1 // SPDX-License-Identifier: GPL-2.0 2 #include <linux/pagewalk.h> 3 #include <linux/mm_inline.h> 4 #include <linux/hugetlb.h> 5 #include <linux/huge_mm.h> 6 #include <linux/mount.h> 7 #include <linux/ksm.h> 8 #include <linux/seq_file.h> 9 #include <linux/highmem.h> 10 #include <linux/ptrace.h> 11 #include <linux/slab.h> 12 #include <linux/pagemap.h> 13 #include <linux/mempolicy.h> 14 #include <linux/rmap.h> 15 #include <linux/swap.h> 16 #include <linux/sched/mm.h> 17 #include <linux/swapops.h> 18 #include <linux/mmu_notifier.h> 19 #include <linux/page_idle.h> 20 #include <linux/shmem_fs.h> 21 #include <linux/uaccess.h> 22 #include <linux/pkeys.h> 23 #include <linux/minmax.h> 24 #include <linux/overflow.h> 25 #include <linux/buildid.h> 26 27 #include <asm/elf.h> 28 #include <asm/tlb.h> 29 #include <asm/tlbflush.h> 30 #include "internal.h" 31 32 #define SEQ_PUT_DEC(str, val) \ 33 seq_put_decimal_ull_width(m, str, (val) << (PAGE_SHIFT-10), 8) 34 void task_mem(struct seq_file *m, struct mm_struct *mm) 35 { 36 unsigned long text, lib, swap, anon, file, shmem; 37 unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss; 38 39 anon = get_mm_counter(mm, MM_ANONPAGES); 40 file = get_mm_counter(mm, MM_FILEPAGES); 41 shmem = get_mm_counter(mm, MM_SHMEMPAGES); 42 43 /* 44 * Note: to minimize their overhead, mm maintains hiwater_vm and 45 * hiwater_rss only when about to *lower* total_vm or rss. Any 46 * collector of these hiwater stats must therefore get total_vm 47 * and rss too, which will usually be the higher. Barriers? not 48 * worth the effort, such snapshots can always be inconsistent. 49 */ 50 hiwater_vm = total_vm = mm->total_vm; 51 if (hiwater_vm < mm->hiwater_vm) 52 hiwater_vm = mm->hiwater_vm; 53 hiwater_rss = total_rss = anon + file + shmem; 54 if (hiwater_rss < mm->hiwater_rss) 55 hiwater_rss = mm->hiwater_rss; 56 57 /* split executable areas between text and lib */ 58 text = PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK); 59 text = min(text, mm->exec_vm << PAGE_SHIFT); 60 lib = (mm->exec_vm << PAGE_SHIFT) - text; 61 62 swap = get_mm_counter(mm, MM_SWAPENTS); 63 SEQ_PUT_DEC("VmPeak:\t", hiwater_vm); 64 SEQ_PUT_DEC(" kB\nVmSize:\t", total_vm); 65 SEQ_PUT_DEC(" kB\nVmLck:\t", mm->locked_vm); 66 SEQ_PUT_DEC(" kB\nVmPin:\t", atomic64_read(&mm->pinned_vm)); 67 SEQ_PUT_DEC(" kB\nVmHWM:\t", hiwater_rss); 68 SEQ_PUT_DEC(" kB\nVmRSS:\t", total_rss); 69 SEQ_PUT_DEC(" kB\nRssAnon:\t", anon); 70 SEQ_PUT_DEC(" kB\nRssFile:\t", file); 71 SEQ_PUT_DEC(" kB\nRssShmem:\t", shmem); 72 SEQ_PUT_DEC(" kB\nVmData:\t", mm->data_vm); 73 SEQ_PUT_DEC(" kB\nVmStk:\t", mm->stack_vm); 74 seq_put_decimal_ull_width(m, 75 " kB\nVmExe:\t", text >> 10, 8); 76 seq_put_decimal_ull_width(m, 77 " kB\nVmLib:\t", lib >> 10, 8); 78 seq_put_decimal_ull_width(m, 79 " kB\nVmPTE:\t", mm_pgtables_bytes(mm) >> 10, 8); 80 SEQ_PUT_DEC(" kB\nVmSwap:\t", swap); 81 seq_puts(m, " kB\n"); 82 hugetlb_report_usage(m, mm); 83 } 84 #undef SEQ_PUT_DEC 85 86 unsigned long task_vsize(struct mm_struct *mm) 87 { 88 return PAGE_SIZE * mm->total_vm; 89 } 90 91 unsigned long task_statm(struct mm_struct *mm, 92 unsigned long *shared, unsigned long *text, 93 unsigned long *data, unsigned long *resident) 94 { 95 *shared = get_mm_counter(mm, MM_FILEPAGES) + 96 get_mm_counter(mm, MM_SHMEMPAGES); 97 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) 98 >> PAGE_SHIFT; 99 *data = mm->data_vm + mm->stack_vm; 100 *resident = *shared + get_mm_counter(mm, MM_ANONPAGES); 101 return mm->total_vm; 102 } 103 104 #ifdef CONFIG_NUMA 105 /* 106 * Save get_task_policy() for show_numa_map(). 107 */ 108 static void hold_task_mempolicy(struct proc_maps_private *priv) 109 { 110 struct task_struct *task = priv->task; 111 112 task_lock(task); 113 priv->task_mempolicy = get_task_policy(task); 114 mpol_get(priv->task_mempolicy); 115 task_unlock(task); 116 } 117 static void release_task_mempolicy(struct proc_maps_private *priv) 118 { 119 mpol_put(priv->task_mempolicy); 120 } 121 #else 122 static void hold_task_mempolicy(struct proc_maps_private *priv) 123 { 124 } 125 static void release_task_mempolicy(struct proc_maps_private *priv) 126 { 127 } 128 #endif 129 130 static struct vm_area_struct *proc_get_vma(struct proc_maps_private *priv, 131 loff_t *ppos) 132 { 133 struct vm_area_struct *vma = vma_next(&priv->iter); 134 135 if (vma) { 136 *ppos = vma->vm_start; 137 } else { 138 *ppos = -2UL; 139 vma = get_gate_vma(priv->mm); 140 } 141 142 return vma; 143 } 144 145 static void *m_start(struct seq_file *m, loff_t *ppos) 146 { 147 struct proc_maps_private *priv = m->private; 148 unsigned long last_addr = *ppos; 149 struct mm_struct *mm; 150 151 /* See m_next(). Zero at the start or after lseek. */ 152 if (last_addr == -1UL) 153 return NULL; 154 155 priv->task = get_proc_task(priv->inode); 156 if (!priv->task) 157 return ERR_PTR(-ESRCH); 158 159 mm = priv->mm; 160 if (!mm || !mmget_not_zero(mm)) { 161 put_task_struct(priv->task); 162 priv->task = NULL; 163 return NULL; 164 } 165 166 if (mmap_read_lock_killable(mm)) { 167 mmput(mm); 168 put_task_struct(priv->task); 169 priv->task = NULL; 170 return ERR_PTR(-EINTR); 171 } 172 173 vma_iter_init(&priv->iter, mm, last_addr); 174 hold_task_mempolicy(priv); 175 if (last_addr == -2UL) 176 return get_gate_vma(mm); 177 178 return proc_get_vma(priv, ppos); 179 } 180 181 static void *m_next(struct seq_file *m, void *v, loff_t *ppos) 182 { 183 if (*ppos == -2UL) { 184 *ppos = -1UL; 185 return NULL; 186 } 187 return proc_get_vma(m->private, ppos); 188 } 189 190 static void m_stop(struct seq_file *m, void *v) 191 { 192 struct proc_maps_private *priv = m->private; 193 struct mm_struct *mm = priv->mm; 194 195 if (!priv->task) 196 return; 197 198 release_task_mempolicy(priv); 199 mmap_read_unlock(mm); 200 mmput(mm); 201 put_task_struct(priv->task); 202 priv->task = NULL; 203 } 204 205 static int proc_maps_open(struct inode *inode, struct file *file, 206 const struct seq_operations *ops, int psize) 207 { 208 struct proc_maps_private *priv = __seq_open_private(file, ops, psize); 209 210 if (!priv) 211 return -ENOMEM; 212 213 priv->inode = inode; 214 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ); 215 if (IS_ERR_OR_NULL(priv->mm)) { 216 int err = priv->mm ? PTR_ERR(priv->mm) : -ESRCH; 217 218 seq_release_private(inode, file); 219 return err; 220 } 221 222 return 0; 223 } 224 225 static int proc_map_release(struct inode *inode, struct file *file) 226 { 227 struct seq_file *seq = file->private_data; 228 struct proc_maps_private *priv = seq->private; 229 230 if (priv->mm) 231 mmdrop(priv->mm); 232 233 return seq_release_private(inode, file); 234 } 235 236 static int do_maps_open(struct inode *inode, struct file *file, 237 const struct seq_operations *ops) 238 { 239 return proc_maps_open(inode, file, ops, 240 sizeof(struct proc_maps_private)); 241 } 242 243 static void get_vma_name(struct vm_area_struct *vma, 244 const struct path **path, 245 const char **name, 246 const char **name_fmt) 247 { 248 struct anon_vma_name *anon_name = vma->vm_mm ? anon_vma_name(vma) : NULL; 249 250 *name = NULL; 251 *path = NULL; 252 *name_fmt = NULL; 253 254 /* 255 * Print the dentry name for named mappings, and a 256 * special [heap] marker for the heap: 257 */ 258 if (vma->vm_file) { 259 /* 260 * If user named this anon shared memory via 261 * prctl(PR_SET_VMA ..., use the provided name. 262 */ 263 if (anon_name) { 264 *name_fmt = "[anon_shmem:%s]"; 265 *name = anon_name->name; 266 } else { 267 *path = file_user_path(vma->vm_file); 268 } 269 return; 270 } 271 272 if (vma->vm_ops && vma->vm_ops->name) { 273 *name = vma->vm_ops->name(vma); 274 if (*name) 275 return; 276 } 277 278 *name = arch_vma_name(vma); 279 if (*name) 280 return; 281 282 if (!vma->vm_mm) { 283 *name = "[vdso]"; 284 return; 285 } 286 287 if (vma_is_initial_heap(vma)) { 288 *name = "[heap]"; 289 return; 290 } 291 292 if (vma_is_initial_stack(vma)) { 293 *name = "[stack]"; 294 return; 295 } 296 297 if (anon_name) { 298 *name_fmt = "[anon:%s]"; 299 *name = anon_name->name; 300 return; 301 } 302 } 303 304 static void show_vma_header_prefix(struct seq_file *m, 305 unsigned long start, unsigned long end, 306 vm_flags_t flags, unsigned long long pgoff, 307 dev_t dev, unsigned long ino) 308 { 309 seq_setwidth(m, 25 + sizeof(void *) * 6 - 1); 310 seq_put_hex_ll(m, NULL, start, 8); 311 seq_put_hex_ll(m, "-", end, 8); 312 seq_putc(m, ' '); 313 seq_putc(m, flags & VM_READ ? 'r' : '-'); 314 seq_putc(m, flags & VM_WRITE ? 'w' : '-'); 315 seq_putc(m, flags & VM_EXEC ? 'x' : '-'); 316 seq_putc(m, flags & VM_MAYSHARE ? 's' : 'p'); 317 seq_put_hex_ll(m, " ", pgoff, 8); 318 seq_put_hex_ll(m, " ", MAJOR(dev), 2); 319 seq_put_hex_ll(m, ":", MINOR(dev), 2); 320 seq_put_decimal_ull(m, " ", ino); 321 seq_putc(m, ' '); 322 } 323 324 static void 325 show_map_vma(struct seq_file *m, struct vm_area_struct *vma) 326 { 327 const struct path *path; 328 const char *name_fmt, *name; 329 vm_flags_t flags = vma->vm_flags; 330 unsigned long ino = 0; 331 unsigned long long pgoff = 0; 332 unsigned long start, end; 333 dev_t dev = 0; 334 335 if (vma->vm_file) { 336 const struct inode *inode = file_user_inode(vma->vm_file); 337 338 dev = inode->i_sb->s_dev; 339 ino = inode->i_ino; 340 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT; 341 } 342 343 start = vma->vm_start; 344 end = vma->vm_end; 345 show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino); 346 347 get_vma_name(vma, &path, &name, &name_fmt); 348 if (path) { 349 seq_pad(m, ' '); 350 seq_path(m, path, "\n"); 351 } else if (name_fmt) { 352 seq_pad(m, ' '); 353 seq_printf(m, name_fmt, name); 354 } else if (name) { 355 seq_pad(m, ' '); 356 seq_puts(m, name); 357 } 358 seq_putc(m, '\n'); 359 } 360 361 static int show_map(struct seq_file *m, void *v) 362 { 363 show_map_vma(m, v); 364 return 0; 365 } 366 367 static const struct seq_operations proc_pid_maps_op = { 368 .start = m_start, 369 .next = m_next, 370 .stop = m_stop, 371 .show = show_map 372 }; 373 374 static int pid_maps_open(struct inode *inode, struct file *file) 375 { 376 return do_maps_open(inode, file, &proc_pid_maps_op); 377 } 378 379 #define PROCMAP_QUERY_VMA_FLAGS ( \ 380 PROCMAP_QUERY_VMA_READABLE | \ 381 PROCMAP_QUERY_VMA_WRITABLE | \ 382 PROCMAP_QUERY_VMA_EXECUTABLE | \ 383 PROCMAP_QUERY_VMA_SHARED \ 384 ) 385 386 #define PROCMAP_QUERY_VALID_FLAGS_MASK ( \ 387 PROCMAP_QUERY_COVERING_OR_NEXT_VMA | \ 388 PROCMAP_QUERY_FILE_BACKED_VMA | \ 389 PROCMAP_QUERY_VMA_FLAGS \ 390 ) 391 392 static int query_vma_setup(struct mm_struct *mm) 393 { 394 return mmap_read_lock_killable(mm); 395 } 396 397 static void query_vma_teardown(struct mm_struct *mm, struct vm_area_struct *vma) 398 { 399 mmap_read_unlock(mm); 400 } 401 402 static struct vm_area_struct *query_vma_find_by_addr(struct mm_struct *mm, unsigned long addr) 403 { 404 return find_vma(mm, addr); 405 } 406 407 static struct vm_area_struct *query_matching_vma(struct mm_struct *mm, 408 unsigned long addr, u32 flags) 409 { 410 struct vm_area_struct *vma; 411 412 next_vma: 413 vma = query_vma_find_by_addr(mm, addr); 414 if (!vma) 415 goto no_vma; 416 417 /* user requested only file-backed VMA, keep iterating */ 418 if ((flags & PROCMAP_QUERY_FILE_BACKED_VMA) && !vma->vm_file) 419 goto skip_vma; 420 421 /* VMA permissions should satisfy query flags */ 422 if (flags & PROCMAP_QUERY_VMA_FLAGS) { 423 u32 perm = 0; 424 425 if (flags & PROCMAP_QUERY_VMA_READABLE) 426 perm |= VM_READ; 427 if (flags & PROCMAP_QUERY_VMA_WRITABLE) 428 perm |= VM_WRITE; 429 if (flags & PROCMAP_QUERY_VMA_EXECUTABLE) 430 perm |= VM_EXEC; 431 if (flags & PROCMAP_QUERY_VMA_SHARED) 432 perm |= VM_MAYSHARE; 433 434 if ((vma->vm_flags & perm) != perm) 435 goto skip_vma; 436 } 437 438 /* found covering VMA or user is OK with the matching next VMA */ 439 if ((flags & PROCMAP_QUERY_COVERING_OR_NEXT_VMA) || vma->vm_start <= addr) 440 return vma; 441 442 skip_vma: 443 /* 444 * If the user needs closest matching VMA, keep iterating. 