1// SPDX-License-Identifier: GPL-2.0
2#include <linux/pagewalk.h>
3#include <linux/mm_inline.h>
4#include <linux/hugetlb.h>
5#include <linux/huge_mm.h>
6#include <linux/mount.h>
7#include <linux/ksm.h>
8#include <linux/seq_file.h>
9#include <linux/highmem.h>
10#include <linux/ptrace.h>
11#include <linux/slab.h>
12#include <linux/pagemap.h>
13#include <linux/mempolicy.h>
14#include <linux/rmap.h>
15#include <linux/swap.h>
16#include <linux/sched/mm.h>
17#include <linux/swapops.h>
18#include <linux/mmu_notifier.h>
19#include <linux/page_idle.h>
20#include <linux/shmem_fs.h>
21#include <linux/uaccess.h>
22#include <linux/pkeys.h>
23#include <linux/minmax.h>
24#include <linux/overflow.h>
25#include <linux/buildid.h>
26
27#include <asm/elf.h>
28#include <asm/tlb.h>
29#include <asm/tlbflush.h>
30#include "internal.h"
31
32#define SENTINEL_VMA_END -1
33#define SENTINEL_VMA_GATE -2
34
35#define SEQ_PUT_DEC(str, val) \
36 seq_put_decimal_ull_width(m, str, (val) << (PAGE_SHIFT-10), 8)
37void task_mem(struct seq_file *m, struct mm_struct *mm)
38{
39 unsigned long text, lib, swap, anon, file, shmem;
40 unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
41
42 anon = get_mm_counter_sum(mm, member: MM_ANONPAGES);
43 file = get_mm_counter_sum(mm, member: MM_FILEPAGES);
44 shmem = get_mm_counter_sum(mm, member: MM_SHMEMPAGES);
45
46 /*
47 * Note: to minimize their overhead, mm maintains hiwater_vm and
48 * hiwater_rss only when about to *lower* total_vm or rss. Any
49 * collector of these hiwater stats must therefore get total_vm
50 * and rss too, which will usually be the higher. Barriers? not
51 * worth the effort, such snapshots can always be inconsistent.
52 */
53 hiwater_vm = total_vm = mm->total_vm;
54 if (hiwater_vm < mm->hiwater_vm)
55 hiwater_vm = mm->hiwater_vm;
56 hiwater_rss = total_rss = anon + file + shmem;
57 if (hiwater_rss < mm->hiwater_rss)
58 hiwater_rss = mm->hiwater_rss;
59
60 /* split executable areas between text and lib */
61 text = PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK);
62 text = min(text, mm->exec_vm << PAGE_SHIFT);
63 lib = (mm->exec_vm << PAGE_SHIFT) - text;
64
65 swap = get_mm_counter_sum(mm, member: MM_SWAPENTS);
66 SEQ_PUT_DEC("VmPeak:\t", hiwater_vm);
67 SEQ_PUT_DEC(" kB\nVmSize:\t", total_vm);
68 SEQ_PUT_DEC(" kB\nVmLck:\t", mm->locked_vm);
69 SEQ_PUT_DEC(" kB\nVmPin:\t", atomic64_read(&mm->pinned_vm));
70 SEQ_PUT_DEC(" kB\nVmHWM:\t", hiwater_rss);
71 SEQ_PUT_DEC(" kB\nVmRSS:\t", total_rss);
72 SEQ_PUT_DEC(" kB\nRssAnon:\t", anon);
73 SEQ_PUT_DEC(" kB\nRssFile:\t", file);
74 SEQ_PUT_DEC(" kB\nRssShmem:\t", shmem);
75 SEQ_PUT_DEC(" kB\nVmData:\t", mm->data_vm);
76 SEQ_PUT_DEC(" kB\nVmStk:\t", mm->stack_vm);
77 seq_put_decimal_ull_width(m,
78 delimiter: " kB\nVmExe:\t", num: text >> 10, width: 8);
79 seq_put_decimal_ull_width(m,
80 delimiter: " kB\nVmLib:\t", num: lib >> 10, width: 8);
81 seq_put_decimal_ull_width(m,
82 delimiter: " kB\nVmPTE:\t", num: mm_pgtables_bytes(mm) >> 10, width: 8);
83 SEQ_PUT_DEC(" kB\nVmSwap:\t", swap);
84 seq_puts(m, s: " kB\n");
85 hugetlb_report_usage(m, mm);
86}
87#undef SEQ_PUT_DEC
88
89unsigned long task_vsize(struct mm_struct *mm)
90{
91 return PAGE_SIZE * mm->total_vm;
92}
93
94unsigned long task_statm(struct mm_struct *mm,
95 unsigned long *shared, unsigned long *text,
96 unsigned long *data, unsigned long *resident)
97{
98 *shared = get_mm_counter_sum(mm, member: MM_FILEPAGES) +
99 get_mm_counter_sum(mm, member: MM_SHMEMPAGES);
100 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
101 >> PAGE_SHIFT;
102 *data = mm->data_vm + mm->stack_vm;
103 *resident = *shared + get_mm_counter_sum(mm, member: MM_ANONPAGES);
104 return mm->total_vm;
105}
106
107#ifdef CONFIG_NUMA
108/*
109 * Save get_task_policy() for show_numa_map().
110 */
111static void hold_task_mempolicy(struct proc_maps_private *priv)
112{
113 struct task_struct *task = priv->task;
114
115 task_lock(p: task);
116 priv->task_mempolicy = get_task_policy(p: task);
117 mpol_get(pol: priv->task_mempolicy);
118 task_unlock(p: task);
119}
120static void release_task_mempolicy(struct proc_maps_private *priv)
121{
122 mpol_put(pol: priv->task_mempolicy);
123}
124#else
125static void hold_task_mempolicy(struct proc_maps_private *priv)
126{
127}
128static void release_task_mempolicy(struct proc_maps_private *priv)
129{
130}
131#endif
132
133#ifdef CONFIG_PER_VMA_LOCK
134
135static void reset_lock_ctx(struct proc_maps_locking_ctx *lock_ctx)
136{
137 lock_ctx->locked_vma = NULL;
138 lock_ctx->mmap_locked = false;
139}
140
141static void unlock_ctx_vma(struct proc_maps_locking_ctx *lock_ctx)
142{
143 if (lock_ctx->locked_vma) {
144 vma_end_read(vma: lock_ctx->locked_vma);
145 lock_ctx->locked_vma = NULL;
146 }
147}
148
149static const struct seq_operations proc_pid_maps_op;
150
151static inline bool lock_vma_range(struct seq_file *m,
152 struct proc_maps_locking_ctx *lock_ctx)
153{
154 /*
155 * smaps and numa_maps perform page table walk, therefore require
156 * mmap_lock but maps can be read with locking just the vma and
157 * walking the vma tree under rcu read protection.
158 */
159 if (m->op != &proc_pid_maps_op) {
160 if (mmap_read_lock_killable(mm: lock_ctx->mm))
161 return false;
162
163 lock_ctx->mmap_locked = true;
164 } else {
165 rcu_read_lock();
166 reset_lock_ctx(lock_ctx);
167 }
168
169 return true;
170}
171
172static inline void unlock_vma_range(struct proc_maps_locking_ctx *lock_ctx)
173{
174 if (lock_ctx->mmap_locked) {
175 mmap_read_unlock(mm: lock_ctx->mm);
176 } else {
177 unlock_ctx_vma(lock_ctx);
178 rcu_read_unlock();
179 }
180}
181
182static struct vm_area_struct *get_next_vma(struct proc_maps_private *priv,
183 loff_t last_pos)
184{
185 struct proc_maps_locking_ctx *lock_ctx = &priv->lock_ctx;
186 struct vm_area_struct *vma;
187
188 if (lock_ctx->mmap_locked)
189 return vma_next(vmi: &priv->iter);
190
191 unlock_ctx_vma(lock_ctx);
192 vma = lock_next_vma(mm: lock_ctx->mm, iter: &priv->iter, address: last_pos);
193 if (!IS_ERR_OR_NULL(ptr: vma))
194 lock_ctx->locked_vma = vma;
195
196 return vma;
197}
198
199static inline bool fallback_to_mmap_lock(struct proc_maps_private *priv,
200 loff_t pos)
201{
202 struct proc_maps_locking_ctx *lock_ctx = &priv->lock_ctx;
203
204 if (lock_ctx->mmap_locked)
205 return false;
206
207 rcu_read_unlock();
208 mmap_read_lock(mm: lock_ctx->mm);
209 /* Reinitialize the iterator after taking mmap_lock */
210 vma_iter_set(vmi: &priv->iter, addr: pos);
211 lock_ctx->mmap_locked = true;
212
213 return true;
214}
215
216#else /* CONFIG_PER_VMA_LOCK */
217
218static inline bool lock_vma_range(struct seq_file *m,
219 struct proc_maps_locking_ctx *lock_ctx)
220{
221 return mmap_read_lock_killable(lock_ctx->mm) == 0;
222}
223
224static inline void unlock_vma_range(struct proc_maps_locking_ctx *lock_ctx)
225{
226 mmap_read_unlock(lock_ctx->mm);
227}
228
229static struct vm_area_struct *get_next_vma(struct proc_maps_private *priv,
230 loff_t last_pos)
231{
232 return vma_next(&priv->iter);
233}
234
235static inline bool fallback_to_mmap_lock(struct proc_maps_private *priv,
236 loff_t pos)
237{
238 return false;
239}
240
241#endif /* CONFIG_PER_VMA_LOCK */
242
243static struct vm_area_struct *proc_get_vma(struct seq_file *m, loff_t *ppos)
244{
245 struct proc_maps_private *priv = m->private;
246 struct vm_area_struct *vma;
247
248retry:
249 vma = get_next_vma(priv, last_pos: *ppos);
250 /* EINTR of EAGAIN is possible */
251 if (IS_ERR(ptr: vma)) {
252 if (PTR_ERR(ptr: vma) == -EAGAIN && fallback_to_mmap_lock(priv, pos: *ppos))
253 goto retry;
254
255 return vma;
256 }
257
258 /* Store previous position to be able to restart if needed */
259 priv->last_pos = *ppos;
260 if (vma) {
261 /*
262 * Track the end of the reported vma to ensure position changes
263 * even if previous vma was merged with the next vma and we
264 * found the extended vma with the same vm_start.
265 */
266 *ppos = vma->vm_end;
267 } else {
268 *ppos = SENTINEL_VMA_GATE;
269 vma = get_gate_vma(mm: priv->lock_ctx.mm);
270 }
271
272 return vma;
273}
274
275static void *m_start(struct seq_file *m, loff_t *ppos)
276{
277 struct proc_maps_private *priv = m->private;
278 struct proc_maps_locking_ctx *lock_ctx;
279 loff_t last_addr = *ppos;
280 struct mm_struct *mm;
281
282 /* See m_next(). Zero at the start or after lseek. */
283 if (last_addr == SENTINEL_VMA_END)
284 return NULL;
285
286 priv->task = get_proc_task(inode: priv->inode);
287 if (!priv->task)
288 return ERR_PTR(error: -ESRCH);
289
290 lock_ctx = &priv->lock_ctx;
291 mm = lock_ctx->mm;
292 if (!mm || !mmget_not_zero(mm)) {
293 put_task_struct(t: priv->task);
294 priv->task = NULL;
295 return NULL;
296 }
297
298 if (!lock_vma_range(m, lock_ctx)) {
299 mmput(mm);
300 put_task_struct(t: priv->task);
301 priv->task = NULL;
302 return ERR_PTR(error: -EINTR);
303 }
304
305 /*
306 * Reset current position if last_addr was set before
307 * and it's not a sentinel.
308 */
309 if (last_addr > 0)
310 *ppos = last_addr = priv->last_pos;
311 vma_iter_init(vmi: &priv->iter, mm, addr: (unsigned long)last_addr);
312 hold_task_mempolicy(priv);
313 if (last_addr == SENTINEL_VMA_GATE)
314 return get_gate_vma(mm);
315
316 return proc_get_vma(m, ppos);
317}
318
319static void *m_next(struct seq_file *m, void *v, loff_t *ppos)
320{
321 if (*ppos == SENTINEL_VMA_GATE) {
322 *ppos = SENTINEL_VMA_END;
323 return NULL;
324 }
325 return proc_get_vma(m, ppos);
326}
327
328static void m_stop(struct seq_file *m, void *v)
329{
330 struct proc_maps_private *priv = m->private;
331 struct mm_struct *mm = priv->lock_ctx.mm;
332
333 if (!priv->task)
334 return;
335
336 release_task_mempolicy(priv);
337 unlock_vma_range(lock_ctx: &priv->lock_ctx);
338 mmput(mm);
339 put_task_struct(t: priv->task);
340 priv->task = NULL;
341}
342
343static int proc_maps_open(struct inode *inode, struct file *file,
344 const struct seq_operations *ops, int psize)
345{
346 struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
347
348 if (!priv)
349 return -ENOMEM;
350
351 priv->inode = inode;
352 priv->lock_ctx.mm = proc_mem_open(inode, PTRACE_MODE_READ);
353 if (IS_ERR(ptr: priv->lock_ctx.mm)) {
354 int err = PTR_ERR(ptr: priv->lock_ctx.mm);
355
356 seq_release_private(inode, file);
357 return err;
358 }
359
360 return 0;
361}
362
363static int proc_map_release(struct inode *inode, struct file *file)
364{
365 struct seq_file *seq = file->private_data;
366 struct proc_maps_private *priv = seq->private;
367
368 if (priv->lock_ctx.mm)
369 mmdrop(mm: priv->lock_ctx.mm);
370
371 return seq_release_private(inode, file);
372}
373
374static int do_maps_open(struct inode *inode, struct file *file,
375 const struct seq_operations *ops)
376{
377 return proc_maps_open(inode, file, ops,
378 psize: sizeof(struct proc_maps_private));
379}
380
381static void get_vma_name(struct vm_area_struct *vma,
382 const struct path **path,
383 const char **name,
384 const char **name_fmt)
385{
386 struct anon_vma_name *anon_name = vma->vm_mm ? anon_vma_name(vma) : NULL;
387
388 *name = NULL;
389 *path = NULL;
390 *name_fmt = NULL;
391
392 /*
393 * Print the dentry name for named mappings, and a
394 * special [heap] marker for the heap:
395 */
396 if (vma->vm_file) {
397 /*
398 * If user named this anon shared memory via
399 * prctl(PR_SET_VMA ..., use the provided name.
