xref: /linux/fs/proc/task_mmu.c (revision f7511d5f66f01fc451747b24e79f3ada7a3af9af)
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/ptrace.h>
9 #include <linux/mempolicy.h>
10 #include <linux/swap.h>
11 #include <linux/swapops.h>
12 #include <linux/seq_file.h>
13 
14 #include <asm/elf.h>
15 #include <asm/uaccess.h>
16 #include <asm/tlbflush.h>
17 #include "internal.h"
18 
19 void task_mem(struct seq_file *m, struct mm_struct *mm)
20 {
21 	unsigned long data, text, lib;
22 	unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
23 
24 	/*
25 	 * Note: to minimize their overhead, mm maintains hiwater_vm and
26 	 * hiwater_rss only when about to *lower* total_vm or rss.  Any
27 	 * collector of these hiwater stats must therefore get total_vm
28 	 * and rss too, which will usually be the higher.  Barriers? not
29 	 * worth the effort, such snapshots can always be inconsistent.
30 	 */
31 	hiwater_vm = total_vm = mm->total_vm;
32 	if (hiwater_vm < mm->hiwater_vm)
33 		hiwater_vm = mm->hiwater_vm;
34 	hiwater_rss = total_rss = get_mm_rss(mm);
35 	if (hiwater_rss < mm->hiwater_rss)
36 		hiwater_rss = mm->hiwater_rss;
37 
38 	data = mm->total_vm - mm->shared_vm - mm->stack_vm;
39 	text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
40 	lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
41 	seq_printf(m,
42 		"VmPeak:\t%8lu kB\n"
43 		"VmSize:\t%8lu kB\n"
44 		"VmLck:\t%8lu kB\n"
45 		"VmHWM:\t%8lu kB\n"
46 		"VmRSS:\t%8lu kB\n"
47 		"VmData:\t%8lu kB\n"
48 		"VmStk:\t%8lu kB\n"
49 		"VmExe:\t%8lu kB\n"
50 		"VmLib:\t%8lu kB\n"
51 		"VmPTE:\t%8lu kB\n",
52 		hiwater_vm << (PAGE_SHIFT-10),
53 		(total_vm - mm->reserved_vm) << (PAGE_SHIFT-10),
54 		mm->locked_vm << (PAGE_SHIFT-10),
55 		hiwater_rss << (PAGE_SHIFT-10),
56 		total_rss << (PAGE_SHIFT-10),
57 		data << (PAGE_SHIFT-10),
58 		mm->stack_vm << (PAGE_SHIFT-10), text, lib,
59 		(PTRS_PER_PTE*sizeof(pte_t)*mm->nr_ptes) >> 10);
60 }
61 
62 unsigned long task_vsize(struct mm_struct *mm)
63 {
64 	return PAGE_SIZE * mm->total_vm;
65 }
66 
67 int task_statm(struct mm_struct *mm, int *shared, int *text,
68 	       int *data, int *resident)
69 {
70 	*shared = get_mm_counter(mm, file_rss);
71 	*text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
72 								>> PAGE_SHIFT;
73 	*data = mm->total_vm - mm->shared_vm;
74 	*resident = *shared + get_mm_counter(mm, anon_rss);
75 	return mm->total_vm;
76 }
77 
78 static void pad_len_spaces(struct seq_file *m, int len)
79 {
80 	len = 25 + sizeof(void*) * 6 - len;
81 	if (len < 1)
82 		len = 1;
83 	seq_printf(m, "%*c", len, ' ');
84 }
85 
86 static void vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma)
87 {
88 	if (vma && vma != priv->tail_vma) {
89 		struct mm_struct *mm = vma->vm_mm;
90 		up_read(&mm->mmap_sem);
91 		mmput(mm);
92 	}
93 }
94 
95 static void *m_start(struct seq_file *m, loff_t *pos)
96 {
97 	struct proc_maps_private *priv = m->private;
98 	unsigned long last_addr = m->version;
99 	struct mm_struct *mm;
100 	struct vm_area_struct *vma, *tail_vma = NULL;
101 	loff_t l = *pos;
102 
103 	/* Clear the per syscall fields in priv */
104 	priv->task = NULL;
105 	priv->tail_vma = NULL;
106 
107 	/*
108 	 * We remember last_addr rather than next_addr to hit with
109 	 * mmap_cache most of the time. We have zero last_addr at
110 	 * the beginning and also after lseek. We will have -1 last_addr
111 	 * after the end of the vmas.
