fs/proc/task_mmu.c

   1 #include <linux/mm.h>
   2 #include <linux/vmacache.h>
   3 #include <linux/hugetlb.h>
   4 #include <linux/huge_mm.h>
   5 #include <linux/mount.h>
   6 #include <linux/seq_file.h>
   7 #include <linux/highmem.h>
   8 #include <linux/ptrace.h>
   9 #include <linux/slab.h>
  10 #include <linux/pagemap.h>
  11 #include <linux/mempolicy.h>
  12 #include <linux/rmap.h>
  13 #include <linux/swap.h>
  14 #include <linux/sched/mm.h>
  15 #include <linux/swapops.h>
  16 #include <linux/mmu_notifier.h>
  17 #include <linux/page_idle.h>
  18 #include <linux/shmem_fs.h>
  19 #include <linux/uaccess.h>
  20
  21 #include <asm/elf.h>
  22 #include <asm/tlb.h>
  23 #include <asm/tlbflush.h>
  24 #include "internal.h"
  25
  26 void task_mem(struct seq_file *m, struct mm_struct *mm)
  27 {
  28         unsigned long text, lib, swap, ptes, pmds, anon, file, shmem;
  29         unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
  30
  31         anon = get_mm_counter(mm, MM_ANONPAGES);
  32         file = get_mm_counter(mm, MM_FILEPAGES);
  33         shmem = get_mm_counter(mm, MM_SHMEMPAGES);
  34
  35         /*
  36          * Note: to minimize their overhead, mm maintains hiwater_vm and
  37          * hiwater_rss only when about to *lower* total_vm or rss.  Any
  38          * collector of these hiwater stats must therefore get total_vm
  39          * and rss too, which will usually be the higher.  Barriers? not
  40          * worth the effort, such snapshots can always be inconsistent.
  41          */
  42         hiwater_vm = total_vm = mm->total_vm;
  43         if (hiwater_vm < mm->hiwater_vm)
  44                 hiwater_vm = mm->hiwater_vm;
  45         hiwater_rss = total_rss = anon + file + shmem;
  46         if (hiwater_rss < mm->hiwater_rss)
  47                 hiwater_rss = mm->hiwater_rss;
  48
  49         text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
  50         lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
  51         swap = get_mm_counter(mm, MM_SWAPENTS);
  52         ptes = PTRS_PER_PTE * sizeof(pte_t) * atomic_long_read(&mm->nr_ptes);
  53         pmds = PTRS_PER_PMD * sizeof(pmd_t) * mm_nr_pmds(mm);
  54         seq_printf(m,
  55                 "VmPeak:\t%8lu kB\n"
  56                 "VmSize:\t%8lu kB\n"
  57                 "VmLck:\t%8lu kB\n"
  58                 "VmPin:\t%8lu kB\n"
  59                 "VmHWM:\t%8lu kB\n"
  60                 "VmRSS:\t%8lu kB\n"
  61                 "RssAnon:\t%8lu kB\n"
  62                 "RssFile:\t%8lu kB\n"
  63                 "RssShmem:\t%8lu kB\n"
  64                 "VmData:\t%8lu kB\n"
  65                 "VmStk:\t%8lu kB\n"
  66                 "VmExe:\t%8lu kB\n"
  67                 "VmLib:\t%8lu kB\n"
  68                 "VmPTE:\t%8lu kB\n"
  69                 "VmPMD:\t%8lu kB\n"
  70                 "VmSwap:\t%8lu kB\n",
  71                 hiwater_vm << (PAGE_SHIFT-10),
  72                 total_vm << (PAGE_SHIFT-10),
  73                 mm->locked_vm << (PAGE_SHIFT-10),
  74                 mm->pinned_vm << (PAGE_SHIFT-10),
  75                 hiwater_rss << (PAGE_SHIFT-10),
  76                 total_rss << (PAGE_SHIFT-10),
  77                 anon << (PAGE_SHIFT-10),
  78                 file << (PAGE_SHIFT-10),
  79                 shmem << (PAGE_SHIFT-10),
  80                 mm->data_vm << (PAGE_SHIFT-10),
  81                 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
  82                 ptes >> 10,
  83                 pmds >> 10,
  84                 swap << (PAGE_SHIFT-10));
  85         hugetlb_report_usage(m, mm);
  86 }
  87
  88 unsigned long task_vsize(struct mm_struct *mm)
  89 {
  90         return PAGE_SIZE * mm->total_vm;
  91 }
  92
  93 unsigned long task_statm(struct mm_struct *mm,
  94                          unsigned long *shared, unsigned long *text,
  95                          unsigned long *data, unsigned long *resident)
  96 {
  97         *shared = get_mm_counter(mm, MM_FILEPAGES) +
  98                         get_mm_counter(mm, MM_SHMEMPAGES);
  99         *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
 100                                                                 >> PAGE_SHIFT;
 101         *data = mm->data_vm + mm->stack_vm;
 102         *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
 103         return mm->total_vm;
 104 }
 105
 106 #ifdef CONFIG_NUMA
 107 /*
 108  * Save get_task_policy() for show_numa_map().
 109  */
 110 static void hold_task_mempolicy(struct proc_maps_private *priv)
 111 {
 112         struct task_struct *task = priv->task;
 113
 114         task_lock(task);
 115         priv->task_mempolicy = get_task_policy(task);
 116         mpol_get(priv->task_mempolicy);
 117         task_unlock(task);
 118 }
 119 static void release_task_mempolicy(struct proc_maps_private *priv)
 120 {
 121         mpol_put(priv->task_mempolicy);
 122 }
 123 #else
 124 static void hold_task_mempolicy(struct proc_maps_private *priv)
 125 {
 126 }
 127 static void release_task_mempolicy(struct proc_maps_private *priv)
 128 {
 129 }
 130 #endif
 131
 132 static void vma_stop(struct proc_maps_private *priv)
 133 {
 134         struct mm_struct *mm = priv->mm;
 135
 136         release_task_mempolicy(priv);
 137         up_read(&mm->mmap_sem);
 138         mmput(mm);
 139 }
 140
 141 static struct vm_area_struct *
 142 m_next_vma(struct proc_maps_private *priv, struct vm_area_struct *vma)
 143 {
 144         if (vma == priv->tail_vma)
 145                 return NULL;
 146         return vma->vm_next ?: priv->tail_vma;
 147 }
 148
 149 static void m_cache_vma(struct seq_file *m, struct vm_area_struct *vma)
 150 {
 151         if (m->count < m->size) /* vma is copied successfully */
 152                 m->version = m_next_vma(m->private, vma) ? vma->vm_end : -1UL;
 153 }
 154
 155 static void *m_start(struct seq_file *m, loff_t *ppos)
 156 {
 157         struct proc_maps_private *priv = m->private;
 158         unsigned long last_addr = m->version;
 159         struct mm_struct *mm;
 160         struct vm_area_struct *vma;
 161         unsigned int pos = *ppos;
 162
 163         /* See m_cache_vma(). Zero at the start or after lseek. */
 164         if (last_addr == -1UL)
 165                 return NULL;
 166
 167         priv->task = get_proc_task(priv->inode);
 168         if (!priv->task)
 169                 return ERR_PTR(-ESRCH);
 170
 171         mm = priv->mm;
 172         if (!mm || !mmget_not_zero(mm))
 173                 return NULL;
 174
 175         down_read(&mm->mmap_sem);
 176         hold_task_mempolicy(priv);
 177         priv->tail_vma = get_gate_vma(mm);
 178
 179         if (last_addr) {
 180                 vma = find_vma(mm, last_addr - 1);
 181                 if (vma && vma->vm_start <= last_addr)
 182                         vma = m_next_vma(priv, vma);
 183                 if (vma)
 184                         return vma;
 185         }
 186
 187         m->version = 0;
 188         if (pos < mm->map_count) {
 189                 for (vma = mm->mmap; pos; pos--) {
 190                         m->version = vma->vm_start;
 191                         vma = vma->vm_next;
 192                 }
 193                 return vma;
 194         }
 195
 196         /* we do not bother to update m->version in this case */
 197         if (pos == mm->map_count && priv->tail_vma)
 198                 return priv->tail_vma;
 199
 200         vma_stop(priv);
 201         return NULL;
 202 }
 203
 204 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
 205 {
 206         struct proc_maps_private *priv = m->private;
 207         struct vm_area_struct *next;
 208
 209         (*pos)++;
 210         next = m_next_vma(priv, v);
 211         if (!next)
 212                 vma_stop(priv);
 213         return next;
 214 }
 215
 216 static void m_stop(struct seq_file *m, void *v)
 217 {
 218         struct proc_maps_private *priv = m->private;
 219
 220         if (!IS_ERR_OR_NULL(v))
 221                 vma_stop(priv);
 222         if (priv->task) {
 223                 put_task_struct(priv->task);
 224                 priv->task = NULL;
 225         }
 226 }
 227
 228 static int proc_maps_open(struct inode *inode, struct file *file,
 229                         const struct seq_operations *ops, int psize)
 230 {
 231         struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
 232
 233         if (!priv)
 234                 return -ENOMEM;
 235
 236         priv->inode = inode;
 237         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
 238         if (IS_ERR(priv->mm)) {
 239                 int err = PTR_ERR(priv->mm);
 240
 241                 seq_release_private(inode, file);
 242                 return err;
 243         }
 244
 245         return 0;
 246 }
 247
 248 static int proc_map_release(struct inode *inode, struct file *file)
 249 {
 250         struct seq_file *seq = file->private_data;
 251         struct proc_maps_private *priv = seq->private;
 252
 253         if (priv->mm)
 254                 mmdrop(priv->mm);
 255
 256         kfree(priv->rollup);
 257         return seq_release_private(inode, file);
 258 }
 259
 260 static int do_maps_open(struct inode *inode, struct file *file,
 261                         const struct seq_operations *ops)
 262 {
 263         return proc_maps_open(inode, file, ops,
 264                                 sizeof(struct proc_maps_private));
 265 }
 266
 267 /*
 268  * Indicate if the VMA is a stack for the given task; for
 269  * /proc/PID/maps that is the stack of the main task.