400 */
401 if (anon_name) {
402 *name_fmt = "[anon_shmem:%s]";
403 *name = anon_name->name;
404 } else {
405 *path = file_user_path(f: vma->vm_file);
406 }
407 return;
408 }
409
410 if (vma->vm_ops && vma->vm_ops->name) {
411 *name = vma->vm_ops->name(vma);
412 if (*name)
413 return;
414 }
415
416 *name = arch_vma_name(vma);
417 if (*name)
418 return;
419
420 if (!vma->vm_mm) {
421 *name = "[vdso]";
422 return;
423 }
424
425 if (vma_is_initial_heap(vma)) {
426 *name = "[heap]";
427 return;
428 }
429
430 if (vma_is_initial_stack(vma)) {
431 *name = "[stack]";
432 return;
433 }
434
435 if (anon_name) {
436 *name_fmt = "[anon:%s]";
437 *name = anon_name->name;
438 return;
439 }
440}
441
442static void show_vma_header_prefix(struct seq_file *m,
443 unsigned long start, unsigned long end,
444 vm_flags_t flags, unsigned long long pgoff,
445 dev_t dev, unsigned long ino)
446{
447 seq_setwidth(m, size: 25 + sizeof(void *) * 6 - 1);
448 seq_put_hex_ll(m, NULL, v: start, width: 8);
449 seq_put_hex_ll(m, delimiter: "-", v: end, width: 8);
450 seq_putc(m, c: ' ');
451 seq_putc(m, c: flags & VM_READ ? 'r' : '-');
452 seq_putc(m, c: flags & VM_WRITE ? 'w' : '-');
453 seq_putc(m, c: flags & VM_EXEC ? 'x' : '-');
454 seq_putc(m, c: flags & VM_MAYSHARE ? 's' : 'p');
455 seq_put_hex_ll(m, delimiter: " ", v: pgoff, width: 8);
456 seq_put_hex_ll(m, delimiter: " ", MAJOR(dev), width: 2);
457 seq_put_hex_ll(m, delimiter: ":", MINOR(dev), width: 2);
458 seq_put_decimal_ull(m, delimiter: " ", num: ino);
459 seq_putc(m, c: ' ');
460}
461
462static void
463show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
464{
465 const struct path *path;
466 const char *name_fmt, *name;
467 vm_flags_t flags = vma->vm_flags;
468 unsigned long ino = 0;
469 unsigned long long pgoff = 0;
470 unsigned long start, end;
471 dev_t dev = 0;
472
473 if (vma->vm_file) {
474 const struct inode *inode = file_user_inode(f: vma->vm_file);
475
476 dev = inode->i_sb->s_dev;
477 ino = inode->i_ino;
478 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
479 }
480
481 start = vma->vm_start;
482 end = vma->vm_end;
483 show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino);
484
485 get_vma_name(vma, path: &path, name: &name, name_fmt: &name_fmt);
486 if (path) {
487 seq_pad(m, c: ' ');
488 seq_path(m, path, "\n");
489 } else if (name_fmt) {
490 seq_pad(m, c: ' ');
491 seq_printf(m, fmt: name_fmt, name);
492 } else if (name) {
493 seq_pad(m, c: ' ');
494 seq_puts(m, s: name);
495 }
496 seq_putc(m, c: '\n');
497}
498
499static int show_map(struct seq_file *m, void *v)
500{
501 show_map_vma(m, vma: v);
502 return 0;
503}
504
505static const struct seq_operations proc_pid_maps_op = {
506 .start = m_start,
507 .next = m_next,
508 .stop = m_stop,
509 .show = show_map
510};
511
512static int pid_maps_open(struct inode *inode, struct file *file)
513{
514 return do_maps_open(inode, file, ops: &proc_pid_maps_op);
515}
516
517#define PROCMAP_QUERY_VMA_FLAGS ( \
518 PROCMAP_QUERY_VMA_READABLE | \
519 PROCMAP_QUERY_VMA_WRITABLE | \
520 PROCMAP_QUERY_VMA_EXECUTABLE | \
521 PROCMAP_QUERY_VMA_SHARED \
522)
523
524#define PROCMAP_QUERY_VALID_FLAGS_MASK ( \
525 PROCMAP_QUERY_COVERING_OR_NEXT_VMA | \
526 PROCMAP_QUERY_FILE_BACKED_VMA | \
527 PROCMAP_QUERY_VMA_FLAGS \
528)
529
530#ifdef CONFIG_PER_VMA_LOCK
531
532static int query_vma_setup(struct proc_maps_locking_ctx *lock_ctx)
533{
534 reset_lock_ctx(lock_ctx);
535
536 return 0;
537}
538
539static void query_vma_teardown(struct proc_maps_locking_ctx *lock_ctx)
540{
541 if (lock_ctx->mmap_locked) {
542 mmap_read_unlock(mm: lock_ctx->mm);
543 lock_ctx->mmap_locked = false;
544 } else {
545 unlock_ctx_vma(lock_ctx);
546 }
547}
548
549static struct vm_area_struct *query_vma_find_by_addr(struct proc_maps_locking_ctx *lock_ctx,
550 unsigned long addr)
551{
552 struct mm_struct *mm = lock_ctx->mm;
553 struct vm_area_struct *vma;
554 struct vma_iterator vmi;
555
556 if (lock_ctx->mmap_locked)
557 return find_vma(mm, addr);
558
559 /* Unlock previously locked VMA and find the next one under RCU */
560 unlock_ctx_vma(lock_ctx);
561 rcu_read_lock();
562 vma_iter_init(vmi: &vmi, mm, addr);
563 vma = lock_next_vma(mm, iter: &vmi, address: addr);
564 rcu_read_unlock();
565
566 if (!vma)
567 return NULL;
568
569 if (!IS_ERR(ptr: vma)) {
570 lock_ctx->locked_vma = vma;
571 return vma;
572 }
573
574 if (PTR_ERR(ptr: vma) == -EAGAIN) {
575 /* Fallback to mmap_lock on vma->vm_refcnt overflow */
576 mmap_read_lock(mm);
577 vma = find_vma(mm, addr);
578 lock_ctx->mmap_locked = true;
579 }
580
581 return vma;
582}
583
584#else /* CONFIG_PER_VMA_LOCK */
585
586static int query_vma_setup(struct proc_maps_locking_ctx *lock_ctx)
587{
588 return mmap_read_lock_killable(lock_ctx->mm);
589}
590
591static void query_vma_teardown(struct proc_maps_locking_ctx *lock_ctx)
592{
593 mmap_read_unlock(lock_ctx->mm);
594}
595
596static struct vm_area_struct *query_vma_find_by_addr(struct proc_maps_locking_ctx *lock_ctx,
597 unsigned long addr)
598{
599 return find_vma(lock_ctx->mm, addr);
600}
601
602#endif /* CONFIG_PER_VMA_LOCK */
603
604static struct vm_area_struct *query_matching_vma(struct proc_maps_locking_ctx *lock_ctx,
605 unsigned long addr, u32 flags)
606{
607 struct vm_area_struct *vma;
608
609next_vma:
610 vma = query_vma_find_by_addr(lock_ctx, addr);
611 if (IS_ERR(ptr: vma))
612 return vma;
613
614 if (!vma)
615 goto no_vma;
616
617 /* user requested only file-backed VMA, keep iterating */
618 if ((flags & PROCMAP_QUERY_FILE_BACKED_VMA) && !vma->vm_file)
619 goto skip_vma;
620
621 /* VMA permissions should satisfy query flags */
622 if (flags & PROCMAP_QUERY_VMA_FLAGS) {
623 u32 perm = 0;
624
625 if (flags & PROCMAP_QUERY_VMA_READABLE)
626 perm |= VM_READ;
627 if (flags & PROCMAP_QUERY_VMA_WRITABLE)
628 perm |= VM_WRITE;
629 if (flags & PROCMAP_QUERY_VMA_EXECUTABLE)
630 perm |= VM_EXEC;
631 if (flags & PROCMAP_QUERY_VMA_SHARED)
632 perm |= VM_MAYSHARE;
633
634 if ((vma->vm_flags & perm) != perm)
635 goto skip_vma;
636 }
637
638 /* found covering VMA or user is OK with the matching next VMA */
639 if ((flags & PROCMAP_QUERY_COVERING_OR_NEXT_VMA) || vma->vm_start <= addr)
640 return vma;
641
642skip_vma:
643 /*
644 * If the user needs closest matching VMA, keep iterating.
645 */
646 addr = vma->vm_end;
647 if (flags & PROCMAP_QUERY_COVERING_OR_NEXT_VMA)
648 goto next_vma;
649
650no_vma:
651 return ERR_PTR(error: -ENOENT);
652}
653
654static int do_procmap_query(struct mm_struct *mm, void __user *uarg)
655{
656 struct proc_maps_locking_ctx lock_ctx = { .mm = mm };
657 struct procmap_query karg;
658 struct vm_area_struct *vma;
659 const char *name = NULL;
660 char build_id_buf[BUILD_ID_SIZE_MAX], *name_buf = NULL;
661 __u64 usize;
662 int err;
663
664 if (copy_from_user(to: &usize, from: (void __user *)uarg, n: sizeof(usize)))
665 return -EFAULT;
666 /* argument struct can never be that large, reject abuse */
667 if (usize > PAGE_SIZE)
668 return -E2BIG;
669 /* argument struct should have at least query_flags and query_addr fields */
670 if (usize < offsetofend(struct procmap_query, query_addr))
671 return -EINVAL;
672 err = copy_struct_from_user(dst: &karg, ksize: sizeof(karg), src: uarg, usize);
673 if (err)
674 return err;
675
676 /* reject unknown flags */
677 if (karg.query_flags & ~PROCMAP_QUERY_VALID_FLAGS_MASK)
678 return -EINVAL;
679 /* either both buffer address and size are set, or both should be zero */
680 if (!!karg.vma_name_size != !!karg.vma_name_addr)
681 return -EINVAL;
682 if (!!karg.build_id_size != !!karg.build_id_addr)
683 return -EINVAL;
684
685 if (!mm || !mmget_not_zero(mm))
686 return -ESRCH;
687
688 err = query_vma_setup(lock_ctx: &lock_ctx);
689 if (err) {
690 mmput(mm);
691 return err;
692 }
693
694 vma = query_matching_vma(lock_ctx: &lock_ctx, addr: karg.query_addr, flags: karg.query_flags);
695 if (IS_ERR(ptr: vma)) {
696 err = PTR_ERR(ptr: vma);
697 vma = NULL;
698 goto out;
699 }
700
701 karg.vma_start = vma->vm_start;
702 karg.vma_end = vma->vm_end;
703
704 karg.vma_flags = 0;
705 if (vma->vm_flags & VM_READ)
706 karg.vma_flags |= PROCMAP_QUERY_VMA_READABLE;
707 if (vma->vm_flags & VM_WRITE)
708 karg.vma_flags |= PROCMAP_QUERY_VMA_WRITABLE;
709 if (vma->vm_flags & VM_EXEC)
710 karg.vma_flags |= PROCMAP_QUERY_VMA_EXECUTABLE;
711 if (vma->vm_flags & VM_MAYSHARE)
712 karg.vma_flags |= PROCMAP_QUERY_VMA_SHARED;
713
714 karg.vma_page_size = vma_kernel_pagesize(vma);
715
716 if (vma->vm_file) {
717 const struct inode *inode = file_user_inode(f: vma->vm_file);
718
719 karg.vma_offset = ((__u64)vma->vm_pgoff) << PAGE_SHIFT;
720 karg.dev_major = MAJOR(inode->i_sb->s_dev);
721 karg.dev_minor = MINOR(inode->i_sb->s_dev);
722 karg.inode = inode->i_ino;
723 } else {
724 karg.vma_offset = 0;
725 karg.dev_major = 0;
726 karg.dev_minor = 0;
727 karg.inode = 0;
728 }
729
730 if (karg.build_id_size) {
731 __u32 build_id_sz;
732
733 err = build_id_parse(vma, build_id: build_id_buf, size: &build_id_sz);
734 if (err) {
735 karg.build_id_size = 0;
736 } else {
737 if (karg.build_id_size < build_id_sz) {
738 err = -ENAMETOOLONG;
739 goto out;
740 }
741 karg.build_id_size = build_id_sz;
742 }
743 }
744
745 if (karg.vma_name_size) {
746 size_t name_buf_sz = min_t(size_t, PATH_MAX, karg.vma_name_size);
747 const struct path *path;
748 const char *name_fmt;
749 size_t name_sz = 0;
750
751 get_vma_name(vma, path: &path, name: &name, name_fmt: &name_fmt);
752
753 if (path || name_fmt || name) {
754 name_buf = kmalloc(name_buf_sz, GFP_KERNEL);
755 if (!name_buf) {
756 err = -ENOMEM;
757 goto out;
758 }
759 }
760 if (path) {
761 name = d_path(path, name_buf, name_buf_sz);
762 if (IS_ERR(ptr: name)) {
763 err = PTR_ERR(ptr: name);
764 goto out;
765 }
766 name_sz = name_buf + name_buf_sz - name;
767 } else if (name || name_fmt) {
768 name_sz = 1 + snprintf(buf: name_buf, size: name_buf_sz, fmt: name_fmt ?: "%s", name);
769 name = name_buf;
770 }
771 if (name_sz > name_buf_sz) {
772 err = -ENAMETOOLONG;
773 goto out;
774 }
775 karg.vma_name_size = name_sz;
776 }
777
778 /* unlock vma or mmap_lock, and put mm_struct before copying data to user */
779 query_vma_teardown(lock_ctx: &lock_ctx);
780 mmput(mm);
781
782 if (karg.vma_name_size && copy_to_user(u64_to_user_ptr(karg.vma_name_addr),
783 from: name, n: karg.vma_name_size)) {
784 kfree(objp: name_buf);
785 return -EFAULT;
786 }
787 kfree(objp: name_buf);
788
789 if (karg.build_id_size && copy_to_user(u64_to_user_ptr(karg.build_id_addr),
790 from: build_id_buf, n: karg.build_id_size))
791 return -EFAULT;
792
793 if (copy_to_user(to: uarg, from: &karg, min_t(size_t, sizeof(karg), usize)))
794 return -EFAULT;
795
796 return 0;
797
798out:
799 query_vma_teardown(lock_ctx: &lock_ctx);
800 mmput(mm);
801 kfree(objp: name_buf);
802 return err;
803}
804
805static long procfs_procmap_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
806{
807 struct seq_file *seq = file->private_data;
808 struct proc_maps_private *priv = seq->private;
809
810 switch (cmd) {
811 case PROCMAP_QUERY:
812 /* priv->lock_ctx.mm is set during file open operation */
813 return do_procmap_query(mm: priv->lock_ctx.mm, uarg: (void __user *)arg);
814 default:
815 return -ENOIOCTLCMD;
816 }
817}
818
819const struct file_operations proc_pid_maps_operations = {
820 .open = pid_maps_open,
821 .read = seq_read,
822 .llseek = seq_lseek,
823 .release = proc_map_release,
824 .unlocked_ioctl = procfs_procmap_ioctl,
825 .compat_ioctl = compat_ptr_ioctl,
826};
827
828/*
829 * Proportional Set Size(PSS): my share of RSS.
830 *
831 * PSS of a process is the count of pages it has in memory, where each
832 * page is divided by the number of processes sharing it. So if a
833 * process has 1000 pages all to itself, and 1000 shared with one other
834 * process, its PSS will be 1500.
835 *
836 * To keep (accumulated) division errors low, we adopt a 64bit
837 * fixed-point pss counter to minimize division errors. So (pss >>
838 * PSS_SHIFT) would be the real byte count.
839 *
840 * A shift of 12 before division means (assuming 4K page size):
841 * - 1M 3-user-pages add up to 8KB errors;
842 * - supports mapcount up to 2^24, or 16M;
843 * - supports PSS up to 2^52 bytes, or 4PB.
