xref: /linux/fs/proc/task_mmu.c (revision 27258e448eb301cf89e351df87aa8cb916653bf2)
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;
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 	seq_printf(m,
40 		"VmPeak:\t%8lu kB\n"
41 		"VmSize:\t%8lu kB\n"
42 		"VmLck:\t%8lu kB\n"
43 		"VmHWM:\t%8lu kB\n"
44 		"VmRSS:\t%8lu kB\n"
45 		"VmData:\t%8lu kB\n"
46 		"VmStk:\t%8lu kB\n"
47 		"VmExe:\t%8lu kB\n"
48 		"VmLib:\t%8lu kB\n"
49 		"VmPTE:\t%8lu kB\n",
50 		hiwater_vm << (PAGE_SHIFT-10),
51 		(total_vm - mm->reserved_vm) << (PAGE_SHIFT-10),
52 		mm->locked_vm << (PAGE_SHIFT-10),
53 		hiwater_rss << (PAGE_SHIFT-10),
54 		total_rss << (PAGE_SHIFT-10),
55 		data << (PAGE_SHIFT-10),
56 		mm->stack_vm << (PAGE_SHIFT-10), text, lib,
57 		(PTRS_PER_PTE*sizeof(pte_t)*mm->nr_ptes) >> 10);
58 }
59 
60 unsigned long task_vsize(struct mm_struct *mm)
61 {
62 	return PAGE_SIZE * mm->total_vm;
63 }
64 
65 int task_statm(struct mm_struct *mm, int *shared, int *text,
66 	       int *data, int *resident)
67 {
68 	*shared = get_mm_counter(mm, file_rss);
69 	*text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
70 								>> PAGE_SHIFT;
71 	*data = mm->total_vm - mm->shared_vm;
72 	*resident = *shared + get_mm_counter(mm, anon_rss);
73 	return mm->total_vm;
74 }
75 
76 static void pad_len_spaces(struct seq_file *m, int len)
77 {
78 	len = 25 + sizeof(void*) * 6 - len;
79 	if (len < 1)
80 		len = 1;
81 	seq_printf(m, "%*c", len, ' ');
82 }
83 
84 static void vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma)
85 {
86 	if (vma && vma != priv->tail_vma) {
87 		struct mm_struct *mm = vma->vm_mm;
88 		up_read(&mm->mmap_sem);
89 		mmput(mm);
90 	}
91 }
92 
93 static void *m_start(struct seq_file *m, loff_t *pos)
94 {
95 	struct proc_maps_private *priv = m->private;
96 	unsigned long last_addr = m->version;
97 	struct mm_struct *mm;
98 	struct vm_area_struct *vma, *tail_vma = NULL;
99 	loff_t l = *pos;
100 
101 	/* Clear the per syscall fields in priv */
102 	priv->task = NULL;
103 	priv->tail_vma = NULL;
104 
105 	/*
106 	 * We remember last_addr rather than next_addr to hit with
107 	 * mmap_cache most of the time. We have zero last_addr at
108 	 * the beginning and also after lseek. We will have -1 last_addr
109 	 * after the end of the vmas.
110 	 */
111 
112 	if (last_addr == -1UL)
113 		return NULL;
114 
115 	priv->task = get_pid_task(priv->pid, PIDTYPE_PID);
116 	if (!priv->task)
117 		return NULL;
118 
119 	mm = mm_for_maps(priv->task);
120 	if (!mm)
121 		return NULL;
122 	down_read(&mm->mmap_sem);
123 
124 	tail_vma = get_gate_vma(priv->task);
125 	priv->tail_vma = tail_vma;
126 
127 	/* Start with last addr hint */
128 	vma = find_vma(mm, last_addr);
129 	if (last_addr && vma) {
130 		vma = vma->vm_next;
131 		goto out;
132 	}
133 
134 	/*
135 	 * Check the vma index is within the range and do
136 	 * sequential scan until m_index.