445 */ 446 addr = vma->vm_end; 447 if (flags & PROCMAP_QUERY_COVERING_OR_NEXT_VMA) 448 goto next_vma; 449 450 no_vma: 451 return ERR_PTR(-ENOENT); 452 } 453 454 static int do_procmap_query(struct proc_maps_private *priv, void __user *uarg) 455 { 456 struct procmap_query karg; 457 struct vm_area_struct *vma; 458 struct mm_struct *mm; 459 const char *name = NULL; 460 char build_id_buf[BUILD_ID_SIZE_MAX], *name_buf = NULL; 461 __u64 usize; 462 int err; 463 464 if (copy_from_user(&usize, (void __user *)uarg, sizeof(usize))) 465 return -EFAULT; 466 /* argument struct can never be that large, reject abuse */ 467 if (usize > PAGE_SIZE) 468 return -E2BIG; 469 /* argument struct should have at least query_flags and query_addr fields */ 470 if (usize < offsetofend(struct procmap_query, query_addr)) 471 return -EINVAL; 472 err = copy_struct_from_user(&karg, sizeof(karg), uarg, usize); 473 if (err) 474 return err; 475 476 /* reject unknown flags */ 477 if (karg.query_flags & ~PROCMAP_QUERY_VALID_FLAGS_MASK) 478 return -EINVAL; 479 /* either both buffer address and size are set, or both should be zero */ 480 if (!!karg.vma_name_size != !!karg.vma_name_addr) 481 return -EINVAL; 482 if (!!karg.build_id_size != !!karg.build_id_addr) 483 return -EINVAL; 484 485 mm = priv->mm; 486 if (!mm || !mmget_not_zero(mm)) 487 return -ESRCH; 488 489 err = query_vma_setup(mm); 490 if (err) { 491 mmput(mm); 492 return err; 493 } 494 495 vma = query_matching_vma(mm, karg.query_addr, karg.query_flags); 496 if (IS_ERR(vma)) { 497 err = PTR_ERR(vma); 498 vma = NULL; 499 goto out; 500 } 501 502 karg.vma_start = vma->vm_start; 503 karg.vma_end = vma->vm_end; 504 505 karg.vma_flags = 0; 506 if (vma->vm_flags & VM_READ) 507 karg.vma_flags |= PROCMAP_QUERY_VMA_READABLE; 508 if (vma->vm_flags & VM_WRITE) 509 karg.vma_flags |= PROCMAP_QUERY_VMA_WRITABLE; 510 if (vma->vm_flags & VM_EXEC) 511 karg.vma_flags |= PROCMAP_QUERY_VMA_EXECUTABLE; 512 if (vma->vm_flags & VM_MAYSHARE) 513 karg.vma_flags |= PROCMAP_QUERY_VMA_SHARED; 514 515 karg.vma_page_size = vma_kernel_pagesize(vma); 516 517 if (vma->vm_file) { 518 const struct inode *inode = file_user_inode(vma->vm_file); 519 520 karg.vma_offset = ((__u64)vma->vm_pgoff) << PAGE_SHIFT; 521 karg.dev_major = MAJOR(inode->i_sb->s_dev); 522 karg.dev_minor = MINOR(inode->i_sb->s_dev); 523 karg.inode = inode->i_ino; 524 } else { 525 karg.vma_offset = 0; 526 karg.dev_major = 0; 527 karg.dev_minor = 0; 528 karg.inode = 0; 529 } 530 531 if (karg.build_id_size) { 532 __u32 build_id_sz; 533 534 err = build_id_parse(vma, build_id_buf, &build_id_sz); 535 if (err) { 536 karg.build_id_size = 0; 537 } else { 538 if (karg.build_id_size < build_id_sz) { 539 err = -ENAMETOOLONG; 540 goto out; 541 } 542 karg.build_id_size = build_id_sz; 543 } 544 } 545 546 if (karg.vma_name_size) { 547 size_t name_buf_sz = min_t(size_t, PATH_MAX, karg.vma_name_size); 548 const struct path *path; 549 const char *name_fmt; 550 size_t name_sz = 0; 551 552 get_vma_name(vma, &path, &name, &name_fmt); 553 554 if (path || name_fmt || name) { 555 name_buf = kmalloc(name_buf_sz, GFP_KERNEL); 556 if (!name_buf) { 557 err = -ENOMEM; 558 goto out; 559 } 560 } 561 if (path) { 562 name = d_path(path, name_buf, name_buf_sz); 563 if (IS_ERR(name)) { 564 err = PTR_ERR(name); 565 goto out; 566 } 567 name_sz = name_buf + name_buf_sz - name; 568 } else if (name || name_fmt) { 569 name_sz = 1 + snprintf(name_buf, name_buf_sz, name_fmt ?: "%s", name); 570 name = name_buf; 571 } 572 if (name_sz > name_buf_sz) { 573 err = -ENAMETOOLONG; 574 goto out; 575 } 576 karg.vma_name_size = name_sz; 577 } 578 579 /* unlock vma or mmap_lock, and put mm_struct before copying data to user */ 580 query_vma_teardown(mm, vma); 581 mmput(mm); 582 583 if (karg.vma_name_size && copy_to_user(u64_to_user_ptr(karg.vma_name_addr), 584 name, karg.vma_name_size)) { 585 kfree(name_buf); 586 return -EFAULT; 587 } 588 kfree(name_buf); 589 590 if (karg.build_id_size && copy_to_user(u64_to_user_ptr(karg.build_id_addr), 591 build_id_buf, karg.build_id_size)) 592 return -EFAULT; 593 594 if (copy_to_user(uarg, &karg, min_t(size_t, sizeof(karg), usize))) 595 return -EFAULT; 596 597 return 0; 598 599 out: 600 query_vma_teardown(mm, vma); 601 mmput(mm); 602 kfree(name_buf); 603 return err; 604 } 605 606 static long procfs_procmap_ioctl(struct file *file, unsigned int cmd, unsigned long arg) 607 { 608 struct seq_file *seq = file->private_data; 609 struct proc_maps_private *priv = seq->private; 610 611 switch (cmd) { 612 case PROCMAP_QUERY: 613 return do_procmap_query(priv, (void __user *)arg); 614 default: 615 return -ENOIOCTLCMD; 616 } 617 } 618 619 const struct file_operations proc_pid_maps_operations = { 620 .open = pid_maps_open, 621 .read = seq_read, 622 .llseek = seq_lseek, 623 .release = proc_map_release, 624 .unlocked_ioctl = procfs_procmap_ioctl, 625 .compat_ioctl = compat_ptr_ioctl, 626 }; 627 628 /* 629 * Proportional Set Size(PSS): my share of RSS. 630 * 631 * PSS of a process is the count of pages it has in memory, where each 632 * page is divided by the number of processes sharing it. So if a 633 * process has 1000 pages all to itself, and 1000 shared with one other 634 * process, its PSS will be 1500. 635 * 636 * To keep (accumulated) division errors low, we adopt a 64bit 637 * fixed-point pss counter to minimize division errors. So (pss >> 638 * PSS_SHIFT) would be the real byte count. 639 * 640 * A shift of 12 before division means (assuming 4K page size): 641 * - 1M 3-user-pages add up to 8KB errors; 642 * - supports mapcount up to 2^24, or 16M; 643 * - supports PSS up to 2^52 bytes, or 4PB. 644 */ 645 #define PSS_SHIFT 12 646 647 #ifdef CONFIG_PROC_PAGE_MONITOR 648 struct mem_size_stats { 649 unsigned long resident; 650 unsigned long shared_clean; 651 unsigned long shared_dirty; 652 unsigned long private_clean; 653 unsigned long private_dirty; 654 unsigned long referenced; 655 unsigned long anonymous; 656 unsigned long lazyfree; 657 unsigned long anonymous_thp; 658 unsigned long shmem_thp; 659 unsigned long file_thp; 660 unsigned long swap; 661 unsigned long shared_hugetlb; 662 unsigned long private_hugetlb; 663 unsigned long ksm; 664 u64 pss; 665 u64 pss_anon; 666 u64 pss_file; 667 u64 pss_shmem; 668 u64 pss_dirty; 669 u64 pss_locked; 670 u64 swap_pss; 671 }; 672 673 static void smaps_page_accumulate(struct mem_size_stats *mss, 674 struct folio *folio, unsigned long size, unsigned long pss, 675 bool dirty, bool locked, bool private) 676 { 677 mss->pss += pss; 678 679 if (folio_test_anon(folio)) 680 mss->pss_anon += pss; 681 else if (folio_test_swapbacked(folio)) 682 mss->pss_shmem += pss; 683 else 684 mss->pss_file += pss; 685 686 if (locked) 687 mss->pss_locked += pss; 688 689 if (dirty || folio_test_dirty(folio)) { 690 mss->pss_dirty += pss; 691 if (private) 692 mss->private_dirty += size; 693 else 694 mss->shared_dirty += size; 695 } else { 696 if (private) 697 mss->private_clean += size; 698 else 699 mss->shared_clean += size; 700 } 701 } 702 703 static void smaps_account(struct mem_size_stats *mss, struct page *page, 704 bool compound, bool young, bool dirty, bool locked, 705 bool present) 706 { 707 struct folio *folio = page_folio(page); 708 int i, nr = compound ? compound_nr(page) : 1; 709 unsigned long size = nr * PAGE_SIZE; 710 bool exclusive; 711 int mapcount; 712 713 /* 714 * First accumulate quantities that depend only on |size| and the type 715 * of the compound page. 716 */ 717 if (folio_test_anon(folio)) { 718 mss->anonymous += size; 719 if (!folio_test_swapbacked(folio) && !dirty && 720 !folio_test_dirty(folio)) 721 mss->lazyfree += size; 722 } 723 724 if (folio_test_ksm(folio)) 725 mss->ksm += size; 726 727 mss->resident += size; 728 /* Accumulate the size in pages that have been accessed. */ 729 if (young || folio_test_young(folio) || folio_test_referenced(folio)) 730 mss->referenced += size; 731 732 /* 733 * Then accumulate quantities that may depend on sharing, or that may 734 * differ page-by-page. 735 * 736 * refcount == 1 for present entries guarantees that the folio is mapped 737 * exactly once. For large folios this implies that exactly one 738 * PTE/PMD/... maps (a part of) this folio. 739 * 740 * Treat all non-present entries (where relying on the mapcount and 741 * refcount doesn't make sense) as "maybe shared, but not sure how 742 * often". We treat device private entries as being fake-present. 743 * 744 * Note that it would not be safe to read the mapcount especially for 745 * pages referenced by migration entries, even with the PTL held. 746 */ 747 if (folio_ref_count(folio) == 1 || !present) { 748 smaps_page_accumulate(mss, folio, size, size << PSS_SHIFT, 749 dirty, locked, present); 750 return; 751 } 752 753 if (IS_ENABLED(CONFIG_NO_PAGE_MAPCOUNT)) { 754 mapcount = folio_average_page_mapcount(folio); 755 exclusive = !folio_maybe_mapped_shared(folio); 756 } 757 758 /* 759 * We obtain a snapshot of the mapcount. Without holding the folio lock 760 * this snapshot can be slightly wrong as we cannot always read the 761 * mapcount atomically. 762 */ 763 for (i = 0; i < nr; i++, page++) { 764 unsigned long pss = PAGE_SIZE << PSS_SHIFT; 765 766 if (IS_ENABLED(CONFIG_PAGE_MAPCOUNT)) { 767 mapcount = folio_precise_page_mapcount(folio, page); 768 exclusive = mapcount < 2; 769 } 770 771 if (mapcount >= 2) 772 pss /= mapcount; 773 smaps_page_accumulate(mss, folio, PAGE_SIZE, pss, 774 dirty, locked, exclusive); 775 } 776 } 777 778 #ifdef CONFIG_SHMEM 779 static int smaps_pte_hole(unsigned long addr, unsigned long end, 780 __always_unused int depth, struct mm_walk *walk) 781 { 782 struct mem_size_stats *mss = walk->private; 783 struct vm_area_struct *vma = walk->vma; 784 785 mss->swap += shmem_partial_swap_usage(walk->vma->vm_file->f_mapping, 786 linear_page_index(vma, addr), 787 linear_page_index(vma, end)); 788 789 return 0; 790 } 791 #else 792 #define smaps_pte_hole NULL 793 #endif /* CONFIG_SHMEM */ 794 795 static void smaps_pte_hole_lookup(unsigned long addr, struct mm_walk *walk) 796 { 797 #ifdef CONFIG_SHMEM 798 if (walk->ops->pte_hole) { 799 /* depth is not used */ 800 smaps_pte_hole(addr, addr + PAGE_SIZE, 0, walk); 801 } 802 #endif 803 } 804 805 static void smaps_pte_entry(pte_t *pte, unsigned long addr, 806 struct mm_walk *walk) 807 { 808 struct mem_size_stats *mss = walk->private; 809 struct vm_area_struct *vma = walk->vma; 810 bool locked = !!