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