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         end = vma->vm_end;
 349
 350         seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
 351         seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu ",
 352                         start,
 353                         end,
 354                         flags & VM_READ ? 'r' : '-',
 355                         flags & VM_WRITE ? 'w' : '-',
 356                         flags & VM_EXEC ? 'x' : '-',
 357                         flags & VM_MAYSHARE ? 's' : 'p',
 358                         pgoff,
 359                         MAJOR(dev), MINOR(dev), ino);
 360
 361         /*
 362          * Print the dentry name for named mappings, and a
 363          * special [heap] marker for the heap:
 364          */
 365         if (file) {
 366                 seq_pad(m, ' ');
 367                 seq_file_path(m, file, "\n");
 368                 goto done;
 369         }
 370
 371         if (vma->vm_ops && vma->vm_ops->name) {
 372                 name = vma->vm_ops->name(vma);
 373                 if (name)
 374                         goto done;
 375         }
 376
 377         name = arch_vma_name(vma);
 378         if (!name) {
 379                 if (!mm) {
 380                         name = "[vdso]";
 381                         goto done;
 382                 }
 383
 384                 if (vma->vm_start <= mm->brk &&
 385                     vma->vm_end >= mm->start_brk) {
 386                         name = "[heap]";
 387                         goto done;
 388                 }
 389
 390                 if (is_stack(priv, vma, is_pid)) {
 391                         name = "[stack]";
 392                         goto done;
 393                 }
 394                 if (vma_get_anon_name(vma)) {
 395                         seq_pad(m, ' ');
 396                         seq_print_vma_name(m, vma);
 397                 }
 398         }
 399
 400 done:
 401         if (name) {
 402                 seq_pad(m, ' ');
 403                 seq_puts(m, name);
 404         }
 405         seq_putc(m, '\n');
 406 }
 407
 408 static int show_map(struct seq_file *m, void *v, int is_pid)
 409 {
 410         show_map_vma(m, v, is_pid);
 411         m_cache_vma(m, v);
 412         return 0;
 413 }
 414
 415 static int show_pid_map(struct seq_file *m, void *v)
 416 {
 417         return show_map(m, v, 1);
 418 }
 419
 420 static int show_tid_map(struct seq_file *m, void *v)
 421 {
 422         return show_map(m, v, 0);
 423 }
 424
 425 static const struct seq_operations proc_pid_maps_op = {
 426         .start  = m_start,
 427         .next   = m_next,
 428         .stop   = m_stop,
 429         .show   = show_pid_map
 430 };
 431
 432 static const struct seq_operations proc_tid_maps_op = {
 433         .start  = m_start,
 434         .next   = m_next,
 435         .stop   = m_stop,
 436         .show   = show_tid_map
 437 };
 438
 439 static int pid_maps_open(struct inode *inode, struct file *file)
 440 {
 441         return do_maps_open(inode, file, &proc_pid_maps_op);
 442 }
 443
 444 static int tid_maps_open(struct inode *inode, struct file *file)
 445 {
 446         return do_maps_open(inode, file, &proc_tid_maps_op);
 447 }
 448
 449 const struct file_operations proc_pid_maps_operations = {
 450         .open           = pid_maps_open,
 451         .read           = seq_read,
 452         .llseek         = seq_lseek,
 453         .release        = proc_map_release,
 454 };
 455
 456 const struct file_operations proc_tid_maps_operations = {
 457         .open           = tid_maps_open,
 458         .read           = seq_read,
 459         .llseek         = seq_lseek,
 460         .release        = proc_map_release,
 461 };
 462
 463 /*
 464  * Proportional Set Size(PSS): my share of RSS.
 465  *
 466  * PSS of a process is the count of pages it has in memory, where each
 467  * page is divided by the number of processes sharing it.  So if a
 468  * process has 1000 pages all to itself, and 1000 shared with one other
 469  * process, its PSS will be 1500.
 470  *
 471  * To keep (accumulated) division errors low, we adopt a 64bit
 472  * fixed-point pss counter to minimize division errors. So (pss >>
 473  * PSS_SHIFT) would be the real byte count.
 474  *
 475  * A shift of 12 before division means (assuming 4K page size):
 476  *      - 1M 3-user-pages add up to 8KB errors;
 477  *      - supports mapcount up to 2^24, or 16M;
 478  *      - supports PSS up to 2^52 bytes, or 4PB.
