xref: /linux/fs/proc/task_mmu.c (revision 60e13231561b3a4c5269bfa1ef6c0569ad6f28ec)
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 		"VmHWM:\t%8lu kB\n"
48 		"VmRSS:\t%8lu kB\n"
49 		"VmData:\t%8lu kB\n"
50 		"VmStk:\t%8lu kB\n"
51 		"VmExe:\t%8lu kB\n"
52 		"VmLib:\t%8lu kB\n"
53 		"VmPTE:\t%8lu kB\n"
54 		"VmSwap:\t%8lu kB\n",
55 		hiwater_vm << (PAGE_SHIFT-10),
56 		(total_vm - mm->reserved_vm) << (PAGE_SHIFT-10),
57 		mm->locked_vm << (PAGE_SHIFT-10),
58 		hiwater_rss << (PAGE_SHIFT-10),
59 		total_rss << (PAGE_SHIFT-10),
60 		data << (PAGE_SHIFT-10),
61 		mm->stack_vm << (PAGE_SHIFT-10), text, lib,
62 		(PTRS_PER_PTE*sizeof(pte_t)*mm->nr_ptes) >> 10,
63 		swap << (PAGE_SHIFT-10));
64 }
65 
66 unsigned long task_vsize(struct mm_struct *mm)
67 {
68 	return PAGE_SIZE * mm->total_vm;
69 }
70 
71 unsigned long task_statm(struct mm_struct *mm,
72 			 unsigned long *shared, unsigned long *text,
73 			 unsigned long *data, unsigned long *resident)
74 {
75 	*shared = get_mm_counter(mm, MM_FILEPAGES);
76 	*text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
77 								>> PAGE_SHIFT;
78 	*data = mm->total_vm - mm->shared_vm;
79 	*resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
80 	return mm->total_vm;
81 }
82 
83 static void pad_len_spaces(struct seq_file *m, int len)
84 {
85 	len = 25 + sizeof(void*) * 6 - len;
86 	if (len < 1)
87 		len = 1;
88 	seq_printf(m, "%*c", len, ' ');
89 }
90 
91 static void vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma)
92 {
93 	if (vma && vma != priv->tail_vma) {
94 		struct mm_struct *mm = vma->vm_mm;
95 		up_read(&mm->mmap_sem);
96 		mmput(mm);
97 	}
98 }
99 
100 static void *m_start(struct seq_file *m, loff_t *pos)
101 {
102 	struct proc_maps_private *priv = m->private;
103 	unsigned long last_addr = m->version;
104 	struct mm_struct *mm;
105 	struct vm_area_struct *vma, *tail_vma = NULL;
106 	loff_t l = *pos;
107 
108 	/* Clear the per syscall fields in priv */
109 	priv->task = NULL;
110 	priv->tail_vma = NULL;
111 
112 	/*
113 	 * We remember last_addr rather than next_addr to hit with
114 	 * mmap_cache most of the time. We have zero last_addr at
115 	 * the beginning and also after lseek. We will have -1 last_addr
116 	 * after the end of the vmas.
117 	 */
118 
119 	if (last_addr == -1UL)
120 		return NULL;
121 
122 	priv->task = get_pid_task(priv->pid, PIDTYPE_PID);
123 	if (!priv->task)
124 		return ERR_PTR(-ESRCH);
125 
126 	mm = mm_for_maps(priv->task);
127 	if (!mm || IS_ERR(mm))
128 		return mm;
129 	down_read(&mm->mmap_sem);
130 
131 	tail_vma = get_gate_vma(priv->task->mm);
132 	priv->tail_vma = tail_vma;
133 
134 	/* Start with last addr hint */
135 	vma = find_vma(mm, last_addr);
136 	if (last_addr && vma) {
137 		vma = vma->vm_next;
138 		goto out;
139 	}
140 
141 	/*
142 	 * Check the vma index is within the range and do
143 	 * sequential scan until m_index.
