Loading...
1#include <linux/mm.h>
2#include <linux/vmacache.h>
3#include <linux/hugetlb.h>
4#include <linux/huge_mm.h>
5#include <linux/mount.h>
6#include <linux/seq_file.h>
7#include <linux/highmem.h>
8#include <linux/ptrace.h>
9#include <linux/slab.h>
10#include <linux/pagemap.h>
11#include <linux/mempolicy.h>
12#include <linux/rmap.h>
13#include <linux/swap.h>
14#include <linux/swapops.h>
15#include <linux/mmu_notifier.h>
16#include <linux/page_idle.h>
17#include <linux/shmem_fs.h>
18
19#include <asm/elf.h>
20#include <asm/uaccess.h>
21#include <asm/tlbflush.h>
22#include "internal.h"
23
24void task_mem(struct seq_file *m, struct mm_struct *mm)
25{
26 unsigned long text, lib, swap, ptes, pmds, anon, file, shmem;
27 unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
28
29 anon = get_mm_counter(mm, MM_ANONPAGES);
30 file = get_mm_counter(mm, MM_FILEPAGES);
31 shmem = get_mm_counter(mm, MM_SHMEMPAGES);
32
33 /*
34 * Note: to minimize their overhead, mm maintains hiwater_vm and
35 * hiwater_rss only when about to *lower* total_vm or rss. Any
36 * collector of these hiwater stats must therefore get total_vm
37 * and rss too, which will usually be the higher. Barriers? not
38 * worth the effort, such snapshots can always be inconsistent.
39 */
40 hiwater_vm = total_vm = mm->total_vm;
41 if (hiwater_vm < mm->hiwater_vm)
42 hiwater_vm = mm->hiwater_vm;
43 hiwater_rss = total_rss = anon + file + shmem;
44 if (hiwater_rss < mm->hiwater_rss)
45 hiwater_rss = mm->hiwater_rss;
46
47 text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
48 lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
49 swap = get_mm_counter(mm, MM_SWAPENTS);
50 ptes = PTRS_PER_PTE * sizeof(pte_t) * atomic_long_read(&mm->nr_ptes);
51 pmds = PTRS_PER_PMD * sizeof(pmd_t) * mm_nr_pmds(mm);
52 seq_printf(m,
53 "VmPeak:\t%8lu kB\n"
54 "VmSize:\t%8lu kB\n"
55 "VmLck:\t%8lu kB\n"
56 "VmPin:\t%8lu kB\n"
57 "VmHWM:\t%8lu kB\n"
58 "VmRSS:\t%8lu kB\n"
59 "RssAnon:\t%8lu kB\n"
60 "RssFile:\t%8lu kB\n"
61 "RssShmem:\t%8lu kB\n"
62 "VmData:\t%8lu kB\n"
63 "VmStk:\t%8lu kB\n"
64 "VmExe:\t%8lu kB\n"
65 "VmLib:\t%8lu kB\n"
66 "VmPTE:\t%8lu kB\n"
67 "VmPMD:\t%8lu kB\n"
68 "VmSwap:\t%8lu kB\n",
69 hiwater_vm << (PAGE_SHIFT-10),
70 total_vm << (PAGE_SHIFT-10),
71 mm->locked_vm << (PAGE_SHIFT-10),
72 mm->pinned_vm << (PAGE_SHIFT-10),
73 hiwater_rss << (PAGE_SHIFT-10),
74 total_rss << (PAGE_SHIFT-10),
75 anon << (PAGE_SHIFT-10),
76 file << (PAGE_SHIFT-10),
77 shmem << (PAGE_SHIFT-10),
78 mm->data_vm << (PAGE_SHIFT-10),
79 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
80 ptes >> 10,
81 pmds >> 10,
82 swap << (PAGE_SHIFT-10));
83 hugetlb_report_usage(m, mm);
84}
85
86unsigned long task_vsize(struct mm_struct *mm)
87{
88 return PAGE_SIZE * mm->total_vm;
89}
90
91unsigned long task_statm(struct mm_struct *mm,
92 unsigned long *shared, unsigned long *text,
93 unsigned long *data, unsigned long *resident)
94{
95 *shared = get_mm_counter(mm, MM_FILEPAGES) +
96 get_mm_counter(mm, MM_SHMEMPAGES);
97 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
98 >> PAGE_SHIFT;
99 *data = mm->data_vm + mm->stack_vm;
100 *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
101 return mm->total_vm;
102}
103
104#ifdef CONFIG_NUMA
105/*
106 * Save get_task_policy() for show_numa_map().
107 */
108static void hold_task_mempolicy(struct proc_maps_private *priv)
109{
110 struct task_struct *task = priv->task;
111
112 task_lock(task);
113 priv->task_mempolicy = get_task_policy(task);
114 mpol_get(priv->task_mempolicy);
115 task_unlock(task);
116}
117static void release_task_mempolicy(struct proc_maps_private *priv)
118{
119 mpol_put(priv->task_mempolicy);
120}
121#else
122static void hold_task_mempolicy(struct proc_maps_private *priv)
123{
124}
125static void release_task_mempolicy(struct proc_maps_private *priv)
126{
127}
128#endif
129
130static void vma_stop(struct proc_maps_private *priv)
131{
132 struct mm_struct *mm = priv->mm;
133
134 release_task_mempolicy(priv);
135 up_read(&mm->mmap_sem);
136 mmput(mm);
137}
138
139static struct vm_area_struct *
140m_next_vma(struct proc_maps_private *priv, struct vm_area_struct *vma)
141{
142 if (vma == priv->tail_vma)
143 return NULL;
144 return vma->vm_next ?: priv->tail_vma;
145}
146
147static void m_cache_vma(struct seq_file *m, struct vm_area_struct *vma)
148{
149 if (m->count < m->size) /* vma is copied successfully */
150 m->version = m_next_vma(m->private, vma) ? vma->vm_start : -1UL;
151}
152
153static void *m_start(struct seq_file *m, loff_t *ppos)
154{
155 struct proc_maps_private *priv = m->private;
156 unsigned long last_addr = m->version;
157 struct mm_struct *mm;
158 struct vm_area_struct *vma;
159 unsigned int pos = *ppos;
160
161 /* See m_cache_vma(). Zero at the start or after lseek. */
162 if (last_addr == -1UL)
163 return NULL;
164
165 priv->task = get_proc_task(priv->inode);
166 if (!priv->task)
167 return ERR_PTR(-ESRCH);
168
169 mm = priv->mm;
170 if (!mm || !atomic_inc_not_zero(&mm->mm_users))
171 return NULL;
172
173 down_read(&mm->mmap_sem);
174 hold_task_mempolicy(priv);
175 priv->tail_vma = get_gate_vma(mm);
176
177 if (last_addr) {
178 vma = find_vma(mm, last_addr);
179 if (vma && (vma = m_next_vma(priv, vma)))
180 return vma;
181 }
182
183 m->version = 0;
184 if (pos < mm->map_count) {
185 for (vma = mm->mmap; pos; pos--) {
186 m->version = vma->vm_start;
187 vma = vma->vm_next;
188 }
189 return vma;
190 }
191
192 /* we do not bother to update m->version in this case */
193 if (pos == mm->map_count && priv->tail_vma)
194 return priv->tail_vma;
195
196 vma_stop(priv);
197 return NULL;
198}
199
200static void *m_next(struct seq_file *m, void *v, loff_t *pos)
201{
202 struct proc_maps_private *priv = m->private;
203 struct vm_area_struct *next;
204
205 (*pos)++;
206 next = m_next_vma(priv, v);
207 if (!next)
208 vma_stop(priv);
209 return next;
210}
211
212static void m_stop(struct seq_file *m, void *v)
213{
214 struct proc_maps_private *priv = m->private;
215
216 if (!IS_ERR_OR_NULL(v))
217 vma_stop(priv);
218 if (priv->task) {
219 put_task_struct(priv->task);
220 priv->task = NULL;
221 }
222}
223
224static int proc_maps_open(struct inode *inode, struct file *file,
225 const struct seq_operations *ops, int psize)
226{
227 struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
228
229 if (!priv)
230 return -ENOMEM;
231
232 priv->inode = inode;
233 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
234 if (IS_ERR(priv->mm)) {
235 int err = PTR_ERR(priv->mm);
236
237 seq_release_private(inode, file);
238 return err;
239 }
240
241 return 0;
242}
243
244static int proc_map_release(struct inode *inode, struct file *file)
245{
246 struct seq_file *seq = file->private_data;
247 struct proc_maps_private *priv = seq->private;
248
249 if (priv->mm)
250 mmdrop(priv->mm);
251
252 return seq_release_private(inode, file);
253}
254
255static int do_maps_open(struct inode *inode, struct file *file,
256 const struct seq_operations *ops)
257{
258 return proc_maps_open(inode, file, ops,
259 sizeof(struct proc_maps_private));
260}
261
262/*
263 * Indicate if the VMA is a stack for the given task; for
264 * /proc/PID/maps that is the stack of the main task.
265 */
266static int is_stack(struct proc_maps_private *priv,
267 struct vm_area_struct *vma, int is_pid)
268{
269 int stack = 0;
270
271 if (is_pid) {
272 stack = vma->vm_start <= vma->vm_mm->start_stack &&
273 vma->vm_end >= vma->vm_mm->start_stack;
274 } else {
275 struct inode *inode = priv->inode;
276 struct task_struct *task;
277
278 rcu_read_lock();
279 task = pid_task(proc_pid(inode), PIDTYPE_PID);
280 if (task)
281 stack = vma_is_stack_for_task(vma, task);
282 rcu_read_unlock();
283 }
284 return stack;
285}
286
287static void
288show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
289{
290 struct mm_struct *mm = vma->vm_mm;
291 struct file *file = vma->vm_file;
292 struct proc_maps_private *priv = m->private;
293 vm_flags_t flags = vma->vm_flags;
294 unsigned long ino = 0;
295 unsigned long long pgoff = 0;
296 unsigned long start, end;
297 dev_t dev = 0;
298 const char *name = NULL;
299
300 if (file) {
301 struct inode *inode = file_inode(vma->vm_file);
302 dev = inode->i_sb->s_dev;
303 ino = inode->i_ino;
304 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
305 }
306
307 /* We don't show the stack guard page in /proc/maps */
308 start = vma->vm_start;
309 if (stack_guard_page_start(vma, start))
310 start += PAGE_SIZE;
311 end = vma->vm_end;
312 if (stack_guard_page_end(vma, end))
313 end -= PAGE_SIZE;
314
315 seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
316 seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu ",
317 start,
318 end,
319 flags & VM_READ ? 'r' : '-',
320 flags & VM_WRITE ? 'w' : '-',
321 flags & VM_EXEC ? 'x' : '-',
322 flags & VM_MAYSHARE ? 's' : 'p',
323 pgoff,
324 MAJOR(dev), MINOR(dev), ino);
325
326 /*
327 * Print the dentry name for named mappings, and a
328 * special [heap] marker for the heap:
329 */
330 if (file) {
331 seq_pad(m, ' ');
332 seq_file_path(m, file, "\n");
333 goto done;
334 }
335
336 if (vma->vm_ops && vma->vm_ops->name) {
337 name = vma->vm_ops->name(vma);
338 if (name)
339 goto done;
340 }
341
342 name = arch_vma_name(vma);
343 if (!name) {
344 if (!mm) {
345 name = "[vdso]";
346 goto done;
347 }
348
349 if (vma->vm_start <= mm->brk &&
350 vma->vm_end >= mm->start_brk) {
351 name = "[heap]";
352 goto done;
353 }
354
355 if (is_stack(priv, vma, is_pid))
356 name = "[stack]";
357 }
358
359done:
360 if (name) {
361 seq_pad(m, ' ');
362 seq_puts(m, name);
363 }
364 seq_putc(m, '\n');
365}
366
367static int show_map(struct seq_file *m, void *v, int is_pid)
368{
369 show_map_vma(m, v, is_pid);
370 m_cache_vma(m, v);
371 return 0;
372}
373
374static int show_pid_map(struct seq_file *m, void *v)
375{
376 return show_map(m, v, 1);
377}
378
379static int show_tid_map(struct seq_file *m, void *v)
380{
381 return show_map(m, v, 0);
382}
383
384static const struct seq_operations proc_pid_maps_op = {
385 .start = m_start,
386 .next = m_next,
387 .stop = m_stop,
388 .show = show_pid_map
389};
390
391static const struct seq_operations proc_tid_maps_op = {
392 .start = m_start,
393 .next = m_next,
394 .stop = m_stop,
395 .show = show_tid_map
396};
397
398static int pid_maps_open(struct inode *inode, struct file *file)
399{
400 return do_maps_open(inode, file, &proc_pid_maps_op);
401}
402
403static int tid_maps_open(struct inode *inode, struct file *file)
404{
405 return do_maps_open(inode, file, &proc_tid_maps_op);
406}
407
408const struct file_operations proc_pid_maps_operations = {
409 .open = pid_maps_open,
410 .read = seq_read,
411 .llseek = seq_lseek,
412 .release = proc_map_release,
413};
414
415const struct file_operations proc_tid_maps_operations = {
416 .open = tid_maps_open,
417 .read = seq_read,
418 .llseek = seq_lseek,
419 .release = proc_map_release,
420};
421
422/*
423 * Proportional Set Size(PSS): my share of RSS.
424 *
425 * PSS of a process is the count of pages it has in memory, where each
426 * page is divided by the number of processes sharing it. So if a
427 * process has 1000 pages all to itself, and 1000 shared with one other
428 * process, its PSS will be 1500.
429 *
430 * To keep (accumulated) division errors low, we adopt a 64bit
431 * fixed-point pss counter to minimize division errors. So (pss >>
432 * PSS_SHIFT) would be the real byte count.
433 *
434 * A shift of 12 before division means (assuming 4K page size):
435 * - 1M 3-user-pages add up to 8KB errors;
436 * - supports mapcount up to 2^24, or 16M;
437 * - supports PSS up to 2^52 bytes, or 4PB.
