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_sem 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_ARCH_HAS_PKEYS
 657                 /* These come out via ProtectionKey: */
 658                 [ilog2(VM_PKEY_BIT0)]   = "",
 659                 [ilog2(VM_PKEY_BIT1)]   = "",
 660                 [ilog2(VM_PKEY_BIT2)]   = "",
 661                 [ilog2(VM_PKEY_BIT3)]   = "",
 662 #if VM_PKEY_BIT4
 663                 [ilog2(VM_PKEY_BIT4)]   = "",
 664 #endif
 665 #endif /* CONFIG_ARCH_HAS_PKEYS */
 666         };
 667         size_t i;
 668
 669         seq_puts(m, "VmFlags: ");
 670         for (i = 0; i < BITS_PER_LONG; i++) {
 671                 if (!mnemonics[i][0])
 672                         continue;
 673                 if (vma->vm_flags & (1UL << i)) {
 674                         seq_putc(m, mnemonics[i][0]);
 675                         seq_putc(m, mnemonics[i][1]);
 676                         seq_putc(m, ' ');
 677                 }
 678         }
 679         seq_putc(m, '\n');
 680 }
 681
 682 #ifdef CONFIG_HUGETLB_PAGE
 683 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
 684                                  unsigned long addr, unsigned long end,
 685                                  struct mm_walk *walk)
 686 {
 687         struct mem_size_stats *mss = walk->private;
 688         struct vm_area_struct *vma = walk->vma;
 689         struct page *page = NULL;
 690
 691         if (pte_present(*pte)) {
 692                 page = vm_normal_page(vma, addr, *pte);
 693         } else if (is_swap_pte(*pte)) {
 694                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 695
 696                 if (is_migration_entry(swpent))
 697                         page = migration_entry_to_page(swpent);
 698                 else if (is_device_private_entry(swpent))
 699                         page = device_private_entry_to_page(swpent);
 700         }
 701         if (page) {
 702                 int mapcount = page_mapcount(page);
 703
 704                 if (mapcount >= 2)
 705                         mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
 706                 else
 707                         mss->private_hugetlb += huge_page_size(hstate_vma(vma));
 708         }
 709         return 0;
 710 }
 711 #else
 712 #define smaps_hugetlb_range     NULL
 713 #endif /* HUGETLB_PAGE */
 714
 715 static const struct mm_walk_ops smaps_walk_ops = {
 716         .pmd_entry              = smaps_pte_range,
 717         .hugetlb_entry          = smaps_hugetlb_range,
 718 };
 719
 720 static const struct mm_walk_ops smaps_shmem_walk_ops = {
 721         .pmd_entry              = smaps_pte_range,
 722         .hugetlb_entry          = smaps_hugetlb_range,
 723         .pte_hole               = smaps_pte_hole,
 724 };
 725
 726 static void smap_gather_stats(struct vm_area_struct *vma,
 727                              struct mem_size_stats *mss)
 728 {
 729 #ifdef CONFIG_SHMEM
 730         /* In case of smaps_rollup, reset the value from previous vma */
 731         mss->check_shmem_swap = false;
 732         if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
 733                 /*
 734                  * For shared or readonly shmem mappings we know that all
 735                  * swapped out pages belong to the shmem object, and we can
 736                  * obtain the swap value much more efficiently. For private
 737                  * writable mappings, we might have COW pages that are
 738                  * not affected by the parent swapped out pages of the shmem
 739                  * object, so we have to distinguish them during the page walk.
 740                  * Unless we know that the shmem object (or the part mapped by
 741                  * our VMA) has no swapped out pages at all.
 742                  */
 743                 unsigned long shmem_swapped = shmem_swap_usage(vma);
 744
 745                 if (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
 746                                         !(vma->vm_flags & VM_WRITE)) {
 747                         mss->swap += shmem_swapped;
 748                 } else {
 749                         mss->check_shmem_swap = true;
 750                         walk_page_vma(vma, &smaps_shmem_walk_ops, mss);
 751                         return;
 752                 }
 753         }
 754 #endif
 755         /* mmap_sem is held in m_start */
 756         walk_page_vma(vma, &smaps_walk_ops, mss);
 757 }
 758
 759 #define SEQ_PUT_DEC(str, val) \
 760                 seq_put_decimal_ull_width(m, str, (val) >> 10, 8)
 761
 762 /* Show the contents common for smaps and smaps_rollup */
 763 static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss,
 764         bool rollup_mode)
 765 {
 766         SEQ_PUT_DEC("Rss:            ", mss->resident);
 767         SEQ_PUT_DEC(" kB\nPss:            ", mss->pss >> PSS_SHIFT);
 768         if (rollup_mode) {
 769                 /*
 770                  * These are meaningful only for smaps_rollup, otherwise two of
 771                  * them are zero, and the other one is the same as Pss.