 479  */
 480 #define PSS_SHIFT 12
 481
 482 #ifdef CONFIG_PROC_PAGE_MONITOR
 483 struct mem_size_stats {
 484         unsigned long resident;
 485         unsigned long shared_clean;
 486         unsigned long shared_dirty;
 487         unsigned long private_clean;
 488         unsigned long private_dirty;
 489         unsigned long referenced;
 490         unsigned long anonymous;
 491         unsigned long anonymous_thp;
 492         unsigned long swap;
 493         unsigned long shared_hugetlb;
 494         unsigned long private_hugetlb;
 495         u64 pss;
 496         u64 swap_pss;
 497 };
 498
 499 static void smaps_account(struct mem_size_stats *mss, struct page *page,
 500                 unsigned long size, bool young, bool dirty)
 501 {
 502         int mapcount;
 503
 504         if (PageAnon(page))
 505                 mss->anonymous += size;
 506
 507         mss->resident += size;
 508         /* Accumulate the size in pages that have been accessed. */
 509         if (young || page_is_young(page) || PageReferenced(page))
 510                 mss->referenced += size;
 511         mapcount = page_mapcount(page);
 512         if (mapcount >= 2) {
 513                 u64 pss_delta;
 514
 515                 if (dirty || PageDirty(page))
 516                         mss->shared_dirty += size;
 517                 else
 518                         mss->shared_clean += size;
 519                 pss_delta = (u64)size << PSS_SHIFT;
 520                 do_div(pss_delta, mapcount);
 521                 mss->pss += pss_delta;
 522         } else {
 523                 if (dirty || PageDirty(page))
 524                         mss->private_dirty += size;
 525                 else
 526                         mss->private_clean += size;
 527                 mss->pss += (u64)size << PSS_SHIFT;
 528         }
 529 }
 530
 531 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
 532                 struct mm_walk *walk)
 533 {
 534         struct mem_size_stats *mss = walk->private;
 535         struct vm_area_struct *vma = walk->vma;
 536         struct page *page = NULL;
 537
 538         if (pte_present(*pte)) {
 539                 page = vm_normal_page(vma, addr, *pte);
 540         } else if (is_swap_pte(*pte)) {
 541                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 542
 543                 if (!non_swap_entry(swpent)) {
 544                         int mapcount;
 545
 546                         mss->swap += PAGE_SIZE;
 547                         mapcount = swp_swapcount(swpent);
 548                         if (mapcount >= 2) {
 549                                 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
 550
 551                                 do_div(pss_delta, mapcount);
 552                                 mss->swap_pss += pss_delta;
 553                         } else {
 554                                 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
 555                         }
 556                 } else if (is_migration_entry(swpent))
 557                         page = migration_entry_to_page(swpent);
 558         }
 559
 560         if (!page)
 561                 return;
 562         smaps_account(mss, page, PAGE_SIZE, pte_young(*pte), pte_dirty(*pte));
 563 }
 564
 565 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 566 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 567                 struct mm_walk *walk)
 568 {
 569         struct mem_size_stats *mss = walk->private;
 570         struct vm_area_struct *vma = walk->vma;
 571         struct page *page;
 572
 573         /* FOLL_DUMP will return -EFAULT on huge zero page */
 574         page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
 575         if (IS_ERR_OR_NULL(page))
 576                 return;
 577         mss->anonymous_thp += HPAGE_PMD_SIZE;
 578         smaps_account(mss, page, HPAGE_PMD_SIZE,
 579                         pmd_young(*pmd), pmd_dirty(*pmd));
 580 }
 581 #else
 582 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 583                 struct mm_walk *walk)
 584 {
 585 }
 586 #endif
 587
 588 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 589                            struct mm_walk *walk)
 590 {
 591         struct vm_area_struct *vma = walk->vma;
 592         pte_t *pte;
 593         spinlock_t *ptl;
 594
 595         if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
 596                 smaps_pmd_entry(pmd, addr, walk);
 597                 spin_unlock(ptl);
 598                 return 0;
 599         }
 600
 601         if (pmd_trans_unstable(pmd))
 602                 return 0;
 603         /*
 604          * The mmap_sem held all the way back in m_start() is what
 605          * keeps khugepaged out of here and from collapsing things
 606          * in here.
 607          */
 608         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 609         for (; addr != end; pte++, addr += PAGE_SIZE)
 610                 smaps_pte_entry(pte, addr, walk);
 611         pte_unmap_unlock(pte - 1, ptl);
 612         cond_resched();
 613         return 0;
 614 }
 615
 616 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
 617 {
 618         /*
 619          * Don't forget to update Documentation/ on changes.
 620          */
 621         static const char mnemonics[BITS_PER_LONG][2] = {
 622                 /*
 623                  * In case if we meet a flag we don't know about.
 624                  */
 625                 [0 ... (BITS_PER_LONG-1)] = "??",
 626
 627                 [ilog2(VM_READ)]        = "rd",
 628                 [ilog2(VM_WRITE)]       = "wr",
 629                 [ilog2(VM_EXEC)]        = "ex",
 630                 [ilog2(VM_SHARED)]      = "sh",
 631                 [ilog2(VM_MAYREAD)]     = "mr",
 632                 [ilog2(VM_MAYWRITE)]    = "mw",
 633                 [ilog2(VM_MAYEXEC)]     = "me",
 634                 [ilog2(VM_MAYSHARE)]    = "ms",
 635                 [ilog2(VM_GROWSDOWN)]   = "gd",
 636                 [ilog2(VM_PFNMAP)]      = "pf",
 637                 [ilog2(VM_DENYWRITE)]   = "dw",
 638 #ifdef CONFIG_X86_INTEL_MPX
 639                 [ilog2(VM_MPX)]         = "mp",
 640 #endif
 641                 [ilog2(VM_LOCKED)]      = "lo",
 642                 [ilog2(VM_IO)]          = "io",
 643                 [ilog2(VM_SEQ_READ)]    = "sr",
 644                 [ilog2(VM_RAND_READ)]   = "rr",
 645                 [ilog2(VM_DONTCOPY)]    = "dc",
 646                 [ilog2(VM_DONTEXPAND)]  = "de",
 647                 [ilog2(VM_ACCOUNT)]     = "ac",
 648                 [ilog2(VM_NORESERVE)]   = "nr",
 649                 [ilog2(VM_HUGETLB)]     = "ht",
 650                 [ilog2(VM_ARCH_1)]      = "ar",
 651                 [ilog2(VM_DONTDUMP)]    = "dd",
 652 #ifdef CONFIG_MEM_SOFT_DIRTY
 653                 [ilog2(VM_SOFTDIRTY)]   = "sd",
 654 #endif
