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 #ifdef CONFIG_MEM_PURGEABLE
24 #include <linux/mm_purgeable.h>
25 #endif
26
27 #include <asm/elf.h>
28 #include <asm/tlb.h>
29 #include <asm/tlbflush.h>
30 #include "internal.h"
31 #include <linux/hck/lite_hck_hideaddr.h>
32
33 #define SEQ_PUT_DEC(str, val) \
34 seq_put_decimal_ull_width(m, str, (val) << (PAGE_SHIFT-10), 8)
task_mem(struct seq_file * m,struct mm_struct * mm)35 void task_mem(struct seq_file *m, struct mm_struct *mm)
36 {
37 unsigned long text, lib, swap, anon, file, shmem;
38 unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
39 #ifdef CONFIG_MEM_PURGEABLE
40 unsigned long nr_purg_sum = 0, nr_purg_pin = 0;
41
42 mm_purg_pages_info(mm, &nr_purg_sum, &nr_purg_pin);
43 #endif
44
45 anon = get_mm_counter_sum(mm, MM_ANONPAGES);
46 file = get_mm_counter_sum(mm, MM_FILEPAGES);
47 shmem = get_mm_counter_sum(mm, MM_SHMEMPAGES);
48
49 /*
50 * Note: to minimize their overhead, mm maintains hiwater_vm and
51 * hiwater_rss only when about to *lower* total_vm or rss. Any
52 * collector of these hiwater stats must therefore get total_vm
53 * and rss too, which will usually be the higher. Barriers? not
54 * worth the effort, such snapshots can always be inconsistent.
55 */
56 hiwater_vm = total_vm = mm->total_vm;
57 if (hiwater_vm < mm->hiwater_vm)
58 hiwater_vm = mm->hiwater_vm;
59 hiwater_rss = total_rss = anon + file + shmem;
60 if (hiwater_rss < mm->hiwater_rss)
61 hiwater_rss = mm->hiwater_rss;
62
63 /* split executable areas between text and lib */
64 text = PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK);
65 text = min(text, mm->exec_vm << PAGE_SHIFT);
66 lib = (mm->exec_vm << PAGE_SHIFT) - text;
67
68 swap = get_mm_counter_sum(mm, MM_SWAPENTS);
69 SEQ_PUT_DEC("VmPeak:\t", hiwater_vm);
70 SEQ_PUT_DEC(" kB\nVmSize:\t", total_vm);
71 SEQ_PUT_DEC(" kB\nVmLck:\t", mm->locked_vm);
72 SEQ_PUT_DEC(" kB\nVmPin:\t", atomic64_read(&mm->pinned_vm));
73 SEQ_PUT_DEC(" kB\nVmHWM:\t", hiwater_rss);
74 SEQ_PUT_DEC(" kB\nVmRSS:\t", total_rss);
75 SEQ_PUT_DEC(" kB\nRssAnon:\t", anon);
76 SEQ_PUT_DEC(" kB\nRssFile:\t", file);
77 SEQ_PUT_DEC(" kB\nRssShmem:\t", shmem);
78 SEQ_PUT_DEC(" kB\nVmData:\t", mm->data_vm);
79 SEQ_PUT_DEC(" kB\nVmStk:\t", mm->stack_vm);
80 seq_put_decimal_ull_width(m,
81 " kB\nVmExe:\t", text >> 10, 8);
82 seq_put_decimal_ull_width(m,
83 " kB\nVmLib:\t", lib >> 10, 8);
84 seq_put_decimal_ull_width(m,
85 " kB\nVmPTE:\t", mm_pgtables_bytes(mm) >> 10, 8);
86 SEQ_PUT_DEC(" kB\nVmSwap:\t", swap);
87 #ifdef CONFIG_MEM_PURGEABLE
88 SEQ_PUT_DEC(" kB\nPurgSum:\t", nr_purg_sum);
89 SEQ_PUT_DEC(" kB\nPurgPin:\t", nr_purg_pin);
90 #endif
91 seq_puts(m, " kB\n");
92 hugetlb_report_usage(m, mm);
93 }
94 #undef SEQ_PUT_DEC
95
task_vsize(struct mm_struct * mm)96 unsigned long task_vsize(struct mm_struct *mm)
97 {
98 return PAGE_SIZE * mm->total_vm;
99 }
100
task_statm(struct mm_struct * mm,unsigned long * shared,unsigned long * text,unsigned long * data,unsigned long * resident)101 unsigned long task_statm(struct mm_struct *mm,
102 unsigned long *shared, unsigned long *text,
103 unsigned long *data, unsigned long *resident)
104 {
105 *shared = get_mm_counter_sum(mm, MM_FILEPAGES) +
106 get_mm_counter_sum(mm, MM_SHMEMPAGES);
107 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
108 >> PAGE_SHIFT;
109 *data = mm->data_vm + mm->stack_vm;
110 *resident = *shared + get_mm_counter_sum(mm, MM_ANONPAGES);
111 return mm->total_vm;
112 }
113
114 #ifdef CONFIG_NUMA
115 /*
116 * Save get_task_policy() for show_numa_map().
117 */
hold_task_mempolicy(struct proc_maps_private * priv)118 static void hold_task_mempolicy(struct proc_maps_private *priv)
119 {
120 struct task_struct *task = priv->task;
121
122 task_lock(task);
123 priv->task_mempolicy = get_task_policy(task);
124 mpol_get(priv->task_mempolicy);
125 task_unlock(task);
126 }
release_task_mempolicy(struct proc_maps_private * priv)127 static void release_task_mempolicy(struct proc_maps_private *priv)
128 {
129 mpol_put(priv->task_mempolicy);
130 }
131 #else
hold_task_mempolicy(struct proc_maps_private * priv)132 static void hold_task_mempolicy(struct proc_maps_private *priv)
133 {
134 }
release_task_mempolicy(struct proc_maps_private * priv)135 static void release_task_mempolicy(struct proc_maps_private *priv)
136 {
137 }
138 #endif
139
proc_get_vma(struct proc_maps_private * priv,loff_t * ppos)140 static struct vm_area_struct *proc_get_vma(struct proc_maps_private *priv,
141 loff_t *ppos)
142 {
143 struct vm_area_struct *vma = vma_next(&priv->iter);
144
145 if (vma) {
146 *ppos = vma->vm_start;
147 } else {
148 *ppos = -2UL;
149 vma = get_gate_vma(priv->mm);
150 }
151
152 return vma;
153 }
154
m_start(struct seq_file * m,loff_t * ppos)155 static void *m_start(struct seq_file *m, loff_t *ppos)
156 {
157 struct proc_maps_private *priv = m->private;
158 unsigned long last_addr = *ppos;
159 struct mm_struct *mm;
160
161 /* See m_next(). Zero at the start or after lseek. */
162 if (last_addr == -1UL)
163 return NULL;
164
165 priv->task = get_proc_task(priv->inode);
166 if (!priv->task)
167 return ERR_PTR(-ESRCH);
168
169 mm = priv->mm;
170 if (!mm || !mmget_not_zero(mm)) {
171 put_task_struct(priv->task);
172 priv->task = NULL;
173 return NULL;
174 }
175
176 if (mmap_read_lock_killable(mm)) {
177 mmput(mm);
178 put_task_struct(priv->task);
179 priv->task = NULL;
180 return ERR_PTR(-EINTR);
181 }
182
183 vma_iter_init(&priv->iter, mm, last_addr);
184 hold_task_mempolicy(priv);
185 if (last_addr == -2UL)
186 return get_gate_vma(mm);
187
188 return proc_get_vma(priv, ppos);
189 }
190
m_next(struct seq_file * m,void * v,loff_t * ppos)191 static void *m_next(struct seq_file *m, void *v, loff_t *ppos)
192 {
193 if (*ppos == -2UL) {
194 *ppos = -1UL;
195 return NULL;
196 }
197 return proc_get_vma(m->private, ppos);
198 }
199
m_stop(struct seq_file * m,void * v)200 static void m_stop(struct seq_file *m, void *v)
201 {
202 struct proc_maps_private *priv = m->private;
203 struct mm_struct *mm = priv->mm;
204
205 if (!priv->task)
206 return;
207
208 release_task_mempolicy(priv);
209 mmap_read_unlock(mm);
210 mmput(mm);
211 put_task_struct(priv->task);
212 priv->task = NULL;
213 }
214
proc_maps_open(struct inode * inode,struct file * file,const struct seq_operations * ops,int psize)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
proc_map_release(struct inode * inode,struct file * file)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
do_maps_open(struct inode * inode,struct file * file,const struct seq_operations * ops)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
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)253 static void show_vma_header_prefix(struct seq_file *m,
254 unsigned long start, unsigned long end,
255 vm_flags_t flags, unsigned long long pgoff,
256 dev_t dev, unsigned long ino)
257 {
258 seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
259 seq_put_hex_ll(m, NULL, start, 8);
260 seq_put_hex_ll(m, "-", end, 8);
261 seq_putc(m, ' ');
262 seq_putc(m, flags & VM_READ ? 'r' : '-');
263 seq_putc(m, flags & VM_WRITE ? 'w' : '-');
264 seq_putc(m, flags & VM_EXEC ? 'x' : '-');
265 seq_putc(m, flags & VM_MAYSHARE ? 's' : 'p');
266 seq_put_hex_ll(m, " ", pgoff, 8);
267 seq_put_hex_ll(m, " ", MAJOR(dev), 2);
268 seq_put_hex_ll(m, ":", MINOR(dev), 2);
269 seq_put_decimal_ull(m, " ", ino);
270 seq_putc(m, ' ');
271 }
272
273 static void
show_map_vma(struct seq_file * m,struct vm_area_struct * vma)274 show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
275 {
276 struct anon_vma_name *anon_name = NULL;
277 struct mm_struct *mm = vma->vm_mm;
278 struct file *file = vma->vm_file;
279 vm_flags_t flags = vma->vm_flags;
280 unsigned long ino = 0;
281 unsigned long long pgoff = 0;
282 unsigned long start, end;
283 dev_t dev = 0;
284 const char *name = NULL;
285
286 if (file) {
287 struct inode *inode = file_inode(vma->vm_file);
288 dev = inode->i_sb->s_dev;
289 ino = inode->i_ino;
290 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
291 }
292
293 start = vma->vm_start;
294 end = vma->vm_end;
295 CALL_HCK_LITE_HOOK(hideaddr_header_prefix_lhck, &start, &end, &flags, m, vma);
296 show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino);
297 if (mm)
298 anon_name = anon_vma_name(vma);
299
300 /*
301 * Print the dentry name for named mappings, and a
302 * special [heap] marker for the heap:
303 */
304 if (file) {
305 seq_pad(m, ' ');
306 /*
307 * If user named this anon shared memory via
308 * prctl(PR_SET_VMA ..., use the provided name.
