xref: /linux/fs/proc/task_mmu.c (revision f2ee442115c9b6219083c019939a9cc0c9abb2f8)
1 #include <linux/mm.h>
2 #include <linux/hugetlb.h>
3 #include <linux/huge_mm.h>
4 #include <linux/mount.h>
5 #include <linux/seq_file.h>
6 #include <linux/highmem.h>
7 #include <linux/ptrace.h>
8 #include <linux/slab.h>
9 #include <linux/pagemap.h>
10 #include <linux/mempolicy.h>
11 #include <linux/rmap.h>
12 #include <linux/swap.h>
13 #include <linux/swapops.h>
14 
15 #include <asm/elf.h>
16 #include <asm/uaccess.h>
17 #include <asm/tlbflush.h>
18 #include "internal.h"
19 
20 void task_mem(struct seq_file *m, struct mm_struct *mm)
21 {
22 	unsigned long data, text, lib, swap;
23 	unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
24 
25 	/*
26 	 * Note: to minimize their overhead, mm maintains hiwater_vm and
27 	 * hiwater_rss only when about to *lower* total_vm or rss.  Any
28 	 * collector of these hiwater stats must therefore get total_vm
29 	 * and rss too, which will usually be the higher.  Barriers? not
30 	 * worth the effort, such snapshots can always be inconsistent.
31 	 */
32 	hiwater_vm = total_vm = mm->total_vm;
33 	if (hiwater_vm < mm->hiwater_vm)
34 		hiwater_vm = mm->hiwater_vm;
35 	hiwater_rss = total_rss = get_mm_rss(mm);
36 	if (hiwater_rss < mm->hiwater_rss)
37 		hiwater_rss = mm->hiwater_rss;
38 
39 	data = mm->total_vm - mm->shared_vm - mm->stack_vm;
40 	text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
41 	lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
42 	swap = get_mm_counter(mm, MM_SWAPENTS);
43 	seq_printf(m,
44 		"VmPeak:\t%8lu kB\n"
45 		"VmSize:\t%8lu kB\n"
46 		"VmLck:\t%8lu kB\n"
47 		"VmPin:\t%8lu kB\n"
48 		"VmHWM:\t%8lu kB\n"
49 		"VmRSS:\t%8lu kB\n"
50 		"VmData:\t%8lu kB\n"
51 		"VmStk:\t%8lu kB\n"
52 		"VmExe:\t%8lu kB\n"
53 		"VmLib:\t%8lu kB\n"
54 		"VmPTE:\t%8lu kB\n"
55 		"VmSwap:\t%8lu kB\n",
56 		hiwater_vm << (PAGE_SHIFT-10),
57 		(total_vm - mm->reserved_vm) << (PAGE_SHIFT-10),
58 		mm->locked_vm << (PAGE_SHIFT-10),
59 		mm->pinned_vm << (PAGE_SHIFT-10),
60 		hiwater_rss << (PAGE_SHIFT-10),
61 		total_rss << (PAGE_SHIFT-10),
62 		data << (PAGE_SHIFT-10),
63 		mm->stack_vm << (PAGE_SHIFT-10), text, lib,
64 		(PTRS_PER_PTE*sizeof(pte_t)*mm->nr_ptes) >> 10,
65 		swap << (PAGE_SHIFT-10));
66 }
67 
68 unsigned long task_vsize(struct mm_struct *mm)
69 {
70 	return PAGE_SIZE * mm->total_vm;
71 }
72 
73 unsigned long task_statm(struct mm_struct *mm,
74 			 unsigned long *shared, unsigned long *text,
75 			 unsigned long *data, unsigned long *resident)
76 {
77 	*shared = get_mm_counter(mm, MM_FILEPAGES);
78 	*text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
79 								>> PAGE_SHIFT;
80 	*data = mm->total_vm - mm->shared_vm;
81 	*resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
82 	return mm->total_vm;
83 }
84 
85 static void pad_len_spaces(struct seq_file *m, int len)
86 {
87 	len = 25 + sizeof(void*) * 6 - len;
88 	if (len < 1)
89 		len = 1;
90 	seq_printf(m, "%*c", len, ' ');
91 }
92 
93 static void vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma)
94 {
95 	if (vma && vma != priv->tail_vma) {
96 		struct mm_struct *mm = vma->vm_mm;
97 		up_read(&mm->mmap_sem);
98 		mmput(mm);
99 	}
100 }
101 
102 static void *m_start(struct seq_file *m, loff_t *pos)
103 {
104 	struct proc_maps_private *priv = m->private;
105 	unsigned long last_addr = m->version;
106 	struct mm_struct *mm;
107 	struct vm_area_struct *vma, *tail_vma = NULL;
108 	loff_t l = *pos;
109 
110 	/* Clear the per syscall fields in priv */
111 	priv->task = NULL;
112 	priv->tail_vma = NULL;
113 
114 	/*
115 	 * We remember last_addr rather than next_addr to hit with
116 	 * mmap_cache most of the time. We have zero last_addr at
117 	 * the beginning and also after lseek. We will have -1 last_addr
118 	 * after the end of the vmas.