 655                 [ilog2(VM_MIXEDMAP)]    = "mm",
 656                 [ilog2(VM_HUGEPAGE)]    = "hg",
 657                 [ilog2(VM_NOHUGEPAGE)]  = "nh",
 658                 [ilog2(VM_MERGEABLE)]   = "mg",
 659                 [ilog2(VM_UFFD_MISSING)]= "um",
 660                 [ilog2(VM_UFFD_WP)]     = "uw",
 661         };
 662         size_t i;
 663
 664         seq_puts(m, "VmFlags: ");
 665         for (i = 0; i < BITS_PER_LONG; i++) {
 666                 if (vma->vm_flags & (1UL << i)) {
 667                         seq_printf(m, "%c%c ",
 668                                    mnemonics[i][0], mnemonics[i][1]);
 669                 }
 670         }
 671         seq_putc(m, '\n');
 672 }
 673
 674 #ifdef CONFIG_HUGETLB_PAGE
 675 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
 676                                  unsigned long addr, unsigned long end,
 677                                  struct mm_walk *walk)
 678 {
 679         struct mem_size_stats *mss = walk->private;
 680         struct vm_area_struct *vma = walk->vma;
 681         struct page *page = NULL;
 682
 683         if (pte_present(*pte)) {
 684                 page = vm_normal_page(vma, addr, *pte);
 685         } else if (is_swap_pte(*pte)) {
 686                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 687
 688                 if (is_migration_entry(swpent))
 689                         page = migration_entry_to_page(swpent);
 690         }
 691         if (page) {
 692                 int mapcount = page_mapcount(page);
 693
 694                 if (mapcount >= 2)
 695                         mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
 696                 else
 697                         mss->private_hugetlb += huge_page_size(hstate_vma(vma));
 698         }
 699         return 0;
 700 }
 701 #endif /* HUGETLB_PAGE */
 702
 703 static int show_smap(struct seq_file *m, void *v, int is_pid)
 704 {
 705         struct vm_area_struct *vma = v;
 706         struct mem_size_stats mss;
 707         struct mm_walk smaps_walk = {
 708                 .pmd_entry = smaps_pte_range,
 709 #ifdef CONFIG_HUGETLB_PAGE
 710                 .hugetlb_entry = smaps_hugetlb_range,
 711 #endif
 712                 .mm = vma->vm_mm,
 713                 .private = &mss,
 714         };
 715
 716         memset(&mss, 0, sizeof mss);
 717         /* mmap_sem is held in m_start */
 718         walk_page_vma(vma, &smaps_walk);
 719
 720         show_map_vma(m, vma, is_pid);
 721
 722         if (vma_get_anon_name(vma)) {
 723                 seq_puts(m, "Name:           ");
 724                 seq_print_vma_name(m, vma);
 725                 seq_putc(m, '\n');
 726         }
 727
 728         seq_printf(m,
 729                    "Size:           %8lu kB\n"
 730                    "Rss:            %8lu kB\n"
 731                    "Pss:            %8lu kB\n"
 732                    "Shared_Clean:   %8lu kB\n"
 733                    "Shared_Dirty:   %8lu kB\n"
 734                    "Private_Clean:  %8lu kB\n"
 735                    "Private_Dirty:  %8lu kB\n"
 736                    "Referenced:     %8lu kB\n"
 737                    "Anonymous:      %8lu kB\n"
 738                    "AnonHugePages:  %8lu kB\n"
 739                    "Shared_Hugetlb: %8lu kB\n"
 740                    "Private_Hugetlb: %7lu kB\n"
 741                    "Swap:           %8lu kB\n"
 742                    "SwapPss:        %8lu kB\n"
 743                    "KernelPageSize: %8lu kB\n"
 744                    "MMUPageSize:    %8lu kB\n"
 745                    "Locked:         %8lu kB\n",
 746                    (vma->vm_end - vma->vm_start) >> 10,
 747                    mss.resident >> 10,
 748                    (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
 749                    mss.shared_clean  >> 10,
 750                    mss.shared_dirty  >> 10,
 751                    mss.private_clean >> 10,
 752                    mss.private_dirty >> 10,
 753                    mss.referenced >> 10,
 754                    mss.anonymous >> 10,
 755                    mss.anonymous_thp >> 10,
 756                    mss.shared_hugetlb >> 10,
 757                    mss.private_hugetlb >> 10,
 758                    mss.swap >> 10,
 759                    (unsigned long)(mss.swap_pss >> (10 + PSS_SHIFT)),
 760                    vma_kernel_pagesize(vma) >> 10,
 761                    vma_mmu_pagesize(vma) >> 10,
 762                    (vma->vm_flags & VM_LOCKED) ?
 763                         (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
 764
 765         show_smap_vma_flags(m, vma);
 766         m_cache_vma(m, vma);
 767         return 0;
 768 }
 769
 770 static int show_pid_smap(struct seq_file *m, void *v)
 771 {
 772         return show_smap(m, v, 1);
 773 }
 774
 775 static int show_tid_smap(struct seq_file *m, void *v)
 776 {
 777         return show_smap(m, v, 0);
 778 }
 779
 780 static const struct seq_operations proc_pid_smaps_op = {
 781         .start  = m_start,
 782         .next   = m_next,
 783         .stop   = m_stop,
 784         .show   = show_pid_smap
 785 };
 786
 787 static const struct seq_operations proc_tid_smaps_op = {
 788         .start  = m_start,
 789         .next   = m_next,
 790         .stop   = m_stop,
 791         .show   = show_tid_smap
 792 };
 793
 794 static int pid_smaps_open(struct inode *inode, struct file *file)
 795 {
 796         return do_maps_open(inode, file, &proc_pid_smaps_op);
 797 }
 798
 799 static int tid_smaps_open(struct inode *inode, struct file *file)
 800 {
 801         return do_maps_open(inode, file, &proc_tid_smaps_op);
 802 }
 803
 804 const struct file_operations proc_pid_smaps_operations = {
 805         .open           = pid_smaps_open,
 806         .read           = seq_read,
 807         .llseek         = seq_lseek,
 808         .release        = proc_map_release,
 809 };
 810
 811 const struct file_operations proc_tid_smaps_operations = {
 812         .open           = tid_smaps_open,
 813         .read           = seq_read,
 814         .llseek         = seq_lseek,
 815         .release        = proc_map_release,
 816 };
 817
 818 enum clear_refs_types {
 819         CLEAR_REFS_ALL = 1,
 820         CLEAR_REFS_ANON,
 821         CLEAR_REFS_MAPPED,
 822         CLEAR_REFS_SOFT_DIRTY,
 823         CLEAR_REFS_MM_HIWATER_RSS,
 824         CLEAR_REFS_LAST,
 825 };
 826
 827 struct clear_refs_private {
 828         enum clear_refs_types type;
 829 };
 830
 831 #ifdef CONFIG_MEM_SOFT_DIRTY
 832 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 833                 unsigned long addr, pte_t *pte)
 834 {
 835         /*
 836          * The soft-dirty tracker uses #PF-s to catch writes
 837          * to pages, so write-protect the pte as well. See the
 838          * Documentation/vm/soft-dirty.txt for full description
 839          * of how soft-dirty works.
