fs/proc/task_mmu.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include <linux/pagewalk.h>
   3 #include <linux/vmacache.h>
   4 #include <linux/hugetlb.h>
   5 #include <linux/huge_mm.h>
   6 #include <linux/mount.h>
   7 #include <linux/seq_file.h>
   8 #include <linux/highmem.h>
   9 #include <linux/ptrace.h>
  10 #include <linux/slab.h>
  11 #include <linux/pagemap.h>
  12 #include <linux/mempolicy.h>
  13 #include <linux/rmap.h>
  14 #include <linux/swap.h>
  15 #include <linux/sched/mm.h>
  16 #include <linux/swapops.h>
  17 #include <linux/mmu_notifier.h>
  18 #include <linux/page_idle.h>
  19 #include <linux/shmem_fs.h>
  20 #include <linux/uaccess.h>
  21 #include <linux/pkeys.h>
  22
  23 #include <asm/elf.h>
  24 #include <asm/tlb.h>
  25 #include <asm/tlbflush.h>
  26 #include "internal.h"
  27
  28 #define SEQ_PUT_DEC(str, val) \
  29                 seq_put_decimal_ull_width(m, str, (val) << (PAGE_SHIFT-10), 8)
  30 void task_mem(struct seq_file *m, struct mm_struct *mm)
  31 {
  32         unsigned long text, lib, swap, anon, file, shmem;
  33         unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
  34
  35         anon = get_mm_counter(mm, MM_ANONPAGES);
  36         file = get_mm_counter(mm, MM_FILEPAGES);
  37         shmem = get_mm_counter(mm, MM_SHMEMPAGES);
  38
  39         /*
  40          * Note: to minimize their overhead, mm maintains hiwater_vm and
  41          * hiwater_rss only when about to *lower* total_vm or rss.  Any
  42          * collector of these hiwater stats must therefore get total_vm
  43          * and rss too, which will usually be the higher.  Barriers? not
  44          * worth the effort, such snapshots can always be inconsistent.
  45          */
  46         hiwater_vm = total_vm = mm->total_vm;
  47         if (hiwater_vm < mm->hiwater_vm)
  48                 hiwater_vm = mm->hiwater_vm;
  49         hiwater_rss = total_rss = anon + file + shmem;
  50         if (hiwater_rss < mm->hiwater_rss)
  51                 hiwater_rss = mm->hiwater_rss;
  52
  53         /* split executable areas between text and lib */
  54         text = PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK);
  55         text = min(text, mm->exec_vm << PAGE_SHIFT);
  56         lib = (mm->exec_vm << PAGE_SHIFT) - text;
  57
  58         swap = get_mm_counter(mm, MM_SWAPENTS);
  59         SEQ_PUT_DEC("VmPeak:\t", hiwater_vm);
  60         SEQ_PUT_DEC(" kB\nVmSize:\t", total_vm);
  61         SEQ_PUT_DEC(" kB\nVmLck:\t", mm->locked_vm);
  62         SEQ_PUT_DEC(" kB\nVmPin:\t", atomic64_read(&mm->pinned_vm));
  63         SEQ_PUT_DEC(" kB\nVmHWM:\t", hiwater_rss);
  64         SEQ_PUT_DEC(" kB\nVmRSS:\t", total_rss);
  65         SEQ_PUT_DEC(" kB\nRssAnon:\t", anon);
  66         SEQ_PUT_DEC(" kB\nRssFile:\t", file);
  67         SEQ_PUT_DEC(" kB\nRssShmem:\t", shmem);
  68         SEQ_PUT_DEC(" kB\nVmData:\t", mm->data_vm);
  69         SEQ_PUT_DEC(" kB\nVmStk:\t", mm->stack_vm);
  70         seq_put_decimal_ull_width(m,
  71                     " kB\nVmExe:\t", text >> 10, 8);
  72         seq_put_decimal_ull_width(m,
  73                     " kB\nVmLib:\t", lib >> 10, 8);
  74         seq_put_decimal_ull_width(m,
  75                     " kB\nVmPTE:\t", mm_pgtables_bytes(mm) >> 10, 8);
  76         SEQ_PUT_DEC(" kB\nVmSwap:\t", swap);
  77         seq_puts(m, " kB\n");
  78         hugetlb_report_usage(m, mm);
  79 }
  80 #undef SEQ_PUT_DEC
  81
  82 unsigned long task_vsize(struct mm_struct *mm)
  83 {
  84         return PAGE_SIZE * mm->total_vm;
  85 }
  86
  87 unsigned long task_statm(struct mm_struct *mm,
  88                          unsigned long *shared, unsigned long *text,
  89                          unsigned long *data, unsigned long *resident)
  90 {
  91         *shared = get_mm_counter(mm, MM_FILEPAGES) +
  92                         get_mm_counter(mm, MM_SHMEMPAGES);
  93         *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
  94                                                                 >> PAGE_SHIFT;
  95         *data = mm->data_vm + mm->stack_vm;
  96         *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
  97         return mm->total_vm;
  98 }
  99
 100 #ifdef CONFIG_NUMA
 101 /*
 102  * Save get_task_policy() for show_numa_map().
 103  */
 104 static void hold_task_mempolicy(struct proc_maps_private *priv)
 105 {
 106         struct task_struct *task = priv->task;
 107
 108         task_lock(task);
 109         priv->task_mempolicy = get_task_policy(task);
 110         mpol_get(priv->task_mempolicy);
 111         task_unlock(task);
 112 }
 113 static void release_task_mempolicy(struct proc_maps_private *priv)
 114 {
 115         mpol_put(priv->task_mempolicy);
 116 }
 117 #else
 118 static void hold_task_mempolicy(struct proc_maps_private *priv)
 119 {
 120 }
 121 static void release_task_mempolicy(struct proc_maps_private *priv)
 122 {
 123 }
 124 #endif
 125
 126 static void *m_start(struct seq_file *m, loff_t *ppos)
 127 {
 128         struct proc_maps_private *priv = m->private;
 129         unsigned long last_addr = *ppos;
 130         struct mm_struct *mm;
 131         struct vm_area_struct *vma;
 132
 133         /* See m_next(). Zero at the start or after lseek. */
 134         if (last_addr == -1UL)
 135                 return NULL;
 136
 137         priv->task = get_proc_task(priv->inode);
 138         if (!priv->task)
 139                 return ERR_PTR(-ESRCH);
 140
 141         mm = priv->mm;
 142         if (!mm || !mmget_not_zero(mm)) {
 143                 put_task_struct(priv->task);
 144                 priv->task = NULL;
 145                 return NULL;
 146         }
 147
 148         if (mmap_read_lock_killable(mm)) {
 149                 mmput(mm);
 150                 put_task_struct(priv->task);
 151                 priv->task = NULL;
 152                 return ERR_PTR(-EINTR);
 153         }
 154
 155         hold_task_mempolicy(priv);
 156         priv->tail_vma = get_gate_vma(mm);
 157
 158         vma = find_vma(mm, last_addr);
 159         if (vma)
 160                 return vma;
 161
 162         return priv->tail_vma;
 163 }
 164
 165 static void *m_next(struct seq_file *m, void *v, loff_t *ppos)
 166 {
 167         struct proc_maps_private *priv = m->private;
 168         struct vm_area_struct *next, *vma = v;
 169
 170         if (vma == priv->tail_vma)
 171                 next = NULL;
 172         else if (vma->vm_next)
 173                 next = vma->vm_next;
 174         else
 175                 next = priv->tail_vma;
 176
 177         *ppos = next ? next->vm_start : -1UL;
 178
 179         return next;
 180 }
 181
 182 static void m_stop(struct seq_file *m, void *v)
 183 {
 184         struct proc_maps_private *priv = m->private;
 185         struct mm_struct *mm = priv->mm;
 186
 187         if (!priv->task)
 188                 return;
 189
 190         release_task_mempolicy(priv);
 191         mmap_read_unlock(mm);
 192         mmput(mm);
 193         put_task_struct(priv->task);
 194         priv->task = NULL;
 195 }
 196
 197 static int proc_maps_open(struct inode *inode, struct file *file,
 198                         const struct seq_operations *ops, int psize)
 199 {
 200         struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
 201
 202         if (!priv)
 203                 return -ENOMEM;
 204
 205         priv->inode = inode;
 206         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
 207         if (IS_ERR(priv->mm)) {
 208                 int err = PTR_ERR(priv->mm);
 209
 210                 seq_release_private(inode, file);
 211                 return err;
 212         }
 213
 214         return 0;
 215 }
 216
 217 static int proc_map_release(struct inode *inode, struct file *file)
 218 {
 219         struct seq_file *seq = file->private_data;
 220         struct proc_maps_private *priv = seq->private;
 221
 222         if (priv->mm)
 223                 mmdrop(priv->mm);
 224
 225         return seq_release_private(inode, file);
 226 }
 227
 228 static int do_maps_open(struct inode *inode, struct file *file,
 229                         const struct seq_operations *ops)
 230 {
 231         return proc_maps_open(inode, file, ops,
 232                                 sizeof(struct proc_maps_private));
 233 }
 234
 235 /*
 236  * Indicate if the VMA is a stack for the given task; for
 237  * /proc/PID/maps that is the stack of the main task.
