xref: /linux/fs/proc/task_mmu.c (revision ff9a79307f89563da6d841da8b7cc4a0afceb0e2)
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/ksm.h>
8 #include <linux/seq_file.h>
9 #include <linux/highmem.h>
10 #include <linux/ptrace.h>
11 #include <linux/slab.h>
12 #include <linux/pagemap.h>
13 #include <linux/mempolicy.h>
14 #include <linux/rmap.h>
15 #include <linux/swap.h>
16 #include <linux/sched/mm.h>
17 #include <linux/swapops.h>
18 #include <linux/mmu_notifier.h>
19 #include <linux/page_idle.h>
20 #include <linux/shmem_fs.h>
21 #include <linux/uaccess.h>
22 #include <linux/pkeys.h>
23 #include <linux/minmax.h>
24 #include <linux/overflow.h>
25 
26 #include <asm/elf.h>
27 #include <asm/tlb.h>
28 #include <asm/tlbflush.h>
29 #include "internal.h"
30 
31 #define SEQ_PUT_DEC(str, val) \
32 		seq_put_decimal_ull_width(m, str, (val) << (PAGE_SHIFT-10), 8)
33 void task_mem(struct seq_file *m, struct mm_struct *mm)
34 {
35 	unsigned long text, lib, swap, anon, file, shmem;
36 	unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
37 
38 	anon = get_mm_counter(mm, MM_ANONPAGES);
39 	file = get_mm_counter(mm, MM_FILEPAGES);
40 	shmem = get_mm_counter(mm, MM_SHMEMPAGES);
41 
42 	/*
43 	 * Note: to minimize their overhead, mm maintains hiwater_vm and
44 	 * hiwater_rss only when about to *lower* total_vm or rss.  Any
45 	 * collector of these hiwater stats must therefore get total_vm
46 	 * and rss too, which will usually be the higher.  Barriers? not
47 	 * worth the effort, such snapshots can always be inconsistent.
48 	 */
49 	hiwater_vm = total_vm = mm->total_vm;
50 	if (hiwater_vm < mm->hiwater_vm)
51 		hiwater_vm = mm->hiwater_vm;
52 	hiwater_rss = total_rss = anon + file + shmem;
53 	if (hiwater_rss < mm->hiwater_rss)
54 		hiwater_rss = mm->hiwater_rss;
55 
56 	/* split executable areas between text and lib */
57 	text = PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK);
58 	text = min(text, mm->exec_vm << PAGE_SHIFT);
59 	lib = (mm->exec_vm << PAGE_SHIFT) - text;
60 
61 	swap = get_mm_counter(mm, MM_SWAPENTS);
62 	SEQ_PUT_DEC("VmPeak:\t", hiwater_vm);
63 	SEQ_PUT_DEC(" kB\nVmSize:\t", total_vm);
64 	SEQ_PUT_DEC(" kB\nVmLck:\t", mm->locked_vm);
65 	SEQ_PUT_DEC(" kB\nVmPin:\t", atomic64_read(&mm->pinned_vm));
66 	SEQ_PUT_DEC(" kB\nVmHWM:\t", hiwater_rss);
67 	SEQ_PUT_DEC(" kB\nVmRSS:\t", total_rss);
68 	SEQ_PUT_DEC(" kB\nRssAnon:\t", anon);
69 	SEQ_PUT_DEC(" kB\nRssFile:\t", file);
70 	SEQ_PUT_DEC(" kB\nRssShmem:\t", shmem);
71 	SEQ_PUT_DEC(" kB\nVmData:\t", mm->data_vm);
72 	SEQ_PUT_DEC(" kB\nVmStk:\t", mm->stack_vm);
73 	seq_put_decimal_ull_width(m,
74 		    " kB\nVmExe:\t", text >> 10, 8);
75 	seq_put_decimal_ull_width(m,
76 		    " kB\nVmLib:\t", lib >> 10, 8);
77 	seq_put_decimal_ull_width(m,
78 		    " kB\nVmPTE:\t", mm_pgtables_bytes(mm) >> 10, 8);
79 	SEQ_PUT_DEC(" kB\nVmSwap:\t", swap);
80 	seq_puts(m, " kB\n");
81 	hugetlb_report_usage(m, mm);
82 }
83 #undef SEQ_PUT_DEC
84 
85 unsigned long task_vsize(struct mm_struct *mm)
86 {
87 	return PAGE_SIZE * mm->total_vm;
88 }
89 
90 unsigned long task_statm(struct mm_struct *mm,
91 			 unsigned long *shared, unsigned long *text,
92 			 unsigned long *data, unsigned long *resident)
93 {
94 	*shared = get_mm_counter(mm, MM_FILEPAGES) +
95 			get_mm_counter(mm, MM_SHMEMPAGES);
96 	*text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
97 								>> PAGE_SHIFT;
98 	*data = mm->data_vm + mm->stack_vm;
99 	*resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
100 	return mm->total_vm;
101 }
102 
103 #ifdef CONFIG_NUMA
104 /*
105  * Save get_task_policy() for show_numa_map().
106  */
107 static void hold_task_mempolicy(struct proc_maps_private *priv)
108 {
109 	struct task_struct *task = priv->task;
110 
111 	task_lock(task);
112 	priv->task_mempolicy = get_task_policy(task);
113 	mpol_get(priv->task_mempolicy);
114 	task_unlock(task);
115 }
116 static void release_task_mempolicy(struct proc_maps_private *priv)
117 {
118 	mpol_put(priv->task_mempolicy);
119 }
120 #else
121 static void hold_task_mempolicy(struct proc_maps_private *priv)
122 {
123 }
124 static void release_task_mempolicy(struct proc_maps_private *priv)
125 {
126 }
127 #endif
128 
129 static struct vm_area_struct *proc_get_vma(struct proc_maps_private *priv,
130 						loff_t *ppos)
131 {
132 	struct vm_area_struct *vma = vma_next(&priv->iter);
133 
134 	if (vma) {
135 		*ppos = vma->vm_start;
136 	} else {
137 		*ppos = -2UL;
138 		vma = get_gate_vma(priv->mm);
139 	}
140 
141 	return vma;
142 }
143 
144 static void *m_start(struct seq_file *m, loff_t *ppos)
145 {
146 	struct proc_maps_private *priv = m->private;
147 	unsigned long last_addr = *ppos;
148 	struct mm_struct *mm;
149 
150 	/* See m_next(). Zero at the start or after lseek. */
151 	if (last_addr == -1UL)
152 		return NULL;
153 
154 	priv->task = get_proc_task(priv->inode);
155 	if (!priv->task)
156 		return ERR_PTR(-ESRCH);
157 
158 	mm = priv->mm;
159 	if (!mm || !mmget_not_zero(mm)) {
160 		put_task_struct(priv->task);
161 		priv->task = NULL;
162 		return NULL;
163 	}
164 
165 	if (mmap_read_lock_killable(mm)) {
166 		mmput(mm);
167 		put_task_struct(priv->task);
168 		priv->task = NULL;
169 		return ERR_PTR(-EINTR);
170 	}
171 
172 	vma_iter_init(&priv->iter, mm, last_addr);
173 	hold_task_mempolicy(priv);
174 	if (last_addr == -2UL)
175 		return get_gate_vma(mm);
176 
177 	return proc_get_vma(priv, ppos);
178 }
179 
180 static void *m_next(struct seq_file *m, void *v, loff_t *ppos)
181 {
182 	if (*ppos == -2UL) {
183 		*ppos = -1UL;
184 		return NULL;
185 	}
186 	return proc_get_vma(m->private, ppos);
187 }
188 
189 static void m_stop(struct seq_file *m, void *v)
190 {
191 	struct proc_maps_private *priv = m->private;
192 	struct mm_struct *mm = priv->mm;
193 
194 	if (!priv->task)
195 		return;
196 
197 	release_task_mempolicy(priv);
198 	mmap_read_unlock(mm);
199 	mmput(mm);
200 	put_task_struct(priv->task);
201 	priv->task = NULL;
202 }
203 
204 static int proc_maps_open(struct inode *inode, struct file *file,
205 			const struct seq_operations *ops, int psize)
206 {
207 	struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
208 
209 	if (!priv)
210 		return -ENOMEM;
211 
212 	priv->inode = inode;
213 	priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
214 	if (IS_ERR(priv->mm)) {
215 		int err = PTR_ERR(priv->mm);
216 
217 		seq_release_private(inode, file);
218 		return err;
219 	}
220 
221 	return 0;
222 }
223 
224 static int proc_map_release(struct inode *inode, struct file *file)
225 {
226 	struct seq_file *seq = file->private_data;
227 	struct proc_maps_private *priv = seq->private;
228 
229 	if (priv->mm)
230 		mmdrop(priv->mm);
231 
232 	return seq_release_private(inode, file);
233 }
234 
235 static int do_maps_open(struct inode *inode, struct file *file,
236 			const struct seq_operations *ops)
237 {
238 	return proc_maps_open(inode, file, ops,
239 				sizeof(struct proc_maps_private));
240 }
241 
242 static void show_vma_header_prefix(struct seq_file *m,
243 				   unsigned long start, unsigned long end,
244 				   vm_flags_t flags, unsigned long long pgoff,
245 				   dev_t dev, unsigned long ino)
246 {
247 	seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
248 	seq_put_hex_ll(m, NULL, start, 8);
249 	seq_put_hex_ll(m, "-", end, 8);
250 	seq_putc(m, ' ');
251 	seq_putc(m, flags & VM_READ ? 'r' : '-');
252 	seq_putc(m, flags & VM_WRITE ? 'w' : '-');
253 	seq_putc(m, flags & VM_EXEC ? 'x' : '-');
254 	seq_putc(m, flags & VM_MAYSHARE ? 's' : 'p');
255 	seq_put_hex_ll(m, " ", pgoff, 8);
256 	seq_put_hex_ll(m, " ", MAJOR(dev), 2);
257 	seq_put_hex_ll(m, ":", MINOR(dev), 2);
258 	seq_put_decimal_ull(m, " ", ino);
259 	seq_putc(m, ' ');
260 }
261 
262 static void
263 show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
264 {
265 	struct anon_vma_name *anon_name = NULL;
266 	struct mm_struct *mm = vma->vm_mm;
267 	struct file *file = vma->vm_file;
268 	vm_flags_t flags = vma->vm_flags;
269 	unsigned long ino = 0;
270 	unsigned long long pgoff = 0;
271 	unsigned long start, end;
272 	dev_t dev = 0;
273 	const char *name = NULL;
274 
275 	if (file) {
276 		const struct inode *inode = file_user_inode(vma->vm_file);
277 
278 		dev = inode->i_sb->s_dev;
279 		ino = inode->i_ino;
280 		pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
281 	}
282 
283 	start = vma->vm_start;
284 	end = vma->vm_end;
285 	show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino);
286 	if (mm)
287 		anon_name = anon_vma_name(vma);
288 
289 	/*
290 	 * Print the dentry name for named mappings, and a
291 	 * special [heap] marker for the heap:
292 	 */
293 	if (file) {
294 		seq_pad(m, ' ');
295 		/*
296 		 * If user named this anon shared memory via
297 		 * prctl(PR_SET_VMA ..., use the provided name.
298 		 */
299 		if (anon_name)
300 			seq_printf(m, "[anon_shmem:%s]", anon_name->name);
301 		else
302 			seq_path(m, file_user_path(file), "\n");
303 		goto done;
304 	}
305 
306 	if (vma->vm_ops && vma->vm_ops->name) {
307 		name = vma->vm_ops->name(vma);
308 		if (name)
309 			goto done;
310 	}
311 
312 	name = arch_vma_name(vma);
313 	if (!name) {
314 		if (!mm) {
315 			name = "[vdso]";
316 			goto done;
317 		}
318 
319 		if (vma_is_initial_heap(vma)) {
320 			name = "[heap]";
321 			goto done;
322 		}
323 
324 		if (vma_is_initial_stack(vma)) {
325 			name = "[stack]";
326 			goto done;
327 		}
328 
329 		if (anon_name) {
330 			seq_pad(m, ' ');
331 			seq_printf(m, "[anon:%s]", anon_name->name);
332 		}
333 	}
334 
335 done:
336 	if (name) {
337 		seq_pad(m, ' ');
338 		seq_puts(m, name);
339 	}
340 	seq_putc(m, '\n');
341 }
342 
343 static int show_map(struct seq_file *m, void *v)
344 {
345 	show_map_vma(m, v);
346 	return 0;
347 }
348 
349 static const struct seq_operations proc_pid_maps_op = {
350 	.start	= m_start,
351 	.next	= m_next,
352 	.stop	= m_stop,
353 	.show	= show_map
354 };
355 
356 static int pid_maps_open(struct inode *inode, struct file *file)
357 {
358 	return do_maps_open(inode, file, &proc_pid_maps_op);
359 }
360 
361 const struct file_operations proc_pid_maps_operations = {
362 	.open		= pid_maps_open,
363 	.read		= seq_read,
364 	.llseek		= seq_lseek,
365 	.release	= proc_map_release,
366 };
367 
368 /*
369  * Proportional Set Size(PSS): my share of RSS.
