fs/proc/task_mmu.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include <linux/mm.h>
   3 #include <linux/vmacache.h>
   4 #include <linux/hugetlb.h>
   5 #include <linux/huge_mm.h>
   6 #include <linux/mount.h>
   7 #include <linux/seq_file.h>
   8 #include <linux/highmem.h>
   9 #include <linux/ptrace.h>
  10 #include <linux/slab.h>
  11 #include <linux/pagemap.h>
  12 #include <linux/mempolicy.h>
  13 #include <linux/rmap.h>
  14 #include <linux/swap.h>
  15 #include <linux/sched/mm.h>
  16 #include <linux/swapops.h>
  17 #include <linux/mmu_notifier.h>
  18 #include <linux/page_idle.h>
  19 #include <linux/shmem_fs.h>
  20 #include <linux/uaccess.h>
  21
  22 #include <asm/elf.h>
  23 #include <asm/tlb.h>
  24 #include <asm/tlbflush.h>
  25 #include "internal.h"
  26
  27 void task_mem(struct seq_file *m, struct mm_struct *mm)
  28 {
  29         unsigned long text, lib, swap, ptes, pmds, anon, file, shmem;
  30         unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
  31
  32         anon = get_mm_counter(mm, MM_ANONPAGES);
  33         file = get_mm_counter(mm, MM_FILEPAGES);
  34         shmem = get_mm_counter(mm, MM_SHMEMPAGES);
  35
  36         /*
  37          * Note: to minimize their overhead, mm maintains hiwater_vm and
  38          * hiwater_rss only when about to *lower* total_vm or rss.  Any
  39          * collector of these hiwater stats must therefore get total_vm
  40          * and rss too, which will usually be the higher.  Barriers? not
  41          * worth the effort, such snapshots can always be inconsistent.
  42          */
  43         hiwater_vm = total_vm = mm->total_vm;
  44         if (hiwater_vm < mm->hiwater_vm)
  45                 hiwater_vm = mm->hiwater_vm;
  46         hiwater_rss = total_rss = anon + file + shmem;
  47         if (hiwater_rss < mm->hiwater_rss)
  48                 hiwater_rss = mm->hiwater_rss;
  49
  50         text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
  51         lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
  52         swap = get_mm_counter(mm, MM_SWAPENTS);
  53         ptes = PTRS_PER_PTE * sizeof(pte_t) * atomic_long_read(&mm->nr_ptes);
  54         pmds = PTRS_PER_PMD * sizeof(pmd_t) * mm_nr_pmds(mm);
  55         seq_printf(m,
  56                 "VmPeak:\t%8lu kB\n"
  57                 "VmSize:\t%8lu kB\n"
  58                 "VmLck:\t%8lu kB\n"
  59                 "VmPin:\t%8lu kB\n"
  60                 "VmHWM:\t%8lu kB\n"
  61                 "VmRSS:\t%8lu kB\n"
  62                 "RssAnon:\t%8lu kB\n"
  63                 "RssFile:\t%8lu kB\n"
  64                 "RssShmem:\t%8lu kB\n"
  65                 "VmData:\t%8lu kB\n"
  66                 "VmStk:\t%8lu kB\n"
  67                 "VmExe:\t%8lu kB\n"
  68                 "VmLib:\t%8lu kB\n"
  69                 "VmPTE:\t%8lu kB\n"
  70                 "VmPMD:\t%8lu kB\n"
  71                 "VmSwap:\t%8lu kB\n",
  72                 hiwater_vm << (PAGE_SHIFT-10),
  73                 total_vm << (PAGE_SHIFT-10),
  74                 mm->locked_vm << (PAGE_SHIFT-10),
  75                 mm->pinned_vm << (PAGE_SHIFT-10),
  76                 hiwater_rss << (PAGE_SHIFT-10),
  77                 total_rss << (PAGE_SHIFT-10),
  78                 anon << (PAGE_SHIFT-10),
  79                 file << (PAGE_SHIFT-10),
  80                 shmem << (PAGE_SHIFT-10),
  81                 mm->data_vm << (PAGE_SHIFT-10),
  82                 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
  83                 ptes >> 10,
  84                 pmds >> 10,
  85                 swap << (PAGE_SHIFT-10));
  86         hugetlb_report_usage(m, mm);
  87 }
  88
  89 unsigned long task_vsize(struct mm_struct *mm)
  90 {
  91         return PAGE_SIZE * mm->total_vm;
  92 }
  93
  94 unsigned long task_statm(struct mm_struct *mm,
  95                          unsigned long *shared, unsigned long *text,
  96                          unsigned long *data, unsigned long *resident)
  97 {
  98         *shared = get_mm_counter(mm, MM_FILEPAGES) +
  99                         get_mm_counter(mm, MM_SHMEMPAGES);
 100         *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
 101                                                                 >> PAGE_SHIFT;
 102         *data = mm->data_vm + mm->stack_vm;
 103         *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
 104         return mm->total_vm;
 105 }
 106
 107 #ifdef CONFIG_NUMA
 108 /*
 109  * Save get_task_policy() for show_numa_map().
 110  */
 111 static void hold_task_mempolicy(struct proc_maps_private *priv)
 112 {
 113         struct task_struct *task = priv->task;
 114
 115         task_lock(task);
 116         priv->task_mempolicy = get_task_policy(task);
 117         mpol_get(priv->task_mempolicy);
 118         task_unlock(task);
 119 }
 120 static void release_task_mempolicy(struct proc_maps_private *priv)
 121 {
 122         mpol_put(priv->task_mempolicy);
 123 }
 124 #else
 125 static void hold_task_mempolicy(struct proc_maps_private *priv)
 126 {
 127 }
 128 static void release_task_mempolicy(struct proc_maps_private *priv)
 129 {
 130 }
 131 #endif
 132
 133 static void vma_stop(struct proc_maps_private *priv)
 134 {
 135         struct mm_struct *mm = priv->mm;
 136
 137         release_task_mempolicy(priv);
 138         up_read(&mm->mmap_sem);
 139         mmput(mm);
 140 }
 141
 142 static struct vm_area_struct *
 143 m_next_vma(struct proc_maps_private *priv, struct vm_area_struct *vma)
 144 {
 145         if (vma == priv->tail_vma)
 146                 return NULL;
 147         return vma->vm_next ?: priv->tail_vma;
 148 }
 149
 150 static void m_cache_vma(struct seq_file *m, struct vm_area_struct *vma)
 151 {
 152         if (m->count < m->size) /* vma is copied successfully */
 153                 m->version = m_next_vma(m->private, vma) ? vma->vm_end : -1UL;
 154 }
 155
 156 static void *m_start(struct seq_file *m, loff_t *ppos)
 157 {
 158         struct proc_maps_private *priv = m->private;
 159         unsigned long last_addr = m->version;
 160         struct mm_struct *mm;
 161         struct vm_area_struct *vma;
 162         unsigned int pos = *ppos;
 163
 164         /* See m_cache_vma(). Zero at the start or after lseek. */
 165         if (last_addr == -1UL)
 166                 return NULL;
 167
 168         priv->task = get_proc_task(priv->inode);
 169         if (!priv->task)
 170                 return ERR_PTR(-ESRCH);
 171
 172         mm = priv->mm;
 173         if (!mm || !mmget_not_zero(mm))
 174                 return NULL;
 175
 176         down_read(&mm->mmap_sem);
 177         hold_task_mempolicy(priv);
 178         priv->tail_vma = get_gate_vma(mm);
 179
 180         if (last_addr) {
 181                 vma = find_vma(mm, last_addr - 1);
 182                 if (vma && vma->vm_start <= last_addr)
 183                         vma = m_next_vma(priv, vma);
 184                 if (vma)
 185                         return vma;
 186         }
 187
 188         m->version = 0;
 189         if (pos < mm->map_count) {
 190                 for (vma = mm->mmap; pos; pos--) {
 191                         m->version = vma->vm_start;
 192                         vma = vma->vm_next;
 193                 }
 194                 return vma;
 195         }
 196
 197         /* we do not bother to update m->version in this case */
 198         if (pos == mm->map_count && priv->tail_vma)
 199                 return priv->tail_vma;
 200
 201         vma_stop(priv);
 202         return NULL;
 203 }
 204
 205 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
 206 {
 207         struct proc_maps_private *priv = m->private;
 208         struct vm_area_struct *next;
 209
 210         (*pos)++;
 211         next = m_next_vma(priv, v);
 212         if (!next)
 213                 vma_stop(priv);
 214         return next;
 215 }
 216
 217 static void m_stop(struct seq_file *m, void *v)
 218 {
 219         struct proc_maps_private *priv = m->private;
 220
 221         if (!IS_ERR_OR_NULL(v))
 222                 vma_stop(priv);
 223         if (priv->task) {
 224                 put_task_struct(priv->task);
 225                 priv->task = NULL;
 226         }
 227 }
 228
 229 static int proc_maps_open(struct inode *inode, struct file *file,
 230                         const struct seq_operations *ops, int psize)
 231 {
 232         struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
 233
 234         if (!priv)
 235                 return -ENOMEM;
 236
 237         priv->inode = inode;
 238         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
 239         if (IS_ERR(priv->mm)) {
 240                 int err = PTR_ERR(priv->mm);
 241
 242                 seq_release_private(inode, file);
 243                 return err;
 244         }
 245
 246         return 0;
 247 }
 248
 249 static int proc_map_release(struct inode *inode, struct file *file)
 250 {
 251         struct seq_file *seq = file->private_data;
 252         struct proc_maps_private *priv = seq->private;
 253
 254         if (priv->mm)
 255                 mmdrop(priv->mm);
 256
 257         kfree(priv->rollup);
 258         return seq_release_private(inode, file);
 259 }
 260
 261 static int do_maps_open(struct inode *inode, struct file *file,
 262                         const struct seq_operations *ops)
 263 {
 264         return proc_maps_open(inode, file, ops,
 265                                 sizeof(struct proc_maps_private));
 266 }
 267
 268 /*
 269  * Indicate if the VMA is a stack for the given task; for
 270  * /proc/PID/maps that is the stack of the main task.