 772                  */
 773                 SEQ_PUT_DEC(" kB\nPss_Anon:       ",
 774                         mss->pss_anon >> PSS_SHIFT);
 775                 SEQ_PUT_DEC(" kB\nPss_File:       ",
 776                         mss->pss_file >> PSS_SHIFT);
 777                 SEQ_PUT_DEC(" kB\nPss_Shmem:      ",
 778                         mss->pss_shmem >> PSS_SHIFT);
 779         }
 780         SEQ_PUT_DEC(" kB\nShared_Clean:   ", mss->shared_clean);
 781         SEQ_PUT_DEC(" kB\nShared_Dirty:   ", mss->shared_dirty);
 782         SEQ_PUT_DEC(" kB\nPrivate_Clean:  ", mss->private_clean);
 783         SEQ_PUT_DEC(" kB\nPrivate_Dirty:  ", mss->private_dirty);
 784         SEQ_PUT_DEC(" kB\nReferenced:     ", mss->referenced);
 785         SEQ_PUT_DEC(" kB\nAnonymous:      ", mss->anonymous);
 786         SEQ_PUT_DEC(" kB\nLazyFree:       ", mss->lazyfree);
 787         SEQ_PUT_DEC(" kB\nAnonHugePages:  ", mss->anonymous_thp);
 788         SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp);
 789         SEQ_PUT_DEC(" kB\nFilePmdMapped: ", mss->file_thp);
 790         SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb);
 791         seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ",
 792                                   mss->private_hugetlb >> 10, 7);
 793         SEQ_PUT_DEC(" kB\nSwap:           ", mss->swap);
 794         SEQ_PUT_DEC(" kB\nSwapPss:        ",
 795                                         mss->swap_pss >> PSS_SHIFT);
 796         SEQ_PUT_DEC(" kB\nLocked:         ",
 797                                         mss->pss_locked >> PSS_SHIFT);
 798         seq_puts(m, " kB\n");
 799 }
 800
 801 static int show_smap(struct seq_file *m, void *v)
 802 {
 803         struct vm_area_struct *vma = v;
 804         struct mem_size_stats mss;
 805
 806         memset(&mss, 0, sizeof(mss));
 807
 808         smap_gather_stats(vma, &mss);
 809
 810         show_map_vma(m, vma);
 811
 812         SEQ_PUT_DEC("Size:           ", vma->vm_end - vma->vm_start);
 813         SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma));
 814         SEQ_PUT_DEC(" kB\nMMUPageSize:    ", vma_mmu_pagesize(vma));
 815         seq_puts(m, " kB\n");
 816
 817         __show_smap(m, &mss, false);
 818
 819         seq_printf(m, "THPeligible:             %d\n",
 820                    transparent_hugepage_enabled(vma));
 821
 822         if (arch_pkeys_enabled())
 823                 seq_printf(m, "ProtectionKey:  %8u\n", vma_pkey(vma));
 824         show_smap_vma_flags(m, vma);
 825
 826         return 0;
 827 }
 828
 829 static int show_smaps_rollup(struct seq_file *m, void *v)
 830 {
 831         struct proc_maps_private *priv = m->private;
 832         struct mem_size_stats mss;
 833         struct mm_struct *mm;
 834         struct vm_area_struct *vma;
 835         unsigned long last_vma_end = 0;
 836         int ret = 0;
 837
 838         priv->task = get_proc_task(priv->inode);
 839         if (!priv->task)
 840                 return -ESRCH;
 841
 842         mm = priv->mm;
 843         if (!mm || !mmget_not_zero(mm)) {
 844                 ret = -ESRCH;
 845                 goto out_put_task;
 846         }
 847
 848         memset(&mss, 0, sizeof(mss));
 849
 850         ret = mmap_read_lock_killable(mm);
 851         if (ret)
 852                 goto out_put_mm;
 853
 854         hold_task_mempolicy(priv);
 855
 856         for (vma = priv->mm->mmap; vma; vma = vma->vm_next) {
 857                 smap_gather_stats(vma, &mss);
 858                 last_vma_end = vma->vm_end;
 859         }
 860
 861         show_vma_header_prefix(m, priv->mm->mmap->vm_start,
 862                                last_vma_end, 0, 0, 0, 0);
 863         seq_pad(m, ' ');
 864         seq_puts(m, "[rollup]\n");
 865
 866         __show_smap(m, &mss, true);
 867
 868         release_task_mempolicy(priv);
 869         mmap_read_unlock(mm);
 870
 871 out_put_mm:
 872         mmput(mm);
 873 out_put_task:
 874         put_task_struct(priv->task);
 875         priv->task = NULL;
 876
 877         return ret;
 878 }
 879 #undef SEQ_PUT_DEC
 880
 881 static const struct seq_operations proc_pid_smaps_op = {
 882         .start  = m_start,
 883         .next   = m_next,
 884         .stop   = m_stop,
 885         .show   = show_smap
 886 };
 887
 888 static int pid_smaps_open(struct inode *inode, struct file *file)
 889 {
 890         return do_maps_open(inode, file, &proc_pid_smaps_op);
 891 }
 892
 893 static int smaps_rollup_open(struct inode *inode, struct file *file)
 894 {
 895         int ret;
 896         struct proc_maps_private *priv;
 897
 898         priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT);
 899         if (!priv)
 900                 return -ENOMEM;
 901
 902         ret = single_open(file, show_smaps_rollup, priv);
 903         if (ret)
 904                 goto out_free;
 905
 906         priv->inode = inode;
 907         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
 908         if (IS_ERR(priv->mm)) {
 909                 ret = PTR_ERR(priv->mm);
 910
 911                 single_release(inode, file);
 912                 goto out_free;
 913         }
 914
 915         return 0;
 916
 917 out_free:
 918         kfree(priv);
 919         return ret;
 920 }
 921
 922 static int smaps_rollup_release(struct inode *inode, struct file *file)
 923 {
 924         struct seq_file *seq = file->private_data;
 925         struct proc_maps_private *priv = seq->private;
 926
 927         if (priv->mm)
 928                 mmdrop(priv->mm);
 929
 930         kfree(priv);
 931         return single_release(inode, file);
 932 }
 933
