xref: /linux/fs/proc/task_mmu.c (revision a115bc070b1fc57ab23f3972401425927b5b465c)
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/pagemap.h>
8 #include <linux/mempolicy.h>
9 #include <linux/swap.h>
10 #include <linux/swapops.h>
11 
12 #include <asm/elf.h>
13 #include <asm/uaccess.h>
14 #include <asm/tlbflush.h>
15 #include "internal.h"
16 
17 void task_mem(struct seq_file *m, struct mm_struct *mm)
18 {
19 	unsigned long data, text, lib, swap;
20 	unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
21 
22 	/*
23 	 * Note: to minimize their overhead, mm maintains hiwater_vm and
24 	 * hiwater_rss only when about to *lower* total_vm or rss.  Any
25 	 * collector of these hiwater stats must therefore get total_vm
26 	 * and rss too, which will usually be the higher.  Barriers? not
27 	 * worth the effort, such snapshots can always be inconsistent.
28 	 */
29 	hiwater_vm = total_vm = mm->total_vm;
30 	if (hiwater_vm < mm->hiwater_vm)
31 		hiwater_vm = mm->hiwater_vm;
32 	hiwater_rss = total_rss = get_mm_rss(mm);
33 	if (hiwater_rss < mm->hiwater_rss)
34 		hiwater_rss = mm->hiwater_rss;
35 
36 	data = mm->total_vm - mm->shared_vm - mm->stack_vm;
37 	text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
38 	lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
39 	swap = get_mm_counter(mm, MM_SWAPENTS);
40 	seq_printf(m,
41 		"VmPeak:\t%8lu kB\n"
42 		"VmSize:\t%8lu kB\n"
43 		"VmLck:\t%8lu kB\n"
44 		"VmHWM:\t%8lu kB\n"
45 		"VmRSS:\t%8lu kB\n"
46 		"VmData:\t%8lu kB\n"
47 		"VmStk:\t%8lu kB\n"
48 		"VmExe:\t%8lu kB\n"
49 		"VmLib:\t%8lu kB\n"
50 		"VmPTE:\t%8lu kB\n"
51 		"VmSwap:\t%8lu kB\n",
52 		hiwater_vm << (PAGE_SHIFT-10),
53 		(total_vm - mm->reserved_vm) << (PAGE_SHIFT-10),
54 		mm->locked_vm << (PAGE_SHIFT-10),
55 		hiwater_rss << (PAGE_SHIFT-10),
56 		total_rss << (PAGE_SHIFT-10),
57 		data << (PAGE_SHIFT-10),
58 		mm->stack_vm << (PAGE_SHIFT-10), text, lib,
59 		(PTRS_PER_PTE*sizeof(pte_t)*mm->nr_ptes) >> 10,
60 		swap << (PAGE_SHIFT-10));
61 }
62 
63 unsigned long task_vsize(struct mm_struct *mm)
64 {
65 	return PAGE_SIZE * mm->total_vm;
66 }
67 
68 int task_statm(struct mm_struct *mm, int *shared, int *text,
69 	       int *data, int *resident)
70 {
71 	*shared = get_mm_counter(mm, MM_FILEPAGES);
72 	*text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
73 								>> PAGE_SHIFT;
74 	*data = mm->total_vm - mm->shared_vm;
75 	*resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
76 	return mm->total_vm;
77 }
78 
79 static void pad_len_spaces(struct seq_file *m, int len)
80 {
81 	len = 25 + sizeof(void*) * 6 - len;
82 	if (len < 1)
83 		len = 1;
84 	seq_printf(m, "%*c", len, ' ');
85 }
86 
87 static void vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma)
88 {
89 	if (vma && vma != priv->tail_vma) {
90 		struct mm_struct *mm = vma->vm_mm;
91 		up_read(&mm->mmap_sem);
92 		mmput(mm);
93 	}
94 }
95 
96 static void *m_start(struct seq_file *m, loff_t *pos)
97 {
98 	struct proc_maps_private *priv = m->private;
99 	unsigned long last_addr = m->version;
100 	struct mm_struct *mm;
101 	struct vm_area_struct *vma, *tail_vma = NULL;
102 	loff_t l = *pos;
103 
104 	/* Clear the per syscall fields in priv */
105 	priv->task = NULL;
106 	priv->tail_vma = NULL;
107 
108 	/*
109 	 * We remember last_addr rather than next_addr to hit with
110 	 * mmap_cache most of the time. We have zero last_addr at
111 	 * the beginning and also after lseek. We will have -1 last_addr
112 	 * after the end of the vmas.
