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
  20 #include <asm/elf.h>
  21 #include <linux/uaccess.h>
  22 #include <asm/tlbflush.h>
  23 #include "internal.h"
  24
  25 void task_mem(struct seq_file *m, struct mm_struct *mm)
  26 {
  27         unsigned long text, lib, swap, ptes, pmds, anon, file, shmem;
  28         unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
  29
  30         anon = get_mm_counter(mm, MM_ANONPAGES);
  31         file = get_mm_counter(mm, MM_FILEPAGES);
  32         shmem = get_mm_counter(mm, MM_SHMEMPAGES);
  33
  34         /*
  35          * Note: to minimize their overhead, mm maintains hiwater_vm and
  36          * hiwater_rss only when about to *lower* total_vm or rss.  Any
  37          * collector of these hiwater stats must therefore get total_vm
  38          * and rss too, which will usually be the higher.  Barriers? not
  39          * worth the effort, such snapshots can always be inconsistent.
  40          */
  41         hiwater_vm = total_vm = mm->total_vm;
  42         if (hiwater_vm < mm->hiwater_vm)
  43                 hiwater_vm = mm->hiwater_vm;
  44         hiwater_rss = total_rss = anon + file + shmem;
  45         if (hiwater_rss < mm->hiwater_rss)
  46                 hiwater_rss = mm->hiwater_rss;
  47
  48         text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
  49         lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
  50         swap = get_mm_counter(mm, MM_SWAPENTS);
  51         ptes = PTRS_PER_PTE * sizeof(pte_t) * atomic_long_read(&mm->nr_ptes);
  52         pmds = PTRS_PER_PMD * sizeof(pmd_t) * mm_nr_pmds(mm);
  53         seq_printf(m,
  54                 "VmPeak:\t%8lu kB\n"
  55                 "VmSize:\t%8lu kB\n"
  56                 "VmLck:\t%8lu kB\n"
  57                 "VmPin:\t%8lu kB\n"
  58                 "VmHWM:\t%8lu kB\n"
  59                 "VmRSS:\t%8lu kB\n"
  60                 "RssAnon:\t%8lu kB\n"
  61                 "RssFile:\t%8lu kB\n"
  62                 "RssShmem:\t%8lu kB\n"
  63                 "VmData:\t%8lu kB\n"
  64                 "VmStk:\t%8lu kB\n"
  65                 "VmExe:\t%8lu kB\n"
  66                 "VmLib:\t%8lu kB\n"
  67                 "VmPTE:\t%8lu kB\n"
  68                 "VmPMD:\t%8lu kB\n"
  69                 "VmSwap:\t%8lu kB\n",
  70                 hiwater_vm << (PAGE_SHIFT-10),
  71                 total_vm << (PAGE_SHIFT-10),
  72                 mm->locked_vm << (PAGE_SHIFT-10),
  73                 mm->pinned_vm << (PAGE_SHIFT-10),
  74                 hiwater_rss << (PAGE_SHIFT-10),
  75                 total_rss << (PAGE_SHIFT-10),
  76                 anon << (PAGE_SHIFT-10),
  77                 file << (PAGE_SHIFT-10),
  78                 shmem << (PAGE_SHIFT-10),
  79                 mm->data_vm << (PAGE_SHIFT-10),
  80                 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
  81                 ptes >> 10,
  82                 pmds >> 10,
  83                 swap << (PAGE_SHIFT-10));
  84         hugetlb_report_usage(m, mm);
  85 }
  86
  87 unsigned long task_vsize(struct mm_struct *mm)
  88 {
  89         return PAGE_SIZE * mm->total_vm;
  90 }
  91
  92 unsigned long task_statm(struct mm_struct *mm,
  93                          unsigned long *shared, unsigned long *text,
  94                          unsigned long *data, unsigned long *resident)
  95 {
  96         *shared = get_mm_counter(mm, MM_FILEPAGES) +
  97                         get_mm_counter(mm, MM_SHMEMPAGES);
  98         *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
  99                                                                 >> PAGE_SHIFT;
 100         *data = mm->data_vm + mm->stack_vm;
 101         *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
 102         return mm->total_vm;
 103 }
 104
 105 #ifdef CONFIG_NUMA
 106 /*
 107  * Save get_task_policy() for show_numa_map().
 108  */
 109 static void hold_task_mempolicy(struct proc_maps_private *priv)
 110 {
 111         struct task_struct *task = priv->task;
 112
 113         task_lock(task);
 114         priv->task_mempolicy = get_task_policy(task);
 115         mpol_get(priv->task_mempolicy);
 116         task_unlock(task);
 117 }
 118 static void release_task_mempolicy(struct proc_maps_private *priv)
 119 {
 120         mpol_put(priv->task_mempolicy);
 121 }
 122 #else
 123 static void hold_task_mempolicy(struct proc_maps_private *priv)
 124 {
 125 }
 126 static void release_task_mempolicy(struct proc_maps_private *priv)
 127 {
 128 }
 129 #endif
 130
 131 static void vma_stop(struct proc_maps_private *priv)
 132 {
 133         struct mm_struct *mm = priv->mm;
 134
 135         release_task_mempolicy(priv);
 136         up_read(&mm->mmap_sem);
 137         mmput(mm);
 138 }
 139
 140 static struct vm_area_struct *
 141 m_next_vma(struct proc_maps_private *priv, struct vm_area_struct *vma)
 142 {
 143         if (vma == priv->tail_vma)
 144                 return NULL;
 145         return vma->vm_next ?: priv->tail_vma;
 146 }
 147
 148 static void m_cache_vma(struct seq_file *m, struct vm_area_struct *vma)
 149 {
 150         if (m->count < m->size) /* vma is copied successfully */
 151                 m->version = m_next_vma(m->private, vma) ? vma->vm_end : -1UL;
 152 }
 153
 154 static void *m_start(struct seq_file *m, loff_t *ppos)
 155 {
 156         struct proc_maps_private *priv = m->private;
 157         unsigned long last_addr = m->version;
 158         struct mm_struct *mm;
 159         struct vm_area_struct *vma;
 160         unsigned int pos = *ppos;
 161
 162         /* See m_cache_vma(). Zero at the start or after lseek. */
 163         if (last_addr == -1UL)
 164                 return NULL;
 165
 166         priv->task = get_proc_task(priv->inode);
 167         if (!priv->task)
 168                 return ERR_PTR(-ESRCH);
 169
 170         mm = priv->mm;
 171         if (!mm || !mmget_not_zero(mm))
 172                 return NULL;
 173
 174         down_read(&mm->mmap_sem);
 175         hold_task_mempolicy(priv);
 176         priv->tail_vma = get_gate_vma(mm);
 177
 178         if (last_addr) {
 179                 vma = find_vma(mm, last_addr - 1);
 180                 if (vma && vma->vm_start <= last_addr)
 181                         vma = m_next_vma(priv, vma);
 182                 if (vma)
 183                         return vma;
 184         }
 185
 186         m->version = 0;
 187         if (pos < mm->map_count) {
 188                 for (vma = mm->mmap; pos; pos--) {
 189                         m->version = vma->vm_start;
 190                         vma = vma->vm_next;
 191                 }
 192                 return vma;
 193         }
 194
 195         /* we do not bother to update m->version in this case */
 196         if (pos == mm->map_count && priv->tail_vma)
 197                 return priv->tail_vma;
 198
 199         vma_stop(priv);
 200         return NULL;
 201 }
 202
 203 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
 204 {
 205         struct proc_maps_private *priv = m->private;
 206         struct vm_area_struct *next;
 207
 208         (*pos)++;
 209         next = m_next_vma(priv, v);
 210         if (!next)
 211                 vma_stop(priv);
 212         return next;
 213 }
 214
 215 static void m_stop(struct seq_file *m, void *v)
 216 {
 217         struct proc_maps_private *priv = m->private;
 218
 219         if (!IS_ERR_OR_NULL(v))
 220                 vma_stop(priv);
 221         if (priv->task) {
 222                 put_task_struct(priv->task);
 223                 priv->task = NULL;
 224         }
 225 }
 226
 227 static int proc_maps_open(struct inode *inode, struct file *file,
 228                         const struct seq_operations *ops, int psize)
 229 {
 230         struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
 231
 232         if (!priv)
 233                 return -ENOMEM;
 234
 235         priv->inode = inode;
 236         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
 237         if (IS_ERR(priv->mm)) {
 238                 int err = PTR_ERR(priv->mm);
 239
 240                 seq_release_private(inode, file);
 241                 return err;
 242         }
 243
 244         return 0;
 245 }
 246
 247 static int proc_map_release(struct inode *inode, struct file *file)
 248 {
 249         struct seq_file *seq = file->private_data;
 250         struct proc_maps_private *priv = seq->private;
 251
 252         if (priv->mm)
 253                 mmdrop(priv->mm);
 254
 255         return seq_release_private(inode, file);
 256 }
 257
 258 static int do_maps_open(struct inode *inode, struct file *file,
 259                         const struct seq_operations *ops)
 260 {
 261         return proc_maps_open(inode, file, ops,
 262                                 sizeof(struct proc_maps_private));
 263 }
 264
 265 /*
 266  * Indicate if the VMA is a stack for the given task; for
 267  * /proc/PID/maps that is the stack of the main task.