 934 const struct file_operations proc_pid_smaps_operations = {
 935         .open           = pid_smaps_open,
 936         .read           = seq_read,
 937         .llseek         = seq_lseek,
 938         .release        = proc_map_release,
 939 };
 940
 941 const struct file_operations proc_pid_smaps_rollup_operations = {
 942         .open           = smaps_rollup_open,
 943         .read           = seq_read,
 944         .llseek         = seq_lseek,
 945         .release        = smaps_rollup_release,
 946 };
 947
 948 enum clear_refs_types {
 949         CLEAR_REFS_ALL = 1,
 950         CLEAR_REFS_ANON,
 951         CLEAR_REFS_MAPPED,
 952         CLEAR_REFS_SOFT_DIRTY,
 953         CLEAR_REFS_MM_HIWATER_RSS,
 954         CLEAR_REFS_LAST,
 955 };
 956
 957 struct clear_refs_private {
 958         enum clear_refs_types type;
 959 };
 960
 961 #ifdef CONFIG_MEM_SOFT_DIRTY
 962 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 963                 unsigned long addr, pte_t *pte)
 964 {
 965         /*
 966          * The soft-dirty tracker uses #PF-s to catch writes
 967          * to pages, so write-protect the pte as well. See the
 968          * Documentation/admin-guide/mm/soft-dirty.rst for full description
 969          * of how soft-dirty works.
 970          */
 971         pte_t ptent = *pte;
 972
 973         if (pte_present(ptent)) {
 974                 pte_t old_pte;
 975
 976                 old_pte = ptep_modify_prot_start(vma, addr, pte);
 977                 ptent = pte_wrprotect(old_pte);
 978                 ptent = pte_clear_soft_dirty(ptent);
 979                 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent);
 980         } else if (is_swap_pte(ptent)) {
 981                 ptent = pte_swp_clear_soft_dirty(ptent);
 982                 set_pte_at(vma->vm_mm, addr, pte, ptent);
 983         }
 984 }
 985 #else
 986 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 987                 unsigned long addr, pte_t *pte)
 988 {
 989 }
 990 #endif
 991
 992 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
 993 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
 994                 unsigned long addr, pmd_t *pmdp)
 995 {
 996         pmd_t old, pmd = *pmdp;
 997
 998         if (pmd_present(pmd)) {
 999                 /* See comment in change_huge_pmd() */
1000                 old = pmdp_invalidate(vma, addr, pmdp);
1001                 if (pmd_dirty(old))
1002                         pmd = pmd_mkdirty(pmd);
1003                 if (pmd_young(old))
1004                         pmd = pmd_mkyoung(pmd);
1005
1006                 pmd = pmd_wrprotect(pmd);
1007                 pmd = pmd_clear_soft_dirty(pmd);
1008
1009                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1010         } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1011                 pmd = pmd_swp_clear_soft_dirty(pmd);
1012                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1013         }
1014 }
1015 #else
1016 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1017                 unsigned long addr, pmd_t *pmdp)
1018 {
1019 }
1020 #endif
1021
1022 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
1023                                 unsigned long end, struct mm_walk *walk)
1024 {
1025         struct clear_refs_private *cp = walk->private;
1026         struct vm_area_struct *vma = walk->vma;
1027         pte_t *pte, ptent;
1028         spinlock_t *ptl;
1029         struct page *page;
1030
1031         ptl = pmd_trans_huge_lock(pmd, vma);
1032         if (ptl) {
1033                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1034                         clear_soft_dirty_pmd(vma, addr, pmd);
1035                         goto out;
1036                 }
1037
1038                 if (!pmd_present(*pmd))
1039                         goto out;
1040
1041                 page = pmd_page(*pmd);
1042
1043                 /* Clear accessed and referenced bits. */
1044                 pmdp_test_and_clear_young(vma, addr, pmd);
1045                 test_and_clear_page_young(page);
1046                 ClearPageReferenced(page);
1047 out:
1048                 spin_unlock(ptl);
1049                 return 0;
1050         }
1051
1052         if (pmd_trans_unstable(pmd))
1053                 return 0;
1054
1055         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1056         for (; addr != end; pte++, addr += PAGE_SIZE) {
1057                 ptent = *pte;
1058
1059                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1060                         clear_soft_dirty(vma, addr, pte);
1061                         continue;
1062                 }
1063
1064                 if (!pte_present(ptent))
1065                         continue;
1066
1067                 page = vm_normal_page(vma, addr, ptent);
1068                 if (!page)
1069                         continue;
1070
1071                 /* Clear accessed and referenced bits. */
1072                 ptep_test_and_clear_young(vma, addr, pte);
1073                 test_and_clear_page_young(page);
1074                 ClearPageReferenced(page);
1075         }
1076         pte_unmap_unlock(pte - 1, ptl);
1077         cond_resched();
1078         return 0;
1079 }
1080
1081 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1082                                 struct mm_walk *walk)
1083 {
1084         struct clear_refs_private *cp = walk->private;
1085         struct vm_area_struct *vma = walk->vma;
1086
1087         if (vma->vm_flags & VM_PFNMAP)
1088                 return 1;
1089
1090         /*
1091          * Writing 1 to /proc/pid/clear_refs affects all pages.
