1 // SPDX-License-Identifier: GPL-2.0 2 #include <linux/pagewalk.h> 3 #include <linux/vmacache.h> 4 #include <linux/mm_inline.h> 5 #include <linux/hugetlb.h> 6 #include <linux/huge_mm.h> 7 #include <linux/mount.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 24 #include <asm/elf.h> 25 #include <asm/tlb.h> 26 #include <asm/tlbflush.h> 27 #include "internal.h" 28 29 #define SEQ_PUT_DEC(str, val) \ 30 seq_put_decimal_ull_width(m, str, (val) << (PAGE_SHIFT-10), 8) 31 void task_mem(struct seq_file *m, struct mm_struct *mm) 32 { 33 unsigned long text, lib, swap, anon, file, shmem; 34 unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss; 35 36 anon = get_mm_counter(mm, MM_ANONPAGES); 37 file = get_mm_counter(mm, MM_FILEPAGES); 38 shmem = get_mm_counter(mm, MM_SHMEMPAGES); 39 40 /* 41 * Note: to minimize their overhead, mm maintains hiwater_vm and 42 * hiwater_rss only when about to *lower* total_vm or rss. Any 43 * collector of these hiwater stats must therefore get total_vm 44 * and rss too, which will usually be the higher. Barriers? not 45 * worth the effort, such snapshots can always be inconsistent. 46 */ 47 hiwater_vm = total_vm = mm->total_vm; 48 if (hiwater_vm < mm->hiwater_vm) 49 hiwater_vm = mm->hiwater_vm; 50 hiwater_rss = total_rss = anon + file + shmem; 51 if (hiwater_rss < mm->hiwater_rss) 52 hiwater_rss = mm->hiwater_rss; 53 54 /* split executable areas between text and lib */ 55 text = PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK); 56 text = min(text, mm->exec_vm << PAGE_SHIFT); 57 lib = (mm->exec_vm << PAGE_SHIFT) - text; 58 59 swap = get_mm_counter(mm, MM_SWAPENTS); 60 SEQ_PUT_DEC("VmPeak:\t", hiwater_vm); 61 SEQ_PUT_DEC(" kB\nVmSize:\t", total_vm); 62 SEQ_PUT_DEC(" kB\nVmLck:\t", mm->locked_vm); 63 SEQ_PUT_DEC(" kB\nVmPin:\t", atomic64_read(&mm->pinned_vm)); 64 SEQ_PUT_DEC(" kB\nVmHWM:\t", hiwater_rss); 65 SEQ_PUT_DEC(" kB\nVmRSS:\t", total_rss); 66 SEQ_PUT_DEC(" kB\nRssAnon:\t", anon); 67 SEQ_PUT_DEC(" kB\nRssFile:\t", file); 68 SEQ_PUT_DEC(" kB\nRssShmem:\t", shmem); 69 SEQ_PUT_DEC(" kB\nVmData:\t", mm->data_vm); 70 SEQ_PUT_DEC(" kB\nVmStk:\t", mm->stack_vm); 71 seq_put_decimal_ull_width(m, 72 " kB\nVmExe:\t", text >> 10, 8); 73 seq_put_decimal_ull_width(m, 74 " kB\nVmLib:\t", lib >> 10, 8); 75 seq_put_decimal_ull_width(m, 76 " kB\nVmPTE:\t", mm_pgtables_bytes(mm) >> 10, 8); 77 SEQ_PUT_DEC(" kB\nVmSwap:\t", swap); 78 seq_puts(m, " kB\n"); 79 hugetlb_report_usage(m, mm); 80 } 81 #undef SEQ_PUT_DEC 82 83 unsigned long task_vsize(struct mm_struct *mm) 84 { 85 return PAGE_SIZE * mm->total_vm; 86 } 87 88 unsigned long task_statm(struct mm_struct *mm, 89 unsigned long *shared, unsigned long *text, 90 unsigned long *data, unsigned long *resident) 91 { 92 *shared = get_mm_counter(mm, MM_FILEPAGES) + 93 get_mm_counter(mm, MM_SHMEMPAGES); 94 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) 95 >> PAGE_SHIFT; 96 *data = mm->data_vm + mm->stack_vm; 97 *resident = *shared + get_mm_counter(mm, MM_ANONPAGES); 98 return mm->total_vm; 99 } 100 101 #ifdef CONFIG_NUMA 102 /* 103 * Save get_task_policy() for show_numa_map(). 104 */ 105 static void hold_task_mempolicy(struct proc_maps_private *priv) 106 { 107 struct task_struct *task = priv->task; 108 109 task_lock(task); 110 priv->task_mempolicy = get_task_policy(task); 111 mpol_get(priv->task_mempolicy); 112 task_unlock(task); 113 } 114 static void release_task_mempolicy(struct proc_maps_private *priv) 115 { 116 mpol_put(priv->task_mempolicy); 117 } 118 #else 119 static void hold_task_mempolicy(struct proc_maps_private *priv) 120 { 121 } 122 static void release_task_mempolicy(struct proc_maps_private *priv) 123 { 124 } 125 #endif 126 127 static void *m_start(struct seq_file *m, loff_t *ppos) 128 { 129 struct proc_maps_private *priv = m->private; 130 unsigned long last_addr = *ppos; 131 struct mm_struct *mm; 132 struct vm_area_struct *vma; 133 134 /* See m_next(). Zero at the start or after lseek. */ 135 if (last_addr == -1UL) 136 return NULL; 137 138 priv->task = get_proc_task(priv->inode); 139 if (!priv->task) 140 return ERR_PTR(-ESRCH); 141 142 mm = priv->mm; 143 if (!mm || !mmget_not_zero(mm)) { 144 put_task_struct(priv->task); 145 priv->task = NULL; 146 return NULL; 147 } 148 149 if (mmap_read_lock_killable(mm)) { 150 mmput(mm); 151 put_task_struct(priv->task); 152 priv->task = NULL; 153 return ERR_PTR(-EINTR); 154 } 155 156 hold_task_mempolicy(priv); 157 priv->tail_vma = get_gate_vma(mm); 158 159 vma = find_vma(mm, last_addr); 160 if (vma) 161 return vma; 162 163 return priv->tail_vma; 164 } 165 166 static void *m_next(struct seq_file *m, void *v, loff_t *ppos) 167 { 168 struct proc_maps_private *priv = m->private; 169 struct vm_area_struct *next, *vma = v; 170 171 if (vma == priv->tail_vma) 172 next = NULL; 173 else if (vma->vm_next) 174 next = vma->vm_next; 175 else 176 next = priv->tail_vma; 177 178 *ppos = next ? next->vm_start : -1UL; 179 180 return next; 181 } 182 183 static void m_stop(struct seq_file *m, void *v) 184 { 185 struct proc_maps_private *priv = m->private; 186 struct mm_struct *mm = priv->mm; 187 188 if (!priv->task) 189 return; 190 191 release_task_mempolicy(priv); 192 mmap_read_unlock(mm); 193 mmput(mm); 194 put_task_struct(priv->task); 195 priv->task = NULL; 196 } 197 198 static int proc_maps_open(struct inode *inode, struct file *file, 199 const struct seq_operations *ops, int psize) 200 { 201 struct proc_maps_private *priv = __seq_open_private(file, ops, psize); 202 203 if (!priv) 204 return -ENOMEM; 205 206 priv->inode = inode; 207 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ); 208 if (IS_ERR(priv->mm)) { 209 int err = PTR_ERR(priv->mm); 210 211 seq_release_private(inode, file); 212 return err; 213 } 214 215 return 0; 216 } 217 218 static int proc_map_release(struct inode *inode, struct file *file) 219 { 220 struct seq_file *seq = file->private_data; 221 struct proc_maps_private *priv = seq->private; 222 223 if (priv->mm) 224 mmdrop(priv->mm); 225 226 return seq_release_private(inode, file); 227 } 228 229 static int do_maps_open(struct inode *inode, struct file *file, 230 const struct seq_operations *ops) 231 { 232 return proc_maps_open(inode, file, ops, 233 sizeof(struct proc_maps_private)); 234 } 235 236 /* 237 * Indicate if the VMA is a stack for the given task; for 238 * /proc/PID/maps that is the stack of the main task. 239 */ 240 static int is_stack(struct vm_area_struct *vma) 241 { 242 /* 243 * We make no effort to guess what a given thread considers to be 244 * its "stack". It's not even well-defined for programs written 245 * languages like Go. 246 */ 247 return vma->vm_start <= vma->vm_mm->start_stack && 248 vma->vm_end >= vma->vm_mm->start_stack; 249 } 250 251 static void show_vma_header_prefix(struct seq_file *m, 252 unsigned long start, unsigned long end, 253 vm_flags_t flags, unsigned long long pgoff, 254 dev_t dev, unsigned long ino) 255 { 256 seq_setwidth(m, 25 + sizeof(void *) * 6 - 1); 257 seq_put_hex_ll(m, NULL, start, 8); 258 seq_put_hex_ll(m, "-", end, 8); 259 seq_putc(m, ' '); 260 seq_putc(m, flags & VM_READ ? 