mm, /proc/pid/pagemap: fix soft dirty marking for PMD migration entry
[sfrench/cifs-2.6.git] / fs / proc / task_mmu.c
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/mm.h>
3 #include <linux/vmacache.h>
4 #include <linux/hugetlb.h>
5 #include <linux/huge_mm.h>
6 #include <linux/mount.h>
7 #include <linux/seq_file.h>
8 #include <linux/highmem.h>
9 #include <linux/ptrace.h>
10 #include <linux/slab.h>
11 #include <linux/pagemap.h>
12 #include <linux/mempolicy.h>
13 #include <linux/rmap.h>
14 #include <linux/swap.h>
15 #include <linux/sched/mm.h>
16 #include <linux/swapops.h>
17 #include <linux/mmu_notifier.h>
18 #include <linux/page_idle.h>
19 #include <linux/shmem_fs.h>
20 #include <linux/uaccess.h>
21
22 #include <asm/elf.h>
23 #include <asm/tlb.h>
24 #include <asm/tlbflush.h>
25 #include "internal.h"
26
27 void task_mem(struct seq_file *m, struct mm_struct *mm)
28 {
29         unsigned long text, lib, swap, ptes, pmds, anon, file, shmem;
30         unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
31
32         anon = get_mm_counter(mm, MM_ANONPAGES);
33         file = get_mm_counter(mm, MM_FILEPAGES);
34         shmem = get_mm_counter(mm, MM_SHMEMPAGES);
35
36         /*
37          * Note: to minimize their overhead, mm maintains hiwater_vm and
38          * hiwater_rss only when about to *lower* total_vm or rss.  Any
39          * collector of these hiwater stats must therefore get total_vm
40          * and rss too, which will usually be the higher.  Barriers? not
41          * worth the effort, such snapshots can always be inconsistent.
42          */
43         hiwater_vm = total_vm = mm->total_vm;
44         if (hiwater_vm < mm->hiwater_vm)
45                 hiwater_vm = mm->hiwater_vm;
46         hiwater_rss = total_rss = anon + file + shmem;
47         if (hiwater_rss < mm->hiwater_rss)
48                 hiwater_rss = mm->hiwater_rss;
49
50         text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
51         lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
52         swap = get_mm_counter(mm, MM_SWAPENTS);
53         ptes = PTRS_PER_PTE * sizeof(pte_t) * atomic_long_read(&mm->nr_ptes);
54         pmds = PTRS_PER_PMD * sizeof(pmd_t) * mm_nr_pmds(mm);
55         seq_printf(m,
56                 "VmPeak:\t%8lu kB\n"
57                 "VmSize:\t%8lu kB\n"
58                 "VmLck:\t%8lu kB\n"
59                 "VmPin:\t%8lu kB\n"
60                 "VmHWM:\t%8lu kB\n"
61                 "VmRSS:\t%8lu kB\n"
62                 "RssAnon:\t%8lu kB\n"
63                 "RssFile:\t%8lu kB\n"
64                 "RssShmem:\t%8lu kB\n"
65                 "VmData:\t%8lu kB\n"
66                 "VmStk:\t%8lu kB\n"
67                 "VmExe:\t%8lu kB\n"
68                 "VmLib:\t%8lu kB\n"
69                 "VmPTE:\t%8lu kB\n"
70                 "VmPMD:\t%8lu kB\n"
71                 "VmSwap:\t%8lu kB\n",
72                 hiwater_vm << (PAGE_SHIFT-10),
73                 total_vm << (PAGE_SHIFT-10),
74                 mm->locked_vm << (PAGE_SHIFT-10),
75                 mm->pinned_vm << (PAGE_SHIFT-10),
76                 hiwater_rss << (PAGE_SHIFT-10),
77                 total_rss << (PAGE_SHIFT-10),
78                 anon << (PAGE_SHIFT-10),
79                 file << (PAGE_SHIFT-10),
80                 shmem << (PAGE_SHIFT-10),
81                 mm->data_vm << (PAGE_SHIFT-10),
82                 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
83                 ptes >> 10,
84                 pmds >> 10,
85                 swap << (PAGE_SHIFT-10));
86         hugetlb_report_usage(m, mm);
87 }
88
89 unsigned long task_vsize(struct mm_struct *mm)
90 {
91         return PAGE_SIZE * mm->total_vm;
92 }
93
94 unsigned long task_statm(struct mm_struct *mm,
95                          unsigned long *shared, unsigned long *text,
96                          unsigned long *data, unsigned long *resident)
97 {
98         *shared = get_mm_counter(mm, MM_FILEPAGES) +
99                         get_mm_counter(mm, MM_SHMEMPAGES);
100         *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
101                                                                 >> PAGE_SHIFT;
102         *data = mm->data_vm + mm->stack_vm;
103         *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
104         return mm->total_vm;
105 }
106
107 #ifdef CONFIG_NUMA
108 /*
109  * Save get_task_policy() for show_numa_map().
110  */
111 static void hold_task_mempolicy(struct proc_maps_private *priv)
112 {
113         struct task_struct *task = priv->task;
114
115         task_lock(task);
116         priv->task_mempolicy = get_task_policy(task);
117         mpol_get(priv->task_mempolicy);
118         task_unlock(task);
119 }
120 static void release_task_mempolicy(struct proc_maps_private *priv)
121 {
122         mpol_put(priv->task_mempolicy);
123 }
124 #else
125 static void hold_task_mempolicy(struct proc_maps_private *priv)
126 {
127 }
128 static void release_task_mempolicy(struct proc_maps_private *priv)
129 {
130 }
131 #endif
132
133 static void vma_stop(struct proc_maps_private *priv)
134 {
135         struct mm_struct *mm = priv->mm;
136
137         release_task_mempolicy(priv);
138         up_read(&mm->mmap_sem);
139         mmput(mm);
140 }
141
142 static struct vm_area_struct *
143 m_next_vma(struct proc_maps_private *priv, struct vm_area_struct *vma)
144 {
145         if (vma == priv->tail_vma)
146                 return NULL;
147         return vma->vm_next ?: priv->tail_vma;
148 }
149
150 static void m_cache_vma(struct seq_file *m, struct vm_area_struct *vma)
151 {
152         if (m->count < m->size) /* vma is copied successfully */
153                 m->version = m_next_vma(m->private, vma) ? vma->vm_end : -1UL;
154 }
155
156 static void *m_start(struct seq_file *m, loff_t *ppos)
157 {
158         struct proc_maps_private *priv = m->private;
159         unsigned long last_addr = m->version;
160         struct mm_struct *mm;
161         struct vm_area_struct *vma;
162         unsigned int pos = *ppos;
163
164         /* See m_cache_vma(). Zero at the start or after lseek. */
165         if (last_addr == -1UL)
166                 return NULL;
167
168         priv->task = get_proc_task(priv->inode);
169         if (!priv->task)
170                 return ERR_PTR(-ESRCH);
171
172         mm = priv->mm;
173         if (!mm || !mmget_not_zero(mm))
174                 return NULL;
175
176         down_read(&mm->mmap_sem);
177         hold_task_mempolicy(priv);
178         priv->tail_vma = get_gate_vma(mm);
179
180         if (last_addr) {
181                 vma = find_vma(mm, last_addr - 1);
182                 if (vma && vma->vm_start <= last_addr)
183                         vma = m_next_vma(priv, vma);
184                 if (vma)
185                         return vma;
186         }
187
188         m->version = 0;
189         if (pos < mm->map_count) {
190                 for (vma = mm->mmap; pos; pos--) {
191                         m->version = vma->vm_start;
192                         vma = vma->vm_next;
193                 }
194                 return vma;
195         }
196
197         /* we do not bother to update m->version in this case */
198         if (pos == mm->map_count && priv->tail_vma)
199                 return priv->tail_vma;
200
201         vma_stop(priv);
202         return NULL;
203 }
204
205 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
206 {
207         struct proc_maps_private *priv = m->private;
208         struct vm_area_struct *next;
209
210         (*pos)++;
211         next = m_next_vma(priv, v);
212         if (!next)
213                 vma_stop(priv);
214         return next;
215 }
216
217 static void m_stop(struct seq_file *m, void *v)
218 {
219         struct proc_maps_private *priv = m->private;
220
221         if (!IS_ERR_OR_NULL(v))
222                 vma_stop(priv);
223         if (priv->task) {
224                 put_task_struct(priv->task);
225                 priv->task = NULL;
226         }
227 }
228
229 static int proc_maps_open(struct inode *inode, struct file *file,
230                         const struct seq_operations *ops, int psize)
231 {
232         struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
233
234         if (!priv)
235                 return -ENOMEM;
236
237         priv->inode = inode;
238         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
239         if (IS_ERR(priv->mm)) {
240                 int err = PTR_ERR(priv->mm);
241
242                 seq_release_private(inode, file);
243                 return err;
244         }
245
246         return 0;
247 }
248
249 static int proc_map_release(struct inode *inode, struct file *file)
250 {
251         struct seq_file *seq = file->private_data;
252         struct proc_maps_private *priv = seq->private;
253
254         if (priv->mm)
255                 mmdrop(priv->mm);
256
257         kfree(priv->rollup);
258         return seq_release_private(inode, file);
259 }
260
261 static int do_maps_open(struct inode *inode, struct file *file,
262                         const struct seq_operations *ops)
263 {
264         return proc_maps_open(inode, file, ops,
265                                 sizeof(struct proc_maps_private));
266 }
267
268 /*
269  * Indicate if the VMA is a stack for the given task; for
270  * /proc/PID/maps that is the stack of the main task.
