xref: /openbmc/linux/fs/proc/task_mmu.c (revision 82ced6fd)
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 
123 	tail_vma = get_gate_vma(priv->task);
124 	priv->tail_vma = tail_vma;
125 
126 	/* Start with last addr hint */
127 	vma = find_vma(mm, last_addr);
128 	if (last_addr && vma) {
129 		vma = vma->vm_next;
130 		goto out;
131 	}
132 
133 	/*
134 	 * Check the vma index is within the range and do
135 	 * sequential scan until m_index.
136 	 */
137 	vma = NULL;
138 	if ((unsigned long)l < mm->map_count) {
139 		vma = mm->mmap;
140 		while (l-- && vma)
141 			vma = vma->vm_next;
142 		goto out;
143 	}
144 
145 	if (l != mm->map_count)
146 		tail_vma = NULL; /* After gate vma */
147 
148 out:
149 	if (vma)
150 		return vma;
151 
152 	/* End of vmas has been reached */
153 	m->version = (tail_vma != NULL)? 0: -1UL;
154 	up_read(&mm->mmap_sem);
155 	mmput(mm);
156 	return tail_vma;
157 }
158 
159 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
160 {
161 	struct proc_maps_private *priv = m->private;
162 	struct vm_area_struct *vma = v;
163 	struct vm_area_struct *tail_vma = priv->tail_vma;
164 
165 	(*pos)++;
166 	if (vma && (vma != tail_vma) && vma->vm_next)
167 		return vma->vm_next;
168 	vma_stop(priv, vma);
169 	return (vma != tail_vma)? tail_vma: NULL;
170 }
171 
172 static void m_stop(struct seq_file *m, void *v)
173 {
174 	struct proc_maps_private *priv = m->private;
175 	struct vm_area_struct *vma = v;
176 
177 	vma_stop(priv, vma);
178 	if (priv->task)
179 		put_task_struct(priv->task);
180 }
181 
182 static int do_maps_open(struct inode *inode, struct file *file,
183 			const struct seq_operations *ops)
184 {
185 	struct proc_maps_private *priv;
186 	int ret = -ENOMEM;
187 	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
188 	if (priv) {
189 		priv->pid = proc_pid(inode);
190 		ret = seq_open(file, ops);
191 		if (!ret) {
192 			struct seq_file *m = file->private_data;
193 			m->private = priv;
194 		} else {
195 			kfree(priv);
196 		}
197 	}
198 	return ret;
199 }
200 
201 static void show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
202 {
203 	struct mm_struct *mm = vma->vm_mm;
204 	struct file *file = vma->vm_file;
205 	int flags = vma->vm_flags;
206 	unsigned long ino = 0;
207 	unsigned long long pgoff = 0;
208 	dev_t dev = 0;
209 	int len;
210 
211 	if (file) {
212 		struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
213 		dev = inode->i_sb->s_dev;
214 		ino = inode->i_ino;
215 		pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
216 	}
217 
218 	seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu %n",
219 			vma->vm_start,
220 			vma->vm_end,
221 			flags & VM_READ ? 'r' : '-',
222 			flags & VM_WRITE ? 'w' : '-',
223 			flags & VM_EXEC ? 'x' : '-',
224 			flags & VM_MAYSHARE ? 's' : 'p',
225 			pgoff,
226 			MAJOR(dev), MINOR(dev), ino, &len);
227 
228 	/*
229 	 * Print the dentry name for named mappings, and a
230 	 * special [heap] marker for the heap:
231 	 */
232 	if (file) {
233 		pad_len_spaces(m, len);
234 		seq_path(m, &file->f_path, "\n");
235 	} else {
236 		const char *name = arch_vma_name(vma);
237 		if (!name) {
238 			if (mm) {
239 				if (vma->vm_start <= mm->start_brk &&
240 						vma->vm_end >= mm->brk) {
241 					name = "[heap]";
242 				} else if (vma->vm_start <= mm->start_stack &&
243 					   vma->vm_end >= mm->start_stack) {
244 					name = "[stack]";
245 				}
246 			} else {
247 				name = "[vdso]";
248 			}
249 		}
250 		if (name) {
251 			pad_len_spaces(m, len);
252 			seq_puts(m, name);
253 		}
254 	}
255 	seq_putc(m, '\n');
256 }
257 
258 static int show_map(struct seq_file *m, void *v)
259 {
260 	struct vm_area_struct *vma = v;
261 	struct proc_maps_private *priv = m->private;
262 	struct task_struct *task = priv->task;
263 
264 	show_map_vma(m, vma);
265 
266 	if (m->count < m->size)  /* vma is copied successfully */
267 		m->version = (vma != get_gate_vma(task))? vma->vm_start: 0;
268 	return 0;
269 }
270 
271 static const struct seq_operations proc_pid_maps_op = {
272 	.start	= m_start,
273 	.next	= m_next,
274 	.stop	= m_stop,
275 	.show	= show_map
276 };
277 
278 static int maps_open(struct inode *inode, struct file *file)
279 {
280 	return do_maps_open(inode, file, &proc_pid_maps_op);
281 }
282 
283 const struct file_operations proc_maps_operations = {
284 	.open		= maps_open,
285 	.read		= seq_read,
286 	.llseek		= seq_lseek,
287 	.release	= seq_release_private,
288 };
289 
290 /*
291  * Proportional Set Size(PSS): my share of RSS.