(vma->vm_flags & VM_LOCKED); 811 struct page *page = NULL; 812 bool present = false, young = false, dirty = false; 813 pte_t ptent = ptep_get(pte); 814 815 if (pte_present(ptent)) { 816 page = vm_normal_page(vma, addr, ptent); 817 young = pte_young(ptent); 818 dirty = pte_dirty(ptent); 819 present = true; 820 } else if (is_swap_pte(ptent)) { 821 swp_entry_t swpent = pte_to_swp_entry(ptent); 822 823 if (!non_swap_entry(swpent)) { 824 int mapcount; 825 826 mss->swap += PAGE_SIZE; 827 mapcount = swp_swapcount(swpent); 828 if (mapcount >= 2) { 829 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT; 830 831 do_div(pss_delta, mapcount); 832 mss->swap_pss += pss_delta; 833 } else { 834 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT; 835 } 836 } else if (is_pfn_swap_entry(swpent)) { 837 if (is_device_private_entry(swpent)) 838 present = true; 839 page = pfn_swap_entry_to_page(swpent); 840 } 841 } else { 842 smaps_pte_hole_lookup(addr, walk); 843 return; 844 } 845 846 if (!page) 847 return; 848 849 smaps_account(mss, page, false, young, dirty, locked, present); 850 } 851 852 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 853 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr, 854 struct mm_walk *walk) 855 { 856 struct mem_size_stats *mss = walk->private; 857 struct vm_area_struct *vma = walk->vma; 858 bool locked = !!(vma->vm_flags & VM_LOCKED); 859 struct page *page = NULL; 860 bool present = false; 861 struct folio *folio; 862 863 if (pmd_present(*pmd)) { 864 page = vm_normal_page_pmd(vma, addr, *pmd); 865 present = true; 866 } else if (unlikely(thp_migration_supported() && is_swap_pmd(*pmd))) { 867 swp_entry_t entry = pmd_to_swp_entry(*pmd); 868 869 if (is_pfn_swap_entry(entry)) 870 page = pfn_swap_entry_to_page(entry); 871 } 872 if (IS_ERR_OR_NULL(page)) 873 return; 874 folio = page_folio(page); 875 if (folio_test_anon(folio)) 876 mss->anonymous_thp += HPAGE_PMD_SIZE; 877 else if (folio_test_swapbacked(folio)) 878 mss->shmem_thp += HPAGE_PMD_SIZE; 879 else if (folio_is_zone_device(folio)) 880 /* pass */; 881 else 882 mss->file_thp += HPAGE_PMD_SIZE; 883 884 smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd), 885 locked, present); 886 } 887 #else 888 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr, 889 struct mm_walk *walk) 890 { 891 } 892 #endif 893 894 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, 895 struct mm_walk *walk) 896 { 897 struct vm_area_struct *vma = walk->vma; 898 pte_t *pte; 899 spinlock_t *ptl; 900 901 ptl = pmd_trans_huge_lock(pmd, vma); 902 if (ptl) { 903 smaps_pmd_entry(pmd, addr, walk); 904 spin_unlock(ptl); 905 goto out; 906 } 907 908 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); 909 if (!pte) { 910 walk->action = ACTION_AGAIN; 911 return 0; 912 } 913 for (; addr != end; pte++, addr += PAGE_SIZE) 914 smaps_pte_entry(pte, addr, walk); 915 pte_unmap_unlock(pte - 1, ptl); 916 out: 917 cond_resched(); 918 return 0; 919 } 920 921 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma) 922 { 923 /* 924 * Don't forget to update Documentation/ on changes. 925 * 926 * The length of the second argument of mnemonics[] 927 * needs to be 3 instead of previously set 2 928 * (i.e. from [BITS_PER_LONG][2] to [BITS_PER_LONG][3]) 929 * to avoid spurious 930 * -Werror=unterminated-string-initialization warning 931 * with GCC 15 932 */ 933 static const char mnemonics[BITS_PER_LONG][3] = { 934 /* 935 * In case if we meet a flag we don't know about. 936 */ 937 [0 ... (BITS_PER_LONG-1)] = "??", 938 939 [ilog2(VM_READ)] = "rd", 940 [ilog2(VM_WRITE)] = "wr", 941 [ilog2(VM_EXEC)] = "ex", 942 [ilog2(VM_SHARED)] = "sh", 943 [ilog2(VM_MAYREAD)] = "mr", 944 [ilog2(VM_MAYWRITE)] = "mw", 945 [ilog2(VM_MAYEXEC)] = "me", 946 [ilog2(VM_MAYSHARE)] = "ms", 947 [ilog2(VM_GROWSDOWN)] = "gd", 948 [ilog2(VM_PFNMAP)] = "pf", 949 [ilog2(VM_LOCKED)] = "lo", 950 [ilog2(VM_IO)] = "io", 951 [ilog2(VM_SEQ_READ)] = "sr", 952 [ilog2(VM_RAND_READ)] = "rr", 953 [ilog2(VM_DONTCOPY)] = "dc", 954 [ilog2(VM_DONTEXPAND)] = "de", 955 [ilog2(VM_LOCKONFAULT)] = "lf", 956 [ilog2(VM_ACCOUNT)] = "ac", 957 [ilog2(VM_NORESERVE)] = "nr", 958 [ilog2(VM_HUGETLB)] = "ht", 959 [ilog2(VM_SYNC)] = "sf", 960 [ilog2(VM_ARCH_1)] = "ar", 961 [ilog2(VM_WIPEONFORK)] = "wf", 962 [ilog2(VM_DONTDUMP)] = "dd", 963 #ifdef CONFIG_ARM64_BTI 964 [ilog2(VM_ARM64_BTI)] = "bt", 965 #endif 966 #ifdef CONFIG_MEM_SOFT_DIRTY 967 [ilog2(VM_SOFTDIRTY)] = "sd", 968 #endif 969 [ilog2(VM_MIXEDMAP)] = "mm", 970 [ilog2(VM_HUGEPAGE)] = "hg", 971 [ilog2(VM_NOHUGEPAGE)] = "nh", 972 [ilog2(VM_MERGEABLE)] = "mg", 973 [ilog2(VM_UFFD_MISSING)]= "um", 974 [ilog2(VM_UFFD_WP)] = "uw", 975 #ifdef CONFIG_ARM64_MTE 976 [ilog2(VM_MTE)] = "mt", 977 [ilog2(VM_MTE_ALLOWED)] = "", 978 #endif 979 #ifdef CONFIG_ARCH_HAS_PKEYS 980 /* These come out via ProtectionKey: */ 981 [ilog2(VM_PKEY_BIT0)] = "", 982 [ilog2(VM_PKEY_BIT1)] = "", 983 [ilog2(VM_PKEY_BIT2)] = "", 984 #if VM_PKEY_BIT3 985 [ilog2(VM_PKEY_BIT3)] = "", 986 #endif 987 #if VM_PKEY_BIT4 988 [ilog2(VM_PKEY_BIT4)] = "", 989 #endif 990 #endif /* CONFIG_ARCH_HAS_PKEYS */ 991 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_MINOR 992 [ilog2(VM_UFFD_MINOR)] = "ui", 993 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_MINOR */ 994 #ifdef CONFIG_ARCH_HAS_USER_SHADOW_STACK 995 [ilog2(VM_SHADOW_STACK)] = "ss", 996 #endif 997 #if defined(CONFIG_64BIT) || defined(CONFIG_PPC32) 998 [ilog2(VM_DROPPABLE)] = "dp", 999 #endif 1000 #ifdef CONFIG_64BIT 1001 [ilog2(VM_SEALED)] = "sl", 1002 #endif 1003 }; 1004 size_t i; 1005 1006 seq_puts(m, "VmFlags: "); 1007 for (i = 0; i < BITS_PER_LONG; i++) { 1008 if (!mnemonics[i][0]) 1009 continue; 1010 if (vma->vm_flags & (1UL << i)) 1011 seq_printf(m, "%s ", mnemonics[i]); 1012 } 1013 seq_putc(m, '\n'); 1014 } 1015 1016 #ifdef CONFIG_HUGETLB_PAGE 1017 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask, 1018 unsigned long addr, unsigned long end, 1019 struct mm_walk *walk) 1020 { 1021 struct mem_size_stats *mss = walk->private; 1022 struct vm_area_struct *vma = walk->vma; 1023 pte_t ptent = huge_ptep_get(walk->mm, addr, pte); 1024 struct folio *folio = NULL; 1025 bool present = false; 1026 1027 if (pte_present(ptent)) { 1028 folio = page_folio(pte_page(ptent)); 1029 present = true; 1030 } else if (is_swap_pte(ptent)) { 1031 swp_entry_t swpent = pte_to_swp_entry(ptent); 1032 1033 if (is_pfn_swap_entry(swpent)) 1034 folio = pfn_swap_entry_folio(swpent); 1035 } 1036 1037 if (folio) { 1038 /* We treat non-present entries as "maybe shared". */ 1039 if (!present || folio_maybe_mapped_shared(folio) || 1040 hugetlb_pmd_shared(pte)) 1041 mss->shared_hugetlb += huge_page_size(hstate_vma(vma)); 1042 else 1043 mss->private_hugetlb += huge_page_size(hstate_vma(vma)); 1044 } 1045 return 0; 1046 } 1047 #else 1048 #define smaps_hugetlb_range NULL 1049 #endif /* HUGETLB_PAGE */ 1050 1051 static const struct mm_walk_ops smaps_walk_ops = { 1052 .pmd_entry = smaps_pte_range, 1053 .hugetlb_entry = smaps_hugetlb_range, 1054 .walk_lock = PGWALK_RDLOCK, 1055 }; 1056 1057 static const struct mm_walk_ops smaps_shmem_walk_ops = { 1058 .pmd_entry = smaps_pte_range, 1059 .hugetlb_entry = smaps_hugetlb_range, 1060 .pte_hole = smaps_pte_hole, 1061 .walk_lock = PGWALK_RDLOCK, 1062 }; 1063 1064 /* 1065 * Gather mem stats from @vma with the indicated beginning 1066 * address @start, and keep them in @mss. 1067 * 1068 * Use vm_start of @vma as the beginning address if @start is 0. 1069 */ 1070 static void smap_gather_stats(struct vm_area_struct *vma, 1071 struct mem_size_stats *mss, unsigned long start) 1072 { 1073 const struct mm_walk_ops *ops = &smaps_walk_ops; 1074 1075 /* Invalid start */ 1076 if (start >= vma->vm_end) 1077 return; 1078 1079 if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) { 1080 /* 1081 * For shared or readonly shmem mappings we know that all 1082 * swapped out pages belong to the shmem object, and we can 1083 * obtain the swap value much more efficiently. For private 1084 * writable mappings, we might have COW pages that are 1085 * not affected by the parent swapped out pages of the shmem 1086 * object, so we have to distinguish them during the page walk. 1087 * Unless we know that the shmem object (or the part mapped by 1088 * our VMA) has no swapped out pages at all. 1089 */ 1090 unsigned long shmem_swapped = shmem_swap_usage(vma); 1091 1092 if (!start && (!shmem_swapped || (vma->vm_flags & VM_SHARED) || 1093 !(vma->vm_flags & VM_WRITE))) { 1094 mss->swap += shmem_swapped; 1095 } else { 1096 ops = &smaps_shmem_walk_ops; 1097 } 1098 } 1099 1100 /* mmap_lock is held in m_start */ 1101 if (!start) 1102 walk_page_vma(vma, ops, mss); 1103 else 1104 walk_page_range(vma->vm_mm, start, vma->vm_end, ops, mss); 1105 } 1106 1107 #define SEQ_PUT_DEC(str, val) \ 1108 seq_put_decimal_ull_width(m, str, (val) >> 10, 8) 1109 1110 /* Show the contents common for smaps and smaps_rollup */ 1111 static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss, 1112 bool rollup_mode) 1113 { 1114 SEQ_PUT_DEC("Rss: ", mss->resident); 1115 SEQ_PUT_DEC(" kB\nPss: ", mss->pss >> PSS_SHIFT); 1116 SEQ_PUT_DEC(" kB\nPss_Dirty: ", mss->pss_dirty >> PSS_SHIFT); 1117 if (rollup_mode) { 1118 /* 1119 * These are meaningful only for smaps_rollup, otherwise two of 1120 * them are zero, and the other one is the same as Pss. 1121 */ 1122 SEQ_PUT_DEC(" kB\nPss_Anon: ", 1123 mss->pss_anon >> PSS_SHIFT); 1124 SEQ_PUT_DEC(" kB\nPss_File: ", 1125 mss->pss_file >> PSS_SHIFT); 1126 SEQ_PUT_DEC(" kB\nPss_Shmem: ", 1127 mss->pss_shmem >> PSS_SHIFT); 1128 } 1129 SEQ_PUT_DEC(" kB\nShared_Clean: ", mss->shared_clean); 1130 SEQ_PUT_DEC(" kB\nShared_Dirty: ", mss->shared_dirty); 1131 SEQ_PUT_DEC(" kB\nPrivate_Clean: ", mss->private_clean); 1132 SEQ_PUT_DEC(" kB\nPrivate_Dirty: ", mss->private_dirty); 1133 SEQ_PUT_DEC(" kB\nReferenced: ", mss->referenced); 1134 SEQ_PUT_DEC(" kB\nAnonymous: ", mss->anonymous); 1135 SEQ_PUT_DEC(" kB\nKSM: ", mss->ksm); 1136 SEQ_PUT_DEC(" kB\nLazyFree: ", mss->lazyfree); 1137 SEQ_PUT_DEC(" kB\nAnonHugePages: ", mss->anonymous_thp); 1138 SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp); 1139 SEQ_PUT_DEC(" kB\nFilePmdMapped: ", mss->file_thp); 1140 SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb); 1141 seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ", 1142 mss->private_hugetlb >> 10, 7); 1143 SEQ_PUT_DEC(" kB\nSwap: ", mss->swap); 1144 SEQ_PUT_DEC(" kB\nSwapPss: ", 1145 