xref: /linux/fs/proc/task_mmu.c (revision d39d0ed196aa1685bb24771e92f78633c66ac9cb)
1 #include <linux/mm.h>
2 #include <linux/hugetlb.h>
3 #include <linux/mount.h>
4 #include <linux/seq_file.h>
5 #include <linux/highmem.h>
6 #include <linux/ptrace.h>
7 #include <linux/slab.h>
8 #include <linux/pagemap.h>
9 #include <linux/mempolicy.h>
10 #include <linux/swap.h>
11 #include <linux/swapops.h>
12 
13 #include <asm/elf.h>
14 #include <asm/uaccess.h>
15 #include <asm/tlbflush.h>
16 #include "internal.h"
17 
18 void task_mem(struct seq_file *m, struct mm_struct *mm)
19 {
20 	unsigned long data, text, lib, swap;
21 	unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
22 
23 	/*
24 	 * Note: to minimize their overhead, mm maintains hiwater_vm and
25 	 * hiwater_rss only when about to *lower* total_vm or rss.  Any
26 	 * collector of these hiwater stats must therefore get total_vm
27 	 * and rss too, which will usually be the higher.  Barriers? not
28 	 * worth the effort, such snapshots can always be inconsistent.
29 	 */
30 	hiwater_vm = total_vm = mm->total_vm;
31 	if (hiwater_vm < mm->hiwater_vm)
32 		hiwater_vm = mm->hiwater_vm;
33 	hiwater_rss = total_rss = get_mm_rss(mm);
34 	if (hiwater_rss < mm->hiwater_rss)
35 		hiwater_rss = mm->hiwater_rss;
36 
37 	data = mm->total_vm - mm->shared_vm - mm->stack_vm;
38 	text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
39 	lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
40 	swap = get_mm_counter(mm, MM_SWAPENTS);
41 	seq_printf(m,
42 		"VmPeak:\t%8lu kB\n"
43 		"VmSize:\t%8lu kB\n"
44 		"VmLck:\t%8lu kB\n"
45 		"VmHWM:\t%8lu kB\n"
46 		"VmRSS:\t%8lu kB\n"
47 		"VmData:\t%8lu kB\n"
48 		"VmStk:\t%8lu kB\n"
49 		"VmExe:\t%8lu kB\n"
50 		"VmLib:\t%8lu kB\n"
51 		"VmPTE:\t%8lu kB\n"
52 		"VmSwap:\t%8lu kB\n",
53 		hiwater_vm << (PAGE_SHIFT-10),
54 		(total_vm - mm->reserved_vm) << (PAGE_SHIFT-10),
55 		mm->locked_vm << (PAGE_SHIFT-10),
56 		hiwater_rss << (PAGE_SHIFT-10),
57 		total_rss << (PAGE_SHIFT-10),
58 		data << (PAGE_SHIFT-10),
59 		mm->stack_vm << (PAGE_SHIFT-10), text, lib,
60 		(PTRS_PER_PTE*sizeof(pte_t)*mm->nr_ptes) >> 10,
61 		swap << (PAGE_SHIFT-10));
62 }
63 
64 unsigned long task_vsize(struct mm_struct *mm)
65 {
66 	return PAGE_SIZE * mm->total_vm;
67 }
68 
69 int task_statm(struct mm_struct *mm, int *shared, int *text,
70 	       int *data, int *resident)
71 {
72 	*shared = get_mm_counter(mm, MM_FILEPAGES);
73 	*text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
74 								>> PAGE_SHIFT;
75 	*data = mm->total_vm - mm->shared_vm;
76 	*resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
77 	return mm->total_vm;
78 }
79 
80 static void pad_len_spaces(struct seq_file *m, int len)
81 {
82 	len = 25 + sizeof(void*) * 6 - len;
83 	if (len < 1)
84 		len = 1;
85 	seq_printf(m, "%*c", len, ' ');
86 }
87 
88 static void vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma)
89 {
90 	if (vma && vma != priv->tail_vma) {
91 		struct mm_struct *mm = vma->vm_mm;
92 		up_read(&mm->mmap_sem);
93 		mmput(mm);
94 	}
95 }
96 
97 static void *m_start(struct seq_file *m, loff_t *pos)
98 {
99 	struct proc_maps_private *priv = m->private;
100 	unsigned long last_addr = m->version;
101 	struct mm_struct *mm;
102 	struct vm_area_struct *vma, *tail_vma = NULL;
103 	loff_t l = *pos;
104 
105 	/* Clear the per syscall fields in priv */
106 	priv->task = NULL;
107 	priv->tail_vma = NULL;
108 
109 	/*
110 	 * We remember last_addr rather than next_addr to hit with
111 	 * mmap_cache most of the time. We have zero last_addr at
112 	 * the beginning and also after lseek. We will have -1 last_addr
113 	 * after the end of the vmas.
