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 "VmHWM:\t%8lu kB\n" 48 "VmRSS:\t%8lu kB\n" 49 "VmData:\t%8lu kB\n" 50 "VmStk:\t%8lu kB\n" 51 "VmExe:\t%8lu kB\n" 52 "VmLib:\t%8lu kB\n" 53 "VmPTE:\t%8lu kB\n" 54 "VmSwap:\t%8lu kB\n", 55 hiwater_vm << (PAGE_SHIFT-10), 56 (total_vm - mm->reserved_vm) << (PAGE_SHIFT-10), 57 mm->locked_vm << (PAGE_SHIFT-10), 58 hiwater_rss << (PAGE_SHIFT-10), 59 total_rss << (PAGE_SHIFT-10), 60 data << (PAGE_SHIFT-10), 61 mm->stack_vm << (PAGE_SHIFT-10), text, lib, 62 (PTRS_PER_PTE*sizeof(pte_t)*mm->nr_ptes) >> 10, 63 swap << (PAGE_SHIFT-10)); 64 } 65 66 unsigned long task_vsize(struct mm_struct *mm) 67 { 68 return PAGE_SIZE * mm->total_vm; 69 } 70 71 unsigned long task_statm(struct mm_struct *mm, 72 unsigned long *shared, unsigned long *text, 73 unsigned long *data, unsigned long *resident) 74 { 75 *shared = get_mm_counter(mm, MM_FILEPAGES); 76 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) 77 >> PAGE_SHIFT; 78 *data = mm->total_vm - mm->shared_vm; 79 *resident = *shared + get_mm_counter(mm, MM_ANONPAGES); 80 return mm->total_vm; 81 } 82 83 static void pad_len_spaces(struct seq_file *m, int len) 84 { 85 len = 25 + sizeof(void*) * 6 - len; 86 if (len < 1) 87 len = 1; 88 seq_printf(m, "%*c", len, ' '); 89 } 90 91 static void vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma) 92 { 93 if (vma && vma != priv->tail_vma) { 94 struct mm_struct *mm = vma->vm_mm; 95 up_read(&mm->mmap_sem); 96 mmput(mm); 97 } 98 } 99 100 static void *m_start(struct seq_file *m, loff_t *pos) 101 { 102 struct proc_maps_private *priv = m->private; 103 unsigned long last_addr = m->version; 104 struct mm_struct *mm; 105 struct vm_area_struct *vma, *tail_vma = NULL; 106 loff_t l = *pos; 107 108 /* Clear the per syscall fields in priv */ 109 priv->task = NULL; 110 priv->tail_vma = NULL; 111 112 /* 113 * We remember last_addr rather than next_addr to hit with 114 * mmap_cache most of the time. We have zero last_addr at 115 * the beginning and also after lseek. We will have -1 last_addr 116 * after the end of the vmas. 117 */ 118 119 if (last_addr == -1UL) 120 return NULL; 121 122 priv->task = get_pid_task(priv->pid, PIDTYPE_PID); 123 if (!priv->task) 124 return ERR_PTR(-ESRCH); 125 126 mm = mm_for_maps(priv->task); 127 if (!mm || IS_ERR(mm)) 128 return mm; 129 down_read(&mm->mmap_sem); 130 131 tail_vma = get_gate_vma(priv->task->mm); 132 priv->tail_vma = tail_vma; 133 134 /* Start with last addr hint */ 135 vma = find_vma(mm, last_addr); 136 if (last_addr && vma) { 137 vma = vma->vm_next; 138 goto out; 139 } 140 141 /* 142 * Check the vma index is within the range and do 143 * sequential scan until m_index. 144 */ 145 vma = NULL; 146 if ((unsigned long)l < mm->map_count) { 147 vma = mm->mmap; 148 while (l-- && vma) 149 vma = vma->vm_next; 150 goto out; 151 } 152 153 if (l != mm->map_count) 154 tail_vma = NULL; /* After gate vma */ 155 156 out: 157 if (vma) 158 return vma; 159 160 /* End of vmas has been reached */ 161 m->version = (tail_vma != NULL)? 