GNU Linux-libre 5.15.137-gnu
[releases.git] / fs / proc / task_mmu.c
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/pagewalk.h>
3 #include <linux/vmacache.h>
4 #include <linux/hugetlb.h>
5 #include <linux/huge_mm.h>
6 #include <linux/mount.h>
7 #include <linux/seq_file.h>
8 #include <linux/highmem.h>
9 #include <linux/ptrace.h>
10 #include <linux/slab.h>
11 #include <linux/pagemap.h>
12 #include <linux/mempolicy.h>
13 #include <linux/rmap.h>
14 #include <linux/swap.h>
15 #include <linux/sched/mm.h>
16 #include <linux/swapops.h>
17 #include <linux/mmu_notifier.h>
18 #include <linux/page_idle.h>
19 #include <linux/shmem_fs.h>
20 #include <linux/uaccess.h>
21 #include <linux/pkeys.h>
22
23 #include <asm/elf.h>
24 #include <asm/tlb.h>
25 #include <asm/tlbflush.h>
26 #include "internal.h"
27
28 #define SEQ_PUT_DEC(str, val) \
29                 seq_put_decimal_ull_width(m, str, (val) << (PAGE_SHIFT-10), 8)
30 void task_mem(struct seq_file *m, struct mm_struct *mm)
31 {
32         unsigned long text, lib, swap, anon, file, shmem;
33         unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
34
35         anon = get_mm_counter(mm, MM_ANONPAGES);
36         file = get_mm_counter(mm, MM_FILEPAGES);
37         shmem = get_mm_counter(mm, MM_SHMEMPAGES);
38
39         /*
40          * Note: to minimize their overhead, mm maintains hiwater_vm and
41          * hiwater_rss only when about to *lower* total_vm or rss.  Any
42          * collector of these hiwater stats must therefore get total_vm
43          * and rss too, which will usually be the higher.  Barriers? not
44          * worth the effort, such snapshots can always be inconsistent.
45          */
46         hiwater_vm = total_vm = mm->total_vm;
47         if (hiwater_vm < mm->hiwater_vm)
48                 hiwater_vm = mm->hiwater_vm;
49         hiwater_rss = total_rss = anon + file + shmem;
50         if (hiwater_rss < mm->hiwater_rss)
51                 hiwater_rss = mm->hiwater_rss;
52
53         /* split executable areas between text and lib */
54         text = PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK);
55         text = min(text, mm->exec_vm << PAGE_SHIFT);
56         lib = (mm->exec_vm << PAGE_SHIFT) - text;
57
58         swap = get_mm_counter(mm, MM_SWAPENTS);
59         SEQ_PUT_DEC("VmPeak:\t", hiwater_vm);
60         SEQ_PUT_DEC(" kB\nVmSize:\t", total_vm);
61         SEQ_PUT_DEC(" kB\nVmLck:\t", mm->locked_vm);
62         SEQ_PUT_DEC(" kB\nVmPin:\t", atomic64_read(&mm->pinned_vm));
63         SEQ_PUT_DEC(" kB\nVmHWM:\t", hiwater_rss);
64         SEQ_PUT_DEC(" kB\nVmRSS:\t", total_rss);
65         SEQ_PUT_DEC(" kB\nRssAnon:\t", anon);
66         SEQ_PUT_DEC(" kB\nRssFile:\t", file);
67         SEQ_PUT_DEC(" kB\nRssShmem:\t", shmem);
68         SEQ_PUT_DEC(" kB\nVmData:\t", mm->data_vm);
69         SEQ_PUT_DEC(" kB\nVmStk:\t", mm->stack_vm);
70         seq_put_decimal_ull_width(m,
71                     " kB\nVmExe:\t", text >> 10, 8);
72         seq_put_decimal_ull_width(m,
73                     " kB\nVmLib:\t", lib >> 10, 8);
74         seq_put_decimal_ull_width(m,
75                     " kB\nVmPTE:\t", mm_pgtables_bytes(mm) >> 10, 8);
76         SEQ_PUT_DEC(" kB\nVmSwap:\t", swap);
77         seq_puts(m, " kB\n");
78         hugetlb_report_usage(m, mm);
79 }
80 #undef SEQ_PUT_DEC
81
82 unsigned long task_vsize(struct mm_struct *mm)
83 {
84         return PAGE_SIZE * mm->total_vm;
85 }
86
87 unsigned long task_statm(struct mm_struct *mm,
88                          unsigned long *shared, unsigned long *text,
89                          unsigned long *data, unsigned long *resident)
90 {
91         *shared = get_mm_counter(mm, MM_FILEPAGES) +
92                         get_mm_counter(mm, MM_SHMEMPAGES);
93         *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
94                                                                 >> PAGE_SHIFT;
95         *data = mm->data_vm + mm->stack_vm;
96         *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
97         return mm->total_vm;
98 }
99
100 #ifdef CONFIG_NUMA
101 /*
102  * Save get_task_policy() for show_numa_map().
103  */
104 static void hold_task_mempolicy(struct proc_maps_private *priv)
105 {
106         struct task_struct *task = priv->task;
107
108         task_lock(task);
109         priv->task_mempolicy = get_task_policy(task);
110         mpol_get(priv->task_mempolicy);
111         task_unlock(task);
112 }
113 static void release_task_mempolicy(struct proc_maps_private *priv)
114 {
115         mpol_put(priv->task_mempolicy);
116 }
117 #else
118 static void hold_task_mempolicy(struct proc_maps_private *priv)
119 {
120 }
121 static void release_task_mempolicy(struct proc_maps_private *priv)
122 {
123 }
124 #endif
125
126 static void *m_start(struct seq_file *m, loff_t *ppos)
127 {
128         struct proc_maps_private *priv = m->private;
129         unsigned long last_addr = *ppos;
130         struct mm_struct *mm;
131         struct vm_area_struct *vma;
132
133         /* See m_next(). Zero at the start or after lseek. */
134         if (last_addr == -1UL)
135                 return NULL;
136
137         priv->task = get_proc_task(priv->inode);
138         if (!priv->task)
139                 return ERR_PTR(-ESRCH);
140
141         mm = priv->mm;
142         if (!mm || !mmget_not_zero(mm)) {
143                 put_task_struct(priv->task);
144                 priv->task = NULL;
145                 return NULL;
146         }
147
148         if (mmap_read_lock_killable(mm)) {
149                 mmput(mm);
150                 put_task_struct(priv->task);
151                 priv->task = NULL;
152                 return ERR_PTR(-EINTR);
153         }
154
155         hold_task_mempolicy(priv);
156         priv->tail_vma = get_gate_vma(mm);
157
158         vma = find_vma(mm, last_addr);
159         if (vma)
160                 return vma;
161
162         return priv->tail_vma;
163 }
164
165 static void *m_next(struct seq_file *m, void *v, loff_t *ppos)
166 {
167         struct proc_maps_private *priv = m->private;
168         struct vm_area_struct *next, *vma = v;
169
170         if (vma == priv->tail_vma)
171                 next = NULL;
172         else if (vma->vm_next)
173                 next = vma->vm_next;
174         else
175                 next = priv->tail_vma;
176
177         *ppos = next ? next->vm_start : -1UL;
178
179         return next;
180 }
181
182 static void m_stop(struct seq_file *m, void *v)
183 {
184         struct proc_maps_private *priv = m->private;
185         struct mm_struct *mm = priv->mm;
186
187         if (!priv->task)
188                 return;
189
190         release_task_mempolicy(priv);
191         mmap_read_unlock(mm);
192         mmput(mm);
193         put_task_struct(priv->task);
194         priv->task = NULL;
195 }
196
197 static int proc_maps_open(struct inode *inode, struct file *file,
198                         const struct seq_operations *ops, int psize)
199 {
200         struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
201
202         if (!priv)
203                 return -ENOMEM;
204
205         priv->inode = inode;
206         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
207         if (IS_ERR(priv->mm)) {
208                 int err = PTR_ERR(priv->mm);
209
210                 seq_release_private(inode, file);
211                 return err;
212         }
213
214         return 0;
215 }
216
217 static int proc_map_release(struct inode *inode, struct file *file)
218 {
219         struct seq_file *seq = file->private_data;
220         struct proc_maps_private *priv = seq->private;
221
222         if (priv->mm)
223                 mmdrop(priv->mm);
224
225         return seq_release_private(inode, file);
226 }
227
228 static int do_maps_open(struct inode *inode, struct file *file,
229                         const struct seq_operations *ops)
230 {
231         return proc_maps_open(inode, file, ops,
232                                 sizeof(struct proc_maps_private));
233 }
234
235 /*
236  * Indicate if the VMA is a stack for the given task; for
237  * /proc/PID/maps that is the stack of the main task.
