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 26 #include <asm/elf.h> 27 #include <asm/tlb.h> 28 #include <asm/tlbflush.h> 29 #include "internal.h" 30 31 #define SEQ_PUT_DEC(str, val) \ 32 seq_put_decimal_ull_width(m, str, (val) << (PAGE_SHIFT-10), 8) 33 void task_mem(struct seq_file *m, struct mm_struct *mm) 34 { 35 unsigned long text, lib, swap, anon, file, shmem; 36 unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss; 37 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 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; 52 hiwater_rss = total_rss = anon + file + shmem; 53 if (hiwater_rss < mm->hiwater_rss) 54 hiwater_rss = mm->hiwater_rss; 55 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 61 swap = get_mm_counter(mm, MM_SWAPENTS); 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); 65 SEQ_PUT_DEC(" kB\nVmPin:\t", atomic64_read(&mm->pinned_vm)); 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"); 81 hugetlb_report_usage(m, mm); 82 } 83 #undef SEQ_PUT_DEC 84 85 unsigned long task_vsize(struct mm_struct *mm) 86 { 87 return PAGE_SIZE * mm->total_vm; 88 } 89 90 unsigned long task_statm(struct mm_struct *mm, 91 unsigned long *shared, unsigned long *text, 92 unsigned long *data, unsigned long *resident) 93 { 94 *shared = get_mm_counter(mm, MM_FILEPAGES) + 95 get_mm_counter(mm, MM_SHMEMPAGES); 96 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) 97 >> PAGE_SHIFT; 98 *data = mm->data_vm + mm->stack_vm; 99 *resident = *shared + get_mm_counter(mm, MM_ANONPAGES); 100 return mm->total_vm; 101 } 102 103 #ifdef CONFIG_NUMA 104 /* 105 * Save get_task_policy() for show_numa_map(). 106 */ 107 static void hold_task_mempolicy(struct proc_maps_private *priv) 108 { 109 struct task_struct *task = priv->task; 110 111 task_lock(task); 112 priv->task_mempolicy = get_task_policy(task); 113 mpol_get(priv->task_mempolicy); 114 task_unlock(task); 115 } 116 static void release_task_mempolicy(struct proc_maps_private *priv) 117 { 118 mpol_put(priv->task_mempolicy); 119 } 120 #else 121 static void hold_task_mempolicy(struct proc_maps_private *priv) 122 { 123 } 124 static void release_task_mempolicy(struct proc_maps_private *priv) 125 { 126 } 127 #endif 128 129 static 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 } 143 144 static void *m_start(struct seq_file *m, loff_t *ppos) 145 { 146 struct proc_maps_private *priv = m->private; 147 unsigned long last_addr = *ppos; 148 struct mm_struct *mm; 149 150 /* See m_next(). Zero at the start or after lseek. */ 151 if (last_addr == -1UL) 152 return NULL; 153 154 priv->task = get_proc_task(priv->inode); 155 if (!priv->task) 156 return ERR_PTR(-ESRCH); 157 158 mm = priv->mm; 159 if (!mm || !mmget_not_zero(mm)) { 160 put_task_struct(priv->task); 161 priv->task = NULL; 162 return NULL; 163 } 164 165 if (mmap_read_lock_killable(mm)) { 166 mmput(mm); 167 put_task_struct(priv->task); 168 priv->task = NULL; 169 return ERR_PTR(-EINTR); 170 } 171 172 vma_iter_init(&priv->iter, mm, last_addr); 173 hold_task_mempolicy(priv); 174 if (last_addr == -2UL) 175 return get_gate_vma(mm); 176 177 return proc_get_vma(priv, ppos); 178 } 179 180 static void *m_next(struct seq_file *m, void *v, loff_t *ppos) 181 { 182 if (*ppos == -2UL) { 183 *ppos = -1UL; 184 return NULL; 185 } 186 return proc_get_vma(m->private, ppos); 187 } 188 189 static void m_stop(struct seq_file *m, void *v) 190 { 191 struct proc_maps_private *priv = m->private; 192 struct mm_struct *mm = priv->mm; 193 194 if (!priv->task) 195 return; 196 197 release_task_mempolicy(priv); 198 mmap_read_unlock(mm); 199 mmput(mm); 200 put_task_struct(priv->task); 201 priv->task = NULL; 202 } 203 204 static 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; 211 212 priv->inode = inode; 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 } 220 221 return 0; 222 } 223 224 static 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 } 234 235 static int do_maps_open(struct inode *inode, struct file *file, 236 const struct seq_operations *ops) 237 { 238 return proc_maps_open(inode, file, ops, 239 sizeof(struct proc_maps_private)); 240 } 241 242 static void get_vma_name(struct vm_area_struct *vma, 243 const struct path **path, 244 const char **name, 245 const char **name_fmt) 246 { 247 struct anon_vma_name *anon_name = vma->vm_mm ? anon_vma_name(vma) : NULL; 248 249 *name = NULL; 250 *path = NULL; 251 *name_fmt = NULL; 252 253 /* 254 * Print the dentry name for named mappings, and a 255 * special [heap] marker for the heap: 256 */ 257 if (vma->vm_file) { 258 /* 259 * If user named this anon shared memory via 260 * prctl(PR_SET_VMA ..., use the provided name. 261 */ 262 if (anon_name) { 263 *name_fmt = "[anon_shmem:%s]"; 264 *name = anon_name->name; 265 } else { 266 *path = file_user_path(vma->vm_file); 267 } 268 return; 269 } 270 271 if (vma->vm_ops && vma->vm_ops->name) { 272 *name = vma->vm_ops->name(vma); 273 if (*name) 274 return; 275 } 276 277 *name = arch_vma_name(vma); 278 if (*name) 279 return; 280 281 if (!vma->vm_mm) { 282 *name = "[vdso]"; 283 return; 284 } 285 286 if (vma_is_initial_heap(vma)) { 287 *name = "[heap]"; 288 return; 289 } 290 291 if (vma_is_initial_stack(vma)) { 292 *name = "[stack]"; 293 return; 294 } 295 296 if (anon_name) { 297 *name_fmt = "[anon:%s]"; 298 *name = anon_name->name; 299 return; 300 } 301 } 302 303 static void show_vma_header_prefix(struct seq_file *m, 304 unsigned long start, unsigned long end, 305 vm_flags_t flags, unsigned long long pgoff, 306 dev_t dev, unsigned long ino) 307 { 308 seq_setwidth(m, 25 + sizeof(void *) * 6 - 1); 309 seq_put_hex_ll(m, NULL, start, 8); 310 seq_put_hex_ll(m, "-", end, 8); 311 seq_putc(m, ' '); 312 seq_putc(m, flags & VM_READ ? 'r' : '-'); 313 seq_putc(m, flags & VM_WRITE ? 'w' : '-'); 314 seq_putc(m, flags & VM_EXEC ? 'x' : '-'); 315 seq_putc(m, flags & VM_MAYSHARE ? 's' : 'p'); 316 seq_put_hex_ll(m, " ", pgoff, 8); 317 seq_put_hex_ll(m, " ", MAJOR(dev), 2); 318 seq_put_hex_ll(m, ":", MINOR(dev), 2); 319 seq_put_decimal_ull(m, " ", ino); 320 seq_putc(m, ' '); 321 } 322 323 static void 324 show_map_vma(struct seq_file *m, struct vm_area_struct *vma) 325 { 326 const struct path *path; 327 const char *name_fmt, *name; 328 vm_flags_t flags = vma->vm_flags; 329 unsigned long ino = 0; 330 unsigned long long pgoff = 0; 331 unsigned long start, end; 332 dev_t dev = 0; 333 334 if (vma->vm_file) { 335 const struct inode *inode = file_user_inode(vma->vm_file); 336 337 dev = inode->i_sb->s_dev; 338 ino = inode->i_ino; 339 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT; 340 } 341 342 start = vma->vm_start; 343 end = vma->vm_end; 344 show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino); 345 346 get_vma_name(vma, &path, &name, &name_fmt); 347 if (path) { 348 seq_pad(m, ' '); 349 seq_path(m, path, "\n"); 350 } else if (name_fmt) { 351 seq_pad(m, ' '); 352 seq_printf(m, name_fmt, name); 353 } else if (name) { 354 seq_pad(m, ' '); 355 seq_puts(m, name); 356 } 357 seq_putc(m, '\n'); 358 } 359 360 static int show_map(struct seq_file *m, void *v) 361 { 362 show_map_vma(m, v); 363 return 0; 364 } 365 366 static const struct seq_operations proc_pid_maps_op = { 367 .start = m_start, 368 .next = m_next, 369 .stop = m_stop, 370 .show = show_map 371 }; 372 373 static int pid_maps_open(struct inode *inode, struct file *file) 374 { 375 return do_maps_open(inode, file, &proc_pid_maps_op); 376 } 377 378 const struct file_operations proc_pid_maps_operations = { 379 .open = pid_maps_open, 380 .read = seq_read, 381 .llseek = seq_lseek, 382 .release = proc_map_release, 383 }; 384 385 /* 386 * Proportional Set Size(PSS): my share of RSS. 387 * 388 * PSS of a process is the count of pages it has in memory, where each 389 * page is divided by the number of processes sharing it. So if a 390 * process has 1000 pages all to itself, and 1000 shared with one other 391 * process, its PSS will be 1500. 392 * 393 * To keep (accumulated) division errors low, we adopt a 64bit 394 * fixed-point pss counter to minimize division errors. So (pss >> 395 * PSS_SHIFT) would be the real byte count. 396 * 397 * A shift of 12 before division means (assuming 4K page size): 398 * - 1M 3-user-pages add up to 8KB errors; 399 * - supports mapcount up to 2^24, or 16M; 400 * - supports PSS up to 2^52 bytes, or 4PB. 401 */ 402 #define PSS_SHIFT 12 403 404 #ifdef CONFIG_PROC_PAGE_MONITOR 405 struct mem_size_stats { 406 unsigned long resident; 407 unsigned long shared_clean; 408 unsigned long shared_dirty; 409 unsigned long private_clean; 410 unsigned long private_dirty; 411 unsigned long referenced; 412 unsigned long anonymous; 413 unsigned long lazyfree; 414 unsigned long anonymous_thp; 415 unsigned long shmem_thp; 416 unsigned long file_thp; 417 unsigned long swap; 418 unsigned long shared_hugetlb; 419 unsigned long private_hugetlb; 420 unsigned long ksm; 421 u64 pss; 422 u64 pss_anon; 423 u64 pss_file; 424 u64 pss_shmem; 425 u64 pss_dirty; 426 u64 pss_locked; 427 u64 swap_pss; 428 }; 429 430 static void smaps_page_accumulate(struct mem_size_stats *mss, 431 struct folio *folio, unsigned long size, unsigned long pss, 432 bool dirty, bool locked, bool private) 433 { 434 mss->pss += pss; 435 436 if (folio_test_anon(folio)) 437 mss->pss_anon += pss; 438 else if (folio_test_swapbacked(folio)) 439 mss->pss_shmem += pss; 440 else 441 mss->pss_file += pss; 442 443 if (locked) 444 mss->pss_locked += pss; 445 446 if (dirty || folio_test_dirty(folio)) { 447 mss->pss_dirty += pss; 448 if (private) 449 mss->private_dirty += size; 450 else 451 mss->shared_dirty += size; 452 } else { 453 if (private) 454 mss->private_clean += size; 455 else 456 mss->shared_clean += size; 457 } 458 } 459 460 static void smaps_account(struct mem_size_stats *mss, struct page *page, 461 bool compound, bool young, bool dirty, bool locked, 462 bool present) 463 { 464 struct folio *folio = page_folio(page); 465 int i, nr = compound ? compound_nr(page) : 1; 466 unsigned long size = nr * PAGE_SIZE; 467 468 /* 469 * First accumulate quantities that depend only on |size| and the type 470 * of the compound page. 471 */ 472 if (folio_test_anon(folio)) { 473 mss->anonymous += size; 474 if (!folio_test_swapbacked(folio) && !dirty && 475 !folio_test_dirty(folio)) 476 mss->lazyfree += size; 477 } 478 479 if (folio_test_ksm(folio)) 480 mss->ksm += size; 481 482 mss->resident += size; 483 /* Accumulate the size in pages that have been accessed. */ 484 if (young || folio_test_young(folio) || folio_test_referenced(folio)) 485 mss->referenced += size; 486 487 /* 488 * Then accumulate quantities that may depend on sharing, or that may 489 * differ page-by-page. 490 * 491 * refcount == 1 for present entries guarantees that the folio is mapped 492 * exactly once. For large folios this implies that exactly one 493 * PTE/PMD/... maps (a part of) this folio. 