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 vm_area_struct *vma)
 272 {
 273         /*
 274          * We make no effort to guess what a given thread considers to be
 275          * its "stack".  It's not even well-defined for programs written
 276          * languages like Go.
 277          */
 278         return vma->vm_start <= vma->vm_mm->start_stack &&
 279                 vma->vm_end >= vma->vm_mm->start_stack;
 280 }
 281
 282 static void show_vma_header_prefix(struct seq_file *m,
 283                                    unsigned long start, unsigned long end,
 284                                    vm_flags_t flags, unsigned long long pgoff,
 285                                    dev_t dev, unsigned long ino)
 286 {
 287         seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
 288         seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu ",
 289                    start,
 290                    end,
 291                    flags & VM_READ ? 'r' : '-',
 292                    flags & VM_WRITE ? 'w' : '-',
 293                    flags & VM_EXEC ? 'x' : '-',
 294                    flags & VM_MAYSHARE ? 's' : 'p',
 295                    pgoff,
 296                    MAJOR(dev), MINOR(dev), ino);
 297 }
 298
 299 static void
 300 show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
 301 {
 302         struct mm_struct *mm = vma->vm_mm;
 303         struct file *file = vma->vm_file;
 304         vm_flags_t flags = vma->vm_flags;
 305         unsigned long ino = 0;
 306         unsigned long long pgoff = 0;
 307         unsigned long start, end;
 308         dev_t dev = 0;
 309         const char *name = NULL;
 310
 311         if (file) {
 312                 struct inode *inode = file_inode(vma->vm_file);
 313                 dev = inode->i_sb->s_dev;
 314                 ino = inode->i_ino;
 315                 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
 316         }
 317
 318         start = vma->vm_start;
 319         end = vma->vm_end;
 320         show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino);
 321
 322         /*
 323          * Print the dentry name for named mappings, and a
 324          * special [heap] marker for the heap:
 325          */
 326         if (file) {
 327                 seq_pad(m, ' ');
 328                 seq_file_path(m, file, "\n");
 329                 goto done;
 330         }
 331
 332         if (vma->vm_ops && vma->vm_ops->name) {
 333                 name = vma->vm_ops->name(vma);
 334                 if (name)
 335                         goto done;
 336         }
 337
 338         name = arch_vma_name(vma);
 339         if (!name) {
 340                 if (!mm) {
 341                         name = "[vdso]";
 342                         goto done;
 343                 }
 344
 345                 if (vma->vm_start <= mm->brk &&
 346                     vma->vm_end >= mm->start_brk) {
 347                         name = "[heap]";
 348                         goto done;
 349                 }
 350
 351                 if (is_stack(vma))
 352                         name = "[stack]";
 353         }
 354
 355 done:
 356         if (name) {
 357                 seq_pad(m, ' ');
 358                 seq_puts(m, name);
 359         }
 360         seq_putc(m, '\n');
 361 }
 362
 363 static int show_map(struct seq_file *m, void *v, int is_pid)
 364 {
 365         show_map_vma(m, v, is_pid);
 366         m_cache_vma(m, v);
 367         return 0;
 368 }
 369
 370 static int show_pid_map(struct seq_file *m, void *v)
 371 {
 372         return show_map(m, v, 1);
 373 }
 374
 375 static int show_tid_map(struct seq_file *m, void *v)
 376 {
 377         return show_map(m, v, 0);
 378 }
 379
 380 static const struct seq_operations proc_pid_maps_op = {
 381         .start  = m_start,
 382         .next   = m_next,
 383         .stop   = m_stop,
 384         .show   = show_pid_map
 385 };
 386
 387 static const struct seq_operations proc_tid_maps_op = {
 388         .start  = m_start,
 389         .next   = m_next,
 390         .stop   = m_stop,
 391         .show   = show_tid_map
 392 };
 393
 394 static int pid_maps_open(struct inode *inode, struct file *file)
 395 {
 396         return do_maps_open(inode, file, &proc_pid_maps_op);
 397 }
 398
 399 static int tid_maps_open(struct inode *inode, struct file *file)
 400 {
 401         return do_maps_open(inode, file, &proc_tid_maps_op);
 402 }
 403
 404 const struct file_operations proc_pid_maps_operations = {
 405         .open           = pid_maps_open,
 406         .read           = seq_read,
 407         .llseek         = seq_lseek,
 408         .release        = proc_map_release,
 409 };
 410
 411 const struct file_operations proc_tid_maps_operations = {
 412         .open           = tid_maps_open,
 413         .read           = seq_read,
 414         .llseek         = seq_lseek,
 415         .release        = proc_map_release,
 416 };
 417
 418 /*
 419  * Proportional Set Size(PSS): my share of RSS.
 420  *
 421  * PSS of a process is the count of pages it has in memory, where each
 422  * page is divided by the number of processes sharing it.  So if a
 423  * process has 1000 pages all to itself, and 1000 shared with one other
 424  * process, its PSS will be 1500.
 425  *
 426  * To keep (accumulated) division errors low, we adopt a 64bit
 427  * fixed-point pss counter to minimize division errors. So (pss >>
 428  * PSS_SHIFT) would be the real byte count.
 429  *
 430  * A shift of 12 before division means (assuming 4K page size):
 431  *      - 1M 3-user-pages add up to 8KB errors;
 432  *      - supports mapcount up to 2^24, or 16M;
 433  *      - supports PSS up to 2^52 bytes, or 4PB.
 434  */
 435 #define PSS_SHIFT 12
 436
 437 #ifdef CONFIG_PROC_PAGE_MONITOR
 438 struct mem_size_stats {
 439         bool first;
 440         unsigned long resident;
 441         unsigned long shared_clean;
 442         unsigned long shared_dirty;
 443         unsigned long private_clean;
 444         unsigned long private_dirty;
 445         unsigned long referenced;
 446         unsigned long anonymous;
 447         unsigned long lazyfree;
 448         unsigned long anonymous_thp;
 449         unsigned long shmem_thp;
 450         unsigned long swap;
 451         unsigned long shared_hugetlb;
 452         unsigned long private_hugetlb;
 453         unsigned long first_vma_start;
 454         u64 pss;
 455         u64 pss_locked;
 456         u64 swap_pss;
 457         bool check_shmem_swap;
 458 };
 459
 460 static void smaps_account(struct mem_size_stats *mss, struct page *page,
 461                 bool compound, bool young, bool dirty)
 462 {
 463         int i, nr = compound ? 1 << compound_order(page) : 1;
 464         unsigned long size = nr * PAGE_SIZE;
 465
 466         if (PageAnon(page)) {
 467                 mss->anonymous += size;
 468                 if (!PageSwapBacked(page) && !dirty && !PageDirty(page))
 469                         mss->lazyfree += size;
 470         }
 471
 472         mss->resident += size;
 473         /* Accumulate the size in pages that have been accessed. */
 474         if (young || page_is_young(page) || PageReferenced(page))
 475                 mss->referenced += size;
 476
 477         /*
 478          * page_count(page) == 1 guarantees the page is mapped exactly once.
