1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/pagewalk.h>
3 #include <linux/mm_inline.h>
4 #include <linux/hugetlb.h>
5 #include <linux/huge_mm.h>
6 #include <linux/mount.h>
7 #include <linux/ksm.h>
8 #include <linux/seq_file.h>
9 #include <linux/highmem.h>
10 #include <linux/ptrace.h>
11 #include <linux/slab.h>
12 #include <linux/pagemap.h>
13 #include <linux/mempolicy.h>
14 #include <linux/rmap.h>
15 #include <linux/swap.h>
16 #include <linux/sched/mm.h>
17 #include <linux/swapops.h>
18 #include <linux/mmu_notifier.h>
19 #include <linux/page_idle.h>
20 #include <linux/shmem_fs.h>
21 #include <linux/uaccess.h>
22 #include <linux/pkeys.h>
23
24 #include <asm/elf.h>
25 #include <asm/tlb.h>
26 #include <asm/tlbflush.h>
27 #include "internal.h"
28
29 #define SEQ_PUT_DEC(str, val) \
30 seq_put_decimal_ull_width(m, str, (val) << (PAGE_SHIFT-10), 8)
task_mem(struct seq_file * m,struct mm_struct * mm)31 void task_mem(struct seq_file *m, struct mm_struct *mm)
32 {
33 unsigned long text, lib, swap, anon, file, shmem;
34 unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
35
36 anon = get_mm_counter(mm, MM_ANONPAGES);
37 file = get_mm_counter(mm, MM_FILEPAGES);
38 shmem = get_mm_counter(mm, MM_SHMEMPAGES);
39
40 /*
41 * Note: to minimize their overhead, mm maintains hiwater_vm and
42 * hiwater_rss only when about to *lower* total_vm or rss. Any
43 * collector of these hiwater stats must therefore get total_vm
44 * and rss too, which will usually be the higher. Barriers? not
45 * worth the effort, such snapshots can always be inconsistent.
46 */
47 hiwater_vm = total_vm = mm->total_vm;
48 if (hiwater_vm < mm->hiwater_vm)
49 hiwater_vm = mm->hiwater_vm;
50 hiwater_rss = total_rss = anon + file + shmem;
51 if (hiwater_rss < mm->hiwater_rss)
52 hiwater_rss = mm->hiwater_rss;
53
54 /* split executable areas between text and lib */
55 text = PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK);
56 text = min(text, mm->exec_vm << PAGE_SHIFT);
57 lib = (mm->exec_vm << PAGE_SHIFT) - text;
58
59 swap = get_mm_counter(mm, MM_SWAPENTS);
60 SEQ_PUT_DEC("VmPeak:\t", hiwater_vm);
61 SEQ_PUT_DEC(" kB\nVmSize:\t", total_vm);
62 SEQ_PUT_DEC(" kB\nVmLck:\t", mm->locked_vm);
63 SEQ_PUT_DEC(" kB\nVmPin:\t", atomic64_read(&mm->pinned_vm));
64 SEQ_PUT_DEC(" kB\nVmHWM:\t", hiwater_rss);
65 SEQ_PUT_DEC(" kB\nVmRSS:\t", total_rss);
66 SEQ_PUT_DEC(" kB\nRssAnon:\t", anon);
67 SEQ_PUT_DEC(" kB\nRssFile:\t", file);
68 SEQ_PUT_DEC(" kB\nRssShmem:\t", shmem);
69 SEQ_PUT_DEC(" kB\nVmData:\t", mm->data_vm);
70 SEQ_PUT_DEC(" kB\nVmStk:\t", mm->stack_vm);
71 seq_put_decimal_ull_width(m,
72 " kB\nVmExe:\t", text >> 10, 8);
73 seq_put_decimal_ull_width(m,
74 " kB\nVmLib:\t", lib >> 10, 8);
75 seq_put_decimal_ull_width(m,
76 " kB\nVmPTE:\t", mm_pgtables_bytes(mm) >> 10, 8);
77 SEQ_PUT_DEC(" kB\nVmSwap:\t", swap);
78 seq_puts(m, " kB\n");
79 hugetlb_report_usage(m, mm);
80 }
81 #undef SEQ_PUT_DEC
82
task_vsize(struct mm_struct * mm)83 unsigned long task_vsize(struct mm_struct *mm)
84 {
85 return PAGE_SIZE * mm->total_vm;
86 }
87
task_statm(struct mm_struct * mm,unsigned long * shared,unsigned long * text,unsigned long * data,unsigned long * resident)88 unsigned long task_statm(struct mm_struct *mm,
89 unsigned long *shared, unsigned long *text,
90 unsigned long *data, unsigned long *resident)
91 {
92 *shared = get_mm_counter(mm, MM_FILEPAGES) +
93 get_mm_counter(mm, MM_SHMEMPAGES);
94 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
95 >> PAGE_SHIFT;
96 *data = mm->data_vm + mm->stack_vm;
97 *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
98 return mm->total_vm;
99 }
100
101 #ifdef CONFIG_NUMA
102 /*
103 * Save get_task_policy() for show_numa_map().
104 */
hold_task_mempolicy(struct proc_maps_private * priv)105 static void hold_task_mempolicy(struct proc_maps_private *priv)
106 {
107 struct task_struct *task = priv->task;
108
109 task_lock(task);
110 priv->task_mempolicy = get_task_policy(task);
111 mpol_get(priv->task_mempolicy);
112 task_unlock(task);
113 }
release_task_mempolicy(struct proc_maps_private * priv)114 static void release_task_mempolicy(struct proc_maps_private *priv)
115 {
116 mpol_put(priv->task_mempolicy);
117 }
118 #else
hold_task_mempolicy(struct proc_maps_private * priv)119 static void hold_task_mempolicy(struct proc_maps_private *priv)
120 {
121 }
release_task_mempolicy(struct proc_maps_private * priv)122 static void release_task_mempolicy(struct proc_maps_private *priv)
123 {
124 }
125 #endif
126
proc_get_vma(struct proc_maps_private * priv,loff_t * ppos)127 static struct vm_area_struct *proc_get_vma(struct proc_maps_private *priv,
128 loff_t *ppos)
129 {
130 struct vm_area_struct *vma = vma_next(&priv->iter);
131
132 if (vma) {
133 *ppos = vma->vm_start;
134 } else {
135 *ppos = -2UL;
136 vma = get_gate_vma(priv->mm);
137 }
138
139 return vma;
140 }
141
m_start(struct seq_file * m,loff_t * ppos)142 static void *m_start(struct seq_file *m, loff_t *ppos)
143 {
144 struct proc_maps_private *priv = m->private;
145 unsigned long last_addr = *ppos;
146 struct mm_struct *mm;
147
148 /* See m_next(). Zero at the start or after lseek. */
149 if (last_addr == -1UL)
150 return NULL;
151
152 priv->task = get_proc_task(priv->inode);
153 if (!priv->task)
154 return ERR_PTR(-ESRCH);
155
156 mm = priv->mm;
157 if (!mm || !mmget_not_zero(mm)) {
158 put_task_struct(priv->task);
159 priv->task = NULL;
160 return NULL;
161 }
162
163 if (mmap_read_lock_killable(mm)) {
164 mmput(mm);
165 put_task_struct(priv->task);
166 priv->task = NULL;
167 return ERR_PTR(-EINTR);
168 }
169
170 vma_iter_init(&priv->iter, mm, last_addr);
171 hold_task_mempolicy(priv);
172 if (last_addr == -2UL)
173 return get_gate_vma(mm);
174
175 return proc_get_vma(priv, ppos);
176 }
177
m_next(struct seq_file * m,void * v,loff_t * ppos)178 static void *m_next(struct seq_file *m, void *v, loff_t *ppos)
179 {
180 if (*ppos == -2UL) {
181 *ppos = -1UL;
182 return NULL;
183 }
184 return proc_get_vma(m->private, ppos);
185 }
186
m_stop(struct seq_file * m,void * v)187 static void m_stop(struct seq_file *m, void *v)
188 {
189 struct proc_maps_private *priv = m->private;
190 struct mm_struct *mm = priv->mm;
191
192 if (!priv->task)
193 return;
194
195 release_task_mempolicy(priv);
196 mmap_read_unlock(mm);
197 mmput(mm);
198 put_task_struct(priv->task);
199 priv->task = NULL;
200 }
201
proc_maps_open(struct inode * inode,struct file * file,const struct seq_operations * ops,int psize)202 static int proc_maps_open(struct inode *inode, struct file *file,
203 const struct seq_operations *ops, int psize)
204 {
205 struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
206
207 if (!priv)
208 return -ENOMEM;
209
210 priv->inode = inode;
211 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
212 if (IS_ERR(priv->mm)) {
213 int err = PTR_ERR(priv->mm);
214
215 seq_release_private(inode, file);
216 return err;
217 }
218
219 return 0;
220 }
221
proc_map_release(struct inode * inode,struct file * file)222 static int proc_map_release(struct inode *inode, struct file *file)
223 {
224 struct seq_file *seq = file->private_data;
225 struct proc_maps_private *priv = seq->private;
226
227 if (priv->mm)
228 mmdrop(priv->mm);
229
230 return seq_release_private(inode, file);
231 }
232
do_maps_open(struct inode * inode,struct file * file,const struct seq_operations * ops)233 static int do_maps_open(struct inode *inode, struct file *file,
234 const struct seq_operations *ops)
235 {
236 return proc_maps_open(inode, file, ops,
237 sizeof(struct proc_maps_private));
238 }
239
show_vma_header_prefix(struct seq_file * m,unsigned long start,unsigned long end,vm_flags_t flags,unsigned long long pgoff,dev_t dev,unsigned long ino)240 static void show_vma_header_prefix(struct seq_file *m,
241 unsigned long start, unsigned long end,
242 vm_flags_t flags, unsigned long long pgoff,
243 dev_t dev, unsigned long ino)
244 {
245 seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
246 seq_put_hex_ll(m, NULL, start, 8);
247 seq_put_hex_ll(m, "-", end, 8);
248 seq_putc(m, ' ');
249 seq_putc(m, flags & VM_READ ? 'r' : '-');
250 seq_putc(m, flags & VM_WRITE ? 'w' : '-');
251 seq_putc(m, flags & VM_EXEC ? 'x' : '-');
252 seq_putc(m, flags & VM_MAYSHARE ? 's' : 'p');
253 seq_put_hex_ll(m, " ", pgoff, 8);
254 seq_put_hex_ll(m, " ", MAJOR(dev), 2);
255 seq_put_hex_ll(m, ":", MINOR(dev), 2);
256 seq_put_decimal_ull(m, " ", ino);
257 seq_putc(m, ' ');
258 }
259
260 static void
show_map_vma(struct seq_file * m,struct vm_area_struct * vma)261 show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
262 {
263 struct anon_vma_name *anon_name = NULL;
264 struct mm_struct *mm = vma->vm_mm;
265 struct file *file = vma->vm_file;
266 vm_flags_t flags = vma->vm_flags;
267 unsigned long ino = 0;
268 unsigned long long pgoff = 0;
269 unsigned long start, end;
270 dev_t dev = 0;
271 const char *name = NULL;
272
273 if (file) {
274 struct inode *inode = file_inode(vma->vm_file);
275 dev = inode->i_sb->s_dev;
276 ino = inode->i_ino;
277 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
278 }
279
280 start = vma->vm_start;
281 end = vma->vm_end;
282 show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino);
283 if (mm)
284 anon_name = anon_vma_name(vma);
285
286 /*
287 * Print the dentry name for named mappings, and a
288 * special [heap] marker for the heap:
289 */
290 if (file) {
291 seq_pad(m, ' ');
292 /*
293 * If user named this anon shared memory via
294 * prctl(PR_SET_VMA ..., use the provided name.
