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 /* 240 * Indicate if the VMA is a stack for the given task; for 241 * /proc/PID/maps that is the stack of the main task. 242 */ 243 static int is_stack(struct vm_area_struct *vma) 244 { 245 /* 246 * We make no effort to guess what a given thread considers to be 247 * its "stack". It's not even well-defined for programs written 248 * languages like Go. 249 */ 250 return vma->vm_start <= vma->vm_mm->start_stack && 251 vma->vm_end >= vma->vm_mm->start_stack; 252 } 253 254 static void show_vma_header_prefix(struct seq_file *m, 255 unsigned long start, unsigned long end, 256 vm_flags_t flags, unsigned long long pgoff, 257 dev_t dev, unsigned long ino) 258 { 259 seq_setwidth(m, 25 + sizeof(void *) * 6 - 1); 260 seq_put_hex_ll(m, NULL, start, 8); 261 seq_put_hex_ll(m, "-", end, 8); 262 seq_putc(m, ' '); 263 seq_putc(m, flags & VM_READ ? 'r' : '-'); 264 seq_putc(m, flags & VM_WRITE ? 'w' : '-'); 265 seq_putc(m, flags & VM_EXEC ? 'x' : '-'); 266 seq_putc(m, flags & VM_MAYSHARE ? 's' : 'p'); 267 seq_put_hex_ll(m, " ", pgoff, 8); 268 seq_put_hex_ll(m, " ", MAJOR(dev), 2); 269 seq_put_hex_ll(m, ":", MINOR(dev), 2); 270 seq_put_decimal_ull(m, " ", ino); 271 seq_putc(m, ' '); 272 } 273 274 static void 275 show_map_vma(struct seq_file *m, struct vm_area_struct *vma) 276 { 277 struct anon_vma_name *anon_name = NULL; 278 struct mm_struct *mm = vma->vm_mm; 279 struct file *file = vma->vm_file; 280 vm_flags_t flags = vma->vm_flags; 281 unsigned long ino = 0; 282 unsigned long long pgoff = 0; 283 unsigned long start, end; 284 dev_t dev = 0; 285 const char *name = NULL; 286 287 if (file) { 288 struct inode *inode = file_inode(vma->vm_file); 289 dev = inode->i_sb->s_dev; 290 ino = inode->i_ino; 291 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT; 292 } 293 294 start = vma->vm_start; 295 end = vma->vm_end; 296 show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino); 297 if (mm) 298 anon_name = anon_vma_name(vma); 299 300 /* 301 * Print the dentry name for named mappings, and a 302 * special [heap] marker for the heap: 303 */ 304 if (file) { 305 seq_pad(m, ' '); 306 /* 307 * If user named this anon shared memory via 308 * prctl(PR_SET_VMA ..., use the provided name. 309 */ 310 if (anon_name) 311 seq_printf(m, "[anon_shmem:%s]", anon_name->name); 312 else 313 seq_file_path(m, file, "\n"); 314 goto done; 315 } 316 317 if (vma->vm_ops && vma->vm_ops->name) { 318 name = vma->vm_ops->name(vma); 319 if (name) 320 goto done; 321 } 322 323 name = arch_vma_name(vma); 324 if (!name) { 325 if (!mm) { 326 name = "[vdso]"; 327 goto done; 328 } 329 330 if (vma->vm_start <= mm->brk && 331 vma->vm_end >= mm->start_brk) { 332 name = "[heap]"; 333 goto done; 334 } 335 336 if (is_stack(vma)) { 337 name = "[stack]"; 338 goto done; 339 } 340 341 if (anon_name) { 342 seq_pad(m, ' '); 343 seq_printf(m, "[anon:%s]", anon_name->name); 344 } 345 } 346 347 done: 348 if (name) { 349 seq_pad(m, ' '); 350 seq_puts(m, name); 351 } 352 seq_putc(m, '\n'); 353 } 354 355 static int show_map(struct seq_file *m, void *v) 356 { 357 show_map_vma(m, v); 358 return 0; 359 } 360 361 static const struct seq_operations proc_pid_maps_op = { 362 .start = m_start, 363 .next = m_next, 364 .stop = m_stop, 365 .show = show_map 366 }; 367 368 static int pid_maps_open(struct inode *inode, struct file *file) 369 { 370 return do_maps_open(inode, file, &proc_pid_maps_op); 371 } 372 373 const struct file_operations proc_pid_maps_operations = { 374 .open = pid_maps_open, 375 .read = seq_read, 376 .llseek = seq_lseek, 377 .release = proc_map_release, 378 }; 379 380 /* 381 * Proportional Set Size(PSS): my share of RSS. 382 * 383 * PSS of a process is the count of pages it has in memory, where each 384 * page is divided by the number of processes sharing it. So if a 385 * process has 1000 pages all to itself, and 1000 shared with one other 386 * process, its PSS will be 1500. 387 * 388 * To keep (accumulated) division errors low, we adopt a 64bit 389 * fixed-point pss counter to minimize division errors. So (pss >> 390 * PSS_SHIFT) would be the real byte count. 391 * 392 * A shift of 12 before division means (assuming 4K page size): 393 * - 1M 3-user-pages add up to 8KB errors; 394 * - supports mapcount up to 2^24, or 16M; 395 * - supports PSS up to 2^52 bytes, or 4PB. 396 */ 397 #define PSS_SHIFT 12 398 399 #ifdef CONFIG_PROC_PAGE_MONITOR 400 struct mem_size_stats { 401 unsigned long resident; 402 unsigned long shared_clean; 403 unsigned long shared_dirty; 404 unsigned long private_clean; 405 unsigned long private_dirty; 406 unsigned long referenced; 407 unsigned long anonymous; 408 unsigned long lazyfree; 409 unsigned long anonymous_thp; 410 unsigned long shmem_thp; 411 unsigned long file_thp; 412 unsigned long swap; 413 unsigned long shared_hugetlb; 414 unsigned long private_hugetlb; 415 u64 pss; 416 u64 pss_anon; 417 u64 pss_file; 418 u64 pss_shmem; 419 u64 pss_dirty; 420 u64 pss_locked; 421 u64 swap_pss; 422 }; 423 424 static void smaps_page_accumulate(struct mem_size_stats *mss, 425 struct page *page, unsigned long size, unsigned long pss, 426 bool dirty, bool locked, bool private) 427 { 428 mss->pss += pss; 429 430 if (PageAnon(page)) 431 mss->pss_anon += pss; 432 else if (PageSwapBacked(page)) 433 mss->pss_shmem += pss; 434 else 435 mss->pss_file += pss; 436 437 if (locked) 438 mss->pss_locked += pss; 439 440 if (dirty || PageDirty(page)) { 441 mss->pss_dirty += pss; 442 if (private) 443 mss->private_dirty += size; 444 else 445 mss->shared_dirty += size; 446 } else { 447 if (private) 448 mss->private_clean += size; 449 else 450 mss->shared_clean += size; 451 } 452 } 453 454 static void smaps_account(struct mem_size_stats *mss, struct page *page, 455 bool compound, bool young, bool dirty, bool locked, 456 bool migration) 457 { 458 int i, nr = compound ? compound_nr(page) : 1; 459 unsigned long size = nr * PAGE_SIZE; 460 461 /* 462 * First accumulate quantities that depend only on |size| and the type 463 * of the compound page. 464 */ 465 if (PageAnon(page)) { 466 mss->anonymous += size; 467 if (!PageSwapBacked(page) && !dirty && !PageDirty(page)) 468 mss->lazyfree += size; 469 } 470 471 mss->resident += size; 472 /* Accumulate the size in pages that have been accessed. */ 473 if (young || page_is_young(page) || PageReferenced(page)) 474 mss->referenced += size; 475 476 /* 477 * Then accumulate quantities that may depend on sharing, or that may 478 * differ page-by-page. 479 * 480 * page_count(page) == 1 guarantees the page is mapped exactly once. 481 * If any subpage of the compound page mapped with PTE it would elevate 482 * page_count(). 