112 	 */
113 
114 	if (last_addr == -1UL)
115 		return NULL;
116 
117 	priv->task = get_pid_task(priv->pid, PIDTYPE_PID);
118 	if (!priv->task)
119 		return NULL;
120 
121 	mm = mm_for_maps(priv->task);
122 	if (!mm)
123 		return NULL;
124 
125 	tail_vma = get_gate_vma(priv->task);
126 	priv->tail_vma = tail_vma;
127 
128 	/* Start with last addr hint */
129 	vma = find_vma(mm, last_addr);
130 	if (last_addr && vma) {
131 		vma = vma->vm_next;
132 		goto out;
133 	}
134 
135 	/*
136 	 * Check the vma index is within the range and do
137 	 * sequential scan until m_index.
138 	 */
139 	vma = NULL;
140 	if ((unsigned long)l < mm->map_count) {
141 		vma = mm->mmap;
142 		while (l-- && vma)
143 			vma = vma->vm_next;
144 		goto out;
145 	}
146 
147 	if (l != mm->map_count)
148 		tail_vma = NULL; /* After gate vma */
149 
150 out:
151 	if (vma)
152 		return vma;
153 
154 	/* End of vmas has been reached */
155 	m->version = (tail_vma != NULL)? 0: -1UL;
156 	up_read(&mm->mmap_sem);
157 	mmput(mm);
158 	return tail_vma;
159 }
160 
161 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
162 {
163 	struct proc_maps_private *priv = m->private;
164 	struct vm_area_struct *vma = v;
165 	struct vm_area_struct *tail_vma = priv->tail_vma;
166 
167 	(*pos)++;
168 	if (vma && (vma != tail_vma) && vma->vm_next)
169 		return vma->vm_next;
170 	vma_stop(priv, vma);
171 	return (vma != tail_vma)? tail_vma: NULL;
172 }
173 
174 static void m_stop(struct seq_file *m, void *v)
175 {
176 	struct proc_maps_private *priv = m->private;
177 	struct vm_area_struct *vma = v;
178 
179 	vma_stop(priv, vma);
180 	if (priv->task)
181 		put_task_struct(priv->task);
182 }
183 
184 static int do_maps_open(struct inode *inode, struct file *file,
185 			const struct seq_operations *ops)
186 {
187 	struct proc_maps_private *priv;
188 	int ret = -ENOMEM;
189 	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
190 	if (priv) {
191 		priv->pid = proc_pid(inode);
192 		ret = seq_open(file, ops);
193 		if (!ret) {
194 			struct seq_file *m = file->private_data;
195 			m->private = priv;
196 		} else {
197 			kfree(priv);
198 		}
199 	}
200 	return ret;
201 }
202 
203 static int show_map(struct seq_file *m, void *v)
204 {
205 	struct proc_maps_private *priv = m->private;
206 	struct task_struct *task = priv->task;
207 	struct vm_area_struct *vma = v;
208 	struct mm_struct *mm = vma->vm_mm;
209 	struct file *file = vma->vm_file;
210 	int flags = vma->vm_flags;
211 	unsigned long ino = 0;
212 	dev_t dev = 0;
213 	int len;
214 
215 	if (maps_protect && !ptrace_may_attach(task))
216 		return -EACCES;
217 
218 	if (file) {
219 		struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
220 		dev = inode->i_sb->s_dev;
221 		ino = inode->i_ino;
222 	}
223 
224 	seq_printf(m, "%08lx-%08lx %c%c%c%c %08lx %02x:%02x %lu %n",
225 			vma->vm_start,
226 			vma->vm_end,
227 			flags & VM_READ ? 'r' : '-',
228 			flags & VM_WRITE ? 'w' : '-',
229 			flags & VM_EXEC ? 'x' : '-',
230 			flags & VM_MAYSHARE ? 's' : 'p',
231 			vma->vm_pgoff << PAGE_SHIFT,
232 			MAJOR(dev), MINOR(dev), ino, &len);
233 
234 	/*
235 	 * Print the dentry name for named mappings, and a
236 	 * special [heap] marker for the heap:
237 	 */
238 	if (file) {
239 		pad_len_spaces(m, len);
240 		seq_path(m, &file->f_path, "\n");
241 	} else {
242 		const char *name = arch_vma_name(vma);
243 		if (!name) {
244 			if (mm) {
245 				if (vma->vm_start <= mm->start_brk &&
246 						vma->vm_end >= mm->brk) {
247 					name = "[heap]";
248 				} else if (vma->vm_start <= mm->start_stack &&
249 					   vma->vm_end >= mm->start_stack) {
250 					name = "[stack]";
251 				}
252 			} else {
253 				name = "[vdso]";
254 			}
255 		}
256 		if (name) {
257 			pad_len_spaces(m, len);
258 			seq_puts(m, name);
259 		}
260 	}
261 	seq_putc(m, '\n');
262 
263 	if (m->count < m->size)  /* vma is copied successfully */
264 		m->version = (vma != get_gate_vma(task))? vma->vm_start: 0;
265 	return 0;
266 }
267 
268 static const struct seq_operations proc_pid_maps_op = {
269 	.start	= m_start,
270 	.next	= m_next,
271 	.stop	= m_stop,
272 	.show	= show_map
273 };
274 
275 static int maps_open(struct inode *inode, struct file *file)
276 {
277 	return do_maps_open(inode, file, &proc_pid_maps_op);
278 }
279 
280 const struct file_operations proc_maps_operations = {
281 	.open		= maps_open,
282 	.read		= seq_read,
283 	.llseek		= seq_lseek,
284 	.release	= seq_release_private,
285 };
286 
287 /*
288  * Proportional Set Size(PSS): my share of RSS.
