Loading...
1#include <linux/mm.h>
2#include <linux/vmacache.h>
3#include <linux/hugetlb.h>
4#include <linux/huge_mm.h>
5#include <linux/mount.h>
6#include <linux/seq_file.h>
7#include <linux/highmem.h>
8#include <linux/ptrace.h>
9#include <linux/slab.h>
10#include <linux/pagemap.h>
11#include <linux/mempolicy.h>
12#include <linux/rmap.h>
13#include <linux/swap.h>
14#include <linux/swapops.h>
15#include <linux/mmu_notifier.h>
16#include <linux/page_idle.h>
17#include <linux/shmem_fs.h>
18
19#include <asm/elf.h>
20#include <asm/uaccess.h>
21#include <asm/tlbflush.h>
22#include "internal.h"
23
24void task_mem(struct seq_file *m, struct mm_struct *mm)
25{
26 unsigned long text, lib, swap, ptes, pmds, anon, file, shmem;
27 unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
28
29 anon = get_mm_counter(mm, MM_ANONPAGES);
30 file = get_mm_counter(mm, MM_FILEPAGES);
31 shmem = get_mm_counter(mm, MM_SHMEMPAGES);
32
33 /*
34 * Note: to minimize their overhead, mm maintains hiwater_vm and
35 * hiwater_rss only when about to *lower* total_vm or rss. Any
36 * collector of these hiwater stats must therefore get total_vm
37 * and rss too, which will usually be the higher. Barriers? not
38 * worth the effort, such snapshots can always be inconsistent.
39 */
40 hiwater_vm = total_vm = mm->total_vm;
41 if (hiwater_vm < mm->hiwater_vm)
42 hiwater_vm = mm->hiwater_vm;
43 hiwater_rss = total_rss = anon + file + shmem;
44 if (hiwater_rss < mm->hiwater_rss)
45 hiwater_rss = mm->hiwater_rss;
46
47 text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
48 lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
49 swap = get_mm_counter(mm, MM_SWAPENTS);
50 ptes = PTRS_PER_PTE * sizeof(pte_t) * atomic_long_read(&mm->nr_ptes);
51 pmds = PTRS_PER_PMD * sizeof(pmd_t) * mm_nr_pmds(mm);
52 seq_printf(m,
53 "VmPeak:\t%8lu kB\n"
54 "VmSize:\t%8lu kB\n"
55 "VmLck:\t%8lu kB\n"
56 "VmPin:\t%8lu kB\n"
57 "VmHWM:\t%8lu kB\n"
58 "VmRSS:\t%8lu kB\n"
59 "RssAnon:\t%8lu kB\n"
60 "RssFile:\t%8lu kB\n"
61 "RssShmem:\t%8lu kB\n"
62 "VmData:\t%8lu kB\n"
63 "VmStk:\t%8lu kB\n"
64 "VmExe:\t%8lu kB\n"
65 "VmLib:\t%8lu kB\n"
66 "VmPTE:\t%8lu kB\n"
67 "VmPMD:\t%8lu kB\n"
68 "VmSwap:\t%8lu kB\n",
69 hiwater_vm << (PAGE_SHIFT-10),
70 total_vm << (PAGE_SHIFT-10),
71 mm->locked_vm << (PAGE_SHIFT-10),
72 mm->pinned_vm << (PAGE_SHIFT-10),
73 hiwater_rss << (PAGE_SHIFT-10),
74 total_rss << (PAGE_SHIFT-10),
75 anon << (PAGE_SHIFT-10),
76 file << (PAGE_SHIFT-10),
77 shmem << (PAGE_SHIFT-10),
78 mm->data_vm << (PAGE_SHIFT-10),
79 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
80 ptes >> 10,
81 pmds >> 10,
82 swap << (PAGE_SHIFT-10));
83 hugetlb_report_usage(m, mm);
84}
85
86unsigned long task_vsize(struct mm_struct *mm)
87{
88 return PAGE_SIZE * mm->total_vm;
89}
90
91unsigned long task_statm(struct mm_struct *mm,
92 unsigned long *shared, unsigned long *text,
93 unsigned long *data, unsigned long *resident)
94{
95 *shared = get_mm_counter(mm, MM_FILEPAGES) +
96 get_mm_counter(mm, MM_SHMEMPAGES);
97 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
98 >> PAGE_SHIFT;
99 *data = mm->data_vm + mm->stack_vm;
100 *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
101 return mm->total_vm;
102}
103
104#ifdef CONFIG_NUMA
105/*
106 * Save get_task_policy() for show_numa_map().
107 */
108static void hold_task_mempolicy(struct proc_maps_private *priv)
109{
110 struct task_struct *task = priv->task;
111
112 task_lock(task);
113 priv->task_mempolicy = get_task_policy(task);
114 mpol_get(priv->task_mempolicy);
115 task_unlock(task);
116}
117static void release_task_mempolicy(struct proc_maps_private *priv)
118{
119 mpol_put(priv->task_mempolicy);
120}
121#else
122static void hold_task_mempolicy(struct proc_maps_private *priv)
123{
124}
125static void release_task_mempolicy(struct proc_maps_private *priv)
126{
127}
128#endif
129
130static void vma_stop(struct proc_maps_private *priv)
131{
132 struct mm_struct *mm = priv->mm;
133
134 release_task_mempolicy(priv);
135 up_read(&mm->mmap_sem);
136 mmput(mm);
137}
138
139static struct vm_area_struct *
140m_next_vma(struct proc_maps_private *priv, struct vm_area_struct *vma)
141{
142 if (vma == priv->tail_vma)
143 return NULL;
144 return vma->vm_next ?: priv->tail_vma;
145}
146
147static void m_cache_vma(struct seq_file *m, struct vm_area_struct *vma)
148{
149 if (m->count < m->size) /* vma is copied successfully */
150 m->version = m_next_vma(m->private, vma) ? vma->vm_start : -1UL;
151}
152
153static void *m_start(struct seq_file *m, loff_t *ppos)
154{
155 struct proc_maps_private *priv = m->private;
156 unsigned long last_addr = m->version;
157 struct mm_struct *mm;
158 struct vm_area_struct *vma;
159 unsigned int pos = *ppos;
160
161 /* See m_cache_vma(). Zero at the start or after lseek. */
162 if (last_addr == -1UL)
163 return NULL;
164
165 priv->task = get_proc_task(priv->inode);
166 if (!priv->task)
167 return ERR_PTR(-ESRCH);
168
169 mm = priv->mm;
170 if (!mm || !atomic_inc_not_zero(&mm->mm_users))
171 return NULL;
172
173 down_read(&mm->mmap_sem);
174 hold_task_mempolicy(priv);
175 priv->tail_vma = get_gate_vma(mm);
176
177 if (last_addr) {
178 vma = find_vma(mm, last_addr);
179 if (vma && (vma = m_next_vma(priv, vma)))
180 return vma;
181 }
182
183 m->version = 0;
184 if (pos < mm->map_count) {
185 for (vma = mm->mmap; pos; pos--) {
186 m->version = vma->vm_start;
187 vma = vma->vm_next;
188 }
189 return vma;
190 }
191
192 /* we do not bother to update m->version in this case */
193 if (pos == mm->map_count && priv->tail_vma)
194 return priv->tail_vma;
195
196 vma_stop(priv);
197 return NULL;
198}
199
200static void *m_next(struct seq_file *m, void *v, loff_t *pos)
201{
202 struct proc_maps_private *priv = m->private;
203 struct vm_area_struct *next;
204
205 (*pos)++;
206 next = m_next_vma(priv, v);
207 if (!next)
208 vma_stop(priv);
209 return next;
210}
211
212static void m_stop(struct seq_file *m, void *v)
213{
214 struct proc_maps_private *priv = m->private;
215
216 if (!IS_ERR_OR_NULL(v))
217 vma_stop(priv);
218 if (priv->task) {
219 put_task_struct(priv->task);
220 priv->task = NULL;
221 }
222}
223
224static int proc_maps_open(struct inode *inode, struct file *file,
225 const struct seq_operations *ops, int psize)
226{
227 struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
228
229 if (!priv)
230 return -ENOMEM;
231
232 priv->inode = inode;
233 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
234 if (IS_ERR(priv->mm)) {
235 int err = PTR_ERR(priv->mm);
236
237 seq_release_private(inode, file);
238 return err;
239 }
240
241 return 0;
242}
243
244static int proc_map_release(struct inode *inode, struct file *file)
245{
246 struct seq_file *seq = file->private_data;
247 struct proc_maps_private *priv = seq->private;
248
249 if (priv->mm)
250 mmdrop(priv->mm);
251
252 return seq_release_private(inode, file);
253}
254
255static int do_maps_open(struct inode *inode, struct file *file,
256 const struct seq_operations *ops)
257{
258 return proc_maps_open(inode, file, ops,
259 sizeof(struct proc_maps_private));
260}
261
262/*
263 * Indicate if the VMA is a stack for the given task; for
264 * /proc/PID/maps that is the stack of the main task.
265 */
266static int is_stack(struct proc_maps_private *priv,
267 struct vm_area_struct *vma, int is_pid)
268{
269 int stack = 0;
270
271 if (is_pid) {
272 stack = vma->vm_start <= vma->vm_mm->start_stack &&
273 vma->vm_end >= vma->vm_mm->start_stack;
274 } else {
275 struct inode *inode = priv->inode;
276 struct task_struct *task;
277
278 rcu_read_lock();
279 task = pid_task(proc_pid(inode), PIDTYPE_PID);
280 if (task)
281 stack = vma_is_stack_for_task(vma, task);
282 rcu_read_unlock();
283 }
284 return stack;
285}
286
287static void
288show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
289{
290 struct mm_struct *mm = vma->vm_mm;
291 struct file *file = vma->vm_file;
292 struct proc_maps_private *priv = m->private;
293 vm_flags_t flags = vma->vm_flags;
294 unsigned long ino = 0;
295 unsigned long long pgoff = 0;
296 unsigned long start, end;
297 dev_t dev = 0;
298 const char *name = NULL;
299
300 if (file) {
301 struct inode *inode = file_inode(vma->vm_file);
302 dev = inode->i_sb->s_dev;
303 ino = inode->i_ino;
304 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
305 }
306
307 /* We don't show the stack guard page in /proc/maps */
308 start = vma->vm_start;
309 if (stack_guard_page_start(vma, start))
310 start += PAGE_SIZE;
311 end = vma->vm_end;
312 if (stack_guard_page_end(vma, end))
313 end -= PAGE_SIZE;
314
315 seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
316 seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu ",
317 start,
318 end,
319 flags & VM_READ ? 'r' : '-',
320 flags & VM_WRITE ? 'w' : '-',
321 flags & VM_EXEC ? 'x' : '-',
322 flags & VM_MAYSHARE ? 's' : 'p',
323 pgoff,
324 MAJOR(dev), MINOR(dev), ino);
325
326 /*
327 * Print the dentry name for named mappings, and a
328 * special [heap] marker for the heap:
329 */
330 if (file) {
331 seq_pad(m, ' ');
332 seq_file_path(m, file, "\n");
333 goto done;
334 }
335
336 if (vma->vm_ops && vma->vm_ops->name) {
337 name = vma->vm_ops->name(vma);
338 if (name)
339 goto done;
340 }
341
342 name = arch_vma_name(vma);
343 if (!name) {
344 if (!mm) {
345 name = "[vdso]";
346 goto done;
347 }
348
349 if (vma->vm_start <= mm->brk &&
350 vma->vm_end >= mm->start_brk) {
351 name = "[heap]";
352 goto done;
353 }
354
355 if (is_stack(priv, vma, is_pid))
356 name = "[stack]";
357 }
358
359done:
360 if (name) {
361 seq_pad(m, ' ');
362 seq_puts(m, name);
363 }
364 seq_putc(m, '\n');
365}
366
367static int show_map(struct seq_file *m, void *v, int is_pid)
368{
369 show_map_vma(m, v, is_pid);
370 m_cache_vma(m, v);
371 return 0;
372}
373
374static int show_pid_map(struct seq_file *m, void *v)
375{
376 return show_map(m, v, 1);
377}
378
379static int show_tid_map(struct seq_file *m, void *v)
380{
381 return show_map(m, v, 0);
382}
383
384static const struct seq_operations proc_pid_maps_op = {
385 .start = m_start,
386 .next = m_next,
387 .stop = m_stop,
388 .show = show_pid_map
389};
390
391static const struct seq_operations proc_tid_maps_op = {
392 .start = m_start,
393 .next = m_next,
394 .stop = m_stop,
395 .show = show_tid_map
396};
397
398static int pid_maps_open(struct inode *inode, struct file *file)
399{
400 return do_maps_open(inode, file, &proc_pid_maps_op);
401}
402
403static int tid_maps_open(struct inode *inode, struct file *file)
404{
405 return do_maps_open(inode, file, &proc_tid_maps_op);
406}
407
408const struct file_operations proc_pid_maps_operations = {
409 .open = pid_maps_open,
410 .read = seq_read,
411 .llseek = seq_lseek,
412 .release = proc_map_release,
413};
414
415const struct file_operations proc_tid_maps_operations = {
416 .open = tid_maps_open,
417 .read = seq_read,
418 .llseek = seq_lseek,
419 .release = proc_map_release,
420};
421
422/*
423 * Proportional Set Size(PSS): my share of RSS.
