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", 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         kfree(priv->rollup);
 251         return seq_release_private(inode, file);
 252 }
 253
 254 static int do_maps_open(struct inode *inode, struct file *file,
 255                         const struct seq_operations *ops)
 256 {
 257         return proc_maps_open(inode, file, ops,
 258                                 sizeof(struct proc_maps_private));
 259 }
 260
 261 /*
 262  * Indicate if the VMA is a stack for the given task; for
 263  * /proc/PID/maps that is the stack of the main task.
 264  */
 265 static int is_stack(struct vm_area_struct *vma)
 266 {
 267         /*
 268          * We make no effort to guess what a given thread considers to be
 269          * its "stack".  It's not even well-defined for programs written
 270          * languages like Go.
 271          */
 272         return vma->vm_start <= vma->vm_mm->start_stack &&
 273                 vma->vm_end >= vma->vm_mm->start_stack;
 274 }
 275
 276 static void show_vma_header_prefix(struct seq_file *m,
 277                                    unsigned long start, unsigned long end,
 278                                    vm_flags_t flags, unsigned long long pgoff,
 279                                    dev_t dev, unsigned long ino)
 280 {
 281         seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
 282         seq_put_hex_ll(m, NULL, start, 8);
 283         seq_put_hex_ll(m, "-", end, 8);
 284         seq_putc(m, ' ');
 285         seq_putc(m, flags & VM_READ ? 'r' : '-');
 286         seq_putc(m, flags & VM_WRITE ? 'w' : '-');
 287         seq_putc(m, flags & VM_EXEC ? 'x' : '-');
 288         seq_putc(m, flags & VM_MAYSHARE ? 's' : 'p');
 289         seq_put_hex_ll(m, " ", pgoff, 8);
 290         seq_put_hex_ll(m, " ", MAJOR(dev), 2);
 291         seq_put_hex_ll(m, ":", MINOR(dev), 2);
 292         seq_put_decimal_ull(m, " ", ino);
 293         seq_putc(m, ' ');
 294 }
 295
 296 static void
 297 show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
 298 {
 299         struct mm_struct *mm = vma->vm_mm;
 300         struct file *file = vma->vm_file;
 301         vm_flags_t flags = vma->vm_flags;
 302         unsigned long ino = 0;
 303         unsigned long long pgoff = 0;
 304         unsigned long start, end;
 305         dev_t dev = 0;
 306         const char *name = NULL;
 307
 308         if (file) {
 309                 struct inode *inode = file_inode(vma->vm_file);
 310                 dev = inode->i_sb->s_dev;
 311                 ino = inode->i_ino;
 312                 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
 313         }
 314
 315         start = vma->vm_start;
 316         end = vma->vm_end;
 317         show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino);
 318
 319         /*
 320          * Print the dentry name for named mappings, and a
 321          * special [heap] marker for the heap:
 322          */
 323         if (file) {
 324                 seq_pad(m, ' ');
 325                 seq_file_path(m, file, "\n");
 326                 goto done;
 327         }
 328
 329         if (vma->vm_ops && vma->vm_ops->name) {
 330                 name = vma->vm_ops->name(vma);
 331                 if (name)
 332                         goto done;
 333         }
 334
 335         name = arch_vma_name(vma);
 336         if (!name) {
 337                 if (!mm) {
 338                         name = "[vdso]";
 339                         goto done;
 340                 }
 341
 342                 if (vma->vm_start <= mm->brk &&
 343                     vma->vm_end >= mm->start_brk) {
 344                         name = "[heap]";
 345                         goto done;
 346                 }
 347
 348                 if (is_stack(vma))
 349                         name = "[stack]";
 350         }
 351
 352 done:
 353         if (name) {
 354                 seq_pad(m, ' ');
 355                 seq_puts(m, name);
 356         }
 357         seq_putc(m, '\n');
 358 }
 359
 360 static int show_map(struct seq_file *m, void *v, int is_pid)
 361 {
 362         show_map_vma(m, v, is_pid);
 363         m_cache_vma(m, v);
 364         return 0;
 365 }
 366
 367 static int show_pid_map(struct seq_file *m, void *v)
 368 {
 369         return show_map(m, v, 1);
 370 }
 371
 372 static int show_tid_map(struct seq_file *m, void *v)
 373 {
 374         return show_map(m, v, 0);
 375 }
 376
 377 static const struct seq_operations proc_pid_maps_op = {
 378         .start  = m_start,
 379         .next   = m_next,
 380         .stop   = m_stop,
 381         .show   = show_pid_map
 382 };
 383
 384 static const struct seq_operations proc_tid_maps_op = {
 385         .start  = m_start,
 386         .next   = m_next,
 387         .stop   = m_stop,
 388         .show   = show_tid_map
 389 };
 390
 391 static int pid_maps_open(struct inode *inode, struct file *file)
 392 {
 393         return do_maps_open(inode, file, &proc_pid_maps_op);
 394 }
 395
 396 static int tid_maps_open(struct inode *inode, struct file *file)
 397 {
 398         return do_maps_open(inode, file, &proc_tid_maps_op);
 399 }
 400
 401 const struct file_operations proc_pid_maps_operations = {
 402         .open           = pid_maps_open,
 403         .read           = seq_read,
 404         .llseek         = seq_lseek,
 405         .release        = proc_map_release,
 406 };
 407
 408 const struct file_operations proc_tid_maps_operations = {
 409         .open           = tid_maps_open,
 410         .read           = seq_read,
 411         .llseek         = seq_lseek,
 412         .release        = proc_map_release,
 413 };
 414
 415 /*
 416  * Proportional Set Size(PSS): my share of RSS.
 417  *
 418  * PSS of a process is the count of pages it has in memory, where each
 419  * page is divided by the number of processes sharing it.  So if a
 420  * process has 1000 pages all to itself, and 1000 shared with one other
 421  * process, its PSS will be 1500.
 422  *
 423  * To keep (accumulated) division errors low, we adopt a 64bit
 424  * fixed-point pss counter to minimize division errors. So (pss >>
 425  * PSS_SHIFT) would be the real byte count.
 426  *
 427  * A shift of 12 before division means (assuming 4K page size):
 428  *      - 1M 3-user-pages add up to 8KB errors;
 429  *      - supports mapcount up to 2^24, or 16M;
 430  *      - supports PSS up to 2^52 bytes, or 4PB.
