1 // SPDX-License-Identifier: GPL-2.0 2 #include <linux/mm.h> 3 #include <linux/vmacache.h> 4 #include <linux/hugetlb.h> 5 #include <linux/huge_mm.h> 6 #include <linux/mount.h> 7 #include <linux/seq_file.h> 8 #include <linux/highmem.h> 9 #include <linux/ptrace.h> 10 #include <linux/slab.h> 11 #include <linux/pagemap.h> 12 #include <linux/mempolicy.h> 13 #include <linux/rmap.h> 14 #include <linux/swap.h> 15 #include <linux/sched/mm.h> 16 #include <linux/swapops.h> 17 #include <linux/mmu_notifier.h> 18 #include <linux/page_idle.h> 19 #include <linux/shmem_fs.h> 20 #include <linux/uaccess.h> 21 22 #include <asm/elf.h> 23 #include <asm/tlb.h> 24 #include <asm/tlbflush.h> 25 #include "internal.h" 26 27 void task_mem(struct seq_file *m, struct mm_struct *mm) 28 { 29 unsigned long text, lib, swap, anon, file, shmem; 30 unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss; 31 32 anon = get_mm_counter(mm, MM_ANONPAGES); 33 file = get_mm_counter(mm, MM_FILEPAGES); 34 shmem = get_mm_counter(mm, MM_SHMEMPAGES); 35 36 /* 37 * Note: to minimize their overhead, mm maintains hiwater_vm and 38 * hiwater_rss only when about to *lower* total_vm or rss. Any 39 * collector of these hiwater stats must therefore get total_vm 40 * and rss too, which will usually be the higher. Barriers? not 41 * worth the effort, such snapshots can always be inconsistent. 42 */ 43 hiwater_vm = total_vm = mm->total_vm; 44 if (hiwater_vm < mm->hiwater_vm) 45 hiwater_vm = mm->hiwater_vm; 46 hiwater_rss = total_rss = anon + file + shmem; 47 if (hiwater_rss < mm->hiwater_rss) 48 hiwater_rss = mm->hiwater_rss; 49 50 text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10; 51 lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text; 52 swap = get_mm_counter(mm, MM_SWAPENTS); 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 "VmSwap:\t%8lu kB\n", 69 hiwater_vm << (PAGE_SHIFT-10), 70 total_vm << (PAGE_SHIFT-10), 71 mm->locked_vm << (PAGE_SHIFT-10), 72 mm->pinned_vm << (PAGE_SHIFT-10), 73 hiwater_rss << (PAGE_SHIFT-10), 74 total_rss << (PAGE_SHIFT-10), 75 anon << (PAGE_SHIFT-10), 76 file << (PAGE_SHIFT-10), 77 shmem << (PAGE_SHIFT-10), 78 mm->data_vm << (PAGE_SHIFT-10), 79 mm->stack_vm << (PAGE_SHIFT-10), text, lib, 80 mm_pgtables_bytes(mm) >> 10, 81 swap << (PAGE_SHIFT-10)); 82 hugetlb_report_usage(m, mm); 83 } 84 85 unsigned long task_vsize(struct mm_struct *mm) 86 { 87 return PAGE_SIZE * mm->total_vm; 88 } 89 90 unsigned long task_statm(struct mm_struct *mm, 91 unsigned long *shared, unsigned long *text, 92 unsigned long *data, unsigned long *resident) 93 { 94 *shared = get_mm_counter(mm, MM_FILEPAGES) + 95 get_mm_counter(mm, MM_SHMEMPAGES); 96 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) 97 >> PAGE_SHIFT; 98 *data = mm->data_vm + mm->stack_vm; 99 *resident = *shared + get_mm_counter(mm, MM_ANONPAGES); 100 return mm->total_vm; 101 } 102 103 #ifdef CONFIG_NUMA 104 /* 105 * Save get_task_policy() for show_numa_map(). 106 */ 107 static void hold_task_mempolicy(struct proc_maps_private *priv) 108 { 109 struct task_struct *task = priv->task; 110 111 task_lock(task); 112 priv->task_mempolicy = get_task_policy(task); 113 mpol_get(priv->task_mempolicy); 114 task_unlock(task); 115 } 116 static void release_task_mempolicy(struct proc_maps_private *priv) 117 { 118 mpol_put(priv->task_mempolicy); 119 } 120 #else 121 static void hold_task_mempolicy(struct proc_maps_private *priv) 122 { 123 } 124 static void release_task_mempolicy(struct proc_maps_private *priv) 125 { 126 } 127 #endif 128 129 static void vma_stop(struct proc_maps_private *priv) 130 { 131 struct mm_struct *mm = priv->mm; 132 133 release_task_mempolicy(priv); 134 up_read(&mm->mmap_sem); 135 mmput(mm); 136 } 137 138 static struct vm_area_struct * 139 m_next_vma(struct proc_maps_private *priv, struct vm_area_struct *vma) 140 { 141 if (vma == priv->tail_vma) 142 return NULL; 143 return vma->vm_next ?: priv->tail_vma; 144 } 145 146 static void m_cache_vma(struct seq_file *m, struct vm_area_struct *vma) 147 { 148 if (m->count < m->size) /* vma is copied successfully */ 149 m->version = m_next_vma(m->private, vma) ? vma->vm_end : -1UL; 150 } 151 152 static void *m_start(struct seq_file *m, loff_t *ppos) 153 { 154 struct proc_maps_private *priv = m->private; 155 unsigned long last_addr = m->version; 156 struct mm_struct *mm; 157 struct vm_area_struct *vma; 158 unsigned int pos = *ppos; 159 160 /* See m_cache_vma(). Zero at the start or after lseek. */ 161 if (last_addr == -1UL) 162 return NULL; 163 164 priv->task = get_proc_task(priv->inode); 165 if (!priv->task) 166 return ERR_PTR(-ESRCH); 167 168 mm = priv->mm; 169 if (!mm || !mmget_not_zero(mm)) 170 return NULL; 171 172 down_read(&mm->mmap_sem); 173 hold_task_mempolicy(priv); 174 priv->tail_vma = get_gate_vma(mm); 175 176 if (last_addr) { 177 vma = find_vma(mm, last_addr - 1); 178 if (vma && vma->vm_start <= last_addr) 179 vma = m_next_vma(priv, vma); 180 if (vma) 181 return vma; 182 } 183 184 m->version = 0; 185 if (pos < mm->map_count) { 186 for (vma = mm->mmap; pos; pos--) { 187 m->version = vma->vm_start; 188 vma = vma->vm_next; 189 } 190 return vma; 191 } 192 193 /* we do not bother to update m->version in this case */ 194 if (pos == mm->map_count && priv->tail_vma) 195 return priv->tail_vma; 196 197 vma_stop(priv); 198 return NULL; 199 } 200 201 static void *m_next(struct seq_file *m, void *v, loff_t *pos) 202 { 203 struct proc_maps_private *priv = m->private; 204 struct vm_area_struct *next; 205 206 (*pos)++; 207 next = m_next_vma(priv, v); 208 if (!next) 209 vma_stop(priv); 210 return next; 211 } 212 213 static void m_stop(struct seq_file *m, void *v) 214 { 215 struct proc_maps_private *priv = m->private; 216 217 if (!IS_ERR_OR_NULL(v)) 218 vma_stop(priv); 219 if (priv->task) { 220 put_task_struct(priv->task); 221 priv->task = NULL; 222 } 223 } 224 225 static int proc_maps_open(struct inode *inode, struct file *file, 226 const struct seq_operations *ops, int psize) 227 { 228 struct proc_maps_private *priv = __seq_open_private(file, ops, psize); 229 230 if (!priv) 231 return -ENOMEM; 232 233 priv->inode = inode; 234 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ); 235 if (IS_ERR(priv->mm)) { 236 int err = PTR_ERR(priv->mm); 237 238 seq_release_private(inode, file); 