xref: /openbmc/linux/fs/proc/task_mmu.c (revision a9a08845)
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/mm.h>
3 #include <linux/vmacache.h>
4 #include <linux/hugetlb.h>
5 #include <linux/huge_mm.h>
6 #include <linux/mount.h>
7 #include <linux/seq_file.h>
8 #include <linux/highmem.h>
9 #include <linux/ptrace.h>
10 #include <linux/slab.h>
11 #include <linux/pagemap.h>
12 #include <linux/mempolicy.h>
13 #include <linux/rmap.h>
14 #include <linux/swap.h>
15 #include <linux/sched/mm.h>
16 #include <linux/swapops.h>
17 #include <linux/mmu_notifier.h>
18 #include <linux/page_idle.h>
19 #include <linux/shmem_fs.h>
20 #include <linux/uaccess.h>
21 
22 #include <asm/elf.h>
23 #include <asm/tlb.h>
24 #include <asm/tlbflush.h>
25 #include "internal.h"
26 
27 void task_mem(struct seq_file *m, struct mm_struct *mm)
28 {
29 	unsigned long text, lib, swap, anon, file, shmem;
30 	unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
31 
32 	anon = get_mm_counter(mm, MM_ANONPAGES);
33 	file = get_mm_counter(mm, MM_FILEPAGES);
34 	shmem = get_mm_counter(mm, MM_SHMEMPAGES);
35 
36 	/*
37 	 * Note: to minimize their overhead, mm maintains hiwater_vm and
38 	 * hiwater_rss only when about to *lower* total_vm or rss.  Any
39 	 * collector of these hiwater stats must therefore get total_vm
40 	 * and rss too, which will usually be the higher.  Barriers? not
41 	 * worth the effort, such snapshots can always be inconsistent.
42 	 */
43 	hiwater_vm = total_vm = mm->total_vm;
44 	if (hiwater_vm < mm->hiwater_vm)
45 		hiwater_vm = mm->hiwater_vm;
46 	hiwater_rss = total_rss = anon + file + shmem;
47 	if (hiwater_rss < mm->hiwater_rss)
48 		hiwater_rss = mm->hiwater_rss;
49 
50 	/* split executable areas between text and lib */
51 	text = PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK);
52 	text = min(text, mm->exec_vm << PAGE_SHIFT);
53 	lib = (mm->exec_vm << PAGE_SHIFT) - text;
54 
55 	swap = get_mm_counter(mm, MM_SWAPENTS);
56 	seq_printf(m,
57 		"VmPeak:\t%8lu kB\n"
58 		"VmSize:\t%8lu kB\n"
59 		"VmLck:\t%8lu kB\n"
60 		"VmPin:\t%8lu kB\n"
61 		"VmHWM:\t%8lu kB\n"
62 		"VmRSS:\t%8lu kB\n"
63 		"RssAnon:\t%8lu kB\n"
64 		"RssFile:\t%8lu kB\n"
65 		"RssShmem:\t%8lu kB\n"
66 		"VmData:\t%8lu kB\n"
67 		"VmStk:\t%8lu kB\n"
68 		"VmExe:\t%8lu kB\n"
69 		"VmLib:\t%8lu kB\n"
70 		"VmPTE:\t%8lu kB\n"
71 		"VmSwap:\t%8lu kB\n",
72 		hiwater_vm << (PAGE_SHIFT-10),
73 		total_vm << (PAGE_SHIFT-10),
74 		mm->locked_vm << (PAGE_SHIFT-10),
75 		mm->pinned_vm << (PAGE_SHIFT-10),
76 		hiwater_rss << (PAGE_SHIFT-10),
77 		total_rss << (PAGE_SHIFT-10),
78 		anon << (PAGE_SHIFT-10),
79 		file << (PAGE_SHIFT-10),
80 		shmem << (PAGE_SHIFT-10),
81 		mm->data_vm << (PAGE_SHIFT-10),
82 		mm->stack_vm << (PAGE_SHIFT-10),
83 		text >> 10,
84 		lib >> 10,
85 		mm_pgtables_bytes(mm) >> 10,
86 		swap << (PAGE_SHIFT-10));
87 	hugetlb_report_usage(m, mm);
88 }
89 
90 unsigned long task_vsize(struct mm_struct *mm)
91 {
92 	return PAGE_SIZE * mm->total_vm;
93 }
94 
95 unsigned long task_statm(struct mm_struct *mm,
96 			 unsigned long *shared, unsigned long *text,
97 			 unsigned long *data, unsigned long *resident)
98 {
99 	*shared = get_mm_counter(mm, MM_FILEPAGES) +
100 			get_mm_counter(mm, MM_SHMEMPAGES);
101 	*text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
102 								>> PAGE_SHIFT;
103 	*data = mm->data_vm + mm->stack_vm;
104 	*resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
105 	return mm->total_vm;
106 }
107 
108 #ifdef CONFIG_NUMA
109 /*
110  * Save get_task_policy() for show_numa_map().
111  */
112 static void hold_task_mempolicy(struct proc_maps_private *priv)
113 {
114 	struct task_struct *task = priv->task;
115 
116 	task_lock(task);
117 	priv->task_mempolicy = get_task_policy(task);
118 	mpol_get(priv->task_mempolicy);
119 	task_unlock(task);
120 }
121 static void release_task_mempolicy(struct proc_maps_private *priv)
122 {
123 	mpol_put(priv->task_mempolicy);
124 }
125 #else
126 static void hold_task_mempolicy(struct proc_maps_private *priv)
127 {
128 }
129 static void release_task_mempolicy(struct proc_maps_private *priv)
130 {
131 }
132 #endif
133 
134 static void vma_stop(struct proc_maps_private *priv)
135 {
136 	struct mm_struct *mm = priv->mm;
137 
138 	release_task_mempolicy(priv);
139 	up_read(&mm->mmap_sem);
140 	mmput(mm);
141 }
142 
143 static struct vm_area_struct *
144 m_next_vma(struct proc_maps_private *priv, struct vm_area_struct *vma)
145 {
146 	if (vma == priv->tail_vma)
147 		return NULL;
148 	return vma->vm_next ?: priv->tail_vma;
149 }
150 
151 static void m_cache_vma(struct seq_file *m, struct vm_area_struct *vma)
152 {
153 	if (m->count < m->size)	/* vma is copied successfully */
154 		m->version = m_next_vma(m->private, vma) ? vma->vm_end : -1UL;
155 }
156 
157 static void *m_start(struct seq_file *m, loff_t *ppos)
158 {
159 	struct proc_maps_private *priv = m->private;
160 	unsigned long last_addr = m->version;
161 	struct mm_struct *mm;
162 	struct vm_area_struct *vma;
163 	unsigned int pos = *ppos;
164 
165 	/* See m_cache_vma(). Zero at the start or after lseek. */
166 	if (last_addr == -1UL)
167 		return NULL;
168 
169 	priv->task = get_proc_task(priv->inode);
170 	if (!priv->task)
171 		return ERR_PTR(-ESRCH);
172 
173 	mm = priv->mm;
174 	if (!mm || !mmget_not_zero(mm))
175 		return NULL;
176 
177 	down_read(&mm->mmap_sem);
178 	hold_task_mempolicy(priv);
179 	priv->tail_vma = get_gate_vma(mm);
180 
181 	if (last_addr) {
182 		vma = find_vma(mm, last_addr - 1);
183 		if (vma && vma->vm_start <= last_addr)
184 			vma = m_next_vma(priv, vma);
185 		if (vma)
186 			return vma;
187 	}
188 
189 	m->version = 0;
190 	if (pos < mm->map_count) {
191 		for (vma = mm->mmap; pos; pos--) {
192 			m->version = vma->vm_start;
193 			vma = vma->vm_next;
194 		}
195 		return vma;
196 	}
197 
198 	/* we do not bother to update m->version in this case */
199 	if (pos == mm->map_count && priv->tail_vma)
200 		return priv->tail_vma;
201 
202 	vma_stop(priv);
203 	return NULL;
204 }
205 
206 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
207 {
208 	struct proc_maps_private *priv = m->private;
209 	struct vm_area_struct *next;
210 
211 	(*pos)++;
212 	next = m_next_vma(priv, v);
213 	if (!next)
214 		vma_stop(priv);
215 	return next;
216 }
217 
218 static void m_stop(struct seq_file *m, void *v)
219 {
220 	struct proc_maps_private *priv = m->private;
221 
222 	if (!IS_ERR_OR_NULL(v))
223 		vma_stop(priv);
224 	if (priv->task) {
225 		put_task_struct(priv->task);
226 		priv->task = NULL;
227 	}
228 }
229 
230 static int proc_maps_open(struct inode *inode, struct file *file,
231 			const struct seq_operations *ops, int psize)
232 {
233 	struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
234 
235 	if (!priv)
236 		return -ENOMEM;
237 
238 	priv->inode = inode;
239 	priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
240 	if (IS_ERR(priv->mm)) {
241 		int err = PTR_ERR(priv->mm);
242 
243 		seq_release_private(inode, file);
244 		return err;
245 	}
246 
247 	return 0;
248 }
249 
250 static int proc_map_release(struct inode *inode, struct file *file)
251 {
252 	struct seq_file *seq = file->private_data;
253 	struct proc_maps_private *priv = seq->private;
254 
255 	if (priv->mm)
256 		mmdrop(priv->mm);
257 
258 	kfree(priv->rollup);
259 	return seq_release_private(inode, file);
260 }
261 
262 static int do_maps_open(struct inode *inode, struct file *file,
263 			const struct seq_operations *ops)
264 {
265 	return proc_maps_open(inode, file, ops,
266 				sizeof(struct proc_maps_private));
267 }
268 
269 /*
270  * Indicate if the VMA is a stack for the given task; for
271  * /proc/PID/maps that is the stack of the main task.
