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)
task_mem(struct seq_file * m,struct mm_struct * mm)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
task_vsize(struct mm_struct * mm)82 unsigned long task_vsize(struct mm_struct *mm)
83 {
84 return PAGE_SIZE * mm->total_vm;
85 }
86
task_statm(struct mm_struct * mm,unsigned long * shared,unsigned long * text,unsigned long * data,unsigned long * resident)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 */
hold_task_mempolicy(struct proc_maps_private * priv)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 }
release_task_mempolicy(struct proc_maps_private * priv)113 static void release_task_mempolicy(struct proc_maps_private *priv)
114 {
115 mpol_put(priv->task_mempolicy);
116 }
117 #else
hold_task_mempolicy(struct proc_maps_private * priv)118 static void hold_task_mempolicy(struct proc_maps_private *priv)
119 {
120 }
release_task_mempolicy(struct proc_maps_private * priv)121 static void release_task_mempolicy(struct proc_maps_private *priv)
122 {
123 }
124 #endif
125
seq_print_vma_name(struct seq_file * m,struct vm_area_struct * vma)126 static void seq_print_vma_name(struct seq_file *m, struct vm_area_struct *vma)
127 {
128 const char __user *name = vma_get_anon_name(vma);
129 struct mm_struct *mm = vma->vm_mm;
130
131 unsigned long page_start_vaddr;
132 unsigned long page_offset;
133 unsigned long num_pages;
134 unsigned long max_len = NAME_MAX;
135 int i;
136
137 page_start_vaddr = (unsigned long)name & PAGE_MASK;
138 page_offset = (unsigned long)name - page_start_vaddr;
139 num_pages = DIV_ROUND_UP(page_offset + max_len, PAGE_SIZE);
140
141 seq_puts(m, "[anon:");
142
143 for (i = 0; i < num_pages; i++) {
144 int len;
145 int write_len;
146 const char *kaddr;
147 long pages_pinned;
148 struct page *page;
149
150 pages_pinned = get_user_pages_remote(current, mm,
151 page_start_vaddr, 1, 0, &page, NULL, NULL);
152 if (pages_pinned < 1) {
153 seq_puts(m, "<fault>]");
154 return;
155 }
156
157 kaddr = (const char *)kmap(page);
158 len = min(max_len, PAGE_SIZE - page_offset);
159 write_len = strnlen(kaddr + page_offset, len);
160 seq_write(m, kaddr + page_offset, write_len);
161 kunmap(page);
162 put_page(page);
163
164 /* if strnlen hit a null terminator then we're done */
165 if (write_len != len)
166 break;
167
168 max_len -= len;
169 page_offset = 0;
170 page_start_vaddr += PAGE_SIZE;
171 }
172
173 seq_putc(m, ']');
174 }
175
vma_stop(struct proc_maps_private * priv)176 static void vma_stop(struct proc_maps_private *priv)
177 {
178 struct mm_struct *mm = priv->mm;
179
180 release_task_mempolicy(priv);
181 up_read(&mm->mmap_sem);
182 mmput(mm);
183 }
184
185 static struct vm_area_struct *
m_next_vma(struct proc_maps_private * priv,struct vm_area_struct * vma)186 m_next_vma(struct proc_maps_private *priv, struct vm_area_struct *vma)
187 {
188 if (vma == priv->tail_vma)
189 return NULL;
190 return vma->vm_next ?: priv->tail_vma;
191 }
192
m_cache_vma(struct seq_file * m,struct vm_area_struct * vma)193 static void m_cache_vma(struct seq_file *m, struct vm_area_struct *vma)
194 {
195 if (m->count < m->size) /* vma is copied successfully */
196 m->version = m_next_vma(m->private, vma) ? vma->vm_end : -1UL;
197 }
198
m_start(struct seq_file * m,loff_t * ppos)199 static void *m_start(struct seq_file *m, loff_t *ppos)
200 {
201 struct proc_maps_private *priv = m->private;
202 unsigned long last_addr = m->version;
203 struct mm_struct *mm;
204 struct vm_area_struct *vma;
205 unsigned int pos = *ppos;
206
207 /* See m_cache_vma(). Zero at the start or after lseek. */
208 if (last_addr == -1UL)
209 return NULL;
210
211 priv->task = get_proc_task(priv->inode);
212 if (!priv->task)
213 return ERR_PTR(-ESRCH);
214
215 mm = priv->mm;
216 if (!mm || !mmget_not_zero(mm))
217 return NULL;
218
219 if (down_read_killable(&mm->mmap_sem)) {
220 mmput(mm);
221 return ERR_PTR(-EINTR);
222 }
223
224 hold_task_mempolicy(priv);
225 priv->tail_vma = get_gate_vma(mm);
226
227 if (last_addr) {
228 vma = find_vma(mm, last_addr - 1);
229 if (vma && vma->vm_start <= last_addr)
230 vma = m_next_vma(priv, vma);
231 if (vma)
232 return vma;
233 }
234
235 m->version = 0;
236 if (pos < mm->map_count) {
237 for (vma = mm->mmap; pos; pos--) {
238 m->version = vma->vm_start;
239 vma = vma->vm_next;
240 }
241 return vma;
242 }
243
244 /* we do not bother to update m->version in this case */
245 if (pos == mm->map_count && priv->tail_vma)
246 return priv->tail_vma;
247
248 vma_stop(priv);
249 return NULL;
250 }
251
m_next(struct seq_file * m,void * v,loff_t * pos)252 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
253 {
254 struct proc_maps_private *priv = m->private;
255 struct vm_area_struct *next;
256
257 (*pos)++;
258 next = m_next_vma(priv, v);
259 if (!next)
260 vma_stop(priv);
261 return next;
262 }
263
m_stop(struct seq_file * m,void * v)264 static void m_stop(struct seq_file *m, void *v)
265 {
266 struct proc_maps_private *priv = m->private;
267
268 if (!IS_ERR_OR_NULL(v))
269 vma_stop(priv);
270 if (priv->task) {
271 put_task_struct(priv->task);
272 priv->task = NULL;
273 }
274 }
275
proc_maps_open(struct inode * inode,struct file * file,const struct seq_operations * ops,int psize)276 static int proc_maps_open(struct inode *inode, struct file *file,
277 const struct seq_operations *ops, int psize)
278 {
279 struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
280
281 if (!priv)
282 return -ENOMEM;
283
284 priv->inode = inode;
285 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
286 if (IS_ERR(priv->mm)) {
287 int err = PTR_ERR(priv->mm);
288
289 seq_release_private(inode, file);
290 return err;
291 }
292
293 return 0;
294 }
295
proc_map_release(struct inode * inode,struct file * file)296 static int proc_map_release(struct inode *inode, struct file *file)
297 {
298 struct seq_file *seq = file->private_data;
299 struct proc_maps_private *priv = seq->private;
300
301 if (priv->mm)
302 mmdrop(priv->mm);
303
304 return seq_release_private(inode, file);
305 }
306
do_maps_open(struct inode * inode,struct file * file,const struct seq_operations * ops)307 static int do_maps_open(struct inode *inode, struct file *file,
308 const struct seq_operations *ops)
309 {
310 return proc_maps_open(inode, file, ops,
311 sizeof(struct proc_maps_private));
312 }
313
314 /*
315 * Indicate if the VMA is a stack for the given task; for
316 * /proc/PID/maps that is the stack of the main task.
317 */
is_stack(struct vm_area_struct * vma)318 static int is_stack(struct vm_area_struct *vma)
319 {
320 /*
321 * We make no effort to guess what a given thread considers to be
322 * its "stack". It's not even well-defined for programs written
323 * languages like Go.
