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/ksm.h> 8#include <linux/seq_file.h> 9#include <linux/highmem.h> 10#include <linux/ptrace.h> 11#include <linux/slab.h> 12#include <linux/pagemap.h> 13#include <linux/mempolicy.h> 14#include <linux/rmap.h> 15#include <linux/swap.h> 16#include <linux/sched/mm.h> 17#include <linux/swapops.h> 18#include <linux/mmu_notifier.h> 19#include <linux/page_idle.h> 20#include <linux/shmem_fs.h> 21#include <linux/uaccess.h> 22#include <linux/pkeys.h> 23#ifdef CONFIG_MEM_PURGEABLE 24#include <linux/mm_purgeable.h> 25#endif 26 27#include <asm/elf.h> 28#include <asm/tlb.h> 29#include <asm/tlbflush.h> 30#include "internal.h" 31#include <linux/hck/lite_hck_hideaddr.h> 32 33#define SEQ_PUT_DEC(str, val) \ 34 seq_put_decimal_ull_width(m, str, (val) << (PAGE_SHIFT-10), 8) 35void task_mem(struct seq_file *m, struct mm_struct *mm) 36{ 37 unsigned long text, lib, swap, anon, file, shmem; 38 unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss; 39#ifdef CONFIG_MEM_PURGEABLE 40 unsigned long nr_purg_sum = 0, nr_purg_pin = 0; 41 42 mm_purg_pages_info(mm, &nr_purg_sum, &nr_purg_pin); 43#endif 44 45 anon = get_mm_counter(mm, MM_ANONPAGES); 46 file = get_mm_counter(mm, MM_FILEPAGES); 47 shmem = get_mm_counter(mm, MM_SHMEMPAGES); 48 49 /* 50 * Note: to minimize their overhead, mm maintains hiwater_vm and 51 * hiwater_rss only when about to *lower* total_vm or rss. Any 52 * collector of these hiwater stats must therefore get total_vm 53 * and rss too, which will usually be the higher. Barriers? not 54 * worth the effort, such snapshots can always be inconsistent. 55 */ 56 hiwater_vm = total_vm = mm->total_vm; 57 if (hiwater_vm < mm->hiwater_vm) 58 hiwater_vm = mm->hiwater_vm; 59 hiwater_rss = total_rss = anon + file + shmem; 60 if (hiwater_rss < mm->hiwater_rss) 61 hiwater_rss = mm->hiwater_rss; 62 63 /* split executable areas between text and lib */ 64 text = PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK); 65 text = min(text, mm->exec_vm << PAGE_SHIFT); 66 lib = (mm->exec_vm << PAGE_SHIFT) - text; 67 68 swap = get_mm_counter(mm, MM_SWAPENTS); 69 SEQ_PUT_DEC("VmPeak:\t", hiwater_vm); 70 SEQ_PUT_DEC(" kB\nVmSize:\t", total_vm); 71 SEQ_PUT_DEC(" kB\nVmLck:\t", mm->locked_vm); 72 SEQ_PUT_DEC(" kB\nVmPin:\t", atomic64_read(&mm->pinned_vm)); 73 SEQ_PUT_DEC(" kB\nVmHWM:\t", hiwater_rss); 74 SEQ_PUT_DEC(" kB\nVmRSS:\t", total_rss); 75 SEQ_PUT_DEC(" kB\nRssAnon:\t", anon); 76 SEQ_PUT_DEC(" kB\nRssFile:\t", file); 77 SEQ_PUT_DEC(" kB\nRssShmem:\t", shmem); 78 SEQ_PUT_DEC(" kB\nVmData:\t", mm->data_vm); 79 SEQ_PUT_DEC(" kB\nVmStk:\t", mm->stack_vm); 80 seq_put_decimal_ull_width(m, 81 " kB\nVmExe:\t", text >> 10, 8); 82 seq_put_decimal_ull_width(m, 83 " kB\nVmLib:\t", lib >> 10, 8); 84 seq_put_decimal_ull_width(m, 85 " kB\nVmPTE:\t", mm_pgtables_bytes(mm) >> 10, 8); 86 SEQ_PUT_DEC(" kB\nVmSwap:\t", swap); 87#ifdef CONFIG_MEM_PURGEABLE 88 SEQ_PUT_DEC(" kB\nPurgSum:\t", nr_purg_sum); 89 SEQ_PUT_DEC(" kB\nPurgPin:\t", nr_purg_pin); 90#endif 91 seq_puts(m, " kB\n"); 92 hugetlb_report_usage(m, mm); 93} 94#undef SEQ_PUT_DEC 95 96unsigned long task_vsize(struct mm_struct *mm) 97{ 98 return PAGE_SIZE * mm->total_vm; 99} 100 101unsigned long task_statm(struct mm_struct *mm, 102 unsigned long *shared, unsigned long *text, 103 unsigned long *data, unsigned long *resident) 104{ 105 *shared = get_mm_counter(mm, MM_FILEPAGES) + 106 get_mm_counter(mm, MM_SHMEMPAGES); 107 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) 108 >> PAGE_SHIFT; 109 *data = mm->data_vm + mm->stack_vm; 110 *resident = *shared + get_mm_counter(mm, MM_ANONPAGES); 111 return mm->total_vm; 112} 113 114#ifdef CONFIG_NUMA 115/* 116 * Save get_task_policy() for show_numa_map(). 117 */ 118static void hold_task_mempolicy(struct proc_maps_private *priv) 119{ 120 struct task_struct *task = priv->task; 121 122 task_lock(task); 123 priv->task_mempolicy = get_task_policy(task); 124 mpol_get(priv->task_mempolicy); 125 task_unlock(task); 126} 127static void release_task_mempolicy(struct proc_maps_private *priv) 128{ 129 mpol_put(priv->task_mempolicy); 130} 131#else 132static void hold_task_mempolicy(struct proc_maps_private *priv) 133{ 134} 135static void release_task_mempolicy(struct proc_maps_private *priv) 136{ 137} 138#endif 139 140static struct vm_area_struct *proc_get_vma(struct proc_maps_private *priv, 141 loff_t *ppos) 142{ 143 struct vm_area_struct *vma = vma_next(&priv->iter); 144 145 if (vma) { 146 *ppos = vma->vm_start; 147 } else { 148 *ppos = -2UL; 149 vma = get_gate_vma(priv->mm); 150 } 151 152 return vma; 153} 154 155static void *m_start(struct seq_file *m, loff_t *ppos) 156{ 157 struct proc_maps_private *priv = m->private; 158 unsigned long last_addr = *ppos; 159 struct mm_struct *mm; 160 161 /* See m_next(). Zero at the start or after lseek. */ 162 if (last_addr == -1UL) 163 return NULL; 164 165 priv->task = get_proc_task(priv->inode); 166 if (!priv->task) 167 return ERR_PTR(-ESRCH); 168 169 mm = priv->mm; 170 if (!mm || !mmget_not_zero(mm)) { 171 put_task_struct(priv->task); 172 priv->task = NULL; 173 return NULL; 174 } 175 176 if (mmap_read_lock_killable(mm)) { 177 mmput(mm); 178 put_task_struct(priv->task); 179 priv->task = NULL; 180 return ERR_PTR(-EINTR); 181 } 182 183 vma_iter_init(&priv->iter, mm, last_addr); 184 hold_task_mempolicy(priv); 185 if (last_addr == -2UL) 186 return get_gate_vma(mm); 187 188 return proc_get_vma(priv, ppos); 189} 190 191static void *m_next(struct seq_file *m, void *v, loff_t *ppos) 192{ 193 if (*ppos == -2UL) { 194 *ppos = -1UL; 195 return NULL; 196 } 197 return proc_get_vma(m->private, ppos); 198} 199 200static void m_stop(struct seq_file *m, void *v) 201{ 202 struct proc_maps_private *priv = m->private; 203 struct mm_struct *mm = priv->mm; 204 205 if (!priv->task) 206 return; 207 208 release_task_mempolicy(priv); 209 mmap_read_unlock(mm); 210 mmput(mm); 211 put_task_struct(priv->task); 212 priv->task = NULL; 213} 214 215static int proc_maps_open(struct inode *inode, struct file *file, 216 const struct seq_operations *ops, int psize) 217{ 218 struct proc_maps_private *priv = __seq_open_private(file, ops, psize); 219 220 if (!priv) 221 return -ENOMEM; 222 223 priv->inode = inode; 224 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ); 225 if (IS_ERR(priv->mm)) { 226 int err = PTR_ERR(priv->mm); 227 228 seq_release_private(inode, file); 229 return err; 230 } 231 232 return 0; 233} 234 235static int proc_map_release(struct inode *inode, struct file *file) 236{ 237 struct seq_file *seq = file->private_data; 238 struct proc_maps_private *priv = seq->private; 239 240 if (priv->mm) 241 mmdrop(priv->mm); 242 243 return seq_release_private(inode, file); 244} 245 246static int do_maps_open(struct inode *inode, struct file *file, 247 const struct seq_operations *ops) 248{ 249 return proc_maps_open(inode, file, ops, 250 sizeof(struct proc_maps_private)); 251} 252 253static void show_vma_header_prefix(struct seq_file *m, 254 unsigned long start, unsigned long end, 255 vm_flags_t flags, unsigned long long pgoff, 256 dev_t dev, unsigned long ino) 257{ 258 seq_setwidth(m, 25 + sizeof(void *) * 6 - 1); 259 seq_put_hex_ll(m, NULL, start, 8); 260 seq_put_hex_ll(m, "-", end, 8); 261 seq_putc(m, ' '); 262 seq_putc(m, flags & VM_READ ? 