1092          * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1093          * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1094          * Writing 4 to /proc/pid/clear_refs affects all pages.
1095          */
1096         if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1097                 return 1;
1098         if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1099                 return 1;
1100         return 0;
1101 }
1102
1103 static const struct mm_walk_ops clear_refs_walk_ops = {
1104         .pmd_entry              = clear_refs_pte_range,
1105         .test_walk              = clear_refs_test_walk,
1106 };
1107
1108 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1109                                 size_t count, loff_t *ppos)
1110 {
1111         struct task_struct *task;
1112         char buffer[PROC_NUMBUF];
1113         struct mm_struct *mm;
1114         struct vm_area_struct *vma;
1115         enum clear_refs_types type;
1116         struct mmu_gather tlb;
1117         int itype;
1118         int rv;
1119
1120         memset(buffer, 0, sizeof(buffer));
1121         if (count > sizeof(buffer) - 1)
1122                 count = sizeof(buffer) - 1;
1123         if (copy_from_user(buffer, buf, count))
1124                 return -EFAULT;
1125         rv = kstrtoint(strstrip(buffer), 10, &itype);
1126         if (rv < 0)
1127                 return rv;
1128         type = (enum clear_refs_types)itype;
1129         if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1130                 return -EINVAL;
1131
1132         task = get_proc_task(file_inode(file));
1133         if (!task)
1134                 return -ESRCH;
1135         mm = get_task_mm(task);
1136         if (mm) {
1137                 struct mmu_notifier_range range;
1138                 struct clear_refs_private cp = {
1139                         .type = type,
1140                 };
1141
1142                 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1143                         if (mmap_write_lock_killable(mm)) {
1144                                 count = -EINTR;
1145                                 goto out_mm;
1146                         }
1147
1148                         /*
1149                          * Writing 5 to /proc/pid/clear_refs resets the peak
1150                          * resident set size to this mm's current rss value.
1151                          */
1152                         reset_mm_hiwater_rss(mm);
1153                         mmap_write_unlock(mm);
1154                         goto out_mm;
1155                 }
1156
1157                 if (mmap_read_lock_killable(mm)) {
1158                         count = -EINTR;
1159                         goto out_mm;
1160                 }
1161                 tlb_gather_mmu(&tlb, mm, 0, -1);
1162                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1163                         for (vma = mm->mmap; vma; vma = vma->vm_next) {
1164                                 if (!(vma->vm_flags & VM_SOFTDIRTY))
1165                                         continue;
1166                                 mmap_read_unlock(mm);
1167                                 if (mmap_write_lock_killable(mm)) {
1168                                         count = -EINTR;
1169                                         goto out_mm;
1170                                 }
1171                                 /*
1172                                  * Avoid to modify vma->vm_flags
1173                                  * without locked ops while the
1174                                  * coredump reads the vm_flags.
1175                                  */
1176                                 if (!mmget_still_valid(mm)) {
1177                                         /*
1178                                          * Silently return "count"
1179                                          * like if get_task_mm()
1180                                          * failed. FIXME: should this
1181                                          * function have returned
1182                                          * -ESRCH if get_task_mm()
1183                                          * failed like if
1184                                          * get_proc_task() fails?
1185                                          */
1186                                         mmap_write_unlock(mm);
1187                                         goto out_mm;
1188                                 }
1189                                 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1190                                         vma->vm_flags &= ~VM_SOFTDIRTY;
1191                                         vma_set_page_prot(vma);
1192                                 }
1193                                 mmap_write_downgrade(mm);
1194                                 break;
1195                         }
1196
1197                         mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY,
1198                                                 0, NULL, mm, 0, -1UL);
1199                         mmu_notifier_invalidate_range_start(&range);
1200                 }
1201                 walk_page_range(mm, 0, mm->highest_vm_end, &clear_refs_walk_ops,
1202                                 &cp);
1203                 if (type == CLEAR_REFS_SOFT_DIRTY)
1204                         mmu_notifier_invalidate_range_end(&range);
1205                 tlb_finish_mmu(&tlb, 0, -1);
1206                 mmap_read_unlock(mm);
1207 out_mm:
1208                 mmput(mm);
1209         }
1210         put_task_struct(task);
1211
1212         return count;
1213 }
1214
1215 const struct file_operations proc_clear_refs_operations = {
1216         .write          = clear_refs_write,
1217         .llseek         = noop_llseek,
1218 };
1219
1220 typedef struct {
1221         u64 pme;
1222 } pagemap_entry_t;
1223
1224 struct pagemapread {
1225         int pos, len;           /* units: PM_ENTRY_BYTES, not bytes */
1226         pagemap_entry_t *buffer;
1227         bool show_pfn;
1228 };
1229
1230 #define PAGEMAP_WALK_SIZE       (PMD_SIZE)
1231 #define PAGEMAP_WALK_MASK       (PMD_MASK)
1232
1233 #define PM_ENTRY_BYTES          sizeof(pagemap_entry_t)
1234 #define PM_PFRAME_BITS          55
1235 #define PM_PFRAME_MASK          GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1236 #define PM_SOFT_DIRTY           BIT_ULL(55)
1237 #define PM_MMAP_EXCLUSIVE       BIT_ULL(56)
1238 #define PM_FILE                 BIT_ULL(61)
1239 #define PM_SWAP                 BIT_ULL(62)
1240 #define PM_PRESENT              BIT_ULL(63)
1241
1242 #define PM_END_OF_BUFFER    1
1243
1244 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1245 {
