xref: /openbmc/linux/fs/proc/task_mmu.c (revision 4800cd83)
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 unsigned long task_statm(struct mm_struct *mm,
70 			 unsigned long *shared, unsigned long *text,
71 			 unsigned long *data, unsigned long *resident)
72 {
73 	*shared = get_mm_counter(mm, MM_FILEPAGES);
74 	*text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
75 								>> PAGE_SHIFT;
76 	*data = mm->total_vm - mm->shared_vm;
77 	*resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
78 	return mm->total_vm;
79 }
80 
81 static void pad_len_spaces(struct seq_file *m, int len)
82 {
83 	len = 25 + sizeof(void*) * 6 - len;
84 	if (len < 1)
85 		len = 1;
86 	seq_printf(m, "%*c", len, ' ');
87 }
88 
89 static void vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma)
90 {
91 	if (vma && vma != priv->tail_vma) {
92 		struct mm_struct *mm = vma->vm_mm;
93 		up_read(&mm->mmap_sem);
94 		mmput(mm);
95 	}
96 }
97 
98 static void *m_start(struct seq_file *m, loff_t *pos)
99 {
100 	struct proc_maps_private *priv = m->private;
101 	unsigned long last_addr = m->version;
102 	struct mm_struct *mm;
103 	struct vm_area_struct *vma, *tail_vma = NULL;
104 	loff_t l = *pos;
105 
106 	/* Clear the per syscall fields in priv */
107 	priv->task = NULL;
108 	priv->tail_vma = NULL;
109 
110 	/*
111 	 * We remember last_addr rather than next_addr to hit with
112 	 * mmap_cache most of the time. We have zero last_addr at
113 	 * the beginning and also after lseek. We will have -1 last_addr
114 	 * after the end of the vmas.
115 	 */
116 
117 	if (last_addr == -1UL)
118 		return NULL;
119 
120 	priv->task = get_pid_task(priv->pid, PIDTYPE_PID);
121 	if (!priv->task)
122 		return NULL;
123 
124 	mm = mm_for_maps(priv->task);
125 	if (!mm)
126 		return NULL;
127 	down_read(&mm->mmap_sem);
128 
129 	tail_vma = get_gate_vma(priv->task);
130 	priv->tail_vma = tail_vma;
131 
132 	/* Start with last addr hint */
133 	vma = find_vma(mm, last_addr);
134 	if (last_addr && vma) {
135 		vma = vma->vm_next;
136 		goto out;
137 	}
138 
139 	/*
140 	 * Check the vma index is within the range and do
141 	 * sequential scan until m_index.
142 	 */
143 	vma = NULL;
144 	if ((unsigned long)l < mm->map_count) {
145 		vma = mm->mmap;
146 		while (l-- && vma)
147 			vma = vma->vm_next;
148 		goto out;
149 	}
150 
151 	if (l != mm->map_count)
152 		tail_vma = NULL; /* After gate vma */
153 
154 out:
155 	if (vma)
156 		return vma;
157 
158 	/* End of vmas has been reached */
159 	m->version = (tail_vma != NULL)? 0: -1UL;
160 	up_read(&mm->mmap_sem);
161 	mmput(mm);
162 	return tail_vma;
163 }
164 
165 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
166 {
167 	struct proc_maps_private *priv = m->private;
168 	struct vm_area_struct *vma = v;
169 	struct vm_area_struct *tail_vma = priv->tail_vma;
170 
171 	(*pos)++;
172 	if (vma && (vma != tail_vma) && vma->vm_next)
173 		return vma->vm_next;
174 	vma_stop(priv, vma);
175 	return (vma != tail_vma)? tail_vma: NULL;
176 }
177 
178 static void m_stop(struct seq_file *m, void *v)
179 {
180 	struct proc_maps_private *priv = m->private;
181 	struct vm_area_struct *vma = v;
182 
183 	vma_stop(priv, vma);
184 	if (priv->task)
185 		put_task_struct(priv->task);
186 }
187 
188 static int do_maps_open(struct inode *inode, struct file *file,
189 			const struct seq_operations *ops)
190 {
191 	struct proc_maps_private *priv;
192 	int ret = -ENOMEM;
193 	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
194 	if (priv) {
195 		priv->pid = proc_pid(inode);
196 		ret = seq_open(file, ops);
