fs/proc/task_mmu.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include <linux/mm.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 vma_stop(struct proc_maps_private *priv)
 127 {
 128         struct mm_struct *mm = priv->mm;
 129
 130         release_task_mempolicy(priv);
 131         up_read(&mm->mmap_sem);
 132         mmput(mm);
 133 }
 134
 135 static struct vm_area_struct *
 136 m_next_vma(struct proc_maps_private *priv, struct vm_area_struct *vma)
 137 {
 138         if (vma == priv->tail_vma)
 139                 return NULL;
 140         return vma->vm_next ?: priv->tail_vma;
 141 }
 142
 143 static void m_cache_vma(struct seq_file *m, struct vm_area_struct *vma)
 144 {
 145         if (m->count < m->size) /* vma is copied successfully */
 146                 m->version = m_next_vma(m->private, vma) ? vma->vm_end : -1UL;
 147 }
 148
 149 static void *m_start(struct seq_file *m, loff_t *ppos)
 150 {
 151         struct proc_maps_private *priv = m->private;
 152         unsigned long last_addr = m->version;
 153         struct mm_struct *mm;
 154         struct vm_area_struct *vma;
 155         unsigned int pos = *ppos;
 156
 157         /* See m_cache_vma(). Zero at the start or after lseek. */
 158         if (last_addr == -1UL)
 159                 return NULL;
 160
 161         priv->task = get_proc_task(priv->inode);
 162         if (!priv->task)
 163                 return ERR_PTR(-ESRCH);
 164
 165         mm = priv->mm;
 166         if (!mm || !mmget_not_zero(mm))
 167                 return NULL;
 168
 169         down_read(&mm->mmap_sem);
 170         hold_task_mempolicy(priv);
 171         priv->tail_vma = get_gate_vma(mm);
 172
 173         if (last_addr) {
 174                 vma = find_vma(mm, last_addr - 1);
 175                 if (vma && vma->vm_start <= last_addr)
 176                         vma = m_next_vma(priv, vma);
 177                 if (vma)
 178                         return vma;
 179         }
 180
 181         m->version = 0;
 182         if (pos < mm->map_count) {
 183                 for (vma = mm->mmap; pos; pos--) {
 184                         m->version = vma->vm_start;
 185                         vma = vma->vm_next;
 186                 }
 187                 return vma;
 188         }
 189
 190         /* we do not bother to update m->version in this case */
 191         if (pos == mm->map_count && priv->tail_vma)
 192                 return priv->tail_vma;
 193
 194         vma_stop(priv);
 195         return NULL;
 196 }
 197
 198 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
 199 {
 200         struct proc_maps_private *priv = m->private;
 201         struct vm_area_struct *next;
 202
 203         (*pos)++;
 204         next = m_next_vma(priv, v);
 205         if (!next)
 206                 vma_stop(priv);
 207         return next;
 208 }
 209
 210 static void m_stop(struct seq_file *m, void *v)
 211 {
 212         struct proc_maps_private *priv = m->private;
 213
 214         if (!IS_ERR_OR_NULL(v))
 215                 vma_stop(priv);
 216         if (priv->task) {
 217                 put_task_struct(priv->task);
 218                 priv->task = NULL;
 219         }
 220 }
 221
 222 static int proc_maps_open(struct inode *inode, struct file *file,
 223                         const struct seq_operations *ops, int psize)
 224 {
 225         struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
 226
 227         if (!priv)
 228                 return -ENOMEM;
 229
 230         priv->inode = inode;
 231         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
 232         if (IS_ERR(priv->mm)) {
 233                 int err = PTR_ERR(priv->mm);
 234
 235                 seq_release_private(inode, file);
 236                 return err;
 237         }
 238
 239         return 0;
 240 }
 241
 242 static int proc_map_release(struct inode *inode, struct file *file)
 243 {
 244         struct seq_file *seq = file->private_data;
 245         struct proc_maps_private *priv = seq->private;
 246
 247         if (priv->mm)
 248                 mmdrop(priv->mm);
 249
 250         return seq_release_private(inode, file);
 251 }
 252
 253 static int do_maps_open(struct inode *inode, struct file *file,
 254                         const struct seq_operations *ops)
 255 {
 256         return proc_maps_open(inode, file, ops,
 257                                 sizeof(struct proc_maps_private));
 258 }
 259
 260 /*
 261  * Indicate if the VMA is a stack for the given task; for
 262  * /proc/PID/maps that is the stack of the main task.
 263  */
 264 static int is_stack(struct vm_area_struct *vma)
 265 {
 266         /*
 267          * We make no effort to guess what a given thread considers to be
 268          * its "stack".  It's not even well-defined for programs written
 269          * languages like Go.
 270          */
 271         return vma->vm_start <= vma->vm_mm->start_stack &&
 272                 vma->vm_end >= vma->vm_mm->start_stack;
 273 }
 274
 275 static void show_vma_header_prefix(struct seq_file *m,
 276                                    unsigned long start, unsigned long end,
 277                                    vm_flags_t flags, unsigned long long pgoff,
 278                                    dev_t dev, unsigned long ino)
 279 {
 280         seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
 281         seq_put_hex_ll(m, NULL, start, 8);
 282         seq_put_hex_ll(m, "-", end, 8);
 283         seq_putc(m, ' ');
 284         seq_putc(m, flags & VM_READ ? 'r' : '-');
 285         seq_putc(m, flags & VM_WRITE ? 'w' : '-');
 286         seq_putc(m, flags & VM_EXEC ? 'x' : '-');
 287         seq_putc(m, flags & VM_MAYSHARE ? 's' : 'p');
 288         seq_put_hex_ll(m, " ", pgoff, 8);
 289         seq_put_hex_ll(m, " ", MAJOR(dev), 2);
 290         seq_put_hex_ll(m, ":", MINOR(dev), 2);
 291         seq_put_decimal_ull(m, " ", ino);
 292         seq_putc(m, ' ');
 293 }
 294
 295 static void
 296 show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
 297 {
 298         struct mm_struct *mm = vma->vm_mm;
 299         struct file *file = vma->vm_file;
 300         vm_flags_t flags = vma->vm_flags;
 301         unsigned long ino = 0;
 302         unsigned long long pgoff = 0;
 303         unsigned long start, end;
 304         dev_t dev = 0;
 305         const char *name = NULL;
 306
 307         if (file) {
 308                 struct inode *inode = file_inode(vma->vm_file);
 309                 dev = inode->i_sb->s_dev;
 310                 ino = inode->i_ino;
 311                 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
 312         }
 313
 314         start = vma->vm_start;
 315         end = vma->vm_end;
 316         show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino);
 317
 318         /*
 319          * Print the dentry name for named mappings, and a
 320          * special [heap] marker for the heap:
 321          */
 322         if (file) {
 323                 seq_pad(m, ' ');
 324                 seq_file_path(m, file, "\n");
 325                 goto done;
 326         }
 327
 328         if (vma->vm_ops && vma->vm_ops->name) {
 329                 name = vma->vm_ops->name(vma);
 330                 if (name)
 331                         goto done;
 332         }
 333
 334         name = arch_vma_name(vma);
 335         if (!name) {
 336                 if (!mm) {
 337                         name = "[vdso]";
 338                         goto done;
 339                 }
 340
 341                 if (vma->vm_start <= mm->brk &&
 342                     vma->vm_end >= mm->start_brk) {
 343                         name = "[heap]";
 344                         goto done;
 345                 }
 346
 347                 if (is_stack(vma))
 348                         name = "[stack]";
 349         }
 350
 351 done:
 352         if (name) {
 353                 seq_pad(m, ' ');
 354                 seq_puts(m, name);
 355         }
 356         seq_putc(m, '\n');
 357 }
 358
 359 static int show_map(struct seq_file *m, void *v)
 360 {
 361         show_map_vma(m, v);
 362         m_cache_vma(m, v);
 363         return 0;
 364 }
 365
 366 static const struct seq_operations proc_pid_maps_op = {
 367         .start  = m_start,
 368         .next   = m_next,
 369         .stop   = m_stop,
 370         .show   = show_map
 371 };
 372
 373 static int pid_maps_open(struct inode *inode, struct file *file)
 374 {
 375         return do_maps_open(inode, file, &proc_pid_maps_op);
 376 }
 377
 378 const struct file_operations proc_pid_maps_operations = {
 379         .open           = pid_maps_open,
 380         .read           = seq_read,
 381         .llseek         = seq_lseek,
 382         .release        = proc_map_release,
 383 };
 384
 385 /*
 386  * Proportional Set Size(PSS): my share of RSS.