1246         return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1247 }
1248
1249 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1250                           struct pagemapread *pm)
1251 {
1252         pm->buffer[pm->pos++] = *pme;
1253         if (pm->pos >= pm->len)
1254                 return PM_END_OF_BUFFER;
1255         return 0;
1256 }
1257
1258 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1259                             __always_unused int depth, struct mm_walk *walk)
1260 {
1261         struct pagemapread *pm = walk->private;
1262         unsigned long addr = start;
1263         int err = 0;
1264
1265         while (addr < end) {
1266                 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1267                 pagemap_entry_t pme = make_pme(0, 0);
1268                 /* End of address space hole, which we mark as non-present. */
1269                 unsigned long hole_end;
1270
1271                 if (vma)
1272                         hole_end = min(end, vma->vm_start);
1273                 else
1274                         hole_end = end;
1275
1276                 for (; addr < hole_end; addr += PAGE_SIZE) {
1277                         err = add_to_pagemap(addr, &pme, pm);
1278                         if (err)
1279                                 goto out;
1280                 }
1281
1282                 if (!vma)
1283                         break;
1284
1285                 /* Addresses in the VMA. */
1286                 if (vma->vm_flags & VM_SOFTDIRTY)
1287                         pme = make_pme(0, PM_SOFT_DIRTY);
1288                 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1289                         err = add_to_pagemap(addr, &pme, pm);
1290                         if (err)
1291                                 goto out;
1292                 }
1293         }
1294 out:
1295         return err;
1296 }
1297
1298 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1299                 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1300 {
1301         u64 frame = 0, flags = 0;
1302         struct page *page = NULL;
1303
1304         if (pte_present(pte)) {
1305                 if (pm->show_pfn)
1306                         frame = pte_pfn(pte);
1307                 flags |= PM_PRESENT;
1308                 page = vm_normal_page(vma, addr, pte);
1309                 if (pte_soft_dirty(pte))
1310                         flags |= PM_SOFT_DIRTY;
1311         } else if (is_swap_pte(pte)) {
1312                 swp_entry_t entry;
1313                 if (pte_swp_soft_dirty(pte))
1314                         flags |= PM_SOFT_DIRTY;
1315                 entry = pte_to_swp_entry(pte);
1316                 if (pm->show_pfn)
1317                         frame = swp_type(entry) |
1318                                 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1319                 flags |= PM_SWAP;
1320                 if (is_migration_entry(entry))
1321                         page = migration_entry_to_page(entry);
1322
1323                 if (is_device_private_entry(entry))
1324                         page = device_private_entry_to_page(entry);
1325         }
1326
1327         if (page && !PageAnon(page))
1328                 flags |= PM_FILE;
1329         if (page && page_mapcount(page) == 1)
1330                 flags |= PM_MMAP_EXCLUSIVE;
1331         if (vma->vm_flags & VM_SOFTDIRTY)
1332                 flags |= PM_SOFT_DIRTY;
1333
1334         return make_pme(frame, flags);
1335 }
1336
1337 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1338                              struct mm_walk *walk)
1339 {
1340         struct vm_area_struct *vma = walk->vma;
1341         struct pagemapread *pm = walk->private;
1342         spinlock_t *ptl;
1343         pte_t *pte, *orig_pte;
1344         int err = 0;
1345
1346 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1347         ptl = pmd_trans_huge_lock(pmdp, vma);
1348         if (ptl) {
1349                 u64 flags = 0, frame = 0;
1350                 pmd_t pmd = *pmdp;
1351                 struct page *page = NULL;
1352
1353                 if (vma->vm_flags & VM_SOFTDIRTY)
1354                         flags |= PM_SOFT_DIRTY;
1355
1356                 if (pmd_present(pmd)) {
1357                         page = pmd_page(pmd);
1358
1359                         flags |= PM_PRESENT;
1360                         if (pmd_soft_dirty(pmd))
1361                                 flags |= PM_SOFT_DIRTY;
1362                         if (pm->show_pfn)
1363                                 frame = pmd_pfn(pmd) +
1364                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1365                 }
1366 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1367                 else if (is_swap_pmd(pmd)) {
1368                         swp_entry_t entry = pmd_to_swp_entry(pmd);
1369                         unsigned long offset;
1370
1371                         if (pm->show_pfn) {
1372                                 offset = swp_offset(entry) +
1373                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1374                                 frame = swp_type(entry) |
1375                                         (offset << MAX_SWAPFILES_SHIFT);
1376                         }
1377                         flags |= PM_SWAP;
1378                         if (pmd_swp_soft_dirty(pmd))
1379                                 flags |= PM_SOFT_DIRTY;
1380                         VM_BUG_ON(!is_pmd_migration_entry(pmd));
1381                         page = migration_entry_to_page(entry);
1382                 }
1383 #endif
1384
1385                 if (page && page_mapcount(page) == 1)
1386                         flags |= PM_MMAP_EXCLUSIVE;
1387
1388                 for (; addr != end; addr += PAGE_SIZE) {
1389                         pagemap_entry_t pme = make_pme(frame, flags);
1390
1391                         err = add_to_pagemap(addr, &pme, pm);
1392                         if (err)
1393                                 break;
1394                         if (pm->show_pfn) {
1395                                 if (flags & PM_PRESENT)
1396                                         frame++;
1397                                 else if (flags & PM_SWAP)
1398                                         frame += (1 << MAX_SWAPFILES_SHIFT);
1399                         }
1400                 }
1401                 spin_unlock(ptl);
1402                 return err;
1403         }
1404
1405         if (pmd_trans_unstable(pmdp))
1406                 return 0;
1407 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1408
1409         /*
1410          * We can assume that @vma always points to a valid one and @end never
1411          * goes beyond vma->vm_end.