 431  */
 432 #define PSS_SHIFT 12
 433
 434 #ifdef CONFIG_PROC_PAGE_MONITOR
 435 struct mem_size_stats {
 436         bool first;
 437         unsigned long resident;
 438         unsigned long shared_clean;
 439         unsigned long shared_dirty;
 440         unsigned long private_clean;
 441         unsigned long private_dirty;
 442         unsigned long referenced;
 443         unsigned long anonymous;
 444         unsigned long lazyfree;
 445         unsigned long anonymous_thp;
 446         unsigned long shmem_thp;
 447         unsigned long swap;
 448         unsigned long shared_hugetlb;
 449         unsigned long private_hugetlb;
 450         unsigned long first_vma_start;
 451         u64 pss;
 452         u64 pss_locked;
 453         u64 swap_pss;
 454         bool check_shmem_swap;
 455 };
 456
 457 static void smaps_account(struct mem_size_stats *mss, struct page *page,
 458                 bool compound, bool young, bool dirty)
 459 {
 460         int i, nr = compound ? 1 << compound_order(page) : 1;
 461         unsigned long size = nr * PAGE_SIZE;
 462
 463         if (PageAnon(page)) {
 464                 mss->anonymous += size;
 465                 if (!PageSwapBacked(page) && !dirty && !PageDirty(page))
 466                         mss->lazyfree += size;
 467         }
 468
 469         mss->resident += size;
 470         /* Accumulate the size in pages that have been accessed. */
 471         if (young || page_is_young(page) || PageReferenced(page))
 472                 mss->referenced += size;
 473
 474         /*
 475          * page_count(page) == 1 guarantees the page is mapped exactly once.
 476          * If any subpage of the compound page mapped with PTE it would elevate
 477          * page_count().
 478          */
 479         if (page_count(page) == 1) {
 480                 if (dirty || PageDirty(page))
 481                         mss->private_dirty += size;
 482                 else
 483                         mss->private_clean += size;
 484                 mss->pss += (u64)size << PSS_SHIFT;
 485                 return;
 486         }
 487
 488         for (i = 0; i < nr; i++, page++) {
 489                 int mapcount = page_mapcount(page);
 490
 491                 if (mapcount >= 2) {
 492                         if (dirty || PageDirty(page))
 493                                 mss->shared_dirty += PAGE_SIZE;
 494                         else
 495                                 mss->shared_clean += PAGE_SIZE;
 496                         mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount;
 497                 } else {
 498                         if (dirty || PageDirty(page))
 499                                 mss->private_dirty += PAGE_SIZE;
 500                         else
 501                                 mss->private_clean += PAGE_SIZE;
 502                         mss->pss += PAGE_SIZE << PSS_SHIFT;
 503                 }
 504         }
 505 }
 506
 507 #ifdef CONFIG_SHMEM
 508 static int smaps_pte_hole(unsigned long addr, unsigned long end,
 509                 struct mm_walk *walk)
 510 {
 511         struct mem_size_stats *mss = walk->private;
 512
 513         mss->swap += shmem_partial_swap_usage(
 514                         walk->vma->vm_file->f_mapping, addr, end);
 515
 516         return 0;
 517 }
 518 #endif
 519
 520 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
 521                 struct mm_walk *walk)
 522 {
 523         struct mem_size_stats *mss = walk->private;
 524         struct vm_area_struct *vma = walk->vma;
 525         struct page *page = NULL;
 526
 527         if (pte_present(*pte)) {
 528                 page = vm_normal_page(vma, addr, *pte);
 529         } else if (is_swap_pte(*pte)) {
 530                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 531
 532                 if (!non_swap_entry(swpent)) {
 533                         int mapcount;
 534
 535                         mss->swap += PAGE_SIZE;
 536                         mapcount = swp_swapcount(swpent);
 537                         if (mapcount >= 2) {
 538                                 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
 539
 540                                 do_div(pss_delta, mapcount);
 541                                 mss->swap_pss += pss_delta;
 542                         } else {
 543                                 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
 544                         }
 545                 } else if (is_migration_entry(swpent))
 546                         page = migration_entry_to_page(swpent);
 547                 else if (is_device_private_entry(swpent))
 548                         page = device_private_entry_to_page(swpent);
 549         } else if (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap
 550                                                         && pte_none(*pte))) {
 551                 page = find_get_entry(vma->vm_file->f_mapping,
 552                                                 linear_page_index(vma, addr));
 553                 if (!page)
 554                         return;
 555
 556                 if (radix_tree_exceptional_entry(page))
 557                         mss->swap += PAGE_SIZE;
 558                 else
 559                         put_page(page);
 560
 561                 return;
 562         }
 563
 564         if (!page)
 565                 return;
 566
 567         smaps_account(mss, page, false, pte_young(*pte), pte_dirty(*pte));
 568 }
 569
 570 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 571 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 572                 struct mm_walk *walk)
 573 {
 574         struct mem_size_stats *mss = walk->private;
 575         struct vm_area_struct *vma = walk->vma;
 576         struct page *page;
 577
 578         /* FOLL_DUMP will return -EFAULT on huge zero page */
 579         page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
 580         if (IS_ERR_OR_NULL(page))
 581                 return;
 582         if (PageAnon(page))
 583                 mss->anonymous_thp += HPAGE_PMD_SIZE;
 584         else if (PageSwapBacked(page))
 585                 mss->shmem_thp += HPAGE_PMD_SIZE;
 586         else if (is_zone_device_page(page))
 587                 /* pass */;
 588         else
 589                 VM_BUG_ON_PAGE(1, page);
 590         smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd));
 591 }
 592 #else
 593 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 594                 struct mm_walk *walk)
 595 {
 596 }
 597 #endif
 598
 599 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 600                            struct mm_walk *walk)
 601 {
 602         struct vm_area_struct *vma = walk->vma;
 603         pte_t *pte;
 604         spinlock_t *ptl;
 605
 606         ptl = pmd_trans_huge_lock(pmd, vma);
 607         if (ptl) {
 608                 if (pmd_present(*pmd))
 609                         smaps_pmd_entry(pmd, addr, walk);
 610                 spin_unlock(ptl);
 611                 goto out;
 612         }
 613
 614         if (pmd_trans_unstable(pmd))
 615                 goto out;
 616         /*
 617          * The mmap_sem held all the way back in m_start() is what
 618          * keeps khugepaged out of here and from collapsing things
 619          * in here.
 620          */
 621         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 622         for (; addr != end; pte++, addr += PAGE_SIZE)
 623                 smaps_pte_entry(pte, addr, walk);
 624         pte_unmap_unlock(pte - 1, ptl);
 625 out:
 626         cond_resched();
 627         return 0;
 628 }
 629
 630 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
 631 {
 632         /*
 633          * Don't forget to update Documentation/ on changes.
 634          */
 635         static const char mnemonics[BITS_PER_LONG][2] = {
 636                 /*
 637                  * In case if we meet a flag we don't know about.