272  */
273 static int is_stack(struct vm_area_struct *vma)
274 {
275 	/*
276 	 * We make no effort to guess what a given thread considers to be
277 	 * its "stack".  It's not even well-defined for programs written
278 	 * languages like Go.
279 	 */
280 	return vma->vm_start <= vma->vm_mm->start_stack &&
281 		vma->vm_end >= vma->vm_mm->start_stack;
282 }
283 
284 static void show_vma_header_prefix(struct seq_file *m,
285 				   unsigned long start, unsigned long end,
286 				   vm_flags_t flags, unsigned long long pgoff,
287 				   dev_t dev, unsigned long ino)
288 {
289 	seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
290 	seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu ",
291 		   start,
292 		   end,
293 		   flags & VM_READ ? 'r' : '-',
294 		   flags & VM_WRITE ? 'w' : '-',
295 		   flags & VM_EXEC ? 'x' : '-',
296 		   flags & VM_MAYSHARE ? 's' : 'p',
297 		   pgoff,
298 		   MAJOR(dev), MINOR(dev), ino);
299 }
300 
301 static void
302 show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
303 {
304 	struct mm_struct *mm = vma->vm_mm;
305 	struct file *file = vma->vm_file;
306 	vm_flags_t flags = vma->vm_flags;
307 	unsigned long ino = 0;
308 	unsigned long long pgoff = 0;
309 	unsigned long start, end;
310 	dev_t dev = 0;
311 	const char *name = NULL;
312 
313 	if (file) {
314 		struct inode *inode = file_inode(vma->vm_file);
315 		dev = inode->i_sb->s_dev;
316 		ino = inode->i_ino;
317 		pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
318 	}
319 
320 	start = vma->vm_start;
321 	end = vma->vm_end;
322 	show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino);
323 
324 	/*
325 	 * Print the dentry name for named mappings, and a
326 	 * special [heap] marker for the heap:
327 	 */
328 	if (file) {
329 		seq_pad(m, ' ');
330 		seq_file_path(m, file, "\n");
331 		goto done;
332 	}
333 
334 	if (vma->vm_ops && vma->vm_ops->name) {
335 		name = vma->vm_ops->name(vma);
336 		if (name)
337 			goto done;
338 	}
339 
340 	name = arch_vma_name(vma);
341 	if (!name) {
342 		if (!mm) {
343 			name = "[vdso]";
344 			goto done;
345 		}
346 
347 		if (vma->vm_start <= mm->brk &&
348 		    vma->vm_end >= mm->start_brk) {
349 			name = "[heap]";
350 			goto done;
351 		}
352 
353 		if (is_stack(vma))
354 			name = "[stack]";
355 	}
356 
357 done:
358 	if (name) {
359 		seq_pad(m, ' ');
360 		seq_puts(m, name);
361 	}
362 	seq_putc(m, '\n');
363 }
364 
365 static int show_map(struct seq_file *m, void *v, int is_pid)
366 {
367 	show_map_vma(m, v, is_pid);
368 	m_cache_vma(m, v);
369 	return 0;
370 }
371 
372 static int show_pid_map(struct seq_file *m, void *v)
373 {
374 	return show_map(m, v, 1);
375 }
376 
377 static int show_tid_map(struct seq_file *m, void *v)
378 {
379 	return show_map(m, v, 0);
380 }
381 
382 static const struct seq_operations proc_pid_maps_op = {
383 	.start	= m_start,
384 	.next	= m_next,
385 	.stop	= m_stop,
386 	.show	= show_pid_map
387 };
388 
389 static const struct seq_operations proc_tid_maps_op = {
390 	.start	= m_start,
391 	.next	= m_next,
392 	.stop	= m_stop,
393 	.show	= show_tid_map
394 };
395 
396 static int pid_maps_open(struct inode *inode, struct file *file)
397 {
398 	return do_maps_open(inode, file, &proc_pid_maps_op);
399 }
400 
401 static int tid_maps_open(struct inode *inode, struct file *file)
402 {
403 	return do_maps_open(inode, file, &proc_tid_maps_op);
404 }
405 
406 const struct file_operations proc_pid_maps_operations = {
407 	.open		= pid_maps_open,
408 	.read		= seq_read,
409 	.llseek		= seq_lseek,
410 	.release	= proc_map_release,
411 };
412 
413 const struct file_operations proc_tid_maps_operations = {
414 	.open		= tid_maps_open,
415 	.read		= seq_read,
416 	.llseek		= seq_lseek,
417 	.release	= proc_map_release,
418 };
419 
420 /*
421  * Proportional Set Size(PSS): my share of RSS.
422  *
423  * PSS of a process is the count of pages it has in memory, where each
424  * page is divided by the number of processes sharing it.  So if a
425  * process has 1000 pages all to itself, and 1000 shared with one other
426  * process, its PSS will be 1500.
427  *
428  * To keep (accumulated) division errors low, we adopt a 64bit
429  * fixed-point pss counter to minimize division errors. So (pss >>
430  * PSS_SHIFT) would be the real byte count.
431  *
432  * A shift of 12 before division means (assuming 4K page size):
433  * 	- 1M 3-user-pages add up to 8KB errors;
434  * 	- supports mapcount up to 2^24, or 16M;
435  * 	- supports PSS up to 2^52 bytes, or 4PB.