 271  */
 272 static int is_stack(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         vm_flags_t flags = vma->vm_flags;
 306         unsigned long ino = 0;
 307         unsigned long long pgoff = 0;
 308         unsigned long start, end;
 309         dev_t dev = 0;
 310         const char *name = NULL;
 311
 312         if (file) {
 313                 struct inode *inode = file_inode(vma->vm_file);
 314                 dev = inode->i_sb->s_dev;
 315                 ino = inode->i_ino;
 316                 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
 317         }
 318
 319         start = vma->vm_start;
 320         end = vma->vm_end;
 321         show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino);
 322
 323         /*
 324          * Print the dentry name for named mappings, and a
 325          * special [heap] marker for the heap:
 326          */
 327         if (file) {
 328                 seq_pad(m, ' ');
 329                 seq_file_path(m, file, "\n");
 330                 goto done;
 331         }
 332
 333         if (vma->vm_ops && vma->vm_ops->name) {
 334                 name = vma->vm_ops->name(vma);
 335                 if (name)
 336                         goto done;
 337         }
 338
 339         name = arch_vma_name(vma);
 340         if (!name) {
 341                 if (!mm) {
 342                         name = "[vdso]";
 343                         goto done;
 344                 }
 345
 346                 if (vma->vm_start <= mm->brk &&
 347                     vma->vm_end >= mm->start_brk) {
 348                         name = "[heap]";
 349                         goto done;
 350                 }
 351
 352                 if (is_stack(vma))
 353                         name = "[stack]";
 354         }
 355
 356 done:
 357         if (name) {
 358                 seq_pad(m, ' ');
 359                 seq_puts(m, name);
 360         }
 361         seq_putc(m, '\n');
 362 }
 363
 364 static int show_map(struct seq_file *m, void *v, int is_pid)
 365 {
 366         show_map_vma(m, v, is_pid);
 367         m_cache_vma(m, v);
 368         return 0;
 369 }
 370
 371 static int show_pid_map(struct seq_file *m, void *v)
 372 {
 373         return show_map(m, v, 1);
 374 }
 375
 376 static int show_tid_map(struct seq_file *m, void *v)
 377 {
 378         return show_map(m, v, 0);
 379 }
 380
 381 static const struct seq_operations proc_pid_maps_op = {
 382         .start  = m_start,
 383         .next   = m_next,
 384         .stop   = m_stop,
 385         .show   = show_pid_map
 386 };
 387
 388 static const struct seq_operations proc_tid_maps_op = {
 389         .start  = m_start,
 390         .next   = m_next,
 391         .stop   = m_stop,
 392         .show   = show_tid_map
 393 };
 394
 395 static int pid_maps_open(struct inode *inode, struct file *file)
 396 {
 397         return do_maps_open(inode, file, &proc_pid_maps_op);
 398 }
 399
 400 static int tid_maps_open(struct inode *inode, struct file *file)
 401 {
 402         return do_maps_open(inode, file, &proc_tid_maps_op);
 403 }
 404
 405 const struct file_operations proc_pid_maps_operations = {
 406         .open           = pid_maps_open,
 407         .read           = seq_read,
 408         .llseek         = seq_lseek,
 409         .release        = proc_map_release,
 410 };
 411
 412 const struct file_operations proc_tid_maps_operations = {
 413         .open           = tid_maps_open,
 414         .read           = seq_read,
 415         .llseek         = seq_lseek,
 416         .release        = proc_map_release,
 417 };
 418
 419 /*
 420  * Proportional Set Size(PSS): my share of RSS.
 421  *
 422  * PSS of a process is the count of pages it has in memory, where each
 423  * page is divided by the number of processes sharing it.  So if a
 424  * process has 1000 pages all to itself, and 1000 shared with one other
 425  * process, its PSS will be 1500.
 426  *
 427  * To keep (accumulated) division errors low, we adopt a 64bit
 428  * fixed-point pss counter to minimize division errors. So (pss >>
 429  * PSS_SHIFT) would be the real byte count.
 430  *
 431  * A shift of 12 before division means (assuming 4K page size):
 432  *      - 1M 3-user-pages add up to 8KB errors;
 433  *      - supports mapcount up to 2^24, or 16M;
 434  *      - supports PSS up to 2^52 bytes, or 4PB.
 435  */
 436 #define PSS_SHIFT 12
 437
 438 #ifdef CONFIG_PROC_PAGE_MONITOR
 439 struct mem_size_stats {
 440         bool first;
 441         unsigned long resident;
 442         unsigned long shared_clean;
 443         unsigned long shared_dirty;
 444         unsigned long private_clean;
 445         unsigned long private_dirty;
 446         unsigned long referenced;
 447         unsigned long anonymous;
 448         unsigned long lazyfree;
 449         unsigned long anonymous_thp;
 450         unsigned long shmem_thp;
 451         unsigned long swap;
 452         unsigned long shared_hugetlb;
 453         unsigned long private_hugetlb;
 454         unsigned long first_vma_start;
 455         u64 pss;
 456         u64 pss_locked;
 457         u64 swap_pss;
 458         bool check_shmem_swap;
 459 };
 460
 461 static void smaps_account(struct mem_size_stats *mss, struct page *page,
 462                 bool compound, bool young, bool dirty, bool locked)
 463 {
 464         int i, nr = compound ? 1 << compound_order(page) : 1;
 465         unsigned long size = nr * PAGE_SIZE;
 466
 467         if (PageAnon(page)) {
 468                 mss->anonymous += size;
 469                 if (!PageSwapBacked(page) && !dirty && !PageDirty(page))
 470                         mss->lazyfree += size;
 471         }
 472
 473         mss->resident += size;
 474         /* Accumulate the size in pages that have been accessed. */
 475         if (young || page_is_young(page) || PageReferenced(page))
 476                 mss->referenced += size;
 477
 478         /*
 479          * page_count(page) == 1 guarantees the page is mapped exactly once.
 480          * If any subpage of the compound page mapped with PTE it would elevate
 481          * page_count().