483 * 484 * The page_mapcount() is called to get a snapshot of the mapcount. 485 * Without holding the page lock this snapshot can be slightly wrong as 486 * we cannot always read the mapcount atomically. It is not safe to 487 * call page_mapcount() even with PTL held if the page is not mapped, 488 * especially for migration entries. Treat regular migration entries 489 * as mapcount == 1. 490 */ 491 if ((page_count(page) == 1) || migration) { 492 smaps_page_accumulate(mss, page, size, size << PSS_SHIFT, dirty, 493 locked, true); 494 return; 495 } 496 for (i = 0; i < nr; i++, page++) { 497 int mapcount = page_mapcount(page); 498 unsigned long pss = PAGE_SIZE << PSS_SHIFT; 499 if (mapcount >= 2) 500 pss /= mapcount; 501 smaps_page_accumulate(mss, page, PAGE_SIZE, pss, dirty, locked, 502 mapcount < 2); 503 } 504 } 505 506 #ifdef CONFIG_SHMEM 507 static int smaps_pte_hole(unsigned long addr, unsigned long end, 508 __always_unused int depth, struct mm_walk *walk) 509 { 510 struct mem_size_stats *mss = walk->private; 511 struct vm_area_struct *vma = walk->vma; 512 513 mss->swap += shmem_partial_swap_usage(walk->vma->vm_file->f_mapping, 514 linear_page_index(vma, addr), 515 linear_page_index(vma, end)); 516 517 return 0; 518 } 519 #else 520 #define smaps_pte_hole NULL 521 #endif /* CONFIG_SHMEM */ 522 523 static void smaps_pte_hole_lookup(unsigned long addr, struct mm_walk *walk) 524 { 525 #ifdef CONFIG_SHMEM 526 if (walk->ops->pte_hole) { 527 /* depth is not used */ 528 smaps_pte_hole(addr, addr + PAGE_SIZE, 0, walk); 529 } 530 #endif 531 } 532 533 static void smaps_pte_entry(pte_t *pte, unsigned long addr, 534 struct mm_walk *walk) 535 { 536 struct mem_size_stats *mss = walk->private; 537 struct vm_area_struct *vma = walk->vma; 538 bool locked = !!(vma->vm_flags & VM_LOCKED); 539 struct page *page = NULL; 540 bool migration = false, young = false, dirty = false; 541 542 if (pte_present(*pte)) { 543 page = vm_normal_page(vma, addr, *pte); 544 young = pte_young(*pte); 545 dirty = pte_dirty(*pte); 546 } else if (is_swap_pte(*pte)) { 547 swp_entry_t swpent = pte_to_swp_entry(*pte); 548 549 if (!non_swap_entry(swpent)) { 550 int mapcount; 551 552 mss->swap += PAGE_SIZE; 553 mapcount = swp_swapcount(swpent); 554 if (mapcount >= 2) { 555 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT; 556 557 do_div(pss_delta, mapcount); 558 mss->swap_pss += pss_delta; 559 } else { 560 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT; 561 } 562 } else if (is_pfn_swap_entry(swpent)) { 563 if (is_migration_entry(swpent)) 564 migration = true; 565 page = pfn_swap_entry_to_page(swpent); 566 } 567 } else { 568 smaps_pte_hole_lookup(addr, walk); 569 return; 570 } 571 572 if (!page) 573 return; 574 575 smaps_account(mss, page, false, young, dirty, locked, migration); 576 } 577 578 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 579 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr, 580 struct mm_walk *walk) 581 { 582 struct mem_size_stats *mss = walk->private; 583 struct vm_area_struct *vma = walk->vma; 584 bool locked = !!(vma->vm_flags & VM_LOCKED); 585 struct page *page = NULL; 586 bool migration = false; 587 588 if (pmd_present(*pmd)) { 589 /* FOLL_DUMP will return -EFAULT on huge zero page */ 590 page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP); 591 } else if (unlikely(thp_migration_supported() && is_swap_pmd(*pmd))) { 592 swp_entry_t entry = pmd_to_swp_entry(*pmd); 593 594 if (is_migration_entry(entry)) { 595 migration = true; 596 page = pfn_swap_entry_to_page(entry); 597 } 598 } 599 if (IS_ERR_OR_NULL(page)) 600 return; 601 if (PageAnon(page)) 602 mss->anonymous_thp += HPAGE_PMD_SIZE; 603 else if (PageSwapBacked(page)) 604 mss->shmem_thp += HPAGE_PMD_SIZE; 605 else if (is_zone_device_page(page)) 606 /* pass */; 607 else 608 mss->file_thp += HPAGE_PMD_SIZE; 609 610 smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd), 611 locked, migration); 612 } 613 #else 614 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr, 615 struct mm_walk *walk) 616 { 617 } 618 #endif 619 620 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, 621 struct mm_walk *walk) 622 { 623 struct vm_area_struct *vma = walk->vma; 624 pte_t *pte; 625 spinlock_t *ptl; 626 627 ptl = pmd_trans_huge_lock(pmd, vma); 628 if (ptl) { 629 smaps_pmd_entry(pmd, addr, walk); 630 spin_unlock(ptl); 631 goto out; 632 } 633 634 if (pmd_trans_unstable(pmd)) 635 goto out; 636 /* 637 * The mmap_lock held all the way back in m_start() is what 638 * keeps khugepaged out of here and from collapsing things 639 * in here. 640 */ 641 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); 642 for (; addr != end; pte++, addr += PAGE_SIZE) 643 smaps_pte_entry(pte, addr, walk); 644 pte_unmap_unlock(pte - 1, ptl); 645 out: 646 cond_resched(); 647 return 0; 648 } 649 650 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma) 651 { 652 /* 653 * Don't forget to update Documentation/ on changes. 654 */ 655 static const char mnemonics[BITS_PER_LONG][2] = { 656 /* 657 * In case if we meet a flag we don't know about. 