Loading...
1// SPDX-License-Identifier: GPL-2.0
2#include <linux/mm.h>
3#include <linux/vmacache.h>
4#include <linux/hugetlb.h>
5#include <linux/huge_mm.h>
6#include <linux/mount.h>
7#include <linux/seq_file.h>
8#include <linux/highmem.h>
9#include <linux/ptrace.h>
10#include <linux/slab.h>
11#include <linux/pagemap.h>
12#include <linux/mempolicy.h>
13#include <linux/rmap.h>
14#include <linux/swap.h>
15#include <linux/sched/mm.h>
16#include <linux/swapops.h>
17#include <linux/mmu_notifier.h>
18#include <linux/page_idle.h>
19#include <linux/shmem_fs.h>
20#include <linux/uaccess.h>
21
22#include <asm/elf.h>
23#include <asm/tlb.h>
24#include <asm/tlbflush.h>
25#include "internal.h"
26
27#define SEQ_PUT_DEC(str, val) \
28 seq_put_decimal_ull_width(m, str, (val) << (PAGE_SHIFT-10), 8)
29void task_mem(struct seq_file *m, struct mm_struct *mm)
30{
31 unsigned long text, lib, swap, anon, file, shmem;
32 unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
33
34 anon = get_mm_counter(mm, MM_ANONPAGES);
35 file = get_mm_counter(mm, MM_FILEPAGES);
36 shmem = get_mm_counter(mm, MM_SHMEMPAGES);
37
38 /*
39 * Note: to minimize their overhead, mm maintains hiwater_vm and
40 * hiwater_rss only when about to *lower* total_vm or rss. Any
41 * collector of these hiwater stats must therefore get total_vm
42 * and rss too, which will usually be the higher. Barriers? not
43 * worth the effort, such snapshots can always be inconsistent.
44 */
45 hiwater_vm = total_vm = mm->total_vm;
46 if (hiwater_vm < mm->hiwater_vm)
47 hiwater_vm = mm->hiwater_vm;
48 hiwater_rss = total_rss = anon + file + shmem;
49 if (hiwater_rss < mm->hiwater_rss)
50 hiwater_rss = mm->hiwater_rss;
51
52 /* split executable areas between text and lib */
53 text = PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK);
54 text = min(text, mm->exec_vm << PAGE_SHIFT);
55 lib = (mm->exec_vm << PAGE_SHIFT) - text;
56
57 swap = get_mm_counter(mm, MM_SWAPENTS);
58 SEQ_PUT_DEC("VmPeak:\t", hiwater_vm);
59 SEQ_PUT_DEC(" kB\nVmSize:\t", total_vm);
60 SEQ_PUT_DEC(" kB\nVmLck:\t", mm->locked_vm);
61 SEQ_PUT_DEC(" kB\nVmPin:\t", mm->pinned_vm);
62 SEQ_PUT_DEC(" kB\nVmHWM:\t", hiwater_rss);
63 SEQ_PUT_DEC(" kB\nVmRSS:\t", total_rss);
64 SEQ_PUT_DEC(" kB\nRssAnon:\t", anon);
65 SEQ_PUT_DEC(" kB\nRssFile:\t", file);
66 SEQ_PUT_DEC(" kB\nRssShmem:\t", shmem);
67 SEQ_PUT_DEC(" kB\nVmData:\t", mm->data_vm);
68 SEQ_PUT_DEC(" kB\nVmStk:\t", mm->stack_vm);
69 seq_put_decimal_ull_width(m,
70 " kB\nVmExe:\t", text >> 10, 8);
71 seq_put_decimal_ull_width(m,
72 " kB\nVmLib:\t", lib >> 10, 8);
73 seq_put_decimal_ull_width(m,
74 " kB\nVmPTE:\t", mm_pgtables_bytes(mm) >> 10, 8);
75 SEQ_PUT_DEC(" kB\nVmSwap:\t", swap);
76 seq_puts(m, " kB\n");
77 hugetlb_report_usage(m, mm);
78}
79#undef SEQ_PUT_DEC
80
81unsigned long task_vsize(struct mm_struct *mm)
82{
83 return PAGE_SIZE * mm->total_vm;
84}
85
86unsigned long task_statm(struct mm_struct *mm,
87 unsigned long *shared, unsigned long *text,
88 unsigned long *data, unsigned long *resident)
89{
90 *shared = get_mm_counter(mm, MM_FILEPAGES) +
91 get_mm_counter(mm, MM_SHMEMPAGES);
92 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
93 >> PAGE_SHIFT;
94 *data = mm->data_vm + mm->stack_vm;
95 *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
96 return mm->total_vm;
97}
98
99#ifdef CONFIG_NUMA
100/*
101 * Save get_task_policy() for show_numa_map().
102 */
103static void hold_task_mempolicy(struct proc_maps_private *priv)
104{
105 struct task_struct *task = priv->task;
106
107 task_lock(task);
108 priv->task_mempolicy = get_task_policy(task);
109 mpol_get(priv->task_mempolicy);
110 task_unlock(task);
111}
112static void release_task_mempolicy(struct proc_maps_private *priv)
113{
114 mpol_put(priv->task_mempolicy);
115}
116#else
117static void hold_task_mempolicy(struct proc_maps_private *priv)
118{
119}
120static void release_task_mempolicy(struct proc_maps_private *priv)
121{
122}
123#endif
124
125static void vma_stop(struct proc_maps_private *priv)
126{
127 struct mm_struct *mm = priv->mm;
128
129 release_task_mempolicy(priv);
130 up_read(&mm->mmap_sem);
131 mmput(mm);
132}
133
134static struct vm_area_struct *
135m_next_vma(struct proc_maps_private *priv, struct vm_area_struct *vma)
136{
137 if (vma == priv->tail_vma)
138 return NULL;
139 return vma->vm_next ?: priv->tail_vma;
140}
141
142static void m_cache_vma(struct seq_file *m, struct vm_area_struct *vma)
143{
144 if (m->count < m->size) /* vma is copied successfully */
145 m->version = m_next_vma(m->private, vma) ? vma->vm_end : -1UL;
146}
147
148static void *m_start(struct seq_file *m, loff_t *ppos)
149{
150 struct proc_maps_private *priv = m->private;
151 unsigned long last_addr = m->version;
152 struct mm_struct *mm;
153 struct vm_area_struct *vma;
154 unsigned int pos = *ppos;
155
156 /* See m_cache_vma(). Zero at the start or after lseek. */
157 if (last_addr == -1UL)
158 return NULL;
159
160 priv->task = get_proc_task(priv->inode);
161 if (!priv->task)
162 return ERR_PTR(-ESRCH);
163
164 mm = priv->mm;
165 if (!mm || !mmget_not_zero(mm))
166 return NULL;
167
168 down_read(&mm->mmap_sem);
169 hold_task_mempolicy(priv);
170 priv->tail_vma = get_gate_vma(mm);
171
172 if (last_addr) {
173 vma = find_vma(mm, last_addr - 1);
174 if (vma && vma->vm_start <= last_addr)
175 vma = m_next_vma(priv, vma);
176 if (vma)
177 return vma;
178 }
179
180 m->version = 0;
181 if (pos < mm->map_count) {
182 for (vma = mm->mmap; pos; pos--) {
183 m->version = vma->vm_start;
184 vma = vma->vm_next;
185 }
186 return vma;
187 }
188
189 /* we do not bother to update m->version in this case */
190 if (pos == mm->map_count && priv->tail_vma)
191 return priv->tail_vma;
192
193 vma_stop(priv);
194 return NULL;
195}
196
197static void *m_next(struct seq_file *m, void *v, loff_t *pos)
198{
199 struct proc_maps_private *priv = m->private;
200 struct vm_area_struct *next;
201
202 (*pos)++;
203 next = m_next_vma(priv, v);
204 if (!next)
205 vma_stop(priv);
206 return next;
207}
208
209static void m_stop(struct seq_file *m, void *v)
210{
211 struct proc_maps_private *priv = m->private;
212
213 if (!IS_ERR_OR_NULL(v))
214 vma_stop(priv);
215 if (priv->task) {
216 put_task_struct(priv->task);
217 priv->task = NULL;
218 }
219}
220
221static int proc_maps_open(struct inode *inode, struct file *file,
222 const struct seq_operations *ops, int psize)
223{
224 struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
225
226 if (!priv)
227 return -ENOMEM;
228
229 priv->inode = inode;
230 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
231 if (IS_ERR(priv->mm)) {
232 int err = PTR_ERR(priv->mm);
233
234 seq_release_private(inode, file);
235 return err;
236 }
237
238 return 0;
239}
240
241static int proc_map_release(struct inode *inode, struct file *file)
242{
243 struct seq_file *seq = file->private_data;
244 struct proc_maps_private *priv = seq->private;
245
246 if (priv->mm)
247 mmdrop(priv->mm);
248
249 kfree(priv->rollup);
250 return seq_release_private(inode, file);
251}
252
253static int do_maps_open(struct inode *inode, struct file *file,
254 const struct seq_operations *ops)
255{
256 return proc_maps_open(inode, file, ops,
257 sizeof(struct proc_maps_private));
258}
259
260/*
261 * Indicate if the VMA is a stack for the given task; for
262 * /proc/PID/maps that is the stack of the main task.