 268  */
 269 static int is_stack(struct proc_maps_private *priv,
 270                     struct vm_area_struct *vma)
 271 {
 272         /*
 273          * We make no effort to guess what a given thread considers to be
 274          * its "stack".  It's not even well-defined for programs written
 275          * languages like Go.
 276          */
 277         return vma->vm_start <= vma->vm_mm->start_stack &&
 278                 vma->vm_end >= vma->vm_mm->start_stack;
 279 }
 280
 281 static void
 282 show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
 283 {
 284         struct mm_struct *mm = vma->vm_mm;
 285         struct file *file = vma->vm_file;
 286         struct proc_maps_private *priv = m->private;
 287         vm_flags_t flags = vma->vm_flags;
 288         unsigned long ino = 0;
 289         unsigned long long pgoff = 0;
 290         unsigned long start, end;
 291         dev_t dev = 0;
 292         const char *name = NULL;
 293
 294         if (file) {
 295                 struct inode *inode = file_inode(vma->vm_file);
 296                 dev = inode->i_sb->s_dev;
 297                 ino = inode->i_ino;
 298                 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
 299         }
 300
 301         /* We don't show the stack guard page in /proc/maps */
 302         start = vma->vm_start;
 303         if (stack_guard_page_start(vma, start))
 304                 start += PAGE_SIZE;
 305         end = vma->vm_end;
 306         if (stack_guard_page_end(vma, end))
 307                 end -= PAGE_SIZE;
 308
 309         seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
 310         seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu ",
 311                         start,
 312                         end,
 313                         flags & VM_READ ? 'r' : '-',
 314                         flags & VM_WRITE ? 'w' : '-',
 315                         flags & VM_EXEC ? 'x' : '-',
 316                         flags & VM_MAYSHARE ? 's' : 'p',
 317                         pgoff,
 318                         MAJOR(dev), MINOR(dev), ino);
 319
 320         /*
 321          * Print the dentry name for named mappings, and a
 322          * special [heap] marker for the heap:
 323          */
 324         if (file) {
 325                 seq_pad(m, ' ');
 326                 seq_file_path(m, file, "\n");
 327                 goto done;
 328         }
 329
 330         if (vma->vm_ops && vma->vm_ops->name) {
 331                 name = vma->vm_ops->name(vma);
 332                 if (name)
 333                         goto done;
 334         }
 335
 336         name = arch_vma_name(vma);
 337         if (!name) {
 338                 if (!mm) {
 339                         name = "[vdso]";
 340                         goto done;
 341                 }
 342
 343                 if (vma->vm_start <= mm->brk &&
 344                     vma->vm_end >= mm->start_brk) {
 345                         name = "[heap]";
 346                         goto done;
 347                 }
 348
 349                 if (is_stack(priv, vma))
 350                         name = "[stack]";
 351         }
 352
 353 done:
 354         if (name) {
 355                 seq_pad(m, ' ');
 356                 seq_puts(m, name);
 357         }
 358         seq_putc(m, '\n');
 359 }
 360
 361 static int show_map(struct seq_file *m, void *v, int is_pid)
 362 {
 363         show_map_vma(m, v, is_pid);
 364         m_cache_vma(m, v);
 365         return 0;
 366 }
 367
 368 static int show_pid_map(struct seq_file *m, void *v)
 369 {
 370         return show_map(m, v, 1);
 371 }
 372
 373 static int show_tid_map(struct seq_file *m, void *v)
 374 {
 375         return show_map(m, v, 0);
 376 }
 377
 378 static const struct seq_operations proc_pid_maps_op = {
 379         .start  = m_start,
 380         .next   = m_next,
 381         .stop   = m_stop,
 382         .show   = show_pid_map
 383 };
 384
 385 static const struct seq_operations proc_tid_maps_op = {
 386         .start  = m_start,
 387         .next   = m_next,
 388         .stop   = m_stop,
 389         .show   = show_tid_map
 390 };
 391
 392 static int pid_maps_open(struct inode *inode, struct file *file)
 393 {
 394         return do_maps_open(inode, file, &proc_pid_maps_op);
 395 }
 396
 397 static int tid_maps_open(struct inode *inode, struct file *file)
 398 {
 399         return do_maps_open(inode, file, &proc_tid_maps_op);
 400 }
 401
 402 const struct file_operations proc_pid_maps_operations = {
 403         .open           = pid_maps_open,
 404         .read           = seq_read,
 405         .llseek         = seq_lseek,
 406         .release        = proc_map_release,
 407 };
 408
 409 const struct file_operations proc_tid_maps_operations = {
 410         .open           = tid_maps_open,
 411         .read           = seq_read,
 412         .llseek         = seq_lseek,
 413         .release        = proc_map_release,
 414 };
 415
 416 /*
 417  * Proportional Set Size(PSS): my share of RSS.
 418  *
 419  * PSS of a process is the count of pages it has in memory, where each
 420  * page is divided by the number of processes sharing it.  So if a
 421  * process has 1000 pages all to itself, and 1000 shared with one other
 422  * process, its PSS will be 1500.
 423  *
 424  * To keep (accumulated) division errors low, we adopt a 64bit
 425  * fixed-point pss counter to minimize division errors. So (pss >>
 426  * PSS_SHIFT) would be the real byte count.
 427  *
 428  * A shift of 12 before division means (assuming 4K page size):
 429  *      - 1M 3-user-pages add up to 8KB errors;
 430  *      - supports mapcount up to 2^24, or 16M;
 431  *      - supports PSS up to 2^52 bytes, or 4PB.
