1 #include <linux/mm.h> 2 #include <linux/vmacache.h> 3 #include <linux/hugetlb.h> 4 #include <linux/huge_mm.h> 5 #include <linux/mount.h> 6 #include <linux/seq_file.h> 7 #include <linux/highmem.h> 8 #include <linux/ptrace.h> 9 #include <linux/slab.h> 10 #include <linux/pagemap.h> 11 #include <linux/mempolicy.h> 12 #include <linux/rmap.h> 13 #include <linux/swap.h> 14 #include <linux/sched/mm.h> 15 #include <linux/swapops.h> 16 #include <linux/mmu_notifier.h> 17 #include <linux/page_idle.h> 18 #include <linux/shmem_fs.h> 19 20 #include <asm/elf.h> 21 #include <linux/uaccess.h> 22 #include <asm/tlbflush.h> 23 #include "internal.h" 24 25 void task_mem(struct seq_file *m, struct mm_struct *mm) 26 { 27 unsigned long text, lib, swap, ptes, pmds, anon, file, shmem; 28 unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss; 29 30 anon = get_mm_counter(mm, MM_ANONPAGES); 31 file = get_mm_counter(mm, MM_FILEPAGES); 32 shmem = get_mm_counter(mm, MM_SHMEMPAGES); 33 34 /* 35 * Note: to minimize their overhead, mm maintains hiwater_vm and 36 * hiwater_rss only when about to *lower* total_vm or rss. Any 37 * collector of these hiwater stats must therefore get total_vm 38 * and rss too, which will usually be the higher. Barriers? not 39 * worth the effort, such snapshots can always be inconsistent. 40 */ 41 hiwater_vm = total_vm = mm->total_vm; 42 if (hiwater_vm < mm->hiwater_vm) 43 hiwater_vm = mm->hiwater_vm; 44 hiwater_rss = total_rss = anon + file + shmem; 45 if (hiwater_rss < mm->hiwater_rss) 46 hiwater_rss = mm->hiwater_rss; 47 48 text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10; 49 lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text; 50 swap = get_mm_counter(mm, MM_SWAPENTS); 51 ptes = PTRS_PER_PTE * sizeof(pte_t) * atomic_long_read(&mm->nr_ptes); 52 pmds = PTRS_PER_PMD * sizeof(pmd_t) * mm_nr_pmds(mm); 53 seq_printf(m, 54 "VmPeak:\t%8lu kB\n" 55 "VmSize:\t%8lu kB\n" 56 "VmLck:\t%8lu kB\n" 57 "VmPin:\t%8lu kB\n" 58 "VmHWM:\t%8lu kB\n" 59 "VmRSS:\t%8lu kB\n" 60 "RssAnon:\t%8lu kB\n" 61 "RssFile:\t%8lu kB\n" 62 "RssShmem:\t%8lu kB\n" 63 "VmData:\t%8lu kB\n" 64 "VmStk:\t%8lu kB\n" 65 "VmExe:\t%8lu kB\n" 66 "VmLib:\t%8lu kB\n" 67 "VmPTE:\t%8lu kB\n" 68 "VmPMD:\t%8lu kB\n" 69 "VmSwap:\t%8lu kB\n", 70 hiwater_vm << (PAGE_SHIFT-10), 71 total_vm << (PAGE_SHIFT-10), 72 mm->locked_vm << (PAGE_SHIFT-10), 73 mm->pinned_vm << (PAGE_SHIFT-10), 74 hiwater_rss << (PAGE_SHIFT-10), 75 total_rss << (PAGE_SHIFT-10), 76 anon << (PAGE_SHIFT-10), 77 file << (PAGE_SHIFT-10), 78 shmem << (PAGE_SHIFT-10), 79 mm->data_vm << (PAGE_SHIFT-10), 80 mm->stack_vm << (PAGE_SHIFT-10), text, lib, 81 ptes >> 10, 82 pmds >> 10, 83 swap << (PAGE_SHIFT-10)); 84 hugetlb_report_usage(m, mm); 85 } 86 87 unsigned long task_vsize(struct mm_struct *mm) 88 { 89 return PAGE_SIZE * mm->total_vm; 90 } 91 92 unsigned long task_statm(struct mm_struct *mm, 93 unsigned long *shared, unsigned long *text, 94 unsigned long *data, unsigned long *resident) 95 { 96 *shared = get_mm_counter(mm, MM_FILEPAGES) + 97 get_mm_counter(mm, MM_SHMEMPAGES); 98 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) 99 >> PAGE_SHIFT; 100 *data = mm->data_vm + mm->stack_vm; 101 *resident = *shared + get_mm_counter(mm, MM_ANONPAGES); 102 return mm->total_vm; 103 } 104 105 #ifdef CONFIG_NUMA 106 /* 107 * Save get_task_policy() for show_numa_map(). 108 */ 109 static void hold_task_mempolicy(struct proc_maps_private *priv) 110 { 111 struct task_struct *task = priv->task; 112 113 task_lock(task); 114 priv->task_mempolicy = get_task_policy(task); 115 mpol_get(priv->task_mempolicy); 116 task_unlock(task); 117 } 118 static void release_task_mempolicy(struct proc_maps_private *priv) 119 { 120 mpol_put(priv->task_mempolicy); 121 } 122 #else 123 static void hold_task_mempolicy(struct proc_maps_private *priv) 124 { 125 } 126 static void release_task_mempolicy(struct proc_maps_private *priv) 127 { 128 } 129 #endif 130 131 static void vma_stop(struct proc_maps_private *priv) 132 { 133 struct mm_struct *mm = priv->mm; 134 135 release_task_mempolicy(priv); 136 up_read(&mm->mmap_sem); 137 mmput(mm); 138 } 139 140 static struct vm_area_struct * 141 m_next_vma(struct proc_maps_private *priv, struct vm_area_struct *vma) 142 { 143 if (vma == priv->tail_vma) 144 return NULL; 145 return vma->vm_next ?: priv->tail_vma; 146 } 147 148 static void m_cache_vma(struct seq_file *m, struct vm_area_struct *vma) 149 { 150 if (m->count < m->size) /* vma is copied successfully */ 151 m->version = m_next_vma(m->private, vma) ? vma->vm_end : -1UL; 152 } 153 154 static void *m_start(struct seq_file *m, loff_t *ppos) 155 { 156 struct proc_maps_private *priv = m->private; 157 unsigned long last_addr = m->version; 158 struct mm_struct *mm; 159 struct vm_area_struct *vma; 160 unsigned int pos = *ppos; 161 162 /* See m_cache_vma(). Zero at the start or after lseek. */ 163 if (last_addr == -1UL) 164 return NULL; 165 166 priv->task = get_proc_task(priv->inode); 167 if (!priv->task) 168 return ERR_PTR(-ESRCH); 169 170 mm = priv->mm; 171 if (!mm || !mmget_not_zero(mm)) 172 return NULL; 173 174 down_read(&mm->mmap_sem); 175 hold_task_mempolicy(priv); 176 priv->tail_vma = get_gate_vma(mm); 177 178 if (last_addr) { 179 vma = find_vma(mm, last_addr - 1); 180 if (vma && vma->vm_start <= last_addr) 181 vma = m_next_vma(priv, vma); 182 if (vma) 183 return vma; 184 } 185 186 m->version = 0; 187 if (pos < mm->map_count) { 188 for (vma = mm->mmap; pos; pos--) { 189 m->version = vma->vm_start; 190 vma = vma->vm_next; 191 } 192 return vma; 193 } 194 195 /* we do not bother to update m->version in this case */ 196 if (pos == mm->map_count && priv->tail_vma) 197 return priv->tail_vma; 198 199 vma_stop(priv); 200 return NULL; 201 } 202 203 static void *m_next(struct seq_file *m, void *v, loff_t *pos) 204 { 205 struct proc_maps_private *priv = m->private; 206 struct vm_area_struct *next; 207 208 (*pos)++; 209 next = m_next_vma(priv, v); 210 if (!next) 211 vma_stop(priv); 212 return next; 213 } 214 215 static void m_stop(struct seq_file *m, void *v) 216 { 217 struct proc_maps_private *priv = m->private; 218 219 if (!IS_ERR_OR_NULL(v)) 220 vma_stop(priv); 221 if (priv->task) { 222 put_task_struct(priv->task); 223 priv->task = NULL; 224 } 225 } 226 227 static int proc_maps_open(struct inode *inode, struct file *file, 228 const struct seq_operations *ops, int psize) 229 { 230 struct proc_maps_private *priv = __seq_open_private(file, ops, psize); 231 232 if (!priv) 233 return -ENOMEM; 234 235 priv->inode = inode; 236 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ); 237 if (IS_ERR(priv->mm)) { 238 int err = PTR_ERR(priv->mm); 239 240 seq_release_private(inode, file); 241 return err; 242 } 243 244 return 0; 245 } 246 247 static int proc_map_release(struct inode *inode, struct file *file) 248 { 249 struct seq_file *seq = file->private_data; 250 struct proc_maps_private *priv = seq->private; 251 252 if (priv->mm) 253 mmdrop(priv->mm); 254 255 return seq_release_private(inode, file); 256 } 257 258 static int do_maps_open(struct inode *inode, struct file *file, 259 const struct seq_operations *ops) 260 { 261 return proc_maps_open(inode, file, ops, 262 sizeof(struct proc_maps_private)); 263 } 264 265 /* 266 * Indicate if the VMA is a stack for the given task; for 267 * /proc/PID/maps that is the stack of the main task. 268 */ 269 static int is_stack(struct proc_maps_private *priv, 270 struct vm_area_struct *vma) 271 { 272 /* 273 * We make no effort to guess what a given thread considers to be 274 * its "stack". It's not even well-defined for programs written 275 * languages like Go. 276 */ 277 return vma->vm_start <= vma->vm_mm->start_stack && 278 vma->vm_end >= vma->vm_mm->start_stack; 279 } 280 281 static void 282 show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid) 283 { 284 struct mm_struct *mm = vma->vm_mm; 285 struct file *file = vma->vm_file; 286 struct proc_maps_private *priv = m->private; 287 vm_flags_t flags = vma->vm_flags; 288 unsigned long ino = 0; 289 unsigned long long pgoff = 0; 290 unsigned long start, end; 291 dev_t dev = 0; 292 const char *name = NULL; 293 294 if (file) { 295 struct inode *inode = file_inode(vma->vm_file); 296 dev = inode->i_sb->s_dev; 297 ino = inode->i_ino; 298 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT; 299 } 300 301 /* We don't show the stack guard page in /proc/maps */ 302 start = vma->vm_start; 303 end = vma->vm_end; 304 305 seq_setwidth(m, 25 + sizeof(void *) * 6 - 1); 306 seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu ", 307 start, 308 end, 309 flags & VM_READ ? 'r' : '-', 310 flags & VM_WRITE ? 'w' : '-', 311 flags & VM_EXEC ? 'x' : '-', 312 flags & VM_MAYSHARE ? 's' : 'p', 313 pgoff, 314 MAJOR(dev), MINOR(dev), ino); 315 316 /* 317 * Print the dentry name for named mappings, and a 318 * special [heap] marker for the heap: 319 */ 320 if (file) { 321 seq_pad(m, ' '); 322 seq_file_path(m, file, "\n"); 323 goto done; 324 } 325 326 if (vma->vm_ops && vma->vm_ops->name) { 327 name = vma->vm_ops->name(vma); 328 if (name) 329 goto done; 330 } 331 332 name = arch_vma_name(vma); 333 if (!name) { 334 if (!mm) { 335 name = "[vdso]"; 336 goto done; 337 } 338 339 if (vma->vm_start <= mm->brk && 340 vma->vm_end >= mm->start_brk) { 341 name = "[heap]"; 342 goto done; 343 } 344 345 if (is_stack(priv, vma)) 346 name = "[stack]"; 347 } 348 349 done: 350 if (name) { 351 seq_pad(m, ' '); 352 seq_puts(m, name); 353 } 354 seq_putc(m, '\n'); 355 } 356 357 static int show_map(struct seq_file *m, void *v, int is_pid) 358 { 359 show_map_vma(m, v, is_pid); 360 m_cache_vma(m, v); 361 return 0; 362 } 363 364 static int show_pid_map(struct seq_file *m, void *v) 365 { 366 return show_map(m, v, 1); 367 } 368 369 static int show_tid_map(struct seq_file *m, void *v) 370 { 371 return show_map(m, v, 0); 372 } 373 374 static const struct seq_operations proc_pid_maps_op = { 375 .start = m_start, 376 .next = m_next, 377 .stop = m_stop, 378 .show = show_pid_map 379 }; 380 381 static const struct seq_operations proc_tid_maps_op = { 382 .start = m_start, 383 .next = m_next, 384 .stop = m_stop, 385 .show = show_tid_map 386 }; 387 388 static int pid_maps_open(struct inode *inode, struct file *file) 389 { 390 return do_maps_open(inode, file, &proc_pid_maps_op); 391 } 392 393 static int tid_maps_open(struct inode *inode, struct file *file) 394 { 395 return do_maps_open(inode, file, &proc_tid_maps_op); 396 } 397 398 const struct file_operations proc_pid_maps_operations = { 399 .open = pid_maps_open, 400 .read = seq_read, 401 .llseek = seq_lseek, 402 .release = proc_map_release, 403 }; 404 405 const struct file_operations proc_tid_maps_operations = { 406 .open = tid_maps_open, 407 .read = seq_read, 408 .llseek = seq_lseek, 409 .release = proc_map_release, 410 }; 411 412 /* 413 * Proportional Set Size(PSS): my share of RSS. 414 * 415 * PSS of a process is the count of pages it has in memory, where each 416 * page is divided by the number of processes sharing it. So if a 417 * process has 1000 pages all to itself, and 1000 shared with one other 418 * process, its PSS will be 1500. 419 * 420 * To keep (accumulated) division errors low, we adopt a 64bit 421 * fixed-point pss counter to minimize division errors. So (pss >> 422 * PSS_SHIFT) would be the real byte count. 423 * 424 * A shift of 12 before division means (assuming 4K page size): 425 * - 1M 3-user-pages add up to 8KB errors; 426 * - supports mapcount up to 2^24, or 16M; 427 * - supports PSS up to 2^52 bytes, or 4PB. 428 */ 429 #define PSS_SHIFT 12 430 431 #ifdef CONFIG_PROC_PAGE_MONITOR 432 struct mem_size_stats { 433 unsigned long resident; 434 unsigned long shared_clean; 435 unsigned long shared_dirty; 436 unsigned long private_clean; 437 unsigned long private_dirty; 438 unsigned long referenced; 439 unsigned long anonymous; 440 unsigned long lazyfree; 441 unsigned long anonymous_thp; 442 unsigned long shmem_thp; 443 unsigned long swap; 444 unsigned long shared_hugetlb; 445 unsigned long private_hugetlb; 446 u64 pss; 447 u64 swap_pss; 448 bool check_shmem_swap; 449 }; 450 451 static void smaps_account(struct mem_size_stats *mss, struct page *page, 452 bool compound, bool young, bool dirty) 453 { 454 int i, nr = compound ? 