fs/proc/task_mmu.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include <linux/mm.h>
   3 #include <linux/vmacache.h>
   4 #include <linux/hugetlb.h>
   5 #include <linux/huge_mm.h>
   6 #include <linux/mount.h>
   7 #include <linux/seq_file.h>
   8 #include <linux/highmem.h>
   9 #include <linux/ptrace.h>
  10 #include <linux/slab.h>
  11 #include <linux/pagemap.h>
  12 #include <linux/mempolicy.h>
  13 #include <linux/rmap.h>
  14 #include <linux/swap.h>
  15 #include <linux/sched/mm.h>
  16 #include <linux/swapops.h>
  17 #include <linux/mmu_notifier.h>
  18 #include <linux/page_idle.h>
  19 #include <linux/shmem_fs.h>
  20 #include <linux/uaccess.h>
  21
  22 #include <asm/elf.h>
  23 #include <asm/tlb.h>
  24 #include <asm/tlbflush.h>
  25 #include "internal.h"
  26
  27 void task_mem(struct seq_file *m, struct mm_struct *mm)
  28 {
  29         unsigned long text, lib, swap, anon, file, shmem;
  30         unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
  31
  32         anon = get_mm_counter(mm, MM_ANONPAGES);
  33         file = get_mm_counter(mm, MM_FILEPAGES);
  34         shmem = get_mm_counter(mm, MM_SHMEMPAGES);
  35
  36         /*
  37          * Note: to minimize their overhead, mm maintains hiwater_vm and
  38          * hiwater_rss only when about to *lower* total_vm or rss.  Any
  39          * collector of these hiwater stats must therefore get total_vm
  40          * and rss too, which will usually be the higher.  Barriers? not
  41          * worth the effort, such snapshots can always be inconsistent.
  42          */
  43         hiwater_vm = total_vm = mm->total_vm;
  44         if (hiwater_vm < mm->hiwater_vm)
  45                 hiwater_vm = mm->hiwater_vm;
  46         hiwater_rss = total_rss = anon + file + shmem;
  47         if (hiwater_rss < mm->hiwater_rss)
  48                 hiwater_rss = mm->hiwater_rss;
  49
  50         text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
  51         lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
  52         swap = get_mm_counter(mm, MM_SWAPENTS);
  53         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                 "VmSwap:\t%8lu kB\n",
  69                 hiwater_vm << (PAGE_SHIFT-10),
  70                 total_vm << (PAGE_SHIFT-10),
  71                 mm->locked_vm << (PAGE_SHIFT-10),
  72                 mm->pinned_vm << (PAGE_SHIFT-10),
  73                 hiwater_rss << (PAGE_SHIFT-10),
  74                 total_rss << (PAGE_SHIFT-10),
  75                 anon << (PAGE_SHIFT-10),
  76                 file << (PAGE_SHIFT-10),
  77                 shmem << (PAGE_SHIFT-10),
  78                 mm->data_vm << (PAGE_SHIFT-10),
  79                 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
  80                 mm_pgtables_bytes(mm) >> 10,
  81                 swap << (PAGE_SHIFT-10));
  82         hugetlb_report_usage(m, mm);
  83 }
  84
  85 unsigned long task_vsize(struct mm_struct *mm)
  86 {
  87         return PAGE_SIZE * mm->total_vm;
  88 }
  89
  90 unsigned long task_statm(struct mm_struct *mm,
  91                          unsigned long *shared, unsigned long *text,
  92                          unsigned long *data, unsigned long *resident)
  93 {
  94         *shared = get_mm_counter(mm, MM_FILEPAGES) +
  95                         get_mm_counter(mm, MM_SHMEMPAGES);
  96         *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
  97                                                                 >> PAGE_SHIFT;
  98         *data = mm->data_vm + mm->stack_vm;
  99         *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
 100         return mm->total_vm;
 101 }
 102
 103 #ifdef CONFIG_NUMA
 104 /*
 105  * Save get_task_policy() for show_numa_map().
 106  */
 107 static void hold_task_mempolicy(struct proc_maps_private *priv)
 108 {
 109         struct task_struct *task = priv->task;
 110
 111         task_lock(task);
 112         priv->task_mempolicy = get_task_policy(task);
 113         mpol_get(priv->task_mempolicy);
 114         task_unlock(task);
 115 }
 116 static void release_task_mempolicy(struct proc_maps_private *priv)
 117 {
 118         mpol_put(priv->task_mempolicy);
 119 }
 120 #else
 121 static void hold_task_mempolicy(struct proc_maps_private *priv)
 122 {
 123 }
 124 static void release_task_mempolicy(struct proc_maps_private *priv)
 125 {
 126 }
 127 #endif
 128
 129 static void vma_stop(struct proc_maps_private *priv)
 130 {
 131         struct mm_struct *mm = priv->mm;
 132
 133         release_task_mempolicy(priv);
 134         up_read(&mm->mmap_sem);
 135         mmput(mm);
 136 }
 137
 138 static struct vm_area_struct *
 139 m_next_vma(struct proc_maps_private *priv, struct vm_area_struct *vma)
 140 {
 141         if (vma == priv->tail_vma)
 142                 return NULL;
 143         return vma->vm_next ?: priv->tail_vma;
 144 }
 145
 146 static void m_cache_vma(struct seq_file *m, struct vm_area_struct *vma)
 147 {
 148         if (m->count < m->size) /* vma is copied successfully */
 149                 m->version = m_next_vma(m->private, vma) ? vma->vm_end : -1UL;
 150 }
 151
 152 static void *m_start(struct seq_file *m, loff_t *ppos)
 153 {
 154         struct proc_maps_private *priv = m->private;
 155         unsigned long last_addr = m->version;
 156         struct mm_struct *mm;
 157         struct vm_area_struct *vma;
 158         unsigned int pos = *ppos;
 159
 160         /* See m_cache_vma(). Zero at the start or after lseek. */
 161         if (last_addr == -1UL)
 162                 return NULL;
 163
 164         priv->task = get_proc_task(priv->inode);
 165         if (!priv->task)
 166                 return ERR_PTR(-ESRCH);
 167
 168         mm = priv->mm;
 169         if (!mm || !mmget_not_zero(mm))
 170                 return NULL;
 171
 172         down_read(&mm->mmap_sem);
 173         hold_task_mempolicy(priv);
 174         priv->tail_vma = get_gate_vma(mm);
 175
 176         if (last_addr) {
 177                 vma = find_vma(mm, last_addr - 1);
 178                 if (vma && vma->vm_start <= last_addr)
 179                         vma = m_next_vma(priv, vma);
 180                 if (vma)
 181                         return vma;
 182         }
 183
 184         m->version = 0;
 185         if (pos < mm->map_count) {
 186                 for (vma = mm->mmap; pos; pos--) {
 187                         m->version = vma->vm_start;
 188                         vma = vma->vm_next;
 189                 }
 190                 return vma;
 191         }
 192
 193         /* we do not bother to update m->version in this case */
 194         if (pos == mm->map_count && priv->tail_vma)
 195                 return priv->tail_vma;
 196
 197         vma_stop(priv);
 198         return NULL;
 199 }
 200
 201 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
 202 {
 203         struct proc_maps_private *priv = m->private;
 204         struct vm_area_struct *next;
 205
 206         (*pos)++;
 207         next = m_next_vma(priv, v);
 208         if (!next)
 209                 vma_stop(priv);
 210         return next;
 211 }
 212
 213 static void m_stop(struct seq_file *m, void *v)
 214 {
 215         struct proc_maps_private *priv = m->private;
 216
 217         if (!IS_ERR_OR_NULL(v))
 218                 vma_stop(priv);
 219         if (priv->task) {
 220                 put_task_struct(priv->task);
 221                 priv->task = NULL;
 222         }
 223 }
 224
 225 static int proc_maps_open(struct inode *inode, struct file *file,
 226                         const struct seq_operations *ops, int psize)
 227 {
 228         struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
 229
 230         if (!priv)
 231                 return -ENOMEM;
 232
 233         priv->inode = inode;
 234         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
 235         if (IS_ERR(priv->mm)) {
 236                 int err = PTR_ERR(priv->mm);
 237
 238                 seq_release_private(inode, file);
 239                 return err;
 240         }
 241
 242         return 0;
 243 }
 244
 245 static int proc_map_release(struct inode *inode, struct file *file)
 246 {
 247         struct seq_file *seq = file->private_data;
 248         struct proc_maps_private *priv = seq->private;
 249
 250         if (priv->mm)
 251                 mmdrop(priv->mm);
 252
 253         kfree(priv->rollup);
 254         return seq_release_private(inode, file);
 255 }
 256
 257 static int do_maps_open(struct inode *inode, struct file *file,
 258                         const struct seq_operations *ops)
 259 {
 260         return proc_maps_open(inode, file, ops,
 261                                 sizeof(struct proc_maps_private));
 262 }
 263
 264 /*
 265  * Indicate if the VMA is a stack for the given task; for
 266  * /proc/PID/maps that is the stack of the main task.