370  *
371  * PSS of a process is the count of pages it has in memory, where each
372  * page is divided by the number of processes sharing it.  So if a
373  * process has 1000 pages all to itself, and 1000 shared with one other
374  * process, its PSS will be 1500.
375  *
376  * To keep (accumulated) division errors low, we adopt a 64bit
377  * fixed-point pss counter to minimize division errors. So (pss >>
378  * PSS_SHIFT) would be the real byte count.
379  *
380  * A shift of 12 before division means (assuming 4K page size):
381  * 	- 1M 3-user-pages add up to 8KB errors;
382  * 	- supports mapcount up to 2^24, or 16M;
383  * 	- supports PSS up to 2^52 bytes, or 4PB.
384  */
385 #define PSS_SHIFT 12
386 
387 #ifdef CONFIG_PROC_PAGE_MONITOR
388 struct mem_size_stats {
389 	unsigned long resident;
390 	unsigned long shared_clean;
391 	unsigned long shared_dirty;
392 	unsigned long private_clean;
393 	unsigned long private_dirty;
394 	unsigned long referenced;
395 	unsigned long anonymous;
396 	unsigned long lazyfree;
397 	unsigned long anonymous_thp;
398 	unsigned long shmem_thp;
399 	unsigned long file_thp;
400 	unsigned long swap;
401 	unsigned long shared_hugetlb;
402 	unsigned long private_hugetlb;
403 	unsigned long ksm;
404 	u64 pss;
405 	u64 pss_anon;
406 	u64 pss_file;
407 	u64 pss_shmem;
408 	u64 pss_dirty;
409 	u64 pss_locked;
410 	u64 swap_pss;
411 };
412 
413 static void smaps_page_accumulate(struct mem_size_stats *mss,
414 		struct page *page, unsigned long size, unsigned long pss,
415 		bool dirty, bool locked, bool private)
416 {
417 	mss->pss += pss;
418 
419 	if (PageAnon(page))
420 		mss->pss_anon += pss;
421 	else if (PageSwapBacked(page))
422 		mss->pss_shmem += pss;
423 	else
424 		mss->pss_file += pss;
425 
426 	if (locked)
427 		mss->pss_locked += pss;
428 
429 	if (dirty || PageDirty(page)) {
430 		mss->pss_dirty += pss;
431 		if (private)
432 			mss->private_dirty += size;
433 		else
434 			mss->shared_dirty += size;
435 	} else {
436 		if (private)
437 			mss->private_clean += size;
438 		else
439 			mss->shared_clean += size;
440 	}
441 }
442 
443 static void smaps_account(struct mem_size_stats *mss, struct page *page,
444 		bool compound, bool young, bool dirty, bool locked,
445 		bool migration)
446 {
447 	int i, nr = compound ? compound_nr(page) : 1;
448 	unsigned long size = nr * PAGE_SIZE;
449 
450 	/*
451 	 * First accumulate quantities that depend only on |size| and the type
452 	 * of the compound page.
453 	 */
454 	if (PageAnon(page)) {
455 		mss->anonymous += size;
456 		if (!PageSwapBacked(page) && !dirty && !PageDirty(page))
457 			mss->lazyfree += size;
458 	}
459 
460 	if (PageKsm(page))
461 		mss->ksm += size;
462 
463 	mss->resident += size;
464 	/* Accumulate the size in pages that have been accessed. */
465 	if (young || page_is_young(page) || PageReferenced(page))
466 		mss->referenced += size;
467 
468 	/*
469 	 * Then accumulate quantities that may depend on sharing, or that may
470 	 * differ page-by-page.
471 	 *
472 	 * page_count(page) == 1 guarantees the page is mapped exactly once.
473 	 * If any subpage of the compound page mapped with PTE it would elevate
474 	 * page_count().
475 	 *
476 	 * The page_mapcount() is called to get a snapshot of the mapcount.
477 	 * Without holding the page lock this snapshot can be slightly wrong as
478 	 * we cannot always read the mapcount atomically.  It is not safe to
479 	 * call page_mapcount() even with PTL held if the page is not mapped,
480 	 * especially for migration entries.  Treat regular migration entries
481 	 * as mapcount == 1.
482 	 */
483 	if ((page_count(page) == 1) || migration) {
484 		smaps_page_accumulate(mss, page, size, size << PSS_SHIFT, dirty,
485 			locked, true);
486 		return;
487 	}
488 	for (i = 0; i < nr; i++, page++) {
489 		int mapcount = page_mapcount(page);
490 		unsigned long pss = PAGE_SIZE << PSS_SHIFT;
491 		if (mapcount >= 2)
492 			pss /= mapcount;
493 		smaps_page_accumulate(mss, page, PAGE_SIZE, pss, dirty, locked,
494 				      mapcount < 2);
495 	}
496 }
497 
498 #ifdef CONFIG_SHMEM
499 static int smaps_pte_hole(unsigned long addr, unsigned long end,
500 			  __always_unused int depth, struct mm_walk *walk)
501 {
502 	struct mem_size_stats *mss = walk->private;
503 	struct vm_area_struct *vma = walk->vma;
504 
505 	mss->swap += shmem_partial_swap_usage(walk->vma->vm_file->f_mapping,
506 					      linear_page_index(vma, addr),
507 					      linear_page_index(vma, end));
508 
509 	return 0;
510 }
511 #else
512 #define smaps_pte_hole		NULL
513 #endif /* CONFIG_SHMEM */
514 
515 static void smaps_pte_hole_lookup(unsigned long addr, struct mm_walk *walk)
516 {
517 #ifdef CONFIG_SHMEM
518 	if (walk->ops->pte_hole) {
519 		/* depth is not used */
520 		smaps_pte_hole(addr, addr + PAGE_SIZE, 0, walk);
521 	}
522 #endif
523 }
524 
525 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
526 		struct mm_walk *walk)
527 {
528 	struct mem_size_stats *mss = walk->private;
529 	struct vm_area_struct *vma = walk->vma;
530 	bool locked = !!(vma->vm_flags & VM_LOCKED);
531 	struct page *page = NULL;
532 	bool migration = false, young = false, dirty = false;
533 	pte_t ptent = ptep_get(pte);
534 
535 	if (pte_present(ptent)) {
536 		page = vm_normal_page(vma, addr, ptent);
537 		young = pte_young(ptent);
538 		dirty = pte_dirty(ptent);
539 	} else if (is_swap_pte(ptent)) {
540 		swp_entry_t swpent = pte_to_swp_entry(ptent);
541 
542 		if (!non_swap_entry(swpent)) {
543 			int mapcount;
544 
545 			mss->swap += PAGE_SIZE;
546 			mapcount = swp_swapcount(swpent);
547 			if (mapcount >= 2) {
548 				u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
549 
550 				do_div(pss_delta, mapcount);
551 				mss->swap_pss += pss_delta;
552 			} else {
553 				mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
554 			}
555 		} else if (is_pfn_swap_entry(swpent)) {
556 			if (is_migration_entry(swpent))
557 				migration = true;
558 			page = pfn_swap_entry_to_page(swpent);
559 		}
560 	} else {
561 		smaps_pte_hole_lookup(addr, walk);
562 		return;
563 	}
564 
565 	if (!page)
566 		return;
567 
568 	smaps_account(mss, page, false, young, dirty, locked, migration);
569 }
570 
571 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
572 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
573 		struct mm_walk *walk)
574 {
575 	struct mem_size_stats *mss = walk->private;
576 	struct vm_area_struct *vma = walk->vma;
577 	bool locked = !!(vma->vm_flags & VM_LOCKED);
578 	struct page *page = NULL;
579 	bool migration = false;
580 
581 	if (pmd_present(*pmd)) {
582 		page = vm_normal_page_pmd(vma, addr, *pmd);
583 	} else if (unlikely(thp_migration_supported() && is_swap_pmd(*pmd))) {
584 		swp_entry_t entry = pmd_to_swp_entry(*pmd);
585 
586 		if (is_migration_entry(entry)) {
587 			migration = true;
588 			page = pfn_swap_entry_to_page(entry);
589 		}
590 	}
591 	if (IS_ERR_OR_NULL(page))
592 		return;
593 	if (PageAnon(page))
594 		mss->anonymous_thp += HPAGE_PMD_SIZE;
595 	else if (PageSwapBacked(page))
596 		mss->shmem_thp += HPAGE_PMD_SIZE;
597 	else if (is_zone_device_page(page))
598 		/* pass */;
599 	else
600 		mss->file_thp += HPAGE_PMD_SIZE;
601 
602 	smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd),
603 		      locked, migration);
604 }
605 #else
606 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
607 		struct mm_walk *walk)
608 {
609 }
610 #endif
611 
612 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
613 			   struct mm_walk *walk)
614 {
615 	struct vm_area_struct *vma = walk->vma;
616 	pte_t *pte;
617 	spinlock_t *ptl;
618 
619 	ptl = pmd_trans_huge_lock(pmd, vma);
620 	if (ptl) {
621 		smaps_pmd_entry(pmd, addr, walk);
622 		spin_unlock(ptl);
623 		goto out;
624 	}
625 
626 	pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
627 	if (!pte) {
628 		walk->action = ACTION_AGAIN;
629 		return 0;
630 	}
631 	for (; addr != end; pte++, addr += PAGE_SIZE)
632 		smaps_pte_entry(pte, addr, walk);
633 	pte_unmap_unlock(pte - 1, ptl);
634 out:
635 	cond_resched();
636 	return 0;
637 }
638 
639 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
640 {
641 	/*
642 	 * Don't forget to update Documentation/ on changes.
643 	 */
644 	static const char mnemonics[BITS_PER_LONG][2] = {
645 		/*
646 		 * In case if we meet a flag we don't know about.
647 		 */
648 		[0 ... (BITS_PER_LONG-1)] = "??",
649 
650 		[ilog2(VM_READ)]	= "rd",
651 		[ilog2(VM_WRITE)]	= "wr",
652 		[ilog2(VM_EXEC)]	= "ex",
653 		[ilog2(VM_SHARED)]	= "sh",
654 		[ilog2(VM_MAYREAD)]	= "mr",
655 		[ilog2(VM_MAYWRITE)]	= "mw",
656 		[ilog2(VM_MAYEXEC)]	= "me",
657 		[ilog2(VM_MAYSHARE)]	= "ms",
658 		[ilog2(VM_GROWSDOWN)]	= "gd",
659 		[ilog2(VM_PFNMAP)]	= "pf",
660 		[ilog2(VM_LOCKED)]	= "lo",
661 		[ilog2(VM_IO)]		= "io",
662 		[ilog2(VM_SEQ_READ)]	= "sr",
663 		[ilog2(VM_RAND_READ)]	= "rr",
664 		[ilog2(VM_DONTCOPY)]	= "dc",
665 		[ilog2(VM_DONTEXPAND)]	= "de",
666 		[ilog2(VM_LOCKONFAULT)]	= "lf",
667 		[ilog2(VM_ACCOUNT)]	= "ac",
668 		[ilog2(VM_NORESERVE)]	= "nr",
669 		[ilog2(VM_HUGETLB)]	= "ht",
670 		[ilog2(VM_SYNC)]	= "sf",
671 		[ilog2(VM_ARCH_1)]	= "ar",
672 		[ilog2(VM_WIPEONFORK)]	= "wf",
673 		[ilog2(VM_DONTDUMP)]	= "dd",
674 #ifdef CONFIG_ARM64_BTI
675 		[ilog2(VM_ARM64_BTI)]	= "bt",
676 #endif
677 #ifdef CONFIG_MEM_SOFT_DIRTY
678 		[ilog2(VM_SOFTDIRTY)]	= "sd",
679 #endif
680 		[ilog2(VM_MIXEDMAP)]	= "mm",
681 		[ilog2(VM_HUGEPAGE)]	= "hg",
682 		[ilog2(VM_NOHUGEPAGE)]	= "nh",
683 		[ilog2(VM_MERGEABLE)]	= "mg",
684 		[ilog2(VM_UFFD_MISSING)]= "um",
685 		[ilog2(VM_UFFD_WP)]	= "uw",
686 #ifdef CONFIG_ARM64_MTE
687 		[ilog2(VM_MTE)]		= "mt",
688 		[ilog2(VM_MTE_ALLOWED)]	= "",
689 #endif
690 #ifdef CONFIG_ARCH_HAS_PKEYS
691 		/* These come out via ProtectionKey: */
692 		[ilog2(VM_PKEY_BIT0)]	= "",
693 		[ilog2(VM_PKEY_BIT1)]	= "",
694 		[ilog2(VM_PKEY_BIT2)]	= "",
695 		[ilog2(VM_PKEY_BIT3)]	= "",
696 #if VM_PKEY_BIT4
697 		[ilog2(VM_PKEY_BIT4)]	= "",
698 #endif
699 #endif /* CONFIG_ARCH_HAS_PKEYS */
700 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_MINOR
701 		[ilog2(VM_UFFD_MINOR)]	= "ui",
702 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_MINOR */
703 #ifdef CONFIG_X86_USER_SHADOW_STACK
704 		[ilog2(VM_SHADOW_STACK)] = "ss",
705 #endif
706 	};
707 	size_t i;
708 
709 	seq_puts(m, "VmFlags: ");
710 	for (i = 0; i < BITS_PER_LONG; i++) {
711 		if (!mnemonics[i][0])
712 			continue;
713 		if (vma->vm_flags & (1UL << i)) {
714 			seq_putc(m, mnemonics[i][0]);
715 			seq_putc(m, mnemonics[i][1]);
716 			seq_putc(m, ' ');
717 		}
718 	}
719 	seq_putc(m, '\n');
720 }
721 
722 #ifdef CONFIG_HUGETLB_PAGE
723 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
724 				 unsigned long addr, unsigned long end,
725 				 struct mm_walk *walk)
726 {
727 	struct mem_size_stats *mss = walk->private;
728 	struct vm_area_struct *vma = walk->vma;
729 	struct page *page = NULL;
730 	pte_t ptent = ptep_get(pte);
731 
732 	if (pte_present(ptent)) {
733 		page = vm_normal_page(vma, addr, ptent);
734 	} else if (is_swap_pte(ptent)) {
735 		swp_entry_t swpent = pte_to_swp_entry(ptent);
736 
737 		if (is_pfn_swap_entry(swpent))
738 			page = pfn_swap_entry_to_page(swpent);
739 	}
740 	if (page) {
741 		if (page_mapcount(page) >= 2 || hugetlb_pmd_shared(pte))
742 			mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
743 		else
744 			mss->private_hugetlb += huge_page_size(hstate_vma(vma));
745 	}
746 	return 0;
747 }
748 #else
749 #define smaps_hugetlb_range	NULL
750 #endif /* HUGETLB_PAGE */
751 
752 static const struct mm_walk_ops smaps_walk_ops = {
753 	.pmd_entry		= smaps_pte_range,
754 	.hugetlb_entry		= smaps_hugetlb_range,
755 	.walk_lock		= PGWALK_RDLOCK,
756 };
757 
758 static const struct mm_walk_ops smaps_shmem_walk_ops = {
759 	.pmd_entry		= smaps_pte_range,
760 	.hugetlb_entry		= smaps_hugetlb_range,
761 	.pte_hole		= smaps_pte_hole,
762 	.walk_lock		= PGWALK_RDLOCK,
763 };
764 
765 /*
766  * Gather mem stats from @vma with the indicated beginning
767  * address @start, and keep them in @mss.