 479          * If any subpage of the compound page mapped with PTE it would elevate
 480          * page_count().
 481          */
 482         if (page_count(page) == 1) {
 483                 if (dirty || PageDirty(page))
 484                         mss->private_dirty += size;
 485                 else
 486                         mss->private_clean += size;
 487                 mss->pss += (u64)size << PSS_SHIFT;
 488                 return;
 489         }
 490
 491         for (i = 0; i < nr; i++, page++) {
 492                 int mapcount = page_mapcount(page);
 493
 494                 if (mapcount >= 2) {
 495                         if (dirty || PageDirty(page))
 496                                 mss->shared_dirty += PAGE_SIZE;
 497                         else
 498                                 mss->shared_clean += PAGE_SIZE;
 499                         mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount;
 500                 } else {
 501                         if (dirty || PageDirty(page))
 502                                 mss->private_dirty += PAGE_SIZE;
 503                         else
 504                                 mss->private_clean += PAGE_SIZE;
 505                         mss->pss += PAGE_SIZE << PSS_SHIFT;
 506                 }
 507         }
 508 }
 509
 510 #ifdef CONFIG_SHMEM
 511 static int smaps_pte_hole(unsigned long addr, unsigned long end,
 512                 struct mm_walk *walk)
 513 {
 514         struct mem_size_stats *mss = walk->private;
 515
 516         mss->swap += shmem_partial_swap_usage(
 517                         walk->vma->vm_file->f_mapping, addr, end);
 518
 519         return 0;
 520 }
 521 #endif
 522
 523 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
 524                 struct mm_walk *walk)
 525 {
 526         struct mem_size_stats *mss = walk->private;
 527         struct vm_area_struct *vma = walk->vma;
 528         struct page *page = NULL;
 529
 530         if (pte_present(*pte)) {
 531                 page = vm_normal_page(vma, addr, *pte);
 532         } else if (is_swap_pte(*pte)) {
 533                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 534
 535                 if (!non_swap_entry(swpent)) {
 536                         int mapcount;
 537
 538                         mss->swap += PAGE_SIZE;
 539                         mapcount = swp_swapcount(swpent);
 540                         if (mapcount >= 2) {
 541                                 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
 542
 543                                 do_div(pss_delta, mapcount);
 544                                 mss->swap_pss += pss_delta;
 545                         } else {
 546                                 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
 547                         }
 548                 } else if (is_migration_entry(swpent))
 549                         page = migration_entry_to_page(swpent);
 550                 else if (is_device_private_entry(swpent))
 551                         page = device_private_entry_to_page(swpent);
 552         } else if (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap
 553                                                         && pte_none(*pte))) {
 554                 page = find_get_entry(vma->vm_file->f_mapping,
 555                                                 linear_page_index(vma, addr));
 556                 if (!page)
 557                         return;
 558
 559                 if (radix_tree_exceptional_entry(page))
 560                         mss->swap += PAGE_SIZE;
 561                 else
 562                         put_page(page);
 563
 564                 return;
 565         }
 566
 567         if (!page)
 568                 return;
 569
 570         smaps_account(mss, page, false, pte_young(*pte), pte_dirty(*pte));
 571 }
 572
 573 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 574 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 575                 struct mm_walk *walk)
 576 {
 577         struct mem_size_stats *mss = walk->private;
 578         struct vm_area_struct *vma = walk->vma;
 579         struct page *page;
 580
 581         /* FOLL_DUMP will return -EFAULT on huge zero page */
 582         page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
 583         if (IS_ERR_OR_NULL(page))
 584                 return;
 585         if (PageAnon(page))
 586                 mss->anonymous_thp += HPAGE_PMD_SIZE;
 587         else if (PageSwapBacked(page))
 588                 mss->shmem_thp += HPAGE_PMD_SIZE;
 589         else if (is_zone_device_page(page))
 590                 /* pass */;
 591         else
 592                 VM_BUG_ON_PAGE(1, page);
 593         smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd));
 594 }
 595 #else
 596 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 597                 struct mm_walk *walk)
 598 {
 599 }
 600 #endif
 601
 602 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 603                            struct mm_walk *walk)
 604 {
 605         struct vm_area_struct *vma = walk->vma;
 606         pte_t *pte;
 607         spinlock_t *ptl;
 608
 609         ptl = pmd_trans_huge_lock(pmd, vma);
 610         if (ptl) {
 611                 if (pmd_present(*pmd))
 612                         smaps_pmd_entry(pmd, addr, walk);
 613                 spin_unlock(ptl);
 614                 goto out;
 615         }
 616
 617         if (pmd_trans_unstable(pmd))
 618                 goto out;
 619         /*
 620          * The mmap_sem held all the way back in m_start() is what
 621          * keeps khugepaged out of here and from collapsing things
 622          * in here.
 623          */
 624         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 625         for (; addr != end; pte++, addr += PAGE_SIZE)
 626                 smaps_pte_entry(pte, addr, walk);
 627         pte_unmap_unlock(pte - 1, ptl);
 628 out:
 629         cond_resched();
 630         return 0;
 631 }
 632
 633 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
 634 {
 635         /*
 636          * Don't forget to update Documentation/ on changes.
 637          */
 638         static const char mnemonics[BITS_PER_LONG][2] = {
 639                 /*
 640                  * In case if we meet a flag we don't know about.
