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, 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, int is_pid)
 307 {
 308         int stack = 0;
 309
 310         if (is_pid) {
 311                 stack = vma->vm_start <= vma->vm_mm->start_stack &&
 312                         vma->vm_end >= vma->vm_mm->start_stack;
 313         } else {
 314                 struct inode *inode = priv->inode;
 315                 struct task_struct *task;
 316
 317                 rcu_read_lock();
 318                 task = pid_task(proc_pid(inode), PIDTYPE_PID);
 319                 if (task)
 320                         stack = vma_is_stack_for_task(vma, task);
 321                 rcu_read_unlock();
 322         }
 323         return stack;
 324 }
 325
 326 static void
 327 show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
 328 {
 329         struct mm_struct *mm = vma->vm_mm;
 330         struct file *file = vma->vm_file;
 331         struct proc_maps_private *priv = m->private;
 332         vm_flags_t flags = vma->vm_flags;
 333         unsigned long ino = 0;
 334         unsigned long long pgoff = 0;
 335         unsigned long start, end;
 336         dev_t dev = 0;
 337         const char *name = NULL;
 338
 339         if (file) {
 340                 struct inode *inode = file_inode(vma->vm_file);
 341                 dev = inode->i_sb->s_dev;
 342                 ino = inode->i_ino;
 343                 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
 344         }
 345
 346         /* We don't show the stack guard page in /proc/maps */
 347         start = vma->vm_start;
 348         if (stack_guard_page_start(vma, start))
 349                 start += PAGE_SIZE;
 350         end = vma->vm_end;
 351         if (stack_guard_page_end(vma, end))
 352                 end -= PAGE_SIZE;
 353
 354         seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
 355         seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu ",
 356                         start,
 357                         end,
 358                         flags & VM_READ ? 'r' : '-',
 359                         flags & VM_WRITE ? 'w' : '-',
 360                         flags & VM_EXEC ? 'x' : '-',
 361                         flags & VM_MAYSHARE ? 's' : 'p',
 362                         pgoff,
 363                         MAJOR(dev), MINOR(dev), ino);
 364
 365         /*
 366          * Print the dentry name for named mappings, and a
 367          * special [heap] marker for the heap:
 368          */
 369         if (file) {
 370                 seq_pad(m, ' ');
 371                 seq_file_path(m, file, "\n");
 372                 goto done;
 373         }
 374
 375         if (vma->vm_ops && vma->vm_ops->name) {
 376                 name = vma->vm_ops->name(vma);
 377                 if (name)
 378                         goto done;
 379         }
 380
 381         name = arch_vma_name(vma);
 382         if (!name) {
 383                 if (!mm) {
 384                         name = "[vdso]";
 385                         goto done;
 386                 }
 387
 388                 if (vma->vm_start <= mm->brk &&
 389                     vma->vm_end >= mm->start_brk) {
 390                         name = "[heap]";
 391                         goto done;
 392                 }
 393
 394                 if (is_stack(priv, vma, is_pid)) {
 395                         name = "[stack]";
 396                         goto done;
 397                 }
 398                 if (vma_get_anon_name(vma)) {
 399                         seq_pad(m, ' ');
 400                         seq_print_vma_name(m, vma);
 401                 }
 402         }
 403
 404 done:
 405         if (name) {
 406                 seq_pad(m, ' ');
 407                 seq_puts(m, name);
 408         }
 409         seq_putc(m, '\n');
 410 }
 411
 412 static int show_map(struct seq_file *m, void *v, int is_pid)
 413 {
 414         show_map_vma(m, v, is_pid);
 415         m_cache_vma(m, v);
 416         return 0;
 417 }
 418
 419 static int show_pid_map(struct seq_file *m, void *v)
 420 {
 421         return show_map(m, v, 1);
 422 }
 423
 424 static int show_tid_map(struct seq_file *m, void *v)
 425 {
 426         return show_map(m, v, 0);
 427 }
 428
 429 static const struct seq_operations proc_pid_maps_op = {
 430         .start  = m_start,
 431         .next   = m_next,
 432         .stop   = m_stop,
 433         .show   = show_pid_map
 434 };
 435
 436 static const struct seq_operations proc_tid_maps_op = {
 437         .start  = m_start,
 438         .next   = m_next,
 439         .stop   = m_stop,
 440         .show   = show_tid_map
 441 };
 442
 443 static int pid_maps_open(struct inode *inode, struct file *file)
 444 {
 445         return do_maps_open(inode, file, &proc_pid_maps_op);
 446 }
 447
 448 static int tid_maps_open(struct inode *inode, struct file *file)
 449 {
 450         return do_maps_open(inode, file, &proc_tid_maps_op);
 451 }
 452
 453 const struct file_operations proc_pid_maps_operations = {
 454         .open           = pid_maps_open,
 455         .read           = seq_read,
 456         .llseek         = seq_lseek,
 457         .release        = proc_map_release,
 458 };
 459
 460 const struct file_operations proc_tid_maps_operations = {
 461         .open           = tid_maps_open,
 462         .read           = seq_read,
 463         .llseek         = seq_lseek,
 464         .release        = proc_map_release,
 465 };
 466
 467 /*
 468  * Proportional Set Size(PSS): my share of RSS.
 469  *
 470  * PSS of a process is the count of pages it has in memory, where each
 471  * page is divided by the number of processes sharing it.  So if a
 472  * process has 1000 pages all to itself, and 1000 shared with one other
 473  * process, its PSS will be 1500.
 474  *
 475  * To keep (accumulated) division errors low, we adopt a 64bit
 476  * fixed-point pss counter to minimize division errors. So (pss >>
 477  * PSS_SHIFT) would be the real byte count.
 478  *
 479  * A shift of 12 before division means (assuming 4K page size):
 480  *      - 1M 3-user-pages add up to 8KB errors;
 481  *      - supports mapcount up to 2^24, or 16M;
 482  *      - supports PSS up to 2^52 bytes, or 4PB.