114 	 */
115 
116 	if (last_addr == -1UL)
117 		return NULL;
118 
119 	priv->task = get_pid_task(priv->pid, PIDTYPE_PID);
120 	if (!priv->task)
121 		return NULL;
122 
123 	mm = mm_for_maps(priv->task);
124 	if (!mm)
125 		return NULL;
126 	down_read(&mm->mmap_sem);
127 
128 	tail_vma = get_gate_vma(priv->task);
129 	priv->tail_vma = tail_vma;
130 
131 	/* Start with last addr hint */
132 	vma = find_vma(mm, last_addr);
133 	if (last_addr && vma) {
134 		vma = vma->vm_next;
135 		goto out;
136 	}
137 
138 	/*
139 	 * Check the vma index is within the range and do
140 	 * sequential scan until m_index.
141 	 */
142 	vma = NULL;
143 	if ((unsigned long)l < mm->map_count) {
144 		vma = mm->mmap;
145 		while (l-- && vma)
146 			vma = vma->vm_next;
147 		goto out;
148 	}
149 
150 	if (l != mm->map_count)
151 		tail_vma = NULL; /* After gate vma */
152 
153 out:
154 	if (vma)
155 		return vma;
156 
157 	/* End of vmas has been reached */
158 	m->version = (tail_vma != NULL)? 0: -1UL;
159 	up_read(&mm->mmap_sem);
160 	mmput(mm);
161 	return tail_vma;
162 }
163 
164 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
165 {
166 	struct proc_maps_private *priv = m->private;
167 	struct vm_area_struct *vma = v;
168 	struct vm_area_struct *tail_vma = priv->tail_vma;
169 
170 	(*pos)++;
171 	if (vma && (vma != tail_vma) && vma->vm_next)
172 		return vma->vm_next;
173 	vma_stop(priv, vma);
174 	return (vma != tail_vma)? tail_vma: NULL;
175 }
176 
177 static void m_stop(struct seq_file *m, void *v)
178 {
179 	struct proc_maps_private *priv = m->private;
180 	struct vm_area_struct *vma = v;
181 
182 	vma_stop(priv, vma);
183 	if (priv->task)
184 		put_task_struct(priv->task);
185 }
186 
187 static int do_maps_open(struct inode *inode, struct file *file,
188 			const struct seq_operations *ops)
189 {
190 	struct proc_maps_private *priv;
191 	int ret = -ENOMEM;
192 	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
193 	if (priv) {
194 		priv->pid = proc_pid(inode);
195 		ret = seq_open(file, ops);
196 		if (!ret) {
197 			struct seq_file *m = file->private_data;
198 			m->private = priv;
199 		} else {
200 			kfree(priv);
201 		}
202 	}
203 	return ret;
204 }
205 
206 static void show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
207 {
208 	struct mm_struct *mm = vma->vm_mm;
209 	struct file *file = vma->vm_file;
210 	int flags = vma->vm_flags;
211 	unsigned long ino = 0;
212 	unsigned long long pgoff = 0;
213 	unsigned long start;
214 	dev_t dev = 0;
215 	int len;
216 
217 	if (file) {
218 		struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
219 		dev = inode->i_sb->s_dev;
220 		ino = inode->i_ino;
221 		pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
222 	}
223 
224 	/* We don't show the stack guard page in /proc/maps */
225 	start = vma->vm_start;
226 	if (vma->vm_flags & VM_GROWSDOWN)
227 		start += PAGE_SIZE;
228 
229 	seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu %n",
230 			start,
231 			vma->vm_end,
232 			flags & VM_READ ? 'r' : '-',
233 			flags & VM_WRITE ? 'w' : '-',
234 			flags & VM_EXEC ? 'x' : '-',
235 			flags & VM_MAYSHARE ? 's' : 'p',
236 			pgoff,
237 			MAJOR(dev), MINOR(dev), ino, &len);
238 
239 	/*
240 	 * Print the dentry name for named mappings, and a
241 	 * special [heap] marker for the heap:
242 	 */
243 	if (file) {
244 		pad_len_spaces(m, len);
245 		seq_path(m, &file->f_path, "\n");
246 	} else {