144 	 */
145 	vma = NULL;
146 	if ((unsigned long)l < mm->map_count) {
147 		vma = mm->mmap;
148 		while (l-- && vma)
149 			vma = vma->vm_next;
150 		goto out;
151 	}
152 
153 	if (l != mm->map_count)
154 		tail_vma = NULL; /* After gate vma */
155 
156 out:
157 	if (vma)
158 		return vma;
159 
160 	/* End of vmas has been reached */
161 	m->version = (tail_vma != NULL)? 0: -1UL;
162 	up_read(&mm->mmap_sem);
163 	mmput(mm);
164 	return tail_vma;
165 }
166 
167 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
168 {
169 	struct proc_maps_private *priv = m->private;
170 	struct vm_area_struct *vma = v;
171 	struct vm_area_struct *tail_vma = priv->tail_vma;
172 
173 	(*pos)++;
174 	if (vma && (vma != tail_vma) && vma->vm_next)
175 		return vma->vm_next;
176 	vma_stop(priv, vma);
177 	return (vma != tail_vma)? tail_vma: NULL;
178 }
179 
180 static void m_stop(struct seq_file *m, void *v)
181 {
182 	struct proc_maps_private *priv = m->private;
183 	struct vm_area_struct *vma = v;
184 
185 	if (!IS_ERR(vma))
186 		vma_stop(priv, vma);
187 	if (priv->task)
188 		put_task_struct(priv->task);
189 }
190 
191 static int do_maps_open(struct inode *inode, struct file *file,
192 			const struct seq_operations *ops)
193 {
194 	struct proc_maps_private *priv;
195 	int ret = -ENOMEM;
196 	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
197 	if (priv) {
198 		priv->pid = proc_pid(inode);
199 		ret = seq_open(file, ops);
200 		if (!ret) {
201 			struct seq_file *m = file->private_data;
202 			m->private = priv;
203 		} else {
204 			kfree(priv);
205 		}
206 	}
207 	return ret;
208 }
209 
210 static void show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
211 {
212 	struct mm_struct *mm = vma->vm_mm;
213 	struct file *file = vma->vm_file;
214 	vm_flags_t flags = vma->vm_flags;
215 	unsigned long ino = 0;
216 	unsigned long long pgoff = 0;
217 	unsigned long start, end;
218 	dev_t dev = 0;
219 	int len;
220 
221 	if (file) {
222 		struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
223 		dev = inode->i_sb->s_dev;
224 		ino = inode->i_ino;
225 		pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
226 	}
227 
228 	/* We don't show the stack guard page in /proc/maps */
229 	start = vma->vm_start;
230 	if (stack_guard_page_start(vma, start))
231 		start += PAGE_SIZE;
232 	end = vma->vm_end;
233 	if (stack_guard_page_end(vma, end))
234 		end -= PAGE_SIZE;
235 
236 	seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu %n",
237 			start,
238 			end,
239 			flags & VM_READ ? 'r' : '-',
240 			flags & VM_WRITE ? 'w' : '-',
241 			flags & VM_EXEC ? 'x' : '-',
242 			flags & VM_MAYSHARE ? 's' : 'p',
243 			pgoff,
244 			MAJOR(dev), MINOR(dev), ino, &len);
245 
246 	/*
247 	 * Print the dentry name for named mappings, and a
248 	 * special [heap] marker for the heap:
249 	 */
250 	if (file) {
251 		pad_len_spaces(m, len);
252 		seq_path(m, &file->f_path, "\n");
253 	} else {
254 		const char *name = arch_vma_name(vma);
255 		if (!name) {
256 			if (mm) {
257 				if (vma->vm_start <= mm->brk &&
258 						vma->vm_end >= mm->start_brk) {
259 					name = "[heap]";
260 				} else if (vma->vm_start <= mm->start_stack &&
261 					   vma->vm_end >= mm->start_stack) {
262 					name = "[stack]";
263 				}
264 			} else {
265 				name = "[vdso]";
266 			}
267 		}
268 		if (name) {
269 			pad_len_spaces(m, len);
270 			seq_puts(m, name);
271 		}
272 	}
273 	seq_putc(m, '\n');
274 }
275 
276 static int show_map(struct seq_file *m, void *v)
277 {
278 	struct vm_area_struct *vma = v;
279 	struct proc_maps_private *priv = m->private;
280 	struct task_struct *task = priv->task;
281 
282 	show_map_vma(m, vma);
283 
284 	if (m->count < m->size)  /* vma is copied successfully */
285 		m->version = (vma != get_gate_vma(task->mm))
286 			? vma->vm_start : 0;
287 	return 0;
288 }
289 
290 static const struct seq_operations proc_pid_maps_op = {
291 	.start	= m_start,
292 	.next	= m_next,
293 	.stop	= m_stop,
294 	.show	= show_map
295 };
296 
297 static int maps_open(struct inode *inode, struct file *file)
298 {
299 	return do_maps_open(inode, file, &proc_pid_maps_op);
300 }
301 
302 const struct file_operations proc_maps_operations = {
303 	.open		= maps_open,
304 	.read		= seq_read,
305 	.llseek		= seq_lseek,
306 	.release	= seq_release_private,
307 };
308 
309 /*
310  * Proportional Set Size(PSS): my share of RSS.
311  *
312  * PSS of a process is the count of pages it has in memory, where each
313  * page is divided by the number of processes sharing it.  So if a
314  * process has 1000 pages all to itself, and 1000 shared with one other
315  * process, its PSS will be 1500.