 482          */
 483         if (page_count(page) == 1) {
 484                 if (dirty || PageDirty(page))
 485                         mss->private_dirty += size;
 486                 else
 487                         mss->private_clean += size;
 488                 mss->pss += (u64)size << PSS_SHIFT;
 489                 if (locked)
 490                         mss->pss_locked += (u64)size << PSS_SHIFT;
 491                 return;
 492         }
 493
 494         for (i = 0; i < nr; i++, page++) {
 495                 int mapcount = page_mapcount(page);
 496                 unsigned long pss = (PAGE_SIZE << PSS_SHIFT);
 497
 498                 if (mapcount >= 2) {
 499                         if (dirty || PageDirty(page))
 500                                 mss->shared_dirty += PAGE_SIZE;
 501                         else
 502                                 mss->shared_clean += PAGE_SIZE;
 503                         mss->pss += pss / mapcount;
 504                         if (locked)
 505                                 mss->pss_locked += pss / mapcount;
 506                 } else {
 507                         if (dirty || PageDirty(page))
 508                                 mss->private_dirty += PAGE_SIZE;
 509                         else
 510                                 mss->private_clean += PAGE_SIZE;
 511                         mss->pss += pss;
 512                         if (locked)
 513                                 mss->pss_locked += pss;
 514                 }
 515         }
 516 }
 517
 518 #ifdef CONFIG_SHMEM
 519 static int smaps_pte_hole(unsigned long addr, unsigned long end,
 520                 struct mm_walk *walk)
 521 {
 522         struct mem_size_stats *mss = walk->private;
 523
 524         mss->swap += shmem_partial_swap_usage(
 525                         walk->vma->vm_file->f_mapping, addr, end);
 526
 527         return 0;
 528 }
 529 #endif
 530
 531 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
 532                 struct mm_walk *walk)
 533 {
 534         struct mem_size_stats *mss = walk->private;
 535         struct vm_area_struct *vma = walk->vma;
 536         bool locked = !!(vma->vm_flags & VM_LOCKED);
 537         struct page *page = NULL;
 538
 539         if (pte_present(*pte)) {
 540                 page = vm_normal_page(vma, addr, *pte);
 541         } else if (is_swap_pte(*pte)) {
 542                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 543
 544                 if (!non_swap_entry(swpent)) {
 545                         int mapcount;
 546
 547                         mss->swap += PAGE_SIZE;
 548                         mapcount = swp_swapcount(swpent);
 549                         if (mapcount >= 2) {
 550                                 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
 551
 552                                 do_div(pss_delta, mapcount);
 553                                 mss->swap_pss += pss_delta;
 554                         } else {
 555                                 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
 556                         }
 557                 } else if (is_migration_entry(swpent))
 558                         page = migration_entry_to_page(swpent);
 559                 else if (is_device_private_entry(swpent))
 560                         page = device_private_entry_to_page(swpent);
 561         } else if (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap
 562                                                         && pte_none(*pte))) {
 563                 page = find_get_entry(vma->vm_file->f_mapping,
 564                                                 linear_page_index(vma, addr));
 565                 if (!page)
 566                         return;
 567
 568                 if (radix_tree_exceptional_entry(page))
 569                         mss->swap += PAGE_SIZE;
 570                 else
 571                         put_page(page);
 572
 573                 return;
 574         }
 575
 576         if (!page)
 577                 return;
 578
 579         smaps_account(mss, page, false, pte_young(*pte), pte_dirty(*pte), locked);
 580 }
 581
 582 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 583 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 584                 struct mm_walk *walk)
 585 {
 586         struct mem_size_stats *mss = walk->private;
 587         struct vm_area_struct *vma = walk->vma;
 588         bool locked = !!(vma->vm_flags & VM_LOCKED);
 589         struct page *page;
 590
 591         /* FOLL_DUMP will return -EFAULT on huge zero page */
 592         page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
 593         if (IS_ERR_OR_NULL(page))
 594                 return;
 595         if (PageAnon(page))
 596                 mss->anonymous_thp += HPAGE_PMD_SIZE;
 597         else if (PageSwapBacked(page))
 598                 mss->shmem_thp += HPAGE_PMD_SIZE;
 599         else if (is_zone_device_page(page))
 600                 /* pass */;
 601         else
 602                 VM_BUG_ON_PAGE(1, page);
 603         smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd), locked);
 604 }
 605 #else
 606 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 607                 struct mm_walk *walk)
 608 {
 609 }
 610 #endif
 611
 612 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 613                            struct mm_walk *walk)
 614 {
 615         struct vm_area_struct *vma = walk->vma;
 616         pte_t *pte;
 617         spinlock_t *ptl;
 618
 619         ptl = pmd_trans_huge_lock(pmd, vma);
 620         if (ptl) {
 621                 if (pmd_present(*pmd))
 622                         smaps_pmd_entry(pmd, addr, walk);
 623                 spin_unlock(ptl);
 624                 goto out;
 625         }
 626
 627         if (pmd_trans_unstable(pmd))
 628                 goto out;
 629         /*
 630          * The mmap_sem held all the way back in m_start() is what
 631          * keeps khugepaged out of here and from collapsing things
 632          * in here.
 633          */
 634         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 635         for (; addr != end; pte++, addr += PAGE_SIZE)
 636                 smaps_pte_entry(pte, addr, walk);
 637         pte_unmap_unlock(pte - 1, ptl);
 638 out:
 639         cond_resched();
 640         return 0;
 641 }
 642
 643 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
 644 {
 645         /*
 646          * Don't forget to update Documentation/ on changes.
 647          */
 648         static const char mnemonics[BITS_PER_LONG][2] = {
 649                 /*
 650                  * In case if we meet a flag we don't know about.
 651                  */
 652                 [0 ... (BITS_PER_LONG-1)] = "??",
 653
 654                 [ilog2(VM_READ)]        = "rd",
 655                 [ilog2(VM_WRITE)]       = "wr",
 656                 [ilog2(VM_EXEC)]        = "ex",
 657                 [ilog2(VM_SHARED)]      = "sh",
 658                 [ilog2(VM_MAYREAD)]     = "mr",
 659                 [ilog2(VM_MAYWRITE)]    = "mw",
 660                 [ilog2(VM_MAYEXEC)]     = "me",
 661                 [ilog2(VM_MAYSHARE)]    = "ms",
 662                 [ilog2(VM_GROWSDOWN)]   = "gd",
 663                 [ilog2(VM_PFNMAP)]      = "pf",
 664                 [ilog2(VM_DENYWRITE)]   = "dw",
 665 #ifdef CONFIG_X86_INTEL_MPX
 666                 [ilog2(VM_MPX)]         = "mp",
 667 #endif
 668                 [ilog2(VM_LOCKED)]      = "lo",
 669                 [ilog2(VM_IO)]          = "io",
 670                 [ilog2(VM_SEQ_READ)]    = "sr",
 671                 [ilog2(VM_RAND_READ)]   = "rr",
 672                 [ilog2(VM_DONTCOPY)]    = "dc",
 673                 [ilog2(VM_DONTEXPAND)]  = "de",
 674                 [ilog2(VM_ACCOUNT)]     = "ac",
 675                 [ilog2(VM_NORESERVE)]   = "nr",
 676                 [ilog2(VM_HUGETLB)]     = "ht",
 677                 [ilog2(VM_ARCH_1)]      = "ar",
 678                 [ilog2(VM_WIPEONFORK)]  = "wf",
 679                 [ilog2(VM_DONTDUMP)]    = "dd",
 680 #ifdef CONFIG_MEM_SOFT_DIRTY
 681                 [ilog2(VM_SOFTDIRTY)]   = "sd",
 682 #endif
 683                 [ilog2(VM_MIXEDMAP)]    = "mm",
 684                 [ilog2(VM_HUGEPAGE)]    = "hg",
 685                 [ilog2(VM_NOHUGEPAGE)]  = "nh",
 686                 [ilog2(VM_MERGEABLE)]   = "mg",
 687                 [ilog2(VM_UFFD_MISSING)]= "um",
 688                 [ilog2(VM_UFFD_WP)]     = "uw",
 689 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
 690                 /* These come out via ProtectionKey: */
 691                 [ilog2(VM_PKEY_BIT0)]   = "",
 692                 [ilog2(VM_PKEY_BIT1)]   = "",
 693                 [ilog2(VM_PKEY_BIT2)]   = "",
 694                 [ilog2(VM_PKEY_BIT3)]   = "",
 695 #endif
 696         };
 697         size_t i;
 698
 699         seq_puts(m, "VmFlags: ");
 700         for (i = 0; i < BITS_PER_LONG; i++) {
 701                 if (!mnemonics[i][0])
 702                         continue;
 703                 if (vma->vm_flags & (1UL << i)) {
 704                         seq_printf(m, "%c%c ",
 705                                    mnemonics[i][0], mnemonics[i][1]);
 706                 }
 707         }
 708         seq_putc(m, '\n');
 709 }
 710
 711 #ifdef CONFIG_HUGETLB_PAGE
 712 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
 713                                  unsigned long addr, unsigned long end,
 714                                  struct mm_walk *walk)
 715 {
 716         struct mem_size_stats *mss = walk->private;
 717         struct vm_area_struct *vma = walk->vma;
 718         struct page *page = NULL;
 719
 720         if (pte_present(*pte)) {
 721                 page = vm_normal_page(vma, addr, *pte);