768  *
769  * Use vm_start of @vma as the beginning address if @start is 0.
770  */
771 static void smap_gather_stats(struct vm_area_struct *vma,
772 		struct mem_size_stats *mss, unsigned long start)
773 {
774 	const struct mm_walk_ops *ops = &smaps_walk_ops;
775 
776 	/* Invalid start */
777 	if (start >= vma->vm_end)
778 		return;
779 
780 	if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
781 		/*
782 		 * For shared or readonly shmem mappings we know that all
783 		 * swapped out pages belong to the shmem object, and we can
784 		 * obtain the swap value much more efficiently. For private
785 		 * writable mappings, we might have COW pages that are
786 		 * not affected by the parent swapped out pages of the shmem
787 		 * object, so we have to distinguish them during the page walk.
788 		 * Unless we know that the shmem object (or the part mapped by
789 		 * our VMA) has no swapped out pages at all.
790 		 */
791 		unsigned long shmem_swapped = shmem_swap_usage(vma);
792 
793 		if (!start && (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
794 					!(vma->vm_flags & VM_WRITE))) {
795 			mss->swap += shmem_swapped;
796 		} else {
797 			ops = &smaps_shmem_walk_ops;
798 		}
799 	}
800 
801 	/* mmap_lock is held in m_start */
802 	if (!start)
803 		walk_page_vma(vma, ops, mss);
804 	else
805 		walk_page_range(vma->vm_mm, start, vma->vm_end, ops, mss);
806 }
807 
808 #define SEQ_PUT_DEC(str, val) \
809 		seq_put_decimal_ull_width(m, str, (val) >> 10, 8)
810 
811 /* Show the contents common for smaps and smaps_rollup */
812 static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss,
813 	bool rollup_mode)
814 {
815 	SEQ_PUT_DEC("Rss:            ", mss->resident);
816 	SEQ_PUT_DEC(" kB\nPss:            ", mss->pss >> PSS_SHIFT);
817 	SEQ_PUT_DEC(" kB\nPss_Dirty:      ", mss->pss_dirty >> PSS_SHIFT);
818 	if (rollup_mode) {
819 		/*
820 		 * These are meaningful only for smaps_rollup, otherwise two of
821 		 * them are zero, and the other one is the same as Pss.
822 		 */
823 		SEQ_PUT_DEC(" kB\nPss_Anon:       ",
824 			mss->pss_anon >> PSS_SHIFT);
825 		SEQ_PUT_DEC(" kB\nPss_File:       ",
826 			mss->pss_file >> PSS_SHIFT);
827 		SEQ_PUT_DEC(" kB\nPss_Shmem:      ",
828 			mss->pss_shmem >> PSS_SHIFT);
829 	}
830 	SEQ_PUT_DEC(" kB\nShared_Clean:   ", mss->shared_clean);
831 	SEQ_PUT_DEC(" kB\nShared_Dirty:   ", mss->shared_dirty);
832 	SEQ_PUT_DEC(" kB\nPrivate_Clean:  ", mss->private_clean);
833 	SEQ_PUT_DEC(" kB\nPrivate_Dirty:  ", mss->private_dirty);
834 	SEQ_PUT_DEC(" kB\nReferenced:     ", mss->referenced);
835 	SEQ_PUT_DEC(" kB\nAnonymous:      ", mss->anonymous);
836 	SEQ_PUT_DEC(" kB\nKSM:            ", mss->ksm);
837 	SEQ_PUT_DEC(" kB\nLazyFree:       ", mss->lazyfree);
838 	SEQ_PUT_DEC(" kB\nAnonHugePages:  ", mss->anonymous_thp);
839 	SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp);
840 	SEQ_PUT_DEC(" kB\nFilePmdMapped:  ", mss->file_thp);
841 	SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb);
842 	seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ",
843 				  mss->private_hugetlb >> 10, 7);
844 	SEQ_PUT_DEC(" kB\nSwap:           ", mss->swap);
845 	SEQ_PUT_DEC(" kB\nSwapPss:        ",
846 					mss->swap_pss >> PSS_SHIFT);
847 	SEQ_PUT_DEC(" kB\nLocked:         ",
848 					mss->pss_locked >> PSS_SHIFT);
849 	seq_puts(m, " kB\n");
850 }
851 
852 static int show_smap(struct seq_file *m, void *v)
853 {
854 	struct vm_area_struct *vma = v;
855 	struct mem_size_stats mss = {};
856 
857 	smap_gather_stats(vma, &mss, 0);
858 
859 	show_map_vma(m, vma);
860 
861 	SEQ_PUT_DEC("Size:           ", vma->vm_end - vma->vm_start);
862 	SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma));
863 	SEQ_PUT_DEC(" kB\nMMUPageSize:    ", vma_mmu_pagesize(vma));
864 	seq_puts(m, " kB\n");
865 
866 	__show_smap(m, &mss, false);
867 
868 	seq_printf(m, "THPeligible:    %8u\n",
869 		   !!thp_vma_allowable_orders(vma, vma->vm_flags, true, false,
870 					      true, THP_ORDERS_ALL));
871 
872 	if (arch_pkeys_enabled())
873 		seq_printf(m, "ProtectionKey:  %8u\n", vma_pkey(vma));
874 	show_smap_vma_flags(m, vma);
875 
876 	return 0;
877 }
878 
879 static int show_smaps_rollup(struct seq_file *m, void *v)
880 {
881 	struct proc_maps_private *priv = m->private;
882 	struct mem_size_stats mss = {};
883 	struct mm_struct *mm = priv->mm;
884 	struct vm_area_struct *vma;
885 	unsigned long vma_start = 0, last_vma_end = 0;
886 	int ret = 0;
887 	VMA_ITERATOR(vmi, mm, 0);
888 
889 	priv->task = get_proc_task(priv->inode);
890 	if (!priv->task)
891 		return -ESRCH;
892 
893 	if (!mm || !mmget_not_zero(mm)) {
894 		ret = -ESRCH;
895 		goto out_put_task;
896 	}
897 
898 	ret = mmap_read_lock_killable(mm);
899 	if (ret)
900 		goto out_put_mm;
901 
902 	hold_task_mempolicy(priv);
903 	vma = vma_next(&vmi);
904 
905 	if (unlikely(!vma))
906 		goto empty_set;
907 
908 	vma_start = vma->vm_start;
909 	do {
910 		smap_gather_stats(vma, &mss, 0);
911 		last_vma_end = vma->vm_end;
912 
913 		/*
914 		 * Release mmap_lock temporarily if someone wants to
915 		 * access it for write request.
916 		 */
917 		if (mmap_lock_is_contended(mm)) {
918 			vma_iter_invalidate(&vmi);
919 			mmap_read_unlock(mm);
920 			ret = mmap_read_lock_killable(mm);
921 			if (ret) {
922 				release_task_mempolicy(priv);
923 				goto out_put_mm;
924 			}
925 
926 			/*
927 			 * After dropping the lock, there are four cases to
928 			 * consider. See the following example for explanation.
929 			 *
930 			 *   +------+------+-----------+
931 			 *   | VMA1 | VMA2 | VMA3      |
932 			 *   +------+------+-----------+
933 			 *   |      |      |           |
934 			 *  4k     8k     16k         400k
935 			 *
936 			 * Suppose we drop the lock after reading VMA2 due to
937 			 * contention, then we get:
938 			 *
939 			 *	last_vma_end = 16k
940 			 *
941 			 * 1) VMA2 is freed, but VMA3 exists:
942 			 *
943 			 *    vma_next(vmi) will return VMA3.
944 			 *    In this case, just continue from VMA3.
945 			 *
946 			 * 2) VMA2 still exists:
947 			 *
948 			 *    vma_next(vmi) will return VMA3.
949 			 *    In this case, just continue from VMA3.
950 			 *
951 			 * 3) No more VMAs can be found:
952 			 *
953 			 *    vma_next(vmi) will return NULL.
954 			 *    No more things to do, just break.
955 			 *
956 			 * 4) (last_vma_end - 1) is the middle of a vma (VMA'):
957 			 *
958 			 *    vma_next(vmi) will return VMA' whose range
959 			 *    contains last_vma_end.
960 			 *    Iterate VMA' from last_vma_end.
961 			 */
962 			vma = vma_next(&vmi);
963 			/* Case 3 above */
964 			if (!vma)
965 				break;
966 
967 			/* Case 1 and 2 above */
968 			if (vma->vm_start >= last_vma_end)
969 				continue;
970 
971 			/* Case 4 above */
972 			if (vma->vm_end > last_vma_end)
973 				smap_gather_stats(vma, &mss, last_vma_end);
974 		}
975 	} for_each_vma(vmi, vma);
976 
977 empty_set:
978 	show_vma_header_prefix(m, vma_start, last_vma_end, 0, 0, 0, 0);
979 	seq_pad(m, ' ');
980 	seq_puts(m, "[rollup]\n");
981 
982 	__show_smap(m, &mss, true);
983 
984 	release_task_mempolicy(priv);
985 	mmap_read_unlock(mm);
986 
987 out_put_mm:
988 	mmput(mm);
989 out_put_task:
990 	put_task_struct(priv->task);
991 	priv->task = NULL;
992 
993 	return ret;
994 }
995 #undef SEQ_PUT_DEC
996 
997 static const struct seq_operations proc_pid_smaps_op = {
998 	.start	= m_start,
999 	.next	= m_next,
1000 	.stop	= m_stop,
1001 	.show	= show_smap
1002 };
1003 
1004 static int pid_smaps_open(struct inode *inode, struct file *file)
1005 {
1006 	return do_maps_open(inode, file, &proc_pid_smaps_op);
1007 }
1008 
1009 static int smaps_rollup_open(struct inode *inode, struct file *file)
1010 {
1011 	int ret;
1012 	struct proc_maps_private *priv;
1013 
1014 	priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT);
1015 	if (!priv)
1016 		return -ENOMEM;
1017 
1018 	ret = single_open(file, show_smaps_rollup, priv);
1019 	if (ret)
1020 		goto out_free;
1021 
1022 	priv->inode = inode;
1023 	priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
1024 	if (IS_ERR(priv->mm)) {
1025 		ret = PTR_ERR(priv->mm);
1026 
1027 		single_release(inode, file);
1028 		goto out_free;
1029 	}
1030 
1031 	return 0;
1032 
1033 out_free:
1034 	kfree(priv);
1035 	return ret;
1036 }
1037 
1038 static int smaps_rollup_release(struct inode *inode, struct file *file)
1039 {
1040 	struct seq_file *seq = file->private_data;
1041 	struct proc_maps_private *priv = seq->private;
1042 
1043 	if (priv->mm)
1044 		mmdrop(priv->mm);
1045 
1046 	kfree(priv);
1047 	return single_release(inode, file);
1048 }
1049 
1050 const struct file_operations proc_pid_smaps_operations = {
1051 	.open		= pid_smaps_open,
1052 	.read		= seq_read,
1053 	.llseek		= seq_lseek,
1054 	.release	= proc_map_release,
1055 };
1056 
1057 const struct file_operations proc_pid_smaps_rollup_operations = {
1058 	.open		= smaps_rollup_open,
1059 	.read		= seq_read,
1060 	.llseek		= seq_lseek,
1061 	.release	= smaps_rollup_release,
1062 };
1063 
1064 enum clear_refs_types {
1065 	CLEAR_REFS_ALL = 1,
1066 	CLEAR_REFS_ANON,
1067 	CLEAR_REFS_MAPPED,
1068 	CLEAR_REFS_SOFT_DIRTY,
1069 	CLEAR_REFS_MM_HIWATER_RSS,
1070 	CLEAR_REFS_LAST,
1071 };
1072 
1073 struct clear_refs_private {
1074 	enum clear_refs_types type;
1075 };
1076 
1077 #ifdef CONFIG_MEM_SOFT_DIRTY
1078 
1079 static inline bool pte_is_pinned(struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1080 {
1081 	struct page *page;
1082 
1083 	if (!pte_write(pte))
1084 		return false;
1085 	if (!is_cow_mapping(vma->vm_flags))
1086 		return false;
1087 	if (likely(!test_bit(MMF_HAS_PINNED, &vma->vm_mm->flags)))
1088 		return false;
1089 	page = vm_normal_page(vma, addr, pte);
1090 	if (!page)
1091 		return false;
1092 	return page_maybe_dma_pinned(page);
1093 }
1094 
1095 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1096 		unsigned long addr, pte_t *pte)
1097 {
1098 	/*
1099 	 * The soft-dirty tracker uses #PF-s to catch writes
1100 	 * to pages, so write-protect the pte as well. See the
1101 	 * Documentation/admin-guide/mm/soft-dirty.rst for full description
1102 	 * of how soft-dirty works.
