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 }; 665 size_t i; 666 667 seq_puts(m, "VmFlags: "); 668 for (i = 0; i < BITS_PER_LONG; i++) { 669 if (!mnemonics[i][0]) 670 continue; 671 if (vma->vm_flags & (1UL << i)) { 672 seq_putc(m, mnemonics[i][0]); 673 seq_putc(m, mnemonics[i][1]); 674 seq_putc(m, ' '); 675 } 676 } 677 seq_putc(m, '\n'); 678 } 679 680 #ifdef CONFIG_HUGETLB_PAGE 681 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask, 682 unsigned long addr, unsigned long end, 683 struct mm_walk *walk) 684 { 685 struct mem_size_stats *mss = walk->private; 686 struct vm_area_struct *vma = walk->vma; 687 struct page *page = NULL; 688 689 if (pte_present(*pte)) { 690 page = vm_normal_page(vma, addr, *pte); 691 } else if (is_swap_pte(*pte)) { 692 swp_entry_t swpent = pte_to_swp_entry(*pte); 693 694 if (is_migration_entry(swpent)) 695 page = migration_entry_to_page(swpent); 696 else if (is_device_private_entry(swpent)) 697 page = device_private_entry_to_page(swpent); 698 } 699 if (page) { 700 int mapcount = page_mapcount(page); 701 702 if (mapcount >= 2) 703 mss->shared_hugetlb += huge_page_size(hstate_vma(vma)); 704 else 705 mss->private_hugetlb += huge_page_size(hstate_vma(vma)); 706 } 707 return 0; 708 } 709 #else 710 #define smaps_hugetlb_range NULL 711 #endif /* HUGETLB_PAGE */ 712 713 static const struct mm_walk_ops smaps_walk_ops = { 714 .pmd_entry = smaps_pte_range, 715 .hugetlb_entry = smaps_hugetlb_range, 716 }; 717 718 static const struct mm_walk_ops smaps_shmem_walk_ops = { 719 .pmd_entry = smaps_pte_range, 720 .hugetlb_entry = smaps_hugetlb_range, 721 .pte_hole = smaps_pte_hole, 722 }; 723 724 /* 725 * Gather mem stats from @vma with the indicated beginning 726 * address @start, and keep them in @mss. 727 * 728 * Use vm_start of @vma as the beginning address if @start is 0. 729 */ 730 static void smap_gather_stats(struct vm_area_struct *vma, 731 struct mem_size_stats *mss, unsigned long start) 732 { 733 const struct mm_walk_ops *ops = &smaps_walk_ops; 734 735 /* Invalid start */ 736 if (start >= vma->vm_end) 737 return; 738 739 #ifdef CONFIG_SHMEM 740 /* In case of smaps_rollup, reset the value from previous vma */ 741 mss->check_shmem_swap = false; 742 if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) { 743 /* 744 * For shared or readonly shmem mappings we know that all 745 * swapped out pages belong to the shmem object, and we can 746 * obtain the swap value much more efficiently. For private 747 * writable mappings, we might have COW pages that are 748 * not affected by the parent swapped out pages of the shmem 749 * object, so we have to distinguish them during the page walk. 750 * Unless we know that the shmem object (or the part mapped by 751 * our VMA) has no swapped out pages at all. 752 */ 753 unsigned long shmem_swapped = shmem_swap_usage(vma); 754 755 if (!start && (!shmem_swapped || (vma->vm_flags & VM_SHARED) || 756 !(vma->vm_flags & VM_WRITE))) { 757 mss->swap += shmem_swapped; 758 } else { 759 mss->check_shmem_swap = true; 760 ops = &smaps_shmem_walk_ops; 761 } 762 } 763 #endif 764 /* mmap_lock is held in m_start */ 765 if (!start) 766 walk_page_vma(vma, ops, mss); 767 else 768 walk_page_range(vma->vm_mm, start, vma->vm_end, ops, mss); 769 } 770 771 #define SEQ_PUT_DEC(str, val) \ 772 seq_put_decimal_ull_width(m, str, (val) >> 10, 8) 773 774 /* Show the contents common for smaps and smaps_rollup */ 775 static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss, 776 bool rollup_mode) 777 { 778 SEQ_PUT_DEC("Rss: ", mss->resident); 779 SEQ_PUT_DEC(" kB\nPss: ", mss->pss >> PSS_SHIFT); 780 if (rollup_mode) { 781 /* 782 * These are meaningful only for smaps_rollup, otherwise two of 783 * them are zero, and the other one is the same as Pss. 784 */ 785 SEQ_PUT_DEC(" kB\nPss_Anon: ", 786 mss->pss_anon >> PSS_SHIFT); 787 SEQ_PUT_DEC(" kB\nPss_File: ", 788 mss->pss_file >> PSS_SHIFT); 789 SEQ_PUT_DEC(" kB\nPss_Shmem: ", 790 mss->pss_shmem >> PSS_SHIFT); 791 } 792 SEQ_PUT_DEC(" kB\nShared_Clean: ", mss->shared_clean); 793 SEQ_PUT_DEC(" kB\nShared_Dirty: ", mss->shared_dirty); 794 SEQ_PUT_DEC(" kB\nPrivate_Clean: ", mss->private_clean); 