'r' : '-'); 261 seq_putc(m, flags & VM_WRITE ? 'w' : '-'); 262 seq_putc(m, flags & VM_EXEC ? 'x' : '-'); 263 seq_putc(m, flags & VM_MAYSHARE ? 's' : 'p'); 264 seq_put_hex_ll(m, " ", pgoff, 8); 265 seq_put_hex_ll(m, " ", MAJOR(dev), 2); 266 seq_put_hex_ll(m, ":", MINOR(dev), 2); 267 seq_put_decimal_ull(m, " ", ino); 268 seq_putc(m, ' '); 269 } 270 271 static void 272 show_map_vma(struct seq_file *m, struct vm_area_struct *vma) 273 { 274 struct mm_struct *mm = vma->vm_mm; 275 struct file *file = vma->vm_file; 276 vm_flags_t flags = vma->vm_flags; 277 unsigned long ino = 0; 278 unsigned long long pgoff = 0; 279 unsigned long start, end; 280 dev_t dev = 0; 281 const char *name = NULL; 282 283 if (file) { 284 struct inode *inode = file_inode(vma->vm_file); 285 dev = inode->i_sb->s_dev; 286 ino = inode->i_ino; 287 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT; 288 } 289 290 start = vma->vm_start; 291 end = vma->vm_end; 292 show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino); 293 294 /* 295 * Print the dentry name for named mappings, and a 296 * special [heap] marker for the heap: 297 */ 298 if (file) { 299 seq_pad(m, ' '); 300 seq_file_path(m, file, "\n"); 301 goto done; 302 } 303 304 if (vma->vm_ops && vma->vm_ops->name) { 305 name = vma->vm_ops->name(vma); 306 if (name) 307 goto done; 308 } 309 310 name = arch_vma_name(vma); 311 if (!name) { 312 struct anon_vma_name *anon_name; 313 314 if (!mm) { 315 name = "[vdso]"; 316 goto done; 317 } 318 319 if (vma->vm_start <= mm->brk && 320 vma->vm_end >= mm->start_brk) { 321 name = "[heap]"; 322 goto done; 323 } 324 325 if (is_stack(vma)) { 326 name = "[stack]"; 327 goto done; 328 } 329 330 anon_name = anon_vma_name(vma); 331 if (anon_name) { 332 seq_pad(m, ' '); 333 seq_printf(m, "[anon:%s]", anon_name->name); 334 } 335 } 336 337 done: 338 if (name) { 339 seq_pad(m, ' '); 340 seq_puts(m, name); 341 } 342 seq_putc(m, '\n'); 343 } 344 345 static int show_map(struct seq_file *m, void *v) 346 { 347 show_map_vma(m, v); 348 return 0; 349 } 350 351 static const struct seq_operations proc_pid_maps_op = { 352 .start = m_start, 353 .next = m_next, 354 .stop = m_stop, 355 .show = show_map 356 }; 357 358 static int pid_maps_open(struct inode *inode, struct file *file) 359 { 360 return do_maps_open(inode, file, &proc_pid_maps_op); 361 } 362 363 const struct file_operations proc_pid_maps_operations = { 364 .open = pid_maps_open, 365 .read = seq_read, 366 .llseek = seq_lseek, 367 .release = proc_map_release, 368 }; 369 370 /* 371 * Proportional Set Size(PSS): my share of RSS. 372 * 373 * PSS of a process is the count of pages it has in memory, where each 374 * page is divided by the number of processes sharing it. So if a 375 * process has 1000 pages all to itself, and 1000 shared with one other 376 * process, its PSS will be 1500. 377 * 378 * To keep (accumulated) division errors low, we adopt a 64bit 379 * fixed-point pss counter to minimize division errors. So (pss >> 380 * PSS_SHIFT) would be the real byte count. 381 * 382 * A shift of 12 before division means (assuming 4K page size): 383 * - 1M 3-user-pages add up to 8KB errors; 384 * - supports mapcount up to 2^24, or 16M; 385 * - supports PSS up to 2^52 bytes, or 4PB. 386 */ 387 #define PSS_SHIFT 12 388 389 #ifdef CONFIG_PROC_PAGE_MONITOR 390 struct mem_size_stats { 391 unsigned long resident; 392 unsigned long shared_clean; 393 unsigned long shared_dirty; 394 unsigned long private_clean; 395 unsigned long private_dirty; 396 unsigned long referenced; 397 unsigned long anonymous; 398 unsigned long lazyfree; 399 unsigned long anonymous_thp; 400 unsigned long shmem_thp; 401 unsigned long file_thp; 402 unsigned long swap; 403 unsigned long shared_hugetlb; 404 unsigned long private_hugetlb; 405 u64 pss; 406 u64 pss_anon; 407 u64 pss_file; 408 u64 pss_shmem; 409 u64 pss_locked; 410 u64 swap_pss; 411 }; 412 413 static void smaps_page_accumulate(struct mem_size_stats *mss, 414 struct page *page, unsigned long size, unsigned long pss, 415 bool dirty, bool locked, bool private) 416 { 417 mss->pss += pss; 418 419 if (PageAnon(page)) 420 mss->pss_anon += pss; 421 else if (PageSwapBacked(page)) 422 mss->pss_shmem += pss; 423 else 424 mss->pss_file += pss; 425 426 if (locked) 427 mss->pss_locked += pss; 428 429 if (dirty || PageDirty(page)) { 430 if (private) 431 mss->private_dirty += size; 432 else 433 mss->shared_dirty += size; 434 } else { 435 if (private) 436 mss->private_clean += size; 437 else 438 mss->shared_clean += size; 439 } 440 } 441 442 static void smaps_account(struct mem_size_stats *mss, struct page *page, 443 bool compound, bool young, bool dirty, bool locked, 444 bool migration) 445 { 446 int i, nr = compound ? compound_nr(page) : 1; 447 unsigned long size = nr * PAGE_SIZE; 448 449 /* 450 * First accumulate quantities that depend only on |size| and the type 451 * of the compound page. 452 */ 453 if (PageAnon(page)) { 454 mss->anonymous += size; 455 if (!PageSwapBacked(page) && !dirty && !PageDirty(page)) 456 mss->lazyfree += size; 457 } 458 459 mss->resident += size; 460 /* Accumulate the size in pages that have been accessed. */ 461 if (young || page_is_young(page) || PageReferenced(page)) 462 mss->referenced += size; 463 464 /* 465 * Then accumulate quantities that may depend on sharing, or that may 466 * differ page-by-page. 467 * 468 * page_count(page) == 1 guarantees the page is mapped exactly once. 469 * If any subpage of the compound page mapped with PTE it would elevate 470 * page_count(). 471 * 472 * The page_mapcount() is called to get a snapshot of the mapcount. 473 * Without holding the page lock this snapshot can be slightly wrong as 474 * we cannot always read the mapcount atomically. It is not safe to 475 * call page_mapcount() even with PTL held if the page is not mapped, 476 * especially for migration entries. Treat regular migration entries 477 * as mapcount == 1. 