494 * 495 * Treat all non-present entries (where relying on the mapcount and 496 * refcount doesn't make sense) as "maybe shared, but not sure how 497 * often". We treat device private entries as being fake-present. 498 * 499 * Note that it would not be safe to read the mapcount especially for 500 * pages referenced by migration entries, even with the PTL held. 501 */ 502 if (folio_ref_count(folio) == 1 || !present) { 503 smaps_page_accumulate(mss, folio, size, size << PSS_SHIFT, 504 dirty, locked, present); 505 return; 506 } 507 /* 508 * We obtain a snapshot of the mapcount. Without holding the folio lock 509 * this snapshot can be slightly wrong as we cannot always read the 510 * mapcount atomically. 511 */ 512 for (i = 0; i < nr; i++, page++) { 513 int mapcount = folio_precise_page_mapcount(folio, page); 514 unsigned long pss = PAGE_SIZE << PSS_SHIFT; 515 if (mapcount >= 2) 516 pss /= mapcount; 517 smaps_page_accumulate(mss, folio, PAGE_SIZE, pss, 518 dirty, locked, mapcount < 2); 519 } 520 } 521 522 #ifdef CONFIG_SHMEM 523 static int smaps_pte_hole(unsigned long addr, unsigned long end, 524 __always_unused int depth, struct mm_walk *walk) 525 { 526 struct mem_size_stats *mss = walk->private; 527 struct vm_area_struct *vma = walk->vma; 528 529 mss->swap += shmem_partial_swap_usage(walk->vma->vm_file->f_mapping, 530 linear_page_index(vma, addr), 531 linear_page_index(vma, end)); 532 533 return 0; 534 } 535 #else 536 #define smaps_pte_hole NULL 537 #endif /* CONFIG_SHMEM */ 538 539 static void smaps_pte_hole_lookup(unsigned long addr, struct mm_walk *walk) 540 { 541 #ifdef CONFIG_SHMEM 542 if (walk->ops->pte_hole) { 543 /* depth is not used */ 544 smaps_pte_hole(addr, addr + PAGE_SIZE, 0, walk); 545 } 546 #endif 547 } 548 549 static void smaps_pte_entry(pte_t *pte, unsigned long addr, 550 struct mm_walk *walk) 551 { 552 struct mem_size_stats *mss = walk->private; 553 struct vm_area_struct *vma = walk->vma; 554 bool locked = !!(vma->vm_flags & VM_LOCKED); 555 struct page *page = NULL; 556 bool present = false, young = false, dirty = false; 557 pte_t ptent = ptep_get(pte); 558 559 if (pte_present(ptent)) { 560 page = vm_normal_page(vma, addr, ptent); 561 young = pte_young(ptent); 562 dirty = pte_dirty(ptent); 563 present = true; 564 } else if (is_swap_pte(ptent)) { 565 swp_entry_t swpent = pte_to_swp_entry(ptent); 566 567 if (!non_swap_entry(swpent)) { 568 int mapcount; 569 570 mss->swap += PAGE_SIZE; 571 mapcount = swp_swapcount(swpent); 572 if (mapcount >= 2) { 573 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT; 574 575 do_div(pss_delta, mapcount); 576 mss->swap_pss += pss_delta; 577 } else { 578 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT; 579 } 580 } else if (is_pfn_swap_entry(swpent)) { 581 if (is_device_private_entry(swpent)) 582 present = true; 583 page = pfn_swap_entry_to_page(swpent); 584 } 585 } else { 586 smaps_pte_hole_lookup(addr, walk); 587 return; 588 } 589 590 if (!page) 591 return; 592 593 smaps_account(mss, page, false, young, dirty, locked, present); 594 } 595 596 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 597 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr, 598 struct mm_walk *walk) 599 { 600 struct mem_size_stats *mss = walk->private; 601 struct vm_area_struct *vma = walk->vma; 602 bool locked = !!(vma->vm_flags & VM_LOCKED); 603 struct page *page = NULL; 604 bool present = false; 605 struct folio *folio; 606 607 if (pmd_present(*pmd)) { 608 page = vm_normal_page_pmd(vma, addr, *pmd); 609 present = true; 610 } else if (unlikely(thp_migration_supported() && is_swap_pmd(*pmd))) { 611 swp_entry_t entry = pmd_to_swp_entry(*pmd); 612 613 if (is_pfn_swap_entry(entry)) 614 page = pfn_swap_entry_to_page(entry); 615 } 616 if (IS_ERR_OR_NULL(page)) 617 return; 618 folio = page_folio(page); 619 if (folio_test_anon(folio)) 620 mss->anonymous_thp += HPAGE_PMD_SIZE; 621 else if (folio_test_swapbacked(folio)) 622 mss->shmem_thp += HPAGE_PMD_SIZE; 623 else if (folio_is_zone_device(folio)) 624 /* pass */; 625 else 626 mss->file_thp += HPAGE_PMD_SIZE; 627 628 smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd), 629 locked, present); 630 } 631 #else 632 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr, 633 struct mm_walk *walk) 634 { 635 } 636 #endif 637 638 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, 639 struct mm_walk *walk) 640 { 641 struct vm_area_struct *vma = walk->vma; 642 pte_t *pte; 643 spinlock_t *ptl; 644 645 ptl = pmd_trans_huge_lock(pmd, vma); 646 if (ptl) { 647 smaps_pmd_entry(pmd, addr, walk); 648 spin_unlock(ptl); 649 goto out; 650 } 651 652 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); 653 if (!pte) { 654 walk->action = ACTION_AGAIN; 655 return 0; 656 } 657 for (; addr != end; pte++, addr += PAGE_SIZE) 658 smaps_pte_entry(pte, addr, walk); 659 pte_unmap_unlock(pte - 1, ptl); 660 out: 661 cond_resched(); 662 return 0; 663 } 664 665 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma) 666 { 667 /* 668 * Don't forget to update Documentation/ on changes. 669 */ 670 static const char mnemonics[BITS_PER_LONG][2] = { 671 /* 672 * In case if we meet a flag we don't know about. 673 */ 674 [0 ... (BITS_PER_LONG-1)] = "??", 675 676 [ilog2(VM_READ)] = "rd", 677 [ilog2(VM_WRITE)] = "wr", 678 [ilog2(VM_EXEC)] = "ex", 679 [ilog2(VM_SHARED)] = "sh", 680 [ilog2(VM_MAYREAD)] = "mr", 681 [ilog2(VM_MAYWRITE)] = "mw", 682 [ilog2(VM_MAYEXEC)] = "me", 683 [ilog2(VM_MAYSHARE)] = "ms", 684 [ilog2(VM_GROWSDOWN)] = "gd", 685 [ilog2(VM_PFNMAP)] = "pf", 686 [ilog2(VM_LOCKED)] = "lo", 687 [ilog2(VM_IO)] = "io", 688 [ilog2(VM_SEQ_READ)] = "sr", 689 [ilog2(VM_RAND_READ)] = "rr", 690 [ilog2(VM_DONTCOPY)] = "dc", 691 [ilog2(VM_DONTEXPAND)] = "de", 692 [ilog2(VM_LOCKONFAULT)] = "lf", 693 [ilog2(VM_ACCOUNT)] = "ac", 694 [ilog2(VM_NORESERVE)] = "nr", 695 [ilog2(VM_HUGETLB)] = "ht", 696 [ilog2(VM_SYNC)] = "sf", 697 [ilog2(VM_ARCH_1)] = "ar", 698 [ilog2(VM_WIPEONFORK)] = "wf", 699 [ilog2(VM_DONTDUMP)] = "dd", 700 #ifdef CONFIG_ARM64_BTI 701 [ilog2(VM_ARM64_BTI)] = "bt", 702 #endif 703 #ifdef CONFIG_MEM_SOFT_DIRTY 704 [ilog2(VM_SOFTDIRTY)] = "sd", 705 #endif 706 [ilog2(VM_MIXEDMAP)] = "mm", 707 [ilog2(VM_HUGEPAGE)] = "hg", 708 [ilog2(VM_NOHUGEPAGE)] = "nh", 709 [ilog2(VM_MERGEABLE)] = "mg", 710 [ilog2(VM_UFFD_MISSING)]= "um", 711 [ilog2(VM_UFFD_WP)] = "uw", 712 #ifdef CONFIG_ARM64_MTE 713 [ilog2(VM_MTE)] = "mt", 714 [ilog2(VM_MTE_ALLOWED)] = "", 715 #endif 716 #ifdef CONFIG_ARCH_HAS_PKEYS 717 /* These come out via ProtectionKey: */ 718 [ilog2(VM_PKEY_BIT0)] = "", 719 [ilog2(VM_PKEY_BIT1)] = "", 720 [ilog2(VM_PKEY_BIT2)] = "", 721 [ilog2(VM_PKEY_BIT3)] = "", 722 #if VM_PKEY_BIT4 723 [ilog2(VM_PKEY_BIT4)] = "", 724 #endif 725 #endif /* CONFIG_ARCH_HAS_PKEYS */ 726 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_MINOR 727 [ilog2(VM_UFFD_MINOR)] = "ui", 728 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_MINOR */ 729 #ifdef CONFIG_X86_USER_SHADOW_STACK 730 [ilog2(VM_SHADOW_STACK)] = "ss", 731 #endif 732 #ifdef CONFIG_64BIT 733 [ilog2(VM_SEALED)] = "sl", 734 #endif 735 }; 736 size_t i; 737 738 seq_puts(m, "VmFlags: "); 739 for (i = 0; i < BITS_PER_LONG; i++) { 740 if (!mnemonics[i][0]) 741 continue; 742 if (vma->vm_flags & (1UL << i)) { 743 seq_putc(m, mnemonics[i][0]); 744 seq_putc(m, mnemonics[i][1]); 745 seq_putc(m, ' '); 746 } 747 } 748 seq_putc(m, '\n'); 749 } 750 751 #ifdef CONFIG_HUGETLB_PAGE 752 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask, 753 unsigned long addr, unsigned long end, 754 struct mm_walk *walk) 755 { 756 struct mem_size_stats *mss = walk->private; 757 struct vm_area_struct *vma = walk->vma; 758 pte_t ptent = huge_ptep_get(pte); 759 struct folio *folio = NULL; 760 bool present = false; 761 762 if (pte_present(ptent)) { 763 folio = page_folio(pte_page(ptent)); 764 present = true; 765 } else if (is_swap_pte(ptent)) { 766 swp_entry_t swpent = pte_to_swp_entry(ptent); 767 768 if (is_pfn_swap_entry(swpent)) 769 folio = pfn_swap_entry_folio(swpent); 770 } 771 772 if (folio) { 773 /* We treat non-present entries as "maybe shared". */ 774 if (!present || folio_likely_mapped_shared(folio) || 775 hugetlb_pmd_shared(pte)) 776 mss->shared_hugetlb += huge_page_size(hstate_vma(vma)); 777 else 778 mss->private_hugetlb += huge_page_size(hstate_vma(vma)); 779 } 780 return 0; 781 } 782 #else 783 #define smaps_hugetlb_range NULL 784 #endif /* HUGETLB_PAGE */ 785 786 static const struct mm_walk_ops smaps_walk_ops = { 787 .pmd_entry = smaps_pte_range, 788 .hugetlb_entry = smaps_hugetlb_range, 789 .walk_lock = PGWALK_RDLOCK, 790 }; 791 792 static const struct mm_walk_ops smaps_shmem_walk_ops = { 793 .pmd_entry = smaps_pte_range, 794 .hugetlb_entry = smaps_hugetlb_range, 795 .pte_hole = smaps_pte_hole, 796 .walk_lock = PGWALK_RDLOCK, 797 }; 798 799 /* 800 * Gather mem stats from @vma with the indicated beginning 801 * address @start, and keep them in @mss. 802 * 803 * Use vm_start of @vma as the beginning address if @start is 0. 