xref: /linux/fs/proc/task_mmu.c (revision 0dd9ac63ce26ec87b080ca9c3e6efed33c23ace6)
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 	dev_t dev = 0;
214 	int len;
215 
216 	if (file) {
217 		struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
218 		dev = inode->i_sb->s_dev;
219 		ino = inode->i_ino;
220 		pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
221 	}
222 
223 	seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu %n",
224 			vma->vm_start,
225 			vma->vm_end,
226 			flags & VM_READ ? 'r' : '-',
227 			flags & VM_WRITE ? 'w' : '-',
228 			flags & VM_EXEC ? 'x' : '-',
229 			flags & VM_MAYSHARE ? 's' : 'p',
230 			pgoff,
231 			MAJOR(dev), MINOR(dev), ino, &len);
232 
233 	/*
234 	 * Print the dentry name for named mappings, and a
235 	 * special [heap] marker for the heap:
236 	 */
237 	if (file) {
238 		pad_len_spaces(m, len);
239 		seq_path(m, &file->f_path, "\n");
240 	} else {
241 		const char *name = arch_vma_name(vma);
242 		if (!name) {
243 			if (mm) {
244 				if (vma->vm_start <= mm->start_brk &&
245 						vma->vm_end >= mm->brk) {
246 					name = "[heap]";
247 				} else if (vma->vm_start <= mm->start_stack &&
248 					   vma->vm_end >= mm->start_stack) {
249 					name = "[stack]";
250 				}
251 			} else {
252 				name = "[vdso]";
253 			}
254 		}
255 		if (name) {
256 			pad_len_spaces(m, len);
257 			seq_puts(m, name);
258 		}
259 	}
260 	seq_putc(m, '\n');
261 }
262 
263 static int show_map(struct seq_file *m, void *v)
264 {
265 	struct vm_area_struct *vma = v;
266 	struct proc_maps_private *priv = m->private;
267 	struct task_struct *task = priv->task;
268 
269 	show_map_vma(m, vma);
270 
271 	if (m->count < m->size)  /* vma is copied successfully */
272 		m->version = (vma != get_gate_vma(task))? vma->vm_start: 0;
273 	return 0;
274 }
275 
276 static const struct seq_operations proc_pid_maps_op = {
277 	.start	= m_start,
278 	.next	= m_next,
279 	.stop	= m_stop,
280 	.show	= show_map
281 };
282 
283 static int maps_open(struct inode *inode, struct file *file)
284 {
285 	return do_maps_open(inode, file, &proc_pid_maps_op);
286 }
287 
288 const struct file_operations proc_maps_operations = {
289 	.open		= maps_open,
290 	.read		= seq_read,
291 	.llseek		= seq_lseek,
292 	.release	= seq_release_private,
293 };
294 
295 /*
296  * Proportional Set Size(PSS): my share of RSS.
297  *
298  * PSS of a process is the count of pages it has in memory, where each
299  * page is divided by the number of processes sharing it.  So if a
300  * process has 1000 pages all to itself, and 1000 shared with one other
301  * process, its PSS will be 1500.
302  *
303  * To keep (accumulated) division errors low, we adopt a 64bit
304  * fixed-point pss counter to minimize division errors. So (pss >>
305  * PSS_SHIFT) would be the real byte count.
306  *
307  * A shift of 12 before division means (assuming 4K page size):
308  * 	- 1M 3-user-pages add up to 8KB errors;
309  * 	- supports mapcount up to 2^24, or 16M;
310  * 	- supports PSS up to 2^52 bytes, or 4PB.