 840          */
 841         pte_t ptent = *pte;
 842
 843         if (pte_present(ptent)) {
 844                 ptent = ptep_modify_prot_start(vma->vm_mm, addr, pte);
 845                 ptent = pte_wrprotect(ptent);
 846                 ptent = pte_clear_soft_dirty(ptent);
 847                 ptep_modify_prot_commit(vma->vm_mm, addr, pte, ptent);
 848         } else if (is_swap_pte(ptent)) {
 849                 ptent = pte_swp_clear_soft_dirty(ptent);
 850                 set_pte_at(vma->vm_mm, addr, pte, ptent);
 851         }
 852 }
 853 #else
 854 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 855                 unsigned long addr, pte_t *pte)
 856 {
 857 }
 858 #endif
 859
 860 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
 861 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
 862                 unsigned long addr, pmd_t *pmdp)
 863 {
 864         pmd_t pmd = *pmdp;
 865
 866         /* See comment in change_huge_pmd() */
 867         pmdp_invalidate(vma, addr, pmdp);
 868         if (pmd_dirty(*pmdp))
 869                 pmd = pmd_mkdirty(pmd);
 870         if (pmd_young(*pmdp))
 871                 pmd = pmd_mkyoung(pmd);
 872
 873         pmd = pmd_wrprotect(pmd);
 874         pmd = pmd_clear_soft_dirty(pmd);
 875
 876         if (vma->vm_flags & VM_SOFTDIRTY)
 877                 vma->vm_flags &= ~VM_SOFTDIRTY;
 878
 879         set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
 880 }
 881 #else
 882 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
 883                 unsigned long addr, pmd_t *pmdp)
 884 {
 885 }
 886 #endif
 887
 888 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
 889                                 unsigned long end, struct mm_walk *walk)
 890 {
 891         struct clear_refs_private *cp = walk->private;
 892         struct vm_area_struct *vma = walk->vma;
 893         pte_t *pte, ptent;
 894         spinlock_t *ptl;
 895         struct page *page;
 896
 897         if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
 898                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
 899                         clear_soft_dirty_pmd(vma, addr, pmd);
 900                         goto out;
 901                 }
 902
 903                 page = pmd_page(*pmd);
 904
 905                 /* Clear accessed and referenced bits. */
 906                 pmdp_test_and_clear_young(vma, addr, pmd);
 907                 test_and_clear_page_young(page);
 908                 ClearPageReferenced(page);
 909 out:
 910                 spin_unlock(ptl);
 911                 return 0;
 912         }
 913
 914         if (pmd_trans_unstable(pmd))
 915                 return 0;
 916
 917         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 918         for (; addr != end; pte++, addr += PAGE_SIZE) {
 919                 ptent = *pte;
 920
 921                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
 922                         clear_soft_dirty(vma, addr, pte);
 923                         continue;
 924                 }
 925
 926                 if (!pte_present(ptent))
 927                         continue;
 928
 929                 page = vm_normal_page(vma, addr, ptent);
 930                 if (!page)
 931                         continue;
 932
 933                 /* Clear accessed and referenced bits. */
 934                 ptep_test_and_clear_young(vma, addr, pte);
 935                 test_and_clear_page_young(page);
 936                 ClearPageReferenced(page);
 937         }
 938         pte_unmap_unlock(pte - 1, ptl);
 939         cond_resched();
 940         return 0;
 941 }
 942
 943 static int clear_refs_test_walk(unsigned long start, unsigned long end,
 944                                 struct mm_walk *walk)
 945 {
 946         struct clear_refs_private *cp = walk->private;
 947         struct vm_area_struct *vma = walk->vma;
 948
 949         if (vma->vm_flags & VM_PFNMAP)
 950                 return 1;
 951
 952         /*
 953          * Writing 1 to /proc/pid/clear_refs affects all pages.
 954          * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
 955          * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
 956          * Writing 4 to /proc/pid/clear_refs affects all pages.
 957          */
 958         if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
 959                 return 1;
 960         if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
 961                 return 1;
 962         return 0;
 963 }
 964
 965 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
 966                                 size_t count, loff_t *ppos)
 967 {
 968         struct task_struct *task;
 969         char buffer[PROC_NUMBUF];
 970         struct mm_struct *mm;
 971         struct vm_area_struct *vma;
 972         enum clear_refs_types type;
 973         int itype;
 974         int rv;
 975
 976         memset(buffer, 0, sizeof(buffer));
 977         if (count > sizeof(buffer) - 1)
 978                 count = sizeof(buffer) - 1;
 979         if (copy_from_user(buffer, buf, count))
 980                 return -EFAULT;
 981         rv = kstrtoint(strstrip(buffer), 10, &itype);
 982         if (rv < 0)
 983                 return rv;
 984         type = (enum clear_refs_types)itype;
 985         if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
 986                 return -EINVAL;
 987
 988         task = get_proc_task(file_inode(file));
 989         if (!task)
 990                 return -ESRCH;
 991         mm = get_task_mm(task);
 992         if (mm) {
 993                 struct clear_refs_private cp = {
 994                         .type = type,
 995                 };
 996                 struct mm_walk clear_refs_walk = {
 997                         .pmd_entry = clear_refs_pte_range,
 998                         .test_walk = clear_refs_test_walk,
 999                         .mm = mm,
1000                         .private = &cp,
1001                 };
1002
1003                 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1004                         /*
1005                          * Writing 5 to /proc/pid/clear_refs resets the peak
1006                          * resident set size to this mm's current rss value.