804 */ 805 static void smap_gather_stats(struct vm_area_struct *vma, 806 struct mem_size_stats *mss, unsigned long start) 807 { 808 const struct mm_walk_ops *ops = &smaps_walk_ops; 809 810 /* Invalid start */ 811 if (start >= vma->vm_end) 812 return; 813 814 if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) { 815 /* 816 * For shared or readonly shmem mappings we know that all 817 * swapped out pages belong to the shmem object, and we can 818 * obtain the swap value much more efficiently. For private 819 * writable mappings, we might have COW pages that are 820 * not affected by the parent swapped out pages of the shmem 821 * object, so we have to distinguish them during the page walk. 822 * Unless we know that the shmem object (or the part mapped by 823 * our VMA) has no swapped out pages at all. 824 */ 825 unsigned long shmem_swapped = shmem_swap_usage(vma); 826 827 if (!start && (!shmem_swapped || (vma->vm_flags & VM_SHARED) || 828 !(vma->vm_flags & VM_WRITE))) { 829 mss->swap += shmem_swapped; 830 } else { 831 ops = &smaps_shmem_walk_ops; 832 } 833 } 834 835 /* mmap_lock is held in m_start */ 836 if (!start) 837 walk_page_vma(vma, ops, mss); 838 else 839 walk_page_range(vma->vm_mm, start, vma->vm_end, ops, mss); 840 } 841 842 #define SEQ_PUT_DEC(str, val) \ 843 seq_put_decimal_ull_width(m, str, (val) >> 10, 8) 844 845 /* Show the contents common for smaps and smaps_rollup */ 846 static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss, 847 bool rollup_mode) 848 { 849 SEQ_PUT_DEC("Rss: ", mss->resident); 850 SEQ_PUT_DEC(" kB\nPss: ", mss->pss >> PSS_SHIFT); 851 SEQ_PUT_DEC(" kB\nPss_Dirty: ", mss->pss_dirty >> PSS_SHIFT); 852 if (rollup_mode) { 853 /* 854 * These are meaningful only for smaps_rollup, otherwise two of 855 * them are zero, and the other one is the same as Pss. 856 */ 857 SEQ_PUT_DEC(" kB\nPss_Anon: ", 858 mss->pss_anon >> PSS_SHIFT); 859 SEQ_PUT_DEC(" kB\nPss_File: ", 860 mss->pss_file >> PSS_SHIFT); 861 SEQ_PUT_DEC(" kB\nPss_Shmem: ", 862 mss->pss_shmem >> PSS_SHIFT); 863 } 864 SEQ_PUT_DEC(" kB\nShared_Clean: ", mss->shared_clean); 865 SEQ_PUT_DEC(" kB\nShared_Dirty: ", mss->shared_dirty); 866 SEQ_PUT_DEC(" kB\nPrivate_Clean: ", mss->private_clean); 867 SEQ_PUT_DEC(" kB\nPrivate_Dirty: ", mss->private_dirty); 868 SEQ_PUT_DEC(" kB\nReferenced: ", mss->referenced); 869 SEQ_PUT_DEC(" kB\nAnonymous: ", mss->anonymous); 870 SEQ_PUT_DEC(" kB\nKSM: ", mss->ksm); 871 SEQ_PUT_DEC(" kB\nLazyFree: ", mss->lazyfree); 872 SEQ_PUT_DEC(" kB\nAnonHugePages: ", mss->anonymous_thp); 873 SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp); 874 SEQ_PUT_DEC(" kB\nFilePmdMapped: ", mss->file_thp); 875 SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb); 876 seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ", 877 mss->private_hugetlb >> 10, 7); 878 SEQ_PUT_DEC(" kB\nSwap: ", mss->swap); 879 SEQ_PUT_DEC(" kB\nSwapPss: ", 880 mss->swap_pss >> PSS_SHIFT); 881 SEQ_PUT_DEC(" kB\nLocked: ", 882 mss->pss_locked >> PSS_SHIFT); 883 seq_puts(m, " kB\n"); 884 } 885 886 static int show_smap(struct seq_file *m, void *v) 887 { 888 struct vm_area_struct *vma = v; 889 struct mem_size_stats mss = {}; 890 891 smap_gather_stats(vma, &mss, 0); 892 893 show_map_vma(m, vma); 894 895 SEQ_PUT_DEC("Size: ", vma->vm_end - vma->vm_start); 896 SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma)); 897 SEQ_PUT_DEC(" kB\nMMUPageSize: ", vma_mmu_pagesize(vma)); 898 seq_puts(m, " kB\n"); 899 900 __show_smap(m, &mss, false); 901 902 seq_printf(m, "THPeligible: %8u\n", 903 !!thp_vma_allowable_orders(vma, vma->vm_flags, 904 TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL)); 905 906 if (arch_pkeys_enabled()) 907 seq_printf(m, "ProtectionKey: %8u\n", vma_pkey(vma)); 908 show_smap_vma_flags(m, vma); 909 910 return 0; 911 } 912 913 static int show_smaps_rollup(struct seq_file *m, void *v) 914 { 915 struct proc_maps_private *priv = m->private; 916 struct mem_size_stats mss = {}; 917 struct mm_struct *mm = priv->mm; 918 struct vm_area_struct *vma; 919 unsigned long vma_start = 0, last_vma_end = 0; 920 int ret = 0; 921 VMA_ITERATOR(vmi, mm, 0); 922 923 priv->task = get_proc_task(priv->inode); 924 if (!priv->task) 925 return -ESRCH; 926 927 if (!mm || !mmget_not_zero(mm)) { 928 ret = -ESRCH; 929 goto out_put_task; 930 } 931 932 ret = mmap_read_lock_killable(mm); 933 if (ret) 934 goto out_put_mm; 935 936 hold_task_mempolicy(priv); 937 vma = vma_next(&vmi); 938 939 if (unlikely(!vma)) 940 goto empty_set; 941 942 vma_start = vma->vm_start; 943 do { 944 smap_gather_stats(vma, &mss, 0); 945 last_vma_end = vma->vm_end; 946 947 /* 948 * Release mmap_lock temporarily if someone wants to 949 * access it for write request. 950 */ 951 if (mmap_lock_is_contended(mm)) { 952 vma_iter_invalidate(&vmi); 953 mmap_read_unlock(mm); 954 ret = mmap_read_lock_killable(mm); 955 if (ret) { 956 release_task_mempolicy(priv); 957 goto out_put_mm; 958 } 959 960 /* 961 * After dropping the lock, there are four cases to 962 * consider. See the following example for explanation. 963 * 964 * +------+------+-----------+ 965 * | VMA1 | VMA2 | VMA3 | 966 * +------+------+-----------+ 967 * | | | | 968 * 4k 8k 16k 400k 969 * 970 * Suppose we drop the lock after reading VMA2 due to 971 * contention, then we get: 972 * 973 * last_vma_end = 16k 974 * 975 * 1) VMA2 is freed, but VMA3 exists: 976 * 977 * vma_next(vmi) will return VMA3. 978 * In this case, just continue from VMA3. 979 * 980 * 2) VMA2 still exists: 981 * 982 * vma_next(vmi) will return VMA3. 983 * In this case, just continue from VMA3. 984 * 985 * 3) No more VMAs can be found: 986 * 987 * vma_next(vmi) will return NULL. 988 * No more things to do, just break. 989 * 990 * 4) (last_vma_end - 1) is the middle of a vma (VMA'): 991 * 992 * vma_next(vmi) will return VMA' whose range 993 * contains last_vma_end. 994 * Iterate VMA' from last_vma_end. 995 */ 996 vma = vma_next(&vmi); 997 /* Case 3 above */ 998 if (!vma) 999 break; 1000 1001 /* Case 1 and 2 above */ 1002 if (vma->vm_start >= last_vma_end) { 1003 smap_gather_stats(vma, &mss, 0); 1004 last_vma_end = vma->vm_end; 1005 continue; 1006 } 1007 1008 /* Case 4 above */ 1009 if (vma->vm_end > last_vma_end) { 1010 smap_gather_stats(vma, &mss, last_vma_end); 1011 last_vma_end = vma->vm_end; 1012 } 1013 } 1014 } for_each_vma(vmi, vma); 1015 1016 empty_set: 1017 show_vma_header_prefix(m, vma_start, last_vma_end, 0, 0, 0, 0); 1018 seq_pad(m, ' '); 1019 seq_puts(m, "[rollup]\n"); 1020 1021 __show_smap(m, &mss, true); 1022 1023 release_task_mempolicy(priv); 1024 mmap_read_unlock(mm); 1025 1026 out_put_mm: 1027 mmput(mm); 1028 out_put_task: 1029 put_task_struct(priv->task); 1030 priv->task = NULL; 1031 1032 return ret; 1033 } 1034 #undef SEQ_PUT_DEC 1035 1036 static const struct seq_operations proc_pid_smaps_op = { 1037 .start = m_start, 1038 .next = m_next, 1039 .stop = m_stop, 1040 .show = show_smap 1041 }; 1042 1043 static int pid_smaps_open(struct inode *inode, struct file *file) 1044 { 1045 return do_maps_open(inode, file, &proc_pid_smaps_op); 1046 } 1047 1048 static int smaps_rollup_open(struct inode *inode, struct file *file) 1049 { 1050 int ret; 1051 struct proc_maps_private *priv; 1052 1053 priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT); 1054 if (!priv) 1055 return -ENOMEM; 1056 1057 ret = single_open(file, show_smaps_rollup, priv); 1058 if (ret) 1059 goto out_free; 1060 1061 priv->inode = inode; 1062 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ); 1063 if (IS_ERR(priv->mm)) { 1064 ret = PTR_ERR(priv->mm); 1065 1066 single_release(inode, file); 1067 goto out_free; 1068 } 1069 1070 return 0; 1071 1072 out_free: 1073 kfree(priv); 1074 return ret; 1075 } 1076 1077 static int smaps_rollup_release(struct inode *inode, struct file *file) 1078 { 1079 struct seq_file *seq = file->private_data; 1080 struct proc_maps_private *priv = seq->private; 1081 1082 if (priv->mm) 1083 mmdrop(priv->mm); 1084 1085 kfree(priv); 1086 return single_release(inode, file); 1087 } 1088 1089 const struct file_operations proc_pid_smaps_operations = { 1090 .open = pid_smaps_open, 1091 .read = seq_read, 1092 .llseek = seq_lseek, 1093 .release = proc_map_release, 1094 }; 1095 1096 const struct file_operations proc_pid_smaps_rollup_operations = { 1097 .open = smaps_rollup_open, 1098 .read = seq_read, 1099 .llseek = seq_lseek, 1100 .release = smaps_rollup_release, 1101 }; 1102 1103 enum clear_refs_types { 1104 CLEAR_REFS_ALL = 1, 1105 CLEAR_REFS_ANON, 1106 CLEAR_REFS_MAPPED, 1107 CLEAR_REFS_SOFT_DIRTY, 1108 CLEAR_REFS_MM_HIWATER_RSS, 1109 CLEAR_REFS_LAST, 1110 }; 1111 1112 struct clear_refs_private { 1113 enum clear_refs_types type; 1114 }; 1115 1116 #ifdef CONFIG_MEM_SOFT_DIRTY 1117 1118 static inline bool pte_is_pinned(struct vm_area_struct *vma, unsigned long addr, pte_t pte) 1119 { 1120 struct folio *folio; 1121 1122 if (!pte_write(pte)) 1123 return false; 1124 if (!is_cow_mapping(vma->vm_flags)) 1125 return false; 1126 if (likely(!test_bit(MMF_HAS_PINNED, &vma->vm_mm->flags))) 1127 return false; 1128 folio = vm_normal_folio(vma, addr, pte); 1129 if (!folio) 1130 return false; 1131 return folio_maybe_dma_pinned(folio); 1132 } 1133 1134 static inline void clear_soft_dirty(struct vm_area_struct *vma, 1135 unsigned long addr, pte_t *pte) 1136 { 1137 /* 1138 * The soft-dirty tracker uses #PF-s to catch writes 1139 * to pages, so write-protect the pte as well. See the 1140 * Documentation/admin-guide/mm/soft-dirty.rst for full description 1141 * of how soft-dirty works. 