 722         } else if (is_swap_pte(*pte)) {
 723                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 724
 725                 if (is_migration_entry(swpent))
 726                         page = migration_entry_to_page(swpent);
 727                 else if (is_device_private_entry(swpent))
 728                         page = device_private_entry_to_page(swpent);
 729         }
 730         if (page) {
 731                 int mapcount = page_mapcount(page);
 732
 733                 if (mapcount >= 2)
 734                         mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
 735                 else
 736                         mss->private_hugetlb += huge_page_size(hstate_vma(vma));
 737         }
 738         return 0;
 739 }
 740 #endif /* HUGETLB_PAGE */
 741
 742 void __weak arch_show_smap(struct seq_file *m, struct vm_area_struct *vma)
 743 {
 744 }
 745
 746 static int show_smap(struct seq_file *m, void *v, int is_pid)
 747 {
 748         struct proc_maps_private *priv = m->private;
 749         struct vm_area_struct *vma = v;
 750         struct mem_size_stats mss_stack;
 751         struct mem_size_stats *mss;
 752         struct mm_walk smaps_walk = {
 753                 .pmd_entry = smaps_pte_range,
 754 #ifdef CONFIG_HUGETLB_PAGE
 755                 .hugetlb_entry = smaps_hugetlb_range,
 756 #endif
 757                 .mm = vma->vm_mm,
 758         };
 759         int ret = 0;
 760         bool rollup_mode;
 761         bool last_vma;
 762
 763         if (priv->rollup) {
 764                 rollup_mode = true;
 765                 mss = priv->rollup;
 766                 if (mss->first) {
 767                         mss->first_vma_start = vma->vm_start;
 768                         mss->first = false;
 769                 }
 770                 last_vma = !m_next_vma(priv, vma);
 771         } else {
 772                 rollup_mode = false;
 773                 memset(&mss_stack, 0, sizeof(mss_stack));
 774                 mss = &mss_stack;
 775         }
 776
 777         smaps_walk.private = mss;
 778
 779 #ifdef CONFIG_SHMEM
 780         /* In case of smaps_rollup, reset the value from previous vma */
 781         mss->check_shmem_swap = false;
 782         if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
 783                 /*
 784                  * For shared or readonly shmem mappings we know that all
 785                  * swapped out pages belong to the shmem object, and we can
 786                  * obtain the swap value much more efficiently. For private
 787                  * writable mappings, we might have COW pages that are
 788                  * not affected by the parent swapped out pages of the shmem
 789                  * object, so we have to distinguish them during the page walk.
 790                  * Unless we know that the shmem object (or the part mapped by
 791                  * our VMA) has no swapped out pages at all.
 792                  */
 793                 unsigned long shmem_swapped = shmem_swap_usage(vma);
 794
 795                 if (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
 796                                         !(vma->vm_flags & VM_WRITE)) {
 797                         mss->swap += shmem_swapped;
 798                 } else {
 799                         mss->check_shmem_swap = true;
 800                         smaps_walk.pte_hole = smaps_pte_hole;
 801                 }
 802         }
 803 #endif
 804         /* mmap_sem is held in m_start */
 805         walk_page_vma(vma, &smaps_walk);
 806
 807         if (!rollup_mode) {
 808                 show_map_vma(m, vma, is_pid);
 809         } else if (last_vma) {
 810                 show_vma_header_prefix(
 811                         m, mss->first_vma_start, vma->vm_end, 0, 0, 0, 0);
 812                 seq_pad(m, ' ');
 813                 seq_puts(m, "[rollup]\n");
 814         } else {
 815                 ret = SEQ_SKIP;
 816         }
 817
 818         if (!rollup_mode)
 819                 seq_printf(m,
 820                            "Size:           %8lu kB\n"
 821                            "KernelPageSize: %8lu kB\n"
 822                            "MMUPageSize:    %8lu kB\n",
 823                            (vma->vm_end - vma->vm_start) >> 10,
 824                            vma_kernel_pagesize(vma) >> 10,
 825                            vma_mmu_pagesize(vma) >> 10);
 826
 827
 828         if (!rollup_mode || last_vma)
 829                 seq_printf(m,
 830                            "Rss:            %8lu kB\n"
 831                            "Pss:            %8lu kB\n"
 832                            "Shared_Clean:   %8lu kB\n"
 833                            "Shared_Dirty:   %8lu kB\n"
 834                            "Private_Clean:  %8lu kB\n"
 835                            "Private_Dirty:  %8lu kB\n"
 836                            "Referenced:     %8lu kB\n"
 837                            "Anonymous:      %8lu kB\n"
 838                            "LazyFree:       %8lu kB\n"
 839                            "AnonHugePages:  %8lu kB\n"
 840                            "ShmemPmdMapped: %8lu kB\n"
 841                            "Shared_Hugetlb: %8lu kB\n"
 842                            "Private_Hugetlb: %7lu kB\n"
 843                            "Swap:           %8lu kB\n"
 844                            "SwapPss:        %8lu kB\n"
 845                            "Locked:         %8lu kB\n",
 846                            mss->resident >> 10,
 847                            (unsigned long)(mss->pss >> (10 + PSS_SHIFT)),
 848                            mss->shared_clean  >> 10,
 849                            mss->shared_dirty  >> 10,
 850                            mss->private_clean >> 10,
 851                            mss->private_dirty >> 10,
 852                            mss->referenced >> 10,
 853                            mss->anonymous >> 10,
 854                            mss->lazyfree >> 10,
 855                            mss->anonymous_thp >> 10,
 856                            mss->shmem_thp >> 10,
 857                            mss->shared_hugetlb >> 10,
 858                            mss->private_hugetlb >> 10,
 859                            mss->swap >> 10,
 860                            (unsigned long)(mss->swap_pss >> (10 + PSS_SHIFT)),
 861                            (unsigned long)(mss->pss_locked >> (10 + PSS_SHIFT)));
 862
 863         if (!rollup_mode) {
 864                 arch_show_smap(m, vma);
 865                 show_smap_vma_flags(m, vma);
 866         }
 867         m_cache_vma(m, vma);
 868         return ret;
 869 }
 870
 871 static int show_pid_smap(struct seq_file *m, void *v)
 872 {
 873         return show_smap(m, v, 1);
 874 }
 875
 876 static int show_tid_smap(struct seq_file *m, void *v)
 877 {
 878         return show_smap(m, v, 0);
 879 }
 880
 881 static const struct seq_operations proc_pid_smaps_op = {
 882         .start  = m_start,
 883         .next   = m_next,
 884         .stop   = m_stop,
 885         .show   = show_pid_smap
 886 };
 887
 888 static const struct seq_operations proc_tid_smaps_op = {
 889         .start  = m_start,
 890         .next   = m_next,
 891         .stop   = m_stop,
 892         .show   = show_tid_smap
 893 };
 894
 895 static int pid_smaps_open(struct inode *inode, struct file *file)
 896 {
 897         return do_maps_open(inode, file, &proc_pid_smaps_op);
 898 }
 899
 900 static int pid_smaps_rollup_open(struct inode *inode, struct file *file)
 901 {
 902         struct seq_file *seq;
 903         struct proc_maps_private *priv;
 904         int ret = do_maps_open(inode, file, &proc_pid_smaps_op);
 905
 906         if (ret < 0)
 907                 return ret;
 908         seq = file->private_data;
 909         priv = seq->private;
 910         priv->rollup = kzalloc(sizeof(*priv->rollup), GFP_KERNEL);
 911         if (!priv->rollup) {
 912                 proc_map_release(inode, file);
 913                 return -ENOMEM;
 914         }
 915         priv->rollup->first = true;
 916         return 0;
 917 }
 918
 919 static int tid_smaps_open(struct inode *inode, struct file *file)
 920 {
 921         return do_maps_open(inode, file, &proc_tid_smaps_op);
 922 }
 923
 924 const struct file_operations proc_pid_smaps_operations = {
 925         .open           = pid_smaps_open,
 926         .read           = seq_read,
 927         .llseek         = seq_lseek,
 928         .release        = proc_map_release,
 929 };
 930
 931 const struct file_operations proc_pid_smaps_rollup_operations = {
 932         .open           = pid_smaps_rollup_open,
 933         .read           = seq_read,
 934         .llseek         = seq_lseek,
 935         .release        = proc_map_release,
 936 };
 937
 938 const struct file_operations proc_tid_smaps_operations = {
 939         .open           = tid_smaps_open,
 940         .read           = seq_read,
 941         .llseek         = seq_lseek,
 942         .release        = proc_map_release,
 943 };
 944
 945 enum clear_refs_types {
 946         CLEAR_REFS_ALL = 1,
 947         CLEAR_REFS_ANON,
 948         CLEAR_REFS_MAPPED,
 949         CLEAR_REFS_SOFT_DIRTY,
 950         CLEAR_REFS_MM_HIWATER_RSS,
 951         CLEAR_REFS_LAST,
 952 };
 953
 954 struct clear_refs_private {
 955         enum clear_refs_types type;
 956 };
 957
 958 #ifdef CONFIG_MEM_SOFT_DIRTY
 959 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 960                 unsigned long addr, pte_t *pte)
 961 {
 962         /*
 963          * The soft-dirty tracker uses #PF-s to catch writes
 964          * to pages, so write-protect the pte as well. See the
 965          * Documentation/vm/soft-dirty.txt for full description
 966          * of how soft-dirty works.