436  */
437 #define PSS_SHIFT 12
438 
439 #ifdef CONFIG_PROC_PAGE_MONITOR
440 struct mem_size_stats {
441 	bool first;
442 	unsigned long resident;
443 	unsigned long shared_clean;
444 	unsigned long shared_dirty;
445 	unsigned long private_clean;
446 	unsigned long private_dirty;
447 	unsigned long referenced;
448 	unsigned long anonymous;
449 	unsigned long lazyfree;
450 	unsigned long anonymous_thp;
451 	unsigned long shmem_thp;
452 	unsigned long swap;
453 	unsigned long shared_hugetlb;
454 	unsigned long private_hugetlb;
455 	unsigned long first_vma_start;
456 	u64 pss;
457 	u64 pss_locked;
458 	u64 swap_pss;
459 	bool check_shmem_swap;
460 };
461 
462 static void smaps_account(struct mem_size_stats *mss, struct page *page,
463 		bool compound, bool young, bool dirty)
464 {
465 	int i, nr = compound ? 1 << compound_order(page) : 1;
466 	unsigned long size = nr * PAGE_SIZE;
467 
468 	if (PageAnon(page)) {
469 		mss->anonymous += size;
470 		if (!PageSwapBacked(page) && !dirty && !PageDirty(page))
471 			mss->lazyfree += size;
472 	}
473 
474 	mss->resident += size;
475 	/* Accumulate the size in pages that have been accessed. */
476 	if (young || page_is_young(page) || PageReferenced(page))
477 		mss->referenced += size;
478 
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 	if (page_count(page) == 1) {
485 		if (dirty || PageDirty(page))
486 			mss->private_dirty += size;
487 		else
488 			mss->private_clean += size;
489 		mss->pss += (u64)size << PSS_SHIFT;
490 		return;
491 	}
492 
493 	for (i = 0; i < nr; i++, page++) {
494 		int mapcount = page_mapcount(page);
495 
496 		if (mapcount >= 2) {
497 			if (dirty || PageDirty(page))
498 				mss->shared_dirty += PAGE_SIZE;
499 			else
500 				mss->shared_clean += PAGE_SIZE;
501 			mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount;
502 		} else {
503 			if (dirty || PageDirty(page))
504 				mss->private_dirty += PAGE_SIZE;
505 			else
506 				mss->private_clean += PAGE_SIZE;
507 			mss->pss += PAGE_SIZE << PSS_SHIFT;
508 		}
509 	}
510 }
511 
512 #ifdef CONFIG_SHMEM
513 static int smaps_pte_hole(unsigned long addr, unsigned long end,
514 		struct mm_walk *walk)
515 {
516 	struct mem_size_stats *mss = walk->private;
517 
518 	mss->swap += shmem_partial_swap_usage(
519 			walk->vma->vm_file->f_mapping, addr, end);
520 
521 	return 0;
522 }
523 #endif
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 	struct page *page = NULL;
531 
532 	if (pte_present(*pte)) {
533 		page = vm_normal_page(vma, addr, *pte);
534 	} else if (is_swap_pte(*pte)) {
535 		swp_entry_t swpent = pte_to_swp_entry(*pte);
536 
537 		if (!non_swap_entry(swpent)) {
538 			int mapcount;
539 
540 			mss->swap += PAGE_SIZE;
541 			mapcount = swp_swapcount(swpent);
542 			if (mapcount >= 2) {
543 				u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
544 
545 				do_div(pss_delta, mapcount);
546 				mss->swap_pss += pss_delta;
547 			} else {
548 				mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
549 			}
550 		} else if (is_migration_entry(swpent))
551 			page = migration_entry_to_page(swpent);
552 		else if (is_device_private_entry(swpent))
553 			page = device_private_entry_to_page(swpent);
554 	} else if (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap
555 							&& pte_none(*pte))) {
556 		page = find_get_entry(vma->vm_file->f_mapping,
557 						linear_page_index(vma, addr));
558 		if (!page)
559 			return;
560 
561 		if (radix_tree_exceptional_entry(page))
562 			mss->swap += PAGE_SIZE;
563 		else
564 			put_page(page);
565 
566 		return;
567 	}
568 
569 	if (!page)
570 		return;
571 
572 	smaps_account(mss, page, false, pte_young(*pte), pte_dirty(*pte));
573 }
574 
575 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
576 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
577 		struct mm_walk *walk)
578 {
579 	struct mem_size_stats *mss = walk->private;
580 	struct vm_area_struct *vma = walk->vma;
581 	struct page *page;
582 
583 	/* FOLL_DUMP will return -EFAULT on huge zero page */
584 	page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
585 	if (IS_ERR_OR_NULL(page))
586 		return;
587 	if (PageAnon(page))
588 		mss->anonymous_thp += HPAGE_PMD_SIZE;
589 	else if (PageSwapBacked(page))
590 		mss->shmem_thp += HPAGE_PMD_SIZE;
591 	else if (is_zone_device_page(page))
592 		/* pass */;
593 	else
594 		VM_BUG_ON_PAGE(1, page);
595 	smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd));
596 }
597 #else
598 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
599 		struct mm_walk *walk)
600 {
601 }
602 #endif
603 
604 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
605 			   struct mm_walk *walk)
606 {
607 	struct vm_area_struct *vma = walk->vma;
608 	pte_t *pte;
609 	spinlock_t *ptl;
610 
611 	ptl = pmd_trans_huge_lock(pmd, vma);
612 	if (ptl) {
613 		if (pmd_present(*pmd))
614 			smaps_pmd_entry(pmd, addr, walk);
615 		spin_unlock(ptl);
616 		goto out;
617 	}
618 
619 	if (pmd_trans_unstable(pmd))
620 		goto out;
621 	/*
622 	 * The mmap_sem held all the way back in m_start() is what
623 	 * keeps khugepaged out of here and from collapsing things
624 	 * in here.
625 	 */
626 	pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
627 	for (; addr != end; pte++, addr += PAGE_SIZE)
628 		smaps_pte_entry(pte, addr, walk);
629 	pte_unmap_unlock(pte - 1, ptl);
630 out:
631 	cond_resched();
632 	return 0;
633 }
634 
635 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
636 {
637 	/*
638 	 * Don't forget to update Documentation/ on changes.
639 	 */
640 	static const char mnemonics[BITS_PER_LONG][2] = {
641 		/*
642 		 * In case if we meet a flag we don't know about.
643 		 */
644 		[0 ... (BITS_PER_LONG-1)] = "??",
645 
646 		[ilog2(VM_READ)]	= "rd",
647 		[ilog2(VM_WRITE)]	= "wr",
648 		[ilog2(VM_EXEC)]	= "ex",
649 		[ilog2(VM_SHARED)]	= "sh",
650 		[ilog2(VM_MAYREAD)]	= "mr",
651 		[ilog2(VM_MAYWRITE)]	= "mw",
652 		[ilog2(VM_MAYEXEC)]	= "me",
653 		[ilog2(VM_MAYSHARE)]	= "ms",
654 		[ilog2(VM_GROWSDOWN)]	= "gd",
655 		[ilog2(VM_PFNMAP)]	= "pf",
656 		[ilog2(VM_DENYWRITE)]	= "dw",
657 #ifdef CONFIG_X86_INTEL_MPX
658 		[ilog2(VM_MPX)]		= "mp",
659 #endif
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_ACCOUNT)]	= "ac",
667 		[ilog2(VM_NORESERVE)]	= "nr",
668 		[ilog2(VM_HUGETLB)]	= "ht",
669 		[ilog2(VM_SYNC)]	= "sf",
670 		[ilog2(VM_ARCH_1)]	= "ar",
671 		[ilog2(VM_WIPEONFORK)]	= "wf",
672 		[ilog2(VM_DONTDUMP)]	= "dd",
673 #ifdef CONFIG_MEM_SOFT_DIRTY
674 		[ilog2(VM_SOFTDIRTY)]	= "sd",
675 #endif
676 		[ilog2(VM_MIXEDMAP)]	= "mm",
677 		[ilog2(VM_HUGEPAGE)]	= "hg",
678 		[ilog2(VM_NOHUGEPAGE)]	= "nh",
679 		[ilog2(VM_MERGEABLE)]	= "mg",
680 		[ilog2(VM_UFFD_MISSING)]= "um",
681 		[ilog2(VM_UFFD_WP)]	= "uw",
682 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
683 		/* These come out via ProtectionKey: */
684 		[ilog2(VM_PKEY_BIT0)]	= "",
685 		[ilog2(VM_PKEY_BIT1)]	= "",
686 		[ilog2(VM_PKEY_BIT2)]	= "",
687 		[ilog2(VM_PKEY_BIT3)]	= "",
688 #endif
689 	};
690 	size_t i;
691 
692 	seq_puts(m, "VmFlags: ");
693 	for (i = 0; i < BITS_PER_LONG; i++) {
694 		if (!mnemonics[i][0])
695 			continue;
696 		if (vma->vm_flags & (1UL << i)) {
697 			seq_printf(m, "%c%c ",
698 				   mnemonics[i][0], mnemonics[i][1]);
699 		}
700 	}
701 	seq_putc(m, '\n');
702 }
703 
704 #ifdef CONFIG_HUGETLB_PAGE
705 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
706 				 unsigned long addr, unsigned long end,
707 				 struct mm_walk *walk)