'r' : '-'); 263 seq_putc(m, flags & VM_WRITE ? 'w' : '-'); 264 seq_putc(m, flags & VM_EXEC ? 'x' : '-'); 265 seq_putc(m, flags & VM_MAYSHARE ? 's' : 'p'); 266 seq_put_hex_ll(m, " ", pgoff, 8); 267 seq_put_hex_ll(m, " ", MAJOR(dev), 2); 268 seq_put_hex_ll(m, ":", MINOR(dev), 2); 269 seq_put_decimal_ull(m, " ", ino); 270 seq_putc(m, ' '); 271} 272 273static void 274show_map_vma(struct seq_file *m, struct vm_area_struct *vma) 275{ 276 struct anon_vma_name *anon_name = NULL; 277 struct mm_struct *mm = vma->vm_mm; 278 struct file *file = vma->vm_file; 279 vm_flags_t flags = vma->vm_flags; 280 unsigned long ino = 0; 281 unsigned long long pgoff = 0; 282 unsigned long start, end; 283 dev_t dev = 0; 284 const char *name = NULL; 285 286 if (file) { 287 struct inode *inode = file_inode(vma->vm_file); 288 dev = inode->i_sb->s_dev; 289 ino = inode->i_ino; 290 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT; 291 } 292 293 start = vma->vm_start; 294 end = vma->vm_end; 295 CALL_HCK_LITE_HOOK(hideaddr_header_prefix_lhck, &start, &end, &flags, m, vma); 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_is_initial_heap(vma)) { 331 name = "[heap]"; 332 goto done; 333 } 334 335 if (vma_is_initial_stack(vma)) { 336 name = "[stack]"; 337 goto done; 338 } 339 340 if (anon_name) { 341 seq_pad(m, ' '); 342 seq_printf(m, "[anon:%s]", anon_name->name); 343 } 344 } 345 346done: 347 if (name) { 348 seq_pad(m, ' '); 349 seq_puts(m, name); 350 } 351 seq_putc(m, '\n'); 352} 353 354static int show_map(struct seq_file *m, void *v) 355{ 356 show_map_vma(m, v); 357 return 0; 358} 359 360static const struct seq_operations proc_pid_maps_op = { 361 .start = m_start, 362 .next = m_next, 363 .stop = m_stop, 364 .show = show_map 365}; 366 367static int pid_maps_open(struct inode *inode, struct file *file) 368{ 369 return do_maps_open(inode, file, &proc_pid_maps_op); 370} 371 372const struct file_operations proc_pid_maps_operations = { 373 .open = pid_maps_open, 374 .read = seq_read, 375 .llseek = seq_lseek, 376 .release = proc_map_release, 377}; 378 379/* 380 * Proportional Set Size(PSS): my share of RSS. 381 * 382 * PSS of a process is the count of pages it has in memory, where each 383 * page is divided by the number of processes sharing it. So if a 384 * process has 1000 pages all to itself, and 1000 shared with one other 385 * process, its PSS will be 1500. 386 * 387 * To keep (accumulated) division errors low, we adopt a 64bit 388 * fixed-point pss counter to minimize division errors. So (pss >> 389 * PSS_SHIFT) would be the real byte count. 390 * 391 * A shift of 12 before division means (assuming 4K page size): 392 * - 1M 3-user-pages add up to 8KB errors; 393 * - supports mapcount up to 2^24, or 16M; 394 * - supports PSS up to 2^52 bytes, or 4PB. 395 */ 396#define PSS_SHIFT 12 397 398#ifdef CONFIG_PROC_PAGE_MONITOR 399struct mem_size_stats { 400 unsigned long resident; 401 unsigned long shared_clean; 402 unsigned long shared_dirty; 403 unsigned long private_clean; 404 unsigned long private_dirty; 405 unsigned long referenced; 406 unsigned long anonymous; 407 unsigned long lazyfree; 408 unsigned long anonymous_thp; 409 unsigned long shmem_thp; 410 unsigned long file_thp; 411 unsigned long swap; 412 unsigned long shared_hugetlb; 413 unsigned long private_hugetlb; 414 unsigned long ksm; 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 424static 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 454static 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 if (PageKsm(page)) 472 mss->ksm += size; 473 474 mss->resident += size; 475 /* Accumulate the size in pages that have been accessed. */ 476 if (young || page_is_young(page) || PageReferenced(page)) 477 mss->referenced += size; 478 479 /* 480 * Then accumulate quantities that may depend on sharing, or that may 481 * differ page-by-page. 482 * 483 * page_count(page) == 1 guarantees the page is mapped exactly once. 484 * If any subpage of the compound page mapped with PTE it would elevate 485 * page_count(). 486 * 487 * The page_mapcount() is called to get a snapshot of the mapcount. 488 * Without holding the page lock this snapshot can be slightly wrong as 489 * we cannot always read the mapcount atomically. It is not safe to 490 * call page_mapcount() even with PTL held if the page is not mapped, 491 * especially for migration entries. Treat regular migration entries 492 * as mapcount == 1. 493 */ 494 if ((page_count(page) == 1) || migration) { 495 smaps_page_accumulate(mss, page, size, size << PSS_SHIFT, dirty, 496 locked, true); 497 return; 498 } 499 for (i = 0; i < nr; i++, page++) { 500 int mapcount = page_mapcount(page); 501 unsigned long pss = PAGE_SIZE << PSS_SHIFT; 502 if (mapcount >= 2) 503 pss /= mapcount; 504 smaps_page_accumulate(mss, page, PAGE_SIZE, pss, dirty, locked, 505 mapcount < 2); 506 } 507} 508 509#ifdef CONFIG_SHMEM 510static int smaps_pte_hole(unsigned long addr, unsigned long end, 511 __always_unused int depth, struct mm_walk *walk) 512{ 513 struct mem_size_stats *mss = walk->private; 514 struct vm_area_struct *vma = walk->vma; 515 516 mss->swap += shmem_partial_swap_usage(walk->vma->vm_file->f_mapping, 517 linear_page_index(vma, addr), 518 linear_page_index(vma, end)); 519 520 return 0; 521} 522#else 523#define smaps_pte_hole NULL 524#endif /* CONFIG_SHMEM */ 525 526static void smaps_pte_hole_lookup(unsigned long addr, struct mm_walk *walk) 527{ 528#ifdef CONFIG_SHMEM 529 if (walk->ops->pte_hole) { 530 /* depth is not used */ 531 smaps_pte_hole(addr, addr + PAGE_SIZE, 0, walk); 532 } 533#endif 534} 535 536static void smaps_pte_entry(pte_t *pte, unsigned long addr, 537 struct mm_walk *walk) 538{ 539 struct mem_size_stats *mss = walk->private; 540 struct vm_area_struct *vma = walk->vma; 541 bool locked = !!