1412          */
1413         orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1414         for (; addr < end; pte++, addr += PAGE_SIZE) {
1415                 pagemap_entry_t pme;
1416
1417                 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1418                 err = add_to_pagemap(addr, &pme, pm);
1419                 if (err)
1420                         break;
1421         }
1422         pte_unmap_unlock(orig_pte, ptl);
1423
1424         cond_resched();
1425
1426         return err;
1427 }
1428
1429 #ifdef CONFIG_HUGETLB_PAGE
1430 /* This function walks within one hugetlb entry in the single call */
1431 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1432                                  unsigned long addr, unsigned long end,
1433                                  struct mm_walk *walk)
1434 {
1435         struct pagemapread *pm = walk->private;
1436         struct vm_area_struct *vma = walk->vma;
1437         u64 flags = 0, frame = 0;
1438         int err = 0;
1439         pte_t pte;
1440
1441         if (vma->vm_flags & VM_SOFTDIRTY)
1442                 flags |= PM_SOFT_DIRTY;
1443
1444         pte = huge_ptep_get(ptep);
1445         if (pte_present(pte)) {
1446                 struct page *page = pte_page(pte);
1447
1448                 if (!PageAnon(page))
1449                         flags |= PM_FILE;
1450
1451                 if (page_mapcount(page) == 1)
1452                         flags |= PM_MMAP_EXCLUSIVE;
1453
1454                 flags |= PM_PRESENT;
1455                 if (pm->show_pfn)
1456                         frame = pte_pfn(pte) +
1457                                 ((addr & ~hmask) >> PAGE_SHIFT);
1458         }
1459
1460         for (; addr != end; addr += PAGE_SIZE) {
1461                 pagemap_entry_t pme = make_pme(frame, flags);
1462
1463                 err = add_to_pagemap(addr, &pme, pm);
1464                 if (err)
1465                         return err;
1466                 if (pm->show_pfn && (flags & PM_PRESENT))
1467                         frame++;
1468         }
1469
1470         cond_resched();
1471
1472         return err;
1473 }
1474 #else
1475 #define pagemap_hugetlb_range   NULL
1476 #endif /* HUGETLB_PAGE */
1477
1478 static const struct mm_walk_ops pagemap_ops = {
1479         .pmd_entry      = pagemap_pmd_range,
1480         .pte_hole       = pagemap_pte_hole,
1481         .hugetlb_entry  = pagemap_hugetlb_range,
1482 };
1483
1484 /*
1485  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1486  *
1487  * For each page in the address space, this file contains one 64-bit entry
1488  * consisting of the following:
1489  *
1490  * Bits 0-54  page frame number (PFN) if present
1491  * Bits 0-4   swap type if swapped
1492  * Bits 5-54  swap offset if swapped
1493  * Bit  55    pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst)
1494  * Bit  56    page exclusively mapped
1495  * Bits 57-60 zero
1496  * Bit  61    page is file-page or shared-anon
1497  * Bit  62    page swapped
1498  * Bit  63    page present
1499  *
1500  * If the page is not present but in swap, then the PFN contains an
1501  * encoding of the swap file number and the page's offset into the
1502  * swap. Unmapped pages return a null PFN. This allows determining
1503  * precisely which pages are mapped (or in swap) and comparing mapped
1504  * pages between processes.
1505  *
1506  * Efficient users of this interface will use /proc/pid/maps to
1507  * determine which areas of memory are actually mapped and llseek to
1508  * skip over unmapped regions.
1509  */
1510 static ssize_t pagemap_read(struct file *file, char __user *buf,
1511                             size_t count, loff_t *ppos)
1512 {
1513         struct mm_struct *mm = file->private_data;
1514         struct pagemapread pm;
1515         unsigned long src;
1516         unsigned long svpfn;
1517         unsigned long start_vaddr;
1518         unsigned long end_vaddr;
1519         int ret = 0, copied = 0;
1520
1521         if (!mm || !mmget_not_zero(mm))
1522                 goto out;
1523
1524         ret = -EINVAL;
1525         /* file position must be aligned */
1526         if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1527                 goto out_mm;
1528
1529         ret = 0;
1530         if (!count)
1531                 goto out_mm;
1532
1533         /* do not disclose physical addresses: attack vector */
1534         pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1535
1536         pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1537         pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL);
1538         ret = -ENOMEM;
1539         if (!pm.buffer)
1540                 goto out_mm;
1541
1542         src = *ppos;
1543         svpfn = src / PM_ENTRY_BYTES;
1544         start_vaddr = svpfn << PAGE_SHIFT;
1545         end_vaddr = mm->task_size;
1546
1547         /* watch out for wraparound */
1548         if (svpfn > mm->task_size >> PAGE_SHIFT)
1549                 start_vaddr = end_vaddr;
1550
1551         /*
1552          * The odds are that this will stop walking way
1553          * before end_vaddr, because the length of the
1554          * user buffer is tracked in "pm", and the walk
1555          * will stop when we hit the end of the buffer.
