1 #include <linux/mm.h> 2 #include <linux/hugetlb.h> 3 #include <linux/huge_mm.h> 4 #include <linux/mount.h> 5 #include <linux/seq_file.h> 6 #include <linux/highmem.h> 7 #include <linux/ptrace.h> 8 #include <linux/slab.h> 9 #include <linux/pagemap.h> 10 #include <linux/mempolicy.h> 11 #include <linux/rmap.h> 12 #include <linux/swap.h> 13 #include <linux/swapops.h> 14 15 #include <asm/elf.h> 16 #include <asm/uaccess.h> 17 #include <asm/tlbflush.h> 18 #include "internal.h" 19 20 void task_mem(struct seq_file *m, struct mm_struct *mm) 21 { 22 unsigned long data, text, lib, swap; 23 unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss; 24 25 /* 26 * Note: to minimize their overhead, mm maintains hiwater_vm and 27 * hiwater_rss only when about to *lower* total_vm or rss. Any 28 * collector of these hiwater stats must therefore get total_vm 29 * and rss too, which will usually be the higher. Barriers? not 30 * worth the effort, such snapshots can always be inconsistent. 31 */ 32 hiwater_vm = total_vm = mm->total_vm; 33 if (hiwater_vm < mm->hiwater_vm) 34 hiwater_vm = mm->hiwater_vm; 35 hiwater_rss = total_rss = get_mm_rss(mm); 36 if (hiwater_rss < mm->hiwater_rss) 37 hiwater_rss = mm->hiwater_rss; 38 39 data = mm->total_vm - mm->shared_vm - mm->stack_vm; 40 text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10; 41 lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text; 42 swap = get_mm_counter(mm, MM_SWAPENTS); 43 seq_printf(m, 44 "VmPeak:\t%8lu kB\n" 45 "VmSize:\t%8lu kB\n" 46 "VmLck:\t%8lu kB\n" 47 "VmPin:\t%8lu kB\n" 48 "VmHWM:\t%8lu kB\n" 49 "VmRSS:\t%8lu kB\n" 50 "VmData:\t%8lu kB\n" 51 "VmStk:\t%8lu kB\n" 52 "VmExe:\t%8lu kB\n" 53 "VmLib:\t%8lu kB\n" 54 "VmPTE:\t%8lu kB\n" 55 "VmSwap:\t%8lu kB\n", 56 hiwater_vm << (PAGE_SHIFT-10), 57 total_vm << (PAGE_SHIFT-10), 58 mm->locked_vm << (PAGE_SHIFT-10), 59 mm->pinned_vm << (PAGE_SHIFT-10), 60 hiwater_rss << (PAGE_SHIFT-10), 61 total_rss << (PAGE_SHIFT-10), 62 data << (PAGE_SHIFT-10), 63 mm->stack_vm << (PAGE_SHIFT-10), text, lib, 64 (PTRS_PER_PTE*sizeof(pte_t)*mm->nr_ptes) >> 10, 65 swap << (PAGE_SHIFT-10)); 66 } 67 68 unsigned long task_vsize(struct mm_struct *mm) 69 { 70 return PAGE_SIZE * mm->total_vm; 71 } 72 73 unsigned long task_statm(struct mm_struct *mm, 74 unsigned long *shared, unsigned long *text, 75 unsigned long *data, unsigned long *resident) 76 { 77 *shared = get_mm_counter(mm, MM_FILEPAGES); 78 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) 79 >> PAGE_SHIFT; 80 *data = mm->total_vm - mm->shared_vm; 81 *resident = *shared + get_mm_counter(mm, MM_ANONPAGES); 82 return mm->total_vm; 83 } 84 85 static void pad_len_spaces(struct seq_file *m, int len) 86 { 87 len = 25 + sizeof(void*) * 6 - len; 88 if (len < 1) 89 len = 1; 90 seq_printf(m, "%*c", len, ' '); 91 } 92 93 #ifdef CONFIG_NUMA 94 /* 95 * These functions are for numa_maps but called in generic **maps seq_file 96 * ->start(), ->stop() ops. 97 * 98 * numa_maps scans all vmas under mmap_sem and checks their mempolicy. 99 * Each mempolicy object is controlled by reference counting. The problem here 100 * is how to avoid accessing dead mempolicy object. 101 * 102 * Because we're holding mmap_sem while reading seq_file, it's safe to access 103 * each vma's mempolicy, no vma objects will never drop refs to mempolicy. 104 * 105 * A task's mempolicy (task->mempolicy) has different behavior. task->mempolicy 106 * is set and replaced under mmap_sem but unrefed and cleared under task_lock(). 107 * So, without task_lock(), we cannot trust get_vma_policy() because we cannot 108 * gurantee the task never exits under us. But taking task_lock() around 109 * get_vma_plicy() causes lock order problem. 110 * 111 * To access task->mempolicy without lock, we hold a reference count of an 112 * object pointed by task->mempolicy and remember it. This will guarantee 113 * that task->mempolicy points to an alive object or NULL in numa_maps accesses. 114 */ 115 static void hold_task_mempolicy(struct proc_maps_private *priv) 116 { 117 struct task_struct *task = priv->task; 118 119 task_lock(task); 120 priv->task_mempolicy = task->mempolicy; 121 mpol_get(priv->task_mempolicy); 122 task_unlock(task); 123 } 124 static void release_task_mempolicy(struct proc_maps_private *priv) 125 { 126 mpol_put(priv->task_mempolicy); 127 } 128 #else 129 static void hold_task_mempolicy(struct proc_maps_private *priv) 130 { 131 } 132 static void release_task_mempolicy(struct proc_maps_private *priv) 133 { 134 } 135 #endif 136 137 static void vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma) 138 { 139 if (vma && vma != priv->tail_vma) { 140 struct mm_struct *mm = vma->vm_mm; 141 release_task_mempolicy(priv); 