xref: /openbmc/linux/fs/proc/task_mmu.c (revision 1d05334d)
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/pagewalk.h>
3 #include <linux/vmacache.h>
4 #include <linux/hugetlb.h>
5 #include <linux/huge_mm.h>
6 #include <linux/mount.h>
7 #include <linux/seq_file.h>
8 #include <linux/highmem.h>
9 #include <linux/ptrace.h>
10 #include <linux/slab.h>
11 #include <linux/pagemap.h>
12 #include <linux/mempolicy.h>
13 #include <linux/rmap.h>
14 #include <linux/swap.h>
15 #include <linux/sched/mm.h>
16 #include <linux/swapops.h>
17 #include <linux/mmu_notifier.h>
18 #include <linux/page_idle.h>
19 #include <linux/shmem_fs.h>
20 #include <linux/uaccess.h>
21 #include <linux/pkeys.h>
22 
23 #include <asm/elf.h>
24 #include <asm/tlb.h>
25 #include <asm/tlbflush.h>
26 #include "internal.h"
27 
28 #define SEQ_PUT_DEC(str, val) \
29 		seq_put_decimal_ull_width(m, str, (val) << (PAGE_SHIFT-10), 8)
30 void task_mem(struct seq_file *m, struct mm_struct *mm)
31 {
32 	unsigned long text, lib, swap, anon, file, shmem;
33 	unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
34 
35 	anon = get_mm_counter(mm, MM_ANONPAGES);
36 	file = get_mm_counter(mm, MM_FILEPAGES);
37 	shmem = get_mm_counter(mm, MM_SHMEMPAGES);
38 
39 	/*
40 	 * Note: to minimize their overhead, mm maintains hiwater_vm and
41 	 * hiwater_rss only when about to *lower* total_vm or rss.  Any
42 	 * collector of these hiwater stats must therefore get total_vm
43 	 * and rss too, which will usually be the higher.  Barriers? not
44 	 * worth the effort, such snapshots can always be inconsistent.
45 	 */
46 	hiwater_vm = total_vm = mm->total_vm;
47 	if (hiwater_vm < mm->hiwater_vm)
48 		hiwater_vm = mm->hiwater_vm;
49 	hiwater_rss = total_rss = anon + file + shmem;
50 	if (hiwater_rss < mm->hiwater_rss)
51 		hiwater_rss = mm->hiwater_rss;
52 
53 	/* split executable areas between text and lib */
54 	text = PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK);
55 	text = min(text, mm->exec_vm << PAGE_SHIFT);
56 	lib = (mm->exec_vm << PAGE_SHIFT) - text;
57 
58 	swap = get_mm_counter(mm, MM_SWAPENTS);
59 	SEQ_PUT_DEC("VmPeak:\t", hiwater_vm);
60 	SEQ_PUT_DEC(" kB\nVmSize:\t", total_vm);
61 	SEQ_PUT_DEC(" kB\nVmLck:\t", mm->locked_vm);
62 	SEQ_PUT_DEC(" kB\nVmPin:\t", atomic64_read(&mm->pinned_vm));
63 	SEQ_PUT_DEC(" kB\nVmHWM:\t", hiwater_rss);
64 	SEQ_PUT_DEC(" kB\nVmRSS:\t", total_rss);
65 	SEQ_PUT_DEC(" kB\nRssAnon:\t", anon);
66 	SEQ_PUT_DEC(" kB\nRssFile:\t", file);
67 	SEQ_PUT_DEC(" kB\nRssShmem:\t", shmem);
68 	SEQ_PUT_DEC(" kB\nVmData:\t", mm->data_vm);
69 	SEQ_PUT_DEC(" kB\nVmStk:\t", mm->stack_vm);
70 	seq_put_decimal_ull_width(m,
71 		    " kB\nVmExe:\t", text >> 10, 8);
72 	seq_put_decimal_ull_width(m,
73 		    " kB\nVmLib:\t", lib >> 10, 8);
74 	seq_put_decimal_ull_width(m,
75 		    " kB\nVmPTE:\t", mm_pgtables_bytes(mm) >> 10, 8);
76 	SEQ_PUT_DEC(" kB\nVmSwap:\t", swap);
77 	seq_puts(m, " kB\n");
78 	hugetlb_report_usage(m, mm);
79 }
80 #undef SEQ_PUT_DEC
81 
82 unsigned long task_vsize(struct mm_struct *mm)
83 {
84 	return PAGE_SIZE * mm->total_vm;
85 }
86 
87 unsigned long task_statm(struct mm_struct *mm,
88 			 unsigned long *shared, unsigned long *text,
89 			 unsigned long *data, unsigned long *resident)
90 {
91 	*shared = get_mm_counter(mm, MM_FILEPAGES) +
92 			get_mm_counter(mm, MM_SHMEMPAGES);
93 	*text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
94 								>> PAGE_SHIFT;
95 	*data = mm->data_vm + mm->stack_vm;
96 	*resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
97 	return mm->total_vm;
98 }
99 
100 #ifdef CONFIG_NUMA
101 /*
102  * Save get_task_policy() for show_numa_map().
103  */
104 static void hold_task_mempolicy(struct proc_maps_private *priv)
105 {
106 	struct task_struct *task = priv->task;
107 
108 	task_lock(task);
109 	priv->task_mempolicy = get_task_policy(task);
110 	mpol_get(priv->task_mempolicy);
111 	task_unlock(task);
112 }
113 static void release_task_mempolicy(struct proc_maps_private *priv)
114 {
115 	mpol_put(priv->task_mempolicy);
116 }
117 #else
118 static void hold_task_mempolicy(struct proc_maps_private *priv)
119 {
120 }
121 static void release_task_mempolicy(struct proc_maps_private *priv)
122 {
123 }
124 #endif
125 
126 static void *m_start(struct seq_file *m, loff_t *ppos)
127 {
128 	struct proc_maps_private *priv = m->private;
129 	unsigned long last_addr = *ppos;
130 	struct mm_struct *mm;
131 	struct vm_area_struct *vma;
132 
133 	/* See m_next(). Zero at the start or after lseek. */
134 	if (last_addr == -1UL)
135 		return NULL;
136 
137 	priv->task = get_proc_task(priv->inode);
138 	if (!priv->task)
139 		return ERR_PTR(-ESRCH);
140 
141 	mm = priv->mm;
142 	if (!mm || !mmget_not_zero(mm)) {
143 		put_task_struct(priv->task);
144 		priv->task = NULL;
145 		return NULL;
146 	}
147 
148 	if (down_read_killable(&mm->mmap_sem)) {
149 		mmput(mm);
150 		put_task_struct(priv->task);
151 		priv->task = NULL;
152 		return ERR_PTR(-EINTR);
153 	}
154 
155 	hold_task_mempolicy(priv);
156 	priv->tail_vma = get_gate_vma(mm);
157 
158 	vma = find_vma(mm, last_addr);
159 	if (vma)
160 		return vma;
161 
162 	return priv->tail_vma;
163 }
164 
165 static void *m_next(struct seq_file *m, void *v, loff_t *ppos)
166 {
167 	struct proc_maps_private *priv = m->private;
168 	struct vm_area_struct *next, *vma = v;
169 
170 	if (vma == priv->tail_vma)
171 		next = NULL;
172 	else if (vma->vm_next)
173 		next = vma->vm_next;
174 	else
175 		next = priv->tail_vma;
176 
177 	*ppos = next ? next->vm_start : -1UL;
178 
179 	return next;
180 }
181 
182 static void m_stop(struct seq_file *m, void *v)
183 {
184 	struct proc_maps_private *priv = m->private;
185 	struct mm_struct *mm = priv->mm;
186 
187 	if (!priv->task)
188 		return;
189 
190 	release_task_mempolicy(priv);
191 	up_read(&mm->mmap_sem);
192 	mmput(mm);
193 	put_task_struct(priv->task);
194 	priv->task = NULL;
195 }
196 
197 static int proc_maps_open(struct inode *inode, struct file *file,
198 			const struct seq_operations *ops, int psize)
199 {
200 	struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
201 
202 	if (!priv)
203 		return -ENOMEM;
204 
205 	priv->inode = inode;
206 	priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
207 	if (IS_ERR(priv->mm)) {
208 		int err = PTR_ERR(priv->mm);
209 
210 		seq_release_private(inode, file);
211 		return err;
212 	}
213 
214 	return 0;
215 }
216 
217 static int proc_map_release(struct inode *inode, struct file *file)
218 {
219 	struct seq_file *seq = file->private_data;
220 	struct proc_maps_private *priv = seq->private;
221 
222 	if (priv->mm)
223 		mmdrop(priv->mm);
224 
225 	return seq_release_private(inode, file);
226 }
227 
228 static int do_maps_open(struct inode *inode, struct file *file,
229 			const struct seq_operations *ops)
230 {
231 	return proc_maps_open(inode, file, ops,
232 				sizeof(struct proc_maps_private));
233 }
234 
235 /*
236  * Indicate if the VMA is a stack for the given task; for
237  * /proc/PID/maps that is the stack of the main task.
