projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
mm, shmem: add internal shmem resident memory accounting
[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 data, text, lib, swap, ptes, pmds;
27 unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
28
29 /*
30 * Note: to minimize their overhead, mm maintains hiwater_vm and
31 * hiwater_rss only when about to *lower* total_vm or rss. Any
32 * collector of these hiwater stats must therefore get total_vm
33 * and rss too, which will usually be the higher. Barriers? not
34 * worth the effort, such snapshots can always be inconsistent.
35 */
36 hiwater_vm = total_vm = mm->total_vm;
37 if (hiwater_vm < mm->hiwater_vm)
38 hiwater_vm = mm->hiwater_vm;
39 hiwater_rss = total_rss = get_mm_rss(mm);
40 if (hiwater_rss < mm->hiwater_rss)
41 hiwater_rss = mm->hiwater_rss;
42
43 data = mm->total_vm - mm->shared_vm - mm->stack_vm;
44 text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
45 lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
46 swap = get_mm_counter(mm, MM_SWAPENTS);
47 ptes = PTRS_PER_PTE * sizeof(pte_t) * atomic_long_read(&mm->nr_ptes);
48 pmds = PTRS_PER_PMD * sizeof(pmd_t) * mm_nr_pmds(mm);
49 seq_printf(m,
50 "VmPeak:\t%8lu kB\n"
51 "VmSize:\t%8lu kB\n"
52 "VmLck:\t%8lu kB\n"
53 "VmPin:\t%8lu kB\n"
54 "VmHWM:\t%8lu kB\n"
55 "VmRSS:\t%8lu kB\n"
56 "VmData:\t%8lu kB\n"
57 "VmStk:\t%8lu kB\n"
58 "VmExe:\t%8lu kB\n"
59 "VmLib:\t%8lu kB\n"
60 "VmPTE:\t%8lu kB\n"
61 "VmPMD:\t%8lu kB\n"
62 "VmSwap:\t%8lu kB\n",
63 hiwater_vm << (PAGE_SHIFT-10),
64 total_vm << (PAGE_SHIFT-10),
65 mm->locked_vm << (PAGE_SHIFT-10),
66 mm->pinned_vm << (PAGE_SHIFT-10),
67 hiwater_rss << (PAGE_SHIFT-10),
68 total_rss << (PAGE_SHIFT-10),
69 data << (PAGE_SHIFT-10),
70 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
71 ptes >> 10,
72 pmds >> 10,
73 swap << (PAGE_SHIFT-10));
74 hugetlb_report_usage(m, mm);
75 }
76
77 unsigned long task_vsize(struct mm_struct *mm)
78 {
79 return PAGE_SIZE * mm->total_vm;
80 }
81
82 unsigned long task_statm(struct mm_struct *mm,
83 unsigned long *shared, unsigned long *text,
84 unsigned long *data, unsigned long *resident)
85 {
86 *shared = get_mm_counter(mm, MM_FILEPAGES) +
87 get_mm_counter(mm, MM_SHMEMPAGES);
88 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
89 >> PAGE_SHIFT;
90 *data = mm->total_vm - mm->shared_vm;
91 *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
92 return mm->total_vm;
93 }
94
95 #ifdef CONFIG_NUMA
96 /*
97 * Save get_task_policy() for show_numa_map().
98 */
99 static void hold_task_mempolicy(struct proc_maps_private *priv)
100 {
101 struct task_struct *task = priv->task;
102
103 task_lock(task);
104 priv->task_mempolicy = get_task_policy(task);
105 mpol_get(priv->task_mempolicy);
106 task_unlock(task);
107 }
108 static void release_task_mempolicy(struct proc_maps_private *priv)
109 {
110 mpol_put(priv->task_mempolicy);
111 }
112 #else
113 static void hold_task_mempolicy(struct proc_maps_private *priv)
114 {
115 }
116 static void release_task_mempolicy(struct proc_maps_private *priv)
117 {
118 }
119 #endif
120
121 static void vma_stop(struct proc_maps_private *priv)
122 {
123 struct mm_struct *mm = priv->mm;
124
125 release_task_mempolicy(priv);
126 up_read(&mm->mmap_sem);
127 mmput(mm);
128 }
129
130 static struct vm_area_struct *
131 m_next_vma(struct proc_maps_private *priv, struct vm_area_struct *vma)
132 {
133 if (vma == priv->tail_vma)
134 return NULL;
135 return vma->vm_next ?: priv->tail_vma;
136 }
137
138 static void m_cache_vma(struct seq_file *m, struct vm_area_struct *vma)
139 {
140 if (m->count < m->size) /* vma is copied successfully */
141 m->version = m_next_vma(m->private, vma) ? vma->vm_start : -1UL;
142 }
143
144 static void *m_start(struct seq_file *m, loff_t *ppos)
145 {
146 struct proc_maps_private *priv = m->private;
147 unsigned long last_addr = m->version;
148 struct mm_struct *mm;
149 struct vm_area_struct *vma;
150 unsigned int pos = *ppos;
151
152 /* See m_cache_vma(). Zero at the start or after lseek. */
153 if (last_addr == -1UL)
154 return NULL;
155
156 priv->task = get_proc_task(priv->inode);
157 if (!priv->task)
158 return ERR_PTR(-ESRCH);
159
160 mm = priv->mm;
161 if (!mm || !atomic_inc_not_zero(&mm->mm_users))
162 return NULL;
163
164 down_read(&mm->mmap_sem);
165 hold_task_mempolicy(priv);
166 priv->tail_vma = get_gate_vma(mm);
167
168 if (last_addr) {
169 vma = find_vma(mm, last_addr);
170 if (vma && (vma = m_next_vma(priv, vma)))
171 return vma;
172 }
173
174 m->version = 0;
175 if (pos < mm->map_count) {
176 for (vma = mm->mmap; pos; pos--) {
177 m->version = vma->vm_start;
178 vma = vma->vm_next;
179 }
180 return vma;
181 }
182
183 /* we do not bother to update m->version in this case */
184 if (pos == mm->map_count && priv->tail_vma)
185 return priv->tail_vma;
186
187 vma_stop(priv);
188 return NULL;
189 }
190
191 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
192 {
193 struct proc_maps_private *priv = m->private;
194 struct vm_area_struct *next;
195
196 (*pos)++;
197 next = m_next_vma(priv, v);
198 if (!next)
199 vma_stop(priv);
200 return next;
201 }
202
203 static void m_stop(struct seq_file *m, void *v)
204 {
205 struct proc_maps_private *priv = m->private;
206
207 if (!IS_ERR_OR_NULL(v))
208 vma_stop(priv);
209 if (priv->task) {
210 put_task_struct(priv->task);
211 priv->task = NULL;
212 }
213 }
214
215 static int proc_maps_open(struct inode *inode, struct file *file,
216 const struct seq_operations *ops, int psize)
217 {
218 struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
219
220 if (!priv)
221 return -ENOMEM;
222
223 priv->inode = inode;
224 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