247 		const char *name = arch_vma_name(vma);
248 		if (!name) {
249 			if (mm) {
250 				if (vma->vm_start <= mm->start_brk &&
251 						vma->vm_end >= mm->brk) {
252 					name = "[heap]";
253 				} else if (vma->vm_start <= mm->start_stack &&
254 					   vma->vm_end >= mm->start_stack) {
255 					name = "[stack]";
256 				}
257 			} else {
258 				name = "[vdso]";
259 			}
260 		}
261 		if (name) {
262 			pad_len_spaces(m, len);
263 			seq_puts(m, name);
264 		}
265 	}
266 	seq_putc(m, '\n');
267 }
268 
269 static int show_map(struct seq_file *m, void *v)
270 {
271 	struct vm_area_struct *vma = v;
272 	struct proc_maps_private *priv = m->private;
273 	struct task_struct *task = priv->task;
274 
275 	show_map_vma(m, vma);
276 
277 	if (m->count < m->size)  /* vma is copied successfully */
278 		m->version = (vma != get_gate_vma(task))? vma->vm_start: 0;
279 	return 0;
280 }
281 
282 static const struct seq_operations proc_pid_maps_op = {
283 	.start	= m_start,
284 	.next	= m_next,
285 	.stop	= m_stop,
286 	.show	= show_map
287 };
288 
289 static int maps_open(struct inode *inode, struct file *file)
290 {
291 	return do_maps_open(inode, file, &proc_pid_maps_op);
292 }
293 
294 const struct file_operations proc_maps_operations = {
295 	.open		= maps_open,
296 	.read		= seq_read,
297 	.llseek		= seq_lseek,
298 	.release	= seq_release_private,
299 };
300 
301 /*
302  * Proportional Set Size(PSS): my share of RSS.
303  *
304  * PSS of a process is the count of pages it has in memory, where each
305  * page is divided by the number of processes sharing it.  So if a
306  * process has 1000 pages all to itself, and 1000 shared with one other
307  * process, its PSS will be 1500.
308  *
309  * To keep (accumulated) division errors low, we adopt a 64bit
310  * fixed-point pss counter to minimize division errors. So (pss >>
311  * PSS_SHIFT) would be the real byte count.
312  *
313  * A shift of 12 before division means (assuming 4K page size):
314  * 	- 1M 3-user-pages add up to 8KB errors;
315  * 	- supports mapcount up to 2^24, or 16M;
316  * 	- supports PSS up to 2^52 bytes, or 4PB.
317  */
318 #define PSS_SHIFT 12
319 
320 #ifdef CONFIG_PROC_PAGE_MONITOR
321 struct mem_size_stats {
322 	struct vm_area_struct *vma;
323 	unsigned long resident;
324 	unsigned long shared_clean;
325 	unsigned long shared_dirty;
326 	unsigned long private_clean;
327 	unsigned long private_dirty;
328 	unsigned long referenced;
329 	unsigned long swap;
330 	u64 pss;
331 };
332 
333 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
334 			   struct mm_walk *walk)
335 {
336 	struct mem_size_stats *mss = walk->private;
337 	struct vm_area_struct *vma = mss->vma;
338 	pte_t *pte, ptent;
339 	spinlock_t *ptl;
340 	struct page *page;
341 	int mapcount;
342 
343 	pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
344 	for (; addr != end; pte++, addr += PAGE_SIZE) {
345 		ptent = *pte;
346 
347 		if (is_swap_pte(ptent)) {
348 			mss->swap += PAGE_SIZE;
349 			continue;
350 		}
351 
352 		if (!pte_present(ptent))
353 			continue;
354 
355 		page = vm_normal_page(vma, addr, ptent);
356 		if (!page)
357 			continue;
358 
359 		mss->resident += PAGE_SIZE;
360 		/* Accumulate the size in pages that have been accessed. */
361 		if (pte_young(ptent) || PageReferenced(page))
362 			mss->referenced += PAGE_SIZE;
363 		mapcount = page_mapcount(page);
364 		if (mapcount >= 2) {