1103 	 */
1104 	pte_t ptent = ptep_get(pte);
1105 
1106 	if (pte_present(ptent)) {
1107 		pte_t old_pte;
1108 
1109 		if (pte_is_pinned(vma, addr, ptent))
1110 			return;
1111 		old_pte = ptep_modify_prot_start(vma, addr, pte);
1112 		ptent = pte_wrprotect(old_pte);
1113 		ptent = pte_clear_soft_dirty(ptent);
1114 		ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent);
1115 	} else if (is_swap_pte(ptent)) {
1116 		ptent = pte_swp_clear_soft_dirty(ptent);
1117 		set_pte_at(vma->vm_mm, addr, pte, ptent);
1118 	}
1119 }
1120 #else
1121 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1122 		unsigned long addr, pte_t *pte)
1123 {
1124 }
1125 #endif
1126 
1127 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
1128 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1129 		unsigned long addr, pmd_t *pmdp)
1130 {
1131 	pmd_t old, pmd = *pmdp;
1132 
1133 	if (pmd_present(pmd)) {
1134 		/* See comment in change_huge_pmd() */
1135 		old = pmdp_invalidate(vma, addr, pmdp);
1136 		if (pmd_dirty(old))
1137 			pmd = pmd_mkdirty(pmd);
1138 		if (pmd_young(old))
1139 			pmd = pmd_mkyoung(pmd);
1140 
1141 		pmd = pmd_wrprotect(pmd);
1142 		pmd = pmd_clear_soft_dirty(pmd);
1143 
1144 		set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1145 	} else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1146 		pmd = pmd_swp_clear_soft_dirty(pmd);
1147 		set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1148 	}
1149 }
1150 #else
1151 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1152 		unsigned long addr, pmd_t *pmdp)
1153 {
1154 }
1155 #endif
1156 
1157 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
1158 				unsigned long end, struct mm_walk *walk)
1159 {
1160 	struct clear_refs_private *cp = walk->private;
1161 	struct vm_area_struct *vma = walk->vma;
1162 	pte_t *pte, ptent;
1163 	spinlock_t *ptl;
1164 	struct page *page;
1165 
1166 	ptl = pmd_trans_huge_lock(pmd, vma);
1167 	if (ptl) {
1168 		if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1169 			clear_soft_dirty_pmd(vma, addr, pmd);
1170 			goto out;
1171 		}
1172 
1173 		if (!pmd_present(*pmd))
1174 			goto out;
1175 
1176 		page = pmd_page(*pmd);
1177 
1178 		/* Clear accessed and referenced bits. */
1179 		pmdp_test_and_clear_young(vma, addr, pmd);
1180 		test_and_clear_page_young(page);
1181 		ClearPageReferenced(page);
1182 out:
1183 		spin_unlock(ptl);
1184 		return 0;
1185 	}
1186 
1187 	pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1188 	if (!pte) {
1189 		walk->action = ACTION_AGAIN;
1190 		return 0;
1191 	}
1192 	for (; addr != end; pte++, addr += PAGE_SIZE) {
1193 		ptent = ptep_get(pte);
1194 
1195 		if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1196 			clear_soft_dirty(vma, addr, pte);
1197 			continue;
1198 		}
1199 
1200 		if (!pte_present(ptent))
1201 			continue;
1202 
1203 		page = vm_normal_page(vma, addr, ptent);
1204 		if (!page)
1205 			continue;
1206 
1207 		/* Clear accessed and referenced bits. */
1208 		ptep_test_and_clear_young(vma, addr, pte);
1209 		test_and_clear_page_young(page);
1210 		ClearPageReferenced(page);
1211 	}
1212 	pte_unmap_unlock(pte - 1, ptl);
1213 	cond_resched();
1214 	return 0;
1215 }
1216 
1217 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1218 				struct mm_walk *walk)
1219 {
1220 	struct clear_refs_private *cp = walk->private;
1221 	struct vm_area_struct *vma = walk->vma;
1222 
1223 	if (vma->vm_flags & VM_PFNMAP)
1224 		return 1;
1225 
1226 	/*
1227 	 * Writing 1 to /proc/pid/clear_refs affects all pages.
1228 	 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1229 	 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1230 	 * Writing 4 to /proc/pid/clear_refs affects all pages.
1231 	 */
1232 	if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1233 		return 1;
1234 	if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1235 		return 1;
1236 	return 0;
1237 }
1238 
1239 static const struct mm_walk_ops clear_refs_walk_ops = {
1240 	.pmd_entry		= clear_refs_pte_range,
1241 	.test_walk		= clear_refs_test_walk,
1242 	.walk_lock		= PGWALK_WRLOCK,
1243 };
1244 
1245 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1246 				size_t count, loff_t *ppos)
1247 {
1248 	struct task_struct *task;
1249 	char buffer[PROC_NUMBUF] = {};
1250 	struct mm_struct *mm;
1251 	struct vm_area_struct *vma;
1252 	enum clear_refs_types type;
1253 	int itype;
1254 	int rv;
1255 
1256 	if (count > sizeof(buffer) - 1)
1257 		count = sizeof(buffer) - 1;
1258 	if (copy_from_user(buffer, buf, count))
1259 		return -EFAULT;
1260 	rv = kstrtoint(strstrip(buffer), 10, &itype);
1261 	if (rv < 0)
1262 		return rv;
1263 	type = (enum clear_refs_types)itype;
1264 	if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1265 		return -EINVAL;
1266 
1267 	task = get_proc_task(file_inode(file));
1268 	if (!task)
1269 		return -ESRCH;
1270 	mm = get_task_mm(task);
1271 	if (mm) {
1272 		VMA_ITERATOR(vmi, mm, 0);
1273 		struct mmu_notifier_range range;
1274 		struct clear_refs_private cp = {
1275 			.type = type,
1276 		};
1277 
1278 		if (mmap_write_lock_killable(mm)) {
1279 			count = -EINTR;
1280 			goto out_mm;
1281 		}
1282 		if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1283 			/*
1284 			 * Writing 5 to /proc/pid/clear_refs resets the peak
1285 			 * resident set size to this mm's current rss value.
1286 			 */
1287 			reset_mm_hiwater_rss(mm);
1288 			goto out_unlock;
1289 		}
1290 
1291 		if (type == CLEAR_REFS_SOFT_DIRTY) {
1292 			for_each_vma(vmi, vma) {
1293 				if (!(vma->vm_flags & VM_SOFTDIRTY))
1294 					continue;
1295 				vm_flags_clear(vma, VM_SOFTDIRTY);
1296 				vma_set_page_prot(vma);
1297 			}
1298 
1299 			inc_tlb_flush_pending(mm);
1300 			mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY,
1301 						0, mm, 0, -1UL);
1302 			mmu_notifier_invalidate_range_start(&range);
1303 		}
1304 		walk_page_range(mm, 0, -1, &clear_refs_walk_ops, &cp);
1305 		if (type == CLEAR_REFS_SOFT_DIRTY) {
1306 			mmu_notifier_invalidate_range_end(&range);
1307 			flush_tlb_mm(mm);
1308 			dec_tlb_flush_pending(mm);
1309 		}
1310 out_unlock:
1311 		mmap_write_unlock(mm);
1312 out_mm:
1313 		mmput(mm);
1314 	}
1315 	put_task_struct(task);
1316 
1317 	return count;
1318 }
1319 
1320 const struct file_operations proc_clear_refs_operations = {
1321 	.write		= clear_refs_write,
1322 	.llseek		= noop_llseek,
1323 };
1324 
1325 typedef struct {
1326 	u64 pme;
1327 } pagemap_entry_t;
1328 
1329 struct pagemapread {
1330 	int pos, len;		/* units: PM_ENTRY_BYTES, not bytes */
1331 	pagemap_entry_t *buffer;
1332 	bool show_pfn;
1333 };
1334 
1335 #define PAGEMAP_WALK_SIZE	(PMD_SIZE)
1336 #define PAGEMAP_WALK_MASK	(PMD_MASK)
1337 
1338 #define PM_ENTRY_BYTES		sizeof(pagemap_entry_t)
1339 #define PM_PFRAME_BITS		55
1340 #define PM_PFRAME_MASK		GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1341 #define PM_SOFT_DIRTY		BIT_ULL(55)
1342 #define PM_MMAP_EXCLUSIVE	BIT_ULL(56)
1343 #define PM_UFFD_WP		BIT_ULL(57)
1344 #define PM_FILE			BIT_ULL(61)
1345 #define PM_SWAP			BIT_ULL(62)
1346 #define PM_PRESENT		BIT_ULL(63)
1347 
1348 #define PM_END_OF_BUFFER    1
1349 
1350 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1351 {
1352 	return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1353 }
1354 
1355 static int add_to_pagemap(pagemap_entry_t *pme, struct pagemapread *pm)
1356 {
1357 	pm->buffer[pm->pos++] = *pme;
1358 	if (pm->pos >= pm->len)
1359 		return PM_END_OF_BUFFER;
1360 	return 0;
1361 }
1362 
1363 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1364 			    __always_unused int depth, struct mm_walk *walk)
1365 {
1366 	struct pagemapread *pm = walk->private;
1367 	unsigned long addr = start;
1368 	int err = 0;
1369 
1370 	while (addr < end) {
1371 		struct vm_area_struct *vma = find_vma(walk->mm, addr);
1372 		pagemap_entry_t pme = make_pme(0, 0);
1373 		/* End of address space hole, which we mark as non-present. */
1374 		unsigned long hole_end;
1375 
1376 		if (vma)
1377 			hole_end = min(end, vma->vm_start);
1378 		else
1379 			hole_end = end;
1380 
1381 		for (; addr < hole_end; addr += PAGE_SIZE) {
1382 			err = add_to_pagemap(&pme, pm);
1383 			if (err)
1384 				goto out;
1385 		}
1386 
1387 		if (!vma)
1388 			break;
1389 
1390 		/* Addresses in the VMA. */
1391 		if (vma->vm_flags & VM_SOFTDIRTY)
1392 			pme = make_pme(0, PM_SOFT_DIRTY);
1393 		for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1394 			err = add_to_pagemap(&pme, pm);
1395 			if (err)
1396 				goto out;
1397 		}
1398 	}
1399 out:
1400 	return err;
1401 }
1402 
1403 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1404 		struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1405 {
1406 	u64 frame = 0, flags = 0;
1407 	struct page *page = NULL;
1408 	bool migration = false;
1409 
1410 	if (pte_present(pte)) {
1411 		if (pm->show_pfn)
1412 			frame = pte_pfn(pte);
1413 		flags |= PM_PRESENT;
1414 		page = vm_normal_page(vma, addr, pte);
1415 		if (pte_soft_dirty(pte))
1416 			flags |= PM_SOFT_DIRTY;
1417 		if (pte_uffd_wp(pte))
1418 			flags |= PM_UFFD_WP;
1419 	} else if (is_swap_pte(pte)) {
1420 		swp_entry_t entry;
1421 		if (pte_swp_soft_dirty(pte))
1422 			flags |= PM_SOFT_DIRTY;
1423 		if (pte_swp_uffd_wp(pte))
1424 			flags |= PM_UFFD_WP;
1425 		entry = pte_to_swp_entry(pte);
1426 		if (pm->show_pfn) {
1427 			pgoff_t offset;
1428 			/*
1429 			 * For PFN swap offsets, keeping the offset field
1430 			 * to be PFN only to be compatible with old smaps.
