fs/proc/task_mmu.c

   1 #include <linux/mm.h>
   2 #include <linux/hugetlb.h>
   3 #include <linux/mount.h>
   4 #include <linux/seq_file.h>
   5 #include <linux/highmem.h>
   6 #include <linux/ptrace.h>
   7 #include <linux/pagemap.h>
   8 #include <linux/mempolicy.h>
   9
  10 #include <asm/elf.h>
  11 #include <asm/uaccess.h>
  12 #include <asm/tlbflush.h>
  13 #include "internal.h"
  14
  15 char *task_mem(struct mm_struct *mm, char *buffer)
  16 {
  17         unsigned long data, text, lib;
  18         unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
  19
  20         /*
  21          * Note: to minimize their overhead, mm maintains hiwater_vm and
  22          * hiwater_rss only when about to *lower* total_vm or rss.  Any
  23          * collector of these hiwater stats must therefore get total_vm
  24          * and rss too, which will usually be the higher.  Barriers? not
  25          * worth the effort, such snapshots can always be inconsistent.
  26          */
  27         hiwater_vm = total_vm = mm->total_vm;
  28         if (hiwater_vm < mm->hiwater_vm)
  29                 hiwater_vm = mm->hiwater_vm;
  30         hiwater_rss = total_rss = get_mm_rss(mm);
  31         if (hiwater_rss < mm->hiwater_rss)
  32                 hiwater_rss = mm->hiwater_rss;
  33
  34         data = mm->total_vm - mm->shared_vm - mm->stack_vm;
  35         text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
  36         lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
  37         buffer += sprintf(buffer,
  38                 "VmPeak:\t%8lu kB\n"
  39                 "VmSize:\t%8lu kB\n"
  40                 "VmLck:\t%8lu kB\n"
  41                 "VmHWM:\t%8lu kB\n"
  42                 "VmRSS:\t%8lu kB\n"
  43                 "VmData:\t%8lu kB\n"
  44                 "VmStk:\t%8lu kB\n"
  45                 "VmExe:\t%8lu kB\n"
  46                 "VmLib:\t%8lu kB\n"
  47                 "VmPTE:\t%8lu kB\n",
  48                 hiwater_vm << (PAGE_SHIFT-10),
  49                 (total_vm - mm->reserved_vm) << (PAGE_SHIFT-10),
  50                 mm->locked_vm << (PAGE_SHIFT-10),
  51                 hiwater_rss << (PAGE_SHIFT-10),
  52                 total_rss << (PAGE_SHIFT-10),
  53                 data << (PAGE_SHIFT-10),
  54                 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
  55                 (PTRS_PER_PTE*sizeof(pte_t)*mm->nr_ptes) >> 10);
  56         return buffer;
  57 }
  58
  59 unsigned long task_vsize(struct mm_struct *mm)
  60 {
  61         return PAGE_SIZE * mm->total_vm;
  62 }
  63
  64 int task_statm(struct mm_struct *mm, int *shared, int *text,
  65                int *data, int *resident)
  66 {
  67         *shared = get_mm_counter(mm, file_rss);
  68         *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
  69                                                                 >> PAGE_SHIFT;
  70         *data = mm->total_vm - mm->shared_vm;
  71         *resident = *shared + get_mm_counter(mm, anon_rss);
  72         return mm->total_vm;
  73 }
  74
  75 int proc_exe_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt)
  76 {
  77         struct vm_area_struct * vma;
  78         int result = -ENOENT;
  79         struct task_struct *task = get_proc_task(inode);
  80         struct mm_struct * mm = NULL;
  81
  82         if (task) {
  83                 mm = get_task_mm(task);
  84                 put_task_struct(task);
  85         }
  86         if (!mm)
  87                 goto out;
  88         down_read(&mm->mmap_sem);
  89
  90         vma = mm->mmap;
  91         while (vma) {
  92                 if ((vma->vm_flags & VM_EXECUTABLE) && vma->vm_file)
  93                         break;
  94                 vma = vma->vm_next;
  95         }
  96
  97         if (vma) {
  98                 *mnt = mntget(vma->vm_file->f_path.mnt);
  99                 *dentry = dget(vma->vm_file->f_path.dentry);
 100                 result = 0;
 101         }
 102
 103         up_read(&mm->mmap_sem);
 104         mmput(mm);
 105 out:
 106         return result;
 107 }
 108
 109 static void pad_len_spaces(struct seq_file *m, int len)
 110 {
 111         len = 25 + sizeof(void*) * 6 - len;
 112         if (len < 1)
 113                 len = 1;
 114         seq_printf(m, "%*c", len, ' ');
 115 }
 116
 117 /*
 118  * Proportional Set Size(PSS): my share of RSS.
