Loading...
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#include <linux/page_idle.h>
17#include <linux/shmem_fs.h>
18
19#include <asm/elf.h>
20#include <asm/uaccess.h>
21#include <asm/tlbflush.h>
22#include "internal.h"
23
24void task_mem(struct seq_file *m, struct mm_struct *mm)
25{
26 unsigned long text, lib, swap, ptes, pmds, anon, file, shmem;
27 unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
28
29 anon = get_mm_counter(mm, MM_ANONPAGES);
30 file = get_mm_counter(mm, MM_FILEPAGES);
31 shmem = get_mm_counter(mm, MM_SHMEMPAGES);
32
33 /*
34 * Note: to minimize their overhead, mm maintains hiwater_vm and
35 * hiwater_rss only when about to *lower* total_vm or rss. Any
36 * collector of these hiwater stats must therefore get total_vm
37 * and rss too, which will usually be the higher. Barriers? not
38 * worth the effort, such snapshots can always be inconsistent.
39 */
40 hiwater_vm = total_vm = mm->total_vm;
41 if (hiwater_vm < mm->hiwater_vm)
42 hiwater_vm = mm->hiwater_vm;
43 hiwater_rss = total_rss = anon + file + shmem;
44 if (hiwater_rss < mm->hiwater_rss)
45 hiwater_rss = mm->hiwater_rss;
46
47 text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
48 lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
49 swap = get_mm_counter(mm, MM_SWAPENTS);
50 ptes = PTRS_PER_PTE * sizeof(pte_t) * atomic_long_read(&mm->nr_ptes);
51 pmds = PTRS_PER_PMD * sizeof(pmd_t) * mm_nr_pmds(mm);
52 seq_printf(m,
53 "VmPeak:\t%8lu kB\n"
54 "VmSize:\t%8lu kB\n"
55 "VmLck:\t%8lu kB\n"
56 "VmPin:\t%8lu kB\n"
57 "VmHWM:\t%8lu kB\n"
58 "VmRSS:\t%8lu kB\n"
59 "RssAnon:\t%8lu kB\n"
60 "RssFile:\t%8lu kB\n"
61 "RssShmem:\t%8lu kB\n"
62 "VmData:\t%8lu kB\n"
63 "VmStk:\t%8lu kB\n"
64 "VmExe:\t%8lu kB\n"
65 "VmLib:\t%8lu kB\n"
66 "VmPTE:\t%8lu kB\n"
67 "VmPMD:\t%8lu kB\n"
68 "VmSwap:\t%8lu kB\n",
69 hiwater_vm << (PAGE_SHIFT-10),
70 total_vm << (PAGE_SHIFT-10),
71 mm->locked_vm << (PAGE_SHIFT-10),
72 mm->pinned_vm << (PAGE_SHIFT-10),
73 hiwater_rss << (PAGE_SHIFT-10),
74 total_rss << (PAGE_SHIFT-10),
75 anon << (PAGE_SHIFT-10),
76 file << (PAGE_SHIFT-10),
77 shmem << (PAGE_SHIFT-10),
78 mm->data_vm << (PAGE_SHIFT-10),
79 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
80 ptes >> 10,
81 pmds >> 10,
82 swap << (PAGE_SHIFT-10));
83 hugetlb_report_usage(m, mm);
84}
85
86unsigned long task_vsize(struct mm_struct *mm)
87{
88 return PAGE_SIZE * mm->total_vm;
89}
90
91unsigned long task_statm(struct mm_struct *mm,
92 unsigned long *shared, unsigned long *text,
93 unsigned long *data, unsigned long *resident)
94{
95 *shared = get_mm_counter(mm, MM_FILEPAGES) +
96 get_mm_counter(mm, MM_SHMEMPAGES);
97 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
98 >> PAGE_SHIFT;
99 *data = mm->data_vm + mm->stack_vm;
100 *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
101 return mm->total_vm;
102}
103
104#ifdef CONFIG_NUMA
105/*
106 * Save get_task_policy() for show_numa_map().
107 */
108static void hold_task_mempolicy(struct proc_maps_private *priv)
109{
110 struct task_struct *task = priv->task;
111
112 task_lock(task);
113 priv->task_mempolicy = get_task_policy(task);
114 mpol_get(priv->task_mempolicy);
115 task_unlock(task);
116}
117static void release_task_mempolicy(struct proc_maps_private *priv)
118{
119 mpol_put(priv->task_mempolicy);
120}
121#else
122static void hold_task_mempolicy(struct proc_maps_private *priv)
123{
124}
125static void release_task_mempolicy(struct proc_maps_private *priv)
126{
127}
128#endif
129
130static void vma_stop(struct proc_maps_private *priv)
131{
132 struct mm_struct *mm = priv->mm;
133
134 release_task_mempolicy(priv);
135 up_read(&mm->mmap_sem);
136 mmput(mm);
137}
138
139static struct vm_area_struct *
140m_next_vma(struct proc_maps_private *priv, struct vm_area_struct *vma)
141{
142 if (vma == priv->tail_vma)
143 return NULL;
144 return vma->vm_next ?: priv->tail_vma;
145}
146
147static void m_cache_vma(struct seq_file *m, struct vm_area_struct *vma)
148{
149 if (m->count < m->size) /* vma is copied successfully */
150 m->version = m_next_vma(m->private, vma) ? vma->vm_start : -1UL;
151}
152
153static void *m_start(struct seq_file *m, loff_t *ppos)
154{
155 struct proc_maps_private *priv = m->private;
156 unsigned long last_addr = m->version;
157 struct mm_struct *mm;
158 struct vm_area_struct *vma;
159 unsigned int pos = *ppos;
160
161 /* See m_cache_vma(). Zero at the start or after lseek. */
162 if (last_addr == -1UL)
163 return NULL;
164
165 priv->task = get_proc_task(priv->inode);
166 if (!priv->task)
167 return ERR_PTR(-ESRCH);
168
169 mm = priv->mm;
170 if (!mm || !atomic_inc_not_zero(&mm->mm_users))
171 return NULL;
172
173 down_read(&mm->mmap_sem);
174 hold_task_mempolicy(priv);
175 priv->tail_vma = get_gate_vma(mm);
176
177 if (last_addr) {
178 vma = find_vma(mm, last_addr);
179 if (vma && (vma = m_next_vma(priv, vma)))
180 return vma;
181 }
182
183 m->version = 0;
184 if (pos < mm->map_count) {
185 for (vma = mm->mmap; pos; pos--) {
186 m->version = vma->vm_start;
187 vma = vma->vm_next;
188 }
189 return vma;
190 }
191
192 /* we do not bother to update m->version in this case */
193 if (pos == mm->map_count && priv->tail_vma)
194 return priv->tail_vma;
195
196 vma_stop(priv);
197 return NULL;
198}
199
200static void *m_next(struct seq_file *m, void *v, loff_t *pos)
201{
202 struct proc_maps_private *priv = m->private;
203 struct vm_area_struct *next;
204
205 (*pos)++;
206 next = m_next_vma(priv, v);
207 if (!next)
208 vma_stop(priv);
209 return next;
210}
211
212static void m_stop(struct seq_file *m, void *v)
213{
214 struct proc_maps_private *priv = m->private;
215
216 if (!IS_ERR_OR_NULL(v))
217 vma_stop(priv);
218 if (priv->task) {
219 put_task_struct(priv->task);
220 priv->task = NULL;
221 }
222}
223
224static int proc_maps_open(struct inode *inode, struct file *file,
225 const struct seq_operations *ops, int psize)
226{
227 struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
228
229 if (!priv)
230 return -ENOMEM;
231
232 priv->inode = inode;
233 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
234 if (IS_ERR(priv->mm)) {
235 int err = PTR_ERR(priv->mm);
236
237 seq_release_private(inode, file);
238 return err;
239 }
240
241 return 0;
242}
243
244static int proc_map_release(struct inode *inode, struct file *file)
245{
246 struct seq_file *seq = file->private_data;
247 struct proc_maps_private *priv = seq->private;
248
249 if (priv->mm)
250 mmdrop(priv->mm);
251
252 return seq_release_private(inode, file);
253}
254
255static int do_maps_open(struct inode *inode, struct file *file,
256 const struct seq_operations *ops)
257{
258 return proc_maps_open(inode, file, ops,
259 sizeof(struct proc_maps_private));
260}
261
262/*
263 * Indicate if the VMA is a stack for the given task; for
264 * /proc/PID/maps that is the stack of the main task.
265 */
266static int is_stack(struct proc_maps_private *priv,
267 struct vm_area_struct *vma, int is_pid)
268{
269 int stack = 0;
270
271 if (is_pid) {
272 stack = vma->vm_start <= vma->vm_mm->start_stack &&
273 vma->vm_end >= vma->vm_mm->start_stack;
274 } else {
275 struct inode *inode = priv->inode;
276 struct task_struct *task;
277
278 rcu_read_lock();
279 task = pid_task(proc_pid(inode), PIDTYPE_PID);
280 if (task)
281 stack = vma_is_stack_for_task(vma, task);
282 rcu_read_unlock();
283 }
284 return stack;
285}
286
287static void
288show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
289{
290 struct mm_struct *mm = vma->vm_mm;
291 struct file *file = vma->vm_file;
292 struct proc_maps_private *priv = m->private;
293 vm_flags_t flags = vma->vm_flags;
294 unsigned long ino = 0;
295 unsigned long long pgoff = 0;
296 unsigned long start, end;
297 dev_t dev = 0;
298 const char *name = NULL;
299
300 if (file) {
301 struct inode *inode = file_inode(vma->vm_file);
302 dev = inode->i_sb->s_dev;
303 ino = inode->i_ino;
304 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
305 }
306
307 /* We don't show the stack guard page in /proc/maps */
308 start = vma->vm_start;
309 if (stack_guard_page_start(vma, start))
310 start += PAGE_SIZE;
311 end = vma->vm_end;
312 if (stack_guard_page_end(vma, end))
313 end -= PAGE_SIZE;
314
315 seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
316 seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu ",
317 start,
318 end,
319 flags & VM_READ ? 'r' : '-',
320 flags & VM_WRITE ? 'w' : '-',
321 flags & VM_EXEC ? 'x' : '-',
322 flags & VM_MAYSHARE ? 's' : 'p',
323 pgoff,
324 MAJOR(dev), MINOR(dev), ino);
325
326 /*
327 * Print the dentry name for named mappings, and a
328 * special [heap] marker for the heap:
329 */
330 if (file) {
331 seq_pad(m, ' ');
332 seq_file_path(m, file, "\n");
333 goto done;
334 }
335
336 if (vma->vm_ops && vma->vm_ops->name) {
337 name = vma->vm_ops->name(vma);
338 if (name)
339 goto done;
340 }
341
342 name = arch_vma_name(vma);
343 if (!name) {
344 if (!mm) {
345 name = "[vdso]";
346 goto done;
347 }
348
349 if (vma->vm_start <= mm->brk &&
350 vma->vm_end >= mm->start_brk) {
351 name = "[heap]";
352 goto done;
353 }
354
355 if (is_stack(priv, vma, is_pid))
356 name = "[stack]";
357 }
358
359done:
360 if (name) {
361 seq_pad(m, ' ');
362 seq_puts(m, name);
363 }
364 seq_putc(m, '\n');
365}
366
367static int show_map(struct seq_file *m, void *v, int is_pid)
368{
369 show_map_vma(m, v, is_pid);
370 m_cache_vma(m, v);
371 return 0;
372}
373
374static int show_pid_map(struct seq_file *m, void *v)
375{
376 return show_map(m, v, 1);
377}
378
379static int show_tid_map(struct seq_file *m, void *v)
380{
381 return show_map(m, v, 0);
382}
383
384static const struct seq_operations proc_pid_maps_op = {
385 .start = m_start,
386 .next = m_next,
387 .stop = m_stop,
388 .show = show_pid_map
389};
390
391static const struct seq_operations proc_tid_maps_op = {
392 .start = m_start,
393 .next = m_next,
394 .stop = m_stop,
395 .show = show_tid_map
396};
397
398static int pid_maps_open(struct inode *inode, struct file *file)
399{
400 return do_maps_open(inode, file, &proc_pid_maps_op);
401}
402
403static int tid_maps_open(struct inode *inode, struct file *file)
404{
405 return do_maps_open(inode, file, &proc_tid_maps_op);
406}
407
408const struct file_operations proc_pid_maps_operations = {
409 .open = pid_maps_open,
410 .read = seq_read,
411 .llseek = seq_lseek,
412 .release = proc_map_release,
413};
414
415const struct file_operations proc_tid_maps_operations = {
416 .open = tid_maps_open,
417 .read = seq_read,
418 .llseek = seq_lseek,
419 .release = proc_map_release,
420};
421
422/*
423 * Proportional Set Size(PSS): my share of RSS.
424 *
425 * PSS of a process is the count of pages it has in memory, where each
426 * page is divided by the number of processes sharing it. So if a
427 * process has 1000 pages all to itself, and 1000 shared with one other
428 * process, its PSS will be 1500.
429 *
430 * To keep (accumulated) division errors low, we adopt a 64bit
431 * fixed-point pss counter to minimize division errors. So (pss >>
432 * PSS_SHIFT) would be the real byte count.
433 *
434 * A shift of 12 before division means (assuming 4K page size):
435 * - 1M 3-user-pages add up to 8KB errors;
436 * - supports mapcount up to 2^24, or 16M;
437 * - supports PSS up to 2^52 bytes, or 4PB.