 387  *
 388  * PSS of a process is the count of pages it has in memory, where each
 389  * page is divided by the number of processes sharing it.  So if a
 390  * process has 1000 pages all to itself, and 1000 shared with one other
 391  * process, its PSS will be 1500.
 392  *
 393  * To keep (accumulated) division errors low, we adopt a 64bit
 394  * fixed-point pss counter to minimize division errors. So (pss >>
 395  * PSS_SHIFT) would be the real byte count.
 396  *
 397  * A shift of 12 before division means (assuming 4K page size):
 398  *      - 1M 3-user-pages add up to 8KB errors;
 399  *      - supports mapcount up to 2^24, or 16M;
 400  *      - supports PSS up to 2^52 bytes, or 4PB.
 401  */
 402 #define PSS_SHIFT 12
 403
 404 #ifdef CONFIG_PROC_PAGE_MONITOR
 405 struct mem_size_stats {
 406         unsigned long resident;
 407         unsigned long shared_clean;
 408         unsigned long shared_dirty;
 409         unsigned long private_clean;
 410         unsigned long private_dirty;
 411         unsigned long referenced;
 412         unsigned long anonymous;
 413         unsigned long lazyfree;
 414         unsigned long anonymous_thp;
 415         unsigned long shmem_thp;
 416         unsigned long swap;
 417         unsigned long shared_hugetlb;
 418         unsigned long private_hugetlb;
 419         u64 pss;
 420         u64 pss_locked;
 421         u64 swap_pss;
 422         bool check_shmem_swap;
 423 };
 424
 425 static void smaps_account(struct mem_size_stats *mss, struct page *page,
 426                 bool compound, bool young, bool dirty, bool locked)
 427 {
 428         int i, nr = compound ? 1 << compound_order(page) : 1;
 429         unsigned long size = nr * PAGE_SIZE;
 430
 431         if (PageAnon(page)) {
 432                 mss->anonymous += size;
 433                 if (!PageSwapBacked(page) && !dirty && !PageDirty(page))
 434                         mss->lazyfree += size;
 435         }
 436
 437         mss->resident += size;
 438         /* Accumulate the size in pages that have been accessed. */
 439         if (young || page_is_young(page) || PageReferenced(page))
 440                 mss->referenced += size;
 441
 442         /*
 443          * page_count(page) == 1 guarantees the page is mapped exactly once.
 444          * If any subpage of the compound page mapped with PTE it would elevate
 445          * page_count().
 446          */
 447         if (page_count(page) == 1) {
 448                 if (dirty || PageDirty(page))
 449                         mss->private_dirty += size;
 450                 else
 451                         mss->private_clean += size;
 452                 mss->pss += (u64)size << PSS_SHIFT;
 453                 if (locked)
 454                         mss->pss_locked += (u64)size << PSS_SHIFT;
 455                 return;
 456         }
 457
 458         for (i = 0; i < nr; i++, page++) {
 459                 int mapcount = page_mapcount(page);
 460                 unsigned long pss = (PAGE_SIZE << PSS_SHIFT);
 461
 462                 if (mapcount >= 2) {
 463                         if (dirty || PageDirty(page))
 464                                 mss->shared_dirty += PAGE_SIZE;
 465                         else
 466                                 mss->shared_clean += PAGE_SIZE;
 467                         mss->pss += pss / mapcount;
 468                         if (locked)
 469                                 mss->pss_locked += pss / mapcount;
 470                 } else {
 471                         if (dirty || PageDirty(page))
 472                                 mss->private_dirty += PAGE_SIZE;
 473                         else
 474                                 mss->private_clean += PAGE_SIZE;
 475                         mss->pss += pss;
 476                         if (locked)
 477                                 mss->pss_locked += pss;
 478                 }
 479         }
 480 }
 481
 482 #ifdef CONFIG_SHMEM
 483 static int smaps_pte_hole(unsigned long addr, unsigned long end,
 484                 struct mm_walk *walk)
 485 {
 486         struct mem_size_stats *mss = walk->private;
 487
 488         mss->swap += shmem_partial_swap_usage(
 489                         walk->vma->vm_file->f_mapping, addr, end);
 490
 491         return 0;
 492 }
 493 #endif
 494
 495 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
 496                 struct mm_walk *walk)
 497 {
 498         struct mem_size_stats *mss = walk->private;
 499         struct vm_area_struct *vma = walk->vma;
 500         bool locked = !!(vma->vm_flags & VM_LOCKED);
 501         struct page *page = NULL;
 502
 503         if (pte_present(*pte)) {
 504                 page = vm_normal_page(vma, addr, *pte);
 505         } else if (is_swap_pte(*pte)) {
 506                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 507
 508                 if (!non_swap_entry(swpent)) {
 509                         int mapcount;
 510
 511                         mss->swap += PAGE_SIZE;
 512                         mapcount = swp_swapcount(swpent);
 513                         if (mapcount >= 2) {
 514                                 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
 515
 516                                 do_div(pss_delta, mapcount);
 517                                 mss->swap_pss += pss_delta;
 518                         } else {
 519                                 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
 520                         }
 521                 } else if (is_migration_entry(swpent))
 522                         page = migration_entry_to_page(swpent);
 523                 else if (is_device_private_entry(swpent))
 524                         page = device_private_entry_to_page(swpent);
 525         } else if (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap
 526                                                         && pte_none(*pte))) {
 527                 page = find_get_entry(vma->vm_file->f_mapping,
 528                                                 linear_page_index(vma, addr));
 529                 if (!page)
 530                         return;
 531
 532                 if (xa_is_value(page))
 533                         mss->swap += PAGE_SIZE;
 534                 else
 535                         put_page(page);
 536
 537                 return;
 538         }
 539
 540         if (!page)
 541                 return;
 542
 543         smaps_account(mss, page, false, pte_young(*pte), pte_dirty(*pte), locked);
 544 }
 545
 546 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 547 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 548                 struct mm_walk *walk)
 549 {
 550         struct mem_size_stats *mss = walk->private;
 551         struct vm_area_struct *vma = walk->vma;
 552         bool locked = !!(vma->vm_flags & VM_LOCKED);
 553         struct page *page;
 554
 555         /* FOLL_DUMP will return -EFAULT on huge zero page */
 556         page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
 557         if (IS_ERR_OR_NULL(page))
 558                 return;
 559         if (PageAnon(page))
 560                 mss->anonymous_thp += HPAGE_PMD_SIZE;
 561         else if (PageSwapBacked(page))
 562                 mss->shmem_thp += HPAGE_PMD_SIZE;
 563         else if (is_zone_device_page(page))
 564                 /* pass */;
 565         else
 566                 VM_BUG_ON_PAGE(1, page);
 567         smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd), locked);
 568 }
 569 #else
 570 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 571                 struct mm_walk *walk)
 572 {
 573 }
 574 #endif
 575
 576 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 577                            struct mm_walk *walk)
 578 {
 579         struct vm_area_struct *vma = walk->vma;
 580         pte_t *pte;
 581         spinlock_t *ptl;
 582
 583         ptl = pmd_trans_huge_lock(pmd, vma);
 584         if (ptl) {
 585                 if (pmd_present(*pmd))
 586                         smaps_pmd_entry(pmd, addr, walk);
 587                 spin_unlock(ptl);
 588                 goto out;
 589         }
 590
 591         if (pmd_trans_unstable(pmd))
 592                 goto out;
 593         /*
 594          * The mmap_sem held all the way back in m_start() is what
 595          * keeps khugepaged out of here and from collapsing things
 596          * in here.