795 SEQ_PUT_DEC(" kB\nPrivate_Dirty: ", mss->private_dirty); 796 SEQ_PUT_DEC(" kB\nReferenced: ", mss->referenced); 797 SEQ_PUT_DEC(" kB\nAnonymous: ", mss->anonymous); 798 SEQ_PUT_DEC(" kB\nLazyFree: ", mss->lazyfree); 799 SEQ_PUT_DEC(" kB\nAnonHugePages: ", mss->anonymous_thp); 800 SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp); 801 SEQ_PUT_DEC(" kB\nFilePmdMapped: ", mss->file_thp); 802 SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb); 803 seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ", 804 mss->private_hugetlb >> 10, 7); 805 SEQ_PUT_DEC(" kB\nSwap: ", mss->swap); 806 SEQ_PUT_DEC(" kB\nSwapPss: ", 807 mss->swap_pss >> PSS_SHIFT); 808 SEQ_PUT_DEC(" kB\nLocked: ", 809 mss->pss_locked >> PSS_SHIFT); 810 seq_puts(m, " kB\n"); 811 } 812 813 static int show_smap(struct seq_file *m, void *v) 814 { 815 struct vm_area_struct *vma = v; 816 struct mem_size_stats mss; 817 818 memset(&mss, 0, sizeof(mss)); 819 820 smap_gather_stats(vma, &mss, 0); 821 822 show_map_vma(m, vma); 823 824 SEQ_PUT_DEC("Size: ", vma->vm_end - vma->vm_start); 825 SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma)); 826 SEQ_PUT_DEC(" kB\nMMUPageSize: ", vma_mmu_pagesize(vma)); 827 seq_puts(m, " kB\n"); 828 829 __show_smap(m, &mss, false); 830 831 seq_printf(m, "THPeligible: %d\n", 832 transparent_hugepage_enabled(vma)); 833 834 if (arch_pkeys_enabled()) 835 seq_printf(m, "ProtectionKey: %8u\n", vma_pkey(vma)); 836 show_smap_vma_flags(m, vma); 837 838 return 0; 839 } 840 841 static int show_smaps_rollup(struct seq_file *m, void *v) 842 { 843 struct proc_maps_private *priv = m->private; 844 struct mem_size_stats mss; 845 struct mm_struct *mm; 846 struct vm_area_struct *vma; 847 unsigned long last_vma_end = 0; 848 int ret = 0; 849 850 priv->task = get_proc_task(priv->inode); 851 if (!priv->task) 852 return -ESRCH; 853 854 mm = priv->mm; 855 if (!mm || !mmget_not_zero(mm)) { 856 ret = -ESRCH; 857 goto out_put_task; 858 } 859 860 memset(&mss, 0, sizeof(mss)); 861 862 ret = mmap_read_lock_killable(mm); 863 if (ret) 864 goto out_put_mm; 865 866 hold_task_mempolicy(priv); 867 868 for (vma = priv->mm->mmap; vma;) { 869 smap_gather_stats(vma, &mss, 0); 870 last_vma_end = vma->vm_end; 871 872 /* 873 * Release mmap_lock temporarily if someone wants to 874 * access it for write request. 875 */ 876 if (mmap_lock_is_contended(mm)) { 877 mmap_read_unlock(mm); 878 ret = mmap_read_lock_killable(mm); 879 if (ret) { 880 release_task_mempolicy(priv); 881 goto out_put_mm; 882 } 883 884 /* 885 * After dropping the lock, there are four cases to 886 * consider. See the following example for explanation. 887 * 888 * +------+------+-----------+ 889 * | VMA1 | VMA2 | VMA3 | 890 * +------+------+-----------+ 891 * | | | | 892 * 4k 8k 16k 400k 893 * 894 * Suppose we drop the lock after reading VMA2 due to 895 * contention, then we get: 896 * 897 * last_vma_end = 16k 898 * 899 * 1) VMA2 is freed, but VMA3 exists: 900 * 901 * find_vma(mm, 16k - 1) will return VMA3. 902 * In this case, just continue from VMA3. 903 * 904 * 2) VMA2 still exists: 905 * 906 * find_vma(mm, 16k - 1) will return VMA2. 907 * Iterate the loop like the original one. 908 * 909 * 3) No more VMAs can be found: 910 * 911 * find_vma(mm, 16k - 1) will return NULL. 912 * No more things to do, just break. 913 * 914 * 4) (last_vma_end - 1) is the middle of a vma (VMA'): 915 * 916 * find_vma(mm, 16k - 1) will return VMA' whose range 917 * contains last_vma_end. 918 * Iterate VMA' from last_vma_end. 919 */ 920 vma = find_vma(mm, last_vma_end - 1); 921 /* Case 3 above */ 922 if (!vma) 923 break; 924 925 /* Case 1 above */ 926 if (vma->vm_start >= last_vma_end) 927 continue; 928 929 /* Case 4 above */ 930 if (vma->vm_end > last_vma_end) 931 smap_gather_stats(vma, &mss, last_vma_end); 932 } 933 /* Case 2 above */ 934 vma = vma->vm_next; 935 } 936 937 show_vma_header_prefix(m, priv->mm->mmap->vm_start, 938 last_vma_end, 0, 0, 0, 0); 939 seq_pad(m, ' '); 940 seq_puts(m, "[rollup]\n"); 941 942 __show_smap(m, &mss, true); 943 944 release_task_mempolicy(priv); 945 mmap_read_unlock(mm); 946 947 out_put_mm: 948 mmput(mm); 949 out_put_task: 950 