844 */
845#define PSS_SHIFT 12
846
847#ifdef CONFIG_PROC_PAGE_MONITOR
848struct mem_size_stats {
849 unsigned long resident;
850 unsigned long shared_clean;
851 unsigned long shared_dirty;
852 unsigned long private_clean;
853 unsigned long private_dirty;
854 unsigned long referenced;
855 unsigned long anonymous;
856 unsigned long lazyfree;
857 unsigned long anonymous_thp;
858 unsigned long shmem_thp;
859 unsigned long file_thp;
860 unsigned long swap;
861 unsigned long shared_hugetlb;
862 unsigned long private_hugetlb;
863 unsigned long ksm;
864 u64 pss;
865 u64 pss_anon;
866 u64 pss_file;
867 u64 pss_shmem;
868 u64 pss_dirty;
869 u64 pss_locked;
870 u64 swap_pss;
871};
872
873static void smaps_page_accumulate(struct mem_size_stats *mss,
874 struct folio *folio, unsigned long size, unsigned long pss,
875 bool dirty, bool locked, bool private)
876{
877 mss->pss += pss;
878
879 if (folio_test_anon(folio))
880 mss->pss_anon += pss;
881 else if (folio_test_swapbacked(folio))
882 mss->pss_shmem += pss;
883 else
884 mss->pss_file += pss;
885
886 if (locked)
887 mss->pss_locked += pss;
888
889 if (dirty || folio_test_dirty(folio)) {
890 mss->pss_dirty += pss;
891 if (private)
892 mss->private_dirty += size;
893 else
894 mss->shared_dirty += size;
895 } else {
896 if (private)
897 mss->private_clean += size;
898 else
899 mss->shared_clean += size;
900 }
901}
902
903static void smaps_account(struct mem_size_stats *mss, struct page *page,
904 bool compound, bool young, bool dirty, bool locked,
905 bool present)
906{
907 struct folio *folio = page_folio(page);
908 int i, nr = compound ? compound_nr(page) : 1;
909 unsigned long size = nr * PAGE_SIZE;
910 bool exclusive;
911 int mapcount;
912
913 /*
914 * First accumulate quantities that depend only on |size| and the type
915 * of the compound page.
916 */
917 if (folio_test_anon(folio)) {
918 mss->anonymous += size;
919 if (!folio_test_swapbacked(folio) && !dirty &&
920 !folio_test_dirty(folio))
921 mss->lazyfree += size;
922 }
923
924 if (folio_test_ksm(folio))
925 mss->ksm += size;
926
927 mss->resident += size;
928 /* Accumulate the size in pages that have been accessed. */
929 if (young || folio_test_young(folio) || folio_test_referenced(folio))
930 mss->referenced += size;
931
932 /*
933 * Then accumulate quantities that may depend on sharing, or that may
934 * differ page-by-page.
935 *
936 * refcount == 1 for present entries guarantees that the folio is mapped
937 * exactly once. For large folios this implies that exactly one
938 * PTE/PMD/... maps (a part of) this folio.
939 *
940 * Treat all non-present entries (where relying on the mapcount and
941 * refcount doesn't make sense) as "maybe shared, but not sure how
942 * often". We treat device private entries as being fake-present.
943 *
944 * Note that it would not be safe to read the mapcount especially for
945 * pages referenced by migration entries, even with the PTL held.
946 */
947 if (folio_ref_count(folio) == 1 || !present) {
948 smaps_page_accumulate(mss, folio, size, pss: size << PSS_SHIFT,
949 dirty, locked, private: present);
950 return;
951 }
952
953 if (IS_ENABLED(CONFIG_NO_PAGE_MAPCOUNT)) {
954 mapcount = folio_average_page_mapcount(folio);
955 exclusive = !folio_maybe_mapped_shared(folio);
956 }
957
958 /*
959 * We obtain a snapshot of the mapcount. Without holding the folio lock
960 * this snapshot can be slightly wrong as we cannot always read the
961 * mapcount atomically.
962 */
963 for (i = 0; i < nr; i++, page++) {
964 unsigned long pss = PAGE_SIZE << PSS_SHIFT;
965
966 if (IS_ENABLED(CONFIG_PAGE_MAPCOUNT)) {
967 mapcount = folio_precise_page_mapcount(folio, page);
968 exclusive = mapcount < 2;
969 }
970
971 if (mapcount >= 2)
972 pss /= mapcount;
973 smaps_page_accumulate(mss, folio, PAGE_SIZE, pss,
974 dirty, locked, private: exclusive);
975 }
976}
977
978#ifdef CONFIG_SHMEM
979static int smaps_pte_hole(unsigned long addr, unsigned long end,
980 __always_unused int depth, struct mm_walk *walk)
981{
982 struct mem_size_stats *mss = walk->private;
983 struct vm_area_struct *vma = walk->vma;
984
985 mss->swap += shmem_partial_swap_usage(mapping: walk->vma->vm_file->f_mapping,
986 start: linear_page_index(vma, address: addr),
987 end: linear_page_index(vma, address: end));
988
989 return 0;
990}
991#else
992#define smaps_pte_hole NULL
993#endif /* CONFIG_SHMEM */
994
995static void smaps_pte_hole_lookup(unsigned long addr, struct mm_walk *walk)
996{
997#ifdef CONFIG_SHMEM
998 if (walk->ops->pte_hole) {
999 /* depth is not used */
1000 smaps_pte_hole(addr, end: addr + PAGE_SIZE, depth: 0, walk);
1001 }
1002#endif
1003}
1004
1005static void smaps_pte_entry(pte_t *pte, unsigned long addr,
1006 struct mm_walk *walk)
1007{
1008 struct mem_size_stats *mss = walk->private;
1009 struct vm_area_struct *vma = walk->vma;
1010 bool locked = !!(vma->vm_flags & VM_LOCKED);
1011 struct page *page = NULL;
1012 bool present = false, young = false, dirty = false;
1013 pte_t ptent = ptep_get(ptep: pte);
1014
1015 if (pte_present(a: ptent)) {
1016 page = vm_normal_page(vma, addr, pte: ptent);
1017 young = pte_young(pte: ptent);
1018 dirty = pte_dirty(pte: ptent);
1019 present = true;
1020 } else if (is_swap_pte(pte: ptent)) {
1021 swp_entry_t swpent = pte_to_swp_entry(pte: ptent);
1022
1023 if (!non_swap_entry(entry: swpent)) {
1024 int mapcount;
1025
1026 mss->swap += PAGE_SIZE;
1027 mapcount = swp_swapcount(entry: swpent);
1028 if (mapcount >= 2) {
1029 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
1030
1031 do_div(pss_delta, mapcount);
1032 mss->swap_pss += pss_delta;
1033 } else {
1034 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
1035 }
1036 } else if (is_pfn_swap_entry(entry: swpent)) {
1037 if (is_device_private_entry(entry: swpent))
1038 present = true;
1039 page = pfn_swap_entry_to_page(entry: swpent);
1040 }
1041 } else {
1042 smaps_pte_hole_lookup(addr, walk);
1043 return;
1044 }
1045
1046 if (!page)
1047 return;
1048
1049 smaps_account(mss, page, compound: false, young, dirty, locked, present);
1050}
1051
1052#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1053static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
1054 struct mm_walk *walk)
1055{
1056 struct mem_size_stats *mss = walk->private;
1057 struct vm_area_struct *vma = walk->vma;
1058 bool locked = !!(vma->vm_flags & VM_LOCKED);
1059 struct page *page = NULL;
1060 bool present = false;
1061 struct folio *folio;
1062
1063 if (pmd_present(*pmd)) {
1064 page = vm_normal_page_pmd(vma, addr, *pmd);
1065 present = true;
1066 } else if (unlikely(thp_migration_supported() && is_swap_pmd(*pmd))) {
1067 swp_entry_t entry = pmd_to_swp_entry(*pmd);
1068
1069 if (is_pfn_swap_entry(entry))
1070 page = pfn_swap_entry_to_page(entry);
1071 }
1072 if (IS_ERR_OR_NULL(page))
1073 return;
1074 folio = page_folio(page);
1075 if (folio_test_anon(folio))
1076 mss->anonymous_thp += HPAGE_PMD_SIZE;
1077 else if (folio_test_swapbacked(folio))
1078 mss->shmem_thp += HPAGE_PMD_SIZE;
1079 else if (folio_is_zone_device(folio))
1080 /* pass */;
1081 else
1082 mss->file_thp += HPAGE_PMD_SIZE;
1083
1084 smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd),
1085 locked, present);
1086}
1087#else
1088static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
1089 struct mm_walk *walk)
1090{
1091}
1092#endif
1093
1094static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
1095 struct mm_walk *walk)
1096{
1097 struct vm_area_struct *vma = walk->vma;
1098 pte_t *pte;
1099 spinlock_t *ptl;
1100
1101 ptl = pmd_trans_huge_lock(pmd, vma);
1102 if (ptl) {
1103 smaps_pmd_entry(pmd, addr, walk);
1104 spin_unlock(lock: ptl);
1105 goto out;
1106 }
1107
1108 pte = pte_offset_map_lock(mm: vma->vm_mm, pmd, addr, ptlp: &ptl);
1109 if (!pte) {
1110 walk->action = ACTION_AGAIN;
1111 return 0;
1112 }
1113 for (; addr != end; pte++, addr += PAGE_SIZE)
1114 smaps_pte_entry(pte, addr, walk);
1115 pte_unmap_unlock(pte - 1, ptl);
1116out:
1117 cond_resched();
1118 return 0;
1119}
1120
1121static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
1122{
1123 /*
1124 * Don't forget to update Documentation/ on changes.
1125 *
1126 * The length of the second argument of mnemonics[]
1127 * needs to be 3 instead of previously set 2
1128 * (i.e. from [BITS_PER_LONG][2] to [BITS_PER_LONG][3])
1129 * to avoid spurious
1130 * -Werror=unterminated-string-initialization warning
1131 * with GCC 15
1132 */
1133 static const char mnemonics[BITS_PER_LONG][3] = {
1134 /*
1135 * In case if we meet a flag we don't know about.
1136 */
1137 [0 ... (BITS_PER_LONG-1)] = "??",
1138
1139 [ilog2(VM_READ)] = "rd",
1140 [ilog2(VM_WRITE)] = "wr",
1141 [ilog2(VM_EXEC)] = "ex",
1142 [ilog2(VM_SHARED)] = "sh",
1143 [ilog2(VM_MAYREAD)] = "mr",
1144 [ilog2(VM_MAYWRITE)] = "mw",
1145 [ilog2(VM_MAYEXEC)] = "me",
1146 [ilog2(VM_MAYSHARE)] = "ms",
1147 [ilog2(VM_GROWSDOWN)] = "gd",
1148 [ilog2(VM_PFNMAP)] = "pf",
1149 [ilog2(VM_LOCKED)] = "lo",
1150 [ilog2(VM_IO)] = "io",
1151 [ilog2(VM_SEQ_READ)] = "sr",
1152 [ilog2(VM_RAND_READ)] = "rr",
1153 [ilog2(VM_DONTCOPY)] = "dc",
1154 [ilog2(VM_DONTEXPAND)] = "de",
1155 [ilog2(VM_LOCKONFAULT)] = "lf",
1156 [ilog2(VM_ACCOUNT)] = "ac",
1157 [ilog2(VM_NORESERVE)] = "nr",
1158 [ilog2(VM_HUGETLB)] = "ht",
1159 [ilog2(VM_SYNC)] = "sf",
1160 [ilog2(VM_ARCH_1)] = "ar",
1161 [ilog2(VM_WIPEONFORK)] = "wf",
1162 [ilog2(VM_DONTDUMP)] = "dd",
1163#ifdef CONFIG_ARM64_BTI
1164 [ilog2(VM_ARM64_BTI)] = "bt",
1165#endif
1166#ifdef CONFIG_MEM_SOFT_DIRTY
1167 [ilog2(VM_SOFTDIRTY)] = "sd",
1168#endif
1169 [ilog2(VM_MIXEDMAP)] = "mm",
1170 [ilog2(VM_HUGEPAGE)] = "hg",
1171 [ilog2(VM_NOHUGEPAGE)] = "nh",
1172 [ilog2(VM_MERGEABLE)] = "mg",
1173 [ilog2(VM_UFFD_MISSING)]= "um",
1174 [ilog2(VM_UFFD_WP)] = "uw",
1175#ifdef CONFIG_ARM64_MTE
1176 [ilog2(VM_MTE)] = "mt",
1177 [ilog2(VM_MTE_ALLOWED)] = "",
1178#endif
1179#ifdef CONFIG_ARCH_HAS_PKEYS
1180 /* These come out via ProtectionKey: */
1181 [ilog2(VM_PKEY_BIT0)] = "",
1182 [ilog2(VM_PKEY_BIT1)] = "",
1183 [ilog2(VM_PKEY_BIT2)] = "",
1184#if VM_PKEY_BIT3
1185 [ilog2(VM_PKEY_BIT3)] = "",
1186#endif
1187#if VM_PKEY_BIT4
1188 [ilog2(VM_PKEY_BIT4)] = "",
1189#endif
1190#endif /* CONFIG_ARCH_HAS_PKEYS */
1191#ifdef CONFIG_HAVE_ARCH_USERFAULTFD_MINOR
1192 [ilog2(VM_UFFD_MINOR)] = "ui",
1193#endif /* CONFIG_HAVE_ARCH_USERFAULTFD_MINOR */
1194#ifdef CONFIG_ARCH_HAS_USER_SHADOW_STACK
1195 [ilog2(VM_SHADOW_STACK)] = "ss",
1196#endif
1197#if defined(CONFIG_64BIT) || defined(CONFIG_PPC32)
1198 [ilog2(VM_DROPPABLE)] = "dp",
1199#endif
1200#ifdef CONFIG_64BIT
1201 [ilog2(VM_SEALED)] = "sl",
1202#endif
1203 };
1204 size_t i;
1205
1206 seq_puts(m, s: "VmFlags: ");
1207 for (i = 0; i < BITS_PER_LONG; i++) {
1208 if (!mnemonics[i][0])
1209 continue;
1210 if (vma->vm_flags & (1UL << i))
1211 seq_printf(m, fmt: "%s ", mnemonics[i]);
1212 }
1213 seq_putc(m, c: '\n');
1214}
1215
1216#ifdef CONFIG_HUGETLB_PAGE
1217static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
1218 unsigned long addr, unsigned long end,
1219 struct mm_walk *walk)
1220{
1221 struct mem_size_stats *mss = walk->private;
1222 struct vm_area_struct *vma = walk->vma;
1223 struct folio *folio = NULL;
1224 bool present = false;
1225 spinlock_t *ptl;
1226 pte_t ptent;
1227
1228 ptl = huge_pte_lock(h: hstate_vma(vma), mm: walk->mm, pte);
1229 ptent = huge_ptep_get(mm: walk->mm, addr, ptep: pte);
1230 if (pte_present(a: ptent)) {
1231 folio = page_folio(pte_page(ptent));
1232 present = true;
1233 } else if (is_swap_pte(pte: ptent)) {
1234 swp_entry_t swpent = pte_to_swp_entry(pte: ptent);
1235
1236 if (is_pfn_swap_entry(entry: swpent))
1237 folio = pfn_swap_entry_folio(entry: swpent);
1238 }
1239
1240 if (folio) {
1241 /* We treat non-present entries as "maybe shared". */
1242 if (!present || folio_maybe_mapped_shared(folio) ||
1243 hugetlb_pmd_shared(pte))
1244 mss->shared_hugetlb += huge_page_size(h: hstate_vma(vma));
1245 else
1246 mss->private_hugetlb += huge_page_size(h: hstate_vma(vma));
1247 }
1248 spin_unlock(lock: ptl);
1249 return 0;
1250}
1251#else
1252#define smaps_hugetlb_range NULL
1253#endif /* HUGETLB_PAGE */
1254
1255static const struct mm_walk_ops smaps_walk_ops = {
1256 .pmd_entry = smaps_pte_range,
1257 .hugetlb_entry = smaps_hugetlb_range,
1258 .walk_lock = PGWALK_RDLOCK,
1259};
1260
1261static const struct mm_walk_ops smaps_shmem_walk_ops = {
1262 .pmd_entry = smaps_pte_range,
1263 .hugetlb_entry = smaps_hugetlb_range,
1264 .pte_hole = smaps_pte_hole,
1265 .walk_lock = PGWALK_RDLOCK,
1266};
1267
1268/*
1269 * Gather mem stats from @vma with the indicated beginning
1270 * address @start, and keep them in @mss.