 597          */
 598         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 599         for (; addr != end; pte++, addr += PAGE_SIZE)
 600                 smaps_pte_entry(pte, addr, walk);
 601         pte_unmap_unlock(pte - 1, ptl);
 602 out:
 603         cond_resched();
 604         return 0;
 605 }
 606
 607 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
 608 {
 609         /*
 610          * Don't forget to update Documentation/ on changes.
 611          */
 612         static const char mnemonics[BITS_PER_LONG][2] = {
 613                 /*
 614                  * In case if we meet a flag we don't know about.
 615                  */
 616                 [0 ... (BITS_PER_LONG-1)] = "??",
 617
 618                 [ilog2(VM_READ)]        = "rd",
 619                 [ilog2(VM_WRITE)]       = "wr",
 620                 [ilog2(VM_EXEC)]        = "ex",
 621                 [ilog2(VM_SHARED)]      = "sh",
 622                 [ilog2(VM_MAYREAD)]     = "mr",
 623                 [ilog2(VM_MAYWRITE)]    = "mw",
 624                 [ilog2(VM_MAYEXEC)]     = "me",
 625                 [ilog2(VM_MAYSHARE)]    = "ms",
 626                 [ilog2(VM_GROWSDOWN)]   = "gd",
 627                 [ilog2(VM_PFNMAP)]      = "pf",
 628                 [ilog2(VM_DENYWRITE)]   = "dw",
 629 #ifdef CONFIG_X86_INTEL_MPX
 630                 [ilog2(VM_MPX)]         = "mp",
 631 #endif
 632                 [ilog2(VM_LOCKED)]      = "lo",
 633                 [ilog2(VM_IO)]          = "io",
 634                 [ilog2(VM_SEQ_READ)]    = "sr",
 635                 [ilog2(VM_RAND_READ)]   = "rr",
 636                 [ilog2(VM_DONTCOPY)]    = "dc",
 637                 [ilog2(VM_DONTEXPAND)]  = "de",
 638                 [ilog2(VM_ACCOUNT)]     = "ac",
 639                 [ilog2(VM_NORESERVE)]   = "nr",
 640                 [ilog2(VM_HUGETLB)]     = "ht",
 641                 [ilog2(VM_SYNC)]        = "sf",
 642                 [ilog2(VM_ARCH_1)]      = "ar",
 643                 [ilog2(VM_WIPEONFORK)]  = "wf",
 644                 [ilog2(VM_DONTDUMP)]    = "dd",
 645 #ifdef CONFIG_MEM_SOFT_DIRTY
 646                 [ilog2(VM_SOFTDIRTY)]   = "sd",
 647 #endif
 648                 [ilog2(VM_MIXEDMAP)]    = "mm",
 649                 [ilog2(VM_HUGEPAGE)]    = "hg",
 650                 [ilog2(VM_NOHUGEPAGE)]  = "nh",
 651                 [ilog2(VM_MERGEABLE)]   = "mg",
 652                 [ilog2(VM_UFFD_MISSING)]= "um",
 653                 [ilog2(VM_UFFD_WP)]     = "uw",
 654 #ifdef CONFIG_ARCH_HAS_PKEYS
 655                 /* These come out via ProtectionKey: */
 656                 [ilog2(VM_PKEY_BIT0)]   = "",
 657                 [ilog2(VM_PKEY_BIT1)]   = "",
 658                 [ilog2(VM_PKEY_BIT2)]   = "",
 659                 [ilog2(VM_PKEY_BIT3)]   = "",
 660 #if VM_PKEY_BIT4
 661                 [ilog2(VM_PKEY_BIT4)]   = "",
 662 #endif
 663 #endif /* CONFIG_ARCH_HAS_PKEYS */
 664         };
 665         size_t i;
 666
 667         seq_puts(m, "VmFlags: ");
 668         for (i = 0; i < BITS_PER_LONG; i++) {
 669                 if (!mnemonics[i][0])
 670                         continue;
 671                 if (vma->vm_flags & (1UL << i)) {
 672                         seq_putc(m, mnemonics[i][0]);
 673                         seq_putc(m, mnemonics[i][1]);
 674                         seq_putc(m, ' ');
 675                 }
 676         }
 677         seq_putc(m, '\n');
 678 }
 679
 680 #ifdef CONFIG_HUGETLB_PAGE
 681 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
 682                                  unsigned long addr, unsigned long end,
 683                                  struct mm_walk *walk)
 684 {
 685         struct mem_size_stats *mss = walk->private;
 686         struct vm_area_struct *vma = walk->vma;
 687         struct page *page = NULL;
 688
 689         if (pte_present(*pte)) {
 690                 page = vm_normal_page(vma, addr, *pte);
 691         } else if (is_swap_pte(*pte)) {
 692                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 693
 694                 if (is_migration_entry(swpent))
 695                         page = migration_entry_to_page(swpent);
 696                 else if (is_device_private_entry(swpent))
 697                         page = device_private_entry_to_page(swpent);
 698         }
 699         if (page) {
 700                 int mapcount = page_mapcount(page);
 701
 702                 if (mapcount >= 2)
 703                         mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
 704                 else
 705                         mss->private_hugetlb += huge_page_size(hstate_vma(vma));
 706         }
 707         return 0;
 708 }
 709 #endif /* HUGETLB_PAGE */
 710
 711 static void smap_gather_stats(struct vm_area_struct *vma,
 712                              struct mem_size_stats *mss)
 713 {
 714         struct mm_walk smaps_walk = {
 715                 .pmd_entry = smaps_pte_range,
 716 #ifdef CONFIG_HUGETLB_PAGE
 717                 .hugetlb_entry = smaps_hugetlb_range,
 718 #endif
 719                 .mm = vma->vm_mm,
 720         };
 721
 722         smaps_walk.private = mss;
 723
 724 #ifdef CONFIG_SHMEM
 725         /* In case of smaps_rollup, reset the value from previous vma */
 726         mss->check_shmem_swap = false;
 727         if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
 728                 /*
 729                  * For shared or readonly shmem mappings we know that all
 730                  * swapped out pages belong to the shmem object, and we can
 731                  * obtain the swap value much more efficiently. For private
 732                  * writable mappings, we might have COW pages that are
 733                  * not affected by the parent swapped out pages of the shmem
 734                  * object, so we have to distinguish them during the page walk.
