1 // SPDX-License-Identifier: GPL-2.0 2 #include <linux/pagewalk.h> 3 #include <linux/mm_inline.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 struct vm_area_struct *proc_get_vma(struct proc_maps_private *priv, 127 loff_t *ppos) 128 { 129 struct vm_area_struct *vma = vma_next(&priv->iter); 130 131 if (vma) { 132 *ppos = vma->vm_start; 133 } else { 134 *ppos = -2UL; 135 vma = get_gate_vma(priv->mm); 136 } 137 138 return vma; 139 } 140 141 static void *m_start(struct seq_file *m, loff_t *ppos) 142 { 143 struct proc_maps_private *priv = m->private; 144 unsigned long last_addr = *ppos; 145 struct mm_struct *mm; 146 147 /* See m_next(). Zero at the start or after lseek. */ 148 if (last_addr == -1UL) 149 return NULL; 150 151 priv->task = get_proc_task(priv->inode); 152 if (!priv->task) 153 return ERR_PTR(-ESRCH); 154 155 mm = priv->mm; 156 if (!mm || !mmget_not_zero(mm)) { 157 put_task_struct(priv->task); 158 priv->task = NULL; 159 return NULL; 160 } 161 162 if (mmap_read_lock_killable(mm)) { 163 mmput(mm); 164 put_task_struct(priv->task); 165 priv->task = NULL; 166 return ERR_PTR(-EINTR); 167 } 168 169 vma_iter_init(&priv->iter, mm, last_addr); 170 hold_task_mempolicy(priv); 171 if (last_addr == -2UL) 172 return get_gate_vma(mm); 173 174 return proc_get_vma(priv, ppos); 175 } 176 177 static void *m_next(struct seq_file *m, void *v, loff_t *ppos) 178 { 179 if (*ppos == -2UL) { 180 *ppos = -1UL; 181 return NULL; 182 } 183 return proc_get_vma(m->private, ppos); 184 } 185 186 static void m_stop(struct seq_file *m, void *v) 187 { 188 struct proc_maps_private *priv = m->private; 189 struct mm_struct *mm = priv->mm; 190 191 if (!priv->task) 192 return; 193 194 release_task_mempolicy(priv); 195 mmap_read_unlock(mm); 196 mmput(mm); 197 put_task_struct(priv->task); 198 priv->task = NULL; 199 } 200 201 static int proc_maps_open(struct inode *inode, struct file *file, 202 const struct seq_operations *ops, int psize) 203 { 204 struct proc_maps_private *priv = __seq_open_private(file, ops, psize); 205 206 if (!priv) 207 return -ENOMEM; 208 209 priv->inode = inode; 210 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ); 211 if (IS_ERR(priv->mm)) { 212 int err = PTR_ERR(priv->mm); 213 214 seq_release_private(inode, file); 215 return err; 216 } 217 218 return 0; 219 } 220 221 static int proc_map_release(struct inode *inode, struct file *file) 222 { 223 struct seq_file *seq = file->private_data; 224 struct proc_maps_private *priv = seq->private; 225 226 if (priv->mm) 227 mmdrop(priv->mm); 228 229 return seq_release_private(inode, file); 230 } 231 232 static int do_maps_open(struct inode *inode, struct file *file, 233 const struct seq_operations *ops) 234 { 235 return proc_maps_open(inode, file, ops, 236 sizeof(struct proc_maps_private)); 237 } 238 239 static void show_vma_header_prefix(struct seq_file *m, 240 unsigned long start, unsigned long end, 241 vm_flags_t flags, unsigned long long pgoff, 242 dev_t dev, unsigned long ino) 243 { 244 seq_setwidth(m, 25 + sizeof(void *) * 6 - 1); 245 seq_put_hex_ll(m, NULL, start, 8); 246 seq_put_hex_ll(m, "-", end, 8); 247 seq_putc(m, ' '); 248 seq_putc(m, flags & VM_READ ? 'r' : '-'); 249 seq_putc(m, flags & VM_WRITE ? 'w' : '-'); 250 seq_putc(m, flags & VM_EXEC ? 'x' : '-'); 251 seq_putc(m, flags & VM_MAYSHARE ? 's' : 'p'); 252 seq_put_hex_ll(m, " ", pgoff, 8); 253 seq_put_hex_ll(m, " ", MAJOR(dev), 2); 254 seq_put_hex_ll(m, ":", MINOR(dev), 2); 255 seq_put_decimal_ull(m, " ", ino); 256 seq_putc(m, ' '); 257 } 258 259 static void 260 show_map_vma(struct seq_file *m, struct vm_area_struct *vma) 261 { 262 struct anon_vma_name *anon_name = NULL; 263 struct mm_struct *mm = vma->vm_mm; 264 struct file *file = vma->vm_file; 265 vm_flags_t flags = vma->vm_flags; 266 unsigned long ino = 0; 267 unsigned long long pgoff = 0; 268 unsigned long start, end; 269 dev_t dev = 0; 270 const char *name = NULL; 271 272 if (file) { 273 struct inode *inode = file_inode(vma->vm_file); 274 dev = inode->i_sb->s_dev; 275 ino = inode->i_ino; 276 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT; 277 } 278 279 start = vma->vm_start; 280 end = vma->vm_end; 281 show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino); 282 if (mm) 283 anon_name = anon_vma_name(vma); 284 285 /* 286 * Print the dentry name for named mappings, and a 287 * special [heap] marker for the heap: 288 */ 289 if (file) { 290 seq_pad(m, ' '); 291 /* 292 * If user named this anon shared memory via 293 * prctl(PR_SET_VMA ..., use the provided name. 294 */ 295 if (anon_name) 296 seq_printf(m, "[anon_shmem:%s]", anon_name->name); 297 else 298 seq_file_path(m, file, "\n"); 299 goto done; 300 } 301 302 if (vma->vm_ops && vma->vm_ops->name) { 303 name = vma->vm_ops->name(vma); 304 if (name) 305 goto done; 306 } 307 308 name = arch_vma_name(vma); 309 if (!name) { 310 if (!mm) { 311 name = "[vdso]"; 312 goto done; 313 } 314 315 if (vma_is_initial_heap(vma)) { 316 name = "[heap]"; 317 goto done; 318 } 319 320 if (vma_is_initial_stack(vma)) { 321 name = "[stack]"; 322 goto done; 323 } 324 325 if (anon_name) { 326 seq_pad(m, ' '); 327 seq_printf(m, "[anon:%s]", anon_name->name); 328 } 329 } 330 331 done: 332 if (name) { 333 seq_pad(m, ' '); 334 seq_puts(m, name); 335 } 336 seq_putc(m, '\n'); 337 } 338 339 static int show_map(struct seq_file *m, void *v) 340 { 341 show_map_vma(m, v); 342 return 0; 343 } 344 345 static const struct seq_operations proc_pid_maps_op = { 346 .start = m_start, 347 .next = m_next, 348 .stop = m_stop, 349 .show = show_map 350 }; 351 352 static int pid_maps_open(struct inode *inode, struct file *file) 353 { 354 return do_maps_open(inode, file, &proc_pid_maps_op); 355 } 356 357 const struct file_operations proc_pid_maps_operations = { 358 .open = pid_maps_open, 359 .read = seq_read, 360 .llseek = seq_lseek, 361 .release = proc_map_release, 362 }; 363 364 /* 365 * Proportional Set Size(PSS): my share of RSS. 366 * 367 * PSS of a process is the count of pages it has in memory, where each 368 * page is divided by the number of processes sharing it. So if a 369 * process has 1000 pages all to itself, and 1000 shared with one other 370 * process, its PSS will be 1500. 371 * 372 * To keep (accumulated) division errors low, we adopt a 64bit 373 * fixed-point pss counter to minimize division errors. So (pss >> 374 * PSS_SHIFT) would be the real byte count. 