stop,
1007 	.show	= show_smap
1008 };
1009 
1010 static int pid_smaps_open(struct inode *inode, struct file *file)
1011 {
1012 	return do_maps_open(inode, file, &proc_pid_smaps_op);
1013 }
1014 
1015 static int smaps_rollup_open(struct inode *inode, struct file *file)
1016 {
1017 	int ret;
1018 	struct proc_maps_private *priv;
1019 
1020 	priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT);
1021 	if (!priv)
1022 		return -ENOMEM;
1023 
1024 	ret = single_open(file, show_smaps_rollup, priv);
1025 	if (ret)
1026 		goto out_free;
1027 
1028 	priv->inode = inode;
1029 	priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
1030 	if (IS_ERR(priv->mm)) {
1031 		ret = PTR_ERR(priv->mm);
1032 
1033 		single_release(inode, file);
1034 		goto out_free;
1035 	}
1036 
1037 	return 0;
1038 
1039 out_free:
1040 	kfree(priv);
1041 	return ret;
1042 }
1043 
1044 static int smaps_rollup_release(struct inode *inode, struct file *file)
1045 {
1046 	struct seq_file *seq = file->private_data;
1047 	struct proc_maps_private *priv = seq->private;
1048 
1049 	if (priv->mm)
1050 		mmdrop(priv->mm);
1051 
1052 	kfree(priv);
1053 	return single_release(inode, file);
1054 }
1055 
1056 const struct file_operations proc_pid_smaps_operations = {
1057 	.open		= pid_smaps_open,
1058 	.read		= seq_read,
1059 	.llseek		= seq_lseek,
1060 	.release	= proc_map_release,
1061 };
1062 
1063 const struct file_operations proc_pid_smaps_rollup_operations = {
1064 	.open		= smaps_rollup_open,
1065 	.read		= seq_read,
1066 	.llseek		= seq_lseek,
1067 	.release	= smaps_rollup_release,
1068 };
1069 
1070 enum clear_refs_types {
1071 	CLEAR_REFS_ALL = 1,
1072 	CLEAR_REFS_ANON,
1073 	CLEAR_REFS_MAPPED,
1074 	CLEAR_REFS_SOFT_DIRTY,
1075 	CLEAR_REFS_MM_HIWATER_RSS,
1076 	CLEAR_REFS_LAST,
1077 };
1078 
1079 struct clear_refs_private {
1080 	enum clear_refs_types type;
1081 };
1082 
1083 #ifdef CONFIG_MEM_SOFT_DIRTY
1084 
1085 static inline bool pte_is_pinned(struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1086 {
1087 	struct page *page;
1088 
1089 	if (!pte_write(pte))
1090 		return false;
1091 	if (!is_cow_mapping(vma->vm_flags))
1092 		return false;
1093 	if (likely(!test_bit(MMF_HAS_PINNED, &vma->vm_mm->flags)))
1094 		return false;
1095 	page = vm_normal_page(vma, addr, pte);
1096 	if (!page)
1097 		return false;
1098 	return page_maybe_dma_pinned(page);
1099 }
1100 
1101 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1102 		unsigned long addr, pte_t *pte)
1103 {
1104 	/*
1105 	 * The soft-dirty tracker uses #PF-s to catch writes
1106 	 * to pages, so write-protect the pte as well. See the
1107 	 * Documentation/admin-guide/mm/soft-dirty.rst for full description
1108 	 * of how soft-dirty works.
