xref: /openbmc/linux/fs/proc/task_mmu.c (revision 63dc02bd)
1 #include <linux/mm.h>
2 #include <linux/hugetlb.h>
3 #include <linux/huge_mm.h>
4 #include <linux/mount.h>
5 #include <linux/seq_file.h>
6 #include <linux/highmem.h>
7 #include <linux/ptrace.h>
8 #include <linux/slab.h>
9 #include <linux/pagemap.h>
10 #include <linux/mempolicy.h>
11 #include <linux/rmap.h>
12 #include <linux/swap.h>
13 #include <linux/swapops.h>
14 
15 #include <asm/elf.h>
16 #include <asm/uaccess.h>
17 #include <asm/tlbflush.h>
18 #include "internal.h"
19 
20 void task_mem(struct seq_file *m, struct mm_struct *mm)
21 {
22 	unsigned long data, text, lib, swap;
23 	unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
24 
25 	/*
26 	 * Note: to minimize their overhead, mm maintains hiwater_vm and
27 	 * hiwater_rss only when about to *lower* total_vm or rss.  Any
28 	 * collector of these hiwater stats must therefore get total_vm
29 	 * and rss too, which will usually be the higher.  Barriers? not
30 	 * worth the effort, such snapshots can always be inconsistent.
31 	 */
32 	hiwater_vm = total_vm = mm->total_vm;
33 	if (hiwater_vm < mm->hiwater_vm)
34 		hiwater_vm = mm->hiwater_vm;
35 	hiwater_rss = total_rss = get_mm_rss(mm);
36 	if (hiwater_rss < mm->hiwater_rss)
37 		hiwater_rss = mm->hiwater_rss;
38 
39 	data = mm->total_vm - mm->shared_vm - mm->stack_vm;
40 	text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
41 	lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
42 	swap = get_mm_counter(mm, MM_SWAPENTS);
43 	seq_printf(m,
44 		"VmPeak:\t%8lu kB\n"
45 		"VmSize:\t%8lu kB\n"
46 		"VmLck:\t%8lu kB\n"
47 		"VmPin:\t%8lu kB\n"
48 		"VmHWM:\t%8lu kB\n"
49 		"VmRSS:\t%8lu kB\n"
50 		"VmData:\t%8lu kB\n"
51 		"VmStk:\t%8lu kB\n"
52 		"VmExe:\t%8lu kB\n"
53 		"VmLib:\t%8lu kB\n"
54 		"VmPTE:\t%8lu kB\n"
55 		"VmSwap:\t%8lu kB\n",
56 		hiwater_vm << (PAGE_SHIFT-10),
57 		(total_vm - mm->reserved_vm) << (PAGE_SHIFT-10),
58 		mm->locked_vm << (PAGE_SHIFT-10),
59 		mm->pinned_vm << (PAGE_SHIFT-10),
60 		hiwater_rss << (PAGE_SHIFT-10),
61 		total_rss << (PAGE_SHIFT-10),
62 		data << (PAGE_SHIFT-10),
63 		mm->stack_vm << (PAGE_SHIFT-10), text, lib,
64 		(PTRS_PER_PTE*sizeof(pte_t)*mm->nr_ptes) >> 10,
65 		swap << (PAGE_SHIFT-10));
66 }
67 
68 unsigned long task_vsize(struct mm_struct *mm)
69 {
70 	return PAGE_SIZE * mm->total_vm;
71 }
72 
73 unsigned long task_statm(struct mm_struct *mm,
74 			 unsigned long *shared, unsigned long *text,
75 			 unsigned long *data, unsigned long *resident)
76 {
77 	*shared = get_mm_counter(mm, MM_FILEPAGES);
78 	*text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
79 								>> PAGE_SHIFT;
80 	*data = mm->total_vm - mm->shared_vm;
81 	*resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
82 	return mm->total_vm;
83 }
84 
85 static void pad_len_spaces(struct seq_file *m, int len)
86 {
87 	len = 25 + sizeof(void*) * 6 - len;
88 	if (len < 1)
89 		len = 1;
90 	seq_printf(m, "%*c", len, ' ');
91 }
92 
93 static void vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma)
94 {
95 	if (vma && vma != priv->tail_vma) {
96 		struct mm_struct *mm = vma->vm_mm;
97 		up_read(&mm->mmap_sem);
98 		mmput(mm);
99 	}
100 }
101 
102 static void *m_start(struct seq_file *m, loff_t *pos)
103 {
104 	struct proc_maps_private *priv = m->private;
105 	unsigned long last_addr = m->version;
106 	struct mm_struct *mm;
107 	struct vm_area_struct *vma, *tail_vma = NULL;
108 	loff_t l = *pos;
109 
110 	/* Clear the per syscall fields in priv */
111 	priv->task = NULL;
112 	priv->tail_vma = NULL;
113 
114 	/*
115 	 * We remember last_addr rather than next_addr to hit with
116 	 * mmap_cache most of the time. We have zero last_addr at
117 	 * the beginning and also after lseek. We will have -1 last_addr
118 	 * after the end of the vmas.
119 	 */
120 
121 	if (last_addr == -1UL)
122 		return NULL;
123 
124 	priv->task = get_pid_task(priv->pid, PIDTYPE_PID);
125 	if (!priv->task)
126 		return ERR_PTR(-ESRCH);
127 
128 	mm = mm_for_maps(priv->task);
129 	if (!mm || IS_ERR(mm))
130 		return mm;
131 	down_read(&mm->mmap_sem);
132 
133 	tail_vma = get_gate_vma(priv->task->mm);
134 	priv->tail_vma = tail_vma;
135 
136 	/* Start with last addr hint */
137 	vma = find_vma(mm, last_addr);
138 	if (last_addr && vma) {
139 		vma = vma->vm_next;
140 		goto out;
141 	}
142 
143 	/*
144 	 * Check the vma index is within the range and do
145 	 * sequential scan until m_index.