316  *
317  * To keep (accumulated) division errors low, we adopt a 64bit
318  * fixed-point pss counter to minimize division errors. So (pss >>
319  * PSS_SHIFT) would be the real byte count.
320  *
321  * A shift of 12 before division means (assuming 4K page size):
322  * 	- 1M 3-user-pages add up to 8KB errors;
323  * 	- supports mapcount up to 2^24, or 16M;
324  * 	- supports PSS up to 2^52 bytes, or 4PB.
325  */
326 #define PSS_SHIFT 12
327 
328 #ifdef CONFIG_PROC_PAGE_MONITOR
329 struct mem_size_stats {
330 	struct vm_area_struct *vma;
331 	unsigned long resident;
332 	unsigned long shared_clean;
333 	unsigned long shared_dirty;
334 	unsigned long private_clean;
335 	unsigned long private_dirty;
336 	unsigned long referenced;
337 	unsigned long anonymous;
338 	unsigned long anonymous_thp;
339 	unsigned long swap;
340 	u64 pss;
341 };
342 
343 
344 static void smaps_pte_entry(pte_t ptent, unsigned long addr,
345 		unsigned long ptent_size, struct mm_walk *walk)
346 {
347 	struct mem_size_stats *mss = walk->private;
348 	struct vm_area_struct *vma = mss->vma;
349 	struct page *page;
350 	int mapcount;
351 
352 	if (is_swap_pte(ptent)) {
353 		mss->swap += ptent_size;
354 		return;
355 	}
356 
357 	if (!pte_present(ptent))
358 		return;
359 
360 	page = vm_normal_page(vma, addr, ptent);
361 	if (!page)
362 		return;
363 
364 	if (PageAnon(page))
365 		mss->anonymous += ptent_size;
366 
367 	mss->resident += ptent_size;
368 	/* Accumulate the size in pages that have been accessed. */
369 	if (pte_young(ptent) || PageReferenced(page))
370 		mss->referenced += ptent_size;
371 	mapcount = page_mapcount(page);
372 	if (mapcount >= 2) {
373 		if (pte_dirty(ptent) || PageDirty(page))
374 			mss->shared_dirty += ptent_size;
375 		else
376 			mss->shared_clean += ptent_size;
377 		mss->pss += (ptent_size << PSS_SHIFT) / mapcount;
378 	} else {
379 		if (pte_dirty(ptent) || PageDirty(page))
380 			mss->private_dirty += ptent_size;
381 		else
382 			mss->private_clean += ptent_size;
383 		mss->pss += (ptent_size << PSS_SHIFT);
384 	}
385 }
386 
387 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
388 			   struct mm_walk *walk)
389 {
390 	struct mem_size_stats *mss = walk->private;
391 	struct vm_area_struct *vma = mss->vma;
392 	pte_t *pte;
393 	spinlock_t *ptl;
394 
395 	spin_lock(&walk->mm->page_table_lock);
396 	if (pmd_trans_huge(*pmd)) {
397 		if (pmd_trans_splitting(*pmd)) {
398 			spin_unlock(&walk->mm->page_table_lock);
399 			wait_split_huge_page(vma->anon_vma, pmd);
400 		} else {
401 			smaps_pte_entry(*(pte_t *)pmd, addr,
402 					HPAGE_PMD_SIZE, walk);
403 			spin_unlock(&walk->mm->page_table_lock);
404 			mss->anonymous_thp += HPAGE_PMD_SIZE;
405 			return 0;
406 		}
407 	} else {
408 		spin_unlock(&walk->mm->page_table_lock);
409 	}
410 	/*
411 	 * The mmap_sem held all the way back in m_start() is what
412 	 * keeps khugepaged out of here and from collapsing things
413 	 * in here.