1271 *
1272 * Use vm_start of @vma as the beginning address if @start is 0.
1273 */
1274static void smap_gather_stats(struct vm_area_struct *vma,
1275 struct mem_size_stats *mss, unsigned long start)
1276{
1277 const struct mm_walk_ops *ops = &smaps_walk_ops;
1278
1279 /* Invalid start */
1280 if (start >= vma->vm_end)
1281 return;
1282
1283 if (vma->vm_file && shmem_mapping(mapping: vma->vm_file->f_mapping)) {
1284 /*
1285 * For shared or readonly shmem mappings we know that all
1286 * swapped out pages belong to the shmem object, and we can
1287 * obtain the swap value much more efficiently. For private
1288 * writable mappings, we might have COW pages that are
1289 * not affected by the parent swapped out pages of the shmem
1290 * object, so we have to distinguish them during the page walk.
1291 * Unless we know that the shmem object (or the part mapped by
1292 * our VMA) has no swapped out pages at all.
1293 */
1294 unsigned long shmem_swapped = shmem_swap_usage(vma);
1295
1296 if (!start && (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
1297 !(vma->vm_flags & VM_WRITE))) {
1298 mss->swap += shmem_swapped;
1299 } else {
1300 ops = &smaps_shmem_walk_ops;
1301 }
1302 }
1303
1304 /* mmap_lock is held in m_start */
1305 if (!start)
1306 walk_page_vma(vma, ops, private: mss);
1307 else
1308 walk_page_range(mm: vma->vm_mm, start, end: vma->vm_end, ops, private: mss);
1309}
1310
1311#define SEQ_PUT_DEC(str, val) \
1312 seq_put_decimal_ull_width(m, str, (val) >> 10, 8)
1313
1314/* Show the contents common for smaps and smaps_rollup */
1315static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss,
1316 bool rollup_mode)
1317{
1318 SEQ_PUT_DEC("Rss: ", mss->resident);
1319 SEQ_PUT_DEC(" kB\nPss: ", mss->pss >> PSS_SHIFT);
1320 SEQ_PUT_DEC(" kB\nPss_Dirty: ", mss->pss_dirty >> PSS_SHIFT);
1321 if (rollup_mode) {
1322 /*
1323 * These are meaningful only for smaps_rollup, otherwise two of
1324 * them are zero, and the other one is the same as Pss.
1325 */
1326 SEQ_PUT_DEC(" kB\nPss_Anon: ",
1327 mss->pss_anon >> PSS_SHIFT);
1328 SEQ_PUT_DEC(" kB\nPss_File: ",
1329 mss->pss_file >> PSS_SHIFT);
1330 SEQ_PUT_DEC(" kB\nPss_Shmem: ",
1331 mss->pss_shmem >> PSS_SHIFT);
1332 }
1333 SEQ_PUT_DEC(" kB\nShared_Clean: ", mss->shared_clean);
1334 SEQ_PUT_DEC(" kB\nShared_Dirty: ", mss->shared_dirty);
1335 SEQ_PUT_DEC(" kB\nPrivate_Clean: ", mss->private_clean);
1336 SEQ_PUT_DEC(" kB\nPrivate_Dirty: ", mss->private_dirty);
1337 SEQ_PUT_DEC(" kB\nReferenced: ", mss->referenced);
1338 SEQ_PUT_DEC(" kB\nAnonymous: ", mss->anonymous);
1339 SEQ_PUT_DEC(" kB\nKSM: ", mss->ksm);
1340 SEQ_PUT_DEC(" kB\nLazyFree: ", mss->lazyfree);
1341 SEQ_PUT_DEC(" kB\nAnonHugePages: ", mss->anonymous_thp);
1342 SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp);
1343 SEQ_PUT_DEC(" kB\nFilePmdMapped: ", mss->file_thp);
1344 SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb);
1345 seq_put_decimal_ull_width(m, delimiter: " kB\nPrivate_Hugetlb: ",
1346 num: mss->private_hugetlb >> 10, width: 7);
1347 SEQ_PUT_DEC(" kB\nSwap: ", mss->swap);
1348 SEQ_PUT_DEC(" kB\nSwapPss: ",
1349 mss->swap_pss >> PSS_SHIFT);
1350 SEQ_PUT_DEC(" kB\nLocked: ",
1351 mss->pss_locked >> PSS_SHIFT);
1352 seq_puts(m, s: " kB\n");
1353}
1354
1355static int show_smap(struct seq_file *m, void *v)
1356{
1357 struct vm_area_struct *vma = v;
1358 struct mem_size_stats mss = {};
1359
1360 smap_gather_stats(vma, mss: &mss, start: 0);
1361
1362 show_map_vma(m, vma);
1363
1364 SEQ_PUT_DEC("Size: ", vma->vm_end - vma->vm_start);
1365 SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma));
1366 SEQ_PUT_DEC(" kB\nMMUPageSize: ", vma_mmu_pagesize(vma));
1367 seq_puts(m, s: " kB\n");
1368
1369 __show_smap(m, mss: &mss, rollup_mode: false);
1370
1371 seq_printf(m, fmt: "THPeligible: %8u\n",
1372 !!thp_vma_allowable_orders(vma, vm_flags: vma->vm_flags, type: TVA_SMAPS,
1373 THP_ORDERS_ALL));
1374
1375 if (arch_pkeys_enabled())
1376 seq_printf(m, fmt: "ProtectionKey: %8u\n", vma_pkey(vma));
1377 show_smap_vma_flags(m, vma);
1378
1379 return 0;
1380}
1381
1382static int show_smaps_rollup(struct seq_file *m, void *v)
1383{
1384 struct proc_maps_private *priv = m->private;
1385 struct mem_size_stats mss = {};
1386 struct mm_struct *mm = priv->lock_ctx.mm;
1387 struct vm_area_struct *vma;
1388 unsigned long vma_start = 0, last_vma_end = 0;
1389 int ret = 0;
1390 VMA_ITERATOR(vmi, mm, 0);
1391
1392 priv->task = get_proc_task(inode: priv->inode);
1393 if (!priv->task)
1394 return -ESRCH;
1395
1396 if (!mm || !mmget_not_zero(mm)) {
1397 ret = -ESRCH;
1398 goto out_put_task;
1399 }
1400
1401 ret = mmap_read_lock_killable(mm);
1402 if (ret)
1403 goto out_put_mm;
1404
1405 hold_task_mempolicy(priv);
1406 vma = vma_next(vmi: &vmi);
1407
1408 if (unlikely(!vma))
1409 goto empty_set;
1410
1411 vma_start = vma->vm_start;
1412 do {
1413 smap_gather_stats(vma, mss: &mss, start: 0);
1414 last_vma_end = vma->vm_end;
1415
1416 /*
1417 * Release mmap_lock temporarily if someone wants to
1418 * access it for write request.
1419 */
1420 if (mmap_lock_is_contended(mm)) {
1421 vma_iter_invalidate(vmi: &vmi);
1422 mmap_read_unlock(mm);
1423 ret = mmap_read_lock_killable(mm);
1424 if (ret) {
1425 release_task_mempolicy(priv);
1426 goto out_put_mm;
1427 }
1428
1429 /*
1430 * After dropping the lock, there are four cases to
1431 * consider. See the following example for explanation.
1432 *
1433 * +------+------+-----------+
1434 * | VMA1 | VMA2 | VMA3 |
1435 * +------+------+-----------+
1436 * | | | |
1437 * 4k 8k 16k 400k
1438 *
1439 * Suppose we drop the lock after reading VMA2 due to
1440 * contention, then we get:
1441 *
1442 * last_vma_end = 16k
1443 *
1444 * 1) VMA2 is freed, but VMA3 exists:
1445 *
1446 * vma_next(vmi) will return VMA3.
1447 * In this case, just continue from VMA3.
1448 *
1449 * 2) VMA2 still exists:
1450 *
1451 * vma_next(vmi) will return VMA3.
1452 * In this case, just continue from VMA3.
1453 *
1454 * 3) No more VMAs can be found:
1455 *
1456 * vma_next(vmi) will return NULL.
1457 * No more things to do, just break.
1458 *
1459 * 4) (last_vma_end - 1) is the middle of a vma (VMA'):
1460 *
1461 * vma_next(vmi) will return VMA' whose range
1462 * contains last_vma_end.
1463 * Iterate VMA' from last_vma_end.
1464 */
1465 vma = vma_next(vmi: &vmi);
1466 /* Case 3 above */
1467 if (!vma)
1468 break;
1469
1470 /* Case 1 and 2 above */
1471 if (vma->vm_start >= last_vma_end) {
1472 smap_gather_stats(vma, mss: &mss, start: 0);
1473 last_vma_end = vma->vm_end;
1474 continue;
1475 }
1476
1477 /* Case 4 above */
1478 if (vma->vm_end > last_vma_end) {
1479 smap_gather_stats(vma, mss: &mss, start: last_vma_end);
1480 last_vma_end = vma->vm_end;
1481 }
1482 }
1483 } for_each_vma(vmi, vma);
1484
1485empty_set:
1486 show_vma_header_prefix(m, start: vma_start, end: last_vma_end, flags: 0, pgoff: 0, dev: 0, ino: 0);
1487 seq_pad(m, c: ' ');
1488 seq_puts(m, s: "[rollup]\n");
1489
1490 __show_smap(m, mss: &mss, rollup_mode: true);
1491
1492 release_task_mempolicy(priv);
1493 mmap_read_unlock(mm);
1494
1495out_put_mm:
1496 mmput(mm);
1497out_put_task:
1498 put_task_struct(t: priv->task);
1499 priv->task = NULL;
1500
1501 return ret;
1502}
1503#undef SEQ_PUT_DEC
1504
1505static const struct seq_operations proc_pid_smaps_op = {
1506 .start = m_start,
1507 .next = m_next,
1508 .stop = m_stop,
1509 .show = show_smap
1510};
1511
1512static int pid_smaps_open(struct inode *inode, struct file *file)
1513{
1514 return do_maps_open(inode, file, ops: &proc_pid_smaps_op);
1515}
1516
1517static int smaps_rollup_open(struct inode *inode, struct file *file)
1518{
1519 int ret;
1520 struct proc_maps_private *priv;
1521
1522 priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT);
1523 if (!priv)
1524 return -ENOMEM;
1525
1526 ret = single_open(file, show_smaps_rollup, priv);
1527 if (ret)
1528 goto out_free;
1529
1530 priv->inode = inode;
1531 priv->lock_ctx.mm = proc_mem_open(inode, PTRACE_MODE_READ);
1532 if (IS_ERR_OR_NULL(ptr: priv->lock_ctx.mm)) {
1533 ret = priv->lock_ctx.mm ? PTR_ERR(ptr: priv->lock_ctx.mm) : -ESRCH;
1534
1535 single_release(inode, file);
1536 goto out_free;
1537 }
1538
1539 return 0;
1540
1541out_free:
1542 kfree(objp: priv);
1543 return ret;
1544}
1545
1546static int smaps_rollup_release(struct inode *inode, struct file *file)
1547{
1548 struct seq_file *seq = file->private_data;
1549 struct proc_maps_private *priv = seq->private;
1550
1551 if (priv->lock_ctx.mm)
1552 mmdrop(mm: priv->lock_ctx.mm);
1553
1554 kfree(objp: priv);
1555 return single_release(inode, file);
1556}
1557
1558const struct file_operations proc_pid_smaps_operations = {
1559 .open = pid_smaps_open,
1560 .read = seq_read,
1561 .llseek = seq_lseek,
1562 .release = proc_map_release,
1563};
1564
1565const struct file_operations proc_pid_smaps_rollup_operations = {
1566 .open = smaps_rollup_open,
1567 .read = seq_read,
1568 .llseek = seq_lseek,
1569 .release = smaps_rollup_release,
1570};
1571
1572enum clear_refs_types {
1573 CLEAR_REFS_ALL = 1,
1574 CLEAR_REFS_ANON,
1575 CLEAR_REFS_MAPPED,
1576 CLEAR_REFS_SOFT_DIRTY,
1577 CLEAR_REFS_MM_HIWATER_RSS,
1578 CLEAR_REFS_LAST,
1579};
1580
1581struct clear_refs_private {
1582 enum clear_refs_types type;
1583};
1584
1585#ifdef CONFIG_MEM_SOFT_DIRTY
1586
1587static inline bool pte_is_pinned(struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1588{
1589 struct folio *folio;
1590
1591 if (!pte_write(pte))
1592 return false;
1593 if (!is_cow_mapping(vma->vm_flags))
1594 return false;
1595 if (likely(!mm_flags_test(MMF_HAS_PINNED, vma->vm_mm)))
1596 return false;
1597 folio = vm_normal_folio(vma, addr, pte);
1598 if (!folio)
1599 return false;
1600 return folio_maybe_dma_pinned(folio);
1601}
1602
1603static inline void clear_soft_dirty(struct vm_area_struct *vma,
1604 unsigned long addr, pte_t *pte)
1605{
1606 /*
1607 * The soft-dirty tracker uses #PF-s to catch writes
1608 * to pages, so write-protect the pte as well. See the
1609 * Documentation/admin-guide/mm/soft-dirty.rst for full description
1610 * of how soft-dirty works.