424 *
425 * PSS of a process is the count of pages it has in memory, where each
426 * page is divided by the number of processes sharing it. So if a
427 * process has 1000 pages all to itself, and 1000 shared with one other
428 * process, its PSS will be 1500.
429 *
430 * To keep (accumulated) division errors low, we adopt a 64bit
431 * fixed-point pss counter to minimize division errors. So (pss >>
432 * PSS_SHIFT) would be the real byte count.
433 *
434 * A shift of 12 before division means (assuming 4K page size):
435 * - 1M 3-user-pages add up to 8KB errors;
436 * - supports mapcount up to 2^24, or 16M;
437 * - supports PSS up to 2^52 bytes, or 4PB.
438 */
439#define PSS_SHIFT 12
440
441#ifdef CONFIG_PROC_PAGE_MONITOR
442struct mem_size_stats {
443 unsigned long resident;
444 unsigned long shared_clean;
445 unsigned long shared_dirty;
446 unsigned long private_clean;
447 unsigned long private_dirty;
448 unsigned long referenced;
449 unsigned long anonymous;
450 unsigned long anonymous_thp;
451 unsigned long swap;
452 unsigned long shared_hugetlb;
453 unsigned long private_hugetlb;
454 u64 pss;
455 u64 swap_pss;
456 bool check_shmem_swap;
457};
458
459static void smaps_account(struct mem_size_stats *mss, struct page *page,
460 bool compound, bool young, bool dirty)
461{
462 int i, nr = compound ? 1 << compound_order(page) : 1;
463 unsigned long size = nr * PAGE_SIZE;
464
465 if (PageAnon(page))
466 mss->anonymous += size;
467
468 mss->resident += size;
469 /* Accumulate the size in pages that have been accessed. */
470 if (young || page_is_young(page) || PageReferenced(page))
471 mss->referenced += size;
472
473 /*
474 * page_count(page) == 1 guarantees the page is mapped exactly once.
475 * If any subpage of the compound page mapped with PTE it would elevate
476 * page_count().
477 */
478 if (page_count(page) == 1) {
479 if (dirty || PageDirty(page))
480 mss->private_dirty += size;
481 else
482 mss->private_clean += size;
483 mss->pss += (u64)size << PSS_SHIFT;
484 return;
485 }
486
487 for (i = 0; i < nr; i++, page++) {
488 int mapcount = page_mapcount(page);
489
490 if (mapcount >= 2) {
491 if (dirty || PageDirty(page))
492 mss->shared_dirty += PAGE_SIZE;
493 else
494 mss->shared_clean += PAGE_SIZE;
495 mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount;
496 } else {
497 if (dirty || PageDirty(page))
498 mss->private_dirty += PAGE_SIZE;
499 else
500 mss->private_clean += PAGE_SIZE;
501 mss->pss += PAGE_SIZE << PSS_SHIFT;
502 }
503 }
504}
505
506#ifdef CONFIG_SHMEM
507static int smaps_pte_hole(unsigned long addr, unsigned long end,
508 struct mm_walk *walk)
509{
510 struct mem_size_stats *mss = walk->private;
511
512 mss->swap += shmem_partial_swap_usage(
513 walk->vma->vm_file->f_mapping, addr, end);
514
515 return 0;
516}
517#endif
518
519static void smaps_pte_entry(pte_t *pte, unsigned long addr,
520 struct mm_walk *walk)
521{
522 struct mem_size_stats *mss = walk->private;
523 struct vm_area_struct *vma = walk->vma;
524 struct page *page = NULL;
525
526 if (pte_present(*pte)) {
527 page = vm_normal_page(vma, addr, *pte);
528 } else if (is_swap_pte(*pte)) {
529 swp_entry_t swpent = pte_to_swp_entry(*pte);
530
531 if (!non_swap_entry(swpent)) {
532 int mapcount;
533
534 mss->swap += PAGE_SIZE;
535 mapcount = swp_swapcount(swpent);
536 if (mapcount >= 2) {
537 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
538
539 do_div(pss_delta, mapcount);
540 mss->swap_pss += pss_delta;
541 } else {
542 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
543 }
544 } else if (is_migration_entry(swpent))
545 page = migration_entry_to_page(swpent);
546 } else if (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap
547 && pte_none(*pte))) {
548 page = find_get_entry(vma->vm_file->f_mapping,
549 linear_page_index(vma, addr));
550 if (!page)
551 return;
552
553 if (radix_tree_exceptional_entry(page))
554 mss->swap += PAGE_SIZE;
555 else
556 put_page(page);
557
558 return;
559 }
560
561 if (!page)
562 return;
563
564 smaps_account(mss, page, false, pte_young(*pte), pte_dirty(*pte));
565}
566
567#ifdef CONFIG_TRANSPARENT_HUGEPAGE
568static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
569 struct mm_walk *walk)
570{
571 struct mem_size_stats *mss = walk->private;
572 struct vm_area_struct *vma = walk->vma;
573 struct page *page;
574
575 /* FOLL_DUMP will return -EFAULT on huge zero page */
576 page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
577 if (IS_ERR_OR_NULL(page))
578 return;
579 mss->anonymous_thp += HPAGE_PMD_SIZE;
580 smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd));
581}
582#else
583static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
584 struct mm_walk *walk)
585{
586}
587#endif
588
589static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
590 struct mm_walk *walk)
591{
592 struct vm_area_struct *vma = walk->vma;
593 pte_t *pte;
594 spinlock_t *ptl;
595
596 ptl = pmd_trans_huge_lock(pmd, vma);
597 if (ptl) {
598 smaps_pmd_entry(pmd, addr, walk);
599 spin_unlock(ptl);
600 return 0;
601 }
602
603 if (pmd_trans_unstable(pmd))
604 return 0;
605 /*
606 * The mmap_sem held all the way back in m_start() is what
607 * keeps khugepaged out of here and from collapsing things
608 * in here.
609 */
610 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
611 for (; addr != end; pte++, addr += PAGE_SIZE)
612 smaps_pte_entry(pte, addr, walk);
613 pte_unmap_unlock(pte - 1, ptl);
614 cond_resched();
615 return 0;
616}
617
618static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
619{
620 /*
621 * Don't forget to update Documentation/ on changes.
622 */
623 static const char mnemonics[BITS_PER_LONG][2] = {
624 /*
625 * In case if we meet a flag we don't know about.
626 */
627 [0 ... (BITS_PER_LONG-1)] = "??",
628
629 [ilog2(VM_READ)] = "rd",
630 [ilog2(VM_WRITE)] = "wr",
631 [ilog2(VM_EXEC)] = "ex",
632 [ilog2(VM_SHARED)] = "sh",
633 [ilog2(VM_MAYREAD)] = "mr",
634 [ilog2(VM_MAYWRITE)] = "mw",
635 [ilog2(VM_MAYEXEC)] = "me",
636 [ilog2(VM_MAYSHARE)] = "ms",
637 [ilog2(VM_GROWSDOWN)] = "gd",
638 [ilog2(VM_PFNMAP)] = "pf",
639 [ilog2(VM_DENYWRITE)] = "dw",
640#ifdef CONFIG_X86_INTEL_MPX
641 [ilog2(VM_MPX)] = "mp",
642#endif
643 [ilog2(VM_LOCKED)] = "lo",
644 [ilog2(VM_IO)] = "io",
645 [ilog2(VM_SEQ_READ)] = "sr",
646 [ilog2(VM_RAND_READ)] = "rr",
647 [ilog2(VM_DONTCOPY)] = "dc",
648 [ilog2(VM_DONTEXPAND)] = "de",
649 [ilog2(VM_ACCOUNT)] = "ac",
650 [ilog2(VM_NORESERVE)] = "nr",
651 [ilog2(VM_HUGETLB)] = "ht",
652 [ilog2(VM_ARCH_1)] = "ar",
653 [ilog2(VM_DONTDUMP)] = "dd",
654#ifdef CONFIG_MEM_SOFT_DIRTY
655 [ilog2(VM_SOFTDIRTY)] = "sd",
656#endif
657 [ilog2(VM_MIXEDMAP)] = "mm",
658 [ilog2(VM_HUGEPAGE)] = "hg",
659 [ilog2(VM_NOHUGEPAGE)] = "nh",
660 [ilog2(VM_MERGEABLE)] = "mg",
661 [ilog2(VM_UFFD_MISSING)]= "um",
662 [ilog2(VM_UFFD_WP)] = "uw",
663#ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
664 /* These come out via ProtectionKey: */
665 [ilog2(VM_PKEY_BIT0)] = "",
666 [ilog2(VM_PKEY_BIT1)] = "",
667 [ilog2(VM_PKEY_BIT2)] = "",
668 [ilog2(VM_PKEY_BIT3)] = "",
669#endif
670 };
671 size_t i;
672
673 seq_puts(m, "VmFlags: ");
674 for (i = 0; i < BITS_PER_LONG; i++) {
675 if (!mnemonics[i][0])
676 continue;
677 if (vma->vm_flags & (1UL << i)) {
678 seq_printf(m, "%c%c ",
679 mnemonics[i][0], mnemonics[i][1]);
680 }
681 }
682 seq_putc(m, '\n');
683}
684
685#ifdef CONFIG_HUGETLB_PAGE
686static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
687 unsigned long addr, unsigned long end,
688 struct mm_walk *walk)
689{
690 struct mem_size_stats *mss = walk->private;
691 struct vm_area_struct *vma = walk->vma;
692 struct page *page = NULL;
693
694 if (pte_present(*pte)) {
695 page = vm_normal_page(vma, addr, *pte);
696 } else if (is_swap_pte(*pte)) {
697 swp_entry_t swpent = pte_to_swp_entry(*pte);
698
699 if (is_migration_entry(swpent))
700 page = migration_entry_to_page(swpent);
701 }
702 if (page) {
703 int mapcount = page_mapcount(page);
704
705 if (mapcount >= 2)
706 mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
707 else
708 mss->private_hugetlb += huge_page_size(hstate_vma(vma));
709 }
710 return 0;
711}
712#endif /* HUGETLB_PAGE */
713
714void __weak arch_show_smap(struct seq_file *m, struct vm_area_struct *vma)
715{
716}
717
718static int show_smap(struct seq_file *m, void *v, int is_pid)
719{
720 struct vm_area_struct *vma = v;
721 struct mem_size_stats mss;
722 struct mm_walk smaps_walk = {
723 .pmd_entry = smaps_pte_range,
724#ifdef CONFIG_HUGETLB_PAGE
725 .hugetlb_entry = smaps_hugetlb_range,
726#endif
727 .mm = vma->vm_mm,
728 .private = &mss,
729 };
730
731 memset(&mss, 0, sizeof mss);
732
733#ifdef CONFIG_SHMEM
734 if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
735 /*
736 * For shared or readonly shmem mappings we know that all
737 * swapped out pages belong to the shmem object, and we can
738 * obtain the swap value much more efficiently. For private
739 * writable mappings, we might have COW pages that are
740 * not affected by the parent swapped out pages of the shmem
741 * object, so we have to distinguish them during the page walk.
742 * Unless we know that the shmem object (or the part mapped by
743 * our VMA) has no swapped out pages at all.
744 */
745 unsigned long shmem_swapped = shmem_swap_usage(vma);
746
747 if (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
748 !(vma->vm_flags & VM_WRITE)) {
749 mss.swap = shmem_swapped;
750 } else {
751 mss.check_shmem_swap = true;
752 smaps_walk.pte_hole = smaps_pte_hole;
753 }
754 }
755#endif
756
757 /* mmap_sem is held in m_start */
758 walk_page_vma(vma, &smaps_walk);
759
760 show_map_vma(m, vma, is_pid);
761
762 seq_printf(m,
763 "Size: %8lu kB\n"
764 "Rss: %8lu kB\n"
765 "Pss: %8lu kB\n"
766 "Shared_Clean: %8lu kB\n"
767 "Shared_Dirty: %8lu kB\n"
768 "Private_Clean: %8lu kB\n"
769 "Private_Dirty: %8lu kB\n"
770 "Referenced: %8lu kB\n"
771 "Anonymous: %8lu kB\n"
772 "AnonHugePages: %8lu kB\n"
773 "Shared_Hugetlb: %8lu kB\n"
774 "Private_Hugetlb: %7lu kB\n"
775 "Swap: %8lu kB\n"
776 "SwapPss: %8lu kB\n"
777 "KernelPageSize: %8lu kB\n"
778 "MMUPageSize: %8lu kB\n"
779 "Locked: %8lu kB\n",
780 (vma->vm_end - vma->vm_start) >> 10,
781 mss.resident >> 10,
782 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
783 mss.shared_clean >> 10,
784 mss.shared_dirty >> 10,
785 mss.private_clean >> 10,
786 mss.private_dirty >> 10,
787 mss.referenced >> 10,
788 mss.anonymous >> 10,
789 mss.anonymous_thp >> 10,
790 mss.shared_hugetlb >> 10,
791 mss.private_hugetlb >> 10,
792 mss.swap >> 10,
793 (unsigned long)(mss.swap_pss >> (10 + PSS_SHIFT)),
794 vma_kernel_pagesize(vma) >> 10,
795 vma_mmu_pagesize(vma) >> 10,
796 (vma->vm_flags & VM_LOCKED) ?