137 	 */
138 	vma = NULL;
139 	if ((unsigned long)l < mm->map_count) {
140 		vma = mm->mmap;
141 		while (l-- && vma)
142 			vma = vma->vm_next;
143 		goto out;
144 	}
145 
146 	if (l != mm->map_count)
147 		tail_vma = NULL; /* After gate vma */
148 
149 out:
150 	if (vma)
151 		return vma;
152 
153 	/* End of vmas has been reached */
154 	m->version = (tail_vma != NULL)? 0: -1UL;
155 	up_read(&mm->mmap_sem);
156 	mmput(mm);
157 	return tail_vma;
158 }
159 
160 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
161 {
162 	struct proc_maps_private *priv = m->private;
163 	struct vm_area_struct *vma = v;
164 	struct vm_area_struct *tail_vma = priv->tail_vma;
165 
166 	(*pos)++;
167 	if (vma && (vma != tail_vma) && vma->vm_next)
168 		return vma->vm_next;
169 	vma_stop(priv, vma);
170 	return (vma != tail_vma)? tail_vma: NULL;
171 }
172 
173 static void m_stop(struct seq_file *m, void *v)
174 {
175 	struct proc_maps_private *priv = m->private;
176 	struct vm_area_struct *vma = v;
177 
178 	vma_stop(priv, vma);
179 	if (priv->task)
180 		put_task_struct(priv->task);
181 }
182 
183 static int do_maps_open(struct inode *inode, struct file *file,
184 			const struct seq_operations *ops)
185 {
186 	struct proc_maps_private *priv;
187 	int ret = -ENOMEM;
188 	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
189 	if (priv) {
190 		priv->pid = proc_pid(inode);
191 		ret = seq_open(file, ops);
192 		if (!ret) {
193 			struct seq_file *m = file->private_data;
194 			m->private = priv;
195 		} else {
196 			kfree(priv);
197 		}
198 	}
199 	return ret;
200 }
201 
202 static void show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
203 {
204 	struct mm_struct *mm = vma->vm_mm;
205 	struct file *file = vma->vm_file;
206 	int flags = vma->vm_flags;
207 	unsigned long ino = 0;
208 	unsigned long long pgoff = 0;
209 	dev_t dev = 0;
210 	int len;
211 
212 	if (file) {
213 		struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
214 		dev = inode->i_sb->s_dev;
215 		ino = inode->i_ino;
216 		pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
217 	}
218 
219 	seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu %n",
220 			vma->vm_start,
221 			vma->vm_end,
222 			flags & VM_READ ? 'r' : '-',
223 			flags & VM_WRITE ? 'w' : '-',
224 			flags & VM_EXEC ? 'x' : '-',
225 			flags & VM_MAYSHARE ? 's' : 'p',
226 			pgoff,
227 			MAJOR(dev), MINOR(dev), ino, &len);
228 
229 	/*
230 	 * Print the dentry name for named mappings, and a
231 	 * special [heap] marker for the heap:
232 	 */
233 	if (file) {
234 		pad_len_spaces(m, len);
235 		seq_path(m, &file->f_path, "\n");
236 	} else {
237 		const char *name = arch_vma_name(vma);
238 		if (!name) {
239 			if (mm) {
240 				if (vma->vm_start <= mm->start_brk &&
241 						vma->vm_end >= mm->brk) {
242 					name = "[heap]";
243 				} else if (vma->vm_start <= mm->start_stack &&
244 					   vma->vm_end >= mm->start_stack) {
245 					name = "[stack]";
246 				}
247 			} else {
248 				name = "[vdso]";
249 			}
250 		}
251 		if (name) {
252 			pad_len_spaces(m, len);
253 			seq_puts(m, name);
254 		}
255 	}
256 	seq_putc(m, '\n');
257 }
258 
259 static int show_map(struct seq_file *m, void *v)
260 {
261 	struct vm_area_struct *vma = v;
262 	struct proc_maps_private *priv = m->private;
263 	struct task_struct *task = priv->task;
264 
265 	show_map_vma(m, vma);
266 
267 	if (m->count < m->size)  /* vma is copied successfully */
268 		m->version = (vma != get_gate_vma(task))? vma->vm_start: 0;
269 	return 0;
270 }
271 
272 static const struct seq_operations proc_pid_maps_op = {
273 	.start	= m_start,
274 	.next	= m_next,
275 	.stop	= m_stop,
276 	.show	= show_map
277 };
278 
279 static int maps_open(struct inode *inode, struct file *file)
280 {
281 	return do_maps_open(inode, file, &proc_pid_maps_op);
282 }
283 
284 const struct file_operations proc_maps_operations = {
285 	.open		= maps_open,
286 	.read		= seq_read,
287 	.llseek		= seq_lseek,
288 	.release	= seq_release_private,
289 };
290 
291 /*
292  * Proportional Set Size(PSS): my share of RSS.
