Loading...
1#include <linux/mm.h>
2#include <linux/hugetlb.h>
3#include <linux/huge_mm.h>
4#include <linux/mount.h>
5#include <linux/seq_file.h>
6#include <linux/highmem.h>
7#include <linux/ptrace.h>
8#include <linux/slab.h>
9#include <linux/pagemap.h>
10#include <linux/mempolicy.h>
11#include <linux/rmap.h>
12#include <linux/swap.h>
13#include <linux/swapops.h>
14
15#include <asm/elf.h>
16#include <asm/uaccess.h>
17#include <asm/tlbflush.h>
18#include "internal.h"
19
20void task_mem(struct seq_file *m, struct mm_struct *mm)
21{
22 unsigned long data, text, lib, swap;
23 unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
24
25 /*
26 * Note: to minimize their overhead, mm maintains hiwater_vm and
27 * hiwater_rss only when about to *lower* total_vm or rss. Any
28 * collector of these hiwater stats must therefore get total_vm
29 * and rss too, which will usually be the higher. Barriers? not
30 * worth the effort, such snapshots can always be inconsistent.
31 */
32 hiwater_vm = total_vm = mm->total_vm;
33 if (hiwater_vm < mm->hiwater_vm)
34 hiwater_vm = mm->hiwater_vm;
35 hiwater_rss = total_rss = get_mm_rss(mm);
36 if (hiwater_rss < mm->hiwater_rss)
37 hiwater_rss = mm->hiwater_rss;
38
39 data = mm->total_vm - mm->shared_vm - mm->stack_vm;
40 text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
41 lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
42 swap = get_mm_counter(mm, MM_SWAPENTS);
43 seq_printf(m,
44 "VmPeak:\t%8lu kB\n"
45 "VmSize:\t%8lu kB\n"
46 "VmLck:\t%8lu kB\n"
47 "VmHWM:\t%8lu kB\n"
48 "VmRSS:\t%8lu kB\n"
49 "VmData:\t%8lu kB\n"
50 "VmStk:\t%8lu kB\n"
51 "VmExe:\t%8lu kB\n"
52 "VmLib:\t%8lu kB\n"
53 "VmPTE:\t%8lu kB\n"
54 "VmSwap:\t%8lu kB\n",
55 hiwater_vm << (PAGE_SHIFT-10),
56 (total_vm - mm->reserved_vm) << (PAGE_SHIFT-10),
57 mm->locked_vm << (PAGE_SHIFT-10),
58 hiwater_rss << (PAGE_SHIFT-10),
59 total_rss << (PAGE_SHIFT-10),
60 data << (PAGE_SHIFT-10),
61 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
62 (PTRS_PER_PTE*sizeof(pte_t)*mm->nr_ptes) >> 10,
63 swap << (PAGE_SHIFT-10));
64}
65
66unsigned long task_vsize(struct mm_struct *mm)
67{
68 return PAGE_SIZE * mm->total_vm;
69}
70
71unsigned long task_statm(struct mm_struct *mm,
72 unsigned long *shared, unsigned long *text,
73 unsigned long *data, unsigned long *resident)
74{
75 *shared = get_mm_counter(mm, MM_FILEPAGES);
76 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
77 >> PAGE_SHIFT;
78 *data = mm->total_vm - mm->shared_vm;
79 *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
80 return mm->total_vm;
81}
82
83static void pad_len_spaces(struct seq_file *m, int len)
84{
85 len = 25 + sizeof(void*) * 6 - len;
86 if (len < 1)
87 len = 1;
88 seq_printf(m, "%*c", len, ' ');
89}
90
91static void vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma)
92{
93 if (vma && vma != priv->tail_vma) {
94 struct mm_struct *mm = vma->vm_mm;
95 up_read(&mm->mmap_sem);
96 mmput(mm);
97 }
98}
99
100static void *m_start(struct seq_file *m, loff_t *pos)
101{
102 struct proc_maps_private *priv = m->private;
103 unsigned long last_addr = m->version;
104 struct mm_struct *mm;
105 struct vm_area_struct *vma, *tail_vma = NULL;
106 loff_t l = *pos;
107
108 /* Clear the per syscall fields in priv */
109 priv->task = NULL;
110 priv->tail_vma = NULL;
111
112 /*
113 * We remember last_addr rather than next_addr to hit with
114 * mmap_cache most of the time. We have zero last_addr at
115 * the beginning and also after lseek. We will have -1 last_addr
116 * after the end of the vmas.
117 */
118
119 if (last_addr == -1UL)
120 return NULL;
121
122 priv->task = get_pid_task(priv->pid, PIDTYPE_PID);
123 if (!priv->task)
124 return ERR_PTR(-ESRCH);
125
126 mm = mm_for_maps(priv->task);
127 if (!mm || IS_ERR(mm))
128 return mm;
129 down_read(&mm->mmap_sem);
130
131 tail_vma = get_gate_vma(priv->task->mm);
132 priv->tail_vma = tail_vma;
133
134 /* Start with last addr hint */
135 vma = find_vma(mm, last_addr);
136 if (last_addr && vma) {
137 vma = vma->vm_next;
138 goto out;
139 }
140
141 /*
142 * Check the vma index is within the range and do
143 * sequential scan until m_index.
144 */
145 vma = NULL;
146 if ((unsigned long)l < mm->map_count) {
147 vma = mm->mmap;
148 while (l-- && vma)
149 vma = vma->vm_next;
150 goto out;
151 }
152
153 if (l != mm->map_count)
154 tail_vma = NULL; /* After gate vma */
155
156out:
157 if (vma)
158 return vma;
159
160 /* End of vmas has been reached */
161 m->version = (tail_vma != NULL)? 0: -1UL;
162 up_read(&mm->mmap_sem);
163 mmput(mm);
164 return tail_vma;
165}
166
167static void *m_next(struct seq_file *m, void *v, loff_t *pos)
168{
169 struct proc_maps_private *priv = m->private;
170 struct vm_area_struct *vma = v;
171 struct vm_area_struct *tail_vma = priv->tail_vma;
172
173 (*pos)++;
174 if (vma && (vma != tail_vma) && vma->vm_next)
175 return vma->vm_next;
176 vma_stop(priv, vma);
177 return (vma != tail_vma)? tail_vma: NULL;
178}
179
180static void m_stop(struct seq_file *m, void *v)
181{
182 struct proc_maps_private *priv = m->private;
183 struct vm_area_struct *vma = v;
184
185 if (!IS_ERR(vma))
186 vma_stop(priv, vma);
187 if (priv->task)
188 put_task_struct(priv->task);
189}
190
191static int do_maps_open(struct inode *inode, struct file *file,
192 const struct seq_operations *ops)
193{
194 struct proc_maps_private *priv;
195 int ret = -ENOMEM;
196 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
197 if (priv) {
198 priv->pid = proc_pid(inode);
199 ret = seq_open(file, ops);
200 if (!ret) {
201 struct seq_file *m = file->private_data;
202 m->private = priv;
203 } else {
204 kfree(priv);
205 }
206 }
207 return ret;
208}
209
210static void show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
211{
212 struct mm_struct *mm = vma->vm_mm;
213 struct file *file = vma->vm_file;
214 vm_flags_t flags = vma->vm_flags;
215 unsigned long ino = 0;
216 unsigned long long pgoff = 0;
217 unsigned long start, end;
218 dev_t dev = 0;
219 int len;
220
221 if (file) {
222 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
223 dev = inode->i_sb->s_dev;
224 ino = inode->i_ino;
225 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
226 }
227
228 /* We don't show the stack guard page in /proc/maps */
229 start = vma->vm_start;
230 if (stack_guard_page_start(vma, start))
231 start += PAGE_SIZE;
232 end = vma->vm_end;
233 if (stack_guard_page_end(vma, end))
234 end -= PAGE_SIZE;
235
236 seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu %n",
237 start,
238 end,
239 flags & VM_READ ? 'r' : '-',
240 flags & VM_WRITE ? 'w' : '-',
241 flags & VM_EXEC ? 'x' : '-',
242 flags & VM_MAYSHARE ? 's' : 'p',
243 pgoff,
244 MAJOR(dev), MINOR(dev), ino, &len);
245
246 /*
247 * Print the dentry name for named mappings, and a
248 * special [heap] marker for the heap:
249 */
250 if (file) {
251 pad_len_spaces(m, len);
252 seq_path(m, &file->f_path, "\n");
253 } else {
254 const char *name = arch_vma_name(vma);
255 if (!name) {
256 if (mm) {
257 if (vma->vm_start <= mm->brk &&
258 vma->vm_end >= mm->start_brk) {
259 name = "[heap]";
260 } else if (vma->vm_start <= mm->start_stack &&
261 vma->vm_end >= mm->start_stack) {
262 name = "[stack]";
263 }
264 } else {
265 name = "[vdso]";
266 }
267 }
268 if (name) {
269 pad_len_spaces(m, len);
270 seq_puts(m, name);
271 }
272 }
273 seq_putc(m, '\n');
274}
275
276static int show_map(struct seq_file *m, void *v)
277{
278 struct vm_area_struct *vma = v;
279 struct proc_maps_private *priv = m->private;
280 struct task_struct *task = priv->task;
281
282 show_map_vma(m, vma);
283
284 if (m->count < m->size) /* vma is copied successfully */
285 m->version = (vma != get_gate_vma(task->mm))
286 ? vma->vm_start : 0;
287 return 0;
288}
289
290static const struct seq_operations proc_pid_maps_op = {
291 .start = m_start,
292 .next = m_next,
293 .stop = m_stop,
294 .show = show_map
295};
296
297static int maps_open(struct inode *inode, struct file *file)
298{
299 return do_maps_open(inode, file, &proc_pid_maps_op);
300}
301
302const struct file_operations proc_maps_operations = {
303 .open = maps_open,
304 .read = seq_read,
305 .llseek = seq_lseek,
306 .release = seq_release_private,
307};
308
309/*
310 * Proportional Set Size(PSS): my share of RSS.
311 *
312 * PSS of a process is the count of pages it has in memory, where each
313 * page is divided by the number of processes sharing it. So if a
314 * process has 1000 pages all to itself, and 1000 shared with one other
315 * process, its PSS will be 1500.
316 *
317 * To keep (accumulated) division errors low, we adopt a 64bit
318 * fixed-point pss counter to minimize division errors. So (pss >>
319 * PSS_SHIFT) would be the real byte count.
320 *
321 * A shift of 12 before division means (assuming 4K page size):
322 * - 1M 3-user-pages add up to 8KB errors;
323 * - supports mapcount up to 2^24, or 16M;
324 * - supports PSS up to 2^52 bytes, or 4PB.
