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(priv->mm)) { 216 int err = PTR_ERR(priv->mm); 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; 710 711 /* 712 * First accumulate quantities that depend only on |size| and the type 713 * of the compound page. 714 */ 715 if (folio_test_anon(folio)) { 716 mss->anonymous += size; 717 if (!folio_test_swapbacked(folio) && !dirty && 718 !folio_test_dirty(folio)) 719 mss->lazyfree += size; 720 } 721 722 if (folio_test_ksm(folio)) 723 mss->ksm += size; 724 725 mss->resident += size; 726 /* Accumulate the size in pages that have been accessed. */ 727 if (young || folio_test_young(folio) || folio_test_referenced(folio)) 728 mss->referenced += size; 729 730 /* 731 * Then accumulate quantities that may depend on sharing, or that may 732 * differ page-by-page. 733 * 734 * refcount == 1 for present entries guarantees that the folio is mapped 735 * exactly once. For large folios this implies that exactly one 736 * PTE/PMD/... maps (a part of) this folio. 737 * 738 * Treat all non-present entries (where relying on the mapcount and 739 * refcount doesn't make sense) as "maybe shared, but not sure how 740 * often". We treat device private entries as being fake-present. 741 * 742 * Note that it would not be safe to read the mapcount especially for 743 * pages referenced by migration entries, even with the PTL held. 744 */ 745 if (folio_ref_count(folio) == 1 || !present) { 746 smaps_page_accumulate(mss, folio, size, size << PSS_SHIFT, 747 dirty, locked, present); 748 return; 749 } 750 /* 751 * We obtain a snapshot of the mapcount. Without holding the folio lock 752 * this snapshot can be slightly wrong as we cannot always read the 753 * mapcount atomically. 754 */ 755 for (i = 0; i < nr; i++, page++) { 756 int mapcount = folio_precise_page_mapcount(folio, page); 757 unsigned long pss = PAGE_SIZE << PSS_SHIFT; 758 if (mapcount >= 2) 759 pss /= mapcount; 760 smaps_page_accumulate(mss, folio, PAGE_SIZE, pss, 761 dirty, locked, mapcount < 2); 762 } 763 } 764 765 #ifdef CONFIG_SHMEM 766 static int smaps_pte_hole(unsigned long addr, unsigned long end, 767 __always_unused int depth, struct mm_walk *walk) 768 { 769 struct mem_size_stats *mss = walk->private; 770 struct vm_area_struct *vma = walk->vma; 771 772 mss->swap += shmem_partial_swap_usage(walk->vma->vm_file->f_mapping, 773 linear_page_index(vma, addr), 774 linear_page_index(vma, end)); 775 776 return 0; 777 } 778 #else 779 #define smaps_pte_hole NULL 780 #endif /* CONFIG_SHMEM */ 781 782 static void smaps_pte_hole_lookup(unsigned long addr, struct mm_walk *walk) 783 { 784 #ifdef CONFIG_SHMEM 785 if (walk->ops->pte_hole) { 786 /* depth is not used */ 787 smaps_pte_hole(addr, addr + PAGE_SIZE, 0, walk); 788 } 789 #endif 790 } 791 792 static void smaps_pte_entry(pte_t *pte, unsigned long addr, 793 struct mm_walk *walk) 794 { 795 struct mem_size_stats *mss = walk->private; 796 struct vm_area_struct *vma = walk->vma; 797 bool locked = !!(vma->vm_flags & VM_LOCKED); 798 struct page *page = NULL; 799 bool present = false, young = false, dirty = false; 800 pte_t ptent = ptep_get(pte); 801 802 if (pte_present(ptent)) { 803 page = vm_normal_page(vma, addr, ptent); 804 young = pte_young(ptent); 805 dirty = pte_dirty(ptent); 806 present = true; 807 } else if (is_swap_pte(ptent)) { 808 swp_entry_t swpent = pte_to_swp_entry(ptent); 809 810 if (!non_swap_entry(swpent)) { 811 int mapcount; 812 813 mss->swap += PAGE_SIZE; 814 mapcount = swp_swapcount(swpent); 815 if (mapcount >= 2) { 816 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT; 817 818 do_div(pss_delta, mapcount); 819 mss->swap_pss += pss_delta; 820 } else { 821 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT; 822 } 823 } else if (is_pfn_swap_entry(swpent)) { 824 if (is_device_private_entry(swpent)) 825 present = true; 826 page = pfn_swap_entry_to_page(swpent); 827 } 828 } else { 829 smaps_pte_hole_lookup(addr, walk); 830 return; 831 } 832 833 if (!page) 834 return; 835 836 smaps_account(mss, page, false, young, dirty, locked, present); 837 } 838 839 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 840 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr, 841 struct mm_walk *walk) 842 { 843 struct mem_size_stats *mss = walk->private; 844 struct vm_area_struct *vma = walk->vma; 845 bool locked = !!(vma->vm_flags & VM_LOCKED); 846 struct page *page = NULL; 847 bool present = false; 848 struct folio *folio; 849 850 if (pmd_present(*pmd)) { 851 page = vm_normal_page_pmd(vma, addr, *pmd); 852 present = true; 853 } else if (unlikely(thp_migration_supported() && is_swap_pmd(*pmd))) { 854 swp_entry_t entry = pmd_to_swp_entry(*pmd); 855 856 if (is_pfn_swap_entry(entry)) 857 page = pfn_swap_entry_to_page(entry); 858 } 859 if (IS_ERR_OR_NULL(page)) 860 return; 861 folio = page_folio(page); 862 if (folio_test_anon(folio)) 863 mss->anonymous_thp += HPAGE_PMD_SIZE; 864 else if (folio_test_swapbacked(folio)) 865 mss->shmem_thp += HPAGE_PMD_SIZE; 866 else if (folio_is_zone_device(folio)) 867 /* pass */; 868 else 869 mss->file_thp += HPAGE_PMD_SIZE; 870 871 smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd), 872 locked, present); 873 } 874 #else 875 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr, 876 struct mm_walk *walk) 877 { 878 } 879 #endif 880 881 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, 882 struct mm_walk *walk) 883 { 884 struct vm_area_struct *vma = walk->vma; 885 pte_t *pte; 886 spinlock_t *ptl; 887 888 ptl = pmd_trans_huge_lock(pmd, vma); 889 if (ptl) { 890 smaps_pmd_entry(pmd, addr, walk); 891 spin_unlock(ptl); 892 goto out; 893 } 894 895 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); 896 if (!pte) { 897 walk->action = ACTION_AGAIN; 898 return 0; 899 } 900 for (; addr != end; pte++, addr += PAGE_SIZE) 901 smaps_pte_entry(pte, addr, walk); 902 pte_unmap_unlock(pte - 1, ptl); 903 out: 904 cond_resched(); 905 return 0; 906 } 907 908 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma) 909 { 910 /* 911 * Don't forget to update Documentation/ on changes. 912 */ 913 static const char mnemonics[BITS_PER_LONG][2] = { 914 /* 915 * In case if we meet a flag we don't know about. 916 */ 917 [0 ... (BITS_PER_LONG-1)] = "??", 918 919 [ilog2(VM_READ)] = "rd", 920 [ilog2(VM_WRITE)] = "wr", 921 [ilog2(VM_EXEC)] = "ex", 922 [ilog2(VM_SHARED)] = "sh", 923 [ilog2(VM_MAYREAD)] = "mr", 924 [ilog2(VM_MAYWRITE)] = "mw", 925 [ilog2(VM_MAYEXEC)] = "me", 926 [ilog2(VM_MAYSHARE)] = "ms", 927 [ilog2(VM_GROWSDOWN)] = "gd", 928 [ilog2(VM_PFNMAP)] = "pf", 929 [ilog2(VM_LOCKED)] = "lo", 930 [ilog2(VM_IO)] = "io", 931 [ilog2(VM_SEQ_READ)] = "sr", 932 [ilog2(VM_RAND_READ)] = "rr", 933 [ilog2(VM_DONTCOPY)] = "dc", 934 [ilog2(VM_DONTEXPAND)] = "de", 935 [ilog2(VM_LOCKONFAULT)] = "lf", 936 [ilog2(VM_ACCOUNT)] = "ac", 937 [ilog2(VM_NORESERVE)] = "nr", 938 [ilog2(VM_HUGETLB)] = "ht", 939 [ilog2(VM_SYNC)] = "sf", 940 [ilog2(VM_ARCH_1)] = "ar", 941 [ilog2(VM_WIPEONFORK)] = "wf", 942 [ilog2(VM_DONTDUMP)] = "dd", 943 #ifdef CONFIG_ARM64_BTI 944 [ilog2(VM_ARM64_BTI)] = "bt", 945 #endif 946 #ifdef CONFIG_MEM_SOFT_DIRTY 947 [ilog2(VM_SOFTDIRTY)] = "sd", 948 #endif 949 [ilog2(VM_MIXEDMAP)] = "mm", 950 [ilog2(VM_HUGEPAGE)] = "hg", 951 [ilog2(VM_NOHUGEPAGE)] = "nh", 952 [ilog2(VM_MERGEABLE)] = "mg", 953 [ilog2(VM_UFFD_MISSING)]= "um", 954 [ilog2(VM_UFFD_WP)] = "uw", 955 #ifdef CONFIG_ARM64_MTE 956 [ilog2(VM_MTE)] = "mt", 957 [ilog2(VM_MTE_ALLOWED)] = "", 958 #endif 959 #ifdef CONFIG_ARCH_HAS_PKEYS 960 /* These come out via ProtectionKey: */ 961 [ilog2(VM_PKEY_BIT0)] = "", 962 [ilog2(VM_PKEY_BIT1)] = "", 963 [ilog2(VM_PKEY_BIT2)] = "", 964 [ilog2(VM_PKEY_BIT3)] = "", 965 #if VM_PKEY_BIT4 966 [ilog2(VM_PKEY_BIT4)] = "", 967 #endif 968 #endif /* CONFIG_ARCH_HAS_PKEYS */ 969 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_MINOR 970 [ilog2(VM_UFFD_MINOR)] = "ui", 971 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_MINOR */ 972 #ifdef CONFIG_X86_USER_SHADOW_STACK 973 [ilog2(VM_SHADOW_STACK)] = "ss", 974 #endif 975 #ifdef CONFIG_64BIT 976 [ilog2(VM_DROPPABLE)] = "dp", 977 [ilog2(VM_SEALED)] = "sl", 978 #endif 979 }; 980 size_t i; 981 982 seq_puts(m, "VmFlags: "); 983 for (i = 0; i < BITS_PER_LONG; i++) { 984 if (!mnemonics[i][0]) 985 continue; 986 if (vma->vm_flags & (1UL << i)) { 987 seq_putc(m, mnemonics[i][0]); 988 seq_putc(m, mnemonics[i][1]); 989 seq_putc(m, ' '); 990 } 