414 	 */
415 	pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
416 	for (; addr != end; pte++, addr += PAGE_SIZE)
417 		smaps_pte_entry(*pte, addr, PAGE_SIZE, walk);
418 	pte_unmap_unlock(pte - 1, ptl);
419 	cond_resched();
420 	return 0;
421 }
422 
423 static int show_smap(struct seq_file *m, void *v)
424 {
425 	struct proc_maps_private *priv = m->private;
426 	struct task_struct *task = priv->task;
427 	struct vm_area_struct *vma = v;
428 	struct mem_size_stats mss;
429 	struct mm_walk smaps_walk = {
430 		.pmd_entry = smaps_pte_range,
431 		.mm = vma->vm_mm,
432 		.private = &mss,
433 	};
434 
435 	memset(&mss, 0, sizeof mss);
436 	mss.vma = vma;
437 	/* mmap_sem is held in m_start */
438 	if (vma->vm_mm && !is_vm_hugetlb_page(vma))
439 		walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk);
440 
441 	show_map_vma(m, vma);
442 
443 	seq_printf(m,
444 		   "Size:           %8lu kB\n"
445 		   "Rss:            %8lu kB\n"
446 		   "Pss:            %8lu kB\n"
447 		   "Shared_Clean:   %8lu kB\n"
448 		   "Shared_Dirty:   %8lu kB\n"
449 		   "Private_Clean:  %8lu kB\n"
450 		   "Private_Dirty:  %8lu kB\n"
451 		   "Referenced:     %8lu kB\n"
452 		   "Anonymous:      %8lu kB\n"
453 		   "AnonHugePages:  %8lu kB\n"
454 		   "Swap:           %8lu kB\n"
455 		   "KernelPageSize: %8lu kB\n"
456 		   "MMUPageSize:    %8lu kB\n"
457 		   "Locked:         %8lu kB\n",
458 		   (vma->vm_end - vma->vm_start) >> 10,
459 		   mss.resident >> 10,
460 		   (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
461 		   mss.shared_clean  >> 10,
462 		   mss.shared_dirty  >> 10,
463 		   mss.private_clean >> 10,
464 		   mss.private_dirty >> 10,
465 		   mss.referenced >> 10,
466 		   mss.anonymous >> 10,
467 		   mss.anonymous_thp >> 10,
468 		   mss.swap >> 10,
469 		   vma_kernel_pagesize(vma) >> 10,
470 		   vma_mmu_pagesize(vma) >> 10,
471 		   (vma->vm_flags & VM_LOCKED) ?
472 			(unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
473 
474 	if (m->count < m->size)  /* vma is copied successfully */
475 		m->version = (vma != get_gate_vma(task->mm))
476 			? vma->vm_start : 0;
477 	return 0;
478 }
479 
480 static const struct seq_operations proc_pid_smaps_op = {
481 	.start	= m_start,
482 	.next	= m_next,
483 	.stop	= m_stop,
484 	.show	= show_smap
485 };
486 
487 static int smaps_open(struct inode *inode, struct file *file)
488 {
489 	return do_maps_open(inode, file, &proc_pid_smaps_op);
490 }
491 
492 const struct file_operations proc_smaps_operations = {
493 	.open		= smaps_open,
494 	.read		= seq_read,
495 	.llseek		= seq_lseek,
496 	.release	= seq_release_private,
497 };
498 
499 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
500 				unsigned long end, struct mm_walk *walk)
501 {
502 	struct vm_area_struct *vma = walk->private;
503 	pte_t *pte, ptent;
504 	spinlock_t *ptl;
505 	struct page *page;
506 
507 	split_huge_page_pmd(walk->mm, pmd);
508 
509 	pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
510 	for (; addr != end; pte++, addr += PAGE_SIZE) {
511 		ptent = *pte;
512 		if (!pte_present(ptent))
513 			continue;
514 
515 		page = vm_normal_page(vma, addr, ptent);
516 		if (!page)
517 			continue;
518 
519 		/* Clear accessed and referenced bits. */
520 		ptep_test_and_clear_young(vma, addr, pte);
521 		ClearPageReferenced(page);
522 	}
523 	pte_unmap_unlock(pte - 1, ptl);
524 	cond_resched();
525 	return 0;
526 }
527 
528 #define CLEAR_REFS_ALL 1
529 #define CLEAR_REFS_ANON 2
530 #define CLEAR_REFS_MAPPED 3
531 
532 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
533 				size_t count, loff_t *ppos)
534 {
535 	struct task_struct *task;
536 	char buffer[PROC_NUMBUF];
537 	struct mm_struct *mm;
538 	struct vm_area_struct *vma;
539 	int type;
540 	int rv;
541 
542 	memset(buffer, 0, sizeof(buffer));
543 	if (count > sizeof(buffer) - 1)
544 		count = sizeof(buffer) - 1;
545 	if (copy_from_user(buffer, buf, count))
546 		return -EFAULT;
547 	rv = kstrtoint(strstrip(buffer), 10, &type);
548 	if (rv < 0)
549 		return rv;
550 	if (type < CLEAR_REFS_ALL || type > CLEAR_REFS_MAPPED)
551 		return -EINVAL;
552 	task = get_proc_task(file->f_path.dentry->d_inode);
553 	if (!task)
554 		return -ESRCH;
555 	mm = get_task_mm(task);
556 	if (mm) {
557 		struct mm_walk clear_refs_walk = {
558 			.pmd_entry = clear_refs_pte_range,
559 			.mm = mm,
560 		};
561 		down_read(&mm->mmap_sem);
562 		for (vma = mm->mmap; vma; vma = vma->vm_next) {
563 			clear_refs_walk.private = vma;
564 			if (is_vm_hugetlb_page(vma))
565 				continue;
566 			/*
567 			 * Writing 1 to /proc/pid/clear_refs affects all pages.