1431 			 */
1432 			if (is_pfn_swap_entry(entry))
1433 				offset = swp_offset_pfn(entry);
1434 			else
1435 				offset = swp_offset(entry);
1436 			frame = swp_type(entry) |
1437 			    (offset << MAX_SWAPFILES_SHIFT);
1438 		}
1439 		flags |= PM_SWAP;
1440 		migration = is_migration_entry(entry);
1441 		if (is_pfn_swap_entry(entry))
1442 			page = pfn_swap_entry_to_page(entry);
1443 		if (pte_marker_entry_uffd_wp(entry))
1444 			flags |= PM_UFFD_WP;
1445 	}
1446 
1447 	if (page && !PageAnon(page))
1448 		flags |= PM_FILE;
1449 	if (page && !migration && page_mapcount(page) == 1)
1450 		flags |= PM_MMAP_EXCLUSIVE;
1451 	if (vma->vm_flags & VM_SOFTDIRTY)
1452 		flags |= PM_SOFT_DIRTY;
1453 
1454 	return make_pme(frame, flags);
1455 }
1456 
1457 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1458 			     struct mm_walk *walk)
1459 {
1460 	struct vm_area_struct *vma = walk->vma;
1461 	struct pagemapread *pm = walk->private;
1462 	spinlock_t *ptl;
1463 	pte_t *pte, *orig_pte;
1464 	int err = 0;
1465 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1466 	bool migration = false;
1467 
1468 	ptl = pmd_trans_huge_lock(pmdp, vma);
1469 	if (ptl) {
1470 		u64 flags = 0, frame = 0;
1471 		pmd_t pmd = *pmdp;
1472 		struct page *page = NULL;
1473 
1474 		if (vma->vm_flags & VM_SOFTDIRTY)
1475 			flags |= PM_SOFT_DIRTY;
1476 
1477 		if (pmd_present(pmd)) {
1478 			page = pmd_page(pmd);
1479 
1480 			flags |= PM_PRESENT;
1481 			if (pmd_soft_dirty(pmd))
1482 				flags |= PM_SOFT_DIRTY;
1483 			if (pmd_uffd_wp(pmd))
1484 				flags |= PM_UFFD_WP;
1485 			if (pm->show_pfn)
1486 				frame = pmd_pfn(pmd) +
1487 					((addr & ~PMD_MASK) >> PAGE_SHIFT);
1488 		}
1489 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1490 		else if (is_swap_pmd(pmd)) {
1491 			swp_entry_t entry = pmd_to_swp_entry(pmd);
1492 			unsigned long offset;
1493 
1494 			if (pm->show_pfn) {
1495 				if (is_pfn_swap_entry(entry))
1496 					offset = swp_offset_pfn(entry);
1497 				else
1498 					offset = swp_offset(entry);
1499 				offset = offset +
1500 					((addr & ~PMD_MASK) >> PAGE_SHIFT);
1501 				frame = swp_type(entry) |
1502 					(offset << MAX_SWAPFILES_SHIFT);
1503 			}
1504 			flags |= PM_SWAP;
1505 			if (pmd_swp_soft_dirty(pmd))
1506 				flags |= PM_SOFT_DIRTY;
1507 			if (pmd_swp_uffd_wp(pmd))
1508 				flags |= PM_UFFD_WP;
1509 			VM_BUG_ON(!is_pmd_migration_entry(pmd));
1510 			migration = is_migration_entry(entry);
1511 			page = pfn_swap_entry_to_page(entry);
1512 		}
1513 #endif
1514 
1515 		if (page && !migration && page_mapcount(page) == 1)
1516 			flags |= PM_MMAP_EXCLUSIVE;
1517 
1518 		for (; addr != end; addr += PAGE_SIZE) {
1519 			pagemap_entry_t pme = make_pme(frame, flags);
1520 
1521 			err = add_to_pagemap(&pme, pm);
1522 			if (err)
1523 				break;
1524 			if (pm->show_pfn) {
1525 				if (flags & PM_PRESENT)
1526 					frame++;
1527 				else if (flags & PM_SWAP)
1528 					frame += (1 << MAX_SWAPFILES_SHIFT);
1529 			}
1530 		}
1531 		spin_unlock(ptl);
1532 		return err;
1533 	}
1534 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1535 
1536 	/*
1537 	 * We can assume that @vma always points to a valid one and @end never
1538 	 * goes beyond vma->vm_end.
1539 	 */
1540 	orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1541 	if (!pte) {
1542 		walk->action = ACTION_AGAIN;
1543 		return err;
1544 	}
1545 	for (; addr < end; pte++, addr += PAGE_SIZE) {
1546 		pagemap_entry_t pme;
1547 
1548 		pme = pte_to_pagemap_entry(pm, vma, addr, ptep_get(pte));
1549 		err = add_to_pagemap(&pme, pm);
1550 		if (err)
1551 			break;
1552 	}
1553 	pte_unmap_unlock(orig_pte, ptl);
1554 
1555 	cond_resched();
1556 
1557 	return err;
1558 }
1559 
1560 #ifdef CONFIG_HUGETLB_PAGE
1561 /* This function walks within one hugetlb entry in the single call */
1562 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1563 				 unsigned long addr, unsigned long end,
1564 				 struct mm_walk *walk)
1565 {
1566 	struct pagemapread *pm = walk->private;
1567 	struct vm_area_struct *vma = walk->vma;
1568 	u64 flags = 0, frame = 0;
1569 	int err = 0;
1570 	pte_t pte;
1571 
1572 	if (vma->vm_flags & VM_SOFTDIRTY)
1573 		flags |= PM_SOFT_DIRTY;
1574 
1575 	pte = huge_ptep_get(ptep);
1576 	if (pte_present(pte)) {
1577 		struct page *page = pte_page(pte);
1578 
1579 		if (!PageAnon(page))
1580 			flags |= PM_FILE;
1581 
1582 		if (page_mapcount(page) == 1)
1583 			flags |= PM_MMAP_EXCLUSIVE;
1584 
1585 		if (huge_pte_uffd_wp(pte))
1586 			flags |= PM_UFFD_WP;
1587 
1588 		flags |= PM_PRESENT;
1589 		if (pm->show_pfn)
1590 			frame = pte_pfn(pte) +
1591 				((addr & ~hmask) >> PAGE_SHIFT);
1592 	} else if (pte_swp_uffd_wp_any(pte)) {
1593 		flags |= PM_UFFD_WP;
1594 	}
1595 
1596 	for (; addr != end; addr += PAGE_SIZE) {
1597 		pagemap_entry_t pme = make_pme(frame, flags);
1598 
1599 		err = add_to_pagemap(&pme, pm);
1600 		if (err)
1601 			return err;
1602 		if (pm->show_pfn && (flags & PM_PRESENT))
1603 			frame++;
1604 	}
1605 
1606 	cond_resched();
1607 
1608 	return err;
1609 }
1610 #else
1611 #define pagemap_hugetlb_range	NULL
1612 #endif /* HUGETLB_PAGE */
1613 
1614 static const struct mm_walk_ops pagemap_ops = {
1615 	.pmd_entry	= pagemap_pmd_range,
1616 	.pte_hole	= pagemap_pte_hole,
1617 	.hugetlb_entry	= pagemap_hugetlb_range,
1618 	.walk_lock	= PGWALK_RDLOCK,
1619 };
1620 
1621 /*
1622  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1623  *
1624  * For each page in the address space, this file contains one 64-bit entry
1625  * consisting of the following:
1626  *
1627  * Bits 0-54  page frame number (PFN) if present
1628  * Bits 0-4   swap type if swapped
1629  * Bits 5-54  swap offset if swapped
1630  * Bit  55    pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst)
1631  * Bit  56    page exclusively mapped
1632  * Bit  57    pte is uffd-wp write-protected
1633  * Bits 58-60 zero
1634  * Bit  61    page is file-page or shared-anon
1635  * Bit  62    page swapped
1636  * Bit  63    page present
1637  *
1638  * If the page is not present but in swap, then the PFN contains an
1639  * encoding of the swap file number and the page's offset into the
1640  * swap. Unmapped pages return a null PFN. This allows determining
1641  * precisely which pages are mapped (or in swap) and comparing mapped
1642  * pages between processes.
1643  *
1644  * Efficient users of this interface will use /proc/pid/maps to
1645  * determine which areas of memory are actually mapped and llseek to
1646  * skip over unmapped regions.
1647  */
1648 static ssize_t pagemap_read(struct file *file, char __user *buf,
1649 			    size_t count, loff_t *ppos)
1650 {
1651 	struct mm_struct *mm = file->private_data;
1652 	struct pagemapread pm;
1653 	unsigned long src;
1654 	unsigned long svpfn;
1655 	unsigned long start_vaddr;
1656 	unsigned long end_vaddr;
1657 	int ret = 0, copied = 0;
1658 
1659 	if (!mm || !mmget_not_zero(mm))
1660 		goto out;
1661 
1662 	ret = -EINVAL;
1663 	/* file position must be aligned */
1664 	if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1665 		goto out_mm;
1666 
1667 	ret = 0;
1668 	if (!count)
1669 		goto out_mm;
1670 
1671 	/* do not disclose physical addresses: attack vector */
1672 	pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1673 
1674 	pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1675 	pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL);
1676 	ret = -ENOMEM;
1677 	if (!pm.buffer)
1678 		goto out_mm;
1679 
1680 	src = *ppos;
1681 	svpfn = src / PM_ENTRY_BYTES;
1682 	end_vaddr = mm->task_size;
1683 
1684 	/* watch out for wraparound */
1685 	start_vaddr = end_vaddr;
1686 	if (svpfn <= (ULONG_MAX >> PAGE_SHIFT)) {
1687 		unsigned long end;
1688 
1689 		ret = mmap_read_lock_killable(mm);
1690 		if (ret)
1691 			goto out_free;
1692 		start_vaddr = untagged_addr_remote(mm, svpfn << PAGE_SHIFT);
1693 		mmap_read_unlock(mm);
1694 
1695 		end = start_vaddr + ((count / PM_ENTRY_BYTES) << PAGE_SHIFT);
1696 		if (end >= start_vaddr && end < mm->task_size)
1697 			end_vaddr = end;
1698 	}
1699 
1700 	/* Ensure the address is inside the task */
1701 	if (start_vaddr > mm->task_size)
1702 		start_vaddr = end_vaddr;
1703 
1704 	ret = 0;
1705 	while (count && (start_vaddr < end_vaddr)) {
1706 		int len;
1707 		unsigned long end;
1708 
1709 		pm.pos = 0;
1710 		end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1711 		/* overflow ? */
1712 		if (end < start_vaddr || end > end_vaddr)
1713 			end = end_vaddr;
1714 		ret = mmap_read_lock_killable(mm);
1715 		if (ret)
1716 			goto out_free;
1717 		ret = walk_page_range(mm, start_vaddr, end, &pagemap_ops, &pm);
1718 		mmap_read_unlock(mm);
1719 		start_vaddr = end;
1720 
1721 		len = min(count, PM_ENTRY_BYTES * pm.pos);
1722 		if (copy_to_user(buf, pm.buffer, len)) {
1723 			ret = -EFAULT;
1724 			goto out_free;
1725 		}
1726 		copied += len;
1727 		buf += len;
1728 		count -= len;
1729 	}
1730 	*ppos += copied;
1731 	if (!ret || ret == PM_END_OF_BUFFER)
1732 		ret = copied;
1733 
1734 out_free:
1735 	kfree(pm.buffer);
1736 out_mm:
1737 	mmput(mm);
1738 out:
1739 	return ret;
1740 }
1741 
1742 static int pagemap_open(struct inode *inode, struct file *file)
1743 {
1744 	struct mm_struct *mm;
1745 
1746 	mm = proc_mem_open(inode, PTRACE_MODE_READ);
1747 	if (IS_ERR(mm))
1748 		return PTR_ERR(mm);
1749 	file->private_data = mm;
1750 	return 0;
1751 }
1752 
1753 static int pagemap_release(struct inode *inode, struct file *file)
1754 {
1755 	struct mm_struct *mm = file->private_data;
1756 
1757 	if (mm)
1758 		mmdrop(mm);
1759 	return 0;
1760 }
1761 
1762 #define PM_SCAN_CATEGORIES	(PAGE_IS_WPALLOWED | PAGE_IS_WRITTEN |	\
1763 				 PAGE_IS_FILE |	PAGE_IS_PRESENT |	\
1764 				 PAGE_IS_SWAPPED | PAGE_IS_PFNZERO |	\
1765 				 PAGE_IS_HUGE | PAGE_IS_SOFT_DIRTY)
1766 #define PM_SCAN_FLAGS		(PM_SCAN_WP_MATCHING | PM_SCAN_CHECK_WPASYNC)
1767 
1768 struct pagemap_scan_private {
1769 	struct pm_scan_arg arg;
1770 	unsigned long masks_of_interest, cur_vma_category;
1771 	struct page_region *vec_buf;
1772 	unsigned long vec_buf_len, vec_buf_index, found_pages;
1773 	struct page_region __user *vec_out;
1774 };
1775 
1776 static unsigned long pagemap_page_category(struct pagemap_scan_private *p,
1777 					   struct vm_area_struct *vma,
1778 					   unsigned long addr, pte_t pte)
1779 {
1780 	unsigned long categories = 0;
1781 
1782 	if (pte_present(pte)) {
1783 		struct page *page;
1784 
1785 		categories |= PAGE_IS_PRESENT;
1786 		if (!pte_uffd_wp(pte))
1787 			categories |= PAGE_IS_WRITTEN;