309 */
310 if (anon_name)
311 seq_printf(m, "[anon_shmem:%s]", anon_name->name);
312 else
313 seq_file_path(m, file, "\n");
314 goto done;
315 }
316
317 if (vma->vm_ops && vma->vm_ops->name) {
318 name = vma->vm_ops->name(vma);
319 if (name)
320 goto done;
321 }
322
323 name = arch_vma_name(vma);
324 if (!name) {
325 if (!mm) {
326 name = "[vdso]";
327 goto done;
328 }
329
330 if (vma_is_initial_heap(vma)) {
331 name = "[heap]";
332 goto done;
333 }
334
335 if (vma_is_initial_stack(vma)) {
336 name = "[stack]";
337 goto done;
338 }
339
340 if (anon_name) {
341 seq_pad(m, ' ');
342 seq_printf(m, "[anon:%s]", anon_name->name);
343 }
344 }
345
346 done:
347 if (name) {
348 seq_pad(m, ' ');
349 seq_puts(m, name);
350 }
351 seq_putc(m, '\n');
352 }
353
show_map(struct seq_file * m,void * v)354 static int show_map(struct seq_file *m, void *v)
355 {
356 show_map_vma(m, v);
357 return 0;
358 }
359
360 static const struct seq_operations proc_pid_maps_op = {
361 .start = m_start,
362 .next = m_next,
363 .stop = m_stop,
364 .show = show_map
365 };
366
pid_maps_open(struct inode * inode,struct file * file)367 static int pid_maps_open(struct inode *inode, struct file *file)
368 {
369 return do_maps_open(inode, file, &proc_pid_maps_op);
370 }
371
372 const struct file_operations proc_pid_maps_operations = {
373 .open = pid_maps_open,
374 .read = seq_read,
375 .llseek = seq_lseek,
376 .release = proc_map_release,
377 };
378
379 /*
380 * Proportional Set Size(PSS): my share of RSS.
381 *
382 * PSS of a process is the count of pages it has in memory, where each
383 * page is divided by the number of processes sharing it. So if a
384 * process has 1000 pages all to itself, and 1000 shared with one other
385 * process, its PSS will be 1500.
386 *
387 * To keep (accumulated) division errors low, we adopt a 64bit
388 * fixed-point pss counter to minimize division errors. So (pss >>
389 * PSS_SHIFT) would be the real byte count.
390 *
391 * A shift of 12 before division means (assuming 4K page size):
392 * - 1M 3-user-pages add up to 8KB errors;
393 * - supports mapcount up to 2^24, or 16M;
394 * - supports PSS up to 2^52 bytes, or 4PB.
395 */
396 #define PSS_SHIFT 12
397
398 #ifdef CONFIG_PROC_PAGE_MONITOR
399 struct mem_size_stats {
400 unsigned long resident;
401 unsigned long shared_clean;
402 unsigned long shared_dirty;
403 unsigned long private_clean;
404 unsigned long private_dirty;
405 unsigned long referenced;
406 unsigned long anonymous;
407 unsigned long lazyfree;
408 unsigned long anonymous_thp;
409 unsigned long shmem_thp;
410 unsigned long file_thp;
411 unsigned long swap;
412 unsigned long shared_hugetlb;
413 unsigned long private_hugetlb;
414 unsigned long ksm;
415 u64 pss;
416 u64 pss_anon;
417 u64 pss_file;
418 u64 pss_shmem;
419 u64 pss_dirty;
420 u64 pss_locked;
421 u64 swap_pss;
422 };
423
smaps_page_accumulate(struct mem_size_stats * mss,struct page * page,unsigned long size,unsigned long pss,bool dirty,bool locked,bool private)424 static void smaps_page_accumulate(struct mem_size_stats *mss,
425 struct page *page, unsigned long size, unsigned long pss,
426 bool dirty, bool locked, bool private)
427 {
428 mss->pss += pss;
429
430 if (PageAnon(page))
431 mss->pss_anon += pss;
432 else if (PageSwapBacked(page))
433 mss->pss_shmem += pss;
434 else
435 mss->pss_file += pss;
436
437 if (locked)
438 mss->pss_locked += pss;
439
440 if (dirty || PageDirty(page)) {
441 mss->pss_dirty += pss;
442 if (private)
443 mss->private_dirty += size;
444 else
445 mss->shared_dirty += size;
446 } else {
447 if (private)
448 mss->private_clean += size;
449 else
450 mss->shared_clean += size;
451 }
452 }
453
smaps_account(struct mem_size_stats * mss,struct page * page,bool compound,bool young,bool dirty,bool locked,bool migration)454 static void smaps_account(struct mem_size_stats *mss, struct page *page,
455 bool compound, bool young, bool dirty, bool locked,
456 bool migration)
457 {
458 int i, nr = compound ? compound_nr(page) : 1;
459 unsigned long size = nr * PAGE_SIZE;
460
461 /*
462 * First accumulate quantities that depend only on |size| and the type
463 * of the compound page.
464 */
465 if (PageAnon(page)) {
466 mss->anonymous += size;
467 if (!PageSwapBacked(page) && !dirty && !PageDirty(page))
468 mss->lazyfree += size;
469 }
470
471 if (PageKsm(page))
472 mss->ksm += size;
473
474 mss->resident += size;
475 /* Accumulate the size in pages that have been accessed. */
476 if (young || page_is_young(page) || PageReferenced(page))
477 mss->referenced += size;
478
479 /*
480 * Then accumulate quantities that may depend on sharing, or that may
481 * differ page-by-page.
482 *
483 * page_count(page) == 1 guarantees the page is mapped exactly once.
484 * If any subpage of the compound page mapped with PTE it would elevate
485 * page_count().
486 *
487 * The page_mapcount() is called to get a snapshot of the mapcount.
488 * Without holding the page lock this snapshot can be slightly wrong as
489 * we cannot always read the mapcount atomically. It is not safe to
490 * call page_mapcount() even with PTL held if the page is not mapped,
491 * especially for migration entries. Treat regular migration entries
492 * as mapcount == 1.