 641                  */
 642                 [0 ... (BITS_PER_LONG-1)] = "??",
 643
 644                 [ilog2(VM_READ)]        = "rd",
 645                 [ilog2(VM_WRITE)]       = "wr",
 646                 [ilog2(VM_EXEC)]        = "ex",
 647                 [ilog2(VM_SHARED)]      = "sh",
 648                 [ilog2(VM_MAYREAD)]     = "mr",
 649                 [ilog2(VM_MAYWRITE)]    = "mw",
 650                 [ilog2(VM_MAYEXEC)]     = "me",
 651                 [ilog2(VM_MAYSHARE)]    = "ms",
 652                 [ilog2(VM_GROWSDOWN)]   = "gd",
 653                 [ilog2(VM_PFNMAP)]      = "pf",
 654                 [ilog2(VM_DENYWRITE)]   = "dw",
 655 #ifdef CONFIG_X86_INTEL_MPX
 656                 [ilog2(VM_MPX)]         = "mp",
 657 #endif
 658                 [ilog2(VM_LOCKED)]      = "lo",
 659                 [ilog2(VM_IO)]          = "io",
 660                 [ilog2(VM_SEQ_READ)]    = "sr",
 661                 [ilog2(VM_RAND_READ)]   = "rr",
 662                 [ilog2(VM_DONTCOPY)]    = "dc",
 663                 [ilog2(VM_DONTEXPAND)]  = "de",
 664                 [ilog2(VM_ACCOUNT)]     = "ac",
 665                 [ilog2(VM_NORESERVE)]   = "nr",
 666                 [ilog2(VM_HUGETLB)]     = "ht",
 667                 [ilog2(VM_ARCH_1)]      = "ar",
 668                 [ilog2(VM_WIPEONFORK)]  = "wf",
 669                 [ilog2(VM_DONTDUMP)]    = "dd",
 670 #ifdef CONFIG_MEM_SOFT_DIRTY
 671                 [ilog2(VM_SOFTDIRTY)]   = "sd",
 672 #endif
 673                 [ilog2(VM_MIXEDMAP)]    = "mm",
 674                 [ilog2(VM_HUGEPAGE)]    = "hg",
 675                 [ilog2(VM_NOHUGEPAGE)]  = "nh",
 676                 [ilog2(VM_MERGEABLE)]   = "mg",
 677                 [ilog2(VM_UFFD_MISSING)]= "um",
 678                 [ilog2(VM_UFFD_WP)]     = "uw",
 679 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
 680                 /* These come out via ProtectionKey: */
 681                 [ilog2(VM_PKEY_BIT0)]   = "",
 682                 [ilog2(VM_PKEY_BIT1)]   = "",
 683                 [ilog2(VM_PKEY_BIT2)]   = "",
 684                 [ilog2(VM_PKEY_BIT3)]   = "",
 685 #endif
 686         };
 687         size_t i;
 688
 689         seq_puts(m, "VmFlags: ");
 690         for (i = 0; i < BITS_PER_LONG; i++) {
 691                 if (!mnemonics[i][0])
 692                         continue;
 693                 if (vma->vm_flags & (1UL << i)) {
 694                         seq_printf(m, "%c%c ",
 695                                    mnemonics[i][0], mnemonics[i][1]);
 696                 }
 697         }
 698         seq_putc(m, '\n');
 699 }
 700
 701 #ifdef CONFIG_HUGETLB_PAGE
 702 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
 703                                  unsigned long addr, unsigned long end,
 704                                  struct mm_walk *walk)
 705 {
 706         struct mem_size_stats *mss = walk->private;
 707         struct vm_area_struct *vma = walk->vma;
 708         struct page *page = NULL;
 709
 710         if (pte_present(*pte)) {
 711                 page = vm_normal_page(vma, addr, *pte);
 712         } else if (is_swap_pte(*pte)) {
 713                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 714
 715                 if (is_migration_entry(swpent))
 716                         page = migration_entry_to_page(swpent);
 717                 else if (is_device_private_entry(swpent))
 718                         page = device_private_entry_to_page(swpent);
 719         }
 720         if (page) {
 721                 int mapcount = page_mapcount(page);
 722
 723                 if (mapcount >= 2)
 724                         mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
 725                 else
 726                         mss->private_hugetlb += huge_page_size(hstate_vma(vma));
 727         }
 728         return 0;
 729 }
 730 #endif /* HUGETLB_PAGE */
 731
 732 void __weak arch_show_smap(struct seq_file *m, struct vm_area_struct *vma)
 733 {
 734 }
 735
 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_printf(m,
 811                            "Size:           %8lu kB\n"
 812                            "KernelPageSize: %8lu kB\n"
 813                            "MMUPageSize:    %8lu kB\n",
 814                            (vma->vm_end - vma->vm_start) >> 10,
 815                            vma_kernel_pagesize(vma) >> 10,
 816                            vma_mmu_pagesize(vma) >> 10);
 817
 818
 819         if (!rollup_mode || last_vma)
 820                 seq_printf(m,
 821                            "Rss:            %8lu kB\n"
 822                            "Pss:            %8lu kB\n"
 823                            "Shared_Clean:   %8lu kB\n"
 824                            "Shared_Dirty:   %8lu kB\n"
 825                            "Private_Clean:  %8lu kB\n"
 826                            "Private_Dirty:  %8lu kB\n"
 827                            "Referenced:     %8lu kB\n"
 828                            "Anonymous:      %8lu kB\n"
 829                            "LazyFree:       %8lu kB\n"
 830                            "AnonHugePages:  %8lu kB\n"
 831                            "ShmemPmdMapped: %8lu kB\n"
 832                            "Shared_Hugetlb: %8lu kB\n"
 833                            "Private_Hugetlb: %7lu kB\n"
 834                            "Swap:           %8lu kB\n"
 835                            "SwapPss:        %8lu kB\n"
 836                            "Locked:         %8lu kB\n",
 837                            mss->resident >> 10,
 838                            (unsigned long)(mss->pss >> (10 + PSS_SHIFT)),
 839                            mss->shared_clean  >> 10,
 840                            mss->shared_dirty  >> 10,
 841                            mss->private_clean >> 10,
 842                            mss->private_dirty >> 10,
 843                            mss->referenced >> 10,
 844                            mss->anonymous >> 10,
 845                            mss->lazyfree >> 10,
 846                            mss->anonymous_thp >> 10,
 847                            mss->shmem_thp >> 10,
 848                            mss->shared_hugetlb >> 10,
 849                            mss->private_hugetlb >> 10,
 850                            mss->swap >> 10,
 851                            (unsigned long)(mss->swap_pss >> (10 + PSS_SHIFT)),
 852                            (unsigned long)(mss->pss >> (10 + PSS_SHIFT)));
 853
 854         if (!rollup_mode) {
 855                 arch_show_smap(m, vma);
 856                 show_smap_vma_flags(m, vma);
 857         }
 858         m_cache_vma(m, vma);
 859         return ret;
 860 }
 861
 862 static int show_pid_smap(struct seq_file *m, void *v)
 863 {
 864         return show_smap(m, v, 1);
 865 }
 866
 867 static int show_tid_smap(struct seq_file *m, void *v)
 868 {
 869         return show_smap(m, v, 0);
 870 }
 871
 872 static const struct seq_operations proc_pid_smaps_op = {
 873         .start  = m_start,
 874         .next   = m_next,
 875         .stop   = m_stop,
