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/swapops.h> 15 #include <linux/mmu_notifier.h> 16 17 #include <asm/elf.h> 18 #include <asm/uaccess.h> 19 #include <asm/tlbflush.h> 20 #include "internal.h" 21 22 void task_mem(struct seq_file *m, struct mm_struct *mm) 23 { 24 unsigned long data, text, lib, swap, ptes, pmds; 25 unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss; 26 27 /* 28 * Note: to minimize their overhead, mm maintains hiwater_vm and 29 * hiwater_rss only when about to *lower* total_vm or rss. Any 30 * collector of these hiwater stats must therefore get total_vm 31 * and rss too, which will usually be the higher. Barriers? not 32 * worth the effort, such snapshots can always be inconsistent. 33 */ 34 hiwater_vm = total_vm = mm->total_vm; 35 if (hiwater_vm < mm->hiwater_vm) 36 hiwater_vm = mm->hiwater_vm; 37 hiwater_rss = total_rss = get_mm_rss(mm); 38 if (hiwater_rss < mm->hiwater_rss) 39 hiwater_rss = mm->hiwater_rss; 40 41 data = mm->total_vm - mm->shared_vm - mm->stack_vm; 42 text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10; 43 lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text; 44 swap = get_mm_counter(mm, MM_SWAPENTS); 45 ptes = PTRS_PER_PTE * sizeof(pte_t) * atomic_long_read(&mm->nr_ptes); 46 pmds = PTRS_PER_PMD * sizeof(pmd_t) * mm_nr_pmds(mm); 47 seq_printf(m, 48 "VmPeak:\t%8lu kB\n" 49 "VmSize:\t%8lu kB\n" 50 "VmLck:\t%8lu kB\n" 51 "VmPin:\t%8lu kB\n" 52 "VmHWM:\t%8lu kB\n" 53 "VmRSS:\t%8lu kB\n" 54 "VmData:\t%8lu kB\n" 55 "VmStk:\t%8lu kB\n" 56 "VmExe:\t%8lu kB\n" 57 "VmLib:\t%8lu kB\n" 58 "VmPTE:\t%8lu kB\n" 59 "VmPMD:\t%8lu kB\n" 60 "VmSwap:\t%8lu kB\n", 61 hiwater_vm << (PAGE_SHIFT-10), 62 total_vm << (PAGE_SHIFT-10), 63 mm->locked_vm << (PAGE_SHIFT-10), 64 mm->pinned_vm << (PAGE_SHIFT-10), 65 hiwater_rss << (PAGE_SHIFT-10), 66 total_rss << (PAGE_SHIFT-10), 67 data << (PAGE_SHIFT-10), 68 mm->stack_vm << (PAGE_SHIFT-10), text, lib, 69 ptes >> 10, 70 pmds >> 10, 71 swap << (PAGE_SHIFT-10)); 72 } 73 74 unsigned long task_vsize(struct mm_struct *mm) 75 { 76 return PAGE_SIZE * mm->total_vm; 77 } 78 79 unsigned long task_statm(struct mm_struct *mm, 80 unsigned long *shared, unsigned long *text, 81 unsigned long *data, unsigned long *resident) 82 { 83 *shared = get_mm_counter(mm, MM_FILEPAGES); 84 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) 85 >> PAGE_SHIFT; 86 *data = mm->total_vm - mm->shared_vm; 87 *resident = *shared + get_mm_counter(mm, MM_ANONPAGES); 88 return mm->total_vm; 89 } 90 91 #ifdef CONFIG_NUMA 92 /* 93 * Save get_task_policy() for show_numa_map(). 94 */ 95 static void hold_task_mempolicy(struct proc_maps_private *priv) 96 { 97 struct task_struct *task = priv->task; 98 99 task_lock(task); 100 priv->task_mempolicy = get_task_policy(task); 101 mpol_get(priv->task_mempolicy); 102 task_unlock(task); 103 } 104 static void release_task_mempolicy(struct proc_maps_private *priv) 105 { 106 mpol_put(priv->task_mempolicy); 107 } 108 #else 109 static void hold_task_mempolicy(struct proc_maps_private *priv) 110 { 111 } 112 static void release_task_mempolicy(struct proc_maps_private *priv) 113 { 114 } 115 #endif 116 117 static void vma_stop(struct proc_maps_private *priv) 118 { 119 struct mm_struct *mm = priv->mm; 120 121 release_task_mempolicy(priv); 122 up_read(&mm->mmap_sem); 123 mmput(mm); 124 } 125 126 static struct vm_area_struct * 127 m_next_vma(struct proc_maps_private *priv, struct vm_area_struct *vma) 128 { 129 if (vma == priv->tail_vma) 130 return NULL; 131 return vma->vm_next ?: priv->tail_vma; 132 } 133 134 static void m_cache_vma(struct seq_file *m, struct vm_area_struct *vma) 135 { 136 if (m->count < m->size) /* vma is copied successfully */ 137 m->version = m_next_vma(m->private, vma) ? vma->vm_start : -1UL; 138 } 139 140 static void *m_start(struct seq_file *m, loff_t *ppos) 141 { 142 struct proc_maps_private *priv = m->private; 143 unsigned long last_addr = m->version; 144 struct mm_struct *mm; 145 struct vm_area_struct *vma; 146 unsigned int pos = *ppos; 147 148 /* See m_cache_vma(). Zero at the start or after lseek. */ 149 if (last_addr == -1UL) 150 return NULL; 151 152 priv->task = get_proc_task(priv->inode); 153 if (!priv->task) 154 return ERR_PTR(-ESRCH); 155 156 mm = priv->mm; 157 if (!mm || !atomic_inc_not_zero(&mm->mm_users)) 158 return NULL; 159 160 down_read(&mm->mmap_sem); 161 hold_task_mempolicy(priv); 162 priv->tail_vma = get_gate_vma(mm); 163 164 if (last_addr) { 165 vma = find_vma(mm, last_addr); 166 if (vma && (vma = m_next_vma(priv, vma))) 167 return vma; 168 } 169 170 m->version = 0; 171 if (pos < mm->map_count) { 172 for (vma = mm->mmap; pos; pos--) { 173 m->version = vma->vm_start; 174 vma = vma->vm_next; 175 } 176 return vma; 177 } 178 179 /* we do not bother to update m->version in this case */ 180 if (pos == mm->map_count && priv->tail_vma) 181 return priv->tail_vma; 182 183 vma_stop(priv); 184 return NULL; 185 } 186 187 static void *m_next(struct seq_file *m, void *v, loff_t *pos) 188 { 189 struct proc_maps_private *priv = m->private; 190 struct vm_area_struct *next; 191 192 (*pos)++; 193 next = m_next_vma(priv, v); 194 if (!next) 195 vma_stop(priv); 196 return next; 197 } 198 199 static void m_stop(struct seq_file *m, void *v) 200 { 201 struct proc_maps_private *priv = m->private; 202 203 if (!IS_ERR_OR_NULL(v)) 204 vma_stop(priv); 205 if (priv->task) { 206 put_task_struct(priv->task); 207 priv->task = NULL; 208 } 209 } 210 211 static int proc_maps_open(struct inode *inode, struct file *file, 212 const