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_migration_entry(swpent)) 518 page = migration_entry_to_page(swpent); 519 else if (is_device_private_entry(swpent)) 520 page = device_private_entry_to_page(swpent); 521 } else if (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap 522 && pte_none(*pte))) { 523 page = xa_load(&vma->vm_file->f_mapping->i_pages, 524 linear_page_index(vma, addr)); 525 if (xa_is_value(page)) 526 mss->swap += PAGE_SIZE; 527 return; 528 } 529 530 if (!page) 531 return; 532 533 smaps_account(mss, page, false, pte_young(*pte), pte_dirty(*pte), locked); 534 } 535 536 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 537 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr, 538 struct mm_walk *walk) 539 { 540 struct mem_size_stats *mss = walk->private; 541 struct vm_area_struct *vma = walk->vma; 542 bool locked = !!(vma->vm_flags & VM_LOCKED); 543 struct page *page = NULL; 544 545 if (pmd_present(*pmd)) { 546 /* FOLL_DUMP will return -EFAULT on huge zero page */ 547 page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP); 548 } else if (unlikely(thp_migration_supported() && is_swap_pmd(*pmd))) { 549 swp_entry_t entry = pmd_to_swp_entry(*pmd); 550 551 if (is_migration_entry(entry)) 552 page = migration_entry_to_page(entry); 553 } 554 if (IS_ERR_OR_NULL(page)) 555 return; 556 if (PageAnon(page)) 557 mss->anonymous_thp += HPAGE_PMD_SIZE; 558 else if (PageSwapBacked(page)) 559 mss->shmem_thp += HPAGE_PMD_SIZE; 560 else if (is_zone_device_page(page)) 561 /* pass */; 562 else 563 mss->file_thp += HPAGE_PMD_SIZE; 564 smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd), locked); 565 } 566 #else 567 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr, 568 struct mm_walk *walk) 569 { 570 } 571 #endif 572 573 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, 574 struct mm_walk *walk) 575 { 576 struct vm_area_struct *vma = walk->vma; 577 pte_t *pte; 578 spinlock_t *ptl; 579 580 ptl = pmd_trans_huge_lock(pmd, vma); 581 if (ptl) { 582 smaps_pmd_entry(pmd, addr, walk); 583 spin_unlock(ptl); 584 goto out; 585 } 586 587 if (pmd_trans_unstable(pmd)) 588 goto out; 589 /* 590 * The mmap_lock held all the way back in m_start() is what 591 * keeps khugepaged out of here and from collapsing things 592 * in here. 593 */ 594 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); 595 for (; addr != end; pte++, addr += PAGE_SIZE) 596 smaps_pte_entry(pte, addr, walk); 597 pte_unmap_unlock(pte - 1, ptl); 598 out: 599 cond_resched(); 600 return 0; 601 } 602 603 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma) 604 { 605 /* 606 * Don't forget to update Documentation/ on changes. 607 */ 608 static const char mnemonics[BITS_PER_LONG][2] = { 609 /* 610 * In case if we meet a flag we don't know about. 611 */ 612 [0 ... (BITS_PER_LONG-1)] = "??", 613 614 [ilog2(VM_READ)] = "rd", 615 [ilog2(VM_WRITE)] = "wr", 616 [ilog2(VM_EXEC)] = "ex", 617 [ilog2(VM_SHARED)] = "sh", 618 [ilog2(VM_MAYREAD)] = "mr", 619 [ilog2(VM_MAYWRITE)] = "mw", 620 [ilog2(VM_MAYEXEC)] = "me", 621 [ilog2(VM_MAYSHARE)] = "ms", 622 [ilog2(VM_GROWSDOWN)] = "gd", 623 [ilog2(VM_PFNMAP)] = "pf", 624 [ilog2(VM_DENYWRITE)] = "dw", 625 [ilog2(VM_LOCKED)] = "lo", 626 [ilog2(VM_IO)] = "io", 627 [ilog2(VM_SEQ_READ)] = "sr", 628 [ilog2(VM_RAND_READ)] = "rr", 629 [ilog2(VM_DONTCOPY)] = "dc", 630 [ilog2(VM_DONTEXPAND)] = "de", 631 [ilog2(VM_ACCOUNT)] = "ac", 632 [ilog2(VM_NORESERVE)] = "nr", 633 [ilog2(VM_HUGETLB)] = "ht", 634 [ilog2(VM_SYNC)] = "sf", 635 [ilog2(VM_ARCH_1)] = "ar", 636 [ilog2(VM_WIPEONFORK)] = "wf", 637 [ilog2(VM_DONTDUMP)] = "dd", 638 #ifdef CONFIG_ARM64_BTI 639 [ilog2(VM_ARM64_BTI)] = "bt", 640 #endif 641 #ifdef CONFIG_MEM_SOFT_DIRTY 642 [ilog2(VM_SOFTDIRTY)] = "sd", 643 #endif 644 [ilog2(VM_MIXEDMAP)] = "mm", 645 [ilog2(VM_HUGEPAGE)] = "hg", 646 [ilog2(VM_NOHUGEPAGE)] = "nh", 647 [ilog2(VM_MERGEABLE)] = "mg", 648 [ilog2(VM_UFFD_MISSING)]= "um", 649 [ilog2(VM_UFFD_WP)] = "uw", 650 #ifdef CONFIG_ARM64_MTE 651 [ilog2(VM_MTE)] = "mt", 652 [ilog2(VM_MTE_ALLOWED)] = "", 653 #endif 654 #ifdef CONFIG_ARCH_HAS_PKEYS 655 /* These come out via ProtectionKey: */ 656 [ilog2(VM_PKEY_BIT0)] = "", 657 [ilog2(VM_PKEY_BIT1)] = "", 658 [ilog2(VM_PKEY_BIT2)] = "", 659 [ilog2(VM_PKEY_BIT3)] = "", 660 #if VM_PKEY_BIT4 661 [ilog2(VM_PKEY_BIT4)] = "", 662 #endif 663 #endif /* CONFIG_ARCH_HAS_PKEYS */ 664 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_MINOR 665 [ilog2(VM_UFFD_MINOR)] = "ui", 666 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_MINOR */ 667 }; 668 size_t i; 669 670 seq_puts(m, "VmFlags: "); 671 for (i = 0; i < BITS_PER_LONG; i++) { 672 if (!mnemonics[i][0]) 673 continue; 674 if (vma->vm_flags & (1UL << i)) { 675 seq_putc(m, mnemonics[i][0]); 676 seq_putc(m, mnemonics[i][1]); 677 seq_putc(m, ' '); 678 } 679 } 680 seq_putc(m, '\n'); 681 } 682 683 #ifdef CONFIG_HUGETLB_PAGE 684 