797 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
798
799 arch_show_smap(m, vma);
800 show_smap_vma_flags(m, vma);
801 m_cache_vma(m, vma);
802 return 0;
803}
804
805static int show_pid_smap(struct seq_file *m, void *v)
806{
807 return show_smap(m, v, 1);
808}
809
810static int show_tid_smap(struct seq_file *m, void *v)
811{
812 return show_smap(m, v, 0);
813}
814
815static const struct seq_operations proc_pid_smaps_op = {
816 .start = m_start,
817 .next = m_next,
818 .stop = m_stop,
819 .show = show_pid_smap
820};
821
822static const struct seq_operations proc_tid_smaps_op = {
823 .start = m_start,
824 .next = m_next,
825 .stop = m_stop,
826 .show = show_tid_smap
827};
828
829static int pid_smaps_open(struct inode *inode, struct file *file)
830{
831 return do_maps_open(inode, file, &proc_pid_smaps_op);
832}
833
834static int tid_smaps_open(struct inode *inode, struct file *file)
835{
836 return do_maps_open(inode, file, &proc_tid_smaps_op);
837}
838
839const struct file_operations proc_pid_smaps_operations = {
840 .open = pid_smaps_open,
841 .read = seq_read,
842 .llseek = seq_lseek,
843 .release = proc_map_release,
844};
845
846const struct file_operations proc_tid_smaps_operations = {
847 .open = tid_smaps_open,
848 .read = seq_read,
849 .llseek = seq_lseek,
850 .release = proc_map_release,
851};
852
853enum clear_refs_types {
854 CLEAR_REFS_ALL = 1,
855 CLEAR_REFS_ANON,
856 CLEAR_REFS_MAPPED,
857 CLEAR_REFS_SOFT_DIRTY,
858 CLEAR_REFS_MM_HIWATER_RSS,
859 CLEAR_REFS_LAST,
860};
861
862struct clear_refs_private {
863 enum clear_refs_types type;
864};
865
866#ifdef CONFIG_MEM_SOFT_DIRTY
867static inline void clear_soft_dirty(struct vm_area_struct *vma,
868 unsigned long addr, pte_t *pte)
869{
870 /*
871 * The soft-dirty tracker uses #PF-s to catch writes
872 * to pages, so write-protect the pte as well. See the
873 * Documentation/vm/soft-dirty.txt for full description
874 * of how soft-dirty works.
875 */
876 pte_t ptent = *pte;
877
878 if (pte_present(ptent)) {
879 ptent = ptep_modify_prot_start(vma->vm_mm, addr, pte);
880 ptent = pte_wrprotect(ptent);
881 ptent = pte_clear_soft_dirty(ptent);
882 ptep_modify_prot_commit(vma->vm_mm, addr, pte, ptent);
883 } else if (is_swap_pte(ptent)) {
884 ptent = pte_swp_clear_soft_dirty(ptent);
885 set_pte_at(vma->vm_mm, addr, pte, ptent);
886 }
887}
888#else
889static inline void clear_soft_dirty(struct vm_area_struct *vma,
890 unsigned long addr, pte_t *pte)
891{
892}
893#endif
894
895#if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
896static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
897 unsigned long addr, pmd_t *pmdp)
898{
899 pmd_t pmd = pmdp_huge_get_and_clear(vma->vm_mm, addr, pmdp);
900
901 pmd = pmd_wrprotect(pmd);
902 pmd = pmd_clear_soft_dirty(pmd);
903
904 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
905}
906#else
907static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
908 unsigned long addr, pmd_t *pmdp)
909{
910}
911#endif
912
913static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
914 unsigned long end, struct mm_walk *walk)
915{
916 struct clear_refs_private *cp = walk->private;
917 struct vm_area_struct *vma = walk->vma;
918 pte_t *pte, ptent;
919 spinlock_t *ptl;
920 struct page *page;
921
922 ptl = pmd_trans_huge_lock(pmd, vma);
923 if (ptl) {
924 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
925 clear_soft_dirty_pmd(vma, addr, pmd);
926 goto out;
927 }
928
929 page = pmd_page(*pmd);
930
931 /* Clear accessed and referenced bits. */
932 pmdp_test_and_clear_young(vma, addr, pmd);
933 test_and_clear_page_young(page);
934 ClearPageReferenced(page);
935out:
936 spin_unlock(ptl);
937 return 0;
938 }
939
940 if (pmd_trans_unstable(pmd))
941 return 0;
942
943 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
944 for (; addr != end; pte++, addr += PAGE_SIZE) {
945 ptent = *pte;
946
947 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
948 clear_soft_dirty(vma, addr, pte);
949 continue;
950 }
951
952 if (!pte_present(ptent))
953 continue;
954
955 page = vm_normal_page(vma, addr, ptent);
956 if (!page)
957 continue;
958
959 /* Clear accessed and referenced bits. */
960 ptep_test_and_clear_young(vma, addr, pte);
961 test_and_clear_page_young(page);
962 ClearPageReferenced(page);
963 }
964 pte_unmap_unlock(pte - 1, ptl);
965 cond_resched();
966 return 0;
967}
968
969static int clear_refs_test_walk(unsigned long start, unsigned long end,
970 struct mm_walk *walk)
971{
972 struct clear_refs_private *cp = walk->private;
973 struct vm_area_struct *vma = walk->vma;
974
975 if (vma->vm_flags & VM_PFNMAP)
976 return 1;
977
978 /*
979 * Writing 1 to /proc/pid/clear_refs affects all pages.
980 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
981 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
982 * Writing 4 to /proc/pid/clear_refs affects all pages.
983 */
984 if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
985 return 1;
986 if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
987 return 1;
988 return 0;
989}
990
991static ssize_t clear_refs_write(struct file *file, const char __user *buf,
992 size_t count, loff_t *ppos)
993{
994 struct task_struct *task;
995 char buffer[PROC_NUMBUF];
996 struct mm_struct *mm;
997 struct vm_area_struct *vma;
998 enum clear_refs_types type;
999 int itype;
1000 int rv;
1001
1002 memset(buffer, 0, sizeof(buffer));
1003 if (count > sizeof(buffer) - 1)
1004 count = sizeof(buffer) - 1;
1005 if (copy_from_user(buffer, buf, count))
1006 return -EFAULT;
1007 rv = kstrtoint(strstrip(buffer), 10, &itype);
1008 if (rv < 0)
1009 return rv;
1010 type = (enum clear_refs_types)itype;
1011 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1012 return -EINVAL;
1013
1014 task = get_proc_task(file_inode(file));
1015 if (!task)
1016 return -ESRCH;
1017 mm = get_task_mm(task);
1018 if (mm) {
1019 struct clear_refs_private cp = {
1020 .type = type,
1021 };
1022 struct mm_walk clear_refs_walk = {
1023 .pmd_entry = clear_refs_pte_range,
1024 .test_walk = clear_refs_test_walk,
1025 .mm = mm,
1026 .private = &cp,
1027 };
1028
1029 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1030 /*
1031 * Writing 5 to /proc/pid/clear_refs resets the peak
1032 * resident set size to this mm's current rss value.
1033 */
1034 down_write(&mm->mmap_sem);
1035 reset_mm_hiwater_rss(mm);
1036 up_write(&mm->mmap_sem);
1037 goto out_mm;
1038 }
1039
1040 down_read(&mm->mmap_sem);
1041 if (type == CLEAR_REFS_SOFT_DIRTY) {
1042 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1043 if (!(vma->vm_flags & VM_SOFTDIRTY))
1044 continue;
1045 up_read(&mm->mmap_sem);
1046 down_write(&mm->mmap_sem);
1047 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1048 vma->vm_flags &= ~VM_SOFTDIRTY;
1049 vma_set_page_prot(vma);
1050 }
1051 downgrade_write(&mm->mmap_sem);
1052 break;
1053 }
1054 mmu_notifier_invalidate_range_start(mm, 0, -1);
1055 }
1056 walk_page_range(0, ~0UL, &clear_refs_walk);
1057 if (type == CLEAR_REFS_SOFT_DIRTY)
1058 mmu_notifier_invalidate_range_end(mm, 0, -1);
1059 flush_tlb_mm(mm);
1060 up_read(&mm->mmap_sem);
1061out_mm:
1062 mmput(mm);
1063 }
1064 put_task_struct(task);
1065
1066 return count;
1067}
1068
1069const struct file_operations proc_clear_refs_operations = {
1070 .write = clear_refs_write,
1071 .llseek = noop_llseek,
1072};
1073
1074typedef struct {
1075 u64 pme;
1076} pagemap_entry_t;
1077
1078struct pagemapread {
1079 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */
1080 pagemap_entry_t *buffer;
1081 bool show_pfn;
1082};
1083
1084#define PAGEMAP_WALK_SIZE (PMD_SIZE)
1085#define PAGEMAP_WALK_MASK (PMD_MASK)
1086
1087#define PM_ENTRY_BYTES sizeof(pagemap_entry_t)
1088#define PM_PFRAME_BITS 55
1089#define PM_PFRAME_MASK GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1090#define PM_SOFT_DIRTY BIT_ULL(55)
1091#define PM_MMAP_EXCLUSIVE BIT_ULL(56)
1092#define PM_FILE BIT_ULL(61)
1093#define PM_SWAP BIT_ULL(62)
1094#define PM_PRESENT BIT_ULL(63)
1095
1096#define PM_END_OF_BUFFER 1
1097
1098static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1099{
1100 return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1101}
1102
1103static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1104 struct pagemapread *pm)
1105{
1106 pm->buffer[pm->pos++] = *pme;
1107 if (pm->pos >= pm->len)
1108 return PM_END_OF_BUFFER;
1109 return 0;
1110}
1111
1112static int pagemap_pte_hole(unsigned long start, unsigned long end,
1113 struct mm_walk *walk)
1114{
1115 struct pagemapread *pm = walk->private;
1116 unsigned long addr = start;
1117 int err = 0;
1118
1119 while (addr < end) {
1120 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1121 pagemap_entry_t pme = make_pme(0, 0);
1122 /* End of address space hole, which we mark as non-present. */
1123 unsigned long hole_end;
1124
1125 if (vma)
1126 hole_end = min(end, vma->vm_start);
1127 else
1128 hole_end = end;
1129
1130 for (; addr < hole_end; addr += PAGE_SIZE) {
1131 err = add_to_pagemap(addr, &pme, pm);
1132 if (err)
1133 goto out;
1134 }
1135
1136 if (!vma)
1137 break;
1138
1139 /* Addresses in the VMA. */
1140 if (vma->vm_flags & VM_SOFTDIRTY)
1141 pme = make_pme(0, PM_SOFT_DIRTY);
1142 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1143 err = add_to_pagemap(addr, &pme, pm);
1144 if (err)
1145 goto out;
1146 }
1147 }
1148out:
1149 return err;
1150}
1151
1152static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1153 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1154{
1155 u64 frame = 0, flags = 0;
1156 struct page *page = NULL;
1157
1158 if (pte_present(pte)) {
1159 if (pm->show_pfn)
1160 frame = pte_pfn(pte);
1161 flags |= PM_PRESENT;
1162 page = vm_normal_page(vma, addr, pte);
1163 if (pte_soft_dirty(pte))
1164 flags |= PM_SOFT_DIRTY;
1165 } else if (is_swap_pte(pte)) {
1166 swp_entry_t entry;
1167 if (pte_swp_soft_dirty(pte))
1168 flags |= PM_SOFT_DIRTY;
1169 entry = pte_to_swp_entry(pte);
1170 frame = swp_type(entry) |
1171 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1172 flags |= PM_SWAP;
1173 if (is_migration_entry(entry))
1174 page = migration_entry_to_page(entry);
1175 }
1176
1177 if (page && !PageAnon(page))
1178 flags |= PM_FILE;
1179 if (page && page_mapcount(page) == 1)
1180 flags |= PM_MMAP_EXCLUSIVE;
1181 if (vma->vm_flags & VM_SOFTDIRTY)
1182 flags |= PM_SOFT_DIRTY;
1183
1184 return make_pme(frame, flags);
1185}
1186
1187static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1188 struct mm_walk *walk)
1189{
1190 struct vm_area_struct *vma = walk->vma;
1191 struct pagemapread *pm = walk->private;
1192 spinlock_t *ptl;
1193 pte_t *pte, *orig_pte;
1194 int err = 0;
1195
1196#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1197 ptl = pmd_trans_huge_lock(pmdp, vma);
1198 if (ptl) {
1199 u64 flags = 0, frame = 0;
1200 pmd_t pmd = *pmdp;
1201
1202 if ((vma->vm_flags & VM_SOFTDIRTY) || pmd_soft_dirty(pmd))
1203 flags |= PM_SOFT_DIRTY;
1204
1205 /*
1206 * Currently pmd for thp is always present because thp
1207 * can not be swapped-out, migrated, or HWPOISONed
1208 * (split in such cases instead.)
1209 * This if-check is just to prepare for future implementation.
1210 */
1211 if (pmd_present(pmd)) {
1212 struct page *page = pmd_page(pmd);
1213
1214 if (page_mapcount(page) == 1)
1215 flags |= PM_MMAP_EXCLUSIVE;
1216
1217 flags |= PM_PRESENT;
1218 if (pm->show_pfn)
1219 frame = pmd_pfn(pmd) +
1220 ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1221 }
1222
1223 for (; addr != end; addr += PAGE_SIZE) {
1224 pagemap_entry_t pme = make_pme(frame, flags);
1225
1226 err = add_to_pagemap(addr, &pme, pm);
1227 if (err)
1228 break;
1229 if (pm->show_pfn && (flags & PM_PRESENT))
1230 frame++;
1231 }
1232 spin_unlock(ptl);
1233 return err;
1234 }
1235
1236 if (pmd_trans_unstable(pmdp))
1237 return 0;
1238#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1239
1240 /*
1241 * We can assume that @vma always points to a valid one and @end never
1242 * goes beyond vma->vm_end.