0: -1UL; 162 up_read(&mm->mmap_sem); 163 mmput(mm); 164 return tail_vma; 165 } 166 167 static void *m_next(struct seq_file *m, void *v, loff_t *pos) 168 { 169 struct proc_maps_private *priv = m->private; 170 struct vm_area_struct *vma = v; 171 struct vm_area_struct *tail_vma = priv->tail_vma; 172 173 (*pos)++; 174 if (vma && (vma != tail_vma) && vma->vm_next) 175 return vma->vm_next; 176 vma_stop(priv, vma); 177 return (vma != tail_vma)? tail_vma: NULL; 178 } 179 180 static void m_stop(struct seq_file *m, void *v) 181 { 182 struct proc_maps_private *priv = m->private; 183 struct vm_area_struct *vma = v; 184 185 if (!IS_ERR(vma)) 186 vma_stop(priv, vma); 187 if (priv->task) 188 put_task_struct(priv->task); 189 } 190 191 static int do_maps_open(struct inode *inode, struct file *file, 192 const struct seq_operations *ops) 193 { 194 struct proc_maps_private *priv; 195 int ret = -ENOMEM; 196 priv = kzalloc(sizeof(*priv), GFP_KERNEL); 197 if (priv) { 198 priv->pid = proc_pid(inode); 199 ret = seq_open(file, ops); 200 if (!ret) { 201 struct seq_file *m = file->private_data; 202 m->private = priv; 203 } else { 204 kfree(priv); 205 } 206 } 207 return ret; 208 } 209 210 static void show_map_vma(struct seq_file *m, struct vm_area_struct *vma) 211 { 212 struct mm_struct *mm = vma->vm_mm; 213 struct file *file = vma->vm_file; 214 int flags = vma->vm_flags; 215 unsigned long ino = 0; 216 unsigned long long pgoff = 0; 217 unsigned long start; 218 dev_t dev = 0; 219 int len; 220 221 if (file) { 222 struct inode *inode = vma->vm_file->f_path.dentry->d_inode; 223 dev = inode->i_sb->s_dev; 224 ino = inode->i_ino; 225 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT; 226 } 227 228 /* We don't show the stack guard page in /proc/maps */ 229 start = vma->vm_start; 230 if (vma->vm_flags & VM_GROWSDOWN) 231 if (!vma_stack_continue(vma->vm_prev, vma->vm_start)) 232 start += PAGE_SIZE; 233 234 seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu %n", 235 start, 236 vma->vm_end, 237 flags & VM_READ ? 'r' : '-', 238 flags & VM_WRITE ? 'w' : '-', 239 flags & VM_EXEC ? 'x' : '-', 240 flags & VM_MAYSHARE ? 's' : 'p', 241 pgoff, 242 MAJOR(dev), MINOR(dev), ino, &len); 243 244 /* 245 * Print the dentry name for named mappings, and a 246 * special [heap] marker for the heap: 247 */ 248 if (file) { 249 pad_len_spaces(m, len); 250 seq_path(m, &file->f_path, "\n"); 251 } else { 252 const char *name = arch_vma_name(vma); 253 if (!name) { 254 if (mm) { 255 if (vma->vm_start <= mm->brk && 256 vma->vm_end >= mm->start_brk) { 257 name = "[heap]"; 258 } else if (vma->vm_start <= mm->start_stack && 259 vma->vm_end >= mm->start_stack) { 260 name = "[stack]"; 261 } 262 } else { 263 name = "[vdso]"; 264 } 265 } 266 if (name) { 267 pad_len_spaces(m, len); 268 seq_puts(m, name); 269 } 270 } 271 seq_putc(m, '\n'); 272 } 273 274 static int show_map(struct seq_file *m, void *v) 275 { 276 struct vm_area_struct *vma = v; 277 struct proc_maps_private *priv = m->private; 278 struct task_struct *task = priv->task; 279 280 show_map_vma(m, vma); 281 282 if (m->count < m->size) /* vma is copied successfully */ 283 m->version = (vma != get_gate_vma(task->mm)) 284 ? vma->vm_start : 0; 285 return 0; 286 } 287 288 static const struct seq_operations proc_pid_maps_op = { 289 .start = m_start, 290 .next = m_next, 291 .stop = m_stop, 292 .show = show_map 293 }; 294 295 static int maps_open(struct inode *inode, struct file *file) 296 { 297 return do_maps_open(inode, file, &proc_pid_maps_op); 298 } 