708 {
709 	struct mem_size_stats *mss = walk->private;
710 	struct vm_area_struct *vma = walk->vma;
711 	struct page *page = NULL;
712 
713 	if (pte_present(*pte)) {
714 		page = vm_normal_page(vma, addr, *pte);
715 	} else if (is_swap_pte(*pte)) {
716 		swp_entry_t swpent = pte_to_swp_entry(*pte);
717 
718 		if (is_migration_entry(swpent))
719 			page = migration_entry_to_page(swpent);
720 		else if (is_device_private_entry(swpent))
721 			page = device_private_entry_to_page(swpent);
722 	}
723 	if (page) {
724 		int mapcount = page_mapcount(page);
725 
726 		if (mapcount >= 2)
727 			mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
728 		else
729 			mss->private_hugetlb += huge_page_size(hstate_vma(vma));
730 	}
731 	return 0;
732 }
733 #endif /* HUGETLB_PAGE */
734 
735 void __weak arch_show_smap(struct seq_file *m, struct vm_area_struct *vma)
736 {
737 }
738 
739 static int show_smap(struct seq_file *m, void *v, int is_pid)
740 {
741 	struct proc_maps_private *priv = m->private;
742 	struct vm_area_struct *vma = v;
743 	struct mem_size_stats mss_stack;
744 	struct mem_size_stats *mss;
745 	struct mm_walk smaps_walk = {
746 		.pmd_entry = smaps_pte_range,
747 #ifdef CONFIG_HUGETLB_PAGE
748 		.hugetlb_entry = smaps_hugetlb_range,
749 #endif
750 		.mm = vma->vm_mm,
751 	};
752 	int ret = 0;
753 	bool rollup_mode;
754 	bool last_vma;
755 
756 	if (priv->rollup) {
757 		rollup_mode = true;
758 		mss = priv->rollup;
759 		if (mss->first) {
760 			mss->first_vma_start = vma->vm_start;
761 			mss->first = false;
762 		}
763 		last_vma = !m_next_vma(priv, vma);
764 	} else {
765 		rollup_mode = false;
766 		memset(&mss_stack, 0, sizeof(mss_stack));
767 		mss = &mss_stack;
768 	}
769 
770 	smaps_walk.private = mss;
771 
772 #ifdef CONFIG_SHMEM
773 	if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
774 		/*
775 		 * For shared or readonly shmem mappings we know that all
776 		 * swapped out pages belong to the shmem object, and we can
777 		 * obtain the swap value much more efficiently. For private
778 		 * writable mappings, we might have COW pages that are
779 		 * not affected by the parent swapped out pages of the shmem
780 		 * object, so we have to distinguish them during the page walk.
781 		 * Unless we know that the shmem object (or the part mapped by
782 		 * our VMA) has no swapped out pages at all.
783 		 */
784 		unsigned long shmem_swapped = shmem_swap_usage(vma);
785 
786 		if (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
787 					!(vma->vm_flags & VM_WRITE)) {
788 			mss->swap = shmem_swapped;
789 		} else {
790 			mss->check_shmem_swap = true;
791 			smaps_walk.pte_hole = smaps_pte_hole;
792 		}
793 	}
794 #endif
795 
796 	/* mmap_sem is held in m_start */
797 	walk_page_vma(vma, &smaps_walk);
798 	if (vma->vm_flags & VM_LOCKED)
799 		mss->pss_locked += mss->pss;
800 
801 	if (!rollup_mode) {
802 		show_map_vma(m, vma, is_pid);
803 	} else if (last_vma) {
804 		show_vma_header_prefix(
805 			m, mss->first_vma_start, vma->vm_end, 0, 0, 0, 0);
806 		seq_pad(m, ' ');
807 		seq_puts(m, "[rollup]\n");
808 	} else {
809 		ret = SEQ_SKIP;
810 	}
811 
812 	if (!rollup_mode)
813 		seq_printf(m,
814 			   "Size:           %8lu kB\n"
815 			   "KernelPageSize: %8lu kB\n"
816 			   "MMUPageSize:    %8lu kB\n",
817 			   (vma->vm_end - vma->vm_start) >> 10,
818 			   vma_kernel_pagesize(vma) >> 10,
819 			   vma_mmu_pagesize(vma) >> 10);
820 
821 
822 	if (!rollup_mode || last_vma)
823 		seq_printf(m,
824 			   "Rss:            %8lu kB\n"
825 			   "Pss:            %8lu kB\n"
826 			   "Shared_Clean:   %8lu kB\n"
827 			   "Shared_Dirty:   %8lu kB\n"
828 			   "Private_Clean:  %8lu kB\n"
829 			   "Private_Dirty:  %8lu kB\n"
830 			   "Referenced:     %8lu kB\n"
831 			   "Anonymous:      %8lu kB\n"
832 			   "LazyFree:       %8lu kB\n"
833 			   "AnonHugePages:  %8lu kB\n"
834 			   "ShmemPmdMapped: %8lu kB\n"
835 			   "Shared_Hugetlb: %8lu kB\n"
836 			   "Private_Hugetlb: %7lu kB\n"
837 			   "Swap:           %8lu kB\n"
838 			   "SwapPss:        %8lu kB\n"
839 			   "Locked:         %8lu kB\n",
840 			   mss->resident >> 10,
841 			   (unsigned long)(mss->pss >> (10 + PSS_SHIFT)),
842 			   mss->shared_clean  >> 10,
843 			   mss->shared_dirty  >> 10,
844 			   mss->private_clean >> 10,
845 			   mss->private_dirty >> 10,
846 			   mss->referenced >> 10,
847 			   mss->anonymous >> 10,
848 			   mss->lazyfree >> 10,
849 			   mss->anonymous_thp >> 10,
850 			   mss->shmem_thp >> 10,
851 			   mss->shared_hugetlb >> 10,
852 			   mss->private_hugetlb >> 10,
853 			   mss->swap >> 10,
854 			   (unsigned long)(mss->swap_pss >> (10 + PSS_SHIFT)),
855 			   (unsigned long)(mss->pss >> (10 + PSS_SHIFT)));
856 
857 	if (!rollup_mode) {
858 		arch_show_smap(m, vma);
859 		show_smap_vma_flags(m, vma);
860 	}
861 	m_cache_vma(m, vma);
862 	return ret;
863 }
864 
865 static int show_pid_smap(struct seq_file *m, void *v)
866 {
867 	return show_smap(m, v, 1);
868 }
869 
870 static int show_tid_smap(struct seq_file *m, void *v)
871 {
872 	return show_smap(m, v, 0);
873 }
874 
875 static const struct seq_operations proc_pid_smaps_op = {
876 	.start	= m_start,
877 	.next	= m_next,
878 	.stop	= m_stop,
879 	.show	= show_pid_smap
880 };
881 
882 static const struct seq_operations proc_tid_smaps_op = {
883 	.start	= m_start,
884 	.next	= m_next,
885 	.stop	= m_stop,
886 	.show	= show_tid_smap
887 };
888 
889 static int pid_smaps_open(struct inode *inode, struct file *file)
890 {
891 	return do_maps_open(inode, file, &proc_pid_smaps_op);
892 }
893 
894 static int pid_smaps_rollup_open(struct inode *inode, struct file *file)
895 {
896 	struct seq_file *seq;
897 	struct proc_maps_private *priv;
898 	int ret = do_maps_open(inode, file, &proc_pid_smaps_op);
899 
900 	if (ret < 0)
901 		return ret;
902 	seq = file->private_data;
903 	priv = seq->private;
904 	priv->rollup = kzalloc(sizeof(*priv->rollup), GFP_KERNEL);
905 	if (!priv->rollup) {
906 		proc_map_release(inode, file);
907 		return -ENOMEM;
908 	}
909 	priv->rollup->first = true;
910 	return 0;
911 }
912 
913 static int tid_smaps_open(struct inode *inode, struct file *file)
914 {
915 	return do_maps_open(inode, file, &proc_tid_smaps_op);
916 }
917 
918 const struct file_operations proc_pid_smaps_operations = {
919 	.open		= pid_smaps_open,
920 	.read		= seq_read,
921 	.llseek		= seq_lseek,
922 	.release	= proc_map_release,
923 };
924 
925 const struct file_operations proc_pid_smaps_rollup_operations = {
926 	.open		= pid_smaps_rollup_open,
927 	.read		= seq_read,
928 	.llseek		= seq_lseek,
929 	.release	= proc_map_release,
930 };
931 
932 const struct file_operations proc_tid_smaps_operations = {
933 	.open		= tid_smaps_open,
934 	.read		= seq_read,
935 	.llseek		= seq_lseek,
936 	.release	= proc_map_release,
937 };
938 
939 enum clear_refs_types {
940 	CLEAR_REFS_ALL = 1,
941 	CLEAR_REFS_ANON,
942 	CLEAR_REFS_MAPPED,
943 	CLEAR_REFS_SOFT_DIRTY,
944 	CLEAR_REFS_MM_HIWATER_RSS,
945 	CLEAR_REFS_LAST,
946 };
947 
948 struct clear_refs_private {
949 	enum clear_refs_types type;
950 };
951 
952 #ifdef CONFIG_MEM_SOFT_DIRTY
953 static inline void clear_soft_dirty(struct vm_area_struct *vma,
954 		unsigned long addr, pte_t *pte)
955 {
956 	/*
957 	 * The soft-dirty tracker uses #PF-s to catch writes
958 	 * to pages, so write-protect the pte as well. See the
959 	 * Documentation/vm/soft-dirty.txt for full description
960 	 * of how soft-dirty works.
