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