658 */ 659 [0 ... (BITS_PER_LONG-1)] = "??", 660 661 [ilog2(VM_READ)] = "rd", 662 [ilog2(VM_WRITE)] = "wr", 663 [ilog2(VM_EXEC)] = "ex", 664 [ilog2(VM_SHARED)] = "sh", 665 [ilog2(VM_MAYREAD)] = "mr", 666 [ilog2(VM_MAYWRITE)] = "mw", 667 [ilog2(VM_MAYEXEC)] = "me", 668 [ilog2(VM_MAYSHARE)] = "ms", 669 [ilog2(VM_GROWSDOWN)] = "gd", 670 [ilog2(VM_PFNMAP)] = "pf", 671 [ilog2(VM_LOCKED)] = "lo", 672 [ilog2(VM_IO)] = "io", 673 [ilog2(VM_SEQ_READ)] = "sr", 674 [ilog2(VM_RAND_READ)] = "rr", 675 [ilog2(VM_DONTCOPY)] = "dc", 676 [ilog2(VM_DONTEXPAND)] = "de", 677 [ilog2(VM_LOCKONFAULT)] = "lf", 678 [ilog2(VM_ACCOUNT)] = "ac", 679 [ilog2(VM_NORESERVE)] = "nr", 680 [ilog2(VM_HUGETLB)] = "ht", 681 [ilog2(VM_SYNC)] = "sf", 682 [ilog2(VM_ARCH_1)] = "ar", 683 [ilog2(VM_WIPEONFORK)] = "wf", 684 [ilog2(VM_DONTDUMP)] = "dd", 685 #ifdef CONFIG_ARM64_BTI 686 [ilog2(VM_ARM64_BTI)] = "bt", 687 #endif 688 #ifdef CONFIG_MEM_SOFT_DIRTY 689 [ilog2(VM_SOFTDIRTY)] = "sd", 690 #endif 691 [ilog2(VM_MIXEDMAP)] = "mm", 692 [ilog2(VM_HUGEPAGE)] = "hg", 693 [ilog2(VM_NOHUGEPAGE)] = "nh", 694 [ilog2(VM_MERGEABLE)] = "mg", 695 [ilog2(VM_UFFD_MISSING)]= "um", 696 [ilog2(VM_UFFD_WP)] = "uw", 697 #ifdef CONFIG_ARM64_MTE 698 [ilog2(VM_MTE)] = "mt", 699 [ilog2(VM_MTE_ALLOWED)] = "", 700 #endif 701 #ifdef CONFIG_ARCH_HAS_PKEYS 702 /* These come out via ProtectionKey: */ 703 [ilog2(VM_PKEY_BIT0)] = "", 704 [ilog2(VM_PKEY_BIT1)] = "", 705 [ilog2(VM_PKEY_BIT2)] = "", 706 [ilog2(VM_PKEY_BIT3)] = "", 707 #if VM_PKEY_BIT4 708 [ilog2(VM_PKEY_BIT4)] = "", 709 #endif 710 #endif /* CONFIG_ARCH_HAS_PKEYS */ 711 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_MINOR 712 [ilog2(VM_UFFD_MINOR)] = "ui", 713 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_MINOR */ 714 }; 715 size_t i; 716 717 seq_puts(m, "VmFlags: "); 718 for (i = 0; i < BITS_PER_LONG; i++) { 719 if (!mnemonics[i][0]) 720 continue; 721 if (vma->vm_flags & (1UL << i)) { 722 seq_putc(m, mnemonics[i][0]); 723 seq_putc(m, mnemonics[i][1]); 724 seq_putc(m, ' '); 725 } 726 } 727 seq_putc(m, '\n'); 728 } 729 730 #ifdef CONFIG_HUGETLB_PAGE 731 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask, 732 unsigned long addr, unsigned long end, 733 struct mm_walk *walk) 734 { 735 struct mem_size_stats *mss = walk->private; 736 struct vm_area_struct *vma = walk->vma; 737 struct page *page = NULL; 738 739 if (pte_present(*pte)) { 740 page = vm_normal_page(vma, addr, *pte); 741 } else if (is_swap_pte(*pte)) { 742 swp_entry_t swpent = pte_to_swp_entry(*pte); 743 744 if (is_pfn_swap_entry(swpent)) 745 page = pfn_swap_entry_to_page(swpent); 746 } 747 if (page) { 748 int mapcount = page_mapcount(page); 749 750 if (mapcount >= 2) 751 mss->shared_hugetlb += huge_page_size(hstate_vma(vma)); 752 else 753 mss->private_hugetlb += huge_page_size(hstate_vma(vma)); 754 } 755 return 0; 756 } 757 #else 758 #define smaps_hugetlb_range NULL 759 #endif /* HUGETLB_PAGE */ 760 761 static const struct mm_walk_ops smaps_walk_ops = { 762 .pmd_entry = smaps_pte_range, 763 .hugetlb_entry = smaps_hugetlb_range, 764 }; 765 766 static const struct mm_walk_ops smaps_shmem_walk_ops = { 767 .pmd_entry = smaps_pte_range, 768 .hugetlb_entry = smaps_hugetlb_range, 769 .pte_hole = smaps_pte_hole, 770 }; 771 772 /* 773 * Gather mem stats from @vma with the indicated beginning 774 * address @start, and keep them in @mss. 775 * 776 * Use vm_start of @vma as the beginning address if @start is 0. 777 */ 778 static void smap_gather_stats(struct vm_area_struct *vma, 779 struct mem_size_stats *mss, unsigned long start) 780 { 781 const struct mm_walk_ops *ops = &smaps_walk_ops; 782 783 /* Invalid start */ 784 if (start >= vma->vm_end) 785 return; 786 787 #ifdef CONFIG_SHMEM 788 if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) { 789 /* 790 * For shared or readonly shmem mappings we know that all 791 * swapped out pages belong to the shmem object, and we can 792 * obtain the swap value much more efficiently. For private 793 * writable mappings, we might have COW pages that are 794 * not affected by the parent swapped out pages of the shmem 795 * object, so we have to distinguish them during the page walk. 796 * Unless we know that the shmem object (or the part mapped by 797 * our VMA) has no swapped out pages at all. 798 */ 799 unsigned long shmem_swapped = shmem_swap_usage(vma); 800 801 if (!start && (!shmem_swapped || (vma->vm_flags & VM_SHARED) || 802 !(vma->vm_flags & VM_WRITE))) { 803 mss->swap += shmem_swapped; 804 } else { 805 ops = &smaps_shmem_walk_ops; 806 } 807 } 808 #endif 809 /* mmap_lock is held in m_start */ 810 if (!start) 811 walk_page_vma(vma, ops, mss); 812 else 813 walk_page_range(vma->vm_mm, start, vma->vm_end, ops, mss); 814 } 815 816 #define SEQ_PUT_DEC(str, val) \ 817 seq_put_decimal_ull_width(m, str, (val) >> 10, 8) 818 819 /* Show the contents common for smaps and smaps_rollup */ 820 static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss, 821 bool rollup_mode) 822 { 823 SEQ_PUT_DEC("Rss: ", mss->resident); 824 SEQ_PUT_DEC(" kB\nPss: ", mss->pss >> PSS_SHIFT); 825 SEQ_PUT_DEC(" kB\nPss_Dirty: ", mss->pss_dirty >> PSS_SHIFT); 826 if (rollup_mode) { 827 /* 828 * These are meaningful only for smaps_rollup, otherwise two of 829 * them are zero, and the other one is the same as Pss. 830 */ 831 SEQ_PUT_DEC(" kB\nPss_Anon: ", 832 mss->pss_anon >> PSS_SHIFT); 833 SEQ_PUT_DEC(" kB\nPss_File: ", 834 mss->pss_file >> PSS_SHIFT); 835 SEQ_PUT_DEC(" kB\nPss_Shmem: ", 836 mss->pss_shmem >> PSS_SHIFT); 837 } 838 SEQ_PUT_DEC(" kB\nShared_Clean: ", mss->shared_clean); 839 SEQ_PUT_DEC(" kB\nShared_Dirty: ", mss->shared_dirty); 840 SEQ_PUT_DEC(" kB\nPrivate_Clean: ", mss->private_clean); 841 SEQ_PUT_DEC(" kB\nPrivate_Dirty: ", mss->private_dirty); 842 SEQ_PUT_DEC(" kB\nReferenced: ", mss->referenced); 843 SEQ_PUT_DEC(" kB\nAnonymous: ", mss->anonymous); 844 SEQ_PUT_DEC(" kB\nLazyFree: ", mss->lazyfree); 845 SEQ_PUT_DEC(" kB\nAnonHugePages: ", mss->anonymous_thp); 846 SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp); 847 SEQ_PUT_DEC(" kB\nFilePmdMapped: ", mss->file_thp); 848 SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb); 849 seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ", 850 mss->private_hugetlb >> 10, 7); 851 SEQ_PUT_DEC(" kB\nSwap: ", mss->swap); 852 SEQ_PUT_DEC(" kB\nSwapPss: ", 853 mss->swap_pss >> PSS_SHIFT); 854 SEQ_PUT_DEC(" kB\nLocked: ", 855 mss->pss_locked >> PSS_SHIFT); 856 seq_puts(m, " kB\n"); 857 } 858 859 static int show_smap(struct seq_file *m, void *v) 860 { 861 struct vm_area_struct *vma = v; 862 struct mem_size_stats mss; 863 864 memset(&mss, 0, sizeof(mss)); 865 866 smap_gather_stats(vma, &mss, 0); 867 868 show_map_vma(m, vma); 869 870 SEQ_PUT_DEC("Size: ", vma->vm_end - vma->vm_start); 871 SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma)); 872 SEQ_PUT_DEC(" kB\nMMUPageSize: ", vma_mmu_pagesize(vma)); 873 seq_puts(m, " kB\n"); 874 875 __show_smap(m, &mss, false); 876 877 seq_printf(m, "THPeligible: %d\n", 878 hugepage_vma_check(vma, vma->vm_flags, true, false, true)); 879 880 if (arch_pkeys_enabled()) 881 seq_printf(m, "ProtectionKey: %8u\n", vma_pkey(vma)); 882 show_smap_vma_flags(m, vma); 883 884 return 0; 885 } 886 887 static int show_smaps_rollup(struct seq_file *m, void *v) 888 { 889 struct proc_maps_private *priv = m->private; 890 struct mem_size_stats mss; 891 struct mm_struct *mm = priv->mm; 892 struct vm_area_struct *vma; 893 unsigned long vma_start = 0, last_vma_end = 0; 894 int ret = 0; 895 MA_STATE(mas, &mm->mm_mt, 0, 0); 896 897 priv->task = get_proc_task(priv->inode); 898 if (!priv->task) 899 return -ESRCH; 900 901 if (!mm || !mmget_not_zero(mm)) { 902 ret = -ESRCH; 903 goto out_put_task; 904 } 905 906 memset(&mss, 0, sizeof(mss)); 907 908 ret = mmap_read_lock_killable(mm); 909 if (ret) 910 goto out_put_mm; 911 912 hold_task_mempolicy(priv); 913 vma = mas_find(&mas, ULONG_MAX); 914 915 if (unlikely(!vma)) 916 goto empty_set; 917 918 vma_start = vma->vm_start; 919 do { 920 smap_gather_stats(vma, &mss, 0); 921 last_vma_end = vma->vm_end; 922 923 /* 924 * Release mmap_lock temporarily if someone wants to 925 * access it for write request. 926 */ 927 if (mmap_lock_is_contended(mm)) { 928 mas_pause(&mas); 929 mmap_read_unlock(mm); 930 ret = mmap_read_lock_killable(mm); 931 if (ret) { 932 release_task_mempolicy(priv); 933 goto out_put_mm; 934 } 935 936 /* 937 * After dropping the lock, there are four cases to 938 * consider. See the following example for explanation. 939 * 940 * +------+------+-----------+ 941 * | VMA1 | VMA2 | VMA3 | 942 * +------+------+-----------+ 943 * | | | | 944 * 4k 8k 16k 400k 945 * 946 * Suppose we drop the lock after reading VMA2 due to 947 * contention, then we get: 948 * 949 * last_vma_end = 16k 950 * 951 * 1) VMA2 is freed, but VMA3 exists: 952 * 953 * find_vma(mm, 16k - 1) will return VMA3. 954 * In this case, just continue from VMA3. 955 * 956 * 2) VMA2 still exists: 957 * 958 * find_vma(mm, 16k - 1) will return VMA2. 959 * Iterate the loop like the original one. 960 * 961 * 3) No more VMAs can be found: 962 * 963 * find_vma(mm, 16k - 1) will return NULL. 964 * No more things to do, just break. 965 * 966 * 4) (last_vma_end - 1) is the middle of a vma (VMA'): 967 * 968 * find_vma(mm, 16k - 1) will return VMA' whose range 969 * contains last_vma_end. 970 * Iterate VMA' from last_vma_end. 971 */ 972 vma = mas_find(&mas, ULONG_MAX); 973 /* Case 3 above */ 974 if (!vma) 975 break; 976 977 /* Case 1 above */ 978 if (vma->vm_start >= last_vma_end) 979 continue; 980 981 /* Case 4 above */ 982 if (vma->vm_end > last_vma_end) 983 smap_gather_stats(vma, &mss, last_vma_end); 984 } 985 /* Case 2 above */ 986 } while ((vma = mas_find(&mas, ULONG_MAX)) != NULL); 987 988 empty_set: 989 show_vma_header_prefix(m, vma_start, last_vma_end, 0, 0, 0, 0); 990 seq_pad(m, ' '); 991 seq_puts(m, "[rollup]\n"); 992 993 __show_smap(m, &mss, true); 994 995 release_task_mempolicy(priv); 996 mmap_read_unlock(mm); 997 998 out_put_mm: 999 mmput(mm); 1000 out_put_task: 1001 put_task_struct(priv->task); 1002 priv->task = NULL; 1003 1004 return ret; 1005 } 1006 #undef SEQ_PUT_DEC 1007 1008 static const struct seq_operations proc_pid_smaps_op = { 1009 .start = m_start, 1010 .next = m_next, 1011 .stop = m_stop, 1012 .show = show_smap 1013 }; 1014 1015 static int pid_smaps_open(struct inode *inode, struct file *file) 1016 { 1017 return do_maps_open(inode, file, &proc_pid_smaps_op); 1018 } 1019 1020 static int smaps_rollup_open(struct inode *inode, struct file *file) 1021 { 1022 int ret; 1023 struct proc_maps_private *priv; 1024 1025 priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT); 1026 if (!priv) 1027 return -ENOMEM; 1028 1029 ret = single_open(file, show_smaps_rollup, priv); 1030 if (ret) 1031 goto out_free; 1032 1033 priv->inode = inode; 1034 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ); 1035 if (IS_ERR(priv->mm)) { 1036 ret = PTR_ERR(priv->mm); 1037 1038 single_release(inode, file); 1039 goto out_free; 1040 } 1041 1042 return 0; 1043 1044 out_free: 1045 kfree(priv); 1046 return ret; 1047 } 1048 1049 static int smaps_rollup_release(struct inode *inode, struct file *file) 1050 { 1051 struct seq_file *seq = file->private_data; 1052 struct proc_maps_private *priv = seq->private; 1053 1054 if (priv->mm) 1055 mmdrop(priv->mm); 1056 1057 kfree(priv); 1058 return single_release(inode, file); 1059 } 1060 1061 const struct file_operations proc_pid_smaps_operations = { 1062 .open = pid_smaps_open, 1063 .read = seq_read, 1064 .llseek = seq_lseek, 1065 .release = proc_map_release, 1066 }; 1067 1068 const struct file_operations proc_pid_smaps_rollup_operations = { 1069 .open = smaps_rollup_open, 1070 .read = seq_read, 1071 .llseek = seq_lseek, 1072 .release = smaps_rollup_release, 1073 }; 1074 1075 enum clear_refs_types { 1076 CLEAR_REFS_ALL = 1, 1077 CLEAR_REFS_ANON, 1078 CLEAR_REFS_MAPPED, 1079 CLEAR_REFS_SOFT_DIRTY, 1080 CLEAR_REFS_MM_HIWATER_RSS, 1081 CLEAR_REFS_LAST, 1082 }; 1083 1084 struct clear_refs_private { 1085 enum clear_refs_types type; 1086 }; 1087 1088 #ifdef CONFIG_MEM_SOFT_DIRTY 1089 1090 static inline bool pte_is_pinned(struct vm_area_struct *vma, unsigned long addr, pte_t pte) 1091 { 1092 struct page *page; 1093 1094 if (!pte_write(pte)) 1095 return false; 1096 if (!is_cow_mapping(vma->vm_flags)) 1097 return false; 1098 if (likely(!test_bit(MMF_HAS_PINNED, &vma->vm_mm->flags))) 1099 return false; 1100 page = vm_normal_page(vma, addr, pte); 1101 if (!page) 1102 return false; 1103 return page_maybe_dma_pinned(page); 1104 } 1105 1106 static inline void clear_soft_dirty(struct vm_area_struct *vma, 1107 unsigned long addr, pte_t *pte) 1108 { 1109 /* 1110 * The soft-dirty tracker uses #PF-s to catch writes 1111 * to pages, so write-protect the pte as well. See the 1112 * Documentation/admin-guide/mm/soft-dirty.rst for full description 1113 * of how soft-dirty works. 