 270  */
 271 static int is_stack(struct proc_maps_private *priv,
 272                     struct vm_area_struct *vma)
 273 {
 274         /*
 275          * We make no effort to guess what a given thread considers to be
 276          * its "stack".  It's not even well-defined for programs written
 277          * languages like Go.
 278          */
 279         return vma->vm_start <= vma->vm_mm->start_stack &&
 280                 vma->vm_end >= vma->vm_mm->start_stack;
 281 }
 282
 283 static void show_vma_header_prefix(struct seq_file *m,
 284                                    unsigned long start, unsigned long end,
 285                                    vm_flags_t flags, unsigned long long pgoff,
 286                                    dev_t dev, unsigned long ino)
 287 {
 288         seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
 289         seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu ",
 290                    start,
 291                    end,
 292                    flags & VM_READ ? 'r' : '-',
 293                    flags & VM_WRITE ? 'w' : '-',
 294                    flags & VM_EXEC ? 'x' : '-',
 295                    flags & VM_MAYSHARE ? 's' : 'p',
 296                    pgoff,
 297                    MAJOR(dev), MINOR(dev), ino);
 298 }
 299
 300 static void
 301 show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
 302 {
 303         struct mm_struct *mm = vma->vm_mm;
 304         struct file *file = vma->vm_file;
 305         struct proc_maps_private *priv = m->private;
 306         vm_flags_t flags = vma->vm_flags;
 307         unsigned long ino = 0;
 308         unsigned long long pgoff = 0;
 309         unsigned long start, end;
 310         dev_t dev = 0;
 311         const char *name = NULL;
 312
 313         if (file) {
 314                 struct inode *inode = file_inode(vma->vm_file);
 315                 dev = inode->i_sb->s_dev;
 316                 ino = inode->i_ino;
 317                 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
 318         }
 319
 320         start = vma->vm_start;
 321         end = vma->vm_end;
 322         show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino);
 323
 324         /*
 325          * Print the dentry name for named mappings, and a
 326          * special [heap] marker for the heap:
 327          */
 328         if (file) {
 329                 seq_pad(m, ' ');
 330                 seq_file_path(m, file, "\n");
 331                 goto done;
 332         }
 333
 334         if (vma->vm_ops && vma->vm_ops->name) {
 335                 name = vma->vm_ops->name(vma);
 336                 if (name)
 337                         goto done;
 338         }
 339
 340         name = arch_vma_name(vma);
 341         if (!name) {
 342                 if (!mm) {
 343                         name = "[vdso]";
 344                         goto done;
 345                 }
 346
 347                 if (vma->vm_start <= mm->brk &&
 348                     vma->vm_end >= mm->start_brk) {
 349                         name = "[heap]";
 350                         goto done;
 351                 }
 352
 353                 if (is_stack(priv, vma))
 354                         name = "[stack]";
 355         }
 356
 357 done:
 358         if (name) {
 359                 seq_pad(m, ' ');
 360                 seq_puts(m, name);
 361         }
 362         seq_putc(m, '\n');
 363 }
 364
 365 static int show_map(struct seq_file *m, void *v, int is_pid)
 366 {
 367         show_map_vma(m, v, is_pid);
 368         m_cache_vma(m, v);
 369         return 0;
 370 }
 371
 372 static int show_pid_map(struct seq_file *m, void *v)
 373 {
 374         return show_map(m, v, 1);
 375 }
 376
 377 static int show_tid_map(struct seq_file *m, void *v)
 378 {
 379         return show_map(m, v, 0);
 380 }
 381
 382 static const struct seq_operations proc_pid_maps_op = {
 383         .start  = m_start,
 384         .next   = m_next,
 385         .stop   = m_stop,
 386         .show   = show_pid_map
 387 };
 388
 389 static const struct seq_operations proc_tid_maps_op = {
 390         .start  = m_start,
 391         .next   = m_next,
 392         .stop   = m_stop,
 393         .show   = show_tid_map
 394 };
 395
 396 static int pid_maps_open(struct inode *inode, struct file *file)
 397 {
 398         return do_maps_open(inode, file, &proc_pid_maps_op);
 399 }
 400
 401 static int tid_maps_open(struct inode *inode, struct file *file)
 402 {
 403         return do_maps_open(inode, file, &proc_tid_maps_op);
 404 }
 405
 406 const struct file_operations proc_pid_maps_operations = {
 407         .open           = pid_maps_open,
 408         .read           = seq_read,
 409         .llseek         = seq_lseek,
 410         .release        = proc_map_release,
 411 };
 412
 413 const struct file_operations proc_tid_maps_operations = {
 414         .open           = tid_maps_open,
 415         .read           = seq_read,
 416         .llseek         = seq_lseek,
 417         .release        = proc_map_release,
 418 };
 419
 420 /*
 421  * Proportional Set Size(PSS): my share of RSS.
 422  *
 423  * PSS of a process is the count of pages it has in memory, where each
 424  * page is divided by the number of processes sharing it.  So if a
 425  * process has 1000 pages all to itself, and 1000 shared with one other
 426  * process, its PSS will be 1500.
 427  *
 428  * To keep (accumulated) division errors low, we adopt a 64bit
 429  * fixed-point pss counter to minimize division errors. So (pss >>
 430  * PSS_SHIFT) would be the real byte count.
 431  *
 432  * A shift of 12 before division means (assuming 4K page size):
 433  *      - 1M 3-user-pages add up to 8KB errors;
 434  *      - supports mapcount up to 2^24, or 16M;
 435  *      - supports PSS up to 2^52 bytes, or 4PB.