119 	 */
120 
121 	if (last_addr == -1UL)
122 		return NULL;
123 
124 	priv->task = get_pid_task(priv->pid, PIDTYPE_PID);
125 	if (!priv->task)
126 		return ERR_PTR(-ESRCH);
127 
128 	mm = mm_for_maps(priv->task);
129 	if (!mm || IS_ERR(mm))
130 		return mm;
131 	down_read(&mm->mmap_sem);
132 
133 	tail_vma = get_gate_vma(priv->task->mm);
134 	priv->tail_vma = tail_vma;
135 
136 	/* Start with last addr hint */
137 	vma = find_vma(mm, last_addr);
138 	if (last_addr && vma) {
139 		vma = vma->vm_next;
140 		goto out;
141 	}
142 
143 	/*
144 	 * Check the vma index is within the range and do
145 	 * sequential scan until m_index.
146 	 */
147 	vma = NULL;
148 	if ((unsigned long)l < mm->map_count) {
149 		vma = mm->mmap;
150 		while (l-- && vma)
151 			vma = vma->vm_next;
152 		goto out;
153 	}
154 
155 	if (l != mm->map_count)
156 		tail_vma = NULL; /* After gate vma */
157 
158 out:
159 	if (vma)
160 		return vma;
161 
162 	/* End of vmas has been reached */
163 	m->version = (tail_vma != NULL)? 0: -1UL;
164 	up_read(&mm->mmap_sem);
165 	mmput(mm);
166 	return tail_vma;
167 }
168 
169 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
170 {
171 	struct proc_maps_private *priv = m->private;
172 	struct vm_area_struct *vma = v;
173 	struct vm_area_struct *tail_vma = priv->tail_vma;
174 
175 	(*pos)++;
176 	if (vma && (vma != tail_vma) && vma->vm_next)
177 		return vma->vm_next;
178 	vma_stop(priv, vma);
179 	return (vma != tail_vma)? tail_vma: NULL;
180 }
181 
182 static void m_stop(struct seq_file *m, void *v)
183 {
184 	struct proc_maps_private *priv = m->private;
185 	struct vm_area_struct *vma = v;
186 
187 	if (!IS_ERR(vma))
188 		vma_stop(priv, vma);
189 	if (priv->task)
190 		put_task_struct(priv->task);
191 }
192 
193 static int do_maps_open(struct inode *inode, struct file *file,
194 			const struct seq_operations *ops)
195 {
196 	struct proc_maps_private *priv;
197 	int ret = -ENOMEM;
198 	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
199 	if (priv) {
200 		priv->pid = proc_pid(inode);
201 		ret = seq_open(file, ops);
202 		if (!ret) {
203 			struct seq_file *m = file->private_data;
204 			m->private = priv;
205 		} else {
206 			kfree(priv);
207 		}
208 	}
209 	return ret;
210 }
211 
212 static void show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
213 {
214 	struct mm_struct *mm = vma->vm_mm;
215 	struct file *file = vma->vm_file;
216 	vm_flags_t flags = vma->vm_flags;
217 	unsigned long ino = 0;
218 	unsigned long long pgoff = 0;
219 	unsigned long start, end;
220 	dev_t dev = 0;
221 	int len;
222 
223 	if (file) {
224 		struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
225 		dev = inode->i_sb->s_dev;
226 		ino = inode->i_ino;
227 		pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
228 	}
229 
230 	/* We don't show the stack guard page in /proc/maps */
231 	start = vma->vm_start;
232 	if (stack_guard_page_start(vma, start))
233 		start += PAGE_SIZE;
234 	end = vma->vm_end;
235 	if (stack_guard_page_end(vma, end))
236 		end -= PAGE_SIZE;
237 
238 	seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu %n",
239 			start,
240 			end,
241 			flags & VM_READ ? 'r' : '-',
242 			flags & VM_WRITE ? 'w' : '-',
243 			flags & VM_EXEC ? 'x' : '-',
244 			flags & VM_MAYSHARE ? 's' : 'p',
245 			pgoff,
246 			MAJOR(dev), MINOR(dev), ino, &len);
247 
248 	/*
249 	 * Print the dentry name for named mappings, and a
250 	 * special [heap] marker for the heap:
251 	 */
252 	if (file) {
253 		pad_len_spaces(m, len);
254 		seq_path(m, &file->f_path, "\n");
255 	} else {
256 		const char *name = arch_vma_name(vma);
257 		if (!name) {
258 			if (mm) {
259 				if (vma->vm_start <= mm->brk &&
260 						vma->vm_end >= mm->start_brk) {
261 					name = "[heap]";
262 				} else if (vma->vm_start <= mm->start_stack &&
263 					   vma->vm_end >= mm->start_stack) {
264 					name = "[stack]";
265 				}
266 			} else {
267 				name = "[vdso]";
268 			}
269 		}
270 		if (name) {
271 			pad_len_spaces(m, len);
272 			seq_puts(m, name);
273 		}
274 	}
275 	seq_putc(m, '\n');
276 }
277 
278 static int show_map(struct seq_file *m, void *v)
279 {
280 	struct vm_area_struct *vma = v;
281 	struct proc_maps_private *priv = m->private;
282 	struct task_struct *task = priv->task;
283 
284 	show_map_vma(m, vma);
285 
286 	if (m->count < m->size)  /* vma is copied successfully */
287 		m->version = (vma != get_gate_vma(task->mm))
288 			? vma->vm_start : 0;
289 	return 0;
290 }
291 
292 static const struct seq_operations proc_pid_maps_op = {
293 	.start	= m_start,
294 	.next	= m_next,
295 	.stop	= m_stop,
296 	.show	= show_map
297 };
298 
299 static int maps_open(struct inode *inode, struct file *file)
300 {
301 	return do_maps_open(inode, file, &proc_pid_maps_op);
302 }
303 
304 const struct file_operations proc_maps_operations = {
305 	.open		= maps_open,
306 	.read		= seq_read,
307 	.llseek		= seq_lseek,
308 	.release	= seq_release_private,
309 };
310 
311 /*
312  * Proportional Set Size(PSS): my share of RSS.