263 */
264static int is_stack(struct vm_area_struct *vma)
265{
266 /*
267 * We make no effort to guess what a given thread considers to be
268 * its "stack". It's not even well-defined for programs written
269 * languages like Go.
270 */
271 return vma->vm_start <= vma->vm_mm->start_stack &&
272 vma->vm_end >= vma->vm_mm->start_stack;
273}
274
275static void show_vma_header_prefix(struct seq_file *m,
276 unsigned long start, unsigned long end,
277 vm_flags_t flags, unsigned long long pgoff,
278 dev_t dev, unsigned long ino)
279{
280 seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
281 seq_put_hex_ll(m, NULL, start, 8);
282 seq_put_hex_ll(m, "-", end, 8);
283 seq_putc(m, ' ');
284 seq_putc(m, flags & VM_READ ? 'r' : '-');
285 seq_putc(m, flags & VM_WRITE ? 'w' : '-');
286 seq_putc(m, flags & VM_EXEC ? 'x' : '-');
287 seq_putc(m, flags & VM_MAYSHARE ? 's' : 'p');
288 seq_put_hex_ll(m, " ", pgoff, 8);
289 seq_put_hex_ll(m, " ", MAJOR(dev), 2);
290 seq_put_hex_ll(m, ":", MINOR(dev), 2);
291 seq_put_decimal_ull(m, " ", ino);
292 seq_putc(m, ' ');
293}
294
295static void
296show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
297{
298 struct mm_struct *mm = vma->vm_mm;
299 struct file *file = vma->vm_file;
300 vm_flags_t flags = vma->vm_flags;
301 unsigned long ino = 0;
302 unsigned long long pgoff = 0;
303 unsigned long start, end;
304 dev_t dev = 0;
305 const char *name = NULL;
306
307 if (file) {
308 struct inode *inode = file_inode(vma->vm_file);
309 dev = inode->i_sb->s_dev;
310 ino = inode->i_ino;
311 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
312 }
313
314 start = vma->vm_start;
315 end = vma->vm_end;
316 show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino);
317
318 /*
319 * Print the dentry name for named mappings, and a
320 * special [heap] marker for the heap:
321 */
322 if (file) {
323 seq_pad(m, ' ');
324 seq_file_path(m, file, "\n");
325 goto done;
326 }
327
328 if (vma->vm_ops && vma->vm_ops->name) {
329 name = vma->vm_ops->name(vma);
330 if (name)
331 goto done;
332 }
333
334 name = arch_vma_name(vma);
335 if (!name) {
336 if (!mm) {
337 name = "[vdso]";
338 goto done;
339 }
340
341 if (vma->vm_start <= mm->brk &&
342 vma->vm_end >= mm->start_brk) {
343 name = "[heap]";
344 goto done;
345 }
346
347 if (is_stack(vma))
348 name = "[stack]";
349 }
350
351done:
352 if (name) {
353 seq_pad(m, ' ');
354 seq_puts(m, name);
355 }
356 seq_putc(m, '\n');
357}
358
359static int show_map(struct seq_file *m, void *v, int is_pid)
360{
361 show_map_vma(m, v, is_pid);
362 m_cache_vma(m, v);
363 return 0;
364}
365
366static int show_pid_map(struct seq_file *m, void *v)
367{
368 return show_map(m, v, 1);
369}
370
371static int show_tid_map(struct seq_file *m, void *v)
372{
373 return show_map(m, v, 0);
374}
375
376static const struct seq_operations proc_pid_maps_op = {
377 .start = m_start,
378 .next = m_next,
379 .stop = m_stop,
380 .show = show_pid_map
381};
382
383static const struct seq_operations proc_tid_maps_op = {
384 .start = m_start,
385 .next = m_next,
386 .stop = m_stop,
387 .show = show_tid_map
388};
389
390static int pid_maps_open(struct inode *inode, struct file *file)
391{
392 return do_maps_open(inode, file, &proc_pid_maps_op);
393}
394
395static int tid_maps_open(struct inode *inode, struct file *file)
396{
397 return do_maps_open(inode, file, &proc_tid_maps_op);
398}
399
400const struct file_operations proc_pid_maps_operations = {
401 .open = pid_maps_open,
402 .read = seq_read,
403 .llseek = seq_lseek,
404 .release = proc_map_release,
405};
406
407const struct file_operations proc_tid_maps_operations = {
408 .open = tid_maps_open,
409 .read = seq_read,
410 .llseek = seq_lseek,
411 .release = proc_map_release,
412};
413
414/*
415 * Proportional Set Size(PSS): my share of RSS.
416 *
417 * PSS of a process is the count of pages it has in memory, where each
418 * page is divided by the number of processes sharing it. So if a
419 * process has 1000 pages all to itself, and 1000 shared with one other
420 * process, its PSS will be 1500.
421 *
422 * To keep (accumulated) division errors low, we adopt a 64bit
423 * fixed-point pss counter to minimize division errors. So (pss >>
424 * PSS_SHIFT) would be the real byte count.
425 *
426 * A shift of 12 before division means (assuming 4K page size):
427 * - 1M 3-user-pages add up to 8KB errors;
428 * - supports mapcount up to 2^24, or 16M;
429 * - supports PSS up to 2^52 bytes, or 4PB.
430 */
431#define PSS_SHIFT 12
432
433#ifdef CONFIG_PROC_PAGE_MONITOR
434struct mem_size_stats {
435 bool first;
436 unsigned long resident;
437 unsigned long shared_clean;
438 unsigned long shared_dirty;
439 unsigned long private_clean;
440 unsigned long private_dirty;
441 unsigned long referenced;
442 unsigned long anonymous;
443 unsigned long lazyfree;
444 unsigned long anonymous_thp;
445 unsigned long shmem_thp;
446 unsigned long swap;
447 unsigned long shared_hugetlb;
448 unsigned long private_hugetlb;
449 unsigned long first_vma_start;
450 u64 pss;
451 u64 pss_locked;
452 u64 swap_pss;
453 bool check_shmem_swap;
454};
455
456static void smaps_account(struct mem_size_stats *mss, struct page *page,
457 bool compound, bool young, bool dirty)
458{
459 int i, nr = compound ? 1 << compound_order(page) : 1;
460 unsigned long size = nr * PAGE_SIZE;
461
462 if (PageAnon(page)) {
463 mss->anonymous += size;
464 if (!PageSwapBacked(page) && !dirty && !PageDirty(page))
465 mss->lazyfree += size;
466 }
467
468 mss->resident += size;
469 /* Accumulate the size in pages that have been accessed. */
470 if (young || page_is_young(page) || PageReferenced(page))
471 mss->referenced += size;
472
473 /*
474 * page_count(page) == 1 guarantees the page is mapped exactly once.