 267  */
 268 static int is_stack(struct vm_area_struct *vma)
 269 {
 270         /*
 271          * We make no effort to guess what a given thread considers to be
 272          * its "stack".  It's not even well-defined for programs written
 273          * languages like Go.
 274          */
 275         return vma->vm_start <= vma->vm_mm->start_stack &&
 276                 vma->vm_end >= vma->vm_mm->start_stack;
 277 }
 278
 279 static void show_vma_header_prefix(struct seq_file *m,
 280                                    unsigned long start, unsigned long end,
 281                                    vm_flags_t flags, unsigned long long pgoff,
 282                                    dev_t dev, unsigned long ino)
 283 {
 284         seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
 285         seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu ",
 286                    start,
 287                    end,
 288                    flags & VM_READ ? 'r' : '-',
 289                    flags & VM_WRITE ? 'w' : '-',
 290                    flags & VM_EXEC ? 'x' : '-',
 291                    flags & VM_MAYSHARE ? 's' : 'p',
 292                    pgoff,
 293                    MAJOR(dev), MINOR(dev), ino);
 294 }
 295
 296 static void
 297 show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
 298 {
 299         struct mm_struct *mm = vma->vm_mm;
 300         struct file *file = vma->vm_file;
 301         vm_flags_t flags = vma->vm_flags;
 302         unsigned long ino = 0;
 303         unsigned long long pgoff = 0;
 304         unsigned long start, end;
 305         dev_t dev = 0;
 306         const char *name = NULL;
 307
 308         if (file) {
 309                 struct inode *inode = file_inode(vma->vm_file);
 310                 dev = inode->i_sb->s_dev;
 311                 ino = inode->i_ino;
 312                 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
 313         }
 314
 315         start = vma->vm_start;
 316         end = vma->vm_end;
 317         show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino);
 318
 319         /*
 320          * Print the dentry name for named mappings, and a
 321          * special [heap] marker for the heap:
 322          */
 323         if (file) {
 324                 seq_pad(m, ' ');
 325                 seq_file_path(m, file, "\n");
 326                 goto done;
 327         }
 328
 329         if (vma->vm_ops && vma->vm_ops->name) {
 330                 name = vma->vm_ops->name(vma);
 331                 if (name)
 332                         goto done;
 333         }
 334
 335         name = arch_vma_name(vma);
 336         if (!name) {
 337                 if (!mm) {
 338                         name = "[vdso]";
 339                         goto done;
 340                 }
 341
 342                 if (vma->vm_start <= mm->brk &&
 343                     vma->vm_end >= mm->start_brk) {
 344                         name = "[heap]";
 345                         goto done;
 346                 }
 347
 348                 if (is_stack(vma))
 349                         name = "[stack]";
 350         }
 351
 352 done:
 353         if (name) {
 354                 seq_pad(m, ' ');
 355                 seq_puts(m, name);
 356         }
 357         seq_putc(m, '\n');
 358 }
 359
 360 static int show_map(struct seq_file *m, void *v, int is_pid)
 361 {
 362         show_map_vma(m, v, is_pid);
 363         m_cache_vma(m, v);
 364         return 0;
 365 }
 366
 367 static int show_pid_map(struct seq_file *m, void *v)
 368 {
 369         return show_map(m, v, 1);
 370 }
 371
 372 static int show_tid_map(struct seq_file *m, void *v)
 373 {
 374         return show_map(m, v, 0);
 375 }
 376
 377 static const struct seq_operations proc_pid_maps_op = {
 378         .start  = m_start,
 379         .next   = m_next,
 380         .stop   = m_stop,
 381         .show   = show_pid_map
 382 };
 383
 384 static const struct seq_operations proc_tid_maps_op = {
 385         .start  = m_start,
 386         .next   = m_next,
 387         .stop   = m_stop,
 388         .show   = show_tid_map
 389 };
 390
 391 static int pid_maps_open(struct inode *inode, struct file *file)
 392 {
 393         return do_maps_open(inode, file, &proc_pid_maps_op);
 394 }
 395
 396 static int tid_maps_open(struct inode *inode, struct file *file)
 397 {
 398         return do_maps_open(inode, file, &proc_tid_maps_op);
 399 }
 400
 401 const struct file_operations proc_pid_maps_operations = {
 402         .open           = pid_maps_open,
 403         .read           = seq_read,
 404         .llseek         = seq_lseek,
 405         .release        = proc_map_release,
 406 };
 407
 408 const struct file_operations proc_tid_maps_operations = {
 409         .open           = tid_maps_open,
 410         .read           = seq_read,
 411         .llseek         = seq_lseek,
 412         .release        = proc_map_release,
 413 };
 414
 415 /*
 416  * Proportional Set Size(PSS): my share of RSS.
 417  *
 418  * PSS of a process is the count of pages it has in memory, where each
 419  * page is divided by the number of processes sharing it.  So if a
 420  * process has 1000 pages all to itself, and 1000 shared with one other
 421  * process, its PSS will be 1500.
 422  *
 423  * To keep (accumulated) division errors low, we adopt a 64bit
 424  * fixed-point pss counter to minimize division errors. So (pss >>
 425  * PSS_SHIFT) would be the real byte count.
 426  *
 427  * A shift of 12 before division means (assuming 4K page size):
 428  *      - 1M 3-user-pages add up to 8KB errors;
 429  *      - supports mapcount up to 2^24, or 16M;
 430  *      - supports PSS up to 2^52 bytes, or 4PB.
 431  */
 432 #define PSS_SHIFT 12
 433
 434 #ifdef CONFIG_PROC_PAGE_MONITOR
 435 struct mem_size_stats {
 436         bool first;
 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         unsigned long first_vma_start;
 451         u64 pss;
 452         u64 pss_locked;
 453         u64 swap_pss;
 454         bool check_shmem_swap;
 455 };
 456
 457 static void smaps_account(struct mem_size_stats *mss, struct page *page,
 458                 bool compound, bool young, bool dirty)
 459 {
 460         int i, nr = compound ? 1 << compound_order(page) : 1;
 461         unsigned long size = nr * PAGE_SIZE;
 462
 463         if (PageAnon(page)) {
 464                 mss->anonymous += size;
 465                 if (!PageSwapBacked(page) && !dirty && !PageDirty(page))
 466                         mss->lazyfree += size;
 467         }
 468
 469         mss->resident += size;
 470         /* Accumulate the size in pages that have been accessed. */
 471         if (young || page_is_young(page) || PageReferenced(page))
 472                 mss->referenced += size;
 473
 474         /*
 475          * page_count(page) == 1 guarantees the page is mapped exactly once.
 476          * If any subpage of the compound page mapped with PTE it would elevate
 477          * page_count().