238  */
239 static int is_stack(struct vm_area_struct *vma)
240 {
241 	/*
242 	 * We make no effort to guess what a given thread considers to be
243 	 * its "stack".  It's not even well-defined for programs written
244 	 * languages like Go.
245 	 */
246 	return vma->vm_start <= vma->vm_mm->start_stack &&
247 		vma->vm_end >= vma->vm_mm->start_stack;
248 }
249 
250 static void show_vma_header_prefix(struct seq_file *m,
251 				   unsigned long start, unsigned long end,
252 				   vm_flags_t flags, unsigned long long pgoff,
253 				   dev_t dev, unsigned long ino)
254 {
255 	seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
256 	seq_put_hex_ll(m, NULL, start, 8);
257 	seq_put_hex_ll(m, "-", end, 8);
258 	seq_putc(m, ' ');
259 	seq_putc(m, flags & VM_READ ? 'r' : '-');
260 	seq_putc(m, flags & VM_WRITE ? 'w' : '-');
261 	seq_putc(m, flags & VM_EXEC ? 'x' : '-');
262 	seq_putc(m, flags & VM_MAYSHARE ? 's' : 'p');
263 	seq_put_hex_ll(m, " ", pgoff, 8);
264 	seq_put_hex_ll(m, " ", MAJOR(dev), 2);
265 	seq_put_hex_ll(m, ":", MINOR(dev), 2);
266 	seq_put_decimal_ull(m, " ", ino);
267 	seq_putc(m, ' ');
268 }
269 
270 static void
271 show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
272 {
273 	struct mm_struct *mm = vma->vm_mm;
274 	struct file *file = vma->vm_file;
275 	vm_flags_t flags = vma->vm_flags;
276 	unsigned long ino = 0;
277 	unsigned long long pgoff = 0;
278 	unsigned long start, end;
279 	dev_t dev = 0;
280 	const char *name = NULL;
281 
282 	if (file) {
283 		struct inode *inode = file_inode(vma->vm_file);
284 		dev = inode->i_sb->s_dev;
285 		ino = inode->i_ino;
286 		pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
287 	}
288 
289 	start = vma->vm_start;
290 	end = vma->vm_end;
291 	show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino);
292 
293 	/*
294 	 * Print the dentry name for named mappings, and a
295 	 * special [heap] marker for the heap:
296 	 */
297 	if (file) {
298 		seq_pad(m, ' ');
299 		seq_file_path(m, file, "\n");
300 		goto done;
301 	}
302 
303 	if (vma->vm_ops && vma->vm_ops->name) {
304 		name = vma->vm_ops->name(vma);
305 		if (name)
306 			goto done;
307 	}
308 
309 	name = arch_vma_name(vma);
310 	if (!name) {
311 		if (!mm) {
312 			name = "[vdso]";
313 			goto done;
314 		}
315 
316 		if (vma->vm_start <= mm->brk &&
317 		    vma->vm_end >= mm->start_brk) {
318 			name = "[heap]";
319 			goto done;
320 		}
321 
322 		if (is_stack(vma))
323 			name = "[stack]";
324 	}
325 
326 done:
327 	if (name) {
328 		seq_pad(m, ' ');
329 		seq_puts(m, name);
330 	}
331 	seq_putc(m, '\n');
332 }
333 
334 static int show_map(struct seq_file *m, void *v)
335 {
336 	show_map_vma(m, v);
337 	return 0;
338 }
339 
340 static const struct seq_operations proc_pid_maps_op = {
341 	.start	= m_start,
342 	.next	= m_next,
343 	.stop	= m_stop,
344 	.show	= show_map
345 };
346 
347 static int pid_maps_open(struct inode *inode, struct file *file)
348 {
349 	return do_maps_open(inode, file, &proc_pid_maps_op);
350 }
351 
352 const struct file_operations proc_pid_maps_operations = {
353 	.open		= pid_maps_open,
354 	.read		= seq_read,
355 	.llseek		= seq_lseek,
356 	.release	= proc_map_release,
357 };
358 
359 /*
360  * Proportional Set Size(PSS): my share of RSS.
361  *
362  * PSS of a process is the count of pages it has in memory, where each
363  * page is divided by the number of processes sharing it.  So if a
364  * process has 1000 pages all to itself, and 1000 shared with one other
365  * process, its PSS will be 1500.
366  *
367  * To keep (accumulated) division errors low, we adopt a 64bit
368  * fixed-point pss counter to minimize division errors. So (pss >>
369  * PSS_SHIFT) would be the real byte count.
370  *
371  * A shift of 12 before division means (assuming 4K page size):
372  * 	- 1M 3-user-pages add up to 8KB errors;
373  * 	- supports mapcount up to 2^24, or 16M;
374  * 	- supports PSS up to 2^52 bytes, or 4PB.
375  */
376 #define PSS_SHIFT 12
377 
378 #ifdef CONFIG_PROC_PAGE_MONITOR
379 struct mem_size_stats {
380 	unsigned long resident;
381 	unsigned long shared_clean;
382 	unsigned long shared_dirty;
383 	unsigned long private_clean;
384 	unsigned long private_dirty;
385 	unsigned long referenced;
386 	unsigned long anonymous;
387 	unsigned long lazyfree;
388 	unsigned long anonymous_thp;
389 	unsigned long shmem_thp;
390 	unsigned long file_thp;
391 	unsigned long swap;
392 	unsigned long shared_hugetlb;
393 	unsigned long private_hugetlb;
394 	u64 pss;
395 	u64 pss_anon;
396 	u64 pss_file;
397 	u64 pss_shmem;
398 	u64 pss_locked;
399 	u64 swap_pss;
400 	bool check_shmem_swap;
401 };
402 
403 static void smaps_page_accumulate(struct mem_size_stats *mss,
404 		struct page *page, unsigned long size, unsigned long pss,
405 		bool dirty, bool locked, bool private)
406 {
407 	mss->pss += pss;
408 
409 	if (PageAnon(page))
410 		mss->pss_anon += pss;
411 	else if (PageSwapBacked(page))
412 		mss->pss_shmem += pss;
413 	else
414 		mss->pss_file += pss;
415 
416 	if (locked)
417 		mss->pss_locked += pss;
418 
419 	if (dirty || PageDirty(page)) {
420 		if (private)
421 			mss->private_dirty += size;
422 		else
423 			mss->shared_dirty += size;
424 	} else {
425 		if (private)
426 			mss->private_clean += size;
427 		else
428 			mss->shared_clean += size;
429 	}
430 }
431 
432 static void smaps_account(struct mem_size_stats *mss, struct page *page,
433 		bool compound, bool young, bool dirty, bool locked)
434 {
435 	int i, nr = compound ? compound_nr(page) : 1;
436 	unsigned long size = nr * PAGE_SIZE;
437 
438 	/*
439 	 * First accumulate quantities that depend only on |size| and the type
440 	 * of the compound page.
441 	 */
442 	if (PageAnon(page)) {
443 		mss->anonymous += size;
444 		if (!PageSwapBacked(page) && !dirty && !PageDirty(page))
445 			mss->lazyfree += size;
446 	}
447 
448 	mss->resident += size;
449 	/* Accumulate the size in pages that have been accessed. */
450 	if (young || page_is_young(page) || PageReferenced(page))
451 		mss->referenced += size;
452 
453 	/*
454 	 * Then accumulate quantities that may depend on sharing, or that may
455 	 * differ page-by-page.
456 	 *
457 	 * page_count(page) == 1 guarantees the page is mapped exactly once.
458 	 * If any subpage of the compound page mapped with PTE it would elevate
459 	 * page_count().