 436  */
 437 #define PSS_SHIFT 12
 438
 439 #ifdef CONFIG_PROC_PAGE_MONITOR
 440 struct mem_size_stats {
 441         bool first;
 442         unsigned long resident;
 443         unsigned long shared_clean;
 444         unsigned long shared_dirty;
 445         unsigned long private_clean;
 446         unsigned long private_dirty;
 447         unsigned long referenced;
 448         unsigned long anonymous;
 449         unsigned long lazyfree;
 450         unsigned long anonymous_thp;
 451         unsigned long shmem_thp;
 452         unsigned long swap;
 453         unsigned long shared_hugetlb;
 454         unsigned long private_hugetlb;
 455         unsigned long first_vma_start;
 456         u64 pss;
 457         u64 pss_locked;
 458         u64 swap_pss;
 459         bool check_shmem_swap;
 460 };
 461
 462 static void smaps_account(struct mem_size_stats *mss, struct page *page,
 463                 bool compound, bool young, bool dirty)
 464 {
 465         int i, nr = compound ? 1 << compound_order(page) : 1;
 466         unsigned long size = nr * PAGE_SIZE;
 467
 468         if (PageAnon(page)) {
 469                 mss->anonymous += size;
 470                 if (!PageSwapBacked(page) && !dirty && !PageDirty(page))
 471                         mss->lazyfree += size;
 472         }
 473
 474         mss->resident += size;
 475         /* Accumulate the size in pages that have been accessed. */
 476         if (young || page_is_young(page) || PageReferenced(page))
 477                 mss->referenced += size;
 478
 479         /*
 480          * page_count(page) == 1 guarantees the page is mapped exactly once.
 481          * If any subpage of the compound page mapped with PTE it would elevate
 482          * page_count().
 483          */
 484         if (page_count(page) == 1) {
 485                 if (dirty || PageDirty(page))
 486                         mss->private_dirty += size;
 487                 else
 488                         mss->private_clean += size;
 489                 mss->pss += (u64)size << PSS_SHIFT;
 490                 return;
 491         }
 492
 493         for (i = 0; i < nr; i++, page++) {
 494                 int mapcount = page_mapcount(page);
 495
 496                 if (mapcount >= 2) {
 497                         if (dirty || PageDirty(page))
 498                                 mss->shared_dirty += PAGE_SIZE;
 499                         else
 500                                 mss->shared_clean += PAGE_SIZE;
 501                         mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount;
 502                 } else {
 503                         if (dirty || PageDirty(page))
 504                                 mss->private_dirty += PAGE_SIZE;
 505                         else
 506                                 mss->private_clean += PAGE_SIZE;
 507                         mss->pss += PAGE_SIZE << PSS_SHIFT;
 508                 }
 509         }
 510 }
 511
 512 #ifdef CONFIG_SHMEM
 513 static int smaps_pte_hole(unsigned long addr, unsigned long end,
 514                 struct mm_walk *walk)
 515 {
 516         struct mem_size_stats *mss = walk->private;
 517
 518         mss->swap += shmem_partial_swap_usage(
 519                         walk->vma->vm_file->f_mapping, addr, end);
 520
 521         return 0;
 522 }
 523 #endif
 524
 525 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
 526                 struct mm_walk *walk)
 527 {
 528         struct mem_size_stats *mss = walk->private;
 529         struct vm_area_struct *vma = walk->vma;
 530         struct page *page = NULL;
 531
 532         if (pte_present(*pte)) {
 533                 page = vm_normal_page(vma, addr, *pte);
 534         } else if (is_swap_pte(*pte)) {
 535                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 536
 537                 if (!non_swap_entry(swpent)) {
 538                         int mapcount;
 539
 540                         mss->swap += PAGE_SIZE;
 541                         mapcount = swp_swapcount(swpent);
 542                         if (mapcount >= 2) {
 543                                 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
 544
 545                                 do_div(pss_delta, mapcount);
 546                                 mss->swap_pss += pss_delta;
 547                         } else {
 548                                 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
 549                         }
 550                 } else if (is_migration_entry(swpent))
 551                         page = migration_entry_to_page(swpent);
 552                 else if (is_device_private_entry(swpent))
 553                         page = device_private_entry_to_page(swpent);
 554         } else if (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap
 555                                                         && pte_none(*pte))) {
 556                 page = find_get_entry(vma->vm_file->f_mapping,
 557                                                 linear_page_index(vma, addr));
 558                 if (!page)
 559                         return;
 560
 561                 if (radix_tree_exceptional_entry(page))
 562                         mss->swap += PAGE_SIZE;
 563                 else
 564                         put_page(page);
 565
 566                 return;
 567         }
 568
 569         if (!page)
 570                 return;
 571
 572         smaps_account(mss, page, false, pte_young(*pte), pte_dirty(*pte));
 573 }
 574
 575 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 576 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 577                 struct mm_walk *walk)
 578 {
 579         struct mem_size_stats *mss = walk->private;
 580         struct vm_area_struct *vma = walk->vma;
 581         struct page *page;
 582
 583         /* FOLL_DUMP will return -EFAULT on huge zero page */
 584         page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
 585         if (IS_ERR_OR_NULL(page))
 586                 return;
 587         if (PageAnon(page))
 588                 mss->anonymous_thp += HPAGE_PMD_SIZE;
 589         else if (PageSwapBacked(page))
 590                 mss->shmem_thp += HPAGE_PMD_SIZE;
 591         else if (is_zone_device_page(page))
 592                 /* pass */;
 593         else
 594                 VM_BUG_ON_PAGE(1, page);
 595         smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd));
 596 }
 597 #else
 598 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 599                 struct mm_walk *walk)
 600 {
 601 }
 602 #endif
 603
 604 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 605                            struct mm_walk *walk)
 606 {
 607         struct vm_area_struct *vma = walk->vma;
 608         pte_t *pte;
 609         spinlock_t *ptl;
 610
 611         ptl = pmd_trans_huge_lock(pmd, vma);
 612         if (ptl) {
 613                 if (pmd_present(*pmd))
 614                         smaps_pmd_entry(pmd, addr, walk);
 615                 spin_unlock(ptl);
 616                 return 0;
 617         }
 618
 619         if (pmd_trans_unstable(pmd))
 620                 return 0;
 621         /*
 622          * The mmap_sem held all the way back in m_start() is what
 623          * keeps khugepaged out of here and from collapsing things
 624          * in here.
 625          */
 626         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 627         for (; addr != end; pte++, addr += PAGE_SIZE)
 628                 smaps_pte_entry(pte, addr, walk);
 629         pte_unmap_unlock(pte - 1, ptl);
 630         cond_resched();
 631         return 0;
 632 }
 633
 634 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
 635 {
 636         /*
 637          * Don't forget to update Documentation/ on changes.
 638          */
 639         static const char mnemonics[BITS_PER_LONG][2] = {
 640                 /*
 641                  * In case if we meet a flag we don't know about.