113 	 */
114 
115 	if (last_addr == -1UL)
116 		return NULL;
117 
118 	priv->task = get_pid_task(priv->pid, PIDTYPE_PID);
119 	if (!priv->task)
120 		return NULL;
121 
122 	mm = mm_for_maps(priv->task);
123 	if (!mm)
124 		return NULL;
125 	down_read(&mm->mmap_sem);
126 
127 	tail_vma = get_gate_vma(priv->task);
128 	priv->tail_vma = tail_vma;
129 
130 	/* Start with last addr hint */
131 	vma = find_vma(mm, last_addr);
132 	if (last_addr && vma) {
133 		vma = vma->vm_next;
134 		goto out;
135 	}
136 
137 	/*
138 	 * Check the vma index is within the range and do
139 	 * sequential scan until m_index.
140 	 */
141 	vma = NULL;
142 	if ((unsigned long)l < mm->map_count) {
143 		vma = mm->mmap;
144 		while (l-- && vma)
145 			vma = vma->vm_next;
146 		goto out;
147 	}
148 
149 	if (l != mm->map_count)
150 		tail_vma = NULL; /* After gate vma */
151 
152 out:
153 	if (vma)
154 		return vma;
155 
156 	/* End of vmas has been reached */
157 	m->version = (tail_vma != NULL)? 0: -1UL;
158 	up_read(&mm->mmap_sem);
159 	mmput(mm);
160 	return tail_vma;
161 }
162 
163 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
164 {
165 	struct proc_maps_private *priv = m->private;
166 	struct vm_area_struct *vma = v;
167 	struct vm_area_struct *tail_vma = priv->tail_vma;
168 
169 	(*pos)++;
170 	if (vma && (vma != tail_vma) && vma->vm_next)
171 		return vma->vm_next;
172 	vma_stop(priv, vma);
173 	return (vma != tail_vma)? tail_vma: NULL;
174 }
175 
176 static void m_stop(struct seq_file *m, void *v)
177 {
178 	struct proc_maps_private *priv = m->private;
179 	struct vm_area_struct *vma = v;
180 
181 	vma_stop(priv, vma);
182 	if (priv->task)
183 		put_task_struct(priv->task);
184 }
185 
186 static int do_maps_open(struct inode *inode, struct file *file,
187 			const struct seq_operations *ops)
188 {
189 	struct proc_maps_private *priv;
190 	int ret = -ENOMEM;
191 	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
192 	if (priv) {
193 		priv->pid = proc_pid(inode);
194 		ret = seq_open(file, ops);
195 		if (!ret) {
196 			struct seq_file *m = file->private_data;
197 			m->private = priv;
198 		} else {
199 			kfree(priv);
200 		}
201 	}
202 	return ret;
203 }
204 
205 static void show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
206 {
207 	struct mm_struct *mm = vma->vm_mm;
208 	struct file *file = vma->vm_file;
209 	int flags = vma->vm_flags;
210 	unsigned long ino = 0;
211 	unsigned long long pgoff = 0;
212 	dev_t dev = 0;
213 	int len;
214 
215 	if (file) {
216 		struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
217 		dev = inode->i_sb->s_dev;
218 		ino = inode->i_ino;
219 		pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
220 	}
221 
222 	seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu %n",
223 			vma->vm_start,
224 			vma->vm_end,
225 			flags & VM_READ ? 'r' : '-',
226 			flags & VM_WRITE ? 'w' : '-',
227 			flags & VM_EXEC ? 'x' : '-',