238  */
239 static int is_stack(struct vm_area_struct *vma)
240 {
241         /*
242          * We make no effort to guess what a given thread considers to be
243          * its "stack".  It's not even well-defined for programs written
244          * languages like Go.
245          */
246         return vma->vm_start <= vma->vm_mm->start_stack &&
247                 vma->vm_end >= vma->vm_mm->start_stack;
248 }
249
250 static void show_vma_header_prefix(struct seq_file *m,
251                                    unsigned long start, unsigned long end,
252                                    vm_flags_t flags, unsigned long long pgoff,
253                                    dev_t dev, unsigned long ino)
254 {
255         seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
256         seq_put_hex_ll(m, NULL, start, 8);
257         seq_put_hex_ll(m, "-", end, 8);
258         seq_putc(m, ' ');
259         seq_putc(m, flags & VM_READ ? 'r' : '-');
260         seq_putc(m, flags & VM_WRITE ? 'w' : '-');
261         seq_putc(m, flags & VM_EXEC ? 'x' : '-');
262         seq_putc(m, flags & VM_MAYSHARE ? 's' : 'p');
263         seq_put_hex_ll(m, " ", pgoff, 8);
264         seq_put_hex_ll(m, " ", MAJOR(dev), 2);
265         seq_put_hex_ll(m, ":", MINOR(dev), 2);
266         seq_put_decimal_ull(m, " ", ino);
267         seq_putc(m, ' ');
268 }
269
270 static void
271 show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
272 {
273         struct mm_struct *mm = vma->vm_mm;
274         struct file *file = vma->vm_file;
275         vm_flags_t flags = vma->vm_flags;
276         unsigned long ino = 0;
277         unsigned long long pgoff = 0;
278         unsigned long start, end;
279         dev_t dev = 0;
280         const char *name = NULL;
281
282         if (file) {
283                 struct inode *inode = file_inode(vma->vm_file);
284                 dev = inode->i_sb->s_dev;
285                 ino = inode->i_ino;
286                 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
287         }
288
289         start = vma->vm_start;
290         end = vma->vm_end;
291         show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino);
292
293         /*
294          * Print the dentry name for named mappings, and a
295          * special [heap] marker for the heap:
296          */
297         if (file) {
298                 seq_pad(m, ' ');
299                 seq_file_path(m, file, "\n");
300                 goto done;
301         }
302
303         if (vma->vm_ops && vma->vm_ops->name) {
304                 name = vma->vm_ops->name(vma);
305                 if (name)
306                         goto done;
307         }
308
309         name = arch_vma_name(vma);
310         if (!name) {
311                 if (!mm) {
312                         name = "[vdso]";
313                         goto done;
314                 }
315
316                 if (vma->vm_start <= mm->brk &&
317                     vma->vm_end >= mm->start_brk) {
318                         name = "[heap]";
319                         goto done;
320                 }
321
322                 if (is_stack(vma))
323                         name = "[stack]";
324         }
325
326 done:
327         if (name) {
328                 seq_pad(m, ' ');
329                 seq_puts(m, name);
330         }
331         seq_putc(m, '\n');
332 }
333
334 static int show_map(struct seq_file *m, void *v)
335 {
336         show_map_vma(m, v);
337         return 0;
338 }
339
340 static const struct seq_operations proc_pid_maps_op = {
341         .start  = m_start,
342         .next   = m_next,
343         .stop   = m_stop,
344         .show   = show_map
345 };
346
347 static int pid_maps_open(struct inode *inode, struct file *file)
348 {
349         return do_maps_open(inode, file, &proc_pid_maps_op);
350 }
351
352 const struct file_operations proc_pid_maps_operations = {
353         .open           = pid_maps_open,
354         .read           = seq_read,
355         .llseek         = seq_lseek,
356         .release        = proc_map_release,
357 };
358
359 /*
360  * Proportional Set Size(PSS): my share of RSS.
361  *
362  * PSS of a process is the count of pages it has in memory, where each
363  * page is divided by the number of processes sharing it.  So if a
364  * process has 1000 pages all to itself, and 1000 shared with one other
365  * process, its PSS will be 1500.
366  *
367  * To keep (accumulated) division errors low, we adopt a 64bit
368  * fixed-point pss counter to minimize division errors. So (pss >>
369  * PSS_SHIFT) would be the real byte count.
370  *
371  * A shift of 12 before division means (assuming 4K page size):
372  *      - 1M 3-user-pages add up to 8KB errors;
373  *      - supports mapcount up to 2^24, or 16M;
374  *      - supports PSS up to 2^52 bytes, or 4PB.
375  */
376 #define PSS_SHIFT 12
377
378 #ifdef CONFIG_PROC_PAGE_MONITOR
379 struct mem_size_stats {
380         unsigned long resident;
381         unsigned long shared_clean;
382         unsigned long shared_dirty;
383         unsigned long private_clean;
384         unsigned long private_dirty;
385         unsigned long referenced;
386         unsigned long anonymous;
387         unsigned long lazyfree;
388         unsigned long anonymous_thp;
389         unsigned long shmem_thp;
390         unsigned long file_thp;
391         unsigned long swap;
392         unsigned long shared_hugetlb;
393         unsigned long private_hugetlb;
394         u64 pss;
395         u64 pss_anon;
396         u64 pss_file;
397         u64 pss_shmem;
398         u64 pss_locked;
399         u64 swap_pss;
400         bool check_shmem_swap;
401 };
402
403 static void smaps_page_accumulate(struct mem_size_stats *mss,
404                 struct page *page, unsigned long size, unsigned long pss,
405                 bool dirty, bool locked, bool private)
406 {
407         mss->pss += pss;
408
409         if (PageAnon(page))
410                 mss->pss_anon += pss;
411         else if (PageSwapBacked(page))
412                 mss->pss_shmem += pss;
413         else
414                 mss->pss_file += pss;
415
416         if (locked)
417                 mss->pss_locked += pss;
418
419         if (dirty || PageDirty(page)) {
420                 if (private)
421                         mss->private_dirty += size;
422                 else
423                         mss->shared_dirty += size;
424         } else {
425                 if (private)
426                         mss->private_clean += size;
427                 else
428                         mss->shared_clean += size;
429         }
430 }
431
432 static void smaps_account(struct mem_size_stats *mss, struct page *page,
433                 bool compound, bool young, bool dirty, bool locked,
434                 bool migration)
435 {
436         int i, nr = compound ? compound_nr(page) : 1;
437         unsigned long size = nr * PAGE_SIZE;
438
439         /*
440          * First accumulate quantities that depend only on |size| and the type
441          * of the compound page.
442          */
443         if (PageAnon(page)) {
444                 mss->anonymous += size;
445                 if (!PageSwapBacked(page) && !dirty && !PageDirty(page))
446                         mss->lazyfree += size;
447         }
448
449         mss->resident += size;
450         /* Accumulate the size in pages that have been accessed. */
451         if (young || page_is_young(page) || PageReferenced(page))
452                 mss->referenced += size;
453
454         /*
455          * Then accumulate quantities that may depend on sharing, or that may
456          * differ page-by-page.
457          *
458          * page_count(page) == 1 guarantees the page is mapped exactly once.
459          * If any subpage of the compound page mapped with PTE it would elevate
460          * page_count().
461          *
462          * The page_mapcount() is called to get a snapshot of the mapcount.
463          * Without holding the page lock this snapshot can be slightly wrong as
464          * we cannot always read the mapcount atomically.  It is not safe to
465          * call page_mapcount() even with PTL held if the page is not mapped,
466          * especially for migration entries.  Treat regular migration entries
467          * as mapcount == 1.
