1 #include <linux/mm.h> 2 #include <linux/hugetlb.h> 3 #include <linux/mount.h> 4 #include <linux/seq_file.h> 5 #include <linux/highmem.h> 6 #include <linux/ptrace.h> 7 #include <linux/slab.h> 8 #include <linux/pagemap.h> 9 #include <linux/mempolicy.h> 10 #include <linux/swap.h> 11 #include <linux/swapops.h> 12 13 #include <asm/elf.h> 14 #include <asm/uaccess.h> 15 #include <asm/tlbflush.h> 16 #include "internal.h" 17 18 void task_mem(struct seq_file *m, struct mm_struct *mm) 19 { 20 unsigned long data, text, lib, swap; 21 unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss; 22 23 /* 24 * Note: to minimize their overhead, mm maintains hiwater_vm and 25 * hiwater_rss only when about to *lower* total_vm or rss. Any 26 * collector of these hiwater stats must therefore get total_vm 27 * and rss too, which will usually be the higher. Barriers? not 28 * worth the effort, such snapshots can always be inconsistent. 29 */ 30 hiwater_vm = total_vm = mm->total_vm; 31 if (hiwater_vm < mm->hiwater_vm) 32 hiwater_vm = mm->hiwater_vm; 33 hiwater_rss = total_rss = get_mm_rss(mm); 34 if (hiwater_rss < mm->hiwater_rss) 35 hiwater_rss = mm->hiwater_rss; 36 37 data = mm->total_vm - mm->shared_vm - mm->stack_vm; 38 text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10; 39 lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text; 40 swap = get_mm_counter(mm, MM_SWAPENTS); 41 seq_printf(m, 42 "VmPeak:\t%8lu kB\n" 43 "VmSize:\t%8lu kB\n" 44 "VmLck:\t%8lu kB\n" 45 "VmHWM:\t%8lu kB\n" 46 "VmRSS:\t%8lu kB\n" 47 "VmData:\t%8lu kB\n" 48 "VmStk:\t%8lu kB\n" 49 "VmExe:\t%8lu kB\n" 50 "VmLib:\t%8lu kB\n" 51 "VmPTE:\t%8lu kB\n" 52 "VmSwap:\t%8lu kB\n", 53 hiwater_vm << (PAGE_SHIFT-10), 54 (total_vm - mm->reserved_vm) << (PAGE_SHIFT-10), 55 mm->locked_vm << (PAGE_SHIFT-10), 56 hiwater_rss << (PAGE_SHIFT-10), 57 total_rss << (PAGE_SHIFT-10), 58 data << (PAGE_SHIFT-10), 59 mm->stack_vm << (PAGE_SHIFT-10), text, lib, 60 (PTRS_PER_PTE*sizeof(pte_t)*mm->nr_ptes) >> 10, 61 swap << (PAGE_SHIFT-10)); 62 } 63 64 unsigned long task_vsize(struct mm_struct *mm) 65 { 66 return PAGE_SIZE * mm->total_vm; 67 } 68 69 int task_statm(struct mm_struct *mm, int *shared, int *text, 70 int *data, int *resident) 71 { 72 *shared = get_mm_counter(mm, MM_FILEPAGES); 73 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) 74 >> PAGE_SHIFT; 75 *data = mm->total_vm - mm->shared_vm; 76 *resident = *shared + get_mm_counter(mm, MM_ANONPAGES); 77 return mm->total_vm; 78 } 79 80 static void pad_len_spaces(struct seq_file *m, int len) 81 { 82 len = 25 + sizeof(void*) * 6 - len; 83 if (len < 1) 84 len = 1; 85 seq_printf(m, "%*c", len, ' '); 86 } 87 88 static void vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma) 89 { 90 if (vma && vma != priv->tail_vma) { 91 struct mm_struct *mm = vma->vm_mm; 92 up_read(&mm->mmap_sem); 93 mmput(mm); 94 } 95 } 96 97 static void *m_start(struct seq_file *m, loff_t *pos) 98 { 99 struct proc_maps_private *priv = m->private; 100 unsigned long last_addr = m->version; 101 struct mm_struct *mm; 102 struct vm_area_struct *vma, *tail_vma = NULL; 103 loff_t l = *pos; 104 105 /* Clear the per syscall fields in priv */ 106 priv->task = NULL; 107 priv->tail_vma = NULL; 108 109 /* 110 * We remember last_addr rather than next_addr