1556          */
1557         ret = 0;
1558         while (count && (start_vaddr < end_vaddr)) {
1559                 int len;
1560                 unsigned long end;
1561
1562                 pm.pos = 0;
1563                 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1564                 /* overflow ? */
1565                 if (end < start_vaddr || end > end_vaddr)
1566                         end = end_vaddr;
1567                 ret = mmap_read_lock_killable(mm);
1568                 if (ret)
1569                         goto out_free;
1570                 ret = walk_page_range(mm, start_vaddr, end, &pagemap_ops, &pm);
1571                 mmap_read_unlock(mm);
1572                 start_vaddr = end;
1573
1574                 len = min(count, PM_ENTRY_BYTES * pm.pos);
1575                 if (copy_to_user(buf, pm.buffer, len)) {
1576                         ret = -EFAULT;
1577                         goto out_free;
1578                 }
1579                 copied += len;
1580                 buf += len;
1581                 count -= len;
1582         }
1583         *ppos += copied;
1584         if (!ret || ret == PM_END_OF_BUFFER)
1585                 ret = copied;
1586
1587 out_free:
1588         kfree(pm.buffer);
1589 out_mm:
1590         mmput(mm);
1591 out:
1592         return ret;
1593 }
1594
1595 static int pagemap_open(struct inode *inode, struct file *file)
1596 {
1597         struct mm_struct *mm;
1598
1599         mm = proc_mem_open(inode, PTRACE_MODE_READ);
1600         if (IS_ERR(mm))
1601                 return PTR_ERR(mm);
1602         file->private_data = mm;
1603         return 0;
1604 }
1605
1606 static int pagemap_release(struct inode *inode, struct file *file)
1607 {
1608         struct mm_struct *mm = file->private_data;
1609
1610         if (mm)
1611                 mmdrop(mm);
1612         return 0;
1613 }
1614
1615 const struct file_operations proc_pagemap_operations = {
1616         .llseek         = mem_lseek, /* borrow this */
1617         .read           = pagemap_read,
1618         .open           = pagemap_open,
1619         .release        = pagemap_release,
1620 };
1621 #endif /* CONFIG_PROC_PAGE_MONITOR */
1622
1623 #ifdef CONFIG_NUMA
1624
1625 struct numa_maps {
1626         unsigned long pages;
1627         unsigned long anon;
1628         unsigned long active;
1629         unsigned long writeback;
1630         unsigned long mapcount_max;
1631         unsigned long dirty;
1632         unsigned long swapcache;
1633         unsigned long node[MAX_NUMNODES];
1634 };
1635
1636 struct numa_maps_private {
1637         struct proc_maps_private proc_maps;
1638         struct numa_maps md;
1639 };
1640
1641 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1642                         unsigned long nr_pages)
1643 {
1644         int count = page_mapcount(page);
1645
1646         md->pages += nr_pages;
1647         if (pte_dirty || PageDirty(page))
1648                 md->dirty += nr_pages;
1649
1650         if (PageSwapCache(page))
1651                 md->swapcache += nr_pages;
1652
1653         if (PageActive(page) || PageUnevictable(page))
1654                 md->active += nr_pages;
1655
1656         if (PageWriteback(page))
1657                 md->writeback += nr_pages;
1658
1659         if (PageAnon(page))
1660                 md->anon += nr_pages;
1661
1662         if (count > md->mapcount_max)
1663                 md->mapcount_max = count;
1664
1665         md->node[page_to_nid(page)] += nr_pages;
1666 }
1667
1668 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1669                 unsigned long addr)
1670 {
1671         struct page *page;
1672         int nid;
1673
1674         if (!pte_present(pte))
1675                 return NULL;
1676
1677         page = vm_normal_page(vma, addr, pte);
1678         if (!page)
1679                 return NULL;
1680
1681         if (PageReserved(page))
1682                 return NULL;
1683
1684         nid = page_to_nid(page);
1685         if (!node_isset(nid, node_states[N_MEMORY]))
1686                 return NULL;
1687
1688         return page;
1689 }
1690
1691 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1692 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1693                                               struct vm_area_struct *vma,
1694                                               unsigned long addr)
1695 {
1696         struct page *page;
1697         int nid;
1698
1699         if (!pmd_present(pmd))
1700                 return NULL;
1701
1702         page = vm_normal_page_pmd(vma, addr, pmd);
1703         if (!page)
1704                 return NULL;
1705
1706         if (PageReserved(page))
1707                 return NULL;
1708
1709         nid = page_to_nid(page);
1710         if (!node_isset(nid, node_states[N_MEMORY]))
1711                 return NULL;
1712
1713         return page;
1714 }
1715 #endif
1716
1717 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1718                 unsigned long end, struct mm_walk *walk)
1719 {
1720         struct numa_maps *md = walk->private;
1721         struct vm_area_struct *vma = walk->vma;
1722         spinlock_t *ptl;
1723         pte_t *orig_pte;
1724         pte_t *pte;
1725
1726 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1727         ptl = pmd_trans_huge_lock(pmd, vma);
1728         if (ptl) {
1729                 struct page *page;