 876         .show   = show_pid_smap
 877 };
 878
 879 static const struct seq_operations proc_tid_smaps_op = {
 880         .start  = m_start,
 881         .next   = m_next,
 882         .stop   = m_stop,
 883         .show   = show_tid_smap
 884 };
 885
 886 static int pid_smaps_open(struct inode *inode, struct file *file)
 887 {
 888         return do_maps_open(inode, file, &proc_pid_smaps_op);
 889 }
 890
 891 static int pid_smaps_rollup_open(struct inode *inode, struct file *file)
 892 {
 893         struct seq_file *seq;
 894         struct proc_maps_private *priv;
 895         int ret = do_maps_open(inode, file, &proc_pid_smaps_op);
 896
 897         if (ret < 0)
 898                 return ret;
 899         seq = file->private_data;
 900         priv = seq->private;
 901         priv->rollup = kzalloc(sizeof(*priv->rollup), GFP_KERNEL);
 902         if (!priv->rollup) {
 903                 proc_map_release(inode, file);
 904                 return -ENOMEM;
 905         }
 906         priv->rollup->first = true;
 907         return 0;
 908 }
 909
 910 static int tid_smaps_open(struct inode *inode, struct file *file)
 911 {
 912         return do_maps_open(inode, file, &proc_tid_smaps_op);
 913 }
 914
 915 const struct file_operations proc_pid_smaps_operations = {
 916         .open           = pid_smaps_open,
 917         .read           = seq_read,
 918         .llseek         = seq_lseek,
 919         .release        = proc_map_release,
 920 };
 921
 922 const struct file_operations proc_pid_smaps_rollup_operations = {
 923         .open           = pid_smaps_rollup_open,
 924         .read           = seq_read,
 925         .llseek         = seq_lseek,
 926         .release        = proc_map_release,
 927 };
 928
 929 const struct file_operations proc_tid_smaps_operations = {
 930         .open           = tid_smaps_open,
 931         .read           = seq_read,
 932         .llseek         = seq_lseek,
 933         .release        = proc_map_release,
 934 };
 935
 936 enum clear_refs_types {
 937         CLEAR_REFS_ALL = 1,
 938         CLEAR_REFS_ANON,
 939         CLEAR_REFS_MAPPED,
 940         CLEAR_REFS_SOFT_DIRTY,
 941         CLEAR_REFS_MM_HIWATER_RSS,
 942         CLEAR_REFS_LAST,
 943 };
 944
 945 struct clear_refs_private {
 946         enum clear_refs_types type;
 947 };
 948
 949 #ifdef CONFIG_MEM_SOFT_DIRTY
 950 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 951                 unsigned long addr, pte_t *pte)
 952 {
 953         /*
 954          * The soft-dirty tracker uses #PF-s to catch writes
 955          * to pages, so write-protect the pte as well. See the
 956          * Documentation/vm/soft-dirty.txt for full description
 957          * of how soft-dirty works.
 958          */
 959         pte_t ptent = *pte;
 960
 961         if (pte_present(ptent)) {
 962                 ptent = ptep_modify_prot_start(vma->vm_mm, addr, pte);
 963                 ptent = pte_wrprotect(ptent);
 964                 ptent = pte_clear_soft_dirty(ptent);
 965                 ptep_modify_prot_commit(vma->vm_mm, addr, pte, ptent);
 966         } else if (is_swap_pte(ptent)) {
 967                 ptent = pte_swp_clear_soft_dirty(ptent);
 968                 set_pte_at(vma->vm_mm, addr, pte, ptent);
 969         }
 970 }
 971 #else
 972 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 973                 unsigned long addr, pte_t *pte)
 974 {
 975 }
 976 #endif
 977
 978 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
 979 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
 980                 unsigned long addr, pmd_t *pmdp)
 981 {
 982         pmd_t pmd = *pmdp;
 983
 984         if (pmd_present(pmd)) {
 985                 /* See comment in change_huge_pmd() */
 986                 pmdp_invalidate(vma, addr, pmdp);
 987                 if (pmd_dirty(*pmdp))
 988                         pmd = pmd_mkdirty(pmd);
 989                 if (pmd_young(*pmdp))
 990                         pmd = pmd_mkyoung(pmd);
 991
 992                 pmd = pmd_wrprotect(pmd);
 993                 pmd = pmd_clear_soft_dirty(pmd);
 994
 995                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
 996         } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
 997                 pmd = pmd_swp_clear_soft_dirty(pmd);
 998                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
 999         }
1000 }
1001 #else
1002 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1003                 unsigned long addr, pmd_t *pmdp)
1004 {
1005 }
1006 #endif
1007
1008 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
1009                                 unsigned long end, struct mm_walk *walk)
1010 {
1011         struct clear_refs_private *cp = walk->private;
1012         struct vm_area_struct *vma = walk->vma;
1013         pte_t *pte, ptent;
1014         spinlock_t *ptl;
1015         struct page *page;
1016
1017         ptl = pmd_trans_huge_lock(pmd, vma);
1018         if (ptl) {
1019                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1020                         clear_soft_dirty_pmd(vma, addr, pmd);
1021                         goto out;
1022                 }
1023
1024                 if (!pmd_present(*pmd))
1025                         goto out;
1026
1027                 page = pmd_page(*pmd);
1028
1029                 /* Clear accessed and referenced bits. */
1030                 pmdp_test_and_clear_young(vma, addr, pmd);
1031                 test_and_clear_page_young(page);
1032                 ClearPageReferenced(page);
1033 out:
1034                 spin_unlock(ptl);
1035                 return 0;
1036         }
1037
1038         if (pmd_trans_unstable(pmd))
1039                 return 0;
1040
1041         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1042         for (; addr != end; pte++, addr += PAGE_SIZE) {
1043                 ptent = *pte;
1044
1045                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1046                         clear_soft_dirty(vma, addr, pte);
1047                         continue;
1048                 }
1049
1050                 if (!pte_present(ptent))
1051                         continue;
1052
1053                 page = vm_normal_page(vma, addr, ptent);
1054                 if (!page)
1055                         continue;
1056
1057                 /* Clear accessed and referenced bits. */
1058                 ptep_test_and_clear_young(vma, addr, pte);
1059                 test_and_clear_page_young(page);
1060                 ClearPageReferenced(page);
1061         }
1062         pte_unmap_unlock(pte - 1, ptl);
1063         cond_resched();
1064         return 0;
1065 }
1066
1067 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1068                                 struct mm_walk *walk)
1069 {
1070         struct clear_refs_private *cp = walk->private;
1071         struct vm_area_struct *vma = walk->vma;
1072
1073         if (vma->vm_flags & VM_PFNMAP)
1074                 return 1;
1075
1076         /*
1077          * Writing 1 to /proc/pid/clear_refs affects all pages.
