projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
fs/proc/task_mmu: fix loss of young/dirty bits during pagemap scan
[linux-2.6-block.git] / fs / proc / task_mmu.c
CommitLineData
b2441318 1// SPDX-License-Identifier: GPL-2.0
a520110e 2#include <linux/pagewalk.h>
17fca131 3#include <linux/mm_inline.h>
1da177e4 4#include <linux/hugetlb.h>
22e057c5 5#include <linux/huge_mm.h>
1da177e4 6#include <linux/mount.h>
8b479335 7#include <linux/ksm.h>
1da177e4 8#include <linux/seq_file.h>
e070ad49 9#include <linux/highmem.h>
5096add8 10#include <linux/ptrace.h>
5a0e3ad6 11#include <linux/slab.h>
6e21c8f1
CL		 12#include <linux/pagemap.h>
13#include <linux/mempolicy.h>
22e057c5 14#include <linux/rmap.h>
85863e47 15#include <linux/swap.h>
6e84f315 16#include <linux/sched/mm.h>
85863e47 17#include <linux/swapops.h>
0f8975ec 18#include <linux/mmu_notifier.h>
33c3fc71 19#include <linux/page_idle.h>
6a15a370 20#include <linux/shmem_fs.h>
b3a81d08 21#include <linux/uaccess.h>
27cca866 22#include <linux/pkeys.h>
52526ca7
MUA		 23#include <linux/minmax.h>
24#include <linux/overflow.h>
e070ad49 25
1da177e4 26#include <asm/elf.h>
b3a81d08 27#include <asm/tlb.h>
e070ad49 28#include <asm/tlbflush.h>
1da177e4
LT		 29#include "internal.h"
30
d1be35cb
AV		 31#define SEQ_PUT_DEC(str, val) \
32 seq_put_decimal_ull_width(m, str, (val) << (PAGE_SHIFT-10), 8)
df5f8314 33void task_mem(struct seq_file *m, struct mm_struct *mm)
1da177e4 34{
af5b0f6a 35 unsigned long text, lib, swap, anon, file, shmem;
365e9c87
HD		 36 unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
37
8cee852e
JM		 38 anon = get_mm_counter(mm, MM_ANONPAGES);
39 file = get_mm_counter(mm, MM_FILEPAGES);
40 shmem = get_mm_counter(mm, MM_SHMEMPAGES);
41
365e9c87
HD		 42 /*
43 * Note: to minimize their overhead, mm maintains hiwater_vm and
44 * hiwater_rss only when about to *lower* total_vm or rss. Any
45 * collector of these hiwater stats must therefore get total_vm
46 * and rss too, which will usually be the higher. Barriers? not
47 * worth the effort, such snapshots can always be inconsistent.
48 */
49 hiwater_vm = total_vm = mm->total_vm;
50 if (hiwater_vm < mm->hiwater_vm)
51 hiwater_vm = mm->hiwater_vm;
8cee852e 52 hiwater_rss = total_rss = anon + file + shmem;
365e9c87
HD		 53 if (hiwater_rss < mm->hiwater_rss)
54 hiwater_rss = mm->hiwater_rss;
1da177e4 55
8526d84f
KK		 56 /* split executable areas between text and lib */
57 text = PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK);
58 text = min(text, mm->exec_vm << PAGE_SHIFT);
59 lib = (mm->exec_vm << PAGE_SHIFT) - text;
60
b084d435 61 swap = get_mm_counter(mm, MM_SWAPENTS);
d1be35cb
AV		 62 SEQ_PUT_DEC("VmPeak:\t", hiwater_vm);
63 SEQ_PUT_DEC(" kB\nVmSize:\t", total_vm);
64 SEQ_PUT_DEC(" kB\nVmLck:\t", mm->locked_vm);
70f8a3ca 65 SEQ_PUT_DEC(" kB\nVmPin:\t", atomic64_read(&mm->pinned_vm));
d1be35cb
AV		 66 SEQ_PUT_DEC(" kB\nVmHWM:\t", hiwater_rss);
67 SEQ_PUT_DEC(" kB\nVmRSS:\t", total_rss);
68 SEQ_PUT_DEC(" kB\nRssAnon:\t", anon);
69 SEQ_PUT_DEC(" kB\nRssFile:\t", file);
70 SEQ_PUT_DEC(" kB\nRssShmem:\t", shmem);
71 SEQ_PUT_DEC(" kB\nVmData:\t", mm->data_vm);
72 SEQ_PUT_DEC(" kB\nVmStk:\t", mm->stack_vm);
73 seq_put_decimal_ull_width(m,
74 " kB\nVmExe:\t", text >> 10, 8);
75 seq_put_decimal_ull_width(m,
76 " kB\nVmLib:\t", lib >> 10, 8);
77 seq_put_decimal_ull_width(m,
78 " kB\nVmPTE:\t", mm_pgtables_bytes(mm) >> 10, 8);
79 SEQ_PUT_DEC(" kB\nVmSwap:\t", swap);
80 seq_puts(m, " kB\n");
5d317b2b 81 hugetlb_report_usage(m, mm);
1da177e4 82}
d1be35cb 83#undef SEQ_PUT_DEC
1da177e4
LT		 84
85unsigned long task_vsize(struct mm_struct *mm)
86{
87 return PAGE_SIZE * mm->total_vm;
88}
89
a2ade7b6
AD		 90unsigned long task_statm(struct mm_struct *mm,
91 unsigned long *shared, unsigned long *text,
92 unsigned long *data, unsigned long *resident)
1da177e4 93{
eca56ff9
JM		 94 *shared = get_mm_counter(mm, MM_FILEPAGES) +
95 get_mm_counter(mm, MM_SHMEMPAGES);
1da177e4
LT		 96 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
97 >> PAGE_SHIFT;
84638335 98 *data = mm->data_vm + mm->stack_vm;
d559db08 99 *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
1da177e4
LT		 100 return mm->total_vm;
101}
102
9e781440
KH		 103#ifdef CONFIG_NUMA
104/*
498f2371 105 * Save get_task_policy() for show_numa_map().
9e781440
KH		 106 */
107static void hold_task_mempolicy(struct proc_maps_private *priv)
108{
109 struct task_struct *task = priv->task;
110
111 task_lock(task);
498f2371 112 priv->task_mempolicy = get_task_policy(task);
9e781440
KH		 113 mpol_get(priv->task_mempolicy);
114 task_unlock(task);
115}
116static void release_task_mempolicy(struct proc_maps_private *priv)
117{
118 mpol_put(priv->task_mempolicy);
119}
120#else
121static void hold_task_mempolicy(struct proc_maps_private *priv)
122{
123}
124static void release_task_mempolicy(struct proc_maps_private *priv)
125{
126}
127#endif
128
c4c84f06
MWO		 129static struct vm_area_struct *proc_get_vma(struct proc_maps_private *priv,
130 loff_t *ppos)
131{
132 struct vm_area_struct *vma = vma_next(&priv->iter);
133
134 if (vma) {
135 *ppos = vma->vm_start;
136 } else {
137 *ppos = -2UL;
138 vma = get_gate_vma(priv->mm);
139 }
140
141 return vma;
142}
143
0c255321 144static void *m_start(struct seq_file *m, loff_t *ppos)
e070ad49 145{
a6198797 146 struct proc_maps_private *priv = m->private;
4781f2c3 147 unsigned long last_addr = *ppos;
a6198797 148 struct mm_struct *mm;
a6198797 149
c2e88d22 150 /* See m_next(). Zero at the start or after lseek. */
b8c20a9b
ON		 151 if (last_addr == -1UL)
152 return NULL;
153
2c03376d 154 priv->task = get_proc_task(priv->inode);
a6198797 155 if (!priv->task)
ec6fd8a4 156 return ERR_PTR(-ESRCH);
a6198797 157
29a40ace 158 mm = priv->mm;
d07ded61
MWO		 159 if (!mm || !mmget_not_zero(mm)) {
160 put_task_struct(priv->task);
161 priv->task = NULL;
29a40ace 162 return NULL;
d07ded61 163 }
a6198797 164
d8ed45c5 165 if (mmap_read_lock_killable(mm)) {
8a713e7d 166 mmput(mm);
d07ded61
MWO		 167 put_task_struct(priv->task);
168 priv->task = NULL;
8a713e7d
KK		 169 return ERR_PTR(-EINTR);
170 }
171
c4c84f06 172 vma_iter_init(&priv->iter, mm, last_addr);
9e781440 173 hold_task_mempolicy(priv);
c4c84f06
MWO		 174 if (last_addr == -2UL)
175 return get_gate_vma(mm);
59b4bf12 176
c4c84f06 177 return proc_get_vma(priv, ppos);
a6198797
MM		 178}
179
4781f2c3 180static void *m_next(struct seq_file *m, void *v, loff_t *ppos)
a6198797 181{
c4c84f06
MWO		 182 if (*ppos == -2UL) {
183 *ppos = -1UL;
184 return NULL;
185 }
186 return proc_get_vma(m->private, ppos);
a6198797
MM		 187}
188
189static void m_stop(struct seq_file *m, void *v)
190{
191 struct proc_maps_private *priv = m->private;
d07ded61 192 struct mm_struct *mm = priv->mm;
a6198797 193
d07ded61
MWO		 194 if (!priv->task)
195 return;
196
197 release_task_mempolicy(priv);
d8ed45c5 198 mmap_read_unlock(mm);
d07ded61
MWO		 199 mmput(mm);
200 put_task_struct(priv->task);
201 priv->task = NULL;
a6198797
MM		 202}
203
4db7d0ee
ON		 204static int proc_maps_open(struct inode *inode, struct file *file,
205 const struct seq_operations *ops, int psize)
206{
207 struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
208
209 if (!priv)
210 return -ENOMEM;
211
2c03376d 212 priv->inode = inode;
29a40ace
ON		 213 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
214 if (IS_ERR(priv->mm)) {
215 int err = PTR_ERR(priv->mm);
216
217 seq_release_private(inode, file);
218 return err;
219 }
220
4db7d0ee
ON		 221 return 0;
222}
223
29a40ace
ON		 224static int proc_map_release(struct inode *inode, struct file *file)
225{
226 struct seq_file *seq = file->private_data;
227 struct proc_maps_private *priv = seq->private;
228
229 if (priv->mm)
230 mmdrop(priv->mm);
231
232 return seq_release_private(inode, file);
233}
234
a6198797 235static int do_maps_open(struct inode *inode, struct file *file,
03a44825 236 const struct seq_operations *ops)
a6198797 237{
4db7d0ee
ON		 238 return proc_maps_open(inode, file, ops,
239 sizeof(struct proc_maps_private));
a6198797 240}
e070ad49 241
493b0e9d
DC		 242static void show_vma_header_prefix(struct seq_file *m,
243 unsigned long start, unsigned long end,
244 vm_flags_t flags, unsigned long long pgoff,
245 dev_t dev, unsigned long ino)
246{
247 seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
0e3dc019
AV		 248 seq_put_hex_ll(m, NULL, start, 8);
249 seq_put_hex_ll(m, "-", end, 8);
250 seq_putc(m, ' ');
251 seq_putc(m, flags & VM_READ ? 'r' : '-');
252 seq_putc(m, flags & VM_WRITE ? 'w' : '-');
253 seq_putc(m, flags & VM_EXEC ? 'x' : '-');
254 seq_putc(m, flags & VM_MAYSHARE ? 's' : 'p');
255 seq_put_hex_ll(m, " ", pgoff, 8);
256 seq_put_hex_ll(m, " ", MAJOR(dev), 2);
257 seq_put_hex_ll(m, ":", MINOR(dev), 2);
258 seq_put_decimal_ull(m, " ", ino);
259 seq_putc(m, ' ');
493b0e9d
DC		 260}
261
b7643757 262static void
871305bb 263show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
1da177e4 264{
d09e8ca6 265 struct anon_vma_name *anon_name = NULL;
e070ad49
ML		 266 struct mm_struct *mm = vma->vm_mm;
267 struct file *file = vma->vm_file;
ca16d140 268 vm_flags_t flags = vma->vm_flags;
1da177e4 269 unsigned long ino = 0;
6260a4b0 270 unsigned long long pgoff = 0;
a09a79f6 271 unsigned long start, end;
1da177e4 272 dev_t dev = 0;
b7643757 273 const char *name = NULL;
1da177e4
LT		 274
275 if (file) {
3efdc78f
AV		 276 const struct inode *inode = file_user_inode(vma->vm_file);
277
1da177e4
LT		 278 dev = inode->i_sb->s_dev;
279 ino = inode->i_ino;
6260a4b0 280 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
1da177e4
LT		 281 }
282
d7824370 283 start = vma->vm_start;
a09a79f6 284 end = vma->vm_end;
493b0e9d 285 show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino);
d09e8ca6
PT		 286 if (mm)
287 anon_name = anon_vma_name(vma);
1da177e4
LT		 288
289 /*
290 * Print the dentry name for named mappings, and a
291 * special [heap] marker for the heap:
292 */
e070ad49 293 if (file) {
652586df 294 seq_pad(m, ' ');
d09e8ca6
PT		 295 /*
296 * If user named this anon shared memory via
297 * prctl(PR_SET_VMA ..., use the provided name.
298 */
299 if (anon_name)
300 seq_printf(m, "[anon_shmem:%s]", anon_name->name);
301 else
08582d67 302 seq_path(m, file_user_path(file), "\n");
b7643757
SP		 303 goto done;
304 }
305
78d683e8
AL		 306 if (vma->vm_ops && vma->vm_ops->name) {
307 name = vma->vm_ops->name(vma);
308 if (name)
309 goto done;
310 }
311
b7643757
SP		 312 name = arch_vma_name(vma);
313 if (!name) {
b7643757
SP		 314 if (!mm) {
315 name = "[vdso]";
316 goto done;
317 }
318
11250fd1 319 if (vma_is_initial_heap(vma)) {
b7643757
SP		 320 name = "[heap]";
321 goto done;
322 }
323
11250fd1 324 if (vma_is_initial_stack(vma)) {
65376df5 325 name = "[stack]";
9a10064f
CC		 326 goto done;
327 }
328
9a10064f
CC		 329 if (anon_name) {
330 seq_pad(m, ' ');
5c26f6ac 331 seq_printf(m, "[anon:%s]", anon_name->name);
9a10064f 332 }
b7643757
SP		 333 }
334
335done:
336 if (name) {
652586df 337 seq_pad(m, ' ');
b7643757 338 seq_puts(m, name);
1da177e4
LT		 339 }
340 seq_putc(m, '\n');
7c88db0c
JK		 341}
342
871305bb 343static int show_map(struct seq_file *m, void *v)
7c88db0c 344{
871305bb 345 show_map_vma(m, v);
1da177e4
LT		 346 return 0;
347}
348
03a44825 349static const struct seq_operations proc_pid_maps_op = {
a6198797
MM		 350 .start = m_start,
351 .next = m_next,
352 .stop = m_stop,
871305bb 353 .show = show_map
a6198797
MM		 354};
355
b7643757 356static int pid_maps_open(struct inode *inode, struct file *file)
a6198797
MM		 357{
358 return do_maps_open(inode, file, &proc_pid_maps_op);
359}
360
b7643757
SP		 361const struct file_operations proc_pid_maps_operations = {
362 .open = pid_maps_open,
363 .read = seq_read,
364 .llseek = seq_lseek,
29a40ace 365 .release = proc_map_release,
b7643757
SP		 366};
367
a6198797
MM		 368/*
369 * Proportional Set Size(PSS): my share of RSS.
