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