271  */
272 static int is_stack(struct vm_area_struct *vma)
273 {
274         /*
275          * We make no effort to guess what a given thread considers to be
276          * its "stack".  It's not even well-defined for programs written
277          * languages like Go.
278          */
279         return vma->vm_start <= vma->vm_mm->start_stack &&
280                 vma->vm_end >= vma->vm_mm->start_stack;
281 }
282
283 static void show_vma_header_prefix(struct seq_file *m,
284                                    unsigned long start, unsigned long end,
285                                    vm_flags_t flags, unsigned long long pgoff,
286                                    dev_t dev, unsigned long ino)
287 {
288         seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
289         seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu ",
290                    start,
291                    end,
292                    flags & VM_READ ? 'r' : '-',
293                    flags & VM_WRITE ? 'w' : '-',
294                    flags & VM_EXEC ? 'x' : '-',
295                    flags & VM_MAYSHARE ? 's' : 'p',
296                    pgoff,
297                    MAJOR(dev), MINOR(dev), ino);
298 }
299
300 static void
301 show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
302 {
303         struct mm_struct *mm = vma->vm_mm;
304         struct file *file = vma->vm_file;
305         vm_flags_t flags = vma->vm_flags;
306         unsigned long ino = 0;
307         unsigned long long pgoff = 0;
308         unsigned long start, end;
309         dev_t dev = 0;
310         const char *name = NULL;
311
312         if (file) {
313                 struct inode *inode = file_inode(vma->vm_file);
314                 dev = inode->i_sb->s_dev;
315                 ino = inode->i_ino;
316                 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
317         }
318
319         start = vma->vm_start;
320         end = vma->vm_end;
321         show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino);
322
323         /*
324          * Print the dentry name for named mappings, and a
325          * special [heap] marker for the heap:
326          */
327         if (file) {
328                 seq_pad(m, ' ');
329                 seq_file_path(m, file, "\n");
330                 goto done;
331         }
332
333         if (vma->vm_ops && vma->vm_ops->name) {
334                 name = vma->vm_ops->name(vma);
335                 if (name)
336                         goto done;
337         }
338
339         name = arch_vma_name(vma);
340         if (!name) {
341                 if (!mm) {
342                         name = "[vdso]";
343                         goto done;
344                 }
345
346                 if (vma->vm_start <= mm->brk &&
347                     vma->vm_end >= mm->start_brk) {
348                         name = "[heap]";
349                         goto done;
350                 }
351
352                 if (is_stack(vma))
353                         name = "[stack]";
354         }
355
356 done:
357         if (name) {
358                 seq_pad(m, ' ');
359                 seq_puts(m, name);
360         }
361         seq_putc(m, '\n');
362 }
363
364 static int show_map(struct seq_file *m, void *v, int is_pid)
365 {
366         show_map_vma(m, v, is_pid);
367         m_cache_vma(m, v);
368         return 0;
369 }
370
371 static int show_pid_map(struct seq_file *m, void *v)
372 {
373         return show_map(m, v, 1);
374 }
375
376 static int show_tid_map(struct seq_file *m, void *v)
377 {
378         return show_map(m, v, 0);
379 }
380
381 static const struct seq_operations proc_pid_maps_op = {
382         .start  = m_start,
383         .next   = m_next,
384         .stop   = m_stop,
385         .show   = show_pid_map
386 };
387
388 static const struct seq_operations proc_tid_maps_op = {
389         .start  = m_start,
390         .next   = m_next,
391         .stop   = m_stop,
392         .show   = show_tid_map
393 };
394
395 static int pid_maps_open(struct inode *inode, struct file *file)
396 {
397         return do_maps_open(inode, file, &proc_pid_maps_op);
398 }
399
400 static int tid_maps_open(struct inode *inode, struct file *file)
401 {
402         return do_maps_open(inode, file, &proc_tid_maps_op);
403 }
404
405 const struct file_operations proc_pid_maps_operations = {
406         .open           = pid_maps_open,
407         .read           = seq_read,
408         .llseek         = seq_lseek,
409         .release        = proc_map_release,
410 };
411
412 const struct file_operations proc_tid_maps_operations = {
413         .open           = tid_maps_open,
414         .read           = seq_read,
415         .llseek         = seq_lseek,
416         .release        = proc_map_release,
417 };
418
419 /*
420  * Proportional Set Size(PSS): my share of RSS.
421  *
422  * PSS of a process is the count of pages it has in memory, where each
423  * page is divided by the number of processes sharing it.  So if a
424  * process has 1000 pages all to itself, and 1000 shared with one other
425  * process, its PSS will be 1500.
426  *
427  * To keep (accumulated) division errors low, we adopt a 64bit
428  * fixed-point pss counter to minimize division errors. So (pss >>
429  * PSS_SHIFT) would be the real byte count.
430  *
431  * A shift of 12 before division means (assuming 4K page size):
432  *      - 1M 3-user-pages add up to 8KB errors;
433  *      - supports mapcount up to 2^24, or 16M;
434  *      - supports PSS up to 2^52 bytes, or 4PB.
435  */
436 #define PSS_SHIFT 12
437
438 #ifdef CONFIG_PROC_PAGE_MONITOR
439 struct mem_size_stats {
440         bool first;
441         unsigned long resident;
442         unsigned long shared_clean;
443         unsigned long shared_dirty;
444         unsigned long private_clean;
445         unsigned long private_dirty;
446         unsigned long referenced;
447         unsigned long anonymous;
448         unsigned long lazyfree;
449         unsigned long anonymous_thp;
450         unsigned long shmem_thp;
451         unsigned long swap;
452         unsigned long shared_hugetlb;
453         unsigned long private_hugetlb;
454         unsigned long first_vma_start;
455         u64 pss;
456         u64 pss_locked;
457         u64 swap_pss;
458         bool check_shmem_swap;
459 };
460
461 static void smaps_account(struct mem_size_stats *mss, struct page *page,
462                 bool compound, bool young, bool dirty)
463 {
464         int i, nr = compound ? 1 << compound_order(page) : 1;
465         unsigned long size = nr * PAGE_SIZE;
466
467         if (PageAnon(page)) {
468                 mss->anonymous += size;
469                 if (!PageSwapBacked(page) && !dirty && !PageDirty(page))
470                         mss->lazyfree += size;
471         }
472
473         mss->resident += size;
474         /* Accumulate the size in pages that have been accessed. */
475         if (young || page_is_young(page) || PageReferenced(page))
476                 mss->referenced += size;
477
478         /*
479          * page_count(page) == 1 guarantees the page is mapped exactly once.