228 			flags & VM_MAYSHARE ? 's' : 'p',
229 			pgoff,
230 			MAJOR(dev), MINOR(dev), ino, &len);
231 
232 	/*
233 	 * Print the dentry name for named mappings, and a
234 	 * special [heap] marker for the heap:
235 	 */
236 	if (file) {
237 		pad_len_spaces(m, len);
238 		seq_path(m, &file->f_path, "\n");
239 	} else {
240 		const char *name = arch_vma_name(vma);
241 		if (!name) {
242 			if (mm) {
243 				if (vma->vm_start <= mm->start_brk &&
244 						vma->vm_end >= mm->brk) {
245 					name = "[heap]";
246 				} else if (vma->vm_start <= mm->start_stack &&
247 					   vma->vm_end >= mm->start_stack) {
248 					name = "[stack]";
249 				} else {
250 					unsigned long stack_start;
251 					struct proc_maps_private *pmp;
252 
253 					pmp = m->private;
254 					stack_start = pmp->task->stack_start;
255 
256 					if (vma->vm_start <= stack_start &&
257 					    vma->vm_end >= stack_start) {
258 						pad_len_spaces(m, len);
259 						seq_printf(m,
260 						 "[threadstack:%08lx]",
261 #ifdef CONFIG_STACK_GROWSUP
262 						 vma->vm_end - stack_start
263 #else
264 						 stack_start - vma->vm_start
265 #endif
266 						);
267 					}
268 				}
269 			} else {
270 				name = "[vdso]";
271 			}
272 		}
273 		if (name) {
274 			pad_len_spaces(m, len);
275 			seq_puts(m, name);
276 		}
277 	}
278 	seq_putc(m, '\n');
279 }
280 
281 static int show_map(struct seq_file *m, void *v)
282 {
283 	struct vm_area_struct *vma = v;
284 	struct proc_maps_private *priv = m->private;
285 	struct task_struct *task = priv->task;
286 
287 	show_map_vma(m, vma);
288 
289 	if (m->count < m->size)  /* vma is copied successfully */
290 		m->version = (vma != get_gate_vma(task))? vma->vm_start: 0;
291 	return 0;
292 }
293 
294 static const struct seq_operations proc_pid_maps_op = {
295 	.start	= m_start,
296 	.next	= m_next,
297 	.stop	= m_stop,
298 	.show	= show_map
299 };
300 
301 static int maps_open(struct inode *inode, struct file *file)
302 {
303 	return do_maps_open(inode, file, &proc_pid_maps_op);
304 }
305 
306 const struct file_operations proc_maps_operations = {
307 	.open		= maps_open,
308 	.read		= seq_read,
309 	.llseek		= seq_lseek,
310 	.release	= seq_release_private,
311 };
312 
313 /*
314  * Proportional Set Size(PSS): my share of RSS.
315  *
316  * PSS of a process is the count of pages it has in memory, where each
317  * page is divided by the number of processes sharing it.  So if a
318  * process has 1000 pages all to itself, and 1000 shared with one other
319  * process, its PSS will be 1500.
320  *
321  * To keep (accumulated) division errors low, we adopt a 64bit
322  * fixed-point pss counter to minimize division errors. So (pss >>
323  * PSS_SHIFT) would be the real byte count.
324  *
325  * A shift of 12 before division means (assuming 4K page size):
326  * 	- 1M 3-user-pages add up to 8KB errors;
327  * 	- supports mapcount up to 2^24, or 16M;
328  * 	- supports PSS up to 2^52 bytes, or 4PB.