460 	 */
461 	if (page_count(page) == 1) {
462 		smaps_page_accumulate(mss, page, size, size << PSS_SHIFT, dirty,
463 			locked, true);
464 		return;
465 	}
466 	for (i = 0; i < nr; i++, page++) {
467 		int mapcount = page_mapcount(page);
468 		unsigned long pss = PAGE_SIZE << PSS_SHIFT;
469 		if (mapcount >= 2)
470 			pss /= mapcount;
471 		smaps_page_accumulate(mss, page, PAGE_SIZE, pss, dirty, locked,
472 				      mapcount < 2);
473 	}
474 }
475 
476 #ifdef CONFIG_SHMEM
477 static int smaps_pte_hole(unsigned long addr, unsigned long end,
478 			  __always_unused int depth, struct mm_walk *walk)
479 {
480 	struct mem_size_stats *mss = walk->private;
481 
482 	mss->swap += shmem_partial_swap_usage(
483 			walk->vma->vm_file->f_mapping, addr, end);
484 
485 	return 0;
486 }
487 #else
488 #define smaps_pte_hole		NULL
489 #endif /* CONFIG_SHMEM */
490 
491 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
492 		struct mm_walk *walk)
493 {
494 	struct mem_size_stats *mss = walk->private;
495 	struct vm_area_struct *vma = walk->vma;
496 	bool locked = !!(vma->vm_flags & VM_LOCKED);
497 	struct page *page = NULL;
498 
499 	if (pte_present(*pte)) {
500 		page = vm_normal_page(vma, addr, *pte);
501 	} else if (is_swap_pte(*pte)) {
502 		swp_entry_t swpent = pte_to_swp_entry(*pte);
503 
504 		if (!non_swap_entry(swpent)) {
505 			int mapcount;
506 
507 			mss->swap += PAGE_SIZE;
508 			mapcount = swp_swapcount(swpent);
509 			if (mapcount >= 2) {
510 				u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
511 
512 				do_div(pss_delta, mapcount);
513 				mss->swap_pss += pss_delta;
514 			} else {
515 				mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
516 			}
517 		} else if (is_migration_entry(swpent))
518 			page = migration_entry_to_page(swpent);
519 		else if (is_device_private_entry(swpent))
520 			page = device_private_entry_to_page(swpent);
521 	} else if (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap
522 							&& pte_none(*pte))) {
523 		page = find_get_entry(vma->vm_file->f_mapping,
524 						linear_page_index(vma, addr));
525 		if (!page)
526 			return;
527 
528 		if (xa_is_value(page))
529 			mss->swap += PAGE_SIZE;
530 		else
531 			put_page(page);
532 
533 		return;
534 	}
535 
536 	if (!page)
537 		return;
538 
539 	smaps_account(mss, page, false, pte_young(*pte), pte_dirty(*pte), locked);
540 }
541 
542 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
543 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
544 		struct mm_walk *walk)
545 {
546 	struct mem_size_stats *mss = walk->private;
547 	struct vm_area_struct *vma = walk->vma;
548 	bool locked = !!(vma->vm_flags & VM_LOCKED);
549 	struct page *page;
550 
551 	/* FOLL_DUMP will return -EFAULT on huge zero page */
552 	page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
553 	if (IS_ERR_OR_NULL(page))
554 		return;
555 	if (PageAnon(page))
556 		mss->anonymous_thp += HPAGE_PMD_SIZE;
557 	else if (PageSwapBacked(page))
558 		mss->shmem_thp += HPAGE_PMD_SIZE;
559 	else if (is_zone_device_page(page))
560 		/* pass */;
561 	else
562 		mss->file_thp += HPAGE_PMD_SIZE;
563 	smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd), locked);
564 }
565 #else
566 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
567 		struct mm_walk *walk)
568 {
569 }
570 #endif
571 
572 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
573 			   struct mm_walk *walk)
574 {
575 	struct vm_area_struct *vma = walk->vma;
576 	pte_t *pte;
577 	spinlock_t *ptl;
578 
579 	ptl = pmd_trans_huge_lock(pmd, vma);
580 	if (ptl) {
581 		if (pmd_present(*pmd))
582 			smaps_pmd_entry(pmd, addr, walk);
583 		spin_unlock(ptl);
584 		goto out;
585 	}
586 
587 	if (pmd_trans_unstable(pmd))
588 		goto out;
589 	/*
590 	 * The mmap_sem held all the way back in m_start() is what
591 	 * keeps khugepaged out of here and from collapsing things
592 	 * in here.
593 	 */
594 	pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
595 	for (; addr != end; pte++, addr += PAGE_SIZE)
596 		smaps_pte_entry(pte, addr, walk);
597 	pte_unmap_unlock(pte - 1, ptl);
598 out:
599 	cond_resched();
600 	return 0;
601 }
602 
603 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
604 {
605 	/*
606 	 * Don't forget to update Documentation/ on changes.
607 	 */
608 	static const char mnemonics[BITS_PER_LONG][2] = {
609 		/*
610 		 * In case if we meet a flag we don't know about.
611 		 */
612 		[0 ... (BITS_PER_LONG-1)] = "??",
613 
614 		[ilog2(VM_READ)]	= "rd",
615 		[ilog2(VM_WRITE)]	= "wr",
616 		[ilog2(VM_EXEC)]	= "ex",
617 		[ilog2(VM_SHARED)]	= "sh",
618 		[ilog2(VM_MAYREAD)]	= "mr",
619 		[ilog2(VM_MAYWRITE)]	= "mw",
620 		[ilog2(VM_MAYEXEC)]	= "me",
621 		[ilog2(VM_MAYSHARE)]	= "ms",
622 		[ilog2(VM_GROWSDOWN)]	= "gd",
623 		[ilog2(VM_PFNMAP)]	= "pf",
624 		[ilog2(VM_DENYWRITE)]	= "dw",
625 #ifdef CONFIG_X86_INTEL_MPX
626 		[ilog2(VM_MPX)]		= "mp",
627 #endif
628 		[ilog2(VM_LOCKED)]	= "lo",
629 		[ilog2(VM_IO)]		= "io",
630 		[ilog2(VM_SEQ_READ)]	= "sr",
631 		[ilog2(VM_RAND_READ)]	= "rr",
632 		[ilog2(VM_DONTCOPY)]	= "dc",
633 		[ilog2(VM_DONTEXPAND)]	= "de",
634 		[ilog2(VM_ACCOUNT)]	= "ac",
635 		[ilog2(VM_NORESERVE)]	= "nr",
636 		[ilog2(VM_HUGETLB)]	= "ht",
637 		[ilog2(VM_SYNC)]	= "sf",
638 		[ilog2(VM_ARCH_1)]	= "ar",
639 		[ilog2(VM_WIPEONFORK)]	= "wf",
640 		[ilog2(VM_DONTDUMP)]	= "dd",
641 #ifdef CONFIG_MEM_SOFT_DIRTY
642 		[ilog2(VM_SOFTDIRTY)]	= "sd",
643 #endif
644 		[ilog2(VM_MIXEDMAP)]	= "mm",
645 		[ilog2(VM_HUGEPAGE)]	= "hg",
646 		[ilog2(VM_NOHUGEPAGE)]	= "nh",
647 		[ilog2(VM_MERGEABLE)]	= "mg",
648 		[ilog2(VM_UFFD_MISSING)]= "um",
649 		[ilog2(VM_UFFD_WP)]	= "uw",
650 #ifdef CONFIG_ARCH_HAS_PKEYS
651 		/* These come out via ProtectionKey: */
652 		[ilog2(VM_PKEY_BIT0)]	= "",
653 		[ilog2(VM_PKEY_BIT1)]	= "",
654 		[ilog2(VM_PKEY_BIT2)]	= "",
655 		[ilog2(VM_PKEY_BIT3)]	= "",
656 #if VM_PKEY_BIT4
657 		[ilog2(VM_PKEY_BIT4)]	= "",
658 #endif
659 #endif /* CONFIG_ARCH_HAS_PKEYS */
660 	};
661 	size_t i;
662 
663 	seq_puts(m, "VmFlags: ");
664 	for (i = 0; i < BITS_PER_LONG; i++) {
665 		if (!mnemonics[i][0])
666 			continue;
667 		if (vma->vm_flags & (1UL << i)) {
668 			seq_putc(m, mnemonics[i][0]);
669 			seq_putc(m, mnemonics[i][1]);
670 			seq_putc(m, ' ');
671 		}
672 	}
673 	seq_putc(m, '\n');
674 }
675 
676 #ifdef CONFIG_HUGETLB_PAGE
677 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
678 				 unsigned long addr, unsigned long end,
679 				 struct mm_walk *walk)
680 {
681 	struct mem_size_stats *mss = walk->private;
682 	struct vm_area_struct *vma = walk->vma;
683 	struct page *page = NULL;
684 
685 	if (pte_present(*pte)) {
686 		page = vm_normal_page(vma, addr, *pte);
687 	} else if (is_swap_pte(*pte)) {
688 		swp_entry_t swpent = pte_to_swp_entry(*pte);
689 
690 		if (is_migration_entry(swpent))
691 			page = migration_entry_to_page(swpent);
692 		else if (is_device_private_entry(swpent))
693 			page = device_private_entry_to_page(swpent);
694 	}
695 	if (page) {
696 		int mapcount = page_mapcount(page);
697 
698 		if (mapcount >= 2)
699 			mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
700 		else
701 			mss->private_hugetlb += huge_page_size(hstate_vma(vma));
702 	}
703 	return 0;
704 }
705 #else
706 #define smaps_hugetlb_range	NULL
707 #endif /* HUGETLB_PAGE */
708 
709 static const struct mm_walk_ops smaps_walk_ops = {
710 	.pmd_entry		= smaps_pte_range,
711 	.hugetlb_entry		= smaps_hugetlb_range,
712 };
713 
714 static const struct mm_walk_ops smaps_shmem_walk_ops = {
715 	.pmd_entry		= smaps_pte_range,
716 	.hugetlb_entry		= smaps_hugetlb_range,
717 	.pte_hole		= smaps_pte_hole,
718 };
719 
720 static void smap_gather_stats(struct vm_area_struct *vma,
721 			     struct mem_size_stats *mss)
722 {
723 #ifdef CONFIG_SHMEM
724 	/* In case of smaps_rollup, reset the value from previous vma */
725 	mss->check_shmem_swap = false;
726 	if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
727 		/*
728 		 * For shared or readonly shmem mappings we know that all
729 		 * swapped out pages belong to the shmem object, and we can
730 		 * obtain the swap value much more efficiently. For private
731 		 * writable mappings, we might have COW pages that are
732 		 * not affected by the parent swapped out pages of the shmem
733 		 * object, so we have to distinguish them during the page walk.