 432  */
 433 #define PSS_SHIFT 12
 434
 435 #ifdef CONFIG_PROC_PAGE_MONITOR
 436 struct mem_size_stats {
 437         unsigned long resident;
 438         unsigned long shared_clean;
 439         unsigned long shared_dirty;
 440         unsigned long private_clean;
 441         unsigned long private_dirty;
 442         unsigned long referenced;
 443         unsigned long anonymous;
 444         unsigned long lazyfree;
 445         unsigned long anonymous_thp;
 446         unsigned long shmem_thp;
 447         unsigned long swap;
 448         unsigned long shared_hugetlb;
 449         unsigned long private_hugetlb;
 450         u64 pss;
 451         u64 swap_pss;
 452         bool check_shmem_swap;
 453 };
 454
 455 static void smaps_account(struct mem_size_stats *mss, struct page *page,
 456                 bool compound, bool young, bool dirty)
 457 {
 458         int i, nr = compound ? 1 << compound_order(page) : 1;
 459         unsigned long size = nr * PAGE_SIZE;
 460
 461         if (PageAnon(page)) {
 462                 mss->anonymous += size;
 463                 if (!PageSwapBacked(page) && !dirty && !PageDirty(page))
 464                         mss->lazyfree += size;
 465         }
 466
 467         mss->resident += size;
 468         /* Accumulate the size in pages that have been accessed. */
 469         if (young || page_is_young(page) || PageReferenced(page))
 470                 mss->referenced += size;
 471
 472         /*
 473          * page_count(page) == 1 guarantees the page is mapped exactly once.
 474          * If any subpage of the compound page mapped with PTE it would elevate
 475          * page_count().
 476          */
 477         if (page_count(page) == 1) {
 478                 if (dirty || PageDirty(page))
 479                         mss->private_dirty += size;
 480                 else
 481                         mss->private_clean += size;
 482                 mss->pss += (u64)size << PSS_SHIFT;
 483                 return;
 484         }
 485
 486         for (i = 0; i < nr; i++, page++) {
 487                 int mapcount = page_mapcount(page);
 488
 489                 if (mapcount >= 2) {
 490                         if (dirty || PageDirty(page))
 491                                 mss->shared_dirty += PAGE_SIZE;
 492                         else
 493                                 mss->shared_clean += PAGE_SIZE;
 494                         mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount;
 495                 } else {
 496                         if (dirty || PageDirty(page))
 497                                 mss->private_dirty += PAGE_SIZE;
 498                         else
 499                                 mss->private_clean += PAGE_SIZE;
 500                         mss->pss += PAGE_SIZE << PSS_SHIFT;
 501                 }
 502         }
 503 }
 504
 505 #ifdef CONFIG_SHMEM
 506 static int smaps_pte_hole(unsigned long addr, unsigned long end,
 507                 struct mm_walk *walk)
 508 {
 509         struct mem_size_stats *mss = walk->private;
 510
 511         mss->swap += shmem_partial_swap_usage(
 512                         walk->vma->vm_file->f_mapping, addr, end);
 513
 514         return 0;
 515 }
 516 #endif
 517
 518 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
 519                 struct mm_walk *walk)
 520 {
 521         struct mem_size_stats *mss = walk->private;
 522         struct vm_area_struct *vma = walk->vma;
 523         struct page *page = NULL;
 524
 525         if (pte_present(*pte)) {
 526                 page = vm_normal_page(vma, addr, *pte);
 527         } else if (is_swap_pte(*pte)) {
 528                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 529
 530                 if (!non_swap_entry(swpent)) {
 531                         int mapcount;
 532
 533                         mss->swap += PAGE_SIZE;
 534                         mapcount = swp_swapcount(swpent);
 535                         if (mapcount >= 2) {
 536                                 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
 537
 538                                 do_div(pss_delta, mapcount);
 539                                 mss->swap_pss += pss_delta;
 540                         } else {
 541                                 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
 542                         }
 543                 } else if (is_migration_entry(swpent))
 544                         page = migration_entry_to_page(swpent);
 545         } else if (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap
 546                                                         && pte_none(*pte))) {
 547                 page = find_get_entry(vma->vm_file->f_mapping,
 548                                                 linear_page_index(vma, addr));
 549                 if (!page)
 550                         return;
 551
 552                 if (radix_tree_exceptional_entry(page))
 553                         mss->swap += PAGE_SIZE;
 554                 else
 555                         put_page(page);
 556
 557                 return;
 558         }
 559
 560         if (!page)
 561                 return;
 562
 563         smaps_account(mss, page, false, pte_young(*pte), pte_dirty(*pte));
 564 }
 565
 566 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 567 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 568                 struct mm_walk *walk)
 569 {
 570         struct mem_size_stats *mss = walk->private;
 571         struct vm_area_struct *vma = walk->vma;
 572         struct page *page;
 573
 574         /* FOLL_DUMP will return -EFAULT on huge zero page */
 575         page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
 576         if (IS_ERR_OR_NULL(page))
 577                 return;
 578         if (PageAnon(page))
 579                 mss->anonymous_thp += HPAGE_PMD_SIZE;
 580         else if (PageSwapBacked(page))
 581                 mss->shmem_thp += HPAGE_PMD_SIZE;
 582         else if (is_zone_device_page(page))
 583                 /* pass */;
 584         else
 585                 VM_BUG_ON_PAGE(1, page);
 586         smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd));
 587 }
 588 #else
 589 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 590                 struct mm_walk *walk)
 591 {
 592 }
 593 #endif
 594
 595 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 596                            struct mm_walk *walk)
 597 {
 598         struct vm_area_struct *vma = walk->vma;
 599         pte_t *pte;
 600         spinlock_t *ptl;
 601
 602         ptl = pmd_trans_huge_lock(pmd, vma);
 603         if (ptl) {
 604                 smaps_pmd_entry(pmd, addr, walk);
 605                 spin_unlock(ptl);
 606                 return 0;
 607         }
 608
 609         if (pmd_trans_unstable(pmd))
 610                 return 0;
 611         /*
 612          * The mmap_sem held all the way back in m_start() is what
 613          * keeps khugepaged out of here and from collapsing things
 614          * in here.
 615          */
 616         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 617         for (; addr != end; pte++, addr += PAGE_SIZE)
 618                 smaps_pte_entry(pte, addr, walk);
 619         pte_unmap_unlock(pte - 1, ptl);
 620         cond_resched();
 621         return 0;
 622 }
 623
 624 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
 625 {
 626         /*
 627          * Don't forget to update Documentation/ on changes.
 628          */
 629         static const char mnemonics[BITS_PER_LONG][2] = {
 630                 /*
 631                  * In case if we meet a flag we don't know about.