370 *
371 * PSS of a process is the count of pages it has in memory, where each
372 * page is divided by the number of processes sharing it. So if a
373 * process has 1000 pages all to itself, and 1000 shared with one other
374 * process, its PSS will be 1500.
375 *
376 * To keep (accumulated) division errors low, we adopt a 64bit
377 * fixed-point pss counter to minimize division errors. So (pss >>
378 * PSS_SHIFT) would be the real byte count.
379 *
380 * A shift of 12 before division means (assuming 4K page size):
381 * - 1M 3-user-pages add up to 8KB errors;
382 * - supports mapcount up to 2^24, or 16M;
383 * - supports PSS up to 2^52 bytes, or 4PB.
384 */
385#define PSS_SHIFT 12
386
1e883281 387#ifdef CONFIG_PROC_PAGE_MONITOR
214e471f 388struct mem_size_stats {
a6198797
MM		 389 unsigned long resident;
390 unsigned long shared_clean;
391 unsigned long shared_dirty;
392 unsigned long private_clean;
393 unsigned long private_dirty;
394 unsigned long referenced;
b40d4f84 395 unsigned long anonymous;
cf8496ea 396 unsigned long lazyfree;
4031a219 397 unsigned long anonymous_thp;
65c45377 398 unsigned long shmem_thp;
60fbf0ab 399 unsigned long file_thp;
214e471f 400 unsigned long swap;
25ee01a2
NH		 401 unsigned long shared_hugetlb;
402 unsigned long private_hugetlb;
8b479335 403 unsigned long ksm;
a6198797 404 u64 pss;
ee2ad71b
LS		 405 u64 pss_anon;
406 u64 pss_file;
407 u64 pss_shmem;
30934843 408 u64 pss_dirty;
493b0e9d 409 u64 pss_locked;
8334b962 410 u64 swap_pss;
a6198797
MM		 411};
412
ee2ad71b
LS		 413static void smaps_page_accumulate(struct mem_size_stats *mss,
414 struct page *page, unsigned long size, unsigned long pss,
415 bool dirty, bool locked, bool private)
416{
417 mss->pss += pss;
418
419 if (PageAnon(page))
420 mss->pss_anon += pss;
421 else if (PageSwapBacked(page))
422 mss->pss_shmem += pss;
423 else
424 mss->pss_file += pss;
425
426 if (locked)
427 mss->pss_locked += pss;
428
429 if (dirty || PageDirty(page)) {
30934843 430 mss->pss_dirty += pss;
ee2ad71b
LS		 431 if (private)
432 mss->private_dirty += size;
433 else
434 mss->shared_dirty += size;
435 } else {
436 if (private)
437 mss->private_clean += size;
438 else
439 mss->shared_clean += size;
440 }
441}
442
c164e038 443static void smaps_account(struct mem_size_stats *mss, struct page *page,
24d7275c
YS		 444 bool compound, bool young, bool dirty, bool locked,
445 bool migration)
c164e038 446{
d8c6546b 447 int i, nr = compound ? compound_nr(page) : 1;
afd9883f 448 unsigned long size = nr * PAGE_SIZE;
c164e038 449
ee2ad71b
LS		 450 /*
451 * First accumulate quantities that depend only on |size| and the type
452 * of the compound page.
453 */
cf8496ea 454 if (PageAnon(page)) {
c164e038 455 mss->anonymous += size;
cf8496ea
SL		 456 if (!PageSwapBacked(page) && !dirty && !PageDirty(page))
457 mss->lazyfree += size;
458 }
c164e038 459
8b479335
SR		 460 if (PageKsm(page))
461 mss->ksm += size;
462
c164e038
KS		 463 mss->resident += size;
464 /* Accumulate the size in pages that have been accessed. */
33c3fc71 465 if (young || page_is_young(page) || PageReferenced(page))
c164e038 466 mss->referenced += size;
c164e038 467
afd9883f 468 /*
ee2ad71b
LS		 469 * Then accumulate quantities that may depend on sharing, or that may
470 * differ page-by-page.
471 *
afd9883f
KS		 472 * page_count(page) == 1 guarantees the page is mapped exactly once.
473 * If any subpage of the compound page mapped with PTE it would elevate
474 * page_count().
24d7275c
YS		 475 *
476 * The page_mapcount() is called to get a snapshot of the mapcount.
477 * Without holding the page lock this snapshot can be slightly wrong as
478 * we cannot always read the mapcount atomically. It is not safe to
479 * call page_mapcount() even with PTL held if the page is not mapped,
480 * especially for migration entries. Treat regular migration entries
481 * as mapcount == 1.
afd9883f 482 */
24d7275c 483 if ((page_count(page) == 1) || migration) {
ee2ad71b
LS		 484 smaps_page_accumulate(mss, page, size, size << PSS_SHIFT, dirty,
485 locked, true);
afd9883f
KS		 486 return;
487 }
afd9883f
KS		 488 for (i = 0; i < nr; i++, page++) {
489 int mapcount = page_mapcount(page);
ee2ad71b
LS		 490 unsigned long pss = PAGE_SIZE << PSS_SHIFT;
491 if (mapcount >= 2)
492 pss /= mapcount;
493 smaps_page_accumulate(mss, page, PAGE_SIZE, pss, dirty, locked,
494 mapcount < 2);
c164e038
KS		 495 }
496}
ae11c4d9 497
c261e7d9 498#ifdef CONFIG_SHMEM
c261e7d9 499static int smaps_pte_hole(unsigned long addr, unsigned long end,
b7a16c7a 500 __always_unused int depth, struct mm_walk *walk)
c261e7d9
VB		 501{
502 struct mem_size_stats *mss = walk->private;
10c848c8 503 struct vm_area_struct *vma = walk->vma;
c261e7d9 504
10c848c8
PX		 505 mss->swap += shmem_partial_swap_usage(walk->vma->vm_file->f_mapping,
506 linear_page_index(vma, addr),
507 linear_page_index(vma, end));
c261e7d9
VB		 508
509 return 0;
510}
7b86ac33
CH		 511#else
512#define smaps_pte_hole NULL
513#endif /* CONFIG_SHMEM */
c261e7d9 514
23010032
PX		 515static void smaps_pte_hole_lookup(unsigned long addr, struct mm_walk *walk)
516{
517#ifdef CONFIG_SHMEM
518 if (walk->ops->pte_hole) {
519 /* depth is not used */
520 smaps_pte_hole(addr, addr + PAGE_SIZE, 0, walk);
521 }
522#endif
523}
524
c164e038
KS		 525static void smaps_pte_entry(pte_t *pte, unsigned long addr,
526 struct mm_walk *walk)
ae11c4d9
DH		 527{
528 struct mem_size_stats *mss = walk->private;
14eb6fdd 529 struct vm_area_struct *vma = walk->vma;
27dd768e 530 bool locked = !!(vma->vm_flags & VM_LOCKED);
b1d4d9e0 531 struct page *page = NULL;
efd41493 532 bool migration = false, young = false, dirty = false;
c33c7948 533 pte_t ptent = ptep_get(pte);
ae11c4d9 534
c33c7948
RR		 535 if (pte_present(ptent)) {
536 page = vm_normal_page(vma, addr, ptent);
537 young = pte_young(ptent);
538 dirty = pte_dirty(ptent);
539 } else if (is_swap_pte(ptent)) {
540 swp_entry_t swpent = pte_to_swp_entry(ptent);
ae11c4d9 541
8334b962
MK		 542 if (!non_swap_entry(swpent)) {
543 int mapcount;
544
c164e038 545 mss->swap += PAGE_SIZE;
8334b962
MK		 546 mapcount = swp_swapcount(swpent);
547 if (mapcount >= 2) {
548 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
549
550 do_div(pss_delta, mapcount);
551 mss->swap_pss += pss_delta;
552 } else {
553 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
554 }
24d7275c
YS		 555 } else if (is_pfn_swap_entry(swpent)) {
556 if (is_migration_entry(swpent))
557 migration = true;
af5cdaf8 558 page = pfn_swap_entry_to_page(swpent);
24d7275c 559 }
23010032
PX		 560 } else {
561 smaps_pte_hole_lookup(addr, walk);
48131e03 562 return;
b1d4d9e0 563 }
ae11c4d9 564
ae11c4d9
DH		 565 if (!page)
566 return;
afd9883f 567
efd41493 568 smaps_account(mss, page, false, young, dirty, locked, migration);
ae11c4d9
DH		 569}
570
c164e038
KS		 571#ifdef CONFIG_TRANSPARENT_HUGEPAGE
572static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
573 struct mm_walk *walk)
574{
575 struct mem_size_stats *mss = walk->private;
14eb6fdd 576 struct vm_area_struct *vma = walk->vma;
27dd768e 577 bool locked = !!(vma->vm_flags & VM_LOCKED);
c94b6923 578 struct page *page = NULL;
24d7275c 579 bool migration = false;
c94b6923
HY		 580
581 if (pmd_present(*pmd)) {
8b9c1cc0 582 page = vm_normal_page_pmd(vma, addr, *pmd);
c94b6923
HY		 583 } else if (unlikely(thp_migration_supported() && is_swap_pmd(*pmd))) {
584 swp_entry_t entry = pmd_to_swp_entry(*pmd);
c164e038 585
24d7275c
YS		 586 if (is_migration_entry(entry)) {
587 migration = true;
af5cdaf8 588 page = pfn_swap_entry_to_page(entry);
24d7275c 589 }
c94b6923 590 }
c164e038
KS		 591 if (IS_ERR_OR_NULL(page))
592 return;
65c45377
KS		 593 if (PageAnon(page))
594 mss->anonymous_thp += HPAGE_PMD_SIZE;
595 else if (PageSwapBacked(page))
596 mss->shmem_thp += HPAGE_PMD_SIZE;
ca120cf6
DW		 597 else if (is_zone_device_page(page))
598 /* pass */;
65c45377 599 else
60fbf0ab 600 mss->file_thp += HPAGE_PMD_SIZE;
24d7275c
YS		 601
602 smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd),
603 locked, migration);
c164e038
KS		 604}
605#else
606static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
607 struct mm_walk *walk)
608{
609}
610#endif
611
b3ae5acb 612static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
2165009b 613 struct mm_walk *walk)
e070ad49 614{
14eb6fdd 615 struct vm_area_struct *vma = walk->vma;
ae11c4d9 616 pte_t *pte;
705e87c0 617 spinlock_t *ptl;
e070ad49 618
b6ec57f4
KS		 619 ptl = pmd_trans_huge_lock(pmd, vma);
620 if (ptl) {
c94b6923 621 smaps_pmd_entry(pmd, addr, walk);
bf929152 622 spin_unlock(ptl);
14038302 623 goto out;
22e057c5 624 }
1a5a9906 625
705e87c0 626 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
7780d040
HD		 627 if (!pte) {
628 walk->action = ACTION_AGAIN;
629 return 0;
630 }
ae11c4d9 631 for (; addr != end; pte++, addr += PAGE_SIZE)
c164e038 632 smaps_pte_entry(pte, addr, walk);
705e87c0 633 pte_unmap_unlock(pte - 1, ptl);
14038302 634out:
705e87c0 635 cond_resched();
b3ae5acb 636 return 0;
e070ad49
ML		 637}
638
834f82e2
CG		 639static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
640{
641 /*
642 * Don't forget to update Documentation/ on changes.
643 */
644 static const char mnemonics[BITS_PER_LONG][2] = {
645 /*
646 * In case if we meet a flag we don't know about.