568 			 *
569 			 * Writing 2 to /proc/pid/clear_refs only affects
570 			 * Anonymous pages.
571 			 *
572 			 * Writing 3 to /proc/pid/clear_refs only affects file
573 			 * mapped pages.
574 			 */
575 			if (type == CLEAR_REFS_ANON && vma->vm_file)
576 				continue;
577 			if (type == CLEAR_REFS_MAPPED && !vma->vm_file)
578 				continue;
579 			walk_page_range(vma->vm_start, vma->vm_end,
580 					&clear_refs_walk);
581 		}
582 		flush_tlb_mm(mm);
583 		up_read(&mm->mmap_sem);
584 		mmput(mm);
585 	}
586 	put_task_struct(task);
587 
588 	return count;
589 }
590 
591 const struct file_operations proc_clear_refs_operations = {
592 	.write		= clear_refs_write,
593 	.llseek		= noop_llseek,
594 };
595 
596 struct pagemapread {
597 	int pos, len;
598 	u64 *buffer;
599 };
600 
601 #define PM_ENTRY_BYTES      sizeof(u64)
602 #define PM_STATUS_BITS      3
603 #define PM_STATUS_OFFSET    (64 - PM_STATUS_BITS)
604 #define PM_STATUS_MASK      (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
605 #define PM_STATUS(nr)       (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
606 #define PM_PSHIFT_BITS      6
607 #define PM_PSHIFT_OFFSET    (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
608 #define PM_PSHIFT_MASK      (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
609 #define PM_PSHIFT(x)        (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
610 #define PM_PFRAME_MASK      ((1LL << PM_PSHIFT_OFFSET) - 1)
611 #define PM_PFRAME(x)        ((x) & PM_PFRAME_MASK)
612 
613 #define PM_PRESENT          PM_STATUS(4LL)
614 #define PM_SWAP             PM_STATUS(2LL)
615 #define PM_NOT_PRESENT      PM_PSHIFT(PAGE_SHIFT)
616 #define PM_END_OF_BUFFER    1
617 
618 static int add_to_pagemap(unsigned long addr, u64 pfn,
619 			  struct pagemapread *pm)
620 {
621 	pm->buffer[pm->pos++] = pfn;
622 	if (pm->pos >= pm->len)
623 		return PM_END_OF_BUFFER;
624 	return 0;
625 }
626 
627 static int pagemap_pte_hole(unsigned long start, unsigned long end,
628 				struct mm_walk *walk)
629 {
630 	struct pagemapread *pm = walk->private;
631 	unsigned long addr;
632 	int err = 0;
633 	for (addr = start; addr < end; addr += PAGE_SIZE) {
634 		err = add_to_pagemap(addr, PM_NOT_PRESENT, pm);
635 		if (err)
636 			break;
637 	}
638 	return err;
639 }
640 
641 static u64 swap_pte_to_pagemap_entry(pte_t pte)
642 {
643 	swp_entry_t e = pte_to_swp_entry(pte);
644 	return swp_type(e) | (swp_offset(e) << MAX_SWAPFILES_SHIFT);
645 }
646 
647 static u64 pte_to_pagemap_entry(pte_t pte)
648 {
649 	u64 pme = 0;
650 	if (is_swap_pte(pte))
651 		pme = PM_PFRAME(swap_pte_to_pagemap_entry(pte))
652 			| PM_PSHIFT(PAGE_SHIFT) | PM_SWAP;
653 	else if (pte_present(pte))
654 		pme = PM_PFRAME(pte_pfn(pte))
655 			| PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
656 	return pme;
657 }
658 
659 static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
660 			     struct mm_walk *walk)
661 {
662 	struct vm_area_struct *vma;
663 	struct pagemapread *pm = walk->private;
664 	pte_t *pte;
665 	int err = 0;
666 
667 	split_huge_page_pmd(walk->mm, pmd);
668 
669 	/* find the first VMA at or above 'addr' */
670 	vma = find_vma(walk->mm, addr);
671 	for (; addr != end; addr += PAGE_SIZE) {
672 		u64 pfn = PM_NOT_PRESENT;
673 
674 		/* check to see if we've left 'vma' behind
675 		 * and need a new, higher one */
676 		if (vma && (addr >= vma->vm_end))
677 			vma = find_vma(walk->mm, addr);
678 
679 		/* check that 'vma' actually covers this address,
680 		 * and that it isn't a huge page vma */
681 		if (vma && (vma->vm_start <= addr) &&
682 		    !is_vm_hugetlb_page(vma)) {
683 			pte = pte_offset_map(pmd, addr);
684 			pfn = pte_to_pagemap_entry(*pte);
685 			/* unmap before userspace copy */
686 			pte_unmap(pte);
687 		}
688 		err = add_to_pagemap(addr, pfn, pm);
689 		if (err)