 483  */
 484 #define PSS_SHIFT 12
 485
 486 #ifdef CONFIG_PROC_PAGE_MONITOR
 487 struct mem_size_stats {
 488         unsigned long resident;
 489         unsigned long shared_clean;
 490         unsigned long shared_dirty;
 491         unsigned long private_clean;
 492         unsigned long private_dirty;
 493         unsigned long referenced;
 494         unsigned long anonymous;
 495         unsigned long anonymous_thp;
 496         unsigned long swap;
 497         unsigned long shared_hugetlb;
 498         unsigned long private_hugetlb;
 499         u64 pss;
 500         u64 swap_pss;
 501 };
 502
 503 static void smaps_account(struct mem_size_stats *mss, struct page *page,
 504                 unsigned long size, bool young, bool dirty)
 505 {
 506         int mapcount;
 507
 508         if (PageAnon(page))
 509                 mss->anonymous += size;
 510
 511         mss->resident += size;
 512         /* Accumulate the size in pages that have been accessed. */
 513         if (young || page_is_young(page) || PageReferenced(page))
 514                 mss->referenced += size;
 515         mapcount = page_mapcount(page);
 516         if (mapcount >= 2) {
 517                 u64 pss_delta;
 518
 519                 if (dirty || PageDirty(page))
 520                         mss->shared_dirty += size;
 521                 else
 522                         mss->shared_clean += size;
 523                 pss_delta = (u64)size << PSS_SHIFT;
 524                 do_div(pss_delta, mapcount);
 525                 mss->pss += pss_delta;
 526         } else {
 527                 if (dirty || PageDirty(page))
 528                         mss->private_dirty += size;
 529                 else
 530                         mss->private_clean += size;
 531                 mss->pss += (u64)size << PSS_SHIFT;
 532         }
 533 }
 534
 535 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
 536                 struct mm_walk *walk)
 537 {
 538         struct mem_size_stats *mss = walk->private;
 539         struct vm_area_struct *vma = walk->vma;
 540         struct page *page = NULL;
 541
 542         if (pte_present(*pte)) {
 543                 page = vm_normal_page(vma, addr, *pte);
 544         } else if (is_swap_pte(*pte)) {
 545                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 546
 547                 if (!non_swap_entry(swpent)) {
 548                         int mapcount;
 549
 550                         mss->swap += PAGE_SIZE;
 551                         mapcount = swp_swapcount(swpent);
 552                         if (mapcount >= 2) {
 553                                 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
 554
 555                                 do_div(pss_delta, mapcount);
 556                                 mss->swap_pss += pss_delta;
 557                         } else {
 558                                 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
 559                         }
 560                 } else if (is_migration_entry(swpent))
 561                         page = migration_entry_to_page(swpent);
 562         }
 563
 564         if (!page)
 565                 return;
 566         smaps_account(mss, page, PAGE_SIZE, 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         mss->anonymous_thp += HPAGE_PMD_SIZE;
 582         smaps_account(mss, page, HPAGE_PMD_SIZE,
 583                         pmd_young(*pmd), pmd_dirty(*pmd));
 584 }
 585 #else
 586 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 587                 struct mm_walk *walk)
 588 {
 589 }
 590 #endif
 591
 592 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 593                            struct mm_walk *walk)
 594 {
 595         struct vm_area_struct *vma = walk->vma;
 596         pte_t *pte;
 597         spinlock_t *ptl;
 598
 599         if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
 600                 smaps_pmd_entry(pmd, addr, walk);
 601                 spin_unlock(ptl);
 602                 return 0;
 603         }
 604
 605         if (pmd_trans_unstable(pmd))
 606                 return 0;
 607         /*
 608          * The mmap_sem held all the way back in m_start() is what
 609          * keeps khugepaged out of here and from collapsing things
 610          * in here.
 611          */
 612         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 613         for (; addr != end; pte++, addr += PAGE_SIZE)
 614                 smaps_pte_entry(pte, addr, walk);
 615         pte_unmap_unlock(pte - 1, ptl);
 616         cond_resched();
 617         return 0;
 618 }
 619
 620 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
 621 {
 622         /*
 623          * Don't forget to update Documentation/ on changes.
 624          */
 625         static const char mnemonics[BITS_PER_LONG][2] = {
 626                 /*
 627                  * In case if we meet a flag we don't know about.