295 */
296 if (anon_name)
297 seq_printf(m, "[anon_shmem:%s]", anon_name->name);
298 else
299 seq_file_path(m, file, "\n");
300 goto done;
301 }
302
303 if (vma->vm_ops && vma->vm_ops->name) {
304 name = vma->vm_ops->name(vma);
305 if (name)
306 goto done;
307 }
308
309 name = arch_vma_name(vma);
310 if (!name) {
311 if (!mm) {
312 name = "[vdso]";
313 goto done;
314 }
315
316 if (vma_is_initial_heap(vma)) {
317 name = "[heap]";
318 goto done;
319 }
320
321 if (vma_is_initial_stack(vma)) {
322 name = "[stack]";
323 goto done;
324 }
325
326 if (anon_name) {
327 seq_pad(m, ' ');
328 seq_printf(m, "[anon:%s]", anon_name->name);
329 }
330 }
331
332 done:
333 if (name) {
334 seq_pad(m, ' ');
335 seq_puts(m, name);
336 }
337 seq_putc(m, '\n');
338 }
339
show_map(struct seq_file * m,void * v)340 static int show_map(struct seq_file *m, void *v)
341 {
342 show_map_vma(m, v);
343 return 0;
344 }
345
346 static const struct seq_operations proc_pid_maps_op = {
347 .start = m_start,
348 .next = m_next,
349 .stop = m_stop,
350 .show = show_map
351 };
352
pid_maps_open(struct inode * inode,struct file * file)353 static int pid_maps_open(struct inode *inode, struct file *file)
354 {
355 return do_maps_open(inode, file, &proc_pid_maps_op);
356 }
357
358 const struct file_operations proc_pid_maps_operations = {
359 .open = pid_maps_open,
360 .read = seq_read,
361 .llseek = seq_lseek,
362 .release = proc_map_release,
363 };
364
365 /*
366 * Proportional Set Size(PSS): my share of RSS.
367 *
368 * PSS of a process is the count of pages it has in memory, where each
369 * page is divided by the number of processes sharing it. So if a
370 * process has 1000 pages all to itself, and 1000 shared with one other
371 * process, its PSS will be 1500.
372 *
373 * To keep (accumulated) division errors low, we adopt a 64bit
374 * fixed-point pss counter to minimize division errors. So (pss >>
375 * PSS_SHIFT) would be the real byte count.
376 *
377 * A shift of 12 before division means (assuming 4K page size):
378 * - 1M 3-user-pages add up to 8KB errors;
379 * - supports mapcount up to 2^24, or 16M;
380 * - supports PSS up to 2^52 bytes, or 4PB.
381 */
382 #define PSS_SHIFT 12
383
384 #ifdef CONFIG_PROC_PAGE_MONITOR
385 struct mem_size_stats {
386 unsigned long resident;
387 unsigned long shared_clean;
388 unsigned long shared_dirty;
389 unsigned long private_clean;
390 unsigned long private_dirty;
391 unsigned long referenced;
392 unsigned long anonymous;
393 unsigned long lazyfree;
394 unsigned long anonymous_thp;
395 unsigned long shmem_thp;
396 unsigned long file_thp;
397 unsigned long swap;
398 unsigned long shared_hugetlb;
399 unsigned long private_hugetlb;
400 unsigned long ksm;
401 u64 pss;
402 u64 pss_anon;
403 u64 pss_file;
404 u64 pss_shmem;
405 u64 pss_dirty;
406 u64 pss_locked;
407 u64 swap_pss;
408 };
409
smaps_page_accumulate(struct mem_size_stats * mss,struct page * page,unsigned long size,unsigned long pss,bool dirty,bool locked,bool private)410 static void smaps_page_accumulate(struct mem_size_stats *mss,
411 struct page *page, unsigned long size, unsigned long pss,
412 bool dirty, bool locked, bool private)
413 {
414 mss->pss += pss;
415
416 if (PageAnon(page))
417 mss->pss_anon += pss;
418 else if (PageSwapBacked(page))
419 mss->pss_shmem += pss;
420 else
421 mss->pss_file += pss;
422
423 if (locked)
424 mss->pss_locked += pss;
425
426 if (dirty || PageDirty(page)) {
427 mss->pss_dirty += pss;
428 if (private)
429 mss->private_dirty += size;
430 else
431 mss->shared_dirty += size;
432 } else {
433 if (private)
434 mss->private_clean += size;
435 else
436 mss->shared_clean += size;
437 }
438 }
439
smaps_account(struct mem_size_stats * mss,struct page * page,bool compound,bool young,bool dirty,bool locked,bool migration)440 static void smaps_account(struct mem_size_stats *mss, struct page *page,
441 bool compound, bool young, bool dirty, bool locked,
442 bool migration)
443 {
444 int i, nr = compound ? compound_nr(page) : 1;
445 unsigned long size = nr * PAGE_SIZE;
446
447 /*
448 * First accumulate quantities that depend only on |size| and the type
449 * of the compound page.
450 */
451 if (PageAnon(page)) {
452 mss->anonymous += size;
453 if (!PageSwapBacked(page) && !dirty && !PageDirty(page))
454 mss->lazyfree += size;
455 }
456
457 if (PageKsm(page))
458 mss->ksm += size;
459
460 mss->resident += size;
461 /* Accumulate the size in pages that have been accessed. */
462 if (young || page_is_young(page) || PageReferenced(page))
463 mss->referenced += size;
464
465 /*
466 * Then accumulate quantities that may depend on sharing, or that may
467 * differ page-by-page.
468 *
469 * page_count(page) == 1 guarantees the page is mapped exactly once.
470 * If any subpage of the compound page mapped with PTE it would elevate
471 * page_count().
472 *
473 * The page_mapcount() is called to get a snapshot of the mapcount.
474 * Without holding the page lock this snapshot can be slightly wrong as
475 * we cannot always read the mapcount atomically. It is not safe to
476 * call page_mapcount() even with PTL held if the page is not mapped,
477 * especially for migration entries. Treat regular migration entries
478 * as mapcount == 1.