mss->swap_pss >> PSS_SHIFT); 1146 SEQ_PUT_DEC(" kB\nLocked: ", 1147 mss->pss_locked >> PSS_SHIFT); 1148 seq_puts(m, " kB\n"); 1149 } 1150 1151 static int show_smap(struct seq_file *m, void *v) 1152 { 1153 struct vm_area_struct *vma = v; 1154 struct mem_size_stats mss = {}; 1155 1156 smap_gather_stats(vma, &mss, 0); 1157 1158 show_map_vma(m, vma); 1159 1160 SEQ_PUT_DEC("Size: ", vma->vm_end - vma->vm_start); 1161 SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma)); 1162 SEQ_PUT_DEC(" kB\nMMUPageSize: ", vma_mmu_pagesize(vma)); 1163 seq_puts(m, " kB\n"); 1164 1165 __show_smap(m, &mss, false); 1166 1167 seq_printf(m, "THPeligible: %8u\n", 1168 !!thp_vma_allowable_orders(vma, vma->vm_flags, 1169 TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL)); 1170 1171 if (arch_pkeys_enabled()) 1172 seq_printf(m, "ProtectionKey: %8u\n", vma_pkey(vma)); 1173 show_smap_vma_flags(m, vma); 1174 1175 return 0; 1176 } 1177 1178 static int show_smaps_rollup(struct seq_file *m, void *v) 1179 { 1180 struct proc_maps_private *priv = m->private; 1181 struct mem_size_stats mss = {}; 1182 struct mm_struct *mm = priv->mm; 1183 struct vm_area_struct *vma; 1184 unsigned long vma_start = 0, last_vma_end = 0; 1185 int ret = 0; 1186 VMA_ITERATOR(vmi, mm, 0); 1187 1188 priv->task = get_proc_task(priv->inode); 1189 if (!priv->task) 1190 return -ESRCH; 1191 1192 if (!mm || !mmget_not_zero(mm)) { 1193 ret = -ESRCH; 1194 goto out_put_task; 1195 } 1196 1197 ret = mmap_read_lock_killable(mm); 1198 if (ret) 1199 goto out_put_mm; 1200 1201 hold_task_mempolicy(priv); 1202 vma = vma_next(&vmi); 1203 1204 if (unlikely(!vma)) 1205 goto empty_set; 1206 1207 vma_start = vma->vm_start; 1208 do { 1209 smap_gather_stats(vma, &mss, 0); 1210 last_vma_end = vma->vm_end; 1211 1212 /* 1213 * Release mmap_lock temporarily if someone wants to 1214 * access it for write request. 1215 */ 1216 if (mmap_lock_is_contended(mm)) { 1217 vma_iter_invalidate(&vmi); 1218 mmap_read_unlock(mm); 1219 ret = mmap_read_lock_killable(mm); 1220 if (ret) { 1221 release_task_mempolicy(priv); 1222 goto out_put_mm; 1223 } 1224 1225 /* 1226 * After dropping the lock, there are four cases to 1227 * consider. See the following example for explanation. 1228 * 1229 * +------+------+-----------+ 1230 * | VMA1 | VMA2 | VMA3 | 1231 * +------+------+-----------+ 1232 * | | | | 1233 * 4k 8k 16k 400k 1234 * 1235 * Suppose we drop the lock after reading VMA2 due to 1236 * contention, then we get: 1237 * 1238 * last_vma_end = 16k 1239 * 1240 * 1) VMA2 is freed, but VMA3 exists: 1241 * 1242 * vma_next(vmi) will return VMA3. 1243 * In this case, just continue from VMA3. 1244 * 1245 * 2) VMA2 still exists: 1246 * 1247 * vma_next(vmi) will return VMA3. 1248 * In this case, just continue from VMA3. 1249 * 1250 * 3) No more VMAs can be found: 1251 * 1252 * vma_next(vmi) will return NULL. 1253 * No more things to do, just break. 1254 * 1255 * 4) (last_vma_end - 1) is the middle of a vma (VMA'): 1256 * 1257 * vma_next(vmi) will return VMA' whose range 1258 * contains last_vma_end. 1259 * Iterate VMA' from last_vma_end. 1260 */ 1261 vma = vma_next(&vmi); 1262 /* Case 3 above */ 1263 if (!vma) 1264 break; 1265 1266 /* Case 1 and 2 above */ 1267 if (vma->vm_start >= last_vma_end) { 1268 smap_gather_stats(vma, &mss, 0); 1269 last_vma_end = vma->vm_end; 1270 continue; 1271 } 1272 1273 /* Case 4 above */ 1274 if (vma->vm_end > last_vma_end) { 1275 smap_gather_stats(vma, &mss, last_vma_end); 1276 last_vma_end = vma->vm_end; 1277 } 1278 } 1279 } for_each_vma(vmi, vma); 1280 1281 empty_set: 1282 show_vma_header_prefix(m, vma_start, last_vma_end, 0, 0, 0, 0); 1283 seq_pad(m, ' '); 1284 seq_puts(m, "[rollup]\n"); 1285 1286 __show_smap(m, &mss, true); 1287 1288 release_task_mempolicy(priv); 1289 mmap_read_unlock(mm); 1290 1291 out_put_mm: 1292 mmput(mm); 1293 out_put_task: 1294 put_task_struct(priv->task); 1295 priv->task = NULL; 1296 1297 return ret; 1298 } 1299 #undef SEQ_PUT_DEC 1300 1301 static const struct seq_operations proc_pid_smaps_op = { 1302 .start = m_start, 1303 .next = m_next, 1304 .stop = m_stop, 1305 .show = show_smap 1306 }; 1307 1308 static int pid_smaps_open(struct inode *inode, struct file *file) 1309 { 1310 return do_maps_open(inode, file, &proc_pid_smaps_op); 1311 } 1312 1313 static int smaps_rollup_open(struct inode *inode, struct file *file) 1314 { 1315 int ret; 1316 struct proc_maps_private *priv; 1317 1318 priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT); 1319 if (!priv) 1320 return -ENOMEM; 1321 1322 ret = single_open(file, show_smaps_rollup, priv); 1323 if (ret) 1324 goto out_free; 1325 1326 priv->inode = inode; 1327 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ); 1328 if (IS_ERR_OR_NULL(priv->mm)) { 1329 ret = priv->mm ? PTR_ERR(priv->mm) : -ESRCH; 1330 1331 single_release(inode, file); 1332 goto out_free; 1333 } 1334 1335 return 0; 1336 1337 out_free: 1338 kfree(priv); 1339 return ret; 1340 } 1341 1342 static int smaps_rollup_release(struct inode *inode, struct file *file) 1343 { 1344 struct seq_file *seq = file->private_data; 1345 struct proc_maps_private *priv = seq->private; 1346 1347 if (priv->mm) 1348 mmdrop(priv->mm); 1349 1350 kfree(priv); 1351 return single_release(inode, file); 1352 } 1353 1354 const struct file_operations proc_pid_smaps_operations = { 1355 .open = pid_smaps_open, 1356 .read = seq_read, 1357 .llseek = seq_lseek, 1358 .release = proc_map_release, 1359 }; 1360 1361 const struct file_operations proc_pid_smaps_rollup_operations = { 1362 .open = smaps_rollup_open, 1363 .read = seq_read, 1364 .llseek = seq_lseek, 1365 .release = smaps_rollup_release, 1366 }; 1367 1368 enum clear_refs_types { 1369 CLEAR_REFS_ALL = 1, 1370 CLEAR_REFS_ANON, 1371 CLEAR_REFS_MAPPED, 1372 CLEAR_REFS_SOFT_DIRTY, 1373 CLEAR_REFS_MM_HIWATER_RSS, 1374 CLEAR_REFS_LAST, 1375 }; 1376 1377 struct clear_refs_private { 1378 enum clear_refs_types type; 1379 }; 1380 1381 #ifdef CONFIG_MEM_SOFT_DIRTY 1382 1383 static inline bool pte_is_pinned(struct vm_area_struct *vma, unsigned long addr, pte_t pte) 1384 { 1385 struct folio *folio; 1386 1387 if (!pte_write(pte)) 1388 return false; 1389 if (!is_cow_mapping(vma->vm_flags)) 1390 return false; 1391 if (likely(!test_bit(MMF_HAS_PINNED, &vma->vm_mm->flags))) 1392 return false; 1393 folio = vm_normal_folio(vma, addr, pte); 1394 if (!folio) 1395 return false; 1396 return folio_maybe_dma_pinned(folio); 1397 } 1398 1399 static inline void clear_soft_dirty(struct vm_area_struct *vma, 1400 unsigned long addr, pte_t *pte) 1401 { 1402 /* 1403 * The soft-dirty tracker uses #PF-s to catch writes 1404 * to pages, so write-protect the pte as well. See the 1405 * Documentation/admin-guide/mm/soft-dirty.rst for full description 1406 * of how soft-dirty works. 1407 */ 1408 pte_t ptent = ptep_get(pte); 1409 1410 if (pte_present(ptent)) { 1411 pte_t old_pte; 1412 1413 if (pte_is_pinned(vma, addr, ptent)) 1414 return; 1415 old_pte = ptep_modify_prot_start(vma, addr, pte); 1416 ptent = pte_wrprotect(old_pte); 1417 ptent = pte_clear_soft_dirty(ptent); 1418 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent); 1419 } else if (is_swap_pte(ptent)) { 1420 ptent = pte_swp_clear_soft_dirty(ptent); 1421 set_pte_at(vma->vm_mm, addr, pte, ptent); 1422 } 1423 } 1424 #else 1425 static inline void clear_soft_dirty(struct vm_area_struct *vma, 1426 unsigned long addr, pte_t *pte) 1427 { 1428 } 1429 #endif 1430 1431 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE) 1432 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma, 1433 unsigned long addr, pmd_t *pmdp) 1434 { 1435 pmd_t old, pmd = *pmdp; 1436 1437 if (pmd_present(pmd)) { 1438 /* See comment in change_huge_pmd() */ 1439 old = pmdp_invalidate(vma, addr, pmdp); 1440 if (pmd_dirty(old)) 1441 pmd = pmd_mkdirty(pmd); 1442 if (pmd_young(old)) 1443 pmd = pmd_mkyoung(pmd); 1444 1445 pmd = pmd_wrprotect(pmd); 1446 pmd = pmd_clear_soft_dirty(pmd); 1447 1448 set_pmd_at(vma->vm_mm, addr, pmdp, pmd); 1449 } else if (is_migration_entry(pmd_to_swp_entry(pmd))) { 1450 pmd = pmd_swp_clear_soft_dirty(pmd); 1451 set_pmd_at(vma->vm_mm, addr, pmdp, pmd); 1452 } 1453 } 1454 #else 1455 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma, 1456 unsigned long addr, pmd_t *pmdp) 1457 { 1458 } 1459 #endif 1460 1461 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr, 1462 unsigned long end, struct mm_walk *walk) 1463 { 1464 struct clear_refs_private *cp = walk->private; 1465 struct vm_area_struct *vma = walk->vma; 1466 pte_t *pte, ptent; 1467 spinlock_t *ptl; 1468 struct folio *folio; 1469 1470 ptl = pmd_trans_huge_lock(pmd, vma); 1471 if (ptl) { 1472 if (cp->type == CLEAR_REFS_SOFT_DIRTY) { 1473 clear_soft_dirty_pmd(vma, addr, pmd); 1474 goto out; 1475 } 1476 1477 if (!pmd_present(*pmd)) 1478 goto out; 1479 1480 folio = pmd_folio(*pmd); 1481 1482 /* Clear accessed and referenced bits. */ 1483 pmdp_test_and_clear_young(vma, addr, pmd); 1484 folio_test_clear_young(folio); 1485 folio_clear_referenced(folio); 1486 out: 1487 spin_unlock(ptl); 1488 return 0; 1489 } 1490 1491 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); 1492 if (!pte) { 1493 walk->action = ACTION_AGAIN; 1494 return 0; 1495 } 1496 for (; addr != end; pte++, addr += PAGE_SIZE) { 1497 ptent = ptep_get(pte); 1498 1499 if (cp->type == CLEAR_REFS_SOFT_DIRTY) { 1500 clear_soft_dirty(vma, addr, pte); 1501 continue; 1502 } 1503 1504 if (!pte_present(ptent)) 1505 continue; 1506 1507 folio = vm_normal_folio(vma, addr, ptent); 1508 if (!folio) 1509 continue; 1510 1511 /* Clear accessed and referenced bits. */ 1512 ptep_test_and_clear_young(vma, addr, pte); 1513 folio_test_clear_young(folio); 1514 folio_clear_referenced(folio); 1515 } 1516 pte_unmap_unlock(pte - 1, ptl); 1517 cond_resched(); 1518 return 0; 1519 } 1520 1521 static int clear_refs_test_walk(unsigned long start, unsigned long end, 1522 struct mm_walk *walk) 1523 { 1524 struct clear_refs_private *cp = walk->private; 1525 struct vm_area_struct *vma = walk->vma; 1526 1527 if (vma->vm_flags & VM_PFNMAP) 1528 return 1; 1529 1530 /* 1531 * Writing 1 to /proc/pid/clear_refs affects all pages. 1532 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages. 1533 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages. 1534 * Writing 4 to /proc/pid/clear_refs affects all pages. 