480          * If any subpage of the compound page mapped with PTE it would elevate
481          * page_count().
482          */
483         if (page_count(page) == 1) {
484                 if (dirty || PageDirty(page))
485                         mss->private_dirty += size;
486                 else
487                         mss->private_clean += size;
488                 mss->pss += (u64)size << PSS_SHIFT;
489                 return;
490         }
491
492         for (i = 0; i < nr; i++, page++) {
493                 int mapcount = page_mapcount(page);
494
495                 if (mapcount >= 2) {
496                         if (dirty || PageDirty(page))
497                                 mss->shared_dirty += PAGE_SIZE;
498                         else
499                                 mss->shared_clean += PAGE_SIZE;
500                         mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount;
501                 } else {
502                         if (dirty || PageDirty(page))
503                                 mss->private_dirty += PAGE_SIZE;
504                         else
505                                 mss->private_clean += PAGE_SIZE;
506                         mss->pss += PAGE_SIZE << PSS_SHIFT;
507                 }
508         }
509 }
510
511 #ifdef CONFIG_SHMEM
512 static int smaps_pte_hole(unsigned long addr, unsigned long end,
513                 struct mm_walk *walk)
514 {
515         struct mem_size_stats *mss = walk->private;
516
517         mss->swap += shmem_partial_swap_usage(
518                         walk->vma->vm_file->f_mapping, addr, end);
519
520         return 0;
521 }
522 #endif
523
524 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
525                 struct mm_walk *walk)
526 {
527         struct mem_size_stats *mss = walk->private;
528         struct vm_area_struct *vma = walk->vma;
529         struct page *page = NULL;
530
531         if (pte_present(*pte)) {
532                 page = vm_normal_page(vma, addr, *pte);
533         } else if (is_swap_pte(*pte)) {
534                 swp_entry_t swpent = pte_to_swp_entry(*pte);
535
536                 if (!non_swap_entry(swpent)) {
537                         int mapcount;
538
539                         mss->swap += PAGE_SIZE;
540                         mapcount = swp_swapcount(swpent);
541                         if (mapcount >= 2) {
542                                 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
543
544                                 do_div(pss_delta, mapcount);
545                                 mss->swap_pss += pss_delta;
546                         } else {
547                                 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
548                         }
549                 } else if (is_migration_entry(swpent))
550                         page = migration_entry_to_page(swpent);
551                 else if (is_device_private_entry(swpent))
552                         page = device_private_entry_to_page(swpent);
553         } else if (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap
554                                                         && pte_none(*pte))) {
555                 page = find_get_entry(vma->vm_file->f_mapping,
556                                                 linear_page_index(vma, addr));
557                 if (!page)
558                         return;
559
560                 if (radix_tree_exceptional_entry(page))
561                         mss->swap += PAGE_SIZE;
562                 else
563                         put_page(page);
564
565                 return;
566         }
567
568         if (!page)
569                 return;
570
571         smaps_account(mss, page, false, pte_young(*pte), pte_dirty(*pte));
572 }
573
574 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
575 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
576                 struct mm_walk *walk)
577 {
578         struct mem_size_stats *mss = walk->private;
579         struct vm_area_struct *vma = walk->vma;
580         struct page *page;
581
582         /* FOLL_DUMP will return -EFAULT on huge zero page */
583         page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
584         if (IS_ERR_OR_NULL(page))
585                 return;
586         if (PageAnon(page))
587                 mss->anonymous_thp += HPAGE_PMD_SIZE;
588         else if (PageSwapBacked(page))
589                 mss->shmem_thp += HPAGE_PMD_SIZE;
590         else if (is_zone_device_page(page))
591                 /* pass */;
592         else
593                 VM_BUG_ON_PAGE(1, page);
594         smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd));
595 }
596 #else
597 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
598                 struct mm_walk *walk)
599 {
600 }
601 #endif
602
603 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
604                            struct mm_walk *walk)
605 {
606         struct vm_area_struct *vma = walk->vma;
607         pte_t *pte;
608         spinlock_t *ptl;
609
610         ptl = pmd_trans_huge_lock(pmd, vma);
611         if (ptl) {
612                 if (pmd_present(*pmd))
613                         smaps_pmd_entry(pmd, addr, walk);
614                 spin_unlock(ptl);
615                 goto out;
616         }
617
618         if (pmd_trans_unstable(pmd))
619                 goto out;
620         /*
621          * The mmap_sem held all the way back in m_start() is what
622          * keeps khugepaged out of here and from collapsing things
623          * in here.
624          */
625         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
626         for (; addr != end; pte++, addr += PAGE_SIZE)
627                 smaps_pte_entry(pte, addr, walk);
628         pte_unmap_unlock(pte - 1, ptl);
629 out:
630         cond_resched();
631         return 0;
632 }
633
634 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
635 {
636         /*
637          * Don't forget to update Documentation/ on changes.
638          */
639         static const char mnemonics[BITS_PER_LONG][2] = {
640                 /*
641                  * In case if we meet a flag we don't know about.
642                  */
643                 [0 ... (BITS_PER_LONG-1)] = "??",
644
645                 [ilog2(VM_READ)]        = "rd",
646                 [ilog2(VM_WRITE)]       = "wr",
647                 [ilog2(VM_EXEC)]        = "ex",
648                 [ilog2(VM_SHARED)]      = "sh",
649                 [ilog2(VM_MAYREAD)]     = "mr",
650                 [ilog2(VM_MAYWRITE)]    = "mw",
651                 [ilog2(VM_MAYEXEC)]     = "me",
652                 [ilog2(VM_MAYSHARE)]    = "ms",
653                 [ilog2(VM_GROWSDOWN)]   = "gd",
654                 [ilog2(VM_PFNMAP)]      = "pf",
655                 [ilog2(VM_DENYWRITE)]   = "dw",
656 #ifdef CONFIG_X86_INTEL_MPX
657                 [ilog2(VM_MPX)]         = "mp",
658 #endif
659                 [ilog2(VM_LOCKED)]      = "lo",
660                 [ilog2(VM_IO)]          = "io",
661                 [ilog2(VM_SEQ_READ)]    = "sr",
662                 [ilog2(VM_RAND_READ)]   = "rr",
663                 [ilog2(VM_DONTCOPY)]    = "dc",
664                 [ilog2(VM_DONTEXPAND)]  = "de",
665                 [ilog2(VM_ACCOUNT)]     = "ac",
666                 [ilog2(VM_NORESERVE)]   = "nr",
667                 [ilog2(VM_HUGETLB)]     = "ht",
668                 [ilog2(VM_ARCH_1)]      = "ar",
669                 [ilog2(VM_WIPEONFORK)]  = "wf",
670                 [ilog2(VM_DONTDUMP)]    = "dd",
671 #ifdef CONFIG_MEM_SOFT_DIRTY
672                 [ilog2(VM_SOFTDIRTY)]   = "sd",
673 #endif
674                 [ilog2(VM_MIXEDMAP)]    = "mm",
675                 [ilog2(VM_HUGEPAGE)]    = "hg",
676                 [ilog2(VM_NOHUGEPAGE)]  = "nh",
677                 [ilog2(VM_MERGEABLE)]   = "mg",
678                 [ilog2(VM_UFFD_MISSING)]= "um",
679                 [ilog2(VM_UFFD_WP)]     = "uw",
680 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
681                 /* These come out via ProtectionKey: */
682                 [ilog2(VM_PKEY_BIT0)]   = "",
683                 [ilog2(VM_PKEY_BIT1)]   = "",
684                 [ilog2(VM_PKEY_BIT2)]   = "",
685                 [ilog2(VM_PKEY_BIT3)]   = "",
686 #endif
687         };
688         size_t i;
689
690         seq_puts(m, "VmFlags: ");
691         for (i = 0; i < BITS_PER_LONG; i++) {
692                 if (!mnemonics[i][0])
693                         continue;
694                 if (vma->vm_flags & (1UL << i)) {
695                         seq_printf(m, "%c%c ",
696                                    mnemonics[i][0], mnemonics[i][1]);
697                 }
698         }
699         seq_putc(m, '\n');
700 }
701
702 #ifdef CONFIG_HUGETLB_PAGE
703 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
704                                  unsigned long addr, unsigned long end,
705                                  struct mm_walk *walk)
706 {
707         struct mem_size_stats *mss = walk->private;
708         struct vm_area_struct *vma = walk->vma;
709         struct page *page = NULL;
710
711         if (pte_present(*pte)) {
712                 page = vm_normal_page(vma, addr, *pte);
713         } else if (is_swap_pte(*pte)) {
714                 swp_entry_t swpent = pte_to_swp_entry(*pte);
715
716                 if (is_migration_entry(swpent))