325 */
326#define PSS_SHIFT 12
327
328#ifdef CONFIG_PROC_PAGE_MONITOR
329struct mem_size_stats {
330 struct vm_area_struct *vma;
331 unsigned long resident;
332 unsigned long shared_clean;
333 unsigned long shared_dirty;
334 unsigned long private_clean;
335 unsigned long private_dirty;
336 unsigned long referenced;
337 unsigned long anonymous;
338 unsigned long anonymous_thp;
339 unsigned long swap;
340 u64 pss;
341};
342
343
344static void smaps_pte_entry(pte_t ptent, unsigned long addr,
345 unsigned long ptent_size, struct mm_walk *walk)
346{
347 struct mem_size_stats *mss = walk->private;
348 struct vm_area_struct *vma = mss->vma;
349 struct page *page;
350 int mapcount;
351
352 if (is_swap_pte(ptent)) {
353 mss->swap += ptent_size;
354 return;
355 }
356
357 if (!pte_present(ptent))
358 return;
359
360 page = vm_normal_page(vma, addr, ptent);
361 if (!page)
362 return;
363
364 if (PageAnon(page))
365 mss->anonymous += ptent_size;
366
367 mss->resident += ptent_size;
368 /* Accumulate the size in pages that have been accessed. */
369 if (pte_young(ptent) || PageReferenced(page))
370 mss->referenced += ptent_size;
371 mapcount = page_mapcount(page);
372 if (mapcount >= 2) {
373 if (pte_dirty(ptent) || PageDirty(page))
374 mss->shared_dirty += ptent_size;
375 else
376 mss->shared_clean += ptent_size;
377 mss->pss += (ptent_size << PSS_SHIFT) / mapcount;
378 } else {
379 if (pte_dirty(ptent) || PageDirty(page))
380 mss->private_dirty += ptent_size;
381 else
382 mss->private_clean += ptent_size;
383 mss->pss += (ptent_size << PSS_SHIFT);
384 }
385}
386
387static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
388 struct mm_walk *walk)
389{
390 struct mem_size_stats *mss = walk->private;
391 struct vm_area_struct *vma = mss->vma;
392 pte_t *pte;
393 spinlock_t *ptl;
394
395 spin_lock(&walk->mm->page_table_lock);
396 if (pmd_trans_huge(*pmd)) {
397 if (pmd_trans_splitting(*pmd)) {
398 spin_unlock(&walk->mm->page_table_lock);
399 wait_split_huge_page(vma->anon_vma, pmd);
400 } else {
401 smaps_pte_entry(*(pte_t *)pmd, addr,
402 HPAGE_PMD_SIZE, walk);
403 spin_unlock(&walk->mm->page_table_lock);
404 mss->anonymous_thp += HPAGE_PMD_SIZE;
405 return 0;
406 }
407 } else {
408 spin_unlock(&walk->mm->page_table_lock);
409 }
410 /*
411 * The mmap_sem held all the way back in m_start() is what
412 * keeps khugepaged out of here and from collapsing things
413 * in here.
414 */
415 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
416 for (; addr != end; pte++, addr += PAGE_SIZE)
417 smaps_pte_entry(*pte, addr, PAGE_SIZE, walk);
418 pte_unmap_unlock(pte - 1, ptl);
419 cond_resched();
420 return 0;
421}
422
423static int show_smap(struct seq_file *m, void *v)
424{
425 struct proc_maps_private *priv = m->private;
426 struct task_struct *task = priv->task;
427 struct vm_area_struct *vma = v;
428 struct mem_size_stats mss;
429 struct mm_walk smaps_walk = {
430 .pmd_entry = smaps_pte_range,
431 .mm = vma->vm_mm,
432 .private = &mss,
433 };
434
435 memset(&mss, 0, sizeof mss);
436 mss.vma = vma;
437 /* mmap_sem is held in m_start */
438 if (vma->vm_mm && !is_vm_hugetlb_page(vma))
439 walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk);
440
441 show_map_vma(m, vma);
442
443 seq_printf(m,
444 "Size: %8lu kB\n"
445 "Rss: %8lu kB\n"
446 "Pss: %8lu kB\n"
447 "Shared_Clean: %8lu kB\n"
448 "Shared_Dirty: %8lu kB\n"
449 "Private_Clean: %8lu kB\n"
450 "Private_Dirty: %8lu kB\n"
451 "Referenced: %8lu kB\n"
452 "Anonymous: %8lu kB\n"
453 "AnonHugePages: %8lu kB\n"
454 "Swap: %8lu kB\n"
455 "KernelPageSize: %8lu kB\n"
456 "MMUPageSize: %8lu kB\n"
457 "Locked: %8lu kB\n",
458 (vma->vm_end - vma->vm_start) >> 10,
459 mss.resident >> 10,
460 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
461 mss.shared_clean >> 10,
462 mss.shared_dirty >> 10,
463 mss.private_clean >> 10,
464 mss.private_dirty >> 10,
465 mss.referenced >> 10,
466 mss.anonymous >> 10,
467 mss.anonymous_thp >> 10,
468 mss.swap >> 10,
469 vma_kernel_pagesize(vma) >> 10,
470 vma_mmu_pagesize(vma) >> 10,
471 (vma->vm_flags & VM_LOCKED) ?
472 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
473
474 if (m->count < m->size) /* vma is copied successfully */
475 m->version = (vma != get_gate_vma(task->mm))
476 ? vma->vm_start : 0;
477 return 0;
478}
479
480static const struct seq_operations proc_pid_smaps_op = {
481 .start = m_start,
482 .next = m_next,
483 .stop = m_stop,
484 .show = show_smap
485};
486
487static int smaps_open(struct inode *inode, struct file *file)
488{
489 return do_maps_open(inode, file, &proc_pid_smaps_op);
490}
491
492const struct file_operations proc_smaps_operations = {
493 .open = smaps_open,
494 .read = seq_read,
495 .llseek = seq_lseek,
496 .release = seq_release_private,
497};
498
499static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
500 unsigned long end, struct mm_walk *walk)
501{
502 struct vm_area_struct *vma = walk->private;
503 pte_t *pte, ptent;
504 spinlock_t *ptl;
505 struct page *page;
506
507 split_huge_page_pmd(walk->mm, pmd);
508
509 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
510 for (; addr != end; pte++, addr += PAGE_SIZE) {
511 ptent = *pte;
512 if (!pte_present(ptent))
513 continue;
514
515 page = vm_normal_page(vma, addr, ptent);
516 if (!page)
517 continue;
518
519 /* Clear accessed and referenced bits. */
520 ptep_test_and_clear_young(vma, addr, pte);
521 ClearPageReferenced(page);
522 }
523 pte_unmap_unlock(pte - 1, ptl);
524 cond_resched();
525 return 0;
526}
527
528#define CLEAR_REFS_ALL 1
529#define CLEAR_REFS_ANON 2
530#define CLEAR_REFS_MAPPED 3
531
532static ssize_t clear_refs_write(struct file *file, const char __user *buf,
533 size_t count, loff_t *ppos)
534{
535 struct task_struct *task;
536 char buffer[PROC_NUMBUF];
537 struct mm_struct *mm;
538 struct vm_area_struct *vma;
539 int type;
540 int rv;
541
542 memset(buffer, 0, sizeof(buffer));
543 if (count > sizeof(buffer) - 1)
544 count = sizeof(buffer) - 1;
545 if (copy_from_user(buffer, buf, count))
546 return -EFAULT;
547 rv = kstrtoint(strstrip(buffer), 10, &type);
548 if (rv < 0)
549 return rv;
550 if (type < CLEAR_REFS_ALL || type > CLEAR_REFS_MAPPED)
551 return -EINVAL;
552 task = get_proc_task(file->f_path.dentry->d_inode);
553 if (!task)
554 return -ESRCH;
555 mm = get_task_mm(task);
556 if (mm) {
557 struct mm_walk clear_refs_walk = {
558 .pmd_entry = clear_refs_pte_range,
559 .mm = mm,
560 };
561 down_read(&mm->mmap_sem);
562 for (vma = mm->mmap; vma; vma = vma->vm_next) {
563 clear_refs_walk.private = vma;
564 if (is_vm_hugetlb_page(vma))
565 continue;
566 /*
567 * Writing 1 to /proc/pid/clear_refs affects all pages.
568 *
569 * Writing 2 to /proc/pid/clear_refs only affects
570 * Anonymous pages.
571 *
572 * Writing 3 to /proc/pid/clear_refs only affects file
573 * mapped pages.