311  */
312 #define PSS_SHIFT 12
313 
314 #ifdef CONFIG_PROC_PAGE_MONITOR
315 struct mem_size_stats {
316 	struct vm_area_struct *vma;
317 	unsigned long resident;
318 	unsigned long shared_clean;
319 	unsigned long shared_dirty;
320 	unsigned long private_clean;
321 	unsigned long private_dirty;
322 	unsigned long referenced;
323 	unsigned long swap;
324 	u64 pss;
325 };
326 
327 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
328 			   struct mm_walk *walk)
329 {
330 	struct mem_size_stats *mss = walk->private;
331 	struct vm_area_struct *vma = mss->vma;
332 	pte_t *pte, ptent;
333 	spinlock_t *ptl;
334 	struct page *page;
335 	int mapcount;
336 
337 	pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
338 	for (; addr != end; pte++, addr += PAGE_SIZE) {
339 		ptent = *pte;
340 
341 		if (is_swap_pte(ptent)) {
342 			mss->swap += PAGE_SIZE;
343 			continue;
344 		}
345 
346 		if (!pte_present(ptent))
347 			continue;
348 
349 		page = vm_normal_page(vma, addr, ptent);
350 		if (!page)
351 			continue;
352 
353 		mss->resident += PAGE_SIZE;
354 		/* Accumulate the size in pages that have been accessed. */
355 		if (pte_young(ptent) || PageReferenced(page))
356 			mss->referenced += PAGE_SIZE;
357 		mapcount = page_mapcount(page);
358 		if (mapcount >= 2) {
359 			if (pte_dirty(ptent))
360 				mss->shared_dirty += PAGE_SIZE;
361 			else
362 				mss->shared_clean += PAGE_SIZE;
363 			mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount;
364 		} else {
365 			if (pte_dirty(ptent))
366 				mss->private_dirty += PAGE_SIZE;
367 			else
368 				mss->private_clean += PAGE_SIZE;
369 			mss->pss += (PAGE_SIZE << PSS_SHIFT);
370 		}
371 	}
372 	pte_unmap_unlock(pte - 1, ptl);
373 	cond_resched();
374 	return 0;
375 }
376 
377 static int show_smap(struct seq_file *m, void *v)
378 {
379 	struct proc_maps_private *priv = m->private;
380 	struct task_struct *task = priv->task;
381 	struct vm_area_struct *vma = v;
382 	struct mem_size_stats mss;
383 	struct mm_walk smaps_walk = {
384 		.pmd_entry = smaps_pte_range,
385 		.mm = vma->vm_mm,
386 		.private = &mss,
387 	};
388 
389 	memset(&mss, 0, sizeof mss);
390 	mss.vma = vma;
391 	/* mmap_sem is held in m_start */
392 	if (vma->vm_mm && !is_vm_hugetlb_page(vma))
393 		walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk);
394 
395 	show_map_vma(m, vma);
396 
397 	seq_printf(m,
398 		   "Size:           %8lu kB\n"
399 		   "Rss:            %8lu kB\n"
400 		   "Pss:            %8lu kB\n"
401 		   "Shared_Clean:   %8lu kB\n"
402 		   "Shared_Dirty:   %8lu kB\n"
403 		   "Private_Clean:  %8lu kB\n"
404 		   "Private_Dirty:  %8lu kB\n"
405 		   "Referenced:     %8lu kB\n"
406 		   "Swap:           %8lu kB\n"
407 		   "KernelPageSize: %8lu kB\n"
408 		   "MMUPageSize:    %8lu kB\n",
409 		   (vma->vm_end - vma->vm_start) >> 10,
410 		   mss.resident >> 10,
411 		   (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
412 		   mss.shared_clean  >> 10,
413 		   mss.shared_dirty  >> 10,
414 		   mss.private_clean >> 10,
415 		   mss.private_dirty >> 10,
416 		   mss.referenced >> 10,
417 		   mss.swap >> 10,
418 		   vma_kernel_pagesize(vma) >> 10,
419 		   vma_mmu_pagesize(vma) >> 10);
420 
421 	if (m->count < m->size)  /* vma is copied successfully */
422 		m->version = (vma != get_gate_vma(task)) ? vma->vm_start : 0;
423 	return 0;
424 }
425 
426 static const struct seq_operations proc_pid_smaps_op = {
427 	.start	= m_start,
428 	.next	= m_next,
429 	.stop	= m_stop,
430 	.show	= show_smap
431 };
432 
433 static int smaps_open(struct inode *inode, struct file *file)
434 {
435 	return do_maps_open(inode, file, &proc_pid_smaps_op);
436 }
437 
438 const struct file_operations proc_smaps_operations = {
439 	.open		= smaps_open,
440 	.read		= seq_read,
441 	.llseek		= seq_lseek,
442 	.release	= seq_release_private,
