1 // SPDX-License-Identifier: GPL-2.0 2 #include <linux/pagewalk.h> 3 #include <linux/vmacache.h> 4 #include <linux/hugetlb.h> 5 #include <linux/huge_mm.h> 6 #include <linux/mount.h> 7 #include <linux/seq_file.h> 8 #include <linux/highmem.h> 9 #include <linux/ptrace.h> 10 #include <linux/slab.h> 11 #include <linux/pagemap.h> 12 #include <linux/mempolicy.h> 13 #include <linux/rmap.h> 14 #include <linux/swap.h> 15 #include <linux/sched/mm.h> 16 #include <linux/swapops.h> 17 #include <linux/mmu_notifier.h> 18 #include <linux/page_idle.h> 19 #include <linux/shmem_fs.h> 20 #include <linux/uaccess.h> 21 #include <linux/pkeys.h> 22 23 #include <asm/elf.h> 24 #include <asm/tlb.h> 25 #include <asm/tlbflush.h> 26 #include "internal.h" 27 28 #define SEQ_PUT_DEC(str, val) \ 29 seq_put_decimal_ull_width(m, str, (val) << (PAGE_SHIFT-10), 8) 30 void task_mem(struct seq_file *m, struct mm_struct *mm) 31 { 32 unsigned long text, lib, swap, anon, file, shmem; 33 unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss; 34 35 anon = get_mm_counter(mm, MM_ANONPAGES); 36 file = get_mm_counter(mm, MM_FILEPAGES); 37 shmem = get_mm_counter(mm, MM_SHMEMPAGES); 38 39 /* 40 * Note: to minimize their overhead, mm maintains hiwater_vm and 41 * hiwater_rss only when about to *lower* total_vm or rss. Any 42 * collector of these hiwater stats must therefore get total_vm 43 * and rss too, which will usually be the higher. Barriers? not 44 * worth the effort, such snapshots can always be inconsistent. 45 */ 46 hiwater_vm = total_vm = mm->total_vm; 47 if (hiwater_vm < mm->hiwater_vm) 48 hiwater_vm = mm->hiwater_vm; 49 hiwater_rss = total_rss = anon + file + shmem; 50 if (hiwater_rss < mm->hiwater_rss) 51 hiwater_rss = mm->hiwater_rss; 52 53 /* split executable areas between text and lib */ 54 text = PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK); 55 text = min(text, mm->exec_vm << PAGE_SHIFT); 56 lib = (mm->exec_vm << PAGE_SHIFT) - text; 57 58 swap = get_mm_counter(mm, MM_SWAPENTS); 59 SEQ_PUT_DEC("VmPeak:\t", hiwater_vm); 60 SEQ_PUT_DEC(" kB\nVmSize:\t", total_vm); 61 SEQ_PUT_DEC(" kB\nVmLck:\t", mm->locked_vm); 62 SEQ_PUT_DEC(" kB\nVmPin:\t", atomic64_read(&mm->pinned_vm)); 63 SEQ_PUT_DEC(" kB\nVmHWM:\t", hiwater_rss); 64 SEQ_PUT_DEC(" kB\nVmRSS:\t", total_rss); 65 SEQ_PUT_DEC(" kB\nRssAnon:\t", anon); 66 SEQ_PUT_DEC(" kB\nRssFile:\t", file); 67 SEQ_PUT_DEC(" kB\nRssShmem:\t", shmem); 68 SEQ_PUT_DEC(" kB\nVmData:\t", mm->data_vm); 69 SEQ_PUT_DEC(" kB\nVmStk:\t", mm->stack_vm); 70 seq_put_decimal_ull_width(m, 71 " kB\nVmExe:\t", text >> 10, 8); 72 seq_put_decimal_ull_width(m, 73 " kB\nVmLib:\t", lib >> 10, 8); 74 seq_put_decimal_ull_width(m, 75 " kB\nVmPTE:\t", mm_pgtables_bytes(mm) >> 10, 8); 76 SEQ_PUT_DEC(" kB\nVmSwap:\t", swap); 77 seq_puts(m, " kB\n"); 78 hugetlb_report_usage(m, mm); 79 } 80 #undef SEQ_PUT_DEC 81 82 unsigned long task_vsize(struct mm_struct *mm) 83 { 84 return PAGE_SIZE * mm->total_vm; 85 } 86 87 unsigned long task_statm(struct mm_struct *mm, 88 unsigned long *shared, unsigned long *text, 89 unsigned long *data, unsigned long *resident) 90 { 91 *shared = get_mm_counter(mm, MM_FILEPAGES) + 92 get_mm_counter(mm, MM_SHMEMPAGES); 93 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) 94 >> PAGE_SHIFT; 95 *data = mm->data_vm + mm->stack_vm; 96 *resident = *shared + get_mm_counter(mm, MM_ANONPAGES); 97 return mm->total_vm; 98 } 99 100 #ifdef CONFIG_NUMA 101 /* 102 * Save get_task_policy() for show_numa_map(). 103 */ 104 static void hold_task_mempolicy(struct proc_maps_private *priv) 105 { 106 struct task_struct *task = priv->task; 107 108 task_lock(task); 109 priv->task_mempolicy = get_task_policy(task); 110 mpol_get(priv->task_mempolicy); 111 task_unlock(task); 112 } 113 static void release_task_mempolicy(struct proc_maps_private *priv) 114 { 115 mpol_put(priv->task_mempolicy); 116 } 117 #else 118 static void hold_task_mempolicy(struct proc_maps_private *priv) 119 { 120 } 121 static void release_task_mempolicy(struct proc_maps_private *priv) 122 { 123 } 124 #endif 125 126 static void *m_start(struct seq_file *m, loff_t *ppos) 127 { 128 struct proc_maps_private *priv = m->private; 129 unsigned long last_addr = *ppos; 130 struct mm_struct *mm; 131 struct vm_area_struct *vma; 132 133 /* See m_next(). Zero at the start or after lseek. */ 134 if (last_addr == -1UL) 135 return NULL; 136 137 priv->task = get_proc_task(priv->inode); 138 if (!priv->task) 139 return ERR_PTR(-ESRCH); 140 141 mm = priv->mm; 142 if (!mm || !mmget_not_zero(mm)) { 143 put_task_struct(priv->task); 144 priv->task = NULL; 145 return NULL; 146 } 147 148 if (mmap_read_lock_killable(mm)) { 149 mmput(mm); 150 put_task_struct(priv->task); 151 priv->task = NULL; 152 return ERR_PTR(-EINTR); 153 } 154 155 hold_task_mempolicy(priv); 156 priv->tail_vma = get_gate_vma(mm); 157 158 vma = find_vma(mm, last_addr); 159 if (vma) 160 return vma; 161 162 return priv->tail_vma; 163 } 164 165 static void *m_next(struct seq_file *m, void *v, loff_t *ppos) 166 { 167 struct proc_maps_private *priv = m->private; 168 struct vm_area_struct *next, *vma = v; 169 170 if (vma == priv->tail_vma) 171 next = NULL; 172 else if (vma->vm_next) 173 next = vma->vm_next; 174 else 175 next = priv->tail_vma; 176 177 *ppos = next ? next->vm_start : -1UL; 178 179 return next; 180 } 181 182 static void m_stop(struct seq_file *m, void *v) 183 { 184 struct proc_maps_private *priv = m->private; 185 struct mm_struct *mm = priv->mm; 186 187 if (!priv->task) 188 return; 189 190 release_task_mempolicy(priv); 191 mmap_read_unlock(mm); 192 mmput(mm); 193 put_task_struct(priv->task); 194 priv->task = NULL; 195 } 196 197 static int proc_maps_open(struct inode *inode, struct file *file, 198 const struct seq_operations *ops, int psize) 199 { 200 struct proc_maps_private *priv = __seq_open_private(file, ops, psize); 201 202 if (!priv) 203 return -ENOMEM; 204 205 priv->inode = inode; 206 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ); 207 if (IS_ERR(priv->mm)) { 208 int err = PTR_ERR(priv->mm); 209 210 seq_release_private(inode, file); 211 return err; 212 } 213 214 return 0; 215 } 216 217 static int proc_map_release(struct inode *inode, struct file *file) 218 { 219 struct seq_file *seq = file->private_data; 220 struct proc_maps_private *priv = seq->private; 221 222 if (priv->mm) 223 mmdrop(priv->mm); 224 225 return seq_release_private(inode, file); 226 } 227 228 static int do_maps_open(struct inode *inode, struct file *file, 229 const struct seq_operations *ops) 230 { 231 return proc_maps_open(inode, file, ops, 232 sizeof(struct proc_maps_private)); 233 } 234 235 /* 236 * Indicate if the VMA is a stack for the given task; for 237 * /proc/PID/maps that is the stack of the main task. 238 */ 239 static int is_stack(struct vm_area_struct *vma) 240 { 241 /* 242 * We make no effort to guess what a given thread considers to be 243 * its "stack". It's not even well-defined for programs written 244 * languages like Go. 245 */ 246 return vma->vm_start <= vma->vm_mm->start_stack && 247 vma->vm_end >= vma->vm_mm->start_stack; 248 } 249 250 static void show_vma_header_prefix(struct seq_file *m, 251 unsigned long start, unsigned long end, 252 vm_flags_t flags, unsigned long long pgoff, 253 dev_t dev, unsigned long ino) 254 { 255 seq_setwidth(m, 25 + sizeof(void *) * 6 - 1); 256 seq_put_hex_ll(m, NULL, start, 8); 257 seq_put_hex_ll(m, "-", end, 8); 258 seq_putc(m, ' '); 259 seq_putc(m, flags & VM_READ ? 