146 	 */
147 	vma = NULL;
148 	if ((unsigned long)l < mm->map_count) {
149 		vma = mm->mmap;
150 		while (l-- && vma)
151 			vma = vma->vm_next;
152 		goto out;
153 	}
154 
155 	if (l != mm->map_count)
156 		tail_vma = NULL; /* After gate vma */
157 
158 out:
159 	if (vma)
160 		return vma;
161 
162 	/* End of vmas has been reached */
163 	m->version = (tail_vma != NULL)? 0: -1UL;
164 	up_read(&mm->mmap_sem);
165 	mmput(mm);
166 	return tail_vma;
167 }
168 
169 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
170 {
171 	struct proc_maps_private *priv = m->private;
172 	struct vm_area_struct *vma = v;
173 	struct vm_area_struct *tail_vma = priv->tail_vma;
174 
175 	(*pos)++;
176 	if (vma && (vma != tail_vma) && vma->vm_next)
177 		return vma->vm_next;
178 	vma_stop(priv, vma);
179 	return (vma != tail_vma)? tail_vma: NULL;
180 }
181 
182 static void m_stop(struct seq_file *m, void *v)
183 {
184 	struct proc_maps_private *priv = m->private;
185 	struct vm_area_struct *vma = v;
186 
187 	if (!IS_ERR(vma))
188 		vma_stop(priv, vma);
189 	if (priv->task)
190 		put_task_struct(priv->task);
191 }
192 
193 static int do_maps_open(struct inode *inode, struct file *file,
194 			const struct seq_operations *ops)
195 {
196 	struct proc_maps_private *priv;
197 	int ret = -ENOMEM;
198 	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
199 	if (priv) {
200 		priv->pid = proc_pid(inode);
201 		ret = seq_open(file, ops);
202 		if (!ret) {
203 			struct seq_file *m = file->private_data;
204 			m->private = priv;
205 		} else {
206 			kfree(priv);
207 		}
208 	}
209 	return ret;
210 }
211 
212 static void
213 show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
214 {
215 	struct mm_struct *mm = vma->vm_mm;
216 	struct file *file = vma->vm_file;
217 	struct proc_maps_private *priv = m->private;
218 	struct task_struct *task = priv->task;
219 	vm_flags_t flags = vma->vm_flags;
220 	unsigned long ino = 0;
221 	unsigned long long pgoff = 0;
222 	unsigned long start, end;
223 	dev_t dev = 0;
224 	int len;
225 	const char *name = NULL;
226 
227 	if (file) {
228 		struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
229 		dev = inode->i_sb->s_dev;
230 		ino = inode->i_ino;
231 		pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
232 	}
233 
234 	/* We don't show the stack guard page in /proc/maps */
235 	start = vma->vm_start;
236 	if (stack_guard_page_start(vma, start))
237 		start += PAGE_SIZE;
238 	end = vma->vm_end;
239 	if (stack_guard_page_end(vma, end))
240 		end -= PAGE_SIZE;
241 
242 	seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu %n",
243 			start,
244 			end,
245 			flags & VM_READ ? 'r' : '-',
246 			flags & VM_WRITE ? 'w' : '-',
247 			flags & VM_EXEC ? 'x' : '-',
248 			flags & VM_MAYSHARE ? 's' : 'p',
249 			pgoff,
250 			MAJOR(dev), MINOR(dev), ino, &len);
251 
252 	/*
253 	 * Print the dentry name for named mappings, and a
254 	 * special [heap] marker for the heap:
255 	 */
256 	if (file) {
257 		pad_len_spaces(m, len);
258 		seq_path(m, &file->f_path, "\n");
259 		goto done;
260 	}
261 
262 	name = arch_vma_name(vma);
263 	if (!name) {
264 		pid_t tid;
265 
266 		if (!mm) {
267 			name = "[vdso]";
268 			goto done;
269 		}
270 
271 		if (vma->vm_start <= mm->brk &&
272 		    vma->vm_end >= mm->start_brk) {
273 			name = "[heap]";
274 			goto done;
275 		}
276 
277 		tid = vm_is_stack(task, vma, is_pid);
278 
279 		if (tid != 0) {
280 			/*
281 			 * Thread stack in /proc/PID/task/TID/maps or
282 			 * the main process stack.