293  *
294  * PSS of a process is the count of pages it has in memory, where each
295  * page is divided by the number of processes sharing it.  So if a
296  * process has 1000 pages all to itself, and 1000 shared with one other
297  * process, its PSS will be 1500.
298  *
299  * To keep (accumulated) division errors low, we adopt a 64bit
300  * fixed-point pss counter to minimize division errors. So (pss >>
301  * PSS_SHIFT) would be the real byte count.
302  *
303  * A shift of 12 before division means (assuming 4K page size):
304  * 	- 1M 3-user-pages add up to 8KB errors;
305  * 	- supports mapcount up to 2^24, or 16M;
306  * 	- supports PSS up to 2^52 bytes, or 4PB.
307  */
308 #define PSS_SHIFT 12
309 
310 #ifdef CONFIG_PROC_PAGE_MONITOR
311 struct mem_size_stats {
312 	struct vm_area_struct *vma;
313 	unsigned long resident;
314 	unsigned long shared_clean;
315 	unsigned long shared_dirty;
316 	unsigned long private_clean;
317 	unsigned long private_dirty;
318 	unsigned long referenced;
319 	unsigned long swap;
320 	u64 pss;
321 };
322 
323 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
324 			   struct mm_walk *walk)
325 {
326 	struct mem_size_stats *mss = walk->private;
327 	struct vm_area_struct *vma = mss->vma;
328 	pte_t *pte, ptent;
329 	spinlock_t *ptl;
330 	struct page *page;
331 	int mapcount;
332 
333 	pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
334 	for (; addr != end; pte++, addr += PAGE_SIZE) {
335 		ptent = *pte;
336 
337 		if (is_swap_pte(ptent)) {
338 			mss->swap += PAGE_SIZE;
339 			continue;
340 		}
341 
342 		if (!pte_present(ptent))
343 			continue;
344 
345 		mss->resident += PAGE_SIZE;
346 
347 		page = vm_normal_page(vma, addr, ptent);
348 		if (!page)
349 			continue;
350 
351 		/* Accumulate the size in pages that have been accessed. */
352 		if (pte_young(ptent) || PageReferenced(page))
353 			mss->referenced += PAGE_SIZE;
354 		mapcount = page_mapcount(page);
355 		if (mapcount >= 2) {
356 			if (pte_dirty(ptent))
357 				mss->shared_dirty += PAGE_SIZE;
358 			else
359 				mss->shared_clean += PAGE_SIZE;
360 			mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount;
361 		} else {
362 			if (pte_dirty(ptent))
363 				mss->private_dirty += PAGE_SIZE;
364 			else
365 				mss->private_clean += PAGE_SIZE;
366 			mss->pss += (PAGE_SIZE << PSS_SHIFT);
367 		}
368 	}
369 	pte_unmap_unlock(pte - 1, ptl);
370 	cond_resched();
371 	return 0;
372 }
373 
374 static int show_smap(struct seq_file *m, void *v)
375 {
376 	struct proc_maps_private *priv = m->private;
377 	struct task_struct *task = priv->task;
378 	struct vm_area_struct *vma = v;
379 	struct mem_size_stats mss;
380 	struct mm_walk smaps_walk = {
381 		.pmd_entry = smaps_pte_range,
382 		.mm = vma->vm_mm,
383 		.private = &mss,
384 	};
385 
386 	memset(&mss, 0, sizeof mss);
387 	mss.vma = vma;
388 	if (vma->vm_mm && !is_vm_hugetlb_page(vma))
389 		walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk);
390 
391 	show_map_vma(m, vma);
392 
393 	seq_printf(m,
394 		   "Size:           %8lu kB\n"
395 		   "Rss:            %8lu kB\n"
396 		   "Pss:            %8lu kB\n"
397 		   "Shared_Clean:   %8lu kB\n"
398 		   "Shared_Dirty:   %8lu kB\n"