991 } 992 seq_putc(m, '\n'); 993 } 994 995 #ifdef CONFIG_HUGETLB_PAGE 996 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask, 997 unsigned long addr, unsigned long end, 998 struct mm_walk *walk) 999 { 1000 struct mem_size_stats *mss = walk->private; 1001 struct vm_area_struct *vma = walk->vma; 1002 pte_t ptent = huge_ptep_get(walk->mm, addr, pte); 1003 struct folio *folio = NULL; 1004 bool present = false; 1005 1006 if (pte_present(ptent)) { 1007 folio = page_folio(pte_page(ptent)); 1008 present = true; 1009 } else if (is_swap_pte(ptent)) { 1010 swp_entry_t swpent = pte_to_swp_entry(ptent); 1011 1012 if (is_pfn_swap_entry(swpent)) 1013 folio = pfn_swap_entry_folio(swpent); 1014 } 1015 1016 if (folio) { 1017 /* We treat non-present entries as "maybe shared". */ 1018 if (!present || folio_likely_mapped_shared(folio) || 1019 hugetlb_pmd_shared(pte)) 1020 mss->shared_hugetlb += huge_page_size(hstate_vma(vma)); 1021 else 1022 mss->private_hugetlb += huge_page_size(hstate_vma(vma)); 1023 } 1024 return 0; 1025 } 1026 #else 1027 #define smaps_hugetlb_range NULL 1028 #endif /* HUGETLB_PAGE */ 1029 1030 static const struct mm_walk_ops smaps_walk_ops = { 1031 .pmd_entry = smaps_pte_range, 1032 .hugetlb_entry = smaps_hugetlb_range, 1033 .walk_lock = PGWALK_RDLOCK, 1034 }; 1035 1036 static const struct mm_walk_ops smaps_shmem_walk_ops = { 1037 .pmd_entry = smaps_pte_range, 1038 .hugetlb_entry = smaps_hugetlb_range, 1039 .pte_hole = smaps_pte_hole, 1040 .walk_lock = PGWALK_RDLOCK, 1041 }; 1042 1043 /* 1044 * Gather mem stats from @vma with the indicated beginning 1045 * address @start, and keep them in @mss. 1046 * 1047 * Use vm_start of @vma as the beginning address if @start is 0. 1048 */ 1049 static void smap_gather_stats(struct vm_area_struct *vma, 1050 struct mem_size_stats *mss, unsigned long start) 1051 { 1052 const struct mm_walk_ops *ops = &smaps_walk_ops; 1053 1054 /* Invalid start */ 1055 if (start >= vma->vm_end) 1056 return; 1057 1058 if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) { 1059 /* 1060 * For shared or readonly shmem mappings we know that all 1061 * swapped out pages belong to the shmem object, and we can 1062 * obtain the swap value much more efficiently. For private 1063 * writable mappings, we might have COW pages that are 1064 * not affected by the parent swapped out pages of the shmem 1065 * object, so we have to distinguish them during the page walk. 1066 * Unless we know that the shmem object (or the part mapped by 1067 * our VMA) has no swapped out pages at all. 1068 */ 1069 unsigned long shmem_swapped = shmem_swap_usage(vma); 1070 1071 if (!start && (!shmem_swapped || (vma->vm_flags & VM_SHARED) || 1072 !(vma->vm_flags & VM_WRITE))) { 1073 mss->swap += shmem_swapped; 1074 } else { 1075 ops = &smaps_shmem_walk_ops; 1076 } 1077 } 1078 1079 /* mmap_lock is held in m_start */ 1080 if (!start) 1081 walk_page_vma(vma, ops, mss); 1082 else 1083 walk_page_range(vma->vm_mm, start, vma->vm_end, ops, mss); 1084 } 1085 1086 #define SEQ_PUT_DEC(str, val) \ 1087 seq_put_decimal_ull_width(m, str, (val) >> 10, 8) 1088 1089 /* Show the contents common for smaps and smaps_rollup */ 1090 static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss, 1091 bool rollup_mode) 1092 { 1093 SEQ_PUT_DEC("Rss: ", mss->resident); 1094 SEQ_PUT_DEC(" kB\nPss: ", mss->pss >> PSS_SHIFT); 1095 SEQ_PUT_DEC(" kB\nPss_Dirty: ", mss->pss_dirty >> PSS_SHIFT); 1096 if (rollup_mode) { 1097 /* 1098 * These are meaningful only for smaps_rollup, otherwise two of 1099 * them are zero, and the other one is the same as Pss. 1100 */ 1101 SEQ_PUT_DEC(" kB\nPss_Anon: ", 1102 mss->pss_anon >> PSS_SHIFT); 1103 SEQ_PUT_DEC(" kB\nPss_File: ", 1104 mss->pss_file >> PSS_SHIFT); 1105 SEQ_PUT_DEC(" kB\nPss_Shmem: ", 1106 mss->pss_shmem >> PSS_SHIFT); 1107 } 1108 SEQ_PUT_DEC(" kB\nShared_Clean: ", mss->shared_clean); 1109 SEQ_PUT_DEC(" kB\nShared_Dirty: ", mss->shared_dirty); 1110 SEQ_PUT_DEC(" kB\nPrivate_Clean: ", mss->private_clean); 1111 SEQ_PUT_DEC(" kB\nPrivate_Dirty: ", mss->private_dirty); 1112 SEQ_PUT_DEC(" kB\nReferenced: ", mss->referenced); 1113 SEQ_PUT_DEC(" kB\nAnonymous: ", mss->anonymous); 1114 SEQ_PUT_DEC(" kB\nKSM: ", mss->ksm); 1115 SEQ_PUT_DEC(" kB\nLazyFree: ", mss->lazyfree); 1116 SEQ_PUT_DEC(" kB\nAnonHugePages: ", mss->anonymous_thp); 1117 SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp); 1118 SEQ_PUT_DEC(" kB\nFilePmdMapped: ", mss->file_thp); 1119 SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb); 1120 seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ", 1121 mss->private_hugetlb >> 10, 7); 1122 SEQ_PUT_DEC(" kB\nSwap: ", mss->swap); 1123 SEQ_PUT_DEC(" kB\nSwapPss: ", 1124 mss->swap_pss >> PSS_SHIFT); 1125 SEQ_PUT_DEC(" kB\nLocked: ", 1126 mss->pss_locked >> PSS_SHIFT); 1127 seq_puts(m, " kB\n"); 1128 } 1129 1130 static int show_smap(struct seq_file *m, void *v) 1131 { 1132 struct vm_area_struct *vma = v; 1133 struct mem_size_stats mss = {}; 1134 1135 smap_gather_stats(vma, &mss, 0); 1136 1137 show_map_vma(m, vma); 1138 1139 SEQ_PUT_DEC("Size: ", vma->vm_end - vma->vm_start); 1140 SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma)); 1141 SEQ_PUT_DEC(" kB\nMMUPageSize: ", vma_mmu_pagesize(vma)); 1142 seq_puts(m, " kB\n"); 1143 1144 __show_smap(m, &mss, false); 1145 1146 seq_printf(m, "THPeligible: %8u\n", 1147 !!thp_vma_allowable_orders(vma, vma->vm_flags, 1148 TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL)); 1149 1150 if (arch_pkeys_enabled()) 1151 seq_printf(m, "ProtectionKey: %8u\n", vma_pkey(vma)); 1152 show_smap_vma_flags(m, vma); 1153 1154 return 0; 1155 } 1156 1157 static int show_smaps_rollup(struct seq_file *m, void *v) 1158 { 1159 struct proc_maps_private *priv = m->private; 1160 struct mem_size_stats mss = {}; 1161 struct mm_struct *mm = priv->mm; 1162 struct vm_area_struct *vma; 1163 unsigned long vma_start = 0, last_vma_end = 0; 1164 int ret = 0; 1165 VMA_ITERATOR(vmi, mm, 0); 1166 1167 priv->task = get_proc_task(priv->inode); 1168 if (!priv->task) 1169 return -ESRCH; 1170 1171 if (!mm || !mmget_not_zero(mm)) { 1172 ret = -ESRCH; 1173 goto out_put_task; 1174 } 1175 1176 ret = mmap_read_lock_killable(mm); 1177 if (ret) 1178 goto out_put_mm; 1179 1180 hold_task_mempolicy(priv); 1181 vma = vma_next(&vmi); 1182 1183 if (unlikely(!vma)) 1184 goto empty_set; 1185 1186 vma_start = vma->vm_start; 1187 do { 1188 smap_gather_stats(vma, &mss, 0); 1189 last_vma_end = vma->vm_end; 1190 1191 /* 1192 * Release mmap_lock temporarily if someone wants to 1193 * access it for write request. 1194 */ 1195 if (mmap_lock_is_contended(mm)) { 1196 vma_iter_invalidate(&vmi); 1197 mmap_read_unlock(mm); 1198 ret = mmap_read_lock_killable(mm); 1199 if (ret) { 1200 release_task_mempolicy(priv); 1201 goto out_put_mm; 1202 } 1203 1204 /* 1205 * After dropping the lock, there are four cases to 1206 * consider. See the following example for explanation. 1207 * 1208 * +------+------+-----------+ 1209 * | VMA1 | VMA2 | VMA3 | 1210 * +------+------+-----------+ 1211 * | | | | 1212 * 4k 8k 16k 400k 1213 * 1214 * Suppose we drop the lock after reading VMA2 due to 1215 * contention, then we get: 1216 * 1217 * last_vma_end = 16k 1218 * 1219 * 1) VMA2 is freed, but VMA3 exists: 1220 * 1221 * vma_next(vmi) will return VMA3. 1222 * In this case, just continue from VMA3. 1223 * 1224 * 2) VMA2 still exists: 1225 * 1226 * vma_next(vmi) will return VMA3. 1227 * In this case, just continue from VMA3. 1228 * 1229 * 3) No more VMAs can be found: 1230 * 1231 * vma_next(vmi) will return NULL. 1232 * No more things to do, just break. 