283 			 */
284 			if (!is_pid || (vma->vm_start <= mm->start_stack &&
285 			    vma->vm_end >= mm->start_stack)) {
286 				name = "[stack]";
287 			} else {
288 				/* Thread stack in /proc/PID/maps */
289 				pad_len_spaces(m, len);
290 				seq_printf(m, "[stack:%d]", tid);
291 			}
292 		}
293 	}
294 
295 done:
296 	if (name) {
297 		pad_len_spaces(m, len);
298 		seq_puts(m, name);
299 	}
300 	seq_putc(m, '\n');
301 }
302 
303 static int show_map(struct seq_file *m, void *v, int is_pid)
304 {
305 	struct vm_area_struct *vma = v;
306 	struct proc_maps_private *priv = m->private;
307 	struct task_struct *task = priv->task;
308 
309 	show_map_vma(m, vma, is_pid);
310 
311 	if (m->count < m->size)  /* vma is copied successfully */
312 		m->version = (vma != get_gate_vma(task->mm))
313 			? vma->vm_start : 0;
314 	return 0;
315 }
316 
317 static int show_pid_map(struct seq_file *m, void *v)
318 {
319 	return show_map(m, v, 1);
320 }
321 
322 static int show_tid_map(struct seq_file *m, void *v)
323 {
324 	return show_map(m, v, 0);
325 }
326 
327 static const struct seq_operations proc_pid_maps_op = {
328 	.start	= m_start,
329 	.next	= m_next,
330 	.stop	= m_stop,
331 	.show	= show_pid_map
332 };
333 
334 static const struct seq_operations proc_tid_maps_op = {
335 	.start	= m_start,
336 	.next	= m_next,
337 	.stop	= m_stop,
338 	.show	= show_tid_map
339 };
340 
341 static int pid_maps_open(struct inode *inode, struct file *file)
342 {
343 	return do_maps_open(inode, file, &proc_pid_maps_op);
344 }
345 
346 static int tid_maps_open(struct inode *inode, struct file *file)
347 {
348 	return do_maps_open(inode, file, &proc_tid_maps_op);
349 }
350 
351 const struct file_operations proc_pid_maps_operations = {
352 	.open		= pid_maps_open,
353 	.read		= seq_read,
354 	.llseek		= seq_lseek,
355 	.release	= seq_release_private,
356 };
357 
358 const struct file_operations proc_tid_maps_operations = {
359 	.open		= tid_maps_open,
360 	.read		= seq_read,
361 	.llseek		= seq_lseek,
362 	.release	= seq_release_private,
363 };
364 
365 /*
366  * Proportional Set Size(PSS): my share of RSS.
367  *
368  * PSS of a process is the count of pages it has in memory, where each
369  * page is divided by the number of processes sharing it.  So if a
370  * process has 1000 pages all to itself, and 1000 shared with one other
371  * process, its PSS will be 1500.
372  *
373  * To keep (accumulated) division errors low, we adopt a 64bit
374  * fixed-point pss counter to minimize division errors. So (pss >>
375  * PSS_SHIFT) would be the real byte count.
376  *
377  * A shift of 12 before division means (assuming 4K page size):
378  * 	- 1M 3-user-pages add up to 8KB errors;
379  * 	- supports mapcount up to 2^24, or 16M;
380  * 	- supports PSS up to 2^52 bytes, or 4PB.
381  */
382 #define PSS_SHIFT 12
383 
384 #ifdef CONFIG_PROC_PAGE_MONITOR
385 struct mem_size_stats {
386 	struct vm_area_struct *vma;
387 	unsigned long resident;
388 	unsigned long shared_clean;
389 	unsigned long shared_dirty;
390 	unsigned long private_clean;
391 	unsigned long private_dirty;
392 	unsigned long referenced;
393 	unsigned long anonymous;
394 	unsigned long anonymous_thp;
395 	unsigned long swap;
396 	u64 pss;
397 };
398 
399 
400 static void smaps_pte_entry(pte_t ptent, unsigned long addr,
401 		unsigned long ptent_size, struct mm_walk *walk)
402 {
403 	struct mem_size_stats *mss = walk->private;
404 	struct vm_area_struct *vma = mss->vma;
405 	struct page *page;
406 	int mapcount;
407 
408 	if (is_swap_pte(ptent)) {
409 		mss->swap += ptent_size;
410 		return;
411 	}
412 
413 	if (!pte_present(ptent))
414 		return;
415 
416 	page = vm_normal_page(vma, addr, ptent);
417 	if (!page)
418 		return;
419 
420 	if (PageAnon(page))
421 		mss->anonymous += ptent_size;
422 
423 	mss->resident += ptent_size;
424 	/* Accumulate the size in pages that have been accessed. */
425 	if (pte_young(ptent) || PageReferenced(page))
426 		mss->referenced += ptent_size;
427 	mapcount = page_mapcount(page);
428 	if (mapcount >= 2) {
429 		if (pte_dirty(ptent) || PageDirty(page))
430 			mss->shared_dirty += ptent_size;
431 		else
432 			mss->shared_clean += ptent_size;
433 		mss->pss += (ptent_size << PSS_SHIFT) / mapcount;
434 	} else {
435 		if (pte_dirty(ptent) || PageDirty(page))
436 			mss->private_dirty += ptent_size;
437 		else
438 			mss->private_clean += ptent_size;
439 		mss->pss += (ptent_size << PSS_SHIFT);
440 	}
441 }
442 
443 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
444 			   struct mm_walk *walk)
445 {
446 	struct mem_size_stats *mss = walk->private;
447 	struct vm_area_struct *vma = mss->vma;
448 	pte_t *pte;
449 	spinlock_t *ptl;
450 
451 	if (pmd_trans_huge_lock(pmd, vma) == 1) {
452 		smaps_pte_entry(*(pte_t *)pmd, addr, HPAGE_PMD_SIZE, walk);
453 		spin_unlock(&walk->mm->page_table_lock);
454 		mss->anonymous_thp += HPAGE_PMD_SIZE;
455 		return 0;
456 	}
457 
458 	if (pmd_trans_unstable(pmd))
459 		return 0;
460 	/*
461 	 * The mmap_sem held all the way back in m_start() is what
462 	 * keeps khugepaged out of here and from collapsing things
463 	 * in here.