734 		 * Unless we know that the shmem object (or the part mapped by
735 		 * our VMA) has no swapped out pages at all.
736 		 */
737 		unsigned long shmem_swapped = shmem_swap_usage(vma);
738 
739 		if (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
740 					!(vma->vm_flags & VM_WRITE)) {
741 			mss->swap += shmem_swapped;
742 		} else {
743 			mss->check_shmem_swap = true;
744 			walk_page_vma(vma, &smaps_shmem_walk_ops, mss);
745 			return;
746 		}
747 	}
748 #endif
749 	/* mmap_sem is held in m_start */
750 	walk_page_vma(vma, &smaps_walk_ops, mss);
751 }
752 
753 #define SEQ_PUT_DEC(str, val) \
754 		seq_put_decimal_ull_width(m, str, (val) >> 10, 8)
755 
756 /* Show the contents common for smaps and smaps_rollup */
757 static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss,
758 	bool rollup_mode)
759 {
760 	SEQ_PUT_DEC("Rss:            ", mss->resident);
761 	SEQ_PUT_DEC(" kB\nPss:            ", mss->pss >> PSS_SHIFT);
762 	if (rollup_mode) {
763 		/*
764 		 * These are meaningful only for smaps_rollup, otherwise two of
765 		 * them are zero, and the other one is the same as Pss.
766 		 */
767 		SEQ_PUT_DEC(" kB\nPss_Anon:       ",
768 			mss->pss_anon >> PSS_SHIFT);
769 		SEQ_PUT_DEC(" kB\nPss_File:       ",
770 			mss->pss_file >> PSS_SHIFT);
771 		SEQ_PUT_DEC(" kB\nPss_Shmem:      ",
772 			mss->pss_shmem >> PSS_SHIFT);
773 	}
774 	SEQ_PUT_DEC(" kB\nShared_Clean:   ", mss->shared_clean);
775 	SEQ_PUT_DEC(" kB\nShared_Dirty:   ", mss->shared_dirty);
776 	SEQ_PUT_DEC(" kB\nPrivate_Clean:  ", mss->private_clean);
777 	SEQ_PUT_DEC(" kB\nPrivate_Dirty:  ", mss->private_dirty);
778 	SEQ_PUT_DEC(" kB\nReferenced:     ", mss->referenced);
779 	SEQ_PUT_DEC(" kB\nAnonymous:      ", mss->anonymous);
780 	SEQ_PUT_DEC(" kB\nLazyFree:       ", mss->lazyfree);
781 	SEQ_PUT_DEC(" kB\nAnonHugePages:  ", mss->anonymous_thp);
782 	SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp);
783 	SEQ_PUT_DEC(" kB\nFilePmdMapped: ", mss->file_thp);
784 	SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb);
785 	seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ",
786 				  mss->private_hugetlb >> 10, 7);
787 	SEQ_PUT_DEC(" kB\nSwap:           ", mss->swap);
788 	SEQ_PUT_DEC(" kB\nSwapPss:        ",
789 					mss->swap_pss >> PSS_SHIFT);
790 	SEQ_PUT_DEC(" kB\nLocked:         ",
791 					mss->pss_locked >> PSS_SHIFT);
792 	seq_puts(m, " kB\n");
793 }
794 
795 static int show_smap(struct seq_file *m, void *v)
796 {
797 	struct vm_area_struct *vma = v;
798 	struct mem_size_stats mss;
799 
800 	memset(&mss, 0, sizeof(mss));
801 
802 	smap_gather_stats(vma, &mss);
803 
804 	show_map_vma(m, vma);
805 
806 	SEQ_PUT_DEC("Size:           ", vma->vm_end - vma->vm_start);
807 	SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma));
808 	SEQ_PUT_DEC(" kB\nMMUPageSize:    ", vma_mmu_pagesize(vma));
809 	seq_puts(m, " kB\n");
810 
811 	__show_smap(m, &mss, false);
812 
813 	seq_printf(m, "THPeligible:		%d\n",
814 		   transparent_hugepage_enabled(vma));
815 
816 	if (arch_pkeys_enabled())
817 		seq_printf(m, "ProtectionKey:  %8u\n", vma_pkey(vma));
818 	show_smap_vma_flags(m, vma);
819 
820 	return 0;
821 }
822 
823 static int show_smaps_rollup(struct seq_file *m, void *v)
824 {
825 	struct proc_maps_private *priv = m->private;
826 	struct mem_size_stats mss;
827 	struct mm_struct *mm;
828 	struct vm_area_struct *vma;
829 	unsigned long last_vma_end = 0;
830 	int ret = 0;
831 
832 	priv->task = get_proc_task(priv->inode);
833 	if (!priv->task)
834 		return -ESRCH;
835 
836 	mm = priv->mm;
837 	if (!mm || !mmget_not_zero(mm)) {
838 		ret = -ESRCH;
839 		goto out_put_task;
840 	}
841 
842 	memset(&mss, 0, sizeof(mss));
843 
844 	ret = down_read_killable(&mm->mmap_sem);
845 	if (ret)
846 		goto out_put_mm;
847 
848 	hold_task_mempolicy(priv);
849 
850 	for (vma = priv->mm->mmap; vma; vma = vma->vm_next) {
851 		smap_gather_stats(vma, &mss);
852 		last_vma_end = vma->vm_end;
853 	}
854 
855 	show_vma_header_prefix(m, priv->mm->mmap->vm_start,
856 			       last_vma_end, 0, 0, 0, 0);
857 	seq_pad(m, ' ');
858 	seq_puts(m, "[rollup]\n");
859 
860 	__show_smap(m, &mss, true);
861 
862 	release_task_mempolicy(priv);
863 	up_read(&mm->mmap_sem);
864 
865 out_put_mm:
866 	mmput(mm);
867 out_put_task:
868 	put_task_struct(priv->task);
869 	priv->task = NULL;
870 
871 	return ret;
872 }
873 #undef SEQ_PUT_DEC
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_smap
880 };
881 
882 static int pid_smaps_open(struct inode *inode, struct file *file)
883 {
884 	return do_maps_open(inode, file, &proc_pid_smaps_op);
885 }
886 
887 static int smaps_rollup_open(struct inode *inode, struct file *file)
888 {
889 	int ret;
890 	struct proc_maps_private *priv;
891 
892 	priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT);
893 	if (!priv)
894 		return -ENOMEM;
895 
896 	ret = single_open(file, show_smaps_rollup, priv);
897 	if (ret)
898 		goto out_free;
899 
900 	priv->inode = inode;
901 	priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
902 	if (IS_ERR(priv->mm)) {
903 		ret = PTR_ERR(priv->mm);
904 
905 		single_release(inode, file);
906 		goto out_free;
907 	}
908 
909 	return 0;
910 
911 out_free:
912 	kfree(priv);
913 	return ret;
914 }
915 
916 static int smaps_rollup_release(struct inode *inode, struct file *file)
917 {
918 	struct seq_file *seq = file->private_data;
919 	struct proc_maps_private *priv = seq->private;
920 
921 	if (priv->mm)
922 		mmdrop(priv->mm);
923 
924 	kfree(priv);
925 	return single_release(inode, file);
926 }
927 
928 const struct file_operations proc_pid_smaps_operations = {
929 	.open		= pid_smaps_open,
930 	.read		= seq_read,
931 	.llseek		= seq_lseek,
932 	.release	= proc_map_release,
933 };
934 
935 const struct file_operations proc_pid_smaps_rollup_operations = {
936 	.open		= smaps_rollup_open,
937 	.read		= seq_read,
938 	.llseek		= seq_lseek,
939 	.release	= smaps_rollup_release,
940 };
941 
942 enum clear_refs_types {
943 	CLEAR_REFS_ALL = 1,
944 	CLEAR_REFS_ANON,
945 	CLEAR_REFS_MAPPED,
946 	CLEAR_REFS_SOFT_DIRTY,
947 	CLEAR_REFS_MM_HIWATER_RSS,
948 	CLEAR_REFS_LAST,
949 };
950 
951 struct clear_refs_private {
952 	enum clear_refs_types type;
953 };
954 
955 #ifdef CONFIG_MEM_SOFT_DIRTY
956 static inline void clear_soft_dirty(struct vm_area_struct *vma,
957 		unsigned long addr, pte_t *pte)
958 {
959 	/*
960 	 * The soft-dirty tracker uses #PF-s to catch writes
961 	 * to pages, so write-protect the pte as well. See the
962 	 * Documentation/admin-guide/mm/soft-dirty.rst for full description
963 	 * of how soft-dirty works.