647 */
648 [0 ... (BITS_PER_LONG-1)] = "??",
649
650 [ilog2(VM_READ)] = "rd",
651 [ilog2(VM_WRITE)] = "wr",
652 [ilog2(VM_EXEC)] = "ex",
653 [ilog2(VM_SHARED)] = "sh",
654 [ilog2(VM_MAYREAD)] = "mr",
655 [ilog2(VM_MAYWRITE)] = "mw",
656 [ilog2(VM_MAYEXEC)] = "me",
657 [ilog2(VM_MAYSHARE)] = "ms",
658 [ilog2(VM_GROWSDOWN)] = "gd",
659 [ilog2(VM_PFNMAP)] = "pf",
834f82e2
CG		 660 [ilog2(VM_LOCKED)] = "lo",
661 [ilog2(VM_IO)] = "io",
662 [ilog2(VM_SEQ_READ)] = "sr",
663 [ilog2(VM_RAND_READ)] = "rr",
664 [ilog2(VM_DONTCOPY)] = "dc",
665 [ilog2(VM_DONTEXPAND)] = "de",
8614d6c5 666 [ilog2(VM_LOCKONFAULT)] = "lf",
834f82e2
CG		 667 [ilog2(VM_ACCOUNT)] = "ac",
668 [ilog2(VM_NORESERVE)] = "nr",
669 [ilog2(VM_HUGETLB)] = "ht",
b6fb293f 670 [ilog2(VM_SYNC)] = "sf",
834f82e2 671 [ilog2(VM_ARCH_1)] = "ar",
d2cd9ede 672 [ilog2(VM_WIPEONFORK)] = "wf",
834f82e2 673 [ilog2(VM_DONTDUMP)] = "dd",
424037b7
DK		 674#ifdef CONFIG_ARM64_BTI
675 [ilog2(VM_ARM64_BTI)] = "bt",
676#endif
ec8e41ae
NH		 677#ifdef CONFIG_MEM_SOFT_DIRTY
678 [ilog2(VM_SOFTDIRTY)] = "sd",
679#endif
834f82e2
CG		 680 [ilog2(VM_MIXEDMAP)] = "mm",
681 [ilog2(VM_HUGEPAGE)] = "hg",
682 [ilog2(VM_NOHUGEPAGE)] = "nh",
683 [ilog2(VM_MERGEABLE)] = "mg",
16ba6f81
AA		 684 [ilog2(VM_UFFD_MISSING)]= "um",
685 [ilog2(VM_UFFD_WP)] = "uw",
9f341931
CM		 686#ifdef CONFIG_ARM64_MTE
687 [ilog2(VM_MTE)] = "mt",
688 [ilog2(VM_MTE_ALLOWED)] = "",
689#endif
5212213a 690#ifdef CONFIG_ARCH_HAS_PKEYS
c1192f84
DH		 691 /* These come out via ProtectionKey: */
692 [ilog2(VM_PKEY_BIT0)] = "",
693 [ilog2(VM_PKEY_BIT1)] = "",
694 [ilog2(VM_PKEY_BIT2)] = "",
695 [ilog2(VM_PKEY_BIT3)] = "",
2c9e0a6f
RP		 696#if VM_PKEY_BIT4
697 [ilog2(VM_PKEY_BIT4)] = "",
c1192f84 698#endif
5212213a 699#endif /* CONFIG_ARCH_HAS_PKEYS */
7677f7fd
AR		 700#ifdef CONFIG_HAVE_ARCH_USERFAULTFD_MINOR
701 [ilog2(VM_UFFD_MINOR)] = "ui",
702#endif /* CONFIG_HAVE_ARCH_USERFAULTFD_MINOR */
54007f81
YY		 703#ifdef CONFIG_X86_USER_SHADOW_STACK
704 [ilog2(VM_SHADOW_STACK)] = "ss",
705#endif
834f82e2
CG		 706 };
707 size_t i;
708
709 seq_puts(m, "VmFlags: ");
710 for (i = 0; i < BITS_PER_LONG; i++) {
c1192f84
DH		 711 if (!mnemonics[i][0])
712 continue;
834f82e2 713 if (vma->vm_flags & (1UL << i)) {
f6640663
AV		 714 seq_putc(m, mnemonics[i][0]);
715 seq_putc(m, mnemonics[i][1]);
716 seq_putc(m, ' ');
834f82e2
CG		 717 }
718 }
719 seq_putc(m, '\n');
720}
721
25ee01a2
NH		 722#ifdef CONFIG_HUGETLB_PAGE
723static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
724 unsigned long addr, unsigned long end,
725 struct mm_walk *walk)
726{
727 struct mem_size_stats *mss = walk->private;
728 struct vm_area_struct *vma = walk->vma;
729 struct page *page = NULL;
c33c7948 730 pte_t ptent = ptep_get(pte);
25ee01a2 731
c33c7948
RR		 732 if (pte_present(ptent)) {
733 page = vm_normal_page(vma, addr, ptent);
734 } else if (is_swap_pte(ptent)) {
735 swp_entry_t swpent = pte_to_swp_entry(ptent);
25ee01a2 736
af5cdaf8
AP		 737 if (is_pfn_swap_entry(swpent))
738 page = pfn_swap_entry_to_page(swpent);
25ee01a2
NH		 739 }
740 if (page) {
3489dbb6 741 if (page_mapcount(page) >= 2 || hugetlb_pmd_shared(pte))
25ee01a2
NH		 742 mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
743 else
744 mss->private_hugetlb += huge_page_size(hstate_vma(vma));
745 }
746 return 0;
747}
7b86ac33
CH		 748#else
749#define smaps_hugetlb_range NULL
25ee01a2
NH		 750#endif /* HUGETLB_PAGE */
751
7b86ac33
CH		 752static const struct mm_walk_ops smaps_walk_ops = {
753 .pmd_entry = smaps_pte_range,
754 .hugetlb_entry = smaps_hugetlb_range,
49b06385 755 .walk_lock = PGWALK_RDLOCK,
7b86ac33
CH		 756};
757
758static const struct mm_walk_ops smaps_shmem_walk_ops = {
759 .pmd_entry = smaps_pte_range,
760 .hugetlb_entry = smaps_hugetlb_range,
761 .pte_hole = smaps_pte_hole,
49b06385 762 .walk_lock = PGWALK_RDLOCK,
7b86ac33
CH		 763};
764
03b4b114
CC		 765/*
766 * Gather mem stats from @vma with the indicated beginning
767 * address @start, and keep them in @mss.
768 *
769 * Use vm_start of @vma as the beginning address if @start is 0.
770 */
8e68d689 771static void smap_gather_stats(struct vm_area_struct *vma,
03b4b114 772 struct mem_size_stats *mss, unsigned long start)
e070ad49 773{
03b4b114
CC		 774 const struct mm_walk_ops *ops = &smaps_walk_ops;
775
776 /* Invalid start */
777 if (start >= vma->vm_end)
778 return;
779
c261e7d9 780 if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
6a15a370
VB		 781 /*
782 * For shared or readonly shmem mappings we know that all
783 * swapped out pages belong to the shmem object, and we can
784 * obtain the swap value much more efficiently. For private
785 * writable mappings, we might have COW pages that are
786 * not affected by the parent swapped out pages of the shmem
787 * object, so we have to distinguish them during the page walk.
788 * Unless we know that the shmem object (or the part mapped by
789 * our VMA) has no swapped out pages at all.
790 */
791 unsigned long shmem_swapped = shmem_swap_usage(vma);
792
03b4b114
CC		 793 if (!start && (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
794 !(vma->vm_flags & VM_WRITE))) {
fa76da46 795 mss->swap += shmem_swapped;
6a15a370 796 } else {
03b4b114 797 ops = &smaps_shmem_walk_ops;
6a15a370 798 }
c261e7d9 799 }
b4aca547 800
c1e8d7c6 801 /* mmap_lock is held in m_start */
03b4b114
CC		 802 if (!start)
803 walk_page_vma(vma, ops, mss);
804 else
805 walk_page_range(vma->vm_mm, start, vma->vm_end, ops, mss);
8e68d689
VB		 806}
807
808#define SEQ_PUT_DEC(str, val) \
809 seq_put_decimal_ull_width(m, str, (val) >> 10, 8)
f1547959
VB		 810
811/* Show the contents common for smaps and smaps_rollup */
ee2ad71b
LS		 812static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss,
813 bool rollup_mode)
f1547959
VB		 814{
815 SEQ_PUT_DEC("Rss: ", mss->resident);
816 SEQ_PUT_DEC(" kB\nPss: ", mss->pss >> PSS_SHIFT);
30934843 817 SEQ_PUT_DEC(" kB\nPss_Dirty: ", mss->pss_dirty >> PSS_SHIFT);
ee2ad71b
LS		 818 if (rollup_mode) {
819 /*
820 * These are meaningful only for smaps_rollup, otherwise two of
821 * them are zero, and the other one is the same as Pss.
822 */
823 SEQ_PUT_DEC(" kB\nPss_Anon: ",
824 mss->pss_anon >> PSS_SHIFT);
825 SEQ_PUT_DEC(" kB\nPss_File: ",
826 mss->pss_file >> PSS_SHIFT);
827 SEQ_PUT_DEC(" kB\nPss_Shmem: ",
828 mss->pss_shmem >> PSS_SHIFT);
829 }
f1547959
VB		 830 SEQ_PUT_DEC(" kB\nShared_Clean: ", mss->shared_clean);
831 SEQ_PUT_DEC(" kB\nShared_Dirty: ", mss->shared_dirty);
832 SEQ_PUT_DEC(" kB\nPrivate_Clean: ", mss->private_clean);
833 SEQ_PUT_DEC(" kB\nPrivate_Dirty: ", mss->private_dirty);
834 SEQ_PUT_DEC(" kB\nReferenced: ", mss->referenced);
835 SEQ_PUT_DEC(" kB\nAnonymous: ", mss->anonymous);
8b479335 836 SEQ_PUT_DEC(" kB\nKSM: ", mss->ksm);
f1547959
VB		 837 SEQ_PUT_DEC(" kB\nLazyFree: ", mss->lazyfree);
838 SEQ_PUT_DEC(" kB\nAnonHugePages: ", mss->anonymous_thp);
839 SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp);
471e78cc 840 SEQ_PUT_DEC(" kB\nFilePmdMapped: ", mss->file_thp);
f1547959
VB		 841 SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb);
842 seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ",
843 mss->private_hugetlb >> 10, 7);
844 SEQ_PUT_DEC(" kB\nSwap: ", mss->swap);
845 SEQ_PUT_DEC(" kB\nSwapPss: ",
846 mss->swap_pss >> PSS_SHIFT);
847 SEQ_PUT_DEC(" kB\nLocked: ",
848 mss->pss_locked >> PSS_SHIFT);
849 seq_puts(m, " kB\n");
850}
851
8e68d689
VB		 852static int show_smap(struct seq_file *m, void *v)
853{
8e68d689 854 struct vm_area_struct *vma = v;
860a2e7f 855 struct mem_size_stats mss = {};
258f669e 856
03b4b114 857 smap_gather_stats(vma, &mss, 0);
258f669e
VB		 858
859 show_map_vma(m, vma);
860
861 SEQ_PUT_DEC("Size: ", vma->vm_end - vma->vm_start);
862 SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma));
863 SEQ_PUT_DEC(" kB\nMMUPageSize: ", vma_mmu_pagesize(vma));
864 seq_puts(m, " kB\n");
865
ee2ad71b 866 __show_smap(m, &mss, false);
258f669e 867
daa60ae6 868 seq_printf(m, "THPeligible: %8u\n",
3485b883
RR		 869 !!thp_vma_allowable_orders(vma, vma->vm_flags, true, false,
870 true, THP_ORDERS_ALL));
7635d9cb 871
258f669e
VB		 872 if (arch_pkeys_enabled())
873 seq_printf(m, "ProtectionKey: %8u\n", vma_pkey(vma));
874 show_smap_vma_flags(m, vma);
875
258f669e
VB		 876 return 0;
877}
878
879static int show_smaps_rollup(struct seq_file *m, void *v)
880{
881 struct proc_maps_private *priv = m->private;
860a2e7f 882 struct mem_size_stats mss = {};
c4c84f06 883 struct mm_struct *mm = priv->mm;
258f669e 884 struct vm_area_struct *vma;
c4c84f06 885 unsigned long vma_start = 0, last_vma_end = 0;
8e68d689 886 int ret = 0;
250cb40f 887 VMA_ITERATOR(vmi, mm, 0);
8e68d689 888
258f669e
VB		 889 priv->task = get_proc_task(priv->inode);
890 if (!priv->task)
891 return -ESRCH;
493b0e9d 892
258f669e
VB		 893 if (!mm || !mmget_not_zero(mm)) {
894 ret = -ESRCH;
895 goto out_put_task;
493b0e9d 896 }
4752c369 897
d8ed45c5 898 ret = mmap_read_lock_killable(mm);
a26a9781
KK		 899 if (ret)
900 goto out_put_mm;
901
258f669e 902 hold_task_mempolicy(priv);
250cb40f 903 vma = vma_next(&vmi);
f1547959 904
c4c84f06
MWO		 905 if (unlikely(!vma))
906 goto empty_set;
907
908 vma_start = vma->vm_start;
909 do {
03b4b114 910 smap_gather_stats(vma, &mss, 0);
258f669e 911 last_vma_end = vma->vm_end;
ff9f47f6
CC		 912
913 /*
914 * Release mmap_lock temporarily if someone wants to
915 * access it for write request.
916 */
917 if (mmap_lock_is_contended(mm)) {
250cb40f 918 vma_iter_invalidate(&vmi);
ff9f47f6
CC		 919 mmap_read_unlock(mm);
920 ret = mmap_read_lock_killable(mm);
921 if (ret) {
922 release_task_mempolicy(priv);
923 goto out_put_mm;
924 }
925
926 /*
927 * After dropping the lock, there are four cases to
928 * consider. See the following example for explanation.
929 *
930 * +------+------+-----------+
931 * | VMA1 | VMA2 | VMA3 |
932 * +------+------+-----------+
933 * | | | |
934 * 4k 8k 16k 400k
935 *
936 * Suppose we drop the lock after reading VMA2 due to
937 * contention, then we get:
938 *
939 * last_vma_end = 16k
940 *
941 * 1) VMA2 is freed, but VMA3 exists:
942 *
250cb40f 943 * vma_next(vmi) will return VMA3.
ff9f47f6
CC		 944 * In this case, just continue from VMA3.
945 *
946 * 2) VMA2 still exists:
947 *
250cb40f
LH		 948 * vma_next(vmi) will return VMA3.
949 * In this case, just continue from VMA3.
ff9f47f6
CC		 950 *
951 * 3) No more VMAs can be found:
952 *
250cb40f 953 * vma_next(vmi) will return NULL.
ff9f47f6
CC		 954 * No more things to do, just break.
955 *
956 * 4) (last_vma_end - 1) is the middle of a vma (VMA'):
957 *
250cb40f 958 * vma_next(vmi) will return VMA' whose range
ff9f47f6
CC		 959 * contains last_vma_end.
960 * Iterate VMA' from last_vma_end.