'r' : '-'); 260 seq_putc(m, flags & VM_WRITE ? 'w' : '-'); 261 seq_putc(m, flags & VM_EXEC ? 'x' : '-'); 262 seq_putc(m, flags & VM_MAYSHARE ? 's' : 'p'); 263 seq_put_hex_ll(m, " ", pgoff, 8); 264 seq_put_hex_ll(m, " ", MAJOR(dev), 2); 265 seq_put_hex_ll(m, ":", MINOR(dev), 2); 266 seq_put_decimal_ull(m, " ", ino); 267 seq_putc(m, ' '); 268 } 269 270 static void 271 show_map_vma(struct seq_file *m, struct vm_area_struct *vma) 272 { 273 struct mm_struct *mm = vma->vm_mm; 274 struct file *file = vma->vm_file; 275 vm_flags_t flags = vma->vm_flags; 276 unsigned long ino = 0; 277 unsigned long long pgoff = 0; 278 unsigned long start, end; 279 dev_t dev = 0; 280 const char *name = NULL; 281 282 if (file) { 283 struct inode *inode = file_inode(vma->vm_file); 284 dev = inode->i_sb->s_dev; 285 ino = inode->i_ino; 286 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT; 287 } 288 289 start = vma->vm_start; 290 end = vma->vm_end; 291 show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino); 292 293 /* 294 * Print the dentry name for named mappings, and a 295 * special [heap] marker for the heap: 296 */ 297 if (file) { 298 seq_pad(m, ' '); 299 seq_file_path(m, file, "\n"); 300 goto done; 301 } 302 303 if (vma->vm_ops && vma->vm_ops->name) { 304 name = vma->vm_ops->name(vma); 305 if (name) 306 goto done; 307 } 308 309 name = arch_vma_name(vma); 310 if (!name) { 311 if (!mm) { 312 name = "[vdso]"; 313 goto done; 314 } 315 316 if (vma->vm_start <= mm->brk && 317 vma->vm_end >= mm->start_brk) { 318 name = "[heap]"; 319 goto done; 320 } 321 322 if (is_stack(vma)) 323 name = "[stack]"; 324 } 325 326 done: 327 if (name) { 328 seq_pad(m, ' '); 329 seq_puts(m, name); 330 } 331 seq_putc(m, '\n'); 332 } 333 334 static int show_map(struct seq_file *m, void *v) 335 { 336 show_map_vma(m, v); 337 return 0; 338 } 339 340 static const struct seq_operations proc_pid_maps_op = { 341 .start = m_start, 342 .next = m_next, 343 .stop = m_stop, 344 .show = show_map 345 }; 346 347 static int pid_maps_open(struct inode *inode, struct file *file) 348 { 349 return do_maps_open(inode, file, &proc_pid_maps_op); 350 } 351 352 const struct file_operations proc_pid_maps_operations = { 353 .open = pid_maps_open, 354 .read = seq_read, 355 .llseek = seq_lseek, 356 .release = proc_map_release, 357 }; 358 359 /* 360 * Proportional Set Size(PSS): my share of RSS. 361 * 362 * PSS of a process is the count of pages it has in memory, where each 363 * page is divided by the number of processes sharing it. So if a 364 * process has 1000 pages all to itself, and 1000 shared with one other 365 * process, its PSS will be 1500. 366 * 367 * To keep (accumulated) division errors low, we adopt a 64bit 368 * fixed-point pss counter to minimize division errors. So (pss >> 369 * PSS_SHIFT) would be the real byte count. 370 * 371 * A shift of 12 before division means (assuming 4K page size): 372 * - 1M 3-user-pages add up to 8KB errors; 373 * - supports mapcount up to 2^24, or 16M; 374 * - supports PSS up to 2^52 bytes, or 4PB. 375 */ 376 #define PSS_SHIFT 12 377 378 #ifdef CONFIG_PROC_PAGE_MONITOR 379 struct mem_size_stats { 380 unsigned long resident; 381 unsigned long shared_clean; 382 unsigned long shared_dirty; 383 unsigned long private_clean; 384 unsigned long private_dirty; 385 unsigned long referenced; 386 unsigned long anonymous; 387 unsigned long lazyfree; 388 unsigned long anonymous_thp; 389 unsigned long shmem_thp; 390 unsigned long file_thp; 391 unsigned long swap; 392 unsigned long shared_hugetlb; 393 unsigned long private_hugetlb; 394 u64 pss; 395 u64 pss_anon; 396 u64 pss_file; 397 u64 pss_shmem; 398 u64 pss_locked; 399 u64 swap_pss; 400 bool check_shmem_swap; 401 }; 402 403 static void smaps_page_accumulate(struct mem_size_stats *mss, 404 struct page *page, unsigned long size, unsigned long pss, 405 bool dirty, bool locked, bool private) 406 { 407 mss->pss += pss; 408 409 if (PageAnon(page)) 410 mss->pss_anon += pss; 411 else if (PageSwapBacked(page)) 412 mss->pss_shmem += pss; 413 else 414 mss->pss_file += pss; 415 416 if (locked) 417 mss->pss_locked += pss; 418 419 if (dirty || PageDirty(page)) { 420 if (private) 421 mss->private_dirty += size; 422 else 423 mss->shared_dirty += size; 424 } else { 425 if (private) 426 mss->private_clean += size; 427 else 428 mss->shared_clean += size; 429 } 430 } 431 432 static void smaps_account(struct mem_size_stats *mss, struct page *page, 433 bool compound, bool young, bool dirty, bool locked) 434 { 435 int i, nr = compound ? compound_nr(page) : 1; 436 unsigned long size = nr * PAGE_SIZE; 437 438 /* 439 * First accumulate quantities that depend only on |size| and the type 440 * of the compound page. 441 */ 442 if (PageAnon(page)) { 443 mss->anonymous += size; 444 if (!PageSwapBacked(page) && !dirty && !PageDirty(page)) 445 mss->lazyfree += size; 446 } 447 448 mss->resident += size; 449 /* Accumulate the size in pages that have been accessed. */ 450 if (young || page_is_young(page) || PageReferenced(page)) 451 mss->referenced += size; 452 453 /* 454 * Then accumulate quantities that may depend on sharing, or that may 455 * differ page-by-page. 456 * 457 * page_count(page) == 1 guarantees the page is mapped exactly once. 458 * If any subpage of the compound page mapped with PTE it would elevate 459 * page_count(). 460 */ 461 if (page_count(page) == 1) { 462 smaps_page_accumulate(mss, page, size, size << PSS_SHIFT, dirty, 463 locked, true); 464 return; 465 } 466 for (i = 0; i < nr; i++, page++) { 467 int mapcount = page_mapcount(page); 468 unsigned long pss = PAGE_SIZE << PSS_SHIFT; 469 if (mapcount >= 2) 470 pss /= mapcount; 471 smaps_page_accumulate(mss, page, PAGE_SIZE, pss, dirty, locked, 472 mapcount < 2); 473 } 474 } 475 476 #ifdef CONFIG_SHMEM 477 static int smaps_pte_hole(unsigned long addr, unsigned long end, 478 __always_unused int depth, struct mm_walk *walk) 479 { 480 struct mem_size_stats *mss = walk->private; 481 482 mss->swap += shmem_partial_swap_usage( 483 walk->vma->vm_file->f_mapping, addr, end); 484 485 return 0; 486 } 487 #else 488 #define smaps_pte_hole NULL 489 #endif /* CONFIG_SHMEM */ 490 491 static void smaps_pte_entry(pte_t *pte, unsigned long addr, 492 struct mm_walk *walk) 493 { 494 struct mem_size_stats *mss = walk->private; 495 struct vm_area_struct *vma = walk->vma; 496 bool locked = !!