225 if (IS_ERR(priv->mm)) {
226 int err = PTR_ERR(priv->mm);
227
228 seq_release_private(inode, file);
229 return err;
230 }
231
232 return 0;
233 }
234
235 static int proc_map_release(struct inode *inode, struct file *file)
236 {
237 struct seq_file *seq = file->private_data;
238 struct proc_maps_private *priv = seq->private;
239
240 if (priv->mm)
241 mmdrop(priv->mm);
242
243 return seq_release_private(inode, file);
244 }
245
246 static int do_maps_open(struct inode *inode, struct file *file,
247 const struct seq_operations *ops)
248 {
249 return proc_maps_open(inode, file, ops,
250 sizeof(struct proc_maps_private));
251 }
252
253 static pid_t pid_of_stack(struct proc_maps_private *priv,
254 struct vm_area_struct *vma, bool is_pid)
255 {
256 struct inode *inode = priv->inode;
257 struct task_struct *task;
258 pid_t ret = 0;
259
260 rcu_read_lock();
261 task = pid_task(proc_pid(inode), PIDTYPE_PID);
262 if (task) {
263 task = task_of_stack(task, vma, is_pid);
264 if (task)
265 ret = task_pid_nr_ns(task, inode->i_sb->s_fs_info);
266 }
267 rcu_read_unlock();
268
269 return ret;
270 }
271
272 static void
273 show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
274 {
275 struct mm_struct *mm = vma->vm_mm;
276 struct file *file = vma->vm_file;
277 struct proc_maps_private *priv = m->private;
278 vm_flags_t flags = vma->vm_flags;
279 unsigned long ino = 0;
280 unsigned long long pgoff = 0;
281 unsigned long start, end;
282 dev_t dev = 0;
283 const char *name = NULL;
284
285 if (file) {
286 struct inode *inode = file_inode(vma->vm_file);
287 dev = inode->i_sb->s_dev;
288 ino = inode->i_ino;
289 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
290 }
291
292 /* We don't show the stack guard page in /proc/maps */
293 start = vma->vm_start;
294 if (stack_guard_page_start(vma, start))
295 start += PAGE_SIZE;
296 end = vma->vm_end;
297 if (stack_guard_page_end(vma, end))
298 end -= PAGE_SIZE;
299
300 seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
301 seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu ",
302 start,
303 end,
304 flags & VM_READ ? 'r' : '-',
305 flags & VM_WRITE ? 'w' : '-',
306 flags & VM_EXEC ? 'x' : '-',
307 flags & VM_MAYSHARE ? 's' : 'p',
308 pgoff,
309 MAJOR(dev), MINOR(dev), ino);
310
311 /*
312 * Print the dentry name for named mappings, and a
313 * special [heap] marker for the heap:
314 */
315 if (file) {
316 seq_pad(m, ' ');
317 seq_file_path(m, file, "\n");
318 goto done;
319 }
320
321 if (vma->vm_ops && vma->vm_ops->name) {
322 name = vma->vm_ops->name(vma);
323 if (name)
324 goto done;
325 }
326
327 name = arch_vma_name(vma);
328 if (!name) {
329 pid_t tid;
330
331 if (!mm) {
332 name = "[vdso]";
333 goto done;
334 }
335
336 if (vma->vm_start <= mm->brk &&
337 vma->vm_end >= mm->start_brk) {
338 name = "[heap]";
339 goto done;
340 }
341
342 tid = pid_of_stack(priv, vma, is_pid);
343 if (tid != 0) {
344 /*
345 * Thread stack in /proc/PID/task/TID/maps or
346 * the main process stack.
347 */
348 if (!is_pid || (vma->vm_start <= mm->start_stack &&
349 vma->vm_end >= mm->start_stack)) {
350 name = "[stack]";
351 } else {
352 /* Thread stack in /proc/PID/maps */
353 seq_pad(m, ' ');
354 seq_printf(m, "[stack:%d]", tid);
355 }
356 }
357 }
358
359 done:
360 if (name) {
361 seq_pad(m, ' ');
362 seq_puts(m, name);
363 }
364 seq_putc(m, '\n');
365 }
366
367 static 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
374 static int show_pid_map(struct seq_file *m, void *v)
375 {
376 return show_map(m, v, 1);
377 }
378
379 static int show_tid_map(struct seq_file *m, void *v)
380 {
381 return show_map(m, v, 0);
382 }
383
384 static 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
391 static 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
398 static int pid_maps_open(struct inode *inode, struct file *file)
399 {
400 return do_maps_open(inode, file, &proc_pid_maps_op);
401 }
402
403 static int tid_maps_open(struct inode *inode, struct file *file)
404 {
405 return do_maps_open(inode, file, &proc_tid_maps_op);
406 }
407
408 const 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
415 const 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
442 struct 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
459 static void smaps_account(struct mem_size_stats *mss, struct page *page,
460 unsigned long size, bool young, bool dirty)
461 {
462 int mapcount;
463
464 if (PageAnon(page))
465 mss->anonymous += size;
466
467 mss->resident += size;
468 /* Accumulate the size in pages that have been accessed. */
469 if (young || page_is_young(page) || PageReferenced(page))
470 mss->referenced += size;
471 mapcount = page_mapcount(page);
472 if (mapcount >= 2) {
473 u64 pss_delta;
474
475 if (dirty || PageDirty(page))
476 mss->shared_dirty += size;
477 else
478 mss->shared_clean += size;
479 pss_delta = (u64)size << PSS_SHIFT;
480 do_div(pss_delta, mapcount);
481 mss->pss += pss_delta;
482 } else {
483 if (dirty || PageDirty(page))
484 mss->private_dirty += size;
485 else
486 mss->private_clean += size;
487 mss->pss += (u64)size << PSS_SHIFT;
488 }
489 }
490
491 #ifdef CONFIG_SHMEM
492 static int smaps_pte_hole(unsigned long addr, unsigned long end,
493 struct mm_walk *walk)
494 {
495 struct mem_size_stats *mss = walk->private;
496
497 mss->swap += shmem_partial_swap_usage(
498 walk->vma->vm_file->f_mapping, addr, end);
499
500 return 0;
501 }
502 #endif
503
504 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
505 struct mm_walk *walk)
506 {
507 struct mem_size_stats *mss = walk->private;
508 struct vm_area_struct *vma = walk->vma;
509 struct page *page = NULL;
510
511 if (pte_present(*pte)) {
512 page = vm_normal_page(vma, addr, *pte);