 642                  */
 643                 [0 ... (BITS_PER_LONG-1)] = "??",
 644
 645                 [ilog2(VM_READ)]        = "rd",
 646                 [ilog2(VM_WRITE)]       = "wr",
 647                 [ilog2(VM_EXEC)]        = "ex",
 648                 [ilog2(VM_SHARED)]      = "sh",
 649                 [ilog2(VM_MAYREAD)]     = "mr",
 650                 [ilog2(VM_MAYWRITE)]    = "mw",
 651                 [ilog2(VM_MAYEXEC)]     = "me",
 652                 [ilog2(VM_MAYSHARE)]    = "ms",
 653                 [ilog2(VM_GROWSDOWN)]   = "gd",
 654                 [ilog2(VM_PFNMAP)]      = "pf",
 655                 [ilog2(VM_DENYWRITE)]   = "dw",
 656 #ifdef CONFIG_X86_INTEL_MPX
 657                 [ilog2(VM_MPX)]         = "mp",
 658 #endif
 659                 [ilog2(VM_LOCKED)]      = "lo",
 660                 [ilog2(VM_IO)]          = "io",
 661                 [ilog2(VM_SEQ_READ)]    = "sr",
 662                 [ilog2(VM_RAND_READ)]   = "rr",
 663                 [ilog2(VM_DONTCOPY)]    = "dc",
 664                 [ilog2(VM_DONTEXPAND)]  = "de",
 665                 [ilog2(VM_ACCOUNT)]     = "ac",
 666                 [ilog2(VM_NORESERVE)]   = "nr",
 667                 [ilog2(VM_HUGETLB)]     = "ht",
 668                 [ilog2(VM_ARCH_1)]      = "ar",
 669                 [ilog2(VM_WIPEONFORK)]  = "wf",
 670                 [ilog2(VM_DONTDUMP)]    = "dd",
 671 #ifdef CONFIG_MEM_SOFT_DIRTY
 672                 [ilog2(VM_SOFTDIRTY)]   = "sd",
 673 #endif
 674                 [ilog2(VM_MIXEDMAP)]    = "mm",
 675                 [ilog2(VM_HUGEPAGE)]    = "hg",
 676                 [ilog2(VM_NOHUGEPAGE)]  = "nh",
 677                 [ilog2(VM_MERGEABLE)]   = "mg",
 678                 [ilog2(VM_UFFD_MISSING)]= "um",
 679                 [ilog2(VM_UFFD_WP)]     = "uw",
 680 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
 681                 /* These come out via ProtectionKey: */
 682                 [ilog2(VM_PKEY_BIT0)]   = "",
 683                 [ilog2(VM_PKEY_BIT1)]   = "",
 684                 [ilog2(VM_PKEY_BIT2)]   = "",
 685                 [ilog2(VM_PKEY_BIT3)]   = "",
 686 #endif
 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_printf(m, "%c%c ",
 696                                    mnemonics[i][0], mnemonics[i][1]);
 697                 }
 698         }
 699         seq_putc(m, '\n');
 700 }
 701
 702 #ifdef CONFIG_HUGETLB_PAGE
 703 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
 704                                  unsigned long addr, unsigned long end,
 705                                  struct mm_walk *walk)
 706 {
 707         struct mem_size_stats *mss = walk->private;
 708         struct vm_area_struct *vma = walk->vma;
 709         struct page *page = NULL;
 710
 711         if (pte_present(*pte)) {
 712                 page = vm_normal_page(vma, addr, *pte);
 713         } else if (is_swap_pte(*pte)) {
 714                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 715
 716                 if (is_migration_entry(swpent))
 717                         page = migration_entry_to_page(swpent);
 718                 else if (is_device_private_entry(swpent))
 719                         page = device_private_entry_to_page(swpent);
 720         }
 721         if (page) {
 722                 int mapcount = page_mapcount(page);
 723
 724                 if (mapcount >= 2)
 725                         mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
 726                 else
 727                         mss->private_hugetlb += huge_page_size(hstate_vma(vma));
 728         }
 729         return 0;
 730 }
 731 #endif /* HUGETLB_PAGE */
 732
 733 void __weak arch_show_smap(struct seq_file *m, struct vm_area_struct *vma)
 734 {
 735 }
 736
 737 static int show_smap(struct seq_file *m, void *v, int is_pid)
 738 {
 739         struct proc_maps_private *priv = m->private;
 740         struct vm_area_struct *vma = v;
 741         struct mem_size_stats mss_stack;
 742         struct mem_size_stats *mss;
 743         struct mm_walk smaps_walk = {
 744                 .pmd_entry = smaps_pte_range,
 745 #ifdef CONFIG_HUGETLB_PAGE
 746                 .hugetlb_entry = smaps_hugetlb_range,
 747 #endif
 748                 .mm = vma->vm_mm,
 749         };
 750         int ret = 0;
 751         bool rollup_mode;
 752         bool last_vma;
 753
 754         if (priv->rollup) {
 755                 rollup_mode = true;
 756                 mss = priv->rollup;
 757                 if (mss->first) {
 758                         mss->first_vma_start = vma->vm_start;
 759                         mss->first = false;
 760                 }
 761                 last_vma = !m_next_vma(priv, vma);
 762         } else {
 763                 rollup_mode = false;
 764                 memset(&mss_stack, 0, sizeof(mss_stack));
 765                 mss = &mss_stack;
 766         }
 767
 768         smaps_walk.private = mss;
 769
 770 #ifdef CONFIG_SHMEM
 771         if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
 772                 /*
 773                  * For shared or readonly shmem mappings we know that all
 774                  * swapped out pages belong to the shmem object, and we can
 775                  * obtain the swap value much more efficiently. For private
 776                  * writable mappings, we might have COW pages that are
 777                  * not affected by the parent swapped out pages of the shmem
 778                  * object, so we have to distinguish them during the page walk.
 779                  * Unless we know that the shmem object (or the part mapped by
 780                  * our VMA) has no swapped out pages at all.
 781                  */
 782                 unsigned long shmem_swapped = shmem_swap_usage(vma);
 783
 784                 if (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
 785                                         !(vma->vm_flags & VM_WRITE)) {
 786                         mss->swap = shmem_swapped;
 787                 } else {
 788                         mss->check_shmem_swap = true;
 789                         smaps_walk.pte_hole = smaps_pte_hole;
 790                 }
 791         }
 792 #endif
 793
 794         /* mmap_sem is held in m_start */
 795         walk_page_vma(vma, &smaps_walk);
 796         if (vma->vm_flags & VM_LOCKED)
 797                 mss->pss_locked += mss->pss;
 798
 799         if (!rollup_mode) {
 800                 show_map_vma(m, vma, is_pid);
 801         } else if (last_vma) {
 802                 show_vma_header_prefix(
 803                         m, mss->first_vma_start, vma->vm_end, 0, 0, 0, 0);
 804                 seq_pad(m, ' ');
 805                 seq_puts(m, "[rollup]\n");
 806         } else {
 807                 ret = SEQ_SKIP;
 808         }
 809
 810         if (!rollup_mode)
 811                 seq_printf(m,
 812                            "Size:           %8lu kB\n"
 813                            "KernelPageSize: %8lu kB\n"
 814                            "MMUPageSize:    %8lu kB\n",
 815                            (vma->vm_end - vma->vm_start) >> 10,
 816                            vma_kernel_pagesize(vma) >> 10,
 817                            vma_mmu_pagesize(vma) >> 10);
 818
 819
 820         if (!rollup_mode || last_vma)
 821                 seq_printf(m,
 822                            "Rss:            %8lu kB\n"
 823                            "Pss:            %8lu kB\n"
 824                            "Shared_Clean:   %8lu kB\n"
 825                            "Shared_Dirty:   %8lu kB\n"
 826                            "Private_Clean:  %8lu kB\n"
 827                            "Private_Dirty:  %8lu kB\n"
 828                            "Referenced:     %8lu kB\n"
 829                            "Anonymous:      %8lu kB\n"
 830                            "LazyFree:       %8lu kB\n"
 831                            "AnonHugePages:  %8lu kB\n"
 832                            "ShmemPmdMapped: %8lu kB\n"
 833                            "Shared_Hugetlb: %8lu kB\n"
 834                            "Private_Hugetlb: %7lu kB\n"
 835                            "Swap:           %8lu kB\n"
 836                            "SwapPss:        %8lu kB\n"
 837                            "Locked:         %8lu kB\n",
 838                            mss->resident >> 10,
 839                            (unsigned long)(mss->pss >> (10 + PSS_SHIFT)),
 840                            mss->shared_clean  >> 10,
 841                            mss->shared_dirty  >> 10,
 842                            mss->private_clean >> 10,
 843                            mss->private_dirty >> 10,
 844                            mss->referenced >> 10,
 845                            mss->anonymous >> 10,
 846                            mss->lazyfree >> 10,
 847                            mss->anonymous_thp >> 10,
 848                            mss->shmem_thp >> 10,
 849                            mss->shared_hugetlb >> 10,
 850                            mss->private_hugetlb >> 10,
 851                            mss->swap >> 10,
 852                            (unsigned long)(mss->swap_pss >> (10 + PSS_SHIFT)),
 853                            (unsigned long)(mss->pss >> (10 + PSS_SHIFT)));
 854
 855         if (!rollup_mode) {
 856                 arch_show_smap(m, vma);
 857                 show_smap_vma_flags(m, vma);
 858         }
 859         m_cache_vma(m, vma);
 860         return ret;
 861 }
 862
 863 static int show_pid_smap(struct seq_file *m, void *v)
 864 {
 865         return show_smap(m, v, 1);
 866 }
 867
 868 static int show_tid_smap(struct seq_file *m, void *v)
 869 {
 870         return show_smap(m, v, 0);