1109 	 */
1110 	pte_t ptent = ptep_get(pte);
1111 
1112 	if (pte_present(ptent)) {
1113 		pte_t old_pte;
1114 
1115 		if (pte_is_pinned(vma, addr, ptent))
1116 			return;
1117 		old_pte = ptep_modify_prot_start(vma, addr, pte);
1118 		ptent = pte_wrprotect(old_pte);
1119 		ptent = pte_clear_soft_dirty(ptent);
1120 		ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent);
1121 	} else if (is_swap_pte(ptent)) {
1122 		ptent = pte_swp_clear_soft_dirty(ptent);
1123 		set_pte_at(vma->vm_mm, addr, pte, ptent);
1124 	}
1125 }
1126 #else
1127 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1128 		unsigned long addr, pte_t *pte)
1129 {
1130 }
1131 #endif
1132 
1133 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
1134 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1135 		unsigned long addr, pmd_t *pmdp)
1136 {
1137 	pmd_t old, pmd = *pmdp;
1138 
1139 	if (pmd_present(pmd)) {
1140 		/* See comment in change_huge_pmd() */
1141 		old = pmdp_invalidate(vma, addr, pmdp);
1142 		if (pmd_dirty(old))
1143 			pmd = pmd_mkdirty(pmd);
1144 		if (pmd_young(old))
1145 			pmd = pmd_mkyoung(pmd);
1146 
1147 		pmd = pmd_wrprotect(pmd);
1148 		pmd = pmd_clear_soft_dirty(pmd);
1149 
1150 		set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1151 	} else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1152 		pmd = pmd_swp_clear_soft_dirty(pmd);
1153 		set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1154 	}
1155 }
1156 #else
1157 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1158 		unsigned long addr, pmd_t *pmdp)
1159 {
1160 }
1161 #endif
1162 
1163 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
1164 				unsigned long end, struct mm_walk *walk)
1165 {
1166 	struct clear_refs_private *cp = walk->private;
1167 	struct vm_area_struct *vma = walk->vma;
1168 	pte_t *pte, ptent;
1169 	spinlock_t *ptl;
1170 	struct page *page;
1171 
1172 	ptl = pmd_trans_huge_lock(pmd, vma);
1173 	if (ptl) {
1174 		if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1175 			clear_soft_dirty_pmd(vma, addr, pmd);
1176 			goto out;
1177 		}
1178 
1179 		if (!pmd_present(*pmd))
1180 			goto out;
1181 
1182 		page = pmd_page(*pmd);
1183 
1184 		/* Clear accessed and referenced bits. */
1185 		pmdp_test_and_clear_young(vma, addr, pmd);
1186 		test_and_clear_page_young(page);
1187 		ClearPageReferenced(page);
1188 out:
1189 		spin_unlock(ptl);
1190 		return 0;
1191 	}
1192 
1193 	pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1194 	if (!pte) {
1195 		walk->action = ACTION_AGAIN;
1196 		return 0;
1197 	}
1198 	for (; addr != end; pte++, addr += PAGE_SIZE) {
1199 		ptent = ptep_get(pte);
1200 
1201 		if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1202 			clear_soft_dirty(vma, addr, pte);
1203 			continue;
1204 		}
1205 
1206 		if (!pte_present(ptent))
1207 			continue;
1208 
1209 		page = vm_normal_page(vma, addr, ptent);
1210 		if (!page)
1211 			continue;
1212 
1213 		/* Clear accessed and referenced bits. */
1214 		ptep_test_and_clear_young(vma, addr, pte);
1215 		test_and_clear_page_young(page);
1216 		ClearPageReferenced(page);
1217 	}
1218 	pte_unmap_unlock(pte - 1, ptl);
1219 	cond_resched();
1220 	return 0;
1221 }
1222 
1223 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1224 				struct mm_walk *walk)
1225 {
1226 	struct clear_refs_private *cp = walk->private;
1227 	struct vm_area_struct *vma = walk->vma;
1228 
1229 	if (vma->vm_flags & VM_PFNMAP)
1230 		return 1;
1231 
1232 	/*
1233 	 * Writing 1 to /proc/pid/clear_refs affects all pages.
1234 	 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1235 	 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1236 	 * Writing 4 to /proc/pid/clear_refs affects all pages.
1237 	 */
1238 	if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1239 		return 1;
1240 	if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1241 		return 1;
1242 	return 0;
1243 }
1244 
1245 static const struct mm_walk_ops clear_refs_walk_ops = {
1246 	.pmd_entry		= clear_refs_pte_range,
1247 	.test_walk		= clear_refs_test_walk,
1248 };
1249 
1250 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1251 				size_t count, loff_t *ppos)
1252 {
1253 	struct task_struct *task;
1254 	char buffer[PROC_NUMBUF];
1255 	struct mm_struct *mm;
1256 	struct vm_area_struct *vma;
1257 	enum clear_refs_types type;
1258 	int itype;
1259 	int rv;
1260 
1261 	memset(buffer, 0, sizeof(buffer));
1262 	if (count > sizeof(buffer) - 1)
1263 		count = sizeof(buffer) - 1;
1264 	if (copy_from_user(buffer, buf, count))
1265 		return -EFAULT;
1266 	rv = kstrtoint(strstrip(buffer), 10, &itype);
1267 	if (rv < 0)
1268 		return rv;
1269 	type = (enum clear_refs_types)itype;
1270 	if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1271 		return -EINVAL;
1272 
1273 	task = get_proc_task(file_inode(file));
1274 	if (!task)
1275 		return -ESRCH;
1276 	mm = get_task_mm(task);
1277 	if (mm) {
1278 		VMA_ITERATOR(vmi, mm, 0);
1279 		struct mmu_notifier_range range;
1280 		struct clear_refs_private cp = {
1281 			.type = type,
1282 		};
1283 
1284 		if (mmap_write_lock_killable(mm)) {
1285 			count = -EINTR;
1286 			goto out_mm;
1287 		}
1288 		if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1289 			/*
1290 			 * Writing 5 to /proc/pid/clear_refs resets the peak
1291 			 * resident set size to this mm's current rss value.