1535 */ 1536 if (cp->type == CLEAR_REFS_ANON && vma->vm_file) 1537 return 1; 1538 if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file) 1539 return 1; 1540 return 0; 1541 } 1542 1543 static const struct mm_walk_ops clear_refs_walk_ops = { 1544 .pmd_entry = clear_refs_pte_range, 1545 .test_walk = clear_refs_test_walk, 1546 .walk_lock = PGWALK_WRLOCK, 1547 }; 1548 1549 static ssize_t clear_refs_write(struct file *file, const char __user *buf, 1550 size_t count, loff_t *ppos) 1551 { 1552 struct task_struct *task; 1553 char buffer[PROC_NUMBUF] = {}; 1554 struct mm_struct *mm; 1555 struct vm_area_struct *vma; 1556 enum clear_refs_types type; 1557 int itype; 1558 int rv; 1559 1560 if (count > sizeof(buffer) - 1) 1561 count = sizeof(buffer) - 1; 1562 if (copy_from_user(buffer, buf, count)) 1563 return -EFAULT; 1564 rv = kstrtoint(strstrip(buffer), 10, &itype); 1565 if (rv < 0) 1566 return rv; 1567 type = (enum clear_refs_types)itype; 1568 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST) 1569 return -EINVAL; 1570 1571 task = get_proc_task(file_inode(file)); 1572 if (!task) 1573 return -ESRCH; 1574 mm = get_task_mm(task); 1575 if (mm) { 1576 VMA_ITERATOR(vmi, mm, 0); 1577 struct mmu_notifier_range range; 1578 struct clear_refs_private cp = { 1579 .type = type, 1580 }; 1581 1582 if (mmap_write_lock_killable(mm)) { 1583 count = -EINTR; 1584 goto out_mm; 1585 } 1586 if (type == CLEAR_REFS_MM_HIWATER_RSS) { 1587 /* 1588 * Writing 5 to /proc/pid/clear_refs resets the peak 1589 * resident set size to this mm's current rss value. 1590 */ 1591 reset_mm_hiwater_rss(mm); 1592 goto out_unlock; 1593 } 1594 1595 if (type == CLEAR_REFS_SOFT_DIRTY) { 1596 for_each_vma(vmi, vma) { 1597 if (!(vma->vm_flags & VM_SOFTDIRTY)) 1598 continue; 1599 vm_flags_clear(vma, VM_SOFTDIRTY); 1600 vma_set_page_prot(vma); 1601 } 1602 1603 inc_tlb_flush_pending(mm); 1604 mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY, 1605 0, mm, 0, -1UL); 1606 mmu_notifier_invalidate_range_start(&range); 1607 } 1608 walk_page_range(mm, 0, -1, &clear_refs_walk_ops, &cp); 1609 if (type == CLEAR_REFS_SOFT_DIRTY) { 1610 mmu_notifier_invalidate_range_end(&range); 1611 flush_tlb_mm(mm); 1612 dec_tlb_flush_pending(mm); 1613 } 1614 out_unlock: 1615 mmap_write_unlock(mm); 1616 out_mm: 1617 mmput(mm); 1618 } 1619 put_task_struct(task); 1620 1621 return count; 1622 } 1623 1624 const struct file_operations proc_clear_refs_operations = { 1625 .write = clear_refs_write, 1626 .llseek = noop_llseek, 1627 }; 1628 1629 typedef struct { 1630 u64 pme; 1631 } pagemap_entry_t; 1632 1633 struct pagemapread { 1634 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */ 1635 pagemap_entry_t *buffer; 1636 bool show_pfn; 1637 }; 1638 1639 #define PAGEMAP_WALK_SIZE (PMD_SIZE) 1640 #define PAGEMAP_WALK_MASK (PMD_MASK) 1641 1642 #define PM_ENTRY_BYTES sizeof(pagemap_entry_t) 1643 #define PM_PFRAME_BITS 55 1644 #define PM_PFRAME_MASK GENMASK_ULL(PM_PFRAME_BITS - 1, 0) 1645 #define PM_SOFT_DIRTY BIT_ULL(55) 1646 #define PM_MMAP_EXCLUSIVE BIT_ULL(56) 1647 #define PM_UFFD_WP BIT_ULL(57) 1648 #define PM_GUARD_REGION BIT_ULL(58) 1649 #define PM_FILE BIT_ULL(61) 1650 #define PM_SWAP BIT_ULL(62) 1651 #define PM_PRESENT BIT_ULL(63) 1652 1653 #define PM_END_OF_BUFFER 1 1654 1655 static inline pagemap_entry_t make_pme(u64 frame, u64 flags) 1656 { 1657 return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags }; 1658 } 1659 1660 static int add_to_pagemap(pagemap_entry_t *pme, struct pagemapread *pm) 1661 { 1662 pm->buffer[pm->pos++] = *pme; 1663 if (pm->pos >= pm->len) 1664 return PM_END_OF_BUFFER; 1665 return 0; 1666 } 1667 1668 static bool __folio_page_mapped_exclusively(struct folio *folio, struct page *page) 1669 { 1670 if (IS_ENABLED(CONFIG_PAGE_MAPCOUNT)) 1671 return folio_precise_page_mapcount(folio, page) == 1; 1672 return !folio_maybe_mapped_shared(folio); 1673 } 1674 1675 static int pagemap_pte_hole(unsigned long start, unsigned long end, 1676 __always_unused int depth, struct mm_walk *walk) 1677 { 1678 struct pagemapread *pm = walk->private; 1679 unsigned long addr = start; 1680 int err = 0; 1681 1682 while (addr < end) { 1683 struct vm_area_struct *vma = find_vma(walk->mm, addr); 1684 pagemap_entry_t pme = make_pme(0, 0); 1685 /* End of address space hole, which we mark as non-present. */ 1686 unsigned long hole_end; 1687 1688 if (vma) 1689 hole_end = min(end, vma->vm_start); 1690 else 1691 hole_end = end; 1692 1693 for (; addr < hole_end; addr += PAGE_SIZE) { 1694 err = add_to_pagemap(&pme, pm); 1695 if (err) 1696 goto out; 1697 } 1698 1699 if (!vma) 1700 break; 1701 1702 /* Addresses in the VMA. */ 1703 if (vma->vm_flags & VM_SOFTDIRTY) 1704 pme = make_pme(0, PM_SOFT_DIRTY); 1705 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) { 1706 err = add_to_pagemap(&pme, pm); 1707 if (err) 1708 goto out; 1709 } 1710 } 1711 out: 1712 return err; 1713 } 1714 1715 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm, 1716 struct vm_area_struct *vma, unsigned long addr, pte_t pte) 1717 { 1718 u64 frame = 0, flags = 0; 1719 struct page *page = NULL; 1720 struct folio *folio; 1721 1722 if (pte_present(pte)) { 1723 if (pm->show_pfn) 1724 frame = pte_pfn(pte); 1725 flags |= PM_PRESENT; 1726 page = vm_normal_page(vma, addr, pte); 1727 if (pte_soft_dirty(pte)) 1728 flags |= PM_SOFT_DIRTY; 1729 if (pte_uffd_wp(pte)) 1730 flags |= PM_UFFD_WP; 1731 } else if (is_swap_pte(pte)) { 1732 swp_entry_t entry; 1733 if (pte_swp_soft_dirty(pte)) 1734 flags |= PM_SOFT_DIRTY; 1735 if (pte_swp_uffd_wp(pte)) 1736 flags |= PM_UFFD_WP; 1737 entry = pte_to_swp_entry(pte); 1738 if (pm->show_pfn) { 1739 pgoff_t offset; 1740 /* 1741 * For PFN swap offsets, keeping the offset field 1742 * to be PFN only to be compatible with old smaps. 1743 */ 1744 if (is_pfn_swap_entry(entry)) 1745 offset = swp_offset_pfn(entry); 1746 else 1747 offset = swp_offset(entry); 1748 frame = swp_type(entry) | 1749 (offset << MAX_SWAPFILES_SHIFT); 1750 } 1751 flags |= PM_SWAP; 1752 if (is_pfn_swap_entry(entry)) 1753 page = pfn_swap_entry_to_page(entry); 1754 if (pte_marker_entry_uffd_wp(entry)) 1755 flags |= PM_UFFD_WP; 1756 if (is_guard_swp_entry(entry)) 1757 flags |= PM_GUARD_REGION; 1758 } 1759 1760 if (page) { 1761 folio = page_folio(page); 1762 if (!folio_test_anon(folio)) 1763 flags |= PM_FILE; 1764 if ((flags & PM_PRESENT) && 1765 __folio_page_mapped_exclusively(folio, page)) 1766 flags |= PM_MMAP_EXCLUSIVE; 1767 } 1768 if (vma->vm_flags & VM_SOFTDIRTY) 1769 flags |= PM_SOFT_DIRTY; 1770 1771 return make_pme(frame, flags); 1772 } 1773 1774 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end, 1775 struct mm_walk *walk) 1776 { 1777 struct vm_area_struct *vma = walk->vma; 1778 struct pagemapread *pm = walk->private; 1779 spinlock_t *ptl; 1780 pte_t *pte, *orig_pte; 1781 int err = 0; 1782 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1783 1784 ptl = pmd_trans_huge_lock(pmdp, vma); 1785 if (ptl) { 1786 unsigned int idx = (addr & ~PMD_MASK) >> PAGE_SHIFT; 1787 u64 flags = 0, frame = 0; 1788 pmd_t pmd = *pmdp; 1789 struct page *page = NULL; 1790 struct folio *folio = NULL; 1791 1792 if (vma->vm_flags & VM_SOFTDIRTY) 1793 flags |= PM_SOFT_DIRTY; 1794 1795 if (pmd_present(pmd)) { 1796 page = pmd_page(pmd); 1797 1798 flags |= PM_PRESENT; 1799 if (pmd_soft_dirty(pmd)) 1800 flags |= PM_SOFT_DIRTY; 1801 if (pmd_uffd_wp(pmd)) 1802 flags |= PM_UFFD_WP; 1803 if (pm->show_pfn) 1804 frame = pmd_pfn(pmd) + idx; 1805 } 1806 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION 1807 else if (is_swap_pmd(pmd)) { 1808 swp_entry_t entry = pmd_to_swp_entry(pmd); 1809 unsigned long offset; 1810 1811 if (pm->show_pfn) { 1812 if (is_pfn_swap_entry(entry)) 1813 offset = swp_offset_pfn(entry) + idx; 1814 else 1815 offset = swp_offset(entry) + idx; 1816 frame = swp_type(entry) | 1817 (offset << MAX_SWAPFILES_SHIFT); 1818 } 1819 flags |= PM_SWAP; 1820 if (pmd_swp_soft_dirty(pmd)) 1821 flags |= PM_SOFT_DIRTY; 1822 if (pmd_swp_uffd_wp(pmd)) 1823 flags |= PM_UFFD_WP; 1824 VM_BUG_ON(!is_pmd_migration_entry(pmd)); 1825 page = pfn_swap_entry_to_page(entry); 1826 } 1827 #endif 1828 1829 if (page) { 1830 folio = page_folio(page); 1831 if (!folio_test_anon(folio)) 1832 flags |= PM_FILE; 1833 } 1834 1835 for (; addr != end; addr += PAGE_SIZE, idx++) { 1836 u64 cur_flags = flags; 1837 pagemap_entry_t pme; 1838 1839 if (folio && (flags & PM_PRESENT) && 1840 __folio_page_mapped_exclusively(folio, page)) 1841 cur_flags |= PM_MMAP_EXCLUSIVE; 1842 1843 pme = make_pme(frame, cur_flags); 1844 err = add_to_pagemap(&pme, pm); 1845 if (err) 1846 break; 1847 if (pm->show_pfn) { 1848 if (flags & PM_PRESENT) 1849 frame++; 1850 else if (flags & PM_SWAP) 1851 frame += (1 << MAX_SWAPFILES_SHIFT); 1852 } 1853 } 1854 spin_unlock(ptl); 1855 return err; 1856 } 1857 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ 1858 1859 /* 1860 * We can assume that @vma always points to a valid one and @end never 1861 * goes beyond vma->vm_end. 1862 */ 1863 orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl); 1864 if (!pte) { 1865 walk->action = ACTION_AGAIN; 1866 return err; 1867 } 1868 for (; addr < end; pte++, addr += PAGE_SIZE) { 1869 pagemap_entry_t pme; 1870 1871 pme = pte_to_pagemap_entry(pm, vma, addr, ptep_get(pte)); 1872 err = add_to_pagemap(&pme, pm); 1873 if (err) 1874 break; 1875 } 1876 pte_unmap_unlock(orig_pte, ptl); 1877 1878 cond_resched(); 1879 1880 return err; 1881 } 1882 1883 #ifdef CONFIG_HUGETLB_PAGE 1884 /* This function walks within one hugetlb entry in the single call */ 1885 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask, 1886 unsigned long addr, unsigned long end, 1887 struct mm_walk *walk) 1888 { 1889 struct pagemapread *pm = walk->private; 1890 struct vm_area_struct *vma = walk->vma; 1891 u64 flags = 0, frame = 0; 1892 int err = 0; 1893 pte_t pte; 1894 1895 if (vma->vm_flags & VM_SOFTDIRTY) 1896 flags |= PM_SOFT_DIRTY; 1897 1898 pte = huge_ptep_get(walk->mm, addr, ptep); 1899 if (pte_present(pte)) { 1900 struct folio *folio = page_folio(pte_page(pte)); 1901 1902 if (!folio_test_anon(folio)) 1903 flags |= PM_FILE; 1904 1905 if (!folio_maybe_mapped_shared(folio) && 1906 !hugetlb_pmd_shared(ptep)) 1907 flags |= PM_MMAP_EXCLUSIVE; 1908 1909 if (huge_pte_uffd_wp(pte)) 1910 flags |= PM_UFFD_WP; 1911 1912 flags |= PM_PRESENT; 1913 if (pm->show_pfn) 1914 frame = pte_pfn(pte) + 1915 ((addr & ~hmask) >> PAGE_SHIFT); 1916 } else if (pte_swp_uffd_wp_any(pte)) { 1917 flags |= PM_UFFD_WP; 1918 } 1919 1920 for (; addr != end; addr += PAGE_SIZE) { 1921 pagemap_entry_t pme = make_pme(frame, flags); 1922 1923 err = add_to_pagemap(&pme, pm); 1924 if (err) 1925 return err; 1926 if (pm->show_pfn && (flags & PM_PRESENT)) 1927 frame++; 1928 } 1929 1930 cond_resched(); 1931 1932 return err; 1933 } 1934 #else 1935 #define pagemap_hugetlb_range NULL 1936 #endif /* HUGETLB_PAGE */ 1937 1938 static const struct mm_walk_ops pagemap_ops = { 1939 .pmd_entry = pagemap_pmd_range, 1940 .pte_hole = pagemap_pte_hole, 1941 .hugetlb_entry = pagemap_hugetlb_range, 1942 .walk_lock = PGWALK_RDLOCK, 1943 }; 1944 1945 /* 1946 * /proc/pid/pagemap - an array mapping virtual pages to pfns 1947 * 1948 * For each page in the address space, this file contains one 64-bit entry 1949 * consisting of the following: 1950 * 1951 * Bits 0-54 page frame number (PFN) if present 1952 * Bits 0-4 swap type if swapped 1953 * Bits 5-54 swap offset if swapped 1954 * Bit 55 pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst) 1955 * Bit 56 page exclusively mapped 1956 * Bit 57 pte is uffd-wp write-protected 1957 * Bit 58 pte is a guard region 1958 * Bits 59-60 zero 1959 * Bit 61 page is file-page or shared-anon 1960 * Bit 62 page swapped 1961 * Bit 63 page present 1962 * 1963 * If the page is not present but in swap, then the PFN contains an 1964 * encoding of the swap file number and the page's offset into the 1965 * swap. Unmapped pages return a null PFN. This allows determining 1966 * precisely which pages are mapped (or in swap) and comparing mapped 1967 * pages between processes. 