324 */
325 return vma->vm_start <= vma->vm_mm->start_stack &&
326 vma->vm_end >= vma->vm_mm->start_stack;
327 }
328
show_vma_header_prefix(struct seq_file * m,unsigned long start,unsigned long end,vm_flags_t flags,unsigned long long pgoff,dev_t dev,unsigned long ino)329 static void show_vma_header_prefix(struct seq_file *m,
330 unsigned long start, unsigned long end,
331 vm_flags_t flags, unsigned long long pgoff,
332 dev_t dev, unsigned long ino)
333 {
334 seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
335 seq_put_hex_ll(m, NULL, start, 8);
336 seq_put_hex_ll(m, "-", end, 8);
337 seq_putc(m, ' ');
338 seq_putc(m, flags & VM_READ ? 'r' : '-');
339 seq_putc(m, flags & VM_WRITE ? 'w' : '-');
340 seq_putc(m, flags & VM_EXEC ? 'x' : '-');
341 seq_putc(m, flags & VM_MAYSHARE ? 's' : 'p');
342 seq_put_hex_ll(m, " ", pgoff, 8);
343 seq_put_hex_ll(m, " ", MAJOR(dev), 2);
344 seq_put_hex_ll(m, ":", MINOR(dev), 2);
345 seq_put_decimal_ull(m, " ", ino);
346 seq_putc(m, ' ');
347 }
348
349 static void
show_map_vma(struct seq_file * m,struct vm_area_struct * vma)350 show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
351 {
352 struct mm_struct *mm = vma->vm_mm;
353 struct file *file = vma->vm_file;
354 vm_flags_t flags = vma->vm_flags;
355 unsigned long ino = 0;
356 unsigned long long pgoff = 0;
357 unsigned long start, end;
358 dev_t dev = 0;
359 const char *name = NULL;
360
361 if (file) {
362 struct inode *inode = file_inode(vma->vm_file);
363 dev = inode->i_sb->s_dev;
364 ino = inode->i_ino;
365 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
366 }
367
368 start = vma->vm_start;
369 end = vma->vm_end;
370 show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino);
371
372 /*
373 * Print the dentry name for named mappings, and a
374 * special [heap] marker for the heap:
375 */
376 if (file) {
377 seq_pad(m, ' ');
378 seq_file_path(m, file, "\n");
379 goto done;
380 }
381
382 if (vma->vm_ops && vma->vm_ops->name) {
383 name = vma->vm_ops->name(vma);
384 if (name)
385 goto done;
386 }
387
388 name = arch_vma_name(vma);
389 if (!name) {
390 if (!mm) {
391 name = "[vdso]";
392 goto done;
393 }
394
395 if (vma->vm_start <= mm->brk &&
396 vma->vm_end >= mm->start_brk) {
397 name = "[heap]";
398 goto done;
399 }
400
401 if (is_stack(vma)) {
402 name = "[stack]";
403 goto done;
404 }
405
406 if (vma_get_anon_name(vma)) {
407 seq_pad(m, ' ');
408 seq_print_vma_name(m, vma);
409 }
410 }
411
412 done:
413 if (name) {
414 seq_pad(m, ' ');
415 seq_puts(m, name);
416 }
417 seq_putc(m, '\n');
418 }
419
show_map(struct seq_file * m,void * v)420 static int show_map(struct seq_file *m, void *v)
421 {
422 show_map_vma(m, v);
423 m_cache_vma(m, v);
424 return 0;
425 }
426
427 static const struct seq_operations proc_pid_maps_op = {
428 .start = m_start,
429 .next = m_next,
430 .stop = m_stop,
431 .show = show_map
432 };
433
pid_maps_open(struct inode * inode,struct file * file)434 static int pid_maps_open(struct inode *inode, struct file *file)
435 {
436 return do_maps_open(inode, file, &proc_pid_maps_op);
437 }
438
439 const struct file_operations proc_pid_maps_operations = {
440 .open = pid_maps_open,
441 .read = seq_read,
442 .llseek = seq_lseek,
443 .release = proc_map_release,
444 };
445
446 /*
447 * Proportional Set Size(PSS): my share of RSS.
448 *
449 * PSS of a process is the count of pages it has in memory, where each
450 * page is divided by the number of processes sharing it. So if a
451 * process has 1000 pages all to itself, and 1000 shared with one other
452 * process, its PSS will be 1500.
453 *
454 * To keep (accumulated) division errors low, we adopt a 64bit
455 * fixed-point pss counter to minimize division errors. So (pss >>
456 * PSS_SHIFT) would be the real byte count.
457 *
458 * A shift of 12 before division means (assuming 4K page size):
459 * - 1M 3-user-pages add up to 8KB errors;
460 * - supports mapcount up to 2^24, or 16M;
461 * - supports PSS up to 2^52 bytes, or 4PB.
462 */
463 #define PSS_SHIFT 12
464
465 #ifdef CONFIG_PROC_PAGE_MONITOR
466 struct mem_size_stats {
467 unsigned long resident;
468 unsigned long shared_clean;
469 unsigned long shared_dirty;
470 unsigned long private_clean;
471 unsigned long private_dirty;
472 unsigned long referenced;
473 unsigned long anonymous;
474 unsigned long lazyfree;
475 unsigned long anonymous_thp;
476 unsigned long shmem_thp;
477 unsigned long file_thp;
478 unsigned long swap;
479 unsigned long shared_hugetlb;
480 unsigned long private_hugetlb;
481 u64 pss;
482 u64 pss_anon;
483 u64 pss_file;
484 u64 pss_shmem;
485 u64 pss_locked;
486 u64 swap_pss;
487 bool check_shmem_swap;
488 };
489
smaps_page_accumulate(struct mem_size_stats * mss,struct page * page,unsigned long size,unsigned long pss,bool dirty,bool locked,bool private)490 static void smaps_page_accumulate(struct mem_size_stats *mss,
491 struct page *page, unsigned long size, unsigned long pss,
492 bool dirty, bool locked, bool private)
493 {
494 mss->pss += pss;
495
496 if (PageAnon(page))
497 mss->pss_anon += pss;
498 else if (PageSwapBacked(page))
499 mss->pss_shmem += pss;
500 else
501 mss->pss_file += pss;
502
503 if (locked)
504 mss->pss_locked += pss;
505
506 if (dirty || PageDirty(page)) {
507 if (private)
508 mss->private_dirty += size;
509 else
510 mss->shared_dirty += size;
511 } else {
512 if (private)
513 mss->private_clean += size;
514 else
515 mss->shared_clean += size;
516 }
517 }
518
smaps_account(struct mem_size_stats * mss,struct page * page,bool compound,bool young,bool dirty,bool locked)519 static void smaps_account(struct mem_size_stats *mss, struct page *page,
520 bool compound, bool young, bool dirty, bool locked)
521 {
522 int i, nr = compound ? compound_nr(page) : 1;
523 unsigned long size = nr * PAGE_SIZE;
524
525 /*
526 * First accumulate quantities that depend only on |size| and the type
527 * of the compound page.
528 */
529 if (PageAnon(page)) {
530 mss->anonymous += size;
531 if (!PageSwapBacked(page) && !dirty && !PageDirty(page))
532 mss->lazyfree += size;
533 }
534
535 mss->resident += size;
536 /* Accumulate the size in pages that have been accessed. */
537 if (young || page_is_young(page) || PageReferenced(page))
538 mss->referenced += size;
539
540 /*
541 * Then accumulate quantities that may depend on sharing, or that may
542 * differ page-by-page.
543 *
544 * page_count(page) == 1 guarantees the page is mapped exactly once.
545 * If any subpage of the compound page mapped with PTE it would elevate
546 * page_count().