375 * 376 * A shift of 12 before division means (assuming 4K page size): 377 * - 1M 3-user-pages add up to 8KB errors; 378 * - supports mapcount up to 2^24, or 16M; 379 * - supports PSS up to 2^52 bytes, or 4PB. 380 */ 381 #define PSS_SHIFT 12 382 383 #ifdef CONFIG_PROC_PAGE_MONITOR 384 struct mem_size_stats { 385 unsigned long resident; 386 unsigned long shared_clean; 387 unsigned long shared_dirty; 388 unsigned long private_clean; 389 unsigned long private_dirty; 390 unsigned long referenced; 391 unsigned long anonymous; 392 unsigned long lazyfree; 393 unsigned long anonymous_thp; 394 unsigned long shmem_thp; 395 unsigned long file_thp; 396 unsigned long swap; 397 unsigned long shared_hugetlb; 398 unsigned long private_hugetlb; 399 u64 pss; 400 u64 pss_anon; 401 u64 pss_file; 402 u64 pss_shmem; 403 u64 pss_dirty; 404 u64 pss_locked; 405 u64 swap_pss; 406 }; 407 408 static void smaps_page_accumulate(struct mem_size_stats *mss, 409 struct page *page, unsigned long size, unsigned long pss, 410 bool dirty, bool locked, bool private) 411 { 412 mss->pss += pss; 413 414 if (PageAnon(page)) 415 mss->pss_anon += pss; 416 else if (PageSwapBacked(page)) 417 mss->pss_shmem += pss; 418 else 419 mss->pss_file += pss; 420 421 if (locked) 422 mss->pss_locked += pss; 423 424 if (dirty || PageDirty(page)) { 425 mss->pss_dirty += pss; 426 if (private) 427 mss->private_dirty += size; 428 else 429 mss->shared_dirty += size; 430 } else { 431 if (private) 432 mss->private_clean += size; 433 else 434 mss->shared_clean += size; 435 } 436 } 437 438 static void smaps_account(struct mem_size_stats *mss, struct page *page, 439 bool compound, bool young, bool dirty, bool locked, 440 bool migration) 441 { 442 int i, nr = compound ? compound_nr(page) : 1; 443 unsigned long size = nr * PAGE_SIZE; 444 445 /* 446 * First accumulate quantities that depend only on |size| and the type 447 * of the compound page. 448 */ 449 if (PageAnon(page)) { 450 mss->anonymous += size; 451 if (!PageSwapBacked(page) && !dirty && !PageDirty(page)) 452 mss->lazyfree += size; 453 } 454 455 mss->resident += size; 456 /* Accumulate the size in pages that have been accessed. */ 457 if (young || page_is_young(page) || PageReferenced(page)) 458 mss->referenced += size; 459 460 /* 461 * Then accumulate quantities that may depend on sharing, or that may 462 * differ page-by-page. 463 * 464 * page_count(page) == 1 guarantees the page is mapped exactly once. 465 * If any subpage of the compound page mapped with PTE it would elevate 466 * page_count(). 467 * 468 * The page_mapcount() is called to get a snapshot of the mapcount. 469 * Without holding the page lock this snapshot can be slightly wrong as 470 * we cannot always read the mapcount atomically. It is not safe to 471 * call page_mapcount() even with PTL held if the page is not mapped, 472 * especially for migration entries. Treat regular migration entries 473 * as mapcount == 1. 474 */ 475 if ((page_count(page) == 1) || migration) { 476 smaps_page_accumulate(mss, page, size, size << PSS_SHIFT, dirty, 477 locked, true); 478 return; 479 } 480 for (i = 0; i < nr; i++, page++) { 481 int mapcount = page_mapcount(page); 482 unsigned long pss = PAGE_SIZE << PSS_SHIFT; 483 if (mapcount >= 2) 484 pss /= mapcount; 485 smaps_page_accumulate(mss, page, PAGE_SIZE, pss, dirty, locked, 486 mapcount < 2); 487 } 488 } 489 490 #ifdef CONFIG_SHMEM 491 static int smaps_pte_hole(unsigned long addr, unsigned long end, 492 __always_unused int depth, struct mm_walk *walk) 493 { 494 struct mem_size_stats *mss = walk->private; 495 struct vm_area_struct *vma = walk->vma; 496 497 mss->swap += shmem_partial_swap_usage(walk->vma->vm_file->f_mapping, 498 linear_page_index(vma, addr), 499 linear_page_index(vma, end)); 500 501 return 0; 502 } 503 #else 504 #define smaps_pte_hole NULL 505 #endif /* CONFIG_SHMEM */ 506 507 static void smaps_pte_hole_lookup(unsigned long addr, struct mm_walk *walk) 508 { 509 #ifdef CONFIG_SHMEM 510 if (walk->ops->pte_hole) { 511 /* depth is not used */ 512 smaps_pte_hole(addr, addr + PAGE_SIZE, 0, walk); 513 } 514 #endif 515 } 516 517 static void smaps_pte_entry(pte_t *pte, unsigned long addr, 518 struct mm_walk *walk) 519 { 520 struct mem_size_stats *mss = walk->private; 521 struct vm_area_struct *vma = walk->vma; 522 bool locked = !!(vma->vm_flags & VM_LOCKED); 523 struct page *page = NULL; 524 bool migration = false, young = false, dirty = false; 525 pte_t ptent = ptep_get(pte); 526 527 if (pte_present(ptent)) { 528 page = vm_normal_page(vma, addr, ptent); 529 young = pte_young(ptent); 530 dirty = pte_dirty(ptent); 531 } else if (is_swap_pte(ptent)) { 532 swp_entry_t swpent = pte_to_swp_entry(ptent); 533 534 if (!non_swap_entry(swpent)) { 535 int mapcount; 536 537 mss->swap += PAGE_SIZE; 538 mapcount = swp_swapcount(swpent); 539 if (mapcount >= 2) { 540 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT; 541 542 do_div(pss_delta, mapcount); 543 mss->swap_pss += pss_delta; 544 } else { 545 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT; 546 } 547 } else if (is_pfn_swap_entry(swpent)) { 548 if (is_migration_entry(swpent)) 549 migration = true; 550 page = pfn_swap_entry_to_page(swpent); 551 } 552 } else { 553 smaps_pte_hole_lookup(addr, walk); 554 return; 555 } 556 557 if (!page) 558 return; 559 560 smaps_account(mss, page, false, young, dirty, locked, migration); 561 } 562 563 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 564 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr, 565 struct mm_walk *walk) 566 { 567 struct mem_size_stats *mss = walk->private; 568 struct vm_area_struct *vma = walk->vma; 569 bool locked = !!(vma->vm_flags & VM_LOCKED); 570 struct page *page = NULL; 571 bool migration = false; 572 573 if (pmd_present(*pmd)) { 574 page = vm_normal_page_pmd(vma, addr, *pmd); 575 } else if (unlikely(thp_migration_supported() && is_swap_pmd(*pmd))) { 576 swp_entry_t entry = pmd_to_swp_entry(*pmd); 577 578 if (is_migration_entry(entry)) { 579 migration = true; 580 page = pfn_swap_entry_to_page(entry); 581 } 582 } 583 if (IS_ERR_OR_NULL(page)) 584 return; 585 if (PageAnon(page)) 586 mss->anonymous_thp += HPAGE_PMD_SIZE; 587 else if (PageSwapBacked(page)) 588 mss->shmem_thp += HPAGE_PMD_SIZE; 589 else if (is_zone_device_page(page)) 590 /* pass */; 591 else 592 mss->file_thp += HPAGE_PMD_SIZE; 593 594 smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd), 595 locked, migration); 596 } 597 #else 598 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr, 599 struct mm_walk *walk) 600 { 601 } 602 #endif 603 604 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, 605 struct mm_walk *walk) 606 { 607 struct vm_area_struct *vma = walk->vma; 608 pte_t *pte; 609 spinlock_t *ptl; 610 611 ptl = pmd_trans_huge_lock(pmd, vma); 612 if (ptl) { 613 smaps_pmd_entry(pmd, addr, walk); 614 spin_unlock(ptl); 615 goto out; 616 } 617 618 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); 619 if (!pte) { 620 walk->action = ACTION_AGAIN; 621 return 0; 622 } 623 for (; addr != end; pte++, addr += PAGE_SIZE) 624 smaps_pte_entry(pte, addr, walk); 625 pte_unmap_unlock(pte - 1, ptl); 626 out: 627 cond_resched(); 628 return 0; 629 } 630 631 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma) 632 { 633 /* 634 * Don't forget to update Documentation/ on changes. 