 871 }
 872
 873 static const struct seq_operations proc_pid_smaps_op = {
 874         .start  = m_start,
 875         .next   = m_next,
 876         .stop   = m_stop,
 877         .show   = show_pid_smap
 878 };
 879
 880 static const struct seq_operations proc_tid_smaps_op = {
 881         .start  = m_start,
 882         .next   = m_next,
 883         .stop   = m_stop,
 884         .show   = show_tid_smap
 885 };
 886
 887 static int pid_smaps_open(struct inode *inode, struct file *file)
 888 {
 889         return do_maps_open(inode, file, &proc_pid_smaps_op);
 890 }
 891
 892 static int pid_smaps_rollup_open(struct inode *inode, struct file *file)
 893 {
 894         struct seq_file *seq;
 895         struct proc_maps_private *priv;
 896         int ret = do_maps_open(inode, file, &proc_pid_smaps_op);
 897
 898         if (ret < 0)
 899                 return ret;
 900         seq = file->private_data;
 901         priv = seq->private;
 902         priv->rollup = kzalloc(sizeof(*priv->rollup), GFP_KERNEL);
 903         if (!priv->rollup) {
 904                 proc_map_release(inode, file);
 905                 return -ENOMEM;
 906         }
 907         priv->rollup->first = true;
 908         return 0;
 909 }
 910
 911 static int tid_smaps_open(struct inode *inode, struct file *file)
 912 {
 913         return do_maps_open(inode, file, &proc_tid_smaps_op);
 914 }
 915
 916 const struct file_operations proc_pid_smaps_operations = {
 917         .open           = pid_smaps_open,
 918         .read           = seq_read,
 919         .llseek         = seq_lseek,
 920         .release        = proc_map_release,
 921 };
 922
 923 const struct file_operations proc_pid_smaps_rollup_operations = {
 924         .open           = pid_smaps_rollup_open,
 925         .read           = seq_read,
 926         .llseek         = seq_lseek,
 927         .release        = proc_map_release,
 928 };
 929
 930 const struct file_operations proc_tid_smaps_operations = {
 931         .open           = tid_smaps_open,
 932         .read           = seq_read,
 933         .llseek         = seq_lseek,
 934         .release        = proc_map_release,
 935 };
 936
 937 enum clear_refs_types {
 938         CLEAR_REFS_ALL = 1,
 939         CLEAR_REFS_ANON,
 940         CLEAR_REFS_MAPPED,
 941         CLEAR_REFS_SOFT_DIRTY,
 942         CLEAR_REFS_MM_HIWATER_RSS,
 943         CLEAR_REFS_LAST,
 944 };
 945
 946 struct clear_refs_private {
 947         enum clear_refs_types type;
 948 };
 949
 950 #ifdef CONFIG_MEM_SOFT_DIRTY
 951 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 952                 unsigned long addr, pte_t *pte)
 953 {
 954         /*
 955          * The soft-dirty tracker uses #PF-s to catch writes
 956          * to pages, so write-protect the pte as well. See the
 957          * Documentation/vm/soft-dirty.txt for full description
 958          * of how soft-dirty works.
 959          */
 960         pte_t ptent = *pte;
 961
 962         if (pte_present(ptent)) {
 963                 ptent = ptep_modify_prot_start(vma->vm_mm, addr, pte);
 964                 ptent = pte_wrprotect(ptent);
 965                 ptent = pte_clear_soft_dirty(ptent);
 966                 ptep_modify_prot_commit(vma->vm_mm, addr, pte, ptent);
 967         } else if (is_swap_pte(ptent)) {
 968                 ptent = pte_swp_clear_soft_dirty(ptent);
 969                 set_pte_at(vma->vm_mm, addr, pte, ptent);
 970         }
 971 }
 972 #else
 973 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 974                 unsigned long addr, pte_t *pte)
 975 {
 976 }
 977 #endif
 978
 979 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
 980 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
 981                 unsigned long addr, pmd_t *pmdp)
 982 {
 983         pmd_t pmd = *pmdp;
 984
 985         if (pmd_present(pmd)) {
 986                 /* See comment in change_huge_pmd() */
 987                 pmdp_invalidate(vma, addr, pmdp);
 988                 if (pmd_dirty(*pmdp))
 989                         pmd = pmd_mkdirty(pmd);
 990                 if (pmd_young(*pmdp))
 991                         pmd = pmd_mkyoung(pmd);
 992
 993                 pmd = pmd_wrprotect(pmd);
 994                 pmd = pmd_clear_soft_dirty(pmd);
 995
 996                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
 997         } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
 998                 pmd = pmd_swp_clear_soft_dirty(pmd);
 999                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1000         }
1001 }
1002 #else
1003 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1004                 unsigned long addr, pmd_t *pmdp)
1005 {
1006 }
1007 #endif
1008
1009 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
1010                                 unsigned long end, struct mm_walk *walk)
1011 {
1012         struct clear_refs_private *cp = walk->private;
1013         struct vm_area_struct *vma = walk->vma;
1014         pte_t *pte, ptent;
1015         spinlock_t *ptl;
1016         struct page *page;
1017
1018         ptl = pmd_trans_huge_lock(pmd, vma);
1019         if (ptl) {
1020                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1021                         clear_soft_dirty_pmd(vma, addr, pmd);
1022                         goto out;
1023                 }
1024
1025                 if (!pmd_present(*pmd))
1026                         goto out;
1027
1028                 page = pmd_page(*pmd);
1029
1030                 /* Clear accessed and referenced bits. */
1031                 pmdp_test_and_clear_young(vma, addr, pmd);
1032                 test_and_clear_page_young(page);
1033                 ClearPageReferenced(page);
1034 out:
1035                 spin_unlock(ptl);
1036                 return 0;
1037         }
1038
1039         if (pmd_trans_unstable(pmd))
1040                 return 0;
1041
1042         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1043         for (; addr != end; pte++, addr += PAGE_SIZE) {
1044                 ptent = *pte;
1045
1046                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1047                         clear_soft_dirty(vma, addr, pte);
1048                         continue;
1049                 }
1050
1051                 if (!pte_present(ptent))
1052                         continue;
1053
1054                 page = vm_normal_page(vma, addr, ptent);
1055                 if (!page)
1056                         continue;
1057
1058                 /* Clear accessed and referenced bits. */
1059                 ptep_test_and_clear_young(vma, addr, pte);
1060                 test_and_clear_page_young(page);
1061                 ClearPageReferenced(page);
1062         }
1063         pte_unmap_unlock(pte - 1, ptl);
1064         cond_resched();
1065         return 0;
1066 }
1067
1068 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1069                                 struct mm_walk *walk)
1070 {
1071         struct clear_refs_private *cp = walk->private;
1072         struct vm_area_struct *vma = walk->vma;
1073
1074         if (vma->vm_flags & VM_PFNMAP)
1075                 return 1;
1076
1077         /*
1078          * Writing 1 to /proc/pid/clear_refs affects all pages.
1079          * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1080          * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1081          * Writing 4 to /proc/pid/clear_refs affects all pages.
1082          */
1083         if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1084                 return 1;
1085         if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1086                 return 1;
1087         return 0;
1088 }
1089
1090 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1091                                 size_t count, loff_t *ppos)
1092 {
1093         struct task_struct *task;
1094         char buffer[PROC_NUMBUF];
1095         struct mm_struct *mm;
1096         struct vm_area_struct *vma;
1097         enum clear_refs_types type;
1098         struct mmu_gather tlb;
1099         int itype;
1100         int rv;
1101
1102         memset(buffer, 0, sizeof(buffer));
1103         if (count > sizeof(buffer) - 1)
1104                 count = sizeof(buffer) - 1;
1105         if (copy_from_user(buffer, buf, count))
1106                 return -EFAULT;
1107         rv = kstrtoint(strstrip(buffer), 10, &itype);
1108         if (rv < 0)
1109                 return rv;
1110         type = (enum clear_refs_types)itype;
1111         if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1112                 return -EINVAL;
1113
1114         task = get_proc_task(file_inode(file));
1115         if (!task)
1116                 return -ESRCH;
1117         mm = get_task_mm(task);
1118         if (mm) {
1119                 struct clear_refs_private cp = {
1120                         .type = type,
1121                 };
1122                 struct mm_walk clear_refs_walk = {
1123                         .pmd_entry = clear_refs_pte_range,
1124                         .test_walk = clear_refs_test_walk,
1125                         .mm = mm,
1126                         .private = &cp,
1127                 };
1128
1129                 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1130                         if (down_write_killable(&mm->mmap_sem)) {
1131                                 count = -EINTR;
1132                                 goto out_mm;
1133                         }
1134
1135                         /*
1136                          * Writing 5 to /proc/pid/clear_refs resets the peak
1137                          * resident set size to this mm's current rss value.