299 300 const struct file_operations proc_maps_operations = { 301 .open = maps_open, 302 .read = seq_read, 303 .llseek = seq_lseek, 304 .release = seq_release_private, 305 }; 306 307 /* 308 * Proportional Set Size(PSS): my share of RSS. 309 * 310 * PSS of a process is the count of pages it has in memory, where each 311 * page is divided by the number of processes sharing it. So if a 312 * process has 1000 pages all to itself, and 1000 shared with one other 313 * process, its PSS will be 1500. 314 * 315 * To keep (accumulated) division errors low, we adopt a 64bit 316 * fixed-point pss counter to minimize division errors. So (pss >> 317 * PSS_SHIFT) would be the real byte count. 318 * 319 * A shift of 12 before division means (assuming 4K page size): 320 * - 1M 3-user-pages add up to 8KB errors; 321 * - supports mapcount up to 2^24, or 16M; 322 * - supports PSS up to 2^52 bytes, or 4PB. 323 */ 324 #define PSS_SHIFT 12 325 326 #ifdef CONFIG_PROC_PAGE_MONITOR 327 struct mem_size_stats { 328 struct vm_area_struct *vma; 329 unsigned long resident; 330 unsigned long shared_clean; 331 unsigned long shared_dirty; 332 unsigned long private_clean; 333 unsigned long private_dirty; 334 unsigned long referenced; 335 unsigned long anonymous; 336 unsigned long anonymous_thp; 337 unsigned long swap; 338 u64 pss; 339 }; 340 341 342 static void smaps_pte_entry(pte_t ptent, unsigned long addr, 343 unsigned long ptent_size, struct mm_walk *walk) 344 { 345 struct mem_size_stats *mss = walk->private; 346 struct vm_area_struct *vma = mss->vma; 347 struct page *page; 348 int mapcount; 349 350 if (is_swap_pte(ptent)) { 351 mss->swap += ptent_size; 352 return; 353 } 354 355 if (!pte_present(ptent)) 356 return; 357 358 page = vm_normal_page(vma, addr, ptent); 359 if (!page) 360 return; 361 362 if (PageAnon(page)) 363 mss->anonymous += ptent_size; 364 365 mss->resident += ptent_size; 366 /* Accumulate the size in pages that have been accessed. */ 367 if (pte_young(ptent) || PageReferenced(page)) 368 mss->referenced += ptent_size; 369 mapcount = page_mapcount(page); 370 if (mapcount >= 2) { 371 if (pte_dirty(ptent) || PageDirty(page)) 372 mss->shared_dirty += ptent_size; 373 else 374 mss->shared_clean += ptent_size; 375 mss->pss += (ptent_size << PSS_SHIFT) / mapcount; 376 } else { 377 if (pte_dirty(ptent) || PageDirty(page)) 378 mss->private_dirty += ptent_size; 379 else 380 mss->private_clean += ptent_size; 381 mss->pss += (ptent_size << PSS_SHIFT); 382 } 383 } 384 385 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, 386 struct mm_walk *walk) 387 { 388 struct mem_size_stats *mss = walk->private; 389 struct vm_area_struct *vma = mss->vma; 390 pte_t *pte; 391 spinlock_t *ptl; 392 393 spin_lock(&walk->mm->page_table_lock); 394 if (pmd_trans_huge(*pmd)) { 395 if (pmd_trans_splitting(*pmd)) { 396 spin_unlock(&walk->mm->page_table_lock); 397 wait_split_huge_page(vma->anon_vma, pmd); 398 } else { 399 smaps_pte_entry(*(pte_t *)pmd, addr, 400 HPAGE_PMD_SIZE, walk); 401 spin_unlock(&walk->mm->page_table_lock); 402 mss->anonymous_thp += HPAGE_PMD_SIZE; 403 return 0; 404 } 405 } else { 406 spin_unlock(&walk->mm->page_table_lock); 407 } 408 /* 409 * The mmap_sem held all the way back in m_start() is what 410 * keeps khugepaged out of