(vma->vm_flags & VM_LOCKED); 497 struct page *page = NULL; 498 499 if (pte_present(*pte)) { 500 page = vm_normal_page(vma, addr, *pte); 501 } else if (is_swap_pte(*pte)) { 502 swp_entry_t swpent = pte_to_swp_entry(*pte); 503 504 if (!non_swap_entry(swpent)) { 505 int mapcount; 506 507 mss->swap += PAGE_SIZE; 508 mapcount = swp_swapcount(swpent); 509 if (mapcount >= 2) { 510 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT; 511 512 do_div(pss_delta, mapcount); 513 mss->swap_pss += pss_delta; 514 } else { 515 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT; 516 } 517 } else if (is_pfn_swap_entry(swpent)) 518 page = pfn_swap_entry_to_page(swpent); 519 } else if (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap 520 && pte_none(*pte))) { 521 page = xa_load(&vma->vm_file->f_mapping->i_pages, 522 linear_page_index(vma, addr)); 523 if (xa_is_value(page)) 524 mss->swap += PAGE_SIZE; 525 return; 526 } 527 528 if (!page) 529 return; 530 531 smaps_account(mss, page, false, pte_young(*pte), pte_dirty(*pte), locked); 532 } 533 534 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 535 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr, 536 struct mm_walk *walk) 537 { 538 struct mem_size_stats *mss = walk->private; 539 struct vm_area_struct *vma = walk->vma; 540 bool locked = !!(vma->vm_flags & VM_LOCKED); 541 struct page *page = NULL; 542 543 if (pmd_present(*pmd)) { 544 /* FOLL_DUMP will return -EFAULT on huge zero page */ 545 page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP); 546 } else if (unlikely(thp_migration_supported() && is_swap_pmd(*pmd))) { 547 swp_entry_t entry = pmd_to_swp_entry(*pmd); 548 549 if (is_migration_entry(entry)) 550 page = pfn_swap_entry_to_page(entry); 551 } 552 if (IS_ERR_OR_NULL(page)) 553 return; 554 if (PageAnon(page)) 555 mss->anonymous_thp += HPAGE_PMD_SIZE; 556 else if (PageSwapBacked(page)) 557 mss->shmem_thp += HPAGE_PMD_SIZE; 558 else if (is_zone_device_page(page)) 559 /* pass */; 560 else 561 mss->file_thp += HPAGE_PMD_SIZE; 562 smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd), locked); 563 } 564 #else 565 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr, 566 struct mm_walk *walk) 567 { 568 } 569 #endif 570 571 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, 572 struct mm_walk *walk) 573 { 574 struct vm_area_struct *vma = walk->vma; 575 pte_t *pte; 576 spinlock_t *ptl; 577 578 ptl = pmd_trans_huge_lock(pmd, vma); 579 if (ptl) { 580 smaps_pmd_entry(pmd, addr, walk); 581 spin_unlock(ptl); 582 goto out; 583 } 584 585 if (pmd_trans_unstable(pmd)) 586 goto out; 587 /* 588 * The mmap_lock held all the way back in m_start() is what 589 * keeps khugepaged out of here and from collapsing things 590 * in here. 591 */ 592 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); 593 for (; addr != end; pte++, addr += PAGE_SIZE) 594 smaps_pte_entry(pte, addr, walk); 595 pte_unmap_unlock(pte - 1, ptl); 596 out: 597 cond_resched(); 598 return 0; 599 } 600 601 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma) 602 { 603 /* 604 * Don't forget to update Documentation/ on changes. 605 */ 606 static const char mnemonics[BITS_PER_LONG][2] = { 607 /* 608 * In case if we meet a flag we don't know about. 609 */ 610 [0 ... (BITS_PER_LONG-1)] = "??", 611 612 [ilog2(VM_READ)] = "rd", 613 [ilog2(VM_WRITE)] = "wr", 614 [ilog2(VM_EXEC)] = "ex", 615 [ilog2(VM_SHARED)] = "sh", 616 [ilog2(VM_MAYREAD)] = "mr", 617 [ilog2(VM_MAYWRITE)] = "mw", 618 [ilog2(VM_MAYEXEC)] = "me", 619 [ilog2(VM_MAYSHARE)] = "ms", 620 [ilog2(VM_GROWSDOWN)] = "gd", 621 [ilog2(VM_PFNMAP)] = "pf", 622 [ilog2(VM_DENYWRITE)] = "dw", 623 [ilog2(VM_LOCKED)] = "lo", 624 [ilog2(VM_IO)] = "io", 625 [ilog2(VM_SEQ_READ)] = "sr", 626 [ilog2(VM_RAND_READ)] = "rr", 627 [ilog2(VM_DONTCOPY)] = "dc", 628 [ilog2(VM_DONTEXPAND)] = "de", 629 [ilog2(VM_ACCOUNT)] = "ac", 630 [ilog2(VM_NORESERVE)] = "nr", 631 [ilog2(VM_HUGETLB)] = "ht", 632 [ilog2(VM_SYNC)] = "sf", 633 [ilog2(VM_ARCH_1)] = "ar", 634 [ilog2(VM_WIPEONFORK)] = "wf", 635 [ilog2(VM_DONTDUMP)] = "dd", 636 #ifdef CONFIG_ARM64_BTI 637 [ilog2(VM_ARM64_BTI)] = "bt", 638 #endif 639 #ifdef CONFIG_MEM_SOFT_DIRTY 640 [ilog2(VM_SOFTDIRTY)] = "sd", 641 #endif 642 [ilog2(VM_MIXEDMAP)] = "mm", 643 [ilog2(VM_HUGEPAGE)] = "hg", 644 [ilog2(VM_NOHUGEPAGE)] = "nh", 645 [ilog2(VM_MERGEABLE)] = "mg", 646 [ilog2(VM_UFFD_MISSING)]= "um", 647 [ilog2(VM_UFFD_WP)] = "uw", 648 #ifdef CONFIG_ARM64_MTE 649 [ilog2(VM_MTE)] = "mt", 650 [ilog2(VM_MTE_ALLOWED)] = "", 651 #endif 652 #ifdef CONFIG_ARCH_HAS_PKEYS 653 /* These come out via ProtectionKey: */ 654 [ilog2(VM_PKEY_BIT0)] = "", 655 [ilog2(VM_PKEY_BIT1)] = "", 656 [ilog2(VM_PKEY_BIT2)] = "", 657 [ilog2(VM_PKEY_BIT3)] = "", 658 #if VM_PKEY_BIT4 659 [ilog2(VM_PKEY_BIT4)] = "", 660 #endif 661 #endif /* CONFIG_ARCH_HAS_PKEYS */ 662 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_MINOR 663 [ilog2(VM_UFFD_MINOR)] = "ui", 664 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_MINOR */ 665 }; 666 size_t i; 667 668 seq_puts(m, "VmFlags: "); 669 for (i = 0; i < BITS_PER_LONG; i++) { 670 if (!mnemonics[i][0]) 671 continue; 672 if (vma->vm_flags & (1UL << i)) { 673 seq_putc(m, mnemonics[i][0]); 674 seq_putc(m, mnemonics[i][1]); 675 seq_putc(m, ' '); 676 } 677 } 678 seq_putc(m, '\n'); 679 } 680 681 #ifdef CONFIG_HUGETLB_PAGE 682 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask, 683 unsigned long addr, unsigned long end, 684 struct mm_walk *walk) 685 { 686 struct mem_size_stats *mss = walk->private; 687 struct vm_area_struct *vma = walk->vma; 688 struct page *page = NULL; 689 690 if (pte_present(*pte)) { 691 page = vm_normal_page(vma, addr, *pte); 692 } else if (is_swap_pte(*pte)) { 693 swp_entry_t swpent = pte_to_swp_entry(*pte); 694 695 if (is_pfn_swap_entry(swpent)) 696 page = pfn_swap_entry_to_page(swpent); 697 } 698 if (page) { 699 int mapcount = page_mapcount(page); 700 701 if (mapcount >= 2) 702 mss->shared_hugetlb += huge_page_size(hstate_vma(vma)); 703 else 704 mss->private_hugetlb += huge_page_size(hstate_vma(vma)); 705 } 706 return 0; 707 } 708 #else 709 #define smaps_hugetlb_range NULL 710 #endif /* HUGETLB_PAGE */ 711 712 static const struct mm_walk_ops smaps_walk_ops = { 713 .pmd_entry = smaps_pte_range, 714 .hugetlb_entry = smaps_hugetlb_range, 715 }; 716 717 static const struct mm_walk_ops smaps_shmem_walk_ops = { 718 .pmd_entry = smaps_pte_range, 719 .hugetlb_entry = smaps_hugetlb_range, 720 .pte_hole = smaps_pte_hole, 721 }; 722 723 /* 724 * Gather mem stats from @vma with the indicated beginning 725 * address @start, and keep them in @mss. 726 * 727 * Use vm_start of @vma as the beginning address if @start is 0. 