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