289  *
290  * PSS of a process is the count of pages it has in memory, where each
291  * page is divided by the number of processes sharing it.  So if a
292  * process has 1000 pages all to itself, and 1000 shared with one other
293  * process, its PSS will be 1500.
294  *
295  * To keep (accumulated) division errors low, we adopt a 64bit
296  * fixed-point pss counter to minimize division errors. So (pss >>
297  * PSS_SHIFT) would be the real byte count.
298  *
299  * A shift of 12 before division means (assuming 4K page size):
300  * 	- 1M 3-user-pages add up to 8KB errors;
301  * 	- supports mapcount up to 2^24, or 16M;
302  * 	- supports PSS up to 2^52 bytes, or 4PB.
303  */
304 #define PSS_SHIFT 12
305 
306 #ifdef CONFIG_PROC_PAGE_MONITOR
307 struct mem_size_stats {
308 	struct vm_area_struct *vma;
309 	unsigned long resident;
310 	unsigned long shared_clean;
311 	unsigned long shared_dirty;
312 	unsigned long private_clean;
313 	unsigned long private_dirty;
314 	unsigned long referenced;
315 	unsigned long swap;
316 	u64 pss;
317 };
318 
319 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
320 			   void *private)
321 {
322 	struct mem_size_stats *mss = private;
323 	struct vm_area_struct *vma = mss->vma;
324 	pte_t *pte, ptent;
325 	spinlock_t *ptl;
326 	struct page *page;
327 	int mapcount;
328 
329 	pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
330 	for (; addr != end; pte++, addr += PAGE_SIZE) {
331 		ptent = *pte;
332 
333 		if (is_swap_pte(ptent)) {
334 			mss->swap += PAGE_SIZE;
335 			continue;
336 		}
337 
338 		if (!pte_present(ptent))
339 			continue;
340 
341 		mss->resident += PAGE_SIZE;
342 
343 		page = vm_normal_page(vma, addr, ptent);
344 		if (!page)
345 			continue;
346 
347 		/* Accumulate the size in pages that have been accessed. */
348 		if (pte_young(ptent) || PageReferenced(page))
349 			mss->referenced += PAGE_SIZE;
350 		mapcount = page_mapcount(page);
351 		if (mapcount >= 2) {
352 			if (pte_dirty(ptent))
353 				mss->shared_dirty += PAGE_SIZE;
354 			else
355 				mss->shared_clean += PAGE_SIZE;
356 			mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount;
357 		} else {
358 			if (pte_dirty(ptent))
359 				mss->private_dirty += PAGE_SIZE;
360 			else
361 				mss->private_clean += PAGE_SIZE;
362 			mss->pss += (PAGE_SIZE << PSS_SHIFT);
363 		}
364 	}
365 	pte_unmap_unlock(pte - 1, ptl);
366 	cond_resched();
367 	return 0;
368 }
369 
370 static struct mm_walk smaps_walk = { .pmd_entry = smaps_pte_range };
371 
372 static int show_smap(struct seq_file *m, void *v)
373 {
374 	struct vm_area_struct *vma = v;
375 	struct mem_size_stats mss;
376 	int ret;
377 
378 	memset(&mss, 0, sizeof mss);
379 	mss.vma = vma;
380 	if (vma->vm_mm && !is_vm_hugetlb_page(vma))
381 		walk_page_range(vma->vm_mm, vma->vm_start, vma->vm_end,
382 				&smaps_walk, &mss);
383 
384 	ret = show_map(m, v);
385 	if (ret)
386 		return ret;
387 
388 	seq_printf(m,
389 		   "Size:           %8lu kB\n"
390 		   "Rss:            %8lu kB\n"
391 		   "Pss:            %8lu kB\n"
392 		   "Shared_Clean:   %8lu kB\n"
393 		   "Shared_Dirty:   %8lu kB\n"
394 		   "Private_Clean:  %8lu kB\n"
395 		   "Private_Dirty:  %8lu kB\n"
396 		   "Referenced:     %8lu kB\n"
397 		   "Swap:           %8lu kB\n",
398 		   (vma->vm_end - vma->vm_start) >> 10,
399 		   mss.resident >> 10,
400 		   (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
401 		   mss.shared_clean  >> 10,
402 		   mss.shared_dirty  >> 10,
403 		   mss.private_clean >> 10,
404 		   mss.private_dirty >> 10,
405 		   mss.referenced >> 10,
406 		   mss.swap >> 10);
407 
408 	return ret;
409 }
410 