728 */ 729 static void smap_gather_stats(struct vm_area_struct *vma, 730 struct mem_size_stats *mss, unsigned long start) 731 { 732 const struct mm_walk_ops *ops = &smaps_walk_ops; 733 734 /* Invalid start */ 735 if (start >= vma->vm_end) 736 return; 737 738 #ifdef CONFIG_SHMEM 739 /* In case of smaps_rollup, reset the value from previous vma */ 740 mss->check_shmem_swap = false; 741 if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) { 742 /* 743 * For shared or readonly shmem mappings we know that all 744 * swapped out pages belong to the shmem object, and we can 745 * obtain the swap value much more efficiently. For private 746 * writable mappings, we might have COW pages that are 747 * not affected by the parent swapped out pages of the shmem 748 * object, so we have to distinguish them during the page walk. 749 * Unless we know that the shmem object (or the part mapped by 750 * our VMA) has no swapped out pages at all. 751 */ 752 unsigned long shmem_swapped = shmem_swap_usage(vma); 753 754 if (!start && (!shmem_swapped || (vma->vm_flags & VM_SHARED) || 755 !(vma->vm_flags & VM_WRITE))) { 756 mss->swap += shmem_swapped; 757 } else { 758 mss->check_shmem_swap = true; 759 ops = &smaps_shmem_walk_ops; 760 } 761 } 762 #endif 763 /* mmap_lock is held in m_start */ 764 if (!start) 765 walk_page_vma(vma, ops, mss); 766 else 767 walk_page_range(vma->vm_mm, start, vma->vm_end, ops, mss); 768 } 769 770 #define SEQ_PUT_DEC(str, val) \ 771 seq_put_decimal_ull_width(m, str, (val) >> 10, 8) 772 773 /* Show the contents common for smaps and smaps_rollup */ 774 static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss, 775 bool rollup_mode) 776 { 777 SEQ_PUT_DEC("Rss: ", mss->resident); 778 SEQ_PUT_DEC(" kB\nPss: ", mss->pss >> PSS_SHIFT); 779 if (rollup_mode) { 780 /* 781 * These are meaningful only for smaps_rollup, otherwise two of 782 * them are zero, and the other one is the same as Pss. 783 */ 784 SEQ_PUT_DEC(" kB\nPss_Anon: ", 785 mss->pss_anon >> PSS_SHIFT); 786 SEQ_PUT_DEC(" kB\nPss_File: ", 787 mss->pss_file >> PSS_SHIFT); 788 SEQ_PUT_DEC(" kB\nPss_Shmem: ", 789 mss->pss_shmem >> PSS_SHIFT); 790 } 791 SEQ_PUT_DEC(" kB\nShared_Clean: ", mss->shared_clean); 792 SEQ_PUT_DEC(" kB\nShared_Dirty: ", mss->shared_dirty); 793 SEQ_PUT_DEC(" kB\nPrivate_Clean: ", mss->private_clean); 794 SEQ_PUT_DEC(" kB\nPrivate_Dirty: ", mss->private_dirty); 795 SEQ_PUT_DEC(" kB\nReferenced: ", mss->referenced); 796 SEQ_PUT_DEC(" kB\nAnonymous: ", mss->anonymous); 797 SEQ_PUT_DEC(" kB\nLazyFree: ", mss->lazyfree); 798 SEQ_PUT_DEC(" kB\nAnonHugePages: ", mss->anonymous_thp); 799 SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp); 800 SEQ_PUT_DEC(" kB\nFilePmdMapped: ", mss->file_thp); 801 SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb); 802 seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ", 803 mss->private_hugetlb >> 10, 7); 804 SEQ_PUT_DEC(" kB\nSwap: ", mss->swap); 805 SEQ_PUT_DEC(" kB\nSwapPss: ", 806 mss->swap_pss >> PSS_SHIFT); 807 SEQ_PUT_DEC(" kB\nLocked: ", 808 mss->pss_locked >> PSS_SHIFT); 809 seq_puts(m, " kB\n"); 810 } 811 812 static int show_smap(struct seq_file *m, void *v) 813 { 814 struct vm_area_struct *vma = v; 815 struct mem_size_stats mss; 816 817 memset(&mss, 0, sizeof(mss)); 818 819 smap_gather_stats(vma, &mss, 0); 820 821 show_map_vma(m, vma); 822 823 SEQ_PUT_DEC("Size: ", vma->vm_end - vma->vm_start); 824 SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma)); 825 SEQ_PUT_DEC(" kB\nMMUPageSize: ", vma_mmu_pagesize(vma)); 826 seq_puts(m, " kB\n"); 827 828 __show_smap(m, &mss, false); 829 830 seq_printf(m, "THPeligible: %d\n", 831 transparent_hugepage_active(vma)); 832 833 if (arch_pkeys_enabled()) 834 seq_printf(m, "ProtectionKey: %8u\n", vma_pkey(vma)); 835 show_smap_vma_flags(m, vma); 836 837 return 0; 838 } 839 840 static int show_smaps_rollup(struct seq_file *m, void *v) 841 { 842 struct proc_maps_private *priv = m->private; 843 struct mem_size_stats mss; 844 struct mm_struct *mm; 845 struct vm_area_struct *vma; 846 unsigned long last_vma_end = 0; 847 int ret = 0; 848 849 priv->task = get_proc_task(priv->inode); 850 if (!priv->task) 851 return -ESRCH; 852 853 mm = priv->mm; 854 if (!mm || !mmget_not_zero(mm)) { 855 ret = -ESRCH; 856 goto out_put_task; 857 } 858 859 memset(&mss, 0, sizeof(mss)); 860 861 ret = mmap_read_lock_killable(mm); 862 if (ret) 863 goto out_put_mm; 864 865 hold_task_mempolicy(priv); 866 867 for (vma = priv->mm->mmap; vma;) { 868 smap_gather_stats(vma, &mss, 0); 869 last_vma_end = vma->vm_end; 870 871 /* 872 * Release mmap_lock temporarily if someone wants to 873 * access it for write request. 874 */ 875 if (mmap_lock_is_contended(mm)) { 876 mmap_read_unlock(mm); 877 ret = mmap_read_lock_killable(mm); 878 if (ret) { 879 release_task_mempolicy(priv); 880 goto out_put_mm; 881 } 882 883 /* 884 * After dropping the lock, there are four cases to 885 * consider. See the following example for explanation. 886 * 887 * +------+------+-----------+ 888 * | VMA1 | VMA2 | VMA3 | 889 * +------+------+-----------+ 890 * | | | | 891 * 4k 8k 16k 400k 892 * 893 * Suppose we drop the lock after reading VMA2 due to 894 * contention, then we get: 895 * 896 * last_vma_end = 16k 897 * 898 * 1) VMA2 is freed, but VMA3 exists: 899 * 900 * find_vma(mm, 16k - 1) will return VMA3. 901 * In this case, just continue from VMA3. 902 * 903 * 2) VMA2 still exists: 904 * 905 * find_vma(mm, 16k - 1) will return VMA2. 906 * Iterate the loop like the original one. 907 * 908 * 3) No more VMAs can be found: 909 * 910 * find_vma(mm, 16k - 1) will return NULL. 911 * No more things to do, just break. 