1078          * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1079          * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1080          * Writing 4 to /proc/pid/clear_refs affects all pages.
1081          */
1082         if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1083                 return 1;
1084         if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1085                 return 1;
1086         return 0;
1087 }
1088
1089 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1090                                 size_t count, loff_t *ppos)
1091 {
1092         struct task_struct *task;
1093         char buffer[PROC_NUMBUF];
1094         struct mm_struct *mm;
1095         struct vm_area_struct *vma;
1096         enum clear_refs_types type;
1097         struct mmu_gather tlb;
1098         int itype;
1099         int rv;
1100
1101         memset(buffer, 0, sizeof(buffer));
1102         if (count > sizeof(buffer) - 1)
1103                 count = sizeof(buffer) - 1;
1104         if (copy_from_user(buffer, buf, count))
1105                 return -EFAULT;
1106         rv = kstrtoint(strstrip(buffer), 10, &itype);
1107         if (rv < 0)
1108                 return rv;
1109         type = (enum clear_refs_types)itype;
1110         if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1111                 return -EINVAL;
1112
1113         task = get_proc_task(file_inode(file));
1114         if (!task)
1115                 return -ESRCH;
1116         mm = get_task_mm(task);
1117         if (mm) {
1118                 struct clear_refs_private cp = {
1119                         .type = type,
1120                 };
1121                 struct mm_walk clear_refs_walk = {
1122                         .pmd_entry = clear_refs_pte_range,
1123                         .test_walk = clear_refs_test_walk,
1124                         .mm = mm,
1125                         .private = &cp,
1126                 };
1127
1128                 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1129                         if (down_write_killable(&mm->mmap_sem)) {
1130                                 count = -EINTR;
1131                                 goto out_mm;
1132                         }
1133
1134                         /*
1135                          * Writing 5 to /proc/pid/clear_refs resets the peak
1136                          * resident set size to this mm's current rss value.
1137                          */
1138                         reset_mm_hiwater_rss(mm);
1139                         up_write(&mm->mmap_sem);
1140                         goto out_mm;
1141                 }
1142
1143                 down_read(&mm->mmap_sem);
1144                 tlb_gather_mmu(&tlb, mm, 0, -1);
1145                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1146                         for (vma = mm->mmap; vma; vma = vma->vm_next) {
1147                                 if (!(vma->vm_flags & VM_SOFTDIRTY))
1148                                         continue;
1149                                 up_read(&mm->mmap_sem);
1150                                 if (down_write_killable(&mm->mmap_sem)) {
1151                                         count = -EINTR;
1152                                         goto out_mm;
1153                                 }
1154                                 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1155                                         vma->vm_flags &= ~VM_SOFTDIRTY;
1156                                         vma_set_page_prot(vma);
1157                                 }
1158                                 downgrade_write(&mm->mmap_sem);
1159                                 break;
1160                         }
1161                         mmu_notifier_invalidate_range_start(mm, 0, -1);
1162                 }
1163                 walk_page_range(0, mm->highest_vm_end, &clear_refs_walk);
1164                 if (type == CLEAR_REFS_SOFT_DIRTY)
1165                         mmu_notifier_invalidate_range_end(mm, 0, -1);
1166                 tlb_finish_mmu(&tlb, 0, -1);
1167                 up_read(&mm->mmap_sem);
1168 out_mm:
1169                 mmput(mm);
1170         }
1171         put_task_struct(task);
1172
1173         return count;
1174 }
1175
1176 const struct file_operations proc_clear_refs_operations = {
1177         .write          = clear_refs_write,
1178         .llseek         = noop_llseek,
1179 };
1180
1181 typedef struct {
1182         u64 pme;
1183 } pagemap_entry_t;
1184
1185 struct pagemapread {
1186         int pos, len;           /* units: PM_ENTRY_BYTES, not bytes */
1187         pagemap_entry_t *buffer;
1188         bool show_pfn;
1189 };
1190
1191 #define PAGEMAP_WALK_SIZE       (PMD_SIZE)
1192 #define PAGEMAP_WALK_MASK       (PMD_MASK)
1193
1194 #define PM_ENTRY_BYTES          sizeof(pagemap_entry_t)
1195 #define PM_PFRAME_BITS          55
1196 #define PM_PFRAME_MASK          GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1197 #define PM_SOFT_DIRTY           BIT_ULL(55)
1198 #define PM_MMAP_EXCLUSIVE       BIT_ULL(56)
1199 #define PM_FILE                 BIT_ULL(61)
1200 #define PM_SWAP                 BIT_ULL(62)
1201 #define PM_PRESENT              BIT_ULL(63)
1202
1203 #define PM_END_OF_BUFFER    1
1204
1205 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1206 {
1207         return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1208 }
1209
1210 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1211                           struct pagemapread *pm)
1212 {
1213         pm->buffer[pm->pos++] = *pme;
1214         if (pm->pos >= pm->len)
1215                 return PM_END_OF_BUFFER;
1216         return 0;
1217 }
1218
1219 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1220                                 struct mm_walk *walk)
1221 {
1222         struct pagemapread *pm = walk->private;
1223         unsigned long addr = start;
1224         int err = 0;
1225
1226         while (addr < end) {
1227                 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1228                 pagemap_entry_t pme = make_pme(0, 0);
1229                 /* End of address space hole, which we mark as non-present. */
1230                 unsigned long hole_end;
1231
1232                 if (vma)
1233                         hole_end = min(end, vma->vm_start);
1234                 else
1235                         hole_end = end;
1236
1237                 for (; addr < hole_end; addr += PAGE_SIZE) {
1238                         err = add_to_pagemap(addr, &pme, pm);
1239                         if (err)
1240                                 goto out;
1241                 }
1242
1243                 if (!vma)
1244                         break;
1245
1246                 /* Addresses in the VMA. */
1247                 if (vma->vm_flags & VM_SOFTDIRTY)
1248                         pme = make_pme(0, PM_SOFT_DIRTY);
1249                 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1250                         err = add_to_pagemap(addr, &pme, pm);
1251                         if (err)
1252                                 goto out;
1253                 }
1254         }
1255 out:
1256         return err;
1257 }
