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/pagemap.h> 8 #include <linux/mempolicy.h> 9 #include <linux/swap.h> 10 #include <linux/swapops.h> 11 12 #include <asm/elf.h> 13 #include <asm/uaccess.h> 14 #include <asm/tlbflush.h> 15 #include "internal.h" 16 17 void task_mem(struct seq_file *m, struct mm_struct *mm) 18 { 19 unsigned long data, text, lib; 20 unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss; 21 22 /* 23 * Note: to minimize their overhead, mm maintains hiwater_vm and 24 * hiwater_rss only when about to *lower* total_vm or rss. Any 25 * collector of these hiwater stats must therefore get total_vm 26 * and rss too, which will usually be the higher. Barriers? not 27 * worth the effort, such snapshots can always be inconsistent. 28 */ 29 hiwater_vm = total_vm = mm->total_vm; 30 if (hiwater_vm < mm->hiwater_vm) 31 hiwater_vm = mm->hiwater_vm; 32 hiwater_rss = total_rss = get_mm_rss(mm); 33 if (hiwater_rss < mm->hiwater_rss) 34 hiwater_rss = mm->hiwater_rss; 35 36 data = mm->total_vm - mm->shared_vm - mm->stack_vm; 37 text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10; 38 lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text; 39 seq_printf(m, 40 "VmPeak:\t%8lu kB\n" 41 "VmSize:\t%8lu kB\n" 42 "VmLck:\t%8lu kB\n" 43 "VmHWM:\t%8lu kB\n" 44 "VmRSS:\t%8lu kB\n" 45 "VmData:\t%8lu kB\n" 46 "VmStk:\t%8lu kB\n" 47 "VmExe:\t%8lu kB\n" 48 "VmLib:\t%8lu kB\n" 49 "VmPTE:\t%8lu kB\n", 50 hiwater_vm << (PAGE_SHIFT-10), 51 (total_vm - mm->reserved_vm) << (PAGE_SHIFT-10), 52 mm->locked_vm << (PAGE_SHIFT-10), 53 hiwater_rss << (PAGE_SHIFT-10), 54 total_rss << (PAGE_SHIFT-10), 55 data << (PAGE_SHIFT-10), 56 mm->stack_vm << (PAGE_SHIFT-10), text, lib, 57 (PTRS_PER_PTE*sizeof(pte_t)*mm->nr_ptes) >> 10); 58 } 59 60 unsigned long task_vsize(struct mm_struct *mm) 61 { 62 return PAGE_SIZE * mm->total_vm; 63 } 64 65 int task_statm(struct mm_struct *mm, int *shared, int *text, 66 int *data, int *resident) 67 { 68 *shared = get_mm_counter(mm, file_rss); 69 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) 70 >> PAGE_SHIFT; 71 *data = mm->total_vm - mm->shared_vm; 72 *resident = *shared + get_mm_counter(mm, anon_rss); 73 return mm->total_vm; 74 } 75 76 static void pad_len_spaces(struct seq_file *m, int len) 77 { 78 len = 25 + sizeof(void*) * 6 - len; 79 if (len < 1) 80 len = 1; 81 seq_printf(m, "%*c", len, ' '); 82 } 83 84 static void vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma) 85 { 86 if (vma && vma != priv->tail_vma) { 87 struct mm_struct *mm = vma->vm_mm; 88 up_read(&mm->mmap_sem); 89 mmput(mm); 90 } 91 } 92 93 static void *m_start(struct seq_file *m, loff_t *pos) 94 { 95 struct proc_maps_private *priv = m->private; 96 unsigned long last_addr = m->version; 97 struct mm_struct *mm; 98 struct vm_area_struct *vma, *tail_vma = NULL; 99 loff_t l = *pos; 100 101 /* Clear the per syscall fields in priv */ 102 priv->task = NULL; 103 priv->tail_vma = NULL; 104 105 /* 106 * We remember last_addr rather than next_addr to hit with 107 * mmap_cache most of the time. We have zero last_addr at 108 * the beginning and also after lseek. We will have -1 last_addr 109 * after the end of the vmas. 110 */ 111 112 if (last_addr == -1UL) 113 return NULL; 114 115 priv->task = get_pid_task(priv->pid, PIDTYPE_PID); 116 if (!priv->task) 117 return NULL; 118 119 mm = mm_for_maps(priv->task); 120 if (!mm) 121 return NULL; 122 down_read(&mm->mmap_sem); 123 124 tail_vma = get_gate_vma(priv->task); 125 priv->tail_vma = tail_vma; 126 127 /* Start with last addr hint */ 128 vma = find_vma(mm, last_addr); 129 if (last_addr && vma) { 130 vma = vma->vm_next; 131 goto out; 132 } 133 134 /* 135 * Check the vma index is within the range and do 136 * sequential scan until m_index. 