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 }; 401 402 static void smaps_page_accumulate(struct mem_size_stats *mss, 403 struct page *page, unsigned long size, unsigned long pss, 404 bool dirty, bool locked, bool private) 405 { 406 mss->pss += pss; 407 408 if (PageAnon(page)) 409 mss->pss_anon += pss; 410 else if (PageSwapBacked(page)) 411 mss->pss_shmem += pss; 412 else 413 mss->pss_file += pss; 414 415 if (locked) 416 mss->pss_locked += pss; 417 418 if (dirty || PageDirty(page)) { 419 if (private) 420 mss->private_dirty += size; 421 else 422 mss->shared_dirty += size; 423 } else { 424 if (private) 425 mss->private_clean += size; 426 else 427 mss->shared_clean += size; 428 } 429 } 430 431 static void smaps_account(struct mem_size_stats *mss, struct page *page, 432 bool compound, bool young, bool dirty, bool locked) 433 { 434 int i, nr = compound ? compound_nr(page) : 1; 435 unsigned long size = nr * PAGE_SIZE; 436 437 /* 438 * First accumulate quantities that depend only on |size| and the type 439 * of the compound page. 440 */ 441 if (PageAnon(page)) { 442 mss->anonymous += size; 443 if (!PageSwapBacked(page) && !dirty && !PageDirty(page)) 444 mss->lazyfree += size; 445 } 446 447 mss->resident += size; 448 /* Accumulate the size in pages that have been accessed. */ 449 if (young || page_is_young(page) || PageReferenced(page)) 450 mss->referenced += size; 451 452 /* 453 * Then accumulate quantities that may depend on sharing, or that may 454 * differ page-by-page. 455 * 456 * page_count(page) == 1 guarantees the page is mapped exactly once. 457 * If any subpage of the compound page mapped with PTE it would elevate 458 * page_count(). 459 */ 460 if (page_count(page) == 1) { 461 smaps_page_accumulate(mss, page, size, size << PSS_SHIFT, dirty, 462 locked, true); 463 return; 464 } 465 for (i = 0; i < nr; i++, page++) { 466 int mapcount = page_mapcount(page); 467 unsigned long pss = PAGE_SIZE << PSS_SHIFT; 468 if (mapcount >= 2) 469 pss /= mapcount; 470 smaps_page_accumulate(mss, page, PAGE_SIZE, pss, dirty, locked, 471 mapcount < 2); 472 } 473 } 474 475 #ifdef CONFIG_SHMEM 476 static int smaps_pte_hole(unsigned long addr, unsigned long end, 477 __always_unused int depth, struct mm_walk *walk) 478 { 479 struct mem_size_stats *mss = walk->private; 480 struct vm_area_struct *vma = walk->vma; 481 482 mss->swap += shmem_partial_swap_usage(walk->vma->vm_file->f_mapping, 483 linear_page_index(vma, addr), 484 linear_page_index(vma, end)); 485 486 return 0; 487 } 488 #else 489 #define smaps_pte_hole NULL 490 #endif /* CONFIG_SHMEM */ 491 492 static void smaps_pte_hole_lookup(unsigned long addr, struct mm_walk *walk) 493 { 494 #ifdef CONFIG_SHMEM 495 if (walk->ops->pte_hole) { 496 /* depth is not used */ 497 smaps_pte_hole(addr, addr + PAGE_SIZE, 0, walk); 498 } 499 #endif 500 } 501 502 static void smaps_pte_entry(pte_t *pte, unsigned long addr, 503 struct mm_walk *walk) 504 { 505 struct mem_size_stats *mss = walk->private; 506 struct vm_area_struct *vma = walk->vma; 507 bool locked = !!(vma->vm_flags & VM_LOCKED); 508 struct page *page = NULL; 509 510 if (pte_present(*pte)) { 511 page = vm_normal_page(vma, addr, *pte); 512 } else if (is_swap_pte(*pte)) { 513 swp_entry_t swpent = pte_to_swp_entry(*pte); 514 515 if (!non_swap_entry(swpent)) { 516 int mapcount; 517 518 mss->swap += PAGE_SIZE; 519 mapcount = swp_swapcount(swpent); 520 if (mapcount >= 2) { 521 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT; 522 523 do_div(pss_delta, mapcount); 524 mss->swap_pss += pss_delta; 525 } else { 526 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT; 527 } 528 } else if (is_pfn_swap_entry(swpent)) 529 page = pfn_swap_entry_to_page(swpent); 530 } else { 531 smaps_pte_hole_lookup(addr, walk); 532 return; 533 } 534 535 if (!page) 536 return; 537 538 smaps_account(mss, page, false, pte_young(*pte), pte_dirty(*pte), locked); 539 } 540 541 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 542 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr, 543 struct mm_walk *walk) 544 { 545 struct mem_size_stats *mss = walk->private; 546 struct vm_area_struct *vma = walk->vma; 547 bool locked = !!(vma->vm_flags & VM_LOCKED); 548 struct page *page = NULL; 549 550 if (pmd_present(*pmd)) { 551 /* FOLL_DUMP will return -EFAULT on huge zero page */ 552 page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP); 553 } else if (unlikely(thp_migration_supported() && is_swap_pmd(*pmd))) { 554 swp_entry_t entry = pmd_to_swp_entry(*pmd); 555 556 if (is_migration_entry(entry)) 557 page = pfn_swap_entry_to_page(entry); 558 } 559 if (IS_ERR_OR_NULL(page)) 560 return; 561 if (PageAnon(page)) 562 mss->anonymous_thp += HPAGE_PMD_SIZE; 563 else if (PageSwapBacked(page)) 564 mss->shmem_thp += HPAGE_PMD_SIZE; 565 else if (is_zone_device_page(page)) 566 /* pass */; 567 else 568 mss->file_thp += HPAGE_PMD_SIZE; 569 smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd), locked); 570 } 571 #else 572 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr, 