 119  *
 120  * PSS of a process is the count of pages it has in memory, where each
 121  * page is divided by the number of processes sharing it.  So if a
 122  * process has 1000 pages all to itself, and 1000 shared with one other
 123  * process, its PSS will be 1500.
 124  *
 125  * To keep (accumulated) division errors low, we adopt a 64bit
 126  * fixed-point pss counter to minimize division errors. So (pss >>
 127  * PSS_SHIFT) would be the real byte count.
 128  *
 129  * A shift of 12 before division means (assuming 4K page size):
 130  *      - 1M 3-user-pages add up to 8KB errors;
 131  *      - supports mapcount up to 2^24, or 16M;
 132  *      - supports PSS up to 2^52 bytes, or 4PB.
 133  */
 134 #define PSS_SHIFT 12
 135
 136 struct mem_size_stats
 137 {
 138         unsigned long resident;
 139         unsigned long shared_clean;
 140         unsigned long shared_dirty;
 141         unsigned long private_clean;
 142         unsigned long private_dirty;
 143         unsigned long referenced;
 144         u64 pss;
 145 };
 146
 147 struct pmd_walker {
 148         struct vm_area_struct *vma;
 149         void *private;
 150         void (*action)(struct vm_area_struct *, pmd_t *, unsigned long,
 151                        unsigned long, void *);
 152 };
 153
 154 static int show_map_internal(struct seq_file *m, void *v, struct mem_size_stats *mss)
 155 {
 156         struct proc_maps_private *priv = m->private;
 157         struct task_struct *task = priv->task;
 158         struct vm_area_struct *vma = v;
 159         struct mm_struct *mm = vma->vm_mm;
 160         struct file *file = vma->vm_file;
 161         int flags = vma->vm_flags;
 162         unsigned long ino = 0;
 163         dev_t dev = 0;
 164         int len;
 165
 166         if (maps_protect && !ptrace_may_attach(task))
 167                 return -EACCES;
 168
 169         if (file) {
 170                 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
 171                 dev = inode->i_sb->s_dev;
 172                 ino = inode->i_ino;
 173         }
 174
 175         seq_printf(m, "%08lx-%08lx %c%c%c%c %08lx %02x:%02x %lu %n",
 176                         vma->vm_start,
 177                         vma->vm_end,
 178                         flags & VM_READ ? 'r' : '-',
 179                         flags & VM_WRITE ? 'w' : '-',
 180                         flags & VM_EXEC ? 'x' : '-',
 181                         flags & VM_MAYSHARE ? 's' : 'p',
 182                         vma->vm_pgoff << PAGE_SHIFT,
 183                         MAJOR(dev), MINOR(dev), ino, &len);
 184
 185         /*
 186          * Print the dentry name for named mappings, and a
 187          * special [heap] marker for the heap:
 188          */
 189         if (file) {
 190                 pad_len_spaces(m, len);
 191                 seq_path(m, file->f_path.mnt, file->f_path.dentry, "\n");
 192         } else {
 193                 const char *name = arch_vma_name(vma);
 194                 if (!name) {
 195                         if (mm) {
 196                                 if (vma->vm_start <= mm->start_brk &&
 197                                                 vma->vm_end >= mm->brk) {
 198                                         name = "[heap]";
 199                                 } else if (vma->vm_start <= mm->start_stack &&
 200                                            vma->vm_end >= mm->start_stack) {
 201                                         name = "[stack]";
 202                                 }
 203                         } else {
 204                                 name = "[vdso]";
 205                         }
 206                 }
 207                 if (name) {
 208                         pad_len_spaces(m, len);
 209                         seq_puts(m, name);
 210                 }
 211         }
 212         seq_putc(m, '\n');
 213
 214         if (mss)
 215                 seq_printf(m,
 216                            "Size:           %8lu kB\n"
 217                            "Rss:            %8lu kB\n"