438 */
439#define PSS_SHIFT 12
440
441#ifdef CONFIG_PROC_PAGE_MONITOR
442struct mem_size_stats {
443 unsigned long resident;
444 unsigned long shared_clean;
445 unsigned long shared_dirty;
446 unsigned long private_clean;
447 unsigned long private_dirty;
448 unsigned long referenced;
449 unsigned long anonymous;
450 unsigned long anonymous_thp;
451 unsigned long swap;
452 unsigned long shared_hugetlb;
453 unsigned long private_hugetlb;
454 u64 pss;
455 u64 swap_pss;
456 bool check_shmem_swap;
457};
458
459static void smaps_account(struct mem_size_stats *mss, struct page *page,
460 bool compound, bool young, bool dirty)
461{
462 int i, nr = compound ? 1 << compound_order(page) : 1;
463 unsigned long size = nr * PAGE_SIZE;
464
465 if (PageAnon(page))
466 mss->anonymous += size;
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 (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap
547 && pte_none(*pte))) {
548 page = find_get_entry(vma->vm_file->f_mapping,
549 linear_page_index(vma, addr));
550 if (!page)
551 return;
552
553 if (radix_tree_exceptional_entry(page))
554 mss->swap += PAGE_SIZE;
555 else
556 put_page(page);
557
558 return;
559 }
560
561 if (!page)
562 return;
563
564 smaps_account(mss, page, false, pte_young(*pte), pte_dirty(*pte));
565}
566
567#ifdef CONFIG_TRANSPARENT_HUGEPAGE
568static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
569 struct mm_walk *walk)
570{
571 struct mem_size_stats *mss = walk->private;
572 struct vm_area_struct *vma = walk->vma;
573 struct page *page;
574
575 /* FOLL_DUMP will return -EFAULT on huge zero page */
576 page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
577 if (IS_ERR_OR_NULL(page))
578 return;
579 mss->anonymous_thp += HPAGE_PMD_SIZE;
580 smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd));
581}
582#else
583static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
584 struct mm_walk *walk)
585{
586}
587#endif
588
589static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
590 struct mm_walk *walk)
591{
592 struct vm_area_struct *vma = walk->vma;
593 pte_t *pte;
594 spinlock_t *ptl;
595
596 ptl = pmd_trans_huge_lock(pmd, vma);
597 if (ptl) {
598 smaps_pmd_entry(pmd, addr, walk);
599 spin_unlock(ptl);
600 return 0;
601 }
602
603 if (pmd_trans_unstable(pmd))
604 return 0;
605 /*
606 * The mmap_sem held all the way back in m_start() is what
607 * keeps khugepaged out of here and from collapsing things
608 * in here.
609 */
610 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
611 for (; addr != end; pte++, addr += PAGE_SIZE)
612 smaps_pte_entry(pte, addr, walk);
613 pte_unmap_unlock(pte - 1, ptl);
614 cond_resched();
615 return 0;
616}
617
618static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
619{
620 /*
621 * Don't forget to update Documentation/ on changes.
622 */
623 static const char mnemonics[BITS_PER_LONG][2] = {
624 /*
625 * In case if we meet a flag we don't know about.
626 */
627 [0 ... (BITS_PER_LONG-1)] = "??",
628
629 [ilog2(VM_READ)] = "rd",
630 [ilog2(VM_WRITE)] = "wr",
631 [ilog2(VM_EXEC)] = "ex",
632 [ilog2(VM_SHARED)] = "sh",
633 [ilog2(VM_MAYREAD)] = "mr",
634 [ilog2(VM_MAYWRITE)] = "mw",
635 [ilog2(VM_MAYEXEC)] = "me",
636 [ilog2(VM_MAYSHARE)] = "ms",
637 [ilog2(VM_GROWSDOWN)] = "gd",
638 [ilog2(VM_PFNMAP)] = "pf",
639 [ilog2(VM_DENYWRITE)] = "dw",
640#ifdef CONFIG_X86_INTEL_MPX
641 [ilog2(VM_MPX)] = "mp",
642#endif
643 [ilog2(VM_LOCKED)] = "lo",
644 [ilog2(VM_IO)] = "io",
645 [ilog2(VM_SEQ_READ)] = "sr",
646 [ilog2(VM_RAND_READ)] = "rr",
647 [ilog2(VM_DONTCOPY)] = "dc",
648 [ilog2(VM_DONTEXPAND)] = "de",
649 [ilog2(VM_ACCOUNT)] = "ac",
650 [ilog2(VM_NORESERVE)] = "nr",
651 [ilog2(VM_HUGETLB)] = "ht",
652 [ilog2(VM_ARCH_1)] = "ar",
653 [ilog2(VM_DONTDUMP)] = "dd",
654#ifdef CONFIG_MEM_SOFT_DIRTY
655 [ilog2(VM_SOFTDIRTY)] = "sd",
656#endif
657 [ilog2(VM_MIXEDMAP)] = "mm",
658 [ilog2(VM_HUGEPAGE)] = "hg",
659 [ilog2(VM_NOHUGEPAGE)] = "nh",
660 [ilog2(VM_MERGEABLE)] = "mg",
661 [ilog2(VM_UFFD_MISSING)]= "um",
662 [ilog2(VM_UFFD_WP)] = "uw",
663#ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
664 /* These come out via ProtectionKey: */
665 [ilog2(VM_PKEY_BIT0)] = "",
666 [ilog2(VM_PKEY_BIT1)] = "",
667 [ilog2(VM_PKEY_BIT2)] = "",
668 [ilog2(VM_PKEY_BIT3)] = "",
669#endif
670 };
671 size_t i;
672
673 seq_puts(m, "VmFlags: ");
674 for (i = 0; i < BITS_PER_LONG; i++) {
675 if (!mnemonics[i][0])
676 continue;
677 if (vma->vm_flags & (1UL << i)) {
678 seq_printf(m, "%c%c ",
679 mnemonics[i][0], mnemonics[i][1]);
680 }
681 }
682 seq_putc(m, '\n');
683}
684
685#ifdef CONFIG_HUGETLB_PAGE
686static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
687 unsigned long addr, unsigned long end,
688 struct mm_walk *walk)
689{
690 struct mem_size_stats *mss = walk->private;
691 struct vm_area_struct *vma = walk->vma;
692 struct page *page = NULL;
693
694 if (pte_present(*pte)) {
695 page = vm_normal_page(vma, addr, *pte);
696 } else if (is_swap_pte(*pte)) {
697 swp_entry_t swpent = pte_to_swp_entry(*pte);
698
699 if (is_migration_entry(swpent))
700 page = migration_entry_to_page(swpent);
701 }
702 if (page) {
703 int mapcount = page_mapcount(page);
704
705 if (mapcount >= 2)
706 mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
707 else
708 mss->private_hugetlb += huge_page_size(hstate_vma(vma));
709 }
710 return 0;
711}
712#endif /* HUGETLB_PAGE */
713
714void __weak arch_show_smap(struct seq_file *m, struct vm_area_struct *vma)
715{
716}
717
718static int show_smap(struct seq_file *m, void *v, int is_pid)
719{
720 struct vm_area_struct *vma = v;
721 struct mem_size_stats mss;
722 struct mm_walk smaps_walk = {
723 .pmd_entry = smaps_pte_range,
724#ifdef CONFIG_HUGETLB_PAGE
725 .hugetlb_entry = smaps_hugetlb_range,
726#endif
727 .mm = vma->vm_mm,
728 .private = &mss,
729 };
730
731 memset(&mss, 0, sizeof mss);
732
733#ifdef CONFIG_SHMEM
734 if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
735 /*
736 * For shared or readonly shmem mappings we know that all
737 * swapped out pages belong to the shmem object, and we can
738 * obtain the swap value much more efficiently. For private
739 * writable mappings, we might have COW pages that are
740 * not affected by the parent swapped out pages of the shmem
741 * object, so we have to distinguish them during the page walk.
742 * Unless we know that the shmem object (or the part mapped by
743 * our VMA) has no swapped out pages at all.
744 */
745 unsigned long shmem_swapped = shmem_swap_usage(vma);
746
747 if (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
748 !(vma->vm_flags & VM_WRITE)) {
749 mss.swap = shmem_swapped;
750 } else {
751 mss.check_shmem_swap = true;
752 smaps_walk.pte_hole = smaps_pte_hole;
753 }
754 }
755#endif
756
757 /* mmap_sem is held in m_start */
758 walk_page_vma(vma, &smaps_walk);
759
760 show_map_vma(m, vma, is_pid);
761
762 seq_printf(m,
763 "Size: %8lu kB\n"
764 "Rss: %8lu kB\n"
765 "Pss: %8lu kB\n"
766 "Shared_Clean: %8lu kB\n"
767 "Shared_Dirty: %8lu kB\n"
768 "Private_Clean: %8lu kB\n"
769 "Private_Dirty: %8lu kB\n"
770 "Referenced: %8lu kB\n"
771 "Anonymous: %8lu kB\n"
772 "AnonHugePages: %8lu kB\n"
773 "Shared_Hugetlb: %8lu kB\n"
774 "Private_Hugetlb: %7lu kB\n"
775 "Swap: %8lu kB\n"
776 "SwapPss: %8lu kB\n"
777 "KernelPageSize: %8lu kB\n"
778 "MMUPageSize: %8lu kB\n"
779 "Locked: %8lu kB\n",
780 (vma->vm_end - vma->vm_start) >> 10,
781 mss.resident >> 10,
782 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
783 mss.shared_clean >> 10,
784 mss.shared_dirty >> 10,
785 mss.private_clean >> 10,
786 mss.private_dirty >> 10,
787 mss.referenced >> 10,
788 mss.anonymous >> 10,
789 mss.anonymous_thp >> 10,
790 mss.shared_hugetlb >> 10,
791 mss.private_hugetlb >> 10,
792 mss.swap >> 10,
793 (unsigned long)(mss.swap_pss >> (10 + PSS_SHIFT)),
794 vma_kernel_pagesize(vma) >> 10,
795 vma_mmu_pagesize(vma) >> 10,
796 (vma->vm_flags & VM_LOCKED) ?
797 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
798
799 arch_show_smap(m, vma);
800 show_smap_vma_flags(m, vma);
801 m_cache_vma(m, vma);
802 return 0;
803}
804
805static int show_pid_smap(struct seq_file *m, void *v)
806{
807 return show_smap(m, v, 1);
808}
809
810static int show_tid_smap(struct seq_file *m, void *v)
811{
812 return show_smap(m, v, 0);
813}
814
815static const struct seq_operations proc_pid_smaps_op = {
816 .start = m_start,
817 .next = m_next,
818 .stop = m_stop,
819 .show = show_pid_smap
820};
821
822static const struct seq_operations proc_tid_smaps_op = {
823 .start = m_start,
824 .next = m_next,
825 .stop = m_stop,
826 .show = show_tid_smap
827};
828
829static int pid_smaps_open(struct inode *inode, struct file *file)
830{
831 return do_maps_open(inode, file, &proc_pid_smaps_op);
832}
833
834static int tid_smaps_open(struct inode *inode, struct file *file)
835{
836 return do_maps_open(inode, file, &proc_tid_smaps_op);
837}
838
839const struct file_operations proc_pid_smaps_operations = {
840 .open = pid_smaps_open,
841 .read = seq_read,
842 .llseek = seq_lseek,
843 .release = proc_map_release,
844};
845
846const struct file_operations proc_tid_smaps_operations = {
847 .open = tid_smaps_open,
848 .read = seq_read,
849 .llseek = seq_lseek,
850 .release = proc_map_release,
851};
852
853enum clear_refs_types {
854 CLEAR_REFS_ALL = 1,
855 CLEAR_REFS_ANON,
856 CLEAR_REFS_MAPPED,
857 CLEAR_REFS_SOFT_DIRTY,
858 CLEAR_REFS_MM_HIWATER_RSS,
859 CLEAR_REFS_LAST,
860};
861
862struct clear_refs_private {
863 enum clear_refs_types type;
864};
865
866#ifdef CONFIG_MEM_SOFT_DIRTY
867static inline void clear_soft_dirty(struct vm_area_struct *vma,
868 unsigned long addr, pte_t *pte)
869{
870 /*
871 * The soft-dirty tracker uses #PF-s to catch writes
872 * to pages, so write-protect the pte as well. See the
873 * Documentation/vm/soft-dirty.txt for full description
874 * of how soft-dirty works.