1114 */ 1115 pte_t ptent = *pte; 1116 1117 if (pte_present(ptent)) { 1118 pte_t old_pte; 1119 1120 if (pte_is_pinned(vma, addr, ptent)) 1121 return; 1122 old_pte = ptep_modify_prot_start(vma, addr, pte); 1123 ptent = pte_wrprotect(old_pte); 1124 ptent = pte_clear_soft_dirty(ptent); 1125 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent); 1126 } else if (is_swap_pte(ptent)) { 1127 ptent = pte_swp_clear_soft_dirty(ptent); 1128 set_pte_at(vma->vm_mm, addr, pte, ptent); 1129 } 1130 } 1131 #else 1132 static inline void clear_soft_dirty(struct vm_area_struct *vma, 1133 unsigned long addr, pte_t *pte) 1134 { 1135 } 1136 #endif 1137 1138 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE) 1139 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma, 1140 unsigned long addr, pmd_t *pmdp) 1141 { 1142 pmd_t old, pmd = *pmdp; 1143 1144 if (pmd_present(pmd)) { 1145 /* See comment in change_huge_pmd() */ 1146 old = pmdp_invalidate(vma, addr, pmdp); 1147 if (pmd_dirty(old)) 1148 pmd = pmd_mkdirty(pmd); 1149 if (pmd_young(old)) 1150 pmd = pmd_mkyoung(pmd); 1151 1152 pmd = pmd_wrprotect(pmd); 1153 pmd = pmd_clear_soft_dirty(pmd); 1154 1155 set_pmd_at(vma->vm_mm, addr, pmdp, pmd); 1156 } else if (is_migration_entry(pmd_to_swp_entry(pmd))) { 1157 pmd = pmd_swp_clear_soft_dirty(pmd); 1158 set_pmd_at(vma->vm_mm, addr, pmdp, pmd); 1159 } 1160 } 1161 #else 1162 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma, 1163 unsigned long addr, pmd_t *pmdp) 1164 { 1165 } 1166 #endif 1167 1168 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr, 1169 unsigned long end, struct mm_walk *walk) 1170 { 1171 struct clear_refs_private *cp = walk->private; 1172 struct vm_area_struct *vma = walk->vma; 1173 pte_t *pte, ptent; 1174 spinlock_t *ptl; 1175 struct page *page; 1176 1177 ptl = pmd_trans_huge_lock(pmd, vma); 1178 if (ptl) { 1179 if (cp->type == CLEAR_REFS_SOFT_DIRTY) { 1180 clear_soft_dirty_pmd(vma, addr, pmd); 1181 goto out; 1182 } 1183 1184 if (!pmd_present(*pmd)) 1185 goto out; 1186 1187 page = pmd_page(*pmd); 1188 1189 /* Clear accessed and referenced bits. */ 1190 pmdp_test_and_clear_young(vma, addr, pmd); 1191 test_and_clear_page_young(page); 1192 ClearPageReferenced(page); 1193 out: 1194 spin_unlock(ptl); 1195 return 0; 1196 } 1197 1198 if (pmd_trans_unstable(pmd)) 1199 return 0; 1200 1201 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); 1202 for (; addr != end; pte++, addr += PAGE_SIZE) { 1203 ptent = *pte; 1204 1205 if (cp->type == CLEAR_REFS_SOFT_DIRTY) { 1206 clear_soft_dirty(vma, addr, pte); 1207 continue; 1208 } 1209 1210 if (!pte_present(ptent)) 1211 continue; 1212 1213 page = vm_normal_page(vma, addr, ptent); 1214 if (!page) 1215 continue; 1216 1217 /* Clear accessed and referenced bits. */ 1218 ptep_test_and_clear_young(vma, addr, pte); 1219 test_and_clear_page_young(page); 1220 ClearPageReferenced(page); 1221 } 1222 pte_unmap_unlock(pte - 1, ptl); 1223 cond_resched(); 1224 return 0; 1225 } 1226 1227 static int clear_refs_test_walk(unsigned long start, unsigned long end, 1228 struct mm_walk *walk) 1229 { 1230 struct clear_refs_private *cp = walk->private; 1231 struct vm_area_struct *vma = walk->vma; 1232 1233 if (vma->vm_flags & VM_PFNMAP) 1234 return 1; 1235 1236 /* 1237 * Writing 1 to /proc/pid/clear_refs affects all pages. 1238 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages. 1239 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages. 1240 * Writing 4 to /proc/pid/clear_refs affects all pages. 1241 */ 1242 if (cp->type == CLEAR_REFS_ANON && vma->vm_file) 1243 return 1; 1244 if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file) 1245 return 1; 1246 return 0; 1247 } 1248 1249 static const struct mm_walk_ops clear_refs_walk_ops = { 1250 .pmd_entry = clear_refs_pte_range, 1251 .test_walk = clear_refs_test_walk, 1252 }; 1253 1254 static ssize_t clear_refs_write(struct file *file, const char __user *buf, 1255 size_t count, loff_t *ppos) 1256 { 1257 struct task_struct *task; 1258 char buffer[PROC_NUMBUF]; 1259 struct mm_struct *mm; 1260 struct vm_area_struct *vma; 1261 enum clear_refs_types type; 1262 int itype; 1263 int rv; 1264 1265 memset(buffer, 0, sizeof(buffer)); 1266 if (count > sizeof(buffer) - 1) 1267 count = sizeof(buffer) - 1; 1268 if (copy_from_user(buffer, buf, count)) 1269 return -EFAULT; 1270 rv = kstrtoint(strstrip(buffer), 10, &itype); 1271 if (rv < 0) 1272 return rv; 1273 type = (enum clear_refs_types)itype; 1274 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST) 1275 return -EINVAL; 1276 1277 task = get_proc_task(file_inode(file)); 1278 if (!task) 1279 return -ESRCH; 1280 mm = get_task_mm(task); 1281 if (mm) { 1282 MA_STATE(mas, &mm->mm_mt, 0, 0); 1283 struct mmu_notifier_range range; 1284 struct clear_refs_private cp = { 1285 .type = type, 1286 }; 1287 1288 if (mmap_write_lock_killable(mm)) { 1289 count = -EINTR; 1290 goto out_mm; 1291 } 1292 if (type == CLEAR_REFS_MM_HIWATER_RSS) { 1293 /* 1294 * Writing 5 to /proc/pid/clear_refs resets the peak 1295 * resident set size to this mm's current rss value. 1296 */ 1297 reset_mm_hiwater_rss(mm); 1298 goto out_unlock; 1299 } 1300 1301 if (type == CLEAR_REFS_SOFT_DIRTY) { 1302 mas_for_each(&mas, vma, ULONG_MAX) { 1303 if (!(vma->vm_flags & VM_SOFTDIRTY)) 1304 continue; 1305 vma->vm_flags &= ~VM_SOFTDIRTY; 1306 vma_set_page_prot(vma); 1307 } 1308 1309 inc_tlb_flush_pending(mm); 1310 mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY, 1311 0, NULL, mm, 0, -1UL); 1312 mmu_notifier_invalidate_range_start(&range); 1313 } 1314 walk_page_range(mm, 0, -1, &clear_refs_walk_ops, &cp); 1315 if (type == CLEAR_REFS_SOFT_DIRTY) { 1316 mmu_notifier_invalidate_range_end(&range); 1317 flush_tlb_mm(mm); 1318 dec_tlb_flush_pending(mm); 1319 } 1320 out_unlock: 1321 mmap_write_unlock(mm); 1322 out_mm: 1323 mmput(mm); 1324 } 1325 put_task_struct(task); 1326 1327 return count; 1328 } 1329 1330 const struct file_operations proc_clear_refs_operations = { 1331 .write = clear_refs_write, 1332 .llseek = noop_llseek, 1333 }; 1334 1335 typedef struct { 1336 u64 pme; 1337 } pagemap_entry_t; 1338 1339 struct pagemapread { 1340 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */ 1341 pagemap_entry_t *buffer; 1342 bool show_pfn; 1343 }; 1344 1345 #define PAGEMAP_WALK_SIZE (PMD_SIZE) 1346 #define PAGEMAP_WALK_MASK (PMD_MASK) 1347 1348 #define PM_ENTRY_BYTES