197 		if (!ret) {
198 			struct seq_file *m = file->private_data;
199 			m->private = priv;
200 		} else {
201 			kfree(priv);
202 		}
203 	}
204 	return ret;
205 }
206 
207 static void show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
208 {
209 	struct mm_struct *mm = vma->vm_mm;
210 	struct file *file = vma->vm_file;
211 	int flags = vma->vm_flags;
212 	unsigned long ino = 0;
213 	unsigned long long pgoff = 0;
214 	unsigned long start;
215 	dev_t dev = 0;
216 	int len;
217 
218 	if (file) {
219 		struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
220 		dev = inode->i_sb->s_dev;
221 		ino = inode->i_ino;
222 		pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
223 	}
224 
225 	/* We don't show the stack guard page in /proc/maps */
226 	start = vma->vm_start;
227 	if (vma->vm_flags & VM_GROWSDOWN)
228 		if (!vma_stack_continue(vma->vm_prev, vma->vm_start))
229 			start += PAGE_SIZE;
230 
231 	seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu %n",
232 			start,
233 			vma->vm_end,
234 			flags & VM_READ ? 'r' : '-',
235 			flags & VM_WRITE ? 'w' : '-',
236 			flags & VM_EXEC ? 'x' : '-',
237 			flags & VM_MAYSHARE ? 's' : 'p',
238 			pgoff,
239 			MAJOR(dev), MINOR(dev), ino, &len);
240 
241 	/*
242 	 * Print the dentry name for named mappings, and a
243 	 * special [heap] marker for the heap:
244 	 */
245 	if (file) {
246 		pad_len_spaces(m, len);
247 		seq_path(m, &file->f_path, "\n");
248 	} else {
249 		const char *name = arch_vma_name(vma);
250 		if (!name) {
251 			if (mm) {
252 				if (vma->vm_start <= mm->start_brk &&
253 						vma->vm_end >= mm->brk) {
254 					name = "[heap]";
255 				} else if (vma->vm_start <= mm->start_stack &&
256 					   vma->vm_end >= mm->start_stack) {
257 					name = "[stack]";
258 				}
259 			} else {
260 				name = "[vdso]";
261 			}
262 		}
263 		if (name) {
264 			pad_len_spaces(m, len);
265 			seq_puts(m, name);
266 		}
267 	}
268 	seq_putc(m, '\n');
269 }
270 
271 static int show_map(struct seq_file *m, void *v)
272 {
273 	struct vm_area_struct *vma = v;
274 	struct proc_maps_private *priv = m->private;
275 	struct task_struct *task = priv->task;
276 
277 	show_map_vma(m, vma);
278 
279 	if (m->count < m->size)  /* vma is copied successfully */
280 		m->version = (vma != get_gate_vma(task))? vma->vm_start: 0;
281 	return 0;
282 }
283 
284 static const struct seq_operations proc_pid_maps_op = {
285 	.start	= m_start,
286 	.next	= m_next,
287 	.stop	= m_stop,
288 	.show	= show_map
289 };
290 
291 static int maps_open(struct inode *inode, struct file *file)
292 {
293 	return do_maps_open(inode, file, &proc_pid_maps_op);
294 }
295 
296 const struct file_operations proc_maps_operations = {
297 	.open		= maps_open,
298 	.read		= seq_read,
299 	.llseek		= seq_lseek,
300 	.release	= seq_release_private,
301 };
302 
303 /*
304  * Proportional Set Size(PSS): my share of RSS.
305  *
306  * PSS of a process is the count of pages it has in memory, where each
307  * page is divided by the number of processes sharing it.  So if a
308  * process has 1000 pages all to itself, and 1000 shared with one other
309  * process, its PSS will be 1500.
310  *
311  * To keep (accumulated) division errors low, we adopt a 64bit
312  * fixed-point pss counter to minimize division errors. So (pss >>
313  * PSS_SHIFT) would be the real byte count.
314  *
315  * A shift of 12 before division means (assuming 4K page size):
316  * 	- 1M 3-user-pages add up to 8KB errors;
317  * 	- supports mapcount up to 2^24, or 16M;
318  * 	- supports PSS up to 2^52 bytes, or 4PB.