1138                          */
1139                         reset_mm_hiwater_rss(mm);
1140                         up_write(&mm->mmap_sem);
1141                         goto out_mm;
1142                 }
1143
1144                 down_read(&mm->mmap_sem);
1145                 tlb_gather_mmu(&tlb, mm, 0, -1);
1146                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1147                         for (vma = mm->mmap; vma; vma = vma->vm_next) {
1148                                 if (!(vma->vm_flags & VM_SOFTDIRTY))
1149                                         continue;
1150                                 up_read(&mm->mmap_sem);
1151                                 if (down_write_killable(&mm->mmap_sem)) {
1152                                         count = -EINTR;
1153                                         goto out_mm;
1154                                 }
1155                                 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1156                                         vma->vm_flags &= ~VM_SOFTDIRTY;
1157                                         vma_set_page_prot(vma);
1158                                 }
1159                                 downgrade_write(&mm->mmap_sem);
1160                                 break;
1161                         }
1162                         mmu_notifier_invalidate_range_start(mm, 0, -1);
1163                 }
1164                 walk_page_range(0, mm->highest_vm_end, &clear_refs_walk);
1165                 if (type == CLEAR_REFS_SOFT_DIRTY)
1166                         mmu_notifier_invalidate_range_end(mm, 0, -1);
1167                 tlb_finish_mmu(&tlb, 0, -1);
1168                 up_read(&mm->mmap_sem);
1169 out_mm:
1170                 mmput(mm);
1171         }
1172         put_task_struct(task);
1173
1174         return count;
1175 }
1176
1177 const struct file_operations proc_clear_refs_operations = {
1178         .write          = clear_refs_write,
1179         .llseek         = noop_llseek,
1180 };
1181
1182 typedef struct {
1183         u64 pme;
1184 } pagemap_entry_t;
1185
1186 struct pagemapread {
1187         int pos, len;           /* units: PM_ENTRY_BYTES, not bytes */
1188         pagemap_entry_t *buffer;
1189         bool show_pfn;
1190 };
1191
1192 #define PAGEMAP_WALK_SIZE       (PMD_SIZE)
1193 #define PAGEMAP_WALK_MASK       (PMD_MASK)
1194
1195 #define PM_ENTRY_BYTES          sizeof(pagemap_entry_t)
1196 #define PM_PFRAME_BITS          55
1197 #define PM_PFRAME_MASK          GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1198 #define PM_SOFT_DIRTY           BIT_ULL(55)
1199 #define PM_MMAP_EXCLUSIVE       BIT_ULL(56)
1200 #define PM_FILE                 BIT_ULL(61)
1201 #define PM_SWAP                 BIT_ULL(62)
1202 #define PM_PRESENT              BIT_ULL(63)
1203
1204 #define PM_END_OF_BUFFER    1
1205
1206 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1207 {
1208         return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1209 }
1210
1211 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1212                           struct pagemapread *pm)
1213 {
1214         pm->buffer[pm->pos++] = *pme;
1215         if (pm->pos >= pm->len)
1216                 return PM_END_OF_BUFFER;
1217         return 0;
1218 }
1219
1220 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1221                                 struct mm_walk *walk)
1222 {
1223         struct pagemapread *pm = walk->private;
1224         unsigned long addr = start;
1225         int err = 0;
1226
1227         while (addr < end) {
1228                 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1229                 pagemap_entry_t pme = make_pme(0, 0);
1230                 /* End of address space hole, which we mark as non-present. */
1231                 unsigned long hole_end;
1232
1233                 if (vma)
1234                         hole_end = min(end, vma->vm_start);
1235                 else
1236                         hole_end = end;
1237
1238                 for (; addr < hole_end; addr += PAGE_SIZE) {
1239                         err = add_to_pagemap(addr, &pme, pm);
1240                         if (err)
1241                                 goto out;
1242                 }
1243
1244                 if (!vma)
1245                         break;
1246
1247                 /* Addresses in the VMA. */
1248                 if (vma->vm_flags & VM_SOFTDIRTY)
1249                         pme = make_pme(0, PM_SOFT_DIRTY);
1250                 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1251                         err = add_to_pagemap(addr, &pme, pm);
1252                         if (err)
1253                                 goto out;
1254                 }
1255         }
1256 out:
1257         return err;
1258 }
1259
1260 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1261                 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1262 {
1263         u64 frame = 0, flags = 0;
1264         struct page *page = NULL;
1265
1266         if (pte_present(pte)) {
1267                 if (pm->show_pfn)
1268                         frame = pte_pfn(pte);
1269                 flags |= PM_PRESENT;
1270                 page = vm_normal_page(vma, addr, pte);
1271                 if (pte_soft_dirty(pte))
1272                         flags |= PM_SOFT_DIRTY;
1273         } else if (is_swap_pte(pte)) {
1274                 swp_entry_t entry;
1275                 if (pte_swp_soft_dirty(pte))
1276                         flags |= PM_SOFT_DIRTY;
1277                 entry = pte_to_swp_entry(pte);
1278                 frame = swp_type(entry) |
1279                         (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1280                 flags |= PM_SWAP;
1281                 if (is_migration_entry(entry))
1282                         page = migration_entry_to_page(entry);
1283
1284                 if (is_device_private_entry(entry))
1285                         page = device_private_entry_to_page(entry);
1286         }
1287
1288         if (page && !PageAnon(page))
1289                 flags |= PM_FILE;
1290         if (page && page_mapcount(page) == 1)
1291                 flags |= PM_MMAP_EXCLUSIVE;
1292         if (vma->vm_flags & VM_SOFTDIRTY)
1293                 flags |= PM_SOFT_DIRTY;
1294
1295         return make_pme(frame, flags);
1296 }
1297
1298 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1299                              struct mm_walk *walk)
1300 {
1301         struct vm_area_struct *vma = walk->vma;
1302         struct pagemapread *pm = walk->private;
1303         spinlock_t *ptl;
1304         pte_t *pte, *orig_pte;
1305         int err = 0;
1306
1307 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1308         ptl = pmd_trans_huge_lock(pmdp, vma);
1309         if (ptl) {
1310                 u64 flags = 0, frame = 0;
1311                 pmd_t pmd = *pmdp;
1312                 struct page *page = NULL;
1313
1314                 if ((vma->vm_flags & VM_SOFTDIRTY) || pmd_soft_dirty(pmd))
1315                         flags |= PM_SOFT_DIRTY;
1316
1317                 if (pmd_present(pmd)) {
1318                         page = pmd_page(pmd);
1319
1320                         flags |= PM_PRESENT;
1321                         if (pm->show_pfn)
1322                                 frame = pmd_pfn(pmd) +
1323                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1324                 }
1325 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1326                 else if (is_swap_pmd(pmd)) {
1327                         swp_entry_t entry = pmd_to_swp_entry(pmd);
1328
1329                         frame = swp_type(entry) |
1330                                 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1331                         flags |= PM_SWAP;
1332                         VM_BUG_ON(!is_pmd_migration_entry(pmd));
1333                         page = migration_entry_to_page(entry);
1334                 }
1335 #endif
1336
1337                 if (page && page_mapcount(page) == 1)
1338                         flags |= PM_MMAP_EXCLUSIVE;
1339
1340                 for (; addr != end; addr += PAGE_SIZE) {
1341                         pagemap_entry_t pme = make_pme(frame, flags);
1342
1343                         err = add_to_pagemap(addr, &pme, pm);
1344                         if (err)
1345                                 break;
1346                         if (pm->show_pfn && (flags & PM_PRESENT))
1347                                 frame++;
1348                 }
1349                 spin_unlock(ptl);
1350                 return err;
1351         }
1352
1353         if (pmd_trans_unstable(pmdp))
1354                 return 0;
1355 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1356
1357         /*
1358          * We can assume that @vma always points to a valid one and @end never
1359          * goes beyond vma->vm_end.