struct seq_operations *ops, int psize) 213 { 214 struct proc_maps_private *priv = __seq_open_private(file, ops, psize); 215 216 if (!priv) 217 return -ENOMEM; 218 219 priv->inode = inode; 220 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ); 221 if (IS_ERR(priv->mm)) { 222 int err = PTR_ERR(priv->mm); 223 224 seq_release_private(inode, file); 225 return err; 226 } 227 228 return 0; 229 } 230 231 static int proc_map_release(struct inode *inode, struct file *file) 232 { 233 struct seq_file *seq = file->private_data; 234 struct proc_maps_private *priv = seq->private; 235 236 if (priv->mm) 237 mmdrop(priv->mm); 238 239 return seq_release_private(inode, file); 240 } 241 242 static int do_maps_open(struct inode *inode, struct file *file, 243 const struct seq_operations *ops) 244 { 245 return proc_maps_open(inode, file, ops, 246 sizeof(struct proc_maps_private)); 247 } 248 249 static pid_t pid_of_stack(struct proc_maps_private *priv, 250 struct vm_area_struct *vma, bool is_pid) 251 { 252 struct inode *inode = priv->inode; 253 struct task_struct *task; 254 pid_t ret = 0; 255 256 rcu_read_lock(); 257 task = pid_task(proc_pid(inode), PIDTYPE_PID); 258 if (task) { 259 task = task_of_stack(task, vma, is_pid); 260 if (task) 261 ret = task_pid_nr_ns(task, inode->i_sb->s_fs_info); 262 } 263 rcu_read_unlock(); 264 265 return ret; 266 } 267 268 static void 269 show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid) 270 { 271 struct mm_struct *mm = vma->vm_mm; 272 struct file *file = vma->vm_file; 273 struct proc_maps_private *priv = m->private; 274 vm_flags_t flags = vma->vm_flags; 275 unsigned long ino = 0; 276 unsigned long long pgoff = 0; 277 unsigned long start, end; 278 dev_t dev = 0; 279 const char *name = NULL; 280 281 if (file) { 282 struct inode *inode = file_inode(vma->vm_file); 283 dev = inode->i_sb->s_dev; 284 ino = inode->i_ino; 285 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT; 286 } 287 288 /* We don't show the stack guard page in /proc/maps */ 289 start = vma->vm_start; 290 if (stack_guard_page_start(vma, start)) 291 start += PAGE_SIZE; 292 end = vma->vm_end; 293 if (stack_guard_page_end(vma, end)) 294 end -= PAGE_SIZE; 295 296 seq_setwidth(m, 25 + sizeof(void *) * 6 - 1); 297 seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu ", 298 start, 299 end, 300 flags & VM_READ ? 'r' : '-', 301 flags & VM_WRITE ? 'w' : '-', 302 flags & VM_EXEC ? 'x' : '-', 303 flags & VM_MAYSHARE ? 's' : 'p', 304 pgoff, 305 MAJOR(dev), MINOR(dev), ino); 306 307 /* 308 * Print the dentry name for named mappings, and a 309 * special [heap] marker for the heap: 310 */ 311 if (file) { 312 seq_pad(m, ' '); 313 seq_file_path(m, file, "\n"); 314 goto done; 315 } 316 317 if (vma->vm_ops && vma->vm_ops->name) { 318 name = vma->vm_ops->name(vma); 319 if (name) 320 goto done; 321 } 322 323 name = arch_vma_name(vma); 324 if (!name) { 325 pid_t tid; 326 327 if (!mm) { 328 name = "[vdso]"; 329 goto done; 330 } 331 332 if (vma->vm_start <= mm->brk && 333 vma->vm_end >= mm->start_brk) { 334 name = "[heap]"; 335 goto done; 336 } 337 338 tid = pid_of_stack(priv, vma, is_pid); 339 if (tid != 0) { 340 /* 341 * Thread stack in /proc/PID/task/TID/maps or 342 * the main process stack. 343 */ 344 if (!is_pid || (vma->vm_start <= mm->start_stack && 345 vma->vm_end >= mm->start_stack)) { 346 name = "[stack]"; 347 } else { 348 /* Thread stack in /proc/PID/maps */ 349 seq_pad(m, ' '); 350 seq_printf(m, "[stack:%d]", tid); 351 } 352 } 353 } 354 355 done: 356 if (name) { 357 seq_pad(m, ' '); 358 seq_puts(m, name); 359 } 360 seq_putc(m, '\n'); 361 } 362 363 static int show_map(struct seq_file *m, void *v, int is_pid) 364 { 365 show_map_vma(m, v, is_pid); 366 m_cache_vma(m, v); 367 return 0; 368 } 369 370 static int show_pid_map(struct seq_file *m, void *v) 371 { 372 return show_map(m, v, 1); 373 } 374 375 static int show_tid_map(struct seq_file *m, void *v) 376 { 377 return show_map(m, v, 0); 378 } 379 380 static const struct seq_operations proc_pid_maps_op = { 381 .start = m_start, 382 .next = m_next, 383 .stop = m_stop, 384 .show = show_pid_map 385 }; 386 387 static const struct seq_operations proc_tid_maps_op = { 388 .start = m_start, 389 .next = m_next, 390 .stop = m_stop, 391 .show = show_tid_map 392 }; 393 394 static int pid_maps_open(struct inode *inode, struct file *file) 395 { 396 return do_maps_open(inode, file, &proc_pid_maps_op); 397 } 398 399 static int tid_maps_open(struct inode *inode, struct file *file) 400 { 401 return do_maps_open(inode, file, &proc_tid_maps_op); 402 } 403 404 const struct file_operations proc_pid_maps_operations = { 405 .open = pid_maps_open, 406 .read = seq_read, 407 .llseek = seq_lseek, 408 .release = proc_map_release, 409 }; 410 411 const struct file_operations proc_tid_maps_operations = { 412 .open = tid_maps_open, 413 .read = seq_read, 414 .llseek = seq_lseek, 415 .release = proc_map_release, 416 }; 417 418 /* 419 * Proportional Set Size(PSS): my share of RSS. 420 * 421 * PSS of a process is the count of pages it has in memory, where each 422 * page is divided by the number of processes sharing it. So if a 423 * process has 1000 pages all to itself, and 1000 shared with one other 424 * process, its PSS will be 1500. 425 * 426 * To keep (accumulated) division errors low, we adopt a 64bit 427 * fixed-point pss counter to minimize division errors. So (pss >> 428 * PSS_SHIFT) would be the real byte count. 