443 };
444 
445 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
446 				unsigned long end, struct mm_walk *walk)
447 {
448 	struct vm_area_struct *vma = walk->private;
449 	pte_t *pte, ptent;
450 	spinlock_t *ptl;
451 	struct page *page;
452 
453 	pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
454 	for (; addr != end; pte++, addr += PAGE_SIZE) {
455 		ptent = *pte;
456 		if (!pte_present(ptent))
457 			continue;
458 
459 		page = vm_normal_page(vma, addr, ptent);
460 		if (!page)
461 			continue;
462 
463 		/* Clear accessed and referenced bits. */
464 		ptep_test_and_clear_young(vma, addr, pte);
465 		ClearPageReferenced(page);
466 	}
467 	pte_unmap_unlock(pte - 1, ptl);
468 	cond_resched();
469 	return 0;
470 }
471 
472 #define CLEAR_REFS_ALL 1
473 #define CLEAR_REFS_ANON 2
474 #define CLEAR_REFS_MAPPED 3
475 
476 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
477 				size_t count, loff_t *ppos)
478 {
479 	struct task_struct *task;
480 	char buffer[PROC_NUMBUF];
481 	struct mm_struct *mm;
482 	struct vm_area_struct *vma;
483 	long type;
484 
485 	memset(buffer, 0, sizeof(buffer));
486 	if (count > sizeof(buffer) - 1)
487 		count = sizeof(buffer) - 1;
488 	if (copy_from_user(buffer, buf, count))
489 		return -EFAULT;
490 	if (strict_strtol(strstrip(buffer), 10, &type))
491 		return -EINVAL;
492 	if (type < CLEAR_REFS_ALL || type > CLEAR_REFS_MAPPED)
493 		return -EINVAL;
494 	task = get_proc_task(file->f_path.dentry->d_inode);
495 	if (!task)
496 		return -ESRCH;
497 	mm = get_task_mm(task);
498 	if (mm) {
499 		struct mm_walk clear_refs_walk = {
500 			.pmd_entry = clear_refs_pte_range,
501 			.mm = mm,
502 		};
503 		down_read(&mm->mmap_sem);
504 		for (vma = mm->mmap; vma; vma = vma->vm_next) {
505 			clear_refs_walk.private = vma;
506 			if (is_vm_hugetlb_page(vma))
507 				continue;
508 			/*
509 			 * Writing 1 to /proc/pid/clear_refs affects all pages.
510 			 *
511 			 * Writing 2 to /proc/pid/clear_refs only affects
512 			 * Anonymous pages.
513 			 *
514 			 * Writing 3 to /proc/pid/clear_refs only affects file
515 			 * mapped pages.
516 			 */
517 			if (type == CLEAR_REFS_ANON && vma->vm_file)
518 				continue;
519 			if (type == CLEAR_REFS_MAPPED && !vma->vm_file)
520 				continue;
521 			walk_page_range(vma->vm_start, vma->vm_end,
522 					&clear_refs_walk);
523 		}
524 		flush_tlb_mm(mm);
525 		up_read(&mm->mmap_sem);
526 		mmput(mm);
527 	}
528 	put_task_struct(task);
529 
530 	return count;
531 }
532 
533 const struct file_operations proc_clear_refs_operations = {
534 	.write		= clear_refs_write,
535 };
536 
537 struct pagemapread {
538 	int pos, len;
539 	u64 *buffer;
540 };
541 
542 #define PM_ENTRY_BYTES      sizeof(u64)
543 #define PM_STATUS_BITS      3
544 #define PM_STATUS_OFFSET    (64 - PM_STATUS_BITS)
545 #define PM_STATUS_MASK      (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
546 #define PM_STATUS(nr)       (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
547 #define PM_PSHIFT_BITS      6
548 #define PM_PSHIFT_OFFSET    (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
549 #define PM_PSHIFT_MASK      (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
550 #define PM_PSHIFT(x)        (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
551 #define PM_PFRAME_MASK      ((1LL << PM_PSHIFT_OFFSET) - 1)
552 #define PM_PFRAME(x)        ((x) & PM_PFRAME_MASK)
553 
554 #define PM_PRESENT          PM_STATUS(4LL)
555 #define PM_SWAP             PM_STATUS(2LL)
556 #define PM_NOT_PRESENT      PM_PSHIFT(PAGE_SHIFT)
557 #define PM_END_OF_BUFFER    1
558 
559 static int add_to_pagemap(unsigned long addr, u64 pfn,
560 			  struct pagemapread *pm)
561 {
562 	pm->buffer[pm->pos++] = pfn;
563 	if (pm->pos >= pm->len)
564 		return PM_END_OF_BUFFER;
565 	return 0;
566 }
567 