313  *
314  * PSS of a process is the count of pages it has in memory, where each
315  * page is divided by the number of processes sharing it.  So if a
316  * process has 1000 pages all to itself, and 1000 shared with one other
317  * process, its PSS will be 1500.
318  *
319  * To keep (accumulated) division errors low, we adopt a 64bit
320  * fixed-point pss counter to minimize division errors. So (pss >>
321  * PSS_SHIFT) would be the real byte count.
322  *
323  * A shift of 12 before division means (assuming 4K page size):
324  * 	- 1M 3-user-pages add up to 8KB errors;
325  * 	- supports mapcount up to 2^24, or 16M;
326  * 	- supports PSS up to 2^52 bytes, or 4PB.
327  */
328 #define PSS_SHIFT 12
329 
330 #ifdef CONFIG_PROC_PAGE_MONITOR
331 struct mem_size_stats {
332 	struct vm_area_struct *vma;
333 	unsigned long resident;
334 	unsigned long shared_clean;
335 	unsigned long shared_dirty;
336 	unsigned long private_clean;
337 	unsigned long private_dirty;
338 	unsigned long referenced;
339 	unsigned long anonymous;
340 	unsigned long anonymous_thp;
341 	unsigned long swap;
342 	u64 pss;
343 };
344 
345 
346 static void smaps_pte_entry(pte_t ptent, unsigned long addr,
347 		unsigned long ptent_size, struct mm_walk *walk)
348 {
349 	struct mem_size_stats *mss = walk->private;
350 	struct vm_area_struct *vma = mss->vma;
351 	struct page *page;
352 	int mapcount;
353 
354 	if (is_swap_pte(ptent)) {
355 		mss->swap += ptent_size;
356 		return;
357 	}
358 
359 	if (!pte_present(ptent))
360 		return;
361 
362 	page = vm_normal_page(vma, addr, ptent);
363 	if (!page)
364 		return;
365 
366 	if (PageAnon(page))
367 		mss->anonymous += ptent_size;
368 
369 	mss->resident += ptent_size;
370 	/* Accumulate the size in pages that have been accessed. */
371 	if (pte_young(ptent) || PageReferenced(page))
372 		mss->referenced += ptent_size;
373 	mapcount = page_mapcount(page);
374 	if (mapcount >= 2) {
375 		if (pte_dirty(ptent) || PageDirty(page))
376 			mss->shared_dirty += ptent_size;
377 		else
378 			mss->shared_clean += ptent_size;
379 		mss->pss += (ptent_size << PSS_SHIFT) / mapcount;
380 	} else {
381 		if (pte_dirty(ptent) || PageDirty(page))
382 			mss->private_dirty += ptent_size;
383 		else
384 			mss->private_clean += ptent_size;
385 		mss->pss += (ptent_size << PSS_SHIFT);
386 	}
387 }
388 
389 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
390 			   struct mm_walk *walk)
391 {
392 	struct mem_size_stats *mss = walk->private;
393 	struct vm_area_struct *vma = mss->vma;
394 	pte_t *pte;
395 	spinlock_t *ptl;
396 
397 	spin_lock(&walk->mm->page_table_lock);
398 	if (pmd_trans_huge(*pmd)) {
399 		if (pmd_trans_splitting(*pmd)) {
400 			spin_unlock(&walk->mm->page_table_lock);
401 			wait_split_huge_page(vma->anon_vma, pmd);
402 		} else {
403 			smaps_pte_entry(*(pte_t *)pmd, addr,
404 					HPAGE_PMD_SIZE, walk);
405 			spin_unlock(&walk->mm->page_table_lock);
406 			mss->anonymous_thp += HPAGE_PMD_SIZE;
407 			return 0;
408 		}
409 	} else {
410 		spin_unlock(&walk->mm->page_table_lock);
411 	}
412 	/*
413 	 * The mmap_sem held all the way back in m_start() is what
414 	 * keeps khugepaged out of here and from collapsing things
415 	 * in here.