513 } else if (is_swap_pte(*pte)) {
514 swp_entry_t swpent = pte_to_swp_entry(*pte);
515
516 if (!non_swap_entry(swpent)) {
517 int mapcount;
518
519 mss->swap += PAGE_SIZE;
520 mapcount = swp_swapcount(swpent);
521 if (mapcount >= 2) {
522 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
523
524 do_div(pss_delta, mapcount);
525 mss->swap_pss += pss_delta;
526 } else {
527 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
528 }
529 } else if (is_migration_entry(swpent))
530 page = migration_entry_to_page(swpent);
531 } else if (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap
532 && pte_none(*pte))) {
533 page = find_get_entry(vma->vm_file->f_mapping,
534 linear_page_index(vma, addr));
535 if (!page)
536 return;
537
538 if (radix_tree_exceptional_entry(page))
539 mss->swap += PAGE_SIZE;
540 else
541 page_cache_release(page);
542
543 return;
544 }
545
546 if (!page)
547 return;
548 smaps_account(mss, page, PAGE_SIZE, pte_young(*pte), pte_dirty(*pte));
549 }
550
551 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
552 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
553 struct mm_walk *walk)
554 {
555 struct mem_size_stats *mss = walk->private;
556 struct vm_area_struct *vma = walk->vma;
557 struct page *page;
558
559 /* FOLL_DUMP will return -EFAULT on huge zero page */
560 page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
561 if (IS_ERR_OR_NULL(page))
562 return;
563 mss->anonymous_thp += HPAGE_PMD_SIZE;
564 smaps_account(mss, page, HPAGE_PMD_SIZE,
565 pmd_young(*pmd), pmd_dirty(*pmd));
566 }
567 #else
568 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
569 struct mm_walk *walk)
570 {
571 }
572 #endif
573
574 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
575 struct mm_walk *walk)
576 {
577 struct vm_area_struct *vma = walk->vma;
578 pte_t *pte;
579 spinlock_t *ptl;
580
581 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
582 smaps_pmd_entry(pmd, addr, walk);
583 spin_unlock(ptl);
584 return 0;
585 }
586
587 if (pmd_trans_unstable(pmd))
588 return 0;
589 /*
590 * The mmap_sem held all the way back in m_start() is what
591 * keeps khugepaged out of here and from collapsing things
592 * in here.
593 */
594 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
595 for (; addr != end; pte++, addr += PAGE_SIZE)
596 smaps_pte_entry(pte, addr, walk);
597 pte_unmap_unlock(pte - 1, ptl);
598 cond_resched();
599 return 0;
600 }
601
602 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
603 {
604 /*
605 * Don't forget to update Documentation/ on changes.
606 */
607 static const char mnemonics[BITS_PER_LONG][2] = {
608 /*
609 * In case if we meet a flag we don't know about.
610 */
611 [0 ... (BITS_PER_LONG-1)] = "??",
612
613 [ilog2(VM_READ)] = "rd",
614 [ilog2(VM_WRITE)] = "wr",
615 [ilog2(VM_EXEC)] = "ex",
616 [ilog2(VM_SHARED)] = "sh",
617 [ilog2(VM_MAYREAD)] = "mr",
618 [ilog2(VM_MAYWRITE)] = "mw",
619 [ilog2(VM_MAYEXEC)] = "me",
620 [ilog2(VM_MAYSHARE)] = "ms",
621 [ilog2(VM_GROWSDOWN)] = "gd",
622 [ilog2(VM_PFNMAP)] = "pf",
623 [ilog2(VM_DENYWRITE)] = "dw",
624 #ifdef CONFIG_X86_INTEL_MPX
625 [ilog2(VM_MPX)] = "mp",
626 #endif
627 [ilog2(VM_LOCKED)] = "lo",
628 [ilog2(VM_IO)] = "io",
629 [ilog2(VM_SEQ_READ)] = "sr",
630 [ilog2(VM_RAND_READ)] = "rr",
631 [ilog2(VM_DONTCOPY)] = "dc",
632 [ilog2(VM_DONTEXPAND)] = "de",
633 [ilog2(VM_ACCOUNT)] = "ac",
634 [ilog2(VM_NORESERVE)] = "nr",
635 [ilog2(VM_HUGETLB)] = "ht",
636 [ilog2(VM_ARCH_1)] = "ar",
637 [ilog2(VM_DONTDUMP)] = "dd",
638 #ifdef CONFIG_MEM_SOFT_DIRTY
639 [ilog2(VM_SOFTDIRTY)] = "sd",
640 #endif
641 [ilog2(VM_MIXEDMAP)] = "mm",
642 [ilog2(VM_HUGEPAGE)] = "hg",
643 [ilog2(VM_NOHUGEPAGE)] = "nh",
644 [ilog2(VM_MERGEABLE)] = "mg",
645 [ilog2(VM_UFFD_MISSING)]= "um",
646 [ilog2(VM_UFFD_WP)] = "uw",
647 };
648 size_t i;
649
650 seq_puts(m, "VmFlags: ");
651 for (i = 0; i < BITS_PER_LONG; i++) {
652 if (vma->vm_flags & (1UL << i)) {
653 seq_printf(m, "%c%c ",
654 mnemonics[i][0], mnemonics[i][1]);
655 }
656 }
657 seq_putc(m, '\n');
658 }
659
660 #ifdef CONFIG_HUGETLB_PAGE
661 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
662 unsigned long addr, unsigned long end,
663 struct mm_walk *walk)
664 {
665 struct mem_size_stats *mss = walk->private;
666 struct vm_area_struct *vma = walk->vma;
667 struct page *page = NULL;
668
669 if (pte_present(*pte)) {
670 page = vm_normal_page(vma, addr, *pte);
671 } else if (is_swap_pte(*pte)) {
672 swp_entry_t swpent = pte_to_swp_entry(*pte);
673
674 if (is_migration_entry(swpent))
675 page = migration_entry_to_page(swpent);
676 }
677 if (page) {
678 int mapcount = page_mapcount(page);
679
680 if (mapcount >= 2)
681 mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
682 else
683 mss->private_hugetlb += huge_page_size(hstate_vma(vma));
684 }
685 return 0;
686 }
687 #endif /* HUGETLB_PAGE */
688
689 static int show_smap(struct seq_file *m, void *v, int is_pid)
690 {
691 struct vm_area_struct *vma = v;
692 struct mem_size_stats mss;
693 struct mm_walk smaps_walk = {
694 .pmd_entry = smaps_pte_range,
695 #ifdef CONFIG_HUGETLB_PAGE
696 .hugetlb_entry = smaps_hugetlb_range,
697 #endif
698 .mm = vma->vm_mm,
699 .private = &mss,
700 };
701
702 memset(&mss, 0, sizeof mss);
703
704 #ifdef CONFIG_SHMEM
705 if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
706 /*
707 * For shared or readonly shmem mappings we know that all
708 * swapped out pages belong to the shmem object, and we can
709 * obtain the swap value much more efficiently. For private
710 * writable mappings, we might have COW pages that are
711 * not affected by the parent swapped out pages of the shmem
712 * object, so we have to distinguish them during the page walk.