 967          */
 968         pte_t ptent = *pte;
 969
 970         if (pte_present(ptent)) {
 971                 ptent = ptep_modify_prot_start(vma->vm_mm, addr, pte);
 972                 ptent = pte_wrprotect(ptent);
 973                 ptent = pte_clear_soft_dirty(ptent);
 974                 ptep_modify_prot_commit(vma->vm_mm, addr, pte, ptent);
 975         } else if (is_swap_pte(ptent)) {
 976                 ptent = pte_swp_clear_soft_dirty(ptent);
 977                 set_pte_at(vma->vm_mm, addr, pte, ptent);
 978         }
 979 }
 980 #else
 981 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 982                 unsigned long addr, pte_t *pte)
 983 {
 984 }
 985 #endif
 986
 987 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
 988 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
 989                 unsigned long addr, pmd_t *pmdp)
 990 {
 991         pmd_t pmd = *pmdp;
 992
 993         if (pmd_present(pmd)) {
 994                 /* See comment in change_huge_pmd() */
 995                 pmdp_invalidate(vma, addr, pmdp);
 996                 if (pmd_dirty(*pmdp))
 997                         pmd = pmd_mkdirty(pmd);
 998                 if (pmd_young(*pmdp))
 999                         pmd = pmd_mkyoung(pmd);
1000
1001                 pmd = pmd_wrprotect(pmd);
1002                 pmd = pmd_clear_soft_dirty(pmd);
1003
1004                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1005         } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1006                 pmd = pmd_swp_clear_soft_dirty(pmd);
1007                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1008         }
1009 }
1010 #else
1011 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1012                 unsigned long addr, pmd_t *pmdp)
1013 {
1014 }
1015 #endif
1016
1017 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
1018                                 unsigned long end, struct mm_walk *walk)
1019 {
1020         struct clear_refs_private *cp = walk->private;
1021         struct vm_area_struct *vma = walk->vma;
1022         pte_t *pte, ptent;
1023         spinlock_t *ptl;
1024         struct page *page;
1025
1026         ptl = pmd_trans_huge_lock(pmd, vma);
1027         if (ptl) {
1028                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1029                         clear_soft_dirty_pmd(vma, addr, pmd);
1030                         goto out;
1031                 }
1032
1033                 if (!pmd_present(*pmd))
1034                         goto out;
1035
1036                 page = pmd_page(*pmd);
1037
1038                 /* Clear accessed and referenced bits. */
1039                 pmdp_test_and_clear_young(vma, addr, pmd);
1040                 test_and_clear_page_young(page);
1041                 ClearPageReferenced(page);
1042 out:
1043                 spin_unlock(ptl);
1044                 return 0;
1045         }
1046
1047         if (pmd_trans_unstable(pmd))
1048                 return 0;
1049
1050         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1051         for (; addr != end; pte++, addr += PAGE_SIZE) {
1052                 ptent = *pte;
1053
1054                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1055                         clear_soft_dirty(vma, addr, pte);
1056                         continue;
1057                 }
1058
1059                 if (!pte_present(ptent))
1060                         continue;
1061
1062                 page = vm_normal_page(vma, addr, ptent);
1063                 if (!page)
1064                         continue;
1065
1066                 /* Clear accessed and referenced bits. */
1067                 ptep_test_and_clear_young(vma, addr, pte);
1068                 test_and_clear_page_young(page);
1069                 ClearPageReferenced(page);
1070         }
1071         pte_unmap_unlock(pte - 1, ptl);
1072         cond_resched();
1073         return 0;
1074 }
1075
1076 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1077                                 struct mm_walk *walk)
1078 {
1079         struct clear_refs_private *cp = walk->private;
1080         struct vm_area_struct *vma = walk->vma;
1081
1082         if (vma->vm_flags & VM_PFNMAP)
1083                 return 1;
1084
1085         /*
1086          * Writing 1 to /proc/pid/clear_refs affects all pages.
1087          * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1088          * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1089          * Writing 4 to /proc/pid/clear_refs affects all pages.
1090          */
1091         if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1092                 return 1;
1093         if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1094                 return 1;
1095         return 0;
1096 }
1097
1098 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1099                                 size_t count, loff_t *ppos)
1100 {
1101         struct task_struct *task;
1102         char buffer[PROC_NUMBUF];
1103         struct mm_struct *mm;
1104         struct vm_area_struct *vma;
1105         enum clear_refs_types type;
1106         struct mmu_gather tlb;
1107         int itype;
1108         int rv;
1109
1110         memset(buffer, 0, sizeof(buffer));
1111         if (count > sizeof(buffer) - 1)
1112                 count = sizeof(buffer) - 1;
1113         if (copy_from_user(buffer, buf, count))
1114                 return -EFAULT;
1115         rv = kstrtoint(strstrip(buffer), 10, &itype);
1116         if (rv < 0)
1117                 return rv;
1118         type = (enum clear_refs_types)itype;
1119         if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1120                 return -EINVAL;
1121
1122         task = get_proc_task(file_inode(file));
1123         if (!task)
1124                 return -ESRCH;
1125         mm = get_task_mm(task);
1126         if (mm) {
1127                 struct clear_refs_private cp = {
1128                         .type = type,
1129                 };
1130                 struct mm_walk clear_refs_walk = {
1131                         .pmd_entry = clear_refs_pte_range,
1132                         .test_walk = clear_refs_test_walk,
1133                         .mm = mm,
1134                         .private = &cp,
1135                 };
1136
1137                 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1138                         if (down_write_killable(&mm->mmap_sem)) {
1139                                 count = -EINTR;
1140                                 goto out_mm;
1141                         }
1142
1143                         /*
1144                          * Writing 5 to /proc/pid/clear_refs resets the peak
1145                          * resident set size to this mm's current rss value.