493 */
494 if ((page_count(page) == 1) || migration) {
495 smaps_page_accumulate(mss, page, size, size << PSS_SHIFT, dirty,
496 locked, true);
497 return;
498 }
499 for (i = 0; i < nr; i++, page++) {
500 int mapcount = page_mapcount(page);
501 unsigned long pss = PAGE_SIZE << PSS_SHIFT;
502 if (mapcount >= 2)
503 pss /= mapcount;
504 smaps_page_accumulate(mss, page, PAGE_SIZE, pss, dirty, locked,
505 mapcount < 2);
506 }
507 }
508
509 #ifdef CONFIG_SHMEM
smaps_pte_hole(unsigned long addr,unsigned long end,__always_unused int depth,struct mm_walk * walk)510 static int smaps_pte_hole(unsigned long addr, unsigned long end,
511 __always_unused int depth, struct mm_walk *walk)
512 {
513 struct mem_size_stats *mss = walk->private;
514 struct vm_area_struct *vma = walk->vma;
515
516 mss->swap += shmem_partial_swap_usage(walk->vma->vm_file->f_mapping,
517 linear_page_index(vma, addr),
518 linear_page_index(vma, end));
519
520 return 0;
521 }
522 #else
523 #define smaps_pte_hole NULL
524 #endif /* CONFIG_SHMEM */
525
smaps_pte_hole_lookup(unsigned long addr,struct mm_walk * walk)526 static void smaps_pte_hole_lookup(unsigned long addr, struct mm_walk *walk)
527 {
528 #ifdef CONFIG_SHMEM
529 if (walk->ops->pte_hole) {
530 /* depth is not used */
531 smaps_pte_hole(addr, addr + PAGE_SIZE, 0, walk);
532 }
533 #endif
534 }
535
smaps_pte_entry(pte_t * pte,unsigned long addr,struct mm_walk * walk)536 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
537 struct mm_walk *walk)
538 {
539 struct mem_size_stats *mss = walk->private;
540 struct vm_area_struct *vma = walk->vma;
541 bool locked = !!(vma->vm_flags & VM_LOCKED);
542 struct page *page = NULL;
543 bool migration = false, young = false, dirty = false;
544 pte_t ptent = ptep_get(pte);
545
546 if (pte_present(ptent)) {
547 page = vm_normal_page(vma, addr, ptent);
548 young = pte_young(ptent);
549 dirty = pte_dirty(ptent);
550 } else if (is_swap_pte(ptent)) {
551 swp_entry_t swpent = pte_to_swp_entry(ptent);
552
553 if (!non_swap_entry(swpent)) {
554 int mapcount;
555
556 mss->swap += PAGE_SIZE;
557 mapcount = swp_swapcount(swpent);
558 if (mapcount >= 2) {
559 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
560
561 do_div(pss_delta, mapcount);
562 mss->swap_pss += pss_delta;
563 } else {
564 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
565 }
566 } else if (is_pfn_swap_entry(swpent)) {
567 if (is_migration_entry(swpent))
568 migration = true;
569 page = pfn_swap_entry_to_page(swpent);
570 }
571 } else {
572 smaps_pte_hole_lookup(addr, walk);
573 return;
574 }
575
576 if (!page)
577 return;
578
579 smaps_account(mss, page, false, young, dirty, locked, migration);
580 }
581
582 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
smaps_pmd_entry(pmd_t * pmd,unsigned long addr,struct mm_walk * walk)583 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
584 struct mm_walk *walk)
585 {
586 struct mem_size_stats *mss = walk->private;
587 struct vm_area_struct *vma = walk->vma;
588 bool locked = !!(vma->vm_flags & VM_LOCKED);
589 struct page *page = NULL;
590 bool migration = false;
591
592 if (pmd_present(*pmd)) {
593 page = vm_normal_page_pmd(vma, addr, *pmd);
594 } else if (unlikely(thp_migration_supported() && is_swap_pmd(*pmd))) {
595 swp_entry_t entry = pmd_to_swp_entry(*pmd);
596
597 if (is_migration_entry(entry)) {
598 migration = true;
599 page = pfn_swap_entry_to_page(entry);
600 }
601 }
602 if (IS_ERR_OR_NULL(page))
603 return;
604 if (PageAnon(page))
605 mss->anonymous_thp += HPAGE_PMD_SIZE;
606 else if (PageSwapBacked(page))
607 mss->shmem_thp += HPAGE_PMD_SIZE;
608 else if (is_zone_device_page(page))
609 /* pass */;
610 else
611 mss->file_thp += HPAGE_PMD_SIZE;
612
613 smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd),
614 locked, migration);
615 }
616 #else
smaps_pmd_entry(pmd_t * pmd,unsigned long addr,struct mm_walk * walk)617 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
618 struct mm_walk *walk)
619 {
620 }
621 #endif
622
smaps_pte_range(pmd_t * pmd,unsigned long addr,unsigned long end,struct mm_walk * walk)623 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
624 struct mm_walk *walk)
625 {
626 struct vm_area_struct *vma = walk->vma;
627 pte_t *pte;
628 spinlock_t *ptl;
629
630 ptl = pmd_trans_huge_lock(pmd, vma);
631 if (ptl) {
632 smaps_pmd_entry(pmd, addr, walk);
633 spin_unlock(ptl);
634 goto out;
635 }
636
637 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
638 if (!pte) {
639 walk->action = ACTION_AGAIN;
640 return 0;
641 }
642 for (; addr != end; pte++, addr += PAGE_SIZE)
643 smaps_pte_entry(pte, addr, walk);
644 pte_unmap_unlock(pte - 1, ptl);
645 out:
646 cond_resched();
647 return 0;
648 }
649
show_smap_vma_flags(struct seq_file * m,struct vm_area_struct * vma)650 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
651 {
652 /*
653 * Don't forget to update Documentation/ on changes.
654 */
655 static const char mnemonics[BITS_PER_LONG][2] = {
656 /*
657 * In case if we meet a flag we don't know about.
658 */
659 [0 ... (BITS_PER_LONG-1)] = "??",
660
661 [ilog2(VM_READ)] = "rd",
662 [ilog2(VM_WRITE)] = "wr",
663 [ilog2(VM_EXEC)] = "ex",
664 [ilog2(VM_SHARED)] = "sh",
665 [ilog2(VM_MAYREAD)] = "mr",
666 [ilog2(VM_MAYWRITE)] = "mw",
667 [ilog2(VM_MAYEXEC)] = "me",
668 [ilog2(VM_MAYSHARE)] = "ms",
669 [ilog2(VM_GROWSDOWN)] = "gd",
670 [ilog2(VM_PFNMAP)] = "pf",
671 [ilog2(VM_LOCKED)] = "lo",
672 [ilog2(VM_IO)] = "io",
673 [ilog2(VM_SEQ_READ)] = "sr",
674 [ilog2(VM_RAND_READ)] = "rr",
675 [ilog2(VM_DONTCOPY)] = "dc",
676 [ilog2(VM_DONTEXPAND)] = "de",
677 [ilog2(VM_LOCKONFAULT)] = "lf",
678 [ilog2(VM_ACCOUNT)] = "ac",
679 [ilog2(VM_NORESERVE)] = "nr",
680 [ilog2(VM_HUGETLB)] = "ht",
681 [ilog2(VM_SYNC)] = "sf",
682 [ilog2(VM_ARCH_1)] = "ar",
683 [ilog2(VM_WIPEONFORK)] = "wf",
684 [ilog2(VM_DONTDUMP)] = "dd",
685 #ifdef CONFIG_ARM64_BTI
686 [ilog2(VM_ARM64_BTI)] = "bt",
687 #endif
688 #ifdef CONFIG_MEM_SOFT_DIRTY
689 [ilog2(VM_SOFTDIRTY)] = "sd",
690 #endif
691 [ilog2(VM_MIXEDMAP)] = "mm",
692 [ilog2(VM_HUGEPAGE)] = "hg",
693 [ilog2(VM_NOHUGEPAGE)] = "nh",
694 [ilog2(VM_MERGEABLE)] = "mg",
695 [ilog2(VM_UFFD_MISSING)]= "um",
696 [ilog2(VM_UFFD_WP)] = "uw",
697 #ifdef CONFIG_ARM64_MTE
698 [ilog2(VM_MTE)] = "mt",
699 [ilog2(VM_MTE_ALLOWED)] = "",
700 #endif
701 #ifdef CONFIG_ARCH_HAS_PKEYS
702 /* These come out via ProtectionKey: */
703 [ilog2(VM_PKEY_BIT0)] = "",
704 [ilog2(VM_PKEY_BIT1)] = "",
705 [ilog2(VM_PKEY_BIT2)] = "",
706 [ilog2(VM_PKEY_BIT3)] = "",
707 #if VM_PKEY_BIT4
708 [ilog2(VM_PKEY_BIT4)] = "",
709 #endif
710 #endif /* CONFIG_ARCH_HAS_PKEYS */
711 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_MINOR
712 [ilog2(VM_UFFD_MINOR)] = "ui",
713 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_MINOR */
714 #ifdef CONFIG_X86_USER_SHADOW_STACK
715 [ilog2(VM_SHADOW_STACK)] = "ss",
716 #endif
717 };
718 size_t i;
719
720 seq_puts(m, "VmFlags: ");
721 for (i = 0; i < BITS_PER_LONG; i++) {
722 if (!mnemonics[i][0])
723 continue;
724 if (vma->vm_flags & (1UL << i)) {
725 seq_putc(m, mnemonics[i][0]);
726 seq_putc(m, mnemonics[i][1]);
727 seq_putc(m, ' ');
728 }
729 }
730 seq_putc(m, '\n');
731 }
732
733 #ifdef CONFIG_HUGETLB_PAGE
smaps_hugetlb_range(pte_t * pte,unsigned long hmask,unsigned long addr,unsigned long end,struct mm_walk * walk)734 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
735 unsigned long addr, unsigned long end,
736 struct mm_walk *walk)
737 {
738 struct mem_size_stats *mss = walk->private;
739 struct vm_area_struct *vma = walk->vma;
740 struct page *page = NULL;
741 pte_t ptent = ptep_get(pte);
742
743 if (pte_present(ptent)) {
744 page = vm_normal_page(vma, addr, ptent);
745 } else if (is_swap_pte(ptent)) {
746 swp_entry_t swpent = pte_to_swp_entry(ptent);
747
748 if (is_pfn_swap_entry(swpent))
749 page = pfn_swap_entry_to_page(swpent);
750 }
751 if (page) {
752 if (page_mapcount(page) >= 2 || hugetlb_pmd_shared(pte))
753 mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
754 else
755 mss->private_hugetlb += huge_page_size(hstate_vma(vma));
756 }
757 return 0;
758 }
759 #else
760 #define smaps_hugetlb_range NULL
761 #endif /* HUGETLB_PAGE */
762
763 static const struct mm_walk_ops smaps_walk_ops = {
764 .pmd_entry = smaps_pte_range,
765 .hugetlb_entry = smaps_hugetlb_range,
766 .walk_lock = PGWALK_RDLOCK,
767 };
768
769 static const struct mm_walk_ops smaps_shmem_walk_ops = {
770 .pmd_entry = smaps_pte_range,
771 .hugetlb_entry = smaps_hugetlb_range,
772 .pte_hole = smaps_pte_hole,
773 .walk_lock = PGWALK_RDLOCK,
774 };
775
776 /*
777 * Gather mem stats from @vma with the indicated beginning
778 * address @start, and keep them in @mss.