961 */
250cb40f 962 vma = vma_next(&vmi);
ff9f47f6
CC		 963 /* Case 3 above */
964 if (!vma)
965 break;
966
250cb40f 967 /* Case 1 and 2 above */
ff9f47f6
CC		 968 if (vma->vm_start >= last_vma_end)
969 continue;
970
971 /* Case 4 above */
972 if (vma->vm_end > last_vma_end)
973 smap_gather_stats(vma, &mss, last_vma_end);
974 }
250cb40f 975 } for_each_vma(vmi, vma);
258f669e 976
c4c84f06
MWO		 977empty_set:
978 show_vma_header_prefix(m, vma_start, last_vma_end, 0, 0, 0, 0);
258f669e
VB		 979 seq_pad(m, ' ');
980 seq_puts(m, "[rollup]\n");
981
ee2ad71b 982 __show_smap(m, &mss, true);
258f669e
VB		 983
984 release_task_mempolicy(priv);
d8ed45c5 985 mmap_read_unlock(mm);
258f669e 986
a26a9781
KK		 987out_put_mm:
988 mmput(mm);
258f669e
VB		 989out_put_task:
990 put_task_struct(priv->task);
991 priv->task = NULL;
992
493b0e9d 993 return ret;
e070ad49 994}
d1be35cb 995#undef SEQ_PUT_DEC
e070ad49 996
03a44825 997static const struct seq_operations proc_pid_smaps_op = {
a6198797
MM		 998 .start = m_start,
999 .next = m_next,
1000 .stop = m_stop,
871305bb 1001 .show = show_smap
a6198797
MM		 1002};
1003
b7643757 1004static int pid_smaps_open(struct inode *inode, struct file *file)
a6198797
MM		 1005{
1006 return do_maps_open(inode, file, &proc_pid_smaps_op);
1007}
1008
258f669e 1009static int smaps_rollup_open(struct inode *inode, struct file *file)
493b0e9d 1010{
258f669e 1011 int ret;
493b0e9d 1012 struct proc_maps_private *priv;
258f669e
VB		 1013
1014 priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT);
1015 if (!priv)
493b0e9d 1016 return -ENOMEM;
258f669e
VB		 1017
1018 ret = single_open(file, show_smaps_rollup, priv);
1019 if (ret)
1020 goto out_free;
1021
1022 priv->inode = inode;
1023 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
1024 if (IS_ERR(priv->mm)) {
1025 ret = PTR_ERR(priv->mm);
1026
1027 single_release(inode, file);
1028 goto out_free;
493b0e9d 1029 }
258f669e 1030
493b0e9d 1031 return 0;
258f669e
VB		 1032
1033out_free:
1034 kfree(priv);
1035 return ret;
1036}
1037
1038static int smaps_rollup_release(struct inode *inode, struct file *file)
1039{
1040 struct seq_file *seq = file->private_data;
1041 struct proc_maps_private *priv = seq->private;
1042
1043 if (priv->mm)
1044 mmdrop(priv->mm);
1045
1046 kfree(priv);
1047 return single_release(inode, file);
493b0e9d
DC		 1048}
1049
b7643757
SP		 1050const struct file_operations proc_pid_smaps_operations = {
1051 .open = pid_smaps_open,
1052 .read = seq_read,
1053 .llseek = seq_lseek,
29a40ace 1054 .release = proc_map_release,
b7643757
SP		 1055};
1056
493b0e9d 1057const struct file_operations proc_pid_smaps_rollup_operations = {
258f669e 1058 .open = smaps_rollup_open,
493b0e9d
DC		 1059 .read = seq_read,
1060 .llseek = seq_lseek,
258f669e 1061 .release = smaps_rollup_release,
493b0e9d
DC		 1062};
1063
040fa020
PE		 1064enum clear_refs_types {
1065 CLEAR_REFS_ALL = 1,
1066 CLEAR_REFS_ANON,
1067 CLEAR_REFS_MAPPED,
0f8975ec 1068 CLEAR_REFS_SOFT_DIRTY,
695f0559 1069 CLEAR_REFS_MM_HIWATER_RSS,
040fa020
PE		 1070 CLEAR_REFS_LAST,
1071};
1072
af9de7eb 1073struct clear_refs_private {
0f8975ec 1074 enum clear_refs_types type;
af9de7eb
PE		 1075};
1076
7d5b3bfa 1077#ifdef CONFIG_MEM_SOFT_DIRTY
9348b73c 1078
9348b73c
LT		 1079static inline bool pte_is_pinned(struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1080{
1081 struct page *page;
1082
1083 if (!pte_write(pte))
1084 return false;
1085 if (!is_cow_mapping(vma->vm_flags))
1086 return false;
a458b76a 1087 if (likely(!test_bit(MMF_HAS_PINNED, &vma->vm_mm->flags)))
9348b73c
LT		 1088 return false;
1089 page = vm_normal_page(vma, addr, pte);
1090 if (!page)
1091 return false;
1092 return page_maybe_dma_pinned(page);
1093}
1094
0f8975ec
PE		 1095static inline void clear_soft_dirty(struct vm_area_struct *vma,
1096 unsigned long addr, pte_t *pte)
1097{
0f8975ec
PE		 1098 /*
1099 * The soft-dirty tracker uses #PF-s to catch writes
1100 * to pages, so write-protect the pte as well. See the
1ad1335d 1101 * Documentation/admin-guide/mm/soft-dirty.rst for full description
0f8975ec
PE		 1102 * of how soft-dirty works.
1103 */
c33c7948 1104 pte_t ptent = ptep_get(pte);
179ef71c
CG		 1105
1106 if (pte_present(ptent)) {
04a86453
AK		 1107 pte_t old_pte;
1108
9348b73c
LT		 1109 if (pte_is_pinned(vma, addr, ptent))
1110 return;
04a86453
AK		 1111 old_pte = ptep_modify_prot_start(vma, addr, pte);
1112 ptent = pte_wrprotect(old_pte);
a7b76174 1113 ptent = pte_clear_soft_dirty(ptent);
04a86453 1114 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent);
179ef71c
CG		 1115 } else if (is_swap_pte(ptent)) {
1116 ptent = pte_swp_clear_soft_dirty(ptent);
326c2597 1117 set_pte_at(vma->vm_mm, addr, pte, ptent);
179ef71c 1118 }
0f8975ec 1119}
5d3875a0
LD		 1120#else
1121static inline void clear_soft_dirty(struct vm_area_struct *vma,
1122 unsigned long addr, pte_t *pte)
1123{
1124}
1125#endif
0f8975ec 1126
5d3875a0 1127#if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
7d5b3bfa
KS		 1128static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1129 unsigned long addr, pmd_t *pmdp)
1130{
a3cf988f 1131 pmd_t old, pmd = *pmdp;
5b7abeae 1132
ab6e3d09
NH		 1133 if (pmd_present(pmd)) {
1134 /* See comment in change_huge_pmd() */
a3cf988f
KS		 1135 old = pmdp_invalidate(vma, addr, pmdp);
1136 if (pmd_dirty(old))
ab6e3d09 1137 pmd = pmd_mkdirty(pmd);
a3cf988f 1138 if (pmd_young(old))
ab6e3d09
NH		 1139 pmd = pmd_mkyoung(pmd);
1140
1141 pmd = pmd_wrprotect(pmd);
1142 pmd = pmd_clear_soft_dirty(pmd);
1143
1144 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1145 } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1146 pmd = pmd_swp_clear_soft_dirty(pmd);
1147 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1148 }
7d5b3bfa 1149}
7d5b3bfa 1150#else
7d5b3bfa
KS		 1151static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1152 unsigned long addr, pmd_t *pmdp)
1153{
1154}
1155#endif
1156
a6198797 1157static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
2165009b 1158 unsigned long end, struct mm_walk *walk)
a6198797 1159{
af9de7eb 1160 struct clear_refs_private *cp = walk->private;
5c64f52a 1161 struct vm_area_struct *vma = walk->vma;
a6198797
MM		 1162 pte_t *pte, ptent;
1163 spinlock_t *ptl;
1164 struct page *page;
1165
b6ec57f4
KS		 1166 ptl = pmd_trans_huge_lock(pmd, vma);
1167 if (ptl) {
7d5b3bfa
KS		 1168 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1169 clear_soft_dirty_pmd(vma, addr, pmd);
1170 goto out;
1171 }
1172
84c3fc4e
ZY		 1173 if (!pmd_present(*pmd))
1174 goto out;
1175
7d5b3bfa
KS		 1176 page = pmd_page(*pmd);
1177
1178 /* Clear accessed and referenced bits. */
1179 pmdp_test_and_clear_young(vma, addr, pmd);
33c3fc71 1180 test_and_clear_page_young(page);
7d5b3bfa
KS		 1181 ClearPageReferenced(page);
1182out:
1183 spin_unlock(ptl);
1184 return 0;
1185 }
1186
a6198797 1187 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
7780d040
HD		 1188 if (!pte) {
1189 walk->action = ACTION_AGAIN;
1190 return 0;
1191 }
a6198797 1192 for (; addr != end; pte++, addr += PAGE_SIZE) {
c33c7948 1193 ptent = ptep_get(pte);
a6198797 1194
0f8975ec
PE		 1195 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1196 clear_soft_dirty(vma, addr, pte);
1197 continue;
1198 }
1199
179ef71c
CG		 1200 if (!pte_present(ptent))
1201 continue;
1202
a6198797
MM		 1203 page = vm_normal_page(vma, addr, ptent);
1204 if (!page)
1205 continue;
1206
1207 /* Clear accessed and referenced bits. */
1208 ptep_test_and_clear_young(vma, addr, pte);
33c3fc71 1209 test_and_clear_page_young(page);
a6198797
MM		 1210 ClearPageReferenced(page);
1211 }
1212 pte_unmap_unlock(pte - 1, ptl);
1213 cond_resched();
1214 return 0;
1215}
1216
5c64f52a
NH		 1217static int clear_refs_test_walk(unsigned long start, unsigned long end,
1218 struct mm_walk *walk)
1219{
1220 struct clear_refs_private *cp = walk->private;
1221 struct vm_area_struct *vma = walk->vma;
1222
48684a65
NH		 1223 if (vma->vm_flags & VM_PFNMAP)
1224 return 1;
1225
5c64f52a
NH		 1226 /*
1227 * Writing 1 to /proc/pid/clear_refs affects all pages.
1228 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1229 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1230 * Writing 4 to /proc/pid/clear_refs affects all pages.
1231 */
1232 if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1233 return 1;
1234 if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1235 return 1;
1236 return 0;
1237}
1238
7b86ac33
CH		 1239static const struct mm_walk_ops clear_refs_walk_ops = {
1240 .pmd_entry = clear_refs_pte_range,
1241 .test_walk = clear_refs_test_walk,
49b06385 1242 .walk_lock = PGWALK_WRLOCK,
7b86ac33
CH		 1243};
1244
f248dcb3
MM		 1245static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1246 size_t count, loff_t *ppos)
b813e931 1247{
f248dcb3 1248 struct task_struct *task;
860a2e7f 1249 char buffer[PROC_NUMBUF] = {};
f248dcb3 1250 struct mm_struct *mm;
b813e931 1251 struct vm_area_struct *vma;
040fa020
PE		 1252 enum clear_refs_types type;
1253 int itype;
0a8cb8e3 1254 int rv;
b813e931 1255
f248dcb3
MM		 1256 if (count > sizeof(buffer) - 1)
1257 count = sizeof(buffer) - 1;
1258 if (copy_from_user(buffer, buf, count))
1259 return -EFAULT;
040fa020 1260 rv = kstrtoint(strstrip(buffer), 10, &itype);
0a8cb8e3
AD		 1261 if (rv < 0)
1262 return rv;
040fa020
PE		 1263 type = (enum clear_refs_types)itype;
1264 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
f248dcb3 1265 return -EINVAL;
541c237c 1266
496ad9aa 1267 task = get_proc_task(file_inode(file));
f248dcb3
MM		 1268 if (!task)
1269 return -ESRCH;
1270 mm = get_task_mm(task);
1271 if (mm) {
250cb40f 1272 VMA_ITERATOR(vmi, mm, 0);
ac46d4f3 1273 struct mmu_notifier_range range;
af9de7eb 1274 struct clear_refs_private cp = {
0f8975ec 1275 .type = type,
af9de7eb 1276 };
695f0559 1277
29a951df
LT		 1278 if (mmap_write_lock_killable(mm)) {
1279 count = -EINTR;
1280 goto out_mm;
1281 }
695f0559
PC		 1282 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1283 /*
1284 * Writing 5 to /proc/pid/clear_refs resets the peak
1285 * resident set size to this mm's current rss value.
1286 */
695f0559 1287 reset_mm_hiwater_rss(mm);
29a951df 1288 goto out_unlock;
695f0559
PC		 1289 }
1290
64e45507 1291 if (type == CLEAR_REFS_SOFT_DIRTY) {
250cb40f 1292 for_each_vma(vmi, vma) {
64e45507
PF		 1293 if (!(vma->vm_flags & VM_SOFTDIRTY))
1294 continue;
1c71222e 1295 vm_flags_clear(vma, VM_SOFTDIRTY);
29a951df 1296 vma_set_page_prot(vma);
64e45507 1297 }
ac46d4f3 1298
912efa17 1299 inc_tlb_flush_pending(mm);
7269f999 1300 mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY,
7d4a8be0 1301 0, mm, 0, -1UL);
ac46d4f3 1302 mmu_notifier_invalidate_range_start(&range);
64e45507 1303 }
c4c84f06 1304 walk_page_range(mm, 0, -1, &clear_refs_walk_ops, &cp);
912efa17 1305 if (type == CLEAR_REFS_SOFT_DIRTY) {
ac46d4f3 1306 mmu_notifier_invalidate_range_end(&range);
912efa17
WD		 1307 flush_tlb_mm(mm);
1308 dec_tlb_flush_pending(mm);
1309 }
29a951df
LT		 1310out_unlock:
1311 mmap_write_unlock(mm);
695f0559 1312out_mm:
f248dcb3
MM		 1313 mmput(mm);
1314 }
1315 put_task_struct(task);
fb92a4b0
VL		 1316
1317 return count;
b813e931
DR		 1318}
1319
f248dcb3
MM		 1320const struct file_operations proc_clear_refs_operations = {
1321 .write = clear_refs_write,
6038f373 1322 .llseek = noop_llseek,
f248dcb3
MM		 1323};
1324
092b50ba
NH		 1325typedef struct {
1326 u64 pme;
1327} pagemap_entry_t;
1328
85863e47 1329struct pagemapread {
8c829622 1330 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */
092b50ba 1331 pagemap_entry_t *buffer;
1c90308e 1332 bool show_pfn;
85863e47
MM		 1333};
1334
5aaabe83
NH		 1335#define PAGEMAP_WALK_SIZE (PMD_SIZE)
1336#define PAGEMAP_WALK_MASK (PMD_MASK)
1337
deb94544
KK		 1338#define PM_ENTRY_BYTES sizeof(pagemap_entry_t)
1339#define PM_PFRAME_BITS 55
1340#define PM_PFRAME_MASK GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1341#define PM_SOFT_DIRTY BIT_ULL(55)
77bb499b 1342#define PM_MMAP_EXCLUSIVE BIT_ULL(56)
fb8e37f3 1343#define PM_UFFD_WP BIT_ULL(57)
deb94544
KK		 1344#define PM_FILE BIT_ULL(61)
1345#define PM_SWAP BIT_ULL(62)
1346#define PM_PRESENT BIT_ULL(63)
1347
85863e47
MM		 1348#define PM_END_OF_BUFFER 1
1349
deb94544 1350static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
092b50ba 1351{
deb94544 1352 return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
092b50ba
NH		 1353}
1354
cabbb6d5 1355static int add_to_pagemap(pagemap_entry_t *pme, struct pagemapread *pm)
85863e47 1356{
092b50ba 1357 pm->buffer[pm->pos++] = *pme;
d82ef020 1358 if (pm->pos >= pm->len)
aae8679b 1359 return PM_END_OF_BUFFER;
85863e47
MM		 1360 return 0;
1361}
1362
1363static int pagemap_pte_hole(unsigned long start, unsigned long end,
b7a16c7a 1364 __always_unused int depth, struct mm_walk *walk)
85863e47 1365{
2165009b 1366 struct pagemapread *pm = walk->private;
68b5a652 1367 unsigned long addr = start;
85863e47 1368 int err = 0;
092b50ba 1369
68b5a652
PF		 1370 while (addr < end) {
1371 struct vm_area_struct *vma = find_vma(walk->mm, addr);
deb94544 1372 pagemap_entry_t pme = make_pme(0, 0);
87e6d49a
PF		 1373 /* End of address space hole, which we mark as non-present. */
1374 unsigned long hole_end;
68b5a652 1375
87e6d49a
PF		 1376 if (vma)
1377 hole_end = min(end, vma->vm_start);
1378 else
1379 hole_end = end;
1380
1381 for (; addr < hole_end; addr += PAGE_SIZE) {
cabbb6d5 1382 err = add_to_pagemap(&pme, pm);
87e6d49a
PF		 1383 if (err)
1384 goto out;
68b5a652
PF		 1385 }
1386
87e6d49a
PF		 1387 if (!vma)
1388 break;
1389
1390 /* Addresses in the VMA. */
1391 if (vma->vm_flags & VM_SOFTDIRTY)
deb94544 1392 pme = make_pme(0, PM_SOFT_DIRTY);
87e6d49a 1393 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
cabbb6d5 1394 err = add_to_pagemap(&pme, pm);
68b5a652
PF		 1395 if (err)
1396 goto out;
1397 }
85863e47 1398 }
68b5a652 1399out:
85863e47
MM		 1400 return err;
1401}
1402
deb94544 1403static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
052fb0d6 1404 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
85863e47 1405{
deb94544 1406 u64 frame = 0, flags = 0;
052fb0d6 1407 struct page *page = NULL;
24d7275c 1408 bool migration = false;
85863e47 1409
052fb0d6 1410 if (pte_present(pte)) {
1c90308e
KK		 1411 if (pm->show_pfn)
1412 frame = pte_pfn(pte);
deb94544 1413 flags |= PM_PRESENT;
25b2995a 1414 page = vm_normal_page(vma, addr, pte);
e9cdd6e7 1415 if (pte_soft_dirty(pte))
deb94544 1416 flags |= PM_SOFT_DIRTY;
fb8e37f3
PX		 1417 if (pte_uffd_wp(pte))
1418 flags |= PM_UFFD_WP;
052fb0d6 1419 } else if (is_swap_pte(pte)) {
179ef71c
CG		 1420 swp_entry_t entry;
1421 if (pte_swp_soft_dirty(pte))
deb94544 1422 flags |= PM_SOFT_DIRTY;
fb8e37f3
PX		 1423 if (pte_swp_uffd_wp(pte))
1424 flags |= PM_UFFD_WP;
179ef71c 1425 entry = pte_to_swp_entry(pte);
0d206b5d
PX		 1426 if (pm->show_pfn) {
1427 pgoff_t offset;
1428 /*
1429 * For PFN swap offsets, keeping the offset field
1430 * to be PFN only to be compatible with old smaps.