1146                          */
1147                         reset_mm_hiwater_rss(mm);
1148                         up_write(&mm->mmap_sem);
1149                         goto out_mm;
1150                 }
1151
1152                 down_read(&mm->mmap_sem);
1153                 tlb_gather_mmu(&tlb, mm, 0, -1);
1154                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1155                         for (vma = mm->mmap; vma; vma = vma->vm_next) {
1156                                 if (!(vma->vm_flags & VM_SOFTDIRTY))
1157                                         continue;
1158                                 up_read(&mm->mmap_sem);
1159                                 if (down_write_killable(&mm->mmap_sem)) {
1160                                         count = -EINTR;
1161                                         goto out_mm;
1162                                 }
1163                                 /*
1164                                  * Avoid to modify vma->vm_flags
1165                                  * without locked ops while the
1166                                  * coredump reads the vm_flags.
1167                                  */
1168                                 if (!mmget_still_valid(mm)) {
1169                                         /*
1170                                          * Silently return "count"
1171                                          * like if get_task_mm()
1172                                          * failed. FIXME: should this
1173                                          * function have returned
1174                                          * -ESRCH if get_task_mm()
1175                                          * failed like if
1176                                          * get_proc_task() fails?
1177                                          */
1178                                         up_write(&mm->mmap_sem);
1179                                         goto out_mm;
1180                                 }
1181                                 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1182                                         vma->vm_flags &= ~VM_SOFTDIRTY;
1183                                         vma_set_page_prot(vma);
1184                                 }
1185                                 downgrade_write(&mm->mmap_sem);
1186                                 break;
1187                         }
1188                         mmu_notifier_invalidate_range_start(mm, 0, -1);
1189                 }
1190                 walk_page_range(0, mm->highest_vm_end, &clear_refs_walk);
1191                 if (type == CLEAR_REFS_SOFT_DIRTY)
1192                         mmu_notifier_invalidate_range_end(mm, 0, -1);
1193                 tlb_finish_mmu(&tlb, 0, -1);
1194                 up_read(&mm->mmap_sem);
1195 out_mm:
1196                 mmput(mm);
1197         }
1198         put_task_struct(task);
1199
1200         return count;
1201 }
1202
1203 const struct file_operations proc_clear_refs_operations = {
1204         .write          = clear_refs_write,
1205         .llseek         = noop_llseek,
1206 };
1207
1208 typedef struct {
1209         u64 pme;
1210 } pagemap_entry_t;
1211
1212 struct pagemapread {
1213         int pos, len;           /* units: PM_ENTRY_BYTES, not bytes */
1214         pagemap_entry_t *buffer;
1215         bool show_pfn;
1216 };
1217
1218 #define PAGEMAP_WALK_SIZE       (PMD_SIZE)
1219 #define PAGEMAP_WALK_MASK       (PMD_MASK)
1220
1221 #define PM_ENTRY_BYTES          sizeof(pagemap_entry_t)
1222 #define PM_PFRAME_BITS          55
1223 #define PM_PFRAME_MASK          GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1224 #define PM_SOFT_DIRTY           BIT_ULL(55)
1225 #define PM_MMAP_EXCLUSIVE       BIT_ULL(56)
1226 #define PM_FILE                 BIT_ULL(61)
1227 #define PM_SWAP                 BIT_ULL(62)
1228 #define PM_PRESENT              BIT_ULL(63)
1229
1230 #define PM_END_OF_BUFFER    1
1231
1232 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1233 {
1234         return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1235 }
1236
1237 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1238                           struct pagemapread *pm)
1239 {
1240         pm->buffer[pm->pos++] = *pme;
1241         if (pm->pos >= pm->len)
1242                 return PM_END_OF_BUFFER;
1243         return 0;
1244 }
1245
1246 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1247                                 struct mm_walk *walk)
1248 {
1249         struct pagemapread *pm = walk->private;
1250         unsigned long addr = start;
1251         int err = 0;
1252
1253         while (addr < end) {
1254                 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1255                 pagemap_entry_t pme = make_pme(0, 0);
1256                 /* End of address space hole, which we mark as non-present. */
1257                 unsigned long hole_end;
1258
1259                 if (vma)
1260                         hole_end = min(end, vma->vm_start);
1261                 else
1262                         hole_end = end;
1263
1264                 for (; addr < hole_end; addr += PAGE_SIZE) {
1265                         err = add_to_pagemap(addr, &pme, pm);
1266                         if (err)
1267                                 goto out;
1268                 }
1269
1270                 if (!vma)
1271                         break;
1272
1273                 /* Addresses in the VMA. */
1274                 if (vma->vm_flags & VM_SOFTDIRTY)
1275                         pme = make_pme(0, PM_SOFT_DIRTY);
1276                 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1277                         err = add_to_pagemap(addr, &pme, pm);
1278                         if (err)
1279                                 goto out;
1280                 }
1281         }
1282 out:
1283         return err;
1284 }
1285
1286 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1287                 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1288 {
1289         u64 frame = 0, flags = 0;
1290         struct page *page = NULL;
1291
1292         if (pte_present(pte)) {
1293                 if (pm->show_pfn)
1294                         frame = pte_pfn(pte);
1295                 flags |= PM_PRESENT;
1296                 page = _vm_normal_page(vma, addr, pte, true);
1297                 if (pte_soft_dirty(pte))
1298                         flags |= PM_SOFT_DIRTY;
1299         } else if (is_swap_pte(pte)) {
1300                 swp_entry_t entry;
1301                 if (pte_swp_soft_dirty(pte))
1302                         flags |= PM_SOFT_DIRTY;
1303                 entry = pte_to_swp_entry(pte);
1304                 if (pm->show_pfn)
1305                         frame = swp_type(entry) |
1306                                 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1307                 flags |= PM_SWAP;
1308                 if (is_migration_entry(entry))
1309                         page = migration_entry_to_page(entry);
1310
1311                 if (is_device_private_entry(entry))
1312                         page = device_private_entry_to_page(entry);
1313         }
1314
1315         if (page && !PageAnon(page))
1316                 flags |= PM_FILE;
1317         if (page && page_mapcount(page) == 1)
1318                 flags |= PM_MMAP_EXCLUSIVE;
1319         if (vma->vm_flags & VM_SOFTDIRTY)
1320                 flags |= PM_SOFT_DIRTY;
1321
1322         return make_pme(frame, flags);
1323 }
1324
1325 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1326                              struct mm_walk *walk)
1327 {
1328         struct vm_area_struct *vma = walk->vma;
1329         struct pagemapread *pm = walk->private;
1330         spinlock_t *ptl;
1331         pte_t *pte, *orig_pte;
1332         int err = 0;
1333
1334 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1335         ptl = pmd_trans_huge_lock(pmdp, vma);
1336         if (ptl) {
1337                 u64 flags = 0, frame = 0;
1338                 pmd_t pmd = *pmdp;
1339                 struct page *page = NULL;
1340
1341                 if (vma->vm_flags & VM_SOFTDIRTY)
1342                         flags |= PM_SOFT_DIRTY;
1343
1344                 if (pmd_present(pmd)) {
1345                         page = pmd_page(pmd);
1346
1347                         flags |= PM_PRESENT;
1348                         if (pmd_soft_dirty(pmd))
1349                                 flags |= PM_SOFT_DIRTY;
1350                         if (pm->show_pfn)
1351                                 frame = pmd_pfn(pmd) +
1352                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1353                 }
1354 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1355                 else if (is_swap_pmd(pmd)) {
1356                         swp_entry_t entry = pmd_to_swp_entry(pmd);
1357                         unsigned long offset;
1358
1359                         if (pm->show_pfn) {
1360                                 offset = swp_offset(entry) +
1361                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1362                                 frame = swp_type(entry) |
1363                                         (offset << MAX_SWAPFILES_SHIFT);
1364                         }
1365                         flags |= PM_SWAP;
1366                         if (pmd_swp_soft_dirty(pmd))
1367                                 flags |= PM_SOFT_DIRTY;
1368                         VM_BUG_ON(!is_pmd_migration_entry(pmd));
1369                         page = migration_entry_to_page(entry);
1370                 }
1371 #endif
1372
1373                 if (page && page_mapcount(page) == 1)
1374                         flags |= PM_MMAP_EXCLUSIVE;
1375
1376                 for (; addr != end; addr += PAGE_SIZE) {
1377                         pagemap_entry_t pme = make_pme(frame, flags);
1378
1379                         err = add_to_pagemap(addr, &pme, pm);
1380                         if (err)
1381                                 break;
1382                         if (pm->show_pfn) {
1383                                 if (flags & PM_PRESENT)
1384                                         frame++;
1385                                 else if (flags & PM_SWAP)
1386                                         frame += (1 << MAX_SWAPFILES_SHIFT);
1387                         }
1388                 }
1389                 spin_unlock(ptl);
1390                 return err;
1391         }
1392
1393         if (pmd_trans_unstable(pmdp))
1394                 return 0;
1395 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1396
1397         /*
1398          * We can assume that @vma always points to a valid one and @end never
1399          * goes beyond vma->vm_end.