399 		   "Private_Clean:  %8lu kB\n"
400 		   "Private_Dirty:  %8lu kB\n"
401 		   "Referenced:     %8lu kB\n"
402 		   "Swap:           %8lu kB\n"
403 		   "KernelPageSize: %8lu kB\n"
404 		   "MMUPageSize:    %8lu kB\n",
405 		   (vma->vm_end - vma->vm_start) >> 10,
406 		   mss.resident >> 10,
407 		   (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
408 		   mss.shared_clean  >> 10,
409 		   mss.shared_dirty  >> 10,
410 		   mss.private_clean >> 10,
411 		   mss.private_dirty >> 10,
412 		   mss.referenced >> 10,
413 		   mss.swap >> 10,
414 		   vma_kernel_pagesize(vma) >> 10,
415 		   vma_mmu_pagesize(vma) >> 10);
416 
417 	if (m->count < m->size)  /* vma is copied successfully */
418 		m->version = (vma != get_gate_vma(task)) ? vma->vm_start : 0;
419 	return 0;
420 }
421 
422 static const struct seq_operations proc_pid_smaps_op = {
423 	.start	= m_start,
424 	.next	= m_next,
425 	.stop	= m_stop,
426 	.show	= show_smap
427 };
428 
429 static int smaps_open(struct inode *inode, struct file *file)
430 {
431 	return do_maps_open(inode, file, &proc_pid_smaps_op);
432 }
433 
434 const struct file_operations proc_smaps_operations = {
435 	.open		= smaps_open,
436 	.read		= seq_read,
437 	.llseek		= seq_lseek,
438 	.release	= seq_release_private,
439 };
440 
441 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
442 				unsigned long end, struct mm_walk *walk)
443 {
444 	struct vm_area_struct *vma = walk->private;
445 	pte_t *pte, ptent;
446 	spinlock_t *ptl;
447 	struct page *page;
448 
449 	pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
450 	for (; addr != end; pte++, addr += PAGE_SIZE) {
451 		ptent = *pte;
452 		if (!pte_present(ptent))
453 			continue;
454 
455 		page = vm_normal_page(vma, addr, ptent);
456 		if (!page)
457 			continue;
458 
459 		/* Clear accessed and referenced bits. */
460 		ptep_test_and_clear_young(vma, addr, pte);
461 		ClearPageReferenced(page);
462 	}
463 	pte_unmap_unlock(pte - 1, ptl);
464 	cond_resched();
465 	return 0;
466 }
467 
468 #define CLEAR_REFS_ALL 1
469 #define CLEAR_REFS_ANON 2
470 #define CLEAR_REFS_MAPPED 3
471 
472 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
473 				size_t count, loff_t *ppos)
474 {
475 	struct task_struct *task;
476 	char buffer[PROC_NUMBUF], *end;
477 	struct mm_struct *mm;
478 	struct vm_area_struct *vma;
479 	int type;
480 
481 	memset(buffer, 0, sizeof(buffer));
482 	if (count > sizeof(buffer) - 1)
483 		count = sizeof(buffer) - 1;
484 	if (copy_from_user(buffer, buf, count))
485 		return -EFAULT;
486 	type = simple_strtol(buffer, &end, 0);
487 	if (type < CLEAR_REFS_ALL || type > CLEAR_REFS_MAPPED)
488 		return -EINVAL;
489 	if (*end == '\n')
490 		end++;
491 	task = get_proc_task(file->f_path.dentry->d_inode);
492 	if (!task)
493 		return -ESRCH;
494 	mm = get_task_mm(task);
495 	if (mm) {
496 		struct mm_walk clear_refs_walk = {
497 			.pmd_entry = clear_refs_pte_range,
498 			.mm = mm,
499 		};
500 		down_read(&mm->mmap_sem);
501 		for (vma = mm->mmap; vma; vma = vma->vm_next) {
502 			clear_refs_walk.private = vma;
503 			if (is_vm_hugetlb_page(vma))
504 				continue;
505 			/*
506 			 * Writing 1 to /proc/pid/clear_refs affects all pages.