 735                  * Unless we know that the shmem object (or the part mapped by
 736                  * our VMA) has no swapped out pages at all.
 737                  */
 738                 unsigned long shmem_swapped = shmem_swap_usage(vma);
 739
 740                 if (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
 741                                         !(vma->vm_flags & VM_WRITE)) {
 742                         mss->swap += shmem_swapped;
 743                 } else {
 744                         mss->check_shmem_swap = true;
 745                         smaps_walk.pte_hole = smaps_pte_hole;
 746                 }
 747         }
 748 #endif
 749         /* mmap_sem is held in m_start */
 750         walk_page_vma(vma, &smaps_walk);
 751 }
 752
 753 #define SEQ_PUT_DEC(str, val) \
 754                 seq_put_decimal_ull_width(m, str, (val) >> 10, 8)
 755
 756 /* Show the contents common for smaps and smaps_rollup */
 757 static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss)
 758 {
 759         SEQ_PUT_DEC("Rss:            ", mss->resident);
 760         SEQ_PUT_DEC(" kB\nPss:            ", mss->pss >> PSS_SHIFT);
 761         SEQ_PUT_DEC(" kB\nShared_Clean:   ", mss->shared_clean);
 762         SEQ_PUT_DEC(" kB\nShared_Dirty:   ", mss->shared_dirty);
 763         SEQ_PUT_DEC(" kB\nPrivate_Clean:  ", mss->private_clean);
 764         SEQ_PUT_DEC(" kB\nPrivate_Dirty:  ", mss->private_dirty);
 765         SEQ_PUT_DEC(" kB\nReferenced:     ", mss->referenced);
 766         SEQ_PUT_DEC(" kB\nAnonymous:      ", mss->anonymous);
 767         SEQ_PUT_DEC(" kB\nLazyFree:       ", mss->lazyfree);
 768         SEQ_PUT_DEC(" kB\nAnonHugePages:  ", mss->anonymous_thp);
 769         SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp);
 770         SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb);
 771         seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ",
 772                                   mss->private_hugetlb >> 10, 7);
 773         SEQ_PUT_DEC(" kB\nSwap:           ", mss->swap);
 774         SEQ_PUT_DEC(" kB\nSwapPss:        ",
 775                                         mss->swap_pss >> PSS_SHIFT);
 776         SEQ_PUT_DEC(" kB\nLocked:         ",
 777                                         mss->pss_locked >> PSS_SHIFT);
 778         seq_puts(m, " kB\n");
 779 }
 780
 781 static int show_smap(struct seq_file *m, void *v)
 782 {
 783         struct vm_area_struct *vma = v;
 784         struct mem_size_stats mss;
 785
 786         memset(&mss, 0, sizeof(mss));
 787
 788         smap_gather_stats(vma, &mss);
 789
 790         show_map_vma(m, vma);
 791
 792         SEQ_PUT_DEC("Size:           ", vma->vm_end - vma->vm_start);
 793         SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma));
 794         SEQ_PUT_DEC(" kB\nMMUPageSize:    ", vma_mmu_pagesize(vma));
 795         seq_puts(m, " kB\n");
 796
 797         __show_smap(m, &mss);
 798
 799         seq_printf(m, "THPeligible:    %d\n", transparent_hugepage_enabled(vma));
 800
 801         if (arch_pkeys_enabled())
 802                 seq_printf(m, "ProtectionKey:  %8u\n", vma_pkey(vma));
 803         show_smap_vma_flags(m, vma);
 804
 805         m_cache_vma(m, vma);
 806
 807         return 0;
 808 }
 809
 810 static int show_smaps_rollup(struct seq_file *m, void *v)
 811 {
 812         struct proc_maps_private *priv = m->private;
 813         struct mem_size_stats mss;
 814         struct mm_struct *mm;
 815         struct vm_area_struct *vma;
 816         unsigned long last_vma_end = 0;
 817         int ret = 0;
 818
 819         priv->task = get_proc_task(priv->inode);
 820         if (!priv->task)
 821                 return -ESRCH;
 822
 823         mm = priv->mm;
 824         if (!mm || !mmget_not_zero(mm)) {
 825                 ret = -ESRCH;
 826                 goto out_put_task;
 827         }
 828
 829         memset(&mss, 0, sizeof(mss));
 830
 831         down_read(&mm->mmap_sem);
 832         hold_task_mempolicy(priv);
 833
 834         for (vma = priv->mm->mmap; vma; vma = vma->vm_next) {
 835                 smap_gather_stats(vma, &mss);
 836                 last_vma_end = vma->vm_end;
 837         }
 838
 839         show_vma_header_prefix(m, priv->mm->mmap->vm_start,
 840                                last_vma_end, 0, 0, 0, 0);
 841         seq_pad(m, ' ');
 842         seq_puts(m, "[rollup]\n");
 843
 844         __show_smap(m, &mss);
 845
 846         release_task_mempolicy(priv);
 847         up_read(&mm->mmap_sem);
 848         mmput(mm);
 849
 850 out_put_task:
 851         put_task_struct(priv->task);
 852         priv->task = NULL;
 853
 854         return ret;
 855 }
 856 #undef SEQ_PUT_DEC
 857
 858 static const struct seq_operations proc_pid_smaps_op = {
 859         .start  = m_start,
 860         .next   = m_next,
 861         .stop   = m_stop,
 862         .show   = show_smap
 863 };
 864
 865 static int pid_smaps_open(struct inode *inode, struct file *file)
 866 {
 867         return do_maps_open(inode, file, &proc_pid_smaps_op);
 868 }
 869
 870 static int smaps_rollup_open(struct inode *inode, struct file *file)
 871 {
 872         int ret;
 873         struct proc_maps_private *priv;
 874
 875         priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT);
 876         if (!priv)
 877                 return -ENOMEM;
 878
 879         ret = single_open(file, show_smaps_rollup, priv);
 880         if (ret)
 881                 goto out_free;
 882
 883         priv->inode = inode;
 884         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
 885         if (IS_ERR(priv->mm)) {
 886                 ret = PTR_ERR(priv->mm);
 887
 888                 single_release(inode, file);
 889                 goto out_free;
 890         }
 891
 892         return 0;
 893
 894 out_free:
 895         kfree(priv);
 896         return ret;
 897 }
 898
 899 static int smaps_rollup_release(struct inode *inode, struct file *file)
 900 {
 901         struct seq_file *seq = file->private_data;
 902         struct proc_maps_private *priv = seq->private;
 903
 904         if (priv->mm)
 905                 mmdrop(priv->mm);
 906
 907         kfree(priv);
 908         return single_release(inode, file);
 909 }
 910
 911 const struct file_operations proc_pid_smaps_operations = {
 912         .open           = pid_smaps_open,
 913         .read           = seq_read,
 914         .llseek         = seq_lseek,
 915         .release        = proc_map_release,
 916 };
 917
 918 const struct file_operations proc_pid_smaps_rollup_operations = {
 919         .open           = smaps_rollup_open,
 920         .read           = seq_read,
 921         .llseek         = seq_lseek,
 922         .release        = smaps_rollup_release,
 923 };
 924
 925 enum clear_refs_types {
 926         CLEAR_REFS_ALL = 1,
 927         CLEAR_REFS_ANON,
 928         CLEAR_REFS_MAPPED,
 929         CLEAR_REFS_SOFT_DIRTY,
 930         CLEAR_REFS_MM_HIWATER_RSS,
 931         CLEAR_REFS_LAST,
 932 };
 933
 934 struct clear_refs_private {
 935         enum clear_refs_types type;
 936 };
 937
 938 #ifdef CONFIG_MEM_SOFT_DIRTY
 939 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 940                 unsigned long addr, pte_t *pte)
 941 {
 942         /*
 943          * The soft-dirty tracker uses #PF-s to catch writes
 944          * to pages, so write-protect the pte as well. See the
 945          * Documentation/admin-guide/mm/soft-dirty.rst for full description
 946          * of how soft-dirty works.