292  *
293  * PSS of a process is the count of pages it has in memory, where each
294  * page is divided by the number of processes sharing it.  So if a
295  * process has 1000 pages all to itself, and 1000 shared with one other
296  * process, its PSS will be 1500.
297  *
298  * To keep (accumulated) division errors low, we adopt a 64bit
299  * fixed-point pss counter to minimize division errors. So (pss >>
300  * PSS_SHIFT) would be the real byte count.
301  *
302  * A shift of 12 before division means (assuming 4K page size):
303  * 	- 1M 3-user-pages add up to 8KB errors;
304  * 	- supports mapcount up to 2^24, or 16M;
305  * 	- supports PSS up to 2^52 bytes, or 4PB.
306  */
307 #define PSS_SHIFT 12
308 
309 #ifdef CONFIG_PROC_PAGE_MONITOR
310 struct mem_size_stats {
311 	struct vm_area_struct *vma;
312 	unsigned long resident;
313 	unsigned long shared_clean;
314 	unsigned long shared_dirty;
315 	unsigned long private_clean;
316 	unsigned long private_dirty;
317 	unsigned long referenced;
318 	unsigned long swap;
319 	u64 pss;
320 };
321 
322 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
323 			   struct mm_walk *walk)
324 {
325 	struct mem_size_stats *mss = walk->private;
326 	struct vm_area_struct *vma = mss->vma;
327 	pte_t *pte, ptent;
328 	spinlock_t *ptl;
329 	struct page *page;
330 	int mapcount;
331 
332 	pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
333 	for (; addr != end; pte++, addr += PAGE_SIZE) {
334 		ptent = *pte;
335 
336 		if (is_swap_pte(ptent)) {
337 			mss->swap += PAGE_SIZE;
338 			continue;
339 		}
340 
341 		if (!pte_present(ptent))
342 			continue;
343 
344 		mss->resident += PAGE_SIZE;
345 
346 		page = vm_normal_page(vma, addr, ptent);
347 		if (!page)
348 			continue;
349 
350 		/* Accumulate the size in pages that have been accessed. */
351 		if (pte_young(ptent) || PageReferenced(page))
352 			mss->referenced += PAGE_SIZE;
353 		mapcount = page_mapcount(page);
354 		if (mapcount >= 2) {
355 			if (pte_dirty(ptent))
356 				mss->shared_dirty += PAGE_SIZE;
357 			else
358 				mss->shared_clean += PAGE_SIZE;
359 			mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount;
360 		} else {
361 			if (pte_dirty(ptent))
362 				mss->private_dirty += PAGE_SIZE;
363 			else
364 				mss->private_clean += PAGE_SIZE;
365 			mss->pss += (PAGE_SIZE << PSS_SHIFT);
366 		}
367 	}
368 	pte_unmap_unlock(pte - 1, ptl);
369 	cond_resched();
370 	return 0;
371 }
372 
373 static int show_smap(struct seq_file *m, void *v)
374 {
375 	struct proc_maps_private *priv = m->private;
376 	struct task_struct *task = priv->task;
377 	struct vm_area_struct *vma = v;
378 	struct mem_size_stats mss;
379 	struct mm_walk smaps_walk = {
380 		.pmd_entry = smaps_pte_range,
381 		.mm = vma->vm_mm,
382 		.private = &mss,
383 	};
384 
385 	memset(&mss, 0, sizeof mss);
386 	mss.vma = vma;
387 	if (vma->vm_mm && !is_vm_hugetlb_page(vma))
388 		walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk);
389 
390 	show_map_vma(m, vma);
391 
392 	seq_printf(m,
393 		   "Size:           %8lu kB\n"
394 		   "Rss:            %8lu kB\n"
395 		   "Pss:            %8lu kB\n"
396 		   "Shared_Clean:   %8lu kB\n"
397 		   "Shared_Dirty:   %8lu kB\n"
398 		   "Private_Clean:  %8lu kB\n"
399 		   "Private_Dirty:  %8lu kB\n"
400 		   "Referenced:     %8lu kB\n"
401 		   "Swap:           %8lu kB\n"
402 		   "KernelPageSize: %8lu kB\n"
403 		   "MMUPageSize:    %8lu kB\n",
404 		   (vma->vm_end - vma->vm_start) >> 10,
405 		   mss.resident >> 10,
406 		   (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