416 	 */
417 	pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
418 	for (; addr != end; pte++, addr += PAGE_SIZE)
419 		smaps_pte_entry(*pte, addr, PAGE_SIZE, walk);
420 	pte_unmap_unlock(pte - 1, ptl);
421 	cond_resched();
422 	return 0;
423 }
424 
425 static int show_smap(struct seq_file *m, void *v)
426 {
427 	struct proc_maps_private *priv = m->private;
428 	struct task_struct *task = priv->task;
429 	struct vm_area_struct *vma = v;
430 	struct mem_size_stats mss;
431 	struct mm_walk smaps_walk = {
432 		.pmd_entry = smaps_pte_range,
433 		.mm = vma->vm_mm,
434 		.private = &mss,
435 	};
436 
437 	memset(&mss, 0, sizeof mss);
438 	mss.vma = vma;
439 	/* mmap_sem is held in m_start */
440 	if (vma->vm_mm && !is_vm_hugetlb_page(vma))
441 		walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk);
442 
443 	show_map_vma(m, vma);
444 
445 	seq_printf(m,
446 		   "Size:           %8lu kB\n"
447 		   "Rss:            %8lu kB\n"
448 		   "Pss:            %8lu kB\n"
449 		   "Shared_Clean:   %8lu kB\n"
450 		   "Shared_Dirty:   %8lu kB\n"
451 		   "Private_Clean:  %8lu kB\n"
452 		   "Private_Dirty:  %8lu kB\n"
453 		   "Referenced:     %8lu kB\n"
454 		   "Anonymous:      %8lu kB\n"
455 		   "AnonHugePages:  %8lu kB\n"
456 		   "Swap:           %8lu kB\n"
457 		   "KernelPageSize: %8lu kB\n"
458 		   "MMUPageSize:    %8lu kB\n"
459 		   "Locked:         %8lu kB\n",
460 		   (vma->vm_end - vma->vm_start) >> 10,
461 		   mss.resident >> 10,
462 		   (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
463 		   mss.shared_clean  >> 10,
464 		   mss.shared_dirty  >> 10,
465 		   mss.private_clean >> 10,
466 		   mss.private_dirty >> 10,
467 		   mss.referenced >> 10,
468 		   mss.anonymous >> 10,
469 		   mss.anonymous_thp >> 10,
470 		   mss.swap >> 10,
471 		   vma_kernel_pagesize(vma) >> 10,
472 		   vma_mmu_pagesize(vma) >> 10,
473 		   (vma->vm_flags & VM_LOCKED) ?
474 			(unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
475 
476 	if (m->count < m->size)  /* vma is copied successfully */
477 		m->version = (vma != get_gate_vma(task->mm))
478 			? vma->vm_start : 0;
479 	return 0;
480 }
481 
482 static const struct seq_operations proc_pid_smaps_op = {
483 	.start	= m_start,
484 	.next	= m_next,
485 	.stop	= m_stop,
486 	.show	= show_smap
487 };
488 
489 static int smaps_open(struct inode *inode, struct file *file)
490 {
491 	return do_maps_open(inode, file, &proc_pid_smaps_op);
492 }
493 
494 const struct file_operations proc_smaps_operations = {
495 	.open		= smaps_open,
496 	.read		= seq_read,
497 	.llseek		= seq_lseek,
498 	.release	= seq_release_private,
499 };
500 
501 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
502 				unsigned long end, struct mm_walk *walk)
503 {
504 	struct vm_area_struct *vma = walk->private;
505 	pte_t *pte, ptent;
506 	spinlock_t *ptl;
507 	struct page *page;
508 
509 	split_huge_page_pmd(walk->mm, pmd);
510 
511 	pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
512 	for (; addr != end; pte++, addr += PAGE_SIZE) {
513 		ptent = *pte;
514 		if (!pte_present(ptent))
515 			continue;
516 
517 		page = vm_normal_page(vma, addr, ptent);
518 		if (!page)
519 			continue;
520 
521 		/* Clear accessed and referenced bits. */
522 		ptep_test_and_clear_young(vma, addr, pte);
523 		ClearPageReferenced(page);
524 	}
525 	pte_unmap_unlock(pte - 1, ptl);
526 	cond_resched();
527 	return 0;
528 }
529 
530 #define CLEAR_REFS_ALL 1
531 #define CLEAR_REFS_ANON 2
532 #define CLEAR_REFS_MAPPED 3
533 
534 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
535 				size_t count, loff_t *ppos)
536 {
537 	struct task_struct *task;
538 	char buffer[PROC_NUMBUF];
539 	struct mm_struct *mm;
540 	struct vm_area_struct *vma;
541 	int type;
542 	int rv;
543 
544 	memset(buffer, 0, sizeof(buffer));
545 	if (count > sizeof(buffer) - 1)
546 		count = sizeof(buffer) - 1;
547 	if (copy_from_user(buffer, buf, count))
548 		return -EFAULT;
549 	rv = kstrtoint(strstrip(buffer), 10, &type);
550 	if (rv < 0)
551 		return rv;
552 	if (type < CLEAR_REFS_ALL || type > CLEAR_REFS_MAPPED)
553 		return -EINVAL;
554 	task = get_proc_task(file->f_path.dentry->d_inode);
555 	if (!task)
556 		return -ESRCH;
557 	mm = get_task_mm(task);
558 	if (mm) {
559 		struct mm_walk clear_refs_walk = {
560 			.pmd_entry = clear_refs_pte_range,
561 			.mm = mm,
562 		};
563 		down_read(&mm->mmap_sem);
564 		for (vma = mm->mmap; vma; vma = vma->vm_next) {
565 			clear_refs_walk.private = vma;
566 			if (is_vm_hugetlb_page(vma))
567 				continue;
568 			/*
569 			 * Writing 1 to /proc/pid/clear_refs affects all pages.