1243 */
1244 orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1245 for (; addr < end; pte++, addr += PAGE_SIZE) {
1246 pagemap_entry_t pme;
1247
1248 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1249 err = add_to_pagemap(addr, &pme, pm);
1250 if (err)
1251 break;
1252 }
1253 pte_unmap_unlock(orig_pte, ptl);
1254
1255 cond_resched();
1256
1257 return err;
1258}
1259
1260#ifdef CONFIG_HUGETLB_PAGE
1261/* This function walks within one hugetlb entry in the single call */
1262static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1263 unsigned long addr, unsigned long end,
1264 struct mm_walk *walk)
1265{
1266 struct pagemapread *pm = walk->private;
1267 struct vm_area_struct *vma = walk->vma;
1268 u64 flags = 0, frame = 0;
1269 int err = 0;
1270 pte_t pte;
1271
1272 if (vma->vm_flags & VM_SOFTDIRTY)
1273 flags |= PM_SOFT_DIRTY;
1274
1275 pte = huge_ptep_get(ptep);
1276 if (pte_present(pte)) {
1277 struct page *page = pte_page(pte);
1278
1279 if (!PageAnon(page))
1280 flags |= PM_FILE;
1281
1282 if (page_mapcount(page) == 1)
1283 flags |= PM_MMAP_EXCLUSIVE;
1284
1285 flags |= PM_PRESENT;
1286 if (pm->show_pfn)
1287 frame = pte_pfn(pte) +
1288 ((addr & ~hmask) >> PAGE_SHIFT);
1289 }
1290
1291 for (; addr != end; addr += PAGE_SIZE) {
1292 pagemap_entry_t pme = make_pme(frame, flags);
1293
1294 err = add_to_pagemap(addr, &pme, pm);
1295 if (err)
1296 return err;
1297 if (pm->show_pfn && (flags & PM_PRESENT))
1298 frame++;
1299 }
1300
1301 cond_resched();
1302
1303 return err;
1304}
1305#endif /* HUGETLB_PAGE */
1306
1307/*
1308 * /proc/pid/pagemap - an array mapping virtual pages to pfns
1309 *
1310 * For each page in the address space, this file contains one 64-bit entry
1311 * consisting of the following:
1312 *
1313 * Bits 0-54 page frame number (PFN) if present
1314 * Bits 0-4 swap type if swapped
1315 * Bits 5-54 swap offset if swapped
1316 * Bit 55 pte is soft-dirty (see Documentation/vm/soft-dirty.txt)
1317 * Bit 56 page exclusively mapped
1318 * Bits 57-60 zero
1319 * Bit 61 page is file-page or shared-anon
1320 * Bit 62 page swapped
1321 * Bit 63 page present
1322 *
1323 * If the page is not present but in swap, then the PFN contains an
1324 * encoding of the swap file number and the page's offset into the
1325 * swap. Unmapped pages return a null PFN. This allows determining
1326 * precisely which pages are mapped (or in swap) and comparing mapped
1327 * pages between processes.
1328 *
1329 * Efficient users of this interface will use /proc/pid/maps to
1330 * determine which areas of memory are actually mapped and llseek to
1331 * skip over unmapped regions.
1332 */
1333static ssize_t pagemap_read(struct file *file, char __user *buf,
1334 size_t count, loff_t *ppos)
1335{
1336 struct mm_struct *mm = file->private_data;
1337 struct pagemapread pm;
1338 struct mm_walk pagemap_walk = {};
1339 unsigned long src;
1340 unsigned long svpfn;
1341 unsigned long start_vaddr;
1342 unsigned long end_vaddr;
1343 int ret = 0, copied = 0;
1344
1345 if (!mm || !atomic_inc_not_zero(&mm->mm_users))
1346 goto out;
1347
1348 ret = -EINVAL;
1349 /* file position must be aligned */
1350 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1351 goto out_mm;
1352
1353 ret = 0;
1354 if (!count)
1355 goto out_mm;
1356
1357 /* do not disclose physical addresses: attack vector */
1358 pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1359
1360 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1361 pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY);
1362 ret = -ENOMEM;
1363 if (!pm.buffer)
1364 goto out_mm;
1365
1366 pagemap_walk.pmd_entry = pagemap_pmd_range;
1367 pagemap_walk.pte_hole = pagemap_pte_hole;
1368#ifdef CONFIG_HUGETLB_PAGE
1369 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1370#endif
1371 pagemap_walk.mm = mm;
1372 pagemap_walk.private = ±
1373
1374 src = *ppos;
1375 svpfn = src / PM_ENTRY_BYTES;
1376 start_vaddr = svpfn << PAGE_SHIFT;
1377 end_vaddr = mm->task_size;
1378
1379 /* watch out for wraparound */
1380 if (svpfn > mm->task_size >> PAGE_SHIFT)
1381 start_vaddr = end_vaddr;
1382
1383 /*
1384 * The odds are that this will stop walking way
1385 * before end_vaddr, because the length of the
1386 * user buffer is tracked in "pm", and the walk
1387 * will stop when we hit the end of the buffer.
1388 */
1389 ret = 0;
1390 while (count && (start_vaddr < end_vaddr)) {
1391 int len;
1392 unsigned long end;
1393
1394 pm.pos = 0;
1395 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1396 /* overflow ? */
1397 if (end < start_vaddr || end > end_vaddr)
1398 end = end_vaddr;
1399 down_read(&mm->mmap_sem);
1400 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1401 up_read(&mm->mmap_sem);
1402 start_vaddr = end;
1403
1404 len = min(count, PM_ENTRY_BYTES * pm.pos);
1405 if (copy_to_user(buf, pm.buffer, len)) {
1406 ret = -EFAULT;
1407 goto out_free;
1408 }
1409 copied += len;
1410 buf += len;
1411 count -= len;
1412 }
1413 *ppos += copied;
1414 if (!ret || ret == PM_END_OF_BUFFER)
1415 ret = copied;
1416
1417out_free:
1418 kfree(pm.buffer);
1419out_mm:
1420 mmput(mm);
1421out:
1422 return ret;
1423}
1424
1425static int pagemap_open(struct inode *inode, struct file *file)
1426{
1427 struct mm_struct *mm;
1428
1429 mm = proc_mem_open(inode, PTRACE_MODE_READ);
1430 if (IS_ERR(mm))
1431 return PTR_ERR(mm);
1432 file->private_data = mm;
1433 return 0;
1434}
1435
1436static int pagemap_release(struct inode *inode, struct file *file)
1437{
1438 struct mm_struct *mm = file->private_data;
1439
1440 if (mm)
1441 mmdrop(mm);
1442 return 0;
1443}
1444
1445const struct file_operations proc_pagemap_operations = {
1446 .llseek = mem_lseek, /* borrow this */
1447 .read = pagemap_read,
1448 .open = pagemap_open,
1449 .release = pagemap_release,
1450};
1451#endif /* CONFIG_PROC_PAGE_MONITOR */
1452
1453#ifdef CONFIG_NUMA
1454
1455struct numa_maps {
1456 unsigned long pages;
1457 unsigned long anon;
1458 unsigned long active;
1459 unsigned long writeback;
1460 unsigned long mapcount_max;
1461 unsigned long dirty;
1462 unsigned long swapcache;
1463 unsigned long node[MAX_NUMNODES];
1464};
1465
1466struct numa_maps_private {
1467 struct proc_maps_private proc_maps;
1468 struct numa_maps md;
1469};
1470
1471static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1472 unsigned long nr_pages)
1473{
1474 int count = page_mapcount(page);
1475
1476 md->pages += nr_pages;
1477 if (pte_dirty || PageDirty(page))
1478 md->dirty += nr_pages;
1479
1480 if (PageSwapCache(page))
1481 md->swapcache += nr_pages;
1482
1483 if (PageActive(page) || PageUnevictable(page))
1484 md->active += nr_pages;
1485
1486 if (PageWriteback(page))
1487 md->writeback += nr_pages;
1488
1489 if (PageAnon(page))
1490 md->anon += nr_pages;
1491
1492 if (count > md->mapcount_max)
1493 md->mapcount_max = count;
1494
1495 md->node[page_to_nid(page)] += nr_pages;
1496}
1497
1498static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1499 unsigned long addr)
1500{
1501 struct page *page;
1502 int nid;
1503
1504 if (!pte_present(pte))
1505 return NULL;
1506
1507 page = vm_normal_page(vma, addr, pte);
1508 if (!page)
1509 return NULL;
1510
1511 if (PageReserved(page))
1512 return NULL;
1513
1514 nid = page_to_nid(page);
1515 if (!node_isset(nid, node_states[N_MEMORY]))
1516 return NULL;
1517
1518 return page;
1519}
1520
1521#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1522static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1523 struct vm_area_struct *vma,
1524 unsigned long addr)
1525{
1526 struct page *page;
1527 int nid;
1528
1529 if (!pmd_present(pmd))
1530 return NULL;
1531
1532 page = vm_normal_page_pmd(vma, addr, pmd);
1533 if (!page)
1534 return NULL;
1535
1536 if (PageReserved(page))
1537 return NULL;
1538
1539 nid = page_to_nid(page);
1540 if (!node_isset(nid, node_states[N_MEMORY]))
1541 return NULL;
1542
1543 return page;
1544}
1545#endif
1546
1547static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1548 unsigned long end, struct mm_walk *walk)
1549{
1550 struct numa_maps *md = walk->private;
1551 struct vm_area_struct *vma = walk->vma;
1552 spinlock_t *ptl;
1553 pte_t *orig_pte;
1554 pte_t *pte;
1555
1556#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1557 ptl = pmd_trans_huge_lock(pmd, vma);
1558 if (ptl) {
1559 struct page *page;
1560
1561 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1562 if (page)
1563 gather_stats(page, md, pmd_dirty(*pmd),
1564 HPAGE_PMD_SIZE/PAGE_SIZE);
1565 spin_unlock(ptl);
1566 return 0;
1567 }
1568
1569 if (pmd_trans_unstable(pmd))
1570 return 0;
1571#endif
1572 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1573 do {
1574 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1575 if (!page)
1576 continue;
1577 gather_stats(page, md, pte_dirty(*pte), 1);
1578
1579 } while (pte++, addr += PAGE_SIZE, addr != end);
1580 pte_unmap_unlock(orig_pte, ptl);
1581 return 0;
1582}
1583#ifdef CONFIG_HUGETLB_PAGE
1584static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1585 unsigned long addr, unsigned long end, struct mm_walk *walk)
1586{
1587 pte_t huge_pte = huge_ptep_get(pte);
1588 struct numa_maps *md;
1589 struct page *page;
1590
1591 if (!pte_present(huge_pte))
1592 return 0;
1593
1594 page = pte_page(huge_pte);
1595 if (!page)
1596 return 0;
1597
1598 md = walk->private;
1599 gather_stats(page, md, pte_dirty(huge_pte), 1);
1600 return 0;
1601}
1602
1603#else
1604static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1605 unsigned long addr, unsigned long end, struct mm_walk *walk)
1606{
1607 return 0;
1608}
1609#endif
1610
1611/*
1612 * Display pages allocated per node and memory policy via /proc.