547 */
548 if (page_count(page) == 1) {
549 smaps_page_accumulate(mss, page, size, size << PSS_SHIFT, dirty,
550 locked, true);
551 return;
552 }
553 for (i = 0; i < nr; i++, page++) {
554 int mapcount = page_mapcount(page);
555 unsigned long pss = PAGE_SIZE << PSS_SHIFT;
556 if (mapcount >= 2)
557 pss /= mapcount;
558 smaps_page_accumulate(mss, page, PAGE_SIZE, pss, dirty, locked,
559 mapcount < 2);
560 }
561 }
562
563 #ifdef CONFIG_SHMEM
smaps_pte_hole(unsigned long addr,unsigned long end,struct mm_walk * walk)564 static int smaps_pte_hole(unsigned long addr, unsigned long end,
565 struct mm_walk *walk)
566 {
567 struct mem_size_stats *mss = walk->private;
568
569 mss->swap += shmem_partial_swap_usage(
570 walk->vma->vm_file->f_mapping, addr, end);
571
572 return 0;
573 }
574 #else
575 #define smaps_pte_hole NULL
576 #endif /* CONFIG_SHMEM */
577
smaps_pte_entry(pte_t * pte,unsigned long addr,struct mm_walk * walk)578 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
579 struct mm_walk *walk)
580 {
581 struct mem_size_stats *mss = walk->private;
582 struct vm_area_struct *vma = walk->vma;
583 bool locked = !!(vma->vm_flags & VM_LOCKED);
584 struct page *page = NULL;
585
586 if (pte_present(*pte)) {
587 page = vm_normal_page(vma, addr, *pte);
588 } else if (is_swap_pte(*pte)) {
589 swp_entry_t swpent = pte_to_swp_entry(*pte);
590
591 if (!non_swap_entry(swpent)) {
592 int mapcount;
593
594 mss->swap += PAGE_SIZE;
595 mapcount = swp_swapcount(swpent);
596 if (mapcount >= 2) {
597 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
598
599 do_div(pss_delta, mapcount);
600 mss->swap_pss += pss_delta;
601 } else {
602 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
603 }
604 } else if (is_migration_entry(swpent))
605 page = migration_entry_to_page(swpent);
606 else if (is_device_private_entry(swpent))
607 page = device_private_entry_to_page(swpent);
608 } else if (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap
609 && pte_none(*pte))) {
610 page = find_get_entry(vma->vm_file->f_mapping,
611 linear_page_index(vma, addr));
612 if (!page)
613 return;
614
615 if (xa_is_value(page))
616 mss->swap += PAGE_SIZE;
617 else
618 put_page(page);
619
620 return;
621 }
622
623 if (!page)
624 return;
625
626 smaps_account(mss, page, false, pte_young(*pte), pte_dirty(*pte), locked);
627 }
628
629 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
smaps_pmd_entry(pmd_t * pmd,unsigned long addr,struct mm_walk * walk)630 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
631 struct mm_walk *walk)
632 {
633 struct mem_size_stats *mss = walk->private;
634 struct vm_area_struct *vma = walk->vma;
635 bool locked = !!(vma->vm_flags & VM_LOCKED);
636 struct page *page;
637
638 /* FOLL_DUMP will return -EFAULT on huge zero page */
639 page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
640 if (IS_ERR_OR_NULL(page))
641 return;
642 if (PageAnon(page))
643 mss->anonymous_thp += HPAGE_PMD_SIZE;
644 else if (PageSwapBacked(page))
645 mss->shmem_thp += HPAGE_PMD_SIZE;
646 else if (is_zone_device_page(page))
647 /* pass */;
648 else
649 mss->file_thp += HPAGE_PMD_SIZE;
650 smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd), locked);
651 }
652 #else
smaps_pmd_entry(pmd_t * pmd,unsigned long addr,struct mm_walk * walk)653 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
654 struct mm_walk *walk)
655 {
656 }
657 #endif
658
smaps_pte_range(pmd_t * pmd,unsigned long addr,unsigned long end,struct mm_walk * walk)659 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
660 struct mm_walk *walk)
661 {
662 struct vm_area_struct *vma = walk->vma;
663 pte_t *pte;
664 spinlock_t *ptl;
665
666 ptl = pmd_trans_huge_lock(pmd, vma);
667 if (ptl) {
668 if (pmd_present(*pmd))
669 smaps_pmd_entry(pmd, addr, walk);
670 spin_unlock(ptl);
671 goto out;
672 }
673
674 if (pmd_trans_unstable(pmd))
675 goto out;
676 /*
677 * The mmap_sem held all the way back in m_start() is what
678 * keeps khugepaged out of here and from collapsing things
679 * in here.
680 */
681 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
682 for (; addr != end; pte++, addr += PAGE_SIZE)
683 smaps_pte_entry(pte, addr, walk);
684 pte_unmap_unlock(pte - 1, ptl);
685 out:
686 cond_resched();
687 return 0;
688 }
689
show_smap_vma_flags(struct seq_file * m,struct vm_area_struct * vma)690 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
691 {
692 /*
693 * Don't forget to update Documentation/ on changes.
694 */
695 static const char mnemonics[BITS_PER_LONG][2] = {
696 /*
697 * In case if we meet a flag we don't know about.
698 */
699 [0 ... (BITS_PER_LONG-1)] = "??",
700
701 [ilog2(VM_READ)] = "rd",
702 [ilog2(VM_WRITE)] = "wr",
703 [ilog2(VM_EXEC)] = "ex",
704 [ilog2(VM_SHARED)] = "sh",
705 [ilog2(VM_MAYREAD)] = "mr",
706 [ilog2(VM_MAYWRITE)] = "mw",
707 [ilog2(VM_MAYEXEC)] = "me",
708 [ilog2(VM_MAYSHARE)] = "ms",
709 [ilog2(VM_GROWSDOWN)] = "gd",
710 [ilog2(VM_PFNMAP)] = "pf",
711 [ilog2(VM_DENYWRITE)] = "dw",
712 #ifdef CONFIG_X86_INTEL_MPX
713 [ilog2(VM_MPX)] = "mp",
714 #endif
715 [ilog2(VM_LOCKED)] = "lo",
716 [ilog2(VM_IO)] = "io",
717 [ilog2(VM_SEQ_READ)] = "sr",
718 [ilog2(VM_RAND_READ)] = "rr",
719 [ilog2(VM_DONTCOPY)] = "dc",
720 [ilog2(VM_DONTEXPAND)] = "de",
721 [ilog2(VM_ACCOUNT)] = "ac",
722 [ilog2(VM_NORESERVE)] = "nr",
723 [ilog2(VM_HUGETLB)] = "ht",
724 [ilog2(VM_SYNC)] = "sf",
725 [ilog2(VM_ARCH_1)] = "ar",
726 [ilog2(VM_WIPEONFORK)] = "wf",
727 [ilog2(VM_DONTDUMP)] = "dd",
728 #ifdef CONFIG_MEM_SOFT_DIRTY
729 [ilog2(VM_SOFTDIRTY)] = "sd",
730 #endif
731 [ilog2(VM_MIXEDMAP)] = "mm",
732 [ilog2(VM_HUGEPAGE)] = "hg",
733 [ilog2(VM_NOHUGEPAGE)] = "nh",
734 [ilog2(VM_MERGEABLE)] = "mg",
735 [ilog2(VM_UFFD_MISSING)]= "um",
736 [ilog2(VM_UFFD_WP)] = "uw",
737 #ifdef CONFIG_ARCH_HAS_PKEYS
738 /* These come out via ProtectionKey: */
739 [ilog2(VM_PKEY_BIT0)] = "",
740 [ilog2(VM_PKEY_BIT1)] = "",
741 [ilog2(VM_PKEY_BIT2)] = "",
742 [ilog2(VM_PKEY_BIT3)] = "",
743 #if VM_PKEY_BIT4
744 [ilog2(VM_PKEY_BIT4)] = "",
745 #endif
746 #endif /* CONFIG_ARCH_HAS_PKEYS */
747 };
748 size_t i;
749
750 seq_puts(m, "VmFlags: ");
751 for (i = 0; i < BITS_PER_LONG; i++) {
752 if (!mnemonics[i][0])
753 continue;
754 if (vma->vm_flags & (1UL << i)) {
755 seq_putc(m, mnemonics[i][0]);
756 seq_putc(m, mnemonics[i][1]);
757 seq_putc(m, ' ');
758 }
759 }
760 seq_putc(m, '\n');
761 }
762
763 #ifdef CONFIG_HUGETLB_PAGE
smaps_hugetlb_range(pte_t * pte,unsigned long hmask,unsigned long addr,unsigned long end,struct mm_walk * walk)764 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
765 unsigned long addr, unsigned long end,
766 struct mm_walk *walk)
767 {
768 struct mem_size_stats *mss = walk->private;
769 struct vm_area_struct *vma = walk->vma;
770 struct page *page = NULL;
771
772 if (pte_present(*pte)) {
773 page = vm_normal_page(vma, addr, *pte);
774 } else if (is_swap_pte(*pte)) {
775 swp_entry_t swpent = pte_to_swp_entry(*pte);
776
777 if (is_migration_entry(swpent))
778 page = migration_entry_to_page(swpent);
779 else if (is_device_private_entry(swpent))
780 page = device_private_entry_to_page(swpent);
781 }
782 if (page) {
783 int mapcount = page_mapcount(page);
784
785 if (mapcount >= 2)
786 mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
787 else
788 mss->private_hugetlb += huge_page_size(hstate_vma(vma));
789 }
790 return 0;
791 }
792 #else
793 #define smaps_hugetlb_range NULL
794 #endif /* HUGETLB_PAGE */
795
796 static const struct mm_walk_ops smaps_walk_ops = {
797 .pmd_entry = smaps_pte_range,
798 .hugetlb_entry = smaps_hugetlb_range,
799 };
800
801 static const struct mm_walk_ops smaps_shmem_walk_ops = {
802 .pmd_entry = smaps_pte_range,
803 .hugetlb_entry = smaps_hugetlb_range,
804 .pte_hole = smaps_pte_hole,
805 };
806
smap_gather_stats(struct vm_area_struct * vma,struct mem_size_stats * mss)807 static void smap_gather_stats(struct vm_area_struct *vma,
808 struct mem_size_stats *mss)
809 {
810 #ifdef CONFIG_SHMEM
811 /* In case of smaps_rollup, reset the value from previous vma */
812 mss->check_shmem_swap = false;
813 if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
814 /*
815 * For shared or readonly shmem mappings we know that all
816 * swapped out pages belong to the shmem object, and we can
817 * obtain the swap value much more efficiently. For private
818 * writable mappings, we might have COW pages that are
819 * not affected by the parent swapped out pages of the shmem
820 * object, so we have to distinguish them during the page walk.