635 */ 636 static const char mnemonics[BITS_PER_LONG][2] = { 637 /* 638 * In case if we meet a flag we don't know about. 639 */ 640 [0 ... (BITS_PER_LONG-1)] = "??", 641 642 [ilog2(VM_READ)] = "rd", 643 [ilog2(VM_WRITE)] = "wr", 644 [ilog2(VM_EXEC)] = "ex", 645 [ilog2(VM_SHARED)] = "sh", 646 [ilog2(VM_MAYREAD)] = "mr", 647 [ilog2(VM_MAYWRITE)] = "mw", 648 [ilog2(VM_MAYEXEC)] = "me", 649 [ilog2(VM_MAYSHARE)] = "ms", 650 [ilog2(VM_GROWSDOWN)] = "gd", 651 [ilog2(VM_PFNMAP)] = "pf", 652 [ilog2(VM_LOCKED)] = "lo", 653 [ilog2(VM_IO)] = "io", 654 [ilog2(VM_SEQ_READ)] = "sr", 655 [ilog2(VM_RAND_READ)] = "rr", 656 [ilog2(VM_DONTCOPY)] = "dc", 657 [ilog2(VM_DONTEXPAND)] = "de", 658 [ilog2(VM_LOCKONFAULT)] = "lf", 659 [ilog2(VM_ACCOUNT)] = "ac", 660 [ilog2(VM_NORESERVE)] = "nr", 661 [ilog2(VM_HUGETLB)] = "ht", 662 [ilog2(VM_SYNC)] = "sf", 663 [ilog2(VM_ARCH_1)] = "ar", 664 [ilog2(VM_WIPEONFORK)] = "wf", 665 [ilog2(VM_DONTDUMP)] = "dd", 666 #ifdef CONFIG_ARM64_BTI 667 [ilog2(VM_ARM64_BTI)] = "bt", 668 #endif 669 #ifdef CONFIG_MEM_SOFT_DIRTY 670 [ilog2(VM_SOFTDIRTY)] = "sd", 671 #endif 672 [ilog2(VM_MIXEDMAP)] = "mm", 673 [ilog2(VM_HUGEPAGE)] = "hg", 674 [ilog2(VM_NOHUGEPAGE)] = "nh", 675 [ilog2(VM_MERGEABLE)] = "mg", 676 [ilog2(VM_UFFD_MISSING)]= "um", 677 [ilog2(VM_UFFD_WP)] = "uw", 678 #ifdef CONFIG_ARM64_MTE 679 [ilog2(VM_MTE)] = "mt", 680 [ilog2(VM_MTE_ALLOWED)] = "", 681 #endif 682 #ifdef CONFIG_ARCH_HAS_PKEYS 683 /* These come out via ProtectionKey: */ 684 [ilog2(VM_PKEY_BIT0)] = "", 685 [ilog2(VM_PKEY_BIT1)] = "", 686 [ilog2(VM_PKEY_BIT2)] = "", 687 [ilog2(VM_PKEY_BIT3)] = "", 688 #if VM_PKEY_BIT4 689 [ilog2(VM_PKEY_BIT4)] = "", 690 #endif 691 #endif /* CONFIG_ARCH_HAS_PKEYS */ 692 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_MINOR 693 [ilog2(VM_UFFD_MINOR)] = "ui", 694 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_MINOR */ 695 }; 696 size_t i; 697 698 seq_puts(m, "VmFlags: "); 699 for (i = 0; i < BITS_PER_LONG; i++) { 700 if (!mnemonics[i][0]) 701 continue; 702 if (vma->vm_flags & (1UL << i)) { 703 seq_putc(m, mnemonics[i][0]); 704 seq_putc(m, mnemonics[i][1]); 705 seq_putc(m, ' '); 706 } 707 } 708 seq_putc(m, '\n'); 709 } 710 711 #ifdef CONFIG_HUGETLB_PAGE 712 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask, 713 unsigned long addr, unsigned long end, 714 struct mm_walk *walk) 715 { 716 struct mem_size_stats *mss = walk->private; 717 struct vm_area_struct *vma = walk->vma; 718 struct page *page = NULL; 719 pte_t ptent = ptep_get(pte); 720 721 if (pte_present(ptent)) { 722 page = vm_normal_page(vma, addr, ptent); 723 } else if (is_swap_pte(ptent)) { 724 swp_entry_t swpent = pte_to_swp_entry(ptent); 725 726 if (is_pfn_swap_entry(swpent)) 727 page = pfn_swap_entry_to_page(swpent); 728 } 729 if (page) { 730 if (page_mapcount(page) >= 2 || hugetlb_pmd_shared(pte)) 731 mss->shared_hugetlb += huge_page_size(hstate_vma(vma)); 732 else 733 mss->private_hugetlb += huge_page_size(hstate_vma(vma)); 734 } 735 return 0; 736 } 737 #else 738 #define smaps_hugetlb_range NULL 739 #endif /* HUGETLB_PAGE */ 740 741 static const struct mm_walk_ops smaps_walk_ops = { 742 .pmd_entry = smaps_pte_range, 743 .hugetlb_entry = smaps_hugetlb_range, 744 .walk_lock = PGWALK_RDLOCK, 745 }; 746 747 static const struct mm_walk_ops smaps_shmem_walk_ops = { 748 .pmd_entry = smaps_pte_range, 749 .hugetlb_entry = smaps_hugetlb_range, 750 .pte_hole = smaps_pte_hole, 751 .walk_lock = PGWALK_RDLOCK, 752 }; 753 754 /* 755 * Gather mem stats from @vma with the indicated beginning 756 * address @start, and keep them in @mss. 757 * 758 * Use vm_start of @vma as the beginning address if @start is 0. 759 */ 760 static void smap_gather_stats(struct vm_area_struct *vma, 761 struct mem_size_stats *mss, unsigned long start) 762 { 763 const struct mm_walk_ops *ops = &smaps_walk_ops; 764 765 /* Invalid start */ 766 if (start >= vma->vm_end) 767 return; 768 769 if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) { 770 /* 771 * For shared or readonly shmem mappings we know that all 772 * swapped out pages belong to the shmem object, and we can 773 * obtain the swap value much more efficiently. For private 774 * writable mappings, we might have COW pages that are 775 * not affected by the parent swapped out pages of the shmem 776 * object, so we have to distinguish them during the page walk. 777 * Unless we know that the shmem object (or the part mapped by 778 * our VMA) has no swapped out pages at all. 779 */ 780 unsigned long shmem_swapped = shmem_swap_usage(vma); 781 782 if (!start && (!shmem_swapped || (vma->vm_flags & VM_SHARED) || 783 !(vma->vm_flags & VM_WRITE))) { 784 mss->swap += shmem_swapped; 785 } else { 786 ops = &smaps_shmem_walk_ops; 787 } 788 } 789 790 /* mmap_lock is held in m_start */ 791 if (!start) 792 walk_page_vma(vma, ops, mss); 793 else 794 walk_page_range(vma->vm_mm, start, vma->vm_end, ops, mss); 795 } 796 797 #define SEQ_PUT_DEC(str, val) \ 798 seq_put_decimal_ull_width(m, str, (val) >> 10, 8) 799 800 /* Show the contents common for smaps and smaps_rollup */ 801 static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss, 802 bool rollup_mode) 803 { 804 SEQ_PUT_DEC("Rss: ", mss->resident); 805 SEQ_PUT_DEC(" kB\nPss: ", mss->pss >> PSS_SHIFT); 806 SEQ_PUT_DEC(" kB\nPss_Dirty: ", mss->pss_dirty >> PSS_SHIFT); 807 if (rollup_mode) { 808 /* 809 * These are meaningful only for smaps_rollup, otherwise two of 810 * them are zero, and the other one is the same as Pss. 811 */ 812 SEQ_PUT_DEC(" kB\nPss_Anon: ", 813 mss->pss_anon >> PSS_SHIFT); 814 SEQ_PUT_DEC(" kB\nPss_File: ", 815 mss->pss_file >> PSS_SHIFT); 816 SEQ_PUT_DEC(" kB\nPss_Shmem: ", 817 mss->pss_shmem >> PSS_SHIFT); 818 } 819 SEQ_PUT_DEC(" kB\nShared_Clean: ", mss->shared_clean); 820 