1400          */
1401         orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1402         for (; addr < end; pte++, addr += PAGE_SIZE) {
1403                 pagemap_entry_t pme;
1404
1405                 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1406                 err = add_to_pagemap(addr, &pme, pm);
1407                 if (err)
1408                         break;
1409         }
1410         pte_unmap_unlock(orig_pte, ptl);
1411
1412         cond_resched();
1413
1414         return err;
1415 }
1416
1417 #ifdef CONFIG_HUGETLB_PAGE
1418 /* This function walks within one hugetlb entry in the single call */
1419 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1420                                  unsigned long addr, unsigned long end,
1421                                  struct mm_walk *walk)
1422 {
1423         struct pagemapread *pm = walk->private;
1424         struct vm_area_struct *vma = walk->vma;
1425         u64 flags = 0, frame = 0;
1426         int err = 0;
1427         pte_t pte;
1428
1429         if (vma->vm_flags & VM_SOFTDIRTY)
1430                 flags |= PM_SOFT_DIRTY;
1431
1432         pte = huge_ptep_get(ptep);
1433         if (pte_present(pte)) {
1434                 struct page *page = pte_page(pte);
1435
1436                 if (!PageAnon(page))
1437                         flags |= PM_FILE;
1438
1439                 if (page_mapcount(page) == 1)
1440                         flags |= PM_MMAP_EXCLUSIVE;
1441
1442                 flags |= PM_PRESENT;
1443                 if (pm->show_pfn)
1444                         frame = pte_pfn(pte) +
1445                                 ((addr & ~hmask) >> PAGE_SHIFT);
1446         }
1447
1448         for (; addr != end; addr += PAGE_SIZE) {
1449                 pagemap_entry_t pme = make_pme(frame, flags);
1450
1451                 err = add_to_pagemap(addr, &pme, pm);
1452                 if (err)
1453                         return err;
1454                 if (pm->show_pfn && (flags & PM_PRESENT))
1455                         frame++;
1456         }
1457
1458         cond_resched();
1459
1460         return err;
1461 }
1462 #endif /* HUGETLB_PAGE */
1463
1464 /*
1465  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1466  *
1467  * For each page in the address space, this file contains one 64-bit entry
1468  * consisting of the following:
1469  *
1470  * Bits 0-54  page frame number (PFN) if present
1471  * Bits 0-4   swap type if swapped
1472  * Bits 5-54  swap offset if swapped
1473  * Bit  55    pte is soft-dirty (see Documentation/vm/soft-dirty.txt)
1474  * Bit  56    page exclusively mapped
1475  * Bits 57-60 zero
1476  * Bit  61    page is file-page or shared-anon
1477  * Bit  62    page swapped
1478  * Bit  63    page present
1479  *
1480  * If the page is not present but in swap, then the PFN contains an
1481  * encoding of the swap file number and the page's offset into the
1482  * swap. Unmapped pages return a null PFN. This allows determining
1483  * precisely which pages are mapped (or in swap) and comparing mapped
1484  * pages between processes.
1485  *
1486  * Efficient users of this interface will use /proc/pid/maps to
1487  * determine which areas of memory are actually mapped and llseek to
1488  * skip over unmapped regions.
1489  */
1490 static ssize_t pagemap_read(struct file *file, char __user *buf,
1491                             size_t count, loff_t *ppos)
1492 {
1493         struct mm_struct *mm = file->private_data;
1494         struct pagemapread pm;
1495         struct mm_walk pagemap_walk = {};
1496         unsigned long src;
1497         unsigned long svpfn;
1498         unsigned long start_vaddr;
1499         unsigned long end_vaddr;
1500         int ret = 0, copied = 0;
1501
1502         if (!mm || !mmget_not_zero(mm))
1503                 goto out;
1504
1505         ret = -EINVAL;
1506         /* file position must be aligned */
1507         if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1508                 goto out_mm;
1509
1510         ret = 0;
1511         if (!count)
1512                 goto out_mm;
1513
1514         /* do not disclose physical addresses: attack vector */
1515         pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1516
1517         pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1518         pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_KERNEL);
1519         ret = -ENOMEM;
1520         if (!pm.buffer)
1521                 goto out_mm;
1522
1523         pagemap_walk.pmd_entry = pagemap_pmd_range;
1524         pagemap_walk.pte_hole = pagemap_pte_hole;
1525 #ifdef CONFIG_HUGETLB_PAGE
1526         pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1527 #endif
1528         pagemap_walk.mm = mm;
1529         pagemap_walk.private = &pm;
1530
1531         src = *ppos;
1532         svpfn = src / PM_ENTRY_BYTES;
1533         start_vaddr = svpfn << PAGE_SHIFT;
1534         end_vaddr = mm->task_size;
1535
1536         /* watch out for wraparound */
1537         if (svpfn > mm->task_size >> PAGE_SHIFT)
1538                 start_vaddr = end_vaddr;
1539
1540         /*
1541          * The odds are that this will stop walking way
1542          * before end_vaddr, because the length of the
1543          * user buffer is tracked in "pm", and the walk
1544          * will stop when we hit the end of the buffer.
1545          */
1546         ret = 0;
1547         while (count && (start_vaddr < end_vaddr)) {
1548                 int len;
1549                 unsigned long end;
1550
1551                 pm.pos = 0;
1552                 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1553                 /* overflow ? */
1554                 if (end < start_vaddr || end > end_vaddr)
1555                         end = end_vaddr;
1556                 down_read(&mm->mmap_sem);
1557                 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1558                 up_read(&mm->mmap_sem);
1559                 start_vaddr = end;
1560
1561                 len = min(count, PM_ENTRY_BYTES * pm.pos);
1562                 if (copy_to_user(buf, pm.buffer, len)) {
1563                         ret = -EFAULT;
1564                         goto out_free;
1565                 }
1566                 copied += len;
1567                 buf += len;
1568                 count -= len;
1569         }
1570         *ppos += copied;
1571         if (!ret || ret == PM_END_OF_BUFFER)
1572                 ret = copied;
1573
1574 out_free:
1575         kfree(pm.buffer);
1576 out_mm:
1577         mmput(mm);
1578 out:
1579         return ret;
1580 }
1581
1582 static int pagemap_open(struct inode *inode, struct file *file)
1583 {
1584         struct mm_struct *mm;
1585
1586         mm = proc_mem_open(inode, PTRACE_MODE_READ);
1587         if (IS_ERR(mm))
1588                 return PTR_ERR(mm);
1589         file->private_data = mm;
1590         return 0;
1591 }
1592
1593 static int pagemap_release(struct inode *inode, struct file *file)
1594 {
1595         struct mm_struct *mm = file->private_data;
1596
1597         if (mm)
1598                 mmdrop(mm);
1599         return 0;
1600 }
1601
1602 const struct file_operations proc_pagemap_operations = {
1603         .llseek         = mem_lseek, /* borrow this */
1604         .read           = pagemap_read,
1605         .open           = pagemap_open,
1606         .release        = pagemap_release,
1607 };
1608 #endif /* CONFIG_PROC_PAGE_MONITOR */
1609
1610 #ifdef CONFIG_NUMA
1611
1612 struct numa_maps {
1613         unsigned long pages;
1614         unsigned long anon;
1615         unsigned long active;
1616         unsigned long writeback;
1617         unsigned long mapcount_max;
1618         unsigned long dirty;
1619         unsigned long swapcache;
1620         unsigned long node[MAX_NUMNODES];
1621 };
1622
1623 struct numa_maps_private {
1624         struct proc_maps_private proc_maps;
1625         struct numa_maps md;
1626 };
1627
1628 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1629                         unsigned long nr_pages)
1630 {
1631         int count = page_mapcount(page);
1632
1633         md->pages += nr_pages;
1634         if (pte_dirty || PageDirty(page))
1635                 md->dirty += nr_pages;
1636
1637         if (PageSwapCache(page))
1638                 md->swapcache += nr_pages;
1639
1640         if (PageActive(page) || PageUnevictable(page))
1641                 md->active += nr_pages;
1642
1643         if (PageWriteback(page))
1644                 md->writeback += nr_pages;
1645
1646         if (PageAnon(page))
1647                 md->anon += nr_pages;
1648