713 * Unless we know that the shmem object (or the part mapped by
714 * our VMA) has no swapped out pages at all.
715 */
716 unsigned long shmem_swapped = shmem_swap_usage(vma);
717
718 if (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
719 !(vma->vm_flags & VM_WRITE)) {
720 mss.swap = shmem_swapped;
721 } else {
722 mss.check_shmem_swap = true;
723 smaps_walk.pte_hole = smaps_pte_hole;
724 }
725 }
726 #endif
727
728 /* mmap_sem is held in m_start */
729 walk_page_vma(vma, &smaps_walk);
730
731 show_map_vma(m, vma, is_pid);
732
733 seq_printf(m,
734 "Size: %8lu kB\n"
735 "Rss: %8lu kB\n"
736 "Pss: %8lu kB\n"
737 "Shared_Clean: %8lu kB\n"
738 "Shared_Dirty: %8lu kB\n"
739 "Private_Clean: %8lu kB\n"
740 "Private_Dirty: %8lu kB\n"
741 "Referenced: %8lu kB\n"
742 "Anonymous: %8lu kB\n"
743 "AnonHugePages: %8lu kB\n"
744 "Shared_Hugetlb: %8lu kB\n"
745 "Private_Hugetlb: %7lu kB\n"
746 "Swap: %8lu kB\n"
747 "SwapPss: %8lu kB\n"
748 "KernelPageSize: %8lu kB\n"
749 "MMUPageSize: %8lu kB\n"
750 "Locked: %8lu kB\n",
751 (vma->vm_end - vma->vm_start) >> 10,
752 mss.resident >> 10,
753 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
754 mss.shared_clean >> 10,
755 mss.shared_dirty >> 10,
756 mss.private_clean >> 10,
757 mss.private_dirty >> 10,
758 mss.referenced >> 10,
759 mss.anonymous >> 10,
760 mss.anonymous_thp >> 10,
761 mss.shared_hugetlb >> 10,
762 mss.private_hugetlb >> 10,
763 mss.swap >> 10,
764 (unsigned long)(mss.swap_pss >> (10 + PSS_SHIFT)),
765 vma_kernel_pagesize(vma) >> 10,
766 vma_mmu_pagesize(vma) >> 10,
767 (vma->vm_flags & VM_LOCKED) ?
768 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
769
770 show_smap_vma_flags(m, vma);
771 m_cache_vma(m, vma);
772 return 0;
773 }
774
775 static int show_pid_smap(struct seq_file *m, void *v)
776 {
777 return show_smap(m, v, 1);
778 }
779
780 static int show_tid_smap(struct seq_file *m, void *v)
781 {
782 return show_smap(m, v, 0);
783 }
784
785 static const struct seq_operations proc_pid_smaps_op = {
786 .start = m_start,
787 .next = m_next,
788 .stop = m_stop,
789 .show = show_pid_smap
790 };
791
792 static const struct seq_operations proc_tid_smaps_op = {
793 .start = m_start,
794 .next = m_next,
795 .stop = m_stop,
796 .show = show_tid_smap
797 };
798
799 static int pid_smaps_open(struct inode *inode, struct file *file)
800 {
801 return do_maps_open(inode, file, &proc_pid_smaps_op);
802 }
803
804 static int tid_smaps_open(struct inode *inode, struct file *file)
805 {
806 return do_maps_open(inode, file, &proc_tid_smaps_op);
807 }
808
809 const struct file_operations proc_pid_smaps_operations = {
810 .open = pid_smaps_open,
811 .read = seq_read,
812 .llseek = seq_lseek,
813 .release = proc_map_release,
814 };
815
816 const struct file_operations proc_tid_smaps_operations = {
817 .open = tid_smaps_open,
818 .read = seq_read,
819 .llseek = seq_lseek,
820 .release = proc_map_release,
821 };
822
823 enum clear_refs_types {
824 CLEAR_REFS_ALL = 1,
825 CLEAR_REFS_ANON,
826 CLEAR_REFS_MAPPED,
827 CLEAR_REFS_SOFT_DIRTY,
828 CLEAR_REFS_MM_HIWATER_RSS,
829 CLEAR_REFS_LAST,
830 };
831
832 struct clear_refs_private {
833 enum clear_refs_types type;
834 };
835
836 #ifdef CONFIG_MEM_SOFT_DIRTY
837 static inline void clear_soft_dirty(struct vm_area_struct *vma,
838 unsigned long addr, pte_t *pte)
839 {
840 /*
841 * The soft-dirty tracker uses #PF-s to catch writes
842 * to pages, so write-protect the pte as well. See the
843 * Documentation/vm/soft-dirty.txt for full description
844 * of how soft-dirty works.
