Merge git://git.kernel.org/pub/scm/linux/kernel/git/agk/linux-2.6-dm
[linux-2.6-block.git] / fs / proc / task_mmu.c
CommitLineData
1da177e4
LT
1#include <linux/mm.h>
2#include <linux/hugetlb.h>
3#include <linux/mount.h>
4#include <linux/seq_file.h>
e070ad49 5#include <linux/highmem.h>
5096add8 6#include <linux/ptrace.h>
6e21c8f1
CL
7#include <linux/pagemap.h>
8#include <linux/mempolicy.h>
85863e47
MM
9#include <linux/swap.h>
10#include <linux/swapops.h>
e070ad49 11
1da177e4
LT
12#include <asm/elf.h>
13#include <asm/uaccess.h>
e070ad49 14#include <asm/tlbflush.h>
1da177e4
LT
15#include "internal.h"
16
df5f8314 17void task_mem(struct seq_file *m, struct mm_struct *mm)
1da177e4
LT
18{
19 unsigned long data, text, lib;
365e9c87
HD
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;
1da177e4
LT
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;
df5f8314 39 seq_printf(m,
365e9c87 40 "VmPeak:\t%8lu kB\n"
1da177e4
LT
41 "VmSize:\t%8lu kB\n"
42 "VmLck:\t%8lu kB\n"
365e9c87 43 "VmHWM:\t%8lu kB\n"
1da177e4
LT
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",
365e9c87
HD
50 hiwater_vm << (PAGE_SHIFT-10),
51 (total_vm - mm->reserved_vm) << (PAGE_SHIFT-10),
1da177e4 52 mm->locked_vm << (PAGE_SHIFT-10),
365e9c87
HD
53 hiwater_rss << (PAGE_SHIFT-10),
54 total_rss << (PAGE_SHIFT-10),
1da177e4
LT
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);
1da177e4
LT
58}
59
60unsigned long task_vsize(struct mm_struct *mm)
61{
62 return PAGE_SIZE * mm->total_vm;
63}
64
65int task_statm(struct mm_struct *mm, int *shared, int *text,
66 int *data, int *resident)
67{
4294621f 68 *shared = get_mm_counter(mm, file_rss);
1da177e4
LT
69 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
70 >> PAGE_SHIFT;
71 *data = mm->total_vm - mm->shared_vm;
4294621f 72 *resident = *shared + get_mm_counter(mm, anon_rss);
1da177e4
LT
73 return mm->total_vm;
74}
75
1da177e4
LT
76static 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
a6198797
MM
84static 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}
ec4dd3eb 92
a6198797 93static void *m_start(struct seq_file *m, loff_t *pos)
e070ad49 94{
a6198797
MM
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
148out:
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
159static 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
172static 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
182static int do_maps_open(struct inode *inode, struct file *file,
03a44825 183 const struct seq_operations *ops)
a6198797
MM
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}
e070ad49 200
7c88db0c 201static void show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
1da177e4 202{
e070ad49
ML
203 struct mm_struct *mm = vma->vm_mm;
204 struct file *file = vma->vm_file;
205 int flags = vma->vm_flags;
1da177e4
LT
206 unsigned long ino = 0;
207 dev_t dev = 0;
208 int len;
209
210 if (file) {
2fddfeef 211 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1da177e4
LT
212 dev = inode->i_sb->s_dev;
213 ino = inode->i_ino;
214 }
215
1804dc6e 216 seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu %n",
e070ad49
ML
217 vma->vm_start,
218 vma->vm_end,
1da177e4
LT
219 flags & VM_READ ? 'r' : '-',
220 flags & VM_WRITE ? 'w' : '-',
221 flags & VM_EXEC ? 'x' : '-',
222 flags & VM_MAYSHARE ? 's' : 'p',
1804dc6e 223 ((loff_t)vma->vm_pgoff) << PAGE_SHIFT,
1da177e4
LT
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 */
e070ad49 230 if (file) {
1da177e4 231 pad_len_spaces(m, len);
c32c2f63 232 seq_path(m, &file->f_path, "\n");
1da177e4 233 } else {
e6e5494c
IM
234 const char *name = arch_vma_name(vma);
235 if (!name) {
236 if (mm) {
237 if (vma->vm_start <= mm->start_brk &&