239 return err; 240 } 241 242 return 0; 243 } 244 245 static int proc_map_release(struct inode *inode, struct file *file) 246 { 247 struct seq_file *seq = file->private_data; 248 struct proc_maps_private *priv = seq->private; 249 250 if (priv->mm) 251 mmdrop(priv->mm); 252 253 kfree(priv->rollup); 254 return seq_release_private(inode, file); 255 } 256 257 static int do_maps_open(struct inode *inode, struct file *file, 258 const struct seq_operations *ops) 259 { 260 return proc_maps_open(inode, file, ops, 261 sizeof(struct proc_maps_private)); 262 } 263 264 /* 265 * Indicate if the VMA is a stack for the given task; for 266 * /proc/PID/maps that is the stack of the main task. 267 */ 268 static int is_stack(struct vm_area_struct *vma) 269 { 270 /* 271 * We make no effort to guess what a given thread considers to be 272 * its "stack". It's not even well-defined for programs written 273 * languages like Go. 274 */ 275 return vma->vm_start <= vma->vm_mm->start_stack && 276 vma->vm_end >= vma->vm_mm->start_stack; 277 } 278 279 static void show_vma_header_prefix(struct seq_file *m, 280 unsigned long start, unsigned long end, 281 vm_flags_t flags, unsigned long long pgoff, 282 dev_t dev, unsigned long ino) 283 { 284 seq_setwidth(m, 25 + sizeof(void *) * 6 - 1); 285 seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu ", 286 start, 287 end, 288 flags & VM_READ ? 'r' : '-', 289 flags & VM_WRITE ? 'w' : '-', 290 flags & VM_EXEC ? 'x' : '-', 291 flags & VM_MAYSHARE ? 's' : 'p', 292 pgoff, 293 MAJOR(dev), MINOR(dev), ino); 294 } 295 296 static void 297 show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid) 298 { 299 struct mm_struct *mm = vma->vm_mm; 300 struct file *file = vma->vm_file; 301 vm_flags_t flags = vma->vm_flags; 302 unsigned long ino = 0; 303 unsigned long long pgoff = 0; 304 unsigned long start, end; 305 dev_t dev = 0; 306 const char *name = NULL; 307 308 if (file) { 309 struct inode *inode = file_inode(vma->vm_file); 310 dev = inode->i_sb->s_dev; 311 ino = inode->i_ino; 312 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT; 313 } 314 315 start = vma->vm_start; 316 end = vma->vm_end; 317 show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino); 318 319 /* 320 * Print the dentry name for named mappings, and a 321 * special [heap] marker for the heap: 322 */ 323 if (file) { 324 seq_pad(m, ' '); 325 seq_file_path(m, file, "\n"); 326 goto done; 327 } 328 329 if (vma->vm_ops && vma->vm_ops->name) { 330 name = vma->vm_ops->name(vma); 331 if (name) 332 goto done; 333 } 334 335 name = arch_vma_name(vma); 336 if (!name) { 337 if (!mm) { 338 name = "[vdso]"; 339 goto done; 340 } 341 342 if (vma->vm_start <= mm->brk && 343 vma->vm_end >= mm->start_brk) { 344 name = "[heap]"; 345 goto done; 346 } 347 348 if (is_stack(vma)) 349 name = "[stack]"; 350 } 351 352 done: 353 if (name) { 354 seq_pad(m, ' '); 355 seq_puts(m, name); 356 } 357 seq_putc(m, '\n'); 358 } 359 360 static int show_map(struct seq_file *m, void *v, int is_pid) 361 { 362 show_map_vma(m, v, is_pid); 363 m_cache_vma(m, v); 364 return 0; 365 } 366 367 static int show_pid_map(struct seq_file *m, void *v) 368 { 369 return show_map(m, v, 1); 370 } 371 372 static int show_tid_map(struct seq_file *m, void *v) 373 { 374 return show_map(m, v, 0); 375 } 376 377 static const struct seq_operations proc_pid_maps_op = { 378 .start = m_start, 379 .next = m_next, 380 .stop = m_stop, 381 .show = show_pid_map 382 }; 383 384 static const struct seq_operations proc_tid_maps_op = { 385 .start = m_start, 386 .next = m_next, 387 .stop = m_stop, 388 .show = show_tid_map 389 }; 390 391 static int pid_maps_open(struct inode *inode, struct file *file) 392 { 393 return do_maps_open(inode, file, &proc_pid_maps_op); 394 } 395 396 static int tid_maps_open(struct inode *inode, struct file *file) 397 { 398 return do_maps_open(inode, file, &proc_tid_maps_op); 399 } 400 401 const struct file_operations proc_pid_maps_operations = { 402 .open = pid_maps_open, 403 .read = seq_read, 404 .llseek = seq_lseek, 405 .release = proc_map_release, 406 }; 407 408 const struct file_operations proc_tid_maps_operations = { 409 .open = tid_maps_open, 410 .read = seq_read, 411 .llseek = seq_lseek, 412 .release = proc_map_release, 413 }; 414 415 /* 416 * Proportional Set Size(PSS): my share of RSS. 417 * 418 * PSS of a process is the count of pages it has in memory, where each 419 * page is divided by the number of processes sharing it. So if a 420 * process has 1000 pages all to itself, and 1000 shared with one other 421 * process, its PSS will be 1500. 422 * 423 * To keep (accumulated) division errors low, we adopt a 64bit 424 * fixed-point pss counter to minimize division errors. So (pss >> 425 * PSS_SHIFT) would be the real byte count. 426 * 427 * A shift of 12 before division means (assuming 4K page size): 428 * - 1M 3-user-pages add up to 8KB errors; 429 * - supports mapcount up to 2^24, or 16M; 430 * - supports PSS up to 2^52 bytes, or 4PB. 431 */ 432 #define PSS_SHIFT 12 433 434 #ifdef CONFIG_PROC_PAGE_MONITOR 435 struct mem_size_stats { 436 bool first; 437 unsigned long resident; 438 unsigned long shared_clean; 439 unsigned long shared_dirty; 440 unsigned long private_clean; 441 unsigned long private_dirty; 442 unsigned long referenced; 443 unsigned long anonymous; 444 unsigned long lazyfree; 445 unsigned long anonymous_thp; 446 unsigned long shmem_thp; 447 unsigned long swap; 448 unsigned long shared_hugetlb; 449 unsigned long private_hugetlb; 450 unsigned long first_vma_start; 451 u64 pss; 452 u64 pss_locked; 453 u64 swap_pss; 454 bool check_shmem_swap; 455 }; 456 457 static void smaps_account(struct mem_size_stats *mss, struct page *page, 458 bool compound, bool young, bool dirty) 459 { 460 int i, nr = compound ? 1 << compound_order(page) : 1; 461 unsigned long size = nr * PAGE_SIZE; 462 463 if (PageAnon(page)) { 464 mss->anonymous += size; 465 if (!PageSwapBacked(page) && !dirty && !PageDirty(page)) 466 mss->lazyfree += size; 467 } 468 469 mss->resident += size; 470 /* Accumulate the size in pages that have been accessed. */ 471 if (young || page_is_young(page) || PageReferenced(page)) 472 mss->referenced += size; 473 474 /* 475 * page_count(page) == 1 guarantees the page is mapped exactly once. 476 * If any subpage of the compound page mapped with PTE it would elevate 477 * page_count(). 