468          */
469         if ((page_count(page) == 1) || migration) {
470                 smaps_page_accumulate(mss, page, size, size << PSS_SHIFT, dirty,
471                         locked, true);
472                 return;
473         }
474         for (i = 0; i < nr; i++, page++) {
475                 int mapcount = page_mapcount(page);
476                 unsigned long pss = PAGE_SIZE << PSS_SHIFT;
477                 if (mapcount >= 2)
478                         pss /= mapcount;
479                 smaps_page_accumulate(mss, page, PAGE_SIZE, pss, dirty, locked,
480                                       mapcount < 2);
481         }
482 }
483
484 #ifdef CONFIG_SHMEM
485 static int smaps_pte_hole(unsigned long addr, unsigned long end,
486                           __always_unused int depth, struct mm_walk *walk)
487 {
488         struct mem_size_stats *mss = walk->private;
489
490         mss->swap += shmem_partial_swap_usage(
491                         walk->vma->vm_file->f_mapping, addr, end);
492
493         return 0;
494 }
495 #else
496 #define smaps_pte_hole          NULL
497 #endif /* CONFIG_SHMEM */
498
499 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
500                 struct mm_walk *walk)
501 {
502         struct mem_size_stats *mss = walk->private;
503         struct vm_area_struct *vma = walk->vma;
504         bool locked = !!(vma->vm_flags & VM_LOCKED);
505         struct page *page = NULL;
506         bool migration = false, young = false, dirty = false;
507
508         if (pte_present(*pte)) {
509                 page = vm_normal_page(vma, addr, *pte);
510                 young = pte_young(*pte);
511                 dirty = pte_dirty(*pte);
512         } else if (is_swap_pte(*pte)) {
513                 swp_entry_t swpent = pte_to_swp_entry(*pte);
514
515                 if (!non_swap_entry(swpent)) {
516                         int mapcount;
517
518                         mss->swap += PAGE_SIZE;
519                         mapcount = swp_swapcount(swpent);
520                         if (mapcount >= 2) {
521                                 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
522
523                                 do_div(pss_delta, mapcount);
524                                 mss->swap_pss += pss_delta;
525                         } else {
526                                 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
527                         }
528                 } else if (is_pfn_swap_entry(swpent)) {
529                         if (is_migration_entry(swpent))
530                                 migration = true;
531                         page = pfn_swap_entry_to_page(swpent);
532                 }
533         } else if (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap
534                                                         && pte_none(*pte))) {
535                 page = xa_load(&vma->vm_file->f_mapping->i_pages,
536                                                 linear_page_index(vma, addr));
537                 if (xa_is_value(page))
538                         mss->swap += PAGE_SIZE;
539                 return;
540         }
541
542         if (!page)
543                 return;
544
545         smaps_account(mss, page, false, young, dirty, locked, migration);
546 }
547
548 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
549 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
550                 struct mm_walk *walk)
551 {
552         struct mem_size_stats *mss = walk->private;
553         struct vm_area_struct *vma = walk->vma;
554         bool locked = !!(vma->vm_flags & VM_LOCKED);
555         struct page *page = NULL;
556         bool migration = false;
557
558         if (pmd_present(*pmd)) {
559                 /* FOLL_DUMP will return -EFAULT on huge zero page */
560                 page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
561         } else if (unlikely(thp_migration_supported() && is_swap_pmd(*pmd))) {
562                 swp_entry_t entry = pmd_to_swp_entry(*pmd);
563
564                 if (is_migration_entry(entry)) {
565                         migration = true;
566                         page = pfn_swap_entry_to_page(entry);
567                 }
568         }
569         if (IS_ERR_OR_NULL(page))
570                 return;
571         if (PageAnon(page))
572                 mss->anonymous_thp += HPAGE_PMD_SIZE;
573         else if (PageSwapBacked(page))
574                 mss->shmem_thp += HPAGE_PMD_SIZE;
575         else if (is_zone_device_page(page))
576                 /* pass */;
577         else
578                 mss->file_thp += HPAGE_PMD_SIZE;
579
580         smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd),
581                       locked, migration);
582 }
583 #else
584 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
585                 struct mm_walk *walk)
586 {
587 }
588 #endif
589
590 static 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                 goto out;
602         }
603
604         if (pmd_trans_unstable(pmd))
605                 goto out;
606         /*
607          * The mmap_lock 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 out:
616         cond_resched();
617         return 0;
618 }
619
620 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
621 {
622         /*
623          * Don't forget to update Documentation/ on changes.
624          */
625         static const char mnemonics[BITS_PER_LONG][2] = {
626                 /*
627                  * In case if we meet a flag we don't know about.
628                  */
629                 [0 ... (BITS_PER_LONG-1)] = "??",
630
631                 [ilog2(VM_READ)]        = "rd",
632                 [ilog2(VM_WRITE)]       = "wr",
633                 [ilog2(VM_EXEC)]        = "ex",
634                 [ilog2(VM_SHARED)]      = "sh",
635                 [ilog2(VM_MAYREAD)]     = "mr",
636                 [ilog2(VM_MAYWRITE)]    = "mw",
637                 [ilog2(VM_MAYEXEC)]     = "me",
638                 [ilog2(VM_MAYSHARE)]    = "ms",
639                 [ilog2(VM_GROWSDOWN)]   = "gd",
640                 [ilog2(VM_PFNMAP)]      = "pf",
641                 [ilog2(VM_LOCKED)]      = "lo",
642                 [ilog2(VM_IO)]          = "io",
643                 [ilog2(VM_SEQ_READ)]    = "sr",
644                 [ilog2(VM_RAND_READ)]   = "rr",
645                 [ilog2(VM_DONTCOPY)]    = "dc",
646                 [ilog2(VM_DONTEXPAND)]  = "de",
647                 [ilog2(VM_ACCOUNT)]     = "ac",
648                 [ilog2(VM_NORESERVE)]   = "nr",
649                 [ilog2(VM_HUGETLB)]     = "ht",
650                 [ilog2(VM_SYNC)]        = "sf",
651                 [ilog2(VM_ARCH_1)]      = "ar",
652                 [ilog2(VM_WIPEONFORK)]  = "wf",
653                 [ilog2(VM_DONTDUMP)]    = "dd",
654 #ifdef CONFIG_ARM64_BTI
655                 [ilog2(VM_ARM64_BTI)]   = "bt",
656 #endif
657 #ifdef CONFIG_MEM_SOFT_DIRTY
658                 [ilog2(VM_SOFTDIRTY)]   = "sd",
659 #endif
660                 [ilog2(VM_MIXEDMAP)]    = "mm",
661                 [ilog2(VM_HUGEPAGE)]    = "hg",
662                 [ilog2(VM_NOHUGEPAGE)]  = "nh",
663                 [ilog2(VM_MERGEABLE)]   = "mg",
664                 [ilog2(VM_UFFD_MISSING)]= "um",
665                 [ilog2(VM_UFFD_WP)]     = "uw",
666 #ifdef CONFIG_ARM64_MTE
667                 [ilog2(VM_MTE)]         = "mt",
668                 [ilog2(VM_MTE_ALLOWED)] = "",
669 #endif
670 #ifdef CONFIG_ARCH_HAS_PKEYS
671                 /* These come out via ProtectionKey: */
672                 [ilog2(VM_PKEY_BIT0)]   = "",
673                 [ilog2(VM_PKEY_BIT1)]   = "",
674                 [ilog2(VM_PKEY_BIT2)]   = "",
675                 [ilog2(VM_PKEY_BIT3)]   = "",
676 #if VM_PKEY_BIT4
677                 [ilog2(VM_PKEY_BIT4)]   = "",
678 #endif
679 #endif /* CONFIG_ARCH_HAS_PKEYS */
680 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_MINOR
681                 [ilog2(VM_UFFD_MINOR)]  = "ui",
682 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_MINOR */
683         };
684         size_t i;
685
686         seq_puts(m, "VmFlags: ");
687         for (i = 0; i < BITS_PER_LONG; i++) {
688                 if (!mnemonics[i][0])
689                         continue;
690                 if (vma->vm_flags & (1UL << i)) {
691                         seq_putc(m, mnemonics[i][0]);
692                         seq_putc(m, mnemonics[i][1]);
693                         seq_putc(m, ' ');
694                 }
695         }
696         seq_putc(m, '\n');
697 }
698
699 #ifdef CONFIG_HUGETLB_PAGE
700 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
701                                  unsigned long addr, unsigned long end,
702                                  struct mm_walk *walk)
703 {
704         struct mem_size_stats *mss = walk->private;
705         struct vm_area_struct *vma = walk->vma;
706         struct page *page = NULL;
707
708         if (pte_present(*pte)) {
709                 page = vm_normal_page(vma, addr, *pte);
710         } else if (is_swap_pte(*pte)) {
711                 swp_entry_t swpent = pte_to_swp_entry(*pte);
712
713                 if (is_pfn_swap_entry(swpent))
714                         page = pfn_swap_entry_to_page(swpent);
715         }
716         if (page) {
717                 if (page_mapcount(page) >= 2 || hugetlb_pmd_shared(pte))
718                         mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
719                 else
720                         mss->private_hugetlb += huge_page_size(hstate_vma(vma));
721         }
722         return 0;
723 }
724 #else
725 #define smaps_hugetlb_range     NULL
726 #endif /* HUGETLB_PAGE */
727
728 static const struct mm_walk_ops smaps_walk_ops = {
729         .pmd_entry              = smaps_pte_range,
730         .hugetlb_entry          = smaps_hugetlb_range,
731 };
732
733 static const struct mm_walk_ops smaps_shmem_walk_ops = {
734         .pmd_entry              = smaps_pte_range,
735         .hugetlb_entry          = smaps_hugetlb_range,
736         .pte_hole               = smaps_pte_hole,
737 };
738
739 /*
740  * Gather mem stats from @vma with the indicated beginning
741  * address @start, and keep them in @mss.