e070ad49 238 vma->vm_end >= mm->brk) {
e6e5494c
IM
239 name = "[heap]";
240 } else if (vma->vm_start <= mm->start_stack &&
241 vma->vm_end >= mm->start_stack) {
242 name = "[stack]";
1da177e4 243 }
e6e5494c
IM
244 } else {
245 name = "[vdso]";
1da177e4 246 }
e6e5494c
IM
247 }
248 if (name) {
1da177e4 249 pad_len_spaces(m, len);
e6e5494c 250 seq_puts(m, name);
1da177e4
LT
251 }
252 }
253 seq_putc(m, '\n');
7c88db0c
JK
254}
255
256static 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);
e070ad49 263
e070ad49
ML
264 if (m->count < m->size) /* vma is copied successfully */
265 m->version = (vma != get_gate_vma(task))? vma->vm_start: 0;
1da177e4
LT
266 return 0;
267}
268
03a44825 269static const struct seq_operations proc_pid_maps_op = {
a6198797
MM
270 .start = m_start,
271 .next = m_next,
272 .stop = m_stop,
273 .show = show_map
274};
275
276static int maps_open(struct inode *inode, struct file *file)
277{
278 return do_maps_open(inode, file, &proc_pid_maps_op);
279}
280
281const 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
1e883281 307#ifdef CONFIG_PROC_PAGE_MONITOR
214e471f 308struct mem_size_stats {
a6198797
MM
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;
214e471f 316 unsigned long swap;
a6198797
MM
317 u64 pss;
318};
319
b3ae5acb 320static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
2165009b 321 struct mm_walk *walk)
e070ad49 322{
2165009b 323 struct mem_size_stats *mss = walk->private;
b3ae5acb 324 struct vm_area_struct *vma = mss->vma;
e070ad49 325 pte_t *pte, ptent;
705e87c0 326 spinlock_t *ptl;
e070ad49 327 struct page *page;
ec4dd3eb 328 int mapcount;
e070ad49 329
705e87c0 330 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
826fad1b 331 for (; addr != end; pte++, addr += PAGE_SIZE) {
e070ad49 332 ptent = *pte;
214e471f
PZ
333
334 if (is_swap_pte(ptent)) {
335 mss->swap += PAGE_SIZE;
336 continue;
337 }
338
705e87c0 339 if (!pte_present(ptent))
e070ad49
ML
340 continue;
341
342 mss->resident += PAGE_SIZE;
ad820c5d
NP
343
344 page = vm_normal_page(vma, addr, ptent);
345 if (!page)
e070ad49
ML
346 continue;
347
f79f177c
DR
348 /* Accumulate the size in pages that have been accessed. */
349 if (pte_young(ptent) || PageReferenced(page))
350 mss->referenced += PAGE_SIZE;
ec4dd3eb
FW
351 mapcount = page_mapcount(page);
352 if (mapcount >= 2) {
e070ad49
ML
353 if (pte_dirty(ptent))
354 mss->shared_dirty += PAGE_SIZE;
355 else
356 mss->shared_clean += PAGE_SIZE;
ec4dd3eb 357 mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount;
e070ad49
ML
358 } else {
359 if (pte_dirty(ptent))
360 mss->private_dirty += PAGE_SIZE;
361 else
362 mss->private_clean += PAGE_SIZE;
ec4dd3eb 363 mss->pss += (PAGE_SIZE << PSS_SHIFT);
e070ad49 364 }
826fad1b 365 }
705e87c0
HD
366 pte_unmap_unlock(pte - 1, ptl);
367 cond_resched();
b3ae5acb 368 return 0;
e070ad49
ML
369}
370
e070ad49
ML
371static int show_smap(struct seq_file *m, void *v)
372{
7c88db0c
JK
373 struct proc_maps_private *priv = m->private;
374 struct task_struct *task = priv->task;
e070ad49 375 struct vm_area_struct *vma = v;
e070ad49 376 struct mem_size_stats mss;
2165009b
DH
377 struct mm_walk smaps_walk = {
378 .pmd_entry = smaps_pte_range,
379 .mm = vma->vm_mm,
380 .private = &mss,
381 };
e070ad49
ML
382
383 memset(&mss, 0, sizeof mss);
b3ae5acb 384 mss.vma = vma;
5ddfae16 385 if (vma->vm_mm && !is_vm_hugetlb_page(vma))
2165009b 386 walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk);
4752c369 387
7c88db0c 388 show_map_vma(m, vma);
4752c369
MM
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"
214e471f
PZ
398 "Referenced: %8lu kB\n"
399 "Swap: %8lu kB\n",
4752c369