507 			 *
508 			 * Writing 2 to /proc/pid/clear_refs only affects
509 			 * Anonymous pages.
510 			 *
511 			 * Writing 3 to /proc/pid/clear_refs only affects file
512 			 * mapped pages.
513 			 */
514 			if (type == CLEAR_REFS_ANON && vma->vm_file)
515 				continue;
516 			if (type == CLEAR_REFS_MAPPED && !vma->vm_file)
517 				continue;
518 			walk_page_range(vma->vm_start, vma->vm_end,
519 					&clear_refs_walk);
520 		}
521 		flush_tlb_mm(mm);
522 		up_read(&mm->mmap_sem);
523 		mmput(mm);
524 	}
525 	put_task_struct(task);
526 	if (end - buffer == 0)
527 		return -EIO;
528 	return end - buffer;
529 }
530 
531 const struct file_operations proc_clear_refs_operations = {
532 	.write		= clear_refs_write,
533 };
534 
535 struct pagemapread {
536 	u64 __user *out, *end;
537 };
538 
539 #define PM_ENTRY_BYTES      sizeof(u64)
540 #define PM_STATUS_BITS      3
541 #define PM_STATUS_OFFSET    (64 - PM_STATUS_BITS)
542 #define PM_STATUS_MASK      (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
543 #define PM_STATUS(nr)       (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
544 #define PM_PSHIFT_BITS      6
545 #define PM_PSHIFT_OFFSET    (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
546 #define PM_PSHIFT_MASK      (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
547 #define PM_PSHIFT(x)        (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
548 #define PM_PFRAME_MASK      ((1LL << PM_PSHIFT_OFFSET) - 1)
549 #define PM_PFRAME(x)        ((x) & PM_PFRAME_MASK)
550 
551 #define PM_PRESENT          PM_STATUS(4LL)
552 #define PM_SWAP             PM_STATUS(2LL)
553 #define PM_NOT_PRESENT      PM_PSHIFT(PAGE_SHIFT)
554 #define PM_END_OF_BUFFER    1
555 
556 static int add_to_pagemap(unsigned long addr, u64 pfn,
557 			  struct pagemapread *pm)
558 {
559 	if (put_user(pfn, pm->out))
560 		return -EFAULT;
561 	pm->out++;
562 	if (pm->out >= pm->end)
563 		return PM_END_OF_BUFFER;
564 	return 0;
565 }
566 
567 static int pagemap_pte_hole(unsigned long start, unsigned long end,
568 				struct mm_walk *walk)
569 {
570 	struct pagemapread *pm = walk->private;
571 	unsigned long addr;
572 	int err = 0;
573 	for (addr = start; addr < end; addr += PAGE_SIZE) {
574 		err = add_to_pagemap(addr, PM_NOT_PRESENT, pm);
575 		if (err)
576 			break;
577 	}
578 	return err;
579 }
580 
581 static u64 swap_pte_to_pagemap_entry(pte_t pte)
582 {
583 	swp_entry_t e = pte_to_swp_entry(pte);