573 struct mm_walk *walk) 574 { 575 } 576 #endif 577 578 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, 579 struct mm_walk *walk) 580 { 581 struct vm_area_struct *vma = walk->vma; 582 pte_t *pte; 583 spinlock_t *ptl; 584 585 ptl = pmd_trans_huge_lock(pmd, vma); 586 if (ptl) { 587 smaps_pmd_entry(pmd, addr, walk); 588 spin_unlock(ptl); 589 goto out; 590 } 591 592 if (pmd_trans_unstable(pmd)) 593 goto out; 594 /* 595 * The mmap_lock held all the way back in m_start() is what 596 * keeps khugepaged out of here and from collapsing things 597 * in here. 598 */ 599 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); 600 for (; addr != end; pte++, addr += PAGE_SIZE) 601 smaps_pte_entry(pte, addr, walk); 602 pte_unmap_unlock(pte - 1, ptl); 603 out: 604 cond_resched(); 605 return 0; 606 } 607 608 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma) 609 { 610 /* 611 * Don't forget to update Documentation/ on changes. 612 */ 613 static const char mnemonics[BITS_PER_LONG][2] = { 614 /* 615 * In case if we meet a flag we don't know about. 616 */ 617 [0 ... (BITS_PER_LONG-1)] = "??", 618 619 [ilog2(VM_READ)] = "rd", 620 [ilog2(VM_WRITE)] = "wr", 621 [ilog2(VM_EXEC)] = "ex", 622 [ilog2(VM_SHARED)] = "sh", 623 [ilog2(VM_MAYREAD)] = "mr", 624 [ilog2(VM_MAYWRITE)] = "mw", 625 [ilog2(VM_MAYEXEC)] = "me", 626 [ilog2(VM_MAYSHARE)] = "ms", 627 [ilog2(VM_GROWSDOWN)] = "gd", 628 [ilog2(VM_PFNMAP)] = "pf", 629 [ilog2(VM_LOCKED)] = "lo", 630 [ilog2(VM_IO)] = "io", 631 [ilog2(VM_SEQ_READ)] = "sr", 632 [ilog2(VM_RAND_READ)] = "rr", 633 [ilog2(VM_DONTCOPY)] = "dc", 634 [ilog2(VM_DONTEXPAND)] = "de", 635 [ilog2(VM_ACCOUNT)] = "ac", 636 [ilog2(VM_NORESERVE)] = "nr", 637 [ilog2(VM_HUGETLB)] = "ht", 638 [ilog2(VM_SYNC)] = "sf", 639 [ilog2(VM_ARCH_1)] = "ar", 640 [ilog2(VM_WIPEONFORK)] = "wf", 641 [ilog2(VM_DONTDUMP)] = "dd", 642 #ifdef CONFIG_ARM64_BTI 643 [ilog2(VM_ARM64_BTI)] = "bt", 644 #endif 645 #ifdef CONFIG_MEM_SOFT_DIRTY 646 [ilog2(VM_SOFTDIRTY)] = "sd", 647 #endif 648 [ilog2(VM_MIXEDMAP)] = "mm", 649 [ilog2(VM_HUGEPAGE)] = "hg", 650 [ilog2(VM_NOHUGEPAGE)] = "nh", 651 [ilog2(VM_MERGEABLE)] = "mg", 652 [ilog2(VM_UFFD_MISSING)]= "um", 653 [ilog2(VM_UFFD_WP)] = "uw", 654 #ifdef CONFIG_ARM64_MTE 655 [ilog2(VM_MTE)] = "mt", 656 [ilog2(VM_MTE_ALLOWED)] = "", 657 #endif 658 #ifdef CONFIG_ARCH_HAS_PKEYS 659 /* These come out via ProtectionKey: */ 660 [ilog2(VM_PKEY_BIT0)] = "", 661 [ilog2(VM_PKEY_BIT1)] = "", 662 [ilog2(VM_PKEY_BIT2)] = "", 663 [ilog2(VM_PKEY_BIT3)] = "", 664 #if VM_PKEY_BIT4 665 [ilog2(VM_PKEY_BIT4)] = "", 666 #endif 667 #endif /* CONFIG_ARCH_HAS_PKEYS */ 668 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_MINOR 669 [ilog2(VM_UFFD_MINOR)] = "ui", 670 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_MINOR */ 671 }; 672 size_t i; 673 674 seq_puts(m, "VmFlags: "); 675 for (i = 0; i < BITS_PER_LONG; i++) { 676 if (!mnemonics[i][0]) 677 continue; 678 if (vma->vm_flags & (1UL << i)) { 679 seq_putc(m, mnemonics[i][0]); 680 seq_putc(m, mnemonics[i][1]); 681 seq_putc(m, ' '); 682 } 683 } 684 seq_putc(m, '\n'); 685 } 686 687 #ifdef CONFIG_HUGETLB_PAGE 688 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask, 689 unsigned long addr, unsigned long end, 690 struct mm_walk *walk) 691 { 692 struct mem_size_stats *mss = walk->private; 693 struct vm_area_struct *vma = walk->vma; 694 struct page *page = NULL; 695 696 if (pte_present(*pte)) { 697 page = vm_normal_page(vma, addr, *pte); 698 } else if (is_swap_pte(*pte)) { 699 swp_entry_t swpent = pte_to_swp_entry(*pte); 700 701 if (is_pfn_swap_entry(swpent)) 702 page = pfn_swap_entry_to_page(swpent); 703 } 704 if (page) { 705 int mapcount = page_mapcount(page); 706 707 if (mapcount >= 2) 708 mss->shared_hugetlb += huge_page_size(hstate_vma(vma)); 709 else 710 mss->private_hugetlb += huge_page_size(hstate_vma(vma)); 711 } 712 return 0; 713 } 714 #else 715 #define smaps_hugetlb_range NULL 716 #endif /* HUGETLB_PAGE */ 717 718 static const struct mm_walk_ops smaps_walk_ops = { 719 .pmd_entry = smaps_pte_range, 720 .hugetlb_entry = smaps_hugetlb_range, 721 }; 722 723 static const struct mm_walk_ops smaps_shmem_walk_ops = { 724 .pmd_entry = smaps_pte_range, 725 .hugetlb_entry = smaps_hugetlb_range, 726 .pte_hole = smaps_pte_hole, 727 }; 728 729 /* 730 * Gather mem stats from @vma with the indicated beginning 731 * address @start, and keep them in @mss. 732 * 733 * Use vm_start of @vma as the beginning address if @start is 0. 