574 */
575 if (type == CLEAR_REFS_ANON && vma->vm_file)
576 continue;
577 if (type == CLEAR_REFS_MAPPED && !vma->vm_file)
578 continue;
579 walk_page_range(vma->vm_start, vma->vm_end,
580 &clear_refs_walk);
581 }
582 flush_tlb_mm(mm);
583 up_read(&mm->mmap_sem);
584 mmput(mm);
585 }
586 put_task_struct(task);
587
588 return count;
589}
590
591const struct file_operations proc_clear_refs_operations = {
592 .write = clear_refs_write,
593 .llseek = noop_llseek,
594};
595
596struct pagemapread {
597 int pos, len;
598 u64 *buffer;
599};
600
601#define PM_ENTRY_BYTES sizeof(u64)
602#define PM_STATUS_BITS 3
603#define PM_STATUS_OFFSET (64 - PM_STATUS_BITS)
604#define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
605#define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
606#define PM_PSHIFT_BITS 6
607#define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
608#define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
609#define PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
610#define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1)
611#define PM_PFRAME(x) ((x) & PM_PFRAME_MASK)
612
613#define PM_PRESENT PM_STATUS(4LL)
614#define PM_SWAP PM_STATUS(2LL)
615#define PM_NOT_PRESENT PM_PSHIFT(PAGE_SHIFT)
616#define PM_END_OF_BUFFER 1
617
618static int add_to_pagemap(unsigned long addr, u64 pfn,
619 struct pagemapread *pm)
620{
621 pm->buffer[pm->pos++] = pfn;
622 if (pm->pos >= pm->len)
623 return PM_END_OF_BUFFER;
624 return 0;
625}
626
627static int pagemap_pte_hole(unsigned long start, unsigned long end,
628 struct mm_walk *walk)
629{
630 struct pagemapread *pm = walk->private;
631 unsigned long addr;
632 int err = 0;
633 for (addr = start; addr < end; addr += PAGE_SIZE) {
634 err = add_to_pagemap(addr, PM_NOT_PRESENT, pm);
635 if (err)
636 break;
637 }
638 return err;
639}
640
641static u64 swap_pte_to_pagemap_entry(pte_t pte)
642{
643 swp_entry_t e = pte_to_swp_entry(pte);
644 return swp_type(e) | (swp_offset(e) << MAX_SWAPFILES_SHIFT);
645}
646
647static u64 pte_to_pagemap_entry(pte_t pte)
648{
649 u64 pme = 0;
650 if (is_swap_pte(pte))
651 pme = PM_PFRAME(swap_pte_to_pagemap_entry(pte))
652 | PM_PSHIFT(PAGE_SHIFT) | PM_SWAP;
653 else if (pte_present(pte))
654 pme = PM_PFRAME(pte_pfn(pte))
655 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
656 return pme;
657}
658
659static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
660 struct mm_walk *walk)
661{
662 struct vm_area_struct *vma;
663 struct pagemapread *pm = walk->private;
664 pte_t *pte;
665 int err = 0;
666
667 split_huge_page_pmd(walk->mm, pmd);
668
669 /* find the first VMA at or above 'addr' */
670 vma = find_vma(walk->mm, addr);
671 for (; addr != end; addr += PAGE_SIZE) {
672 u64 pfn = PM_NOT_PRESENT;
673
674 /* check to see if we've left 'vma' behind
675 * and need a new, higher one */
676 if (vma && (addr >= vma->vm_end))
677 vma = find_vma(walk->mm, addr);
678
679 /* check that 'vma' actually covers this address,
680 * and that it isn't a huge page vma */
681 if (vma && (vma->vm_start <= addr) &&
682 !is_vm_hugetlb_page(vma)) {
683 pte = pte_offset_map(pmd, addr);
684 pfn = pte_to_pagemap_entry(*pte);
685 /* unmap before userspace copy */
686 pte_unmap(pte);
687 }
688 err = add_to_pagemap(addr, pfn, pm);
689 if (err)
690 return err;
691 }
692
693 cond_resched();
694
695 return err;
696}
697
698#ifdef CONFIG_HUGETLB_PAGE
699static u64 huge_pte_to_pagemap_entry(pte_t pte, int offset)
700{
701 u64 pme = 0;
702 if (pte_present(pte))
703 pme = PM_PFRAME(pte_pfn(pte) + offset)
704 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
705 return pme;
706}
707
708/* This function walks within one hugetlb entry in the single call */
709static int pagemap_hugetlb_range(pte_t *pte, unsigned long hmask,
710 unsigned long addr, unsigned long end,
711 struct mm_walk *walk)
712{
713 struct pagemapread *pm = walk->private;
714 int err = 0;
715 u64 pfn;
716
717 for (; addr != end; addr += PAGE_SIZE) {
718 int offset = (addr & ~hmask) >> PAGE_SHIFT;
719 pfn = huge_pte_to_pagemap_entry(*pte, offset);
720 err = add_to_pagemap(addr, pfn, pm);
721 if (err)
722 return err;
723 }
724
725 cond_resched();
726
727 return err;
728}
729#endif /* HUGETLB_PAGE */
730
731/*
732 * /proc/pid/pagemap - an array mapping virtual pages to pfns
733 *
734 * For each page in the address space, this file contains one 64-bit entry
735 * consisting of the following:
736 *
737 * Bits 0-55 page frame number (PFN) if present
738 * Bits 0-4 swap type if swapped
739 * Bits 5-55 swap offset if swapped
740 * Bits 55-60 page shift (page size = 1<<page shift)
741 * Bit 61 reserved for future use
742 * Bit 62 page swapped
743 * Bit 63 page present
744 *
745 * If the page is not present but in swap, then the PFN contains an
746 * encoding of the swap file number and the page's offset into the
747 * swap. Unmapped pages return a null PFN. This allows determining
748 * precisely which pages are mapped (or in swap) and comparing mapped
749 * pages between processes.
750 *
751 * Efficient users of this interface will use /proc/pid/maps to
752 * determine which areas of memory are actually mapped and llseek to
753 * skip over unmapped regions.
754 */
755#define PAGEMAP_WALK_SIZE (PMD_SIZE)
756#define PAGEMAP_WALK_MASK (PMD_MASK)
757static ssize_t pagemap_read(struct file *file, char __user *buf,
758 size_t count, loff_t *ppos)
759{
760 struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
761 struct mm_struct *mm;
762 struct pagemapread pm;
763 int ret = -ESRCH;
764 struct mm_walk pagemap_walk = {};
765 unsigned long src;
766 unsigned long svpfn;
767 unsigned long start_vaddr;
768 unsigned long end_vaddr;
769 int copied = 0;
770
771 if (!task)
772 goto out;
773
774 ret = -EINVAL;
775 /* file position must be aligned */
776 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
777 goto out_task;
778
779 ret = 0;
780 if (!count)
781 goto out_task;
782
783 pm.len = PM_ENTRY_BYTES * (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
784 pm.buffer = kmalloc(pm.len, GFP_TEMPORARY);
785 ret = -ENOMEM;
786 if (!pm.buffer)
787 goto out_task;
788
789 mm = mm_for_maps(task);
790 ret = PTR_ERR(mm);
791 if (!mm || IS_ERR(mm))
792 goto out_free;
793
794 pagemap_walk.pmd_entry = pagemap_pte_range;
795 pagemap_walk.pte_hole = pagemap_pte_hole;
796#ifdef CONFIG_HUGETLB_PAGE
797 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
798#endif
799 pagemap_walk.mm = mm;
800 pagemap_walk.private = ±
801
802 src = *ppos;
803 svpfn = src / PM_ENTRY_BYTES;
804 start_vaddr = svpfn << PAGE_SHIFT;
805 end_vaddr = TASK_SIZE_OF(task);
806
807 /* watch out for wraparound */
808 if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
809 start_vaddr = end_vaddr;
810
811 /*
812 * The odds are that this will stop walking way
813 * before end_vaddr, because the length of the
814 * user buffer is tracked in "pm", and the walk
815 * will stop when we hit the end of the buffer.
816 */
817 ret = 0;
818 while (count && (start_vaddr < end_vaddr)) {
819 int len;
820 unsigned long end;
821
822 pm.pos = 0;
823 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
824 /* overflow ? */
825 if (end < start_vaddr || end > end_vaddr)
826 end = end_vaddr;
827 down_read(&mm->mmap_sem);
828 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
829 up_read(&mm->mmap_sem);
830 start_vaddr = end;
831
832 len = min(count, PM_ENTRY_BYTES * pm.pos);
833 if (copy_to_user(buf, pm.buffer, len)) {
834 ret = -EFAULT;
835 goto out_mm;
836 }
837 copied += len;
838 buf += len;
839 count -= len;
840 }
841 *ppos += copied;
842 if (!ret || ret == PM_END_OF_BUFFER)
843 ret = copied;
844
845out_mm:
846 mmput(mm);
847out_free:
848 kfree(pm.buffer);
849out_task:
850 put_task_struct(task);
851out:
852 return ret;
853}
854
855const struct file_operations proc_pagemap_operations = {
856 .llseek = mem_lseek, /* borrow this */
857 .read = pagemap_read,
858};
859#endif /* CONFIG_PROC_PAGE_MONITOR */
860
861#ifdef CONFIG_NUMA
862
863struct numa_maps {
864 struct vm_area_struct *vma;
865 unsigned long pages;
866 unsigned long anon;
867 unsigned long active;
868 unsigned long writeback;
869 unsigned long mapcount_max;
870 unsigned long dirty;
871 unsigned long swapcache;
872 unsigned long node[MAX_NUMNODES];
873};
874
875struct numa_maps_private {
876 struct proc_maps_private proc_maps;
877 struct numa_maps md;
878};
879
880static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
881 unsigned long nr_pages)
882{
883 int count = page_mapcount(page);
884
885 md->pages += nr_pages;
886 if (pte_dirty || PageDirty(page))
887 md->dirty += nr_pages;
888
889 if (PageSwapCache(page))
890 md->swapcache += nr_pages;
891
892 if (PageActive(page) || PageUnevictable(page))
893 md->active += nr_pages;
894
895 if (PageWriteback(page))
896 md->writeback += nr_pages;
897
898 if (PageAnon(page))
899 md->anon += nr_pages;
900
901 if (count > md->mapcount_max)
902 md->mapcount_max = count;
903
904 md->node[page_to_nid(page)] += nr_pages;
905}
906
907static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
908 unsigned long addr)
909{
910 struct page *page;
911 int nid;
912
913 if (!pte_present(pte))
914 return NULL;
915
916 page = vm_normal_page(vma, addr, pte);
917 if (!page)
918 return NULL;
919
920 if (PageReserved(page))
921 return NULL;
922
923 nid = page_to_nid(page);
924 if (!node_isset(nid, node_states[N_HIGH_MEMORY]))
925 return NULL;
926
927 return page;
928}
929
930static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
931 unsigned long end, struct mm_walk *walk)
932{
933 struct numa_maps *md;
934 spinlock_t *ptl;
935 pte_t *orig_pte;
936 pte_t *pte;
937
938 md = walk->private;
939 spin_lock(&walk->mm->page_table_lock);
940 if (pmd_trans_huge(*pmd)) {
941 if (pmd_trans_splitting(*pmd)) {
942 spin_unlock(&walk->mm->page_table_lock);
943 wait_split_huge_page(md->vma->anon_vma, pmd);
944 } else {
945 pte_t huge_pte = *(pte_t *)pmd;
946 struct page *page;
947
948 page = can_gather_numa_stats(huge_pte, md->vma, addr);
949 if (page)
950 gather_stats(page, md, pte_dirty(huge_pte),
951 HPAGE_PMD_SIZE/PAGE_SIZE);
952 spin_unlock(&walk->mm->page_table_lock);
953 return 0;
954 }
955 } else {
956 spin_unlock(&walk->mm->page_table_lock);
957 }
958
959 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
960 do {
961 struct page *page = can_gather_numa_stats(*pte, md->vma, addr);
962 if (!page)
963 continue;
964 gather_stats(page, md, pte_dirty(*pte), 1);
965
966 } while (pte++, addr += PAGE_SIZE, addr != end);
967 pte_unmap_unlock(orig_pte, ptl);
968 return 0;
969}
970#ifdef CONFIG_HUGETLB_PAGE
971static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
972 unsigned long addr, unsigned long end, struct mm_walk *walk)
973{
974 struct numa_maps *md;
975 struct page *page;
976
977 if (pte_none(*pte))
978 return 0;
979
980 page = pte_page(*pte);
981 if (!page)
982 return 0;
983
984 md = walk->private;
985 gather_stats(page, md, pte_dirty(*pte), 1);
986 return 0;
987}
988
989#else
990static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
991 unsigned long addr, unsigned long end, struct mm_walk *walk)
992{
993 return 0;
994}
995#endif
996
997/*
998 * Display pages allocated per node and memory policy via /proc.