1611 */
1612 pte_t ptent = ptep_get(pte);
1613
1614 if (pte_present(ptent)) {
1615 pte_t old_pte;
1616
1617 if (pte_is_pinned(vma, addr, ptent))
1618 return;
1619 old_pte = ptep_modify_prot_start(vma, addr, pte);
1620 ptent = pte_wrprotect(old_pte);
1621 ptent = pte_clear_soft_dirty(ptent);
1622 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent);
1623 } else if (is_swap_pte(ptent)) {
1624 ptent = pte_swp_clear_soft_dirty(ptent);
1625 set_pte_at(vma->vm_mm, addr, pte, ptent);
1626 }
1627}
1628#else
1629static inline void clear_soft_dirty(struct vm_area_struct *vma,
1630 unsigned long addr, pte_t *pte)
1631{
1632}
1633#endif
1634
1635#if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
1636static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1637 unsigned long addr, pmd_t *pmdp)
1638{
1639 pmd_t old, pmd = *pmdp;
1640
1641 if (pmd_present(pmd)) {
1642 /* See comment in change_huge_pmd() */
1643 old = pmdp_invalidate(vma, addr, pmdp);
1644 if (pmd_dirty(old))
1645 pmd = pmd_mkdirty(pmd);
1646 if (pmd_young(old))
1647 pmd = pmd_mkyoung(pmd);
1648
1649 pmd = pmd_wrprotect(pmd);
1650 pmd = pmd_clear_soft_dirty(pmd);
1651
1652 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1653 } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1654 pmd = pmd_swp_clear_soft_dirty(pmd);
1655 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1656 }
1657}
1658#else
1659static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1660 unsigned long addr, pmd_t *pmdp)
1661{
1662}
1663#endif
1664
1665static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
1666 unsigned long end, struct mm_walk *walk)
1667{
1668 struct clear_refs_private *cp = walk->private;
1669 struct vm_area_struct *vma = walk->vma;
1670 pte_t *pte, ptent;
1671 spinlock_t *ptl;
1672 struct folio *folio;
1673
1674 ptl = pmd_trans_huge_lock(pmd, vma);
1675 if (ptl) {
1676 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1677 clear_soft_dirty_pmd(vma, addr, pmdp: pmd);
1678 goto out;
1679 }
1680
1681 if (!pmd_present(pmd: *pmd))
1682 goto out;
1683
1684 folio = pmd_folio(*pmd);
1685
1686 /* Clear accessed and referenced bits. */
1687 pmdp_test_and_clear_young(vma, addr, pmdp: pmd);
1688 folio_test_clear_young(folio);
1689 folio_clear_referenced(folio);
1690out:
1691 spin_unlock(lock: ptl);
1692 return 0;
1693 }
1694
1695 pte = pte_offset_map_lock(mm: vma->vm_mm, pmd, addr, ptlp: &ptl);
1696 if (!pte) {
1697 walk->action = ACTION_AGAIN;
1698 return 0;
1699 }
1700 for (; addr != end; pte++, addr += PAGE_SIZE) {
1701 ptent = ptep_get(ptep: pte);
1702
1703 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1704 clear_soft_dirty(vma, addr, pte);
1705 continue;
1706 }
1707
1708 if (!pte_present(a: ptent))
1709 continue;
1710
1711 folio = vm_normal_folio(vma, addr, pte: ptent);
1712 if (!folio)
1713 continue;
1714
1715 /* Clear accessed and referenced bits. */
1716 ptep_test_and_clear_young(vma, addr, ptep: pte);
1717 folio_test_clear_young(folio);
1718 folio_clear_referenced(folio);
1719 }
1720 pte_unmap_unlock(pte - 1, ptl);
1721 cond_resched();
1722 return 0;
1723}
1724
1725static int clear_refs_test_walk(unsigned long start, unsigned long end,
1726 struct mm_walk *walk)
1727{
1728 struct clear_refs_private *cp = walk->private;
1729 struct vm_area_struct *vma = walk->vma;
1730
1731 if (vma->vm_flags & VM_PFNMAP)
1732 return 1;
1733
1734 /*
1735 * Writing 1 to /proc/pid/clear_refs affects all pages.
1736 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1737 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1738 * Writing 4 to /proc/pid/clear_refs affects all pages.
1739 */
1740 if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1741 return 1;
1742 if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1743 return 1;
1744 return 0;
1745}
1746
1747static const struct mm_walk_ops clear_refs_walk_ops = {
1748 .pmd_entry = clear_refs_pte_range,
1749 .test_walk = clear_refs_test_walk,
1750 .walk_lock = PGWALK_WRLOCK,
1751};
1752
1753static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1754 size_t count, loff_t *ppos)
1755{
1756 struct task_struct *task;
1757 char buffer[PROC_NUMBUF] = {};
1758 struct mm_struct *mm;
1759 struct vm_area_struct *vma;
1760 enum clear_refs_types type;
1761 int itype;
1762 int rv;
1763
1764 if (count > sizeof(buffer) - 1)
1765 count = sizeof(buffer) - 1;
1766 if (copy_from_user(to: buffer, from: buf, n: count))
1767 return -EFAULT;
1768 rv = kstrtoint(s: strstrip(str: buffer), base: 10, res: &itype);
1769 if (rv < 0)
1770 return rv;
1771 type = (enum clear_refs_types)itype;
1772 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1773 return -EINVAL;
1774
1775 task = get_proc_task(inode: file_inode(f: file));
1776 if (!task)
1777 return -ESRCH;
1778 mm = get_task_mm(task);
1779 if (mm) {
1780 VMA_ITERATOR(vmi, mm, 0);
1781 struct mmu_notifier_range range;
1782 struct clear_refs_private cp = {
1783 .type = type,
1784 };
1785
1786 if (mmap_write_lock_killable(mm)) {
1787 count = -EINTR;
1788 goto out_mm;
1789 }
1790 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1791 /*
1792 * Writing 5 to /proc/pid/clear_refs resets the peak
1793 * resident set size to this mm's current rss value.
1794 */
1795 reset_mm_hiwater_rss(mm);
1796 goto out_unlock;
1797 }
1798
1799 if (type == CLEAR_REFS_SOFT_DIRTY) {
1800 for_each_vma(vmi, vma) {
1801 if (!(vma->vm_flags & VM_SOFTDIRTY))
1802 continue;
1803 vm_flags_clear(vma, VM_SOFTDIRTY);
1804 vma_set_page_prot(vma);
1805 }
1806
1807 inc_tlb_flush_pending(mm);
1808 mmu_notifier_range_init(range: &range, event: MMU_NOTIFY_SOFT_DIRTY,
1809 flags: 0, mm, start: 0, end: -1UL);
1810 mmu_notifier_invalidate_range_start(range: &range);
1811 }
1812 walk_page_range(mm, start: 0, end: -1, ops: &clear_refs_walk_ops, private: &cp);
1813 if (type == CLEAR_REFS_SOFT_DIRTY) {
1814 mmu_notifier_invalidate_range_end(range: &range);
1815 flush_tlb_mm(mm);
1816 dec_tlb_flush_pending(mm);
1817 }
1818out_unlock:
1819 mmap_write_unlock(mm);
1820out_mm:
1821 mmput(mm);
1822 }
1823 put_task_struct(t: task);
1824
1825 return count;
1826}
1827
1828const struct file_operations proc_clear_refs_operations = {
1829 .write = clear_refs_write,
1830 .llseek = noop_llseek,
1831};
1832
1833typedef struct {
1834 u64 pme;
1835} pagemap_entry_t;
1836
1837struct pagemapread {
1838 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */
1839 pagemap_entry_t *buffer;
1840 bool show_pfn;
1841};
1842
1843#define PAGEMAP_WALK_SIZE (PMD_SIZE)
1844#define PAGEMAP_WALK_MASK (PMD_MASK)
1845
1846#define PM_ENTRY_BYTES sizeof(pagemap_entry_t)
1847#define PM_PFRAME_BITS 55
1848#define PM_PFRAME_MASK GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1849#define PM_SOFT_DIRTY BIT_ULL(55)
1850#define PM_MMAP_EXCLUSIVE BIT_ULL(56)
1851#define PM_UFFD_WP BIT_ULL(57)
1852#define PM_GUARD_REGION BIT_ULL(58)
1853#define PM_FILE BIT_ULL(61)
1854#define PM_SWAP BIT_ULL(62)
1855#define PM_PRESENT BIT_ULL(63)
1856
1857#define PM_END_OF_BUFFER 1
1858
1859static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1860{
1861 return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1862}
1863
1864static int add_to_pagemap(pagemap_entry_t *pme, struct pagemapread *pm)
1865{
1866 pm->buffer[pm->pos++] = *pme;
1867 if (pm->pos >= pm->len)
1868 return PM_END_OF_BUFFER;
1869 return 0;
1870}
1871
1872static bool __folio_page_mapped_exclusively(struct folio *folio, struct page *page)
1873{
1874 if (IS_ENABLED(CONFIG_PAGE_MAPCOUNT))
1875 return folio_precise_page_mapcount(folio, page) == 1;
1876 return !folio_maybe_mapped_shared(folio);
1877}
1878
1879static int pagemap_pte_hole(unsigned long start, unsigned long end,
1880 __always_unused int depth, struct mm_walk *walk)
1881{
1882 struct pagemapread *pm = walk->private;
1883 unsigned long addr = start;
1884 int err = 0;
1885
1886 while (addr < end) {
1887 struct vm_area_struct *vma = find_vma(mm: walk->mm, addr);
1888 pagemap_entry_t pme = make_pme(frame: 0, flags: 0);
1889 /* End of address space hole, which we mark as non-present. */
1890 unsigned long hole_end;
1891
1892 if (vma)
1893 hole_end = min(end, vma->vm_start);
1894 else
1895 hole_end = end;
1896
1897 for (; addr < hole_end; addr += PAGE_SIZE) {
1898 err = add_to_pagemap(pme: &pme, pm);
1899 if (err)
1900 goto out;
1901 }
1902
1903 if (!vma)
1904 break;
1905
1906 /* Addresses in the VMA. */
1907 if (vma->vm_flags & VM_SOFTDIRTY)
1908 pme = make_pme(frame: 0, PM_SOFT_DIRTY);
1909 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1910 err = add_to_pagemap(pme: &pme, pm);
1911 if (err)
1912 goto out;
1913 }
1914 }
1915out:
1916 return err;
1917}
1918
1919static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1920 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1921{
1922 u64 frame = 0, flags = 0;
1923 struct page *page = NULL;
1924 struct folio *folio;
1925
1926 if (pte_present(a: pte)) {
1927 if (pm->show_pfn)
1928 frame = pte_pfn(pte);
1929 flags |= PM_PRESENT;
1930 page = vm_normal_page(vma, addr, pte);
1931 if (pte_soft_dirty(pte))
1932 flags |= PM_SOFT_DIRTY;
1933 if (pte_uffd_wp(pte))
1934 flags |= PM_UFFD_WP;
1935 } else if (is_swap_pte(pte)) {
1936 swp_entry_t entry;
1937 if (pte_swp_soft_dirty(pte))
1938 flags |= PM_SOFT_DIRTY;
1939 if (pte_swp_uffd_wp(pte))
1940 flags |= PM_UFFD_WP;
1941 entry = pte_to_swp_entry(pte);
1942 if (pm->show_pfn) {
1943 pgoff_t offset;
1944 /*
1945 * For PFN swap offsets, keeping the offset field
1946 * to be PFN only to be compatible with old smaps.
1947 */
1948 if (is_pfn_swap_entry(entry))
1949 offset = swp_offset_pfn(entry);
1950 else
1951 offset = swp_offset(entry);
1952 frame = swp_type(entry) |
1953 (offset << MAX_SWAPFILES_SHIFT);
1954 }
1955 flags |= PM_SWAP;
1956 if (is_pfn_swap_entry(entry))
1957 page = pfn_swap_entry_to_page(entry);
1958 if (pte_marker_entry_uffd_wp(entry))
1959 flags |= PM_UFFD_WP;
1960 if (is_guard_swp_entry(entry))
1961 flags |= PM_GUARD_REGION;
1962 }
1963
1964 if (page) {
1965 folio = page_folio(page);
1966 if (!folio_test_anon(folio))
1967 flags |= PM_FILE;
1968 if ((flags & PM_PRESENT) &&
1969 __folio_page_mapped_exclusively(folio, page))
1970 flags |= PM_MMAP_EXCLUSIVE;
1971 }
1972 if (vma->vm_flags & VM_SOFTDIRTY)
1973 flags |= PM_SOFT_DIRTY;
1974
1975 return make_pme(frame, flags);
1976}
1977
1978static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1979 struct mm_walk *walk)
1980{
1981 struct vm_area_struct *vma = walk->vma;
1982 struct pagemapread *pm = walk->private;
1983 spinlock_t *ptl;
1984 pte_t *pte, *orig_pte;
1985 int err = 0;
1986#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1987
1988 ptl = pmd_trans_huge_lock(pmdp, vma);
1989 if (ptl) {
1990 unsigned int idx = (addr & ~PMD_MASK) >> PAGE_SHIFT;
1991 u64 flags = 0, frame = 0;
1992 pmd_t pmd = *pmdp;
1993 struct page *page = NULL;
1994 struct folio *folio = NULL;
1995
1996 if (vma->vm_flags & VM_SOFTDIRTY)
1997 flags |= PM_SOFT_DIRTY;
1998
1999 if (pmd_present(pmd)) {
2000 page = pmd_page(pmd);
2001
2002 flags |= PM_PRESENT;
2003 if (pmd_soft_dirty(pmd))
2004 flags |= PM_SOFT_DIRTY;
2005 if (pmd_uffd_wp(pmd))
2006 flags |= PM_UFFD_WP;
2007 if (pm->show_pfn)
2008 frame = pmd_pfn(pmd) + idx;
2009 }
2010#ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
2011 else if (is_swap_pmd(pmd)) {
2012 swp_entry_t entry = pmd_to_swp_entry(pmd);
2013 unsigned long offset;
2014
2015 if (pm->show_pfn) {
2016 if (is_pfn_swap_entry(entry))
2017 offset = swp_offset_pfn(entry) + idx;
2018 else
2019 offset = swp_offset(entry) + idx;
2020 frame = swp_type(entry) |
2021 (offset << MAX_SWAPFILES_SHIFT);
2022 }
2023 flags |= PM_SWAP;
2024 if (pmd_swp_soft_dirty(pmd))
2025 flags |= PM_SOFT_DIRTY;
2026 if (pmd_swp_uffd_wp(pmd))
2027 flags |= PM_UFFD_WP;
2028 VM_BUG_ON(!is_pmd_migration_entry(pmd));
2029 page = pfn_swap_entry_to_page(entry);
2030 }
2031#endif
2032
2033 if (page) {
2034 folio = page_folio(page);
2035 if (!folio_test_anon(folio))
2036 flags |= PM_FILE;
2037 }
2038
2039 for (; addr != end; addr += PAGE_SIZE, idx++) {
2040 u64 cur_flags = flags;
2041 pagemap_entry_t pme;
2042
2043 if (folio && (flags & PM_PRESENT) &&
2044 __folio_page_mapped_exclusively(folio, page))
2045 cur_flags |= PM_MMAP_EXCLUSIVE;
2046
2047 pme = make_pme(frame, cur_flags);
2048 err = add_to_pagemap(&pme, pm);
2049 if (err)
2050 break;
2051 if (pm->show_pfn) {
2052 if (flags & PM_PRESENT)
2053 frame++;
2054 else if (flags & PM_SWAP)
2055 frame += (1 << MAX_SWAPFILES_SHIFT);
2056 }
2057 }
2058 spin_unlock(ptl);
2059 return err;
2060 }
2061#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
2062
2063 /*
2064 * We can assume that @vma always points to a valid one and @end never
2065 * goes beyond vma->vm_end.