put_task_struct(priv->task); 951 priv->task = NULL; 952 953 return ret; 954 } 955 #undef SEQ_PUT_DEC 956 957 static const struct seq_operations proc_pid_smaps_op = { 958 .start = m_start, 959 .next = m_next, 960 .stop = m_stop, 961 .show = show_smap 962 }; 963 964 static int pid_smaps_open(struct inode *inode, struct file *file) 965 { 966 return do_maps_open(inode, file, &proc_pid_smaps_op); 967 } 968 969 static int smaps_rollup_open(struct inode *inode, struct file *file) 970 { 971 int ret; 972 struct proc_maps_private *priv; 973 974 priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT); 975 if (!priv) 976 return -ENOMEM; 977 978 ret = single_open(file, show_smaps_rollup, priv); 979 if (ret) 980 goto out_free; 981 982 priv->inode = inode; 983 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ); 984 if (IS_ERR(priv->mm)) { 985 ret = PTR_ERR(priv->mm); 986 987 single_release(inode, file); 988 goto out_free; 989 } 990 991 return 0; 992 993 out_free: 994 kfree(priv); 995 return ret; 996 } 997 998 static int smaps_rollup_release(struct inode *inode, struct file *file) 999 { 1000 struct seq_file *seq = file->private_data; 1001 struct proc_maps_private *priv = seq->private; 1002 1003 if (priv->mm) 1004 mmdrop(priv->mm); 1005 1006 kfree(priv); 1007 return single_release(inode, file); 1008 } 1009 1010 const struct file_operations proc_pid_smaps_operations = { 1011 .open = pid_smaps_open, 1012 .read = seq_read, 1013 .llseek = seq_lseek, 1014 .release = proc_map_release, 1015 }; 1016 1017 const struct file_operations proc_pid_smaps_rollup_operations = { 1018 .open = smaps_rollup_open, 1019 .read = seq_read, 1020 .llseek = seq_lseek, 1021 .release = smaps_rollup_release, 1022 }; 1023 1024 enum clear_refs_types { 1025 CLEAR_REFS_ALL = 1, 1026 CLEAR_REFS_ANON, 1027 CLEAR_REFS_MAPPED, 1028 CLEAR_REFS_SOFT_DIRTY, 1029 CLEAR_REFS_MM_HIWATER_RSS, 1030 CLEAR_REFS_LAST, 1031 }; 1032 1033 struct clear_refs_private { 1034 enum clear_refs_types type; 1035 }; 1036 1037 #ifdef CONFIG_MEM_SOFT_DIRTY 1038 1039 static inline bool pte_is_pinned(struct vm_area_struct *vma, unsigned long addr, pte_t pte) 1040 { 1041 struct page *page; 1042 1043 if (!pte_write(pte)) 1044 return false; 1045 if (!is_cow_mapping(vma->vm_flags)) 1046 return false; 1047 if (likely(!atomic_read(&vma->vm_mm->has_pinned))) 1048 return false; 1049 page = vm_normal_page(vma, addr, pte); 1050 if (!page) 1051 return false; 1052 return page_maybe_dma_pinned(page); 1053 } 1054 1055 static inline void clear_soft_dirty(struct vm_area_struct *vma, 1056 unsigned long addr, pte_t *pte) 1057 { 1058 /* 1059 * The soft-dirty tracker uses #PF-s to catch writes 1060 * to pages, so write-protect the pte as well. See the 1061 * Documentation/admin-guide/mm/soft-dirty.rst for full description 1062 * of how soft-dirty works. 1063 */ 1064 pte_t ptent = *pte; 1065 1066 if (pte_present(ptent)) { 1067 pte_t old_pte; 1068 1069 if (pte_is_pinned(vma, addr, ptent)) 1070 return; 1071 old_pte = ptep_modify_prot_start(vma, addr, pte); 1072 ptent = pte_wrprotect(old_pte); 1073 ptent = pte_clear_soft_dirty(ptent); 1074 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent); 1075 } else if (is_swap_pte(ptent)) { 1076 ptent = pte_swp_clear_soft_dirty(ptent); 1077 set_pte_at(vma->vm_mm, addr, pte, ptent); 1078 } 1079 } 1080 #else 1081 static inline void clear_soft_dirty(struct vm_area_struct *vma, 1082 unsigned long addr, pte_t *pte) 1083 { 1084 } 1085 #endif 1086 1087 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE) 1088 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma, 1089 unsigned long addr, pmd_t *pmdp) 1090 { 1091 pmd_t old, pmd = *pmdp; 1092 1093 if (pmd_present(pmd)) { 1094 /* See comment in change_huge_pmd() */ 1095 old = pmdp_invalidate(vma, addr, pmdp); 1096 if (pmd_dirty(old)) 1097 pmd = pmd_mkdirty(pmd); 1098 if (pmd_young(old)) 1099 pmd = pmd_mkyoung(pmd); 1100 1101 pmd = pmd_wrprotect(pmd); 1102 pmd = pmd_clear_soft_dirty(pmd); 1103 1104 set_pmd_at(vma->vm_mm, addr, pmdp, pmd); 1105 } else if (is_migration_entry(pmd_to_swp_entry(pmd))) { 1106 pmd = pmd_swp_clear_soft_dirty(pmd); 1107 set_pmd_at(vma->vm_mm, addr, pmdp, pmd); 