1233 * 1234 * 4) (last_vma_end - 1) is the middle of a vma (VMA'): 1235 * 1236 * vma_next(vmi) will return VMA' whose range 1237 * contains last_vma_end. 1238 * Iterate VMA' from last_vma_end. 1239 */ 1240 vma = vma_next(&vmi); 1241 /* Case 3 above */ 1242 if (!vma) 1243 break; 1244 1245 /* Case 1 and 2 above */ 1246 if (vma->vm_start >= last_vma_end) { 1247 smap_gather_stats(vma, &mss, 0); 1248 last_vma_end = vma->vm_end; 1249 continue; 1250 } 1251 1252 /* Case 4 above */ 1253 if (vma->vm_end > last_vma_end) { 1254 smap_gather_stats(vma, &mss, last_vma_end); 1255 last_vma_end = vma->vm_end; 1256 } 1257 } 1258 } for_each_vma(vmi, vma); 1259 1260 empty_set: 1261 show_vma_header_prefix(m, vma_start, last_vma_end, 0, 0, 0, 0); 1262 seq_pad(m, ' '); 1263 seq_puts(m, "[rollup]\n"); 1264 1265 __show_smap(m, &mss, true); 1266 1267 release_task_mempolicy(priv); 1268 mmap_read_unlock(mm); 1269 1270 out_put_mm: 1271 mmput(mm); 1272 out_put_task: 1273 put_task_struct(priv->task); 1274 priv->task = NULL; 1275 1276 return ret; 1277 } 1278 #undef SEQ_PUT_DEC 1279 1280 static const struct seq_operations proc_pid_smaps_op = { 1281 .start = m_start, 1282 .next = m_next, 1283 .stop = m_stop, 1284 .show = show_smap 1285 }; 1286 1287 static int pid_smaps_open(struct inode *inode, struct file *file) 1288 { 1289 return do_maps_open(inode, file, &proc_pid_smaps_op); 1290 } 1291 1292 static int smaps_rollup_open(struct inode *inode, struct file *file) 1293 { 1294 int ret; 1295 struct proc_maps_private *priv; 1296 1297 priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT); 1298 if (!priv) 1299 return -ENOMEM; 1300 1301 ret = single_open(file, show_smaps_rollup, priv); 1302 if (ret) 1303 goto out_free; 1304 1305 priv->inode = inode; 1306 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ); 1307 if (IS_ERR(priv->mm)) { 1308 ret = PTR_ERR(priv->mm); 1309 1310 single_release(inode, file); 1311 goto out_free; 1312 } 1313 1314 return 0; 1315 1316 out_free: 1317 kfree(priv); 1318 return ret; 1319 } 1320 1321 static int smaps_rollup_release(struct inode *inode, struct file *file) 1322 { 1323 struct seq_file *seq = file->private_data; 1324 struct proc_maps_private *priv = seq->private; 1325 1326 if (priv->mm) 1327 mmdrop(priv->mm); 1328 1329 kfree(priv); 1330 return single_release(inode, file); 1331 } 1332 1333 const struct file_operations proc_pid_smaps_operations = { 1334 .open = pid_smaps_open, 1335 .read = seq_read, 1336 .llseek = seq_lseek, 1337 .release = proc_map_release, 1338 }; 1339 1340 const struct file_operations proc_pid_smaps_rollup_operations = { 1341 .open = smaps_rollup_open, 1342 .read = seq_read, 1343 .llseek = seq_lseek, 1344 .release = smaps_rollup_release, 1345 }; 1346 1347 enum clear_refs_types { 1348 CLEAR_REFS_ALL = 1, 1349 CLEAR_REFS_ANON, 1350 CLEAR_REFS_MAPPED, 1351 CLEAR_REFS_SOFT_DIRTY, 1352 CLEAR_REFS_MM_HIWATER_RSS, 1353 CLEAR_REFS_LAST, 1354 }; 1355 1356 struct clear_refs_private { 1357 enum clear_refs_types type; 1358 }; 1359 1360 #ifdef CONFIG_MEM_SOFT_DIRTY 1361 1362 static inline bool pte_is_pinned(struct vm_area_struct *vma, unsigned long addr, pte_t pte) 1363 { 1364 struct folio *folio; 1365 1366 if (!pte_write(pte)) 1367 return false; 1368 if (!is_cow_mapping(vma->vm_flags)) 1369 return false; 1370 if (likely(!test_bit(MMF_HAS_PINNED, &vma->vm_mm->flags))) 1371 return false; 1372 folio = vm_normal_folio(vma, addr, pte); 1373 if (!folio) 1374 return false; 1375 return folio_maybe_dma_pinned(folio); 1376 } 1377 1378 static inline void clear_soft_dirty(struct vm_area_struct *vma, 1379 unsigned long addr, pte_t *pte) 1380 { 1381 /* 1382 * The soft-dirty tracker uses #PF-s to catch writes 1383 * to pages, so write-protect the pte as well. See the 1384 * Documentation/admin-guide/mm/soft-dirty.rst for full description 1385 * of how soft-dirty works. 1386 */ 1387 pte_t ptent = ptep_get(pte); 1388 1389 if (pte_present(ptent)) { 1390 pte_t old_pte; 1391 1392 if (pte_is_pinned(vma, addr, ptent)) 1393 return; 1394 old_pte = ptep_modify_prot_start(vma, addr, pte); 1395 ptent = pte_wrprotect(old_pte); 1396 ptent = pte_clear_soft_dirty(ptent); 1397 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent); 1398 } else if (is_swap_pte(ptent)) { 1399 ptent = pte_swp_clear_soft_dirty(ptent); 1400 set_pte_at(vma->vm_mm, addr, pte, ptent); 1401 } 1402 } 1403 #else 1404 static inline void clear_soft_dirty(struct vm_area_struct *vma, 1405 unsigned long addr, pte_t *pte) 1406 { 1407 } 1408 #endif 1409 1410 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE) 1411 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma, 1412 unsigned long addr, pmd_t *pmdp) 1413 { 1414 pmd_t old, pmd = *pmdp; 1415 1416 if (pmd_present(pmd)) { 1417 /* See comment in change_huge_pmd() */ 1418 old = pmdp_invalidate(vma, addr, pmdp); 1419 if (pmd_dirty(old)) 1420 pmd = pmd_mkdirty(pmd); 1421 if (pmd_young(old)) 1422 pmd = pmd_mkyoung(pmd); 1423 1424 pmd = pmd_wrprotect(pmd); 1425 pmd = pmd_clear_soft_dirty(pmd); 1426 1427 set_pmd_at(vma->vm_mm, addr, pmdp, pmd); 1428 } else if (is_migration_entry(pmd_to_swp_entry(pmd))) { 1429 pmd = pmd_swp_clear_soft_dirty(pmd); 1430 set_pmd_at(vma->vm_mm, addr, pmdp, pmd); 1431 } 1432 } 1433 #else 1434 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma, 1435 unsigned long addr, pmd_t *pmdp) 1436 { 1437 } 1438 #endif 1439 1440 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr, 1441 unsigned long end, struct mm_walk *walk) 1442 { 1443 struct clear_refs_private *cp = walk->private; 1444 struct vm_area_struct *vma = walk->vma; 1445 pte_t *pte, ptent; 1446 spinlock_t *ptl; 1447 struct folio *folio; 1448 1449 ptl = pmd_trans_huge_lock(pmd, vma); 1450 if (ptl) { 1451 if (cp->type == CLEAR_REFS_SOFT_DIRTY) { 1452 clear_soft_dirty_pmd(vma, addr, pmd); 1453 goto out; 1454 } 1455 1456 if (!pmd_present(*pmd)) 1457 goto out; 1458 1459 folio = pmd_folio(*pmd); 1460 1461 /* Clear accessed and referenced bits. */ 1462 pmdp_test_and_clear_young(vma, addr, pmd); 1463 folio_test_clear_young(folio); 1464 folio_clear_referenced(folio); 1465 out: 1466 spin_unlock(ptl); 1467 return 0; 1468 } 1469 1470 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); 1471 if (!pte) { 1472 walk->action = ACTION_AGAIN; 1473 return 0; 1474 } 1475 for (; addr != end; pte++, addr += PAGE_SIZE) { 1476 ptent = ptep_get(pte); 1477 1478 if (cp->type == CLEAR_REFS_SOFT_DIRTY) { 1479 clear_soft_dirty(vma, addr, pte); 1480 continue; 1481 } 1482 1483 if (!pte_present(ptent)) 1484 continue; 1485 1486 folio = vm_normal_folio(vma, addr, ptent); 1487 if (!folio) 1488 continue; 1489 1490 /* Clear accessed and referenced bits. */ 1491 ptep_test_and_clear_young(vma, addr, pte); 1492 folio_test_clear_young(folio); 1493 folio_clear_referenced(folio); 1494 } 1495 pte_unmap_unlock(pte - 1, ptl); 1496 cond_resched(); 1497 return 0; 1498 } 1499 1500 static int clear_refs_test_walk(unsigned long start, unsigned long end, 1501 struct mm_walk *walk) 1502 { 1503 struct clear_refs_private *cp = walk->private; 1504 struct vm_area_struct *vma = walk->vma; 1505 1506 if (vma->vm_flags & VM_PFNMAP) 1507 return 1; 1508 1509 /* 1510 * Writing 1 to /proc/pid/clear_refs affects all pages. 1511 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages. 1512 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages. 1513 * Writing 4 to /proc/pid/clear_refs affects all pages. 1514 */ 1515 if (cp->type == CLEAR_REFS_ANON && vma->vm_file) 1516 return 1; 1517 if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file) 1518 return 1; 1519 return 0; 1520 } 1521 1522 static const struct mm_walk_ops clear_refs_walk_ops = { 1523 .pmd_entry = clear_refs_pte_range, 1524 .test_walk = clear_refs_test_walk, 1525 .walk_lock = PGWALK_WRLOCK, 1526 }; 1527 1528 static ssize_t clear_refs_write(struct file *file, const char __user *buf, 1529 size_t count, loff_t *ppos) 1530 { 1531 struct task_struct *task; 1532 char buffer[PROC_NUMBUF] = {}; 1533 struct mm_struct *mm; 1534 struct vm_area_struct *vma; 1535 enum clear_refs_types type; 1536 int itype; 1537 int rv; 1538 1539 if (count > sizeof(buffer) - 1) 1540 count = sizeof(buffer) - 1; 1541 if (copy_from_user(buffer, buf, count)) 1542 return -EFAULT; 1543 rv = kstrtoint(strstrip(buffer), 10, &itype); 1544 if (rv < 0) 1545 return rv; 1546 type = (enum clear_refs_types)itype; 1547 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST) 1548 return -EINVAL; 1549 1550 task = get_proc_task(file_inode(file)); 1551 if (!task) 1552 return -ESRCH; 1553 mm = get_task_mm(task); 1554 if (mm) { 1555 VMA_ITERATOR(vmi, mm, 0); 1556 struct mmu_notifier_range range; 1557 struct clear_refs_private cp = { 1558 .type = type, 1559 }; 1560 1561 if (mmap_write_lock_killable(mm)) { 1562 count = -EINTR; 1563 goto out_mm; 1564 } 1565 if (type == CLEAR_REFS_MM_HIWATER_RSS) { 1566 /* 1567 * Writing 5 to /proc/pid/clear_refs resets the peak 1568 * resident set size to this mm's current rss value. 1569 */ 1570 reset_mm_hiwater_rss(mm); 1571 goto out_unlock; 1572 } 1573 1574 if (type == CLEAR_REFS_SOFT_DIRTY) { 1575 for_each_vma(vmi, vma) { 1576 if (!