 628                  */
 629                 [0 ... (BITS_PER_LONG-1)] = "??",
 630
 631                 [ilog2(VM_READ)]        = "rd",
 632                 [ilog2(VM_WRITE)]       = "wr",
 633                 [ilog2(VM_EXEC)]        = "ex",
 634                 [ilog2(VM_SHARED)]      = "sh",
 635                 [ilog2(VM_MAYREAD)]     = "mr",
 636                 [ilog2(VM_MAYWRITE)]    = "mw",
 637                 [ilog2(VM_MAYEXEC)]     = "me",
 638                 [ilog2(VM_MAYSHARE)]    = "ms",
 639                 [ilog2(VM_GROWSDOWN)]   = "gd",
 640                 [ilog2(VM_PFNMAP)]      = "pf",
 641                 [ilog2(VM_DENYWRITE)]   = "dw",
 642 #ifdef CONFIG_X86_INTEL_MPX
 643                 [ilog2(VM_MPX)]         = "mp",
 644 #endif
 645                 [ilog2(VM_LOCKED)]      = "lo",
 646                 [ilog2(VM_IO)]          = "io",
 647                 [ilog2(VM_SEQ_READ)]    = "sr",
 648                 [ilog2(VM_RAND_READ)]   = "rr",
 649                 [ilog2(VM_DONTCOPY)]    = "dc",
 650                 [ilog2(VM_DONTEXPAND)]  = "de",
 651                 [ilog2(VM_ACCOUNT)]     = "ac",
 652                 [ilog2(VM_NORESERVE)]   = "nr",
 653                 [ilog2(VM_HUGETLB)]     = "ht",
 654                 [ilog2(VM_ARCH_1)]      = "ar",
 655                 [ilog2(VM_DONTDUMP)]    = "dd",
 656 #ifdef CONFIG_MEM_SOFT_DIRTY
 657                 [ilog2(VM_SOFTDIRTY)]   = "sd",
 658 #endif
 659                 [ilog2(VM_MIXEDMAP)]    = "mm",
 660                 [ilog2(VM_HUGEPAGE)]    = "hg",
 661                 [ilog2(VM_NOHUGEPAGE)]  = "nh",
 662                 [ilog2(VM_MERGEABLE)]   = "mg",
 663                 [ilog2(VM_UFFD_MISSING)]= "um",
 664                 [ilog2(VM_UFFD_WP)]     = "uw",
 665         };
 666         size_t i;
 667
 668         seq_puts(m, "VmFlags: ");
 669         for (i = 0; i < BITS_PER_LONG; i++) {
 670                 if (vma->vm_flags & (1UL << i)) {
 671                         seq_printf(m, "%c%c ",
 672                                    mnemonics[i][0], mnemonics[i][1]);
 673                 }
 674         }
 675         seq_putc(m, '\n');
 676 }
 677
 678 #ifdef CONFIG_HUGETLB_PAGE
 679 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
 680                                  unsigned long addr, unsigned long end,
 681                                  struct mm_walk *walk)
 682 {
 683         struct mem_size_stats *mss = walk->private;
 684         struct vm_area_struct *vma = walk->vma;
 685         struct page *page = NULL;
 686
 687         if (pte_present(*pte)) {
 688                 page = vm_normal_page(vma, addr, *pte);
 689         } else if (is_swap_pte(*pte)) {
 690                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 691
 692                 if (is_migration_entry(swpent))
 693                         page = migration_entry_to_page(swpent);
 694         }
 695         if (page) {
 696                 int mapcount = page_mapcount(page);
 697
 698                 if (mapcount >= 2)
 699                         mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
 700                 else
 701                         mss->private_hugetlb += huge_page_size(hstate_vma(vma));
 702         }
 703         return 0;
 704 }
 705 #endif /* HUGETLB_PAGE */
 706
 707 static int show_smap(struct seq_file *m, void *v, int is_pid)
 708 {
 709         struct vm_area_struct *vma = v;
 710         struct mem_size_stats mss;
 711         struct mm_walk smaps_walk = {
 712                 .pmd_entry = smaps_pte_range,
 713 #ifdef CONFIG_HUGETLB_PAGE
 714                 .hugetlb_entry = smaps_hugetlb_range,
 715 #endif
 716                 .mm = vma->vm_mm,
 717                 .private = &mss,
 718         };
 719
 720         memset(&mss, 0, sizeof mss);
 721         /* mmap_sem is held in m_start */
 722         walk_page_vma(vma, &smaps_walk);
 723
 724         show_map_vma(m, vma, is_pid);
 725
 726         if (vma_get_anon_name(vma)) {
 727                 seq_puts(m, "Name:           ");
 728                 seq_print_vma_name(m, vma);
 729                 seq_putc(m, '\n');
 730         }
 731
 732         seq_printf(m,
 733                    "Size:           %8lu kB\n"
 734                    "Rss:            %8lu kB\n"
 735                    "Pss:            %8lu kB\n"
 736                    "Shared_Clean:   %8lu kB\n"
 737                    "Shared_Dirty:   %8lu kB\n"
 738                    "Private_Clean:  %8lu kB\n"
 739                    "Private_Dirty:  %8lu kB\n"
 740                    "Referenced:     %8lu kB\n"
 741                    "Anonymous:      %8lu kB\n"
 742                    "AnonHugePages:  %8lu kB\n"
 743                    "Shared_Hugetlb: %8lu kB\n"
 744                    "Private_Hugetlb: %7lu kB\n"
 745                    "Swap:           %8lu kB\n"
 746                    "SwapPss:        %8lu kB\n"
 747                    "KernelPageSize: %8lu kB\n"
 748                    "MMUPageSize:    %8lu kB\n"
 749                    "Locked:         %8lu kB\n",
 750                    (vma->vm_end - vma->vm_start) >> 10,
 751                    mss.resident >> 10,
 752                    (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
 753                    mss.shared_clean  >> 10,
 754                    mss.shared_dirty  >> 10,
 755                    mss.private_clean >> 10,
 756                    mss.private_dirty >> 10,
 757                    mss.referenced >> 10,
 758                    mss.anonymous >> 10,
 759                    mss.anonymous_thp >> 10,
 760                    mss.shared_hugetlb >> 10,
 761                    mss.private_hugetlb >> 10,
 762                    mss.swap >> 10,
 763                    (unsigned long)(mss.swap_pss >> (10 + PSS_SHIFT)),
 764                    vma_kernel_pagesize(vma) >> 10,
 765                    vma_mmu_pagesize(vma) >> 10,
 766                    (vma->vm_flags & VM_LOCKED) ?
