projects / deliverable / linux.git / blob
? search:


a353b4c6e86e5d24007ac4f9ead1d80d577037d6
[deliverable/linux.git] / fs / proc / task_mmu.c
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
24 void 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
86 unsigned long task_vsize(struct mm_struct *mm)
87 {
88 return PAGE_SIZE * mm->total_vm;
89 }
90
91 unsigned 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 */
108 static 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 }
117 static void release_task_mempolicy(struct proc_maps_private *priv)
118 {
119 mpol_put(priv->task_mempolicy);
120 }
121 #else
122 static void hold_task_mempolicy(struct proc_maps_private *priv)
123 {
124 }
125 static void release_task_mempolicy(struct proc_maps_private *priv)
126 {
127 }
128 #endif
129
130 static 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
139 static struct vm_area_struct *
140 m_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
147 static 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
153 static 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
200 static 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
212 static 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
224 static 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
244 static 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
255 static 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 static pid_t pid_of_stack(struct proc_maps_private *priv,
263 struct vm_area_struct *vma, bool is_pid)
264 {
265 struct inode *inode = priv->inode;
266 struct task_struct *task;
267 pid_t ret = 0;
268
269 rcu_read_lock();
270 task = pid_task(proc_pid(inode), PIDTYPE_PID);
271 if (task) {
272 task = task_of_stack(task, vma, is_pid);
273 if (task)
274 ret = task_pid_nr_ns(task, inode->i_sb->s_fs_info);
275 }
276 rcu_read_unlock();
277
278 return ret;
279 }
280
281 static void
282 show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
283 {
284 struct mm_struct *mm = vma->vm_mm;
285 struct file *file = vma->vm_file;
286 struct proc_maps_private *priv = m->private;
287 vm_flags_t flags = vma->vm_flags;
288 unsigned long ino = 0;
289 unsigned long long pgoff = 0;
290 unsigned long start, end;
291 dev_t dev = 0;
292 const char *name = NULL;
293
294 if (file) {
295 struct inode *inode = file_inode(vma->vm_file);
296 dev = inode->i_sb->s_dev;
297 ino = inode->i_ino;
298 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
299 }
300
301 /* We don't show the stack guard page in /proc/maps */
302 start = vma->vm_start;
303 if (stack_guard_page_start(vma, start))
304 start += PAGE_SIZE;
305 end = vma->vm_end;
306 if (stack_guard_page_end(vma, end))
307 end -= PAGE_SIZE;
308
309 seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
310 seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu ",
311 start,
312 end,
313 flags & VM_READ ? 'r' : '-',
314 flags & VM_WRITE ? 'w' : '-',
315 flags & VM_EXEC ? 'x' : '-',
316 flags & VM_MAYSHARE ? 's' : 'p',
317 pgoff,
318 MAJOR(dev), MINOR(dev), ino);
319
320 /*
321 * Print the dentry name for named mappings, and a
322 * special [heap] marker for the heap:
323 */
324 if (file) {
325 seq_pad(m, ' ');
326 seq_file_path(m, file, "\n");
327 goto done;
328 }
329
330 if (vma->vm_ops && vma->vm_ops->name) {
331 name = vma->vm_ops->name(vma);
332 if (name)
333 goto done;
334 }
335
336 name = arch_vma_name(vma);
337 if (!name) {
338 pid_t tid;
339
340 if (!mm) {
341 name = "[vdso]";
342 goto done;
343 }
344
345 if (vma->vm_start <= mm->brk &&
346 vma->vm_end >= mm->start_brk) {
347 name = "[heap]";
348 goto done;
349 }
350
351 tid = pid_of_stack(priv, vma, is_pid);
352 if (tid != 0) {
353 /*
354 * Thread stack in /proc/PID/task/TID/maps or
355 * the main process stack.
356 */
357 if (!is_pid || (vma->vm_start <= mm->start_stack &&
358 vma->vm_end >= mm->start_stack)) {
359 name = "[stack]";
360 } else {
361 /* Thread stack in /proc/PID/maps */
362 seq_pad(m, ' ');
363 seq_printf(m, "[stack:%d]", tid);
364 }
365 }
366 }
367
368 done:
369 if (name) {
370 seq_pad(m, ' ');
371 seq_puts(m, name);
372 }
373 seq_putc(m, '\n');
374 }
375
376 static int show_map(struct seq_file *m, void *v, int is_pid)
377 {
378 show_map_vma(m, v, is_pid);
379 m_cache_vma(m, v);
380 return 0;
381 }
382
383 static int show_pid_map(struct seq_file *m, void *v)
384 {
385 return show_map(m, v, 1);
386 }
387
388 static int show_tid_map(struct seq_file *m, void *v)
389 {
390 return show_map(m, v, 0);
391 }
392
393 static const struct seq_operations proc_pid_maps_op = {
394 .start = m_start,
395 .next = m_next,
396 .stop = m_stop,
397 .show = show_pid_map
398 };
399
400 static const struct seq_operations proc_tid_maps_op = {
401 .start = m_start,
402 .next = m_next,
403 .stop = m_stop,
404 .show = show_tid_map
405 };
406
407 static int pid_maps_open(struct inode *inode, struct file *file)
408 {
409 return do_maps_open(inode, file, &proc_pid_maps_op);
410 }
411
412 static int tid_maps_open(struct inode *inode, struct file *file)
413 {
414 return do_maps_open(inode, file, &proc_tid_maps_op);
415 }
416
417 const struct file_operations proc_pid_maps_operations = {
418 .open = pid_maps_open,
419 .read = seq_read,
420 .llseek = seq_lseek,
421 .release = proc_map_release,
422 };
423
424 const struct file_operations proc_tid_maps_operations = {
425 .open = tid_maps_open,
426 .read = seq_read,
427 .llseek = seq_lseek,
428 .release = proc_map_release,
429 };
430
431 /*
432 * Proportional Set Size(PSS): my share of RSS.