438 */
439#define PSS_SHIFT 12
440
441#ifdef CONFIG_PROC_PAGE_MONITOR
442struct mem_size_stats {
443 unsigned long resident;
444 unsigned long shared_clean;
445 unsigned long shared_dirty;
446 unsigned long private_clean;
447 unsigned long private_dirty;
448 unsigned long referenced;
449 unsigned long anonymous;
450 unsigned long anonymous_thp;
451 unsigned long swap;
452 unsigned long shared_hugetlb;
453 unsigned long private_hugetlb;
454 u64 pss;
455 u64 swap_pss;
456 bool check_shmem_swap;
457};
458
459static void smaps_account(struct mem_size_stats *mss, struct page *page,
460 bool compound, bool young, bool dirty)
461{
462 int i, nr = compound ? 1 << compound_order(page) : 1;
463 unsigned long size = nr * PAGE_SIZE;
464
465 if (PageAnon(page))
466 mss->anonymous += size;
467
468 mss->resident += size;
469 /* Accumulate the size in pages that have been accessed. */
470 if (young || page_is_young(page) || PageReferenced(page))
471 mss->referenced += size;
472
473 /*
474 * page_count(page) == 1 guarantees the page is mapped exactly once.
475 * If any subpage of the compound page mapped with PTE it would elevate
476 * page_count().
477 */
478 if (page_count(page) == 1) {
479 if (dirty || PageDirty(page))
480 mss->private_dirty += size;
481 else
482 mss->private_clean += size;
483 mss->pss += (u64)size << PSS_SHIFT;
484 return;
485 }
486
487 for (i = 0; i < nr; i++, page++) {
488 int mapcount = page_mapcount(page);
489
490 if (mapcount >= 2) {
491 if (dirty || PageDirty(page))
492 mss->shared_dirty += PAGE_SIZE;
493 else
494 mss->shared_clean += PAGE_SIZE;
495 mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount;
496 } else {
497 if (dirty || PageDirty(page))
498 mss->private_dirty += PAGE_SIZE;
499 else
500 mss->private_clean += PAGE_SIZE;
501 mss->pss += PAGE_SIZE << PSS_SHIFT;
502 }
503 }
504}
505
506#ifdef CONFIG_SHMEM
507static int smaps_pte_hole(unsigned long addr, unsigned long end,
508 struct mm_walk *walk)
509{
510 struct mem_size_stats *mss = walk->private;
511
512 mss->swap += shmem_partial_swap_usage(
513 walk->vma->vm_file->f_mapping, addr, end);
514
515 return 0;
516}
517#endif
518
519static void smaps_pte_entry(pte_t *pte, unsigned long addr,
520 struct mm_walk *walk)
521{
522 struct mem_size_stats *mss = walk->private;
523 struct vm_area_struct *vma = walk->vma;
524 struct page *page = NULL;
525
526 if (pte_present(*pte)) {
527 page = vm_normal_page(vma, addr, *pte);
528 } else if (is_swap_pte(*pte)) {
529 swp_entry_t swpent = pte_to_swp_entry(*pte);
530
531 if (!non_swap_entry(swpent)) {
532 int mapcount;
533
534 mss->swap += PAGE_SIZE;
535 mapcount = swp_swapcount(swpent);
536 if (mapcount >= 2) {
537 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
538
539 do_div(pss_delta, mapcount);
540 mss->swap_pss += pss_delta;
541 } else {
542 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
543 }
544 } else if (is_migration_entry(swpent))
545 page = migration_entry_to_page(swpent);
546 } else if (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap
547 && pte_none(*pte))) {
548 page = find_get_entry(vma->vm_file->f_mapping,
549 linear_page_index(vma, addr));
550 if (!page)
551 return;
552
553 if (radix_tree_exceptional_entry(page))
554 mss->swap += PAGE_SIZE;
555 else
556 put_page(page);
557
558 return;
559 }
560
561 if (!page)
562 return;
563
564 smaps_account(mss, page, false, pte_young(*pte), pte_dirty(*pte));
565}
566
567#ifdef CONFIG_TRANSPARENT_HUGEPAGE
568static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
569 struct mm_walk *walk)
570{
571 struct mem_size_stats *mss = walk->private;
572 struct vm_area_struct *vma = walk->vma;
573 struct page *page;
574
575 /* FOLL_DUMP will return -EFAULT on huge zero page */
576 page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
577 if (IS_ERR_OR_NULL(page))
578 return;
579 mss->anonymous_thp += HPAGE_PMD_SIZE;
580 smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd));
581}
582#else
583static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
584 struct mm_walk *walk)
585{
586}
587#endif
588
589static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
590 struct mm_walk *walk)
591{
592 struct vm_area_struct *vma = walk->vma;
593 pte_t *pte;
594 spinlock_t *ptl;
595
596 ptl = pmd_trans_huge_lock(pmd, vma);
597 if (ptl) {
598 smaps_pmd_entry(pmd, addr, walk);
599 spin_unlock(ptl);
600 return 0;
601 }
602
603 if (pmd_trans_unstable(pmd))
604 return 0;
605 /*
606 * The mmap_sem held all the way back in m_start() is what
607 * keeps khugepaged out of here and from collapsing things
608 * in here.
609 */
610 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
611 for (; addr != end; pte++, addr += PAGE_SIZE)
612 smaps_pte_entry(pte, addr, walk);
613 pte_unmap_unlock(pte - 1, ptl);
614 cond_resched();
615 return 0;
616}
617
618static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
619{
620 /*
621 * Don't forget to update Documentation/ on changes.
622 */
623 static const char mnemonics[BITS_PER_LONG][2] = {
624 /*
625 * In case if we meet a flag we don't know about.
626 */
627 [0 ... (BITS_PER_LONG-1)] = "??",
628
629 [ilog2(VM_READ)] = "rd",
630 [ilog2(VM_WRITE)] = "wr",
631 [ilog2(VM_EXEC)] = "ex",
632 [ilog2(VM_SHARED)] = "sh",
633 [ilog2(VM_MAYREAD)] = "mr",
634 [ilog2(VM_MAYWRITE)] = "mw",
635 [ilog2(VM_MAYEXEC)] = "me",
636 [ilog2(VM_MAYSHARE)] = "ms",
637 [ilog2(VM_GROWSDOWN)] = "gd",
638 [ilog2(VM_PFNMAP)] = "pf",
639 [ilog2(VM_DENYWRITE)] = "dw",
640#ifdef CONFIG_X86_INTEL_MPX
641 [ilog2(VM_MPX)] = "mp",
642#endif
643 [ilog2(VM_LOCKED)] = "lo",
644 [ilog2(VM_IO)] = "io",
645 [ilog2(VM_SEQ_READ)] = "sr",
646 [ilog2(VM_RAND_READ)] = "rr",
647 [ilog2(VM_DONTCOPY)] = "dc",
648 [ilog2(VM_DONTEXPAND)] = "de",
649 [ilog2(VM_ACCOUNT)] = "ac",
650 [ilog2(VM_NORESERVE)] = "nr",
651 [ilog2(VM_HUGETLB)] = "ht",
652 [ilog2(VM_ARCH_1)] = "ar",
653 [ilog2(VM_DONTDUMP)] = "dd",
654#ifdef CONFIG_MEM_SOFT_DIRTY
655 [ilog2(VM_SOFTDIRTY)] = "sd",
656#endif
657 [ilog2(VM_MIXEDMAP)] = "mm",
658 [ilog2(VM_HUGEPAGE)] = "hg",
659 [ilog2(VM_NOHUGEPAGE)] = "nh",
660 [ilog2(VM_MERGEABLE)] = "mg",
661 [ilog2(VM_UFFD_MISSING)]= "um",
662 [ilog2(VM_UFFD_WP)] = "uw",
663#ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
664 /* These come out via ProtectionKey: */
665 [ilog2(VM_PKEY_BIT0)] = "",
666 [ilog2(VM_PKEY_BIT1)] = "",
667 [ilog2(VM_PKEY_BIT2)] = "",
668 [ilog2(VM_PKEY_BIT3)] = "",
669#endif
670 };
671 size_t i;
672
673 seq_puts(m, "VmFlags: ");
674 for (i = 0; i < BITS_PER_LONG; i++) {
675 if (!mnemonics[i][0])
676 continue;
677 if (vma->vm_flags & (1UL << i)) {
678 seq_printf(m, "%c%c ",
679 mnemonics[i][0], mnemonics[i][1]);
680 }
681 }
682 seq_putc(m, '\n');
683}
684
685#ifdef CONFIG_HUGETLB_PAGE
686static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
687 unsigned long addr, unsigned long end,
688 struct mm_walk *walk)
689{
690 struct mem_size_stats *mss = walk->private;
691 struct vm_area_struct *vma = walk->vma;
692 struct page *page = NULL;
693
694 if (pte_present(*pte)) {
695 page = vm_normal_page(vma, addr, *pte);
696 } else if (is_swap_pte(*pte)) {
697 swp_entry_t swpent = pte_to_swp_entry(*pte);
698
699 if (is_migration_entry(swpent))
700 page = migration_entry_to_page(swpent);
701 }
702 if (page) {
703 int mapcount = page_mapcount(page);
704
705 if (mapcount >= 2)
706 mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
707 else
708 mss->private_hugetlb += huge_page_size(hstate_vma(vma));
709 }
710 return 0;
711}
712#endif /* HUGETLB_PAGE */
713
714void __weak arch_show_smap(struct seq_file *m, struct vm_area_struct *vma)
715{
716}
717
718static int show_smap(struct seq_file *m, void *v, int is_pid)
719{
720 struct vm_area_struct *vma = v;
721 struct mem_size_stats mss;
722 struct mm_walk smaps_walk = {
723 .pmd_entry = smaps_pte_range,
724#ifdef CONFIG_HUGETLB_PAGE
725 .hugetlb_entry = smaps_hugetlb_range,
726#endif
727 .mm = vma->vm_mm,
728 .private = &mss,
729 };
730
731 memset(&mss, 0, sizeof mss);
732
733#ifdef CONFIG_SHMEM
734 if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
735 /*
736 * For shared or readonly shmem mappings we know that all
737 * swapped out pages belong to the shmem object, and we can
738 * obtain the swap value much more efficiently. For private
739 * writable mappings, we might have COW pages that are
740 * not affected by the parent swapped out pages of the shmem
741 * object, so we have to distinguish them during the page walk.
742 * Unless we know that the shmem object (or the part mapped by
743 * our VMA) has no swapped out pages at all.
744 */
745 unsigned long shmem_swapped = shmem_swap_usage(vma);
746
747 if (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
748 !(vma->vm_flags & VM_WRITE)) {
749 mss.swap = shmem_swapped;
750 } else {
751 mss.check_shmem_swap = true;
752 smaps_walk.pte_hole = smaps_pte_hole;
753 }
754 }
755#endif
756
757 /* mmap_sem is held in m_start */
758 walk_page_vma(vma, &smaps_walk);
759
760 show_map_vma(m, vma, is_pid);
761
762 seq_printf(m,
763 "Size: %8lu kB\n"
764 "Rss: %8lu kB\n"
765 "Pss: %8lu kB\n"
766 "Shared_Clean: %8lu kB\n"
767 "Shared_Dirty: %8lu kB\n"
768 "Private_Clean: %8lu kB\n"
769 "Private_Dirty: %8lu kB\n"
770 "Referenced: %8lu kB\n"
771 "Anonymous: %8lu kB\n"
772 "AnonHugePages: %8lu kB\n"
773 "Shared_Hugetlb: %8lu kB\n"
774 "Private_Hugetlb: %7lu kB\n"
775 "Swap: %8lu kB\n"
776 "SwapPss: %8lu kB\n"
777 "KernelPageSize: %8lu kB\n"
778 "MMUPageSize: %8lu kB\n"
779 "Locked: %8lu kB\n",
780 (vma->vm_end - vma->vm_start) >> 10,
781 mss.resident >> 10,
782 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
783 mss.shared_clean >> 10,
784 mss.shared_dirty >> 10,
785 mss.private_clean >> 10,
786 mss.private_dirty >> 10,
787 mss.referenced >> 10,
788 mss.anonymous >> 10,
789 mss.anonymous_thp >> 10,
790 mss.shared_hugetlb >> 10,
791 mss.private_hugetlb >> 10,
792 mss.swap >> 10,
793 (unsigned long)(mss.swap_pss >> (10 + PSS_SHIFT)),
794 vma_kernel_pagesize(vma) >> 10,
795 vma_mmu_pagesize(vma) >> 10,
796 (vma->vm_flags & VM_LOCKED) ?
797 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
798
799 arch_show_smap(m, vma);
800 show_smap_vma_flags(m, vma);
801 m_cache_vma(m, vma);
802 return 0;
803}
804
805static int show_pid_smap(struct seq_file *m, void *v)
806{
807 return show_smap(m, v, 1);
808}
809
810static int show_tid_smap(struct seq_file *m, void *v)
811{
812 return show_smap(m, v, 0);
813}
814
815static const struct seq_operations proc_pid_smaps_op = {
816 .start = m_start,
817 .next = m_next,
818 .stop = m_stop,
819 .show = show_pid_smap
820};
821
822static const struct seq_operations proc_tid_smaps_op = {
823 .start = m_start,
824 .next = m_next,
825 .stop = m_stop,
826 .show = show_tid_smap
827};
828
829static int pid_smaps_open(struct inode *inode, struct file *file)
830{
831 return do_maps_open(inode, file, &proc_pid_smaps_op);
832}
833
834static int tid_smaps_open(struct inode *inode, struct file *file)
835{
836 return do_maps_open(inode, file, &proc_tid_smaps_op);
837}
838
839const struct file_operations proc_pid_smaps_operations = {
840 .open = pid_smaps_open,
841 .read = seq_read,
842 .llseek = seq_lseek,
843 .release = proc_map_release,
844};
845
846const struct file_operations proc_tid_smaps_operations = {
847 .open = tid_smaps_open,
848 .read = seq_read,
849 .llseek = seq_lseek,
850 .release = proc_map_release,
851};
852
853enum clear_refs_types {
854 CLEAR_REFS_ALL = 1,
855 CLEAR_REFS_ANON,
856 CLEAR_REFS_MAPPED,
857 CLEAR_REFS_SOFT_DIRTY,
858 CLEAR_REFS_MM_HIWATER_RSS,
859 CLEAR_REFS_LAST,
860};
861
862struct clear_refs_private {
863 enum clear_refs_types type;
864};
865
866#ifdef CONFIG_MEM_SOFT_DIRTY
867static inline void clear_soft_dirty(struct vm_area_struct *vma,
868 unsigned long addr, pte_t *pte)
869{
870 /*
871 * The soft-dirty tracker uses #PF-s to catch writes
872 * to pages, so write-protect the pte as well. See the
873 * Documentation/vm/soft-dirty.txt for full description
874 * of how soft-dirty works.