845 */
846 pte_t ptent = *pte;
847
848 if (pte_present(ptent)) {
849 ptent = ptep_modify_prot_start(vma->vm_mm, addr, pte);
850 ptent = pte_wrprotect(ptent);
851 ptent = pte_clear_soft_dirty(ptent);
852 ptep_modify_prot_commit(vma->vm_mm, addr, pte, ptent);
853 } else if (is_swap_pte(ptent)) {
854 ptent = pte_swp_clear_soft_dirty(ptent);
855 set_pte_at(vma->vm_mm, addr, pte, ptent);
856 }
857 }
858 #else
859 static inline void clear_soft_dirty(struct vm_area_struct *vma,
860 unsigned long addr, pte_t *pte)
861 {
862 }
863 #endif
864
865 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
866 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
867 unsigned long addr, pmd_t *pmdp)
868 {
869 pmd_t pmd = pmdp_huge_get_and_clear(vma->vm_mm, addr, pmdp);
870
871 pmd = pmd_wrprotect(pmd);
872 pmd = pmd_clear_soft_dirty(pmd);
873
874 if (vma->vm_flags & VM_SOFTDIRTY)
875 vma->vm_flags &= ~VM_SOFTDIRTY;
876
877 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
878 }
879 #else
880 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
881 unsigned long addr, pmd_t *pmdp)
882 {
883 }
884 #endif
885
886 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
887 unsigned long end, struct mm_walk *walk)
888 {
889 struct clear_refs_private *cp = walk->private;
890 struct vm_area_struct *vma = walk->vma;
891 pte_t *pte, ptent;
892 spinlock_t *ptl;
893 struct page *page;
894
895 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
896 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
897 clear_soft_dirty_pmd(vma, addr, pmd);
898 goto out;
899 }
900
901 page = pmd_page(*pmd);
902
903 /* Clear accessed and referenced bits. */
904 pmdp_test_and_clear_young(vma, addr, pmd);
905 test_and_clear_page_young(page);
906 ClearPageReferenced(page);
907 out:
908 spin_unlock(ptl);
909 return 0;
910 }
911
912 if (pmd_trans_unstable(pmd))
913 return 0;
914
915 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
916 for (; addr != end; pte++, addr += PAGE_SIZE) {
917 ptent = *pte;
918
919 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
920 clear_soft_dirty(vma, addr, pte);
921 continue;
922 }
923
924 if (!pte_present(ptent))
925 continue;
926
927 page = vm_normal_page(vma, addr, ptent);
928 if (!page)
929 continue;
930
931 /* Clear accessed and referenced bits. */
932 ptep_test_and_clear_young(vma, addr, pte);
933 test_and_clear_page_young(page);
934 ClearPageReferenced(page);
935 }
936 pte_unmap_unlock(pte - 1, ptl);
937 cond_resched();
938 return 0;
939 }
940
941 static int clear_refs_test_walk(unsigned long start, unsigned long end,
942 struct mm_walk *walk)
943 {
944 struct clear_refs_private *cp = walk->private;
945 struct vm_area_struct *vma = walk->vma;
946
947 if (vma->vm_flags & VM_PFNMAP)
948 return 1;
949
950 /*
951 * Writing 1 to /proc/pid/clear_refs affects all pages.
952 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
953 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
954 * Writing 4 to /proc/pid/clear_refs affects all pages.
955 */
956 if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
957 return 1;
958 if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
959 return 1;
960 return 0;
961 }
962
963 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
964 size_t count, loff_t *ppos)
965 {
966 struct task_struct *task;
967 char buffer[PROC_NUMBUF];
968 struct mm_struct *mm;
969 struct vm_area_struct *vma;
970 enum clear_refs_types type;
971 int itype;
972 int rv;
973
974 memset(buffer, 0, sizeof(buffer));
975 if (count > sizeof(buffer) - 1)
976 count = sizeof(buffer) - 1;
977 if (copy_from_user(buffer, buf, count))
978 return -EFAULT;
979 rv = kstrtoint(strstrip(buffer), 10, &itype);
980 if (rv < 0)
981 return rv;
982 type = (enum clear_refs_types)itype;
983 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
984 return -EINVAL;
985
986 task = get_proc_task(file_inode(file));
987 if (!task)
988 return -ESRCH;
989 mm = get_task_mm(task);
990 if (mm) {
991 struct clear_refs_private cp = {
992 .type = type,
993 };
994 struct mm_walk clear_refs_walk = {
995 .pmd_entry = clear_refs_pte_range,
996 .test_walk = clear_refs_test_walk,
997 .mm = mm,
998 .private = &cp,
999 };
1000
1001 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1002 /*
1003 * Writing 5 to /proc/pid/clear_refs resets the peak
1004 * resident set size to this mm's current rss value.
1005 */
1006 down_write(&mm->mmap_sem);
1007 reset_mm_hiwater_rss(mm);
1008 up_write(&mm->mmap_sem);
1009 goto out_mm;
1010 }
1011
1012 down_read(&mm->mmap_sem);
1013 if (type == CLEAR_REFS_SOFT_DIRTY) {
1014 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1015 if (!(vma->vm_flags & VM_SOFTDIRTY))
1016 continue;
1017 up_read(&mm->mmap_sem);
1018 down_write(&mm->mmap_sem);
1019 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1020 vma->vm_flags &= ~VM_SOFTDIRTY;
1021 vma_set_page_prot(vma);
1022 }
1023 downgrade_write(&mm->mmap_sem);
1024 break;
1025 }
1026 mmu_notifier_invalidate_range_start(mm, 0, -1);
1027 }
1028 walk_page_range(0, ~0UL, &clear_refs_walk);
1029 if (type == CLEAR_REFS_SOFT_DIRTY)
1030 mmu_notifier_invalidate_range_end(mm, 0, -1);
1031 flush_tlb_mm(mm);
1032 up_read(&mm->mmap_sem);
1033 out_mm:
1034 mmput(mm);
1035 }
1036 put_task_struct(task);
1037
1038 return count;
1039 }
1040
1041 const struct file_operations proc_clear_refs_operations = {
1042 .write = clear_refs_write,
1043 .llseek = noop_llseek,
1044 };
1045
1046 typedef struct {
1047 u64 pme;
1048 } pagemap_entry_t;
1049
1050 struct pagemapread {
1051 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */
1052 pagemap_entry_t *buffer;
1053 bool show_pfn;
1054 };
1055
1056 #define PAGEMAP_WALK_SIZE (PMD_SIZE)
1057 #define PAGEMAP_WALK_MASK (PMD_MASK)
1058
1059 #define PM_ENTRY_BYTES sizeof(pagemap_entry_t)
1060 #define PM_PFRAME_BITS 55
1061 #define PM_PFRAME_MASK GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1062 #define PM_SOFT_DIRTY BIT_ULL(55)
1063 #define PM_MMAP_EXCLUSIVE BIT_ULL(56)
1064 #define PM_FILE BIT_ULL(61)
1065 #define PM_SWAP BIT_ULL(62)
1066 #define PM_PRESENT BIT_ULL(63)
1067
1068 #define PM_END_OF_BUFFER 1
1069
1070 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1071 {
1072 return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1073 }