SEQ_PUT_DEC(" kB\nShared_Dirty: ", mss->shared_dirty); 821 SEQ_PUT_DEC(" kB\nPrivate_Clean: ", mss->private_clean); 822 SEQ_PUT_DEC(" kB\nPrivate_Dirty: ", mss->private_dirty); 823 SEQ_PUT_DEC(" kB\nReferenced: ", mss->referenced); 824 SEQ_PUT_DEC(" kB\nAnonymous: ", mss->anonymous); 825 SEQ_PUT_DEC(" kB\nLazyFree: ", mss->lazyfree); 826 SEQ_PUT_DEC(" kB\nAnonHugePages: ", mss->anonymous_thp); 827 SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp); 828 SEQ_PUT_DEC(" kB\nFilePmdMapped: ", mss->file_thp); 829 SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb); 830 seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ", 831 mss->private_hugetlb >> 10, 7); 832 SEQ_PUT_DEC(" kB\nSwap: ", mss->swap); 833 SEQ_PUT_DEC(" kB\nSwapPss: ", 834 mss->swap_pss >> PSS_SHIFT); 835 SEQ_PUT_DEC(" kB\nLocked: ", 836 mss->pss_locked >> PSS_SHIFT); 837 seq_puts(m, " kB\n"); 838 } 839 840 static int show_smap(struct seq_file *m, void *v) 841 { 842 struct vm_area_struct *vma = v; 843 struct mem_size_stats mss; 844 845 memset(&mss, 0, sizeof(mss)); 846 847 smap_gather_stats(vma, &mss, 0); 848 849 show_map_vma(m, vma); 850 851 SEQ_PUT_DEC("Size: ", vma->vm_end - vma->vm_start); 852 SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma)); 853 SEQ_PUT_DEC(" kB\nMMUPageSize: ", vma_mmu_pagesize(vma)); 854 seq_puts(m, " kB\n"); 855 856 __show_smap(m, &mss, false); 857 858 seq_printf(m, "THPeligible: %8u\n", 859 hugepage_vma_check(vma, vma->vm_flags, true, false, true)); 860 861 if (arch_pkeys_enabled()) 862 seq_printf(m, "ProtectionKey: %8u\n", vma_pkey(vma)); 863 show_smap_vma_flags(m, vma); 864 865 return 0; 866 } 867 868 static int show_smaps_rollup(struct seq_file *m, void *v) 869 { 870 struct proc_maps_private *priv = m->private; 871 struct mem_size_stats mss; 872 struct mm_struct *mm = priv->mm; 873 struct vm_area_struct *vma; 874 unsigned long vma_start = 0, last_vma_end = 0; 875 int ret = 0; 876 VMA_ITERATOR(vmi, mm, 0); 877 878 priv->task = get_proc_task(priv->inode); 879 if (!priv->task) 880 return -ESRCH; 881 882 if (!mm || !mmget_not_zero(mm)) { 883 ret = -ESRCH; 884 goto out_put_task; 885 } 886 887 memset(&mss, 0, sizeof(mss)); 888 889 ret = mmap_read_lock_killable(mm); 890 if (ret) 891 goto out_put_mm; 892 893 hold_task_mempolicy(priv); 894 vma = vma_next(&vmi); 895 896 if (unlikely(!vma)) 897 goto empty_set; 898 899 vma_start = vma->vm_start; 900 do { 901 smap_gather_stats(vma, &mss, 0); 902 last_vma_end = vma->vm_end; 903 904 /* 905 * Release mmap_lock temporarily if someone wants to 906 * access it for write request. 907 */ 908 if (mmap_lock_is_contended(mm)) { 909 vma_iter_invalidate(&vmi); 910 mmap_read_unlock(mm); 911 ret = mmap_read_lock_killable(mm); 912 if (ret) { 913 release_task_mempolicy(priv); 914 goto out_put_mm; 915 } 916 917 /* 918 * After dropping the lock, there are four cases to 919 * consider. See the following example for explanation. 920 * 921 * +------+------+-----------+ 922 * | VMA1 | VMA2 | VMA3 | 923 * +------+------+-----------+ 924 * | | | | 925 * 4k 8k 16k 400k 926 * 927 * Suppose we drop the lock after reading VMA2 due to 928 * contention, then we get: 929 * 930 * last_vma_end = 16k 931 * 932 * 1) VMA2 is freed, but VMA3 exists: 933 * 934 * vma_next(vmi) will return VMA3. 935 * In this case, just continue from VMA3. 936 * 937 * 2) VMA2 still exists: 938 * 939 * vma_next(vmi) will return VMA3. 940 * In this case, just continue from VMA3. 941 * 942 * 3) No more VMAs can be found: 943 * 944 * vma_next(vmi) will return NULL. 945 * No more things to do, just break. 946 * 947 * 4) (last_vma_end - 1) is the middle of a vma (VMA'): 948 * 949 * vma_next(vmi) will return VMA' whose range 950 * contains last_vma_end. 951 * Iterate VMA' from last_vma_end. 952 */ 953 vma = vma_next(&vmi); 954 /* Case 3 above */ 955 if (!vma) 956 break; 957 958 /* Case 1 and 2 above */ 959 if (vma->vm_start >= last_vma_end) 960 continue; 961 962 /* Case 4 above */ 963 if (vma->vm_end > last_vma_end) 964 smap_gather_stats(vma, &mss, last_vma_end); 965 } 966 } for_each_vma(vmi, vma); 967 968 empty_set: 969 show_vma_header_prefix(m, vma_start, last_vma_end, 0, 0, 0, 0); 970 seq_pad(m, ' '); 971 seq_puts(m, "[rollup]\n"); 972 973 __show_smap(m, &mss, true); 974 975 release_task_mempolicy(priv); 976 mmap_read_unlock(mm); 977 978 out_put_mm: 979 mmput(mm); 980 out_put_task: 981 put_task_struct(priv->task); 982 priv->task = NULL; 983 984 return ret; 985 } 986 #undef SEQ_PUT_DEC 987 988 static const struct seq_operations proc_pid_smaps_op = { 989 .start = m_start, 990 .next = m_next, 991 .stop = m_stop, 992 .show = show_smap 993 }; 994 995 static int pid_smaps_open(struct inode *inode, struct file *file) 996 { 997 return do_maps_open(inode, file, &proc_pid_smaps_op); 998 } 999 1000 static int smaps_rollup_open(struct inode *inode, struct file *file) 1001 { 1002 int ret; 1003 struct proc_maps_private *priv; 1004 1005 priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT); 1006 if (!priv) 1007 return -ENOMEM; 1008 1009 ret = single_open(file, show_smaps_rollup, priv); 1010 if (ret) 1011 goto out_free; 1012 1013 priv->inode = inode; 1014 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ); 1015 if (IS_ERR(priv->mm)) { 1016 ret = PTR_ERR(priv->mm); 1017 1018 single_release(inode, file); 1019 goto out_free; 1020 } 1021 1022 return 0; 1023 1024 out_free: 1025 kfree(priv); 1026 return ret; 1027 } 1028 1029 static int smaps_rollup_release(struct inode *inode, struct file *file) 1030 { 1031 struct seq_file *seq = file->private_data; 1032 struct proc_maps_private *priv = seq->private; 1033 1034 if (priv->mm) 1035 mmdrop(priv->mm); 1036 1037 kfree(priv); 1038 return single_release(inode, file); 1039 } 1040 1041 const struct file_operations proc_pid_smaps_operations = { 1042 .open = pid_smaps_open, 1043 .read = seq_read, 1044 .llseek = seq_lseek, 1045 .release = proc_map_release, 1046 }; 1047 1048 const struct file_operations proc_pid_smaps_rollup_operations = { 1049 .open = smaps_rollup_open, 1050 .read = seq_read, 1051 .llseek = seq_lseek, 1052 .release = smaps_rollup_release, 1053 }; 1054 1055 enum clear_refs_types { 1056 CLEAR_REFS_ALL = 1, 1057 CLEAR_REFS_ANON, 1058 CLEAR_REFS_MAPPED, 1059 CLEAR_REFS_SOFT_DIRTY, 1060 CLEAR_REFS_MM_HIWATER_RSS, 1061 CLEAR_REFS_LAST, 1062 }; 1063 1064 struct clear_refs_private { 1065 enum clear_refs_types type; 1066 }; 1067 1068 #ifdef CONFIG_MEM_SOFT_DIRTY 1069 1070 static inline bool pte_is_pinned(struct vm_area_struct *vma, unsigned long addr, pte_t pte) 1071 { 1072 struct page *page; 1073 1074 if (!pte_write(pte)) 1075 return false; 1076 if (!is_cow_mapping(vma->vm_flags)) 1077 return false; 1078 if (likely(!test_bit(MMF_HAS_PINNED, &vma->vm_mm->flags))) 1079 return false; 1080 page = vm_normal_page(vma, addr, pte); 1081 if (!page) 1082 return false; 1083 return page_maybe_dma_pinned(page); 1084 } 1085 1086 static inline void clear_soft_dirty(struct vm_area_struct *vma, 1087 unsigned long addr, pte_t *pte) 1088 { 1089 /* 1090 * The soft-dirty tracker uses #PF-s to catch writes 1091 * to pages, so write-protect the pte as well. See the 1092 * Documentation/admin-guide/mm/soft-dirty.rst for full description 1093 * of how soft-dirty works. 