464 	 */
465 	pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
466 	for (; addr != end; pte++, addr += PAGE_SIZE)
467 		smaps_pte_entry(*pte, addr, PAGE_SIZE, walk);
468 	pte_unmap_unlock(pte - 1, ptl);
469 	cond_resched();
470 	return 0;
471 }
472 
473 static int show_smap(struct seq_file *m, void *v, int is_pid)
474 {
475 	struct proc_maps_private *priv = m->private;
476 	struct task_struct *task = priv->task;
477 	struct vm_area_struct *vma = v;
478 	struct mem_size_stats mss;
479 	struct mm_walk smaps_walk = {
480 		.pmd_entry = smaps_pte_range,
481 		.mm = vma->vm_mm,
482 		.private = &mss,
483 	};
484 
485 	memset(&mss, 0, sizeof mss);
486 	mss.vma = vma;
487 	/* mmap_sem is held in m_start */
488 	if (vma->vm_mm && !is_vm_hugetlb_page(vma))
489 		walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk);
490 
491 	show_map_vma(m, vma, is_pid);
492 
493 	seq_printf(m,
494 		   "Size:           %8lu kB\n"
495 		   "Rss:            %8lu kB\n"
496 		   "Pss:            %8lu kB\n"
497 		   "Shared_Clean:   %8lu kB\n"
498 		   "Shared_Dirty:   %8lu kB\n"
499 		   "Private_Clean:  %8lu kB\n"
500 		   "Private_Dirty:  %8lu kB\n"
501 		   "Referenced:     %8lu kB\n"
502 		   "Anonymous:      %8lu kB\n"
503 		   "AnonHugePages:  %8lu kB\n"
504 		   "Swap:           %8lu kB\n"
505 		   "KernelPageSize: %8lu kB\n"
506 		   "MMUPageSize:    %8lu kB\n"
507 		   "Locked:         %8lu kB\n",
508 		   (vma->vm_end - vma->vm_start) >> 10,
509 		   mss.resident >> 10,
510 		   (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
511 		   mss.shared_clean  >> 10,
512 		   mss.shared_dirty  >> 10,
513 		   mss.private_clean >> 10,
514 		   mss.private_dirty >> 10,
515 		   mss.referenced >> 10,
516 		   mss.anonymous >> 10,
517 		   mss.anonymous_thp >> 10,
518 		   mss.swap >> 10,
519 		   vma_kernel_pagesize(vma) >> 10,
520 		   vma_mmu_pagesize(vma) >> 10,
521 		   (vma->vm_flags & VM_LOCKED) ?
522 			(unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
523 
524 	if (m->count < m->size)  /* vma is copied successfully */
525 		m->version = (vma != get_gate_vma(task->mm))
526 			? vma->vm_start : 0;
527 	return 0;
528 }
529 
530 static int show_pid_smap(struct seq_file *m, void *v)
531 {
532 	return show_smap(m, v, 1);
533 }
534 
535 static int show_tid_smap(struct seq_file *m, void *v)
536 {
537 	return show_smap(m, v, 0);
538 }
539 
540 static const struct seq_operations proc_pid_smaps_op = {
541 	.start	= m_start,
542 	.next	= m_next,
543 	.stop	= m_stop,
544 	.show	= show_pid_smap
545 };
546 
547 static const struct seq_operations proc_tid_smaps_op = {
548 	.start	= m_start,
549 	.next	= m_next,
550 	.stop	= m_stop,
551 	.show	= show_tid_smap
552 };
553 
554 static int pid_smaps_open(struct inode *inode, struct file *file)
555 {
556 	return do_maps_open(inode, file, &proc_pid_smaps_op);
557 }
558 
559 static int tid_smaps_open(struct inode *inode, struct file *file)
560 {
561 	return do_maps_open(inode, file, &proc_tid_smaps_op);
562 }
563 
564 const struct file_operations proc_pid_smaps_operations = {
565 	.open		= pid_smaps_open,
566 	.read		= seq_read,
567 	.llseek		= seq_lseek,
568 	.release	= seq_release_private,
569 };
570 
571 const struct file_operations proc_tid_smaps_operations = {
572 	.open		= tid_smaps_open,
573 	.read		= seq_read,
574 	.llseek		= seq_lseek,
575 	.release	= seq_release_private,
576 };
577 
578 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
579 				unsigned long end, struct mm_walk *walk)
580 {
581 	struct vm_area_struct *vma = walk->private;
582 	pte_t *pte, ptent;
583 	spinlock_t *ptl;
584 	struct page *page;
585 
586 	split_huge_page_pmd(walk->mm, pmd);
587 	if (pmd_trans_unstable(pmd))
588 		return 0;
589 
590 	pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
591 	for (; addr != end; pte++, addr += PAGE_SIZE) {
592 		ptent = *pte;
593 		if (!pte_present(ptent))
594 			continue;
595 
596 		page = vm_normal_page(vma, addr, ptent);
597 		if (!page)
598 			continue;
599 
600 		/* Clear accessed and referenced bits. */
601 		ptep_test_and_clear_young(vma, addr, pte);
602 		ClearPageReferenced(page);
603 	}
604 	pte_unmap_unlock(pte - 1, ptl);
605 	cond_resched();
606 	return 0;
607 }
608 
609 #define CLEAR_REFS_ALL 1
610 #define CLEAR_REFS_ANON 2
611 #define CLEAR_REFS_MAPPED 3
612 
613 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
614 				size_t count, loff_t *ppos)
615 {
616 	struct task_struct *task;
617 	char buffer[PROC_NUMBUF];
618 	struct mm_struct *mm;
619 	struct vm_area_struct *vma;
620 	int type;
621 	int rv;
622 
623 	memset(buffer, 0, sizeof(buffer));
624 	if (count > sizeof(buffer) - 1)
625 		count = sizeof(buffer) - 1;
626 	if (copy_from_user(buffer, buf, count))
627 		return -EFAULT;
628 	rv = kstrtoint(strstrip(buffer), 10, &type);
629 	if (rv < 0)
630 		return rv;
631 	if (type < CLEAR_REFS_ALL || type > CLEAR_REFS_MAPPED)
632 		return -EINVAL;
633 	task = get_proc_task(file->f_path.dentry->d_inode);
634 	if (!task)
635 		return -ESRCH;
636 	mm = get_task_mm(task);
637 	if (mm) {
638 		struct mm_walk clear_refs_walk = {
639 			.pmd_entry = clear_refs_pte_range,
640 			.mm = mm,
641 		};
642 		down_read(&mm->mmap_sem);
643 		for (vma = mm->mmap; vma; vma = vma->vm_next) {
644 			clear_refs_walk.private = vma;
645 			if (is_vm_hugetlb_page(vma))
646 				continue;
647 			/*
648 			 * Writing 1 to /proc/pid/clear_refs affects all pages.