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask, 685 unsigned long addr, unsigned long end, 686 struct mm_walk *walk) 687 { 688 struct mem_size_stats *mss = walk->private; 689 struct vm_area_struct *vma = walk->vma; 690 struct page *page = NULL; 691 692 if (pte_present(*pte)) { 693 page = vm_normal_page(vma, addr, *pte); 694 } else if (is_swap_pte(*pte)) { 695 swp_entry_t swpent = pte_to_swp_entry(*pte); 696 697 if (is_migration_entry(swpent)) 698 page = migration_entry_to_page(swpent); 699 else if (is_device_private_entry(swpent)) 700 page = device_private_entry_to_page(swpent); 701 } 702 if (page) { 703 int mapcount = page_mapcount(page); 704 705 if (mapcount >= 2) 706 mss->shared_hugetlb += huge_page_size(hstate_vma(vma)); 707 else 708 mss->private_hugetlb += huge_page_size(hstate_vma(vma)); 709 } 710 return 0; 711 } 712 #else 713 #define smaps_hugetlb_range NULL 714 #endif /* HUGETLB_PAGE */ 715 716 static const struct mm_walk_ops smaps_walk_ops = { 717 .pmd_entry = smaps_pte_range, 718 .hugetlb_entry = smaps_hugetlb_range, 719 }; 720 721 static const struct mm_walk_ops smaps_shmem_walk_ops = { 722 .pmd_entry = smaps_pte_range, 723 .hugetlb_entry = smaps_hugetlb_range, 724 .pte_hole = smaps_pte_hole, 725 }; 726 727 /* 728 * Gather mem stats from @vma with the indicated beginning 729 * address @start, and keep them in @mss. 730 * 731 * Use vm_start of @vma as the beginning address if @start is 0. 732 */ 733 static void smap_gather_stats(struct vm_area_struct *vma, 734 struct mem_size_stats *mss, unsigned long start) 735 { 736 const struct mm_walk_ops *ops = &smaps_walk_ops; 737 738 /* Invalid start */ 739 if (start >= vma->vm_end) 740 return; 741 742 #ifdef CONFIG_SHMEM 743 /* In case of smaps_rollup, reset the value from previous vma */ 744 mss->check_shmem_swap = false; 745 if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) { 746 /* 747 * For shared or readonly shmem mappings we know that all 748 * swapped out pages belong to the shmem object, and we can 749 * obtain the swap value much more efficiently. For private 750 * writable mappings, we might have COW pages that are 751 * not affected by the parent swapped out pages of the shmem 752 * object, so we have to distinguish them during the page walk. 753 * Unless we know that the shmem object (or the part mapped by 754 * our VMA) has no swapped out pages at all. 755 */ 756 unsigned long shmem_swapped = shmem_swap_usage(vma); 757 758 if (!start && (!shmem_swapped || (vma->vm_flags & VM_SHARED) || 759 !(vma->vm_flags & VM_WRITE))) { 760 mss->swap += shmem_swapped; 761 } else { 762 mss->check_shmem_swap = true; 763 ops = &smaps_shmem_walk_ops; 764 } 765 } 766 #endif 767 /* mmap_lock is held in m_start */ 768 if (!start) 769 walk_page_vma(vma, ops, mss); 770 else 771 walk_page_range(vma->vm_mm, start, vma->vm_end, ops, mss); 772 } 773 774 #define SEQ_PUT_DEC(str, val) \ 775 seq_put_decimal_ull_width(m, str, (val) >> 10, 8) 776 777 /* Show the contents common for smaps and smaps_rollup */ 778 static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss, 779 bool rollup_mode) 780 { 781 SEQ_PUT_DEC("Rss: ", mss->resident); 782 SEQ_PUT_DEC(" kB\nPss: ", mss->pss >> PSS_SHIFT); 783 if (rollup_mode) { 784 /* 785 * These are meaningful only for smaps_rollup, otherwise two of 786 * them are zero, and the other one is the same as Pss. 787 */ 788 SEQ_PUT_DEC(" kB\nPss_Anon: ", 789 mss->pss_anon >> PSS_SHIFT); 790 SEQ_PUT_DEC(" kB\nPss_File: ", 791 mss->pss_file >> PSS_SHIFT); 792 SEQ_PUT_DEC(" kB\nPss_Shmem: ", 793 mss->pss_shmem >> PSS_SHIFT); 794 } 795 SEQ_PUT_DEC(" kB\nShared_Clean: ", mss->shared_clean); 796 SEQ_PUT_DEC(" kB\nShared_Dirty: ", mss->shared_dirty); 797 SEQ_PUT_DEC(" kB\nPrivate_Clean: ", mss->private_clean); 798 SEQ_PUT_DEC(" kB\nPrivate_Dirty: ", mss->private_dirty); 799 SEQ_PUT_DEC(" kB\nReferenced: ", mss->referenced); 800 SEQ_PUT_DEC(" kB\nAnonymous: ", mss->anonymous); 801 SEQ_PUT_DEC(" kB\nLazyFree: ", mss->lazyfree); 802 SEQ_PUT_DEC(" kB\nAnonHugePages: ", mss->anonymous_thp); 803 SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp); 804 SEQ_PUT_DEC(" kB\nFilePmdMapped: ", mss->file_thp); 805 SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb); 806 seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ", 807 mss->private_hugetlb >> 10, 7); 808 SEQ_PUT_DEC(" kB\nSwap: ", mss->swap); 809 SEQ_PUT_DEC(" kB\nSwapPss: ", 810 mss->swap_pss >> PSS_SHIFT); 811 SEQ_PUT_DEC(" kB\nLocked: ", 812 mss->pss_locked >> PSS_SHIFT); 813 seq_puts(m, " kB\n"); 814 } 815 816 static int show_smap(struct seq_file *m, void *v) 817 { 818 struct vm_area_struct *vma = v; 819 struct mem_size_stats mss; 820 821 memset(&mss, 0, sizeof(mss)); 822 823 smap_gather_stats(vma, &mss, 0); 824 825 show_map_vma(m, vma); 826 827 SEQ_PUT_DEC("Size: ", vma->vm_end - vma->vm_start); 828 SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma)); 829 