 638                  */
 639                 [0 ... (BITS_PER_LONG-1)] = "??",
 640
 641                 [ilog2(VM_READ)]        = "rd",
 642                 [ilog2(VM_WRITE)]       = "wr",
 643                 [ilog2(VM_EXEC)]        = "ex",
 644                 [ilog2(VM_SHARED)]      = "sh",
 645                 [ilog2(VM_MAYREAD)]     = "mr",
 646                 [ilog2(VM_MAYWRITE)]    = "mw",
 647                 [ilog2(VM_MAYEXEC)]     = "me",
 648                 [ilog2(VM_MAYSHARE)]    = "ms",
 649                 [ilog2(VM_GROWSDOWN)]   = "gd",
 650                 [ilog2(VM_PFNMAP)]      = "pf",
 651                 [ilog2(VM_DENYWRITE)]   = "dw",
 652 #ifdef CONFIG_X86_INTEL_MPX
 653                 [ilog2(VM_MPX)]         = "mp",
 654 #endif
 655                 [ilog2(VM_LOCKED)]      = "lo",
 656                 [ilog2(VM_IO)]          = "io",
 657                 [ilog2(VM_SEQ_READ)]    = "sr",
 658                 [ilog2(VM_RAND_READ)]   = "rr",
 659                 [ilog2(VM_DONTCOPY)]    = "dc",
 660                 [ilog2(VM_DONTEXPAND)]  = "de",
 661                 [ilog2(VM_ACCOUNT)]     = "ac",
 662                 [ilog2(VM_NORESERVE)]   = "nr",
 663                 [ilog2(VM_HUGETLB)]     = "ht",
 664                 [ilog2(VM_SYNC)]        = "sf",
 665                 [ilog2(VM_ARCH_1)]      = "ar",
 666                 [ilog2(VM_WIPEONFORK)]  = "wf",
 667                 [ilog2(VM_DONTDUMP)]    = "dd",
 668 #ifdef CONFIG_MEM_SOFT_DIRTY
 669                 [ilog2(VM_SOFTDIRTY)]   = "sd",
 670 #endif
 671                 [ilog2(VM_MIXEDMAP)]    = "mm",
 672                 [ilog2(VM_HUGEPAGE)]    = "hg",
 673                 [ilog2(VM_NOHUGEPAGE)]  = "nh",
 674                 [ilog2(VM_MERGEABLE)]   = "mg",
 675                 [ilog2(VM_UFFD_MISSING)]= "um",
 676                 [ilog2(VM_UFFD_WP)]     = "uw",
 677 #ifdef CONFIG_ARCH_HAS_PKEYS
 678                 /* These come out via ProtectionKey: */
 679                 [ilog2(VM_PKEY_BIT0)]   = "",
 680                 [ilog2(VM_PKEY_BIT1)]   = "",
 681                 [ilog2(VM_PKEY_BIT2)]   = "",
 682                 [ilog2(VM_PKEY_BIT3)]   = "",
 683 #if VM_PKEY_BIT4
 684                 [ilog2(VM_PKEY_BIT4)]   = "",
 685 #endif
 686 #endif /* CONFIG_ARCH_HAS_PKEYS */
 687         };
 688         size_t i;
 689
 690         seq_puts(m, "VmFlags: ");
 691         for (i = 0; i < BITS_PER_LONG; i++) {
 692                 if (!mnemonics[i][0])
 693                         continue;
 694                 if (vma->vm_flags & (1UL << i)) {
 695                         seq_putc(m, mnemonics[i][0]);
 696                         seq_putc(m, mnemonics[i][1]);
 697                         seq_putc(m, ' ');
 698                 }
 699         }
 700         seq_putc(m, '\n');
 701 }
 702
 703 #ifdef CONFIG_HUGETLB_PAGE
 704 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
 705                                  unsigned long addr, unsigned long end,
 706                                  struct mm_walk *walk)
 707 {
 708         struct mem_size_stats *mss = walk->private;
 709         struct vm_area_struct *vma = walk->vma;
 710         struct page *page = NULL;
 711
 712         if (pte_present(*pte)) {
 713                 page = vm_normal_page(vma, addr, *pte);
 714         } else if (is_swap_pte(*pte)) {
 715                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 716
 717                 if (is_migration_entry(swpent))
 718                         page = migration_entry_to_page(swpent);
 719                 else if (is_device_private_entry(swpent))
 720                         page = device_private_entry_to_page(swpent);
 721         }
 722         if (page) {
 723                 int mapcount = page_mapcount(page);
 724
 725                 if (mapcount >= 2)
 726                         mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
 727                 else
 728                         mss->private_hugetlb += huge_page_size(hstate_vma(vma));
 729         }
 730         return 0;
 731 }
 732 #endif /* HUGETLB_PAGE */
 733
 734 #define SEQ_PUT_DEC(str, val) \
 735                 seq_put_decimal_ull_width(m, str, (val) >> 10, 8)
 736 static int show_smap(struct seq_file *m, void *v, int is_pid)
 737 {
 738         struct proc_maps_private *priv = m->private;
 739         struct vm_area_struct *vma = v;
 740         struct mem_size_stats mss_stack;
 741         struct mem_size_stats *mss;
 742         struct mm_walk smaps_walk = {
 743                 .pmd_entry = smaps_pte_range,
 744 #ifdef CONFIG_HUGETLB_PAGE
 745                 .hugetlb_entry = smaps_hugetlb_range,
 746 #endif
 747                 .mm = vma->vm_mm,
 748         };
 749         int ret = 0;
 750         bool rollup_mode;
 751         bool last_vma;
 752
 753         if (priv->rollup) {
 754                 rollup_mode = true;
 755                 mss = priv->rollup;
 756                 if (mss->first) {
 757                         mss->first_vma_start = vma->vm_start;
 758                         mss->first = false;
 759                 }
 760                 last_vma = !m_next_vma(priv, vma);
 761         } else {
 762                 rollup_mode = false;
 763                 memset(&mss_stack, 0, sizeof(mss_stack));
 764                 mss = &mss_stack;
 765         }
 766
 767         smaps_walk.private = mss;
 768
 769 #ifdef CONFIG_SHMEM
 770         if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
 771                 /*
 772                  * For shared or readonly shmem mappings we know that all
 773                  * swapped out pages belong to the shmem object, and we can
 774                  * obtain the swap value much more efficiently. For private
 775                  * writable mappings, we might have COW pages that are
 776                  * not affected by the parent swapped out pages of the shmem
 777                  * object, so we have to distinguish them during the page walk.
 778                  * Unless we know that the shmem object (or the part mapped by
 779                  * our VMA) has no swapped out pages at all.