649 			 *
650 			 * Writing 2 to /proc/pid/clear_refs only affects
651 			 * Anonymous pages.
652 			 *
653 			 * Writing 3 to /proc/pid/clear_refs only affects file
654 			 * mapped pages.
655 			 */
656 			if (type == CLEAR_REFS_ANON && vma->vm_file)
657 				continue;
658 			if (type == CLEAR_REFS_MAPPED && !vma->vm_file)
659 				continue;
660 			walk_page_range(vma->vm_start, vma->vm_end,
661 					&clear_refs_walk);
662 		}
663 		flush_tlb_mm(mm);
664 		up_read(&mm->mmap_sem);
665 		mmput(mm);
666 	}
667 	put_task_struct(task);
668 
669 	return count;
670 }
671 
672 const struct file_operations proc_clear_refs_operations = {
673 	.write		= clear_refs_write,
674 	.llseek		= noop_llseek,
675 };
676 
677 typedef struct {
678 	u64 pme;
679 } pagemap_entry_t;
680 
681 struct pagemapread {
682 	int pos, len;
683 	pagemap_entry_t *buffer;
684 };
685 
686 #define PAGEMAP_WALK_SIZE	(PMD_SIZE)
687 #define PAGEMAP_WALK_MASK	(PMD_MASK)
688 
689 #define PM_ENTRY_BYTES      sizeof(u64)
690 #define PM_STATUS_BITS      3
691 #define PM_STATUS_OFFSET    (64 - PM_STATUS_BITS)
692 #define PM_STATUS_MASK      (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
693 #define PM_STATUS(nr)       (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
694 #define PM_PSHIFT_BITS      6
695 #define PM_PSHIFT_OFFSET    (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
696 #define PM_PSHIFT_MASK      (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
697 #define PM_PSHIFT(x)        (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
698 #define PM_PFRAME_MASK      ((1LL << PM_PSHIFT_OFFSET) - 1)
699 #define PM_PFRAME(x)        ((x) & PM_PFRAME_MASK)
700 
701 #define PM_PRESENT          PM_STATUS(4LL)
702 #define PM_SWAP             PM_STATUS(2LL)
703 #define PM_NOT_PRESENT      PM_PSHIFT(PAGE_SHIFT)
704 #define PM_END_OF_BUFFER    1
705 
706 static inline pagemap_entry_t make_pme(u64 val)
707 {
708 	return (pagemap_entry_t) { .pme = val };
709 }
710 
711 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
712 			  struct pagemapread *pm)
713 {
714 	pm->buffer[pm->pos++] = *pme;
715 	if (pm->pos >= pm->len)
716 		return PM_END_OF_BUFFER;
717 	return 0;
718 }
719 
720 static int pagemap_pte_hole(unsigned long start, unsigned long end,
721 				struct mm_walk *walk)
722 {
723 	struct pagemapread *pm = walk->private;
724 	unsigned long addr;
725 	int err = 0;
726 	pagemap_entry_t pme = make_pme(PM_NOT_PRESENT);
727 
728 	for (addr = start; addr < end; addr += PAGE_SIZE) {
729 		err = add_to_pagemap(addr, &pme, pm);
730 		if (err)
731 			break;
732 	}
733 	return err;
734 }
735 
736 static u64 swap_pte_to_pagemap_entry(pte_t pte)
737 {
738 	swp_entry_t e = pte_to_swp_entry(pte);
739 	return swp_type(e) | (swp_offset(e) << MAX_SWAPFILES_SHIFT);
740 }
741 
742 static void pte_to_pagemap_entry(pagemap_entry_t *pme, pte_t pte)
743 {
744 	if (is_swap_pte(pte))
745 		*pme = make_pme(PM_PFRAME(swap_pte_to_pagemap_entry(pte))
746 				| PM_PSHIFT(PAGE_SHIFT) | PM_SWAP);
747 	else if (pte_present(pte))
748 		*pme = make_pme(PM_PFRAME(pte_pfn(pte))
749 				| PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT);
750 	else
751 		*pme = make_pme(PM_NOT_PRESENT);
752 }
753 
754 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
755 static void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme,
756 					pmd_t pmd, int offset)
757 {
758 	/*
759 	 * Currently pmd for thp is always present because thp can not be
760 	 * swapped-out, migrated, or HWPOISONed (split in such cases instead.)
761 	 * This if-check is just to prepare for future implementation.