 767                         (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
 768
 769         show_smap_vma_flags(m, vma);
 770         m_cache_vma(m, vma);
 771         return 0;
 772 }
 773
 774 static int show_pid_smap(struct seq_file *m, void *v)
 775 {
 776         return show_smap(m, v, 1);
 777 }
 778
 779 static int show_tid_smap(struct seq_file *m, void *v)
 780 {
 781         return show_smap(m, v, 0);
 782 }
 783
 784 static const struct seq_operations proc_pid_smaps_op = {
 785         .start  = m_start,
 786         .next   = m_next,
 787         .stop   = m_stop,
 788         .show   = show_pid_smap
 789 };
 790
 791 static const struct seq_operations proc_tid_smaps_op = {
 792         .start  = m_start,
 793         .next   = m_next,
 794         .stop   = m_stop,
 795         .show   = show_tid_smap
 796 };
 797
 798 static int pid_smaps_open(struct inode *inode, struct file *file)
 799 {
 800         return do_maps_open(inode, file, &proc_pid_smaps_op);
 801 }
 802
 803 static int tid_smaps_open(struct inode *inode, struct file *file)
 804 {
 805         return do_maps_open(inode, file, &proc_tid_smaps_op);
 806 }
 807
 808 const struct file_operations proc_pid_smaps_operations = {
 809         .open           = pid_smaps_open,
 810         .read           = seq_read,
 811         .llseek         = seq_lseek,
 812         .release        = proc_map_release,
 813 };
 814
 815 const struct file_operations proc_tid_smaps_operations = {
 816         .open           = tid_smaps_open,
 817         .read           = seq_read,
 818         .llseek         = seq_lseek,
 819         .release        = proc_map_release,
 820 };
 821
 822 enum clear_refs_types {
 823         CLEAR_REFS_ALL = 1,
 824         CLEAR_REFS_ANON,
 825         CLEAR_REFS_MAPPED,
 826         CLEAR_REFS_SOFT_DIRTY,
 827         CLEAR_REFS_MM_HIWATER_RSS,
 828         CLEAR_REFS_LAST,
 829 };
 830
 831 struct clear_refs_private {
 832         enum clear_refs_types type;
 833 };
 834
 835 #ifdef CONFIG_MEM_SOFT_DIRTY
 836 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 837                 unsigned long addr, pte_t *pte)
 838 {
 839         /*
 840          * The soft-dirty tracker uses #PF-s to catch writes
 841          * to pages, so write-protect the pte as well. See the
 842          * Documentation/vm/soft-dirty.txt for full description
 843          * of how soft-dirty works.
 844          */
 845         pte_t ptent = *pte;
 846
 847         if (pte_present(ptent)) {
 848                 ptent = ptep_modify_prot_start(vma->vm_mm, addr, pte);
 849                 ptent = pte_wrprotect(ptent);
 850                 ptent = pte_clear_soft_dirty(ptent);
 851                 ptep_modify_prot_commit(vma->vm_mm, addr, pte, ptent);
 852         } else if (is_swap_pte(ptent)) {
 853                 ptent = pte_swp_clear_soft_dirty(ptent);
 854                 set_pte_at(vma->vm_mm, addr, pte, ptent);
 855         }
 856 }
 857 #else
 858 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 859                 unsigned long addr, pte_t *pte)
 860 {
 861 }
 862 #endif
 863
 864 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
 865 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
 866                 unsigned long addr, pmd_t *pmdp)
 867 {
 868         pmd_t pmd = pmdp_huge_get_and_clear(vma->vm_mm, addr, pmdp);
 869
 870         pmd = pmd_wrprotect(pmd);
 871         pmd = pmd_clear_soft_dirty(pmd);
 872
 873         if (vma->vm_flags & VM_SOFTDIRTY)
 874                 vma->vm_flags &= ~VM_SOFTDIRTY;
 875
 876         set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
 877 }
 878 #else
 879 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
 880                 unsigned long addr, pmd_t *pmdp)
 881 {
 882 }
 883 #endif
 884
 885 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
 886                                 unsigned long end, struct mm_walk *walk)
 887 {
 888         struct clear_refs_private *cp = walk->private;
 889         struct vm_area_struct *vma = walk->vma;
 890         pte_t *pte, ptent;
 891         spinlock_t *ptl;
 892         struct page *page;
 893
 894         if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
 895                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
 896                         clear_soft_dirty_pmd(vma, addr, pmd);
 897                         goto out;
 898                 }
 899
 900                 page = pmd_page(*pmd);
 901
 902                 /* Clear accessed and referenced bits. */
 903                 pmdp_test_and_clear_young(vma, addr, pmd);
 904                 test_and_clear_page_young(page);
 905                 ClearPageReferenced(page);
 906 out:
 907                 spin_unlock(ptl);
 908                 return 0;
 909         }
 910
 911         if (pmd_trans_unstable(pmd))
 912                 return 0;
 913
 914         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 915         for (; addr != end; pte++, addr += PAGE_SIZE) {
 916                 ptent = *pte;
 917
 918                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
 919                         clear_soft_dirty(vma, addr, pte);
 920                         continue;
 921                 }
 922
 923                 if (!pte_present(ptent))
 924                         continue;
 925
 926                 page = vm_normal_page(vma, addr, ptent);
 927                 if (!page)
 928                         continue;
 929
 930                 /* Clear accessed and referenced bits. */
 931                 ptep_test_and_clear_young(vma, addr, pte);
 932                 test_and_clear_page_young(page);
 933                 ClearPageReferenced(page);
 934         }
 935         pte_unmap_unlock(pte - 1, ptl);
 936         cond_resched();
 937         return 0;
 938 }
 939
 940 static int clear_refs_test_walk(unsigned long start, unsigned long end,
 941                                 struct mm_walk *walk)
 942 {
 943         struct clear_refs_private *cp = walk->private;
 944         struct vm_area_struct *vma = walk->vma;
 945
 946         if (vma->vm_flags & VM_PFNMAP)
 947                 return 1;
 948
 949         /*
 950          * Writing 1 to /proc/pid/clear_refs affects all pages.
 951          * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
 952          * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
 953          * Writing 4 to /proc/pid/clear_refs affects all pages.
 954          */
 955         if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
 956                 return 1;
 957         if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
 958                 return 1;
 959         return 0;
 960 }
 961
 962 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
 963                                 size_t count, loff_t *ppos)
 964 {
 965         struct task_struct *task;
 966         char buffer[PROC_NUMBUF];
 967         struct mm_struct *mm;
 968         struct vm_area_struct *vma;
 969         enum clear_refs_types type;
 970         int itype;
 971         int rv;
 972
 973         memset(buffer, 0, sizeof(buffer));
 974         if (count > sizeof(buffer) - 1)
 975                 count = sizeof(buffer) - 1;
 976         if (copy_from_user(buffer, buf, count))
 977                 return -EFAULT;
 978         rv = kstrtoint(strstrip(buffer), 10, &itype);
 979         if (rv < 0)
 980                 return rv;
 981         type = (enum clear_refs_types)itype;
 982         if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
 983                 return -EINVAL;
 984
 985         task = get_proc_task(file_inode(file));
 986         if (!task)
 987                 return -ESRCH;
 988         mm = get_task_mm(task);
 989         if (mm) {
 990                 struct clear_refs_private cp = {
 991                         .type = type,
 992                 };
 993                 struct mm_walk clear_refs_walk = {
 994                         .pmd_entry = clear_refs_pte_range,
 995                         .test_walk = clear_refs_test_walk,
 996                         .mm = mm,
 997                         .private = &cp,
 998                 };
 999
1000                 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1001                         /*
1002                          * Writing 5 to /proc/pid/clear_refs resets the peak
1003                          * resident set size to this mm's current rss value.