999 */
1000static int show_numa_map(struct seq_file *m, void *v)
1001{
1002 struct numa_maps_private *numa_priv = m->private;
1003 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1004 struct vm_area_struct *vma = v;
1005 struct numa_maps *md = &numa_priv->md;
1006 struct file *file = vma->vm_file;
1007 struct mm_struct *mm = vma->vm_mm;
1008 struct mm_walk walk = {};
1009 struct mempolicy *pol;
1010 int n;
1011 char buffer[50];
1012
1013 if (!mm)
1014 return 0;
1015
1016 /* Ensure we start with an empty set of numa_maps statistics. */
1017 memset(md, 0, sizeof(*md));
1018
1019 md->vma = vma;
1020
1021 walk.hugetlb_entry = gather_hugetbl_stats;
1022 walk.pmd_entry = gather_pte_stats;
1023 walk.private = md;
1024 walk.mm = mm;
1025
1026 pol = get_vma_policy(proc_priv->task, vma, vma->vm_start);
1027 mpol_to_str(buffer, sizeof(buffer), pol, 0);
1028 mpol_cond_put(pol);
1029
1030 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1031
1032 if (file) {
1033 seq_printf(m, " file=");
1034 seq_path(m, &file->f_path, "\n\t= ");
1035 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1036 seq_printf(m, " heap");
1037 } else if (vma->vm_start <= mm->start_stack &&
1038 vma->vm_end >= mm->start_stack) {
1039 seq_printf(m, " stack");
1040 }
1041
1042 walk_page_range(vma->vm_start, vma->vm_end, &walk);
1043
1044 if (!md->pages)
1045 goto out;
1046
1047 if (md->anon)
1048 seq_printf(m, " anon=%lu", md->anon);
1049
1050 if (md->dirty)
1051 seq_printf(m, " dirty=%lu", md->dirty);
1052
1053 if (md->pages != md->anon && md->pages != md->dirty)
1054 seq_printf(m, " mapped=%lu", md->pages);
1055
1056 if (md->mapcount_max > 1)
1057 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1058
1059 if (md->swapcache)
1060 seq_printf(m, " swapcache=%lu", md->swapcache);
1061
1062 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1063 seq_printf(m, " active=%lu", md->active);
1064
1065 if (md->writeback)
1066 seq_printf(m, " writeback=%lu", md->writeback);
1067
1068 for_each_node_state(n, N_HIGH_MEMORY)
1069 if (md->node[n])
1070 seq_printf(m, " N%d=%lu", n, md->node[n]);
1071out:
1072 seq_putc(m, '\n');
1073
1074 if (m->count < m->size)
1075 m->version = (vma != proc_priv->tail_vma) ? vma->vm_start : 0;
1076 return 0;
1077}
1078
1079static const struct seq_operations proc_pid_numa_maps_op = {
1080 .start = m_start,
1081 .next = m_next,
1082 .stop = m_stop,
1083 .show = show_numa_map,
1084};
1085
1086static int numa_maps_open(struct inode *inode, struct file *file)
1087{
1088 struct numa_maps_private *priv;
1089 int ret = -ENOMEM;
1090 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
1091 if (priv) {
1092 priv->proc_maps.pid = proc_pid(inode);
1093 ret = seq_open(file, &proc_pid_numa_maps_op);
1094 if (!ret) {
1095 struct seq_file *m = file->private_data;
1096 m->private = priv;
1097 } else {
1098 kfree(priv);
1099 }
1100 }
1101 return ret;
1102}
1103
1104const struct file_operations proc_numa_maps_operations = {
1105 .open = numa_maps_open,
1106 .read = seq_read,
1107 .llseek = seq_lseek,
1108 .release = seq_release_private,
1109};
1110#endif /* CONFIG_NUMA */
1#include <linux/mm.h>
2#include <linux/vmacache.h>
3#include <linux/hugetlb.h>
4#include <linux/huge_mm.h>
5#include <linux/mount.h>
6#include <linux/seq_file.h>
7#include <linux/highmem.h>
8#include <linux/ptrace.h>
9#include <linux/slab.h>
10#include <linux/pagemap.h>
11#include <linux/mempolicy.h>
12#include <linux/rmap.h>
13#include <linux/swap.h>
14#include <linux/swapops.h>
15#include <linux/mmu_notifier.h>
16
17#include <asm/elf.h>
18#include <asm/uaccess.h>
19#include <asm/tlbflush.h>
20#include "internal.h"
21
22void task_mem(struct seq_file *m, struct mm_struct *mm)
23{
24 unsigned long data, text, lib, swap;
25 unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
26
27 /*
28 * Note: to minimize their overhead, mm maintains hiwater_vm and
29 * hiwater_rss only when about to *lower* total_vm or rss. Any
30 * collector of these hiwater stats must therefore get total_vm
31 * and rss too, which will usually be the higher. Barriers? not
32 * worth the effort, such snapshots can always be inconsistent.
33 */
34 hiwater_vm = total_vm = mm->total_vm;
35 if (hiwater_vm < mm->hiwater_vm)
36 hiwater_vm = mm->hiwater_vm;
37 hiwater_rss = total_rss = get_mm_rss(mm);
38 if (hiwater_rss < mm->hiwater_rss)
39 hiwater_rss = mm->hiwater_rss;
40
41 data = mm->total_vm - mm->shared_vm - mm->stack_vm;
42 text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
43 lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
44 swap = get_mm_counter(mm, MM_SWAPENTS);
45 seq_printf(m,
46 "VmPeak:\t%8lu kB\n"
47 "VmSize:\t%8lu kB\n"
48 "VmLck:\t%8lu kB\n"
49 "VmPin:\t%8lu kB\n"
50 "VmHWM:\t%8lu kB\n"
51 "VmRSS:\t%8lu kB\n"
52 "VmData:\t%8lu kB\n"
53 "VmStk:\t%8lu kB\n"
54 "VmExe:\t%8lu kB\n"
55 "VmLib:\t%8lu kB\n"
56 "VmPTE:\t%8lu kB\n"
57 "VmSwap:\t%8lu kB\n",
58 hiwater_vm << (PAGE_SHIFT-10),
59 total_vm << (PAGE_SHIFT-10),
60 mm->locked_vm << (PAGE_SHIFT-10),
61 mm->pinned_vm << (PAGE_SHIFT-10),
62 hiwater_rss << (PAGE_SHIFT-10),
63 total_rss << (PAGE_SHIFT-10),
64 data << (PAGE_SHIFT-10),
65 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
66 (PTRS_PER_PTE * sizeof(pte_t) *
67 atomic_long_read(&mm->nr_ptes)) >> 10,
68 swap << (PAGE_SHIFT-10));
69}
70
71unsigned long task_vsize(struct mm_struct *mm)
72{
73 return PAGE_SIZE * mm->total_vm;
74}
75
76unsigned long task_statm(struct mm_struct *mm,
77 unsigned long *shared, unsigned long *text,
78 unsigned long *data, unsigned long *resident)
79{
80 *shared = get_mm_counter(mm, MM_FILEPAGES);
81 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
82 >> PAGE_SHIFT;
83 *data = mm->total_vm - mm->shared_vm;
84 *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
85 return mm->total_vm;
86}
87
88#ifdef CONFIG_NUMA
89/*
90 * These functions are for numa_maps but called in generic **maps seq_file
91 * ->start(), ->stop() ops.
92 *
93 * numa_maps scans all vmas under mmap_sem and checks their mempolicy.
94 * Each mempolicy object is controlled by reference counting. The problem here
95 * is how to avoid accessing dead mempolicy object.
96 *
97 * Because we're holding mmap_sem while reading seq_file, it's safe to access
98 * each vma's mempolicy, no vma objects will never drop refs to mempolicy.
99 *
100 * A task's mempolicy (task->mempolicy) has different behavior. task->mempolicy
101 * is set and replaced under mmap_sem but unrefed and cleared under task_lock().
102 * So, without task_lock(), we cannot trust get_vma_policy() because we cannot
103 * gurantee the task never exits under us. But taking task_lock() around
104 * get_vma_plicy() causes lock order problem.
105 *
106 * To access task->mempolicy without lock, we hold a reference count of an
107 * object pointed by task->mempolicy and remember it. This will guarantee
108 * that task->mempolicy points to an alive object or NULL in numa_maps accesses.
109 */
110static void hold_task_mempolicy(struct proc_maps_private *priv)
111{
112 struct task_struct *task = priv->task;
113
114 task_lock(task);
115 priv->task_mempolicy = task->mempolicy;
116 mpol_get(priv->task_mempolicy);
117 task_unlock(task);
118}
119static void release_task_mempolicy(struct proc_maps_private *priv)
120{
121 mpol_put(priv->task_mempolicy);
122}
123#else
124static void hold_task_mempolicy(struct proc_maps_private *priv)
125{
126}
127static void release_task_mempolicy(struct proc_maps_private *priv)
128{
129}
130#endif
131
132static void vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma)
133{
134 if (vma && vma != priv->tail_vma) {
135 struct mm_struct *mm = vma->vm_mm;
136 release_task_mempolicy(priv);
137 up_read(&mm->mmap_sem);
138 mmput(mm);
139 }
140}
141
142static void *m_start(struct seq_file *m, loff_t *pos)
143{
144 struct proc_maps_private *priv = m->private;
145 unsigned long last_addr = m->version;
146 struct mm_struct *mm;
147 struct vm_area_struct *vma, *tail_vma = NULL;
148 loff_t l = *pos;
149
150 /* Clear the per syscall fields in priv */
151 priv->task = NULL;
152 priv->tail_vma = NULL;
153
154 /*
155 * We remember last_addr rather than next_addr to hit with
156 * vmacache most of the time. We have zero last_addr at
157 * the beginning and also after lseek. We will have -1 last_addr
158 * after the end of the vmas.
159 */
160
161 if (last_addr == -1UL)
162 return NULL;
163
164 priv->task = get_pid_task(priv->pid, PIDTYPE_PID);
165 if (!priv->task)
166 return ERR_PTR(-ESRCH);
167
168 mm = mm_access(priv->task, PTRACE_MODE_READ);
169 if (!mm || IS_ERR(mm))
170 return mm;
171 down_read(&mm->mmap_sem);
172
173 tail_vma = get_gate_vma(priv->task->mm);
174 priv->tail_vma = tail_vma;
175 hold_task_mempolicy(priv);
176 /* Start with last addr hint */
177 vma = find_vma(mm, last_addr);
178 if (last_addr && vma) {
179 vma = vma->vm_next;
180 goto out;
181 }
182
183 /*
184 * Check the vma index is within the range and do
185 * sequential scan until m_index.