(vma->vm_flags & VM_SOFTDIRTY)) 1577 continue; 1578 vm_flags_clear(vma, VM_SOFTDIRTY); 1579 vma_set_page_prot(vma); 1580 } 1581 1582 inc_tlb_flush_pending(mm); 1583 mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY, 1584 0, mm, 0, -1UL); 1585 mmu_notifier_invalidate_range_start(&range); 1586 } 1587 walk_page_range(mm, 0, -1, &clear_refs_walk_ops, &cp); 1588 if (type == CLEAR_REFS_SOFT_DIRTY) { 1589 mmu_notifier_invalidate_range_end(&range); 1590 flush_tlb_mm(mm); 1591 dec_tlb_flush_pending(mm); 1592 } 1593 out_unlock: 1594 mmap_write_unlock(mm); 1595 out_mm: 1596 mmput(mm); 1597 } 1598 put_task_struct(task); 1599 1600 return count; 1601 } 1602 1603 const struct file_operations proc_clear_refs_operations = { 1604 .write = clear_refs_write, 1605 .llseek = noop_llseek, 1606 }; 1607 1608 typedef struct { 1609 u64 pme; 1610 } pagemap_entry_t; 1611 1612 struct pagemapread { 1613 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */ 1614 pagemap_entry_t *buffer; 1615 bool show_pfn; 1616 }; 1617 1618 #define PAGEMAP_WALK_SIZE (PMD_SIZE) 1619 #define PAGEMAP_WALK_MASK (PMD_MASK) 1620 1621 #define PM_ENTRY_BYTES sizeof(pagemap_entry_t) 1622 #define PM_PFRAME_BITS 55 1623 #define PM_PFRAME_MASK GENMASK_ULL(PM_PFRAME_BITS - 1, 0) 1624 #define PM_SOFT_DIRTY BIT_ULL(55) 1625 #define PM_MMAP_EXCLUSIVE BIT_ULL(56) 1626 #define PM_UFFD_WP BIT_ULL(57) 1627 #define PM_FILE BIT_ULL(61) 1628 #define PM_SWAP BIT_ULL(62) 1629 #define PM_PRESENT BIT_ULL(63) 1630 1631 #define PM_END_OF_BUFFER 1 1632 1633 static inline pagemap_entry_t make_pme(u64 frame, u64 flags) 1634 { 1635 return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags }; 1636 } 1637 1638 static int add_to_pagemap(pagemap_entry_t *pme, struct pagemapread *pm) 1639 { 1640 pm->buffer[pm->pos++] = *pme; 1641 if (pm->pos >= pm->len) 1642 return PM_END_OF_BUFFER; 1643 return 0; 1644 } 1645 1646 static int pagemap_pte_hole(unsigned long start, unsigned long end, 1647 __always_unused int depth, struct mm_walk *walk) 1648 { 1649 struct pagemapread *pm = walk->private; 1650 unsigned long addr = start; 1651 int err = 0; 1652 1653 while (addr < end) { 1654 struct vm_area_struct *vma = find_vma(walk->mm, addr); 1655 pagemap_entry_t pme = make_pme(0, 0); 1656 /* End of address space hole, which we mark as non-present. */ 1657 unsigned long hole_end; 1658 1659 if (vma) 1660 hole_end = min(end, vma->vm_start); 1661 else 1662 hole_end = end; 1663 1664 for (; addr < hole_end; addr += PAGE_SIZE) { 1665 err = add_to_pagemap(&pme, pm); 1666 if (err) 1667 goto out; 1668 } 1669 1670 if (!vma) 1671 break; 1672 1673 /* Addresses in the VMA. */ 1674 if (vma->vm_flags & VM_SOFTDIRTY) 1675 pme = make_pme(0, PM_SOFT_DIRTY); 1676 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) { 1677 err = add_to_pagemap(&pme, pm); 1678 if (err) 1679 goto out; 1680 } 1681 } 1682 out: 1683 return err; 1684 } 1685 1686 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm, 1687 struct vm_area_struct *vma, unsigned long addr, pte_t pte) 1688 { 1689 u64 frame = 0, flags = 0; 1690 struct page *page = NULL; 1691 struct folio *folio; 1692 1693 if (pte_present(pte)) { 1694 if (pm->show_pfn) 1695 frame = pte_pfn(pte); 1696 flags |= PM_PRESENT; 1697 page = vm_normal_page(vma, addr, pte); 1698 if (pte_soft_dirty(pte)) 1699 flags |= PM_SOFT_DIRTY; 1700 if (pte_uffd_wp(pte)) 1701 flags |= PM_UFFD_WP; 1702 } else if (is_swap_pte(pte)) { 1703 swp_entry_t entry; 1704 if (pte_swp_soft_dirty(pte)) 1705 flags |= PM_SOFT_DIRTY; 1706 if (pte_swp_uffd_wp(pte)) 1707 flags |= PM_UFFD_WP; 1708 entry = pte_to_swp_entry(pte); 1709 if (pm->show_pfn) { 1710 pgoff_t offset; 1711 /* 1712 * For PFN swap offsets, keeping the offset field 1713 * to be PFN only to be compatible with old smaps. 1714 */ 1715 if (is_pfn_swap_entry(entry)) 1716 offset = swp_offset_pfn(entry); 1717 else 1718 offset = swp_offset(entry); 1719 frame = swp_type(entry) | 1720 (offset << MAX_SWAPFILES_SHIFT); 1721 } 1722 flags |= PM_SWAP; 1723 if (is_pfn_swap_entry(entry)) 1724 page = pfn_swap_entry_to_page(entry); 1725 if (pte_marker_entry_uffd_wp(entry)) 1726 flags |= PM_UFFD_WP; 1727 } 1728 1729 if (page) { 1730 folio = page_folio(page); 1731 if (!folio_test_anon(folio)) 1732 flags |= PM_FILE; 1733 if ((flags & PM_PRESENT) && 1734 folio_precise_page_mapcount(folio, page) == 1) 1735 flags |= PM_MMAP_EXCLUSIVE; 1736 } 1737 if (vma->vm_flags & VM_SOFTDIRTY) 1738 flags |= PM_SOFT_DIRTY; 1739 1740 return make_pme(frame, flags); 1741 } 1742 1743 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end, 1744 struct mm_walk *walk) 1745 { 1746 struct vm_area_struct *vma = walk->vma; 1747 struct pagemapread *pm = walk->private; 1748 spinlock_t *ptl; 1749 pte_t *pte, *orig_pte; 1750 int err = 0; 1751 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1752 1753 ptl = pmd_trans_huge_lock(pmdp, vma); 1754 if (ptl) { 1755 unsigned int idx = (addr & ~PMD_MASK) >> PAGE_SHIFT; 1756 u64 flags = 0, frame = 0; 1757 pmd_t pmd = *pmdp; 1758 struct page *page = NULL; 1759 struct folio *folio = NULL; 1760 1761 if (vma->vm_flags & VM_SOFTDIRTY) 1762 flags |= PM_SOFT_DIRTY; 1763 1764 if (pmd_present(pmd)) { 1765 page = pmd_page(pmd); 1766 1767 flags |= PM_PRESENT; 1768 if (pmd_soft_dirty(pmd)) 1769 flags |= PM_SOFT_DIRTY; 1770 if (pmd_uffd_wp(pmd)) 1771 flags |= PM_UFFD_WP; 1772 if (pm->show_pfn) 1773 frame = pmd_pfn(pmd) + idx; 1774 } 1775 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION 1776 else if (is_swap_pmd(pmd)) { 1777 swp_entry_t entry = pmd_to_swp_entry(pmd); 1778 unsigned long offset; 1779 1780 if (pm->show_pfn) { 1781 if (is_pfn_swap_entry(entry)) 1782 offset = swp_offset_pfn(entry) + idx; 1783 else 1784 offset = swp_offset(entry) + idx; 1785 frame = swp_type(entry) | 1786 (offset << MAX_SWAPFILES_SHIFT); 1787 } 1788 flags |= PM_SWAP; 1789 if (pmd_swp_soft_dirty(pmd)) 1790 flags |= PM_SOFT_DIRTY; 1791 if (pmd_swp_uffd_wp(pmd)) 1792 flags |= PM_UFFD_WP; 1793 VM_BUG_ON(!is_pmd_migration_entry(pmd)); 1794 page = pfn_swap_entry_to_page(entry); 1795 } 1796 #endif 1797 1798 if (page) { 1799 folio = page_folio(page); 1800 if (!folio_test_anon(folio)) 1801 flags |= PM_FILE; 1802 } 1803 1804 for (; addr != end; addr += PAGE_SIZE, idx++) { 1805 unsigned long cur_flags = flags; 1806 pagemap_entry_t pme; 1807 1808 if (folio && (flags & PM_PRESENT) && 1809 folio_precise_page_mapcount(folio, page + idx) == 1) 1810 cur_flags |= PM_MMAP_EXCLUSIVE; 1811 1812 pme = make_pme(frame, cur_flags); 1813 err = add_to_pagemap(&pme, pm); 1814 if (err) 1815 break; 1816 if (pm->show_pfn) { 1817 if (flags & PM_PRESENT) 1818 frame++; 1819 else if (flags & PM_SWAP) 1820 frame += (1 << MAX_SWAPFILES_SHIFT); 1821 } 1822 } 1823 spin_unlock(ptl); 1824 return err; 1825 } 1826 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ 1827 1828 /* 1829 * We can assume that @vma always points to a valid one and @end never 1830 * goes beyond vma->vm_end. 1831 */ 1832 orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl); 1833 if (!pte) { 1834 walk->action = ACTION_AGAIN; 1835 return err; 1836 } 1837 for (; addr < end; pte++, addr += PAGE_SIZE) { 1838 pagemap_entry_t pme; 1839 1840 pme = pte_to_pagemap_entry(pm, vma, addr, ptep_get(pte)); 1841 err = add_to_pagemap(&pme, pm); 1842 if (err) 1843 break; 1844 } 1845 pte_unmap_unlock(orig_pte, ptl); 1846 1847 cond_resched(); 1848 1849 return err; 1850 } 1851 1852 #ifdef CONFIG_HUGETLB_PAGE 1853 /* This function walks within one hugetlb entry in the single call */ 1854 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask, 1855 unsigned long addr, unsigned long end, 1856 struct mm_walk *walk) 1857 { 1858 struct pagemapread *pm = walk->private; 1859 struct vm_area_struct *vma = walk->vma; 1860 u64 flags = 0, frame = 0; 1861 int err = 0; 1862 pte_t pte; 1863 1864 if (vma->vm_flags & VM_SOFTDIRTY) 1865 flags |= PM_SOFT_DIRTY; 1866 1867 pte = huge_ptep_get(walk->mm, addr, ptep); 1868 if (pte_present(pte)) { 1869 struct folio *folio = page_folio(pte_page(pte)); 1870 1871 if (!folio_test_anon(folio)) 1872 flags |= PM_FILE; 1873 1874 if (!folio_likely_mapped_shared(folio) && 1875 !hugetlb_pmd_shared(ptep)) 1876 flags |= PM_MMAP_EXCLUSIVE; 1877 1878 if (huge_pte_uffd_wp(pte)) 1879 flags |= PM_UFFD_WP; 1880 1881 flags |= PM_PRESENT; 1882 if (pm->show_pfn) 1883 frame = pte_pfn(pte) + 1884 ((addr & ~hmask) >> PAGE_SHIFT); 1885 } else if (pte_swp_uffd_wp_any(pte)) { 1886 flags |= PM_UFFD_WP; 1887 } 1888 1889 for (; addr != end; addr += PAGE_SIZE) { 1890 pagemap_entry_t pme = make_pme(frame, flags); 1891 1892 err = add_to_pagemap(&pme, pm); 1893 if (err) 1894 return err; 1895 if (pm->show_pfn && (flags & PM_PRESENT)) 1896 frame++; 1897 } 1898 1899 cond_resched(); 1900 1901 return err; 1902 } 1903 #else 1904 #define pagemap_hugetlb_range NULL 1905 #endif /* HUGETLB_PAGE */ 1906 1907 static const struct mm_walk_ops pagemap_ops = { 1908 .pmd_entry = pagemap_pmd_range, 1909 .pte_hole = pagemap_pte_hole, 1910 .hugetlb_entry = pagemap_hugetlb_range, 1911 .walk_lock = PGWALK_RDLOCK, 1912 }; 1913 1914 /* 1915 * /proc/pid/pagemap - an array mapping virtual pages to pfns 1916 * 1917 * For each page in the address space, this file contains one 64-bit entry 1918 * consisting of the following: 1919 * 1920 * Bits 0-54 page frame number (PFN) if present 1921 * Bits 0-4 swap type if swapped 1922 * Bits 5-54 swap offset if swapped 1923 * Bit 55 pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst) 1924 * Bit 56 page exclusively mapped 1925 * Bit 57 pte is uffd-wp write-protected 1926 * Bits 58-60 zero 1927 * Bit 61 page is file-page or shared-anon 1928 * Bit 62 page swapped 1929 * Bit 63 page present 1930 * 1931 * If the page is not present but in swap, then the PFN contains an 1932 * encoding of the swap file number and the page's offset into the 1933 * swap. Unmapped pages return a null PFN. This allows determining 1934 * precisely which pages are mapped (or in swap) and comparing mapped 1935 * pages between processes. 1936 * 1937 * Efficient users of this interface will use /proc/pid/maps to 1938 * determine which areas of memory are actually mapped and llseek to 1939 * skip over unmapped regions. 1940 */ 1941 static ssize_t pagemap_read(struct file *file, char __user *buf, 1942 size_t count, loff_t *ppos) 1943 { 1944 struct mm_struct *mm = file->private_data; 1945 struct pagemapread pm; 1946 unsigned long src; 1947 unsigned long svpfn; 1948 unsigned long start_vaddr; 1949 unsigned long end_vaddr; 1950 int ret = 0, copied = 0; 1951 1952 if (!mm || !mmget_not_zero(mm)) 1953 goto out; 1954 1955 ret = -EINVAL; 1956 /* file position must be aligned */ 1957 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES)) 1958 goto out_mm; 1959 1960 ret = 0; 1961 if (!count) 1962 goto out_mm; 1963 1964 /* do not disclose physical addresses: attack vector */ 1965 pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN); 1966 1967 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT); 1968 pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL); 1969 ret = -ENOMEM; 1970 if (!pm.buffer) 1971 goto out_mm; 1972 1973 src = *ppos; 1974 svpfn = src / PM_ENTRY_BYTES; 1975 end_vaddr = mm->task_size; 1976 1977 /* watch out for wraparound */ 1978 start_vaddr = end_vaddr; 1979 if (svpfn <= (ULONG_MAX >> PAGE_SHIFT)) { 1980 unsigned long end; 1981 1982 ret = mmap_read_lock_killable(mm); 1983 if (ret) 1984 goto out_free; 1985 start_vaddr = untagged_addr_remote(mm, svpfn << PAGE_SHIFT); 1986 mmap_read_unlock(mm); 1987 1988 end = start_vaddr + ((count / PM_ENTRY_BYTES) << PAGE_SHIFT); 1989 if (end >= start_vaddr && end < mm->task_size) 1990 end_vaddr = end; 1991 } 1992 1993 /* Ensure the address is inside the task */ 1994 if (start_vaddr > mm->task_size) 1995 start_vaddr = end_vaddr; 1996 1997 ret = 0; 1998 while (count && (start_vaddr < end_vaddr)) { 1999 int len; 2000 unsigned long end; 2001 2002 pm.pos = 0; 2003 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK; 2004 /* overflow ? */ 2005 if (end < start_vaddr || end > end_vaddr) 2006 end = end_vaddr; 2007 ret = mmap_read_lock_killable(mm); 2008 if (ret) 2009 goto out_free; 2010 ret = walk_page_range(mm, start_vaddr, end, &pagemap_ops, &pm); 2011 mmap_read_unlock(mm); 2012 start_vaddr = end; 2013 2014 len = min(count, PM_ENTRY_BYTES * pm.pos); 2015 if (copy_to_user(buf, pm.buffer, len)) { 2016 ret = -EFAULT; 2017 goto out_free; 2018 } 2019 copied += len; 2020 buf += len; 2021 count -= len; 2022 } 2023 *ppos += copied; 2024 if (!ret || ret == PM_END_OF_BUFFER) 2025 ret = copied; 2026 2027 out_free: 2028 kfree(pm.buffer); 2029 out_mm: 2030 mmput(mm); 2031 out: 2032 return ret; 2033 } 2034 2035 static int pagemap_open(struct inode *inode, struct file *file) 2036 { 2037 struct mm_struct *mm; 2038 2039 mm = proc_mem_open(inode, PTRACE_MODE_READ); 2040 if (IS_ERR(mm)) 2041 return PTR_ERR(mm); 2042 file->private_data = mm; 2043 return 0; 2044 } 2045 2046 static int pagemap_release(struct inode *inode, struct file *file) 2047 { 2048 struct mm_struct *mm = file->private_data; 2049 2050 if (mm) 2051 mmdrop(mm); 2052 return 0; 2053 } 2054 2055 #define PM_SCAN_CATEGORIES (PAGE_IS_WPALLOWED | PAGE_IS_WRITTEN | \ 2056 PAGE_IS_FILE | PAGE_IS_PRESENT | \ 2057 PAGE_IS_SWAPPED | PAGE_IS_PFNZERO | \ 2058 PAGE_IS_HUGE | PAGE_IS_SOFT_DIRTY) 2059 #define PM_SCAN_FLAGS (PM_SCAN_WP_MATCHING | PM_SCAN_CHECK_WPASYNC) 2060 2061 struct pagemap_scan_private { 2062 struct pm_scan_arg arg; 2063 unsigned long masks_of_interest, cur_vma_category; 2064 struct page_region *vec_buf; 2065 unsigned long vec_buf_len, vec_buf_index, found_pages; 2066 struct page_region __user *vec_out; 2067 }; 2068 2069 static unsigned long pagemap_page_category(struct pagemap_scan_private *p, 2070 struct vm_area_struct *vma, 2071 unsigned long addr, pte_t pte) 2072 { 2073 unsigned long categories = 0; 2074 2075 if (pte_present(pte)) { 2076 struct page *page; 2077 2078 categories |= PAGE_IS_PRESENT; 2079 if (!pte_uffd_wp(pte)) 2080 categories |= PAGE_IS_WRITTEN; 2081 2082 if (p->masks_of_interest & PAGE_IS_FILE) { 2083 page = vm_normal_page(vma, addr, pte); 2084 if (page && !PageAnon(page)) 2085 categories |= PAGE_IS_FILE; 2086 } 2087 2088 if (is_zero_pfn(pte_pfn(pte))) 2089 categories |= PAGE_IS_PFNZERO; 2090 if (pte_soft_dirty(pte)) 2091 categories |= PAGE_IS_SOFT_DIRTY; 2092 } else if (is_swap_pte(pte)) { 2093 swp_entry_t swp; 2094 2095 categories |= PAGE_IS_SWAPPED; 2096 if (!pte_swp_uffd_wp_any(pte)) 2097 categories |= PAGE_IS_WRITTEN; 2098 2099 if (p->masks_of_interest & PAGE_IS_FILE) { 2100 swp = pte_to_swp_entry(pte); 2101 if (is_pfn_swap_entry(swp) && 2102 !folio_test_anon(pfn_swap_entry_folio(swp))) 2103 categories |= PAGE_IS_FILE; 2104 } 2105 if (pte_swp_soft_dirty(pte)) 2106 categories |= PAGE_IS_SOFT_DIRTY; 2107 } 2108 2109 return categories; 2110 } 2111 2112 static void make_uffd_wp_pte(struct vm_area_struct *vma, 2113 unsigned long addr, pte_t *pte, pte_t ptent) 2114 { 2115 if (pte_present(ptent)) { 2116 pte_t old_pte; 2117 2118 old_pte = ptep_modify_prot_start(vma, addr, pte); 2119 ptent = pte_mkuffd_wp(old_pte); 2120 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent); 2121 } else if (is_swap_pte(ptent)) { 2122 ptent = pte_swp_mkuffd_wp(ptent); 2123 set_pte_at(vma->vm_mm, addr, pte, ptent); 2124 } else { 2125 set_pte_at(vma->vm_mm, addr, pte, 2126 make_pte_marker(PTE_MARKER_UFFD_WP)); 2127 } 2128 } 2129 2130 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 2131 static unsigned long pagemap_thp_category(struct pagemap_scan_private *p, 2132 struct vm_area_struct *vma, 2133 unsigned long addr, pmd_t pmd) 2134 { 2135 unsigned long categories = PAGE_IS_HUGE; 2136 2137 if (pmd_present(pmd)) { 2138 struct page *page; 2139 2140 categories |= PAGE_IS_PRESENT; 2141 if (!pmd_uffd_wp(pmd)) 2142 categories |= PAGE_IS_WRITTEN; 2143 2144 if (p->masks_of_interest & PAGE_IS_FILE) { 2145 page = vm_normal_page_pmd(vma, addr, pmd); 2146 if (page && !PageAnon(page)) 2147 categories |= PAGE_IS_FILE; 2148 } 2149 2150 if (is_zero_pfn(pmd_pfn(pmd))) 2151 categories |= PAGE_IS_PFNZERO; 2152 if (pmd_soft_dirty(pmd)) 2153 categories |= PAGE_IS_SOFT_DIRTY; 2154 } else if (is_swap_pmd(pmd)) { 2155 swp_entry_t swp; 2156 2157 categories |= PAGE_IS_SWAPPED; 2158 if (!pmd_swp_uffd_wp(pmd)) 2159 categories |= PAGE_IS_WRITTEN; 2160 if (pmd_swp_soft_dirty(pmd)) 2161 categories |= PAGE_IS_SOFT_DIRTY; 2162 2163 if (p->masks_of_interest & PAGE_IS_FILE) { 2164 swp = pmd_to_swp_entry(pmd); 2165 if (is_pfn_swap_entry(swp) && 2166 !folio_test_anon(pfn_swap_entry_folio(swp))) 2167 categories |= PAGE_IS_FILE; 2168 } 2169 } 2170 2171 return categories; 2172 } 2173 2174 static void make_uffd_wp_pmd(struct vm_area_struct *vma, 2175 unsigned long addr, pmd_t *pmdp) 2176 { 2177 pmd_t old, pmd = *pmdp; 2178 2179 if (pmd_present(pmd)) { 2180 old = pmdp_invalidate_ad(vma, addr, pmdp); 2181 pmd = pmd_mkuffd_wp(old); 2182 set_pmd_at(vma->vm_mm, addr, pmdp, pmd); 2183 } else if (is_migration_entry(pmd_to_swp_entry(pmd))) { 2184 pmd = pmd_swp_mkuffd_wp(pmd); 2185 set_pmd_at(vma->vm_mm, addr, pmdp, pmd); 2186 } 2187 } 2188 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ 2189 2190 #ifdef CONFIG_HUGETLB_PAGE 2191 static unsigned long pagemap_hugetlb_category(pte_t pte) 2192 { 2193 unsigned long categories = PAGE_IS_HUGE; 2194 2195 /* 2196 * According to pagemap_hugetlb_range(), file-backed HugeTLB 2197 * page cannot be swapped. So PAGE_IS_FILE is not checked for 2198 * swapped pages. 