568 static int pagemap_pte_hole(unsigned long start, unsigned long end,
569 				struct mm_walk *walk)
570 {
571 	struct pagemapread *pm = walk->private;
572 	unsigned long addr;
573 	int err = 0;
574 	for (addr = start; addr < end; addr += PAGE_SIZE) {
575 		err = add_to_pagemap(addr, PM_NOT_PRESENT, pm);
576 		if (err)
577 			break;
578 	}
579 	return err;
580 }
581 
582 static u64 swap_pte_to_pagemap_entry(pte_t pte)
583 {
584 	swp_entry_t e = pte_to_swp_entry(pte);
585 	return swp_type(e) | (swp_offset(e) << MAX_SWAPFILES_SHIFT);
586 }
587 
588 static u64 pte_to_pagemap_entry(pte_t pte)
589 {
590 	u64 pme = 0;
591 	if (is_swap_pte(pte))
592 		pme = PM_PFRAME(swap_pte_to_pagemap_entry(pte))
593 			| PM_PSHIFT(PAGE_SHIFT) | PM_SWAP;
594 	else if (pte_present(pte))
595 		pme = PM_PFRAME(pte_pfn(pte))
596 			| PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
597 	return pme;
598 }
599 
600 static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
601 			     struct mm_walk *walk)
602 {
603 	struct vm_area_struct *vma;
604 	struct pagemapread *pm = walk->private;
605 	pte_t *pte;
606 	int err = 0;
607 
608 	/* find the first VMA at or above 'addr' */
609 	vma = find_vma(walk->mm, addr);
610 	for (; addr != end; addr += PAGE_SIZE) {
611 		u64 pfn = PM_NOT_PRESENT;
612 
613 		/* check to see if we've left 'vma' behind
614 		 * and need a new, higher one */
615 		if (vma && (addr >= vma->vm_end))
616 			vma = find_vma(walk->mm, addr);
617 
618 		/* check that 'vma' actually covers this address,
619 		 * and that it isn't a huge page vma */
620 		if (vma && (vma->vm_start <= addr) &&
621 		    !is_vm_hugetlb_page(vma)) {
622 			pte = pte_offset_map(pmd, addr);
623 			pfn = pte_to_pagemap_entry(*pte);
624 			/* unmap before userspace copy */
625 			pte_unmap(pte);
626 		}
627 		err = add_to_pagemap(addr, pfn, pm);
628 		if (err)
629 			return err;
630 	}
631 
632 	cond_resched();
633 
634 	return err;
635 }
636 
637 #ifdef CONFIG_HUGETLB_PAGE
638 static u64 huge_pte_to_pagemap_entry(pte_t pte, int offset)
639 {
640 	u64 pme = 0;
641 	if (pte_present(pte))
642 		pme = PM_PFRAME(pte_pfn(pte) + offset)
643 			| PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
644 	return pme;
645 }
646 
647 /* This function walks within one hugetlb entry in the single call */
648 static int pagemap_hugetlb_range(pte_t *pte, unsigned long hmask,
649 				 unsigned long addr, unsigned long end,
650 				 struct mm_walk *walk)
651 {
652 	struct pagemapread *pm = walk->private;
653 	int err = 0;
654 	u64 pfn;
655 
656 	for (; addr != end; addr += PAGE_SIZE) {
657 		int offset = (addr & ~hmask) >> PAGE_SHIFT;
658 		pfn = huge_pte_to_pagemap_entry(*pte, offset);
659 		err = add_to_pagemap(addr, pfn, pm);
660 		if (err)
661 			return err;
662 	}
663 
664 	cond_resched();
665 
666 	return err;
667 }
668 #endif /* HUGETLB_PAGE */
669 
670 /*
671  * /proc/pid/pagemap - an array mapping virtual pages to pfns
672  *
673  * For each page in the address space, this file contains one 64-bit entry
674  * consisting of the following:
675  *
676  * Bits 0-55  page frame number (PFN) if present
677  * Bits 0-4   swap type if swapped
678  * Bits 5-55  swap offset if swapped
679  * Bits 55-60 page shift (page size = 1<<page shift)
680  * Bit  61    reserved for future use
681  * Bit  62    page swapped
682  * Bit  63    page present
683  *
684  * If the page is not present but in swap, then the PFN contains an
685  * encoding of the swap file number and the page's offset into the
686  * swap. Unmapped pages return a null PFN. This allows determining
687  * precisely which pages are mapped (or in swap) and comparing mapped
688  * pages between processes.
689  *
690  * Efficient users of this interface will use /proc/pid/maps to
691  * determine which areas of memory are actually mapped and llseek to
692  * skip over unmapped regions.