186 */
187 vma = NULL;
188 if ((unsigned long)l < mm->map_count) {
189 vma = mm->mmap;
190 while (l-- && vma)
191 vma = vma->vm_next;
192 goto out;
193 }
194
195 if (l != mm->map_count)
196 tail_vma = NULL; /* After gate vma */
197
198out:
199 if (vma)
200 return vma;
201
202 release_task_mempolicy(priv);
203 /* End of vmas has been reached */
204 m->version = (tail_vma != NULL)? 0: -1UL;
205 up_read(&mm->mmap_sem);
206 mmput(mm);
207 return tail_vma;
208}
209
210static void *m_next(struct seq_file *m, void *v, loff_t *pos)
211{
212 struct proc_maps_private *priv = m->private;
213 struct vm_area_struct *vma = v;
214 struct vm_area_struct *tail_vma = priv->tail_vma;
215
216 (*pos)++;
217 if (vma && (vma != tail_vma) && vma->vm_next)
218 return vma->vm_next;
219 vma_stop(priv, vma);
220 return (vma != tail_vma)? tail_vma: NULL;
221}
222
223static void m_stop(struct seq_file *m, void *v)
224{
225 struct proc_maps_private *priv = m->private;
226 struct vm_area_struct *vma = v;
227
228 if (!IS_ERR(vma))
229 vma_stop(priv, vma);
230 if (priv->task)
231 put_task_struct(priv->task);
232}
233
234static int do_maps_open(struct inode *inode, struct file *file,
235 const struct seq_operations *ops)
236{
237 struct proc_maps_private *priv;
238 int ret = -ENOMEM;
239 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
240 if (priv) {
241 priv->pid = proc_pid(inode);
242 ret = seq_open(file, ops);
243 if (!ret) {
244 struct seq_file *m = file->private_data;
245 m->private = priv;
246 } else {
247 kfree(priv);
248 }
249 }
250 return ret;
251}
252
253static void
254show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
255{
256 struct mm_struct *mm = vma->vm_mm;
257 struct file *file = vma->vm_file;
258 struct proc_maps_private *priv = m->private;
259 struct task_struct *task = priv->task;
260 vm_flags_t flags = vma->vm_flags;
261 unsigned long ino = 0;
262 unsigned long long pgoff = 0;
263 unsigned long start, end;
264 dev_t dev = 0;
265 const char *name = NULL;
266
267 if (file) {
268 struct inode *inode = file_inode(vma->vm_file);
269 dev = inode->i_sb->s_dev;
270 ino = inode->i_ino;
271 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
272 }
273
274 /* We don't show the stack guard page in /proc/maps */
275 start = vma->vm_start;
276 if (stack_guard_page_start(vma, start))
277 start += PAGE_SIZE;
278 end = vma->vm_end;
279 if (stack_guard_page_end(vma, end))
280 end -= PAGE_SIZE;
281
282 seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
283 seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu ",
284 start,
285 end,
286 flags & VM_READ ? 'r' : '-',
287 flags & VM_WRITE ? 'w' : '-',
288 flags & VM_EXEC ? 'x' : '-',
289 flags & VM_MAYSHARE ? 's' : 'p',
290 pgoff,
291 MAJOR(dev), MINOR(dev), ino);
292
293 /*
294 * Print the dentry name for named mappings, and a
295 * special [heap] marker for the heap:
296 */
297 if (file) {
298 seq_pad(m, ' ');
299 seq_path(m, &file->f_path, "\n");
300 goto done;
301 }
302
303 name = arch_vma_name(vma);
304 if (!name) {
305 pid_t tid;
306
307 if (!mm) {
308 name = "[vdso]";
309 goto done;
310 }
311
312 if (vma->vm_start <= mm->brk &&
313 vma->vm_end >= mm->start_brk) {
314 name = "[heap]";
315 goto done;
316 }
317
318 tid = vm_is_stack(task, vma, is_pid);
319
320 if (tid != 0) {
321 /*
322 * Thread stack in /proc/PID/task/TID/maps or
323 * the main process stack.
324 */
325 if (!is_pid || (vma->vm_start <= mm->start_stack &&
326 vma->vm_end >= mm->start_stack)) {
327 name = "[stack]";
328 } else {
329 /* Thread stack in /proc/PID/maps */
330 seq_pad(m, ' ');
331 seq_printf(m, "[stack:%d]", tid);
332 }
333 }
334 }
335
336done:
337 if (name) {
338 seq_pad(m, ' ');
339 seq_puts(m, name);
340 }
341 seq_putc(m, '\n');
342}
343
344static int show_map(struct seq_file *m, void *v, int is_pid)
345{
346 struct vm_area_struct *vma = v;
347 struct proc_maps_private *priv = m->private;
348 struct task_struct *task = priv->task;
349
350 show_map_vma(m, vma, is_pid);
351
352 if (m->count < m->size) /* vma is copied successfully */
353 m->version = (vma != get_gate_vma(task->mm))
354 ? vma->vm_start : 0;
355 return 0;
356}
357
358static int show_pid_map(struct seq_file *m, void *v)
359{
360 return show_map(m, v, 1);
361}
362
363static int show_tid_map(struct seq_file *m, void *v)
364{
365 return show_map(m, v, 0);
366}
367
368static const struct seq_operations proc_pid_maps_op = {
369 .start = m_start,
370 .next = m_next,
371 .stop = m_stop,
372 .show = show_pid_map
373};
374
375static const struct seq_operations proc_tid_maps_op = {
376 .start = m_start,
377 .next = m_next,
378 .stop = m_stop,
379 .show = show_tid_map
380};
381
382static int pid_maps_open(struct inode *inode, struct file *file)
383{
384 return do_maps_open(inode, file, &proc_pid_maps_op);
385}
386
387static int tid_maps_open(struct inode *inode, struct file *file)
388{
389 return do_maps_open(inode, file, &proc_tid_maps_op);
390}
391
392const struct file_operations proc_pid_maps_operations = {
393 .open = pid_maps_open,
394 .read = seq_read,
395 .llseek = seq_lseek,
396 .release = seq_release_private,
397};
398
399const struct file_operations proc_tid_maps_operations = {
400 .open = tid_maps_open,
401 .read = seq_read,
402 .llseek = seq_lseek,
403 .release = seq_release_private,
404};
405
406/*
407 * Proportional Set Size(PSS): my share of RSS.
408 *
409 * PSS of a process is the count of pages it has in memory, where each
410 * page is divided by the number of processes sharing it. So if a
411 * process has 1000 pages all to itself, and 1000 shared with one other
412 * process, its PSS will be 1500.
413 *
414 * To keep (accumulated) division errors low, we adopt a 64bit
415 * fixed-point pss counter to minimize division errors. So (pss >>
416 * PSS_SHIFT) would be the real byte count.
417 *
418 * A shift of 12 before division means (assuming 4K page size):
419 * - 1M 3-user-pages add up to 8KB errors;
420 * - supports mapcount up to 2^24, or 16M;
421 * - supports PSS up to 2^52 bytes, or 4PB.
422 */
423#define PSS_SHIFT 12
424
425#ifdef CONFIG_PROC_PAGE_MONITOR
426struct mem_size_stats {
427 struct vm_area_struct *vma;
428 unsigned long resident;
429 unsigned long shared_clean;
430 unsigned long shared_dirty;
431 unsigned long private_clean;
432 unsigned long private_dirty;
433 unsigned long referenced;
434 unsigned long anonymous;
435 unsigned long anonymous_thp;
436 unsigned long swap;
437 unsigned long nonlinear;
438 u64 pss;
439};
440
441
442static void smaps_pte_entry(pte_t ptent, unsigned long addr,
443 unsigned long ptent_size, struct mm_walk *walk)
444{
445 struct mem_size_stats *mss = walk->private;
446 struct vm_area_struct *vma = mss->vma;
447 pgoff_t pgoff = linear_page_index(vma, addr);
448 struct page *page = NULL;
449 int mapcount;
450
451 if (pte_present(ptent)) {
452 page = vm_normal_page(vma, addr, ptent);
453 } else if (is_swap_pte(ptent)) {
454 swp_entry_t swpent = pte_to_swp_entry(ptent);
455
456 if (!non_swap_entry(swpent))
457 mss->swap += ptent_size;
458 else if (is_migration_entry(swpent))
459 page = migration_entry_to_page(swpent);
460 } else if (pte_file(ptent)) {
461 if (pte_to_pgoff(ptent) != pgoff)
462 mss->nonlinear += ptent_size;
463 }
464
465 if (!page)
466 return;
467
468 if (PageAnon(page))
469 mss->anonymous += ptent_size;
470
471 if (page->index != pgoff)
472 mss->nonlinear += ptent_size;
473
474 mss->resident += ptent_size;
475 /* Accumulate the size in pages that have been accessed. */
476 if (pte_young(ptent) || PageReferenced(page))
477 mss->referenced += ptent_size;
478 mapcount = page_mapcount(page);
479 if (mapcount >= 2) {
480 if (pte_dirty(ptent) || PageDirty(page))
481 mss->shared_dirty += ptent_size;
482 else
483 mss->shared_clean += ptent_size;
484 mss->pss += (ptent_size << PSS_SHIFT) / mapcount;
485 } else {
486 if (pte_dirty(ptent) || PageDirty(page))
487 mss->private_dirty += ptent_size;
488 else
489 mss->private_clean += ptent_size;
490 mss->pss += (ptent_size << PSS_SHIFT);
491 }
492}
493
494static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
495 struct mm_walk *walk)
496{
497 struct mem_size_stats *mss = walk->private;
498 struct vm_area_struct *vma = mss->vma;
499 pte_t *pte;
500 spinlock_t *ptl;
501
502 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
503 smaps_pte_entry(*(pte_t *)pmd, addr, HPAGE_PMD_SIZE, walk);
504 spin_unlock(ptl);
505 mss->anonymous_thp += HPAGE_PMD_SIZE;
506 return 0;
507 }
508
509 if (pmd_trans_unstable(pmd))
510 return 0;
511 /*
512 * The mmap_sem held all the way back in m_start() is what
513 * keeps khugepaged out of here and from collapsing things
514 * in here.
515 */
516 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
517 for (; addr != end; pte++, addr += PAGE_SIZE)
518 smaps_pte_entry(*pte, addr, PAGE_SIZE, walk);
519 pte_unmap_unlock(pte - 1, ptl);
520 cond_resched();
521 return 0;
522}
523
524static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
525{
526 /*
527 * Don't forget to update Documentation/ on changes.
528 */
529 static const char mnemonics[BITS_PER_LONG][2] = {
530 /*
531 * In case if we meet a flag we don't know about.