 238  */
 239 static int is_stack(struct vm_area_struct *vma)
 240 {
 241         /*
 242          * We make no effort to guess what a given thread considers to be
 243          * its "stack".  It's not even well-defined for programs written
 244          * languages like Go.
 245          */
 246         return vma->vm_start <= vma->vm_mm->start_stack &&
 247                 vma->vm_end >= vma->vm_mm->start_stack;
 248 }
 249
 250 static void show_vma_header_prefix(struct seq_file *m,
 251                                    unsigned long start, unsigned long end,
 252                                    vm_flags_t flags, unsigned long long pgoff,
 253                                    dev_t dev, unsigned long ino)
 254 {
 255         seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
 256         seq_put_hex_ll(m, NULL, start, 8);
 257         seq_put_hex_ll(m, "-", end, 8);
 258         seq_putc(m, ' ');
 259         seq_putc(m, flags & VM_READ ? 'r' : '-');
 260         seq_putc(m, flags & VM_WRITE ? 'w' : '-');
 261         seq_putc(m, flags & VM_EXEC ? 'x' : '-');
 262         seq_putc(m, flags & VM_MAYSHARE ? 's' : 'p');
 263         seq_put_hex_ll(m, " ", pgoff, 8);
 264         seq_put_hex_ll(m, " ", MAJOR(dev), 2);
 265         seq_put_hex_ll(m, ":", MINOR(dev), 2);
 266         seq_put_decimal_ull(m, " ", ino);
 267         seq_putc(m, ' ');
 268 }
 269
 270 static void
 271 show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
 272 {
 273         struct mm_struct *mm = vma->vm_mm;
 274         struct file *file = vma->vm_file;
 275         vm_flags_t flags = vma->vm_flags;
 276         unsigned long ino = 0;
 277         unsigned long long pgoff = 0;
 278         unsigned long start, end;
 279         dev_t dev = 0;
 280         const char *name = NULL;
 281
 282         if (file) {
 283                 struct inode *inode = file_inode(vma->vm_file);
 284                 dev = inode->i_sb->s_dev;
 285                 ino = inode->i_ino;
 286                 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
 287         }
 288
 289         start = vma->vm_start;
 290         end = vma->vm_end;
 291         show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino);
 292
 293         /*
 294          * Print the dentry name for named mappings, and a
 295          * special [heap] marker for the heap:
 296          */
 297         if (file) {
 298                 seq_pad(m, ' ');
 299                 seq_file_path(m, file, "\n");
 300                 goto done;
 301         }
 302
 303         if (vma->vm_ops && vma->vm_ops->name) {
 304                 name = vma->vm_ops->name(vma);
 305                 if (name)
 306                         goto done;
 307         }
 308
 309         name = arch_vma_name(vma);
 310         if (!name) {
 311                 if (!mm) {
 312                         name = "[vdso]";
 313                         goto done;
 314                 }
 315
 316                 if (vma->vm_start <= mm->brk &&
 317                     vma->vm_end >= mm->start_brk) {
 318                         name = "[heap]";
 319                         goto done;
 320                 }
 321
 322                 if (is_stack(vma))
 323                         name = "[stack]";
 324         }
 325
 326 done:
 327         if (name) {
 328                 seq_pad(m, ' ');
 329                 seq_puts(m, name);
 330         }
 331         seq_putc(m, '\n');
 332 }
 333
 334 static int show_map(struct seq_file *m, void *v)
 335 {
 336         show_map_vma(m, v);
 337         return 0;
 338 }
 339
 340 static const struct seq_operations proc_pid_maps_op = {
 341         .start  = m_start,
 342         .next   = m_next,
 343         .stop   = m_stop,
 344         .show   = show_map
 345 };
 346
 347 static int pid_maps_open(struct inode *inode, struct file *file)
 348 {
 349         return do_maps_open(inode, file, &proc_pid_maps_op);
 350 }
 351
 352 const struct file_operations proc_pid_maps_operations = {
 353         .open           = pid_maps_open,
 354         .read           = seq_read,
 355         .llseek         = seq_lseek,
 356         .release        = proc_map_release,
 357 };
 358
 359 /*
 360  * Proportional Set Size(PSS): my share of RSS.
 361  *
 362  * PSS of a process is the count of pages it has in memory, where each
 363  * page is divided by the number of processes sharing it.  So if a
 364  * process has 1000 pages all to itself, and 1000 shared with one other
 365  * process, its PSS will be 1500.
 366  *
 367  * To keep (accumulated) division errors low, we adopt a 64bit
 368  * fixed-point pss counter to minimize division errors. So (pss >>
 369  * PSS_SHIFT) would be the real byte count.
 370  *
 371  * A shift of 12 before division means (assuming 4K page size):
 372  *      - 1M 3-user-pages add up to 8KB errors;
 373  *      - supports mapcount up to 2^24, or 16M;
 374  *      - supports PSS up to 2^52 bytes, or 4PB.
 375  */
 376 #define PSS_SHIFT 12
 377
 378 #ifdef CONFIG_PROC_PAGE_MONITOR
 379 struct mem_size_stats {
 380         unsigned long resident;
 381         unsigned long shared_clean;
 382         unsigned long shared_dirty;
 383         unsigned long private_clean;
 384         unsigned long private_dirty;
 385         unsigned long referenced;
 386         unsigned long anonymous;
 387         unsigned long lazyfree;
 388         unsigned long anonymous_thp;
 389         unsigned long shmem_thp;
 390         unsigned long file_thp;
 391         unsigned long swap;
 392         unsigned long shared_hugetlb;
 393         unsigned long private_hugetlb;
 394         u64 pss;
 395         u64 pss_anon;
 396         u64 pss_file;
 397         u64 pss_shmem;
 398         u64 pss_locked;
 399         u64 swap_pss;
 400         bool check_shmem_swap;
 401 };
 402
 403 static void smaps_page_accumulate(struct mem_size_stats *mss,
 404                 struct page *page, unsigned long size, unsigned long pss,
 405                 bool dirty, bool locked, bool private)
 406 {
 407         mss->pss += pss;
 408
 409         if (PageAnon(page))
 410                 mss->pss_anon += pss;
 411         else if (PageSwapBacked(page))
 412                 mss->pss_shmem += pss;
 413         else
 414                 mss->pss_file += pss;
 415
 416         if (locked)
 417                 mss->pss_locked += pss;
 418
 419         if (dirty || PageDirty(page)) {
 420                 if (private)
 421                         mss->private_dirty += size;
 422                 else
 423                         mss->shared_dirty += size;
 424         } else {
 425                 if (private)
 426                         mss->private_clean += size;
 427                 else
 428                         mss->shared_clean += size;
 429         }
 430 }
 431
 432 static void smaps_account(struct mem_size_stats *mss, struct page *page,
 433                 bool compound, bool young, bool dirty, bool locked)
 434 {
 435         int i, nr = compound ? compound_nr(page) : 1;
 436         unsigned long size = nr * PAGE_SIZE;
 437
 438         /*
 439          * First accumulate quantities that depend only on |size| and the type
 440          * of the compound page.
 441          */
 442         if (PageAnon(page)) {
 443                 mss->anonymous += size;
 444                 if (!PageSwapBacked(page) && !dirty && !PageDirty(page))
 445                         mss->lazyfree += size;
 446         }
 447
 448         mss->resident += size;
 449         /* Accumulate the size in pages that have been accessed. */
 450         if (young || page_is_young(page) || PageReferenced(page))
 451                 mss->referenced += size;
 452
 453         /*
 454          * Then accumulate quantities that may depend on sharing, or that may
 455          * differ page-by-page.