478 */ 479 if (page_count(page) == 1) { 480 if (dirty || PageDirty(page)) 481 mss->private_dirty += size; 482 else 483 mss->private_clean += size; 484 mss->pss += (u64)size << PSS_SHIFT; 485 return; 486 } 487 488 for (i = 0; i < nr; i++, page++) { 489 int mapcount = page_mapcount(page); 490 491 if (mapcount >= 2) { 492 if (dirty || PageDirty(page)) 493 mss->shared_dirty += PAGE_SIZE; 494 else 495 mss->shared_clean += PAGE_SIZE; 496 mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount; 497 } else { 498 if (dirty || PageDirty(page)) 499 mss->private_dirty += PAGE_SIZE; 500 else 501 mss->private_clean += PAGE_SIZE; 502 mss->pss += PAGE_SIZE << PSS_SHIFT; 503 } 504 } 505 } 506 507 #ifdef CONFIG_SHMEM 508 static int smaps_pte_hole(unsigned long addr, unsigned long end, 509 struct mm_walk *walk) 510 { 511 struct mem_size_stats *mss = walk->private; 512 513 mss->swap += shmem_partial_swap_usage( 514 walk->vma->vm_file->f_mapping, addr, end); 515 516 return 0; 517 } 518 #endif 519 520 static void smaps_pte_entry(pte_t *pte, unsigned long addr, 521 struct mm_walk *walk) 522 { 523 struct mem_size_stats *mss = walk->private; 524 struct vm_area_struct *vma = walk->vma; 525 struct page *page = NULL; 526 527 if (pte_present(*pte)) { 528 page = vm_normal_page(vma, addr, *pte); 529 } else if (is_swap_pte(*pte)) { 530 swp_entry_t swpent = pte_to_swp_entry(*pte); 531 532 if (!non_swap_entry(swpent)) { 533 int mapcount; 534 535 mss->swap += PAGE_SIZE; 536 mapcount = swp_swapcount(swpent); 537 if (mapcount >= 2) { 538 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT; 539 540 do_div(pss_delta, mapcount); 541 mss->swap_pss += pss_delta; 542 } else { 543 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT; 544 } 545 } else if (is_migration_entry(swpent)) 546 page = migration_entry_to_page(swpent); 547 else if (is_device_private_entry(swpent)) 548 page = device_private_entry_to_page(swpent); 549 } else if (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap 550 && pte_none(*pte))) { 551 page = find_get_entry(vma->vm_file->f_mapping, 552 linear_page_index(vma, addr)); 553 if (!page) 554 return; 555 556 if (radix_tree_exceptional_entry(page)) 557 mss->swap += PAGE_SIZE; 558 else 559 put_page(page); 560 561 return; 562 } 563 564 if (!page) 565 return; 566 567 smaps_account(mss, page, false, pte_young(*pte), pte_dirty(*pte)); 568 } 569 570 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 571 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr, 572 struct mm_walk *walk) 573 { 574 struct mem_size_stats *mss = walk->private; 575 struct vm_area_struct *vma = walk->vma; 576 struct page *page; 577 578 /* FOLL_DUMP will return -EFAULT on huge zero page */ 579 page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP); 580 if (IS_ERR_OR_NULL(page)) 581 return; 582 if (PageAnon(page)) 583 mss->anonymous_thp += HPAGE_PMD_SIZE; 584 else if (PageSwapBacked(page)) 585 mss->shmem_thp += HPAGE_PMD_SIZE; 586 else if (is_zone_device_page(page)) 587 /* pass */; 588 else 589 VM_BUG_ON_PAGE(1, page); 590 smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd)); 591 } 592 #else 593 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr, 594 struct mm_walk *walk) 595 { 596 } 597 #endif 598 599 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, 600 struct mm_walk *walk) 601 { 602 struct vm_area_struct *vma = walk->vma; 603 pte_t *pte; 604 spinlock_t *ptl; 605 606 ptl = pmd_trans_huge_lock(pmd, vma); 607 if (ptl) { 608 if (pmd_present(*pmd)) 609 smaps_pmd_entry(pmd, addr, walk); 610 spin_unlock(ptl); 611 goto out; 612 } 613 614 if (pmd_trans_unstable(pmd)) 615 goto out; 616 /* 617 * The mmap_sem held all the way back in m_start() is what 618 * keeps khugepaged out of here and from collapsing things 619 * in here. 620 */ 621 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); 622 for (; addr != end; pte++, addr += PAGE_SIZE) 623 smaps_pte_entry(pte, addr, walk); 624 pte_unmap_unlock(pte - 1, ptl); 625 out: 626 cond_resched(); 627 return 0; 628 } 629 630 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma) 631 { 632 /* 633 * Don't forget to update Documentation/ on changes. 634 */ 635 static const char mnemonics[BITS_PER_LONG][2] = { 636 /* 637 * In case if we meet a flag we don't know about. 638 */ 639 [0 ... (BITS_PER_LONG-1)] = "??", 640 641 [ilog2(VM_READ)] = "rd", 642 [ilog2(VM_WRITE)] = "wr", 643 [ilog2(VM_EXEC)] = "ex", 644 [ilog2(VM_SHARED)] = "sh", 645 [ilog2(VM_MAYREAD)] = "mr", 646 [ilog2(VM_MAYWRITE)] = "mw", 647 [ilog2(VM_MAYEXEC)] = "me", 648 [ilog2(VM_MAYSHARE)] = "ms", 649 [ilog2(VM_GROWSDOWN)] = "gd", 650 [ilog2(VM_PFNMAP)] = "pf", 651 [ilog2(VM_DENYWRITE)] = "dw", 652 #ifdef CONFIG_X86_INTEL_MPX 653 [ilog2(VM_MPX)] = "mp", 654 #endif 655 [ilog2(VM_LOCKED)] = "lo", 656 [ilog2(VM_IO)] = "io", 657 [ilog2(VM_SEQ_READ)] = "sr", 658 [ilog2(VM_RAND_READ)] = "rr", 659 [ilog2(VM_DONTCOPY)] = "dc", 660 [ilog2(VM_DONTEXPAND)] = "de", 661 [ilog2(VM_ACCOUNT)] = "ac", 662 [ilog2(VM_NORESERVE)] = "nr", 663 [ilog2(VM_HUGETLB)] = "ht", 664 [ilog2(VM_ARCH_1)] = "ar", 665 [ilog2(VM_WIPEONFORK)] = "wf", 666 [ilog2(VM_DONTDUMP)] = "dd", 667 #ifdef CONFIG_MEM_SOFT_DIRTY 668 [ilog2(VM_SOFTDIRTY)] = "sd", 669 #endif 670 [ilog2(VM_MIXEDMAP)] = "mm", 671 [ilog2(VM_HUGEPAGE)] = "hg", 672 [ilog2(VM_NOHUGEPAGE)] = "nh", 673 [ilog2(VM_MERGEABLE)] = "mg", 674 [ilog2(VM_UFFD_MISSING)]= "um", 675 [ilog2(VM_UFFD_WP)] = "uw", 676 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS 677 /* These come out via ProtectionKey: */ 678 [ilog2(VM_PKEY_BIT0)] = "", 679 [ilog2(VM_PKEY_BIT1)] = "", 680 [ilog2(VM_PKEY_BIT2)] = "", 681 [ilog2(VM_PKEY_BIT3)] = "", 682 #endif 683 }; 684 size_t i; 685 686 seq_puts(m, "VmFlags: "); 687 for (i = 0; i < BITS_PER_LONG; i++) { 688 if (!mnemonics[i][0]) 689 continue; 690 if (vma->vm_flags & (1UL << i)) { 691 seq_printf(m, "%c%c ", 692 mnemonics[i][0], mnemonics[i][1]); 693 } 694 } 695 seq_putc(m, '\n'); 696 } 697 698 #ifdef CONFIG_HUGETLB_PAGE 699 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask, 700 unsigned long addr, unsigned long end, 701 struct mm_walk *walk) 702 { 703 struct mem_size_stats *mss = walk->private; 704 struct vm_area_struct *vma = walk->vma; 705 struct page *page = NULL; 706 707 if (pte_present(*pte)) { 708 page = vm_normal_page(vma, addr, *pte); 709 } else if (is_swap_pte(*pte)) { 710 swp_entry_t swpent = pte_to_swp_entry(*pte); 711 712 if (is_migration_entry(swpent)) 713 page = migration_entry_to_page(swpent); 714 else if (is_device_private_entry(swpent)) 715 page = device_private_entry_to_page(swpent); 716 } 717 if (page) { 718 int mapcount = page_mapcount(page); 719 720 if (mapcount >= 2) 721 mss->shared_hugetlb += huge_page_size(hstate_vma(vma)); 722 else 723 mss->private_hugetlb += huge_page_size(hstate_vma(vma)); 724 } 725 return 0; 726 } 727 #endif /* HUGETLB_PAGE */ 728 729 void __weak arch_show_smap(struct seq_file *m, struct vm_area_struct *vma) 730 { 731 } 732 733 static int show_smap(struct seq_file *m, void *v, int is_pid) 734 { 735 struct proc_maps_private *priv = m->private; 736 struct vm_area_struct *vma = v; 737 struct mem_size_stats mss_stack; 738 struct mem_size_stats *mss; 739 struct mm_walk smaps_walk = { 740 .pmd_entry = smaps_pte_range, 741 #ifdef CONFIG_HUGETLB_PAGE 742 .hugetlb_entry = smaps_hugetlb_range, 743 #endif 744 .mm = vma->vm_mm, 745 }; 746 int ret = 0; 747 bool rollup_mode; 748 bool last_vma; 749 750 if (priv->rollup) { 751 rollup_mode = true; 752 mss = priv->rollup; 753 if (mss->first) { 754 mss->first_vma_start = vma->vm_start; 755 mss->first = false; 756 } 757 last_vma = !m_next_vma(priv, vma); 758 } else { 759 rollup_mode = false; 760 memset(&mss_stack, 0, sizeof(mss_stack)); 761 mss = &mss_stack; 762 } 763 764 smaps_walk.private = mss; 765 766 #ifdef CONFIG_SHMEM 767 if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) { 768 /* 769 * For shared or readonly shmem mappings we know that all 770 * swapped out pages belong to the shmem object, and we can 771 * obtain the swap value much more efficiently. For private 772 * writable mappings, we might have COW pages that are 773 * not affected by the parent swapped out pages of the shmem 774 * object, so we have to distinguish them during the page walk. 775 * Unless we know that the shmem object (or the part mapped by 776 * our VMA) has no swapped out pages at all. 777 */ 778 unsigned long shmem_swapped = shmem_swap_usage(vma); 779 780 if (!shmem_swapped || (vma->vm_flags & VM_SHARED) || 781 !(vma->vm_flags & VM_WRITE)) { 782 mss->swap = shmem_swapped; 783 } else { 784 mss->check_shmem_swap = true; 785 smaps_walk.pte_hole = smaps_pte_hole; 786 } 787 } 788 #endif 789 790 /* mmap_sem is held in m_start */ 791 walk_page_vma(vma, &smaps_walk); 792 if (vma->vm_flags & VM_LOCKED) 793 mss->pss_locked += mss->pss; 794 795 if (!rollup_mode) { 796 show_map_vma(m, vma, is_pid); 797 } else if (last_vma) { 798 show_vma_header_prefix( 799 m, mss->first_vma_start, vma->vm_end, 0, 0, 0, 0); 800 seq_pad(m, ' '); 801 seq_puts(m, "[rollup]\n"); 802 } else { 803 ret = SEQ_SKIP; 804 } 805 806 if (!rollup_mode) 807 seq_printf(m, 808 "Size: %8lu kB\n" 809 "KernelPageSize: %8lu kB\n" 810 "MMUPageSize: %8lu kB\n", 811 (vma->vm_end - vma->vm_start) >> 10, 812 vma_kernel_pagesize(vma) >> 10, 813 vma_mmu_pagesize(vma) >> 10); 814 815 816 if (!rollup_mode || last_vma) 817 seq_printf(m, 818 "Rss: %8lu kB\n" 819 "Pss: %8lu kB\n" 820 "Shared_Clean: %8lu kB\n" 821 "Shared_Dirty: %8lu kB\n" 822 "Private_Clean: %8lu kB\n" 823 "Private_Dirty: %8lu kB\n" 824 "Referenced: %8lu kB\n" 825 "Anonymous: %8lu kB\n" 826 "LazyFree: %8lu kB\n" 827 "AnonHugePages: %8lu kB\n" 828 "ShmemPmdMapped: %8lu kB\n" 829 "Shared_Hugetlb: %8lu kB\n" 830 "Private_Hugetlb: %7lu kB\n" 831 "Swap: %8lu kB\n" 832 "SwapPss: %8lu kB\n" 833 "Locked: %8lu kB\n", 834 mss->resident >> 10, 835 (unsigned long)(mss->pss >> (10 + PSS_SHIFT)), 836 mss->shared_clean >> 10, 837 mss->shared_dirty >> 10, 838 mss->private_clean >> 10, 839 mss->private_dirty >> 10, 840 mss->referenced >> 10, 841 mss->anonymous >> 10, 842 mss->lazyfree >> 10, 843 mss->anonymous_thp >> 10, 844 mss->shmem_thp >> 10, 845 mss->shared_hugetlb >> 10, 846 mss->private_hugetlb >> 10, 847 mss->swap >> 10, 848 (unsigned long)(mss->swap_pss >> (10 + PSS_SHIFT)), 849 (unsigned long)(mss->pss >> (10 + PSS_SHIFT))); 850 851 if (!rollup_mode) { 852 arch_show_smap(m, vma); 853 show_smap_vma_flags(m, vma); 854 } 855 m_cache_vma(m, vma); 856 return ret; 857 } 858 859 static int show_pid_smap(struct seq_file *m, void *v) 860 { 861 return show_smap(m, v, 1); 862 } 863 864 static int show_tid_smap(struct seq_file *m, void *v) 865 { 866 return show_smap(m, v, 0); 867 } 868 869 static const struct seq_operations proc_pid_smaps_op = { 870 .start = m_start, 871 .next = m_next, 872 .stop = m_stop, 873 .show = show_pid_smap 874 }; 875 876 static const struct seq_operations proc_tid_smaps_op = { 877 .start = m_start, 878 .next = m_next, 879 .stop = m_stop, 880 .show = show_tid_smap 881 }; 882 883 static int pid_smaps_open(struct inode *inode, struct file *file) 884 { 885 return do_maps_open(inode, file, &proc_pid_smaps_op); 886 } 887 888 static int pid_smaps_rollup_open(struct inode *inode, struct file *file) 889 { 890 struct seq_file *seq; 891 struct proc_maps_private *priv; 892 int ret = do_maps_open(inode, file, &proc_pid_smaps_op); 893 894 if (ret < 0) 895 return ret; 896 seq = file->private_data; 897 priv = seq->private; 898 priv->rollup = kzalloc(sizeof(*priv->rollup), GFP_KERNEL); 899 if (!priv->rollup) { 900 proc_map_release(inode, file); 901 return -ENOMEM; 902 } 903 priv->rollup->first = true; 904 return 0; 905 } 906 907 static int tid_smaps_open(struct inode *inode, struct file *file) 908 { 909 return do_maps_open(inode, file, &proc_tid_smaps_op); 910 } 911 912 const struct file_operations proc_pid_smaps_operations = { 913 .open = pid_smaps_open, 914 .read = seq_read, 915 .llseek = seq_lseek, 916 .release = proc_map_release, 917 }; 918 919 const struct file_operations proc_pid_smaps_rollup_operations = { 920 .open = pid_smaps_rollup_open, 921 .read = seq_read, 922 .llseek = seq_lseek, 923 .release = proc_map_release, 924 }; 925 926 const struct file_operations proc_tid_smaps_operations = { 927 .open = tid_smaps_open, 928 .read = seq_read, 929 .llseek = seq_lseek, 930 .release = proc_map_release, 931 }; 932 933 enum clear_refs_types { 934 CLEAR_REFS_ALL = 1, 935 CLEAR_REFS_ANON, 936 CLEAR_REFS_MAPPED, 937 CLEAR_REFS_SOFT_DIRTY, 938 CLEAR_REFS_MM_HIWATER_RSS, 939 CLEAR_REFS_LAST, 940 }; 941 942 struct clear_refs_private { 943 enum clear_refs_types type; 944 }; 945 946 #ifdef CONFIG_MEM_SOFT_DIRTY 947 static inline void clear_soft_dirty(struct vm_area_struct *vma, 948 unsigned long addr, pte_t *pte) 949 { 950 /* 951 * The soft-dirty tracker uses #PF-s to catch writes 952 * to pages, so write-protect the pte as well. See the 953 * Documentation/vm/soft-dirty.txt for full description 954 * of how soft-dirty works. 