 780                  */
 781                 unsigned long shmem_swapped = shmem_swap_usage(vma);
 782
 783                 if (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
 784                                         !(vma->vm_flags & VM_WRITE)) {
 785                         mss->swap = shmem_swapped;
 786                 } else {
 787                         mss->check_shmem_swap = true;
 788                         smaps_walk.pte_hole = smaps_pte_hole;
 789                 }
 790         }
 791 #endif
 792
 793         /* mmap_sem is held in m_start */
 794         walk_page_vma(vma, &smaps_walk);
 795         if (vma->vm_flags & VM_LOCKED)
 796                 mss->pss_locked += mss->pss;
 797
 798         if (!rollup_mode) {
 799                 show_map_vma(m, vma, is_pid);
 800         } else if (last_vma) {
 801                 show_vma_header_prefix(
 802                         m, mss->first_vma_start, vma->vm_end, 0, 0, 0, 0);
 803                 seq_pad(m, ' ');
 804                 seq_puts(m, "[rollup]\n");
 805         } else {
 806                 ret = SEQ_SKIP;
 807         }
 808
 809         if (!rollup_mode) {
 810                 SEQ_PUT_DEC("Size:           ", vma->vm_end - vma->vm_start);
 811                 SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma));
 812                 SEQ_PUT_DEC(" kB\nMMUPageSize:    ", vma_mmu_pagesize(vma));
 813                 seq_puts(m, " kB\n");
 814         }
 815
 816         if (!rollup_mode || last_vma) {
 817                 SEQ_PUT_DEC("Rss:            ", mss->resident);
 818                 SEQ_PUT_DEC(" kB\nPss:            ", mss->pss >> PSS_SHIFT);
 819                 SEQ_PUT_DEC(" kB\nShared_Clean:   ", mss->shared_clean);
 820                 SEQ_PUT_DEC(" kB\nShared_Dirty:   ", mss->shared_dirty);
 821                 SEQ_PUT_DEC(" kB\nPrivate_Clean:  ", mss->private_clean);
 822                 SEQ_PUT_DEC(" kB\nPrivate_Dirty:  ", mss->private_dirty);
 823                 SEQ_PUT_DEC(" kB\nReferenced:     ", mss->referenced);
 824                 SEQ_PUT_DEC(" kB\nAnonymous:      ", mss->anonymous);
 825                 SEQ_PUT_DEC(" kB\nLazyFree:       ", mss->lazyfree);
 826                 SEQ_PUT_DEC(" kB\nAnonHugePages:  ", mss->anonymous_thp);
 827                 SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp);
 828                 SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb);
 829                 seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ",
 830                                           mss->private_hugetlb >> 10, 7);
 831                 SEQ_PUT_DEC(" kB\nSwap:           ", mss->swap);
 832                 SEQ_PUT_DEC(" kB\nSwapPss:        ",
 833                                                 mss->swap_pss >> PSS_SHIFT);
 834                 SEQ_PUT_DEC(" kB\nLocked:         ", mss->pss >> PSS_SHIFT);
 835                 seq_puts(m, " kB\n");
 836         }
 837         if (!rollup_mode) {
 838                 if (arch_pkeys_enabled())
 839                         seq_printf(m, "ProtectionKey:  %8u\n", vma_pkey(vma));
 840                 show_smap_vma_flags(m, vma);
 841         }
 842         m_cache_vma(m, vma);
 843         return ret;
 844 }
 845 #undef SEQ_PUT_DEC
 846
 847 static int show_pid_smap(struct seq_file *m, void *v)
 848 {
 849         return show_smap(m, v, 1);
 850 }
 851
 852 static int show_tid_smap(struct seq_file *m, void *v)
 853 {
 854         return show_smap(m, v, 0);
 855 }
 856
 857 static const struct seq_operations proc_pid_smaps_op = {
 858         .start  = m_start,
 859         .next   = m_next,
 860         .stop   = m_stop,
 861         .show   = show_pid_smap
 862 };
 863
 864 static const struct seq_operations proc_tid_smaps_op = {
 865         .start  = m_start,
 866         .next   = m_next,
 867         .stop   = m_stop,
 868         .show   = show_tid_smap
 869 };
 870
 871 static int pid_smaps_open(struct inode *inode, struct file *file)
 872 {
 873         return do_maps_open(inode, file, &proc_pid_smaps_op);
 874 }
 875
 876 static int pid_smaps_rollup_open(struct inode *inode, struct file *file)
 877 {
 878         struct seq_file *seq;
 879         struct proc_maps_private *priv;
 880         int ret = do_maps_open(inode, file, &proc_pid_smaps_op);
 881
 882         if (ret < 0)
 883                 return ret;
 884         seq = file->private_data;
 885         priv = seq->private;
 886         priv->rollup = kzalloc(sizeof(*priv->rollup), GFP_KERNEL);
 887         if (!priv->rollup) {
 888                 proc_map_release(inode, file);
 889                 return -ENOMEM;
 890         }
 891         priv->rollup->first = true;
 892         return 0;
 893 }
 894
 895 static int tid_smaps_open(struct inode *inode, struct file *file)
 896 {
 897         return do_maps_open(inode, file, &proc_tid_smaps_op);
 898 }
 899
 900 const struct file_operations proc_pid_smaps_operations = {
 901         .open           = pid_smaps_open,
 902         .read           = seq_read,
 903         .llseek         = seq_lseek,
 904         .release        = proc_map_release,
 905 };
 906
 907 const struct file_operations proc_pid_smaps_rollup_operations = {
 908         .open           = pid_smaps_rollup_open,
 909         .read           = seq_read,
 910         .llseek         = seq_lseek,
 911         .release        = proc_map_release,
 912 };
 913
 914 const struct file_operations proc_tid_smaps_operations = {
 915         .open           = tid_smaps_open,
 916         .read           = seq_read,
 917         .llseek         = seq_lseek,
 918         .release        = proc_map_release,
 919 };
 920
 921 enum clear_refs_types {
 922         CLEAR_REFS_ALL = 1,
 923         CLEAR_REFS_ANON,
 924         CLEAR_REFS_MAPPED,
 925         CLEAR_REFS_SOFT_DIRTY,
 926         CLEAR_REFS_MM_HIWATER_RSS,
 927         CLEAR_REFS_LAST,
 928 };
 929
 930 struct clear_refs_private {
 931         enum clear_refs_types type;
 932 };
 933
 934 #ifdef CONFIG_MEM_SOFT_DIRTY
 935 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 936                 unsigned long addr, pte_t *pte)
 937 {
 938         /*
 939          * The soft-dirty tracker uses #PF-s to catch writes
 940          * to pages, so write-protect the pte as well. See the
 941          * Documentation/admin-guide/mm/soft-dirty.rst for full description
 942          * of how soft-dirty works.