 478          */
 479         if (page_count(page) == 1) {
 480                 if (dirty || PageDirty(page))
 481                         mss->private_dirty += size;
 482                 else
 483                         mss->private_clean += size;
 484                 mss->pss += (u64)size << PSS_SHIFT;
 485                 return;
 486         }
 487
 488         for (i = 0; i < nr; i++, page++) {
 489                 int mapcount = page_mapcount(page);
 490
 491                 if (mapcount >= 2) {
 492                         if (dirty || PageDirty(page))
 493                                 mss->shared_dirty += PAGE_SIZE;
 494                         else
 495                                 mss->shared_clean += PAGE_SIZE;
 496                         mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount;
 497                 } else {
 498                         if (dirty || PageDirty(page))
 499                                 mss->private_dirty += PAGE_SIZE;
 500                         else
 501                                 mss->private_clean += PAGE_SIZE;
 502                         mss->pss += PAGE_SIZE << PSS_SHIFT;
 503                 }
 504         }
 505 }
 506
 507 #ifdef CONFIG_SHMEM
 508 static int smaps_pte_hole(unsigned long addr, unsigned long end,
 509                 struct mm_walk *walk)
 510 {
 511         struct mem_size_stats *mss = walk->private;
 512
 513         mss->swap += shmem_partial_swap_usage(
 514                         walk->vma->vm_file->f_mapping, addr, end);
 515
 516         return 0;
 517 }
 518 #endif
 519
 520 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
 521                 struct mm_walk *walk)
 522 {
 523         struct mem_size_stats *mss = walk->private;
 524         struct vm_area_struct *vma = walk->vma;
 525         struct page *page = NULL;
 526
 527         if (pte_present(*pte)) {
 528                 page = vm_normal_page(vma, addr, *pte);
 529         } else if (is_swap_pte(*pte)) {
 530                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 531
 532                 if (!non_swap_entry(swpent)) {
 533                         int mapcount;
 534
 535                         mss->swap += PAGE_SIZE;
 536                         mapcount = swp_swapcount(swpent);
 537                         if (mapcount >= 2) {
 538                                 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
 539
 540                                 do_div(pss_delta, mapcount);
 541                                 mss->swap_pss += pss_delta;
 542                         } else {
 543                                 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
 544                         }
 545                 } else if (is_migration_entry(swpent))
 546                         page = migration_entry_to_page(swpent);
 547                 else if (is_device_private_entry(swpent))
 548                         page = device_private_entry_to_page(swpent);
 549         } else if (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap
 550                                                         && pte_none(*pte))) {
 551                 page = find_get_entry(vma->vm_file->f_mapping,
 552                                                 linear_page_index(vma, addr));
 553                 if (!page)
 554                         return;
 555
 556                 if (radix_tree_exceptional_entry(page))
 557                         mss->swap += PAGE_SIZE;
 558                 else
 559                         put_page(page);
 560
 561                 return;
 562         }
 563
 564         if (!page)
 565                 return;
 566
 567         smaps_account(mss, page, false, pte_young(*pte), pte_dirty(*pte));
 568 }
 569
 570 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 571 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 572                 struct mm_walk *walk)
 573 {
 574         struct mem_size_stats *mss = walk->private;
 575         struct vm_area_struct *vma = walk->vma;
 576         struct page *page;
 577
 578         /* FOLL_DUMP will return -EFAULT on huge zero page */
 579         page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
 580         if (IS_ERR_OR_NULL(page))
 581                 return;
 582         if (PageAnon(page))
 583                 mss->anonymous_thp += HPAGE_PMD_SIZE;
 584         else if (PageSwapBacked(page))
 585                 mss->shmem_thp += HPAGE_PMD_SIZE;
 586         else if (is_zone_device_page(page))
 587                 /* pass */;
 588         else
 589                 VM_BUG_ON_PAGE(1, page);
 590         smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd));
 591 }
 592 #else
 593 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 594                 struct mm_walk *walk)
 595 {
 596 }
 597 #endif
 598
 599 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 600                            struct mm_walk *walk)
 601 {
 602         struct vm_area_struct *vma = walk->vma;
 603         pte_t *pte;
 604         spinlock_t *ptl;
 605
 606         ptl = pmd_trans_huge_lock(pmd, vma);
 607         if (ptl) {
 608                 if (pmd_present(*pmd))
 609                         smaps_pmd_entry(pmd, addr, walk);
 610                 spin_unlock(ptl);
 611                 goto out;
 612         }
 613
 614         if (pmd_trans_unstable(pmd))
 615                 goto out;
 616         /*
 617          * The mmap_sem held all the way back in m_start() is what
 618          * keeps khugepaged out of here and from collapsing things
 619          * in here.
 620          */
 621         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 622         for (; addr != end; pte++, addr += PAGE_SIZE)
 623                 smaps_pte_entry(pte, addr, walk);
 624         pte_unmap_unlock(pte - 1, ptl);
 625 out:
 626         cond_resched();
 627         return 0;
 628 }
 629
 630 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
 631 {
 632         /*
 633          * Don't forget to update Documentation/ on changes.
 634          */
 635         static const char mnemonics[BITS_PER_LONG][2] = {
 636                 /*
 637                  * In case if we meet a flag we don't know about.
 638                  */
 639                 [0 ... (BITS_PER_LONG-1)] = "??",
 640
 641                 [ilog2(VM_READ)]        = "rd",
 642                 [ilog2(VM_WRITE)]       = "wr",
 643                 [ilog2(VM_EXEC)]        = "ex",
 644                 [ilog2(VM_SHARED)]      = "sh",
 645                 [ilog2(VM_MAYREAD)]     = "mr",
 646                 [ilog2(VM_MAYWRITE)]    = "mw",
 647                 [ilog2(VM_MAYEXEC)]     = "me",
 648                 [ilog2(VM_MAYSHARE)]    = "ms",
 649                 [ilog2(VM_GROWSDOWN)]   = "gd",
 650                 [ilog2(VM_PFNMAP)]      = "pf",
 651                 [ilog2(VM_DENYWRITE)]   = "dw",
 652 #ifdef CONFIG_X86_INTEL_MPX
 653                 [ilog2(VM_MPX)]         = "mp",
 654 #endif
 655                 [ilog2(VM_LOCKED)]      = "lo",
 656                 [ilog2(VM_IO)]          = "io",
 657                 [ilog2(VM_SEQ_READ)]    = "sr",
 658                 [ilog2(VM_RAND_READ)]   = "rr",
 659                 [ilog2(VM_DONTCOPY)]    = "dc",
 660                 [ilog2(VM_DONTEXPAND)]  = "de",
 661                 [ilog2(VM_ACCOUNT)]     = "ac",
 662                 [ilog2(VM_NORESERVE)]   = "nr",
 663                 [ilog2(VM_HUGETLB)]     = "ht",
 664                 [ilog2(VM_SYNC)]        = "sf",
 665                 [ilog2(VM_ARCH_1)]      = "ar",
 666                 [ilog2(VM_WIPEONFORK)]  = "wf",
 667                 [ilog2(VM_DONTDUMP)]    = "dd",
 668 #ifdef CONFIG_MEM_SOFT_DIRTY
 669                 [ilog2(VM_SOFTDIRTY)]   = "sd",
 670 #endif
 671                 [ilog2(VM_MIXEDMAP)]    = "mm",
 672                 [ilog2(VM_HUGEPAGE)]    = "hg",
 673                 [ilog2(VM_NOHUGEPAGE)]  = "nh",
 674                 [ilog2(VM_MERGEABLE)]   = "mg",
 675                 [ilog2(VM_UFFD_MISSING)]= "um",
 676                 [ilog2(VM_UFFD_WP)]     = "uw",
 677 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
 678                 /* These come out via ProtectionKey: */
 679                 [ilog2(VM_PKEY_BIT0)]   = "",
 680                 [ilog2(VM_PKEY_BIT1)]   = "",
 681                 [ilog2(VM_PKEY_BIT2)]   = "",
 682                 [ilog2(VM_PKEY_BIT3)]   = "",
 683 #endif
 684         };
 685         size_t i;
 686
 687         seq_puts(m, "VmFlags: ");
 688         for (i = 0; i < BITS_PER_LONG; i++) {
 689                 if (!mnemonics[i][0])
 690                         continue;
 691                 if (vma->vm_flags & (1UL << i)) {
 692                         seq_printf(m, "%c%c ",
 693                                    mnemonics[i][0], mnemonics[i][1]);
 694                 }
 695         }
 696         seq_putc(m, '\n');
 697 }
 698
 699 #ifdef CONFIG_HUGETLB_PAGE
 700 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
 701                                  unsigned long addr, unsigned long end,
 702                                  struct mm_walk *walk)
 703 {
 704         struct mem_size_stats *mss = walk->private;
 705         struct vm_area_struct *vma = walk->vma;
 706         struct page *page = NULL;
 707
 708         if (pte_present(*pte)) {
 709                 page = vm_normal_page(vma, addr, *pte);
 710         } else if (is_swap_pte(*pte)) {
 711                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 712
 713                 if (is_migration_entry(swpent))