1108 } 1109 } 1110 #else 1111 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma, 1112 unsigned long addr, pmd_t *pmdp) 1113 { 1114 } 1115 #endif 1116 1117 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr, 1118 unsigned long end, struct mm_walk *walk) 1119 { 1120 struct clear_refs_private *cp = walk->private; 1121 struct vm_area_struct *vma = walk->vma; 1122 pte_t *pte, ptent; 1123 spinlock_t *ptl; 1124 struct page *page; 1125 1126 ptl = pmd_trans_huge_lock(pmd, vma); 1127 if (ptl) { 1128 if (cp->type == CLEAR_REFS_SOFT_DIRTY) { 1129 clear_soft_dirty_pmd(vma, addr, pmd); 1130 goto out; 1131 } 1132 1133 if (!pmd_present(*pmd)) 1134 goto out; 1135 1136 page = pmd_page(*pmd); 1137 1138 /* Clear accessed and referenced bits. */ 1139 pmdp_test_and_clear_young(vma, addr, pmd); 1140 test_and_clear_page_young(page); 1141 ClearPageReferenced(page); 1142 out: 1143 spin_unlock(ptl); 1144 return 0; 1145 } 1146 1147 if (pmd_trans_unstable(pmd)) 1148 return 0; 1149 1150 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); 1151 for (; addr != end; pte++, addr += PAGE_SIZE) { 1152 ptent = *pte; 1153 1154 if (cp->type == CLEAR_REFS_SOFT_DIRTY) { 1155 clear_soft_dirty(vma, addr, pte); 1156 continue; 1157 } 1158 1159 if (!pte_present(ptent)) 1160 continue; 1161 1162 page = vm_normal_page(vma, addr, ptent); 1163 if (!page) 1164 continue; 1165 1166 /* Clear accessed and referenced bits. */ 1167 ptep_test_and_clear_young(vma, addr, pte); 1168 test_and_clear_page_young(page); 1169 ClearPageReferenced(page); 1170 } 1171 pte_unmap_unlock(pte - 1, ptl); 1172 cond_resched(); 1173 return 0; 1174 } 1175 1176 static int clear_refs_test_walk(unsigned long start, unsigned long end, 1177 struct mm_walk *walk) 1178 { 1179 struct clear_refs_private *cp = walk->private; 1180 struct vm_area_struct *vma = walk->vma; 1181 1182 if (vma->vm_flags & VM_PFNMAP) 1183 return 1; 1184 1185 /* 1186 * Writing 1 to /proc/pid/clear_refs affects all pages. 1187 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages. 1188 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages. 1189 * Writing 4 to /proc/pid/clear_refs affects all pages. 1190 */ 1191 if (cp->type == CLEAR_REFS_ANON && vma->vm_file) 1192 return 1; 1193 if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file) 1194 return 1; 1195 return 0; 1196 } 1197 1198 static const struct mm_walk_ops clear_refs_walk_ops = { 1199 .pmd_entry = clear_refs_pte_range, 1200 .test_walk = clear_refs_test_walk, 1201 }; 1202 1203 static ssize_t clear_refs_write(struct file *file, const char __user *buf, 1204 size_t count, loff_t *ppos) 1205 { 1206 struct task_struct *task; 1207 char buffer[PROC_NUMBUF]; 1208 struct mm_struct *mm; 1209 struct vm_area_struct *vma; 1210 enum clear_refs_types type; 1211 int itype; 1212 int rv; 1213 1214 memset(buffer, 0, sizeof(buffer)); 1215 if (count > sizeof(buffer) - 1) 1216 count = sizeof(buffer) - 1; 1217 if (copy_from_user(buffer, buf, count)) 1218 return -EFAULT; 1219 rv = kstrtoint(strstrip(buffer), 10, &itype); 1220 if (rv < 0) 1221 return rv; 1222 type = (enum clear_refs_types)itype; 1223 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST) 1224 return -EINVAL; 1225 1226 task = get_proc_task(file_inode(file)); 1227 if (!task) 1228 return -ESRCH; 1229 mm = get_task_mm(task); 1230 if (mm) { 1231 struct mmu_notifier_range range; 1232 struct clear_refs_private cp = { 1233 .type = type, 1234 }; 1235 1236 if (mmap_write_lock_killable(mm)) { 1237 count = -EINTR; 1238 goto out_mm; 1239 } 1240 if (type == CLEAR_REFS_MM_HIWATER_RSS) { 1241 /* 1242 * Writing 5 to /proc/pid/clear_refs resets the peak 1243 * resident set size to this mm's current rss value. 1244 */ 1245 reset_mm_hiwater_rss(mm); 1246 goto out_unlock; 1247 } 1248 1249 if (type == CLEAR_REFS_SOFT_DIRTY) { 1250 for (vma = mm->mmap; vma; vma = vma->vm_next) { 1251 if (!