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>
15 #include <asm/uaccess.h>
16 #include <asm/tlbflush.h>
19 void task_mem(struct seq_file
*m
, struct mm_struct
*mm
)
21 unsigned long data
, text
, lib
;
22 unsigned long hiwater_vm
, total_vm
, hiwater_rss
, total_rss
;
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.
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
;
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
;
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);
62 unsigned long task_vsize(struct mm_struct
*mm
)
64 return PAGE_SIZE
* mm
->total_vm
;
67 int task_statm(struct mm_struct
*mm
, int *shared
, int *text
,
68 int *data
, int *resident
)
70 *shared
= get_mm_counter(mm
, file_rss
);
71 *text
= (PAGE_ALIGN(mm
->end_code
) - (mm
->start_code
& PAGE_MASK
))
73 *data
= mm
->total_vm
- mm
->shared_vm
;
74 *resident
= *shared
+ get_mm_counter(mm
, anon_rss
);
78 static void pad_len_spaces(struct seq_file
*m
, int len
)
80 len
= 25 + sizeof(void*) * 6 - len
;
83 seq_printf(m
, "%*c", len
, ' ');
86 static void vma_stop(struct proc_maps_private
*priv
, struct vm_area_struct
*vma
)
88 if (vma
&& vma
!= priv
->tail_vma
) {
89 struct mm_struct
*mm
= vma
->vm_mm
;
90 up_read(&mm
->mmap_sem
);
95 static void *m_start(struct seq_file
*m
, loff_t
*pos
)
97 struct proc_maps_private
*priv
= m
->private;
98 unsigned long last_addr
= m
->version
;
100 struct vm_area_struct
*vma
, *tail_vma
= NULL
;
103 /* Clear the per syscall fields in priv */
105 priv
->tail_vma
= NULL
;
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.
114 if (last_addr
== -1UL)
117 priv
->task
= get_pid_task(priv
->pid
, PIDTYPE_PID
);
121 mm
= mm_for_maps(priv
->task
);
125 tail_vma
= get_gate_vma(priv
->task
);
126 priv
->tail_vma
= tail_vma
;
128 /* Start with last addr hint */
129 vma
= find_vma(mm
, last_addr
);
130 if (last_addr
&& vma
) {
136 * Check the vma index is within the range and do
137 * sequential scan until m_index.
140 if ((unsigned long)l
< mm
->map_count
) {
147 if (l
!= mm
->map_count
)
148 tail_vma
= NULL
; /* After gate vma */
154 /* End of vmas has been reached */
155 m
->version
= (tail_vma
!= NULL
)? 0: -1UL;
156 up_read(&mm
->mmap_sem
);
161 static void *m_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
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
;
168 if (vma
&& (vma
!= tail_vma
) && vma
->vm_next
)
171 return (vma
!= tail_vma
)? tail_vma
: NULL
;
174 static void m_stop(struct seq_file
*m
, void *v
)
176 struct proc_maps_private
*priv
= m
->private;
177 struct vm_area_struct
*vma
= v
;
181 put_task_struct(priv
->task
);
184 static int do_maps_open(struct inode
*inode
, struct file
*file
,
185 const struct seq_operations
*ops
)
187 struct proc_maps_private
*priv
;
189 priv
= kzalloc(sizeof(*priv
), GFP_KERNEL
);
191 priv
->pid
= proc_pid(inode
);
192 ret
= seq_open(file
, ops
);
194 struct seq_file
*m
= file
->private_data
;
203 static int show_map(struct seq_file
*m
, void *v
)
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;
215 if (maps_protect
&& !ptrace_may_attach(task
))
219 struct inode
*inode
= vma
->vm_file
->f_path
.dentry
->d_inode
;
220 dev
= inode
->i_sb
->s_dev
;
224 seq_printf(m
, "%08lx-%08lx %c%c%c%c %08lx %02x:%02x %lu %n",
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
);
235 * Print the dentry name for named mappings, and a
236 * special [heap] marker for the heap:
239 pad_len_spaces(m
, len
);
240 seq_path(m
, &file
->f_path
, "\n");
242 const char *name
= arch_vma_name(vma
);
245 if (vma
->vm_start
<= mm
->start_brk
&&
246 vma
->vm_end
>= mm
->brk
) {
248 } else if (vma
->vm_start
<= mm
->start_stack
&&
249 vma
->vm_end
>= mm
->start_stack
) {
257 pad_len_spaces(m
, len
);
263 if (m
->count
< m
->size
) /* vma is copied successfully */
264 m
->version
= (vma
!= get_gate_vma(task
))? vma
->vm_start
: 0;
268 static const struct seq_operations proc_pid_maps_op
= {
275 static int maps_open(struct inode
*inode
, struct file
*file
)
277 return do_maps_open(inode
, file
, &proc_pid_maps_op
);
280 const struct file_operations proc_maps_operations
= {
284 .release
= seq_release_private
,
288 * Proportional Set Size(PSS): my share of RSS.