475 * If any subpage of the compound page mapped with PTE it would elevate
476 * page_count().
477 */
478 if (page_count(page) == 1) {
479 if (dirty || PageDirty(page))
480 mss->private_dirty += size;
481 else
482 mss->private_clean += size;
483 mss->pss += (u64)size << PSS_SHIFT;
484 return;
485 }
486
487 for (i = 0; i < nr; i++, page++) {
488 int mapcount = page_mapcount(page);
489
490 if (mapcount >= 2) {
491 if (dirty || PageDirty(page))
492 mss->shared_dirty += PAGE_SIZE;
493 else
494 mss->shared_clean += PAGE_SIZE;
495 mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount;
496 } else {
497 if (dirty || PageDirty(page))
498 mss->private_dirty += PAGE_SIZE;
499 else
500 mss->private_clean += PAGE_SIZE;
501 mss->pss += PAGE_SIZE << PSS_SHIFT;
502 }
503 }
504}
505
506#ifdef CONFIG_SHMEM
507static int smaps_pte_hole(unsigned long addr, unsigned long end,
508 struct mm_walk *walk)
509{
510 struct mem_size_stats *mss = walk->private;
511
512 mss->swap += shmem_partial_swap_usage(
513 walk->vma->vm_file->f_mapping, addr, end);
514
515 return 0;
516}
517#endif
518
519static void smaps_pte_entry(pte_t *pte, unsigned long addr,
520 struct mm_walk *walk)
521{
522 struct mem_size_stats *mss = walk->private;
523 struct vm_area_struct *vma = walk->vma;
524 struct page *page = NULL;
525
526 if (pte_present(*pte)) {
527 page = vm_normal_page(vma, addr, *pte);
528 } else if (is_swap_pte(*pte)) {
529 swp_entry_t swpent = pte_to_swp_entry(*pte);
530
531 if (!non_swap_entry(swpent)) {
532 int mapcount;
533
534 mss->swap += PAGE_SIZE;
535 mapcount = swp_swapcount(swpent);
536 if (mapcount >= 2) {
537 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
538
539 do_div(pss_delta, mapcount);
540 mss->swap_pss += pss_delta;
541 } else {
542 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
543 }
544 } else if (is_migration_entry(swpent))
545 page = migration_entry_to_page(swpent);
546 else if (is_device_private_entry(swpent))
547 page = device_private_entry_to_page(swpent);
548 } else if (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap
549 && pte_none(*pte))) {
550 page = find_get_entry(vma->vm_file->f_mapping,
551 linear_page_index(vma, addr));
552 if (!page)
553 return;
554
555 if (radix_tree_exceptional_entry(page))
556 mss->swap += PAGE_SIZE;
557 else
558 put_page(page);
559
560 return;
561 }
562
563 if (!page)
564 return;
565
566 smaps_account(mss, page, false, pte_young(*pte), pte_dirty(*pte));
567}
568
569#ifdef CONFIG_TRANSPARENT_HUGEPAGE
570static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
571 struct mm_walk *walk)
572{
573 struct mem_size_stats *mss = walk->private;
574 struct vm_area_struct *vma = walk->vma;
575 struct page *page;
576
577 /* FOLL_DUMP will return -EFAULT on huge zero page */
578 page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
579 if (IS_ERR_OR_NULL(page))
580 return;
581 if (PageAnon(page))
582 mss->anonymous_thp += HPAGE_PMD_SIZE;
583 else if (PageSwapBacked(page))
584 mss->shmem_thp += HPAGE_PMD_SIZE;
585 else if (is_zone_device_page(page))
586 /* pass */;
587 else
588 VM_BUG_ON_PAGE(1, page);
589 smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd));
590}
591#else
592static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
593 struct mm_walk *walk)
594{
595}
596#endif
597
598static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
599 struct mm_walk *walk)
600{
601 struct vm_area_struct *vma = walk->vma;
602 pte_t *pte;
603 spinlock_t *ptl;
604
605 ptl = pmd_trans_huge_lock(pmd, vma);
606 if (ptl) {
607 if (pmd_present(*pmd))
608 smaps_pmd_entry(pmd, addr, walk);
609 spin_unlock(ptl);
610 goto out;
611 }
612
613 if (pmd_trans_unstable(pmd))
614 goto out;
615 /*
616 * The mmap_sem held all the way back in m_start() is what
617 * keeps khugepaged out of here and from collapsing things
618 * in here.
619 */
620 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
621 for (; addr != end; pte++, addr += PAGE_SIZE)
622 smaps_pte_entry(pte, addr, walk);
623 pte_unmap_unlock(pte - 1, ptl);
624out:
625 cond_resched();
626 return 0;
627}
628
629static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
630{
631 /*
632 * Don't forget to update Documentation/ on changes.
633 */
634 static const char mnemonics[BITS_PER_LONG][2] = {
635 /*
636 * In case if we meet a flag we don't know about.
