fs/proc/task_mmu.c

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