779 *
780 * Use vm_start of @vma as the beginning address if @start is 0.
781 */
smap_gather_stats(struct vm_area_struct * vma,struct mem_size_stats * mss,unsigned long start)782 static void smap_gather_stats(struct vm_area_struct *vma,
783 struct mem_size_stats *mss, unsigned long start)
784 {
785 const struct mm_walk_ops *ops = &smaps_walk_ops;
786
787 /* Invalid start */
788 if (start >= vma->vm_end)
789 return;
790
791 if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
792 /*
793 * For shared or readonly shmem mappings we know that all
794 * swapped out pages belong to the shmem object, and we can
795 * obtain the swap value much more efficiently. For private
796 * writable mappings, we might have COW pages that are
797 * not affected by the parent swapped out pages of the shmem
798 * object, so we have to distinguish them during the page walk.
799 * Unless we know that the shmem object (or the part mapped by
800 * our VMA) has no swapped out pages at all.
801 */
802 unsigned long shmem_swapped = shmem_swap_usage(vma);
803
804 if (!start && (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
805 !(vma->vm_flags & VM_WRITE))) {
806 mss->swap += shmem_swapped;
807 } else {
808 ops = &smaps_shmem_walk_ops;
809 }
810 }
811
812 /* mmap_lock is held in m_start */
813 if (!start)
814 walk_page_vma(vma, ops, mss);
815 else
816 walk_page_range(vma->vm_mm, start, vma->vm_end, ops, mss);
817 }
818
819 #define SEQ_PUT_DEC(str, val) \
820 seq_put_decimal_ull_width(m, str, (val) >> 10, 8)
821
822 /* Show the contents common for smaps and smaps_rollup */
__show_smap(struct seq_file * m,const struct mem_size_stats * mss,bool rollup_mode)823 static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss,
824 bool rollup_mode)
825 {
826 SEQ_PUT_DEC("Rss: ", mss->resident);
827 SEQ_PUT_DEC(" kB\nPss: ", mss->pss >> PSS_SHIFT);
828 SEQ_PUT_DEC(" kB\nPss_Dirty: ", mss->pss_dirty >> PSS_SHIFT);
829 if (rollup_mode) {
830 /*
831 * These are meaningful only for smaps_rollup, otherwise two of
832 * them are zero, and the other one is the same as Pss.
833 */
834 SEQ_PUT_DEC(" kB\nPss_Anon: ",
835 mss->pss_anon >> PSS_SHIFT);
836 SEQ_PUT_DEC(" kB\nPss_File: ",
837 mss->pss_file >> PSS_SHIFT);
838 SEQ_PUT_DEC(" kB\nPss_Shmem: ",
839 mss->pss_shmem >> PSS_SHIFT);
840 }
841 SEQ_PUT_DEC(" kB\nShared_Clean: ", mss->shared_clean);
842 SEQ_PUT_DEC(" kB\nShared_Dirty: ", mss->shared_dirty);
843 SEQ_PUT_DEC(" kB\nPrivate_Clean: ", mss->private_clean);
844 SEQ_PUT_DEC(" kB\nPrivate_Dirty: ", mss->private_dirty);
845 SEQ_PUT_DEC(" kB\nReferenced: ", mss->referenced);
846 SEQ_PUT_DEC(" kB\nAnonymous: ", mss->anonymous);
847 SEQ_PUT_DEC(" kB\nKSM: ", mss->ksm);
848 SEQ_PUT_DEC(" kB\nLazyFree: ", mss->lazyfree);
849 SEQ_PUT_DEC(" kB\nAnonHugePages: ", mss->anonymous_thp);
850 SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp);
851 SEQ_PUT_DEC(" kB\nFilePmdMapped: ", mss->file_thp);
852 SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb);
853 seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ",
854 mss->private_hugetlb >> 10, 7);
855 SEQ_PUT_DEC(" kB\nSwap: ", mss->swap);
856 SEQ_PUT_DEC(" kB\nSwapPss: ",
857 mss->swap_pss >> PSS_SHIFT);
858 SEQ_PUT_DEC(" kB\nLocked: ",
859 mss->pss_locked >> PSS_SHIFT);
860 seq_puts(m, " kB\n");
861 }
862
show_smap(struct seq_file * m,void * v)863 static int show_smap(struct seq_file *m, void *v)
864 {
865 struct vm_area_struct *vma = v;
866 struct mem_size_stats mss;
867
868 memset(&mss, 0, sizeof(mss));
869
870 smap_gather_stats(vma, &mss, 0);
871
872 show_map_vma(m, vma);
873
874 SEQ_PUT_DEC("Size: ", vma->vm_end - vma->vm_start);
875 SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma));
876 SEQ_PUT_DEC(" kB\nMMUPageSize: ", vma_mmu_pagesize(vma));
877 seq_puts(m, " kB\n");
878
879 __show_smap(m, &mss, false);
880
881 seq_printf(m, "THPeligible: %8u\n",
882 hugepage_vma_check(vma, vma->vm_flags, true, false, true));
883
884 if (arch_pkeys_enabled())
885 seq_printf(m, "ProtectionKey: %8u\n", vma_pkey(vma));
886 show_smap_vma_flags(m, vma);
887
888 return 0;
889 }
890
show_smaps_rollup(struct seq_file * m,void * v)891 static int show_smaps_rollup(struct seq_file *m, void *v)
892 {
893 struct proc_maps_private *priv = m->private;
894 struct mem_size_stats mss;
895 struct mm_struct *mm = priv->mm;
896 struct vm_area_struct *vma;
897 unsigned long vma_start = 0, last_vma_end = 0;
898 int ret = 0;
899 VMA_ITERATOR(vmi, mm, 0);
900
901 priv->task = get_proc_task(priv->inode);
902 if (!priv->task)
903 return -ESRCH;
904
905 if (!mm || !mmget_not_zero(mm)) {
906 ret = -ESRCH;
907 goto out_put_task;
908 }
909
910 memset(&mss, 0, sizeof(mss));
911
912 ret = mmap_read_lock_killable(mm);
913 if (ret)
914 goto out_put_mm;
915
916 hold_task_mempolicy(priv);
917 vma = vma_next(&vmi);
918
919 if (unlikely(!vma))
920 goto empty_set;
921
922 vma_start = vma->vm_start;
923 do {
924 smap_gather_stats(vma, &mss, 0);
925 last_vma_end = vma->vm_end;
926
927 /*
928 * Release mmap_lock temporarily if someone wants to
929 * access it for write request.
930 */
931 if (mmap_lock_is_contended(mm)) {
932 vma_iter_invalidate(&vmi);
933 mmap_read_unlock(mm);
934 ret = mmap_read_lock_killable(mm);
935 if (ret) {
936 release_task_mempolicy(priv);
937 goto out_put_mm;
938 }
939
940 /*
941 * After dropping the lock, there are four cases to
942 * consider. See the following example for explanation.
943 *
944 * +------+------+-----------+
945 * | VMA1 | VMA2 | VMA3 |
946 * +------+------+-----------+
947 * | | | |
948 * 4k 8k 16k 400k
949 *
950 * Suppose we drop the lock after reading VMA2 due to
951 * contention, then we get:
952 *
953 * last_vma_end = 16k
954 *
955 * 1) VMA2 is freed, but VMA3 exists:
956 *
957 * vma_next(vmi) will return VMA3.
958 * In this case, just continue from VMA3.
959 *
960 * 2) VMA2 still exists:
961 *
962 * vma_next(vmi) will return VMA3.
963 * In this case, just continue from VMA3.
964 *
965 * 3) No more VMAs can be found:
966 *
967 * vma_next(vmi) will return NULL.
968 * No more things to do, just break.
969 *
970 * 4) (last_vma_end - 1) is the middle of a vma (VMA'):
971 *
972 * vma_next(vmi) will return VMA' whose range
973 * contains last_vma_end.
974 * Iterate VMA' from last_vma_end.