 943          */
 944         pte_t ptent = *pte;
 945
 946         if (pte_present(ptent)) {
 947                 ptent = ptep_modify_prot_start(vma->vm_mm, addr, pte);
 948                 ptent = pte_wrprotect(ptent);
 949                 ptent = pte_clear_soft_dirty(ptent);
 950                 ptep_modify_prot_commit(vma->vm_mm, addr, pte, ptent);
 951         } else if (is_swap_pte(ptent)) {
 952                 ptent = pte_swp_clear_soft_dirty(ptent);
 953                 set_pte_at(vma->vm_mm, addr, pte, ptent);
 954         }
 955 }
 956 #else
 957 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 958                 unsigned long addr, pte_t *pte)
 959 {
 960 }
 961 #endif
 962
 963 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
 964 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
 965                 unsigned long addr, pmd_t *pmdp)
 966 {
 967         pmd_t old, pmd = *pmdp;
 968
 969         if (pmd_present(pmd)) {
 970                 /* See comment in change_huge_pmd() */
 971                 old = pmdp_invalidate(vma, addr, pmdp);
 972                 if (pmd_dirty(old))
 973                         pmd = pmd_mkdirty(pmd);
 974                 if (pmd_young(old))
 975                         pmd = pmd_mkyoung(pmd);
 976
 977                 pmd = pmd_wrprotect(pmd);
 978                 pmd = pmd_clear_soft_dirty(pmd);
 979
 980                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
 981         } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
 982                 pmd = pmd_swp_clear_soft_dirty(pmd);
 983                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
 984         }
 985 }
 986 #else
 987 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
 988                 unsigned long addr, pmd_t *pmdp)
 989 {
 990 }
 991 #endif
 992
 993 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
 994                                 unsigned long end, struct mm_walk *walk)
 995 {
 996         struct clear_refs_private *cp = walk->private;
 997         struct vm_area_struct *vma = walk->vma;
 998         pte_t *pte, ptent;
 999         spinlock_t *ptl;
1000         struct page *page;
1001
1002         ptl = pmd_trans_huge_lock(pmd, vma);
1003         if (ptl) {
1004                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1005                         clear_soft_dirty_pmd(vma, addr, pmd);
1006                         goto out;
1007                 }
1008
1009                 if (!pmd_present(*pmd))
1010                         goto out;
1011
1012                 page = pmd_page(*pmd);
1013
1014                 /* Clear accessed and referenced bits. */
1015                 pmdp_test_and_clear_young(vma, addr, pmd);
1016                 test_and_clear_page_young(page);
1017                 ClearPageReferenced(page);
1018 out:
1019                 spin_unlock(ptl);
1020                 return 0;
1021         }
1022
1023         if (pmd_trans_unstable(pmd))
1024                 return 0;
1025
1026         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1027         for (; addr != end; pte++, addr += PAGE_SIZE) {
1028                 ptent = *pte;
1029
1030                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1031                         clear_soft_dirty(vma, addr, pte);
1032                         continue;
1033                 }
1034
1035                 if (!pte_present(ptent))
1036                         continue;
1037
1038                 page = vm_normal_page(vma, addr, ptent);
1039                 if (!page)
1040                         continue;
1041
1042                 /* Clear accessed and referenced bits. */
1043                 ptep_test_and_clear_young(vma, addr, pte);
1044                 test_and_clear_page_young(page);
1045                 ClearPageReferenced(page);
1046         }
1047         pte_unmap_unlock(pte - 1, ptl);
1048         cond_resched();
1049         return 0;
1050 }
1051
1052 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1053                                 struct mm_walk *walk)
1054 {
1055         struct clear_refs_private *cp = walk->private;
1056         struct vm_area_struct *vma = walk->vma;
1057
1058         if (vma->vm_flags & VM_PFNMAP)
1059                 return 1;
1060
1061         /*
1062          * Writing 1 to /proc/pid/clear_refs affects all pages.
1063          * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1064          * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1065          * Writing 4 to /proc/pid/clear_refs affects all pages.
1066          */
1067         if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1068                 return 1;
1069         if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1070                 return 1;
1071         return 0;
1072 }
1073
1074 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1075                                 size_t count, loff_t *ppos)
1076 {
1077         struct task_struct *task;
1078         char buffer[PROC_NUMBUF];
1079         struct mm_struct *mm;
1080         struct vm_area_struct *vma;
1081         enum clear_refs_types type;
1082         struct mmu_gather tlb;
1083         int itype;
1084         int rv;
1085
1086         memset(buffer, 0, sizeof(buffer));
1087         if (count > sizeof(buffer) - 1)
1088                 count = sizeof(buffer) - 1;
1089         if (copy_from_user(buffer, buf, count))
1090                 return -EFAULT;
1091         rv = kstrtoint(strstrip(buffer), 10, &itype);
1092         if (rv < 0)
1093                 return rv;
1094         type = (enum clear_refs_types)itype;
1095         if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1096                 return -EINVAL;
1097
1098         task = get_proc_task(file_inode(file));
1099         if (!task)
1100                 return -ESRCH;
1101         mm = get_task_mm(task);
1102         if (mm) {
1103                 struct clear_refs_private cp = {
1104                         .type = type,
1105                 };
1106                 struct mm_walk clear_refs_walk = {
1107                         .pmd_entry = clear_refs_pte_range,
1108                         .test_walk = clear_refs_test_walk,
1109                         .mm = mm,
1110                         .private = &cp,
1111                 };
1112
1113                 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1114                         if (down_write_killable(&mm->mmap_sem)) {
1115                                 count = -EINTR;
1116                                 goto out_mm;
1117                         }
1118
1119                         /*
1120                          * Writing 5 to /proc/pid/clear_refs resets the peak
1121                          * resident set size to this mm's current rss value.
1122                          */
1123                         reset_mm_hiwater_rss(mm);
1124                         up_write(&mm->mmap_sem);
1125                         goto out_mm;
1126                 }
1127
1128                 down_read(&mm->mmap_sem);
1129                 tlb_gather_mmu(&tlb, mm, 0, -1);
1130                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1131                         for (vma = mm->mmap; vma; vma = vma->vm_next) {
1132                                 if (!(vma->vm_flags & VM_SOFTDIRTY))
1133                                         continue;
1134                                 up_read(&mm->mmap_sem);
1135                                 if (down_write_killable(&mm->mmap_sem)) {
1136                                         count = -EINTR;
1137                                         goto out_mm;
1138                                 }
1139                                 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1140                                         vma->vm_flags &= ~VM_SOFTDIRTY;
1141                                         vma_set_page_prot(vma);
1142                                 }
1143                                 downgrade_write(&mm->mmap_sem);
1144                                 break;
1145                         }
1146                         mmu_notifier_invalidate_range_start(mm, 0, -1);
1147                 }
1148                 walk_page_range(0, mm->highest_vm_end, &clear_refs_walk);
1149                 if (type == CLEAR_REFS_SOFT_DIRTY)
1150                         mmu_notifier_invalidate_range_end(mm, 0, -1);
1151                 tlb_finish_mmu(&tlb, 0, -1);
1152                 up_read(&mm->mmap_sem);
1153 out_mm:
1154                 mmput(mm);
1155         }
1156         put_task_struct(task);
1157
1158         return count;
1159 }
1160
1161 const struct file_operations proc_clear_refs_operations = {
1162         .write          = clear_refs_write,
1163         .llseek         = noop_llseek,
1164 };
1165
1166 typedef struct {
1167         u64 pme;
1168 } pagemap_entry_t;
1169
1170 struct pagemapread {
1171         int pos, len;           /* units: PM_ENTRY_BYTES, not bytes */
1172         pagemap_entry_t *buffer;
1173         bool show_pfn;
1174 };
1175
1176 #define PAGEMAP_WALK_SIZE       (PMD_SIZE)
1177 #define PAGEMAP_WALK_MASK       (PMD_MASK)
1178
1179 #define PM_ENTRY_BYTES          sizeof(pagemap_entry_t)
1180 #define PM_PFRAME_BITS          55
1181 #define PM_PFRAME_MASK          GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1182 #define PM_SOFT_DIRTY           BIT_ULL(55)
1183 #define PM_MMAP_EXCLUSIVE       BIT_ULL(56)
1184 #define PM_FILE                 BIT_ULL(61)
1185 #define PM_SWAP                 BIT_ULL(62)
1186 #define PM_PRESENT              BIT_ULL(63)
1187
1188 #define PM_END_OF_BUFFER    1
1189
1190 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1191 {
1192         return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1193 }