821 * Unless we know that the shmem object (or the part mapped by
822 * our VMA) has no swapped out pages at all.
823 */
824 unsigned long shmem_swapped = shmem_swap_usage(vma);
825
826 if (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
827 !(vma->vm_flags & VM_WRITE)) {
828 mss->swap += shmem_swapped;
829 } else {
830 mss->check_shmem_swap = true;
831 walk_page_vma(vma, &smaps_shmem_walk_ops, mss);
832 return;
833 }
834 }
835 #endif
836 /* mmap_sem is held in m_start */
837 walk_page_vma(vma, &smaps_walk_ops, mss);
838 }
839
840 #define SEQ_PUT_DEC(str, val) \
841 seq_put_decimal_ull_width(m, str, (val) >> 10, 8)
842
843 /* Show the contents common for smaps and smaps_rollup */
__show_smap(struct seq_file * m,const struct mem_size_stats * mss,bool rollup_mode)844 static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss,
845 bool rollup_mode)
846 {
847 SEQ_PUT_DEC("Rss: ", mss->resident);
848 SEQ_PUT_DEC(" kB\nPss: ", mss->pss >> PSS_SHIFT);
849 if (rollup_mode) {
850 /*
851 * These are meaningful only for smaps_rollup, otherwise two of
852 * them are zero, and the other one is the same as Pss.
853 */
854 SEQ_PUT_DEC(" kB\nPss_Anon: ",
855 mss->pss_anon >> PSS_SHIFT);
856 SEQ_PUT_DEC(" kB\nPss_File: ",
857 mss->pss_file >> PSS_SHIFT);
858 SEQ_PUT_DEC(" kB\nPss_Shmem: ",
859 mss->pss_shmem >> PSS_SHIFT);
860 }
861 SEQ_PUT_DEC(" kB\nShared_Clean: ", mss->shared_clean);
862 SEQ_PUT_DEC(" kB\nShared_Dirty: ", mss->shared_dirty);
863 SEQ_PUT_DEC(" kB\nPrivate_Clean: ", mss->private_clean);
864 SEQ_PUT_DEC(" kB\nPrivate_Dirty: ", mss->private_dirty);
865 SEQ_PUT_DEC(" kB\nReferenced: ", mss->referenced);
866 SEQ_PUT_DEC(" kB\nAnonymous: ", mss->anonymous);
867 SEQ_PUT_DEC(" kB\nLazyFree: ", mss->lazyfree);
868 SEQ_PUT_DEC(" kB\nAnonHugePages: ", mss->anonymous_thp);
869 SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp);
870 SEQ_PUT_DEC(" kB\nFilePmdMapped: ", mss->file_thp);
871 SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb);
872 seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ",
873 mss->private_hugetlb >> 10, 7);
874 SEQ_PUT_DEC(" kB\nSwap: ", mss->swap);
875 SEQ_PUT_DEC(" kB\nSwapPss: ",
876 mss->swap_pss >> PSS_SHIFT);
877 SEQ_PUT_DEC(" kB\nLocked: ",
878 mss->pss_locked >> PSS_SHIFT);
879 seq_puts(m, " kB\n");
880 }
881
show_smap(struct seq_file * m,void * v)882 static int show_smap(struct seq_file *m, void *v)
883 {
884 struct vm_area_struct *vma = v;
885 struct mem_size_stats mss;
886
887 memset(&mss, 0, sizeof(mss));
888
889 smap_gather_stats(vma, &mss);
890
891 show_map_vma(m, vma);
892 if (vma_get_anon_name(vma)) {
893 seq_puts(m, "Name: ");
894 seq_print_vma_name(m, vma);
895 seq_putc(m, '\n');
896 }
897
898 SEQ_PUT_DEC("Size: ", vma->vm_end - vma->vm_start);
899 SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma));
900 SEQ_PUT_DEC(" kB\nMMUPageSize: ", vma_mmu_pagesize(vma));
901 seq_puts(m, " kB\n");
902
903 __show_smap(m, &mss, false);
904
905 seq_printf(m, "THPeligible: %d\n",
906 transparent_hugepage_enabled(vma));
907
908 if (arch_pkeys_enabled())
909 seq_printf(m, "ProtectionKey: %8u\n", vma_pkey(vma));
910 show_smap_vma_flags(m, vma);
911
912 m_cache_vma(m, vma);
913
914 return 0;
915 }
916
show_smaps_rollup(struct seq_file * m,void * v)917 static int show_smaps_rollup(struct seq_file *m, void *v)
918 {
919 struct proc_maps_private *priv = m->private;
920 struct mem_size_stats mss;
921 struct mm_struct *mm;
922 struct vm_area_struct *vma;
923 unsigned long last_vma_end = 0;
924 int ret = 0;
925
926 priv->task = get_proc_task(priv->inode);
927 if (!priv->task)
928 return -ESRCH;
929
930 mm = priv->mm;
931 if (!mm || !mmget_not_zero(mm)) {
932 ret = -ESRCH;
933 goto out_put_task;
934 }
935
936 memset(&mss, 0, sizeof(mss));
937
938 ret = down_read_killable(&mm->mmap_sem);
939 if (ret)
940 goto out_put_mm;
941
942 hold_task_mempolicy(priv);
943
944 for (vma = priv->mm->mmap; vma; vma = vma->vm_next) {
945 smap_gather_stats(vma, &mss);
946 last_vma_end = vma->vm_end;
947 }
948
949 show_vma_header_prefix(m, priv->mm->mmap->vm_start,
950 last_vma_end, 0, 0, 0, 0);
951 seq_pad(m, ' ');
952 seq_puts(m, "[rollup]\n");
953
954 __show_smap(m, &mss, true);
955
956 release_task_mempolicy(priv);
957 up_read(&mm->mmap_sem);
958
959 out_put_mm:
960 mmput(mm);
961 out_put_task:
962 put_task_struct(priv->task);
963 priv->task = NULL;
964
965 return ret;
966 }
967 #undef SEQ_PUT_DEC
968
969 static const struct seq_operations proc_pid_smaps_op = {
970 .start = m_start,
971 .next = m_next,
972 .stop = m_stop,
973 .show = show_smap
974 };
975
pid_smaps_open(struct inode * inode,struct file * file)976 static int pid_smaps_open(struct inode *inode, struct file *file)
977 {
978 return do_maps_open(inode, file, &proc_pid_smaps_op);
979 }
980
smaps_rollup_open(struct inode * inode,struct file * file)981 static int smaps_rollup_open(struct inode *inode, struct file *file)
982 {
983 int ret;
984 struct proc_maps_private *priv;
985
986 priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT);
987 if (!priv)
988 return -ENOMEM;
989
990 ret = single_open(file, show_smaps_rollup, priv);
991 if (ret)
992 goto out_free;
993
994 priv->inode = inode;
995 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
996 if (IS_ERR(priv->mm)) {
997 ret = PTR_ERR(priv->mm);
998
999 single_release(inode, file);
1000 goto out_free;
1001 }
1002
1003 return 0;
1004
1005 out_free:
1006 kfree(priv);
1007 return ret;
1008 }
1009
smaps_rollup_release(struct inode * inode,struct file * file)1010 static int smaps_rollup_release(struct inode *inode, struct file *file)
1011 {
1012 struct seq_file *seq = file->private_data;
1013 struct proc_maps_private *priv = seq->private;
1014
1015 if (priv->mm)
1016 mmdrop(priv->mm);
1017
1018 kfree(priv);
1019 return single_release(inode, file);
1020 }
1021
1022 const struct file_operations proc_pid_smaps_operations = {
1023 .open = pid_smaps_open,
1024 .read = seq_read,
1025 .llseek = seq_lseek,
1026 .release = proc_map_release,
1027 };
1028
1029 const struct file_operations proc_pid_smaps_rollup_operations = {
1030 .open = smaps_rollup_open,
1031 .read = seq_read,
1032 .llseek = seq_lseek,
1033 .release = smaps_rollup_release,
1034 };
1035
1036 enum clear_refs_types {
1037 CLEAR_REFS_ALL = 1,
1038 CLEAR_REFS_ANON,
1039 CLEAR_REFS_MAPPED,
1040 CLEAR_REFS_SOFT_DIRTY,
1041 CLEAR_REFS_MM_HIWATER_RSS,
1042 CLEAR_REFS_LAST,
1043 };
1044
1045 struct clear_refs_private {
1046 enum clear_refs_types type;
1047 };
1048
1049 #ifdef CONFIG_MEM_SOFT_DIRTY
clear_soft_dirty(struct vm_area_struct * vma,unsigned long addr,pte_t * pte)1050 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1051 unsigned long addr, pte_t *pte)
1052 {
1053 /*
1054 * The soft-dirty tracker uses #PF-s to catch writes
1055 * to pages, so write-protect the pte as well. See the
1056 * Documentation/admin-guide/mm/soft-dirty.rst for full description
1057 * of how soft-dirty works.