975 */
976 vma = vma_next(&vmi);
977 /* Case 3 above */
978 if (!vma)
979 break;
980
981 /* Case 1 and 2 above */
982 if (vma->vm_start >= last_vma_end) {
983 smap_gather_stats(vma, &mss, 0);
984 last_vma_end = vma->vm_end;
985 continue;
986 }
987
988 /* Case 4 above */
989 if (vma->vm_end > last_vma_end) {
990 smap_gather_stats(vma, &mss, last_vma_end);
991 last_vma_end = vma->vm_end;
992 }
993 }
994 } for_each_vma(vmi, vma);
995
996 empty_set:
997 show_vma_header_prefix(m, vma_start, last_vma_end, 0, 0, 0, 0);
998 seq_pad(m, ' ');
999 seq_puts(m, "[rollup]\n");
1000
1001 __show_smap(m, &mss, true);
1002
1003 release_task_mempolicy(priv);
1004 mmap_read_unlock(mm);
1005
1006 out_put_mm:
1007 mmput(mm);
1008 out_put_task:
1009 put_task_struct(priv->task);
1010 priv->task = NULL;
1011
1012 return ret;
1013 }
1014 #undef SEQ_PUT_DEC
1015
1016 static const struct seq_operations proc_pid_smaps_op = {
1017 .start = m_start,
1018 .next = m_next,
1019 .stop = m_stop,
1020 .show = show_smap
1021 };
1022
pid_smaps_open(struct inode * inode,struct file * file)1023 static int pid_smaps_open(struct inode *inode, struct file *file)
1024 {
1025 return do_maps_open(inode, file, &proc_pid_smaps_op);
1026 }
1027
smaps_rollup_open(struct inode * inode,struct file * file)1028 static int smaps_rollup_open(struct inode *inode, struct file *file)
1029 {
1030 int ret;
1031 struct proc_maps_private *priv;
1032
1033 priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT);
1034 if (!priv)
1035 return -ENOMEM;
1036
1037 ret = single_open(file, show_smaps_rollup, priv);
1038 if (ret)
1039 goto out_free;
1040
1041 priv->inode = inode;
1042 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
1043 if (IS_ERR(priv->mm)) {
1044 ret = PTR_ERR(priv->mm);
1045
1046 single_release(inode, file);
1047 goto out_free;
1048 }
1049
1050 return 0;
1051
1052 out_free:
1053 kfree(priv);
1054 return ret;
1055 }
1056
smaps_rollup_release(struct inode * inode,struct file * file)1057 static int smaps_rollup_release(struct inode *inode, struct file *file)
1058 {
1059 struct seq_file *seq = file->private_data;
1060 struct proc_maps_private *priv = seq->private;
1061
1062 if (priv->mm)
1063 mmdrop(priv->mm);
1064
1065 kfree(priv);
1066 return single_release(inode, file);
1067 }
1068
1069 const struct file_operations proc_pid_smaps_operations = {
1070 .open = pid_smaps_open,
1071 .read = seq_read,
1072 .llseek = seq_lseek,
1073 .release = proc_map_release,
1074 };
1075
1076 const struct file_operations proc_pid_smaps_rollup_operations = {
1077 .open = smaps_rollup_open,
1078 .read = seq_read,
1079 .llseek = seq_lseek,
1080 .release = smaps_rollup_release,
1081 };
1082
1083 enum clear_refs_types {
1084 CLEAR_REFS_ALL = 1,
1085 CLEAR_REFS_ANON,
1086 CLEAR_REFS_MAPPED,
1087 CLEAR_REFS_SOFT_DIRTY,
1088 CLEAR_REFS_MM_HIWATER_RSS,
1089 CLEAR_REFS_LAST,
1090 };
1091
1092 struct clear_refs_private {
1093 enum clear_refs_types type;
1094 };
1095
1096 #ifdef CONFIG_MEM_SOFT_DIRTY
1097
pte_is_pinned(struct vm_area_struct * vma,unsigned long addr,pte_t pte)1098 static inline bool pte_is_pinned(struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1099 {
1100 struct page *page;
1101
1102 if (!pte_write(pte))
1103 return false;
1104 if (!is_cow_mapping(vma->vm_flags))
1105 return false;
1106 if (likely(!test_bit(MMF_HAS_PINNED, &vma->vm_mm->flags)))
1107 return false;
1108 page = vm_normal_page(vma, addr, pte);
1109 if (!page)
1110 return false;
1111 return page_maybe_dma_pinned(page);
1112 }
1113
clear_soft_dirty(struct vm_area_struct * vma,unsigned long addr,pte_t * pte)1114 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1115 unsigned long addr, pte_t *pte)
1116 {
1117 /*
1118 * The soft-dirty tracker uses #PF-s to catch writes
1119 * to pages, so write-protect the pte as well. See the
1120 * Documentation/admin-guide/mm/soft-dirty.rst for full description
1121 * of how soft-dirty works.
1122 */
1123 pte_t ptent = ptep_get(pte);
1124
1125 if (pte_present(ptent)) {
1126 pte_t old_pte;
1127
1128 if (pte_is_pinned(vma, addr, ptent))
1129 return;
1130 old_pte = ptep_modify_prot_start(vma, addr, pte);
1131 ptent = pte_wrprotect(old_pte);
1132 ptent = pte_clear_soft_dirty(ptent);
1133 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent);
1134 } else if (is_swap_pte(ptent)) {
1135 ptent = pte_swp_clear_soft_dirty(ptent);
1136 set_pte_at(vma->vm_mm, addr, pte, ptent);
1137 }
1138 }
1139 #else
clear_soft_dirty(struct vm_area_struct * vma,unsigned long addr,pte_t * pte)1140 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1141 unsigned long addr, pte_t *pte)
1142 {
1143 }
1144 #endif
1145
1146 #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)1147 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1148 unsigned long addr, pmd_t *pmdp)
1149 {
1150 pmd_t old, pmd = *pmdp;
1151
1152 if (pmd_present(pmd)) {
1153 /* See comment in change_huge_pmd() */
1154 old = pmdp_invalidate(vma, addr, pmdp);
1155 if (pmd_dirty(old))
1156 pmd = pmd_mkdirty(pmd);
1157 if (pmd_young(old))
1158 pmd = pmd_mkyoung(pmd);
1159
1160 pmd = pmd_wrprotect(pmd);
1161 pmd = pmd_clear_soft_dirty(pmd);
1162
1163 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1164 } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1165 pmd = pmd_swp_clear_soft_dirty(pmd);
1166 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1167 }
1168 }
1169 #else
clear_soft_dirty_pmd(struct vm_area_struct * vma,unsigned long addr,pmd_t * pmdp)1170 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1171 unsigned long addr, pmd_t *pmdp)
1172 {
1173 }
1174 #endif
1175
clear_refs_pte_range(pmd_t * pmd,unsigned long addr,unsigned long end,struct mm_walk * walk)1176 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
1177 unsigned long end, struct mm_walk *walk)
1178 {
1179 struct clear_refs_private *cp = walk->private;
1180 struct vm_area_struct *vma = walk->vma;
1181 pte_t *pte, ptent;
1182 spinlock_t *ptl;
1183 struct page *page;
1184
1185 ptl = pmd_trans_huge_lock(pmd, vma);
1186 if (ptl) {
1187 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1188 clear_soft_dirty_pmd(vma, addr, pmd);
1189 goto out;
1190 }
1191
1192 if (!pmd_present(*pmd))
1193 goto out;
1194
1195 page = pmd_page(*pmd);
1196
1197 /* Clear accessed and referenced bits. */
1198 pmdp_test_and_clear_young(vma, addr, pmd);
1199 test_and_clear_page_young(page);
1200 ClearPageReferenced(page);
1201 out:
1202 spin_unlock(ptl);
1203 return 0;
1204 }
1205
1206 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1207 if (!pte) {
1208 walk->action = ACTION_AGAIN;
1209 return 0;
1210 }
1211 for (; addr != end; pte++, addr += PAGE_SIZE) {
1212 ptent = ptep_get(pte);
1213
1214 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1215 clear_soft_dirty(vma, addr, pte);
1216 continue;
1217 }
1218
1219 if (!pte_present(ptent))
1220 continue;
1221
1222 page = vm_normal_page(vma, addr, ptent);
1223 if (!page)
1224 continue;
1225
1226 /* Clear accessed and referenced bits. */
1227 ptep_test_and_clear_young(vma, addr, pte);
1228 test_and_clear_page_young(page);
1229 ClearPageReferenced(page);
1230 }
1231 pte_unmap_unlock(pte - 1, ptl);
1232 cond_resched();
1233 return 0;
1234 }
1235
clear_refs_test_walk(unsigned long start,unsigned long end,struct mm_walk * walk)1236 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1237 struct mm_walk *walk)
1238 {
1239 struct clear_refs_private *cp = walk->private;
1240 struct vm_area_struct *vma = walk->vma;
1241
1242 if (vma->vm_flags & VM_PFNMAP)
1243 return 1;
1244
1245 /*
1246 * Writing 1 to /proc/pid/clear_refs affects all pages.
1247 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1248 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1249 * Writing 4 to /proc/pid/clear_refs affects all pages.