2066 */
2067 orig_pte = pte = pte_offset_map_lock(mm: walk->mm, pmd: pmdp, addr, ptlp: &ptl);
2068 if (!pte) {
2069 walk->action = ACTION_AGAIN;
2070 return err;
2071 }
2072 for (; addr < end; pte++, addr += PAGE_SIZE) {
2073 pagemap_entry_t pme;
2074
2075 pme = pte_to_pagemap_entry(pm, vma, addr, pte: ptep_get(ptep: pte));
2076 err = add_to_pagemap(pme: &pme, pm);
2077 if (err)
2078 break;
2079 }
2080 pte_unmap_unlock(orig_pte, ptl);
2081
2082 cond_resched();
2083
2084 return err;
2085}
2086
2087#ifdef CONFIG_HUGETLB_PAGE
2088/* This function walks within one hugetlb entry in the single call */
2089static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
2090 unsigned long addr, unsigned long end,
2091 struct mm_walk *walk)
2092{
2093 struct pagemapread *pm = walk->private;
2094 struct vm_area_struct *vma = walk->vma;
2095 u64 flags = 0, frame = 0;
2096 spinlock_t *ptl;
2097 int err = 0;
2098 pte_t pte;
2099
2100 if (vma->vm_flags & VM_SOFTDIRTY)
2101 flags |= PM_SOFT_DIRTY;
2102
2103 ptl = huge_pte_lock(h: hstate_vma(vma), mm: walk->mm, pte: ptep);
2104 pte = huge_ptep_get(mm: walk->mm, addr, ptep);
2105 if (pte_present(a: pte)) {
2106 struct folio *folio = page_folio(pte_page(pte));
2107
2108 if (!folio_test_anon(folio))
2109 flags |= PM_FILE;
2110
2111 if (!folio_maybe_mapped_shared(folio) &&
2112 !hugetlb_pmd_shared(pte: ptep))
2113 flags |= PM_MMAP_EXCLUSIVE;
2114
2115 if (huge_pte_uffd_wp(pte))
2116 flags |= PM_UFFD_WP;
2117
2118 flags |= PM_PRESENT;
2119 if (pm->show_pfn)
2120 frame = pte_pfn(pte) +
2121 ((addr & ~hmask) >> PAGE_SHIFT);
2122 } else if (pte_swp_uffd_wp_any(pte)) {
2123 flags |= PM_UFFD_WP;
2124 }
2125
2126 for (; addr != end; addr += PAGE_SIZE) {
2127 pagemap_entry_t pme = make_pme(frame, flags);
2128
2129 err = add_to_pagemap(pme: &pme, pm);
2130 if (err)
2131 break;
2132 if (pm->show_pfn && (flags & PM_PRESENT))
2133 frame++;
2134 }
2135
2136 spin_unlock(lock: ptl);
2137 cond_resched();
2138
2139 return err;
2140}
2141#else
2142#define pagemap_hugetlb_range NULL
2143#endif /* HUGETLB_PAGE */
2144
2145static const struct mm_walk_ops pagemap_ops = {
2146 .pmd_entry = pagemap_pmd_range,
2147 .pte_hole = pagemap_pte_hole,
2148 .hugetlb_entry = pagemap_hugetlb_range,
2149 .walk_lock = PGWALK_RDLOCK,
2150};
2151
2152/*
2153 * /proc/pid/pagemap - an array mapping virtual pages to pfns
2154 *
2155 * For each page in the address space, this file contains one 64-bit entry
2156 * consisting of the following:
2157 *
2158 * Bits 0-54 page frame number (PFN) if present
2159 * Bits 0-4 swap type if swapped
2160 * Bits 5-54 swap offset if swapped
2161 * Bit 55 pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst)
2162 * Bit 56 page exclusively mapped
2163 * Bit 57 pte is uffd-wp write-protected
2164 * Bit 58 pte is a guard region
2165 * Bits 59-60 zero
2166 * Bit 61 page is file-page or shared-anon
2167 * Bit 62 page swapped
2168 * Bit 63 page present
2169 *
2170 * If the page is not present but in swap, then the PFN contains an
2171 * encoding of the swap file number and the page's offset into the
2172 * swap. Unmapped pages return a null PFN. This allows determining
2173 * precisely which pages are mapped (or in swap) and comparing mapped
2174 * pages between processes.
2175 *
2176 * Efficient users of this interface will use /proc/pid/maps to
2177 * determine which areas of memory are actually mapped and llseek to
2178 * skip over unmapped regions.
2179 */
2180static ssize_t pagemap_read(struct file *file, char __user *buf,
2181 size_t count, loff_t *ppos)
2182{
2183 struct mm_struct *mm = file->private_data;
2184 struct pagemapread pm;
2185 unsigned long src;
2186 unsigned long svpfn;
2187 unsigned long start_vaddr;
2188 unsigned long end_vaddr;
2189 int ret = 0, copied = 0;
2190
2191 if (!mm || !mmget_not_zero(mm))
2192 goto out;
2193
2194 ret = -EINVAL;
2195 /* file position must be aligned */
2196 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
2197 goto out_mm;
2198
2199 ret = 0;
2200 if (!count)
2201 goto out_mm;
2202
2203 /* do not disclose physical addresses: attack vector */
2204 pm.show_pfn = file_ns_capable(file, ns: &init_user_ns, CAP_SYS_ADMIN);
2205
2206 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
2207 pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL);
2208 ret = -ENOMEM;
2209 if (!pm.buffer)
2210 goto out_mm;
2211
2212 src = *ppos;
2213 svpfn = src / PM_ENTRY_BYTES;
2214 end_vaddr = mm->task_size;
2215
2216 /* watch out for wraparound */
2217 start_vaddr = end_vaddr;
2218 if (svpfn <= (ULONG_MAX >> PAGE_SHIFT)) {
2219 unsigned long end;
2220
2221 ret = mmap_read_lock_killable(mm);
2222 if (ret)
2223 goto out_free;
2224 start_vaddr = untagged_addr_remote(mm, svpfn << PAGE_SHIFT);
2225 mmap_read_unlock(mm);
2226
2227 end = start_vaddr + ((count / PM_ENTRY_BYTES) << PAGE_SHIFT);
2228 if (end >= start_vaddr && end < mm->task_size)
2229 end_vaddr = end;
2230 }
2231
2232 /* Ensure the address is inside the task */
2233 if (start_vaddr > mm->task_size)
2234 start_vaddr = end_vaddr;
2235
2236 ret = 0;
2237 while (count && (start_vaddr < end_vaddr)) {
2238 int len;
2239 unsigned long end;
2240
2241 pm.pos = 0;
2242 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
2243 /* overflow ? */
2244 if (end < start_vaddr || end > end_vaddr)
2245 end = end_vaddr;
2246 ret = mmap_read_lock_killable(mm);
2247 if (ret)
2248 goto out_free;
2249 ret = walk_page_range(mm, start: start_vaddr, end, ops: &pagemap_ops, private: &pm);
2250 mmap_read_unlock(mm);
2251 start_vaddr = end;
2252
2253 len = min(count, PM_ENTRY_BYTES * pm.pos);
2254 if (copy_to_user(to: buf, from: pm.buffer, n: len)) {
2255 ret = -EFAULT;
2256 goto out_free;
2257 }
2258 copied += len;
2259 buf += len;
2260 count -= len;
2261 }
2262 *ppos += copied;
2263 if (!ret || ret == PM_END_OF_BUFFER)
2264 ret = copied;
2265
2266out_free:
2267 kfree(objp: pm.buffer);
2268out_mm:
2269 mmput(mm);
2270out:
2271 return ret;
2272}
2273
2274static int pagemap_open(struct inode *inode, struct file *file)
2275{
2276 struct mm_struct *mm;
2277
2278 mm = proc_mem_open(inode, PTRACE_MODE_READ);
2279 if (IS_ERR_OR_NULL(ptr: mm))
2280 return mm ? PTR_ERR(ptr: mm) : -ESRCH;
2281 file->private_data = mm;
2282 return 0;
2283}
2284
2285static int pagemap_release(struct inode *inode, struct file *file)
2286{
2287 struct mm_struct *mm = file->private_data;
2288
2289 if (mm)
2290 mmdrop(mm);
2291 return 0;
2292}
2293
2294#define PM_SCAN_CATEGORIES (PAGE_IS_WPALLOWED | PAGE_IS_WRITTEN | \
2295 PAGE_IS_FILE | PAGE_IS_PRESENT | \
2296 PAGE_IS_SWAPPED | PAGE_IS_PFNZERO | \
2297 PAGE_IS_HUGE | PAGE_IS_SOFT_DIRTY | \
2298 PAGE_IS_GUARD)
2299#define PM_SCAN_FLAGS (PM_SCAN_WP_MATCHING | PM_SCAN_CHECK_WPASYNC)
2300
2301struct pagemap_scan_private {
2302 struct pm_scan_arg arg;
2303 unsigned long masks_of_interest, cur_vma_category;
2304 struct page_region *vec_buf;
2305 unsigned long vec_buf_len, vec_buf_index, found_pages;
2306 struct page_region __user *vec_out;
2307};
2308
2309static unsigned long pagemap_page_category(struct pagemap_scan_private *p,
2310 struct vm_area_struct *vma,
2311 unsigned long addr, pte_t pte)
2312{
2313 unsigned long categories = 0;
2314
2315 if (pte_present(a: pte)) {
2316 struct page *page;
2317
2318 categories |= PAGE_IS_PRESENT;
2319 if (!pte_uffd_wp(pte))
2320 categories |= PAGE_IS_WRITTEN;
2321
2322 if (p->masks_of_interest & PAGE_IS_FILE) {
2323 page = vm_normal_page(vma, addr, pte);
2324 if (page && !PageAnon(page))
2325 categories |= PAGE_IS_FILE;
2326 }
2327
2328 if (is_zero_pfn(pfn: pte_pfn(pte)))
2329 categories |= PAGE_IS_PFNZERO;
2330 if (pte_soft_dirty(pte))
2331 categories |= PAGE_IS_SOFT_DIRTY;
2332 } else if (is_swap_pte(pte)) {
2333 swp_entry_t swp;
2334
2335 categories |= PAGE_IS_SWAPPED;
2336 if (!pte_swp_uffd_wp_any(pte))
2337 categories |= PAGE_IS_WRITTEN;
2338
2339 swp = pte_to_swp_entry(pte);
2340 if (is_guard_swp_entry(entry: swp))
2341 categories |= PAGE_IS_GUARD;
2342 else if ((p->masks_of_interest & PAGE_IS_FILE) &&
2343 is_pfn_swap_entry(entry: swp) &&
2344 !folio_test_anon(folio: pfn_swap_entry_folio(entry: swp)))
2345 categories |= PAGE_IS_FILE;
2346
2347 if (pte_swp_soft_dirty(pte))
2348 categories |= PAGE_IS_SOFT_DIRTY;
2349 }
2350
2351 return categories;
2352}
2353
2354static void make_uffd_wp_pte(struct vm_area_struct *vma,
2355 unsigned long addr, pte_t *pte, pte_t ptent)
2356{
2357 if (pte_present(a: ptent)) {
2358 pte_t old_pte;
2359
2360 old_pte = ptep_modify_prot_start(vma, addr, ptep: pte);
2361 ptent = pte_mkuffd_wp(pte: old_pte);
2362 ptep_modify_prot_commit(vma, addr, ptep: pte, old_pte, pte: ptent);
2363 } else if (is_swap_pte(pte: ptent)) {
2364 ptent = pte_swp_mkuffd_wp(pte: ptent);
2365 set_pte_at(vma->vm_mm, addr, pte, ptent);
2366 } else {
2367 set_pte_at(vma->vm_mm, addr, pte,
2368 make_pte_marker(PTE_MARKER_UFFD_WP));
2369 }
2370}
2371
2372#ifdef CONFIG_TRANSPARENT_HUGEPAGE
2373static unsigned long pagemap_thp_category(struct pagemap_scan_private *p,
2374 struct vm_area_struct *vma,
2375 unsigned long addr, pmd_t pmd)
2376{
2377 unsigned long categories = PAGE_IS_HUGE;
2378
2379 if (pmd_present(pmd)) {
2380 struct page *page;
2381
2382 categories |= PAGE_IS_PRESENT;
2383 if (!pmd_uffd_wp(pmd))
2384 categories |= PAGE_IS_WRITTEN;
2385
2386 if (p->masks_of_interest & PAGE_IS_FILE) {
2387 page = vm_normal_page_pmd(vma, addr, pmd);
2388 if (page && !PageAnon(page))
2389 categories |= PAGE_IS_FILE;
2390 }
2391
2392 if (is_huge_zero_pmd(pmd))
2393 categories |= PAGE_IS_PFNZERO;
2394 if (pmd_soft_dirty(pmd))
2395 categories |= PAGE_IS_SOFT_DIRTY;
2396 } else if (is_swap_pmd(pmd)) {
2397 swp_entry_t swp;
2398
2399 categories |= PAGE_IS_SWAPPED;
2400 if (!pmd_swp_uffd_wp(pmd))
2401 categories |= PAGE_IS_WRITTEN;
2402 if (pmd_swp_soft_dirty(pmd))
2403 categories |= PAGE_IS_SOFT_DIRTY;
2404
2405 if (p->masks_of_interest & PAGE_IS_FILE) {
2406 swp = pmd_to_swp_entry(pmd);
2407 if (is_pfn_swap_entry(swp) &&
2408 !folio_test_anon(pfn_swap_entry_folio(swp)))
2409 categories |= PAGE_IS_FILE;
2410 }
2411 }
2412
2413 return categories;
2414}
2415
2416static void make_uffd_wp_pmd(struct vm_area_struct *vma,
2417 unsigned long addr, pmd_t *pmdp)
2418{
2419 pmd_t old, pmd = *pmdp;
2420
2421 if (pmd_present(pmd)) {
2422 old = pmdp_invalidate_ad(vma, addr, pmdp);
2423 pmd = pmd_mkuffd_wp(old);
2424 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
2425 } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
2426 pmd = pmd_swp_mkuffd_wp(pmd);
2427 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
2428 }
2429}
2430#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
2431
2432#ifdef CONFIG_HUGETLB_PAGE
2433static unsigned long pagemap_hugetlb_category(pte_t pte)
2434{
2435 unsigned long categories = PAGE_IS_HUGE;
2436
2437 /*
2438 * According to pagemap_hugetlb_range(), file-backed HugeTLB
2439 * page cannot be swapped. So PAGE_IS_FILE is not checked for
2440 * swapped pages.
2441 */
2442 if (pte_present(a: pte)) {
2443 categories |= PAGE_IS_PRESENT;
2444 if (!huge_pte_uffd_wp(pte))
2445 categories |= PAGE_IS_WRITTEN;
2446 if (!PageAnon(pte_page(pte)))
2447 categories |= PAGE_IS_FILE;
2448 if (is_zero_pfn(pfn: pte_pfn(pte)))
2449 categories |= PAGE_IS_PFNZERO;
2450 if (pte_soft_dirty(pte))
2451 categories |= PAGE_IS_SOFT_DIRTY;
2452 } else if (is_swap_pte(pte)) {
2453 categories |= PAGE_IS_SWAPPED;
2454 if (!pte_swp_uffd_wp_any(pte))
2455 categories |= PAGE_IS_WRITTEN;
2456 if (pte_swp_soft_dirty(pte))
2457 categories |= PAGE_IS_SOFT_DIRTY;
2458 }
2459
2460 return categories;
2461}
2462
2463static void make_uffd_wp_huge_pte(struct vm_area_struct *vma,
2464 unsigned long addr, pte_t *ptep,
2465 pte_t ptent)
2466{
2467 unsigned long psize;
2468
2469 if (is_hugetlb_entry_hwpoisoned(pte: ptent) || is_pte_marker(pte: ptent))
2470 return;
2471
2472 psize = huge_page_size(h: hstate_vma(vma));
2473
2474 if (is_hugetlb_entry_migration(pte: ptent))
2475 set_huge_pte_at(mm: vma->vm_mm, addr, ptep,
2476 pte: pte_swp_mkuffd_wp(pte: ptent), sz: psize);
2477 else if (!huge_pte_none(pte: ptent))
2478 huge_ptep_modify_prot_commit(vma, addr, ptep, old_pte: ptent,
2479 pte: huge_pte_mkuffd_wp(pte: ptent));
2480 else
2481 set_huge_pte_at(mm: vma->vm_mm, addr, ptep,
2482 pte: make_pte_marker(PTE_MARKER_UFFD_WP), sz: psize);
2483}
2484#endif /* CONFIG_HUGETLB_PAGE */
2485
2486#if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLB_PAGE)
2487static void pagemap_scan_backout_range(struct pagemap_scan_private *p,
2488 unsigned long addr, unsigned long end)
2489{
2490 struct page_region *cur_buf = &p->vec_buf[p->vec_buf_index];
2491
2492 if (!p->vec_buf)
2493 return;
2494
2495 if (cur_buf->start != addr)
2496 cur_buf->end = addr;
2497 else
2498 cur_buf->start = cur_buf->end = 0;
2499
2500 p->found_pages -= (end - addr) / PAGE_SIZE;
2501}
2502#endif
2503
2504static bool pagemap_scan_is_interesting_page(unsigned long categories,
2505 const struct pagemap_scan_private *p)
2506{
2507 categories ^= p->arg.category_inverted;
2508 if ((categories & p->arg.category_mask) != p->arg.category_mask)
2509 return false;
2510 if (p->arg.category_anyof_mask && !(categories & p->arg.category_anyof_mask))
2511 return false;
2512
2513 return true;
2514}
2515
2516static bool pagemap_scan_is_interesting_vma(unsigned long categories,
2517 const struct pagemap_scan_private *p)
2518{
2519 unsigned long required = p->arg.category_mask & PAGE_IS_WPALLOWED;
2520
2521 categories ^= p->arg.category_inverted;
2522 if ((categories & required) != required)
2523 return false;
2524
2525 return true;
2526}
2527
2528static int pagemap_scan_test_walk(unsigned long start, unsigned long end,
2529 struct mm_walk *walk)
2530{
2531 struct pagemap_scan_private *p = walk->private;
2532 struct vm_area_struct *vma = walk->vma;
2533 unsigned long vma_category = 0;
2534 bool wp_allowed = userfaultfd_wp_async(vma) &&
2535 userfaultfd_wp_use_markers(vma);
2536
2537 if (!wp_allowed) {
2538 /* User requested explicit failure over wp-async capability */
2539 if (p->arg.flags & PM_SCAN_CHECK_WPASYNC)
2540 return -EPERM;
2541 /*
2542 * User requires wr-protect, and allows silently skipping
2543 * unsupported vmas.