637 */
638 [0 ... (BITS_PER_LONG-1)] = "??",
639
640 [ilog2(VM_READ)] = "rd",
641 [ilog2(VM_WRITE)] = "wr",
642 [ilog2(VM_EXEC)] = "ex",
643 [ilog2(VM_SHARED)] = "sh",
644 [ilog2(VM_MAYREAD)] = "mr",
645 [ilog2(VM_MAYWRITE)] = "mw",
646 [ilog2(VM_MAYEXEC)] = "me",
647 [ilog2(VM_MAYSHARE)] = "ms",
648 [ilog2(VM_GROWSDOWN)] = "gd",
649 [ilog2(VM_PFNMAP)] = "pf",
650 [ilog2(VM_DENYWRITE)] = "dw",
651#ifdef CONFIG_X86_INTEL_MPX
652 [ilog2(VM_MPX)] = "mp",
653#endif
654 [ilog2(VM_LOCKED)] = "lo",
655 [ilog2(VM_IO)] = "io",
656 [ilog2(VM_SEQ_READ)] = "sr",
657 [ilog2(VM_RAND_READ)] = "rr",
658 [ilog2(VM_DONTCOPY)] = "dc",
659 [ilog2(VM_DONTEXPAND)] = "de",
660 [ilog2(VM_ACCOUNT)] = "ac",
661 [ilog2(VM_NORESERVE)] = "nr",
662 [ilog2(VM_HUGETLB)] = "ht",
663 [ilog2(VM_SYNC)] = "sf",
664 [ilog2(VM_ARCH_1)] = "ar",
665 [ilog2(VM_WIPEONFORK)] = "wf",
666 [ilog2(VM_DONTDUMP)] = "dd",
667#ifdef CONFIG_MEM_SOFT_DIRTY
668 [ilog2(VM_SOFTDIRTY)] = "sd",
669#endif
670 [ilog2(VM_MIXEDMAP)] = "mm",
671 [ilog2(VM_HUGEPAGE)] = "hg",
672 [ilog2(VM_NOHUGEPAGE)] = "nh",
673 [ilog2(VM_MERGEABLE)] = "mg",
674 [ilog2(VM_UFFD_MISSING)]= "um",
675 [ilog2(VM_UFFD_WP)] = "uw",
676#ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
677 /* These come out via ProtectionKey: */
678 [ilog2(VM_PKEY_BIT0)] = "",
679 [ilog2(VM_PKEY_BIT1)] = "",
680 [ilog2(VM_PKEY_BIT2)] = "",
681 [ilog2(VM_PKEY_BIT3)] = "",
682#endif
683 };
684 size_t i;
685
686 seq_puts(m, "VmFlags: ");
687 for (i = 0; i < BITS_PER_LONG; i++) {
688 if (!mnemonics[i][0])
689 continue;
690 if (vma->vm_flags & (1UL << i)) {
691 seq_putc(m, mnemonics[i][0]);
692 seq_putc(m, mnemonics[i][1]);
693 seq_putc(m, ' ');
694 }
695 }
696 seq_putc(m, '\n');
697}
698
699#ifdef CONFIG_HUGETLB_PAGE
700static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
701 unsigned long addr, unsigned long end,
702 struct mm_walk *walk)
703{
704 struct mem_size_stats *mss = walk->private;
705 struct vm_area_struct *vma = walk->vma;
706 struct page *page = NULL;
707
708 if (pte_present(*pte)) {
709 page = vm_normal_page(vma, addr, *pte);
710 } else if (is_swap_pte(*pte)) {
711 swp_entry_t swpent = pte_to_swp_entry(*pte);
712
713 if (is_migration_entry(swpent))
714 page = migration_entry_to_page(swpent);
715 else if (is_device_private_entry(swpent))
716 page = device_private_entry_to_page(swpent);
717 }
718 if (page) {
719 int mapcount = page_mapcount(page);
720
721 if (mapcount >= 2)
722 mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
723 else
724 mss->private_hugetlb += huge_page_size(hstate_vma(vma));
725 }
726 return 0;
727}
728#endif /* HUGETLB_PAGE */
729
730void __weak arch_show_smap(struct seq_file *m, struct vm_area_struct *vma)
731{
732}
733
734#define SEQ_PUT_DEC(str, val) \
735 seq_put_decimal_ull_width(m, str, (val) >> 10, 8)
736static int show_smap(struct seq_file *m, void *v, int is_pid)
737{
738 struct proc_maps_private *priv = m->private;
739 struct vm_area_struct *vma = v;
740 struct mem_size_stats mss_stack;
741 struct mem_size_stats *mss;
742 struct mm_walk smaps_walk = {
743 .pmd_entry = smaps_pte_range,
744#ifdef CONFIG_HUGETLB_PAGE
745 .hugetlb_entry = smaps_hugetlb_range,
746#endif
747 .mm = vma->vm_mm,
748 };
749 int ret = 0;
750 bool rollup_mode;
751 bool last_vma;
752
753 if (priv->rollup) {
754 rollup_mode = true;
755 mss = priv->rollup;
756 if (mss->first) {
757 mss->first_vma_start = vma->vm_start;
758 mss->first = false;
759 }
760 last_vma = !m_next_vma(priv, vma);
761 } else {
762 rollup_mode = false;
763 memset(&mss_stack, 0, sizeof(mss_stack));
764 mss = &mss_stack;
765 }
766
767 smaps_walk.private = mss;
768
769#ifdef CONFIG_SHMEM
770 if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
771 /*
772 * For shared or readonly shmem mappings we know that all
773 * swapped out pages belong to the shmem object, and we can
774 * obtain the swap value much more efficiently. For private
775 * writable mappings, we might have COW pages that are
776 * not affected by the parent swapped out pages of the shmem
777 * object, so we have to distinguish them during the page walk.
778 * Unless we know that the shmem object (or the part mapped by
779 * our VMA) has no swapped out pages at all.
780 */
781 unsigned long shmem_swapped = shmem_swap_usage(vma);
782
783 if (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
784 !(vma->vm_flags & VM_WRITE)) {
785 mss->swap = shmem_swapped;
786 } else {
787 mss->check_shmem_swap = true;
788 smaps_walk.pte_hole = smaps_pte_hole;
789 }
790 }
791#endif
792
793 /* mmap_sem is held in m_start */
794 walk_page_vma(vma, &smaps_walk);
795 if (vma->vm_flags & VM_LOCKED)
796 mss->pss_locked += mss->pss;
797
798 if (!rollup_mode) {
799 show_map_vma(m, vma, is_pid);
800 } else if (last_vma) {
801 show_vma_header_prefix(
802 m, mss->first_vma_start, vma->vm_end, 0, 0, 0, 0);
803 seq_pad(m, ' ');
804 seq_puts(m, "[rollup]\n");
805 } else {
806 ret = SEQ_SKIP;
807 }
808
809 if (!rollup_mode) {
810 SEQ_PUT_DEC("Size: ", vma->vm_end - vma->vm_start);
811 SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma));
812 SEQ_PUT_DEC(" kB\nMMUPageSize: ", vma_mmu_pagesize(vma));
813 seq_puts(m, " kB\n");
814 }
815
816 if (!rollup_mode || last_vma) {
817 SEQ_PUT_DEC("Rss: ", mss->resident);
818 SEQ_PUT_DEC(" kB\nPss: ", mss->pss >> PSS_SHIFT);
819 SEQ_PUT_DEC(" kB\nShared_Clean: ", mss->shared_clean);
820 SEQ_PUT_DEC(" kB\nShared_Dirty: ", mss->shared_dirty);
821 SEQ_PUT_DEC(" kB\nPrivate_Clean: ", mss->private_clean);
822 SEQ_PUT_DEC(" kB\nPrivate_Dirty: ", mss->private_dirty);
823 SEQ_PUT_DEC(" kB\nReferenced: ", mss->referenced);
824 SEQ_PUT_DEC(" kB\nAnonymous: ", mss->anonymous);
825 SEQ_PUT_DEC(" kB\nLazyFree: ", mss->lazyfree);
826 SEQ_PUT_DEC(" kB\nAnonHugePages: ", mss->anonymous_thp);
827 SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp);
828 SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb);
829 seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ",
830 mss->private_hugetlb >> 10, 7);
831 SEQ_PUT_DEC(" kB\nSwap: ", mss->swap);
832 SEQ_PUT_DEC(" kB\nSwapPss: ",
833 mss->swap_pss >> PSS_SHIFT);
834 SEQ_PUT_DEC(" kB\nLocked: ", mss->pss >> PSS_SHIFT);
835 seq_puts(m, " kB\n");
836 }
837 if (!rollup_mode) {
838 arch_show_smap(m, vma);
839 show_smap_vma_flags(m, vma);
840 }
841 m_cache_vma(m, vma);
842 return ret;
843}
844#undef SEQ_PUT_DEC
845
846static int show_pid_smap(struct seq_file *m, void *v)
847{
848 return show_smap(m, v, 1);
849}
850
851static int show_tid_smap(struct seq_file *m, void *v)
852{
853 return show_smap(m, v, 0);
854}
855
856static const struct seq_operations proc_pid_smaps_op = {
857 .start = m_start,
858 .next = m_next,
859 .stop = m_stop,
860 .show = show_pid_smap
861};
862
863static const struct seq_operations proc_tid_smaps_op = {
864 .start = m_start,
865 .next = m_next,
866 .stop = m_stop,
867 .show = show_tid_smap
868};
869
870static int pid_smaps_open(struct inode *inode, struct file *file)
871{
872 return do_maps_open(inode, file, &proc_pid_smaps_op);
873}
874
875static int pid_smaps_rollup_open(struct inode *inode, struct file *file)
876{
877 struct seq_file *seq;
878 struct proc_maps_private *priv;
879 int ret = do_maps_open(inode, file, &proc_pid_smaps_op);
880
881 if (ret < 0)
882 return ret;
883 seq = file->private_data;
884 priv = seq->private;
885 priv->rollup = kzalloc(sizeof(*priv->rollup), GFP_KERNEL);
886 if (!priv->rollup) {
887 proc_map_release(inode, file);
888 return -ENOMEM;
889 }
890 priv->rollup->first = true;
891 return 0;
892}
893
894static int tid_smaps_open(struct inode *inode, struct file *file)
895{
896 return do_maps_open(inode, file, &proc_tid_smaps_op);
897}
898
899const struct file_operations proc_pid_smaps_operations = {
900 .open = pid_smaps_open,
901 .read = seq_read,
902 .llseek = seq_lseek,
903 .release = proc_map_release,
904};
905
906const struct file_operations proc_pid_smaps_rollup_operations = {
907 .open = pid_smaps_rollup_open,
908 .read = seq_read,
909 .llseek = seq_lseek,
910 .release = proc_map_release,
911};
912
913const struct file_operations proc_tid_smaps_operations = {
914 .open = tid_smaps_open,
915 .read = seq_read,
916 .llseek = seq_lseek,
917 .release = proc_map_release,
918};
919
920enum clear_refs_types {
921 CLEAR_REFS_ALL = 1,
922 CLEAR_REFS_ANON,
923 CLEAR_REFS_MAPPED,
924 CLEAR_REFS_SOFT_DIRTY,
925 CLEAR_REFS_MM_HIWATER_RSS,
926 CLEAR_REFS_LAST,
927};
928
929struct clear_refs_private {
930 enum clear_refs_types type;
931};
932
933#ifdef CONFIG_MEM_SOFT_DIRTY
934static inline void clear_soft_dirty(struct vm_area_struct *vma,
935 unsigned long addr, pte_t *pte)
936{
937 /*
938 * The soft-dirty tracker uses #PF-s to catch writes
939 * to pages, so write-protect the pte as well. See the
940 * Documentation/vm/soft-dirty.txt for full description
941 * of how soft-dirty works.