1360          */
1361         orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1362         for (; addr < end; pte++, addr += PAGE_SIZE) {
1363                 pagemap_entry_t pme;
1364
1365                 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1366                 err = add_to_pagemap(addr, &pme, pm);
1367                 if (err)
1368                         break;
1369         }
1370         pte_unmap_unlock(orig_pte, ptl);
1371
1372         cond_resched();
1373
1374         return err;
1375 }
1376
1377 #ifdef CONFIG_HUGETLB_PAGE
1378 /* This function walks within one hugetlb entry in the single call */
1379 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1380                                  unsigned long addr, unsigned long end,
1381                                  struct mm_walk *walk)
1382 {
1383         struct pagemapread *pm = walk->private;
1384         struct vm_area_struct *vma = walk->vma;
1385         u64 flags = 0, frame = 0;
1386         int err = 0;
1387         pte_t pte;
1388
1389         if (vma->vm_flags & VM_SOFTDIRTY)
1390                 flags |= PM_SOFT_DIRTY;
1391
1392         pte = huge_ptep_get(ptep);
1393         if (pte_present(pte)) {
1394                 struct page *page = pte_page(pte);
1395
1396                 if (!PageAnon(page))
1397                         flags |= PM_FILE;
1398
1399                 if (page_mapcount(page) == 1)
1400                         flags |= PM_MMAP_EXCLUSIVE;
1401
1402                 flags |= PM_PRESENT;
1403                 if (pm->show_pfn)
1404                         frame = pte_pfn(pte) +
1405                                 ((addr & ~hmask) >> PAGE_SHIFT);
1406         }
1407
1408         for (; addr != end; addr += PAGE_SIZE) {
1409                 pagemap_entry_t pme = make_pme(frame, flags);
1410
1411                 err = add_to_pagemap(addr, &pme, pm);
1412                 if (err)
1413                         return err;
1414                 if (pm->show_pfn && (flags & PM_PRESENT))
1415                         frame++;
1416         }
1417
1418         cond_resched();
1419
1420         return err;
1421 }
1422 #endif /* HUGETLB_PAGE */
1423
1424 /*
1425  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1426  *
1427  * For each page in the address space, this file contains one 64-bit entry
1428  * consisting of the following:
1429  *
1430  * Bits 0-54  page frame number (PFN) if present
1431  * Bits 0-4   swap type if swapped
1432  * Bits 5-54  swap offset if swapped
1433  * Bit  55    pte is soft-dirty (see Documentation/vm/soft-dirty.txt)
1434  * Bit  56    page exclusively mapped
1435  * Bits 57-60 zero
1436  * Bit  61    page is file-page or shared-anon
1437  * Bit  62    page swapped
1438  * Bit  63    page present
1439  *
1440  * If the page is not present but in swap, then the PFN contains an
1441  * encoding of the swap file number and the page's offset into the
1442  * swap. Unmapped pages return a null PFN. This allows determining
1443  * precisely which pages are mapped (or in swap) and comparing mapped
1444  * pages between processes.
1445  *
1446  * Efficient users of this interface will use /proc/pid/maps to
1447  * determine which areas of memory are actually mapped and llseek to
1448  * skip over unmapped regions.
1449  */
1450 static ssize_t pagemap_read(struct file *file, char __user *buf,
1451                             size_t count, loff_t *ppos)
1452 {
1453         struct mm_struct *mm = file->private_data;
1454         struct pagemapread pm;
1455         struct mm_walk pagemap_walk = {};
1456         unsigned long src;
1457         unsigned long svpfn;
1458         unsigned long start_vaddr;
1459         unsigned long end_vaddr;
1460         int ret = 0, copied = 0;
1461
1462         if (!mm || !mmget_not_zero(mm))
1463                 goto out;
1464
1465         ret = -EINVAL;
1466         /* file position must be aligned */
1467         if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1468                 goto out_mm;
1469
1470         ret = 0;
1471         if (!count)
1472                 goto out_mm;
1473
1474         /* do not disclose physical addresses: attack vector */
1475         pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1476
1477         pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1478         pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY);
1479         ret = -ENOMEM;
1480         if (!pm.buffer)
1481                 goto out_mm;
1482
1483         pagemap_walk.pmd_entry = pagemap_pmd_range;
1484         pagemap_walk.pte_hole = pagemap_pte_hole;
1485 #ifdef CONFIG_HUGETLB_PAGE
1486         pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1487 #endif
1488         pagemap_walk.mm = mm;
1489         pagemap_walk.private = &pm;
1490
1491         src = *ppos;
1492         svpfn = src / PM_ENTRY_BYTES;
1493         start_vaddr = svpfn << PAGE_SHIFT;
1494         end_vaddr = mm->task_size;
1495
1496         /* watch out for wraparound */
1497         if (svpfn > mm->task_size >> PAGE_SHIFT)
1498                 start_vaddr = end_vaddr;
1499
1500         /*
1501          * The odds are that this will stop walking way
1502          * before end_vaddr, because the length of the
1503          * user buffer is tracked in "pm", and the walk
1504          * will stop when we hit the end of the buffer.
1505          */
1506         ret = 0;
1507         while (count && (start_vaddr < end_vaddr)) {
1508                 int len;
1509                 unsigned long end;
1510
1511                 pm.pos = 0;
1512                 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1513                 /* overflow ? */
1514                 if (end < start_vaddr || end > end_vaddr)
1515                         end = end_vaddr;
1516                 down_read(&mm->mmap_sem);
1517                 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1518                 up_read(&mm->mmap_sem);
1519                 start_vaddr = end;
1520
1521                 len = min(count, PM_ENTRY_BYTES * pm.pos);
1522                 if (copy_to_user(buf, pm.buffer, len)) {
1523                         ret = -EFAULT;
1524                         goto out_free;
1525                 }
1526                 copied += len;
1527                 buf += len;
1528                 count -= len;
1529         }
1530         *ppos += copied;
1531         if (!ret || ret == PM_END_OF_BUFFER)
1532                 ret = copied;
1533
1534 out_free:
1535         kfree(pm.buffer);
1536 out_mm:
1537         mmput(mm);
1538 out:
1539         return ret;
1540 }
1541
1542 static int pagemap_open(struct inode *inode, struct file *file)
1543 {
1544         struct mm_struct *mm;
1545
1546         mm = proc_mem_open(inode, PTRACE_MODE_READ);
1547         if (IS_ERR(mm))
1548                 return PTR_ERR(mm);
1549         file->private_data = mm;
1550         return 0;
1551 }
1552
1553 static int pagemap_release(struct inode *inode, struct file *file)
1554 {
1555         struct mm_struct *mm = file->private_data;
1556
1557         if (mm)
1558                 mmdrop(mm);
1559         return 0;
1560 }
1561
1562 const struct file_operations proc_pagemap_operations = {
1563         .llseek         = mem_lseek, /* borrow this */
1564         .read           = pagemap_read,
1565         .open           = pagemap_open,
1566         .release        = pagemap_release,
1567 };
1568 #endif /* CONFIG_PROC_PAGE_MONITOR */
1569
1570 #ifdef CONFIG_NUMA
1571
1572 struct numa_maps {
1573         unsigned long pages;
1574         unsigned long anon;
1575         unsigned long active;
1576         unsigned long writeback;
1577         unsigned long mapcount_max;
1578         unsigned long dirty;
1579         unsigned long swapcache;
1580         unsigned long node[MAX_NUMNODES];
1581 };
1582
1583 struct numa_maps_private {
1584         struct proc_maps_private proc_maps;
1585         struct numa_maps md;
1586 };
1587
1588 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1589                         unsigned long nr_pages)
1590 {
1591         int count = page_mapcount(page);
1592
1593         md->pages += nr_pages;
1594         if (pte_dirty || PageDirty(page))
1595                 md->dirty += nr_pages;
1596
1597         if (PageSwapCache(page))
1598                 md->swapcache += nr_pages;
1599
1600         if (PageActive(page) || PageUnevictable(page))
1601                 md->active += nr_pages;
1602
1603         if (PageWriteback(page))
1604                 md->writeback += nr_pages;
1605
1606         if (PageAnon(page))
1607                 md->anon += nr_pages;
1608