 456          *
 457          * page_count(page) == 1 guarantees the page is mapped exactly once.
 458          * If any subpage of the compound page mapped with PTE it would elevate
 459          * page_count().
 460          */
 461         if (page_count(page) == 1) {
 462                 smaps_page_accumulate(mss, page, size, size << PSS_SHIFT, dirty,
 463                         locked, true);
 464                 return;
 465         }
 466         for (i = 0; i < nr; i++, page++) {
 467                 int mapcount = page_mapcount(page);
 468                 unsigned long pss = PAGE_SIZE << PSS_SHIFT;
 469                 if (mapcount >= 2)
 470                         pss /= mapcount;
 471                 smaps_page_accumulate(mss, page, PAGE_SIZE, pss, dirty, locked,
 472                                       mapcount < 2);
 473         }
 474 }
 475
 476 #ifdef CONFIG_SHMEM
 477 static int smaps_pte_hole(unsigned long addr, unsigned long end,
 478                           __always_unused int depth, struct mm_walk *walk)
 479 {
 480         struct mem_size_stats *mss = walk->private;
 481
 482         mss->swap += shmem_partial_swap_usage(
 483                         walk->vma->vm_file->f_mapping, addr, end);
 484
 485         return 0;
 486 }
 487 #else
 488 #define smaps_pte_hole          NULL
 489 #endif /* CONFIG_SHMEM */
 490
 491 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
 492                 struct mm_walk *walk)
 493 {
 494         struct mem_size_stats *mss = walk->private;
 495         struct vm_area_struct *vma = walk->vma;
 496         bool locked = !!(vma->vm_flags & VM_LOCKED);
 497         struct page *page = NULL;
 498
 499         if (pte_present(*pte)) {
 500                 page = vm_normal_page(vma, addr, *pte);
 501         } else if (is_swap_pte(*pte)) {
 502                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 503
 504                 if (!non_swap_entry(swpent)) {
 505                         int mapcount;
 506
 507                         mss->swap += PAGE_SIZE;
 508                         mapcount = swp_swapcount(swpent);
 509                         if (mapcount >= 2) {
 510                                 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
 511
 512                                 do_div(pss_delta, mapcount);
 513                                 mss->swap_pss += pss_delta;
 514                         } else {
 515                                 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
 516                         }
 517                 } else if (is_migration_entry(swpent))
 518                         page = migration_entry_to_page(swpent);
 519                 else if (is_device_private_entry(swpent))
 520                         page = device_private_entry_to_page(swpent);
 521         } else if (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap
 522                                                         && pte_none(*pte))) {
 523                 page = find_get_entry(vma->vm_file->f_mapping,
 524                                                 linear_page_index(vma, addr));
 525                 if (!page)
 526                         return;
 527
 528                 if (xa_is_value(page))
 529                         mss->swap += PAGE_SIZE;
 530                 else
 531                         put_page(page);
 532
 533                 return;
 534         }
 535
 536         if (!page)
 537                 return;
 538
 539         smaps_account(mss, page, false, pte_young(*pte), pte_dirty(*pte), locked);
 540 }
 541
 542 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 543 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 544                 struct mm_walk *walk)
 545 {
 546         struct mem_size_stats *mss = walk->private;
 547         struct vm_area_struct *vma = walk->vma;
 548         bool locked = !!(vma->vm_flags & VM_LOCKED);
 549         struct page *page = NULL;
 550
 551         if (pmd_present(*pmd)) {
 552                 /* FOLL_DUMP will return -EFAULT on huge zero page */
 553                 page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
 554         } else if (unlikely(thp_migration_supported() && is_swap_pmd(*pmd))) {
 555                 swp_entry_t entry = pmd_to_swp_entry(*pmd);
 556
 557                 if (is_migration_entry(entry))
 558                         page = migration_entry_to_page(entry);
 559         }
 560         if (IS_ERR_OR_NULL(page))
 561                 return;
 562         if (PageAnon(page))
 563                 mss->anonymous_thp += HPAGE_PMD_SIZE;
 564         else if (PageSwapBacked(page))
 565                 mss->shmem_thp += HPAGE_PMD_SIZE;
 566         else if (is_zone_device_page(page))
 567                 /* pass */;
 568         else
 569                 mss->file_thp += HPAGE_PMD_SIZE;
 570         smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd), locked);
 571 }
 572 #else
 573 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 574                 struct mm_walk *walk)
 575 {
 576 }
 577 #endif
 578
 579 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 580                            struct mm_walk *walk)
 581 {
 582         struct vm_area_struct *vma = walk->vma;
 583         pte_t *pte;
 584         spinlock_t *ptl;
 585
 586         ptl = pmd_trans_huge_lock(pmd, vma);
 587         if (ptl) {
 588                 smaps_pmd_entry(pmd, addr, walk);
 589                 spin_unlock(ptl);
 590                 goto out;
 591         }
 592
 593         if (pmd_trans_unstable(pmd))
 594                 goto out;
 595         /*
 596          * The mmap_lock held all the way back in m_start() is what
 597          * keeps khugepaged out of here and from collapsing things
 598          * in here.
 599          */
 600         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 601         for (; addr != end; pte++, addr += PAGE_SIZE)
 602                 smaps_pte_entry(pte, addr, walk);
 603         pte_unmap_unlock(pte - 1, ptl);
 604 out:
 605         cond_resched();
 606         return 0;
 607 }
 608
 609 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
 610 {
 611         /*
 612          * Don't forget to update Documentation/ on changes.
 613          */
 614         static const char mnemonics[BITS_PER_LONG][2] = {
 615                 /*
 616                  * In case if we meet a flag we don't know about.
 617                  */
 618                 [0 ... (BITS_PER_LONG-1)] = "??",
 619
 620                 [ilog2(VM_READ)]        = "rd",
 621                 [ilog2(VM_WRITE)]       = "wr",
 622                 [ilog2(VM_EXEC)]        = "ex",
 623                 [ilog2(VM_SHARED)]      = "sh",
 624                 [ilog2(VM_MAYREAD)]     = "mr",
 625                 [ilog2(VM_MAYWRITE)]    = "mw",
 626                 [ilog2(VM_MAYEXEC)]     = "me",
 627                 [ilog2(VM_MAYSHARE)]    = "ms",
 628                 [ilog2(VM_GROWSDOWN)]   = "gd",
 629                 [ilog2(VM_PFNMAP)]      = "pf",
 630                 [ilog2(VM_DENYWRITE)]   = "dw",
 631                 [ilog2(VM_LOCKED)]      = "lo",
 632                 [ilog2(VM_IO)]          = "io",
 633                 [ilog2(VM_SEQ_READ)]    = "sr",
 634                 [ilog2(VM_RAND_READ)]   = "rr",
 635                 [ilog2(VM_DONTCOPY)]    = "dc",
 636                 [ilog2(VM_DONTEXPAND)]  = "de",
 637                 [ilog2(VM_ACCOUNT)]     = "ac",
 638                 [ilog2(VM_NORESERVE)]   = "nr",
 639                 [ilog2(VM_HUGETLB)]     = "ht",
 640                 [ilog2(VM_SYNC)]        = "sf",
 641                 [ilog2(VM_ARCH_1)]      = "ar",
 642                 [ilog2(VM_WIPEONFORK)]  = "wf",
 643                 [ilog2(VM_DONTDUMP)]    = "dd",
 644 #ifdef CONFIG_ARM64_BTI
 645                 [ilog2(VM_ARM64_BTI)]   = "bt",
 646 #endif
 647 #ifdef CONFIG_MEM_SOFT_DIRTY
 648                 [ilog2(VM_SOFTDIRTY)]   = "sd",
 649 #endif
 650                 [ilog2(VM_MIXEDMAP)]    = "mm",
 651                 [ilog2(VM_HUGEPAGE)]    = "hg",
 652                 [ilog2(VM_NOHUGEPAGE)]  = "nh",
 653                 [ilog2(VM_MERGEABLE)]   = "mg",
 654                 [ilog2(VM_UFFD_MISSING)]= "um",
 655                 [ilog2(VM_UFFD_WP)]     = "uw",
 656 #ifdef CONFIG_ARM64_MTE