875 */
876 pte_t ptent = *pte;
877
878 if (pte_present(ptent)) {
879 ptent = ptep_modify_prot_start(vma->vm_mm, addr, pte);
880 ptent = pte_wrprotect(ptent);
881 ptent = pte_clear_soft_dirty(ptent);
882 ptep_modify_prot_commit(vma->vm_mm, addr, pte, ptent);
883 } else if (is_swap_pte(ptent)) {
884 ptent = pte_swp_clear_soft_dirty(ptent);
885 set_pte_at(vma->vm_mm, addr, pte, ptent);
886 }
887}
888#else
889static inline void clear_soft_dirty(struct vm_area_struct *vma,
890 unsigned long addr, pte_t *pte)
891{
892}
893#endif
894
895#if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
896static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
897 unsigned long addr, pmd_t *pmdp)
898{
899 pmd_t pmd = pmdp_huge_get_and_clear(vma->vm_mm, addr, pmdp);
900
901 pmd = pmd_wrprotect(pmd);
902 pmd = pmd_clear_soft_dirty(pmd);
903
904 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
905}
906#else
907static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
908 unsigned long addr, pmd_t *pmdp)
909{
910}
911#endif
912
913static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
914 unsigned long end, struct mm_walk *walk)
915{
916 struct clear_refs_private *cp = walk->private;
917 struct vm_area_struct *vma = walk->vma;
918 pte_t *pte, ptent;
919 spinlock_t *ptl;
920 struct page *page;
921
922 ptl = pmd_trans_huge_lock(pmd, vma);
923 if (ptl) {
924 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
925 clear_soft_dirty_pmd(vma, addr, pmd);
926 goto out;
927 }
928
929 page = pmd_page(*pmd);
930
931 /* Clear accessed and referenced bits. */
932 pmdp_test_and_clear_young(vma, addr, pmd);
933 test_and_clear_page_young(page);
934 ClearPageReferenced(page);
935out:
936 spin_unlock(ptl);
937 return 0;
938 }
939
940 if (pmd_trans_unstable(pmd))
941 return 0;
942
943 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
944 for (; addr != end; pte++, addr += PAGE_SIZE) {
945 ptent = *pte;
946
947 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
948 clear_soft_dirty(vma, addr, pte);
949 continue;
950 }
951
952 if (!pte_present(ptent))
953 continue;
954
955 page = vm_normal_page(vma, addr, ptent);
956 if (!page)
957 continue;
958
959 /* Clear accessed and referenced bits. */
960 ptep_test_and_clear_young(vma, addr, pte);
961 test_and_clear_page_young(page);
962 ClearPageReferenced(page);
963 }
964 pte_unmap_unlock(pte - 1, ptl);
965 cond_resched();
966 return 0;
967}
968
969static int clear_refs_test_walk(unsigned long start, unsigned long end,
970 struct mm_walk *walk)
971{
972 struct clear_refs_private *cp = walk->private;
973 struct vm_area_struct *vma = walk->vma;
974
975 if (vma->vm_flags & VM_PFNMAP)
976 return 1;
977
978 /*
979 * Writing 1 to /proc/pid/clear_refs affects all pages.
980 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
981 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
982 * Writing 4 to /proc/pid/clear_refs affects all pages.
983 */
984 if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
985 return 1;
986 if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
987 return 1;
988 return 0;
989}
990
991static ssize_t clear_refs_write(struct file *file, const char __user *buf,
992 size_t count, loff_t *ppos)
993{
994 struct task_struct *task;
995 char buffer[PROC_NUMBUF];
996 struct mm_struct *mm;
997 struct vm_area_struct *vma;
998 enum clear_refs_types type;
999 int itype;
1000 int rv;
1001
1002 memset(buffer, 0, sizeof(buffer));
1003 if (count > sizeof(buffer) - 1)
1004 count = sizeof(buffer) - 1;
1005 if (copy_from_user(buffer, buf, count))
1006 return -EFAULT;
1007 rv = kstrtoint(strstrip(buffer), 10, &itype);
1008 if (rv < 0)
1009 return rv;
1010 type = (enum clear_refs_types)itype;
1011 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1012 return -EINVAL;
1013
1014 task = get_proc_task(file_inode(file));
1015 if (!task)
1016 return -ESRCH;
1017 mm = get_task_mm(task);
1018 if (mm) {
1019 struct clear_refs_private cp = {
1020 .type = type,
1021 };
1022 struct mm_walk clear_refs_walk = {
1023 .pmd_entry = clear_refs_pte_range,
1024 .test_walk = clear_refs_test_walk,
1025 .mm = mm,
1026 .private = &cp,
1027 };
1028
1029 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1030 /*
1031 * Writing 5 to /proc/pid/clear_refs resets the peak
1032 * resident set size to this mm's current rss value.
1033 */
1034 down_write(&mm->mmap_sem);
1035 reset_mm_hiwater_rss(mm);
1036 up_write(&mm->mmap_sem);
1037 goto out_mm;
1038 }
1039
1040 down_read(&mm->mmap_sem);
1041 if (type == CLEAR_REFS_SOFT_DIRTY) {
1042 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1043 if (!(vma->vm_flags & VM_SOFTDIRTY))
1044 continue;
1045 up_read(&mm->mmap_sem);
1046 down_write(&mm->mmap_sem);
1047 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1048 vma->vm_flags &= ~VM_SOFTDIRTY;
1049 vma_set_page_prot(vma);
1050 }
1051 downgrade_write(&mm->mmap_sem);
1052 break;
1053 }
1054 mmu_notifier_invalidate_range_start(mm, 0, -1);
1055 }
1056 walk_page_range(0, ~0UL, &clear_refs_walk);
1057 if (type == CLEAR_REFS_SOFT_DIRTY)
1058 mmu_notifier_invalidate_range_end(mm, 0, -1);
1059 flush_tlb_mm(mm);
1060 up_read(&mm->mmap_sem);
1061out_mm:
1062 mmput(mm);
1063 }
1064 put_task_struct(task);
1065
1066 return count;
1067}
1068
1069const struct file_operations proc_clear_refs_operations = {
1070 .write = clear_refs_write,
1071 .llseek = noop_llseek,
1072};
1073
1074typedef struct {
1075 u64 pme;
1076} pagemap_entry_t;
1077
1078struct pagemapread {
1079 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */
1080 pagemap_entry_t *buffer;
1081 bool show_pfn;
1082};
1083
1084#define PAGEMAP_WALK_SIZE (PMD_SIZE)
1085#define PAGEMAP_WALK_MASK (PMD_MASK)
1086
1087#define PM_ENTRY_BYTES sizeof(pagemap_entry_t)
1088#define PM_PFRAME_BITS 55
1089#define PM_PFRAME_MASK GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1090#define PM_SOFT_DIRTY BIT_ULL(55)
1091#define PM_MMAP_EXCLUSIVE BIT_ULL(56)
1092#define PM_FILE BIT_ULL(61)
1093#define PM_SWAP BIT_ULL(62)
1094#define PM_PRESENT BIT_ULL(63)
1095
1096#define PM_END_OF_BUFFER 1
1097
1098static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1099{
1100 return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1101}
1102
1103static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1104 struct pagemapread *pm)
1105{
1106 pm->buffer[pm->pos++] = *pme;
1107 if (pm->pos >= pm->len)
1108 return PM_END_OF_BUFFER;
1109 return 0;
1110}
1111
1112static int pagemap_pte_hole(unsigned long start, unsigned long end,
1113 struct mm_walk *walk)
1114{
1115 struct pagemapread *pm = walk->private;
1116 unsigned long addr = start;
1117 int err = 0;
1118
1119 while (addr < end) {
1120 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1121 pagemap_entry_t pme = make_pme(0, 0);
1122 /* End of address space hole, which we mark as non-present. */
1123 unsigned long hole_end;
1124
1125 if (vma)
1126 hole_end = min(end, vma->vm_start);
1127 else
1128 hole_end = end;
1129
1130 for (; addr < hole_end; addr += PAGE_SIZE) {
1131 err = add_to_pagemap(addr, &pme, pm);
1132 if (err)
1133 goto out;
1134 }
1135
1136 if (!vma)
1137 break;
1138
1139 /* Addresses in the VMA. */
1140 if (vma->vm_flags & VM_SOFTDIRTY)
1141 pme = make_pme(0, PM_SOFT_DIRTY);
1142 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1143 err = add_to_pagemap(addr, &pme, pm);
1144 if (err)
1145 goto out;
1146 }
1147 }
1148out:
1149 return err;
1150}
1151
1152static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1153 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1154{
1155 u64 frame = 0, flags = 0;
1156 struct page *page = NULL;
1157
1158 if (pte_present(pte)) {
1159 if (pm->show_pfn)
1160 frame = pte_pfn(pte);
1161 flags |= PM_PRESENT;
1162 page = vm_normal_page(vma, addr, pte);
1163 if (pte_soft_dirty(pte))
1164 flags |= PM_SOFT_DIRTY;
1165 } else if (is_swap_pte(pte)) {
1166 swp_entry_t entry;
1167 if (pte_swp_soft_dirty(pte))
1168 flags |= PM_SOFT_DIRTY;
1169 entry = pte_to_swp_entry(pte);
1170 frame = swp_type(entry) |
1171 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1172 flags |= PM_SWAP;
1173 if (is_migration_entry(entry))
1174 page = migration_entry_to_page(entry);
1175 }
1176
1177 if (page && !PageAnon(page))
1178 flags |= PM_FILE;
1179 if (page && page_mapcount(page) == 1)
1180 flags |= PM_MMAP_EXCLUSIVE;
1181 if (vma->vm_flags & VM_SOFTDIRTY)
1182 flags |= PM_SOFT_DIRTY;
1183
1184 return make_pme(frame, flags);
1185}
1186
1187static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1188 struct mm_walk *walk)
1189{
1190 struct vm_area_struct *vma = walk->vma;
1191 struct pagemapread *pm = walk->private;
1192 spinlock_t *ptl;
1193 pte_t *pte, *orig_pte;
1194 int err = 0;
1195
1196#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1197 ptl = pmd_trans_huge_lock(pmdp, vma);
1198 if (ptl) {
1199 u64 flags = 0, frame = 0;
1200 pmd_t pmd = *pmdp;
1201
1202 if ((vma->vm_flags & VM_SOFTDIRTY) || pmd_soft_dirty(pmd))
1203 flags |= PM_SOFT_DIRTY;
1204
1205 /*
1206 * Currently pmd for thp is always present because thp
1207 * can not be swapped-out, migrated, or HWPOISONed
1208 * (split in such cases instead.)
1209 * This if-check is just to prepare for future implementation.
1210 */
1211 if (pmd_present(pmd)) {
1212 struct page *page = pmd_page(pmd);
1213
1214 if (page_mapcount(page) == 1)
1215 flags |= PM_MMAP_EXCLUSIVE;
1216
1217 flags |= PM_PRESENT;
1218 if (pm->show_pfn)
1219 frame = pmd_pfn(pmd) +
1220 ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1221 }
1222
1223 for (; addr != end; addr += PAGE_SIZE) {
1224 pagemap_entry_t pme = make_pme(frame, flags);
1225
1226 err = add_to_pagemap(addr, &pme, pm);
1227 if (err)
1228 break;
1229 if (pm->show_pfn && (flags & PM_PRESENT))
1230 frame++;
1231 }
1232 spin_unlock(ptl);
1233 return err;
1234 }
1235
1236 if (pmd_trans_unstable(pmdp))
1237 return 0;
1238#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1239
1240 /*
1241 * We can assume that @vma always points to a valid one and @end never
1242 * goes beyond vma->vm_end.
1243 */
1244 orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1245 for (; addr < end; pte++, addr += PAGE_SIZE) {
1246 pagemap_entry_t pme;
1247
1248 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1249 err = add_to_pagemap(addr, &pme, pm);
1250 if (err)
1251 break;
1252 }
1253 pte_unmap_unlock(orig_pte, ptl);
1254
1255 cond_resched();
1256
1257 return err;
1258}
1259
1260#ifdef CONFIG_HUGETLB_PAGE
1261/* This function walks within one hugetlb entry in the single call */
1262static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1263 unsigned long addr, unsigned long end,
1264 struct mm_walk *walk)
1265{
1266 struct pagemapread *pm = walk->private;
1267 struct vm_area_struct *vma = walk->vma;
1268 u64 flags = 0, frame = 0;
1269 int err = 0;
1270 pte_t pte;
1271
1272 if (vma->vm_flags & VM_SOFTDIRTY)
1273 flags |= PM_SOFT_DIRTY;
1274
1275 pte = huge_ptep_get(ptep);
1276 if (pte_present(pte)) {
1277 struct page *page = pte_page(pte);
1278
1279 if (!PageAnon(page))
1280 flags |= PM_FILE;
1281
1282 if (page_mapcount(page) == 1)
1283 flags |= PM_MMAP_EXCLUSIVE;
1284
1285 flags |= PM_PRESENT;
1286 if (pm->show_pfn)
1287 frame = pte_pfn(pte) +
1288 ((addr & ~hmask) >> PAGE_SHIFT);
1289 }
1290
1291 for (; addr != end; addr += PAGE_SIZE) {
1292 pagemap_entry_t pme = make_pme(frame, flags);
1293
1294 err = add_to_pagemap(addr, &pme, pm);
1295 if (err)
1296 return err;
1297 if (pm->show_pfn && (flags & PM_PRESENT))
1298 frame++;
1299 }
1300
1301 cond_resched();
1302
1303 return err;
1304}
1305#endif /* HUGETLB_PAGE */
1306
1307/*
1308 * /proc/pid/pagemap - an array mapping virtual pages to pfns
1309 *
1310 * For each page in the address space, this file contains one 64-bit entry
1311 * consisting of the following:
1312 *
1313 * Bits 0-54 page frame number (PFN) if present
1314 * Bits 0-4 swap type if swapped
1315 * Bits 5-54 swap offset if swapped
1316 * Bit 55 pte is soft-dirty (see Documentation/vm/soft-dirty.txt)
1317 * Bit 56 page exclusively mapped
1318 * Bits 57-60 zero
1319 * Bit 61 page is file-page or shared-anon
1320 * Bit 62 page swapped
1321 * Bit 63 page present
1322 *
1323 * If the page is not present but in swap, then the PFN contains an
1324 * encoding of the swap file number and the page's offset into the
1325 * swap. Unmapped pages return a null PFN. This allows determining
1326 * precisely which pages are mapped (or in swap) and comparing mapped
1327 * pages between processes.