1609         if (count > md->mapcount_max)
1610                 md->mapcount_max = count;
1611
1612         md->node[page_to_nid(page)] += nr_pages;
1613 }
1614
1615 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1616                 unsigned long addr)
1617 {
1618         struct page *page;
1619         int nid;
1620
1621         if (!pte_present(pte))
1622                 return NULL;
1623
1624         page = vm_normal_page(vma, addr, pte);
1625         if (!page)
1626                 return NULL;
1627
1628         if (PageReserved(page))
1629                 return NULL;
1630
1631         nid = page_to_nid(page);
1632         if (!node_isset(nid, node_states[N_MEMORY]))
1633                 return NULL;
1634
1635         return page;
1636 }
1637
1638 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1639 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1640                                               struct vm_area_struct *vma,
1641                                               unsigned long addr)
1642 {
1643         struct page *page;
1644         int nid;
1645
1646         if (!pmd_present(pmd))
1647                 return NULL;
1648
1649         page = vm_normal_page_pmd(vma, addr, pmd);
1650         if (!page)
1651                 return NULL;
1652
1653         if (PageReserved(page))
1654                 return NULL;
1655
1656         nid = page_to_nid(page);
1657         if (!node_isset(nid, node_states[N_MEMORY]))
1658                 return NULL;
1659
1660         return page;
1661 }
1662 #endif
1663
1664 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1665                 unsigned long end, struct mm_walk *walk)
1666 {
1667         struct numa_maps *md = walk->private;
1668         struct vm_area_struct *vma = walk->vma;
1669         spinlock_t *ptl;
1670         pte_t *orig_pte;
1671         pte_t *pte;
1672
1673 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1674         ptl = pmd_trans_huge_lock(pmd, vma);
1675         if (ptl) {
1676                 struct page *page;
1677
1678                 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1679                 if (page)
1680                         gather_stats(page, md, pmd_dirty(*pmd),
1681                                      HPAGE_PMD_SIZE/PAGE_SIZE);
1682                 spin_unlock(ptl);
1683                 return 0;
1684         }
1685
1686         if (pmd_trans_unstable(pmd))
1687                 return 0;
1688 #endif
1689         orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1690         do {
1691                 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1692                 if (!page)
1693                         continue;
1694                 gather_stats(page, md, pte_dirty(*pte), 1);
1695
1696         } while (pte++, addr += PAGE_SIZE, addr != end);
1697         pte_unmap_unlock(orig_pte, ptl);
1698         cond_resched();
1699         return 0;
1700 }
1701 #ifdef CONFIG_HUGETLB_PAGE
1702 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1703                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1704 {
1705         pte_t huge_pte = huge_ptep_get(pte);
1706         struct numa_maps *md;
1707         struct page *page;
1708
1709         if (!pte_present(huge_pte))
1710                 return 0;
1711
1712         page = pte_page(huge_pte);
1713         if (!page)
1714                 return 0;
1715
1716         md = walk->private;
1717         gather_stats(page, md, pte_dirty(huge_pte), 1);
1718         return 0;
1719 }
1720
1721 #else
1722 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1723                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1724 {
1725         return 0;
1726 }
1727 #endif
1728
1729 /*
1730  * Display pages allocated per node and memory policy via /proc.
1731  */
1732 static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1733 {
1734         struct numa_maps_private *numa_priv = m->private;
1735         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1736         struct vm_area_struct *vma = v;
1737         struct numa_maps *md = &numa_priv->md;
1738         struct file *file = vma->vm_file;
1739         struct mm_struct *mm = vma->vm_mm;
1740         struct mm_walk walk = {
1741                 .hugetlb_entry = gather_hugetlb_stats,
1742                 .pmd_entry = gather_pte_stats,
1743                 .private = md,
1744                 .mm = mm,
1745         };
1746         struct mempolicy *pol;
1747         char buffer[64];
1748         int nid;
1749
1750         if (!mm)
1751                 return 0;
1752
1753         /* Ensure we start with an empty set of numa_maps statistics. */
1754         memset(md, 0, sizeof(*md));
1755
1756         pol = __get_vma_policy(vma, vma->vm_start);
1757         if (pol) {
1758                 mpol_to_str(buffer, sizeof(buffer), pol);
1759                 mpol_cond_put(pol);
1760         } else {
1761                 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1762         }
1763
1764         seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1765
1766         if (file) {
1767                 seq_puts(m, " file=");
1768                 seq_file_path(m, file, "\n\t= ");
1769         } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1770                 seq_puts(m, " heap");
1771         } else if (is_stack(proc_priv, vma)) {
1772                 seq_puts(m, " stack");
1773         }
1774
1775         if (is_vm_hugetlb_page(vma))
1776                 seq_puts(m, " huge");
1777
1778         /* mmap_sem is held by m_start */
1779         walk_page_vma(vma, &walk);
1780
1781         if (!md->pages)
1782                 goto out;
1783
1784         if (md->anon)
1785                 seq_printf(m, " anon=%lu", md->anon);
1786
1787         if (md->dirty)
1788                 seq_printf(m, " dirty=%lu", md->dirty);
1789
1790         if (md->pages != md->anon && md->pages != md->dirty)
1791                 seq_printf(m, " mapped=%lu", md->pages);
1792
1793         if (md->mapcount_max > 1)
1794                 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1795
1796         if (md->swapcache)
1797                 seq_printf(m, " swapcache=%lu", md->swapcache);
1798
1799         if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1800                 seq_printf(m, " active=%lu", md->active);
1801
1802         if (md->writeback)
1803                 seq_printf(m, " writeback=%lu", md->writeback);
1804
1805         for_each_node_state(nid, N_MEMORY)
1806                 if (md->node[nid])
1807                         seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1808
1809         seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1810 out:
1811         seq_putc(m, '\n');
1812         m_cache_vma(m, vma);
1813         return 0;
1814 }
1815
1816 static int show_pid_numa_map(struct seq_file *m, void *v)
1817 {
1818         return show_numa_map(m, v, 1);
1819 }
1820
1821 static int show_tid_numa_map(struct seq_file *m, void *v)
1822 {
1823         return show_numa_map(m, v, 0);
1824 }
1825
1826 static const struct seq_operations proc_pid_numa_maps_op = {
1827         .start  = m_start,
1828         .next   = m_next,
1829         .stop   = m_stop,
1830         .show   = show_pid_numa_map,
1831 };
1832
1833 static const struct seq_operations proc_tid_numa_maps_op = {
1834         .start  = m_start,
1835         .next   = m_next,
1836         .stop   = m_stop,
1837         .show   = show_tid_numa_map,
1838 };
1839
1840 static int numa_maps_open(struct inode *inode, struct file *file,
1841                           const struct seq_operations *ops)
1842 {
1843         return proc_maps_open(inode, file, ops,
1844                                 sizeof(struct numa_maps_private));
1845 }
1846
1847 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1848 {
1849         return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1850 }
1851
1852 static int tid_numa_maps_open(struct inode *inode, struct file *file)
1853 {
1854         return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1855 }
1856
1857 const struct file_operations proc_pid_numa_maps_operations = {
1858         .open           = pid_numa_maps_open,
1859         .read           = seq_read,
1860         .llseek         = seq_lseek,
1861         .release        = proc_map_release,
1862 };
1863
1864 const struct file_operations proc_tid_numa_maps_operations = {
1865         .open           = tid_numa_maps_open,
1866         .read           = seq_read,
1867         .llseek         = seq_lseek,
1868         .release        = proc_map_release,
1869 };
1870 #endif /* CONFIG_NUMA */