429 * 430 * A shift of 12 before division means (assuming 4K page size): 431 * - 1M 3-user-pages add up to 8KB errors; 432 * - supports mapcount up to 2^24, or 16M; 433 * - supports PSS up to 2^52 bytes, or 4PB. 434 */ 435 #define PSS_SHIFT 12 436 437 #ifdef CONFIG_PROC_PAGE_MONITOR 438 struct mem_size_stats { 439 unsigned long resident; 440 unsigned long shared_clean; 441 unsigned long shared_dirty; 442 unsigned long private_clean; 443 unsigned long private_dirty; 444 unsigned long referenced; 445 unsigned long anonymous; 446 unsigned long anonymous_thp; 447 unsigned long swap; 448 u64 pss; 449 }; 450 451 static void smaps_account(struct mem_size_stats *mss, struct page *page, 452 unsigned long size, bool young, bool dirty) 453 { 454 int mapcount; 455 456 if (PageAnon(page)) 457 mss->anonymous += size; 458 459 mss->resident += size; 460 /* Accumulate the size in pages that have been accessed. */ 461 if (young || PageReferenced(page)) 462 mss->referenced += size; 463 mapcount = page_mapcount(page); 464 if (mapcount >= 2) { 465 u64 pss_delta; 466 467 if (dirty || PageDirty(page)) 468 mss->shared_dirty += size; 469 else 470 mss->shared_clean += size; 471 pss_delta = (u64)size << PSS_SHIFT; 472 do_div(pss_delta, mapcount); 473 mss->pss += pss_delta; 474 } else { 475 if (dirty || PageDirty(page)) 476 mss->private_dirty += size; 477 else 478 mss->private_clean += size; 479 mss->pss += (u64)size << PSS_SHIFT; 480 } 481 } 482 483 static void smaps_pte_entry(pte_t *pte, unsigned long addr, 484 struct mm_walk *walk) 485 { 486 struct mem_size_stats *mss = walk->private; 487 struct vm_area_struct *vma = walk->vma; 488 struct page *page = NULL; 489 490 if (pte_present(*pte)) { 491 page = vm_normal_page(vma, addr, *pte); 492 } else if (is_swap_pte(*pte)) { 493 swp_entry_t swpent = pte_to_swp_entry(*pte); 494 495 if (!non_swap_entry(swpent)) 496 mss->swap += PAGE_SIZE; 497 else if (is_migration_entry(swpent)) 498 page = migration_entry_to_page(swpent); 499 } 500 501 if (!page) 502 return; 503 smaps_account(mss, page, PAGE_SIZE, pte_young(*pte), pte_dirty(*pte)); 504 } 505 506 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 507 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr, 508 struct mm_walk *walk) 509 { 510 struct mem_size_stats *mss = walk->private; 511 struct vm_area_struct *vma = walk->vma; 512 struct page *page; 513 514 /* FOLL_DUMP will return -EFAULT on huge zero page */ 515 page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP); 516 if (IS_ERR_OR_NULL(page)) 517 return; 518 mss->anonymous_thp += HPAGE_PMD_SIZE; 519 smaps_account(mss, page, HPAGE_PMD_SIZE, 520 pmd_young(*pmd), pmd_dirty(*pmd)); 521 } 522 #else 523 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr, 524 struct mm_walk *walk) 525 { 526 } 527 #endif 528 529 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, 530 struct mm_walk *walk) 531 { 532 struct vm_area_struct *vma = walk->vma; 533 pte_t *pte; 534 spinlock_t *ptl; 535 536 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { 537 smaps_pmd_entry(pmd, addr, walk); 538 spin_unlock(ptl); 539 return 0; 540 } 541 542 if (pmd_trans_unstable(pmd)) 543 return 0; 544 /* 545 * The mmap_sem held all the way back in m_start() is what 546 * keeps khugepaged out of here and from collapsing things 547 * in here. 548 */ 549 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); 550 for (; addr != end; pte++, addr += PAGE_SIZE) 551 smaps_pte_entry(pte, addr, walk); 552 pte_unmap_unlock(pte - 1, ptl); 553 cond_resched(); 554 return 0; 555 } 556 557 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma) 558 { 559 /* 560 * Don't forget to update Documentation/ on changes. 561 */ 562 static const char mnemonics[BITS_PER_LONG][2] = { 563 /* 564 * In case if we meet a flag we don't know about. 565 */ 566 [0 ... (BITS_PER_LONG-1)] = "??", 567 568 [ilog2(VM_READ)] = "rd", 569 [ilog2(VM_WRITE)] = "wr", 570 [ilog2(VM_EXEC)] = "ex", 571 [ilog2(VM_SHARED)] = "sh", 572 [ilog2(VM_MAYREAD)] = "mr", 573 [ilog2(VM_MAYWRITE)] = "mw", 574 [ilog2(VM_MAYEXEC)] = "me", 575 [ilog2(VM_MAYSHARE)] = "ms", 576 [ilog2(VM_GROWSDOWN)] = "gd", 577 [ilog2(VM_PFNMAP)] = "pf", 578 [ilog2(VM_DENYWRITE)] = "dw", 579 #ifdef CONFIG_X86_INTEL_MPX 580 [ilog2(VM_MPX)] = "mp", 581 #endif 582 [ilog2(VM_LOCKED)] = "lo", 583 [ilog2(VM_IO)] = "io", 584 [ilog2(VM_SEQ_READ)] = "sr", 585 [ilog2(VM_RAND_READ)] = "rr", 586 [ilog2(VM_DONTCOPY)] = "dc", 587 [ilog2(VM_DONTEXPAND)] = "de", 588 [ilog2(VM_ACCOUNT)] = "ac", 589 [ilog2(VM_NORESERVE)] = "nr", 590 [ilog2(VM_HUGETLB)] = "ht", 591 [ilog2(VM_ARCH_1)] = "ar", 592 [ilog2(VM_DONTDUMP)] = "dd", 593 #ifdef CONFIG_MEM_SOFT_DIRTY 594 [ilog2(VM_SOFTDIRTY)] = "sd", 595 #endif 596 [ilog2(VM_MIXEDMAP)] = "mm", 597 [ilog2(VM_HUGEPAGE)] = "hg", 598 [ilog2(VM_NOHUGEPAGE)] = "nh", 599 [ilog2(VM_MERGEABLE)] = "mg", 600 [ilog2(VM_UFFD_MISSING)]= "um", 601 [ilog2(VM_UFFD_WP)] = "uw", 602 }; 603 size_t i; 604 605 seq_puts(m, "VmFlags: "); 606 for (i = 0; i < BITS_PER_LONG; i++) { 607 if (vma->vm_flags & (1UL << i)) { 608 seq_printf(m, "%c%c ", 609 mnemonics[i][0], mnemonics[i][1]); 610 } 611 } 612 seq_putc(m, '\n'); 613 } 614 615 static int show_smap(struct seq_file *m, void *v, int is_pid) 616 { 617 struct vm_area_struct *vma = v; 618 struct mem_size_stats mss; 619 struct mm_walk smaps_walk = { 620 .pmd_entry = smaps_pte_range, 621 .mm = vma->vm_mm, 622 .private = &mss, 623 }; 624 625 memset(&mss, 0, sizeof mss); 626 /* mmap_sem is held in m_start */ 627 walk_page_vma(vma, &smaps_walk); 628 629 show_map_vma(m, vma, is_pid); 630 631 seq_printf(m, 632 "Size: %8lu kB\n" 633 "Rss: %8lu kB\n" 634 "Pss: %8lu kB\n" 635 "Shared_Clean: %8lu kB\n" 636 "Shared_Dirty: %8lu kB\n" 637 "Private_Clean: %8lu kB\n" 638 "Private_Dirty: %8lu kB\n" 639 "Referenced: %8lu kB\n" 640 "Anonymous: %8lu kB\n" 641 "AnonHugePages: %8lu kB\n" 642 "Swap: %8lu kB\n" 643 "KernelPageSize: %8lu kB\n" 644 "MMUPageSize: %8lu kB\n" 645 "Locked: %8lu kB\n", 646 (vma->vm_end - vma->vm_start) >> 10, 647 mss.resident >> 10, 648 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)), 649 mss.shared_clean >> 10, 650 mss.shared_dirty >> 10, 651 mss.private_clean >> 10, 652 mss.private_dirty >> 10, 653 mss.referenced >> 10, 654 mss.anonymous >> 10, 655 mss.anonymous_thp >> 10, 656 mss.swap >> 10, 657 vma_kernel_pagesize(vma) >> 10, 658 vma_mmu_pagesize(vma) >> 10, 659 (vma->vm_flags & VM_LOCKED) ? 