955 */ 956 pte_t ptent = *pte; 957 958 if (pte_present(ptent)) { 959 ptent = ptep_modify_prot_start(vma->vm_mm, addr, pte); 960 ptent = pte_wrprotect(ptent); 961 ptent = pte_clear_soft_dirty(ptent); 962 ptep_modify_prot_commit(vma->vm_mm, addr, pte, ptent); 963 } else if (is_swap_pte(ptent)) { 964 ptent = pte_swp_clear_soft_dirty(ptent); 965 set_pte_at(vma->vm_mm, addr, pte, ptent); 966 } 967 } 968 #else 969 static inline void clear_soft_dirty(struct vm_area_struct *vma, 970 unsigned long addr, pte_t *pte) 971 { 972 } 973 #endif 974 975 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE) 976 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma, 977 unsigned long addr, pmd_t *pmdp) 978 { 979 pmd_t pmd = *pmdp; 980 981 if (pmd_present(pmd)) { 982 /* See comment in change_huge_pmd() */ 983 pmdp_invalidate(vma, addr, pmdp); 984 if (pmd_dirty(*pmdp)) 985 pmd = pmd_mkdirty(pmd); 986 if (pmd_young(*pmdp)) 987 pmd = pmd_mkyoung(pmd); 988 989 pmd = pmd_wrprotect(pmd); 990 pmd = pmd_clear_soft_dirty(pmd); 991 992 set_pmd_at(vma->vm_mm, addr, pmdp, pmd); 993 } else if (is_migration_entry(pmd_to_swp_entry(pmd))) { 994 pmd = pmd_swp_clear_soft_dirty(pmd); 995 set_pmd_at(vma->vm_mm, addr, pmdp, pmd); 996 } 997 } 998 #else 999 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma, 1000 unsigned long addr, pmd_t *pmdp) 1001 { 1002 } 1003 #endif 1004 1005 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr, 1006 unsigned long end, struct mm_walk *walk) 1007 { 1008 struct clear_refs_private *cp = walk->private; 1009 struct vm_area_struct *vma = walk->vma; 1010 pte_t *pte, ptent; 1011 spinlock_t *ptl; 1012 struct page *page; 1013 1014 ptl = pmd_trans_huge_lock(pmd, vma); 1015 if (ptl) { 1016 if (cp->type == CLEAR_REFS_SOFT_DIRTY) { 1017 clear_soft_dirty_pmd(vma, addr, pmd); 1018 goto out; 1019 } 1020 1021 if (!pmd_present(*pmd)) 1022 goto out; 1023 1024 page = pmd_page(*pmd); 1025 1026 /* Clear accessed and referenced bits. */ 1027 pmdp_test_and_clear_young(vma, addr, pmd); 1028 test_and_clear_page_young(page); 1029 ClearPageReferenced(page); 1030 out: 1031 spin_unlock(ptl); 1032 return 0; 1033 } 1034 1035 if (pmd_trans_unstable(pmd)) 1036 return 0; 1037 1038 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); 1039 for (; addr != end; pte++, addr += PAGE_SIZE) { 1040 ptent = *pte; 1041 1042 if (cp->type == CLEAR_REFS_SOFT_DIRTY) { 1043 clear_soft_dirty(vma, addr, pte); 1044 continue; 1045 } 1046 1047 if (!pte_present(ptent)) 1048 continue; 1049 1050 page = vm_normal_page(vma, addr, ptent); 1051 if (!page) 1052 continue; 1053 1054 /* Clear accessed and referenced bits. */ 1055 ptep_test_and_clear_young(vma, addr, pte); 1056 test_and_clear_page_young(page); 1057 ClearPageReferenced(page); 1058 } 1059 pte_unmap_unlock(pte - 1, ptl); 1060 cond_resched(); 1061 return 0; 1062 } 1063 1064 static int clear_refs_test_walk(unsigned long start, unsigned long end, 1065 struct mm_walk *walk) 1066 { 1067 struct clear_refs_private *cp = walk->private; 1068 struct vm_area_struct *vma = walk->vma; 1069 1070 if (vma->vm_flags & VM_PFNMAP) 1071 return 1; 1072 1073 /* 1074 * Writing 1 to /proc/pid/clear_refs affects all pages. 1075 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages. 1076 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages. 1077 * Writing 4 to /proc/pid/clear_refs affects all pages. 1078 */ 1079 if (cp->type == CLEAR_REFS_ANON && vma->vm_file) 1080 return 1; 1081 if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file) 1082 return 1; 1083 return 0; 1084 } 1085 1086 static ssize_t clear_refs_write(struct file *file, const char __user *buf, 1087 size_t count, loff_t *ppos) 1088 { 1089 struct task_struct *task; 1090 char buffer[PROC_NUMBUF]; 1091 struct mm_struct *mm; 1092 struct vm_area_struct *vma; 1093 enum clear_refs_types type; 1094 struct mmu_gather tlb; 1095 int itype; 1096 int rv; 1097 1098 memset(buffer, 0, sizeof(buffer)); 1099 if (count > sizeof(buffer) - 1) 1100 count = sizeof(buffer) - 1; 1101 if (copy_from_user(buffer, buf, count)) 1102 return -EFAULT; 1103 rv = kstrtoint(strstrip(buffer), 10, &itype); 1104 if (rv < 0) 1105 return rv; 1106 type = (enum clear_refs_types)itype; 1107 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST) 1108 return -EINVAL; 1109 1110 task = get_proc_task(file_inode(file)); 1111 if (!task) 1112 return -ESRCH; 1113 mm = get_task_mm(task); 1114 if (mm) { 1115 struct clear_refs_private cp = { 1116 .type = type, 1117 }; 1118 struct mm_walk clear_refs_walk = { 1119 .pmd_entry = clear_refs_pte_range, 1120 .test_walk = clear_refs_test_walk, 1121 .mm = mm, 1122 .private = &cp, 1123 }; 1124 1125 if (type == CLEAR_REFS_MM_HIWATER_RSS) { 1126 if (down_write_killable(&mm->mmap_sem)) { 1127 count = -EINTR; 1128 goto out_mm; 1129 } 1130 1131 /* 1132 * Writing 5 to /proc/pid/clear_refs resets the peak 1133 * resident set size to this mm's current rss value. 1134 */ 1135 reset_mm_hiwater_rss(mm); 1136 up_write(&mm->mmap_sem); 1137 goto out_mm; 1138 } 1139 1140 down_read(&mm->mmap_sem); 1141 tlb_gather_mmu(&tlb, mm, 0, -1); 1142 if (type == CLEAR_REFS_SOFT_DIRTY) { 1143 for (vma = mm->mmap; vma; vma = vma->vm_next) { 1144 if (!