1 << compound_order(page) : 1; 455 unsigned long size = nr * PAGE_SIZE; 456 457 if (PageAnon(page)) { 458 mss->anonymous += size; 459 if (!PageSwapBacked(page) && !dirty && !PageDirty(page)) 460 mss->lazyfree += size; 461 } 462 463 mss->resident += size; 464 /* Accumulate the size in pages that have been accessed. */ 465 if (young || page_is_young(page) || PageReferenced(page)) 466 mss->referenced += size; 467 468 /* 469 * page_count(page) == 1 guarantees the page is mapped exactly once. 470 * If any subpage of the compound page mapped with PTE it would elevate 471 * page_count(). 472 */ 473 if (page_count(page) == 1) { 474 if (dirty || PageDirty(page)) 475 mss->private_dirty += size; 476 else 477 mss->private_clean += size; 478 mss->pss += (u64)size << PSS_SHIFT; 479 return; 480 } 481 482 for (i = 0; i < nr; i++, page++) { 483 int mapcount = page_mapcount(page); 484 485 if (mapcount >= 2) { 486 if (dirty || PageDirty(page)) 487 mss->shared_dirty += PAGE_SIZE; 488 else 489 mss->shared_clean += PAGE_SIZE; 490 mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount; 491 } else { 492 if (dirty || PageDirty(page)) 493 mss->private_dirty += PAGE_SIZE; 494 else 495 mss->private_clean += PAGE_SIZE; 496 mss->pss += PAGE_SIZE << PSS_SHIFT; 497 } 498 } 499 } 500 501 #ifdef CONFIG_SHMEM 502 static int smaps_pte_hole(unsigned long addr, unsigned long end, 503 struct mm_walk *walk) 504 { 505 struct mem_size_stats *mss = walk->private; 506 507 mss->swap += shmem_partial_swap_usage( 508 walk->vma->vm_file->f_mapping, addr, end); 509 510 return 0; 511 } 512 #endif 513 514 static void smaps_pte_entry(pte_t *pte, unsigned long addr, 515 struct mm_walk *walk) 516 { 517 struct mem_size_stats *mss = walk->private; 518 struct vm_area_struct *vma = walk->vma; 519 struct page *page = NULL; 520 521 if (pte_present(*pte)) { 522 page = vm_normal_page(vma, addr, *pte); 523 } else if (is_swap_pte(*pte)) { 524 swp_entry_t swpent = pte_to_swp_entry(*pte); 525 526 if (!non_swap_entry(swpent)) { 527 int mapcount; 528 529 mss->swap += PAGE_SIZE; 530 mapcount = swp_swapcount(swpent); 531 if (mapcount >= 2) { 532 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT; 533 534 do_div(pss_delta, mapcount); 535 mss->swap_pss += pss_delta; 536 } else { 537 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT; 538 } 539 } else if (is_migration_entry(swpent)) 540 page = migration_entry_to_page(swpent); 541 } else if (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap 542 && pte_none(*pte))) { 543 page = find_get_entry(vma->vm_file->f_mapping, 544 linear_page_index(vma, addr)); 545 if (!page) 546 return; 547 548 if (radix_tree_exceptional_entry(page)) 549 mss->swap += PAGE_SIZE; 550 else 551 put_page(page); 552 553 return; 554 } 555 556 if (!page) 557 return; 558 559 smaps_account(mss, page, false, pte_young(*pte), pte_dirty(*pte)); 560 } 561 562 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 563 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr, 564 struct mm_walk *walk) 565 { 566 struct mem_size_stats *mss = walk->private; 567 struct vm_area_struct *vma = walk->vma; 568 struct page *page; 569 570 /* FOLL_DUMP will return -EFAULT on huge zero page */ 571 page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP); 572 if (IS_ERR_OR_NULL(page)) 573 return; 574 if (PageAnon(page)) 575 mss->anonymous_thp += HPAGE_PMD_SIZE; 576 else if (PageSwapBacked(page)) 577 mss->shmem_thp += HPAGE_PMD_SIZE; 578 else if (is_zone_device_page(page)) 579 /* pass */; 580 else 581 VM_BUG_ON_PAGE(1, page); 582 smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd)); 583 } 584 #else 585 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr, 586 struct mm_walk *walk) 587 { 588 } 589 #endif 590 591 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, 592 struct mm_walk *walk) 593 { 594 struct vm_area_struct *vma = walk->vma; 595 pte_t *pte; 596 spinlock_t *ptl; 597 598 ptl = pmd_trans_huge_lock(pmd, vma); 599 if (ptl) { 600 smaps_pmd_entry(pmd, addr, walk); 601 spin_unlock(ptl); 602 return 0; 603 } 604 605 if (pmd_trans_unstable(pmd)) 606 return 0; 607 /* 608 * The mmap_sem held all the way back in m_start() is what 609 * keeps khugepaged out of here and from collapsing things 610 * in here. 611 */ 612 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); 613 for (; addr != end; pte++, addr += PAGE_SIZE) 614 smaps_pte_entry(pte, addr, walk); 615 pte_unmap_unlock(pte - 1, ptl); 616 cond_resched(); 617 return 0; 618 } 619 620 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma) 621 { 622 /* 623 * Don't forget to update Documentation/ on changes. 624 */ 625 static const char mnemonics[BITS_PER_LONG][2] = { 626 /* 627 * In case if we meet a flag we don't know about. 628 */ 629 [0 ... (BITS_PER_LONG-1)] = "??", 630 631 [ilog2(VM_READ)] = "rd", 632 [ilog2(VM_WRITE)] = "wr", 633 [ilog2(VM_EXEC)] = "ex", 634 [ilog2(VM_SHARED)] = "sh", 635 [ilog2(VM_MAYREAD)] = "mr", 636 [ilog2(VM_MAYWRITE)] = "mw", 637 [ilog2(VM_MAYEXEC)] = "me", 638 [ilog2(VM_MAYSHARE)] = "ms", 639 [ilog2(VM_GROWSDOWN)] = "gd", 640 [ilog2(VM_PFNMAP)] = "pf", 641 [ilog2(VM_DENYWRITE)] = "dw", 642 #ifdef CONFIG_X86_INTEL_MPX 643 [ilog2(VM_MPX)] = "mp", 644 #endif 645 [ilog2(VM_LOCKED)] = "lo", 646 [ilog2(VM_IO)] = "io", 647 [ilog2(VM_SEQ_READ)] = "sr", 648 [ilog2(VM_RAND_READ)] = "rr", 649 [ilog2(VM_DONTCOPY)] = "dc", 650 [ilog2(VM_DONTEXPAND)] = "de", 651 [ilog2(VM_ACCOUNT)] = "ac", 652 [ilog2(VM_NORESERVE)] = "nr", 653 [ilog2(VM_HUGETLB)] = "ht", 654 [ilog2(VM_ARCH_1)] = "ar", 655 [ilog2(VM_DONTDUMP)] = "dd", 656 #ifdef CONFIG_MEM_SOFT_DIRTY 657 [ilog2(VM_SOFTDIRTY)] = "sd", 658 #endif 659 [ilog2(VM_MIXEDMAP)] = "mm", 660 [ilog2(VM_HUGEPAGE)] = "hg", 661 [ilog2(VM_NOHUGEPAGE)] = "nh", 662 [ilog2(VM_MERGEABLE)] = "mg", 663 [ilog2(VM_UFFD_MISSING)]= "um", 664 [ilog2(VM_UFFD_WP)] = "uw", 665 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS 666 /* These come out via ProtectionKey: */ 667 [ilog2(VM_PKEY_BIT0)] = "", 668 [ilog2(VM_PKEY_BIT1)] = "", 669 [ilog2(VM_PKEY_BIT2)] = "", 670 [ilog2(VM_PKEY_BIT3)] = "", 671 #endif 672 }; 673 size_t i; 674 675 seq_puts(m, "VmFlags: "); 676 for (i = 0; i < BITS_PER_LONG; i++) { 677 if (!mnemonics[i][0]) 678 continue; 679 if (vma->vm_flags & (1UL << i)) { 680 seq_printf(m, "%c%c ", 681 mnemonics[i][0], mnemonics[i][1]); 682 } 683 } 684 seq_putc(m, '\n'); 685 } 686 687 #ifdef CONFIG_HUGETLB_PAGE 688 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask, 689 unsigned long addr, unsigned long end, 690 struct mm_walk *walk) 691 { 692 struct mem_size_stats *mss = walk->private; 693 struct vm_area_struct *vma = walk->vma; 694 struct page *page = NULL; 695 696 if (pte_present(*pte)) { 697 page = vm_normal_page(vma, addr, *pte); 698 } else if (is_swap_pte(*pte)) { 699 swp_entry_t swpent = pte_to_swp_entry(*pte); 700 701 if (is_migration_entry(swpent)) 702 page = migration_entry_to_page(swpent); 703 } 704 if (page) { 705 int mapcount = page_mapcount(page); 706 707 if (mapcount >= 2) 708 mss->shared_hugetlb += huge_page_size(hstate_vma(vma)); 709 else 710 mss->private_hugetlb += huge_page_size(hstate_vma(vma)); 711 } 712 return 0; 713 } 714 #endif /* HUGETLB_PAGE */ 715 716 void __weak arch_show_smap(struct seq_file *m, struct vm_area_struct *vma) 717 { 718 } 719 720 static int show_smap(struct seq_file *m, void *v, int is_pid) 721 { 722 struct vm_area_struct *vma = v; 723 struct mem_size_stats mss; 724 struct mm_walk smaps_walk = { 725 .pmd_entry = smaps_pte_range, 726 #ifdef CONFIG_HUGETLB_PAGE 727 .hugetlb_entry = smaps_hugetlb_range, 728 #endif 729 .mm = vma->vm_mm, 730 .private = &mss, 731 }; 732 733 memset(&mss, 0, sizeof mss); 734 735 #ifdef CONFIG_SHMEM 736 if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) { 737 /* 738 * For shared or readonly shmem mappings we know that all 739 * swapped out pages belong to the shmem object, and we can 740 * obtain the swap value much more efficiently. For private 741 * writable mappings, we might have COW pages that are 742 * not affected by the parent swapped out pages of the shmem 743 * object, so we have to distinguish them during the page walk. 744 * Unless we know that the shmem object (or the part mapped by 745 * our VMA) has no swapped out pages at all. 746 */ 747 unsigned long shmem_swapped = shmem_swap_usage(vma); 748 749 if (!shmem_swapped || (vma->vm_flags & VM_SHARED) || 750 !(vma->vm_flags & VM_WRITE)) { 751 mss.swap = shmem_swapped; 752 } else { 753 mss.check_shmem_swap = true; 754 smaps_walk.pte_hole = smaps_pte_hole; 755 } 756 } 757 #endif 758 759 /* mmap_sem is held in m_start */ 760 walk_page_vma(vma, &smaps_walk); 761 762 show_map_vma(m, vma, is_pid); 763 764 seq_printf(m, 765 "Size: %8lu kB\n" 766 "Rss: %8lu kB\n" 767 "Pss: %8lu kB\n" 768 "Shared_Clean: %8lu kB\n" 769 "Shared_Dirty: %8lu kB\n" 770 "Private_Clean: %8lu kB\n" 771 "Private_Dirty: %8lu kB\n" 772 "Referenced: %8lu kB\n" 773 "Anonymous: %8lu kB\n" 774 "LazyFree: %8lu kB\n" 775 "AnonHugePages: %8lu kB\n" 776 "ShmemPmdMapped: %8lu kB\n" 777 "Shared_Hugetlb: %8lu kB\n" 778 "Private_Hugetlb: %7lu kB\n" 779 "Swap: %8lu kB\n" 780 "SwapPss: %8lu kB\n" 781 "KernelPageSize: %8lu kB\n" 782 "MMUPageSize: %8lu kB\n" 783 "Locked: %8lu kB\n", 784 (vma->vm_end - vma->vm_start) >> 10, 785 mss.resident >> 10, 786 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)), 787 mss.shared_clean >> 10, 788 mss.shared_dirty >> 10, 789 mss.private_clean >> 10, 790 mss.private_dirty >> 10, 791 mss.referenced >> 10, 792 mss.anonymous >> 10, 793 mss.lazyfree >> 10, 794 mss.anonymous_thp >> 10, 795 mss.shmem_thp >> 10, 796 mss.shared_hugetlb >> 10, 797 mss.private_hugetlb >> 10, 798 mss.swap >> 10, 799 (unsigned long)(mss.swap_pss >> (10 + PSS_SHIFT)), 800 vma_kernel_pagesize(vma) >> 10, 801 vma_mmu_pagesize(vma) >> 10, 802 (vma->vm_flags & VM_LOCKED) ? 803 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0); 804 805 arch_show_smap(m, vma); 806 show_smap_vma_flags(m, vma); 807 m_cache_vma(m, vma); 808 return 0; 809 } 810 811 static int show_pid_smap(struct seq_file *m, void *v) 812 { 813 return show_smap(m, v, 1); 814 } 815 816 static int show_tid_smap(struct seq_file *m, void *v) 817 { 818 return show_smap(m, v, 0); 819 } 820 821 static const struct seq_operations proc_pid_smaps_op = { 822 .start = m_start, 823 .next = m_next, 824 .stop = m_stop, 825 .show = show_pid_smap 826 }; 827 828 static const struct seq_operations proc_tid_smaps_op = { 829 .start = m_start, 830 .next = m_next, 831 .stop = m_stop, 832 .show = show_tid_smap 833 }; 834 835 static int pid_smaps_open(struct inode *inode, struct file *file) 836 { 837 return do_maps_open(inode, file, &proc_pid_smaps_op); 838 } 839 840 static int tid_smaps_open(struct inode *inode, struct file *file) 841 { 842 return do_maps_open(inode, file, &proc_tid_smaps_op); 843 } 844 845 const struct file_operations proc_pid_smaps_operations = { 846 .open = pid_smaps_open, 847 .read = seq_read, 848 .llseek = seq_lseek, 849 .release = proc_map_release, 850 }; 851 852 const struct file_operations proc_tid_smaps_operations = { 853 .open = tid_smaps_open, 854 .read = seq_read, 855 .llseek = seq_lseek, 856 .release = proc_map_release, 857 }; 858 859 enum clear_refs_types { 860 CLEAR_REFS_ALL = 1, 861 CLEAR_REFS_ANON, 862 CLEAR_REFS_MAPPED, 863 CLEAR_REFS_SOFT_DIRTY, 864 CLEAR_REFS_MM_HIWATER_RSS, 865 CLEAR_REFS_LAST, 866 }; 867 868 struct clear_refs_private { 869 enum clear_refs_types type; 870 }; 871 872 #ifdef CONFIG_MEM_SOFT_DIRTY 873 static inline void clear_soft_dirty(struct vm_area_struct *vma, 874 unsigned long addr, pte_t *pte) 875 { 876 /* 877 * The soft-dirty tracker uses #PF-s to catch writes 878 * to pages, so write-protect the pte as well. See the 879 * Documentation/vm/soft-dirty.txt for full description 880 * of how soft-dirty works. 