660 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0); 661 662 show_smap_vma_flags(m, vma); 663 m_cache_vma(m, vma); 664 return 0; 665 } 666 667 static int show_pid_smap(struct seq_file *m, void *v) 668 { 669 return show_smap(m, v, 1); 670 } 671 672 static int show_tid_smap(struct seq_file *m, void *v) 673 { 674 return show_smap(m, v, 0); 675 } 676 677 static const struct seq_operations proc_pid_smaps_op = { 678 .start = m_start, 679 .next = m_next, 680 .stop = m_stop, 681 .show = show_pid_smap 682 }; 683 684 static const struct seq_operations proc_tid_smaps_op = { 685 .start = m_start, 686 .next = m_next, 687 .stop = m_stop, 688 .show = show_tid_smap 689 }; 690 691 static int pid_smaps_open(struct inode *inode, struct file *file) 692 { 693 return do_maps_open(inode, file, &proc_pid_smaps_op); 694 } 695 696 static int tid_smaps_open(struct inode *inode, struct file *file) 697 { 698 return do_maps_open(inode, file, &proc_tid_smaps_op); 699 } 700 701 const struct file_operations proc_pid_smaps_operations = { 702 .open = pid_smaps_open, 703 .read = seq_read, 704 .llseek = seq_lseek, 705 .release = proc_map_release, 706 }; 707 708 const struct file_operations proc_tid_smaps_operations = { 709 .open = tid_smaps_open, 710 .read = seq_read, 711 .llseek = seq_lseek, 712 .release = proc_map_release, 713 }; 714 715 /* 716 * We do not want to have constant page-shift bits sitting in 717 * pagemap entries and are about to reuse them some time soon. 718 * 719 * Here's the "migration strategy": 720 * 1. when the system boots these bits remain what they are, 721 * but a warning about future change is printed in log; 722 * 2. once anyone clears soft-dirty bits via clear_refs file, 723 * these flag is set to denote, that user is aware of the 724 * new API and those page-shift bits change their meaning. 725 * The respective warning is printed in dmesg; 726 * 3. In a couple of releases we will remove all the mentions 727 * of page-shift in pagemap entries. 728 */ 729 730 static bool soft_dirty_cleared __read_mostly; 731 732 enum clear_refs_types { 733 CLEAR_REFS_ALL = 1, 734 CLEAR_REFS_ANON, 735 CLEAR_REFS_MAPPED, 736 CLEAR_REFS_SOFT_DIRTY, 737 CLEAR_REFS_MM_HIWATER_RSS, 738 CLEAR_REFS_LAST, 739 }; 740 741 struct clear_refs_private { 742 enum clear_refs_types type; 743 }; 744 745 #ifdef CONFIG_MEM_SOFT_DIRTY 746 static inline void clear_soft_dirty(struct vm_area_struct *vma, 747 unsigned long addr, pte_t *pte) 748 { 749 /* 750 * The soft-dirty tracker uses #PF-s to catch writes 751 * to pages, so write-protect the pte as well. See the 752 * Documentation/vm/soft-dirty.txt for full description 753 * of how soft-dirty works. 754 */ 755 pte_t ptent = *pte; 756 757 if (pte_present(ptent)) { 758 ptent = pte_wrprotect(ptent); 759 ptent = pte_clear_flags(ptent, _PAGE_SOFT_DIRTY); 760 } else if (is_swap_pte(ptent)) { 761 ptent = pte_swp_clear_soft_dirty(ptent); 762 } 763 764 set_pte_at(vma->vm_mm, addr, pte, ptent); 765 } 766 767 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma, 768 unsigned long addr, pmd_t *pmdp) 769 { 770 pmd_t pmd = *pmdp; 771 772 pmd = pmd_wrprotect(pmd); 773 pmd = pmd_clear_flags(pmd, _PAGE_SOFT_DIRTY); 774 775 if (vma->vm_flags & VM_SOFTDIRTY) 776 vma->vm_flags &= ~VM_SOFTDIRTY; 777 778 set_pmd_at(vma->vm_mm, addr, pmdp, pmd); 779 } 780 781 #else 782 783 static inline void clear_soft_dirty(struct vm_area_struct *vma, 784 unsigned long addr, pte_t *pte) 785 { 786 } 787 788 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma, 789 unsigned long addr, pmd_t *pmdp) 790 { 791 } 792 #endif 793 794 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr, 795 unsigned long end, struct mm_walk *walk) 796 { 797 struct clear_refs_private *cp = walk->private; 798 struct vm_area_struct *vma = walk->vma; 799 pte_t *pte, ptent; 800 spinlock_t *ptl; 801 struct page *page; 802 803 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { 804 if (cp->type == CLEAR_REFS_SOFT_DIRTY) { 805 clear_soft_dirty_pmd(vma, addr, pmd); 806 goto out; 807 } 808 809 page = pmd_page(*pmd); 810 811 /* Clear accessed and referenced bits. */ 812 pmdp_test_and_clear_young(vma, addr, pmd); 813 ClearPageReferenced(page); 814 out: 815 spin_unlock(ptl); 816 return 0; 817 } 818 819 if (pmd_trans_unstable(pmd)) 820 return 0; 821 822 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); 823 for (; addr != end; pte++, addr += PAGE_SIZE) { 824 ptent = *pte; 825 826 if (cp->type == CLEAR_REFS_SOFT_DIRTY) { 827 clear_soft_dirty(vma, addr, pte); 828 continue; 829 } 830 831 if (!pte_present(ptent)) 832 continue; 833 834 page = vm_normal_page(vma, addr, ptent); 835 if (!page) 836 continue; 837 838 /* Clear accessed and referenced bits. */ 839 ptep_test_and_clear_young(vma, addr, pte); 840 ClearPageReferenced(page); 841 } 842 pte_unmap_unlock(pte - 1, ptl); 843 cond_resched(); 844 return 0; 845 } 846 847 static int clear_refs_test_walk(unsigned long start, unsigned long end, 848 struct mm_walk *walk) 849 { 850 struct clear_refs_private *cp = walk->private; 851 struct vm_area_struct *vma = walk->vma; 852 853 if (vma->vm_flags & VM_PFNMAP) 854 return 1; 855 856 /* 857 * Writing 1 to /proc/pid/clear_refs affects all pages. 