(vma->vm_flags & VM_LOCKED); 542 struct page *page = NULL; 543 bool migration = false, young = false, dirty = false; 544 pte_t ptent = ptep_get(pte); 545 546 if (pte_present(ptent)) { 547 page = vm_normal_page(vma, addr, ptent); 548 young = pte_young(ptent); 549 dirty = pte_dirty(ptent); 550 } else if (is_swap_pte(ptent)) { 551 swp_entry_t swpent = pte_to_swp_entry(ptent); 552 553 if (!non_swap_entry(swpent)) { 554 int mapcount; 555 556 mss->swap += PAGE_SIZE; 557 mapcount = swp_swapcount(swpent); 558 if (mapcount >= 2) { 559 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT; 560 561 do_div(pss_delta, mapcount); 562 mss->swap_pss += pss_delta; 563 } else { 564 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT; 565 } 566 } else if (is_pfn_swap_entry(swpent)) { 567 if (is_migration_entry(swpent)) 568 migration = true; 569 page = pfn_swap_entry_to_page(swpent); 570 } 571 } else { 572 smaps_pte_hole_lookup(addr, walk); 573 return; 574 } 575 576 if (!page) 577 return; 578 579 smaps_account(mss, page, false, young, dirty, locked, migration); 580} 581 582#ifdef CONFIG_TRANSPARENT_HUGEPAGE 583static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr, 584 struct mm_walk *walk) 585{ 586 struct mem_size_stats *mss = walk->private; 587 struct vm_area_struct *vma = walk->vma; 588 bool locked = !!(vma->vm_flags & VM_LOCKED); 589 struct page *page = NULL; 590 bool migration = false; 591 592 if (pmd_present(*pmd)) { 593 page = vm_normal_page_pmd(vma, addr, *pmd); 594 } else if (unlikely(thp_migration_supported() && is_swap_pmd(*pmd))) { 595 swp_entry_t entry = pmd_to_swp_entry(*pmd); 596 597 if (is_migration_entry(entry)) { 598 migration = true; 599 page = pfn_swap_entry_to_page(entry); 600 } 601 } 602 if (IS_ERR_OR_NULL(page)) 603 return; 604 if (PageAnon(page)) 605 mss->anonymous_thp += HPAGE_PMD_SIZE; 606 else if (PageSwapBacked(page)) 607 mss->shmem_thp += HPAGE_PMD_SIZE; 608 else if (is_zone_device_page(page)) 609 /* pass */; 610 else 611 mss->file_thp += HPAGE_PMD_SIZE; 612 613 smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd), 614 locked, migration); 615} 616#else 617static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr, 618 struct mm_walk *walk) 619{ 620} 621#endif 622 623static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, 624 struct mm_walk *walk) 625{ 626 struct vm_area_struct *vma = walk->vma; 627 pte_t *pte; 628 spinlock_t *ptl; 629 630 ptl = pmd_trans_huge_lock(pmd, vma); 631 if (ptl) { 632 smaps_pmd_entry(pmd, addr, walk); 633 spin_unlock(ptl); 634 goto out; 635 } 636 637 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); 638 if (!pte) { 639 walk->action = ACTION_AGAIN; 640 return 0; 641 } 642 for (; addr != end; pte++, addr += PAGE_SIZE) 643 smaps_pte_entry(pte, addr, walk); 644 pte_unmap_unlock(pte - 1, ptl); 645out: 646 cond_resched(); 647 return 0; 648} 649 650static 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#ifdef CONFIG_X86_USER_SHADOW_STACK 715 [ilog2(VM_SHADOW_STACK)] = "ss", 716#endif 717 }; 718 size_t i; 719 720 seq_puts(m, "VmFlags: "); 721 for (i = 0; i < BITS_PER_LONG; i++) { 722 if (!mnemonics[i][0]) 723 continue; 724 if (vma->vm_flags & (1UL << i)) { 725 seq_putc(m, mnemonics[i][0]); 726 seq_putc(m, mnemonics[i][1]); 727 seq_putc(m, ' '); 728 } 729 } 730 seq_putc(m, '\n'); 731} 732 733#ifdef CONFIG_HUGETLB_PAGE 734static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask, 735 unsigned long addr, unsigned long end, 736 struct mm_walk *walk) 737{ 738 struct mem_size_stats *mss = walk->private; 739 struct vm_area_struct *vma = walk->vma; 740 struct page *page = NULL; 741 pte_t ptent = ptep_get(pte); 742 743 if (pte_present(ptent)) { 744 page = vm_normal_page(vma, addr, ptent); 745 } else if (is_swap_pte(ptent)) { 746 swp_entry_t swpent = pte_to_swp_entry(ptent); 747 748 if (is_pfn_swap_entry(swpent)) 749 page = pfn_swap_entry_to_page(swpent); 750 } 751 if (page) { 752 if (page_mapcount(page) >= 2 || hugetlb_pmd_shared(pte)) 753 mss->shared_hugetlb += huge_page_size(hstate_vma(vma)); 754 else 755 mss->private_hugetlb += huge_page_size(hstate_vma(vma)); 756 } 757 return 0; 758} 759#else 760#define smaps_hugetlb_range NULL 761#endif /* HUGETLB_PAGE */ 762 763static const struct mm_walk_ops smaps_walk_ops = { 764 .pmd_entry = smaps_pte_range, 765 .hugetlb_entry = smaps_hugetlb_range, 766 .walk_lock = PGWALK_RDLOCK, 767}; 768 769static const struct mm_walk_ops smaps_shmem_walk_ops = { 770 .pmd_entry = smaps_pte_range, 771 .hugetlb_entry = smaps_hugetlb_range, 772 .pte_hole = smaps_pte_hole, 773 .walk_lock = PGWALK_RDLOCK, 774}; 775 776/* 777 * Gather mem stats from @vma with the indicated beginning 778 * address @start, and keep them in @mss. 779 * 780 * Use vm_start of @vma as the beginning address if @start is 0. 781 */ 782static void smap_gather_stats(struct vm_area_struct *vma, 783 struct mem_size_stats *mss, unsigned long start) 784{ 785 const struct mm_walk_ops *ops = &smaps_walk_ops; 786 787 /* Invalid start */ 788 if (start >= vma->vm_end) 789 return; 790 791 if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) { 792 /* 793 * For shared or readonly shmem mappings we know that all 794 * swapped out pages belong to the shmem object, and we can 795 * obtain the swap value much more efficiently. For private 796 * writable mappings, we might have COW pages that are 797 * not affected by the parent swapped out pages of the shmem 798 * object, so we have to distinguish them during the page walk. 799 * Unless we know that the shmem object (or the part mapped by 800 * our VMA) has no swapped out pages at all. 801 */ 802 unsigned long shmem_swapped = shmem_swap_usage(vma); 803 804 if (!start && (!shmem_swapped || (vma->vm_flags & VM_SHARED) || 805 !