717                         page = migration_entry_to_page(swpent);
718                 else if (is_device_private_entry(swpent))
719                         page = device_private_entry_to_page(swpent);
720         }
721         if (page) {
722                 int mapcount = page_mapcount(page);
723
724                 if (mapcount >= 2)
725                         mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
726                 else
727                         mss->private_hugetlb += huge_page_size(hstate_vma(vma));
728         }
729         return 0;
730 }
731 #endif /* HUGETLB_PAGE */
732
733 void __weak arch_show_smap(struct seq_file *m, struct vm_area_struct *vma)
734 {
735 }
736
737 static int show_smap(struct seq_file *m, void *v, int is_pid)
738 {
739         struct proc_maps_private *priv = m->private;
740         struct vm_area_struct *vma = v;
741         struct mem_size_stats mss_stack;
742         struct mem_size_stats *mss;
743         struct mm_walk smaps_walk = {
744                 .pmd_entry = smaps_pte_range,
745 #ifdef CONFIG_HUGETLB_PAGE
746                 .hugetlb_entry = smaps_hugetlb_range,
747 #endif
748                 .mm = vma->vm_mm,
749         };
750         int ret = 0;
751         bool rollup_mode;
752         bool last_vma;
753
754         if (priv->rollup) {
755                 rollup_mode = true;
756                 mss = priv->rollup;
757                 if (mss->first) {
758                         mss->first_vma_start = vma->vm_start;
759                         mss->first = false;
760                 }
761                 last_vma = !m_next_vma(priv, vma);
762         } else {
763                 rollup_mode = false;
764                 memset(&mss_stack, 0, sizeof(mss_stack));
765                 mss = &mss_stack;
766         }
767
768         smaps_walk.private = mss;
769
770 #ifdef CONFIG_SHMEM
771         if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
772                 /*
773                  * For shared or readonly shmem mappings we know that all
774                  * swapped out pages belong to the shmem object, and we can
775                  * obtain the swap value much more efficiently. For private
776                  * writable mappings, we might have COW pages that are
777                  * not affected by the parent swapped out pages of the shmem
778                  * object, so we have to distinguish them during the page walk.
779                  * Unless we know that the shmem object (or the part mapped by
780                  * our VMA) has no swapped out pages at all.
781                  */
782                 unsigned long shmem_swapped = shmem_swap_usage(vma);
783
784                 if (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
785                                         !(vma->vm_flags & VM_WRITE)) {
786                         mss->swap = shmem_swapped;
787                 } else {
788                         mss->check_shmem_swap = true;
789                         smaps_walk.pte_hole = smaps_pte_hole;
790                 }
791         }
792 #endif
793
794         /* mmap_sem is held in m_start */
795         walk_page_vma(vma, &smaps_walk);
796         if (vma->vm_flags & VM_LOCKED)
797                 mss->pss_locked += mss->pss;
798
799         if (!rollup_mode) {
800                 show_map_vma(m, vma, is_pid);
801         } else if (last_vma) {
802                 show_vma_header_prefix(
803                         m, mss->first_vma_start, vma->vm_end, 0, 0, 0, 0);
804                 seq_pad(m, ' ');
805                 seq_puts(m, "[rollup]\n");
806         } else {
807                 ret = SEQ_SKIP;
808         }
809
810         if (!rollup_mode)
811                 seq_printf(m,
812                            "Size:           %8lu kB\n"
813                            "KernelPageSize: %8lu kB\n"
814                            "MMUPageSize:    %8lu kB\n",
815                            (vma->vm_end - vma->vm_start) >> 10,
816                            vma_kernel_pagesize(vma) >> 10,
817                            vma_mmu_pagesize(vma) >> 10);
818
819
820         if (!rollup_mode || last_vma)
821                 seq_printf(m,
822                            "Rss:            %8lu kB\n"
823                            "Pss:            %8lu kB\n"
824                            "Shared_Clean:   %8lu kB\n"
825                            "Shared_Dirty:   %8lu kB\n"
826                            "Private_Clean:  %8lu kB\n"
827                            "Private_Dirty:  %8lu kB\n"
828                            "Referenced:     %8lu kB\n"
829                            "Anonymous:      %8lu kB\n"
830                            "LazyFree:       %8lu kB\n"
831                            "AnonHugePages:  %8lu kB\n"
832                            "ShmemPmdMapped: %8lu kB\n"
833                            "Shared_Hugetlb: %8lu kB\n"
834                            "Private_Hugetlb: %7lu kB\n"
835                            "Swap:           %8lu kB\n"
836                            "SwapPss:        %8lu kB\n"
837                            "Locked:         %8lu kB\n",
838                            mss->resident >> 10,
839                            (unsigned long)(mss->pss >> (10 + PSS_SHIFT)),
840                            mss->shared_clean  >> 10,
841                            mss->shared_dirty  >> 10,
842                            mss->private_clean >> 10,
843                            mss->private_dirty >> 10,
844                            mss->referenced >> 10,
845                            mss->anonymous >> 10,
846                            mss->lazyfree >> 10,
847                            mss->anonymous_thp >> 10,
848                            mss->shmem_thp >> 10,
849                            mss->shared_hugetlb >> 10,
850                            mss->private_hugetlb >> 10,
851                            mss->swap >> 10,
852                            (unsigned long)(mss->swap_pss >> (10 + PSS_SHIFT)),
853                            (unsigned long)(mss->pss >> (10 + PSS_SHIFT)));
854
855         if (!rollup_mode) {
856                 arch_show_smap(m, vma);
857                 show_smap_vma_flags(m, vma);
858         }
859         m_cache_vma(m, vma);
860         return ret;
861 }
862
863 static int show_pid_smap(struct seq_file *m, void *v)
864 {
865         return show_smap(m, v, 1);
866 }
867
868 static int show_tid_smap(struct seq_file *m, void *v)
869 {
870         return show_smap(m, v, 0);
871 }
872
873 static const struct seq_operations proc_pid_smaps_op = {
874         .start  = m_start,
875         .next   = m_next,
876         .stop   = m_stop,
877         .show   = show_pid_smap
878 };
879
880 static const struct seq_operations proc_tid_smaps_op = {
881         .start  = m_start,
882         .next   = m_next,
883         .stop   = m_stop,
884         .show   = show_tid_smap
885 };
886
887 static int pid_smaps_open(struct inode *inode, struct file *file)
888 {
889         return do_maps_open(inode, file, &proc_pid_smaps_op);
890 }
891
892 static int pid_smaps_rollup_open(struct inode *inode, struct file *file)
893 {
894         struct seq_file *seq;
895         struct proc_maps_private *priv;
896         int ret = do_maps_open(inode, file, &proc_pid_smaps_op);
897
898         if (ret < 0)
899                 return ret;
900         seq = file->private_data;
901         priv = seq->private;
902         priv->rollup = kzalloc(sizeof(*priv->rollup), GFP_KERNEL);
903         if (!priv->rollup) {
904                 proc_map_release(inode, file);
905                 return -ENOMEM;
906         }
907         priv->rollup->first = true;
908         return 0;
909 }
910
911 static int tid_smaps_open(struct inode *inode, struct file *file)
912 {
913         return do_maps_open(inode, file, &proc_tid_smaps_op);
914 }
915
916 const struct file_operations proc_pid_smaps_operations = {
917         .open           = pid_smaps_open,
918         .read           = seq_read,
919         .llseek         = seq_lseek,
920         .release        = proc_map_release,
921 };
922
923 const struct file_operations proc_pid_smaps_rollup_operations = {
924         .open           = pid_smaps_rollup_open,
925         .read           = seq_read,
926         .llseek         = seq_lseek,
927         .release        = proc_map_release,
928 };
929
930 const struct file_operations proc_tid_smaps_operations = {
931         .open           = tid_smaps_open,
932         .read           = seq_read,
933         .llseek         = seq_lseek,
934         .release        = proc_map_release,
935 };
936
937 enum clear_refs_types {
938         CLEAR_REFS_ALL = 1,
939         CLEAR_REFS_ANON,
940         CLEAR_REFS_MAPPED,
941         CLEAR_REFS_SOFT_DIRTY,
942         CLEAR_REFS_MM_HIWATER_RSS,
943         CLEAR_REFS_LAST,
944 };
945
946 struct clear_refs_private {
947         enum clear_refs_types type;
948 };
949
950 #ifdef CONFIG_MEM_SOFT_DIRTY
951 static inline void clear_soft_dirty(struct vm_area_struct *vma,
952                 unsigned long addr, pte_t *pte)
953 {
954         /*
955          * The soft-dirty tracker uses #PF-s to catch writes
956          * to pages, so write-protect the pte as well. See the
957          * Documentation/vm/soft-dirty.txt for full description
958          * of how soft-dirty works.
