fs/proc/task_mmu.c

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