SEQ_PUT_DEC(" kB\nMMUPageSize: ", vma_mmu_pagesize(vma)); 830 seq_puts(m, " kB\n"); 831 832 __show_smap(m, &mss, false); 833 834 seq_printf(m, "THPeligible: %d\n", 835 transparent_hugepage_enabled(vma)); 836 837 if (arch_pkeys_enabled()) 838 seq_printf(m, "ProtectionKey: %8u\n", vma_pkey(vma)); 839 show_smap_vma_flags(m, vma); 840 841 return 0; 842 } 843 844 static int show_smaps_rollup(struct seq_file *m, void *v) 845 { 846 struct proc_maps_private *priv = m->private; 847 struct mem_size_stats mss; 848 struct mm_struct *mm; 849 struct vm_area_struct *vma; 850 unsigned long last_vma_end = 0; 851 int ret = 0; 852 853 priv->task = get_proc_task(priv->inode); 854 if (!priv->task) 855 return -ESRCH; 856 857 mm = priv->mm; 858 if (!mm || !mmget_not_zero(mm)) { 859 ret = -ESRCH; 860 goto out_put_task; 861 } 862 863 memset(&mss, 0, sizeof(mss)); 864 865 ret = mmap_read_lock_killable(mm); 866 if (ret) 867 goto out_put_mm; 868 869 hold_task_mempolicy(priv); 870 871 for (vma = priv->mm->mmap; vma;) { 872 smap_gather_stats(vma, &mss, 0); 873 last_vma_end = vma->vm_end; 874 875 /* 876 * Release mmap_lock temporarily if someone wants to 877 * access it for write request. 878 */ 879 if (mmap_lock_is_contended(mm)) { 880 mmap_read_unlock(mm); 881 ret = mmap_read_lock_killable(mm); 882 if (ret) { 883 release_task_mempolicy(priv); 884 goto out_put_mm; 885 } 886 887 /* 888 * After dropping the lock, there are four cases to 889 * consider. See the following example for explanation. 890 * 891 * +------+------+-----------+ 892 * | VMA1 | VMA2 | VMA3 | 893 * +------+------+-----------+ 894 * | | | | 895 * 4k 8k 16k 400k 896 * 897 * Suppose we drop the lock after reading VMA2 due to 898 * contention, then we get: 899 * 900 * last_vma_end = 16k 901 * 902 * 1) VMA2 is freed, but VMA3 exists: 903 * 904 * find_vma(mm, 16k - 1) will return VMA3. 905 * In this case, just continue from VMA3. 906 * 907 * 2) VMA2 still exists: 908 * 909 * find_vma(mm, 16k - 1) will return VMA2. 910 * Iterate the loop like the original one. 911 * 912 * 3) No more VMAs can be found: 913 * 914 * find_vma(mm, 16k - 1) will return NULL. 915 * No more things to do, just break. 916 * 917 * 4) (last_vma_end - 1) is the middle of a vma (VMA'): 918 * 919 * find_vma(mm, 16k - 1) will return VMA' whose range 920 * contains last_vma_end. 921 * Iterate VMA' from last_vma_end. 922 */ 923 vma = find_vma(mm, last_vma_end - 1); 924 /* Case 3 above */ 925 if (!vma) 926 break; 927 928 /* Case 1 above */ 929 if (vma->vm_start >= last_vma_end) 930 continue; 931 932 /* Case 4 above */ 933 if (vma->vm_end > last_vma_end) 934 smap_gather_stats(vma, &mss, last_vma_end); 935 } 936 /* Case 2 above */ 937 vma = vma->vm_next; 938 } 939 940 show_vma_header_prefix(m, priv->mm->mmap->vm_start, 941 last_vma_end, 0, 0, 0, 0); 942 seq_pad(m, ' '); 943 seq_puts(m, "[rollup]\n"); 944 945 __show_smap(m, &mss, true); 946 947 release_task_mempolicy(priv); 948 mmap_read_unlock(mm); 949 950 out_put_mm: 951 mmput(mm); 952 out_put_task: 953 put_task_struct(priv->task); 954 priv->task = NULL; 955 956 return ret; 957 } 958 #undef SEQ_PUT_DEC 959 960 static const struct seq_operations proc_pid_smaps_op = { 961 .start = m_start, 962 .next = m_next, 963 .stop = m_stop, 964 .show = show_smap 965 }; 966 967 static int pid_smaps_open(struct inode *inode, struct file *file) 968 { 969 return do_maps_open(inode, file, &proc_pid_smaps_op); 970 } 971 972 static int smaps_rollup_open(struct inode *inode, struct file *file) 973 { 974 int ret; 975 struct proc_maps_private *priv; 976 977 priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT); 978 if (!priv) 979 return -ENOMEM; 980 981 ret = single_open(file, show_smaps_rollup, priv); 982 if (ret) 983 goto out_free; 984 985 priv->inode = inode; 986 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ); 987 if (IS_ERR(priv->mm)) { 988 ret = PTR_ERR(priv->mm); 989 990 single_release(inode, file); 991 goto out_free; 992 } 993 994 return 0; 995 996 out_free: 997 kfree(priv); 998 return ret; 999 } 1000 1001 static int smaps_rollup_release(struct inode *inode, struct file *file) 1002 { 1003 struct seq_file *seq = file->private_data; 1004 struct proc_maps_private *priv = seq->private; 1005 1006 if (priv->mm) 1007 mmdrop(priv->mm); 1008 1009 kfree(priv); 1010 return single_release(inode, file); 1011 } 1012 1013 const struct file_operations proc_pid_smaps_operations = { 1014 .open = pid_smaps_open, 1015 .read = seq_read, 1016 .llseek = seq_lseek, 1017 .release = proc_map_release, 1018 }; 1019 1020 const struct file_operations proc_pid_smaps_rollup_operations = { 1021 .open = smaps_rollup_open, 1022 .read = seq_read, 1023 .llseek = seq_lseek, 1024 .release = smaps_rollup_release, 1025 }; 1026 1027 enum clear_refs_types { 1028 CLEAR_REFS_ALL = 1, 1029 CLEAR_REFS_ANON, 1030 CLEAR_REFS_MAPPED, 1031 CLEAR_REFS_SOFT_DIRTY, 1032 CLEAR_REFS_MM_HIWATER_RSS, 1033 