to hit with 111 * mmap_cache most of the time. We have zero last_addr at 112 * the beginning and also after lseek. We will have -1 last_addr 113 * after the end of the vmas. 114 */ 115 116 if (last_addr == -1UL) 117 return NULL; 118 119 priv->task = get_pid_task(priv->pid, PIDTYPE_PID); 120 if (!priv->task) 121 return NULL; 122 123 mm = mm_for_maps(priv->task); 124 if (!mm) 125 return NULL; 126 down_read(&mm->mmap_sem); 127 128 tail_vma = get_gate_vma(priv->task); 129 priv->tail_vma = tail_vma; 130 131 /* Start with last addr hint */ 132 vma = find_vma(mm, last_addr); 133 if (last_addr && vma) { 134 vma = vma->vm_next; 135 goto out; 136 } 137 138 /* 139 * Check the vma index is within the range and do 140 * sequential scan until m_index. 141 */ 142 vma = NULL; 143 if ((unsigned long)l < mm->map_count) { 144 vma = mm->mmap; 145 while (l-- && vma) 146 vma = vma->vm_next; 147 goto out; 148 } 149 150 if (l != mm->map_count) 151 tail_vma = NULL; /* After gate vma */ 152 153 out: 154 if (vma) 155 return vma; 156 157 /* End of vmas has been reached */ 158 m->version = (tail_vma != NULL)? 0: -1UL; 159 up_read(&mm->mmap_sem); 160 mmput(mm); 161 return tail_vma; 162 } 163 164 static void *m_next(struct seq_file *m, void *v, loff_t *pos) 165 { 166 struct proc_maps_private *priv = m->private; 167 struct vm_area_struct *vma = v; 168 struct vm_area_struct *tail_vma = priv->tail_vma; 169 170 (*pos)++; 171 if (vma && (vma != tail_vma) && vma->vm_next) 172 return vma->vm_next; 173 vma_stop(priv, vma); 174 return (vma != tail_vma)? tail_vma: NULL; 175 } 176 177 static void m_stop(struct seq_file *m, void *v) 178 { 179 struct proc_maps_private *priv = m->private; 180 struct vm_area_struct *vma = v; 181 182 vma_stop(priv, vma); 183 if (priv->task) 184 put_task_struct(priv->task); 185 } 186 187 static int do_maps_open(struct inode *inode, struct file *file, 188 const struct seq_operations *ops) 189 { 190 struct proc_maps_private *priv; 191 int ret = -ENOMEM; 192 priv = kzalloc(sizeof(*priv), GFP_KERNEL); 193 if (priv) { 194 priv->pid = proc_pid(inode); 195 ret = seq_open(file, ops); 196 if (!ret) { 197 struct seq_file *m = file->private_data; 198 m->private = priv; 199 } else { 200 kfree(priv); 201 } 202 } 203 return ret; 204 } 205 206 static void show_map_vma(struct seq_file *m, struct vm_area_struct *vma) 207 { 208 struct mm_struct *mm = vma->vm_mm; 209 struct file *file = vma->vm_file; 210 int flags = vma->vm_flags; 211 unsigned long ino = 0; 212 unsigned long long pgoff = 0; 213 unsigned long start; 214 dev_t dev = 0; 215 int len; 216 217 if (file) { 218 struct inode *inode = vma->vm_file->f_path.dentry->d_inode; 219 dev = inode->i_sb->s_dev; 220 ino = inode->i_ino; 221 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT; 222 } 223 224 /* We don't show the stack guard page in /proc/maps */ 225 start = vma->vm_start; 226 if (vma->vm_flags & VM_GROWSDOWN) 227 if (!vma_stack_continue(vma->vm_prev, vma->vm_start)) 228 start += PAGE_SIZE; 229 230 seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu %n", 231 start, 232 vma->vm_end, 233 flags & VM_READ ? 'r' : '-', 234 flags & VM_WRITE ? 'w' : '-', 235 flags & VM_EXEC ? 'x' : '-', 236 flags & VM_MAYSHARE ? 