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.
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.
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.
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
;
319 static int smaps_pte_range(pmd_t
*pmd
, unsigned long addr
, unsigned long end
,
322 struct mem_size_stats
*mss
= private;
323 struct vm_area_struct
*vma
= mss
->vma
;
329 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
330 for (; addr
!= end
; pte
++, addr
+= PAGE_SIZE
) {
333 if (is_swap_pte(ptent
)) {
334 mss
->swap
+= PAGE_SIZE
;
338 if (!pte_present(ptent
))
341 mss
->resident
+= PAGE_SIZE
;
343 page
= vm_normal_page(vma
, addr
, ptent
);
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
);
352 if (pte_dirty(ptent
))
353 mss
->shared_dirty
+= PAGE_SIZE
;
355 mss
->shared_clean
+= PAGE_SIZE
;
356 mss
->pss
+= (PAGE_SIZE
<< PSS_SHIFT
) / mapcount
;
358 if (pte_dirty(ptent
))
359 mss
->private_dirty
+= PAGE_SIZE
;
361 mss
->private_clean
+= PAGE_SIZE
;
362 mss
->pss
+= (PAGE_SIZE
<< PSS_SHIFT
);
365 pte_unmap_unlock(pte
- 1, ptl
);
370 static struct mm_walk smaps_walk
= { .pmd_entry
= smaps_pte_range
};
372 static int show_smap(struct seq_file
*m
, void *v
)
374 struct vm_area_struct
*vma
= v
;
375 struct mem_size_stats mss
;
378 memset(&mss
, 0, sizeof mss
);
380 if (vma
->vm_mm
&& !is_vm_hugetlb_page(vma
))
381 walk_page_range(vma
->vm_mm
, vma
->vm_start
, vma
->vm_end
,
384 ret
= show_map(m
, v
);
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"
398 (vma
->vm_end
- vma
->vm_start
) >> 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,
411 static const struct seq_operations proc_pid_smaps_op
= {
418 static int smaps_open(struct inode
*inode
, struct file
*file
)
420 return do_maps_open(inode
, file
, &proc_pid_smaps_op
);
423 const struct file_operations proc_smaps_operations
= {
427 .release
= seq_release_private
,
430 static int clear_refs_pte_range(pmd_t
*pmd
, unsigned long addr
,
431 unsigned long end
, void *private)
433 struct vm_area_struct
*vma
= private;
438 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
439 for (; addr
!= end
; pte
++, addr
+= PAGE_SIZE
) {
441 if (!pte_present(ptent
))
444 page
= vm_normal_page(vma
, addr
, ptent
);
448 /* Clear accessed and referenced bits. */
449 ptep_test_and_clear_young(vma
, addr
, pte
);
450 ClearPageReferenced(page
);
452 pte_unmap_unlock(pte
- 1, ptl
);
457 static struct mm_walk clear_refs_walk
= { .pmd_entry
= clear_refs_pte_range
};
459 static ssize_t
clear_refs_write(struct file
*file
, const char __user
*buf
,
460 size_t count
, loff_t
*ppos
)
462 struct task_struct
*task
;
463 char buffer
[PROC_NUMBUF
], *end
;
464 struct mm_struct
*mm
;
465 struct vm_area_struct
*vma
;
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
))
472 if (!simple_strtol(buffer
, &end
, 0))
476 task
= get_proc_task(file
->f_path
.dentry
->d_inode
);
479 mm
= get_task_mm(task
);
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
);
487 up_read(&mm
->mmap_sem
);
490 put_task_struct(task
);
491 if (end
- buffer
== 0)
496 const struct file_operations proc_clear_refs_operations
= {
497 .write
= clear_refs_write
,
501 char __user
*out
, *end
;
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)
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
521 static int add_to_pagemap(unsigned long addr
, u64 pfn
,
522 struct pagemapread
*pm
)