942 */
943 pte_t ptent = *pte;
944
945 if (pte_present(ptent)) {
946 ptent = ptep_modify_prot_start(vma->vm_mm, addr, pte);
947 ptent = pte_wrprotect(ptent);
948 ptent = pte_clear_soft_dirty(ptent);
949 ptep_modify_prot_commit(vma->vm_mm, addr, pte, ptent);
950 } else if (is_swap_pte(ptent)) {
951 ptent = pte_swp_clear_soft_dirty(ptent);
952 set_pte_at(vma->vm_mm, addr, pte, ptent);
953 }
954}
955#else
956static inline void clear_soft_dirty(struct vm_area_struct *vma,
957 unsigned long addr, pte_t *pte)
958{
959}
960#endif
961
962#if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
963static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
964 unsigned long addr, pmd_t *pmdp)
965{
966 pmd_t old, pmd = *pmdp;
967
968 if (pmd_present(pmd)) {
969 /* See comment in change_huge_pmd() */
970 old = pmdp_invalidate(vma, addr, pmdp);
971 if (pmd_dirty(old))
972 pmd = pmd_mkdirty(pmd);
973 if (pmd_young(old))
974 pmd = pmd_mkyoung(pmd);
975
976 pmd = pmd_wrprotect(pmd);
977 pmd = pmd_clear_soft_dirty(pmd);
978
979 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
980 } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
981 pmd = pmd_swp_clear_soft_dirty(pmd);
982 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
983 }
984}
985#else
986static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
987 unsigned long addr, pmd_t *pmdp)
988{
989}
990#endif
991
992static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
993 unsigned long end, struct mm_walk *walk)
994{
995 struct clear_refs_private *cp = walk->private;
996 struct vm_area_struct *vma = walk->vma;
997 pte_t *pte, ptent;
998 spinlock_t *ptl;
999 struct page *page;
1000
1001 ptl = pmd_trans_huge_lock(pmd, vma);
1002 if (ptl) {
1003 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1004 clear_soft_dirty_pmd(vma, addr, pmd);
1005 goto out;
1006 }
1007
1008 if (!pmd_present(*pmd))
1009 goto out;
1010
1011 page = pmd_page(*pmd);
1012
1013 /* Clear accessed and referenced bits. */
1014 pmdp_test_and_clear_young(vma, addr, pmd);
1015 test_and_clear_page_young(page);
1016 ClearPageReferenced(page);
1017out:
1018 spin_unlock(ptl);
1019 return 0;
1020 }
1021
1022 if (pmd_trans_unstable(pmd))
1023 return 0;
1024
1025 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1026 for (; addr != end; pte++, addr += PAGE_SIZE) {
1027 ptent = *pte;
1028
1029 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1030 clear_soft_dirty(vma, addr, pte);
1031 continue;
1032 }
1033
1034 if (!pte_present(ptent))
1035 continue;
1036
1037 page = vm_normal_page(vma, addr, ptent);
1038 if (!page)
1039 continue;
1040
1041 /* Clear accessed and referenced bits. */
1042 ptep_test_and_clear_young(vma, addr, pte);
1043 test_and_clear_page_young(page);
1044 ClearPageReferenced(page);
1045 }
1046 pte_unmap_unlock(pte - 1, ptl);
1047 cond_resched();
1048 return 0;
1049}
1050
1051static int clear_refs_test_walk(unsigned long start, unsigned long end,
1052 struct mm_walk *walk)
1053{
1054 struct clear_refs_private *cp = walk->private;
1055 struct vm_area_struct *vma = walk->vma;
1056
1057 if (vma->vm_flags & VM_PFNMAP)
1058 return 1;
1059
1060 /*
1061 * Writing 1 to /proc/pid/clear_refs affects all pages.
1062 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1063 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1064 * Writing 4 to /proc/pid/clear_refs affects all pages.
1065 */
1066 if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1067 return 1;
1068 if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1069 return 1;
1070 return 0;
1071}
1072
1073static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1074 size_t count, loff_t *ppos)
1075{
1076 struct task_struct *task;
1077 char buffer[PROC_NUMBUF];
1078 struct mm_struct *mm;
1079 struct vm_area_struct *vma;
1080 enum clear_refs_types type;
1081 struct mmu_gather tlb;
1082 int itype;
1083 int rv;
1084
1085 memset(buffer, 0, sizeof(buffer));
1086 if (count > sizeof(buffer) - 1)
1087 count = sizeof(buffer) - 1;
1088 if (copy_from_user(buffer, buf, count))
1089 return -EFAULT;
1090 rv = kstrtoint(strstrip(buffer), 10, &itype);
1091 if (rv < 0)
1092 return rv;
1093 type = (enum clear_refs_types)itype;
1094 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1095 return -EINVAL;
1096
1097 task = get_proc_task(file_inode(file));
1098 if (!task)
1099 return -ESRCH;
1100 mm = get_task_mm(task);
1101 if (mm) {
1102 struct clear_refs_private cp = {
1103 .type = type,
1104 };
1105 struct mm_walk clear_refs_walk = {
1106 .pmd_entry = clear_refs_pte_range,
1107 .test_walk = clear_refs_test_walk,
1108 .mm = mm,
1109 .private = &cp,
1110 };
1111
1112 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1113 if (down_write_killable(&mm->mmap_sem)) {
1114 count = -EINTR;
1115 goto out_mm;
1116 }
1117
1118 /*
1119 * Writing 5 to /proc/pid/clear_refs resets the peak
1120 * resident set size to this mm's current rss value.