433 *
434 * PSS of a process is the count of pages it has in memory, where each
435 * page is divided by the number of processes sharing it. So if a
436 * process has 1000 pages all to itself, and 1000 shared with one other
437 * process, its PSS will be 1500.
438 *
439 * To keep (accumulated) division errors low, we adopt a 64bit
440 * fixed-point pss counter to minimize division errors. So (pss >>
441 * PSS_SHIFT) would be the real byte count.
442 *
443 * A shift of 12 before division means (assuming 4K page size):
444 * - 1M 3-user-pages add up to 8KB errors;
445 * - supports mapcount up to 2^24, or 16M;
446 * - supports PSS up to 2^52 bytes, or 4PB.
447 */
448 #define PSS_SHIFT 12
449
450 #ifdef CONFIG_PROC_PAGE_MONITOR
451 struct mem_size_stats {
452 unsigned long resident;
453 unsigned long shared_clean;
454 unsigned long shared_dirty;
455 unsigned long private_clean;
456 unsigned long private_dirty;
457 unsigned long referenced;
458 unsigned long anonymous;
459 unsigned long anonymous_thp;
460 unsigned long swap;
461 unsigned long shared_hugetlb;
462 unsigned long private_hugetlb;
463 u64 pss;
464 u64 swap_pss;
465 bool check_shmem_swap;
466 };
467
468 static void smaps_account(struct mem_size_stats *mss, struct page *page,
469 unsigned long size, bool young, bool dirty)
470 {
471 int mapcount;
472
473 if (PageAnon(page))
474 mss->anonymous += size;
475
476 mss->resident += size;
477 /* Accumulate the size in pages that have been accessed. */
478 if (young || page_is_young(page) || PageReferenced(page))
479 mss->referenced += size;
480 mapcount = page_mapcount(page);
481 if (mapcount >= 2) {
482 u64 pss_delta;
483
484 if (dirty || PageDirty(page))
485 mss->shared_dirty += size;
486 else
487 mss->shared_clean += size;
488 pss_delta = (u64)size << PSS_SHIFT;
489 do_div(pss_delta, mapcount);
490 mss->pss += pss_delta;
491 } else {
492 if (dirty || PageDirty(page))
493 mss->private_dirty += size;
494 else
495 mss->private_clean += size;
496 mss->pss += (u64)size << PSS_SHIFT;
497 }
498 }
499
500 #ifdef CONFIG_SHMEM
501 static int smaps_pte_hole(unsigned long addr, unsigned long end,
502 struct mm_walk *walk)
503 {
504 struct mem_size_stats *mss = walk->private;
505
506 mss->swap += shmem_partial_swap_usage(
507 walk->vma->vm_file->f_mapping, addr, end);
508
509 return 0;
510 }
511 #endif
512
513 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
514 struct mm_walk *walk)
515 {
516 struct mem_size_stats *mss = walk->private;
517 struct vm_area_struct *vma = walk->vma;
518 struct page *page = NULL;
519
520 if (pte_present(*pte)) {
521 page = vm_normal_page(vma, addr, *pte);
522 } else if (is_swap_pte(*pte)) {
523 swp_entry_t swpent = pte_to_swp_entry(*pte);
524
525 if (!non_swap_entry(swpent)) {
526 int mapcount;
527
528 mss->swap += PAGE_SIZE;
529 mapcount = swp_swapcount(swpent);
530 if (mapcount >= 2) {
531 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
532
533 do_div(pss_delta, mapcount);
534 mss->swap_pss += pss_delta;
535 } else {
536 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
537 }
538 } else if (is_migration_entry(swpent))
539 page = migration_entry_to_page(swpent);
540 } else if (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap
541 && pte_none(*pte))) {
542 page = find_get_entry(vma->vm_file->f_mapping,
543 linear_page_index(vma, addr));
544 if (!page)
545 return;
546
547 if (radix_tree_exceptional_entry(page))
548 mss->swap += PAGE_SIZE;
549 else
550 page_cache_release(page);
551
552 return;
553 }
554
555 if (!page)
556 return;
557 smaps_account(mss, page, PAGE_SIZE, pte_young(*pte), pte_dirty(*pte));
558 }
559
560 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
561 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
562 struct mm_walk *walk)
563 {
564 struct mem_size_stats *mss = walk->private;
565 struct vm_area_struct *vma = walk->vma;
566 struct page *page;
567
568 /* FOLL_DUMP will return -EFAULT on huge zero page */
569 page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
570 if (IS_ERR_OR_NULL(page))
571 return;
572 mss->anonymous_thp += HPAGE_PMD_SIZE;
573 smaps_account(mss, page, HPAGE_PMD_SIZE,
574 pmd_young(*pmd), pmd_dirty(*pmd));
575 }
576 #else
577 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
578 struct mm_walk *walk)
579 {
580 }
581 #endif
582
583 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
584 struct mm_walk *walk)
585 {
586 struct vm_area_struct *vma = walk->vma;
587 pte_t *pte;
588 spinlock_t *ptl;
589
590 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
591 smaps_pmd_entry(pmd, addr, walk);
592 spin_unlock(ptl);
593 return 0;
594 }
595
596 if (pmd_trans_unstable(pmd))
597 return 0;
598 /*
599 * The mmap_sem held all the way back in m_start() is what
600 * keeps khugepaged out of here and from collapsing things
601 * in here.
602 */
603 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
604 for (; addr != end; pte++, addr += PAGE_SIZE)
605 smaps_pte_entry(pte, addr, walk);
606 pte_unmap_unlock(pte - 1, ptl);
607 cond_resched();
608 return 0;
609 }
610
611 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
612 {
613 /*
614 * Don't forget to update Documentation/ on changes.
615 */
616 static const char mnemonics[BITS_PER_LONG][2] = {
617 /*
618 * In case if we meet a flag we don't know about.