912 * 913 * 4) (last_vma_end - 1) is the middle of a vma (VMA'): 914 * 915 * find_vma(mm, 16k - 1) will return VMA' whose range 916 * contains last_vma_end. 917 * Iterate VMA' from last_vma_end. 918 */ 919 vma = find_vma(mm, last_vma_end - 1); 920 /* Case 3 above */ 921 if (!vma) 922 break; 923 924 /* Case 1 above */ 925 if (vma->vm_start >= last_vma_end) 926 continue; 927 928 /* Case 4 above */ 929 if (vma->vm_end > last_vma_end) 930 smap_gather_stats(vma, &mss, last_vma_end); 931 } 932 /* Case 2 above */ 933 vma = vma->vm_next; 934 } 935 936 show_vma_header_prefix(m, priv->mm->mmap->vm_start, 937 last_vma_end, 0, 0, 0, 0); 938 seq_pad(m, ' '); 939 seq_puts(m, "[rollup]\n"); 940 941 __show_smap(m, &mss, true); 942 943 release_task_mempolicy(priv); 944 mmap_read_unlock(mm); 945 946 out_put_mm: 947 mmput(mm); 948 out_put_task: 949 put_task_struct(priv->task); 950 priv->task = NULL; 951 952 return ret; 953 } 954 #undef SEQ_PUT_DEC 955 956 static const struct seq_operations proc_pid_smaps_op = { 957 .start = m_start, 958 .next = m_next, 959 .stop = m_stop, 960 .show = show_smap 961 }; 962 963 static int pid_smaps_open(struct inode *inode, struct file *file) 964 { 965 return do_maps_open(inode, file, &proc_pid_smaps_op); 966 } 967 968 static int smaps_rollup_open(struct inode *inode, struct file *file) 969 { 970 int ret; 971 struct proc_maps_private *priv; 972 973 priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT); 974 if (!priv) 975 return -ENOMEM; 976 977 ret = single_open(file, show_smaps_rollup, priv); 978 if (ret) 979 goto out_free; 980 981 priv->inode = inode; 982 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ); 983 if (IS_ERR(priv->mm)) { 984 ret = PTR_ERR(priv->mm); 985 986 single_release(inode, file); 987 goto out_free; 988 } 989 990 return 0; 991 992 out_free: 993 kfree(priv); 994 return ret; 995 } 996 997 static int smaps_rollup_release(struct inode *inode, struct file *file) 998 { 999 struct seq_file *seq = file->private_data; 1000 struct proc_maps_private *priv = seq->private; 1001 1002 if (priv->mm) 1003 mmdrop(priv->mm); 1004 1005 kfree(priv); 1006 return single_release(inode, file); 1007 } 1008 1009 const struct file_operations proc_pid_smaps_operations = { 1010 .open = pid_smaps_open, 1011 .read = seq_read, 1012 .llseek = seq_lseek, 1013 .release = proc_map_release, 1014 }; 1015 1016 const struct file_operations proc_pid_smaps_rollup_operations = { 1017 .open = smaps_rollup_open, 1018 .read = seq_read, 1019 .llseek = seq_lseek, 1020 .release = smaps_rollup_release, 1021 }; 1022 1023 enum clear_refs_types { 1024 CLEAR_REFS_ALL = 1, 1025 CLEAR_REFS_ANON, 1026 CLEAR_REFS_MAPPED, 1027 CLEAR_REFS_SOFT_DIRTY, 1028 CLEAR_REFS_MM_HIWATER_RSS, 1029 CLEAR_REFS_LAST, 1030 }; 1031 1032 struct clear_refs_private { 1033 enum clear_refs_types type; 1034 }; 1035 1036 #ifdef CONFIG_MEM_SOFT_DIRTY 1037 1038 static inline bool pte_is_pinned(struct vm_area_struct *vma, unsigned long addr, pte_t pte) 1039 { 1040 struct page *page; 1041 1042 if (!pte_write(pte)) 1043 return false; 1044 if (!is_cow_mapping(vma->vm_flags)) 1045 return false; 1046 if (likely(!test_bit(MMF_HAS_PINNED, &vma->vm_mm->flags))) 1047 return false; 1048 page = vm_normal_page(vma, addr, pte); 1049 if (!page) 1050 return false; 1051 return page_maybe_dma_pinned(page); 1052 } 1053 1054 static inline void clear_soft_dirty(struct vm_area_struct *vma, 1055 unsigned long addr, pte_t *pte) 1056 { 1057 /* 1058 * The soft-dirty tracker uses #PF-s to catch writes 1059 * to pages, so write-protect the pte as well. See the 1060 * Documentation/admin-guide/mm/soft-dirty.rst for full description 1061 * of how soft-dirty works. 1062 */ 1063 pte_t ptent = *pte; 1064 1065 if (pte_present(ptent)) { 1066 pte_t old_pte; 1067 1068 if (pte_is_pinned(vma, addr, ptent)) 1069 return; 1070 old_pte = ptep_modify_prot_start(vma, addr, pte); 1071 ptent = pte_wrprotect(old_pte); 1072 ptent = pte_clear_soft_dirty(ptent); 1073 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent); 1074 } else if (is_swap_pte(ptent)) { 1075 ptent = pte_swp_clear_soft_dirty(ptent); 1076 set_pte_at(vma->vm_mm, addr, pte, ptent); 1077 } 1078 } 1079 #else 1080 static inline void clear_soft_dirty(struct vm_area_struct *vma, 1081 unsigned long addr, pte_t *pte) 1082 { 1083 } 1084 #endif 1085 1086 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE) 1087 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma, 1088 unsigned long addr, pmd_t *pmdp) 1089 { 1090 pmd_t old, pmd = *pmdp; 1091 1092 if (pmd_present(pmd)) { 1093 /* See comment in change_huge_pmd() */ 1094 old = pmdp_invalidate(vma, addr, pmdp); 1095 if (pmd_dirty(old)) 1096 pmd = pmd_mkdirty(pmd); 1097 if (pmd_young(old)) 1098 pmd = pmd_mkyoung(pmd); 1099 1100 pmd = pmd_wrprotect(pmd); 1101 pmd = pmd_clear_soft_dirty(pmd); 1102 1103 set_pmd_at(vma->vm_mm, addr, pmdp, pmd); 1104 } else if (is_migration_entry(pmd_to_swp_entry(pmd))) { 1105 pmd = pmd_swp_clear_soft_dirty(pmd); 1106 set_pmd_at(vma->vm_mm, addr, pmdp, pmd); 1107 } 1108 } 1109 #else 1110 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma, 1111 unsigned long addr, pmd_t *pmdp) 1112 { 1113 } 1114 #endif 1115 1116 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr, 1117 unsigned long end, struct mm_walk *walk) 1118 { 1119 struct clear_refs_private *cp = walk->private; 1120 struct vm_area_struct *vma = walk->vma; 1121 pte_t *pte, ptent; 1122 spinlock_t *ptl; 1123 struct page *page; 1124 1125 ptl = pmd_trans_huge_lock(pmd, vma); 1126 if (ptl) { 1127 if (cp->type == CLEAR_REFS_SOFT_DIRTY) { 1128 clear_soft_dirty_pmd(vma, addr, pmd); 1129 goto out; 1130 } 1131 1132 if (!pmd_present(*pmd)) 1133 goto out; 1134 1135 page = pmd_page(*pmd); 1136 1137 /* Clear accessed and referenced bits. */ 1138 pmdp_test_and_clear_young(vma, addr, pmd); 1139 test_and_clear_page_young(page); 1140 ClearPageReferenced(page); 1141 out: 1142 spin_unlock(ptl); 1143 return 0; 1144 } 1145 1146 if (pmd_trans_unstable(pmd)) 1147 return 0; 1148 1149 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); 1150 for (; addr != end; pte++, addr += PAGE_SIZE) { 1151 ptent = *pte; 1152 1153 if (cp->type == CLEAR_REFS_SOFT_DIRTY) { 1154 clear_soft_dirty(vma, addr, pte); 1155 continue; 1156 } 1157 1158 if (!pte_present(ptent)) 1159 continue; 1160 1161 page = vm_normal_page(vma, addr, ptent); 1162 if (!page) 1163 continue; 1164 1165 /* Clear accessed and referenced bits. */ 1166 ptep_test_and_clear_young(vma, addr, pte); 1167 test_and_clear_page_young(page); 1168 ClearPageReferenced(page); 1169 } 1170 pte_unmap_unlock(pte - 1, ptl); 1171 cond_resched(); 1172 return 0; 1173 } 1174 1175 static int clear_refs_test_walk(unsigned long start, unsigned long end, 1176 struct mm_walk *walk) 1177 { 1178 struct clear_refs_private *cp = walk->private; 1179 struct vm_area_struct *vma = walk->vma; 1180 1181 if (vma->vm_flags & VM_PFNMAP) 1182 return 1; 1183 1184 /* 1185 * Writing 1 to /proc/pid/clear_refs affects all pages. 1186 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages. 1187 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages. 1188 * Writing 4 to /proc/pid/clear_refs affects all pages. 