479 */
480 if ((page_count(page) == 1) || migration) {
481 smaps_page_accumulate(mss, page, size, size << PSS_SHIFT, dirty,
482 locked, true);
483 return;
484 }
485 for (i = 0; i < nr; i++, page++) {
486 int mapcount = page_mapcount(page);
487 unsigned long pss = PAGE_SIZE << PSS_SHIFT;
488 if (mapcount >= 2)
489 pss /= mapcount;
490 smaps_page_accumulate(mss, page, PAGE_SIZE, pss, dirty, locked,
491 mapcount < 2);
492 }
493 }
494
495 #ifdef CONFIG_SHMEM
smaps_pte_hole(unsigned long addr,unsigned long end,__always_unused int depth,struct mm_walk * walk)496 static int smaps_pte_hole(unsigned long addr, unsigned long end,
497 __always_unused int depth, struct mm_walk *walk)
498 {
499 struct mem_size_stats *mss = walk->private;
500 struct vm_area_struct *vma = walk->vma;
501
502 mss->swap += shmem_partial_swap_usage(walk->vma->vm_file->f_mapping,
503 linear_page_index(vma, addr),
504 linear_page_index(vma, end));
505
506 return 0;
507 }
508 #else
509 #define smaps_pte_hole NULL
510 #endif /* CONFIG_SHMEM */
511
smaps_pte_hole_lookup(unsigned long addr,struct mm_walk * walk)512 static void smaps_pte_hole_lookup(unsigned long addr, struct mm_walk *walk)
513 {
514 #ifdef CONFIG_SHMEM
515 if (walk->ops->pte_hole) {
516 /* depth is not used */
517 smaps_pte_hole(addr, addr + PAGE_SIZE, 0, walk);
518 }
519 #endif
520 }
521
smaps_pte_entry(pte_t * pte,unsigned long addr,struct mm_walk * walk)522 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
523 struct mm_walk *walk)
524 {
525 struct mem_size_stats *mss = walk->private;
526 struct vm_area_struct *vma = walk->vma;
527 bool locked = !!(vma->vm_flags & VM_LOCKED);
528 struct page *page = NULL;
529 bool migration = false, young = false, dirty = false;
530 pte_t ptent = ptep_get(pte);
531
532 if (pte_present(ptent)) {
533 page = vm_normal_page(vma, addr, ptent);
534 young = pte_young(ptent);
535 dirty = pte_dirty(ptent);
536 } else if (is_swap_pte(ptent)) {
537 swp_entry_t swpent = pte_to_swp_entry(ptent);
538
539 if (!non_swap_entry(swpent)) {
540 int mapcount;
541
542 mss->swap += PAGE_SIZE;
543 mapcount = swp_swapcount(swpent);
544 if (mapcount >= 2) {
545 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
546
547 do_div(pss_delta, mapcount);
548 mss->swap_pss += pss_delta;
549 } else {
550 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
551 }
552 } else if (is_pfn_swap_entry(swpent)) {
553 if (is_migration_entry(swpent))
554 migration = true;
555 page = pfn_swap_entry_to_page(swpent);
556 }
557 } else {
558 smaps_pte_hole_lookup(addr, walk);
559 return;
560 }
561
562 if (!page)
563 return;
564
565 smaps_account(mss, page, false, young, dirty, locked, migration);
566 }
567
568 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
smaps_pmd_entry(pmd_t * pmd,unsigned long addr,struct mm_walk * walk)569 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
570 struct mm_walk *walk)
571 {
572 struct mem_size_stats *mss = walk->private;
573 struct vm_area_struct *vma = walk->vma;
574 bool locked = !!(vma->vm_flags & VM_LOCKED);
575 struct page *page = NULL;
576 bool migration = false;
577
578 if (pmd_present(*pmd)) {
579 page = vm_normal_page_pmd(vma, addr, *pmd);
580 } else if (unlikely(thp_migration_supported() && is_swap_pmd(*pmd))) {
581 swp_entry_t entry = pmd_to_swp_entry(*pmd);
582
583 if (is_migration_entry(entry)) {
584 migration = true;
585 page = pfn_swap_entry_to_page(entry);
586 }
587 }
588 if (IS_ERR_OR_NULL(page))
589 return;
590 if (PageAnon(page))
591 mss->anonymous_thp += HPAGE_PMD_SIZE;
592 else if (PageSwapBacked(page))
593 mss->shmem_thp += HPAGE_PMD_SIZE;
594 else if (is_zone_device_page(page))
595 /* pass */;
596 else
597 mss->file_thp += HPAGE_PMD_SIZE;
598
599 smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd),
600 locked, migration);
601 }
602 #else
smaps_pmd_entry(pmd_t * pmd,unsigned long addr,struct mm_walk * walk)603 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
604 struct mm_walk *walk)
605 {
606 }
607 #endif
608
smaps_pte_range(pmd_t * pmd,unsigned long addr,unsigned long end,struct mm_walk * walk)609 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
610 struct mm_walk *walk)
611 {
612 struct vm_area_struct *vma = walk->vma;
613 pte_t *pte;
614 spinlock_t *ptl;
615
616 ptl = pmd_trans_huge_lock(pmd, vma);
617 if (ptl) {
618 smaps_pmd_entry(pmd, addr, walk);
619 spin_unlock(ptl);
620 goto out;
621 }
622
623 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
624 if (!pte) {
625 walk->action = ACTION_AGAIN;
626 return 0;
627 }
628 for (; addr != end; pte++, addr += PAGE_SIZE)
629 smaps_pte_entry(pte, addr, walk);
630 pte_unmap_unlock(pte - 1, ptl);
631 out:
632 cond_resched();
633 return 0;
634 }
635
show_smap_vma_flags(struct seq_file * m,struct vm_area_struct * vma)636 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
637 {
638 /*
639 * Don't forget to update Documentation/ on changes.
640 */
641 static const char mnemonics[BITS_PER_LONG][2] = {
642 /*
643 * In case if we meet a flag we don't know about.
644 */
645 [0 ... (BITS_PER_LONG-1)] = "??",
646
647 [ilog2(VM_READ)] = "rd",
648 [ilog2(VM_WRITE)] = "wr",
649 [ilog2(VM_EXEC)] = "ex",
650 [ilog2(VM_SHARED)] = "sh",
651 [ilog2(VM_MAYREAD)] = "mr",
652 [ilog2(VM_MAYWRITE)] = "mw",
653 [ilog2(VM_MAYEXEC)] = "me",
654 [ilog2(VM_MAYSHARE)] = "ms",
655 [ilog2(VM_GROWSDOWN)] = "gd",
656 [ilog2(VM_PFNMAP)] = "pf",
657 [ilog2(VM_LOCKED)] = "lo",
658 [ilog2(VM_IO)] = "io",
659 [ilog2(VM_SEQ_READ)] = "sr",
660 [ilog2(VM_RAND_READ)] = "rr",
661 [ilog2(VM_DONTCOPY)] = "dc",
662 [ilog2(VM_DONTEXPAND)] = "de",
663 [ilog2(VM_LOCKONFAULT)] = "lf",
664 [ilog2(VM_ACCOUNT)] = "ac",
665 [ilog2(VM_NORESERVE)] = "nr",
666 [ilog2(VM_HUGETLB)] = "ht",
667 [ilog2(VM_SYNC)] = "sf",
668 [ilog2(VM_ARCH_1)] = "ar",
669 [ilog2(VM_WIPEONFORK)] = "wf",
670 [ilog2(VM_DONTDUMP)] = "dd",
671 #ifdef CONFIG_ARM64_BTI
672 [ilog2(VM_ARM64_BTI)] = "bt",
673 #endif
674 #ifdef CONFIG_MEM_SOFT_DIRTY
675 [ilog2(VM_SOFTDIRTY)] = "sd",
676 #endif
677 [ilog2(VM_MIXEDMAP)] = "mm",
678 [ilog2(VM_HUGEPAGE)] = "hg",
679 [ilog2(VM_NOHUGEPAGE)] = "nh",
680 [ilog2(VM_MERGEABLE)] = "mg",
681 [ilog2(VM_UFFD_MISSING)]= "um",
682 [ilog2(VM_UFFD_WP)] = "uw",
683 #ifdef CONFIG_ARM64_MTE
684 [ilog2(VM_MTE)] = "mt",
685 [ilog2(VM_MTE_ALLOWED)] = "",
686 #endif
687 #ifdef CONFIG_ARCH_HAS_PKEYS
688 /* These come out via ProtectionKey: */
689 [ilog2(VM_PKEY_BIT0)] = "",
690 [ilog2(VM_PKEY_BIT1)] = "",
691 [ilog2(VM_PKEY_BIT2)] = "",
692 [ilog2(VM_PKEY_BIT3)] = "",
693 #if VM_PKEY_BIT4
694 [ilog2(VM_PKEY_BIT4)] = "",
695 #endif
696 #endif /* CONFIG_ARCH_HAS_PKEYS */
697 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_MINOR
698 [ilog2(VM_UFFD_MINOR)] = "ui",
699 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_MINOR */
700 #ifdef CONFIG_X86_USER_SHADOW_STACK
701 [ilog2(VM_SHADOW_STACK)] = "ss",
702 #endif
703 };
704 size_t i;
705
706 seq_puts(m, "VmFlags: ");
707 for (i = 0; i < BITS_PER_LONG; i++) {
708 if (!mnemonics[i][0])
709 continue;
710 if (vma->vm_flags & (1UL << i)) {
711 seq_putc(m, mnemonics[i][0]);
712 seq_putc(m, mnemonics[i][1]);
713 seq_putc(m, ' ');
714 }
715 }
716 seq_putc(m, '\n');
717 }
718
719 #ifdef CONFIG_HUGETLB_PAGE
smaps_hugetlb_range(pte_t * pte,unsigned long hmask,unsigned long addr,unsigned long end,struct mm_walk * walk)720 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
721 unsigned long addr, unsigned long end,
722 struct mm_walk *walk)
723 {
724 struct mem_size_stats *mss = walk->private;
725 struct vm_area_struct *vma = walk->vma;
726 struct page *page = NULL;
727 pte_t ptent = ptep_get(pte);
728
729 if (pte_present(ptent)) {
730 page = vm_normal_page(vma, addr, ptent);
731 } else if (is_swap_pte(ptent)) {
732 swp_entry_t swpent = pte_to_swp_entry(ptent);
733
734 if (is_pfn_swap_entry(swpent))
735 page = pfn_swap_entry_to_page(swpent);
736 }
737 if (page) {
738 if (page_mapcount(page) >= 2 || hugetlb_pmd_shared(pte))
739 mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
740 else
741 mss->private_hugetlb += huge_page_size(hstate_vma(vma));
742 }
743 return 0;
744 }
745 #else
746 #define smaps_hugetlb_range NULL
747 #endif /* HUGETLB_PAGE */
748
749 static const struct mm_walk_ops smaps_walk_ops = {
750 .pmd_entry = smaps_pte_range,
751 .hugetlb_entry = smaps_hugetlb_range,
752 .walk_lock = PGWALK_RDLOCK,
753 };
754
755 static const struct mm_walk_ops smaps_shmem_walk_ops = {
756 .pmd_entry = smaps_pte_range,
757 .hugetlb_entry = smaps_hugetlb_range,
758 .pte_hole = smaps_pte_hole,
759 .walk_lock = PGWALK_RDLOCK,
760 };
761
762 /*
763 * Gather mem stats from @vma with the indicated beginning
764 * address @start, and keep them in @mss.