137 */ 138 vma = NULL; 139 if ((unsigned long)l < mm->map_count) { 140 vma = mm->mmap; 141 while (l-- && vma) 142 vma = vma->vm_next; 143 goto out; 144 } 145 146 if (l != mm->map_count) 147 tail_vma = NULL; /* After gate vma */ 148 149 out: 150 if (vma) 151 return vma; 152 153 /* End of vmas has been reached */ 154 m->version = (tail_vma != NULL)? 0: -1UL; 155 up_read(&mm->mmap_sem); 156 mmput(mm); 157 return tail_vma; 158 } 159 160 static void *m_next(struct seq_file *m, void *v, loff_t *pos) 161 { 162 struct proc_maps_private *priv = m->private; 163 struct vm_area_struct *vma = v; 164 struct vm_area_struct *tail_vma = priv->tail_vma; 165 166 (*pos)++; 167 if (vma && (vma != tail_vma) && vma->vm_next) 168 return vma->vm_next; 169 vma_stop(priv, vma); 170 return (vma != tail_vma)? tail_vma: NULL; 171 } 172 173 static void m_stop(struct seq_file *m, void *v) 174 { 175 struct proc_maps_private *priv = m->private; 176 struct vm_area_struct *vma = v; 177 178 vma_stop(priv, vma); 179 if (priv->task) 180 put_task_struct(priv->task); 181 } 182 183 static int do_maps_open(struct inode *inode, struct file *file, 184 const struct seq_operations *ops) 185 { 186 struct proc_maps_private *priv; 187 int ret = -ENOMEM; 188 priv = kzalloc(sizeof(*priv), GFP_KERNEL); 189 if (priv) { 190 priv->pid = proc_pid(inode); 191 ret = seq_open(file, ops); 192 if (!ret) { 193 struct seq_file *m = file->private_data; 194 m->private = priv; 195 } else { 196 kfree(priv); 197 } 198 } 199 return ret; 200 } 201 202 static void show_map_vma(struct seq_file *m, struct vm_area_struct *vma) 203 { 204 struct mm_struct *mm = vma->vm_mm; 205 struct file *file = vma->vm_file; 206 int flags = vma->vm_flags; 207 unsigned long ino = 0; 208 unsigned long long pgoff = 0; 209 dev_t dev = 0; 210 int len; 211 212 if (file) { 213 struct inode *inode = vma->vm_file->f_path.dentry->d_inode; 214 dev = inode->i_sb->s_dev; 215 ino = inode->i_ino; 216 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT; 217 } 218 219 seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu %n", 220 vma->vm_start, 221 vma->vm_end, 222 flags & VM_READ ? 'r' : '-', 223 flags & VM_WRITE ? 'w' : '-', 224 flags & VM_EXEC ? 'x' : '-', 225 flags & VM_MAYSHARE ? 's' : 'p', 226 pgoff, 227 MAJOR(dev), MINOR(dev), ino, &len); 228 229 /* 230 * Print the dentry name for named mappings, and a 231 * special [heap] marker for the heap: 232 */ 233 if (file) { 234 pad_len_spaces(m, len); 235 seq_path(m, &file->f_path, "\n"); 236 } else { 237 const char *name = arch_vma_name(vma); 238 if (!name) { 239 if (mm) { 240 if (vma->vm_start <= mm->start_brk && 241 vma->vm_end >= mm->brk) { 242 name = "[heap]"; 243 } else if (vma->vm_start <= mm->start_stack && 244 vma->vm_end >= mm->start_stack) { 245 name = "[stack]"; 246 } else { 247 unsigned long stack_start; 248 struct proc_maps_private *pmp; 249 250 pmp = m->private; 251 stack_start = pmp->task->stack_start; 252 253 if (vma->vm_start <= stack_start && 254 vma->vm_end >= stack_start) { 255 pad_len_spaces(m, len); 256 seq_printf(m, 257 "[threadstack:%08lx]", 258 #ifdef CONFIG_STACK_GROWSUP 259 vma->vm_end - stack_start 260 #else 261 stack_start - vma->vm_start 262 #endif 263 ); 264 } 265 } 266 } else { 267 name = "[vdso]"; 268 } 269 } 270 if (name) { 271 pad_len_spaces(m, len); 272 seq_puts(m, name); 273 } 274 } 275 seq_putc(m, '\n'); 276 } 277 278 static int show_map(struct seq_file *m, void *v) 279 { 280 struct vm_area_struct *vma = v; 281 struct proc_maps_private *priv = m->private; 282 struct task_struct *task = priv->task; 283 284 show_map_vma(m, vma); 285 286 if (m->count < m->size) /* vma is copied successfully */ 287 m->version = (vma != get_gate_vma(task))? vma->vm_start: 0; 288 return 0; 289 } 290 291 static const struct seq_operations proc_pid_maps_op = { 292 .start = m_start, 293 .next = m_next, 294 .stop = m_stop, 295 .show = show_map 296 }; 297 298 static int maps_open(struct inode *inode, struct file *file) 299 { 300 return do_maps_open(inode, file, &proc_pid_maps_op); 301 } 302 303 const struct file_operations proc_maps_operations = { 304 .open = maps_open, 305 .read = seq_read, 306 .llseek = seq_lseek, 307 .release = seq_release_private, 308 }; 309 310 /* 311 * Proportional Set Size(PSS): my share of RSS. 312 * 313 * PSS of a process is the count of pages it has in memory, where each 314 * page is divided by the number of processes sharing it. So if a 315 * process has 1000 pages all to itself, and 1000 shared with one other 316 * process, its PSS will be 1500. 317 * 318 * To keep (accumulated) division errors low, we adopt a 64bit 319 * fixed-point pss counter to minimize division errors. So (pss >> 320 * PSS_SHIFT) would be the real byte count. 321 * 322 * A shift of 12 before division means (assuming 4K page size): 323 * - 1M 3-user-pages add up to 8KB errors; 324 * - supports mapcount up to 2^24, or 16M; 325 * - supports PSS up to 2^52 bytes, or 4PB. 326 */ 327 #define PSS_SHIFT 12 328 329 #ifdef CONFIG_PROC_PAGE_MONITOR 330 struct mem_size_stats { 331 struct vm_area_struct *vma; 332 unsigned long resident; 333 unsigned long shared_clean; 334 unsigned long shared_dirty; 335 unsigned long private_clean; 336 unsigned long private_dirty; 337 unsigned long referenced; 338 unsigned long swap; 339 u64 pss; 340 }; 341 342 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, 343 struct mm_walk *walk) 344 { 345 struct mem_size_stats *mss = walk->private; 346 struct vm_area_struct *vma = mss->vma; 347 pte_t *pte, ptent; 348 spinlock_t *ptl; 349 struct page *page; 350 int mapcount; 351 352 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); 353 for (; addr != end; pte++, addr += PAGE_SIZE) { 354 ptent = *pte; 355 356 if (is_swap_pte(ptent)) { 357 mss->swap += PAGE_SIZE; 358 continue; 359 } 360 361 if (!pte_present(ptent)) 362 continue; 363 364 page = vm_normal_page(vma, addr, ptent); 365 if (!page) 366 continue; 367 368 mss->resident += PAGE_SIZE; 369 /* Accumulate the size in pages that have been accessed. */ 370 if (pte_young(ptent) || PageReferenced(page)) 371 mss->referenced += PAGE_SIZE; 372 mapcount = page_mapcount(page); 373 if (mapcount >= 2) { 374 if (pte_dirty(ptent)) 375 mss->shared_dirty += PAGE_SIZE; 376 else 377 