CLEAR_REFS_LAST, 1034 }; 1035 1036 struct clear_refs_private { 1037 enum clear_refs_types type; 1038 }; 1039 1040 #ifdef CONFIG_MEM_SOFT_DIRTY 1041 1042 static inline bool pte_is_pinned(struct vm_area_struct *vma, unsigned long addr, pte_t pte) 1043 { 1044 struct page *page; 1045 1046 if (!pte_write(pte)) 1047 return false; 1048 if (!is_cow_mapping(vma->vm_flags)) 1049 return false; 1050 if (likely(!atomic_read(&vma->vm_mm->has_pinned))) 1051 return false; 1052 page = vm_normal_page(vma, addr, pte); 1053 if (!page) 1054 return false; 1055 return page_maybe_dma_pinned(page); 1056 } 1057 1058 static inline void clear_soft_dirty(struct vm_area_struct *vma, 1059 unsigned long addr, pte_t *pte) 1060 { 1061 /* 1062 * The soft-dirty tracker uses #PF-s to catch writes 1063 * to pages, so write-protect the pte as well. See the 1064 * Documentation/admin-guide/mm/soft-dirty.rst for full description 1065 * of how soft-dirty works. 1066 */ 1067 pte_t ptent = *pte; 1068 1069 if (pte_present(ptent)) { 1070 pte_t old_pte; 1071 1072 if (pte_is_pinned(vma, addr, ptent)) 1073 return; 1074 old_pte = ptep_modify_prot_start(vma, addr, pte); 1075 ptent = pte_wrprotect(old_pte); 1076 ptent = pte_clear_soft_dirty(ptent); 1077 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent); 1078 } else if (is_swap_pte(ptent)) { 1079 ptent = pte_swp_clear_soft_dirty(ptent); 1080 set_pte_at(vma->vm_mm, addr, pte, ptent); 1081 } 1082 } 1083 #else 1084 static inline void clear_soft_dirty(struct vm_area_struct *vma, 1085 unsigned long addr, pte_t *pte) 1086 { 1087 } 1088 #endif 1089 1090 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE) 1091 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma, 1092 unsigned long addr, pmd_t *pmdp) 1093 { 1094 pmd_t old, pmd = *pmdp; 1095 1096 if (pmd_present(pmd)) { 1097 /* See comment in change_huge_pmd() */ 1098 old = pmdp_invalidate(vma, addr, pmdp); 1099 if (pmd_dirty(old)) 1100 pmd = pmd_mkdirty(pmd); 1101 if (pmd_young(old)) 1102 pmd = pmd_mkyoung(pmd); 1103 1104 pmd = pmd_wrprotect(pmd); 1105 pmd = pmd_clear_soft_dirty(pmd); 1106 1107 set_pmd_at(vma->vm_mm, addr, pmdp, pmd); 1108 } else if (is_migration_entry(pmd_to_swp_entry(pmd))) { 1109 pmd = pmd_swp_clear_soft_dirty(pmd); 1110 set_pmd_at(vma->vm_mm, addr, pmdp, pmd); 1111 } 1112 } 1113 #else 1114 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma, 1115 unsigned long addr, pmd_t *pmdp) 1116 { 1117 } 1118 #endif 1119 1120 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr, 1121 unsigned long end, struct mm_walk *walk) 1122 { 1123 struct clear_refs_private *cp = walk->private; 1124 struct vm_area_struct *vma = walk->vma; 1125 pte_t *pte, ptent; 1126 spinlock_t *ptl; 1127 struct page *page; 1128 1129 ptl = pmd_trans_huge_lock(pmd, vma); 1130 if (ptl) { 1131 if (cp->type == CLEAR_REFS_SOFT_DIRTY) { 1132 clear_soft_dirty_pmd(vma, addr, pmd); 1133 goto out; 1134 } 1135 1136 if (!pmd_present(*pmd)) 1137 goto out; 1138 1139 page = pmd_page(*pmd); 1140 1141 /* Clear accessed and referenced bits. */ 1142 pmdp_test_and_clear_young(vma, addr, pmd); 1143 test_and_clear_page_young(page); 1144 ClearPageReferenced(page); 1145 out: 1146 spin_unlock(ptl); 1147 return 0; 1148 } 1149 1150 if (pmd_trans_unstable(pmd)) 1151 return 0; 1152 1153 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); 1154 for (; addr != end; pte++, addr += PAGE_SIZE) { 1155 ptent = *pte; 1156 1157 if (cp->type == CLEAR_REFS_SOFT_DIRTY) { 1158 clear_soft_dirty(vma, addr, pte); 1159 continue; 1160 } 1161 1162 if (!pte_present(ptent)) 1163 continue; 1164 1165 page = vm_normal_page(vma, addr, ptent); 1166 if (!page) 1167 continue; 1168 1169 /* Clear accessed and referenced bits. */ 1170 ptep_test_and_clear_young(vma, addr, pte); 1171 test_and_clear_page_young(page); 1172 ClearPageReferenced(page); 1173 } 1174 pte_unmap_unlock(pte - 1, ptl); 1175 cond_resched(); 1176 return 0; 1177 } 1178 1179 static int clear_refs_test_walk(unsigned long start, unsigned long end, 1180 struct mm_walk *walk) 1181 { 1182 struct clear_refs_private *cp = walk->private; 1183 struct vm_area_struct *vma = walk->vma; 1184 1185 if (vma->vm_flags & VM_PFNMAP) 1186 return 1; 1187 1188 /* 1189 * Writing 1 to /proc/pid/clear_refs affects all pages. 1190 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages. 1191 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages. 1192 * Writing 4 to /proc/pid/clear_refs affects all pages. 