142 up_read(&mm->mmap_sem); 143 mmput(mm); 144 } 145 } 146 147 static void *m_start(struct seq_file *m, loff_t *pos) 148 { 149 struct proc_maps_private *priv = m->private; 150 unsigned long last_addr = m->version; 151 struct mm_struct *mm; 152 struct vm_area_struct *vma, *tail_vma = NULL; 153 loff_t l = *pos; 154 155 /* Clear the per syscall fields in priv */ 156 priv->task = NULL; 157 priv->tail_vma = NULL; 158 159 /* 160 * We remember last_addr rather than next_addr to hit with 161 * mmap_cache most of the time. We have zero last_addr at 162 * the beginning and also after lseek. We will have -1 last_addr 163 * after the end of the vmas. 164 */ 165 166 if (last_addr == -1UL) 167 return NULL; 168 169 priv->task = get_pid_task(priv->pid, PIDTYPE_PID); 170 if (!priv->task) 171 return ERR_PTR(-ESRCH); 172 173 mm = mm_access(priv->task, PTRACE_MODE_READ); 174 if (!mm || IS_ERR(mm)) 175 return mm; 176 down_read(&mm->mmap_sem); 177 178 tail_vma = get_gate_vma(priv->task->mm); 179 priv->tail_vma = tail_vma; 180 hold_task_mempolicy(priv); 181 /* Start with last addr hint */ 182 vma = find_vma(mm, last_addr); 183 if (last_addr && vma) { 184 vma = vma->vm_next; 185 goto out; 186 } 187 188 /* 189 * Check the vma index is within the range and do 190 * sequential scan until m_index. 191 */ 192 vma = NULL; 193 if ((unsigned long)l < mm->map_count) { 194 vma = mm->mmap; 195 while (l-- && vma) 196 vma = vma->vm_next; 197 goto out; 198 } 199 200 if (l != mm->map_count) 201 tail_vma = NULL; /* After gate vma */ 202 203 out: 204 if (vma) 205 return vma; 206 207 release_task_mempolicy(priv); 208 /* End of vmas has been reached */ 209 m->version = (tail_vma != NULL)? 0: -1UL; 210 up_read(&mm->mmap_sem); 211 mmput(mm); 212 return tail_vma; 213 } 214 215 static void *m_next(struct seq_file *m, void *v, loff_t *pos) 216 { 217 struct proc_maps_private *priv = m->private; 218 struct vm_area_struct *vma = v; 219 struct vm_area_struct *tail_vma = priv->tail_vma; 220 221 (*pos)++; 222 if (vma && (vma != tail_vma) && vma->vm_next) 223 return vma->vm_next; 224 vma_stop(priv, vma); 225 return (vma != tail_vma)? tail_vma: NULL; 226 } 227 228 static void m_stop(struct seq_file *m, void *v) 229 { 230 struct proc_maps_private *priv = m->private; 231 struct vm_area_struct *vma = v; 232 233 if (!IS_ERR(vma)) 234 vma_stop(priv, vma); 235 if (priv->task) 236 put_task_struct(priv->task); 237 } 238 239 static int do_maps_open(struct inode *inode, struct file *file, 240 const struct seq_operations *ops) 241 { 242 struct proc_maps_private *priv; 243 int ret = -ENOMEM; 244 priv = kzalloc(sizeof(*priv), GFP_KERNEL); 245 if (priv) { 246 priv->pid = proc_pid(inode); 247 ret = seq_open(file, ops); 248 if (!ret) { 249 struct seq_file *m = file->private_data; 250 m->private = priv; 251 } else { 252 kfree(priv); 253 } 254 } 255 return ret; 256 } 257 258 static void 259 show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid) 260 { 261 struct mm_struct *mm = vma->vm_mm; 262 struct file *file = vma->vm_file; 263 struct proc_maps_private *priv = m->private; 264 struct task_struct *task = priv->task; 265 vm_flags_t flags = vma->vm_flags; 266 unsigned long ino = 0; 267 unsigned long long pgoff = 0; 268 unsigned long start, end; 269 dev_t dev = 0; 270 int len; 271 const char *name = NULL; 272 273 if (file) { 274 struct inode *inode = vma->vm_file->f_path.dentry->d_inode; 275 dev = inode->i_sb->s_dev; 276 ino = inode->i_ino; 277 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT; 278 } 279 280 /* We don't show the stack guard page in /proc/maps */ 281 start = vma->vm_start; 282 if (stack_guard_page_start(vma, start)) 283 start += PAGE_SIZE; 284 end = vma->vm_end; 285 if (stack_guard_page_end(vma, end)) 286 end -= PAGE_SIZE; 287 288 seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu %n", 289 start, 290 end, 291 flags & VM_READ ? 'r' : '-', 292 flags & VM_WRITE ? 'w' : '-', 293 flags & VM_EXEC ? 'x' : '-', 294 flags & VM_MAYSHARE ? 's' : 'p', 295 pgoff, 296 MAJOR(dev), MINOR(dev), ino, &len); 297 298 /* 299 * Print the dentry name for named mappings, and a 300 * special [heap] marker for the heap: 301 */ 302 if (file) { 303 pad_len_spaces(m, len); 304 seq_path(m, &file->f_path, "\n"); 305 goto done; 306 } 307 308 name = arch_vma_name(vma); 309 if (!name) { 310 pid_t tid; 311 312 if (!mm) { 313 name = "[vdso]"; 314 goto done; 315 } 316 317 if (vma->vm_start <= mm->brk && 318 vma->vm_end >= mm->start_brk) { 319 name = "[heap]"; 320 goto done; 321 } 322 323 tid = vm_is_stack(task, vma, is_pid); 324 325 if (tid != 0) { 326 /* 327 * Thread stack in /proc/PID/task/TID/maps or 328 * the main process stack. 