734 */ 735 static void smap_gather_stats(struct vm_area_struct *vma, 736 struct mem_size_stats *mss, unsigned long start) 737 { 738 const struct mm_walk_ops *ops = &smaps_walk_ops; 739 740 /* Invalid start */ 741 if (start >= vma->vm_end) 742 return; 743 744 #ifdef CONFIG_SHMEM 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 ops = &smaps_shmem_walk_ops; 763 } 764 } 765 #endif 766 /* mmap_lock is held in m_start */ 767 if (!start) 768 walk_page_vma(vma, ops, mss); 769 else 770 walk_page_range(vma->vm_mm, start, vma->vm_end, ops, mss); 771 } 772 773 #define SEQ_PUT_DEC(str, val) \ 774 seq_put_decimal_ull_width(m, str, (val) >> 10, 8) 775 776 /* Show the contents common for smaps and smaps_rollup */ 777 static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss, 778 bool rollup_mode) 779 { 780 SEQ_PUT_DEC("Rss: ", mss->resident); 781 SEQ_PUT_DEC(" kB\nPss: ", mss->pss >> PSS_SHIFT); 782 if (rollup_mode) { 783 /* 784 * These are meaningful only for smaps_rollup, otherwise two of 785 * them are zero, and the other one is the same as Pss. 786 */ 787 SEQ_PUT_DEC(" kB\nPss_Anon: ", 788 mss->pss_anon >> PSS_SHIFT); 789 SEQ_PUT_DEC(" kB\nPss_File: ", 790 mss->pss_file >> PSS_SHIFT); 791 SEQ_PUT_DEC(" kB\nPss_Shmem: ", 792 mss->pss_shmem >> PSS_SHIFT); 793 } 794 SEQ_PUT_DEC(" kB\nShared_Clean: ", mss->shared_clean); 795 SEQ_PUT_DEC(" kB\nShared_Dirty: ", mss->shared_dirty); 796 SEQ_PUT_DEC(" kB\nPrivate_Clean: ", mss->private_clean); 797 SEQ_PUT_DEC(" kB\nPrivate_Dirty: ", mss->private_dirty); 798 SEQ_PUT_DEC(" kB\nReferenced: ", mss->referenced); 799 SEQ_PUT_DEC(" kB\nAnonymous: ", mss->anonymous); 800 SEQ_PUT_DEC(" kB\nLazyFree: ", mss->lazyfree); 801 SEQ_PUT_DEC(" kB\nAnonHugePages: ", mss->anonymous_thp); 802 SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp); 803 SEQ_PUT_DEC(" kB\nFilePmdMapped: ", mss->file_thp); 804 SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb); 805 seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ", 806 mss->private_hugetlb >> 10, 7); 807 SEQ_PUT_DEC(" kB\nSwap: ", mss->swap); 808 SEQ_PUT_DEC(" kB\nSwapPss: ", 809 mss->swap_pss >> PSS_SHIFT); 810 SEQ_PUT_DEC(" kB\nLocked: ", 811 mss->pss_locked >> PSS_SHIFT); 812 seq_puts(m, " kB\n"); 813 } 814 815 static int show_smap(struct seq_file *m, void *v) 816 { 817 struct vm_area_struct *vma = v; 818 struct mem_size_stats mss; 819 820 memset(&mss, 0, sizeof(mss)); 821 822 smap_gather_stats(vma, &mss, 0); 823 824 show_map_vma(m, vma); 825 826 SEQ_PUT_DEC("Size: ", vma->vm_end - vma->vm_start); 827 SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma)); 828 SEQ_PUT_DEC(" kB\nMMUPageSize: ", vma_mmu_pagesize(vma)); 829 seq_puts(m, " kB\n"); 830 831 __show_smap(m, &mss, false); 832 833 seq_printf(m, "THPeligible: %d\n", 834 transparent_hugepage_active(vma)); 835 836 if (arch_pkeys_enabled()) 837 seq_printf(m, "ProtectionKey: %8u\n", vma_pkey(vma)); 838 show_smap_vma_flags(m, vma); 839 840 return 0; 841 } 842 843 static int show_smaps_rollup(struct seq_file *m, void *v) 844 { 845 struct proc_maps_private *priv = m->private; 846 struct mem_size_stats mss; 847 struct mm_struct *mm; 848 struct vm_area_struct *vma; 849 unsigned long last_vma_end = 0; 850 int ret = 0; 851 852 priv->task = get_proc_task(priv->inode); 853 if (!priv->task) 854 return -ESRCH; 855 856 mm = priv->mm; 857 if (!mm || !mmget_not_zero(mm)) { 858 ret = -ESRCH; 859 goto out_put_task; 860 } 861 862 memset(&mss, 0, sizeof(mss)); 863 864 ret = mmap_read_lock_killable(mm); 865 if (ret) 866 goto out_put_mm; 867 868 hold_task_mempolicy(priv); 869 870 for (vma = priv->mm->mmap; vma;) { 871 smap_gather_stats(vma, &mss, 0); 872 last_vma_end = vma->vm_end; 873 874 /* 875 * Release mmap_lock temporarily if someone wants to 876 * access it for write request. 877 */ 878 if (mmap_lock_is_contended(mm)) { 879 mmap_read_unlock(mm); 880 ret = mmap_read_lock_killable(mm); 881 if (ret) { 882 release_task_mempolicy(priv); 883 goto out_put_mm; 884 } 885 886 /* 887 * After dropping the lock, there are four cases to 888 * consider. See the following example for explanation. 889 * 890 * +------+------+-----------+ 891 * | VMA1 | VMA2 | VMA3 | 892 * +------+------+-----------+ 893 * | | | | 894 * 4k 8k 16k 400k 895 * 896 * Suppose we drop the lock after reading VMA2 due to 897 * contention, then we get: 898 * 899 * last_vma_end = 16k 900 * 901 * 1) VMA2 is freed, but VMA3 exists: 902 * 903 * find_vma(mm, 16k - 1) will return VMA3. 904 * In this case, just continue from VMA3. 905 * 906 * 2) VMA2 still exists: 907 * 908 * find_vma(mm, 16k - 1) will return VMA2. 909 * Iterate the loop like the original one. 910 * 911 * 3) No more VMAs can be found: 912 * 913 * find_vma(mm, 16k - 1) will return NULL. 914 * No more things to do, just break. 915 * 916 * 4) (last_vma_end - 1) is the middle of a vma (VMA'): 917 * 918 * find_vma(mm, 16k - 1) will return VMA' whose range 919 * contains last_vma_end. 920 * Iterate VMA' from last_vma_end. 921 */ 922 vma = find_vma(mm, last_vma_end - 1); 923 /* Case 3 above */ 924 if (!vma) 925 break; 926 927 /* Case 1 above */ 928 if (vma->vm_start >= last_vma_end) 929 continue; 930 931 /* Case 4 above */ 932 if (vma->vm_end > last_vma_end) 933 smap_gather_stats(vma, &mss, last_vma_end); 934 } 935 /* Case 2 above */ 936 vma = vma->vm_next; 937 } 938 939 show_vma_header_prefix(m, priv->mm->mmap->vm_start, 940 last_vma_end, 0, 0, 0, 0); 941 seq_pad(m, ' '); 942 seq_puts(m, "[rollup]\n"); 943 944 __show_smap(m, &mss, true); 945 946 release_task_mempolicy(priv); 947 mmap_read_unlock(mm); 948 949 out_put_mm: 950 mmput(mm); 951 