's' : 'p', 237 pgoff, 238 MAJOR(dev), MINOR(dev), ino, &len); 239 240 /* 241 * Print the dentry name for named mappings, and a 242 * special [heap] marker for the heap: 243 */ 244 if (file) { 245 pad_len_spaces(m, len); 246 seq_path(m, &file->f_path, "\n"); 247 } else { 248 const char *name = arch_vma_name(vma); 249 if (!name) { 250 if (mm) { 251 if (vma->vm_start <= mm->start_brk && 252 vma->vm_end >= mm->brk) { 253 name = "[heap]"; 254 } else if (vma->vm_start <= mm->start_stack && 255 vma->vm_end >= mm->start_stack) { 256 name = "[stack]"; 257 } 258 } else { 259 name = "[vdso]"; 260 } 261 } 262 if (name) { 263 pad_len_spaces(m, len); 264 seq_puts(m, name); 265 } 266 } 267 seq_putc(m, '\n'); 268 } 269 270 static int show_map(struct seq_file *m, void *v) 271 { 272 struct vm_area_struct *vma = v; 273 struct proc_maps_private *priv = m->private; 274 struct task_struct *task = priv->task; 275 276 show_map_vma(m, vma); 277 278 if (m->count < m->size) /* vma is copied successfully */ 279 m->version = (vma != get_gate_vma(task))? vma->vm_start: 0; 280 return 0; 281 } 282 283 static const struct seq_operations proc_pid_maps_op = { 284 .start = m_start, 285 .next = m_next, 286 .stop = m_stop, 287 .show = show_map 288 }; 289 290 static int maps_open(struct inode *inode, struct file *file) 291 { 292 return do_maps_open(inode, file, &proc_pid_maps_op); 293 } 294 295 const struct file_operations proc_maps_operations = { 296 .open = maps_open, 297 .read = seq_read, 298 .llseek = seq_lseek, 299 .release = seq_release_private, 300 }; 301 302 /* 303 * Proportional Set Size(PSS): my share of RSS. 304 * 305 * PSS of a process is the count of pages it has in memory, where each 306 * page is divided by the number of processes sharing it. So if a 307 * process has 1000 pages all to itself, and 1000 shared with one other 308 * process, its PSS will be 1500. 309 * 310 * To keep (accumulated) division errors low, we adopt a 64bit 311 * fixed-point pss counter to minimize division errors. So (pss >> 312 * PSS_SHIFT) would be the real byte count. 313 * 314 * A shift of 12 before division means (assuming 4K page size): 315 * - 1M 3-user-pages add up to 8KB errors; 316 * - supports mapcount up to 2^24, or 16M; 317 * - supports PSS up to 2^52 bytes, or 4PB. 318 */ 319 #define PSS_SHIFT 12 320 321 #ifdef CONFIG_PROC_PAGE_MONITOR 322 struct mem_size_stats { 323 struct vm_area_struct *vma; 324 unsigned long resident; 325 unsigned long shared_clean; 326 unsigned long shared_dirty; 327 unsigned long private_clean; 328 unsigned long private_dirty; 329 unsigned long referenced; 330 unsigned long swap; 331 u64 pss; 332 }; 333 334 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, 335 struct mm_walk *walk) 336 { 337 struct mem_size_stats *mss = walk->private; 338 struct vm_area_struct *vma = mss->vma; 339 pte_t *pte, ptent; 340 spinlock_t *ptl; 341 struct page *page; 342 int mapcount; 343 344 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); 345 for (; addr != end; pte++, addr += PAGE_SIZE) { 346 ptent = *pte; 347 348 if (is_swap_pte(ptent)) { 349 mss->swap += PAGE_SIZE; 350 continue; 351 } 352 353 if (!pte_present(ptent)) 354 continue; 355 356 page = vm_normal_page(vma, addr, ptent); 357 if (!page) 358 continue; 359 360 mss->resident += PAGE_SIZE; 361 /* Accumulate the size in pages that have