1250 */
1251 if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1252 return 1;
1253 if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1254 return 1;
1255 return 0;
1256 }
1257
1258 static const struct mm_walk_ops clear_refs_walk_ops = {
1259 .pmd_entry = clear_refs_pte_range,
1260 .test_walk = clear_refs_test_walk,
1261 .walk_lock = PGWALK_WRLOCK,
1262 };
1263
clear_refs_write(struct file * file,const char __user * buf,size_t count,loff_t * ppos)1264 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1265 size_t count, loff_t *ppos)
1266 {
1267 struct task_struct *task;
1268 char buffer[PROC_NUMBUF];
1269 struct mm_struct *mm;
1270 struct vm_area_struct *vma;
1271 enum clear_refs_types type;
1272 int itype;
1273 int rv;
1274
1275 memset(buffer, 0, sizeof(buffer));
1276 if (count > sizeof(buffer) - 1)
1277 count = sizeof(buffer) - 1;
1278 if (copy_from_user(buffer, buf, count))
1279 return -EFAULT;
1280 rv = kstrtoint(strstrip(buffer), 10, &itype);
1281 if (rv < 0)
1282 return rv;
1283 type = (enum clear_refs_types)itype;
1284 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1285 return -EINVAL;
1286
1287 task = get_proc_task(file_inode(file));
1288 if (!task)
1289 return -ESRCH;
1290 mm = get_task_mm(task);
1291 if (mm) {
1292 VMA_ITERATOR(vmi, mm, 0);
1293 struct mmu_notifier_range range;
1294 struct clear_refs_private cp = {
1295 .type = type,
1296 };
1297
1298 if (mmap_write_lock_killable(mm)) {
1299 count = -EINTR;
1300 goto out_mm;
1301 }
1302 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1303 /*
1304 * Writing 5 to /proc/pid/clear_refs resets the peak
1305 * resident set size to this mm's current rss value.
1306 */
1307 reset_mm_hiwater_rss(mm);
1308 goto out_unlock;
1309 }
1310
1311 if (type == CLEAR_REFS_SOFT_DIRTY) {
1312 for_each_vma(vmi, vma) {
1313 if (!(vma->vm_flags & VM_SOFTDIRTY))
1314 continue;
1315 vm_flags_clear(vma, VM_SOFTDIRTY);
1316 vma_set_page_prot(vma);
1317 }
1318
1319 inc_tlb_flush_pending(mm);
1320 mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY,
1321 0, mm, 0, -1UL);
1322 mmu_notifier_invalidate_range_start(&range);
1323 }
1324 walk_page_range(mm, 0, -1, &clear_refs_walk_ops, &cp);
1325 if (type == CLEAR_REFS_SOFT_DIRTY) {
1326 mmu_notifier_invalidate_range_end(&range);
1327 flush_tlb_mm(mm);
1328 dec_tlb_flush_pending(mm);
1329 }
1330 out_unlock:
1331 mmap_write_unlock(mm);
1332 out_mm:
1333 mmput(mm);
1334 }
1335 put_task_struct(task);
1336
1337 return count;
1338 }
1339
1340 const struct file_operations proc_clear_refs_operations = {
1341 .write = clear_refs_write,
1342 .llseek = noop_llseek,
1343 };
1344
1345 typedef struct {
1346 u64 pme;
1347 } pagemap_entry_t;
1348
1349 struct pagemapread {
1350 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */
1351 pagemap_entry_t *buffer;
1352 bool show_pfn;
1353 };
1354
1355 #define PAGEMAP_WALK_SIZE (PMD_SIZE)
1356 #define PAGEMAP_WALK_MASK (PMD_MASK)
1357
1358 #define PM_ENTRY_BYTES sizeof(pagemap_entry_t)
1359 #define PM_PFRAME_BITS 55
1360 #define PM_PFRAME_MASK GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1361 #define PM_SOFT_DIRTY BIT_ULL(55)
1362 #define PM_MMAP_EXCLUSIVE BIT_ULL(56)
1363 #define PM_UFFD_WP BIT_ULL(57)
1364 #define PM_FILE BIT_ULL(61)
1365 #define PM_SWAP BIT_ULL(62)
1366 #define PM_PRESENT BIT_ULL(63)
1367
1368 #define PM_END_OF_BUFFER 1
1369
make_pme(u64 frame,u64 flags)1370 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1371 {
1372 return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1373 }
1374
add_to_pagemap(pagemap_entry_t * pme,struct pagemapread * pm)1375 static int add_to_pagemap(pagemap_entry_t *pme, struct pagemapread *pm)
1376 {
1377 pm->buffer[pm->pos++] = *pme;
1378 if (pm->pos >= pm->len)
1379 return PM_END_OF_BUFFER;
1380 return 0;
1381 }
1382
pagemap_pte_hole(unsigned long start,unsigned long end,__always_unused int depth,struct mm_walk * walk)1383 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1384 __always_unused int depth, struct mm_walk *walk)
1385 {
1386 struct pagemapread *pm = walk->private;
1387 unsigned long addr = start;
1388 int err = 0;
1389
1390 while (addr < end) {
1391 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1392 pagemap_entry_t pme = make_pme(0, 0);
1393 /* End of address space hole, which we mark as non-present. */
1394 unsigned long hole_end;
1395
1396 if (vma)
1397 hole_end = min(end, vma->vm_start);
1398 else
1399 hole_end = end;
1400
1401 for (; addr < hole_end; addr += PAGE_SIZE) {
1402 err = add_to_pagemap(&pme, pm);
1403 if (err)
1404 goto out;
1405 }
1406
1407 if (!vma)
1408 break;
1409
1410 /* Addresses in the VMA. */
1411 if (vma->vm_flags & VM_SOFTDIRTY)
1412 pme = make_pme(0, PM_SOFT_DIRTY);
1413 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1414 err = add_to_pagemap(&pme, pm);
1415 if (err)
1416 goto out;
1417 }
1418 }
1419 out:
1420 return err;
1421 }
1422
pte_to_pagemap_entry(struct pagemapread * pm,struct vm_area_struct * vma,unsigned long addr,pte_t pte)1423 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1424 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1425 {
1426 u64 frame = 0, flags = 0;
1427 struct page *page = NULL;
1428
1429 if (pte_present(pte)) {
1430 if (pm->show_pfn)
1431 frame = pte_pfn(pte);
1432 flags |= PM_PRESENT;
1433 page = vm_normal_page(vma, addr, pte);
1434 if (pte_soft_dirty(pte))
1435 flags |= PM_SOFT_DIRTY;
1436 if (pte_uffd_wp(pte))
1437 flags |= PM_UFFD_WP;
1438 } else if (is_swap_pte(pte)) {
1439 swp_entry_t entry;
1440 if (pte_swp_soft_dirty(pte))
1441 flags |= PM_SOFT_DIRTY;
1442 if (pte_swp_uffd_wp(pte))
1443 flags |= PM_UFFD_WP;
1444 entry = pte_to_swp_entry(pte);
1445 if (pm->show_pfn) {
1446 pgoff_t offset;
1447 /*
1448 * For PFN swap offsets, keeping the offset field
1449 * to be PFN only to be compatible with old smaps.
1450 */
1451 if (is_pfn_swap_entry(entry))
1452 offset = swp_offset_pfn(entry);
1453 else
1454 offset = swp_offset(entry);
1455 frame = swp_type(entry) |
1456 (offset << MAX_SWAPFILES_SHIFT);
1457 }
1458 flags |= PM_SWAP;
1459 if (is_pfn_swap_entry(entry))
1460 page = pfn_swap_entry_to_page(entry);
1461 if (pte_marker_entry_uffd_wp(entry))
1462 flags |= PM_UFFD_WP;
1463 }
1464
1465 if (page && !PageAnon(page))
1466 flags |= PM_FILE;
1467 if (page && (flags & PM_PRESENT) && page_mapcount(page) == 1)
1468 flags |= PM_MMAP_EXCLUSIVE;
1469 if (vma->vm_flags & VM_SOFTDIRTY)
1470 flags |= PM_SOFT_DIRTY;
1471
1472 return make_pme(frame, flags);
1473 }
1474
pagemap_pmd_range(pmd_t * pmdp,unsigned long addr,unsigned long end,struct mm_walk * walk)1475 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1476 struct mm_walk *walk)
1477 {
1478 struct vm_area_struct *vma = walk->vma;
1479 struct pagemapread *pm = walk->private;
1480 spinlock_t *ptl;
1481 pte_t *pte, *orig_pte;
1482 int err = 0;
1483 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1484
1485 ptl = pmd_trans_huge_lock(pmdp, vma);
1486 if (ptl) {
1487 unsigned int idx = (addr & ~PMD_MASK) >> PAGE_SHIFT;
1488 u64 flags = 0, frame = 0;
1489 pmd_t pmd = *pmdp;
1490 struct page *page = NULL;
1491
1492 if (vma->vm_flags & VM_SOFTDIRTY)
1493 flags |= PM_SOFT_DIRTY;
1494
1495 if (pmd_present(pmd)) {
1496 page = pmd_page(pmd);
1497
1498 flags |= PM_PRESENT;
1499 if (pmd_soft_dirty(pmd))
1500 flags |= PM_SOFT_DIRTY;
1501 if (pmd_uffd_wp(pmd))
1502 flags |= PM_UFFD_WP;
1503 if (pm->show_pfn)
1504 frame = pmd_pfn(pmd) + idx;
1505 }
1506 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1507 else if (is_swap_pmd(pmd)) {
1508 swp_entry_t entry = pmd_to_swp_entry(pmd);
1509 unsigned long offset;
1510
1511 if (pm->show_pfn) {
1512 if (is_pfn_swap_entry(entry))
1513 offset = swp_offset_pfn(entry) + idx;
1514 else
1515 offset = swp_offset(entry) + idx;
1516 frame = swp_type(entry) |
1517 (offset << MAX_SWAPFILES_SHIFT);
1518 }
1519 flags |= PM_SWAP;
1520 if (pmd_swp_soft_dirty(pmd))
1521 flags |= PM_SOFT_DIRTY;
1522 if (pmd_swp_uffd_wp(pmd))
1523 flags |= PM_UFFD_WP;
1524 VM_BUG_ON(!is_pmd_migration_entry(pmd));
1525 page = pfn_swap_entry_to_page(entry);
1526 }
1527 #endif
1528
1529 if (page && !PageAnon(page))
1530 flags |= PM_FILE;
1531
1532 for (; addr != end; addr += PAGE_SIZE, idx++) {
1533 u64 cur_flags = flags;
1534 pagemap_entry_t pme;
1535
1536 if (page && (flags & PM_PRESENT) &&
1537 page_mapcount(page + idx) == 1)
1538 cur_flags |= PM_MMAP_EXCLUSIVE;
1539
1540 pme = make_pme(frame, cur_flags);
1541 err = add_to_pagemap(&pme, pm);
1542 if (err)
1543 break;
1544 if (pm->show_pfn) {
1545 if (flags & PM_PRESENT)
1546 frame++;
1547 else if (flags & PM_SWAP)
1548 frame += (1 << MAX_SWAPFILES_SHIFT);
1549 }
1550 }
1551 spin_unlock(ptl);
1552 return err;
1553 }
1554 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1555
1556 /*
1557 * We can assume that @vma always points to a valid one and @end never
1558 * goes beyond vma->vm_end.