 218                            "Pss:            %8lu kB\n"
 219                            "Shared_Clean:   %8lu kB\n"
 220                            "Shared_Dirty:   %8lu kB\n"
 221                            "Private_Clean:  %8lu kB\n"
 222                            "Private_Dirty:  %8lu kB\n"
 223                            "Referenced:     %8lu kB\n",
 224                            (vma->vm_end - vma->vm_start) >> 10,
 225                            mss->resident >> 10,
 226                            (unsigned long)(mss->pss >> (10 + PSS_SHIFT)),
 227                            mss->shared_clean  >> 10,
 228                            mss->shared_dirty  >> 10,
 229                            mss->private_clean >> 10,
 230                            mss->private_dirty >> 10,
 231                            mss->referenced >> 10);
 232
 233         if (m->count < m->size)  /* vma is copied successfully */
 234                 m->version = (vma != get_gate_vma(task))? vma->vm_start: 0;
 235         return 0;
 236 }
 237
 238 static int show_map(struct seq_file *m, void *v)
 239 {
 240         return show_map_internal(m, v, NULL);
 241 }
 242
 243 static void smaps_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
 244                             unsigned long addr, unsigned long end,
 245                             void *private)
 246 {
 247         struct mem_size_stats *mss = private;
 248         pte_t *pte, ptent;
 249         spinlock_t *ptl;
 250         struct page *page;
 251         int mapcount;
 252
 253         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 254         for (; addr != end; pte++, addr += PAGE_SIZE) {
 255                 ptent = *pte;
 256                 if (!pte_present(ptent))
 257                         continue;
 258
 259                 mss->resident += PAGE_SIZE;
 260
 261                 page = vm_normal_page(vma, addr, ptent);
 262                 if (!page)
 263                         continue;
 264
 265                 /* Accumulate the size in pages that have been accessed. */
 266                 if (pte_young(ptent) || PageReferenced(page))
 267                         mss->referenced += PAGE_SIZE;
 268                 mapcount = page_mapcount(page);
 269                 if (mapcount >= 2) {
 270                         if (pte_dirty(ptent))
 271                                 mss->shared_dirty += PAGE_SIZE;
 272                         else
 273                                 mss->shared_clean += PAGE_SIZE;
 274                         mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount;
 275                 } else {
 276                         if (pte_dirty(ptent))
 277                                 mss->private_dirty += PAGE_SIZE;
 278                         else
 279                                 mss->private_clean += PAGE_SIZE;
 280                         mss->pss += (PAGE_SIZE << PSS_SHIFT);
 281                 }
 282         }
 283         pte_unmap_unlock(pte - 1, ptl);
 284         cond_resched();
 285 }
 286
 287 static void clear_refs_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
 288                                  unsigned long addr, unsigned long end,
 289                                  void *private)
 290 {
 291         pte_t *pte, ptent;
 292         spinlock_t *ptl;
 293         struct page *page;
 294
 295         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 296         for (; addr != end; pte++, addr += PAGE_SIZE) {
 297                 ptent = *pte;
 298                 if (!pte_present(ptent))
 299                         continue;
 300
 301                 page = vm_normal_page(vma, addr, ptent);
 302                 if (!page)
 303                         continue;
 304
 305                 /* Clear accessed and referenced bits. */
 306                 ptep_test_and_clear_young(vma, addr, pte);
 307                 ClearPageReferenced(page);
 308         }
 309         pte_unmap_unlock(pte - 1, ptl);
 310         cond_resched();
 311 }
 312
 313 static inline void walk_pmd_range(struct pmd_walker *walker, pud_t *pud,