 657                 [ilog2(VM_MTE)]         = "mt",
 658                 [ilog2(VM_MTE_ALLOWED)] = "",
 659 #endif
 660 #ifdef CONFIG_ARCH_HAS_PKEYS
 661                 /* These come out via ProtectionKey: */
 662                 [ilog2(VM_PKEY_BIT0)]   = "",
 663                 [ilog2(VM_PKEY_BIT1)]   = "",
 664                 [ilog2(VM_PKEY_BIT2)]   = "",
 665                 [ilog2(VM_PKEY_BIT3)]   = "",
 666 #if VM_PKEY_BIT4
 667                 [ilog2(VM_PKEY_BIT4)]   = "",
 668 #endif
 669 #endif /* CONFIG_ARCH_HAS_PKEYS */
 670         };
 671         size_t i;
 672
 673         seq_puts(m, "VmFlags: ");
 674         for (i = 0; i < BITS_PER_LONG; i++) {
 675                 if (!mnemonics[i][0])
 676                         continue;
 677                 if (vma->vm_flags & (1UL << i)) {
 678                         seq_putc(m, mnemonics[i][0]);
 679                         seq_putc(m, mnemonics[i][1]);
 680                         seq_putc(m, ' ');
 681                 }
 682         }
 683         seq_putc(m, '\n');
 684 }
 685
 686 #ifdef CONFIG_HUGETLB_PAGE
 687 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
 688                                  unsigned long addr, unsigned long end,
 689                                  struct mm_walk *walk)
 690 {
 691         struct mem_size_stats *mss = walk->private;
 692         struct vm_area_struct *vma = walk->vma;
 693         struct page *page = NULL;
 694
 695         if (pte_present(*pte)) {
 696                 page = vm_normal_page(vma, addr, *pte);
 697         } else if (is_swap_pte(*pte)) {
 698                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 699
 700                 if (is_migration_entry(swpent))
 701                         page = migration_entry_to_page(swpent);
 702                 else if (is_device_private_entry(swpent))
 703                         page = device_private_entry_to_page(swpent);
 704         }
 705         if (page) {
 706                 int mapcount = page_mapcount(page);
 707
 708                 if (mapcount >= 2)
 709                         mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
 710                 else
 711                         mss->private_hugetlb += huge_page_size(hstate_vma(vma));
 712         }
 713         return 0;
 714 }
 715 #else
 716 #define smaps_hugetlb_range     NULL
 717 #endif /* HUGETLB_PAGE */
 718
 719 static const struct mm_walk_ops smaps_walk_ops = {
 720         .pmd_entry              = smaps_pte_range,
 721         .hugetlb_entry          = smaps_hugetlb_range,
 722 };
 723
 724 static const struct mm_walk_ops smaps_shmem_walk_ops = {
 725         .pmd_entry              = smaps_pte_range,
 726         .hugetlb_entry          = smaps_hugetlb_range,
 727         .pte_hole               = smaps_pte_hole,
 728 };
 729
 730 static void smap_gather_stats(struct vm_area_struct *vma,
 731                              struct mem_size_stats *mss)
 732 {
 733 #ifdef CONFIG_SHMEM
 734         /* In case of smaps_rollup, reset the value from previous vma */
 735         mss->check_shmem_swap = false;
 736         if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
 737                 /*
 738                  * For shared or readonly shmem mappings we know that all
 739                  * swapped out pages belong to the shmem object, and we can
 740                  * obtain the swap value much more efficiently. For private
 741                  * writable mappings, we might have COW pages that are
 742                  * not affected by the parent swapped out pages of the shmem
 743                  * object, so we have to distinguish them during the page walk.
 744                  * Unless we know that the shmem object (or the part mapped by
 745                  * our VMA) has no swapped out pages at all.
 746                  */
 747                 unsigned long shmem_swapped = shmem_swap_usage(vma);
 748
 749                 if (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
 750                                         !(vma->vm_flags & VM_WRITE)) {
 751                         mss->swap += shmem_swapped;
 752                 } else {
 753                         mss->check_shmem_swap = true;
 754                         walk_page_vma(vma, &smaps_shmem_walk_ops, mss);
 755                         return;
 756                 }
 757         }
 758 #endif
 759         /* mmap_lock is held in m_start */
 760         walk_page_vma(vma, &smaps_walk_ops, mss);
 761 }
 762
 763 #define SEQ_PUT_DEC(str, val) \
 764                 seq_put_decimal_ull_width(m, str, (val) >> 10, 8)
 765
 766 /* Show the contents common for smaps and smaps_rollup */
 767 static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss,
 768         bool rollup_mode)
 769 {
 770         SEQ_PUT_DEC("Rss:            ", mss->resident);
 771         SEQ_PUT_DEC(" kB\nPss:            ", mss->pss >> PSS_SHIFT);
 772         if (rollup_mode) {
 773                 /*
 774                  * These are meaningful only for smaps_rollup, otherwise two of
 775                  * them are zero, and the other one is the same as Pss.
 776                  */
 777                 SEQ_PUT_DEC(" kB\nPss_Anon:       ",
 778                         mss->pss_anon >> PSS_SHIFT);
 779                 SEQ_PUT_DEC(" kB\nPss_File:       ",
 780                         mss->pss_file >> PSS_SHIFT);
 781                 SEQ_PUT_DEC(" kB\nPss_Shmem:      ",
 782                         mss->pss_shmem >> PSS_SHIFT);
 783         }
 784         SEQ_PUT_DEC(" kB\nShared_Clean:   ", mss->shared_clean);
 785         SEQ_PUT_DEC(" kB\nShared_Dirty:   ", mss->shared_dirty);
 786         SEQ_PUT_DEC(" kB\nPrivate_Clean:  ", mss->private_clean);
 787         SEQ_PUT_DEC(" kB\nPrivate_Dirty:  ", mss->private_dirty);
 788         SEQ_PUT_DEC(" kB\nReferenced:     ", mss->referenced);
 789         SEQ_PUT_DEC(" kB\nAnonymous:      ", mss->anonymous);
 790         SEQ_PUT_DEC(" kB\nLazyFree:       ", mss->lazyfree);
 791         SEQ_PUT_DEC(" kB\nAnonHugePages:  ", mss->anonymous_thp);
 792         SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp);
 793         SEQ_PUT_DEC(" kB\nFilePmdMapped:  ", mss->file_thp);
 794         SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb);
 795         seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ",
 796                                   mss->private_hugetlb >> 10, 7);
 797         SEQ_PUT_DEC(" kB\nSwap:           ", mss->swap);
 798         SEQ_PUT_DEC(" kB\nSwapPss:        ",
 799                                         mss->swap_pss >> PSS_SHIFT);
 800         SEQ_PUT_DEC(" kB\nLocked:         ",
 801                                         mss->pss_locked >> PSS_SHIFT);
 802         seq_puts(m, " kB\n");
 803 }
 804
 805 static int show_smap(struct seq_file *m, void *v)
 806 {
 807         struct vm_area_struct *vma = v;
 808         struct mem_size_stats mss;
 809
 810         memset(&mss, 0, sizeof(mss));
 811
 812         smap_gather_stats(vma, &mss);
 813
 814         show_map_vma(m, vma);
 815
 816         SEQ_PUT_DEC("Size:           ", vma->vm_end - vma->vm_start);
 817         SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma));
 818         SEQ_PUT_DEC(" kB\nMMUPageSize:    ", vma_mmu_pagesize(vma));
 819         seq_puts(m, " kB\n");
 820
 821         __show_smap(m, &mss, false);
 822
 823         seq_printf(m, "THPeligible:    %d\n",
 824                    transparent_hugepage_enabled(vma));
 825
 826         if (arch_pkeys_enabled())
 827                 seq_printf(m, "ProtectionKey:  %8u\n", vma_pkey(vma));
 828         show_smap_vma_flags(m, vma);
 829
 830         return 0;
 831 }
 832
 833 static int show_smaps_rollup(struct seq_file *m, void *v)
 834 {
 835         struct proc_maps_private *priv = m->private;
 836         struct mem_size_stats mss;
 837         struct mm_struct *mm;
 838         struct vm_area_struct *vma;
 839         unsigned long last_vma_end = 0;
 840         int ret = 0;
 841
 842         priv->task = get_proc_task(priv->inode);
 843         if (!priv->task)
 844                 return -ESRCH;
 845
 846         mm = priv->mm;
 847         if (!mm || !mmget_not_zero(mm)) {
 848                 ret = -ESRCH;
 849                 goto out_put_task;
 850         }
 851