762 	 */
763 	if (pmd_present(pmd))
764 		*pme = make_pme(PM_PFRAME(pmd_pfn(pmd) + offset)
765 				| PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT);
766 	else
767 		*pme = make_pme(PM_NOT_PRESENT);
768 }
769 #else
770 static inline void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme,
771 						pmd_t pmd, int offset)
772 {
773 }
774 #endif
775 
776 static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
777 			     struct mm_walk *walk)
778 {
779 	struct vm_area_struct *vma;
780 	struct pagemapread *pm = walk->private;
781 	pte_t *pte;
782 	int err = 0;
783 	pagemap_entry_t pme = make_pme(PM_NOT_PRESENT);
784 
785 	/* find the first VMA at or above 'addr' */
786 	vma = find_vma(walk->mm, addr);
787 	if (pmd_trans_huge_lock(pmd, vma) == 1) {
788 		for (; addr != end; addr += PAGE_SIZE) {
789 			unsigned long offset;
790 
791 			offset = (addr & ~PAGEMAP_WALK_MASK) >>
792 					PAGE_SHIFT;
793 			thp_pmd_to_pagemap_entry(&pme, *pmd, offset);
794 			err = add_to_pagemap(addr, &pme, pm);
795 			if (err)
796 				break;
797 		}
798 		spin_unlock(&walk->mm->page_table_lock);
799 		return err;
800 	}
801 
802 	if (pmd_trans_unstable(pmd))
803 		return 0;
804 	for (; addr != end; addr += PAGE_SIZE) {
805 
806 		/* check to see if we've left 'vma' behind
807 		 * and need a new, higher one */
808 		if (vma && (addr >= vma->vm_end)) {
809 			vma = find_vma(walk->mm, addr);
810 			pme = make_pme(PM_NOT_PRESENT);
811 		}
812 
813 		/* check that 'vma' actually covers this address,
814 		 * and that it isn't a huge page vma */
815 		if (vma && (vma->vm_start <= addr) &&
816 		    !is_vm_hugetlb_page(vma)) {
817 			pte = pte_offset_map(pmd, addr);
818 			pte_to_pagemap_entry(&pme, *pte);
819 			/* unmap before userspace copy */
820 			pte_unmap(pte);
821 		}
822 		err = add_to_pagemap(addr, &pme, pm);
823 		if (err)
824 			return err;
825 	}
826 
827 	cond_resched();
828 
829 	return err;
830 }
831 
832 #ifdef CONFIG_HUGETLB_PAGE
833 static void huge_pte_to_pagemap_entry(pagemap_entry_t *pme,
834 					pte_t pte, int offset)
835 {
836 	if (pte_present(pte))
837 		*pme = make_pme(PM_PFRAME(pte_pfn(pte) + offset)
838 				| PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT);
839 	else
840 		*pme = make_pme(PM_NOT_PRESENT);
841 }
842 
843 /* This function walks within one hugetlb entry in the single call */
844 static int pagemap_hugetlb_range(pte_t *pte, unsigned long hmask,
845 				 unsigned long addr, unsigned long end,
846 				 struct mm_walk *walk)
847 {
848 	struct pagemapread *pm = walk->private;
849 	int err = 0;
850 	pagemap_entry_t pme;
851 
852 	for (; addr != end; addr += PAGE_SIZE) {
853 		int offset = (addr & ~hmask) >> PAGE_SHIFT;
854 		huge_pte_to_pagemap_entry(&pme, *pte, offset);
855 		err = add_to_pagemap(addr, &pme, pm);
856 		if (err)
857 			return err;
858 	}
859 
860 	cond_resched();
861 
862 	return err;
863 }
864 #endif /* HUGETLB_PAGE */
865 
866 /*
867  * /proc/pid/pagemap - an array mapping virtual pages to pfns
868  *
869  * For each page in the address space, this file contains one 64-bit entry
870  * consisting of the following:
871  *
872  * Bits 0-55  page frame number (PFN) if present
873  * Bits 0-4   swap type if swapped
874  * Bits 5-55  swap offset if swapped
875  * Bits 55-60 page shift (page size = 1<<page shift)
876  * Bit  61    reserved for future use
877  * Bit  62    page swapped
878  * Bit  63    page present
879  *
880  * If the page is not present but in swap, then the PFN contains an
881  * encoding of the swap file number and the page's offset into the
882  * swap. Unmapped pages return a null PFN. This allows determining
883  * precisely which pages are mapped (or in swap) and comparing mapped
884  * pages between processes.
885  *
886  * Efficient users of this interface will use /proc/pid/maps to
887  * determine which areas of memory are actually mapped and llseek to
888  * skip over unmapped regions.
889  */
890 static ssize_t pagemap_read(struct file *file, char __user *buf,
891 			    size_t count, loff_t *ppos)
892 {
893 	struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
894 	struct mm_struct *mm;
895 	struct pagemapread pm;
896 	int ret = -ESRCH;
897 	struct mm_walk pagemap_walk = {};
898 	unsigned long src;
899 	unsigned long svpfn;
900 	unsigned long start_vaddr;
901 	unsigned long end_vaddr;
902 	int copied = 0;
903 
904 	if (!task)
905 		goto out;
906 
907 	ret = -EINVAL;
908 	/* file position must be aligned */
909 	if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
910 		goto out_task;
911 
912 	ret = 0;
913 	if (!count)
914 		goto out_task;
915 
916 	pm.len = PM_ENTRY_BYTES * (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
917 	pm.buffer = kmalloc(pm.len, GFP_TEMPORARY);
918 	ret = -ENOMEM;
919 	if (!pm.buffer)
920 		goto out_task;
921 
922 	mm = mm_for_maps(task);
923 	ret = PTR_ERR(mm);
924 	if (!mm || IS_ERR(mm))
925 		goto out_free;
926 
927 	pagemap_walk.pmd_entry = pagemap_pte_range;
928 	pagemap_walk.pte_hole = pagemap_pte_hole;
929 #ifdef CONFIG_HUGETLB_PAGE
930 	pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
931 #endif
932 	pagemap_walk.mm = mm;
933 	pagemap_walk.private = &pm;
934 
935 	src = *ppos;
936 	svpfn = src / PM_ENTRY_BYTES;
937 	start_vaddr = svpfn << PAGE_SHIFT;
938 	end_vaddr = TASK_SIZE_OF(task);
939 
940 	/* watch out for wraparound */
941 	if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
942 		start_vaddr = end_vaddr;
943 
944 	/*
945 	 * The odds are that this will stop walking way
946 	 * before end_vaddr, because the length of the
947 	 * user buffer is tracked in "pm", and the walk
948 	 * will stop when we hit the end of the buffer.
