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