619 */
620 [0 ... (BITS_PER_LONG-1)] = "??",
621
622 [ilog2(VM_READ)] = "rd",
623 [ilog2(VM_WRITE)] = "wr",
624 [ilog2(VM_EXEC)] = "ex",
625 [ilog2(VM_SHARED)] = "sh",
626 [ilog2(VM_MAYREAD)] = "mr",
627 [ilog2(VM_MAYWRITE)] = "mw",
628 [ilog2(VM_MAYEXEC)] = "me",
629 [ilog2(VM_MAYSHARE)] = "ms",
630 [ilog2(VM_GROWSDOWN)] = "gd",
631 [ilog2(VM_PFNMAP)] = "pf",
632 [ilog2(VM_DENYWRITE)] = "dw",
633 #ifdef CONFIG_X86_INTEL_MPX
634 [ilog2(VM_MPX)] = "mp",
635 #endif
636 [ilog2(VM_LOCKED)] = "lo",
637 [ilog2(VM_IO)] = "io",
638 [ilog2(VM_SEQ_READ)] = "sr",
639 [ilog2(VM_RAND_READ)] = "rr",
640 [ilog2(VM_DONTCOPY)] = "dc",
641 [ilog2(VM_DONTEXPAND)] = "de",
642 [ilog2(VM_ACCOUNT)] = "ac",
643 [ilog2(VM_NORESERVE)] = "nr",
644 [ilog2(VM_HUGETLB)] = "ht",
645 [ilog2(VM_ARCH_1)] = "ar",
646 [ilog2(VM_DONTDUMP)] = "dd",
647 #ifdef CONFIG_MEM_SOFT_DIRTY
648 [ilog2(VM_SOFTDIRTY)] = "sd",
649 #endif
650 [ilog2(VM_MIXEDMAP)] = "mm",
651 [ilog2(VM_HUGEPAGE)] = "hg",
652 [ilog2(VM_NOHUGEPAGE)] = "nh",
653 [ilog2(VM_MERGEABLE)] = "mg",
654 [ilog2(VM_UFFD_MISSING)]= "um",
655 [ilog2(VM_UFFD_WP)] = "uw",
656 };
657 size_t i;
658
659 seq_puts(m, "VmFlags: ");
660 for (i = 0; i < BITS_PER_LONG; i++) {
661 if (vma->vm_flags & (1UL << i)) {
662 seq_printf(m, "%c%c ",
663 mnemonics[i][0], mnemonics[i][1]);
664 }
665 }
666 seq_putc(m, '\n');
667 }
668
669 #ifdef CONFIG_HUGETLB_PAGE
670 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
671 unsigned long addr, unsigned long end,
672 struct mm_walk *walk)
673 {
674 struct mem_size_stats *mss = walk->private;
675 struct vm_area_struct *vma = walk->vma;
676 struct page *page = NULL;
677
678 if (pte_present(*pte)) {
679 page = vm_normal_page(vma, addr, *pte);
680 } else if (is_swap_pte(*pte)) {
681 swp_entry_t swpent = pte_to_swp_entry(*pte);
682
683 if (is_migration_entry(swpent))
684 page = migration_entry_to_page(swpent);
685 }
686 if (page) {
687 int mapcount = page_mapcount(page);
688
689 if (mapcount >= 2)
690 mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
691 else
692 mss->private_hugetlb += huge_page_size(hstate_vma(vma));
693 }
694 return 0;
695 }
696 #endif /* HUGETLB_PAGE */
697
698 static int show_smap(struct seq_file *m, void *v, int is_pid)
699 {
700 struct vm_area_struct *vma = v;
701 struct mem_size_stats mss;
702 struct mm_walk smaps_walk = {
703 .pmd_entry = smaps_pte_range,
704 #ifdef CONFIG_HUGETLB_PAGE
705 .hugetlb_entry = smaps_hugetlb_range,
706 #endif
707 .mm = vma->vm_mm,
708 .private = &mss,
709 };
710
711 memset(&mss, 0, sizeof mss);
712
713 #ifdef CONFIG_SHMEM
714 if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
715 /*
716 * For shared or readonly shmem mappings we know that all
717 * swapped out pages belong to the shmem object, and we can
718 * obtain the swap value much more efficiently. For private
719 * writable mappings, we might have COW pages that are
720 * not affected by the parent swapped out pages of the shmem
721 * object, so we have to distinguish them during the page walk.
722 * Unless we know that the shmem object (or the part mapped by
723 * our VMA) has no swapped out pages at all.
724 */
725 unsigned long shmem_swapped = shmem_swap_usage(vma);
726
727 if (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
728 !(vma->vm_flags & VM_WRITE)) {
729 mss.swap = shmem_swapped;
730 } else {
731 mss.check_shmem_swap = true;
732 smaps_walk.pte_hole = smaps_pte_hole;
733 }
734 }
735 #endif
736
737 /* mmap_sem is held in m_start */
738 walk_page_vma(vma, &smaps_walk);
739
740 show_map_vma(m, vma, is_pid);
741
742 seq_printf(m,
743 "Size: %8lu kB\n"
744 "Rss: %8lu kB\n"
745 "Pss: %8lu kB\n"
746 "Shared_Clean: %8lu kB\n"
747 "Shared_Dirty: %8lu kB\n"
748 "Private_Clean: %8lu kB\n"
749 "Private_Dirty: %8lu kB\n"
750 "Referenced: %8lu kB\n"
751 "Anonymous: %8lu kB\n"
752 "AnonHugePages: %8lu kB\n"
753 "Shared_Hugetlb: %8lu kB\n"
754 "Private_Hugetlb: %7lu kB\n"
755 "Swap: %8lu kB\n"
756 "SwapPss: %8lu kB\n"
757 "KernelPageSize: %8lu kB\n"
758 "MMUPageSize: %8lu kB\n"
759 "Locked: %8lu kB\n",
760 (vma->vm_end - vma->vm_start) >> 10,
761 mss.resident >> 10,
762 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
763 mss.shared_clean >> 10,
764 mss.shared_dirty >> 10,
765 mss.private_clean >> 10,
766 mss.private_dirty >> 10,
767 mss.referenced >> 10,
768 mss.anonymous >> 10,
769 mss.anonymous_thp >> 10,
770 mss.shared_hugetlb >> 10,
771 mss.private_hugetlb >> 10,
772 mss.swap >> 10,
773 (unsigned long)(mss.swap_pss >> (10 + PSS_SHIFT)),
774 vma_kernel_pagesize(vma) >> 10,
775 vma_mmu_pagesize(vma) >> 10,
776 (vma->vm_flags & VM_LOCKED) ?