875 */
876 pte_t ptent = *pte;
877
878 if (pte_present(ptent)) {
879 ptent = ptep_modify_prot_start(vma->vm_mm, addr, pte);
880 ptent = pte_wrprotect(ptent);
881 ptent = pte_clear_soft_dirty(ptent);
882 ptep_modify_prot_commit(vma->vm_mm, addr, pte, ptent);
883 } else if (is_swap_pte(ptent)) {
884 ptent = pte_swp_clear_soft_dirty(ptent);
885 set_pte_at(vma->vm_mm, addr, pte, ptent);
886 }
887}
888#else
889static inline void clear_soft_dirty(struct vm_area_struct *vma,
890 unsigned long addr, pte_t *pte)
891{
892}
893#endif
894
895#if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
896static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
897 unsigned long addr, pmd_t *pmdp)
898{
899 pmd_t pmd = pmdp_huge_get_and_clear(vma->vm_mm, addr, pmdp);
900
901 pmd = pmd_wrprotect(pmd);
902 pmd = pmd_clear_soft_dirty(pmd);
903
904 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
905}
906#else
907static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
908 unsigned long addr, pmd_t *pmdp)
909{
910}
911#endif
912
913static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
914 unsigned long end, struct mm_walk *walk)
915{
916 struct clear_refs_private *cp = walk->private;
917 struct vm_area_struct *vma = walk->vma;
918 pte_t *pte, ptent;
919 spinlock_t *ptl;
920 struct page *page;
921
922 ptl = pmd_trans_huge_lock(pmd, vma);
923 if (ptl) {
924 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
925 clear_soft_dirty_pmd(vma, addr, pmd);
926 goto out;
927 }
928
929 page = pmd_page(*pmd);
930
931 /* Clear accessed and referenced bits. */
932 pmdp_test_and_clear_young(vma, addr, pmd);
933 test_and_clear_page_young(page);
934 ClearPageReferenced(page);
935out:
936 spin_unlock(ptl);
937 return 0;
938 }
939
940 if (pmd_trans_unstable(pmd))
941 return 0;
942
943 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
944 for (; addr != end; pte++, addr += PAGE_SIZE) {
945 ptent = *pte;
946
947 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
948 clear_soft_dirty(vma, addr, pte);
949 continue;
950 }
951
952 if (!pte_present(ptent))
953 continue;
954
955 page = vm_normal_page(vma, addr, ptent);
956 if (!page)
957 continue;
958
959 /* Clear accessed and referenced bits. */
960 ptep_test_and_clear_young(vma, addr, pte);
961 test_and_clear_page_young(page);
962 ClearPageReferenced(page);
963 }
964 pte_unmap_unlock(pte - 1, ptl);
965 cond_resched();
966 return 0;
967}
968
969static int clear_refs_test_walk(unsigned long start, unsigned long end,
970 struct mm_walk *walk)
971{
972 struct clear_refs_private *cp = walk->private;
973 struct vm_area_struct *vma = walk->vma;
974
975 if (vma->vm_flags & VM_PFNMAP)
976 return 1;
977
978 /*
979 * Writing 1 to /proc/pid/clear_refs affects all pages.
980 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
981 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
982 * Writing 4 to /proc/pid/clear_refs affects all pages.
983 */
984 if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
985 return 1;
986 if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
987 return 1;
988 return 0;
989}
990
991static ssize_t clear_refs_write(struct file *file, const char __user *buf,
992 size_t count, loff_t *ppos)
993{
994 struct task_struct *task;
995 char buffer[PROC_NUMBUF];
996 struct mm_struct *mm;
997 struct vm_area_struct *vma;
998 enum clear_refs_types type;
999 int itype;
1000 int rv;
1001
1002 memset(buffer, 0, sizeof(buffer));
1003 if (count > sizeof(buffer) - 1)
1004 count = sizeof(buffer) - 1;
1005 if (copy_from_user(buffer, buf, count))
1006 return -EFAULT;
1007 rv = kstrtoint(strstrip(buffer), 10, &itype);
1008 if (rv < 0)
1009 return rv;
1010 type = (enum clear_refs_types)itype;
1011 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1012 return -EINVAL;
1013
1014 task = get_proc_task(file_inode(file));
1015 if (!task)
1016 return -ESRCH;
1017 mm = get_task_mm(task);
1018 if (mm) {
1019 struct clear_refs_private cp = {
1020 .type = type,
1021 };
1022 struct mm_walk clear_refs_walk = {
1023 .pmd_entry = clear_refs_pte_range,
1024 .test_walk = clear_refs_test_walk,
1025 .mm = mm,
1026 .private = &cp,
1027 };
1028
1029 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1030 /*
1031 * Writing 5 to /proc/pid/clear_refs resets the peak
1032 * resident set size to this mm's current rss value.
1033 */
1034 down_write(&mm->mmap_sem);
1035 reset_mm_hiwater_rss(mm);
1036 up_write(&mm->mmap_sem);
1037 goto out_mm;
1038 }
1039
1040 down_read(&mm->mmap_sem);
1041 if (type == CLEAR_REFS_SOFT_DIRTY) {
1042 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1043 if (!(vma->vm_flags & VM_SOFTDIRTY))
1044 continue;
1045 up_read(&mm->mmap_sem);
1046 down_write(&mm->mmap_sem);
1047 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1048 vma->vm_flags &= ~VM_SOFTDIRTY;
1049 vma_set_page_prot(vma);
1050 }
1051 downgrade_write(&mm->mmap_sem);
1052 break;
1053 }
1054 mmu_notifier_invalidate_range_start(mm, 0, -1);
1055 }
1056 walk_page_range(0, ~0UL, &clear_refs_walk);
1057 if (type == CLEAR_REFS_SOFT_DIRTY)
1058 mmu_notifier_invalidate_range_end(mm, 0, -1);
1059 flush_tlb_mm(mm);
1060 up_read(&mm->mmap_sem);
1061out_mm:
1062 mmput(mm);
1063 }
1064 put_task_struct(task);
1065
1066 return count;
1067}
1068
1069const struct file_operations proc_clear_refs_operations = {
1070 .write = clear_refs_write,
1071 .llseek = noop_llseek,
1072};
1073
1074typedef struct {
1075 u64 pme;
1076} pagemap_entry_t;
1077
1078struct pagemapread {
1079 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */
1080 pagemap_entry_t *buffer;
1081 bool show_pfn;
1082};
1083
1084#define PAGEMAP_WALK_SIZE (PMD_SIZE)
1085#define PAGEMAP_WALK_MASK (PMD_MASK)
1086
1087#define PM_ENTRY_BYTES sizeof(pagemap_entry_t)
1088#define PM_PFRAME_BITS 55
1089#define PM_PFRAME_MASK GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1090#define PM_SOFT_DIRTY BIT_ULL(55)
1091#define PM_MMAP_EXCLUSIVE BIT_ULL(56)
1092#define PM_FILE BIT_ULL(61)
1093#define PM_SWAP BIT_ULL(62)
1094#define PM_PRESENT BIT_ULL(63)
1095
1096#define PM_END_OF_BUFFER 1
1097
1098static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1099{
1100 return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1101}
1102
1103static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1104 struct pagemapread *pm)
1105{
1106 pm->buffer[pm->pos++] = *pme;
1107 if (pm->pos >= pm->len)
1108 return PM_END_OF_BUFFER;
1109 return 0;
1110}
1111
1112static int pagemap_pte_hole(unsigned long start, unsigned long end,
1113 struct mm_walk *walk)
1114{
1115 struct pagemapread *pm = walk->private;
1116 unsigned long addr = start;
1117 int err = 0;
1118
1119 while (addr < end) {
1120 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1121 pagemap_entry_t pme = make_pme(0, 0);
1122 /* End of address space hole, which we mark as non-present. */
1123 unsigned long hole_end;
1124
1125 if (vma)
1126 hole_end = min(end, vma->vm_start);
1127 else
1128 hole_end = end;
1129
1130 for (; addr < hole_end; addr += PAGE_SIZE) {
1131 err = add_to_pagemap(addr, &pme, pm);
1132 if (err)
1133 goto out;
1134 }
1135
1136 if (!vma)
1137 break;
1138
1139 /* Addresses in the VMA. */
1140 if (vma->vm_flags & VM_SOFTDIRTY)
1141 pme = make_pme(0, PM_SOFT_DIRTY);
1142 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1143 err = add_to_pagemap(addr, &pme, pm);
1144 if (err)
1145 goto out;
1146 }
1147 }
1148out:
1149 return err;
1150}
1151
1152static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1153 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1154{
1155 u64 frame = 0, flags = 0;
1156 struct page *page = NULL;
1157
1158 if (pte_present(pte)) {
1159 if (pm->show_pfn)
1160 frame = pte_pfn(pte);
1161 flags |= PM_PRESENT;
1162 page = vm_normal_page(vma, addr, pte);
1163 if (pte_soft_dirty(pte))
1164 flags |= PM_SOFT_DIRTY;
1165 } else if (is_swap_pte(pte)) {
1166 swp_entry_t entry;
1167 if (pte_swp_soft_dirty(pte))
1168 flags |= PM_SOFT_DIRTY;
1169 entry = pte_to_swp_entry(pte);
1170 frame = swp_type(entry) |
1171 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1172 flags |= PM_SWAP;
1173 if (is_migration_entry(entry))
1174 page = migration_entry_to_page(entry);
1175 }
1176
1177 if (page && !PageAnon(page))
1178 flags |= PM_FILE;
1179 if (page && page_mapcount(page) == 1)
1180 flags |= PM_MMAP_EXCLUSIVE;
1181 if (vma->vm_flags & VM_SOFTDIRTY)
1182 flags |= PM_SOFT_DIRTY;
1183
1184 return make_pme(frame, flags);
1185}
1186
1187static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1188 struct mm_walk *walk)
1189{
1190 struct vm_area_struct *vma = walk->vma;
1191 struct pagemapread *pm = walk->private;
1192 spinlock_t *ptl;
1193 pte_t *pte, *orig_pte;
1194 int err = 0;
1195
1196#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1197 ptl = pmd_trans_huge_lock(pmdp, vma);
1198 if (ptl) {
1199 u64 flags = 0, frame = 0;
1200 pmd_t pmd = *pmdp;
1201
1202 if ((vma->vm_flags & VM_SOFTDIRTY) || pmd_soft_dirty(pmd))
1203 flags |= PM_SOFT_DIRTY;
1204
1205 /*
1206 * Currently pmd for thp is always present because thp
1207 * can not be swapped-out, migrated, or HWPOISONed
1208 * (split in such cases instead.)
1209 * This if-check is just to prepare for future implementation.
1210 */
1211 if (pmd_present(pmd)) {
1212 struct page *page = pmd_page(pmd);
1213
1214 if (page_mapcount(page) == 1)
1215 flags |= PM_MMAP_EXCLUSIVE;
1216
1217 flags |= PM_PRESENT;
1218 if (pm->show_pfn)
1219 frame = pmd_pfn(pmd) +
1220 ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1221 }
1222
1223 for (; addr != end; addr += PAGE_SIZE) {
1224 pagemap_entry_t pme = make_pme(frame, flags);
1225
1226 err = add_to_pagemap(addr, &pme, pm);
1227 if (err)
1228 break;
1229 if (pm->show_pfn && (flags & PM_PRESENT))
1230 frame++;
1231 }
1232 spin_unlock(ptl);
1233 return err;
1234 }
1235
1236 if (pmd_trans_unstable(pmdp))
1237 return 0;
1238#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1239
1240 /*
1241 * We can assume that @vma always points to a valid one and @end never
1242 * goes beyond vma->vm_end.
1243 */
1244 orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1245 for (; addr < end; pte++, addr += PAGE_SIZE) {
1246 pagemap_entry_t pme;
1247
1248 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1249 err = add_to_pagemap(addr, &pme, pm);
1250 if (err)
1251 break;
1252 }
1253 pte_unmap_unlock(orig_pte, ptl);
1254
1255 cond_resched();
1256
1257 return err;
1258}
1259
1260#ifdef CONFIG_HUGETLB_PAGE
1261/* This function walks within one hugetlb entry in the single call */
1262static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1263 unsigned long addr, unsigned long end,
1264 struct mm_walk *walk)
1265{
1266 struct pagemapread *pm = walk->private;
1267 struct vm_area_struct *vma = walk->vma;
1268 u64 flags = 0, frame = 0;
1269 int err = 0;
1270 pte_t pte;
1271
1272 if (vma->vm_flags & VM_SOFTDIRTY)
1273 flags |= PM_SOFT_DIRTY;
1274
1275 pte = huge_ptep_get(ptep);
1276 if (pte_present(pte)) {
1277 struct page *page = pte_page(pte);
1278
1279 if (!PageAnon(page))
1280 flags |= PM_FILE;
1281
1282 if (page_mapcount(page) == 1)
1283 flags |= PM_MMAP_EXCLUSIVE;
1284
1285 flags |= PM_PRESENT;
1286 if (pm->show_pfn)
1287 frame = pte_pfn(pte) +
1288 ((addr & ~hmask) >> PAGE_SHIFT);
1289 }
1290
1291 for (; addr != end; addr += PAGE_SIZE) {
1292 pagemap_entry_t pme = make_pme(frame, flags);
1293
1294 err = add_to_pagemap(addr, &pme, pm);
1295 if (err)
1296 return err;
1297 if (pm->show_pfn && (flags & PM_PRESENT))
1298 frame++;
1299 }
1300
1301 cond_resched();
1302
1303 return err;
1304}
1305#endif /* HUGETLB_PAGE */
1306
1307/*
1308 * /proc/pid/pagemap - an array mapping virtual pages to pfns
1309 *
1310 * For each page in the address space, this file contains one 64-bit entry
1311 * consisting of the following:
1312 *
1313 * Bits 0-54 page frame number (PFN) if present
1314 * Bits 0-4 swap type if swapped
1315 * Bits 5-54 swap offset if swapped
1316 * Bit 55 pte is soft-dirty (see Documentation/vm/soft-dirty.txt)
1317 * Bit 56 page exclusively mapped
1318 * Bits 57-60 zero
1319 * Bit 61 page is file-page or shared-anon
1320 * Bit 62 page swapped
1321 * Bit 63 page present
1322 *
1323 * If the page is not present but in swap, then the PFN contains an
1324 * encoding of the swap file number and the page's offset into the
1325 * swap. Unmapped pages return a null PFN. This allows determining
1326 * precisely which pages are mapped (or in swap) and comparing mapped
1327 * pages between processes.
1328 *
1329 * Efficient users of this interface will use /proc/pid/maps to
1330 * determine which areas of memory are actually mapped and llseek to
1331 * skip over unmapped regions.