329 */ 330 if (!is_pid || (vma->vm_start <= mm->start_stack && 331 vma->vm_end >= mm->start_stack)) { 332 name = "[stack]"; 333 } else { 334 /* Thread stack in /proc/PID/maps */ 335 pad_len_spaces(m, len); 336 seq_printf(m, "[stack:%d]", tid); 337 } 338 } 339 } 340 341 done: 342 if (name) { 343 pad_len_spaces(m, len); 344 seq_puts(m, name); 345 } 346 seq_putc(m, '\n'); 347 } 348 349 static int show_map(struct seq_file *m, void *v, int is_pid) 350 { 351 struct vm_area_struct *vma = v; 352 struct proc_maps_private *priv = m->private; 353 struct task_struct *task = priv->task; 354 355 show_map_vma(m, vma, is_pid); 356 357 if (m->count < m->size) /* vma is copied successfully */ 358 m->version = (vma != get_gate_vma(task->mm)) 359 ? vma->vm_start : 0; 360 return 0; 361 } 362 363 static int show_pid_map(struct seq_file *m, void *v) 364 { 365 return show_map(m, v, 1); 366 } 367 368 static int show_tid_map(struct seq_file *m, void *v) 369 { 370 return show_map(m, v, 0); 371 } 372 373 static const struct seq_operations proc_pid_maps_op = { 374 .start = m_start, 375 .next = m_next, 376 .stop = m_stop, 377 .show = show_pid_map 378 }; 379 380 static const struct seq_operations proc_tid_maps_op = { 381 .start = m_start, 382 .next = m_next, 383 .stop = m_stop, 384 .show = show_tid_map 385 }; 386 387 static int pid_maps_open(struct inode *inode, struct file *file) 388 { 389 return do_maps_open(inode, file, &proc_pid_maps_op); 390 } 391 392 static int tid_maps_open(struct inode *inode, struct file *file) 393 { 394 return do_maps_open(inode, file, &proc_tid_maps_op); 395 } 396 397 const struct file_operations proc_pid_maps_operations = { 398 .open = pid_maps_open, 399 .read = seq_read, 400 .llseek = seq_lseek, 401 .release = seq_release_private, 402 }; 403 404 const struct file_operations proc_tid_maps_operations = { 405 .open = tid_maps_open, 406 .read = seq_read, 407 .llseek = seq_lseek, 408 .release = seq_release_private, 409 }; 410 411 /* 412 * Proportional Set Size(PSS): my share of RSS. 413 * 414 * PSS of a process is the count of pages it has in memory, where each 415 * page is divided by the number of processes sharing it. So if a 416 * process has 1000 pages all to itself, and 1000 shared with one other 417 * process, its PSS will be 1500. 418 * 419 * To keep (accumulated) division errors low, we adopt a 64bit 420 * fixed-point pss counter to minimize division errors. So (pss >> 421 * PSS_SHIFT) would be the real byte count. 422 * 423 * A shift of 12 before division means (assuming 4K page size): 424 * - 1M 3-user-pages add up to 8KB errors; 425 * - supports mapcount up to 2^24, or 16M; 426 * - supports PSS up to 2^52 bytes, or 4PB. 427 */ 428 #define PSS_SHIFT 12 429 430 #ifdef CONFIG_PROC_PAGE_MONITOR 431 struct mem_size_stats { 432 struct vm_area_struct *vma; 433 unsigned long resident; 434 unsigned long shared_clean; 435 unsigned long shared_dirty; 436 unsigned long private_clean; 437 unsigned long private_dirty; 438 unsigned long referenced; 439 unsigned long anonymous; 440 unsigned long anonymous_thp; 441 unsigned long swap; 442 unsigned long nonlinear; 443 u64 pss; 444 }; 445 446 447 static void smaps_pte_entry(pte_t ptent, unsigned long addr, 448 unsigned long ptent_size, struct mm_walk *walk) 449 { 450 struct mem_size_stats *mss = walk->private; 451 struct vm_area_struct *vma = mss->vma; 452 pgoff_t pgoff = linear_page_index(vma, addr); 453 struct page *page = NULL; 454 int mapcount; 455 456 if (pte_present(ptent)) { 457 page = vm_normal_page(vma, addr, ptent); 458 } else if (is_swap_pte(ptent)) { 459 swp_entry_t swpent = pte_to_swp_entry(ptent); 460 461 if (!non_swap_entry(swpent)) 462 mss->swap += ptent_size; 463 else if (is_migration_entry(swpent)) 464 page = migration_entry_to_page(swpent); 465 } else if (pte_file(ptent)) { 466 if (pte_to_pgoff(ptent) != pgoff) 467 mss->nonlinear += ptent_size; 468 } 469 470 if (!page) 471 return; 472 473 if (PageAnon(page)) 474 mss->anonymous += ptent_size; 475 476 if (page->index != pgoff) 477 mss->nonlinear += ptent_size; 478 479 mss->resident += ptent_size; 480 /* Accumulate the size in pages that have been accessed. */ 481 if (pte_young(ptent) || PageReferenced(page)) 482 mss->referenced += ptent_size; 483 mapcount = page_mapcount(page); 484 if (mapcount >= 2) { 485 if (pte_dirty(ptent) || PageDirty(page)) 486 mss->shared_dirty += ptent_size; 487 else 488 mss->shared_clean += ptent_size; 489 mss->pss += (ptent_size << PSS_SHIFT) / mapcount; 490 } else { 491 if (pte_dirty(ptent) || PageDirty(page)) 492 mss->private_dirty += ptent_size; 493 else 494 mss->private_clean += ptent_size; 495 mss->pss += (ptent_size << PSS_SHIFT); 496 } 497 } 498 499 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, 500 