777 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
778
779 show_smap_vma_flags(m, vma);
780 m_cache_vma(m, vma);
781 return 0;
782 }
783
784 static int show_pid_smap(struct seq_file *m, void *v)
785 {
786 return show_smap(m, v, 1);
787 }
788
789 static int show_tid_smap(struct seq_file *m, void *v)
790 {
791 return show_smap(m, v, 0);
792 }
793
794 static const struct seq_operations proc_pid_smaps_op = {
795 .start = m_start,
796 .next = m_next,
797 .stop = m_stop,
798 .show = show_pid_smap
799 };
800
801 static const struct seq_operations proc_tid_smaps_op = {
802 .start = m_start,
803 .next = m_next,
804 .stop = m_stop,
805 .show = show_tid_smap
806 };
807
808 static int pid_smaps_open(struct inode *inode, struct file *file)
809 {
810 return do_maps_open(inode, file, &proc_pid_smaps_op);
811 }
812
813 static int tid_smaps_open(struct inode *inode, struct file *file)
814 {
815 return do_maps_open(inode, file, &proc_tid_smaps_op);
816 }
817
818 const struct file_operations proc_pid_smaps_operations = {
819 .open = pid_smaps_open,
820 .read = seq_read,
821 .llseek = seq_lseek,
822 .release = proc_map_release,
823 };
824
825 const struct file_operations proc_tid_smaps_operations = {
826 .open = tid_smaps_open,
827 .read = seq_read,
828 .llseek = seq_lseek,
829 .release = proc_map_release,
830 };
831
832 enum clear_refs_types {
833 CLEAR_REFS_ALL = 1,
834 CLEAR_REFS_ANON,
835 CLEAR_REFS_MAPPED,
836 CLEAR_REFS_SOFT_DIRTY,
837 CLEAR_REFS_MM_HIWATER_RSS,
838 CLEAR_REFS_LAST,
839 };
840
841 struct clear_refs_private {
842 enum clear_refs_types type;
843 };
844
845 #ifdef CONFIG_MEM_SOFT_DIRTY
846 static inline void clear_soft_dirty(struct vm_area_struct *vma,
847 unsigned long addr, pte_t *pte)
848 {
849 /*
850 * The soft-dirty tracker uses #PF-s to catch writes
851 * to pages, so write-protect the pte as well. See the
852 * Documentation/vm/soft-dirty.txt for full description
853 * of how soft-dirty works.
854 */
855 pte_t ptent = *pte;
856
857 if (pte_present(ptent)) {
858 ptent = ptep_modify_prot_start(vma->vm_mm, addr, pte);
859 ptent = pte_wrprotect(ptent);
860 ptent = pte_clear_soft_dirty(ptent);
861 ptep_modify_prot_commit(vma->vm_mm, addr, pte, ptent);
862 } else if (is_swap_pte(ptent)) {
863 ptent = pte_swp_clear_soft_dirty(ptent);
864 set_pte_at(vma->vm_mm, addr, pte, ptent);
865 }
866 }
867 #else
868 static inline void clear_soft_dirty(struct vm_area_struct *vma,
869 unsigned long addr, pte_t *pte)
870 {
871 }
872 #endif
873
874 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
875 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
876 unsigned long addr, pmd_t *pmdp)
877 {
878 pmd_t pmd = pmdp_huge_get_and_clear(vma->vm_mm, addr, pmdp);
879
880 pmd = pmd_wrprotect(pmd);
881 pmd = pmd_clear_soft_dirty(pmd);
882
883 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
884 }
885 #else
886 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
887 unsigned long addr, pmd_t *pmdp)
888 {
889 }
890 #endif
891
892 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
893 unsigned long end, struct mm_walk *walk)
894 {
895 struct clear_refs_private *cp = walk->private;
896 struct vm_area_struct *vma = walk->vma;
897 pte_t *pte, ptent;
898 spinlock_t *ptl;
899 struct page *page;
900
901 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
902 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
903 clear_soft_dirty_pmd(vma, addr, pmd);
904 goto out;
905 }
906
907 page = pmd_page(*pmd);
908
909 /* Clear accessed and referenced bits. */
910 pmdp_test_and_clear_young(vma, addr, pmd);
911 test_and_clear_page_young(page);
912 ClearPageReferenced(page);
913 out:
914 spin_unlock(ptl);
915 return 0;
916 }
917
918 if (pmd_trans_unstable(pmd))
919 return 0;
920
921 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
922 for (; addr != end; pte++, addr += PAGE_SIZE) {
923 ptent = *pte;
924
925 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
926 clear_soft_dirty(vma, addr, pte);
927 continue;
928 }
929
930 if (!pte_present(ptent))
931 continue;
932
933 page = vm_normal_page(vma, addr, ptent);
934 if (!page)
935 continue;
936
937 /* Clear accessed and referenced bits. */
938 ptep_test_and_clear_young(vma, addr, pte);
939 test_and_clear_page_young(page);
940 ClearPageReferenced(page);
941 }
942 pte_unmap_unlock(pte - 1, ptl);
943 cond_resched();
944 return 0;
945 }
946
947 static int clear_refs_test_walk(unsigned long start, unsigned long end,
948 struct mm_walk *walk)
949 {
950 struct clear_refs_private *cp = walk->private;
951 struct vm_area_struct *vma = walk->vma;
952
953 if (vma->vm_flags & VM_PFNMAP)
954 return 1;
955
956 /*
957 * Writing 1 to /proc/pid/clear_refs affects all pages.
958 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
959 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
960 * Writing 4 to /proc/pid/clear_refs affects all pages.