been accessed. */ 362 if (pte_young(ptent) || PageReferenced(page)) 363 mss->referenced += PAGE_SIZE; 364 mapcount = page_mapcount(page); 365 if (mapcount >= 2) { 366 if (pte_dirty(ptent) || PageDirty(page)) 367 mss->shared_dirty += PAGE_SIZE; 368 else 369 mss->shared_clean += PAGE_SIZE; 370 mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount; 371 } else { 372 if (pte_dirty(ptent) || PageDirty(page)) 373 mss->private_dirty += PAGE_SIZE; 374 else 375 mss->private_clean += PAGE_SIZE; 376 mss->pss += (PAGE_SIZE << PSS_SHIFT); 377 } 378 } 379 pte_unmap_unlock(pte - 1, ptl); 380 cond_resched(); 381 return 0; 382 } 383 384 static int show_smap(struct seq_file *m, void *v) 385 { 386 struct proc_maps_private *priv = m->private; 387 struct task_struct *task = priv->task; 388 struct vm_area_struct *vma = v; 389 struct mem_size_stats mss; 390 struct mm_walk smaps_walk = { 391 .pmd_entry = smaps_pte_range, 392 .mm = vma->vm_mm, 393 .private = &mss, 394 }; 395 396 memset(&mss, 0, sizeof mss); 397 mss.vma = vma; 398 /* mmap_sem is held in m_start */ 399 if (vma->vm_mm && !is_vm_hugetlb_page(vma)) 400 walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk); 401 402 show_map_vma(m, vma); 403 404 seq_printf(m, 405 "Size: %8lu kB\n" 406 "Rss: %8lu kB\n" 407 "Pss: %8lu kB\n" 408 "Shared_Clean: %8lu kB\n" 409 "Shared_Dirty: %8lu kB\n" 410 "Private_Clean: %8lu kB\n" 411 "Private_Dirty: %8lu kB\n" 412 "Referenced: %8lu kB\n" 413 "Swap: %8lu kB\n" 414 "KernelPageSize: %8lu kB\n" 415 "MMUPageSize: %8lu kB\n", 416 (vma->vm_end - vma->vm_start) >> 10, 417 mss.resident >> 10, 418 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)), 419 mss.shared_clean >> 10, 420 mss.shared_dirty >> 10, 421 mss.private_clean >> 10, 422 mss.private_dirty >> 10, 423 mss.referenced >> 10, 424 mss.swap >> 10, 425 vma_kernel_pagesize(vma) >> 10, 426 vma_mmu_pagesize(vma) >> 10); 427 428 if (m->count < m->size) /* vma is copied successfully */ 429 m->version = (vma != get_gate_vma(task)) ? vma->vm_start : 0; 430 return 0; 431 } 432 433 static const struct seq_operations proc_pid_smaps_op = { 434 .start = m_start, 435 .next = m_next, 436 .stop = m_stop, 437 .show = show_smap 438 }; 439 440 static int smaps_open(struct inode *inode, struct file *file) 441 { 442 return do_maps_open(inode, file, &proc_pid_smaps_op); 443 } 444 445 const struct file_operations proc_smaps_operations = { 446 .open = smaps_open, 447 .read = seq_read, 448 .llseek = seq_lseek, 449 .release = seq_release_private, 450 }; 451 452 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr, 453 unsigned long end, struct mm_walk *walk) 454 { 455 struct vm_area_struct *vma = walk->private; 456 pte_t *pte, ptent; 457 spinlock_t *ptl; 458 struct page *page; 459 460 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); 461 for (; addr != end; pte++, addr += PAGE_SIZE) { 462 ptent = *pte; 463 if (!pte_present(ptent)) 464 continue; 465 466 page = vm_normal_page(vma, addr, ptent); 467 if (!page) 468 continue; 469 470 /* Clear accessed and referenced bits. */ 471 ptep_test_and_clear_young(vma, addr, pte); 472 ClearPageReferenced(page); 473 } 474 pte_unmap_unlock(pte - 1, ptl); 475 cond_resched(); 476 return 0; 477 } 478 479 #define CLEAR_REFS_ALL 1 480 #define CLEAR_REFS_ANON 2 481 #define CLEAR_REFS_MAPPED 3 482 483 static ssize_t clear_refs_write(struct file *file, const char __user *buf, 484 size_t count, loff_t *ppos) 485 { 486 struct task_struct *task; 487 char buffer[PROC_NUMBUF]; 488 struct mm_struct *mm; 489 struct vm_area_struct *vma; 490 long type; 491 492 memset(buffer, 0, sizeof(buffer)); 493 if (count > sizeof(buffer) - 1) 494 count = sizeof(buffer) - 1; 495 if (copy_from_user(buffer, buf, count)) 496 return -EFAULT; 497 if (strict_strtol(strstrip(buffer), 10, &type)) 498 return -EINVAL; 499 if (type < CLEAR_REFS_ALL || type > CLEAR_REFS_MAPPED) 500 return -EINVAL; 501 task = get_proc_task(file->f_path.dentry->d_inode); 502 if (!task) 503 return -ESRCH; 504 mm = get_task_mm(task); 505 if (mm) { 506 struct mm_walk clear_refs_walk = { 507 .pmd_entry = clear_refs_pte_range, 508 .mm = mm, 509 }; 510 down_read(&mm->mmap_sem); 511 for (vma = mm->mmap; vma; vma = vma->vm_next) { 512 clear_refs_walk.private = vma; 513 if (is_vm_hugetlb_page(vma)) 514 continue; 515 /* 516 * Writing 1 to /proc/pid/clear_refs affects all pages. 517 * 518 * Writing 2 to /proc/pid/clear_refs only affects 519 * Anonymous pages. 520 * 521 * Writing 3 to /proc/pid/clear_refs only affects file 522 * mapped pages. 523 */ 524 if (type == CLEAR_REFS_ANON && vma->vm_file) 525 continue; 526 if (type == CLEAR_REFS_MAPPED && !vma->vm_file) 527 continue; 528 walk_page_range(vma->vm_start, vma->vm_end, 529 &clear_refs_walk); 530 } 531 flush_tlb_mm(mm); 532 up_read(&mm->mmap_sem); 533 mmput(mm); 534 } 535 put_task_struct(task); 536 537 return count; 538 } 539 540 const struct file_operations proc_clear_refs_operations = { 541 .write = clear_refs_write, 542 .llseek = noop_llseek, 543 }; 544 545 struct pagemapread { 546 int pos, len; 547 u64 *buffer; 548 }; 549 550 #define PM_ENTRY_BYTES sizeof(u64) 551 #define PM_STATUS_BITS 3 552 #define PM_STATUS_OFFSET (64 - PM_STATUS_BITS) 553 #define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET) 554 #define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK) 555 #define PM_PSHIFT_BITS 6 556 #define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS) 557 #define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET) 558 #define PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK) 559 #define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1) 560 #define PM_PFRAME(x) ((x) & PM_PFRAME_MASK) 561 562 #define PM_PRESENT PM_STATUS(4LL) 563 #define PM_SWAP PM_STATUS(2LL) 564 #define PM_NOT_PRESENT PM_PSHIFT(PAGE_SHIFT) 565 #define PM_END_OF_BUFFER 1 566 567 static int add_to_pagemap(unsigned long addr, u64 pfn, 568 struct pagemapread *pm) 569 { 570 pm->buffer[pm->pos++] = pfn; 571 if (pm->pos >= pm->len) 572 return PM_END_OF_BUFFER; 573 return 0; 574 } 575 576 static int pagemap_pte_hole(unsigned long start, unsigned long end, 577 struct mm_walk *walk) 578 { 579 struct pagemapread *pm = walk->private; 580 unsigned long addr; 581 int err = 0; 582 for (addr = start; addr < end; addr += PAGE_SIZE) { 583 err = add_to_pagemap(addr, PM_NOT_PRESENT, pm); 584 if (err) 585 break; 586 } 587 return err; 588 } 589 590 static u64 swap_pte_to_pagemap_entry(pte_t pte) 591 { 592 swp_entry_t e = pte_to_swp_entry(pte); 593 return swp_type(e) | (swp_offset(e) << MAX_SWAPFILES_SHIFT); 