478 */ 479 if ((page_count(page) == 1) || migration) { 480 smaps_page_accumulate(mss, page, size, size << PSS_SHIFT, dirty, 481 locked, true); 482 return; 483 } 484 for (i = 0; i < nr; i++, page++) { 485 int mapcount = page_mapcount(page); 486 unsigned long pss = PAGE_SIZE << PSS_SHIFT; 487 if (mapcount >= 2) 488 pss /= mapcount; 489 smaps_page_accumulate(mss, page, PAGE_SIZE, pss, dirty, locked, 490 mapcount < 2); 491 } 492 } 493 494 #ifdef CONFIG_SHMEM 495 static int smaps_pte_hole(unsigned long addr, unsigned long end, 496 __always_unused int depth, struct mm_walk *walk) 497 { 498 struct mem_size_stats *mss = walk->private; 499 struct vm_area_struct *vma = walk->vma; 500 501 mss->swap += shmem_partial_swap_usage(walk->vma->vm_file->f_mapping, 502 linear_page_index(vma, addr), 503 linear_page_index(vma, end)); 504 505 return 0; 506 } 507 #else 508 #define smaps_pte_hole NULL 509 #endif /* CONFIG_SHMEM */ 510 511 static void smaps_pte_hole_lookup(unsigned long addr, struct mm_walk *walk) 512 { 513 #ifdef CONFIG_SHMEM 514 if (walk->ops->pte_hole) { 515 /* depth is not used */ 516 smaps_pte_hole(addr, addr + PAGE_SIZE, 0, walk); 517 } 518 #endif 519 } 520 521 static void smaps_pte_entry(pte_t *pte, unsigned long addr, 522 struct mm_walk *walk) 523 { 524 struct mem_size_stats *mss = walk->private; 525 struct vm_area_struct *vma = walk->vma; 526 bool locked = !!(vma->vm_flags & VM_LOCKED); 527 struct page *page = NULL; 528 bool migration = false; 529 530 if (pte_present(*pte)) { 531 page = vm_normal_page(vma, addr, *pte); 532 } else if (is_swap_pte(*pte)) { 533 swp_entry_t swpent = pte_to_swp_entry(*pte); 534 535 if (!non_swap_entry(swpent)) { 536 int mapcount; 537 538 mss->swap += PAGE_SIZE; 539 mapcount = swp_swapcount(swpent); 540 if (mapcount >= 2) { 541 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT; 542 543 do_div(pss_delta, mapcount); 544 mss->swap_pss += pss_delta; 545 } else { 546 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT; 547 } 548 } else if (is_pfn_swap_entry(swpent)) { 549 if (is_migration_entry(swpent)) 550 migration = true; 551 page = pfn_swap_entry_to_page(swpent); 552 } 553 } else { 554 smaps_pte_hole_lookup(addr, walk); 555 return; 556 } 557 558 if (!page) 559 return; 560 561 smaps_account(mss, page, false, pte_young(*pte), pte_dirty(*pte), 562 locked, migration); 563 } 564 565 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 566 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr, 567 struct mm_walk *walk) 568 { 569 struct mem_size_stats *mss = walk->private; 570 struct vm_area_struct *vma = walk->vma; 571 bool locked = !!(vma->vm_flags & VM_LOCKED); 572 struct page *page = NULL; 573 bool migration = false; 574 575 if (pmd_present(*pmd)) { 576 /* FOLL_DUMP will return -EFAULT on huge zero page */ 577 page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP); 578 } else if (unlikely(thp_migration_supported() && is_swap_pmd(*pmd))) { 579 swp_entry_t entry = pmd_to_swp_entry(*pmd); 580 581 if (is_migration_entry(entry)) { 582 migration = true; 583 page = pfn_swap_entry_to_page(entry); 584 } 585 } 586 if (IS_ERR_OR_NULL(page)) 587 return; 588 if (PageAnon(page)) 589 mss->anonymous_thp += HPAGE_PMD_SIZE; 590 else if (PageSwapBacked(page)) 591 mss->shmem_thp += HPAGE_PMD_SIZE; 592 else if (is_zone_device_page(page)) 593 /* pass */; 594 else 595 mss->file_thp += HPAGE_PMD_SIZE; 596 597 smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd), 598 locked, migration); 599 } 600 #else 601 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr, 602 struct mm_walk *walk) 603 { 604 } 605 #endif 606 607 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, 608 struct mm_walk *walk) 609 { 610 struct vm_area_struct *vma = walk->vma; 611 pte_t *pte; 612 spinlock_t *ptl; 613 614 ptl = pmd_trans_huge_lock(pmd, vma); 615 if (ptl) { 616 smaps_pmd_entry(pmd, addr, walk); 617 spin_unlock(ptl); 618 goto out; 619 } 620 621 if (pmd_trans_unstable(pmd)) 622 goto out; 623 /* 624 * The mmap_lock held all the way back in m_start() is what 625 * keeps khugepaged out of here and from collapsing things 626 * in here. 627 */ 628 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); 629 for (; addr != end; pte++, addr += PAGE_SIZE) 630 smaps_pte_entry(pte, addr, walk); 631 pte_unmap_unlock(pte - 1, ptl); 632 out: 633 cond_resched(); 634 return 0; 635 } 636 637 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma) 638 { 639 /* 640 * Don't forget to update Documentation/ on changes. 641 */ 642 static const char mnemonics[BITS_PER_LONG][2] = { 643 /* 644 * In case if we meet a flag we don't know about. 645 */ 646 [0 ... (BITS_PER_LONG-1)] = "??", 647 648 [ilog2(VM_READ)] = "rd", 649 [ilog2(VM_WRITE)] = "wr", 650 [ilog2(VM_EXEC)] = "ex", 651 [ilog2(VM_SHARED)] = "sh", 652 [ilog2(VM_MAYREAD)] = "mr", 653 [ilog2(VM_MAYWRITE)] = "mw", 654 [ilog2(VM_MAYEXEC)] = "me", 655 [ilog2(VM_MAYSHARE)] = "ms", 656 [ilog2(VM_GROWSDOWN)] = "gd", 657 [ilog2(VM_PFNMAP)] = "pf", 658 [ilog2(VM_LOCKED)] = "lo", 659 [ilog2(VM_IO)] = "io", 660 [ilog2(VM_SEQ_READ)] = "sr", 661 [ilog2(VM_RAND_READ)] = "rr", 662 [ilog2(VM_DONTCOPY)] = "dc", 663 [ilog2(VM_DONTEXPAND)] = "de", 664 [ilog2(VM_ACCOUNT)] = "ac", 665 [ilog2(VM_NORESERVE)] = "nr", 666 [ilog2(VM_HUGETLB)] = "ht", 667 [ilog2(VM_SYNC)] = "sf", 668 [ilog2(VM_ARCH_1)] = "ar", 669 [ilog2(VM_WIPEONFORK)] = "wf", 670 [ilog2(VM_DONTDUMP)] = "dd", 671 #ifdef CONFIG_ARM64_BTI 672 [ilog2(VM_ARM64_BTI)] = "bt", 673 #endif 674 #ifdef CONFIG_MEM_SOFT_DIRTY 675 [ilog2(VM_SOFTDIRTY)] = "sd", 676 #endif 677 [ilog2(VM_MIXEDMAP)] = "mm", 678 [ilog2(VM_HUGEPAGE)] = "hg", 679 [ilog2(VM_NOHUGEPAGE)] = "nh", 680 [ilog2(VM_MERGEABLE)] = "mg", 681 [ilog2(VM_UFFD_MISSING)]= "um", 682 [ilog2(VM_UFFD_WP)] = "uw", 683 #ifdef CONFIG_ARM64_MTE 684 [ilog2(VM_MTE)] = "mt", 685 [ilog2(VM_MTE_ALLOWED)] = "", 686 #endif 687 #ifdef CONFIG_ARCH_HAS_PKEYS 688 /* These come out via ProtectionKey: */ 689 [ilog2(VM_PKEY_BIT0)] = "", 690 [ilog2(VM_PKEY_BIT1)] = "", 691 [ilog2(VM_PKEY_BIT2)] = "", 692 [ilog2(VM_PKEY_BIT3)] = "", 693 #if VM_PKEY_BIT4 694 [ilog2(VM_PKEY_BIT4)] = "", 695 #endif 696 #endif /* CONFIG_ARCH_HAS_PKEYS */ 697 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_MINOR 698 [ilog2(VM_UFFD_MINOR)] = "ui", 699 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_MINOR */ 700 }; 701 size_t i; 702 703 seq_puts(m, "VmFlags: "); 704 for (i = 0; i < BITS_PER_LONG; i++) { 705 if (!mnemonics[i][0]) 706 continue; 707 if (vma->vm_flags & (1UL << i)) { 708 seq_putc(m, mnemonics[i][0]); 709 seq_putc(m, mnemonics[i][1]); 710 seq_putc(m, ' '); 711 } 712 } 713 seq_putc(m, '\n'); 714 } 715 716 #ifdef CONFIG_HUGETLB_PAGE 717 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask, 718 unsigned long addr, unsigned long end, 719 struct mm_walk *walk) 720 { 721 struct mem_size_stats *mss = walk->private; 722 struct vm_area_struct *vma = walk->vma; 723 struct page *page = NULL; 724 725 if (pte_present(*pte)) { 726 page = vm_normal_page(vma, addr, *pte); 727 } else if (is_swap_pte(*pte)) { 728 swp_entry_t swpent = pte_to_swp_entry(*pte); 729 730 if (is_pfn_swap_entry(swpent)) 731 page = pfn_swap_entry_to_page(swpent); 732 } 733 if (page) { 734 int mapcount = page_mapcount(page); 735 736 if (mapcount >= 2) 737 mss->shared_hugetlb += huge_page_size(hstate_vma(vma)); 738 else 739 mss->private_hugetlb += huge_page_size(hstate_vma(vma)); 740 } 741 return 0; 742 } 743 #else 744 #define smaps_hugetlb_range NULL 745 #endif /* HUGETLB_PAGE */ 746 747 static const struct mm_walk_ops smaps_walk_ops = { 748 .pmd_entry = smaps_pte_range, 749 .hugetlb_entry = smaps_hugetlb_range, 750 }; 751 752 static const struct mm_walk_ops smaps_shmem_walk_ops = { 753 .pmd_entry = smaps_pte_range, 754 .hugetlb_entry = smaps_hugetlb_range, 755 .pte_hole = smaps_pte_hole, 756 }; 757 758 /* 759 * Gather mem stats from @vma with the indicated beginning 760 * address @start, and keep them in @mss. 761 * 762 * Use vm_start of @vma as the beginning address if @start is 0. 