365 			if (pte_dirty(ptent))
366 				mss->shared_dirty += PAGE_SIZE;
367 			else
368 				mss->shared_clean += PAGE_SIZE;
369 			mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount;
370 		} else {
371 			if (pte_dirty(ptent))
372 				mss->private_dirty += PAGE_SIZE;
373 			else
374 				mss->private_clean += PAGE_SIZE;
375 			mss->pss += (PAGE_SIZE << PSS_SHIFT);
376 		}
377 	}
378 	pte_unmap_unlock(pte - 1, ptl);
379 	cond_resched();
380 	return 0;
381 }
382 
383 static int show_smap(struct seq_file *m, void *v)
384 {
385 	struct proc_maps_private *priv = m->private;
386 	struct task_struct *task = priv->task;
387 	struct vm_area_struct *vma = v;
388 	struct mem_size_stats mss;
389 	struct mm_walk smaps_walk = {
390 		.pmd_entry = smaps_pte_range,
391 		.mm = vma->vm_mm,
392 		.private = &mss,
393 	};
394 
395 	memset(&mss, 0, sizeof mss);
396 	mss.vma = vma;
397 	/* mmap_sem is held in m_start */
398 	if (vma->vm_mm && !is_vm_hugetlb_page(vma))
399 		walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk);
400 
401 	show_map_vma(m, vma);
402 
403 	seq_printf(m,
404 		   "Size:           %8lu kB\n"
405 		   "Rss:            %8lu kB\n"
406 		   "Pss:            %8lu kB\n"
407 		   "Shared_Clean:   %8lu kB\n"
408 		   "Shared_Dirty:   %8lu kB\n"
409 		   "Private_Clean:  %8lu kB\n"
410 		   "Private_Dirty:  %8lu kB\n"
411 		   "Referenced:     %8lu kB\n"
412 		   "Swap:           %8lu kB\n"
413 		   "KernelPageSize: %8lu kB\n"
414 		   "MMUPageSize:    %8lu kB\n",
415 		   (vma->vm_end - vma->vm_start) >> 10,
416 		   mss.resident >> 10,
417 		   (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
418 		   mss.shared_clean  >> 10,
419 		   mss.shared_dirty  >> 10,
420 		   mss.private_clean >> 10,
421 		   mss.private_dirty >> 10,
422 		   mss.referenced >> 10,
423 		   mss.swap >> 10,
424 		   vma_kernel_pagesize(vma) >> 10,
425 		   vma_mmu_pagesize(vma) >> 10);
426 
427 	if (m->count < m->size)  /* vma is copied successfully */
428 		m->version = (vma != get_gate_vma(task)) ? vma->vm_start : 0;
429 	return 0;
430 }
431 
432 static const struct seq_operations proc_pid_smaps_op = {
433 	.start	= m_start,
434 	.next	= m_next,
435 	.stop	= m_stop,
436 	.show	= show_smap
437 };
438 
439 static int smaps_open(struct inode *inode, struct file *file)
440 {
441 	return do_maps_open(inode, file, &proc_pid_smaps_op);
442 }
443 
444 const struct file_operations proc_smaps_operations = {
445 	.open		= smaps_open,
446 	.read		= seq_read,
447 	.llseek		= seq_lseek,
448 	.release	= seq_release_private,
449 };
450 
451 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
452 				unsigned long end, struct mm_walk *walk)
453 {
454 	struct vm_area_struct *vma = walk->private;
455 	pte_t *pte, ptent;
456 	spinlock_t *ptl;
457 	struct page *page;
458 
459 	pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
460 	for (; addr != end; pte++, addr += PAGE_SIZE) {
461 		ptent = *pte;
462 		if (!pte_present(ptent))
463 			continue;
464 
465 		page = vm_normal_page(vma, addr, ptent);
466 		if (!page)
467 			continue;
468 
469 		/* Clear accessed and referenced bits. */
470 		ptep_test_and_clear_young(vma, addr, pte);
471 		ClearPageReferenced(page);
472 	}
473 	pte_unmap_unlock(pte - 1, ptl);
474 	cond_resched();
475 	return 0;
476 }
477 
478 #define CLEAR_REFS_ALL 1
479 #define CLEAR_REFS_ANON 2
480 #define CLEAR_REFS_MAPPED 3
481 