961 	 */
962 	pte_t ptent = *pte;
963 
964 	if (pte_present(ptent)) {
965 		ptent = ptep_modify_prot_start(vma->vm_mm, addr, pte);
966 		ptent = pte_wrprotect(ptent);
967 		ptent = pte_clear_soft_dirty(ptent);
968 		ptep_modify_prot_commit(vma->vm_mm, addr, pte, ptent);
969 	} else if (is_swap_pte(ptent)) {
970 		ptent = pte_swp_clear_soft_dirty(ptent);
971 		set_pte_at(vma->vm_mm, addr, pte, ptent);
972 	}
973 }
974 #else
975 static inline void clear_soft_dirty(struct vm_area_struct *vma,
976 		unsigned long addr, pte_t *pte)
977 {
978 }
979 #endif
980 
981 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
982 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
983 		unsigned long addr, pmd_t *pmdp)
984 {
985 	pmd_t old, pmd = *pmdp;
986 
987 	if (pmd_present(pmd)) {
988 		/* See comment in change_huge_pmd() */
989 		old = pmdp_invalidate(vma, addr, pmdp);
990 		if (pmd_dirty(old))
991 			pmd = pmd_mkdirty(pmd);
992 		if (pmd_young(old))
993 			pmd = pmd_mkyoung(pmd);
994 
995 		pmd = pmd_wrprotect(pmd);
996 		pmd = pmd_clear_soft_dirty(pmd);
997 
998 		set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
999 	} else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1000 		pmd = pmd_swp_clear_soft_dirty(pmd);
1001 		set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1002 	}
1003 }
1004 #else
1005 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1006 		unsigned long addr, pmd_t *pmdp)
1007 {
1008 }
1009 #endif
1010 
1011 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
1012 				unsigned long end, struct mm_walk *walk)
1013 {
1014 	struct clear_refs_private *cp = walk->private;
1015 	struct vm_area_struct *vma = walk->vma;
1016 	pte_t *pte, ptent;
1017 	spinlock_t *ptl;
1018 	struct page *page;
1019 
1020 	ptl = pmd_trans_huge_lock(pmd, vma);
1021 	if (ptl) {
1022 		if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1023 			clear_soft_dirty_pmd(vma, addr, pmd);
1024 			goto out;
1025 		}
1026 
1027 		if (!pmd_present(*pmd))
1028 			goto out;
1029 
1030 		page = pmd_page(*pmd);
1031 
1032 		/* Clear accessed and referenced bits. */
1033 		pmdp_test_and_clear_young(vma, addr, pmd);
1034 		test_and_clear_page_young(page);
1035 		ClearPageReferenced(page);
1036 out:
1037 		spin_unlock(ptl);
1038 		return 0;
1039 	}
1040 
1041 	if (pmd_trans_unstable(pmd))
1042 		return 0;
1043 
1044 	pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1045 	for (; addr != end; pte++, addr += PAGE_SIZE) {
1046 		ptent = *pte;
1047 
1048 		if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1049 			clear_soft_dirty(vma, addr, pte);
1050 			continue;
1051 		}
1052 
1053 		if (!pte_present(ptent))
1054 			continue;
1055 
1056 		page = vm_normal_page(vma, addr, ptent);
1057 		if (!page)
1058 			continue;
1059 
1060 		/* Clear accessed and referenced bits. */
1061 		ptep_test_and_clear_young(vma, addr, pte);
1062 		test_and_clear_page_young(page);
1063 		ClearPageReferenced(page);
1064 	}
1065 	pte_unmap_unlock(pte - 1, ptl);
1066 	cond_resched();
1067 	return 0;
1068 }
1069 
1070 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1071 				struct mm_walk *walk)
1072 {
1073 	struct clear_refs_private *cp = walk->private;
1074 	struct vm_area_struct *vma = walk->vma;
1075 
1076 	if (vma->vm_flags & VM_PFNMAP)
1077 		return 1;
1078 
1079 	/*
1080 	 * Writing 1 to /proc/pid/clear_refs affects all pages.
1081 	 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1082 	 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1083 	 * Writing 4 to /proc/pid/clear_refs affects all pages.
1084 	 */
1085 	if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1086 		return 1;
1087 	if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1088 		return 1;
1089 	return 0;
1090 }
1091 
1092 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1093 				size_t count, loff_t *ppos)
1094 {
1095 	struct task_struct *task;
1096 	char buffer[PROC_NUMBUF];
1097 	struct mm_struct *mm;
1098 	struct vm_area_struct *vma;
1099 	enum clear_refs_types type;
1100 	struct mmu_gather tlb;
1101 	int itype;
1102 	int rv;
1103 
1104 	memset(buffer, 0, sizeof(buffer));
1105 	if (count > sizeof(buffer) - 1)
1106 		count = sizeof(buffer) - 1;
1107 	if (copy_from_user(buffer, buf, count))
1108 		return -EFAULT;
1109 	rv = kstrtoint(strstrip(buffer), 10, &itype);
1110 	if (rv < 0)
1111 		return rv;
1112 	type = (enum clear_refs_types)itype;
1113 	if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1114 		return -EINVAL;
1115 
1116 	task = get_proc_task(file_inode(file));
1117 	if (!task)
1118 		return -ESRCH;
1119 	mm = get_task_mm(task);
1120 	if (mm) {
1121 		struct clear_refs_private cp = {
1122 			.type = type,
1123 		};
1124 		struct mm_walk clear_refs_walk = {
1125 			.pmd_entry = clear_refs_pte_range,
1126 			.test_walk = clear_refs_test_walk,
1127 			.mm = mm,
1128 			.private = &cp,
1129 		};
1130 
1131 		if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1132 			if (down_write_killable(&mm->mmap_sem)) {
1133 				count = -EINTR;
1134 				goto out_mm;
1135 			}
1136 
1137 			/*
1138 			 * Writing 5 to /proc/pid/clear_refs resets the peak
1139 			 * resident set size to this mm's current rss value.