(vma->vm_flags & VM_SOFTDIRTY)) 1145 continue; 1146 up_read(&mm->mmap_sem); 1147 if (down_write_killable(&mm->mmap_sem)) { 1148 count = -EINTR; 1149 goto out_mm; 1150 } 1151 for (vma = mm->mmap; vma; vma = vma->vm_next) { 1152 vma->vm_flags &= ~VM_SOFTDIRTY; 1153 vma_set_page_prot(vma); 1154 } 1155 downgrade_write(&mm->mmap_sem); 1156 break; 1157 } 1158 mmu_notifier_invalidate_range_start(mm, 0, -1); 1159 } 1160 walk_page_range(0, mm->highest_vm_end, &clear_refs_walk); 1161 if (type == CLEAR_REFS_SOFT_DIRTY) 1162 mmu_notifier_invalidate_range_end(mm, 0, -1); 1163 tlb_finish_mmu(&tlb, 0, -1); 1164 up_read(&mm->mmap_sem); 1165 out_mm: 1166 mmput(mm); 1167 } 1168 put_task_struct(task); 1169 1170 return count; 1171 } 1172 1173 const struct file_operations proc_clear_refs_operations = { 1174 .write = clear_refs_write, 1175 .llseek = noop_llseek, 1176 }; 1177 1178 typedef struct { 1179 u64 pme; 1180 } pagemap_entry_t; 1181 1182 struct pagemapread { 1183 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */ 1184 pagemap_entry_t *buffer; 1185 bool show_pfn; 1186 }; 1187 1188 #define PAGEMAP_WALK_SIZE (PMD_SIZE) 1189 #define PAGEMAP_WALK_MASK (PMD_MASK) 1190 1191 #define PM_ENTRY_BYTES sizeof(pagemap_entry_t) 1192 #define PM_PFRAME_BITS 55 1193 #define PM_PFRAME_MASK GENMASK_ULL(PM_PFRAME_BITS - 1, 0) 1194 #define PM_SOFT_DIRTY BIT_ULL(55) 1195 #define PM_MMAP_EXCLUSIVE BIT_ULL(56) 1196 #define PM_FILE BIT_ULL(61) 1197 #define PM_SWAP BIT_ULL(62) 1198 #define PM_PRESENT BIT_ULL(63) 1199 1200 #define PM_END_OF_BUFFER 1 1201 1202 static inline pagemap_entry_t make_pme(u64 frame, u64 flags) 1203 { 1204 return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags }; 1205 } 1206 1207 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme, 1208 struct pagemapread *pm) 1209 { 1210 pm->buffer[pm->pos++] = *pme; 1211 if (pm->pos >= pm->len) 1212 return PM_END_OF_BUFFER; 1213 return 0; 1214 } 1215 1216 static int pagemap_pte_hole(unsigned long start, unsigned long end, 1217 struct mm_walk *walk) 1218 { 1219 struct pagemapread *pm = walk->private; 1220 unsigned long addr = start; 1221 int err = 0; 1222 1223 while (addr < end) { 1224 struct vm_area_struct *vma = find_vma(walk->mm, addr); 1225 pagemap_entry_t pme = make_pme(0, 0); 1226 /* End of address space hole, which we mark as non-present. */ 1227 unsigned long hole_end; 1228 1229 if (vma) 1230 hole_end = min(end, vma->vm_start); 1231 else 1232 hole_end = end; 1233 1234 for (; addr < hole_end; addr += PAGE_SIZE) { 1235 err = add_to_pagemap(addr, &pme, pm); 1236 if (err) 1237 goto out; 1238 } 1239 1240 if (!vma) 1241 break; 1242 1243 /* Addresses in the VMA. */ 1244 if (vma->vm_flags & VM_SOFTDIRTY) 1245 pme = make_pme(0, PM_SOFT_DIRTY); 1246 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) { 1247 err = add_to_pagemap(addr, &pme, pm); 1248 if (err) 1249 goto out; 1250 } 1251 } 1252 out: 1253 return err; 1254 } 1255 1256 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm, 1257 struct vm_area_struct *vma, unsigned long addr, pte_t pte) 1258 { 1259 u64 frame = 0, flags = 0; 1260 struct page *page = NULL; 1261 1262 if (pte_present(pte)) { 1263 if (pm->show_pfn) 1264 frame = pte_pfn(pte); 1265 flags |= PM_PRESENT; 1266 page = _vm_normal_page(vma, addr, pte, true); 1267 if (pte_soft_dirty(pte)) 1268 flags |= PM_SOFT_DIRTY; 1269 } else if (is_swap_pte(pte)) { 1270 swp_entry_t entry; 1271 if (pte_swp_soft_dirty(pte)) 1272 flags |= PM_SOFT_DIRTY; 1273 entry = pte_to_swp_entry(pte); 1274 frame = swp_type(entry) | 1275 (swp_offset(entry) << MAX_SWAPFILES_SHIFT); 1276 flags |= PM_SWAP; 1277 if (is_migration_entry(entry)) 1278 page = migration_entry_to_page(entry); 1279 1280 if (is_device_private_entry(entry)) 1281 page = device_private_entry_to_page(entry); 1282 } 1283 1284 if (page && !PageAnon(page)) 1285 flags |= PM_FILE; 1286 if (page && page_mapcount(page) == 1) 1287 flags |= PM_MMAP_EXCLUSIVE; 1288 if (vma->vm_flags & VM_SOFTDIRTY) 1289 flags |= PM_SOFT_DIRTY; 1290 1291 return make_pme(frame, flags); 1292 } 1293 1294 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end, 1295 struct mm_walk *walk) 1296 { 1297 struct vm_area_struct *vma = walk->vma; 1298 struct pagemapread *pm = walk->private; 1299 spinlock_t *ptl; 1300 pte_t *pte, *orig_pte; 1301 int err = 0; 1302 1303 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1304 ptl = pmd_trans_huge_lock(pmdp, vma); 1305 if (ptl) { 1306 u64 flags = 0, frame = 0; 1307 pmd_t pmd = *pmdp; 1308 struct page *page = NULL; 1309 1310 if (vma->vm_flags & VM_SOFTDIRTY) 1311 flags |= PM_SOFT_DIRTY; 1312 1313 if (pmd_present(pmd)) { 1314 page = pmd_page(pmd); 1315 1316 flags |= PM_PRESENT; 1317 if (pmd_soft_dirty(pmd)) 1318 flags |= PM_SOFT_DIRTY; 1319 if (pm->show_pfn) 1320 frame = pmd_pfn(pmd) + 1321 ((addr & ~PMD_MASK) >> PAGE_SHIFT); 1322 } 1323 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION 1324 else if (is_swap_pmd(pmd)) { 1325 swp_entry_t entry = pmd_to_swp_entry(pmd); 1326 1327 frame = swp_type(entry) | 1328 (swp_offset(entry) << MAX_SWAPFILES_SHIFT); 1329 flags |= PM_SWAP; 1330 if (pmd_swp_soft_dirty(pmd)) 1331 flags |= PM_SOFT_DIRTY; 1332 VM_BUG_ON(!is_pmd_migration_entry(pmd)); 1333 page = migration_entry_to_page(entry); 1334 } 1335 #endif 1336 1337 if (page && page_mapcount(page) == 1) 1338 flags |= PM_MMAP_EXCLUSIVE; 1339 1340 for (; addr != end; addr += PAGE_SIZE) { 1341 pagemap_entry_t pme = make_pme(frame, flags); 1342 1343 err = add_to_pagemap(addr, &pme, pm); 1344 if (err) 1345 break; 1346 if (pm->show_pfn && (flags & PM_PRESENT)) 1347 frame++; 1348 } 1349 spin_unlock(ptl); 1350 return err; 1351 } 1352 1353 if (pmd_trans_unstable(pmdp)) 1354 return 0; 1355 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ 1356 1357 /* 1358 * We can assume that @vma always points to a valid one and @end never 1359 * goes beyond vma->vm_end. 1360 */ 1361 orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl); 1362 for (; addr < end; pte++, addr += PAGE_SIZE) { 1363 pagemap_entry_t pme; 1364 1365 pme = pte_to_pagemap_entry(pm, vma, addr, *pte); 1366 err = add_to_pagemap(addr, &pme, pm); 1367 if (err) 1368 break; 1369 } 1370 pte_unmap_unlock(orig_pte, ptl); 1371 1372 cond_resched(); 1373 1374 return err; 1375 } 1376 1377 #ifdef CONFIG_HUGETLB_PAGE 1378 /* This function walks within one hugetlb entry in the single call */ 1379 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask, 1380 unsigned long addr, unsigned long end, 1381 struct mm_walk *walk) 1382 { 1383 struct pagemapread *pm = walk->private; 1384 struct vm_area_struct *vma = walk->vma; 1385 u64 flags = 0, frame = 0; 1386 int err = 0; 1387 pte_t pte; 1388 1389 if (vma->vm_flags & VM_SOFTDIRTY) 1390 flags |= PM_SOFT_DIRTY; 1391 1392 pte = huge_ptep_get(ptep); 1393 if (pte_present(pte)) { 1394 struct page *page = pte_page(pte); 1395 1396 if (!PageAnon(page)) 1397 flags |= PM_FILE; 1398 1399 if (page_mapcount(page) == 