(vma->vm_flags & VM_WRITE))) { 806 mss->swap += shmem_swapped; 807 } else { 808 ops = &smaps_shmem_walk_ops; 809 } 810 } 811 812 /* mmap_lock is held in m_start */ 813 if (!start) 814 walk_page_vma(vma, ops, mss); 815 else 816 walk_page_range(vma->vm_mm, start, vma->vm_end, ops, mss); 817} 818 819#define SEQ_PUT_DEC(str, val) \ 820 seq_put_decimal_ull_width(m, str, (val) >> 10, 8) 821 822/* Show the contents common for smaps and smaps_rollup */ 823static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss, 824 bool rollup_mode) 825{ 826 SEQ_PUT_DEC("Rss: ", mss->resident); 827 SEQ_PUT_DEC(" kB\nPss: ", mss->pss >> PSS_SHIFT); 828 SEQ_PUT_DEC(" kB\nPss_Dirty: ", mss->pss_dirty >> PSS_SHIFT); 829 if (rollup_mode) { 830 /* 831 * These are meaningful only for smaps_rollup, otherwise two of 832 * them are zero, and the other one is the same as Pss. 833 */ 834 SEQ_PUT_DEC(" kB\nPss_Anon: ", 835 mss->pss_anon >> PSS_SHIFT); 836 SEQ_PUT_DEC(" kB\nPss_File: ", 837 mss->pss_file >> PSS_SHIFT); 838 SEQ_PUT_DEC(" kB\nPss_Shmem: ", 839 mss->pss_shmem >> PSS_SHIFT); 840 } 841 SEQ_PUT_DEC(" kB\nShared_Clean: ", mss->shared_clean); 842 SEQ_PUT_DEC(" kB\nShared_Dirty: ", mss->shared_dirty); 843 SEQ_PUT_DEC(" kB\nPrivate_Clean: ", mss->private_clean); 844 SEQ_PUT_DEC(" kB\nPrivate_Dirty: ", mss->private_dirty); 845 SEQ_PUT_DEC(" kB\nReferenced: ", mss->referenced); 846 SEQ_PUT_DEC(" kB\nAnonymous: ", mss->anonymous); 847 SEQ_PUT_DEC(" kB\nKSM: ", mss->ksm); 848 SEQ_PUT_DEC(" kB\nLazyFree: ", mss->lazyfree); 849 SEQ_PUT_DEC(" kB\nAnonHugePages: ", mss->anonymous_thp); 850 SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp); 851 SEQ_PUT_DEC(" kB\nFilePmdMapped: ", mss->file_thp); 852 SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb); 853 seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ", 854 mss->private_hugetlb >> 10, 7); 855 SEQ_PUT_DEC(" kB\nSwap: ", mss->swap); 856 SEQ_PUT_DEC(" kB\nSwapPss: ", 857 mss->swap_pss >> PSS_SHIFT); 858 SEQ_PUT_DEC(" kB\nLocked: ", 859 mss->pss_locked >> PSS_SHIFT); 860 seq_puts(m, " kB\n"); 861} 862 863static int show_smap(struct seq_file *m, void *v) 864{ 865 struct vm_area_struct *vma = v; 866 struct mem_size_stats mss; 867 868 memset(&mss, 0, sizeof(mss)); 869 870 smap_gather_stats(vma, &mss, 0); 871 872 show_map_vma(m, vma); 873 874 SEQ_PUT_DEC("Size: ", vma->vm_end - vma->vm_start); 875 SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma)); 876 SEQ_PUT_DEC(" kB\nMMUPageSize: ", vma_mmu_pagesize(vma)); 877 seq_puts(m, " kB\n"); 878 879 __show_smap(m, &mss, false); 880 881 seq_printf(m, "THPeligible: %8u\n", 882 hugepage_vma_check(vma, vma->vm_flags, true, false, true)); 883 884 if (arch_pkeys_enabled()) 885 seq_printf(m, "ProtectionKey: %8u\n", vma_pkey(vma)); 886 show_smap_vma_flags(m, vma); 887 888 return 0; 889} 890 891static int show_smaps_rollup(struct seq_file *m, void *v) 892{ 893 struct proc_maps_private *priv = m->private; 894 struct mem_size_stats mss; 895 struct mm_struct *mm = priv->mm; 896 struct vm_area_struct *vma; 897 unsigned long vma_start = 0, last_vma_end = 0; 898 int ret = 0; 899 VMA_ITERATOR(vmi, mm, 0); 900 901 priv->task = get_proc_task(priv->inode); 902 if (!priv->task) 903 return -ESRCH; 904 905 if (!mm || !mmget_not_zero(mm)) { 906 ret = -ESRCH; 907 goto out_put_task; 908 } 909 910 memset(&mss, 0, sizeof(mss)); 911 912 ret = mmap_read_lock_killable(mm); 913 if (ret) 914 goto out_put_mm; 915 916 hold_task_mempolicy(priv); 917 vma = vma_next(&vmi); 918 919 if (unlikely(!vma)) 920 goto empty_set; 921 922 vma_start = vma->vm_start; 923 do { 924 smap_gather_stats(vma, &mss, 0); 925 last_vma_end = vma->vm_end; 926 927 /* 928 * Release mmap_lock temporarily if someone wants to 929 * access it for write request. 930 */ 931 if (mmap_lock_is_contended(mm)) { 932 vma_iter_invalidate(&vmi); 933 mmap_read_unlock(mm); 934 ret = mmap_read_lock_killable(mm); 935 if (ret) { 936 release_task_mempolicy(priv); 937 goto out_put_mm; 938 } 939 940 /* 941 * After dropping the lock, there are four cases to 942 * consider. See the following example for explanation. 943 * 944 * +------+------+-----------+ 945 * | VMA1 | VMA2 | VMA3 | 946 * +------+------+-----------+ 947 * | | | | 948 * 4k 8k 16k 400k 949 * 950 * Suppose we drop the lock after reading VMA2 due to 951 * contention, then we get: 952 * 953 * last_vma_end = 16k 954 * 955 * 1) VMA2 is freed, but VMA3 exists: 956 * 957 * vma_next(vmi) will return VMA3. 958 * In this case, just continue from VMA3. 959 * 960 * 2) VMA2 still exists: 961 * 962 * vma_next(vmi) will return VMA3. 963 * In this case, just continue from VMA3. 964 * 965 * 3) No more VMAs can be found: 966 * 967 * vma_next(vmi) will return NULL. 968 * No more things to do, just break. 969 * 970 * 4) (last_vma_end - 1) is the middle of a vma (VMA'): 971 * 972 * vma_next(vmi) will return VMA' whose range 973 * contains last_vma_end. 974 * Iterate VMA' from last_vma_end. 975 */ 976 vma = vma_next(&vmi); 977 /* Case 3 above */ 978 if (!vma) 979 break; 980 981 /* Case 1 and 2 above */ 982 if (vma->vm_start >= last_vma_end) 983 continue; 984 985 /* Case 4 above */ 986 if (vma->vm_end > last_vma_end) 987 smap_gather_stats(vma, &mss, last_vma_end); 988 } 989 } for_each_vma(vmi, vma); 990 991empty_set: 992 show_vma_header_prefix(m, vma_start, last_vma_end, 0, 0, 0, 0); 993 seq_pad(m, ' '); 994 seq_puts(m, "[rollup]\n"); 995 996 __show_smap(m, &mss, true); 997 998 release_task_mempolicy(priv); 999 mmap_read_unlock(mm); 1000 1001out_put_mm: 1002 mmput(mm); 1003out_put_task: 1004 put_task_struct(priv->task); 1005 priv->task = NULL; 1006 1007 return ret; 1008} 1009#undef SEQ_PUT_DEC 1010 1011static const struct seq_operations proc_pid_smaps_op = { 1012 .start = m_start, 1013 .next = m_next, 1014 .stop = m_stop, 1015 .show = show_smap 1016}; 1017 1018static int pid_smaps_open(struct inode *inode, struct file *file) 1019{ 1020 return do_maps_open(inode, file, &proc_pid_smaps_op); 1021} 1022 1023static int smaps_rollup_open(struct inode *inode, struct file *file) 1024{ 1025 int ret; 1026 struct proc_maps_private *priv; 1027 1028 priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT); 1029 if (!priv) 1030 return -ENOMEM; 1031 1032 ret = single_open(file, show_smaps_rollup, priv); 1033 if (ret) 1034 goto out_free; 1035 1036 priv->inode = inode; 1037 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ); 1038 if (IS_ERR(priv->mm)) { 1039 ret = PTR_ERR(priv->mm); 1040 1041 single_release(inode, file); 1042 goto out_free; 1043 } 1044 1045 return 0; 1046 1047out_free: 1048 kfree(priv); 1049 return ret; 1050} 1051 1052static int smaps_rollup_release(struct inode *inode, struct file *file) 1053{ 1054 struct seq_file *seq = file->private_data; 1055 struct proc_maps_private *priv = seq->private; 1056 1057 if (priv->mm) 1058 mmdrop(priv->mm); 1059 1060 kfree(priv); 1061 return single_release(inode, file); 1062} 1063 1064const struct file_operations proc_pid_smaps_operations = { 1065 .open = pid_smaps_open, 1066 .read = seq_read, 1067 .llseek = seq_lseek, 1068 .release = proc_map_release, 1069}; 1070 1071const struct file_operations proc_pid_smaps_rollup_operations = { 1072 .open = smaps_rollup_open, 1073 .read = seq_read, 1074 .llseek = seq_lseek, 1075 .release = smaps_rollup_release, 1076}; 1077 1078enum clear_refs_types { 1079 CLEAR_REFS_ALL = 1, 1080 CLEAR_REFS_ANON, 1081 CLEAR_REFS_MAPPED, 1082 CLEAR_REFS_SOFT_DIRTY, 1083 CLEAR_REFS_MM_HIWATER_RSS, 1084 CLEAR_REFS_LAST, 1085}; 1086 1087struct clear_refs_private { 1088 enum clear_refs_types type; 1089}; 1090 1091#ifdef CONFIG_MEM_SOFT_DIRTY 1092 1093static inline bool pte_is_pinned(struct vm_area_struct *vma, unsigned long addr, pte_t pte) 1094{ 1095 struct page *page; 1096 1097 if (!pte_write(pte)) 1098 return false; 1099 if (!is_cow_mapping(vma->vm_flags)) 1100 return false; 1101 if (likely(!test_bit(MMF_HAS_PINNED, &vma->vm_mm->flags))) 1102 return false; 1103 page = vm_normal_page(vma, addr, pte); 1104 if (!page) 1105 return false; 1106 return page_maybe_dma_pinned(page); 1107} 1108 1109static inline void clear_soft_dirty(struct vm_area_struct *vma, 1110 unsigned long addr, pte_t *pte) 1111{ 1112 /* 1113 * The soft-dirty tracker uses #PF-s to catch writes 1114 * to pages, so write-protect the pte as well. See the 1115 * Documentation/admin-guide/mm/soft-dirty.rst for full description 1116 * of how soft-dirty works. 1117 */ 1118 pte_t ptent = ptep_get(pte); 1119 1120 if (pte_present(ptent)) { 1121 pte_t old_pte; 1122 1123 if (pte_is_pinned(vma, addr, ptent)) 1124 return; 1125 old_pte = ptep_modify_prot_start(vma, addr, pte); 1126 ptent = pte_wrprotect(old_pte); 1127 ptent = pte_clear_soft_dirty(ptent); 1128 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent); 1129 } else if (is_swap_pte(ptent)) { 1130 ptent = pte_swp_clear_soft_dirty(ptent); 1131 set_pte_at(vma->vm_mm, addr, pte, ptent); 1132 } 1133} 1134#else 1135static inline void clear_soft_dirty(struct vm_area_struct *vma, 1136 unsigned long addr, pte_t *pte) 1137{ 1138} 1139#endif 1140 1141#if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE) 1142static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma, 1143 unsigned long addr, pmd_t *pmdp) 1144{ 1145 pmd_t old, pmd = *pmdp; 1146 1147 if (pmd_present(pmd)) { 1148 /* See comment in change_huge_pmd() */ 1149 old = pmdp_invalidate(vma, addr, pmdp); 1150 if (pmd_dirty(old)) 1151 pmd = pmd_mkdirty(pmd); 1152 if (pmd_young(old)) 1153 pmd = pmd_mkyoung(pmd); 1154 1155 pmd = pmd_wrprotect(pmd); 1156 pmd = pmd_clear_soft_dirty(pmd); 1157 1158 set_pmd_at(vma->vm_mm, addr, pmdp, pmd); 1159 } else if (is_migration_entry(pmd_to_swp_entry(pmd))) { 1160 pmd = pmd_swp_clear_soft_dirty(pmd); 1161 set_pmd_at(vma->vm_mm, addr, pmdp, pmd); 1162 } 1163} 1164#else 1165static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma, 1166 unsigned long addr, pmd_t *pmdp) 1167{ 1168} 1169#endif 1170 1171static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr, 1172 unsigned long end, struct mm_walk *walk) 1173{ 1174 struct clear_refs_private *cp = walk->private; 1175 struct vm_area_struct *vma = walk->vma; 1176 pte_t *pte, ptent; 1177 spinlock_t *ptl; 1178 struct page *page; 1179 1180 ptl = pmd_trans_huge_lock(pmd, vma); 1181 if (ptl) { 1182 if (cp->type == CLEAR_REFS_SOFT_DIRTY) { 1183 clear_soft_dirty_pmd(vma, addr, pmd); 1184 goto out; 1185 } 1186 1187 if (!pmd_present(*pmd)) 1188 goto out; 1189 1190 page = pmd_page(*pmd); 1191 1192 /* Clear accessed and referenced bits. */ 1193 pmdp_test_and_clear_young(vma, addr, pmd); 1194 test_and_clear_page_young(page); 1195 ClearPageReferenced(page); 1196out: 1197 spin_unlock(ptl); 1198 return 0; 1199 } 1200 1201 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); 1202 if (!pte) { 1203 walk->action = ACTION_AGAIN; 1204 return 0; 1205 } 1206 for (; addr != end; pte++, addr += PAGE_SIZE) { 1207 ptent = ptep_get(pte); 1208 1209 if (cp->type == CLEAR_REFS_SOFT_DIRTY) { 1210 clear_soft_dirty(vma, addr, pte); 1211 continue; 1212 } 1213 1214 if (!pte_present(ptent)) 1215 continue; 1216 1217 page = vm_normal_page(vma, addr, ptent); 1218 if (!page) 1219 continue; 1220 1221 /* Clear accessed and referenced bits. */ 1222 ptep_test_and_clear_young(vma, addr, pte); 1223 test_and_clear_page_young(page); 1224 ClearPageReferenced(page); 1225 } 1226 pte_unmap_unlock(pte - 1, ptl); 1227 cond_resched(); 1228 return 0; 1229} 1230 1231static int clear_refs_test_walk(unsigned long start, unsigned long end, 1232 struct mm_walk *walk) 1233{ 1234 struct clear_refs_private *cp = walk->private; 1235 struct vm_area_struct *vma = walk->vma; 1236 1237 if (vma->vm_flags & VM_PFNMAP) 1238 return 1; 1239 1240 /* 1241 * Writing 1 to /proc/pid/clear_refs affects all pages. 1242 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages. 1243 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages. 1244 * Writing 4 to /proc/pid/clear_refs affects all pages. 1245 */ 1246 if (cp->type == CLEAR_REFS_ANON && vma->vm_file) 1247 return 1; 1248 if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file) 1249 return 1; 1250 return 0; 1251} 1252 1253static const struct mm_walk_ops clear_refs_walk_ops = { 1254 .pmd_entry = clear_refs_pte_range, 1255 .test_walk = clear_refs_test_walk, 1256 .walk_lock = PGWALK_WRLOCK, 1257}; 1258 1259static ssize_t clear_refs_write(struct file *file, const char __user *buf, 1260 size_t count, loff_t *ppos) 1261{ 1262 struct task_struct *task; 1263 char buffer[PROC_NUMBUF]; 1264 struct mm_struct *mm; 1265 struct vm_area_struct *vma; 1266 enum clear_refs_types type; 1267 int itype; 1268 int rv; 1269 1270 memset(buffer, 0, sizeof(buffer)); 1271 if (count > sizeof(buffer) - 1) 1272 count = sizeof(buffer) - 1; 1273 if (copy_from_user(buffer, buf, count)) 1274 return -EFAULT; 1275 rv = kstrtoint(strstrip(buffer), 10, &itype); 1276 if (rv < 0) 1277 return rv; 1278 type = (enum clear_refs_types)itype; 1279 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST) 1280 return -EINVAL; 1281 1282 task = get_proc_task(file_inode(file)); 1283 if (!task) 1284 return -ESRCH; 1285 mm = get_task_mm(task); 1286 if (mm) { 1287 VMA_ITERATOR(vmi, mm, 0); 1288 struct mmu_notifier_range range; 1289 struct clear_refs_private cp = { 1290 .type = type, 1291 }; 1292 1293 if (mmap_write_lock_killable(mm)) { 1294 count = -EINTR; 1295 goto out_mm; 1296 } 1297 if (type == CLEAR_REFS_MM_HIWATER_RSS) { 1298 /* 1299 * Writing 5 to /proc/pid/clear_refs resets the peak 1300 * resident set size to this mm's current rss value. 1301 */ 1302 reset_mm_hiwater_rss(mm); 1303 goto out_unlock; 1304 } 1305 1306 if (type == CLEAR_REFS_SOFT_DIRTY) { 1307 for_each_vma(vmi, vma) { 1308 if (!