1194
1195 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1196                           struct pagemapread *pm)
1197 {
1198         pm->buffer[pm->pos++] = *pme;
1199         if (pm->pos >= pm->len)
1200                 return PM_END_OF_BUFFER;
1201         return 0;
1202 }
1203
1204 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1205                                 struct mm_walk *walk)
1206 {
1207         struct pagemapread *pm = walk->private;
1208         unsigned long addr = start;
1209         int err = 0;
1210
1211         while (addr < end) {
1212                 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1213                 pagemap_entry_t pme = make_pme(0, 0);
1214                 /* End of address space hole, which we mark as non-present. */
1215                 unsigned long hole_end;
1216
1217                 if (vma)
1218                         hole_end = min(end, vma->vm_start);
1219                 else
1220                         hole_end = end;
1221
1222                 for (; addr < hole_end; addr += PAGE_SIZE) {
1223                         err = add_to_pagemap(addr, &pme, pm);
1224                         if (err)
1225                                 goto out;
1226                 }
1227
1228                 if (!vma)
1229                         break;
1230
1231                 /* Addresses in the VMA. */
1232                 if (vma->vm_flags & VM_SOFTDIRTY)
1233                         pme = make_pme(0, PM_SOFT_DIRTY);
1234                 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1235                         err = add_to_pagemap(addr, &pme, pm);
1236                         if (err)
1237                                 goto out;
1238                 }
1239         }
1240 out:
1241         return err;
1242 }
1243
1244 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1245                 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1246 {
1247         u64 frame = 0, flags = 0;
1248         struct page *page = NULL;
1249
1250         if (pte_present(pte)) {
1251                 if (pm->show_pfn)
1252                         frame = pte_pfn(pte);
1253                 flags |= PM_PRESENT;
1254                 page = _vm_normal_page(vma, addr, pte, true);
1255                 if (pte_soft_dirty(pte))
1256                         flags |= PM_SOFT_DIRTY;
1257         } else if (is_swap_pte(pte)) {
1258                 swp_entry_t entry;
1259                 if (pte_swp_soft_dirty(pte))
1260                         flags |= PM_SOFT_DIRTY;
1261                 entry = pte_to_swp_entry(pte);
1262                 if (pm->show_pfn)
1263                         frame = swp_type(entry) |
1264                                 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1265                 flags |= PM_SWAP;
1266                 if (is_migration_entry(entry))
1267                         page = migration_entry_to_page(entry);
1268
1269                 if (is_device_private_entry(entry))
1270                         page = device_private_entry_to_page(entry);
1271         }
1272
1273         if (page && !PageAnon(page))
1274                 flags |= PM_FILE;
1275         if (page && page_mapcount(page) == 1)
1276                 flags |= PM_MMAP_EXCLUSIVE;
1277         if (vma->vm_flags & VM_SOFTDIRTY)
1278                 flags |= PM_SOFT_DIRTY;
1279
1280         return make_pme(frame, flags);
1281 }
1282
1283 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1284                              struct mm_walk *walk)
1285 {
1286         struct vm_area_struct *vma = walk->vma;
1287         struct pagemapread *pm = walk->private;
1288         spinlock_t *ptl;
1289         pte_t *pte, *orig_pte;
1290         int err = 0;
1291
1292 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1293         ptl = pmd_trans_huge_lock(pmdp, vma);
1294         if (ptl) {
1295                 u64 flags = 0, frame = 0;
1296                 pmd_t pmd = *pmdp;
1297                 struct page *page = NULL;
1298
1299                 if (vma->vm_flags & VM_SOFTDIRTY)
1300                         flags |= PM_SOFT_DIRTY;
1301
1302                 if (pmd_present(pmd)) {
1303                         page = pmd_page(pmd);
1304
1305                         flags |= PM_PRESENT;
1306                         if (pmd_soft_dirty(pmd))
1307                                 flags |= PM_SOFT_DIRTY;
1308                         if (pm->show_pfn)
1309                                 frame = pmd_pfn(pmd) +
1310                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1311                 }
1312 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1313                 else if (is_swap_pmd(pmd)) {
1314                         swp_entry_t entry = pmd_to_swp_entry(pmd);
1315                         unsigned long offset;
1316
1317                         if (pm->show_pfn) {
1318                                 offset = swp_offset(entry) +
1319                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1320                                 frame = swp_type(entry) |
1321                                         (offset << MAX_SWAPFILES_SHIFT);
1322                         }
1323                         flags |= PM_SWAP;
1324                         if (pmd_swp_soft_dirty(pmd))
1325                                 flags |= PM_SOFT_DIRTY;
1326                         VM_BUG_ON(!is_pmd_migration_entry(pmd));
1327                         page = migration_entry_to_page(entry);
1328                 }
1329 #endif
1330
1331                 if (page && page_mapcount(page) == 1)
1332                         flags |= PM_MMAP_EXCLUSIVE;
1333
1334                 for (; addr != end; addr += PAGE_SIZE) {
1335                         pagemap_entry_t pme = make_pme(frame, flags);
1336
1337                         err = add_to_pagemap(addr, &pme, pm);
1338                         if (err)
1339                                 break;
1340                         if (pm->show_pfn) {
1341                                 if (flags & PM_PRESENT)
1342                                         frame++;
1343                                 else if (flags & PM_SWAP)
1344                                         frame += (1 << MAX_SWAPFILES_SHIFT);
1345                         }
1346                 }
1347                 spin_unlock(ptl);
1348                 return err;
1349         }
1350
1351         if (pmd_trans_unstable(pmdp))
1352                 return 0;
1353 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1354
1355         /*
1356          * We can assume that @vma always points to a valid one and @end never
1357          * goes beyond vma->vm_end.
1358          */
1359         orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1360         for (; addr < end; pte++, addr += PAGE_SIZE) {
1361                 pagemap_entry_t pme;
1362
1363                 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1364                 err = add_to_pagemap(addr, &pme, pm);
1365                 if (err)
1366                         break;
1367         }
1368         pte_unmap_unlock(orig_pte, ptl);
1369
1370         cond_resched();
1371
1372         return err;
1373 }
1374
1375 #ifdef CONFIG_HUGETLB_PAGE
1376 /* This function walks within one hugetlb entry in the single call */
1377 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1378                                  unsigned long addr, unsigned long end,
1379                                  struct mm_walk *walk)
1380 {
1381         struct pagemapread *pm = walk->private;
1382         struct vm_area_struct *vma = walk->vma;
1383         u64 flags = 0, frame = 0;
1384         int err = 0;
1385         pte_t pte;
1386
1387         if (vma->vm_flags & VM_SOFTDIRTY)
1388                 flags |= PM_SOFT_DIRTY;
1389
1390         pte = huge_ptep_get(ptep);
1391         if (pte_present(pte)) {
1392                 struct page *page = pte_page(pte);
1393
1394                 if (!PageAnon(page))
1395                         flags |= PM_FILE;
1396
1397                 if (page_mapcount(page) == 1)
1398                         flags |= PM_MMAP_EXCLUSIVE;
1399
1400                 flags |= PM_PRESENT;
1401                 if (pm->show_pfn)
1402                         frame = pte_pfn(pte) +
1403                                 ((addr & ~hmask) >> PAGE_SHIFT);
1404         }
1405
1406         for (; addr != end; addr += PAGE_SIZE) {
1407                 pagemap_entry_t pme = make_pme(frame, flags);
1408
1409                 err = add_to_pagemap(addr, &pme, pm);
1410                 if (err)
1411                         return err;
1412                 if (pm->show_pfn && (flags & PM_PRESENT))
1413                         frame++;
1414         }
1415
1416         cond_resched();
1417
1418         return err;
1419 }
1420 #endif /* HUGETLB_PAGE */
1421
1422 /*
1423  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1424  *
1425  * For each page in the address space, this file contains one 64-bit entry
1426  * consisting of the following:
1427  *
1428  * Bits 0-54  page frame number (PFN) if present
1429  * Bits 0-4   swap type if swapped
1430  * Bits 5-54  swap offset if swapped
1431  * Bit  55    pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst)
1432  * Bit  56    page exclusively mapped
1433  * Bits 57-60 zero
1434  * Bit  61    page is file-page or shared-anon
1435  * Bit  62    page swapped
1436  * Bit  63    page present
1437  *
1438  * If the page is not present but in swap, then the PFN contains an
1439  * encoding of the swap file number and the page's offset into the
1440  * swap. Unmapped pages return a null PFN. This allows determining
1441  * precisely which pages are mapped (or in swap) and comparing mapped
1442  * pages between processes.