964 	 */
965 	pte_t ptent = *pte;
966 
967 	if (pte_present(ptent)) {
968 		pte_t old_pte;
969 
970 		old_pte = ptep_modify_prot_start(vma, addr, pte);
971 		ptent = pte_wrprotect(old_pte);
972 		ptent = pte_clear_soft_dirty(ptent);
973 		ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent);
974 	} else if (is_swap_pte(ptent)) {
975 		ptent = pte_swp_clear_soft_dirty(ptent);
976 		set_pte_at(vma->vm_mm, addr, pte, ptent);
977 	}
978 }
979 #else
980 static inline void clear_soft_dirty(struct vm_area_struct *vma,
981 		unsigned long addr, pte_t *pte)
982 {
983 }
984 #endif
985 
986 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
987 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
988 		unsigned long addr, pmd_t *pmdp)
989 {
990 	pmd_t old, pmd = *pmdp;
991 
992 	if (pmd_present(pmd)) {
993 		/* See comment in change_huge_pmd() */
994 		old = pmdp_invalidate(vma, addr, pmdp);
995 		if (pmd_dirty(old))
996 			pmd = pmd_mkdirty(pmd);
997 		if (pmd_young(old))
998 			pmd = pmd_mkyoung(pmd);
999 
1000 		pmd = pmd_wrprotect(pmd);
1001 		pmd = pmd_clear_soft_dirty(pmd);
1002 
1003 		set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1004 	} else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1005 		pmd = pmd_swp_clear_soft_dirty(pmd);
1006 		set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1007 	}
1008 }
1009 #else
1010 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1011 		unsigned long addr, pmd_t *pmdp)
1012 {
1013 }
1014 #endif
1015 
1016 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
1017 				unsigned long end, struct mm_walk *walk)
1018 {
1019 	struct clear_refs_private *cp = walk->private;
1020 	struct vm_area_struct *vma = walk->vma;
1021 	pte_t *pte, ptent;
1022 	spinlock_t *ptl;
1023 	struct page *page;
1024 
1025 	ptl = pmd_trans_huge_lock(pmd, vma);
1026 	if (ptl) {
1027 		if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1028 			clear_soft_dirty_pmd(vma, addr, pmd);
1029 			goto out;
1030 		}
1031 
1032 		if (!pmd_present(*pmd))
1033 			goto out;
1034 
1035 		page = pmd_page(*pmd);
1036 
1037 		/* Clear accessed and referenced bits. */
1038 		pmdp_test_and_clear_young(vma, addr, pmd);
1039 		test_and_clear_page_young(page);
1040 		ClearPageReferenced(page);
1041 out:
1042 		spin_unlock(ptl);
1043 		return 0;
1044 	}
1045 
1046 	if (pmd_trans_unstable(pmd))
1047 		return 0;
1048 
1049 	pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1050 	for (; addr != end; pte++, addr += PAGE_SIZE) {
1051 		ptent = *pte;
1052 
1053 		if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1054 			clear_soft_dirty(vma, addr, pte);
1055 			continue;
1056 		}
1057 
1058 		if (!pte_present(ptent))
1059 			continue;
1060 
1061 		page = vm_normal_page(vma, addr, ptent);
1062 		if (!page)
1063 			continue;
1064 
1065 		/* Clear accessed and referenced bits. */
1066 		ptep_test_and_clear_young(vma, addr, pte);
1067 		test_and_clear_page_young(page);
1068 		ClearPageReferenced(page);
1069 	}
1070 	pte_unmap_unlock(pte - 1, ptl);
1071 	cond_resched();
1072 	return 0;
1073 }
1074 
1075 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1076 				struct mm_walk *walk)
1077 {
1078 	struct clear_refs_private *cp = walk->private;
1079 	struct vm_area_struct *vma = walk->vma;
1080 
1081 	if (vma->vm_flags & VM_PFNMAP)
1082 		return 1;
1083 
1084 	/*
1085 	 * Writing 1 to /proc/pid/clear_refs affects all pages.
1086 	 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1087 	 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1088 	 * Writing 4 to /proc/pid/clear_refs affects all pages.
1089 	 */
1090 	if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1091 		return 1;
1092 	if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1093 		return 1;
1094 	return 0;
1095 }
1096 
1097 static const struct mm_walk_ops clear_refs_walk_ops = {
1098 	.pmd_entry		= clear_refs_pte_range,
1099 	.test_walk		= clear_refs_test_walk,
1100 };
1101 
1102 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1103 				size_t count, loff_t *ppos)
1104 {
1105 	struct task_struct *task;
1106 	char buffer[PROC_NUMBUF];
1107 	struct mm_struct *mm;
1108 	struct vm_area_struct *vma;
1109 	enum clear_refs_types type;
1110 	struct mmu_gather tlb;
1111 	int itype;
1112 	int rv;
1113 
1114 	memset(buffer, 0, sizeof(buffer));
1115 	if (count > sizeof(buffer) - 1)
1116 		count = sizeof(buffer) - 1;
1117 	if (copy_from_user(buffer, buf, count))
1118 		return -EFAULT;
1119 	rv = kstrtoint(strstrip(buffer), 10, &itype);
1120 	if (rv < 0)
1121 		return rv;
1122 	type = (enum clear_refs_types)itype;
1123 	if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1124 		return -EINVAL;
1125 
1126 	task = get_proc_task(file_inode(file));
1127 	if (!task)
1128 		return -ESRCH;
1129 	mm = get_task_mm(task);
1130 	if (mm) {
1131 		struct mmu_notifier_range range;
1132 		struct clear_refs_private cp = {
1133 			.type = type,
1134 		};
1135 
1136 		if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1137 			if (down_write_killable(&mm->mmap_sem)) {
1138 				count = -EINTR;
1139 				goto out_mm;
1140 			}
1141 
1142 			/*
1143 			 * Writing 5 to /proc/pid/clear_refs resets the peak
1144 			 * resident set size to this mm's current rss value.