319  */
320 #define PSS_SHIFT 12
321 
322 #ifdef CONFIG_PROC_PAGE_MONITOR
323 struct mem_size_stats {
324 	struct vm_area_struct *vma;
325 	unsigned long resident;
326 	unsigned long shared_clean;
327 	unsigned long shared_dirty;
328 	unsigned long private_clean;
329 	unsigned long private_dirty;
330 	unsigned long referenced;
331 	unsigned long anonymous;
332 	unsigned long swap;
333 	u64 pss;
334 };
335 
336 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
337 			   struct mm_walk *walk)
338 {
339 	struct mem_size_stats *mss = walk->private;
340 	struct vm_area_struct *vma = mss->vma;
341 	pte_t *pte, ptent;
342 	spinlock_t *ptl;
343 	struct page *page;
344 	int mapcount;
345 
346 	pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
347 	for (; addr != end; pte++, addr += PAGE_SIZE) {
348 		ptent = *pte;
349 
350 		if (is_swap_pte(ptent)) {
351 			mss->swap += PAGE_SIZE;
352 			continue;
353 		}
354 
355 		if (!pte_present(ptent))
356 			continue;
357 
358 		page = vm_normal_page(vma, addr, ptent);
359 		if (!page)
360 			continue;
361 
362 		if (PageAnon(page))
363 			mss->anonymous += PAGE_SIZE;
364 
365 		mss->resident += PAGE_SIZE;
366 		/* Accumulate the size in pages that have been accessed. */
367 		if (pte_young(ptent) || PageReferenced(page))
368 			mss->referenced += PAGE_SIZE;
369 		mapcount = page_mapcount(page);
370 		if (mapcount >= 2) {
371 			if (pte_dirty(ptent) || PageDirty(page))
372 				mss->shared_dirty += PAGE_SIZE;
373 			else
374 				mss->shared_clean += PAGE_SIZE;
375 			mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount;
376 		} else {
377 			if (pte_dirty(ptent) || PageDirty(page))
378 				mss->private_dirty += PAGE_SIZE;
379 			else
380 				mss->private_clean += PAGE_SIZE;
381 			mss->pss += (PAGE_SIZE << PSS_SHIFT);
382 		}
383 	}
384 	pte_unmap_unlock(pte - 1, ptl);
385 	cond_resched();
386 	return 0;
387 }
388 
389 static int show_smap(struct seq_file *m, void *v)
390 {
391 	struct proc_maps_private *priv = m->private;
392 	struct task_struct *task = priv->task;
393 	struct vm_area_struct *vma = v;
394 	struct mem_size_stats mss;
395 	struct mm_walk smaps_walk = {
396 		.pmd_entry = smaps_pte_range,
397 		.mm = vma->vm_mm,
398 		.private = &mss,
399 	};
400 
401 	memset(&mss, 0, sizeof mss);
402 	mss.vma = vma;
403 	/* mmap_sem is held in m_start */
404 	if (vma->vm_mm && !is_vm_hugetlb_page(vma))
405 		walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk);
406 
407 	show_map_vma(m, vma);
408 
409 	seq_printf(m,
410 		   "Size:           %8lu kB\n"
411 		   "Rss:            %8lu kB\n"
412 		   "Pss:            %8lu kB\n"
413 		   "Shared_Clean:   %8lu kB\n"
414 		   "Shared_Dirty:   %8lu kB\n"
415 		   "Private_Clean:  %8lu kB\n"
416 		   "Private_Dirty:  %8lu kB\n"
417 		   "Referenced:     %8lu kB\n"
418 		   "Anonymous:      %8lu kB\n"
419 		   "Swap:           %8lu kB\n"
420 		   "KernelPageSize: %8lu kB\n"
421 		   "MMUPageSize:    %8lu kB\n"
422 		   "Locked:         %8lu kB\n",
423 		   (vma->vm_end - vma->vm_start) >> 10,
424 		   mss.resident >> 10,
425 		   (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
426 		   mss.shared_clean  >> 10,
427 		   mss.shared_dirty  >> 10,
428 		   mss.private_clean >> 10,
429 		   mss.private_dirty >> 10,
430 		   mss.referenced >> 10,
431 		   mss.anonymous >> 10,
432 		   mss.swap >> 10,
433 		   vma_kernel_pagesize(vma) >> 10,
434 		   vma_mmu_pagesize(vma) >> 10,
435 		   (vma->vm_flags & VM_LOCKED) ?