1007                          */
1008                         down_write(&mm->mmap_sem);
1009                         reset_mm_hiwater_rss(mm);
1010                         up_write(&mm->mmap_sem);
1011                         goto out_mm;
1012                 }
1013
1014                 down_read(&mm->mmap_sem);
1015                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1016                         for (vma = mm->mmap; vma; vma = vma->vm_next) {
1017                                 if (!(vma->vm_flags & VM_SOFTDIRTY))
1018                                         continue;
1019                                 up_read(&mm->mmap_sem);
1020                                 down_write(&mm->mmap_sem);
1021                                 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1022                                         vma->vm_flags &= ~VM_SOFTDIRTY;
1023                                         vma_set_page_prot(vma);
1024                                 }
1025                                 downgrade_write(&mm->mmap_sem);
1026                                 break;
1027                         }
1028                         mmu_notifier_invalidate_range_start(mm, 0, -1);
1029                 }
1030                 walk_page_range(0, ~0UL, &clear_refs_walk);
1031                 if (type == CLEAR_REFS_SOFT_DIRTY)
1032                         mmu_notifier_invalidate_range_end(mm, 0, -1);
1033                 flush_tlb_mm(mm);
1034                 up_read(&mm->mmap_sem);
1035 out_mm:
1036                 mmput(mm);
1037         }
1038         put_task_struct(task);
1039
1040         return count;
1041 }
1042
1043 const struct file_operations proc_clear_refs_operations = {
1044         .write          = clear_refs_write,
1045         .llseek         = noop_llseek,
1046 };
1047
1048 typedef struct {
1049         u64 pme;
1050 } pagemap_entry_t;
1051
1052 struct pagemapread {
1053         int pos, len;           /* units: PM_ENTRY_BYTES, not bytes */
1054         pagemap_entry_t *buffer;
1055         bool show_pfn;
1056 };
1057
1058 #define PAGEMAP_WALK_SIZE       (PMD_SIZE)
1059 #define PAGEMAP_WALK_MASK       (PMD_MASK)
1060
1061 #define PM_ENTRY_BYTES          sizeof(pagemap_entry_t)
1062 #define PM_PFRAME_BITS          55
1063 #define PM_PFRAME_MASK          GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1064 #define PM_SOFT_DIRTY           BIT_ULL(55)
1065 #define PM_MMAP_EXCLUSIVE       BIT_ULL(56)
1066 #define PM_FILE                 BIT_ULL(61)
1067 #define PM_SWAP                 BIT_ULL(62)
1068 #define PM_PRESENT              BIT_ULL(63)
1069
1070 #define PM_END_OF_BUFFER    1
1071
1072 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1073 {
1074         return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1075 }
1076
1077 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1078                           struct pagemapread *pm)
1079 {
1080         pm->buffer[pm->pos++] = *pme;
1081         if (pm->pos >= pm->len)
1082                 return PM_END_OF_BUFFER;
1083         return 0;
1084 }
1085
1086 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1087                                 struct mm_walk *walk)
1088 {
1089         struct pagemapread *pm = walk->private;
1090         unsigned long addr = start;
1091         int err = 0;
1092
1093         while (addr < end) {
1094                 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1095                 pagemap_entry_t pme = make_pme(0, 0);
1096                 /* End of address space hole, which we mark as non-present. */
1097                 unsigned long hole_end;
1098
1099                 if (vma)
1100                         hole_end = min(end, vma->vm_start);
1101                 else
1102                         hole_end = end;
1103
1104                 for (; addr < hole_end; addr += PAGE_SIZE) {
1105                         err = add_to_pagemap(addr, &pme, pm);
1106                         if (err)
1107                                 goto out;
1108                 }
1109
1110                 if (!vma)
1111                         break;
1112
1113                 /* Addresses in the VMA. */
1114                 if (vma->vm_flags & VM_SOFTDIRTY)
1115                         pme = make_pme(0, PM_SOFT_DIRTY);
1116                 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1117                         err = add_to_pagemap(addr, &pme, pm);
1118                         if (err)
1119                                 goto out;
1120                 }
1121         }
1122 out:
1123         return err;
1124 }
1125
1126 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1127                 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1128 {
1129         u64 frame = 0, flags = 0;
1130         struct page *page = NULL;
1131
1132         if (pte_present(pte)) {
1133                 if (pm->show_pfn)
1134                         frame = pte_pfn(pte);
1135                 flags |= PM_PRESENT;
1136                 page = vm_normal_page(vma, addr, pte);
1137                 if (pte_soft_dirty(pte))
1138                         flags |= PM_SOFT_DIRTY;
1139         } else if (is_swap_pte(pte)) {
1140                 swp_entry_t entry;
1141                 if (pte_swp_soft_dirty(pte))
1142                         flags |= PM_SOFT_DIRTY;
1143                 entry = pte_to_swp_entry(pte);
1144                 frame = swp_type(entry) |
1145                         (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1146                 flags |= PM_SWAP;
1147                 if (is_migration_entry(entry))
1148                         page = migration_entry_to_page(entry);
1149         }
1150
1151         if (page && !PageAnon(page))
1152                 flags |= PM_FILE;
1153         if (page && page_mapcount(page) == 1)
1154                 flags |= PM_MMAP_EXCLUSIVE;
1155         if (vma->vm_flags & VM_SOFTDIRTY)
1156                 flags |= PM_SOFT_DIRTY;
1157
1158         return make_pme(frame, flags);
1159 }
1160
1161 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1162                              struct mm_walk *walk)
1163 {
1164         struct vm_area_struct *vma = walk->vma;
1165         struct pagemapread *pm = walk->private;
1166         spinlock_t *ptl;
1167         pte_t *pte, *orig_pte;
1168         int err = 0;
1169
1170 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1171         if (pmd_trans_huge_lock(pmdp, vma, &ptl) == 1) {
1172                 u64 flags = 0, frame = 0;
1173                 pmd_t pmd = *pmdp;
1174
1175                 if ((vma->vm_flags & VM_SOFTDIRTY) || pmd_soft_dirty(pmd))
1176                         flags |= PM_SOFT_DIRTY;
1177
1178                 /*
1179                  * Currently pmd for thp is always present because thp
1180                  * can not be swapped-out, migrated, or HWPOISONed
1181                  * (split in such cases instead.)