1292 			 */
1293 			reset_mm_hiwater_rss(mm);
1294 			goto out_unlock;
1295 		}
1296 
1297 		if (type == CLEAR_REFS_SOFT_DIRTY) {
1298 			for_each_vma(vmi, vma) {
1299 				if (!(vma->vm_flags & VM_SOFTDIRTY))
1300 					continue;
1301 				vm_flags_clear(vma, VM_SOFTDIRTY);
1302 				vma_set_page_prot(vma);
1303 			}
1304 
1305 			inc_tlb_flush_pending(mm);
1306 			mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY,
1307 						0, mm, 0, -1UL);
1308 			mmu_notifier_invalidate_range_start(&range);
1309 		}
1310 		walk_page_range(mm, 0, -1, &clear_refs_walk_ops, &cp);
1311 		if (type == CLEAR_REFS_SOFT_DIRTY) {
1312 			mmu_notifier_invalidate_range_end(&range);
1313 			flush_tlb_mm(mm);
1314 			dec_tlb_flush_pending(mm);
1315 		}
1316 out_unlock:
1317 		mmap_write_unlock(mm);
1318 out_mm:
1319 		mmput(mm);
1320 	}
1321 	put_task_struct(task);
1322 
1323 	return count;
1324 }
1325 
1326 const struct file_operations proc_clear_refs_operations = {
1327 	.write		= clear_refs_write,
1328 	.llseek		= noop_llseek,
1329 };
1330 
1331 typedef struct {
1332 	u64 pme;
1333 } pagemap_entry_t;
1334 
1335 struct pagemapread {
1336 	int pos, len;		/* units: PM_ENTRY_BYTES, not bytes */
1337 	pagemap_entry_t *buffer;
1338 	bool show_pfn;
1339 };
1340 
1341 #define PAGEMAP_WALK_SIZE	(PMD_SIZE)
1342 #define PAGEMAP_WALK_MASK	(PMD_MASK)
1343 
1344 #define PM_ENTRY_BYTES		sizeof(pagemap_entry_t)
1345 #define PM_PFRAME_BITS		55
1346 #define PM_PFRAME_MASK		GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1347 #define PM_SOFT_DIRTY		BIT_ULL(55)
1348 #define PM_MMAP_EXCLUSIVE	BIT_ULL(56)
1349 #define PM_UFFD_WP		BIT_ULL(57)
1350 #define PM_FILE			BIT_ULL(61)
1351 #define PM_SWAP			BIT_ULL(62)
1352 #define PM_PRESENT		BIT_ULL(63)
1353 
1354 #define PM_END_OF_BUFFER    1
1355 
1356 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1357 {
1358 	return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1359 }
1360 
1361 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1362 			  struct pagemapread *pm)
1363 {
1364 	pm->buffer[pm->pos++] = *pme;
1365 	if (pm->pos >= pm->len)
1366 		return PM_END_OF_BUFFER;
1367 	return 0;
1368 }
1369 
1370 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1371 			    __always_unused int depth, struct mm_walk *walk)
1372 {
1373 	struct pagemapread *pm = walk->private;
1374 	unsigned long addr = start;
1375 	int err = 0;
1376 
1377 	while (addr < end) {
1378 		struct vm_area_struct *vma = find_vma(walk->mm, addr);
1379 		pagemap_entry_t pme = make_pme(0, 0);
1380 		/* End of address space hole, which we mark as non-present. */
1381 		unsigned long hole_end;
1382 
1383 		if (vma)
1384 			hole_end = min(end, vma->vm_start);
1385 		else
1386 			hole_end = end;
1387 
1388 		for (; addr < hole_end; addr += PAGE_SIZE) {
1389 			err = add_to_pagemap(addr, &pme, pm);
1390 			if (err)
1391 				goto out;
1392 		}
1393 
1394 		if (!vma)
1395 			break;
1396 
1397 		/* Addresses in the VMA. */
1398 		if (vma->vm_flags & VM_SOFTDIRTY)
1399 			pme = make_pme(0, PM_SOFT_DIRTY);
1400 		for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1401 			err = add_to_pagemap(addr, &pme, pm);
1402 			if (err)
1403 				goto out;
1404 		}
1405 	}
1406 out:
1407 	return err;
1408 }
1409 
1410 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1411 		struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1412 {
1413 	u64 frame = 0, flags = 0;
1414 	struct page *page = NULL;
1415 	bool migration = false;
1416 
1417 	if (pte_present(pte)) {
1418 		if (pm->show_pfn)
1419 			frame = pte_pfn(pte);
1420 		flags |= PM_PRESENT;
1421 		page = vm_normal_page(vma, addr, pte);
1422 		if (pte_soft_dirty(pte))
1423 			flags |= PM_SOFT_DIRTY;
1424 		if (pte_uffd_wp(pte))
1425 			flags |= PM_UFFD_WP;
1426 	} else if (is_swap_pte(pte)) {
1427 		swp_entry_t entry;
1428 		if (pte_swp_soft_dirty(pte))
1429 			flags |= PM_SOFT_DIRTY;
1430 		if (pte_swp_uffd_wp(pte))
1431 			flags |= PM_UFFD_WP;
1432 		entry = pte_to_swp_entry(pte);
1433 		if (pm->show_pfn) {
1434 			pgoff_t offset;
1435 			/*
1436 			 * For PFN swap offsets, keeping the offset field
1437 			 * to be PFN only to be compatible with old smaps.
1438 			 */
1439 			if (is_pfn_swap_entry(entry))
1440 				offset = swp_offset_pfn(entry);
1441 			else
1442 				offset = swp_offset(entry);
1443 			frame = swp_type(entry) |
1444 			    (offset << MAX_SWAPFILES_SHIFT);
1445 		}
1446 		flags |= PM_SWAP;
1447 		migration = is_migration_entry(entry);
1448 		if (is_pfn_swap_entry(entry))
1449 			page = pfn_swap_entry_to_page(entry);
1450 		if (pte_marker_entry_uffd_wp(entry))
1451 			flags |= PM_UFFD_WP;
1452 	}
1453 
1454 	if (page && !PageAnon(page))
1455 		flags |= PM_FILE;
1456 	if (page && !migration && page_mapcount(page) == 1)
1457 		flags |= PM_MMAP_EXCLUSIVE;
1458 	if (vma->vm_flags & VM_SOFTDIRTY)
1459 		flags |= PM_SOFT_DIRTY;
1460 
1461 	return make_pme(frame, flags);
1462 }
1463 
1464 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1465 			     struct mm_walk *walk)
1466 {
1467 	struct vm_area_struct *vma = walk->vma;
1468 	struct pagemapread *pm = walk->private;
1469 	spinlock_t *ptl;
1470 	pte_t *pte, *orig_pte;
1471 	int err = 0;
1472 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1473 	bool migration = false;
1474 
1475 	ptl = pmd_trans_huge_lock(pmdp, vma);
1476 	if (ptl) {
1477 		u64 flags = 0, frame = 0;
1478 		pmd_t pmd = *pmdp;
1479 		struct page *page = NULL;
1480 
1481 		if (vma->vm_flags & VM_SOFTDIRTY)
1482 			flags |= PM_SOFT_DIRTY;
1483 
1484 		if (pmd_present(pmd)) {
1485 			page = pmd_page(pmd);
1486 
1487 			flags |= PM_PRESENT;
1488 			if (pmd_soft_dirty(pmd))
1489 				flags |= PM_SOFT_DIRTY;
1490 			if (pmd_uffd_wp(pmd))
1491 				flags |= PM_UFFD_WP;
1492 			if (pm->show_pfn)
1493 				frame = pmd_pfn(pmd) +
1494 					((addr & ~PMD_MASK) >> PAGE_SHIFT);
1495 		}
1496 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1497 		else if (is_swap_pmd(pmd)) {
1498 			swp_entry_t entry = pmd_to_swp_entry(pmd);
1499 			unsigned long offset;
1500 
1501 			if (pm->show_pfn) {
1502 				if (is_pfn_swap_entry(entry))
1503 					offset = swp_offset_pfn(entry);
1504 				else
1505 					offset = swp_offset(entry);
1506 				offset = offset +
1507 					((addr & ~PMD_MASK) >> PAGE_SHIFT);
1508 				frame = swp_type(entry) |
1509 					(offset << MAX_SWAPFILES_SHIFT);
1510 			}
1511 			flags |= PM_SWAP;
1512 			if (pmd_swp_soft_dirty(pmd))
1513 				flags |= PM_SOFT_DIRTY;