1121 */
1122 reset_mm_hiwater_rss(mm);
1123 up_write(&mm->mmap_sem);
1124 goto out_mm;
1125 }
1126
1127 down_read(&mm->mmap_sem);
1128 tlb_gather_mmu(&tlb, mm, 0, -1);
1129 if (type == CLEAR_REFS_SOFT_DIRTY) {
1130 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1131 if (!(vma->vm_flags & VM_SOFTDIRTY))
1132 continue;
1133 up_read(&mm->mmap_sem);
1134 if (down_write_killable(&mm->mmap_sem)) {
1135 count = -EINTR;
1136 goto out_mm;
1137 }
1138 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1139 vma->vm_flags &= ~VM_SOFTDIRTY;
1140 vma_set_page_prot(vma);
1141 }
1142 downgrade_write(&mm->mmap_sem);
1143 break;
1144 }
1145 mmu_notifier_invalidate_range_start(mm, 0, -1);
1146 }
1147 walk_page_range(0, mm->highest_vm_end, &clear_refs_walk);
1148 if (type == CLEAR_REFS_SOFT_DIRTY)
1149 mmu_notifier_invalidate_range_end(mm, 0, -1);
1150 tlb_finish_mmu(&tlb, 0, -1);
1151 up_read(&mm->mmap_sem);
1152out_mm:
1153 mmput(mm);
1154 }
1155 put_task_struct(task);
1156
1157 return count;
1158}
1159
1160const struct file_operations proc_clear_refs_operations = {
1161 .write = clear_refs_write,
1162 .llseek = noop_llseek,
1163};
1164
1165typedef struct {
1166 u64 pme;
1167} pagemap_entry_t;
1168
1169struct pagemapread {
1170 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */
1171 pagemap_entry_t *buffer;
1172 bool show_pfn;
1173};
1174
1175#define PAGEMAP_WALK_SIZE (PMD_SIZE)
1176#define PAGEMAP_WALK_MASK (PMD_MASK)
1177
1178#define PM_ENTRY_BYTES sizeof(pagemap_entry_t)
1179#define PM_PFRAME_BITS 55
1180#define PM_PFRAME_MASK GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1181#define PM_SOFT_DIRTY BIT_ULL(55)
1182#define PM_MMAP_EXCLUSIVE BIT_ULL(56)
1183#define PM_FILE BIT_ULL(61)
1184#define PM_SWAP BIT_ULL(62)
1185#define PM_PRESENT BIT_ULL(63)
1186
1187#define PM_END_OF_BUFFER 1
1188
1189static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1190{
1191 return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1192}
1193
1194static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1195 struct pagemapread *pm)
1196{
1197 pm->buffer[pm->pos++] = *pme;
1198 if (pm->pos >= pm->len)
1199 return PM_END_OF_BUFFER;
1200 return 0;
1201}
1202
1203static int pagemap_pte_hole(unsigned long start, unsigned long end,
1204 struct mm_walk *walk)
1205{
1206 struct pagemapread *pm = walk->private;
1207 unsigned long addr = start;
1208 int err = 0;
1209
1210 while (addr < end) {
1211 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1212 pagemap_entry_t pme = make_pme(0, 0);
1213 /* End of address space hole, which we mark as non-present. */
1214 unsigned long hole_end;
1215
1216 if (vma)
1217 hole_end = min(end, vma->vm_start);
1218 else
1219 hole_end = end;
1220
1221 for (; addr < hole_end; addr += PAGE_SIZE) {
1222 err = add_to_pagemap(addr, &pme, pm);
1223 if (err)
1224 goto out;
1225 }
1226
1227 if (!vma)
1228 break;
1229
1230 /* Addresses in the VMA. */
1231 if (vma->vm_flags & VM_SOFTDIRTY)
1232 pme = make_pme(0, PM_SOFT_DIRTY);
1233 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1234 err = add_to_pagemap(addr, &pme, pm);
1235 if (err)
1236 goto out;
1237 }
1238 }
1239out:
1240 return err;
1241}
1242
1243static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1244 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1245{
1246 u64 frame = 0, flags = 0;
1247 struct page *page = NULL;
1248
1249 if (pte_present(pte)) {
1250 if (pm->show_pfn)
1251 frame = pte_pfn(pte);
1252 flags |= PM_PRESENT;
1253 page = _vm_normal_page(vma, addr, pte, true);
1254 if (pte_soft_dirty(pte))
1255 flags |= PM_SOFT_DIRTY;
1256 } else if (is_swap_pte(pte)) {
1257 swp_entry_t entry;
1258 if (pte_swp_soft_dirty(pte))
1259 flags |= PM_SOFT_DIRTY;
1260 entry = pte_to_swp_entry(pte);
1261 frame = swp_type(entry) |
1262 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1263 flags |= PM_SWAP;
1264 if (is_migration_entry(entry))
1265 page = migration_entry_to_page(entry);
1266
1267 if (is_device_private_entry(entry))
1268 page = device_private_entry_to_page(entry);
1269 }
1270
1271 if (page && !PageAnon(page))
1272 flags |= PM_FILE;
1273 if (page && page_mapcount(page) == 1)
1274 flags |= PM_MMAP_EXCLUSIVE;
1275 if (vma->vm_flags & VM_SOFTDIRTY)
1276 flags |= PM_SOFT_DIRTY;
1277
1278 return make_pme(frame, flags);
1279}
1280
1281static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1282 struct mm_walk *walk)
1283{
1284 struct vm_area_struct *vma = walk->vma;
1285 struct pagemapread *pm = walk->private;
1286 spinlock_t *ptl;
1287 pte_t *pte, *orig_pte;
1288 int err = 0;
1289
1290#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1291 ptl = pmd_trans_huge_lock(pmdp, vma);
1292 if (ptl) {
1293 u64 flags = 0, frame = 0;
1294 pmd_t pmd = *pmdp;
1295 struct page *page = NULL;
1296
1297 if (vma->vm_flags & VM_SOFTDIRTY)
1298 flags |= PM_SOFT_DIRTY;
1299
1300 if (pmd_present(pmd)) {
1301 page = pmd_page(pmd);
1302
1303 flags |= PM_PRESENT;
1304 if (pmd_soft_dirty(pmd))
1305 flags |= PM_SOFT_DIRTY;
1306 if (pm->show_pfn)
1307 frame = pmd_pfn(pmd) +
1308 ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1309 }
1310#ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1311 else if (is_swap_pmd(pmd)) {
1312 swp_entry_t entry = pmd_to_swp_entry(pmd);
1313 unsigned long offset = swp_offset(entry);
1314
1315 offset += (addr & ~PMD_MASK) >> PAGE_SHIFT;
1316 frame = swp_type(entry) |
1317 (offset << MAX_SWAPFILES_SHIFT);
1318 flags |= PM_SWAP;
1319 if (pmd_swp_soft_dirty(pmd))
1320 flags |= PM_SOFT_DIRTY;
1321 VM_BUG_ON(!is_pmd_migration_entry(pmd));
1322 page = migration_entry_to_page(entry);
1323 }
1324#endif
1325
1326 if (page && page_mapcount(page) == 1)
1327 flags |= PM_MMAP_EXCLUSIVE;
1328
1329 for (; addr != end; addr += PAGE_SIZE) {
1330 pagemap_entry_t pme = make_pme(frame, flags);
1331
1332 err = add_to_pagemap(addr, &pme, pm);
1333 if (err)
1334 break;
1335 if (pm->show_pfn && (flags & PM_PRESENT))
1336 frame++;
1337 else if (flags & PM_SWAP)
1338 frame += (1 << MAX_SWAPFILES_SHIFT);
1339 }
1340 spin_unlock(ptl);
1341 return err;
1342 }
1343
1344 if (pmd_trans_unstable(pmdp))
1345 return 0;
1346#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1347
1348 /*
1349 * We can assume that @vma always points to a valid one and @end never
1350 * goes beyond vma->vm_end.