1) 1400 flags |= PM_MMAP_EXCLUSIVE; 1401 1402 flags |= PM_PRESENT; 1403 if (pm->show_pfn) 1404 frame = pte_pfn(pte) + 1405 ((addr & ~hmask) >> PAGE_SHIFT); 1406 } 1407 1408 for (; addr != end; addr += PAGE_SIZE) { 1409 pagemap_entry_t pme = make_pme(frame, flags); 1410 1411 err = add_to_pagemap(addr, &pme, pm); 1412 if (err) 1413 return err; 1414 if (pm->show_pfn && (flags & PM_PRESENT)) 1415 frame++; 1416 } 1417 1418 cond_resched(); 1419 1420 return err; 1421 } 1422 #endif /* HUGETLB_PAGE */ 1423 1424 /* 1425 * /proc/pid/pagemap - an array mapping virtual pages to pfns 1426 * 1427 * For each page in the address space, this file contains one 64-bit entry 1428 * consisting of the following: 1429 * 1430 * Bits 0-54 page frame number (PFN) if present 1431 * Bits 0-4 swap type if swapped 1432 * Bits 5-54 swap offset if swapped 1433 * Bit 55 pte is soft-dirty (see Documentation/vm/soft-dirty.txt) 1434 * Bit 56 page exclusively mapped 1435 * Bits 57-60 zero 1436 * Bit 61 page is file-page or shared-anon 1437 * Bit 62 page swapped 1438 * Bit 63 page present 1439 * 1440 * If the page is not present but in swap, then the PFN contains an 1441 * encoding of the swap file number and the page's offset into the 1442 * swap. Unmapped pages return a null PFN. This allows determining 1443 * precisely which pages are mapped (or in swap) and comparing mapped 1444 * pages between processes. 1445 * 1446 * Efficient users of this interface will use /proc/pid/maps to 1447 * determine which areas of memory are actually mapped and llseek to 1448 * skip over unmapped regions. 1449 */ 1450 static ssize_t pagemap_read(struct file *file, char __user *buf, 1451 size_t count, loff_t *ppos) 1452 { 1453 struct mm_struct *mm = file->private_data; 1454 struct pagemapread pm; 1455 struct mm_walk pagemap_walk = {}; 1456 unsigned long src; 1457 unsigned long svpfn; 1458 unsigned long start_vaddr; 1459 unsigned long end_vaddr; 1460 int ret = 0, copied = 0; 1461 1462 if (!mm || !mmget_not_zero(mm)) 1463 goto out; 1464 1465 ret = -EINVAL; 1466 /* file position must be aligned */ 1467 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES)) 1468 goto out_mm; 1469 1470 ret = 0; 1471 if (!count) 1472 goto out_mm; 1473 1474 /* do not disclose physical addresses: attack vector */ 1475 pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN); 1476 1477 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT); 1478 pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_KERNEL); 1479 ret = -ENOMEM; 1480 if (!pm.buffer) 1481 goto out_mm; 1482 1483 pagemap_walk.pmd_entry = pagemap_pmd_range; 1484 pagemap_walk.pte_hole = pagemap_pte_hole; 1485 #ifdef CONFIG_HUGETLB_PAGE 1486 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range; 1487 #endif 1488 pagemap_walk.mm = mm; 1489 pagemap_walk.private = ± 1490 1491 src = *ppos; 1492 svpfn = src / PM_ENTRY_BYTES; 1493 start_vaddr = svpfn << PAGE_SHIFT; 1494 end_vaddr = mm->task_size; 1495 1496 /* watch out for wraparound */ 1497 if (svpfn > mm->task_size >> PAGE_SHIFT) 1498 start_vaddr = end_vaddr; 1499 1500 /* 1501 * The odds are that this will stop walking way 1502 * before end_vaddr, because the length of the 1503 * user buffer is tracked in "pm", and the walk 1504 * will stop when we hit the end of the buffer. 1505 */ 1506 ret = 0; 1507 while (count && (start_vaddr < end_vaddr)) { 1508 int len; 1509 unsigned long end; 1510 1511 pm.pos = 0; 1512 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK; 1513 /* overflow ? */ 1514 if (end < start_vaddr || end > end_vaddr) 1515 end = end_vaddr; 1516 down_read(&mm->mmap_sem); 1517 ret = walk_page_range(start_vaddr, end, &pagemap_walk); 1518 up_read(&mm->mmap_sem); 1519 start_vaddr = end; 1520 1521 len = min(count, PM_ENTRY_BYTES * pm.pos); 1522 if (copy_to_user(buf, pm.buffer, len)) { 1523 ret = -EFAULT; 1524 goto out_free; 1525 } 1526 copied += len; 1527 buf += len; 1528 count -= len; 1529 } 1530 *ppos += copied; 1531 if (!ret || ret == PM_END_OF_BUFFER) 1532 ret = copied; 1533 1534 out_free: 1535 kfree(pm.buffer); 1536 out_mm: 1537 mmput(mm); 1538 out: 1539 return ret; 1540 } 1541 1542 static int pagemap_open(struct inode *inode, struct file *file) 1543 { 1544 struct mm_struct *mm; 1545 1546 mm = proc_mem_open(inode, PTRACE_MODE_READ); 1547 if (IS_ERR(mm)) 1548 return PTR_ERR(mm); 1549 file->private_data = mm; 1550 return 0; 1551 } 1552 1553 static int pagemap_release(struct inode *inode, struct file *file) 1554 { 1555 struct mm_struct *mm = file->private_data; 1556 1557 if (mm) 1558 mmdrop(mm); 1559 return 0; 1560 } 1561 1562 const struct file_operations proc_pagemap_operations = { 1563 .llseek = mem_lseek, /* borrow this */ 1564 .read = pagemap_read, 1565 .open = pagemap_open, 1566 .release = pagemap_release, 1567 }; 1568 #endif /* CONFIG_PROC_PAGE_MONITOR */ 1569 1570 #ifdef CONFIG_NUMA 1571 1572 struct numa_maps { 1573 unsigned long pages; 1574 unsigned long anon; 1575 unsigned long active; 1576 unsigned long writeback; 1577 unsigned long mapcount_max; 1578 unsigned long dirty; 1579 unsigned long swapcache; 1580 unsigned long node[MAX_NUMNODES]; 1581 }; 1582 1583 struct numa_maps_private { 1584 struct proc_maps_private proc_maps; 1585 struct numa_maps md; 1586 }; 1587 1588 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty, 1589 unsigned long nr_pages) 1590 { 1591 int count = page_mapcount(page); 1592 1593 md->pages += nr_pages; 1594 if (pte_dirty || PageDirty(page)) 1595 md->dirty += nr_pages; 1596 1597 if (PageSwapCache(page)) 1598 md->swapcache += nr_pages; 1599 1600 if (PageActive(page) || PageUnevictable(page)) 1601 md->active += nr_pages; 1602 1603 if (PageWriteback(page)) 1604 md->writeback += nr_pages; 1605 1606 if (PageAnon(page)) 1607 md->anon += nr_pages; 1608 1609 if (count > md->mapcount_max) 1610 md->mapcount_max = count; 1611 1612 md->node[page_to_nid(page)] += nr_pages; 1613 } 1614 1615 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma, 1616 unsigned long addr) 1617 { 1618 struct page *page; 1619 int nid; 1620 1621 if (!pte_present(pte)) 1622 return NULL; 1623 1624 page = vm_normal_page(vma, addr, pte); 1625 if (!page) 1626 return NULL; 1627 1628 if (PageReserved(page)) 1629 return NULL; 1630 1631 nid = page_to_nid(page); 1632 if (!node_isset(nid, node_states[N_MEMORY])) 1633 