1074
1075 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1076 struct pagemapread *pm)
1077 {
1078 pm->buffer[pm->pos++] = *pme;
1079 if (pm->pos >= pm->len)
1080 return PM_END_OF_BUFFER;
1081 return 0;
1082 }
1083
1084 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1085 struct mm_walk *walk)
1086 {
1087 struct pagemapread *pm = walk->private;
1088 unsigned long addr = start;
1089 int err = 0;
1090
1091 while (addr < end) {
1092 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1093 pagemap_entry_t pme = make_pme(0, 0);
1094 /* End of address space hole, which we mark as non-present. */
1095 unsigned long hole_end;
1096
1097 if (vma)
1098 hole_end = min(end, vma->vm_start);
1099 else
1100 hole_end = end;
1101
1102 for (; addr < hole_end; addr += PAGE_SIZE) {
1103 err = add_to_pagemap(addr, &pme, pm);
1104 if (err)
1105 goto out;
1106 }
1107
1108 if (!vma)
1109 break;
1110
1111 /* Addresses in the VMA. */
1112 if (vma->vm_flags & VM_SOFTDIRTY)
1113 pme = make_pme(0, PM_SOFT_DIRTY);
1114 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1115 err = add_to_pagemap(addr, &pme, pm);
1116 if (err)
1117 goto out;
1118 }
1119 }
1120 out:
1121 return err;
1122 }
1123
1124 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1125 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1126 {
1127 u64 frame = 0, flags = 0;
1128 struct page *page = NULL;
1129
1130 if (pte_present(pte)) {
1131 if (pm->show_pfn)
1132 frame = pte_pfn(pte);
1133 flags |= PM_PRESENT;
1134 page = vm_normal_page(vma, addr, pte);
1135 if (pte_soft_dirty(pte))
1136 flags |= PM_SOFT_DIRTY;
1137 } else if (is_swap_pte(pte)) {
1138 swp_entry_t entry;
1139 if (pte_swp_soft_dirty(pte))
1140 flags |= PM_SOFT_DIRTY;
1141 entry = pte_to_swp_entry(pte);
1142 frame = swp_type(entry) |
1143 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1144 flags |= PM_SWAP;
1145 if (is_migration_entry(entry))
1146 page = migration_entry_to_page(entry);
1147 }
1148
1149 if (page && !PageAnon(page))
1150 flags |= PM_FILE;
1151 if (page && page_mapcount(page) == 1)
1152 flags |= PM_MMAP_EXCLUSIVE;
1153 if (vma->vm_flags & VM_SOFTDIRTY)
1154 flags |= PM_SOFT_DIRTY;
1155
1156 return make_pme(frame, flags);
1157 }
1158
1159 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1160 struct mm_walk *walk)
1161 {
1162 struct vm_area_struct *vma = walk->vma;
1163 struct pagemapread *pm = walk->private;
1164 spinlock_t *ptl;
1165 pte_t *pte, *orig_pte;
1166 int err = 0;
1167
1168 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1169 if (pmd_trans_huge_lock(pmdp, vma, &ptl) == 1) {
1170 u64 flags = 0, frame = 0;
1171 pmd_t pmd = *pmdp;
1172
1173 if ((vma->vm_flags & VM_SOFTDIRTY) || pmd_soft_dirty(pmd))
1174 flags |= PM_SOFT_DIRTY;
1175
1176 /*
1177 * Currently pmd for thp is always present because thp
1178 * can not be swapped-out, migrated, or HWPOISONed
1179 * (split in such cases instead.)
1180 * This if-check is just to prepare for future implementation.
1181 */
1182 if (pmd_present(pmd)) {
1183 struct page *page = pmd_page(pmd);
1184
1185 if (page_mapcount(page) == 1)
1186 flags |= PM_MMAP_EXCLUSIVE;
1187
1188 flags |= PM_PRESENT;
1189 if (pm->show_pfn)
1190 frame = pmd_pfn(pmd) +
1191 ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1192 }
1193
1194 for (; addr != end; addr += PAGE_SIZE) {
1195 pagemap_entry_t pme = make_pme(frame, flags);
1196
1197 err = add_to_pagemap(addr, &pme, pm);
1198 if (err)
1199 break;
1200 if (pm->show_pfn && (flags & PM_PRESENT))
1201 frame++;
1202 }
1203 spin_unlock(ptl);
1204 return err;
1205 }
1206
1207 if (pmd_trans_unstable(pmdp))
1208 return 0;
1209 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1210
1211 /*
1212 * We can assume that @vma always points to a valid one and @end never
1213 * goes beyond vma->vm_end.
1214 */
1215 orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1216 for (; addr < end; pte++, addr += PAGE_SIZE) {
1217 pagemap_entry_t pme;
1218
1219 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1220 err = add_to_pagemap(addr, &pme, pm);
1221 if (err)
1222 break;
1223 }
1224 pte_unmap_unlock(orig_pte, ptl);
1225
1226 cond_resched();
1227
1228 return err;
1229 }
1230
1231 #ifdef CONFIG_HUGETLB_PAGE
1232 /* This function walks within one hugetlb entry in the single call */
1233 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1234 unsigned long addr, unsigned long end,
1235 struct mm_walk *walk)
1236 {
1237 struct pagemapread *pm = walk->private;
1238 struct vm_area_struct *vma = walk->vma;
1239 u64 flags = 0, frame = 0;
1240 int err = 0;
1241 pte_t pte;
1242
1243 if (vma->vm_flags & VM_SOFTDIRTY)
1244 flags |= PM_SOFT_DIRTY;
1245
1246 pte = huge_ptep_get(ptep);
1247 if (pte_present(pte)) {
1248 struct page *page = pte_page(pte);
1249
1250 if (!PageAnon(page))
1251 flags |= PM_FILE;
1252
1253 if (page_mapcount(page) == 1)
1254 flags |= PM_MMAP_EXCLUSIVE;
1255
1256 flags |= PM_PRESENT;
1257 if (pm->show_pfn)
1258 frame = pte_pfn(pte) +
1259 ((addr & ~hmask) >> PAGE_SHIFT);
1260 }
1261
1262 for (; addr != end; addr += PAGE_SIZE) {
1263 pagemap_entry_t pme = make_pme(frame, flags);
1264
1265 err = add_to_pagemap(addr, &pme, pm);
1266 if (err)
1267 return err;
1268 if (pm->show_pfn && (flags & PM_PRESENT))
1269 frame++;
1270 }
1271
1272 cond_resched();
1273
1274 return err;
1275 }
1276 #endif /* HUGETLB_PAGE */
1277
1278 /*
1279 * /proc/pid/pagemap - an array mapping virtual pages to pfns
1280 *
1281 * For each page in the address space, this file contains one 64-bit entry
1282 * consisting of the following:
1283 *
1284 * Bits 0-54 page frame number (PFN) if present
1285 * Bits 0-4 swap type if swapped
1286 * Bits 5-54 swap offset if swapped
1287 * Bit 55 pte is soft-dirty (see Documentation/vm/soft-dirty.txt)
1288 * Bit 56 page exclusively mapped
1289 * Bits 57-60 zero
1290 * Bit 61 page is file-page or shared-anon
1291 * Bit 62 page swapped
1292 * Bit 63 page present
1293 *
1294 * If the page is not present but in swap, then the PFN contains an
1295 * encoding of the swap file number and the page's offset into the
1296 * swap. Unmapped pages return a null PFN. This allows determining
1297 * precisely which pages are mapped (or in swap) and comparing mapped
1298 * pages between processes.