1649         if (count > md->mapcount_max)
1650                 md->mapcount_max = count;
1651
1652         md->node[page_to_nid(page)] += nr_pages;
1653 }
1654
1655 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1656                 unsigned long addr)
1657 {
1658         struct page *page;
1659         int nid;
1660
1661         if (!pte_present(pte))
1662                 return NULL;
1663
1664         page = vm_normal_page(vma, addr, pte);
1665         if (!page)
1666                 return NULL;
1667
1668         if (PageReserved(page))
1669                 return NULL;
1670
1671         nid = page_to_nid(page);
1672         if (!node_isset(nid, node_states[N_MEMORY]))
1673                 return NULL;
1674
1675         return page;
1676 }
1677
1678 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1679 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1680                                               struct vm_area_struct *vma,
1681                                               unsigned long addr)
1682 {
1683         struct page *page;
1684         int nid;
1685
1686         if (!pmd_present(pmd))
1687                 return NULL;
1688
1689         page = vm_normal_page_pmd(vma, addr, pmd);
1690         if (!page)
1691                 return NULL;
1692
1693         if (PageReserved(page))
1694                 return NULL;
1695
1696         nid = page_to_nid(page);
1697         if (!node_isset(nid, node_states[N_MEMORY]))
1698                 return NULL;
1699
1700         return page;
1701 }
1702 #endif
1703
1704 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1705                 unsigned long end, struct mm_walk *walk)
1706 {
1707         struct numa_maps *md = walk->private;
1708         struct vm_area_struct *vma = walk->vma;
1709         spinlock_t *ptl;
1710         pte_t *orig_pte;
1711         pte_t *pte;
1712
1713 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1714         ptl = pmd_trans_huge_lock(pmd, vma);
1715         if (ptl) {
1716                 struct page *page;
1717
1718                 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1719                 if (page)
1720                         gather_stats(page, md, pmd_dirty(*pmd),
1721                                      HPAGE_PMD_SIZE/PAGE_SIZE);
1722                 spin_unlock(ptl);
1723                 return 0;
1724         }
1725
1726         if (pmd_trans_unstable(pmd))
1727                 return 0;
1728 #endif
1729         orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1730         do {
1731                 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1732                 if (!page)
1733                         continue;
1734                 gather_stats(page, md, pte_dirty(*pte), 1);
1735
1736         } while (pte++, addr += PAGE_SIZE, addr != end);
1737         pte_unmap_unlock(orig_pte, ptl);
1738         cond_resched();
1739         return 0;
1740 }
1741 #ifdef CONFIG_HUGETLB_PAGE
1742 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1743                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1744 {
1745         pte_t huge_pte = huge_ptep_get(pte);
1746         struct numa_maps *md;
1747         struct page *page;
1748
1749         if (!pte_present(huge_pte))
1750                 return 0;
1751
1752         page = pte_page(huge_pte);
1753         if (!page)
1754                 return 0;
1755
1756         md = walk->private;
1757         gather_stats(page, md, pte_dirty(huge_pte), 1);
1758         return 0;
1759 }
1760
1761 #else
1762 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1763                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1764 {
1765         return 0;
1766 }
1767 #endif
1768
1769 /*
1770  * Display pages allocated per node and memory policy via /proc.
1771  */
1772 static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1773 {
1774         struct numa_maps_private *numa_priv = m->private;
1775         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1776         struct vm_area_struct *vma = v;
1777         struct numa_maps *md = &numa_priv->md;
1778         struct file *file = vma->vm_file;
1779         struct mm_struct *mm = vma->vm_mm;
1780         struct mm_walk walk = {
1781                 .hugetlb_entry = gather_hugetlb_stats,
1782                 .pmd_entry = gather_pte_stats,
1783                 .private = md,
1784                 .mm = mm,
1785         };
1786         struct mempolicy *pol;
1787         char buffer[64];
1788         int nid;
1789
1790         if (!mm)
1791                 return 0;
1792
1793         /* Ensure we start with an empty set of numa_maps statistics. */
1794         memset(md, 0, sizeof(*md));
1795
1796         pol = __get_vma_policy(vma, vma->vm_start);
1797         if (pol) {
1798                 mpol_to_str(buffer, sizeof(buffer), pol);
1799                 mpol_cond_put(pol);
1800         } else {
1801                 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1802         }
1803
1804         seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1805
1806         if (file) {
1807                 seq_puts(m, " file=");
1808                 seq_file_path(m, file, "\n\t= ");
1809         } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1810                 seq_puts(m, " heap");
1811         } else if (is_stack(vma)) {
1812                 seq_puts(m, " stack");
1813         }
1814
1815         if (is_vm_hugetlb_page(vma))
1816                 seq_puts(m, " huge");
1817
1818         /* mmap_sem is held by m_start */
1819         walk_page_vma(vma, &walk);
1820
1821         if (!md->pages)
1822                 goto out;
1823
1824         if (md->anon)
1825                 seq_printf(m, " anon=%lu", md->anon);
1826
1827         if (md->dirty)
1828                 seq_printf(m, " dirty=%lu", md->dirty);
1829
1830         if (md->pages != md->anon && md->pages != md->dirty)
1831                 seq_printf(m, " mapped=%lu", md->pages);
1832
1833         if (md->mapcount_max > 1)
1834                 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1835
1836         if (md->swapcache)
1837                 seq_printf(m, " swapcache=%lu", md->swapcache);
1838
1839         if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1840                 seq_printf(m, " active=%lu", md->active);
1841
1842         if (md->writeback)
1843                 seq_printf(m, " writeback=%lu", md->writeback);
1844
1845         for_each_node_state(nid, N_MEMORY)
1846                 if (md->node[nid])
1847                         seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1848
1849         seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1850 out:
1851         seq_putc(m, '\n');
1852         m_cache_vma(m, vma);
1853         return 0;
1854 }
1855
1856 static int show_pid_numa_map(struct seq_file *m, void *v)
1857 {
1858         return show_numa_map(m, v, 1);
1859 }
1860
1861 static int show_tid_numa_map(struct seq_file *m, void *v)
1862 {
1863         return show_numa_map(m, v, 0);
1864 }
1865
1866 static const struct seq_operations proc_pid_numa_maps_op = {
1867         .start  = m_start,
1868         .next   = m_next,
1869         .stop   = m_stop,
1870         .show   = show_pid_numa_map,
1871 };
1872
1873 static const struct seq_operations proc_tid_numa_maps_op = {
1874         .start  = m_start,
1875         .next   = m_next,
1876         .stop   = m_stop,
1877         .show   = show_tid_numa_map,
1878 };
1879
1880 static int numa_maps_open(struct inode *inode, struct file *file,
1881                           const struct seq_operations *ops)
1882 {
1883         return proc_maps_open(inode, file, ops,
1884                                 sizeof(struct numa_maps_private));
1885 }
1886
1887 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1888 {
1889         return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1890 }
1891
1892 static int tid_numa_maps_open(struct inode *inode, struct file *file)
1893 {
1894         return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1895 }
1896
1897 const struct file_operations proc_pid_numa_maps_operations = {
1898         .open           = pid_numa_maps_open,
1899         .read           = seq_read,
1900         .llseek         = seq_lseek,
1901         .release        = proc_map_release,
1902 };
1903
1904 const struct file_operations proc_tid_numa_maps_operations = {
1905         .open           = tid_numa_maps_open,
1906         .read           = seq_read,
1907         .llseek         = seq_lseek,
1908         .release        = proc_map_release,
1909 };
1910 #endif /* CONFIG_NUMA */