 947          */
 948         pte_t ptent = *pte;
 949
 950         if (pte_present(ptent)) {
 951                 pte_t old_pte;
 952
 953                 old_pte = ptep_modify_prot_start(vma, addr, pte);
 954                 ptent = pte_wrprotect(old_pte);
 955                 ptent = pte_clear_soft_dirty(ptent);
 956                 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent);
 957         } else if (is_swap_pte(ptent)) {
 958                 ptent = pte_swp_clear_soft_dirty(ptent);
 959                 set_pte_at(vma->vm_mm, addr, pte, ptent);
 960         }
 961 }
 962 #else
 963 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 964                 unsigned long addr, pte_t *pte)
 965 {
 966 }
 967 #endif
 968
 969 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
 970 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
 971                 unsigned long addr, pmd_t *pmdp)
 972 {
 973         pmd_t old, pmd = *pmdp;
 974
 975         if (pmd_present(pmd)) {
 976                 /* See comment in change_huge_pmd() */
 977                 old = pmdp_invalidate(vma, addr, pmdp);
 978                 if (pmd_dirty(old))
 979                         pmd = pmd_mkdirty(pmd);
 980                 if (pmd_young(old))
 981                         pmd = pmd_mkyoung(pmd);
 982
 983                 pmd = pmd_wrprotect(pmd);
 984                 pmd = pmd_clear_soft_dirty(pmd);
 985
 986                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
 987         } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
 988                 pmd = pmd_swp_clear_soft_dirty(pmd);
 989                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
 990         }
 991 }
 992 #else
 993 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
 994                 unsigned long addr, pmd_t *pmdp)
 995 {
 996 }
 997 #endif
 998
 999 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
1000                                 unsigned long end, struct mm_walk *walk)
1001 {
1002         struct clear_refs_private *cp = walk->private;
1003         struct vm_area_struct *vma = walk->vma;
1004         pte_t *pte, ptent;
1005         spinlock_t *ptl;
1006         struct page *page;
1007
1008         ptl = pmd_trans_huge_lock(pmd, vma);
1009         if (ptl) {
1010                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1011                         clear_soft_dirty_pmd(vma, addr, pmd);
1012                         goto out;
1013                 }
1014
1015                 if (!pmd_present(*pmd))
1016                         goto out;
1017
1018                 page = pmd_page(*pmd);
1019
1020                 /* Clear accessed and referenced bits. */
1021                 pmdp_test_and_clear_young(vma, addr, pmd);
1022                 test_and_clear_page_young(page);
1023                 ClearPageReferenced(page);
1024 out:
1025                 spin_unlock(ptl);
1026                 return 0;
1027         }
1028
1029         if (pmd_trans_unstable(pmd))
1030                 return 0;
1031
1032         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1033         for (; addr != end; pte++, addr += PAGE_SIZE) {
1034                 ptent = *pte;
1035
1036                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1037                         clear_soft_dirty(vma, addr, pte);
1038                         continue;
1039                 }
1040
1041                 if (!pte_present(ptent))
1042                         continue;
1043
1044                 page = vm_normal_page(vma, addr, ptent);
1045                 if (!page)
1046                         continue;
1047
1048                 /* Clear accessed and referenced bits. */
1049                 ptep_test_and_clear_young(vma, addr, pte);
1050                 test_and_clear_page_young(page);
1051                 ClearPageReferenced(page);
1052         }
1053         pte_unmap_unlock(pte - 1, ptl);
1054         cond_resched();
1055         return 0;
1056 }
1057
1058 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1059                                 struct mm_walk *walk)
1060 {
1061         struct clear_refs_private *cp = walk->private;
1062         struct vm_area_struct *vma = walk->vma;
1063
1064         if (vma->vm_flags & VM_PFNMAP)
1065                 return 1;
1066
1067         /*
1068          * Writing 1 to /proc/pid/clear_refs affects all pages.
1069          * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1070          * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1071          * Writing 4 to /proc/pid/clear_refs affects all pages.
1072          */
1073         if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1074                 return 1;
1075         if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1076                 return 1;
1077         return 0;
1078 }
1079
1080 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1081                                 size_t count, loff_t *ppos)
1082 {
1083         struct task_struct *task;
1084         char buffer[PROC_NUMBUF];
1085         struct mm_struct *mm;
1086         struct vm_area_struct *vma;
1087         enum clear_refs_types type;
1088         struct mmu_gather tlb;
1089         int itype;
1090         int rv;
1091
1092         memset(buffer, 0, sizeof(buffer));
1093         if (count > sizeof(buffer) - 1)
1094                 count = sizeof(buffer) - 1;
1095         if (copy_from_user(buffer, buf, count))
1096                 return -EFAULT;
1097         rv = kstrtoint(strstrip(buffer), 10, &itype);
1098         if (rv < 0)
1099                 return rv;
1100         type = (enum clear_refs_types)itype;
1101         if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1102                 return -EINVAL;
1103
1104         task = get_proc_task(file_inode(file));
1105         if (!task)
1106                 return -ESRCH;
1107         mm = get_task_mm(task);
1108         if (mm) {
1109                 struct mmu_notifier_range range;
1110                 struct clear_refs_private cp = {
1111                         .type = type,
1112                 };
1113                 struct mm_walk clear_refs_walk = {
1114                         .pmd_entry = clear_refs_pte_range,
1115                         .test_walk = clear_refs_test_walk,
1116                         .mm = mm,
1117                         .private = &cp,
1118                 };
1119
1120                 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1121                         if (down_write_killable(&mm->mmap_sem)) {
1122                                 count = -EINTR;
1123                                 goto out_mm;
1124                         }
1125
1126                         /*
1127                          * Writing 5 to /proc/pid/clear_refs resets the peak
1128                          * resident set size to this mm's current rss value.