 852         memset(&mss, 0, sizeof(mss));
 853
 854         ret = mmap_read_lock_killable(mm);
 855         if (ret)
 856                 goto out_put_mm;
 857
 858         hold_task_mempolicy(priv);
 859
 860         for (vma = priv->mm->mmap; vma; vma = vma->vm_next) {
 861                 smap_gather_stats(vma, &mss);
 862                 last_vma_end = vma->vm_end;
 863         }
 864
 865         show_vma_header_prefix(m, priv->mm->mmap->vm_start,
 866                                last_vma_end, 0, 0, 0, 0);
 867         seq_pad(m, ' ');
 868         seq_puts(m, "[rollup]\n");
 869
 870         __show_smap(m, &mss, true);
 871
 872         release_task_mempolicy(priv);
 873         mmap_read_unlock(mm);
 874
 875 out_put_mm:
 876         mmput(mm);
 877 out_put_task:
 878         put_task_struct(priv->task);
 879         priv->task = NULL;
 880
 881         return ret;
 882 }
 883 #undef SEQ_PUT_DEC
 884
 885 static const struct seq_operations proc_pid_smaps_op = {
 886         .start  = m_start,
 887         .next   = m_next,
 888         .stop   = m_stop,
 889         .show   = show_smap
 890 };
 891
 892 static int pid_smaps_open(struct inode *inode, struct file *file)
 893 {
 894         return do_maps_open(inode, file, &proc_pid_smaps_op);
 895 }
 896
 897 static int smaps_rollup_open(struct inode *inode, struct file *file)
 898 {
 899         int ret;
 900         struct proc_maps_private *priv;
 901
 902         priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT);
 903         if (!priv)
 904                 return -ENOMEM;
 905
 906         ret = single_open(file, show_smaps_rollup, priv);
 907         if (ret)
 908                 goto out_free;
 909
 910         priv->inode = inode;
 911         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
 912         if (IS_ERR(priv->mm)) {
 913                 ret = PTR_ERR(priv->mm);
 914
 915                 single_release(inode, file);
 916                 goto out_free;
 917         }
 918
 919         return 0;
 920
 921 out_free:
 922         kfree(priv);
 923         return ret;
 924 }
 925
 926 static int smaps_rollup_release(struct inode *inode, struct file *file)
 927 {
 928         struct seq_file *seq = file->private_data;
 929         struct proc_maps_private *priv = seq->private;
 930
 931         if (priv->mm)
 932                 mmdrop(priv->mm);
 933
 934         kfree(priv);
 935         return single_release(inode, file);
 936 }
 937
 938 const struct file_operations proc_pid_smaps_operations = {
 939         .open           = pid_smaps_open,
 940         .read           = seq_read,
 941         .llseek         = seq_lseek,
 942         .release        = proc_map_release,
 943 };
 944
 945 const struct file_operations proc_pid_smaps_rollup_operations = {
 946         .open           = smaps_rollup_open,
 947         .read           = seq_read,
 948         .llseek         = seq_lseek,
 949         .release        = smaps_rollup_release,
 950 };
 951
 952 enum clear_refs_types {
 953         CLEAR_REFS_ALL = 1,
 954         CLEAR_REFS_ANON,
 955         CLEAR_REFS_MAPPED,
 956         CLEAR_REFS_SOFT_DIRTY,
 957         CLEAR_REFS_MM_HIWATER_RSS,
 958         CLEAR_REFS_LAST,
 959 };
 960
 961 struct clear_refs_private {
 962         enum clear_refs_types type;
 963 };
 964
 965 #ifdef CONFIG_MEM_SOFT_DIRTY
 966 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 967                 unsigned long addr, pte_t *pte)
 968 {
 969         /*
 970          * The soft-dirty tracker uses #PF-s to catch writes
 971          * to pages, so write-protect the pte as well. See the
 972          * Documentation/admin-guide/mm/soft-dirty.rst for full description
 973          * of how soft-dirty works.
 974          */
 975         pte_t ptent = *pte;
 976
 977         if (pte_present(ptent)) {
 978                 pte_t old_pte;
 979
 980                 old_pte = ptep_modify_prot_start(vma, addr, pte);
 981                 ptent = pte_wrprotect(old_pte);
 982                 ptent = pte_clear_soft_dirty(ptent);
 983                 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent);
 984         } else if (is_swap_pte(ptent)) {
 985                 ptent = pte_swp_clear_soft_dirty(ptent);
 986                 set_pte_at(vma->vm_mm, addr, pte, ptent);
 987         }
 988 }
 989 #else
 990 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 991                 unsigned long addr, pte_t *pte)
 992 {
 993 }
 994 #endif
 995
 996 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
 997 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
 998                 unsigned long addr, pmd_t *pmdp)
 999 {
1000         pmd_t old, pmd = *pmdp;
1001
1002         if (pmd_present(pmd)) {
1003                 /* See comment in change_huge_pmd() */
1004                 old = pmdp_invalidate(vma, addr, pmdp);
1005                 if (pmd_dirty(old))
1006                         pmd = pmd_mkdirty(pmd);
1007                 if (pmd_young(old))
1008                         pmd = pmd_mkyoung(pmd);
1009
1010                 pmd = pmd_wrprotect(pmd);
1011                 pmd = pmd_clear_soft_dirty(pmd);
1012
1013                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1014         } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1015                 pmd = pmd_swp_clear_soft_dirty(pmd);
1016                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1017         }
1018 }
1019 #else
1020 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1021                 unsigned long addr, pmd_t *pmdp)
1022 {
1023 }
1024 #endif
1025
1026 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
1027                                 unsigned long end, struct mm_walk *walk)
1028 {
1029         struct clear_refs_private *cp = walk->private;
1030         struct vm_area_struct *vma = walk->vma;
1031         pte_t *pte, ptent;
1032         spinlock_t *ptl;
1033         struct page *page;
1034
1035         ptl = pmd_trans_huge_lock(pmd, vma);
1036         if (ptl) {
1037                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1038                         clear_soft_dirty_pmd(vma, addr, pmd);
1039                         goto out;
1040                 }
1041
1042                 if (!pmd_present(*pmd))
1043                         goto out;
1044
1045                 page = pmd_page(*pmd);
1046
1047                 /* Clear accessed and referenced bits. */
1048                 pmdp_test_and_clear_young(vma, addr, pmd);
1049                 test_and_clear_page_young(page);
1050                 ClearPageReferenced(page);
1051 out:
1052                 spin_unlock(ptl);
1053                 return 0;
1054         }
1055
1056         if (pmd_trans_unstable(pmd))
1057                 return 0;
1058
1059         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1060         for (; addr != end; pte++, addr += PAGE_SIZE) {
1061                 ptent = *pte;
1062
1063                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1064                         clear_soft_dirty(vma, addr, pte);
1065                         continue;
1066                 }
1067
1068                 if (!pte_present(ptent))
1069                         continue;
1070
1071                 page = vm_normal_page(vma, addr, ptent);
1072                 if (!page)
1073                         continue;
1074
1075                 /* Clear accessed and referenced bits. */
1076                 ptep_test_and_clear_young(vma, addr, pte);
1077                 test_and_clear_page_young(page);
1078                 ClearPageReferenced(page);
1079         }
1080         pte_unmap_unlock(pte - 1, ptl);
1081         cond_resched();
1082         return 0;
1083 }
1084
1085 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1086                                 struct mm_walk *walk)
1087 {
1088         struct clear_refs_private *cp = walk->private;
1089         struct vm_area_struct *vma = walk->vma;
1090
1091         if (vma->vm_flags & VM_PFNMAP)
1092                 return 1;
1093
1094         /*
1095          * Writing 1 to /proc/pid/clear_refs affects all pages.
1096          * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1097          * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1098          * Writing 4 to /proc/pid/clear_refs affects all pages.