1332 */
1333static ssize_t pagemap_read(struct file *file, char __user *buf,
1334 size_t count, loff_t *ppos)
1335{
1336 struct mm_struct *mm = file->private_data;
1337 struct pagemapread pm;
1338 struct mm_walk pagemap_walk = {};
1339 unsigned long src;
1340 unsigned long svpfn;
1341 unsigned long start_vaddr;
1342 unsigned long end_vaddr;
1343 int ret = 0, copied = 0;
1344
1345 if (!mm || !atomic_inc_not_zero(&mm->mm_users))
1346 goto out;
1347
1348 ret = -EINVAL;
1349 /* file position must be aligned */
1350 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1351 goto out_mm;
1352
1353 ret = 0;
1354 if (!count)
1355 goto out_mm;
1356
1357 /* do not disclose physical addresses: attack vector */
1358 pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1359
1360 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1361 pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY);
1362 ret = -ENOMEM;
1363 if (!pm.buffer)
1364 goto out_mm;
1365
1366 pagemap_walk.pmd_entry = pagemap_pmd_range;
1367 pagemap_walk.pte_hole = pagemap_pte_hole;
1368#ifdef CONFIG_HUGETLB_PAGE
1369 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1370#endif
1371 pagemap_walk.mm = mm;
1372 pagemap_walk.private = ±
1373
1374 src = *ppos;
1375 svpfn = src / PM_ENTRY_BYTES;
1376 start_vaddr = svpfn << PAGE_SHIFT;
1377 end_vaddr = mm->task_size;
1378
1379 /* watch out for wraparound */
1380 if (svpfn > mm->task_size >> PAGE_SHIFT)
1381 start_vaddr = end_vaddr;
1382
1383 /*
1384 * The odds are that this will stop walking way
1385 * before end_vaddr, because the length of the
1386 * user buffer is tracked in "pm", and the walk
1387 * will stop when we hit the end of the buffer.
1388 */
1389 ret = 0;
1390 while (count && (start_vaddr < end_vaddr)) {
1391 int len;
1392 unsigned long end;
1393
1394 pm.pos = 0;
1395 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1396 /* overflow ? */
1397 if (end < start_vaddr || end > end_vaddr)
1398 end = end_vaddr;
1399 down_read(&mm->mmap_sem);
1400 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1401 up_read(&mm->mmap_sem);
1402 start_vaddr = end;
1403
1404 len = min(count, PM_ENTRY_BYTES * pm.pos);
1405 if (copy_to_user(buf, pm.buffer, len)) {
1406 ret = -EFAULT;
1407 goto out_free;
1408 }
1409 copied += len;
1410 buf += len;
1411 count -= len;
1412 }
1413 *ppos += copied;
1414 if (!ret || ret == PM_END_OF_BUFFER)
1415 ret = copied;
1416
1417out_free:
1418 kfree(pm.buffer);
1419out_mm:
1420 mmput(mm);
1421out:
1422 return ret;
1423}
1424
1425static int pagemap_open(struct inode *inode, struct file *file)
1426{
1427 struct mm_struct *mm;
1428
1429 mm = proc_mem_open(inode, PTRACE_MODE_READ);
1430 if (IS_ERR(mm))
1431 return PTR_ERR(mm);
1432 file->private_data = mm;
1433 return 0;
1434}
1435
1436static int pagemap_release(struct inode *inode, struct file *file)
1437{
1438 struct mm_struct *mm = file->private_data;
1439
1440 if (mm)
1441 mmdrop(mm);
1442 return 0;
1443}
1444
1445const struct file_operations proc_pagemap_operations = {
1446 .llseek = mem_lseek, /* borrow this */
1447 .read = pagemap_read,
1448 .open = pagemap_open,
1449 .release = pagemap_release,
1450};
1451#endif /* CONFIG_PROC_PAGE_MONITOR */
1452
1453#ifdef CONFIG_NUMA
1454
1455struct numa_maps {
1456 unsigned long pages;
1457 unsigned long anon;
1458 unsigned long active;
1459 unsigned long writeback;
1460 unsigned long mapcount_max;
1461 unsigned long dirty;
1462 unsigned long swapcache;
1463 unsigned long node[MAX_NUMNODES];
1464};
1465
1466struct numa_maps_private {
1467 struct proc_maps_private proc_maps;
1468 struct numa_maps md;
1469};
1470
1471static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1472 unsigned long nr_pages)
1473{
1474 int count = page_mapcount(page);
1475
1476 md->pages += nr_pages;
1477 if (pte_dirty || PageDirty(page))
1478 md->dirty += nr_pages;
1479
1480 if (PageSwapCache(page))
1481 md->swapcache += nr_pages;
1482
1483 if (PageActive(page) || PageUnevictable(page))
1484 md->active += nr_pages;
1485
1486 if (PageWriteback(page))
1487 md->writeback += nr_pages;
1488
1489 if (PageAnon(page))
1490 md->anon += nr_pages;
1491
1492 if (count > md->mapcount_max)
1493 md->mapcount_max = count;
1494
1495 md->node[page_to_nid(page)] += nr_pages;
1496}
1497
1498static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1499 unsigned long addr)
1500{
1501 struct page *page;
1502 int nid;
1503
1504 if (!pte_present(pte))
1505 return NULL;
1506
1507 page = vm_normal_page(vma, addr, pte);
1508 if (!page)
1509 return NULL;
1510
1511 if (PageReserved(page))
1512 return NULL;
1513
1514 nid = page_to_nid(page);
1515 if (!node_isset(nid, node_states[N_MEMORY]))
1516 return NULL;
1517
1518 return page;
1519}
1520
1521#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1522static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1523 struct vm_area_struct *vma,
1524 unsigned long addr)
1525{
1526 struct page *page;
1527 int nid;
1528
1529 if (!pmd_present(pmd))
1530 return NULL;
1531
1532 page = vm_normal_page_pmd(vma, addr, pmd);
1533 if (!page)
1534 return NULL;
1535
1536 if (PageReserved(page))
1537 return NULL;
1538
1539 nid = page_to_nid(page);
1540 if (!node_isset(nid, node_states[N_MEMORY]))
1541 return NULL;
1542
1543 return page;
1544}
1545#endif
1546
1547static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1548 unsigned long end, struct mm_walk *walk)
1549{
1550 struct numa_maps *md = walk->private;
1551 struct vm_area_struct *vma = walk->vma;
1552 spinlock_t *ptl;
1553 pte_t *orig_pte;
1554 pte_t *pte;
1555
1556#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1557 ptl = pmd_trans_huge_lock(pmd, vma);
1558 if (ptl) {
1559 struct page *page;
1560
1561 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1562 if (page)
1563 gather_stats(page, md, pmd_dirty(*pmd),
1564 HPAGE_PMD_SIZE/PAGE_SIZE);
1565 spin_unlock(ptl);
1566 return 0;
1567 }
1568
1569 if (pmd_trans_unstable(pmd))
1570 return 0;
1571#endif
1572 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1573 do {
1574 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1575 if (!page)
1576 continue;
1577 gather_stats(page, md, pte_dirty(*pte), 1);
1578
1579 } while (pte++, addr += PAGE_SIZE, addr != end);
1580 pte_unmap_unlock(orig_pte, ptl);
1581 return 0;
1582}
1583#ifdef CONFIG_HUGETLB_PAGE
1584static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1585 unsigned long addr, unsigned long end, struct mm_walk *walk)
1586{
1587 pte_t huge_pte = huge_ptep_get(pte);
1588 struct numa_maps *md;
1589 struct page *page;
1590
1591 if (!pte_present(huge_pte))
1592 return 0;
1593
1594 page = pte_page(huge_pte);
1595 if (!page)
1596 return 0;
1597
1598 md = walk->private;
1599 gather_stats(page, md, pte_dirty(huge_pte), 1);
1600 return 0;
1601}
1602
1603#else
1604static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1605 unsigned long addr, unsigned long end, struct mm_walk *walk)
1606{
1607 return 0;
1608}
1609#endif
1610
1611/*
1612 * Display pages allocated per node and memory policy via /proc.
1613 */
1614static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1615{
1616 struct numa_maps_private *numa_priv = m->private;
1617 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1618 struct vm_area_struct *vma = v;
1619 struct numa_maps *md = &numa_priv->md;
1620 struct file *file = vma->vm_file;
1621 struct mm_struct *mm = vma->vm_mm;
1622 struct mm_walk walk = {
1623 .hugetlb_entry = gather_hugetlb_stats,
1624 .pmd_entry = gather_pte_stats,
1625 .private = md,
1626 .mm = mm,
1627 };
1628 struct mempolicy *pol;
1629 char buffer[64];
1630 int nid;
1631
1632 if (!mm)
1633 return 0;
1634
1635 /* Ensure we start with an empty set of numa_maps statistics. */
1636 memset(md, 0, sizeof(*md));
1637
1638 pol = __get_vma_policy(vma, vma->vm_start);
1639 if (pol) {
1640 mpol_to_str(buffer, sizeof(buffer), pol);
1641 mpol_cond_put(pol);
1642 } else {
1643 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1644 }
1645
1646 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1647
1648 if (file) {
1649 seq_puts(m, " file=");
1650 seq_file_path(m, file, "\n\t= ");
1651 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1652 seq_puts(m, " heap");
1653 } else if (is_stack(proc_priv, vma, is_pid)) {
1654 seq_puts(m, " stack");
1655 }
1656
1657 if (is_vm_hugetlb_page(vma))
1658 seq_puts(m, " huge");
1659
1660 /* mmap_sem is held by m_start */
1661 walk_page_vma(vma, &walk);
1662
1663 if (!md->pages)
1664 goto out;
1665
1666 if (md->anon)
1667 seq_printf(m, " anon=%lu", md->anon);
1668
1669 if (md->dirty)
1670 seq_printf(m, " dirty=%lu", md->dirty);
1671
1672 if (md->pages != md->anon && md->pages != md->dirty)
1673 seq_printf(m, " mapped=%lu", md->pages);
1674
1675 if (md->mapcount_max > 1)
1676 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1677
1678 if (md->swapcache)
1679 seq_printf(m, " swapcache=%lu", md->swapcache);
1680
1681 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1682 seq_printf(m, " active=%lu", md->active);
1683
1684 if (md->writeback)
1685 seq_printf(m, " writeback=%lu", md->writeback);
1686
1687 for_each_node_state(nid, N_MEMORY)
1688 if (md->node[nid])
1689 seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1690
1691 seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1692out:
1693 seq_putc(m, '\n');
1694 m_cache_vma(m, vma);
1695 return 0;
1696}
1697
1698static int show_pid_numa_map(struct seq_file *m, void *v)
1699{
1700 return show_numa_map(m, v, 1);
1701}
1702
1703static int show_tid_numa_map(struct seq_file *m, void *v)
1704{
1705 return show_numa_map(m, v, 0);
1706}
1707
1708static const struct seq_operations proc_pid_numa_maps_op = {
1709 .start = m_start,
1710 .next = m_next,
1711 .stop = m_stop,
1712 .show = show_pid_numa_map,
1713};
1714
1715static const struct seq_operations proc_tid_numa_maps_op = {
1716 .start = m_start,
1717 .next = m_next,
1718 .stop = m_stop,
1719 .show = show_tid_numa_map,
1720};
1721
1722static int numa_maps_open(struct inode *inode, struct file *file,
1723 const struct seq_operations *ops)
1724{
1725 return proc_maps_open(inode, file, ops,
1726 sizeof(struct numa_maps_private));
1727}
1728
1729static int pid_numa_maps_open(struct inode *inode, struct file *file)
1730{
1731 return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1732}
1733
1734static int tid_numa_maps_open(struct inode *inode, struct file *file)
1735{
1736 return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1737}
1738
1739const struct file_operations proc_pid_numa_maps_operations = {
1740 .open = pid_numa_maps_open,
1741 .read = seq_read,
1742 .llseek = seq_lseek,
1743 .release = proc_map_release,
1744};
1745
1746const struct file_operations proc_tid_numa_maps_operations = {
1747 .open = tid_numa_maps_open,
1748 .read = seq_read,
1749 .llseek = seq_lseek,
1750 .release = proc_map_release,
1751};
1752#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#include <linux/page_idle.h>
17#include <linux/shmem_fs.h>
18
19#include <asm/elf.h>
20#include <linux/uaccess.h>
21#include <asm/tlbflush.h>
22#include "internal.h"
23
24void task_mem(struct seq_file *m, struct mm_struct *mm)
25{
26 unsigned long text, lib, swap, ptes, pmds, anon, file, shmem;
27 unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
28
29 anon = get_mm_counter(mm, MM_ANONPAGES);
30 file = get_mm_counter(mm, MM_FILEPAGES);
31 shmem = get_mm_counter(mm, MM_SHMEMPAGES);
32
33 /*
34 * Note: to minimize their overhead, mm maintains hiwater_vm and
35 * hiwater_rss only when about to *lower* total_vm or rss. Any
36 * collector of these hiwater stats must therefore get total_vm
37 * and rss too, which will usually be the higher. Barriers? not
38 * worth the effort, such snapshots can always be inconsistent.