2199 */ 2200 if (pte_present(pte)) { 2201 categories |= PAGE_IS_PRESENT; 2202 if (!huge_pte_uffd_wp(pte)) 2203 categories |= PAGE_IS_WRITTEN; 2204 if (!PageAnon(pte_page(pte))) 2205 categories |= PAGE_IS_FILE; 2206 if (is_zero_pfn(pte_pfn(pte))) 2207 categories |= PAGE_IS_PFNZERO; 2208 if (pte_soft_dirty(pte)) 2209 categories |= PAGE_IS_SOFT_DIRTY; 2210 } else if (is_swap_pte(pte)) { 2211 categories |= PAGE_IS_SWAPPED; 2212 if (!pte_swp_uffd_wp_any(pte)) 2213 categories |= PAGE_IS_WRITTEN; 2214 if (pte_swp_soft_dirty(pte)) 2215 categories |= PAGE_IS_SOFT_DIRTY; 2216 } 2217 2218 return categories; 2219 } 2220 2221 static void make_uffd_wp_huge_pte(struct vm_area_struct *vma, 2222 unsigned long addr, pte_t *ptep, 2223 pte_t ptent) 2224 { 2225 unsigned long psize; 2226 2227 if (is_hugetlb_entry_hwpoisoned(ptent) || is_pte_marker(ptent)) 2228 return; 2229 2230 psize = huge_page_size(hstate_vma(vma)); 2231 2232 if (is_hugetlb_entry_migration(ptent)) 2233 set_huge_pte_at(vma->vm_mm, addr, ptep, 2234 pte_swp_mkuffd_wp(ptent), psize); 2235 else if (!huge_pte_none(ptent)) 2236 huge_ptep_modify_prot_commit(vma, addr, ptep, ptent, 2237 huge_pte_mkuffd_wp(ptent)); 2238 else 2239 set_huge_pte_at(vma->vm_mm, addr, ptep, 2240 make_pte_marker(PTE_MARKER_UFFD_WP), psize); 2241 } 2242 #endif /* CONFIG_HUGETLB_PAGE */ 2243 2244 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLB_PAGE) 2245 static void pagemap_scan_backout_range(struct pagemap_scan_private *p, 2246 unsigned long addr, unsigned long end) 2247 { 2248 struct page_region *cur_buf = &p->vec_buf[p->vec_buf_index]; 2249 2250 if (cur_buf->start != addr) 2251 cur_buf->end = addr; 2252 else 2253 cur_buf->start = cur_buf->end = 0; 2254 2255 p->found_pages -= (end - addr) / PAGE_SIZE; 2256 } 2257 #endif 2258 2259 static bool pagemap_scan_is_interesting_page(unsigned long categories, 2260 const struct pagemap_scan_private *p) 2261 { 2262 categories ^= p->arg.category_inverted; 2263 if ((categories & p->arg.category_mask) != p->arg.category_mask) 2264 return false; 2265 if (p->arg.category_anyof_mask && !(categories & p->arg.category_anyof_mask)) 2266 return false; 2267 2268 return true; 2269 } 2270 2271 static bool pagemap_scan_is_interesting_vma(unsigned long categories, 2272 const struct pagemap_scan_private *p) 2273 { 2274 unsigned long required = p->arg.category_mask & PAGE_IS_WPALLOWED; 2275 2276 categories ^= p->arg.category_inverted; 2277 if ((categories & required) != required) 2278 return false; 2279 2280 return true; 2281 } 2282 2283 static int pagemap_scan_test_walk(unsigned long start, unsigned long end, 2284 struct mm_walk *walk) 2285 { 2286 struct pagemap_scan_private *p = walk->private; 2287 struct vm_area_struct *vma = walk->vma; 2288 unsigned long vma_category = 0; 2289 bool wp_allowed = userfaultfd_wp_async(vma) && 2290 userfaultfd_wp_use_markers(vma); 2291 2292 if (!wp_allowed) { 2293 /* User requested explicit failure over wp-async capability */ 2294 if (p->arg.flags & PM_SCAN_CHECK_WPASYNC) 2295 return -EPERM; 2296 /* 2297 * User requires wr-protect, and allows silently skipping 2298 * unsupported vmas. 2299 */ 2300 if (p->arg.flags & PM_SCAN_WP_MATCHING) 2301 return 1; 2302 /* 2303 * Then the request doesn't involve wr-protects at all, 2304 * fall through to the rest checks, and allow vma walk. 2305 */ 2306 } 2307 2308 if (vma->vm_flags & VM_PFNMAP) 2309 return 1; 2310 2311 if (wp_allowed) 2312 vma_category |= PAGE_IS_WPALLOWED; 2313 2314 if (vma->vm_flags & VM_SOFTDIRTY) 2315 vma_category |= PAGE_IS_SOFT_DIRTY; 2316 2317 if (!pagemap_scan_is_interesting_vma(vma_category, p)) 2318 return 1; 2319 2320 p->cur_vma_category = vma_category; 2321 2322 return 0; 2323 } 2324 2325 static bool pagemap_scan_push_range(unsigned long categories, 2326 struct pagemap_scan_private *p, 2327 unsigned long addr, unsigned long end) 2328 { 2329 struct page_region *cur_buf = &p->vec_buf[p->vec_buf_index]; 2330 2331 /* 2332 * When there is no output buffer provided at all, the sentinel values 2333 * won't match here. There is no other way for `cur_buf->end` to be 2334 * non-zero other than it being non-empty. 2335 */ 2336 if (addr == cur_buf->end && categories == cur_buf->categories) { 2337 cur_buf->end = end; 2338 return true; 2339 } 2340 2341 if (cur_buf->end) { 2342 if (p->vec_buf_index >= p->vec_buf_len - 1) 2343 return false; 2344 2345 cur_buf = &p->vec_buf[++p->vec_buf_index]; 2346 } 2347 2348 cur_buf->start = addr; 2349 cur_buf->end = end; 2350 cur_buf->categories = categories; 2351 2352 return true; 2353 } 2354 2355 static int pagemap_scan_output(unsigned long categories, 2356 struct pagemap_scan_private *p, 2357 unsigned long addr, unsigned long *end) 2358 { 2359 unsigned long n_pages, total_pages; 2360 int ret = 0; 2361 2362 if (!p->vec_buf) 2363 return 0; 2364 2365 categories &= p->arg.return_mask; 2366 2367 n_pages = (*end - addr) / PAGE_SIZE; 2368 if (check_add_overflow(p->found_pages, n_pages, &total_pages) || 2369 total_pages > p->arg.max_pages) { 2370 size_t n_too_much = total_pages - p->arg.max_pages; 2371 *end -= n_too_much * PAGE_SIZE; 2372 n_pages -= n_too_much; 2373 ret = -ENOSPC; 2374 } 2375 2376 if (!pagemap_scan_push_range(categories, p, addr, *end)) { 2377 *end = addr; 2378 n_pages = 0; 2379 ret = -ENOSPC; 2380 } 2381 2382 p->found_pages += n_pages; 2383 if (ret) 2384 p->arg.walk_end = *end; 2385 2386 return ret; 2387 } 2388 2389 static int pagemap_scan_thp_entry(pmd_t *pmd, unsigned long start, 2390 unsigned long end, struct mm_walk *walk) 2391 { 2392 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 2393 struct pagemap_scan_private *p = walk->private; 2394 struct vm_area_struct *vma = walk->vma; 2395 unsigned long categories; 2396 spinlock_t *ptl; 2397 int ret = 0; 2398 2399 ptl = pmd_trans_huge_lock(pmd, vma); 2400 if (!ptl) 2401 return -ENOENT; 2402 2403 categories = p->cur_vma_category | 2404 pagemap_thp_category(p, vma, start, *pmd); 2405 2406 if (!pagemap_scan_is_interesting_page(categories, p)) 2407 goto out_unlock; 2408 2409 ret = pagemap_scan_output(categories, p, start, &end); 2410 if (start == end) 2411 goto out_unlock; 2412 2413 if (~p->arg.flags & PM_SCAN_WP_MATCHING) 2414 goto out_unlock; 2415 if (~categories & PAGE_IS_WRITTEN) 2416 goto out_unlock; 2417 2418 /* 2419 * Break huge page into small pages if the WP operation 2420 * needs to be performed on a portion of the huge page. 2421 */ 2422 if (end != start + HPAGE_SIZE) { 2423 spin_unlock(ptl); 2424 split_huge_pmd(vma, pmd, start); 2425 pagemap_scan_backout_range(p, start, end); 2426 /* Report as if there was no THP */ 2427 return -ENOENT; 2428 } 2429 2430 make_uffd_wp_pmd(vma, start, pmd); 2431 flush_tlb_range(vma, start, end); 2432 out_unlock: 2433 spin_unlock(ptl); 2434 return ret; 2435 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */ 2436 return -ENOENT; 2437 #endif 2438 } 2439 2440 static int pagemap_scan_pmd_entry(pmd_t *pmd, unsigned long start, 2441 unsigned long end, struct mm_walk *walk) 2442 { 2443 struct pagemap_scan_private *p = walk->private; 2444 struct vm_area_struct *vma = walk->vma; 2445 unsigned long addr, flush_end = 0; 2446 pte_t *pte, *start_pte; 2447 spinlock_t *ptl; 2448 int ret; 2449 2450 arch_enter_lazy_mmu_mode(); 2451 2452 ret = pagemap_scan_thp_entry(pmd, start, end, walk); 2453 if (ret != -ENOENT) { 2454 arch_leave_lazy_mmu_mode(); 2455 return ret; 2456 } 2457 2458 ret = 0; 2459 start_pte = pte = pte_offset_map_lock(vma->vm_mm, pmd, start, &ptl); 2460 if (!pte) { 2461 arch_leave_lazy_mmu_mode(); 2462 walk->action = ACTION_AGAIN; 2463 return 0; 2464 } 2465 2466 if ((p->arg.flags & PM_SCAN_WP_MATCHING) && !p->vec_out) { 2467 /* Fast path for performing exclusive WP */ 2468 for (addr = start; addr != end; pte++, addr += PAGE_SIZE) { 2469 pte_t ptent = ptep_get(pte); 2470 2471 if ((pte_present(ptent) && pte_uffd_wp(ptent)) || 