881 */ 882 pte_t ptent = *pte; 883 884 if (pte_present(ptent)) { 885 ptent = ptep_modify_prot_start(vma->vm_mm, addr, pte); 886 ptent = pte_wrprotect(ptent); 887 ptent = pte_clear_soft_dirty(ptent); 888 ptep_modify_prot_commit(vma->vm_mm, addr, pte, ptent); 889 } else if (is_swap_pte(ptent)) { 890 ptent = pte_swp_clear_soft_dirty(ptent); 891 set_pte_at(vma->vm_mm, addr, pte, ptent); 892 } 893 } 894 #else 895 static inline void clear_soft_dirty(struct vm_area_struct *vma, 896 unsigned long addr, pte_t *pte) 897 { 898 } 899 #endif 900 901 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE) 902 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma, 903 unsigned long addr, pmd_t *pmdp) 904 { 905 pmd_t pmd = *pmdp; 906 907 /* See comment in change_huge_pmd() */ 908 pmdp_invalidate(vma, addr, pmdp); 909 if (pmd_dirty(*pmdp)) 910 pmd = pmd_mkdirty(pmd); 911 if (pmd_young(*pmdp)) 912 pmd = pmd_mkyoung(pmd); 913 914 pmd = pmd_wrprotect(pmd); 915 pmd = pmd_clear_soft_dirty(pmd); 916 917 set_pmd_at(vma->vm_mm, addr, pmdp, pmd); 918 } 919 #else 920 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma, 921 unsigned long addr, pmd_t *pmdp) 922 { 923 } 924 #endif 925 926 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr, 927 unsigned long end, struct mm_walk *walk) 928 { 929 struct clear_refs_private *cp = walk->private; 930 struct vm_area_struct *vma = walk->vma; 931 pte_t *pte, ptent; 932 spinlock_t *ptl; 933 struct page *page; 934 935 ptl = pmd_trans_huge_lock(pmd, vma); 936 if (ptl) { 937 if (cp->type == CLEAR_REFS_SOFT_DIRTY) { 938 clear_soft_dirty_pmd(vma, addr, pmd); 939 goto out; 940 } 941 942 page = pmd_page(*pmd); 943 944 /* Clear accessed and referenced bits. */ 945 pmdp_test_and_clear_young(vma, addr, pmd); 946 test_and_clear_page_young(page); 947 ClearPageReferenced(page); 948 out: 949 spin_unlock(ptl); 950 return 0; 951 } 952 953 if (pmd_trans_unstable(pmd)) 954 return 0; 955 956 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); 957 for (; addr != end; pte++, addr += PAGE_SIZE) { 958 ptent = *pte; 959 960 if (cp->type == CLEAR_REFS_SOFT_DIRTY) { 961 clear_soft_dirty(vma, addr, pte); 962 continue; 963 } 964 965 if (!pte_present(ptent)) 966 continue; 967 968 page = vm_normal_page(vma, addr, ptent); 969 if (!page) 970 continue; 971 972 /* Clear accessed and referenced bits. */ 973 ptep_test_and_clear_young(vma, addr, pte); 974 test_and_clear_page_young(page); 975 ClearPageReferenced(page); 976 } 977 pte_unmap_unlock(pte - 1, ptl); 978 cond_resched(); 979 return 0; 980 } 981 982 static int clear_refs_test_walk(unsigned long start, unsigned long end, 983 struct mm_walk *walk) 984 { 985 struct clear_refs_private *cp = walk->private; 986 struct vm_area_struct *vma = walk->vma; 987 988 if (vma->vm_flags & VM_PFNMAP) 989 return 1; 990 991 /* 992 * Writing 1 to /proc/pid/clear_refs affects all pages. 993 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages. 994 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages. 995 * Writing 4 to /proc/pid/clear_refs affects all pages. 996 */ 997 if (cp->type == CLEAR_REFS_ANON && vma->vm_file) 998 return 1; 999 if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file) 1000 return 1; 1001 return 0; 1002 } 1003 1004 static ssize_t clear_refs_write(struct file *file, const char __user *buf, 1005 size_t count, loff_t *ppos) 1006 { 1007 struct task_struct *task; 1008 char buffer[PROC_NUMBUF]; 1009 struct mm_struct *mm; 1010 struct vm_area_struct *vma; 1011 enum clear_refs_types type; 1012 int itype; 1013 int rv; 1014 1015 memset(buffer, 0, sizeof(buffer)); 1016 if (count > sizeof(buffer) - 1) 1017 count = sizeof(buffer) - 1; 1018 if (copy_from_user(buffer, buf, count)) 1019 return -EFAULT; 1020 rv = kstrtoint(strstrip(buffer), 10, &itype); 1021 if (rv < 0) 1022 return rv; 1023 type = (enum clear_refs_types)itype; 1024 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST) 1025 return -EINVAL; 1026 1027 task = get_proc_task(file_inode(file)); 1028 if (!task) 1029 return -ESRCH; 1030 mm = get_task_mm(task); 1031 if (mm) { 1032 struct clear_refs_private cp = { 1033 .type = type, 1034 }; 1035 struct mm_walk clear_refs_walk = { 1036 .pmd_entry = clear_refs_pte_range, 1037 .test_walk = clear_refs_test_walk, 1038 .mm = mm, 1039 .private = &cp, 1040 }; 1041 1042 if (type == CLEAR_REFS_MM_HIWATER_RSS) { 1043 if (down_write_killable(&mm->mmap_sem)) { 1044 count = -EINTR; 1045 goto out_mm; 1046 } 1047 1048 /* 1049 * Writing 5 to /proc/pid/clear_refs resets the peak 1050 * resident set size to this mm's current rss value. 1051 */ 1052 reset_mm_hiwater_rss(mm); 1053 up_write(&mm->mmap_sem); 1054 goto out_mm; 1055 } 1056 1057 down_read(&mm->mmap_sem); 1058 if (type == CLEAR_REFS_SOFT_DIRTY) { 1059 for (vma = mm->mmap; vma; vma = vma->vm_next) { 1060 if (!(vma->vm_flags & VM_SOFTDIRTY)) 1061 continue; 1062 up_read(&mm->mmap_sem); 1063 if (down_write_killable(&mm->mmap_sem)) { 1064 count = -EINTR; 1065 goto out_mm; 1066 } 1067 for (vma = mm->mmap; vma; vma = vma->vm_next) { 1068 vma->vm_flags &= ~VM_SOFTDIRTY; 1069 vma_set_page_prot(vma); 1070 } 1071 downgrade_write(&mm->mmap_sem); 1072 break; 1073 } 1074 mmu_notifier_invalidate_range_start(mm, 0, -1); 1075 } 1076 walk_page_range(0, mm->highest_vm_end, &clear_refs_walk); 1077 if (type == CLEAR_REFS_SOFT_DIRTY) 1078 mmu_notifier_invalidate_range_end(mm, 0, -1); 1079 flush_tlb_mm(mm); 1080 up_read(&mm->mmap_sem); 1081 out_mm: 1082 mmput(mm); 1083 } 1084 put_task_struct(task); 1085 1086 return count; 1087 } 1088 1089 const struct file_operations proc_clear_refs_operations = { 1090 .write = clear_refs_write, 1091 .llseek = noop_llseek, 1092 }; 1093 1094 typedef struct { 1095 u64 pme; 1096 } pagemap_entry_t; 1097 1098 struct pagemapread { 1099 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */ 1100 pagemap_entry_t *buffer; 1101 bool show_pfn; 1102 }; 1103 1104 #define PAGEMAP_WALK_SIZE (PMD_SIZE) 1105 #define PAGEMAP_WALK_MASK (PMD_MASK) 1106 1107 #define PM_ENTRY_BYTES sizeof(pagemap_entry_t) 1108 #define PM_PFRAME_BITS 55 1109 #define PM_PFRAME_MASK GENMASK_ULL(PM_PFRAME_BITS - 1, 0) 1110 #define PM_SOFT_DIRTY BIT_ULL(55) 1111 #define PM_MMAP_EXCLUSIVE BIT_ULL(56) 1112 #define PM_FILE BIT_ULL(61) 1113 #define PM_SWAP BIT_ULL(62) 1114 #define PM_PRESENT BIT_ULL(63) 1115 1116 #define PM_END_OF_BUFFER 1 1117 1118 static inline pagemap_entry_t make_pme(u64 frame, u64 flags) 1119 { 1120 return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags }; 1121 } 1122 1123 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme, 1124 struct pagemapread *pm) 1125 { 1126 pm->buffer[pm->pos++] = *pme; 1127 if (pm->pos >= pm->len) 1128 return PM_END_OF_BUFFER; 1129 return 0; 1130 } 1131 1132 static int pagemap_pte_hole(unsigned long start, unsigned long end, 1133 struct mm_walk *walk) 1134 { 1135 struct pagemapread *pm = walk->private; 1136 unsigned long addr = start; 1137 int err = 0; 1138 1139 while (addr < end) { 1140 struct vm_area_struct *vma = find_vma(walk->mm, addr); 1141 pagemap_entry_t pme = make_pme(0, 0); 1142 /* End of address space hole, which we mark as non-present. */ 1143 unsigned long hole_end; 1144 1145 if (vma) 1146 hole_end = min(end, vma->vm_start); 1147 else 1148 hole_end = end; 1149 1150 for (; addr < hole_end; addr += PAGE_SIZE) { 1151 err = add_to_pagemap(addr, &pme, pm); 1152 if (err) 1153 goto out; 1154 } 1155 1156 if (!vma) 1157 break; 1158 1159 /* Addresses in the VMA. */ 1160 if (vma->vm_flags & VM_SOFTDIRTY) 1161 pme = make_pme(0, PM_SOFT_DIRTY); 1162 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) { 1163 err = add_to_pagemap(addr, &pme, pm); 1164 if (err) 1165 goto out; 1166 } 1167 } 1168 out: 1169 return err; 1170 } 1171 1172 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm, 1173 struct vm_area_struct *vma, unsigned long addr, pte_t pte) 1174 { 1175 u64 frame = 0, flags = 0; 1176 struct page *page = NULL; 1177 1178 if (pte_present(pte)) { 1179 if (pm->show_pfn) 1180 frame = pte_pfn(pte); 1181 flags |= PM_PRESENT; 1182 page = vm_normal_page(vma, addr, pte); 1183 if (pte_soft_dirty(pte)) 1184 flags |= PM_SOFT_DIRTY; 1185 } else if (is_swap_pte(pte)) { 1186 swp_entry_t entry; 1187 if (pte_swp_soft_dirty(pte)) 1188 flags |= PM_SOFT_DIRTY; 1189 entry = pte_to_swp_entry(pte); 1190 frame = swp_type(entry) | 1191 (swp_offset(entry) << MAX_SWAPFILES_SHIFT); 1192 flags |= PM_SWAP; 1193 if (is_migration_entry(entry)) 1194 page = migration_entry_to_page(entry); 1195 } 1196 1197 if (page && !PageAnon(page)) 1198 flags |= PM_FILE; 1199 if (page && page_mapcount(page) == 1) 1200 flags |= PM_MMAP_EXCLUSIVE; 1201 if (vma->vm_flags & VM_SOFTDIRTY) 1202 flags |= PM_SOFT_DIRTY; 1203 1204 return make_pme(frame, flags); 1205 } 1206 1207 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end, 1208 struct mm_walk *walk) 1209 { 1210 struct vm_area_struct *vma = walk->vma; 1211 struct pagemapread *pm = walk->private; 1212 spinlock_t *ptl; 1213 pte_t *pte, *orig_pte; 1214 int err = 0; 1215 1216 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1217 ptl = pmd_trans_huge_lock(pmdp, vma); 1218 if (ptl) { 1219 u64 flags = 0, frame = 0; 1220 pmd_t pmd = *pmdp; 1221 1222 if ((vma->vm_flags & VM_SOFTDIRTY) || pmd_soft_dirty(pmd)) 1223 flags |= PM_SOFT_DIRTY; 1224 1225 /* 1226 * Currently pmd for thp is always present because thp 1227 * can not be swapped-out, migrated, or HWPOISONed 1228 * (split in such cases instead.) 1229 * This if-check is just to prepare for future implementation. 1230 */ 1231 if (pmd_present(pmd)) { 1232 struct page *page = pmd_page(pmd); 1233 1234 if (page_mapcount(page) == 1) 1235 flags |= PM_MMAP_EXCLUSIVE; 1236 1237 flags |= PM_PRESENT; 1238 if (pm->show_pfn) 1239 frame = pmd_pfn(pmd) + 1240 ((addr & ~PMD_MASK) >> PAGE_SHIFT); 1241 } 1242 1243 for (; addr != end; addr += PAGE_SIZE) { 1244 pagemap_entry_t pme = make_pme(frame, flags); 1245 1246 err = add_to_pagemap(addr, &pme, pm); 1247 if (err) 1248 break; 1249 if (pm->show_pfn && (flags & PM_PRESENT)) 1250 frame++; 1251 } 1252 spin_unlock(ptl); 1253 return err; 1254 } 1255 1256 if (pmd_trans_unstable(pmdp)) 1257 return 0; 1258 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ 1259 1260 /* 1261 * We can assume that @vma always points to a valid one and @end never 1262 * goes beyond vma->vm_end. 1263 */ 1264 orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl); 1265 for (; addr < end; pte++, addr += PAGE_SIZE) { 1266 pagemap_entry_t pme; 1267 1268 pme = pte_to_pagemap_entry(pm, vma, addr, *pte); 1269 err = add_to_pagemap(addr, &pme, pm); 1270 if (err) 1271 break; 1272 } 1273 pte_unmap_unlock(orig_pte, ptl); 1274 1275 cond_resched(); 1276 1277 return err; 1278 } 1279 1280 #ifdef CONFIG_HUGETLB_PAGE 1281 /* This function walks within one hugetlb entry in the single call */ 1282 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask, 1283 unsigned long addr, unsigned long end, 1284 struct mm_walk *walk) 1285 { 1286 struct pagemapread *pm = walk->private; 1287 struct vm_area_struct *vma = walk->vma; 1288 u64 flags = 0, frame = 0; 1289 int err = 0; 1290 pte_t pte; 1291 1292 if (vma->vm_flags & VM_SOFTDIRTY) 1293 flags |= PM_SOFT_DIRTY; 1294 1295 pte = huge_ptep_get(ptep); 1296 if (pte_present(pte)) { 1297 struct page *page = pte_page(pte); 1298 1299 if (!PageAnon(page)) 1300 flags |= PM_FILE; 1301 1302 if (page_mapcount(page) == 1) 1303 flags |= PM_MMAP_EXCLUSIVE; 1304 1305 flags |= PM_PRESENT; 1306 if (pm->show_pfn) 1307 frame = pte_pfn(pte) + 1308 ((addr & ~hmask) >> PAGE_SHIFT); 1309 } 1310 1311 for (; addr != end; addr += PAGE_SIZE) { 1312 pagemap_entry_t pme = make_pme(frame, flags); 1313 1314 err = add_to_pagemap(addr, &pme, pm); 1315 if (err) 1316 return err; 1317 if (pm->show_pfn && (flags & PM_PRESENT)) 1318 frame++; 1319 } 1320 1321 cond_resched(); 1322 1323 return err; 1324 } 1325 #endif /* HUGETLB_PAGE */ 1326 1327 /* 1328 * /proc/pid/pagemap - an array mapping virtual pages to pfns 1329 * 1330 * For each page in the address space, this file contains one 64-bit entry 1331 * consisting of the following: 1332 * 1333 * Bits 0-54 page frame number (PFN) if present 1334 * Bits 0-4 swap type if swapped 1335 * Bits 5-54 swap offset if swapped 1336 * Bit 55 pte is soft-dirty (see Documentation/vm/soft-dirty.txt) 1337 * Bit 56 page exclusively mapped 1338 * Bits 57-60 zero 1339 * Bit 61 page is file-page or shared-anon 1340 * Bit 62 page swapped 1341 * Bit 63 page present 1342 * 1343 * If the page is not present but in swap, then the PFN contains an 1344 * encoding of the swap file number and the page's offset into the 1345 * swap. Unmapped pages return a null PFN. This allows determining 1346 * precisely which pages are mapped (or in swap) and comparing mapped 1347 * pages between processes. 