959          */
960         pte_t ptent = *pte;
961
962         if (pte_present(ptent)) {
963                 ptent = ptep_modify_prot_start(vma->vm_mm, addr, pte);
964                 ptent = pte_wrprotect(ptent);
965                 ptent = pte_clear_soft_dirty(ptent);
966                 ptep_modify_prot_commit(vma->vm_mm, addr, pte, ptent);
967         } else if (is_swap_pte(ptent)) {
968                 ptent = pte_swp_clear_soft_dirty(ptent);
969                 set_pte_at(vma->vm_mm, addr, pte, ptent);
970         }
971 }
972 #else
973 static inline void clear_soft_dirty(struct vm_area_struct *vma,
974                 unsigned long addr, pte_t *pte)
975 {
976 }
977 #endif
978
979 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
980 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
981                 unsigned long addr, pmd_t *pmdp)
982 {
983         pmd_t pmd = *pmdp;
984
985         if (pmd_present(pmd)) {
986                 /* See comment in change_huge_pmd() */
987                 pmdp_invalidate(vma, addr, pmdp);
988                 if (pmd_dirty(*pmdp))
989                         pmd = pmd_mkdirty(pmd);
990                 if (pmd_young(*pmdp))
991                         pmd = pmd_mkyoung(pmd);
992
993                 pmd = pmd_wrprotect(pmd);
994                 pmd = pmd_clear_soft_dirty(pmd);
995
996                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
997         } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
998                 pmd = pmd_swp_clear_soft_dirty(pmd);
999                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1000         }
1001 }
1002 #else
1003 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1004                 unsigned long addr, pmd_t *pmdp)
1005 {
1006 }
1007 #endif
1008
1009 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
1010                                 unsigned long end, struct mm_walk *walk)
1011 {
1012         struct clear_refs_private *cp = walk->private;
1013         struct vm_area_struct *vma = walk->vma;
1014         pte_t *pte, ptent;
1015         spinlock_t *ptl;
1016         struct page *page;
1017
1018         ptl = pmd_trans_huge_lock(pmd, vma);
1019         if (ptl) {
1020                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1021                         clear_soft_dirty_pmd(vma, addr, pmd);
1022                         goto out;
1023                 }
1024
1025                 if (!pmd_present(*pmd))
1026                         goto out;
1027
1028                 page = pmd_page(*pmd);
1029
1030                 /* Clear accessed and referenced bits. */
1031                 pmdp_test_and_clear_young(vma, addr, pmd);
1032                 test_and_clear_page_young(page);
1033                 ClearPageReferenced(page);
1034 out:
1035                 spin_unlock(ptl);
1036                 return 0;
1037         }
1038
1039         if (pmd_trans_unstable(pmd))
1040                 return 0;
1041
1042         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1043         for (; addr != end; pte++, addr += PAGE_SIZE) {
1044                 ptent = *pte;
1045
1046                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1047                         clear_soft_dirty(vma, addr, pte);
1048                         continue;
1049                 }
1050
1051                 if (!pte_present(ptent))
1052                         continue;
1053
1054                 page = vm_normal_page(vma, addr, ptent);
1055                 if (!page)
1056                         continue;
1057
1058                 /* Clear accessed and referenced bits. */
1059                 ptep_test_and_clear_young(vma, addr, pte);
1060                 test_and_clear_page_young(page);
1061                 ClearPageReferenced(page);
1062         }
1063         pte_unmap_unlock(pte - 1, ptl);
1064         cond_resched();
1065         return 0;
1066 }
1067
1068 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1069                                 struct mm_walk *walk)
1070 {
1071         struct clear_refs_private *cp = walk->private;
1072         struct vm_area_struct *vma = walk->vma;
1073
1074         if (vma->vm_flags & VM_PFNMAP)
1075                 return 1;
1076
1077         /*
1078          * Writing 1 to /proc/pid/clear_refs affects all pages.
1079          * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1080          * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1081          * Writing 4 to /proc/pid/clear_refs affects all pages.
1082          */
1083         if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1084                 return 1;
1085         if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1086                 return 1;
1087         return 0;
1088 }
1089
1090 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1091                                 size_t count, loff_t *ppos)
1092 {
1093         struct task_struct *task;
1094         char buffer[PROC_NUMBUF];
1095         struct mm_struct *mm;
1096         struct vm_area_struct *vma;
1097         enum clear_refs_types type;
1098         struct mmu_gather tlb;
1099         int itype;
1100         int rv;
1101
1102         memset(buffer, 0, sizeof(buffer));
1103         if (count > sizeof(buffer) - 1)
1104                 count = sizeof(buffer) - 1;
1105         if (copy_from_user(buffer, buf, count))
1106                 return -EFAULT;
1107         rv = kstrtoint(strstrip(buffer), 10, &itype);
1108         if (rv < 0)
1109                 return rv;
1110         type = (enum clear_refs_types)itype;
1111         if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1112                 return -EINVAL;
1113
1114         task = get_proc_task(file_inode(file));
1115         if (!task)
1116                 return -ESRCH;
1117         mm = get_task_mm(task);
1118         if (mm) {
1119                 struct clear_refs_private cp = {
1120                         .type = type,
1121                 };
1122                 struct mm_walk clear_refs_walk = {
1123                         .pmd_entry = clear_refs_pte_range,
1124                         .test_walk = clear_refs_test_walk,
1125                         .mm = mm,
1126                         .private = &cp,
1127                 };
1128
1129                 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1130                         if (down_write_killable(&mm->mmap_sem)) {
1131                                 count = -EINTR;
1132                                 goto out_mm;
1133                         }
1134
1135                         /*
1136                          * Writing 5 to /proc/pid/clear_refs resets the peak
1137                          * resident set size to this mm's current rss value.