482 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
483 				size_t count, loff_t *ppos)
484 {
485 	struct task_struct *task;
486 	char buffer[PROC_NUMBUF];
487 	struct mm_struct *mm;
488 	struct vm_area_struct *vma;
489 	long type;
490 
491 	memset(buffer, 0, sizeof(buffer));
492 	if (count > sizeof(buffer) - 1)
493 		count = sizeof(buffer) - 1;
494 	if (copy_from_user(buffer, buf, count))
495 		return -EFAULT;
496 	if (strict_strtol(strstrip(buffer), 10, &type))
497 		return -EINVAL;
498 	if (type < CLEAR_REFS_ALL || type > CLEAR_REFS_MAPPED)
499 		return -EINVAL;
500 	task = get_proc_task(file->f_path.dentry->d_inode);
501 	if (!task)
502 		return -ESRCH;
503 	mm = get_task_mm(task);
504 	if (mm) {
505 		struct mm_walk clear_refs_walk = {
506 			.pmd_entry = clear_refs_pte_range,
507 			.mm = mm,
508 		};
509 		down_read(&mm->mmap_sem);
510 		for (vma = mm->mmap; vma; vma = vma->vm_next) {
511 			clear_refs_walk.private = vma;
512 			if (is_vm_hugetlb_page(vma))
513 				continue;
514 			/*
515 			 * Writing 1 to /proc/pid/clear_refs affects all pages.
516 			 *
517 			 * Writing 2 to /proc/pid/clear_refs only affects
518 			 * Anonymous pages.
519 			 *
520 			 * Writing 3 to /proc/pid/clear_refs only affects file
521 			 * mapped pages.
522 			 */
523 			if (type == CLEAR_REFS_ANON && vma->vm_file)
524 				continue;
525 			if (type == CLEAR_REFS_MAPPED && !vma->vm_file)
526 				continue;
527 			walk_page_range(vma->vm_start, vma->vm_end,
528 					&clear_refs_walk);
529 		}
530 		flush_tlb_mm(mm);
531 		up_read(&mm->mmap_sem);
532 		mmput(mm);
533 	}
534 	put_task_struct(task);
535 
536 	return count;
537 }
538 
539 const struct file_operations proc_clear_refs_operations = {
540 	.write		= clear_refs_write,
541 };
542 
543 struct pagemapread {
544 	int pos, len;
545 	u64 *buffer;
546 };
547 
548 #define PM_ENTRY_BYTES      sizeof(u64)
549 #define PM_STATUS_BITS      3
550 #define PM_STATUS_OFFSET    (64 - PM_STATUS_BITS)
551 #define PM_STATUS_MASK      (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
552 #define PM_STATUS(nr)       (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
553 #define PM_PSHIFT_BITS      6
554 #define PM_PSHIFT_OFFSET    (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
555 #define PM_PSHIFT_MASK      (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
556 #define PM_PSHIFT(x)        (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
557 #define PM_PFRAME_MASK      ((1LL << PM_PSHIFT_OFFSET) - 1)
558 #define PM_PFRAME(x)        ((x) & PM_PFRAME_MASK)
559 
560 #define PM_PRESENT          PM_STATUS(4LL)
561 #define PM_SWAP             PM_STATUS(2LL)
562 #define PM_NOT_PRESENT      PM_PSHIFT(PAGE_SHIFT)
563 #define PM_END_OF_BUFFER    1
564 
565 static int add_to_pagemap(unsigned long addr, u64 pfn,
566 			  struct pagemapread *pm)
567 {
568 	pm->buffer[pm->pos++] = pfn;
569 	if (pm->pos >= pm->len)
570 		return PM_END_OF_BUFFER;
571 	return 0;
572 }
573 
574 static int pagemap_pte_hole(unsigned long start, unsigned long end,
575 				struct mm_walk *walk)
576 {
577 	struct pagemapread *pm = walk->private;
578 	unsigned long addr;
579 	int err = 0;
580 	for (addr = start; addr < end; addr += PAGE_SIZE) {
581 		err = add_to_pagemap(addr, PM_NOT_PRESENT, pm);
582 		if (err)
583 			break;
584 	}
585 	return err;
586 }
587 
588 static u64 swap_pte_to_pagemap_entry(pte_t pte)
589 {
590 	swp_entry_t e = pte_to_swp_entry(pte);