742  *
743  * Use vm_start of @vma as the beginning address if @start is 0.
744  */
745 static void smap_gather_stats(struct vm_area_struct *vma,
746                 struct mem_size_stats *mss, unsigned long start)
747 {
748         const struct mm_walk_ops *ops = &smaps_walk_ops;
749
750         /* Invalid start */
751         if (start >= vma->vm_end)
752                 return;
753
754 #ifdef CONFIG_SHMEM
755         /* In case of smaps_rollup, reset the value from previous vma */
756         mss->check_shmem_swap = false;
757         if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
758                 /*
759                  * For shared or readonly shmem mappings we know that all
760                  * swapped out pages belong to the shmem object, and we can
761                  * obtain the swap value much more efficiently. For private
762                  * writable mappings, we might have COW pages that are
763                  * not affected by the parent swapped out pages of the shmem
764                  * object, so we have to distinguish them during the page walk.
765                  * Unless we know that the shmem object (or the part mapped by
766                  * our VMA) has no swapped out pages at all.
767                  */
768                 unsigned long shmem_swapped = shmem_swap_usage(vma);
769
770                 if (!start && (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
771                                         !(vma->vm_flags & VM_WRITE))) {
772                         mss->swap += shmem_swapped;
773                 } else {
774                         mss->check_shmem_swap = true;
775                         ops = &smaps_shmem_walk_ops;
776                 }
777         }
778 #endif
779         /* mmap_lock is held in m_start */
780         if (!start)
781                 walk_page_vma(vma, ops, mss);
782         else
783                 walk_page_range(vma->vm_mm, start, vma->vm_end, ops, mss);
784 }
785
786 #define SEQ_PUT_DEC(str, val) \
787                 seq_put_decimal_ull_width(m, str, (val) >> 10, 8)
788
789 /* Show the contents common for smaps and smaps_rollup */
790 static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss,
791         bool rollup_mode)
792 {
793         SEQ_PUT_DEC("Rss:            ", mss->resident);
794         SEQ_PUT_DEC(" kB\nPss:            ", mss->pss >> PSS_SHIFT);
795         if (rollup_mode) {
796                 /*
797                  * These are meaningful only for smaps_rollup, otherwise two of
798                  * them are zero, and the other one is the same as Pss.
799                  */
800                 SEQ_PUT_DEC(" kB\nPss_Anon:       ",
801                         mss->pss_anon >> PSS_SHIFT);
802                 SEQ_PUT_DEC(" kB\nPss_File:       ",
803                         mss->pss_file >> PSS_SHIFT);
804                 SEQ_PUT_DEC(" kB\nPss_Shmem:      ",
805                         mss->pss_shmem >> PSS_SHIFT);
806         }
807         SEQ_PUT_DEC(" kB\nShared_Clean:   ", mss->shared_clean);
808         SEQ_PUT_DEC(" kB\nShared_Dirty:   ", mss->shared_dirty);
809         SEQ_PUT_DEC(" kB\nPrivate_Clean:  ", mss->private_clean);
810         SEQ_PUT_DEC(" kB\nPrivate_Dirty:  ", mss->private_dirty);
811         SEQ_PUT_DEC(" kB\nReferenced:     ", mss->referenced);
812         SEQ_PUT_DEC(" kB\nAnonymous:      ", mss->anonymous);
813         SEQ_PUT_DEC(" kB\nLazyFree:       ", mss->lazyfree);
814         SEQ_PUT_DEC(" kB\nAnonHugePages:  ", mss->anonymous_thp);
815         SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp);
816         SEQ_PUT_DEC(" kB\nFilePmdMapped:  ", mss->file_thp);
817         SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb);
818         seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ",
819                                   mss->private_hugetlb >> 10, 7);
820         SEQ_PUT_DEC(" kB\nSwap:           ", mss->swap);
821         SEQ_PUT_DEC(" kB\nSwapPss:        ",
822                                         mss->swap_pss >> PSS_SHIFT);
823         SEQ_PUT_DEC(" kB\nLocked:         ",
824                                         mss->pss_locked >> PSS_SHIFT);
825         seq_puts(m, " kB\n");
826 }
827
828 static int show_smap(struct seq_file *m, void *v)
829 {
830         struct vm_area_struct *vma = v;
831         struct mem_size_stats mss;
832
833         memset(&mss, 0, sizeof(mss));
834
835         smap_gather_stats(vma, &mss, 0);
836
837         show_map_vma(m, vma);
838
839         SEQ_PUT_DEC("Size:           ", vma->vm_end - vma->vm_start);
840         SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma));
841         SEQ_PUT_DEC(" kB\nMMUPageSize:    ", vma_mmu_pagesize(vma));
842         seq_puts(m, " kB\n");
843
844         __show_smap(m, &mss, false);
845
846         seq_printf(m, "THPeligible:    %d\n",
847                    transparent_hugepage_active(vma));
848
849         if (arch_pkeys_enabled())
850                 seq_printf(m, "ProtectionKey:  %8u\n", vma_pkey(vma));
851         show_smap_vma_flags(m, vma);
852
853         return 0;
854 }
855
856 static int show_smaps_rollup(struct seq_file *m, void *v)
857 {
858         struct proc_maps_private *priv = m->private;
859         struct mem_size_stats mss;
860         struct mm_struct *mm;
861         struct vm_area_struct *vma;
862         unsigned long last_vma_end = 0;
863         int ret = 0;
864
865         priv->task = get_proc_task(priv->inode);
866         if (!priv->task)
867                 return -ESRCH;
868
869         mm = priv->mm;
870         if (!mm || !mmget_not_zero(mm)) {
871                 ret = -ESRCH;
872                 goto out_put_task;
873         }
874
875         memset(&mss, 0, sizeof(mss));
876
877         ret = mmap_read_lock_killable(mm);
878         if (ret)
879                 goto out_put_mm;
880
881         hold_task_mempolicy(priv);
882
883         for (vma = priv->mm->mmap; vma;) {
884                 smap_gather_stats(vma, &mss, 0);
885                 last_vma_end = vma->vm_end;
886
887                 /*
888                  * Release mmap_lock temporarily if someone wants to
889                  * access it for write request.
890                  */
891                 if (mmap_lock_is_contended(mm)) {
892                         mmap_read_unlock(mm);
893                         ret = mmap_read_lock_killable(mm);
894                         if (ret) {
895                                 release_task_mempolicy(priv);
896                                 goto out_put_mm;
897                         }
898
899                         /*
900                          * After dropping the lock, there are four cases to
901                          * consider. See the following example for explanation.
902                          *
903                          *   +------+------+-----------+
904                          *   | VMA1 | VMA2 | VMA3      |
905                          *   +------+------+-----------+
906                          *   |      |      |           |
907                          *  4k     8k     16k         400k
908                          *
909                          * Suppose we drop the lock after reading VMA2 due to
910                          * contention, then we get:
911                          *
912                          *      last_vma_end = 16k
913                          *
914                          * 1) VMA2 is freed, but VMA3 exists:
915                          *
916                          *    find_vma(mm, 16k - 1) will return VMA3.
917                          *    In this case, just continue from VMA3.
918                          *
919                          * 2) VMA2 still exists:
920                          *
921                          *    find_vma(mm, 16k - 1) will return VMA2.
922                          *    Iterate the loop like the original one.
923                          *
924                          * 3) No more VMAs can be found:
925                          *
926                          *    find_vma(mm, 16k - 1) will return NULL.
927                          *    No more things to do, just break.
928                          *
929                          * 4) (last_vma_end - 1) is the middle of a vma (VMA'):
930                          *
931                          *    find_vma(mm, 16k - 1) will return VMA' whose range
932                          *    contains last_vma_end.