39 */
40 hiwater_vm = total_vm = mm->total_vm;
41 if (hiwater_vm < mm->hiwater_vm)
42 hiwater_vm = mm->hiwater_vm;
43 hiwater_rss = total_rss = anon + file + shmem;
44 if (hiwater_rss < mm->hiwater_rss)
45 hiwater_rss = mm->hiwater_rss;
46
47 text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
48 lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
49 swap = get_mm_counter(mm, MM_SWAPENTS);
50 ptes = PTRS_PER_PTE * sizeof(pte_t) * atomic_long_read(&mm->nr_ptes);
51 pmds = PTRS_PER_PMD * sizeof(pmd_t) * mm_nr_pmds(mm);
52 seq_printf(m,
53 "VmPeak:\t%8lu kB\n"
54 "VmSize:\t%8lu kB\n"
55 "VmLck:\t%8lu kB\n"
56 "VmPin:\t%8lu kB\n"
57 "VmHWM:\t%8lu kB\n"
58 "VmRSS:\t%8lu kB\n"
59 "RssAnon:\t%8lu kB\n"
60 "RssFile:\t%8lu kB\n"
61 "RssShmem:\t%8lu kB\n"
62 "VmData:\t%8lu kB\n"
63 "VmStk:\t%8lu kB\n"
64 "VmExe:\t%8lu kB\n"
65 "VmLib:\t%8lu kB\n"
66 "VmPTE:\t%8lu kB\n"
67 "VmPMD:\t%8lu kB\n"
68 "VmSwap:\t%8lu kB\n",
69 hiwater_vm << (PAGE_SHIFT-10),
70 total_vm << (PAGE_SHIFT-10),
71 mm->locked_vm << (PAGE_SHIFT-10),
72 mm->pinned_vm << (PAGE_SHIFT-10),
73 hiwater_rss << (PAGE_SHIFT-10),
74 total_rss << (PAGE_SHIFT-10),
75 anon << (PAGE_SHIFT-10),
76 file << (PAGE_SHIFT-10),
77 shmem << (PAGE_SHIFT-10),
78 mm->data_vm << (PAGE_SHIFT-10),
79 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
80 ptes >> 10,
81 pmds >> 10,
82 swap << (PAGE_SHIFT-10));
83 hugetlb_report_usage(m, mm);
84}
85
86unsigned long task_vsize(struct mm_struct *mm)
87{
88 return PAGE_SIZE * mm->total_vm;
89}
90
91unsigned long task_statm(struct mm_struct *mm,
92 unsigned long *shared, unsigned long *text,
93 unsigned long *data, unsigned long *resident)
94{
95 *shared = get_mm_counter(mm, MM_FILEPAGES) +
96 get_mm_counter(mm, MM_SHMEMPAGES);
97 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
98 >> PAGE_SHIFT;
99 *data = mm->data_vm + mm->stack_vm;
100 *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
101 return mm->total_vm;
102}
103
104#ifdef CONFIG_NUMA
105/*
106 * Save get_task_policy() for show_numa_map().
107 */
108static void hold_task_mempolicy(struct proc_maps_private *priv)
109{
110 struct task_struct *task = priv->task;
111
112 task_lock(task);
113 priv->task_mempolicy = get_task_policy(task);
114 mpol_get(priv->task_mempolicy);
115 task_unlock(task);
116}
117static void release_task_mempolicy(struct proc_maps_private *priv)
118{
119 mpol_put(priv->task_mempolicy);
120}
121#else
122static void hold_task_mempolicy(struct proc_maps_private *priv)
123{
124}
125static void release_task_mempolicy(struct proc_maps_private *priv)
126{
127}
128#endif
129
130static void vma_stop(struct proc_maps_private *priv)
131{
132 struct mm_struct *mm = priv->mm;
133
134 release_task_mempolicy(priv);
135 up_read(&mm->mmap_sem);
136 mmput(mm);
137}
138
139static struct vm_area_struct *
140m_next_vma(struct proc_maps_private *priv, struct vm_area_struct *vma)
141{
142 if (vma == priv->tail_vma)
143 return NULL;
144 return vma->vm_next ?: priv->tail_vma;
145}
146
147static void m_cache_vma(struct seq_file *m, struct vm_area_struct *vma)
148{
149 if (m->count < m->size) /* vma is copied successfully */
150 m->version = m_next_vma(m->private, vma) ? vma->vm_end : -1UL;
151}
152
153static void *m_start(struct seq_file *m, loff_t *ppos)
154{
155 struct proc_maps_private *priv = m->private;
156 unsigned long last_addr = m->version;
157 struct mm_struct *mm;
158 struct vm_area_struct *vma;
159 unsigned int pos = *ppos;
160
161 /* See m_cache_vma(). Zero at the start or after lseek. */
162 if (last_addr == -1UL)
163 return NULL;
164
165 priv->task = get_proc_task(priv->inode);
166 if (!priv->task)
167 return ERR_PTR(-ESRCH);
168
169 mm = priv->mm;
170 if (!mm || !atomic_inc_not_zero(&mm->mm_users))
171 return NULL;
172
173 down_read(&mm->mmap_sem);
174 hold_task_mempolicy(priv);
175 priv->tail_vma = get_gate_vma(mm);
176
177 if (last_addr) {
178 vma = find_vma(mm, last_addr - 1);
179 if (vma && vma->vm_start <= last_addr)
180 vma = m_next_vma(priv, vma);
181 if (vma)
182 return vma;
183 }
184
185 m->version = 0;
186 if (pos < mm->map_count) {
187 for (vma = mm->mmap; pos; pos--) {
188 m->version = vma->vm_start;
189 vma = vma->vm_next;
190 }
191 return vma;
192 }
193
194 /* we do not bother to update m->version in this case */
195 if (pos == mm->map_count && priv->tail_vma)
196 return priv->tail_vma;
197
198 vma_stop(priv);
199 return NULL;
200}
201
202static void *m_next(struct seq_file *m, void *v, loff_t *pos)
203{
204 struct proc_maps_private *priv = m->private;
205 struct vm_area_struct *next;
206
207 (*pos)++;
208 next = m_next_vma(priv, v);
209 if (!next)
210 vma_stop(priv);
211 return next;
212}
213
214static void m_stop(struct seq_file *m, void *v)
215{
216 struct proc_maps_private *priv = m->private;
217
218 if (!IS_ERR_OR_NULL(v))
219 vma_stop(priv);
220 if (priv->task) {
221 put_task_struct(priv->task);
222 priv->task = NULL;
223 }
224}
225
226static int proc_maps_open(struct inode *inode, struct file *file,
227 const struct seq_operations *ops, int psize)
228{
229 struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
230
231 if (!priv)
232 return -ENOMEM;
233
234 priv->inode = inode;
235 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
236 if (IS_ERR(priv->mm)) {
237 int err = PTR_ERR(priv->mm);
238
239 seq_release_private(inode, file);
240 return err;
241 }
242
243 return 0;
244}
245
246static int proc_map_release(struct inode *inode, struct file *file)
247{
248 struct seq_file *seq = file->private_data;
249 struct proc_maps_private *priv = seq->private;
250
251 if (priv->mm)
252 mmdrop(priv->mm);
253
254 return seq_release_private(inode, file);
255}
256
257static int do_maps_open(struct inode *inode, struct file *file,
258 const struct seq_operations *ops)
259{
260 return proc_maps_open(inode, file, ops,
261 sizeof(struct proc_maps_private));
262}
263
264/*
265 * Indicate if the VMA is a stack for the given task; for
266 * /proc/PID/maps that is the stack of the main task.
267 */
268static int is_stack(struct proc_maps_private *priv,
269 struct vm_area_struct *vma)
270{
271 /*
272 * We make no effort to guess what a given thread considers to be
273 * its "stack". It's not even well-defined for programs written
274 * languages like Go.
275 */
276 return vma->vm_start <= vma->vm_mm->start_stack &&
277 vma->vm_end >= vma->vm_mm->start_stack;
278}
279
280static void
281show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
282{
283 struct mm_struct *mm = vma->vm_mm;
284 struct file *file = vma->vm_file;
285 struct proc_maps_private *priv = m->private;
286 vm_flags_t flags = vma->vm_flags;
287 unsigned long ino = 0;
288 unsigned long long pgoff = 0;
289 unsigned long start, end;
290 dev_t dev = 0;
291 const char *name = NULL;
292
293 if (file) {
294 struct inode *inode = file_inode(vma->vm_file);
295 dev = inode->i_sb->s_dev;
296 ino = inode->i_ino;
297 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
298 }
299
300 /* We don't show the stack guard page in /proc/maps */
301 start = vma->vm_start;
302 if (stack_guard_page_start(vma, start))
303 start += PAGE_SIZE;
304 end = vma->vm_end;
305 if (stack_guard_page_end(vma, end))
306 end -= PAGE_SIZE;
307
308 seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
309 seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu ",
310 start,
311 end,
312 flags & VM_READ ? 'r' : '-',
313 flags & VM_WRITE ? 'w' : '-',
314 flags & VM_EXEC ? 'x' : '-',
315 flags & VM_MAYSHARE ? 's' : 'p',
316 pgoff,
317 MAJOR(dev), MINOR(dev), ino);
318
319 /*
320 * Print the dentry name for named mappings, and a
321 * special [heap] marker for the heap:
322 */
323 if (file) {
324 seq_pad(m, ' ');
325 seq_file_path(m, file, "\n");
326 goto done;
327 }
328
329 if (vma->vm_ops && vma->vm_ops->name) {
330 name = vma->vm_ops->name(vma);
331 if (name)
332 goto done;
333 }
334
335 name = arch_vma_name(vma);
336 if (!name) {
337 if (!mm) {
338 name = "[vdso]";
339 goto done;
340 }
341
342 if (vma->vm_start <= mm->brk &&
343 vma->vm_end >= mm->start_brk) {
344 name = "[heap]";
345 goto done;
346 }
347
348 if (is_stack(priv, vma))
349 name = "[stack]";
350 }
351
352done:
353 if (name) {
354 seq_pad(m, ' ');
355 seq_puts(m, name);
356 }
357 seq_putc(m, '\n');
358}
359
360static int show_map(struct seq_file *m, void *v, int is_pid)
361{
362 show_map_vma(m, v, is_pid);
363 m_cache_vma(m, v);
364 return 0;
365}
366
367static int show_pid_map(struct seq_file *m, void *v)
368{
369 return show_map(m, v, 1);
370}
371
372static int show_tid_map(struct seq_file *m, void *v)
373{
374 return show_map(m, v, 0);
375}
376
377static const struct seq_operations proc_pid_maps_op = {
378 .start = m_start,
379 .next = m_next,
380 .stop = m_stop,
381 .show = show_pid_map
382};
383
384static const struct seq_operations proc_tid_maps_op = {
385 .start = m_start,
386 .next = m_next,
387 .stop = m_stop,
388 .show = show_tid_map
389};
390
391static int pid_maps_open(struct inode *inode, struct file *file)
392{
393 return do_maps_open(inode, file, &proc_pid_maps_op);
394}
395
396static int tid_maps_open(struct inode *inode, struct file *file)
397{
398 return do_maps_open(inode, file, &proc_tid_maps_op);
399}
400
401const struct file_operations proc_pid_maps_operations = {
402 .open = pid_maps_open,
403 .read = seq_read,
404 .llseek = seq_lseek,
405 .release = proc_map_release,
406};
407
408const struct file_operations proc_tid_maps_operations = {
409 .open = tid_maps_open,
410 .read = seq_read,
411 .llseek = seq_lseek,
412 .release = proc_map_release,
413};
414
415/*
416 * Proportional Set Size(PSS): my share of RSS.
417 *
418 * PSS of a process is the count of pages it has in memory, where each
419 * page is divided by the number of processes sharing it. So if a
420 * process has 1000 pages all to itself, and 1000 shared with one other
421 * process, its PSS will be 1500.
422 *
423 * To keep (accumulated) division errors low, we adopt a 64bit
424 * fixed-point pss counter to minimize division errors. So (pss >>
425 * PSS_SHIFT) would be the real byte count.
426 *
427 * A shift of 12 before division means (assuming 4K page size):
428 * - 1M 3-user-pages add up to 8KB errors;
429 * - supports mapcount up to 2^24, or 16M;
430 * - supports PSS up to 2^52 bytes, or 4PB.
431 */
432#define PSS_SHIFT 12
433
434#ifdef CONFIG_PROC_PAGE_MONITOR
435struct mem_size_stats {
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 anonymous_thp;
444 unsigned long shmem_thp;
445 unsigned long swap;
446 unsigned long shared_hugetlb;
447 unsigned long private_hugetlb;
448 u64 pss;
449 u64 swap_pss;
450 bool check_shmem_swap;
451};
452
453static void smaps_account(struct mem_size_stats *mss, struct page *page,
454 bool compound, bool young, bool dirty)
455{
456 int i, nr = compound ? 1 << compound_order(page) : 1;
457 unsigned long size = nr * PAGE_SIZE;
458
459 if (PageAnon(page))
460 mss->anonymous += size;
461
462 mss->resident += size;
463 /* Accumulate the size in pages that have been accessed. */
464 if (young || page_is_young(page) || PageReferenced(page))
465 mss->referenced += size;
466
467 /*
468 * page_count(page) == 1 guarantees the page is mapped exactly once.
469 * If any subpage of the compound page mapped with PTE it would elevate
470 * page_count().