1514 			if (pmd_swp_uffd_wp(pmd))
1515 				flags |= PM_UFFD_WP;
1516 			VM_BUG_ON(!is_pmd_migration_entry(pmd));
1517 			migration = is_migration_entry(entry);
1518 			page = pfn_swap_entry_to_page(entry);
1519 		}
1520 #endif
1521 
1522 		if (page && !migration && page_mapcount(page) == 1)
1523 			flags |= PM_MMAP_EXCLUSIVE;
1524 
1525 		for (; addr != end; addr += PAGE_SIZE) {
1526 			pagemap_entry_t pme = make_pme(frame, flags);
1527 
1528 			err = add_to_pagemap(addr, &pme, pm);
1529 			if (err)
1530 				break;
1531 			if (pm->show_pfn) {
1532 				if (flags & PM_PRESENT)
1533 					frame++;
1534 				else if (flags & PM_SWAP)
1535 					frame += (1 << MAX_SWAPFILES_SHIFT);
1536 			}
1537 		}
1538 		spin_unlock(ptl);
1539 		return err;
1540 	}
1541 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1542 
1543 	/*
1544 	 * We can assume that @vma always points to a valid one and @end never
1545 	 * goes beyond vma->vm_end.
1546 	 */
1547 	orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1548 	if (!pte) {
1549 		walk->action = ACTION_AGAIN;
1550 		return err;
1551 	}
1552 	for (; addr < end; pte++, addr += PAGE_SIZE) {
1553 		pagemap_entry_t pme;
1554 
1555 		pme = pte_to_pagemap_entry(pm, vma, addr, ptep_get(pte));
1556 		err = add_to_pagemap(addr, &pme, pm);
1557 		if (err)
1558 			break;
1559 	}
1560 	pte_unmap_unlock(orig_pte, ptl);
1561 
1562 	cond_resched();
1563 
1564 	return err;
1565 }
1566 
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)
1572 {
1573 	struct pagemapread *pm = walk->private;
1574 	struct vm_area_struct *vma = walk->vma;
1575 	u64 flags = 0, frame = 0;
1576 	int err = 0;
1577 	pte_t pte;
1578 
1579 	if (vma->vm_flags & VM_SOFTDIRTY)
1580 		flags |= PM_SOFT_DIRTY;
1581 
1582 	pte = huge_ptep_get(ptep);
1583 	if (pte_present(pte)) {
1584 		struct page *page = pte_page(pte);
1585 
1586 		if (!PageAnon(page))
1587 			flags |= PM_FILE;
1588 
1589 		if (page_mapcount(page) == 1)
1590 			flags |= PM_MMAP_EXCLUSIVE;
1591 
1592 		if (huge_pte_uffd_wp(pte))
1593 			flags |= PM_UFFD_WP;
1594 
1595 		flags |= PM_PRESENT;
1596 		if (pm->show_pfn)
1597 			frame = pte_pfn(pte) +
1598 				((addr & ~hmask) >> PAGE_SHIFT);
1599 	} else if (pte_swp_uffd_wp_any(pte)) {
1600 		flags |= PM_UFFD_WP;
1601 	}
1602 
1603 	for (; addr != end; addr += PAGE_SIZE) {
1604 		pagemap_entry_t pme = make_pme(frame, flags);
1605 
1606 		err = add_to_pagemap(addr, &pme, pm);
1607 		if (err)
1608 			return err;
1609 		if (pm->show_pfn && (flags & PM_PRESENT))
1610 			frame++;
1611 	}
1612 
1613 	cond_resched();
1614 
1615 	return err;
1616 }
1617 #else
1618 #define pagemap_hugetlb_range	NULL
1619 #endif /* HUGETLB_PAGE */
1620 
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,
1625 };
1626 
1627 /*
1628  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1629  *
1630  * For each page in the address space, this file contains one 64-bit entry
1631  * consisting of the following:
1632  *
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
1639  * Bits 58-60 zero
1640  * Bit  61    page is file-page or shared-anon
1641  * Bit  62    page swapped
1642  * Bit  63    page present
1643  *
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.