(vma->vm_flags & VM_SOFTDIRTY)) 1252 continue; 1253 vma->vm_flags &= ~VM_SOFTDIRTY; 1254 vma_set_page_prot(vma); 1255 } 1256 1257 inc_tlb_flush_pending(mm); 1258 mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY, 1259 0, NULL, mm, 0, -1UL); 1260 mmu_notifier_invalidate_range_start(&range); 1261 } 1262 walk_page_range(mm, 0, mm->highest_vm_end, &clear_refs_walk_ops, 1263 &cp); 1264 if (type == CLEAR_REFS_SOFT_DIRTY) { 1265 mmu_notifier_invalidate_range_end(&range); 1266 flush_tlb_mm(mm); 1267 dec_tlb_flush_pending(mm); 1268 } 1269 out_unlock: 1270 mmap_write_unlock(mm); 1271 out_mm: 1272 mmput(mm); 1273 } 1274 put_task_struct(task); 1275 1276 return count; 1277 } 1278 1279 const struct file_operations proc_clear_refs_operations = { 1280 .write = clear_refs_write, 1281 .llseek = noop_llseek, 1282 }; 1283 1284 typedef struct { 1285 u64 pme; 1286 } pagemap_entry_t; 1287 1288 struct pagemapread { 1289 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */ 1290 pagemap_entry_t *buffer; 1291 bool show_pfn; 1292 }; 1293 1294 #define PAGEMAP_WALK_SIZE (PMD_SIZE) 1295 #define PAGEMAP_WALK_MASK (PMD_MASK) 1296 1297 #define PM_ENTRY_BYTES sizeof(pagemap_entry_t) 1298 #define PM_PFRAME_BITS 55 1299 #define PM_PFRAME_MASK GENMASK_ULL(PM_PFRAME_BITS - 1, 0) 1300 #define PM_SOFT_DIRTY BIT_ULL(55) 1301 #define PM_MMAP_EXCLUSIVE BIT_ULL(56) 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 } else if (is_swap_pte(pte)) { 1376 swp_entry_t entry; 1377 if (pte_swp_soft_dirty(pte)) 1378 flags |= PM_SOFT_DIRTY; 1379 entry = pte_to_swp_entry(pte); 1380 if (pm->show_pfn) 1381 frame = swp_type(entry) | 1382 (swp_offset(entry) << MAX_SWAPFILES_SHIFT); 1383 flags |= PM_SWAP; 1384 if (is_migration_entry(entry)) 1385 page = migration_entry_to_page(entry); 1386 1387 if (is_device_private_entry(entry)) 1388 page = device_private_entry_to_page(entry); 1389 } 1390 1391 if (page && !PageAnon(page)) 1392 flags |= PM_FILE; 1393 if (page && page_mapcount(page) == 1) 1394 flags |= PM_MMAP_EXCLUSIVE; 1395 if (vma->vm_flags & VM_SOFTDIRTY) 1396 flags |= PM_SOFT_DIRTY; 1397 1398 return make_pme(frame, flags); 1399 } 1400 1401 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end, 1402 struct mm_walk *walk) 1403 { 1404 struct vm_area_struct *vma = walk->vma; 1405 struct pagemapread *pm = walk->private; 1406 spinlock_t *ptl; 1407 pte_t *pte, *orig_pte; 1408 int err = 0; 1409 1410 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1411 ptl = pmd_trans_huge_lock(pmdp, vma); 1412 if (ptl) { 1413 u64 flags = 0, frame = 0; 1414 pmd_t pmd = *pmdp; 1415 struct page *page = NULL; 1416 1417 if (vma->vm_flags & VM_SOFTDIRTY) 1418 flags |= PM_SOFT_DIRTY; 1419 1420 if (pmd_present(pmd)) { 1421 page = pmd_page(pmd); 1422 1423 flags |= PM_PRESENT; 1424 if (pmd_soft_dirty(pmd)) 1425 flags |= PM_SOFT_DIRTY; 1426 if (pm->show_pfn) 1427 frame = pmd_pfn(pmd) + 1428 ((addr & ~PMD_MASK) >> PAGE_SHIFT); 1429 } 1430 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION 1431 else if (is_swap_pmd(pmd)) { 1432 swp_entry_t entry = pmd_to_swp_entry(pmd); 1433 unsigned long offset; 1434 1435 if (pm->show_pfn) { 1436 offset = swp_offset(entry) + 1437 ((addr & ~PMD_MASK) >> PAGE_SHIFT); 1438 frame = swp_type(entry) | 1439 (offset << MAX_SWAPFILES_SHIFT); 1440 } 1441 flags |= PM_SWAP; 1442 if (pmd_swp_soft_dirty(pmd)) 1443 flags |= PM_SOFT_DIRTY; 1444 VM_BUG_ON(!is_pmd_migration_entry(pmd)); 1445 page = migration_entry_to_page(entry); 1446 } 1447 #endif 1448 1449 if (page && page_mapcount(page) == 1) 1450 flags |= PM_MMAP_EXCLUSIVE; 1451 1452 for (; addr != end; addr += PAGE_SIZE) { 1453 pagemap_entry_t pme = make_pme(frame, flags); 1454 1455 err = add_to_pagemap(addr, &pme, pm); 1456 if (err) 1457 break; 1458 if (pm->show_pfn) { 1459 if (flags & PM_PRESENT) 1460 frame++; 1461 else if (flags & PM_SWAP) 1462 frame += (1 << MAX_SWAPFILES_SHIFT); 1463 } 1464 } 1465 spin_unlock(ptl); 1466 return err; 1467 } 1468 1469 if (pmd_trans_unstable(pmdp)) 1470 return 0; 1471 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ 1472 1473 /* 1474 * We can assume that @vma always points to a valid one and @end never 1475 * goes beyond vma->vm_end. 1476 */ 1477 orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl); 1478 for (; addr < end; pte++, addr += PAGE_SIZE) { 1479 pagemap_entry_t pme; 1480 1481 pme = pte_to_pagemap_entry(pm, vma, addr, *pte); 1482 err = add_to_pagemap(addr, &pme, pm); 1483 if (err) 1484 break; 1485 } 1486 