1099          */
1100         if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1101                 return 1;
1102         if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1103                 return 1;
1104         return 0;
1105 }
1106
1107 static const struct mm_walk_ops clear_refs_walk_ops = {
1108         .pmd_entry              = clear_refs_pte_range,
1109         .test_walk              = clear_refs_test_walk,
1110 };
1111
1112 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1113                                 size_t count, loff_t *ppos)
1114 {
1115         struct task_struct *task;
1116         char buffer[PROC_NUMBUF];
1117         struct mm_struct *mm;
1118         struct vm_area_struct *vma;
1119         enum clear_refs_types type;
1120         struct mmu_gather tlb;
1121         int itype;
1122         int rv;
1123
1124         memset(buffer, 0, sizeof(buffer));
1125         if (count > sizeof(buffer) - 1)
1126                 count = sizeof(buffer) - 1;
1127         if (copy_from_user(buffer, buf, count))
1128                 return -EFAULT;
1129         rv = kstrtoint(strstrip(buffer), 10, &itype);
1130         if (rv < 0)
1131                 return rv;
1132         type = (enum clear_refs_types)itype;
1133         if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1134                 return -EINVAL;
1135
1136         task = get_proc_task(file_inode(file));
1137         if (!task)
1138                 return -ESRCH;
1139         mm = get_task_mm(task);
1140         if (mm) {
1141                 struct mmu_notifier_range range;
1142                 struct clear_refs_private cp = {
1143                         .type = type,
1144                 };
1145
1146                 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1147                         if (mmap_write_lock_killable(mm)) {
1148                                 count = -EINTR;
1149                                 goto out_mm;
1150                         }
1151
1152                         /*
1153                          * Writing 5 to /proc/pid/clear_refs resets the peak
1154                          * resident set size to this mm's current rss value.
1155                          */
1156                         reset_mm_hiwater_rss(mm);
1157                         mmap_write_unlock(mm);
1158                         goto out_mm;
1159                 }
1160
1161                 if (mmap_read_lock_killable(mm)) {
1162                         count = -EINTR;
1163                         goto out_mm;
1164                 }
1165                 tlb_gather_mmu(&tlb, mm, 0, -1);
1166                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1167                         for (vma = mm->mmap; vma; vma = vma->vm_next) {
1168                                 if (!(vma->vm_flags & VM_SOFTDIRTY))
1169                                         continue;
1170                                 mmap_read_unlock(mm);
1171                                 if (mmap_write_lock_killable(mm)) {
1172                                         count = -EINTR;
1173                                         goto out_mm;
1174                                 }
1175                                 /*
1176                                  * Avoid to modify vma->vm_flags
1177                                  * without locked ops while the
1178                                  * coredump reads the vm_flags.
1179                                  */
1180                                 if (!mmget_still_valid(mm)) {
1181                                         /*
1182                                          * Silently return "count"
1183                                          * like if get_task_mm()
1184                                          * failed. FIXME: should this
1185                                          * function have returned
1186                                          * -ESRCH if get_task_mm()
1187                                          * failed like if
1188                                          * get_proc_task() fails?
1189                                          */
1190                                         mmap_write_unlock(mm);
1191                                         goto out_mm;
1192                                 }
1193                                 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1194                                         vma->vm_flags &= ~VM_SOFTDIRTY;
1195                                         vma_set_page_prot(vma);
1196                                 }
1197                                 mmap_write_downgrade(mm);
1198                                 break;
1199                         }
1200
1201                         mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY,
1202                                                 0, NULL, mm, 0, -1UL);
1203                         mmu_notifier_invalidate_range_start(&range);
1204                 }
1205                 walk_page_range(mm, 0, mm->highest_vm_end, &clear_refs_walk_ops,
1206                                 &cp);
1207                 if (type == CLEAR_REFS_SOFT_DIRTY)
1208                         mmu_notifier_invalidate_range_end(&range);
1209                 tlb_finish_mmu(&tlb, 0, -1);
1210                 mmap_read_unlock(mm);
1211 out_mm:
1212                 mmput(mm);
1213         }
1214         put_task_struct(task);
1215
1216         return count;
1217 }
1218
1219 const struct file_operations proc_clear_refs_operations = {
1220         .write          = clear_refs_write,
1221         .llseek         = noop_llseek,
1222 };
1223
1224 typedef struct {
1225         u64 pme;
1226 } pagemap_entry_t;
1227
1228 struct pagemapread {
1229         int pos, len;           /* units: PM_ENTRY_BYTES, not bytes */
1230         pagemap_entry_t *buffer;
1231         bool show_pfn;
1232 };
1233
1234 #define PAGEMAP_WALK_SIZE       (PMD_SIZE)
1235 #define PAGEMAP_WALK_MASK       (PMD_MASK)
1236
1237 #define PM_ENTRY_BYTES          sizeof(pagemap_entry_t)
1238 #define PM_PFRAME_BITS          55
1239 #define PM_PFRAME_MASK          GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1240 #define PM_SOFT_DIRTY           BIT_ULL(55)
1241 #define PM_MMAP_EXCLUSIVE       BIT_ULL(56)
1242 #define PM_FILE                 BIT_ULL(61)
1243 #define PM_SWAP                 BIT_ULL(62)
1244 #define PM_PRESENT              BIT_ULL(63)
1245
1246 #define PM_END_OF_BUFFER    1
1247
1248 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1249 {
1250         return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1251 }
1252
1253 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1254                           struct pagemapread *pm)
1255 {
1256         pm->buffer[pm->pos++] = *pme;
1257         if (pm->pos >= pm->len)
1258                 return PM_END_OF_BUFFER;
1259         return 0;
1260 }
1261
1262 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1263                             __always_unused int depth, struct mm_walk *walk)
1264 {
1265         struct pagemapread *pm = walk->private;
1266         unsigned long addr = start;
1267         int err = 0;
1268
1269         while (addr < end) {
1270                 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1271                 pagemap_entry_t pme = make_pme(0, 0);
1272                 /* End of address space hole, which we mark as non-present. */
1273                 unsigned long hole_end;
1274
1275                 if (vma)
1276                         hole_end = min(end, vma->vm_start);
1277                 else
1278                         hole_end = end;
1279
1280                 for (; addr < hole_end; addr += PAGE_SIZE) {
1281                         err = add_to_pagemap(addr, &pme, pm);
1282                         if (err)
1283                                 goto out;
1284                 }
1285
1286                 if (!vma)
1287                         break;
1288
1289                 /* Addresses in the VMA. */
1290                 if (vma->vm_flags & VM_SOFTDIRTY)
1291                         pme = make_pme(0, PM_SOFT_DIRTY);
1292                 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1293                         err = add_to_pagemap(addr, &pme, pm);
1294                         if (err)
1295                                 goto out;
1296                 }
1297         }
1298 out:
1299         return err;
1300 }
1301
1302 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1303                 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1304 {
1305         u64 frame = 0, flags = 0;
1306         struct page *page = NULL;
1307
1308         if (pte_present(pte)) {
1309                 if (pm->show_pfn)
1310                         frame = pte_pfn(pte);
1311                 flags |= PM_PRESENT;
1312                 page = vm_normal_page(vma, addr, pte);
1313                 if (pte_soft_dirty(pte))
1314                         flags |= PM_SOFT_DIRTY;
1315         } else if (is_swap_pte(pte)) {
1316                 swp_entry_t entry;
1317                 if (pte_swp_soft_dirty(pte))
1318                         flags |= PM_SOFT_DIRTY;
1319                 entry = pte_to_swp_entry(pte);
1320                 if (pm->show_pfn)
1321                         frame = swp_type(entry) |
1322                                 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1323                 flags |= PM_SWAP;
1324                 if (is_migration_entry(entry))
1325                         page = migration_entry_to_page(entry);
1326
1327                 if (is_device_private_entry(entry))
1328                         page = device_private_entry_to_page(entry);
1329         }
1330
1331         if (page && !PageAnon(page))
1332                 flags |= PM_FILE;
1333         if (page && page_mapcount(page) == 1)
1334                 flags |= PM_MMAP_EXCLUSIVE;
1335         if (vma->vm_flags & VM_SOFTDIRTY)
1336                 flags |= PM_SOFT_DIRTY;
1337
1338         return make_pme(frame, flags);
1339 }
1340
1341 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1342                              struct mm_walk *walk)
1343 {
1344         struct vm_area_struct *vma = walk->vma;
1345         struct pagemapread *pm = walk->private;
1346         spinlock_t *ptl;
1347         pte_t *pte, *orig_pte;
1348         int err = 0;
1349
1350 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1351         ptl = pmd_trans_huge_lock(pmdp, vma);
1352         if (ptl) {
1353                 u64 flags = 0, frame = 0;
1354                 pmd_t pmd = *pmdp;
1355                 struct page *page = NULL;
1356
1357                 if (vma->vm_flags & VM_SOFTDIRTY)
1358                         flags |= PM_SOFT_DIRTY;
1359
1360                 if (pmd_present(pmd)) {
1361                         page = pmd_page(pmd);
1362
1363                         flags |= PM_PRESENT;
1364                         if (pmd_soft_dirty(pmd))
1365                                 flags |= PM_SOFT_DIRTY;
1366                         if (pm->show_pfn)
1367                                 frame = pmd_pfn(pmd) +
1368                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1369                 }
1370 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1371                 else if (is_swap_pmd(pmd)) {
1372                         swp_entry_t entry = pmd_to_swp_entry(pmd);
1373                         unsigned long offset;
1374
1375                         if (pm->show_pfn) {
1376                                 offset = swp_offset(entry) +
1377                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1378                                 frame = swp_type(entry) |
1379                                         (offset << MAX_SWAPFILES_SHIFT);
1380                         }
1381                         flags |= PM_SWAP;
1382                         if (pmd_swp_soft_dirty(pmd))
1383                                 flags |= PM_SOFT_DIRTY;
1384                         VM_BUG_ON(!is_pmd_migration_entry(pmd));
1385                         page = migration_entry_to_page(entry);
1386                 }
1387 #endif
1388
1389                 if (page && page_mapcount(page) == 1)
1390                         flags |= PM_MMAP_EXCLUSIVE;
1391
1392                 for (; addr != end; addr += PAGE_SIZE) {
1393                         pagemap_entry_t pme = make_pme(frame, flags);
1394
1395                         err = add_to_pagemap(addr, &pme, pm);
1396                         if (err)
1397                                 break;
1398                         if (pm->show_pfn) {
1399                                 if (flags & PM_PRESENT)
1400                                         frame++;
1401                                 else if (flags & PM_SWAP)
1402                                         frame += (1 << MAX_SWAPFILES_SHIFT);
1403                         }
1404                 }
1405                 spin_unlock(ptl);
1406                 return err;
1407         }
1408
1409         if (pmd_trans_unstable(pmdp))
1410                 return 0;
1411 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1412
1413         /*
1414          * We can assume that @vma always points to a valid one and @end never
1415          * goes beyond vma->vm_end.