594 } 595 596 static u64 pte_to_pagemap_entry(pte_t pte) 597 { 598 u64 pme = 0; 599 if (is_swap_pte(pte)) 600 pme = PM_PFRAME(swap_pte_to_pagemap_entry(pte)) 601 | PM_PSHIFT(PAGE_SHIFT) | PM_SWAP; 602 else if (pte_present(pte)) 603 pme = PM_PFRAME(pte_pfn(pte)) 604 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT; 605 return pme; 606 } 607 608 static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, 609 struct mm_walk *walk) 610 { 611 struct vm_area_struct *vma; 612 struct pagemapread *pm = walk->private; 613 pte_t *pte; 614 int err = 0; 615 616 /* find the first VMA at or above 'addr' */ 617 vma = find_vma(walk->mm, addr); 618 for (; addr != end; addr += PAGE_SIZE) { 619 u64 pfn = PM_NOT_PRESENT; 620 621 /* check to see if we've left 'vma' behind 622 * and need a new, higher one */ 623 if (vma && (addr >= vma->vm_end)) 624 vma = find_vma(walk->mm, addr); 625 626 /* check that 'vma' actually covers this address, 627 * and that it isn't a huge page vma */ 628 if (vma && (vma->vm_start <= addr) && 629 !is_vm_hugetlb_page(vma)) { 630 pte = pte_offset_map(pmd, addr); 631 pfn = pte_to_pagemap_entry(*pte); 632 /* unmap before userspace copy */ 633 pte_unmap(pte); 634 } 635 err = add_to_pagemap(addr, pfn, pm); 636 if (err) 637 return err; 638 } 639 640 cond_resched(); 641 642 return err; 643 } 644 645 #ifdef CONFIG_HUGETLB_PAGE 646 static u64 huge_pte_to_pagemap_entry(pte_t pte, int offset) 647 { 648 u64 pme = 0; 649 if (pte_present(pte)) 650 pme = PM_PFRAME(pte_pfn(pte) + offset) 651 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT; 652 return pme; 653 } 654 655 /* This function walks within one hugetlb entry in the single call */ 656 static int pagemap_hugetlb_range(pte_t *pte, unsigned long hmask, 657 unsigned long addr, unsigned long end, 658 struct mm_walk *walk) 659 { 660 struct pagemapread *pm = walk->private; 661 int err = 0; 662 u64 pfn; 663 664 for (; addr != end; addr += PAGE_SIZE) { 665 int offset = (addr & ~hmask) >> PAGE_SHIFT; 666 pfn = huge_pte_to_pagemap_entry(*pte, offset); 667 err = add_to_pagemap(addr, pfn, pm); 668 if (err) 669 return err; 670 } 671 672 cond_resched(); 673 674 return err; 675 } 676 #endif /* HUGETLB_PAGE */ 677 678 /* 679 * /proc/pid/pagemap - an array mapping virtual pages to pfns 680 * 681 * For each page in the address space, this file contains one 64-bit entry 682 * consisting of the following: 683 * 684 * Bits 0-55 page frame number (PFN) if present 685 * Bits 0-4 swap type if swapped 686 * Bits 5-55 swap offset if swapped 687 * Bits 55-60 page shift (page size = 1<<page shift) 688 * Bit 61 reserved for future use 689 * Bit 62 page swapped 690 * Bit 63 page present 691 * 692 * If the page is not present but in swap, then the PFN contains an 693 * encoding of the swap file number and the page's offset into the 694 * swap. Unmapped pages return a null PFN. This allows determining 695 * precisely which pages are mapped (or in swap) and comparing mapped 696 * pages between processes. 697 * 698 * Efficient users of this interface will use /proc/pid/maps to 699 * determine which areas of memory are actually mapped and llseek to 700 * skip over unmapped regions. 