591 	return swp_type(e) | (swp_offset(e) << MAX_SWAPFILES_SHIFT);
592 }
593 
594 static u64 pte_to_pagemap_entry(pte_t pte)
595 {
596 	u64 pme = 0;
597 	if (is_swap_pte(pte))
598 		pme = PM_PFRAME(swap_pte_to_pagemap_entry(pte))
599 			| PM_PSHIFT(PAGE_SHIFT) | PM_SWAP;
600 	else if (pte_present(pte))
601 		pme = PM_PFRAME(pte_pfn(pte))
602 			| PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
603 	return pme;
604 }
605 
606 static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
607 			     struct mm_walk *walk)
608 {
609 	struct vm_area_struct *vma;
610 	struct pagemapread *pm = walk->private;
611 	pte_t *pte;
612 	int err = 0;
613 
614 	/* find the first VMA at or above 'addr' */
615 	vma = find_vma(walk->mm, addr);
616 	for (; addr != end; addr += PAGE_SIZE) {
617 		u64 pfn = PM_NOT_PRESENT;
618 
619 		/* check to see if we've left 'vma' behind
620 		 * and need a new, higher one */
621 		if (vma && (addr >= vma->vm_end))
622 			vma = find_vma(walk->mm, addr);
623 
624 		/* check that 'vma' actually covers this address,
625 		 * and that it isn't a huge page vma */
626 		if (vma && (vma->vm_start <= addr) &&
627 		    !is_vm_hugetlb_page(vma)) {
628 			pte = pte_offset_map(pmd, addr);
629 			pfn = pte_to_pagemap_entry(*pte);
630 			/* unmap before userspace copy */
631 			pte_unmap(pte);
632 		}
633 		err = add_to_pagemap(addr, pfn, pm);
634 		if (err)
635 			return err;
636 	}
637 
638 	cond_resched();
639 
640 	return err;
641 }
642 
643 #ifdef CONFIG_HUGETLB_PAGE
644 static u64 huge_pte_to_pagemap_entry(pte_t pte, int offset)
645 {
646 	u64 pme = 0;
647 	if (pte_present(pte))
648 		pme = PM_PFRAME(pte_pfn(pte) + offset)
649 			| PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
650 	return pme;
651 }
652 
653 /* This function walks within one hugetlb entry in the single call */
654 static int pagemap_hugetlb_range(pte_t *pte, unsigned long hmask,
655 				 unsigned long addr, unsigned long end,
656 				 struct mm_walk *walk)
657 {
658 	struct pagemapread *pm = walk->private;
659 	int err = 0;
660 	u64 pfn;
661 
662 	for (; addr != end; addr += PAGE_SIZE) {
663 		int offset = (addr & ~hmask) >> PAGE_SHIFT;
664 		pfn = huge_pte_to_pagemap_entry(*pte, offset);
665 		err = add_to_pagemap(addr, pfn, pm);
666 		if (err)
667 			return err;
668 	}
669 
670 	cond_resched();
671 
672 	return err;
673 }
674 #endif /* HUGETLB_PAGE */
675 
676 /*
677  * /proc/pid/pagemap - an array mapping virtual pages to pfns
678  *
679  * For each page in the address space, this file contains one 64-bit entry
680  * consisting of the following:
681  *
682  * Bits 0-55  page frame number (PFN) if present
683  * Bits 0-4   swap type if swapped
684  * Bits 5-55  swap offset if swapped
685  * Bits 55-60 page shift (page size = 1<<page shift)
686  * Bit  61    reserved for future use
687  * Bit  62    page swapped
688  * Bit  63    page present
689  *
690  * If the page is not present but in swap, then the PFN contains an
691  * encoding of the swap file number and the page's offset into the
692  * swap. Unmapped pages return a null PFN. This allows determining
693  * precisely which pages are mapped (or in swap) and comparing mapped
694  * pages between processes.
695  *
696  * Efficient users of this interface will use /proc/pid/maps to
697  * determine which areas of memory are actually mapped and llseek to
698  * skip over unmapped regions.