690 			return err;
691 	}
692 
693 	cond_resched();
694 
695 	return err;
696 }
697 
698 #ifdef CONFIG_HUGETLB_PAGE
699 static u64 huge_pte_to_pagemap_entry(pte_t pte, int offset)
700 {
701 	u64 pme = 0;
702 	if (pte_present(pte))
703 		pme = PM_PFRAME(pte_pfn(pte) + offset)
704 			| PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
705 	return pme;
706 }
707 
708 /* This function walks within one hugetlb entry in the single call */
709 static int pagemap_hugetlb_range(pte_t *pte, unsigned long hmask,
710 				 unsigned long addr, unsigned long end,
711 				 struct mm_walk *walk)
712 {
713 	struct pagemapread *pm = walk->private;
714 	int err = 0;
715 	u64 pfn;
716 
717 	for (; addr != end; addr += PAGE_SIZE) {
718 		int offset = (addr & ~hmask) >> PAGE_SHIFT;
719 		pfn = huge_pte_to_pagemap_entry(*pte, offset);
720 		err = add_to_pagemap(addr, pfn, pm);
721 		if (err)
722 			return err;
723 	}
724 
725 	cond_resched();
726 
727 	return err;
728 }
729 #endif /* HUGETLB_PAGE */
730 
731 /*
732  * /proc/pid/pagemap - an array mapping virtual pages to pfns
733  *
734  * For each page in the address space, this file contains one 64-bit entry
735  * consisting of the following:
736  *
737  * Bits 0-55  page frame number (PFN) if present
738  * Bits 0-4   swap type if swapped
739  * Bits 5-55  swap offset if swapped
740  * Bits 55-60 page shift (page size = 1<<page shift)
741  * Bit  61    reserved for future use
742  * Bit  62    page swapped
743  * Bit  63    page present
744  *
745  * If the page is not present but in swap, then the PFN contains an
746  * encoding of the swap file number and the page's offset into the
747  * swap. Unmapped pages return a null PFN. This allows determining
748  * precisely which pages are mapped (or in swap) and comparing mapped
749  * pages between processes.
750  *
751  * Efficient users of this interface will use /proc/pid/maps to
752  * determine which areas of memory are actually mapped and llseek to
753  * skip over unmapped regions.
754  */
755 #define PAGEMAP_WALK_SIZE	(PMD_SIZE)
756 #define PAGEMAP_WALK_MASK	(PMD_MASK)
757 static ssize_t pagemap_read(struct file *file, char __user *buf,
758 			    size_t count, loff_t *ppos)
759 {
760 	struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
761 	struct mm_struct *mm;
762 	struct pagemapread pm;
763 	int ret = -ESRCH;
764 	struct mm_walk pagemap_walk = {};
765 	unsigned long src;
766 	unsigned long svpfn;
767 	unsigned long start_vaddr;
768 	unsigned long end_vaddr;
769 	int copied = 0;
770 
771 	if (!task)
772 		goto out;
773 
774 	ret = -EINVAL;
775 	/* file position must be aligned */
776 	if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
777 		goto out_task;
778 
779 	ret = 0;
780 	if (!count)
781 		goto out_task;
782 
783 	pm.len = PM_ENTRY_BYTES * (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
784 	pm.buffer = kmalloc(pm.len, GFP_TEMPORARY);
785 	ret = -ENOMEM;
786 	if (!pm.buffer)
787 		goto out_task;
788 
789 	mm = mm_for_maps(task);
790 	ret = PTR_ERR(mm);
791 	if (!mm || IS_ERR(mm))
792 		goto out_free;
793 
794 	pagemap_walk.pmd_entry = pagemap_pte_range;
795 	pagemap_walk.pte_hole = pagemap_pte_hole;
796 #ifdef CONFIG_HUGETLB_PAGE
797 	pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
798 #endif
799 	pagemap_walk.mm = mm;
800 	pagemap_walk.private = &pm;
801 
802 	src = *ppos;
803 	svpfn = src / PM_ENTRY_BYTES;
804 	start_vaddr = svpfn << PAGE_SHIFT;
805 	end_vaddr = TASK_SIZE_OF(task);
806 
807 	/* watch out for wraparound */
808 	if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
809 		start_vaddr = end_vaddr;
810 
811 	/*
812 	 * The odds are that this will stop walking way
813 	 * before end_vaddr, because the length of the
814 	 * user buffer is tracked in "pm", and the walk
815 	 * will stop when we hit the end of the buffer.