1351 */
1352 orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1353 for (; addr < end; pte++, addr += PAGE_SIZE) {
1354 pagemap_entry_t pme;
1355
1356 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1357 err = add_to_pagemap(addr, &pme, pm);
1358 if (err)
1359 break;
1360 }
1361 pte_unmap_unlock(orig_pte, ptl);
1362
1363 cond_resched();
1364
1365 return err;
1366}
1367
1368#ifdef CONFIG_HUGETLB_PAGE
1369/* This function walks within one hugetlb entry in the single call */
1370static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1371 unsigned long addr, unsigned long end,
1372 struct mm_walk *walk)
1373{
1374 struct pagemapread *pm = walk->private;
1375 struct vm_area_struct *vma = walk->vma;
1376 u64 flags = 0, frame = 0;
1377 int err = 0;
1378 pte_t pte;
1379
1380 if (vma->vm_flags & VM_SOFTDIRTY)
1381 flags |= PM_SOFT_DIRTY;
1382
1383 pte = huge_ptep_get(ptep);
1384 if (pte_present(pte)) {
1385 struct page *page = pte_page(pte);
1386
1387 if (!PageAnon(page))
1388 flags |= PM_FILE;
1389
1390 if (page_mapcount(page) == 1)
1391 flags |= PM_MMAP_EXCLUSIVE;
1392
1393 flags |= PM_PRESENT;
1394 if (pm->show_pfn)
1395 frame = pte_pfn(pte) +
1396 ((addr & ~hmask) >> PAGE_SHIFT);
1397 }
1398
1399 for (; addr != end; addr += PAGE_SIZE) {
1400 pagemap_entry_t pme = make_pme(frame, flags);
1401
1402 err = add_to_pagemap(addr, &pme, pm);
1403 if (err)
1404 return err;
1405 if (pm->show_pfn && (flags & PM_PRESENT))
1406 frame++;
1407 }
1408
1409 cond_resched();
1410
1411 return err;
1412}
1413#endif /* HUGETLB_PAGE */
1414
1415/*
1416 * /proc/pid/pagemap - an array mapping virtual pages to pfns
1417 *
1418 * For each page in the address space, this file contains one 64-bit entry
1419 * consisting of the following:
1420 *
1421 * Bits 0-54 page frame number (PFN) if present
1422 * Bits 0-4 swap type if swapped
1423 * Bits 5-54 swap offset if swapped
1424 * Bit 55 pte is soft-dirty (see Documentation/vm/soft-dirty.txt)
1425 * Bit 56 page exclusively mapped
1426 * Bits 57-60 zero
1427 * Bit 61 page is file-page or shared-anon
1428 * Bit 62 page swapped
1429 * Bit 63 page present
1430 *
1431 * If the page is not present but in swap, then the PFN contains an
1432 * encoding of the swap file number and the page's offset into the
1433 * swap. Unmapped pages return a null PFN. This allows determining
1434 * precisely which pages are mapped (or in swap) and comparing mapped
1435 * pages between processes.
1436 *
1437 * Efficient users of this interface will use /proc/pid/maps to
1438 * determine which areas of memory are actually mapped and llseek to
1439 * skip over unmapped regions.
1440 */
1441static ssize_t pagemap_read(struct file *file, char __user *buf,
1442 size_t count, loff_t *ppos)
1443{
1444 struct mm_struct *mm = file->private_data;
1445 struct pagemapread pm;
1446 struct mm_walk pagemap_walk = {};
1447 unsigned long src;
1448 unsigned long svpfn;
1449 unsigned long start_vaddr;
1450 unsigned long end_vaddr;
1451 int ret = 0, copied = 0;
1452
1453 if (!mm || !mmget_not_zero(mm))
1454 goto out;
1455
1456 ret = -EINVAL;
1457 /* file position must be aligned */
1458 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1459 goto out_mm;
1460
1461 ret = 0;
1462 if (!count)
1463 goto out_mm;
1464
1465 /* do not disclose physical addresses: attack vector */
1466 pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1467
1468 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1469 pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_KERNEL);
1470 ret = -ENOMEM;
1471 if (!pm.buffer)
1472 goto out_mm;
1473
1474 pagemap_walk.pmd_entry = pagemap_pmd_range;
1475 pagemap_walk.pte_hole = pagemap_pte_hole;
1476#ifdef CONFIG_HUGETLB_PAGE
1477 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1478#endif
1479 pagemap_walk.mm = mm;
1480 pagemap_walk.private = ±
1481
1482 src = *ppos;
1483 svpfn = src / PM_ENTRY_BYTES;
1484 start_vaddr = svpfn << PAGE_SHIFT;
1485 end_vaddr = mm->task_size;
1486
1487 /* watch out for wraparound */
1488 if (svpfn > mm->task_size >> PAGE_SHIFT)
1489 start_vaddr = end_vaddr;
1490
1491 /*
1492 * The odds are that this will stop walking way
1493 * before end_vaddr, because the length of the
1494 * user buffer is tracked in "pm", and the walk
1495 * will stop when we hit the end of the buffer.
1496 */
1497 ret = 0;
1498 while (count && (start_vaddr < end_vaddr)) {
1499 int len;
1500 unsigned long end;
1501
1502 pm.pos = 0;
1503 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1504 /* overflow ? */
1505 if (end < start_vaddr || end > end_vaddr)
1506 end = end_vaddr;
1507 down_read(&mm->mmap_sem);
1508 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1509 up_read(&mm->mmap_sem);
1510 start_vaddr = end;
1511
1512 len = min(count, PM_ENTRY_BYTES * pm.pos);
1513 if (copy_to_user(buf, pm.buffer, len)) {
1514 ret = -EFAULT;
1515 goto out_free;
1516 }
1517 copied += len;
1518 buf += len;
1519 count -= len;
1520 }
1521 *ppos += copied;
1522 if (!ret || ret == PM_END_OF_BUFFER)
1523 ret = copied;
1524
1525out_free:
1526 kfree(pm.buffer);
1527out_mm:
1528 mmput(mm);
1529out:
1530 return ret;
1531}
1532
1533static int pagemap_open(struct inode *inode, struct file *file)
1534{
1535 struct mm_struct *mm;
1536
1537 mm = proc_mem_open(inode, PTRACE_MODE_READ);
1538 if (IS_ERR(mm))
1539 return PTR_ERR(mm);
1540 file->private_data = mm;
1541 return 0;
1542}
1543
1544static int pagemap_release(struct inode *inode, struct file *file)
1545{
1546 struct mm_struct *mm = file->private_data;
1547
1548 if (mm)
1549 mmdrop(mm);
1550 return 0;
1551}
1552
1553const struct file_operations proc_pagemap_operations = {
1554 .llseek = mem_lseek, /* borrow this */
1555 .read = pagemap_read,
1556 .open = pagemap_open,
1557 .release = pagemap_release,
1558};
1559#endif /* CONFIG_PROC_PAGE_MONITOR */
1560
1561#ifdef CONFIG_NUMA
1562
1563struct numa_maps {
1564 unsigned long pages;
1565 unsigned long anon;
1566 unsigned long active;
1567 unsigned long writeback;
1568 unsigned long mapcount_max;
1569 unsigned long dirty;
1570 unsigned long swapcache;
1571 unsigned long node[MAX_NUMNODES];
1572};
1573
1574struct numa_maps_private {
1575 struct proc_maps_private proc_maps;
1576 struct numa_maps md;
1577};
1578
1579static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1580 unsigned long nr_pages)
1581{
1582 int count = page_mapcount(page);
1583
1584 md->pages += nr_pages;
1585 if (pte_dirty || PageDirty(page))
1586 md->dirty += nr_pages;
1587
1588 if (PageSwapCache(page))
1589 md->swapcache += nr_pages;
1590
1591 if (PageActive(page) || PageUnevictable(page))
1592 md->active += nr_pages;
1593
1594 if (PageWriteback(page))
1595 md->writeback += nr_pages;
1596
1597 if (PageAnon(page))
1598 md->anon += nr_pages;
1599