1348 * 1349 * Efficient users of this interface will use /proc/pid/maps to 1350 * determine which areas of memory are actually mapped and llseek to 1351 * skip over unmapped regions. 1352 */ 1353 static ssize_t pagemap_read(struct file *file, char __user *buf, 1354 size_t count, loff_t *ppos) 1355 { 1356 struct mm_struct *mm = file->private_data; 1357 struct pagemapread pm; 1358 struct mm_walk pagemap_walk = {}; 1359 unsigned long src; 1360 unsigned long svpfn; 1361 unsigned long start_vaddr; 1362 unsigned long end_vaddr; 1363 int ret = 0, copied = 0; 1364 1365 if (!mm || !mmget_not_zero(mm)) 1366 goto out; 1367 1368 ret = -EINVAL; 1369 /* file position must be aligned */ 1370 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES)) 1371 goto out_mm; 1372 1373 ret = 0; 1374 if (!count) 1375 goto out_mm; 1376 1377 /* do not disclose physical addresses: attack vector */ 1378 pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN); 1379 1380 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT); 1381 pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY); 1382 ret = -ENOMEM; 1383 if (!pm.buffer) 1384 goto out_mm; 1385 1386 pagemap_walk.pmd_entry = pagemap_pmd_range; 1387 pagemap_walk.pte_hole = pagemap_pte_hole; 1388 #ifdef CONFIG_HUGETLB_PAGE 1389 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range; 1390 #endif 1391 pagemap_walk.mm = mm; 1392 pagemap_walk.private = ± 1393 1394 src = *ppos; 1395 svpfn = src / PM_ENTRY_BYTES; 1396 start_vaddr = svpfn << PAGE_SHIFT; 1397 end_vaddr = mm->task_size; 1398 1399 /* watch out for wraparound */ 1400 if (svpfn > mm->task_size >> PAGE_SHIFT) 1401 start_vaddr = end_vaddr; 1402 1403 /* 1404 * The odds are that this will stop walking way 1405 * before end_vaddr, because the length of the 1406 * user buffer is tracked in "pm", and the walk 1407 * will stop when we hit the end of the buffer. 1408 */ 1409 ret = 0; 1410 while (count && (start_vaddr < end_vaddr)) { 1411 int len; 1412 unsigned long end; 1413 1414 pm.pos = 0; 1415 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK; 1416 /* overflow ? */ 1417 if (end < start_vaddr || end > end_vaddr) 1418 end = end_vaddr; 1419 down_read(&mm->mmap_sem); 1420 ret = walk_page_range(start_vaddr, end, &pagemap_walk); 1421 up_read(&mm->mmap_sem); 1422 start_vaddr = end; 1423 1424 len = min(count, PM_ENTRY_BYTES * pm.pos); 1425 if (copy_to_user(buf, pm.buffer, len)) { 1426 ret = -EFAULT; 1427 goto out_free; 1428 } 1429 copied += len; 1430 buf += len; 1431 count -= len; 1432 } 1433 *ppos += copied; 1434 if (!ret || ret == PM_END_OF_BUFFER) 1435 ret = copied; 1436 1437 out_free: 1438 kfree(pm.buffer); 1439 out_mm: 1440 mmput(mm); 1441 out: 1442 return ret; 1443 } 1444 1445 static int pagemap_open(struct inode *inode, struct file *file) 1446 { 1447 struct mm_struct *mm; 1448 1449 mm = proc_mem_open(inode, PTRACE_MODE_READ); 1450 if (IS_ERR(mm)) 1451 return PTR_ERR(mm); 1452 file->private_data = mm; 1453 return 0; 1454 } 1455 1456 static int pagemap_release(struct inode *inode, struct file *file) 1457 { 1458 struct mm_struct *mm = file->private_data; 1459 1460 if (mm) 1461 mmdrop(mm); 1462 return 0; 1463 } 1464 1465 const struct file_operations proc_pagemap_operations = { 1466 .llseek = mem_lseek, /* borrow this */ 1467 .read = pagemap_read, 1468 .open = pagemap_open, 1469 .release = pagemap_release, 1470 }; 1471 #endif /* CONFIG_PROC_PAGE_MONITOR */ 1472 1473 #ifdef CONFIG_NUMA 1474 1475 struct numa_maps { 1476 unsigned long pages; 1477 unsigned long anon; 1478 unsigned long active; 1479 unsigned long writeback; 1480 unsigned long mapcount_max; 1481 unsigned long dirty; 1482 unsigned long swapcache; 1483 unsigned long node[MAX_NUMNODES]; 1484 }; 1485 1486 struct numa_maps_private { 1487 struct proc_maps_private proc_maps; 1488 struct numa_maps md; 1489 }; 1490 1491 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty, 1492 unsigned long nr_pages) 1493 { 1494 int count = page_mapcount(page); 1495 1496 md->pages += nr_pages; 1497 if (pte_dirty || PageDirty(page)) 1498 md->dirty += nr_pages; 1499 1500 if (PageSwapCache(page)) 1501 md->swapcache += nr_pages; 1502 1503 if (PageActive(page) || PageUnevictable(page)) 1504 md->active += nr_pages; 1505 1506 if (PageWriteback(page)) 1507 md->writeback += nr_pages; 1508 1509 if (PageAnon(page)) 1510 md->anon += nr_pages; 1511 1512 if (count > md->mapcount_max) 1513 md->mapcount_max = count; 1514 1515 md->node[page_to_nid(page)] += nr_pages; 1516 } 1517 1518 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma, 1519 unsigned long addr) 1520 { 1521 struct page *page; 1522 int nid; 1523 1524 if (!pte_present(pte)) 1525 return NULL; 1526 1527 page = vm_normal_page(vma, addr, pte); 1528 if (!page) 1529 return NULL; 1530 1531 if (PageReserved(page)) 1532 return NULL; 1533 1534 nid = page_to_nid(page); 1535 if (!node_isset(nid, node_states[N_MEMORY])) 1536 return NULL; 1537 1538 return page; 1539 } 1540 1541 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1542 static struct page *can_gather_numa_stats_pmd(pmd_t pmd, 1543 struct vm_area_struct *vma, 1544 unsigned long addr) 1545 { 1546 struct page *page; 1547 int nid; 1548 1549 if (!pmd_present(pmd)) 