1431 */
1432 if (is_pfn_swap_entry(entry))
1433 offset = swp_offset_pfn(entry);
1434 else
1435 offset = swp_offset(entry);
ab6ecf24 1436 frame = swp_type(entry) |
0d206b5d
PX		 1437 (offset << MAX_SWAPFILES_SHIFT);
1438 }
deb94544 1439 flags |= PM_SWAP;
24d7275c 1440 migration = is_migration_entry(entry);
af5cdaf8
AP		 1441 if (is_pfn_swap_entry(entry))
1442 page = pfn_swap_entry_to_page(entry);
8e165e73
PX		 1443 if (pte_marker_entry_uffd_wp(entry))
1444 flags |= PM_UFFD_WP;
052fb0d6
KK		 1445 }
1446
1447 if (page && !PageAnon(page))
1448 flags |= PM_FILE;
24d7275c 1449 if (page && !migration && page_mapcount(page) == 1)
77bb499b 1450 flags |= PM_MMAP_EXCLUSIVE;
deb94544
KK		 1451 if (vma->vm_flags & VM_SOFTDIRTY)
1452 flags |= PM_SOFT_DIRTY;
052fb0d6 1453
deb94544 1454 return make_pme(frame, flags);
bcf8039e
DH		 1455}
1456
356515e7 1457static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
2165009b 1458 struct mm_walk *walk)
85863e47 1459{
f995ece2 1460 struct vm_area_struct *vma = walk->vma;
2165009b 1461 struct pagemapread *pm = walk->private;
bf929152 1462 spinlock_t *ptl;
05fbf357 1463 pte_t *pte, *orig_pte;
85863e47 1464 int err = 0;
356515e7 1465#ifdef CONFIG_TRANSPARENT_HUGEPAGE
24d7275c
YS		 1466 bool migration = false;
1467
b6ec57f4
KS		 1468 ptl = pmd_trans_huge_lock(pmdp, vma);
1469 if (ptl) {
356515e7
KK		 1470 u64 flags = 0, frame = 0;
1471 pmd_t pmd = *pmdp;
84c3fc4e 1472 struct page *page = NULL;
0f8975ec 1473
b83d7e43 1474 if (vma->vm_flags & VM_SOFTDIRTY)
deb94544 1475 flags |= PM_SOFT_DIRTY;
d9104d1c 1476
356515e7 1477 if (pmd_present(pmd)) {
84c3fc4e 1478 page = pmd_page(pmd);
77bb499b 1479
356515e7 1480 flags |= PM_PRESENT;
b83d7e43
HY		 1481 if (pmd_soft_dirty(pmd))
1482 flags |= PM_SOFT_DIRTY;
fb8e37f3
PX		 1483 if (pmd_uffd_wp(pmd))
1484 flags |= PM_UFFD_WP;
1c90308e
KK		 1485 if (pm->show_pfn)
1486 frame = pmd_pfn(pmd) +
1487 ((addr & ~PMD_MASK) >> PAGE_SHIFT);
356515e7 1488 }
84c3fc4e
ZY		 1489#ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1490 else if (is_swap_pmd(pmd)) {
1491 swp_entry_t entry = pmd_to_swp_entry(pmd);
ab6ecf24 1492 unsigned long offset;
84c3fc4e 1493
ab6ecf24 1494 if (pm->show_pfn) {
0d206b5d
PX		 1495 if (is_pfn_swap_entry(entry))
1496 offset = swp_offset_pfn(entry);
1497 else
1498 offset = swp_offset(entry);
1499 offset = offset +
ab6ecf24
HY		 1500 ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1501 frame = swp_type(entry) |
1502 (offset << MAX_SWAPFILES_SHIFT);
1503 }
84c3fc4e 1504 flags |= PM_SWAP;
b83d7e43
HY		 1505 if (pmd_swp_soft_dirty(pmd))
1506 flags |= PM_SOFT_DIRTY;
fb8e37f3
PX		 1507 if (pmd_swp_uffd_wp(pmd))
1508 flags |= PM_UFFD_WP;
84c3fc4e 1509 VM_BUG_ON(!is_pmd_migration_entry(pmd));
24d7275c 1510 migration = is_migration_entry(entry);
af5cdaf8 1511 page = pfn_swap_entry_to_page(entry);
84c3fc4e
ZY		 1512 }
1513#endif
1514
24d7275c 1515 if (page && !migration && page_mapcount(page) == 1)
84c3fc4e 1516 flags |= PM_MMAP_EXCLUSIVE;
356515e7 1517
025c5b24 1518 for (; addr != end; addr += PAGE_SIZE) {
356515e7 1519 pagemap_entry_t pme = make_pme(frame, flags);
025c5b24 1520
cabbb6d5 1521 err = add_to_pagemap(&pme, pm);
025c5b24
NH		 1522 if (err)
1523 break;
ab6ecf24
HY		 1524 if (pm->show_pfn) {
1525 if (flags & PM_PRESENT)
1526 frame++;
1527 else if (flags & PM_SWAP)
1528 frame += (1 << MAX_SWAPFILES_SHIFT);
1529 }
5aaabe83 1530 }
bf929152 1531 spin_unlock(ptl);
025c5b24 1532 return err;
5aaabe83 1533 }
356515e7 1534#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
81d0fa62 1535
f995ece2
NH		 1536 /*
1537 * We can assume that @vma always points to a valid one and @end never
1538 * goes beyond vma->vm_end.
1539 */
356515e7 1540 orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
7780d040
HD		 1541 if (!pte) {
1542 walk->action = ACTION_AGAIN;
1543 return err;
1544 }
f995ece2
NH		 1545 for (; addr < end; pte++, addr += PAGE_SIZE) {
1546 pagemap_entry_t pme;
05fbf357 1547
c33c7948 1548 pme = pte_to_pagemap_entry(pm, vma, addr, ptep_get(pte));
cabbb6d5 1549 err = add_to_pagemap(&pme, pm);
05fbf357 1550 if (err)
81d0fa62 1551 break;
85863e47 1552 }
f995ece2 1553 pte_unmap_unlock(orig_pte, ptl);
85863e47
MM		 1554
1555 cond_resched();
1556
1557 return err;
1558}
1559
1a5cb814 1560#ifdef CONFIG_HUGETLB_PAGE
116354d1 1561/* This function walks within one hugetlb entry in the single call */
356515e7 1562static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
116354d1
NH		 1563 unsigned long addr, unsigned long end,
1564 struct mm_walk *walk)
5dc37642 1565{
5dc37642 1566 struct pagemapread *pm = walk->private;
f995ece2 1567 struct vm_area_struct *vma = walk->vma;
356515e7 1568 u64 flags = 0, frame = 0;
5dc37642 1569 int err = 0;
356515e7 1570 pte_t pte;
5dc37642 1571
f995ece2 1572 if (vma->vm_flags & VM_SOFTDIRTY)
deb94544 1573 flags |= PM_SOFT_DIRTY;
d9104d1c 1574
356515e7
KK		 1575 pte = huge_ptep_get(ptep);
1576 if (pte_present(pte)) {
1577 struct page *page = pte_page(pte);
1578
1579 if (!PageAnon(page))
1580 flags |= PM_FILE;
1581
77bb499b
KK		 1582 if (page_mapcount(page) == 1)
1583 flags |= PM_MMAP_EXCLUSIVE;
1584
8e165e73
PX		 1585 if (huge_pte_uffd_wp(pte))
1586 flags |= PM_UFFD_WP;
1587
356515e7 1588 flags |= PM_PRESENT;
1c90308e
KK		 1589 if (pm->show_pfn)
1590 frame = pte_pfn(pte) +
1591 ((addr & ~hmask) >> PAGE_SHIFT);
8e165e73
PX		 1592 } else if (pte_swp_uffd_wp_any(pte)) {
1593 flags |= PM_UFFD_WP;
356515e7
KK		 1594 }
1595
5dc37642 1596 for (; addr != end; addr += PAGE_SIZE) {
356515e7
KK		 1597 pagemap_entry_t pme = make_pme(frame, flags);
1598
cabbb6d5 1599 err = add_to_pagemap(&pme, pm);
5dc37642
NH		 1600 if (err)
1601 return err;
1c90308e 1602 if (pm->show_pfn && (flags & PM_PRESENT))
356515e7 1603 frame++;
5dc37642
NH		 1604 }
1605
1606 cond_resched();
1607
1608 return err;
1609}
7b86ac33
CH		 1610#else
1611#define pagemap_hugetlb_range NULL
1a5cb814 1612#endif /* HUGETLB_PAGE */
5dc37642 1613
7b86ac33
CH		 1614static const struct mm_walk_ops pagemap_ops = {
1615 .pmd_entry = pagemap_pmd_range,
1616 .pte_hole = pagemap_pte_hole,
1617 .hugetlb_entry = pagemap_hugetlb_range,
49b06385 1618 .walk_lock = PGWALK_RDLOCK,
7b86ac33
CH		 1619};
1620
85863e47
MM		 1621/*
1622 * /proc/pid/pagemap - an array mapping virtual pages to pfns
1623 *
f16278c6
HR		 1624 * For each page in the address space, this file contains one 64-bit entry
1625 * consisting of the following:
1626 *
052fb0d6 1627 * Bits 0-54 page frame number (PFN) if present
f16278c6 1628 * Bits 0-4 swap type if swapped
052fb0d6 1629 * Bits 5-54 swap offset if swapped
1ad1335d 1630 * Bit 55 pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst)
77bb499b 1631 * Bit 56 page exclusively mapped
dd21bfa4
YZ		 1632 * Bit 57 pte is uffd-wp write-protected
1633 * Bits 58-60 zero
052fb0d6 1634 * Bit 61 page is file-page or shared-anon
f16278c6
HR		 1635 * Bit 62 page swapped
1636 * Bit 63 page present
1637 *
1638 * If the page is not present but in swap, then the PFN contains an
1639 * encoding of the swap file number and the page's offset into the
1640 * swap. Unmapped pages return a null PFN. This allows determining
85863e47
MM		 1641 * precisely which pages are mapped (or in swap) and comparing mapped
1642 * pages between processes.
1643 *
1644 * Efficient users of this interface will use /proc/pid/maps to
1645 * determine which areas of memory are actually mapped and llseek to
1646 * skip over unmapped regions.