1189 */ 1190 if (cp->type == CLEAR_REFS_ANON && vma->vm_file) 1191 return 1; 1192 if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file) 1193 return 1; 1194 return 0; 1195 } 1196 1197 static const struct mm_walk_ops clear_refs_walk_ops = { 1198 .pmd_entry = clear_refs_pte_range, 1199 .test_walk = clear_refs_test_walk, 1200 }; 1201 1202 static ssize_t clear_refs_write(struct file *file, const char __user *buf, 1203 size_t count, loff_t *ppos) 1204 { 1205 struct task_struct *task; 1206 char buffer[PROC_NUMBUF]; 1207 struct mm_struct *mm; 1208 struct vm_area_struct *vma; 1209 enum clear_refs_types type; 1210 int itype; 1211 int rv; 1212 1213 memset(buffer, 0, sizeof(buffer)); 1214 if (count > sizeof(buffer) - 1) 1215 count = sizeof(buffer) - 1; 1216 if (copy_from_user(buffer, buf, count)) 1217 return -EFAULT; 1218 rv = kstrtoint(strstrip(buffer), 10, &itype); 1219 if (rv < 0) 1220 return rv; 1221 type = (enum clear_refs_types)itype; 1222 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST) 1223 return -EINVAL; 1224 1225 task = get_proc_task(file_inode(file)); 1226 if (!task) 1227 return -ESRCH; 1228 mm = get_task_mm(task); 1229 if (mm) { 1230 struct mmu_notifier_range range; 1231 struct clear_refs_private cp = { 1232 .type = type, 1233 }; 1234 1235 if (mmap_write_lock_killable(mm)) { 1236 count = -EINTR; 1237 goto out_mm; 1238 } 1239 if (type == CLEAR_REFS_MM_HIWATER_RSS) { 1240 /* 1241 * Writing 5 to /proc/pid/clear_refs resets the peak 1242 * resident set size to this mm's current rss value. 1243 */ 1244 reset_mm_hiwater_rss(mm); 1245 goto out_unlock; 1246 } 1247 1248 if (type == CLEAR_REFS_SOFT_DIRTY) { 1249 for (vma = mm->mmap; vma; vma = vma->vm_next) { 1250 if (!(vma->vm_flags & VM_SOFTDIRTY)) 1251 continue; 1252 vma->vm_flags &= ~VM_SOFTDIRTY; 1253 vma_set_page_prot(vma); 1254 } 1255 1256 inc_tlb_flush_pending(mm); 1257 mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY, 1258 0, NULL, mm, 0, -1UL); 1259 mmu_notifier_invalidate_range_start(&range); 1260 } 1261 walk_page_range(mm, 0, mm->highest_vm_end, &clear_refs_walk_ops, 1262 &cp); 1263 if (type == CLEAR_REFS_SOFT_DIRTY) { 1264 mmu_notifier_invalidate_range_end(&range); 1265 flush_tlb_mm(mm); 1266 dec_tlb_flush_pending(mm); 1267 } 1268 out_unlock: 1269 mmap_write_unlock(mm); 1270 out_mm: 1271 mmput(mm); 1272 } 1273 put_task_struct(task); 1274 1275 return count; 1276 } 1277 1278 const struct file_operations proc_clear_refs_operations = { 1279 .write = clear_refs_write, 1280 .llseek = noop_llseek, 1281 }; 1282 1283 typedef struct { 1284 u64 pme; 1285 } pagemap_entry_t; 1286 1287 struct pagemapread { 1288 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */ 1289 pagemap_entry_t *buffer; 1290 bool show_pfn; 1291 }; 1292 1293 #define PAGEMAP_WALK_SIZE (PMD_SIZE) 1294 #define PAGEMAP_WALK_MASK (PMD_MASK) 1295 1296 #define PM_ENTRY_BYTES sizeof(pagemap_entry_t) 1297 #define PM_PFRAME_BITS 55 1298 #define PM_PFRAME_MASK GENMASK_ULL(PM_PFRAME_BITS - 1, 0) 1299 #define PM_SOFT_DIRTY BIT_ULL(55) 1300 #define PM_MMAP_EXCLUSIVE BIT_ULL(56) 1301 #define PM_UFFD_WP BIT_ULL(57) 1302 #define PM_FILE BIT_ULL(61) 1303 #define PM_SWAP BIT_ULL(62) 1304 #define PM_PRESENT BIT_ULL(63) 1305 1306 #define PM_END_OF_BUFFER 1 1307 1308 static inline pagemap_entry_t make_pme(u64 frame, u64 flags) 1309 { 1310 return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags }; 1311 } 1312 1313 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme, 1314 struct pagemapread *pm) 1315 { 1316 pm->buffer[pm->pos++] = *pme; 1317 if (pm->pos >= pm->len) 1318 return PM_END_OF_BUFFER; 1319 return 0; 1320 } 1321 1322 static int pagemap_pte_hole(unsigned long start, unsigned long end, 1323 __always_unused int depth, struct mm_walk *walk) 1324 { 1325 struct pagemapread *pm = walk->private; 1326 unsigned long addr = start; 1327 int err = 0; 1328 1329 while (addr < end) { 1330 struct vm_area_struct *vma = find_vma(walk->mm, addr); 1331 pagemap_entry_t pme = make_pme(0, 0); 1332 /* End of address space hole, which we mark as non-present. */ 1333 unsigned long hole_end; 1334 1335 if (vma) 1336 hole_end = min(end, vma->vm_start); 1337 else 1338 hole_end = end; 1339 1340 for (; addr < hole_end; addr += PAGE_SIZE) { 1341 err = add_to_pagemap(addr, &pme, pm); 1342 if (err) 1343 goto out; 1344 } 1345 1346 if (!vma) 1347 break; 1348 1349 /* Addresses in the VMA. */ 1350 if (vma->vm_flags & VM_SOFTDIRTY) 1351 pme = make_pme(0, PM_SOFT_DIRTY); 1352 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) { 1353 err = add_to_pagemap(addr, &pme, pm); 1354 if (err) 1355 goto out; 1356 } 1357 } 1358 out: 1359 return err; 1360 } 1361 1362 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm, 1363 struct vm_area_struct *vma, unsigned long addr, pte_t pte) 1364 { 1365 u64 frame = 0, flags = 0; 1366 struct page *page = NULL; 1367 1368 if (pte_present(pte)) { 1369 if (pm->show_pfn) 1370 frame = pte_pfn(pte); 1371 flags |= PM_PRESENT; 1372 page = vm_normal_page(vma, addr, pte); 1373 if (pte_soft_dirty(pte)) 1374 flags |= PM_SOFT_DIRTY; 1375 if (pte_uffd_wp(pte)) 1376 flags |= PM_UFFD_WP; 1377 } else if (is_swap_pte(pte)) { 1378 swp_entry_t entry; 1379 if (pte_swp_soft_dirty(pte)) 1380 flags |= PM_SOFT_DIRTY; 1381 if (pte_swp_uffd_wp(pte)) 1382 flags |= PM_UFFD_WP; 1383 entry = pte_to_swp_entry(pte); 1384 if (pm->show_pfn) 1385 frame = swp_type(entry) | 1386 (swp_offset(entry) << MAX_SWAPFILES_SHIFT); 1387 flags |= PM_SWAP; 1388 if (is_pfn_swap_entry(entry)) 1389 page = pfn_swap_entry_to_page(entry); 1390 } 1391 1392 if (page && !PageAnon(page)) 1393 flags |= PM_FILE; 1394 if (page && page_mapcount(page) == 1) 1395 flags |= PM_MMAP_EXCLUSIVE; 1396 if (vma->vm_flags & VM_SOFTDIRTY) 1397 flags |= PM_SOFT_DIRTY; 1398 1399 return make_pme(frame, flags); 1400 } 1401 1402 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end, 1403 struct mm_walk *walk) 1404 { 1405 struct vm_area_struct *vma = walk->vma; 1406 struct pagemapread *pm = walk->private; 1407 spinlock_t *ptl; 1408 pte_t *pte, *orig_pte; 1409 int err = 0; 1410 1411 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1412 ptl = pmd_trans_huge_lock(pmdp, vma); 1413 if (ptl) { 1414 u64 flags = 0, frame = 0; 1415 pmd_t pmd = *pmdp; 1416 struct page *page = NULL; 1417 1418 if (vma->vm_flags & VM_SOFTDIRTY) 1419 flags |= PM_SOFT_DIRTY; 1420 1421 if (pmd_present(pmd)) { 1422 page = pmd_page(pmd); 1423 1424 flags |= PM_PRESENT; 1425 if (pmd_soft_dirty(pmd)) 1426 flags |= PM_SOFT_DIRTY; 1427 if (pmd_uffd_wp(pmd)) 1428 flags |= PM_UFFD_WP; 1429 if (pm->show_pfn) 1430 frame = pmd_pfn(pmd) + 1431 ((addr & ~PMD_MASK) >> PAGE_SHIFT); 1432 } 1433 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION 1434 else if (is_swap_pmd(pmd)) { 1435 swp_entry_t entry = pmd_to_swp_entry(pmd); 1436 unsigned long offset; 1437 1438 if (pm->show_pfn) { 1439 offset = swp_offset(entry) + 1440 ((addr & ~PMD_MASK) >> PAGE_SHIFT); 1441 frame = swp_type(entry) | 1442 (offset << MAX_SWAPFILES_SHIFT); 1443 } 1444 flags |= PM_SWAP; 1445 if (pmd_swp_soft_dirty(pmd)) 1446 flags |= PM_SOFT_DIRTY; 1447 if (pmd_swp_uffd_wp(pmd)) 1448 flags |= PM_UFFD_WP; 1449 VM_BUG_ON(!is_pmd_migration_entry(pmd)); 1450 page = pfn_swap_entry_to_page(entry); 