1788 
1789 		if (p->masks_of_interest & PAGE_IS_FILE) {
1790 			page = vm_normal_page(vma, addr, pte);
1791 			if (page && !PageAnon(page))
1792 				categories |= PAGE_IS_FILE;
1793 		}
1794 
1795 		if (is_zero_pfn(pte_pfn(pte)))
1796 			categories |= PAGE_IS_PFNZERO;
1797 		if (pte_soft_dirty(pte))
1798 			categories |= PAGE_IS_SOFT_DIRTY;
1799 	} else if (is_swap_pte(pte)) {
1800 		swp_entry_t swp;
1801 
1802 		categories |= PAGE_IS_SWAPPED;
1803 		if (!pte_swp_uffd_wp_any(pte))
1804 			categories |= PAGE_IS_WRITTEN;
1805 
1806 		if (p->masks_of_interest & PAGE_IS_FILE) {
1807 			swp = pte_to_swp_entry(pte);
1808 			if (is_pfn_swap_entry(swp) &&
1809 			    !folio_test_anon(pfn_swap_entry_folio(swp)))
1810 				categories |= PAGE_IS_FILE;
1811 		}
1812 		if (pte_swp_soft_dirty(pte))
1813 			categories |= PAGE_IS_SOFT_DIRTY;
1814 	}
1815 
1816 	return categories;
1817 }
1818 
1819 static void make_uffd_wp_pte(struct vm_area_struct *vma,
1820 			     unsigned long addr, pte_t *pte, pte_t ptent)
1821 {
1822 	if (pte_present(ptent)) {
1823 		pte_t old_pte;
1824 
1825 		old_pte = ptep_modify_prot_start(vma, addr, pte);
1826 		ptent = pte_mkuffd_wp(old_pte);
1827 		ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent);
1828 	} else if (is_swap_pte(ptent)) {
1829 		ptent = pte_swp_mkuffd_wp(ptent);
1830 		set_pte_at(vma->vm_mm, addr, pte, ptent);
1831 	} else {
1832 		set_pte_at(vma->vm_mm, addr, pte,
1833 			   make_pte_marker(PTE_MARKER_UFFD_WP));
1834 	}
1835 }
1836 
1837 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1838 static unsigned long pagemap_thp_category(struct pagemap_scan_private *p,
1839 					  struct vm_area_struct *vma,
1840 					  unsigned long addr, pmd_t pmd)
1841 {
1842 	unsigned long categories = PAGE_IS_HUGE;
1843 
1844 	if (pmd_present(pmd)) {
1845 		struct page *page;
1846 
1847 		categories |= PAGE_IS_PRESENT;
1848 		if (!pmd_uffd_wp(pmd))
1849 			categories |= PAGE_IS_WRITTEN;
1850 
1851 		if (p->masks_of_interest & PAGE_IS_FILE) {
1852 			page = vm_normal_page_pmd(vma, addr, pmd);
1853 			if (page && !PageAnon(page))
1854 				categories |= PAGE_IS_FILE;
1855 		}
1856 
1857 		if (is_zero_pfn(pmd_pfn(pmd)))
1858 			categories |= PAGE_IS_PFNZERO;
1859 		if (pmd_soft_dirty(pmd))
1860 			categories |= PAGE_IS_SOFT_DIRTY;
1861 	} else if (is_swap_pmd(pmd)) {
1862 		swp_entry_t swp;
1863 
1864 		categories |= PAGE_IS_SWAPPED;
1865 		if (!pmd_swp_uffd_wp(pmd))
1866 			categories |= PAGE_IS_WRITTEN;
1867 		if (pmd_swp_soft_dirty(pmd))
1868 			categories |= PAGE_IS_SOFT_DIRTY;
1869 
1870 		if (p->masks_of_interest & PAGE_IS_FILE) {
1871 			swp = pmd_to_swp_entry(pmd);
1872 			if (is_pfn_swap_entry(swp) &&
1873 			    !folio_test_anon(pfn_swap_entry_folio(swp)))
1874 				categories |= PAGE_IS_FILE;
1875 		}
1876 	}
1877 
1878 	return categories;
1879 }
1880 
1881 static void make_uffd_wp_pmd(struct vm_area_struct *vma,
1882 			     unsigned long addr, pmd_t *pmdp)
1883 {
1884 	pmd_t old, pmd = *pmdp;
1885 
1886 	if (pmd_present(pmd)) {
1887 		old = pmdp_invalidate_ad(vma, addr, pmdp);
1888 		pmd = pmd_mkuffd_wp(old);
1889 		set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1890 	} else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1891 		pmd = pmd_swp_mkuffd_wp(pmd);
1892 		set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1893 	}
1894 }
1895 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1896 
1897 #ifdef CONFIG_HUGETLB_PAGE
1898 static unsigned long pagemap_hugetlb_category(pte_t pte)
1899 {
1900 	unsigned long categories = PAGE_IS_HUGE;
1901 
1902 	/*
1903 	 * According to pagemap_hugetlb_range(), file-backed HugeTLB
1904 	 * page cannot be swapped. So PAGE_IS_FILE is not checked for
1905 	 * swapped pages.
1906 	 */
1907 	if (pte_present(pte)) {
1908 		categories |= PAGE_IS_PRESENT;
1909 		if (!huge_pte_uffd_wp(pte))
1910 			categories |= PAGE_IS_WRITTEN;
1911 		if (!PageAnon(pte_page(pte)))
1912 			categories |= PAGE_IS_FILE;
1913 		if (is_zero_pfn(pte_pfn(pte)))
1914 			categories |= PAGE_IS_PFNZERO;
1915 		if (pte_soft_dirty(pte))
1916 			categories |= PAGE_IS_SOFT_DIRTY;
1917 	} else if (is_swap_pte(pte)) {
1918 		categories |= PAGE_IS_SWAPPED;
1919 		if (!pte_swp_uffd_wp_any(pte))
1920 			categories |= PAGE_IS_WRITTEN;
1921 		if (pte_swp_soft_dirty(pte))
1922 			categories |= PAGE_IS_SOFT_DIRTY;
1923 	}
1924 
1925 	return categories;
1926 }
1927 
1928 static void make_uffd_wp_huge_pte(struct vm_area_struct *vma,
1929 				  unsigned long addr, pte_t *ptep,
1930 				  pte_t ptent)
1931 {
1932 	unsigned long psize;
1933 
1934 	if (is_hugetlb_entry_hwpoisoned(ptent) || is_pte_marker(ptent))
1935 		return;
1936 
1937 	psize = huge_page_size(hstate_vma(vma));
1938 
1939 	if (is_hugetlb_entry_migration(ptent))
1940 		set_huge_pte_at(vma->vm_mm, addr, ptep,
1941 				pte_swp_mkuffd_wp(ptent), psize);
1942 	else if (!huge_pte_none(ptent))
1943 		huge_ptep_modify_prot_commit(vma, addr, ptep, ptent,
1944 					     huge_pte_mkuffd_wp(ptent));
1945 	else
1946 		set_huge_pte_at(vma->vm_mm, addr, ptep,
1947 				make_pte_marker(PTE_MARKER_UFFD_WP), psize);
1948 }
1949 #endif /* CONFIG_HUGETLB_PAGE */
1950 
1951 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLB_PAGE)
1952 static void pagemap_scan_backout_range(struct pagemap_scan_private *p,
1953 				       unsigned long addr, unsigned long end)
1954 {
1955 	struct page_region *cur_buf = &p->vec_buf[p->vec_buf_index];
1956 
1957 	if (cur_buf->start != addr)
1958 		cur_buf->end = addr;
1959 	else
1960 		cur_buf->start = cur_buf->end = 0;
1961 
1962 	p->found_pages -= (end - addr) / PAGE_SIZE;
1963 }
1964 #endif
1965 
1966 static bool pagemap_scan_is_interesting_page(unsigned long categories,
1967 					     const struct pagemap_scan_private *p)
1968 {
1969 	categories ^= p->arg.category_inverted;
1970 	if ((categories & p->arg.category_mask) != p->arg.category_mask)
1971 		return false;
1972 	if (p->arg.category_anyof_mask && !(categories & p->arg.category_anyof_mask))
1973 		return false;
1974 
1975 	return true;
1976 }
1977 
1978 static bool pagemap_scan_is_interesting_vma(unsigned long categories,
1979 					    const struct pagemap_scan_private *p)
1980 {
1981 	unsigned long required = p->arg.category_mask & PAGE_IS_WPALLOWED;
1982 
1983 	categories ^= p->arg.category_inverted;
1984 	if ((categories & required) != required)
1985 		return false;
1986 
1987 	return true;
1988 }
1989 
1990 static int pagemap_scan_test_walk(unsigned long start, unsigned long end,
1991 				  struct mm_walk *walk)
1992 {
1993 	struct pagemap_scan_private *p = walk->private;
1994 	struct vm_area_struct *vma = walk->vma;
1995 	unsigned long vma_category = 0;
1996 	bool wp_allowed = userfaultfd_wp_async(vma) &&
1997 	    userfaultfd_wp_use_markers(vma);
1998 
1999 	if (!wp_allowed) {
2000 		/* User requested explicit failure over wp-async capability */
2001 		if (p->arg.flags & PM_SCAN_CHECK_WPASYNC)
2002 			return -EPERM;
2003 		/*
2004 		 * User requires wr-protect, and allows silently skipping
2005 		 * unsupported vmas.
2006 		 */
2007 		if (p->arg.flags & PM_SCAN_WP_MATCHING)
2008 			return 1;
2009 		/*
2010 		 * Then the request doesn't involve wr-protects at all,
2011 		 * fall through to the rest checks, and allow vma walk.
2012 		 */
2013 	}
2014 
2015 	if (vma->vm_flags & VM_PFNMAP)
2016 		return 1;
2017 
2018 	if (wp_allowed)
2019 		vma_category |= PAGE_IS_WPALLOWED;
2020 
2021 	if (vma->vm_flags & VM_SOFTDIRTY)
2022 		vma_category |= PAGE_IS_SOFT_DIRTY;
2023 
2024 	if (!pagemap_scan_is_interesting_vma(vma_category, p))
2025 		return 1;
2026 
2027 	p->cur_vma_category = vma_category;
2028 
2029 	return 0;
2030 }
2031 
2032 static bool pagemap_scan_push_range(unsigned long categories,
2033 				    struct pagemap_scan_private *p,
2034 				    unsigned long addr, unsigned long end)
2035 {
2036 	struct page_region *cur_buf = &p->vec_buf[p->vec_buf_index];
2037 
2038 	/*
2039 	 * When there is no output buffer provided at all, the sentinel values
2040 	 * won't match here. There is no other way for `cur_buf->end` to be
2041 	 * non-zero other than it being non-empty.
2042 	 */
2043 	if (addr == cur_buf->end && categories == cur_buf->categories) {
2044 		cur_buf->end = end;
2045 		return true;
2046 	}
2047 
2048 	if (cur_buf->end) {
2049 		if (p->vec_buf_index >= p->vec_buf_len - 1)
2050 			return false;
2051 
2052 		cur_buf = &p->vec_buf[++p->vec_buf_index];
2053 	}
2054 
2055 	cur_buf->start = addr;
2056 	cur_buf->end = end;
2057 	cur_buf->categories = categories;
2058 
2059 	return true;
2060 }
2061 
2062 static int pagemap_scan_output(unsigned long categories,
2063 			       struct pagemap_scan_private *p,
2064 			       unsigned long addr, unsigned long *end)
2065 {
2066 	unsigned long n_pages, total_pages;
2067 	int ret = 0;
2068 
2069 	if (!p->vec_buf)
2070 		return 0;
2071 
2072 	categories &= p->arg.return_mask;
2073 
2074 	n_pages = (*end - addr) / PAGE_SIZE;
2075 	if (check_add_overflow(p->found_pages, n_pages, &total_pages) ||
2076 	    total_pages > p->arg.max_pages) {
2077 		size_t n_too_much = total_pages - p->arg.max_pages;
2078 		*end -= n_too_much * PAGE_SIZE;
2079 		n_pages -= n_too_much;
2080 		ret = -ENOSPC;
2081 	}
2082 
2083 	if (!pagemap_scan_push_range(categories, p, addr, *end)) {
2084 		*end = addr;
2085 		n_pages = 0;
2086 		ret = -ENOSPC;
2087 	}
2088 
2089 	p->found_pages += n_pages;
2090 	if (ret)
2091 		p->arg.walk_end = *end;
2092 
2093 	return ret;
2094 }
2095 
2096 static int pagemap_scan_thp_entry(pmd_t *pmd, unsigned long start,
2097 				  unsigned long end, struct mm_walk *walk)
2098 {
2099 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2100 	struct pagemap_scan_private *p = walk->private;
2101 	struct vm_area_struct *vma = walk->vma;
2102 	unsigned long categories;
2103 	spinlock_t *ptl;
2104 	int ret = 0;
2105 
2106 	ptl = pmd_trans_huge_lock(pmd, vma);
2107 	if (!ptl)
2108 		return -ENOENT;
2109 
2110 	categories = p->cur_vma_category |
2111 		     pagemap_thp_category(p, vma, start, *pmd);
2112 
2113 	if (!pagemap_scan_is_interesting_page(categories, p))
2114 		goto out_unlock;
2115 
2116 	ret = pagemap_scan_output(categories, p, start, &end);
2117 	if (start == end)
2118 		goto out_unlock;
2119 
2120 	if (~p->arg.flags & PM_SCAN_WP_MATCHING)
2121 		goto out_unlock;
2122 	if (~categories & PAGE_IS_WRITTEN)
2123 		goto out_unlock;
2124 
2125 	/*
2126 	 * Break huge page into small pages if the WP operation
2127 	 * needs to be performed on a portion of the huge page.