1416          */
1417         orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1418         for (; addr < end; pte++, addr += PAGE_SIZE) {
1419                 pagemap_entry_t pme;
1420
1421                 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1422                 err = add_to_pagemap(addr, &pme, pm);
1423                 if (err)
1424                         break;
1425         }
1426         pte_unmap_unlock(orig_pte, ptl);
1427
1428         cond_resched();
1429
1430         return err;
1431 }
1432
1433 #ifdef CONFIG_HUGETLB_PAGE
1434 /* This function walks within one hugetlb entry in the single call */
1435 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1436                                  unsigned long addr, unsigned long end,
1437                                  struct mm_walk *walk)
1438 {
1439         struct pagemapread *pm = walk->private;
1440         struct vm_area_struct *vma = walk->vma;
1441         u64 flags = 0, frame = 0;
1442         int err = 0;
1443         pte_t pte;
1444
1445         if (vma->vm_flags & VM_SOFTDIRTY)
1446                 flags |= PM_SOFT_DIRTY;
1447
1448         pte = huge_ptep_get(ptep);
1449         if (pte_present(pte)) {
1450                 struct page *page = pte_page(pte);
1451
1452                 if (!PageAnon(page))
1453                         flags |= PM_FILE;
1454
1455                 if (page_mapcount(page) == 1)
1456                         flags |= PM_MMAP_EXCLUSIVE;
1457
1458                 flags |= PM_PRESENT;
1459                 if (pm->show_pfn)
1460                         frame = pte_pfn(pte) +
1461                                 ((addr & ~hmask) >> PAGE_SHIFT);
1462         }
1463
1464         for (; addr != end; addr += PAGE_SIZE) {
1465                 pagemap_entry_t pme = make_pme(frame, flags);
1466
1467                 err = add_to_pagemap(addr, &pme, pm);
1468                 if (err)
1469                         return err;
1470                 if (pm->show_pfn && (flags & PM_PRESENT))
1471                         frame++;
1472         }
1473
1474         cond_resched();
1475
1476         return err;
1477 }
1478 #else
1479 #define pagemap_hugetlb_range   NULL
1480 #endif /* HUGETLB_PAGE */
1481
1482 static const struct mm_walk_ops pagemap_ops = {
1483         .pmd_entry      = pagemap_pmd_range,
1484         .pte_hole       = pagemap_pte_hole,
1485         .hugetlb_entry  = pagemap_hugetlb_range,
1486 };
1487
1488 /*
1489  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1490  *
1491  * For each page in the address space, this file contains one 64-bit entry
1492  * consisting of the following:
1493  *
1494  * Bits 0-54  page frame number (PFN) if present
1495  * Bits 0-4   swap type if swapped
1496  * Bits 5-54  swap offset if swapped
1497  * Bit  55    pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst)
1498  * Bit  56    page exclusively mapped
1499  * Bits 57-60 zero
1500  * Bit  61    page is file-page or shared-anon
1501  * Bit  62    page swapped
1502  * Bit  63    page present
1503  *
1504  * If the page is not present but in swap, then the PFN contains an
1505  * encoding of the swap file number and the page's offset into the
1506  * swap. Unmapped pages return a null PFN. This allows determining
1507  * precisely which pages are mapped (or in swap) and comparing mapped
1508  * pages between processes.
1509  *
1510  * Efficient users of this interface will use /proc/pid/maps to
1511  * determine which areas of memory are actually mapped and llseek to
1512  * skip over unmapped regions.
1513  */
1514 static ssize_t pagemap_read(struct file *file, char __user *buf,
1515                             size_t count, loff_t *ppos)
1516 {
1517         struct mm_struct *mm = file->private_data;
1518         struct pagemapread pm;
1519         unsigned long src;
1520         unsigned long svpfn;
1521         unsigned long start_vaddr;
1522         unsigned long end_vaddr;
1523         int ret = 0, copied = 0;
1524
1525         if (!mm || !mmget_not_zero(mm))
1526                 goto out;
1527
1528         ret = -EINVAL;
1529         /* file position must be aligned */
1530         if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1531                 goto out_mm;
1532
1533         ret = 0;
1534         if (!count)
1535                 goto out_mm;
1536
1537         /* do not disclose physical addresses: attack vector */
1538         pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1539
1540         pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1541         pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL);
1542         ret = -ENOMEM;
1543         if (!pm.buffer)
1544                 goto out_mm;
1545
1546         src = *ppos;
1547         svpfn = src / PM_ENTRY_BYTES;
1548         start_vaddr = svpfn << PAGE_SHIFT;
1549         end_vaddr = mm->task_size;
1550
1551         /* watch out for wraparound */
1552         if (svpfn > mm->task_size >> PAGE_SHIFT)
1553                 start_vaddr = end_vaddr;
1554
1555         /*
1556          * The odds are that this will stop walking way
1557          * before end_vaddr, because the length of the
1558          * user buffer is tracked in "pm", and the walk
1559          * will stop when we hit the end of the buffer.