1129                          */
1130                         reset_mm_hiwater_rss(mm);
1131                         up_write(&mm->mmap_sem);
1132                         goto out_mm;
1133                 }
1134
1135                 down_read(&mm->mmap_sem);
1136                 tlb_gather_mmu(&tlb, mm, 0, -1);
1137                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1138                         for (vma = mm->mmap; vma; vma = vma->vm_next) {
1139                                 if (!(vma->vm_flags & VM_SOFTDIRTY))
1140                                         continue;
1141                                 up_read(&mm->mmap_sem);
1142                                 if (down_write_killable(&mm->mmap_sem)) {
1143                                         count = -EINTR;
1144                                         goto out_mm;
1145                                 }
1146                                 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1147                                         vma->vm_flags &= ~VM_SOFTDIRTY;
1148                                         vma_set_page_prot(vma);
1149                                 }
1150                                 downgrade_write(&mm->mmap_sem);
1151                                 break;
1152                         }
1153
1154                         mmu_notifier_range_init(&range, mm, 0, -1UL);
1155                         mmu_notifier_invalidate_range_start(&range);
1156                 }
1157                 walk_page_range(0, mm->highest_vm_end, &clear_refs_walk);
1158                 if (type == CLEAR_REFS_SOFT_DIRTY)
1159                         mmu_notifier_invalidate_range_end(&range);
1160                 tlb_finish_mmu(&tlb, 0, -1);
1161                 up_read(&mm->mmap_sem);
1162 out_mm:
1163                 mmput(mm);
1164         }
1165         put_task_struct(task);
1166
1167         return count;
1168 }
1169
1170 const struct file_operations proc_clear_refs_operations = {
1171         .write          = clear_refs_write,
1172         .llseek         = noop_llseek,
1173 };
1174
1175 typedef struct {
1176         u64 pme;
1177 } pagemap_entry_t;
1178
1179 struct pagemapread {
1180         int pos, len;           /* units: PM_ENTRY_BYTES, not bytes */
1181         pagemap_entry_t *buffer;
1182         bool show_pfn;
1183 };
1184
1185 #define PAGEMAP_WALK_SIZE       (PMD_SIZE)
1186 #define PAGEMAP_WALK_MASK       (PMD_MASK)
1187
1188 #define PM_ENTRY_BYTES          sizeof(pagemap_entry_t)
1189 #define PM_PFRAME_BITS          55
1190 #define PM_PFRAME_MASK          GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1191 #define PM_SOFT_DIRTY           BIT_ULL(55)
1192 #define PM_MMAP_EXCLUSIVE       BIT_ULL(56)
1193 #define PM_FILE                 BIT_ULL(61)
1194 #define PM_SWAP                 BIT_ULL(62)
1195 #define PM_PRESENT              BIT_ULL(63)
1196
1197 #define PM_END_OF_BUFFER    1
1198
1199 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1200 {
1201         return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1202 }
1203
1204 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1205                           struct pagemapread *pm)
1206 {
1207         pm->buffer[pm->pos++] = *pme;
1208         if (pm->pos >= pm->len)
1209                 return PM_END_OF_BUFFER;
1210         return 0;
1211 }
1212
1213 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1214                                 struct mm_walk *walk)
1215 {
1216         struct pagemapread *pm = walk->private;
1217         unsigned long addr = start;
1218         int err = 0;
1219
1220         while (addr < end) {
1221                 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1222                 pagemap_entry_t pme = make_pme(0, 0);
1223                 /* End of address space hole, which we mark as non-present. */
1224                 unsigned long hole_end;
1225
1226                 if (vma)
1227                         hole_end = min(end, vma->vm_start);
1228                 else
1229                         hole_end = end;
1230
1231                 for (; addr < hole_end; addr += PAGE_SIZE) {
1232                         err = add_to_pagemap(addr, &pme, pm);
1233                         if (err)
1234                                 goto out;
1235                 }
1236
1237                 if (!vma)
1238                         break;
1239
1240                 /* Addresses in the VMA. */
1241                 if (vma->vm_flags & VM_SOFTDIRTY)
1242                         pme = make_pme(0, PM_SOFT_DIRTY);
1243                 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1244                         err = add_to_pagemap(addr, &pme, pm);
1245                         if (err)
1246                                 goto out;
1247                 }
1248         }
1249 out:
1250         return err;
1251 }
1252
1253 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1254                 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1255 {
1256         u64 frame = 0, flags = 0;
1257         struct page *page = NULL;
1258
1259         if (pte_present(pte)) {
1260                 if (pm->show_pfn)
1261                         frame = pte_pfn(pte);
1262                 flags |= PM_PRESENT;
1263                 page = _vm_normal_page(vma, addr, pte, true);
1264                 if (pte_soft_dirty(pte))
1265                         flags |= PM_SOFT_DIRTY;
1266         } else if (is_swap_pte(pte)) {
1267                 swp_entry_t entry;
1268                 if (pte_swp_soft_dirty(pte))
1269                         flags |= PM_SOFT_DIRTY;
1270                 entry = pte_to_swp_entry(pte);
1271                 if (pm->show_pfn)
1272                         frame = swp_type(entry) |
1273                                 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1274                 flags |= PM_SWAP;
1275                 if (is_migration_entry(entry))
1276                         page = migration_entry_to_page(entry);
1277
1278                 if (is_device_private_entry(entry))
1279                         page = device_private_entry_to_page(entry);
1280         }
1281
1282         if (page && !PageAnon(page))
1283                 flags |= PM_FILE;
1284         if (page && page_mapcount(page) == 1)
1285                 flags |= PM_MMAP_EXCLUSIVE;
1286         if (vma->vm_flags & VM_SOFTDIRTY)
1287                 flags |= PM_SOFT_DIRTY;
1288
1289         return make_pme(frame, flags);
1290 }
1291
1292 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1293                              struct mm_walk *walk)
1294 {
1295         struct vm_area_struct *vma = walk->vma;
1296         struct pagemapread *pm = walk->private;
1297         spinlock_t *ptl;
1298         pte_t *pte, *orig_pte;
1299         int err = 0;
1300
1301 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1302         ptl = pmd_trans_huge_lock(pmdp, vma);
1303         if (ptl) {
1304                 u64 flags = 0, frame = 0;
1305                 pmd_t pmd = *pmdp;
1306                 struct page *page = NULL;
1307
1308                 if (vma->vm_flags & VM_SOFTDIRTY)
1309                         flags |= PM_SOFT_DIRTY;
1310
1311                 if (pmd_present(pmd)) {
1312                         page = pmd_page(pmd);
1313
1314                         flags |= PM_PRESENT;
1315                         if (pmd_soft_dirty(pmd))
1316                                 flags |= PM_SOFT_DIRTY;
1317                         if (pm->show_pfn)
1318                                 frame = pmd_pfn(pmd) +
1319                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1320                 }
1321 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1322                 else if (is_swap_pmd(pmd)) {
1323                         swp_entry_t entry = pmd_to_swp_entry(pmd);
1324                         unsigned long offset;
1325
1326                         if (pm->show_pfn) {
1327                                 offset = swp_offset(entry) +
1328                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1329                                 frame = swp_type(entry) |
1330                                         (offset << MAX_SWAPFILES_SHIFT);
1331                         }
1332                         flags |= PM_SWAP;
1333                         if (pmd_swp_soft_dirty(pmd))
1334                                 flags |= PM_SOFT_DIRTY;
1335                         VM_BUG_ON(!is_pmd_migration_entry(pmd));
1336                         page = migration_entry_to_page(entry);
1337                 }
1338 #endif
1339
1340                 if (page && page_mapcount(page) == 1)
1341                         flags |= PM_MMAP_EXCLUSIVE;
1342
1343                 for (; addr != end; addr += PAGE_SIZE) {
1344                         pagemap_entry_t pme = make_pme(frame, flags);
1345
1346                         err = add_to_pagemap(addr, &pme, pm);
1347                         if (err)
1348                                 break;
1349                         if (pm->show_pfn) {
1350                                 if (flags & PM_PRESENT)
1351                                         frame++;
1352                                 else if (flags & PM_SWAP)
1353                                         frame += (1 << MAX_SWAPFILES_SHIFT);
1354                         }
1355                 }
1356                 spin_unlock(ptl);
1357                 return err;
1358         }
1359
1360         if (pmd_trans_unstable(pmdp))
1361                 return 0;
1362 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1363
1364         /*
1365          * We can assume that @vma always points to a valid one and @end never
1366          * goes beyond vma->vm_end.