949 	 */
950 	ret = 0;
951 	while (count && (start_vaddr < end_vaddr)) {
952 		int len;
953 		unsigned long end;
954 
955 		pm.pos = 0;
956 		end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
957 		/* overflow ? */
958 		if (end < start_vaddr || end > end_vaddr)
959 			end = end_vaddr;
960 		down_read(&mm->mmap_sem);
961 		ret = walk_page_range(start_vaddr, end, &pagemap_walk);
962 		up_read(&mm->mmap_sem);
963 		start_vaddr = end;
964 
965 		len = min(count, PM_ENTRY_BYTES * pm.pos);
966 		if (copy_to_user(buf, pm.buffer, len)) {
967 			ret = -EFAULT;
968 			goto out_mm;
969 		}
970 		copied += len;
971 		buf += len;
972 		count -= len;
973 	}
974 	*ppos += copied;
975 	if (!ret || ret == PM_END_OF_BUFFER)
976 		ret = copied;
977 
978 out_mm:
979 	mmput(mm);
980 out_free:
981 	kfree(pm.buffer);
982 out_task:
983 	put_task_struct(task);
984 out:
985 	return ret;
986 }
987 
988 const struct file_operations proc_pagemap_operations = {
989 	.llseek		= mem_lseek, /* borrow this */
990 	.read		= pagemap_read,
991 };
992 #endif /* CONFIG_PROC_PAGE_MONITOR */
993 
994 #ifdef CONFIG_NUMA
995 
996 struct numa_maps {
997 	struct vm_area_struct *vma;
998 	unsigned long pages;
999 	unsigned long anon;
1000 	unsigned long active;
1001 	unsigned long writeback;
1002 	unsigned long mapcount_max;
1003 	unsigned long dirty;
1004 	unsigned long swapcache;
1005 	unsigned long node[MAX_NUMNODES];
1006 };
1007 
1008 struct numa_maps_private {
1009 	struct proc_maps_private proc_maps;
1010 	struct numa_maps md;
1011 };
1012 
1013 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1014 			unsigned long nr_pages)
1015 {
1016 	int count = page_mapcount(page);
1017 
1018 	md->pages += nr_pages;
1019 	if (pte_dirty || PageDirty(page))
1020 		md->dirty += nr_pages;
1021 
1022 	if (PageSwapCache(page))
1023 		md->swapcache += nr_pages;
1024 
1025 	if (PageActive(page) || PageUnevictable(page))
1026 		md->active += nr_pages;
1027 
1028 	if (PageWriteback(page))
1029 		md->writeback += nr_pages;
1030 
1031 	if (PageAnon(page))
1032 		md->anon += nr_pages;
1033 
1034 	if (count > md->mapcount_max)
1035 		md->mapcount_max = count;
1036 
1037 	md->node[page_to_nid(page)] += nr_pages;
1038 }
1039 
1040 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1041 		unsigned long addr)
1042 {
1043 	struct page *page;
1044 	int nid;
1045 
1046 	if (!pte_present(pte))
1047 		return NULL;
1048 
1049 	page = vm_normal_page(vma, addr, pte);
1050 	if (!page)
1051 		return NULL;
1052 
1053 	if (PageReserved(page))
1054 		return NULL;
1055 
1056 	nid = page_to_nid(page);
1057 	if (!node_isset(nid, node_states[N_HIGH_MEMORY]))
1058 		return NULL;
1059 
1060 	return page;
1061 }
1062 
1063 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1064 		unsigned long end, struct mm_walk *walk)
1065 {
1066 	struct numa_maps *md;
1067 	spinlock_t *ptl;
1068 	pte_t *orig_pte;
1069 	pte_t *pte;
1070 
1071 	md = walk->private;
1072 
1073 	if (pmd_trans_huge_lock(pmd, md->vma) == 1) {
1074 		pte_t huge_pte = *(pte_t *)pmd;
1075 		struct page *page;
1076 
1077 		page = can_gather_numa_stats(huge_pte, md->vma, addr);
1078 		if (page)
1079 			gather_stats(page, md, pte_dirty(huge_pte),
1080 				     HPAGE_PMD_SIZE/PAGE_SIZE);
1081 		spin_unlock(&walk->mm->page_table_lock);
1082 		return 0;
1083 	}
1084 
1085 	if (pmd_trans_unstable(pmd))
1086 		return 0;
1087 	orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1088 	do {
1089 		struct page *page = can_gather_numa_stats(*pte, md->vma, addr);
1090 		if (!page)
1091 			continue;
1092 		gather_stats(page, md, pte_dirty(*pte), 1);
1093 
1094 	} while (pte++, addr += PAGE_SIZE, addr != end);
1095 	pte_unmap_unlock(orig_pte, ptl);
1096 	return 0;
1097 }
1098 #ifdef CONFIG_HUGETLB_PAGE
1099 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
1100 		unsigned long addr, unsigned long end, struct mm_walk *walk)
1101 {
1102 	struct numa_maps *md;
1103 	struct page *page;
1104 
1105 	if (pte_none(*pte))
1106 		return 0;
1107 
1108 	page = pte_page(*pte);
1109 	if (!page)
1110 		return 0;
1111 
1112 	md = walk->private;
1113 	gather_stats(page, md, pte_dirty(*pte), 1);
1114 	return 0;
1115 }
1116 
1117 #else
1118 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
1119 		unsigned long addr, unsigned long end, struct mm_walk *walk)
1120 {
1121 	return 0;
1122 }
1123 #endif
1124 
1125 /*
1126  * Display pages allocated per node and memory policy via /proc.