763 */ 764 static void smap_gather_stats(struct vm_area_struct *vma, 765 struct mem_size_stats *mss, unsigned long start) 766 { 767 const struct mm_walk_ops *ops = &smaps_walk_ops; 768 769 /* Invalid start */ 770 if (start >= vma->vm_end) 771 return; 772 773 #ifdef CONFIG_SHMEM 774 if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) { 775 /* 776 * For shared or readonly shmem mappings we know that all 777 * swapped out pages belong to the shmem object, and we can 778 * obtain the swap value much more efficiently. For private 779 * writable mappings, we might have COW pages that are 780 * not affected by the parent swapped out pages of the shmem 781 * object, so we have to distinguish them during the page walk. 782 * Unless we know that the shmem object (or the part mapped by 783 * our VMA) has no swapped out pages at all. 784 */ 785 unsigned long shmem_swapped = shmem_swap_usage(vma); 786 787 if (!start && (!shmem_swapped || (vma->vm_flags & VM_SHARED) || 788 !(vma->vm_flags & VM_WRITE))) { 789 mss->swap += shmem_swapped; 790 } else { 791 ops = &smaps_shmem_walk_ops; 792 } 793 } 794 #endif 795 /* mmap_lock is held in m_start */ 796 if (!start) 797 walk_page_vma(vma, ops, mss); 798 else 799 walk_page_range(vma->vm_mm, start, vma->vm_end, ops, mss); 800 } 801 802 #define SEQ_PUT_DEC(str, val) \ 803 seq_put_decimal_ull_width(m, str, (val) >> 10, 8) 804 805 /* Show the contents common for smaps and smaps_rollup */ 806 static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss, 807 bool rollup_mode) 808 { 809 SEQ_PUT_DEC("Rss: ", mss->resident); 810 SEQ_PUT_DEC(" kB\nPss: ", mss->pss >> PSS_SHIFT); 811 if (rollup_mode) { 812 /* 813 * These are meaningful only for smaps_rollup, otherwise two of 814 * them are zero, and the other one is the same as Pss. 815 */ 816 SEQ_PUT_DEC(" kB\nPss_Anon: ", 817 mss->pss_anon >> PSS_SHIFT); 818 SEQ_PUT_DEC(" kB\nPss_File: ", 819 mss->pss_file >> PSS_SHIFT); 820 SEQ_PUT_DEC(" kB\nPss_Shmem: ", 821 mss->pss_shmem >> PSS_SHIFT); 822 } 823 SEQ_PUT_DEC(" kB\nShared_Clean: ", mss->shared_clean); 824 SEQ_PUT_DEC(" kB\nShared_Dirty: ", mss->shared_dirty); 825 SEQ_PUT_DEC(" kB\nPrivate_Clean: ", mss->private_clean); 826 SEQ_PUT_DEC(" kB\nPrivate_Dirty: ", mss->private_dirty); 827 SEQ_PUT_DEC(" kB\nReferenced: ", mss->referenced); 828 SEQ_PUT_DEC(" kB\nAnonymous: ", mss->anonymous); 829 SEQ_PUT_DEC(" kB\nLazyFree: ", mss->lazyfree); 830 SEQ_PUT_DEC(" kB\nAnonHugePages: ", mss->anonymous_thp); 831 SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp); 832 SEQ_PUT_DEC(" kB\nFilePmdMapped: ", mss->file_thp); 833 SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb); 834 seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ", 835 mss->private_hugetlb >> 10, 7); 836 SEQ_PUT_DEC(" kB\nSwap: ", mss->swap); 837 SEQ_PUT_DEC(" kB\nSwapPss: ", 838 mss->swap_pss >> PSS_SHIFT); 839 SEQ_PUT_DEC(" kB\nLocked: ", 840 mss->pss_locked >> PSS_SHIFT); 841 seq_puts(m, " kB\n"); 842 } 843 844 static int show_smap(struct seq_file *m, void *v) 845 { 846 struct vm_area_struct *vma = v; 847 struct mem_size_stats mss; 848 849 memset(&mss, 0, sizeof(mss)); 850 851 smap_gather_stats(vma, &mss, 0); 852 853 show_map_vma(m, vma); 854 855 SEQ_PUT_DEC("Size: ", vma->vm_end - vma->vm_start); 856 SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma)); 857 SEQ_PUT_DEC(" kB\nMMUPageSize: ", vma_mmu_pagesize(vma)); 858 seq_puts(m, " kB\n"); 859 860 __show_smap(m, &mss, false); 861 862 seq_printf(m, "THPeligible: %d\n", 863 transparent_hugepage_active(vma)); 864 865 if (arch_pkeys_enabled()) 866 seq_printf(m, "ProtectionKey: %8u\n", vma_pkey(vma)); 867 show_smap_vma_flags(m, vma); 868 869 return 0; 870 } 871 872 static int show_smaps_rollup(struct seq_file *m, void *v) 873 { 874 struct proc_maps_private *priv = m->private; 875 struct mem_size_stats mss; 876 struct mm_struct *mm; 877 struct vm_area_struct *vma; 878 unsigned long last_vma_end = 0; 879 int ret = 0; 880 881 priv->task = get_proc_task(priv->inode); 882 if (!priv->task) 883 return -ESRCH; 884 885 mm = priv->mm; 886 if (!mm || !mmget_not_zero(mm)) { 887 ret = -ESRCH; 888 goto out_put_task; 889 } 890 891 memset(&mss, 0, sizeof(mss)); 892 893 ret = mmap_read_lock_killable(mm); 894 if (ret) 895 goto out_put_mm; 896 897 hold_task_mempolicy(priv); 898 899 for (vma = priv->mm->mmap; vma;) { 900 smap_gather_stats(vma, &mss, 0); 901 last_vma_end = vma->vm_end; 902 903 /* 904 * Release mmap_lock temporarily if someone wants to 905 * access it for write request. 906 */ 907 if (mmap_lock_is_contended(mm)) { 908 mmap_read_unlock(mm); 909 ret = mmap_read_lock_killable(mm); 910 if (ret) { 911 release_task_mempolicy(priv); 912 goto out_put_mm; 913 } 914 915 /* 916 * After dropping the lock, there are four cases to 917 * consider. See the following example for explanation. 918 * 919 * +------+------+-----------+ 920 * | VMA1 | VMA2 | VMA3 | 921 * +------+------+-----------+ 922 * | | | | 923 * 4k 8k 16k 400k 924 * 925 * Suppose we drop the lock after reading VMA2 due to 926 * contention, then we get: 927 * 928 * last_vma_end = 16k 929 * 930 * 1) VMA2 is freed, but VMA3 exists: 931 * 932 * find_vma(mm, 16k - 1) will return VMA3. 933 * In this case, just continue from VMA3. 934 * 935 * 2) VMA2 still exists: 936 * 937 * find_vma(mm, 16k - 1) will return VMA2. 938 * Iterate the loop like the original one. 939 * 940 * 3) No more VMAs can be found: 941 * 942 * find_vma(mm, 16k - 1) will return NULL. 943 * No more things to do, just break. 