1193 */ 1194 if (cp->type == CLEAR_REFS_ANON && vma->vm_file) 1195 return 1; 1196 if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file) 1197 return 1; 1198 return 0; 1199 } 1200 1201 static const struct mm_walk_ops clear_refs_walk_ops = { 1202 .pmd_entry = clear_refs_pte_range, 1203 .test_walk = clear_refs_test_walk, 1204 }; 1205 1206 static ssize_t clear_refs_write(struct file *file, const char __user *buf, 1207 size_t count, loff_t *ppos) 1208 { 1209 struct task_struct *task; 1210 char buffer[PROC_NUMBUF]; 1211 struct mm_struct *mm; 1212 struct vm_area_struct *vma; 1213 enum clear_refs_types type; 1214 int itype; 1215 int rv; 1216 1217 memset(buffer, 0, sizeof(buffer)); 1218 if (count > sizeof(buffer) - 1) 1219 count = sizeof(buffer) - 1; 1220 if (copy_from_user(buffer, buf, count)) 1221 return -EFAULT; 1222 rv = kstrtoint(strstrip(buffer), 10, &itype); 1223 if (rv < 0) 1224 return rv; 1225 type = (enum clear_refs_types)itype; 1226 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST) 1227 return -EINVAL; 1228 1229 task = get_proc_task(file_inode(file)); 1230 if (!task) 1231 return -ESRCH; 1232 mm = get_task_mm(task); 1233 if (mm) { 1234 struct mmu_notifier_range range; 1235 struct clear_refs_private cp = { 1236 .type = type, 1237 }; 1238 1239 if (mmap_write_lock_killable(mm)) { 1240 count = -EINTR; 1241 goto out_mm; 1242 } 1243 if (type == CLEAR_REFS_MM_HIWATER_RSS) { 1244 /* 1245 * Writing 5 to /proc/pid/clear_refs resets the peak 1246 * resident set size to this mm's current rss value. 1247 */ 1248 reset_mm_hiwater_rss(mm); 1249 goto out_unlock; 1250 } 1251 1252 if (type == CLEAR_REFS_SOFT_DIRTY) { 1253 for (vma = mm->mmap; vma; vma = vma->vm_next) { 1254 if (!(vma->vm_flags & VM_SOFTDIRTY)) 1255 continue; 1256 vma->vm_flags &= ~VM_SOFTDIRTY; 1257 vma_set_page_prot(vma); 1258 } 1259 1260 inc_tlb_flush_pending(mm); 1261 mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY, 1262 0, NULL, mm, 0, -1UL); 1263 mmu_notifier_invalidate_range_start(&range); 1264 } 1265 walk_page_range(mm, 0, mm->highest_vm_end, &clear_refs_walk_ops, 1266 &cp); 1267 if (type == CLEAR_REFS_SOFT_DIRTY) { 1268 mmu_notifier_invalidate_range_end(&range); 1269 flush_tlb_mm(mm); 1270 dec_tlb_flush_pending(mm); 1271 } 1272 out_unlock: 1273 mmap_write_unlock(mm); 1274 out_mm: 1275 mmput(mm); 1276 } 1277 put_task_struct(task); 1278 1279 return count; 1280 } 1281 1282 const struct file_operations proc_clear_refs_operations = { 1283 .write = clear_refs_write, 1284 .llseek = noop_llseek, 1285 }; 1286 1287 typedef struct { 1288 u64 pme; 1289 } pagemap_entry_t; 1290 1291 struct pagemapread { 1292 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */ 1293 pagemap_entry_t *buffer; 1294 bool show_pfn; 1295 }; 1296 1297 #define PAGEMAP_WALK_SIZE (PMD_SIZE) 1298 #define PAGEMAP_WALK_MASK (PMD_MASK) 1299 1300 #define PM_ENTRY_BYTES sizeof(pagemap_entry_t) 1301 #define PM_PFRAME_BITS 55 1302 #define PM_PFRAME_MASK GENMASK_ULL(PM_PFRAME_BITS - 1, 0) 1303 #define PM_SOFT_DIRTY BIT_ULL(55) 1304 #define PM_MMAP_EXCLUSIVE BIT_ULL(56) 1305 #define PM_FILE BIT_ULL(61) 1306 #define PM_SWAP BIT_ULL(62) 1307 #define PM_PRESENT BIT_ULL(63) 1308 1309 #define PM_END_OF_BUFFER 1 1310 1311 static inline pagemap_entry_t make_pme(u64 frame, u64 flags) 1312 { 1313 return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags }; 1314 } 1315 1316 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme, 1317 struct pagemapread *pm) 1318 { 1319 pm->buffer[pm->pos++] = *pme; 1320 if (pm->pos >= pm->len) 1321 return PM_END_OF_BUFFER; 1322 return 0; 1323 } 1324 1325 static int pagemap_pte_hole(unsigned long start, unsigned long end, 1326 __always_unused int depth, struct mm_walk *walk) 1327 { 1328 struct pagemapread *pm = walk->private; 1329 unsigned long addr = start; 1330 int err = 0; 1331 1332 while (addr < end) { 1333 struct vm_area_struct *vma = find_vma(walk->mm, addr); 1334 pagemap_entry_t pme = make_pme(0, 0); 1335 /* End of address space hole, which we mark as non-present. */ 1336 unsigned long hole_end; 1337 1338 if (vma) 1339 hole_end = min(end, vma->vm_start); 1340 else 1341 hole_end = end; 1342 1343 for (; addr < hole_end; addr += PAGE_SIZE) { 1344 err = add_to_pagemap(addr, &pme, pm); 1345 if (err) 1346 goto out; 1347 } 1348 1349 if (!vma) 1350 break; 1351 1352 /* Addresses in the VMA. */ 1353 if (vma->vm_flags & VM_SOFTDIRTY) 1354 pme = make_pme(0, PM_SOFT_DIRTY); 1355 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) { 1356 err = add_to_pagemap(addr, &pme, pm); 1357 if (err) 1358 goto out; 1359 } 1360 } 1361 out: 1362 return err; 1363 } 1364 1365 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm, 1366 struct vm_area_struct *vma, unsigned long addr, pte_t pte) 1367 { 1368 u64 frame = 0, flags = 0; 1369 struct page *page = NULL; 1370 1371 if (pte_present(pte)) { 1372 if (pm->show_pfn) 1373 frame = pte_pfn(pte); 1374 flags |= PM_PRESENT; 1375 page = vm_normal_page(vma, addr, pte); 1376 if (pte_soft_dirty(pte)) 1377 flags |= PM_SOFT_DIRTY; 1378 } else if (is_swap_pte(pte)) { 1379 swp_entry_t entry; 1380 if (pte_swp_soft_dirty(pte)) 1381 flags |= PM_SOFT_DIRTY; 1382 entry = pte_to_swp_entry(pte); 1383 if (pm->show_pfn) 1384 frame = swp_type(entry) | 1385 (swp_offset(entry) << MAX_SWAPFILES_SHIFT); 1386 flags |= PM_SWAP; 1387 if (is_migration_entry(entry)) 1388 page = migration_entry_to_page(entry); 