return NULL; 1634 1635 return page; 1636 } 1637 1638 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1639 static struct page *can_gather_numa_stats_pmd(pmd_t pmd, 1640 struct vm_area_struct *vma, 1641 unsigned long addr) 1642 { 1643 struct page *page; 1644 int nid; 1645 1646 if (!pmd_present(pmd)) 1647 return NULL; 1648 1649 page = vm_normal_page_pmd(vma, addr, pmd); 1650 if (!page) 1651 return NULL; 1652 1653 if (PageReserved(page)) 1654 return NULL; 1655 1656 nid = page_to_nid(page); 1657 if (!node_isset(nid, node_states[N_MEMORY])) 1658 return NULL; 1659 1660 return page; 1661 } 1662 #endif 1663 1664 static int gather_pte_stats(pmd_t *pmd, unsigned long addr, 1665 unsigned long end, struct mm_walk *walk) 1666 { 1667 struct numa_maps *md = walk->private; 1668 struct vm_area_struct *vma = walk->vma; 1669 spinlock_t *ptl; 1670 pte_t *orig_pte; 1671 pte_t *pte; 1672 1673 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1674 ptl = pmd_trans_huge_lock(pmd, vma); 1675 if (ptl) { 1676 struct page *page; 1677 1678 page = can_gather_numa_stats_pmd(*pmd, vma, addr); 1679 if (page) 1680 gather_stats(page, md, pmd_dirty(*pmd), 1681 HPAGE_PMD_SIZE/PAGE_SIZE); 1682 spin_unlock(ptl); 1683 return 0; 1684 } 1685 1686 if (pmd_trans_unstable(pmd)) 1687 return 0; 1688 #endif 1689 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl); 1690 do { 1691 struct page *page = can_gather_numa_stats(*pte, vma, addr); 1692 if (!page) 1693 continue; 1694 gather_stats(page, md, pte_dirty(*pte), 1); 1695 1696 } while (pte++, addr += PAGE_SIZE, addr != end); 1697 pte_unmap_unlock(orig_pte, ptl); 1698 cond_resched(); 1699 return 0; 1700 } 1701 #ifdef CONFIG_HUGETLB_PAGE 1702 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask, 1703 unsigned long addr, unsigned long end, struct mm_walk *walk) 1704 { 1705 pte_t huge_pte = huge_ptep_get(pte); 1706 struct numa_maps *md; 1707 struct page *page; 1708 1709 if (!pte_present(huge_pte)) 1710 return 0; 1711 1712 page = pte_page(huge_pte); 1713 if (!page) 1714 return 0; 1715 1716 md = walk->private; 1717 gather_stats(page, md, pte_dirty(huge_pte), 1); 1718 return 0; 1719 } 1720 1721 #else 1722 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask, 1723 unsigned long addr, unsigned long end, struct mm_walk *walk) 1724 { 1725 return 0; 1726 } 1727 #endif 1728 1729 /* 1730 * Display pages allocated per node and memory policy via /proc. 1731 */ 1732 static int show_numa_map(struct seq_file *m, void *v, int is_pid) 1733 { 1734 struct numa_maps_private *numa_priv = m->private; 1735 struct proc_maps_private *proc_priv = &numa_priv->proc_maps; 1736 struct vm_area_struct *vma = v; 1737 struct numa_maps *md = &numa_priv->md; 1738 struct file *file = vma->vm_file; 1739 struct mm_struct *mm = vma->vm_mm; 1740 struct mm_walk walk = { 1741 .hugetlb_entry = gather_hugetlb_stats, 1742 .pmd_entry = gather_pte_stats, 1743 .private = md, 1744 .mm = mm, 1745 }; 1746 struct mempolicy *pol; 1747 char buffer[64]; 1748 int nid; 1749 1750 if (!mm) 1751 return 0; 1752 1753 /* Ensure we start with an empty set of numa_maps statistics. */ 1754 memset(md, 0, sizeof(*md)); 1755 1756 pol = __get_vma_policy(vma, vma->vm_start); 1757 if (pol) { 1758 mpol_to_str(buffer, sizeof(buffer), pol); 1759 mpol_cond_put(pol); 1760 } else { 1761 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy); 1762 } 1763 1764 seq_printf(m, "%08lx %s", vma->vm_start, buffer); 1765 1766 if (file) { 1767 seq_puts(m, " file="); 1768 seq_file_path(m, file, "\n\t= "); 1769 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) { 1770 seq_puts(m, " heap"); 1771 } else if (is_stack(vma)) { 1772 seq_puts(m, " stack"); 1773 } 1774 1775 if (is_vm_hugetlb_page(vma)) 1776 seq_puts(m, " huge"); 1777 1778 /* mmap_sem is held by m_start */ 1779 walk_page_vma(vma, &walk); 1780 1781 if (!md->pages) 1782 goto out; 1783 1784 if (md->anon) 1785 seq_printf(m, " anon=%lu", md->anon); 1786 1787 if (md->dirty) 1788 seq_printf(m, " dirty=%lu", md->dirty); 1789 1790 if (md->pages != md->anon && md->pages != md->dirty) 1791 seq_printf(m, " mapped=%lu", md->pages); 1792 1793 if (md->mapcount_max > 1) 1794 seq_printf(m, " mapmax=%lu", md->mapcount_max); 1795 1796 if (md->swapcache) 1797 seq_printf(m, " swapcache=%lu", md->swapcache); 1798 1799 if (md->active < md->pages && !is_vm_hugetlb_page(vma)) 1800 seq_printf(m, " active=%lu", md->active); 1801 1802 if (md->writeback) 1803 seq_printf(m, " writeback=%lu", md->writeback); 1804 1805 for_each_node_state(nid, N_MEMORY) 1806 if (md->node[nid]) 1807 seq_printf(m, " N%d=%lu", nid, md->node[nid]); 1808 1809 seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10); 1810 out: 1811 seq_putc(m, '\n'); 1812 m_cache_vma(m, vma); 1813 return 0; 1814 } 1815 1816 static int show_pid_numa_map(struct seq_file *m, void *v) 1817 { 1818 return show_numa_map(m, v, 1); 1819 } 1820 1821 static int show_tid_numa_map(struct seq_file *m, void *v) 1822 { 1823 return show_numa_map(m, v, 0); 1824 } 1825 1826 static const struct seq_operations proc_pid_numa_maps_op = { 1827 .start = m_start, 1828 .next = m_next, 1829 .stop = m_stop, 1830 .show = show_pid_numa_map, 1831 }; 1832 1833 static const struct seq_operations proc_tid_numa_maps_op = { 1834 .start = m_start, 1835 .next = m_next, 1836 .stop = m_stop, 1837 .show = show_tid_numa_map, 1838 }; 1839 1840 static int numa_maps_open(struct inode *inode, struct file *file, 1841 const struct seq_operations *ops) 1842 { 1843 return proc_maps_open(inode, file, ops, 1844 sizeof(struct numa_maps_private)); 1845 } 1846 1847 static int pid_numa_maps_open(struct inode *inode, struct file *file) 1848 { 1849 return numa_maps_open(inode, file, &proc_pid_numa_maps_op); 1850 } 1851 1852 static int tid_numa_maps_open(struct inode *inode, struct file *file) 1853 { 1854 return numa_maps_open(inode, file, &proc_tid_numa_maps_op); 1855 } 1856 1857 const struct file_operations proc_pid_numa_maps_operations = { 1858 .open = pid_numa_maps_open, 1859 .read = seq_read, 1860 .llseek = seq_lseek, 1861 .release = proc_map_release, 1862 }; 1863 1864 const struct file_operations proc_tid_numa_maps_operations = { 1865 .open = tid_numa_maps_open, 1866 .read = seq_read, 1867 .llseek = seq_lseek, 1868 .release = proc_map_release, 1869 }; 1870 #endif /* CONFIG_NUMA */ 1871