816 	 */
817 	ret = 0;
818 	while (count && (start_vaddr < end_vaddr)) {
819 		int len;
820 		unsigned long end;
821 
822 		pm.pos = 0;
823 		end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
824 		/* overflow ? */
825 		if (end < start_vaddr || end > end_vaddr)
826 			end = end_vaddr;
827 		down_read(&mm->mmap_sem);
828 		ret = walk_page_range(start_vaddr, end, &pagemap_walk);
829 		up_read(&mm->mmap_sem);
830 		start_vaddr = end;
831 
832 		len = min(count, PM_ENTRY_BYTES * pm.pos);
833 		if (copy_to_user(buf, pm.buffer, len)) {
834 			ret = -EFAULT;
835 			goto out_mm;
836 		}
837 		copied += len;
838 		buf += len;
839 		count -= len;
840 	}
841 	*ppos += copied;
842 	if (!ret || ret == PM_END_OF_BUFFER)
843 		ret = copied;
844 
845 out_mm:
846 	mmput(mm);
847 out_free:
848 	kfree(pm.buffer);
849 out_task:
850 	put_task_struct(task);
851 out:
852 	return ret;
853 }
854 
855 const struct file_operations proc_pagemap_operations = {
856 	.llseek		= mem_lseek, /* borrow this */
857 	.read		= pagemap_read,
858 };
859 #endif /* CONFIG_PROC_PAGE_MONITOR */
860 
861 #ifdef CONFIG_NUMA
862 
863 struct numa_maps {
864 	struct vm_area_struct *vma;
865 	unsigned long pages;
866 	unsigned long anon;
867 	unsigned long active;
868 	unsigned long writeback;
869 	unsigned long mapcount_max;
870 	unsigned long dirty;
871 	unsigned long swapcache;
872 	unsigned long node[MAX_NUMNODES];
873 };
874 
875 struct numa_maps_private {
876 	struct proc_maps_private proc_maps;
877 	struct numa_maps md;
878 };
879 
880 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
881 			unsigned long nr_pages)
882 {
883 	int count = page_mapcount(page);
884 
885 	md->pages += nr_pages;
886 	if (pte_dirty || PageDirty(page))
887 		md->dirty += nr_pages;
888 
889 	if (PageSwapCache(page))
890 		md->swapcache += nr_pages;
891 
892 	if (PageActive(page) || PageUnevictable(page))
893 		md->active += nr_pages;
894 
895 	if (PageWriteback(page))
896 		md->writeback += nr_pages;
897 
898 	if (PageAnon(page))
899 		md->anon += nr_pages;
900 
901 	if (count > md->mapcount_max)
902 		md->mapcount_max = count;
903 
904 	md->node[page_to_nid(page)] += nr_pages;
905 }
906 
907 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
908 		unsigned long addr)
909 {
910 	struct page *page;
911 	int nid;
912 
913 	if (!pte_present(pte))
914 		return NULL;
915 
916 	page = vm_normal_page(vma, addr, pte);
917 	if (!page)
918 		return NULL;
919 
920 	if (PageReserved(page))
921 		return NULL;
922 
923 	nid = page_to_nid(page);
924 	if (!node_isset(nid, node_states[N_HIGH_MEMORY]))
925 		return NULL;
926 
927 	return page;
928 }
929 
930 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
931 		unsigned long end, struct mm_walk *walk)
932 {
933 	struct numa_maps *md;
934 	spinlock_t *ptl;
935 	pte_t *orig_pte;
936 	pte_t *pte;
937 
938 	md = walk->private;
939 	spin_lock(&walk->mm->page_table_lock);
940 	if (pmd_trans_huge(*pmd)) {
941 		if (pmd_trans_splitting(*pmd)) {
942 			spin_unlock(&walk->mm->page_table_lock);
943 			wait_split_huge_page(md->vma->anon_vma, pmd);
944 		} else {
945 			pte_t huge_pte = *(pte_t *)pmd;
946 			struct page *page;
947 
948 			page = can_gather_numa_stats(huge_pte, md->vma, addr);
949 			if (page)
950 				gather_stats(page, md, pte_dirty(huge_pte),
951 						HPAGE_PMD_SIZE/PAGE_SIZE);
952 			spin_unlock(&walk->mm->page_table_lock);
953 			return 0;
954 		}
955 	} else {
956 		spin_unlock(&walk->mm->page_table_lock);
957 	}
958 
959 	orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
960 	do {
961 		struct page *page = can_gather_numa_stats(*pte, md->vma, addr);
962 		if (!page)
963 			continue;
964 		gather_stats(page, md, pte_dirty(*pte), 1);
965 
966 	} while (pte++, addr += PAGE_SIZE, addr != end);
967 	pte_unmap_unlock(orig_pte, ptl);
968 	return 0;
969 }
970 #ifdef CONFIG_HUGETLB_PAGE
971 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
972 		unsigned long addr, unsigned long end, struct mm_walk *walk)
973 {
974 	struct numa_maps *md;
975 	struct page *page;
976 
977 	if (pte_none(*pte))
978 		return 0;
979 
980 	page = pte_page(*pte);
981 	if (!page)
982 		return 0;
983 
984 	md = walk->private;
985 	gather_stats(page, md, pte_dirty(*pte), 1);
986 	return 0;
987 }
988 
989 #else
990 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
991 		unsigned long addr, unsigned long end, struct mm_walk *walk)
992 {
993 	return 0;
994 }
995 #endif
996 
997 /*
998  * Display pages allocated per node and memory policy via /proc.
