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