1328 *
1329 * Efficient users of this interface will use /proc/pid/maps to
1330 * determine which areas of memory are actually mapped and llseek to
1331 * skip over unmapped regions.
1332 */
1333static ssize_t pagemap_read(struct file *file, char __user *buf,
1334 size_t count, loff_t *ppos)
1335{
1336 struct mm_struct *mm = file->private_data;
1337 struct pagemapread pm;
1338 struct mm_walk pagemap_walk = {};
1339 unsigned long src;
1340 unsigned long svpfn;
1341 unsigned long start_vaddr;
1342 unsigned long end_vaddr;
1343 int ret = 0, copied = 0;
1344
1345 if (!mm || !atomic_inc_not_zero(&mm->mm_users))
1346 goto out;
1347
1348 ret = -EINVAL;
1349 /* file position must be aligned */
1350 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1351 goto out_mm;
1352
1353 ret = 0;
1354 if (!count)
1355 goto out_mm;
1356
1357 /* do not disclose physical addresses: attack vector */
1358 pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1359
1360 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1361 pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY);
1362 ret = -ENOMEM;
1363 if (!pm.buffer)
1364 goto out_mm;
1365
1366 pagemap_walk.pmd_entry = pagemap_pmd_range;
1367 pagemap_walk.pte_hole = pagemap_pte_hole;
1368#ifdef CONFIG_HUGETLB_PAGE
1369 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1370#endif
1371 pagemap_walk.mm = mm;
1372 pagemap_walk.private = ±
1373
1374 src = *ppos;
1375 svpfn = src / PM_ENTRY_BYTES;
1376 start_vaddr = svpfn << PAGE_SHIFT;
1377 end_vaddr = mm->task_size;
1378
1379 /* watch out for wraparound */
1380 if (svpfn > mm->task_size >> PAGE_SHIFT)
1381 start_vaddr = end_vaddr;
1382
1383 /*
1384 * The odds are that this will stop walking way
1385 * before end_vaddr, because the length of the
1386 * user buffer is tracked in "pm", and the walk
1387 * will stop when we hit the end of the buffer.
1388 */
1389 ret = 0;
1390 while (count && (start_vaddr < end_vaddr)) {
1391 int len;
1392 unsigned long end;
1393
1394 pm.pos = 0;
1395 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1396 /* overflow ? */
1397 if (end < start_vaddr || end > end_vaddr)
1398 end = end_vaddr;
1399 down_read(&mm->mmap_sem);
1400 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1401 up_read(&mm->mmap_sem);
1402 start_vaddr = end;
1403
1404 len = min(count, PM_ENTRY_BYTES * pm.pos);
1405 if (copy_to_user(buf, pm.buffer, len)) {
1406 ret = -EFAULT;
1407 goto out_free;
1408 }
1409 copied += len;
1410 buf += len;
1411 count -= len;
1412 }
1413 *ppos += copied;
1414 if (!ret || ret == PM_END_OF_BUFFER)
1415 ret = copied;
1416
1417out_free:
1418 kfree(pm.buffer);
1419out_mm:
1420 mmput(mm);
1421out:
1422 return ret;
1423}
1424
1425static int pagemap_open(struct inode *inode, struct file *file)
1426{
1427 struct mm_struct *mm;
1428
1429 mm = proc_mem_open(inode, PTRACE_MODE_READ);
1430 if (IS_ERR(mm))
1431 return PTR_ERR(mm);
1432 file->private_data = mm;
1433 return 0;
1434}
1435
1436static int pagemap_release(struct inode *inode, struct file *file)
1437{
1438 struct mm_struct *mm = file->private_data;
1439
1440 if (mm)
1441 mmdrop(mm);
1442 return 0;
1443}
1444
1445const struct file_operations proc_pagemap_operations = {
1446 .llseek = mem_lseek, /* borrow this */
1447 .read = pagemap_read,
1448 .open = pagemap_open,
1449 .release = pagemap_release,
1450};
1451#endif /* CONFIG_PROC_PAGE_MONITOR */
1452
1453#ifdef CONFIG_NUMA
1454
1455struct numa_maps {
1456 unsigned long pages;
1457 unsigned long anon;
1458 unsigned long active;
1459 unsigned long writeback;
1460 unsigned long mapcount_max;
1461 unsigned long dirty;
1462 unsigned long swapcache;
1463 unsigned long node[MAX_NUMNODES];
1464};
1465
1466struct numa_maps_private {
1467 struct proc_maps_private proc_maps;
1468 struct numa_maps md;
1469};
1470
1471static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1472 unsigned long nr_pages)
1473{
1474 int count = page_mapcount(page);
1475
1476 md->pages += nr_pages;
1477 if (pte_dirty || PageDirty(page))
1478 md->dirty += nr_pages;
1479
1480 if (PageSwapCache(page))
1481 md->swapcache += nr_pages;
1482
1483 if (PageActive(page) || PageUnevictable(page))
1484 md->active += nr_pages;
1485
1486 if (PageWriteback(page))
1487 md->writeback += nr_pages;
1488
1489 if (PageAnon(page))
1490 md->anon += nr_pages;
1491
1492 if (count > md->mapcount_max)
1493 md->mapcount_max = count;
1494
1495 md->node[page_to_nid(page)] += nr_pages;
1496}
1497
1498static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1499 unsigned long addr)
1500{
1501 struct page *page;
1502 int nid;
1503
1504 if (!pte_present(pte))
1505 return NULL;
1506
1507 page = vm_normal_page(vma, addr, pte);
1508 if (!page)
1509 return NULL;
1510
1511 if (PageReserved(page))
1512 return NULL;
1513
1514 nid = page_to_nid(page);
1515 if (!node_isset(nid, node_states[N_MEMORY]))
1516 return NULL;
1517
1518 return page;
1519}
1520
1521#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1522static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1523 struct vm_area_struct *vma,
1524 unsigned long addr)
1525{
1526 struct page *page;
1527 int nid;
1528
1529 if (!pmd_present(pmd))
1530 return NULL;
1531
1532 page = vm_normal_page_pmd(vma, addr, pmd);
1533 if (!page)
1534 return NULL;
1535
1536 if (PageReserved(page))
1537 return NULL;
1538
1539 nid = page_to_nid(page);
1540 if (!node_isset(nid, node_states[N_MEMORY]))
1541 return NULL;
1542
1543 return page;
1544}
1545#endif
1546
1547static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1548 unsigned long end, struct mm_walk *walk)
1549{
1550 struct numa_maps *md = walk->private;
1551 struct vm_area_struct *vma = walk->vma;
1552 spinlock_t *ptl;
1553 pte_t *orig_pte;
1554 pte_t *pte;
1555
1556#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1557 ptl = pmd_trans_huge_lock(pmd, vma);
1558 if (ptl) {
1559 struct page *page;
1560
1561 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1562 if (page)
1563 gather_stats(page, md, pmd_dirty(*pmd),
1564 HPAGE_PMD_SIZE/PAGE_SIZE);
1565 spin_unlock(ptl);
1566 return 0;
1567 }
1568
1569 if (pmd_trans_unstable(pmd))
1570 return 0;
1571#endif
1572 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1573 do {
1574 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1575 if (!page)
1576 continue;
1577 gather_stats(page, md, pte_dirty(*pte), 1);
1578
1579 } while (pte++, addr += PAGE_SIZE, addr != end);
1580 pte_unmap_unlock(orig_pte, ptl);
1581 return 0;
1582}
1583#ifdef CONFIG_HUGETLB_PAGE
1584static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1585 unsigned long addr, unsigned long end, struct mm_walk *walk)
1586{
1587 pte_t huge_pte = huge_ptep_get(pte);
1588 struct numa_maps *md;
1589 struct page *page;
1590
1591 if (!pte_present(huge_pte))
1592 return 0;
1593
1594 page = pte_page(huge_pte);
1595 if (!page)
1596 return 0;
1597
1598 md = walk->private;
1599 gather_stats(page, md, pte_dirty(huge_pte), 1);
1600 return 0;
1601}
1602
1603#else
1604static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1605 unsigned long addr, unsigned long end, struct mm_walk *walk)
1606{
1607 return 0;
1608}
1609#endif
1610
1611/*
1612 * Display pages allocated per node and memory policy via /proc.
1613 */
1614static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1615{
1616 struct numa_maps_private *numa_priv = m->private;
1617 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1618 struct vm_area_struct *vma = v;
1619 struct numa_maps *md = &numa_priv->md;
1620 struct file *file = vma->vm_file;
1621 struct mm_struct *mm = vma->vm_mm;
1622 struct mm_walk walk = {
1623 .hugetlb_entry = gather_hugetlb_stats,
1624 .pmd_entry = gather_pte_stats,
1625 .private = md,
1626 .mm = mm,
1627 };
1628 struct mempolicy *pol;
1629 char buffer[64];
1630 int nid;
1631
1632 if (!mm)
1633 return 0;
1634
1635 /* Ensure we start with an empty set of numa_maps statistics. */
1636 memset(md, 0, sizeof(*md));
1637
1638 pol = __get_vma_policy(vma, vma->vm_start);
1639 if (pol) {
1640 mpol_to_str(buffer, sizeof(buffer), pol);
1641 mpol_cond_put(pol);
1642 } else {
1643 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1644 }
1645
1646 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1647
1648 if (file) {
1649 seq_puts(m, " file=");
1650 seq_file_path(m, file, "\n\t= ");
1651 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1652 seq_puts(m, " heap");
1653 } else if (is_stack(proc_priv, vma, is_pid)) {
1654 seq_puts(m, " stack");
1655 }
1656
1657 if (is_vm_hugetlb_page(vma))
1658 seq_puts(m, " huge");
1659
1660 /* mmap_sem is held by m_start */
1661 walk_page_vma(vma, &walk);
1662
1663 if (!md->pages)
1664 goto out;
1665
1666 if (md->anon)
1667 seq_printf(m, " anon=%lu", md->anon);
1668
1669 if (md->dirty)
1670 seq_printf(m, " dirty=%lu", md->dirty);
1671
1672 if (md->pages != md->anon && md->pages != md->dirty)
1673 seq_printf(m, " mapped=%lu", md->pages);
1674
1675 if (md->mapcount_max > 1)
1676 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1677
1678 if (md->swapcache)
1679 seq_printf(m, " swapcache=%lu", md->swapcache);
1680
1681 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1682 seq_printf(m, " active=%lu", md->active);
1683
1684 if (md->writeback)
1685 seq_printf(m, " writeback=%lu", md->writeback);
1686
1687 for_each_node_state(nid, N_MEMORY)
1688 if (md->node[nid])
1689 seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1690
1691 seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1692out:
1693 seq_putc(m, '\n');
1694 m_cache_vma(m, vma);
1695 return 0;
1696}
1697
1698static int show_pid_numa_map(struct seq_file *m, void *v)
1699{
1700 return show_numa_map(m, v, 1);
1701}
1702
1703static int show_tid_numa_map(struct seq_file *m, void *v)
1704{
1705 return show_numa_map(m, v, 0);
1706}
1707
1708static const struct seq_operations proc_pid_numa_maps_op = {
1709 .start = m_start,
1710 .next = m_next,
1711 .stop = m_stop,
1712 .show = show_pid_numa_map,
1713};
1714
1715static const struct seq_operations proc_tid_numa_maps_op = {
1716 .start = m_start,
1717 .next = m_next,
1718 .stop = m_stop,
1719 .show = show_tid_numa_map,
1720};
1721
1722static int numa_maps_open(struct inode *inode, struct file *file,
1723 const struct seq_operations *ops)
1724{
1725 return proc_maps_open(inode, file, ops,
1726 sizeof(struct numa_maps_private));
1727}
1728
1729static int pid_numa_maps_open(struct inode *inode, struct file *file)
1730{
1731 return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1732}
1733
1734static int tid_numa_maps_open(struct inode *inode, struct file *file)
1735{
1736 return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1737}
1738
1739const struct file_operations proc_pid_numa_maps_operations = {
1740 .open = pid_numa_maps_open,
1741 .read = seq_read,
1742 .llseek = seq_lseek,
1743 .release = proc_map_release,
1744};
1745
1746const struct file_operations proc_tid_numa_maps_operations = {
1747 .open = tid_numa_maps_open,
1748 .read = seq_read,
1749 .llseek = seq_lseek,
1750 .release = proc_map_release,
1751};
1752#endif /* CONFIG_NUMA */
1// 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)
30void 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
82unsigned long task_vsize(struct mm_struct *mm)
83{
84 return PAGE_SIZE * mm->total_vm;
85}
86
87unsigned 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 */
104static 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}
113static void release_task_mempolicy(struct proc_maps_private *priv)
114{
115 mpol_put(priv->task_mempolicy);
116}
117#else
118static void hold_task_mempolicy(struct proc_maps_private *priv)
119{
120}
121static void release_task_mempolicy(struct proc_maps_private *priv)
122{
123}
124#endif
125
126static void *m_start(struct seq_file *m, loff_t *ppos)
127{
128 struct proc_maps_private *priv = m->private;
129 unsigned long last_addr = *ppos;
130 struct mm_struct *mm;
131 struct vm_area_struct *vma;
132
133 /* See m_next(). Zero at the start or after lseek. */
134 if (last_addr == -1UL)
135 return NULL;
136
137 priv->task = get_proc_task(priv->inode);
138 if (!priv->task)
139 return ERR_PTR(-ESRCH);
140
141 mm = priv->mm;
142 if (!mm || !mmget_not_zero(mm)) {
143 put_task_struct(priv->task);
144 priv->task = NULL;
145 return NULL;
146 }
147
148 if (mmap_read_lock_killable(mm)) {
149 mmput(mm);
150 put_task_struct(priv->task);
151 priv->task = NULL;
152 return ERR_PTR(-EINTR);
153 }
154
155 hold_task_mempolicy(priv);
156 priv->tail_vma = get_gate_vma(mm);
157
158 vma = find_vma(mm, last_addr);
159 if (vma)
160 return vma;
161
162 return priv->tail_vma;
163}
164
165static void *m_next(struct seq_file *m, void *v, loff_t *ppos)
166{
167 struct proc_maps_private *priv = m->private;
168 struct vm_area_struct *next, *vma = v;
169
170 if (vma == priv->tail_vma)
171 next = NULL;
172 else if (vma->vm_next)
173 next = vma->vm_next;
174 else
175 next = priv->tail_vma;
176
177 *ppos = next ? next->vm_start : -1UL;
178
179 return next;
180}
181
182static void m_stop(struct seq_file *m, void *v)
183{
184 struct proc_maps_private *priv = m->private;
185 struct mm_struct *mm = priv->mm;
186
187 if (!priv->task)
188 return;
189
190 release_task_mempolicy(priv);
191 mmap_read_unlock(mm);
192 mmput(mm);
193 put_task_struct(priv->task);
194 priv->task = NULL;
195}
196
197static int proc_maps_open(struct inode *inode, struct file *file,
198 const struct seq_operations *ops, int psize)
199{
200 struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
201
202 if (!priv)
203 return -ENOMEM;
204
205 priv->inode = inode;
206 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
207 if (IS_ERR(priv->mm)) {
208 int err = PTR_ERR(priv->mm);
209
210 seq_release_private(inode, file);
211 return err;
212 }
213
214 return 0;
215}
216
217static int proc_map_release(struct inode *inode, struct file *file)
218{
219 struct seq_file *seq = file->private_data;
220 struct proc_maps_private *priv = seq->private;
221
222 if (priv->mm)
223 mmdrop(priv->mm);
224
225 return seq_release_private(inode, file);
226}
227
228static int do_maps_open(struct inode *inode, struct file *file,
229 const struct seq_operations *ops)
230{
231 return proc_maps_open(inode, file, ops,
232 sizeof(struct proc_maps_private));
233}
234
235/*
236 * Indicate if the VMA is a stack for the given task; for
237 * /proc/PID/maps that is the stack of the main task.