pte_unmap_unlock(orig_pte, ptl); 1487 1488 cond_resched(); 1489 1490 return err; 1491 } 1492 1493 #ifdef CONFIG_HUGETLB_PAGE 1494 /* This function walks within one hugetlb entry in the single call */ 1495 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask, 1496 unsigned long addr, unsigned long end, 1497 struct mm_walk *walk) 1498 { 1499 struct pagemapread *pm = walk->private; 1500 struct vm_area_struct *vma = walk->vma; 1501 u64 flags = 0, frame = 0; 1502 int err = 0; 1503 pte_t pte; 1504 1505 if (vma->vm_flags & VM_SOFTDIRTY) 1506 flags |= PM_SOFT_DIRTY; 1507 1508 pte = huge_ptep_get(ptep); 1509 if (pte_present(pte)) { 1510 struct page *page = pte_page(pte); 1511 1512 if (!PageAnon(page)) 1513 flags |= PM_FILE; 1514 1515 if (page_mapcount(page) == 1) 1516 flags |= PM_MMAP_EXCLUSIVE; 1517 1518 flags |= PM_PRESENT; 1519 if (pm->show_pfn) 1520 frame = pte_pfn(pte) + 1521 ((addr & ~hmask) >> PAGE_SHIFT); 1522 } 1523 1524 for (; addr != end; addr += PAGE_SIZE) { 1525 pagemap_entry_t pme = make_pme(frame, flags); 1526 1527 err = add_to_pagemap(addr, &pme, pm); 1528 if (err) 1529 return err; 1530 if (pm->show_pfn && (flags & PM_PRESENT)) 1531 frame++; 1532 } 1533 1534 cond_resched(); 1535 1536 return err; 1537 } 1538 #else 1539 #define pagemap_hugetlb_range NULL 1540 #endif /* HUGETLB_PAGE */ 1541 1542 static const struct mm_walk_ops pagemap_ops = { 1543 .pmd_entry = pagemap_pmd_range, 1544 .pte_hole = pagemap_pte_hole, 1545 .hugetlb_entry = pagemap_hugetlb_range, 1546 }; 1547 1548 /* 1549 * /proc/pid/pagemap - an array mapping virtual pages to pfns 1550 * 1551 * For each page in the address space, this file contains one 64-bit entry 1552 * consisting of the following: 1553 * 1554 * Bits 0-54 page frame number (PFN) if present 1555 * Bits 0-4 swap type if swapped 1556 * Bits 5-54 swap offset if swapped 1557 * Bit 55 pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst) 1558 * Bit 56 page exclusively mapped 1559 * Bits 57-60 zero 1560 * Bit 61 page is file-page or shared-anon 1561 * Bit 62 page swapped 1562 * Bit 63 page present 1563 * 1564 * If the page is not present but in swap, then the PFN contains an 1565 * encoding of the swap file number and the page's offset into the 1566 * swap. Unmapped pages return a null PFN. This allows determining 1567 * precisely which pages are mapped (or in swap) and comparing mapped 1568 * pages between processes. 1569 * 1570 * Efficient users of this interface will use /proc/pid/maps to 1571 * determine which areas of memory are actually mapped and llseek to 1572 * skip over unmapped regions. 1573 */ 1574 static ssize_t pagemap_read(struct file *file, char __user *buf, 1575 size_t count, loff_t *ppos) 1576 { 1577 struct mm_struct *mm = file->private_data; 1578 struct pagemapread pm; 1579 unsigned long src; 1580 unsigned long svpfn; 1581 unsigned long start_vaddr; 1582 unsigned long end_vaddr; 1583 int ret = 0, copied = 0; 1584 1585 if (!mm || !mmget_not_zero(mm)) 1586 goto out; 1587 1588 ret = -EINVAL; 1589 /* file position must be aligned */ 1590 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES)) 1591 goto out_mm; 1592 1593 ret = 0; 1594 if (!count) 1595 goto out_mm; 1596 1597 /* do not disclose physical addresses: attack vector */ 1598 pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN); 1599 1600 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT); 1601 pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL); 1602 ret = -ENOMEM; 1603 if (!pm.buffer) 1604 goto out_mm; 1605 1606 src = *ppos; 1607 svpfn = src / PM_ENTRY_BYTES; 1608 end_vaddr = mm->task_size; 1609 1610 /* watch out for wraparound */ 1611 start_vaddr = end_vaddr; 1612 if (svpfn <= (ULONG_MAX >> PAGE_SHIFT)) 1613 start_vaddr = untagged_addr(svpfn << PAGE_SHIFT); 1614 1615 /* Ensure the address is inside the task */ 1616 if (start_vaddr > mm->task_size) 1617 start_vaddr = end_vaddr; 1618 1619 /* 1620 * The odds are that this will stop walking way 1621 * before end_vaddr, because the length of the 1622 * user buffer is tracked in "pm", and the walk 1623 * will stop when we hit the end of the buffer. 