sizeof(pagemap_entry_t) 1349 #define PM_PFRAME_BITS 55 1350 #define PM_PFRAME_MASK GENMASK_ULL(PM_PFRAME_BITS - 1, 0) 1351 #define PM_SOFT_DIRTY BIT_ULL(55) 1352 #define PM_MMAP_EXCLUSIVE BIT_ULL(56) 1353 #define PM_UFFD_WP BIT_ULL(57) 1354 #define PM_FILE BIT_ULL(61) 1355 #define PM_SWAP BIT_ULL(62) 1356 #define PM_PRESENT BIT_ULL(63) 1357 1358 #define PM_END_OF_BUFFER 1 1359 1360 static inline pagemap_entry_t make_pme(u64 frame, u64 flags) 1361 { 1362 return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags }; 1363 } 1364 1365 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme, 1366 struct pagemapread *pm) 1367 { 1368 pm->buffer[pm->pos++] = *pme; 1369 if (pm->pos >= pm->len) 1370 return PM_END_OF_BUFFER; 1371 return 0; 1372 } 1373 1374 static int pagemap_pte_hole(unsigned long start, unsigned long end, 1375 __always_unused int depth, struct mm_walk *walk) 1376 { 1377 struct pagemapread *pm = walk->private; 1378 unsigned long addr = start; 1379 int err = 0; 1380 1381 while (addr < end) { 1382 struct vm_area_struct *vma = find_vma(walk->mm, addr); 1383 pagemap_entry_t pme = make_pme(0, 0); 1384 /* End of address space hole, which we mark as non-present. */ 1385 unsigned long hole_end; 1386 1387 if (vma) 1388 hole_end = min(end, vma->vm_start); 1389 else 1390 hole_end = end; 1391 1392 for (; addr < hole_end; addr += PAGE_SIZE) { 1393 err = add_to_pagemap(addr, &pme, pm); 1394 if (err) 1395 goto out; 1396 } 1397 1398 if (!vma) 1399 break; 1400 1401 /* Addresses in the VMA. */ 1402 if (vma->vm_flags & VM_SOFTDIRTY) 1403 pme = make_pme(0, PM_SOFT_DIRTY); 1404 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) { 1405 err = add_to_pagemap(addr, &pme, pm); 1406 if (err) 1407 goto out; 1408 } 1409 } 1410 out: 1411 return err; 1412 } 1413 1414 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm, 1415 struct vm_area_struct *vma, unsigned long addr, pte_t pte) 1416 { 1417 u64 frame = 0, flags = 0; 1418 struct page *page = NULL; 1419 bool migration = false; 1420 1421 if (pte_present(pte)) { 1422 if (pm->show_pfn) 1423 frame = pte_pfn(pte); 1424 flags |= PM_PRESENT; 1425 page = vm_normal_page(vma, addr, pte); 1426 if (pte_soft_dirty(pte)) 1427 flags |= PM_SOFT_DIRTY; 1428 if (pte_uffd_wp(pte)) 1429 flags |= PM_UFFD_WP; 1430 } else if (is_swap_pte(pte)) { 1431 swp_entry_t entry; 1432 if (pte_swp_soft_dirty(pte)) 1433 flags |= PM_SOFT_DIRTY; 1434 if (pte_swp_uffd_wp(pte)) 1435 flags |= PM_UFFD_WP; 1436 entry = pte_to_swp_entry(pte); 1437 if (pm->show_pfn) { 1438 pgoff_t offset; 1439 /* 1440 * For PFN swap offsets, keeping the offset field 1441 * to be PFN only to be compatible with old smaps. 1442 */ 1443 if (is_pfn_swap_entry(entry)) 1444 offset = swp_offset_pfn(entry); 1445 else 1446 offset = swp_offset(entry); 1447 frame = swp_type(entry) | 1448 (offset << MAX_SWAPFILES_SHIFT); 1449 } 1450 flags |= PM_SWAP; 1451 migration = is_migration_entry(entry); 1452 if (is_pfn_swap_entry(entry)) 1453 page = pfn_swap_entry_to_page(entry); 1454 if (pte_marker_entry_uffd_wp(entry)) 1455 flags |= PM_UFFD_WP; 1456 } 1457 1458 if (page && !PageAnon(page)) 1459 flags |= PM_FILE; 1460 if (page && !migration && page_mapcount(page) == 1) 1461 flags |= PM_MMAP_EXCLUSIVE; 1462 if (vma->vm_flags & VM_SOFTDIRTY) 1463 flags |= PM_SOFT_DIRTY; 1464 1465 return make_pme(frame, flags); 1466 } 1467 1468 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end, 1469 struct mm_walk *walk) 1470 { 1471 struct vm_area_struct *vma = walk->vma; 1472 struct pagemapread *pm = walk->private; 1473 spinlock_t *ptl; 1474 pte_t *pte, *orig_pte; 1475 int err = 0; 1476 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1477 bool migration = false; 1478 1479 ptl = pmd_trans_huge_lock(pmdp, vma); 1480 if (ptl) { 1481 u64 flags = 0, frame = 0; 1482 pmd_t pmd = *pmdp; 1483 struct page *page = NULL; 1484 1485 if (vma->vm_flags & VM_SOFTDIRTY) 1486 flags |= PM_SOFT_DIRTY; 1487 1488 if (pmd_present(pmd)) { 1489 page = pmd_page(pmd); 1490 1491 flags |= PM_PRESENT; 1492 if (pmd_soft_dirty(pmd)) 1493 flags |= PM_SOFT_DIRTY; 1494 if (pmd_uffd_wp(pmd)) 1495 flags |= PM_UFFD_WP; 1496 if (pm->show_pfn) 1497 frame = pmd_pfn(pmd) + 1498 ((addr & ~PMD_MASK) >> PAGE_SHIFT); 1499 } 1500 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION 1501 else if (is_swap_pmd(pmd)) { 1502 swp_entry_t entry = pmd_to_swp_entry(pmd); 1503 unsigned long offset; 1504 1505 if (pm->show_pfn) { 1506 if (is_pfn_swap_entry(entry)) 1507 offset = swp_offset_pfn(entry); 1508 else 1509 offset = swp_offset(entry); 1510 offset = offset + 1511 ((addr & ~PMD_MASK) >> PAGE_SHIFT); 1512 frame = swp_type(entry) | 1513 (offset << MAX_SWAPFILES_SHIFT); 1514 } 1515 flags |= PM_SWAP; 1516 if (pmd_swp_soft_dirty(pmd)) 1517 flags |= PM_SOFT_DIRTY; 1518 if (pmd_swp_uffd_wp(pmd)) 1519 flags |= PM_UFFD_WP; 1520 VM_BUG_ON(!is_pmd_migration_entry(pmd)); 1521 migration = is_migration_entry(entry); 1522 page = pfn_swap_entry_to_page(entry); 1523 } 1524 #endif 1525 1526 if (page && !migration && page_mapcount(page) == 1) 1527 flags |= PM_MMAP_EXCLUSIVE; 1528 1529 for (; addr != end; addr += PAGE_SIZE) { 1530 pagemap_entry_t pme = make_pme(frame, flags); 1531 1532 err = add_to_pagemap(addr, &pme, pm); 1533 if (err) 1534 break; 1535 if (pm->show_pfn) { 1536 if (flags & PM_PRESENT) 1537 frame++; 1538 else if (flags & PM_SWAP) 1539 frame += (1 << MAX_SWAPFILES_SHIFT); 1540 } 1541 } 1542 spin_unlock(ptl); 1543 return err; 1544 } 1545 1546 if (pmd_trans_unstable(pmdp)) 1547 return 0; 1548 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ 1549 1550 /* 1551 * We can assume that @vma always points to a valid one and @end never 1552 * goes beyond vma->vm_end. 1553 */ 1554 orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl); 1555 for (; addr < end; pte++, addr += PAGE_SIZE) { 1556 pagemap_entry_t pme; 1557 1558 pme = pte_to_pagemap_entry(pm, vma, addr, *pte); 1559 err = add_to_pagemap(addr, &pme, pm); 1560 if (err) 1561 break; 1562 } 1563 pte_unmap_unlock(orig_pte, ptl); 1564 1565 cond_resched(); 1566 1567 return err; 1568 } 1569 1570 #ifdef CONFIG_HUGETLB_PAGE 1571 /* This function walks within one hugetlb entry in the single call */ 1572 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask, 