mss->shared_clean += PAGE_SIZE; 378 mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount; 379 } else { 380 if (pte_dirty(ptent)) 381 mss->private_dirty += PAGE_SIZE; 382 else 383 mss->private_clean += PAGE_SIZE; 384 mss->pss += (PAGE_SIZE << PSS_SHIFT); 385 } 386 } 387 pte_unmap_unlock(pte - 1, ptl); 388 cond_resched(); 389 return 0; 390 } 391 392 static int show_smap(struct seq_file *m, void *v) 393 { 394 struct proc_maps_private *priv = m->private; 395 struct task_struct *task = priv->task; 396 struct vm_area_struct *vma = v; 397 struct mem_size_stats mss; 398 struct mm_walk smaps_walk = { 399 .pmd_entry = smaps_pte_range, 400 .mm = vma->vm_mm, 401 .private = &mss, 402 }; 403 404 memset(&mss, 0, sizeof mss); 405 mss.vma = vma; 406 if (vma->vm_mm && !is_vm_hugetlb_page(vma)) 407 walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk); 408 409 show_map_vma(m, vma); 410 411 seq_printf(m, 412 "Size: %8lu kB\n" 413 "Rss: %8lu kB\n" 414 "Pss: %8lu kB\n" 415 "Shared_Clean: %8lu kB\n" 416 "Shared_Dirty: %8lu kB\n" 417 "Private_Clean: %8lu kB\n" 418 "Private_Dirty: %8lu kB\n" 419 "Referenced: %8lu kB\n" 420 "Swap: %8lu kB\n" 421 "KernelPageSize: %8lu kB\n" 422 "MMUPageSize: %8lu kB\n", 423 (vma->vm_end - vma->vm_start) >> 10, 424 mss.resident >> 10, 425 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)), 426 mss.shared_clean >> 10, 427 mss.shared_dirty >> 10, 428 mss.private_clean >> 10, 429 mss.private_dirty >> 10, 430 mss.referenced >> 10, 431 mss.swap >> 10, 432 vma_kernel_pagesize(vma) >> 10, 433 vma_mmu_pagesize(vma) >> 10); 434 435 if (m->count < m->size) /* vma is copied successfully */ 436 m->version = (vma != get_gate_vma(task)) ? vma->vm_start : 0; 437 return 0; 438 } 439 440 static const struct seq_operations proc_pid_smaps_op = { 441 .start = m_start, 442 .next = m_next, 443 .stop = m_stop, 444 .show = show_smap 445 }; 446 447 static int smaps_open(struct inode *inode, struct file *file) 448 { 449 return do_maps_open(inode, file, &proc_pid_smaps_op); 450 } 451 452 const struct file_operations proc_smaps_operations = { 453 .open = smaps_open, 454 .read = seq_read, 455 .llseek = seq_lseek, 456 .release = seq_release_private, 457 }; 458 459 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr, 460 unsigned long end, struct mm_walk *walk) 461 { 462 struct vm_area_struct *vma = walk->private; 463 pte_t *pte, ptent; 464 spinlock_t *ptl; 465 struct page *page; 466 467 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); 468 for (; addr != end; pte++, addr += PAGE_SIZE) { 469 ptent = *pte; 470 if (!pte_present(ptent)) 471 continue; 472 473 page = vm_normal_page(vma, addr, ptent); 474 if (!page) 475 continue; 476 477 /* Clear accessed and referenced bits. */ 478 ptep_test_and_clear_young(vma, addr, pte); 479 ClearPageReferenced(page); 480 } 481 pte_unmap_unlock(pte - 1, ptl); 482 cond_resched(); 483 return 0; 484 } 485 486 #define CLEAR_REFS_ALL 1 487 #define CLEAR_REFS_ANON 2 488 #define CLEAR_REFS_MAPPED 3 489 490 static ssize_t clear_refs_write(struct file *file, const char __user *buf, 491 size_t count, loff_t *ppos) 492 { 493 struct task_struct *task; 494 char buffer[PROC_NUMBUF]; 495 struct mm_struct *mm; 496 struct vm_area_struct *vma; 497 long type; 498 499 memset(buffer, 0, sizeof(buffer)); 500 if (count > sizeof(buffer) - 1) 501 count = sizeof(buffer) - 1; 502 if (copy_from_user(buffer, buf, count)) 503 return -EFAULT; 504 if (strict_strtol(strstrip(buffer), 10, &type)) 505 return -EINVAL; 506 if (type < CLEAR_REFS_ALL || type > CLEAR_REFS_MAPPED) 507 return -EINVAL; 508 task = get_proc_task(file->f_path.dentry->d_inode); 509 if (!task) 510 return -ESRCH; 511 mm = get_task_mm(task); 512 if (mm) { 513 struct mm_walk clear_refs_walk = { 514 .pmd_entry = clear_refs_pte_range, 515 .mm = mm, 516 }; 517 down_read(&mm->mmap_sem); 518 for (vma = mm->mmap; vma; vma = vma->vm_next) { 519 clear_refs_walk.private = vma; 520 if (is_vm_hugetlb_page(vma)) 521 continue; 522 /* 523 * Writing 1 to /proc/pid/clear_refs affects all pages. 524 * 525 * Writing 2 to /proc/pid/clear_refs only affects 526 * Anonymous pages. 527 * 528 * Writing 3 to /proc/pid/clear_refs only affects file 529 * mapped pages. 530 */ 531 if (type == CLEAR_REFS_ANON && vma->vm_file) 532 continue; 533 if (type == CLEAR_REFS_MAPPED && !vma->vm_file) 534 continue; 535 walk_page_range(vma->vm_start, vma->vm_end, 536 &clear_refs_walk); 537 } 538 flush_tlb_mm(mm); 539 up_read(&mm->mmap_sem); 540 mmput(mm); 541 } 542 put_task_struct(task); 543 544 return count; 545 } 546 547 const struct file_operations proc_clear_refs_operations = { 548 .write = clear_refs_write, 549 }; 550 551 struct pagemapread { 552 u64 __user *out, *end; 553 }; 554 555 #define PM_ENTRY_BYTES sizeof(u64) 556 #define PM_STATUS_BITS 3 557 #define PM_STATUS_OFFSET (64 - PM_STATUS_BITS) 558 #define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET) 559 #define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK) 560 #define PM_PSHIFT_BITS 6 561 #define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS) 562 #define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET) 563 #define PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK) 564 #define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1) 565 #define PM_PFRAME(x) ((x) & PM_PFRAME_MASK) 566 567 #define PM_PRESENT PM_STATUS(4LL) 568 #define PM_SWAP PM_STATUS(2LL) 569 #define PM_NOT_PRESENT PM_PSHIFT(PAGE_SHIFT) 570 #define PM_END_OF_BUFFER 1 571 572 static int add_to_pagemap(unsigned long addr, u64 pfn, 573 struct pagemapread *pm) 574 { 575 if (put_user(pfn, pm->out)) 576 return -EFAULT; 577 pm->out++; 578 if (pm->out >= pm->end) 579 return PM_END_OF_BUFFER; 580 return 0; 581 } 582 583 static int pagemap_pte_hole(unsigned long start, unsigned long end, 584 struct mm_walk *walk) 585 { 586 struct pagemapread *pm = walk->private; 587 unsigned long addr; 588 int err = 0; 589 for (addr = start; addr < end; addr += PAGE_SIZE) { 590 err = add_to_pagemap(addr, PM_NOT_PRESENT, pm); 591 if (err) 592 break; 593 } 594 return err; 595 } 596 597 static u64 swap_pte_to_pagemap_entry(pte_t pte) 598 { 599 swp_entry_t e = pte_to_swp_entry(pte); 600 return swp_type(e) | (swp_offset(e) << MAX_SWAPFILES_SHIFT); 601 } 602 603 static u64 pte_to_pagemap_entry(pte_t pte) 604 { 605 u64 pme = 0; 606 if (is_swap_pte(pte)) 607 pme = PM_PFRAME(swap_pte_to_pagemap_entry(pte)) 608 | PM_PSHIFT(PAGE_SHIFT) | PM_SWAP; 609 else if (pte_present(pte)) 610 pme = PM_PFRAME(pte_pfn(pte)) 611 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT; 612 return pme; 613 } 614 615 static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, 616 struct