238 */
239static int is_stack(struct vm_area_struct *vma)
240{
241 /*
242 * We make no effort to guess what a given thread considers to be
243 * its "stack". It's not even well-defined for programs written
244 * languages like Go.
245 */
246 return vma->vm_start <= vma->vm_mm->start_stack &&
247 vma->vm_end >= vma->vm_mm->start_stack;
248}
249
250static void show_vma_header_prefix(struct seq_file *m,
251 unsigned long start, unsigned long end,
252 vm_flags_t flags, unsigned long long pgoff,
253 dev_t dev, unsigned long ino)
254{
255 seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
256 seq_put_hex_ll(m, NULL, start, 8);
257 seq_put_hex_ll(m, "-", end, 8);
258 seq_putc(m, ' ');
259 seq_putc(m, flags & VM_READ ? 'r' : '-');
260 seq_putc(m, flags & VM_WRITE ? 'w' : '-');
261 seq_putc(m, flags & VM_EXEC ? 'x' : '-');
262 seq_putc(m, flags & VM_MAYSHARE ? 's' : 'p');
263 seq_put_hex_ll(m, " ", pgoff, 8);
264 seq_put_hex_ll(m, " ", MAJOR(dev), 2);
265 seq_put_hex_ll(m, ":", MINOR(dev), 2);
266 seq_put_decimal_ull(m, " ", ino);
267 seq_putc(m, ' ');
268}
269
270static void
271show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
272{
273 struct mm_struct *mm = vma->vm_mm;
274 struct file *file = vma->vm_file;
275 vm_flags_t flags = vma->vm_flags;
276 unsigned long ino = 0;
277 unsigned long long pgoff = 0;
278 unsigned long start, end;
279 dev_t dev = 0;
280 const char *name = NULL;
281
282 if (file) {
283 struct inode *inode = file_inode(vma->vm_file);
284 dev = inode->i_sb->s_dev;
285 ino = inode->i_ino;
286 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
287 }
288
289 start = vma->vm_start;
290 end = vma->vm_end;
291 show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino);
292
293 /*
294 * Print the dentry name for named mappings, and a
295 * special [heap] marker for the heap:
296 */
297 if (file) {
298 seq_pad(m, ' ');
299 seq_file_path(m, file, "\n");
300 goto done;
301 }
302
303 if (vma->vm_ops && vma->vm_ops->name) {
304 name = vma->vm_ops->name(vma);
305 if (name)
306 goto done;
307 }
308
309 name = arch_vma_name(vma);
310 if (!name) {
311 if (!mm) {
312 name = "[vdso]";
313 goto done;
314 }
315
316 if (vma->vm_start <= mm->brk &&
317 vma->vm_end >= mm->start_brk) {
318 name = "[heap]";
319 goto done;
320 }
321
322 if (is_stack(vma))
323 name = "[stack]";
324 }
325
326done:
327 if (name) {
328 seq_pad(m, ' ');
329 seq_puts(m, name);
330 }
331 seq_putc(m, '\n');
332}
333
334static int show_map(struct seq_file *m, void *v)
335{
336 show_map_vma(m, v);
337 return 0;
338}
339
340static const struct seq_operations proc_pid_maps_op = {
341 .start = m_start,
342 .next = m_next,
343 .stop = m_stop,
344 .show = show_map
345};
346
347static int pid_maps_open(struct inode *inode, struct file *file)
348{
349 return do_maps_open(inode, file, &proc_pid_maps_op);
350}
351
352const struct file_operations proc_pid_maps_operations = {
353 .open = pid_maps_open,
354 .read = seq_read,
355 .llseek = seq_lseek,
356 .release = proc_map_release,
357};
358
359/*
360 * Proportional Set Size(PSS): my share of RSS.
361 *
362 * PSS of a process is the count of pages it has in memory, where each
363 * page is divided by the number of processes sharing it. So if a
364 * process has 1000 pages all to itself, and 1000 shared with one other
365 * process, its PSS will be 1500.
366 *
367 * To keep (accumulated) division errors low, we adopt a 64bit
368 * fixed-point pss counter to minimize division errors. So (pss >>
369 * PSS_SHIFT) would be the real byte count.
370 *
371 * A shift of 12 before division means (assuming 4K page size):
372 * - 1M 3-user-pages add up to 8KB errors;
373 * - supports mapcount up to 2^24, or 16M;
374 * - supports PSS up to 2^52 bytes, or 4PB.
375 */
376#define PSS_SHIFT 12
377
378#ifdef CONFIG_PROC_PAGE_MONITOR
379struct mem_size_stats {
380 unsigned long resident;
381 unsigned long shared_clean;
382 unsigned long shared_dirty;
383 unsigned long private_clean;
384 unsigned long private_dirty;
385 unsigned long referenced;
386 unsigned long anonymous;
387 unsigned long lazyfree;
388 unsigned long anonymous_thp;
389 unsigned long shmem_thp;
390 unsigned long file_thp;
391 unsigned long swap;
392 unsigned long shared_hugetlb;
393 unsigned long private_hugetlb;
394 u64 pss;
395 u64 pss_anon;
396 u64 pss_file;
397 u64 pss_shmem;
398 u64 pss_locked;
399 u64 swap_pss;
400 bool check_shmem_swap;
401};
402
403static void smaps_page_accumulate(struct mem_size_stats *mss,
404 struct page *page, unsigned long size, unsigned long pss,
405 bool dirty, bool locked, bool private)
406{
407 mss->pss += pss;
408
409 if (PageAnon(page))
410 mss->pss_anon += pss;
411 else if (PageSwapBacked(page))
412 mss->pss_shmem += pss;
413 else
414 mss->pss_file += pss;
415
416 if (locked)
417 mss->pss_locked += pss;
418
419 if (dirty || PageDirty(page)) {
420 if (private)
421 mss->private_dirty += size;
422 else
423 mss->shared_dirty += size;
424 } else {
425 if (private)
426 mss->private_clean += size;
427 else
428 mss->shared_clean += size;
429 }
430}
431
432static void smaps_account(struct mem_size_stats *mss, struct page *page,
433 bool compound, bool young, bool dirty, bool locked)
434{
435 int i, nr = compound ? compound_nr(page) : 1;
436 unsigned long size = nr * PAGE_SIZE;
437
438 /*
439 * First accumulate quantities that depend only on |size| and the type
440 * of the compound page.
441 */
442 if (PageAnon(page)) {
443 mss->anonymous += size;
444 if (!PageSwapBacked(page) && !dirty && !PageDirty(page))
445 mss->lazyfree += size;
446 }
447
448 mss->resident += size;
449 /* Accumulate the size in pages that have been accessed. */
450 if (young || page_is_young(page) || PageReferenced(page))
451 mss->referenced += size;
452
453 /*
454 * Then accumulate quantities that may depend on sharing, or that may
455 * differ page-by-page.
456 *
457 * page_count(page) == 1 guarantees the page is mapped exactly once.
458 * If any subpage of the compound page mapped with PTE it would elevate
459 * page_count().
460 */
461 if (page_count(page) == 1) {
462 smaps_page_accumulate(mss, page, size, size << PSS_SHIFT, dirty,
463 locked, true);
464 return;
465 }
466 for (i = 0; i < nr; i++, page++) {
467 int mapcount = page_mapcount(page);
468 unsigned long pss = PAGE_SIZE << PSS_SHIFT;
469 if (mapcount >= 2)
470 pss /= mapcount;
471 smaps_page_accumulate(mss, page, PAGE_SIZE, pss, dirty, locked,
472 mapcount < 2);
473 }
474}
475
476#ifdef CONFIG_SHMEM
477static int smaps_pte_hole(unsigned long addr, unsigned long end,
478 __always_unused int depth, struct mm_walk *walk)
479{
480 struct mem_size_stats *mss = walk->private;
481
482 mss->swap += shmem_partial_swap_usage(
483 walk->vma->vm_file->f_mapping, addr, end);
484
485 return 0;
486}
487#else
488#define smaps_pte_hole NULL
489#endif /* CONFIG_SHMEM */
490
491static void smaps_pte_entry(pte_t *pte, unsigned long addr,
492 struct mm_walk *walk)
493{
494 struct mem_size_stats *mss = walk->private;
495 struct vm_area_struct *vma = walk->vma;
496 bool locked = !!(vma->vm_flags & VM_LOCKED);
497 struct page *page = NULL;
498
499 if (pte_present(*pte)) {
500 page = vm_normal_page(vma, addr, *pte);
501 } else if (is_swap_pte(*pte)) {
502 swp_entry_t swpent = pte_to_swp_entry(*pte);
503
504 if (!non_swap_entry(swpent)) {
505 int mapcount;
506
507 mss->swap += PAGE_SIZE;
508 mapcount = swp_swapcount(swpent);
509 if (mapcount >= 2) {
510 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
511
512 do_div(pss_delta, mapcount);
513 mss->swap_pss += pss_delta;
514 } else {
515 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
516 }
517 } else if (is_pfn_swap_entry(swpent))
518 page = pfn_swap_entry_to_page(swpent);
519 } else if (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap
520 && pte_none(*pte))) {
521 page = xa_load(&vma->vm_file->f_mapping->i_pages,
522 linear_page_index(vma, addr));
523 if (xa_is_value(page))
524 mss->swap += PAGE_SIZE;
525 return;
526 }
527
528 if (!page)
529 return;
530
531 smaps_account(mss, page, false, pte_young(*pte), pte_dirty(*pte), locked);
532}
533
534#ifdef CONFIG_TRANSPARENT_HUGEPAGE
535static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
536 struct mm_walk *walk)
537{
538 struct mem_size_stats *mss = walk->private;
539 struct vm_area_struct *vma = walk->vma;
540 bool locked = !!(vma->vm_flags & VM_LOCKED);
541 struct page *page = NULL;
542
543 if (pmd_present(*pmd)) {
544 /* FOLL_DUMP will return -EFAULT on huge zero page */
545 page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
546 } else if (unlikely(thp_migration_supported() && is_swap_pmd(*pmd))) {
547 swp_entry_t entry = pmd_to_swp_entry(*pmd);
548
549 if (is_migration_entry(entry))
550 page = pfn_swap_entry_to_page(entry);
551 }
552 if (IS_ERR_OR_NULL(page))
553 return;
554 if (PageAnon(page))
555 mss->anonymous_thp += HPAGE_PMD_SIZE;
556 else if (PageSwapBacked(page))
557 mss->shmem_thp += HPAGE_PMD_SIZE;
558 else if (is_zone_device_page(page))
559 /* pass */;
560 else
561 mss->file_thp += HPAGE_PMD_SIZE;
562 smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd), locked);
563}
564#else
565static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
566 struct mm_walk *walk)
567{
568}
569#endif
570
571static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
572 struct mm_walk *walk)
573{
574 struct vm_area_struct *vma = walk->vma;
575 pte_t *pte;
576 spinlock_t *ptl;
577
578 ptl = pmd_trans_huge_lock(pmd, vma);
579 if (ptl) {
580 smaps_pmd_entry(pmd, addr, walk);
581 spin_unlock(ptl);
582 goto out;
583 }
584
585 if (pmd_trans_unstable(pmd))
586 goto out;
587 /*
588 * The mmap_lock held all the way back in m_start() is what
589 * keeps khugepaged out of here and from collapsing things
590 * in here.
591 */
592 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
593 for (; addr != end; pte++, addr += PAGE_SIZE)
594 smaps_pte_entry(pte, addr, walk);
595 pte_unmap_unlock(pte - 1, ptl);
596out:
597 cond_resched();
598 return 0;
599}
600
601static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
602{
603 /*
604 * Don't forget to update Documentation/ on changes.
605 */
606 static const char mnemonics[BITS_PER_LONG][2] = {
607 /*
608 * In case if we meet a flag we don't know about.