1647 */
1648static ssize_t pagemap_read(struct file *file, char __user *buf,
1649 size_t count, loff_t *ppos)
1650{
a06db751 1651 struct mm_struct *mm = file->private_data;
85863e47 1652 struct pagemapread pm;
5d7e0d2b
AM		 1653 unsigned long src;
1654 unsigned long svpfn;
1655 unsigned long start_vaddr;
1656 unsigned long end_vaddr;
a06db751 1657 int ret = 0, copied = 0;
85863e47 1658
388f7934 1659 if (!mm || !mmget_not_zero(mm))
85863e47
MM		 1660 goto out;
1661
85863e47
MM		 1662 ret = -EINVAL;
1663 /* file position must be aligned */
aae8679b 1664 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
a06db751 1665 goto out_mm;
85863e47
MM		 1666
1667 ret = 0;
08161786 1668 if (!count)
a06db751 1669 goto out_mm;
08161786 1670
1c90308e
KK		 1671 /* do not disclose physical addresses: attack vector */
1672 pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1673
8c829622 1674 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
6da2ec56 1675 pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL);
5d7e0d2b 1676 ret = -ENOMEM;
d82ef020 1677 if (!pm.buffer)
a06db751 1678 goto out_mm;
85863e47 1679
5d7e0d2b
AM		 1680 src = *ppos;
1681 svpfn = src / PM_ENTRY_BYTES;
a06db751 1682 end_vaddr = mm->task_size;
5d7e0d2b
AM		 1683
1684 /* watch out for wraparound */
40d6366e 1685 start_vaddr = end_vaddr;
428e106a 1686 if (svpfn <= (ULONG_MAX >> PAGE_SHIFT)) {
7bab8dfb
YX		 1687 unsigned long end;
1688
428e106a
KS		 1689 ret = mmap_read_lock_killable(mm);
1690 if (ret)
1691 goto out_free;
1692 start_vaddr = untagged_addr_remote(mm, svpfn << PAGE_SHIFT);
1693 mmap_read_unlock(mm);
7bab8dfb
YX		 1694
1695 end = start_vaddr + ((count / PM_ENTRY_BYTES) << PAGE_SHIFT);
1696 if (end >= start_vaddr && end < mm->task_size)
1697 end_vaddr = end;
428e106a 1698 }
40d6366e
MC		 1699
1700 /* Ensure the address is inside the task */
1701 if (start_vaddr > mm->task_size)
5d7e0d2b
AM		 1702 start_vaddr = end_vaddr;
1703
d82ef020
KH		 1704 ret = 0;
1705 while (count && (start_vaddr < end_vaddr)) {
1706 int len;
1707 unsigned long end;
1708
1709 pm.pos = 0;
ea251c1d 1710 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
d82ef020
KH		 1711 /* overflow ? */
1712 if (end < start_vaddr || end > end_vaddr)
1713 end = end_vaddr;
d8ed45c5 1714 ret = mmap_read_lock_killable(mm);
ad80b932
KK		 1715 if (ret)
1716 goto out_free;
7b86ac33 1717 ret = walk_page_range(mm, start_vaddr, end, &pagemap_ops, &pm);
d8ed45c5 1718 mmap_read_unlock(mm);
d82ef020
KH		 1719 start_vaddr = end;
1720
1721 len = min(count, PM_ENTRY_BYTES * pm.pos);
309361e0 1722 if (copy_to_user(buf, pm.buffer, len)) {
d82ef020 1723 ret = -EFAULT;
a06db751 1724 goto out_free;
d82ef020
KH		 1725 }
1726 copied += len;
1727 buf += len;
1728 count -= len;
85863e47 1729 }
d82ef020
KH		 1730 *ppos += copied;
1731 if (!ret || ret == PM_END_OF_BUFFER)
1732 ret = copied;
1733
98bc93e5
KM		 1734out_free:
1735 kfree(pm.buffer);
a06db751
KK		 1736out_mm:
1737 mmput(mm);
85863e47
MM		 1738out:
1739 return ret;
1740}
1741
541c237c
PE		 1742static int pagemap_open(struct inode *inode, struct file *file)
1743{
a06db751
KK		 1744 struct mm_struct *mm;
1745
a06db751
KK		 1746 mm = proc_mem_open(inode, PTRACE_MODE_READ);
1747 if (IS_ERR(mm))
1748 return PTR_ERR(mm);
1749 file->private_data = mm;
1750 return 0;
1751}
1752
1753static int pagemap_release(struct inode *inode, struct file *file)
1754{
1755 struct mm_struct *mm = file->private_data;
1756
1757 if (mm)
1758 mmdrop(mm);
541c237c
PE		 1759 return 0;
1760}
1761
52526ca7
MUA		 1762#define PM_SCAN_CATEGORIES (PAGE_IS_WPALLOWED | PAGE_IS_WRITTEN | \
1763 PAGE_IS_FILE | PAGE_IS_PRESENT | \
1764 PAGE_IS_SWAPPED | PAGE_IS_PFNZERO | \
e6a9a2cb 1765 PAGE_IS_HUGE | PAGE_IS_SOFT_DIRTY)
52526ca7
MUA		 1766#define PM_SCAN_FLAGS (PM_SCAN_WP_MATCHING | PM_SCAN_CHECK_WPASYNC)
1767
1768struct pagemap_scan_private {
1769 struct pm_scan_arg arg;
1770 unsigned long masks_of_interest, cur_vma_category;
1771 struct page_region *vec_buf;
1772 unsigned long vec_buf_len, vec_buf_index, found_pages;
1773 struct page_region __user *vec_out;
1774};
1775
1776static unsigned long pagemap_page_category(struct pagemap_scan_private *p,
1777 struct vm_area_struct *vma,
1778 unsigned long addr, pte_t pte)
1779{
1780 unsigned long categories = 0;
1781
1782 if (pte_present(pte)) {
1783 struct page *page;
1784
1785 categories |= PAGE_IS_PRESENT;
1786 if (!pte_uffd_wp(pte))
1787 categories |= PAGE_IS_WRITTEN;
1788
1789 if (p->masks_of_interest & PAGE_IS_FILE) {
1790 page = vm_normal_page(vma, addr, pte);
1791 if (page && !PageAnon(page))
1792 categories |= PAGE_IS_FILE;
1793 }
1794
1795 if (is_zero_pfn(pte_pfn(pte)))
1796 categories |= PAGE_IS_PFNZERO;
e6a9a2cb
AV		 1797 if (pte_soft_dirty(pte))
1798 categories |= PAGE_IS_SOFT_DIRTY;
52526ca7
MUA		 1799 } else if (is_swap_pte(pte)) {
1800 swp_entry_t swp;
1801
1802 categories |= PAGE_IS_SWAPPED;
1803 if (!pte_swp_uffd_wp_any(pte))
1804 categories |= PAGE_IS_WRITTEN;
1805
1806 if (p->masks_of_interest & PAGE_IS_FILE) {
1807 swp = pte_to_swp_entry(pte);
1808 if (is_pfn_swap_entry(swp) &&
71014224 1809 !folio_test_anon(pfn_swap_entry_folio(swp)))
52526ca7
MUA		 1810 categories |= PAGE_IS_FILE;
1811 }
e6a9a2cb
AV		 1812 if (pte_swp_soft_dirty(pte))
1813 categories |= PAGE_IS_SOFT_DIRTY;
52526ca7
MUA		 1814 }
1815
1816 return categories;
1817}
1818
1819static void make_uffd_wp_pte(struct vm_area_struct *vma,
1820 unsigned long addr, pte_t *pte)
1821{
1822 pte_t ptent = ptep_get(pte);
1823
1824 if (pte_present(ptent)) {
1825 pte_t old_pte;
1826
1827 old_pte = ptep_modify_prot_start(vma, addr, pte);
c70dce49 1828 ptent = pte_mkuffd_wp(old_pte);
52526ca7
MUA		 1829 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent);
1830 } else if (is_swap_pte(ptent)) {
1831 ptent = pte_swp_mkuffd_wp(ptent);
1832 set_pte_at(vma->vm_mm, addr, pte, ptent);
1833 } else {
1834 set_pte_at(vma->vm_mm, addr, pte,
1835 make_pte_marker(PTE_MARKER_UFFD_WP));
1836 }
1837}
1838
1839#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1840static unsigned long pagemap_thp_category(struct pagemap_scan_private *p,
1841 struct vm_area_struct *vma,
1842 unsigned long addr, pmd_t pmd)
1843{
1844 unsigned long categories = PAGE_IS_HUGE;
1845
1846 if (pmd_present(pmd)) {
1847 struct page *page;
1848
1849 categories |= PAGE_IS_PRESENT;
1850 if (!pmd_uffd_wp(pmd))
1851 categories |= PAGE_IS_WRITTEN;
1852
1853 if (p->masks_of_interest & PAGE_IS_FILE) {
1854 page = vm_normal_page_pmd(vma, addr, pmd);
1855 if (page && !PageAnon(page))
1856 categories |= PAGE_IS_FILE;
1857 }
1858
1859 if (is_zero_pfn(pmd_pfn(pmd)))
1860 categories |= PAGE_IS_PFNZERO;
e6a9a2cb
AV		 1861 if (pmd_soft_dirty(pmd))
1862 categories |= PAGE_IS_SOFT_DIRTY;
52526ca7
MUA		 1863 } else if (is_swap_pmd(pmd)) {
1864 swp_entry_t swp;
1865
1866 categories |= PAGE_IS_SWAPPED;
1867 if (!pmd_swp_uffd_wp(pmd))
1868 categories |= PAGE_IS_WRITTEN;
e6a9a2cb
AV		 1869 if (pmd_swp_soft_dirty(pmd))
1870 categories |= PAGE_IS_SOFT_DIRTY;
52526ca7
MUA		 1871
1872 if (p->masks_of_interest & PAGE_IS_FILE) {
1873 swp = pmd_to_swp_entry(pmd);
1874 if (is_pfn_swap_entry(swp) &&
71014224 1875 !folio_test_anon(pfn_swap_entry_folio(swp)))
52526ca7
MUA		 1876 categories |= PAGE_IS_FILE;
1877 }
1878 }
1879
1880 return categories;
1881}
1882
1883static void make_uffd_wp_pmd(struct vm_area_struct *vma,
1884 unsigned long addr, pmd_t *pmdp)
1885{
1886 pmd_t old, pmd = *pmdp;
1887
1888 if (pmd_present(pmd)) {
1889 old = pmdp_invalidate_ad(vma, addr, pmdp);
1890 pmd = pmd_mkuffd_wp(old);
1891 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1892 } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1893 pmd = pmd_swp_mkuffd_wp(pmd);
1894 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1895 }
1896}
1897#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1898
1899#ifdef CONFIG_HUGETLB_PAGE
1900static unsigned long pagemap_hugetlb_category(pte_t pte)
1901{
1902 unsigned long categories = PAGE_IS_HUGE;
1903
1904 /*
1905 * According to pagemap_hugetlb_range(), file-backed HugeTLB
1906 * page cannot be swapped. So PAGE_IS_FILE is not checked for
1907 * swapped pages.
1908 */
1909 if (pte_present(pte)) {
1910 categories |= PAGE_IS_PRESENT;
1911 if (!huge_pte_uffd_wp(pte))
1912 categories |= PAGE_IS_WRITTEN;
1913 if (!PageAnon(pte_page(pte)))
1914 categories |= PAGE_IS_FILE;
1915 if (is_zero_pfn(pte_pfn(pte)))
1916 categories |= PAGE_IS_PFNZERO;
e6a9a2cb
AV		 1917 if (pte_soft_dirty(pte))
1918 categories |= PAGE_IS_SOFT_DIRTY;
52526ca7
MUA		 1919 } else if (is_swap_pte(pte)) {
1920 categories |= PAGE_IS_SWAPPED;
1921 if (!pte_swp_uffd_wp_any(pte))
1922 categories |= PAGE_IS_WRITTEN;
e6a9a2cb
AV		 1923 if (pte_swp_soft_dirty(pte))
1924 categories |= PAGE_IS_SOFT_DIRTY;
52526ca7
MUA		 1925 }
1926
1927 return categories;
1928}
1929
1930static void make_uffd_wp_huge_pte(struct vm_area_struct *vma,
1931 unsigned long addr, pte_t *ptep,
1932 pte_t ptent)
1933{
1934 unsigned long psize;
1935
1936 if (is_hugetlb_entry_hwpoisoned(ptent) || is_pte_marker(ptent))
1937 return;
1938
1939 psize = huge_page_size(hstate_vma(vma));
1940
1941 if (is_hugetlb_entry_migration(ptent))
1942 set_huge_pte_at(vma->vm_mm, addr, ptep,
1943 pte_swp_mkuffd_wp(ptent), psize);
1944 else if (!huge_pte_none(ptent))
1945 huge_ptep_modify_prot_commit(vma, addr, ptep, ptent,
1946 huge_pte_mkuffd_wp(ptent));
1947 else
1948 set_huge_pte_at(vma->vm_mm, addr, ptep,
1949 make_pte_marker(PTE_MARKER_UFFD_WP), psize);
1950}
1951#endif /* CONFIG_HUGETLB_PAGE */
1952
1953#if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLB_PAGE)
1954static void pagemap_scan_backout_range(struct pagemap_scan_private *p,
1955 unsigned long addr, unsigned long end)
1956{
1957 struct page_region *cur_buf = &p->vec_buf[p->vec_buf_index];
1958
1959 if (cur_buf->start != addr)
1960 cur_buf->end = addr;
1961 else
1962 cur_buf->start = cur_buf->end = 0;
1963
1964 p->found_pages -= (end - addr) / PAGE_SIZE;
1965}
1966#endif
1967
1968static bool pagemap_scan_is_interesting_page(unsigned long categories,
1969 const struct pagemap_scan_private *p)
1970{
1971 categories ^= p->arg.category_inverted;
1972 if ((categories & p->arg.category_mask) != p->arg.category_mask)
1973 return false;
1974 if (p->arg.category_anyof_mask && !(categories & p->arg.category_anyof_mask))
1975 return false;
1976
1977 return true;
1978}
1979
1980static bool pagemap_scan_is_interesting_vma(unsigned long categories,
1981 const struct pagemap_scan_private *p)
1982{
1983 unsigned long required = p->arg.category_mask & PAGE_IS_WPALLOWED;
1984
1985 categories ^= p->arg.category_inverted;
1986 if ((categories & required) != required)
1987 return false;
1988
1989 return true;
1990}
1991
1992static int pagemap_scan_test_walk(unsigned long start, unsigned long end,
1993 struct mm_walk *walk)
1994{
1995 struct pagemap_scan_private *p = walk->private;
1996 struct vm_area_struct *vma = walk->vma;
1997 unsigned long vma_category = 0;
0dff1b40
PX		 1998 bool wp_allowed = userfaultfd_wp_async(vma) &&
1999 userfaultfd_wp_use_markers(vma);
52526ca7 2000
0dff1b40
PX		 2001 if (!wp_allowed) {
2002 /* User requested explicit failure over wp-async capability */
2003 if (p->arg.flags & PM_SCAN_CHECK_WPASYNC)
2004 return -EPERM;
2005 /*
2006 * User requires wr-protect, and allows silently skipping
2007 * unsupported vmas.
2008 */
2009 if (p->arg.flags & PM_SCAN_WP_MATCHING)
2010 return 1;
2011 /*
2012 * Then the request doesn't involve wr-protects at all,
2013 * fall through to the rest checks, and allow vma walk.