1145 			 */
1146 			reset_mm_hiwater_rss(mm);
1147 			up_write(&mm->mmap_sem);
1148 			goto out_mm;
1149 		}
1150 
1151 		if (down_read_killable(&mm->mmap_sem)) {
1152 			count = -EINTR;
1153 			goto out_mm;
1154 		}
1155 		tlb_gather_mmu(&tlb, mm, 0, -1);
1156 		if (type == CLEAR_REFS_SOFT_DIRTY) {
1157 			for (vma = mm->mmap; vma; vma = vma->vm_next) {
1158 				if (!(vma->vm_flags & VM_SOFTDIRTY))
1159 					continue;
1160 				up_read(&mm->mmap_sem);
1161 				if (down_write_killable(&mm->mmap_sem)) {
1162 					count = -EINTR;
1163 					goto out_mm;
1164 				}
1165 				/*
1166 				 * Avoid to modify vma->vm_flags
1167 				 * without locked ops while the
1168 				 * coredump reads the vm_flags.
1169 				 */
1170 				if (!mmget_still_valid(mm)) {
1171 					/*
1172 					 * Silently return "count"
1173 					 * like if get_task_mm()
1174 					 * failed. FIXME: should this
1175 					 * function have returned
1176 					 * -ESRCH if get_task_mm()
1177 					 * failed like if
1178 					 * get_proc_task() fails?
1179 					 */
1180 					up_write(&mm->mmap_sem);
1181 					goto out_mm;
1182 				}
1183 				for (vma = mm->mmap; vma; vma = vma->vm_next) {
1184 					vma->vm_flags &= ~VM_SOFTDIRTY;
1185 					vma_set_page_prot(vma);
1186 				}
1187 				downgrade_write(&mm->mmap_sem);
1188 				break;
1189 			}
1190 
1191 			mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY,
1192 						0, NULL, mm, 0, -1UL);
1193 			mmu_notifier_invalidate_range_start(&range);
1194 		}
1195 		walk_page_range(mm, 0, mm->highest_vm_end, &clear_refs_walk_ops,
1196 				&cp);
1197 		if (type == CLEAR_REFS_SOFT_DIRTY)
1198 			mmu_notifier_invalidate_range_end(&range);
1199 		tlb_finish_mmu(&tlb, 0, -1);
1200 		up_read(&mm->mmap_sem);
1201 out_mm:
1202 		mmput(mm);
1203 	}
1204 	put_task_struct(task);
1205 
1206 	return count;
1207 }
1208 
1209 const struct file_operations proc_clear_refs_operations = {
1210 	.write		= clear_refs_write,
1211 	.llseek		= noop_llseek,
1212 };
1213 
1214 typedef struct {
1215 	u64 pme;
1216 } pagemap_entry_t;
1217 
1218 struct pagemapread {
1219 	int pos, len;		/* units: PM_ENTRY_BYTES, not bytes */
1220 	pagemap_entry_t *buffer;
1221 	bool show_pfn;
1222 };
1223 
1224 #define PAGEMAP_WALK_SIZE	(PMD_SIZE)
1225 #define PAGEMAP_WALK_MASK	(PMD_MASK)
1226 
1227 #define PM_ENTRY_BYTES		sizeof(pagemap_entry_t)
1228 #define PM_PFRAME_BITS		55
1229 #define PM_PFRAME_MASK		GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1230 #define PM_SOFT_DIRTY		BIT_ULL(55)
1231 #define PM_MMAP_EXCLUSIVE	BIT_ULL(56)
1232 #define PM_FILE			BIT_ULL(61)
1233 #define PM_SWAP			BIT_ULL(62)
1234 #define PM_PRESENT		BIT_ULL(63)
1235 
1236 #define PM_END_OF_BUFFER    1
1237 
1238 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1239 {
1240 	return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1241 }
1242 
1243 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1244 			  struct pagemapread *pm)
1245 {
1246 	pm->buffer[pm->pos++] = *pme;
1247 	if (pm->pos >= pm->len)
1248 		return PM_END_OF_BUFFER;
1249 	return 0;
1250 }
1251 
1252 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1253 			    __always_unused int depth, struct mm_walk *walk)
1254 {
1255 	struct pagemapread *pm = walk->private;
1256 	unsigned long addr = start;
1257 	int err = 0;
1258 
1259 	while (addr < end) {
1260 		struct vm_area_struct *vma = find_vma(walk->mm, addr);
1261 		pagemap_entry_t pme = make_pme(0, 0);
1262 		/* End of address space hole, which we mark as non-present. */
1263 		unsigned long hole_end;
1264 
1265 		if (vma)
1266 			hole_end = min(end, vma->vm_start);
1267 		else
1268 			hole_end = end;
1269 
1270 		for (; addr < hole_end; addr += PAGE_SIZE) {
1271 			err = add_to_pagemap(addr, &pme, pm);
1272 			if (err)
1273 				goto out;
1274 		}
1275 
1276 		if (!vma)
1277 			break;
1278 
1279 		/* Addresses in the VMA. */
1280 		if (vma->vm_flags & VM_SOFTDIRTY)
1281 			pme = make_pme(0, PM_SOFT_DIRTY);
1282 		for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1283 			err = add_to_pagemap(addr, &pme, pm);
1284 			if (err)
1285 				goto out;
1286 		}
1287 	}
1288 out:
1289 	return err;
1290 }
1291 
1292 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1293 		struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1294 {
1295 	u64 frame = 0, flags = 0;
1296 	struct page *page = NULL;
1297 
1298 	if (pte_present(pte)) {
1299 		if (pm->show_pfn)
1300 			frame = pte_pfn(pte);
1301 		flags |= PM_PRESENT;
1302 		page = vm_normal_page(vma, addr, pte);
1303 		if (pte_soft_dirty(pte))
1304 			flags |= PM_SOFT_DIRTY;
1305 	} else if (is_swap_pte(pte)) {
1306 		swp_entry_t entry;
1307 		if (pte_swp_soft_dirty(pte))
1308 			flags |= PM_SOFT_DIRTY;
1309 		entry = pte_to_swp_entry(pte);
1310 		if (pm->show_pfn)
1311 			frame = swp_type(entry) |
1312 				(swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1313 		flags |= PM_SWAP;
1314 		if (is_migration_entry(entry))
1315 			page = migration_entry_to_page(entry);
1316 
1317 		if (is_device_private_entry(entry))
1318 			page = device_private_entry_to_page(entry);
1319 	}
1320 
1321 	if (page && !PageAnon(page))
1322 		flags |= PM_FILE;
1323 	if (page && page_mapcount(page) == 1)
1324 		flags |= PM_MMAP_EXCLUSIVE;
1325 	if (vma->vm_flags & VM_SOFTDIRTY)
1326 		flags |= PM_SOFT_DIRTY;
1327 
1328 	return make_pme(frame, flags);
1329 }
1330 
1331 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1332 			     struct mm_walk *walk)
1333 {
1334 	struct vm_area_struct *vma = walk->vma;
1335 	struct pagemapread *pm = walk->private;
1336 	spinlock_t *ptl;
1337 	pte_t *pte, *orig_pte;
1338 	int err = 0;
1339 
1340 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1341 	ptl = pmd_trans_huge_lock(pmdp, vma);
1342 	if (ptl) {
1343 		u64 flags = 0, frame = 0;
1344 		pmd_t pmd = *pmdp;
1345 		struct page *page = NULL;
1346 
1347 		if (vma->vm_flags & VM_SOFTDIRTY)
1348 			flags |= PM_SOFT_DIRTY;
1349 
1350 		if (pmd_present(pmd)) {
1351 			page = pmd_page(pmd);
1352 
1353 			flags |= PM_PRESENT;
1354 			if (pmd_soft_dirty(pmd))
1355 				flags |= PM_SOFT_DIRTY;
1356 			if (pm->show_pfn)
1357 				frame = pmd_pfn(pmd) +
1358 					((addr & ~PMD_MASK) >> PAGE_SHIFT);
1359 		}
1360 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1361 		else if (is_swap_pmd(pmd)) {
1362 			swp_entry_t entry = pmd_to_swp_entry(pmd);
1363 			unsigned long offset;
1364 
1365 			if (pm->show_pfn) {
1366 				offset = swp_offset(entry) +
1367 					((addr & ~PMD_MASK) >> PAGE_SHIFT);
1368 				frame = swp_type(entry) |
1369 					(offset << MAX_SWAPFILES_SHIFT);
1370 			}
1371 			flags |= PM_SWAP;
1372 			if (pmd_swp_soft_dirty(pmd))
1373 				flags |= PM_SOFT_DIRTY;
1374 			VM_BUG_ON(!is_pmd_migration_entry(pmd));
1375 			page = migration_entry_to_page(entry);
1376 		}
1377 #endif
1378 
1379 		if (page && page_mapcount(page) == 1)
1380 			flags |= PM_MMAP_EXCLUSIVE;
1381 
1382 		for (; addr != end; addr += PAGE_SIZE) {
1383 			pagemap_entry_t pme = make_pme(frame, flags);
1384 
1385 			err = add_to_pagemap(addr, &pme, pm);
1386 			if (err)
1387 				break;
1388 			if (pm->show_pfn) {
1389 				if (flags & PM_PRESENT)
1390 					frame++;
1391 				else if (flags & PM_SWAP)
1392 					frame += (1 << MAX_SWAPFILES_SHIFT);
1393 			}
1394 		}
1395 		spin_unlock(ptl);
1396 		return err;
1397 	}
1398 
1399 	if (pmd_trans_unstable(pmdp))
1400 		return 0;
1401 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1402 
1403 	/*
1404 	 * We can assume that @vma always points to a valid one and @end never
1405 	 * goes beyond vma->vm_end.