693  */
694 #define PAGEMAP_WALK_SIZE	(PMD_SIZE)
695 static ssize_t pagemap_read(struct file *file, char __user *buf,
696 			    size_t count, loff_t *ppos)
697 {
698 	struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
699 	struct mm_struct *mm;
700 	struct pagemapread pm;
701 	int ret = -ESRCH;
702 	struct mm_walk pagemap_walk = {};
703 	unsigned long src;
704 	unsigned long svpfn;
705 	unsigned long start_vaddr;
706 	unsigned long end_vaddr;
707 	int copied = 0;
708 
709 	if (!task)
710 		goto out;
711 
712 	ret = -EACCES;
713 	if (!ptrace_may_access(task, PTRACE_MODE_READ))
714 		goto out_task;
715 
716 	ret = -EINVAL;
717 	/* file position must be aligned */
718 	if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
719 		goto out_task;
720 
721 	ret = 0;
722 
723 	if (!count)
724 		goto out_task;
725 
726 	mm = get_task_mm(task);
727 	if (!mm)
728 		goto out_task;
729 
730 	pm.len = PM_ENTRY_BYTES * (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
731 	pm.buffer = kmalloc(pm.len, GFP_TEMPORARY);
732 	ret = -ENOMEM;
733 	if (!pm.buffer)
734 		goto out_mm;
735 
736 	pagemap_walk.pmd_entry = pagemap_pte_range;
737 	pagemap_walk.pte_hole = pagemap_pte_hole;
738 #ifdef CONFIG_HUGETLB_PAGE
739 	pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
740 #endif
741 	pagemap_walk.mm = mm;
742 	pagemap_walk.private = &pm;
743 
744 	src = *ppos;
745 	svpfn = src / PM_ENTRY_BYTES;
746 	start_vaddr = svpfn << PAGE_SHIFT;
747 	end_vaddr = TASK_SIZE_OF(task);
748 
749 	/* watch out for wraparound */
750 	if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
751 		start_vaddr = end_vaddr;
752 
753 	/*
754 	 * The odds are that this will stop walking way
755 	 * before end_vaddr, because the length of the
756 	 * user buffer is tracked in "pm", and the walk
757 	 * will stop when we hit the end of the buffer.
758 	 */
759 	ret = 0;
760 	while (count && (start_vaddr < end_vaddr)) {
761 		int len;
762 		unsigned long end;
763 
764 		pm.pos = 0;
765 		end = start_vaddr + PAGEMAP_WALK_SIZE;
766 		/* overflow ? */
767 		if (end < start_vaddr || end > end_vaddr)
768 			end = end_vaddr;
769 		down_read(&mm->mmap_sem);
770 		ret = walk_page_range(start_vaddr, end, &pagemap_walk);
771 		up_read(&mm->mmap_sem);
772 		start_vaddr = end;
773 
774 		len = min(count, PM_ENTRY_BYTES * pm.pos);
775 		if (copy_to_user(buf, pm.buffer, len)) {
776 			ret = -EFAULT;
777 			goto out_free;
778 		}
779 		copied += len;
780 		buf += len;
781 		count -= len;
782 	}
783 	*ppos += copied;
784 	if (!ret || ret == PM_END_OF_BUFFER)
785 		ret = copied;
786 
787 out_free:
788 	kfree(pm.buffer);
789 out_mm:
790 	mmput(mm);
791 out_task:
792 	put_task_struct(task);
793 out:
794 	return ret;
795 }
796 
797 const struct file_operations proc_pagemap_operations = {
798 	.llseek		= mem_lseek, /* borrow this */
799 	.read		= pagemap_read,
800 };
801 #endif /* CONFIG_PROC_PAGE_MONITOR */
802 
803 #ifdef CONFIG_NUMA
804 extern int show_numa_map(struct seq_file *m, void *v);
805 
806 static const struct seq_operations proc_pid_numa_maps_op = {
807         .start  = m_start,
808         .next   = m_next,
809         .stop   = m_stop,
810         .show   = show_numa_map,
811 };
812 
813 static int numa_maps_open(struct inode *inode, struct file *file)
814 {
815 	return do_maps_open(inode, file, &proc_pid_numa_maps_op);
816 }
817 
818 const struct file_operations proc_numa_maps_operations = {
819 	.open		= numa_maps_open,
820 	.read		= seq_read,
821 	.llseek		= seq_lseek,
822 	.release	= seq_release_private,
823 };
824 #endif
825