out_put_task: 952 put_task_struct(priv->task); 953 priv->task = NULL; 954 955 return ret; 956 } 957 #undef SEQ_PUT_DEC 958 959 static const struct seq_operations proc_pid_smaps_op = { 960 .start = m_start, 961 .next = m_next, 962 .stop = m_stop, 963 .show = show_smap 964 }; 965 966 static int pid_smaps_open(struct inode *inode, struct file *file) 967 { 968 return do_maps_open(inode, file, &proc_pid_smaps_op); 969 } 970 971 static int smaps_rollup_open(struct inode *inode, struct file *file) 972 { 973 int ret; 974 struct proc_maps_private *priv; 975 976 priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT); 977 if (!priv) 978 return -ENOMEM; 979 980 ret = single_open(file, show_smaps_rollup, priv); 981 if (ret) 982 goto out_free; 983 984 priv->inode = inode; 985 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ); 986 if (IS_ERR(priv->mm)) { 987 ret = PTR_ERR(priv->mm); 988 989 single_release(inode, file); 990 goto out_free; 991 } 992 993 return 0; 994 995 out_free: 996 kfree(priv); 997 return ret; 998 } 999 1000 static int smaps_rollup_release(struct inode *inode, struct file *file) 1001 { 1002 struct seq_file *seq = file->private_data; 1003 struct proc_maps_private *priv = seq->private; 1004 1005 if (priv->mm) 1006 mmdrop(priv->mm); 1007 1008 kfree(priv); 1009 return single_release(inode, file); 1010 } 1011 1012 const struct file_operations proc_pid_smaps_operations = { 1013 .open = pid_smaps_open, 1014 .read = seq_read, 1015 .llseek = seq_lseek, 1016 .release = proc_map_release, 1017 }; 1018 1019 const struct file_operations proc_pid_smaps_rollup_operations = { 1020 .open = smaps_rollup_open, 1021 .read = seq_read, 1022 .llseek = seq_lseek, 1023 .release = smaps_rollup_release, 1024 }; 1025 1026 enum clear_refs_types { 1027 CLEAR_REFS_ALL = 1, 1028 CLEAR_REFS_ANON, 1029 CLEAR_REFS_MAPPED, 1030 CLEAR_REFS_SOFT_DIRTY, 1031 CLEAR_REFS_MM_HIWATER_RSS, 1032 CLEAR_REFS_LAST, 1033 }; 1034 1035 struct clear_refs_private { 1036 enum clear_refs_types type; 1037 }; 1038 1039 #ifdef CONFIG_MEM_SOFT_DIRTY 1040 1041 static inline bool pte_is_pinned(struct vm_area_struct *vma, unsigned long addr, pte_t pte) 1042 { 1043 struct page *page; 1044 1045 if (!pte_write(pte)) 1046 return false; 1047 if (!is_cow_mapping(vma->vm_flags)) 1048 return false; 1049 if (likely(!test_bit(MMF_HAS_PINNED, &vma->vm_mm->flags))) 1050 return false; 1051 page = vm_normal_page(vma, addr, pte); 1052 if (!page) 1053 return false; 1054 return page_maybe_dma_pinned(page); 1055 } 1056 1057 static inline void clear_soft_dirty(struct vm_area_struct *vma, 1058 unsigned long addr, pte_t *pte) 1059 { 1060 /* 1061 * The soft-dirty tracker uses #PF-s to catch writes 1062 * to pages, so write-protect the pte as well. See the 1063 * Documentation/admin-guide/mm/soft-dirty.rst for full description 1064 * of how soft-dirty works. 1065 */ 1066 pte_t ptent = *pte; 1067 1068 if (pte_present(ptent)) { 1069 pte_t old_pte; 1070 1071 if (pte_is_pinned(vma, addr, ptent)) 1072 return; 1073 old_pte = ptep_modify_prot_start(vma, addr, pte); 1074 ptent = pte_wrprotect(old_pte); 1075 ptent = pte_clear_soft_dirty(ptent); 1076 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent); 1077 } else if (is_swap_pte(ptent)) { 1078 ptent = pte_swp_clear_soft_dirty(ptent); 1079 set_pte_at(vma->vm_mm, addr, pte, ptent); 1080 } 1081 } 1082 #else 1083 static inline void clear_soft_dirty(struct vm_area_struct *vma, 1084 unsigned long addr, pte_t *pte) 1085 { 1086 } 1087 #endif 1088 1089 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE) 1090 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma, 1091 unsigned long addr, pmd_t *pmdp) 1092 { 1093 pmd_t old, pmd = *pmdp; 1094 1095 if (pmd_present(pmd)) { 1096 /* See comment in change_huge_pmd() */ 1097 old = pmdp_invalidate(vma, addr, pmdp); 1098 if (pmd_dirty(old)) 1099 pmd = pmd_mkdirty(pmd); 1100 if (pmd_young(old)) 1101 pmd = pmd_mkyoung(pmd); 1102 1103 pmd = pmd_wrprotect(pmd); 1104 pmd = pmd_clear_soft_dirty(pmd); 1105 1106 set_pmd_at(vma->vm_mm, addr, pmdp, pmd); 1107 } else if (is_migration_entry(pmd_to_swp_entry(pmd))) { 1108 pmd = pmd_swp_clear_soft_dirty(pmd); 1109 set_pmd_at(vma->vm_mm, addr, pmdp, pmd); 1110 } 1111 } 1112 #else 1113 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma, 1114 unsigned long addr, pmd_t *pmdp) 1115 { 1116 } 1117 #endif 1118 1119 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr, 1120 unsigned long end, struct mm_walk *walk) 1121 { 1122 struct clear_refs_private *cp = walk->private; 1123 struct vm_area_struct *vma = walk->vma; 1124 pte_t *pte, ptent; 1125 spinlock_t *ptl; 1126 struct page *page; 1127 1128 ptl = pmd_trans_huge_lock(pmd, vma); 1129 if (ptl) { 1130 if (cp->type == CLEAR_REFS_SOFT_DIRTY) { 1131 clear_soft_dirty_pmd(vma, addr, pmd); 1132 goto out; 1133 } 1134 1135 if (!pmd_present(*pmd)) 1136 goto out; 1137 1138 page = pmd_page(*pmd); 1139 1140 /* Clear accessed and referenced bits. */ 1141 pmdp_test_and_clear_young(vma, addr, pmd); 1142 test_and_clear_page_young(page); 1143 