1142 */ 1143 pte_t ptent = ptep_get(pte); 1144 1145 if (pte_present(ptent)) { 1146 pte_t old_pte; 1147 1148 if (pte_is_pinned(vma, addr, ptent)) 1149 return; 1150 old_pte = ptep_modify_prot_start(vma, addr, pte); 1151 ptent = pte_wrprotect(old_pte); 1152 ptent = pte_clear_soft_dirty(ptent); 1153 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent); 1154 } else if (is_swap_pte(ptent)) { 1155 ptent = pte_swp_clear_soft_dirty(ptent); 1156 set_pte_at(vma->vm_mm, addr, pte, ptent); 1157 } 1158 } 1159 #else 1160 static inline void clear_soft_dirty(struct vm_area_struct *vma, 1161 unsigned long addr, pte_t *pte) 1162 { 1163 } 1164 #endif 1165 1166 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE) 1167 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma, 1168 unsigned long addr, pmd_t *pmdp) 1169 { 1170 pmd_t old, pmd = *pmdp; 1171 1172 if (pmd_present(pmd)) { 1173 /* See comment in change_huge_pmd() */ 1174 old = pmdp_invalidate(vma, addr, pmdp); 1175 if (pmd_dirty(old)) 1176 pmd = pmd_mkdirty(pmd); 1177 if (pmd_young(old)) 1178 pmd = pmd_mkyoung(pmd); 1179 1180 pmd = pmd_wrprotect(pmd); 1181 pmd = pmd_clear_soft_dirty(pmd); 1182 1183 set_pmd_at(vma->vm_mm, addr, pmdp, pmd); 1184 } else if (is_migration_entry(pmd_to_swp_entry(pmd))) { 1185 pmd = pmd_swp_clear_soft_dirty(pmd); 1186 set_pmd_at(vma->vm_mm, addr, pmdp, pmd); 1187 } 1188 } 1189 #else 1190 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma, 1191 unsigned long addr, pmd_t *pmdp) 1192 { 1193 } 1194 #endif 1195 1196 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr, 1197 unsigned long end, struct mm_walk *walk) 1198 { 1199 struct clear_refs_private *cp = walk->private; 1200 struct vm_area_struct *vma = walk->vma; 1201 pte_t *pte, ptent; 1202 spinlock_t *ptl; 1203 struct folio *folio; 1204 1205 ptl = pmd_trans_huge_lock(pmd, vma); 1206 if (ptl) { 1207 if (cp->type == CLEAR_REFS_SOFT_DIRTY) { 1208 clear_soft_dirty_pmd(vma, addr, pmd); 1209 goto out; 1210 } 1211 1212 if (!pmd_present(*pmd)) 1213 goto out; 1214 1215 folio = pmd_folio(*pmd); 1216 1217 /* Clear accessed and referenced bits. */ 1218 pmdp_test_and_clear_young(vma, addr, pmd); 1219 folio_test_clear_young(folio); 1220 folio_clear_referenced(folio); 1221 out: 1222 spin_unlock(ptl); 1223 return 0; 1224 } 1225 1226 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); 1227 if (!pte) { 1228 walk->action = ACTION_AGAIN; 1229 return 0; 1230 } 1231 for (; addr != end; pte++, addr += PAGE_SIZE) { 1232 ptent = ptep_get(pte); 1233 1234 if (cp->type == CLEAR_REFS_SOFT_DIRTY) { 1235 clear_soft_dirty(vma, addr, pte); 1236 continue; 1237 } 1238 1239 if (!pte_present(ptent)) 1240 continue; 1241 1242 folio = vm_normal_folio(vma, addr, ptent); 1243 if (!folio) 1244 continue; 1245 1246 /* Clear accessed and referenced bits. */ 1247 ptep_test_and_clear_young(vma, addr, pte); 1248 folio_test_clear_young(folio); 1249 folio_clear_referenced(folio); 1250 } 1251 pte_unmap_unlock(pte - 1, ptl); 1252 cond_resched(); 1253 return 0; 1254 } 1255 1256 static int clear_refs_test_walk(unsigned long start, unsigned long end, 1257 struct mm_walk *walk) 1258 { 1259 struct clear_refs_private *cp = walk->private; 1260 struct vm_area_struct *vma = walk->vma; 1261 1262 if (vma->vm_flags & VM_PFNMAP) 1263 return 1; 1264 1265 /* 1266 * Writing 1 to /proc/pid/clear_refs affects all pages. 1267 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages. 1268 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages. 1269 * Writing 4 to /proc/pid/clear_refs affects all pages. 1270 */ 1271 if (cp->type == CLEAR_REFS_ANON && vma->vm_file) 1272 return 1; 1273 if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file) 1274 return 1; 1275 return 0; 1276 } 1277 1278 static const struct mm_walk_ops clear_refs_walk_ops = { 1279 .pmd_entry = clear_refs_pte_range, 1280 .test_walk = clear_refs_test_walk, 1281 .walk_lock = PGWALK_WRLOCK, 1282 }; 1283 1284 static ssize_t clear_refs_write(struct file *file, const char __user *buf, 1285 size_t count, loff_t *ppos) 1286 { 1287 struct task_struct *task; 1288 char buffer[PROC_NUMBUF] = {}; 1289 struct mm_struct *mm; 1290 struct vm_area_struct *vma; 1291 enum clear_refs_types type; 1292 int itype; 1293 int rv; 1294 1295 if (count > sizeof(buffer) - 1) 1296 count = sizeof(buffer) - 1; 1297 if (copy_from_user(buffer, buf, count)) 1298 return -EFAULT; 1299 rv = kstrtoint(strstrip(buffer), 10, &itype); 1300 if (rv < 0) 1301 return rv; 1302 type = (enum clear_refs_types)itype; 1303 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST) 1304 return -EINVAL; 1305 1306 task = get_proc_task(file_inode(file)); 1307 if (!task) 1308 return -ESRCH; 1309 mm = get_task_mm(task); 1310 if (mm) { 1311 VMA_ITERATOR(vmi, mm, 0); 1312 struct mmu_notifier_range range; 1313 struct clear_refs_private cp = { 1314 .type = type, 1315 }; 1316 1317 if (mmap_write_lock_killable(mm)) { 1318 count = -EINTR; 1319 goto out_mm; 1320 } 1321 if (type == CLEAR_REFS_MM_HIWATER_RSS) { 1322 /* 1323 * Writing 5 to /proc/pid/clear_refs resets the peak 1324 * resident set size to this mm's current rss value. 1325 */ 1326 reset_mm_hiwater_rss(mm); 1327 goto out_unlock; 1328 } 1329 1330 if (type == CLEAR_REFS_SOFT_DIRTY) { 1331 for_each_vma(vmi, vma) { 1332 if (!(vma->vm_flags & VM_SOFTDIRTY)) 1333 continue; 1334 vm_flags_clear(vma, VM_SOFTDIRTY); 1335 vma_set_page_prot(vma); 1336 } 1337 1338 inc_tlb_flush_pending(mm); 1339 mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY, 1340 0, mm, 0, -1UL); 1341 mmu_notifier_invalidate_range_start(&range); 1342 } 1343 walk_page_range(mm, 0, -1, &clear_refs_walk_ops, &cp); 1344 if (type == CLEAR_REFS_SOFT_DIRTY) { 1345 mmu_notifier_invalidate_range_end(&range); 1346 flush_tlb_mm(mm); 1347 dec_tlb_flush_pending(mm); 1348 } 1349 out_unlock: 1350 mmap_write_unlock(mm); 1351 out_mm: 1352 mmput(mm); 1353 } 1354 put_task_struct(task); 1355 1356 return count; 1357 } 1358 1359 const struct file_operations proc_clear_refs_operations = { 1360 .write = clear_refs_write, 1361 .llseek = noop_llseek, 1362 }; 1363 1364 typedef struct { 1365 u64 pme; 1366 } pagemap_entry_t; 1367 1368 struct pagemapread { 1369 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */ 1370 pagemap_entry_t *buffer; 1371 bool show_pfn; 1372 }; 1373 1374 #define PAGEMAP_WALK_SIZE (PMD_SIZE) 1375 #define PAGEMAP_WALK_MASK (PMD_MASK) 1376 1377 #define PM_ENTRY_BYTES sizeof(pagemap_entry_t) 1378 #define PM_PFRAME_BITS 55 1379 #define PM_PFRAME_MASK GENMASK_ULL(PM_PFRAME_BITS - 1, 0) 1380 #define PM_SOFT_DIRTY BIT_ULL(55) 1381 #define PM_MMAP_EXCLUSIVE BIT_ULL(56) 1382 #define PM_UFFD_WP BIT_ULL(57) 1383 #define PM_FILE BIT_ULL(61) 1384 #define PM_SWAP BIT_ULL(62) 1385 #define PM_PRESENT BIT_ULL(63) 1386 1387 #define PM_END_OF_BUFFER 1 1388 1389 static inline pagemap_entry_t make_pme(u64 frame, u64 flags) 1390 { 1391 return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags }; 1392 } 1393 1394 static int add_to_pagemap(pagemap_entry_t *pme, struct pagemapread *pm) 1395 { 1396 pm->buffer[pm->pos++] = *pme; 1397 if (pm->pos >= pm->len) 1398 return PM_END_OF_BUFFER; 1399 return 0; 1400 } 1401 1402 static int pagemap_pte_hole(unsigned long start, unsigned long end, 1403 __always_unused int depth, struct mm_walk *walk) 1404 { 1405 struct pagemapread *pm = walk->private; 1406 unsigned long addr = start; 1407 int err = 0; 1408 1409 while (addr < end) { 1410 struct vm_area_struct *vma = find_vma(walk->mm, addr); 1411 pagemap_entry_t pme = make_pme(0, 0); 1412 /* End of address space hole, which we mark as non-present. */ 1413 unsigned long hole_end; 1414 1415 if (vma) 1416 hole_end = min(end, vma->vm_start); 1417 else 1418 hole_end = end; 1419 1420 for (; addr < hole_end; addr += PAGE_SIZE) { 1421 err = add_to_pagemap(&pme, pm); 1422 if (err) 1423 goto out; 1424 } 1425 1426 if (!vma) 1427 break; 1428 1429 /* Addresses in the VMA. */ 1430 if (vma->vm_flags & VM_SOFTDIRTY) 1431 pme = make_pme(0, PM_SOFT_DIRTY); 1432 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) { 1433 err = add_to_pagemap(&pme, pm); 1434 if (err) 1435 goto out; 1436 } 1437 } 1438 out: 1439 return err; 1440 } 1441 1442 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm, 1443 struct vm_area_struct *vma, unsigned long addr, pte_t pte) 1444 { 1445 u64 frame = 0, flags = 0; 1446 struct page *page = NULL; 1447 struct folio *folio; 1448 1449 if (pte_present(pte)) { 1450 if (pm->show_pfn) 1451 frame = pte_pfn(pte); 1452 flags |= PM_PRESENT; 1453 page = vm_normal_page(vma, addr, pte); 1454 if (pte_soft_dirty(pte)) 1455 flags |= PM_SOFT_DIRTY; 1456 if (pte_uffd_wp(pte)) 1457 flags |= PM_UFFD_WP; 1458 } else if (is_swap_pte(pte)) { 1459 swp_entry_t entry; 1460 if (pte_swp_soft_dirty(pte)) 1461 flags |= PM_SOFT_DIRTY; 1462 if (pte_swp_uffd_wp(pte)) 1463 flags |= PM_UFFD_WP; 1464 entry = pte_to_swp_entry(pte); 1465 if (pm->show_pfn) { 1466 pgoff_t offset; 1467 /* 1468 * For PFN swap offsets, keeping the offset field 1469 * to be PFN only to be compatible with old smaps. 