765 *
766 * Use vm_start of @vma as the beginning address if @start is 0.
767 */
smap_gather_stats(struct vm_area_struct * vma,struct mem_size_stats * mss,unsigned long start)768 static void smap_gather_stats(struct vm_area_struct *vma,
769 struct mem_size_stats *mss, unsigned long start)
770 {
771 const struct mm_walk_ops *ops = &smaps_walk_ops;
772
773 /* Invalid start */
774 if (start >= vma->vm_end)
775 return;
776
777 if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
778 /*
779 * For shared or readonly shmem mappings we know that all
780 * swapped out pages belong to the shmem object, and we can
781 * obtain the swap value much more efficiently. For private
782 * writable mappings, we might have COW pages that are
783 * not affected by the parent swapped out pages of the shmem
784 * object, so we have to distinguish them during the page walk.
785 * Unless we know that the shmem object (or the part mapped by
786 * our VMA) has no swapped out pages at all.
787 */
788 unsigned long shmem_swapped = shmem_swap_usage(vma);
789
790 if (!start && (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
791 !(vma->vm_flags & VM_WRITE))) {
792 mss->swap += shmem_swapped;
793 } else {
794 ops = &smaps_shmem_walk_ops;
795 }
796 }
797
798 /* mmap_lock is held in m_start */
799 if (!start)
800 walk_page_vma(vma, ops, mss);
801 else
802 walk_page_range(vma->vm_mm, start, vma->vm_end, ops, mss);
803 }
804
805 #define SEQ_PUT_DEC(str, val) \
806 seq_put_decimal_ull_width(m, str, (val) >> 10, 8)
807
808 /* Show the contents common for smaps and smaps_rollup */
__show_smap(struct seq_file * m,const struct mem_size_stats * mss,bool rollup_mode)809 static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss,
810 bool rollup_mode)
811 {
812 SEQ_PUT_DEC("Rss: ", mss->resident);
813 SEQ_PUT_DEC(" kB\nPss: ", mss->pss >> PSS_SHIFT);
814 SEQ_PUT_DEC(" kB\nPss_Dirty: ", mss->pss_dirty >> PSS_SHIFT);
815 if (rollup_mode) {
816 /*
817 * These are meaningful only for smaps_rollup, otherwise two of
818 * them are zero, and the other one is the same as Pss.
819 */
820 SEQ_PUT_DEC(" kB\nPss_Anon: ",
821 mss->pss_anon >> PSS_SHIFT);
822 SEQ_PUT_DEC(" kB\nPss_File: ",
823 mss->pss_file >> PSS_SHIFT);
824 SEQ_PUT_DEC(" kB\nPss_Shmem: ",
825 mss->pss_shmem >> PSS_SHIFT);
826 }
827 SEQ_PUT_DEC(" kB\nShared_Clean: ", mss->shared_clean);
828 SEQ_PUT_DEC(" kB\nShared_Dirty: ", mss->shared_dirty);
829 SEQ_PUT_DEC(" kB\nPrivate_Clean: ", mss->private_clean);
830 SEQ_PUT_DEC(" kB\nPrivate_Dirty: ", mss->private_dirty);
831 SEQ_PUT_DEC(" kB\nReferenced: ", mss->referenced);
832 SEQ_PUT_DEC(" kB\nAnonymous: ", mss->anonymous);
833 SEQ_PUT_DEC(" kB\nKSM: ", mss->ksm);
834 SEQ_PUT_DEC(" kB\nLazyFree: ", mss->lazyfree);
835 SEQ_PUT_DEC(" kB\nAnonHugePages: ", mss->anonymous_thp);
836 SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp);
837 SEQ_PUT_DEC(" kB\nFilePmdMapped: ", mss->file_thp);
838 SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb);
839 seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ",
840 mss->private_hugetlb >> 10, 7);
841 SEQ_PUT_DEC(" kB\nSwap: ", mss->swap);
842 SEQ_PUT_DEC(" kB\nSwapPss: ",
843 mss->swap_pss >> PSS_SHIFT);
844 SEQ_PUT_DEC(" kB\nLocked: ",
845 mss->pss_locked >> PSS_SHIFT);
846 seq_puts(m, " kB\n");
847 }
848
show_smap(struct seq_file * m,void * v)849 static int show_smap(struct seq_file *m, void *v)
850 {
851 struct vm_area_struct *vma = v;
852 struct mem_size_stats mss;
853
854 memset(&mss, 0, sizeof(mss));
855
856 smap_gather_stats(vma, &mss, 0);
857
858 show_map_vma(m, vma);
859
860 SEQ_PUT_DEC("Size: ", vma->vm_end - vma->vm_start);
861 SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma));
862 SEQ_PUT_DEC(" kB\nMMUPageSize: ", vma_mmu_pagesize(vma));
863 seq_puts(m, " kB\n");
864
865 __show_smap(m, &mss, false);
866
867 seq_printf(m, "THPeligible: %8u\n",
868 hugepage_vma_check(vma, vma->vm_flags, true, false, true));
869
870 if (arch_pkeys_enabled())
871 seq_printf(m, "ProtectionKey: %8u\n", vma_pkey(vma));
872 show_smap_vma_flags(m, vma);
873
874 return 0;
875 }
876
show_smaps_rollup(struct seq_file * m,void * v)877 static int show_smaps_rollup(struct seq_file *m, void *v)
878 {
879 struct proc_maps_private *priv = m->private;
880 struct mem_size_stats mss;
881 struct mm_struct *mm = priv->mm;
882 struct vm_area_struct *vma;
883 unsigned long vma_start = 0, last_vma_end = 0;
884 int ret = 0;
885 VMA_ITERATOR(vmi, mm, 0);
886
887 priv->task = get_proc_task(priv->inode);
888 if (!priv->task)
889 return -ESRCH;
890
891 if (!mm || !mmget_not_zero(mm)) {
892 ret = -ESRCH;
893 goto out_put_task;
894 }
895
896 memset(&mss, 0, sizeof(mss));
897
898 ret = mmap_read_lock_killable(mm);
899 if (ret)
900 goto out_put_mm;
901
902 hold_task_mempolicy(priv);
903 vma = vma_next(&vmi);
904
905 if (unlikely(!vma))
906 goto empty_set;
907
908 vma_start = vma->vm_start;
909 do {
910 smap_gather_stats(vma, &mss, 0);
911 last_vma_end = vma->vm_end;
912
913 /*
914 * Release mmap_lock temporarily if someone wants to
915 * access it for write request.
916 */
917 if (mmap_lock_is_contended(mm)) {
918 vma_iter_invalidate(&vmi);
919 mmap_read_unlock(mm);
920 ret = mmap_read_lock_killable(mm);
921 if (ret) {
922 release_task_mempolicy(priv);
923 goto out_put_mm;
924 }
925
926 /*
927 * After dropping the lock, there are four cases to
928 * consider. See the following example for explanation.
929 *
930 * +------+------+-----------+
931 * | VMA1 | VMA2 | VMA3 |
932 * +------+------+-----------+
933 * | | | |
934 * 4k 8k 16k 400k
935 *
936 * Suppose we drop the lock after reading VMA2 due to
937 * contention, then we get:
938 *
939 * last_vma_end = 16k
940 *
941 * 1) VMA2 is freed, but VMA3 exists:
942 *
943 * vma_next(vmi) will return VMA3.
944 * In this case, just continue from VMA3.
945 *
946 * 2) VMA2 still exists:
947 *
948 * vma_next(vmi) will return VMA3.
949 * In this case, just continue from VMA3.
950 *
951 * 3) No more VMAs can be found:
952 *
953 * vma_next(vmi) will return NULL.
954 * No more things to do, just break.
955 *
956 * 4) (last_vma_end - 1) is the middle of a vma (VMA'):
957 *
958 * vma_next(vmi) will return VMA' whose range
959 * contains last_vma_end.
960 * Iterate VMA' from last_vma_end.