2544 */
2545 if (p->arg.flags & PM_SCAN_WP_MATCHING)
2546 return 1;
2547 /*
2548 * Then the request doesn't involve wr-protects at all,
2549 * fall through to the rest checks, and allow vma walk.
2550 */
2551 }
2552
2553 if (vma->vm_flags & VM_PFNMAP)
2554 return 1;
2555
2556 if (wp_allowed)
2557 vma_category |= PAGE_IS_WPALLOWED;
2558
2559 if (vma->vm_flags & VM_SOFTDIRTY)
2560 vma_category |= PAGE_IS_SOFT_DIRTY;
2561
2562 if (!pagemap_scan_is_interesting_vma(categories: vma_category, p))
2563 return 1;
2564
2565 p->cur_vma_category = vma_category;
2566
2567 return 0;
2568}
2569
2570static bool pagemap_scan_push_range(unsigned long categories,
2571 struct pagemap_scan_private *p,
2572 unsigned long addr, unsigned long end)
2573{
2574 struct page_region *cur_buf = &p->vec_buf[p->vec_buf_index];
2575
2576 /*
2577 * When there is no output buffer provided at all, the sentinel values
2578 * won't match here. There is no other way for `cur_buf->end` to be
2579 * non-zero other than it being non-empty.
2580 */
2581 if (addr == cur_buf->end && categories == cur_buf->categories) {
2582 cur_buf->end = end;
2583 return true;
2584 }
2585
2586 if (cur_buf->end) {
2587 if (p->vec_buf_index >= p->vec_buf_len - 1)
2588 return false;
2589
2590 cur_buf = &p->vec_buf[++p->vec_buf_index];
2591 }
2592
2593 cur_buf->start = addr;
2594 cur_buf->end = end;
2595 cur_buf->categories = categories;
2596
2597 return true;
2598}
2599
2600static int pagemap_scan_output(unsigned long categories,
2601 struct pagemap_scan_private *p,
2602 unsigned long addr, unsigned long *end)
2603{
2604 unsigned long n_pages, total_pages;
2605 int ret = 0;
2606
2607 if (!p->vec_buf)
2608 return 0;
2609
2610 categories &= p->arg.return_mask;
2611
2612 n_pages = (*end - addr) / PAGE_SIZE;
2613 if (check_add_overflow(p->found_pages, n_pages, &total_pages) ||
2614 total_pages > p->arg.max_pages) {
2615 size_t n_too_much = total_pages - p->arg.max_pages;
2616 *end -= n_too_much * PAGE_SIZE;
2617 n_pages -= n_too_much;
2618 ret = -ENOSPC;
2619 }
2620
2621 if (!pagemap_scan_push_range(categories, p, addr, end: *end)) {
2622 *end = addr;
2623 n_pages = 0;
2624 ret = -ENOSPC;
2625 }
2626
2627 p->found_pages += n_pages;
2628 if (ret)
2629 p->arg.walk_end = *end;
2630
2631 return ret;
2632}
2633
2634static int pagemap_scan_thp_entry(pmd_t *pmd, unsigned long start,
2635 unsigned long end, struct mm_walk *walk)
2636{
2637#ifdef CONFIG_TRANSPARENT_HUGEPAGE
2638 struct pagemap_scan_private *p = walk->private;
2639 struct vm_area_struct *vma = walk->vma;
2640 unsigned long categories;
2641 spinlock_t *ptl;
2642 int ret = 0;
2643
2644 ptl = pmd_trans_huge_lock(pmd, vma);
2645 if (!ptl)
2646 return -ENOENT;
2647
2648 categories = p->cur_vma_category |
2649 pagemap_thp_category(p, vma, start, *pmd);
2650
2651 if (!pagemap_scan_is_interesting_page(categories, p))
2652 goto out_unlock;
2653
2654 ret = pagemap_scan_output(categories, p, start, &end);
2655 if (start == end)
2656 goto out_unlock;
2657
2658 if (~p->arg.flags & PM_SCAN_WP_MATCHING)
2659 goto out_unlock;
2660 if (~categories & PAGE_IS_WRITTEN)
2661 goto out_unlock;
2662
2663 /*
2664 * Break huge page into small pages if the WP operation
2665 * needs to be performed on a portion of the huge page.
2666 */
2667 if (end != start + HPAGE_SIZE) {
2668 spin_unlock(ptl);
2669 split_huge_pmd(vma, pmd, start);
2670 pagemap_scan_backout_range(p, start, end);
2671 /* Report as if there was no THP */
2672 return -ENOENT;
2673 }
2674
2675 make_uffd_wp_pmd(vma, start, pmd);
2676 flush_tlb_range(vma, start, end);
2677out_unlock:
2678 spin_unlock(ptl);
2679 return ret;
2680#else /* !CONFIG_TRANSPARENT_HUGEPAGE */
2681 return -ENOENT;
2682#endif
2683}
2684
2685static int pagemap_scan_pmd_entry(pmd_t *pmd, unsigned long start,
2686 unsigned long end, struct mm_walk *walk)
2687{
2688 struct pagemap_scan_private *p = walk->private;
2689 struct vm_area_struct *vma = walk->vma;
2690 unsigned long addr, flush_end = 0;
2691 pte_t *pte, *start_pte;
2692 spinlock_t *ptl;
2693 int ret;
2694
2695 ret = pagemap_scan_thp_entry(pmd, start, end, walk);
2696 if (ret != -ENOENT)
2697 return ret;
2698
2699 ret = 0;
2700 start_pte = pte = pte_offset_map_lock(mm: vma->vm_mm, pmd, addr: start, ptlp: &ptl);
2701 if (!pte) {
2702 walk->action = ACTION_AGAIN;
2703 return 0;
2704 }
2705
2706 arch_enter_lazy_mmu_mode();
2707
2708 if ((p->arg.flags & PM_SCAN_WP_MATCHING) && !p->vec_out) {
2709 /* Fast path for performing exclusive WP */
2710 for (addr = start; addr != end; pte++, addr += PAGE_SIZE) {
2711 pte_t ptent = ptep_get(ptep: pte);
2712
2713 if ((pte_present(a: ptent) && pte_uffd_wp(pte: ptent)) ||
2714 pte_swp_uffd_wp_any(pte: ptent))
2715 continue;
2716 make_uffd_wp_pte(vma, addr, pte, ptent);
2717 if (!flush_end)
2718 start = addr;
2719 flush_end = addr + PAGE_SIZE;
2720 }
2721 goto flush_and_return;
2722 }
2723
2724 if (!p->arg.category_anyof_mask && !p->arg.category_inverted &&
2725 p->arg.category_mask == PAGE_IS_WRITTEN &&
2726 p->arg.return_mask == PAGE_IS_WRITTEN) {
2727 for (addr = start; addr < end; pte++, addr += PAGE_SIZE) {
2728 unsigned long next = addr + PAGE_SIZE;
2729 pte_t ptent = ptep_get(ptep: pte);
2730
2731 if ((pte_present(a: ptent) && pte_uffd_wp(pte: ptent)) ||
2732 pte_swp_uffd_wp_any(pte: ptent))
2733 continue;
2734 ret = pagemap_scan_output(categories: p->cur_vma_category | PAGE_IS_WRITTEN,
2735 p, addr, end: &next);
2736 if (next == addr)
2737 break;
2738 if (~p->arg.flags & PM_SCAN_WP_MATCHING)
2739 continue;
2740 make_uffd_wp_pte(vma, addr, pte, ptent);
2741 if (!flush_end)
2742 start = addr;
2743 flush_end = next;
2744 }
2745 goto flush_and_return;
2746 }
2747
2748 for (addr = start; addr != end; pte++, addr += PAGE_SIZE) {
2749 pte_t ptent = ptep_get(ptep: pte);
2750 unsigned long categories = p->cur_vma_category |
2751 pagemap_page_category(p, vma, addr, pte: ptent);
2752 unsigned long next = addr + PAGE_SIZE;
2753
2754 if (!pagemap_scan_is_interesting_page(categories, p))
2755 continue;
2756
2757 ret = pagemap_scan_output(categories, p, addr, end: &next);
2758 if (next == addr)
2759 break;
2760
2761 if (~p->arg.flags & PM_SCAN_WP_MATCHING)
2762 continue;
2763 if (~categories & PAGE_IS_WRITTEN)
2764 continue;
2765
2766 make_uffd_wp_pte(vma, addr, pte, ptent);
2767 if (!flush_end)
2768 start = addr;
2769 flush_end = next;
2770 }
2771
2772flush_and_return:
2773 if (flush_end)
2774 flush_tlb_range(vma, start, addr);
2775
2776 arch_leave_lazy_mmu_mode();
2777 pte_unmap_unlock(start_pte, ptl);
2778
2779 cond_resched();
2780 return ret;
2781}
2782
2783#ifdef CONFIG_HUGETLB_PAGE
2784static int pagemap_scan_hugetlb_entry(pte_t *ptep, unsigned long hmask,
2785 unsigned long start, unsigned long end,
2786 struct mm_walk *walk)
2787{
2788 struct pagemap_scan_private *p = walk->private;
2789 struct vm_area_struct *vma = walk->vma;
2790 unsigned long categories;
2791 spinlock_t *ptl;
2792 int ret = 0;
2793 pte_t pte;
2794
2795 if (~p->arg.flags & PM_SCAN_WP_MATCHING) {
2796 /* Go the short route when not write-protecting pages. */
2797
2798 pte = huge_ptep_get(mm: walk->mm, addr: start, ptep);
2799 categories = p->cur_vma_category | pagemap_hugetlb_category(pte);
2800
2801 if (!pagemap_scan_is_interesting_page(categories, p))
2802 return 0;
2803
2804 return pagemap_scan_output(categories, p, addr: start, end: &end);
2805 }
2806
2807 i_mmap_lock_write(mapping: vma->vm_file->f_mapping);
2808 ptl = huge_pte_lock(h: hstate_vma(vma), mm: vma->vm_mm, pte: ptep);
2809
2810 pte = huge_ptep_get(mm: walk->mm, addr: start, ptep);
2811 categories = p->cur_vma_category | pagemap_hugetlb_category(pte);
2812
2813 if (!pagemap_scan_is_interesting_page(categories, p))
2814 goto out_unlock;
2815
2816 ret = pagemap_scan_output(categories, p, addr: start, end: &end);
2817 if (start == end)
2818 goto out_unlock;
2819
2820 if (~categories & PAGE_IS_WRITTEN)
2821 goto out_unlock;
2822
2823 if (end != start + HPAGE_SIZE) {
2824 /* Partial HugeTLB page WP isn't possible. */
2825 pagemap_scan_backout_range(p, addr: start, end);
2826 p->arg.walk_end = start;
2827 ret = 0;
2828 goto out_unlock;
2829 }
2830
2831 make_uffd_wp_huge_pte(vma, addr: start, ptep, ptent: pte);
2832 flush_hugetlb_tlb_range(vma, start, end);
2833
2834out_unlock:
2835 spin_unlock(lock: ptl);
2836 i_mmap_unlock_write(mapping: vma->vm_file->f_mapping);
2837
2838 return ret;
2839}
2840#else
2841#define pagemap_scan_hugetlb_entry NULL
2842#endif
2843
2844static int pagemap_scan_pte_hole(unsigned long addr, unsigned long end,
2845 int depth, struct mm_walk *walk)
2846{
2847 struct pagemap_scan_private *p = walk->private;
2848 struct vm_area_struct *vma = walk->vma;
2849 int ret, err;
2850
2851 if (!vma || !pagemap_scan_is_interesting_page(categories: p->cur_vma_category, p))
2852 return 0;
2853
2854 ret = pagemap_scan_output(categories: p->cur_vma_category, p, addr, end: &end);
2855 if (addr == end)
2856 return ret;
2857
2858 if (~p->arg.flags & PM_SCAN_WP_MATCHING)
2859 return ret;
2860
2861 err = uffd_wp_range(vma, start: addr, len: end - addr, enable_wp: true);
2862 if (err < 0)
2863 ret = err;
2864
2865 return ret;
2866}
2867
2868static const struct mm_walk_ops pagemap_scan_ops = {
2869 .test_walk = pagemap_scan_test_walk,
2870 .pmd_entry = pagemap_scan_pmd_entry,
2871 .pte_hole = pagemap_scan_pte_hole,
2872 .hugetlb_entry = pagemap_scan_hugetlb_entry,
2873};
2874
2875static int pagemap_scan_get_args(struct pm_scan_arg *arg,
2876 unsigned long uarg)
2877{
2878 if (copy_from_user(to: arg, from: (void __user *)uarg, n: sizeof(*arg)))
2879 return -EFAULT;
2880
2881 if (arg->size != sizeof(struct pm_scan_arg))
2882 return -EINVAL;
2883
2884 /* Validate requested features */
2885 if (arg->flags & ~PM_SCAN_FLAGS)
2886 return -EINVAL;
2887 if ((arg->category_inverted | arg->category_mask |
2888 arg->category_anyof_mask | arg->return_mask) & ~PM_SCAN_CATEGORIES)
2889 return -EINVAL;
2890
2891 arg->start = untagged_addr((unsigned long)arg->start);
2892 arg->end = untagged_addr((unsigned long)arg->end);
2893 arg->vec = untagged_addr((unsigned long)arg->vec);
2894
2895 /* Validate memory pointers */
2896 if (!IS_ALIGNED(arg->start, PAGE_SIZE))
2897 return -EINVAL;
2898 if (!access_ok((void __user *)(long)arg->start, arg->end - arg->start))
2899 return -EFAULT;
2900 if (!arg->vec && arg->vec_len)
2901 return -EINVAL;
2902 if (UINT_MAX == SIZE_MAX && arg->vec_len > SIZE_MAX)
2903 return -EINVAL;
2904 if (arg->vec && !access_ok((void __user *)(long)arg->vec,
2905 size_mul(arg->vec_len, sizeof(struct page_region))))
2906 return -EFAULT;
2907
2908 /* Fixup default values */
2909 arg->end = ALIGN(arg->end, PAGE_SIZE);
2910 arg->walk_end = 0;
2911 if (!arg->max_pages)
2912 arg->max_pages = ULONG_MAX;
2913
2914 return 0;
2915}
2916
2917static int pagemap_scan_writeback_args(struct pm_scan_arg *arg,
2918 unsigned long uargl)
2919{
2920 struct pm_scan_arg __user *uarg = (void __user *)uargl;
2921
2922 if (copy_to_user(to: &uarg->walk_end, from: &arg->walk_end, n: sizeof(arg->walk_end)))
2923 return -EFAULT;
2924
2925 return 0;
2926}
2927
2928static int pagemap_scan_init_bounce_buffer(struct pagemap_scan_private *p)