471 */
472 if (page_count(page) == 1) {
473 if (dirty || PageDirty(page))
474 mss->private_dirty += size;
475 else
476 mss->private_clean += size;
477 mss->pss += (u64)size << PSS_SHIFT;
478 return;
479 }
480
481 for (i = 0; i < nr; i++, page++) {
482 int mapcount = page_mapcount(page);
483
484 if (mapcount >= 2) {
485 if (dirty || PageDirty(page))
486 mss->shared_dirty += PAGE_SIZE;
487 else
488 mss->shared_clean += PAGE_SIZE;
489 mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount;
490 } else {
491 if (dirty || PageDirty(page))
492 mss->private_dirty += PAGE_SIZE;
493 else
494 mss->private_clean += PAGE_SIZE;
495 mss->pss += PAGE_SIZE << PSS_SHIFT;
496 }
497 }
498}
499
500#ifdef CONFIG_SHMEM
501static int smaps_pte_hole(unsigned long addr, unsigned long end,
502 struct mm_walk *walk)
503{
504 struct mem_size_stats *mss = walk->private;
505
506 mss->swap += shmem_partial_swap_usage(
507 walk->vma->vm_file->f_mapping, addr, end);
508
509 return 0;
510}
511#endif
512
513static void smaps_pte_entry(pte_t *pte, unsigned long addr,
514 struct mm_walk *walk)
515{
516 struct mem_size_stats *mss = walk->private;
517 struct vm_area_struct *vma = walk->vma;
518 struct page *page = NULL;
519
520 if (pte_present(*pte)) {
521 page = vm_normal_page(vma, addr, *pte);
522 } else if (is_swap_pte(*pte)) {
523 swp_entry_t swpent = pte_to_swp_entry(*pte);
524
525 if (!non_swap_entry(swpent)) {
526 int mapcount;
527
528 mss->swap += PAGE_SIZE;
529 mapcount = swp_swapcount(swpent);
530 if (mapcount >= 2) {
531 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
532
533 do_div(pss_delta, mapcount);
534 mss->swap_pss += pss_delta;
535 } else {
536 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
537 }
538 } else if (is_migration_entry(swpent))
539 page = migration_entry_to_page(swpent);
540 } else if (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap
541 && pte_none(*pte))) {
542 page = find_get_entry(vma->vm_file->f_mapping,
543 linear_page_index(vma, addr));
544 if (!page)
545 return;
546
547 if (radix_tree_exceptional_entry(page))
548 mss->swap += PAGE_SIZE;
549 else
550 put_page(page);
551
552 return;
553 }
554
555 if (!page)
556 return;
557
558 smaps_account(mss, page, false, pte_young(*pte), pte_dirty(*pte));
559}
560
561#ifdef CONFIG_TRANSPARENT_HUGEPAGE
562static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
563 struct mm_walk *walk)
564{
565 struct mem_size_stats *mss = walk->private;
566 struct vm_area_struct *vma = walk->vma;
567 struct page *page;
568
569 /* FOLL_DUMP will return -EFAULT on huge zero page */
570 page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
571 if (IS_ERR_OR_NULL(page))
572 return;
573 if (PageAnon(page))
574 mss->anonymous_thp += HPAGE_PMD_SIZE;
575 else if (PageSwapBacked(page))
576 mss->shmem_thp += HPAGE_PMD_SIZE;
577 else if (is_zone_device_page(page))
578 /* pass */;
579 else
580 VM_BUG_ON_PAGE(1, page);
581 smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd));
582}
583#else
584static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
585 struct mm_walk *walk)
586{
587}
588#endif
589
590static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
591 struct mm_walk *walk)
592{
593 struct vm_area_struct *vma = walk->vma;
594 pte_t *pte;
595 spinlock_t *ptl;
596
597 ptl = pmd_trans_huge_lock(pmd, vma);
598 if (ptl) {
599 smaps_pmd_entry(pmd, addr, walk);
600 spin_unlock(ptl);
601 return 0;
602 }
603
604 if (pmd_trans_unstable(pmd))
605 return 0;
606 /*
607 * The mmap_sem held all the way back in m_start() is what
608 * keeps khugepaged out of here and from collapsing things
609 * in here.
610 */
611 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
612 for (; addr != end; pte++, addr += PAGE_SIZE)
613 smaps_pte_entry(pte, addr, walk);
614 pte_unmap_unlock(pte - 1, ptl);
615 cond_resched();
616 return 0;
617}
618
619static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
620{
621 /*
622 * Don't forget to update Documentation/ on changes.
623 */
624 static const char mnemonics[BITS_PER_LONG][2] = {
625 /*
626 * In case if we meet a flag we don't know about.
627 */
628 [0 ... (BITS_PER_LONG-1)] = "??",
629
630 [ilog2(VM_READ)] = "rd",
631 [ilog2(VM_WRITE)] = "wr",
632 [ilog2(VM_EXEC)] = "ex",
633 [ilog2(VM_SHARED)] = "sh",
634 [ilog2(VM_MAYREAD)] = "mr",
635 [ilog2(VM_MAYWRITE)] = "mw",
636 [ilog2(VM_MAYEXEC)] = "me",
637 [ilog2(VM_MAYSHARE)] = "ms",
638 [ilog2(VM_GROWSDOWN)] = "gd",
639 [ilog2(VM_PFNMAP)] = "pf",
640 [ilog2(VM_DENYWRITE)] = "dw",
641#ifdef CONFIG_X86_INTEL_MPX
642 [ilog2(VM_MPX)] = "mp",
643#endif
644 [ilog2(VM_LOCKED)] = "lo",
645 [ilog2(VM_IO)] = "io",
646 [ilog2(VM_SEQ_READ)] = "sr",
647 [ilog2(VM_RAND_READ)] = "rr",
648 [ilog2(VM_DONTCOPY)] = "dc",
649 [ilog2(VM_DONTEXPAND)] = "de",
650 [ilog2(VM_ACCOUNT)] = "ac",
651 [ilog2(VM_NORESERVE)] = "nr",
652 [ilog2(VM_HUGETLB)] = "ht",
653 [ilog2(VM_ARCH_1)] = "ar",
654 [ilog2(VM_DONTDUMP)] = "dd",
655#ifdef CONFIG_MEM_SOFT_DIRTY
656 [ilog2(VM_SOFTDIRTY)] = "sd",
657#endif
658 [ilog2(VM_MIXEDMAP)] = "mm",
659 [ilog2(VM_HUGEPAGE)] = "hg",
660 [ilog2(VM_NOHUGEPAGE)] = "nh",
661 [ilog2(VM_MERGEABLE)] = "mg",
662 [ilog2(VM_UFFD_MISSING)]= "um",
663 [ilog2(VM_UFFD_WP)] = "uw",
664#ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
665 /* These come out via ProtectionKey: */
666 [ilog2(VM_PKEY_BIT0)] = "",
667 [ilog2(VM_PKEY_BIT1)] = "",
668 [ilog2(VM_PKEY_BIT2)] = "",
669 [ilog2(VM_PKEY_BIT3)] = "",
670#endif
671 };
672 size_t i;
673
674 seq_puts(m, "VmFlags: ");
675 for (i = 0; i < BITS_PER_LONG; i++) {
676 if (!mnemonics[i][0])
677 continue;
678 if (vma->vm_flags & (1UL << i)) {
679 seq_printf(m, "%c%c ",
680 mnemonics[i][0], mnemonics[i][1]);
681 }
682 }
683 seq_putc(m, '\n');
684}
685
686#ifdef CONFIG_HUGETLB_PAGE
687static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
688 unsigned long addr, unsigned long end,
689 struct mm_walk *walk)
690{
691 struct mem_size_stats *mss = walk->private;
692 struct vm_area_struct *vma = walk->vma;
693 struct page *page = NULL;
694
695 if (pte_present(*pte)) {
696 page = vm_normal_page(vma, addr, *pte);
697 } else if (is_swap_pte(*pte)) {
698 swp_entry_t swpent = pte_to_swp_entry(*pte);
699
700 if (is_migration_entry(swpent))
701 page = migration_entry_to_page(swpent);
702 }
703 if (page) {
704 int mapcount = page_mapcount(page);
705
706 if (mapcount >= 2)
707 mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
708 else
709 mss->private_hugetlb += huge_page_size(hstate_vma(vma));
710 }
711 return 0;
712}
713#endif /* HUGETLB_PAGE */
714
715void __weak arch_show_smap(struct seq_file *m, struct vm_area_struct *vma)
716{
717}
718
719static int show_smap(struct seq_file *m, void *v, int is_pid)
720{
721 struct vm_area_struct *vma = v;
722 struct mem_size_stats mss;
723 struct mm_walk smaps_walk = {
724 .pmd_entry = smaps_pte_range,
725#ifdef CONFIG_HUGETLB_PAGE
726 .hugetlb_entry = smaps_hugetlb_range,
727#endif
728 .mm = vma->vm_mm,
729 .private = &mss,
730 };
731
732 memset(&mss, 0, sizeof mss);
733
734#ifdef CONFIG_SHMEM
735 if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
736 /*
737 * For shared or readonly shmem mappings we know that all
738 * swapped out pages belong to the shmem object, and we can
739 * obtain the swap value much more efficiently. For private
740 * writable mappings, we might have COW pages that are
741 * not affected by the parent swapped out pages of the shmem
742 * object, so we have to distinguish them during the page walk.
743 * Unless we know that the shmem object (or the part mapped by
744 * our VMA) has no swapped out pages at all.
745 */
746 unsigned long shmem_swapped = shmem_swap_usage(vma);
747
748 if (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
749 !(vma->vm_flags & VM_WRITE)) {
750 mss.swap = shmem_swapped;
751 } else {
752 mss.check_shmem_swap = true;
753 smaps_walk.pte_hole = smaps_pte_hole;
754 }
755 }
756#endif
757
758 /* mmap_sem is held in m_start */
759 walk_page_vma(vma, &smaps_walk);
760
761 show_map_vma(m, vma, is_pid);
762
763 seq_printf(m,
764 "Size: %8lu kB\n"
765 "Rss: %8lu kB\n"
766 "Pss: %8lu kB\n"
767 "Shared_Clean: %8lu kB\n"
768 "Shared_Dirty: %8lu kB\n"
769 "Private_Clean: %8lu kB\n"
770 "Private_Dirty: %8lu kB\n"
771 "Referenced: %8lu kB\n"
772 "Anonymous: %8lu kB\n"
773 "AnonHugePages: %8lu kB\n"
774 "ShmemPmdMapped: %8lu kB\n"
775 "Shared_Hugetlb: %8lu kB\n"
776 "Private_Hugetlb: %7lu kB\n"
777 "Swap: %8lu kB\n"
778 "SwapPss: %8lu kB\n"
779 "KernelPageSize: %8lu kB\n"
780 "MMUPageSize: %8lu kB\n"
781 "Locked: %8lu kB\n",
782 (vma->vm_end - vma->vm_start) >> 10,
783 mss.resident >> 10,
784 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
785 mss.shared_clean >> 10,
786 mss.shared_dirty >> 10,
787 mss.private_clean >> 10,
788 mss.private_dirty >> 10,
789 mss.referenced >> 10,
790 mss.anonymous >> 10,
791 mss.anonymous_thp >> 10,
792 mss.shmem_thp >> 10,
793 mss.shared_hugetlb >> 10,
794 mss.private_hugetlb >> 10,
795 mss.swap >> 10,
796 (unsigned long)(mss.swap_pss >> (10 + PSS_SHIFT)),
797 vma_kernel_pagesize(vma) >> 10,
798 vma_mmu_pagesize(vma) >> 10,
799 (vma->vm_flags & VM_LOCKED) ?
800 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
801
802 arch_show_smap(m, vma);
803 show_smap_vma_flags(m, vma);
804 m_cache_vma(m, vma);
805 return 0;
806}
807
808static int show_pid_smap(struct seq_file *m, void *v)
809{
810 return show_smap(m, v, 1);
811}
812
813static int show_tid_smap(struct seq_file *m, void *v)
814{
815 return show_smap(m, v, 0);
816}
817
818static const struct seq_operations proc_pid_smaps_op = {
819 .start = m_start,
820 .next = m_next,
821 .stop = m_stop,
822 .show = show_pid_smap
823};
824
825static const struct seq_operations proc_tid_smaps_op = {
826 .start = m_start,
827 .next = m_next,
828 .stop = m_stop,
829 .show = show_tid_smap
830};
831
832static int pid_smaps_open(struct inode *inode, struct file *file)
833{
834 return do_maps_open(inode, file, &proc_pid_smaps_op);
835}
836
837static int tid_smaps_open(struct inode *inode, struct file *file)
838{
839 return do_maps_open(inode, file, &proc_tid_smaps_op);
840}
841
842const struct file_operations proc_pid_smaps_operations = {
843 .open = pid_smaps_open,
844 .read = seq_read,
845 .llseek = seq_lseek,
846 .release = proc_map_release,
847};
848
849const struct file_operations proc_tid_smaps_operations = {
850 .open = tid_smaps_open,
851 .read = seq_read,
852 .llseek = seq_lseek,
853 .release = proc_map_release,
854};
855
856enum clear_refs_types {
857 CLEAR_REFS_ALL = 1,
858 CLEAR_REFS_ANON,
859 CLEAR_REFS_MAPPED,
860 CLEAR_REFS_SOFT_DIRTY,
861 CLEAR_REFS_MM_HIWATER_RSS,
862 CLEAR_REFS_LAST,
863};
864
865struct clear_refs_private {
866 enum clear_refs_types type;
867};
868
869#ifdef CONFIG_MEM_SOFT_DIRTY
870static inline void clear_soft_dirty(struct vm_area_struct *vma,
871 unsigned long addr, pte_t *pte)
872{
873 /*
874 * The soft-dirty tracker uses #PF-s to catch writes
875 * to pages, so write-protect the pte as well. See the
876 * Documentation/vm/soft-dirty.txt for full description
877 * of how soft-dirty works.