here and from collapsing things 411 * in here. 412 */ 413 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); 414 for (; addr != end; pte++, addr += PAGE_SIZE) 415 smaps_pte_entry(*pte, addr, PAGE_SIZE, walk); 416 pte_unmap_unlock(pte - 1, ptl); 417 cond_resched(); 418 return 0; 419 } 420 421 static int show_smap(struct seq_file *m, void *v) 422 { 423 struct proc_maps_private *priv = m->private; 424 struct task_struct *task = priv->task; 425 struct vm_area_struct *vma = v; 426 struct mem_size_stats mss; 427 struct mm_walk smaps_walk = { 428 .pmd_entry = smaps_pte_range, 429 .mm = vma->vm_mm, 430 .private = &mss, 431 }; 432 433 memset(&mss, 0, sizeof mss); 434 mss.vma = vma; 435 /* mmap_sem is held in m_start */ 436 if (vma->vm_mm && !is_vm_hugetlb_page(vma)) 437 walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk); 438 439 show_map_vma(m, vma); 440 441 seq_printf(m, 442 "Size: %8lu kB\n" 443 "Rss: %8lu kB\n" 444 "Pss: %8lu kB\n" 445 "Shared_Clean: %8lu kB\n" 446 "Shared_Dirty: %8lu kB\n" 447 "Private_Clean: %8lu kB\n" 448 "Private_Dirty: %8lu kB\n" 449 "Referenced: %8lu kB\n" 450 "Anonymous: %8lu kB\n" 451 "AnonHugePages: %8lu kB\n" 452 "Swap: %8lu kB\n" 453 "KernelPageSize: %8lu kB\n" 454 "MMUPageSize: %8lu kB\n" 455 "Locked: %8lu kB\n", 456 (vma->vm_end - vma->vm_start) >> 10, 457 mss.resident >> 10, 458 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)), 459 mss.shared_clean >> 10, 460 mss.shared_dirty >> 10, 461 mss.private_clean >> 10, 462 mss.private_dirty >> 10, 463 mss.referenced >> 10, 464 mss.anonymous >> 10, 465 mss.anonymous_thp >> 10, 466 mss.swap >> 10, 467 vma_kernel_pagesize(vma) >> 10, 468 vma_mmu_pagesize(vma) >> 10, 469 (vma->vm_flags & VM_LOCKED) ? 470 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0); 471 472 if (m->count < m->size) /* vma is copied successfully */ 473 m->version = (vma != get_gate_vma(task->mm)) 474 ? vma->vm_start : 0; 475 return 0; 476 } 477 478 static const struct seq_operations proc_pid_smaps_op = { 479 .start = m_start, 480 .next = m_next, 481 .stop = m_stop, 482 .show = show_smap 483 }; 484 485 static int smaps_open(struct inode *inode, struct file *file) 486 { 487 return do_maps_open(inode, file, &proc_pid_smaps_op); 488 } 489 490 const struct file_operations proc_smaps_operations = { 491 .open = smaps_open, 492 .read = seq_read, 493 .llseek = seq_lseek, 494 .release = seq_release_private, 495 }; 496 497 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr, 498 unsigned long end, struct mm_walk *walk) 499 { 500 struct vm_area_struct *vma = walk->private; 501 pte_t *pte, ptent; 502 spinlock_t *ptl; 503 struct page *page; 504 505 split_huge_page_pmd(walk->mm, pmd); 506 507 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); 508 for (; addr != end; pte++, addr += PAGE_SIZE) { 509 ptent = *pte; 510 if (!pte_present(ptent)) 511 continue; 512 513 page = vm_normal_page(vma, addr, ptent); 514 if (!page) 515 continue; 516 517 /* Clear accessed and referenced bits. */ 518 ptep_test_and_clear_young(vma, addr, pte); 519 ClearPageReferenced(page); 520 } 521 pte_unmap_unlock(pte - 1, ptl); 522 cond_resched(); 523 return 0; 524 } 525 526 #define CLEAR_REFS_ALL 1 527 #define CLEAR_REFS_ANON 2 528 #define CLEAR_REFS_MAPPED 3 529 530 static ssize_t