1406 	 */
1407 	orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1408 	for (; addr < end; pte++, addr += PAGE_SIZE) {
1409 		pagemap_entry_t pme;
1410 
1411 		pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1412 		err = add_to_pagemap(addr, &pme, pm);
1413 		if (err)
1414 			break;
1415 	}
1416 	pte_unmap_unlock(orig_pte, ptl);
1417 
1418 	cond_resched();
1419 
1420 	return err;
1421 }
1422 
1423 #ifdef CONFIG_HUGETLB_PAGE
1424 /* This function walks within one hugetlb entry in the single call */
1425 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1426 				 unsigned long addr, unsigned long end,
1427 				 struct mm_walk *walk)
1428 {
1429 	struct pagemapread *pm = walk->private;
1430 	struct vm_area_struct *vma = walk->vma;
1431 	u64 flags = 0, frame = 0;
1432 	int err = 0;
1433 	pte_t pte;
1434 
1435 	if (vma->vm_flags & VM_SOFTDIRTY)
1436 		flags |= PM_SOFT_DIRTY;
1437 
1438 	pte = huge_ptep_get(ptep);
1439 	if (pte_present(pte)) {
1440 		struct page *page = pte_page(pte);
1441 
1442 		if (!PageAnon(page))
1443 			flags |= PM_FILE;
1444 
1445 		if (page_mapcount(page) == 1)
1446 			flags |= PM_MMAP_EXCLUSIVE;
1447 
1448 		flags |= PM_PRESENT;
1449 		if (pm->show_pfn)
1450 			frame = pte_pfn(pte) +
1451 				((addr & ~hmask) >> PAGE_SHIFT);
1452 	}
1453 
1454 	for (; addr != end; addr += PAGE_SIZE) {
1455 		pagemap_entry_t pme = make_pme(frame, flags);
1456 
1457 		err = add_to_pagemap(addr, &pme, pm);
1458 		if (err)
1459 			return err;
1460 		if (pm->show_pfn && (flags & PM_PRESENT))
1461 			frame++;
1462 	}
1463 
1464 	cond_resched();
1465 
1466 	return err;
1467 }
1468 #else
1469 #define pagemap_hugetlb_range	NULL
1470 #endif /* HUGETLB_PAGE */
1471 
1472 static const struct mm_walk_ops pagemap_ops = {
1473 	.pmd_entry	= pagemap_pmd_range,
1474 	.pte_hole	= pagemap_pte_hole,
1475 	.hugetlb_entry	= pagemap_hugetlb_range,
1476 };
1477 
1478 /*
1479  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1480  *
1481  * For each page in the address space, this file contains one 64-bit entry
1482  * consisting of the following:
1483  *
1484  * Bits 0-54  page frame number (PFN) if present
1485  * Bits 0-4   swap type if swapped
1486  * Bits 5-54  swap offset if swapped
1487  * Bit  55    pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst)
1488  * Bit  56    page exclusively mapped
1489  * Bits 57-60 zero
1490  * Bit  61    page is file-page or shared-anon
1491  * Bit  62    page swapped
1492  * Bit  63    page present
1493  *
1494  * If the page is not present but in swap, then the PFN contains an
1495  * encoding of the swap file number and the page's offset into the
1496  * swap. Unmapped pages return a null PFN. This allows determining
1497  * precisely which pages are mapped (or in swap) and comparing mapped
1498  * pages between processes.
1499  *
1500  * Efficient users of this interface will use /proc/pid/maps to
1501  * determine which areas of memory are actually mapped and llseek to
1502  * skip over unmapped regions.
1503  */
1504 static ssize_t pagemap_read(struct file *file, char __user *buf,
1505 			    size_t count, loff_t *ppos)
1506 {
1507 	struct mm_struct *mm = file->private_data;
1508 	struct pagemapread pm;
1509 	unsigned long src;
1510 	unsigned long svpfn;
1511 	unsigned long start_vaddr;
1512 	unsigned long end_vaddr;
1513 	int ret = 0, copied = 0;
1514 
1515 	if (!mm || !mmget_not_zero(mm))
1516 		goto out;
1517 
1518 	ret = -EINVAL;
1519 	/* file position must be aligned */
1520 	if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1521 		goto out_mm;
1522 
1523 	ret = 0;
1524 	if (!count)
1525 		goto out_mm;
1526 
1527 	/* do not disclose physical addresses: attack vector */
1528 	pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1529 
1530 	pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1531 	pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL);
1532 	ret = -ENOMEM;
1533 	if (!pm.buffer)
1534 		goto out_mm;
1535 
1536 	src = *ppos;
1537 	svpfn = src / PM_ENTRY_BYTES;
1538 	start_vaddr = svpfn << PAGE_SHIFT;
1539 	end_vaddr = mm->task_size;
1540 
1541 	/* watch out for wraparound */
1542 	if (svpfn > mm->task_size >> PAGE_SHIFT)
1543 		start_vaddr = end_vaddr;
1544 
1545 	/*
1546 	 * The odds are that this will stop walking way
1547 	 * before end_vaddr, because the length of the
1548 	 * user buffer is tracked in "pm", and the walk
1549 	 * will stop when we hit the end of the buffer.