1367          */
1368         orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1369         for (; addr < end; pte++, addr += PAGE_SIZE) {
1370                 pagemap_entry_t pme;
1371
1372                 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1373                 err = add_to_pagemap(addr, &pme, pm);
1374                 if (err)
1375                         break;
1376         }
1377         pte_unmap_unlock(orig_pte, ptl);
1378
1379         cond_resched();
1380
1381         return err;
1382 }
1383
1384 #ifdef CONFIG_HUGETLB_PAGE
1385 /* This function walks within one hugetlb entry in the single call */
1386 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1387                                  unsigned long addr, unsigned long end,
1388                                  struct mm_walk *walk)
1389 {
1390         struct pagemapread *pm = walk->private;
1391         struct vm_area_struct *vma = walk->vma;
1392         u64 flags = 0, frame = 0;
1393         int err = 0;
1394         pte_t pte;
1395
1396         if (vma->vm_flags & VM_SOFTDIRTY)
1397                 flags |= PM_SOFT_DIRTY;
1398
1399         pte = huge_ptep_get(ptep);
1400         if (pte_present(pte)) {
1401                 struct page *page = pte_page(pte);
1402
1403                 if (!PageAnon(page))
1404                         flags |= PM_FILE;
1405
1406                 if (page_mapcount(page) == 1)
1407                         flags |= PM_MMAP_EXCLUSIVE;
1408
1409                 flags |= PM_PRESENT;
1410                 if (pm->show_pfn)
1411                         frame = pte_pfn(pte) +
1412                                 ((addr & ~hmask) >> PAGE_SHIFT);
1413         }
1414
1415         for (; addr != end; addr += PAGE_SIZE) {
1416                 pagemap_entry_t pme = make_pme(frame, flags);
1417
1418                 err = add_to_pagemap(addr, &pme, pm);
1419                 if (err)
1420                         return err;
1421                 if (pm->show_pfn && (flags & PM_PRESENT))
1422                         frame++;
1423         }
1424
1425         cond_resched();
1426
1427         return err;
1428 }
1429 #endif /* HUGETLB_PAGE */
1430
1431 /*
1432  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1433  *
1434  * For each page in the address space, this file contains one 64-bit entry
1435  * consisting of the following:
1436  *
1437  * Bits 0-54  page frame number (PFN) if present
1438  * Bits 0-4   swap type if swapped
1439  * Bits 5-54  swap offset if swapped
1440  * Bit  55    pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst)
1441  * Bit  56    page exclusively mapped
1442  * Bits 57-60 zero
1443  * Bit  61    page is file-page or shared-anon
1444  * Bit  62    page swapped
1445  * Bit  63    page present
1446  *
1447  * If the page is not present but in swap, then the PFN contains an
1448  * encoding of the swap file number and the page's offset into the
1449  * swap. Unmapped pages return a null PFN. This allows determining
1450  * precisely which pages are mapped (or in swap) and comparing mapped
1451  * pages between processes.
1452  *
1453  * Efficient users of this interface will use /proc/pid/maps to
1454  * determine which areas of memory are actually mapped and llseek to
1455  * skip over unmapped regions.
1456  */
1457 static ssize_t pagemap_read(struct file *file, char __user *buf,
1458                             size_t count, loff_t *ppos)
1459 {
1460         struct mm_struct *mm = file->private_data;
1461         struct pagemapread pm;
1462         struct mm_walk pagemap_walk = {};
1463         unsigned long src;
1464         unsigned long svpfn;
1465         unsigned long start_vaddr;
1466         unsigned long end_vaddr;
1467         int ret = 0, copied = 0;
1468
1469         if (!mm || !mmget_not_zero(mm))
1470                 goto out;
1471
1472         ret = -EINVAL;
1473         /* file position must be aligned */
1474         if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1475                 goto out_mm;
1476
1477         ret = 0;
1478         if (!count)
1479                 goto out_mm;
1480
1481         /* do not disclose physical addresses: attack vector */
1482         pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1483
1484         pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1485         pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL);
1486         ret = -ENOMEM;
1487         if (!pm.buffer)
1488                 goto out_mm;
1489
1490         pagemap_walk.pmd_entry = pagemap_pmd_range;
1491         pagemap_walk.pte_hole = pagemap_pte_hole;
1492 #ifdef CONFIG_HUGETLB_PAGE
1493         pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1494 #endif
1495         pagemap_walk.mm = mm;
1496         pagemap_walk.private = &pm;
1497
1498         src = *ppos;
1499         svpfn = src / PM_ENTRY_BYTES;
1500         start_vaddr = svpfn << PAGE_SHIFT;
1501         end_vaddr = mm->task_size;
1502
1503         /* watch out for wraparound */
1504         if (svpfn > mm->task_size >> PAGE_SHIFT)
1505                 start_vaddr = end_vaddr;
1506
1507         /*
1508          * The odds are that this will stop walking way
1509          * before end_vaddr, because the length of the
1510          * user buffer is tracked in "pm", and the walk
1511          * will stop when we hit the end of the buffer.
1512          */
1513         ret = 0;
1514         while (count && (start_vaddr < end_vaddr)) {
1515                 int len;
1516                 unsigned long end;
1517
1518                 pm.pos = 0;
1519                 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1520                 /* overflow ? */
1521                 if (end < start_vaddr || end > end_vaddr)
1522                         end = end_vaddr;
1523                 down_read(&mm->mmap_sem);
1524                 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1525                 up_read(&mm->mmap_sem);
1526                 start_vaddr = end;
1527
1528                 len = min(count, PM_ENTRY_BYTES * pm.pos);
1529                 if (copy_to_user(buf, pm.buffer, len)) {
1530                         ret = -EFAULT;
1531                         goto out_free;
1532                 }
1533                 copied += len;
1534                 buf += len;
1535                 count -= len;
1536         }
1537         *ppos += copied;
1538         if (!ret || ret == PM_END_OF_BUFFER)
1539                 ret = copied;
1540
1541 out_free:
1542         kfree(pm.buffer);
1543 out_mm:
1544         mmput(mm);
1545 out:
1546         return ret;
1547 }
1548
1549 static int pagemap_open(struct inode *inode, struct file *file)
1550 {
1551         struct mm_struct *mm;
1552
1553         mm = proc_mem_open(inode, PTRACE_MODE_READ);
1554         if (IS_ERR(mm))
1555                 return PTR_ERR(mm);
1556         file->private_data = mm;
1557         return 0;
1558 }
1559
1560 static int pagemap_release(struct inode *inode, struct file *file)
1561 {
1562         struct mm_struct *mm = file->private_data;
1563
1564         if (mm)
1565                 mmdrop(mm);
1566         return 0;
1567 }
1568
1569 const struct file_operations proc_pagemap_operations = {
1570         .llseek         = mem_lseek, /* borrow this */
1571         .read           = pagemap_read,
1572         .open           = pagemap_open,
1573         .release        = pagemap_release,
1574 };
1575 #endif /* CONFIG_PROC_PAGE_MONITOR */
1576
1577 #ifdef CONFIG_NUMA
1578
1579 struct numa_maps {
1580         unsigned long pages;
1581         unsigned long anon;
1582         unsigned long active;
1583         unsigned long writeback;
1584         unsigned long mapcount_max;
1585         unsigned long dirty;
1586         unsigned long swapcache;
1587         unsigned long node[MAX_NUMNODES];
1588 };
1589
1590 struct numa_maps_private {
1591         struct proc_maps_private proc_maps;
1592         struct numa_maps md;
1593 };
1594
1595 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1596                         unsigned long nr_pages)
1597 {