329  */
330 #define PSS_SHIFT 12
331 
332 #ifdef CONFIG_PROC_PAGE_MONITOR
333 struct mem_size_stats {
334 	struct vm_area_struct *vma;
335 	unsigned long resident;
336 	unsigned long shared_clean;
337 	unsigned long shared_dirty;
338 	unsigned long private_clean;
339 	unsigned long private_dirty;
340 	unsigned long referenced;
341 	unsigned long swap;
342 	u64 pss;
343 };
344 
345 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
346 			   struct mm_walk *walk)
347 {
348 	struct mem_size_stats *mss = walk->private;
349 	struct vm_area_struct *vma = mss->vma;
350 	pte_t *pte, ptent;
351 	spinlock_t *ptl;
352 	struct page *page;
353 	int mapcount;
354 
355 	pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
356 	for (; addr != end; pte++, addr += PAGE_SIZE) {
357 		ptent = *pte;
358 
359 		if (is_swap_pte(ptent)) {
360 			mss->swap += PAGE_SIZE;
361 			continue;
362 		}
363 
364 		if (!pte_present(ptent))
365 			continue;
366 
367 		page = vm_normal_page(vma, addr, ptent);
368 		if (!page)
369 			continue;
370 
371 		mss->resident += PAGE_SIZE;
372 		/* Accumulate the size in pages that have been accessed. */
373 		if (pte_young(ptent) || PageReferenced(page))
374 			mss->referenced += PAGE_SIZE;
375 		mapcount = page_mapcount(page);
376 		if (mapcount >= 2) {
377 			if (pte_dirty(ptent))
378 				mss->shared_dirty += PAGE_SIZE;
379 			else
380 				mss->shared_clean += PAGE_SIZE;
381 			mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount;
382 		} else {
383 			if (pte_dirty(ptent))
384 				mss->private_dirty += PAGE_SIZE;
385 			else
386 				mss->private_clean += PAGE_SIZE;
387 			mss->pss += (PAGE_SIZE << PSS_SHIFT);
388 		}
389 	}
390 	pte_unmap_unlock(pte - 1, ptl);
391 	cond_resched();
392 	return 0;
393 }
394 
395 static int show_smap(struct seq_file *m, void *v)
396 {
397 	struct proc_maps_private *priv = m->private;
398 	struct task_struct *task = priv->task;
399 	struct vm_area_struct *vma = v;
400 	struct mem_size_stats mss;
401 	struct mm_walk smaps_walk = {
402 		.pmd_entry = smaps_pte_range,
403 		.mm = vma->vm_mm,
404 		.private = &mss,
405 	};
406 
407 	memset(&mss, 0, sizeof mss);
408 	mss.vma = vma;
409 	if (vma->vm_mm && !is_vm_hugetlb_page(vma))
410 		walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk);
411 
412 	show_map_vma(m, vma);
413 
414 	seq_printf(m,
415 		   "Size:           %8lu kB\n"
416 		   "Rss:            %8lu kB\n"
417 		   "Pss:            %8lu kB\n"
418 		   "Shared_Clean:   %8lu kB\n"
419 		   "Shared_Dirty:   %8lu kB\n"
420 		   "Private_Clean:  %8lu kB\n"
421 		   "Private_Dirty:  %8lu kB\n"
422 		   "Referenced:     %8lu kB\n"
423 		   "Swap:           %8lu kB\n"
424 		   "KernelPageSize: %8lu kB\n"
425 		   "MMUPageSize:    %8lu kB\n",
426 		   (vma->vm_end - vma->vm_start) >> 10,
427 		   mss.resident >> 10,
428 		   (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
429 		   mss.shared_clean  >> 10,
430 		   mss.shared_dirty  >> 10,