1613 */
1614static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1615{
1616 struct numa_maps_private *numa_priv = m->private;
1617 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1618 struct vm_area_struct *vma = v;
1619 struct numa_maps *md = &numa_priv->md;
1620 struct file *file = vma->vm_file;
1621 struct mm_struct *mm = vma->vm_mm;
1622 struct mm_walk walk = {
1623 .hugetlb_entry = gather_hugetlb_stats,
1624 .pmd_entry = gather_pte_stats,
1625 .private = md,
1626 .mm = mm,
1627 };
1628 struct mempolicy *pol;
1629 char buffer[64];
1630 int nid;
1631
1632 if (!mm)
1633 return 0;
1634
1635 /* Ensure we start with an empty set of numa_maps statistics. */
1636 memset(md, 0, sizeof(*md));
1637
1638 pol = __get_vma_policy(vma, vma->vm_start);
1639 if (pol) {
1640 mpol_to_str(buffer, sizeof(buffer), pol);
1641 mpol_cond_put(pol);
1642 } else {
1643 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1644 }
1645
1646 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1647
1648 if (file) {
1649 seq_puts(m, " file=");
1650 seq_file_path(m, file, "\n\t= ");
1651 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1652 seq_puts(m, " heap");
1653 } else if (is_stack(proc_priv, vma, is_pid)) {
1654 seq_puts(m, " stack");
1655 }
1656
1657 if (is_vm_hugetlb_page(vma))
1658 seq_puts(m, " huge");
1659
1660 /* mmap_sem is held by m_start */
1661 walk_page_vma(vma, &walk);
1662
1663 if (!md->pages)
1664 goto out;
1665
1666 if (md->anon)
1667 seq_printf(m, " anon=%lu", md->anon);
1668
1669 if (md->dirty)
1670 seq_printf(m, " dirty=%lu", md->dirty);
1671
1672 if (md->pages != md->anon && md->pages != md->dirty)
1673 seq_printf(m, " mapped=%lu", md->pages);
1674
1675 if (md->mapcount_max > 1)
1676 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1677
1678 if (md->swapcache)
1679 seq_printf(m, " swapcache=%lu", md->swapcache);
1680
1681 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1682 seq_printf(m, " active=%lu", md->active);
1683
1684 if (md->writeback)
1685 seq_printf(m, " writeback=%lu", md->writeback);
1686
1687 for_each_node_state(nid, N_MEMORY)
1688 if (md->node[nid])
1689 seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1690
1691 seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1692out:
1693 seq_putc(m, '\n');
1694 m_cache_vma(m, vma);
1695 return 0;
1696}
1697
1698static int show_pid_numa_map(struct seq_file *m, void *v)
1699{
1700 return show_numa_map(m, v, 1);
1701}
1702
1703static int show_tid_numa_map(struct seq_file *m, void *v)
1704{
1705 return show_numa_map(m, v, 0);
1706}
1707
1708static const struct seq_operations proc_pid_numa_maps_op = {
1709 .start = m_start,
1710 .next = m_next,
1711 .stop = m_stop,
1712 .show = show_pid_numa_map,
1713};
1714
1715static const struct seq_operations proc_tid_numa_maps_op = {
1716 .start = m_start,
1717 .next = m_next,
1718 .stop = m_stop,
1719 .show = show_tid_numa_map,
1720};
1721
1722static int numa_maps_open(struct inode *inode, struct file *file,
1723 const struct seq_operations *ops)
1724{
1725 return proc_maps_open(inode, file, ops,
1726 sizeof(struct numa_maps_private));
1727}
1728
1729static int pid_numa_maps_open(struct inode *inode, struct file *file)
1730{
1731 return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1732}
1733
1734static int tid_numa_maps_open(struct inode *inode, struct file *file)
1735{
1736 return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1737}
1738
1739const struct file_operations proc_pid_numa_maps_operations = {
1740 .open = pid_numa_maps_open,
1741 .read = seq_read,
1742 .llseek = seq_lseek,
1743 .release = proc_map_release,
1744};
1745
1746const struct file_operations proc_tid_numa_maps_operations = {
1747 .open = tid_numa_maps_open,
1748 .read = seq_read,
1749 .llseek = seq_lseek,
1750 .release = proc_map_release,
1751};
1752#endif /* CONFIG_NUMA */
1#include <linux/mm.h>
2#include <linux/vmacache.h>
3#include <linux/hugetlb.h>
4#include <linux/huge_mm.h>
5#include <linux/mount.h>
6#include <linux/seq_file.h>
7#include <linux/highmem.h>
8#include <linux/ptrace.h>
9#include <linux/slab.h>
10#include <linux/pagemap.h>
11#include <linux/mempolicy.h>
12#include <linux/rmap.h>
13#include <linux/swap.h>
14#include <linux/swapops.h>
15#include <linux/mmu_notifier.h>
16
17#include <asm/elf.h>
18#include <asm/uaccess.h>
19#include <asm/tlbflush.h>
20#include "internal.h"
21
22void task_mem(struct seq_file *m, struct mm_struct *mm)
23{
24 unsigned long data, text, lib, swap;
25 unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
26
27 /*
28 * Note: to minimize their overhead, mm maintains hiwater_vm and
29 * hiwater_rss only when about to *lower* total_vm or rss. Any
30 * collector of these hiwater stats must therefore get total_vm
31 * and rss too, which will usually be the higher. Barriers? not
32 * worth the effort, such snapshots can always be inconsistent.
33 */
34 hiwater_vm = total_vm = mm->total_vm;
35 if (hiwater_vm < mm->hiwater_vm)
36 hiwater_vm = mm->hiwater_vm;
37 hiwater_rss = total_rss = get_mm_rss(mm);
38 if (hiwater_rss < mm->hiwater_rss)
39 hiwater_rss = mm->hiwater_rss;
40
41 data = mm->total_vm - mm->shared_vm - mm->stack_vm;
42 text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
43 lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
44 swap = get_mm_counter(mm, MM_SWAPENTS);
45 seq_printf(m,
46 "VmPeak:\t%8lu kB\n"
47 "VmSize:\t%8lu kB\n"
48 "VmLck:\t%8lu kB\n"
49 "VmPin:\t%8lu kB\n"
50 "VmHWM:\t%8lu kB\n"
51 "VmRSS:\t%8lu kB\n"
52 "VmData:\t%8lu kB\n"
53 "VmStk:\t%8lu kB\n"
54 "VmExe:\t%8lu kB\n"
55 "VmLib:\t%8lu kB\n"
56 "VmPTE:\t%8lu kB\n"
57 "VmSwap:\t%8lu kB\n",
58 hiwater_vm << (PAGE_SHIFT-10),
59 total_vm << (PAGE_SHIFT-10),
60 mm->locked_vm << (PAGE_SHIFT-10),
61 mm->pinned_vm << (PAGE_SHIFT-10),
62 hiwater_rss << (PAGE_SHIFT-10),
63 total_rss << (PAGE_SHIFT-10),
64 data << (PAGE_SHIFT-10),
65 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
66 (PTRS_PER_PTE * sizeof(pte_t) *
67 atomic_long_read(&mm->nr_ptes)) >> 10,
68 swap << (PAGE_SHIFT-10));
69}
70
71unsigned long task_vsize(struct mm_struct *mm)
72{
73 return PAGE_SIZE * mm->total_vm;
74}
75
76unsigned long task_statm(struct mm_struct *mm,
77 unsigned long *shared, unsigned long *text,
78 unsigned long *data, unsigned long *resident)
79{
80 *shared = get_mm_counter(mm, MM_FILEPAGES);
81 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
82 >> PAGE_SHIFT;
83 *data = mm->total_vm - mm->shared_vm;
84 *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
85 return mm->total_vm;
86}
87
88#ifdef CONFIG_NUMA
89/*
90 * These functions are for numa_maps but called in generic **maps seq_file
91 * ->start(), ->stop() ops.
92 *
93 * numa_maps scans all vmas under mmap_sem and checks their mempolicy.
94 * Each mempolicy object is controlled by reference counting. The problem here
95 * is how to avoid accessing dead mempolicy object.
96 *
97 * Because we're holding mmap_sem while reading seq_file, it's safe to access
98 * each vma's mempolicy, no vma objects will never drop refs to mempolicy.
99 *
100 * A task's mempolicy (task->mempolicy) has different behavior. task->mempolicy
101 * is set and replaced under mmap_sem but unrefed and cleared under task_lock().
102 * So, without task_lock(), we cannot trust get_vma_policy() because we cannot
103 * gurantee the task never exits under us. But taking task_lock() around
104 * get_vma_plicy() causes lock order problem.
105 *
106 * To access task->mempolicy without lock, we hold a reference count of an
107 * object pointed by task->mempolicy and remember it. This will guarantee
108 * that task->mempolicy points to an alive object or NULL in numa_maps accesses.
109 */
110static void hold_task_mempolicy(struct proc_maps_private *priv)
111{
112 struct task_struct *task = priv->task;
113
114 task_lock(task);
115 priv->task_mempolicy = task->mempolicy;
116 mpol_get(priv->task_mempolicy);
117 task_unlock(task);
118}
119static void release_task_mempolicy(struct proc_maps_private *priv)
120{
121 mpol_put(priv->task_mempolicy);
122}
123#else
124static void hold_task_mempolicy(struct proc_maps_private *priv)
125{
126}
127static void release_task_mempolicy(struct proc_maps_private *priv)
128{
129}
130#endif
131
132static void vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma)
133{
134 if (vma && vma != priv->tail_vma) {
135 struct mm_struct *mm = vma->vm_mm;
136 release_task_mempolicy(priv);
137 up_read(&mm->mmap_sem);
138 mmput(mm);
139 }
140}
141
142static void *m_start(struct seq_file *m, loff_t *pos)
143{
144 struct proc_maps_private *priv = m->private;
145 unsigned long last_addr = m->version;
146 struct mm_struct *mm;
147 struct vm_area_struct *vma, *tail_vma = NULL;
148 loff_t l = *pos;
149
150 /* Clear the per syscall fields in priv */
151 priv->task = NULL;
152 priv->tail_vma = NULL;
153
154 /*
155 * We remember last_addr rather than next_addr to hit with
156 * vmacache most of the time. We have zero last_addr at
157 * the beginning and also after lseek. We will have -1 last_addr
158 * after the end of the vmas.
159 */
160
161 if (last_addr == -1UL)
162 return NULL;
163
164 priv->task = get_pid_task(priv->pid, PIDTYPE_PID);
165 if (!priv->task)
166 return ERR_PTR(-ESRCH);
167
168 mm = mm_access(priv->task, PTRACE_MODE_READ);
169 if (!mm || IS_ERR(mm))
170 return mm;
171 down_read(&mm->mmap_sem);
172
173 tail_vma = get_gate_vma(priv->task->mm);
174 priv->tail_vma = tail_vma;
175 hold_task_mempolicy(priv);
176 /* Start with last addr hint */
177 vma = find_vma(mm, last_addr);
178 if (last_addr && vma) {
179 vma = vma->vm_next;
180 goto out;
181 }
182
183 /*
184 * Check the vma index is within the range and do
185 * sequential scan until m_index.
186 */
187 vma = NULL;
188 if ((unsigned long)l < mm->map_count) {
189 vma = mm->mmap;
190 while (l-- && vma)
191 vma = vma->vm_next;
192 goto out;
193 }
194
195 if (l != mm->map_count)
196 tail_vma = NULL; /* After gate vma */
197
198out:
199 if (vma)
200 return vma;
201
202 release_task_mempolicy(priv);
203 /* End of vmas has been reached */
204 m->version = (tail_vma != NULL)? 0: -1UL;
205 up_read(&mm->mmap_sem);
206 mmput(mm);
207 return tail_vma;
208}
209
210static void *m_next(struct seq_file *m, void *v, loff_t *pos)
211{
212 struct proc_maps_private *priv = m->private;
213 struct vm_area_struct *vma = v;
214 struct vm_area_struct *tail_vma = priv->tail_vma;
215
216 (*pos)++;
217 if (vma && (vma != tail_vma) && vma->vm_next)
218 return vma->vm_next;
219 vma_stop(priv, vma);
220 return (vma != tail_vma)? tail_vma: NULL;
221}
222
223static void m_stop(struct seq_file *m, void *v)
224{
225 struct proc_maps_private *priv = m->private;
226 struct vm_area_struct *vma = v;
227
228 if (!IS_ERR(vma))
229 vma_stop(priv, vma);
230 if (priv->task)
231 put_task_struct(priv->task);
232}
233
234static int do_maps_open(struct inode *inode, struct file *file,
235 const struct seq_operations *ops)
236{
237 struct proc_maps_private *priv;
238 int ret = -ENOMEM;
239 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
240 if (priv) {
241 priv->pid = proc_pid(inode);
242 ret = seq_open(file, ops);
243 if (!ret) {
244 struct seq_file *m = file->private_data;
245 m->private = priv;
246 } else {
247 kfree(priv);
248 }
249 }
250 return ret;
251}
252
253static void
254show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
255{
256 struct mm_struct *mm = vma->vm_mm;
257 struct file *file = vma->vm_file;
258 struct proc_maps_private *priv = m->private;
259 struct task_struct *task = priv->task;
260 vm_flags_t flags = vma->vm_flags;
261 unsigned long ino = 0;
262 unsigned long long pgoff = 0;
263 unsigned long start, end;
264 dev_t dev = 0;
265 const char *name = NULL;
266
267 if (file) {
268 struct inode *inode = file_inode(vma->vm_file);
269 dev = inode->i_sb->s_dev;
270 ino = inode->i_ino;
271 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
272 }
273
274 /* We don't show the stack guard page in /proc/maps */
275 start = vma->vm_start;
276 if (stack_guard_page_start(vma, start))
277 start += PAGE_SIZE;
278 end = vma->vm_end;
279 if (stack_guard_page_end(vma, end))
280 end -= PAGE_SIZE;
281
282 seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
283 seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu ",
284 start,
285 end,
286 flags & VM_READ ? 'r' : '-',
287 flags & VM_WRITE ? 'w' : '-',
288 flags & VM_EXEC ? 'x' : '-',
289 flags & VM_MAYSHARE ? 's' : 'p',
290 pgoff,
291 MAJOR(dev), MINOR(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_path(m, &file->f_path, "\n");
300 goto done;
301 }
302
303 name = arch_vma_name(vma);
304 if (!name) {
305 pid_t tid;
306
307 if (!mm) {
308 name = "[vdso]";
309 goto done;
310 }
311
312 if (vma->vm_start <= mm->brk &&
313 vma->vm_end >= mm->start_brk) {
314 name = "[heap]";
315 goto done;
316 }
317
318 tid = vm_is_stack(task, vma, is_pid);
319
320 if (tid != 0) {
321 /*
322 * Thread stack in /proc/PID/task/TID/maps or
323 * the main process stack.