2929{
2930 if (!p->arg.vec_len)
2931 return 0;
2932
2933 p->vec_buf_len = min_t(size_t, PAGEMAP_WALK_SIZE >> PAGE_SHIFT,
2934 p->arg.vec_len);
2935 p->vec_buf = kmalloc_array(p->vec_buf_len, sizeof(*p->vec_buf),
2936 GFP_KERNEL);
2937 if (!p->vec_buf)
2938 return -ENOMEM;
2939
2940 p->vec_buf->start = p->vec_buf->end = 0;
2941 p->vec_out = (struct page_region __user *)(long)p->arg.vec;
2942
2943 return 0;
2944}
2945
2946static long pagemap_scan_flush_buffer(struct pagemap_scan_private *p)
2947{
2948 const struct page_region *buf = p->vec_buf;
2949 long n = p->vec_buf_index;
2950
2951 if (!p->vec_buf)
2952 return 0;
2953
2954 if (buf[n].end != buf[n].start)
2955 n++;
2956
2957 if (!n)
2958 return 0;
2959
2960 if (copy_to_user(to: p->vec_out, from: buf, n: n * sizeof(*buf)))
2961 return -EFAULT;
2962
2963 p->arg.vec_len -= n;
2964 p->vec_out += n;
2965
2966 p->vec_buf_index = 0;
2967 p->vec_buf_len = min_t(size_t, p->vec_buf_len, p->arg.vec_len);
2968 p->vec_buf->start = p->vec_buf->end = 0;
2969
2970 return n;
2971}
2972
2973static long do_pagemap_scan(struct mm_struct *mm, unsigned long uarg)
2974{
2975 struct pagemap_scan_private p = {0};
2976 unsigned long walk_start;
2977 size_t n_ranges_out = 0;
2978 int ret;
2979
2980 ret = pagemap_scan_get_args(arg: &p.arg, uarg);
2981 if (ret)
2982 return ret;
2983
2984 p.masks_of_interest = p.arg.category_mask | p.arg.category_anyof_mask |
2985 p.arg.return_mask;
2986 ret = pagemap_scan_init_bounce_buffer(p: &p);
2987 if (ret)
2988 return ret;
2989
2990 for (walk_start = p.arg.start; walk_start < p.arg.end;
2991 walk_start = p.arg.walk_end) {
2992 struct mmu_notifier_range range;
2993 long n_out;
2994
2995 if (fatal_signal_pending(current)) {
2996 ret = -EINTR;
2997 break;
2998 }
2999
3000 ret = mmap_read_lock_killable(mm);
3001 if (ret)
3002 break;
3003
3004 /* Protection change for the range is going to happen. */
3005 if (p.arg.flags & PM_SCAN_WP_MATCHING) {
3006 mmu_notifier_range_init(range: &range, event: MMU_NOTIFY_PROTECTION_VMA, flags: 0,
3007 mm, start: walk_start, end: p.arg.end);
3008 mmu_notifier_invalidate_range_start(range: &range);
3009 }
3010
3011 ret = walk_page_range(mm, start: walk_start, end: p.arg.end,
3012 ops: &pagemap_scan_ops, private: &p);
3013
3014 if (p.arg.flags & PM_SCAN_WP_MATCHING)
3015 mmu_notifier_invalidate_range_end(range: &range);
3016
3017 mmap_read_unlock(mm);
3018
3019 n_out = pagemap_scan_flush_buffer(p: &p);
3020 if (n_out < 0)
3021 ret = n_out;
3022 else
3023 n_ranges_out += n_out;
3024
3025 if (ret != -ENOSPC)
3026 break;
3027
3028 if (p.arg.vec_len == 0 || p.found_pages == p.arg.max_pages)
3029 break;
3030 }
3031
3032 /* ENOSPC signifies early stop (buffer full) from the walk. */
3033 if (!ret || ret == -ENOSPC)
3034 ret = n_ranges_out;
3035
3036 /* The walk_end isn't set when ret is zero */
3037 if (!p.arg.walk_end)
3038 p.arg.walk_end = p.arg.end;
3039 if (pagemap_scan_writeback_args(arg: &p.arg, uargl: uarg))
3040 ret = -EFAULT;
3041
3042 kfree(objp: p.vec_buf);
3043 return ret;
3044}
3045
3046static long do_pagemap_cmd(struct file *file, unsigned int cmd,
3047 unsigned long arg)
3048{
3049 struct mm_struct *mm = file->private_data;
3050
3051 switch (cmd) {
3052 case PAGEMAP_SCAN:
3053 return do_pagemap_scan(mm, uarg: arg);
3054
3055 default:
3056 return -EINVAL;
3057 }
3058}
3059
3060const struct file_operations proc_pagemap_operations = {
3061 .llseek = mem_lseek, /* borrow this */
3062 .read = pagemap_read,
3063 .open = pagemap_open,
3064 .release = pagemap_release,
3065 .unlocked_ioctl = do_pagemap_cmd,
3066 .compat_ioctl = do_pagemap_cmd,
3067};
3068#endif /* CONFIG_PROC_PAGE_MONITOR */
3069
3070#ifdef CONFIG_NUMA
3071
3072struct numa_maps {
3073 unsigned long pages;
3074 unsigned long anon;
3075 unsigned long active;
3076 unsigned long writeback;
3077 unsigned long mapcount_max;
3078 unsigned long dirty;
3079 unsigned long swapcache;
3080 unsigned long node[MAX_NUMNODES];
3081};
3082
3083struct numa_maps_private {
3084 struct proc_maps_private proc_maps;
3085 struct numa_maps md;
3086};
3087
3088static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
3089 unsigned long nr_pages)
3090{
3091 struct folio *folio = page_folio(page);
3092 int count;
3093
3094 if (IS_ENABLED(CONFIG_PAGE_MAPCOUNT))
3095 count = folio_precise_page_mapcount(folio, page);
3096 else
3097 count = folio_average_page_mapcount(folio);
3098
3099 md->pages += nr_pages;
3100 if (pte_dirty || folio_test_dirty(folio))
3101 md->dirty += nr_pages;
3102
3103 if (folio_test_swapcache(folio))
3104 md->swapcache += nr_pages;
3105
3106 if (folio_test_active(folio) || folio_test_unevictable(folio))
3107 md->active += nr_pages;
3108
3109 if (folio_test_writeback(folio))
3110 md->writeback += nr_pages;
3111
3112 if (folio_test_anon(folio))
3113 md->anon += nr_pages;
3114
3115 if (count > md->mapcount_max)
3116 md->mapcount_max = count;
3117
3118 md->node[folio_nid(folio)] += nr_pages;
3119}
3120
3121static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
3122 unsigned long addr)
3123{
3124 struct page *page;
3125 int nid;
3126
3127 if (!pte_present(a: pte))
3128 return NULL;
3129
3130 page = vm_normal_page(vma, addr, pte);
3131 if (!page || is_zone_device_page(page))
3132 return NULL;
3133
3134 if (PageReserved(page))
3135 return NULL;
3136
3137 nid = page_to_nid(page);
3138 if (!node_isset(nid, node_states[N_MEMORY]))
3139 return NULL;
3140
3141 return page;
3142}
3143
3144#ifdef CONFIG_TRANSPARENT_HUGEPAGE
3145static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
3146 struct vm_area_struct *vma,
3147 unsigned long addr)
3148{
3149 struct page *page;
3150 int nid;
3151
3152 if (!pmd_present(pmd))
3153 return NULL;
3154
3155 page = vm_normal_page_pmd(vma, addr, pmd);
3156 if (!page)
3157 return NULL;
3158
3159 if (PageReserved(page))
3160 return NULL;
3161
3162 nid = page_to_nid(page);
3163 if (!node_isset(nid, node_states[N_MEMORY]))
3164 return NULL;
3165
3166 return page;
3167}
3168#endif
3169
3170static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
3171 unsigned long end, struct mm_walk *walk)
3172{
3173 struct numa_maps *md = walk->private;
3174 struct vm_area_struct *vma = walk->vma;
3175 spinlock_t *ptl;
3176 pte_t *orig_pte;
3177 pte_t *pte;
3178
3179#ifdef CONFIG_TRANSPARENT_HUGEPAGE
3180 ptl = pmd_trans_huge_lock(pmd, vma);
3181 if (ptl) {
3182 struct page *page;
3183
3184 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
3185 if (page)
3186 gather_stats(page, md, pmd_dirty(*pmd),
3187 HPAGE_PMD_SIZE/PAGE_SIZE);
3188 spin_unlock(ptl);
3189 return 0;
3190 }
3191#endif
3192 orig_pte = pte = pte_offset_map_lock(mm: walk->mm, pmd, addr, ptlp: &ptl);
3193 if (!pte) {
3194 walk->action = ACTION_AGAIN;
3195 return 0;
3196 }
3197 do {
3198 pte_t ptent = ptep_get(ptep: pte);
3199 struct page *page = can_gather_numa_stats(pte: ptent, vma, addr);
3200 if (!page)
3201 continue;
3202 gather_stats(page, md, pte_dirty: pte_dirty(pte: ptent), nr_pages: 1);
3203
3204 } while (pte++, addr += PAGE_SIZE, addr != end);
3205 pte_unmap_unlock(orig_pte, ptl);
3206 cond_resched();
3207 return 0;
3208}
3209#ifdef CONFIG_HUGETLB_PAGE
3210static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
3211 unsigned long addr, unsigned long end, struct mm_walk *walk)
3212{
3213 pte_t huge_pte;
3214 struct numa_maps *md;
3215 struct page *page;
3216 spinlock_t *ptl;
3217
3218 ptl = huge_pte_lock(h: hstate_vma(vma: walk->vma), mm: walk->mm, pte);
3219 huge_pte = huge_ptep_get(mm: walk->mm, addr, ptep: pte);
3220 if (!pte_present(a: huge_pte))
3221 goto out;
3222
3223 page = pte_page(huge_pte);
3224
3225 md = walk->private;
3226 gather_stats(page, md, pte_dirty: pte_dirty(pte: huge_pte), nr_pages: 1);
3227out:
3228 spin_unlock(lock: ptl);
3229 return 0;
3230}
3231
3232#else
3233static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
3234 unsigned long addr, unsigned long end, struct mm_walk *walk)
3235{
3236 return 0;
3237}
3238#endif
3239
3240static const struct mm_walk_ops show_numa_ops = {
3241 .hugetlb_entry = gather_hugetlb_stats,
3242 .pmd_entry = gather_pte_stats,
3243 .walk_lock = PGWALK_RDLOCK,
3244};
3245
3246/*
3247 * Display pages allocated per node and memory policy via /proc.
3248 */
3249static int show_numa_map(struct seq_file *m, void *v)
3250{
3251 struct numa_maps_private *numa_priv = m->private;
3252 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
3253 struct vm_area_struct *vma = v;
3254 struct numa_maps *md = &numa_priv->md;
3255 struct file *file = vma->vm_file;
3256 struct mm_struct *mm = vma->vm_mm;
3257 char buffer[64];
3258 struct mempolicy *pol;
3259 pgoff_t ilx;
3260 int nid;
3261
3262 if (!mm)
3263 return 0;
3264
3265 /* Ensure we start with an empty set of numa_maps statistics. */
3266 memset(s: md, c: 0, n: sizeof(*md));
3267
3268 pol = __get_vma_policy(vma, addr: vma->vm_start, ilx: &ilx);
3269 if (pol) {
3270 mpol_to_str(buffer, maxlen: sizeof(buffer), pol);
3271 mpol_cond_put(pol);
3272 } else {
3273 mpol_to_str(buffer, maxlen: sizeof(buffer), pol: proc_priv->task_mempolicy);
3274 }
3275
3276 seq_printf(m, fmt: "%08lx %s", vma->vm_start, buffer);
3277
3278 if (file) {
3279 seq_puts(m, s: " file=");
3280 seq_path(m, file_user_path(f: file), "\n\t= ");
3281 } else if (vma_is_initial_heap(vma)) {
3282 seq_puts(m, s: " heap");
3283 } else if (vma_is_initial_stack(vma)) {
3284 seq_puts(m, s: " stack");
3285 }
3286
3287 if (is_vm_hugetlb_page(vma))
3288 seq_puts(m, s: " huge");
3289
3290 /* mmap_lock is held by m_start */
3291 walk_page_vma(vma, ops: &show_numa_ops, private: md);
3292
3293 if (!md->pages)
3294 goto out;
3295
3296 if (md->anon)
3297 seq_printf(m, fmt: " anon=%lu", md->anon);
3298
3299 if (md->dirty)
3300 seq_printf(m, fmt: " dirty=%lu", md->dirty);
3301
3302 if (md->pages != md->anon && md->pages != md->dirty)
3303 seq_printf(m, fmt: " mapped=%lu", md->pages);
3304
3305 if (md->mapcount_max > 1)
3306 seq_printf(m, fmt: " mapmax=%lu", md->mapcount_max);
3307
3308 if (md->swapcache)
3309 seq_printf(m, fmt: " swapcache=%lu", md->swapcache);
3310
3311 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
3312 seq_printf(m, fmt: " active=%lu", md->active);
3313
3314 if (md->writeback)
3315 seq_printf(m, fmt: " writeback=%lu", md->writeback);
3316
3317 for_each_node_state(nid, N_MEMORY)
3318 if (md->node[nid])
3319 seq_printf(m, fmt: " N%d=%lu", nid, md->node[nid]);
3320
3321 seq_printf(m, fmt: " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
3322out:
3323 seq_putc(m, c: '\n');
3324 return 0;
3325}
3326
3327static const struct seq_operations proc_pid_numa_maps_op = {
3328 .start = m_start,
3329 .next = m_next,
3330 .stop = m_stop,
3331 .show = show_numa_map,
3332};
3333
3334static int pid_numa_maps_open(struct inode *inode, struct file *file)
3335{
3336 return proc_maps_open(inode, file, ops: &proc_pid_numa_maps_op,
3337 psize: sizeof(struct numa_maps_private));
3338}
3339
3340const struct file_operations proc_pid_numa_maps_operations = {
3341 .open = pid_numa_maps_open,
3342 .read = seq_read,
3343 .llseek = seq_lseek,
3344 .release = proc_map_release,
3345};
3346
3347#endif /* CONFIG_NUMA */
3348