584 	return swp_type(e) | (swp_offset(e) << MAX_SWAPFILES_SHIFT);
585 }
586 
587 static u64 pte_to_pagemap_entry(pte_t pte)
588 {
589 	u64 pme = 0;
590 	if (is_swap_pte(pte))
591 		pme = PM_PFRAME(swap_pte_to_pagemap_entry(pte))
592 			| PM_PSHIFT(PAGE_SHIFT) | PM_SWAP;
593 	else if (pte_present(pte))
594 		pme = PM_PFRAME(pte_pfn(pte))
595 			| PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
596 	return pme;
597 }
598 
599 static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
600 			     struct mm_walk *walk)
601 {
602 	struct vm_area_struct *vma;
603 	struct pagemapread *pm = walk->private;
604 	pte_t *pte;
605 	int err = 0;
606 
607 	/* find the first VMA at or above 'addr' */
608 	vma = find_vma(walk->mm, addr);
609 	for (; addr != end; addr += PAGE_SIZE) {
610 		u64 pfn = PM_NOT_PRESENT;
611 
612 		/* check to see if we've left 'vma' behind
613 		 * and need a new, higher one */
614 		if (vma && (addr >= vma->vm_end))
615 			vma = find_vma(walk->mm, addr);
616 
617 		/* check that 'vma' actually covers this address,
618 		 * and that it isn't a huge page vma */
619 		if (vma && (vma->vm_start <= addr) &&
620 		    !is_vm_hugetlb_page(vma)) {
621 			pte = pte_offset_map(pmd, addr);
622 			pfn = pte_to_pagemap_entry(*pte);
623 			/* unmap before userspace copy */
624 			pte_unmap(pte);
625 		}
626 		err = add_to_pagemap(addr, pfn, pm);
627 		if (err)
628 			return err;
629 	}
630 
631 	cond_resched();
632 
633 	return err;
634 }
635 
636 /*
637  * /proc/pid/pagemap - an array mapping virtual pages to pfns
638  *
639  * For each page in the address space, this file contains one 64-bit entry
640  * consisting of the following:
641  *
642  * Bits 0-55  page frame number (PFN) if present
643  * Bits 0-4   swap type if swapped
644  * Bits 5-55  swap offset if swapped
645  * Bits 55-60 page shift (page size = 1<<page shift)
646  * Bit  61    reserved for future use
647  * Bit  62    page swapped
648  * Bit  63    page present
649  *
650  * If the page is not present but in swap, then the PFN contains an
651  * encoding of the swap file number and the page's offset into the
652  * swap. Unmapped pages return a null PFN. This allows determining
653  * precisely which pages are mapped (or in swap) and comparing mapped
654  * pages between processes.
655  *
656  * Efficient users of this interface will use /proc/pid/maps to
657  * determine which areas of memory are actually mapped and llseek to
658  * skip over unmapped regions.