 632                  */
 633                 [0 ... (BITS_PER_LONG-1)] = "??",
 634
 635                 [ilog2(VM_READ)]        = "rd",
 636                 [ilog2(VM_WRITE)]       = "wr",
 637                 [ilog2(VM_EXEC)]        = "ex",
 638                 [ilog2(VM_SHARED)]      = "sh",
 639                 [ilog2(VM_MAYREAD)]     = "mr",
 640                 [ilog2(VM_MAYWRITE)]    = "mw",
 641                 [ilog2(VM_MAYEXEC)]     = "me",
 642                 [ilog2(VM_MAYSHARE)]    = "ms",
 643                 [ilog2(VM_GROWSDOWN)]   = "gd",
 644                 [ilog2(VM_PFNMAP)]      = "pf",
 645                 [ilog2(VM_DENYWRITE)]   = "dw",
 646 #ifdef CONFIG_X86_INTEL_MPX
 647                 [ilog2(VM_MPX)]         = "mp",
 648 #endif
 649                 [ilog2(VM_LOCKED)]      = "lo",
 650                 [ilog2(VM_IO)]          = "io",
 651                 [ilog2(VM_SEQ_READ)]    = "sr",
 652                 [ilog2(VM_RAND_READ)]   = "rr",
 653                 [ilog2(VM_DONTCOPY)]    = "dc",
 654                 [ilog2(VM_DONTEXPAND)]  = "de",
 655                 [ilog2(VM_ACCOUNT)]     = "ac",
 656                 [ilog2(VM_NORESERVE)]   = "nr",
 657                 [ilog2(VM_HUGETLB)]     = "ht",
 658                 [ilog2(VM_ARCH_1)]      = "ar",
 659                 [ilog2(VM_DONTDUMP)]    = "dd",
 660 #ifdef CONFIG_MEM_SOFT_DIRTY
 661                 [ilog2(VM_SOFTDIRTY)]   = "sd",
 662 #endif
 663                 [ilog2(VM_MIXEDMAP)]    = "mm",
 664                 [ilog2(VM_HUGEPAGE)]    = "hg",
 665                 [ilog2(VM_NOHUGEPAGE)]  = "nh",
 666                 [ilog2(VM_MERGEABLE)]   = "mg",
 667                 [ilog2(VM_UFFD_MISSING)]= "um",
 668                 [ilog2(VM_UFFD_WP)]     = "uw",
 669 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
 670                 /* These come out via ProtectionKey: */
 671                 [ilog2(VM_PKEY_BIT0)]   = "",
 672                 [ilog2(VM_PKEY_BIT1)]   = "",
 673                 [ilog2(VM_PKEY_BIT2)]   = "",
 674                 [ilog2(VM_PKEY_BIT3)]   = "",
 675 #endif
 676         };
 677         size_t i;
 678
 679         seq_puts(m, "VmFlags: ");
 680         for (i = 0; i < BITS_PER_LONG; i++) {
 681                 if (!mnemonics[i][0])
 682                         continue;
 683                 if (vma->vm_flags & (1UL << i)) {
 684                         seq_printf(m, "%c%c ",
 685                                    mnemonics[i][0], mnemonics[i][1]);
 686                 }
 687         }
 688         seq_putc(m, '\n');
 689 }
 690
 691 #ifdef CONFIG_HUGETLB_PAGE
 692 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
 693                                  unsigned long addr, unsigned long end,
 694                                  struct mm_walk *walk)
 695 {
 696         struct mem_size_stats *mss = walk->private;
 697         struct vm_area_struct *vma = walk->vma;
 698         struct page *page = NULL;
 699
 700         if (pte_present(*pte)) {
 701                 page = vm_normal_page(vma, addr, *pte);
 702         } else if (is_swap_pte(*pte)) {
 703                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 704
 705                 if (is_migration_entry(swpent))
 706                         page = migration_entry_to_page(swpent);
 707         }
 708         if (page) {
 709                 int mapcount = page_mapcount(page);
 710
 711                 if (mapcount >= 2)
 712                         mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
 713                 else
 714                         mss->private_hugetlb += huge_page_size(hstate_vma(vma));
 715         }
 716         return 0;
 717 }
 718 #endif /* HUGETLB_PAGE */
 719
 720 void __weak arch_show_smap(struct seq_file *m, struct vm_area_struct *vma)
 721 {
 722 }
 723
 724 static int show_smap(struct seq_file *m, void *v, int is_pid)
 725 {
 726         struct vm_area_struct *vma = v;
 727         struct mem_size_stats mss;
 728         struct mm_walk smaps_walk = {
 729                 .pmd_entry = smaps_pte_range,
 730 #ifdef CONFIG_HUGETLB_PAGE
 731                 .hugetlb_entry = smaps_hugetlb_range,
 732 #endif
 733                 .mm = vma->vm_mm,
 734                 .private = &mss,
 735         };
 736
 737         memset(&mss, 0, sizeof mss);
 738
 739 #ifdef CONFIG_SHMEM
 740         if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
 741                 /*
 742                  * For shared or readonly shmem mappings we know that all
 743                  * swapped out pages belong to the shmem object, and we can
 744                  * obtain the swap value much more efficiently. For private
 745                  * writable mappings, we might have COW pages that are
 746                  * not affected by the parent swapped out pages of the shmem
 747                  * object, so we have to distinguish them during the page walk.
 748                  * Unless we know that the shmem object (or the part mapped by
 749                  * our VMA) has no swapped out pages at all.
 750                  */
 751                 unsigned long shmem_swapped = shmem_swap_usage(vma);
 752
 753                 if (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
 754                                         !(vma->vm_flags & VM_WRITE)) {
 755                         mss.swap = shmem_swapped;
 756                 } else {
 757                         mss.check_shmem_swap = true;
 758                         smaps_walk.pte_hole = smaps_pte_hole;
 759                 }
 760         }
 761 #endif
 762
 763         /* mmap_sem is held in m_start */
 764         walk_page_vma(vma, &smaps_walk);
 765
 766         show_map_vma(m, vma, is_pid);
 767
 768         seq_printf(m,
 769                    "Size:           %8lu kB\n"
 770                    "Rss:            %8lu kB\n"
 771                    "Pss:            %8lu kB\n"
 772                    "Shared_Clean:   %8lu kB\n"
 773                    "Shared_Dirty:   %8lu kB\n"
 774                    "Private_Clean:  %8lu kB\n"
 775                    "Private_Dirty:  %8lu kB\n"
 776                    "Referenced:     %8lu kB\n"
 777                    "Anonymous:      %8lu kB\n"
 778                    "LazyFree:       %8lu kB\n"
 779                    "AnonHugePages:  %8lu kB\n"
 780                    "ShmemPmdMapped: %8lu kB\n"
 781                    "Shared_Hugetlb: %8lu kB\n"
 782                    "Private_Hugetlb: %7lu kB\n"
 783                    "Swap:           %8lu kB\n"
 784                    "SwapPss:        %8lu kB\n"
 785                    "KernelPageSize: %8lu kB\n"
 786                    "MMUPageSize:    %8lu kB\n"
 787                    "Locked:         %8lu kB\n",
 788                    (vma->vm_end - vma->vm_start) >> 10,
 789                    mss.resident >> 10,
 790                    (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
 791                    mss.shared_clean  >> 10,
 792                    mss.shared_dirty  >> 10,
 793                    mss.private_clean >> 10,
 794                    mss.private_dirty >> 10,
 795                    mss.referenced >> 10,
 796                    mss.anonymous >> 10,
 797                    mss.lazyfree >> 10,
 798                    mss.anonymous_thp >> 10,
 799                    mss.shmem_thp >> 10,
 800                    mss.shared_hugetlb >> 10,
 801                    mss.private_hugetlb >> 10,
 802                    mss.swap >> 10,
 803                    (unsigned long)(mss.swap_pss >> (10 + PSS_SHIFT)),
 804                    vma_kernel_pagesize(vma) >> 10,
 805                    vma_mmu_pagesize(vma) >> 10,
 806                    (vma->vm_flags & VM_LOCKED) ?