1730
1731                 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1732                 if (page)
1733                         gather_stats(page, md, pmd_dirty(*pmd),
1734                                      HPAGE_PMD_SIZE/PAGE_SIZE);
1735                 spin_unlock(ptl);
1736                 return 0;
1737         }
1738
1739         if (pmd_trans_unstable(pmd))
1740                 return 0;
1741 #endif
1742         orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1743         do {
1744                 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1745                 if (!page)
1746                         continue;
1747                 gather_stats(page, md, pte_dirty(*pte), 1);
1748
1749         } while (pte++, addr += PAGE_SIZE, addr != end);
1750         pte_unmap_unlock(orig_pte, ptl);
1751         cond_resched();
1752         return 0;
1753 }
1754 #ifdef CONFIG_HUGETLB_PAGE
1755 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1756                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1757 {
1758         pte_t huge_pte = huge_ptep_get(pte);
1759         struct numa_maps *md;
1760         struct page *page;
1761
1762         if (!pte_present(huge_pte))
1763                 return 0;
1764
1765         page = pte_page(huge_pte);
1766         if (!page)
1767                 return 0;
1768
1769         md = walk->private;
1770         gather_stats(page, md, pte_dirty(huge_pte), 1);
1771         return 0;
1772 }
1773
1774 #else
1775 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1776                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1777 {
1778         return 0;
1779 }
1780 #endif
1781
1782 static const struct mm_walk_ops show_numa_ops = {
1783         .hugetlb_entry = gather_hugetlb_stats,
1784         .pmd_entry = gather_pte_stats,
1785 };
1786
1787 /*
1788  * Display pages allocated per node and memory policy via /proc.
1789  */
1790 static int show_numa_map(struct seq_file *m, void *v)
1791 {
1792         struct numa_maps_private *numa_priv = m->private;
1793         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1794         struct vm_area_struct *vma = v;
1795         struct numa_maps *md = &numa_priv->md;
1796         struct file *file = vma->vm_file;
1797         struct mm_struct *mm = vma->vm_mm;
1798         struct mempolicy *pol;
1799         char buffer[64];
1800         int nid;
1801
1802         if (!mm)
1803                 return 0;
1804
1805         /* Ensure we start with an empty set of numa_maps statistics. */
1806         memset(md, 0, sizeof(*md));
1807
1808         pol = __get_vma_policy(vma, vma->vm_start);
1809         if (pol) {
1810                 mpol_to_str(buffer, sizeof(buffer), pol);
1811                 mpol_cond_put(pol);
1812         } else {
1813                 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1814         }
1815
1816         seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1817
1818         if (file) {
1819                 seq_puts(m, " file=");
1820                 seq_file_path(m, file, "\n\t= ");
1821         } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1822                 seq_puts(m, " heap");
1823         } else if (is_stack(vma)) {
1824                 seq_puts(m, " stack");
1825         }
1826
1827         if (is_vm_hugetlb_page(vma))
1828                 seq_puts(m, " huge");
1829
1830         /* mmap_sem is held by m_start */
1831         walk_page_vma(vma, &show_numa_ops, md);
1832
1833         if (!md->pages)
1834                 goto out;
1835
1836         if (md->anon)
1837                 seq_printf(m, " anon=%lu", md->anon);
1838
1839         if (md->dirty)
1840                 seq_printf(m, " dirty=%lu", md->dirty);
1841
1842         if (md->pages != md->anon && md->pages != md->dirty)
1843                 seq_printf(m, " mapped=%lu", md->pages);
1844
1845         if (md->mapcount_max > 1)
1846                 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1847
1848         if (md->swapcache)
1849                 seq_printf(m, " swapcache=%lu", md->swapcache);
1850
1851         if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1852                 seq_printf(m, " active=%lu", md->active);
1853
1854         if (md->writeback)
1855                 seq_printf(m, " writeback=%lu", md->writeback);
1856
1857         for_each_node_state(nid, N_MEMORY)
1858                 if (md->node[nid])
1859                         seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1860
1861         seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1862 out:
1863         seq_putc(m, '\n');
1864         return 0;
1865 }
1866
1867 static const struct seq_operations proc_pid_numa_maps_op = {
1868         .start  = m_start,
1869         .next   = m_next,
1870         .stop   = m_stop,
1871         .show   = show_numa_map,
1872 };
1873
1874 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1875 {
1876         return proc_maps_open(inode, file, &proc_pid_numa_maps_op,
1877                                 sizeof(struct numa_maps_private));
1878 }
1879
1880 const struct file_operations proc_pid_numa_maps_operations = {
1881         .open           = pid_numa_maps_open,
1882         .read           = seq_read,
1883         .llseek         = seq_lseek,
1884         .release        = proc_map_release,
1885 };
1886
1887 #endif /* CONFIG_NUMA */