944 * 945 * 4) (last_vma_end - 1) is the middle of a vma (VMA'): 946 * 947 * find_vma(mm, 16k - 1) will return VMA' whose range 948 * contains last_vma_end. 949 * Iterate VMA' from last_vma_end. 950 */ 951 vma = find_vma(mm, last_vma_end - 1); 952 /* Case 3 above */ 953 if (!vma) 954 break; 955 956 /* Case 1 above */ 957 if (vma->vm_start >= last_vma_end) 958 continue; 959 960 /* Case 4 above */ 961 if (vma->vm_end > last_vma_end) 962 smap_gather_stats(vma, &mss, last_vma_end); 963 } 964 /* Case 2 above */ 965 vma = vma->vm_next; 966 } 967 968 show_vma_header_prefix(m, priv->mm->mmap->vm_start, 969 last_vma_end, 0, 0, 0, 0); 970 seq_pad(m, ' '); 971 seq_puts(m, "[rollup]\n"); 972 973 __show_smap(m, &mss, true); 974 975 release_task_mempolicy(priv); 976 mmap_read_unlock(mm); 977 978 out_put_mm: 979 mmput(mm); 980 out_put_task: 981 put_task_struct(priv->task); 982 priv->task = NULL; 983 984 return ret; 985 } 986 #undef SEQ_PUT_DEC 987 988 static const struct seq_operations proc_pid_smaps_op = { 989 .start = m_start, 990 .next = m_next, 991 .stop = m_stop, 992 .show = show_smap 993 }; 994 995 static int pid_smaps_open(struct inode *inode, struct file *file) 996 { 997 return do_maps_open(inode, file, &proc_pid_smaps_op); 998 } 999 1000 static int smaps_rollup_open(struct inode *inode, struct file *file) 1001 { 1002 int ret; 1003 struct proc_maps_private *priv; 1004 1005 priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT); 1006 if (!priv) 1007 return -ENOMEM; 1008 1009 ret = single_open(file, show_smaps_rollup, priv); 1010 if (ret) 1011 goto out_free; 1012 1013 priv->inode = inode; 1014 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ); 1015 if (IS_ERR(priv->mm)) { 1016 ret = PTR_ERR(priv->mm); 1017 1018 single_release(inode, file); 1019 goto out_free; 1020 } 1021 1022 return 0; 1023 1024 out_free: 1025 kfree(priv); 1026 return ret; 1027 } 1028 1029 static int smaps_rollup_release(struct inode *inode, struct file *file) 1030 { 1031 struct seq_file *seq = file->private_data; 1032 struct proc_maps_private *priv = seq->private; 1033 1034 if (priv->mm) 1035 mmdrop(priv->mm); 1036 1037 kfree(priv); 1038 return single_release(inode, file); 1039 } 1040 1041 const struct file_operations proc_pid_smaps_operations = { 1042 .open = pid_smaps_open, 1043 .read = seq_read, 1044 .llseek = seq_lseek, 1045 .release = proc_map_release, 1046 }; 1047 1048 const struct file_operations proc_pid_smaps_rollup_operations = { 1049 .open = smaps_rollup_open, 1050 .read = seq_read, 1051 .llseek = seq_lseek, 1052 .release = smaps_rollup_release, 1053 }; 1054 1055 enum clear_refs_types { 1056 CLEAR_REFS_ALL = 1, 1057 CLEAR_REFS_ANON, 1058 CLEAR_REFS_MAPPED, 1059 CLEAR_REFS_SOFT_DIRTY, 1060 CLEAR_REFS_MM_HIWATER_RSS, 1061 CLEAR_REFS_LAST, 1062 }; 1063 1064 struct clear_refs_private { 1065 enum clear_refs_types type; 1066 }; 1067 1068 #ifdef CONFIG_MEM_SOFT_DIRTY 1069 1070 static inline bool pte_is_pinned(struct vm_area_struct *vma, unsigned long addr, pte_t pte) 1071 { 1072 struct page *page; 1073 1074 if (!pte_write(pte)) 1075 return false; 1076 if (!is_cow_mapping(vma->vm_flags)) 1077 return false; 1078 if (likely(!test_bit(MMF_HAS_PINNED, &vma->vm_mm->flags))) 1079 return false; 1080 page = vm_normal_page(vma, addr, pte); 1081 if (!page) 1082 return false; 1083 return page_maybe_dma_pinned(page); 1084 } 1085 1086 static inline void clear_soft_dirty(struct vm_area_struct *vma, 1087 unsigned long addr, pte_t *pte) 1088 { 1089 /* 1090 * The soft-dirty tracker uses #PF-s to catch writes 1091 * to pages, so write-protect the pte as well. See the 1092 * Documentation/admin-guide/mm/soft-dirty.rst for full description 1093 * of how soft-dirty works. 1094 */ 1095 pte_t ptent = *pte; 1096 1097 if (pte_present(ptent)) { 1098 pte_t old_pte; 1099 1100 if (pte_is_pinned(vma, addr, ptent)) 1101 return; 1102 old_pte = ptep_modify_prot_start(vma, addr, pte); 1103 ptent = pte_wrprotect(old_pte); 1104 ptent = pte_clear_soft_dirty(ptent); 1105 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent); 1106 } else if (is_swap_pte(ptent)) { 1107 ptent = pte_swp_clear_soft_dirty(ptent); 1108 set_pte_at(vma->vm_mm, addr, pte, ptent); 1109 } 1110 } 1111 #else 1112 static inline void clear_soft_dirty(struct vm_area_struct *vma, 1113 unsigned long addr, pte_t *pte) 1114 { 1115 } 1116 #endif 1117 1118 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE) 1119 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma, 1120 unsigned long addr, pmd_t *pmdp) 1121 { 1122 pmd_t old, pmd = *pmdp; 1123 1124 if (pmd_present(pmd)) { 1125 /* See comment in change_huge_pmd() */ 1126 old = pmdp_invalidate(vma, addr, pmdp); 1127 if (pmd_dirty(old)) 1128 pmd = pmd_mkdirty(pmd); 1129 if (pmd_young(old)) 1130 pmd = pmd_mkyoung(pmd); 1131 1132 pmd = pmd_wrprotect(pmd); 1133 pmd = pmd_clear_soft_dirty(pmd); 1134 1135 set_pmd_at(vma->vm_mm, addr, pmdp, pmd); 1136 } else if (is_migration_entry(pmd_to_swp_entry(pmd))) { 1137 pmd = pmd_swp_clear_soft_dirty(pmd); 1138 set_pmd_at(vma->vm_mm, addr, pmdp, pmd); 1139 } 1140 } 1141 #else 1142 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma, 1143 unsigned long addr, pmd_t *pmdp) 1144 { 1145 } 1146 #endif 1147 1148 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr, 1149 unsigned long end, struct mm_walk *walk) 1150 { 1151 struct clear_refs_private *cp = walk->private; 1152 struct vm_area_struct *vma = walk->vma; 1153 pte_t *pte, ptent; 1154 spinlock_t *ptl; 1155 struct page *page; 1156 1157 ptl = pmd_trans_huge_lock(pmd, vma); 1158 if (ptl) { 1159 if (cp->type == CLEAR_REFS_SOFT_DIRTY) { 1160 clear_soft_dirty_pmd(vma, addr, pmd); 1161 goto out; 1162 } 1163 1164 if (!pmd_present(*pmd)) 1165 goto out; 1166 1167 page = pmd_page(*pmd); 1168 1169 /* Clear accessed and referenced bits. */ 1170 pmdp_test_and_clear_young(vma, addr, pmd); 1171 test_and_clear_page_young(page); 1172 ClearPageReferenced(page); 1173 out: 1174 spin_unlock(ptl); 1175 return 0; 1176 } 1177 1178 if (pmd_trans_unstable(pmd)) 1179 return 0; 1180 1181 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); 1182 for (; addr != end; pte++, addr += PAGE_SIZE) { 1183 ptent = *pte; 1184 1185 if (cp->type == CLEAR_REFS_SOFT_DIRTY) { 1186 clear_soft_dirty(vma, addr, pte); 1187 continue; 1188 } 1189 1190 if (!pte_present(ptent)) 1191 continue; 1192 1193 page = vm_normal_page(vma, addr, ptent); 1194 if (!page) 1195 continue; 1196 1197 /* Clear accessed and referenced bits. */ 1198 ptep_test_and_clear_young(vma, addr, pte); 1199 test_and_clear_page_young(page); 1200 ClearPageReferenced(page); 1201 } 1202 pte_unmap_unlock(pte - 1, ptl); 1203 cond_resched(); 1204 return 0; 1205 } 1206 1207 static int clear_refs_test_walk(unsigned long start, unsigned long end, 1208 struct mm_walk *walk) 1209 { 1210 struct clear_refs_private *cp = walk->private; 1211 struct vm_area_struct *vma = walk->vma; 1212 1213 if (vma->vm_flags & VM_PFNMAP) 1214 return 1; 1215 1216 /* 1217 * Writing 1 to /proc/pid/clear_refs affects all pages. 1218 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages. 1219 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages. 1220 * Writing 4 to /proc/pid/clear_refs affects all pages. 