1004                          */
1005                         down_write(&mm->mmap_sem);
1006                         reset_mm_hiwater_rss(mm);
1007                         up_write(&mm->mmap_sem);
1008                         goto out_mm;
1009                 }
1010
1011                 down_read(&mm->mmap_sem);
1012                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1013                         for (vma = mm->mmap; vma; vma = vma->vm_next) {
1014                                 if (!(vma->vm_flags & VM_SOFTDIRTY))
1015                                         continue;
1016                                 up_read(&mm->mmap_sem);
1017                                 down_write(&mm->mmap_sem);
1018                                 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1019                                         vma->vm_flags &= ~VM_SOFTDIRTY;
1020                                         vma_set_page_prot(vma);
1021                                 }
1022                                 downgrade_write(&mm->mmap_sem);
1023                                 break;
1024                         }
1025                         mmu_notifier_invalidate_range_start(mm, 0, -1);
1026                 }
1027                 walk_page_range(0, ~0UL, &clear_refs_walk);
1028                 if (type == CLEAR_REFS_SOFT_DIRTY)
1029                         mmu_notifier_invalidate_range_end(mm, 0, -1);
1030                 flush_tlb_mm(mm);
1031                 up_read(&mm->mmap_sem);
1032 out_mm:
1033                 mmput(mm);
1034         }
1035         put_task_struct(task);
1036
1037         return count;
1038 }
1039
1040 const struct file_operations proc_clear_refs_operations = {
1041         .write          = clear_refs_write,
1042         .llseek         = noop_llseek,
1043 };
1044
1045 typedef struct {
1046         u64 pme;
1047 } pagemap_entry_t;
1048
1049 struct pagemapread {
1050         int pos, len;           /* units: PM_ENTRY_BYTES, not bytes */
1051         pagemap_entry_t *buffer;
1052         bool show_pfn;
1053 };
1054
1055 #define PAGEMAP_WALK_SIZE       (PMD_SIZE)
1056 #define PAGEMAP_WALK_MASK       (PMD_MASK)
1057
1058 #define PM_ENTRY_BYTES          sizeof(pagemap_entry_t)
1059 #define PM_PFRAME_BITS          55
1060 #define PM_PFRAME_MASK          GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1061 #define PM_SOFT_DIRTY           BIT_ULL(55)
1062 #define PM_MMAP_EXCLUSIVE       BIT_ULL(56)
1063 #define PM_FILE                 BIT_ULL(61)
1064 #define PM_SWAP                 BIT_ULL(62)
1065 #define PM_PRESENT              BIT_ULL(63)
1066
1067 #define PM_END_OF_BUFFER    1
1068
1069 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1070 {
1071         return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1072 }
1073
1074 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1075                           struct pagemapread *pm)
1076 {
1077         pm->buffer[pm->pos++] = *pme;
1078         if (pm->pos >= pm->len)
1079                 return PM_END_OF_BUFFER;
1080         return 0;
1081 }
1082
1083 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1084                                 struct mm_walk *walk)
1085 {
1086         struct pagemapread *pm = walk->private;
1087         unsigned long addr = start;
1088         int err = 0;
1089
1090         while (addr < end) {
1091                 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1092                 pagemap_entry_t pme = make_pme(0, 0);
1093                 /* End of address space hole, which we mark as non-present. */
1094                 unsigned long hole_end;
1095
1096                 if (vma)
1097                         hole_end = min(end, vma->vm_start);
1098                 else
1099                         hole_end = end;
1100
1101                 for (; addr < hole_end; addr += PAGE_SIZE) {
1102                         err = add_to_pagemap(addr, &pme, pm);
1103                         if (err)
1104                                 goto out;
1105                 }
1106
1107                 if (!vma)
1108                         break;
1109
1110                 /* Addresses in the VMA. */
1111                 if (vma->vm_flags & VM_SOFTDIRTY)
1112                         pme = make_pme(0, PM_SOFT_DIRTY);
1113                 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1114                         err = add_to_pagemap(addr, &pme, pm);
1115                         if (err)
1116                                 goto out;
1117                 }
1118         }
1119 out:
1120         return err;
1121 }
1122
1123 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1124                 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1125 {
1126         u64 frame = 0, flags = 0;
1127         struct page *page = NULL;
1128
1129         if (pte_present(pte)) {
1130                 if (pm->show_pfn)
1131                         frame = pte_pfn(pte);
1132                 flags |= PM_PRESENT;
1133                 page = vm_normal_page(vma, addr, pte);
1134                 if (pte_soft_dirty(pte))
1135                         flags |= PM_SOFT_DIRTY;
1136         } else if (is_swap_pte(pte)) {
1137                 swp_entry_t entry;
1138                 if (pte_swp_soft_dirty(pte))
1139                         flags |= PM_SOFT_DIRTY;
1140                 entry = pte_to_swp_entry(pte);
1141                 frame = swp_type(entry) |
1142                         (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1143                 flags |= PM_SWAP;
1144                 if (is_migration_entry(entry))
1145                         page = migration_entry_to_page(entry);
1146         }
1147
1148         if (page && !PageAnon(page))
1149                 flags |= PM_FILE;
1150         if (page && page_mapcount(page) == 1)
1151                 flags |= PM_MMAP_EXCLUSIVE;
1152         if (vma->vm_flags & VM_SOFTDIRTY)
1153                 flags |= PM_SOFT_DIRTY;
1154
1155         return make_pme(frame, flags);
1156 }
1157
1158 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1159                              struct mm_walk *walk)
1160 {
1161         struct vm_area_struct *vma = walk->vma;
1162         struct pagemapread *pm = walk->private;
1163         spinlock_t *ptl;
1164         pte_t *pte, *orig_pte;
1165         int err = 0;
1166
1167 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1168         if (pmd_trans_huge_lock(pmdp, vma, &ptl) == 1) {
1169                 u64 flags = 0, frame = 0;
1170                 pmd_t pmd = *pmdp;
1171
1172                 if ((vma->vm_flags & VM_SOFTDIRTY) || pmd_soft_dirty(pmd))
1173                         flags |= PM_SOFT_DIRTY;
1174
1175                 /*
1176                  * Currently pmd for thp is always present because thp
1177                  * can not be swapped-out, migrated, or HWPOISONed
1178                  * (split in such cases instead.)