324 */
325 if (!is_pid || (vma->vm_start <= mm->start_stack &&
326 vma->vm_end >= mm->start_stack)) {
327 name = "[stack]";
328 } else {
329 /* Thread stack in /proc/PID/maps */
330 seq_pad(m, ' ');
331 seq_printf(m, "[stack:%d]", tid);
332 }
333 }
334 }
335
336done:
337 if (name) {
338 seq_pad(m, ' ');
339 seq_puts(m, name);
340 }
341 seq_putc(m, '\n');
342}
343
344static int show_map(struct seq_file *m, void *v, int is_pid)
345{
346 struct vm_area_struct *vma = v;
347 struct proc_maps_private *priv = m->private;
348 struct task_struct *task = priv->task;
349
350 show_map_vma(m, vma, is_pid);
351
352 if (m->count < m->size) /* vma is copied successfully */
353 m->version = (vma != get_gate_vma(task->mm))
354 ? vma->vm_start : 0;
355 return 0;
356}
357
358static int show_pid_map(struct seq_file *m, void *v)
359{
360 return show_map(m, v, 1);
361}
362
363static int show_tid_map(struct seq_file *m, void *v)
364{
365 return show_map(m, v, 0);
366}
367
368static const struct seq_operations proc_pid_maps_op = {
369 .start = m_start,
370 .next = m_next,
371 .stop = m_stop,
372 .show = show_pid_map
373};
374
375static const struct seq_operations proc_tid_maps_op = {
376 .start = m_start,
377 .next = m_next,
378 .stop = m_stop,
379 .show = show_tid_map
380};
381
382static int pid_maps_open(struct inode *inode, struct file *file)
383{
384 return do_maps_open(inode, file, &proc_pid_maps_op);
385}
386
387static int tid_maps_open(struct inode *inode, struct file *file)
388{
389 return do_maps_open(inode, file, &proc_tid_maps_op);
390}
391
392const struct file_operations proc_pid_maps_operations = {
393 .open = pid_maps_open,
394 .read = seq_read,
395 .llseek = seq_lseek,
396 .release = seq_release_private,
397};
398
399const struct file_operations proc_tid_maps_operations = {
400 .open = tid_maps_open,
401 .read = seq_read,
402 .llseek = seq_lseek,
403 .release = seq_release_private,
404};
405
406/*
407 * Proportional Set Size(PSS): my share of RSS.
408 *
409 * PSS of a process is the count of pages it has in memory, where each
410 * page is divided by the number of processes sharing it. So if a
411 * process has 1000 pages all to itself, and 1000 shared with one other
412 * process, its PSS will be 1500.
413 *
414 * To keep (accumulated) division errors low, we adopt a 64bit
415 * fixed-point pss counter to minimize division errors. So (pss >>
416 * PSS_SHIFT) would be the real byte count.
417 *
418 * A shift of 12 before division means (assuming 4K page size):
419 * - 1M 3-user-pages add up to 8KB errors;
420 * - supports mapcount up to 2^24, or 16M;
421 * - supports PSS up to 2^52 bytes, or 4PB.
422 */
423#define PSS_SHIFT 12
424
425#ifdef CONFIG_PROC_PAGE_MONITOR
426struct mem_size_stats {
427 struct vm_area_struct *vma;
428 unsigned long resident;
429 unsigned long shared_clean;
430 unsigned long shared_dirty;
431 unsigned long private_clean;
432 unsigned long private_dirty;
433 unsigned long referenced;
434 unsigned long anonymous;
435 unsigned long anonymous_thp;
436 unsigned long swap;
437 unsigned long nonlinear;
438 u64 pss;
439};
440
441
442static void smaps_pte_entry(pte_t ptent, unsigned long addr,
443 unsigned long ptent_size, struct mm_walk *walk)
444{
445 struct mem_size_stats *mss = walk->private;
446 struct vm_area_struct *vma = mss->vma;
447 pgoff_t pgoff = linear_page_index(vma, addr);
448 struct page *page = NULL;
449 int mapcount;
450
451 if (pte_present(ptent)) {
452 page = vm_normal_page(vma, addr, ptent);
453 } else if (is_swap_pte(ptent)) {
454 swp_entry_t swpent = pte_to_swp_entry(ptent);
455
456 if (!non_swap_entry(swpent))
457 mss->swap += ptent_size;
458 else if (is_migration_entry(swpent))
459 page = migration_entry_to_page(swpent);
460 } else if (pte_file(ptent)) {
461 if (pte_to_pgoff(ptent) != pgoff)
462 mss->nonlinear += ptent_size;
463 }
464
465 if (!page)
466 return;
467
468 if (PageAnon(page))
469 mss->anonymous += ptent_size;
470
471 if (page->index != pgoff)
472 mss->nonlinear += ptent_size;
473
474 mss->resident += ptent_size;
475 /* Accumulate the size in pages that have been accessed. */
476 if (pte_young(ptent) || PageReferenced(page))
477 mss->referenced += ptent_size;
478 mapcount = page_mapcount(page);
479 if (mapcount >= 2) {
480 if (pte_dirty(ptent) || PageDirty(page))
481 mss->shared_dirty += ptent_size;
482 else
483 mss->shared_clean += ptent_size;
484 mss->pss += (ptent_size << PSS_SHIFT) / mapcount;
485 } else {
486 if (pte_dirty(ptent) || PageDirty(page))
487 mss->private_dirty += ptent_size;
488 else
489 mss->private_clean += ptent_size;
490 mss->pss += (ptent_size << PSS_SHIFT);
491 }
492}
493
494static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
495 struct mm_walk *walk)
496{
497 struct mem_size_stats *mss = walk->private;
498 struct vm_area_struct *vma = mss->vma;
499 pte_t *pte;
500 spinlock_t *ptl;
501
502 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
503 smaps_pte_entry(*(pte_t *)pmd, addr, HPAGE_PMD_SIZE, walk);
504 spin_unlock(ptl);
505 mss->anonymous_thp += HPAGE_PMD_SIZE;
506 return 0;
507 }
508
509 if (pmd_trans_unstable(pmd))
510 return 0;
511 /*
512 * The mmap_sem held all the way back in m_start() is what
513 * keeps khugepaged out of here and from collapsing things
514 * in here.
515 */
516 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
517 for (; addr != end; pte++, addr += PAGE_SIZE)
518 smaps_pte_entry(*pte, addr, PAGE_SIZE, walk);
519 pte_unmap_unlock(pte - 1, ptl);
520 cond_resched();
521 return 0;
522}
523
524static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
525{
526 /*
527 * Don't forget to update Documentation/ on changes.
528 */
529 static const char mnemonics[BITS_PER_LONG][2] = {
530 /*
531 * In case if we meet a flag we don't know about.
532 */
533 [0 ... (BITS_PER_LONG-1)] = "??",
534
535 [ilog2(VM_READ)] = "rd",
536 [ilog2(VM_WRITE)] = "wr",
537 [ilog2(VM_EXEC)] = "ex",
538 [ilog2(VM_SHARED)] = "sh",
539 [ilog2(VM_MAYREAD)] = "mr",
540 [ilog2(VM_MAYWRITE)] = "mw",
541 [ilog2(VM_MAYEXEC)] = "me",
542 [ilog2(VM_MAYSHARE)] = "ms",
543 [ilog2(VM_GROWSDOWN)] = "gd",
544 [ilog2(VM_PFNMAP)] = "pf",
545 [ilog2(VM_DENYWRITE)] = "dw",
546 [ilog2(VM_LOCKED)] = "lo",
547 [ilog2(VM_IO)] = "io",
548 [ilog2(VM_SEQ_READ)] = "sr",
549 [ilog2(VM_RAND_READ)] = "rr",
550 [ilog2(VM_DONTCOPY)] = "dc",
551 [ilog2(VM_DONTEXPAND)] = "de",
552 [ilog2(VM_ACCOUNT)] = "ac",
553 [ilog2(VM_NORESERVE)] = "nr",
554 [ilog2(VM_HUGETLB)] = "ht",
555 [ilog2(VM_NONLINEAR)] = "nl",
556 [ilog2(VM_ARCH_1)] = "ar",
557 [ilog2(VM_DONTDUMP)] = "dd",
558#ifdef CONFIG_MEM_SOFT_DIRTY
559 [ilog2(VM_SOFTDIRTY)] = "sd",
560#endif
561 [ilog2(VM_MIXEDMAP)] = "mm",
562 [ilog2(VM_HUGEPAGE)] = "hg",
563 [ilog2(VM_NOHUGEPAGE)] = "nh",
564 [ilog2(VM_MERGEABLE)] = "mg",
565 };
566 size_t i;
567
568 seq_puts(m, "VmFlags: ");
569 for (i = 0; i < BITS_PER_LONG; i++) {
570 if (vma->vm_flags & (1UL << i)) {
571 seq_printf(m, "%c%c ",
572 mnemonics[i][0], mnemonics[i][1]);
573 }
574 }
575 seq_putc(m, '\n');
576}
577
578static int show_smap(struct seq_file *m, void *v, int is_pid)
579{
580 struct proc_maps_private *priv = m->private;
581 struct task_struct *task = priv->task;
582 struct vm_area_struct *vma = v;
583 struct mem_size_stats mss;
584 struct mm_walk smaps_walk = {
585 .pmd_entry = smaps_pte_range,
586 .mm = vma->vm_mm,
587 .private = &mss,
588 };
589
590 memset(&mss, 0, sizeof mss);
591 mss.vma = vma;
592 /* mmap_sem is held in m_start */
593 if (vma->vm_mm && !is_vm_hugetlb_page(vma))
594 walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk);
595
596 show_map_vma(m, vma, is_pid);
597
598 seq_printf(m,
599 "Size: %8lu kB\n"
600 "Rss: %8lu kB\n"
601 "Pss: %8lu kB\n"
602 "Shared_Clean: %8lu kB\n"
603 "Shared_Dirty: %8lu kB\n"
604 "Private_Clean: %8lu kB\n"
605 "Private_Dirty: %8lu kB\n"
606 "Referenced: %8lu kB\n"
607 "Anonymous: %8lu kB\n"
608 "AnonHugePages: %8lu kB\n"
609 "Swap: %8lu kB\n"
610 "KernelPageSize: %8lu kB\n"
611 "MMUPageSize: %8lu kB\n"
612 "Locked: %8lu kB\n",
613 (vma->vm_end - vma->vm_start) >> 10,
614 mss.resident >> 10,
615 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
616 mss.shared_clean >> 10,
617 mss.shared_dirty >> 10,
618 mss.private_clean >> 10,
619 mss.private_dirty >> 10,
620 mss.referenced >> 10,
621 mss.anonymous >> 10,
622 mss.anonymous_thp >> 10,
623 mss.swap >> 10,
624 vma_kernel_pagesize(vma) >> 10,
625 vma_mmu_pagesize(vma) >> 10,
626 (vma->vm_flags & VM_LOCKED) ?
627 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
628
629 if (vma->vm_flags & VM_NONLINEAR)
630 seq_printf(m, "Nonlinear: %8lu kB\n",
631 mss.nonlinear >> 10);
632
633 show_smap_vma_flags(m, vma);
634
635 if (m->count < m->size) /* vma is copied successfully */
636 m->version = (vma != get_gate_vma(task->mm))
637 ? vma->vm_start : 0;
638 return 0;
639}
640
641static int show_pid_smap(struct seq_file *m, void *v)
642{
643 return show_smap(m, v, 1);
644}
645
646static int show_tid_smap(struct seq_file *m, void *v)
647{
648 return show_smap(m, v, 0);
649}
650
651static const struct seq_operations proc_pid_smaps_op = {
652 .start = m_start,
653 .next = m_next,
654 .stop = m_stop,
655 .show = show_pid_smap
656};
657
658static const struct seq_operations proc_tid_smaps_op = {
659 .start = m_start,
660 .next = m_next,
661 .stop = m_stop,
662 .show = show_tid_smap
663};
664
665static int pid_smaps_open(struct inode *inode, struct file *file)
666{
667 return do_maps_open(inode, file, &proc_pid_smaps_op);
668}
669
670static int tid_smaps_open(struct inode *inode, struct file *file)
671{
672 return do_maps_open(inode, file, &proc_tid_smaps_op);
673}
674
675const struct file_operations proc_pid_smaps_operations = {
676 .open = pid_smaps_open,
677 .read = seq_read,
678 .llseek = seq_lseek,
679 .release = seq_release_private,
680};
681
682const struct file_operations proc_tid_smaps_operations = {
683 .open = tid_smaps_open,
684 .read = seq_read,
685 .llseek = seq_lseek,
686 .release = seq_release_private,
687};
688
689/*
690 * We do not want to have constant page-shift bits sitting in
691 * pagemap entries and are about to reuse them some time soon.
692 *
693 * Here's the "migration strategy":
694 * 1. when the system boots these bits remain what they are,
695 * but a warning about future change is printed in log;
696 * 2. once anyone clears soft-dirty bits via clear_refs file,
697 * these flag is set to denote, that user is aware of the
698 * new API and those page-shift bits change their meaning.
699 * The respective warning is printed in dmesg;
700 * 3. In a couple of releases we will remove all the mentions
701 * of page-shift in pagemap entries.
702 */
703
704static bool soft_dirty_cleared __read_mostly;
705
706enum clear_refs_types {
707 CLEAR_REFS_ALL = 1,
708 CLEAR_REFS_ANON,
709 CLEAR_REFS_MAPPED,
710 CLEAR_REFS_SOFT_DIRTY,
711 CLEAR_REFS_LAST,
712};
713
714struct clear_refs_private {
715 struct vm_area_struct *vma;
716 enum clear_refs_types type;
717};
718
719static inline void clear_soft_dirty(struct vm_area_struct *vma,
720 unsigned long addr, pte_t *pte)
721{
722#ifdef CONFIG_MEM_SOFT_DIRTY
723 /*
724 * The soft-dirty tracker uses #PF-s to catch writes
725 * to pages, so write-protect the pte as well. See the
726 * Documentation/vm/soft-dirty.txt for full description
727 * of how soft-dirty works.