1094 */ 1095 pte_t ptent = ptep_get(pte); 1096 1097 if (pte_present(ptent)) { 1098 pte_t old_pte; 1099 1100 if (pte_is_pinned(vma, addr, ptent)) 1101 return; 1102 old_pte = ptep_modify_prot_start(vma, addr, pte); 1103 ptent = pte_wrprotect(old_pte); 1104 ptent = pte_clear_soft_dirty(ptent); 1105 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent); 1106 } else if (is_swap_pte(ptent)) { 1107 ptent = pte_swp_clear_soft_dirty(ptent); 1108 set_pte_at(vma->vm_mm, addr, pte, ptent); 1109 } 1110 } 1111 #else 1112 static inline void clear_soft_dirty(struct vm_area_struct *vma, 1113 unsigned long addr, pte_t *pte) 1114 { 1115 } 1116 #endif 1117 1118 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE) 1119 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma, 1120 unsigned long addr, pmd_t *pmdp) 1121 { 1122 pmd_t old, pmd = *pmdp; 1123 1124 if (pmd_present(pmd)) { 1125 /* See comment in change_huge_pmd() */ 1126 old = pmdp_invalidate(vma, addr, pmdp); 1127 if (pmd_dirty(old)) 1128 pmd = pmd_mkdirty(pmd); 1129 if (pmd_young(old)) 1130 pmd = pmd_mkyoung(pmd); 1131 1132 pmd = pmd_wrprotect(pmd); 1133 pmd = pmd_clear_soft_dirty(pmd); 1134 1135 set_pmd_at(vma->vm_mm, addr, pmdp, pmd); 1136 } else if (is_migration_entry(pmd_to_swp_entry(pmd))) { 1137 pmd = pmd_swp_clear_soft_dirty(pmd); 1138 set_pmd_at(vma->vm_mm, addr, pmdp, pmd); 1139 } 1140 } 1141 #else 1142 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma, 1143 unsigned long addr, pmd_t *pmdp) 1144 { 1145 } 1146 #endif 1147 1148 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr, 1149 unsigned long end, struct mm_walk *walk) 1150 { 1151 struct clear_refs_private *cp = walk->private; 1152 struct vm_area_struct *vma = walk->vma; 1153 pte_t *pte, ptent; 1154 spinlock_t *ptl; 1155 struct page *page; 1156 1157 ptl = pmd_trans_huge_lock(pmd, vma); 1158 if (ptl) { 1159 if (cp->type == CLEAR_REFS_SOFT_DIRTY) { 1160 clear_soft_dirty_pmd(vma, addr, pmd); 1161 goto out; 1162 } 1163 1164 if (!pmd_present(*pmd)) 1165 goto out; 1166 1167 page = pmd_page(*pmd); 1168 1169 /* Clear accessed and referenced bits. */ 1170 pmdp_test_and_clear_young(vma, addr, pmd); 1171 test_and_clear_page_young(page); 1172 ClearPageReferenced(page); 1173 out: 1174 spin_unlock(ptl); 1175 return 0; 1176 } 1177 1178 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); 1179 if (!pte) { 1180 walk->action = ACTION_AGAIN; 1181 return 0; 1182 } 1183 for (; addr != end; pte++, addr += PAGE_SIZE) { 1184 ptent = ptep_get(pte); 1185 1186 if (cp->type == CLEAR_REFS_SOFT_DIRTY) { 1187 clear_soft_dirty(vma, addr, pte); 1188 continue; 1189 } 1190 1191 if (!pte_present(ptent)) 1192 continue; 1193 1194 page = vm_normal_page(vma, addr, ptent); 1195 if (!page) 1196 continue; 1197 1198 /* Clear accessed and referenced bits. */ 1199 ptep_test_and_clear_young(vma, addr, pte); 1200 test_and_clear_page_young(page); 1201 ClearPageReferenced(page); 1202 } 1203 pte_unmap_unlock(pte - 1, ptl); 1204 cond_resched(); 1205 return 0; 1206 } 1207 1208 static int clear_refs_test_walk(unsigned long start, unsigned long end, 1209 struct mm_walk *walk) 1210 { 1211 struct clear_refs_private *cp = walk->private; 1212 struct vm_area_struct *vma = walk->vma; 1213 1214 if (vma->vm_flags & VM_PFNMAP) 1215 return 1; 1216 1217 /* 1218 * Writing 1 to /proc/pid/clear_refs affects all pages. 1219 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages. 1220 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages. 1221 * Writing 4 to /proc/pid/clear_refs affects all pages. 1222 */ 1223 if (cp->type == CLEAR_REFS_ANON && vma->vm_file) 1224 return 1; 1225 if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file) 1226 return 1; 1227 return 0; 1228 } 1229 1230 static const struct mm_walk_ops clear_refs_walk_ops = { 1231 .pmd_entry = clear_refs_pte_range, 1232 .test_walk = clear_refs_test_walk, 1233 .walk_lock = PGWALK_WRLOCK, 1234 }; 1235 1236 static ssize_t clear_refs_write(struct file *file, const char __user *buf, 1237 size_t count, loff_t *ppos) 1238 { 1239 struct task_struct *task; 1240 char buffer[PROC_NUMBUF]; 1241 struct mm_struct *mm; 1242 struct vm_area_struct *vma; 1243 enum clear_refs_types type; 1244 int itype; 1245 int rv; 1246 1247 memset(buffer, 0, sizeof(buffer)); 1248 if (count > sizeof(buffer) - 1) 1249 count = sizeof(buffer) - 1; 1250 if (copy_from_user(buffer, buf, count)) 1251 return -EFAULT; 1252 rv = kstrtoint(strstrip(buffer), 10, &itype); 1253 if (rv < 0) 1254 return rv; 1255 type = (enum clear_refs_types)itype; 1256 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST) 1257 return -EINVAL; 1258 1259 task = get_proc_task(file_inode(file)); 1260 if (!task) 1261 return -ESRCH; 1262 mm = get_task_mm(task); 1263 if (mm) { 1264 VMA_ITERATOR(vmi, mm, 0); 1265 struct mmu_notifier_range range; 1266 struct clear_refs_private cp = { 1267 .type = type, 1268 }; 1269 1270 if (mmap_write_lock_killable(mm)) { 1271 count = -EINTR; 1272 goto out_mm; 1273 } 1274 if (type == CLEAR_REFS_MM_HIWATER_RSS) { 1275 /* 1276 * Writing 5 to /proc/pid/clear_refs resets the peak 1277 * resident set size to this mm's current rss value. 1278 */ 1279 reset_mm_hiwater_rss(mm); 1280 goto out_unlock; 1281 } 1282 1283 if (type == CLEAR_REFS_SOFT_DIRTY) { 1284 for_each_vma(vmi, vma) { 1285 if (!