1258
1259 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1260                 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1261 {
1262         u64 frame = 0, flags = 0;
1263         struct page *page = NULL;
1264
1265         if (pte_present(pte)) {
1266                 if (pm->show_pfn)
1267                         frame = pte_pfn(pte);
1268                 flags |= PM_PRESENT;
1269                 page = _vm_normal_page(vma, addr, pte, true);
1270                 if (pte_soft_dirty(pte))
1271                         flags |= PM_SOFT_DIRTY;
1272         } else if (is_swap_pte(pte)) {
1273                 swp_entry_t entry;
1274                 if (pte_swp_soft_dirty(pte))
1275                         flags |= PM_SOFT_DIRTY;
1276                 entry = pte_to_swp_entry(pte);
1277                 frame = swp_type(entry) |
1278                         (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1279                 flags |= PM_SWAP;
1280                 if (is_migration_entry(entry))
1281                         page = migration_entry_to_page(entry);
1282
1283                 if (is_device_private_entry(entry))
1284                         page = device_private_entry_to_page(entry);
1285         }
1286
1287         if (page && !PageAnon(page))
1288                 flags |= PM_FILE;
1289         if (page && page_mapcount(page) == 1)
1290                 flags |= PM_MMAP_EXCLUSIVE;
1291         if (vma->vm_flags & VM_SOFTDIRTY)
1292                 flags |= PM_SOFT_DIRTY;
1293
1294         return make_pme(frame, flags);
1295 }
1296
1297 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1298                              struct mm_walk *walk)
1299 {
1300         struct vm_area_struct *vma = walk->vma;
1301         struct pagemapread *pm = walk->private;
1302         spinlock_t *ptl;
1303         pte_t *pte, *orig_pte;
1304         int err = 0;
1305
1306 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1307         ptl = pmd_trans_huge_lock(pmdp, vma);
1308         if (ptl) {
1309                 u64 flags = 0, frame = 0;
1310                 pmd_t pmd = *pmdp;
1311                 struct page *page = NULL;
1312
1313                 if ((vma->vm_flags & VM_SOFTDIRTY) || pmd_soft_dirty(pmd))
1314                         flags |= PM_SOFT_DIRTY;
1315
1316                 if (pmd_present(pmd)) {
1317                         page = pmd_page(pmd);
1318
1319                         flags |= PM_PRESENT;
1320                         if (pm->show_pfn)
1321                                 frame = pmd_pfn(pmd) +
1322                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1323                 }
1324 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1325                 else if (is_swap_pmd(pmd)) {
1326                         swp_entry_t entry = pmd_to_swp_entry(pmd);
1327
1328                         frame = swp_type(entry) |
1329                                 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1330                         flags |= PM_SWAP;
1331                         VM_BUG_ON(!is_pmd_migration_entry(pmd));
1332                         page = migration_entry_to_page(entry);
1333                 }
1334 #endif
1335
1336                 if (page && page_mapcount(page) == 1)
1337                         flags |= PM_MMAP_EXCLUSIVE;
1338
1339                 for (; addr != end; addr += PAGE_SIZE) {
1340                         pagemap_entry_t pme = make_pme(frame, flags);
1341
1342                         err = add_to_pagemap(addr, &pme, pm);
1343                         if (err)
1344                                 break;
1345                         if (pm->show_pfn && (flags & PM_PRESENT))
1346                                 frame++;
1347                 }
1348                 spin_unlock(ptl);
1349                 return err;
1350         }
1351
1352         if (pmd_trans_unstable(pmdp))
1353                 return 0;
1354 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1355
1356         /*
1357          * We can assume that @vma always points to a valid one and @end never
1358          * goes beyond vma->vm_end.
1359          */
1360         orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1361         for (; addr < end; pte++, addr += PAGE_SIZE) {
1362                 pagemap_entry_t pme;
1363
1364                 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1365                 err = add_to_pagemap(addr, &pme, pm);
1366                 if (err)
1367                         break;
1368         }
1369         pte_unmap_unlock(orig_pte, ptl);
1370
1371         cond_resched();
1372
1373         return err;
1374 }
1375
1376 #ifdef CONFIG_HUGETLB_PAGE
1377 /* This function walks within one hugetlb entry in the single call */
1378 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1379                                  unsigned long addr, unsigned long end,
1380                                  struct mm_walk *walk)
1381 {
1382         struct pagemapread *pm = walk->private;
1383         struct vm_area_struct *vma = walk->vma;
1384         u64 flags = 0, frame = 0;
1385         int err = 0;
1386         pte_t pte;
1387
1388         if (vma->vm_flags & VM_SOFTDIRTY)
1389                 flags |= PM_SOFT_DIRTY;
1390
1391         pte = huge_ptep_get(ptep);
1392         if (pte_present(pte)) {
1393                 struct page *page = pte_page(pte);
1394
1395                 if (!PageAnon(page))
1396                         flags |= PM_FILE;
1397
1398                 if (page_mapcount(page) == 1)
1399                         flags |= PM_MMAP_EXCLUSIVE;
1400
1401                 flags |= PM_PRESENT;
1402                 if (pm->show_pfn)
1403                         frame = pte_pfn(pte) +
1404                                 ((addr & ~hmask) >> PAGE_SHIFT);
1405         }
1406
1407         for (; addr != end; addr += PAGE_SIZE) {
1408                 pagemap_entry_t pme = make_pme(frame, flags);
1409
1410                 err = add_to_pagemap(addr, &pme, pm);
1411                 if (err)
1412                         return err;
1413                 if (pm->show_pfn && (flags & PM_PRESENT))
1414                         frame++;
1415         }
1416
1417         cond_resched();
1418
1419         return err;
1420 }
1421 #endif /* HUGETLB_PAGE */
1422
1423 /*
1424  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1425  *
1426  * For each page in the address space, this file contains one 64-bit entry
1427  * consisting of the following:
1428  *
1429  * Bits 0-54  page frame number (PFN) if present
1430  * Bits 0-4   swap type if swapped
1431  * Bits 5-54  swap offset if swapped
1432  * Bit  55    pte is soft-dirty (see Documentation/vm/soft-dirty.txt)
1433  * Bit  56    page exclusively mapped
1434  * Bits 57-60 zero
1435  * Bit  61    page is file-page or shared-anon
1436  * Bit  62    page swapped
1437  * Bit  63    page present
1438  *
1439  * If the page is not present but in swap, then the PFN contains an
1440  * encoding of the swap file number and the page's offset into the
1441  * swap. Unmapped pages return a null PFN. This allows determining
1442  * precisely which pages are mapped (or in swap) and comparing mapped
1443  * pages between processes.