MM
400 (vma->vm_end - vma->vm_start) >> 10,
401 mss.resident >> 10,
402 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
403 mss.shared_clean >> 10,
404 mss.shared_dirty >> 10,
405 mss.private_clean >> 10,
406 mss.private_dirty >> 10,
214e471f
PZ
407 mss.referenced >> 10,
408 mss.swap >> 10);
4752c369 409
7c88db0c
JK
410 if (m->count < m->size) /* vma is copied successfully */
411 m->version = (vma != get_gate_vma(task)) ? vma->vm_start : 0;
412 return 0;
e070ad49
ML
413}
414
03a44825 415static const struct seq_operations proc_pid_smaps_op = {
a6198797
MM
416 .start = m_start,
417 .next = m_next,
418 .stop = m_stop,
419 .show = show_smap
420};
421
422static int smaps_open(struct inode *inode, struct file *file)
423{
424 return do_maps_open(inode, file, &proc_pid_smaps_op);
425}
426
427const struct file_operations proc_smaps_operations = {
428 .open = smaps_open,
429 .read = seq_read,
430 .llseek = seq_lseek,
431 .release = seq_release_private,
432};
433
434static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
2165009b 435 unsigned long end, struct mm_walk *walk)
a6198797 436{
2165009b 437 struct vm_area_struct *vma = walk->private;
a6198797
MM
438 pte_t *pte, ptent;
439 spinlock_t *ptl;
440 struct page *page;
441
442 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
443 for (; addr != end; pte++, addr += PAGE_SIZE) {
444 ptent = *pte;
445 if (!pte_present(ptent))
446 continue;
447
448 page = vm_normal_page(vma, addr, ptent);
449 if (!page)
450 continue;
451
452 /* Clear accessed and referenced bits. */
453 ptep_test_and_clear_young(vma, addr, pte);
454 ClearPageReferenced(page);
455 }
456 pte_unmap_unlock(pte - 1, ptl);
457 cond_resched();
458 return 0;
459}
460
f248dcb3
MM
461static ssize_t clear_refs_write(struct file *file, const char __user *buf,
462 size_t count, loff_t *ppos)
b813e931 463{
f248dcb3
MM
464 struct task_struct *task;
465 char buffer[PROC_NUMBUF], *end;
466 struct mm_struct *mm;
b813e931
DR
467 struct vm_area_struct *vma;
468
f248dcb3
MM
469 memset(buffer, 0, sizeof(buffer));
470 if (count > sizeof(buffer) - 1)
471 count = sizeof(buffer) - 1;
472 if (copy_from_user(buffer, buf, count))
473 return -EFAULT;
474 if (!simple_strtol(buffer, &end, 0))
475 return -EINVAL;
476 if (*end == '\n')
477 end++;
478 task = get_proc_task(file->f_path.dentry->d_inode);
479 if (!task)
480 return -ESRCH;
481 mm = get_task_mm(task);
482 if (mm) {
20cbc972
AM
483 struct mm_walk clear_refs_walk = {
484 .pmd_entry = clear_refs_pte_range,
485 .mm = mm,
486 };
f248dcb3 487 down_read(&mm->mmap_sem);
2165009b
DH
488 for (vma = mm->mmap; vma; vma = vma->vm_next) {
489 clear_refs_walk.private = vma;
f248dcb3 490 if (!is_vm_hugetlb_page(vma))
2165009b
DH
491 walk_page_range(vma->vm_start, vma->vm_end,
492 &clear_refs_walk);
493 }
f248dcb3
MM
494 flush_tlb_mm(mm);
495 up_read(&mm->mmap_sem);
496 mmput(mm);
497 }
498 put_task_struct(task);
499 if (end - buffer == 0)
500 return -EIO;
501 return end - buffer;
b813e931
DR
502}
503
f248dcb3
MM
504const struct file_operations proc_clear_refs_operations = {
505 .write = clear_refs_write,
506};
507
85863e47 508struct pagemapread {
aae8679b 509 u64 __user *out, *end;
85863e47
MM
510};
511
f16278c6
HR
512#define PM_ENTRY_BYTES sizeof(u64)
513#define PM_STATUS_BITS 3
514#define PM_STATUS_OFFSET (64 - PM_STATUS_BITS)
515#define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
516#define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
517#define PM_PSHIFT_BITS 6
518#define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
519#define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
520#define PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