570 			 *
571 			 * Writing 2 to /proc/pid/clear_refs only affects
572 			 * Anonymous pages.
573 			 *
574 			 * Writing 3 to /proc/pid/clear_refs only affects file
575 			 * mapped pages.
576 			 */
577 			if (type == CLEAR_REFS_ANON && vma->vm_file)
578 				continue;
579 			if (type == CLEAR_REFS_MAPPED && !vma->vm_file)
580 				continue;
581 			walk_page_range(vma->vm_start, vma->vm_end,
582 					&clear_refs_walk);
583 		}
584 		flush_tlb_mm(mm);
585 		up_read(&mm->mmap_sem);
586 		mmput(mm);
587 	}
588 	put_task_struct(task);
589 
590 	return count;
591 }
592 
593 const struct file_operations proc_clear_refs_operations = {
594 	.write		= clear_refs_write,
595 	.llseek		= noop_llseek,
596 };
597 
598 struct pagemapread {
599 	int pos, len;
600 	u64 *buffer;
601 };
602 
603 #define PM_ENTRY_BYTES      sizeof(u64)
604 #define PM_STATUS_BITS      3
605 #define PM_STATUS_OFFSET    (64 - PM_STATUS_BITS)
606 #define PM_STATUS_MASK      (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
607 #define PM_STATUS(nr)       (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
608 #define PM_PSHIFT_BITS      6
609 #define PM_PSHIFT_OFFSET    (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
610 #define PM_PSHIFT_MASK      (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
611 #define PM_PSHIFT(x)        (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
612 #define PM_PFRAME_MASK      ((1LL << PM_PSHIFT_OFFSET) - 1)
613 #define PM_PFRAME(x)        ((x) & PM_PFRAME_MASK)
614 
615 #define PM_PRESENT          PM_STATUS(4LL)
616 #define PM_SWAP             PM_STATUS(2LL)
617 #define PM_NOT_PRESENT      PM_PSHIFT(PAGE_SHIFT)
618 #define PM_END_OF_BUFFER    1
619 
620 static int add_to_pagemap(unsigned long addr, u64 pfn,
621 			  struct pagemapread *pm)
622 {
623 	pm->buffer[pm->pos++] = pfn;
624 	if (pm->pos >= pm->len)
625 		return PM_END_OF_BUFFER;
626 	return 0;
627 }
628 
629 static int pagemap_pte_hole(unsigned long start, unsigned long end,
630 				struct mm_walk *walk)
631 {
632 	struct pagemapread *pm = walk->private;
633 	unsigned long addr;
634 	int err = 0;
635 	for (addr = start; addr < end; addr += PAGE_SIZE) {
636 		err = add_to_pagemap(addr, PM_NOT_PRESENT, pm);
637 		if (err)
638 			break;
639 	}
640 	return err;
641 }
642 
643 static u64 swap_pte_to_pagemap_entry(pte_t pte)
644 {
645 	swp_entry_t e = pte_to_swp_entry(pte);
646 	return swp_type(e) | (swp_offset(e) << MAX_SWAPFILES_SHIFT);
647 }
648 
649 static u64 pte_to_pagemap_entry(pte_t pte)
650 {
651 	u64 pme = 0;
652 	if (is_swap_pte(pte))
653 		pme = PM_PFRAME(swap_pte_to_pagemap_entry(pte))
654 			| PM_PSHIFT(PAGE_SHIFT) | PM_SWAP;
655 	else if (pte_present(pte))
656 		pme = PM_PFRAME(pte_pfn(pte))
657 			| PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
658 	return pme;
659 }
660 
661 static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
662 			     struct mm_walk *walk)
663 {
664 	struct vm_area_struct *vma;
665 	struct pagemapread *pm = walk->private;
666 	pte_t *pte;
667 	int err = 0;
668 
669 	split_huge_page_pmd(walk->mm, pmd);
670 
671 	/* find the first VMA at or above 'addr' */
672 	vma = find_vma(walk->mm, addr);
673 	for (; addr != end; addr += PAGE_SIZE) {
674 		u64 pfn = PM_NOT_PRESENT;
675 
676 		/* check to see if we've left 'vma' behind
677 		 * and need a new, higher one */
678 		if (vma && (addr >= vma->vm_end))
679 			vma = find_vma(walk->mm, addr);
680 
681 		/* check that 'vma' actually covers this address,
682 		 * and that it isn't a huge page vma */
683 		if (vma && (vma->vm_start <= addr) &&
684 		    !is_vm_hugetlb_page(vma)) {
685 			pte = pte_offset_map(pmd, addr);
686 			pfn = pte_to_pagemap_entry(*pte);
687 			/* unmap before userspace copy */
688 			pte_unmap(pte);
689 		}
690 		err = add_to_pagemap(addr, pfn, pm);
691 		if (err)