431 		   mss.private_clean >> 10,
432 		   mss.private_dirty >> 10,
433 		   mss.referenced >> 10,
434 		   mss.swap >> 10,
435 		   vma_kernel_pagesize(vma) >> 10,
436 		   vma_mmu_pagesize(vma) >> 10);
437 
438 	if (m->count < m->size)  /* vma is copied successfully */
439 		m->version = (vma != get_gate_vma(task)) ? vma->vm_start : 0;
440 	return 0;
441 }
442 
443 static const struct seq_operations proc_pid_smaps_op = {
444 	.start	= m_start,
445 	.next	= m_next,
446 	.stop	= m_stop,
447 	.show	= show_smap
448 };
449 
450 static int smaps_open(struct inode *inode, struct file *file)
451 {
452 	return do_maps_open(inode, file, &proc_pid_smaps_op);
453 }
454 
455 const struct file_operations proc_smaps_operations = {
456 	.open		= smaps_open,
457 	.read		= seq_read,
458 	.llseek		= seq_lseek,
459 	.release	= seq_release_private,
460 };
461 
462 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
463 				unsigned long end, struct mm_walk *walk)
464 {
465 	struct vm_area_struct *vma = walk->private;
466 	pte_t *pte, ptent;
467 	spinlock_t *ptl;
468 	struct page *page;
469 
470 	pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
471 	for (; addr != end; pte++, addr += PAGE_SIZE) {
472 		ptent = *pte;
473 		if (!pte_present(ptent))
474 			continue;
475 
476 		page = vm_normal_page(vma, addr, ptent);
477 		if (!page)
478 			continue;
479 
480 		/* Clear accessed and referenced bits. */
481 		ptep_test_and_clear_young(vma, addr, pte);
482 		ClearPageReferenced(page);
483 	}
484 	pte_unmap_unlock(pte - 1, ptl);
485 	cond_resched();
486 	return 0;
487 }
488 
489 #define CLEAR_REFS_ALL 1
490 #define CLEAR_REFS_ANON 2
491 #define CLEAR_REFS_MAPPED 3
492 
493 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
494 				size_t count, loff_t *ppos)
495 {
496 	struct task_struct *task;
497 	char buffer[PROC_NUMBUF];
498 	struct mm_struct *mm;
499 	struct vm_area_struct *vma;
500 	long type;
501 
502 	memset(buffer, 0, sizeof(buffer));
503 	if (count > sizeof(buffer) - 1)
504 		count = sizeof(buffer) - 1;
505 	if (copy_from_user(buffer, buf, count))
506 		return -EFAULT;
507 	if (strict_strtol(strstrip(buffer), 10, &type))
508 		return -EINVAL;
509 	if (type < CLEAR_REFS_ALL || type > CLEAR_REFS_MAPPED)
510 		return -EINVAL;
511 	task = get_proc_task(file->f_path.dentry->d_inode);
512 	if (!task)
513 		return -ESRCH;
514 	mm = get_task_mm(task);
515 	if (mm) {
516 		struct mm_walk clear_refs_walk = {
517 			.pmd_entry = clear_refs_pte_range,
518 			.mm = mm,
519 		};
520 		down_read(&mm->mmap_sem);
521 		for (vma = mm->mmap; vma; vma = vma->vm_next) {
522 			clear_refs_walk.private = vma;
523 			if (is_vm_hugetlb_page(vma))
524 				continue;
525 			/*
526 			 * Writing 1 to /proc/pid/clear_refs affects all pages.
527 			 *
528 			 * Writing 2 to /proc/pid/clear_refs only affects
529 			 * Anonymous pages.
530 			 *
531 			 * Writing 3 to /proc/pid/clear_refs only affects file
532 			 * mapped pages.