728 */
729 pte_t ptent = *pte;
730
731 if (pte_present(ptent)) {
732 ptent = pte_wrprotect(ptent);
733 ptent = pte_clear_flags(ptent, _PAGE_SOFT_DIRTY);
734 } else if (is_swap_pte(ptent)) {
735 ptent = pte_swp_clear_soft_dirty(ptent);
736 } else if (pte_file(ptent)) {
737 ptent = pte_file_clear_soft_dirty(ptent);
738 }
739
740 if (vma->vm_flags & VM_SOFTDIRTY)
741 vma->vm_flags &= ~VM_SOFTDIRTY;
742
743 set_pte_at(vma->vm_mm, addr, pte, ptent);
744#endif
745}
746
747static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
748 unsigned long end, struct mm_walk *walk)
749{
750 struct clear_refs_private *cp = walk->private;
751 struct vm_area_struct *vma = cp->vma;
752 pte_t *pte, ptent;
753 spinlock_t *ptl;
754 struct page *page;
755
756 split_huge_page_pmd(vma, addr, pmd);
757 if (pmd_trans_unstable(pmd))
758 return 0;
759
760 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
761 for (; addr != end; pte++, addr += PAGE_SIZE) {
762 ptent = *pte;
763
764 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
765 clear_soft_dirty(vma, addr, pte);
766 continue;
767 }
768
769 if (!pte_present(ptent))
770 continue;
771
772 page = vm_normal_page(vma, addr, ptent);
773 if (!page)
774 continue;
775
776 /* Clear accessed and referenced bits. */
777 ptep_test_and_clear_young(vma, addr, pte);
778 ClearPageReferenced(page);
779 }
780 pte_unmap_unlock(pte - 1, ptl);
781 cond_resched();
782 return 0;
783}
784
785static ssize_t clear_refs_write(struct file *file, const char __user *buf,
786 size_t count, loff_t *ppos)
787{
788 struct task_struct *task;
789 char buffer[PROC_NUMBUF];
790 struct mm_struct *mm;
791 struct vm_area_struct *vma;
792 enum clear_refs_types type;
793 int itype;
794 int rv;
795
796 memset(buffer, 0, sizeof(buffer));
797 if (count > sizeof(buffer) - 1)
798 count = sizeof(buffer) - 1;
799 if (copy_from_user(buffer, buf, count))
800 return -EFAULT;
801 rv = kstrtoint(strstrip(buffer), 10, &itype);
802 if (rv < 0)
803 return rv;
804 type = (enum clear_refs_types)itype;
805 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
806 return -EINVAL;
807
808 if (type == CLEAR_REFS_SOFT_DIRTY) {
809 soft_dirty_cleared = true;
810 pr_warn_once("The pagemap bits 55-60 has changed their meaning! "
811 "See the linux/Documentation/vm/pagemap.txt for details.\n");
812 }
813
814 task = get_proc_task(file_inode(file));
815 if (!task)
816 return -ESRCH;
817 mm = get_task_mm(task);
818 if (mm) {
819 struct clear_refs_private cp = {
820 .type = type,
821 };
822 struct mm_walk clear_refs_walk = {
823 .pmd_entry = clear_refs_pte_range,
824 .mm = mm,
825 .private = &cp,
826 };
827 down_read(&mm->mmap_sem);
828 if (type == CLEAR_REFS_SOFT_DIRTY)
829 mmu_notifier_invalidate_range_start(mm, 0, -1);
830 for (vma = mm->mmap; vma; vma = vma->vm_next) {
831 cp.vma = vma;
832 if (is_vm_hugetlb_page(vma))
833 continue;
834 /*
835 * Writing 1 to /proc/pid/clear_refs affects all pages.
836 *
837 * Writing 2 to /proc/pid/clear_refs only affects
838 * Anonymous pages.
839 *
840 * Writing 3 to /proc/pid/clear_refs only affects file
841 * mapped pages.
842 */
843 if (type == CLEAR_REFS_ANON && vma->vm_file)
844 continue;
845 if (type == CLEAR_REFS_MAPPED && !vma->vm_file)
846 continue;
847 walk_page_range(vma->vm_start, vma->vm_end,
848 &clear_refs_walk);
849 }
850 if (type == CLEAR_REFS_SOFT_DIRTY)
851 mmu_notifier_invalidate_range_end(mm, 0, -1);
852 flush_tlb_mm(mm);
853 up_read(&mm->mmap_sem);
854 mmput(mm);
855 }
856 put_task_struct(task);
857
858 return count;
859}
860
861const struct file_operations proc_clear_refs_operations = {
862 .write = clear_refs_write,
863 .llseek = noop_llseek,
864};
865
866typedef struct {
867 u64 pme;
868} pagemap_entry_t;
869
870struct pagemapread {
871 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */
872 pagemap_entry_t *buffer;
873 bool v2;
874};
875
876#define PAGEMAP_WALK_SIZE (PMD_SIZE)
877#define PAGEMAP_WALK_MASK (PMD_MASK)
878
879#define PM_ENTRY_BYTES sizeof(pagemap_entry_t)
880#define PM_STATUS_BITS 3
881#define PM_STATUS_OFFSET (64 - PM_STATUS_BITS)
882#define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
883#define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
884#define PM_PSHIFT_BITS 6
885#define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
886#define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
887#define __PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
888#define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1)
889#define PM_PFRAME(x) ((x) & PM_PFRAME_MASK)
890/* in "new" pagemap pshift bits are occupied with more status bits */
891#define PM_STATUS2(v2, x) (__PM_PSHIFT(v2 ? x : PAGE_SHIFT))
892
893#define __PM_SOFT_DIRTY (1LL)
894#define PM_PRESENT PM_STATUS(4LL)
895#define PM_SWAP PM_STATUS(2LL)
896#define PM_FILE PM_STATUS(1LL)
897#define PM_NOT_PRESENT(v2) PM_STATUS2(v2, 0)
898#define PM_END_OF_BUFFER 1
899
900static inline pagemap_entry_t make_pme(u64 val)
901{
902 return (pagemap_entry_t) { .pme = val };
903}
904
905static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
906 struct pagemapread *pm)
907{
908 pm->buffer[pm->pos++] = *pme;
909 if (pm->pos >= pm->len)
910 return PM_END_OF_BUFFER;
911 return 0;
912}
913
914static int pagemap_pte_hole(unsigned long start, unsigned long end,
915 struct mm_walk *walk)
916{
917 struct pagemapread *pm = walk->private;
918 unsigned long addr;
919 int err = 0;
920 pagemap_entry_t pme = make_pme(PM_NOT_PRESENT(pm->v2));
921
922 for (addr = start; addr < end; addr += PAGE_SIZE) {
923 err = add_to_pagemap(addr, &pme, pm);
924 if (err)
925 break;
926 }
927 return err;
928}
929
930static void pte_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm,
931 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
932{
933 u64 frame, flags;
934 struct page *page = NULL;
935 int flags2 = 0;
936
937 if (pte_present(pte)) {
938 frame = pte_pfn(pte);
939 flags = PM_PRESENT;
940 page = vm_normal_page(vma, addr, pte);
941 if (pte_soft_dirty(pte))
942 flags2 |= __PM_SOFT_DIRTY;
943 } else if (is_swap_pte(pte)) {
944 swp_entry_t entry;
945 if (pte_swp_soft_dirty(pte))
946 flags2 |= __PM_SOFT_DIRTY;
947 entry = pte_to_swp_entry(pte);
948 frame = swp_type(entry) |
949 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
950 flags = PM_SWAP;
951 if (is_migration_entry(entry))
952 page = migration_entry_to_page(entry);
953 } else {
954 if (vma->vm_flags & VM_SOFTDIRTY)
955 flags2 |= __PM_SOFT_DIRTY;
956 *pme = make_pme(PM_NOT_PRESENT(pm->v2) | PM_STATUS2(pm->v2, flags2));
957 return;
958 }
959
960 if (page && !PageAnon(page))
961 flags |= PM_FILE;
962 if ((vma->vm_flags & VM_SOFTDIRTY))
963 flags2 |= __PM_SOFT_DIRTY;
964
965 *pme = make_pme(PM_PFRAME(frame) | PM_STATUS2(pm->v2, flags2) | flags);
966}
967
968#ifdef CONFIG_TRANSPARENT_HUGEPAGE
969static void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm,
970 pmd_t pmd, int offset, int pmd_flags2)
971{
972 /*
973 * Currently pmd for thp is always present because thp can not be
974 * swapped-out, migrated, or HWPOISONed (split in such cases instead.)
975 * This if-check is just to prepare for future implementation.
976 */
977 if (pmd_present(pmd))
978 *pme = make_pme(PM_PFRAME(pmd_pfn(pmd) + offset)
979 | PM_STATUS2(pm->v2, pmd_flags2) | PM_PRESENT);
980 else
981 *pme = make_pme(PM_NOT_PRESENT(pm->v2) | PM_STATUS2(pm->v2, pmd_flags2));
982}
983#else
984static inline void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm,
985 pmd_t pmd, int offset, int pmd_flags2)
986{
987}
988#endif
989
990static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
991 struct mm_walk *walk)
992{
993 struct vm_area_struct *vma;
994 struct pagemapread *pm = walk->private;
995 spinlock_t *ptl;
996 pte_t *pte;
997 int err = 0;
998 pagemap_entry_t pme = make_pme(PM_NOT_PRESENT(pm->v2));
999
1000 /* find the first VMA at or above 'addr' */
1001 vma = find_vma(walk->mm, addr);
1002 if (vma && pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
1003 int pmd_flags2;
1004
1005 if ((vma->vm_flags & VM_SOFTDIRTY) || pmd_soft_dirty(*pmd))
1006 pmd_flags2 = __PM_SOFT_DIRTY;
1007 else
1008 pmd_flags2 = 0;
1009
1010 for (; addr != end; addr += PAGE_SIZE) {
1011 unsigned long offset;
1012
1013 offset = (addr & ~PAGEMAP_WALK_MASK) >>
1014 PAGE_SHIFT;
1015 thp_pmd_to_pagemap_entry(&pme, pm, *pmd, offset, pmd_flags2);
1016 err = add_to_pagemap(addr, &pme, pm);
1017 if (err)
1018 break;
1019 }
1020 spin_unlock(ptl);
1021 return err;
1022 }
1023
1024 if (pmd_trans_unstable(pmd))
1025 return 0;
1026 for (; addr != end; addr += PAGE_SIZE) {
1027 int flags2;
1028
1029 /* check to see if we've left 'vma' behind
1030 * and need a new, higher one */
1031 if (vma && (addr >= vma->vm_end)) {
1032 vma = find_vma(walk->mm, addr);
1033 if (vma && (vma->vm_flags & VM_SOFTDIRTY))
1034 flags2 = __PM_SOFT_DIRTY;
1035 else
1036 flags2 = 0;
1037 pme = make_pme(PM_NOT_PRESENT(pm->v2) | PM_STATUS2(pm->v2, flags2));
1038 }
1039
1040 /* check that 'vma' actually covers this address,
1041 * and that it isn't a huge page vma */
1042 if (vma && (vma->vm_start <= addr) &&
1043 !is_vm_hugetlb_page(vma)) {
1044 pte = pte_offset_map(pmd, addr);
1045 pte_to_pagemap_entry(&pme, pm, vma, addr, *pte);
1046 /* unmap before userspace copy */
1047 pte_unmap(pte);
1048 }
1049 err = add_to_pagemap(addr, &pme, pm);
1050 if (err)
1051 return err;
1052 }
1053
1054 cond_resched();
1055
1056 return err;
1057}
1058
1059#ifdef CONFIG_HUGETLB_PAGE
1060static void huge_pte_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm,
1061 pte_t pte, int offset, int flags2)
1062{
1063 if (pte_present(pte))
1064 *pme = make_pme(PM_PFRAME(pte_pfn(pte) + offset) |
1065 PM_STATUS2(pm->v2, flags2) |
1066 PM_PRESENT);
1067 else
1068 *pme = make_pme(PM_NOT_PRESENT(pm->v2) |
1069 PM_STATUS2(pm->v2, flags2));
1070}
1071
1072/* This function walks within one hugetlb entry in the single call */
1073static int pagemap_hugetlb_range(pte_t *pte, unsigned long hmask,
1074 unsigned long addr, unsigned long end,
1075 struct mm_walk *walk)
1076{
1077 struct pagemapread *pm = walk->private;
1078 struct vm_area_struct *vma;
1079 int err = 0;
1080 int flags2;
1081 pagemap_entry_t pme;
1082
1083 vma = find_vma(walk->mm, addr);
1084 WARN_ON_ONCE(!vma);
1085
1086 if (vma && (vma->vm_flags & VM_SOFTDIRTY))
1087 flags2 = __PM_SOFT_DIRTY;
1088 else
1089 flags2 = 0;
1090
1091 for (; addr != end; addr += PAGE_SIZE) {
1092 int offset = (addr & ~hmask) >> PAGE_SHIFT;
1093 huge_pte_to_pagemap_entry(&pme, pm, *pte, offset, flags2);
1094 err = add_to_pagemap(addr, &pme, pm);
1095 if (err)
1096 return err;
1097 }
1098
1099 cond_resched();
1100
1101 return err;
1102}
1103#endif /* HUGETLB_PAGE */
1104
1105/*
1106 * /proc/pid/pagemap - an array mapping virtual pages to pfns
1107 *
1108 * For each page in the address space, this file contains one 64-bit entry
1109 * consisting of the following:
1110 *
1111 * Bits 0-54 page frame number (PFN) if present
1112 * Bits 0-4 swap type if swapped
1113 * Bits 5-54 swap offset if swapped
1114 * Bits 55-60 page shift (page size = 1<<page shift)
1115 * Bit 61 page is file-page or shared-anon
1116 * Bit 62 page swapped
1117 * Bit 63 page present
1118 *
1119 * If the page is not present but in swap, then the PFN contains an
1120 * encoding of the swap file number and the page's offset into the
1121 * swap. Unmapped pages return a null PFN. This allows determining
1122 * precisely which pages are mapped (or in swap) and comparing mapped
1123 * pages between processes.