1179                  * This if-check is just to prepare for future implementation.
1180                  */
1181                 if (pmd_present(pmd)) {
1182                         struct page *page = pmd_page(pmd);
1183
1184                         if (page_mapcount(page) == 1)
1185                                 flags |= PM_MMAP_EXCLUSIVE;
1186
1187                         flags |= PM_PRESENT;
1188                         if (pm->show_pfn)
1189                                 frame = pmd_pfn(pmd) +
1190                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1191                 }
1192
1193                 for (; addr != end; addr += PAGE_SIZE) {
1194                         pagemap_entry_t pme = make_pme(frame, flags);
1195
1196                         err = add_to_pagemap(addr, &pme, pm);
1197                         if (err)
1198                                 break;
1199                         if (pm->show_pfn && (flags & PM_PRESENT))
1200                                 frame++;
1201                 }
1202                 spin_unlock(ptl);
1203                 return err;
1204         }
1205
1206         if (pmd_trans_unstable(pmdp))
1207                 return 0;
1208 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1209
1210         /*
1211          * We can assume that @vma always points to a valid one and @end never
1212          * goes beyond vma->vm_end.
1213          */
1214         orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1215         for (; addr < end; pte++, addr += PAGE_SIZE) {
1216                 pagemap_entry_t pme;
1217
1218                 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1219                 err = add_to_pagemap(addr, &pme, pm);
1220                 if (err)
1221                         break;
1222         }
1223         pte_unmap_unlock(orig_pte, ptl);
1224
1225         cond_resched();
1226
1227         return err;
1228 }
1229
1230 #ifdef CONFIG_HUGETLB_PAGE
1231 /* This function walks within one hugetlb entry in the single call */
1232 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1233                                  unsigned long addr, unsigned long end,
1234                                  struct mm_walk *walk)
1235 {
1236         struct pagemapread *pm = walk->private;
1237         struct vm_area_struct *vma = walk->vma;
1238         u64 flags = 0, frame = 0;
1239         int err = 0;
1240         pte_t pte;
1241
1242         if (vma->vm_flags & VM_SOFTDIRTY)
1243                 flags |= PM_SOFT_DIRTY;
1244
1245         pte = huge_ptep_get(ptep);
1246         if (pte_present(pte)) {
1247                 struct page *page = pte_page(pte);
1248
1249                 if (!PageAnon(page))
1250                         flags |= PM_FILE;
1251
1252                 if (page_mapcount(page) == 1)
1253                         flags |= PM_MMAP_EXCLUSIVE;
1254
1255                 flags |= PM_PRESENT;
1256                 if (pm->show_pfn)
1257                         frame = pte_pfn(pte) +
1258                                 ((addr & ~hmask) >> PAGE_SHIFT);
1259         }
1260
1261         for (; addr != end; addr += PAGE_SIZE) {
1262                 pagemap_entry_t pme = make_pme(frame, flags);
1263
1264                 err = add_to_pagemap(addr, &pme, pm);
1265                 if (err)
1266                         return err;
1267                 if (pm->show_pfn && (flags & PM_PRESENT))
1268                         frame++;
1269         }
1270
1271         cond_resched();
1272
1273         return err;
1274 }
1275 #endif /* HUGETLB_PAGE */
1276
1277 /*
1278  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1279  *
1280  * For each page in the address space, this file contains one 64-bit entry
1281  * consisting of the following:
1282  *
1283  * Bits 0-54  page frame number (PFN) if present
1284  * Bits 0-4   swap type if swapped
1285  * Bits 5-54  swap offset if swapped
1286  * Bit  55    pte is soft-dirty (see Documentation/vm/soft-dirty.txt)
1287  * Bit  56    page exclusively mapped
1288  * Bits 57-60 zero
1289  * Bit  61    page is file-page or shared-anon
1290  * Bit  62    page swapped
1291  * Bit  63    page present
1292  *
1293  * If the page is not present but in swap, then the PFN contains an
1294  * encoding of the swap file number and the page's offset into the
1295  * swap. Unmapped pages return a null PFN. This allows determining
1296  * precisely which pages are mapped (or in swap) and comparing mapped
1297  * pages between processes.
1298  *
1299  * Efficient users of this interface will use /proc/pid/maps to
1300  * determine which areas of memory are actually mapped and llseek to
1301  * skip over unmapped regions.