1598         int count = page_mapcount(page);
1599
1600         md->pages += nr_pages;
1601         if (pte_dirty || PageDirty(page))
1602                 md->dirty += nr_pages;
1603
1604         if (PageSwapCache(page))
1605                 md->swapcache += nr_pages;
1606
1607         if (PageActive(page) || PageUnevictable(page))
1608                 md->active += nr_pages;
1609
1610         if (PageWriteback(page))
1611                 md->writeback += nr_pages;
1612
1613         if (PageAnon(page))
1614                 md->anon += nr_pages;
1615
1616         if (count > md->mapcount_max)
1617                 md->mapcount_max = count;
1618
1619         md->node[page_to_nid(page)] += nr_pages;
1620 }
1621
1622 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1623                 unsigned long addr)
1624 {
1625         struct page *page;
1626         int nid;
1627
1628         if (!pte_present(pte))
1629                 return NULL;
1630
1631         page = vm_normal_page(vma, addr, pte);
1632         if (!page)
1633                 return NULL;
1634
1635         if (PageReserved(page))
1636                 return NULL;
1637
1638         nid = page_to_nid(page);
1639         if (!node_isset(nid, node_states[N_MEMORY]))
1640                 return NULL;
1641
1642         return page;
1643 }
1644
1645 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1646 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1647                                               struct vm_area_struct *vma,
1648                                               unsigned long addr)
1649 {
1650         struct page *page;
1651         int nid;
1652
1653         if (!pmd_present(pmd))
1654                 return NULL;
1655
1656         page = vm_normal_page_pmd(vma, addr, pmd);
1657         if (!page)
1658                 return NULL;
1659
1660         if (PageReserved(page))
1661                 return NULL;
1662
1663         nid = page_to_nid(page);
1664         if (!node_isset(nid, node_states[N_MEMORY]))
1665                 return NULL;
1666
1667         return page;
1668 }
1669 #endif
1670
1671 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1672                 unsigned long end, struct mm_walk *walk)
1673 {
1674         struct numa_maps *md = walk->private;
1675         struct vm_area_struct *vma = walk->vma;
1676         spinlock_t *ptl;
1677         pte_t *orig_pte;
1678         pte_t *pte;
1679
1680 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1681         ptl = pmd_trans_huge_lock(pmd, vma);
1682         if (ptl) {
1683                 struct page *page;
1684
1685                 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1686                 if (page)
1687                         gather_stats(page, md, pmd_dirty(*pmd),
1688                                      HPAGE_PMD_SIZE/PAGE_SIZE);
1689                 spin_unlock(ptl);
1690                 return 0;
1691         }
1692
1693         if (pmd_trans_unstable(pmd))
1694                 return 0;
1695 #endif
1696         orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1697         do {
1698                 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1699                 if (!page)
1700                         continue;
1701                 gather_stats(page, md, pte_dirty(*pte), 1);
1702
1703         } while (pte++, addr += PAGE_SIZE, addr != end);
1704         pte_unmap_unlock(orig_pte, ptl);
1705         cond_resched();
1706         return 0;
1707 }
1708 #ifdef CONFIG_HUGETLB_PAGE
1709 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1710                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1711 {
1712         pte_t huge_pte = huge_ptep_get(pte);
1713         struct numa_maps *md;
1714         struct page *page;
1715
1716         if (!pte_present(huge_pte))
1717                 return 0;
1718
1719         page = pte_page(huge_pte);
1720         if (!page)
1721                 return 0;
1722
1723         md = walk->private;
1724         gather_stats(page, md, pte_dirty(huge_pte), 1);
1725         return 0;
1726 }
1727
1728 #else
1729 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1730                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1731 {
1732         return 0;
1733 }
1734 #endif
1735
1736 /*
1737  * Display pages allocated per node and memory policy via /proc.
1738  */
1739 static int show_numa_map(struct seq_file *m, void *v)
1740 {
1741         struct numa_maps_private *numa_priv = m->private;
1742         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1743         struct vm_area_struct *vma = v;
1744         struct numa_maps *md = &numa_priv->md;
1745         struct file *file = vma->vm_file;
1746         struct mm_struct *mm = vma->vm_mm;
1747         struct mm_walk walk = {
1748                 .hugetlb_entry = gather_hugetlb_stats,
1749                 .pmd_entry = gather_pte_stats,
1750                 .private = md,
1751                 .mm = mm,
1752         };
1753         struct mempolicy *pol;
1754         char buffer[64];
1755         int nid;
1756
1757         if (!mm)
1758                 return 0;
1759
1760         /* Ensure we start with an empty set of numa_maps statistics. */
1761         memset(md, 0, sizeof(*md));
1762
1763         pol = __get_vma_policy(vma, vma->vm_start);
1764         if (pol) {
1765                 mpol_to_str(buffer, sizeof(buffer), pol);
1766                 mpol_cond_put(pol);
1767         } else {
1768                 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1769         }
1770
1771         seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1772
1773         if (file) {
1774                 seq_puts(m, " file=");
1775                 seq_file_path(m, file, "\n\t= ");
1776         } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1777                 seq_puts(m, " heap");
1778         } else if (is_stack(vma)) {
1779                 seq_puts(m, " stack");
1780         }
1781
1782         if (is_vm_hugetlb_page(vma))
1783                 seq_puts(m, " huge");
1784
1785         /* mmap_sem is held by m_start */
1786         walk_page_vma(vma, &walk);
1787
1788         if (!md->pages)
1789                 goto out;
1790
1791         if (md->anon)
1792                 seq_printf(m, " anon=%lu", md->anon);
1793
1794         if (md->dirty)
1795                 seq_printf(m, " dirty=%lu", md->dirty);
1796
1797         if (md->pages != md->anon && md->pages != md->dirty)
1798                 seq_printf(m, " mapped=%lu", md->pages);
1799
1800         if (md->mapcount_max > 1)
1801                 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1802
1803         if (md->swapcache)
1804                 seq_printf(m, " swapcache=%lu", md->swapcache);
1805
1806         if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1807                 seq_printf(m, " active=%lu", md->active);
1808
1809         if (md->writeback)
1810                 seq_printf(m, " writeback=%lu", md->writeback);
1811
1812         for_each_node_state(nid, N_MEMORY)
1813                 if (md->node[nid])
1814                         seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1815
1816         seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1817 out:
1818         seq_putc(m, '\n');
1819         m_cache_vma(m, vma);
1820         return 0;
1821 }
1822
1823 static const struct seq_operations proc_pid_numa_maps_op = {
1824         .start  = m_start,
1825         .next   = m_next,
1826         .stop   = m_stop,
1827         .show   = show_numa_map,
1828 };
1829
1830 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1831 {
1832         return proc_maps_open(inode, file, &proc_pid_numa_maps_op,
1833                                 sizeof(struct numa_maps_private));
1834 }
1835
1836 const struct file_operations proc_pid_numa_maps_operations = {
1837         .open           = pid_numa_maps_open,
1838         .read           = seq_read,
1839         .llseek         = seq_lseek,
1840         .release        = proc_map_release,
1841 };
1842
1843 #endif /* CONFIG_NUMA */