 714                         page = migration_entry_to_page(swpent);
 715                 else if (is_device_private_entry(swpent))
 716                         page = device_private_entry_to_page(swpent);
 717         }
 718         if (page) {
 719                 int mapcount = page_mapcount(page);
 720
 721                 if (mapcount >= 2)
 722                         mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
 723                 else
 724                         mss->private_hugetlb += huge_page_size(hstate_vma(vma));
 725         }
 726         return 0;
 727 }
 728 #endif /* HUGETLB_PAGE */
 729
 730 void __weak arch_show_smap(struct seq_file *m, struct vm_area_struct *vma)
 731 {
 732 }
 733
 734 static int show_smap(struct seq_file *m, void *v, int is_pid)
 735 {
 736         struct proc_maps_private *priv = m->private;
 737         struct vm_area_struct *vma = v;
 738         struct mem_size_stats mss_stack;
 739         struct mem_size_stats *mss;
 740         struct mm_walk smaps_walk = {
 741                 .pmd_entry = smaps_pte_range,
 742 #ifdef CONFIG_HUGETLB_PAGE
 743                 .hugetlb_entry = smaps_hugetlb_range,
 744 #endif
 745                 .mm = vma->vm_mm,
 746         };
 747         int ret = 0;
 748         bool rollup_mode;
 749         bool last_vma;
 750
 751         if (priv->rollup) {
 752                 rollup_mode = true;
 753                 mss = priv->rollup;
 754                 if (mss->first) {
 755                         mss->first_vma_start = vma->vm_start;
 756                         mss->first = false;
 757                 }
 758                 last_vma = !m_next_vma(priv, vma);
 759         } else {
 760                 rollup_mode = false;
 761                 memset(&mss_stack, 0, sizeof(mss_stack));
 762                 mss = &mss_stack;
 763         }
 764
 765         smaps_walk.private = mss;
 766
 767 #ifdef CONFIG_SHMEM
 768         if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
 769                 /*
 770                  * For shared or readonly shmem mappings we know that all
 771                  * swapped out pages belong to the shmem object, and we can
 772                  * obtain the swap value much more efficiently. For private
 773                  * writable mappings, we might have COW pages that are
 774                  * not affected by the parent swapped out pages of the shmem
 775                  * object, so we have to distinguish them during the page walk.
 776                  * Unless we know that the shmem object (or the part mapped by
 777                  * our VMA) has no swapped out pages at all.
 778                  */
 779                 unsigned long shmem_swapped = shmem_swap_usage(vma);
 780
 781                 if (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
 782                                         !(vma->vm_flags & VM_WRITE)) {
 783                         mss->swap = shmem_swapped;
 784                 } else {
 785                         mss->check_shmem_swap = true;
 786                         smaps_walk.pte_hole = smaps_pte_hole;
 787                 }
 788         }
 789 #endif
 790
 791         /* mmap_sem is held in m_start */
 792         walk_page_vma(vma, &smaps_walk);
 793         if (vma->vm_flags & VM_LOCKED)
 794                 mss->pss_locked += mss->pss;
 795
 796         if (!rollup_mode) {
 797                 show_map_vma(m, vma, is_pid);
 798         } else if (last_vma) {
 799                 show_vma_header_prefix(
 800                         m, mss->first_vma_start, vma->vm_end, 0, 0, 0, 0);
 801                 seq_pad(m, ' ');
 802                 seq_puts(m, "[rollup]\n");
 803         } else {
 804                 ret = SEQ_SKIP;
 805         }
 806
 807         if (!rollup_mode)
 808                 seq_printf(m,
 809                            "Size:           %8lu kB\n"
 810                            "KernelPageSize: %8lu kB\n"
 811                            "MMUPageSize:    %8lu kB\n",
 812                            (vma->vm_end - vma->vm_start) >> 10,
 813                            vma_kernel_pagesize(vma) >> 10,
 814                            vma_mmu_pagesize(vma) >> 10);
 815
 816
 817         if (!rollup_mode || last_vma)
 818                 seq_printf(m,
 819                            "Rss:            %8lu kB\n"
 820                            "Pss:            %8lu kB\n"
 821                            "Shared_Clean:   %8lu kB\n"
 822                            "Shared_Dirty:   %8lu kB\n"
 823                            "Private_Clean:  %8lu kB\n"
 824                            "Private_Dirty:  %8lu kB\n"
 825                            "Referenced:     %8lu kB\n"
 826                            "Anonymous:      %8lu kB\n"
 827                            "LazyFree:       %8lu kB\n"
 828                            "AnonHugePages:  %8lu kB\n"
 829                            "ShmemPmdMapped: %8lu kB\n"
 830                            "Shared_Hugetlb: %8lu kB\n"
 831                            "Private_Hugetlb: %7lu kB\n"
 832                            "Swap:           %8lu kB\n"
 833                            "SwapPss:        %8lu kB\n"
 834                            "Locked:         %8lu kB\n",
 835                            mss->resident >> 10,
 836                            (unsigned long)(mss->pss >> (10 + PSS_SHIFT)),
 837                            mss->shared_clean  >> 10,
 838                            mss->shared_dirty  >> 10,
 839                            mss->private_clean >> 10,
 840                            mss->private_dirty >> 10,
 841                            mss->referenced >> 10,
 842                            mss->anonymous >> 10,
 843                            mss->lazyfree >> 10,
 844                            mss->anonymous_thp >> 10,
 845                            mss->shmem_thp >> 10,
 846                            mss->shared_hugetlb >> 10,
 847                            mss->private_hugetlb >> 10,
 848                            mss->swap >> 10,
 849                            (unsigned long)(mss->swap_pss >> (10 + PSS_SHIFT)),
 850                            (unsigned long)(mss->pss >> (10 + PSS_SHIFT)));
 851
 852         if (!rollup_mode) {
 853                 arch_show_smap(m, vma);
 854                 show_smap_vma_flags(m, vma);
 855         }
 856         m_cache_vma(m, vma);
 857         return ret;
 858 }
 859
 860 static int show_pid_smap(struct seq_file *m, void *v)
 861 {
 862         return show_smap(m, v, 1);
 863 }
 864
 865 static int show_tid_smap(struct seq_file *m, void *v)
 866 {
 867         return show_smap(m, v, 0);
 868 }
 869
 870 static const struct seq_operations proc_pid_smaps_op = {
 871         .start  = m_start,
 872         .next   = m_next,
 873         .stop   = m_stop,
 874         .show   = show_pid_smap
 875 };
 876
 877 static const struct seq_operations proc_tid_smaps_op = {
 878         .start  = m_start,
 879         .next   = m_next,
 880         .stop   = m_stop,
 881         .show   = show_tid_smap
 882 };
 883
 884 static int pid_smaps_open(struct inode *inode, struct file *file)
 885 {
 886         return do_maps_open(inode, file, &proc_pid_smaps_op);
 887 }
 888
 889 static int pid_smaps_rollup_open(struct inode *inode, struct file *file)
 890 {
 891         struct seq_file *seq;
 892         struct proc_maps_private *priv;
 893         int ret = do_maps_open(inode, file, &proc_pid_smaps_op);
 894
 895         if (ret < 0)
 896                 return ret;
 897         seq = file->private_data;
 898         priv = seq->private;
 899         priv->rollup = kzalloc(sizeof(*priv->rollup), GFP_KERNEL);
 900         if (!priv->rollup) {
 901                 proc_map_release(inode, file);
 902                 return -ENOMEM;
 903         }
 904         priv->rollup->first = true;
 905         return 0;
 906 }
 907
 908 static int tid_smaps_open(struct inode *inode, struct file *file)
 909 {
 910         return do_maps_open(inode, file, &proc_tid_smaps_op);
 911 }
 912
 913 const struct file_operations proc_pid_smaps_operations = {
 914         .open           = pid_smaps_open,
 915         .read           = seq_read,
 916         .llseek         = seq_lseek,
 917         .release        = proc_map_release,
 918 };
 919
 920 const struct file_operations proc_pid_smaps_rollup_operations = {
 921         .open           = pid_smaps_rollup_open,
 922         .read           = seq_read,
 923         .llseek         = seq_lseek,
 924         .release        = proc_map_release,
 925 };
 926
 927 const struct file_operations proc_tid_smaps_operations = {
 928         .open           = tid_smaps_open,
 929         .read           = seq_read,
 930         .llseek         = seq_lseek,
 931         .release        = proc_map_release,
 932 };
 933
 934 enum clear_refs_types {
 935         CLEAR_REFS_ALL = 1,
 936         CLEAR_REFS_ANON,
 937         CLEAR_REFS_MAPPED,
 938         CLEAR_REFS_SOFT_DIRTY,
 939         CLEAR_REFS_MM_HIWATER_RSS,
 940         CLEAR_REFS_LAST,
 941 };
 942
 943 struct clear_refs_private {
 944         enum clear_refs_types type;
 945 };
 946
 947 #ifdef CONFIG_MEM_SOFT_DIRTY
 948 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 949                 unsigned long addr, pte_t *pte)
 950 {
 951         /*
 952          * The soft-dirty tracker uses #PF-s to catch writes
 953          * to pages, so write-protect the pte as well. See the
 954          * Documentation/vm/soft-dirty.txt for full description
 955          * of how soft-dirty works.