961 */
962 if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
963 return 1;
964 if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
965 return 1;
966 return 0;
967 }
968
969 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
970 size_t count, loff_t *ppos)
971 {
972 struct task_struct *task;
973 char buffer[PROC_NUMBUF];
974 struct mm_struct *mm;
975 struct vm_area_struct *vma;
976 enum clear_refs_types type;
977 int itype;
978 int rv;
979
980 memset(buffer, 0, sizeof(buffer));
981 if (count > sizeof(buffer) - 1)
982 count = sizeof(buffer) - 1;
983 if (copy_from_user(buffer, buf, count))
984 return -EFAULT;
985 rv = kstrtoint(strstrip(buffer), 10, &itype);
986 if (rv < 0)
987 return rv;
988 type = (enum clear_refs_types)itype;
989 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
990 return -EINVAL;
991
992 task = get_proc_task(file_inode(file));
993 if (!task)
994 return -ESRCH;
995 mm = get_task_mm(task);
996 if (mm) {
997 struct clear_refs_private cp = {
998 .type = type,
999 };
1000 struct mm_walk clear_refs_walk = {
1001 .pmd_entry = clear_refs_pte_range,
1002 .test_walk = clear_refs_test_walk,
1003 .mm = mm,
1004 .private = &cp,
1005 };
1006
1007 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1008 /*
1009 * Writing 5 to /proc/pid/clear_refs resets the peak
1010 * resident set size to this mm's current rss value.
1011 */
1012 down_write(&mm->mmap_sem);
1013 reset_mm_hiwater_rss(mm);
1014 up_write(&mm->mmap_sem);
1015 goto out_mm;
1016 }
1017
1018 down_read(&mm->mmap_sem);
1019 if (type == CLEAR_REFS_SOFT_DIRTY) {
1020 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1021 if (!(vma->vm_flags & VM_SOFTDIRTY))
1022 continue;
1023 up_read(&mm->mmap_sem);
1024 down_write(&mm->mmap_sem);
1025 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1026 vma->vm_flags &= ~VM_SOFTDIRTY;
1027 vma_set_page_prot(vma);
1028 }
1029 downgrade_write(&mm->mmap_sem);
1030 break;
1031 }
1032 mmu_notifier_invalidate_range_start(mm, 0, -1);
1033 }
1034 walk_page_range(0, ~0UL, &clear_refs_walk);
1035 if (type == CLEAR_REFS_SOFT_DIRTY)
1036 mmu_notifier_invalidate_range_end(mm, 0, -1);
1037 flush_tlb_mm(mm);
1038 up_read(&mm->mmap_sem);
1039 out_mm:
1040 mmput(mm);
1041 }
1042 put_task_struct(task);
1043
1044 return count;
1045 }
1046
1047 const struct file_operations proc_clear_refs_operations = {
1048 .write = clear_refs_write,
1049 .llseek = noop_llseek,
1050 };
1051
1052 typedef struct {
1053 u64 pme;
1054 } pagemap_entry_t;
1055
1056 struct pagemapread {
1057 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */
1058 pagemap_entry_t *buffer;
1059 bool show_pfn;
1060 };
1061
1062 #define PAGEMAP_WALK_SIZE (PMD_SIZE)
1063 #define PAGEMAP_WALK_MASK (PMD_MASK)
1064
1065 #define PM_ENTRY_BYTES sizeof(pagemap_entry_t)
1066 #define PM_PFRAME_BITS 55
1067 #define PM_PFRAME_MASK GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1068 #define PM_SOFT_DIRTY BIT_ULL(55)
1069 #define PM_MMAP_EXCLUSIVE BIT_ULL(56)
1070 #define PM_FILE BIT_ULL(61)
1071 #define PM_SWAP BIT_ULL(62)
1072 #define PM_PRESENT BIT_ULL(63)
1073
1074 #define PM_END_OF_BUFFER 1
1075
1076 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1077 {
1078 return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1079 }
1080
1081 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1082 struct pagemapread *pm)
1083 {
1084 pm->buffer[pm->pos++] = *pme;
1085 if (pm->pos >= pm->len)
1086 return PM_END_OF_BUFFER;
1087 return 0;
1088 }
1089
1090 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1091 struct mm_walk *walk)
1092 {
1093 struct pagemapread *pm = walk->private;
1094 unsigned long addr = start;
1095 int err = 0;
1096
1097 while (addr < end) {
1098 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1099 pagemap_entry_t pme = make_pme(0, 0);
1100 /* End of address space hole, which we mark as non-present. */
1101 unsigned long hole_end;
1102
1103 if (vma)
1104 hole_end = min(end, vma->vm_start);
1105 else
1106 hole_end = end;
1107
1108 for (; addr < hole_end; addr += PAGE_SIZE) {
1109 err = add_to_pagemap(addr, &pme, pm);
1110 if (err)
1111 goto out;
1112 }
1113
1114 if (!vma)
1115 break;
1116
1117 /* Addresses in the VMA. */
1118 if (vma->vm_flags & VM_SOFTDIRTY)
1119 pme = make_pme(0, PM_SOFT_DIRTY);
1120 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1121 err = add_to_pagemap(addr, &pme, pm);
1122 if (err)
1123 goto out;
1124 }
1125 }
1126 out:
1127 return err;
1128 }
1129
1130 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1131 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1132 {
1133 u64 frame = 0, flags = 0;
1134 struct page *page = NULL;
1135
1136 if (pte_present(pte)) {
1137 if (pm->show_pfn)
1138 frame = pte_pfn(pte);
1139 flags |= PM_PRESENT;
1140 page = vm_normal_page(vma, addr, pte);
1141 if (pte_soft_dirty(pte))
1142 flags |= PM_SOFT_DIRTY;
1143 } else if (is_swap_pte(pte)) {
1144 swp_entry_t entry;
1145 if (pte_swp_soft_dirty(pte))
1146 flags |= PM_SOFT_DIRTY;
1147 entry = pte_to_swp_entry(pte);
1148 frame = swp_type(entry) |
1149 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1150 flags |= PM_SWAP;
1151 if (is_migration_entry(entry))
1152 page = migration_entry_to_page(entry);
1153 }
1154
1155 if (page && !PageAnon(page))
1156 flags |= PM_FILE;
1157 if (page && page_mapcount(page) == 1)
1158 flags |= PM_MMAP_EXCLUSIVE;
1159 if (vma->vm_flags & VM_SOFTDIRTY)
1160 flags |= PM_SOFT_DIRTY;
1161
1162 return make_pme(frame, flags);
1163 }
1164
1165 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1166 struct mm_walk *walk)
1167 {
1168 struct vm_area_struct *vma = walk->vma;
1169 struct pagemapread *pm = walk->private;
1170 spinlock_t *ptl;
1171 pte_t *pte, *orig_pte;
1172 int err = 0;
1173
1174 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1175 if (pmd_trans_huge_lock(pmdp, vma, &ptl) == 1) {
1176 u64 flags = 0, frame = 0;
1177 pmd_t pmd = *pmdp;
1178
1179 if ((vma->vm_flags & VM_SOFTDIRTY) || pmd_soft_dirty(pmd))
1180 flags |= PM_SOFT_DIRTY;
1181
1182 /*
1183 * Currently pmd for thp is always present because thp
1184 * can not be swapped-out, migrated, or HWPOISONed
1185 * (split in such cases instead.)