692 			return err;
693 	}
694 
695 	cond_resched();
696 
697 	return err;
698 }
699 
700 #ifdef CONFIG_HUGETLB_PAGE
701 static u64 huge_pte_to_pagemap_entry(pte_t pte, int offset)
702 {
703 	u64 pme = 0;
704 	if (pte_present(pte))
705 		pme = PM_PFRAME(pte_pfn(pte) + offset)
706 			| PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
707 	return pme;
708 }
709 
710 /* This function walks within one hugetlb entry in the single call */
711 static int pagemap_hugetlb_range(pte_t *pte, unsigned long hmask,
712 				 unsigned long addr, unsigned long end,
713 				 struct mm_walk *walk)
714 {
715 	struct pagemapread *pm = walk->private;
716 	int err = 0;
717 	u64 pfn;
718 
719 	for (; addr != end; addr += PAGE_SIZE) {
720 		int offset = (addr & ~hmask) >> PAGE_SHIFT;
721 		pfn = huge_pte_to_pagemap_entry(*pte, offset);
722 		err = add_to_pagemap(addr, pfn, pm);
723 		if (err)
724 			return err;
725 	}
726 
727 	cond_resched();
728 
729 	return err;
730 }
731 #endif /* HUGETLB_PAGE */
732 
733 /*
734  * /proc/pid/pagemap - an array mapping virtual pages to pfns
735  *
736  * For each page in the address space, this file contains one 64-bit entry
737  * consisting of the following:
738  *
739  * Bits 0-55  page frame number (PFN) if present
740  * Bits 0-4   swap type if swapped
741  * Bits 5-55  swap offset if swapped
742  * Bits 55-60 page shift (page size = 1<<page shift)
743  * Bit  61    reserved for future use
744  * Bit  62    page swapped
745  * Bit  63    page present
746  *
747  * If the page is not present but in swap, then the PFN contains an
748  * encoding of the swap file number and the page's offset into the
749  * swap. Unmapped pages return a null PFN. This allows determining
750  * precisely which pages are mapped (or in swap) and comparing mapped
751  * pages between processes.
752  *
753  * Efficient users of this interface will use /proc/pid/maps to
754  * determine which areas of memory are actually mapped and llseek to
755  * skip over unmapped regions.
756  */
757 #define PAGEMAP_WALK_SIZE	(PMD_SIZE)
758 #define PAGEMAP_WALK_MASK	(PMD_MASK)
759 static ssize_t pagemap_read(struct file *file, char __user *buf,
760 			    size_t count, loff_t *ppos)
761 {
762 	struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
763 	struct mm_struct *mm;
764 	struct pagemapread pm;
765 	int ret = -ESRCH;
766 	struct mm_walk pagemap_walk = {};
767 	unsigned long src;
768 	unsigned long svpfn;
769 	unsigned long start_vaddr;
770 	unsigned long end_vaddr;
771 	int copied = 0;
772 
773 	if (!task)
774 		goto out;
775 
776 	ret = -EINVAL;
777 	/* file position must be aligned */
778 	if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
779 		goto out_task;
780 
781 	ret = 0;
782 	if (!count)
783 		goto out_task;
784 
785 	pm.len = PM_ENTRY_BYTES * (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
786 	pm.buffer = kmalloc(pm.len, GFP_TEMPORARY);
787 	ret = -ENOMEM;
788 	if (!pm.buffer)
789 		goto out_task;
790 
791 	mm = mm_for_maps(task);
792 	ret = PTR_ERR(mm);
793 	if (!mm || IS_ERR(mm))
794 		goto out_free;
795 
796 	pagemap_walk.pmd_entry = pagemap_pte_range;
797 	pagemap_walk.pte_hole = pagemap_pte_hole;
798 #ifdef CONFIG_HUGETLB_PAGE
799 	pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
800 #endif
801 	pagemap_walk.mm = mm;
802 	pagemap_walk.private = &pm;
803 
804 	src = *ppos;
805 	svpfn = src / PM_ENTRY_BYTES;
806 	start_vaddr = svpfn << PAGE_SHIFT;
807 	end_vaddr = TASK_SIZE_OF(task);
808 
809 	/* watch out for wraparound */
810 	if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
811 		start_vaddr = end_vaddr;
812 
813 	/*
814 	 * The odds are that this will stop walking way
815 	 * before end_vaddr, because the length of the
816 	 * user buffer is tracked in "pm", and the walk
817 	 * will stop when we hit the end of the buffer.