1302  */
1303 static ssize_t pagemap_read(struct file *file, char __user *buf,
1304                             size_t count, loff_t *ppos)
1305 {
1306         struct mm_struct *mm = file->private_data;
1307         struct pagemapread pm;
1308         struct mm_walk pagemap_walk = {};
1309         unsigned long src;
1310         unsigned long svpfn;
1311         unsigned long start_vaddr;
1312         unsigned long end_vaddr;
1313         int ret = 0, copied = 0;
1314
1315         if (!mm || !atomic_inc_not_zero(&mm->mm_users))
1316                 goto out;
1317
1318         ret = -EINVAL;
1319         /* file position must be aligned */
1320         if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1321                 goto out_mm;
1322
1323         ret = 0;
1324         if (!count)
1325                 goto out_mm;
1326
1327         /* do not disclose physical addresses: attack vector */
1328         pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1329
1330         pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1331         pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY);
1332         ret = -ENOMEM;
1333         if (!pm.buffer)
1334                 goto out_mm;
1335
1336         pagemap_walk.pmd_entry = pagemap_pmd_range;
1337         pagemap_walk.pte_hole = pagemap_pte_hole;
1338 #ifdef CONFIG_HUGETLB_PAGE
1339         pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1340 #endif
1341         pagemap_walk.mm = mm;
1342         pagemap_walk.private = &pm;
1343
1344         src = *ppos;
1345         svpfn = src / PM_ENTRY_BYTES;
1346         start_vaddr = svpfn << PAGE_SHIFT;
1347         end_vaddr = mm->task_size;
1348
1349         /* watch out for wraparound */
1350         if (svpfn > mm->task_size >> PAGE_SHIFT)
1351                 start_vaddr = end_vaddr;
1352
1353         /*
1354          * The odds are that this will stop walking way
1355          * before end_vaddr, because the length of the
1356          * user buffer is tracked in "pm", and the walk
1357          * will stop when we hit the end of the buffer.
1358          */
1359         ret = 0;
1360         while (count && (start_vaddr < end_vaddr)) {
1361                 int len;
1362                 unsigned long end;
1363
1364                 pm.pos = 0;
1365                 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1366                 /* overflow ? */
1367                 if (end < start_vaddr || end > end_vaddr)
1368                         end = end_vaddr;
1369                 down_read(&mm->mmap_sem);
1370                 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1371                 up_read(&mm->mmap_sem);
1372                 start_vaddr = end;
1373
1374                 len = min(count, PM_ENTRY_BYTES * pm.pos);
1375                 if (copy_to_user(buf, pm.buffer, len)) {
1376                         ret = -EFAULT;
1377                         goto out_free;
1378                 }
1379                 copied += len;
1380                 buf += len;
1381                 count -= len;
1382         }
1383         *ppos += copied;
1384         if (!ret || ret == PM_END_OF_BUFFER)
1385                 ret = copied;
1386
1387 out_free:
1388         kfree(pm.buffer);
1389 out_mm:
1390         mmput(mm);
1391 out:
1392         return ret;
1393 }
1394
1395 static int pagemap_open(struct inode *inode, struct file *file)
1396 {
1397         struct mm_struct *mm;
1398
1399         mm = proc_mem_open(inode, PTRACE_MODE_READ);
1400         if (IS_ERR(mm))
1401                 return PTR_ERR(mm);
1402         file->private_data = mm;
1403         return 0;
1404 }
1405
1406 static int pagemap_release(struct inode *inode, struct file *file)
1407 {
1408         struct mm_struct *mm = file->private_data;
1409
1410         if (mm)
1411                 mmdrop(mm);
1412         return 0;
1413 }
1414
1415 const struct file_operations proc_pagemap_operations = {
1416         .llseek         = mem_lseek, /* borrow this */
1417         .read           = pagemap_read,
1418         .open           = pagemap_open,
1419         .release        = pagemap_release,
1420 };
1421 #endif /* CONFIG_PROC_PAGE_MONITOR */
1422
1423 #ifdef CONFIG_NUMA
1424
1425 struct numa_maps {
1426         unsigned long pages;
1427         unsigned long anon;
1428         unsigned long active;
1429         unsigned long writeback;
1430         unsigned long mapcount_max;
1431         unsigned long dirty;
1432         unsigned long swapcache;
1433         unsigned long node[MAX_NUMNODES];
1434 };
1435
1436 struct numa_maps_private {
1437         struct proc_maps_private proc_maps;
1438         struct numa_maps md;
1439 };
1440
1441 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1442                         unsigned long nr_pages)
1443 {
1444         int count = page_mapcount(page);
1445
1446         md->pages += nr_pages;
1447         if (pte_dirty || PageDirty(page))
1448                 md->dirty += nr_pages;
1449
1450         if (PageSwapCache(page))
1451                 md->swapcache += nr_pages;
1452
1453         if (PageActive(page) || PageUnevictable(page))
1454                 md->active += nr_pages;
1455
1456         if (PageWriteback(page))
1457                 md->writeback += nr_pages;
1458
1459         if (PageAnon(page))
1460                 md->anon += nr_pages;
1461
1462         if (count > md->mapcount_max)
1463                 md->mapcount_max = count;
1464
1465         md->node[page_to_nid(page)] += nr_pages;
1466 }
1467
1468 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1469                 unsigned long addr)
1470 {
1471         struct page *page;
1472         int nid;
1473
1474         if (!pte_present(pte))
1475                 return NULL;
1476
1477         page = vm_normal_page(vma, addr, pte);
1478         if (!page)
1479                 return NULL;
1480
1481         if (PageReserved(page))
1482                 return NULL;
1483
1484         nid = page_to_nid(page);
1485         if (!node_isset(nid, node_states[N_MEMORY]))
1486                 return NULL;
1487
1488         return page;
1489 }
1490
1491 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1492 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1493                                               struct vm_area_struct *vma,
1494                                               unsigned long addr)
1495 {
1496         struct page *page;
1497         int nid;
1498
1499         if (!pmd_present(pmd))
1500                 return NULL;
1501
1502         page = vm_normal_page_pmd(vma, addr, pmd);
1503         if (!page)
1504                 return NULL;
1505
1506         if (PageReserved(page))
1507                 return NULL;
1508
1509         nid = page_to_nid(page);
1510         if (!node_isset(nid, node_states[N_MEMORY]))
1511                 return NULL;
1512
1513         return page;
1514 }
1515 #endif
1516
1517 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1518                 unsigned long end, struct mm_walk *walk)
1519 {
1520         struct numa_maps *md = walk->private;
1521         struct vm_area_struct *vma = walk->vma;
1522         spinlock_t *ptl;
1523         pte_t *orig_pte;
1524         pte_t *pte;
1525
1526 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1527         if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
1528                 struct page *page;
1529
1530                 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1531                 if (page)
1532                         gather_stats(page, md, pmd_dirty(*pmd),
1533                                      HPAGE_PMD_SIZE/PAGE_SIZE);
1534                 spin_unlock(ptl);
1535                 return 0;
1536         }
1537
1538         if (pmd_trans_unstable(pmd))
1539                 return 0;
1540 #endif
1541         orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1542         do {
1543                 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1544                 if (!page)
1545                         continue;
1546                 gather_stats(page, md, pte_dirty(*pte), 1);
1547
1548         } while (pte++, addr += PAGE_SIZE, addr != end);
1549         pte_unmap_unlock(orig_pte, ptl);
1550         return 0;
1551 }
1552 #ifdef CONFIG_HUGETLB_PAGE
1553 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1554                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1555 {
1556         pte_t huge_pte = huge_ptep_get(pte);
1557         struct numa_maps *md;
1558         struct page *page;
1559
1560         if (!pte_present(huge_pte))
1561                 return 0;
1562
1563         page = pte_page(huge_pte);
1564         if (!page)
1565                 return 0;
1566
1567         md = walk->private;
1568         gather_stats(page, md, pte_dirty(huge_pte), 1);
1569         return 0;
1570 }
1571
1572 #else
1573 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1574                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1575 {
1576         return 0;
1577 }
1578 #endif
1579
1580 /*
1581  * Display pages allocated per node and memory policy via /proc.