1140 			 */
1141 			reset_mm_hiwater_rss(mm);
1142 			up_write(&mm->mmap_sem);
1143 			goto out_mm;
1144 		}
1145 
1146 		down_read(&mm->mmap_sem);
1147 		tlb_gather_mmu(&tlb, mm, 0, -1);
1148 		if (type == CLEAR_REFS_SOFT_DIRTY) {
1149 			for (vma = mm->mmap; vma; vma = vma->vm_next) {
1150 				if (!(vma->vm_flags & VM_SOFTDIRTY))
1151 					continue;
1152 				up_read(&mm->mmap_sem);
1153 				if (down_write_killable(&mm->mmap_sem)) {
1154 					count = -EINTR;
1155 					goto out_mm;
1156 				}
1157 				for (vma = mm->mmap; vma; vma = vma->vm_next) {
1158 					vma->vm_flags &= ~VM_SOFTDIRTY;
1159 					vma_set_page_prot(vma);
1160 				}
1161 				downgrade_write(&mm->mmap_sem);
1162 				break;
1163 			}
1164 			mmu_notifier_invalidate_range_start(mm, 0, -1);
1165 		}
1166 		walk_page_range(0, mm->highest_vm_end, &clear_refs_walk);
1167 		if (type == CLEAR_REFS_SOFT_DIRTY)
1168 			mmu_notifier_invalidate_range_end(mm, 0, -1);
1169 		tlb_finish_mmu(&tlb, 0, -1);
1170 		up_read(&mm->mmap_sem);
1171 out_mm:
1172 		mmput(mm);
1173 	}
1174 	put_task_struct(task);
1175 
1176 	return count;
1177 }
1178 
1179 const struct file_operations proc_clear_refs_operations = {
1180 	.write		= clear_refs_write,
1181 	.llseek		= noop_llseek,
1182 };
1183 
1184 typedef struct {
1185 	u64 pme;
1186 } pagemap_entry_t;
1187 
1188 struct pagemapread {
1189 	int pos, len;		/* units: PM_ENTRY_BYTES, not bytes */
1190 	pagemap_entry_t *buffer;
1191 	bool show_pfn;
1192 };
1193 
1194 #define PAGEMAP_WALK_SIZE	(PMD_SIZE)
1195 #define PAGEMAP_WALK_MASK	(PMD_MASK)
1196 
1197 #define PM_ENTRY_BYTES		sizeof(pagemap_entry_t)
1198 #define PM_PFRAME_BITS		55
1199 #define PM_PFRAME_MASK		GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1200 #define PM_SOFT_DIRTY		BIT_ULL(55)
1201 #define PM_MMAP_EXCLUSIVE	BIT_ULL(56)
1202 #define PM_FILE			BIT_ULL(61)
1203 #define PM_SWAP			BIT_ULL(62)
1204 #define PM_PRESENT		BIT_ULL(63)
1205 
1206 #define PM_END_OF_BUFFER    1
1207 
1208 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1209 {
1210 	return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1211 }
1212 
1213 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1214 			  struct pagemapread *pm)
1215 {
1216 	pm->buffer[pm->pos++] = *pme;
1217 	if (pm->pos >= pm->len)
1218 		return PM_END_OF_BUFFER;
1219 	return 0;
1220 }
1221 
1222 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1223 				struct mm_walk *walk)
1224 {
1225 	struct pagemapread *pm = walk->private;
1226 	unsigned long addr = start;
1227 	int err = 0;
1228 
1229 	while (addr < end) {
1230 		struct vm_area_struct *vma = find_vma(walk->mm, addr);
1231 		pagemap_entry_t pme = make_pme(0, 0);
1232 		/* End of address space hole, which we mark as non-present. */
1233 		unsigned long hole_end;
1234 
1235 		if (vma)
1236 			hole_end = min(end, vma->vm_start);
1237 		else
1238 			hole_end = end;
1239 
1240 		for (; addr < hole_end; addr += PAGE_SIZE) {
1241 			err = add_to_pagemap(addr, &pme, pm);
1242 			if (err)
1243 				goto out;
1244 		}
1245 
1246 		if (!vma)
1247 			break;
1248 
1249 		/* Addresses in the VMA. */
1250 		if (vma->vm_flags & VM_SOFTDIRTY)
1251 			pme = make_pme(0, PM_SOFT_DIRTY);
1252 		for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1253 			err = add_to_pagemap(addr, &pme, pm);
1254 			if (err)
1255 				goto out;
1256 		}
1257 	}
1258 out:
1259 	return err;
1260 }
1261 
1262 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1263 		struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1264 {
1265 	u64 frame = 0, flags = 0;
1266 	struct page *page = NULL;
1267 
1268 	if (pte_present(pte)) {
1269 		if (pm->show_pfn)
1270 			frame = pte_pfn(pte);
1271 		flags |= PM_PRESENT;
1272 		page = _vm_normal_page(vma, addr, pte, true);
1273 		if (pte_soft_dirty(pte))
1274 			flags |= PM_SOFT_DIRTY;
1275 	} else if (is_swap_pte(pte)) {
1276 		swp_entry_t entry;
1277 		if (pte_swp_soft_dirty(pte))
1278 			flags |= PM_SOFT_DIRTY;
1279 		entry = pte_to_swp_entry(pte);
1280 		frame = swp_type(entry) |
1281 			(swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1282 		flags |= PM_SWAP;
1283 		if (is_migration_entry(entry))
1284 			page = migration_entry_to_page(entry);
1285 
1286 		if (is_device_private_entry(entry))
1287 			page = device_private_entry_to_page(entry);
1288 	}
1289 
1290 	if (page && !PageAnon(page))
1291 		flags |= PM_FILE;
1292 	if (page && page_mapcount(page) == 1)
1293 		flags |= PM_MMAP_EXCLUSIVE;
1294 	if (vma->vm_flags & VM_SOFTDIRTY)
1295 		flags |= PM_SOFT_DIRTY;
1296 
1297 	return make_pme(frame, flags);
1298 }
1299 
1300 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1301 			     struct mm_walk *walk)
1302 {
1303 	struct vm_area_struct *vma = walk->vma;
1304 	struct pagemapread *pm = walk->private;
1305 	spinlock_t *ptl;
1306 	pte_t *pte, *orig_pte;
1307 	int err = 0;
1308 
1309 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1310 	ptl = pmd_trans_huge_lock(pmdp, vma);
1311 	if (ptl) {
1312 		u64 flags = 0, frame = 0;
1313 		pmd_t pmd = *pmdp;
1314 		struct page *page = NULL;
1315 
1316 		if (vma->vm_flags & VM_SOFTDIRTY)
1317 			flags |= PM_SOFT_DIRTY;
1318 
1319 		if (pmd_present(pmd)) {
1320 			page = pmd_page(pmd);
1321 
1322 			flags |= PM_PRESENT;
1323 			if (pmd_soft_dirty(pmd))
1324 				flags |= PM_SOFT_DIRTY;
1325 			if (pm->show_pfn)
1326 				frame = pmd_pfn(pmd) +
1327 					((addr & ~PMD_MASK) >> PAGE_SHIFT);
1328 		}
1329 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1330 		else if (is_swap_pmd(pmd)) {
1331 			swp_entry_t entry = pmd_to_swp_entry(pmd);
1332 
1333 			frame = swp_type(entry) |
1334 				(swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1335 			flags |= PM_SWAP;
1336 			if (pmd_swp_soft_dirty(pmd))
1337 				flags |= PM_SOFT_DIRTY;
1338 			VM_BUG_ON(!is_pmd_migration_entry(pmd));
1339 			page = migration_entry_to_page(entry);
1340 		}
1341 #endif
1342 
1343 		if (page && page_mapcount(page) == 1)
1344 			flags |= PM_MMAP_EXCLUSIVE;
1345 
1346 		for (; addr != end; addr += PAGE_SIZE) {
1347 			pagemap_entry_t pme = make_pme(frame, flags);
1348 
1349 			err = add_to_pagemap(addr, &pme, pm);
1350 			if (err)
1351 				break;
1352 			if (pm->show_pfn && (flags & PM_PRESENT))
1353 				frame++;
1354 		}
1355 		spin_unlock(ptl);
1356 		return err;
1357 	}
1358 
1359 	if (pmd_trans_unstable(pmdp))
1360 		return 0;
1361 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1362 
1363 	/*
1364 	 * We can assume that @vma always points to a valid one and @end never
1365 	 * goes beyond vma->vm_end.