 956          */
 957         pte_t ptent = *pte;
 958
 959         if (pte_present(ptent)) {
 960                 ptent = ptep_modify_prot_start(vma->vm_mm, addr, pte);
 961                 ptent = pte_wrprotect(ptent);
 962                 ptent = pte_clear_soft_dirty(ptent);
 963                 ptep_modify_prot_commit(vma->vm_mm, addr, pte, ptent);
 964         } else if (is_swap_pte(ptent)) {
 965                 ptent = pte_swp_clear_soft_dirty(ptent);
 966                 set_pte_at(vma->vm_mm, addr, pte, ptent);
 967         }
 968 }
 969 #else
 970 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 971                 unsigned long addr, pte_t *pte)
 972 {
 973 }
 974 #endif
 975
 976 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
 977 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
 978                 unsigned long addr, pmd_t *pmdp)
 979 {
 980         pmd_t pmd = *pmdp;
 981
 982         if (pmd_present(pmd)) {
 983                 /* See comment in change_huge_pmd() */
 984                 pmdp_invalidate(vma, addr, pmdp);
 985                 if (pmd_dirty(*pmdp))
 986                         pmd = pmd_mkdirty(pmd);
 987                 if (pmd_young(*pmdp))
 988                         pmd = pmd_mkyoung(pmd);
 989
 990                 pmd = pmd_wrprotect(pmd);
 991                 pmd = pmd_clear_soft_dirty(pmd);
 992
 993                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
 994         } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
 995                 pmd = pmd_swp_clear_soft_dirty(pmd);
 996                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
 997         }
 998 }
 999 #else
1000 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1001                 unsigned long addr, pmd_t *pmdp)
1002 {
1003 }
1004 #endif
1005
1006 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
1007                                 unsigned long end, struct mm_walk *walk)
1008 {
1009         struct clear_refs_private *cp = walk->private;
1010         struct vm_area_struct *vma = walk->vma;
1011         pte_t *pte, ptent;
1012         spinlock_t *ptl;
1013         struct page *page;
1014
1015         ptl = pmd_trans_huge_lock(pmd, vma);
1016         if (ptl) {
1017                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1018                         clear_soft_dirty_pmd(vma, addr, pmd);
1019                         goto out;
1020                 }
1021
1022                 if (!pmd_present(*pmd))
1023                         goto out;
1024
1025                 page = pmd_page(*pmd);
1026
1027                 /* Clear accessed and referenced bits. */
1028                 pmdp_test_and_clear_young(vma, addr, pmd);
1029                 test_and_clear_page_young(page);
1030                 ClearPageReferenced(page);
1031 out:
1032                 spin_unlock(ptl);
1033                 return 0;
1034         }
1035
1036         if (pmd_trans_unstable(pmd))
1037                 return 0;
1038
1039         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1040         for (; addr != end; pte++, addr += PAGE_SIZE) {
1041                 ptent = *pte;
1042
1043                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1044                         clear_soft_dirty(vma, addr, pte);
1045                         continue;
1046                 }
1047
1048                 if (!pte_present(ptent))
1049                         continue;
1050
1051                 page = vm_normal_page(vma, addr, ptent);
1052                 if (!page)
1053                         continue;
1054
1055                 /* Clear accessed and referenced bits. */
1056                 ptep_test_and_clear_young(vma, addr, pte);
1057                 test_and_clear_page_young(page);
1058                 ClearPageReferenced(page);
1059         }
1060         pte_unmap_unlock(pte - 1, ptl);
1061         cond_resched();
1062         return 0;
1063 }
1064
1065 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1066                                 struct mm_walk *walk)
1067 {
1068         struct clear_refs_private *cp = walk->private;
1069         struct vm_area_struct *vma = walk->vma;
1070
1071         if (vma->vm_flags & VM_PFNMAP)
1072                 return 1;
1073
1074         /*
1075          * Writing 1 to /proc/pid/clear_refs affects all pages.
1076          * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1077          * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1078          * Writing 4 to /proc/pid/clear_refs affects all pages.
1079          */
1080         if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1081                 return 1;
1082         if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1083                 return 1;
1084         return 0;
1085 }
1086
1087 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1088                                 size_t count, loff_t *ppos)
1089 {
1090         struct task_struct *task;
1091         char buffer[PROC_NUMBUF];
1092         struct mm_struct *mm;
1093         struct vm_area_struct *vma;
1094         enum clear_refs_types type;
1095         struct mmu_gather tlb;
1096         int itype;
1097         int rv;
1098
1099         memset(buffer, 0, sizeof(buffer));
1100         if (count > sizeof(buffer) - 1)
1101                 count = sizeof(buffer) - 1;
1102         if (copy_from_user(buffer, buf, count))
1103                 return -EFAULT;
1104         rv = kstrtoint(strstrip(buffer), 10, &itype);
1105         if (rv < 0)
1106                 return rv;
1107         type = (enum clear_refs_types)itype;
1108         if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1109                 return -EINVAL;
1110
1111         task = get_proc_task(file_inode(file));
1112         if (!task)
1113                 return -ESRCH;
1114         mm = get_task_mm(task);
1115         if (mm) {
1116                 struct clear_refs_private cp = {
1117                         .type = type,
1118                 };
1119                 struct mm_walk clear_refs_walk = {
1120                         .pmd_entry = clear_refs_pte_range,
1121                         .test_walk = clear_refs_test_walk,
1122                         .mm = mm,
1123                         .private = &cp,
1124                 };
1125
1126                 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1127                         if (down_write_killable(&mm->mmap_sem)) {
1128                                 count = -EINTR;
1129                                 goto out_mm;
1130                         }
1131
1132                         /*
1133                          * Writing 5 to /proc/pid/clear_refs resets the peak
1134                          * resident set size to this mm's current rss value.