1186 * This if-check is just to prepare for future implementation.
1187 */
1188 if (pmd_present(pmd)) {
1189 struct page *page = pmd_page(pmd);
1190
1191 if (page_mapcount(page) == 1)
1192 flags |= PM_MMAP_EXCLUSIVE;
1193
1194 flags |= PM_PRESENT;
1195 if (pm->show_pfn)
1196 frame = pmd_pfn(pmd) +
1197 ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1198 }
1199
1200 for (; addr != end; addr += PAGE_SIZE) {
1201 pagemap_entry_t pme = make_pme(frame, flags);
1202
1203 err = add_to_pagemap(addr, &pme, pm);
1204 if (err)
1205 break;
1206 if (pm->show_pfn && (flags & PM_PRESENT))
1207 frame++;
1208 }
1209 spin_unlock(ptl);
1210 return err;
1211 }
1212
1213 if (pmd_trans_unstable(pmdp))
1214 return 0;
1215 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1216
1217 /*
1218 * We can assume that @vma always points to a valid one and @end never
1219 * goes beyond vma->vm_end.
1220 */
1221 orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1222 for (; addr < end; pte++, addr += PAGE_SIZE) {
1223 pagemap_entry_t pme;
1224
1225 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1226 err = add_to_pagemap(addr, &pme, pm);
1227 if (err)
1228 break;
1229 }
1230 pte_unmap_unlock(orig_pte, ptl);
1231
1232 cond_resched();
1233
1234 return err;
1235 }
1236
1237 #ifdef CONFIG_HUGETLB_PAGE
1238 /* This function walks within one hugetlb entry in the single call */
1239 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1240 unsigned long addr, unsigned long end,
1241 struct mm_walk *walk)
1242 {
1243 struct pagemapread *pm = walk->private;
1244 struct vm_area_struct *vma = walk->vma;
1245 u64 flags = 0, frame = 0;
1246 int err = 0;
1247 pte_t pte;
1248
1249 if (vma->vm_flags & VM_SOFTDIRTY)
1250 flags |= PM_SOFT_DIRTY;
1251
1252 pte = huge_ptep_get(ptep);
1253 if (pte_present(pte)) {
1254 struct page *page = pte_page(pte);
1255
1256 if (!PageAnon(page))
1257 flags |= PM_FILE;
1258
1259 if (page_mapcount(page) == 1)
1260 flags |= PM_MMAP_EXCLUSIVE;
1261
1262 flags |= PM_PRESENT;
1263 if (pm->show_pfn)
1264 frame = pte_pfn(pte) +
1265 ((addr & ~hmask) >> PAGE_SHIFT);
1266 }
1267
1268 for (; addr != end; addr += PAGE_SIZE) {
1269 pagemap_entry_t pme = make_pme(frame, flags);
1270
1271 err = add_to_pagemap(addr, &pme, pm);
1272 if (err)
1273 return err;
1274 if (pm->show_pfn && (flags & PM_PRESENT))
1275 frame++;
1276 }
1277
1278 cond_resched();
1279
1280 return err;
1281 }
1282 #endif /* HUGETLB_PAGE */
1283
1284 /*
1285 * /proc/pid/pagemap - an array mapping virtual pages to pfns
1286 *
1287 * For each page in the address space, this file contains one 64-bit entry
1288 * consisting of the following:
1289 *
1290 * Bits 0-54 page frame number (PFN) if present
1291 * Bits 0-4 swap type if swapped
1292 * Bits 5-54 swap offset if swapped
1293 * Bit 55 pte is soft-dirty (see Documentation/vm/soft-dirty.txt)
1294 * Bit 56 page exclusively mapped
1295 * Bits 57-60 zero
1296 * Bit 61 page is file-page or shared-anon
1297 * Bit 62 page swapped
1298 * Bit 63 page present
1299 *
1300 * If the page is not present but in swap, then the PFN contains an
1301 * encoding of the swap file number and the page's offset into the
1302 * swap. Unmapped pages return a null PFN. This allows determining
1303 * precisely which pages are mapped (or in swap) and comparing mapped
1304 * pages between processes.
1305 *
1306 * Efficient users of this interface will use /proc/pid/maps to
1307 * determine which areas of memory are actually mapped and llseek to
1308 * skip over unmapped regions.