1550 return NULL; 1551 1552 page = vm_normal_page_pmd(vma, addr, pmd); 1553 if (!page) 1554 return NULL; 1555 1556 if (PageReserved(page)) 1557 return NULL; 1558 1559 nid = page_to_nid(page); 1560 if (!node_isset(nid, node_states[N_MEMORY])) 1561 return NULL; 1562 1563 return page; 1564 } 1565 #endif 1566 1567 static int gather_pte_stats(pmd_t *pmd, unsigned long addr, 1568 unsigned long end, struct mm_walk *walk) 1569 { 1570 struct numa_maps *md = walk->private; 1571 struct vm_area_struct *vma = walk->vma; 1572 spinlock_t *ptl; 1573 pte_t *orig_pte; 1574 pte_t *pte; 1575 1576 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1577 ptl = pmd_trans_huge_lock(pmd, vma); 1578 if (ptl) { 1579 struct page *page; 1580 1581 page = can_gather_numa_stats_pmd(*pmd, vma, addr); 1582 if (page) 1583 gather_stats(page, md, pmd_dirty(*pmd), 1584 HPAGE_PMD_SIZE/PAGE_SIZE); 1585 spin_unlock(ptl); 1586 return 0; 1587 } 1588 1589 if (pmd_trans_unstable(pmd)) 1590 return 0; 1591 #endif 1592 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl); 1593 do { 1594 struct page *page = can_gather_numa_stats(*pte, vma, addr); 1595 if (!page) 1596 continue; 1597 gather_stats(page, md, pte_dirty(*pte), 1); 1598 1599 } while (pte++, addr += PAGE_SIZE, addr != end); 1600 pte_unmap_unlock(orig_pte, ptl); 1601 cond_resched(); 1602 return 0; 1603 } 1604 #ifdef CONFIG_HUGETLB_PAGE 1605 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask, 1606 unsigned long addr, unsigned long end, struct mm_walk *walk) 1607 { 1608 pte_t huge_pte = huge_ptep_get(pte); 1609 struct numa_maps *md; 1610 struct page *page; 1611 1612 if (!pte_present(huge_pte)) 1613 return 0; 1614 1615 page = pte_page(huge_pte); 1616 if (!page) 1617 return 0; 1618 1619 md = walk->private; 1620 gather_stats(page, md, pte_dirty(huge_pte), 1); 1621 return 0; 1622 } 1623 1624 #else 1625 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask, 1626 unsigned long addr, unsigned long end, struct mm_walk *walk) 1627 { 1628 return 0; 1629 } 1630 #endif 1631 1632 /* 1633 * Display pages allocated per node and memory policy via /proc. 1634 */ 1635 static int show_numa_map(struct seq_file *m, void *v, int is_pid) 1636 { 1637 struct numa_maps_private *numa_priv = m->private; 1638 struct proc_maps_private *proc_priv = &numa_priv->proc_maps; 1639 struct vm_area_struct *vma = v; 1640 struct numa_maps *md = &numa_priv->md; 1641 struct file *file = vma->vm_file; 1642 struct mm_struct *mm = vma->vm_mm; 1643 struct mm_walk walk = { 1644 .hugetlb_entry = gather_hugetlb_stats, 1645 .pmd_entry = gather_pte_stats, 1646 .private = md, 1647 .mm = mm, 1648 }; 1649 struct mempolicy *pol; 1650 char buffer[64]; 1651 int nid; 1652 1653 if (!mm) 1654 return 0; 1655 1656 /* Ensure we start with an empty set of numa_maps statistics. */ 1657 memset(md, 0, sizeof(*md)); 1658 1659 pol = __get_vma_policy(vma, vma->vm_start); 1660 if (pol) { 1661 mpol_to_str(buffer, sizeof(buffer), pol); 1662 mpol_cond_put(pol); 1663 } else { 1664 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy); 1665 } 1666 1667 seq_printf(m, "%08lx %s", vma->vm_start, buffer); 1668 1669 if (file) { 1670 seq_puts(m, " file="); 1671 seq_file_path(m, file, "\n\t= "); 1672 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) { 1673 seq_puts(m, " heap"); 1674 } else if (is_stack(proc_priv, vma)) { 1675 seq_puts(m, " stack"); 1676 } 1677 1678 if (is_vm_hugetlb_page(vma)) 1679 seq_puts(m, " huge"); 1680 1681 /* mmap_sem is held by m_start */ 1682 walk_page_vma(vma, &walk); 1683 1684 if (!md->pages) 1685 goto out; 1686 1687 if (md->anon) 1688 seq_printf(m, " anon=%lu", md->anon); 1689 1690 if (md->dirty) 1691 seq_printf(m, " dirty=%lu", md->dirty); 1692 1693 if (md->pages != md->anon && md->pages != md->dirty) 1694 seq_printf(m, " mapped=%lu", md->pages); 1695 1696 if (md->mapcount_max > 1) 1697 seq_printf(m, " mapmax=%lu", md->mapcount_max); 1698 1699 if (md->swapcache) 1700 seq_printf(m, " swapcache=%lu", md->swapcache); 1701 1702 if (md->active < md->pages && !is_vm_hugetlb_page(vma)) 1703 seq_printf(m, " active=%lu", md->active); 1704 1705 if (md->writeback) 1706 seq_printf(m, " writeback=%lu", md->writeback); 1707 1708 for_each_node_state(nid, N_MEMORY) 1709 if (md->node[nid]) 1710 seq_printf(m, " N%d=%lu", nid, md->node[nid]); 1711 1712 seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10); 1713 out: 1714 seq_putc(m, '\n'); 1715 m_cache_vma(m, vma); 1716 return 0; 1717 } 1718 1719 static int show_pid_numa_map(struct seq_file *m, void *v) 1720 { 1721 return show_numa_map(m, v, 1); 1722 } 1723 1724 static int show_tid_numa_map(struct seq_file *m, void *v) 1725 { 1726 return show_numa_map(m, v, 0); 1727 } 1728 1729 static const struct seq_operations proc_pid_numa_maps_op = { 1730 .start = m_start, 1731 .next = m_next, 1732 .stop = m_stop, 1733 .show = show_pid_numa_map, 1734 }; 1735 1736 static const struct seq_operations proc_tid_numa_maps_op = { 1737 .start = m_start, 1738 .next = m_next, 1739 .stop = m_stop, 1740 .show = show_tid_numa_map, 1741 }; 1742 1743 static int numa_maps_open(struct inode *inode, struct file *file, 1744 const struct seq_operations *ops) 1745 { 1746 return proc_maps_open(inode, file, ops, 1747 sizeof(struct numa_maps_private)); 1748 } 1749 1750 static int pid_numa_maps_open(struct inode *inode, struct file *file) 1751 { 1752 return numa_maps_open(inode, file, &proc_pid_numa_maps_op); 1753 } 1754 1755 static int tid_numa_maps_open(struct inode *inode, struct file *file) 1756 { 1757 return numa_maps_open(inode, file, &proc_tid_numa_maps_op); 1758 } 1759 1760 const struct file_operations proc_pid_numa_maps_operations = { 1761 .open = pid_numa_maps_open, 1762 .read = seq_read, 1763 .llseek = seq_lseek, 1764 .release = proc_map_release, 1765 }; 1766 1767 const struct file_operations proc_tid_numa_maps_operations = { 1768 .open = tid_numa_maps_open, 1769 .read = seq_read, 1770 .llseek = seq_lseek, 1771 .release = proc_map_release, 1772 }; 1773 #endif /* CONFIG_NUMA */ 1774