407 		   mss.shared_clean  >> 10,
408 		   mss.shared_dirty  >> 10,
409 		   mss.private_clean >> 10,
410 		   mss.private_dirty >> 10,
411 		   mss.referenced >> 10,
412 		   mss.swap >> 10,
413 		   vma_kernel_pagesize(vma) >> 10,
414 		   vma_mmu_pagesize(vma) >> 10);
415 
416 	if (m->count < m->size)  /* vma is copied successfully */
417 		m->version = (vma != get_gate_vma(task)) ? vma->vm_start : 0;
418 	return 0;
419 }
420 
421 static const struct seq_operations proc_pid_smaps_op = {
422 	.start	= m_start,
423 	.next	= m_next,
424 	.stop	= m_stop,
425 	.show	= show_smap
426 };
427 
428 static int smaps_open(struct inode *inode, struct file *file)
429 {
430 	return do_maps_open(inode, file, &proc_pid_smaps_op);
431 }
432 
433 const struct file_operations proc_smaps_operations = {
434 	.open		= smaps_open,
435 	.read		= seq_read,
436 	.llseek		= seq_lseek,
437 	.release	= seq_release_private,
438 };
439 
440 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
441 				unsigned long end, struct mm_walk *walk)
442 {
443 	struct vm_area_struct *vma = walk->private;
444 	pte_t *pte, ptent;
445 	spinlock_t *ptl;
446 	struct page *page;
447 
448 	pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
449 	for (; addr != end; pte++, addr += PAGE_SIZE) {
450 		ptent = *pte;
451 		if (!pte_present(ptent))
452 			continue;
453 
454 		page = vm_normal_page(vma, addr, ptent);
455 		if (!page)
456 			continue;
457 
458 		/* Clear accessed and referenced bits. */
459 		ptep_test_and_clear_young(vma, addr, pte);
460 		ClearPageReferenced(page);
461 	}
462 	pte_unmap_unlock(pte - 1, ptl);
463 	cond_resched();
464 	return 0;
465 }
466 
467 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
468 				size_t count, loff_t *ppos)
469 {
470 	struct task_struct *task;
471 	char buffer[PROC_NUMBUF], *end;
472 	struct mm_struct *mm;
473 	struct vm_area_struct *vma;
474 
475 	memset(buffer, 0, sizeof(buffer));
476 	if (count > sizeof(buffer) - 1)
477 		count = sizeof(buffer) - 1;
478 	if (copy_from_user(buffer, buf, count))
479 		return -EFAULT;
480 	if (!simple_strtol(buffer, &end, 0))
481 		return -EINVAL;
482 	if (*end == '\n')
483 		end++;
484 	task = get_proc_task(file->f_path.dentry->d_inode);
485 	if (!task)
486 		return -ESRCH;
487 	mm = get_task_mm(task);
488 	if (mm) {
489 		struct mm_walk clear_refs_walk = {
490 			.pmd_entry = clear_refs_pte_range,
491 			.mm = mm,
492 		};
493 		down_read(&mm->mmap_sem);
494 		for (vma = mm->mmap; vma; vma = vma->vm_next) {
495 			clear_refs_walk.private = vma;
496 			if (!is_vm_hugetlb_page(vma))
497 				walk_page_range(vma->vm_start, vma->vm_end,
498 						&clear_refs_walk);
499 		}
500 		flush_tlb_mm(mm);
501 		up_read(&mm->mmap_sem);
502 		mmput(mm);
503 	}
504 	put_task_struct(task);
505 	if (end - buffer == 0)
506 		return -EIO;
507 	return end - buffer;
508 }
509 
510 const struct file_operations proc_clear_refs_operations = {
511 	.write		= clear_refs_write,
512 };
513 
514 struct pagemapread {
515 	u64 __user *out, *end;
516 };
517 
518 #define PM_ENTRY_BYTES      sizeof(u64)
519 #define PM_STATUS_BITS      3
520 #define PM_STATUS_OFFSET    (64 - PM_STATUS_BITS)
521 #define PM_STATUS_MASK      (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
522 #define PM_STATUS(nr)       (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
523 #define PM_PSHIFT_BITS      6
524 #define PM_PSHIFT_OFFSET    (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
525 #define PM_PSHIFT_MASK      (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