1444  *
1445  * Efficient users of this interface will use /proc/pid/maps to
1446  * determine which areas of memory are actually mapped and llseek to
1447  * skip over unmapped regions.
1448  */
1449 static ssize_t pagemap_read(struct file *file, char __user *buf,
1450                             size_t count, loff_t *ppos)
1451 {
1452         struct mm_struct *mm = file->private_data;
1453         struct pagemapread pm;
1454         struct mm_walk pagemap_walk = {};
1455         unsigned long src;
1456         unsigned long svpfn;
1457         unsigned long start_vaddr;
1458         unsigned long end_vaddr;
1459         int ret = 0, copied = 0;
1460
1461         if (!mm || !mmget_not_zero(mm))
1462                 goto out;
1463
1464         ret = -EINVAL;
1465         /* file position must be aligned */
1466         if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1467                 goto out_mm;
1468
1469         ret = 0;
1470         if (!count)
1471                 goto out_mm;
1472
1473         /* do not disclose physical addresses: attack vector */
1474         pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1475
1476         pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1477         pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY);
1478         ret = -ENOMEM;
1479         if (!pm.buffer)
1480                 goto out_mm;
1481
1482         pagemap_walk.pmd_entry = pagemap_pmd_range;
1483         pagemap_walk.pte_hole = pagemap_pte_hole;
1484 #ifdef CONFIG_HUGETLB_PAGE
1485         pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1486 #endif
1487         pagemap_walk.mm = mm;
1488         pagemap_walk.private = &pm;
1489
1490         src = *ppos;
1491         svpfn = src / PM_ENTRY_BYTES;
1492         start_vaddr = svpfn << PAGE_SHIFT;
1493         end_vaddr = mm->task_size;
1494
1495         /* watch out for wraparound */
1496         if (svpfn > mm->task_size >> PAGE_SHIFT)
1497                 start_vaddr = end_vaddr;
1498
1499         /*
1500          * The odds are that this will stop walking way
1501          * before end_vaddr, because the length of the
1502          * user buffer is tracked in "pm", and the walk
1503          * will stop when we hit the end of the buffer.
1504          */
1505         ret = 0;
1506         while (count && (start_vaddr < end_vaddr)) {
1507                 int len;
1508                 unsigned long end;
1509
1510                 pm.pos = 0;
1511                 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1512                 /* overflow ? */
1513                 if (end < start_vaddr || end > end_vaddr)
1514                         end = end_vaddr;
1515                 down_read(&mm->mmap_sem);
1516                 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1517                 up_read(&mm->mmap_sem);
1518                 start_vaddr = end;
1519
1520                 len = min(count, PM_ENTRY_BYTES * pm.pos);
1521                 if (copy_to_user(buf, pm.buffer, len)) {
1522                         ret = -EFAULT;
1523                         goto out_free;
1524                 }
1525                 copied += len;
1526                 buf += len;
1527                 count -= len;
1528         }
1529         *ppos += copied;
1530         if (!ret || ret == PM_END_OF_BUFFER)
1531                 ret = copied;
1532
1533 out_free:
1534         kfree(pm.buffer);
1535 out_mm:
1536         mmput(mm);
1537 out:
1538         return ret;
1539 }
1540
1541 static int pagemap_open(struct inode *inode, struct file *file)
1542 {
1543         struct mm_struct *mm;
1544
1545         mm = proc_mem_open(inode, PTRACE_MODE_READ);
1546         if (IS_ERR(mm))
1547                 return PTR_ERR(mm);
1548         file->private_data = mm;
1549         return 0;
1550 }
1551
1552 static int pagemap_release(struct inode *inode, struct file *file)
1553 {
1554         struct mm_struct *mm = file->private_data;
1555
1556         if (mm)
1557                 mmdrop(mm);
1558         return 0;
1559 }
1560
1561 const struct file_operations proc_pagemap_operations = {
1562         .llseek         = mem_lseek, /* borrow this */
1563         .read           = pagemap_read,
1564         .open           = pagemap_open,
1565         .release        = pagemap_release,
1566 };
1567 #endif /* CONFIG_PROC_PAGE_MONITOR */
1568
1569 #ifdef CONFIG_NUMA
1570
1571 struct numa_maps {
1572         unsigned long pages;
1573         unsigned long anon;
1574         unsigned long active;
1575         unsigned long writeback;
1576         unsigned long mapcount_max;
1577         unsigned long dirty;
1578         unsigned long swapcache;
1579         unsigned long node[MAX_NUMNODES];
1580 };
1581
1582 struct numa_maps_private {
1583         struct proc_maps_private proc_maps;
1584         struct numa_maps md;
1585 };
1586
1587 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1588                         unsigned long nr_pages)
1589 {
1590         int count = page_mapcount(page);
1591
1592         md->pages += nr_pages;
1593         if (pte_dirty || PageDirty(page))
1594                 md->dirty += nr_pages;
1595
1596         if (PageSwapCache(page))
1597                 md->swapcache += nr_pages;
1598
1599         if (PageActive(page) || PageUnevictable(page))
1600                 md->active += nr_pages;
1601
1602         if (PageWriteback(page))
1603                 md->writeback += nr_pages;
1604
1605         if (PageAnon(page))
1606                 md->anon += nr_pages;
1607
1608         if (count > md->mapcount_max)
1609                 md->mapcount_max = count;
1610
1611         md->node[page_to_nid(page)] += nr_pages;
1612 }
1613
1614 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1615                 unsigned long addr)
1616 {
1617         struct page *page;
1618         int nid;
1619
1620         if (!pte_present(pte))
1621                 return NULL;
1622
1623         page = vm_normal_page(vma, addr, pte);
1624         if (!page)
1625                 return NULL;
1626
1627         if (PageReserved(page))
1628                 return NULL;
1629
1630         nid = page_to_nid(page);
1631         if (!node_isset(nid, node_states[N_MEMORY]))
1632                 return NULL;
1633
1634         return page;
1635 }
1636
1637 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1638 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1639                                               struct vm_area_struct *vma,
1640                                               unsigned long addr)
1641 {
1642         struct page *page;
1643         int nid;
1644
1645         if (!pmd_present(pmd))
1646                 return NULL;
1647
1648         page = vm_normal_page_pmd(vma, addr, pmd);
1649         if (!page)
1650                 return NULL;
1651
1652         if (PageReserved(page))
1653                 return NULL;
1654
1655         nid = page_to_nid(page);
1656         if (!node_isset(nid, node_states[N_MEMORY]))
1657                 return NULL;
1658
1659         return page;
1660 }
1661 #endif
1662