1559 */
1560 orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1561 if (!pte) {
1562 walk->action = ACTION_AGAIN;
1563 return err;
1564 }
1565 for (; addr < end; pte++, addr += PAGE_SIZE) {
1566 pagemap_entry_t pme;
1567
1568 pme = pte_to_pagemap_entry(pm, vma, addr, ptep_get(pte));
1569 err = add_to_pagemap(&pme, pm);
1570 if (err)
1571 break;
1572 }
1573 pte_unmap_unlock(orig_pte, ptl);
1574
1575 cond_resched();
1576
1577 return err;
1578 }
1579
1580 #ifdef CONFIG_HUGETLB_PAGE
1581 /* 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)1582 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1583 unsigned long addr, unsigned long end,
1584 struct mm_walk *walk)
1585 {
1586 struct pagemapread *pm = walk->private;
1587 struct vm_area_struct *vma = walk->vma;
1588 u64 flags = 0, frame = 0;
1589 int err = 0;
1590 pte_t pte;
1591
1592 if (vma->vm_flags & VM_SOFTDIRTY)
1593 flags |= PM_SOFT_DIRTY;
1594
1595 pte = huge_ptep_get(ptep);
1596 if (pte_present(pte)) {
1597 struct page *page = pte_page(pte);
1598
1599 if (!PageAnon(page))
1600 flags |= PM_FILE;
1601
1602 if (page_mapcount(page) == 1)
1603 flags |= PM_MMAP_EXCLUSIVE;
1604
1605 if (huge_pte_uffd_wp(pte))
1606 flags |= PM_UFFD_WP;
1607
1608 flags |= PM_PRESENT;
1609 if (pm->show_pfn)
1610 frame = pte_pfn(pte) +
1611 ((addr & ~hmask) >> PAGE_SHIFT);
1612 } else if (pte_swp_uffd_wp_any(pte)) {
1613 flags |= PM_UFFD_WP;
1614 }
1615
1616 for (; addr != end; addr += PAGE_SIZE) {
1617 pagemap_entry_t pme = make_pme(frame, flags);
1618
1619 err = add_to_pagemap(&pme, pm);
1620 if (err)
1621 return err;
1622 if (pm->show_pfn && (flags & PM_PRESENT))
1623 frame++;
1624 }
1625
1626 cond_resched();
1627
1628 return err;
1629 }
1630 #else
1631 #define pagemap_hugetlb_range NULL
1632 #endif /* HUGETLB_PAGE */
1633
1634 static const struct mm_walk_ops pagemap_ops = {
1635 .pmd_entry = pagemap_pmd_range,
1636 .pte_hole = pagemap_pte_hole,
1637 .hugetlb_entry = pagemap_hugetlb_range,
1638 .walk_lock = PGWALK_RDLOCK,
1639 };
1640
1641 /*
1642 * /proc/pid/pagemap - an array mapping virtual pages to pfns
1643 *
1644 * For each page in the address space, this file contains one 64-bit entry
1645 * consisting of the following:
1646 *
1647 * Bits 0-54 page frame number (PFN) if present
1648 * Bits 0-4 swap type if swapped
1649 * Bits 5-54 swap offset if swapped
1650 * Bit 55 pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst)
1651 * Bit 56 page exclusively mapped
1652 * Bit 57 pte is uffd-wp write-protected
1653 * Bits 58-60 zero
1654 * Bit 61 page is file-page or shared-anon
1655 * Bit 62 page swapped
1656 * Bit 63 page present
1657 *
1658 * If the page is not present but in swap, then the PFN contains an
1659 * encoding of the swap file number and the page's offset into the
1660 * swap. Unmapped pages return a null PFN. This allows determining
1661 * precisely which pages are mapped (or in swap) and comparing mapped
1662 * pages between processes.
1663 *
1664 * Efficient users of this interface will use /proc/pid/maps to
1665 * determine which areas of memory are actually mapped and llseek to
1666 * skip over unmapped regions.
1667 */
pagemap_read(struct file * file,char __user * buf,size_t count,loff_t * ppos)1668 static ssize_t pagemap_read(struct file *file, char __user *buf,
1669 size_t count, loff_t *ppos)
1670 {
1671 struct mm_struct *mm = file->private_data;
1672 struct pagemapread pm;
1673 unsigned long src;
1674 unsigned long svpfn;
1675 unsigned long start_vaddr;
1676 unsigned long end_vaddr;
1677 int ret = 0, copied = 0;
1678
1679 if (!mm || !mmget_not_zero(mm))
1680 goto out;
1681
1682 ret = -EINVAL;
1683 /* file position must be aligned */
1684 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1685 goto out_mm;
1686
1687 ret = 0;
1688 if (!count)
1689 goto out_mm;
1690
1691 /* do not disclose physical addresses: attack vector */
1692 pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1693
1694 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1695 pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL);
1696 ret = -ENOMEM;
1697 if (!pm.buffer)
1698 goto out_mm;
1699
1700 src = *ppos;
1701 svpfn = src / PM_ENTRY_BYTES;
1702 end_vaddr = mm->task_size;
1703
1704 /* watch out for wraparound */
1705 start_vaddr = end_vaddr;
1706 if (svpfn <= (ULONG_MAX >> PAGE_SHIFT)) {
1707 unsigned long end;
1708
1709 ret = mmap_read_lock_killable(mm);
1710 if (ret)
1711 goto out_free;
1712 start_vaddr = untagged_addr_remote(mm, svpfn << PAGE_SHIFT);
1713 mmap_read_unlock(mm);
1714
1715 end = start_vaddr + ((count / PM_ENTRY_BYTES) << PAGE_SHIFT);
1716 if (end >= start_vaddr && end < mm->task_size)
1717 end_vaddr = end;
1718 }
1719
1720 /* Ensure the address is inside the task */
1721 if (start_vaddr > mm->task_size)
1722 start_vaddr = end_vaddr;
1723
1724 ret = 0;
1725 while (count && (start_vaddr < end_vaddr)) {
1726 int len;
1727 unsigned long end;
1728
1729 pm.pos = 0;
1730 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1731 /* overflow ? */
1732 if (end < start_vaddr || end > end_vaddr)
1733 end = end_vaddr;
1734 ret = mmap_read_lock_killable(mm);
1735 if (ret)
1736 goto out_free;
1737 ret = walk_page_range(mm, start_vaddr, end, &pagemap_ops, &pm);
1738 mmap_read_unlock(mm);
1739 start_vaddr = end;
1740
1741 len = min(count, PM_ENTRY_BYTES * pm.pos);
1742 if (copy_to_user(buf, pm.buffer, len)) {
1743 ret = -EFAULT;
1744 goto out_free;
1745 }
1746 copied += len;
1747 buf += len;
1748 count -= len;
1749 }
1750 *ppos += copied;
1751 if (!ret || ret == PM_END_OF_BUFFER)
1752 ret = copied;
1753
1754 out_free:
1755 kfree(pm.buffer);
1756 out_mm:
1757 mmput(mm);
1758 out:
1759 return ret;
1760 }
1761
pagemap_open(struct inode * inode,struct file * file)1762 static int pagemap_open(struct inode *inode, struct file *file)
1763 {
1764 struct mm_struct *mm;
1765
1766 mm = proc_mem_open(inode, PTRACE_MODE_READ);
1767 if (IS_ERR(mm))
1768 return PTR_ERR(mm);
1769 file->private_data = mm;
1770 return 0;
1771 }
1772
pagemap_release(struct inode * inode,struct file * file)1773 static int pagemap_release(struct inode *inode, struct file *file)
1774 {
1775 struct mm_struct *mm = file->private_data;
1776
1777 if (mm)
1778 mmdrop(mm);
1779 return 0;
1780 }
1781
1782 const struct file_operations proc_pagemap_operations = {
1783 .llseek = mem_lseek, /* borrow this */
1784 .read = pagemap_read,
1785 .open = pagemap_open,
1786 .release = pagemap_release,
1787 };
1788 #endif /* CONFIG_PROC_PAGE_MONITOR */
1789
1790 #ifdef CONFIG_NUMA
1791
1792 struct numa_maps {
1793 unsigned long pages;
1794 unsigned long anon;
1795 unsigned long active;
1796 unsigned long writeback;
1797 unsigned long mapcount_max;
1798 unsigned long dirty;
1799 unsigned long swapcache;
1800 unsigned long node[MAX_NUMNODES];
1801 };
1802
1803 struct numa_maps_private {
1804 struct proc_maps_private proc_maps;
1805 struct numa_maps md;
1806 };
1807