533 			 */
534 			if (type == CLEAR_REFS_ANON && vma->vm_file)
535 				continue;
536 			if (type == CLEAR_REFS_MAPPED && !vma->vm_file)
537 				continue;
538 			walk_page_range(vma->vm_start, vma->vm_end,
539 					&clear_refs_walk);
540 		}
541 		flush_tlb_mm(mm);
542 		up_read(&mm->mmap_sem);
543 		mmput(mm);
544 	}
545 	put_task_struct(task);
546 
547 	return count;
548 }
549 
550 const struct file_operations proc_clear_refs_operations = {
551 	.write		= clear_refs_write,
552 };
553 
554 struct pagemapread {
555 	u64 __user *out, *end;
556 };
557 
558 #define PM_ENTRY_BYTES      sizeof(u64)
559 #define PM_STATUS_BITS      3
560 #define PM_STATUS_OFFSET    (64 - PM_STATUS_BITS)
561 #define PM_STATUS_MASK      (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
562 #define PM_STATUS(nr)       (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
563 #define PM_PSHIFT_BITS      6
564 #define PM_PSHIFT_OFFSET    (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
565 #define PM_PSHIFT_MASK      (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
566 #define PM_PSHIFT(x)        (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
567 #define PM_PFRAME_MASK      ((1LL << PM_PSHIFT_OFFSET) - 1)
568 #define PM_PFRAME(x)        ((x) & PM_PFRAME_MASK)
569 
570 #define PM_PRESENT          PM_STATUS(4LL)
571 #define PM_SWAP             PM_STATUS(2LL)
572 #define PM_NOT_PRESENT      PM_PSHIFT(PAGE_SHIFT)
573 #define PM_END_OF_BUFFER    1
574 
575 static int add_to_pagemap(unsigned long addr, u64 pfn,
576 			  struct pagemapread *pm)
577 {
578 	if (put_user(pfn, pm->out))
579 		return -EFAULT;
580 	pm->out++;
581 	if (pm->out >= pm->end)
582 		return PM_END_OF_BUFFER;
583 	return 0;
584 }
585 
586 static int pagemap_pte_hole(unsigned long start, unsigned long end,
587 				struct mm_walk *walk)
588 {
589 	struct pagemapread *pm = walk->private;
590 	unsigned long addr;
591 	int err = 0;
592 	for (addr = start; addr < end; addr += PAGE_SIZE) {
593 		err = add_to_pagemap(addr, PM_NOT_PRESENT, pm);
594 		if (err)
595 			break;
596 	}
597 	return err;
598 }
599 
600 static u64 swap_pte_to_pagemap_entry(pte_t pte)
601 {
602 	swp_entry_t e = pte_to_swp_entry(pte);
603 	return swp_type(e) | (swp_offset(e) << MAX_SWAPFILES_SHIFT);
604 }
605 
606 static u64 pte_to_pagemap_entry(pte_t pte)
607 {
608 	u64 pme = 0;
609 	if (is_swap_pte(pte))
610 		pme = PM_PFRAME(swap_pte_to_pagemap_entry(pte))
611 			| PM_PSHIFT(PAGE_SHIFT) | PM_SWAP;
612 	else if (pte_present(pte))
613 		pme = PM_PFRAME(pte_pfn(pte))
614 			| PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
615 	return pme;
616 }
617 
618 static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
619 			     struct mm_walk *walk)
620 {
621 	struct vm_area_struct *vma;
622 	struct pagemapread *pm = walk->private;
623 	pte_t *pte;
624 	int err = 0;
625 
626 	/* find the first VMA at or above 'addr' */
627 	vma = find_vma(walk->mm, addr);