1451 } 1452 #endif 1453 1454 if (page && page_mapcount(page) == 1) 1455 flags |= PM_MMAP_EXCLUSIVE; 1456 1457 for (; addr != end; addr += PAGE_SIZE) { 1458 pagemap_entry_t pme = make_pme(frame, flags); 1459 1460 err = add_to_pagemap(addr, &pme, pm); 1461 if (err) 1462 break; 1463 if (pm->show_pfn) { 1464 if (flags & PM_PRESENT) 1465 frame++; 1466 else if (flags & PM_SWAP) 1467 frame += (1 << MAX_SWAPFILES_SHIFT); 1468 } 1469 } 1470 spin_unlock(ptl); 1471 return err; 1472 } 1473 1474 if (pmd_trans_unstable(pmdp)) 1475 return 0; 1476 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ 1477 1478 /* 1479 * We can assume that @vma always points to a valid one and @end never 1480 * goes beyond vma->vm_end. 1481 */ 1482 orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl); 1483 for (; addr < end; pte++, addr += PAGE_SIZE) { 1484 pagemap_entry_t pme; 1485 1486 pme = pte_to_pagemap_entry(pm, vma, addr, *pte); 1487 err = add_to_pagemap(addr, &pme, pm); 1488 if (err) 1489 break; 1490 } 1491 pte_unmap_unlock(orig_pte, ptl); 1492 1493 cond_resched(); 1494 1495 return err; 1496 } 1497 1498 #ifdef CONFIG_HUGETLB_PAGE 1499 /* This function walks within one hugetlb entry in the single call */ 1500 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask, 1501 unsigned long addr, unsigned long end, 1502 struct mm_walk *walk) 1503 { 1504 struct pagemapread *pm = walk->private; 1505 struct vm_area_struct *vma = walk->vma; 1506 u64 flags = 0, frame = 0; 1507 int err = 0; 1508 pte_t pte; 1509 1510 if (vma->vm_flags & VM_SOFTDIRTY) 1511 flags |= PM_SOFT_DIRTY; 1512 1513 pte = huge_ptep_get(ptep); 1514 if (pte_present(pte)) { 1515 struct page *page = pte_page(pte); 1516 1517 if (!PageAnon(page)) 1518 flags |= PM_FILE; 1519 1520 if (page_mapcount(page) == 1) 1521 flags |= PM_MMAP_EXCLUSIVE; 1522 1523 flags |= PM_PRESENT; 1524 if (pm->show_pfn) 1525 frame = pte_pfn(pte) + 1526 ((addr & ~hmask) >> PAGE_SHIFT); 1527 } 1528 1529 for (; addr != end; addr += PAGE_SIZE) { 1530 pagemap_entry_t pme = make_pme(frame, flags); 1531 1532 err = add_to_pagemap(addr, &pme, pm); 1533 if (err) 1534 return err; 1535 if (pm->show_pfn && (flags & PM_PRESENT)) 1536 frame++; 1537 } 1538 1539 cond_resched(); 1540 1541 return err; 1542 } 1543 #else 1544 #define pagemap_hugetlb_range NULL 1545 #endif /* HUGETLB_PAGE */ 1546 1547 static const struct mm_walk_ops pagemap_ops = { 1548 .pmd_entry = pagemap_pmd_range, 1549 .pte_hole = pagemap_pte_hole, 1550 .hugetlb_entry = pagemap_hugetlb_range, 1551 }; 1552 1553 /* 1554 * /proc/pid/pagemap - an array mapping virtual pages to pfns 1555 * 1556 * For each page in the address space, this file contains one 64-bit entry 1557 * consisting of the following: 1558 * 1559 * Bits 0-54 page frame number (PFN) if present 1560 * Bits 0-4 swap type if swapped 1561 * Bits 5-54 swap offset if swapped 1562 * Bit 55 pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst) 1563 * Bit 56 page exclusively mapped 1564 * Bits 57-60 zero 1565 * Bit 61 page is file-page or shared-anon 1566 * Bit 62 page swapped 1567 * Bit 63 page present 1568 * 1569 * If the page is not present but in swap, then the PFN contains an 1570 * encoding of the swap file number and the page's offset into the 1571 * swap. Unmapped pages return a null PFN. This allows determining 1572 * precisely which pages are mapped (or in swap) and comparing mapped 1573 * pages between processes. 1574 * 1575 * Efficient users of this interface will use /proc/pid/maps to 1576 * determine which areas of memory are actually mapped and llseek to 1577 * skip over unmapped regions. 1578 */ 1579 static ssize_t pagemap_read(struct file *file, char __user *buf, 1580 size_t count, loff_t *ppos) 1581 { 1582 struct mm_struct *mm = file->private_data; 1583 struct pagemapread pm; 1584 unsigned long src; 1585 unsigned long svpfn; 1586 unsigned long start_vaddr; 1587 unsigned long end_vaddr; 1588 int ret = 0, copied = 0; 1589 1590 if (!mm || !mmget_not_zero(mm)) 1591 goto out; 1592 1593 ret = -EINVAL; 1594 /* file position must be aligned */ 1595 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES)) 1596 goto out_mm; 1597 1598 ret = 0; 1599 if (!count) 1600 goto out_mm; 1601 1602 /* do not disclose physical addresses: attack vector */ 1603 pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN); 1604 1605 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT); 1606 pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL); 1607 ret = -ENOMEM; 1608 if (!pm.buffer) 1609 goto out_mm; 1610 1611 src = *ppos; 1612 svpfn = src / PM_ENTRY_BYTES; 1613 end_vaddr = mm->task_size; 1614 1615 /* watch out for wraparound */ 1616 start_vaddr = end_vaddr; 1617 if (svpfn <= (ULONG_MAX >> PAGE_SHIFT)) 1618 start_vaddr = untagged_addr(svpfn << PAGE_SHIFT); 1619 1620 /* Ensure the address is inside the task */ 1621 if (start_vaddr > mm->task_size) 1622 start_vaddr = end_vaddr; 1623 1624 /* 1625 * The odds are that this will stop walking way 1626 * before end_vaddr, because the length of the 1627 * user buffer is tracked in "pm", and the walk 1628 * will stop when we hit the end of the buffer. 1629 */ 1630 ret = 0; 1631 while (count && (start_vaddr < end_vaddr)) { 1632 int len; 1633 unsigned long end; 1634 1635 pm.pos = 0; 1636 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK; 1637 /* overflow ? */ 1638 if (end < start_vaddr || end > end_vaddr) 1639 end = end_vaddr; 1640 ret = mmap_read_lock_killable(mm); 1641 if (ret) 1642 goto out_free; 1643 ret = walk_page_range(mm, start_vaddr, end, &pagemap_ops, &pm); 1644 mmap_read_unlock(mm); 1645 start_vaddr = end; 1646 1647 len = min(count, PM_ENTRY_BYTES * pm.pos); 1648 if (copy_to_user(buf, pm.buffer, len)) { 1649 ret = -EFAULT; 1650 goto out_free; 1651 } 1652 copied += len; 1653 buf += len; 1654 count -= len; 1655 } 1656 *ppos += copied; 1657 if (!ret || ret == PM_END_OF_BUFFER) 1658 ret = copied; 1659 1660 out_free: 1661 kfree(pm.buffer); 1662 out_mm: 1663 mmput(mm); 1664 out: 1665 return ret; 1666 } 1667 1668 static int pagemap_open(struct inode *inode, struct file *file) 1669 { 1670 struct mm_struct *mm; 1671 1672 mm = proc_mem_open(inode, PTRACE_MODE_READ); 1673 if (IS_ERR(mm)) 1674 return PTR_ERR(mm); 1675 file->private_data = mm; 1676 return 0; 1677 } 1678 1679 static int pagemap_release(struct inode *inode, struct file *file) 1680 { 1681 struct mm_struct *mm = file->private_data; 1682 1683 if (mm) 1684 mmdrop(mm); 1685 return 0; 1686 } 1687 1688 const struct file_operations proc_pagemap_operations = { 1689 .llseek = mem_lseek, /* borrow this */ 1690 .read = pagemap_read, 1691 .open = pagemap_open, 1692 .release = pagemap_release, 1693 }; 1694 #endif /* CONFIG_PROC_PAGE_MONITOR */ 1695 1696 #ifdef CONFIG_NUMA 1697 1698 struct