1582  */
1583 static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1584 {
1585         struct numa_maps_private *numa_priv = m->private;
1586         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1587         struct vm_area_struct *vma = v;
1588         struct numa_maps *md = &numa_priv->md;
1589         struct file *file = vma->vm_file;
1590         struct mm_struct *mm = vma->vm_mm;
1591         struct mm_walk walk = {
1592                 .hugetlb_entry = gather_hugetlb_stats,
1593                 .pmd_entry = gather_pte_stats,
1594                 .private = md,
1595                 .mm = mm,
1596         };
1597         struct mempolicy *pol;
1598         char buffer[64];
1599         int nid;
1600
1601         if (!mm)
1602                 return 0;
1603
1604         /* Ensure we start with an empty set of numa_maps statistics. */
1605         memset(md, 0, sizeof(*md));
1606
1607         pol = __get_vma_policy(vma, vma->vm_start);
1608         if (pol) {
1609                 mpol_to_str(buffer, sizeof(buffer), pol);
1610                 mpol_cond_put(pol);
1611         } else {
1612                 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1613         }
1614
1615         seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1616
1617         if (file) {
1618                 seq_puts(m, " file=");
1619                 seq_file_path(m, file, "\n\t= ");
1620         } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1621                 seq_puts(m, " heap");
1622         } else if (is_stack(proc_priv, vma, is_pid)) {
1623                 seq_puts(m, " stack");
1624         }
1625
1626         if (is_vm_hugetlb_page(vma))
1627                 seq_puts(m, " huge");
1628
1629         /* mmap_sem is held by m_start */
1630         walk_page_vma(vma, &walk);
1631
1632         if (!md->pages)
1633                 goto out;
1634
1635         if (md->anon)
1636                 seq_printf(m, " anon=%lu", md->anon);
1637
1638         if (md->dirty)
1639                 seq_printf(m, " dirty=%lu", md->dirty);
1640
1641         if (md->pages != md->anon && md->pages != md->dirty)
1642                 seq_printf(m, " mapped=%lu", md->pages);
1643
1644         if (md->mapcount_max > 1)
1645                 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1646
1647         if (md->swapcache)
1648                 seq_printf(m, " swapcache=%lu", md->swapcache);
1649
1650         if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1651                 seq_printf(m, " active=%lu", md->active);
1652
1653         if (md->writeback)
1654                 seq_printf(m, " writeback=%lu", md->writeback);
1655
1656         for_each_node_state(nid, N_MEMORY)
1657                 if (md->node[nid])
1658                         seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1659
1660         seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1661 out:
1662         seq_putc(m, '\n');
1663         m_cache_vma(m, vma);
1664         return 0;
1665 }
1666
1667 static int show_pid_numa_map(struct seq_file *m, void *v)
1668 {
1669         return show_numa_map(m, v, 1);
1670 }
1671
1672 static int show_tid_numa_map(struct seq_file *m, void *v)
1673 {
1674         return show_numa_map(m, v, 0);
1675 }
1676
1677 static const struct seq_operations proc_pid_numa_maps_op = {
1678         .start  = m_start,
1679         .next   = m_next,
1680         .stop   = m_stop,
1681         .show   = show_pid_numa_map,
1682 };
1683
1684 static const struct seq_operations proc_tid_numa_maps_op = {
1685         .start  = m_start,
1686         .next   = m_next,
1687         .stop   = m_stop,
1688         .show   = show_tid_numa_map,
1689 };
1690
1691 static int numa_maps_open(struct inode *inode, struct file *file,
1692                           const struct seq_operations *ops)
1693 {
1694         return proc_maps_open(inode, file, ops,
1695                                 sizeof(struct numa_maps_private));
1696 }
1697
1698 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1699 {
1700         return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1701 }
1702
1703 static int tid_numa_maps_open(struct inode *inode, struct file *file)
1704 {
1705         return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1706 }
1707
1708 const struct file_operations proc_pid_numa_maps_operations = {
1709         .open           = pid_numa_maps_open,
1710         .read           = seq_read,
1711         .llseek         = seq_lseek,
1712         .release        = proc_map_release,
1713 };
1714
1715 const struct file_operations proc_tid_numa_maps_operations = {
1716         .open           = tid_numa_maps_open,
1717         .read           = seq_read,
1718         .llseek         = seq_lseek,
1719         .release        = proc_map_release,
1720 };
1721 #endif /* CONFIG_NUMA */