961 */
962 vma = vma_next(&vmi);
963 /* Case 3 above */
964 if (!vma)
965 break;
966
967 /* Case 1 and 2 above */
968 if (vma->vm_start >= last_vma_end) {
969 smap_gather_stats(vma, &mss, 0);
970 last_vma_end = vma->vm_end;
971 continue;
972 }
973
974 /* Case 4 above */
975 if (vma->vm_end > last_vma_end) {
976 smap_gather_stats(vma, &mss, last_vma_end);
977 last_vma_end = vma->vm_end;
978 }
979 }
980 } for_each_vma(vmi, vma);
981
982 empty_set:
983 show_vma_header_prefix(m, vma_start, last_vma_end, 0, 0, 0, 0);
984 seq_pad(m, ' ');
985 seq_puts(m, "[rollup]\n");
986
987 __show_smap(m, &mss, true);
988
989 release_task_mempolicy(priv);
990 mmap_read_unlock(mm);
991
992 out_put_mm:
993 mmput(mm);
994 out_put_task:
995 put_task_struct(priv->task);
996 priv->task = NULL;
997
998 return ret;
999 }
1000 #undef SEQ_PUT_DEC
1001
1002 static const struct seq_operations proc_pid_smaps_op = {
1003 .start = m_start,
1004 .next = m_next,
1005 .stop = m_stop,
1006 .show = show_smap
1007 };
1008
pid_smaps_open(struct inode * inode,struct file * file)1009 static int pid_smaps_open(struct inode *inode, struct file *file)
1010 {
1011 return do_maps_open(inode, file, &proc_pid_smaps_op);
1012 }
1013
smaps_rollup_open(struct inode * inode,struct file * file)1014 static int smaps_rollup_open(struct inode *inode, struct file *file)
1015 {
1016 int ret;
1017 struct proc_maps_private *priv;
1018
1019 priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT);
1020 if (!priv)
1021 return -ENOMEM;
1022
1023 ret = single_open(file, show_smaps_rollup, priv);
1024 if (ret)
1025 goto out_free;
1026
1027 priv->inode = inode;
1028 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
1029 if (IS_ERR(priv->mm)) {
1030 ret = PTR_ERR(priv->mm);
1031
1032 single_release(inode, file);
1033 goto out_free;
1034 }
1035
1036 return 0;
1037
1038 out_free:
1039 kfree(priv);
1040 return ret;
1041 }
1042
smaps_rollup_release(struct inode * inode,struct file * file)1043 static int smaps_rollup_release(struct inode *inode, struct file *file)
1044 {
1045 struct seq_file *seq = file->private_data;
1046 struct proc_maps_private *priv = seq->private;
1047
1048 if (priv->mm)
1049 mmdrop(priv->mm);
1050
1051 kfree(priv);
1052 return single_release(inode, file);
1053 }
1054
1055 const struct file_operations proc_pid_smaps_operations = {
1056 .open = pid_smaps_open,
1057 .read = seq_read,
1058 .llseek = seq_lseek,
1059 .release = proc_map_release,
1060 };
1061
1062 const struct file_operations proc_pid_smaps_rollup_operations = {
1063 .open = smaps_rollup_open,
1064 .read = seq_read,
1065 .llseek = seq_lseek,
1066 .release = smaps_rollup_release,
1067 };
1068
1069 enum clear_refs_types {
1070 CLEAR_REFS_ALL = 1,
1071 CLEAR_REFS_ANON,
1072 CLEAR_REFS_MAPPED,
1073 CLEAR_REFS_SOFT_DIRTY,
1074 CLEAR_REFS_MM_HIWATER_RSS,
1075 CLEAR_REFS_LAST,
1076 };
1077
1078 struct clear_refs_private {
1079 enum clear_refs_types type;
1080 };
1081
1082 #ifdef CONFIG_MEM_SOFT_DIRTY
1083
pte_is_pinned(struct vm_area_struct * vma,unsigned long addr,pte_t pte)1084 static inline bool pte_is_pinned(struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1085 {
1086 struct page *page;
1087
1088 if (!pte_write(pte))
1089 return false;
1090 if (!is_cow_mapping(vma->vm_flags))
1091 return false;
1092 if (likely(!test_bit(MMF_HAS_PINNED, &vma->vm_mm->flags)))
1093 return false;
1094 page = vm_normal_page(vma, addr, pte);
1095 if (!page)
1096 return false;
1097 return page_maybe_dma_pinned(page);
1098 }
1099
clear_soft_dirty(struct vm_area_struct * vma,unsigned long addr,pte_t * pte)1100 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1101 unsigned long addr, pte_t *pte)
1102 {
1103 /*
1104 * The soft-dirty tracker uses #PF-s to catch writes
1105 * to pages, so write-protect the pte as well. See the
1106 * Documentation/admin-guide/mm/soft-dirty.rst for full description
1107 * of how soft-dirty works.
1108 */
1109 pte_t ptent = ptep_get(pte);
1110
1111 if (pte_present(ptent)) {
1112 pte_t old_pte;
1113
1114 if (pte_is_pinned(vma, addr, ptent))
1115 return;
1116 old_pte = ptep_modify_prot_start(vma, addr, pte);
1117 ptent = pte_wrprotect(old_pte);
1118 ptent = pte_clear_soft_dirty(ptent);
1119 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent);
1120 } else if (is_swap_pte(ptent)) {
1121 ptent = pte_swp_clear_soft_dirty(ptent);
1122 set_pte_at(vma->vm_mm, addr, pte, ptent);
1123 }
1124 }
1125 #else
clear_soft_dirty(struct vm_area_struct * vma,unsigned long addr,pte_t * pte)1126 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1127 unsigned long addr, pte_t *pte)
1128 {
1129 }
1130 #endif
1131
1132 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
clear_soft_dirty_pmd(struct vm_area_struct * vma,unsigned long addr,pmd_t * pmdp)1133 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1134 unsigned long addr, pmd_t *pmdp)
1135 {
1136 pmd_t old, pmd = *pmdp;
1137
1138 if (pmd_present(pmd)) {
1139 /* See comment in change_huge_pmd() */
1140 old = pmdp_invalidate(vma, addr, pmdp);
1141 if (pmd_dirty(old))
1142 pmd = pmd_mkdirty(pmd);
1143 if (pmd_young(old))
1144 pmd = pmd_mkyoung(pmd);
1145
1146 pmd = pmd_wrprotect(pmd);
1147 pmd = pmd_clear_soft_dirty(pmd);
1148
1149 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1150 } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1151 pmd = pmd_swp_clear_soft_dirty(pmd);
1152 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1153 }
1154 }
1155 #else
clear_soft_dirty_pmd(struct vm_area_struct * vma,unsigned long addr,pmd_t * pmdp)1156 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1157 unsigned long addr, pmd_t *pmdp)
1158 {
1159 }
1160 #endif
1161
clear_refs_pte_range(pmd_t * pmd,unsigned long addr,unsigned long end,struct mm_walk * walk)1162 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
1163 unsigned long end, struct mm_walk *walk)
1164 {
1165 struct clear_refs_private *cp = walk->private;
1166 struct vm_area_struct *vma = walk->vma;
1167 pte_t *pte, ptent;
1168 spinlock_t *ptl;
1169 struct page *page;
1170
1171 ptl = pmd_trans_huge_lock(pmd, vma);
1172 if (ptl) {
1173 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1174 clear_soft_dirty_pmd(vma, addr, pmd);
1175 goto out;
1176 }
1177
1178 if (!pmd_present(*pmd))
1179 goto out;
1180
1181 page = pmd_page(*pmd);
1182
1183 /* Clear accessed and referenced bits. */
1184 pmdp_test_and_clear_young(vma, addr, pmd);
1185 test_and_clear_page_young(page);
1186 ClearPageReferenced(page);
1187 out:
1188 spin_unlock(ptl);
1189 return 0;
1190 }
1191
1192 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1193 if (!pte) {
1194 walk->action = ACTION_AGAIN;
1195 return 0;
1196 }
1197 for (; addr != end; pte++, addr += PAGE_SIZE) {
1198 ptent = ptep_get(pte);
1199
1200 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1201 clear_soft_dirty(vma, addr, pte);
1202 continue;
1203 }
1204
1205 if (!pte_present(ptent))
1206 continue;
1207
1208 page = vm_normal_page(vma, addr, ptent);
1209 if (!page)
1210 continue;
1211
1212 /* Clear accessed and referenced bits. */
1213 ptep_test_and_clear_young(vma, addr, pte);
1214 test_and_clear_page_young(page);
1215 ClearPageReferenced(page);
1216 }
1217 pte_unmap_unlock(pte - 1, ptl);
1218 cond_resched();
1219 return 0;
1220 }
1221
clear_refs_test_walk(unsigned long start,unsigned long end,struct mm_walk * walk)1222 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1223 struct mm_walk *walk)
1224 {
1225 struct clear_refs_private *cp = walk->private;
1226 struct vm_area_struct *vma = walk->vma;
1227
1228 if (vma->vm_flags & VM_PFNMAP)
1229 return 1;
1230
1231 /*
1232 * Writing 1 to /proc/pid/clear_refs affects all pages.
1233 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1234 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1235 * Writing 4 to /proc/pid/clear_refs affects all pages.