numa_maps { 1699 unsigned long pages; 1700 unsigned long anon; 1701 unsigned long active; 1702 unsigned long writeback; 1703 unsigned long mapcount_max; 1704 unsigned long dirty; 1705 unsigned long swapcache; 1706 unsigned long node[MAX_NUMNODES]; 1707 }; 1708 1709 struct numa_maps_private { 1710 struct proc_maps_private proc_maps; 1711 struct numa_maps md; 1712 }; 1713 1714 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty, 1715 unsigned long nr_pages) 1716 { 1717 int count = page_mapcount(page); 1718 1719 md->pages += nr_pages; 1720 if (pte_dirty || PageDirty(page)) 1721 md->dirty += nr_pages; 1722 1723 if (PageSwapCache(page)) 1724 md->swapcache += nr_pages; 1725 1726 if (PageActive(page) || PageUnevictable(page)) 1727 md->active += nr_pages; 1728 1729 if (PageWriteback(page)) 1730 md->writeback += nr_pages; 1731 1732 if (PageAnon(page)) 1733 md->anon += nr_pages; 1734 1735 if (count > md->mapcount_max) 1736 md->mapcount_max = count; 1737 1738 md->node[page_to_nid(page)] += nr_pages; 1739 } 1740 1741 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma, 1742 unsigned long addr) 1743 { 1744 struct page *page; 1745 int nid; 1746 1747 if (!pte_present(pte)) 1748 return NULL; 1749 1750 page = vm_normal_page(vma, addr, pte); 1751 if (!page) 1752 return NULL; 1753 1754 if (PageReserved(page)) 1755 return NULL; 1756 1757 nid = page_to_nid(page); 1758 if (!node_isset(nid, node_states[N_MEMORY])) 1759 return NULL; 1760 1761 return page; 1762 } 1763 1764 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1765 static struct page *can_gather_numa_stats_pmd(pmd_t pmd, 1766 struct vm_area_struct *vma, 1767 unsigned long addr) 1768 { 1769 struct page *page; 1770 int nid; 1771 1772 if (!pmd_present(pmd)) 1773 return NULL; 1774 1775 page = vm_normal_page_pmd(vma, addr, pmd); 1776 if (!page) 1777 return NULL; 1778 1779 if (PageReserved(page)) 1780 return NULL; 1781 1782 nid = page_to_nid(page); 1783 if (!node_isset(nid, node_states[N_MEMORY])) 1784 return NULL; 1785 1786 return page; 1787 } 1788 #endif 1789 1790 static int gather_pte_stats(pmd_t *pmd, unsigned long addr, 1791 unsigned long end, struct mm_walk *walk) 1792 { 1793 struct numa_maps *md = walk->private; 1794 struct vm_area_struct *vma = walk->vma; 1795 spinlock_t *ptl; 1796 pte_t *orig_pte; 1797 pte_t *pte; 1798 1799 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1800 ptl = pmd_trans_huge_lock(pmd, vma); 1801 if (ptl) { 1802 struct page *page; 1803 1804 page = can_gather_numa_stats_pmd(*pmd, vma, addr); 1805 if (page) 1806 gather_stats(page, md, pmd_dirty(*pmd), 1807 HPAGE_PMD_SIZE/PAGE_SIZE); 1808 spin_unlock(ptl); 1809 return 0; 1810 } 1811 1812 if (pmd_trans_unstable(pmd)) 1813 return 0; 1814 #endif 1815 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl); 1816 do { 1817 struct page *page = can_gather_numa_stats(*pte, vma, addr); 1818 if (!page) 1819 continue; 1820 gather_stats(page, md, pte_dirty(*pte), 1); 1821 1822 } while (pte++, addr += PAGE_SIZE, addr != end); 1823 pte_unmap_unlock(orig_pte, ptl); 1824 cond_resched(); 1825 return 0; 1826 } 1827 #ifdef CONFIG_HUGETLB_PAGE 1828 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask, 1829 unsigned long addr, unsigned long end, struct mm_walk *walk) 1830 { 1831 pte_t huge_pte = huge_ptep_get(pte); 1832 struct numa_maps *md; 1833 struct page *page; 1834 1835 if (!pte_present(huge_pte)) 1836 return 0; 1837 1838 page = pte_page(huge_pte); 1839 if (!page) 1840 return 0; 1841 1842 md = walk->private; 1843 gather_stats(page, md, pte_dirty(huge_pte), 1); 1844 return 0; 1845 } 1846 1847 #else 1848 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask, 1849 unsigned long addr, unsigned long end, struct mm_walk *walk) 1850 { 1851 return 0; 1852 } 1853 #endif 1854 1855 static const struct mm_walk_ops show_numa_ops = { 1856 .hugetlb_entry = gather_hugetlb_stats, 1857 .pmd_entry = gather_pte_stats, 1858 }; 1859 1860 /* 1861 * Display pages allocated per node and memory policy via /proc. 1862 */ 1863 static int show_numa_map(struct seq_file *m, void *v) 1864 { 1865 struct numa_maps_private *numa_priv = m->private; 1866 struct proc_maps_private *proc_priv = &numa_priv->proc_maps; 1867 struct vm_area_struct *vma = v; 1868 struct numa_maps *md = &numa_priv->md; 1869 struct file *file = vma->vm_file; 1870 struct mm_struct *mm = vma->vm_mm; 1871 struct mempolicy *pol; 1872 char buffer[64]; 1873 int nid; 1874 1875 if (!mm) 1876 return 0; 1877 1878 /* Ensure we start with an empty set of numa_maps statistics. */ 1879 memset(md, 0, sizeof(*md)); 1880 1881 pol = __get_vma_policy(vma, vma->vm_start); 1882 if (pol) { 1883 mpol_to_str(buffer, sizeof(buffer), pol); 1884 mpol_cond_put(pol); 1885 } else { 1886 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy); 1887 } 1888 1889 seq_printf(m, "%08lx %s", vma->vm_start, buffer); 1890 1891 if (file) { 1892 seq_puts(m, " file="); 1893 seq_file_path(m, file, "\n\t= "); 1894 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) { 1895 seq_puts(m, " heap"); 1896 } else if (is_stack(vma)) { 1897 seq_puts(m, " stack"); 1898 } 1899 1900 if (is_vm_hugetlb_page(vma)) 1901 seq_puts(m, " huge"); 1902 1903 /* mmap_lock is held by m_start */ 1904 walk_page_vma(vma, &show_numa_ops, md); 1905 1906 if (!md->pages) 1907 goto out; 1908 1909 if (md->anon) 1910 seq_printf(m, " anon=%lu", md->anon); 1911 1912 if (md->dirty) 1913 seq_printf(m, " dirty=%lu", md->dirty); 1914 1915 if (md->pages != md->anon && md->pages != md->dirty) 1916 seq_printf(m, " mapped=%lu", md->pages); 1917 1918 if (md->mapcount_max > 1) 1919 seq_printf(m, " mapmax=%lu", md->mapcount_max); 1920 1921 if (md->swapcache) 1922 seq_printf(m, " swapcache=%lu", md->swapcache); 1923 1924 if (md->active < md->pages && !is_vm_hugetlb_page(vma)) 1925 seq_printf(m, " active=%lu", md->active); 1926 1927 if (md->writeback) 1928 seq_printf(m, " writeback=%lu", md->writeback); 1929 1930 for_each_node_state(nid, N_MEMORY) 1931 if (md->node[nid]) 1932 seq_printf(m, " N%d=%lu", nid, md->node[nid]); 1933 1934 seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10); 1935 out: 1936 seq_putc(m, '\n'); 1937 return 0; 1938 } 1939 1940 static const struct seq_operations proc_pid_numa_maps_op = { 1941 .start = m_start, 1942 .next = m_next, 1943 .stop = m_stop, 1944 .show = show_numa_map, 1945 }; 1946 1947 static int pid_numa_maps_open(struct inode *inode, struct file *file) 1948 { 1949 return proc_maps_open(inode, file, &proc_pid_numa_maps_op, 1950 sizeof(struct numa_maps_private)); 1951 } 1952 1953 const struct file_operations proc_pid_numa_maps_operations = { 1954 .open = pid_numa_maps_open, 1955 .read = seq_read, 1956 .llseek = seq_lseek, 1957 .release = proc_map_release, 1958 }; 1959 1960 #endif /* CONFIG_NUMA */ 1961