1649  *
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.
1653  */
1654 static ssize_t pagemap_read(struct file *file, char __user *buf,
1655 			    size_t count, loff_t *ppos)
1656 {
1657 	struct mm_struct *mm = file->private_data;
1658 	struct pagemapread pm;
1659 	unsigned long src;
1660 	unsigned long svpfn;
1661 	unsigned long start_vaddr;
1662 	unsigned long end_vaddr;
1663 	int ret = 0, copied = 0;
1664 
1665 	if (!mm || !mmget_not_zero(mm))
1666 		goto out;
1667 
1668 	ret = -EINVAL;
1669 	/* file position must be aligned */
1670 	if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1671 		goto out_mm;
1672 
1673 	ret = 0;
1674 	if (!count)
1675 		goto out_mm;
1676 
1677 	/* do not disclose physical addresses: attack vector */
1678 	pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1679 
1680 	pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1681 	pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL);
1682 	ret = -ENOMEM;
1683 	if (!pm.buffer)
1684 		goto out_mm;
1685 
1686 	src = *ppos;
1687 	svpfn = src / PM_ENTRY_BYTES;
1688 	end_vaddr = mm->task_size;
1689 
1690 	/* watch out for wraparound */
1691 	start_vaddr = end_vaddr;
1692 	if (svpfn <= (ULONG_MAX >> PAGE_SHIFT)) {
1693 		unsigned long end;
1694 
1695 		ret = mmap_read_lock_killable(mm);
1696 		if (ret)
1697 			goto out_free;
1698 		start_vaddr = untagged_addr_remote(mm, svpfn << PAGE_SHIFT);
1699 		mmap_read_unlock(mm);
1700 
1701 		end = start_vaddr + ((count / PM_ENTRY_BYTES) << PAGE_SHIFT);
1702 		if (end >= start_vaddr && end < mm->task_size)
1703 			end_vaddr = end;
1704 	}
1705 
1706 	/* Ensure the address is inside the task */
1707 	if (start_vaddr > mm->task_size)
1708 		start_vaddr = end_vaddr;
1709 
1710 	ret = 0;
1711 	while (count && (start_vaddr < end_vaddr)) {
1712 		int len;
1713 		unsigned long end;
1714 
1715 		pm.pos = 0;
1716 		end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1717 		/* overflow ? */
1718 		if (end < start_vaddr || end > end_vaddr)
1719 			end = end_vaddr;
1720 		ret = mmap_read_lock_killable(mm);
1721 		if (ret)
1722 			goto out_free;
1723 		ret = walk_page_range(mm, start_vaddr, end, &pagemap_ops, &pm);
1724 		mmap_read_unlock(mm);
1725 		start_vaddr = end;
1726 
1727 		len = min(count, PM_ENTRY_BYTES * pm.pos);
1728 		if (copy_to_user(buf, pm.buffer, len)) {
1729 			ret = -EFAULT;
1730 			goto out_free;
1731 		}
1732 		copied += len;
1733 		buf += len;
1734 		count -= len;
1735 	}
1736 	*ppos += copied;
1737 	if (!ret || ret == PM_END_OF_BUFFER)
1738 		ret = copied;
1739 
1740 out_free:
1741 	kfree(pm.buffer);
1742 out_mm:
1743 	mmput(mm);
1744 out:
1745 	return ret;
1746 }
1747 
1748 static int pagemap_open(struct inode *inode, struct file *file)
1749 {
1750 	struct mm_struct *mm;
1751 
1752 	mm = proc_mem_open(inode, PTRACE_MODE_READ);
1753 	if (IS_ERR(mm))
1754 		return PTR_ERR(mm);
1755 	file->private_data = mm;
1756 	return 0;
1757 }
1758 
1759 static int pagemap_release(struct inode *inode, struct file *file)
1760 {
1761 	struct mm_struct *mm = file->private_data;
1762 
1763 	if (mm)
1764 		mmdrop(mm);
1765 	return 0;
1766 }
1767 
1768 const struct file_operations proc_pagemap_operations = {
1769 	.llseek		= mem_lseek, /* borrow this */
1770 	.read		= pagemap_read,
1771 	.open		= pagemap_open,
1772 	.release	= pagemap_release,
1773 };
1774 #endif /* CONFIG_PROC_PAGE_MONITOR */
1775 
1776 #ifdef CONFIG_NUMA
1777 
1778 struct numa_maps {
1779 	unsigned long pages;
1780 	unsigned long anon;
1781 	unsigned long active;
1782 	unsigned long writeback;
1783 	unsigned long mapcount_max;
1784 	unsigned long dirty;
1785 	unsigned long swapcache;
1786 	unsigned long node[MAX_NUMNODES];
1787 };
1788 
1789 struct numa_maps_private {
1790 	struct proc_maps_private proc_maps;
1791 	struct numa_maps md;
1792 };
1793 