628 	for (; addr != end; addr += PAGE_SIZE) {
629 		u64 pfn = PM_NOT_PRESENT;
630 
631 		/* check to see if we've left 'vma' behind
632 		 * and need a new, higher one */
633 		if (vma && (addr >= vma->vm_end))
634 			vma = find_vma(walk->mm, addr);
635 
636 		/* check that 'vma' actually covers this address,
637 		 * and that it isn't a huge page vma */
638 		if (vma && (vma->vm_start <= addr) &&
639 		    !is_vm_hugetlb_page(vma)) {
640 			pte = pte_offset_map(pmd, addr);
641 			pfn = pte_to_pagemap_entry(*pte);
642 			/* unmap before userspace copy */
643 			pte_unmap(pte);
644 		}
645 		err = add_to_pagemap(addr, pfn, pm);
646 		if (err)
647 			return err;
648 	}
649 
650 	cond_resched();
651 
652 	return err;
653 }
654 
655 static u64 huge_pte_to_pagemap_entry(pte_t pte, int offset)
656 {
657 	u64 pme = 0;
658 	if (pte_present(pte))
659 		pme = PM_PFRAME(pte_pfn(pte) + offset)
660 			| PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
661 	return pme;
662 }
663 
664 static int pagemap_hugetlb_range(pte_t *pte, unsigned long addr,
665 				 unsigned long end, struct mm_walk *walk)
666 {
667 	struct vm_area_struct *vma;
668 	struct pagemapread *pm = walk->private;
669 	struct hstate *hs = NULL;
670 	int err = 0;
671 
672 	vma = find_vma(walk->mm, addr);
673 	if (vma)
674 		hs = hstate_vma(vma);
675 	for (; addr != end; addr += PAGE_SIZE) {
676 		u64 pfn = PM_NOT_PRESENT;
677 
678 		if (vma && (addr >= vma->vm_end)) {
679 			vma = find_vma(walk->mm, addr);
680 			if (vma)
681 				hs = hstate_vma(vma);
682 		}
683 
684 		if (vma && (vma->vm_start <= addr) && is_vm_hugetlb_page(vma)) {
685 			/* calculate pfn of the "raw" page in the hugepage. */
686 			int offset = (addr & ~huge_page_mask(hs)) >> PAGE_SHIFT;
687 			pfn = huge_pte_to_pagemap_entry(*pte, offset);
688 		}
689 		err = add_to_pagemap(addr, pfn, pm);
690 		if (err)
691 			return err;
692 	}
693 
694 	cond_resched();
695 
696 	return err;
697 }
698 
699 /*
700  * /proc/pid/pagemap - an array mapping virtual pages to pfns
701  *
702  * For each page in the address space, this file contains one 64-bit entry
703  * consisting of the following:
704  *
705  * Bits 0-55  page frame number (PFN) if present
706  * Bits 0-4   swap type if swapped
707  * Bits 5-55  swap offset if swapped
708  * Bits 55-60 page shift (page size = 1<<page shift)
709  * Bit  61    reserved for future use
710  * Bit  62    page swapped
711  * Bit  63    page present
712  *
713  * If the page is not present but in swap, then the PFN contains an
714  * encoding of the swap file number and the page's offset into the
715  * swap. Unmapped pages return a null PFN. This allows determining
716  * precisely which pages are mapped (or in swap) and comparing mapped
717  * pages between processes.
718  *
719  * Efficient users of this interface will use /proc/pid/maps to
720  * determine which areas of memory are actually mapped and llseek to
721  * skip over unmapped regions.