1236 */
1237 if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1238 return 1;
1239 if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1240 return 1;
1241 return 0;
1242 }
1243
1244 static const struct mm_walk_ops clear_refs_walk_ops = {
1245 .pmd_entry = clear_refs_pte_range,
1246 .test_walk = clear_refs_test_walk,
1247 .walk_lock = PGWALK_WRLOCK,
1248 };
1249
clear_refs_write(struct file * file,const char __user * buf,size_t count,loff_t * ppos)1250 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1251 size_t count, loff_t *ppos)
1252 {
1253 struct task_struct *task;
1254 char buffer[PROC_NUMBUF];
1255 struct mm_struct *mm;
1256 struct vm_area_struct *vma;
1257 enum clear_refs_types type;
1258 int itype;
1259 int rv;
1260
1261 memset(buffer, 0, sizeof(buffer));
1262 if (count > sizeof(buffer) - 1)
1263 count = sizeof(buffer) - 1;
1264 if (copy_from_user(buffer, buf, count))
1265 return -EFAULT;
1266 rv = kstrtoint(strstrip(buffer), 10, &itype);
1267 if (rv < 0)
1268 return rv;
1269 type = (enum clear_refs_types)itype;
1270 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1271 return -EINVAL;
1272
1273 task = get_proc_task(file_inode(file));
1274 if (!task)
1275 return -ESRCH;
1276 mm = get_task_mm(task);
1277 if (mm) {
1278 VMA_ITERATOR(vmi, mm, 0);
1279 struct mmu_notifier_range range;
1280 struct clear_refs_private cp = {
1281 .type = type,
1282 };
1283
1284 if (mmap_write_lock_killable(mm)) {
1285 count = -EINTR;
1286 goto out_mm;
1287 }
1288 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1289 /*
1290 * Writing 5 to /proc/pid/clear_refs resets the peak
1291 * resident set size to this mm's current rss value.
1292 */
1293 reset_mm_hiwater_rss(mm);
1294 goto out_unlock;
1295 }
1296
1297 if (type == CLEAR_REFS_SOFT_DIRTY) {
1298 for_each_vma(vmi, vma) {
1299 if (!(vma->vm_flags & VM_SOFTDIRTY))
1300 continue;
1301 vm_flags_clear(vma, VM_SOFTDIRTY);
1302 vma_set_page_prot(vma);
1303 }
1304
1305 inc_tlb_flush_pending(mm);
1306 mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY,
1307 0, mm, 0, -1UL);
1308 mmu_notifier_invalidate_range_start(&range);
1309 }
1310 walk_page_range(mm, 0, -1, &clear_refs_walk_ops, &cp);
1311 if (type == CLEAR_REFS_SOFT_DIRTY) {
1312 mmu_notifier_invalidate_range_end(&range);
1313 flush_tlb_mm(mm);
1314 dec_tlb_flush_pending(mm);
1315 }
1316 out_unlock:
1317 mmap_write_unlock(mm);
1318 out_mm:
1319 mmput(mm);
1320 }
1321 put_task_struct(task);
1322
1323 return count;
1324 }
1325
1326 const struct file_operations proc_clear_refs_operations = {
1327 .write = clear_refs_write,
1328 .llseek = noop_llseek,
1329 };
1330
1331 typedef struct {
1332 u64 pme;
1333 } pagemap_entry_t;
1334
1335 struct pagemapread {
1336 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */
1337 pagemap_entry_t *buffer;
1338 bool show_pfn;
1339 };
1340
1341 #define PAGEMAP_WALK_SIZE (PMD_SIZE)
1342 #define PAGEMAP_WALK_MASK (PMD_MASK)
1343
1344 #define PM_ENTRY_BYTES sizeof(pagemap_entry_t)
1345 #define PM_PFRAME_BITS 55
1346 #define PM_PFRAME_MASK GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1347 #define PM_SOFT_DIRTY BIT_ULL(55)
1348 #define PM_MMAP_EXCLUSIVE BIT_ULL(56)
1349 #define PM_UFFD_WP BIT_ULL(57)
1350 #define PM_FILE BIT_ULL(61)
1351 #define PM_SWAP BIT_ULL(62)
1352 #define PM_PRESENT BIT_ULL(63)
1353
1354 #define PM_END_OF_BUFFER 1
1355
make_pme(u64 frame,u64 flags)1356 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1357 {
1358 return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1359 }
1360
add_to_pagemap(unsigned long addr,pagemap_entry_t * pme,struct pagemapread * pm)1361 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1362 struct pagemapread *pm)
1363 {
1364 pm->buffer[pm->pos++] = *pme;
1365 if (pm->pos >= pm->len)
1366 return PM_END_OF_BUFFER;
1367 return 0;
1368 }
1369
pagemap_pte_hole(unsigned long start,unsigned long end,__always_unused int depth,struct mm_walk * walk)1370 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1371 __always_unused int depth, struct mm_walk *walk)
1372 {
1373 struct pagemapread *pm = walk->private;
1374 unsigned long addr = start;
1375 int err = 0;
1376
1377 while (addr < end) {
1378 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1379 pagemap_entry_t pme = make_pme(0, 0);
1380 /* End of address space hole, which we mark as non-present. */
1381 unsigned long hole_end;
1382
1383 if (vma)
1384 hole_end = min(end, vma->vm_start);
1385 else
1386 hole_end = end;
1387
1388 for (; addr < hole_end; addr += PAGE_SIZE) {
1389 err = add_to_pagemap(addr, &pme, pm);
1390 if (err)
1391 goto out;
1392 }
1393
1394 if (!vma)
1395 break;
1396
1397 /* Addresses in the VMA. */
1398 if (vma->vm_flags & VM_SOFTDIRTY)
1399 pme = make_pme(0, PM_SOFT_DIRTY);
1400 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1401 err = add_to_pagemap(addr, &pme, pm);
1402 if (err)
1403 goto out;
1404 }
1405 }
1406 out:
1407 return err;
1408 }
1409
pte_to_pagemap_entry(struct pagemapread * pm,struct vm_area_struct * vma,unsigned long addr,pte_t pte)1410 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1411 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1412 {
1413 u64 frame = 0, flags = 0;
1414 struct page *page = NULL;
1415 bool migration = false;
1416
1417 if (pte_present(pte)) {
1418 if (pm->show_pfn)
1419 frame = pte_pfn(pte);
1420 flags |= PM_PRESENT;
1421 page = vm_normal_page(vma, addr, pte);
1422 if (pte_soft_dirty(pte))
1423 flags |= PM_SOFT_DIRTY;
1424 if (pte_uffd_wp(pte))
1425 flags |= PM_UFFD_WP;
1426 } else if (is_swap_pte(pte)) {
1427 swp_entry_t entry;
1428 if (pte_swp_soft_dirty(pte))
1429 flags |= PM_SOFT_DIRTY;
1430 if (pte_swp_uffd_wp(pte))
1431 flags |= PM_UFFD_WP;
1432 entry = pte_to_swp_entry(pte);
1433 if (pm->show_pfn) {
1434 pgoff_t offset;
1435 /*
1436 * For PFN swap offsets, keeping the offset field
1437 * to be PFN only to be compatible with old smaps.
1438 */
1439 if (is_pfn_swap_entry(entry))
1440 offset = swp_offset_pfn(entry);
1441 else
1442 offset = swp_offset(entry);
1443 frame = swp_type(entry) |
1444 (offset << MAX_SWAPFILES_SHIFT);
1445 }
1446 flags |= PM_SWAP;
1447 migration = is_migration_entry(entry);
1448 if (is_pfn_swap_entry(entry))
1449 page = pfn_swap_entry_to_page(entry);
1450 if (pte_marker_entry_uffd_wp(entry))
1451 flags |= PM_UFFD_WP;
1452 }
1453
1454 if (page && !PageAnon(page))
1455 flags |= PM_FILE;
1456 if (page && !migration && page_mapcount(page) == 1)
1457 flags |= PM_MMAP_EXCLUSIVE;
1458 if (vma->vm_flags & VM_SOFTDIRTY)
1459 flags |= PM_SOFT_DIRTY;
1460
1461 return make_pme(frame, flags);
1462 }
1463
pagemap_pmd_range(pmd_t * pmdp,unsigned long addr,unsigned long end,struct mm_walk * walk)1464 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1465 struct mm_walk *walk)
1466 {
1467 struct vm_area_struct *vma = walk->vma;
1468 struct pagemapread *pm = walk->private;
1469 spinlock_t *ptl;
1470 pte_t *pte, *orig_pte;
1471 int err = 0;
1472 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1473 bool migration = false;
1474
1475 ptl = pmd_trans_huge_lock(pmdp, vma);
1476 if (ptl) {
1477 u64 flags = 0, frame = 0;
1478 pmd_t pmd = *pmdp;
1479 struct page *page = NULL;
1480
1481 if (vma->vm_flags & VM_SOFTDIRTY)
1482 flags |= PM_SOFT_DIRTY;
1483
1484 if (pmd_present(pmd)) {
1485 page = pmd_page(pmd);
1486
1487 flags |= PM_PRESENT;
1488 if (pmd_soft_dirty(pmd))
1489 flags |= PM_SOFT_DIRTY;
1490 if (pmd_uffd_wp(pmd))
1491 flags |= PM_UFFD_WP;
1492 if (pm->show_pfn)
1493 frame = pmd_pfn(pmd) +
1494 ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1495 }
1496 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1497 else if (is_swap_pmd(pmd)) {
1498 swp_entry_t entry = pmd_to_swp_entry(pmd);
1499 unsigned long offset;
1500
1501 if (pm->show_pfn) {
1502 if (is_pfn_swap_entry(entry))
1503 offset = swp_offset_pfn(entry);
1504 else
1505 offset = swp_offset(entry);
1506 offset = offset +
1507 ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1508 frame = swp_type(entry) |
1509 (offset << MAX_SWAPFILES_SHIFT);
1510 }
1511 flags |= PM_SWAP;
1512 if (pmd_swp_soft_dirty(pmd))
1513 flags |= PM_SOFT_DIRTY;
1514 if (pmd_swp_uffd_wp(pmd))
1515 flags |= PM_UFFD_WP;
1516 VM_BUG_ON(!is_pmd_migration_entry(pmd));
1517 migration = is_migration_entry(entry);
1518 page = pfn_swap_entry_to_page(entry);
1519 }
1520 #endif
1521
1522 if (page && !migration && page_mapcount(page) == 1)
1523 flags |= PM_MMAP_EXCLUSIVE;
1524
1525 for (; addr != end; addr += PAGE_SIZE) {
1526 pagemap_entry_t pme = make_pme(frame, flags);
1527
1528 err = add_to_pagemap(addr, &pme, pm);
1529 if (err)
1530 break;
1531 if (pm->show_pfn) {
1532 if (flags & PM_PRESENT)
1533 frame++;
1534 else if (flags & PM_SWAP)
1535 frame += (1 << MAX_SWAPFILES_SHIFT);
1536 }
1537 }
1538 spin_unlock(ptl);
1539 return err;
1540 }
1541 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1542
1543 /*
1544 * We can assume that @vma always points to a valid one and @end never
1545 * goes beyond vma->vm_end.