1573 unsigned long addr, unsigned long end, 1574 struct mm_walk *walk) 1575 { 1576 struct pagemapread *pm = walk->private; 1577 struct vm_area_struct *vma = walk->vma; 1578 u64 flags = 0, frame = 0; 1579 int err = 0; 1580 pte_t pte; 1581 1582 if (vma->vm_flags & VM_SOFTDIRTY) 1583 flags |= PM_SOFT_DIRTY; 1584 1585 pte = huge_ptep_get(ptep); 1586 if (pte_present(pte)) { 1587 struct page *page = pte_page(pte); 1588 1589 if (!PageAnon(page)) 1590 flags |= PM_FILE; 1591 1592 if (page_mapcount(page) == 1) 1593 flags |= PM_MMAP_EXCLUSIVE; 1594 1595 if (huge_pte_uffd_wp(pte)) 1596 flags |= PM_UFFD_WP; 1597 1598 flags |= PM_PRESENT; 1599 if (pm->show_pfn) 1600 frame = pte_pfn(pte) + 1601 ((addr & ~hmask) >> PAGE_SHIFT); 1602 } else if (pte_swp_uffd_wp_any(pte)) { 1603 flags |= PM_UFFD_WP; 1604 } 1605 1606 for (; addr != end; addr += PAGE_SIZE) { 1607 pagemap_entry_t pme = make_pme(frame, flags); 1608 1609 err = add_to_pagemap(addr, &pme, pm); 1610 if (err) 1611 return err; 1612 if (pm->show_pfn && (flags & PM_PRESENT)) 1613 frame++; 1614 } 1615 1616 cond_resched(); 1617 1618 return err; 1619 } 1620 #else 1621 #define pagemap_hugetlb_range NULL 1622 #endif /* HUGETLB_PAGE */ 1623 1624 static const struct mm_walk_ops pagemap_ops = { 1625 .pmd_entry = pagemap_pmd_range, 1626 .pte_hole = pagemap_pte_hole, 1627 .hugetlb_entry = pagemap_hugetlb_range, 1628 }; 1629 1630 /* 1631 * /proc/pid/pagemap - an array mapping virtual pages to pfns 1632 * 1633 * For each page in the address space, this file contains one 64-bit entry 1634 * consisting of the following: 1635 * 1636 * Bits 0-54 page frame number (PFN) if present 1637 * Bits 0-4 swap type if swapped 1638 * Bits 5-54 swap offset if swapped 1639 * Bit 55 pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst) 1640 * Bit 56 page exclusively mapped 1641 * Bit 57 pte is uffd-wp write-protected 1642 * Bits 58-60 zero 1643 * Bit 61 page is file-page or shared-anon 1644 * Bit 62 page swapped 1645 * Bit 63 page present 1646 * 1647 * If the page is not present but in swap, then the PFN contains an 1648 * encoding of the swap file number and the page's offset into the 1649 * swap. Unmapped pages return a null PFN. This allows determining 1650 * precisely which pages are mapped (or in swap) and comparing mapped 1651 * pages between processes. 1652 * 1653 * Efficient users of this interface will use /proc/pid/maps to 1654 * determine which areas of memory are actually mapped and llseek to 1655 * skip over unmapped regions. 1656 */ 1657 static ssize_t pagemap_read(struct file *file, char __user *buf, 1658 size_t count, loff_t *ppos) 1659 { 1660 struct mm_struct *mm = file->private_data; 1661 struct pagemapread pm; 1662 unsigned long src; 1663 unsigned long svpfn; 1664 unsigned long start_vaddr; 1665 unsigned long end_vaddr; 1666 int ret = 0, copied = 0; 1667 1668 if (!mm || !mmget_not_zero(mm)) 1669 goto out; 1670 1671 ret = -EINVAL; 1672 /* file position must be aligned */ 1673 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES)) 1674 goto out_mm; 1675 1676 ret = 0; 1677 if (!count) 1678 goto out_mm; 1679 1680 /* do not disclose physical addresses: attack vector */ 1681 pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN); 1682 1683 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT); 1684 pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL); 1685 ret = -ENOMEM; 1686 if (!pm.buffer) 1687 goto out_mm; 1688 1689 src = *ppos; 1690 svpfn = src / PM_ENTRY_BYTES; 1691 end_vaddr = mm->task_size; 1692 1693 /* watch out for wraparound */ 1694 start_vaddr = end_vaddr; 1695 if (svpfn <= (ULONG_MAX >> PAGE_SHIFT)) 1696 start_vaddr = untagged_addr(svpfn << PAGE_SHIFT); 1697 1698 /* Ensure the address is inside the task */ 1699 if (start_vaddr > mm->task_size) 1700 start_vaddr = end_vaddr; 1701 1702 /* 1703 * The odds are that this will stop walking way 1704 * before end_vaddr, because the length of the 1705 * user buffer is tracked in "pm", and the walk 1706 * will stop when we hit the end of the buffer. 1707 */ 1708 ret = 0; 1709 while (count && (start_vaddr < end_vaddr)) { 1710 int len; 1711 unsigned long end; 1712 1713 pm.pos = 0; 1714 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK; 1715 /* overflow ? */ 1716 if (end < start_vaddr || end > end_vaddr) 1717 end = end_vaddr; 1718 ret = mmap_read_lock_killable(mm); 1719 if (ret) 1720 goto out_free; 1721 ret = walk_page_range(mm, start_vaddr, end, &pagemap_ops, &pm); 1722 mmap_read_unlock(mm); 1723 start_vaddr = end; 1724 1725 len = min(count, PM_ENTRY_BYTES * pm.pos); 1726 if (copy_to_user(buf, pm.buffer, len)) { 1727 ret = -EFAULT; 1728 goto out_free; 1729 } 1730 copied += len; 1731 buf += len; 1732 count -= len; 1733 } 1734 *ppos += copied; 1735 if (!ret || ret == PM_END_OF_BUFFER) 1736 ret = copied; 1737 1738 out_free: 1739 kfree(pm.buffer); 1740 out_mm: 1741 mmput(mm); 1742 out: 1743 return ret; 1744 } 1745 1746 static int pagemap_open(struct inode *inode, struct file *file) 1747 { 1748 struct mm_struct *mm; 1749 1750 mm = proc_mem_open(inode, PTRACE_MODE_READ); 1751 if (IS_ERR(mm)) 1752 return PTR_ERR(mm); 1753 file->private_data = mm; 1754 return 0; 1755 } 1756 1757 static int pagemap_release(struct inode *inode, struct file *file) 1758 { 1759 struct mm_struct *mm = file->private_data; 1760 1761 if (mm) 1762 mmdrop(mm); 1763 return 0; 1764 } 1765 1766 const struct file_operations proc_pagemap_operations = { 1767 .llseek = mem_lseek, /* borrow this */ 1768 .read = pagemap_read, 1769 .open = pagemap_open, 1770 .release = pagemap_release, 1771 }; 1772 #endif /* CONFIG_PROC_PAGE_MONITOR */ 1773 1774 #ifdef CONFIG_NUMA 1775 1776 struct numa_maps { 1777 unsigned long pages; 1778 unsigned long anon; 1779 unsigned long active; 1780 unsigned long writeback; 1781 unsigned long mapcount_max; 1782 unsigned long dirty; 1783 unsigned long swapcache; 1784 unsigned long node[MAX_NUMNODES]; 1785 }; 1786 1787 struct numa_maps_private { 1788 struct proc_maps_private proc_maps; 1789 struct numa_maps md; 1790 }; 1791 1792 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty, 1793 unsigned long nr_pages) 1794 { 1795 int count = page_mapcount(page); 1796 1797 md->pages += nr_pages; 