1550 	 */
1551 	ret = 0;
1552 	while (count && (start_vaddr < end_vaddr)) {
1553 		int len;
1554 		unsigned long end;
1555 
1556 		pm.pos = 0;
1557 		end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1558 		/* overflow ? */
1559 		if (end < start_vaddr || end > end_vaddr)
1560 			end = end_vaddr;
1561 		ret = down_read_killable(&mm->mmap_sem);
1562 		if (ret)
1563 			goto out_free;
1564 		ret = walk_page_range(mm, start_vaddr, end, &pagemap_ops, &pm);
1565 		up_read(&mm->mmap_sem);
1566 		start_vaddr = end;
1567 
1568 		len = min(count, PM_ENTRY_BYTES * pm.pos);
1569 		if (copy_to_user(buf, pm.buffer, len)) {
1570 			ret = -EFAULT;
1571 			goto out_free;
1572 		}
1573 		copied += len;
1574 		buf += len;
1575 		count -= len;
1576 	}
1577 	*ppos += copied;
1578 	if (!ret || ret == PM_END_OF_BUFFER)
1579 		ret = copied;
1580 
1581 out_free:
1582 	kfree(pm.buffer);
1583 out_mm:
1584 	mmput(mm);
1585 out:
1586 	return ret;
1587 }
1588 
1589 static int pagemap_open(struct inode *inode, struct file *file)
1590 {
1591 	struct mm_struct *mm;
1592 
1593 	mm = proc_mem_open(inode, PTRACE_MODE_READ);
1594 	if (IS_ERR(mm))
1595 		return PTR_ERR(mm);
1596 	file->private_data = mm;
1597 	return 0;
1598 }
1599 
1600 static int pagemap_release(struct inode *inode, struct file *file)
1601 {
1602 	struct mm_struct *mm = file->private_data;
1603 
1604 	if (mm)
1605 		mmdrop(mm);
1606 	return 0;
1607 }
1608 
1609 const struct file_operations proc_pagemap_operations = {
1610 	.llseek		= mem_lseek, /* borrow this */
1611 	.read		= pagemap_read,
1612 	.open		= pagemap_open,
1613 	.release	= pagemap_release,
1614 };
1615 #endif /* CONFIG_PROC_PAGE_MONITOR */
1616 
1617 #ifdef CONFIG_NUMA
1618 
1619 struct numa_maps {
1620 	unsigned long pages;
1621 	unsigned long anon;
1622 	unsigned long active;
1623 	unsigned long writeback;
1624 	unsigned long mapcount_max;
1625 	unsigned long dirty;
1626 	unsigned long swapcache;
1627 	unsigned long node[MAX_NUMNODES];
1628 };
1629 
1630 struct numa_maps_private {
1631 	struct proc_maps_private proc_maps;
1632 	struct numa_maps md;
1633 };
1634 
1635 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1636 			unsigned long nr_pages)
1637 {
1638 	int count = page_mapcount(page);
1639 
1640 	md->pages += nr_pages;
1641 	if (pte_dirty || PageDirty(page))
1642 		md->dirty += nr_pages;
1643 
1644 	if (PageSwapCache(page))
1645 		md->swapcache += nr_pages;
1646 
1647 	if (PageActive(page) || PageUnevictable(page))
1648 		md->active += nr_pages;
1649 
1650 	if (PageWriteback(page))
1651 		md->writeback += nr_pages;
1652 
1653 	if (PageAnon(page))
1654 		md->anon += nr_pages;
1655 
1656 	if (count > md->mapcount_max)
1657 		md->mapcount_max = count;
1658 
1659 	md->node[page_to_nid(page)] += nr_pages;
1660 }
1661 
1662 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1663 		unsigned long addr)
1664 {
1665 	struct page *page;
1666 	int nid;
1667 
1668 	if (!pte_present(pte))
1669 		return NULL;
1670 
1671 	page = vm_normal_page(vma, addr, pte);
1672 	if (!page)
1673 		return NULL;
1674 
1675 	if (PageReserved(page))
1676 		return NULL;
1677 
1678 	nid = page_to_nid(page);
1679 	if (!node_isset(nid, node_states[N_MEMORY]))
1680 		return NULL;
1681 
1682 	return page;
1683 }
1684 
1685 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1686 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1687 					      struct vm_area_struct *vma,
1688 					      unsigned long addr)
1689 {
1690 	struct page *page;
1691 	int nid;
1692 
1693 	if (!pmd_present(pmd))
1694 		return NULL;
1695 
1696 	page = vm_normal_page_pmd(vma, addr, pmd);
1697 	if (!page)
1698 		return NULL;
1699 
1700 	if (PageReserved(page))
1701 		return NULL;
1702 
1703 	nid = page_to_nid(page);
1704 	if (!node_isset(nid, node_states[N_MEMORY]))
1705 		return NULL;
1706 
1707 	return page;
1708 }
1709 #endif
1710 
1711 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1712 		unsigned long end, struct mm_walk *walk)
1713 {
1714 	struct numa_maps *md = walk->private;
1715 	struct vm_area_struct *vma = walk->vma;
1716 	spinlock_t *ptl;
1717 	pte_t *orig_pte;
1718 	pte_t *pte;
1719 
1720 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1721 	ptl = pmd_trans_huge_lock(pmd, vma);
1722 	if (ptl) {
1723 		struct page *page;
1724 
1725 		page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1726 		if (page)
1727 			gather_stats(page, md, pmd_dirty(*pmd),
1728 				     HPAGE_PMD_SIZE/PAGE_SIZE);
1729 		spin_unlock(ptl);
1730 		return 0;
1731 	}
1732 
1733 	if (pmd_trans_unstable(pmd))
1734 		return 0;
1735 #endif
1736 	orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1737 	do {
1738 		struct page *page = can_gather_numa_stats(*pte, vma, addr);
1739 		if (!page)
1740 			continue;
1741 		gather_stats(page, md, pte_dirty(*pte), 1);
1742 
1743 	} while (pte++, addr += PAGE_SIZE, addr != end);
1744 	pte_unmap_unlock(orig_pte, ptl);
1745 	cond_resched();
1746 	return 0;
1747 }
1748 #ifdef CONFIG_HUGETLB_PAGE
1749 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1750 		unsigned long addr, unsigned long end, struct mm_walk *walk)
1751 {
1752 	pte_t huge_pte = huge_ptep_get(pte);
1753 	struct numa_maps *md;
1754 	struct page *page;
1755 
1756 	if (!pte_present(huge_pte))
1757 		return 0;
1758 
1759 	page = pte_page(huge_pte);
1760 	if (!page)
1761 		return 0;
1762 
1763 	md = walk->private;
1764 	gather_stats(page, md, pte_dirty(huge_pte), 1);
1765 	return 0;
1766 }
1767 
1768 #else
1769 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1770 		unsigned long addr, unsigned long end, struct mm_walk *walk)
1771 {
1772 	return 0;
1773 }
1774 #endif
1775 
1776 static const struct mm_walk_ops show_numa_ops = {
1777 	.hugetlb_entry = gather_hugetlb_stats,
1778 	.pmd_entry = gather_pte_stats,
1779 };
1780 
1781 /*
1782  * Display pages allocated per node and memory policy via /proc.
1783  */
1784 static int show_numa_map(struct seq_file *m, void *v)
1785 {
1786 	struct numa_maps_private *numa_priv = m->private;
1787 	struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1788 	struct vm_area_struct *vma = v;
1789 	struct numa_maps *md = &numa_priv->md;
1790 	struct file *file = vma->vm_file;
1791 	struct mm_struct *mm = vma->vm_mm;
1792 	struct mempolicy *pol;
1793 	char buffer[64];
1794 	int nid;
1795 
1796 	if (!mm)
1797 		return 0;
1798 
1799 	/* Ensure we start with an empty set of numa_maps statistics. */
1800 	memset(md, 0, sizeof(*md));
1801 
1802 	pol = __get_vma_policy(vma, vma->vm_start);
1803 	if (pol) {
1804 		mpol_to_str(buffer, sizeof(buffer), pol);
1805 		mpol_cond_put(pol);
1806 	} else {
1807 		mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1808 	}
1809 
1810 	seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1811 
1812 	if (file) {
1813 		seq_puts(m, " file=");
1814 		seq_file_path(m, file, "\n\t= ");
1815 	} else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1816 		seq_puts(m, " heap");
1817 	} else if (is_stack(vma)) {
1818 		seq_puts(m, " stack");
1819 	}
1820 
1821 	if (is_vm_hugetlb_page(vma))
1822 		seq_puts(m, " huge");
1823 
1824 	/* mmap_sem is held by m_start */
1825 	walk_page_vma(vma, &show_numa_ops, md);
1826 
1827 	if (!md->pages)
1828 		goto out;
1829 
1830 	if (md->anon)
1831 		seq_printf(m, " anon=%lu", md->anon);
1832 
1833 	if (md->dirty)
1834 		seq_printf(m, " dirty=%lu", md->dirty);
1835 
1836 	if (md->pages != md->anon && md->pages != md->dirty)
1837 		seq_printf(m, " mapped=%lu", md->pages);
1838 
1839 	if (md->mapcount_max > 1)
1840 		seq_printf(m, " mapmax=%lu", md->mapcount_max);
1841 
1842 	if (md->swapcache)
1843 		seq_printf(m, " swapcache=%lu", md->swapcache);
1844 
1845 	if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1846 		seq_printf(m, " active=%lu", md->active);
1847 
1848 	if (md->writeback)
1849 		seq_printf(m, " writeback=%lu", md->writeback);
1850 
1851 	for_each_node_state(nid, N_MEMORY)
1852 		if (md->node[nid])
1853 			seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1854 
1855 	seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1856 out:
1857 	seq_putc(m, '\n');
1858 	return 0;
1859 }
1860 
1861 static const struct seq_operations proc_pid_numa_maps_op = {
1862 	.start  = m_start,
1863 	.next   = m_next,
1864 	.stop   = m_stop,
1865 	.show   = show_numa_map,
1866 };
1867 
1868 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1869 {
1870 	return proc_maps_open(inode, file, &proc_pid_numa_maps_op,
1871 				sizeof(struct numa_maps_private));
1872 }
1873 
1874 const struct file_operations proc_pid_numa_maps_operations = {
1875 	.open		= pid_numa_maps_open,
1876 	.read		= seq_read,
1877 	.llseek		= seq_lseek,
1878 	.release	= proc_map_release,
1879 };
1880 
1881 #endif /* CONFIG_NUMA */
1882