1138                          */
1139                         reset_mm_hiwater_rss(mm);
1140                         up_write(&mm->mmap_sem);
1141                         goto out_mm;
1142                 }
1143
1144                 down_read(&mm->mmap_sem);
1145                 tlb_gather_mmu(&tlb, mm, 0, -1);
1146                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1147                         for (vma = mm->mmap; vma; vma = vma->vm_next) {
1148                                 if (!(vma->vm_flags & VM_SOFTDIRTY))
1149                                         continue;
1150                                 up_read(&mm->mmap_sem);
1151                                 if (down_write_killable(&mm->mmap_sem)) {
1152                                         count = -EINTR;
1153                                         goto out_mm;
1154                                 }
1155                                 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1156                                         vma->vm_flags &= ~VM_SOFTDIRTY;
1157                                         vma_set_page_prot(vma);
1158                                 }
1159                                 downgrade_write(&mm->mmap_sem);
1160                                 break;
1161                         }
1162                         mmu_notifier_invalidate_range_start(mm, 0, -1);
1163                 }
1164                 walk_page_range(0, mm->highest_vm_end, &clear_refs_walk);
1165                 if (type == CLEAR_REFS_SOFT_DIRTY)
1166                         mmu_notifier_invalidate_range_end(mm, 0, -1);
1167                 tlb_finish_mmu(&tlb, 0, -1);
1168                 up_read(&mm->mmap_sem);
1169 out_mm:
1170                 mmput(mm);
1171         }
1172         put_task_struct(task);
1173
1174         return count;
1175 }
1176
1177 const struct file_operations proc_clear_refs_operations = {
1178         .write          = clear_refs_write,
1179         .llseek         = noop_llseek,
1180 };
1181
1182 typedef struct {
1183         u64 pme;
1184 } pagemap_entry_t;
1185
1186 struct pagemapread {
1187         int pos, len;           /* units: PM_ENTRY_BYTES, not bytes */
1188         pagemap_entry_t *buffer;
1189         bool show_pfn;
1190 };
1191
1192 #define PAGEMAP_WALK_SIZE       (PMD_SIZE)
1193 #define PAGEMAP_WALK_MASK       (PMD_MASK)
1194
1195 #define PM_ENTRY_BYTES          sizeof(pagemap_entry_t)
1196 #define PM_PFRAME_BITS          55
1197 #define PM_PFRAME_MASK          GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1198 #define PM_SOFT_DIRTY           BIT_ULL(55)
1199 #define PM_MMAP_EXCLUSIVE       BIT_ULL(56)
1200 #define PM_FILE                 BIT_ULL(61)
1201 #define PM_SWAP                 BIT_ULL(62)
1202 #define PM_PRESENT              BIT_ULL(63)
1203
1204 #define PM_END_OF_BUFFER    1
1205
1206 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1207 {
1208         return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1209 }
1210
1211 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1212                           struct pagemapread *pm)
1213 {
1214         pm->buffer[pm->pos++] = *pme;
1215         if (pm->pos >= pm->len)
1216                 return PM_END_OF_BUFFER;
1217         return 0;
1218 }
1219
1220 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1221                                 struct mm_walk *walk)
1222 {
1223         struct pagemapread *pm = walk->private;
1224         unsigned long addr = start;
1225         int err = 0;
1226
1227         while (addr < end) {
1228                 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1229                 pagemap_entry_t pme = make_pme(0, 0);
1230                 /* End of address space hole, which we mark as non-present. */
1231                 unsigned long hole_end;
1232
1233                 if (vma)
1234                         hole_end = min(end, vma->vm_start);
1235                 else
1236                         hole_end = end;
1237
1238                 for (; addr < hole_end; addr += PAGE_SIZE) {
1239                         err = add_to_pagemap(addr, &pme, pm);
1240                         if (err)
1241                                 goto out;
1242                 }
1243
1244                 if (!vma)
1245                         break;
1246
1247                 /* Addresses in the VMA. */
1248                 if (vma->vm_flags & VM_SOFTDIRTY)
1249                         pme = make_pme(0, PM_SOFT_DIRTY);
1250                 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1251                         err = add_to_pagemap(addr, &pme, pm);
1252                         if (err)
1253                                 goto out;
1254                 }
1255         }
1256 out:
1257         return err;
1258 }
1259
1260 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1261                 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1262 {
1263         u64 frame = 0, flags = 0;
1264         struct page *page = NULL;
1265
1266         if (pte_present(pte)) {
1267                 if (pm->show_pfn)
1268                         frame = pte_pfn(pte);
1269                 flags |= PM_PRESENT;
1270                 page = _vm_normal_page(vma, addr, pte, true);
1271                 if (pte_soft_dirty(pte))
1272                         flags |= PM_SOFT_DIRTY;
1273         } else if (is_swap_pte(pte)) {
1274                 swp_entry_t entry;
1275                 if (pte_swp_soft_dirty(pte))
1276                         flags |= PM_SOFT_DIRTY;
1277                 entry = pte_to_swp_entry(pte);
1278                 frame = swp_type(entry) |
1279                         (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1280                 flags |= PM_SWAP;
1281                 if (is_migration_entry(entry))
1282                         page = migration_entry_to_page(entry);
1283
1284                 if (is_device_private_entry(entry))
1285                         page = device_private_entry_to_page(entry);
1286         }
1287
1288         if (page && !PageAnon(page))
1289                 flags |= PM_FILE;
1290         if (page && page_mapcount(page) == 1)
1291                 flags |= PM_MMAP_EXCLUSIVE;
1292         if (vma->vm_flags & VM_SOFTDIRTY)
1293                 flags |= PM_SOFT_DIRTY;
1294
1295         return make_pme(frame, flags);
1296 }
1297
1298 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1299                              struct mm_walk *walk)
1300 {
1301         struct vm_area_struct *vma = walk->vma;
1302         struct pagemapread *pm = walk->private;
1303         spinlock_t *ptl;
1304         pte_t *pte, *orig_pte;
1305         int err = 0;
1306
1307 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1308         ptl = pmd_trans_huge_lock(pmdp, vma);
1309         if (ptl) {
1310                 u64 flags = 0, frame = 0;
1311                 pmd_t pmd = *pmdp;
1312                 struct page *page = NULL;
1313
1314                 if (vma->vm_flags & VM_SOFTDIRTY)
1315                         flags |= PM_SOFT_DIRTY;
1316
1317                 if (pmd_present(pmd)) {
1318                         page = pmd_page(pmd);
1319
1320                         flags |= PM_PRESENT;
1321                         if (pmd_soft_dirty(pmd))
1322                                 flags |= PM_SOFT_DIRTY;
1323                         if (pm->show_pfn)
1324                                 frame = pmd_pfn(pmd) +
1325                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1326                 }
1327 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1328                 else if (is_swap_pmd(pmd)) {
1329                         swp_entry_t entry = pmd_to_swp_entry(pmd);
1330
1331                         frame = swp_type(entry) |
1332                                 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1333                         flags |= PM_SWAP;
1334                         if (pmd_swp_soft_dirty(pmd))
1335                                 flags |= PM_SOFT_DIRTY;
1336                         VM_BUG_ON(!is_pmd_migration_entry(pmd));
1337                         page = migration_entry_to_page(entry);
1338                 }
1339 #endif
1340
1341                 if (page && page_mapcount(page) == 1)
1342                         flags |= PM_MMAP_EXCLUSIVE;
1343
1344                 for (; addr != end; addr += PAGE_SIZE) {
1345                         pagemap_entry_t pme = make_pme(frame, flags);
1346
1347                         err = add_to_pagemap(addr, &pme, pm);
1348                         if (err)
1349                                 break;
1350                         if (pm->show_pfn && (flags & PM_PRESENT))
1351                                 frame++;
1352                 }
1353                 spin_unlock(ptl);
1354                 return err;
1355         }
1356
1357         if (pmd_trans_unstable(pmdp))
1358                 return 0;
1359 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1360
1361         /*
1362          * We can assume that @vma always points to a valid one and @end never
1363          * goes beyond vma->vm_end.