1182                  * This if-check is just to prepare for future implementation.
1183                  */
1184                 if (pmd_present(pmd)) {
1185                         struct page *page = pmd_page(pmd);
1186
1187                         if (page_mapcount(page) == 1)
1188                                 flags |= PM_MMAP_EXCLUSIVE;
1189
1190                         flags |= PM_PRESENT;
1191                         if (pm->show_pfn)
1192                                 frame = pmd_pfn(pmd) +
1193                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1194                 }
1195
1196                 for (; addr != end; addr += PAGE_SIZE) {
1197                         pagemap_entry_t pme = make_pme(frame, flags);
1198
1199                         err = add_to_pagemap(addr, &pme, pm);
1200                         if (err)
1201                                 break;
1202                         if (pm->show_pfn && (flags & PM_PRESENT))
1203                                 frame++;
1204                 }
1205                 spin_unlock(ptl);
1206                 return err;
1207         }
1208
1209         if (pmd_trans_unstable(pmdp))
1210                 return 0;
1211 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1212
1213         /*
1214          * We can assume that @vma always points to a valid one and @end never
1215          * goes beyond vma->vm_end.
1216          */
1217         orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1218         for (; addr < end; pte++, addr += PAGE_SIZE) {
1219                 pagemap_entry_t pme;
1220
1221                 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1222                 err = add_to_pagemap(addr, &pme, pm);
1223                 if (err)
1224                         break;
1225         }
1226         pte_unmap_unlock(orig_pte, ptl);
1227
1228         cond_resched();
1229
1230         return err;
1231 }
1232
1233 #ifdef CONFIG_HUGETLB_PAGE
1234 /* This function walks within one hugetlb entry in the single call */
1235 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1236                                  unsigned long addr, unsigned long end,
1237                                  struct mm_walk *walk)
1238 {
1239         struct pagemapread *pm = walk->private;
1240         struct vm_area_struct *vma = walk->vma;
1241         u64 flags = 0, frame = 0;
1242         int err = 0;
1243         pte_t pte;
1244
1245         if (vma->vm_flags & VM_SOFTDIRTY)
1246                 flags |= PM_SOFT_DIRTY;
1247
1248         pte = huge_ptep_get(ptep);
1249         if (pte_present(pte)) {
1250                 struct page *page = pte_page(pte);
1251
1252                 if (!PageAnon(page))
1253                         flags |= PM_FILE;
1254
1255                 if (page_mapcount(page) == 1)
1256                         flags |= PM_MMAP_EXCLUSIVE;
1257
1258                 flags |= PM_PRESENT;
1259                 if (pm->show_pfn)
1260                         frame = pte_pfn(pte) +
1261                                 ((addr & ~hmask) >> PAGE_SHIFT);
1262         }
1263
1264         for (; addr != end; addr += PAGE_SIZE) {
1265                 pagemap_entry_t pme = make_pme(frame, flags);
1266
1267                 err = add_to_pagemap(addr, &pme, pm);
1268                 if (err)
1269                         return err;
1270                 if (pm->show_pfn && (flags & PM_PRESENT))
1271                         frame++;
1272         }
1273
1274         cond_resched();
1275
1276         return err;
1277 }
1278 #endif /* HUGETLB_PAGE */
1279
1280 /*
1281  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1282  *
1283  * For each page in the address space, this file contains one 64-bit entry
1284  * consisting of the following:
1285  *
1286  * Bits 0-54  page frame number (PFN) if present
1287  * Bits 0-4   swap type if swapped
1288  * Bits 5-54  swap offset if swapped
1289  * Bit  55    pte is soft-dirty (see Documentation/vm/soft-dirty.txt)
1290  * Bit  56    page exclusively mapped
1291  * Bits 57-60 zero
1292  * Bit  61    page is file-page or shared-anon
1293  * Bit  62    page swapped
1294  * Bit  63    page present
1295  *
1296  * If the page is not present but in swap, then the PFN contains an
1297  * encoding of the swap file number and the page's offset into the
1298  * swap. Unmapped pages return a null PFN. This allows determining
1299  * precisely which pages are mapped (or in swap) and comparing mapped
1300  * pages between processes.
1301  *
1302  * Efficient users of this interface will use /proc/pid/maps to
1303  * determine which areas of memory are actually mapped and llseek to
1304  * skip over unmapped regions.
1305  */
1306 static ssize_t pagemap_read(struct file *file, char __user *buf,
1307                             size_t count, loff_t *ppos)
1308 {
1309         struct mm_struct *mm = file->private_data;
1310         struct pagemapread pm;
1311         struct mm_walk pagemap_walk = {};
1312         unsigned long src;
1313         unsigned long svpfn;
1314         unsigned long start_vaddr;
1315         unsigned long end_vaddr;
1316         int ret = 0, copied = 0;
1317
1318         if (!mm || !atomic_inc_not_zero(&mm->mm_users))
1319                 goto out;
1320
1321         ret = -EINVAL;
1322         /* file position must be aligned */
1323         if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1324                 goto out_mm;
1325
1326         ret = 0;
1327         if (!count)
1328                 goto out_mm;
1329
1330         /* do not disclose physical addresses: attack vector */
1331         pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1332
1333         pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1334         pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY);
1335         ret = -ENOMEM;
1336         if (!pm.buffer)
1337                 goto out_mm;
1338
1339         pagemap_walk.pmd_entry = pagemap_pmd_range;
1340         pagemap_walk.pte_hole = pagemap_pte_hole;
1341 #ifdef CONFIG_HUGETLB_PAGE
1342         pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1343 #endif
1344         pagemap_walk.mm = mm;
1345         pagemap_walk.private = &pm;
1346
1347         src = *ppos;
1348         svpfn = src / PM_ENTRY_BYTES;
1349         start_vaddr = svpfn << PAGE_SHIFT;
1350         end_vaddr = mm->task_size;
1351
1352         /* watch out for wraparound */
1353         if (svpfn > mm->task_size >> PAGE_SHIFT)
1354                 start_vaddr = end_vaddr;
1355
1356         /*
1357          * The odds are that this will stop walking way
1358          * before end_vaddr, because the length of the
1359          * user buffer is tracked in "pm", and the walk
1360          * will stop when we hit the end of the buffer.