(vma->vm_flags & VM_SOFTDIRTY)) 1286 continue; 1287 vm_flags_clear(vma, VM_SOFTDIRTY); 1288 vma_set_page_prot(vma); 1289 } 1290 1291 inc_tlb_flush_pending(mm); 1292 mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY, 1293 0, mm, 0, -1UL); 1294 mmu_notifier_invalidate_range_start(&range); 1295 } 1296 walk_page_range(mm, 0, -1, &clear_refs_walk_ops, &cp); 1297 if (type == CLEAR_REFS_SOFT_DIRTY) { 1298 mmu_notifier_invalidate_range_end(&range); 1299 flush_tlb_mm(mm); 1300 dec_tlb_flush_pending(mm); 1301 } 1302 out_unlock: 1303 mmap_write_unlock(mm); 1304 out_mm: 1305 mmput(mm); 1306 } 1307 put_task_struct(task); 1308 1309 return count; 1310 } 1311 1312 const struct file_operations proc_clear_refs_operations = { 1313 .write = clear_refs_write, 1314 .llseek = noop_llseek, 1315 }; 1316 1317 typedef struct { 1318 u64 pme; 1319 } pagemap_entry_t; 1320 1321 struct pagemapread { 1322 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */ 1323 pagemap_entry_t *buffer; 1324 bool show_pfn; 1325 }; 1326 1327 #define PAGEMAP_WALK_SIZE (PMD_SIZE) 1328 #define PAGEMAP_WALK_MASK (PMD_MASK) 1329 1330 #define PM_ENTRY_BYTES sizeof(pagemap_entry_t) 1331 #define PM_PFRAME_BITS 55 1332 #define PM_PFRAME_MASK GENMASK_ULL(PM_PFRAME_BITS - 1, 0) 1333 #define PM_SOFT_DIRTY BIT_ULL(55) 1334 #define PM_MMAP_EXCLUSIVE BIT_ULL(56) 1335 #define PM_UFFD_WP BIT_ULL(57) 1336 #define PM_FILE BIT_ULL(61) 1337 #define PM_SWAP BIT_ULL(62) 1338 #define PM_PRESENT BIT_ULL(63) 1339 1340 #define PM_END_OF_BUFFER 1 1341 1342 static inline pagemap_entry_t make_pme(u64 frame, u64 flags) 1343 { 1344 return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags }; 1345 } 1346 1347 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme, 1348 struct pagemapread *pm) 1349 { 1350 pm->buffer[pm->pos++] = *pme; 1351 if (pm->pos >= pm->len) 1352 return PM_END_OF_BUFFER; 1353 return 0; 1354 } 1355 1356 static int pagemap_pte_hole(unsigned long start, unsigned long end, 1357 __always_unused int depth, struct mm_walk *walk) 1358 { 1359 struct pagemapread *pm = walk->private; 1360 unsigned long addr = start; 1361 int err = 0; 1362 1363 while (addr < end) { 1364 struct vm_area_struct *vma = find_vma(walk->mm, addr); 1365 pagemap_entry_t pme = make_pme(0, 0); 1366 /* End of address space hole, which we mark as non-present. */ 1367 unsigned long hole_end; 1368 1369 if (vma) 1370 hole_end = min(end, vma->vm_start); 1371 else 1372 hole_end = end; 1373 1374 for (; addr < hole_end; addr += PAGE_SIZE) { 1375 err = add_to_pagemap(addr, &pme, pm); 1376 if (err) 1377 goto out; 1378 } 1379 1380 if (!vma) 1381 break; 1382 1383 /* Addresses in the VMA. */ 1384 if (vma->vm_flags & VM_SOFTDIRTY) 1385 pme = make_pme(0, PM_SOFT_DIRTY); 1386 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) { 1387 err = add_to_pagemap(addr, &pme, pm); 1388 if (err) 1389 goto out; 1390 } 1391 } 1392 out: 1393 return err; 1394 } 1395 1396 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm, 1397 struct vm_area_struct *vma, unsigned long addr, pte_t pte) 1398 { 1399 u64 frame = 0, flags = 0; 1400 struct page *page = NULL; 1401 bool migration = false; 1402 1403 if (pte_present(pte)) { 1404 if (pm->show_pfn) 1405 frame = pte_pfn(pte); 1406 flags |= PM_PRESENT; 1407 page = vm_normal_page(vma, addr, pte); 1408 if (pte_soft_dirty(pte)) 1409 flags |= PM_SOFT_DIRTY; 1410 if (pte_uffd_wp(pte)) 1411 flags |= PM_UFFD_WP; 1412 } else if (is_swap_pte(pte)) { 1413 swp_entry_t entry; 1414 if (pte_swp_soft_dirty(pte)) 1415 flags |= PM_SOFT_DIRTY; 1416 if (pte_swp_uffd_wp(pte)) 1417 flags |= PM_UFFD_WP; 1418 entry = pte_to_swp_entry(pte); 1419 if (pm->show_pfn) { 1420 pgoff_t offset; 1421 /* 1422 * For PFN swap offsets, keeping the offset field 1423 * to be PFN only to be compatible with old smaps. 1424 */ 1425 if (is_pfn_swap_entry(entry)) 1426 offset = swp_offset_pfn(entry); 1427 else 1428 offset = swp_offset(entry); 1429 frame = swp_type(entry) | 1430 (offset << MAX_SWAPFILES_SHIFT); 1431 } 1432 flags |= PM_SWAP; 1433 migration = is_migration_entry(entry); 1434 if (is_pfn_swap_entry(entry)) 1435 page = pfn_swap_entry_to_page(entry); 1436 if (pte_marker_entry_uffd_wp(entry)) 1437 flags |= PM_UFFD_WP; 1438 } 1439 1440 if (page && !PageAnon(page)) 1441 flags |= PM_FILE; 1442 if (page && !migration && page_mapcount(page) == 1) 1443 flags |= PM_MMAP_EXCLUSIVE; 1444 if (vma->vm_flags & VM_SOFTDIRTY) 1445 flags |= PM_SOFT_DIRTY; 1446 1447 return make_pme(frame, flags); 1448 } 1449 1450 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end, 1451 struct mm_walk *walk) 1452 { 1453 struct vm_area_struct *vma = walk->vma; 1454 struct pagemapread *pm = walk->private; 1455 spinlock_t *ptl; 1456 pte_t *pte, *orig_pte; 1457 int err = 0; 1458 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1459 bool migration = false; 1460 1461 ptl = pmd_trans_huge_lock(pmdp, vma); 1462 if (ptl) { 1463 u64 flags = 0, frame = 0; 1464 pmd_t pmd = *pmdp; 1465 struct page *page = NULL; 1466 1467 if (vma->vm_flags & VM_SOFTDIRTY) 1468 flags |= PM_SOFT_DIRTY; 1469 1470 if (pmd_present(pmd)) { 1471 page = pmd_page(pmd); 1472 1473 flags |= PM_PRESENT; 1474 if (pmd_soft_dirty(pmd)) 1475 flags |= PM_SOFT_DIRTY; 1476 if (pmd_uffd_wp(pmd)) 1477 flags |= PM_UFFD_WP; 1478 if (pm->show_pfn) 1479 frame = pmd_pfn(pmd) + 1480 ((addr & ~PMD_MASK) >> PAGE_SHIFT); 1481 } 1482 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION 1483 else if (is_swap_pmd(pmd)) { 1484 swp_entry_t entry = pmd_to_swp_entry(pmd); 1485 unsigned long offset; 1486 1487 if (pm->show_pfn) { 1488 if (is_pfn_swap_entry(entry)) 1489 offset = swp_offset_pfn(entry); 1490 else 1491 offset = swp_offset(entry); 1492 offset = offset + 1493 ((addr & ~PMD_MASK) >> PAGE_SHIFT); 1494 frame = swp_type(entry) | 1495 (offset << MAX_SWAPFILES_SHIFT); 1496 } 1497 flags |= PM_SWAP; 1498 if (pmd_swp_soft_dirty(pmd)) 1499 flags |= PM_SOFT_DIRTY; 1500 if (pmd_swp_uffd_wp(pmd)) 1501 flags |= PM_UFFD_WP; 1502 VM_BUG_ON(!is_pmd_migration_entry(pmd)); 1503 migration = is_migration_entry(entry); 1504 page = pfn_swap_entry_to_page(entry); 1505 } 1506 #endif 1507 1508 if (page && !migration && page_mapcount(page) == 1) 1509 flags |= PM_MMAP_EXCLUSIVE; 1510 1511 for (; addr != end; addr += PAGE_SIZE) { 1512 pagemap_entry_t pme = make_pme(frame, flags); 1513 1514 err = add_to_pagemap(addr, &pme, pm); 1515 if (err) 1516 break; 1517 if (pm->show_pfn) { 1518 if (flags & PM_PRESENT) 1519 frame++; 1520 else if (flags & PM_SWAP) 1521 frame += (1 << MAX_SWAPFILES_SHIFT); 1522 } 1523 } 1524 spin_unlock(ptl); 1525 return err; 1526 } 1527 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ 1528 1529 /* 1530 * We can assume that @vma always points to a valid one and @end never 1531 * goes beyond vma->vm_end. 1532 */ 1533 orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl); 1534 if (!pte) { 1535 walk->action = ACTION_AGAIN; 1536 return err; 1537 } 1538 for (; addr < end; pte++, addr += PAGE_SIZE) { 1539 pagemap_entry_t pme; 1540 1541 