701 */ 702 #define PAGEMAP_WALK_SIZE (PMD_SIZE) 703 static ssize_t pagemap_read(struct file *file, char __user *buf, 704 size_t count, loff_t *ppos) 705 { 706 struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode); 707 struct mm_struct *mm; 708 struct pagemapread pm; 709 int ret = -ESRCH; 710 struct mm_walk pagemap_walk = {}; 711 unsigned long src; 712 unsigned long svpfn; 713 unsigned long start_vaddr; 714 unsigned long end_vaddr; 715 int copied = 0; 716 717 if (!task) 718 goto out; 719 720 ret = -EACCES; 721 if (!ptrace_may_access(task, PTRACE_MODE_READ)) 722 goto out_task; 723 724 ret = -EINVAL; 725 /* file position must be aligned */ 726 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES)) 727 goto out_task; 728 729 ret = 0; 730 731 if (!count) 732 goto out_task; 733 734 mm = get_task_mm(task); 735 if (!mm) 736 goto out_task; 737 738 pm.len = PM_ENTRY_BYTES * (PAGEMAP_WALK_SIZE >> PAGE_SHIFT); 739 pm.buffer = kmalloc(pm.len, GFP_TEMPORARY); 740 ret = -ENOMEM; 741 if (!pm.buffer) 742 goto out_mm; 743 744 pagemap_walk.pmd_entry = pagemap_pte_range; 745 pagemap_walk.pte_hole = pagemap_pte_hole; 746 #ifdef CONFIG_HUGETLB_PAGE 747 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range; 748 #endif 749 pagemap_walk.mm = mm; 750 pagemap_walk.private = ± 751 752 src = *ppos; 753 svpfn = src / PM_ENTRY_BYTES; 754 start_vaddr = svpfn << PAGE_SHIFT; 755 end_vaddr = TASK_SIZE_OF(task); 756 757 /* watch out for wraparound */ 758 if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT) 759 start_vaddr = end_vaddr; 760 761 /* 762 * The odds are that this will stop walking way 763 * before end_vaddr, because the length of the 764 * user buffer is tracked in "pm", and the walk 765 * will stop when we hit the end of the buffer. 766 */ 767 ret = 0; 768 while (count && (start_vaddr < end_vaddr)) { 769 int len; 770 unsigned long end; 771 772 pm.pos = 0; 773 end = start_vaddr + PAGEMAP_WALK_SIZE; 774 /* overflow ? */ 775 if (end < start_vaddr || end > end_vaddr) 776 end = end_vaddr; 777 down_read(&mm->mmap_sem); 778 ret = walk_page_range(start_vaddr, end, &pagemap_walk); 779 up_read(&mm->mmap_sem); 780 start_vaddr = end; 781 782 len = min(count, PM_ENTRY_BYTES * pm.pos); 783 if (copy_to_user(buf, pm.buffer, len)) { 784 ret = -EFAULT; 785 goto out_free; 786 } 787 copied += len; 788 buf += len; 789 count -= len; 790 } 791 *ppos += copied; 792 if (!ret || ret == PM_END_OF_BUFFER) 793 ret = copied; 794 795 out_free: 796 kfree(pm.buffer); 797 out_mm: 798 mmput(mm); 799 out_task: 800 put_task_struct(task); 801 out: 802 return ret; 803 } 804 805 const struct file_operations proc_pagemap_operations = { 806 .llseek = mem_lseek, /* borrow this */ 807 .read = pagemap_read, 808 }; 809 #endif /* CONFIG_PROC_PAGE_MONITOR */ 810 811 #ifdef CONFIG_NUMA 812 extern int show_numa_map(struct seq_file *m, void *v); 813 814 static const struct seq_operations proc_pid_numa_maps_op = { 815 .start = m_start, 816 .next = m_next, 817 .stop = m_stop, 818 .show = show_numa_map, 819 }; 820 821 static int numa_maps_open(struct inode *inode, struct file *file) 822 { 823 return do_maps_open(inode, file, &proc_pid_numa_maps_op); 824 } 825 826 const struct file_operations proc_numa_maps_operations = { 827 .open = numa_maps_open, 828 .read = seq_read, 829 .llseek = seq_lseek, 830 .release = seq_release_private, 831 }; 832 #endif 833