933                          *    Iterate VMA' from last_vma_end.
934                          */
935                         vma = find_vma(mm, last_vma_end - 1);
936                         /* Case 3 above */
937                         if (!vma)
938                                 break;
939
940                         /* Case 1 above */
941                         if (vma->vm_start >= last_vma_end)
942                                 continue;
943
944                         /* Case 4 above */
945                         if (vma->vm_end > last_vma_end)
946                                 smap_gather_stats(vma, &mss, last_vma_end);
947                 }
948                 /* Case 2 above */
949                 vma = vma->vm_next;
950         }
951
952         show_vma_header_prefix(m, priv->mm->mmap ? priv->mm->mmap->vm_start : 0,
953                                last_vma_end, 0, 0, 0, 0);
954         seq_pad(m, ' ');
955         seq_puts(m, "[rollup]\n");
956
957         __show_smap(m, &mss, true);
958
959         release_task_mempolicy(priv);
960         mmap_read_unlock(mm);
961
962 out_put_mm:
963         mmput(mm);
964 out_put_task:
965         put_task_struct(priv->task);
966         priv->task = NULL;
967
968         return ret;
969 }
970 #undef SEQ_PUT_DEC
971
972 static const struct seq_operations proc_pid_smaps_op = {
973         .start  = m_start,
974         .next   = m_next,
975         .stop   = m_stop,
976         .show   = show_smap
977 };
978
979 static int pid_smaps_open(struct inode *inode, struct file *file)
980 {
981         return do_maps_open(inode, file, &proc_pid_smaps_op);
982 }
983
984 static int smaps_rollup_open(struct inode *inode, struct file *file)
985 {
986         int ret;
987         struct proc_maps_private *priv;
988
989         priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT);
990         if (!priv)
991                 return -ENOMEM;
992
993         ret = single_open(file, show_smaps_rollup, priv);
994         if (ret)
995                 goto out_free;
996
997         priv->inode = inode;
998         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
999         if (IS_ERR(priv->mm)) {
1000                 ret = PTR_ERR(priv->mm);
1001
1002                 single_release(inode, file);
1003                 goto out_free;
1004         }
1005
1006         return 0;
1007
1008 out_free:
1009         kfree(priv);
1010         return ret;
1011 }
1012
1013 static int smaps_rollup_release(struct inode *inode, struct file *file)
1014 {
1015         struct seq_file *seq = file->private_data;
1016         struct proc_maps_private *priv = seq->private;
1017
1018         if (priv->mm)
1019                 mmdrop(priv->mm);
1020
1021         kfree(priv);
1022         return single_release(inode, file);
1023 }
1024
1025 const struct file_operations proc_pid_smaps_operations = {
1026         .open           = pid_smaps_open,
1027         .read           = seq_read,
1028         .llseek         = seq_lseek,
1029         .release        = proc_map_release,
1030 };
1031
1032 const struct file_operations proc_pid_smaps_rollup_operations = {
1033         .open           = smaps_rollup_open,
1034         .read           = seq_read,
1035         .llseek         = seq_lseek,
1036         .release        = smaps_rollup_release,
1037 };
1038
1039 enum clear_refs_types {
1040         CLEAR_REFS_ALL = 1,
1041         CLEAR_REFS_ANON,
1042         CLEAR_REFS_MAPPED,
1043         CLEAR_REFS_SOFT_DIRTY,
1044         CLEAR_REFS_MM_HIWATER_RSS,
1045         CLEAR_REFS_LAST,
1046 };
1047
1048 struct clear_refs_private {
1049         enum clear_refs_types type;
1050 };
1051
1052 #ifdef CONFIG_MEM_SOFT_DIRTY
1053
1054 static inline bool pte_is_pinned(struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1055 {
1056         struct page *page;
1057
1058         if (!pte_write(pte))
1059                 return false;
1060         if (!is_cow_mapping(vma->vm_flags))
1061                 return false;
1062         if (likely(!test_bit(MMF_HAS_PINNED, &vma->vm_mm->flags)))
1063                 return false;
1064         page = vm_normal_page(vma, addr, pte);
1065         if (!page)
1066                 return false;
1067         return page_maybe_dma_pinned(page);
1068 }
1069
1070 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1071                 unsigned long addr, pte_t *pte)
1072 {
1073         /*
1074          * The soft-dirty tracker uses #PF-s to catch writes
1075          * to pages, so write-protect the pte as well. See the
1076          * Documentation/admin-guide/mm/soft-dirty.rst for full description
1077          * of how soft-dirty works.
1078          */
1079         pte_t ptent = *pte;
1080
1081         if (pte_present(ptent)) {
1082                 pte_t old_pte;
1083
1084                 if (pte_is_pinned(vma, addr, ptent))
1085                         return;
1086                 old_pte = ptep_modify_prot_start(vma, addr, pte);
1087                 ptent = pte_wrprotect(old_pte);
1088                 ptent = pte_clear_soft_dirty(ptent);
1089                 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent);
1090         } else if (is_swap_pte(ptent)) {
1091                 ptent = pte_swp_clear_soft_dirty(ptent);
1092                 set_pte_at(vma->vm_mm, addr, pte, ptent);
1093         }
1094 }
1095 #else
1096 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1097                 unsigned long addr, pte_t *pte)
1098 {
1099 }
1100 #endif
1101
1102 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
1103 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1104                 unsigned long addr, pmd_t *pmdp)
1105 {
1106         pmd_t old, pmd = *pmdp;
1107
1108         if (pmd_present(pmd)) {
1109                 /* See comment in change_huge_pmd() */
1110                 old = pmdp_invalidate(vma, addr, pmdp);
1111                 if (pmd_dirty(old))
1112                         pmd = pmd_mkdirty(pmd);
1113                 if (pmd_young(old))
1114                         pmd = pmd_mkyoung(pmd);
1115
1116                 pmd = pmd_wrprotect(pmd);
1117                 pmd = pmd_clear_soft_dirty(pmd);
1118
1119                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1120         } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1121                 pmd = pmd_swp_clear_soft_dirty(pmd);
1122                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1123         }
1124 }
1125 #else
1126 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1127                 unsigned long addr, pmd_t *pmdp)
1128 {
1129 }
1130 #endif
1131
1132 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
1133                                 unsigned long end, struct mm_walk *walk)
1134 {
1135         struct clear_refs_private *cp = walk->private;
1136         struct vm_area_struct *vma = walk->vma;
1137         pte_t *pte, ptent;
1138         spinlock_t *ptl;
1139         struct page *page;
1140
1141         ptl = pmd_trans_huge_lock(pmd, vma);
1142         if (ptl) {
1143                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1144                         clear_soft_dirty_pmd(vma, addr, pmd);
1145                         goto out;
1146                 }
1147
1148                 if (!pmd_present(*pmd))
1149                         goto out;
1150
1151                 page = pmd_page(*pmd);
1152
1153                 /* Clear accessed and referenced bits. */
1154                 pmdp_test_and_clear_young(vma, addr, pmd);
1155                 test_and_clear_page_young(page);
1156                 ClearPageReferenced(page);
1157 out:
1158                 spin_unlock(ptl);
1159                 return 0;
1160         }
1161
1162         if (pmd_trans_unstable(pmd))
1163                 return 0;
1164
1165         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1166         for (; addr != end; pte++, addr += PAGE_SIZE) {
1167                 ptent = *pte;
1168
1169                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1170                         clear_soft_dirty(vma, addr, pte);
1171                         continue;
1172                 }
1173
1174                 if (!pte_present(ptent))
1175                         continue;
1176
1177                 page = vm_normal_page(vma, addr, ptent);
1178                 if (!page)
1179                         continue;
1180
1181                 /* Clear accessed and referenced bits. */
1182                 ptep_test_and_clear_young(vma, addr, pte);
1183                 test_and_clear_page_young(page);
1184                 ClearPageReferenced(page);
1185         }
1186         pte_unmap_unlock(pte - 1, ptl);
1187         cond_resched();
1188         return 0;
1189 }
1190
1191 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1192                                 struct mm_walk *walk)
1193 {
1194         struct clear_refs_private *cp = walk->private;
1195         struct vm_area_struct *vma = walk->vma;
1196
1197         if (vma->vm_flags & VM_PFNMAP)
1198                 return 1;
1199
1200         /*
1201          * Writing 1 to /proc/pid/clear_refs affects all pages.
1202          * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1203          * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1204          * Writing 4 to /proc/pid/clear_refs affects all pages.