1546 */
1547 orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1548 if (!pte) {
1549 walk->action = ACTION_AGAIN;
1550 return err;
1551 }
1552 for (; addr < end; pte++, addr += PAGE_SIZE) {
1553 pagemap_entry_t pme;
1554
1555 pme = pte_to_pagemap_entry(pm, vma, addr, ptep_get(pte));
1556 err = add_to_pagemap(addr, &pme, pm);
1557 if (err)
1558 break;
1559 }
1560 pte_unmap_unlock(orig_pte, ptl);
1561
1562 cond_resched();
1563
1564 return err;
1565 }
1566
1567 #ifdef CONFIG_HUGETLB_PAGE
1568 /* This function walks within one hugetlb entry in the single call */
pagemap_hugetlb_range(pte_t * ptep,unsigned long hmask,unsigned long addr,unsigned long end,struct mm_walk * walk)1569 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1570 unsigned long addr, unsigned long end,
1571 struct mm_walk *walk)
1572 {
1573 struct pagemapread *pm = walk->private;
1574 struct vm_area_struct *vma = walk->vma;
1575 u64 flags = 0, frame = 0;
1576 int err = 0;
1577 pte_t pte;
1578
1579 if (vma->vm_flags & VM_SOFTDIRTY)
1580 flags |= PM_SOFT_DIRTY;
1581
1582 pte = huge_ptep_get(ptep);
1583 if (pte_present(pte)) {
1584 struct page *page = pte_page(pte);
1585
1586 if (!PageAnon(page))
1587 flags |= PM_FILE;
1588
1589 if (page_mapcount(page) == 1)
1590 flags |= PM_MMAP_EXCLUSIVE;
1591
1592 if (huge_pte_uffd_wp(pte))
1593 flags |= PM_UFFD_WP;
1594
1595 flags |= PM_PRESENT;
1596 if (pm->show_pfn)
1597 frame = pte_pfn(pte) +
1598 ((addr & ~hmask) >> PAGE_SHIFT);
1599 } else if (pte_swp_uffd_wp_any(pte)) {
1600 flags |= PM_UFFD_WP;
1601 }
1602
1603 for (; addr != end; addr += PAGE_SIZE) {
1604 pagemap_entry_t pme = make_pme(frame, flags);
1605
1606 err = add_to_pagemap(addr, &pme, pm);
1607 if (err)
1608 return err;
1609 if (pm->show_pfn && (flags & PM_PRESENT))
1610 frame++;
1611 }
1612
1613 cond_resched();
1614
1615 return err;
1616 }
1617 #else
1618 #define pagemap_hugetlb_range NULL
1619 #endif /* HUGETLB_PAGE */
1620
1621 static const struct mm_walk_ops pagemap_ops = {
1622 .pmd_entry = pagemap_pmd_range,
1623 .pte_hole = pagemap_pte_hole,
1624 .hugetlb_entry = pagemap_hugetlb_range,
1625 .walk_lock = PGWALK_RDLOCK,
1626 };
1627
1628 /*
1629 * /proc/pid/pagemap - an array mapping virtual pages to pfns
1630 *
1631 * For each page in the address space, this file contains one 64-bit entry
1632 * consisting of the following:
1633 *
1634 * Bits 0-54 page frame number (PFN) if present
1635 * Bits 0-4 swap type if swapped
1636 * Bits 5-54 swap offset if swapped
1637 * Bit 55 pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst)
1638 * Bit 56 page exclusively mapped
1639 * Bit 57 pte is uffd-wp write-protected
1640 * Bits 58-60 zero
1641 * Bit 61 page is file-page or shared-anon
1642 * Bit 62 page swapped
1643 * Bit 63 page present
1644 *
1645 * If the page is not present but in swap, then the PFN contains an
1646 * encoding of the swap file number and the page's offset into the
1647 * swap. Unmapped pages return a null PFN. This allows determining
1648 * precisely which pages are mapped (or in swap) and comparing mapped
1649 * pages between processes.
1650 *
1651 * Efficient users of this interface will use /proc/pid/maps to
1652 * determine which areas of memory are actually mapped and llseek to
1653 * skip over unmapped regions.
1654 */
pagemap_read(struct file * file,char __user * buf,size_t count,loff_t * ppos)1655 static ssize_t pagemap_read(struct file *file, char __user *buf,
1656 size_t count, loff_t *ppos)
1657 {
1658 struct mm_struct *mm = file->private_data;
1659 struct pagemapread pm;
1660 unsigned long src;
1661 unsigned long svpfn;
1662 unsigned long start_vaddr;
1663 unsigned long end_vaddr;
1664 int ret = 0, copied = 0;
1665
1666 if (!mm || !mmget_not_zero(mm))
1667 goto out;
1668
1669 ret = -EINVAL;
1670 /* file position must be aligned */
1671 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1672 goto out_mm;
1673
1674 ret = 0;
1675 if (!count)
1676 goto out_mm;
1677
1678 /* do not disclose physical addresses: attack vector */
1679 pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1680
1681 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1682 pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL);
1683 ret = -ENOMEM;
1684 if (!pm.buffer)
1685 goto out_mm;
1686
1687 src = *ppos;
1688 svpfn = src / PM_ENTRY_BYTES;
1689 end_vaddr = mm->task_size;
1690
1691 /* watch out for wraparound */
1692 start_vaddr = end_vaddr;
1693 if (svpfn <= (ULONG_MAX >> PAGE_SHIFT)) {
1694 unsigned long end;
1695
1696 ret = mmap_read_lock_killable(mm);
1697 if (ret)
1698 goto out_free;
1699 start_vaddr = untagged_addr_remote(mm, svpfn << PAGE_SHIFT);
1700 mmap_read_unlock(mm);
1701
1702 end = start_vaddr + ((count / PM_ENTRY_BYTES) << PAGE_SHIFT);
1703 if (end >= start_vaddr && end < mm->task_size)
1704 end_vaddr = end;
1705 }
1706
1707 /* Ensure the address is inside the task */
1708 if (start_vaddr > mm->task_size)
1709 start_vaddr = end_vaddr;
1710
1711 ret = 0;
1712 while (count && (start_vaddr < end_vaddr)) {
1713 int len;
1714 unsigned long end;
1715
1716 pm.pos = 0;
1717 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1718 /* overflow ? */
1719 if (end < start_vaddr || end > end_vaddr)
1720 end = end_vaddr;
1721 ret = mmap_read_lock_killable(mm);
1722 if (ret)
1723 goto out_free;
1724 ret = walk_page_range(mm, start_vaddr, end, &pagemap_ops, &pm);
1725 mmap_read_unlock(mm);
1726 start_vaddr = end;
1727
1728 len = min(count, PM_ENTRY_BYTES * pm.pos);
1729 if (copy_to_user(buf, pm.buffer, len)) {
1730 ret = -EFAULT;
1731 goto out_free;
1732 }
1733 copied += len;
1734 buf += len;
1735 count -= len;
1736 }
1737 *ppos += copied;
1738 if (!ret || ret == PM_END_OF_BUFFER)
1739 ret = copied;
1740
1741 out_free:
1742 kfree(pm.buffer);
1743 out_mm:
1744 mmput(mm);
1745 out:
1746 return ret;
1747 }
1748
pagemap_open(struct inode * inode,struct file * file)1749 static int pagemap_open(struct inode *inode, struct file *file)
1750 {
1751 struct mm_struct *mm;
1752
1753 mm = proc_mem_open(inode, PTRACE_MODE_READ);
1754 if (IS_ERR(mm))
1755 return PTR_ERR(mm);
1756 file->private_data = mm;
1757 return 0;
1758 }
1759
pagemap_release(struct inode * inode,struct file * file)1760 static int pagemap_release(struct inode *inode, struct file *file)
1761 {
1762 struct mm_struct *mm = file->private_data;
1763
1764 if (mm)
1765 mmdrop(mm);
1766 return 0;
1767 }
1768
1769 const struct file_operations proc_pagemap_operations = {
1770 .llseek = mem_lseek, /* borrow this */
1771 .read = pagemap_read,
1772 .open = pagemap_open,
1773 .release = pagemap_release,
1774 };
1775 #endif /* CONFIG_PROC_PAGE_MONITOR */
1776
1777 #ifdef CONFIG_NUMA
1778
1779 struct numa_maps {
1780 unsigned long pages;
1781 unsigned long anon;
1782 unsigned long active;
1783 unsigned long writeback;
1784 unsigned long mapcount_max;
1785 unsigned long dirty;
1786 unsigned long swapcache;
1787 unsigned long node[MAX_NUMNODES];
1788 };
1789
1790 struct numa_maps_private {
1791 struct proc_maps_private proc_maps;
1792 struct numa_maps md;
1793 };
1794