mm_walk *walk) 617 { 618 struct vm_area_struct *vma; 619 struct pagemapread *pm = walk->private; 620 pte_t *pte; 621 int err = 0; 622 623 /* find the first VMA at or above 'addr' */ 624 vma = find_vma(walk->mm, addr); 625 for (; addr != end; addr += PAGE_SIZE) { 626 u64 pfn = PM_NOT_PRESENT; 627 628 /* check to see if we've left 'vma' behind 629 * and need a new, higher one */ 630 if (vma && (addr >= vma->vm_end)) 631 vma = find_vma(walk->mm, addr); 632 633 /* check that 'vma' actually covers this address, 634 * and that it isn't a huge page vma */ 635 if (vma && (vma->vm_start <= addr) && 636 !is_vm_hugetlb_page(vma)) { 637 pte = pte_offset_map(pmd, addr); 638 pfn = pte_to_pagemap_entry(*pte); 639 /* unmap before userspace copy */ 640 pte_unmap(pte); 641 } 642 err = add_to_pagemap(addr, pfn, pm); 643 if (err) 644 return err; 645 } 646 647 cond_resched(); 648 649 return err; 650 } 651 652 static u64 huge_pte_to_pagemap_entry(pte_t pte, int offset) 653 { 654 u64 pme = 0; 655 if (pte_present(pte)) 656 pme = PM_PFRAME(pte_pfn(pte) + offset) 657 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT; 658 return pme; 659 } 660 661 static int pagemap_hugetlb_range(pte_t *pte, unsigned long addr, 662 unsigned long end, struct mm_walk *walk) 663 { 664 struct vm_area_struct *vma; 665 struct pagemapread *pm = walk->private; 666 struct hstate *hs = NULL; 667 int err = 0; 668 669 vma = find_vma(walk->mm, addr); 670 if (vma) 671 hs = hstate_vma(vma); 672 for (; addr != end; addr += PAGE_SIZE) { 673 u64 pfn = PM_NOT_PRESENT; 674 675 if (vma && (addr >= vma->vm_end)) { 676 vma = find_vma(walk->mm, addr); 677 if (vma) 678 hs = hstate_vma(vma); 679 } 680 681 if (vma && (vma->vm_start <= addr) && is_vm_hugetlb_page(vma)) { 682 /* calculate pfn of the "raw" page in the hugepage. */ 683 int offset = (addr & ~huge_page_mask(hs)) >> PAGE_SHIFT; 684 pfn = huge_pte_to_pagemap_entry(*pte, offset); 685 } 686 err = add_to_pagemap(addr, pfn, pm); 687 if (err) 688 return err; 689 } 690 691 cond_resched(); 692 693 return err; 694 } 695 696 /* 697 * /proc/pid/pagemap - an array mapping virtual pages to pfns 698 * 699 * For each page in the address space, this file contains one 64-bit entry 700 * consisting of the following: 701 * 702 * Bits 0-55 page frame number (PFN) if present 703 * Bits 0-4 swap type if swapped 704 * Bits 5-55 swap offset if swapped 705 * Bits 55-60 page shift (page size = 1<<page shift) 706 * Bit 61 reserved for future use 707 * Bit 62 page swapped 708 * Bit 63 page present 709 * 710 * If the page is not present but in swap, then the PFN contains an 711 * encoding of the swap file number and the page's offset into the 712 * swap. Unmapped pages return a null PFN. This allows determining 713 * precisely which pages are mapped (or in swap) and comparing mapped 714 * pages between processes. 715 * 716 * Efficient users of this interface will use /proc/pid/maps to 717 * determine which areas of memory are actually mapped and llseek to 718 * skip over unmapped regions. 