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