526 #define PM_PSHIFT(x)        (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
527 #define PM_PFRAME_MASK      ((1LL << PM_PSHIFT_OFFSET) - 1)
528 #define PM_PFRAME(x)        ((x) & PM_PFRAME_MASK)
529 
530 #define PM_PRESENT          PM_STATUS(4LL)
531 #define PM_SWAP             PM_STATUS(2LL)
532 #define PM_NOT_PRESENT      PM_PSHIFT(PAGE_SHIFT)
533 #define PM_END_OF_BUFFER    1
534 
535 static int add_to_pagemap(unsigned long addr, u64 pfn,
536 			  struct pagemapread *pm)
537 {
538 	if (put_user(pfn, pm->out))
539 		return -EFAULT;
540 	pm->out++;
541 	if (pm->out >= pm->end)
542 		return PM_END_OF_BUFFER;
543 	return 0;
544 }
545 
546 static int pagemap_pte_hole(unsigned long start, unsigned long end,
547 				struct mm_walk *walk)
548 {
549 	struct pagemapread *pm = walk->private;
550 	unsigned long addr;
551 	int err = 0;
552 	for (addr = start; addr < end; addr += PAGE_SIZE) {
553 		err = add_to_pagemap(addr, PM_NOT_PRESENT, pm);
554 		if (err)
555 			break;
556 	}
557 	return err;
558 }
559 
560 static u64 swap_pte_to_pagemap_entry(pte_t pte)
561 {
562 	swp_entry_t e = pte_to_swp_entry(pte);
563 	return swp_type(e) | (swp_offset(e) << MAX_SWAPFILES_SHIFT);
564 }
565 
566 static u64 pte_to_pagemap_entry(pte_t pte)
567 {
568 	u64 pme = 0;
569 	if (is_swap_pte(pte))
570 		pme = PM_PFRAME(swap_pte_to_pagemap_entry(pte))
571 			| PM_PSHIFT(PAGE_SHIFT) | PM_SWAP;
572 	else if (pte_present(pte))
573 		pme = PM_PFRAME(pte_pfn(pte))
574 			| PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
575 	return pme;
576 }
577 
578 static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
579 			     struct mm_walk *walk)
580 {
581 	struct vm_area_struct *vma;
582 	struct pagemapread *pm = walk->private;
583 	pte_t *pte;
584 	int err = 0;
585 
586 	/* find the first VMA at or above 'addr' */
587 	vma = find_vma(walk->mm, addr);
588 	for (; addr != end; addr += PAGE_SIZE) {
589 		u64 pfn = PM_NOT_PRESENT;
590 
591 		/* check to see if we've left 'vma' behind
592 		 * and need a new, higher one */
593 		if (vma && (addr >= vma->vm_end))
594 			vma = find_vma(walk->mm, addr);
595 
596 		/* check that 'vma' actually covers this address,
597 		 * and that it isn't a huge page vma */
598 		if (vma && (vma->vm_start <= addr) &&
599 		    !is_vm_hugetlb_page(vma)) {
600 			pte = pte_offset_map(pmd, addr);
601 			pfn = pte_to_pagemap_entry(*pte);
602 			/* unmap before userspace copy */
603 			pte_unmap(pte);
604 		}
605 		err = add_to_pagemap(addr, pfn, pm);
606 		if (err)
607 			return err;
608 	}
609 
610 	cond_resched();
611 
612 	return err;
613 }
614 
615 /*
616  * /proc/pid/pagemap - an array mapping virtual pages to pfns
617  *
618  * For each page in the address space, this file contains one 64-bit entry
619  * consisting of the following:
620  *
621  * Bits 0-55  page frame number (PFN) if present
622  * Bits 0-4   swap type if swapped
623  * Bits 5-55  swap offset if swapped
624  * Bits 55-60 page shift (page size = 1<<page shift)
625  * Bit  61    reserved for future use
626  * Bit  62    page swapped
627  * Bit  63    page present
628  *
629  * If the page is not present but in swap, then the PFN contains an
630  * encoding of the swap file number and the page's offset into the
631  * swap. Unmapped pages return a null PFN. This allows determining
632  * precisely which pages are mapped (or in swap) and comparing mapped
633  * pages between processes.
634  *
635  * Efficient users of this interface will use /proc/pid/maps to
636  * determine which areas of memory are actually mapped and llseek to
637  * skip over unmapped regions.