1794 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1795 			unsigned long nr_pages)
1796 {
1797 	int count = page_mapcount(page);
1798 
1799 	md->pages += nr_pages;
1800 	if (pte_dirty || PageDirty(page))
1801 		md->dirty += nr_pages;
1802 
1803 	if (PageSwapCache(page))
1804 		md->swapcache += nr_pages;
1805 
1806 	if (PageActive(page) || PageUnevictable(page))
1807 		md->active += nr_pages;
1808 
1809 	if (PageWriteback(page))
1810 		md->writeback += nr_pages;
1811 
1812 	if (PageAnon(page))
1813 		md->anon += nr_pages;
1814 
1815 	if (count > md->mapcount_max)
1816 		md->mapcount_max = count;
1817 
1818 	md->node[page_to_nid(page)] += nr_pages;
1819 }
1820 
1821 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1822 		unsigned long addr)
1823 {
1824 	struct page *page;
1825 	int nid;
1826 
1827 	if (!pte_present(pte))
1828 		return NULL;
1829 
1830 	page = vm_normal_page(vma, addr, pte);
1831 	if (!page || is_zone_device_page(page))
1832 		return NULL;
1833 
1834 	if (PageReserved(page))
1835 		return NULL;
1836 
1837 	nid = page_to_nid(page);
1838 	if (!node_isset(nid, node_states[N_MEMORY]))
1839 		return NULL;
1840 
1841 	return page;
1842 }
1843 
1844 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1845 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1846 					      struct vm_area_struct *vma,
1847 					      unsigned long addr)
1848 {
1849 	struct page *page;
1850 	int nid;
1851 
1852 	if (!pmd_present(pmd))
1853 		return NULL;
1854 
1855 	page = vm_normal_page_pmd(vma, addr, pmd);
1856 	if (!page)
1857 		return NULL;
1858 
1859 	if (PageReserved(page))
1860 		return NULL;
1861 
1862 	nid = page_to_nid(page);
1863 	if (!node_isset(nid, node_states[N_MEMORY]))
1864 		return NULL;
1865 
1866 	return page;
1867 }
1868 #endif
1869 
1870 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1871 		unsigned long end, struct mm_walk *walk)
1872 {
1873 	struct numa_maps *md = walk->private;
1874 	struct vm_area_struct *vma = walk->vma;
1875 	spinlock_t *ptl;
1876 	pte_t *orig_pte;
1877 	pte_t *pte;
1878 
1879 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1880 	ptl = pmd_trans_huge_lock(pmd, vma);
1881 	if (ptl) {
1882 		struct page *page;
1883 
1884 		page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1885 		if (page)
1886 			gather_stats(page, md, pmd_dirty(*pmd),
1887 				     HPAGE_PMD_SIZE/PAGE_SIZE);
1888 		spin_unlock(ptl);
1889 		return 0;
1890 	}
1891 #endif
1892 	orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1893 	if (!pte) {
1894 		walk->action = ACTION_AGAIN;
1895 		return 0;
1896 	}
1897 	do {
1898 		pte_t ptent = ptep_get(pte);
1899 		struct page *page = can_gather_numa_stats(ptent, vma, addr);
1900 		if (!page)
1901 			continue;
1902 		gather_stats(page, md, pte_dirty(ptent), 1);
1903 
1904 	} while (pte++, addr += PAGE_SIZE, addr != end);
1905 	pte_unmap_unlock(orig_pte, ptl);
1906 	cond_resched();
1907 	return 0;
1908 }
1909 #ifdef CONFIG_HUGETLB_PAGE
1910 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1911 		unsigned long addr, unsigned long end, struct mm_walk *walk)
1912 {
1913 	pte_t huge_pte = huge_ptep_get(pte);
1914 	struct numa_maps *md;
1915 	struct page *page;
1916 
1917 	if (!pte_present(huge_pte))
1918 		return 0;
1919 
1920 	page = pte_page(huge_pte);
1921 
1922 	md = walk->private;
1923 	gather_stats(page, md, pte_dirty(huge_pte), 1);
1924 	return 0;
1925 }
1926 
1927 #else
1928 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1929 		unsigned long addr, unsigned long end, struct mm_walk *walk)
1930 {
1931 	return 0;
1932 }
1933 #endif
1934 
1935 static const struct mm_walk_ops show_numa_ops = {
1936 	.hugetlb_entry = gather_hugetlb_stats,
1937 	.pmd_entry = gather_pte_stats,
1938 };
1939 
1940 /*
1941  * Display pages allocated per node and memory policy via /proc.