(vma->vm_flags & VM_SOFTDIRTY)) 1309 continue; 1310 vm_flags_clear(vma, VM_SOFTDIRTY); 1311 vma_set_page_prot(vma); 1312 } 1313 1314 inc_tlb_flush_pending(mm); 1315 mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY, 1316 0, mm, 0, -1UL); 1317 mmu_notifier_invalidate_range_start(&range); 1318 } 1319 walk_page_range(mm, 0, -1, &clear_refs_walk_ops, &cp); 1320 if (type == CLEAR_REFS_SOFT_DIRTY) { 1321 mmu_notifier_invalidate_range_end(&range); 1322 flush_tlb_mm(mm); 1323 dec_tlb_flush_pending(mm); 1324 } 1325out_unlock: 1326 mmap_write_unlock(mm); 1327out_mm: 1328 mmput(mm); 1329 } 1330 put_task_struct(task); 1331 1332 return count; 1333} 1334 1335const struct file_operations proc_clear_refs_operations = { 1336 .write = clear_refs_write, 1337 .llseek = noop_llseek, 1338}; 1339 1340typedef struct { 1341 u64 pme; 1342} pagemap_entry_t; 1343 1344struct pagemapread { 1345 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */ 1346 pagemap_entry_t *buffer; 1347 bool show_pfn; 1348}; 1349 1350#define PAGEMAP_WALK_SIZE (PMD_SIZE) 1351#define PAGEMAP_WALK_MASK (PMD_MASK) 1352 1353#define PM_ENTRY_BYTES sizeof(pagemap_entry_t) 1354#define PM_PFRAME_BITS 55 1355#define PM_PFRAME_MASK GENMASK_ULL(PM_PFRAME_BITS - 1, 0) 1356#define PM_SOFT_DIRTY BIT_ULL(55) 1357#define PM_MMAP_EXCLUSIVE BIT_ULL(56) 1358#define PM_UFFD_WP BIT_ULL(57) 1359#define PM_FILE BIT_ULL(61) 1360#define PM_SWAP BIT_ULL(62) 1361#define PM_PRESENT BIT_ULL(63) 1362 1363#define PM_END_OF_BUFFER 1 1364 1365static inline pagemap_entry_t make_pme(u64 frame, u64 flags) 1366{ 1367 return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags }; 1368} 1369 1370static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme, 1371 struct pagemapread *pm) 1372{ 1373 pm->buffer[pm->pos++] = *pme; 1374 if (pm->pos >= pm->len) 1375 return PM_END_OF_BUFFER; 1376 return 0; 1377} 1378 1379static int pagemap_pte_hole(unsigned long start, unsigned long end, 1380 __always_unused int depth, struct mm_walk *walk) 1381{ 1382 struct pagemapread *pm = walk->private; 1383 unsigned long addr = start; 1384 int err = 0; 1385 1386 while (addr < end) { 1387 struct vm_area_struct *vma = find_vma(walk->mm, addr); 1388 pagemap_entry_t pme = make_pme(0, 0); 1389 /* End of address space hole, which we mark as non-present. */ 1390 unsigned long hole_end; 1391 1392 if (vma) 1393 hole_end = min(end, vma->vm_start); 1394 else 1395 hole_end = end; 1396 1397 for (; addr < hole_end; addr += PAGE_SIZE) { 1398 err = add_to_pagemap(addr, &pme, pm); 1399 if (err) 1400 goto out; 1401 } 1402 1403 if (!vma) 1404 break; 1405 1406 /* Addresses in the VMA. */ 1407 if (vma->vm_flags & VM_SOFTDIRTY) 1408 pme = make_pme(0, PM_SOFT_DIRTY); 1409 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) { 1410 err = add_to_pagemap(addr, &pme, pm); 1411 if (err) 1412 goto out; 1413 } 1414 } 1415out: 1416 return err; 1417} 1418 1419static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm, 1420 struct vm_area_struct *vma, unsigned long addr, pte_t pte) 1421{ 1422 u64 frame = 0, flags = 0; 1423 struct page *page = NULL; 1424 bool migration = false; 1425 1426 if (pte_present(pte)) { 1427 if (pm->show_pfn) 1428 frame = pte_pfn(pte); 1429 flags |= PM_PRESENT; 1430 page = vm_normal_page(vma, addr, pte); 1431 if (pte_soft_dirty(pte)) 1432 flags |= PM_SOFT_DIRTY; 1433 if (pte_uffd_wp(pte)) 1434 flags |= PM_UFFD_WP; 1435 } else if (is_swap_pte(pte)) { 1436 swp_entry_t entry; 1437 if (pte_swp_soft_dirty(pte)) 1438 flags |= PM_SOFT_DIRTY; 1439 if (pte_swp_uffd_wp(pte)) 1440 flags |= PM_UFFD_WP; 1441 entry = pte_to_swp_entry(pte); 1442 if (pm->show_pfn) { 1443 pgoff_t offset; 1444 /* 1445 * For PFN swap offsets, keeping the offset field 1446 * to be PFN only to be compatible with old smaps. 1447 */ 1448 if (is_pfn_swap_entry(entry)) 1449 offset = swp_offset_pfn(entry); 1450 else 1451 offset = swp_offset(entry); 1452 frame = swp_type(entry) | 1453 (offset << MAX_SWAPFILES_SHIFT); 1454 } 1455 flags |= PM_SWAP; 1456 migration = is_migration_entry(entry); 1457 if (is_pfn_swap_entry(entry)) 1458 page = pfn_swap_entry_to_page(entry); 1459 if (pte_marker_entry_uffd_wp(entry)) 1460 flags |= PM_UFFD_WP; 1461 } 1462 1463 if (page && !PageAnon(page)) 1464 flags |= PM_FILE; 1465 if (page && !migration && page_mapcount(page) == 1) 1466 flags |= PM_MMAP_EXCLUSIVE; 1467 if (vma->vm_flags & VM_SOFTDIRTY) 1468 flags |= PM_SOFT_DIRTY; 1469 1470 return make_pme(frame, flags); 1471} 1472 1473static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end, 1474 struct mm_walk *walk) 1475{ 1476 struct vm_area_struct *vma = walk->vma; 1477 struct pagemapread *pm = walk->private; 1478 spinlock_t *ptl; 1479 pte_t *pte, *orig_pte; 1480 int err = 0; 1481#ifdef CONFIG_TRANSPARENT_HUGEPAGE 1482 bool migration = false; 1483 1484 ptl = pmd_trans_huge_lock(pmdp, vma); 1485 if (ptl) { 1486 u64 flags = 0, frame = 0; 1487 pmd_t pmd = *pmdp; 1488 struct page *page = NULL; 1489 1490 if (vma->vm_flags & VM_SOFTDIRTY) 1491 flags |= PM_SOFT_DIRTY; 1492 1493 if (pmd_present(pmd)) { 1494 page = pmd_page(pmd); 1495 1496 flags |= PM_PRESENT; 1497 if (pmd_soft_dirty(pmd)) 1498 flags |= PM_SOFT_DIRTY; 1499 if (pmd_uffd_wp(pmd)) 1500 flags |= PM_UFFD_WP; 1501 if (pm->show_pfn) 1502 frame = pmd_pfn(pmd) + 1503 ((addr & ~PMD_MASK) >> PAGE_SHIFT); 1504 } 1505#ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION 1506 else if (is_swap_pmd(pmd)) { 1507 swp_entry_t entry = pmd_to_swp_entry(pmd); 1508 unsigned long offset; 1509 1510 if (pm->show_pfn) { 1511 if (is_pfn_swap_entry(entry)) 1512 offset = swp_offset_pfn(entry); 1513 else 1514 offset = swp_offset(entry); 1515 offset = offset + 1516 ((addr & ~PMD_MASK) >> PAGE_SHIFT); 1517 frame = swp_type(entry) | 1518 (offset << MAX_SWAPFILES_SHIFT); 1519 } 1520 flags |= PM_SWAP; 1521 if (pmd_swp_soft_dirty(pmd)) 1522 flags |= PM_SOFT_DIRTY; 1523 if (pmd_swp_uffd_wp(pmd)) 1524 flags |= PM_UFFD_WP; 1525 VM_BUG_ON(!is_pmd_migration_entry(pmd)); 1526 migration = is_migration_entry(entry); 1527 page = pfn_swap_entry_to_page(entry); 1528 } 1529#endif 1530 1531 if (page && !migration && page_mapcount(page) == 1) 1532 flags |= PM_MMAP_EXCLUSIVE; 1533 1534 for (; addr != end; addr += PAGE_SIZE) { 1535 pagemap_entry_t pme = make_pme(frame, flags); 1536 1537 err = add_to_pagemap(addr, &pme, pm); 1538 if (err) 1539 break; 1540 if (pm->show_pfn) { 1541 if (flags & PM_PRESENT) 1542 frame++; 1543 else if (flags & PM_SWAP) 1544 frame += (1 << MAX_SWAPFILES_SHIFT); 1545 } 1546 } 1547 spin_unlock(ptl); 1548 return err; 1549 } 1550#endif /* CONFIG_TRANSPARENT_HUGEPAGE */ 1551 1552 /* 1553 * We can assume that @vma always points to a valid one and @end never 1554 * goes beyond vma->vm_end. 1555 */ 1556 orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl); 