 807                         (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
 808
 809         arch_show_smap(m, vma);
 810         show_smap_vma_flags(m, vma);
 811         m_cache_vma(m, vma);
 812         return 0;
 813 }
 814
 815 static int show_pid_smap(struct seq_file *m, void *v)
 816 {
 817         return show_smap(m, v, 1);
 818 }
 819
 820 static int show_tid_smap(struct seq_file *m, void *v)
 821 {
 822         return show_smap(m, v, 0);
 823 }
 824
 825 static const struct seq_operations proc_pid_smaps_op = {
 826         .start  = m_start,
 827         .next   = m_next,
 828         .stop   = m_stop,
 829         .show   = show_pid_smap
 830 };
 831
 832 static const struct seq_operations proc_tid_smaps_op = {
 833         .start  = m_start,
 834         .next   = m_next,
 835         .stop   = m_stop,
 836         .show   = show_tid_smap
 837 };
 838
 839 static int pid_smaps_open(struct inode *inode, struct file *file)
 840 {
 841         return do_maps_open(inode, file, &proc_pid_smaps_op);
 842 }
 843
 844 static int tid_smaps_open(struct inode *inode, struct file *file)
 845 {
 846         return do_maps_open(inode, file, &proc_tid_smaps_op);
 847 }
 848
 849 const struct file_operations proc_pid_smaps_operations = {
 850         .open           = pid_smaps_open,
 851         .read           = seq_read,
 852         .llseek         = seq_lseek,
 853         .release        = proc_map_release,
 854 };
 855
 856 const struct file_operations proc_tid_smaps_operations = {
 857         .open           = tid_smaps_open,
 858         .read           = seq_read,
 859         .llseek         = seq_lseek,
 860         .release        = proc_map_release,
 861 };
 862
 863 enum clear_refs_types {
 864         CLEAR_REFS_ALL = 1,
 865         CLEAR_REFS_ANON,
 866         CLEAR_REFS_MAPPED,
 867         CLEAR_REFS_SOFT_DIRTY,
 868         CLEAR_REFS_MM_HIWATER_RSS,
 869         CLEAR_REFS_LAST,
 870 };
 871
 872 struct clear_refs_private {
 873         enum clear_refs_types type;
 874 };
 875
 876 #ifdef CONFIG_MEM_SOFT_DIRTY
 877 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 878                 unsigned long addr, pte_t *pte)
 879 {
 880         /*
 881          * The soft-dirty tracker uses #PF-s to catch writes
 882          * to pages, so write-protect the pte as well. See the
 883          * Documentation/vm/soft-dirty.txt for full description
 884          * of how soft-dirty works.
 885          */
 886         pte_t ptent = *pte;
 887
 888         if (pte_present(ptent)) {
 889                 ptent = ptep_modify_prot_start(vma->vm_mm, addr, pte);
 890                 ptent = pte_wrprotect(ptent);
 891                 ptent = pte_clear_soft_dirty(ptent);
 892                 ptep_modify_prot_commit(vma->vm_mm, addr, pte, ptent);
 893         } else if (is_swap_pte(ptent)) {
 894                 ptent = pte_swp_clear_soft_dirty(ptent);
 895                 set_pte_at(vma->vm_mm, addr, pte, ptent);
 896         }
 897 }
 898 #else
 899 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 900                 unsigned long addr, pte_t *pte)
 901 {
 902 }
 903 #endif
 904
 905 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
 906 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
 907                 unsigned long addr, pmd_t *pmdp)
 908 {
 909         pmd_t pmd = *pmdp;
 910
 911         /* See comment in change_huge_pmd() */
 912         pmdp_invalidate(vma, addr, pmdp);
 913         if (pmd_dirty(*pmdp))
 914                 pmd = pmd_mkdirty(pmd);
 915         if (pmd_young(*pmdp))
 916                 pmd = pmd_mkyoung(pmd);
 917
 918         pmd = pmd_wrprotect(pmd);
 919         pmd = pmd_clear_soft_dirty(pmd);
 920
 921         set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
 922 }
 923 #else
 924 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
 925                 unsigned long addr, pmd_t *pmdp)
 926 {
 927 }
 928 #endif
 929
 930 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
 931                                 unsigned long end, struct mm_walk *walk)
 932 {
 933         struct clear_refs_private *cp = walk->private;
 934         struct vm_area_struct *vma = walk->vma;
 935         pte_t *pte, ptent;
 936         spinlock_t *ptl;
 937         struct page *page;
 938
 939         ptl = pmd_trans_huge_lock(pmd, vma);
 940         if (ptl) {
 941                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
 942                         clear_soft_dirty_pmd(vma, addr, pmd);
 943                         goto out;
 944                 }
 945
 946                 page = pmd_page(*pmd);
 947
 948                 /* Clear accessed and referenced bits. */
 949                 pmdp_test_and_clear_young(vma, addr, pmd);
 950                 test_and_clear_page_young(page);
 951                 ClearPageReferenced(page);
 952 out:
 953                 spin_unlock(ptl);
 954                 return 0;
 955         }
 956
 957         if (pmd_trans_unstable(pmd))
 958                 return 0;
 959
 960         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 961         for (; addr != end; pte++, addr += PAGE_SIZE) {
 962                 ptent = *pte;
 963
 964                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
 965                         clear_soft_dirty(vma, addr, pte);
 966                         continue;
 967                 }
 968
 969                 if (!pte_present(ptent))
 970                         continue;
 971
 972                 page = vm_normal_page(vma, addr, ptent);
 973                 if (!page)
 974                         continue;
 975
 976                 /* Clear accessed and referenced bits. */
 977                 ptep_test_and_clear_young(vma, addr, pte);
 978                 test_and_clear_page_young(page);
 979                 ClearPageReferenced(page);
 980         }
 981         pte_unmap_unlock(pte - 1, ptl);
 982         cond_resched();
 983         return 0;
 984 }
 985
 986 static int clear_refs_test_walk(unsigned long start, unsigned long end,
 987                                 struct mm_walk *walk)
 988 {
 989         struct clear_refs_private *cp = walk->private;
 990         struct vm_area_struct *vma = walk->vma;
 991
 992         if (vma->vm_flags & VM_PFNMAP)
 993                 return 1;
 994
 995         /*
 996          * Writing 1 to /proc/pid/clear_refs affects all pages.
 997          * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
 998          * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
 999          * Writing 4 to /proc/pid/clear_refs affects all pages.
1000          */
1001         if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1002                 return 1;
1003         if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1004                 return 1;
1005         return 0;
1006 }
1007
1008 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1009                                 size_t count, loff_t *ppos)
1010 {
1011         struct task_struct *task;
1012         char buffer[PROC_NUMBUF];
1013         struct mm_struct *mm;
1014         struct vm_area_struct *vma;
1015         enum clear_refs_types type;
1016         int itype;
1017         int rv;
1018
1019         memset(buffer, 0, sizeof(buffer));
1020         if (count > sizeof(buffer) - 1)
1021                 count = sizeof(buffer) - 1;
1022         if (copy_from_user(buffer, buf, count))
1023                 return -EFAULT;
1024         rv = kstrtoint(strstrip(buffer), 10, &itype);
1025         if (rv < 0)
1026                 return rv;
1027         type = (enum clear_refs_types)itype;
1028         if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1029                 return -EINVAL;
1030
1031         task = get_proc_task(file_inode(file));
1032         if (!task)
1033                 return -ESRCH;
1034         mm = get_task_mm(task);
1035         if (mm) {
1036                 struct clear_refs_private cp = {
1037                         .type = type,
1038                 };
1039                 struct mm_walk clear_refs_walk = {
1040                         .pmd_entry = clear_refs_pte_range,
1041                         .test_walk = clear_refs_test_walk,
1042                         .mm = mm,
1043                         .private = &cp,
1044                 };
1045
1046                 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1047                         if (down_write_killable(&mm->mmap_sem)) {
1048                                 count = -EINTR;
1049                                 goto out_mm;
1050                         }
1051
1052                         /*
1053                          * Writing 5 to /proc/pid/clear_refs resets the peak
1054                          * resident set size to this mm's current rss value.