1299 *
1300 * Efficient users of this interface will use /proc/pid/maps to
1301 * determine which areas of memory are actually mapped and llseek to
1302 * skip over unmapped regions.
1303 */
1304 static ssize_t pagemap_read(struct file *file, char __user *buf,
1305 size_t count, loff_t *ppos)
1306 {
1307 struct mm_struct *mm = file->private_data;
1308 struct pagemapread pm;
1309 struct mm_walk pagemap_walk = {};
1310 unsigned long src;
1311 unsigned long svpfn;
1312 unsigned long start_vaddr;
1313 unsigned long end_vaddr;
1314 int ret = 0, copied = 0;
1315
1316 if (!mm || !atomic_inc_not_zero(&mm->mm_users))
1317 goto out;
1318
1319 ret = -EINVAL;
1320 /* file position must be aligned */
1321 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1322 goto out_mm;
1323
1324 ret = 0;
1325 if (!count)
1326 goto out_mm;
1327
1328 /* do not disclose physical addresses: attack vector */
1329 pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1330
1331 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1332 pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY);
1333 ret = -ENOMEM;
1334 if (!pm.buffer)
1335 goto out_mm;
1336
1337 pagemap_walk.pmd_entry = pagemap_pmd_range;
1338 pagemap_walk.pte_hole = pagemap_pte_hole;
1339 #ifdef CONFIG_HUGETLB_PAGE
1340 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1341 #endif
1342 pagemap_walk.mm = mm;
1343 pagemap_walk.private = &pm;
1344
1345 src = *ppos;
1346 svpfn = src / PM_ENTRY_BYTES;
1347 start_vaddr = svpfn << PAGE_SHIFT;
1348 end_vaddr = mm->task_size;
1349
1350 /* watch out for wraparound */
1351 if (svpfn > mm->task_size >> PAGE_SHIFT)
1352 start_vaddr = end_vaddr;
1353
1354 /*
1355 * The odds are that this will stop walking way
1356 * before end_vaddr, because the length of the
1357 * user buffer is tracked in "pm", and the walk
1358 * will stop when we hit the end of the buffer.
1359 */
1360 ret = 0;
1361 while (count && (start_vaddr < end_vaddr)) {
1362 int len;
1363 unsigned long end;
1364
1365 pm.pos = 0;
1366 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1367 /* overflow ? */
1368 if (end < start_vaddr || end > end_vaddr)
1369 end = end_vaddr;
1370 down_read(&mm->mmap_sem);
1371 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1372 up_read(&mm->mmap_sem);
1373 start_vaddr = end;
1374
1375 len = min(count, PM_ENTRY_BYTES * pm.pos);
1376 if (copy_to_user(buf, pm.buffer, len)) {
1377 ret = -EFAULT;
1378 goto out_free;
1379 }
1380 copied += len;
1381 buf += len;
1382 count -= len;
1383 }
1384 *ppos += copied;
1385 if (!ret || ret == PM_END_OF_BUFFER)
1386 ret = copied;
1387
1388 out_free:
1389 kfree(pm.buffer);
1390 out_mm:
1391 mmput(mm);
1392 out:
1393 return ret;
1394 }
1395
1396 static int pagemap_open(struct inode *inode, struct file *file)
1397 {
1398 struct mm_struct *mm;
1399
1400 mm = proc_mem_open(inode, PTRACE_MODE_READ);
1401 if (IS_ERR(mm))
1402 return PTR_ERR(mm);
1403 file->private_data = mm;
1404 return 0;
1405 }
1406
1407 static int pagemap_release(struct inode *inode, struct file *file)
1408 {
1409 struct mm_struct *mm = file->private_data;
1410
1411 if (mm)
1412 mmdrop(mm);
1413 return 0;
1414 }
1415
1416 const struct file_operations proc_pagemap_operations = {
1417 .llseek = mem_lseek, /* borrow this */
1418 .read = pagemap_read,
1419 .open = pagemap_open,
1420 .release = pagemap_release,
1421 };
1422 #endif /* CONFIG_PROC_PAGE_MONITOR */
1423
1424 #ifdef CONFIG_NUMA
1425
1426 struct numa_maps {
1427 unsigned long pages;
1428 unsigned long anon;
1429 unsigned long active;
1430 unsigned long writeback;
1431 unsigned long mapcount_max;
1432 unsigned long dirty;
1433 unsigned long swapcache;
1434 unsigned long node[MAX_NUMNODES];
1435 };
1436
1437 struct numa_maps_private {
1438 struct proc_maps_private proc_maps;
1439 struct numa_maps md;
1440 };
1441
1442 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1443 unsigned long nr_pages)
1444 {
1445 int count = page_mapcount(page);
1446
1447 md->pages += nr_pages;
1448 if (pte_dirty || PageDirty(page))
1449 md->dirty += nr_pages;
1450
1451 if (PageSwapCache(page))
1452 md->swapcache += nr_pages;
1453
1454 if (PageActive(page) || PageUnevictable(page))
1455 md->active += nr_pages;
1456
1457 if (PageWriteback(page))
1458 md->writeback += nr_pages;
1459
1460 if (PageAnon(page))
1461 md->anon += nr_pages;
1462
1463 if (count > md->mapcount_max)
1464 md->mapcount_max = count;
1465
1466 md->node[page_to_nid(page)] += nr_pages;
1467 }
1468
1469 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1470 unsigned long addr)
1471 {
1472 struct page *page;
1473 int nid;
1474
1475 if (!pte_present(pte))
1476 return NULL;
1477
1478 page = vm_normal_page(vma, addr, pte);
1479 if (!page)
1480 return NULL;
1481
1482 if (PageReserved(page))
1483 return NULL;
1484
1485 nid = page_to_nid(page);
1486 if (!node_isset(nid, node_states[N_MEMORY]))
1487 return NULL;
1488
1489 return page;
1490 }
1491
1492 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1493 unsigned long end, struct mm_walk *walk)
1494 {
1495 struct numa_maps *md = walk->private;
1496 struct vm_area_struct *vma = walk->vma;
1497 spinlock_t *ptl;
1498 pte_t *orig_pte;
1499 pte_t *pte;
1500
1501 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
1502 pte_t huge_pte = *(pte_t *)pmd;
1503 struct page *page;
1504
1505 page = can_gather_numa_stats(huge_pte, vma, addr);
1506 if (page)
1507 gather_stats(page, md, pte_dirty(huge_pte),
1508 HPAGE_PMD_SIZE/PAGE_SIZE);
1509 spin_unlock(ptl);
1510 return 0;
1511 }
1512
1513 if (pmd_trans_unstable(pmd))
1514 return 0;
1515 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1516 do {
1517 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1518 if (!page)
1519 continue;
1520 gather_stats(page, md, pte_dirty(*pte), 1);
1521
1522 } while (pte++, addr += PAGE_SIZE, addr != end);
1523 pte_unmap_unlock(orig_pte, ptl);
1524 return 0;
1525 }
1526 #ifdef CONFIG_HUGETLB_PAGE
1527 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1528 unsigned long addr, unsigned long end, struct mm_walk *walk)
1529 {
1530 struct numa_maps *md;
1531 struct page *page;
1532
1533 if (!pte_present(*pte))
1534 return 0;
1535
1536 page = pte_page(*pte);
1537 if (!page)
1538 return 0;
1539
1540 md = walk->private;
1541 gather_stats(page, md, pte_dirty(*pte), 1);
1542 return 0;
1543 }
1544
1545 #else
1546 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1547 unsigned long addr, unsigned long end, struct mm_walk *walk)
1548 {
1549 return 0;
1550 }
1551 #endif
1552
1553 /*
1554 * Display pages allocated per node and memory policy via /proc.