411 static const struct seq_operations proc_pid_smaps_op = {
412 	.start	= m_start,
413 	.next	= m_next,
414 	.stop	= m_stop,
415 	.show	= show_smap
416 };
417 
418 static int smaps_open(struct inode *inode, struct file *file)
419 {
420 	return do_maps_open(inode, file, &proc_pid_smaps_op);
421 }
422 
423 const struct file_operations proc_smaps_operations = {
424 	.open		= smaps_open,
425 	.read		= seq_read,
426 	.llseek		= seq_lseek,
427 	.release	= seq_release_private,
428 };
429 
430 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
431 				unsigned long end, void *private)
432 {
433 	struct vm_area_struct *vma = private;
434 	pte_t *pte, ptent;
435 	spinlock_t *ptl;
436 	struct page *page;
437 
438 	pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
439 	for (; addr != end; pte++, addr += PAGE_SIZE) {
440 		ptent = *pte;
441 		if (!pte_present(ptent))
442 			continue;
443 
444 		page = vm_normal_page(vma, addr, ptent);
445 		if (!page)
446 			continue;
447 
448 		/* Clear accessed and referenced bits. */
449 		ptep_test_and_clear_young(vma, addr, pte);
450 		ClearPageReferenced(page);
451 	}
452 	pte_unmap_unlock(pte - 1, ptl);
453 	cond_resched();
454 	return 0;
455 }
456 
457 static struct mm_walk clear_refs_walk = { .pmd_entry = clear_refs_pte_range };
458 
459 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
460 				size_t count, loff_t *ppos)
461 {
462 	struct task_struct *task;
463 	char buffer[PROC_NUMBUF], *end;
464 	struct mm_struct *mm;
465 	struct vm_area_struct *vma;
466 
467 	memset(buffer, 0, sizeof(buffer));
468 	if (count > sizeof(buffer) - 1)
469 		count = sizeof(buffer) - 1;
470 	if (copy_from_user(buffer, buf, count))
471 		return -EFAULT;
472 	if (!simple_strtol(buffer, &end, 0))
473 		return -EINVAL;
474 	if (*end == '\n')
475 		end++;
476 	task = get_proc_task(file->f_path.dentry->d_inode);
477 	if (!task)
478 		return -ESRCH;
479 	mm = get_task_mm(task);
480 	if (mm) {
481 		down_read(&mm->mmap_sem);
482 		for (vma = mm->mmap; vma; vma = vma->vm_next)
483 			if (!is_vm_hugetlb_page(vma))
484 				walk_page_range(mm, vma->vm_start, vma->vm_end,
485 						&clear_refs_walk, vma);
486 		flush_tlb_mm(mm);
487 		up_read(&mm->mmap_sem);
488 		mmput(mm);
489 	}
490 	put_task_struct(task);
491 	if (end - buffer == 0)
492 		return -EIO;
493 	return end - buffer;
494 }
495 
496 const struct file_operations proc_clear_refs_operations = {
497 	.write		= clear_refs_write,
498 };
499 
500 struct pagemapread {
501 	char __user *out, *end;
502 };
503 
504 #define PM_ENTRY_BYTES      sizeof(u64)
505 #define PM_STATUS_BITS      3
506 #define PM_STATUS_OFFSET    (64 - PM_STATUS_BITS)
507 #define PM_STATUS_MASK      (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
508 #define PM_STATUS(nr)       (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
509 #define PM_PSHIFT_BITS      6
510 #define PM_PSHIFT_OFFSET    (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
511 #define PM_PSHIFT_MASK      (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
512 #define PM_PSHIFT(x)        (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
513 #define PM_PFRAME_MASK      ((1LL << PM_PSHIFT_OFFSET) - 1)
514 #define PM_PFRAME(x)        ((x) & PM_PFRAME_MASK)
515 
516 #define PM_PRESENT          PM_STATUS(4LL)
517 #define PM_SWAP             PM_STATUS(2LL)
518 #define PM_NOT_PRESENT      PM_PSHIFT(PAGE_SHIFT)
519 #define PM_END_OF_BUFFER    1
520 
521 static int add_to_pagemap(unsigned long addr, u64 pfn,
522 			  struct pagemapread *pm)
523 {
524 	/*
525 	 * Make sure there's room in the buffer for an
526 	 * entire entry.  Otherwise, only copy part of
527 	 * the pfn.