gather_stats(struct page * page,struct numa_maps * md,int pte_dirty,unsigned long nr_pages)1795 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1796 unsigned long nr_pages)
1797 {
1798 int count = page_mapcount(page);
1799
1800 md->pages += nr_pages;
1801 if (pte_dirty || PageDirty(page))
1802 md->dirty += nr_pages;
1803
1804 if (PageSwapCache(page))
1805 md->swapcache += nr_pages;
1806
1807 if (PageActive(page) || PageUnevictable(page))
1808 md->active += nr_pages;
1809
1810 if (PageWriteback(page))
1811 md->writeback += nr_pages;
1812
1813 if (PageAnon(page))
1814 md->anon += nr_pages;
1815
1816 if (count > md->mapcount_max)
1817 md->mapcount_max = count;
1818
1819 md->node[page_to_nid(page)] += nr_pages;
1820 }
1821
can_gather_numa_stats(pte_t pte,struct vm_area_struct * vma,unsigned long addr)1822 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1823 unsigned long addr)
1824 {
1825 struct page *page;
1826 int nid;
1827
1828 if (!pte_present(pte))
1829 return NULL;
1830
1831 page = vm_normal_page(vma, addr, pte);
1832 if (!page || is_zone_device_page(page))
1833 return NULL;
1834
1835 if (PageReserved(page))
1836 return NULL;
1837
1838 nid = page_to_nid(page);
1839 if (!node_isset(nid, node_states[N_MEMORY]))
1840 return NULL;
1841
1842 return page;
1843 }
1844
1845 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
can_gather_numa_stats_pmd(pmd_t pmd,struct vm_area_struct * vma,unsigned long addr)1846 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1847 struct vm_area_struct *vma,
1848 unsigned long addr)
1849 {
1850 struct page *page;
1851 int nid;
1852
1853 if (!pmd_present(pmd))
1854 return NULL;
1855
1856 page = vm_normal_page_pmd(vma, addr, pmd);
1857 if (!page)
1858 return NULL;
1859
1860 if (PageReserved(page))
1861 return NULL;
1862
1863 nid = page_to_nid(page);
1864 if (!node_isset(nid, node_states[N_MEMORY]))
1865 return NULL;
1866
1867 return page;
1868 }
1869 #endif
1870
gather_pte_stats(pmd_t * pmd,unsigned long addr,unsigned long end,struct mm_walk * walk)1871 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1872 unsigned long end, struct mm_walk *walk)
1873 {
1874 struct numa_maps *md = walk->private;
1875 struct vm_area_struct *vma = walk->vma;
1876 spinlock_t *ptl;
1877 pte_t *orig_pte;
1878 pte_t *pte;
1879
1880 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1881 ptl = pmd_trans_huge_lock(pmd, vma);
1882 if (ptl) {
1883 struct page *page;
1884
1885 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1886 if (page)
1887 gather_stats(page, md, pmd_dirty(*pmd),
1888 HPAGE_PMD_SIZE/PAGE_SIZE);
1889 spin_unlock(ptl);
1890 return 0;
1891 }
1892 #endif
1893 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1894 if (!pte) {
1895 walk->action = ACTION_AGAIN;
1896 return 0;
1897 }
1898 do {
1899 pte_t ptent = ptep_get(pte);
1900 struct page *page = can_gather_numa_stats(ptent, vma, addr);
1901 if (!page)
1902 continue;
1903 gather_stats(page, md, pte_dirty(ptent), 1);
1904
1905 } while (pte++, addr += PAGE_SIZE, addr != end);
1906 pte_unmap_unlock(orig_pte, ptl);
1907 cond_resched();
1908 return 0;
1909 }
1910 #ifdef CONFIG_HUGETLB_PAGE
gather_hugetlb_stats(pte_t * pte,unsigned long hmask,unsigned long addr,unsigned long end,struct mm_walk * walk)1911 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1912 unsigned long addr, unsigned long end, struct mm_walk *walk)
1913 {
1914 pte_t huge_pte = huge_ptep_get(pte);
1915 struct numa_maps *md;
1916 struct page *page;
1917
1918 if (!pte_present(huge_pte))
1919 return 0;
1920
1921 page = pte_page(huge_pte);
1922
1923 md = walk->private;
1924 gather_stats(page, md, pte_dirty(huge_pte), 1);
1925 return 0;
1926 }
1927
1928 #else
gather_hugetlb_stats(pte_t * pte,unsigned long hmask,unsigned long addr,unsigned long end,struct mm_walk * walk)1929 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1930 unsigned long addr, unsigned long end, struct mm_walk *walk)
1931 {
1932 return 0;
1933 }
1934 #endif
1935
1936 static const struct mm_walk_ops show_numa_ops = {
1937 .hugetlb_entry = gather_hugetlb_stats,
1938 .pmd_entry = gather_pte_stats,
1939 .walk_lock = PGWALK_RDLOCK,
1940 };
1941
1942 /*
1943 * Display pages allocated per node and memory policy via /proc.
1944 */
show_numa_map(struct seq_file * m,void * v)1945 static int show_numa_map(struct seq_file *m, void *v)
1946 {
1947 struct numa_maps_private *numa_priv = m->private;
1948 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1949 struct vm_area_struct *vma = v;
1950 struct numa_maps *md = &numa_priv->md;
1951 struct file *file = vma->vm_file;
1952 struct mm_struct *mm = vma->vm_mm;
1953 struct mempolicy *pol;
1954 char buffer[64];
1955 int nid;
1956
1957 if (!mm)
1958 return 0;
1959
1960 /* Ensure we start with an empty set of numa_maps statistics. */
1961 memset(md, 0, sizeof(*md));
1962
1963 pol = __get_vma_policy(vma, vma->vm_start);
1964 if (pol) {
1965 mpol_to_str(buffer, sizeof(buffer), pol);
1966 mpol_cond_put(pol);
1967 } else {
1968 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1969 }
1970
1971 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1972
1973 if (file) {
1974 seq_puts(m, " file=");
1975 seq_file_path(m, file, "\n\t= ");
1976 } else if (vma_is_initial_heap(vma)) {
1977 seq_puts(m, " heap");
1978 } else if (vma_is_initial_stack(vma)) {
1979 seq_puts(m, " stack");
1980 }
1981
1982 if (is_vm_hugetlb_page(vma))
1983 seq_puts(m, " huge");
1984
1985 /* mmap_lock is held by m_start */
1986 walk_page_vma(vma, &show_numa_ops, md);
1987
1988 if (!md->pages)
1989 goto out;
1990
1991 if (md->anon)
1992 seq_printf(m, " anon=%lu", md->anon);
1993
1994 if (md->dirty)
1995 seq_printf(m, " dirty=%lu", md->dirty);
1996
1997 if (md->pages != md->anon && md->pages != md->dirty)
1998 seq_printf(m, " mapped=%lu", md->pages);
1999
2000 if (md->mapcount_max > 1)
2001 seq_printf(m, " mapmax=%lu", md->mapcount_max);
2002
2003 if (md->swapcache)
2004 seq_printf(m, " swapcache=%lu", md->swapcache);
2005
2006 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
2007 seq_printf(m, " active=%lu", md->active);
2008
2009 if (md->writeback)
2010 seq_printf(m, " writeback=%lu", md->writeback);
2011
2012 for_each_node_state(nid, N_MEMORY)
2013 if (md->node[nid])
2014 seq_printf(m, " N%d=%lu", nid, md->node[nid]);
2015
2016 seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
2017 out:
2018 seq_putc(m, '\n');
2019 return 0;
2020 }
2021
2022 static const struct seq_operations proc_pid_numa_maps_op = {
2023 .start = m_start,
2024 .next = m_next,
2025 .stop = m_stop,
2026 .show = show_numa_map,
2027 };
2028
pid_numa_maps_open(struct inode * inode,struct file * file)2029 static int pid_numa_maps_open(struct inode *inode, struct file *file)
2030 {
2031 return proc_maps_open(inode, file, &proc_pid_numa_maps_op,
2032 sizeof(struct numa_maps_private));
2033 }
2034
2035 const struct file_operations proc_pid_numa_maps_operations = {
2036 .open = pid_numa_maps_open,
2037 .read = seq_read,
2038 .llseek = seq_lseek,
2039 .release = proc_map_release,
2040 };
2041
2042 #endif /* CONFIG_NUMA */
2043