2128 	 */
2129 	if (end != start + HPAGE_SIZE) {
2130 		spin_unlock(ptl);
2131 		split_huge_pmd(vma, pmd, start);
2132 		pagemap_scan_backout_range(p, start, end);
2133 		/* Report as if there was no THP */
2134 		return -ENOENT;
2135 	}
2136 
2137 	make_uffd_wp_pmd(vma, start, pmd);
2138 	flush_tlb_range(vma, start, end);
2139 out_unlock:
2140 	spin_unlock(ptl);
2141 	return ret;
2142 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
2143 	return -ENOENT;
2144 #endif
2145 }
2146 
2147 static int pagemap_scan_pmd_entry(pmd_t *pmd, unsigned long start,
2148 				  unsigned long end, struct mm_walk *walk)
2149 {
2150 	struct pagemap_scan_private *p = walk->private;
2151 	struct vm_area_struct *vma = walk->vma;
2152 	unsigned long addr, flush_end = 0;
2153 	pte_t *pte, *start_pte;
2154 	spinlock_t *ptl;
2155 	int ret;
2156 
2157 	arch_enter_lazy_mmu_mode();
2158 
2159 	ret = pagemap_scan_thp_entry(pmd, start, end, walk);
2160 	if (ret != -ENOENT) {
2161 		arch_leave_lazy_mmu_mode();
2162 		return ret;
2163 	}
2164 
2165 	ret = 0;
2166 	start_pte = pte = pte_offset_map_lock(vma->vm_mm, pmd, start, &ptl);
2167 	if (!pte) {
2168 		arch_leave_lazy_mmu_mode();
2169 		walk->action = ACTION_AGAIN;
2170 		return 0;
2171 	}
2172 
2173 	if ((p->arg.flags & PM_SCAN_WP_MATCHING) && !p->vec_out) {
2174 		/* Fast path for performing exclusive WP */
2175 		for (addr = start; addr != end; pte++, addr += PAGE_SIZE) {
2176 			pte_t ptent = ptep_get(pte);
2177 
2178 			if ((pte_present(ptent) && pte_uffd_wp(ptent)) ||
2179 			    pte_swp_uffd_wp_any(ptent))
2180 				continue;
2181 			make_uffd_wp_pte(vma, addr, pte, ptent);
2182 			if (!flush_end)
2183 				start = addr;
2184 			flush_end = addr + PAGE_SIZE;
2185 		}
2186 		goto flush_and_return;
2187 	}
2188 
2189 	if (!p->arg.category_anyof_mask && !p->arg.category_inverted &&
2190 	    p->arg.category_mask == PAGE_IS_WRITTEN &&
2191 	    p->arg.return_mask == PAGE_IS_WRITTEN) {
2192 		for (addr = start; addr < end; pte++, addr += PAGE_SIZE) {
2193 			unsigned long next = addr + PAGE_SIZE;
2194 			pte_t ptent = ptep_get(pte);
2195 
2196 			if ((pte_present(ptent) && pte_uffd_wp(ptent)) ||
2197 			    pte_swp_uffd_wp_any(ptent))
2198 				continue;
2199 			ret = pagemap_scan_output(p->cur_vma_category | PAGE_IS_WRITTEN,
2200 						  p, addr, &next);
2201 			if (next == addr)
2202 				break;
2203 			if (~p->arg.flags & PM_SCAN_WP_MATCHING)
2204 				continue;
2205 			make_uffd_wp_pte(vma, addr, pte, ptent);
2206 			if (!flush_end)
2207 				start = addr;
2208 			flush_end = next;
2209 		}
2210 		goto flush_and_return;
2211 	}
2212 
2213 	for (addr = start; addr != end; pte++, addr += PAGE_SIZE) {
2214 		pte_t ptent = ptep_get(pte);
2215 		unsigned long categories = p->cur_vma_category |
2216 					   pagemap_page_category(p, vma, addr, ptent);
2217 		unsigned long next = addr + PAGE_SIZE;
2218 
2219 		if (!pagemap_scan_is_interesting_page(categories, p))
2220 			continue;
2221 
2222 		ret = pagemap_scan_output(categories, p, addr, &next);
2223 		if (next == addr)
2224 			break;
2225 
2226 		if (~p->arg.flags & PM_SCAN_WP_MATCHING)
2227 			continue;
2228 		if (~categories & PAGE_IS_WRITTEN)
2229 			continue;
2230 
2231 		make_uffd_wp_pte(vma, addr, pte, ptent);
2232 		if (!flush_end)
2233 			start = addr;
2234 		flush_end = next;
2235 	}
2236 
2237 flush_and_return:
2238 	if (flush_end)
2239 		flush_tlb_range(vma, start, addr);
2240 
2241 	pte_unmap_unlock(start_pte, ptl);
2242 	arch_leave_lazy_mmu_mode();
2243 
2244 	cond_resched();
2245 	return ret;
2246 }
2247 
2248 #ifdef CONFIG_HUGETLB_PAGE
2249 static int pagemap_scan_hugetlb_entry(pte_t *ptep, unsigned long hmask,
2250 				      unsigned long start, unsigned long end,
2251 				      struct mm_walk *walk)
2252 {
2253 	struct pagemap_scan_private *p = walk->private;
2254 	struct vm_area_struct *vma = walk->vma;
2255 	unsigned long categories;
2256 	spinlock_t *ptl;
2257 	int ret = 0;
2258 	pte_t pte;
2259 
2260 	if (~p->arg.flags & PM_SCAN_WP_MATCHING) {
2261 		/* Go the short route when not write-protecting pages. */
2262 
2263 		pte = huge_ptep_get(ptep);
2264 		categories = p->cur_vma_category | pagemap_hugetlb_category(pte);
2265 
2266 		if (!pagemap_scan_is_interesting_page(categories, p))
2267 			return 0;
2268 
2269 		return pagemap_scan_output(categories, p, start, &end);
2270 	}
2271 
2272 	i_mmap_lock_write(vma->vm_file->f_mapping);
2273 	ptl = huge_pte_lock(hstate_vma(vma), vma->vm_mm, ptep);
2274 
2275 	pte = huge_ptep_get(ptep);
2276 	categories = p->cur_vma_category | pagemap_hugetlb_category(pte);
2277 
2278 	if (!pagemap_scan_is_interesting_page(categories, p))
2279 		goto out_unlock;
2280 
2281 	ret = pagemap_scan_output(categories, p, start, &end);
2282 	if (start == end)
2283 		goto out_unlock;
2284 
2285 	if (~categories & PAGE_IS_WRITTEN)
2286 		goto out_unlock;
2287 
2288 	if (end != start + HPAGE_SIZE) {
2289 		/* Partial HugeTLB page WP isn't possible. */
2290 		pagemap_scan_backout_range(p, start, end);
2291 		p->arg.walk_end = start;
2292 		ret = 0;
2293 		goto out_unlock;
2294 	}
2295 
2296 	make_uffd_wp_huge_pte(vma, start, ptep, pte);
2297 	flush_hugetlb_tlb_range(vma, start, end);
2298 
2299 out_unlock:
2300 	spin_unlock(ptl);
2301 	i_mmap_unlock_write(vma->vm_file->f_mapping);
2302 
2303 	return ret;
2304 }
2305 #else
2306 #define pagemap_scan_hugetlb_entry NULL
2307 #endif
2308 
2309 static int pagemap_scan_pte_hole(unsigned long addr, unsigned long end,
2310 				 int depth, struct mm_walk *walk)
2311 {
2312 	struct pagemap_scan_private *p = walk->private;
2313 	struct vm_area_struct *vma = walk->vma;
2314 	int ret, err;
2315 
2316 	if (!vma || !pagemap_scan_is_interesting_page(p->cur_vma_category, p))
2317 		return 0;
2318 
2319 	ret = pagemap_scan_output(p->cur_vma_category, p, addr, &end);
2320 	if (addr == end)
2321 		return ret;
2322 
2323 	if (~p->arg.flags & PM_SCAN_WP_MATCHING)
2324 		return ret;
2325 
2326 	err = uffd_wp_range(vma, addr, end - addr, true);
2327 	if (err < 0)
2328 		ret = err;
2329 
2330 	return ret;
2331 }
2332 
2333 static const struct mm_walk_ops pagemap_scan_ops = {
2334 	.test_walk = pagemap_scan_test_walk,
2335 	.pmd_entry = pagemap_scan_pmd_entry,
2336 	.pte_hole = pagemap_scan_pte_hole,
2337 	.hugetlb_entry = pagemap_scan_hugetlb_entry,
2338 };
2339 
2340 static int pagemap_scan_get_args(struct pm_scan_arg *arg,
2341 				 unsigned long uarg)
2342 {
2343 	if (copy_from_user(arg, (void __user *)uarg, sizeof(*arg)))
2344 		return -EFAULT;
2345 
2346 	if (arg->size != sizeof(struct pm_scan_arg))
2347 		return -EINVAL;
2348 
2349 	/* Validate requested features */
2350 	if (arg->flags & ~PM_SCAN_FLAGS)
2351 		return -EINVAL;
2352 	if ((arg->category_inverted | arg->category_mask |
2353 	     arg->category_anyof_mask | arg->return_mask) & ~PM_SCAN_CATEGORIES)
2354 		return -EINVAL;
2355 
2356 	arg->start = untagged_addr((unsigned long)arg->start);
2357 	arg->end = untagged_addr((unsigned long)arg->end);
2358 	arg->vec = untagged_addr((unsigned long)arg->vec);
2359 
2360 	/* Validate memory pointers */
2361 	if (!IS_ALIGNED(arg->start, PAGE_SIZE))
2362 		return -EINVAL;
2363 	if (!access_ok((void __user *)(long)arg->start, arg->end - arg->start))
2364 		return -EFAULT;
2365 	if (!arg->vec && arg->vec_len)
2366 		return -EINVAL;
2367 	if (arg->vec && !access_ok((void __user *)(long)arg->vec,
2368 			      arg->vec_len * sizeof(struct page_region)))
2369 		return -EFAULT;
2370 
2371 	/* Fixup default values */
2372 	arg->end = ALIGN(arg->end, PAGE_SIZE);
2373 	arg->walk_end = 0;
2374 	if (!arg->max_pages)
2375 		arg->max_pages = ULONG_MAX;
2376 
2377 	return 0;
2378 }
2379 
2380 static int pagemap_scan_writeback_args(struct pm_scan_arg *arg,
2381 				       unsigned long uargl)
2382 {
2383 	struct pm_scan_arg __user *uarg	= (void __user *)uargl;
2384 
2385 	if (copy_to_user(&uarg->walk_end, &arg->walk_end, sizeof(arg->walk_end)))
2386 		return -EFAULT;
2387 
2388 	return 0;
2389 }
2390 
2391 static int pagemap_scan_init_bounce_buffer(struct pagemap_scan_private *p)
2392 {
2393 	if (!p->arg.vec_len)
2394 		return 0;
2395 
2396 	p->vec_buf_len = min_t(size_t, PAGEMAP_WALK_SIZE >> PAGE_SHIFT,
2397 			       p->arg.vec_len);
2398 	p->vec_buf = kmalloc_array(p->vec_buf_len, sizeof(*p->vec_buf),
2399 				   GFP_KERNEL);
2400 	if (!p->vec_buf)
2401 		return -ENOMEM;
2402 
2403 	p->vec_buf->start = p->vec_buf->end = 0;
2404 	p->vec_out = (struct page_region __user *)(long)p->arg.vec;
2405 
2406 	return 0;
2407 }
2408 
2409 static long pagemap_scan_flush_buffer(struct pagemap_scan_private *p)
2410 {
2411 	const struct page_region *buf = p->vec_buf;
2412 	long n = p->vec_buf_index;
2413 
2414 	if (!p->vec_buf)
2415 		return 0;
2416 
2417 	if (buf[n].end != buf[n].start)
2418 		n++;
2419 
2420 	if (!n)
2421 		return 0;
2422 
2423 	if (copy_to_user(p->vec_out, buf, n * sizeof(*buf)))
2424 		return -EFAULT;
2425 
2426 	p->arg.vec_len -= n;
2427 	p->vec_out += n;
2428 
2429 	p->vec_buf_index = 0;
2430 	p->vec_buf_len = min_t(size_t, p->vec_buf_len, p->arg.vec_len);
2431 	p->vec_buf->start = p->vec_buf->end = 0;
2432 
2433 	return n;
2434 }
2435 
2436 static long do_pagemap_scan(struct mm_struct *mm, unsigned long uarg)
2437 {
2438 	struct pagemap_scan_private p = {0};
2439 	unsigned long walk_start;
2440 	size_t n_ranges_out = 0;
2441 	int ret;
2442 
2443 	ret = pagemap_scan_get_args(&p.arg, uarg);
2444 	if (ret)
2445 		return ret;
2446 
2447 	p.masks_of_interest = p.arg.category_mask | p.arg.category_anyof_mask |