gather_stats(struct page * page,struct numa_maps * md,int pte_dirty,unsigned long nr_pages)1808 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1809 unsigned long nr_pages)
1810 {
1811 int count = page_mapcount(page);
1812
1813 md->pages += nr_pages;
1814 if (pte_dirty || PageDirty(page))
1815 md->dirty += nr_pages;
1816
1817 if (PageSwapCache(page))
1818 md->swapcache += nr_pages;
1819
1820 if (PageActive(page) || PageUnevictable(page))
1821 md->active += nr_pages;
1822
1823 if (PageWriteback(page))
1824 md->writeback += nr_pages;
1825
1826 if (PageAnon(page))
1827 md->anon += nr_pages;
1828
1829 if (count > md->mapcount_max)
1830 md->mapcount_max = count;
1831
1832 md->node[page_to_nid(page)] += nr_pages;
1833 }
1834
can_gather_numa_stats(pte_t pte,struct vm_area_struct * vma,unsigned long addr)1835 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1836 unsigned long addr)
1837 {
1838 struct page *page;
1839 int nid;
1840
1841 if (!pte_present(pte))
1842 return NULL;
1843
1844 page = vm_normal_page(vma, addr, pte);
1845 if (!page || is_zone_device_page(page))
1846 return NULL;
1847
1848 if (PageReserved(page))
1849 return NULL;
1850
1851 nid = page_to_nid(page);
1852 if (!node_isset(nid, node_states[N_MEMORY]))
1853 return NULL;
1854
1855 return page;
1856 }
1857
1858 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
can_gather_numa_stats_pmd(pmd_t pmd,struct vm_area_struct * vma,unsigned long addr)1859 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1860 struct vm_area_struct *vma,
1861 unsigned long addr)
1862 {
1863 struct page *page;
1864 int nid;
1865
1866 if (!pmd_present(pmd))
1867 return NULL;
1868
1869 page = vm_normal_page_pmd(vma, addr, pmd);
1870 if (!page)
1871 return NULL;
1872
1873 if (PageReserved(page))
1874 return NULL;
1875
1876 nid = page_to_nid(page);
1877 if (!node_isset(nid, node_states[N_MEMORY]))
1878 return NULL;
1879
1880 return page;
1881 }
1882 #endif
1883
gather_pte_stats(pmd_t * pmd,unsigned long addr,unsigned long end,struct mm_walk * walk)1884 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1885 unsigned long end, struct mm_walk *walk)
1886 {
1887 struct numa_maps *md = walk->private;
1888 struct vm_area_struct *vma = walk->vma;
1889 spinlock_t *ptl;
1890 pte_t *orig_pte;
1891 pte_t *pte;
1892
1893 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1894 ptl = pmd_trans_huge_lock(pmd, vma);
1895 if (ptl) {
1896 struct page *page;
1897
1898 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1899 if (page)
1900 gather_stats(page, md, pmd_dirty(*pmd),
1901 HPAGE_PMD_SIZE/PAGE_SIZE);
1902 spin_unlock(ptl);
1903 return 0;
1904 }
1905 #endif
1906 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1907 if (!pte) {
1908 walk->action = ACTION_AGAIN;
1909 return 0;
1910 }
1911 do {
1912 pte_t ptent = ptep_get(pte);
1913 struct page *page = can_gather_numa_stats(ptent, vma, addr);
1914 if (!page)
1915 continue;
1916 gather_stats(page, md, pte_dirty(ptent), 1);
1917
1918 } while (pte++, addr += PAGE_SIZE, addr != end);
1919 pte_unmap_unlock(orig_pte, ptl);
1920 cond_resched();
1921 return 0;
1922 }
1923 #ifdef CONFIG_HUGETLB_PAGE
gather_hugetlb_stats(pte_t * pte,unsigned long hmask,unsigned long addr,unsigned long end,struct mm_walk * walk)1924 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1925 unsigned long addr, unsigned long end, struct mm_walk *walk)
1926 {
1927 pte_t huge_pte = huge_ptep_get(pte);
1928 struct numa_maps *md;
1929 struct page *page;
1930
1931 if (!pte_present(huge_pte))
1932 return 0;
1933
1934 page = pte_page(huge_pte);
1935
1936 md = walk->private;
1937 gather_stats(page, md, pte_dirty(huge_pte), 1);
1938 return 0;
1939 }
1940
1941 #else
gather_hugetlb_stats(pte_t * pte,unsigned long hmask,unsigned long addr,unsigned long end,struct mm_walk * walk)1942 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1943 unsigned long addr, unsigned long end, struct mm_walk *walk)
1944 {
1945 return 0;
1946 }
1947 #endif
1948
1949 static const struct mm_walk_ops show_numa_ops = {
1950 .hugetlb_entry = gather_hugetlb_stats,
1951 .pmd_entry = gather_pte_stats,
1952 .walk_lock = PGWALK_RDLOCK,
1953 };
1954
1955 /*
1956 * Display pages allocated per node and memory policy via /proc.
1957 */
show_numa_map(struct seq_file * m,void * v)1958 static int show_numa_map(struct seq_file *m, void *v)
1959 {
1960 struct numa_maps_private *numa_priv = m->private;
1961 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1962 struct vm_area_struct *vma = v;
1963 struct numa_maps *md = &numa_priv->md;
1964 struct file *file = vma->vm_file;
1965 struct mm_struct *mm = vma->vm_mm;
1966 struct mempolicy *pol;
1967 char buffer[64];
1968 int nid;
1969
1970 if (!mm)
1971 return 0;
1972
1973 /* Ensure we start with an empty set of numa_maps statistics. */
1974 memset(md, 0, sizeof(*md));
1975
1976 pol = __get_vma_policy(vma, vma->vm_start);
1977 if (pol) {
1978 mpol_to_str(buffer, sizeof(buffer), pol);
1979 mpol_cond_put(pol);
1980 } else {
1981 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1982 }
1983
1984 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1985
1986 if (file) {
1987 seq_puts(m, " file=");
1988 seq_file_path(m, file, "\n\t= ");
1989 } else if (vma_is_initial_heap(vma)) {
1990 seq_puts(m, " heap");
1991 } else if (vma_is_initial_stack(vma)) {
1992 seq_puts(m, " stack");
1993 }
1994
1995 if (is_vm_hugetlb_page(vma))
1996 seq_puts(m, " huge");
1997
1998 /* mmap_lock is held by m_start */
1999 walk_page_vma(vma, &show_numa_ops, md);
2000
2001 if (!md->pages)
2002 goto out;
2003
2004 if (md->anon)
2005 seq_printf(m, " anon=%lu", md->anon);
2006
2007 if (md->dirty)
2008 seq_printf(m, " dirty=%lu", md->dirty);
2009
2010 if (md->pages != md->anon && md->pages != md->dirty)
2011 seq_printf(m, " mapped=%lu", md->pages);
2012
2013 if (md->mapcount_max > 1)
2014 seq_printf(m, " mapmax=%lu", md->mapcount_max);
2015
2016 if (md->swapcache)
2017 seq_printf(m, " swapcache=%lu", md->swapcache);
2018
2019 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
2020 seq_printf(m, " active=%lu", md->active);
2021
2022 if (md->writeback)
2023 seq_printf(m, " writeback=%lu", md->writeback);
2024
2025 for_each_node_state(nid, N_MEMORY)
2026 if (md->node[nid])
2027 seq_printf(m, " N%d=%lu", nid, md->node[nid]);
2028
2029 seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
2030 out:
2031 seq_putc(m, '\n');
2032 return 0;
2033 }
2034
2035 static const struct seq_operations proc_pid_numa_maps_op = {
2036 .start = m_start,
2037 .next = m_next,
2038 .stop = m_stop,
2039 .show = show_numa_map,
2040 };
2041
pid_numa_maps_open(struct inode * inode,struct file * file)2042 static int pid_numa_maps_open(struct inode *inode, struct file *file)
2043 {
2044 return proc_maps_open(inode, file, &proc_pid_numa_maps_op,
2045 sizeof(struct numa_maps_private));
2046 }
2047
2048 const struct file_operations proc_pid_numa_maps_operations = {
2049 .open = pid_numa_maps_open,
2050 .read = seq_read,
2051 .llseek = seq_lseek,
2052 .release = proc_map_release,
2053 };
2054
2055 #endif /* CONFIG_NUMA */
2056