858 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages. 859 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages. 860 * Writing 4 to /proc/pid/clear_refs affects all pages. 861 */ 862 if (cp->type == CLEAR_REFS_ANON && vma->vm_file) 863 return 1; 864 if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file) 865 return 1; 866 return 0; 867 } 868 869 static ssize_t clear_refs_write(struct file *file, const char __user *buf, 870 size_t count, loff_t *ppos) 871 { 872 struct task_struct *task; 873 char buffer[PROC_NUMBUF]; 874 struct mm_struct *mm; 875 struct vm_area_struct *vma; 876 enum clear_refs_types type; 877 int itype; 878 int rv; 879 880 memset(buffer, 0, sizeof(buffer)); 881 if (count > sizeof(buffer) - 1) 882 count = sizeof(buffer) - 1; 883 if (copy_from_user(buffer, buf, count)) 884 return -EFAULT; 885 rv = kstrtoint(strstrip(buffer), 10, &itype); 886 if (rv < 0) 887 return rv; 888 type = (enum clear_refs_types)itype; 889 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST) 890 return -EINVAL; 891 892 if (type == CLEAR_REFS_SOFT_DIRTY) { 893 soft_dirty_cleared = true; 894 pr_warn_once("The pagemap bits 55-60 has changed their meaning!" 895 " See the linux/Documentation/vm/pagemap.txt for " 896 "details.\n"); 897 } 898 899 task = get_proc_task(file_inode(file)); 900 if (!task) 901 return -ESRCH; 902 mm = get_task_mm(task); 903 if (mm) { 904 struct clear_refs_private cp = { 905 .type = type, 906 }; 907 struct mm_walk clear_refs_walk = { 908 .pmd_entry = clear_refs_pte_range, 909 .test_walk = clear_refs_test_walk, 910 .mm = mm, 911 .private = &cp, 912 }; 913 914 if (type == CLEAR_REFS_MM_HIWATER_RSS) { 915 /* 916 * Writing 5 to /proc/pid/clear_refs resets the peak 917 * resident set size to this mm's current rss value. 918 */ 919 down_write(&mm->mmap_sem); 920 reset_mm_hiwater_rss(mm); 921 up_write(&mm->mmap_sem); 922 goto out_mm; 923 } 924 925 down_read(&mm->mmap_sem); 926 if (type == CLEAR_REFS_SOFT_DIRTY) { 927 for (vma = mm->mmap; vma; vma = vma->vm_next) { 928 if (!(vma->vm_flags & VM_SOFTDIRTY)) 929 continue; 930 up_read(&mm->mmap_sem); 931 down_write(&mm->mmap_sem); 932 for (vma = mm->mmap; vma; vma = vma->vm_next) { 933 vma->vm_flags &= ~VM_SOFTDIRTY; 934 vma_set_page_prot(vma); 935 } 936 downgrade_write(&mm->mmap_sem); 937 break; 938 } 939 mmu_notifier_invalidate_range_start(mm, 0, -1); 940 } 941 walk_page_range(0, ~0UL, &clear_refs_walk); 942 if (type == CLEAR_REFS_SOFT_DIRTY) 943 mmu_notifier_invalidate_range_end(mm, 0, -1); 944 flush_tlb_mm(mm); 945 up_read(&mm->mmap_sem); 946 out_mm: 947 mmput(mm); 948 } 949 put_task_struct(task); 950 951 return count; 952 } 953 954 const struct file_operations proc_clear_refs_operations = { 955 .write = clear_refs_write, 956 .llseek = noop_llseek, 957 }; 958 959 typedef struct { 960 u64 pme; 961 } pagemap_entry_t; 962 963 struct pagemapread { 964 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */ 965 pagemap_entry_t *buffer; 966 bool v2; 967 }; 968 969 #define PAGEMAP_WALK_SIZE (PMD_SIZE) 970 #define PAGEMAP_WALK_MASK (PMD_MASK) 971 972 #define PM_ENTRY_BYTES sizeof(pagemap_entry_t) 973 #define PM_STATUS_BITS 3 974 #define PM_STATUS_OFFSET (64 - PM_STATUS_BITS) 975 #define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET) 976 #define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK) 977 #define PM_PSHIFT_BITS 6 978 #define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS) 979 #define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET) 980 #define __PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK) 981 #define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1) 982 #define PM_PFRAME(x) ((x) & PM_PFRAME_MASK) 983 /* in "new" pagemap pshift bits are occupied with more status bits */ 984 #define PM_STATUS2(v2, x) (__PM_PSHIFT(v2 ? x : PAGE_SHIFT)) 985 986 #define __PM_SOFT_DIRTY (1LL) 987 #define PM_PRESENT PM_STATUS(4LL) 988 #define PM_SWAP PM_STATUS(2LL) 989 #define PM_FILE PM_STATUS(1LL) 990 #define PM_NOT_PRESENT(v2) PM_STATUS2(v2, 0) 991 #define PM_END_OF_BUFFER 1 992 993 static inline pagemap_entry_t make_pme(u64 val) 994 { 995 return (pagemap_entry_t) { .pme = val }; 996 } 997 998 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme, 999 struct pagemapread *pm) 1000 { 1001 pm->buffer[pm->pos++] = *pme; 1002 if (pm->pos >= pm->len) 1003 return PM_END_OF_BUFFER; 1004 return 0; 1005 } 1006 1007 static int pagemap_pte_hole(unsigned long start, unsigned long end, 1008 struct mm_walk *walk) 1009 { 1010 struct pagemapread *pm = walk->private; 1011 unsigned long addr = start; 1012 int err = 0; 1013 1014 while (addr < end) { 1015 struct vm_area_struct *vma = find_vma(walk->mm, addr); 1016 pagemap_entry_t pme = make_pme(PM_NOT_PRESENT(pm->v2)); 1017 /* End of address space hole, which we mark as non-present. */ 1018 unsigned long hole_end; 1019 1020 if (vma) 1021 hole_end = min(end, vma->vm_start); 1022 else 1023 hole_end = end; 1024 1025 for (; addr < hole_end; addr += PAGE_SIZE) { 1026 err = add_to_pagemap(addr, &pme, pm); 1027 if (err) 1028 goto out; 1029 } 1030 1031 if (!vma) 1032 break; 