1560          */
1561         ret = 0;
1562         while (count && (start_vaddr < end_vaddr)) {
1563                 int len;
1564                 unsigned long end;
1565
1566                 pm.pos = 0;
1567                 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1568                 /* overflow ? */
1569                 if (end < start_vaddr || end > end_vaddr)
1570                         end = end_vaddr;
1571                 ret = mmap_read_lock_killable(mm);
1572                 if (ret)
1573                         goto out_free;
1574                 ret = walk_page_range(mm, start_vaddr, end, &pagemap_ops, &pm);
1575                 mmap_read_unlock(mm);
1576                 start_vaddr = end;
1577
1578                 len = min(count, PM_ENTRY_BYTES * pm.pos);
1579                 if (copy_to_user(buf, pm.buffer, len)) {
1580                         ret = -EFAULT;
1581                         goto out_free;
1582                 }
1583                 copied += len;
1584                 buf += len;
1585                 count -= len;
1586         }
1587         *ppos += copied;
1588         if (!ret || ret == PM_END_OF_BUFFER)
1589                 ret = copied;
1590
1591 out_free:
1592         kfree(pm.buffer);
1593 out_mm:
1594         mmput(mm);
1595 out:
1596         return ret;
1597 }
1598
1599 static int pagemap_open(struct inode *inode, struct file *file)
1600 {
1601         struct mm_struct *mm;
1602
1603         mm = proc_mem_open(inode, PTRACE_MODE_READ);
1604         if (IS_ERR(mm))
1605                 return PTR_ERR(mm);
1606         file->private_data = mm;
1607         return 0;
1608 }
1609
1610 static int pagemap_release(struct inode *inode, struct file *file)
1611 {
1612         struct mm_struct *mm = file->private_data;
1613
1614         if (mm)
1615                 mmdrop(mm);
1616         return 0;
1617 }
1618
1619 const struct file_operations proc_pagemap_operations = {
1620         .llseek         = mem_lseek, /* borrow this */
1621         .read           = pagemap_read,
1622         .open           = pagemap_open,
1623         .release        = pagemap_release,
1624 };
1625 #endif /* CONFIG_PROC_PAGE_MONITOR */
1626
1627 #ifdef CONFIG_NUMA
1628
1629 struct numa_maps {
1630         unsigned long pages;
1631         unsigned long anon;
1632         unsigned long active;
1633         unsigned long writeback;
1634         unsigned long mapcount_max;
1635         unsigned long dirty;
1636         unsigned long swapcache;
1637         unsigned long node[MAX_NUMNODES];
1638 };
1639
1640 struct numa_maps_private {
1641         struct proc_maps_private proc_maps;
1642         struct numa_maps md;
1643 };
1644
1645 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1646                         unsigned long nr_pages)
1647 {
1648         int count = page_mapcount(page);
1649
1650         md->pages += nr_pages;
1651         if (pte_dirty || PageDirty(page))
1652                 md->dirty += nr_pages;
1653
1654         if (PageSwapCache(page))
1655                 md->swapcache += nr_pages;
1656
1657         if (PageActive(page) || PageUnevictable(page))
1658                 md->active += nr_pages;
1659
1660         if (PageWriteback(page))
1661                 md->writeback += nr_pages;
1662
1663         if (PageAnon(page))
1664                 md->anon += nr_pages;
1665
1666         if (count > md->mapcount_max)
1667                 md->mapcount_max = count;
1668
1669         md->node[page_to_nid(page)] += nr_pages;
1670 }
1671
1672 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1673                 unsigned long addr)
1674 {
1675         struct page *page;
1676         int nid;
1677
1678         if (!pte_present(pte))
1679                 return NULL;
1680
1681         page = vm_normal_page(vma, addr, pte);
1682         if (!page)
1683                 return NULL;
1684
1685         if (PageReserved(page))
1686                 return NULL;
1687
1688         nid = page_to_nid(page);
1689         if (!node_isset(nid, node_states[N_MEMORY]))
1690                 return NULL;
1691
1692         return page;
1693 }
1694
1695 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1696 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1697                                               struct vm_area_struct *vma,
1698                                               unsigned long addr)
1699 {
1700         struct page *page;
1701         int nid;
1702
1703         if (!pmd_present(pmd))
1704                 return NULL;
1705
1706         page = vm_normal_page_pmd(vma, addr, pmd);
1707         if (!page)
1708                 return NULL;
1709
1710         if (PageReserved(page))
1711                 return NULL;
1712
1713         nid = page_to_nid(page);
1714         if (!node_isset(nid, node_states[N_MEMORY]))
1715                 return NULL;
1716
1717         return page;
1718 }
1719 #endif
1720
1721 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1722                 unsigned long end, struct mm_walk *walk)
1723 {
1724         struct numa_maps *md = walk->private;
1725         struct vm_area_struct *vma = walk->vma;
1726         spinlock_t *ptl;
1727         pte_t *orig_pte;
1728         pte_t *pte;
1729
1730 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1731         ptl = pmd_trans_huge_lock(pmd, vma);
1732         if (ptl) {
1733                 struct page *page;
1734
1735                 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1736                 if (page)
1737                         gather_stats(page, md, pmd_dirty(*pmd),
1738                                      HPAGE_PMD_SIZE/PAGE_SIZE);
1739                 spin_unlock(ptl);
1740                 return 0;
1741         }
1742
1743         if (pmd_trans_unstable(pmd))
1744                 return 0;
1745 #endif
1746         orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1747         do {
1748                 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1749                 if (!page)
1750                         continue;
1751                 gather_stats(page, md, pte_dirty(*pte), 1);
1752
1753         } while (pte++, addr += PAGE_SIZE, addr != end);
1754         pte_unmap_unlock(orig_pte, ptl);
1755         cond_resched();
1756         return 0;
1757 }
1758 #ifdef CONFIG_HUGETLB_PAGE
1759 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1760                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1761 {
1762         pte_t huge_pte = huge_ptep_get(pte);
1763         struct numa_maps *md;
1764         struct page *page;
1765
1766         if (!pte_present(huge_pte))
1767                 return 0;
1768
1769         page = pte_page(huge_pte);
1770         if (!page)
1771                 return 0;
1772
1773         md = walk->private;
1774         gather_stats(page, md, pte_dirty(huge_pte), 1);
1775         return 0;
1776 }
1777
1778 #else
1779 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1780                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1781 {
1782         return 0;
1783 }
1784 #endif
1785
1786 static const struct mm_walk_ops show_numa_ops = {
1787         .hugetlb_entry = gather_hugetlb_stats,
1788         .pmd_entry = gather_pte_stats,
1789 };
1790
1791 /*
1792  * Display pages allocated per node and memory policy via /proc.
1793  */
1794 static int show_numa_map(struct seq_file *m, void *v)
1795 {
1796         struct numa_maps_private *numa_priv = m->private;
1797         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1798         struct vm_area_struct *vma = v;
1799         struct numa_maps *md = &numa_priv->md;
1800         struct file *file = vma->vm_file;
1801         struct mm_struct *mm = vma->vm_mm;
1802         struct mempolicy *pol;
1803         char buffer[64];
1804         int nid;
1805
1806         if (!mm)
1807                 return 0;
1808
1809         /* Ensure we start with an empty set of numa_maps statistics. */
1810         memset(md, 0, sizeof(*md));
1811
1812         pol = __get_vma_policy(vma, vma->vm_start);
1813         if (pol) {
1814                 mpol_to_str(buffer, sizeof(buffer), pol);
1815                 mpol_cond_put(pol);
1816         } else {
1817                 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1818         }
1819
1820         seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1821
1822         if (file) {
1823                 seq_puts(m, " file=");
1824                 seq_file_path(m, file, "\n\t= ");
1825         } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1826                 seq_puts(m, " heap");
1827         } else if (is_stack(vma)) {
1828                 seq_puts(m, " stack");
1829         }
1830
1831         if (is_vm_hugetlb_page(vma))
1832                 seq_puts(m, " huge");
1833
1834         /* mmap_lock is held by m_start */
1835         walk_page_vma(vma, &show_numa_ops, md);
1836
1837         if (!md->pages)
1838                 goto out;
1839
1840         if (md->anon)
1841                 seq_printf(m, " anon=%lu", md->anon);
1842
1843         if (md->dirty)
1844                 seq_printf(m, " dirty=%lu", md->dirty);
1845
1846         if (md->pages != md->anon && md->pages != md->dirty)
1847                 seq_printf(m, " mapped=%lu", md->pages);
1848
1849         if (md->mapcount_max > 1)
1850                 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1851
1852         if (md->swapcache)
1853                 seq_printf(m, " swapcache=%lu", md->swapcache);
1854
1855         if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1856                 seq_printf(m, " active=%lu", md->active);
1857
1858         if (md->writeback)
1859                 seq_printf(m, " writeback=%lu", md->writeback);
1860
1861         for_each_node_state(nid, N_MEMORY)
1862                 if (md->node[nid])
1863                         seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1864
1865         seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1866 out:
1867         seq_putc(m, '\n');
1868         return 0;
1869 }
1870
1871 static const struct seq_operations proc_pid_numa_maps_op = {
1872         .start  = m_start,
1873         .next   = m_next,
1874         .stop   = m_stop,
1875         .show   = show_numa_map,
1876 };
1877
1878 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1879 {
1880         return proc_maps_open(inode, file, &proc_pid_numa_maps_op,
1881                                 sizeof(struct numa_maps_private));
1882 }
1883
1884 const struct file_operations proc_pid_numa_maps_operations = {
1885         .open           = pid_numa_maps_open,
1886         .read           = seq_read,
1887         .llseek         = seq_lseek,
1888         .release        = proc_map_release,
1889 };
1890
1891 #endif /* CONFIG_NUMA */