1364          */
1365         orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1366         for (; addr < end; pte++, addr += PAGE_SIZE) {
1367                 pagemap_entry_t pme;
1368
1369                 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1370                 err = add_to_pagemap(addr, &pme, pm);
1371                 if (err)
1372                         break;
1373         }
1374         pte_unmap_unlock(orig_pte, ptl);
1375
1376         cond_resched();
1377
1378         return err;
1379 }
1380
1381 #ifdef CONFIG_HUGETLB_PAGE
1382 /* This function walks within one hugetlb entry in the single call */
1383 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1384                                  unsigned long addr, unsigned long end,
1385                                  struct mm_walk *walk)
1386 {
1387         struct pagemapread *pm = walk->private;
1388         struct vm_area_struct *vma = walk->vma;
1389         u64 flags = 0, frame = 0;
1390         int err = 0;
1391         pte_t pte;
1392
1393         if (vma->vm_flags & VM_SOFTDIRTY)
1394                 flags |= PM_SOFT_DIRTY;
1395
1396         pte = huge_ptep_get(ptep);
1397         if (pte_present(pte)) {
1398                 struct page *page = pte_page(pte);
1399
1400                 if (!PageAnon(page))
1401                         flags |= PM_FILE;
1402
1403                 if (page_mapcount(page) == 1)
1404                         flags |= PM_MMAP_EXCLUSIVE;
1405
1406                 flags |= PM_PRESENT;
1407                 if (pm->show_pfn)
1408                         frame = pte_pfn(pte) +
1409                                 ((addr & ~hmask) >> PAGE_SHIFT);
1410         }
1411
1412         for (; addr != end; addr += PAGE_SIZE) {
1413                 pagemap_entry_t pme = make_pme(frame, flags);
1414
1415                 err = add_to_pagemap(addr, &pme, pm);
1416                 if (err)
1417                         return err;
1418                 if (pm->show_pfn && (flags & PM_PRESENT))
1419                         frame++;
1420         }
1421
1422         cond_resched();
1423
1424         return err;
1425 }
1426 #endif /* HUGETLB_PAGE */
1427
1428 /*
1429  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1430  *
1431  * For each page in the address space, this file contains one 64-bit entry
1432  * consisting of the following:
1433  *
1434  * Bits 0-54  page frame number (PFN) if present
1435  * Bits 0-4   swap type if swapped
1436  * Bits 5-54  swap offset if swapped
1437  * Bit  55    pte is soft-dirty (see Documentation/vm/soft-dirty.txt)
1438  * Bit  56    page exclusively mapped
1439  * Bits 57-60 zero
1440  * Bit  61    page is file-page or shared-anon
1441  * Bit  62    page swapped
1442  * Bit  63    page present
1443  *
1444  * If the page is not present but in swap, then the PFN contains an
1445  * encoding of the swap file number and the page's offset into the
1446  * swap. Unmapped pages return a null PFN. This allows determining
1447  * precisely which pages are mapped (or in swap) and comparing mapped
1448  * pages between processes.
1449  *
1450  * Efficient users of this interface will use /proc/pid/maps to
1451  * determine which areas of memory are actually mapped and llseek to
1452  * skip over unmapped regions.
1453  */
1454 static ssize_t pagemap_read(struct file *file, char __user *buf,
1455                             size_t count, loff_t *ppos)
1456 {
1457         struct mm_struct *mm = file->private_data;
1458         struct pagemapread pm;
1459         struct mm_walk pagemap_walk = {};
1460         unsigned long src;
1461         unsigned long svpfn;
1462         unsigned long start_vaddr;
1463         unsigned long end_vaddr;
1464         int ret = 0, copied = 0;
1465
1466         if (!mm || !mmget_not_zero(mm))
1467                 goto out;
1468
1469         ret = -EINVAL;
1470         /* file position must be aligned */
1471         if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1472                 goto out_mm;
1473
1474         ret = 0;
1475         if (!count)
1476                 goto out_mm;
1477
1478         /* do not disclose physical addresses: attack vector */
1479         pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1480
1481         pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1482         pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_KERNEL);
1483         ret = -ENOMEM;
1484         if (!pm.buffer)
1485                 goto out_mm;
1486
1487         pagemap_walk.pmd_entry = pagemap_pmd_range;
1488         pagemap_walk.pte_hole = pagemap_pte_hole;
1489 #ifdef CONFIG_HUGETLB_PAGE
1490         pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1491 #endif
1492         pagemap_walk.mm = mm;
1493         pagemap_walk.private = &pm;
1494
1495         src = *ppos;
1496         svpfn = src / PM_ENTRY_BYTES;
1497         start_vaddr = svpfn << PAGE_SHIFT;
1498         end_vaddr = mm->task_size;
1499
1500         /* watch out for wraparound */
1501         if (svpfn > mm->task_size >> PAGE_SHIFT)
1502                 start_vaddr = end_vaddr;
1503
1504         /*
1505          * The odds are that this will stop walking way
1506          * before end_vaddr, because the length of the
1507          * user buffer is tracked in "pm", and the walk
1508          * will stop when we hit the end of the buffer.
1509          */
1510         ret = 0;
1511         while (count && (start_vaddr < end_vaddr)) {
1512                 int len;
1513                 unsigned long end;
1514
1515                 pm.pos = 0;
1516                 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1517                 /* overflow ? */
1518                 if (end < start_vaddr || end > end_vaddr)
1519                         end = end_vaddr;
1520                 down_read(&mm->mmap_sem);
1521                 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1522                 up_read(&mm->mmap_sem);
1523                 start_vaddr = end;
1524
1525                 len = min(count, PM_ENTRY_BYTES * pm.pos);
1526                 if (copy_to_user(buf, pm.buffer, len)) {
1527                         ret = -EFAULT;
1528                         goto out_free;
1529                 }
1530                 copied += len;
1531                 buf += len;
1532                 count -= len;
1533         }
1534         *ppos += copied;
1535         if (!ret || ret == PM_END_OF_BUFFER)
1536                 ret = copied;
1537
1538 out_free:
1539         kfree(pm.buffer);
1540 out_mm:
1541         mmput(mm);
1542 out:
1543         return ret;
1544 }
1545
1546 static int pagemap_open(struct inode *inode, struct file *file)
1547 {
1548         struct mm_struct *mm;
1549
1550         mm = proc_mem_open(inode, PTRACE_MODE_READ);
1551         if (IS_ERR(mm))
1552                 return PTR_ERR(mm);
1553         file->private_data = mm;
1554         return 0;
1555 }
1556
1557 static int pagemap_release(struct inode *inode, struct file *file)
1558 {
1559         struct mm_struct *mm = file->private_data;
1560
1561         if (mm)
1562                 mmdrop(mm);
1563         return 0;
1564 }
1565
1566 const struct file_operations proc_pagemap_operations = {
1567         .llseek         = mem_lseek, /* borrow this */
1568         .read           = pagemap_read,
1569         .open           = pagemap_open,
1570         .release        = pagemap_release,
1571 };
1572 #endif /* CONFIG_PROC_PAGE_MONITOR */
1573
1574 #ifdef CONFIG_NUMA
1575
1576 struct numa_maps {
1577         unsigned long pages;
1578         unsigned long anon;
1579         unsigned long active;
1580         unsigned long writeback;
1581         unsigned long mapcount_max;
1582         unsigned long dirty;
1583         unsigned long swapcache;
1584         unsigned long node[MAX_NUMNODES];
1585 };
1586
1587 struct numa_maps_private {
1588         struct proc_maps_private proc_maps;
1589         struct numa_maps md;
1590 };
1591
1592 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1593                         unsigned long nr_pages)
1594 {
1595         int count = page_mapcount(page);
1596
1597         md->pages += nr_pages;
1598         if (pte_dirty || PageDirty(page))
1599                 md->dirty += nr_pages;
1600
1601         if (PageSwapCache(page))
1602                 md->swapcache += nr_pages;
1603
1604         if (PageActive(page) || PageUnevictable(page))
1605                 md->active += nr_pages;
1606
1607         if (PageWriteback(page))
1608                 md->writeback += nr_pages;
1609
1610         if (PageAnon(page))
1611                 md->anon += nr_pages;
1612