1968 * 1969 * Efficient users of this interface will use /proc/pid/maps to 1970 * determine which areas of memory are actually mapped and llseek to 1971 * skip over unmapped regions. 1972 */ 1973 static ssize_t pagemap_read(struct file *file, char __user *buf, 1974 size_t count, loff_t *ppos) 1975 { 1976 struct mm_struct *mm = file->private_data; 1977 struct pagemapread pm; 1978 unsigned long src; 1979 unsigned long svpfn; 1980 unsigned long start_vaddr; 1981 unsigned long end_vaddr; 1982 int ret = 0, copied = 0; 1983 1984 if (!mm || !mmget_not_zero(mm)) 1985 goto out; 1986 1987 ret = -EINVAL; 1988 /* file position must be aligned */ 1989 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES)) 1990 goto out_mm; 1991 1992 ret = 0; 1993 if (!count) 1994 goto out_mm; 1995 1996 /* do not disclose physical addresses: attack vector */ 1997 pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN); 1998 1999 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT); 2000 pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL); 2001 ret = -ENOMEM; 2002 if (!pm.buffer) 2003 goto out_mm; 2004 2005 src = *ppos; 2006 svpfn = src / PM_ENTRY_BYTES; 2007 end_vaddr = mm->task_size; 2008 2009 /* watch out for wraparound */ 2010 start_vaddr = end_vaddr; 2011 if (svpfn <= (ULONG_MAX >> PAGE_SHIFT)) { 2012 unsigned long end; 2013 2014 ret = mmap_read_lock_killable(mm); 2015 if (ret) 2016 goto out_free; 2017 start_vaddr = untagged_addr_remote(mm, svpfn << PAGE_SHIFT); 2018 mmap_read_unlock(mm); 2019 2020 end = start_vaddr + ((count / PM_ENTRY_BYTES) << PAGE_SHIFT); 2021 if (end >= start_vaddr && end < mm->task_size) 2022 end_vaddr = end; 2023 } 2024 2025 /* Ensure the address is inside the task */ 2026 if (start_vaddr > mm->task_size) 2027 start_vaddr = end_vaddr; 2028 2029 ret = 0; 2030 while (count && (start_vaddr < end_vaddr)) { 2031 int len; 2032 unsigned long end; 2033 2034 pm.pos = 0; 2035 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK; 2036 /* overflow ? */ 2037 if (end < start_vaddr || end > end_vaddr) 2038 end = end_vaddr; 2039 ret = mmap_read_lock_killable(mm); 2040 if (ret) 2041 goto out_free; 2042 ret = walk_page_range(mm, start_vaddr, end, &pagemap_ops, &pm); 2043 mmap_read_unlock(mm); 2044 start_vaddr = end; 2045 2046 len = min(count, PM_ENTRY_BYTES * pm.pos); 2047 if (copy_to_user(buf, pm.buffer, len)) { 2048 ret = -EFAULT; 2049 goto out_free; 2050 } 2051 copied += len; 2052 buf += len; 2053 count -= len; 2054 } 2055 *ppos += copied; 2056 if (!ret || ret == PM_END_OF_BUFFER) 2057 ret = copied; 2058 2059 out_free: 2060 kfree(pm.buffer); 2061 out_mm: 2062 mmput(mm); 2063 out: 2064 return ret; 2065 } 2066 2067 static int pagemap_open(struct inode *inode, struct file *file) 2068 { 2069 struct mm_struct *mm; 2070 2071 mm = proc_mem_open(inode, PTRACE_MODE_READ); 2072 if (IS_ERR_OR_NULL(mm)) 2073 return mm ? PTR_ERR(mm) : -ESRCH; 2074 file->private_data = mm; 2075 return 0; 2076 } 2077 2078 static int pagemap_release(struct inode *inode, struct file *file) 2079 { 2080 struct mm_struct *mm = file->private_data; 2081 2082 if (mm) 2083 mmdrop(mm); 2084 return 0; 2085 } 2086 2087 #define PM_SCAN_CATEGORIES (PAGE_IS_WPALLOWED | PAGE_IS_WRITTEN | \ 2088 PAGE_IS_FILE | PAGE_IS_PRESENT | \ 2089 PAGE_IS_SWAPPED | PAGE_IS_PFNZERO | \ 2090 PAGE_IS_HUGE | PAGE_IS_SOFT_DIRTY | \ 2091 PAGE_IS_GUARD) 2092 #define PM_SCAN_FLAGS (PM_SCAN_WP_MATCHING | PM_SCAN_CHECK_WPASYNC) 2093 2094 struct pagemap_scan_private { 2095 struct pm_scan_arg arg; 2096 unsigned long masks_of_interest, cur_vma_category; 2097 struct page_region *vec_buf; 2098 unsigned long vec_buf_len, vec_buf_index, found_pages; 2099 struct page_region __user *vec_out; 2100 }; 2101 2102 static unsigned long pagemap_page_category(struct pagemap_scan_private *p, 2103 struct vm_area_struct *vma, 2104 unsigned long addr, pte_t pte) 2105 { 2106 unsigned long categories = 0; 2107 2108 if (pte_present(pte)) { 2109 struct page *page; 2110 2111 categories |= PAGE_IS_PRESENT; 2112 if (!pte_uffd_wp(pte)) 2113 categories |= PAGE_IS_WRITTEN; 2114 2115 if (p->masks_of_interest & PAGE_IS_FILE) { 2116 page = vm_normal_page(vma, addr, pte); 2117 if (page && !PageAnon(page)) 2118 categories |= PAGE_IS_FILE; 2119 } 2120 2121 if (is_zero_pfn(pte_pfn(pte))) 2122 categories |= PAGE_IS_PFNZERO; 2123 if (pte_soft_dirty(pte)) 2124 categories |= PAGE_IS_SOFT_DIRTY; 2125 } else if (is_swap_pte(pte)) { 2126 swp_entry_t swp; 2127 2128 categories |= PAGE_IS_SWAPPED; 2129 if (!pte_swp_uffd_wp_any(pte)) 2130 categories |= PAGE_IS_WRITTEN; 2131 2132 swp = pte_to_swp_entry(pte); 2133 if (is_guard_swp_entry(swp)) 2134 categories |= PAGE_IS_GUARD; 2135 else if ((p->masks_of_interest & PAGE_IS_FILE) && 2136 is_pfn_swap_entry(swp) && 2137 !folio_test_anon(pfn_swap_entry_folio(swp))) 2138 categories |= PAGE_IS_FILE; 2139 2140 if (pte_swp_soft_dirty(pte)) 2141 categories |= PAGE_IS_SOFT_DIRTY; 2142 } 2143 2144 return categories; 2145 } 2146 2147 static void make_uffd_wp_pte(struct vm_area_struct *vma, 2148 unsigned long addr, pte_t *pte, pte_t ptent) 2149 { 2150 if (pte_present(ptent)) { 2151 pte_t old_pte; 2152 2153 old_pte = ptep_modify_prot_start(vma, addr, pte); 2154 ptent = pte_mkuffd_wp(old_pte); 2155 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent); 2156 } else if (is_swap_pte(ptent)) { 2157 ptent = pte_swp_mkuffd_wp(ptent); 2158 set_pte_at(vma->vm_mm, addr, pte, ptent); 2159 } else { 2160 set_pte_at(vma->vm_mm, addr, pte, 2161 make_pte_marker(PTE_MARKER_UFFD_WP)); 2162 } 2163 } 2164 2165 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 2166 static unsigned long pagemap_thp_category(struct pagemap_scan_private *p, 2167 struct vm_area_struct *vma, 2168 unsigned long addr, pmd_t pmd) 2169 { 2170 unsigned long categories = PAGE_IS_HUGE; 2171 2172 if (pmd_present(pmd)) { 2173 struct page *page; 2174 2175 categories |= PAGE_IS_PRESENT; 2176 if (!pmd_uffd_wp(pmd)) 2177 categories |= PAGE_IS_WRITTEN; 2178 2179 if (p->masks_of_interest & PAGE_IS_FILE) { 2180 page = vm_normal_page_pmd(vma, addr, pmd); 2181 if (page && !PageAnon(page)) 2182 categories |= PAGE_IS_FILE; 2183 } 2184 2185 if (is_zero_pfn(pmd_pfn(pmd))) 2186 categories |= PAGE_IS_PFNZERO; 2187 if (pmd_soft_dirty(pmd)) 2188 categories |= PAGE_IS_SOFT_DIRTY; 2189 } else if (is_swap_pmd(pmd)) { 2190 swp_entry_t swp; 2191 2192 categories |= PAGE_IS_SWAPPED; 2193 if (!pmd_swp_uffd_wp(pmd)) 2194 categories |= PAGE_IS_WRITTEN; 2195 if (pmd_swp_soft_dirty(pmd)) 2196 categories |= PAGE_IS_SOFT_DIRTY; 2197 2198 if (p->masks_of_interest & PAGE_IS_FILE) { 2199 swp = pmd_to_swp_entry(pmd); 2200 if (is_pfn_swap_entry(swp) && 2201 !folio_test_anon(pfn_swap_entry_folio(swp))) 2202 categories |= PAGE_IS_FILE; 2203 } 2204 } 2205 2206 return categories; 2207 } 2208 2209 static void make_uffd_wp_pmd(struct vm_area_struct *vma, 2210 unsigned long addr, pmd_t *pmdp) 2211 { 2212 pmd_t old, pmd = *pmdp; 2213 2214 if (pmd_present(pmd)) { 2215 old = pmdp_invalidate_ad(vma, addr, pmdp); 2216 pmd = pmd_mkuffd_wp(old); 2217 set_pmd_at(vma->vm_mm, addr, pmdp, pmd); 2218 } else if (is_migration_entry(pmd_to_swp_entry(pmd))) { 2219 pmd = pmd_swp_mkuffd_wp(pmd); 2220 set_pmd_at(vma->vm_mm, addr, pmdp, pmd); 2221 } 2222 } 2223 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ 2224 2225 #ifdef CONFIG_HUGETLB_PAGE 2226 static unsigned long pagemap_hugetlb_category(pte_t pte) 2227 { 2228 unsigned long categories = PAGE_IS_HUGE; 2229 2230 /* 2231 * According to pagemap_hugetlb_range(), file-backed HugeTLB 2232 * page cannot be swapped. So PAGE_IS_FILE is not checked for 2233 * swapped pages. 2234 */ 2235 if (pte_present(pte)) { 2236 categories |= PAGE_IS_PRESENT; 2237 if (!huge_pte_uffd_wp(pte)) 2238 categories |= PAGE_IS_WRITTEN; 2239 if (!PageAnon(pte_page(pte))) 2240 categories |= PAGE_IS_FILE; 2241 if (is_zero_pfn(pte_pfn(pte))) 2242 categories |= PAGE_IS_PFNZERO; 2243 if (pte_soft_dirty(pte)) 2244 categories |= PAGE_IS_SOFT_DIRTY; 2245 } else if (is_swap_pte(pte)) { 2246 categories |= PAGE_IS_SWAPPED; 2247 if (!pte_swp_uffd_wp_any(pte)) 2248 categories |= PAGE_IS_WRITTEN; 2249 if (pte_swp_soft_dirty(pte)) 2250 categories |= PAGE_IS_SOFT_DIRTY; 2251 } 2252 2253 return categories; 2254 } 2255 2256 static void make_uffd_wp_huge_pte(struct vm_area_struct *vma, 2257 unsigned long addr, pte_t *ptep, 2258 pte_t ptent) 2259 { 2260 unsigned long psize; 2261 2262 if (is_hugetlb_entry_hwpoisoned(ptent) || is_pte_marker(ptent)) 2263 return; 2264 2265 psize = huge_page_size(hstate_vma(vma)); 2266 2267 if (is_hugetlb_entry_migration(ptent)) 2268 set_huge_pte_at(vma->vm_mm, addr, ptep, 2269 pte_swp_mkuffd_wp(ptent), psize); 2270 else if (!huge_pte_none(ptent)) 2271 huge_ptep_modify_prot_commit(vma, addr, ptep, ptent, 2272 huge_pte_mkuffd_wp(ptent)); 2273 else 2274 set_huge_pte_at(vma->vm_mm, addr, ptep, 2275 make_pte_marker(PTE_MARKER_UFFD_WP), psize); 2276 } 2277 #endif /* CONFIG_HUGETLB_PAGE */ 2278 2279 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLB_PAGE) 2280 static void pagemap_scan_backout_range(struct pagemap_scan_private *p, 2281 unsigned long addr, unsigned long end) 2282 { 2283 struct page_region *cur_buf = &p->vec_buf[p->vec_buf_index]; 2284 2285 if (cur_buf->start != addr) 2286 cur_buf->end = addr; 2287 else 2288 cur_buf->start = cur_buf->end = 0; 2289 2290 p->found_pages -= (end - addr) / PAGE_SIZE; 2291 } 2292 #endif 2293 2294 static bool pagemap_scan_is_interesting_page(unsigned long categories, 2295 const struct pagemap_scan_private *p) 2296 { 2297 categories ^= p->arg.category_inverted; 2298 if ((categories & p->arg.category_mask) != p->arg.category_mask) 2299 return false; 2300 if (p->arg.category_anyof_mask && !