1058 */
1059 pte_t ptent = *pte;
1060
1061 if (pte_present(ptent)) {
1062 pte_t old_pte;
1063
1064 old_pte = ptep_modify_prot_start(vma, addr, pte);
1065 ptent = pte_wrprotect(old_pte);
1066 ptent = pte_clear_soft_dirty(ptent);
1067 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent);
1068 } else if (is_swap_pte(ptent)) {
1069 ptent = pte_swp_clear_soft_dirty(ptent);
1070 set_pte_at(vma->vm_mm, addr, pte, ptent);
1071 }
1072 }
1073 #else
clear_soft_dirty(struct vm_area_struct * vma,unsigned long addr,pte_t * pte)1074 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1075 unsigned long addr, pte_t *pte)
1076 {
1077 }
1078 #endif
1079
1080 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
clear_soft_dirty_pmd(struct vm_area_struct * vma,unsigned long addr,pmd_t * pmdp)1081 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1082 unsigned long addr, pmd_t *pmdp)
1083 {
1084 pmd_t old, pmd = *pmdp;
1085
1086 if (pmd_present(pmd)) {
1087 /* See comment in change_huge_pmd() */
1088 old = pmdp_invalidate(vma, addr, pmdp);
1089 if (pmd_dirty(old))
1090 pmd = pmd_mkdirty(pmd);
1091 if (pmd_young(old))
1092 pmd = pmd_mkyoung(pmd);
1093
1094 pmd = pmd_wrprotect(pmd);
1095 pmd = pmd_clear_soft_dirty(pmd);
1096
1097 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1098 } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1099 pmd = pmd_swp_clear_soft_dirty(pmd);
1100 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1101 }
1102 }
1103 #else
clear_soft_dirty_pmd(struct vm_area_struct * vma,unsigned long addr,pmd_t * pmdp)1104 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1105 unsigned long addr, pmd_t *pmdp)
1106 {
1107 }
1108 #endif
1109
clear_refs_pte_range(pmd_t * pmd,unsigned long addr,unsigned long end,struct mm_walk * walk)1110 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
1111 unsigned long end, struct mm_walk *walk)
1112 {
1113 struct clear_refs_private *cp = walk->private;
1114 struct vm_area_struct *vma = walk->vma;
1115 pte_t *pte, ptent;
1116 spinlock_t *ptl;
1117 struct page *page;
1118
1119 ptl = pmd_trans_huge_lock(pmd, vma);
1120 if (ptl) {
1121 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1122 clear_soft_dirty_pmd(vma, addr, pmd);
1123 goto out;
1124 }
1125
1126 if (!pmd_present(*pmd))
1127 goto out;
1128
1129 page = pmd_page(*pmd);
1130
1131 /* Clear accessed and referenced bits. */
1132 pmdp_test_and_clear_young(vma, addr, pmd);
1133 test_and_clear_page_young(page);
1134 ClearPageReferenced(page);
1135 out:
1136 spin_unlock(ptl);
1137 return 0;
1138 }
1139
1140 if (pmd_trans_unstable(pmd))
1141 return 0;
1142
1143 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1144 for (; addr != end; pte++, addr += PAGE_SIZE) {
1145 ptent = *pte;
1146
1147 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1148 clear_soft_dirty(vma, addr, pte);
1149 continue;
1150 }
1151
1152 if (!pte_present(ptent))
1153 continue;
1154
1155 page = vm_normal_page(vma, addr, ptent);
1156 if (!page)
1157 continue;
1158
1159 /* Clear accessed and referenced bits. */
1160 ptep_test_and_clear_young(vma, addr, pte);
1161 test_and_clear_page_young(page);
1162 ClearPageReferenced(page);
1163 }
1164 pte_unmap_unlock(pte - 1, ptl);
1165 cond_resched();
1166 return 0;
1167 }
1168
clear_refs_test_walk(unsigned long start,unsigned long end,struct mm_walk * walk)1169 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1170 struct mm_walk *walk)
1171 {
1172 struct clear_refs_private *cp = walk->private;
1173 struct vm_area_struct *vma = walk->vma;
1174
1175 if (vma->vm_flags & VM_PFNMAP)
1176 return 1;
1177
1178 /*
1179 * Writing 1 to /proc/pid/clear_refs affects all pages.
1180 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1181 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1182 * Writing 4 to /proc/pid/clear_refs affects all pages.
1183 */
1184 if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1185 return 1;
1186 if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1187 return 1;
1188 return 0;
1189 }
1190
1191 static const struct mm_walk_ops clear_refs_walk_ops = {
1192 .pmd_entry = clear_refs_pte_range,
1193 .test_walk = clear_refs_test_walk,
1194 };
1195
clear_refs_write(struct file * file,const char __user * buf,size_t count,loff_t * ppos)1196 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1197 size_t count, loff_t *ppos)
1198 {
1199 struct task_struct *task;
1200 char buffer[PROC_NUMBUF];
1201 struct mm_struct *mm;
1202 struct vm_area_struct *vma;
1203 enum clear_refs_types type;
1204 struct mmu_gather tlb;
1205 int itype;
1206 int rv;
1207
1208 memset(buffer, 0, sizeof(buffer));
1209 if (count > sizeof(buffer) - 1)
1210 count = sizeof(buffer) - 1;
1211 if (copy_from_user(buffer, buf, count))
1212 return -EFAULT;
1213 rv = kstrtoint(strstrip(buffer), 10, &itype);
1214 if (rv < 0)
1215 return rv;
1216 type = (enum clear_refs_types)itype;
1217 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1218 return -EINVAL;
1219
1220 task = get_proc_task(file_inode(file));
1221 if (!task)
1222 return -ESRCH;
1223 mm = get_task_mm(task);
1224 if (mm) {
1225 struct mmu_notifier_range range;
1226 struct clear_refs_private cp = {
1227 .type = type,
1228 };
1229
1230 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1231 if (down_write_killable(&mm->mmap_sem)) {
1232 count = -EINTR;
1233 goto out_mm;
1234 }
1235
1236 /*
1237 * Writing 5 to /proc/pid/clear_refs resets the peak
1238 * resident set size to this mm's current rss value.
