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