818 	 */
819 	ret = 0;
820 	while (count && (start_vaddr < end_vaddr)) {
821 		int len;
822 		unsigned long end;
823 
824 		pm.pos = 0;
825 		end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
826 		/* overflow ? */
827 		if (end < start_vaddr || end > end_vaddr)
828 			end = end_vaddr;
829 		down_read(&mm->mmap_sem);
830 		ret = walk_page_range(start_vaddr, end, &pagemap_walk);
831 		up_read(&mm->mmap_sem);
832 		start_vaddr = end;
833 
834 		len = min(count, PM_ENTRY_BYTES * pm.pos);
835 		if (copy_to_user(buf, pm.buffer, len)) {
836 			ret = -EFAULT;
837 			goto out_mm;
838 		}
839 		copied += len;
840 		buf += len;
841 		count -= len;
842 	}
843 	*ppos += copied;
844 	if (!ret || ret == PM_END_OF_BUFFER)
845 		ret = copied;
846 
847 out_mm:
848 	mmput(mm);
849 out_free:
850 	kfree(pm.buffer);
851 out_task:
852 	put_task_struct(task);
853 out:
854 	return ret;
855 }
856 
857 const struct file_operations proc_pagemap_operations = {
858 	.llseek		= mem_lseek, /* borrow this */
859 	.read		= pagemap_read,
860 };
861 #endif /* CONFIG_PROC_PAGE_MONITOR */
862 
863 #ifdef CONFIG_NUMA
864 
865 struct numa_maps {
866 	struct vm_area_struct *vma;
867 	unsigned long pages;
868 	unsigned long anon;
869 	unsigned long active;
870 	unsigned long writeback;
871 	unsigned long mapcount_max;
872 	unsigned long dirty;
873 	unsigned long swapcache;
874 	unsigned long node[MAX_NUMNODES];
875 };
876 
877 struct numa_maps_private {
878 	struct proc_maps_private proc_maps;
879 	struct numa_maps md;
880 };
881 
882 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
883 			unsigned long nr_pages)
884 {
885 	int count = page_mapcount(page);
886 
887 	md->pages += nr_pages;
888 	if (pte_dirty || PageDirty(page))
889 		md->dirty += nr_pages;
890 
891 	if (PageSwapCache(page))
892 		md->swapcache += nr_pages;
893 
894 	if (PageActive(page) || PageUnevictable(page))
895 		md->active += nr_pages;
896 
897 	if (PageWriteback(page))
898 		md->writeback += nr_pages;
899 
900 	if (PageAnon(page))
901 		md->anon += nr_pages;
902 
903 	if (count > md->mapcount_max)
904 		md->mapcount_max = count;
905 
906 	md->node[page_to_nid(page)] += nr_pages;
907 }
908 
909 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
910 		unsigned long addr)
911 {
912 	struct page *page;
913 	int nid;
914 
915 	if (!pte_present(pte))
916 		return NULL;
917 
918 	page = vm_normal_page(vma, addr, pte);
919 	if (!page)
920 		return NULL;
921 
922 	if (PageReserved(page))
923 		return NULL;
924 
925 	nid = page_to_nid(page);
926 	if (!node_isset(nid, node_states[N_HIGH_MEMORY]))
927 		return NULL;
928 
929 	return page;
930 }
931 
932 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
933 		unsigned long end, struct mm_walk *walk)
934 {
935 	struct numa_maps *md;
936 	spinlock_t *ptl;
937 	pte_t *orig_pte;
938 	pte_t *pte;
939 
940 	md = walk->private;
941 	spin_lock(&walk->mm->page_table_lock);
942 	if (pmd_trans_huge(*pmd)) {
943 		if (pmd_trans_splitting(*pmd)) {
944 			spin_unlock(&walk->mm->page_table_lock);
945 			wait_split_huge_page(md->vma->anon_vma, pmd);
946 		} else {
947 			pte_t huge_pte = *(pte_t *)pmd;
948 			struct page *page;
949 
950 			page = can_gather_numa_stats(huge_pte, md->vma, addr);
951 			if (page)
952 				gather_stats(page, md, pte_dirty(huge_pte),
953 						HPAGE_PMD_SIZE/PAGE_SIZE);
954 			spin_unlock(&walk->mm->page_table_lock);
955 			return 0;
956 		}
957 	} else {
958 		spin_unlock(&walk->mm->page_table_lock);
959 	}
960 
961 	orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
962 	do {
963 		struct page *page = can_gather_numa_stats(*pte, md->vma, addr);
964 		if (!page)
965 			continue;
966 		gather_stats(page, md, pte_dirty(*pte), 1);
967 
968 	} while (pte++, addr += PAGE_SIZE, addr != end);
969 	pte_unmap_unlock(orig_pte, ptl);
970 	return 0;
971 }
972 #ifdef CONFIG_HUGETLB_PAGE
973 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
974 		unsigned long addr, unsigned long end, struct mm_walk *walk)
975 {
976 	struct numa_maps *md;
977 	struct page *page;
978 
979 	if (pte_none(*pte))
980 		return 0;
981 
982 	page = pte_page(*pte);
983 	if (!page)
984 		return 0;
985 
986 	md = walk->private;
987 	gather_stats(page, md, pte_dirty(*pte), 1);
988 	return 0;
989 }
990 
991 #else
992 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
993 		unsigned long addr, unsigned long end, struct mm_walk *walk)
994 {
995 	return 0;
996 }
997 #endif
998 
999 /*
1000  * Display pages allocated per node and memory policy via /proc.