1055                          */
1056                         reset_mm_hiwater_rss(mm);
1057                         up_write(&mm->mmap_sem);
1058                         goto out_mm;
1059                 }
1060
1061                 down_read(&mm->mmap_sem);
1062                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1063                         for (vma = mm->mmap; vma; vma = vma->vm_next) {
1064                                 if (!(vma->vm_flags & VM_SOFTDIRTY))
1065                                         continue;
1066                                 up_read(&mm->mmap_sem);
1067                                 if (down_write_killable(&mm->mmap_sem)) {
1068                                         count = -EINTR;
1069                                         goto out_mm;
1070                                 }
1071                                 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1072                                         vma->vm_flags &= ~VM_SOFTDIRTY;
1073                                         vma_set_page_prot(vma);
1074                                 }
1075                                 downgrade_write(&mm->mmap_sem);
1076                                 break;
1077                         }
1078                         mmu_notifier_invalidate_range_start(mm, 0, -1);
1079                 }
1080                 walk_page_range(0, mm->highest_vm_end, &clear_refs_walk);
1081                 if (type == CLEAR_REFS_SOFT_DIRTY)
1082                         mmu_notifier_invalidate_range_end(mm, 0, -1);
1083                 flush_tlb_mm(mm);
1084                 up_read(&mm->mmap_sem);
1085 out_mm:
1086                 mmput(mm);
1087         }
1088         put_task_struct(task);
1089
1090         return count;
1091 }
1092
1093 const struct file_operations proc_clear_refs_operations = {
1094         .write          = clear_refs_write,
1095         .llseek         = noop_llseek,
1096 };
1097
1098 typedef struct {
1099         u64 pme;
1100 } pagemap_entry_t;
1101
1102 struct pagemapread {
1103         int pos, len;           /* units: PM_ENTRY_BYTES, not bytes */
1104         pagemap_entry_t *buffer;
1105         bool show_pfn;
1106 };
1107
1108 #define PAGEMAP_WALK_SIZE       (PMD_SIZE)
1109 #define PAGEMAP_WALK_MASK       (PMD_MASK)
1110
1111 #define PM_ENTRY_BYTES          sizeof(pagemap_entry_t)
1112 #define PM_PFRAME_BITS          55
1113 #define PM_PFRAME_MASK          GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1114 #define PM_SOFT_DIRTY           BIT_ULL(55)
1115 #define PM_MMAP_EXCLUSIVE       BIT_ULL(56)
1116 #define PM_FILE                 BIT_ULL(61)
1117 #define PM_SWAP                 BIT_ULL(62)
1118 #define PM_PRESENT              BIT_ULL(63)
1119
1120 #define PM_END_OF_BUFFER    1
1121
1122 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1123 {
1124         return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1125 }
1126
1127 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1128                           struct pagemapread *pm)
1129 {
1130         pm->buffer[pm->pos++] = *pme;
1131         if (pm->pos >= pm->len)
1132                 return PM_END_OF_BUFFER;
1133         return 0;
1134 }
1135
1136 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1137                                 struct mm_walk *walk)
1138 {
1139         struct pagemapread *pm = walk->private;
1140         unsigned long addr = start;
1141         int err = 0;
1142
1143         while (addr < end) {
1144                 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1145                 pagemap_entry_t pme = make_pme(0, 0);
1146                 /* End of address space hole, which we mark as non-present. */
1147                 unsigned long hole_end;
1148
1149                 if (vma)
1150                         hole_end = min(end, vma->vm_start);
1151                 else
1152                         hole_end = end;
1153
1154                 for (; addr < hole_end; addr += PAGE_SIZE) {
1155                         err = add_to_pagemap(addr, &pme, pm);
1156                         if (err)
1157                                 goto out;
1158                 }
1159
1160                 if (!vma)
1161                         break;
1162
1163                 /* Addresses in the VMA. */
1164                 if (vma->vm_flags & VM_SOFTDIRTY)
1165                         pme = make_pme(0, PM_SOFT_DIRTY);
1166                 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1167                         err = add_to_pagemap(addr, &pme, pm);
1168                         if (err)
1169                                 goto out;
1170                 }
1171         }
1172 out:
1173         return err;
1174 }
1175
1176 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1177                 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1178 {
1179         u64 frame = 0, flags = 0;
1180         struct page *page = NULL;
1181
1182         if (pte_present(pte)) {
1183                 if (pm->show_pfn)
1184                         frame = pte_pfn(pte);
1185                 flags |= PM_PRESENT;
1186                 page = vm_normal_page(vma, addr, pte);
1187                 if (pte_soft_dirty(pte))
1188                         flags |= PM_SOFT_DIRTY;
1189         } else if (is_swap_pte(pte)) {
1190                 swp_entry_t entry;
1191                 if (pte_swp_soft_dirty(pte))
1192                         flags |= PM_SOFT_DIRTY;
1193                 entry = pte_to_swp_entry(pte);
1194                 frame = swp_type(entry) |
1195                         (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1196                 flags |= PM_SWAP;
1197                 if (is_migration_entry(entry))
1198                         page = migration_entry_to_page(entry);
1199         }
1200
1201         if (page && !PageAnon(page))
1202                 flags |= PM_FILE;
1203         if (page && page_mapcount(page) == 1)
1204                 flags |= PM_MMAP_EXCLUSIVE;
1205         if (vma->vm_flags & VM_SOFTDIRTY)
1206                 flags |= PM_SOFT_DIRTY;
1207
1208         return make_pme(frame, flags);
1209 }
1210
1211 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1212                              struct mm_walk *walk)
1213 {
1214         struct vm_area_struct *vma = walk->vma;
1215         struct pagemapread *pm = walk->private;
1216         spinlock_t *ptl;
1217         pte_t *pte, *orig_pte;
1218         int err = 0;
1219
1220 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1221         ptl = pmd_trans_huge_lock(pmdp, vma);
1222         if (ptl) {
1223                 u64 flags = 0, frame = 0;
1224                 pmd_t pmd = *pmdp;
1225
1226                 if ((vma->vm_flags & VM_SOFTDIRTY) || pmd_soft_dirty(pmd))
1227                         flags |= PM_SOFT_DIRTY;
1228
1229                 /*
1230                  * Currently pmd for thp is always present because thp
1231                  * can not be swapped-out, migrated, or HWPOISONed
1232                  * (split in such cases instead.)
1233                  * This if-check is just to prepare for future implementation.
1234                  */
1235                 if (pmd_present(pmd)) {
1236                         struct page *page = pmd_page(pmd);
1237
1238                         if (page_mapcount(page) == 1)
1239                                 flags |= PM_MMAP_EXCLUSIVE;
1240
1241                         flags |= PM_PRESENT;
1242                         if (pm->show_pfn)
1243                                 frame = pmd_pfn(pmd) +
1244                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1245                 }
1246
1247                 for (; addr != end; addr += PAGE_SIZE) {
1248                         pagemap_entry_t pme = make_pme(frame, flags);
1249
1250                         err = add_to_pagemap(addr, &pme, pm);
1251                         if (err)
1252                                 break;
1253                         if (pm->show_pfn && (flags & PM_PRESENT))
1254                                 frame++;
1255                 }
1256                 spin_unlock(ptl);
1257                 return err;
1258         }
1259
1260         if (pmd_trans_unstable(pmdp))
1261                 return 0;
1262 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1263
1264         /*
1265          * We can assume that @vma always points to a valid one and @end never
1266          * goes beyond vma->vm_end.