999  */
1000 static int show_numa_map(struct seq_file *m, void *v)
1001 {
1002 	struct numa_maps_private *numa_priv = m->private;
1003 	struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1004 	struct vm_area_struct *vma = v;
1005 	struct numa_maps *md = &numa_priv->md;
1006 	struct file *file = vma->vm_file;
1007 	struct mm_struct *mm = vma->vm_mm;
1008 	struct mm_walk walk = {};
1009 	struct mempolicy *pol;
1010 	int n;
1011 	char buffer[50];
1012 
1013 	if (!mm)
1014 		return 0;
1015 
1016 	/* Ensure we start with an empty set of numa_maps statistics. */
1017 	memset(md, 0, sizeof(*md));
1018 
1019 	md->vma = vma;
1020 
1021 	walk.hugetlb_entry = gather_hugetbl_stats;
1022 	walk.pmd_entry = gather_pte_stats;
1023 	walk.private = md;
1024 	walk.mm = mm;
1025 
1026 	pol = get_vma_policy(proc_priv->task, vma, vma->vm_start);
1027 	mpol_to_str(buffer, sizeof(buffer), pol, 0);
1028 	mpol_cond_put(pol);
1029 
1030 	seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1031 
1032 	if (file) {
1033 		seq_printf(m, " file=");
1034 		seq_path(m, &file->f_path, "\n\t= ");
1035 	} else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1036 		seq_printf(m, " heap");
1037 	} else if (vma->vm_start <= mm->start_stack &&
1038 			vma->vm_end >= mm->start_stack) {
1039 		seq_printf(m, " stack");
1040 	}
1041 
1042 	walk_page_range(vma->vm_start, vma->vm_end, &walk);
1043 
1044 	if (!md->pages)
1045 		goto out;
1046 
1047 	if (md->anon)
1048 		seq_printf(m, " anon=%lu", md->anon);
1049 
1050 	if (md->dirty)
1051 		seq_printf(m, " dirty=%lu", md->dirty);
1052 
1053 	if (md->pages != md->anon && md->pages != md->dirty)
1054 		seq_printf(m, " mapped=%lu", md->pages);
1055 
1056 	if (md->mapcount_max > 1)
1057 		seq_printf(m, " mapmax=%lu", md->mapcount_max);
1058 
1059 	if (md->swapcache)
1060 		seq_printf(m, " swapcache=%lu", md->swapcache);
1061 
1062 	if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1063 		seq_printf(m, " active=%lu", md->active);
1064 
1065 	if (md->writeback)
1066 		seq_printf(m, " writeback=%lu", md->writeback);
1067 
1068 	for_each_node_state(n, N_HIGH_MEMORY)
1069 		if (md->node[n])
1070 			seq_printf(m, " N%d=%lu", n, md->node[n]);
1071 out:
1072 	seq_putc(m, '\n');
1073 
1074 	if (m->count < m->size)
1075 		m->version = (vma != proc_priv->tail_vma) ? vma->vm_start : 0;
1076 	return 0;
1077 }
1078 
1079 static const struct seq_operations proc_pid_numa_maps_op = {
1080         .start  = m_start,
1081         .next   = m_next,
1082         .stop   = m_stop,
1083         .show   = show_numa_map,
1084 };
1085 
1086 static int numa_maps_open(struct inode *inode, struct file *file)
1087 {
1088 	struct numa_maps_private *priv;
1089 	int ret = -ENOMEM;
1090 	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
1091 	if (priv) {
1092 		priv->proc_maps.pid = proc_pid(inode);
1093 		ret = seq_open(file, &proc_pid_numa_maps_op);
1094 		if (!ret) {
1095 			struct seq_file *m = file->private_data;
1096 			m->private = priv;
1097 		} else {
1098 			kfree(priv);
1099 		}
1100 	}
1101 	return ret;
1102 }
1103 
1104 const struct file_operations proc_numa_maps_operations = {
1105 	.open		= numa_maps_open,
1106 	.read		= seq_read,
1107 	.llseek		= seq_lseek,
1108 	.release	= seq_release_private,
1109 };
1110 #endif /* CONFIG_NUMA */
1111