1124 *
1125 * Efficient users of this interface will use /proc/pid/maps to
1126 * determine which areas of memory are actually mapped and llseek to
1127 * skip over unmapped regions.
1128 */
1129static ssize_t pagemap_read(struct file *file, char __user *buf,
1130 size_t count, loff_t *ppos)
1131{
1132 struct task_struct *task = get_proc_task(file_inode(file));
1133 struct mm_struct *mm;
1134 struct pagemapread pm;
1135 int ret = -ESRCH;
1136 struct mm_walk pagemap_walk = {};
1137 unsigned long src;
1138 unsigned long svpfn;
1139 unsigned long start_vaddr;
1140 unsigned long end_vaddr;
1141 int copied = 0;
1142
1143 if (!task)
1144 goto out;
1145
1146 ret = -EINVAL;
1147 /* file position must be aligned */
1148 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1149 goto out_task;
1150
1151 ret = 0;
1152 if (!count)
1153 goto out_task;
1154
1155 pm.v2 = soft_dirty_cleared;
1156 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1157 pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY);
1158 ret = -ENOMEM;
1159 if (!pm.buffer)
1160 goto out_task;
1161
1162 mm = mm_access(task, PTRACE_MODE_READ);
1163 ret = PTR_ERR(mm);
1164 if (!mm || IS_ERR(mm))
1165 goto out_free;
1166
1167 pagemap_walk.pmd_entry = pagemap_pte_range;
1168 pagemap_walk.pte_hole = pagemap_pte_hole;
1169#ifdef CONFIG_HUGETLB_PAGE
1170 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1171#endif
1172 pagemap_walk.mm = mm;
1173 pagemap_walk.private = ±
1174
1175 src = *ppos;
1176 svpfn = src / PM_ENTRY_BYTES;
1177 start_vaddr = svpfn << PAGE_SHIFT;
1178 end_vaddr = TASK_SIZE_OF(task);
1179
1180 /* watch out for wraparound */
1181 if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
1182 start_vaddr = end_vaddr;
1183
1184 /*
1185 * The odds are that this will stop walking way
1186 * before end_vaddr, because the length of the
1187 * user buffer is tracked in "pm", and the walk
1188 * will stop when we hit the end of the buffer.
1189 */
1190 ret = 0;
1191 while (count && (start_vaddr < end_vaddr)) {
1192 int len;
1193 unsigned long end;
1194
1195 pm.pos = 0;
1196 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1197 /* overflow ? */
1198 if (end < start_vaddr || end > end_vaddr)
1199 end = end_vaddr;
1200 down_read(&mm->mmap_sem);
1201 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1202 up_read(&mm->mmap_sem);
1203 start_vaddr = end;
1204
1205 len = min(count, PM_ENTRY_BYTES * pm.pos);
1206 if (copy_to_user(buf, pm.buffer, len)) {
1207 ret = -EFAULT;
1208 goto out_mm;
1209 }
1210 copied += len;
1211 buf += len;
1212 count -= len;
1213 }
1214 *ppos += copied;
1215 if (!ret || ret == PM_END_OF_BUFFER)
1216 ret = copied;
1217
1218out_mm:
1219 mmput(mm);
1220out_free:
1221 kfree(pm.buffer);
1222out_task:
1223 put_task_struct(task);
1224out:
1225 return ret;
1226}
1227
1228static int pagemap_open(struct inode *inode, struct file *file)
1229{
1230 pr_warn_once("Bits 55-60 of /proc/PID/pagemap entries are about "
1231 "to stop being page-shift some time soon. See the "
1232 "linux/Documentation/vm/pagemap.txt for details.\n");
1233 return 0;
1234}
1235
1236const struct file_operations proc_pagemap_operations = {
1237 .llseek = mem_lseek, /* borrow this */
1238 .read = pagemap_read,
1239 .open = pagemap_open,
1240};
1241#endif /* CONFIG_PROC_PAGE_MONITOR */
1242
1243#ifdef CONFIG_NUMA
1244
1245struct numa_maps {
1246 struct vm_area_struct *vma;
1247 unsigned long pages;
1248 unsigned long anon;
1249 unsigned long active;
1250 unsigned long writeback;
1251 unsigned long mapcount_max;
1252 unsigned long dirty;
1253 unsigned long swapcache;
1254 unsigned long node[MAX_NUMNODES];
1255};
1256
1257struct numa_maps_private {
1258 struct proc_maps_private proc_maps;
1259 struct numa_maps md;
1260};
1261
1262static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1263 unsigned long nr_pages)
1264{
1265 int count = page_mapcount(page);
1266
1267 md->pages += nr_pages;
1268 if (pte_dirty || PageDirty(page))
1269 md->dirty += nr_pages;
1270
1271 if (PageSwapCache(page))
1272 md->swapcache += nr_pages;
1273
1274 if (PageActive(page) || PageUnevictable(page))
1275 md->active += nr_pages;
1276
1277 if (PageWriteback(page))
1278 md->writeback += nr_pages;
1279
1280 if (PageAnon(page))
1281 md->anon += nr_pages;
1282
1283 if (count > md->mapcount_max)
1284 md->mapcount_max = count;
1285
1286 md->node[page_to_nid(page)] += nr_pages;
1287}
1288
1289static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1290 unsigned long addr)
1291{
1292 struct page *page;
1293 int nid;
1294
1295 if (!pte_present(pte))
1296 return NULL;
1297
1298 page = vm_normal_page(vma, addr, pte);
1299 if (!page)
1300 return NULL;
1301
1302 if (PageReserved(page))
1303 return NULL;
1304
1305 nid = page_to_nid(page);
1306 if (!node_isset(nid, node_states[N_MEMORY]))
1307 return NULL;
1308
1309 return page;
1310}
1311
1312static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1313 unsigned long end, struct mm_walk *walk)
1314{
1315 struct numa_maps *md;
1316 spinlock_t *ptl;
1317 pte_t *orig_pte;
1318 pte_t *pte;
1319
1320 md = walk->private;
1321
1322 if (pmd_trans_huge_lock(pmd, md->vma, &ptl) == 1) {
1323 pte_t huge_pte = *(pte_t *)pmd;
1324 struct page *page;
1325
1326 page = can_gather_numa_stats(huge_pte, md->vma, addr);
1327 if (page)
1328 gather_stats(page, md, pte_dirty(huge_pte),
1329 HPAGE_PMD_SIZE/PAGE_SIZE);
1330 spin_unlock(ptl);
1331 return 0;
1332 }
1333
1334 if (pmd_trans_unstable(pmd))
1335 return 0;
1336 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1337 do {
1338 struct page *page = can_gather_numa_stats(*pte, md->vma, addr);
1339 if (!page)
1340 continue;
1341 gather_stats(page, md, pte_dirty(*pte), 1);
1342
1343 } while (pte++, addr += PAGE_SIZE, addr != end);
1344 pte_unmap_unlock(orig_pte, ptl);
1345 return 0;
1346}
1347#ifdef CONFIG_HUGETLB_PAGE
1348static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
1349 unsigned long addr, unsigned long end, struct mm_walk *walk)
1350{
1351 struct numa_maps *md;
1352 struct page *page;
1353
1354 if (!pte_present(*pte))
1355 return 0;
1356
1357 page = pte_page(*pte);
1358 if (!page)
1359 return 0;
1360
1361 md = walk->private;
1362 gather_stats(page, md, pte_dirty(*pte), 1);
1363 return 0;
1364}
1365
1366#else
1367static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
1368 unsigned long addr, unsigned long end, struct mm_walk *walk)
1369{
1370 return 0;
1371}
1372#endif
1373
1374/*
1375 * Display pages allocated per node and memory policy via /proc.
1376 */
1377static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1378{
1379 struct numa_maps_private *numa_priv = m->private;
1380 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1381 struct vm_area_struct *vma = v;
1382 struct numa_maps *md = &numa_priv->md;
1383 struct file *file = vma->vm_file;
1384 struct task_struct *task = proc_priv->task;
1385 struct mm_struct *mm = vma->vm_mm;
1386 struct mm_walk walk = {};
1387 struct mempolicy *pol;
1388 char buffer[64];
1389 int nid;
1390
1391 if (!mm)
1392 return 0;
1393
1394 /* Ensure we start with an empty set of numa_maps statistics. */
1395 memset(md, 0, sizeof(*md));
1396
1397 md->vma = vma;
1398
1399 walk.hugetlb_entry = gather_hugetbl_stats;
1400 walk.pmd_entry = gather_pte_stats;
1401 walk.private = md;
1402 walk.mm = mm;
1403
1404 pol = get_vma_policy(task, vma, vma->vm_start);
1405 mpol_to_str(buffer, sizeof(buffer), pol);
1406 mpol_cond_put(pol);
1407
1408 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1409
1410 if (file) {
1411 seq_printf(m, " file=");
1412 seq_path(m, &file->f_path, "\n\t= ");
1413 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1414 seq_printf(m, " heap");
1415 } else {
1416 pid_t tid = vm_is_stack(task, vma, is_pid);
1417 if (tid != 0) {
1418 /*
1419 * Thread stack in /proc/PID/task/TID/maps or
1420 * the main process stack.
1421 */
1422 if (!is_pid || (vma->vm_start <= mm->start_stack &&
1423 vma->vm_end >= mm->start_stack))
1424 seq_printf(m, " stack");
1425 else
1426 seq_printf(m, " stack:%d", tid);
1427 }
1428 }
1429
1430 if (is_vm_hugetlb_page(vma))
1431 seq_printf(m, " huge");
1432
1433 walk_page_range(vma->vm_start, vma->vm_end, &walk);
1434
1435 if (!md->pages)
1436 goto out;
1437
1438 if (md->anon)
1439 seq_printf(m, " anon=%lu", md->anon);
1440
1441 if (md->dirty)
1442 seq_printf(m, " dirty=%lu", md->dirty);
1443
1444 if (md->pages != md->anon && md->pages != md->dirty)
1445 seq_printf(m, " mapped=%lu", md->pages);
1446
1447 if (md->mapcount_max > 1)
1448 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1449
1450 if (md->swapcache)
1451 seq_printf(m, " swapcache=%lu", md->swapcache);
1452
1453 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1454 seq_printf(m, " active=%lu", md->active);
1455
1456 if (md->writeback)
1457 seq_printf(m, " writeback=%lu", md->writeback);
1458
1459 for_each_node_state(nid, N_MEMORY)
1460 if (md->node[nid])
1461 seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1462out:
1463 seq_putc(m, '\n');
1464
1465 if (m->count < m->size)
1466 m->version = (vma != proc_priv->tail_vma) ? vma->vm_start : 0;
1467 return 0;
1468}
1469
1470static int show_pid_numa_map(struct seq_file *m, void *v)
1471{
1472 return show_numa_map(m, v, 1);
1473}
1474
1475static int show_tid_numa_map(struct seq_file *m, void *v)
1476{
1477 return show_numa_map(m, v, 0);
1478}
1479
1480static const struct seq_operations proc_pid_numa_maps_op = {
1481 .start = m_start,
1482 .next = m_next,
1483 .stop = m_stop,
1484 .show = show_pid_numa_map,
1485};
1486
1487static const struct seq_operations proc_tid_numa_maps_op = {
1488 .start = m_start,
1489 .next = m_next,
1490 .stop = m_stop,
1491 .show = show_tid_numa_map,
1492};
1493
1494static int numa_maps_open(struct inode *inode, struct file *file,
1495 const struct seq_operations *ops)
1496{
1497 struct numa_maps_private *priv;
1498 int ret = -ENOMEM;
1499 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
1500 if (priv) {
1501 priv->proc_maps.pid = proc_pid(inode);
1502 ret = seq_open(file, ops);
1503 if (!ret) {
1504 struct seq_file *m = file->private_data;
1505 m->private = priv;
1506 } else {
1507 kfree(priv);
1508 }
1509 }
1510 return ret;
1511}
1512
1513static int pid_numa_maps_open(struct inode *inode, struct file *file)
1514{
1515 return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1516}
1517
1518static int tid_numa_maps_open(struct inode *inode, struct file *file)
1519{
1520 return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1521}
1522
1523const struct file_operations proc_pid_numa_maps_operations = {
1524 .open = pid_numa_maps_open,
1525 .read = seq_read,
1526 .llseek = seq_lseek,
1527 .release = seq_release_private,
1528};
1529
1530const struct file_operations proc_tid_numa_maps_operations = {
1531 .open = tid_numa_maps_open,
1532 .read = seq_read,
1533 .llseek = seq_lseek,
1534 .release = seq_release_private,
1535};
1536#endif /* CONFIG_NUMA */