1557 if (!pte) { 1558 walk->action = ACTION_AGAIN; 1559 return err; 1560 } 1561 for (; addr < end; pte++, addr += PAGE_SIZE) { 1562 pagemap_entry_t pme; 1563 1564 pme = pte_to_pagemap_entry(pm, vma, addr, ptep_get(pte)); 1565 err = add_to_pagemap(addr, &pme, pm); 1566 if (err) 1567 break; 1568 } 1569 pte_unmap_unlock(orig_pte, ptl); 1570 1571 cond_resched(); 1572 1573 return err; 1574} 1575 1576#ifdef CONFIG_HUGETLB_PAGE 1577/* This function walks within one hugetlb entry in the single call */ 1578static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask, 1579 unsigned long addr, unsigned long end, 1580 struct mm_walk *walk) 1581{ 1582 struct pagemapread *pm = walk->private; 1583 struct vm_area_struct *vma = walk->vma; 1584 u64 flags = 0, frame = 0; 1585 int err = 0; 1586 pte_t pte; 1587 1588 if (vma->vm_flags & VM_SOFTDIRTY) 1589 flags |= PM_SOFT_DIRTY; 1590 1591 pte = huge_ptep_get(ptep); 1592 if (pte_present(pte)) { 1593 struct page *page = pte_page(pte); 1594 1595 if (!PageAnon(page)) 1596 flags |= PM_FILE; 1597 1598 if (page_mapcount(page) == 1) 1599 flags |= PM_MMAP_EXCLUSIVE; 1600 1601 if (huge_pte_uffd_wp(pte)) 1602 flags |= PM_UFFD_WP; 1603 1604 flags |= PM_PRESENT; 1605 if (pm->show_pfn) 1606 frame = pte_pfn(pte) + 1607 ((addr & ~hmask) >> PAGE_SHIFT); 1608 } else if (pte_swp_uffd_wp_any(pte)) { 1609 flags |= PM_UFFD_WP; 1610 } 1611 1612 for (; addr != end; addr += PAGE_SIZE) { 1613 pagemap_entry_t pme = make_pme(frame, flags); 1614 1615 err = add_to_pagemap(addr, &pme, pm); 1616 if (err) 1617 return err; 1618 if (pm->show_pfn && (flags & PM_PRESENT)) 1619 frame++; 1620 } 1621 1622 cond_resched(); 1623 1624 return err; 1625} 1626#else 1627#define pagemap_hugetlb_range NULL 1628#endif /* HUGETLB_PAGE */ 1629 1630static const struct mm_walk_ops pagemap_ops = { 1631 .pmd_entry = pagemap_pmd_range, 1632 .pte_hole = pagemap_pte_hole, 1633 .hugetlb_entry = pagemap_hugetlb_range, 1634 .walk_lock = PGWALK_RDLOCK, 1635}; 1636 1637/* 1638 * /proc/pid/pagemap - an array mapping virtual pages to pfns 1639 * 1640 * For each page in the address space, this file contains one 64-bit entry 1641 * consisting of the following: 1642 * 1643 * Bits 0-54 page frame number (PFN) if present 1644 * Bits 0-4 swap type if swapped 1645 * Bits 5-54 swap offset if swapped 1646 * Bit 55 pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst) 1647 * Bit 56 page exclusively mapped 1648 * Bit 57 pte is uffd-wp write-protected 1649 * Bits 58-60 zero 1650 * Bit 61 page is file-page or shared-anon 1651 * Bit 62 page swapped 1652 * Bit 63 page present 1653 * 1654 * If the page is not present but in swap, then the PFN contains an 1655 * encoding of the swap file number and the page's offset into the 1656 * swap. Unmapped pages return a null PFN. This allows determining 1657 * precisely which pages are mapped (or in swap) and comparing mapped 1658 * pages between processes. 1659 * 1660 * Efficient users of this interface will use /proc/pid/maps to 1661 * determine which areas of memory are actually mapped and llseek to 1662 * skip over unmapped regions. 1663 */ 1664static ssize_t pagemap_read(struct file *file, char __user *buf, 1665 size_t count, loff_t *ppos) 1666{ 1667 struct mm_struct *mm = file->private_data; 1668 struct pagemapread pm; 1669 unsigned long src; 1670 unsigned long svpfn; 1671 unsigned long start_vaddr; 1672 unsigned long end_vaddr; 1673 int ret = 0, copied = 0; 1674 1675 if (!mm || !mmget_not_zero(mm)) 1676 goto out; 1677 1678 ret = -EINVAL; 1679 /* file position must be aligned */ 1680 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES)) 1681 goto out_mm; 1682 1683 ret = 0; 1684 if (!count) 1685 goto out_mm; 1686 1687 /* do not disclose physical addresses: attack vector */ 1688 pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN); 1689 1690 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT); 1691 pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL); 1692 ret = -ENOMEM; 1693 if (!pm.buffer) 1694 goto out_mm; 1695 1696 src = *ppos; 1697 svpfn = src / PM_ENTRY_BYTES; 1698 end_vaddr = mm->task_size; 1699 1700 /* watch out for wraparound */ 1701 start_vaddr = end_vaddr; 1702 if (svpfn <= (ULONG_MAX >> PAGE_SHIFT)) { 1703 unsigned long end; 1704 1705 ret = mmap_read_lock_killable(mm); 1706 if (ret) 1707 goto out_free; 1708 start_vaddr = untagged_addr_remote(mm, svpfn << PAGE_SHIFT); 1709 mmap_read_unlock(mm); 1710 1711 end = start_vaddr + ((count / PM_ENTRY_BYTES) << PAGE_SHIFT); 1712 if (end >= start_vaddr && end < mm->task_size) 1713 end_vaddr = end; 1714 } 1715 1716 /* Ensure the address is inside the task */ 1717 if (start_vaddr > mm->task_size) 1718 start_vaddr = end_vaddr; 1719 1720 ret = 0; 1721 while (count && (start_vaddr < end_vaddr)) { 1722 int len; 1723 unsigned long end; 1724 1725 pm.pos = 0; 1726 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK; 1727 /* overflow ? */ 1728 if (end < start_vaddr || end > end_vaddr) 1729 end = end_vaddr; 1730 ret = mmap_read_lock_killable(mm); 1731 if (ret) 1732 goto out_free; 1733 ret = walk_page_range(mm, start_vaddr, end, &pagemap_ops, &pm); 1734 mmap_read_unlock(mm); 1735 start_vaddr = end; 1736 1737 len = min(count, PM_ENTRY_BYTES * pm.pos); 1738 if (copy_to_user(buf, pm.buffer, len)) { 1739 ret = -EFAULT; 1740 goto out_free; 1741 } 1742 copied += len; 1743 buf += len; 1744 count -= len; 1745 } 1746 *ppos += copied; 1747 if (!ret || ret == PM_END_OF_BUFFER) 1748 ret = copied; 1749 1750out_free: 1751 kfree(pm.buffer); 1752out_mm: 1753 mmput(mm); 1754out: 1755 return ret; 1756} 1757 1758static int pagemap_open(struct inode *inode, struct file *file) 1759{ 1760 struct mm_struct *mm; 1761 1762 mm = proc_mem_open(inode, PTRACE_MODE_READ); 1763 if (IS_ERR(mm)) 1764 return PTR_ERR(mm); 1765 file->private_data = mm; 1766 return 0; 1767} 1768 1769static int pagemap_release(struct inode *inode, struct file *file) 1770{ 1771 struct mm_struct *mm = file->private_data; 1772 1773 if (mm) 1774 mmdrop(mm); 1775 return 0; 1776} 1777 1778const struct file_operations proc_pagemap_operations = { 1779 .llseek = mem_lseek, /* borrow this */ 1780 .read = pagemap_read, 1781 .open = pagemap_open, 1782 .release = pagemap_release, 1783}; 1784#endif /* CONFIG_PROC_PAGE_MONITOR */ 1785 1786#ifdef CONFIG_NUMA 1787 1788struct numa_maps { 1789 unsigned long pages; 1790 unsigned long anon; 1791 unsigned long active; 1792 unsigned long writeback; 1793 unsigned long mapcount_max; 1794 unsigned long dirty; 1795 unsigned long swapcache; 1796 unsigned long