pme = pte_to_pagemap_entry(pm, vma, addr, ptep_get(pte)); 1542 err = add_to_pagemap(addr, &pme, pm); 1543 if (err) 1544 break; 1545 } 1546 pte_unmap_unlock(orig_pte, ptl); 1547 1548 cond_resched(); 1549 1550 return err; 1551 } 1552 1553 #ifdef CONFIG_HUGETLB_PAGE 1554 /* This function walks within one hugetlb entry in the single call */ 1555 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask, 1556 unsigned long addr, unsigned long end, 1557 struct mm_walk *walk) 1558 { 1559 struct pagemapread *pm = walk->private; 1560 struct vm_area_struct *vma = walk->vma; 1561 u64 flags = 0, frame = 0; 1562 int err = 0; 1563 pte_t pte; 1564 1565 if (vma->vm_flags & VM_SOFTDIRTY) 1566 flags |= PM_SOFT_DIRTY; 1567 1568 pte = huge_ptep_get(ptep); 1569 if (pte_present(pte)) { 1570 struct page *page = pte_page(pte); 1571 1572 if (!PageAnon(page)) 1573 flags |= PM_FILE; 1574 1575 if (page_mapcount(page) == 1) 1576 flags |= PM_MMAP_EXCLUSIVE; 1577 1578 if (huge_pte_uffd_wp(pte)) 1579 flags |= PM_UFFD_WP; 1580 1581 flags |= PM_PRESENT; 1582 if (pm->show_pfn) 1583 frame = pte_pfn(pte) + 1584 ((addr & ~hmask) >> PAGE_SHIFT); 1585 } else if (pte_swp_uffd_wp_any(pte)) { 1586 flags |= PM_UFFD_WP; 1587 } 1588 1589 for (; addr != end; addr += PAGE_SIZE) { 1590 pagemap_entry_t pme = make_pme(frame, flags); 1591 1592 err = add_to_pagemap(addr, &pme, pm); 1593 if (err) 1594 return err; 1595 if (pm->show_pfn && (flags & PM_PRESENT)) 1596 frame++; 1597 } 1598 1599 cond_resched(); 1600 1601 return err; 1602 } 1603 #else 1604 #define pagemap_hugetlb_range NULL 1605 #endif /* HUGETLB_PAGE */ 1606 1607 static const struct mm_walk_ops pagemap_ops = { 1608 .pmd_entry = pagemap_pmd_range, 1609 .pte_hole = pagemap_pte_hole, 1610 .hugetlb_entry = pagemap_hugetlb_range, 1611 .walk_lock = PGWALK_RDLOCK, 1612 }; 1613 1614 /* 1615 * /proc/pid/pagemap - an array mapping virtual pages to pfns 1616 * 1617 * For each page in the address space, this file contains one 64-bit entry 1618 * consisting of the following: 1619 * 1620 * Bits 0-54 page frame number (PFN) if present 1621 * Bits 0-4 swap type if swapped 1622 * Bits 5-54 swap offset if swapped 1623 * Bit 55 pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst) 1624 * Bit 56 page exclusively mapped 1625 * Bit 57 pte is uffd-wp write-protected 1626 * Bits 58-60 zero 1627 * Bit 61 page is file-page or shared-anon 1628 * Bit 62 page swapped 1629 * Bit 63 page present 1630 * 1631 * If the page is not present but in swap, then the PFN contains an 1632 * encoding of the swap file number and the page's offset into the 1633 * swap. Unmapped pages return a null PFN. This allows determining 1634 * precisely which pages are mapped (or in swap) and comparing mapped 1635 * pages between processes. 1636 * 1637 * Efficient users of this interface will use /proc/pid/maps to 1638 * determine which areas of memory are actually mapped and llseek to 1639 * skip over unmapped regions. 1640 */ 1641 static ssize_t pagemap_read(struct file *file, char __user *buf, 1642 size_t count, loff_t *ppos) 1643 { 1644 struct mm_struct *mm = file->private_data; 1645 struct pagemapread pm; 1646 unsigned long src; 1647 unsigned long svpfn; 1648 unsigned long start_vaddr; 1649 unsigned long end_vaddr; 1650 int ret = 0, copied = 0; 1651 1652 if (!mm || !mmget_not_zero(mm)) 1653 goto out; 1654 1655 ret = -EINVAL; 1656 /* file position must be aligned */ 1657 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES)) 1658 goto out_mm; 1659 1660 ret = 0; 1661 if (!count) 1662 goto out_mm; 1663 1664 /* do not disclose physical addresses: attack vector */ 1665 pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN); 1666 1667 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT); 1668 pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL); 1669 ret = -ENOMEM; 1670 if (!pm.buffer) 1671 goto out_mm; 1672 1673 src = *ppos; 1674 svpfn = src / PM_ENTRY_BYTES; 1675 end_vaddr = mm->task_size; 1676 1677 /* watch out for wraparound */ 1678 start_vaddr = end_vaddr; 1679 if (svpfn <= (ULONG_MAX >> PAGE_SHIFT)) { 1680 unsigned long end; 1681 1682 ret = mmap_read_lock_killable(mm); 1683 if (ret) 1684 goto out_free; 1685 start_vaddr = untagged_addr_remote(mm, svpfn << PAGE_SHIFT); 1686 mmap_read_unlock(mm); 1687 1688 end = start_vaddr + ((count / PM_ENTRY_BYTES) << PAGE_SHIFT); 1689 if (end >= start_vaddr && end < mm->task_size) 1690 end_vaddr = end; 1691 } 1692 1693 /* Ensure the address is inside the task */ 1694 if (start_vaddr > mm->task_size) 1695 start_vaddr = end_vaddr; 1696 1697 ret = 0; 1698 while (count && (start_vaddr < end_vaddr)) { 1699 int len; 1700 unsigned long end; 1701 1702 pm.pos = 0; 1703 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK; 1704 /* overflow ? */ 1705 if (end < start_vaddr || end > end_vaddr) 1706 end = end_vaddr; 1707 ret = mmap_read_lock_killable(mm); 1708 if (ret) 1709 goto out_free; 1710 ret = walk_page_range(mm, start_vaddr, end, &pagemap_ops, &pm); 1711 mmap_read_unlock(mm); 1712 start_vaddr = end; 1713 1714 len = min(count, PM_ENTRY_BYTES * pm.pos); 1715 if (copy_to_user(buf, pm.buffer, len)) { 1716 ret = -EFAULT; 1717 goto out_free; 1718 } 1719 copied += len; 1720 buf += len; 1721 count -= len; 1722 } 1723 *ppos += copied; 1724 if (!ret || ret == PM_END_OF_BUFFER) 1725 ret = copied; 1726 1727 out_free: 1728 kfree(pm.buffer); 1729 out_mm: 1730 mmput(mm); 1731 out: 1732 return ret; 1733 } 1734 1735 static int pagemap_open(struct inode *inode, struct file *file) 1736 { 1737 struct mm_struct *mm; 1738 1739 mm = proc_mem_open(inode, PTRACE_MODE_READ); 1740 if (IS_ERR(mm)) 1741 return PTR_ERR(mm); 1742 file->private_data = mm; 1743 return 0; 1744 } 1745 1746 static int pagemap_release(struct inode *inode, struct file *file) 1747 { 1748 struct mm_struct *mm = file->private_data; 1749 1750 if (mm) 1751 mmdrop(mm); 1752 return 0; 1753 } 1754 1755 const struct file_operations proc_pagemap_operations = { 1756 .llseek = mem_lseek, /* borrow this */ 1757 .read = pagemap_read, 1758 .open = pagemap_open, 1759 .release = pagemap_release, 1760 }; 1761 #endif /* CONFIG_PROC_PAGE_MONITOR */ 1762 1763 #ifdef CONFIG_NUMA 1764 1765 struct numa_maps { 1766 unsigned long pages; 1767 unsigned long anon; 1768 unsigned long active; 1769 unsigned long writeback; 1770 unsigned long mapcount_max; 1771 unsigned long dirty; 1772 unsigned long swapcache; 1773 unsigned long node[MAX_NUMNODES]; 1774 }; 1775 1776 struct numa_maps_private { 1777 struct proc_maps_private proc_maps; 