532 */
533 [0 ... (BITS_PER_LONG-1)] = "??",
534
535 [ilog2(VM_READ)] = "rd",
536 [ilog2(VM_WRITE)] = "wr",
537 [ilog2(VM_EXEC)] = "ex",
538 [ilog2(VM_SHARED)] = "sh",
539 [ilog2(VM_MAYREAD)] = "mr",
540 [ilog2(VM_MAYWRITE)] = "mw",
541 [ilog2(VM_MAYEXEC)] = "me",
542 [ilog2(VM_MAYSHARE)] = "ms",
543 [ilog2(VM_GROWSDOWN)] = "gd",
544 [ilog2(VM_PFNMAP)] = "pf",
545 [ilog2(VM_DENYWRITE)] = "dw",
546 [ilog2(VM_LOCKED)] = "lo",
547 [ilog2(VM_IO)] = "io",
548 [ilog2(VM_SEQ_READ)] = "sr",
549 [ilog2(VM_RAND_READ)] = "rr",
550 [ilog2(VM_DONTCOPY)] = "dc",
551 [ilog2(VM_DONTEXPAND)] = "de",
552 [ilog2(VM_ACCOUNT)] = "ac",
553 [ilog2(VM_NORESERVE)] = "nr",
554 [ilog2(VM_HUGETLB)] = "ht",
555 [ilog2(VM_NONLINEAR)] = "nl",
556 [ilog2(VM_ARCH_1)] = "ar",
557 [ilog2(VM_DONTDUMP)] = "dd",
558#ifdef CONFIG_MEM_SOFT_DIRTY
559 [ilog2(VM_SOFTDIRTY)] = "sd",
560#endif
561 [ilog2(VM_MIXEDMAP)] = "mm",
562 [ilog2(VM_HUGEPAGE)] = "hg",
563 [ilog2(VM_NOHUGEPAGE)] = "nh",
564 [ilog2(VM_MERGEABLE)] = "mg",
565 };
566 size_t i;
567
568 seq_puts(m, "VmFlags: ");
569 for (i = 0; i < BITS_PER_LONG; i++) {
570 if (vma->vm_flags & (1UL << i)) {
571 seq_printf(m, "%c%c ",
572 mnemonics[i][0], mnemonics[i][1]);
573 }
574 }
575 seq_putc(m, '\n');
576}
577
578static int show_smap(struct seq_file *m, void *v, int is_pid)
579{
580 struct proc_maps_private *priv = m->private;
581 struct task_struct *task = priv->task;
582 struct vm_area_struct *vma = v;
583 struct mem_size_stats mss;
584 struct mm_walk smaps_walk = {
585 .pmd_entry = smaps_pte_range,
586 .mm = vma->vm_mm,
587 .private = &mss,
588 };
589
590 memset(&mss, 0, sizeof mss);
591 mss.vma = vma;
592 /* mmap_sem is held in m_start */
593 if (vma->vm_mm && !is_vm_hugetlb_page(vma))
594 walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk);
595
596 show_map_vma(m, vma, is_pid);
597
598 seq_printf(m,
599 "Size: %8lu kB\n"
600 "Rss: %8lu kB\n"
601 "Pss: %8lu kB\n"
602 "Shared_Clean: %8lu kB\n"
603 "Shared_Dirty: %8lu kB\n"
604 "Private_Clean: %8lu kB\n"
605 "Private_Dirty: %8lu kB\n"
606 "Referenced: %8lu kB\n"
607 "Anonymous: %8lu kB\n"
608 "AnonHugePages: %8lu kB\n"
609 "Swap: %8lu kB\n"
610 "KernelPageSize: %8lu kB\n"
611 "MMUPageSize: %8lu kB\n"
612 "Locked: %8lu kB\n",
613 (vma->vm_end - vma->vm_start) >> 10,
614 mss.resident >> 10,
615 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
616 mss.shared_clean >> 10,
617 mss.shared_dirty >> 10,
618 mss.private_clean >> 10,
619 mss.private_dirty >> 10,
620 mss.referenced >> 10,
621 mss.anonymous >> 10,
622 mss.anonymous_thp >> 10,
623 mss.swap >> 10,
624 vma_kernel_pagesize(vma) >> 10,
625 vma_mmu_pagesize(vma) >> 10,
626 (vma->vm_flags & VM_LOCKED) ?
627 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
628
629 if (vma->vm_flags & VM_NONLINEAR)
630 seq_printf(m, "Nonlinear: %8lu kB\n",
631 mss.nonlinear >> 10);
632
633 show_smap_vma_flags(m, vma);
634
635 if (m->count < m->size) /* vma is copied successfully */
636 m->version = (vma != get_gate_vma(task->mm))
637 ? vma->vm_start : 0;
638 return 0;
639}
640
641static int show_pid_smap(struct seq_file *m, void *v)
642{
643 return show_smap(m, v, 1);
644}
645
646static int show_tid_smap(struct seq_file *m, void *v)
647{
648 return show_smap(m, v, 0);
649}
650
651static const struct seq_operations proc_pid_smaps_op = {
652 .start = m_start,
653 .next = m_next,
654 .stop = m_stop,
655 .show = show_pid_smap
656};
657
658static const struct seq_operations proc_tid_smaps_op = {
659 .start = m_start,
660 .next = m_next,
661 .stop = m_stop,
662 .show = show_tid_smap
663};
664
665static int pid_smaps_open(struct inode *inode, struct file *file)
666{
667 return do_maps_open(inode, file, &proc_pid_smaps_op);
668}
669
670static int tid_smaps_open(struct inode *inode, struct file *file)
671{
672 return do_maps_open(inode, file, &proc_tid_smaps_op);
673}
674
675const struct file_operations proc_pid_smaps_operations = {
676 .open = pid_smaps_open,
677 .read = seq_read,
678 .llseek = seq_lseek,
679 .release = seq_release_private,
680};
681
682const struct file_operations proc_tid_smaps_operations = {
683 .open = tid_smaps_open,
684 .read = seq_read,
685 .llseek = seq_lseek,
686 .release = seq_release_private,
687};
688
689/*
690 * We do not want to have constant page-shift bits sitting in
691 * pagemap entries and are about to reuse them some time soon.
692 *
693 * Here's the "migration strategy":
694 * 1. when the system boots these bits remain what they are,
695 * but a warning about future change is printed in log;
696 * 2. once anyone clears soft-dirty bits via clear_refs file,
697 * these flag is set to denote, that user is aware of the
698 * new API and those page-shift bits change their meaning.
699 * The respective warning is printed in dmesg;
700 * 3. In a couple of releases we will remove all the mentions
701 * of page-shift in pagemap entries.
702 */
703
704static bool soft_dirty_cleared __read_mostly;
705
706enum clear_refs_types {
707 CLEAR_REFS_ALL = 1,
708 CLEAR_REFS_ANON,
709 CLEAR_REFS_MAPPED,
710 CLEAR_REFS_SOFT_DIRTY,
711 CLEAR_REFS_LAST,
712};
713
714struct clear_refs_private {
715 struct vm_area_struct *vma;
716 enum clear_refs_types type;
717};
718
719static inline void clear_soft_dirty(struct vm_area_struct *vma,
720 unsigned long addr, pte_t *pte)
721{
722#ifdef CONFIG_MEM_SOFT_DIRTY
723 /*
724 * The soft-dirty tracker uses #PF-s to catch writes
725 * to pages, so write-protect the pte as well. See the
726 * Documentation/vm/soft-dirty.txt for full description
727 * of how soft-dirty works.
728 */
729 pte_t ptent = *pte;
730
731 if (pte_present(ptent)) {
732 ptent = pte_wrprotect(ptent);
733 ptent = pte_clear_flags(ptent, _PAGE_SOFT_DIRTY);
734 } else if (is_swap_pte(ptent)) {
735 ptent = pte_swp_clear_soft_dirty(ptent);
736 } else if (pte_file(ptent)) {
737 ptent = pte_file_clear_soft_dirty(ptent);
738 }
739
740 if (vma->vm_flags & VM_SOFTDIRTY)
741 vma->vm_flags &= ~VM_SOFTDIRTY;
742
743 set_pte_at(vma->vm_mm, addr, pte, ptent);
744#endif
745}
746
747static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
748 unsigned long end, struct mm_walk *walk)
749{
750 struct clear_refs_private *cp = walk->private;
751 struct vm_area_struct *vma = cp->vma;
752 pte_t *pte, ptent;
753 spinlock_t *ptl;
754 struct page *page;
755
756 split_huge_page_pmd(vma, addr, pmd);
757 if (pmd_trans_unstable(pmd))
758 return 0;
759
760 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
761 for (; addr != end; pte++, addr += PAGE_SIZE) {
762 ptent = *pte;
763
764 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
765 clear_soft_dirty(vma, addr, pte);
766 continue;
767 }
768
769 if (!pte_present(ptent))
770 continue;
771
772 page = vm_normal_page(vma, addr, ptent);
773 if (!page)
774 continue;
775
776 /* Clear accessed and referenced bits. */
777 ptep_test_and_clear_young(vma, addr, pte);
778 ClearPageReferenced(page);
779 }
780 pte_unmap_unlock(pte - 1, ptl);
781 cond_resched();
782 return 0;
783}
784
785static ssize_t clear_refs_write(struct file *file, const char __user *buf,
786 size_t count, loff_t *ppos)
787{
788 struct task_struct *task;
789 char buffer[PROC_NUMBUF];
790 struct mm_struct *mm;
791 struct vm_area_struct *vma;
792 enum clear_refs_types type;
793 int itype;
794 int rv;
795
796 memset(buffer, 0, sizeof(buffer));
797 if (count > sizeof(buffer) - 1)
798 count = sizeof(buffer) - 1;
799 if (copy_from_user(buffer, buf, count))
800 return -EFAULT;
801 rv = kstrtoint(strstrip(buffer), 10, &itype);
802 if (rv < 0)
803 return rv;
804 type = (enum clear_refs_types)itype;
805 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
806 return -EINVAL;
807
808 if (type == CLEAR_REFS_SOFT_DIRTY) {
809 soft_dirty_cleared = true;
810 pr_warn_once("The pagemap bits 55-60 has changed their meaning! "
811 "See the linux/Documentation/vm/pagemap.txt for details.\n");
812 }
813
814 task = get_proc_task(file_inode(file));
815 if (!task)
816 return -ESRCH;
817 mm = get_task_mm(task);
818 if (mm) {
819 struct clear_refs_private cp = {
820 .type = type,
821 };
822 struct mm_walk clear_refs_walk = {
823 .pmd_entry = clear_refs_pte_range,
824 .mm = mm,
825 .private = &cp,
826 };
827 down_read(&mm->mmap_sem);
828 if (type == CLEAR_REFS_SOFT_DIRTY)
829 mmu_notifier_invalidate_range_start(mm, 0, -1);
830 for (vma = mm->mmap; vma; vma = vma->vm_next) {
831 cp.vma = vma;
832 if (is_vm_hugetlb_page(vma))
833 continue;
834 /*
835 * Writing 1 to /proc/pid/clear_refs affects all pages.
836 *
837 * Writing 2 to /proc/pid/clear_refs only affects
838 * Anonymous pages.
839 *
840 * Writing 3 to /proc/pid/clear_refs only affects file
841 * mapped pages.
