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