1221 */ 1222 if (cp->type == CLEAR_REFS_ANON && vma->vm_file) 1223 return 1; 1224 if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file) 1225 return 1; 1226 return 0; 1227 } 1228 1229 static const struct mm_walk_ops clear_refs_walk_ops = { 1230 .pmd_entry = clear_refs_pte_range, 1231 .test_walk = clear_refs_test_walk, 1232 }; 1233 1234 static ssize_t clear_refs_write(struct file *file, const char __user *buf, 1235 size_t count, loff_t *ppos) 1236 { 1237 struct task_struct *task; 1238 char buffer[PROC_NUMBUF]; 1239 struct mm_struct *mm; 1240 struct vm_area_struct *vma; 1241 enum clear_refs_types type; 1242 int itype; 1243 int rv; 1244 1245 memset(buffer, 0, sizeof(buffer)); 1246 if (count > sizeof(buffer) - 1) 1247 count = sizeof(buffer) - 1; 1248 if (copy_from_user(buffer, buf, count)) 1249 return -EFAULT; 1250 rv = kstrtoint(strstrip(buffer), 10, &itype); 1251 if (rv < 0) 1252 return rv; 1253 type = (enum clear_refs_types)itype; 1254 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST) 1255 return -EINVAL; 1256 1257 task = get_proc_task(file_inode(file)); 1258 if (!task) 1259 return -ESRCH; 1260 mm = get_task_mm(task); 1261 if (mm) { 1262 struct mmu_notifier_range range; 1263 struct clear_refs_private cp = { 1264 .type = type, 1265 }; 1266 1267 if (mmap_write_lock_killable(mm)) { 1268 count = -EINTR; 1269 goto out_mm; 1270 } 1271 if (type == CLEAR_REFS_MM_HIWATER_RSS) { 1272 /* 1273 * Writing 5 to /proc/pid/clear_refs resets the peak 1274 * resident set size to this mm's current rss value. 1275 */ 1276 reset_mm_hiwater_rss(mm); 1277 goto out_unlock; 1278 } 1279 1280 if (type == CLEAR_REFS_SOFT_DIRTY) { 1281 for (vma = mm->mmap; vma; vma = vma->vm_next) { 1282 if (!(vma->vm_flags & VM_SOFTDIRTY)) 1283 continue; 1284 vma->vm_flags &= ~VM_SOFTDIRTY; 1285 vma_set_page_prot(vma); 1286 } 1287 1288 inc_tlb_flush_pending(mm); 1289 mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY, 1290 0, NULL, mm, 0, -1UL); 1291 mmu_notifier_invalidate_range_start(&range); 1292 } 1293 walk_page_range(mm, 0, mm->highest_vm_end, &clear_refs_walk_ops, 1294 &cp); 1295 if (type == CLEAR_REFS_SOFT_DIRTY) { 1296 mmu_notifier_invalidate_range_end(&range); 1297 flush_tlb_mm(mm); 1298 dec_tlb_flush_pending(mm); 1299 } 1300 out_unlock: 1301 mmap_write_unlock(mm); 1302 out_mm: 1303 mmput(mm); 1304 } 1305 put_task_struct(task); 1306 1307 return count; 1308 } 1309 1310 const struct file_operations proc_clear_refs_operations = { 1311 .write = clear_refs_write, 1312 .llseek = noop_llseek, 1313 }; 1314 1315 typedef struct { 1316 u64 pme; 1317 } pagemap_entry_t; 1318 1319 struct pagemapread { 1320 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */ 1321 pagemap_entry_t *buffer; 1322 bool show_pfn; 1323 }; 1324 1325 #define PAGEMAP_WALK_SIZE (PMD_SIZE) 1326 #define PAGEMAP_WALK_MASK (PMD_MASK) 1327 1328 #define PM_ENTRY_BYTES sizeof(pagemap_entry_t) 1329 #define PM_PFRAME_BITS 55 1330 #define PM_PFRAME_MASK GENMASK_ULL(PM_PFRAME_BITS - 1, 0) 1331 #define PM_SOFT_DIRTY BIT_ULL(55) 1332 #define PM_MMAP_EXCLUSIVE BIT_ULL(56) 1333 #define PM_UFFD_WP BIT_ULL(57) 1334 #define PM_FILE BIT_ULL(61) 1335 #define PM_SWAP BIT_ULL(62) 1336 #define PM_PRESENT BIT_ULL(63) 1337 1338 #define PM_END_OF_BUFFER 1 1339 1340 static inline pagemap_entry_t make_pme(u64 frame, u64 flags) 1341 { 1342 return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags }; 1343 } 1344 1345 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme, 1346 struct pagemapread *pm) 1347 { 1348 pm->buffer[pm->pos++] = *pme; 1349 if (pm->pos >= pm->len) 1350 return PM_END_OF_BUFFER; 1351 return 0; 1352 } 1353 1354 static int pagemap_pte_hole(unsigned long start, unsigned long end, 1355 __always_unused int depth, struct mm_walk *walk) 1356 { 1357 struct pagemapread *pm = walk->private; 1358 unsigned long addr = start; 1359 int err = 0; 1360 1361 while (addr < end) { 1362 struct vm_area_struct *vma = find_vma(walk->mm, addr); 1363 pagemap_entry_t pme = make_pme(0, 0); 1364 /* End of address space hole, which we mark as non-present. */ 1365 unsigned long hole_end; 1366 1367 if (vma) 1368 hole_end = min(end, vma->vm_start); 1369 else 1370 hole_end = end; 1371 1372 for (; addr < hole_end; addr += PAGE_SIZE) { 1373 err = add_to_pagemap(addr, &pme, pm); 1374 if (err) 1375 goto out; 1376 } 1377 1378 if (!vma) 1379 break; 1380 1381 /* Addresses in the VMA. */ 1382 if (vma->vm_flags & VM_SOFTDIRTY) 1383 pme = make_pme(0, PM_SOFT_DIRTY); 1384 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) { 1385 err = add_to_pagemap(addr, &pme, pm); 1386 if (err) 1387 goto out; 1388 } 1389 } 1390 out: 1391 return err; 1392 } 1393 1394 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm, 1395 struct vm_area_struct *vma, unsigned long addr, pte_t pte) 1396 { 1397 u64 frame = 0, flags = 0; 1398 struct page *page = NULL; 1399 bool migration = false; 1400 1401 if (pte_present(pte)) { 1402 if (pm->show_pfn) 1403 frame = pte_pfn(pte); 1404 flags |= PM_PRESENT; 1405 page = vm_normal_page(vma, addr, pte); 1406 if (pte_soft_dirty(pte)) 1407 flags |= PM_SOFT_DIRTY; 1408 if (pte_uffd_wp(pte)) 1409 flags |= PM_UFFD_WP; 1410 } else if (is_swap_pte(pte)) { 1411 swp_entry_t entry; 1412 if (pte_swp_soft_dirty(pte)) 1413 flags |= PM_SOFT_DIRTY; 1414 if (pte_swp_uffd_wp(pte)) 1415 flags |= PM_UFFD_WP; 1416 entry = pte_to_swp_entry(pte); 1417 if (pm->show_pfn) 1418 frame = swp_type(entry) | 1419 (swp_offset(entry) << MAX_SWAPFILES_SHIFT); 1420 flags |= PM_SWAP; 1421 migration = is_migration_entry(entry); 1422 if (is_pfn_swap_entry(entry)) 1423 page = pfn_swap_entry_to_page(entry); 1424 } 1425 1426 if (page && !PageAnon(page)) 1427 flags |= PM_FILE; 1428 if (page && !migration && page_mapcount(page) == 1) 1429 flags |= PM_MMAP_EXCLUSIVE; 1430 if (vma->vm_flags & VM_SOFTDIRTY) 1431 flags |= PM_SOFT_DIRTY; 1432 1433 return make_pme(frame, flags); 1434 } 1435 1436 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end, 1437 struct mm_walk *walk) 1438 { 1439 struct vm_area_struct *vma = walk->vma; 1440 struct pagemapread *pm = walk->private; 1441 spinlock_t *ptl; 1442 pte_t *pte, *orig_pte; 1443 int err = 0; 1444 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1445 bool migration = false; 1446 1447 ptl = pmd_trans_huge_lock(pmdp, vma); 1448 if (ptl) { 1449 u64 flags = 0, frame = 0; 1450 pmd_t pmd = *pmdp; 1451 struct page *page = NULL; 1452 1453 if (vma->vm_flags & VM_SOFTDIRTY) 1454 flags |= PM_SOFT_DIRTY; 1455 1456 if (pmd_present(pmd)) { 1457 page = pmd_page(pmd); 1458 1459 flags |= PM_PRESENT; 1460 if (pmd_soft_dirty(pmd)) 1461 flags |= PM_SOFT_DIRTY; 1462 if (pmd_uffd_wp(pmd)) 1463 flags |= PM_UFFD_WP; 1464 if (pm->show_pfn) 1465 frame = pmd_pfn(pmd) + 1466 ((addr & ~PMD_MASK) >> PAGE_SHIFT); 1467 } 1468 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION 1469 else if (is_swap_pmd(pmd)) { 1470 swp_entry_t entry = pmd_to_swp_entry(pmd); 1471 unsigned long offset; 1472 1473 if (pm->show_pfn) { 1474 offset = swp_offset(entry) + 1475 ((addr & ~PMD_MASK) >> PAGE_SHIFT); 1476 frame = swp_type(entry) | 1477 (offset << MAX_SWAPFILES_SHIFT); 1478 } 1479 flags |= PM_SWAP; 1480 if (pmd_swp_soft_dirty(pmd)) 1481 flags |= PM_SOFT_DIRTY; 1482 if (pmd_swp_uffd_wp(pmd)) 1483 flags |= PM_UFFD_WP; 1484 