struct mm_walk *walk) 501 { 502 struct mem_size_stats *mss = walk->private; 503 struct vm_area_struct *vma = mss->vma; 504 pte_t *pte; 505 spinlock_t *ptl; 506 507 if (pmd_trans_huge_lock(pmd, vma) == 1) { 508 smaps_pte_entry(*(pte_t *)pmd, addr, HPAGE_PMD_SIZE, walk); 509 spin_unlock(&walk->mm->page_table_lock); 510 mss->anonymous_thp += HPAGE_PMD_SIZE; 511 return 0; 512 } 513 514 if (pmd_trans_unstable(pmd)) 515 return 0; 516 /* 517 * The mmap_sem held all the way back in m_start() is what 518 * keeps khugepaged out of here and from collapsing things 519 * in here. 520 */ 521 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); 522 for (; addr != end; pte++, addr += PAGE_SIZE) 523 smaps_pte_entry(*pte, addr, PAGE_SIZE, walk); 524 pte_unmap_unlock(pte - 1, ptl); 525 cond_resched(); 526 return 0; 527 } 528 529 static int show_smap(struct seq_file *m, void *v, int is_pid) 530 { 531 struct proc_maps_private *priv = m->private; 532 struct task_struct *task = priv->task; 533 struct vm_area_struct *vma = v; 534 struct mem_size_stats mss; 535 struct mm_walk smaps_walk = { 536 .pmd_entry = smaps_pte_range, 537 .mm = vma->vm_mm, 538 .private = &mss, 539 }; 540 541 memset(&mss, 0, sizeof mss); 542 mss.vma = vma; 543 /* mmap_sem is held in m_start */ 544 if (vma->vm_mm && !is_vm_hugetlb_page(vma)) 545 walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk); 546 547 show_map_vma(m, vma, is_pid); 548 549 seq_printf(m, 550 "Size: %8lu kB\n" 551 "Rss: %8lu kB\n" 552 "Pss: %8lu kB\n" 553 "Shared_Clean: %8lu kB\n" 554 "Shared_Dirty: %8lu kB\n" 555 "Private_Clean: %8lu kB\n" 556 "Private_Dirty: %8lu kB\n" 557 "Referenced: %8lu kB\n" 558 "Anonymous: %8lu kB\n" 559 "AnonHugePages: %8lu kB\n" 560 "Swap: %8lu kB\n" 561 "KernelPageSize: %8lu kB\n" 562 "MMUPageSize: %8lu kB\n" 563 "Locked: %8lu kB\n", 564 (vma->vm_end - vma->vm_start) >> 10, 565 mss.resident >> 10, 566 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)), 567 mss.shared_clean >> 10, 568 mss.shared_dirty >> 10, 569 mss.private_clean >> 10, 570 mss.private_dirty >> 10, 571 mss.referenced >> 10, 572 mss.anonymous >> 10, 573 mss.anonymous_thp >> 10, 574 mss.swap >> 10, 575 vma_kernel_pagesize(vma) >> 10, 576 vma_mmu_pagesize(vma) >> 10, 577 (vma->vm_flags & VM_LOCKED) ? 578 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0); 579 580 if (vma->vm_flags & VM_NONLINEAR) 581 seq_printf(m, "Nonlinear: %8lu kB\n", 582 mss.nonlinear >> 10); 583 584 if (m->count < m->size) /* vma is copied successfully */ 585 m->version = (vma != get_gate_vma(task->mm)) 586 ? vma->vm_start : 0; 587 return 0; 588 } 589 590 static int show_pid_smap(struct seq_file *m, void *v) 591 { 592 return show_smap(m, v, 1); 593 } 594 595 static int show_tid_smap(struct seq_file *m, void *v) 596 { 597 return show_smap(m, v, 0); 598 } 599 600 static const struct seq_operations proc_pid_smaps_op = { 601 .start = m_start, 602 .next = m_next, 603 .stop = m_stop, 604 .show = show_pid_smap 605 }; 606 607 static const struct seq_operations proc_tid_smaps_op = { 608 .start = m_start, 609 .next = m_next, 610 .stop = m_stop, 611 .show = show_tid_smap 612 }; 613 614 static int pid_smaps_open(struct inode *inode, struct file *file) 615 { 616 return do_maps_open(inode, file, &proc_pid_smaps_op); 617 } 618 619 static int tid_smaps_open(struct inode *inode, struct file *file) 620 { 621 return do_maps_open(inode, file, &proc_tid_smaps_op); 622 } 623 624 const struct file_operations proc_pid_smaps_operations = { 625 .open = pid_smaps_open, 626 .read = seq_read, 627 .llseek = seq_lseek, 628 .release = seq_release_private, 629 }; 630 631 const struct file_operations proc_tid_smaps_operations = { 632 .open = tid_smaps_open, 633 .read = seq_read, 634 .llseek = seq_lseek, 635 .release = seq_release_private, 636 }; 637 638 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr, 639 unsigned long end, struct mm_walk *walk) 640 { 641 struct vm_area_struct *vma = walk->private; 642 pte_t *pte, ptent; 643 spinlock_t *ptl; 644 struct page *page; 645 646 split_huge_page_pmd(walk->mm, pmd); 647 if (pmd_trans_unstable(pmd)) 648 return 0; 649 650 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); 651 for (; addr != end; pte++, addr += PAGE_SIZE) { 652 ptent = *pte; 653 if (!pte_present(ptent)) 654 continue; 655 656 page = vm_normal_page(vma, addr, ptent); 657 if (!page) 658 continue; 659 660 /* Clear accessed and referenced bits. */ 661 ptep_test_and_clear_young(vma, addr, pte); 662 ClearPageReferenced(page); 663 } 664 pte_unmap_unlock(pte - 1, ptl); 665 cond_resched(); 666 return 0; 667 } 668 669 #define CLEAR_REFS_ALL 1 670 #define CLEAR_REFS_ANON 2 671 #define CLEAR_REFS_MAPPED 3 672 673 static ssize_t