1001  */
1002 static int show_numa_map(struct seq_file *m, void *v)
1003 {
1004 	struct numa_maps_private *numa_priv = m->private;
1005 	struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1006 	struct vm_area_struct *vma = v;
1007 	struct numa_maps *md = &numa_priv->md;
1008 	struct file *file = vma->vm_file;
1009 	struct mm_struct *mm = vma->vm_mm;
1010 	struct mm_walk walk = {};
1011 	struct mempolicy *pol;
1012 	int n;
1013 	char buffer[50];
1014 
1015 	if (!mm)
1016 		return 0;
1017 
1018 	/* Ensure we start with an empty set of numa_maps statistics. */
1019 	memset(md, 0, sizeof(*md));
1020 
1021 	md->vma = vma;
1022 
1023 	walk.hugetlb_entry = gather_hugetbl_stats;
1024 	walk.pmd_entry = gather_pte_stats;
1025 	walk.private = md;
1026 	walk.mm = mm;
1027 
1028 	pol = get_vma_policy(proc_priv->task, vma, vma->vm_start);
1029 	mpol_to_str(buffer, sizeof(buffer), pol, 0);
1030 	mpol_cond_put(pol);
1031 
1032 	seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1033 
1034 	if (file) {
1035 		seq_printf(m, " file=");
1036 		seq_path(m, &file->f_path, "\n\t= ");
1037 	} else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1038 		seq_printf(m, " heap");
1039 	} else if (vma->vm_start <= mm->start_stack &&
1040 			vma->vm_end >= mm->start_stack) {
1041 		seq_printf(m, " stack");
1042 	}
1043 
1044 	if (is_vm_hugetlb_page(vma))
1045 		seq_printf(m, " huge");
1046 
1047 	walk_page_range(vma->vm_start, vma->vm_end, &walk);
1048 
1049 	if (!md->pages)
1050 		goto out;
1051 
1052 	if (md->anon)
1053 		seq_printf(m, " anon=%lu", md->anon);
1054 
1055 	if (md->dirty)
1056 		seq_printf(m, " dirty=%lu", md->dirty);
1057 
1058 	if (md->pages != md->anon && md->pages != md->dirty)
1059 		seq_printf(m, " mapped=%lu", md->pages);
1060 
1061 	if (md->mapcount_max > 1)
1062 		seq_printf(m, " mapmax=%lu", md->mapcount_max);
1063 
1064 	if (md->swapcache)
1065 		seq_printf(m, " swapcache=%lu", md->swapcache);
1066 
1067 	if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1068 		seq_printf(m, " active=%lu", md->active);
1069 
1070 	if (md->writeback)
1071 		seq_printf(m, " writeback=%lu", md->writeback);
1072 
1073 	for_each_node_state(n, N_HIGH_MEMORY)
1074 		if (md->node[n])
1075 			seq_printf(m, " N%d=%lu", n, md->node[n]);
1076 out:
1077 	seq_putc(m, '\n');
1078 
1079 	if (m->count < m->size)
1080 		m->version = (vma != proc_priv->tail_vma) ? vma->vm_start : 0;
1081 	return 0;
1082 }
1083 
1084 static const struct seq_operations proc_pid_numa_maps_op = {
1085         .start  = m_start,
1086         .next   = m_next,
1087         .stop   = m_stop,
1088         .show   = show_numa_map,
1089 };
1090 
1091 static int numa_maps_open(struct inode *inode, struct file *file)
1092 {
1093 	struct numa_maps_private *priv;
1094 	int ret = -ENOMEM;
1095 	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
1096 	if (priv) {
1097 		priv->proc_maps.pid = proc_pid(inode);
1098 		ret = seq_open(file, &proc_pid_numa_maps_op);
1099 		if (!ret) {
1100 			struct seq_file *m = file->private_data;
1101 			m->private = priv;
1102 		} else {
1103 			kfree(priv);
1104 		}
1105 	}
1106 	return ret;
1107 }
1108 
1109 const struct file_operations proc_numa_maps_operations = {
1110 	.open		= numa_maps_open,
1111 	.read		= seq_read,
1112 	.llseek		= seq_lseek,
1113 	.release	= seq_release_private,
1114 };
1115 #endif /* CONFIG_NUMA */
1116