2448 			      p.arg.return_mask;
2449 	ret = pagemap_scan_init_bounce_buffer(&p);
2450 	if (ret)
2451 		return ret;
2452 
2453 	for (walk_start = p.arg.start; walk_start < p.arg.end;
2454 			walk_start = p.arg.walk_end) {
2455 		struct mmu_notifier_range range;
2456 		long n_out;
2457 
2458 		if (fatal_signal_pending(current)) {
2459 			ret = -EINTR;
2460 			break;
2461 		}
2462 
2463 		ret = mmap_read_lock_killable(mm);
2464 		if (ret)
2465 			break;
2466 
2467 		/* Protection change for the range is going to happen. */
2468 		if (p.arg.flags & PM_SCAN_WP_MATCHING) {
2469 			mmu_notifier_range_init(&range, MMU_NOTIFY_PROTECTION_VMA, 0,
2470 						mm, walk_start, p.arg.end);
2471 			mmu_notifier_invalidate_range_start(&range);
2472 		}
2473 
2474 		ret = walk_page_range(mm, walk_start, p.arg.end,
2475 				      &pagemap_scan_ops, &p);
2476 
2477 		if (p.arg.flags & PM_SCAN_WP_MATCHING)
2478 			mmu_notifier_invalidate_range_end(&range);
2479 
2480 		mmap_read_unlock(mm);
2481 
2482 		n_out = pagemap_scan_flush_buffer(&p);
2483 		if (n_out < 0)
2484 			ret = n_out;
2485 		else
2486 			n_ranges_out += n_out;
2487 
2488 		if (ret != -ENOSPC)
2489 			break;
2490 
2491 		if (p.arg.vec_len == 0 || p.found_pages == p.arg.max_pages)
2492 			break;
2493 	}
2494 
2495 	/* ENOSPC signifies early stop (buffer full) from the walk. */
2496 	if (!ret || ret == -ENOSPC)
2497 		ret = n_ranges_out;
2498 
2499 	/* The walk_end isn't set when ret is zero */
2500 	if (!p.arg.walk_end)
2501 		p.arg.walk_end = p.arg.end;
2502 	if (pagemap_scan_writeback_args(&p.arg, uarg))
2503 		ret = -EFAULT;
2504 
2505 	kfree(p.vec_buf);
2506 	return ret;
2507 }
2508 
2509 static long do_pagemap_cmd(struct file *file, unsigned int cmd,
2510 			   unsigned long arg)
2511 {
2512 	struct mm_struct *mm = file->private_data;
2513 
2514 	switch (cmd) {
2515 	case PAGEMAP_SCAN:
2516 		return do_pagemap_scan(mm, arg);
2517 
2518 	default:
2519 		return -EINVAL;
2520 	}
2521 }
2522 
2523 const struct file_operations proc_pagemap_operations = {
2524 	.llseek		= mem_lseek, /* borrow this */
2525 	.read		= pagemap_read,
2526 	.open		= pagemap_open,
2527 	.release	= pagemap_release,
2528 	.unlocked_ioctl = do_pagemap_cmd,
2529 	.compat_ioctl	= do_pagemap_cmd,
2530 };
2531 #endif /* CONFIG_PROC_PAGE_MONITOR */
2532 
2533 #ifdef CONFIG_NUMA
2534 
2535 struct numa_maps {
2536 	unsigned long pages;
2537 	unsigned long anon;
2538 	unsigned long active;
2539 	unsigned long writeback;
2540 	unsigned long mapcount_max;
2541 	unsigned long dirty;
2542 	unsigned long swapcache;
2543 	unsigned long node[MAX_NUMNODES];
2544 };
2545 
2546 struct numa_maps_private {
2547 	struct proc_maps_private proc_maps;
2548 	struct numa_maps md;
2549 };
2550 
2551 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
2552 			unsigned long nr_pages)
2553 {
2554 	int count = page_mapcount(page);
2555 
2556 	md->pages += nr_pages;
2557 	if (pte_dirty || PageDirty(page))
2558 		md->dirty += nr_pages;
2559 
2560 	if (PageSwapCache(page))
2561 		md->swapcache += nr_pages;
2562 
2563 	if (PageActive(page) || PageUnevictable(page))
2564 		md->active += nr_pages;
2565 
2566 	if (PageWriteback(page))
2567 		md->writeback += nr_pages;
2568 
2569 	if (PageAnon(page))
2570 		md->anon += nr_pages;
2571 
2572 	if (count > md->mapcount_max)
2573 		md->mapcount_max = count;
2574 
2575 	md->node[page_to_nid(page)] += nr_pages;
2576 }
2577 
2578 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
2579 		unsigned long addr)
2580 {
2581 	struct page *page;
2582 	int nid;
2583 
2584 	if (!pte_present(pte))
2585 		return NULL;
2586 
2587 	page = vm_normal_page(vma, addr, pte);
2588 	if (!page || is_zone_device_page(page))
2589 		return NULL;
2590 
2591 	if (PageReserved(page))
2592 		return NULL;
2593 
2594 	nid = page_to_nid(page);
2595 	if (!node_isset(nid, node_states[N_MEMORY]))
2596 		return NULL;
2597 
2598 	return page;
2599 }
2600 
2601 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2602 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
2603 					      struct vm_area_struct *vma,
2604 					      unsigned long addr)
2605 {
2606 	struct page *page;
2607 	int nid;
2608 
2609 	if (!pmd_present(pmd))
2610 		return NULL;
2611 
2612 	page = vm_normal_page_pmd(vma, addr, pmd);
2613 	if (!page)
2614 		return NULL;
2615 
2616 	if (PageReserved(page))
2617 		return NULL;
2618 
2619 	nid = page_to_nid(page);
2620 	if (!node_isset(nid, node_states[N_MEMORY]))
2621 		return NULL;
2622 
2623 	return page;
2624 }
2625 #endif
2626 
2627 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
2628 		unsigned long end, struct mm_walk *walk)
2629 {
2630 	struct numa_maps *md = walk->private;
2631 	struct vm_area_struct *vma = walk->vma;
2632 	spinlock_t *ptl;
2633 	pte_t *orig_pte;
2634 	pte_t *pte;
2635 
2636 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2637 	ptl = pmd_trans_huge_lock(pmd, vma);
2638 	if (ptl) {
2639 		struct page *page;
2640 
2641 		page = can_gather_numa_stats_pmd(*pmd, vma, addr);
2642 		if (page)
2643 			gather_stats(page, md, pmd_dirty(*pmd),
2644 				     HPAGE_PMD_SIZE/PAGE_SIZE);
2645 		spin_unlock(ptl);
2646 		return 0;
2647 	}
2648 #endif
2649 	orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
2650 	if (!pte) {
2651 		walk->action = ACTION_AGAIN;
2652 		return 0;
2653 	}
2654 	do {
2655 		pte_t ptent = ptep_get(pte);
2656 		struct page *page = can_gather_numa_stats(ptent, vma, addr);
2657 		if (!page)
2658 			continue;
2659 		gather_stats(page, md, pte_dirty(ptent), 1);
2660 
2661 	} while (pte++, addr += PAGE_SIZE, addr != end);
2662 	pte_unmap_unlock(orig_pte, ptl);
2663 	cond_resched();
2664 	return 0;
2665 }
2666 #ifdef CONFIG_HUGETLB_PAGE
2667 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
2668 		unsigned long addr, unsigned long end, struct mm_walk *walk)
2669 {
2670 	pte_t huge_pte = huge_ptep_get(pte);
2671 	struct numa_maps *md;
2672 	struct page *page;
2673 
2674 	if (!pte_present(huge_pte))
2675 		return 0;
2676 
2677 	page = pte_page(huge_pte);
2678 
2679 	md = walk->private;
2680 	gather_stats(page, md, pte_dirty(huge_pte), 1);
2681 	return 0;
2682 }
2683 
2684 #else
2685 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
2686 		unsigned long addr, unsigned long end, struct mm_walk *walk)
2687 {
2688 	return 0;
2689 }
2690 #endif
2691 
2692 static const struct mm_walk_ops show_numa_ops = {
2693 	.hugetlb_entry = gather_hugetlb_stats,
2694 	.pmd_entry = gather_pte_stats,
2695 	.walk_lock = PGWALK_RDLOCK,
2696 };
2697 
2698 /*
2699  * Display pages allocated per node and memory policy via /proc.
2700  */
2701 static int show_numa_map(struct seq_file *m, void *v)
2702 {
2703 	struct numa_maps_private *numa_priv = m->private;
2704 	struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
2705 	struct vm_area_struct *vma = v;
2706 	struct numa_maps *md = &numa_priv->md;
2707 	struct file *file = vma->vm_file;
2708 	struct mm_struct *mm = vma->vm_mm;
2709 	char buffer[64];
2710 	struct mempolicy *pol;
2711 	pgoff_t ilx;
2712 	int nid;
2713 
2714 	if (!mm)
2715 		return 0;
2716 
2717 	/* Ensure we start with an empty set of numa_maps statistics. */
2718 	memset(md, 0, sizeof(*md));
2719 
2720 	pol = __get_vma_policy(vma, vma->vm_start, &ilx);
2721 	if (pol) {
2722 		mpol_to_str(buffer, sizeof(buffer), pol);
2723 		mpol_cond_put(pol);
2724 	} else {
2725 		mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
2726 	}
2727 
2728 	seq_printf(m, "%08lx %s", vma->vm_start, buffer);
2729 
2730 	if (file) {
2731 		seq_puts(m, " file=");
2732 		seq_path(m, file_user_path(file), "\n\t= ");
2733 	} else if (vma_is_initial_heap(vma)) {
2734 		seq_puts(m, " heap");
2735 	} else if (vma_is_initial_stack(vma)) {
2736 		seq_puts(m, " stack");
2737 	}
2738 
2739 	if (is_vm_hugetlb_page(vma))
2740 		seq_puts(m, " huge");
2741 
2742 	/* mmap_lock is held by m_start */
2743 	walk_page_vma(vma, &show_numa_ops, md);
2744 
2745 	if (!md->pages)
2746 		goto out;
2747 
2748 	if (md->anon)
2749 		seq_printf(m, " anon=%lu", md->anon);
2750 
2751 	if (md->dirty)
2752 		seq_printf(m, " dirty=%lu", md->dirty);
2753 
2754 	if (md->pages != md->anon && md->pages != md->dirty)
2755 		seq_printf(m, " mapped=%lu", md->pages);
2756 
2757 	if (md->mapcount_max > 1)
2758 		seq_printf(m, " mapmax=%lu", md->mapcount_max);
2759 
2760 	if (md->swapcache)
2761 		seq_printf(m, " swapcache=%lu", md->swapcache);
2762 
2763 	if (md->active < md->pages && !is_vm_hugetlb_page(vma))
2764 		seq_printf(m, " active=%lu", md->active);
2765 
2766 	if (md->writeback)
2767 		seq_printf(m, " writeback=%lu", md->writeback);
2768 
2769 	for_each_node_state(nid, N_MEMORY)
2770 		if (md->node[nid])
2771 			seq_printf(m, " N%d=%lu", nid, md->node[nid]);
2772 
2773 	seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
2774 out:
2775 	seq_putc(m, '\n');
2776 	return 0;
2777 }
2778 
2779 static const struct seq_operations proc_pid_numa_maps_op = {
2780 	.start  = m_start,
2781 	.next   = m_next,
2782 	.stop   = m_stop,
2783 	.show   = show_numa_map,
2784 };
2785 
2786 static int pid_numa_maps_open(struct inode *inode, struct file *file)
2787 {
2788 	return proc_maps_open(inode, file, &proc_pid_numa_maps_op,
2789 				sizeof(struct numa_maps_private));
2790 }
2791 
2792 const struct file_operations proc_pid_numa_maps_operations = {
2793 	.open		= pid_numa_maps_open,
2794 	.read		= seq_read,
2795 	.llseek		= seq_lseek,
2796 	.release	= proc_map_release,
2797 };
2798 
2799 #endif /* CONFIG_NUMA */
2800