clear_refs_write(struct file *file, const char __user *buf, 674 size_t count, loff_t *ppos) 675 { 676 struct task_struct *task; 677 char buffer[PROC_NUMBUF]; 678 struct mm_struct *mm; 679 struct vm_area_struct *vma; 680 int type; 681 int rv; 682 683 memset(buffer, 0, sizeof(buffer)); 684 if (count > sizeof(buffer) - 1) 685 count = sizeof(buffer) - 1; 686 if (copy_from_user(buffer, buf, count)) 687 return -EFAULT; 688 rv = kstrtoint(strstrip(buffer), 10, &type); 689 if (rv < 0) 690 return rv; 691 if (type < CLEAR_REFS_ALL || type > CLEAR_REFS_MAPPED) 692 return -EINVAL; 693 task = get_proc_task(file->f_path.dentry->d_inode); 694 if (!task) 695 return -ESRCH; 696 mm = get_task_mm(task); 697 if (mm) { 698 struct mm_walk clear_refs_walk = { 699 .pmd_entry = clear_refs_pte_range, 700 .mm = mm, 701 }; 702 down_read(&mm->mmap_sem); 703 for (vma = mm->mmap; vma; vma = vma->vm_next) { 704 clear_refs_walk.private = vma; 705 if (is_vm_hugetlb_page(vma)) 706 continue; 707 /* 708 * Writing 1 to /proc/pid/clear_refs affects all pages. 709 * 710 * Writing 2 to /proc/pid/clear_refs only affects 711 * Anonymous pages. 712 * 713 * Writing 3 to /proc/pid/clear_refs only affects file 714 * mapped pages. 715 */ 716 if (type == CLEAR_REFS_ANON && vma->vm_file) 717 continue; 718 if (type == CLEAR_REFS_MAPPED && !vma->vm_file) 719 continue; 720 walk_page_range(vma->vm_start, vma->vm_end, 721 &clear_refs_walk); 722 } 723 flush_tlb_mm(mm); 724 up_read(&mm->mmap_sem); 725 mmput(mm); 726 } 727 put_task_struct(task); 728 729 return count; 730 } 731 732 const struct file_operations proc_clear_refs_operations = { 733 .write = clear_refs_write, 734 .llseek = noop_llseek, 735 }; 736 737 typedef struct { 738 u64 pme; 739 } pagemap_entry_t; 740 741 struct pagemapread { 742 int pos, len; 743 pagemap_entry_t *buffer; 744 }; 745 746 #define PAGEMAP_WALK_SIZE (PMD_SIZE) 747 #define PAGEMAP_WALK_MASK (PMD_MASK) 748 749 #define PM_ENTRY_BYTES sizeof(u64) 750 #define PM_STATUS_BITS 3 751 #define PM_STATUS_OFFSET (64 - PM_STATUS_BITS) 752 #define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET) 753 #define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK) 754 #define PM_PSHIFT_BITS 6 755 #define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS) 756 #define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET) 757 #define PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK) 758 #define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1) 759 #define PM_PFRAME(x) ((x) & PM_PFRAME_MASK) 760 761 #define PM_PRESENT PM_STATUS(4LL) 762 #define PM_SWAP PM_STATUS(2LL) 763 #define PM_FILE PM_STATUS(1LL) 764 #define PM_NOT_PRESENT PM_PSHIFT(PAGE_SHIFT) 765 #define PM_END_OF_BUFFER 1 766 767 static inline pagemap_entry_t make_pme(u64 val) 768 { 769 return (pagemap_entry_t) { .pme = val }; 770 } 771 772 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme, 773 struct pagemapread *pm) 774 { 775 pm->buffer[pm->pos++] = *pme; 776 if (pm->pos >= pm->len) 777 return PM_END_OF_BUFFER; 778 return 0; 779 } 780 781 static int pagemap_pte_hole(unsigned long start, unsigned long end, 782 struct mm_walk *walk) 783 { 784 struct pagemapread *pm = walk->private; 785 unsigned long addr; 786 int err = 0; 787 pagemap_entry_t pme = make_pme(PM_NOT_PRESENT); 788 789 for (addr = start; addr < end; addr += PAGE_SIZE) { 790 err = add_to_pagemap(addr, &pme, pm); 791 if (err) 792 break; 793 } 794 return err; 795 } 796 797 static void pte_to_pagemap_entry(pagemap_entry_t *pme, 798 struct vm_area_struct *vma, unsigned long addr, pte_t pte) 799 { 800 u64 frame, flags; 801 struct page *page = NULL; 802 803 if (pte_present(pte)) { 804 frame = pte_pfn(pte); 805 flags = PM_PRESENT; 806 page = vm_normal_page(vma, addr, pte); 807 } else if (is_swap_pte(pte)) { 808 swp_entry_t entry = pte_to_swp_entry(pte); 809 810 frame = swp_type(entry) | 811 (swp_offset(entry) << MAX_SWAPFILES_SHIFT); 812 flags = PM_SWAP; 813 if (is_migration_entry(entry)) 814 page = migration_entry_to_page(entry); 815 } else { 816 *pme = make_pme(PM_NOT_PRESENT); 817 return; 818 } 819 820 if (page && !PageAnon(page)) 821 flags |= PM_FILE; 822 823 *pme = make_pme(PM_PFRAME(frame) | PM_PSHIFT(PAGE_SHIFT) | flags); 824 } 825 826 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 827 static void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme, 828 pmd_t pmd, int offset) 829 { 830 /* 831 * Currently pmd for thp is always present because thp can not be 832 * swapped-out, migrated, or HWPOISONed (split in such cases instead.) 833 * This if-check is just to prepare for future implementation. 