1135                          */
1136                         reset_mm_hiwater_rss(mm);
1137                         up_write(&mm->mmap_sem);
1138                         goto out_mm;
1139                 }
1140
1141                 down_read(&mm->mmap_sem);
1142                 tlb_gather_mmu(&tlb, mm, 0, -1);
1143                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1144                         for (vma = mm->mmap; vma; vma = vma->vm_next) {
1145                                 if (!(vma->vm_flags & VM_SOFTDIRTY))
1146                                         continue;
1147                                 up_read(&mm->mmap_sem);
1148                                 if (down_write_killable(&mm->mmap_sem)) {
1149                                         count = -EINTR;
1150                                         goto out_mm;
1151                                 }
1152                                 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1153                                         vma->vm_flags &= ~VM_SOFTDIRTY;
1154                                         vma_set_page_prot(vma);
1155                                 }
1156                                 downgrade_write(&mm->mmap_sem);
1157                                 break;
1158                         }
1159                         mmu_notifier_invalidate_range_start(mm, 0, -1);
1160                 }
1161                 walk_page_range(0, mm->highest_vm_end, &clear_refs_walk);
1162                 if (type == CLEAR_REFS_SOFT_DIRTY)
1163                         mmu_notifier_invalidate_range_end(mm, 0, -1);
1164                 tlb_finish_mmu(&tlb, 0, -1);
1165                 up_read(&mm->mmap_sem);
1166 out_mm:
1167                 mmput(mm);
1168         }
1169         put_task_struct(task);
1170
1171         return count;
1172 }
1173
1174 const struct file_operations proc_clear_refs_operations = {
1175         .write          = clear_refs_write,
1176         .llseek         = noop_llseek,
1177 };
1178
1179 typedef struct {
1180         u64 pme;
1181 } pagemap_entry_t;
1182
1183 struct pagemapread {
1184         int pos, len;           /* units: PM_ENTRY_BYTES, not bytes */
1185         pagemap_entry_t *buffer;
1186         bool show_pfn;
1187 };
1188
1189 #define PAGEMAP_WALK_SIZE       (PMD_SIZE)
1190 #define PAGEMAP_WALK_MASK       (PMD_MASK)
1191
1192 #define PM_ENTRY_BYTES          sizeof(pagemap_entry_t)
1193 #define PM_PFRAME_BITS          55
1194 #define PM_PFRAME_MASK          GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1195 #define PM_SOFT_DIRTY           BIT_ULL(55)
1196 #define PM_MMAP_EXCLUSIVE       BIT_ULL(56)
1197 #define PM_FILE                 BIT_ULL(61)
1198 #define PM_SWAP                 BIT_ULL(62)
1199 #define PM_PRESENT              BIT_ULL(63)
1200
1201 #define PM_END_OF_BUFFER    1
1202
1203 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1204 {
1205         return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1206 }
1207
1208 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1209                           struct pagemapread *pm)
1210 {
1211         pm->buffer[pm->pos++] = *pme;
1212         if (pm->pos >= pm->len)
1213                 return PM_END_OF_BUFFER;
1214         return 0;
1215 }
1216
1217 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1218                                 struct mm_walk *walk)
1219 {
1220         struct pagemapread *pm = walk->private;
1221         unsigned long addr = start;
1222         int err = 0;
1223
1224         while (addr < end) {
1225                 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1226                 pagemap_entry_t pme = make_pme(0, 0);
1227                 /* End of address space hole, which we mark as non-present. */
1228                 unsigned long hole_end;
1229
1230                 if (vma)
1231                         hole_end = min(end, vma->vm_start);
1232                 else
1233                         hole_end = end;
1234
1235                 for (; addr < hole_end; addr += PAGE_SIZE) {
1236                         err = add_to_pagemap(addr, &pme, pm);
1237                         if (err)
1238                                 goto out;
1239                 }
1240
1241                 if (!vma)
1242                         break;
1243
1244                 /* Addresses in the VMA. */
1245                 if (vma->vm_flags & VM_SOFTDIRTY)
1246                         pme = make_pme(0, PM_SOFT_DIRTY);
1247                 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1248                         err = add_to_pagemap(addr, &pme, pm);
1249                         if (err)
1250                                 goto out;
1251                 }
1252         }
1253 out:
1254         return err;
1255 }
1256
1257 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1258                 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1259 {
1260         u64 frame = 0, flags = 0;
1261         struct page *page = NULL;
1262
1263         if (pte_present(pte)) {
1264                 if (pm->show_pfn)
1265                         frame = pte_pfn(pte);
1266                 flags |= PM_PRESENT;
1267                 page = _vm_normal_page(vma, addr, pte, true);
1268                 if (pte_soft_dirty(pte))
1269                         flags |= PM_SOFT_DIRTY;
1270         } else if (is_swap_pte(pte)) {
1271                 swp_entry_t entry;
1272                 if (pte_swp_soft_dirty(pte))
1273                         flags |= PM_SOFT_DIRTY;
1274                 entry = pte_to_swp_entry(pte);
1275                 frame = swp_type(entry) |
1276                         (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1277                 flags |= PM_SWAP;
1278                 if (is_migration_entry(entry))
1279                         page = migration_entry_to_page(entry);
1280
1281                 if (is_device_private_entry(entry))
1282                         page = device_private_entry_to_page(entry);
1283         }
1284
1285         if (page && !PageAnon(page))
1286                 flags |= PM_FILE;
1287         if (page && page_mapcount(page) == 1)
1288                 flags |= PM_MMAP_EXCLUSIVE;
1289         if (vma->vm_flags & VM_SOFTDIRTY)
1290                 flags |= PM_SOFT_DIRTY;
1291
1292         return make_pme(frame, flags);
1293 }
1294
1295 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1296                              struct mm_walk *walk)
1297 {
1298         struct vm_area_struct *vma = walk->vma;
1299         struct pagemapread *pm = walk->private;
1300         spinlock_t *ptl;
1301         pte_t *pte, *orig_pte;
1302         int err = 0;
1303
1304 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1305         ptl = pmd_trans_huge_lock(pmdp, vma);
1306         if (ptl) {
1307                 u64 flags = 0, frame = 0;
1308                 pmd_t pmd = *pmdp;
1309                 struct page *page = NULL;
1310
1311                 if (vma->vm_flags & VM_SOFTDIRTY)
1312                         flags |= PM_SOFT_DIRTY;
1313
1314                 if (pmd_present(pmd)) {
1315                         page = pmd_page(pmd);
1316
1317                         flags |= PM_PRESENT;
1318                         if (pmd_soft_dirty(pmd))
1319                                 flags |= PM_SOFT_DIRTY;
1320                         if (pm->show_pfn)
1321                                 frame = pmd_pfn(pmd) +
1322                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1323                 }
1324 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1325                 else if (is_swap_pmd(pmd)) {
1326                         swp_entry_t entry = pmd_to_swp_entry(pmd);
1327
1328                         frame = swp_type(entry) |
1329                                 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1330                         flags |= PM_SWAP;
1331                         if (pmd_swp_soft_dirty(pmd))
1332                                 flags |= PM_SOFT_DIRTY;
1333                         VM_BUG_ON(!is_pmd_migration_entry(pmd));
1334                         page = migration_entry_to_page(entry);
1335                 }
1336 #endif
1337
1338                 if (page && page_mapcount(page) == 1)
1339                         flags |= PM_MMAP_EXCLUSIVE;
1340
1341                 for (; addr != end; addr += PAGE_SIZE) {
1342                         pagemap_entry_t pme = make_pme(frame, flags);
1343
1344                         err = add_to_pagemap(addr, &pme, pm);
1345                         if (err)
1346                                 break;
1347                         if (pm->show_pfn && (flags & PM_PRESENT))
1348                                 frame++;
1349                 }
1350                 spin_unlock(ptl);
1351                 return err;
1352         }
1353
1354         if (pmd_trans_unstable(pmdp))
1355                 return 0;
1356 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1357
1358         /*
1359          * We can assume that @vma always points to a valid one and @end never
1360          * goes beyond vma->vm_end.