ClearPageReferenced(page); 1144 out: 1145 spin_unlock(ptl); 1146 return 0; 1147 } 1148 1149 if (pmd_trans_unstable(pmd)) 1150 return 0; 1151 1152 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); 1153 for (; addr != end; pte++, addr += PAGE_SIZE) { 1154 ptent = *pte; 1155 1156 if (cp->type == CLEAR_REFS_SOFT_DIRTY) { 1157 clear_soft_dirty(vma, addr, pte); 1158 continue; 1159 } 1160 1161 if (!pte_present(ptent)) 1162 continue; 1163 1164 page = vm_normal_page(vma, addr, ptent); 1165 if (!page) 1166 continue; 1167 1168 /* Clear accessed and referenced bits. */ 1169 ptep_test_and_clear_young(vma, addr, pte); 1170 test_and_clear_page_young(page); 1171 ClearPageReferenced(page); 1172 } 1173 pte_unmap_unlock(pte - 1, ptl); 1174 cond_resched(); 1175 return 0; 1176 } 1177 1178 static int clear_refs_test_walk(unsigned long start, unsigned long end, 1179 struct mm_walk *walk) 1180 { 1181 struct clear_refs_private *cp = walk->private; 1182 struct vm_area_struct *vma = walk->vma; 1183 1184 if (vma->vm_flags & VM_PFNMAP) 1185 return 1; 1186 1187 /* 1188 * Writing 1 to /proc/pid/clear_refs affects all pages. 1189 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages. 1190 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages. 1191 * Writing 4 to /proc/pid/clear_refs affects all pages. 1192 */ 1193 if (cp->type == CLEAR_REFS_ANON && vma->vm_file) 1194 return 1; 1195 if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file) 1196 return 1; 1197 return 0; 1198 } 1199 1200 static const struct mm_walk_ops clear_refs_walk_ops = { 1201 .pmd_entry = clear_refs_pte_range, 1202 .test_walk = clear_refs_test_walk, 1203 }; 1204 1205 static ssize_t clear_refs_write(struct file *file, const char __user *buf, 1206 size_t count, loff_t *ppos) 1207 { 1208 struct task_struct *task; 1209 char buffer[PROC_NUMBUF]; 1210 struct mm_struct *mm; 1211 struct vm_area_struct *vma; 1212 enum clear_refs_types type; 1213 int itype; 1214 int rv; 1215 1216 memset(buffer, 0, sizeof(buffer)); 1217 if (count > sizeof(buffer) - 1) 1218 count = sizeof(buffer) - 1; 1219 if (copy_from_user(buffer, buf, count)) 1220 return -EFAULT; 1221 rv = kstrtoint(strstrip(buffer), 10, &itype); 1222 if (rv < 0) 1223 return rv; 1224 type = (enum clear_refs_types)itype; 1225 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST) 1226 return -EINVAL; 1227 1228 task = get_proc_task(file_inode(file)); 1229 if (!task) 1230 return -ESRCH; 1231 mm = get_task_mm(task); 1232 if (mm) { 1233 struct mmu_notifier_range range; 1234 struct clear_refs_private cp = { 1235 .type = type, 1236 }; 1237 1238 if (mmap_write_lock_killable(mm)) { 1239 count = -EINTR; 1240 goto out_mm; 1241 } 1242 if (type == CLEAR_REFS_MM_HIWATER_RSS) { 1243 /* 1244 * Writing 5 to /proc/pid/clear_refs resets the peak 1245 * resident set size to this mm's current rss value. 1246 */ 1247 reset_mm_hiwater_rss(mm); 1248 goto out_unlock; 1249 } 1250 1251 if (type == CLEAR_REFS_SOFT_DIRTY) { 1252 for (vma = mm->mmap; vma; vma = vma->vm_next) { 1253 if (!(vma->vm_flags & VM_SOFTDIRTY)) 1254 continue; 1255 vma->vm_flags &= ~VM_SOFTDIRTY; 1256 vma_set_page_prot(vma); 1257 } 1258 1259 inc_tlb_flush_pending(mm); 1260 mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY, 1261 0, NULL, mm, 0, -1UL); 1262 mmu_notifier_invalidate_range_start(&range); 1263 } 1264 walk_page_range(mm, 0, mm->highest_vm_end, &clear_refs_walk_ops, 1265 &cp); 1266 if (type == CLEAR_REFS_SOFT_DIRTY) { 1267 mmu_notifier_invalidate_range_end(&range); 1268 flush_tlb_mm(mm); 1269 dec_tlb_flush_pending(mm); 1270 } 1271 out_unlock: 1272 mmap_write_unlock(mm); 1273 out_mm: 1274 mmput(mm); 1275 } 1276 put_task_struct(task); 1277 1278 return count; 1279 } 1280 1281 const struct file_operations proc_clear_refs_operations = { 1282 .write = clear_refs_write, 1283 .llseek = noop_llseek, 1284 }; 1285 1286 typedef struct { 1287 u64 pme; 1288 } pagemap_entry_t; 1289 1290 struct pagemapread { 1291 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */ 1292 pagemap_entry_t *buffer; 1293 bool show_pfn; 1294 }; 1295 1296 #define PAGEMAP_WALK_SIZE (PMD_SIZE) 1297 #define PAGEMAP_WALK_MASK (PMD_MASK) 1298 1299 #define PM_ENTRY_BYTES sizeof(pagemap_entry_t) 1300 #define PM_PFRAME_BITS 55 1301 #define PM_PFRAME_MASK GENMASK_ULL(PM_PFRAME_BITS - 1, 0) 1302 #define PM_SOFT_DIRTY BIT_ULL(55) 1303 #define PM_MMAP_EXCLUSIVE BIT_ULL(56) 1304 #define PM_UFFD_WP BIT_ULL(57) 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 if (pte_uffd_wp(pte)) 1379 flags |= PM_UFFD_WP; 1380 } else if (is_swap_pte(pte)) { 1381 swp_entry_t entry; 1382 if (pte_swp_soft_dirty(pte)) 1383 flags |= PM_SOFT_DIRTY; 1384 if (pte_swp_uffd_wp(pte)) 1385 flags |= PM_UFFD_WP; 1386 entry = pte_to_swp_entry(pte); 1387 if (pm->show_pfn) 1388 frame = swp_type(entry) | 1389 (swp_offset(entry) << MAX_SWAPFILES_SHIFT); 1390 flags |= PM_SWAP; 1391 if (is_pfn_swap_entry(entry)) 1392 page = pfn_swap_entry_to_page(entry); 1393 } 1394 1395 if (page && !PageAnon(page)) 1396 flags |= PM_FILE; 1397 if (page && page_mapcount(page) == 1) 1398 flags |= PM_MMAP_EXCLUSIVE; 1399 if (vma->vm_flags & VM_SOFTDIRTY) 1400 flags |= PM_SOFT_DIRTY; 1401 1402 return make_pme(frame, flags); 1403 } 1404 1405 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end, 1406 struct mm_walk *walk) 1407 { 1408 struct vm_area_struct *vma = walk->vma; 1409 struct pagemapread *pm = walk->private; 1410 spinlock_t *ptl; 1411 pte_t *pte, *orig_pte; 1412 int err = 0; 1413 1414 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1415 ptl = pmd_trans_huge_lock(pmdp, vma); 1416 if (ptl) { 1417 u64 flags = 0, frame = 0; 1418 pmd_t pmd = *pmdp; 1419 struct page *page = NULL; 1420 1421 if (vma->vm_flags & VM_SOFTDIRTY) 1422 flags |= PM_SOFT_DIRTY; 1423 1424 if (pmd_present(pmd)) { 1425 page = pmd_page(pmd); 1426 1427 flags |= PM_PRESENT; 1428 if (pmd_soft_dirty(pmd)) 1429 flags |= PM_SOFT_DIRTY; 1430 if (pmd_uffd_wp(pmd)) 1431 flags |= PM_UFFD_WP; 1432 if (pm->show_pfn) 1433 frame = pmd_pfn(pmd) + 1434 ((addr & ~PMD_MASK) >> PAGE_SHIFT); 1435 } 1436 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION 1437 else if (is_swap_pmd(pmd)) { 1438 swp_entry_t entry = pmd_to_swp_entry(pmd); 1439 unsigned long offset; 1440 1441 if (pm->show_pfn) { 1442 offset = swp_offset(entry) + 1443 ((addr & ~PMD_MASK) >> PAGE_SHIFT); 1444 frame = swp_type(entry) | 1445 (offset << MAX_SWAPFILES_SHIFT); 1446 } 1447 flags |= PM_SWAP; 1448 if (pmd_swp_soft_dirty(pmd)) 1449 flags |= PM_SOFT_DIRTY; 1450 if (pmd_swp_uffd_wp(pmd)) 1451 flags |= PM_UFFD_WP; 1452 VM_BUG_ON(!is_pmd_migration_entry(pmd)); 1453 page = pfn_swap_entry_to_page(entry); 1454 } 1455 #endif 1456 1457 if (page && page_mapcount(page) == 1) 1458 flags |= PM_MMAP_EXCLUSIVE; 1459 1460 for (; addr != end; addr += PAGE_SIZE) { 1461 pagemap_entry_t pme = make_pme(frame, flags); 1462 1463 err = add_to_pagemap(addr, &pme, pm); 1464 if (err) 1465 break; 1466 if (pm->show_pfn) { 1467 if (flags & PM_PRESENT) 1468 frame++; 1469 else if (flags & PM_SWAP) 1470 frame += (1 << MAX_SWAPFILES_SHIFT); 1471 } 1472 } 1473 spin_unlock(ptl); 1474 return err; 1475 } 1476 1477 if (pmd_trans_unstable(pmdp)) 1478 return 0; 1479 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ 1480 1481 /* 1482 * We can assume that @vma always points to a valid one and @end never 1483 * goes beyond vma->vm_end. 1484 */ 1485 orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl); 1486 for (; addr < end; pte++, addr += PAGE_SIZE) { 1487 pagemap_entry_t pme; 1488 1489 pme = pte_to_pagemap_entry(pm, vma, addr, *pte); 1490 err = add_to_pagemap(addr, &pme, pm); 1491 if (err) 1492 break; 1493 } 1494 pte_unmap_unlock(orig_pte, ptl); 1495 1496 cond_resched(); 1497 1498 return err; 1499 } 1500 1501 #ifdef CONFIG_HUGETLB_PAGE 1502 /* This function walks within one hugetlb entry in the single call */ 1503 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask, 1504 unsigned long addr, unsigned long end, 1505 struct mm_walk *walk) 1506 { 1507 struct pagemapread *pm = walk->private; 1508 struct vm_area_struct *vma = walk->vma; 1509 u64 flags = 0, frame = 0; 1510 int err = 0; 1511 pte_t pte; 1512 1513 if (vma->vm_flags & VM_SOFTDIRTY) 1514 flags |= PM_SOFT_DIRTY; 1515 1516 pte = huge_ptep_get(ptep); 1517 if (pte_present(pte)) { 1518 struct page *page = pte_page(pte); 1519 1520 if (!PageAnon(page)) 1521 flags |= PM_FILE; 1522 1523 if (page_mapcount(page) == 1) 1524 flags |= PM_MMAP_EXCLUSIVE; 1525 1526 flags |= PM_PRESENT; 1527 if (pm->show_pfn) 1528 frame = pte_pfn(pte) + 1529 ((addr & ~hmask) >> PAGE_SHIFT); 1530 } 1531 1532 for (; addr != end; addr += PAGE_SIZE) { 1533 pagemap_entry_t pme = make_pme(frame, flags); 1534 1535 err = add_to_pagemap(addr, &pme, pm); 1536 if (err) 1537 return err; 1538 if (pm->show_pfn && (flags & PM_PRESENT)) 1539 frame++; 1540 } 1541 1542 cond_resched(); 1543 1544 return err; 1545 } 1546 #else 1547 #define pagemap_hugetlb_range NULL 1548 #endif /* HUGETLB_PAGE */ 1549 1550 static const struct mm_walk_ops pagemap_ops = { 1551 .pmd_entry = pagemap_pmd_range, 1552 .pte_hole = pagemap_pte_hole, 1553 .hugetlb_entry = pagemap_hugetlb_range, 1554 }; 1555 1556 /* 1557 * /proc/pid/pagemap - an array mapping virtual pages to pfns 1558 * 1559 * For each page in the address space, this file contains one 64-bit entry 1560 * consisting of the following: 1561 * 1562 * Bits 0-54 page frame number (PFN) if present 1563 * Bits 0-4 swap type if swapped 1564 * Bits 5-54 swap offset if swapped 1565 * Bit 55 pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst) 1566 * Bit 56 page exclusively mapped 1567 * Bits 57-60 zero 1568 * Bit 61 page is file-page or shared-anon 1569 * Bit 62 page swapped 1570 * Bit 63 page present 1571 * 1572 * If the page is not present but in swap, then the PFN contains an 1573 * encoding of the swap file number and the page's offset into the 1574 * swap. Unmapped pages return a null PFN. This allows determining 1575 * precisely which pages are mapped (or in swap) and comparing mapped 1576 * pages between processes. 