1366 	 */
1367 	orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1368 	for (; addr < end; pte++, addr += PAGE_SIZE) {
1369 		pagemap_entry_t pme;
1370 
1371 		pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1372 		err = add_to_pagemap(addr, &pme, pm);
1373 		if (err)
1374 			break;
1375 	}
1376 	pte_unmap_unlock(orig_pte, ptl);
1377 
1378 	cond_resched();
1379 
1380 	return err;
1381 }
1382 
1383 #ifdef CONFIG_HUGETLB_PAGE
1384 /* This function walks within one hugetlb entry in the single call */
1385 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1386 				 unsigned long addr, unsigned long end,
1387 				 struct mm_walk *walk)
1388 {
1389 	struct pagemapread *pm = walk->private;
1390 	struct vm_area_struct *vma = walk->vma;
1391 	u64 flags = 0, frame = 0;
1392 	int err = 0;
1393 	pte_t pte;
1394 
1395 	if (vma->vm_flags & VM_SOFTDIRTY)
1396 		flags |= PM_SOFT_DIRTY;
1397 
1398 	pte = huge_ptep_get(ptep);
1399 	if (pte_present(pte)) {
1400 		struct page *page = pte_page(pte);
1401 
1402 		if (!PageAnon(page))
1403 			flags |= PM_FILE;
1404 
1405 		if (page_mapcount(page) == 1)
1406 			flags |= PM_MMAP_EXCLUSIVE;
1407 
1408 		flags |= PM_PRESENT;
1409 		if (pm->show_pfn)
1410 			frame = pte_pfn(pte) +
1411 				((addr & ~hmask) >> PAGE_SHIFT);
1412 	}
1413 
1414 	for (; addr != end; addr += PAGE_SIZE) {
1415 		pagemap_entry_t pme = make_pme(frame, flags);
1416 
1417 		err = add_to_pagemap(addr, &pme, pm);
1418 		if (err)
1419 			return err;
1420 		if (pm->show_pfn && (flags & PM_PRESENT))
1421 			frame++;
1422 	}
1423 
1424 	cond_resched();
1425 
1426 	return err;
1427 }
1428 #endif /* HUGETLB_PAGE */
1429 
1430 /*
1431  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1432  *
1433  * For each page in the address space, this file contains one 64-bit entry
1434  * consisting of the following:
1435  *
1436  * Bits 0-54  page frame number (PFN) if present
1437  * Bits 0-4   swap type if swapped
1438  * Bits 5-54  swap offset if swapped
1439  * Bit  55    pte is soft-dirty (see Documentation/vm/soft-dirty.txt)
1440  * Bit  56    page exclusively mapped
1441  * Bits 57-60 zero
1442  * Bit  61    page is file-page or shared-anon
1443  * Bit  62    page swapped
1444  * Bit  63    page present
1445  *
1446  * If the page is not present but in swap, then the PFN contains an
1447  * encoding of the swap file number and the page's offset into the
1448  * swap. Unmapped pages return a null PFN. This allows determining
1449  * precisely which pages are mapped (or in swap) and comparing mapped
1450  * pages between processes.
1451  *
1452  * Efficient users of this interface will use /proc/pid/maps to
1453  * determine which areas of memory are actually mapped and llseek to
1454  * skip over unmapped regions.
1455  */
1456 static ssize_t pagemap_read(struct file *file, char __user *buf,
1457 			    size_t count, loff_t *ppos)
1458 {
1459 	struct mm_struct *mm = file->private_data;
1460 	struct pagemapread pm;
1461 	struct mm_walk pagemap_walk = {};
1462 	unsigned long src;
1463 	unsigned long svpfn;
1464 	unsigned long start_vaddr;
1465 	unsigned long end_vaddr;
1466 	int ret = 0, copied = 0;
1467 
1468 	if (!mm || !mmget_not_zero(mm))
1469 		goto out;
1470 
1471 	ret = -EINVAL;
1472 	/* file position must be aligned */
1473 	if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1474 		goto out_mm;
1475 
1476 	ret = 0;
1477 	if (!count)
1478 		goto out_mm;
1479 
1480 	/* do not disclose physical addresses: attack vector */
1481 	pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1482 
1483 	pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1484 	pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_KERNEL);
1485 	ret = -ENOMEM;
1486 	if (!pm.buffer)
1487 		goto out_mm;
1488 
1489 	pagemap_walk.pmd_entry = pagemap_pmd_range;
1490 	pagemap_walk.pte_hole = pagemap_pte_hole;
1491 #ifdef CONFIG_HUGETLB_PAGE
1492 	pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1493 #endif
1494 	pagemap_walk.mm = mm;
1495 	pagemap_walk.private = &pm;
1496 
1497 	src = *ppos;
1498 	svpfn = src / PM_ENTRY_BYTES;
1499 	start_vaddr = svpfn << PAGE_SHIFT;
1500 	end_vaddr = mm->task_size;
1501 
1502 	/* watch out for wraparound */
1503 	if (svpfn > mm->task_size >> PAGE_SHIFT)
1504 		start_vaddr = end_vaddr;
1505 
1506 	/*
1507 	 * The odds are that this will stop walking way
1508 	 * before end_vaddr, because the length of the
1509 	 * user buffer is tracked in "pm", and the walk
1510 	 * will stop when we hit the end of the buffer.
1511 	 */
1512 	ret = 0;
1513 	while (count && (start_vaddr < end_vaddr)) {
1514 		int len;
1515 		unsigned long end;
1516 
1517 		pm.pos = 0;
1518 		end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1519 		/* overflow ? */
1520 		if (end < start_vaddr || end > end_vaddr)
1521 			end = end_vaddr;
1522 		down_read(&mm->mmap_sem);
1523 		ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1524 		up_read(&mm->mmap_sem);
1525 		start_vaddr = end;
1526 
1527 		len = min(count, PM_ENTRY_BYTES * pm.pos);
1528 		if (copy_to_user(buf, pm.buffer, len)) {
1529 			ret = -EFAULT;
1530 			goto out_free;
1531 		}
1532 		copied += len;
1533 		buf += len;
1534 		count -= len;
1535 	}
1536 	*ppos += copied;
1537 	if (!ret || ret == PM_END_OF_BUFFER)
1538 		ret = copied;
1539 
1540 out_free:
1541 	kfree(pm.buffer);
1542 out_mm:
1543 	mmput(mm);
1544 out:
1545 	return ret;
1546 }
1547 
1548 static int pagemap_open(struct inode *inode, struct file *file)
1549 {
1550 	struct mm_struct *mm;
1551 
1552 	mm = proc_mem_open(inode, PTRACE_MODE_READ);
1553 	if (IS_ERR(mm))
1554 		return PTR_ERR(mm);
1555 	file->private_data = mm;
1556 	return 0;
1557 }
1558 
1559 static int pagemap_release(struct inode *inode, struct file *file)
1560 {
1561 	struct mm_struct *mm = file->private_data;
1562 
1563 	if (mm)
1564 		mmdrop(mm);
1565 	return 0;
1566 }
1567 
1568 const struct file_operations proc_pagemap_operations = {
1569 	.llseek		= mem_lseek, /* borrow this */
1570 	.read		= pagemap_read,
1571 	.open		= pagemap_open,
1572 	.release	= pagemap_release,
1573 };
1574 #endif /* CONFIG_PROC_PAGE_MONITOR */
1575 
1576 #ifdef CONFIG_NUMA
1577 
1578 struct numa_maps {
1579 	unsigned long pages;
1580 	unsigned long anon;
1581 	unsigned long active;
1582 	unsigned long writeback;
1583 	unsigned long mapcount_max;
1584 	unsigned long dirty;
1585 	unsigned long swapcache;
1586 	unsigned long node[MAX_NUMNODES];
1587 };
1588 
1589 struct numa_maps_private {
1590 	struct proc_maps_private proc_maps;
1591 	struct numa_maps md;
1592 };
1593 
1594 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1595 			unsigned long nr_pages)
1596 {
1597 	int count = page_mapcount(page);
1598 
1599 	md->pages += nr_pages;
1600 	if (pte_dirty || PageDirty(page))
1601 		md->dirty += nr_pages;
1602 
1603 	if (PageSwapCache(page))
1604 		md->swapcache += nr_pages;
1605 
1606 	if (PageActive(page) || PageUnevictable(page))
1607 		md->active += nr_pages;
1608 
1609 	if (PageWriteback(page))
1610 		md->writeback += nr_pages;
1611 
1612 	if (PageAnon(page))
1613 		md->anon += nr_pages;
1614 