1033 1034 /* Addresses in the VMA. */ 1035 if (vma->vm_flags & VM_SOFTDIRTY) 1036 pme.pme |= PM_STATUS2(pm->v2, __PM_SOFT_DIRTY); 1037 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) { 1038 err = add_to_pagemap(addr, &pme, pm); 1039 if (err) 1040 goto out; 1041 } 1042 } 1043 out: 1044 return err; 1045 } 1046 1047 static void pte_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm, 1048 struct vm_area_struct *vma, unsigned long addr, pte_t pte) 1049 { 1050 u64 frame, flags; 1051 struct page *page = NULL; 1052 int flags2 = 0; 1053 1054 if (pte_present(pte)) { 1055 frame = pte_pfn(pte); 1056 flags = PM_PRESENT; 1057 page = vm_normal_page(vma, addr, pte); 1058 if (pte_soft_dirty(pte)) 1059 flags2 |= __PM_SOFT_DIRTY; 1060 } else if (is_swap_pte(pte)) { 1061 swp_entry_t entry; 1062 if (pte_swp_soft_dirty(pte)) 1063 flags2 |= __PM_SOFT_DIRTY; 1064 entry = pte_to_swp_entry(pte); 1065 frame = swp_type(entry) | 1066 (swp_offset(entry) << MAX_SWAPFILES_SHIFT); 1067 flags = PM_SWAP; 1068 if (is_migration_entry(entry)) 1069 page = migration_entry_to_page(entry); 1070 } else { 1071 if (vma->vm_flags & VM_SOFTDIRTY) 1072 flags2 |= __PM_SOFT_DIRTY; 1073 *pme = make_pme(PM_NOT_PRESENT(pm->v2) | PM_STATUS2(pm->v2, flags2)); 1074 return; 1075 } 1076 1077 if (page && !PageAnon(page)) 1078 flags |= PM_FILE; 1079 if ((vma->vm_flags & VM_SOFTDIRTY)) 1080 flags2 |= __PM_SOFT_DIRTY; 1081 1082 *pme = make_pme(PM_PFRAME(frame) | PM_STATUS2(pm->v2, flags2) | flags); 1083 } 1084 1085 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1086 static void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm, 1087 pmd_t pmd, int offset, int pmd_flags2) 1088 { 1089 /* 1090 * Currently pmd for thp is always present because thp can not be 1091 * swapped-out, migrated, or HWPOISONed (split in such cases instead.) 1092 * This if-check is just to prepare for future implementation. 1093 */ 1094 if (pmd_present(pmd)) 1095 *pme = make_pme(PM_PFRAME(pmd_pfn(pmd) + offset) 1096 | PM_STATUS2(pm->v2, pmd_flags2) | PM_PRESENT); 1097 else 1098 *pme = make_pme(PM_NOT_PRESENT(pm->v2) | PM_STATUS2(pm->v2, pmd_flags2)); 1099 } 1100 #else 1101 static inline void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm, 1102 pmd_t pmd, int offset, int pmd_flags2) 1103 { 1104 } 1105 #endif 1106 1107 static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, 1108 struct mm_walk *walk) 1109 { 1110 struct vm_area_struct *vma = walk->vma; 1111 struct pagemapread *pm = walk->private; 1112 spinlock_t *ptl; 1113 pte_t *pte, *orig_pte; 1114 int err = 0; 1115 1116 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { 1117 int pmd_flags2; 1118 1119 if ((vma->vm_flags & VM_SOFTDIRTY) || pmd_soft_dirty(*pmd)) 1120 pmd_flags2 = __PM_SOFT_DIRTY; 1121 else 1122 pmd_flags2 = 0; 1123 1124 for (; addr != end; addr += PAGE_SIZE) { 1125 unsigned long offset; 1126 pagemap_entry_t pme; 1127 1128 offset = (addr & ~PAGEMAP_WALK_MASK) >> 1129 PAGE_SHIFT; 1130 thp_pmd_to_pagemap_entry(&pme, pm, *pmd, offset, pmd_flags2); 1131 err = add_to_pagemap(addr, &pme, pm); 1132 if (err) 1133 break; 1134 } 1135 spin_unlock(ptl); 1136 return err; 1137 } 1138 1139 if (pmd_trans_unstable(pmd)) 1140 return 0; 1141 1142 /* 1143 * We can assume that @vma always points to a valid one and @end never 1144 * goes beyond vma->vm_end. 1145 */ 1146 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl); 1147 for (; addr < end; pte++, addr += PAGE_SIZE) { 1148 pagemap_entry_t pme; 1149 1150 pte_to_pagemap_entry(&pme, pm, vma, addr, *pte); 1151 err = add_to_pagemap(addr, &pme, pm); 1152 if (err) 1153 break; 1154 } 1155 pte_unmap_unlock(orig_pte, ptl); 1156 1157 cond_resched(); 1158 1159 return err; 1160 } 1161 1162 #ifdef CONFIG_HUGETLB_PAGE 1163 static void huge_pte_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm, 1164 pte_t pte, int offset, int flags2) 1165 { 1166 if (pte_present(pte)) 1167 *pme = make_pme(PM_PFRAME(pte_pfn(pte) + offset) | 1168 PM_STATUS2(pm->v2, flags2) | 1169 PM_PRESENT); 1170 else 1171 *pme = make_pme(PM_NOT_PRESENT(pm->v2) | 1172 PM_STATUS2(pm->v2, flags2)); 1173 } 1174 1175 /* This function walks within one hugetlb entry in the single call */ 1176 static int pagemap_hugetlb_range(pte_t *pte, unsigned long hmask, 1177 unsigned long addr, unsigned long end, 1178 struct mm_walk *walk) 1179 { 1180 struct pagemapread *pm = walk->private; 1181 struct vm_area_struct *vma = walk->vma; 1182 int err = 0; 1183 int flags2; 1184 pagemap_entry_t pme; 1185 1186 if (vma->vm_flags & VM_SOFTDIRTY) 1187 flags2 = __PM_SOFT_DIRTY; 1188 else 1189 flags2 = 0; 1190 1191 for (; addr != end; addr += PAGE_SIZE) { 1192 int offset = (addr & ~hmask) >> PAGE_SHIFT; 1193 huge_pte_to_pagemap_entry(&pme, pm, *pte, offset, flags2); 1194 err = add_to_pagemap(addr, &pme, pm); 1195 if (err) 1196 return err; 1197 } 1198 1199 cond_resched(); 1200 1201 return err; 1202 } 1203 #endif /* HUGETLB_PAGE */ 1204 1205 /* 1206 * /proc/pid/pagemap - an array mapping virtual pages to pfns 1207 * 1208 * For each page in the address space, this file contains one 64-bit entry 1209 * consisting of the following: 1210 * 1211 * Bits 0-54 page frame number (PFN) if present 1212 * Bits 0-4 swap type if swapped 1213 * Bits 5-54 swap offset if swapped 1214 * Bits 55-60 page shift (page size = 1<<page shift) 1215 * Bit 61 page is file-page or shared-anon 1216 * Bit 62 page swapped 1217 * Bit 63 page present 1218 * 1219 * If the page is not present but in swap, then the PFN contains an 1220 * encoding of the swap file number and the page's offset into the 1221 * swap. Unmapped pages return a null PFN. This allows determining 1222 * precisely which pages are mapped (or in swap) and comparing mapped 1223 * pages between processes. 