528 	 */
529 	if (pm->out + PM_ENTRY_BYTES >= pm->end) {
530 		if (copy_to_user(pm->out, &pfn, pm->end - pm->out))
531 			return -EFAULT;
532 		pm->out = pm->end;
533 		return PM_END_OF_BUFFER;
534 	}
535 
536 	if (put_user(pfn, pm->out))
537 		return -EFAULT;
538 	pm->out += PM_ENTRY_BYTES;
539 	return 0;
540 }
541 
542 static int pagemap_pte_hole(unsigned long start, unsigned long end,
543 				void *private)
544 {
545 	struct pagemapread *pm = private;
546 	unsigned long addr;
547 	int err = 0;
548 	for (addr = start; addr < end; addr += PAGE_SIZE) {
549 		err = add_to_pagemap(addr, PM_NOT_PRESENT, pm);
550 		if (err)
551 			break;
552 	}
553 	return err;
554 }
555 
556 static u64 swap_pte_to_pagemap_entry(pte_t pte)
557 {
558 	swp_entry_t e = pte_to_swp_entry(pte);
559 	return swp_type(e) | (swp_offset(e) << MAX_SWAPFILES_SHIFT);
560 }
561 
562 static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
563 			     void *private)
564 {
565 	struct pagemapread *pm = private;
566 	pte_t *pte;
567 	int err = 0;
568 
569 	for (; addr != end; addr += PAGE_SIZE) {
570 		u64 pfn = PM_NOT_PRESENT;
571 		pte = pte_offset_map(pmd, addr);
572 		if (is_swap_pte(*pte))
573 			pfn = PM_PFRAME(swap_pte_to_pagemap_entry(*pte))
574 				| PM_PSHIFT(PAGE_SHIFT) | PM_SWAP;
575 		else if (pte_present(*pte))
576 			pfn = PM_PFRAME(pte_pfn(*pte))
577 				| PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
578 		/* unmap so we're not in atomic when we copy to userspace */
579 		pte_unmap(pte);
580 		err = add_to_pagemap(addr, pfn, pm);
581 		if (err)
582 			return err;
583 	}
584 
585 	cond_resched();
586 
587 	return err;
588 }
589 
590 static struct mm_walk pagemap_walk = {
591 	.pmd_entry = pagemap_pte_range,
592 	.pte_hole = pagemap_pte_hole
593 };
594 
595 /*
596  * /proc/pid/pagemap - an array mapping virtual pages to pfns
597  *
598  * For each page in the address space, this file contains one 64-bit entry
599  * consisting of the following:
600  *
601  * Bits 0-55  page frame number (PFN) if present
602  * Bits 0-4   swap type if swapped
603  * Bits 5-55  swap offset if swapped
604  * Bits 55-60 page shift (page size = 1<<page shift)
605  * Bit  61    reserved for future use
606  * Bit  62    page swapped
607  * Bit  63    page present
608  *
609  * If the page is not present but in swap, then the PFN contains an
610  * encoding of the swap file number and the page's offset into the
611  * swap. Unmapped pages return a null PFN. This allows determining
612  * precisely which pages are mapped (or in swap) and comparing mapped
613  * pages between processes.
614  *
615  * Efficient users of this interface will use /proc/pid/maps to
616  * determine which areas of memory are actually mapped and llseek to
617  * skip over unmapped regions.