1361          */
1362         orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1363         for (; addr < end; pte++, addr += PAGE_SIZE) {
1364                 pagemap_entry_t pme;
1365
1366                 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1367                 err = add_to_pagemap(addr, &pme, pm);
1368                 if (err)
1369                         break;
1370         }
1371         pte_unmap_unlock(orig_pte, ptl);
1372
1373         cond_resched();
1374
1375         return err;
1376 }
1377
1378 #ifdef CONFIG_HUGETLB_PAGE
1379 /* This function walks within one hugetlb entry in the single call */
1380 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1381                                  unsigned long addr, unsigned long end,
1382                                  struct mm_walk *walk)
1383 {
1384         struct pagemapread *pm = walk->private;
1385         struct vm_area_struct *vma = walk->vma;
1386         u64 flags = 0, frame = 0;
1387         int err = 0;
1388         pte_t pte;
1389
1390         if (vma->vm_flags & VM_SOFTDIRTY)
1391                 flags |= PM_SOFT_DIRTY;
1392
1393         pte = huge_ptep_get(ptep);
1394         if (pte_present(pte)) {
1395                 struct page *page = pte_page(pte);
1396
1397                 if (!PageAnon(page))
1398                         flags |= PM_FILE;
1399
1400                 if (page_mapcount(page) == 1)
1401                         flags |= PM_MMAP_EXCLUSIVE;
1402
1403                 flags |= PM_PRESENT;
1404                 if (pm->show_pfn)
1405                         frame = pte_pfn(pte) +
1406                                 ((addr & ~hmask) >> PAGE_SHIFT);
1407         }
1408
1409         for (; addr != end; addr += PAGE_SIZE) {
1410                 pagemap_entry_t pme = make_pme(frame, flags);
1411
1412                 err = add_to_pagemap(addr, &pme, pm);
1413                 if (err)
1414                         return err;
1415                 if (pm->show_pfn && (flags & PM_PRESENT))
1416                         frame++;
1417         }
1418
1419         cond_resched();
1420
1421         return err;
1422 }
1423 #endif /* HUGETLB_PAGE */
1424
1425 /*
1426  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1427  *
1428  * For each page in the address space, this file contains one 64-bit entry
1429  * consisting of the following:
1430  *
1431  * Bits 0-54  page frame number (PFN) if present
1432  * Bits 0-4   swap type if swapped
1433  * Bits 5-54  swap offset if swapped
1434  * Bit  55    pte is soft-dirty (see Documentation/vm/soft-dirty.txt)
1435  * Bit  56    page exclusively mapped
1436  * Bits 57-60 zero
1437  * Bit  61    page is file-page or shared-anon
1438  * Bit  62    page swapped
1439  * Bit  63    page present
1440  *
1441  * If the page is not present but in swap, then the PFN contains an
1442  * encoding of the swap file number and the page's offset into the
1443  * swap. Unmapped pages return a null PFN. This allows determining
1444  * precisely which pages are mapped (or in swap) and comparing mapped
1445  * pages between processes.
1446  *
1447  * Efficient users of this interface will use /proc/pid/maps to
1448  * determine which areas of memory are actually mapped and llseek to
1449  * skip over unmapped regions.
1450  */
1451 static ssize_t pagemap_read(struct file *file, char __user *buf,
1452                             size_t count, loff_t *ppos)
1453 {
1454         struct mm_struct *mm = file->private_data;
1455         struct pagemapread pm;
1456         struct mm_walk pagemap_walk = {};
1457         unsigned long src;
1458         unsigned long svpfn;
1459         unsigned long start_vaddr;
1460         unsigned long end_vaddr;
1461         int ret = 0, copied = 0;
1462
1463         if (!mm || !mmget_not_zero(mm))
1464                 goto out;
1465
1466         ret = -EINVAL;
1467         /* file position must be aligned */
1468         if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1469                 goto out_mm;
1470
1471         ret = 0;
1472         if (!count)
1473                 goto out_mm;
1474
1475         /* do not disclose physical addresses: attack vector */
1476         pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1477
1478         pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1479         pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_KERNEL);
1480         ret = -ENOMEM;
1481         if (!pm.buffer)
1482                 goto out_mm;
1483
1484         pagemap_walk.pmd_entry = pagemap_pmd_range;
1485         pagemap_walk.pte_hole = pagemap_pte_hole;
1486 #ifdef CONFIG_HUGETLB_PAGE
1487         pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1488 #endif
1489         pagemap_walk.mm = mm;
1490         pagemap_walk.private = &pm;
1491
1492         src = *ppos;
1493         svpfn = src / PM_ENTRY_BYTES;
1494         start_vaddr = svpfn << PAGE_SHIFT;
1495         end_vaddr = mm->task_size;
1496
1497         /* watch out for wraparound */
1498         if (svpfn > mm->task_size >> PAGE_SHIFT)
1499                 start_vaddr = end_vaddr;
1500
1501         /*
1502          * The odds are that this will stop walking way
1503          * before end_vaddr, because the length of the
1504          * user buffer is tracked in "pm", and the walk
1505          * will stop when we hit the end of the buffer.
1506          */
1507         ret = 0;
1508         while (count && (start_vaddr < end_vaddr)) {
1509                 int len;
1510                 unsigned long end;
1511
1512                 pm.pos = 0;
1513                 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1514                 /* overflow ? */
1515                 if (end < start_vaddr || end > end_vaddr)
1516                         end = end_vaddr;
1517                 down_read(&mm->mmap_sem);
1518                 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1519                 up_read(&mm->mmap_sem);
1520                 start_vaddr = end;
1521
1522                 len = min(count, PM_ENTRY_BYTES * pm.pos);
1523                 if (copy_to_user(buf, pm.buffer, len)) {
1524                         ret = -EFAULT;
1525                         goto out_free;
1526                 }
1527                 copied += len;
1528                 buf += len;
1529                 count -= len;
1530         }
1531         *ppos += copied;
1532         if (!ret || ret == PM_END_OF_BUFFER)
1533                 ret = copied;
1534
1535 out_free:
1536         kfree(pm.buffer);
1537 out_mm:
1538         mmput(mm);
1539 out:
1540         return ret;
1541 }
1542
1543 static int pagemap_open(struct inode *inode, struct file *file)
1544 {
1545         struct mm_struct *mm;
1546
1547         mm = proc_mem_open(inode, PTRACE_MODE_READ);
1548         if (IS_ERR(mm))
1549                 return PTR_ERR(mm);
1550         file->private_data = mm;
1551         return 0;
1552 }
1553
1554 static int pagemap_release(struct inode *inode, struct file *file)
1555 {
1556         struct mm_struct *mm = file->private_data;
1557
1558         if (mm)
1559                 mmdrop(mm);
1560         return 0;
1561 }
1562
1563 const struct file_operations proc_pagemap_operations = {
1564         .llseek         = mem_lseek, /* borrow this */
1565         .read           = pagemap_read,
1566         .open           = pagemap_open,
1567         .release        = pagemap_release,
1568 };
1569 #endif /* CONFIG_PROC_PAGE_MONITOR */
1570
1571 #ifdef CONFIG_NUMA
1572
1573 struct numa_maps {
1574         unsigned long pages;
1575         unsigned long anon;
1576         unsigned long active;
1577         unsigned long writeback;
1578         unsigned long mapcount_max;
1579         unsigned long dirty;
1580         unsigned long swapcache;
1581         unsigned long node[MAX_NUMNODES];
1582 };
1583
1584 struct numa_maps_private {
1585         struct proc_maps_private proc_maps;
1586         struct numa_maps md;
1587 };
1588
1589 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1590                         unsigned long nr_pages)
1591 {
1592         int count = page_mapcount(page);
1593
1594         md->pages += nr_pages;
1595         if (pte_dirty || PageDirty(page))
1596                 md->dirty += nr_pages;
1597
1598         if (PageSwapCache(page))
1599                 md->swapcache += nr_pages;
1600
1601         if (PageActive(page) || PageUnevictable(page))
1602                 md->active += nr_pages;
1603
1604         if (PageWriteback(page))
1605                 md->writeback += nr_pages;
1606
1607         if (PageAnon(page))
1608                 md->anon += nr_pages;
1609