1239 */
1240 reset_mm_hiwater_rss(mm);
1241 up_write(&mm->mmap_sem);
1242 goto out_mm;
1243 }
1244
1245 if (down_read_killable(&mm->mmap_sem)) {
1246 count = -EINTR;
1247 goto out_mm;
1248 }
1249 tlb_gather_mmu(&tlb, mm, 0, -1);
1250 if (type == CLEAR_REFS_SOFT_DIRTY) {
1251 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1252 if (!(vma->vm_flags & VM_SOFTDIRTY))
1253 continue;
1254 up_read(&mm->mmap_sem);
1255 if (down_write_killable(&mm->mmap_sem)) {
1256 count = -EINTR;
1257 goto out_mm;
1258 }
1259 /*
1260 * Avoid to modify vma->vm_flags
1261 * without locked ops while the
1262 * coredump reads the vm_flags.
1263 */
1264 if (!mmget_still_valid(mm)) {
1265 /*
1266 * Silently return "count"
1267 * like if get_task_mm()
1268 * failed. FIXME: should this
1269 * function have returned
1270 * -ESRCH if get_task_mm()
1271 * failed like if
1272 * get_proc_task() fails?
1273 */
1274 up_write(&mm->mmap_sem);
1275 goto out_mm;
1276 }
1277 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1278 vma->vm_flags &= ~VM_SOFTDIRTY;
1279 vma_set_page_prot(vma);
1280 }
1281 downgrade_write(&mm->mmap_sem);
1282 break;
1283 }
1284
1285 mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY,
1286 0, NULL, mm, 0, -1UL);
1287 mmu_notifier_invalidate_range_start(&range);
1288 }
1289 walk_page_range(mm, 0, mm->highest_vm_end, &clear_refs_walk_ops,
1290 &cp);
1291 if (type == CLEAR_REFS_SOFT_DIRTY)
1292 mmu_notifier_invalidate_range_end(&range);
1293 tlb_finish_mmu(&tlb, 0, -1);
1294 up_read(&mm->mmap_sem);
1295 out_mm:
1296 mmput(mm);
1297 }
1298 put_task_struct(task);
1299
1300 return count;
1301 }
1302
1303 const struct file_operations proc_clear_refs_operations = {
1304 .write = clear_refs_write,
1305 .llseek = noop_llseek,
1306 };
1307
1308 typedef struct {
1309 u64 pme;
1310 } pagemap_entry_t;
1311
1312 struct pagemapread {
1313 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */
1314 pagemap_entry_t *buffer;
1315 bool show_pfn;
1316 };
1317
1318 #define PAGEMAP_WALK_SIZE (PMD_SIZE)
1319 #define PAGEMAP_WALK_MASK (PMD_MASK)
1320
1321 #define PM_ENTRY_BYTES sizeof(pagemap_entry_t)
1322 #define PM_PFRAME_BITS 55
1323 #define PM_PFRAME_MASK GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1324 #define PM_SOFT_DIRTY BIT_ULL(55)
1325 #define PM_MMAP_EXCLUSIVE BIT_ULL(56)
1326 #define PM_FILE BIT_ULL(61)
1327 #define PM_SWAP BIT_ULL(62)
1328 #define PM_PRESENT BIT_ULL(63)
1329
1330 #define PM_END_OF_BUFFER 1
1331
make_pme(u64 frame,u64 flags)1332 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1333 {
1334 return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1335 }
1336
add_to_pagemap(unsigned long addr,pagemap_entry_t * pme,struct pagemapread * pm)1337 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1338 struct pagemapread *pm)
1339 {
1340 pm->buffer[pm->pos++] = *pme;
1341 if (pm->pos >= pm->len)
1342 return PM_END_OF_BUFFER;
1343 return 0;
1344 }
1345
pagemap_pte_hole(unsigned long start,unsigned long end,struct mm_walk * walk)1346 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1347 struct mm_walk *walk)
1348 {
1349 struct pagemapread *pm = walk->private;
1350 unsigned long addr = start;
1351 int err = 0;
1352
1353 while (addr < end) {
1354 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1355 pagemap_entry_t pme = make_pme(0, 0);
1356 /* End of address space hole, which we mark as non-present. */
1357 unsigned long hole_end;
1358
1359 if (vma)
1360 hole_end = min(end, vma->vm_start);
1361 else
1362 hole_end = end;
1363
1364 for (; addr < hole_end; addr += PAGE_SIZE) {
1365 err = add_to_pagemap(addr, &pme, pm);
1366 if (err)
1367 goto out;
1368 }
1369
1370 if (!vma)
1371 break;
1372
1373 /* Addresses in the VMA. */
1374 if (vma->vm_flags & VM_SOFTDIRTY)
1375 pme = make_pme(0, PM_SOFT_DIRTY);
1376 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1377 err = add_to_pagemap(addr, &pme, pm);
1378 if (err)
1379 goto out;
1380 }
1381 }
1382 out:
1383 return err;
1384 }
1385
pte_to_pagemap_entry(struct pagemapread * pm,struct vm_area_struct * vma,unsigned long addr,pte_t pte)1386 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1387 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1388 {
1389 u64 frame = 0, flags = 0;
1390 struct page *page = NULL;
1391
1392 if (pte_present(pte)) {
1393 if (pm->show_pfn)
1394 frame = pte_pfn(pte);
1395 flags |= PM_PRESENT;
1396 page = vm_normal_page(vma, addr, pte);
1397 if (pte_soft_dirty(pte))
1398 flags |= PM_SOFT_DIRTY;
1399 } else if (is_swap_pte(pte)) {
1400 swp_entry_t entry;
1401 if (pte_swp_soft_dirty(pte))
1402 flags |= PM_SOFT_DIRTY;
1403 entry = pte_to_swp_entry(pte);
1404 if (pm->show_pfn)
1405 frame = swp_type(entry) |
1406 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1407 flags |= PM_SWAP;
1408 if (is_migration_entry(entry))
1409 page = migration_entry_to_page(entry);
1410
1411 if (is_device_private_entry(entry))
1412 page = device_private_entry_to_page(entry);
1413 }
1414
1415 if (page && !PageAnon(page))
1416 flags |= PM_FILE;
1417 if (page && page_mapcount(page) == 1)
1418 flags |= PM_MMAP_EXCLUSIVE;
1419 if (vma->vm_flags & VM_SOFTDIRTY)
1420 flags |= PM_SOFT_DIRTY;
1421
1422 return make_pme(frame, flags);
1423 }
1424
pagemap_pmd_range(pmd_t * pmdp,unsigned long addr,unsigned long end,struct mm_walk * walk)1425 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1426 struct mm_walk *walk)
1427 {
1428 struct vm_area_struct *vma = walk->vma;
1429 struct pagemapread *pm = walk->private;
1430 spinlock_t *ptl;
1431 pte_t *pte, *orig_pte;
1432 int err = 0;
1433
1434 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1435 ptl = pmd_trans_huge_lock(pmdp, vma);
1436 if (ptl) {
1437 u64 flags = 0, frame = 0;
1438 pmd_t pmd = *pmdp;
1439 struct page *page = NULL;
1440
1441 if (vma->vm_flags & VM_SOFTDIRTY)
1442 flags |= PM_SOFT_DIRTY;
1443
1444 if (pmd_present(pmd)) {
1445 page = pmd_page(pmd);
1446
1447 flags |= PM_PRESENT;
1448 if (pmd_soft_dirty(pmd))
1449 flags |= PM_SOFT_DIRTY;
1450 if (pm->show_pfn)
1451 frame = pmd_pfn(pmd) +
1452 ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1453 }
1454 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1455 else if (is_swap_pmd(pmd)) {
1456 swp_entry_t entry = pmd_to_swp_entry(pmd);
1457 unsigned long offset;
1458
1459 if (pm->show_pfn) {
1460 offset = swp_offset(entry) +
1461 ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1462 frame = swp_type(entry) |
1463 (offset << MAX_SWAPFILES_SHIFT);
1464 }
1465 flags |= PM_SWAP;
1466 if (pmd_swp_soft_dirty(pmd))
1467 flags |= PM_SOFT_DIRTY;
1468 VM_BUG_ON(!is_pmd_migration_entry(pmd));
1469 page = migration_entry_to_page(entry);
1470 }
1471 #endif
1472
1473 if (page && page_mapcount(page) == 1)
1474 flags |= PM_MMAP_EXCLUSIVE;
1475
1476 for (; addr != end; addr += PAGE_SIZE) {
1477 pagemap_entry_t pme = make_pme(frame, flags);
1478
1479 err = add_to_pagemap(addr, &pme, pm);
1480 if (err)
1481 break;
1482 if (pm->show_pfn) {
1483 if (flags & PM_PRESENT)
1484 frame++;
1485 else if (flags & PM_SWAP)
1486 frame += (1 << MAX_SWAPFILES_SHIFT);
1487 }
1488 }
1489 spin_unlock(ptl);
1490 return err;
1491 }
1492
1493 if (pmd_trans_unstable(pmdp))
1494 return 0;
1495 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1496
1497 /*
1498 * We can assume that @vma always points to a valid one and @end never
1499 * goes beyond vma->vm_end.