1267          */
1268         orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1269         for (; addr < end; pte++, addr += PAGE_SIZE) {
1270                 pagemap_entry_t pme;
1271
1272                 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1273                 err = add_to_pagemap(addr, &pme, pm);
1274                 if (err)
1275                         break;
1276         }
1277         pte_unmap_unlock(orig_pte, ptl);
1278
1279         cond_resched();
1280
1281         return err;
1282 }
1283
1284 #ifdef CONFIG_HUGETLB_PAGE
1285 /* This function walks within one hugetlb entry in the single call */
1286 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1287                                  unsigned long addr, unsigned long end,
1288                                  struct mm_walk *walk)
1289 {
1290         struct pagemapread *pm = walk->private;
1291         struct vm_area_struct *vma = walk->vma;
1292         u64 flags = 0, frame = 0;
1293         int err = 0;
1294         pte_t pte;
1295
1296         if (vma->vm_flags & VM_SOFTDIRTY)
1297                 flags |= PM_SOFT_DIRTY;
1298
1299         pte = huge_ptep_get(ptep);
1300         if (pte_present(pte)) {
1301                 struct page *page = pte_page(pte);
1302
1303                 if (!PageAnon(page))
1304                         flags |= PM_FILE;
1305
1306                 if (page_mapcount(page) == 1)
1307                         flags |= PM_MMAP_EXCLUSIVE;
1308
1309                 flags |= PM_PRESENT;
1310                 if (pm->show_pfn)
1311                         frame = pte_pfn(pte) +
1312                                 ((addr & ~hmask) >> PAGE_SHIFT);
1313         }
1314
1315         for (; addr != end; addr += PAGE_SIZE) {
1316                 pagemap_entry_t pme = make_pme(frame, flags);
1317
1318                 err = add_to_pagemap(addr, &pme, pm);
1319                 if (err)
1320                         return err;
1321                 if (pm->show_pfn && (flags & PM_PRESENT))
1322                         frame++;
1323         }
1324
1325         cond_resched();
1326
1327         return err;
1328 }
1329 #endif /* HUGETLB_PAGE */
1330
1331 /*
1332  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1333  *
1334  * For each page in the address space, this file contains one 64-bit entry
1335  * consisting of the following:
1336  *
1337  * Bits 0-54  page frame number (PFN) if present
1338  * Bits 0-4   swap type if swapped
1339  * Bits 5-54  swap offset if swapped
1340  * Bit  55    pte is soft-dirty (see Documentation/vm/soft-dirty.txt)
1341  * Bit  56    page exclusively mapped
1342  * Bits 57-60 zero
1343  * Bit  61    page is file-page or shared-anon
1344  * Bit  62    page swapped
1345  * Bit  63    page present
1346  *
1347  * If the page is not present but in swap, then the PFN contains an
1348  * encoding of the swap file number and the page's offset into the
1349  * swap. Unmapped pages return a null PFN. This allows determining
1350  * precisely which pages are mapped (or in swap) and comparing mapped
1351  * pages between processes.
1352  *
1353  * Efficient users of this interface will use /proc/pid/maps to
1354  * determine which areas of memory are actually mapped and llseek to
1355  * skip over unmapped regions.
1356  */
1357 static ssize_t pagemap_read(struct file *file, char __user *buf,
1358                             size_t count, loff_t *ppos)
1359 {
1360         struct mm_struct *mm = file->private_data;
1361         struct pagemapread pm;
1362         struct mm_walk pagemap_walk = {};
1363         unsigned long src;
1364         unsigned long svpfn;
1365         unsigned long start_vaddr;
1366         unsigned long end_vaddr;
1367         int ret = 0, copied = 0;
1368
1369         if (!mm || !mmget_not_zero(mm))
1370                 goto out;
1371
1372         ret = -EINVAL;
1373         /* file position must be aligned */
1374         if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1375                 goto out_mm;
1376
1377         ret = 0;
1378         if (!count)
1379                 goto out_mm;
1380
1381         /* do not disclose physical addresses: attack vector */
1382         pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1383
1384         pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1385         pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY);
1386         ret = -ENOMEM;
1387         if (!pm.buffer)
1388                 goto out_mm;
1389
1390         pagemap_walk.pmd_entry = pagemap_pmd_range;
1391         pagemap_walk.pte_hole = pagemap_pte_hole;
1392 #ifdef CONFIG_HUGETLB_PAGE
1393         pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1394 #endif
1395         pagemap_walk.mm = mm;
1396         pagemap_walk.private = &pm;
1397
1398         src = *ppos;
1399         svpfn = src / PM_ENTRY_BYTES;
1400         start_vaddr = svpfn << PAGE_SHIFT;
1401         end_vaddr = mm->task_size;
1402
1403         /* watch out for wraparound */
1404         if (svpfn > mm->task_size >> PAGE_SHIFT)
1405                 start_vaddr = end_vaddr;
1406
1407         /*
1408          * The odds are that this will stop walking way
1409          * before end_vaddr, because the length of the
1410          * user buffer is tracked in "pm", and the walk
1411          * will stop when we hit the end of the buffer.
1412          */
1413         ret = 0;
1414         while (count && (start_vaddr < end_vaddr)) {
1415                 int len;
1416                 unsigned long end;
1417
1418                 pm.pos = 0;
1419                 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1420                 /* overflow ? */
1421                 if (end < start_vaddr || end > end_vaddr)
1422                         end = end_vaddr;
1423                 down_read(&mm->mmap_sem);
1424                 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1425                 up_read(&mm->mmap_sem);
1426                 start_vaddr = end;
1427
1428                 len = min(count, PM_ENTRY_BYTES * pm.pos);
1429                 if (copy_to_user(buf, pm.buffer, len)) {
1430                         ret = -EFAULT;
1431                         goto out_free;
1432                 }
1433                 copied += len;
1434                 buf += len;
1435                 count -= len;
1436         }
1437         *ppos += copied;
1438         if (!ret || ret == PM_END_OF_BUFFER)
1439                 ret = copied;
1440
1441 out_free:
1442         kfree(pm.buffer);
1443 out_mm:
1444         mmput(mm);
1445 out:
1446         return ret;
1447 }
1448
1449 static int pagemap_open(struct inode *inode, struct file *file)
1450 {
1451         struct mm_struct *mm;
1452
1453         mm = proc_mem_open(inode, PTRACE_MODE_READ);
1454         if (IS_ERR(mm))
1455                 return PTR_ERR(mm);
1456         file->private_data = mm;
1457         return 0;
1458 }
1459
1460 static int pagemap_release(struct inode *inode, struct file *file)
1461 {
1462         struct mm_struct *mm = file->private_data;
1463
1464         if (mm)
1465                 mmdrop(mm);
1466         return 0;
1467 }
1468
1469 const struct file_operations proc_pagemap_operations = {
1470         .llseek         = mem_lseek, /* borrow this */
1471         .read           = pagemap_read,
1472         .open           = pagemap_open,
1473         .release        = pagemap_release,
1474 };
1475 #endif /* CONFIG_PROC_PAGE_MONITOR */
1476
1477 #ifdef CONFIG_NUMA
1478
1479 struct numa_maps {
1480         unsigned long pages;
1481         unsigned long anon;
1482         unsigned long active;
1483         unsigned long writeback;
1484         unsigned long mapcount_max;
1485         unsigned long dirty;
1486         unsigned long swapcache;
1487         unsigned long node[MAX_NUMNODES];
1488 };
1489
1490 struct numa_maps_private {
1491         struct proc_maps_private proc_maps;
1492         struct numa_maps md;
1493 };
1494
1495 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1496                         unsigned long nr_pages)
1497 {
1498         int count = page_mapcount(page);
1499
1500         md->pages += nr_pages;
1501         if (pte_dirty || PageDirty(page))
1502                 md->dirty += nr_pages;
1503
1504         if (PageSwapCache(page))
1505                 md->swapcache += nr_pages;
1506
1507         if (PageActive(page) || PageUnevictable(page))
1508                 md->active += nr_pages;
1509
1510         if (PageWriteback(page))
1511                 md->writeback += nr_pages;
1512
1513         if (PageAnon(page))
1514                 md->anon += nr_pages;
1515