1663 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1664                 unsigned long end, struct mm_walk *walk)
1665 {
1666         struct numa_maps *md = walk->private;
1667         struct vm_area_struct *vma = walk->vma;
1668         spinlock_t *ptl;
1669         pte_t *orig_pte;
1670         pte_t *pte;
1671
1672 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1673         ptl = pmd_trans_huge_lock(pmd, vma);
1674         if (ptl) {
1675                 struct page *page;
1676
1677                 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1678                 if (page)
1679                         gather_stats(page, md, pmd_dirty(*pmd),
1680                                      HPAGE_PMD_SIZE/PAGE_SIZE);
1681                 spin_unlock(ptl);
1682                 return 0;
1683         }
1684
1685         if (pmd_trans_unstable(pmd))
1686                 return 0;
1687 #endif
1688         orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1689         do {
1690                 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1691                 if (!page)
1692                         continue;
1693                 gather_stats(page, md, pte_dirty(*pte), 1);
1694
1695         } while (pte++, addr += PAGE_SIZE, addr != end);
1696         pte_unmap_unlock(orig_pte, ptl);
1697         cond_resched();
1698         return 0;
1699 }
1700 #ifdef CONFIG_HUGETLB_PAGE
1701 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1702                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1703 {
1704         pte_t huge_pte = huge_ptep_get(pte);
1705         struct numa_maps *md;
1706         struct page *page;
1707
1708         if (!pte_present(huge_pte))
1709                 return 0;
1710
1711         page = pte_page(huge_pte);
1712         if (!page)
1713                 return 0;
1714
1715         md = walk->private;
1716         gather_stats(page, md, pte_dirty(huge_pte), 1);
1717         return 0;
1718 }
1719
1720 #else
1721 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1722                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1723 {
1724         return 0;
1725 }
1726 #endif
1727
1728 /*
1729  * Display pages allocated per node and memory policy via /proc.
1730  */
1731 static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1732 {
1733         struct numa_maps_private *numa_priv = m->private;
1734         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1735         struct vm_area_struct *vma = v;
1736         struct numa_maps *md = &numa_priv->md;
1737         struct file *file = vma->vm_file;
1738         struct mm_struct *mm = vma->vm_mm;
1739         struct mm_walk walk = {
1740                 .hugetlb_entry = gather_hugetlb_stats,
1741                 .pmd_entry = gather_pte_stats,
1742                 .private = md,
1743                 .mm = mm,
1744         };
1745         struct mempolicy *pol;
1746         char buffer[64];
1747         int nid;
1748
1749         if (!mm)
1750                 return 0;
1751
1752         /* Ensure we start with an empty set of numa_maps statistics. */
1753         memset(md, 0, sizeof(*md));
1754
1755         pol = __get_vma_policy(vma, vma->vm_start);
1756         if (pol) {
1757                 mpol_to_str(buffer, sizeof(buffer), pol);
1758                 mpol_cond_put(pol);
1759         } else {
1760                 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1761         }
1762
1763         seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1764
1765         if (file) {
1766                 seq_puts(m, " file=");
1767                 seq_file_path(m, file, "\n\t= ");
1768         } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1769                 seq_puts(m, " heap");
1770         } else if (is_stack(vma)) {
1771                 seq_puts(m, " stack");
1772         }
1773
1774         if (is_vm_hugetlb_page(vma))
1775                 seq_puts(m, " huge");
1776
1777         /* mmap_sem is held by m_start */
1778         walk_page_vma(vma, &walk);
1779
1780         if (!md->pages)
1781                 goto out;
1782
1783         if (md->anon)
1784                 seq_printf(m, " anon=%lu", md->anon);
1785
1786         if (md->dirty)
1787                 seq_printf(m, " dirty=%lu", md->dirty);
1788
1789         if (md->pages != md->anon && md->pages != md->dirty)
1790                 seq_printf(m, " mapped=%lu", md->pages);
1791
1792         if (md->mapcount_max > 1)
1793                 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1794
1795         if (md->swapcache)
1796                 seq_printf(m, " swapcache=%lu", md->swapcache);
1797
1798         if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1799                 seq_printf(m, " active=%lu", md->active);
1800
1801         if (md->writeback)
1802                 seq_printf(m, " writeback=%lu", md->writeback);
1803
1804         for_each_node_state(nid, N_MEMORY)
1805                 if (md->node[nid])
1806                         seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1807
1808         seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1809 out:
1810         seq_putc(m, '\n');
1811         m_cache_vma(m, vma);
1812         return 0;
1813 }
1814
1815 static int show_pid_numa_map(struct seq_file *m, void *v)
1816 {
1817         return show_numa_map(m, v, 1);
1818 }
1819
1820 static int show_tid_numa_map(struct seq_file *m, void *v)
1821 {
1822         return show_numa_map(m, v, 0);
1823 }
1824
1825 static const struct seq_operations proc_pid_numa_maps_op = {
1826         .start  = m_start,
1827         .next   = m_next,
1828         .stop   = m_stop,
1829         .show   = show_pid_numa_map,
1830 };
1831
1832 static const struct seq_operations proc_tid_numa_maps_op = {
1833         .start  = m_start,
1834         .next   = m_next,
1835         .stop   = m_stop,
1836         .show   = show_tid_numa_map,
1837 };
1838
1839 static int numa_maps_open(struct inode *inode, struct file *file,
1840                           const struct seq_operations *ops)
1841 {
1842         return proc_maps_open(inode, file, ops,
1843                                 sizeof(struct numa_maps_private));
1844 }
1845
1846 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1847 {
1848         return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1849 }
1850
1851 static int tid_numa_maps_open(struct inode *inode, struct file *file)
1852 {
1853         return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1854 }
1855
1856 const struct file_operations proc_pid_numa_maps_operations = {
1857         .open           = pid_numa_maps_open,
1858         .read           = seq_read,
1859         .llseek         = seq_lseek,
1860         .release        = proc_map_release,
1861 };
1862
1863 const struct file_operations proc_tid_numa_maps_operations = {
1864         .open           = tid_numa_maps_open,
1865         .read           = seq_read,
1866         .llseek         = seq_lseek,
1867         .release        = proc_map_release,
1868 };
1869 #endif /* CONFIG_NUMA */