clear_refs_write(struct file *file, const char __user *buf, 531 size_t count, loff_t *ppos) 532 { 533 struct task_struct *task; 534 char buffer[PROC_NUMBUF]; 535 struct mm_struct *mm; 536 struct vm_area_struct *vma; 537 long type; 538 539 memset(buffer, 0, sizeof(buffer)); 540 if (count > sizeof(buffer) - 1) 541 count = sizeof(buffer) - 1; 542 if (copy_from_user(buffer, buf, count)) 543 return -EFAULT; 544 if (strict_strtol(strstrip(buffer), 10, &type)) 545 return -EINVAL; 546 if (type < CLEAR_REFS_ALL || type > CLEAR_REFS_MAPPED) 547 return -EINVAL; 548 task = get_proc_task(file->f_path.dentry->d_inode); 549 if (!task) 550 return -ESRCH; 551 mm = get_task_mm(task); 552 if (mm) { 553 struct mm_walk clear_refs_walk = { 554 .pmd_entry = clear_refs_pte_range, 555 .mm = mm, 556 }; 557 down_read(&mm->mmap_sem); 558 for (vma = mm->mmap; vma; vma = vma->vm_next) { 559 clear_refs_walk.private = vma; 560 if (is_vm_hugetlb_page(vma)) 561 continue; 562 /* 563 * Writing 1 to /proc/pid/clear_refs affects all pages. 564 * 565 * Writing 2 to /proc/pid/clear_refs only affects 566 * Anonymous pages. 567 * 568 * Writing 3 to /proc/pid/clear_refs only affects file 569 * mapped pages. 570 */ 571 if (type == CLEAR_REFS_ANON && vma->vm_file) 572 continue; 573 if (type == CLEAR_REFS_MAPPED && !vma->vm_file) 574 continue; 575 walk_page_range(vma->vm_start, vma->vm_end, 576 &clear_refs_walk); 577 } 578 flush_tlb_mm(mm); 579 up_read(&mm->mmap_sem); 580 mmput(mm); 581 } 582 put_task_struct(task); 583 584 return count; 585 } 586 587 const struct file_operations proc_clear_refs_operations = { 588 .write = clear_refs_write, 589 .llseek = noop_llseek, 590 }; 591 592 struct pagemapread { 593 int pos, len; 594 u64 *buffer; 595 }; 596 597 #define PM_ENTRY_BYTES sizeof(u64) 598 #define PM_STATUS_BITS 3 599 #define PM_STATUS_OFFSET (64 - PM_STATUS_BITS) 600 #define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET) 601 #define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK) 602 #define PM_PSHIFT_BITS 6 603 #define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS) 604 #define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET) 605 #define PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK) 606 #define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1) 607 #define PM_PFRAME(x) ((x) & PM_PFRAME_MASK) 608 609 #define PM_PRESENT PM_STATUS(4LL) 610 #define PM_SWAP PM_STATUS(2LL) 611 #define PM_NOT_PRESENT PM_PSHIFT(PAGE_SHIFT) 612 #define PM_END_OF_BUFFER 1 613 614 static int add_to_pagemap(unsigned long addr, u64 pfn, 615 struct pagemapread *pm) 616 { 617 pm->buffer[pm->pos++] = pfn; 618 if (pm->pos >= pm->len) 619 return PM_END_OF_BUFFER; 620 return 0; 621 } 622 623 static int pagemap_pte_hole(unsigned long start, unsigned long end, 624 struct mm_walk *walk) 625 { 626 struct pagemapread *pm = walk->private; 627 unsigned long addr; 628 int err = 0; 629 for (addr = start; addr < end; addr += PAGE_SIZE) { 630 err = add_to_pagemap(addr, PM_NOT_PRESENT, pm); 631 if (err) 632 break; 633 } 634 return err; 635 } 636 637 static u64 swap_pte_to_pagemap_entry(pte_t pte) 638 { 639 swp_entry_t e = pte_to_swp_entry(pte); 640 return swp_type(e) | (swp_offset(e) << MAX_SWAPFILES_SHIFT); 641 } 642 643 static u64 pte_to_pagemap_entry(pte_t