1624 */ 1625 ret = 0; 1626 while (count && (start_vaddr < end_vaddr)) { 1627 int len; 1628 unsigned long end; 1629 1630 pm.pos = 0; 1631 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK; 1632 /* overflow ? */ 1633 if (end < start_vaddr || end > end_vaddr) 1634 end = end_vaddr; 1635 ret = mmap_read_lock_killable(mm); 1636 if (ret) 1637 goto out_free; 1638 ret = walk_page_range(mm, start_vaddr, end, &pagemap_ops, &pm); 1639 mmap_read_unlock(mm); 1640 start_vaddr = end; 1641 1642 len = min(count, PM_ENTRY_BYTES * pm.pos); 1643 if (copy_to_user(buf, pm.buffer, len)) { 1644 ret = -EFAULT; 1645 goto out_free; 1646 } 1647 copied += len; 1648 buf += len; 1649 count -= len; 1650 } 1651 *ppos += copied; 1652 if (!ret || ret == PM_END_OF_BUFFER) 1653 ret = copied; 1654 1655 out_free: 1656 kfree(pm.buffer); 1657 out_mm: 1658 mmput(mm); 1659 out: 1660 return ret; 1661 } 1662 1663 static int pagemap_open(struct inode *inode, struct file *file) 1664 { 1665 struct mm_struct *mm; 1666 1667 mm = proc_mem_open(inode, PTRACE_MODE_READ); 1668 if (IS_ERR(mm)) 1669 return PTR_ERR(mm); 1670 file->private_data = mm; 1671 return 0; 1672 } 1673 1674 static int pagemap_release(struct inode *inode, struct file *file) 1675 { 1676 struct mm_struct *mm = file->private_data; 1677 1678 if (mm) 1679 mmdrop(mm); 1680 return 0; 1681 } 1682 1683 const struct file_operations proc_pagemap_operations = { 1684 .llseek = mem_lseek, /* borrow this */ 1685 .read = pagemap_read, 1686 .open = pagemap_open, 1687 .release = pagemap_release, 1688 }; 1689 #endif /* CONFIG_PROC_PAGE_MONITOR */ 1690 1691 #ifdef CONFIG_NUMA 1692 1693 struct numa_maps { 1694 unsigned long pages; 1695 unsigned long anon; 1696 unsigned long active; 1697 unsigned long writeback; 1698 unsigned long mapcount_max; 1699 unsigned long dirty; 1700 unsigned long swapcache; 1701 unsigned long node[MAX_NUMNODES]; 1702 }; 1703 1704 struct numa_maps_private { 1705 struct proc_maps_private proc_maps; 1706 struct numa_maps md; 1707 }; 1708 1709 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty, 1710 unsigned long nr_pages) 1711 { 1712 int count = page_mapcount(page); 1713 1714 md->pages += nr_pages; 1715 if (pte_dirty || PageDirty(page)) 1716 md->dirty += nr_pages; 1717 1718 if (PageSwapCache(page)) 1719 md->swapcache += nr_pages; 1720 1721 if (PageActive(page) || PageUnevictable(page)) 1722 md->active += nr_pages; 1723 1724 if (PageWriteback(page)) 1725 md->writeback += nr_pages; 1726 1727 if (PageAnon(page)) 1728 md->anon += nr_pages; 1729 1730 if (count > md->mapcount_max) 1731 md->mapcount_max = count; 1732 1733 md->node[page_to_nid(page)] += nr_pages; 1734 } 1735 1736 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma, 1737 unsigned long addr) 1738 { 1739 struct page *page; 1740 int nid; 1741 1742 if (!pte_present(pte)) 1743 return NULL; 1744 1745 page = vm_normal_page(vma, addr, pte); 1746 if (!page) 1747 return NULL; 1748 1749 if (PageReserved(page)) 1750 return NULL; 1751 1752 nid = page_to_nid(page); 1753 if (!node_isset(nid, node_states[N_MEMORY])) 1754 return NULL; 1755 1756 return page; 1757 } 1758 1759 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1760 static struct page *can_gather_numa_stats_pmd(pmd_t pmd, 1761 struct vm_area_struct *vma, 1762 unsigned long addr) 1763 { 1764 struct page *page; 1765 int nid; 1766 1767 if (!pmd_present(pmd)) 1768 return NULL; 1769 1770 page = vm_normal_page_pmd(vma, addr, pmd); 1771 if (!page) 1772 return NULL; 1773 1774 if (PageReserved(page)) 1775 return NULL; 1776 1777 nid = page_to_nid(page); 1778 if (!node_isset(nid, node_states[N_MEMORY])) 1779 return NULL; 1780 1781 return page; 1782 } 1783 #endif 1784 1785 static int gather_pte_stats(pmd_t *pmd, unsigned long addr, 1786 unsigned long end, struct mm_walk *walk) 1787 { 1788 struct numa_maps *md = walk->private; 1789 struct vm_area_struct *vma = walk->vma; 1790 spinlock_t *ptl; 1791 pte_t *orig_pte; 1792 pte_t *pte; 1793 1794 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1795 ptl = pmd_trans_huge_lock(pmd, vma); 1796 if (ptl) { 1797 struct page *page; 1798 1799 page = can_gather_numa_stats_pmd(*pmd, vma, addr); 1800 if (page) 1801 gather_stats(page, md, pmd_dirty(*pmd), 1802 