842 */
843 if (type == CLEAR_REFS_ANON && vma->vm_file)
844 continue;
845 if (type == CLEAR_REFS_MAPPED && !vma->vm_file)
846 continue;
847 walk_page_range(vma->vm_start, vma->vm_end,
848 &clear_refs_walk);
849 }
850 if (type == CLEAR_REFS_SOFT_DIRTY)
851 mmu_notifier_invalidate_range_end(mm, 0, -1);
852 flush_tlb_mm(mm);
853 up_read(&mm->mmap_sem);
854 mmput(mm);
855 }
856 put_task_struct(task);
857
858 return count;
859}
860
861const struct file_operations proc_clear_refs_operations = {
862 .write = clear_refs_write,
863 .llseek = noop_llseek,
864};
865
866typedef struct {
867 u64 pme;
868} pagemap_entry_t;
869
870struct pagemapread {
871 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */
872 pagemap_entry_t *buffer;
873 bool v2;
874};
875
876#define PAGEMAP_WALK_SIZE (PMD_SIZE)
877#define PAGEMAP_WALK_MASK (PMD_MASK)
878
879#define PM_ENTRY_BYTES sizeof(pagemap_entry_t)
880#define PM_STATUS_BITS 3
881#define PM_STATUS_OFFSET (64 - PM_STATUS_BITS)
882#define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
883#define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
884#define PM_PSHIFT_BITS 6
885#define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
886#define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
887#define __PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
888#define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1)
889#define PM_PFRAME(x) ((x) & PM_PFRAME_MASK)
890/* in "new" pagemap pshift bits are occupied with more status bits */
891#define PM_STATUS2(v2, x) (__PM_PSHIFT(v2 ? x : PAGE_SHIFT))
892
893#define __PM_SOFT_DIRTY (1LL)
894#define PM_PRESENT PM_STATUS(4LL)
895#define PM_SWAP PM_STATUS(2LL)
896#define PM_FILE PM_STATUS(1LL)
897#define PM_NOT_PRESENT(v2) PM_STATUS2(v2, 0)
898#define PM_END_OF_BUFFER 1
899
900static inline pagemap_entry_t make_pme(u64 val)
901{
902 return (pagemap_entry_t) { .pme = val };
903}
904
905static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
906 struct pagemapread *pm)
907{
908 pm->buffer[pm->pos++] = *pme;
909 if (pm->pos >= pm->len)
910 return PM_END_OF_BUFFER;
911 return 0;
912}
913
914static int pagemap_pte_hole(unsigned long start, unsigned long end,
915 struct mm_walk *walk)
916{
917 struct pagemapread *pm = walk->private;
918 unsigned long addr;
919 int err = 0;
920 pagemap_entry_t pme = make_pme(PM_NOT_PRESENT(pm->v2));
921
922 for (addr = start; addr < end; addr += PAGE_SIZE) {
923 err = add_to_pagemap(addr, &pme, pm);
924 if (err)
925 break;
926 }
927 return err;
928}
929
930static void pte_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm,
931 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
932{
933 u64 frame, flags;
934 struct page *page = NULL;
935 int flags2 = 0;
936
937 if (pte_present(pte)) {
938 frame = pte_pfn(pte);
939 flags = PM_PRESENT;
940 page = vm_normal_page(vma, addr, pte);
941 if (pte_soft_dirty(pte))
942 flags2 |= __PM_SOFT_DIRTY;
943 } else if (is_swap_pte(pte)) {
944 swp_entry_t entry;
945 if (pte_swp_soft_dirty(pte))
946 flags2 |= __PM_SOFT_DIRTY;
947 entry = pte_to_swp_entry(pte);
948 frame = swp_type(entry) |
949 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
950 flags = PM_SWAP;
951 if (is_migration_entry(entry))
952 page = migration_entry_to_page(entry);
953 } else {
954 if (vma->vm_flags & VM_SOFTDIRTY)
955 flags2 |= __PM_SOFT_DIRTY;
956 *pme = make_pme(PM_NOT_PRESENT(pm->v2) | PM_STATUS2(pm->v2, flags2));
957 return;
958 }
959
960 if (page && !PageAnon(page))
961 flags |= PM_FILE;
962 if ((vma->vm_flags & VM_SOFTDIRTY))
963 flags2 |= __PM_SOFT_DIRTY;
964
965 *pme = make_pme(PM_PFRAME(frame) | PM_STATUS2(pm->v2, flags2) | flags);
966}
967
968#ifdef CONFIG_TRANSPARENT_HUGEPAGE
969static void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm,
970 pmd_t pmd, int offset, int pmd_flags2)
971{
972 /*
973 * Currently pmd for thp is always present because thp can not be
974 * swapped-out, migrated, or HWPOISONed (split in such cases instead.)
975 * This if-check is just to prepare for future implementation.
976 */
977 if (pmd_present(pmd))
978 *pme = make_pme(PM_PFRAME(pmd_pfn(pmd) + offset)
979 | PM_STATUS2(pm->v2, pmd_flags2) | PM_PRESENT);
980 else
981 *pme = make_pme(PM_NOT_PRESENT(pm->v2) | PM_STATUS2(pm->v2, pmd_flags2));
982}
983#else
984static inline void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm,
985 pmd_t pmd, int offset, int pmd_flags2)
986{
987}
988#endif
989
990static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
991 struct mm_walk *walk)
992{
993 struct vm_area_struct *vma;
994 struct pagemapread *pm = walk->private;
995 spinlock_t *ptl;
996 pte_t *pte;
997 int err = 0;
998 pagemap_entry_t pme = make_pme(PM_NOT_PRESENT(pm->v2));
999
1000 /* find the first VMA at or above 'addr' */
1001 vma = find_vma(walk->mm, addr);
1002 if (vma && pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
1003 int pmd_flags2;
1004
1005 if ((vma->vm_flags & VM_SOFTDIRTY) || pmd_soft_dirty(*pmd))
1006 pmd_flags2 = __PM_SOFT_DIRTY;
1007 else
1008 pmd_flags2 = 0;
1009
1010 for (; addr != end; addr += PAGE_SIZE) {
1011 unsigned long offset;
1012
1013 offset = (addr & ~PAGEMAP_WALK_MASK) >>
1014 PAGE_SHIFT;
1015 thp_pmd_to_pagemap_entry(&pme, pm, *pmd, offset, pmd_flags2);
1016 err = add_to_pagemap(addr, &pme, pm);
1017 if (err)
1018 break;
1019 }
1020 spin_unlock(ptl);
1021 return err;
1022 }
1023
1024 if (pmd_trans_unstable(pmd))
1025 return 0;
1026 for (; addr != end; addr += PAGE_SIZE) {
1027 int flags2;
1028
1029 /* check to see if we've left 'vma' behind
1030 * and need a new, higher one */
1031 if (vma && (addr >= vma->vm_end)) {
1032 vma = find_vma(walk->mm, addr);
1033 if (vma && (vma->vm_flags & VM_SOFTDIRTY))
1034 flags2 = __PM_SOFT_DIRTY;
1035 else
1036 flags2 = 0;
1037 pme = make_pme(PM_NOT_PRESENT(pm->v2) | PM_STATUS2(pm->v2, flags2));
1038 }
1039
1040 /* check that 'vma' actually covers this address,
1041 * and that it isn't a huge page vma */
1042 if (vma && (vma->vm_start <= addr) &&
1043 !is_vm_hugetlb_page(vma)) {
1044 pte = pte_offset_map(pmd, addr);
1045 pte_to_pagemap_entry(&pme, pm, vma, addr, *pte);
1046 /* unmap before userspace copy */
1047 pte_unmap(pte);
1048 }
1049 err = add_to_pagemap(addr, &pme, pm);
1050 if (err)
1051 return err;
1052 }
1053
1054 cond_resched();
1055
1056 return err;
1057}
1058
1059#ifdef CONFIG_HUGETLB_PAGE
1060static void huge_pte_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm,
1061 pte_t pte, int offset, int flags2)
1062{
1063 if (pte_present(pte))
1064 *pme = make_pme(PM_PFRAME(pte_pfn(pte) + offset) |
1065 PM_STATUS2(pm->v2, flags2) |
1066 PM_PRESENT);
1067 else
1068 *pme = make_pme(PM_NOT_PRESENT(pm->v2) |
1069 PM_STATUS2(pm->v2, flags2));
1070}
1071
1072/* This function walks within one hugetlb entry in the single call */
1073static int pagemap_hugetlb_range(pte_t *pte, unsigned long hmask,
1074 unsigned long addr, unsigned long end,
1075 struct mm_walk *walk)
1076{
1077 struct pagemapread *pm = walk->private;
1078 struct vm_area_struct *vma;
1079 int err = 0;
1080 int flags2;
1081 pagemap_entry_t pme;
1082
1083 vma = find_vma(walk->mm, addr);
1084 WARN_ON_ONCE(!vma);
1085
1086 if (vma && (vma->vm_flags & VM_SOFTDIRTY))
1087 flags2 = __PM_SOFT_DIRTY;
1088 else
1089 flags2 = 0;
1090
1091 for (; addr != end; addr += PAGE_SIZE) {
1092 int offset = (addr & ~hmask) >> PAGE_SHIFT;
1093 huge_pte_to_pagemap_entry(&pme, pm, *pte, offset, flags2);
1094 err = add_to_pagemap(addr, &pme, pm);
1095 if (err)
1096 return err;
1097 }
1098
1099 cond_resched();
1100
1101 return err;
1102}
1103#endif /* HUGETLB_PAGE */
1104
1105/*
1106 * /proc/pid/pagemap - an array mapping virtual pages to pfns
1107 *
1108 * For each page in the address space, this file contains one 64-bit entry
1109 * consisting of the following:
1110 *
1111 * Bits 0-54 page frame number (PFN) if present
1112 * Bits 0-4 swap type if swapped
1113 * Bits 5-54 swap offset if swapped
1114 * Bits 55-60 page shift (page size = 1<<page shift)
1115 * Bit 61 page is file-page or shared-anon
1116 * Bit 62 page swapped
1117 * Bit 63 page present
1118 *
1119 * If the page is not present but in swap, then the PFN contains an
1120 * encoding of the swap file number and the page's offset into the
1121 * swap. Unmapped pages return a null PFN. This allows determining
1122 * precisely which pages are mapped (or in swap) and comparing mapped
1123 * pages between processes.
1124 *
1125 * Efficient users of this interface will use /proc/pid/maps to
1126 * determine which areas of memory are actually mapped and llseek to
1127 * skip over unmapped regions.
1128 */
1129static ssize_t pagemap_read(struct file *file, char __user *buf,
1130 size_t count, loff_t *ppos)
1131{
1132 struct task_struct *task = get_proc_task(file_inode(file));
1133 struct mm_struct *mm;
1134 struct pagemapread pm;
1135 int ret = -ESRCH;
1136 struct mm_walk pagemap_walk = {};
1137 unsigned long src;
1138 unsigned long svpfn;
1139 unsigned long start_vaddr;
1140 unsigned long end_vaddr;
1141 int copied = 0;
1142
1143 if (!task)
1144 goto out;
1145
1146 ret = -EINVAL;
1147 /* file position must be aligned */
1148 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1149 goto out_task;
1150
1151 ret = 0;
1152 if (!count)
1153 goto out_task;
1154
1155 pm.v2 = soft_dirty_cleared;
1156 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1157 pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY);
1158 ret = -ENOMEM;
1159 if (!pm.buffer)
1160 goto out_task;
1161
1162 mm = mm_access(task, PTRACE_MODE_READ);
1163 ret = PTR_ERR(mm);
1164 if (!mm || IS_ERR(mm))
1165 goto out_free;
1166
1167 pagemap_walk.pmd_entry = pagemap_pte_range;
1168 pagemap_walk.pte_hole = pagemap_pte_hole;
1169#ifdef CONFIG_HUGETLB_PAGE
1170 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1171#endif
1172 pagemap_walk.mm = mm;
1173 pagemap_walk.private = ±
1174
1175 src = *ppos;
1176 svpfn = src / PM_ENTRY_BYTES;
1177 start_vaddr = svpfn << PAGE_SHIFT;
1178 end_vaddr = TASK_SIZE_OF(task);
1179
1180 /* watch out for wraparound */
1181 if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
1182 start_vaddr = end_vaddr;
1183
1184 /*
1185 * The odds are that this will stop walking way
1186 * before end_vaddr, because the length of the
1187 * user buffer is tracked in "pm", and the walk
1188 * will stop when we hit the end of the buffer.