1470 */ 1471 if (is_pfn_swap_entry(entry)) 1472 offset = swp_offset_pfn(entry); 1473 else 1474 offset = swp_offset(entry); 1475 frame = swp_type(entry) | 1476 (offset << MAX_SWAPFILES_SHIFT); 1477 } 1478 flags |= PM_SWAP; 1479 if (is_pfn_swap_entry(entry)) 1480 page = pfn_swap_entry_to_page(entry); 1481 if (pte_marker_entry_uffd_wp(entry)) 1482 flags |= PM_UFFD_WP; 1483 } 1484 1485 if (page) { 1486 folio = page_folio(page); 1487 if (!folio_test_anon(folio)) 1488 flags |= PM_FILE; 1489 if ((flags & PM_PRESENT) && 1490 folio_precise_page_mapcount(folio, page) == 1) 1491 flags |= PM_MMAP_EXCLUSIVE; 1492 } 1493 if (vma->vm_flags & VM_SOFTDIRTY) 1494 flags |= PM_SOFT_DIRTY; 1495 1496 return make_pme(frame, flags); 1497 } 1498 1499 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end, 1500 struct mm_walk *walk) 1501 { 1502 struct vm_area_struct *vma = walk->vma; 1503 struct pagemapread *pm = walk->private; 1504 spinlock_t *ptl; 1505 pte_t *pte, *orig_pte; 1506 int err = 0; 1507 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1508 1509 ptl = pmd_trans_huge_lock(pmdp, vma); 1510 if (ptl) { 1511 unsigned int idx = (addr & ~PMD_MASK) >> PAGE_SHIFT; 1512 u64 flags = 0, frame = 0; 1513 pmd_t pmd = *pmdp; 1514 struct page *page = NULL; 1515 struct folio *folio = NULL; 1516 1517 if (vma->vm_flags & VM_SOFTDIRTY) 1518 flags |= PM_SOFT_DIRTY; 1519 1520 if (pmd_present(pmd)) { 1521 page = pmd_page(pmd); 1522 1523 flags |= PM_PRESENT; 1524 if (pmd_soft_dirty(pmd)) 1525 flags |= PM_SOFT_DIRTY; 1526 if (pmd_uffd_wp(pmd)) 1527 flags |= PM_UFFD_WP; 1528 if (pm->show_pfn) 1529 frame = pmd_pfn(pmd) + idx; 1530 } 1531 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION 1532 else if (is_swap_pmd(pmd)) { 1533 swp_entry_t entry = pmd_to_swp_entry(pmd); 1534 unsigned long offset; 1535 1536 if (pm->show_pfn) { 1537 if (is_pfn_swap_entry(entry)) 1538 offset = swp_offset_pfn(entry) + idx; 1539 else 1540 offset = swp_offset(entry) + idx; 1541 frame = swp_type(entry) | 1542 (offset << MAX_SWAPFILES_SHIFT); 1543 } 1544 flags |= PM_SWAP; 1545 if (pmd_swp_soft_dirty(pmd)) 1546 flags |= PM_SOFT_DIRTY; 1547 if (pmd_swp_uffd_wp(pmd)) 1548 flags |= PM_UFFD_WP; 1549 VM_BUG_ON(!is_pmd_migration_entry(pmd)); 1550 page = pfn_swap_entry_to_page(entry); 1551 } 1552 #endif 1553 1554 if (page) { 1555 folio = page_folio(page); 1556 if (!folio_test_anon(folio)) 1557 flags |= PM_FILE; 1558 } 1559 1560 for (; addr != end; addr += PAGE_SIZE, idx++) { 1561 unsigned long cur_flags = flags; 1562 pagemap_entry_t pme; 1563 1564 if (folio && (flags & PM_PRESENT) && 1565 folio_precise_page_mapcount(folio, page + idx) == 1) 1566 cur_flags |= PM_MMAP_EXCLUSIVE; 1567 1568 pme = make_pme(frame, cur_flags); 1569 err = add_to_pagemap(&pme, pm); 1570 if (err) 1571 break; 1572 if (pm->show_pfn) { 1573 if (flags & PM_PRESENT) 1574 frame++; 1575 else if (flags & PM_SWAP) 1576 frame += (1 << MAX_SWAPFILES_SHIFT); 1577 } 1578 } 1579 spin_unlock(ptl); 1580 return err; 1581 } 1582 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ 1583 1584 /* 1585 * We can assume that @vma always points to a valid one and @end never 1586 * goes beyond vma->vm_end. 1587 */ 1588 orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl); 1589 if (!pte) { 1590 walk->action = ACTION_AGAIN; 1591 return err; 1592 } 1593 for (; addr < end; pte++, addr += PAGE_SIZE) { 1594 pagemap_entry_t pme; 1595 1596 pme = pte_to_pagemap_entry(pm, vma, addr, ptep_get(pte)); 1597 err = add_to_pagemap(&pme, pm); 1598 if (err) 1599 break; 1600 } 1601 pte_unmap_unlock(orig_pte, ptl); 1602 1603 cond_resched(); 1604 1605 return err; 1606 } 1607 1608 #ifdef CONFIG_HUGETLB_PAGE 1609 /* This function walks within one hugetlb entry in the single call */ 1610 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask, 1611 unsigned long addr, unsigned long end, 1612 struct mm_walk *walk) 1613 { 1614 struct pagemapread *pm = walk->private; 1615 struct vm_area_struct *vma = walk->vma; 1616 u64 flags = 0, frame = 0; 1617 int err = 0; 1618 pte_t pte; 1619 1620 if (vma->vm_flags & VM_SOFTDIRTY) 1621 flags |= PM_SOFT_DIRTY; 1622 1623 pte = huge_ptep_get(ptep); 1624 if (pte_present(pte)) { 1625 struct folio *folio = page_folio(pte_page(pte)); 1626 1627 if (!folio_test_anon(folio)) 1628 flags |= PM_FILE; 1629 1630 if (!folio_likely_mapped_shared(folio) && 1631 !hugetlb_pmd_shared(ptep)) 1632 flags |= PM_MMAP_EXCLUSIVE; 1633 1634 if (huge_pte_uffd_wp(pte)) 1635 flags |= PM_UFFD_WP; 1636 1637 flags |= PM_PRESENT; 1638 if (pm->show_pfn) 1639 frame = pte_pfn(pte) + 1640 ((addr & ~hmask) >> PAGE_SHIFT); 1641 } else if (pte_swp_uffd_wp_any(pte)) { 1642 flags |= PM_UFFD_WP; 1643 } 1644 1645 for (; addr != end; addr += PAGE_SIZE) { 1646 pagemap_entry_t pme = make_pme(frame, flags); 1647 1648 err = add_to_pagemap(&pme, pm); 1649 if (err) 1650 return err; 1651 if (pm->show_pfn && (flags & PM_PRESENT)) 1652 frame++; 1653 } 1654 1655 cond_resched(); 1656 1657 return err; 1658 } 1659 #else 1660 #define pagemap_hugetlb_range NULL 1661 #endif /* HUGETLB_PAGE */ 1662 1663 static const struct mm_walk_ops pagemap_ops = { 1664 .pmd_entry = pagemap_pmd_range, 1665 .pte_hole = pagemap_pte_hole, 1666 .hugetlb_entry = pagemap_hugetlb_range, 1667 .walk_lock = PGWALK_RDLOCK, 1668 }; 1669 1670 /* 1671 * /proc/pid/pagemap - an array mapping virtual pages to pfns 1672 * 1673 * For each page in the address space, this file contains one 64-bit entry 1674 * consisting of the following: 1675 * 1676 * Bits 0-54 page frame number (PFN) if present 1677 * Bits 0-4 swap type if swapped 1678 * Bits 5-54 swap offset if swapped 1679 * Bit 55 pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst) 1680 * Bit 56 page exclusively mapped 1681 * Bit 57 pte is uffd-wp write-protected 1682 * Bits 58-60 zero 1683 * Bit 61 page is file-page or shared-anon 1684 * Bit 62 page swapped 1685 * Bit 63 page present 1686 * 1687 * If the page is not present but in swap, then the PFN contains an 1688 * encoding of the swap file number and the page's offset into the 1689 * swap. Unmapped pages return a null PFN. This allows determining 1690 * precisely which pages are mapped (or in swap) and comparing mapped 1691 * pages between processes. 1692 * 1693 * Efficient users of this interface will use /proc/pid/maps to 1694 * determine which areas of memory are actually mapped and llseek to 1695 * skip over unmapped regions. 1696 */ 1697 static ssize_t pagemap_read(struct file *file, char __user *buf, 1698 size_t count, loff_t *ppos) 1699 { 1700 struct mm_struct *mm = file->private_data; 1701 struct pagemapread pm; 1702 unsigned long src; 1703 unsigned long svpfn; 1704 unsigned long start_vaddr; 1705 unsigned long end_vaddr; 1706 int ret = 0, copied = 0; 1707 1708 if (!mm || !mmget_not_zero(mm)) 1709 goto out; 1710 1711 ret = -EINVAL; 1712 /* file position must be aligned */ 1713 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES)) 1714 goto out_mm; 1715 1716 ret = 0; 1717 if (!count) 1718 goto out_mm; 1719 1720 /* do not disclose physical addresses: attack vector */ 1721 pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN); 1722 1723 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT); 1724 pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL); 1725 ret = -ENOMEM; 1726 if (!pm.buffer) 1727 goto out_mm; 1728 1729 src = *ppos; 1730 svpfn = src / PM_ENTRY_BYTES; 1731 end_vaddr = mm->task_size; 1732 1733 /* watch out for wraparound */ 1734 start_vaddr = end_vaddr; 1735 if (svpfn <= (ULONG_MAX >> PAGE_SHIFT)) { 1736 unsigned long end; 1737 1738 ret = mmap_read_lock_killable(mm); 1739 if (ret) 1740 goto out_free; 1741 start_vaddr = untagged_addr_remote(mm, svpfn << PAGE_SHIFT); 1742 mmap_read_unlock(mm); 1743 1744 end = start_vaddr + ((count / PM_ENTRY_BYTES) << PAGE_SHIFT); 1745 if (end >= start_vaddr && end < mm->task_size) 1746 end_vaddr = end; 1747 } 1748 1749 /* Ensure the address is inside the task */ 1750 if (start_vaddr > mm->task_size) 1751 start_vaddr = end_vaddr; 1752 1753 ret = 0; 1754 while (count && (start_vaddr < end_vaddr)) { 1755 int len; 1756 unsigned long end; 1757 1758 pm.pos = 0; 1759 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK; 1760 /* overflow ? */ 1761 if (end < start_vaddr || end > end_vaddr) 1762 end = end_vaddr; 1763 ret = mmap_read_lock_killable(mm); 1764 if (ret) 1765 goto out_free; 1766 ret = walk_page_range(mm, start_vaddr, end, &pagemap_ops, &pm); 1767 mmap_read_unlock(mm); 1768 start_vaddr = end; 1769 1770 len = min(count, PM_ENTRY_BYTES * pm.pos); 1771 if (copy_to_user(buf, pm.buffer, len)) { 1772 ret = -EFAULT; 1773 goto out_free; 1774 } 1775 copied += len; 1776 buf += len; 1777 count -= len; 1778 } 1779 *ppos += copied; 1780 if (!ret || ret == PM_END_OF_BUFFER) 1781 ret = copied; 1782 1783 out_free: 1784 kfree(pm.buffer); 1785 out_mm: 1786 mmput(mm); 1787 out: 1788 return ret; 1789 } 1790 1791 static int pagemap_open(struct inode *inode, struct file *file) 1792 { 1793 struct mm_struct *mm; 1794 1795 mm = proc_mem_open(inode, PTRACE_MODE_READ); 1796 if (IS_ERR(mm)) 1797 return PTR_ERR(mm); 1798 file->private_data = mm; 1799 return 0; 1800 } 1801 1802 static int pagemap_release(struct inode *inode, struct file *file) 1803 { 1804 struct mm_struct *mm = file->private_data; 1805 1806 if (mm) 1807 mmdrop(mm); 1808 return 0; 1809 } 1810 1811 #define PM_SCAN_CATEGORIES (PAGE_IS_WPALLOWED | PAGE_IS_WRITTEN | \ 1812 PAGE_IS_FILE | PAGE_IS_PRESENT | \ 1813 PAGE_IS_SWAPPED | PAGE_IS_PFNZERO | \ 1814 PAGE_IS_HUGE | PAGE_IS_SOFT_DIRTY) 1815 #define PM_SCAN_FLAGS (PM_SCAN_WP_MATCHING | PM_SCAN_CHECK_WPASYNC) 1816 1817 struct pagemap_scan_private { 1818 struct pm_scan_arg arg; 1819 unsigned long masks_of_interest, cur_vma_category; 1820 struct page_region *vec_buf; 1821 unsigned long vec_buf_len, vec_buf_index, found_pages; 1822 struct page_region __user *vec_out; 1823 }; 1824 1825 static unsigned long pagemap_page_category(struct pagemap_scan_private *p, 1826 struct vm_area_struct *vma, 1827 unsigned long addr, pte_t pte) 1828 { 1829 unsigned long categories = 0; 1830 1831 if (pte_present(pte)) { 1832 struct page *page; 1833 1834 categories |= PAGE_IS_PRESENT; 1835 if (!pte_uffd_wp(pte)) 1836 categories |= PAGE_IS_WRITTEN; 1837 1838 if (p->masks_of_interest & PAGE_IS_FILE) { 1839 page = vm_normal_page(vma, addr, pte); 1840 if (page && !PageAnon(page)) 1841 categories |= PAGE_IS_FILE; 1842 } 1843 1844 if (is_zero_pfn(pte_pfn(pte))) 1845 categories |= PAGE_IS_PFNZERO; 1846 if (pte_soft_dirty(pte)) 1847 categories |= PAGE_IS_SOFT_DIRTY; 1848 } else if (is_swap_pte(pte)) { 1849 swp_entry_t swp; 1850 1851 categories |= PAGE_IS_SWAPPED; 1852 if (!pte_swp_uffd_wp_any(pte)) 1853 categories |= PAGE_IS_WRITTEN; 1854 1855 if (p->masks_of_interest & PAGE_IS_FILE) { 1856 swp = pte_to_swp_entry(pte); 1857 if (is_pfn_swap_entry(swp) && 1858 !folio_test_anon(pfn_swap_entry_folio(swp))) 1859 categories |= PAGE_IS_FILE; 1860 } 1861 if (pte_swp_soft_dirty(pte)) 1862 categories |= PAGE_IS_SOFT_DIRTY; 1863 } 1864 1865 return categories; 1866 } 1867 1868 static void make_uffd_wp_pte(struct vm_area_struct *vma, 1869 unsigned long addr, pte_t *pte, pte_t ptent) 1870 { 1871 if (pte_present(ptent)) { 1872 pte_t old_pte; 1873 1874 old_pte = ptep_modify_prot_start(vma, addr, pte); 1875 ptent = pte_mkuffd_wp(old_pte); 1876 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent); 1877 } else if (is_swap_pte(ptent)) { 1878 ptent = pte_swp_mkuffd_wp(ptent); 1879 set_pte_at(vma->vm_mm, addr, pte, ptent); 1880 } else { 1881 set_pte_at(vma->vm_mm, addr, pte, 1882 make_pte_marker(PTE_MARKER_UFFD_WP)); 1883 } 1884 } 1885 1886 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1887 static unsigned long pagemap_thp_category(struct pagemap_scan_private *p, 1888 struct vm_area_struct *vma, 1889 unsigned long addr, pmd_t pmd) 1890 { 1891 unsigned long categories = PAGE_IS_HUGE; 1892 1893 if (pmd_present(pmd)) { 1894 struct page *page; 1895 1896 categories |= PAGE_IS_PRESENT; 1897 if (!pmd_uffd_wp(pmd)) 1898 categories |= PAGE_IS_WRITTEN; 1899 1900 if (p->masks_of_interest & PAGE_IS_FILE) { 1901 page = vm_normal_page_pmd(vma, addr, pmd); 1902 if (page && !PageAnon(page)) 1903 categories |= PAGE_IS_FILE; 1904 } 1905 1906 if (is_zero_pfn(pmd_pfn(pmd))) 1907 categories |= PAGE_IS_PFNZERO; 1908 if (pmd_soft_dirty(pmd)) 1909 categories |= PAGE_IS_SOFT_DIRTY; 1910 } else if (is_swap_pmd(pmd)) { 1911 swp_entry_t swp; 1912 1913 categories |= PAGE_IS_SWAPPED; 1914 if (!pmd_swp_uffd_wp(pmd)) 1915 categories |= PAGE_IS_WRITTEN; 1916 if (pmd_swp_soft_dirty(pmd)) 1917 categories |= PAGE_IS_SOFT_DIRTY; 1918 1919 if (p->masks_of_interest & PAGE_IS_FILE) { 1920 swp = pmd_to_swp_entry(pmd); 1921 if (is_pfn_swap_entry(swp) && 1922 !folio_test_anon(pfn_swap_entry_folio(swp))) 1923 categories |= PAGE_IS_FILE; 1924 } 1925 } 1926 1927 return categories; 1928 } 1929 1930 static void make_uffd_wp_pmd(struct vm_area_struct *vma, 1931 unsigned long addr, pmd_t *pmdp) 1932 { 1933 pmd_t old, pmd = *pmdp; 1934 1935 if (pmd_present(pmd)) { 1936 old = pmdp_invalidate_ad(vma, addr, pmdp); 1937 pmd = pmd_mkuffd_wp(old); 1938 set_pmd_at(vma->vm_mm, addr, pmdp, pmd); 1939 } else if (is_migration_entry(pmd_to_swp_entry(pmd))) { 1940 pmd = pmd_swp_mkuffd_wp(pmd); 1941 set_pmd_at(vma->vm_mm, addr, pmdp, pmd); 1942 } 1943 } 1944 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ 1945 1946 #ifdef CONFIG_HUGETLB_PAGE 1947 static unsigned long pagemap_hugetlb_category(pte_t pte) 1948 { 1949 unsigned long categories = PAGE_IS_HUGE; 1950 1951 /* 1952 * According to pagemap_hugetlb_range(), file-backed HugeTLB 1953 * page cannot be swapped. So PAGE_IS_FILE is not checked for 1954 * swapped pages. 1955 */ 1956 if (pte_present(pte)) { 1957 categories |= PAGE_IS_PRESENT; 1958 if (!huge_pte_uffd_wp(pte)) 1959 categories |= PAGE_IS_WRITTEN; 1960 if (!PageAnon(pte_page(pte))) 1961 categories |= PAGE_IS_FILE; 1962 if (is_zero_pfn(pte_pfn(pte))) 1963 categories |= PAGE_IS_PFNZERO; 1964 if (pte_soft_dirty(pte)) 1965 categories |= PAGE_IS_SOFT_DIRTY; 1966 } else if (is_swap_pte(pte)) { 1967 categories |= PAGE_IS_SWAPPED; 1968 if (!pte_swp_uffd_wp_any(pte)) 1969 categories |= PAGE_IS_WRITTEN; 1970 if (pte_swp_soft_dirty(pte)) 1971 categories |= PAGE_IS_SOFT_DIRTY; 1972 } 1973 1974 return categories; 1975 } 1976 1977 static void make_uffd_wp_huge_pte(struct vm_area_struct *vma, 1978 unsigned long addr, pte_t *ptep, 1979 pte_t ptent) 1980 { 1981 unsigned long psize; 1982 1983 if (is_hugetlb_entry_hwpoisoned(ptent) || is_pte_marker(ptent)) 1984 return; 1985 1986 psize = huge_page_size(hstate_vma(vma)); 1987 1988 if (is_hugetlb_entry_migration(ptent)) 1989 set_huge_pte_at(vma->vm_mm, addr, ptep, 1990 pte_swp_mkuffd_wp(ptent), psize); 1991 else if (!huge_pte_none(ptent)) 1992 huge_ptep_modify_prot_commit(vma, addr, ptep, ptent, 1993 huge_pte_mkuffd_wp(ptent)); 1994 else 1995 set_huge_pte_at(vma->vm_mm, addr, ptep, 1996 make_pte_marker(PTE_MARKER_UFFD_WP), psize); 1997 } 1998 #endif /* CONFIG_HUGETLB_PAGE */ 1999 2000 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLB_PAGE) 2001 static void pagemap_scan_backout_range(struct pagemap_scan_private *p, 2002 unsigned long addr, unsigned long end) 2003 { 2004 struct page_region *cur_buf = &p->vec_buf[p->vec_buf_index]; 2005 2006 if (cur_buf->start != addr) 2007 cur_buf->end = addr; 2008 else 2009 cur_buf->start = cur_buf->end = 0; 2010 2011 p->found_pages -= (end - addr) / PAGE_SIZE; 2012 } 2013 #endif 2014 2015 static bool pagemap_scan_is_interesting_page(unsigned long categories, 2016 const struct pagemap_scan_private *p) 2017 { 2018 categories ^= p->arg.category_inverted; 2019 if ((categories & p->arg.category_mask) != p->arg.category_mask) 2020 return false; 2021 if (p->arg.category_anyof_mask && !(categories & p->arg.category_anyof_mask)) 2022 return false; 2023 2024 return true; 2025 } 2026 2027 static bool pagemap_scan_is_interesting_vma(unsigned long categories, 2028 const struct pagemap_scan_private *p) 2029 { 2030 unsigned long required = p->arg.category_mask & PAGE_IS_WPALLOWED; 2031 2032 categories ^= p->arg.category_inverted; 2033 if ((categories & required) != required) 2034 return false; 2035 2036 return true; 2037 } 2038 2039 static int pagemap_scan_test_walk(unsigned long start, unsigned long end, 2040 struct mm_walk *walk) 2041 { 2042 struct pagemap_scan_private *p = walk->private; 2043 struct vm_area_struct *vma = walk->vma; 2044 unsigned long vma_category = 0; 2045 bool wp_allowed = userfaultfd_wp_async(vma) && 2046 userfaultfd_wp_use_markers(vma); 2047 2048 if (!wp_allowed) { 2049 /* User requested explicit failure over wp-async capability */ 2050 if (p->arg.flags & PM_SCAN_CHECK_WPASYNC) 2051 return -EPERM; 2052 /* 2053 * User requires wr-protect, and allows silently skipping 2054 * unsupported vmas. 2055 */ 2056 if (p->arg.flags & PM_SCAN_WP_MATCHING) 2057 return 1; 2058 /* 2059 * Then the request doesn't involve wr-protects at all, 2060 * fall through to the rest checks, and allow vma walk. 2061 */ 2062 } 2063 2064 if (vma->vm_flags & VM_PFNMAP) 2065 return 1; 2066 2067 if (wp_allowed) 2068 vma_category |= PAGE_IS_WPALLOWED; 2069 2070 if (vma->vm_flags & VM_SOFTDIRTY) 2071 vma_category |= PAGE_IS_SOFT_DIRTY; 2072 2073 if (!pagemap_scan_is_interesting_vma(vma_category, p)) 2074 return 1; 2075 2076 p->cur_vma_category = vma_category; 2077 2078 return 0; 2079 } 2080 2081 static bool pagemap_scan_push_range(unsigned long categories, 2082 struct pagemap_scan_private *p, 2083 unsigned long addr, unsigned long end) 2084 { 2085 struct page_region *cur_buf = &p->vec_buf[p->vec_buf_index]; 2086 2087 /* 2088 * When there is no output buffer provided at all, the sentinel values 2089 * won't match here. There is no other way for `cur_buf->end` to be 2090 * non-zero other than it being non-empty. 