1500 */
1501 orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1502 for (; addr < end; pte++, addr += PAGE_SIZE) {
1503 pagemap_entry_t pme;
1504
1505 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1506 err = add_to_pagemap(addr, &pme, pm);
1507 if (err)
1508 break;
1509 }
1510 pte_unmap_unlock(orig_pte, ptl);
1511
1512 cond_resched();
1513
1514 return err;
1515 }
1516
1517 #ifdef CONFIG_HUGETLB_PAGE
1518 /* This function walks within one hugetlb entry in the single call */
pagemap_hugetlb_range(pte_t * ptep,unsigned long hmask,unsigned long addr,unsigned long end,struct mm_walk * walk)1519 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1520 unsigned long addr, unsigned long end,
1521 struct mm_walk *walk)
1522 {
1523 struct pagemapread *pm = walk->private;
1524 struct vm_area_struct *vma = walk->vma;
1525 u64 flags = 0, frame = 0;
1526 int err = 0;
1527 pte_t pte;
1528
1529 if (vma->vm_flags & VM_SOFTDIRTY)
1530 flags |= PM_SOFT_DIRTY;
1531
1532 pte = huge_ptep_get(ptep);
1533 if (pte_present(pte)) {
1534 struct page *page = pte_page(pte);
1535
1536 if (!PageAnon(page))
1537 flags |= PM_FILE;
1538
1539 if (page_mapcount(page) == 1)
1540 flags |= PM_MMAP_EXCLUSIVE;
1541
1542 flags |= PM_PRESENT;
1543 if (pm->show_pfn)
1544 frame = pte_pfn(pte) +
1545 ((addr & ~hmask) >> PAGE_SHIFT);
1546 }
1547
1548 for (; addr != end; addr += PAGE_SIZE) {
1549 pagemap_entry_t pme = make_pme(frame, flags);
1550
1551 err = add_to_pagemap(addr, &pme, pm);
1552 if (err)
1553 return err;
1554 if (pm->show_pfn && (flags & PM_PRESENT))
1555 frame++;
1556 }
1557
1558 cond_resched();
1559
1560 return err;
1561 }
1562 #else
1563 #define pagemap_hugetlb_range NULL
1564 #endif /* HUGETLB_PAGE */
1565
1566 static const struct mm_walk_ops pagemap_ops = {
1567 .pmd_entry = pagemap_pmd_range,
1568 .pte_hole = pagemap_pte_hole,
1569 .hugetlb_entry = pagemap_hugetlb_range,
1570 };
1571
1572 /*
1573 * /proc/pid/pagemap - an array mapping virtual pages to pfns
1574 *
1575 * For each page in the address space, this file contains one 64-bit entry
1576 * consisting of the following:
1577 *
1578 * Bits 0-54 page frame number (PFN) if present
1579 * Bits 0-4 swap type if swapped
1580 * Bits 5-54 swap offset if swapped
1581 * Bit 55 pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst)
1582 * Bit 56 page exclusively mapped
1583 * Bits 57-60 zero
1584 * Bit 61 page is file-page or shared-anon
1585 * Bit 62 page swapped
1586 * Bit 63 page present
1587 *
1588 * If the page is not present but in swap, then the PFN contains an
1589 * encoding of the swap file number and the page's offset into the
1590 * swap. Unmapped pages return a null PFN. This allows determining
1591 * precisely which pages are mapped (or in swap) and comparing mapped
1592 * pages between processes.
1593 *
1594 * Efficient users of this interface will use /proc/pid/maps to
1595 * determine which areas of memory are actually mapped and llseek to
1596 * skip over unmapped regions.
1597 */
pagemap_read(struct file * file,char __user * buf,size_t count,loff_t * ppos)1598 static ssize_t pagemap_read(struct file *file, char __user *buf,
1599 size_t count, loff_t *ppos)
1600 {
1601 struct mm_struct *mm = file->private_data;
1602 struct pagemapread pm;
1603 unsigned long src;
1604 unsigned long svpfn;
1605 unsigned long start_vaddr;
1606 unsigned long end_vaddr;
1607 int ret = 0, copied = 0;
1608
1609 if (!mm || !mmget_not_zero(mm))
1610 goto out;
1611
1612 ret = -EINVAL;
1613 /* file position must be aligned */
1614 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1615 goto out_mm;
1616
1617 ret = 0;
1618 if (!count)
1619 goto out_mm;
1620
1621 /* do not disclose physical addresses: attack vector */
1622 pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1623
1624 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1625 pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL);
1626 ret = -ENOMEM;
1627 if (!pm.buffer)
1628 goto out_mm;
1629
1630 src = *ppos;
1631 svpfn = src / PM_ENTRY_BYTES;
1632 start_vaddr = svpfn << PAGE_SHIFT;
1633 end_vaddr = mm->task_size;
1634
1635 /* watch out for wraparound */
1636 if (svpfn > mm->task_size >> PAGE_SHIFT)
1637 start_vaddr = end_vaddr;
1638
1639 /*
1640 * The odds are that this will stop walking way
1641 * before end_vaddr, because the length of the
1642 * user buffer is tracked in "pm", and the walk
1643 * will stop when we hit the end of the buffer.