2014 */
2015 }
52526ca7
MUA		 2016
2017 if (vma->vm_flags & VM_PFNMAP)
2018 return 1;
2019
0dff1b40
PX		 2020 if (wp_allowed)
2021 vma_category |= PAGE_IS_WPALLOWED;
2022
e6a9a2cb
AV		 2023 if (vma->vm_flags & VM_SOFTDIRTY)
2024 vma_category |= PAGE_IS_SOFT_DIRTY;
2025
52526ca7
MUA		 2026 if (!pagemap_scan_is_interesting_vma(vma_category, p))
2027 return 1;
2028
2029 p->cur_vma_category = vma_category;
2030
2031 return 0;
2032}
2033
2034static bool pagemap_scan_push_range(unsigned long categories,
2035 struct pagemap_scan_private *p,
2036 unsigned long addr, unsigned long end)
2037{
2038 struct page_region *cur_buf = &p->vec_buf[p->vec_buf_index];
2039
2040 /*
2041 * When there is no output buffer provided at all, the sentinel values
2042 * won't match here. There is no other way for `cur_buf->end` to be
2043 * non-zero other than it being non-empty.
2044 */
2045 if (addr == cur_buf->end && categories == cur_buf->categories) {
2046 cur_buf->end = end;
2047 return true;
2048 }
2049
2050 if (cur_buf->end) {
2051 if (p->vec_buf_index >= p->vec_buf_len - 1)
2052 return false;
2053
2054 cur_buf = &p->vec_buf[++p->vec_buf_index];
2055 }
2056
2057 cur_buf->start = addr;
2058 cur_buf->end = end;
2059 cur_buf->categories = categories;
2060
2061 return true;
2062}
2063
2064static int pagemap_scan_output(unsigned long categories,
2065 struct pagemap_scan_private *p,
2066 unsigned long addr, unsigned long *end)
2067{
2068 unsigned long n_pages, total_pages;
2069 int ret = 0;
2070
2071 if (!p->vec_buf)
2072 return 0;
2073
2074 categories &= p->arg.return_mask;
2075
2076 n_pages = (*end - addr) / PAGE_SIZE;
2077 if (check_add_overflow(p->found_pages, n_pages, &total_pages) ||
2078 total_pages > p->arg.max_pages) {
2079 size_t n_too_much = total_pages - p->arg.max_pages;
2080 *end -= n_too_much * PAGE_SIZE;
2081 n_pages -= n_too_much;
2082 ret = -ENOSPC;
2083 }
2084
2085 if (!pagemap_scan_push_range(categories, p, addr, *end)) {
2086 *end = addr;
2087 n_pages = 0;
2088 ret = -ENOSPC;
2089 }
2090
2091 p->found_pages += n_pages;
2092 if (ret)
2093 p->arg.walk_end = *end;
2094
2095 return ret;
2096}
2097
2098static int pagemap_scan_thp_entry(pmd_t *pmd, unsigned long start,
2099 unsigned long end, struct mm_walk *walk)
2100{
2101#ifdef CONFIG_TRANSPARENT_HUGEPAGE
2102 struct pagemap_scan_private *p = walk->private;
2103 struct vm_area_struct *vma = walk->vma;
2104 unsigned long categories;
2105 spinlock_t *ptl;
2106 int ret = 0;
2107
2108 ptl = pmd_trans_huge_lock(pmd, vma);
2109 if (!ptl)
2110 return -ENOENT;
2111
2112 categories = p->cur_vma_category |
2113 pagemap_thp_category(p, vma, start, *pmd);
2114
2115 if (!pagemap_scan_is_interesting_page(categories, p))
2116 goto out_unlock;
2117
2118 ret = pagemap_scan_output(categories, p, start, &end);
2119 if (start == end)
2120 goto out_unlock;
2121
2122 if (~p->arg.flags & PM_SCAN_WP_MATCHING)
2123 goto out_unlock;
2124 if (~categories & PAGE_IS_WRITTEN)
2125 goto out_unlock;
2126
2127 /*
2128 * Break huge page into small pages if the WP operation
2129 * needs to be performed on a portion of the huge page.
2130 */
2131 if (end != start + HPAGE_SIZE) {
2132 spin_unlock(ptl);
2133 split_huge_pmd(vma, pmd, start);
2134 pagemap_scan_backout_range(p, start, end);
2135 /* Report as if there was no THP */
2136 return -ENOENT;
2137 }
2138
2139 make_uffd_wp_pmd(vma, start, pmd);
2140 flush_tlb_range(vma, start, end);
2141out_unlock:
2142 spin_unlock(ptl);
2143 return ret;
2144#else /* !CONFIG_TRANSPARENT_HUGEPAGE */
2145 return -ENOENT;
2146#endif
2147}
2148
2149static int pagemap_scan_pmd_entry(pmd_t *pmd, unsigned long start,
2150 unsigned long end, struct mm_walk *walk)
2151{
2152 struct pagemap_scan_private *p = walk->private;
2153 struct vm_area_struct *vma = walk->vma;
2154 unsigned long addr, flush_end = 0;
2155 pte_t *pte, *start_pte;
2156 spinlock_t *ptl;
2157 int ret;
2158
2159 arch_enter_lazy_mmu_mode();
2160
2161 ret = pagemap_scan_thp_entry(pmd, start, end, walk);
2162 if (ret != -ENOENT) {
2163 arch_leave_lazy_mmu_mode();
2164 return ret;
2165 }
2166
2167 ret = 0;
2168 start_pte = pte = pte_offset_map_lock(vma->vm_mm, pmd, start, &ptl);
2169 if (!pte) {
2170 arch_leave_lazy_mmu_mode();
2171 walk->action = ACTION_AGAIN;
2172 return 0;
2173 }
2174
4980e837 2175 if ((p->arg.flags & PM_SCAN_WP_MATCHING) && !p->vec_out) {
12f6b01a
MUA		 2176 /* Fast path for performing exclusive WP */
2177 for (addr = start; addr != end; pte++, addr += PAGE_SIZE) {
2178 if (pte_uffd_wp(ptep_get(pte)))
2179 continue;
2180 make_uffd_wp_pte(vma, addr, pte);
2181 if (!flush_end)
2182 start = addr;
2183 flush_end = addr + PAGE_SIZE;
2184 }
2185 goto flush_and_return;
2186 }
2187
2188 if (!p->arg.category_anyof_mask && !p->arg.category_inverted &&
2189 p->arg.category_mask == PAGE_IS_WRITTEN &&
2190 p->arg.return_mask == PAGE_IS_WRITTEN) {
2191 for (addr = start; addr < end; pte++, addr += PAGE_SIZE) {
2192 unsigned long next = addr + PAGE_SIZE;
2193
2194 if (pte_uffd_wp(ptep_get(pte)))
2195 continue;
2196 ret = pagemap_scan_output(p->cur_vma_category | PAGE_IS_WRITTEN,
2197 p, addr, &next);
2198 if (next == addr)
2199 break;
2200 if (~p->arg.flags & PM_SCAN_WP_MATCHING)
2201 continue;
2202 make_uffd_wp_pte(vma, addr, pte);
2203 if (!flush_end)
2204 start = addr;
2205 flush_end = next;
2206 }
2207 goto flush_and_return;
2208 }
2209
52526ca7
MUA		 2210 for (addr = start; addr != end; pte++, addr += PAGE_SIZE) {
2211 unsigned long categories = p->cur_vma_category |
2212 pagemap_page_category(p, vma, addr, ptep_get(pte));
2213 unsigned long next = addr + PAGE_SIZE;
2214
2215 if (!pagemap_scan_is_interesting_page(categories, p))
2216 continue;
2217
2218 ret = pagemap_scan_output(categories, p, addr, &next);
2219 if (next == addr)
2220 break;
2221
2222 if (~p->arg.flags & PM_SCAN_WP_MATCHING)
2223 continue;
2224 if (~categories & PAGE_IS_WRITTEN)
2225 continue;
2226
2227 make_uffd_wp_pte(vma, addr, pte);
2228 if (!flush_end)
2229 start = addr;
2230 flush_end = next;
2231 }
2232
12f6b01a 2233flush_and_return:
52526ca7
MUA		 2234 if (flush_end)
2235 flush_tlb_range(vma, start, addr);
2236
2237 pte_unmap_unlock(start_pte, ptl);
2238 arch_leave_lazy_mmu_mode();
2239
2240 cond_resched();
2241 return ret;
2242}
2243
2244#ifdef CONFIG_HUGETLB_PAGE
2245static int pagemap_scan_hugetlb_entry(pte_t *ptep, unsigned long hmask,
2246 unsigned long start, unsigned long end,
2247 struct mm_walk *walk)
2248{
2249 struct pagemap_scan_private *p = walk->private;
2250 struct vm_area_struct *vma = walk->vma;
2251 unsigned long categories;
2252 spinlock_t *ptl;
2253 int ret = 0;
2254 pte_t pte;
2255
2256 if (~p->arg.flags & PM_SCAN_WP_MATCHING) {
2257 /* Go the short route when not write-protecting pages. */
2258
2259 pte = huge_ptep_get(ptep);
2260 categories = p->cur_vma_category | pagemap_hugetlb_category(pte);
2261
2262 if (!pagemap_scan_is_interesting_page(categories, p))
2263 return 0;
2264
2265 return pagemap_scan_output(categories, p, start, &end);
2266 }
2267
2268 i_mmap_lock_write(vma->vm_file->f_mapping);
2269 ptl = huge_pte_lock(hstate_vma(vma), vma->vm_mm, ptep);
2270
2271 pte = huge_ptep_get(ptep);
2272 categories = p->cur_vma_category | pagemap_hugetlb_category(pte);
2273
2274 if (!pagemap_scan_is_interesting_page(categories, p))
2275 goto out_unlock;
2276
2277 ret = pagemap_scan_output(categories, p, start, &end);
2278 if (start == end)
2279 goto out_unlock;
2280
2281 if (~categories & PAGE_IS_WRITTEN)
2282 goto out_unlock;
2283
2284 if (end != start + HPAGE_SIZE) {
2285 /* Partial HugeTLB page WP isn't possible. */
2286 pagemap_scan_backout_range(p, start, end);
2287 p->arg.walk_end = start;
2288 ret = 0;
2289 goto out_unlock;
2290 }
2291
2292 make_uffd_wp_huge_pte(vma, start, ptep, pte);
2293 flush_hugetlb_tlb_range(vma, start, end);
2294
2295out_unlock:
2296 spin_unlock(ptl);
2297 i_mmap_unlock_write(vma->vm_file->f_mapping);
2298
2299 return ret;
2300}
2301#else
2302#define pagemap_scan_hugetlb_entry NULL
2303#endif
2304
2305static int pagemap_scan_pte_hole(unsigned long addr, unsigned long end,
2306 int depth, struct mm_walk *walk)
2307{
2308 struct pagemap_scan_private *p = walk->private;
2309 struct vm_area_struct *vma = walk->vma;
2310 int ret, err;
2311
2312 if (!vma || !pagemap_scan_is_interesting_page(p->cur_vma_category, p))
2313 return 0;
2314
2315 ret = pagemap_scan_output(p->cur_vma_category, p, addr, &end);
2316 if (addr == end)
2317 return ret;
2318
2319 if (~p->arg.flags & PM_SCAN_WP_MATCHING)
2320 return ret;
2321
2322 err = uffd_wp_range(vma, addr, end - addr, true);
2323 if (err < 0)
2324 ret = err;
2325
2326 return ret;
2327}
2328
2329static const struct mm_walk_ops pagemap_scan_ops = {
2330 .test_walk = pagemap_scan_test_walk,
2331 .pmd_entry = pagemap_scan_pmd_entry,
2332 .pte_hole = pagemap_scan_pte_hole,
2333 .hugetlb_entry = pagemap_scan_hugetlb_entry,
2334};
2335
2336static int pagemap_scan_get_args(struct pm_scan_arg *arg,
2337 unsigned long uarg)
2338{
2339 if (copy_from_user(arg, (void __user *)uarg, sizeof(*arg)))
2340 return -EFAULT;
2341
2342 if (arg->size != sizeof(struct pm_scan_arg))
2343 return -EINVAL;
2344
2345 /* Validate requested features */
2346 if (arg->flags & ~PM_SCAN_FLAGS)
2347 return -EINVAL;
2348 if ((arg->category_inverted | arg->category_mask |
2349 arg->category_anyof_mask | arg->return_mask) & ~PM_SCAN_CATEGORIES)
2350 return -EINVAL;
2351
2352 arg->start = untagged_addr((unsigned long)arg->start);
2353 arg->end = untagged_addr((unsigned long)arg->end);
2354 arg->vec = untagged_addr((unsigned long)arg->vec);
2355
2356 /* Validate memory pointers */
2357 if (!IS_ALIGNED(arg->start, PAGE_SIZE))
2358 return -EINVAL;
2359 if (!access_ok((void __user *)(long)arg->start, arg->end - arg->start))
2360 return -EFAULT;
2361 if (!arg->vec && arg->vec_len)
2362 return -EINVAL;
2363 if (arg->vec && !access_ok((void __user *)(long)arg->vec,
2364 arg->vec_len * sizeof(struct page_region)))
2365 return -EFAULT;
2366
2367 /* Fixup default values */
2368 arg->end = ALIGN(arg->end, PAGE_SIZE);
2369 arg->walk_end = 0;
2370 if (!arg->max_pages)
2371 arg->max_pages = ULONG_MAX;
2372
2373 return 0;
2374}
2375
2376static int pagemap_scan_writeback_args(struct pm_scan_arg *arg,
2377 unsigned long uargl)
2378{
2379 struct pm_scan_arg __user *uarg = (void __user *)uargl;
2380
2381 if (copy_to_user(&uarg->walk_end, &arg->walk_end, sizeof(arg->walk_end)))
2382 return -EFAULT;
2383
2384 return 0;
2385}
2386
2387static int pagemap_scan_init_bounce_buffer(struct pagemap_scan_private *p)
2388{
2389 if (!p->arg.vec_len)
2390 return 0;
2391
2392 p->vec_buf_len = min_t(size_t, PAGEMAP_WALK_SIZE >> PAGE_SHIFT,
2393 p->arg.vec_len);
2394 p->vec_buf = kmalloc_array(p->vec_buf_len, sizeof(*p->vec_buf),
2395 GFP_KERNEL);
2396 if (!p->vec_buf)
2397 return -ENOMEM;
2398
2399 p->vec_buf->start = p->vec_buf->end = 0;
2400 p->vec_out = (struct page_region __user *)(long)p->arg.vec;
2401
2402 return 0;
2403}
2404
2405static long pagemap_scan_flush_buffer(struct pagemap_scan_private *p)
2406{
2407 const struct page_region *buf = p->vec_buf;
2408 long n = p->vec_buf_index;
2409
2410 if (!p->vec_buf)
2411 return 0;
2412
2413 if (buf[n].end != buf[n].start)