1443  *
1444  * Efficient users of this interface will use /proc/pid/maps to
1445  * determine which areas of memory are actually mapped and llseek to
1446  * skip over unmapped regions.
1447  */
1448 static ssize_t pagemap_read(struct file *file, char __user *buf,
1449                             size_t count, loff_t *ppos)
1450 {
1451         struct mm_struct *mm = file->private_data;
1452         struct pagemapread pm;
1453         struct mm_walk pagemap_walk = {};
1454         unsigned long src;
1455         unsigned long svpfn;
1456         unsigned long start_vaddr;
1457         unsigned long end_vaddr;
1458         int ret = 0, copied = 0;
1459
1460         if (!mm || !mmget_not_zero(mm))
1461                 goto out;
1462
1463         ret = -EINVAL;
1464         /* file position must be aligned */
1465         if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1466                 goto out_mm;
1467
1468         ret = 0;
1469         if (!count)
1470                 goto out_mm;
1471
1472         /* do not disclose physical addresses: attack vector */
1473         pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1474
1475         pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1476         pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL);
1477         ret = -ENOMEM;
1478         if (!pm.buffer)
1479                 goto out_mm;
1480
1481         pagemap_walk.pmd_entry = pagemap_pmd_range;
1482         pagemap_walk.pte_hole = pagemap_pte_hole;
1483 #ifdef CONFIG_HUGETLB_PAGE
1484         pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1485 #endif
1486         pagemap_walk.mm = mm;
1487         pagemap_walk.private = &pm;
1488
1489         src = *ppos;
1490         svpfn = src / PM_ENTRY_BYTES;
1491         start_vaddr = svpfn << PAGE_SHIFT;
1492         end_vaddr = mm->task_size;
1493
1494         /* watch out for wraparound */
1495         if (svpfn > mm->task_size >> PAGE_SHIFT)
1496                 start_vaddr = end_vaddr;
1497
1498         /*
1499          * The odds are that this will stop walking way
1500          * before end_vaddr, because the length of the
1501          * user buffer is tracked in "pm", and the walk
1502          * will stop when we hit the end of the buffer.
1503          */
1504         ret = 0;
1505         while (count && (start_vaddr < end_vaddr)) {
1506                 int len;
1507                 unsigned long end;
1508
1509                 pm.pos = 0;
1510                 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1511                 /* overflow ? */
1512                 if (end < start_vaddr || end > end_vaddr)
1513                         end = end_vaddr;
1514                 down_read(&mm->mmap_sem);
1515                 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1516                 up_read(&mm->mmap_sem);
1517                 start_vaddr = end;
1518
1519                 len = min(count, PM_ENTRY_BYTES * pm.pos);
1520                 if (copy_to_user(buf, pm.buffer, len)) {
1521                         ret = -EFAULT;
1522                         goto out_free;
1523                 }
1524                 copied += len;
1525                 buf += len;
1526                 count -= len;
1527         }
1528         *ppos += copied;
1529         if (!ret || ret == PM_END_OF_BUFFER)
1530                 ret = copied;
1531
1532 out_free:
1533         kfree(pm.buffer);
1534 out_mm:
1535         mmput(mm);
1536 out:
1537         return ret;
1538 }
1539
1540 static int pagemap_open(struct inode *inode, struct file *file)
1541 {
1542         struct mm_struct *mm;
1543
1544         mm = proc_mem_open(inode, PTRACE_MODE_READ);
1545         if (IS_ERR(mm))
1546                 return PTR_ERR(mm);
1547         file->private_data = mm;
1548         return 0;
1549 }
1550
1551 static int pagemap_release(struct inode *inode, struct file *file)
1552 {
1553         struct mm_struct *mm = file->private_data;
1554
1555         if (mm)
1556                 mmdrop(mm);
1557         return 0;
1558 }
1559
1560 const struct file_operations proc_pagemap_operations = {
1561         .llseek         = mem_lseek, /* borrow this */
1562         .read           = pagemap_read,
1563         .open           = pagemap_open,
1564         .release        = pagemap_release,
1565 };
1566 #endif /* CONFIG_PROC_PAGE_MONITOR */
1567
1568 #ifdef CONFIG_NUMA
1569
1570 struct numa_maps {
1571         unsigned long pages;
1572         unsigned long anon;
1573         unsigned long active;
1574         unsigned long writeback;
1575         unsigned long mapcount_max;
1576         unsigned long dirty;
1577         unsigned long swapcache;
1578         unsigned long node[MAX_NUMNODES];
1579 };
1580
1581 struct numa_maps_private {
1582         struct proc_maps_private proc_maps;
1583         struct numa_maps md;
1584 };
1585
1586 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1587                         unsigned long nr_pages)
1588 {
1589         int count = page_mapcount(page);
1590
1591         md->pages += nr_pages;
1592         if (pte_dirty || PageDirty(page))
1593                 md->dirty += nr_pages;
1594
1595         if (PageSwapCache(page))
1596                 md->swapcache += nr_pages;
1597
1598         if (PageActive(page) || PageUnevictable(page))
1599                 md->active += nr_pages;
1600
1601         if (PageWriteback(page))
1602                 md->writeback += nr_pages;
1603
1604         if (PageAnon(page))
1605                 md->anon += nr_pages;
1606
1607         if (count > md->mapcount_max)
1608                 md->mapcount_max = count;
1609
1610         md->node[page_to_nid(page)] += nr_pages;
1611 }
1612
1613 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1614                 unsigned long addr)
1615 {
1616         struct page *page;
1617         int nid;
1618
1619         if (!pte_present(pte))
1620                 return NULL;
1621
1622         page = vm_normal_page(vma, addr, pte);
1623         if (!page)
1624                 return NULL;
1625
1626         if (PageReserved(page))
1627                 return NULL;
1628
1629         nid = page_to_nid(page);
1630         if (!node_isset(nid, node_states[N_MEMORY]))
1631                 return NULL;
1632
1633         return page;
1634 }
1635
1636 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1637 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1638                                               struct vm_area_struct *vma,
1639                                               unsigned long addr)
1640 {
1641         struct page *page;
1642         int nid;
1643
1644         if (!pmd_present(pmd))
1645                 return NULL;
1646
1647         page = vm_normal_page_pmd(vma, addr, pmd);
1648         if (!page)
1649                 return NULL;
1650
1651         if (PageReserved(page))
1652                 return NULL;
1653
1654         nid = page_to_nid(page);
1655         if (!node_isset(nid, node_states[N_MEMORY]))
1656                 return NULL;
1657
1658         return page;
1659 }
1660 #endif
1661