1577 * 1578 * Efficient users of this interface will use /proc/pid/maps to 1579 * determine which areas of memory are actually mapped and llseek to 1580 * skip over unmapped regions. 1581 */ 1582 static ssize_t pagemap_read(struct file *file, char __user *buf, 1583 size_t count, loff_t *ppos) 1584 { 1585 struct mm_struct *mm = file->private_data; 1586 struct pagemapread pm; 1587 unsigned long src; 1588 unsigned long svpfn; 1589 unsigned long start_vaddr; 1590 unsigned long end_vaddr; 1591 int ret = 0, copied = 0; 1592 1593 if (!mm || !mmget_not_zero(mm)) 1594 goto out; 1595 1596 ret = -EINVAL; 1597 /* file position must be aligned */ 1598 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES)) 1599 goto out_mm; 1600 1601 ret = 0; 1602 if (!count) 1603 goto out_mm; 1604 1605 /* do not disclose physical addresses: attack vector */ 1606 pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN); 1607 1608 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT); 1609 pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL); 1610 ret = -ENOMEM; 1611 if (!pm.buffer) 1612 goto out_mm; 1613 1614 src = *ppos; 1615 svpfn = src / PM_ENTRY_BYTES; 1616 end_vaddr = mm->task_size; 1617 1618 /* watch out for wraparound */ 1619 start_vaddr = end_vaddr; 1620 if (svpfn <= (ULONG_MAX >> PAGE_SHIFT)) 1621 start_vaddr = untagged_addr(svpfn << PAGE_SHIFT); 1622 1623 /* Ensure the address is inside the task */ 1624 if (start_vaddr > mm->task_size) 1625 start_vaddr = end_vaddr; 1626 1627 /* 1628 * The odds are that this will stop walking way 1629 * before end_vaddr, because the length of the 1630 * user buffer is tracked in "pm", and the walk 1631 * will stop when we hit the end of the buffer. 1632 */ 1633 ret = 0; 1634 while (count && (start_vaddr < end_vaddr)) { 1635 int len; 1636 unsigned long end; 1637 1638 pm.pos = 0; 1639 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK; 1640 /* overflow ? */ 1641 if (end < start_vaddr || end > end_vaddr) 1642 end = end_vaddr; 1643 ret = mmap_read_lock_killable(mm); 1644 if (ret) 1645 goto out_free; 1646 ret = walk_page_range(mm, start_vaddr, end, &pagemap_ops, &pm); 1647 mmap_read_unlock(mm); 1648 start_vaddr = end; 1649 1650 len = min(count, PM_ENTRY_BYTES * pm.pos); 1651 if (copy_to_user(buf, pm.buffer, len)) { 1652 ret = -EFAULT; 1653 goto out_free; 1654 } 1655 copied += len; 1656 buf += len; 1657 count -= len; 1658 } 1659 *ppos += copied; 1660 if (!ret || ret == PM_END_OF_BUFFER) 1661 ret = copied; 1662 1663 out_free: 1664 kfree(pm.buffer); 1665 out_mm: 1666 mmput(mm); 1667 out: 1668 return ret; 1669 } 1670 1671 static int pagemap_open(struct inode *inode, struct file *file) 1672 { 1673 struct mm_struct *mm; 1674 1675 mm = proc_mem_open(inode, PTRACE_MODE_READ); 1676 if (IS_ERR(mm)) 1677 return PTR_ERR(mm); 1678 file->private_data = mm; 1679 return 0; 1680 } 1681 1682 static int pagemap_release(struct inode *inode, struct file *file) 1683 { 1684 struct mm_struct *mm = file->private_data; 1685 1686 if (mm) 1687 mmdrop(mm); 1688 return 0; 1689 } 1690 1691 const struct file_operations proc_pagemap_operations = { 1692 .llseek = mem_lseek, /* borrow this */ 1693 .read = pagemap_read, 1694 .open = pagemap_open, 1695 .release = pagemap_release, 1696 }; 1697 #endif /* CONFIG_PROC_PAGE_MONITOR */ 1698 1699 #ifdef CONFIG_NUMA 1700 1701 struct numa_maps { 1702 unsigned long pages; 1703 unsigned long anon; 1704 unsigned long active; 1705 unsigned long writeback; 1706 unsigned long mapcount_max; 1707 unsigned long dirty; 1708 unsigned long swapcache; 1709 unsigned long node[MAX_NUMNODES]; 1710 }; 1711 1712 struct numa_maps_private { 1713 struct proc_maps_private proc_maps; 1714 struct numa_maps md; 1715 }; 1716 1717 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty, 1718 unsigned long nr_pages) 1719 { 1720 int count = page_mapcount(page); 1721 1722 md->pages += nr_pages; 1723 if (pte_dirty || PageDirty(page)) 1724 md->dirty += nr_pages; 1725 1726 if (PageSwapCache(page)) 1727 md->swapcache += nr_pages; 1728 1729 if (PageActive(page) || PageUnevictable(page)) 1730 md->active += nr_pages; 1731 1732 if (PageWriteback(page)) 1733 md->writeback += nr_pages; 1734 1735 if (PageAnon(page)) 1736 md->anon += nr_pages; 1737 1738 if (count > md->mapcount_max) 1739 md->mapcount_max = count; 1740 1741 md->node[page_to_nid(page)] += nr_pages; 1742 } 1743 1744 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma, 1745 unsigned long addr) 1746 { 1747 struct page *page; 1748 int nid; 1749 1750 if (!pte_present(pte)) 1751 return NULL; 1752 1753 page = vm_normal_page(vma, addr, pte); 1754 if (!page) 1755 return NULL; 1756 1757 if (PageReserved(page)) 1758 return NULL; 1759 1760 nid = page_to_nid(page); 1761 if (!node_isset(nid, node_states[N_MEMORY])) 1762 return NULL; 1763 1764 return page; 1765 } 1766 1767 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1768 static struct page *can_gather_numa_stats_pmd(pmd_t pmd, 1769 struct vm_area_struct *vma, 1770 unsigned long addr) 1771 { 1772 struct page *page; 1773 int nid; 1774 1775 if (!pmd_present(pmd)) 1776 return NULL; 1777 1778 page = vm_normal_page_pmd(vma, addr, pmd); 1779 if (!page) 1780 return NULL; 1781 1782 if (PageReserved(page)) 1783 return NULL; 1784 1785 nid = page_to_nid(page); 1786 if (!node_isset(nid, node_states[N_MEMORY])) 1787 return NULL; 1788 1789 return page; 1790 } 1791 #endif 1792 1793 static int gather_pte_stats(pmd_t *pmd, unsigned long addr, 1794 unsigned long end, struct mm_walk *walk) 1795 { 1796 struct numa_maps *md = walk->private; 1797 struct vm_area_struct *vma = walk->vma; 1798 spinlock_t *ptl; 1799 pte_t *orig_pte; 1800 pte_t *pte; 1801 1802 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1803 ptl = pmd_trans_huge_lock(pmd, vma); 1804 if (ptl) { 1805 struct page *page; 1806 1807 page = can_gather_numa_stats_pmd(*pmd, vma, addr); 1808 if (page) 1809 gather_stats(page, md, pmd_dirty(*pmd), 1810 HPAGE_PMD_SIZE/PAGE_SIZE); 1811 spin_unlock(ptl); 1812 return 0; 1813 } 1814 1815 if (pmd_trans_unstable(pmd)) 1816 return 0; 1817 #endif 1818 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl); 1819 do { 1820 struct page *page = can_gather_numa_stats(*pte, vma, addr); 1821 if (!page) 1822 continue; 1823 gather_stats(page, md, pte_dirty(*pte), 1); 1824 1825 } while (pte++, addr += PAGE_SIZE, addr != end); 1826 pte_unmap_unlock(orig_pte, ptl); 1827 cond_resched(); 1828 return 0; 1829 } 1830 #ifdef CONFIG_HUGETLB_PAGE 1831 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask, 1832 unsigned long addr, unsigned long end, struct mm_walk *walk) 1833 { 1834 pte_t huge_pte = huge_ptep_get(pte); 1835 struct numa_maps *md; 1836 struct page *page; 1837 1838 if (!pte_present(huge_pte)) 1839 return 0; 1840 1841 page = pte_page(huge_pte); 1842 if (!page) 1843 return 0; 1844 1845 md = walk->private; 1846 gather_stats(page, md, pte_dirty(huge_pte), 1); 1847 return 0; 1848 } 1849 1850 #else 1851 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask, 1852 unsigned long addr, unsigned long end, struct mm_walk *walk) 1853 { 1854 return 0; 1855 } 1856 #endif 1857 1858 static const struct mm_walk_ops show_numa_ops = { 1859 .hugetlb_entry = gather_hugetlb_stats, 1860 .pmd_entry = gather_pte_stats, 1861 }; 1862 1863 /* 1864 * Display pages allocated per node and memory policy via /proc. 1865 */ 1866 static int show_numa_map(struct seq_file *m, void *v) 1867 { 1868 struct numa_maps_private *numa_priv = m->private; 1869 struct proc_maps_private *proc_priv = &numa_priv->proc_maps; 1870 struct vm_area_struct *vma = v; 1871 struct numa_maps *md = &numa_priv->md; 1872 struct file *file = vma->vm_file; 1873 struct mm_struct *mm = vma->vm_mm; 1874 struct mempolicy *pol; 1875 char buffer[64]; 1876 int nid; 1877 1878 if (!mm) 1879 return 0; 1880 1881 /* Ensure we start with an empty set of numa_maps statistics. */ 1882 memset(md, 0, sizeof(*md)); 1883 1884 pol = __get_vma_policy(vma, vma->vm_start); 1885 if (pol) { 1886 mpol_to_str(buffer, sizeof(buffer), pol); 1887 mpol_cond_put(pol); 1888 } else { 1889 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy); 1890 } 1891 1892 seq_printf(m, "%08lx %s", vma->vm_start, buffer); 1893 1894 if (file) { 1895 seq_puts(m, " file="); 1896 seq_file_path(m, file, "\n\t= "); 1897 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) { 1898 seq_puts(m, " heap"); 1899 } else if (is_stack(vma)) { 1900 seq_puts(m, " stack"); 1901 } 1902 1903 if (is_vm_hugetlb_page(vma)) 1904 seq_puts(m, " huge"); 1905 1906 /* mmap_lock is held by m_start */ 1907 walk_page_vma(vma, &show_numa_ops, md); 1908 1909 if (!md->pages) 1910 goto out; 1911 1912 if (md->anon) 1913 seq_printf(m, " anon=%lu", md->anon); 1914 1915 if (md->dirty) 1916 seq_printf(m, " dirty=%lu", md->dirty); 1917 1918 if (md->pages != md->anon && md->pages != md->dirty) 1919 seq_printf(m, " mapped=%lu", md->pages); 1920 1921 if (md->mapcount_max > 1) 1922 seq_printf(m, " mapmax=%lu", md->mapcount_max); 1923 1924 if (md->swapcache) 1925 seq_printf(m, " swapcache=%lu", md->swapcache); 1926 1927 if (md->active < md->pages && !is_vm_hugetlb_page(vma)) 1928 seq_printf(m, " active=%lu", md->active); 1929 1930 if (md->writeback) 1931 seq_printf(m, " writeback=%lu", md->writeback); 1932 1933 for_each_node_state(nid, N_MEMORY) 1934 if (md->node[nid]) 1935 seq_printf(m, " N%d=%lu", nid, md->node[nid]); 1936 1937 seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10); 1938 out: 1939 seq_putc(m, '\n'); 1940 return 0; 1941 } 1942 1943 static const struct seq_operations proc_pid_numa_maps_op = { 1944 .start = m_start, 1945 .next = m_next, 1946 .stop = m_stop, 1947 .show = show_numa_map, 1948 }; 1949 1950 static int pid_numa_maps_open(struct inode *inode, struct file *file) 1951 { 1952 return proc_maps_open(inode, file, &proc_pid_numa_maps_op, 1953 sizeof(struct numa_maps_private)); 1954 } 1955 1956 const struct file_operations proc_pid_numa_maps_operations = { 1957 .open = pid_numa_maps_open, 1958 .read = seq_read, 1959 .llseek = seq_lseek, 1960 .release = proc_map_release, 1961 }; 1962 1963 #endif /* CONFIG_NUMA */ 1964