2472 pte_swp_uffd_wp_any(ptent)) 2473 continue; 2474 make_uffd_wp_pte(vma, addr, pte, ptent); 2475 if (!flush_end) 2476 start = addr; 2477 flush_end = addr + PAGE_SIZE; 2478 } 2479 goto flush_and_return; 2480 } 2481 2482 if (!p->arg.category_anyof_mask && !p->arg.category_inverted && 2483 p->arg.category_mask == PAGE_IS_WRITTEN && 2484 p->arg.return_mask == PAGE_IS_WRITTEN) { 2485 for (addr = start; addr < end; pte++, addr += PAGE_SIZE) { 2486 unsigned long next = addr + PAGE_SIZE; 2487 pte_t ptent = ptep_get(pte); 2488 2489 if ((pte_present(ptent) && pte_uffd_wp(ptent)) || 2490 pte_swp_uffd_wp_any(ptent)) 2491 continue; 2492 ret = pagemap_scan_output(p->cur_vma_category | PAGE_IS_WRITTEN, 2493 p, addr, &next); 2494 if (next == addr) 2495 break; 2496 if (~p->arg.flags & PM_SCAN_WP_MATCHING) 2497 continue; 2498 make_uffd_wp_pte(vma, addr, pte, ptent); 2499 if (!flush_end) 2500 start = addr; 2501 flush_end = next; 2502 } 2503 goto flush_and_return; 2504 } 2505 2506 for (addr = start; addr != end; pte++, addr += PAGE_SIZE) { 2507 pte_t ptent = ptep_get(pte); 2508 unsigned long categories = p->cur_vma_category | 2509 pagemap_page_category(p, vma, addr, ptent); 2510 unsigned long next = addr + PAGE_SIZE; 2511 2512 if (!pagemap_scan_is_interesting_page(categories, p)) 2513 continue; 2514 2515 ret = pagemap_scan_output(categories, p, addr, &next); 2516 if (next == addr) 2517 break; 2518 2519 if (~p->arg.flags & PM_SCAN_WP_MATCHING) 2520 continue; 2521 if (~categories & PAGE_IS_WRITTEN) 2522 continue; 2523 2524 make_uffd_wp_pte(vma, addr, pte, ptent); 2525 if (!flush_end) 2526 start = addr; 2527 flush_end = next; 2528 } 2529 2530 flush_and_return: 2531 if (flush_end) 2532 flush_tlb_range(vma, start, addr); 2533 2534 pte_unmap_unlock(start_pte, ptl); 2535 arch_leave_lazy_mmu_mode(); 2536 2537 cond_resched(); 2538 return ret; 2539 } 2540 2541 #ifdef CONFIG_HUGETLB_PAGE 2542 static int pagemap_scan_hugetlb_entry(pte_t *ptep, unsigned long hmask, 2543 unsigned long start, unsigned long end, 2544 struct mm_walk *walk) 2545 { 2546 struct pagemap_scan_private *p = walk->private; 2547 struct vm_area_struct *vma = walk->vma; 2548 unsigned long categories; 2549 spinlock_t *ptl; 2550 int ret = 0; 2551 pte_t pte; 2552 2553 if (~p->arg.flags & PM_SCAN_WP_MATCHING) { 2554 /* Go the short route when not write-protecting pages. */ 2555 2556 pte = huge_ptep_get(walk->mm, start, ptep); 2557 categories = p->cur_vma_category | pagemap_hugetlb_category(pte); 2558 2559 if (!pagemap_scan_is_interesting_page(categories, p)) 2560 return 0; 2561 2562 return pagemap_scan_output(categories, p, start, &end); 2563 } 2564 2565 i_mmap_lock_write(vma->vm_file->f_mapping); 2566 ptl = huge_pte_lock(hstate_vma(vma), vma->vm_mm, ptep); 2567 2568 pte = huge_ptep_get(walk->mm, start, ptep); 2569 categories = p->cur_vma_category | pagemap_hugetlb_category(pte); 2570 2571 if (!pagemap_scan_is_interesting_page(categories, p)) 2572 goto out_unlock; 2573 2574 ret = pagemap_scan_output(categories, p, start, &end); 2575 if (start == end) 2576 goto out_unlock; 2577 2578 if (~categories & PAGE_IS_WRITTEN) 2579 goto out_unlock; 2580 2581 if (end != start + HPAGE_SIZE) { 2582 /* Partial HugeTLB page WP isn't possible. */ 2583 pagemap_scan_backout_range(p, start, end); 2584 p->arg.walk_end = start; 2585 ret = 0; 2586 goto out_unlock; 2587 } 2588 2589 make_uffd_wp_huge_pte(vma, start, ptep, pte); 2590 flush_hugetlb_tlb_range(vma, start, end); 2591 2592 out_unlock: 2593 spin_unlock(ptl); 2594 i_mmap_unlock_write(vma->vm_file->f_mapping); 2595 2596 return ret; 2597 } 2598 #else 2599 #define pagemap_scan_hugetlb_entry NULL 2600 #endif 2601 2602 static int pagemap_scan_pte_hole(unsigned long addr, unsigned long end, 2603 int depth, struct mm_walk *walk) 2604 { 2605 struct pagemap_scan_private *p = walk->private; 2606 struct vm_area_struct *vma = walk->vma; 2607 int ret, err; 2608 2609 if (!vma || !pagemap_scan_is_interesting_page(p->cur_vma_category, p)) 2610 return 0; 2611 2612 ret = pagemap_scan_output(p->cur_vma_category, p, addr, &end); 2613 if (addr == end) 2614 return ret; 2615 2616 if (~p->arg.flags & PM_SCAN_WP_MATCHING) 2617 return ret; 2618 2619 err = uffd_wp_range(vma, addr, end - addr, true); 2620 if (err < 0) 2621 ret = err; 2622 2623 return ret; 2624 } 2625 2626 static const struct mm_walk_ops pagemap_scan_ops = { 2627 .test_walk = pagemap_scan_test_walk, 2628 .pmd_entry = pagemap_scan_pmd_entry, 2629 .pte_hole = pagemap_scan_pte_hole, 2630 .hugetlb_entry = pagemap_scan_hugetlb_entry, 2631 }; 2632 2633 static int pagemap_scan_get_args(struct pm_scan_arg *arg, 2634 unsigned long uarg) 2635 { 2636 if (copy_from_user(arg, (void __user *)uarg, sizeof(*arg))) 2637 return -EFAULT; 2638 2639 if (arg->size != sizeof(struct pm_scan_arg)) 2640 return -EINVAL; 2641 2642 /* Validate requested features */ 2643 if (arg->flags & ~PM_SCAN_FLAGS) 2644 return -EINVAL; 2645 if ((arg->category_inverted | arg->category_mask | 2646 arg->category_anyof_mask | arg->return_mask) & ~PM_SCAN_CATEGORIES) 2647 return -EINVAL; 2648 2649 arg->start = untagged_addr((unsigned long)arg->start); 2650 arg->end = untagged_addr((unsigned long)arg->end); 2651 arg->vec = untagged_addr((unsigned long)arg->vec); 2652 2653 /* Validate memory pointers */ 2654 if (!IS_ALIGNED(arg->start, PAGE_SIZE)) 2655 return -EINVAL; 2656 if (!access_ok((void __user *)(long)arg->start, arg->end - arg->start)) 2657 return -EFAULT; 2658 if (!arg->vec && arg->vec_len) 2659 return -EINVAL; 2660 if (arg->vec && !access_ok((void __user *)(long)arg->vec, 2661 arg->vec_len * sizeof(struct page_region))) 2662 return -EFAULT; 2663 2664 /* Fixup default values */ 2665 arg->end = ALIGN(arg->end, PAGE_SIZE); 2666 arg->walk_end = 0; 2667 if (!arg->max_pages) 2668 arg->max_pages = ULONG_MAX; 2669 2670 return 0; 2671 } 2672 2673 static int pagemap_scan_writeback_args(struct pm_scan_arg *arg, 2674 unsigned long uargl) 2675 { 2676 struct pm_scan_arg __user *uarg = (void __user *)uargl; 2677 2678 if (copy_to_user(&uarg->walk_end, &arg->walk_end, sizeof(arg->walk_end))) 2679 return -EFAULT; 2680 2681 return 0; 2682 } 2683 2684 static int pagemap_scan_init_bounce_buffer(struct pagemap_scan_private *p) 2685 { 2686 if (!p->arg.vec_len) 2687 return 0; 2688 2689 p->vec_buf_len = min_t(size_t, PAGEMAP_WALK_SIZE >> PAGE_SHIFT, 2690 p->arg.vec_len); 2691 p->vec_buf = kmalloc_array(p->vec_buf_len, sizeof(*p->vec_buf), 2692 GFP_KERNEL); 2693 if (!p->vec_buf) 2694 return -ENOMEM; 2695 2696 p->vec_buf->start = p->vec_buf->end = 0; 2697 p->vec_out = (struct page_region __user *)(long)p->arg.vec; 2698 2699 return 0; 2700 } 2701 2702 static long pagemap_scan_flush_buffer(struct pagemap_scan_private *p) 2703 { 2704 const struct page_region *buf = p->vec_buf; 2705 long n = p->vec_buf_index; 2706 2707 if (!p->vec_buf) 2708 return 0; 2709 2710 if (buf[n].end != buf[n].start) 2711 n++; 2712 2713 if (!n) 2714 return 0; 2715 2716 if (copy_to_user(p->vec_out, buf, n * sizeof(*buf))) 2717 return -EFAULT; 2718 2719 p->arg.vec_len -= n; 2720 p->vec_out += n; 2721 2722 p->vec_buf_index = 0; 2723 p->vec_buf_len = min_t(size_t, p->vec_buf_len, p->arg.vec_len); 2724 p->vec_buf->start = p->vec_buf->end = 0; 2725 2726 return n; 2727 } 2728 2729 static long do_pagemap_scan(struct mm_struct *mm, unsigned long uarg) 2730 { 2731 struct pagemap_scan_private p = {0}; 2732 unsigned long walk_start; 2733 size_t n_ranges_out = 0; 2734 int ret; 2735 2736 ret = pagemap_scan_get_args(&p.arg, uarg); 2737 if (ret) 2738 return ret; 2739 2740 p.masks_of_interest = p.arg.category_mask | p.arg.category_anyof_mask | 2741 p.arg.return_mask; 2742 ret = pagemap_scan_init_bounce_buffer(&p); 2743 if (ret) 2744 return ret; 2745 2746 for (walk_start = p.arg.start; walk_start < p.arg.end; 2747 walk_start = p.arg.walk_end) { 2748 struct mmu_notifier_range range; 2749 long n_out; 2750 2751 if (fatal_signal_pending(current)) { 2752 ret = -EINTR; 2753 break; 2754 } 2755 2756 ret = mmap_read_lock_killable(mm); 2757 if (ret) 2758 break; 2759 2760 /* Protection change for the range is going to happen. */ 2761 if (p.arg.flags & PM_SCAN_WP_MATCHING) { 2762 mmu_notifier_range_init(&range, MMU_NOTIFY_PROTECTION_VMA, 0, 2763 mm, walk_start, p.arg.end); 2764 mmu_notifier_invalidate_range_start(&range); 2765 } 2766 2767 ret = walk_page_range(mm, walk_start, p.arg.end, 2768 &pagemap_scan_ops, &p); 2769 2770 if (p.arg.flags & PM_SCAN_WP_MATCHING) 2771 mmu_notifier_invalidate_range_end(&range); 2772 2773 mmap_read_unlock(mm); 2774 2775 n_out = pagemap_scan_flush_buffer(&p); 2776 if (n_out < 0) 2777 ret = n_out; 2778 else 2779 n_ranges_out += n_out; 2780 2781 if (ret != -ENOSPC) 2782 break; 2783 2784 if (p.arg.vec_len == 0 || p.found_pages == p.arg.max_pages) 2785 break; 2786 } 2787 2788 /* ENOSPC signifies early stop (buffer full) from the walk. */ 2789 if (!ret || ret == -ENOSPC) 2790 ret = n_ranges_out; 2791 2792 /* The walk_end isn't set when ret is zero */ 2793 if (!p.arg.walk_end) 2794 p.arg.walk_end = p.arg.end; 2795 if (pagemap_scan_writeback_args(&p.arg, uarg)) 2796 ret = -EFAULT; 2797 2798 kfree(p.vec_buf); 2799 return ret; 2800 } 2801 2802 static long do_pagemap_cmd(struct file *file, unsigned int cmd, 2803 unsigned long arg) 2804 { 2805 struct mm_struct *mm = file->private_data; 2806 2807 switch (cmd) { 2808 case PAGEMAP_SCAN: 2809 return do_pagemap_scan(mm, arg); 2810 2811 default: 2812 return -EINVAL; 2813 } 2814 } 2815 2816 const struct file_operations proc_pagemap_operations = { 2817 .llseek = mem_lseek, /* borrow this */ 2818 .read = pagemap_read, 2819 .open = pagemap_open, 2820 .release = pagemap_release, 2821 .unlocked_ioctl = do_pagemap_cmd, 2822 .compat_ioctl = do_pagemap_cmd, 2823 }; 2824 #endif /* CONFIG_PROC_PAGE_MONITOR */ 2825 2826 #ifdef CONFIG_NUMA 2827 2828 struct numa_maps { 2829 unsigned long pages; 2830 unsigned long anon; 2831 unsigned long active; 2832 unsigned long writeback; 2833 unsigned long mapcount_max; 2834 unsigned long dirty; 2835 unsigned long swapcache; 2836 unsigned long node[MAX_NUMNODES]; 2837 }; 2838 2839 struct numa_maps_private { 2840 struct proc_maps_private proc_maps; 2841 struct numa_maps md; 2842 }; 2843 2844 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty, 2845 unsigned long nr_pages) 2846 { 2847 struct folio *folio = page_folio(page); 2848 int count = folio_precise_page_mapcount(folio, page); 2849 2850 md->pages += nr_pages; 2851 if (pte_dirty || folio_test_dirty(folio)) 2852 md->dirty += nr_pages; 2853 2854 if (folio_test_swapcache(folio)) 2855 md->swapcache += nr_pages; 2856 2857 if (folio_test_active(folio) || folio_test_unevictable(folio)) 2858 md->active += nr_pages; 2859 2860 if (folio_test_writeback(folio)) 2861 md->writeback += nr_pages; 2862 2863 if (folio_test_anon(folio)) 2864 md->anon += nr_pages; 2865 2866 if (count > md->mapcount_max) 2867 md->mapcount_max = count; 2868 2869 md->node[folio_nid(folio)] += nr_pages; 2870 } 2871 2872 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma, 2873 unsigned long addr) 2874 { 2875 struct page *page; 2876 int nid; 2877 2878 if (!pte_present(pte)) 2879 return NULL; 2880 2881 page = vm_normal_page(vma, addr, pte); 2882 if (!page || is_zone_device_page(page)) 2883 return NULL; 2884 2885 if (PageReserved(page)) 2886 return NULL; 2887 2888 nid = page_to_nid(page); 2889 if (!node_isset(nid, node_states[N_MEMORY])) 2890 return NULL; 2891 2892 return page; 2893 } 2894 2895 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 2896 static struct page *can_gather_numa_stats_pmd(pmd_t pmd, 2897 struct vm_area_struct *vma, 2898 unsigned long addr) 2899 { 2900 struct page *page; 2901 int nid; 2902 2903 if (!pmd_present(pmd)) 2904 return NULL; 2905 2906 page = vm_normal_page_pmd(vma, addr, pmd); 2907 if (!page) 2908 return NULL; 2909 2910 if (PageReserved(page)) 2911 return NULL; 2912 2913 nid = page_to_nid(page); 2914 if (!node_isset(nid, node_states[N_MEMORY])) 2915 return NULL; 2916 2917 return page; 2918 } 2919 #endif 2920 2921 static int gather_pte_stats(pmd_t *pmd, unsigned long addr, 2922 unsigned long end, struct mm_walk *walk) 2923 { 2924 struct numa_maps *md = walk->private; 2925 struct vm_area_struct *vma = walk->vma; 2926 spinlock_t *ptl; 2927 pte_t *orig_pte; 2928 pte_t *pte; 2929 2930 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 2931 ptl = pmd_trans_huge_lock(pmd, vma); 2932 if (ptl) { 2933 struct page *page; 2934 2935 page = can_gather_numa_stats_pmd(*pmd, vma, addr); 2936 if (page) 2937 gather_stats(page, md, pmd_dirty(*pmd), 2938 HPAGE_PMD_SIZE/PAGE_SIZE); 2939 spin_unlock(ptl); 2940 return 0; 2941 } 2942 #endif 2943 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl); 2944 if (!pte) { 2945 walk->action = ACTION_AGAIN; 2946 return 0; 2947 } 2948 do { 2949 pte_t ptent = ptep_get(pte); 2950 struct page *page = can_gather_numa_stats(ptent, vma, addr); 2951 if (!page) 2952 continue; 2953 gather_stats(page, md, pte_dirty(ptent), 1); 2954 2955 } while (pte++, addr += PAGE_SIZE, addr != end); 2956 pte_unmap_unlock(orig_pte, ptl); 2957 cond_resched(); 2958 return 0; 2959 } 2960 #ifdef CONFIG_HUGETLB_PAGE 2961 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask, 2962 unsigned long addr, unsigned long end, struct mm_walk *walk) 2963 { 2964 pte_t huge_pte = huge_ptep_get(walk->mm, addr, pte); 2965 struct numa_maps *md; 2966 struct page *page; 2967 2968 if (!pte_present(huge_pte)) 2969 return 0; 2970 2971 page = pte_page(huge_pte); 2972 2973 md = walk->private; 2974 gather_stats(page, md, pte_dirty(huge_pte), 1); 2975 return 0; 2976 } 2977 2978 #else 2979 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask, 2980 unsigned long addr, unsigned long end, struct mm_walk *walk) 2981 { 2982 return 0; 2983 } 2984 #endif 2985 2986 static const struct mm_walk_ops show_numa_ops = { 2987 .hugetlb_entry = gather_hugetlb_stats, 2988 .pmd_entry = gather_pte_stats, 2989 .walk_lock = PGWALK_RDLOCK, 2990 }; 2991 2992 /* 2993 * Display pages allocated per node and memory policy via /proc. 2994 */ 2995 static int show_numa_map(struct seq_file *m, void *v) 2996 { 2997 struct numa_maps_private *numa_priv = m->private; 2998 struct proc_maps_private *proc_priv = &numa_priv->proc_maps; 2999 struct vm_area_struct *vma = v; 3000 struct numa_maps *md = &numa_priv->md; 3001 struct file *file = vma->vm_file; 3002 struct mm_struct *mm = vma->vm_mm; 3003 char buffer[64]; 3004 struct mempolicy *pol; 3005 pgoff_t ilx; 3006 int nid; 3007 3008 if (!mm) 3009 return 0; 3010 3011 /* Ensure we start with an empty set of numa_maps statistics. */ 3012 memset(md, 0, sizeof(*md)); 3013 3014 pol = __get_vma_policy(vma, vma->vm_start, &ilx); 3015 if (pol) { 3016 mpol_to_str(buffer, sizeof(buffer), pol); 3017 mpol_cond_put(pol); 3018 } else { 3019 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy); 3020 } 3021 3022 seq_printf(m, "%08lx %s", vma->vm_start, buffer); 3023 3024 if (file) { 3025 seq_puts(m, " file="); 3026 seq_path(m, file_user_path(file), "\n\t= "); 3027 } else if (vma_is_initial_heap(vma)) { 3028 seq_puts(m, " heap"); 3029 } else if (vma_is_initial_stack(vma)) { 3030 seq_puts(m, " stack"); 3031 } 3032 3033 if (is_vm_hugetlb_page(vma)) 3034 seq_puts(m, " huge"); 3035 3036 /* mmap_lock is held by m_start */ 3037 walk_page_vma(vma, &show_numa_ops, md); 3038 3039 if (!md->pages) 3040 goto out; 3041 3042 if (md->anon) 3043 seq_printf(m, " anon=%lu", md->anon); 3044 3045 if (md->dirty) 3046 seq_printf(m, " dirty=%lu", md->dirty); 3047 3048 if (md->pages != md->anon && md->pages != md->dirty) 3049 seq_printf(m, " mapped=%lu", md->pages); 3050 3051 if (md->mapcount_max > 1) 3052 seq_printf(m, " mapmax=%lu", md->mapcount_max); 3053 3054 if (md->swapcache) 3055 seq_printf(m, " swapcache=%lu", md->swapcache); 3056 3057 if (md->active < md->pages && !is_vm_hugetlb_page(vma)) 3058 seq_printf(m, " active=%lu", md->active); 3059 3060 if (md->writeback) 3061 seq_printf(m, " writeback=%lu", md->writeback); 3062 3063 for_each_node_state(nid, N_MEMORY) 3064 if (md->node[nid]) 3065 seq_printf(m, " N%d=%lu", nid, md->node[nid]); 3066 3067 seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10); 3068 out: 3069 seq_putc(m, '\n'); 3070 return 0; 3071 } 3072 3073 static const struct seq_operations proc_pid_numa_maps_op = { 3074 .start = m_start, 3075 .next = m_next, 3076 .stop = m_stop, 3077 .show = show_numa_map, 3078 }; 3079 3080 static int pid_numa_maps_open(struct inode *inode, struct file *file) 3081 { 3082 return proc_maps_open(inode, file, &proc_pid_numa_maps_op, 3083 sizeof(struct numa_maps_private)); 3084 } 3085 3086 const struct file_operations proc_pid_numa_maps_operations = { 3087 .open = pid_numa_maps_open, 3088 .read = seq_read, 3089 .llseek = seq_lseek, 3090 .release = proc_map_release, 3091 }; 3092 3093 #endif /* CONFIG_NUMA */ 3094