525 * Make sure there's room in the buffer for an
526 * entire entry. Otherwise, only copy part of
529 if (pm
->out
+ PM_ENTRY_BYTES
>= pm
->end
) {
530 if (copy_to_user(pm
->out
, &pfn
, pm
->end
- pm
->out
))
533 return PM_END_OF_BUFFER
;
536 if (put_user(pfn
, pm
->out
))
538 pm
->out
+= PM_ENTRY_BYTES
;
542 static int pagemap_pte_hole(unsigned long start
, unsigned long end
,
545 struct pagemapread
*pm
= private;
548 for (addr
= start
; addr
< end
; addr
+= PAGE_SIZE
) {
549 err
= add_to_pagemap(addr
, PM_NOT_PRESENT
, pm
);
556 static u64
swap_pte_to_pagemap_entry(pte_t pte
)
558 swp_entry_t e
= pte_to_swp_entry(pte
);
559 return swp_type(e
) | (swp_offset(e
) << MAX_SWAPFILES_SHIFT
);
562 static int pagemap_pte_range(pmd_t
*pmd
, unsigned long addr
, unsigned long end
,
565 struct pagemapread
*pm
= private;
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 */
580 err
= add_to_pagemap(addr
, pfn
, pm
);
590 static struct mm_walk pagemap_walk
= {
591 .pmd_entry
= pagemap_pte_range
,
592 .pte_hole
= pagemap_pte_hole
596 * /proc/pid/pagemap - an array mapping virtual pages to pfns
598 * For each page in the address space, this file contains one 64-bit entry
599 * consisting of the following:
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
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.
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.
619 static ssize_t
pagemap_read(struct file
*file
, char __user
*buf
,
620 size_t count
, loff_t
*ppos
)
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
;
634 if (!ptrace_may_attach(task
))
638 /* file position must be aligned */
639 if (*ppos
% PM_ENTRY_BYTES
)
643 mm
= get_task_mm(task
);
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
);
655 down_read(¤t
->mm
->mmap_sem
);
656 ret
= get_user_pages(current
, current
->mm
, uaddr
, pagecount
,
658 up_read(¤t
->mm
->mmap_sem
);
663 if (ret
!= pagecount
) {
670 pm
.end
= buf
+ count
;
672 if (!ptrace_may_attach(task
)) {
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
);
680 /* watch out for wraparound */
681 if (svpfn
> TASK_SIZE_OF(task
) >> PAGE_SHIFT
)
682 start_vaddr
= end_vaddr
;
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.
690 ret
= walk_page_range(mm
, start_vaddr
, end_vaddr
,
692 if (ret
== PM_END_OF_BUFFER
)
694 /* don't need mmap_sem for these, but this looks cleaner */
695 *ppos
+= pm
.out
- buf
;
701 for (; pagecount
; pagecount
--) {
702 page
= pages
[pagecount
-1];
703 if (!PageReserved(page
))
705 page_cache_release(page
);
712 put_task_struct(task
);
717 const struct file_operations proc_pagemap_operations
= {
718 .llseek
= mem_lseek
, /* borrow this */
719 .read
= pagemap_read
,
721 #endif /* CONFIG_PROC_PAGE_MONITOR */
724 extern int show_numa_map(struct seq_file
*m
, void *v
);
726 static int show_numa_map_checked(struct seq_file
*m
, void *v
)
728 struct proc_maps_private
*priv
= m
->private;
729 struct task_struct
*task
= priv
->task
;
731 if (maps_protect
&& !ptrace_may_attach(task
))
734 return show_numa_map(m
, v
);
737 static const struct seq_operations proc_pid_numa_maps_op
= {
741 .show
= show_numa_map_checked
744 static int numa_maps_open(struct inode
*inode
, struct file
*file
)
746 return do_maps_open(inode
, file
, &proc_pid_numa_maps_op
);
749 const struct file_operations proc_numa_maps_operations
= {
750 .open
= numa_maps_open
,
753 .release
= seq_release_private
,