1615 	if (count > md->mapcount_max)
1616 		md->mapcount_max = count;
1617 
1618 	md->node[page_to_nid(page)] += nr_pages;
1619 }
1620 
1621 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1622 		unsigned long addr)
1623 {
1624 	struct page *page;
1625 	int nid;
1626 
1627 	if (!pte_present(pte))
1628 		return NULL;
1629 
1630 	page = vm_normal_page(vma, addr, pte);
1631 	if (!page)
1632 		return NULL;
1633 
1634 	if (PageReserved(page))
1635 		return NULL;
1636 
1637 	nid = page_to_nid(page);
1638 	if (!node_isset(nid, node_states[N_MEMORY]))
1639 		return NULL;
1640 
1641 	return page;
1642 }
1643 
1644 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1645 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1646 					      struct vm_area_struct *vma,
1647 					      unsigned long addr)
1648 {
1649 	struct page *page;
1650 	int nid;
1651 
1652 	if (!pmd_present(pmd))
1653 		return NULL;
1654 
1655 	page = vm_normal_page_pmd(vma, addr, pmd);
1656 	if (!page)
1657 		return NULL;
1658 
1659 	if (PageReserved(page))
1660 		return NULL;
1661 
1662 	nid = page_to_nid(page);
1663 	if (!node_isset(nid, node_states[N_MEMORY]))
1664 		return NULL;
1665 
1666 	return page;
1667 }
1668 #endif
1669 
1670 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1671 		unsigned long end, struct mm_walk *walk)
1672 {
1673 	struct numa_maps *md = walk->private;
1674 	struct vm_area_struct *vma = walk->vma;
1675 	spinlock_t *ptl;
1676 	pte_t *orig_pte;
1677 	pte_t *pte;
1678 
1679 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1680 	ptl = pmd_trans_huge_lock(pmd, vma);
1681 	if (ptl) {
1682 		struct page *page;
1683 
1684 		page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1685 		if (page)
1686 			gather_stats(page, md, pmd_dirty(*pmd),
1687 				     HPAGE_PMD_SIZE/PAGE_SIZE);
1688 		spin_unlock(ptl);
1689 		return 0;
1690 	}
1691 
1692 	if (pmd_trans_unstable(pmd))
1693 		return 0;
1694 #endif
1695 	orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1696 	do {
1697 		struct page *page = can_gather_numa_stats(*pte, vma, addr);
1698 		if (!page)
1699 			continue;
1700 		gather_stats(page, md, pte_dirty(*pte), 1);
1701 
1702 	} while (pte++, addr += PAGE_SIZE, addr != end);
1703 	pte_unmap_unlock(orig_pte, ptl);
1704 	cond_resched();
1705 	return 0;
1706 }
1707 #ifdef CONFIG_HUGETLB_PAGE
1708 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1709 		unsigned long addr, unsigned long end, struct mm_walk *walk)
1710 {
1711 	pte_t huge_pte = huge_ptep_get(pte);
1712 	struct numa_maps *md;
1713 	struct page *page;
1714 
1715 	if (!pte_present(huge_pte))
1716 		return 0;
1717 
1718 	page = pte_page(huge_pte);
1719 	if (!page)
1720 		return 0;
1721 
1722 	md = walk->private;
1723 	gather_stats(page, md, pte_dirty(huge_pte), 1);
1724 	return 0;
1725 }
1726 
1727 #else
1728 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1729 		unsigned long addr, unsigned long end, struct mm_walk *walk)
1730 {
1731 	return 0;
1732 }
1733 #endif
1734 
1735 /*
1736  * Display pages allocated per node and memory policy via /proc.
1737  */
1738 static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1739 {
1740 	struct numa_maps_private *numa_priv = m->private;
1741 	struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1742 	struct vm_area_struct *vma = v;
1743 	struct numa_maps *md = &numa_priv->md;
1744 	struct file *file = vma->vm_file;
1745 	struct mm_struct *mm = vma->vm_mm;
1746 	struct mm_walk walk = {
1747 		.hugetlb_entry = gather_hugetlb_stats,
1748 		.pmd_entry = gather_pte_stats,
1749 		.private = md,
1750 		.mm = mm,
1751 	};
1752 	struct mempolicy *pol;
1753 	char buffer[64];
1754 	int nid;
1755 
1756 	if (!mm)
1757 		return 0;
1758 
1759 	/* Ensure we start with an empty set of numa_maps statistics. */
1760 	memset(md, 0, sizeof(*md));
1761 
1762 	pol = __get_vma_policy(vma, vma->vm_start);
1763 	if (pol) {
1764 		mpol_to_str(buffer, sizeof(buffer), pol);
1765 		mpol_cond_put(pol);
1766 	} else {
1767 		mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1768 	}
1769 
1770 	seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1771 
1772 	if (file) {
1773 		seq_puts(m, " file=");
1774 		seq_file_path(m, file, "\n\t= ");
1775 	} else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1776 		seq_puts(m, " heap");
1777 	} else if (is_stack(vma)) {
1778 		seq_puts(m, " stack");
1779 	}
1780 
1781 	if (is_vm_hugetlb_page(vma))
1782 		seq_puts(m, " huge");
1783 
1784 	/* mmap_sem is held by m_start */
1785 	walk_page_vma(vma, &walk);
1786 
1787 	if (!md->pages)
1788 		goto out;
1789 
1790 	if (md->anon)
1791 		seq_printf(m, " anon=%lu", md->anon);
1792 
1793 	if (md->dirty)
1794 		seq_printf(m, " dirty=%lu", md->dirty);
1795 
1796 	if (md->pages != md->anon && md->pages != md->dirty)
1797 		seq_printf(m, " mapped=%lu", md->pages);
1798 
1799 	if (md->mapcount_max > 1)
1800 		seq_printf(m, " mapmax=%lu", md->mapcount_max);
1801 
1802 	if (md->swapcache)
1803 		seq_printf(m, " swapcache=%lu", md->swapcache);
1804 
1805 	if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1806 		seq_printf(m, " active=%lu", md->active);
1807 
1808 	if (md->writeback)
1809 		seq_printf(m, " writeback=%lu", md->writeback);
1810 
1811 	for_each_node_state(nid, N_MEMORY)
1812 		if (md->node[nid])
1813 			seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1814 
1815 	seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1816 out:
1817 	seq_putc(m, '\n');
1818 	m_cache_vma(m, vma);
1819 	return 0;
1820 }
1821 
1822 static int show_pid_numa_map(struct seq_file *m, void *v)
1823 {
1824 	return show_numa_map(m, v, 1);
1825 }
1826 
1827 static int show_tid_numa_map(struct seq_file *m, void *v)
1828 {
1829 	return show_numa_map(m, v, 0);
1830 }
1831 
1832 static const struct seq_operations proc_pid_numa_maps_op = {
1833 	.start  = m_start,
1834 	.next   = m_next,
1835 	.stop   = m_stop,
1836 	.show   = show_pid_numa_map,
1837 };
1838 
1839 static const struct seq_operations proc_tid_numa_maps_op = {
1840 	.start  = m_start,
1841 	.next   = m_next,
1842 	.stop   = m_stop,
1843 	.show   = show_tid_numa_map,
1844 };
1845 
1846 static int numa_maps_open(struct inode *inode, struct file *file,
1847 			  const struct seq_operations *ops)
1848 {
1849 	return proc_maps_open(inode, file, ops,
1850 				sizeof(struct numa_maps_private));
1851 }
1852 
1853 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1854 {
1855 	return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1856 }
1857 
1858 static int tid_numa_maps_open(struct inode *inode, struct file *file)
1859 {
1860 	return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1861 }
1862 
1863 const struct file_operations proc_pid_numa_maps_operations = {
1864 	.open		= pid_numa_maps_open,
1865 	.read		= seq_read,
1866 	.llseek		= seq_lseek,
1867 	.release	= proc_map_release,
1868 };
1869 
1870 const struct file_operations proc_tid_numa_maps_operations = {
1871 	.open		= tid_numa_maps_open,
1872 	.read		= seq_read,
1873 	.llseek		= seq_lseek,
1874 	.release	= proc_map_release,
1875 };
1876 #endif /* CONFIG_NUMA */
1877