436 			(unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
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 	.llseek		= noop_llseek,
553 };
554 
555 struct pagemapread {
556 	int pos, len;
557 	u64 *buffer;
558 };
559 
560 #define PM_ENTRY_BYTES      sizeof(u64)
561 #define PM_STATUS_BITS      3
562 #define PM_STATUS_OFFSET    (64 - PM_STATUS_BITS)
563 #define PM_STATUS_MASK      (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
564 #define PM_STATUS(nr)       (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
565 #define PM_PSHIFT_BITS      6
566 #define PM_PSHIFT_OFFSET    (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
567 #define PM_PSHIFT_MASK      (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
568 #define PM_PSHIFT(x)        (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
569 #define PM_PFRAME_MASK      ((1LL << PM_PSHIFT_OFFSET) - 1)
570 #define PM_PFRAME(x)        ((x) & PM_PFRAME_MASK)
571 
572 #define PM_PRESENT          PM_STATUS(4LL)
573 #define PM_SWAP             PM_STATUS(2LL)
574 #define PM_NOT_PRESENT      PM_PSHIFT(PAGE_SHIFT)
575 #define PM_END_OF_BUFFER    1
576 
577 static int add_to_pagemap(unsigned long addr, u64 pfn,
578 			  struct pagemapread *pm)
579 {
580 	pm->buffer[pm->pos++] = pfn;
581 	if (pm->pos >= pm->len)
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 #ifdef CONFIG_HUGETLB_PAGE
656 static u64 huge_pte_to_pagemap_entry(pte_t pte, int offset)
657 {
658 	u64 pme = 0;
659 	if (pte_present(pte))
660 		pme = PM_PFRAME(pte_pfn(pte) + offset)
661 			| PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
662 	return pme;
663 }
664 
665 /* This function walks within one hugetlb entry in the single call */
666 static int pagemap_hugetlb_range(pte_t *pte, unsigned long hmask,
667 				 unsigned long addr, unsigned long end,
668 				 struct mm_walk *walk)
669 {
670 	struct pagemapread *pm = walk->private;
671 	int err = 0;
672 	u64 pfn;
673 
674 	for (; addr != end; addr += PAGE_SIZE) {
675 		int offset = (addr & ~hmask) >> PAGE_SHIFT;
676 		pfn = huge_pte_to_pagemap_entry(*pte, offset);
677 		err = add_to_pagemap(addr, pfn, pm);
678 		if (err)
679 			return err;
680 	}
681 
682 	cond_resched();
683 
684 	return err;
685 }
686 #endif /* HUGETLB_PAGE */
687 
688 /*
689  * /proc/pid/pagemap - an array mapping virtual pages to pfns
690  *
691  * For each page in the address space, this file contains one 64-bit entry
692  * consisting of the following:
693  *
694  * Bits 0-55  page frame number (PFN) if present
695  * Bits 0-4   swap type if swapped
696  * Bits 5-55  swap offset if swapped
697  * Bits 55-60 page shift (page size = 1<<page shift)
698  * Bit  61    reserved for future use
699  * Bit  62    page swapped
700  * Bit  63    page present
701  *
702  * If the page is not present but in swap, then the PFN contains an
703  * encoding of the swap file number and the page's offset into the
704  * swap. Unmapped pages return a null PFN. This allows determining
705  * precisely which pages are mapped (or in swap) and comparing mapped
706  * pages between processes.
707  *
708  * Efficient users of this interface will use /proc/pid/maps to
709  * determine which areas of memory are actually mapped and llseek to
710  * skip over unmapped regions.