VM_BUG_ON(!is_pmd_migration_entry(pmd)); 1485 migration = is_migration_entry(entry); 1486 page = pfn_swap_entry_to_page(entry); 1487 } 1488 #endif 1489 1490 if (page && !migration && page_mapcount(page) == 1) 1491 flags |= PM_MMAP_EXCLUSIVE; 1492 1493 for (; addr != end; addr += PAGE_SIZE) { 1494 pagemap_entry_t pme = make_pme(frame, flags); 1495 1496 err = add_to_pagemap(addr, &pme, pm); 1497 if (err) 1498 break; 1499 if (pm->show_pfn) { 1500 if (flags & PM_PRESENT) 1501 frame++; 1502 else if (flags & PM_SWAP) 1503 frame += (1 << MAX_SWAPFILES_SHIFT); 1504 } 1505 } 1506 spin_unlock(ptl); 1507 return err; 1508 } 1509 1510 if (pmd_trans_unstable(pmdp)) 1511 return 0; 1512 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ 1513 1514 /* 1515 * We can assume that @vma always points to a valid one and @end never 1516 * goes beyond vma->vm_end. 1517 */ 1518 orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl); 1519 for (; addr < end; pte++, addr += PAGE_SIZE) { 1520 pagemap_entry_t pme; 1521 1522 pme = pte_to_pagemap_entry(pm, vma, addr, *pte); 1523 err = add_to_pagemap(addr, &pme, pm); 1524 if (err) 1525 break; 1526 } 1527 pte_unmap_unlock(orig_pte, ptl); 1528 1529 cond_resched(); 1530 1531 return err; 1532 } 1533 1534 #ifdef CONFIG_HUGETLB_PAGE 1535 /* This function walks within one hugetlb entry in the single call */ 1536 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask, 1537 unsigned long addr, unsigned long end, 1538 struct mm_walk *walk) 1539 { 1540 struct pagemapread *pm = walk->private; 1541 struct vm_area_struct *vma = walk->vma; 1542 u64 flags = 0, frame = 0; 1543 int err = 0; 1544 pte_t pte; 1545 1546 if (vma->vm_flags & VM_SOFTDIRTY) 1547 flags |= PM_SOFT_DIRTY; 1548 1549 pte = huge_ptep_get(ptep); 1550 if (pte_present(pte)) { 1551 struct page *page = pte_page(pte); 1552 1553 if (!PageAnon(page)) 1554 flags |= PM_FILE; 1555 1556 if (page_mapcount(page) == 1) 1557 flags |= PM_MMAP_EXCLUSIVE; 1558 1559 flags |= PM_PRESENT; 1560 if (pm->show_pfn) 1561 frame = pte_pfn(pte) + 1562 ((addr & ~hmask) >> PAGE_SHIFT); 1563 } 1564 1565 for (; addr != end; addr += PAGE_SIZE) { 1566 pagemap_entry_t pme = make_pme(frame, flags); 1567 1568 err = add_to_pagemap(addr, &pme, pm); 1569 if (err) 1570 return err; 1571 if (pm->show_pfn && (flags & PM_PRESENT)) 1572 frame++; 1573 } 1574 1575 cond_resched(); 1576 1577 return err; 1578 } 1579 #else 1580 #define pagemap_hugetlb_range NULL 1581 #endif /* HUGETLB_PAGE */ 1582 1583 static const struct mm_walk_ops pagemap_ops = { 1584 .pmd_entry = pagemap_pmd_range, 1585 .pte_hole = pagemap_pte_hole, 1586 .hugetlb_entry = pagemap_hugetlb_range, 1587 }; 1588 1589 /* 1590 * /proc/pid/pagemap - an array mapping virtual pages to pfns 1591 * 1592 * For each page in the address space, this file contains one 64-bit entry 1593 * consisting of the following: 1594 * 1595 * Bits 0-54 page frame number (PFN) if present 1596 * Bits 0-4 swap type if swapped 1597 * Bits 5-54 swap offset if swapped 1598 * Bit 55 pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst) 1599 * Bit 56 page exclusively mapped 1600 * Bit 57 pte is uffd-wp write-protected 1601 * Bits 58-60 zero 1602 * Bit 61 page is file-page or shared-anon 1603 * Bit 62 page swapped 1604 * Bit 63 page present 1605 * 1606 * If the page is not present but in swap, then the PFN contains an 1607 * encoding of the swap file number and the page's offset into the 1608 * swap. Unmapped pages return a null PFN. This allows determining 1609 * precisely which pages are mapped (or in swap) and comparing mapped 1610 * pages between processes. 1611 * 1612 * Efficient users of this interface will use /proc/pid/maps to 1613 * determine which areas of memory are actually mapped and llseek to 1614 * skip over unmapped regions. 1615 */ 1616 static ssize_t pagemap_read(struct file *file, char __user *buf, 1617 size_t count, loff_t *ppos) 1618 { 1619 struct mm_struct *mm = file->private_data; 1620 struct pagemapread pm; 1621 unsigned long src; 1622 unsigned long svpfn; 1623 unsigned long start_vaddr; 1624 unsigned long end_vaddr; 1625 int ret = 0, copied = 0; 1626 1627 if (!mm || !mmget_not_zero(mm)) 1628 goto out; 1629 1630 ret = -EINVAL; 1631 /* file position must be aligned */ 1632 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES)) 1633 goto out_mm; 1634 1635 ret = 0; 1636 if (!count) 1637 goto out_mm; 1638 1639 /* do not disclose physical addresses: attack vector */ 1640 pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN); 1641 1642 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT); 1643 pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL); 1644 ret = -ENOMEM; 1645 if (!pm.buffer) 1646 goto out_mm; 1647 1648 src = *ppos; 1649 svpfn = src / PM_ENTRY_BYTES; 1650 end_vaddr = mm->task_size; 1651 1652 /* watch out for wraparound */ 1653 start_vaddr = end_vaddr; 1654 if (svpfn <= (ULONG_MAX >> PAGE_SHIFT)) 1655 start_vaddr = untagged_addr(svpfn << PAGE_SHIFT); 1656 1657 /* Ensure the address is inside the task */ 1658 if (start_vaddr > mm->task_size) 1659 start_vaddr = end_vaddr; 1660 1661 /* 1662 * The odds are that this will stop walking way 1663 * before end_vaddr, because the length of the 1664 * user buffer is tracked in "pm", and the walk 1665 * will stop when we hit the end of the buffer. 1666 */ 1667 ret = 0; 1668 while (count && (start_vaddr < end_vaddr)) { 1669 int len; 1670 unsigned long end; 1671 1672 pm.pos = 0; 1673 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK; 1674 /* overflow ? */ 1675 if (end < start_vaddr || end > end_vaddr) 1676 end = end_vaddr; 1677 ret = mmap_read_lock_killable(mm); 1678 if (ret) 1679 goto out_free; 1680 ret = walk_page_range(mm, start_vaddr, end, &pagemap_ops, &pm); 1681 mmap_read_unlock(mm); 1682 start_vaddr = end; 1683 1684 len = min(count, PM_ENTRY_BYTES * pm.pos); 1685 if (copy_to_user(buf, pm.buffer, len)) { 1686 ret = -EFAULT; 1687 goto out_free; 1688 } 1689 copied += len; 1690 buf += len; 1691 count -= len; 1692 } 1693 *ppos += copied; 1694 if (!ret || ret == PM_END_OF_BUFFER) 1695 ret = copied; 1696 1697 out_free: 1698 kfree(pm.buffer); 1699 out_mm: 1700 mmput(mm); 1701 out: 1702 return ret; 1703 } 1704 1705 static int pagemap_open(struct inode *inode, struct file *file) 1706 { 1707 struct mm_struct *mm; 1708 1709 mm = proc_mem_open(inode, PTRACE_MODE_READ); 1710 if (IS_ERR(mm)) 1711 return PTR_ERR(mm); 1712 file->private_data = mm; 1713 return 0; 1714 } 1715 1716 static int pagemap_release(struct inode *inode, struct file *file) 1717 { 1718 struct mm_struct *mm = file->private_data; 1719 1720 if (mm) 1721 mmdrop(mm); 1722 return 0; 1723 } 1724 1725 const struct file_operations proc_pagemap_operations = { 1726 .llseek = mem_lseek, /* borrow this */ 1727 .read = pagemap_read, 1728 .open = pagemap_open, 1729 .release = pagemap_release, 1730 }; 1731 #endif /* CONFIG_PROC_PAGE_MONITOR */ 1732 1733 #ifdef CONFIG_NUMA 1734 1735 struct numa_maps { 1736 unsigned long pages; 1737 unsigned long anon; 1738 unsigned long active; 1739 unsigned long writeback; 1740 unsigned long mapcount_max; 1741 unsigned long dirty; 1742 unsigned long swapcache; 1743 unsigned long node[MAX_NUMNODES]; 1744 }; 1745 1746 struct numa_maps_private { 1747 struct proc_maps_private proc_maps; 1748 struct numa_maps md; 1749 }; 1750 1751 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty, 1752 unsigned long nr_pages) 1753 { 1754 int count = page_mapcount(page); 1755 1756 md->pages += nr_pages; 1757 if (pte_dirty || PageDirty(page)) 1758 md->dirty += nr_pages; 1759 1760 if (PageSwapCache(page)) 1761 md->swapcache += nr_pages; 1762 1763 if (PageActive(page) || PageUnevictable(page)) 1764 md->active += nr_pages; 1765 1766 if (PageWriteback(page)) 1767 md->writeback += nr_pages; 1768 1769 if (PageAnon(page)) 1770 md->anon += nr_pages; 1771 1772 if (count > md->mapcount_max) 1773 md->mapcount_max = count; 1774 1775 md->node[page_to_nid(page)] += nr_pages; 1776 } 1777 1778 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma, 1779 unsigned long addr) 1780 { 1781 struct page *page; 1782 int nid; 1783 1784 if (!pte_present(pte)) 1785 return NULL; 1786 1787 page = vm_normal_page(vma, addr, pte); 1788 if (!page) 1789 return NULL; 1790 1791 if (PageReserved(page)) 1792 return NULL; 1793 1794 nid = page_to_nid(page); 1795 if (!node_isset(nid, node_states[N_MEMORY])) 1796 return NULL; 1797 1798 return page; 1799 } 1800 1801 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1802 static struct page *can_gather_numa_stats_pmd(pmd_t pmd, 1803 struct vm_area_struct *vma, 1804 unsigned long addr) 1805 { 1806 struct page *page; 1807 int nid; 1808 1809 if (!pmd_present(pmd)) 1810 return NULL; 1811 1812 page = vm_normal_page_pmd(vma, addr, pmd); 1813 if (!page) 1814 return NULL; 1815 1816 if (PageReserved(page)) 1817 return NULL; 1818 1819 nid = page_to_nid(page); 1820 if (!node_isset(nid, node_states[N_MEMORY])) 1821 return NULL; 1822 1823 return page; 1824 } 1825 #endif 1826 1827 static int gather_pte_stats(pmd_t *pmd, unsigned long addr, 1828 unsigned long end, struct mm_walk *walk) 1829 { 1830 struct numa_maps *md = walk->private; 1831 struct vm_area_struct *vma = walk->vma; 1832 spinlock_t *ptl; 1833 pte_t *orig_pte; 1834 pte_t *pte; 1835 1836 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1837 ptl = pmd_trans_huge_lock(pmd, vma); 1838 if (ptl) { 1839 struct page *page; 1840 1841 page = can_gather_numa_stats_pmd(*pmd, vma, addr); 1842 if (page) 1843 gather_stats(page, md, pmd_dirty(*pmd), 1844 HPAGE_PMD_SIZE/PAGE_SIZE); 1845 spin_unlock(ptl); 1846 return 0; 1847 } 1848 1849 if (pmd_trans_unstable(pmd)) 1850 return 0; 1851 #endif 1852 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl); 1853 do { 1854 struct page *page = can_gather_numa_stats(*pte, vma, addr); 1855 if (!page) 1856 continue; 1857 gather_stats(page, md, pte_dirty(*pte), 1); 1858 1859 } while (pte++, addr += PAGE_SIZE, addr != end); 1860 pte_unmap_unlock(orig_pte, ptl); 1861 cond_resched(); 1862 return 0; 1863 } 1864 #ifdef CONFIG_HUGETLB_PAGE 1865 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask, 1866 unsigned long addr, unsigned long end, struct mm_walk *walk) 1867 { 1868 pte_t huge_pte = huge_ptep_get(pte); 1869 struct numa_maps *md; 1870 struct page *page; 1871 1872 if (!pte_present(huge_pte)) 1873 return 0; 1874 1875 page = pte_page(huge_pte); 1876 if (!page) 1877 return 0; 1878 1879 md = walk->private; 1880 gather_stats(page, md, pte_dirty(huge_pte), 1); 1881 return 0; 1882 } 1883 1884 #else 1885 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask, 1886 unsigned long addr, unsigned long end, struct mm_walk *walk) 1887 { 1888 return 0; 1889 } 1890 #endif 1891 1892 static const struct mm_walk_ops show_numa_ops = { 1893 .hugetlb_entry = gather_hugetlb_stats, 1894 .pmd_entry = gather_pte_stats, 1895 }; 1896 1897 /* 1898 * Display pages allocated per node and memory policy via /proc. 1899 */ 1900 static int show_numa_map(struct seq_file *m, void *v) 1901 { 1902 struct numa_maps_private *numa_priv = m->private; 1903 struct proc_maps_private *proc_priv = &numa_priv->proc_maps; 1904 struct vm_area_struct *vma = v; 1905 struct numa_maps *md = &numa_priv->md; 1906 struct file *file = vma->vm_file; 1907 struct mm_struct *mm = vma->vm_mm; 1908 struct mempolicy *pol; 1909 char buffer[64]; 1910 int nid; 1911 1912 if (!mm) 1913 return 0; 1914 1915 /* Ensure we start with an empty set of numa_maps statistics. */ 1916 memset(md, 0, sizeof(*md)); 1917 1918 pol = __get_vma_policy(vma, vma->vm_start); 1919 if (pol) { 1920 mpol_to_str(buffer, sizeof(buffer), pol); 1921 mpol_cond_put(pol); 1922 } else { 1923 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy); 1924 } 1925 1926 seq_printf(m, "%08lx %s", vma->vm_start, buffer); 1927 1928 if (file) { 1929 seq_puts(m, " file="); 1930 seq_file_path(m, file, "\n\t= "); 1931 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) { 1932 seq_puts(m, " heap"); 1933 } else if (is_stack(vma)) { 1934 seq_puts(m, " stack"); 1935 } 1936 1937 if (is_vm_hugetlb_page(vma)) 1938 seq_puts(m, " huge"); 1939 1940 /* mmap_lock is held by m_start */ 1941 walk_page_vma(vma, &show_numa_ops, md); 1942 1943 if (!md->pages) 1944 goto out; 1945 1946 if (md->anon) 1947 seq_printf(m, " anon=%lu", md->anon); 1948 1949 if (md->dirty) 1950 seq_printf(m, " dirty=%lu", md->dirty); 1951 1952 if (md->pages != md->anon && md->pages != md->dirty) 1953 seq_printf(m, " mapped=%lu", md->pages); 1954 1955 if (md->mapcount_max > 1) 1956 seq_printf(m, " mapmax=%lu", md->mapcount_max); 1957 1958 if (md->swapcache) 1959 seq_printf(m, " swapcache=%lu", md->swapcache); 1960 1961 if (md->active < md->pages && !is_vm_hugetlb_page(vma)) 1962 seq_printf(m, " active=%lu", md->active); 1963 1964 if (md->writeback) 1965 seq_printf(m, " writeback=%lu", md->writeback); 1966 1967 for_each_node_state(nid, N_MEMORY) 1968 if (md->node[nid]) 1969 seq_printf(m, " N%d=%lu", nid, md->node[nid]); 1970 1971 seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10); 1972 out: 1973 seq_putc(m, '\n'); 1974 return 0; 1975 } 1976 1977 static const struct seq_operations proc_pid_numa_maps_op = { 1978 .start = m_start, 1979 .next = m_next, 1980 .stop = m_stop, 1981 .show = show_numa_map, 1982 }; 1983 1984 static int pid_numa_maps_open(struct inode *inode, struct file *file) 1985 { 1986 return proc_maps_open(inode, file, &proc_pid_numa_maps_op, 1987 sizeof(struct numa_maps_private)); 1988 } 1989 1990 const struct file_operations proc_pid_numa_maps_operations = { 1991 .open = pid_numa_maps_open, 1992 .read = seq_read, 1993 .llseek = seq_lseek, 1994 .release = proc_map_release, 1995 }; 1996 1997 #endif /* CONFIG_NUMA */ 1998