1205          */
1206         if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1207                 return 1;
1208         if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1209                 return 1;
1210         return 0;
1211 }
1212
1213 static const struct mm_walk_ops clear_refs_walk_ops = {
1214         .pmd_entry              = clear_refs_pte_range,
1215         .test_walk              = clear_refs_test_walk,
1216 };
1217
1218 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1219                                 size_t count, loff_t *ppos)
1220 {
1221         struct task_struct *task;
1222         char buffer[PROC_NUMBUF];
1223         struct mm_struct *mm;
1224         struct vm_area_struct *vma;
1225         enum clear_refs_types type;
1226         int itype;
1227         int rv;
1228
1229         memset(buffer, 0, sizeof(buffer));
1230         if (count > sizeof(buffer) - 1)
1231                 count = sizeof(buffer) - 1;
1232         if (copy_from_user(buffer, buf, count))
1233                 return -EFAULT;
1234         rv = kstrtoint(strstrip(buffer), 10, &itype);
1235         if (rv < 0)
1236                 return rv;
1237         type = (enum clear_refs_types)itype;
1238         if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1239                 return -EINVAL;
1240
1241         task = get_proc_task(file_inode(file));
1242         if (!task)
1243                 return -ESRCH;
1244         mm = get_task_mm(task);
1245         if (mm) {
1246                 struct mmu_notifier_range range;
1247                 struct clear_refs_private cp = {
1248                         .type = type,
1249                 };
1250
1251                 if (mmap_write_lock_killable(mm)) {
1252                         count = -EINTR;
1253                         goto out_mm;
1254                 }
1255                 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1256                         /*
1257                          * Writing 5 to /proc/pid/clear_refs resets the peak
1258                          * resident set size to this mm's current rss value.
1259                          */
1260                         reset_mm_hiwater_rss(mm);
1261                         goto out_unlock;
1262                 }
1263
1264                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1265                         for (vma = mm->mmap; vma; vma = vma->vm_next) {
1266                                 if (!(vma->vm_flags & VM_SOFTDIRTY))
1267                                         continue;
1268                                 vma->vm_flags &= ~VM_SOFTDIRTY;
1269                                 vma_set_page_prot(vma);
1270                         }
1271
1272                         inc_tlb_flush_pending(mm);
1273                         mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY,
1274                                                 0, NULL, mm, 0, -1UL);
1275                         mmu_notifier_invalidate_range_start(&range);
1276                 }
1277                 walk_page_range(mm, 0, mm->highest_vm_end, &clear_refs_walk_ops,
1278                                 &cp);
1279                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1280                         mmu_notifier_invalidate_range_end(&range);
1281                         flush_tlb_mm(mm);
1282                         dec_tlb_flush_pending(mm);
1283                 }
1284 out_unlock:
1285                 mmap_write_unlock(mm);
1286 out_mm:
1287                 mmput(mm);
1288         }
1289         put_task_struct(task);
1290
1291         return count;
1292 }
1293
1294 const struct file_operations proc_clear_refs_operations = {
1295         .write          = clear_refs_write,
1296         .llseek         = noop_llseek,
1297 };
1298
1299 typedef struct {
1300         u64 pme;
1301 } pagemap_entry_t;
1302
1303 struct pagemapread {
1304         int pos, len;           /* units: PM_ENTRY_BYTES, not bytes */
1305         pagemap_entry_t *buffer;
1306         bool show_pfn;
1307 };
1308
1309 #define PAGEMAP_WALK_SIZE       (PMD_SIZE)
1310 #define PAGEMAP_WALK_MASK       (PMD_MASK)
1311
1312 #define PM_ENTRY_BYTES          sizeof(pagemap_entry_t)
1313 #define PM_PFRAME_BITS          55
1314 #define PM_PFRAME_MASK          GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1315 #define PM_SOFT_DIRTY           BIT_ULL(55)
1316 #define PM_MMAP_EXCLUSIVE       BIT_ULL(56)
1317 #define PM_UFFD_WP              BIT_ULL(57)
1318 #define PM_FILE                 BIT_ULL(61)
1319 #define PM_SWAP                 BIT_ULL(62)
1320 #define PM_PRESENT              BIT_ULL(63)
1321
1322 #define PM_END_OF_BUFFER    1
1323
1324 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1325 {
1326         return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1327 }
1328
1329 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1330                           struct pagemapread *pm)
1331 {
1332         pm->buffer[pm->pos++] = *pme;
1333         if (pm->pos >= pm->len)
1334                 return PM_END_OF_BUFFER;
1335         return 0;
1336 }
1337
1338 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1339                             __always_unused int depth, struct mm_walk *walk)
1340 {
1341         struct pagemapread *pm = walk->private;
1342         unsigned long addr = start;
1343         int err = 0;
1344
1345         while (addr < end) {
1346                 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1347                 pagemap_entry_t pme = make_pme(0, 0);
1348                 /* End of address space hole, which we mark as non-present. */
1349                 unsigned long hole_end;
1350
1351                 if (vma)
1352                         hole_end = min(end, vma->vm_start);
1353                 else
1354                         hole_end = end;
1355
1356                 for (; addr < hole_end; addr += PAGE_SIZE) {
1357                         err = add_to_pagemap(addr, &pme, pm);
1358                         if (err)
1359                                 goto out;
1360                 }
1361
1362                 if (!vma)
1363                         break;
1364
1365                 /* Addresses in the VMA. */
1366                 if (vma->vm_flags & VM_SOFTDIRTY)
1367                         pme = make_pme(0, PM_SOFT_DIRTY);
1368                 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1369                         err = add_to_pagemap(addr, &pme, pm);
1370                         if (err)
1371                                 goto out;
1372                 }
1373         }
1374 out:
1375         return err;
1376 }
1377
1378 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1379                 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1380 {
1381         u64 frame = 0, flags = 0;
1382         struct page *page = NULL;
1383         bool migration = false;
1384
1385         if (pte_present(pte)) {
1386                 if (pm->show_pfn)
1387                         frame = pte_pfn(pte);
1388                 flags |= PM_PRESENT;
1389                 page = vm_normal_page(vma, addr, pte);
1390                 if (pte_soft_dirty(pte))
1391                         flags |= PM_SOFT_DIRTY;
1392                 if (pte_uffd_wp(pte))
1393                         flags |= PM_UFFD_WP;
1394         } else if (is_swap_pte(pte)) {
1395                 swp_entry_t entry;
1396                 if (pte_swp_soft_dirty(pte))
1397                         flags |= PM_SOFT_DIRTY;
1398                 if (pte_swp_uffd_wp(pte))
1399                         flags |= PM_UFFD_WP;
1400                 entry = pte_to_swp_entry(pte);
1401                 if (pm->show_pfn)
1402                         frame = swp_type(entry) |
1403                                 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1404                 flags |= PM_SWAP;
1405                 migration = is_migration_entry(entry);
1406                 if (is_pfn_swap_entry(entry))
1407                         page = pfn_swap_entry_to_page(entry);
1408         }
1409
1410         if (page && !PageAnon(page))
1411                 flags |= PM_FILE;
1412         if (page && !migration && page_mapcount(page) == 1)
1413                 flags |= PM_MMAP_EXCLUSIVE;
1414         if (vma->vm_flags & VM_SOFTDIRTY)
1415                 flags |= PM_SOFT_DIRTY;
1416
1417         return make_pme(frame, flags);
1418 }
1419
1420 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1421                              struct mm_walk *walk)
1422 {
1423         struct vm_area_struct *vma = walk->vma;
1424         struct pagemapread *pm = walk->private;
1425         spinlock_t *ptl;
1426         pte_t *pte, *orig_pte;
1427         int err = 0;
1428 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1429         bool migration = false;
1430
1431         ptl = pmd_trans_huge_lock(pmdp, vma);
1432         if (ptl) {
1433                 u64 flags = 0, frame = 0;
1434                 pmd_t pmd = *pmdp;
1435                 struct page *page = NULL;
1436
1437                 if (vma->vm_flags & VM_SOFTDIRTY)
1438                         flags |= PM_SOFT_DIRTY;
1439
1440                 if (pmd_present(pmd)) {
1441                         page = pmd_page(pmd);
1442
1443                         flags |= PM_PRESENT;
1444                         if (pmd_soft_dirty(pmd))
1445                                 flags |= PM_SOFT_DIRTY;
1446                         if (pmd_uffd_wp(pmd))
1447                                 flags |= PM_UFFD_WP;
1448                         if (pm->show_pfn)
1449                                 frame = pmd_pfn(pmd) +
1450                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1451                 }
1452 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1453                 else if (is_swap_pmd(pmd)) {
1454                         swp_entry_t entry = pmd_to_swp_entry(pmd);
1455                         unsigned long offset;
1456
1457                         if (pm->show_pfn) {
1458                                 offset = swp_offset(entry) +
1459                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1460                                 frame = swp_type(entry) |
1461                                         (offset << MAX_SWAPFILES_SHIFT);
1462                         }
1463                         flags |= PM_SWAP;
1464                         if (pmd_swp_soft_dirty(pmd))
1465                                 flags |= PM_SOFT_DIRTY;
1466                         if (pmd_swp_uffd_wp(pmd))
1467                                 flags |= PM_UFFD_WP;
1468                         VM_BUG_ON(!is_pmd_migration_entry(pmd));
1469                         migration = is_migration_entry(entry);
1470                         page = pfn_swap_entry_to_page(entry);
1471                 }
1472 #endif
1473
1474                 if (page && !migration && page_mapcount(page) == 1)
1475                         flags |= PM_MMAP_EXCLUSIVE;
1476
1477                 for (; addr != end; addr += PAGE_SIZE) {
1478                         pagemap_entry_t pme = make_pme(frame, flags);
1479
1480                         err = add_to_pagemap(addr, &pme, pm);
1481                         if (err)
1482                                 break;
1483                         if (pm->show_pfn) {
1484                                 if (flags & PM_PRESENT)
1485                                         frame++;
1486                                 else if (flags & PM_SWAP)
1487                                         frame += (1 << MAX_SWAPFILES_SHIFT);
1488                         }
1489                 }
1490                 spin_unlock(ptl);
1491                 return err;
1492         }
1493
1494         if (pmd_trans_unstable(pmdp))
1495                 return 0;
1496 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1497
1498         /*
1499          * We can assume that @vma always points to a valid one and @end never
1500          * goes beyond vma->vm_end.