pte) 644 { 645 u64 pme = 0; 646 if (is_swap_pte(pte)) 647 pme = PM_PFRAME(swap_pte_to_pagemap_entry(pte)) 648 | PM_PSHIFT(PAGE_SHIFT) | PM_SWAP; 649 else if (pte_present(pte)) 650 pme = PM_PFRAME(pte_pfn(pte)) 651 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT; 652 return pme; 653 } 654 655 static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, 656 struct mm_walk *walk) 657 { 658 struct vm_area_struct *vma; 659 struct pagemapread *pm = walk->private; 660 pte_t *pte; 661 int err = 0; 662 663 split_huge_page_pmd(walk->mm, pmd); 664 665 /* find the first VMA at or above 'addr' */ 666 vma = find_vma(walk->mm, addr); 667 for (; addr != end; addr += PAGE_SIZE) { 668 u64 pfn = PM_NOT_PRESENT; 669 670 /* check to see if we've left 'vma' behind 671 * and need a new, higher one */ 672 if (vma && (addr >= vma->vm_end)) 673 vma = find_vma(walk->mm, addr); 674 675 /* check that 'vma' actually covers this address, 676 * and that it isn't a huge page vma */ 677 if (vma && (vma->vm_start <= addr) && 678 !is_vm_hugetlb_page(vma)) { 679 pte = pte_offset_map(pmd, addr); 680 pfn = pte_to_pagemap_entry(*pte); 681 /* unmap before userspace copy */ 682 pte_unmap(pte); 683 } 684 err = add_to_pagemap(addr, pfn, pm); 685 if (err) 686 return err; 687 } 688 689 cond_resched(); 690 691 return err; 692 } 693 694 #ifdef CONFIG_HUGETLB_PAGE 695 static u64 huge_pte_to_pagemap_entry(pte_t pte, int offset) 696 { 697 u64 pme = 0; 698 if (pte_present(pte)) 699 pme = PM_PFRAME(pte_pfn(pte) + offset) 700 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT; 701 return pme; 702 } 703 704 /* This function walks within one hugetlb entry in the single call */ 705 static int pagemap_hugetlb_range(pte_t *pte, unsigned long hmask, 706 unsigned long addr, unsigned long end, 707 struct mm_walk *walk) 708 { 709 struct pagemapread *pm = walk->private; 710 int err = 0; 711 u64 pfn; 712 713 for (; addr != end; addr += PAGE_SIZE) { 714 int offset = (addr & ~hmask) >> PAGE_SHIFT; 715 pfn = huge_pte_to_pagemap_entry(*pte, offset); 716 err = add_to_pagemap(addr, pfn, pm); 717 if (err) 718 return err; 719 } 720 721 cond_resched(); 722 723 return err; 724 } 725 #endif /* HUGETLB_PAGE */ 726 727 /* 728 * /proc/pid/pagemap - an array mapping virtual pages to pfns 729 * 730 * For each page in the address space, this file contains one 64-bit entry 731 * consisting of the following: 732 * 733 * Bits 0-55 page frame number (PFN) if present 734 * Bits 0-4 swap type if swapped 735 * Bits 5-55 swap offset if swapped 736 * Bits 55-60 page shift (page size = 1<<page shift) 737 * Bit 61 reserved for future use 738 * Bit 62 page swapped 739 * Bit 63 page present 740 * 741 * If the page is not present but in swap, then the PFN contains an 742 * encoding of the swap file number and the page's offset into the 743 * swap. Unmapped pages return a null PFN. This allows determining 744 * precisely which pages are mapped (or in swap) and comparing mapped 745 * pages between processes. 746 * 747 * Efficient users of this interface will use /proc/pid/maps to 748 * determine which areas of memory are actually mapped and llseek to 749 * skip over unmapped regions. 