878 */
879 pte_t ptent = *pte;
880
881 if (pte_present(ptent)) {
882 ptent = ptep_modify_prot_start(vma->vm_mm, addr, pte);
883 ptent = pte_wrprotect(ptent);
884 ptent = pte_clear_soft_dirty(ptent);
885 ptep_modify_prot_commit(vma->vm_mm, addr, pte, ptent);
886 } else if (is_swap_pte(ptent)) {
887 ptent = pte_swp_clear_soft_dirty(ptent);
888 set_pte_at(vma->vm_mm, addr, pte, ptent);
889 }
890}
891#else
892static inline void clear_soft_dirty(struct vm_area_struct *vma,
893 unsigned long addr, pte_t *pte)
894{
895}
896#endif
897
898#if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
899static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
900 unsigned long addr, pmd_t *pmdp)
901{
902 pmd_t pmd = pmdp_huge_get_and_clear(vma->vm_mm, addr, pmdp);
903
904 pmd = pmd_wrprotect(pmd);
905 pmd = pmd_clear_soft_dirty(pmd);
906
907 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
908}
909#else
910static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
911 unsigned long addr, pmd_t *pmdp)
912{
913}
914#endif
915
916static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
917 unsigned long end, struct mm_walk *walk)
918{
919 struct clear_refs_private *cp = walk->private;
920 struct vm_area_struct *vma = walk->vma;
921 pte_t *pte, ptent;
922 spinlock_t *ptl;
923 struct page *page;
924
925 ptl = pmd_trans_huge_lock(pmd, vma);
926 if (ptl) {
927 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
928 clear_soft_dirty_pmd(vma, addr, pmd);
929 goto out;
930 }
931
932 page = pmd_page(*pmd);
933
934 /* Clear accessed and referenced bits. */
935 pmdp_test_and_clear_young(vma, addr, pmd);
936 test_and_clear_page_young(page);
937 ClearPageReferenced(page);
938out:
939 spin_unlock(ptl);
940 return 0;
941 }
942
943 if (pmd_trans_unstable(pmd))
944 return 0;
945
946 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
947 for (; addr != end; pte++, addr += PAGE_SIZE) {
948 ptent = *pte;
949
950 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
951 clear_soft_dirty(vma, addr, pte);
952 continue;
953 }
954
955 if (!pte_present(ptent))
956 continue;
957
958 page = vm_normal_page(vma, addr, ptent);
959 if (!page)
960 continue;
961
962 /* Clear accessed and referenced bits. */
963 ptep_test_and_clear_young(vma, addr, pte);
964 test_and_clear_page_young(page);
965 ClearPageReferenced(page);
966 }
967 pte_unmap_unlock(pte - 1, ptl);
968 cond_resched();
969 return 0;
970}
971
972static int clear_refs_test_walk(unsigned long start, unsigned long end,
973 struct mm_walk *walk)
974{
975 struct clear_refs_private *cp = walk->private;
976 struct vm_area_struct *vma = walk->vma;
977
978 if (vma->vm_flags & VM_PFNMAP)
979 return 1;
980
981 /*
982 * Writing 1 to /proc/pid/clear_refs affects all pages.
983 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
984 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
985 * Writing 4 to /proc/pid/clear_refs affects all pages.
986 */
987 if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
988 return 1;
989 if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
990 return 1;
991 return 0;
992}
993
994static ssize_t clear_refs_write(struct file *file, const char __user *buf,
995 size_t count, loff_t *ppos)
996{
997 struct task_struct *task;
998 char buffer[PROC_NUMBUF];
999 struct mm_struct *mm;
1000 struct vm_area_struct *vma;
1001 enum clear_refs_types type;
1002 int itype;
1003 int rv;
1004
1005 memset(buffer, 0, sizeof(buffer));
1006 if (count > sizeof(buffer) - 1)
1007 count = sizeof(buffer) - 1;
1008 if (copy_from_user(buffer, buf, count))
1009 return -EFAULT;
1010 rv = kstrtoint(strstrip(buffer), 10, &itype);
1011 if (rv < 0)
1012 return rv;
1013 type = (enum clear_refs_types)itype;
1014 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1015 return -EINVAL;
1016
1017 task = get_proc_task(file_inode(file));
1018 if (!task)
1019 return -ESRCH;
1020 mm = get_task_mm(task);
1021 if (mm) {
1022 struct clear_refs_private cp = {
1023 .type = type,
1024 };
1025 struct mm_walk clear_refs_walk = {
1026 .pmd_entry = clear_refs_pte_range,
1027 .test_walk = clear_refs_test_walk,
1028 .mm = mm,
1029 .private = &cp,
1030 };
1031
1032 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1033 if (down_write_killable(&mm->mmap_sem)) {
1034 count = -EINTR;
1035 goto out_mm;
1036 }
1037
1038 /*
1039 * Writing 5 to /proc/pid/clear_refs resets the peak
1040 * resident set size to this mm's current rss value.
1041 */
1042 reset_mm_hiwater_rss(mm);
1043 up_write(&mm->mmap_sem);
1044 goto out_mm;
1045 }
1046
1047 down_read(&mm->mmap_sem);
1048 if (type == CLEAR_REFS_SOFT_DIRTY) {
1049 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1050 if (!(vma->vm_flags & VM_SOFTDIRTY))
1051 continue;
1052 up_read(&mm->mmap_sem);
1053 if (down_write_killable(&mm->mmap_sem)) {
1054 count = -EINTR;
1055 goto out_mm;
1056 }
1057 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1058 vma->vm_flags &= ~VM_SOFTDIRTY;
1059 vma_set_page_prot(vma);
1060 }
1061 downgrade_write(&mm->mmap_sem);
1062 break;
1063 }
1064 mmu_notifier_invalidate_range_start(mm, 0, -1);
1065 }
1066 walk_page_range(0, mm->highest_vm_end, &clear_refs_walk);
1067 if (type == CLEAR_REFS_SOFT_DIRTY)
1068 mmu_notifier_invalidate_range_end(mm, 0, -1);
1069 flush_tlb_mm(mm);
1070 up_read(&mm->mmap_sem);
1071out_mm:
1072 mmput(mm);
1073 }
1074 put_task_struct(task);
1075
1076 return count;
1077}
1078
1079const struct file_operations proc_clear_refs_operations = {
1080 .write = clear_refs_write,
1081 .llseek = noop_llseek,
1082};
1083
1084typedef struct {
1085 u64 pme;
1086} pagemap_entry_t;
1087
1088struct pagemapread {
1089 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */
1090 pagemap_entry_t *buffer;
1091 bool show_pfn;
1092};
1093
1094#define PAGEMAP_WALK_SIZE (PMD_SIZE)
1095#define PAGEMAP_WALK_MASK (PMD_MASK)
1096
1097#define PM_ENTRY_BYTES sizeof(pagemap_entry_t)
1098#define PM_PFRAME_BITS 55
1099#define PM_PFRAME_MASK GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1100#define PM_SOFT_DIRTY BIT_ULL(55)
1101#define PM_MMAP_EXCLUSIVE BIT_ULL(56)
1102#define PM_FILE BIT_ULL(61)
1103#define PM_SWAP BIT_ULL(62)
1104#define PM_PRESENT BIT_ULL(63)
1105
1106#define PM_END_OF_BUFFER 1
1107
1108static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1109{
1110 return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1111}
1112
1113static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1114 struct pagemapread *pm)
1115{
1116 pm->buffer[pm->pos++] = *pme;
1117 if (pm->pos >= pm->len)
1118 return PM_END_OF_BUFFER;
1119 return 0;
1120}
1121
1122static int pagemap_pte_hole(unsigned long start, unsigned long end,
1123 struct mm_walk *walk)
1124{
1125 struct pagemapread *pm = walk->private;
1126 unsigned long addr = start;
1127 int err = 0;
1128
1129 while (addr < end) {
1130 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1131 pagemap_entry_t pme = make_pme(0, 0);
1132 /* End of address space hole, which we mark as non-present. */
1133 unsigned long hole_end;
1134
1135 if (vma)
1136 hole_end = min(end, vma->vm_start);
1137 else
1138 hole_end = end;
1139
1140 for (; addr < hole_end; addr += PAGE_SIZE) {
1141 err = add_to_pagemap(addr, &pme, pm);
1142 if (err)
1143 goto out;
1144 }
1145
1146 if (!vma)
1147 break;
1148
1149 /* Addresses in the VMA. */
1150 if (vma->vm_flags & VM_SOFTDIRTY)
1151 pme = make_pme(0, PM_SOFT_DIRTY);
1152 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1153 err = add_to_pagemap(addr, &pme, pm);
1154 if (err)
1155 goto out;
1156 }
1157 }
1158out:
1159 return err;
1160}
1161
1162static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1163 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1164{
1165 u64 frame = 0, flags = 0;
1166 struct page *page = NULL;
1167
1168 if (pte_present(pte)) {
1169 if (pm->show_pfn)
1170 frame = pte_pfn(pte);
1171 flags |= PM_PRESENT;
1172 page = vm_normal_page(vma, addr, pte);
1173 if (pte_soft_dirty(pte))
1174 flags |= PM_SOFT_DIRTY;
1175 } else if (is_swap_pte(pte)) {
1176 swp_entry_t entry;
1177 if (pte_swp_soft_dirty(pte))
1178 flags |= PM_SOFT_DIRTY;
1179 entry = pte_to_swp_entry(pte);
1180 frame = swp_type(entry) |
1181 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1182 flags |= PM_SWAP;
1183 if (is_migration_entry(entry))
1184 page = migration_entry_to_page(entry);
1185 }
1186
1187 if (page && !PageAnon(page))
1188 flags |= PM_FILE;
1189 if (page && page_mapcount(page) == 1)
1190 flags |= PM_MMAP_EXCLUSIVE;
1191 if (vma->vm_flags & VM_SOFTDIRTY)
1192 flags |= PM_SOFT_DIRTY;
1193
1194 return make_pme(frame, flags);
1195}
1196
1197static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1198 struct mm_walk *walk)
1199{
1200 struct vm_area_struct *vma = walk->vma;
1201 struct pagemapread *pm = walk->private;
1202 spinlock_t *ptl;
1203 pte_t *pte, *orig_pte;
1204 int err = 0;
1205
1206#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1207 ptl = pmd_trans_huge_lock(pmdp, vma);
1208 if (ptl) {
1209 u64 flags = 0, frame = 0;
1210 pmd_t pmd = *pmdp;
1211
1212 if ((vma->vm_flags & VM_SOFTDIRTY) || pmd_soft_dirty(pmd))
1213 flags |= PM_SOFT_DIRTY;
1214
1215 /*
1216 * Currently pmd for thp is always present because thp
1217 * can not be swapped-out, migrated, or HWPOISONed
1218 * (split in such cases instead.)
1219 * This if-check is just to prepare for future implementation.
1220 */
1221 if (pmd_present(pmd)) {
1222 struct page *page = pmd_page(pmd);
1223
1224 if (page_mapcount(page) == 1)
1225 flags |= PM_MMAP_EXCLUSIVE;
1226
1227 flags |= PM_PRESENT;
1228 if (pm->show_pfn)
1229 frame = pmd_pfn(pmd) +
1230 ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1231 }
1232
1233 for (; addr != end; addr += PAGE_SIZE) {
1234 pagemap_entry_t pme = make_pme(frame, flags);
1235
1236 err = add_to_pagemap(addr, &pme, pm);
1237 if (err)
1238 break;
1239 if (pm->show_pfn && (flags & PM_PRESENT))
1240 frame++;
1241 }
1242 spin_unlock(ptl);
1243 return err;
1244 }
1245
1246 if (pmd_trans_unstable(pmdp))
1247 return 0;
1248#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1249
1250 /*
1251 * We can assume that @vma always points to a valid one and @end never
1252 * goes beyond vma->vm_end.
1253 */
1254 orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1255 for (; addr < end; pte++, addr += PAGE_SIZE) {
1256 pagemap_entry_t pme;
1257
1258 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1259 err = add_to_pagemap(addr, &pme, pm);
1260 if (err)
1261 break;
1262 }
1263 pte_unmap_unlock(orig_pte, ptl);
1264
1265 cond_resched();
1266
1267 return err;
1268}
1269
1270#ifdef CONFIG_HUGETLB_PAGE
1271/* This function walks within one hugetlb entry in the single call */
1272static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1273 unsigned long addr, unsigned long end,
1274 struct mm_walk *walk)
1275{
1276 struct pagemapread *pm = walk->private;
1277 struct vm_area_struct *vma = walk->vma;
1278 u64 flags = 0, frame = 0;
1279 int err = 0;
1280 pte_t pte;
1281
1282 if (vma->vm_flags & VM_SOFTDIRTY)
1283 flags |= PM_SOFT_DIRTY;
1284
1285 pte = huge_ptep_get(ptep);
1286 if (pte_present(pte)) {
1287 struct page *page = pte_page(pte);
1288
1289 if (!PageAnon(page))
1290 flags |= PM_FILE;
1291
1292 if (page_mapcount(page) == 1)
1293 flags |= PM_MMAP_EXCLUSIVE;
1294
1295 flags |= PM_PRESENT;
1296 if (pm->show_pfn)
1297 frame = pte_pfn(pte) +
1298 ((addr & ~hmask) >> PAGE_SHIFT);
1299 }
1300
1301 for (; addr != end; addr += PAGE_SIZE) {
1302 pagemap_entry_t pme = make_pme(frame, flags);
1303
1304 err = add_to_pagemap(addr, &pme, pm);
1305 if (err)
1306 return err;
1307 if (pm->show_pfn && (flags & PM_PRESENT))
1308 frame++;
1309 }
1310
1311 cond_resched();
1312
1313 return err;
1314}
1315#endif /* HUGETLB_PAGE */
1316
1317/*
1318 * /proc/pid/pagemap - an array mapping virtual pages to pfns
1319 *
1320 * For each page in the address space, this file contains one 64-bit entry
1321 * consisting of the following:
1322 *
1323 * Bits 0-54 page frame number (PFN) if present
1324 * Bits 0-4 swap type if swapped
1325 * Bits 5-54 swap offset if swapped
1326 * Bit 55 pte is soft-dirty (see Documentation/vm/soft-dirty.txt)
1327 * Bit 56 page exclusively mapped
1328 * Bits 57-60 zero
1329 * Bit 61 page is file-page or shared-anon
1330 * Bit 62 page swapped
1331 * Bit 63 page present
1332 *
1333 * If the page is not present but in swap, then the PFN contains an
1334 * encoding of the swap file number and the page's offset into the
1335 * swap. Unmapped pages return a null PFN. This allows determining
1336 * precisely which pages are mapped (or in swap) and comparing mapped
1337 * pages between processes.