722  */
723 static ssize_t pagemap_read(struct file *file, char __user *buf,
724 			    size_t count, loff_t *ppos)
725 {
726 	struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
727 	struct page **pages, *page;
728 	unsigned long uaddr, uend;
729 	struct mm_struct *mm;
730 	struct pagemapread pm;
731 	int pagecount;
732 	int ret = -ESRCH;
733 	struct mm_walk pagemap_walk = {};
734 	unsigned long src;
735 	unsigned long svpfn;
736 	unsigned long start_vaddr;
737 	unsigned long end_vaddr;
738 
739 	if (!task)
740 		goto out;
741 
742 	ret = -EACCES;
743 	if (!ptrace_may_access(task, PTRACE_MODE_READ))
744 		goto out_task;
745 
746 	ret = -EINVAL;
747 	/* file position must be aligned */
748 	if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
749 		goto out_task;
750 
751 	ret = 0;
752 
753 	if (!count)
754 		goto out_task;
755 
756 	mm = get_task_mm(task);
757 	if (!mm)
758 		goto out_task;
759 
760 
761 	uaddr = (unsigned long)buf & PAGE_MASK;
762 	uend = (unsigned long)(buf + count);
763 	pagecount = (PAGE_ALIGN(uend) - uaddr) / PAGE_SIZE;
764 	ret = 0;
765 	if (pagecount == 0)
766 		goto out_mm;
767 	pages = kcalloc(pagecount, sizeof(struct page *), GFP_KERNEL);
768 	ret = -ENOMEM;
769 	if (!pages)
770 		goto out_mm;
771 
772 	down_read(&current->mm->mmap_sem);
773 	ret = get_user_pages(current, current->mm, uaddr, pagecount,
774 			     1, 0, pages, NULL);
775 	up_read(&current->mm->mmap_sem);
776 
777 	if (ret < 0)
778 		goto out_free;
779 
780 	if (ret != pagecount) {
781 		pagecount = ret;
782 		ret = -EFAULT;
783 		goto out_pages;
784 	}
785 
786 	pm.out = (u64 __user *)buf;
787 	pm.end = (u64 __user *)(buf + count);
788 
789 	pagemap_walk.pmd_entry = pagemap_pte_range;
790 	pagemap_walk.pte_hole = pagemap_pte_hole;
791 	pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
792 	pagemap_walk.mm = mm;
793 	pagemap_walk.private = &pm;
794 
795 	src = *ppos;
796 	svpfn = src / PM_ENTRY_BYTES;
797 	start_vaddr = svpfn << PAGE_SHIFT;
798 	end_vaddr = TASK_SIZE_OF(task);
799 
800 	/* watch out for wraparound */
801 	if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
802 		start_vaddr = end_vaddr;
803 
804 	/*
805 	 * The odds are that this will stop walking way
806 	 * before end_vaddr, because the length of the
807 	 * user buffer is tracked in "pm", and the walk
808 	 * will stop when we hit the end of the buffer.
809 	 */
810 	ret = walk_page_range(start_vaddr, end_vaddr, &pagemap_walk);
811 	if (ret == PM_END_OF_BUFFER)
812 		ret = 0;
813 	/* don't need mmap_sem for these, but this looks cleaner */
814 	*ppos += (char __user *)pm.out - buf;
815 	if (!ret)
816 		ret = (char __user *)pm.out - buf;
817 
818 out_pages:
819 	for (; pagecount; pagecount--) {
820 		page = pages[pagecount-1];
821 		if (!PageReserved(page))
822 			SetPageDirty(page);
823 		page_cache_release(page);
824 	}
825 out_free:
826 	kfree(pages);
827 out_mm:
828 	mmput(mm);
829 out_task:
830 	put_task_struct(task);
831 out:
832 	return ret;
833 }
834 
835 const struct file_operations proc_pagemap_operations = {
836 	.llseek		= mem_lseek, /* borrow this */
837 	.read		= pagemap_read,
838 };
839 #endif /* CONFIG_PROC_PAGE_MONITOR */
840 
841 #ifdef CONFIG_NUMA
842 extern int show_numa_map(struct seq_file *m, void *v);
843 
844 static const struct seq_operations proc_pid_numa_maps_op = {
845         .start  = m_start,
846         .next   = m_next,
847         .stop   = m_stop,
848         .show   = show_numa_map,
849 };
850 
851 static int numa_maps_open(struct inode *inode, struct file *file)
852 {
853 	return do_maps_open(inode, file, &proc_pid_numa_maps_op);
854 }
855 
856 const struct file_operations proc_numa_maps_operations = {
857 	.open		= numa_maps_open,
858 	.read		= seq_read,
859 	.llseek		= seq_lseek,
860 	.release	= seq_release_private,
861 };
862 #endif
863