1309 */
1310 static ssize_t pagemap_read(struct file *file, char __user *buf,
1311 size_t count, loff_t *ppos)
1312 {
1313 struct mm_struct *mm = file->private_data;
1314 struct pagemapread pm;
1315 struct mm_walk pagemap_walk = {};
1316 unsigned long src;
1317 unsigned long svpfn;
1318 unsigned long start_vaddr;
1319 unsigned long end_vaddr;
1320 int ret = 0, copied = 0;
1321
1322 if (!mm || !atomic_inc_not_zero(&mm->mm_users))
1323 goto out;
1324
1325 ret = -EINVAL;
1326 /* file position must be aligned */
1327 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1328 goto out_mm;
1329
1330 ret = 0;
1331 if (!count)
1332 goto out_mm;
1333
1334 /* do not disclose physical addresses: attack vector */
1335 pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1336
1337 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1338 pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY);
1339 ret = -ENOMEM;
1340 if (!pm.buffer)
1341 goto out_mm;
1342
1343 pagemap_walk.pmd_entry = pagemap_pmd_range;
1344 pagemap_walk.pte_hole = pagemap_pte_hole;
1345 #ifdef CONFIG_HUGETLB_PAGE
1346 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1347 #endif
1348 pagemap_walk.mm = mm;
1349 pagemap_walk.private = &pm;
1350
1351 src = *ppos;
1352 svpfn = src / PM_ENTRY_BYTES;
1353 start_vaddr = svpfn << PAGE_SHIFT;
1354 end_vaddr = mm->task_size;
1355
1356 /* watch out for wraparound */
1357 if (svpfn > mm->task_size >> PAGE_SHIFT)
1358 start_vaddr = end_vaddr;
1359
1360 /*
1361 * The odds are that this will stop walking way
1362 * before end_vaddr, because the length of the
1363 * user buffer is tracked in "pm", and the walk
1364 * will stop when we hit the end of the buffer.
1365 */
1366 ret = 0;
1367 while (count && (start_vaddr < end_vaddr)) {
1368 int len;
1369 unsigned long end;
1370
1371 pm.pos = 0;
1372 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1373 /* overflow ? */
1374 if (end < start_vaddr || end > end_vaddr)
1375 end = end_vaddr;
1376 down_read(&mm->mmap_sem);
1377 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1378 up_read(&mm->mmap_sem);
1379 start_vaddr = end;
1380
1381 len = min(count, PM_ENTRY_BYTES * pm.pos);
1382 if (copy_to_user(buf, pm.buffer, len)) {
1383 ret = -EFAULT;
1384 goto out_free;
1385 }
1386 copied += len;
1387 buf += len;
1388 count -= len;
1389 }
1390 *ppos += copied;
1391 if (!ret || ret == PM_END_OF_BUFFER)
1392 ret = copied;
1393
1394 out_free:
1395 kfree(pm.buffer);
1396 out_mm:
1397 mmput(mm);
1398 out:
1399 return ret;
1400 }
1401
1402 static int pagemap_open(struct inode *inode, struct file *file)
1403 {
1404 struct mm_struct *mm;
1405
1406 mm = proc_mem_open(inode, PTRACE_MODE_READ);
1407 if (IS_ERR(mm))
1408 return PTR_ERR(mm);
1409 file->private_data = mm;
1410 return 0;
1411 }
1412
1413 static int pagemap_release(struct inode *inode, struct file *file)
1414 {
1415 struct mm_struct *mm = file->private_data;
1416
1417 if (mm)
1418 mmdrop(mm);
1419 return 0;
1420 }
1421
1422 const struct file_operations proc_pagemap_operations = {
1423 .llseek = mem_lseek, /* borrow this */
1424 .read = pagemap_read,
1425 .open = pagemap_open,
1426 .release = pagemap_release,
1427 };
1428 #endif /* CONFIG_PROC_PAGE_MONITOR */
1429
1430 #ifdef CONFIG_NUMA
1431
1432 struct numa_maps {
1433 unsigned long pages;
1434 unsigned long anon;
1435 unsigned long active;
1436 unsigned long writeback;
1437 unsigned long mapcount_max;
1438 unsigned long dirty;
1439 unsigned long swapcache;
1440 unsigned long node[MAX_NUMNODES];
1441 };
1442
1443 struct numa_maps_private {
1444 struct proc_maps_private proc_maps;
1445 struct numa_maps md;
1446 };
1447
1448 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1449 unsigned long nr_pages)
1450 {
1451 int count = page_mapcount(page);
1452
1453 md->pages += nr_pages;
1454 if (pte_dirty || PageDirty(page))
1455 md->dirty += nr_pages;
1456
1457 if (PageSwapCache(page))
1458 md->swapcache += nr_pages;
1459
1460 if (PageActive(page) || PageUnevictable(page))
1461 md->active += nr_pages;
1462
1463 if (PageWriteback(page))
1464 md->writeback += nr_pages;
1465
1466 if (PageAnon(page))
1467 md->anon += nr_pages;
1468
1469 if (count > md->mapcount_max)
1470 md->mapcount_max = count;
1471
1472 md->node[page_to_nid(page)] += nr_pages;
1473 }
1474
1475 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1476 unsigned long addr)
1477 {
1478 struct page *page;
1479 int nid;
1480
1481 if (!pte_present(pte))
1482 return NULL;
1483
1484 page = vm_normal_page(vma, addr, pte);
1485 if (!page)
1486 return NULL;
1487
1488 if (PageReserved(page))
1489 return NULL;
1490
1491 nid = page_to_nid(page);
1492 if (!node_isset(nid, node_states[N_MEMORY]))
1493 return NULL;
1494
1495 return page;
1496 }
1497
1498 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1499 unsigned long end, struct mm_walk *walk)
1500 {
1501 struct numa_maps *md = walk->private;
1502 struct vm_area_struct *vma = walk->vma;
1503 spinlock_t *ptl;
1504 pte_t *orig_pte;
1505 pte_t *pte;
1506
1507 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
1508 pte_t huge_pte = *(pte_t *)pmd;
1509 struct page *page;
1510
1511 page = can_gather_numa_stats(huge_pte, vma, addr);
1512 if (page)
1513 gather_stats(page, md, pte_dirty(huge_pte),
1514 HPAGE_PMD_SIZE/PAGE_SIZE);
1515 spin_unlock(ptl);
1516 return 0;
1517 }
1518
1519 if (pmd_trans_unstable(pmd))
1520 return 0;
1521 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1522 do {
1523 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1524 if (!page)
1525 continue;
1526 gather_stats(page, md, pte_dirty(*pte), 1);
1527
1528 } while (pte++, addr += PAGE_SIZE, addr != end);
1529 pte_unmap_unlock(orig_pte, ptl);
1530 return 0;
1531 }
1532 #ifdef CONFIG_HUGETLB_PAGE
1533 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1534 unsigned long addr, unsigned long end, struct mm_walk *walk)
1535 {
1536 struct numa_maps *md;
1537 struct page *page;
1538
1539 if (!pte_present(*pte))
1540 return 0;
1541
1542 page = pte_page(*pte);
1543 if (!page)
1544 return 0;
1545
1546 md = walk->private;
1547 gather_stats(page, md, pte_dirty(*pte), 1);
1548 return 0;
1549 }
1550
1551 #else
1552 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1553 unsigned long addr, unsigned long end, struct mm_walk *walk)
1554 {
1555 return 0;
1556 }
1557 #endif
1558
1559 /*
1560 * Display pages allocated per node and memory policy via /proc.