1389 1390 if (is_device_private_entry(entry)) 1391 page = device_private_entry_to_page(entry); 1392 } 1393 1394 if (page && !PageAnon(page)) 1395 flags |= PM_FILE; 1396 if (page && page_mapcount(page) == 1) 1397 flags |= PM_MMAP_EXCLUSIVE; 1398 if (vma->vm_flags & VM_SOFTDIRTY) 1399 flags |= PM_SOFT_DIRTY; 1400 1401 return make_pme(frame, flags); 1402 } 1403 1404 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end, 1405 struct mm_walk *walk) 1406 { 1407 struct vm_area_struct *vma = walk->vma; 1408 struct pagemapread *pm = walk->private; 1409 spinlock_t *ptl; 1410 pte_t *pte, *orig_pte; 1411 int err = 0; 1412 1413 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1414 ptl = pmd_trans_huge_lock(pmdp, vma); 1415 if (ptl) { 1416 u64 flags = 0, frame = 0; 1417 pmd_t pmd = *pmdp; 1418 struct page *page = NULL; 1419 1420 if (vma->vm_flags & VM_SOFTDIRTY) 1421 flags |= PM_SOFT_DIRTY; 1422 1423 if (pmd_present(pmd)) { 1424 page = pmd_page(pmd); 1425 1426 flags |= PM_PRESENT; 1427 if (pmd_soft_dirty(pmd)) 1428 flags |= PM_SOFT_DIRTY; 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 VM_BUG_ON(!is_pmd_migration_entry(pmd)); 1448 page = migration_entry_to_page(entry); 1449 } 1450 #endif 1451 1452 if (page && page_mapcount(page) == 1) 1453 flags |= PM_MMAP_EXCLUSIVE; 1454 1455 for (; addr != end; addr += PAGE_SIZE) { 1456 pagemap_entry_t pme = make_pme(frame, flags); 1457 1458 err = add_to_pagemap(addr, &pme, pm); 1459 if (err) 1460 break; 1461 if (pm->show_pfn) { 1462 if (flags & PM_PRESENT) 1463 frame++; 1464 else if (flags & PM_SWAP) 1465 frame += (1 << MAX_SWAPFILES_SHIFT); 1466 } 1467 } 1468 spin_unlock(ptl); 1469 return err; 1470 } 1471 1472 if (pmd_trans_unstable(pmdp)) 1473 return 0; 1474 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ 1475 1476 /* 1477 * We can assume that @vma always points to a valid one and @end never 1478 * goes beyond vma->vm_end. 1479 */ 1480 orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl); 1481 for (; addr < end; pte++, addr += PAGE_SIZE) { 1482 pagemap_entry_t pme; 1483 1484 pme = pte_to_pagemap_entry(pm, vma, addr, *pte); 1485 err = add_to_pagemap(addr, &pme, pm); 1486 if (err) 1487 break; 1488 } 1489 pte_unmap_unlock(orig_pte, ptl); 1490 1491 cond_resched(); 1492 1493 return err; 1494 } 1495 1496 #ifdef CONFIG_HUGETLB_PAGE 1497 /* This function walks within one hugetlb entry in the single call */ 1498 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask, 1499 unsigned long addr, unsigned long end, 1500 struct mm_walk *walk) 1501 { 1502 struct pagemapread *pm = walk->private; 1503 struct vm_area_struct *vma = walk->vma; 1504 u64 flags = 0, frame = 0; 1505 int err = 0; 1506 pte_t pte; 1507 1508 if (vma->vm_flags & VM_SOFTDIRTY) 1509 flags |= PM_SOFT_DIRTY; 1510 1511 pte = huge_ptep_get(ptep); 1512 if (pte_present(pte)) { 1513 struct page *page = pte_page(pte); 1514 1515 if (!PageAnon(page)) 1516 flags |= PM_FILE; 1517 1518 if (page_mapcount(page) == 1) 1519 flags |= PM_MMAP_EXCLUSIVE; 1520 1521 flags |= PM_PRESENT; 1522 if (pm->show_pfn) 1523 frame = pte_pfn(pte) + 1524 ((addr & ~hmask) >> PAGE_SHIFT); 1525 } 1526 1527 for (; addr != end; addr += PAGE_SIZE) { 1528 pagemap_entry_t pme = make_pme(frame, flags); 1529 1530 err = add_to_pagemap(addr, &pme, pm); 1531 if (err) 1532 return err; 1533 if (pm->show_pfn && (flags & PM_PRESENT)) 1534 frame++; 1535 } 1536 1537 cond_resched(); 1538 1539 return err; 1540 } 1541 #else 1542 #define pagemap_hugetlb_range NULL 1543 #endif /* HUGETLB_PAGE */ 1544 1545 static const struct mm_walk_ops pagemap_ops = { 1546 .pmd_entry = pagemap_pmd_range, 1547 .pte_hole = pagemap_pte_hole, 1548 .hugetlb_entry = pagemap_hugetlb_range, 1549 }; 1550 1551 /* 1552 * /proc/pid/pagemap - an array mapping virtual pages to pfns 1553 * 1554 * For each page in the address space, this file contains one 64-bit entry 1555 * consisting of the following: 1556 * 1557 * Bits 0-54 page frame number (PFN) if present 1558 * Bits 0-4 swap type if swapped 1559 * Bits 5-54 swap offset if swapped 1560 * Bit 55 pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst) 1561 * Bit 56 page exclusively mapped 1562 * Bits 57-60 zero 1563 * Bit 61 page is file-page or shared-anon 1564 * Bit 62 page swapped 1565 * Bit 63 page present 1566 * 1567 * If the page is not present but in swap, then the PFN contains an 1568 * encoding of the swap file number and the page's offset into the 1569 * swap. Unmapped pages return a null PFN. This allows determining 1570 * precisely which pages are mapped (or in swap) and comparing mapped 1571 * pages between processes. 1572 * 1573 * Efficient users of this interface will use /proc/pid/maps to 1574 * determine which areas of memory are actually mapped and llseek to 1575 * skip over unmapped regions. 