699  */
700 #define PAGEMAP_WALK_SIZE	(PMD_SIZE)
701 static ssize_t pagemap_read(struct file *file, char __user *buf,
702 			    size_t count, loff_t *ppos)
703 {
704 	struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
705 	struct mm_struct *mm;
706 	struct pagemapread pm;
707 	int ret = -ESRCH;
708 	struct mm_walk pagemap_walk = {};
709 	unsigned long src;
710 	unsigned long svpfn;
711 	unsigned long start_vaddr;
712 	unsigned long end_vaddr;
713 	int copied = 0;
714 
715 	if (!task)
716 		goto out;
717 
718 	ret = -EACCES;
719 	if (!ptrace_may_access(task, PTRACE_MODE_READ))
720 		goto out_task;
721 
722 	ret = -EINVAL;
723 	/* file position must be aligned */
724 	if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
725 		goto out_task;
726 
727 	ret = 0;
728 
729 	if (!count)
730 		goto out_task;
731 
732 	mm = get_task_mm(task);
733 	if (!mm)
734 		goto out_task;
735 
736 	pm.len = PM_ENTRY_BYTES * (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
737 	pm.buffer = kmalloc(pm.len, GFP_TEMPORARY);
738 	ret = -ENOMEM;
739 	if (!pm.buffer)
740 		goto out_mm;
741 
742 	pagemap_walk.pmd_entry = pagemap_pte_range;
743 	pagemap_walk.pte_hole = pagemap_pte_hole;
744 #ifdef CONFIG_HUGETLB_PAGE
745 	pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
746 #endif
747 	pagemap_walk.mm = mm;
748 	pagemap_walk.private = &pm;
749 
750 	src = *ppos;
751 	svpfn = src / PM_ENTRY_BYTES;
752 	start_vaddr = svpfn << PAGE_SHIFT;
753 	end_vaddr = TASK_SIZE_OF(task);
754 
755 	/* watch out for wraparound */
756 	if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
757 		start_vaddr = end_vaddr;
758 
759 	/*
760 	 * The odds are that this will stop walking way
761 	 * before end_vaddr, because the length of the
762 	 * user buffer is tracked in "pm", and the walk
763 	 * will stop when we hit the end of the buffer.
764 	 */
765 	ret = 0;
766 	while (count && (start_vaddr < end_vaddr)) {
767 		int len;
768 		unsigned long end;
769 
770 		pm.pos = 0;
771 		end = start_vaddr + PAGEMAP_WALK_SIZE;
772 		/* overflow ? */
773 		if (end < start_vaddr || end > end_vaddr)
774 			end = end_vaddr;
775 		down_read(&mm->mmap_sem);
776 		ret = walk_page_range(start_vaddr, end, &pagemap_walk);
777 		up_read(&mm->mmap_sem);
778 		start_vaddr = end;
779 
780 		len = min(count, PM_ENTRY_BYTES * pm.pos);
781 		if (copy_to_user(buf, pm.buffer, len)) {
782 			ret = -EFAULT;
783 			goto out_free;
784 		}
785 		copied += len;
786 		buf += len;
787 		count -= len;
788 	}
789 	*ppos += copied;
790 	if (!ret || ret == PM_END_OF_BUFFER)
791 		ret = copied;
792 
793 out_free:
794 	kfree(pm.buffer);
795 out_mm:
796 	mmput(mm);
797 out_task:
798 	put_task_struct(task);
799 out:
800 	return ret;
801 }
802 
803 const struct file_operations proc_pagemap_operations = {
804 	.llseek		= mem_lseek, /* borrow this */
805 	.read		= pagemap_read,
806 };
807 #endif /* CONFIG_PROC_PAGE_MONITOR */
808 
809 #ifdef CONFIG_NUMA
810 extern int show_numa_map(struct seq_file *m, void *v);
811 
812 static const struct seq_operations proc_pid_numa_maps_op = {
813         .start  = m_start,
814         .next   = m_next,
815         .stop   = m_stop,
816         .show   = show_numa_map,
817 };
818 
819 static int numa_maps_open(struct inode *inode, struct file *file)
820 {
821 	return do_maps_open(inode, file, &proc_pid_numa_maps_op);
822 }
823 
824 const struct file_operations proc_numa_maps_operations = {
825 	.open		= numa_maps_open,
826 	.read		= seq_read,
827 	.llseek		= seq_lseek,
828 	.release	= seq_release_private,
829 };
830 #endif
831