1561 */
1562 static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1563 {
1564 struct numa_maps_private *numa_priv = m->private;
1565 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1566 struct vm_area_struct *vma = v;
1567 struct numa_maps *md = &numa_priv->md;
1568 struct file *file = vma->vm_file;
1569 struct mm_struct *mm = vma->vm_mm;
1570 struct mm_walk walk = {
1571 .hugetlb_entry = gather_hugetlb_stats,
1572 .pmd_entry = gather_pte_stats,
1573 .private = md,
1574 .mm = mm,
1575 };
1576 struct mempolicy *pol;
1577 char buffer[64];
1578 int nid;
1579
1580 if (!mm)
1581 return 0;
1582
1583 /* Ensure we start with an empty set of numa_maps statistics. */
1584 memset(md, 0, sizeof(*md));
1585
1586 pol = __get_vma_policy(vma, vma->vm_start);
1587 if (pol) {
1588 mpol_to_str(buffer, sizeof(buffer), pol);
1589 mpol_cond_put(pol);
1590 } else {
1591 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1592 }
1593
1594 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1595
1596 if (file) {
1597 seq_puts(m, " file=");
1598 seq_file_path(m, file, "\n\t= ");
1599 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1600 seq_puts(m, " heap");
1601 } else {
1602 pid_t tid = pid_of_stack(proc_priv, vma, is_pid);
1603 if (tid != 0) {
1604 /*
1605 * Thread stack in /proc/PID/task/TID/maps or
1606 * the main process stack.
1607 */
1608 if (!is_pid || (vma->vm_start <= mm->start_stack &&
1609 vma->vm_end >= mm->start_stack))
1610 seq_puts(m, " stack");
1611 else
1612 seq_printf(m, " stack:%d", tid);
1613 }
1614 }
1615
1616 if (is_vm_hugetlb_page(vma))
1617 seq_puts(m, " huge");
1618
1619 /* mmap_sem is held by m_start */
1620 walk_page_vma(vma, &walk);
1621
1622 if (!md->pages)
1623 goto out;
1624
1625 if (md->anon)
1626 seq_printf(m, " anon=%lu", md->anon);
1627
1628 if (md->dirty)
1629 seq_printf(m, " dirty=%lu", md->dirty);
1630
1631 if (md->pages != md->anon && md->pages != md->dirty)
1632 seq_printf(m, " mapped=%lu", md->pages);
1633
1634 if (md->mapcount_max > 1)
1635 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1636
1637 if (md->swapcache)
1638 seq_printf(m, " swapcache=%lu", md->swapcache);
1639
1640 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1641 seq_printf(m, " active=%lu", md->active);
1642
1643 if (md->writeback)
1644 seq_printf(m, " writeback=%lu", md->writeback);
1645
1646 for_each_node_state(nid, N_MEMORY)
1647 if (md->node[nid])
1648 seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1649
1650 seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1651 out:
1652 seq_putc(m, '\n');
1653 m_cache_vma(m, vma);
1654 return 0;
1655 }
1656
1657 static int show_pid_numa_map(struct seq_file *m, void *v)
1658 {
1659 return show_numa_map(m, v, 1);
1660 }
1661
1662 static int show_tid_numa_map(struct seq_file *m, void *v)
1663 {
1664 return show_numa_map(m, v, 0);
1665 }
1666
1667 static const struct seq_operations proc_pid_numa_maps_op = {
1668 .start = m_start,
1669 .next = m_next,
1670 .stop = m_stop,
1671 .show = show_pid_numa_map,
1672 };
1673
1674 static const struct seq_operations proc_tid_numa_maps_op = {
1675 .start = m_start,
1676 .next = m_next,
1677 .stop = m_stop,
1678 .show = show_tid_numa_map,
1679 };
1680
1681 static int numa_maps_open(struct inode *inode, struct file *file,
1682 const struct seq_operations *ops)
1683 {
1684 return proc_maps_open(inode, file, ops,
1685 sizeof(struct numa_maps_private));
1686 }
1687
1688 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1689 {
1690 return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1691 }
1692
1693 static int tid_numa_maps_open(struct inode *inode, struct file *file)
1694 {
1695 return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1696 }
1697
1698 const struct file_operations proc_pid_numa_maps_operations = {
1699 .open = pid_numa_maps_open,
1700 .read = seq_read,
1701 .llseek = seq_lseek,
1702 .release = proc_map_release,
1703 };
1704
1705 const struct file_operations proc_tid_numa_maps_operations = {
1706 .open = tid_numa_maps_open,
1707 .read = seq_read,
1708 .llseek = seq_lseek,
1709 .release = proc_map_release,
1710 };
1711 #endif /* CONFIG_NUMA */
Linux kernel - Deliverables
This page took 0.078381 seconds and 4 git commands to generate.