1224 * 1225 * Efficient users of this interface will use /proc/pid/maps to 1226 * determine which areas of memory are actually mapped and llseek to 1227 * skip over unmapped regions. 1228 */ 1229 static ssize_t pagemap_read(struct file *file, char __user *buf, 1230 size_t count, loff_t *ppos) 1231 { 1232 struct task_struct *task = get_proc_task(file_inode(file)); 1233 struct mm_struct *mm; 1234 struct pagemapread pm; 1235 int ret = -ESRCH; 1236 struct mm_walk pagemap_walk = {}; 1237 unsigned long src; 1238 unsigned long svpfn; 1239 unsigned long start_vaddr; 1240 unsigned long end_vaddr; 1241 int copied = 0; 1242 1243 if (!task) 1244 goto out; 1245 1246 ret = -EINVAL; 1247 /* file position must be aligned */ 1248 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES)) 1249 goto out_task; 1250 1251 ret = 0; 1252 if (!count) 1253 goto out_task; 1254 1255 pm.v2 = soft_dirty_cleared; 1256 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT); 1257 pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY); 1258 ret = -ENOMEM; 1259 if (!pm.buffer) 1260 goto out_task; 1261 1262 mm = mm_access(task, PTRACE_MODE_READ); 1263 ret = PTR_ERR(mm); 1264 if (!mm || IS_ERR(mm)) 1265 goto out_free; 1266 1267 pagemap_walk.pmd_entry = pagemap_pte_range; 1268 pagemap_walk.pte_hole = pagemap_pte_hole; 1269 #ifdef CONFIG_HUGETLB_PAGE 1270 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range; 1271 #endif 1272 pagemap_walk.mm = mm; 1273 pagemap_walk.private = ± 1274 1275 src = *ppos; 1276 svpfn = src / PM_ENTRY_BYTES; 1277 start_vaddr = svpfn << PAGE_SHIFT; 1278 end_vaddr = TASK_SIZE_OF(task); 1279 1280 /* watch out for wraparound */ 1281 if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT) 1282 start_vaddr = end_vaddr; 1283 1284 /* 1285 * The odds are that this will stop walking way 1286 * before end_vaddr, because the length of the 1287 * user buffer is tracked in "pm", and the walk 1288 * will stop when we hit the end of the buffer. 1289 */ 1290 ret = 0; 1291 while (count && (start_vaddr < end_vaddr)) { 1292 int len; 1293 unsigned long end; 1294 1295 pm.pos = 0; 1296 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK; 1297 /* overflow ? */ 1298 if (end < start_vaddr || end > end_vaddr) 1299 end = end_vaddr; 1300 down_read(&mm->mmap_sem); 1301 ret = walk_page_range(start_vaddr, end, &pagemap_walk); 1302 up_read(&mm->mmap_sem); 1303 start_vaddr = end; 1304 1305 len = min(count, PM_ENTRY_BYTES * pm.pos); 1306 if (copy_to_user(buf, pm.buffer, len)) { 1307 ret = -EFAULT; 1308 goto out_mm; 1309 } 1310 copied += len; 1311 buf += len; 1312 count -= len; 1313 } 1314 *ppos += copied; 1315 if (!ret || ret == PM_END_OF_BUFFER) 1316 ret = copied; 1317 1318 out_mm: 1319 mmput(mm); 1320 out_free: 1321 kfree(pm.buffer); 1322 out_task: 1323 put_task_struct(task); 1324 out: 1325 return ret; 1326 } 1327 1328 static int pagemap_open(struct inode *inode, struct file *file) 1329 { 1330 /* do not disclose physical addresses: attack vector */ 1331 if (!capable(CAP_SYS_ADMIN)) 1332 return -EPERM; 1333 pr_warn_once("Bits 55-60 of /proc/PID/pagemap entries are about " 1334 "to stop being page-shift some time soon. See the " 1335 "linux/Documentation/vm/pagemap.txt for details.\n"); 1336 return 0; 1337 } 1338 1339 const struct file_operations proc_pagemap_operations = { 1340 .llseek = mem_lseek, /* borrow this */ 1341 .read = pagemap_read, 1342 .open = pagemap_open, 1343 }; 1344 #endif /* CONFIG_PROC_PAGE_MONITOR */ 1345 1346 #ifdef CONFIG_NUMA 1347 1348 struct numa_maps { 1349 unsigned long pages; 1350 unsigned long anon; 1351 unsigned long active; 1352 unsigned long writeback; 1353 unsigned long mapcount_max; 1354 unsigned long dirty; 1355 unsigned long swapcache; 1356 unsigned long node[MAX_NUMNODES]; 1357 }; 1358 1359 struct numa_maps_private { 1360 struct proc_maps_private proc_maps; 1361 struct numa_maps md; 1362 }; 1363 1364 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty, 1365 unsigned long nr_pages) 1366 { 1367 int count = page_mapcount(page); 1368 1369 md->pages += nr_pages; 1370 if (pte_dirty || PageDirty(page)) 1371 md->dirty += nr_pages; 1372 1373 if (PageSwapCache(page)) 1374 md->swapcache += nr_pages; 1375 1376 if (PageActive(page) || PageUnevictable(page)) 1377 md->active += nr_pages; 1378 1379 if (PageWriteback(page)) 1380 md->writeback += nr_pages; 1381 1382 if (PageAnon(page)) 1383 md->anon += nr_pages; 1384 1385 if (count > md->mapcount_max) 1386 md->mapcount_max = count; 1387 1388 md->node[page_to_nid(page)] += nr_pages; 1389 } 1390 1391 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma, 1392 unsigned long addr) 1393 { 1394 struct page *page; 1395 int nid; 1396 1397 if (!pte_present(pte)) 1398 return NULL; 1399 1400 page = vm_normal_page(vma, addr, pte); 1401 if (!page) 1402 return NULL; 1403 1404 if (PageReserved(page)) 1405 return NULL; 1406 1407 nid = page_to_nid(page); 1408 if (!node_isset(nid, node_states[N_MEMORY])) 1409 return NULL; 1410 1411 return page; 1412 } 1413 1414 static int gather_pte_stats(pmd_t *pmd, unsigned long addr, 1415 unsigned long end, struct mm_walk *walk) 1416 { 1417 struct numa_maps *md = walk->private; 1418 struct vm_area_struct *vma = walk->vma; 1419 spinlock_t *ptl; 1420 pte_t *orig_pte; 1421 pte_t *pte; 1422 1423 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { 1424 pte_t huge_pte = *(pte_t *)pmd; 1425 struct page *page; 1426 1427 page = can_gather_numa_stats(huge_pte, vma, addr); 1428 if (page) 1429 gather_stats(page, md, pte_dirty(huge_pte), 1430 HPAGE_PMD_SIZE/PAGE_SIZE); 1431 spin_unlock(ptl); 1432 return 0; 1433 } 1434 1435 if (pmd_trans_unstable(pmd)) 1436 return 0; 1437 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl); 1438 do { 1439 struct page *page = can_gather_numa_stats(*pte, vma, addr); 1440 if (!page) 1441 continue; 1442 gather_stats(page, md, pte_dirty(*pte), 1); 1443 1444 } while (pte++, addr += PAGE_SIZE, addr != end); 1445 pte_unmap_unlock(orig_pte, ptl); 1446 return 0; 1447 } 1448 #ifdef CONFIG_HUGETLB_PAGE 1449 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask, 1450 unsigned long addr, unsigned long end, struct mm_walk *walk) 1451 { 1452 struct numa_maps *md; 1453 struct page *page; 1454 1455 if (!pte_present(*pte)) 1456 return 0; 1457 1458 page = pte_page(*pte); 1459 if (!page) 1460 return 0; 1461 1462 md = walk->private; 1463 gather_stats(page, md, pte_dirty(*pte), 1); 1464 return 0; 1465 } 1466 1467 #else 1468 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask, 1469 unsigned long addr, unsigned long end, struct mm_walk *walk) 1470 { 1471 return 0; 1472 } 1473 #endif 1474 1475 /* 1476 * Display pages allocated per node and memory policy via /proc. 1477 */ 1478 static int show_numa_map(struct seq_file *m, void *v, int is_pid) 1479 { 1480 struct numa_maps_private *numa_priv = m->private; 1481 struct proc_maps_private *proc_priv = &numa_priv->proc_maps; 1482 struct vm_area_struct *vma = v; 1483 struct numa_maps *md = &numa_priv->md; 1484 struct file *file = vma->vm_file; 1485 struct mm_struct *mm = vma->vm_mm; 1486 struct mm_walk walk = { 1487 .hugetlb_entry = gather_hugetlb_stats, 1488 .pmd_entry = gather_pte_stats, 1489 .private = md, 1490 .mm = mm, 1491 }; 1492 struct mempolicy *pol; 1493 char buffer[64]; 1494 int nid; 1495 1496 if (!mm) 1497 return 0; 1498 1499 /* Ensure we start with an empty set of numa_maps statistics. */ 1500 memset(md, 0, sizeof(*md)); 1501 1502 pol = __get_vma_policy(vma, vma->vm_start); 1503 if (pol) { 1504 mpol_to_str(buffer, sizeof(buffer), pol); 1505 mpol_cond_put(pol); 1506 } else { 1507 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy); 1508 } 1509 1510 seq_printf(m, "%08lx %s", vma->vm_start, buffer); 1511 1512 if (file) { 1513 seq_puts(m, " file="); 1514 seq_file_path(m, file, "\n\t= "); 1515 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) { 1516 seq_puts(m, " heap"); 1517 } else { 1518 pid_t tid = pid_of_stack(proc_priv, vma, is_pid); 1519 if (tid != 0) { 1520 /* 1521 * Thread stack in /proc/PID/task/TID/maps or 1522 * the main process stack. 1523 */ 1524 if (!is_pid || (vma->vm_start <= mm->start_stack && 1525 vma->vm_end >= mm->start_stack)) 1526 seq_puts(m, " stack"); 1527 else 1528 seq_printf(m, " stack:%d", tid); 1529 } 1530 } 1531 1532 if (is_vm_hugetlb_page(vma)) 1533 seq_puts(m, " huge"); 1534 1535 /* mmap_sem is held by m_start */ 1536 walk_page_vma(vma, &walk); 1537 1538 if (!md->pages) 1539 goto out; 1540 1541 if (md->anon) 1542 seq_printf(m, " anon=%lu", md->anon); 1543 1544 if (md->dirty) 1545 seq_printf(m, " dirty=%lu", md->dirty); 1546 1547 if (md->pages != md->anon && md->pages != md->dirty) 1548 seq_printf(m, " mapped=%lu", md->pages); 1549 1550 if (md->mapcount_max > 1) 1551 seq_printf(m, " mapmax=%lu", md->mapcount_max); 1552 1553 if (md->swapcache) 1554 seq_printf(m, " swapcache=%lu", md->swapcache); 1555 1556 if (md->active < md->pages && !is_vm_hugetlb_page(vma)) 1557 seq_printf(m, " active=%lu", md->active); 1558 1559 if (md->writeback) 1560 seq_printf(m, " writeback=%lu", md->writeback); 1561 1562 for_each_node_state(nid, N_MEMORY) 1563 if (md->node[nid]) 1564 seq_printf(m, " N%d=%lu", nid, md->node[nid]); 1565 1566 seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10); 1567 out: 1568 seq_putc(m, '\n'); 1569 m_cache_vma(m, vma); 1570 return 0; 1571 } 1572 1573 static int show_pid_numa_map(struct seq_file *m, void *v) 1574 { 1575 return show_numa_map(m, v, 1); 1576 } 1577 1578 static int show_tid_numa_map(struct seq_file *m, void *v) 1579 { 1580 return show_numa_map(m, v, 0); 1581 } 1582 1583 static const struct seq_operations proc_pid_numa_maps_op = { 1584 .start = m_start, 1585 .next = m_next, 1586 .stop = m_stop, 1587 .show = show_pid_numa_map, 1588 }; 1589 1590 static const struct seq_operations proc_tid_numa_maps_op = { 1591 .start = m_start, 1592 .next = m_next, 1593 .stop = m_stop, 1594 .show = show_tid_numa_map, 1595 }; 1596 1597 static int numa_maps_open(struct inode *inode, struct file *file, 1598 const struct seq_operations *ops) 1599 { 1600 return proc_maps_open(inode, file, ops, 1601 sizeof(struct numa_maps_private)); 1602 } 1603 1604 static int pid_numa_maps_open(struct inode *inode, struct file *file) 1605 { 1606 return numa_maps_open(inode, file, &proc_pid_numa_maps_op); 1607 } 1608 1609 static int tid_numa_maps_open(struct inode *inode, struct file *file) 1610 { 1611 return numa_maps_open(inode, file, &proc_tid_numa_maps_op); 1612 } 1613 1614 const struct file_operations proc_pid_numa_maps_operations = { 1615 .open = pid_numa_maps_open, 1616 .read = seq_read, 1617 .llseek = seq_lseek, 1618 .release = proc_map_release, 1619 }; 1620 1621 const struct file_operations proc_tid_numa_maps_operations = { 1622 .open = tid_numa_maps_open, 1623 .read = seq_read, 1624 .llseek = seq_lseek, 1625 .release = proc_map_release, 1626 }; 1627 #endif /* CONFIG_NUMA */ 1628