659  */
660 static ssize_t pagemap_read(struct file *file, char __user *buf,
661 			    size_t count, loff_t *ppos)
662 {
663 	struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
664 	struct page **pages, *page;
665 	unsigned long uaddr, uend;
666 	struct mm_struct *mm;
667 	struct pagemapread pm;
668 	int pagecount;
669 	int ret = -ESRCH;
670 	struct mm_walk pagemap_walk = {};
671 	unsigned long src;
672 	unsigned long svpfn;
673 	unsigned long start_vaddr;
674 	unsigned long end_vaddr;
675 
676 	if (!task)
677 		goto out;
678 
679 	ret = -EACCES;
680 	if (!ptrace_may_access(task, PTRACE_MODE_READ))
681 		goto out_task;
682 
683 	ret = -EINVAL;
684 	/* file position must be aligned */
685 	if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
686 		goto out_task;
687 
688 	ret = 0;
689 
690 	if (!count)
691 		goto out_task;
692 
693 	mm = get_task_mm(task);
694 	if (!mm)
695 		goto out_task;
696 
697 
698 	uaddr = (unsigned long)buf & PAGE_MASK;
699 	uend = (unsigned long)(buf + count);
700 	pagecount = (PAGE_ALIGN(uend) - uaddr) / PAGE_SIZE;
701 	ret = 0;
702 	if (pagecount == 0)
703 		goto out_mm;
704 	pages = kcalloc(pagecount, sizeof(struct page *), GFP_KERNEL);
705 	ret = -ENOMEM;
706 	if (!pages)
707 		goto out_mm;
708 
709 	down_read(&current->mm->mmap_sem);
710 	ret = get_user_pages(current, current->mm, uaddr, pagecount,
711 			     1, 0, pages, NULL);
712 	up_read(&current->mm->mmap_sem);
713 
714 	if (ret < 0)
715 		goto out_free;
716 
717 	if (ret != pagecount) {
718 		pagecount = ret;
719 		ret = -EFAULT;
720 		goto out_pages;
721 	}
722 
723 	pm.out = (u64 __user *)buf;
724 	pm.end = (u64 __user *)(buf + count);
725 
726 	pagemap_walk.pmd_entry = pagemap_pte_range;
727 	pagemap_walk.pte_hole = pagemap_pte_hole;
728 	pagemap_walk.mm = mm;
729 	pagemap_walk.private = &pm;
730 
731 	src = *ppos;
732 	svpfn = src / PM_ENTRY_BYTES;
733 	start_vaddr = svpfn << PAGE_SHIFT;
734 	end_vaddr = TASK_SIZE_OF(task);
735 
736 	/* watch out for wraparound */
737 	if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
738 		start_vaddr = end_vaddr;
739 
740 	/*
741 	 * The odds are that this will stop walking way
742 	 * before end_vaddr, because the length of the
743 	 * user buffer is tracked in "pm", and the walk
744 	 * will stop when we hit the end of the buffer.
745 	 */
746 	ret = walk_page_range(start_vaddr, end_vaddr, &pagemap_walk);
747 	if (ret == PM_END_OF_BUFFER)
748 		ret = 0;
749 	/* don't need mmap_sem for these, but this looks cleaner */
750 	*ppos += (char __user *)pm.out - buf;
751 	if (!ret)
752 		ret = (char __user *)pm.out - buf;
753 
754 out_pages:
755 	for (; pagecount; pagecount--) {
756 		page = pages[pagecount-1];
757 		if (!PageReserved(page))
758 			SetPageDirty(page);
759 		page_cache_release(page);
760 	}
761 out_free:
762 	kfree(pages);
763 out_mm:
764 	mmput(mm);
765 out_task:
766 	put_task_struct(task);
767 out:
768 	return ret;
769 }
770 
771 const struct file_operations proc_pagemap_operations = {
772 	.llseek		= mem_lseek, /* borrow this */
773 	.read		= pagemap_read,
774 };
775 #endif /* CONFIG_PROC_PAGE_MONITOR */
776 
777 #ifdef CONFIG_NUMA
778 extern int show_numa_map(struct seq_file *m, void *v);
779 
780 static const struct seq_operations proc_pid_numa_maps_op = {
781         .start  = m_start,
782         .next   = m_next,
783         .stop   = m_stop,
784         .show   = show_numa_map,
785 };
786 
787 static int numa_maps_open(struct inode *inode, struct file *file)
788 {
789 	return do_maps_open(inode, file, &proc_pid_numa_maps_op);
790 }
791 
792 const struct file_operations proc_numa_maps_operations = {
793 	.open		= numa_maps_open,
794 	.read		= seq_read,
795 	.llseek		= seq_lseek,
796 	.release	= seq_release_private,
797 };
798 #endif
799