834 */ 835 if (pmd_present(pmd)) 836 *pme = make_pme(PM_PFRAME(pmd_pfn(pmd) + offset) 837 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT); 838 else 839 *pme = make_pme(PM_NOT_PRESENT); 840 } 841 #else 842 static inline void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme, 843 pmd_t pmd, int offset) 844 { 845 } 846 #endif 847 848 static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, 849 struct mm_walk *walk) 850 { 851 struct vm_area_struct *vma; 852 struct pagemapread *pm = walk->private; 853 pte_t *pte; 854 int err = 0; 855 pagemap_entry_t pme = make_pme(PM_NOT_PRESENT); 856 857 /* find the first VMA at or above 'addr' */ 858 vma = find_vma(walk->mm, addr); 859 if (vma && pmd_trans_huge_lock(pmd, vma) == 1) { 860 for (; addr != end; addr += PAGE_SIZE) { 861 unsigned long offset; 862 863 offset = (addr & ~PAGEMAP_WALK_MASK) >> 864 PAGE_SHIFT; 865 thp_pmd_to_pagemap_entry(&pme, *pmd, offset); 866 err = add_to_pagemap(addr, &pme, pm); 867 if (err) 868 break; 869 } 870 spin_unlock(&walk->mm->page_table_lock); 871 return err; 872 } 873 874 if (pmd_trans_unstable(pmd)) 875 return 0; 876 for (; addr != end; addr += PAGE_SIZE) { 877 878 /* check to see if we've left 'vma' behind 879 * and need a new, higher one */ 880 if (vma && (addr >= vma->vm_end)) { 881 vma = find_vma(walk->mm, addr); 882 pme = make_pme(PM_NOT_PRESENT); 883 } 884 885 /* check that 'vma' actually covers this address, 886 * and that it isn't a huge page vma */ 887 if (vma && (vma->vm_start <= addr) && 888 !is_vm_hugetlb_page(vma)) { 889 pte = pte_offset_map(pmd, addr); 890 pte_to_pagemap_entry(&pme, vma, addr, *pte); 891 /* unmap before userspace copy */ 892 pte_unmap(pte); 893 } 894 err = add_to_pagemap(addr, &pme, pm); 895 if (err) 896 return err; 897 } 898 899 cond_resched(); 900 901 return err; 902 } 903 904 #ifdef CONFIG_HUGETLB_PAGE 905 static void huge_pte_to_pagemap_entry(pagemap_entry_t *pme, 906 pte_t pte, int offset) 907 { 908 if (pte_present(pte)) 909 *pme = make_pme(PM_PFRAME(pte_pfn(pte) + offset) 910 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT); 911 else 912 *pme = make_pme(PM_NOT_PRESENT); 913 } 914 915 /* This function walks within one hugetlb entry in the single call */ 916 static int pagemap_hugetlb_range(pte_t *pte, unsigned long hmask, 917 unsigned long addr, unsigned long end, 918 struct mm_walk *walk) 919 { 920 struct pagemapread *pm = walk->private; 921 int err = 0; 922 pagemap_entry_t pme; 923 924 for (; addr != end; addr += PAGE_SIZE) { 925 int offset = (addr & ~hmask) >> PAGE_SHIFT; 926 huge_pte_to_pagemap_entry(&pme, *pte, offset); 927 err = add_to_pagemap(addr, &pme, pm); 928 if (err) 929 return err; 930 } 931 932 cond_resched(); 933 934 return err; 935 } 936 #endif /* HUGETLB_PAGE */ 937 938 /* 939 * /proc/pid/pagemap - an array mapping virtual pages to pfns 940 * 941 * For each page in the address space, this file contains one 64-bit entry 942 * consisting of the following: 943 * 944 * Bits 0-54 page frame number (PFN) if present 945 * Bits 0-4 swap type if swapped 946 * Bits 5-54 swap offset if swapped 947 * Bits 55-60 page shift (page size = 1<<page shift) 948 * Bit 61 page is file-page or shared-anon 949 * Bit 62 page swapped 950 * Bit 63 page present 951 * 952 * If the page is not present but in swap, then the PFN contains an 953 * encoding of the swap file number and the page's offset into the 954 * swap. Unmapped pages return a null PFN. This allows determining 955 * precisely which pages are mapped (or in swap) and comparing mapped 956 * pages between processes. 957 * 958 * Efficient users of this interface will use /proc/pid/maps to 959 * determine which areas of memory are actually mapped and llseek to 960 * skip over unmapped regions. 961 */ 962 static ssize_t pagemap_read(struct file *file, char __user *buf, 963 size_t count, loff_t *ppos) 964 { 965 struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode); 966 struct mm_struct *mm; 967 struct pagemapread pm; 968 int ret = -ESRCH; 969 struct mm_walk pagemap_walk = {}; 970 unsigned long src; 971 unsigned long svpfn; 972 unsigned long start_vaddr; 973 unsigned long end_vaddr; 974 int copied = 0; 975 976 if (!task) 977 goto out; 978 979 ret = -EINVAL; 980 /* file position must be aligned */ 981 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES)) 982 goto out_task; 983 984 ret = 0; 985 if (!count) 986 goto out_task; 987 988 pm.len = PM_ENTRY_BYTES * (PAGEMAP_WALK_SIZE >> PAGE_SHIFT); 989 pm.buffer = kmalloc(pm.len, GFP_TEMPORARY); 990 ret = -ENOMEM; 991 if (!pm.buffer) 992 goto out_task; 993 994 mm = mm_access(task, PTRACE_MODE_READ); 995 ret = PTR_ERR(mm); 996 if (!mm || IS_ERR(mm)) 997 goto out_free; 998 999 pagemap_walk.pmd_entry = pagemap_pte_range; 1000 pagemap_walk.pte_hole = pagemap_pte_hole; 1001 #ifdef CONFIG_HUGETLB_PAGE 1002 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range; 1003 #endif 1004 pagemap_walk.mm = mm; 1005 pagemap_walk.private = ± 1006 1007 src = *ppos; 1008 svpfn = src / PM_ENTRY_BYTES; 1009 start_vaddr = svpfn << PAGE_SHIFT; 1010 end_vaddr = TASK_SIZE_OF(task); 1011 1012 /* watch out for wraparound */ 1013 if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT) 1014 start_vaddr = end_vaddr; 1015 1016 /* 1017 * The odds are that this will stop walking way 1018 * before end_vaddr, because the length of the 1019 * user buffer is tracked in "pm", and the walk 1020 * will stop when we hit the end of the buffer. 