2414 n++;
2415
2416 if (!n)
2417 return 0;
2418
2419 if (copy_to_user(p->vec_out, buf, n * sizeof(*buf)))
2420 return -EFAULT;
2421
2422 p->arg.vec_len -= n;
2423 p->vec_out += n;
2424
2425 p->vec_buf_index = 0;
2426 p->vec_buf_len = min_t(size_t, p->vec_buf_len, p->arg.vec_len);
2427 p->vec_buf->start = p->vec_buf->end = 0;
2428
2429 return n;
2430}
2431
2432static long do_pagemap_scan(struct mm_struct *mm, unsigned long uarg)
2433{
52526ca7
MUA		 2434 struct pagemap_scan_private p = {0};
2435 unsigned long walk_start;
2436 size_t n_ranges_out = 0;
2437 int ret;
2438
2439 ret = pagemap_scan_get_args(&p.arg, uarg);
2440 if (ret)
2441 return ret;
2442
2443 p.masks_of_interest = p.arg.category_mask | p.arg.category_anyof_mask |
2444 p.arg.return_mask;
2445 ret = pagemap_scan_init_bounce_buffer(&p);
2446 if (ret)
2447 return ret;
2448
52526ca7
MUA		 2449 for (walk_start = p.arg.start; walk_start < p.arg.end;
2450 walk_start = p.arg.walk_end) {
4cccb622 2451 struct mmu_notifier_range range;
52526ca7
MUA		 2452 long n_out;
2453
2454 if (fatal_signal_pending(current)) {
2455 ret = -EINTR;
2456 break;
2457 }
2458
2459 ret = mmap_read_lock_killable(mm);
2460 if (ret)
2461 break;
4cccb622
MUA		 2462
2463 /* Protection change for the range is going to happen. */
2464 if (p.arg.flags & PM_SCAN_WP_MATCHING) {
2465 mmu_notifier_range_init(&range, MMU_NOTIFY_PROTECTION_VMA, 0,
2466 mm, walk_start, p.arg.end);
2467 mmu_notifier_invalidate_range_start(&range);
2468 }
2469
52526ca7
MUA		 2470 ret = walk_page_range(mm, walk_start, p.arg.end,
2471 &pagemap_scan_ops, &p);
4cccb622
MUA		 2472
2473 if (p.arg.flags & PM_SCAN_WP_MATCHING)
2474 mmu_notifier_invalidate_range_end(&range);
2475
52526ca7
MUA		 2476 mmap_read_unlock(mm);
2477
2478 n_out = pagemap_scan_flush_buffer(&p);
2479 if (n_out < 0)
2480 ret = n_out;
2481 else
2482 n_ranges_out += n_out;
2483
2484 if (ret != -ENOSPC)
2485 break;
2486
2487 if (p.arg.vec_len == 0 || p.found_pages == p.arg.max_pages)
2488 break;
2489 }
2490
2491 /* ENOSPC signifies early stop (buffer full) from the walk. */
2492 if (!ret || ret == -ENOSPC)
2493 ret = n_ranges_out;
2494
2495 /* The walk_end isn't set when ret is zero */
2496 if (!p.arg.walk_end)
2497 p.arg.walk_end = p.arg.end;
2498 if (pagemap_scan_writeback_args(&p.arg, uarg))
2499 ret = -EFAULT;
2500
52526ca7
MUA		 2501 kfree(p.vec_buf);
2502 return ret;
2503}
2504
2505static long do_pagemap_cmd(struct file *file, unsigned int cmd,
2506 unsigned long arg)
2507{
2508 struct mm_struct *mm = file->private_data;
2509
2510 switch (cmd) {
2511 case PAGEMAP_SCAN:
2512 return do_pagemap_scan(mm, arg);
2513
2514 default:
2515 return -EINVAL;
2516 }
2517}
2518
85863e47
MM		 2519const struct file_operations proc_pagemap_operations = {
2520 .llseek = mem_lseek, /* borrow this */
2521 .read = pagemap_read,
541c237c 2522 .open = pagemap_open,
a06db751 2523 .release = pagemap_release,
52526ca7
MUA		 2524 .unlocked_ioctl = do_pagemap_cmd,
2525 .compat_ioctl = do_pagemap_cmd,
85863e47 2526};
1e883281 2527#endif /* CONFIG_PROC_PAGE_MONITOR */
85863e47 2528
6e21c8f1 2529#ifdef CONFIG_NUMA
6e21c8f1 2530
f69ff943 2531struct numa_maps {
f69ff943
SW		 2532 unsigned long pages;
2533 unsigned long anon;
2534 unsigned long active;
2535 unsigned long writeback;
2536 unsigned long mapcount_max;
2537 unsigned long dirty;
2538 unsigned long swapcache;
2539 unsigned long node[MAX_NUMNODES];
2540};
2541
5b52fc89
SW		 2542struct numa_maps_private {
2543 struct proc_maps_private proc_maps;
2544 struct numa_maps md;
2545};
2546
eb4866d0
DH		 2547static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
2548 unsigned long nr_pages)
f69ff943
SW		 2549{
2550 int count = page_mapcount(page);
2551
eb4866d0 2552 md->pages += nr_pages;
f69ff943 2553 if (pte_dirty || PageDirty(page))
eb4866d0 2554 md->dirty += nr_pages;
f69ff943
SW		 2555
2556 if (PageSwapCache(page))
eb4866d0 2557 md->swapcache += nr_pages;
f69ff943
SW		 2558
2559 if (PageActive(page) || PageUnevictable(page))
eb4866d0 2560 md->active += nr_pages;
f69ff943
SW		 2561
2562 if (PageWriteback(page))
eb4866d0 2563 md->writeback += nr_pages;
f69ff943
SW		 2564
2565 if (PageAnon(page))
eb4866d0 2566 md->anon += nr_pages;
f69ff943
SW		 2567
2568 if (count > md->mapcount_max)
2569 md->mapcount_max = count;
2570
eb4866d0 2571 md->node[page_to_nid(page)] += nr_pages;
f69ff943
SW		 2572}
2573
3200a8aa
DH		 2574static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
2575 unsigned long addr)
2576{
2577 struct page *page;
2578 int nid;
2579
2580 if (!pte_present(pte))
2581 return NULL;
2582
2583 page = vm_normal_page(vma, addr, pte);
3218f871 2584 if (!page || is_zone_device_page(page))
3200a8aa
DH		 2585 return NULL;
2586
2587 if (PageReserved(page))
2588 return NULL;
2589
2590 nid = page_to_nid(page);
4ff1b2c2 2591 if (!node_isset(nid, node_states[N_MEMORY]))
3200a8aa
DH		 2592 return NULL;
2593
2594 return page;
2595}
2596
28093f9f
GS		 2597#ifdef CONFIG_TRANSPARENT_HUGEPAGE
2598static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
2599 struct vm_area_struct *vma,
2600 unsigned long addr)
2601{
2602 struct page *page;
2603 int nid;
2604
2605 if (!pmd_present(pmd))
2606 return NULL;
2607
2608 page = vm_normal_page_pmd(vma, addr, pmd);
2609 if (!page)
2610 return NULL;
2611
2612 if (PageReserved(page))
2613 return NULL;
2614
2615 nid = page_to_nid(page);
2616 if (!node_isset(nid, node_states[N_MEMORY]))
2617 return NULL;
2618
2619 return page;
2620}
2621#endif
2622
f69ff943
SW		 2623static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
2624 unsigned long end, struct mm_walk *walk)
2625{
d85f4d6d
NH		 2626 struct numa_maps *md = walk->private;
2627 struct vm_area_struct *vma = walk->vma;
f69ff943
SW		 2628 spinlock_t *ptl;
2629 pte_t *orig_pte;
2630 pte_t *pte;
2631
28093f9f 2632#ifdef CONFIG_TRANSPARENT_HUGEPAGE
b6ec57f4
KS		 2633 ptl = pmd_trans_huge_lock(pmd, vma);
2634 if (ptl) {
025c5b24
NH		 2635 struct page *page;
2636
28093f9f 2637 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
025c5b24 2638 if (page)
28093f9f 2639 gather_stats(page, md, pmd_dirty(*pmd),
025c5b24 2640 HPAGE_PMD_SIZE/PAGE_SIZE);
bf929152 2641 spin_unlock(ptl);
025c5b24 2642 return 0;
32ef4384 2643 }
28093f9f 2644#endif
f69ff943 2645 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
7780d040
HD		 2646 if (!pte) {
2647 walk->action = ACTION_AGAIN;
2648 return 0;
2649 }
f69ff943 2650 do {
c33c7948
RR		 2651 pte_t ptent = ptep_get(pte);
2652 struct page *page = can_gather_numa_stats(ptent, vma, addr);
f69ff943
SW		 2653 if (!page)
2654 continue;
c33c7948 2655 gather_stats(page, md, pte_dirty(ptent), 1);
f69ff943
SW		 2656
2657 } while (pte++, addr += PAGE_SIZE, addr != end);
2658 pte_unmap_unlock(orig_pte, ptl);
a66c0410 2659 cond_resched();
f69ff943
SW		 2660 return 0;
2661}
2662#ifdef CONFIG_HUGETLB_PAGE
632fd60f 2663static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
f69ff943
SW		 2664 unsigned long addr, unsigned long end, struct mm_walk *walk)
2665{
5c2ff95e 2666 pte_t huge_pte = huge_ptep_get(pte);
f69ff943
SW		 2667 struct numa_maps *md;
2668 struct page *page;
2669
5c2ff95e 2670 if (!pte_present(huge_pte))
f69ff943
SW		 2671 return 0;
2672
5c2ff95e 2673 page = pte_page(huge_pte);
f69ff943
SW		 2674
2675 md = walk->private;
5c2ff95e 2676 gather_stats(page, md, pte_dirty(huge_pte), 1);
f69ff943
SW		 2677 return 0;
2678}
2679
2680#else
632fd60f 2681static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
f69ff943
SW		 2682 unsigned long addr, unsigned long end, struct mm_walk *walk)
2683{
2684 return 0;
2685}
2686#endif
2687
7b86ac33
CH		 2688static const struct mm_walk_ops show_numa_ops = {
2689 .hugetlb_entry = gather_hugetlb_stats,
2690 .pmd_entry = gather_pte_stats,
49b06385 2691 .walk_lock = PGWALK_RDLOCK,
7b86ac33
CH		 2692};
2693
f69ff943
SW		 2694/*
2695 * Display pages allocated per node and memory policy via /proc.
2696 */
871305bb 2697static int show_numa_map(struct seq_file *m, void *v)
f69ff943 2698{
5b52fc89
SW		 2699 struct numa_maps_private *numa_priv = m->private;
2700 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
f69ff943 2701 struct vm_area_struct *vma = v;
5b52fc89 2702 struct numa_maps *md = &numa_priv->md;
f69ff943
SW		 2703 struct file *file = vma->vm_file;
2704 struct mm_struct *mm = vma->vm_mm;
948927ee 2705 char buffer[64];
ddc1a5cb
HD		 2706 struct mempolicy *pol;
2707 pgoff_t ilx;
948927ee 2708 int nid;
f69ff943
SW		 2709
2710 if (!mm)
2711 return 0;
2712
5b52fc89
SW		 2713 /* Ensure we start with an empty set of numa_maps statistics. */
2714 memset(md, 0, sizeof(*md));
f69ff943 2715
ddc1a5cb 2716 pol = __get_vma_policy(vma, vma->vm_start, &ilx);
498f2371
ON		 2717 if (pol) {
2718 mpol_to_str(buffer, sizeof(buffer), pol);
2719 mpol_cond_put(pol);
2720 } else {
2721 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
2722 }
f69ff943
SW		 2723
2724 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
2725
2726 if (file) {
17c2b4ee 2727 seq_puts(m, " file=");
08582d67 2728 seq_path(m, file_user_path(file), "\n\t= ");
11250fd1 2729 } else if (vma_is_initial_heap(vma)) {
17c2b4ee 2730 seq_puts(m, " heap");
11250fd1 2731 } else if (vma_is_initial_stack(vma)) {
65376df5 2732 seq_puts(m, " stack");
f69ff943
SW		 2733 }
2734
fc360bd9 2735 if (is_vm_hugetlb_page(vma))
17c2b4ee 2736 seq_puts(m, " huge");
fc360bd9 2737
c1e8d7c6 2738 /* mmap_lock is held by m_start */
7b86ac33 2739 walk_page_vma(vma, &show_numa_ops, md);
f69ff943
SW		 2740
2741 if (!md->pages)
2742 goto out;
2743
2744 if (md->anon)
2745 seq_printf(m, " anon=%lu", md->anon);
2746
2747 if (md->dirty)
2748 seq_printf(m, " dirty=%lu", md->dirty);
2749
2750 if (md->pages != md->anon && md->pages != md->dirty)
2751 seq_printf(m, " mapped=%lu", md->pages);
2752
2753 if (md->mapcount_max > 1)
2754 seq_printf(m, " mapmax=%lu", md->mapcount_max);
2755
2756 if (md->swapcache)
2757 seq_printf(m, " swapcache=%lu", md->swapcache);
2758
2759 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
2760 seq_printf(m, " active=%lu", md->active);
2761
2762 if (md->writeback)
2763 seq_printf(m, " writeback=%lu", md->writeback);
2764
948927ee
DR		 2765 for_each_node_state(nid, N_MEMORY)
2766 if (md->node[nid])
2767 seq_printf(m, " N%d=%lu", nid, md->node[nid]);
198d1597
RA		 2768
2769 seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
f69ff943
SW		 2770out:
2771 seq_putc(m, '\n');
f69ff943
SW		 2772 return 0;
2773}
5b52fc89 2774
03a44825 2775static const struct seq_operations proc_pid_numa_maps_op = {
b7643757
SP		 2776 .start = m_start,
2777 .next = m_next,
2778 .stop = m_stop,
871305bb 2779 .show = show_numa_map,
6e21c8f1 2780};
662795de 2781
b7643757
SP		 2782static int pid_numa_maps_open(struct inode *inode, struct file *file)
2783{
871305bb
VB		 2784 return proc_maps_open(inode, file, &proc_pid_numa_maps_op,
2785 sizeof(struct numa_maps_private));
b7643757
SP		 2786}
2787
2788const struct file_operations proc_pid_numa_maps_operations = {
2789 .open = pid_numa_maps_open,
2790 .read = seq_read,
2791 .llseek = seq_lseek,
29a40ace 2792 .release = proc_map_release,
b7643757
SP		 2793};
2794
f69ff943 2795#endif /* CONFIG_NUMA */
Linux 6.x block layer and io_uring tree(s)