1778 struct numa_maps md; 1779 }; 1780 1781 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty, 1782 unsigned long nr_pages) 1783 { 1784 int count = page_mapcount(page); 1785 1786 md->pages += nr_pages; 1787 if (pte_dirty || PageDirty(page)) 1788 md->dirty += nr_pages; 1789 1790 if (PageSwapCache(page)) 1791 md->swapcache += nr_pages; 1792 1793 if (PageActive(page) || PageUnevictable(page)) 1794 md->active += nr_pages; 1795 1796 if (PageWriteback(page)) 1797 md->writeback += nr_pages; 1798 1799 if (PageAnon(page)) 1800 md->anon += nr_pages; 1801 1802 if (count > md->mapcount_max) 1803 md->mapcount_max = count; 1804 1805 md->node[page_to_nid(page)] += nr_pages; 1806 } 1807 1808 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma, 1809 unsigned long addr) 1810 { 1811 struct page *page; 1812 int nid; 1813 1814 if (!pte_present(pte)) 1815 return NULL; 1816 1817 page = vm_normal_page(vma, addr, pte); 1818 if (!page || is_zone_device_page(page)) 1819 return NULL; 1820 1821 if (PageReserved(page)) 1822 return NULL; 1823 1824 nid = page_to_nid(page); 1825 if (!node_isset(nid, node_states[N_MEMORY])) 1826 return NULL; 1827 1828 return page; 1829 } 1830 1831 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1832 static struct page *can_gather_numa_stats_pmd(pmd_t pmd, 1833 struct vm_area_struct *vma, 1834 unsigned long addr) 1835 { 1836 struct page *page; 1837 int nid; 1838 1839 if (!pmd_present(pmd)) 1840 return NULL; 1841 1842 page = vm_normal_page_pmd(vma, addr, pmd); 1843 if (!page) 1844 return NULL; 1845 1846 if (PageReserved(page)) 1847 return NULL; 1848 1849 nid = page_to_nid(page); 1850 if (!node_isset(nid, node_states[N_MEMORY])) 1851 return NULL; 1852 1853 return page; 1854 } 1855 #endif 1856 1857 static int gather_pte_stats(pmd_t *pmd, unsigned long addr, 1858 unsigned long end, struct mm_walk *walk) 1859 { 1860 struct numa_maps *md = walk->private; 1861 struct vm_area_struct *vma = walk->vma; 1862 spinlock_t *ptl; 1863 pte_t *orig_pte; 1864 pte_t *pte; 1865 1866 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1867 ptl = pmd_trans_huge_lock(pmd, vma); 1868 if (ptl) { 1869 struct page *page; 1870 1871 page = can_gather_numa_stats_pmd(*pmd, vma, addr); 1872 if (page) 1873 gather_stats(page, md, pmd_dirty(*pmd), 1874 HPAGE_PMD_SIZE/PAGE_SIZE); 1875 spin_unlock(ptl); 1876 return 0; 1877 } 1878 #endif 1879 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl); 1880 if (!pte) { 1881 walk->action = ACTION_AGAIN; 1882 return 0; 1883 } 1884 do { 1885 pte_t ptent = ptep_get(pte); 1886 struct page *page = can_gather_numa_stats(ptent, vma, addr); 1887 if (!page) 1888 continue; 1889 gather_stats(page, md, pte_dirty(ptent), 1); 1890 1891 } while (pte++, addr += PAGE_SIZE, addr != end); 1892 pte_unmap_unlock(orig_pte, ptl); 1893 cond_resched(); 1894 return 0; 1895 } 1896 #ifdef CONFIG_HUGETLB_PAGE 1897 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask, 1898 unsigned long addr, unsigned long end, struct mm_walk *walk) 1899 { 1900 pte_t huge_pte = huge_ptep_get(pte); 1901 struct numa_maps *md; 1902 struct page *page; 1903 1904 if (!pte_present(huge_pte)) 1905 return 0; 1906 1907 page = pte_page(huge_pte); 1908 1909 md = walk->private; 1910 gather_stats(page, md, pte_dirty(huge_pte), 1); 1911 return 0; 1912 } 1913 1914 #else 1915 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask, 1916 unsigned long addr, unsigned long end, struct mm_walk *walk) 1917 { 1918 return 0; 1919 } 1920 #endif 1921 1922 static const struct mm_walk_ops show_numa_ops = { 1923 .hugetlb_entry = gather_hugetlb_stats, 1924 .pmd_entry = gather_pte_stats, 1925 .walk_lock = PGWALK_RDLOCK, 1926 }; 1927 1928 /* 1929 * Display pages allocated per node and memory policy via /proc. 1930 */ 1931 static int show_numa_map(struct seq_file *m, void *v) 1932 { 1933 struct numa_maps_private *numa_priv = m->private; 1934 struct proc_maps_private *proc_priv = &numa_priv->proc_maps; 1935 struct vm_area_struct *vma = v; 1936 struct numa_maps *md = &numa_priv->md; 1937 struct file *file = vma->vm_file; 1938 struct mm_struct *mm = vma->vm_mm; 1939 struct mempolicy *pol; 1940 char buffer[64]; 1941 int nid; 1942 1943 if (!mm) 1944 return 0; 1945 1946 /* Ensure we start with an empty set of numa_maps statistics. */ 1947 memset(md, 0, sizeof(*md)); 1948 1949 pol = __get_vma_policy(vma, vma->vm_start); 1950 if (pol) { 1951 mpol_to_str(buffer, sizeof(buffer), pol); 1952 mpol_cond_put(pol); 1953 } else { 1954 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy); 1955 } 1956 1957 seq_printf(m, "%08lx %s", vma->vm_start, buffer); 1958 1959 if (file) { 1960 seq_puts(m, " file="); 1961 seq_file_path(m, file, "\n\t= "); 1962 } else if (vma_is_initial_heap(vma)) { 1963 seq_puts(m, " heap"); 1964 } else if (vma_is_initial_stack(vma)) { 1965 seq_puts(m, " stack"); 1966 } 1967 1968 if (is_vm_hugetlb_page(vma)) 1969 seq_puts(m, " huge"); 1970 1971 /* mmap_lock is held by m_start */ 1972 walk_page_vma(vma, &show_numa_ops, md); 1973 1974 if (!md->pages) 1975 goto out; 1976 1977 if (md->anon) 1978 seq_printf(m, " anon=%lu", md->anon); 1979 1980 if (md->dirty) 1981 seq_printf(m, " dirty=%lu", md->dirty); 1982 1983 if (md->pages != md->anon && md->pages != md->dirty) 1984 seq_printf(m, " mapped=%lu", md->pages); 1985 1986 if (md->mapcount_max > 1) 1987 seq_printf(m, " mapmax=%lu", md->mapcount_max); 1988 1989 if (md->swapcache) 1990 seq_printf(m, " swapcache=%lu", md->swapcache); 1991 1992 if (md->active < md->pages && !is_vm_hugetlb_page(vma)) 1993 seq_printf(m, " active=%lu", md->active); 1994 1995 if (md->writeback) 1996 seq_printf(m, " writeback=%lu", md->writeback); 1997 1998 for_each_node_state(nid, N_MEMORY) 1999 if (md->node[nid]) 2000 seq_printf(m, " N%d=%lu", nid, md->node[nid]); 2001 2002 seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10); 2003 out: 2004 seq_putc(m, '\n'); 2005 return 0; 2006 } 2007 2008 static const struct seq_operations proc_pid_numa_maps_op = { 2009 .start = m_start, 2010 .next = m_next, 2011 .stop = m_stop, 2012 .show = show_numa_map, 2013 }; 2014 2015 static int pid_numa_maps_open(struct inode *inode, struct file *file) 2016 { 2017 return proc_maps_open(inode, file, &proc_pid_numa_maps_op, 2018 sizeof(struct numa_maps_private)); 2019 } 2020 2021 const struct file_operations proc_pid_numa_maps_operations = { 2022 .open = pid_numa_maps_open, 2023 .read = seq_read, 2024 .llseek = seq_lseek, 2025 .release = proc_map_release, 2026 }; 2027 2028 #endif /* CONFIG_NUMA */ 2029