1189 */
1190 ret = 0;
1191 while (count && (start_vaddr < end_vaddr)) {
1192 int len;
1193 unsigned long end;
1194
1195 pm.pos = 0;
1196 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1197 /* overflow ? */
1198 if (end < start_vaddr || end > end_vaddr)
1199 end = end_vaddr;
1200 down_read(&mm->mmap_sem);
1201 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1202 up_read(&mm->mmap_sem);
1203 start_vaddr = end;
1204
1205 len = min(count, PM_ENTRY_BYTES * pm.pos);
1206 if (copy_to_user(buf, pm.buffer, len)) {
1207 ret = -EFAULT;
1208 goto out_mm;
1209 }
1210 copied += len;
1211 buf += len;
1212 count -= len;
1213 }
1214 *ppos += copied;
1215 if (!ret || ret == PM_END_OF_BUFFER)
1216 ret = copied;
1217
1218out_mm:
1219 mmput(mm);
1220out_free:
1221 kfree(pm.buffer);
1222out_task:
1223 put_task_struct(task);
1224out:
1225 return ret;
1226}
1227
1228static int pagemap_open(struct inode *inode, struct file *file)
1229{
1230 pr_warn_once("Bits 55-60 of /proc/PID/pagemap entries are about "
1231 "to stop being page-shift some time soon. See the "
1232 "linux/Documentation/vm/pagemap.txt for details.\n");
1233 return 0;
1234}
1235
1236const struct file_operations proc_pagemap_operations = {
1237 .llseek = mem_lseek, /* borrow this */
1238 .read = pagemap_read,
1239 .open = pagemap_open,
1240};
1241#endif /* CONFIG_PROC_PAGE_MONITOR */
1242
1243#ifdef CONFIG_NUMA
1244
1245struct numa_maps {
1246 struct vm_area_struct *vma;
1247 unsigned long pages;
1248 unsigned long anon;
1249 unsigned long active;
1250 unsigned long writeback;
1251 unsigned long mapcount_max;
1252 unsigned long dirty;
1253 unsigned long swapcache;
1254 unsigned long node[MAX_NUMNODES];
1255};
1256
1257struct numa_maps_private {
1258 struct proc_maps_private proc_maps;
1259 struct numa_maps md;
1260};
1261
1262static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1263 unsigned long nr_pages)
1264{
1265 int count = page_mapcount(page);
1266
1267 md->pages += nr_pages;
1268 if (pte_dirty || PageDirty(page))
1269 md->dirty += nr_pages;
1270
1271 if (PageSwapCache(page))
1272 md->swapcache += nr_pages;
1273
1274 if (PageActive(page) || PageUnevictable(page))
1275 md->active += nr_pages;
1276
1277 if (PageWriteback(page))
1278 md->writeback += nr_pages;
1279
1280 if (PageAnon(page))
1281 md->anon += nr_pages;
1282
1283 if (count > md->mapcount_max)
1284 md->mapcount_max = count;
1285
1286 md->node[page_to_nid(page)] += nr_pages;
1287}
1288
1289static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1290 unsigned long addr)
1291{
1292 struct page *page;
1293 int nid;
1294
1295 if (!pte_present(pte))
1296 return NULL;
1297
1298 page = vm_normal_page(vma, addr, pte);
1299 if (!page)
1300 return NULL;
1301
1302 if (PageReserved(page))
1303 return NULL;
1304
1305 nid = page_to_nid(page);
1306 if (!node_isset(nid, node_states[N_MEMORY]))
1307 return NULL;
1308
1309 return page;
1310}
1311
1312static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1313 unsigned long end, struct mm_walk *walk)
1314{
1315 struct numa_maps *md;
1316 spinlock_t *ptl;
1317 pte_t *orig_pte;
1318 pte_t *pte;
1319
1320 md = walk->private;
1321
1322 if (pmd_trans_huge_lock(pmd, md->vma, &ptl) == 1) {
1323 pte_t huge_pte = *(pte_t *)pmd;
1324 struct page *page;
1325
1326 page = can_gather_numa_stats(huge_pte, md->vma, addr);
1327 if (page)
1328 gather_stats(page, md, pte_dirty(huge_pte),
1329 HPAGE_PMD_SIZE/PAGE_SIZE);
1330 spin_unlock(ptl);
1331 return 0;
1332 }
1333
1334 if (pmd_trans_unstable(pmd))
1335 return 0;
1336 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1337 do {
1338 struct page *page = can_gather_numa_stats(*pte, md->vma, addr);
1339 if (!page)
1340 continue;
1341 gather_stats(page, md, pte_dirty(*pte), 1);
1342
1343 } while (pte++, addr += PAGE_SIZE, addr != end);
1344 pte_unmap_unlock(orig_pte, ptl);
1345 return 0;
1346}
1347#ifdef CONFIG_HUGETLB_PAGE
1348static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
1349 unsigned long addr, unsigned long end, struct mm_walk *walk)
1350{
1351 struct numa_maps *md;
1352 struct page *page;
1353
1354 if (!pte_present(*pte))
1355 return 0;
1356
1357 page = pte_page(*pte);
1358 if (!page)
1359 return 0;
1360
1361 md = walk->private;
1362 gather_stats(page, md, pte_dirty(*pte), 1);
1363 return 0;
1364}
1365
1366#else
1367static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
1368 unsigned long addr, unsigned long end, struct mm_walk *walk)
1369{
1370 return 0;
1371}
1372#endif
1373
1374/*
1375 * Display pages allocated per node and memory policy via /proc.
1376 */
1377static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1378{
1379 struct numa_maps_private *numa_priv = m->private;
1380 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1381 struct vm_area_struct *vma = v;
1382 struct numa_maps *md = &numa_priv->md;
1383 struct file *file = vma->vm_file;
1384 struct task_struct *task = proc_priv->task;
1385 struct mm_struct *mm = vma->vm_mm;
1386 struct mm_walk walk = {};
1387 struct mempolicy *pol;
1388 char buffer[64];
1389 int nid;
1390
1391 if (!mm)
1392 return 0;
1393
1394 /* Ensure we start with an empty set of numa_maps statistics. */
1395 memset(md, 0, sizeof(*md));
1396
1397 md->vma = vma;
1398
1399 walk.hugetlb_entry = gather_hugetbl_stats;
1400 walk.pmd_entry = gather_pte_stats;
1401 walk.private = md;
1402 walk.mm = mm;
1403
1404 pol = get_vma_policy(task, vma, vma->vm_start);
1405 mpol_to_str(buffer, sizeof(buffer), pol);
1406 mpol_cond_put(pol);
1407
1408 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1409
1410 if (file) {
1411 seq_printf(m, " file=");
1412 seq_path(m, &file->f_path, "\n\t= ");
1413 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1414 seq_printf(m, " heap");
1415 } else {
1416 pid_t tid = vm_is_stack(task, vma, is_pid);
1417 if (tid != 0) {
1418 /*
1419 * Thread stack in /proc/PID/task/TID/maps or
1420 * the main process stack.
1421 */
1422 if (!is_pid || (vma->vm_start <= mm->start_stack &&
1423 vma->vm_end >= mm->start_stack))
1424 seq_printf(m, " stack");
1425 else
1426 seq_printf(m, " stack:%d", tid);
1427 }
1428 }
1429
1430 if (is_vm_hugetlb_page(vma))
1431 seq_printf(m, " huge");
1432
1433 walk_page_range(vma->vm_start, vma->vm_end, &walk);
1434
1435 if (!md->pages)
1436 goto out;
1437
1438 if (md->anon)
1439 seq_printf(m, " anon=%lu", md->anon);
1440
1441 if (md->dirty)
1442 seq_printf(m, " dirty=%lu", md->dirty);
1443
1444 if (md->pages != md->anon && md->pages != md->dirty)
1445 seq_printf(m, " mapped=%lu", md->pages);
1446
1447 if (md->mapcount_max > 1)
1448 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1449
1450 if (md->swapcache)
1451 seq_printf(m, " swapcache=%lu", md->swapcache);
1452
1453 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1454 seq_printf(m, " active=%lu", md->active);
1455
1456 if (md->writeback)
1457 seq_printf(m, " writeback=%lu", md->writeback);
1458
1459 for_each_node_state(nid, N_MEMORY)
1460 if (md->node[nid])
1461 seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1462out:
1463 seq_putc(m, '\n');
1464
1465 if (m->count < m->size)
1466 m->version = (vma != proc_priv->tail_vma) ? vma->vm_start : 0;
1467 return 0;
1468}
1469
1470static int show_pid_numa_map(struct seq_file *m, void *v)
1471{
1472 return show_numa_map(m, v, 1);
1473}
1474
1475static int show_tid_numa_map(struct seq_file *m, void *v)
1476{
1477 return show_numa_map(m, v, 0);
1478}
1479
1480static const struct seq_operations proc_pid_numa_maps_op = {
1481 .start = m_start,
1482 .next = m_next,
1483 .stop = m_stop,
1484 .show = show_pid_numa_map,
1485};
1486
1487static const struct seq_operations proc_tid_numa_maps_op = {
1488 .start = m_start,
1489 .next = m_next,
1490 .stop = m_stop,
1491 .show = show_tid_numa_map,
1492};
1493
1494static int numa_maps_open(struct inode *inode, struct file *file,
1495 const struct seq_operations *ops)
1496{
1497 struct numa_maps_private *priv;
1498 int ret = -ENOMEM;
1499 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
1500 if (priv) {
1501 priv->proc_maps.pid = proc_pid(inode);
1502 ret = seq_open(file, ops);
1503 if (!ret) {
1504 struct seq_file *m = file->private_data;
1505 m->private = priv;
1506 } else {
1507 kfree(priv);
1508 }
1509 }
1510 return ret;
1511}
1512
1513static int pid_numa_maps_open(struct inode *inode, struct file *file)
1514{
1515 return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1516}
1517
1518static int tid_numa_maps_open(struct inode *inode, struct file *file)
1519{
1520 return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1521}
1522
1523const struct file_operations proc_pid_numa_maps_operations = {
1524 .open = pid_numa_maps_open,
1525 .read = seq_read,
1526 .llseek = seq_lseek,
1527 .release = seq_release_private,
1528};
1529
1530const struct file_operations proc_tid_numa_maps_operations = {
1531 .open = tid_numa_maps_open,
1532 .read = seq_read,
1533 .llseek = seq_lseek,
1534 .release = seq_release_private,
1535};
1536#endif /* CONFIG_NUMA */