1613         if (count > md->mapcount_max)
1614                 md->mapcount_max = count;
1615
1616         md->node[page_to_nid(page)] += nr_pages;
1617 }
1618
1619 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1620                 unsigned long addr)
1621 {
1622         struct page *page;
1623         int nid;
1624
1625         if (!pte_present(pte))
1626                 return NULL;
1627
1628         page = vm_normal_page(vma, addr, pte);
1629         if (!page)
1630                 return NULL;
1631
1632         if (PageReserved(page))
1633                 return NULL;
1634
1635         nid = page_to_nid(page);
1636         if (!node_isset(nid, node_states[N_MEMORY]))
1637                 return NULL;
1638
1639         return page;
1640 }
1641
1642 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1643 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1644                                               struct vm_area_struct *vma,
1645                                               unsigned long addr)
1646 {
1647         struct page *page;
1648         int nid;
1649
1650         if (!pmd_present(pmd))
1651                 return NULL;
1652
1653         page = vm_normal_page_pmd(vma, addr, pmd);
1654         if (!page)
1655                 return NULL;
1656
1657         if (PageReserved(page))
1658                 return NULL;
1659
1660         nid = page_to_nid(page);
1661         if (!node_isset(nid, node_states[N_MEMORY]))
1662                 return NULL;
1663
1664         return page;
1665 }
1666 #endif
1667
1668 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1669                 unsigned long end, struct mm_walk *walk)
1670 {
1671         struct numa_maps *md = walk->private;
1672         struct vm_area_struct *vma = walk->vma;
1673         spinlock_t *ptl;
1674         pte_t *orig_pte;
1675         pte_t *pte;
1676
1677 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1678         ptl = pmd_trans_huge_lock(pmd, vma);
1679         if (ptl) {
1680                 struct page *page;
1681
1682                 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1683                 if (page)
1684                         gather_stats(page, md, pmd_dirty(*pmd),
1685                                      HPAGE_PMD_SIZE/PAGE_SIZE);
1686                 spin_unlock(ptl);
1687                 return 0;
1688         }
1689
1690         if (pmd_trans_unstable(pmd))
1691                 return 0;
1692 #endif
1693         orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1694         do {
1695                 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1696                 if (!page)
1697                         continue;
1698                 gather_stats(page, md, pte_dirty(*pte), 1);
1699
1700         } while (pte++, addr += PAGE_SIZE, addr != end);
1701         pte_unmap_unlock(orig_pte, ptl);
1702         cond_resched();
1703         return 0;
1704 }
1705 #ifdef CONFIG_HUGETLB_PAGE
1706 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1707                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1708 {
1709         pte_t huge_pte = huge_ptep_get(pte);
1710         struct numa_maps *md;
1711         struct page *page;
1712
1713         if (!pte_present(huge_pte))
1714                 return 0;
1715
1716         page = pte_page(huge_pte);
1717         if (!page)
1718                 return 0;
1719
1720         md = walk->private;
1721         gather_stats(page, md, pte_dirty(huge_pte), 1);
1722         return 0;
1723 }
1724
1725 #else
1726 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1727                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1728 {
1729         return 0;
1730 }
1731 #endif
1732
1733 /*
1734  * Display pages allocated per node and memory policy via /proc.
1735  */
1736 static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1737 {
1738         struct numa_maps_private *numa_priv = m->private;
1739         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1740         struct vm_area_struct *vma = v;
1741         struct numa_maps *md = &numa_priv->md;
1742         struct file *file = vma->vm_file;
1743         struct mm_struct *mm = vma->vm_mm;
1744         struct mm_walk walk = {
1745                 .hugetlb_entry = gather_hugetlb_stats,
1746                 .pmd_entry = gather_pte_stats,
1747                 .private = md,
1748                 .mm = mm,
1749         };
1750         struct mempolicy *pol;
1751         char buffer[64];
1752         int nid;
1753
1754         if (!mm)
1755                 return 0;
1756
1757         /* Ensure we start with an empty set of numa_maps statistics. */
1758         memset(md, 0, sizeof(*md));
1759
1760         pol = __get_vma_policy(vma, vma->vm_start);
1761         if (pol) {
1762                 mpol_to_str(buffer, sizeof(buffer), pol);
1763                 mpol_cond_put(pol);
1764         } else {
1765                 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1766         }
1767
1768         seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1769
1770         if (file) {
1771                 seq_puts(m, " file=");
1772                 seq_file_path(m, file, "\n\t= ");
1773         } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1774                 seq_puts(m, " heap");
1775         } else if (is_stack(vma)) {
1776                 seq_puts(m, " stack");
1777         }
1778
1779         if (is_vm_hugetlb_page(vma))
1780                 seq_puts(m, " huge");
1781
1782         /* mmap_sem is held by m_start */
1783         walk_page_vma(vma, &walk);
1784
1785         if (!md->pages)
1786                 goto out;
1787
1788         if (md->anon)
1789                 seq_printf(m, " anon=%lu", md->anon);
1790
1791         if (md->dirty)
1792                 seq_printf(m, " dirty=%lu", md->dirty);
1793
1794         if (md->pages != md->anon && md->pages != md->dirty)
1795                 seq_printf(m, " mapped=%lu", md->pages);
1796
1797         if (md->mapcount_max > 1)
1798                 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1799
1800         if (md->swapcache)
1801                 seq_printf(m, " swapcache=%lu", md->swapcache);
1802
1803         if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1804                 seq_printf(m, " active=%lu", md->active);
1805
1806         if (md->writeback)
1807                 seq_printf(m, " writeback=%lu", md->writeback);
1808
1809         for_each_node_state(nid, N_MEMORY)
1810                 if (md->node[nid])
1811                         seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1812
1813         seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1814 out:
1815         seq_putc(m, '\n');
1816         m_cache_vma(m, vma);
1817         return 0;
1818 }
1819
1820 static int show_pid_numa_map(struct seq_file *m, void *v)
1821 {
1822         return show_numa_map(m, v, 1);
1823 }
1824
1825 static int show_tid_numa_map(struct seq_file *m, void *v)
1826 {
1827         return show_numa_map(m, v, 0);
1828 }
1829
1830 static const struct seq_operations proc_pid_numa_maps_op = {
1831         .start  = m_start,
1832         .next   = m_next,
1833         .stop   = m_stop,
1834         .show   = show_pid_numa_map,
1835 };
1836
1837 static const struct seq_operations proc_tid_numa_maps_op = {
1838         .start  = m_start,
1839         .next   = m_next,
1840         .stop   = m_stop,
1841         .show   = show_tid_numa_map,
1842 };
1843
1844 static int numa_maps_open(struct inode *inode, struct file *file,
1845                           const struct seq_operations *ops)
1846 {
1847         return proc_maps_open(inode, file, ops,
1848                                 sizeof(struct numa_maps_private));
1849 }
1850
1851 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1852 {
1853         return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1854 }
1855
1856 static int tid_numa_maps_open(struct inode *inode, struct file *file)
1857 {
1858         return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1859 }
1860
1861 const struct file_operations proc_pid_numa_maps_operations = {
1862         .open           = pid_numa_maps_open,
1863         .read           = seq_read,
1864         .llseek         = seq_lseek,
1865         .release        = proc_map_release,
1866 };
1867
1868 const struct file_operations proc_tid_numa_maps_operations = {
1869         .open           = tid_numa_maps_open,
1870         .read           = seq_read,
1871         .llseek         = seq_lseek,
1872         .release        = proc_map_release,
1873 };
1874 #endif /* CONFIG_NUMA */