1338 *
1339 * Efficient users of this interface will use /proc/pid/maps to
1340 * determine which areas of memory are actually mapped and llseek to
1341 * skip over unmapped regions.
1342 */
1343static ssize_t pagemap_read(struct file *file, char __user *buf,
1344 size_t count, loff_t *ppos)
1345{
1346 struct mm_struct *mm = file->private_data;
1347 struct pagemapread pm;
1348 struct mm_walk pagemap_walk = {};
1349 unsigned long src;
1350 unsigned long svpfn;
1351 unsigned long start_vaddr;
1352 unsigned long end_vaddr;
1353 int ret = 0, copied = 0;
1354
1355 if (!mm || !atomic_inc_not_zero(&mm->mm_users))
1356 goto out;
1357
1358 ret = -EINVAL;
1359 /* file position must be aligned */
1360 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1361 goto out_mm;
1362
1363 ret = 0;
1364 if (!count)
1365 goto out_mm;
1366
1367 /* do not disclose physical addresses: attack vector */
1368 pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1369
1370 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1371 pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY);
1372 ret = -ENOMEM;
1373 if (!pm.buffer)
1374 goto out_mm;
1375
1376 pagemap_walk.pmd_entry = pagemap_pmd_range;
1377 pagemap_walk.pte_hole = pagemap_pte_hole;
1378#ifdef CONFIG_HUGETLB_PAGE
1379 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1380#endif
1381 pagemap_walk.mm = mm;
1382 pagemap_walk.private = ±
1383
1384 src = *ppos;
1385 svpfn = src / PM_ENTRY_BYTES;
1386 start_vaddr = svpfn << PAGE_SHIFT;
1387 end_vaddr = mm->task_size;
1388
1389 /* watch out for wraparound */
1390 if (svpfn > mm->task_size >> PAGE_SHIFT)
1391 start_vaddr = end_vaddr;
1392
1393 /*
1394 * The odds are that this will stop walking way
1395 * before end_vaddr, because the length of the
1396 * user buffer is tracked in "pm", and the walk
1397 * will stop when we hit the end of the buffer.
1398 */
1399 ret = 0;
1400 while (count && (start_vaddr < end_vaddr)) {
1401 int len;
1402 unsigned long end;
1403
1404 pm.pos = 0;
1405 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1406 /* overflow ? */
1407 if (end < start_vaddr || end > end_vaddr)
1408 end = end_vaddr;
1409 down_read(&mm->mmap_sem);
1410 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1411 up_read(&mm->mmap_sem);
1412 start_vaddr = end;
1413
1414 len = min(count, PM_ENTRY_BYTES * pm.pos);
1415 if (copy_to_user(buf, pm.buffer, len)) {
1416 ret = -EFAULT;
1417 goto out_free;
1418 }
1419 copied += len;
1420 buf += len;
1421 count -= len;
1422 }
1423 *ppos += copied;
1424 if (!ret || ret == PM_END_OF_BUFFER)
1425 ret = copied;
1426
1427out_free:
1428 kfree(pm.buffer);
1429out_mm:
1430 mmput(mm);
1431out:
1432 return ret;
1433}
1434
1435static int pagemap_open(struct inode *inode, struct file *file)
1436{
1437 struct mm_struct *mm;
1438
1439 mm = proc_mem_open(inode, PTRACE_MODE_READ);
1440 if (IS_ERR(mm))
1441 return PTR_ERR(mm);
1442 file->private_data = mm;
1443 return 0;
1444}
1445
1446static int pagemap_release(struct inode *inode, struct file *file)
1447{
1448 struct mm_struct *mm = file->private_data;
1449
1450 if (mm)
1451 mmdrop(mm);
1452 return 0;
1453}
1454
1455const struct file_operations proc_pagemap_operations = {
1456 .llseek = mem_lseek, /* borrow this */
1457 .read = pagemap_read,
1458 .open = pagemap_open,
1459 .release = pagemap_release,
1460};
1461#endif /* CONFIG_PROC_PAGE_MONITOR */
1462
1463#ifdef CONFIG_NUMA
1464
1465struct numa_maps {
1466 unsigned long pages;
1467 unsigned long anon;
1468 unsigned long active;
1469 unsigned long writeback;
1470 unsigned long mapcount_max;
1471 unsigned long dirty;
1472 unsigned long swapcache;
1473 unsigned long node[MAX_NUMNODES];
1474};
1475
1476struct numa_maps_private {
1477 struct proc_maps_private proc_maps;
1478 struct numa_maps md;
1479};
1480
1481static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1482 unsigned long nr_pages)
1483{
1484 int count = page_mapcount(page);
1485
1486 md->pages += nr_pages;
1487 if (pte_dirty || PageDirty(page))
1488 md->dirty += nr_pages;
1489
1490 if (PageSwapCache(page))
1491 md->swapcache += nr_pages;
1492
1493 if (PageActive(page) || PageUnevictable(page))
1494 md->active += nr_pages;
1495
1496 if (PageWriteback(page))
1497 md->writeback += nr_pages;
1498
1499 if (PageAnon(page))
1500 md->anon += nr_pages;
1501
1502 if (count > md->mapcount_max)
1503 md->mapcount_max = count;
1504
1505 md->node[page_to_nid(page)] += nr_pages;
1506}
1507
1508static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1509 unsigned long addr)
1510{
1511 struct page *page;
1512 int nid;
1513
1514 if (!pte_present(pte))
1515 return NULL;
1516
1517 page = vm_normal_page(vma, addr, pte);
1518 if (!page)
1519 return NULL;
1520
1521 if (PageReserved(page))
1522 return NULL;
1523
1524 nid = page_to_nid(page);
1525 if (!node_isset(nid, node_states[N_MEMORY]))
1526 return NULL;
1527
1528 return page;
1529}
1530
1531#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1532static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1533 struct vm_area_struct *vma,
1534 unsigned long addr)
1535{
1536 struct page *page;
1537 int nid;
1538
1539 if (!pmd_present(pmd))
1540 return NULL;
1541
1542 page = vm_normal_page_pmd(vma, addr, pmd);
1543 if (!page)
1544 return NULL;
1545
1546 if (PageReserved(page))
1547 return NULL;
1548
1549 nid = page_to_nid(page);
1550 if (!node_isset(nid, node_states[N_MEMORY]))
1551 return NULL;
1552
1553 return page;
1554}
1555#endif
1556
1557static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1558 unsigned long end, struct mm_walk *walk)
1559{
1560 struct numa_maps *md = walk->private;
1561 struct vm_area_struct *vma = walk->vma;
1562 spinlock_t *ptl;
1563 pte_t *orig_pte;
1564 pte_t *pte;
1565
1566#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1567 ptl = pmd_trans_huge_lock(pmd, vma);
1568 if (ptl) {
1569 struct page *page;
1570
1571 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1572 if (page)
1573 gather_stats(page, md, pmd_dirty(*pmd),
1574 HPAGE_PMD_SIZE/PAGE_SIZE);
1575 spin_unlock(ptl);
1576 return 0;
1577 }
1578
1579 if (pmd_trans_unstable(pmd))
1580 return 0;
1581#endif
1582 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1583 do {
1584 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1585 if (!page)
1586 continue;
1587 gather_stats(page, md, pte_dirty(*pte), 1);
1588
1589 } while (pte++, addr += PAGE_SIZE, addr != end);
1590 pte_unmap_unlock(orig_pte, ptl);
1591 cond_resched();
1592 return 0;
1593}
1594#ifdef CONFIG_HUGETLB_PAGE
1595static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1596 unsigned long addr, unsigned long end, struct mm_walk *walk)
1597{
1598 pte_t huge_pte = huge_ptep_get(pte);
1599 struct numa_maps *md;
1600 struct page *page;
1601
1602 if (!pte_present(huge_pte))
1603 return 0;
1604
1605 page = pte_page(huge_pte);
1606 if (!page)
1607 return 0;
1608
1609 md = walk->private;
1610 gather_stats(page, md, pte_dirty(huge_pte), 1);
1611 return 0;
1612}
1613
1614#else
1615static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1616 unsigned long addr, unsigned long end, struct mm_walk *walk)
1617{
1618 return 0;
1619}
1620#endif
1621
1622/*
1623 * Display pages allocated per node and memory policy via /proc.
1624 */
1625static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1626{
1627 struct numa_maps_private *numa_priv = m->private;
1628 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1629 struct vm_area_struct *vma = v;
1630 struct numa_maps *md = &numa_priv->md;
1631 struct file *file = vma->vm_file;
1632 struct mm_struct *mm = vma->vm_mm;
1633 struct mm_walk walk = {
1634 .hugetlb_entry = gather_hugetlb_stats,
1635 .pmd_entry = gather_pte_stats,
1636 .private = md,
1637 .mm = mm,
1638 };
1639 struct mempolicy *pol;
1640 char buffer[64];
1641 int nid;
1642
1643 if (!mm)
1644 return 0;
1645
1646 /* Ensure we start with an empty set of numa_maps statistics. */
1647 memset(md, 0, sizeof(*md));
1648
1649 pol = __get_vma_policy(vma, vma->vm_start);
1650 if (pol) {
1651 mpol_to_str(buffer, sizeof(buffer), pol);
1652 mpol_cond_put(pol);
1653 } else {
1654 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1655 }
1656
1657 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1658
1659 if (file) {
1660 seq_puts(m, " file=");
1661 seq_file_path(m, file, "\n\t= ");
1662 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1663 seq_puts(m, " heap");
1664 } else if (is_stack(proc_priv, vma)) {
1665 seq_puts(m, " stack");
1666 }
1667
1668 if (is_vm_hugetlb_page(vma))
1669 seq_puts(m, " huge");
1670
1671 /* mmap_sem is held by m_start */
1672 walk_page_vma(vma, &walk);
1673
1674 if (!md->pages)
1675 goto out;
1676
1677 if (md->anon)
1678 seq_printf(m, " anon=%lu", md->anon);
1679
1680 if (md->dirty)
1681 seq_printf(m, " dirty=%lu", md->dirty);
1682
1683 if (md->pages != md->anon && md->pages != md->dirty)
1684 seq_printf(m, " mapped=%lu", md->pages);
1685
1686 if (md->mapcount_max > 1)
1687 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1688
1689 if (md->swapcache)
1690 seq_printf(m, " swapcache=%lu", md->swapcache);
1691
1692 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1693 seq_printf(m, " active=%lu", md->active);
1694
1695 if (md->writeback)
1696 seq_printf(m, " writeback=%lu", md->writeback);
1697
1698 for_each_node_state(nid, N_MEMORY)
1699 if (md->node[nid])
1700 seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1701
1702 seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1703out:
1704 seq_putc(m, '\n');
1705 m_cache_vma(m, vma);
1706 return 0;
1707}
1708
1709static int show_pid_numa_map(struct seq_file *m, void *v)
1710{
1711 return show_numa_map(m, v, 1);
1712}
1713
1714static int show_tid_numa_map(struct seq_file *m, void *v)
1715{
1716 return show_numa_map(m, v, 0);
1717}
1718
1719static const struct seq_operations proc_pid_numa_maps_op = {
1720 .start = m_start,
1721 .next = m_next,
1722 .stop = m_stop,
1723 .show = show_pid_numa_map,
1724};
1725
1726static const struct seq_operations proc_tid_numa_maps_op = {
1727 .start = m_start,
1728 .next = m_next,
1729 .stop = m_stop,
1730 .show = show_tid_numa_map,
1731};
1732
1733static int numa_maps_open(struct inode *inode, struct file *file,
1734 const struct seq_operations *ops)
1735{
1736 return proc_maps_open(inode, file, ops,
1737 sizeof(struct numa_maps_private));
1738}
1739
1740static int pid_numa_maps_open(struct inode *inode, struct file *file)
1741{
1742 return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1743}
1744
1745static int tid_numa_maps_open(struct inode *inode, struct file *file)
1746{
1747 return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1748}
1749
1750const struct file_operations proc_pid_numa_maps_operations = {
1751 .open = pid_numa_maps_open,
1752 .read = seq_read,
1753 .llseek = seq_lseek,
1754 .release = proc_map_release,
1755};
1756
1757const struct file_operations proc_tid_numa_maps_operations = {
1758 .open = tid_numa_maps_open,
1759 .read = seq_read,
1760 .llseek = seq_lseek,
1761 .release = proc_map_release,
1762};
1763#endif /* CONFIG_NUMA */