719 */ 720 static ssize_t pagemap_read(struct file *file, char __user *buf, 721 size_t count, loff_t *ppos) 722 { 723 struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode); 724 struct page **pages, *page; 725 unsigned long uaddr, uend; 726 struct mm_struct *mm; 727 struct pagemapread pm; 728 int pagecount; 729 int ret = -ESRCH; 730 struct mm_walk pagemap_walk = {}; 731 unsigned long src; 732 unsigned long svpfn; 733 unsigned long start_vaddr; 734 unsigned long end_vaddr; 735 736 if (!task) 737 goto out; 738 739 ret = -EACCES; 740 if (!ptrace_may_access(task, PTRACE_MODE_READ)) 741 goto out_task; 742 743 ret = -EINVAL; 744 /* file position must be aligned */ 745 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES)) 746 goto out_task; 747 748 ret = 0; 749 750 if (!count) 751 goto out_task; 752 753 mm = get_task_mm(task); 754 if (!mm) 755 goto out_task; 756 757 758 uaddr = (unsigned long)buf & PAGE_MASK; 759 uend = (unsigned long)(buf + count); 760 pagecount = (PAGE_ALIGN(uend) - uaddr) / PAGE_SIZE; 761 ret = 0; 762 if (pagecount == 0) 763 goto out_mm; 764 pages = kcalloc(pagecount, sizeof(struct page *), GFP_KERNEL); 765 ret = -ENOMEM; 766 if (!pages) 767 goto out_mm; 768 769 down_read(¤t->mm->mmap_sem); 770 ret = get_user_pages(current, current->mm, uaddr, pagecount, 771 1, 0, pages, NULL); 772 up_read(¤t->mm->mmap_sem); 773 774 if (ret < 0) 775 goto out_free; 776 777 if (ret != pagecount) { 778 pagecount = ret; 779 ret = -EFAULT; 780 goto out_pages; 781 } 782 783 pm.out = (u64 __user *)buf; 784 pm.end = (u64 __user *)(buf + count); 785 786 pagemap_walk.pmd_entry = pagemap_pte_range; 787 pagemap_walk.pte_hole = pagemap_pte_hole; 788 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range; 789 pagemap_walk.mm = mm; 790 pagemap_walk.private = ± 791 792 src = *ppos; 793 svpfn = src / PM_ENTRY_BYTES; 794 start_vaddr = svpfn << PAGE_SHIFT; 795 end_vaddr = TASK_SIZE_OF(task); 796 797 /* watch out for wraparound */ 798 if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT) 799 start_vaddr = end_vaddr; 800 801 /* 802 * The odds are that this will stop walking way 803 * before end_vaddr, because the length of the 804 * user buffer is tracked in "pm", and the walk 805 * will stop when we hit the end of the buffer. 806 */ 807 ret = walk_page_range(start_vaddr, end_vaddr, &pagemap_walk); 808 if (ret == PM_END_OF_BUFFER) 809 ret = 0; 810 /* don't need mmap_sem for these, but this looks cleaner */ 811 *ppos += (char __user *)pm.out - buf; 812 if (!ret) 813 ret = (char __user *)pm.out - buf; 814 815 out_pages: 816 for (; pagecount; pagecount--) { 817 page = pages[pagecount-1]; 818 if (!PageReserved(page)) 819 SetPageDirty(page); 820 page_cache_release(page); 821 } 822 out_free: 823 kfree(pages); 824 out_mm: 825 mmput(mm); 826 out_task: 827 put_task_struct(task); 828 out: 829 return ret; 830 } 831 832 const struct file_operations proc_pagemap_operations = { 833 .llseek = mem_lseek, /* borrow this */ 834 .read = pagemap_read, 835 }; 836 #endif /* CONFIG_PROC_PAGE_MONITOR */ 837 838 #ifdef CONFIG_NUMA 839 extern int show_numa_map(struct seq_file *m, void *v); 840 841 static const struct seq_operations proc_pid_numa_maps_op = { 842 .start = m_start, 843 .next = m_next, 844 .stop = m_stop, 845 .show = show_numa_map, 846 }; 847 848 static int numa_maps_open(struct inode *inode, struct file *file) 849 { 850 return do_maps_open(inode, file, &proc_pid_numa_maps_op); 851 } 852 853 const struct file_operations proc_numa_maps_operations = { 854 .open = numa_maps_open, 855 .read = seq_read, 856 .llseek = seq_lseek, 857 .release = seq_release_private, 858 }; 859 #endif 860