1127  */
1128 static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1129 {
1130 	struct numa_maps_private *numa_priv = m->private;
1131 	struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1132 	struct vm_area_struct *vma = v;
1133 	struct numa_maps *md = &numa_priv->md;
1134 	struct file *file = vma->vm_file;
1135 	struct mm_struct *mm = vma->vm_mm;
1136 	struct mm_walk walk = {};
1137 	struct mempolicy *pol;
1138 	int n;
1139 	char buffer[50];
1140 
1141 	if (!mm)
1142 		return 0;
1143 
1144 	/* Ensure we start with an empty set of numa_maps statistics. */
1145 	memset(md, 0, sizeof(*md));
1146 
1147 	md->vma = vma;
1148 
1149 	walk.hugetlb_entry = gather_hugetbl_stats;
1150 	walk.pmd_entry = gather_pte_stats;
1151 	walk.private = md;
1152 	walk.mm = mm;
1153 
1154 	pol = get_vma_policy(proc_priv->task, vma, vma->vm_start);
1155 	mpol_to_str(buffer, sizeof(buffer), pol, 0);
1156 	mpol_cond_put(pol);
1157 
1158 	seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1159 
1160 	if (file) {
1161 		seq_printf(m, " file=");
1162 		seq_path(m, &file->f_path, "\n\t= ");
1163 	} else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1164 		seq_printf(m, " heap");
1165 	} else {
1166 		pid_t tid = vm_is_stack(proc_priv->task, vma, is_pid);
1167 		if (tid != 0) {
1168 			/*
1169 			 * Thread stack in /proc/PID/task/TID/maps or
1170 			 * the main process stack.
1171 			 */
1172 			if (!is_pid || (vma->vm_start <= mm->start_stack &&
1173 			    vma->vm_end >= mm->start_stack))
1174 				seq_printf(m, " stack");
1175 			else
1176 				seq_printf(m, " stack:%d", tid);
1177 		}
1178 	}
1179 
1180 	if (is_vm_hugetlb_page(vma))
1181 		seq_printf(m, " huge");
1182 
1183 	walk_page_range(vma->vm_start, vma->vm_end, &walk);
1184 
1185 	if (!md->pages)
1186 		goto out;
1187 
1188 	if (md->anon)
1189 		seq_printf(m, " anon=%lu", md->anon);
1190 
1191 	if (md->dirty)
1192 		seq_printf(m, " dirty=%lu", md->dirty);
1193 
1194 	if (md->pages != md->anon && md->pages != md->dirty)
1195 		seq_printf(m, " mapped=%lu", md->pages);
1196 
1197 	if (md->mapcount_max > 1)
1198 		seq_printf(m, " mapmax=%lu", md->mapcount_max);
1199 
1200 	if (md->swapcache)
1201 		seq_printf(m, " swapcache=%lu", md->swapcache);
1202 
1203 	if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1204 		seq_printf(m, " active=%lu", md->active);
1205 
1206 	if (md->writeback)
1207 		seq_printf(m, " writeback=%lu", md->writeback);
1208 
1209 	for_each_node_state(n, N_HIGH_MEMORY)
1210 		if (md->node[n])
1211 			seq_printf(m, " N%d=%lu", n, md->node[n]);
1212 out:
1213 	seq_putc(m, '\n');
1214 
1215 	if (m->count < m->size)
1216 		m->version = (vma != proc_priv->tail_vma) ? vma->vm_start : 0;
1217 	return 0;
1218 }
1219 
1220 static int show_pid_numa_map(struct seq_file *m, void *v)
1221 {
1222 	return show_numa_map(m, v, 1);
1223 }
1224 
1225 static int show_tid_numa_map(struct seq_file *m, void *v)
1226 {
1227 	return show_numa_map(m, v, 0);
1228 }
1229 
1230 static const struct seq_operations proc_pid_numa_maps_op = {
1231 	.start  = m_start,
1232 	.next   = m_next,
1233 	.stop   = m_stop,
1234 	.show   = show_pid_numa_map,
1235 };
1236 
1237 static const struct seq_operations proc_tid_numa_maps_op = {
1238 	.start  = m_start,
1239 	.next   = m_next,
1240 	.stop   = m_stop,
1241 	.show   = show_tid_numa_map,
1242 };
1243 
1244 static int numa_maps_open(struct inode *inode, struct file *file,
1245 			  const struct seq_operations *ops)
1246 {
1247 	struct numa_maps_private *priv;
1248 	int ret = -ENOMEM;
1249 	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
1250 	if (priv) {
1251 		priv->proc_maps.pid = proc_pid(inode);
1252 		ret = seq_open(file, ops);
1253 		if (!ret) {
1254 			struct seq_file *m = file->private_data;
1255 			m->private = priv;
1256 		} else {
1257 			kfree(priv);
1258 		}
1259 	}
1260 	return ret;
1261 }
1262 
1263 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1264 {
1265 	return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1266 }
1267 
1268 static int tid_numa_maps_open(struct inode *inode, struct file *file)
1269 {
1270 	return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1271 }
1272 
1273 const struct file_operations proc_pid_numa_maps_operations = {
1274 	.open		= pid_numa_maps_open,
1275 	.read		= seq_read,
1276 	.llseek		= seq_lseek,
1277 	.release	= seq_release_private,
1278 };
1279 
1280 const struct file_operations proc_tid_numa_maps_operations = {
1281 	.open		= tid_numa_maps_open,
1282 	.read		= seq_read,
1283 	.llseek		= seq_lseek,
1284 	.release	= seq_release_private,
1285 };
1286 #endif /* CONFIG_NUMA */
1287