521#define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1)
522#define PM_PFRAME(x) ((x) & PM_PFRAME_MASK)
523
524#define PM_PRESENT PM_STATUS(4LL)
525#define PM_SWAP PM_STATUS(2LL)
526#define PM_NOT_PRESENT PM_PSHIFT(PAGE_SHIFT)
85863e47
MM
527#define PM_END_OF_BUFFER 1
528
529static int add_to_pagemap(unsigned long addr, u64 pfn,
530 struct pagemapread *pm)
531{
85863e47
MM
532 if (put_user(pfn, pm->out))
533 return -EFAULT;
aae8679b
TT
534 pm->out++;
535 if (pm->out >= pm->end)
536 return PM_END_OF_BUFFER;
85863e47
MM
537 return 0;
538}
539
540static int pagemap_pte_hole(unsigned long start, unsigned long end,
2165009b 541 struct mm_walk *walk)
85863e47 542{
2165009b 543 struct pagemapread *pm = walk->private;
85863e47
MM
544 unsigned long addr;
545 int err = 0;
546 for (addr = start; addr < end; addr += PAGE_SIZE) {
547 err = add_to_pagemap(addr, PM_NOT_PRESENT, pm);
548 if (err)
549 break;
550 }
551 return err;
552}
553
9d02dbc8 554static u64 swap_pte_to_pagemap_entry(pte_t pte)
85863e47
MM
555{
556 swp_entry_t e = pte_to_swp_entry(pte);
f16278c6 557 return swp_type(e) | (swp_offset(e) << MAX_SWAPFILES_SHIFT);
85863e47
MM
558}
559
49c50342 560static u64 pte_to_pagemap_entry(pte_t pte)
bcf8039e 561{
49c50342 562 u64 pme = 0;
bcf8039e
DH
563 if (is_swap_pte(pte))
564 pme = PM_PFRAME(swap_pte_to_pagemap_entry(pte))
565 | PM_PSHIFT(PAGE_SHIFT) | PM_SWAP;
566 else if (pte_present(pte))
567 pme = PM_PFRAME(pte_pfn(pte))
568 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
569 return pme;
570}
571
85863e47 572static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
2165009b 573 struct mm_walk *walk)
85863e47 574{
bcf8039e 575 struct vm_area_struct *vma;
2165009b 576 struct pagemapread *pm = walk->private;
85863e47
MM
577 pte_t *pte;
578 int err = 0;
579
bcf8039e
DH
580 /* find the first VMA at or above 'addr' */
581 vma = find_vma(walk->mm, addr);
85863e47
MM
582 for (; addr != end; addr += PAGE_SIZE) {
583 u64 pfn = PM_NOT_PRESENT;
bcf8039e
DH
584
585 /* check to see if we've left 'vma' behind
586 * and need a new, higher one */
587 if (vma && (addr >= vma->vm_end))
588 vma = find_vma(walk->mm, addr);
589
590 /* check that 'vma' actually covers this address,
591 * and that it isn't a huge page vma */
592 if (vma && (vma->vm_start <= addr) &&
593 !is_vm_hugetlb_page(vma)) {
594 pte = pte_offset_map(pmd, addr);
595 pfn = pte_to_pagemap_entry(*pte);
596 /* unmap before userspace copy */
597 pte_unmap(pte);
598 }
85863e47
MM
599 err = add_to_pagemap(addr, pfn, pm);
600 if (err)
601 return err;
602 }
603
604 cond_resched();
605
606 return err;
607}
608
85863e47
MM
609/*
610 * /proc/pid/pagemap - an array mapping virtual pages to pfns
611 *
f16278c6
HR
612 * For each page in the address space, this file contains one 64-bit entry
613 * consisting of the following:
614 *
615 * Bits 0-55 page frame number (PFN) if present
616 * Bits 0-4 swap type if swapped
617 * Bits 5-55 swap offset if swapped
618 * Bits 55-60 page shift (page size = 1<<page shift)
619 * Bit 61 reserved for future use
620 * Bit 62 page swapped
621 * Bit 63 page present
622 *
623 * If the page is not present but in swap, then the PFN contains an
624 * encoding of the swap file number and the page's offset into the
625 * swap. Unmapped pages return a null PFN. This allows determining
85863e47
MM
626 * precisely which pages are mapped (or in swap) and comparing mapped
627 * pages between processes.
628 *
629 * Efficient users of this interface will use /proc/pid/maps to
630 * determine which areas of memory are actually mapped and llseek to
631 * skip over unmapped regions.