1516         if (count > md->mapcount_max)
1517                 md->mapcount_max = count;
1518
1519         md->node[page_to_nid(page)] += nr_pages;
1520 }
1521
1522 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1523                 unsigned long addr)
1524 {
1525         struct page *page;
1526         int nid;
1527
1528         if (!pte_present(pte))
1529                 return NULL;
1530
1531         page = vm_normal_page(vma, addr, pte);
1532         if (!page)
1533                 return NULL;
1534
1535         if (PageReserved(page))
1536                 return NULL;
1537
1538         nid = page_to_nid(page);
1539         if (!node_isset(nid, node_states[N_MEMORY]))
1540                 return NULL;
1541
1542         return page;
1543 }
1544
1545 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1546 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1547                                               struct vm_area_struct *vma,
1548                                               unsigned long addr)
1549 {
1550         struct page *page;
1551         int nid;
1552
1553         if (!pmd_present(pmd))
1554                 return NULL;
1555
1556         page = vm_normal_page_pmd(vma, addr, pmd);
1557         if (!page)
1558                 return NULL;
1559
1560         if (PageReserved(page))
1561                 return NULL;
1562
1563         nid = page_to_nid(page);
1564         if (!node_isset(nid, node_states[N_MEMORY]))
1565                 return NULL;
1566
1567         return page;
1568 }
1569 #endif
1570
1571 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1572                 unsigned long end, struct mm_walk *walk)
1573 {
1574         struct numa_maps *md = walk->private;
1575         struct vm_area_struct *vma = walk->vma;
1576         spinlock_t *ptl;
1577         pte_t *orig_pte;
1578         pte_t *pte;
1579
1580 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1581         ptl = pmd_trans_huge_lock(pmd, vma);
1582         if (ptl) {
1583                 struct page *page;
1584
1585                 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1586                 if (page)
1587                         gather_stats(page, md, pmd_dirty(*pmd),
1588                                      HPAGE_PMD_SIZE/PAGE_SIZE);
1589                 spin_unlock(ptl);
1590                 return 0;
1591         }
1592
1593         if (pmd_trans_unstable(pmd))
1594                 return 0;
1595 #endif
1596         orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1597         do {
1598                 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1599                 if (!page)
1600                         continue;
1601                 gather_stats(page, md, pte_dirty(*pte), 1);
1602
1603         } while (pte++, addr += PAGE_SIZE, addr != end);
1604         pte_unmap_unlock(orig_pte, ptl);
1605         cond_resched();
1606         return 0;
1607 }
1608 #ifdef CONFIG_HUGETLB_PAGE
1609 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1610                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1611 {
1612         pte_t huge_pte = huge_ptep_get(pte);
1613         struct numa_maps *md;
1614         struct page *page;
1615
1616         if (!pte_present(huge_pte))
1617                 return 0;
1618
1619         page = pte_page(huge_pte);
1620         if (!page)
1621                 return 0;
1622
1623         md = walk->private;
1624         gather_stats(page, md, pte_dirty(huge_pte), 1);
1625         return 0;
1626 }
1627
1628 #else
1629 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1630                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1631 {
1632         return 0;
1633 }
1634 #endif
1635
1636 /*
1637  * Display pages allocated per node and memory policy via /proc.
1638  */
1639 static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1640 {
1641         struct numa_maps_private *numa_priv = m->private;
1642         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1643         struct vm_area_struct *vma = v;
1644         struct numa_maps *md = &numa_priv->md;
1645         struct file *file = vma->vm_file;
1646         struct mm_struct *mm = vma->vm_mm;
1647         struct mm_walk walk = {
1648                 .hugetlb_entry = gather_hugetlb_stats,
1649                 .pmd_entry = gather_pte_stats,
1650                 .private = md,
1651                 .mm = mm,
1652         };
1653         struct mempolicy *pol;
1654         char buffer[64];
1655         int nid;
1656
1657         if (!mm)
1658                 return 0;
1659
1660         /* Ensure we start with an empty set of numa_maps statistics. */
1661         memset(md, 0, sizeof(*md));
1662
1663         pol = __get_vma_policy(vma, vma->vm_start);
1664         if (pol) {
1665                 mpol_to_str(buffer, sizeof(buffer), pol);
1666                 mpol_cond_put(pol);
1667         } else {
1668                 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1669         }
1670
1671         seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1672
1673         if (file) {
1674                 seq_puts(m, " file=");
1675                 seq_file_path(m, file, "\n\t= ");
1676         } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1677                 seq_puts(m, " heap");
1678         } else if (is_stack(proc_priv, vma)) {
1679                 seq_puts(m, " stack");
1680         }
1681
1682         if (is_vm_hugetlb_page(vma))
1683                 seq_puts(m, " huge");
1684
1685         /* mmap_sem is held by m_start */
1686         walk_page_vma(vma, &walk);
1687
1688         if (!md->pages)
1689                 goto out;
1690
1691         if (md->anon)
1692                 seq_printf(m, " anon=%lu", md->anon);
1693
1694         if (md->dirty)
1695                 seq_printf(m, " dirty=%lu", md->dirty);
1696
1697         if (md->pages != md->anon && md->pages != md->dirty)
1698                 seq_printf(m, " mapped=%lu", md->pages);
1699
1700         if (md->mapcount_max > 1)
1701                 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1702
1703         if (md->swapcache)
1704                 seq_printf(m, " swapcache=%lu", md->swapcache);
1705
1706         if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1707                 seq_printf(m, " active=%lu", md->active);
1708
1709         if (md->writeback)
1710                 seq_printf(m, " writeback=%lu", md->writeback);
1711
1712         for_each_node_state(nid, N_MEMORY)
1713                 if (md->node[nid])
1714                         seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1715
1716         seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1717 out:
1718         seq_putc(m, '\n');
1719         m_cache_vma(m, vma);
1720         return 0;
1721 }
1722
1723 static int show_pid_numa_map(struct seq_file *m, void *v)
1724 {
1725         return show_numa_map(m, v, 1);
1726 }
1727
1728 static int show_tid_numa_map(struct seq_file *m, void *v)
1729 {
1730         return show_numa_map(m, v, 0);
1731 }
1732
1733 static const struct seq_operations proc_pid_numa_maps_op = {
1734         .start  = m_start,
1735         .next   = m_next,
1736         .stop   = m_stop,
1737         .show   = show_pid_numa_map,
1738 };
1739
1740 static const struct seq_operations proc_tid_numa_maps_op = {
1741         .start  = m_start,
1742         .next   = m_next,
1743         .stop   = m_stop,
1744         .show   = show_tid_numa_map,
1745 };
1746
1747 static int numa_maps_open(struct inode *inode, struct file *file,
1748                           const struct seq_operations *ops)
1749 {
1750         return proc_maps_open(inode, file, ops,
1751                                 sizeof(struct numa_maps_private));
1752 }
1753
1754 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1755 {
1756         return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1757 }
1758
1759 static int tid_numa_maps_open(struct inode *inode, struct file *file)
1760 {
1761         return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1762 }
1763
1764 const struct file_operations proc_pid_numa_maps_operations = {
1765         .open           = pid_numa_maps_open,
1766         .read           = seq_read,
1767         .llseek         = seq_lseek,
1768         .release        = proc_map_release,
1769 };
1770
1771 const struct file_operations proc_tid_numa_maps_operations = {
1772         .open           = tid_numa_maps_open,
1773         .read           = seq_read,
1774         .llseek         = seq_lseek,
1775         .release        = proc_map_release,
1776 };
1777 #endif /* CONFIG_NUMA */