1600 if (count > md->mapcount_max)
1601 md->mapcount_max = count;
1602
1603 md->node[page_to_nid(page)] += nr_pages;
1604}
1605
1606static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1607 unsigned long addr)
1608{
1609 struct page *page;
1610 int nid;
1611
1612 if (!pte_present(pte))
1613 return NULL;
1614
1615 page = vm_normal_page(vma, addr, pte);
1616 if (!page)
1617 return NULL;
1618
1619 if (PageReserved(page))
1620 return NULL;
1621
1622 nid = page_to_nid(page);
1623 if (!node_isset(nid, node_states[N_MEMORY]))
1624 return NULL;
1625
1626 return page;
1627}
1628
1629#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1630static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1631 struct vm_area_struct *vma,
1632 unsigned long addr)
1633{
1634 struct page *page;
1635 int nid;
1636
1637 if (!pmd_present(pmd))
1638 return NULL;
1639
1640 page = vm_normal_page_pmd(vma, addr, pmd);
1641 if (!page)
1642 return NULL;
1643
1644 if (PageReserved(page))
1645 return NULL;
1646
1647 nid = page_to_nid(page);
1648 if (!node_isset(nid, node_states[N_MEMORY]))
1649 return NULL;
1650
1651 return page;
1652}
1653#endif
1654
1655static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1656 unsigned long end, struct mm_walk *walk)
1657{
1658 struct numa_maps *md = walk->private;
1659 struct vm_area_struct *vma = walk->vma;
1660 spinlock_t *ptl;
1661 pte_t *orig_pte;
1662 pte_t *pte;
1663
1664#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1665 ptl = pmd_trans_huge_lock(pmd, vma);
1666 if (ptl) {
1667 struct page *page;
1668
1669 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1670 if (page)
1671 gather_stats(page, md, pmd_dirty(*pmd),
1672 HPAGE_PMD_SIZE/PAGE_SIZE);
1673 spin_unlock(ptl);
1674 return 0;
1675 }
1676
1677 if (pmd_trans_unstable(pmd))
1678 return 0;
1679#endif
1680 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1681 do {
1682 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1683 if (!page)
1684 continue;
1685 gather_stats(page, md, pte_dirty(*pte), 1);
1686
1687 } while (pte++, addr += PAGE_SIZE, addr != end);
1688 pte_unmap_unlock(orig_pte, ptl);
1689 cond_resched();
1690 return 0;
1691}
1692#ifdef CONFIG_HUGETLB_PAGE
1693static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1694 unsigned long addr, unsigned long end, struct mm_walk *walk)
1695{
1696 pte_t huge_pte = huge_ptep_get(pte);
1697 struct numa_maps *md;
1698 struct page *page;
1699
1700 if (!pte_present(huge_pte))
1701 return 0;
1702
1703 page = pte_page(huge_pte);
1704 if (!page)
1705 return 0;
1706
1707 md = walk->private;
1708 gather_stats(page, md, pte_dirty(huge_pte), 1);
1709 return 0;
1710}
1711
1712#else
1713static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1714 unsigned long addr, unsigned long end, struct mm_walk *walk)
1715{
1716 return 0;
1717}
1718#endif
1719
1720/*
1721 * Display pages allocated per node and memory policy via /proc.
1722 */
1723static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1724{
1725 struct numa_maps_private *numa_priv = m->private;
1726 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1727 struct vm_area_struct *vma = v;
1728 struct numa_maps *md = &numa_priv->md;
1729 struct file *file = vma->vm_file;
1730 struct mm_struct *mm = vma->vm_mm;
1731 struct mm_walk walk = {
1732 .hugetlb_entry = gather_hugetlb_stats,
1733 .pmd_entry = gather_pte_stats,
1734 .private = md,
1735 .mm = mm,
1736 };
1737 struct mempolicy *pol;
1738 char buffer[64];
1739 int nid;
1740
1741 if (!mm)
1742 return 0;
1743
1744 /* Ensure we start with an empty set of numa_maps statistics. */
1745 memset(md, 0, sizeof(*md));
1746
1747 pol = __get_vma_policy(vma, vma->vm_start);
1748 if (pol) {
1749 mpol_to_str(buffer, sizeof(buffer), pol);
1750 mpol_cond_put(pol);
1751 } else {
1752 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1753 }
1754
1755 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1756
1757 if (file) {
1758 seq_puts(m, " file=");
1759 seq_file_path(m, file, "\n\t= ");
1760 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1761 seq_puts(m, " heap");
1762 } else if (is_stack(vma)) {
1763 seq_puts(m, " stack");
1764 }
1765
1766 if (is_vm_hugetlb_page(vma))
1767 seq_puts(m, " huge");
1768
1769 /* mmap_sem is held by m_start */
1770 walk_page_vma(vma, &walk);
1771
1772 if (!md->pages)
1773 goto out;
1774
1775 if (md->anon)
1776 seq_printf(m, " anon=%lu", md->anon);
1777
1778 if (md->dirty)
1779 seq_printf(m, " dirty=%lu", md->dirty);
1780
1781 if (md->pages != md->anon && md->pages != md->dirty)
1782 seq_printf(m, " mapped=%lu", md->pages);
1783
1784 if (md->mapcount_max > 1)
1785 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1786
1787 if (md->swapcache)
1788 seq_printf(m, " swapcache=%lu", md->swapcache);
1789
1790 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1791 seq_printf(m, " active=%lu", md->active);
1792
1793 if (md->writeback)
1794 seq_printf(m, " writeback=%lu", md->writeback);
1795
1796 for_each_node_state(nid, N_MEMORY)
1797 if (md->node[nid])
1798 seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1799
1800 seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1801out:
1802 seq_putc(m, '\n');
1803 m_cache_vma(m, vma);
1804 return 0;
1805}
1806
1807static int show_pid_numa_map(struct seq_file *m, void *v)
1808{
1809 return show_numa_map(m, v, 1);
1810}
1811
1812static int show_tid_numa_map(struct seq_file *m, void *v)
1813{
1814 return show_numa_map(m, v, 0);
1815}
1816
1817static const struct seq_operations proc_pid_numa_maps_op = {
1818 .start = m_start,
1819 .next = m_next,
1820 .stop = m_stop,
1821 .show = show_pid_numa_map,
1822};
1823
1824static const struct seq_operations proc_tid_numa_maps_op = {
1825 .start = m_start,
1826 .next = m_next,
1827 .stop = m_stop,
1828 .show = show_tid_numa_map,
1829};
1830
1831static int numa_maps_open(struct inode *inode, struct file *file,
1832 const struct seq_operations *ops)
1833{
1834 return proc_maps_open(inode, file, ops,
1835 sizeof(struct numa_maps_private));
1836}
1837
1838static int pid_numa_maps_open(struct inode *inode, struct file *file)
1839{
1840 return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1841}
1842
1843static int tid_numa_maps_open(struct inode *inode, struct file *file)
1844{
1845 return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1846}
1847
1848const struct file_operations proc_pid_numa_maps_operations = {
1849 .open = pid_numa_maps_open,
1850 .read = seq_read,
1851 .llseek = seq_lseek,
1852 .release = proc_map_release,
1853};
1854
1855const struct file_operations proc_tid_numa_maps_operations = {
1856 .open = tid_numa_maps_open,
1857 .read = seq_read,
1858 .llseek = seq_lseek,
1859 .release = proc_map_release,
1860};
1861#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 */