HPAGE_PMD_SIZE/PAGE_SIZE); 1803 spin_unlock(ptl); 1804 return 0; 1805 } 1806 1807 if (pmd_trans_unstable(pmd)) 1808 return 0; 1809 #endif 1810 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl); 1811 do { 1812 struct page *page = can_gather_numa_stats(*pte, vma, addr); 1813 if (!page) 1814 continue; 1815 gather_stats(page, md, pte_dirty(*pte), 1); 1816 1817 } while (pte++, addr += PAGE_SIZE, addr != end); 1818 pte_unmap_unlock(orig_pte, ptl); 1819 cond_resched(); 1820 return 0; 1821 } 1822 #ifdef CONFIG_HUGETLB_PAGE 1823 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask, 1824 unsigned long addr, unsigned long end, struct mm_walk *walk) 1825 { 1826 pte_t huge_pte = huge_ptep_get(pte); 1827 struct numa_maps *md; 1828 struct page *page; 1829 1830 if (!pte_present(huge_pte)) 1831 return 0; 1832 1833 page = pte_page(huge_pte); 1834 if (!page) 1835 return 0; 1836 1837 md = walk->private; 1838 gather_stats(page, md, pte_dirty(huge_pte), 1); 1839 return 0; 1840 } 1841 1842 #else 1843 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask, 1844 unsigned long addr, unsigned long end, struct mm_walk *walk) 1845 { 1846 return 0; 1847 } 1848 #endif 1849 1850 static const struct mm_walk_ops show_numa_ops = { 1851 .hugetlb_entry = gather_hugetlb_stats, 1852 .pmd_entry = gather_pte_stats, 1853 }; 1854 1855 /* 1856 * Display pages allocated per node and memory policy via /proc. 1857 */ 1858 static int show_numa_map(struct seq_file *m, void *v) 1859 { 1860 struct numa_maps_private *numa_priv = m->private; 1861 struct proc_maps_private *proc_priv = &numa_priv->proc_maps; 1862 struct vm_area_struct *vma = v; 1863 struct numa_maps *md = &numa_priv->md; 1864 struct file *file = vma->vm_file; 1865 struct mm_struct *mm = vma->vm_mm; 1866 struct mempolicy *pol; 1867 char buffer[64]; 1868 int nid; 1869 1870 if (!mm) 1871 return 0; 1872 1873 /* Ensure we start with an empty set of numa_maps statistics. */ 1874 memset(md, 0, sizeof(*md)); 1875 1876 pol = __get_vma_policy(vma, vma->vm_start); 1877 if (pol) { 1878 mpol_to_str(buffer, sizeof(buffer), pol); 1879 mpol_cond_put(pol); 1880 } else { 1881 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy); 1882 } 1883 1884 seq_printf(m, "%08lx %s", vma->vm_start, buffer); 1885 1886 if (file) { 1887 seq_puts(m, " file="); 1888 seq_file_path(m, file, "\n\t= "); 1889 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) { 1890 seq_puts(m, " heap"); 1891 } else if (is_stack(vma)) { 1892 seq_puts(m, " stack"); 1893 } 1894 1895 if (is_vm_hugetlb_page(vma)) 1896 seq_puts(m, " huge"); 1897 1898 /* mmap_lock is held by m_start */ 1899 walk_page_vma(vma, &show_numa_ops, md); 1900 1901 if (!md->pages) 1902 goto out; 1903 1904 if (md->anon) 1905 seq_printf(m, " anon=%lu", md->anon); 1906 1907 if (md->dirty) 1908 seq_printf(m, " dirty=%lu", md->dirty); 1909 1910 if (md->pages != md->anon && md->pages != md->dirty) 1911 seq_printf(m, " mapped=%lu", md->pages); 1912 1913 if (md->mapcount_max > 1) 1914 seq_printf(m, " mapmax=%lu", md->mapcount_max); 1915 1916 if (md->swapcache) 1917 seq_printf(m, " swapcache=%lu", md->swapcache); 1918 1919 if (md->active < md->pages && !is_vm_hugetlb_page(vma)) 1920 seq_printf(m, " active=%lu", md->active); 1921 1922 if (md->writeback) 1923 seq_printf(m, " writeback=%lu", md->writeback); 1924 1925 for_each_node_state(nid, N_MEMORY) 1926 if (md->node[nid]) 1927 seq_printf(m, " N%d=%lu", nid, md->node[nid]); 1928 1929 seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10); 1930 out: 1931 seq_putc(m, '\n'); 1932 return 0; 1933 } 1934 1935 static const struct seq_operations proc_pid_numa_maps_op = { 1936 .start = m_start, 1937 .next = m_next, 1938 .stop = m_stop, 1939 .show = show_numa_map, 1940 }; 1941 1942 static int pid_numa_maps_open(struct inode *inode, struct file *file) 1943 { 1944 return proc_maps_open(inode, file, &proc_pid_numa_maps_op, 1945 sizeof(struct numa_maps_private)); 1946 } 1947 1948 const struct file_operations proc_pid_numa_maps_operations = { 1949 .open = pid_numa_maps_open, 1950 .read = seq_read, 1951 .llseek = seq_lseek, 1952 .release = proc_map_release, 1953 }; 1954 1955 #endif /* CONFIG_NUMA */ 1956