618  */
619 static ssize_t pagemap_read(struct file *file, char __user *buf,
620 			    size_t count, loff_t *ppos)
621 {
622 	struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
623 	struct page **pages, *page;
624 	unsigned long uaddr, uend;
625 	struct mm_struct *mm;
626 	struct pagemapread pm;
627 	int pagecount;
628 	int ret = -ESRCH;
629 
630 	if (!task)
631 		goto out;
632 
633 	ret = -EACCES;
634 	if (!ptrace_may_attach(task))
635 		goto out_task;
636 
637 	ret = -EINVAL;
638 	/* file position must be aligned */
639 	if (*ppos % PM_ENTRY_BYTES)
640 		goto out_task;
641 
642 	ret = 0;
643 	mm = get_task_mm(task);
644 	if (!mm)
645 		goto out_task;
646 
647 	ret = -ENOMEM;
648 	uaddr = (unsigned long)buf & PAGE_MASK;
649 	uend = (unsigned long)(buf + count);
650 	pagecount = (PAGE_ALIGN(uend) - uaddr) / PAGE_SIZE;
651 	pages = kmalloc(pagecount * sizeof(struct page *), GFP_KERNEL);
652 	if (!pages)
653 		goto out_mm;
654 
655 	down_read(&current->mm->mmap_sem);
656 	ret = get_user_pages(current, current->mm, uaddr, pagecount,
657 			     1, 0, pages, NULL);
658 	up_read(&current->mm->mmap_sem);
659 
660 	if (ret < 0)
661 		goto out_free;
662 
663 	if (ret != pagecount) {
664 		pagecount = ret;
665 		ret = -EFAULT;
666 		goto out_pages;
667 	}
668 
669 	pm.out = buf;
670 	pm.end = buf + count;
671 
672 	if (!ptrace_may_attach(task)) {
673 		ret = -EIO;
674 	} else {
675 		unsigned long src = *ppos;
676 		unsigned long svpfn = src / PM_ENTRY_BYTES;
677 		unsigned long start_vaddr = svpfn << PAGE_SHIFT;
678 		unsigned long end_vaddr = TASK_SIZE_OF(task);
679 
680 		/* watch out for wraparound */
681 		if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
682 			start_vaddr = end_vaddr;
683 
684 		/*
685 		 * The odds are that this will stop walking way
686 		 * before end_vaddr, because the length of the
687 		 * user buffer is tracked in "pm", and the walk
688 		 * will stop when we hit the end of the buffer.
689 		 */
690 		ret = walk_page_range(mm, start_vaddr, end_vaddr,
691 					&pagemap_walk, &pm);
692 		if (ret == PM_END_OF_BUFFER)
693 			ret = 0;
694 		/* don't need mmap_sem for these, but this looks cleaner */
695 		*ppos += pm.out - buf;
696 		if (!ret)
697 			ret = pm.out - buf;
698 	}
699 
700 out_pages:
701 	for (; pagecount; pagecount--) {
702 		page = pages[pagecount-1];
703 		if (!PageReserved(page))
704 			SetPageDirty(page);
705 		page_cache_release(page);
706 	}
707 out_free:
708 	kfree(pages);
709 out_mm:
710 	mmput(mm);
711 out_task:
712 	put_task_struct(task);
713 out:
714 	return ret;
715 }
716 
717 const struct file_operations proc_pagemap_operations = {
718 	.llseek		= mem_lseek, /* borrow this */
719 	.read		= pagemap_read,
720 };
721 #endif /* CONFIG_PROC_PAGE_MONITOR */
722 
723 #ifdef CONFIG_NUMA
724 extern int show_numa_map(struct seq_file *m, void *v);
725 
726 static int show_numa_map_checked(struct seq_file *m, void *v)
727 {
728 	struct proc_maps_private *priv = m->private;
729 	struct task_struct *task = priv->task;
730 
731 	if (maps_protect && !ptrace_may_attach(task))
732 		return -EACCES;
733 
734 	return show_numa_map(m, v);
735 }
736 
737 static const struct seq_operations proc_pid_numa_maps_op = {
738         .start  = m_start,
739         .next   = m_next,
740         .stop   = m_stop,
741         .show   = show_numa_map_checked
742 };
743 
744 static int numa_maps_open(struct inode *inode, struct file *file)
745 {
746 	return do_maps_open(inode, file, &proc_pid_numa_maps_op);
747 }
748 
749 const struct file_operations proc_numa_maps_operations = {
750 	.open		= numa_maps_open,
751 	.read		= seq_read,
752 	.llseek		= seq_lseek,
753 	.release	= seq_release_private,
754 };
755 #endif
756