1576 */ 1577 static ssize_t pagemap_read(struct file *file, char __user *buf, 1578 size_t count, loff_t *ppos) 1579 { 1580 struct mm_struct *mm = file->private_data; 1581 struct pagemapread pm; 1582 unsigned long src; 1583 unsigned long svpfn; 1584 unsigned long start_vaddr; 1585 unsigned long end_vaddr; 1586 int ret = 0, copied = 0; 1587 1588 if (!mm || !mmget_not_zero(mm)) 1589 goto out; 1590 1591 ret = -EINVAL; 1592 /* file position must be aligned */ 1593 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES)) 1594 goto out_mm; 1595 1596 ret = 0; 1597 if (!count) 1598 goto out_mm; 1599 1600 /* do not disclose physical addresses: attack vector */ 1601 pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN); 1602 1603 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT); 1604 pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL); 1605 ret = -ENOMEM; 1606 if (!pm.buffer) 1607 goto out_mm; 1608 1609 src = *ppos; 1610 svpfn = src / PM_ENTRY_BYTES; 1611 end_vaddr = mm->task_size; 1612 1613 /* watch out for wraparound */ 1614 start_vaddr = end_vaddr; 1615 if (svpfn <= (ULONG_MAX >> PAGE_SHIFT)) 1616 start_vaddr = untagged_addr(svpfn << PAGE_SHIFT); 1617 1618 /* Ensure the address is inside the task */ 1619 if (start_vaddr > mm->task_size) 1620 start_vaddr = end_vaddr; 1621 1622 /* 1623 * The odds are that this will stop walking way 1624 * before end_vaddr, because the length of the 1625 * user buffer is tracked in "pm", and the walk 1626 * will stop when we hit the end of the buffer. 1627 */ 1628 ret = 0; 1629 while (count && (start_vaddr < end_vaddr)) { 1630 int len; 1631 unsigned long end; 1632 1633 pm.pos = 0; 1634 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK; 1635 /* overflow ? */ 1636 if (end < start_vaddr || end > end_vaddr) 1637 end = end_vaddr; 1638 ret = mmap_read_lock_killable(mm); 1639 if (ret) 1640 goto out_free; 1641 ret = walk_page_range(mm, start_vaddr, end, &pagemap_ops, &pm); 1642 mmap_read_unlock(mm); 1643 start_vaddr = end; 1644 1645 len = min(count, PM_ENTRY_BYTES * pm.pos); 1646 if (copy_to_user(buf, pm.buffer, len)) { 1647 ret = -EFAULT; 1648 goto out_free; 1649 } 1650 copied += len; 1651 buf += len; 1652 count -= len; 1653 } 1654 *ppos += copied; 1655 if (!ret || ret == PM_END_OF_BUFFER) 1656 ret = copied; 1657 1658 out_free: 1659 kfree(pm.buffer); 1660 out_mm: 1661 mmput(mm); 1662 out: 1663 return ret; 1664 } 1665 1666 static int pagemap_open(struct inode *inode, struct file *file) 1667 { 1668 struct mm_struct *mm; 1669 1670 mm = proc_mem_open(inode, PTRACE_MODE_READ); 1671 if (IS_ERR(mm)) 1672 return PTR_ERR(mm); 1673 file->private_data = mm; 1674 return 0; 1675 } 1676 1677 static int pagemap_release(struct inode *inode, struct file *file) 1678 { 1679 struct mm_struct *mm = file->private_data; 1680 1681 if (mm) 1682 mmdrop(mm); 1683 return 0; 1684 } 1685 1686 const struct file_operations proc_pagemap_operations = { 1687 .llseek = mem_lseek, /* borrow this */ 1688 .read = pagemap_read, 1689 .open = pagemap_open, 1690 .release = pagemap_release, 1691 }; 1692 #endif /* CONFIG_PROC_PAGE_MONITOR */ 1693 1694 #ifdef CONFIG_NUMA 1695 1696 struct numa_maps { 1697 unsigned long pages; 1698 unsigned long anon; 1699 unsigned long active; 1700 unsigned long writeback; 1701 unsigned long mapcount_max; 1702 unsigned long dirty; 1703 unsigned long swapcache; 1704 unsigned long node[MAX_NUMNODES]; 1705 }; 1706 1707 struct numa_maps_private { 1708 struct proc_maps_private proc_maps; 1709 struct numa_maps md; 1710 }; 1711 1712 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty, 1713 unsigned long nr_pages) 1714 { 1715 int count = page_mapcount(page); 1716 1717 md->pages += nr_pages; 1718 if (pte_dirty || PageDirty(page)) 1719 md->dirty += nr_pages; 1720 1721 if (PageSwapCache(page)) 1722 md->swapcache += nr_pages; 1723 1724 if (PageActive(page) || PageUnevictable(page)) 1725 md->active += nr_pages; 1726 1727 if (PageWriteback(page)) 1728 md->writeback += nr_pages; 1729 1730 if (PageAnon(page)) 1731 md->anon += nr_pages; 1732 1733 if (count > md->mapcount_max) 1734 md->mapcount_max = count; 1735 1736 md->node[page_to_nid(page)] += nr_pages; 1737 } 1738 1739 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma, 1740 unsigned long addr) 1741 { 1742 struct page *page; 1743 int nid; 1744 1745 if (!pte_present(pte)) 1746 return NULL; 1747 1748 page = vm_normal_page(vma, addr, pte); 1749 if (!page) 1750 return NULL; 1751 1752 if (PageReserved(page)) 1753 return NULL; 1754 1755 nid = page_to_nid(page); 1756 if (!node_isset(nid, node_states[N_MEMORY])) 1757 return NULL; 1758 1759 return page; 1760 } 1761 1762 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1763 static struct page *can_gather_numa_stats_pmd(pmd_t pmd, 1764 struct vm_area_struct *vma, 1765 unsigned long addr) 1766 { 1767 struct page *page; 1768 int nid; 1769 1770 if (!pmd_present(pmd)) 1771 return NULL; 1772 1773 page = vm_normal_page_pmd(vma, addr, pmd); 1774 if (!page) 1775 return NULL; 1776 1777 