609 */
610 [0 ... (BITS_PER_LONG-1)] = "??",
611
612 [ilog2(VM_READ)] = "rd",
613 [ilog2(VM_WRITE)] = "wr",
614 [ilog2(VM_EXEC)] = "ex",
615 [ilog2(VM_SHARED)] = "sh",
616 [ilog2(VM_MAYREAD)] = "mr",
617 [ilog2(VM_MAYWRITE)] = "mw",
618 [ilog2(VM_MAYEXEC)] = "me",
619 [ilog2(VM_MAYSHARE)] = "ms",
620 [ilog2(VM_GROWSDOWN)] = "gd",
621 [ilog2(VM_PFNMAP)] = "pf",
622 [ilog2(VM_DENYWRITE)] = "dw",
623 [ilog2(VM_LOCKED)] = "lo",
624 [ilog2(VM_IO)] = "io",
625 [ilog2(VM_SEQ_READ)] = "sr",
626 [ilog2(VM_RAND_READ)] = "rr",
627 [ilog2(VM_DONTCOPY)] = "dc",
628 [ilog2(VM_DONTEXPAND)] = "de",
629 [ilog2(VM_ACCOUNT)] = "ac",
630 [ilog2(VM_NORESERVE)] = "nr",
631 [ilog2(VM_HUGETLB)] = "ht",
632 [ilog2(VM_SYNC)] = "sf",
633 [ilog2(VM_ARCH_1)] = "ar",
634 [ilog2(VM_WIPEONFORK)] = "wf",
635 [ilog2(VM_DONTDUMP)] = "dd",
636#ifdef CONFIG_ARM64_BTI
637 [ilog2(VM_ARM64_BTI)] = "bt",
638#endif
639#ifdef CONFIG_MEM_SOFT_DIRTY
640 [ilog2(VM_SOFTDIRTY)] = "sd",
641#endif
642 [ilog2(VM_MIXEDMAP)] = "mm",
643 [ilog2(VM_HUGEPAGE)] = "hg",
644 [ilog2(VM_NOHUGEPAGE)] = "nh",
645 [ilog2(VM_MERGEABLE)] = "mg",
646 [ilog2(VM_UFFD_MISSING)]= "um",
647 [ilog2(VM_UFFD_WP)] = "uw",
648#ifdef CONFIG_ARM64_MTE
649 [ilog2(VM_MTE)] = "mt",
650 [ilog2(VM_MTE_ALLOWED)] = "",
651#endif
652#ifdef CONFIG_ARCH_HAS_PKEYS
653 /* These come out via ProtectionKey: */
654 [ilog2(VM_PKEY_BIT0)] = "",
655 [ilog2(VM_PKEY_BIT1)] = "",
656 [ilog2(VM_PKEY_BIT2)] = "",
657 [ilog2(VM_PKEY_BIT3)] = "",
658#if VM_PKEY_BIT4
659 [ilog2(VM_PKEY_BIT4)] = "",
660#endif
661#endif /* CONFIG_ARCH_HAS_PKEYS */
662#ifdef CONFIG_HAVE_ARCH_USERFAULTFD_MINOR
663 [ilog2(VM_UFFD_MINOR)] = "ui",
664#endif /* CONFIG_HAVE_ARCH_USERFAULTFD_MINOR */
665 };
666 size_t i;
667
668 seq_puts(m, "VmFlags: ");
669 for (i = 0; i < BITS_PER_LONG; i++) {
670 if (!mnemonics[i][0])
671 continue;
672 if (vma->vm_flags & (1UL << i)) {
673 seq_putc(m, mnemonics[i][0]);
674 seq_putc(m, mnemonics[i][1]);
675 seq_putc(m, ' ');
676 }
677 }
678 seq_putc(m, '\n');
679}
680
681#ifdef CONFIG_HUGETLB_PAGE
682static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
683 unsigned long addr, unsigned long end,
684 struct mm_walk *walk)
685{
686 struct mem_size_stats *mss = walk->private;
687 struct vm_area_struct *vma = walk->vma;
688 struct page *page = NULL;
689
690 if (pte_present(*pte)) {
691 page = vm_normal_page(vma, addr, *pte);
692 } else if (is_swap_pte(*pte)) {
693 swp_entry_t swpent = pte_to_swp_entry(*pte);
694
695 if (is_pfn_swap_entry(swpent))
696 page = pfn_swap_entry_to_page(swpent);
697 }
698 if (page) {
699 int mapcount = page_mapcount(page);
700
701 if (mapcount >= 2)
702 mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
703 else
704 mss->private_hugetlb += huge_page_size(hstate_vma(vma));
705 }
706 return 0;
707}
708#else
709#define smaps_hugetlb_range NULL
710#endif /* HUGETLB_PAGE */
711
712static const struct mm_walk_ops smaps_walk_ops = {
713 .pmd_entry = smaps_pte_range,
714 .hugetlb_entry = smaps_hugetlb_range,
715};
716
717static const struct mm_walk_ops smaps_shmem_walk_ops = {
718 .pmd_entry = smaps_pte_range,
719 .hugetlb_entry = smaps_hugetlb_range,
720 .pte_hole = smaps_pte_hole,
721};
722
723/*
724 * Gather mem stats from @vma with the indicated beginning
725 * address @start, and keep them in @mss.
726 *
727 * Use vm_start of @vma as the beginning address if @start is 0.
728 */
729static void smap_gather_stats(struct vm_area_struct *vma,
730 struct mem_size_stats *mss, unsigned long start)
731{
732 const struct mm_walk_ops *ops = &smaps_walk_ops;
733
734 /* Invalid start */
735 if (start >= vma->vm_end)
736 return;
737
738#ifdef CONFIG_SHMEM
739 /* In case of smaps_rollup, reset the value from previous vma */
740 mss->check_shmem_swap = false;
741 if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
742 /*
743 * For shared or readonly shmem mappings we know that all
744 * swapped out pages belong to the shmem object, and we can
745 * obtain the swap value much more efficiently. For private
746 * writable mappings, we might have COW pages that are
747 * not affected by the parent swapped out pages of the shmem
748 * object, so we have to distinguish them during the page walk.
749 * Unless we know that the shmem object (or the part mapped by
750 * our VMA) has no swapped out pages at all.
751 */
752 unsigned long shmem_swapped = shmem_swap_usage(vma);
753
754 if (!start && (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
755 !(vma->vm_flags & VM_WRITE))) {
756 mss->swap += shmem_swapped;
757 } else {
758 mss->check_shmem_swap = true;
759 ops = &smaps_shmem_walk_ops;
760 }
761 }
762#endif
763 /* mmap_lock is held in m_start */
764 if (!start)
765 walk_page_vma(vma, ops, mss);
766 else
767 walk_page_range(vma->vm_mm, start, vma->vm_end, ops, mss);
768}
769
770#define SEQ_PUT_DEC(str, val) \
771 seq_put_decimal_ull_width(m, str, (val) >> 10, 8)
772
773/* Show the contents common for smaps and smaps_rollup */
774static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss,
775 bool rollup_mode)
776{
777 SEQ_PUT_DEC("Rss: ", mss->resident);
778 SEQ_PUT_DEC(" kB\nPss: ", mss->pss >> PSS_SHIFT);
779 if (rollup_mode) {
780 /*
781 * These are meaningful only for smaps_rollup, otherwise two of
782 * them are zero, and the other one is the same as Pss.
783 */
784 SEQ_PUT_DEC(" kB\nPss_Anon: ",
785 mss->pss_anon >> PSS_SHIFT);
786 SEQ_PUT_DEC(" kB\nPss_File: ",
787 mss->pss_file >> PSS_SHIFT);
788 SEQ_PUT_DEC(" kB\nPss_Shmem: ",
789 mss->pss_shmem >> PSS_SHIFT);
790 }
791 SEQ_PUT_DEC(" kB\nShared_Clean: ", mss->shared_clean);
792 SEQ_PUT_DEC(" kB\nShared_Dirty: ", mss->shared_dirty);
793 SEQ_PUT_DEC(" kB\nPrivate_Clean: ", mss->private_clean);
794 SEQ_PUT_DEC(" kB\nPrivate_Dirty: ", mss->private_dirty);
795 SEQ_PUT_DEC(" kB\nReferenced: ", mss->referenced);
796 SEQ_PUT_DEC(" kB\nAnonymous: ", mss->anonymous);
797 SEQ_PUT_DEC(" kB\nLazyFree: ", mss->lazyfree);
798 SEQ_PUT_DEC(" kB\nAnonHugePages: ", mss->anonymous_thp);
799 SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp);
800 SEQ_PUT_DEC(" kB\nFilePmdMapped: ", mss->file_thp);
801 SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb);
802 seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ",
803 mss->private_hugetlb >> 10, 7);
804 SEQ_PUT_DEC(" kB\nSwap: ", mss->swap);
805 SEQ_PUT_DEC(" kB\nSwapPss: ",
806 mss->swap_pss >> PSS_SHIFT);
807 SEQ_PUT_DEC(" kB\nLocked: ",
808 mss->pss_locked >> PSS_SHIFT);
809 seq_puts(m, " kB\n");
810}
811
812static int show_smap(struct seq_file *m, void *v)
813{
814 struct vm_area_struct *vma = v;
815 struct mem_size_stats mss;
816
817 memset(&mss, 0, sizeof(mss));
818
819 smap_gather_stats(vma, &mss, 0);
820
821 show_map_vma(m, vma);
822
823 SEQ_PUT_DEC("Size: ", vma->vm_end - vma->vm_start);
824 SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma));
825 SEQ_PUT_DEC(" kB\nMMUPageSize: ", vma_mmu_pagesize(vma));
826 seq_puts(m, " kB\n");
827
828 __show_smap(m, &mss, false);
829
830 seq_printf(m, "THPeligible: %d\n",
831 transparent_hugepage_active(vma));
832
833 if (arch_pkeys_enabled())
834 seq_printf(m, "ProtectionKey: %8u\n", vma_pkey(vma));
835 show_smap_vma_flags(m, vma);
836
837 return 0;
838}
839
840static int show_smaps_rollup(struct seq_file *m, void *v)
841{
842 struct proc_maps_private *priv = m->private;
843 struct mem_size_stats mss;
844 struct mm_struct *mm;
845 struct vm_area_struct *vma;
846 unsigned long last_vma_end = 0;
847 int ret = 0;
848
849 priv->task = get_proc_task(priv->inode);
850 if (!priv->task)
851 return -ESRCH;
852
853 mm = priv->mm;
854 if (!mm || !mmget_not_zero(mm)) {
855 ret = -ESRCH;
856 goto out_put_task;
857 }
858
859 memset(&mss, 0, sizeof(mss));
860
861 ret = mmap_read_lock_killable(mm);
862 if (ret)
863 goto out_put_mm;
864
865 hold_task_mempolicy(priv);
866
867 for (vma = priv->mm->mmap; vma;) {
868 smap_gather_stats(vma, &mss, 0);
869 last_vma_end = vma->vm_end;
870
871 /*
872 * Release mmap_lock temporarily if someone wants to
873 * access it for write request.
874 */
875 if (mmap_lock_is_contended(mm)) {
876 mmap_read_unlock(mm);
877 ret = mmap_read_lock_killable(mm);
878 if (ret) {
879 release_task_mempolicy(priv);
880 goto out_put_mm;
881 }
882
883 /*
884 * After dropping the lock, there are four cases to
885 * consider. See the following example for explanation.
886 *
887 * +------+------+-----------+
888 * | VMA1 | VMA2 | VMA3 |
889 * +------+------+-----------+
890 * | | | |
891 * 4k 8k 16k 400k
892 *
893 * Suppose we drop the lock after reading VMA2 due to
894 * contention, then we get:
895 *
896 * last_vma_end = 16k
897 *
898 * 1) VMA2 is freed, but VMA3 exists:
899 *
900 * find_vma(mm, 16k - 1) will return VMA3.
901 * In this case, just continue from VMA3.
902 *
903 * 2) VMA2 still exists:
904 *
905 * find_vma(mm, 16k - 1) will return VMA2.
906 * Iterate the loop like the original one.
907 *
908 * 3) No more VMAs can be found:
909 *
910 * find_vma(mm, 16k - 1) will return NULL.
911 * No more things to do, just break.
912 *
913 * 4) (last_vma_end - 1) is the middle of a vma (VMA'):
914 *
915 * find_vma(mm, 16k - 1) will return VMA' whose range
916 * contains last_vma_end.
917 * Iterate VMA' from last_vma_end.
918 */
919 vma = find_vma(mm, last_vma_end - 1);
920 /* Case 3 above */
921 if (!vma)
922 break;
923
924 /* Case 1 above */
925 if (vma->vm_start >= last_vma_end)
926 continue;
927
928 /* Case 4 above */
929 if (vma->vm_end > last_vma_end)
930 smap_gather_stats(vma, &mss, last_vma_end);
931 }
932 /* Case 2 above */
933 vma = vma->vm_next;
934 }
935
936 show_vma_header_prefix(m, priv->mm->mmap->vm_start,
937 last_vma_end, 0, 0, 0, 0);
938 seq_pad(m, ' ');
939 seq_puts(m, "[rollup]\n");
940
941 __show_smap(m, &mss, true);
942
943 release_task_mempolicy(priv);
944 mmap_read_unlock(mm);
945
946out_put_mm:
947 mmput(mm);
948out_put_task:
949 put_task_struct(priv->task);
950 priv->task = NULL;
951
952 return ret;
953}
954#undef SEQ_PUT_DEC
955
956static const struct seq_operations proc_pid_smaps_op = {
957 .start = m_start,
958 .next = m_next,
959 .stop = m_stop,
960 .show = show_smap
961};
962
963static int pid_smaps_open(struct inode *inode, struct file *file)
964{
965 return do_maps_open(inode, file, &proc_pid_smaps_op);
966}
967
968static int smaps_rollup_open(struct inode *inode, struct file *file)
969{
970 int ret;
971 struct proc_maps_private *priv;
972
973 priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT);
974 if (!priv)
975 return -ENOMEM;
976
977 ret = single_open(file, show_smaps_rollup, priv);
978 if (ret)
979 goto out_free;
980
981 priv->inode = inode;
982 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
983 if (IS_ERR(priv->mm)) {
984 ret = PTR_ERR(priv->mm);
985
986 single_release(inode, file);
987 goto out_free;
988 }
989
990 return 0;
991
992out_free:
993 kfree(priv);
994 return ret;
995}
996
997static int smaps_rollup_release(struct inode *inode, struct file *file)
998{
999 struct seq_file *seq = file->private_data;
1000 struct proc_maps_private *priv = seq->private;
1001
1002 if (priv->mm)
1003 mmdrop(priv->mm);
1004
1005 kfree(priv);
1006 return single_release(inode, file);
1007}
1008
1009const struct file_operations proc_pid_smaps_operations = {
1010 .open = pid_smaps_open,
1011 .read = seq_read,
1012 .llseek = seq_lseek,
1013 .release = proc_map_release,
1014};
1015
1016const struct file_operations proc_pid_smaps_rollup_operations = {
1017 .open = smaps_rollup_open,
1018 .read = seq_read,
1019 .llseek = seq_lseek,
1020 .release = smaps_rollup_release,
1021};
1022
1023enum clear_refs_types {
1024 CLEAR_REFS_ALL = 1,
1025 CLEAR_REFS_ANON,
1026 CLEAR_REFS_MAPPED,
1027 CLEAR_REFS_SOFT_DIRTY,
1028 CLEAR_REFS_MM_HIWATER_RSS,
1029 CLEAR_REFS_LAST,
1030};
1031
1032struct clear_refs_private {
1033 enum clear_refs_types type;
1034};
1035
1036#ifdef CONFIG_MEM_SOFT_DIRTY
1037
1038static inline bool pte_is_pinned(struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1039{
1040 struct page *page;
1041
1042 if (!pte_write(pte))
1043 return false;
1044 if (!is_cow_mapping(vma->vm_flags))
1045 return false;
1046 if (likely(!test_bit(MMF_HAS_PINNED, &vma->vm_mm->flags)))
1047 return false;
1048 page = vm_normal_page(vma, addr, pte);
1049 if (!page)
1050 return false;
1051 return page_maybe_dma_pinned(page);
1052}
1053
1054static inline void clear_soft_dirty(struct vm_area_struct *vma,
1055 unsigned long addr, pte_t *pte)
1056{
1057 /*
1058 * The soft-dirty tracker uses #PF-s to catch writes
1059 * to pages, so write-protect the pte as well. See the
1060 * Documentation/admin-guide/mm/soft-dirty.rst for full description
1061 * of how soft-dirty works.