(categories & p->arg.category_anyof_mask)) 2301 return false; 2302 2303 return true; 2304 } 2305 2306 static bool pagemap_scan_is_interesting_vma(unsigned long categories, 2307 const struct pagemap_scan_private *p) 2308 { 2309 unsigned long required = p->arg.category_mask & PAGE_IS_WPALLOWED; 2310 2311 categories ^= p->arg.category_inverted; 2312 if ((categories & required) != required) 2313 return false; 2314 2315 return true; 2316 } 2317 2318 static int pagemap_scan_test_walk(unsigned long start, unsigned long end, 2319 struct mm_walk *walk) 2320 { 2321 struct pagemap_scan_private *p = walk->private; 2322 struct vm_area_struct *vma = walk->vma; 2323 unsigned long vma_category = 0; 2324 bool wp_allowed = userfaultfd_wp_async(vma) && 2325 userfaultfd_wp_use_markers(vma); 2326 2327 if (!wp_allowed) { 2328 /* User requested explicit failure over wp-async capability */ 2329 if (p->arg.flags & PM_SCAN_CHECK_WPASYNC) 2330 return -EPERM; 2331 /* 2332 * User requires wr-protect, and allows silently skipping 2333 * unsupported vmas. 2334 */ 2335 if (p->arg.flags & PM_SCAN_WP_MATCHING) 2336 return 1; 2337 /* 2338 * Then the request doesn't involve wr-protects at all, 2339 * fall through to the rest checks, and allow vma walk. 2340 */ 2341 } 2342 2343 if (vma->vm_flags & VM_PFNMAP) 2344 return 1; 2345 2346 if (wp_allowed) 2347 vma_category |= PAGE_IS_WPALLOWED; 2348 2349 if (vma->vm_flags & VM_SOFTDIRTY) 2350 vma_category |= PAGE_IS_SOFT_DIRTY; 2351 2352 if (!pagemap_scan_is_interesting_vma(vma_category, p)) 2353 return 1; 2354 2355 p->cur_vma_category = vma_category; 2356 2357 return 0; 2358 } 2359 2360 static bool pagemap_scan_push_range(unsigned long categories, 2361 struct pagemap_scan_private *p, 2362 unsigned long addr, unsigned long end) 2363 { 2364 struct page_region *cur_buf = &p->vec_buf[p->vec_buf_index]; 2365 2366 /* 2367 * When there is no output buffer provided at all, the sentinel values 2368 * won't match here. There is no other way for `cur_buf->end` to be 2369 * non-zero other than it being non-empty. 2370 */ 2371 if (addr == cur_buf->end && categories == cur_buf->categories) { 2372 cur_buf->end = end; 2373 return true; 2374 } 2375 2376 if (cur_buf->end) { 2377 if (p->vec_buf_index >= p->vec_buf_len - 1) 2378 return false; 2379 2380 cur_buf = &p->vec_buf[++p->vec_buf_index]; 2381 } 2382 2383 cur_buf->start = addr; 2384 cur_buf->end = end; 2385 cur_buf->categories = categories; 2386 2387 return true; 2388 } 2389 2390 static int pagemap_scan_output(unsigned long categories, 2391 struct pagemap_scan_private *p, 2392 unsigned long addr, unsigned long *end) 2393 { 2394 unsigned long n_pages, total_pages; 2395 int ret = 0; 2396 2397 if (!p->vec_buf) 2398 return 0; 2399 2400 categories &= p->arg.return_mask; 2401 2402 n_pages = (*end - addr) / PAGE_SIZE; 2403 if (check_add_overflow(p->found_pages, n_pages, &total_pages) || 2404 total_pages > p->arg.max_pages) { 2405 size_t n_too_much = total_pages - p->arg.max_pages; 2406 *end -= n_too_much * PAGE_SIZE; 2407 n_pages -= n_too_much; 2408 ret = -ENOSPC; 2409 } 2410 2411 if (!pagemap_scan_push_range(categories, p, addr, *end)) { 2412 *end = addr; 2413 n_pages = 0; 2414 ret = -ENOSPC; 2415 } 2416 2417 p->found_pages += n_pages; 2418 if (ret) 2419 p->arg.walk_end = *end; 2420 2421 return ret; 2422 } 2423 2424 static int pagemap_scan_thp_entry(pmd_t *pmd, unsigned long start, 2425 unsigned long end, struct mm_walk *walk) 2426 { 2427 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 2428 struct pagemap_scan_private *p = walk->private; 2429 struct vm_area_struct *vma = walk->vma; 2430 unsigned long categories; 2431 spinlock_t *ptl; 2432 int ret = 0; 2433 2434 ptl = pmd_trans_huge_lock(pmd, vma); 2435 if (!ptl) 2436 return -ENOENT; 2437 2438 categories = p->cur_vma_category | 2439 pagemap_thp_category(p, vma, start, *pmd); 2440 2441 if (!pagemap_scan_is_interesting_page(categories, p)) 2442 goto out_unlock; 2443 2444 ret = pagemap_scan_output(categories, p, start, &end); 2445 if (start == end) 2446 goto out_unlock; 2447 2448 if (~p->arg.flags & PM_SCAN_WP_MATCHING) 2449 goto out_unlock; 2450 if (~categories & PAGE_IS_WRITTEN) 2451 goto out_unlock; 2452 2453 /* 2454 * Break huge page into small pages if the WP operation 2455 * needs to be performed on a portion of the huge page. 2456 */ 2457 if (end != start + HPAGE_SIZE) { 2458 spin_unlock(ptl); 2459 split_huge_pmd(vma, pmd, start); 2460 pagemap_scan_backout_range(p, start, end); 2461 /* Report as if there was no THP */ 2462 return -ENOENT; 2463 } 2464 2465 make_uffd_wp_pmd(vma, start, pmd); 2466 flush_tlb_range(vma, start, end); 2467 out_unlock: 2468 spin_unlock(ptl); 2469 return ret; 2470 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */ 2471 return -ENOENT; 2472 #endif 2473 } 2474 2475 static int pagemap_scan_pmd_entry(pmd_t *pmd, unsigned long start, 2476 unsigned long end, struct mm_walk *walk) 2477 { 2478 struct pagemap_scan_private *p = walk->private; 2479 struct vm_area_struct *vma = walk->vma; 2480 unsigned long addr, flush_end = 0; 2481 pte_t *pte, *start_pte; 2482 spinlock_t *ptl; 2483 int ret; 2484 2485 ret = pagemap_scan_thp_entry(pmd, start, end, walk); 2486 if (ret != -ENOENT) 2487 return ret; 2488 2489 ret = 0; 2490 start_pte = pte = pte_offset_map_lock(vma->vm_mm, pmd, start, &ptl); 2491 if (!pte) { 2492 walk->action = ACTION_AGAIN; 2493 return 0; 2494 } 2495 2496 arch_enter_lazy_mmu_mode(); 2497 2498 if ((p->arg.flags & PM_SCAN_WP_MATCHING) && !p->vec_out) { 2499 /* Fast path for performing exclusive WP */ 2500 for (addr = start; addr != end; pte++, addr += PAGE_SIZE) { 2501 pte_t ptent = ptep_get(pte); 2502 2503 if ((pte_present(ptent) && pte_uffd_wp(ptent)) || 2504 pte_swp_uffd_wp_any(ptent)) 2505 continue; 2506 make_uffd_wp_pte(vma, addr, pte, ptent); 2507 if (!flush_end) 2508 start = addr; 2509 flush_end = addr + PAGE_SIZE; 2510 } 2511 goto flush_and_return; 2512 } 2513 2514 if (!p->arg.category_anyof_mask && !p->arg.category_inverted && 2515 p->arg.category_mask == PAGE_IS_WRITTEN && 2516 p->arg.return_mask == PAGE_IS_WRITTEN) { 2517 for (addr = start; addr < end; pte++, addr += PAGE_SIZE) { 2518 unsigned long next = addr + PAGE_SIZE; 2519 pte_t ptent = ptep_get(pte); 2520 2521 if ((pte_present(ptent) && pte_uffd_wp(ptent)) || 2522 pte_swp_uffd_wp_any(ptent)) 2523 continue; 2524 ret = pagemap_scan_output(p->cur_vma_category | PAGE_IS_WRITTEN, 2525 p, addr, &next); 2526 if (next == addr) 2527 break; 2528 if (~p->arg.flags & PM_SCAN_WP_MATCHING) 2529 continue; 2530 make_uffd_wp_pte(vma, addr, pte, ptent); 2531 if (!flush_end) 2532 start = addr; 2533 flush_end = next; 2534 } 2535 goto flush_and_return; 2536 } 2537 2538 for (addr = start; addr != end; pte++, addr += PAGE_SIZE) { 2539 pte_t ptent = ptep_get(pte); 2540 unsigned long categories = p->cur_vma_category | 2541 pagemap_page_category(p, vma, addr, ptent); 2542 unsigned long next = addr + PAGE_SIZE; 2543 2544 if (!pagemap_scan_is_interesting_page(categories, p)) 2545 continue; 2546 2547 ret = pagemap_scan_output(categories, p, addr, &next); 2548 if (next == addr) 2549 break; 2550 2551 if (~p->arg.flags & PM_SCAN_WP_MATCHING) 2552 continue; 2553 if (~categories & PAGE_IS_WRITTEN) 2554 continue; 2555 2556 make_uffd_wp_pte(vma, addr, pte, ptent); 2557 if (!flush_end) 2558 start = addr; 2559 flush_end = next; 2560 } 2561 2562 flush_and_return: 2563 if (flush_end) 2564 flush_tlb_range(vma, start, addr); 2565 2566 arch_leave_lazy_mmu_mode(); 2567 pte_unmap_unlock(start_pte, ptl); 2568 2569 cond_resched(); 2570 return ret; 2571 } 2572 2573 #ifdef CONFIG_HUGETLB_PAGE 2574 static int pagemap_scan_hugetlb_entry(pte_t *ptep, unsigned long hmask, 2575 unsigned long start, unsigned long end, 2576 struct mm_walk *walk) 2577 { 2578 struct pagemap_scan_private *p = walk->private; 2579 struct vm_area_struct *vma = walk->vma; 2580 unsigned long categories; 2581 spinlock_t *ptl; 2582 int ret = 0; 2583 pte_t pte; 2584 2585 if (~p->arg.flags & PM_SCAN_WP_MATCHING) { 2586 /* Go the short route when not write-protecting pages. */ 2587 2588 pte = huge_ptep_get(walk->mm, start, ptep); 2589 categories = p->cur_vma_category | pagemap_hugetlb_category(pte); 2590 2591 if (!pagemap_scan_is_interesting_page(categories, p)) 2592 return 0; 2593 2594 return pagemap_scan_output(categories, p, start, &end); 2595 } 2596 2597 i_mmap_lock_write(vma->vm_file->f_mapping); 2598 ptl = huge_pte_lock(hstate_vma(vma), vma->vm_mm, ptep); 2599 2600 pte = huge_ptep_get(walk->mm, start, ptep); 2601 categories = p->cur_vma_category | pagemap_hugetlb_category(pte); 2602 2603 if (!pagemap_scan_is_interesting_page(categories, p)) 2604 goto out_unlock; 2605 2606 ret = pagemap_scan_output(categories, p, start, &end); 2607 if (start == end) 2608 goto out_unlock; 2609 2610 if (~categories & PAGE_IS_WRITTEN) 2611 goto out_unlock; 2612 2613 if (end != start + HPAGE_SIZE) { 2614 /* Partial HugeTLB page WP isn't possible. */ 2615 pagemap_scan_backout_range(p, start, end); 2616 p->arg.walk_end = start; 2617 ret = 0; 2618 goto out_unlock; 2619 } 2620 2621 make_uffd_wp_huge_pte(vma, start, ptep, pte); 2622 flush_hugetlb_tlb_range(vma, start, end); 2623 2624 out_unlock: 2625 spin_unlock(ptl); 2626 i_mmap_unlock_write(vma->vm_file->f_mapping); 2627 2628 return ret; 2629 } 2630 #else 2631 #define pagemap_scan_hugetlb_entry NULL 2632 #endif 2633 2634 static int pagemap_scan_pte_hole(unsigned long addr, unsigned long end, 2635 int depth, struct mm_walk *walk) 2636 { 2637 struct pagemap_scan_private *p = walk->private; 2638 struct vm_area_struct *vma = walk->vma; 2639 int ret, err; 2640 2641 if (!vma || !pagemap_scan_is_interesting_page(p->cur_vma_category, p)) 2642 return 0; 2643 2644 ret = pagemap_scan_output(p->cur_vma_category, p, addr, &end); 2645 if (addr == end) 2646 return ret; 2647 2648 if (~p->arg.flags & PM_SCAN_WP_MATCHING) 2649 return ret; 2650 2651 err = uffd_wp_range(vma, addr, end - addr, true); 2652 if (err < 0) 2653 ret = err; 2654 2655 return ret; 2656 } 2657 2658 static const struct mm_walk_ops pagemap_scan_ops = { 2659 .test_walk = pagemap_scan_test_walk, 2660 .pmd_entry = pagemap_scan_pmd_entry, 2661 .pte_hole = pagemap_scan_pte_hole, 2662 .hugetlb_entry = pagemap_scan_hugetlb_entry, 2663 }; 2664 2665 static int pagemap_scan_get_args(struct pm_scan_arg *arg, 2666 unsigned long uarg) 2667 { 2668 if (copy_from_user(arg, (void __user *)uarg, sizeof(*arg))) 2669 return -EFAULT; 2670 2671 if (arg->size != sizeof(struct pm_scan_arg)) 2672 return -EINVAL; 2673 2674 /* Validate requested features */ 2675 if (arg->flags & ~PM_SCAN_FLAGS) 2676 return -EINVAL; 2677 if ((arg->category_inverted | arg->category_mask | 2678 arg->category_anyof_mask | arg->return_mask) & ~PM_SCAN_CATEGORIES) 2679 return -EINVAL; 2680 2681 arg->start = untagged_addr((unsigned long)arg->start); 2682 arg->end = untagged_addr((unsigned long)arg->end); 2683 arg->vec = untagged_addr((unsigned long)arg->vec); 2684 2685 /* Validate memory pointers */ 2686 if (!IS_ALIGNED(arg->start, PAGE_SIZE)) 2687 return -EINVAL; 2688 if (!access_ok((void __user *)(long)arg->start, arg->end - arg->start)) 2689 return -EFAULT; 2690 if (!arg->vec && arg->vec_len) 2691 return -EINVAL; 2692 if (UINT_MAX == SIZE_MAX && arg->vec_len > SIZE_MAX) 2693 return -EINVAL; 2694 if (arg->vec && !access_ok((void __user *)(long)arg->vec, 2695 size_mul(arg->vec_len, sizeof(struct page_region)))) 2696 return -EFAULT; 2697 2698 /* Fixup default values */ 2699 arg->end = ALIGN(arg->end, PAGE_SIZE); 2700 arg->walk_end = 0; 2701 if (!arg->max_pages) 2702 arg->max_pages = ULONG_MAX; 2703 2704 return 0; 2705 } 2706 2707 static int pagemap_scan_writeback_args(struct pm_scan_arg *arg, 2708 unsigned long uargl) 2709 { 2710 struct pm_scan_arg __user *uarg = (void __user *)uargl; 2711 2712 if (copy_to_user(&uarg->walk_end, &arg->walk_end, sizeof(arg->walk_end))) 2713 return -EFAULT; 2714 2715 return 0; 2716 } 2717 2718 static int pagemap_scan_init_bounce_buffer(struct pagemap_scan_private *p) 2719 { 2720 if (!p->arg.vec_len) 2721 return 0; 2722 2723 p->vec_buf_len = min_t(size_t, PAGEMAP_WALK_SIZE >> PAGE_SHIFT, 2724 p->arg.vec_len); 2725 p->vec_buf = kmalloc_array(p->vec_buf_len, sizeof(*p->vec_buf), 2726 GFP_KERNEL); 2727 if (!p->vec_buf) 2728 return -ENOMEM; 2729 2730 p->vec_buf->start = p->vec_buf->end = 0; 2731 p->vec_out = (struct page_region __user *)(long)p->arg.vec; 2732 2733 return 0; 2734 } 2735 2736 static long pagemap_scan_flush_buffer(struct pagemap_scan_private *p) 2737 { 2738 const struct page_region *buf = p->vec_buf; 2739 long n = p->vec_buf_index; 2740 2741 if (!p->vec_buf) 2742 return 0; 2743 2744 if (buf[n].end != buf[n].start) 2745 n++; 2746 2747 if (!n) 2748 return 0; 2749 2750 if (copy_to_user(p->vec_out, buf, n * sizeof(*buf))) 2751 return -EFAULT; 2752 2753 p->arg.vec_len -= n; 2754 p->vec_out += n; 2755 2756 p->vec_buf_index = 0; 2757 p->vec_buf_len = min_t(size_t, p->vec_buf_len, p->arg.vec_len); 2758 p->vec_buf->start = p->vec_buf->end = 0; 2759 2760 return n; 2761 } 2762 2763 static long do_pagemap_scan(struct mm_struct *mm, unsigned long uarg) 2764 { 2765 struct pagemap_scan_private p = {0}; 2766 unsigned long walk_start; 2767 size_t n_ranges_out = 0; 2768 int ret; 2769 2770 ret = pagemap_scan_get_args(&p.arg, uarg); 2771 if (ret) 2772 return ret; 2773 2774 p.masks_of_interest = p.arg.category_mask | p.arg.category_anyof_mask | 2775 p.arg.return_mask; 2776 ret = pagemap_scan_init_bounce_buffer(&p); 2777 if (ret) 2778 return ret; 2779 2780 for (walk_start = p.arg.start; walk_start < p.arg.end; 2781 walk_start = p.arg.walk_end) { 2782 struct mmu_notifier_range range; 2783 long n_out; 2784 2785 if (fatal_signal_pending(current)) { 2786 ret = -EINTR; 2787 break; 2788 } 2789 2790 ret = mmap_read_lock_killable(mm); 2791 if (ret) 2792 break; 2793 2794 /* Protection change for the range is going to happen. */ 2795 if (p.arg.flags & PM_SCAN_WP_MATCHING) { 2796 mmu_notifier_range_init(&range, MMU_NOTIFY_PROTECTION_VMA, 0, 2797 mm, walk_start, p.arg.end); 2798 mmu_notifier_invalidate_range_start(&range); 2799 } 2800 2801 ret = walk_page_range(mm, walk_start, p.arg.end, 2802 &pagemap_scan_ops, &p); 2803 2804 if (p.arg.flags & PM_SCAN_WP_MATCHING) 2805 mmu_notifier_invalidate_range_end(&range); 2806 2807 mmap_read_unlock(mm); 2808 2809 n_out = pagemap_scan_flush_buffer(&p); 2810 if (n_out < 0) 2811 ret = n_out; 2812 else 2813 n_ranges_out += n_out; 2814 2815 if (ret != -ENOSPC) 2816 break; 2817 2818 if (p.arg.vec_len == 0 || p.found_pages == p.arg.max_pages) 2819 break; 2820 } 2821 2822 /* ENOSPC signifies early stop (buffer full) from the walk. */ 2823 if (!ret || ret == -ENOSPC) 2824 ret = n_ranges_out; 2825 2826 /* The walk_end isn't set when ret is zero */ 2827 if (!p.arg.walk_end) 2828 p.arg.walk_end = p.arg.end; 2829 if (pagemap_scan_writeback_args(&p.arg, uarg)) 2830 ret = -EFAULT; 2831 2832 kfree(p.vec_buf); 2833 return ret; 2834 } 2835 2836 static long do_pagemap_cmd(struct file *file, unsigned int cmd, 2837 unsigned long arg) 2838 { 2839 struct mm_struct *mm = file->private_data; 2840 2841 switch (cmd) { 2842 case PAGEMAP_SCAN: 2843 return do_pagemap_scan(mm, arg); 2844 2845 default: 2846 return -EINVAL; 2847 } 2848 } 2849 2850 const struct file_operations proc_pagemap_operations = { 2851 .llseek = mem_lseek, /* borrow this */ 2852 .read = pagemap_read, 2853 .open = pagemap_open, 2854 .release = pagemap_release, 2855 .unlocked_ioctl = do_pagemap_cmd, 2856 .compat_ioctl = do_pagemap_cmd, 2857 }; 2858 #endif /* CONFIG_PROC_PAGE_MONITOR */ 2859 2860 #ifdef CONFIG_NUMA 2861 2862 struct numa_maps { 2863 unsigned long pages; 2864 unsigned long anon; 2865 unsigned long active; 2866 unsigned long writeback; 2867 unsigned long mapcount_max; 2868 unsigned long dirty; 2869 unsigned long swapcache; 2870 unsigned long node[MAX_NUMNODES]; 2871 }; 2872 2873 struct numa_maps_private { 2874 struct proc_maps_private proc_maps; 2875 struct numa_maps md; 2876 }; 2877 2878 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty, 2879 unsigned long nr_pages) 2880 { 2881 struct folio *folio = page_folio(page); 2882 int count; 2883 2884 if (IS_ENABLED(CONFIG_PAGE_MAPCOUNT)) 2885 count = folio_precise_page_mapcount(folio, page); 2886 else 2887 count = folio_average_page_mapcount(folio); 2888 2889 md->pages += nr_pages; 2890 if (pte_dirty || folio_test_dirty(folio)) 2891 md->dirty += nr_pages; 2892 2893 if (folio_test_swapcache(folio)) 2894 md->swapcache += nr_pages; 2895 2896 if (folio_test_active(folio) || folio_test_unevictable(folio)) 2897 md->active += nr_pages; 2898 2899 if (folio_test_writeback(folio)) 2900 md->writeback += nr_pages; 2901 2902 if (folio_test_anon(folio)) 2903 md->anon += nr_pages; 2904 2905 if (count > md->mapcount_max) 2906 md->mapcount_max = count; 2907 2908 md->node[folio_nid(folio)] += nr_pages; 2909 } 2910 2911 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma, 2912 unsigned long addr) 2913 { 2914 struct page *page; 2915 int nid; 2916 2917 if (!pte_present(pte)) 2918 return NULL; 2919 2920 page = vm_normal_page(vma, addr, pte); 2921 if (!page || is_zone_device_page(page)) 2922 return NULL; 2923 2924 if (PageReserved(page)) 2925 return NULL; 2926 2927 nid = page_to_nid(page); 2928 if (!node_isset(nid, node_states[N_MEMORY])) 2929 return NULL; 2930 2931 return page; 2932 } 2933 2934 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 2935 static struct page *can_gather_numa_stats_pmd(pmd_t pmd, 2936 struct vm_area_struct *vma, 2937 unsigned long addr) 2938 { 2939 struct page *page; 2940 int nid; 2941 2942 if (!pmd_present(pmd)) 2943 return NULL; 2944 2945 page = vm_normal_page_pmd(vma, addr, pmd); 2946 if (!page) 2947 return NULL; 2948 2949 if (PageReserved(page)) 2950 return NULL; 2951 2952 nid = page_to_nid(page); 2953 if (!node_isset(nid, node_states[N_MEMORY])) 2954 return NULL; 2955 2956 return page; 2957 } 2958 #endif 2959 2960 static int gather_pte_stats(pmd_t *pmd, unsigned long addr, 2961 unsigned long end, struct mm_walk *walk) 2962 { 2963 struct numa_maps *md = walk->private; 2964 struct vm_area_struct *vma = walk->vma; 2965 spinlock_t *ptl; 2966 pte_t *orig_pte; 2967 pte_t *pte; 2968 2969 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 2970 ptl = pmd_trans_huge_lock(pmd, vma); 2971 if (ptl) { 2972 struct page *page; 2973 2974 page = can_gather_numa_stats_pmd(*pmd, vma, addr); 2975 if (page) 2976 gather_stats(page, md, pmd_dirty(*pmd), 2977 HPAGE_PMD_SIZE/PAGE_SIZE); 2978 spin_unlock(ptl); 2979 return 0; 2980 } 2981 #endif 2982 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl); 2983 if (!pte) { 2984 walk->action = ACTION_AGAIN; 2985 return 0; 2986 } 2987 do { 2988 pte_t ptent = ptep_get(pte); 2989 struct page *page = can_gather_numa_stats(ptent, vma, addr); 2990 if (!page) 2991 continue; 2992 gather_stats(page, md, pte_dirty(ptent), 1); 2993 2994 } while (pte++, addr += PAGE_SIZE, addr != end); 2995 pte_unmap_unlock(orig_pte, ptl); 2996 cond_resched(); 2997 return 0; 2998 } 2999 #ifdef CONFIG_HUGETLB_PAGE 3000 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask, 3001 unsigned long addr, unsigned long end, struct mm_walk *walk) 3002 { 3003 pte_t huge_pte = huge_ptep_get(walk->mm, addr, pte); 3004 struct numa_maps *md; 3005 struct page *page; 3006 3007 if (!pte_present(huge_pte)) 3008 return 0; 3009 3010 page = pte_page(huge_pte); 3011 3012 md = walk->private; 3013 gather_stats(page, md, pte_dirty(huge_pte), 1); 3014 return 0; 3015 } 3016 3017 #else 3018 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask, 3019 unsigned long addr, unsigned long end, struct mm_walk *walk) 3020 { 3021 return 0; 3022 } 3023 #endif 3024 3025 static const struct mm_walk_ops show_numa_ops = { 3026 .hugetlb_entry = gather_hugetlb_stats, 3027 .pmd_entry = gather_pte_stats, 3028 .walk_lock = PGWALK_RDLOCK, 3029 }; 3030 3031 /* 3032 * Display pages allocated per node and memory policy via /proc. 3033 */ 3034 static int show_numa_map(struct seq_file *m, void *v) 3035 { 3036 struct numa_maps_private *numa_priv = m->private; 3037 struct proc_maps_private *proc_priv = &numa_priv->proc_maps; 3038 struct vm_area_struct *vma = v; 3039 struct numa_maps *md = &numa_priv->md; 3040 struct file *file = vma->vm_file; 3041 struct mm_struct *mm = vma->vm_mm; 3042 char buffer[64]; 3043 struct mempolicy *pol; 3044 pgoff_t ilx; 3045 int nid; 3046 3047 if (!mm) 3048 return 0; 3049 3050 /* Ensure we start with an empty set of numa_maps statistics. */ 3051 memset(md, 0, sizeof(*md)); 3052 3053 pol = __get_vma_policy(vma, vma->vm_start, &ilx); 3054 if (pol) { 3055 mpol_to_str(buffer, sizeof(buffer), pol); 3056 mpol_cond_put(pol); 3057 } else { 3058 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy); 3059 } 3060 3061 seq_printf(m, "%08lx %s", vma->vm_start, buffer); 3062 3063 if (file) { 3064 seq_puts(m, " file="); 3065 seq_path(m, file_user_path(file), "\n\t= "); 3066 } else if (vma_is_initial_heap(vma)) { 3067 seq_puts(m, " heap"); 3068 } else if (vma_is_initial_stack(vma)) { 3069 seq_puts(m, " stack"); 3070 } 3071 3072 if (is_vm_hugetlb_page(vma)) 3073 seq_puts(m, " huge"); 3074 3075 /* mmap_lock is held by m_start */ 3076 walk_page_vma(vma, &show_numa_ops, md); 3077 3078 if (!md->pages) 3079 goto out; 3080 3081 if (md->anon) 3082 seq_printf(m, " anon=%lu", md->anon); 3083 3084 if (md->dirty) 3085 seq_printf(m, " dirty=%lu", md->dirty); 3086 3087 if (md->pages != md->anon && md->pages != md->dirty) 3088 seq_printf(m, " mapped=%lu", md->pages); 3089 3090 if (md->mapcount_max > 1) 3091 seq_printf(m, " mapmax=%lu", md->mapcount_max); 3092 3093 if (md->swapcache) 3094 seq_printf(m, " swapcache=%lu", md->swapcache); 3095 3096 if (md->active < md->pages && !is_vm_hugetlb_page(vma)) 3097 seq_printf(m, " active=%lu", md->active); 3098 3099 if (md->writeback) 3100 seq_printf(m, " writeback=%lu", md->writeback); 3101 3102 for_each_node_state(nid, N_MEMORY) 3103 if (md->node[nid]) 3104 seq_printf(m, " N%d=%lu", nid, md->node[nid]); 3105 3106 seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10); 3107 out: 3108 seq_putc(m, '\n'); 3109 return 0; 3110 } 3111 3112 static const struct seq_operations proc_pid_numa_maps_op = { 3113 .start = m_start, 3114 .next = m_next, 3115 .stop = m_stop, 3116 .show = show_numa_map, 3117 }; 3118 3119 static int pid_numa_maps_open(struct inode *inode, struct file *file) 3120 { 3121 return proc_maps_open(inode, file, &proc_pid_numa_maps_op, 3122 sizeof(struct numa_maps_private)); 3123 } 3124 3125 const struct file_operations proc_pid_numa_maps_operations = { 3126 .open = pid_numa_maps_open, 3127 .read = seq_read, 3128 .llseek = seq_lseek, 3129 .release = proc_map_release, 3130 }; 3131 3132 #endif /* CONFIG_NUMA */ 3133