1361          */
1362         ret = 0;
1363         while (count && (start_vaddr < end_vaddr)) {
1364                 int len;
1365                 unsigned long end;
1366
1367                 pm.pos = 0;
1368                 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1369                 /* overflow ? */
1370                 if (end < start_vaddr || end > end_vaddr)
1371                         end = end_vaddr;
1372                 down_read(&mm->mmap_sem);
1373                 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1374                 up_read(&mm->mmap_sem);
1375                 start_vaddr = end;
1376
1377                 len = min(count, PM_ENTRY_BYTES * pm.pos);
1378                 if (copy_to_user(buf, pm.buffer, len)) {
1379                         ret = -EFAULT;
1380                         goto out_free;
1381                 }
1382                 copied += len;
1383                 buf += len;
1384                 count -= len;
1385         }
1386         *ppos += copied;
1387         if (!ret || ret == PM_END_OF_BUFFER)
1388                 ret = copied;
1389
1390 out_free:
1391         kfree(pm.buffer);
1392 out_mm:
1393         mmput(mm);
1394 out:
1395         return ret;
1396 }
1397
1398 static int pagemap_open(struct inode *inode, struct file *file)
1399 {
1400         struct mm_struct *mm;
1401
1402 #if defined(CONFIG_ARCH_ROCKCHIP) && defined(CONFIG_ANDROID)
1403         /*
1404          * For pass CTS
1405          * FileSystemPermissionTest: Assert /proc/self/pagemap not readable
1406          */
1407         /* do not disclose physical addresses: attack vector */
1408         if (!capable(CAP_SYS_ADMIN))
1409                 return -EPERM;
1410 #endif
1411
1412         mm = proc_mem_open(inode, PTRACE_MODE_READ);
1413         if (IS_ERR(mm))
1414                 return PTR_ERR(mm);
1415         file->private_data = mm;
1416         return 0;
1417 }
1418
1419 static int pagemap_release(struct inode *inode, struct file *file)
1420 {
1421         struct mm_struct *mm = file->private_data;
1422
1423         if (mm)
1424                 mmdrop(mm);
1425         return 0;
1426 }
1427
1428 const struct file_operations proc_pagemap_operations = {
1429         .llseek         = mem_lseek, /* borrow this */
1430         .read           = pagemap_read,
1431         .open           = pagemap_open,
1432         .release        = pagemap_release,
1433 };
1434 #endif /* CONFIG_PROC_PAGE_MONITOR */
1435
1436 #ifdef CONFIG_NUMA
1437
1438 struct numa_maps {
1439         unsigned long pages;
1440         unsigned long anon;
1441         unsigned long active;
1442         unsigned long writeback;
1443         unsigned long mapcount_max;
1444         unsigned long dirty;
1445         unsigned long swapcache;
1446         unsigned long node[MAX_NUMNODES];
1447 };
1448
1449 struct numa_maps_private {
1450         struct proc_maps_private proc_maps;
1451         struct numa_maps md;
1452 };
1453
1454 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1455                         unsigned long nr_pages)
1456 {
1457         int count = page_mapcount(page);
1458
1459         md->pages += nr_pages;
1460         if (pte_dirty || PageDirty(page))
1461                 md->dirty += nr_pages;
1462
1463         if (PageSwapCache(page))
1464                 md->swapcache += nr_pages;
1465
1466         if (PageActive(page) || PageUnevictable(page))
1467                 md->active += nr_pages;
1468
1469         if (PageWriteback(page))
1470                 md->writeback += nr_pages;
1471
1472         if (PageAnon(page))
1473                 md->anon += nr_pages;
1474
1475         if (count > md->mapcount_max)
1476                 md->mapcount_max = count;
1477
1478         md->node[page_to_nid(page)] += nr_pages;
1479 }
1480
1481 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1482                 unsigned long addr)
1483 {
1484         struct page *page;
1485         int nid;
1486
1487         if (!pte_present(pte))
1488                 return NULL;
1489
1490         page = vm_normal_page(vma, addr, pte);
1491         if (!page)
1492                 return NULL;
1493
1494         if (PageReserved(page))
1495                 return NULL;
1496
1497         nid = page_to_nid(page);
1498         if (!node_isset(nid, node_states[N_MEMORY]))
1499                 return NULL;
1500
1501         return page;
1502 }
1503
1504 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1505 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1506                                               struct vm_area_struct *vma,
1507                                               unsigned long addr)
1508 {
1509         struct page *page;
1510         int nid;
1511
1512         if (!pmd_present(pmd))
1513                 return NULL;
1514
1515         page = vm_normal_page_pmd(vma, addr, pmd);
1516         if (!page)
1517                 return NULL;
1518
1519         if (PageReserved(page))
1520                 return NULL;
1521
1522         nid = page_to_nid(page);
1523         if (!node_isset(nid, node_states[N_MEMORY]))
1524                 return NULL;
1525
1526         return page;
1527 }
1528 #endif
1529
1530 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1531                 unsigned long end, struct mm_walk *walk)
1532 {
1533         struct numa_maps *md = walk->private;
1534         struct vm_area_struct *vma = walk->vma;
1535         spinlock_t *ptl;
1536         pte_t *orig_pte;
1537         pte_t *pte;
1538
1539 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1540         if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
1541                 struct page *page;
1542
1543                 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1544                 if (page)
1545                         gather_stats(page, md, pmd_dirty(*pmd),
1546                                      HPAGE_PMD_SIZE/PAGE_SIZE);
1547                 spin_unlock(ptl);
1548                 return 0;
1549         }
1550
1551         if (pmd_trans_unstable(pmd))
1552                 return 0;
1553 #endif
1554         orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1555         do {
1556                 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1557                 if (!page)
1558                         continue;
1559                 gather_stats(page, md, pte_dirty(*pte), 1);
1560
1561         } while (pte++, addr += PAGE_SIZE, addr != end);
1562         pte_unmap_unlock(orig_pte, ptl);
1563         return 0;
1564 }
1565 #ifdef CONFIG_HUGETLB_PAGE
1566 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1567                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1568 {
1569         pte_t huge_pte = huge_ptep_get(pte);
1570         struct numa_maps *md;
1571         struct page *page;
1572
1573         if (!pte_present(huge_pte))
1574                 return 0;
1575
1576         page = pte_page(huge_pte);
1577         if (!page)
1578                 return 0;
1579
1580         md = walk->private;
1581         gather_stats(page, md, pte_dirty(huge_pte), 1);
1582         return 0;
1583 }
1584
1585 #else
1586 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1587                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1588 {
1589         return 0;
1590 }
1591 #endif
1592
1593 /*
1594  * Display pages allocated per node and memory policy via /proc.