1062 */
1063 pte_t ptent = *pte;
1064
1065 if (pte_present(ptent)) {
1066 pte_t old_pte;
1067
1068 if (pte_is_pinned(vma, addr, ptent))
1069 return;
1070 old_pte = ptep_modify_prot_start(vma, addr, pte);
1071 ptent = pte_wrprotect(old_pte);
1072 ptent = pte_clear_soft_dirty(ptent);
1073 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent);
1074 } else if (is_swap_pte(ptent)) {
1075 ptent = pte_swp_clear_soft_dirty(ptent);
1076 set_pte_at(vma->vm_mm, addr, pte, ptent);
1077 }
1078}
1079#else
1080static inline void clear_soft_dirty(struct vm_area_struct *vma,
1081 unsigned long addr, pte_t *pte)
1082{
1083}
1084#endif
1085
1086#if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
1087static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1088 unsigned long addr, pmd_t *pmdp)
1089{
1090 pmd_t old, pmd = *pmdp;
1091
1092 if (pmd_present(pmd)) {
1093 /* See comment in change_huge_pmd() */
1094 old = pmdp_invalidate(vma, addr, pmdp);
1095 if (pmd_dirty(old))
1096 pmd = pmd_mkdirty(pmd);
1097 if (pmd_young(old))
1098 pmd = pmd_mkyoung(pmd);
1099
1100 pmd = pmd_wrprotect(pmd);
1101 pmd = pmd_clear_soft_dirty(pmd);
1102
1103 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1104 } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1105 pmd = pmd_swp_clear_soft_dirty(pmd);
1106 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1107 }
1108}
1109#else
1110static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1111 unsigned long addr, pmd_t *pmdp)
1112{
1113}
1114#endif
1115
1116static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
1117 unsigned long end, struct mm_walk *walk)
1118{
1119 struct clear_refs_private *cp = walk->private;
1120 struct vm_area_struct *vma = walk->vma;
1121 pte_t *pte, ptent;
1122 spinlock_t *ptl;
1123 struct page *page;
1124
1125 ptl = pmd_trans_huge_lock(pmd, vma);
1126 if (ptl) {
1127 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1128 clear_soft_dirty_pmd(vma, addr, pmd);
1129 goto out;
1130 }
1131
1132 if (!pmd_present(*pmd))
1133 goto out;
1134
1135 page = pmd_page(*pmd);
1136
1137 /* Clear accessed and referenced bits. */
1138 pmdp_test_and_clear_young(vma, addr, pmd);
1139 test_and_clear_page_young(page);
1140 ClearPageReferenced(page);
1141out:
1142 spin_unlock(ptl);
1143 return 0;
1144 }
1145
1146 if (pmd_trans_unstable(pmd))
1147 return 0;
1148
1149 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1150 for (; addr != end; pte++, addr += PAGE_SIZE) {
1151 ptent = *pte;
1152
1153 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1154 clear_soft_dirty(vma, addr, pte);
1155 continue;
1156 }
1157
1158 if (!pte_present(ptent))
1159 continue;
1160
1161 page = vm_normal_page(vma, addr, ptent);
1162 if (!page)
1163 continue;
1164
1165 /* Clear accessed and referenced bits. */
1166 ptep_test_and_clear_young(vma, addr, pte);
1167 test_and_clear_page_young(page);
1168 ClearPageReferenced(page);
1169 }
1170 pte_unmap_unlock(pte - 1, ptl);
1171 cond_resched();
1172 return 0;
1173}
1174
1175static int clear_refs_test_walk(unsigned long start, unsigned long end,
1176 struct mm_walk *walk)
1177{
1178 struct clear_refs_private *cp = walk->private;
1179 struct vm_area_struct *vma = walk->vma;
1180
1181 if (vma->vm_flags & VM_PFNMAP)
1182 return 1;
1183
1184 /*
1185 * Writing 1 to /proc/pid/clear_refs affects all pages.
1186 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1187 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1188 * Writing 4 to /proc/pid/clear_refs affects all pages.
1189 */
1190 if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1191 return 1;
1192 if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1193 return 1;
1194 return 0;
1195}
1196
1197static const struct mm_walk_ops clear_refs_walk_ops = {
1198 .pmd_entry = clear_refs_pte_range,
1199 .test_walk = clear_refs_test_walk,
1200};
1201
1202static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1203 size_t count, loff_t *ppos)
1204{
1205 struct task_struct *task;
1206 char buffer[PROC_NUMBUF];
1207 struct mm_struct *mm;
1208 struct vm_area_struct *vma;
1209 enum clear_refs_types type;
1210 int itype;
1211 int rv;
1212
1213 memset(buffer, 0, sizeof(buffer));
1214 if (count > sizeof(buffer) - 1)
1215 count = sizeof(buffer) - 1;
1216 if (copy_from_user(buffer, buf, count))
1217 return -EFAULT;
1218 rv = kstrtoint(strstrip(buffer), 10, &itype);
1219 if (rv < 0)
1220 return rv;
1221 type = (enum clear_refs_types)itype;
1222 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1223 return -EINVAL;
1224
1225 task = get_proc_task(file_inode(file));
1226 if (!task)
1227 return -ESRCH;
1228 mm = get_task_mm(task);
1229 if (mm) {
1230 struct mmu_notifier_range range;
1231 struct clear_refs_private cp = {
1232 .type = type,
1233 };
1234
1235 if (mmap_write_lock_killable(mm)) {
1236 count = -EINTR;
1237 goto out_mm;
1238 }
1239 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1240 /*
1241 * Writing 5 to /proc/pid/clear_refs resets the peak
1242 * resident set size to this mm's current rss value.
1243 */
1244 reset_mm_hiwater_rss(mm);
1245 goto out_unlock;
1246 }
1247
1248 if (type == CLEAR_REFS_SOFT_DIRTY) {
1249 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1250 if (!(vma->vm_flags & VM_SOFTDIRTY))
1251 continue;
1252 vma->vm_flags &= ~VM_SOFTDIRTY;
1253 vma_set_page_prot(vma);
1254 }
1255
1256 inc_tlb_flush_pending(mm);
1257 mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY,
1258 0, NULL, mm, 0, -1UL);
1259 mmu_notifier_invalidate_range_start(&range);
1260 }
1261 walk_page_range(mm, 0, mm->highest_vm_end, &clear_refs_walk_ops,
1262 &cp);
1263 if (type == CLEAR_REFS_SOFT_DIRTY) {
1264 mmu_notifier_invalidate_range_end(&range);
1265 flush_tlb_mm(mm);
1266 dec_tlb_flush_pending(mm);
1267 }
1268out_unlock:
1269 mmap_write_unlock(mm);
1270out_mm:
1271 mmput(mm);
1272 }
1273 put_task_struct(task);
1274
1275 return count;
1276}
1277
1278const struct file_operations proc_clear_refs_operations = {
1279 .write = clear_refs_write,
1280 .llseek = noop_llseek,
1281};
1282
1283typedef struct {
1284 u64 pme;
1285} pagemap_entry_t;
1286
1287struct pagemapread {
1288 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */
1289 pagemap_entry_t *buffer;
1290 bool show_pfn;
1291};
1292
1293#define PAGEMAP_WALK_SIZE (PMD_SIZE)
1294#define PAGEMAP_WALK_MASK (PMD_MASK)
1295
1296#define PM_ENTRY_BYTES sizeof(pagemap_entry_t)
1297#define PM_PFRAME_BITS 55
1298#define PM_PFRAME_MASK GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1299#define PM_SOFT_DIRTY BIT_ULL(55)
1300#define PM_MMAP_EXCLUSIVE BIT_ULL(56)
1301#define PM_UFFD_WP BIT_ULL(57)
1302#define PM_FILE BIT_ULL(61)
1303#define PM_SWAP BIT_ULL(62)
1304#define PM_PRESENT BIT_ULL(63)
1305
1306#define PM_END_OF_BUFFER 1
1307
1308static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1309{
1310 return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1311}
1312
1313static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1314 struct pagemapread *pm)
1315{
1316 pm->buffer[pm->pos++] = *pme;
1317 if (pm->pos >= pm->len)
1318 return PM_END_OF_BUFFER;
1319 return 0;
1320}
1321
1322static int pagemap_pte_hole(unsigned long start, unsigned long end,
1323 __always_unused int depth, struct mm_walk *walk)
1324{
1325 struct pagemapread *pm = walk->private;
1326 unsigned long addr = start;
1327 int err = 0;
1328
1329 while (addr < end) {
1330 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1331 pagemap_entry_t pme = make_pme(0, 0);
1332 /* End of address space hole, which we mark as non-present. */
1333 unsigned long hole_end;
1334
1335 if (vma)
1336 hole_end = min(end, vma->vm_start);
1337 else
1338 hole_end = end;
1339
1340 for (; addr < hole_end; addr += PAGE_SIZE) {
1341 err = add_to_pagemap(addr, &pme, pm);
1342 if (err)
1343 goto out;
1344 }
1345
1346 if (!vma)
1347 break;
1348
1349 /* Addresses in the VMA. */
1350 if (vma->vm_flags & VM_SOFTDIRTY)
1351 pme = make_pme(0, PM_SOFT_DIRTY);
1352 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1353 err = add_to_pagemap(addr, &pme, pm);
1354 if (err)
1355 goto out;
1356 }
1357 }
1358out:
1359 return err;
1360}
1361
1362static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1363 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1364{
1365 u64 frame = 0, flags = 0;
1366 struct page *page = NULL;
1367
1368 if (pte_present(pte)) {
1369 if (pm->show_pfn)
1370 frame = pte_pfn(pte);
1371 flags |= PM_PRESENT;
1372 page = vm_normal_page(vma, addr, pte);
1373 if (pte_soft_dirty(pte))
1374 flags |= PM_SOFT_DIRTY;
1375 if (pte_uffd_wp(pte))
1376 flags |= PM_UFFD_WP;
1377 } else if (is_swap_pte(pte)) {
1378 swp_entry_t entry;
1379 if (pte_swp_soft_dirty(pte))
1380 flags |= PM_SOFT_DIRTY;
1381 if (pte_swp_uffd_wp(pte))
1382 flags |= PM_UFFD_WP;
1383 entry = pte_to_swp_entry(pte);
1384 if (pm->show_pfn)
1385 frame = swp_type(entry) |
1386 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1387 flags |= PM_SWAP;
1388 if (is_pfn_swap_entry(entry))
1389 page = pfn_swap_entry_to_page(entry);
1390 }
1391
1392 if (page && !PageAnon(page))
1393 flags |= PM_FILE;
1394 if (page && page_mapcount(page) == 1)
1395 flags |= PM_MMAP_EXCLUSIVE;
1396 if (vma->vm_flags & VM_SOFTDIRTY)
1397 flags |= PM_SOFT_DIRTY;
1398
1399 return make_pme(frame, flags);
1400}
1401
1402static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1403 struct mm_walk *walk)
1404{
1405 struct vm_area_struct *vma = walk->vma;
1406 struct pagemapread *pm = walk->private;
1407 spinlock_t *ptl;
1408 pte_t *pte, *orig_pte;
1409 int err = 0;
1410
1411#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1412 ptl = pmd_trans_huge_lock(pmdp, vma);
1413 if (ptl) {
1414 u64 flags = 0, frame = 0;
1415 pmd_t pmd = *pmdp;
1416 struct page *page = NULL;
1417
1418 if (vma->vm_flags & VM_SOFTDIRTY)
1419 flags |= PM_SOFT_DIRTY;
1420
1421 if (pmd_present(pmd)) {
1422 page = pmd_page(pmd);
1423
1424 flags |= PM_PRESENT;
1425 if (pmd_soft_dirty(pmd))
1426 flags |= PM_SOFT_DIRTY;
1427 if (pmd_uffd_wp(pmd))
1428 flags |= PM_UFFD_WP;
1429 if (pm->show_pfn)
1430 frame = pmd_pfn(pmd) +
1431 ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1432 }
1433#ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1434 else if (is_swap_pmd(pmd)) {
1435 swp_entry_t entry = pmd_to_swp_entry(pmd);
1436 unsigned long offset;
1437
1438 if (pm->show_pfn) {
1439 offset = swp_offset(entry) +
1440 ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1441 frame = swp_type(entry) |
1442 (offset << MAX_SWAPFILES_SHIFT);
1443 }
1444 flags |= PM_SWAP;
1445 if (pmd_swp_soft_dirty(pmd))
1446 flags |= PM_SOFT_DIRTY;
1447 if (pmd_swp_uffd_wp(pmd))
1448 flags |= PM_UFFD_WP;
1449 VM_BUG_ON(!is_pmd_migration_entry(pmd));
1450 page = pfn_swap_entry_to_page(entry);
1451 }
1452#endif
1453
1454 if (page && page_mapcount(page) == 1)
1455 flags |= PM_MMAP_EXCLUSIVE;
1456
1457 for (; addr != end; addr += PAGE_SIZE) {
1458 pagemap_entry_t pme = make_pme(frame, flags);
1459
1460 err = add_to_pagemap(addr, &pme, pm);
1461 if (err)
1462 break;
1463 if (pm->show_pfn) {
1464 if (flags & PM_PRESENT)
1465 frame++;
1466 else if (flags & PM_SWAP)
1467 frame += (1 << MAX_SWAPFILES_SHIFT);
1468 }
1469 }
1470 spin_unlock(ptl);
1471 return err;
1472 }
1473
1474 if (pmd_trans_unstable(pmdp))
1475 return 0;
1476#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1477
1478 /*
1479 * We can assume that @vma always points to a valid one and @end never
1480 * goes beyond vma->vm_end.