1021 */ 1022 ret = 0; 1023 while (count && (start_vaddr < end_vaddr)) { 1024 int len; 1025 unsigned long end; 1026 1027 pm.pos = 0; 1028 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK; 1029 /* overflow ? */ 1030 if (end < start_vaddr || end > end_vaddr) 1031 end = end_vaddr; 1032 down_read(&mm->mmap_sem); 1033 ret = walk_page_range(start_vaddr, end, &pagemap_walk); 1034 up_read(&mm->mmap_sem); 1035 start_vaddr = end; 1036 1037 len = min(count, PM_ENTRY_BYTES * pm.pos); 1038 if (copy_to_user(buf, pm.buffer, len)) { 1039 ret = -EFAULT; 1040 goto out_mm; 1041 } 1042 copied += len; 1043 buf += len; 1044 count -= len; 1045 } 1046 *ppos += copied; 1047 if (!ret || ret == PM_END_OF_BUFFER) 1048 ret = copied; 1049 1050 out_mm: 1051 mmput(mm); 1052 out_free: 1053 kfree(pm.buffer); 1054 out_task: 1055 put_task_struct(task); 1056 out: 1057 return ret; 1058 } 1059 1060 const struct file_operations proc_pagemap_operations = { 1061 .llseek = mem_lseek, /* borrow this */ 1062 .read = pagemap_read, 1063 }; 1064 #endif /* CONFIG_PROC_PAGE_MONITOR */ 1065 1066 #ifdef CONFIG_NUMA 1067 1068 struct numa_maps { 1069 struct vm_area_struct *vma; 1070 unsigned long pages; 1071 unsigned long anon; 1072 unsigned long active; 1073 unsigned long writeback; 1074 unsigned long mapcount_max; 1075 unsigned long dirty; 1076 unsigned long swapcache; 1077 unsigned long node[MAX_NUMNODES]; 1078 }; 1079 1080 struct numa_maps_private { 1081 struct proc_maps_private proc_maps; 1082 struct numa_maps md; 1083 }; 1084 1085 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty, 1086 unsigned long nr_pages) 1087 { 1088 int count = page_mapcount(page); 1089 1090 md->pages += nr_pages; 1091 if (pte_dirty || PageDirty(page)) 1092 md->dirty += nr_pages; 1093 1094 if (PageSwapCache(page)) 1095 md->swapcache += nr_pages; 1096 1097 if (PageActive(page) || PageUnevictable(page)) 1098 md->active += nr_pages; 1099 1100 if (PageWriteback(page)) 1101 md->writeback += nr_pages; 1102 1103 if (PageAnon(page)) 1104 md->anon += nr_pages; 1105 1106 if (count > md->mapcount_max) 1107 md->mapcount_max = count; 1108 1109 md->node[page_to_nid(page)] += nr_pages; 1110 } 1111 1112 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma, 1113 unsigned long addr) 1114 { 1115 struct page *page; 1116 int nid; 1117 1118 if (!pte_present(pte)) 1119 return NULL; 1120 1121 page = vm_normal_page(vma, addr, pte); 1122 if (!page) 1123 return NULL; 1124 1125 if (PageReserved(page)) 1126 return NULL; 1127 1128 nid = page_to_nid(page); 1129 if (!node_isset(nid, node_states[N_HIGH_MEMORY])) 1130 return NULL; 1131 1132 return page; 1133 } 1134 1135 static int gather_pte_stats(pmd_t *pmd, unsigned long addr, 1136 unsigned long end, struct mm_walk *walk) 1137 { 1138 struct numa_maps *md; 1139 spinlock_t *ptl; 1140 pte_t *orig_pte; 1141 pte_t *pte; 1142 1143 md = walk->private; 1144 1145 if (pmd_trans_huge_lock(pmd, md->vma) == 1) { 1146 pte_t huge_pte = *(pte_t *)pmd; 1147 struct page *page; 1148 1149 page = can_gather_numa_stats(huge_pte, md->vma, addr); 1150 if (page) 1151 gather_stats(page, md, pte_dirty(huge_pte), 1152 HPAGE_PMD_SIZE/PAGE_SIZE); 1153 spin_unlock(&walk->mm->page_table_lock); 1154 return 0; 1155 } 1156 1157 if (pmd_trans_unstable(pmd)) 1158 return 0; 1159 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl); 1160 do { 1161 struct page *page = can_gather_numa_stats(*pte, md->vma, addr); 1162 if (!page) 1163 continue; 1164 gather_stats(page, md, pte_dirty(*pte), 1); 1165 1166 } while (pte++, addr += PAGE_SIZE, addr != end); 1167 pte_unmap_unlock(orig_pte, ptl); 1168 return 0; 1169 } 1170 #ifdef CONFIG_HUGETLB_PAGE 1171 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask, 1172 unsigned long addr, unsigned long end, struct mm_walk *walk) 1173 { 1174 struct numa_maps *md; 1175 struct page *page; 1176 1177 if (pte_none(*pte)) 1178 return 0; 1179 1180 page = pte_page(*pte); 1181 if (!page) 1182 return 0; 1183 1184 md = walk->private; 1185 gather_stats(page, md, pte_dirty(*pte), 1); 1186 return 0; 1187 } 1188 1189 #else 1190 