1798 if (pte_dirty || PageDirty(page)) 1799 md->dirty += nr_pages; 1800 1801 if (PageSwapCache(page)) 1802 md->swapcache += nr_pages; 1803 1804 if (PageActive(page) || PageUnevictable(page)) 1805 md->active += nr_pages; 1806 1807 if (PageWriteback(page)) 1808 md->writeback += nr_pages; 1809 1810 if (PageAnon(page)) 1811 md->anon += nr_pages; 1812 1813 if (count > md->mapcount_max) 1814 md->mapcount_max = count; 1815 1816 md->node[page_to_nid(page)] += nr_pages; 1817 } 1818 1819 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma, 1820 unsigned long addr) 1821 { 1822 struct page *page; 1823 int nid; 1824 1825 if (!pte_present(pte)) 1826 return NULL; 1827 1828 page = vm_normal_page(vma, addr, pte); 1829 if (!page || is_zone_device_page(page)) 1830 return NULL; 1831 1832 if (PageReserved(page)) 1833 return NULL; 1834 1835 nid = page_to_nid(page); 1836 if (!node_isset(nid, node_states[N_MEMORY])) 1837 return NULL; 1838 1839 return page; 1840 } 1841 1842 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1843 static struct page *can_gather_numa_stats_pmd(pmd_t pmd, 1844 struct vm_area_struct *vma, 1845 unsigned long addr) 1846 { 1847 struct page *page; 1848 int nid; 1849 1850 if (!pmd_present(pmd)) 1851 return NULL; 1852 1853 page = vm_normal_page_pmd(vma, addr, pmd); 1854 if (!page) 1855 return NULL; 1856 1857 if (PageReserved(page)) 1858 return NULL; 1859 1860 nid = page_to_nid(page); 1861 if (!node_isset(nid, node_states[N_MEMORY])) 1862 return NULL; 1863 1864 return page; 1865 } 1866 #endif 1867 1868 static int gather_pte_stats(pmd_t *pmd, unsigned long addr, 1869 unsigned long end, struct mm_walk *walk) 1870 { 1871 struct numa_maps *md = walk->private; 1872 struct vm_area_struct *vma = walk->vma; 1873 spinlock_t *ptl; 1874 pte_t *orig_pte; 1875 pte_t *pte; 1876 1877 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1878 ptl = pmd_trans_huge_lock(pmd, vma); 1879 if (ptl) { 1880 struct page *page; 1881 1882 page = can_gather_numa_stats_pmd(*pmd, vma, addr); 1883 if (page) 1884 gather_stats(page, md, pmd_dirty(*pmd), 1885 HPAGE_PMD_SIZE/PAGE_SIZE); 1886 spin_unlock(ptl); 1887 return 0; 1888 } 1889 1890 if (pmd_trans_unstable(pmd)) 1891 return 0; 1892 #endif 1893 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl); 1894 do { 1895 struct page *page = can_gather_numa_stats(*pte, vma, addr); 1896 if (!page) 1897 continue; 1898 gather_stats(page, md, pte_dirty(*pte), 1); 1899 1900 } while (pte++, addr += PAGE_SIZE, addr != end); 1901 pte_unmap_unlock(orig_pte, ptl); 1902 cond_resched(); 1903 return 0; 1904 } 1905 #ifdef CONFIG_HUGETLB_PAGE 1906 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask, 1907 unsigned long addr, unsigned long end, struct mm_walk *walk) 1908 { 1909 pte_t huge_pte = huge_ptep_get(pte); 1910 struct numa_maps *md; 1911 struct page *page; 1912 1913 if (!pte_present(huge_pte)) 1914 return 0; 1915 1916 page = pte_page(huge_pte); 1917 1918 md = walk->private; 1919 gather_stats(page, md, pte_dirty(huge_pte), 1); 1920 return 0; 1921 } 1922 1923 #else 1924 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask, 1925 unsigned long addr, unsigned long end, struct mm_walk *walk) 1926 { 1927 return 0; 1928 } 1929 #endif 1930 1931 static const struct mm_walk_ops show_numa_ops = { 1932 .hugetlb_entry = gather_hugetlb_stats, 1933 .pmd_entry = gather_pte_stats, 1934 }; 1935 1936 /* 1937 * Display pages allocated per node and memory policy via /proc. 1938 */ 1939 static int show_numa_map(struct seq_file *m, void *v) 1940 { 1941 struct numa_maps_private *numa_priv = m->private; 1942 struct proc_maps_private *proc_priv = &numa_priv->proc_maps; 1943 struct vm_area_struct *vma = v; 1944 struct numa_maps *md = &numa_priv->md; 1945 struct file *file = vma->vm_file; 1946 struct mm_struct *mm = vma->vm_mm; 1947 struct mempolicy *pol; 1948 char buffer[64]; 1949 int nid; 1950 1951 if (!mm) 1952 return 0; 1953 1954 /* Ensure we start with an empty set of numa_maps statistics. */ 1955 memset(md, 0, sizeof(*md)); 1956 1957 pol = __get_vma_policy(vma, vma->vm_start); 1958 if (pol) { 1959 mpol_to_str(buffer, sizeof(buffer), pol); 1960 mpol_cond_put(pol); 1961 } else { 1962 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy); 1963 } 1964 1965 seq_printf(m, "%08lx %s", vma->vm_start, buffer); 1966 1967 if (file) { 1968 seq_puts(m, " file="); 1969 seq_file_path(m, file, "\n\t= "); 1970 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) { 1971 seq_puts(m, " heap"); 1972 } else if (is_stack(vma)) { 1973 seq_puts(m, " stack"); 1974 } 1975 1976 if (is_vm_hugetlb_page(vma)) 1977 seq_puts(m, " huge"); 1978 1979 /* mmap_lock is held by m_start */ 1980 walk_page_vma(vma, &show_numa_ops, md); 1981 1982 if (!md->pages) 1983 goto out; 1984 1985 if (md->anon) 1986 seq_printf(m, " anon=%lu", md->anon); 1987 1988 if (md->dirty) 1989 seq_printf(m, " dirty=%lu", md->dirty); 1990 1991 if (md->pages != md->anon && md->pages != md->dirty) 1992 seq_printf(m, " mapped=%lu", md->pages); 1993 1994 if (md->mapcount_max > 1) 1995 seq_printf(m, " mapmax=%lu", md->mapcount_max); 1996 1997 if (md->swapcache) 1998 seq_printf(m, " swapcache=%lu", md->swapcache); 1999 2000 if (md->active < md->pages && !is_vm_hugetlb_page(vma)) 2001 seq_printf(m, " active=%lu", md->active); 2002 2003 if (md->writeback) 2004 seq_printf(m, " writeback=%lu", md->writeback); 2005 2006 for_each_node_state(nid, N_MEMORY) 2007 if (md->node[nid]) 2008 seq_printf(m, " N%d=%lu", nid, md->node[nid]); 2009 2010 seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10); 2011 out: 2012 seq_putc(m, '\n'); 2013 return 0; 2014 } 2015 2016 static const struct seq_operations proc_pid_numa_maps_op = { 2017 .start = m_start, 2018 .next = m_next, 2019 .stop = m_stop, 2020 .show = show_numa_map, 2021 }; 2022 2023 static int pid_numa_maps_open(struct inode *inode, struct file *file) 2024 { 2025 return proc_maps_open(inode, file, &proc_pid_numa_maps_op, 2026 sizeof(struct numa_maps_private)); 2027 } 2028 2029 const struct file_operations proc_pid_numa_maps_operations = { 2030 .open = pid_numa_maps_open, 2031 .read = seq_read, 2032 .llseek = seq_lseek, 2033 .release = proc_map_release, 2034 }; 2035 2036 #endif /* CONFIG_NUMA */ 2037