1662 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1663                 unsigned long end, struct mm_walk *walk)
1664 {
1665         struct numa_maps *md = walk->private;
1666         struct vm_area_struct *vma = walk->vma;
1667         spinlock_t *ptl;
1668         pte_t *orig_pte;
1669         pte_t *pte;
1670
1671 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1672         ptl = pmd_trans_huge_lock(pmd, vma);
1673         if (ptl) {
1674                 struct page *page;
1675
1676                 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1677                 if (page)
1678                         gather_stats(page, md, pmd_dirty(*pmd),
1679                                      HPAGE_PMD_SIZE/PAGE_SIZE);
1680                 spin_unlock(ptl);
1681                 return 0;
1682         }
1683
1684         if (pmd_trans_unstable(pmd))
1685                 return 0;
1686 #endif
1687         orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1688         do {
1689                 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1690                 if (!page)
1691                         continue;
1692                 gather_stats(page, md, pte_dirty(*pte), 1);
1693
1694         } while (pte++, addr += PAGE_SIZE, addr != end);
1695         pte_unmap_unlock(orig_pte, ptl);
1696         cond_resched();
1697         return 0;
1698 }
1699 #ifdef CONFIG_HUGETLB_PAGE
1700 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1701                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1702 {
1703         pte_t huge_pte = huge_ptep_get(pte);
1704         struct numa_maps *md;
1705         struct page *page;
1706
1707         if (!pte_present(huge_pte))
1708                 return 0;
1709
1710         page = pte_page(huge_pte);
1711         if (!page)
1712                 return 0;
1713
1714         md = walk->private;
1715         gather_stats(page, md, pte_dirty(huge_pte), 1);
1716         return 0;
1717 }
1718
1719 #else
1720 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1721                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1722 {
1723         return 0;
1724 }
1725 #endif
1726
1727 /*
1728  * Display pages allocated per node and memory policy via /proc.
1729  */
1730 static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1731 {
1732         struct numa_maps_private *numa_priv = m->private;
1733         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1734         struct vm_area_struct *vma = v;
1735         struct numa_maps *md = &numa_priv->md;
1736         struct file *file = vma->vm_file;
1737         struct mm_struct *mm = vma->vm_mm;
1738         struct mm_walk walk = {
1739                 .hugetlb_entry = gather_hugetlb_stats,
1740                 .pmd_entry = gather_pte_stats,
1741                 .private = md,
1742                 .mm = mm,
1743         };
1744         struct mempolicy *pol;
1745         char buffer[64];
1746         int nid;
1747
1748         if (!mm)
1749                 return 0;
1750
1751         /* Ensure we start with an empty set of numa_maps statistics. */
1752         memset(md, 0, sizeof(*md));
1753
1754         pol = __get_vma_policy(vma, vma->vm_start);
1755         if (pol) {
1756                 mpol_to_str(buffer, sizeof(buffer), pol);
1757                 mpol_cond_put(pol);
1758         } else {
1759                 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1760         }
1761
1762         seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1763
1764         if (file) {
1765                 seq_puts(m, " file=");
1766                 seq_file_path(m, file, "\n\t= ");
1767         } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1768                 seq_puts(m, " heap");
1769         } else if (is_stack(vma)) {
1770                 seq_puts(m, " stack");
1771         }
1772
1773         if (is_vm_hugetlb_page(vma))
1774                 seq_puts(m, " huge");
1775
1776         /* mmap_sem is held by m_start */
1777         walk_page_vma(vma, &walk);
1778
1779         if (!md->pages)
1780                 goto out;
1781
1782         if (md->anon)
1783                 seq_printf(m, " anon=%lu", md->anon);
1784
1785         if (md->dirty)
1786                 seq_printf(m, " dirty=%lu", md->dirty);
1787
1788         if (md->pages != md->anon && md->pages != md->dirty)
1789                 seq_printf(m, " mapped=%lu", md->pages);
1790
1791         if (md->mapcount_max > 1)
1792                 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1793
1794         if (md->swapcache)
1795                 seq_printf(m, " swapcache=%lu", md->swapcache);
1796
1797         if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1798                 seq_printf(m, " active=%lu", md->active);
1799
1800         if (md->writeback)
1801                 seq_printf(m, " writeback=%lu", md->writeback);
1802
1803         for_each_node_state(nid, N_MEMORY)
1804                 if (md->node[nid])
1805                         seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1806
1807         seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1808 out:
1809         seq_putc(m, '\n');
1810         m_cache_vma(m, vma);
1811         return 0;
1812 }
1813
1814 static int show_pid_numa_map(struct seq_file *m, void *v)
1815 {
1816         return show_numa_map(m, v, 1);
1817 }
1818
1819 static int show_tid_numa_map(struct seq_file *m, void *v)
1820 {
1821         return show_numa_map(m, v, 0);
1822 }
1823
1824 static const struct seq_operations proc_pid_numa_maps_op = {
1825         .start  = m_start,
1826         .next   = m_next,
1827         .stop   = m_stop,
1828         .show   = show_pid_numa_map,
1829 };
1830
1831 static const struct seq_operations proc_tid_numa_maps_op = {
1832         .start  = m_start,
1833         .next   = m_next,
1834         .stop   = m_stop,
1835         .show   = show_tid_numa_map,
1836 };
1837
1838 static int numa_maps_open(struct inode *inode, struct file *file,
1839                           const struct seq_operations *ops)
1840 {
1841         return proc_maps_open(inode, file, ops,
1842                                 sizeof(struct numa_maps_private));
1843 }
1844
1845 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1846 {
1847         return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1848 }
1849
1850 static int tid_numa_maps_open(struct inode *inode, struct file *file)
1851 {
1852         return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1853 }
1854
1855 const struct file_operations proc_pid_numa_maps_operations = {
1856         .open           = pid_numa_maps_open,
1857         .read           = seq_read,
1858         .llseek         = seq_lseek,
1859         .release        = proc_map_release,
1860 };
1861
1862 const struct file_operations proc_tid_numa_maps_operations = {
1863         .open           = tid_numa_maps_open,
1864         .read           = seq_read,
1865         .llseek         = seq_lseek,
1866         .release        = proc_map_release,
1867 };
1868 #endif /* CONFIG_NUMA */