750 */ 751 #define PAGEMAP_WALK_SIZE (PMD_SIZE) 752 #define PAGEMAP_WALK_MASK (PMD_MASK) 753 static ssize_t pagemap_read(struct file *file, char __user *buf, 754 size_t count, loff_t *ppos) 755 { 756 struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode); 757 struct mm_struct *mm; 758 struct pagemapread pm; 759 int ret = -ESRCH; 760 struct mm_walk pagemap_walk = {}; 761 unsigned long src; 762 unsigned long svpfn; 763 unsigned long start_vaddr; 764 unsigned long end_vaddr; 765 int copied = 0; 766 767 if (!task) 768 goto out; 769 770 mm = mm_for_maps(task); 771 ret = PTR_ERR(mm); 772 if (!mm || IS_ERR(mm)) 773 goto out_task; 774 775 ret = -EINVAL; 776 /* file position must be aligned */ 777 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES)) 778 goto out_task; 779 780 ret = 0; 781 782 if (!count) 783 goto out_task; 784 785 pm.len = PM_ENTRY_BYTES * (PAGEMAP_WALK_SIZE >> PAGE_SHIFT); 786 pm.buffer = kmalloc(pm.len, GFP_TEMPORARY); 787 ret = -ENOMEM; 788 if (!pm.buffer) 789 goto out_mm; 790 791 pagemap_walk.pmd_entry = pagemap_pte_range; 792 pagemap_walk.pte_hole = pagemap_pte_hole; 793 #ifdef CONFIG_HUGETLB_PAGE 794 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range; 795 #endif 796 pagemap_walk.mm = mm; 797 pagemap_walk.private = ± 798 799 src = *ppos; 800 svpfn = src / PM_ENTRY_BYTES; 801 start_vaddr = svpfn << PAGE_SHIFT; 802 end_vaddr = TASK_SIZE_OF(task); 803 804 /* watch out for wraparound */ 805 if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT) 806 start_vaddr = end_vaddr; 807 808 /* 809 * The odds are that this will stop walking way 810 * before end_vaddr, because the length of the 811 * user buffer is tracked in "pm", and the walk 812 * will stop when we hit the end of the buffer. 813 */ 814 ret = 0; 815 while (count && (start_vaddr < end_vaddr)) { 816 int len; 817 unsigned long end; 818 819 pm.pos = 0; 820 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK; 821 /* overflow ? */ 822 if (end < start_vaddr || end > end_vaddr) 823 end = end_vaddr; 824 down_read(&mm->mmap_sem); 825 ret = walk_page_range(start_vaddr, end, &pagemap_walk); 826 up_read(&mm->mmap_sem); 827 start_vaddr = end; 828 829 len = min(count, PM_ENTRY_BYTES * pm.pos); 830 if (copy_to_user(buf, pm.buffer, len)) { 831 ret = -EFAULT; 832 goto out_free; 833 } 834 copied += len; 835 buf += len; 836 count -= len; 837 } 838 *ppos += copied; 839 if (!ret || ret == PM_END_OF_BUFFER) 840 ret = copied; 841 842 out_free: 843 kfree(pm.buffer); 844 out_mm: 845 mmput(mm); 846 out_task: 847 put_task_struct(task); 848 out: 849 return ret; 850 } 851 852 const struct file_operations proc_pagemap_operations = { 853 .llseek = mem_lseek, /* borrow this */ 854 .read = pagemap_read, 855 }; 856 #endif /* CONFIG_PROC_PAGE_MONITOR */ 857 858 #ifdef CONFIG_NUMA 859 extern int show_numa_map(struct seq_file *m, void *v); 860 861 static const struct seq_operations proc_pid_numa_maps_op = { 862 .start = m_start, 863 .next = m_next, 864 .stop = m_stop, 865 .show = show_numa_map, 866 }; 867 868 static int numa_maps_open(struct inode *inode, struct file *file) 869 { 870 return do_maps_open(inode, file, &proc_pid_numa_maps_op); 871 } 872 873 const struct file_operations proc_numa_maps_operations = { 874 .open = numa_maps_open, 875 .read = seq_read, 876 .llseek = seq_lseek, 877 .release = seq_release_private, 878 }; 879 #endif 880