1555 */
1556 static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1557 {
1558 struct numa_maps_private *numa_priv = m->private;
1559 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1560 struct vm_area_struct *vma = v;
1561 struct numa_maps *md = &numa_priv->md;
1562 struct file *file = vma->vm_file;
1563 struct mm_struct *mm = vma->vm_mm;
1564 struct mm_walk walk = {
1565 .hugetlb_entry = gather_hugetlb_stats,
1566 .pmd_entry = gather_pte_stats,
1567 .private = md,
1568 .mm = mm,
1569 };
1570 struct mempolicy *pol;
1571 char buffer[64];
1572 int nid;
1573
1574 if (!mm)
1575 return 0;
1576
1577 /* Ensure we start with an empty set of numa_maps statistics. */
1578 memset(md, 0, sizeof(*md));
1579
1580 pol = __get_vma_policy(vma, vma->vm_start);
1581 if (pol) {
1582 mpol_to_str(buffer, sizeof(buffer), pol);
1583 mpol_cond_put(pol);
1584 } else {
1585 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1586 }
1587
1588 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1589
1590 if (file) {
1591 seq_puts(m, " file=");
1592 seq_file_path(m, file, "\n\t= ");
1593 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1594 seq_puts(m, " heap");
1595 } else {
1596 pid_t tid = pid_of_stack(proc_priv, vma, is_pid);
1597 if (tid != 0) {
1598 /*
1599 * Thread stack in /proc/PID/task/TID/maps or
1600 * the main process stack.
1601 */
1602 if (!is_pid || (vma->vm_start <= mm->start_stack &&
1603 vma->vm_end >= mm->start_stack))
1604 seq_puts(m, " stack");
1605 else
1606 seq_printf(m, " stack:%d", tid);
1607 }
1608 }
1609
1610 if (is_vm_hugetlb_page(vma))
1611 seq_puts(m, " huge");
1612
1613 /* mmap_sem is held by m_start */
1614 walk_page_vma(vma, &walk);
1615
1616 if (!md->pages)
1617 goto out;
1618
1619 if (md->anon)
1620 seq_printf(m, " anon=%lu", md->anon);
1621
1622 if (md->dirty)
1623 seq_printf(m, " dirty=%lu", md->dirty);
1624
1625 if (md->pages != md->anon && md->pages != md->dirty)
1626 seq_printf(m, " mapped=%lu", md->pages);
1627
1628 if (md->mapcount_max > 1)
1629 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1630
1631 if (md->swapcache)
1632 seq_printf(m, " swapcache=%lu", md->swapcache);
1633
1634 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1635 seq_printf(m, " active=%lu", md->active);
1636
1637 if (md->writeback)
1638 seq_printf(m, " writeback=%lu", md->writeback);
1639
1640 for_each_node_state(nid, N_MEMORY)
1641 if (md->node[nid])
1642 seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1643
1644 seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1645 out:
1646 seq_putc(m, '\n');
1647 m_cache_vma(m, vma);
1648 return 0;
1649 }
1650
1651 static int show_pid_numa_map(struct seq_file *m, void *v)
1652 {
1653 return show_numa_map(m, v, 1);
1654 }
1655
1656 static int show_tid_numa_map(struct seq_file *m, void *v)
1657 {
1658 return show_numa_map(m, v, 0);
1659 }
1660
1661 static const struct seq_operations proc_pid_numa_maps_op = {
1662 .start = m_start,
1663 .next = m_next,
1664 .stop = m_stop,
1665 .show = show_pid_numa_map,
1666 };
1667
1668 static const struct seq_operations proc_tid_numa_maps_op = {
1669 .start = m_start,
1670 .next = m_next,
1671 .stop = m_stop,
1672 .show = show_tid_numa_map,
1673 };
1674
1675 static int numa_maps_open(struct inode *inode, struct file *file,
1676 const struct seq_operations *ops)
1677 {
1678 return proc_maps_open(inode, file, ops,
1679 sizeof(struct numa_maps_private));
1680 }
1681
1682 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1683 {
1684 return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1685 }
1686
1687 static int tid_numa_maps_open(struct inode *inode, struct file *file)
1688 {
1689 return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1690 }
1691
1692 const struct file_operations proc_pid_numa_maps_operations = {
1693 .open = pid_numa_maps_open,
1694 .read = seq_read,
1695 .llseek = seq_lseek,
1696 .release = proc_map_release,
1697 };
1698
1699 const struct file_operations proc_tid_numa_maps_operations = {
1700 .open = tid_numa_maps_open,
1701 .read = seq_read,
1702 .llseek = seq_lseek,
1703 .release = proc_map_release,
1704 };
1705 #endif /* CONFIG_NUMA */
Linux kernel - Deliverables
This page took 0.120289 seconds and 6 git commands to generate.