1501          */
1502         orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1503         for (; addr < end; pte++, addr += PAGE_SIZE) {
1504                 pagemap_entry_t pme;
1505
1506                 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1507                 err = add_to_pagemap(addr, &pme, pm);
1508                 if (err)
1509                         break;
1510         }
1511         pte_unmap_unlock(orig_pte, ptl);
1512
1513         cond_resched();
1514
1515         return err;
1516 }
1517
1518 #ifdef CONFIG_HUGETLB_PAGE
1519 /* This function walks within one hugetlb entry in the single call */
1520 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1521                                  unsigned long addr, unsigned long end,
1522                                  struct mm_walk *walk)
1523 {
1524         struct pagemapread *pm = walk->private;
1525         struct vm_area_struct *vma = walk->vma;
1526         u64 flags = 0, frame = 0;
1527         int err = 0;
1528         pte_t pte;
1529
1530         if (vma->vm_flags & VM_SOFTDIRTY)
1531                 flags |= PM_SOFT_DIRTY;
1532
1533         pte = huge_ptep_get(ptep);
1534         if (pte_present(pte)) {
1535                 struct page *page = pte_page(pte);
1536
1537                 if (!PageAnon(page))
1538                         flags |= PM_FILE;
1539
1540                 if (page_mapcount(page) == 1)
1541                         flags |= PM_MMAP_EXCLUSIVE;
1542
1543                 flags |= PM_PRESENT;
1544                 if (pm->show_pfn)
1545                         frame = pte_pfn(pte) +
1546                                 ((addr & ~hmask) >> PAGE_SHIFT);
1547         }
1548
1549         for (; addr != end; addr += PAGE_SIZE) {
1550                 pagemap_entry_t pme = make_pme(frame, flags);
1551
1552                 err = add_to_pagemap(addr, &pme, pm);
1553                 if (err)
1554                         return err;
1555                 if (pm->show_pfn && (flags & PM_PRESENT))
1556                         frame++;
1557         }
1558
1559         cond_resched();
1560
1561         return err;
1562 }
1563 #else
1564 #define pagemap_hugetlb_range   NULL
1565 #endif /* HUGETLB_PAGE */
1566
1567 static const struct mm_walk_ops pagemap_ops = {
1568         .pmd_entry      = pagemap_pmd_range,
1569         .pte_hole       = pagemap_pte_hole,
1570         .hugetlb_entry  = pagemap_hugetlb_range,
1571 };
1572
1573 /*
1574  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1575  *
1576  * For each page in the address space, this file contains one 64-bit entry
1577  * consisting of the following:
1578  *
1579  * Bits 0-54  page frame number (PFN) if present
1580  * Bits 0-4   swap type if swapped
1581  * Bits 5-54  swap offset if swapped
1582  * Bit  55    pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst)
1583  * Bit  56    page exclusively mapped
1584  * Bit  57    pte is uffd-wp write-protected
1585  * Bits 58-60 zero
1586  * Bit  61    page is file-page or shared-anon
1587  * Bit  62    page swapped
1588  * Bit  63    page present
1589  *
1590  * If the page is not present but in swap, then the PFN contains an
1591  * encoding of the swap file number and the page's offset into the
1592  * swap. Unmapped pages return a null PFN. This allows determining
1593  * precisely which pages are mapped (or in swap) and comparing mapped
1594  * pages between processes.
1595  *
1596  * Efficient users of this interface will use /proc/pid/maps to
1597  * determine which areas of memory are actually mapped and llseek to
1598  * skip over unmapped regions.
1599  */
1600 static ssize_t pagemap_read(struct file *file, char __user *buf,
1601                             size_t count, loff_t *ppos)
1602 {
1603         struct mm_struct *mm = file->private_data;
1604         struct pagemapread pm;
1605         unsigned long src;
1606         unsigned long svpfn;
1607         unsigned long start_vaddr;
1608         unsigned long end_vaddr;
1609         int ret = 0, copied = 0;
1610
1611         if (!mm || !mmget_not_zero(mm))
1612                 goto out;
1613
1614         ret = -EINVAL;
1615         /* file position must be aligned */
1616         if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1617                 goto out_mm;
1618
1619         ret = 0;
1620         if (!count)
1621                 goto out_mm;
1622
1623         /* do not disclose physical addresses: attack vector */
1624         pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1625
1626         pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1627         pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL);
1628         ret = -ENOMEM;
1629         if (!pm.buffer)
1630                 goto out_mm;
1631
1632         src = *ppos;
1633         svpfn = src / PM_ENTRY_BYTES;
1634         end_vaddr = mm->task_size;
1635
1636         /* watch out for wraparound */
1637         start_vaddr = end_vaddr;
1638         if (svpfn <= (ULONG_MAX >> PAGE_SHIFT))
1639                 start_vaddr = untagged_addr(svpfn << PAGE_SHIFT);
1640
1641         /* Ensure the address is inside the task */
1642         if (start_vaddr > mm->task_size)
1643                 start_vaddr = end_vaddr;
1644
1645         /*
1646          * The odds are that this will stop walking way
1647          * before end_vaddr, because the length of the
1648          * user buffer is tracked in "pm", and the walk
1649          * will stop when we hit the end of the buffer.