1610         if (count > md->mapcount_max)
1611                 md->mapcount_max = count;
1612
1613         md->node[page_to_nid(page)] += nr_pages;
1614 }
1615
1616 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1617                 unsigned long addr)
1618 {
1619         struct page *page;
1620         int nid;
1621
1622         if (!pte_present(pte))
1623                 return NULL;
1624
1625         page = vm_normal_page(vma, addr, pte);
1626         if (!page)
1627                 return NULL;
1628
1629         if (PageReserved(page))
1630                 return NULL;
1631
1632         nid = page_to_nid(page);
1633         if (!node_isset(nid, node_states[N_MEMORY]))
1634                 return NULL;
1635
1636         return page;
1637 }
1638
1639 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1640 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1641                                               struct vm_area_struct *vma,
1642                                               unsigned long addr)
1643 {
1644         struct page *page;
1645         int nid;
1646
1647         if (!pmd_present(pmd))
1648                 return NULL;
1649
1650         page = vm_normal_page_pmd(vma, addr, pmd);
1651         if (!page)
1652                 return NULL;
1653
1654         if (PageReserved(page))
1655                 return NULL;
1656
1657         nid = page_to_nid(page);
1658         if (!node_isset(nid, node_states[N_MEMORY]))
1659                 return NULL;
1660
1661         return page;
1662 }
1663 #endif
1664
1665 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1666                 unsigned long end, struct mm_walk *walk)
1667 {
1668         struct numa_maps *md = walk->private;
1669         struct vm_area_struct *vma = walk->vma;
1670         spinlock_t *ptl;
1671         pte_t *orig_pte;
1672         pte_t *pte;
1673
1674 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1675         ptl = pmd_trans_huge_lock(pmd, vma);
1676         if (ptl) {
1677                 struct page *page;
1678
1679                 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1680                 if (page)
1681                         gather_stats(page, md, pmd_dirty(*pmd),
1682                                      HPAGE_PMD_SIZE/PAGE_SIZE);
1683                 spin_unlock(ptl);
1684                 return 0;
1685         }
1686
1687         if (pmd_trans_unstable(pmd))
1688                 return 0;
1689 #endif
1690         orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1691         do {
1692                 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1693                 if (!page)
1694                         continue;
1695                 gather_stats(page, md, pte_dirty(*pte), 1);
1696
1697         } while (pte++, addr += PAGE_SIZE, addr != end);
1698         pte_unmap_unlock(orig_pte, ptl);
1699         cond_resched();
1700         return 0;
1701 }
1702 #ifdef CONFIG_HUGETLB_PAGE
1703 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1704                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1705 {
1706         pte_t huge_pte = huge_ptep_get(pte);
1707         struct numa_maps *md;
1708         struct page *page;
1709
1710         if (!pte_present(huge_pte))
1711                 return 0;
1712
1713         page = pte_page(huge_pte);
1714         if (!page)
1715                 return 0;
1716
1717         md = walk->private;
1718         gather_stats(page, md, pte_dirty(huge_pte), 1);
1719         return 0;
1720 }
1721
1722 #else
1723 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1724                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1725 {
1726         return 0;
1727 }
1728 #endif
1729
1730 /*
1731  * Display pages allocated per node and memory policy via /proc.
1732  */
1733 static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1734 {
1735         struct numa_maps_private *numa_priv = m->private;
1736         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1737         struct vm_area_struct *vma = v;
1738         struct numa_maps *md = &numa_priv->md;
1739         struct file *file = vma->vm_file;
1740         struct mm_struct *mm = vma->vm_mm;
1741         struct mm_walk walk = {
1742                 .hugetlb_entry = gather_hugetlb_stats,
1743                 .pmd_entry = gather_pte_stats,
1744                 .private = md,
1745                 .mm = mm,
1746         };
1747         struct mempolicy *pol;
1748         char buffer[64];
1749         int nid;
1750
1751         if (!mm)
1752                 return 0;
1753
1754         /* Ensure we start with an empty set of numa_maps statistics. */
1755         memset(md, 0, sizeof(*md));
1756
1757         pol = __get_vma_policy(vma, vma->vm_start);
1758         if (pol) {
1759                 mpol_to_str(buffer, sizeof(buffer), pol);
1760                 mpol_cond_put(pol);
1761         } else {
1762                 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1763         }
1764
1765         seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1766
1767         if (file) {
1768                 seq_puts(m, " file=");
1769                 seq_file_path(m, file, "\n\t= ");
1770         } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1771                 seq_puts(m, " heap");
1772         } else if (is_stack(vma)) {
1773                 seq_puts(m, " stack");
1774         }
1775
1776         if (is_vm_hugetlb_page(vma))
1777                 seq_puts(m, " huge");
1778
1779         /* mmap_sem is held by m_start */
1780         walk_page_vma(vma, &walk);
1781
1782         if (!md->pages)
1783                 goto out;
1784
1785         if (md->anon)
1786                 seq_printf(m, " anon=%lu", md->anon);
1787
1788         if (md->dirty)
1789                 seq_printf(m, " dirty=%lu", md->dirty);
1790
1791         if (md->pages != md->anon && md->pages != md->dirty)
1792                 seq_printf(m, " mapped=%lu", md->pages);
1793
1794         if (md->mapcount_max > 1)
1795                 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1796
1797         if (md->swapcache)
1798                 seq_printf(m, " swapcache=%lu", md->swapcache);
1799
1800         if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1801                 seq_printf(m, " active=%lu", md->active);
1802
1803         if (md->writeback)
1804                 seq_printf(m, " writeback=%lu", md->writeback);
1805
1806         for_each_node_state(nid, N_MEMORY)
1807                 if (md->node[nid])
1808                         seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1809
1810         seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1811 out:
1812         seq_putc(m, '\n');
1813         m_cache_vma(m, vma);
1814         return 0;
1815 }
1816
1817 static int show_pid_numa_map(struct seq_file *m, void *v)
1818 {
1819         return show_numa_map(m, v, 1);
1820 }
1821
1822 static int show_tid_numa_map(struct seq_file *m, void *v)
1823 {
1824         return show_numa_map(m, v, 0);
1825 }
1826
1827 static const struct seq_operations proc_pid_numa_maps_op = {
1828         .start  = m_start,
1829         .next   = m_next,
1830         .stop   = m_stop,
1831         .show   = show_pid_numa_map,
1832 };
1833
1834 static const struct seq_operations proc_tid_numa_maps_op = {
1835         .start  = m_start,
1836         .next   = m_next,
1837         .stop   = m_stop,
1838         .show   = show_tid_numa_map,
1839 };
1840
1841 static int numa_maps_open(struct inode *inode, struct file *file,
1842                           const struct seq_operations *ops)
1843 {
1844         return proc_maps_open(inode, file, ops,
1845                                 sizeof(struct numa_maps_private));
1846 }
1847
1848 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1849 {
1850         return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1851 }
1852
1853 static int tid_numa_maps_open(struct inode *inode, struct file *file)
1854 {
1855         return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1856 }
1857
1858 const struct file_operations proc_pid_numa_maps_operations = {
1859         .open           = pid_numa_maps_open,
1860         .read           = seq_read,
1861         .llseek         = seq_lseek,
1862         .release        = proc_map_release,
1863 };
1864
1865 const struct file_operations proc_tid_numa_maps_operations = {
1866         .open           = tid_numa_maps_open,
1867         .read           = seq_read,
1868         .llseek         = seq_lseek,
1869         .release        = proc_map_release,
1870 };
1871 #endif /* CONFIG_NUMA */