1942  */
1943 static int show_numa_map(struct seq_file *m, void *v)
1944 {
1945 	struct numa_maps_private *numa_priv = m->private;
1946 	struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1947 	struct vm_area_struct *vma = v;
1948 	struct numa_maps *md = &numa_priv->md;
1949 	struct file *file = vma->vm_file;
1950 	struct mm_struct *mm = vma->vm_mm;
1951 	struct mempolicy *pol;
1952 	char buffer[64];
1953 	int nid;
1954 
1955 	if (!mm)
1956 		return 0;
1957 
1958 	/* Ensure we start with an empty set of numa_maps statistics. */
1959 	memset(md, 0, sizeof(*md));
1960 
1961 	pol = __get_vma_policy(vma, vma->vm_start);
1962 	if (pol) {
1963 		mpol_to_str(buffer, sizeof(buffer), pol);
1964 		mpol_cond_put(pol);
1965 	} else {
1966 		mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1967 	}
1968 
1969 	seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1970 
1971 	if (file) {
1972 		seq_puts(m, " file=");
1973 		seq_file_path(m, file, "\n\t= ");
1974 	} else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1975 		seq_puts(m, " heap");
1976 	} else if (is_stack(vma)) {
1977 		seq_puts(m, " stack");
1978 	}
1979 
1980 	if (is_vm_hugetlb_page(vma))
1981 		seq_puts(m, " huge");
1982 
1983 	/* mmap_lock is held by m_start */
1984 	walk_page_vma(vma, &show_numa_ops, md);
1985 
1986 	if (!md->pages)
1987 		goto out;
1988 
1989 	if (md->anon)
1990 		seq_printf(m, " anon=%lu", md->anon);
1991 
1992 	if (md->dirty)
1993 		seq_printf(m, " dirty=%lu", md->dirty);
1994 
1995 	if (md->pages != md->anon && md->pages != md->dirty)
1996 		seq_printf(m, " mapped=%lu", md->pages);
1997 
1998 	if (md->mapcount_max > 1)
1999 		seq_printf(m, " mapmax=%lu", md->mapcount_max);
2000 
2001 	if (md->swapcache)
2002 		seq_printf(m, " swapcache=%lu", md->swapcache);
2003 
2004 	if (md->active < md->pages && !is_vm_hugetlb_page(vma))
2005 		seq_printf(m, " active=%lu", md->active);
2006 
2007 	if (md->writeback)
2008 		seq_printf(m, " writeback=%lu", md->writeback);
2009 
2010 	for_each_node_state(nid, N_MEMORY)
2011 		if (md->node[nid])
2012 			seq_printf(m, " N%d=%lu", nid, md->node[nid]);
2013 
2014 	seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
2015 out:
2016 	seq_putc(m, '\n');
2017 	return 0;
2018 }
2019 
2020 static const struct seq_operations proc_pid_numa_maps_op = {
2021 	.start  = m_start,
2022 	.next   = m_next,
2023 	.stop   = m_stop,
2024 	.show   = show_numa_map,
2025 };
2026 
2027 static int pid_numa_maps_open(struct inode *inode, struct file *file)
2028 {
2029 	return proc_maps_open(inode, file, &proc_pid_numa_maps_op,
2030 				sizeof(struct numa_maps_private));
2031 }
2032 
2033 const struct file_operations proc_pid_numa_maps_operations = {
2034 	.open		= pid_numa_maps_open,
2035 	.read		= seq_read,
2036 	.llseek		= seq_lseek,
2037 	.release	= proc_map_release,
2038 };
2039 
2040 #endif /* CONFIG_NUMA */
2041