1650          */
1651         ret = 0;
1652         while (count && (start_vaddr < end_vaddr)) {
1653                 int len;
1654                 unsigned long end;
1655
1656                 pm.pos = 0;
1657                 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1658                 /* overflow ? */
1659                 if (end < start_vaddr || end > end_vaddr)
1660                         end = end_vaddr;
1661                 ret = mmap_read_lock_killable(mm);
1662                 if (ret)
1663                         goto out_free;
1664                 ret = walk_page_range(mm, start_vaddr, end, &pagemap_ops, &pm);
1665                 mmap_read_unlock(mm);
1666                 start_vaddr = end;
1667
1668                 len = min(count, PM_ENTRY_BYTES * pm.pos);
1669                 if (copy_to_user(buf, pm.buffer, len)) {
1670                         ret = -EFAULT;
1671                         goto out_free;
1672                 }
1673                 copied += len;
1674                 buf += len;
1675                 count -= len;
1676         }
1677         *ppos += copied;
1678         if (!ret || ret == PM_END_OF_BUFFER)
1679                 ret = copied;
1680
1681 out_free:
1682         kfree(pm.buffer);
1683 out_mm:
1684         mmput(mm);
1685 out:
1686         return ret;
1687 }
1688
1689 static int pagemap_open(struct inode *inode, struct file *file)
1690 {
1691         struct mm_struct *mm;
1692
1693         mm = proc_mem_open(inode, PTRACE_MODE_READ);
1694         if (IS_ERR(mm))
1695                 return PTR_ERR(mm);
1696         file->private_data = mm;
1697         return 0;
1698 }
1699
1700 static int pagemap_release(struct inode *inode, struct file *file)
1701 {
1702         struct mm_struct *mm = file->private_data;
1703
1704         if (mm)
1705                 mmdrop(mm);
1706         return 0;
1707 }
1708
1709 const struct file_operations proc_pagemap_operations = {
1710         .llseek         = mem_lseek, /* borrow this */
1711         .read           = pagemap_read,
1712         .open           = pagemap_open,
1713         .release        = pagemap_release,
1714 };
1715 #endif /* CONFIG_PROC_PAGE_MONITOR */
1716
1717 #ifdef CONFIG_NUMA
1718
1719 struct numa_maps {
1720         unsigned long pages;
1721         unsigned long anon;
1722         unsigned long active;
1723         unsigned long writeback;
1724         unsigned long mapcount_max;
1725         unsigned long dirty;
1726         unsigned long swapcache;
1727         unsigned long node[MAX_NUMNODES];
1728 };
1729
1730 struct numa_maps_private {
1731         struct proc_maps_private proc_maps;
1732         struct numa_maps md;
1733 };
1734
1735 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1736                         unsigned long nr_pages)
1737 {
1738         int count = page_mapcount(page);
1739
1740         md->pages += nr_pages;
1741         if (pte_dirty || PageDirty(page))
1742                 md->dirty += nr_pages;
1743
1744         if (PageSwapCache(page))
1745                 md->swapcache += nr_pages;
1746
1747         if (PageActive(page) || PageUnevictable(page))
1748                 md->active += nr_pages;
1749
1750         if (PageWriteback(page))
1751                 md->writeback += nr_pages;
1752
1753         if (PageAnon(page))
1754                 md->anon += nr_pages;
1755
1756         if (count > md->mapcount_max)
1757                 md->mapcount_max = count;
1758
1759         md->node[page_to_nid(page)] += nr_pages;
1760 }
1761
1762 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1763                 unsigned long addr)
1764 {
1765         struct page *page;
1766         int nid;
1767
1768         if (!pte_present(pte))
1769                 return NULL;
1770
1771         page = vm_normal_page(vma, addr, pte);
1772         if (!page)
1773                 return NULL;
1774
1775         if (PageReserved(page))
1776                 return NULL;
1777
1778         nid = page_to_nid(page);
1779         if (!node_isset(nid, node_states[N_MEMORY]))
1780                 return NULL;
1781
1782         return page;
1783 }
1784
1785 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1786 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1787                                               struct vm_area_struct *vma,
1788                                               unsigned long addr)
1789 {
1790         struct page *page;
1791         int nid;
1792
1793         if (!pmd_present(pmd))
1794                 return NULL;
1795
1796         page = vm_normal_page_pmd(vma, addr, pmd);
1797         if (!page)
1798                 return NULL;
1799
1800         if (PageReserved(page))
1801                 return NULL;
1802
1803         nid = page_to_nid(page);
1804         if (!node_isset(nid, node_states[N_MEMORY]))
1805                 return NULL;
1806
1807         return page;
1808 }
1809 #endif
1810
1811 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1812                 unsigned long end, struct mm_walk *walk)
1813 {
1814         struct numa_maps *md = walk->private;
1815         struct vm_area_struct *vma = walk->vma;
1816         spinlock_t *ptl;
1817         pte_t *orig_pte;
1818         pte_t *pte;
1819
1820 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1821         ptl = pmd_trans_huge_lock(pmd, vma);
1822         if (ptl) {
1823                 struct page *page;
1824
1825                 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1826                 if (page)
1827                         gather_stats(page, md, pmd_dirty(*pmd),
1828                                      HPAGE_PMD_SIZE/PAGE_SIZE);
1829                 spin_unlock(ptl);
1830                 return 0;
1831         }
1832
1833         if (pmd_trans_unstable(pmd))
1834                 return 0;
1835 #endif
1836         orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1837         do {
1838                 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1839                 if (!page)
1840                         continue;
1841                 gather_stats(page, md, pte_dirty(*pte), 1);
1842
1843         } while (pte++, addr += PAGE_SIZE, addr != end);
1844         pte_unmap_unlock(orig_pte, ptl);
1845         cond_resched();
1846         return 0;
1847 }
1848 #ifdef CONFIG_HUGETLB_PAGE
1849 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1850                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1851 {
1852         pte_t huge_pte = huge_ptep_get(pte);
1853         struct numa_maps *md;
1854         struct page *page;
1855
1856         if (!pte_present(huge_pte))
1857                 return 0;
1858
1859         page = pte_page(huge_pte);
1860         if (!page)
1861                 return 0;
1862
1863         md = walk->private;
1864         gather_stats(page, md, pte_dirty(huge_pte), 1);
1865         return 0;
1866 }
1867
1868 #else
1869 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1870                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1871 {
1872         return 0;
1873 }
1874 #endif
1875
1876 static const struct mm_walk_ops show_numa_ops = {
1877         .hugetlb_entry = gather_hugetlb_stats,
1878         .pmd_entry = gather_pte_stats,
1879 };
1880
1881 /*
1882  * Display pages allocated per node and memory policy via /proc.
1883  */
1884 static int show_numa_map(struct seq_file *m, void *v)
1885 {
1886         struct numa_maps_private *numa_priv = m->private;
1887         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1888         struct vm_area_struct *vma = v;
1889         struct numa_maps *md = &numa_priv->md;
1890         struct file *file = vma->vm_file;
1891         struct mm_struct *mm = vma->vm_mm;
1892         struct mempolicy *pol;
1893         char buffer[64];
1894         int nid;
1895
1896         if (!mm)
1897                 return 0;
1898
1899         /* Ensure we start with an empty set of numa_maps statistics. */
1900         memset(md, 0, sizeof(*md));
1901
1902         pol = __get_vma_policy(vma, vma->vm_start);
1903         if (pol) {
1904                 mpol_to_str(buffer, sizeof(buffer), pol);
1905                 mpol_cond_put(pol);
1906         } else {
1907                 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1908         }
1909
1910         seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1911
1912         if (file) {
1913                 seq_puts(m, " file=");
1914                 seq_file_path(m, file, "\n\t= ");
1915         } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1916                 seq_puts(m, " heap");
1917         } else if (is_stack(vma)) {
1918                 seq_puts(m, " stack");
1919         }
1920
1921         if (is_vm_hugetlb_page(vma))
1922                 seq_puts(m, " huge");
1923
1924         /* mmap_lock is held by m_start */
1925         walk_page_vma(vma, &show_numa_ops, md);
1926
1927         if (!md->pages)
1928                 goto out;
1929
1930         if (md->anon)
1931                 seq_printf(m, " anon=%lu", md->anon);
1932
1933         if (md->dirty)
1934                 seq_printf(m, " dirty=%lu", md->dirty);
1935
1936         if (md->pages != md->anon && md->pages != md->dirty)
1937                 seq_printf(m, " mapped=%lu", md->pages);
1938
1939         if (md->mapcount_max > 1)
1940                 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1941
1942         if (md->swapcache)
1943                 seq_printf(m, " swapcache=%lu", md->swapcache);
1944
1945         if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1946                 seq_printf(m, " active=%lu", md->active);
1947
1948         if (md->writeback)
1949                 seq_printf(m, " writeback=%lu", md->writeback);
1950
1951         for_each_node_state(nid, N_MEMORY)
1952                 if (md->node[nid])
1953                         seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1954
1955         seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1956 out:
1957         seq_putc(m, '\n');
1958         return 0;
1959 }
1960
1961 static const struct seq_operations proc_pid_numa_maps_op = {
1962         .start  = m_start,
1963         .next   = m_next,
1964         .stop   = m_stop,
1965         .show   = show_numa_map,
1966 };
1967
1968 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1969 {
1970         return proc_maps_open(inode, file, &proc_pid_numa_maps_op,
1971                                 sizeof(struct numa_maps_private));
1972 }
1973
1974 const struct file_operations proc_pid_numa_maps_operations = {
1975         .open           = pid_numa_maps_open,
1976         .read           = seq_read,
1977         .llseek         = seq_lseek,
1978         .release        = proc_map_release,
1979 };
1980
1981 #endif /* CONFIG_NUMA */