632 */
633static ssize_t pagemap_read(struct file *file, char __user *buf,
634 size_t count, loff_t *ppos)
635{
636 struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
637 struct page **pages, *page;
638 unsigned long uaddr, uend;
639 struct mm_struct *mm;
640 struct pagemapread pm;
641 int pagecount;
642 int ret = -ESRCH;
ee1e6ab6 643 struct mm_walk pagemap_walk = {};
5d7e0d2b
AM
644 unsigned long src;
645 unsigned long svpfn;
646 unsigned long start_vaddr;
647 unsigned long end_vaddr;
85863e47
MM
648
649 if (!task)
650 goto out;
651
652 ret = -EACCES;
006ebb40 653 if (!ptrace_may_access(task, PTRACE_MODE_READ))
fb39380b 654 goto out_task;
85863e47
MM
655
656 ret = -EINVAL;
657 /* file position must be aligned */
aae8679b 658 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
fb39380b 659 goto out_task;
85863e47
MM
660
661 ret = 0;
662 mm = get_task_mm(task);
663 if (!mm)
fb39380b 664 goto out_task;
85863e47 665
5d7e0d2b 666
85863e47
MM
667 uaddr = (unsigned long)buf & PAGE_MASK;
668 uend = (unsigned long)(buf + count);
669 pagecount = (PAGE_ALIGN(uend) - uaddr) / PAGE_SIZE;
5d7e0d2b
AM
670 ret = 0;
671 if (pagecount == 0)
672 goto out_mm;
673 pages = kcalloc(pagecount, sizeof(struct page *), GFP_KERNEL);
674 ret = -ENOMEM;
85863e47 675 if (!pages)
fb39380b 676 goto out_mm;
85863e47
MM
677
678 down_read(&current->mm->mmap_sem);
679 ret = get_user_pages(current, current->mm, uaddr, pagecount,
680 1, 0, pages, NULL);
681 up_read(&current->mm->mmap_sem);
682
683 if (ret < 0)
684 goto out_free;
685
fb39380b
MT
686 if (ret != pagecount) {
687 pagecount = ret;
688 ret = -EFAULT;
689 goto out_pages;
690 }
691
aae8679b
TT
692 pm.out = (u64 *)buf;
693 pm.end = (u64 *)(buf + count);
85863e47 694
5d7e0d2b
AM
695 pagemap_walk.pmd_entry = pagemap_pte_range;
696 pagemap_walk.pte_hole = pagemap_pte_hole;
697 pagemap_walk.mm = mm;
698 pagemap_walk.private = &pm;
699
700 src = *ppos;
701 svpfn = src / PM_ENTRY_BYTES;
702 start_vaddr = svpfn << PAGE_SHIFT;
703 end_vaddr = TASK_SIZE_OF(task);
704
705 /* watch out for wraparound */
706 if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
707 start_vaddr = end_vaddr;
708
709 /*
710 * The odds are that this will stop walking way
711 * before end_vaddr, because the length of the
712 * user buffer is tracked in "pm", and the walk
713 * will stop when we hit the end of the buffer.
714 */
715 ret = walk_page_range(start_vaddr, end_vaddr, &pagemap_walk);
716 if (ret == PM_END_OF_BUFFER)
717 ret = 0;
718 /* don't need mmap_sem for these, but this looks cleaner */
719 *ppos += (char *)pm.out - buf;
720 if (!ret)
721 ret = (char *)pm.out - buf;
85863e47 722
fb39380b 723out_pages:
85863e47
MM
724 for (; pagecount; pagecount--) {
725 page = pages[pagecount-1];
726 if (!PageReserved(page))
727 SetPageDirty(page);
728 page_cache_release(page);
729 }
85863e47
MM
730out_free:
731 kfree(pages);
fb39380b
MT
732out_mm:
733 mmput(mm);
85863e47
MM
734out_task:
735 put_task_struct(task);
736out:
737 return ret;
738}
739
740const struct file_operations proc_pagemap_operations = {
741 .llseek = mem_lseek, /* borrow this */
742 .read = pagemap_read,
743};
1e883281 744#endif /* CONFIG_PROC_PAGE_MONITOR */
85863e47 745
6e21c8f1 746#ifdef CONFIG_NUMA
1a75a6c8 747extern int show_numa_map(struct seq_file *m, void *v);
6e21c8f1 748
03a44825 749static const struct seq_operations proc_pid_numa_maps_op = {
1a75a6c8
CL
750 .start = m_start,
751 .next = m_next,
752 .stop = m_stop,
3bbfe059 753 .show = show_numa_map,
6e21c8f1 754};
662795de
EB
755
756static int numa_maps_open(struct inode *inode, struct file *file)
757{
758 return do_maps_open(inode, file, &proc_pid_numa_maps_op);
759}
760
00977a59 761const struct file_operations proc_numa_maps_operations = {
662795de
EB
762 .open = numa_maps_open,
763 .read = seq_read,
764 .llseek = seq_lseek,
99f89551 765 .release = seq_release_private,
662795de 766};
6e21c8f1 767#endif