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask, 1191 unsigned long addr, unsigned long end, struct mm_walk *walk) 1192 { 1193 return 0; 1194 } 1195 #endif 1196 1197 /* 1198 * Display pages allocated per node and memory policy via /proc. 1199 */ 1200 static int show_numa_map(struct seq_file *m, void *v, int is_pid) 1201 { 1202 struct numa_maps_private *numa_priv = m->private; 1203 struct proc_maps_private *proc_priv = &numa_priv->proc_maps; 1204 struct vm_area_struct *vma = v; 1205 struct numa_maps *md = &numa_priv->md; 1206 struct file *file = vma->vm_file; 1207 struct task_struct *task = proc_priv->task; 1208 struct mm_struct *mm = vma->vm_mm; 1209 struct mm_walk walk = {}; 1210 struct mempolicy *pol; 1211 int n; 1212 char buffer[50]; 1213 1214 if (!mm) 1215 return 0; 1216 1217 /* Ensure we start with an empty set of numa_maps statistics. */ 1218 memset(md, 0, sizeof(*md)); 1219 1220 md->vma = vma; 1221 1222 walk.hugetlb_entry = gather_hugetbl_stats; 1223 walk.pmd_entry = gather_pte_stats; 1224 walk.private = md; 1225 walk.mm = mm; 1226 1227 pol = get_vma_policy(task, vma, vma->vm_start); 1228 mpol_to_str(buffer, sizeof(buffer), pol, 0); 1229 mpol_cond_put(pol); 1230 1231 seq_printf(m, "%08lx %s", vma->vm_start, buffer); 1232 1233 if (file) { 1234 seq_printf(m, " file="); 1235 seq_path(m, &file->f_path, "\n\t= "); 1236 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) { 1237 seq_printf(m, " heap"); 1238 } else { 1239 pid_t tid = vm_is_stack(task, vma, is_pid); 1240 if (tid != 0) { 1241 /* 1242 * Thread stack in /proc/PID/task/TID/maps or 1243 * the main process stack. 1244 */ 1245 if (!is_pid || (vma->vm_start <= mm->start_stack && 1246 vma->vm_end >= mm->start_stack)) 1247 seq_printf(m, " stack"); 1248 else 1249 seq_printf(m, " stack:%d", tid); 1250 } 1251 } 1252 1253 if (is_vm_hugetlb_page(vma)) 1254 seq_printf(m, " huge"); 1255 1256 walk_page_range(vma->vm_start, vma->vm_end, &walk); 1257 1258 if (!md->pages) 1259 goto out; 1260 1261 if (md->anon) 1262 seq_printf(m, " anon=%lu", md->anon); 1263 1264 if (md->dirty) 1265 seq_printf(m, " dirty=%lu", md->dirty); 1266 1267 if (md->pages != md->anon && md->pages != md->dirty) 1268 seq_printf(m, " mapped=%lu", md->pages); 1269 1270 if (md->mapcount_max > 1) 1271 seq_printf(m, " mapmax=%lu", md->mapcount_max); 1272 1273 if (md->swapcache) 1274 seq_printf(m, " swapcache=%lu", md->swapcache); 1275 1276 if (md->active < md->pages && !is_vm_hugetlb_page(vma)) 1277 seq_printf(m, " active=%lu", md->active); 1278 1279 if (md->writeback) 1280 seq_printf(m, " writeback=%lu", md->writeback); 1281 1282 for_each_node_state(n, N_HIGH_MEMORY) 1283 if (md->node[n]) 1284 seq_printf(m, " N%d=%lu", n, md->node[n]); 1285 out: 1286 seq_putc(m, '\n'); 1287 1288 if (m->count < m->size) 1289 m->version = (vma != proc_priv->tail_vma) ? vma->vm_start : 0; 1290 return 0; 1291 } 1292 1293 static int show_pid_numa_map(struct seq_file *m, void *v) 1294 { 1295 return show_numa_map(m, v, 1); 1296 } 1297 1298 static int show_tid_numa_map(struct seq_file *m, void *v) 1299 { 1300 return show_numa_map(m, v, 0); 1301 } 1302 1303 static const struct seq_operations proc_pid_numa_maps_op = { 1304 .start = m_start, 1305 .next = m_next, 1306 .stop = m_stop, 1307 .show = show_pid_numa_map, 1308 }; 1309 1310 static const struct seq_operations proc_tid_numa_maps_op = { 1311 .start = m_start, 1312 .next = m_next, 1313 .stop = m_stop, 1314 .show = show_tid_numa_map, 1315 }; 1316 1317 static int numa_maps_open(struct inode *inode, struct file *file, 1318 const struct seq_operations *ops) 1319 { 1320 struct numa_maps_private *priv; 1321 int ret = -ENOMEM; 1322 priv = kzalloc(sizeof(*priv), GFP_KERNEL); 1323 if (priv) { 1324 priv->proc_maps.pid = proc_pid(inode); 1325 ret = seq_open(file, ops); 1326 if (!ret) { 1327 struct seq_file *m = file->private_data; 1328 m->private = priv; 1329 } else { 1330 kfree(priv); 1331 } 1332 } 1333 return ret; 1334 } 1335 1336 static int pid_numa_maps_open(struct inode *inode, struct file *file) 1337 { 1338 return numa_maps_open(inode, file, &proc_pid_numa_maps_op); 1339 } 1340 1341 static int tid_numa_maps_open(struct inode *inode, struct file *file) 1342 { 1343 return numa_maps_open(inode, file, &proc_tid_numa_maps_op); 1344 } 1345 1346 const struct file_operations proc_pid_numa_maps_operations = { 1347 .open = pid_numa_maps_open, 1348 .read = seq_read, 1349 .llseek = seq_lseek, 1350 .release = seq_release_private, 1351 }; 1352 1353 const struct file_operations proc_tid_numa_maps_operations = { 1354 .open = tid_numa_maps_open, 1355 .read = seq_read, 1356 .llseek = seq_lseek, 1357 .release = seq_release_private, 1358 }; 1359 #endif /* CONFIG_NUMA */ 1360