638  */
639 static ssize_t pagemap_read(struct file *file, char __user *buf,
640 			    size_t count, loff_t *ppos)
641 {
642 	struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
643 	struct page **pages, *page;
644 	unsigned long uaddr, uend;
645 	struct mm_struct *mm;
646 	struct pagemapread pm;
647 	int pagecount;
648 	int ret = -ESRCH;
649 	struct mm_walk pagemap_walk = {};
650 	unsigned long src;
651 	unsigned long svpfn;
652 	unsigned long start_vaddr;
653 	unsigned long end_vaddr;
654 
655 	if (!task)
656 		goto out;
657 
658 	ret = -EACCES;
659 	if (!ptrace_may_access(task, PTRACE_MODE_READ))
660 		goto out_task;
661 
662 	ret = -EINVAL;
663 	/* file position must be aligned */
664 	if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
665 		goto out_task;
666 
667 	ret = 0;
668 
669 	if (!count)
670 		goto out_task;
671 
672 	mm = get_task_mm(task);
673 	if (!mm)
674 		goto out_task;
675 
676 
677 	uaddr = (unsigned long)buf & PAGE_MASK;
678 	uend = (unsigned long)(buf + count);
679 	pagecount = (PAGE_ALIGN(uend) - uaddr) / PAGE_SIZE;
680 	ret = 0;
681 	if (pagecount == 0)
682 		goto out_mm;
683 	pages = kcalloc(pagecount, sizeof(struct page *), GFP_KERNEL);
684 	ret = -ENOMEM;
685 	if (!pages)
686 		goto out_mm;
687 
688 	down_read(&current->mm->mmap_sem);
689 	ret = get_user_pages(current, current->mm, uaddr, pagecount,
690 			     1, 0, pages, NULL);
691 	up_read(&current->mm->mmap_sem);
692 
693 	if (ret < 0)
694 		goto out_free;
695 
696 	if (ret != pagecount) {
697 		pagecount = ret;
698 		ret = -EFAULT;
699 		goto out_pages;
700 	}
701 
702 	pm.out = (u64 __user *)buf;
703 	pm.end = (u64 __user *)(buf + count);
704 
705 	pagemap_walk.pmd_entry = pagemap_pte_range;
706 	pagemap_walk.pte_hole = pagemap_pte_hole;
707 	pagemap_walk.mm = mm;
708 	pagemap_walk.private = &pm;
709 
710 	src = *ppos;
711 	svpfn = src / PM_ENTRY_BYTES;
712 	start_vaddr = svpfn << PAGE_SHIFT;
713 	end_vaddr = TASK_SIZE_OF(task);
714 
715 	/* watch out for wraparound */
716 	if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
717 		start_vaddr = end_vaddr;
718 
719 	/*
720 	 * The odds are that this will stop walking way
721 	 * before end_vaddr, because the length of the
722 	 * user buffer is tracked in "pm", and the walk
723 	 * will stop when we hit the end of the buffer.
724 	 */
725 	ret = walk_page_range(start_vaddr, end_vaddr, &pagemap_walk);
726 	if (ret == PM_END_OF_BUFFER)
727 		ret = 0;
728 	/* don't need mmap_sem for these, but this looks cleaner */
729 	*ppos += (char __user *)pm.out - buf;
730 	if (!ret)
731 		ret = (char __user *)pm.out - buf;
732 
733 out_pages:
734 	for (; pagecount; pagecount--) {
735 		page = pages[pagecount-1];
736 		if (!PageReserved(page))
737 			SetPageDirty(page);
738 		page_cache_release(page);
739 	}
740 out_free:
741 	kfree(pages);
742 out_mm:
743 	mmput(mm);
744 out_task:
745 	put_task_struct(task);
746 out:
747 	return ret;
748 }
749 
750 const struct file_operations proc_pagemap_operations = {
751 	.llseek		= mem_lseek, /* borrow this */
752 	.read		= pagemap_read,
753 };
754 #endif /* CONFIG_PROC_PAGE_MONITOR */
755 
756 #ifdef CONFIG_NUMA
757 extern int show_numa_map(struct seq_file *m, void *v);
758 
759 static const struct seq_operations proc_pid_numa_maps_op = {
760         .start  = m_start,
761         .next   = m_next,
762         .stop   = m_stop,
763         .show   = show_numa_map,
764 };
765 
766 static int numa_maps_open(struct inode *inode, struct file *file)
767 {
768 	return do_maps_open(inode, file, &proc_pid_numa_maps_op);
769 }
770 
771 const struct file_operations proc_numa_maps_operations = {
772 	.open		= numa_maps_open,
773 	.read		= seq_read,
774 	.llseek		= seq_lseek,
775 	.release	= seq_release_private,
776 };
777 #endif
778