1595  */
1596 static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1597 {
1598         struct numa_maps_private *numa_priv = m->private;
1599         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1600         struct vm_area_struct *vma = v;
1601         struct numa_maps *md = &numa_priv->md;
1602         struct file *file = vma->vm_file;
1603         struct mm_struct *mm = vma->vm_mm;
1604         struct mm_walk walk = {
1605                 .hugetlb_entry = gather_hugetlb_stats,
1606                 .pmd_entry = gather_pte_stats,
1607                 .private = md,
1608                 .mm = mm,
1609         };
1610         struct mempolicy *pol;
1611         char buffer[64];
1612         int nid;
1613
1614         if (!mm)
1615                 return 0;
1616
1617         /* Ensure we start with an empty set of numa_maps statistics. */
1618         memset(md, 0, sizeof(*md));
1619
1620         pol = __get_vma_policy(vma, vma->vm_start);
1621         if (pol) {
1622                 mpol_to_str(buffer, sizeof(buffer), pol);
1623                 mpol_cond_put(pol);
1624         } else {
1625                 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1626         }
1627
1628         seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1629
1630         if (file) {
1631                 seq_puts(m, " file=");
1632                 seq_file_path(m, file, "\n\t= ");
1633         } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1634                 seq_puts(m, " heap");
1635         } else if (is_stack(proc_priv, vma, is_pid)) {
1636                 seq_puts(m, " stack");
1637         }
1638
1639         if (is_vm_hugetlb_page(vma))
1640                 seq_puts(m, " huge");
1641
1642         /* mmap_sem is held by m_start */
1643         walk_page_vma(vma, &walk);
1644
1645         if (!md->pages)
1646                 goto out;
1647
1648         if (md->anon)
1649                 seq_printf(m, " anon=%lu", md->anon);
1650
1651         if (md->dirty)
1652                 seq_printf(m, " dirty=%lu", md->dirty);
1653
1654         if (md->pages != md->anon && md->pages != md->dirty)
1655                 seq_printf(m, " mapped=%lu", md->pages);
1656
1657         if (md->mapcount_max > 1)
1658                 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1659
1660         if (md->swapcache)
1661                 seq_printf(m, " swapcache=%lu", md->swapcache);
1662
1663         if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1664                 seq_printf(m, " active=%lu", md->active);
1665
1666         if (md->writeback)
1667                 seq_printf(m, " writeback=%lu", md->writeback);
1668
1669         for_each_node_state(nid, N_MEMORY)
1670                 if (md->node[nid])
1671                         seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1672
1673         seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1674 out:
1675         seq_putc(m, '\n');
1676         m_cache_vma(m, vma);
1677         return 0;
1678 }
1679
1680 static int show_pid_numa_map(struct seq_file *m, void *v)
1681 {
1682         return show_numa_map(m, v, 1);
1683 }
1684
1685 static int show_tid_numa_map(struct seq_file *m, void *v)
1686 {
1687         return show_numa_map(m, v, 0);
1688 }
1689
1690 static const struct seq_operations proc_pid_numa_maps_op = {
1691         .start  = m_start,
1692         .next   = m_next,
1693         .stop   = m_stop,
1694         .show   = show_pid_numa_map,
1695 };
1696
1697 static const struct seq_operations proc_tid_numa_maps_op = {
1698         .start  = m_start,
1699         .next   = m_next,
1700         .stop   = m_stop,
1701         .show   = show_tid_numa_map,
1702 };
1703
1704 static int numa_maps_open(struct inode *inode, struct file *file,
1705                           const struct seq_operations *ops)
1706 {
1707         return proc_maps_open(inode, file, ops,
1708                                 sizeof(struct numa_maps_private));
1709 }
1710
1711 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1712 {
1713         return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1714 }
1715
1716 static int tid_numa_maps_open(struct inode *inode, struct file *file)
1717 {
1718         return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1719 }
1720
1721 const struct file_operations proc_pid_numa_maps_operations = {
1722         .open           = pid_numa_maps_open,
1723         .read           = seq_read,
1724         .llseek         = seq_lseek,
1725         .release        = proc_map_release,
1726 };
1727
1728 const struct file_operations proc_tid_numa_maps_operations = {
1729         .open           = tid_numa_maps_open,
1730         .read           = seq_read,
1731         .llseek         = seq_lseek,
1732         .release        = proc_map_release,
1733 };
1734 #endif /* CONFIG_NUMA */