if (PageReserved(page)) 1778 return NULL; 1779 1780 nid = page_to_nid(page); 1781 if (!node_isset(nid, node_states[N_MEMORY])) 1782 return NULL; 1783 1784 return page; 1785 } 1786 #endif 1787 1788 static int gather_pte_stats(pmd_t *pmd, unsigned long addr, 1789 unsigned long end, struct mm_walk *walk) 1790 { 1791 struct numa_maps *md = walk->private; 1792 struct vm_area_struct *vma = walk->vma; 1793 spinlock_t *ptl; 1794 pte_t *orig_pte; 1795 pte_t *pte; 1796 1797 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1798 ptl = pmd_trans_huge_lock(pmd, vma); 1799 if (ptl) { 1800 struct page *page; 1801 1802 page = can_gather_numa_stats_pmd(*pmd, vma, addr); 1803 if (page) 1804 gather_stats(page, md, pmd_dirty(*pmd), 1805 HPAGE_PMD_SIZE/PAGE_SIZE); 1806 spin_unlock(ptl); 1807 return 0; 1808 } 1809 1810 if (pmd_trans_unstable(pmd)) 1811 return 0; 1812 #endif 1813 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl); 1814 do { 1815 struct page *page = can_gather_numa_stats(*pte, vma, addr); 1816 if (!page) 1817 continue; 1818 gather_stats(page, md, pte_dirty(*pte), 1); 1819 1820 } while (pte++, addr += PAGE_SIZE, addr != end); 1821 pte_unmap_unlock(orig_pte, ptl); 1822 cond_resched(); 1823 return 0; 1824 } 1825 #ifdef CONFIG_HUGETLB_PAGE 1826 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask, 1827 unsigned long addr, unsigned long end, struct mm_walk *walk) 1828 { 1829 pte_t huge_pte = huge_ptep_get(pte); 1830 struct numa_maps *md; 1831 struct page *page; 1832 1833 if (!pte_present(huge_pte)) 1834 return 0; 1835 1836 page = pte_page(huge_pte); 1837 if (!page) 1838 return 0; 1839 1840 md = walk->private; 1841 gather_stats(page, md, pte_dirty(huge_pte), 1); 1842 return 0; 1843 } 1844 1845 #else 1846 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask, 1847 unsigned long addr, unsigned long end, struct mm_walk *walk) 1848 { 1849 return 0; 1850 } 1851 #endif 1852 1853 static const struct mm_walk_ops show_numa_ops = { 1854 .hugetlb_entry = gather_hugetlb_stats, 1855 .pmd_entry = gather_pte_stats, 1856 }; 1857 1858 /* 1859 * Display pages allocated per node and memory policy via /proc. 1860 */ 1861 static int show_numa_map(struct seq_file *m, void *v) 1862 { 1863 struct numa_maps_private *numa_priv = m->private; 1864 struct proc_maps_private *proc_priv = &numa_priv->proc_maps; 1865 struct vm_area_struct *vma = v; 1866 struct numa_maps *md = &numa_priv->md; 1867 struct file *file = vma->vm_file; 1868 struct mm_struct *mm = vma->vm_mm; 1869 struct mempolicy *pol; 1870 char buffer[64]; 1871 int nid; 1872 1873 if (!mm) 1874 return 0; 1875 1876 /* Ensure we start with an empty set of numa_maps statistics. */ 1877 memset(md, 0, sizeof(*md)); 1878 1879 pol = __get_vma_policy(vma, vma->vm_start); 1880 if (pol) { 1881 mpol_to_str(buffer, sizeof(buffer), pol); 1882 mpol_cond_put(pol); 1883 } else { 1884 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy); 1885 } 1886 1887 seq_printf(m, "%08lx %s", vma->vm_start, buffer); 1888 1889 if (file) { 1890 seq_puts(m, " file="); 1891 seq_file_path(m, file, "\n\t= "); 1892 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) { 1893 seq_puts(m, " heap"); 1894 } else if (is_stack(vma)) { 1895 seq_puts(m, " stack"); 1896 } 1897 1898 if (is_vm_hugetlb_page(vma)) 1899 seq_puts(m, " huge"); 1900 1901 /* mmap_lock is held by m_start */ 1902 walk_page_vma(vma, &show_numa_ops, md); 1903 1904 if (!md->pages) 1905 goto out; 1906 1907 if (md->anon) 1908 seq_printf(m, " anon=%lu", md->anon); 1909 1910 if (md->dirty) 1911 seq_printf(m, " dirty=%lu", md->dirty); 1912 1913 if (md->pages != md->anon && md->pages != md->dirty) 1914 seq_printf(m, " mapped=%lu", md->pages); 1915 1916 if (md->mapcount_max > 1) 1917 seq_printf(m, " mapmax=%lu", md->mapcount_max); 1918 1919 if (md->swapcache) 1920 seq_printf(m, " swapcache=%lu", md->swapcache); 1921 1922 if (md->active < md->pages && !is_vm_hugetlb_page(vma)) 1923 seq_printf(m, " active=%lu", md->active); 1924 1925 if (md->writeback) 1926 seq_printf(m, " writeback=%lu", md->writeback); 1927 1928 for_each_node_state(nid, N_MEMORY) 1929 if (md->node[nid]) 1930 seq_printf(m, " N%d=%lu", nid, md->node[nid]); 1931 1932 seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10); 1933 out: 1934 seq_putc(m, '\n'); 1935 return 0; 1936 } 1937 1938 static const struct seq_operations proc_pid_numa_maps_op = { 1939 .start = m_start, 1940 .next = m_next, 1941 .stop = m_stop, 1942 .show = show_numa_map, 1943 }; 1944 1945 static int pid_numa_maps_open(struct inode *inode, struct file *file) 1946 { 1947 return proc_maps_open(inode, file, &proc_pid_numa_maps_op, 1948 sizeof(struct numa_maps_private)); 1949 } 1950 1951 const struct file_operations proc_pid_numa_maps_operations = { 1952 .open = pid_numa_maps_open, 1953 .read = seq_read, 1954 .llseek = seq_lseek, 1955 .release = proc_map_release, 1956 }; 1957 1958 #endif /* CONFIG_NUMA */ 1959