711  */
712 #define PAGEMAP_WALK_SIZE	(PMD_SIZE)
713 #define PAGEMAP_WALK_MASK	(PMD_MASK)
714 static ssize_t pagemap_read(struct file *file, char __user *buf,
715 			    size_t count, loff_t *ppos)
716 {
717 	struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
718 	struct mm_struct *mm;
719 	struct pagemapread pm;
720 	int ret = -ESRCH;
721 	struct mm_walk pagemap_walk = {};
722 	unsigned long src;
723 	unsigned long svpfn;
724 	unsigned long start_vaddr;
725 	unsigned long end_vaddr;
726 	int copied = 0;
727 
728 	if (!task)
729 		goto out;
730 
731 	ret = -EACCES;
732 	if (!ptrace_may_access(task, PTRACE_MODE_READ))
733 		goto out_task;
734 
735 	ret = -EINVAL;
736 	/* file position must be aligned */
737 	if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
738 		goto out_task;
739 
740 	ret = 0;
741 
742 	if (!count)
743 		goto out_task;
744 
745 	mm = get_task_mm(task);
746 	if (!mm)
747 		goto out_task;
748 
749 	pm.len = PM_ENTRY_BYTES * (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
750 	pm.buffer = kmalloc(pm.len, GFP_TEMPORARY);
751 	ret = -ENOMEM;
752 	if (!pm.buffer)
753 		goto out_mm;
754 
755 	pagemap_walk.pmd_entry = pagemap_pte_range;
756 	pagemap_walk.pte_hole = pagemap_pte_hole;
757 #ifdef CONFIG_HUGETLB_PAGE
758 	pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
759 #endif
760 	pagemap_walk.mm = mm;
761 	pagemap_walk.private = &pm;
762 
763 	src = *ppos;
764 	svpfn = src / PM_ENTRY_BYTES;
765 	start_vaddr = svpfn << PAGE_SHIFT;
766 	end_vaddr = TASK_SIZE_OF(task);
767 
768 	/* watch out for wraparound */
769 	if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
770 		start_vaddr = end_vaddr;
771 
772 	/*
773 	 * The odds are that this will stop walking way
774 	 * before end_vaddr, because the length of the
775 	 * user buffer is tracked in "pm", and the walk
776 	 * will stop when we hit the end of the buffer.
777 	 */
778 	ret = 0;
779 	while (count && (start_vaddr < end_vaddr)) {
780 		int len;
781 		unsigned long end;
782 
783 		pm.pos = 0;
784 		end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
785 		/* overflow ? */
786 		if (end < start_vaddr || end > end_vaddr)
787 			end = end_vaddr;
788 		down_read(&mm->mmap_sem);
789 		ret = walk_page_range(start_vaddr, end, &pagemap_walk);
790 		up_read(&mm->mmap_sem);
791 		start_vaddr = end;
792 
793 		len = min(count, PM_ENTRY_BYTES * pm.pos);
794 		if (copy_to_user(buf, pm.buffer, len)) {
795 			ret = -EFAULT;
796 			goto out_free;
797 		}
798 		copied += len;
799 		buf += len;
800 		count -= len;
801 	}
802 	*ppos += copied;
803 	if (!ret || ret == PM_END_OF_BUFFER)
804 		ret = copied;
805 
806 out_free:
807 	kfree(pm.buffer);
808 out_mm:
809 	mmput(mm);
810 out_task:
811 	put_task_struct(task);
812 out:
813 	return ret;
814 }
815 
816 const struct file_operations proc_pagemap_operations = {
817 	.llseek		= mem_lseek, /* borrow this */
818 	.read		= pagemap_read,
819 };
820 #endif /* CONFIG_PROC_PAGE_MONITOR */
821 
822 #ifdef CONFIG_NUMA
823 extern int show_numa_map(struct seq_file *m, void *v);
824 
825 static const struct seq_operations proc_pid_numa_maps_op = {
826         .start  = m_start,
827         .next   = m_next,
828         .stop   = m_stop,
829         .show   = show_numa_map,
830 };
831 
832 static int numa_maps_open(struct inode *inode, struct file *file)
833 {
834 	return do_maps_open(inode, file, &proc_pid_numa_maps_op);
835 }
836 
837 const struct file_operations proc_numa_maps_operations = {
838 	.open		= numa_maps_open,
839 	.read		= seq_read,
840 	.llseek		= seq_lseek,
841 	.release	= seq_release_private,
842 };
843 #endif
844