1644 */
1645 ret = 0;
1646 while (count && (start_vaddr < end_vaddr)) {
1647 int len;
1648 unsigned long end;
1649
1650 pm.pos = 0;
1651 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1652 /* overflow ? */
1653 if (end < start_vaddr || end > end_vaddr)
1654 end = end_vaddr;
1655 ret = down_read_killable(&mm->mmap_sem);
1656 if (ret)
1657 goto out_free;
1658 ret = walk_page_range(mm, start_vaddr, end, &pagemap_ops, &pm);
1659 up_read(&mm->mmap_sem);
1660 start_vaddr = end;
1661
1662 len = min(count, PM_ENTRY_BYTES * pm.pos);
1663 if (copy_to_user(buf, pm.buffer, len)) {
1664 ret = -EFAULT;
1665 goto out_free;
1666 }
1667 copied += len;
1668 buf += len;
1669 count -= len;
1670 }
1671 *ppos += copied;
1672 if (!ret || ret == PM_END_OF_BUFFER)
1673 ret = copied;
1674
1675 out_free:
1676 kfree(pm.buffer);
1677 out_mm:
1678 mmput(mm);
1679 out:
1680 return ret;
1681 }
1682
pagemap_open(struct inode * inode,struct file * file)1683 static int pagemap_open(struct inode *inode, struct file *file)
1684 {
1685 struct mm_struct *mm;
1686
1687 mm = proc_mem_open(inode, PTRACE_MODE_READ);
1688 if (IS_ERR(mm))
1689 return PTR_ERR(mm);
1690 file->private_data = mm;
1691 return 0;
1692 }
1693
pagemap_release(struct inode * inode,struct file * file)1694 static int pagemap_release(struct inode *inode, struct file *file)
1695 {
1696 struct mm_struct *mm = file->private_data;
1697
1698 if (mm)
1699 mmdrop(mm);
1700 return 0;
1701 }
1702
1703 const struct file_operations proc_pagemap_operations = {
1704 .llseek = mem_lseek, /* borrow this */
1705 .read = pagemap_read,
1706 .open = pagemap_open,
1707 .release = pagemap_release,
1708 };
1709 #endif /* CONFIG_PROC_PAGE_MONITOR */
1710
1711 #ifdef CONFIG_NUMA
1712
1713 struct numa_maps {
1714 unsigned long pages;
1715 unsigned long anon;
1716 unsigned long active;
1717 unsigned long writeback;
1718 unsigned long mapcount_max;
1719 unsigned long dirty;
1720 unsigned long swapcache;
1721 unsigned long node[MAX_NUMNODES];
1722 };
1723
1724 struct numa_maps_private {
1725 struct proc_maps_private proc_maps;
1726 struct numa_maps md;
1727 };
1728
gather_stats(struct page * page,struct numa_maps * md,int pte_dirty,unsigned long nr_pages)1729 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1730 unsigned long nr_pages)
1731 {
1732 int count = page_mapcount(page);
1733
1734 md->pages += nr_pages;
1735 if (pte_dirty || PageDirty(page))
1736 md->dirty += nr_pages;
1737
1738 if (PageSwapCache(page))
1739 md->swapcache += nr_pages;
1740
1741 if (PageActive(page) || PageUnevictable(page))
1742 md->active += nr_pages;
1743
1744 if (PageWriteback(page))
1745 md->writeback += nr_pages;
1746
1747 if (PageAnon(page))
1748 md->anon += nr_pages;
1749
1750 if (count > md->mapcount_max)
1751 md->mapcount_max = count;
1752
1753 md->node[page_to_nid(page)] += nr_pages;
1754 }
1755
can_gather_numa_stats(pte_t pte,struct vm_area_struct * vma,unsigned long addr)1756 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1757 unsigned long addr)
1758 {
1759 struct page *page;
1760 int nid;
1761
1762 if (!pte_present(pte))
1763 return NULL;
1764
1765 page = vm_normal_page(vma, addr, pte);
1766 if (!page)
1767 return NULL;
1768
1769 if (PageReserved(page))
1770 return NULL;
1771
1772 nid = page_to_nid(page);
1773 if (!node_isset(nid, node_states[N_MEMORY]))
1774 return NULL;
1775
1776 return page;
1777 }
1778
1779 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
can_gather_numa_stats_pmd(pmd_t pmd,struct vm_area_struct * vma,unsigned long addr)1780 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1781 struct vm_area_struct *vma,
1782 unsigned long addr)
1783 {
1784 struct page *page;
1785 int nid;
1786
1787 if (!pmd_present(pmd))
1788 return NULL;
1789
1790 page = vm_normal_page_pmd(vma, addr, pmd);
1791 if (!page)
1792 return NULL;
1793
1794 if (PageReserved(page))
1795 return NULL;
1796
1797 nid = page_to_nid(page);
1798 if (!node_isset(nid, node_states[N_MEMORY]))
1799 return NULL;
1800
1801 return page;
1802 }
1803 #endif
1804
gather_pte_stats(pmd_t * pmd,unsigned long addr,unsigned long end,struct mm_walk * walk)1805 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1806 unsigned long end, struct mm_walk *walk)
1807 {
1808 struct numa_maps *md = walk->private;
1809 struct vm_area_struct *vma = walk->vma;
1810 spinlock_t *ptl;
1811 pte_t *orig_pte;
1812 pte_t *pte;
1813
1814 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1815 ptl = pmd_trans_huge_lock(pmd, vma);
1816 if (ptl) {
1817 struct page *page;
1818
1819 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1820 if (page)
1821 gather_stats(page, md, pmd_dirty(*pmd),
1822 HPAGE_PMD_SIZE/PAGE_SIZE);
1823 spin_unlock(ptl);
1824 return 0;
1825 }
1826
1827 if (pmd_trans_unstable(pmd))
1828 return 0;
1829 #endif
1830 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1831 do {
1832 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1833 if (!page)
1834 continue;
1835 gather_stats(page, md, pte_dirty(*pte), 1);
1836
1837 } while (pte++, addr += PAGE_SIZE, addr != end);
1838 pte_unmap_unlock(orig_pte, ptl);
1839 cond_resched();
1840 return 0;
1841 }
1842 #ifdef CONFIG_HUGETLB_PAGE
gather_hugetlb_stats(pte_t * pte,unsigned long hmask,unsigned long addr,unsigned long end,struct mm_walk * walk)1843 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1844 unsigned long addr, unsigned long end, struct mm_walk *walk)
1845 {
1846 pte_t huge_pte = huge_ptep_get(pte);
1847 struct numa_maps *md;
1848 struct page *page;
1849
1850 if (!pte_present(huge_pte))
1851 return 0;
1852
1853 page = pte_page(huge_pte);
1854 if (!page)
1855 return 0;
1856
1857 md = walk->private;
1858 gather_stats(page, md, pte_dirty(huge_pte), 1);
1859 return 0;
1860 }
1861
1862 #else
gather_hugetlb_stats(pte_t * pte,unsigned long hmask,unsigned long addr,unsigned long end,struct mm_walk * walk)1863 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1864 unsigned long addr, unsigned long end, struct mm_walk *walk)
1865 {
1866 return 0;
1867 }
1868 #endif
1869
1870 static const struct mm_walk_ops show_numa_ops = {
1871 .hugetlb_entry = gather_hugetlb_stats,
1872 .pmd_entry = gather_pte_stats,
1873 };
1874
1875 /*
1876 * Display pages allocated per node and memory policy via /proc.
1877 */
show_numa_map(struct seq_file * m,void * v)1878 static int show_numa_map(struct seq_file *m, void *v)
1879 {
1880 struct numa_maps_private *numa_priv = m->private;
1881 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1882 struct vm_area_struct *vma = v;
1883 struct numa_maps *md = &numa_priv->md;
1884 struct file *file = vma->vm_file;
1885 struct mm_struct *mm = vma->vm_mm;
1886 struct mempolicy *pol;
1887 char buffer[64];
1888 int nid;
1889
1890 if (!mm)
1891 return 0;
1892
1893 /* Ensure we start with an empty set of numa_maps statistics. */
1894 memset(md, 0, sizeof(*md));
1895
1896 pol = __get_vma_policy(vma, vma->vm_start);
1897 if (pol) {
1898 mpol_to_str(buffer, sizeof(buffer), pol);
1899 mpol_cond_put(pol);
1900 } else {
1901 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1902 }
1903
1904 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1905
1906 if (file) {
1907 seq_puts(m, " file=");
1908 seq_file_path(m, file, "\n\t= ");
1909 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1910 seq_puts(m, " heap");
1911 } else if (is_stack(vma)) {
1912 seq_puts(m, " stack");
1913 }
1914
1915 if (is_vm_hugetlb_page(vma))
1916 seq_puts(m, " huge");
1917
1918 /* mmap_sem is held by m_start */
1919 walk_page_vma(vma, &show_numa_ops, md);
1920
1921 if (!md->pages)
1922 goto out;
1923
1924 if (md->anon)
1925 seq_printf(m, " anon=%lu", md->anon);
1926
1927 if (md->dirty)
1928 seq_printf(m, " dirty=%lu", md->dirty);
1929
1930 if (md->pages != md->anon && md->pages != md->dirty)
1931 seq_printf(m, " mapped=%lu", md->pages);
1932
1933 if (md->mapcount_max > 1)
1934 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1935
1936 if (md->swapcache)
1937 seq_printf(m, " swapcache=%lu", md->swapcache);
1938
1939 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1940 seq_printf(m, " active=%lu", md->active);
1941
1942 if (md->writeback)
1943 seq_printf(m, " writeback=%lu", md->writeback);
1944
1945 for_each_node_state(nid, N_MEMORY)
1946 if (md->node[nid])
1947 seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1948
1949 seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1950 out:
1951 seq_putc(m, '\n');
1952 m_cache_vma(m, vma);
1953 return 0;
1954 }
1955
1956 static const struct seq_operations proc_pid_numa_maps_op = {
1957 .start = m_start,
1958 .next = m_next,
1959 .stop = m_stop,
1960 .show = show_numa_map,
1961 };
1962
pid_numa_maps_open(struct inode * inode,struct file * file)1963 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1964 {
1965 return proc_maps_open(inode, file, &proc_pid_numa_maps_op,
1966 sizeof(struct numa_maps_private));
1967 }
1968
1969 const struct file_operations proc_pid_numa_maps_operations = {
1970 .open = pid_numa_maps_open,
1971 .read = seq_read,
1972 .llseek = seq_lseek,
1973 .release = proc_map_release,
1974 };
1975
1976 #endif /* CONFIG_NUMA */
1977