1481 */
1482 orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1483 for (; addr < end; pte++, addr += PAGE_SIZE) {
1484 pagemap_entry_t pme;
1485
1486 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1487 err = add_to_pagemap(addr, &pme, pm);
1488 if (err)
1489 break;
1490 }
1491 pte_unmap_unlock(orig_pte, ptl);
1492
1493 cond_resched();
1494
1495 return err;
1496}
1497
1498#ifdef CONFIG_HUGETLB_PAGE
1499/* This function walks within one hugetlb entry in the single call */
1500static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1501 unsigned long addr, unsigned long end,
1502 struct mm_walk *walk)
1503{
1504 struct pagemapread *pm = walk->private;
1505 struct vm_area_struct *vma = walk->vma;
1506 u64 flags = 0, frame = 0;
1507 int err = 0;
1508 pte_t pte;
1509
1510 if (vma->vm_flags & VM_SOFTDIRTY)
1511 flags |= PM_SOFT_DIRTY;
1512
1513 pte = huge_ptep_get(ptep);
1514 if (pte_present(pte)) {
1515 struct page *page = pte_page(pte);
1516
1517 if (!PageAnon(page))
1518 flags |= PM_FILE;
1519
1520 if (page_mapcount(page) == 1)
1521 flags |= PM_MMAP_EXCLUSIVE;
1522
1523 flags |= PM_PRESENT;
1524 if (pm->show_pfn)
1525 frame = pte_pfn(pte) +
1526 ((addr & ~hmask) >> PAGE_SHIFT);
1527 }
1528
1529 for (; addr != end; addr += PAGE_SIZE) {
1530 pagemap_entry_t pme = make_pme(frame, flags);
1531
1532 err = add_to_pagemap(addr, &pme, pm);
1533 if (err)
1534 return err;
1535 if (pm->show_pfn && (flags & PM_PRESENT))
1536 frame++;
1537 }
1538
1539 cond_resched();
1540
1541 return err;
1542}
1543#else
1544#define pagemap_hugetlb_range NULL
1545#endif /* HUGETLB_PAGE */
1546
1547static const struct mm_walk_ops pagemap_ops = {
1548 .pmd_entry = pagemap_pmd_range,
1549 .pte_hole = pagemap_pte_hole,
1550 .hugetlb_entry = pagemap_hugetlb_range,
1551};
1552
1553/*
1554 * /proc/pid/pagemap - an array mapping virtual pages to pfns
1555 *
1556 * For each page in the address space, this file contains one 64-bit entry
1557 * consisting of the following:
1558 *
1559 * Bits 0-54 page frame number (PFN) if present
1560 * Bits 0-4 swap type if swapped
1561 * Bits 5-54 swap offset if swapped
1562 * Bit 55 pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst)
1563 * Bit 56 page exclusively mapped
1564 * Bits 57-60 zero
1565 * Bit 61 page is file-page or shared-anon
1566 * Bit 62 page swapped
1567 * Bit 63 page present
1568 *
1569 * If the page is not present but in swap, then the PFN contains an
1570 * encoding of the swap file number and the page's offset into the
1571 * swap. Unmapped pages return a null PFN. This allows determining
1572 * precisely which pages are mapped (or in swap) and comparing mapped
1573 * pages between processes.
1574 *
1575 * Efficient users of this interface will use /proc/pid/maps to
1576 * determine which areas of memory are actually mapped and llseek to
1577 * skip over unmapped regions.
1578 */
1579static ssize_t pagemap_read(struct file *file, char __user *buf,
1580 size_t count, loff_t *ppos)
1581{
1582 struct mm_struct *mm = file->private_data;
1583 struct pagemapread pm;
1584 unsigned long src;
1585 unsigned long svpfn;
1586 unsigned long start_vaddr;
1587 unsigned long end_vaddr;
1588 int ret = 0, copied = 0;
1589
1590 if (!mm || !mmget_not_zero(mm))
1591 goto out;
1592
1593 ret = -EINVAL;
1594 /* file position must be aligned */
1595 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1596 goto out_mm;
1597
1598 ret = 0;
1599 if (!count)
1600 goto out_mm;
1601
1602 /* do not disclose physical addresses: attack vector */
1603 pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1604
1605 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1606 pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL);
1607 ret = -ENOMEM;
1608 if (!pm.buffer)
1609 goto out_mm;
1610
1611 src = *ppos;
1612 svpfn = src / PM_ENTRY_BYTES;
1613 end_vaddr = mm->task_size;
1614
1615 /* watch out for wraparound */
1616 start_vaddr = end_vaddr;
1617 if (svpfn <= (ULONG_MAX >> PAGE_SHIFT))
1618 start_vaddr = untagged_addr(svpfn << PAGE_SHIFT);
1619
1620 /* Ensure the address is inside the task */
1621 if (start_vaddr > mm->task_size)
1622 start_vaddr = end_vaddr;
1623
1624 /*
1625 * The odds are that this will stop walking way
1626 * before end_vaddr, because the length of the
1627 * user buffer is tracked in "pm", and the walk
1628 * will stop when we hit the end of the buffer.
1629 */
1630 ret = 0;
1631 while (count && (start_vaddr < end_vaddr)) {
1632 int len;
1633 unsigned long end;
1634
1635 pm.pos = 0;
1636 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1637 /* overflow ? */
1638 if (end < start_vaddr || end > end_vaddr)
1639 end = end_vaddr;
1640 ret = mmap_read_lock_killable(mm);
1641 if (ret)
1642 goto out_free;
1643 ret = walk_page_range(mm, start_vaddr, end, &pagemap_ops, &pm);
1644 mmap_read_unlock(mm);
1645 start_vaddr = end;
1646
1647 len = min(count, PM_ENTRY_BYTES * pm.pos);
1648 if (copy_to_user(buf, pm.buffer, len)) {
1649 ret = -EFAULT;
1650 goto out_free;
1651 }
1652 copied += len;
1653 buf += len;
1654 count -= len;
1655 }
1656 *ppos += copied;
1657 if (!ret || ret == PM_END_OF_BUFFER)
1658 ret = copied;
1659
1660out_free:
1661 kfree(pm.buffer);
1662out_mm:
1663 mmput(mm);
1664out:
1665 return ret;
1666}
1667
1668static int pagemap_open(struct inode *inode, struct file *file)
1669{
1670 struct mm_struct *mm;
1671
1672 mm = proc_mem_open(inode, PTRACE_MODE_READ);
1673 if (IS_ERR(mm))
1674 return PTR_ERR(mm);
1675 file->private_data = mm;
1676 return 0;
1677}
1678
1679static int pagemap_release(struct inode *inode, struct file *file)
1680{
1681 struct mm_struct *mm = file->private_data;
1682
1683 if (mm)
1684 mmdrop(mm);
1685 return 0;
1686}
1687
1688const struct file_operations proc_pagemap_operations = {
1689 .llseek = mem_lseek, /* borrow this */
1690 .read = pagemap_read,
1691 .open = pagemap_open,
1692 .release = pagemap_release,
1693};
1694#endif /* CONFIG_PROC_PAGE_MONITOR */
1695
1696#ifdef CONFIG_NUMA
1697
1698struct numa_maps {
1699 unsigned long pages;
1700 unsigned long anon;
1701 unsigned long active;
1702 unsigned long writeback;
1703 unsigned long mapcount_max;
1704 unsigned long dirty;
1705 unsigned long swapcache;
1706 unsigned long node[MAX_NUMNODES];
1707};
1708
1709struct numa_maps_private {
1710 struct proc_maps_private proc_maps;
1711 struct numa_maps md;
1712};
1713
1714static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1715 unsigned long nr_pages)
1716{
1717 int count = page_mapcount(page);
1718
1719 md->pages += nr_pages;
1720 if (pte_dirty || PageDirty(page))
1721 md->dirty += nr_pages;
1722
1723 if (PageSwapCache(page))
1724 md->swapcache += nr_pages;
1725
1726 if (PageActive(page) || PageUnevictable(page))
1727 md->active += nr_pages;
1728
1729 if (PageWriteback(page))
1730 md->writeback += nr_pages;
1731
1732 if (PageAnon(page))
1733 md->anon += nr_pages;
1734
1735 if (count > md->mapcount_max)
1736 md->mapcount_max = count;
1737
1738 md->node[page_to_nid(page)] += nr_pages;
1739}
1740
1741static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1742 unsigned long addr)
1743{
1744 struct page *page;
1745 int nid;
1746
1747 if (!pte_present(pte))
1748 return NULL;
1749
1750 page = vm_normal_page(vma, addr, pte);
1751 if (!page)
1752 return NULL;
1753
1754 if (PageReserved(page))
1755 return NULL;
1756
1757 nid = page_to_nid(page);
1758 if (!node_isset(nid, node_states[N_MEMORY]))
1759 return NULL;
1760
1761 return page;
1762}
1763
1764#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1765static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1766 struct vm_area_struct *vma,
1767 unsigned long addr)
1768{
1769 struct page *page;
1770 int nid;
1771
1772 if (!pmd_present(pmd))
1773 return NULL;
1774
1775 page = vm_normal_page_pmd(vma, addr, pmd);
1776 if (!page)
1777 return NULL;
1778
1779 if (PageReserved(page))
1780 return NULL;
1781
1782 nid = page_to_nid(page);
1783 if (!node_isset(nid, node_states[N_MEMORY]))
1784 return NULL;
1785
1786 return page;
1787}
1788#endif
1789
1790static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1791 unsigned long end, struct mm_walk *walk)
1792{
1793 struct numa_maps *md = walk->private;
1794 struct vm_area_struct *vma = walk->vma;
1795 spinlock_t *ptl;
1796 pte_t *orig_pte;
1797 pte_t *pte;
1798
1799#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1800 ptl = pmd_trans_huge_lock(pmd, vma);
1801 if (ptl) {
1802 struct page *page;
1803
1804 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1805 if (page)
1806 gather_stats(page, md, pmd_dirty(*pmd),
1807 HPAGE_PMD_SIZE/PAGE_SIZE);
1808 spin_unlock(ptl);
1809 return 0;
1810 }
1811
1812 if (pmd_trans_unstable(pmd))
1813 return 0;
1814#endif
1815 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1816 do {
1817 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1818 if (!page)
1819 continue;
1820 gather_stats(page, md, pte_dirty(*pte), 1);
1821
1822 } while (pte++, addr += PAGE_SIZE, addr != end);
1823 pte_unmap_unlock(orig_pte, ptl);
1824 cond_resched();
1825 return 0;
1826}
1827#ifdef CONFIG_HUGETLB_PAGE
1828static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1829 unsigned long addr, unsigned long end, struct mm_walk *walk)
1830{
1831 pte_t huge_pte = huge_ptep_get(pte);
1832 struct numa_maps *md;
1833 struct page *page;
1834
1835 if (!pte_present(huge_pte))
1836 return 0;
1837
1838 page = pte_page(huge_pte);
1839 if (!page)
1840 return 0;
1841
1842 md = walk->private;
1843 gather_stats(page, md, pte_dirty(huge_pte), 1);
1844 return 0;
1845}
1846
1847#else
1848static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1849 unsigned long addr, unsigned long end, struct mm_walk *walk)
1850{
1851 return 0;
1852}
1853#endif
1854
1855static const struct mm_walk_ops show_numa_ops = {
1856 .hugetlb_entry = gather_hugetlb_stats,
1857 .pmd_entry = gather_pte_stats,
1858};
1859
1860/*
1861 * Display pages allocated per node and memory policy via /proc.
1862 */
1863static int show_numa_map(struct seq_file *m, void *v)
1864{
1865 struct numa_maps_private *numa_priv = m->private;
1866 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1867 struct vm_area_struct *vma = v;
1868 struct numa_maps *md = &numa_priv->md;
1869 struct file *file = vma->vm_file;
1870 struct mm_struct *mm = vma->vm_mm;
1871 struct mempolicy *pol;
1872 char buffer[64];
1873 int nid;
1874
1875 if (!mm)
1876 return 0;
1877
1878 /* Ensure we start with an empty set of numa_maps statistics. */
1879 memset(md, 0, sizeof(*md));
1880
1881 pol = __get_vma_policy(vma, vma->vm_start);
1882 if (pol) {
1883 mpol_to_str(buffer, sizeof(buffer), pol);
1884 mpol_cond_put(pol);
1885 } else {
1886 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1887 }
1888
1889 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1890
1891 if (file) {
1892 seq_puts(m, " file=");
1893 seq_file_path(m, file, "\n\t= ");
1894 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1895 seq_puts(m, " heap");
1896 } else if (is_stack(vma)) {
1897 seq_puts(m, " stack");
1898 }
1899
1900 if (is_vm_hugetlb_page(vma))
1901 seq_puts(m, " huge");
1902
1903 /* mmap_lock is held by m_start */
1904 walk_page_vma(vma, &show_numa_ops, md);
1905
1906 if (!md->pages)
1907 goto out;
1908
1909 if (md->anon)
1910 seq_printf(m, " anon=%lu", md->anon);
1911
1912 if (md->dirty)
1913 seq_printf(m, " dirty=%lu", md->dirty);
1914
1915 if (md->pages != md->anon && md->pages != md->dirty)
1916 seq_printf(m, " mapped=%lu", md->pages);
1917
1918 if (md->mapcount_max > 1)
1919 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1920
1921 if (md->swapcache)
1922 seq_printf(m, " swapcache=%lu", md->swapcache);
1923
1924 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1925 seq_printf(m, " active=%lu", md->active);
1926
1927 if (md->writeback)
1928 seq_printf(m, " writeback=%lu", md->writeback);
1929
1930 for_each_node_state(nid, N_MEMORY)
1931 if (md->node[nid])
1932 seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1933
1934 seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1935out:
1936 seq_putc(m, '\n');
1937 return 0;
1938}
1939
1940static const struct seq_operations proc_pid_numa_maps_op = {
1941 .start = m_start,
1942 .next = m_next,
1943 .stop = m_stop,
1944 .show = show_numa_map,
1945};
1946
1947static int pid_numa_maps_open(struct inode *inode, struct file *file)
1948{
1949 return proc_maps_open(inode, file, &proc_pid_numa_maps_op,
1950 sizeof(struct numa_maps_private));
1951}
1952
1953const struct file_operations proc_pid_numa_maps_operations = {
1954 .open = pid_numa_maps_open,
1955 .read = seq_read,
1956 .llseek = seq_lseek,
1957 .release = proc_map_release,
1958};
1959
1960#endif /* CONFIG_NUMA */