2091 */ 2092 if (addr == cur_buf->end && categories == cur_buf->categories) { 2093 cur_buf->end = end; 2094 return true; 2095 } 2096 2097 if (cur_buf->end) { 2098 if (p->vec_buf_index >= p->vec_buf_len - 1) 2099 return false; 2100 2101 cur_buf = &p->vec_buf[++p->vec_buf_index]; 2102 } 2103 2104 cur_buf->start = addr; 2105 cur_buf->end = end; 2106 cur_buf->categories = categories; 2107 2108 return true; 2109 } 2110 2111 static int pagemap_scan_output(unsigned long categories, 2112 struct pagemap_scan_private *p, 2113 unsigned long addr, unsigned long *end) 2114 { 2115 unsigned long n_pages, total_pages; 2116 int ret = 0; 2117 2118 if (!p->vec_buf) 2119 return 0; 2120 2121 categories &= p->arg.return_mask; 2122 2123 n_pages = (*end - addr) / PAGE_SIZE; 2124 if (check_add_overflow(p->found_pages, n_pages, &total_pages) || 2125 total_pages > p->arg.max_pages) { 2126 size_t n_too_much = total_pages - p->arg.max_pages; 2127 *end -= n_too_much * PAGE_SIZE; 2128 n_pages -= n_too_much; 2129 ret = -ENOSPC; 2130 } 2131 2132 if (!pagemap_scan_push_range(categories, p, addr, *end)) { 2133 *end = addr; 2134 n_pages = 0; 2135 ret = -ENOSPC; 2136 } 2137 2138 p->found_pages += n_pages; 2139 if (ret) 2140 p->arg.walk_end = *end; 2141 2142 return ret; 2143 } 2144 2145 static int pagemap_scan_thp_entry(pmd_t *pmd, unsigned long start, 2146 unsigned long end, struct mm_walk *walk) 2147 { 2148 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 2149 struct pagemap_scan_private *p = walk->private; 2150 struct vm_area_struct *vma = walk->vma; 2151 unsigned long categories; 2152 spinlock_t *ptl; 2153 int ret = 0; 2154 2155 ptl = pmd_trans_huge_lock(pmd, vma); 2156 if (!ptl) 2157 return -ENOENT; 2158 2159 categories = p->cur_vma_category | 2160 pagemap_thp_category(p, vma, start, *pmd); 2161 2162 if (!pagemap_scan_is_interesting_page(categories, p)) 2163 goto out_unlock; 2164 2165 ret = pagemap_scan_output(categories, p, start, &end); 2166 if (start == end) 2167 goto out_unlock; 2168 2169 if (~p->arg.flags & PM_SCAN_WP_MATCHING) 2170 goto out_unlock; 2171 if (~categories & PAGE_IS_WRITTEN) 2172 goto out_unlock; 2173 2174 /* 2175 * Break huge page into small pages if the WP operation 2176 * needs to be performed on a portion of the huge page. 2177 */ 2178 if (end != start + HPAGE_SIZE) { 2179 spin_unlock(ptl); 2180 split_huge_pmd(vma, pmd, start); 2181 pagemap_scan_backout_range(p, start, end); 2182 /* Report as if there was no THP */ 2183 return -ENOENT; 2184 } 2185 2186 make_uffd_wp_pmd(vma, start, pmd); 2187 flush_tlb_range(vma, start, end); 2188 out_unlock: 2189 spin_unlock(ptl); 2190 return ret; 2191 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */ 2192 return -ENOENT; 2193 #endif 2194 } 2195 2196 static int pagemap_scan_pmd_entry(pmd_t *pmd, unsigned long start, 2197 unsigned long end, struct mm_walk *walk) 2198 { 2199 struct pagemap_scan_private *p = walk->private; 2200 struct vm_area_struct *vma = walk->vma; 2201 unsigned long addr, flush_end = 0; 2202 pte_t *pte, *start_pte; 2203 spinlock_t *ptl; 2204 int ret; 2205 2206 arch_enter_lazy_mmu_mode(); 2207 2208 ret = pagemap_scan_thp_entry(pmd, start, end, walk); 2209 if (ret != -ENOENT) { 2210 arch_leave_lazy_mmu_mode(); 2211 return ret; 2212 } 2213 2214 ret = 0; 2215 start_pte = pte = pte_offset_map_lock(vma->vm_mm, pmd, start, &ptl); 2216 if (!pte) { 2217 arch_leave_lazy_mmu_mode(); 2218 walk->action = ACTION_AGAIN; 2219 return 0; 2220 } 2221 2222 if ((p->arg.flags & PM_SCAN_WP_MATCHING) && !p->vec_out) { 2223 /* Fast path for performing exclusive WP */ 2224 for (addr = start; addr != end; pte++, addr += PAGE_SIZE) { 2225 pte_t ptent = ptep_get(pte); 2226 2227 if ((pte_present(ptent) && pte_uffd_wp(ptent)) || 2228 pte_swp_uffd_wp_any(ptent)) 2229 continue; 2230 make_uffd_wp_pte(vma, addr, pte, ptent); 2231 if (!flush_end) 2232 start = addr; 2233 flush_end = addr + PAGE_SIZE; 2234 } 2235 goto flush_and_return; 2236 } 2237 2238 if (!p->arg.category_anyof_mask && !p->arg.category_inverted && 2239 p->arg.category_mask == PAGE_IS_WRITTEN && 2240 p->arg.return_mask == PAGE_IS_WRITTEN) { 2241 for (addr = start; addr < end; pte++, addr += PAGE_SIZE) { 2242 unsigned long next = addr + PAGE_SIZE; 2243 pte_t ptent = ptep_get(pte); 2244 2245 if ((pte_present(ptent) && pte_uffd_wp(ptent)) || 2246 pte_swp_uffd_wp_any(ptent)) 2247 continue; 2248 ret = pagemap_scan_output(p->cur_vma_category | PAGE_IS_WRITTEN, 2249 p, addr, &next); 2250 if (next == addr) 2251 break; 2252 if (~p->arg.flags & PM_SCAN_WP_MATCHING) 2253 continue; 2254 make_uffd_wp_pte(vma, addr, pte, ptent); 2255 if (!flush_end) 2256 start = addr; 2257 flush_end = next; 2258 } 2259 goto flush_and_return; 2260 } 2261 2262 for (addr = start; addr != end; pte++, addr += PAGE_SIZE) { 2263 pte_t ptent = ptep_get(pte); 2264 unsigned long categories = p->cur_vma_category | 2265 pagemap_page_category(p, vma, addr, ptent); 2266 unsigned long next = addr + PAGE_SIZE; 2267 2268 if (!pagemap_scan_is_interesting_page(categories, p)) 2269 continue; 2270 2271 ret = pagemap_scan_output(categories, p, addr, &next); 2272 if (next == addr) 2273 break; 2274 2275 if (~p->arg.flags & PM_SCAN_WP_MATCHING) 2276 continue; 2277 if (~categories & PAGE_IS_WRITTEN) 2278 continue; 2279 2280 make_uffd_wp_pte(vma, addr, pte, ptent); 2281 if (!flush_end) 2282 start = addr; 2283 flush_end = next; 2284 } 2285 2286 flush_and_return: 2287 if (flush_end) 2288 flush_tlb_range(vma, start, addr); 2289 2290 pte_unmap_unlock(start_pte, ptl); 2291 arch_leave_lazy_mmu_mode(); 2292 2293 cond_resched(); 2294 return ret; 2295 } 2296 2297 #ifdef CONFIG_HUGETLB_PAGE 2298 static int pagemap_scan_hugetlb_entry(pte_t *ptep, unsigned long hmask, 2299 unsigned long start, unsigned long end, 2300 struct mm_walk *walk) 2301 { 2302 struct pagemap_scan_private *p = walk->private; 2303 struct vm_area_struct *vma = walk->vma; 2304 unsigned long categories; 2305 spinlock_t *ptl; 2306 int ret = 0; 2307 pte_t pte; 2308 2309 if (~p->arg.flags & PM_SCAN_WP_MATCHING) { 2310 /* Go the short route when not write-protecting pages. */ 2311 2312 pte = huge_ptep_get(ptep); 2313 categories = p->cur_vma_category | pagemap_hugetlb_category(pte); 2314 2315 if (!pagemap_scan_is_interesting_page(categories, p)) 2316 return 0; 2317 2318 return pagemap_scan_output(categories, p, start, &end); 2319 } 2320 2321 i_mmap_lock_write(vma->vm_file->f_mapping); 2322 ptl = huge_pte_lock(hstate_vma(vma), vma->vm_mm, ptep); 2323 2324 pte = huge_ptep_get(ptep); 2325 categories = p->cur_vma_category | pagemap_hugetlb_category(pte); 2326 2327 if (!pagemap_scan_is_interesting_page(categories, p)) 2328 goto out_unlock; 2329 2330 ret = pagemap_scan_output(categories, p, start, &end); 2331 if (start == end) 2332 goto out_unlock; 2333 2334 if (~categories & PAGE_IS_WRITTEN) 2335 goto out_unlock; 2336 2337 if (end != start + HPAGE_SIZE) { 2338 /* Partial HugeTLB page WP isn't possible. */ 2339 pagemap_scan_backout_range(p, start, end); 2340 p->arg.walk_end = start; 2341 ret = 0; 2342 goto out_unlock; 2343 } 2344 2345 make_uffd_wp_huge_pte(vma, start, ptep, pte); 2346 flush_hugetlb_tlb_range(vma, start, end); 2347 2348 out_unlock: 2349 spin_unlock(ptl); 2350 i_mmap_unlock_write(vma->vm_file->f_mapping); 2351 2352 return ret; 2353 } 2354 #else 2355 #define pagemap_scan_hugetlb_entry NULL 2356 #endif 2357 2358 static int pagemap_scan_pte_hole(unsigned long addr, unsigned long end, 2359 int depth, struct mm_walk *walk) 2360 { 2361 struct pagemap_scan_private *p = walk->private; 2362 struct vm_area_struct *vma = walk->vma; 2363 int ret, err; 2364 2365 if (!vma || !pagemap_scan_is_interesting_page(p->cur_vma_category, p)) 2366 return 0; 2367 2368 ret = pagemap_scan_output(p->cur_vma_category, p, addr, &end); 2369 if (addr == end) 2370 return ret; 2371 2372 if (~p->arg.flags & PM_SCAN_WP_MATCHING) 2373 return ret; 2374 2375 err = uffd_wp_range(vma, addr, end - addr, true); 2376 if (err < 0) 2377 ret = err; 2378 2379 return ret; 2380 } 2381 2382 static const struct mm_walk_ops pagemap_scan_ops = { 2383 .test_walk = pagemap_scan_test_walk, 2384 .pmd_entry = pagemap_scan_pmd_entry, 2385 .pte_hole = pagemap_scan_pte_hole, 2386 .hugetlb_entry = pagemap_scan_hugetlb_entry, 2387 }; 2388 2389 static int pagemap_scan_get_args(struct pm_scan_arg *arg, 2390 unsigned long uarg) 2391 { 2392 if (copy_from_user(arg, (void __user *)uarg, sizeof(*arg))) 2393 return -EFAULT; 2394 2395 if (arg->size != sizeof(struct pm_scan_arg)) 2396 return -EINVAL; 2397 2398 /* Validate requested features */ 2399 if (arg->flags & ~PM_SCAN_FLAGS) 2400 return -EINVAL; 2401 if ((arg->category_inverted | arg->category_mask | 2402 arg->category_anyof_mask | arg->return_mask) & ~PM_SCAN_CATEGORIES) 2403 return -EINVAL; 2404 2405 arg->start = untagged_addr((unsigned long)arg->start); 2406 arg->end = untagged_addr((unsigned long)arg->end); 2407 arg->vec = untagged_addr((unsigned long)arg->vec); 2408 2409 /* Validate memory pointers */ 2410 if (!IS_ALIGNED(arg->start, PAGE_SIZE)) 2411 return -EINVAL; 2412 if (!access_ok((void __user *)(long)arg->start, arg->end - arg->start)) 2413 return -EFAULT; 2414 if (!arg->vec && arg->vec_len) 2415 return -EINVAL; 2416 if (arg->vec && !access_ok((void __user *)(long)arg->vec, 2417 arg->vec_len * sizeof(struct page_region))) 2418 return -EFAULT; 2419 2420 /* Fixup default values */ 2421 arg->end = ALIGN(arg->end, PAGE_SIZE); 2422 arg->walk_end = 0; 2423 if (!arg->max_pages) 2424 arg->max_pages = ULONG_MAX; 2425 2426 return 0; 2427 } 2428 2429 static int pagemap_scan_writeback_args(struct pm_scan_arg *arg, 2430 unsigned long uargl) 2431 { 2432 struct pm_scan_arg __user *uarg = (void __user *)uargl; 2433 2434 if (copy_to_user(&uarg->walk_end, &arg->walk_end, sizeof(arg->walk_end))) 2435 return -EFAULT; 2436 2437 return 0; 2438 } 2439 2440 static int pagemap_scan_init_bounce_buffer(struct pagemap_scan_private *p) 2441 { 2442 if (!p->arg.vec_len) 2443 return 0; 2444 2445 p->vec_buf_len = min_t(size_t, PAGEMAP_WALK_SIZE >> PAGE_SHIFT, 2446 p->arg.vec_len); 2447 p->vec_buf = kmalloc_array(p->vec_buf_len, sizeof(*p->vec_buf), 2448 GFP_KERNEL); 2449 if (!p->vec_buf) 2450 return -ENOMEM; 2451 2452 p->vec_buf->start = p->vec_buf->end = 0; 2453 p->vec_out = (struct page_region __user *)(long)p->arg.vec; 2454 2455 return 0; 2456 } 2457 2458 static long pagemap_scan_flush_buffer(struct pagemap_scan_private *p) 2459 { 2460 const struct page_region *buf = p->vec_buf; 2461 long n = p->vec_buf_index; 2462 2463 if (!p->vec_buf) 2464 return 0; 2465 2466 if (buf[n].end != buf[n].start) 2467 n++; 2468 2469 if (!n) 2470 return 0; 2471 2472 if (copy_to_user(p->vec_out, buf, n * sizeof(*buf))) 2473 return -EFAULT; 2474 2475 p->arg.vec_len -= n; 2476 p->vec_out += n; 2477 2478 p->vec_buf_index = 0; 2479 p->vec_buf_len = min_t(size_t, p->vec_buf_len, p->arg.vec_len); 2480 p->vec_buf->start = p->vec_buf->end = 0; 2481 2482 return n; 2483 } 2484 2485 static long do_pagemap_scan(struct mm_struct *mm, unsigned long uarg) 2486 { 2487 struct pagemap_scan_private p = {0}; 2488 unsigned long walk_start; 2489 size_t n_ranges_out = 0; 2490 int ret; 2491 2492 ret = pagemap_scan_get_args(&p.arg, uarg); 2493 if (ret) 2494 return ret; 2495 2496 p.masks_of_interest = p.arg.category_mask | p.arg.category_anyof_mask | 2497 p.arg.return_mask; 2498 ret = pagemap_scan_init_bounce_buffer(&p); 2499 if (ret) 2500 return ret; 2501 2502 for (walk_start = p.arg.start; walk_start < p.arg.end; 2503 walk_start = p.arg.walk_end) { 2504 struct mmu_notifier_range range; 2505 long n_out; 2506 2507 if (fatal_signal_pending(current)) { 2508 ret = -EINTR; 2509 break; 2510 } 2511 2512 ret = mmap_read_lock_killable(mm); 2513 if (ret) 2514 break; 2515 2516 /* Protection change for the range is going to happen. */ 2517 if (p.arg.flags & PM_SCAN_WP_MATCHING) { 2518 mmu_notifier_range_init(&range, MMU_NOTIFY_PROTECTION_VMA, 0, 2519 mm, walk_start, p.arg.end); 2520 mmu_notifier_invalidate_range_start(&range); 2521 } 2522 2523 ret = walk_page_range(mm, walk_start, p.arg.end, 2524 &pagemap_scan_ops, &p); 2525 2526 if (p.arg.flags & PM_SCAN_WP_MATCHING) 2527 mmu_notifier_invalidate_range_end(&range); 2528 2529 mmap_read_unlock(mm); 2530 2531 n_out = pagemap_scan_flush_buffer(&p); 2532 if (n_out < 0) 2533 ret = n_out; 2534 else 2535 n_ranges_out += n_out; 2536 2537 if (ret != -ENOSPC) 2538 break; 2539 2540 if (p.arg.vec_len == 0 || p.found_pages == p.arg.max_pages) 2541 break; 2542 } 2543 2544 /* ENOSPC signifies early stop (buffer full) from the walk. */ 2545 if (!ret || ret == -ENOSPC) 2546 ret = n_ranges_out; 2547 2548 /* The walk_end isn't set when ret is zero */ 2549 if (!p.arg.walk_end) 2550 p.arg.walk_end = p.arg.end; 2551 if (pagemap_scan_writeback_args(&p.arg, uarg)) 2552 ret = -EFAULT; 2553 2554 kfree(p.vec_buf); 2555 return ret; 2556 } 2557 2558 static long do_pagemap_cmd(struct file *file, unsigned int cmd, 2559 unsigned long arg) 2560 { 2561 struct mm_struct *mm = file->private_data; 2562 2563 switch (cmd) { 2564 case PAGEMAP_SCAN: 2565 return do_pagemap_scan(mm, arg); 2566 2567 default: 2568 return -EINVAL; 2569 } 2570 } 2571 2572 const struct file_operations proc_pagemap_operations = { 2573 .llseek = mem_lseek, /* borrow this */ 2574 .read = pagemap_read, 2575 .open = pagemap_open, 2576 .release = pagemap_release, 2577 .unlocked_ioctl = do_pagemap_cmd, 2578 .compat_ioctl = do_pagemap_cmd, 2579 }; 2580 #endif /* CONFIG_PROC_PAGE_MONITOR */ 2581 2582 #ifdef CONFIG_NUMA 2583 2584 struct numa_maps { 2585 unsigned long pages; 2586 unsigned long anon; 2587 unsigned long active; 2588 unsigned long writeback; 2589 unsigned long mapcount_max; 2590 unsigned long dirty; 2591 unsigned long swapcache; 2592 unsigned long node[MAX_NUMNODES]; 2593 }; 2594 2595 struct numa_maps_private { 2596 struct proc_maps_private proc_maps; 2597 struct numa_maps md; 2598 }; 2599 2600 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty, 2601 unsigned long nr_pages) 2602 { 2603 struct folio *folio = page_folio(page); 2604 int count = folio_precise_page_mapcount(folio, page); 2605 2606 md->pages += nr_pages; 2607 if (pte_dirty || folio_test_dirty(folio)) 2608 md->dirty += nr_pages; 2609 2610 if (folio_test_swapcache(folio)) 2611 md->swapcache += nr_pages; 2612 2613 if (folio_test_active(folio) || folio_test_unevictable(folio)) 2614 md->active += nr_pages; 2615 2616 if (folio_test_writeback(folio)) 2617 md->writeback += nr_pages; 2618 2619 if (folio_test_anon(folio)) 2620 md->anon += nr_pages; 2621 2622 if (count > md->mapcount_max) 2623 md->mapcount_max = count; 2624 2625 md->node[folio_nid(folio)] += nr_pages; 2626 } 2627 2628 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma, 2629 unsigned long addr) 2630 { 2631 struct page *page; 2632 int nid; 2633 2634 if (!pte_present(pte)) 2635 return NULL; 2636 2637 page = vm_normal_page(vma, addr, pte); 2638 if (!page || is_zone_device_page(page)) 2639 return NULL; 2640 2641 if (PageReserved(page)) 2642 return NULL; 2643 2644 nid = page_to_nid(page); 2645 if (!node_isset(nid, node_states[N_MEMORY])) 2646 return NULL; 2647 2648 return page; 2649 } 2650 2651 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 2652 static struct page *can_gather_numa_stats_pmd(pmd_t pmd, 2653 struct vm_area_struct *vma, 2654 unsigned long addr) 2655 { 2656 struct page *page; 2657 int nid; 2658 2659 if (!pmd_present(pmd)) 2660 return NULL; 2661 2662 page = vm_normal_page_pmd(vma, addr, pmd); 2663 if (!page) 2664 return NULL; 2665 2666 if (PageReserved(page)) 2667 return NULL; 2668 2669 nid = page_to_nid(page); 2670 if (!node_isset(nid, node_states[N_MEMORY])) 2671 return NULL; 2672 2673 return page; 2674 } 2675 #endif 2676 2677 static int gather_pte_stats(pmd_t *pmd, unsigned long addr, 2678 unsigned long end, struct mm_walk *walk) 2679 { 2680 struct numa_maps *md = walk->private; 2681 struct vm_area_struct *vma = walk->vma; 2682 spinlock_t *ptl; 2683 pte_t *orig_pte; 2684 pte_t *pte; 2685 2686 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 2687 ptl = pmd_trans_huge_lock(pmd, vma); 2688 if (ptl) { 2689 struct page *page; 2690 2691 page = can_gather_numa_stats_pmd(*pmd, vma, addr); 2692 if (page) 2693 gather_stats(page, md, pmd_dirty(*pmd), 2694 HPAGE_PMD_SIZE/PAGE_SIZE); 2695 spin_unlock(ptl); 2696 return 0; 2697 } 2698 #endif 2699 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl); 2700 if (!pte) { 2701 walk->action = ACTION_AGAIN; 2702 return 0; 2703 } 2704 do { 2705 pte_t ptent = ptep_get(pte); 2706 struct page *page = can_gather_numa_stats(ptent, vma, addr); 2707 if (!page) 2708 continue; 2709 gather_stats(page, md, pte_dirty(ptent), 1); 2710 2711 } while (pte++, addr += PAGE_SIZE, addr != end); 2712 pte_unmap_unlock(orig_pte, ptl); 2713 cond_resched(); 2714 return 0; 2715 } 2716 #ifdef CONFIG_HUGETLB_PAGE 2717 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask, 2718 unsigned long addr, unsigned long end, struct mm_walk *walk) 2719 { 2720 pte_t huge_pte = huge_ptep_get(pte); 2721 struct numa_maps *md; 2722 struct page *page; 2723 2724 if (!pte_present(huge_pte)) 2725 return 0; 2726 2727 page = pte_page(huge_pte); 2728 2729 md = walk->private; 2730 gather_stats(page, md, pte_dirty(huge_pte), 1); 2731 return 0; 2732 } 2733 2734 #else 2735 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask, 2736 unsigned long addr, unsigned long end, struct mm_walk *walk) 2737 { 2738 return 0; 2739 } 2740 #endif 2741 2742 static const struct mm_walk_ops show_numa_ops = { 2743 .hugetlb_entry = gather_hugetlb_stats, 2744 .pmd_entry = gather_pte_stats, 2745 .walk_lock = PGWALK_RDLOCK, 2746 }; 2747 2748 /* 2749 * Display pages allocated per node and memory policy via /proc. 2750 */ 2751 static int show_numa_map(struct seq_file *m, void *v) 2752 { 2753 struct numa_maps_private *numa_priv = m->private; 2754 struct proc_maps_private *proc_priv = &numa_priv->proc_maps; 2755 struct vm_area_struct *vma = v; 2756 struct numa_maps *md = &numa_priv->md; 2757 struct file *file = vma->vm_file; 2758 struct mm_struct *mm = vma->vm_mm; 2759 char buffer[64]; 2760 struct mempolicy *pol; 2761 pgoff_t ilx; 2762 int nid; 2763 2764 if (!mm) 2765 return 0; 2766 2767 /* Ensure we start with an empty set of numa_maps statistics. */ 2768 memset(md, 0, sizeof(*md)); 2769 2770 pol = __get_vma_policy(vma, vma->vm_start, &ilx); 2771 if (pol) { 2772 mpol_to_str(buffer, sizeof(buffer), pol); 2773 mpol_cond_put(pol); 2774 } else { 2775 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy); 2776 } 2777 2778 seq_printf(m, "%08lx %s", vma->vm_start, buffer); 2779 2780 if (file) { 2781 seq_puts(m, " file="); 2782 seq_path(m, file_user_path(file), "\n\t= "); 2783 } else if (vma_is_initial_heap(vma)) { 2784 seq_puts(m, " heap"); 2785 } else if (vma_is_initial_stack(vma)) { 2786 seq_puts(m, " stack"); 2787 } 2788 2789 if (is_vm_hugetlb_page(vma)) 2790 seq_puts(m, " huge"); 2791 2792 /* mmap_lock is held by m_start */ 2793 walk_page_vma(vma, &show_numa_ops, md); 2794 2795 if (!md->pages) 2796 goto out; 2797 2798 if (md->anon) 2799 seq_printf(m, " anon=%lu", md->anon); 2800 2801 if (md->dirty) 2802 seq_printf(m, " dirty=%lu", md->dirty); 2803 2804 if (md->pages != md->anon && md->pages != md->dirty) 2805 seq_printf(m, " mapped=%lu", md->pages); 2806 2807 if (md->mapcount_max > 1) 2808 seq_printf(m, " mapmax=%lu", md->mapcount_max); 2809 2810 if (md->swapcache) 2811 seq_printf(m, " swapcache=%lu", md->swapcache); 2812 2813 if (md->active < md->pages && !is_vm_hugetlb_page(vma)) 2814 seq_printf(m, " active=%lu", md->active); 2815 2816 if (md->writeback) 2817 seq_printf(m, " writeback=%lu", md->writeback); 2818 2819 for_each_node_state(nid, N_MEMORY) 2820 if (md->node[nid]) 2821 seq_printf(m, " N%d=%lu", nid, md->node[nid]); 2822 2823 seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10); 2824 out: 2825 seq_putc(m, '\n'); 2826 return 0; 2827 } 2828 2829 static const struct seq_operations proc_pid_numa_maps_op = { 2830 .start = m_start, 2831 .next = m_next, 2832 .stop = m_stop, 2833 .show = show_numa_map, 2834 }; 2835 2836 static int pid_numa_maps_open(struct inode *inode, struct file *file) 2837 { 2838 return proc_maps_open(inode, file, &proc_pid_numa_maps_op, 2839 sizeof(struct numa_maps_private)); 2840 } 2841 2842 const struct file_operations proc_pid_numa_maps_operations = { 2843 .open = pid_numa_maps_open, 2844 .read = seq_read, 2845 .llseek = seq_lseek, 2846 .release = proc_map_release, 2847 }; 2848 2849 #endif /* CONFIG_NUMA */ 2850