node[MAX_NUMNODES]; 1797}; 1798 1799struct numa_maps_private { 1800 struct proc_maps_private proc_maps; 1801 struct numa_maps md; 1802}; 1803 1804static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty, 1805 unsigned long nr_pages) 1806{ 1807 int count = page_mapcount(page); 1808 1809 md->pages += nr_pages; 1810 if (pte_dirty || PageDirty(page)) 1811 md->dirty += nr_pages; 1812 1813 if (PageSwapCache(page)) 1814 md->swapcache += nr_pages; 1815 1816 if (PageActive(page) || PageUnevictable(page)) 1817 md->active += nr_pages; 1818 1819 if (PageWriteback(page)) 1820 md->writeback += nr_pages; 1821 1822 if (PageAnon(page)) 1823 md->anon += nr_pages; 1824 1825 if (count > md->mapcount_max) 1826 md->mapcount_max = count; 1827 1828 md->node[page_to_nid(page)] += nr_pages; 1829} 1830 1831static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma, 1832 unsigned long addr) 1833{ 1834 struct page *page; 1835 int nid; 1836 1837 if (!pte_present(pte)) 1838 return NULL; 1839 1840 page = vm_normal_page(vma, addr, pte); 1841 if (!page || is_zone_device_page(page)) 1842 return NULL; 1843 1844 if (PageReserved(page)) 1845 return NULL; 1846 1847 nid = page_to_nid(page); 1848 if (!node_isset(nid, node_states[N_MEMORY])) 1849 return NULL; 1850 1851 return page; 1852} 1853 1854#ifdef CONFIG_TRANSPARENT_HUGEPAGE 1855static struct page *can_gather_numa_stats_pmd(pmd_t pmd, 1856 struct vm_area_struct *vma, 1857 unsigned long addr) 1858{ 1859 struct page *page; 1860 int nid; 1861 1862 if (!pmd_present(pmd)) 1863 return NULL; 1864 1865 page = vm_normal_page_pmd(vma, addr, pmd); 1866 if (!page) 1867 return NULL; 1868 1869 if (PageReserved(page)) 1870 return NULL; 1871 1872 nid = page_to_nid(page); 1873 if (!node_isset(nid, node_states[N_MEMORY])) 1874 return NULL; 1875 1876 return page; 1877} 1878#endif 1879 1880static int gather_pte_stats(pmd_t *pmd, unsigned long addr, 1881 unsigned long end, struct mm_walk *walk) 1882{ 1883 struct numa_maps *md = walk->private; 1884 struct vm_area_struct *vma = walk->vma; 1885 spinlock_t *ptl; 1886 pte_t *orig_pte; 1887 pte_t *pte; 1888 1889#ifdef CONFIG_TRANSPARENT_HUGEPAGE 1890 ptl = pmd_trans_huge_lock(pmd, vma); 1891 if (ptl) { 1892 struct page *page; 1893 1894 page = can_gather_numa_stats_pmd(*pmd, vma, addr); 1895 if (page) 1896 gather_stats(page, md, pmd_dirty(*pmd), 1897 HPAGE_PMD_SIZE/PAGE_SIZE); 1898 spin_unlock(ptl); 1899 return 0; 1900 } 1901#endif 1902 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl); 1903 if (!pte) { 1904 walk->action = ACTION_AGAIN; 1905 return 0; 1906 } 1907 do { 1908 pte_t ptent = ptep_get(pte); 1909 struct page *page = can_gather_numa_stats(ptent, vma, addr); 1910 if (!page) 1911 continue; 1912 gather_stats(page, md, pte_dirty(ptent), 1); 1913 1914 } while (pte++, addr += PAGE_SIZE, addr != end); 1915 pte_unmap_unlock(orig_pte, ptl); 1916 cond_resched(); 1917 return 0; 1918} 1919#ifdef CONFIG_HUGETLB_PAGE 1920static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask, 1921 unsigned long addr, unsigned long end, struct mm_walk *walk) 1922{ 1923 pte_t huge_pte = huge_ptep_get(pte); 1924 struct numa_maps *md; 1925 struct page *page; 1926 1927 if (!pte_present(huge_pte)) 1928 return 0; 1929 1930 page = pte_page(huge_pte); 1931 1932 md = walk->private; 1933 gather_stats(page, md, pte_dirty(huge_pte), 1); 1934 return 0; 1935} 1936 1937#else 1938static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask, 1939 unsigned long addr, unsigned long end, struct mm_walk *walk) 1940{ 1941 return 0; 1942} 1943#endif 1944 1945static const struct mm_walk_ops show_numa_ops = { 1946 .hugetlb_entry = gather_hugetlb_stats, 1947 .pmd_entry = gather_pte_stats, 1948 .walk_lock = PGWALK_RDLOCK, 1949}; 1950 1951/* 1952 * Display pages allocated per node and memory policy via /proc. 1953 */ 1954static int show_numa_map(struct seq_file *m, void *v) 1955{ 1956 struct numa_maps_private *numa_priv = m->private; 1957 struct proc_maps_private *proc_priv = &numa_priv->proc_maps; 1958 struct vm_area_struct *vma = v; 1959 struct numa_maps *md = &numa_priv->md; 1960 struct file *file = vma->vm_file; 1961 struct mm_struct *mm = vma->vm_mm; 1962 struct mempolicy *pol; 1963 char buffer[64]; 1964 int nid; 1965 1966 if (!mm) 1967 return 0; 1968 1969 /* Ensure we start with an empty set of numa_maps statistics. */ 1970 memset(md, 0, sizeof(*md)); 1971 1972 pol = __get_vma_policy(vma, vma->vm_start); 1973 if (pol) { 1974 mpol_to_str(buffer, sizeof(buffer), pol); 1975 mpol_cond_put(pol); 1976 } else { 1977 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy); 1978 } 1979 1980 seq_printf(m, "%08lx %s", vma->vm_start, buffer); 1981 1982 if (file) { 1983 seq_puts(m, " file="); 1984 seq_file_path(m, file, "\n\t= "); 1985 } else if (vma_is_initial_heap(vma)) { 1986 seq_puts(m, " heap"); 1987 } else if (vma_is_initial_stack(vma)) { 1988 seq_puts(m, " stack"); 1989 } 1990 1991 if (is_vm_hugetlb_page(vma)) 1992 seq_puts(m, " huge"); 1993 1994 /* mmap_lock is held by m_start */ 1995 walk_page_vma(vma, &show_numa_ops, md); 1996 1997 if (!md->pages) 1998 goto out; 1999 2000 if (md->anon) 2001 seq_printf(m, " anon=%lu", md->anon); 2002 2003 if (md->dirty) 2004 seq_printf(m, " dirty=%lu", md->dirty); 2005 2006 if (md->pages != md->anon && md->pages != md->dirty) 2007 seq_printf(m, " mapped=%lu", md->pages); 2008 2009 if (md->mapcount_max > 1) 2010 seq_printf(m, " mapmax=%lu", md->mapcount_max); 2011 2012 if (md->swapcache) 2013 seq_printf(m, " swapcache=%lu", md->swapcache); 2014 2015 if (md->active < md->pages && !is_vm_hugetlb_page(vma)) 2016 seq_printf(m, " active=%lu", md->active); 2017 2018 if (md->writeback) 2019 seq_printf(m, " writeback=%lu", md->writeback); 2020 2021 for_each_node_state(nid, N_MEMORY) 2022 if (md->node[nid]) 2023 seq_printf(m, " N%d=%lu", nid, md->node[nid]); 2024 2025 seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10); 2026out: 2027 seq_putc(m, '\n'); 2028 return 0; 2029} 2030 2031static const struct seq_operations proc_pid_numa_maps_op = { 2032 .start = m_start, 2033 .next = m_next, 2034 .stop = m_stop, 2035 .show = show_numa_map, 2036}; 2037 2038static int pid_numa_maps_open(struct inode *inode, struct file *file) 2039{ 2040 return proc_maps_open(inode, file, &proc_pid_numa_maps_op, 2041 sizeof(struct numa_maps_private)); 2042} 2043 2044const struct file_operations proc_pid_numa_maps_operations = { 2045 .open = pid_numa_maps_open, 2046 .read = seq_read, 2047 .llseek = seq_lseek, 2048 .release = proc_map_release, 2049}; 2050 2051#endif /* CONFIG_NUMA */ 2052