 314                                   unsigned long addr, unsigned long end)
 315 {
 316         pmd_t *pmd;
 317         unsigned long next;
 318
 319         for (pmd = pmd_offset(pud, addr); addr != end;
 320              pmd++, addr = next) {
 321                 next = pmd_addr_end(addr, end);
 322                 if (pmd_none_or_clear_bad(pmd))
 323                         continue;
 324                 walker->action(walker->vma, pmd, addr, next, walker->private);
 325         }
 326 }
 327
 328 static inline void walk_pud_range(struct pmd_walker *walker, pgd_t *pgd,
 329                                   unsigned long addr, unsigned long end)
 330 {
 331         pud_t *pud;
 332         unsigned long next;
 333
 334         for (pud = pud_offset(pgd, addr); addr != end;
 335              pud++, addr = next) {
 336                 next = pud_addr_end(addr, end);
 337                 if (pud_none_or_clear_bad(pud))
 338                         continue;
 339                 walk_pmd_range(walker, pud, addr, next);
 340         }
 341 }
 342
 343 /*
 344  * walk_page_range - walk the page tables of a VMA with a callback
 345  * @vma - VMA to walk
 346  * @action - callback invoked for every bottom-level (PTE) page table
 347  * @private - private data passed to the callback function
 348  *
 349  * Recursively walk the page table for the memory area in a VMA, calling
 350  * a callback for every bottom-level (PTE) page table.
 351  */
 352 static inline void walk_page_range(struct vm_area_struct *vma,
 353                                    void (*action)(struct vm_area_struct *,
 354                                                   pmd_t *, unsigned long,
 355                                                   unsigned long, void *),
 356                                    void *private)
 357 {
 358         unsigned long addr = vma->vm_start;
 359         unsigned long end = vma->vm_end;
 360         struct pmd_walker walker = {
 361                 .vma            = vma,
 362                 .private        = private,
 363                 .action         = action,
 364         };
 365         pgd_t *pgd;
 366         unsigned long next;
 367
 368         for (pgd = pgd_offset(vma->vm_mm, addr); addr != end;
 369              pgd++, addr = next) {
 370                 next = pgd_addr_end(addr, end);
 371                 if (pgd_none_or_clear_bad(pgd))
 372                         continue;
 373                 walk_pud_range(&walker, pgd, addr, next);
 374         }
 375 }
 376
 377 static int show_smap(struct seq_file *m, void *v)
 378 {
 379         struct vm_area_struct *vma = v;
 380         struct mem_size_stats mss;
 381
 382         memset(&mss, 0, sizeof mss);
 383         if (vma->vm_mm && !is_vm_hugetlb_page(vma))
 384                 walk_page_range(vma, smaps_pte_range, &mss);
 385         return show_map_internal(m, v, &mss);
 386 }
 387
 388 void clear_refs_smap(struct mm_struct *mm)
 389 {
 390         struct vm_area_struct *vma;
 391
 392         down_read(&mm->mmap_sem);
 393         for (vma = mm->mmap; vma; vma = vma->vm_next)
 394                 if (vma->vm_mm && !is_vm_hugetlb_page(vma))
 395                         walk_page_range(vma, clear_refs_pte_range, NULL);
 396         flush_tlb_mm(mm);
 397         up_read(&mm->mmap_sem);
 398 }
 399
 400 static void *m_start(struct seq_file *m, loff_t *pos)
 401 {
 402         struct proc_maps_private *priv = m->private;
 403         unsigned long last_addr = m->version;
 404         struct mm_struct *mm;
 405         struct vm_area_struct *vma, *tail_vma = NULL;
 406         loff_t l = *pos;
 407
 408         /* Clear the per syscall fields in priv */
 409         priv->task = NULL;
 410         priv->tail_vma = NULL;
 411
 412         /*
 413          * We remember last_addr rather than next_addr to hit with
 414          * mmap_cache most of the time. We have zero last_addr at
 415          * the beginning and also after lseek. We will have -1 last_addr
 416          * after the end of the vmas.
 417          */
 418
 419         if (last_addr == -1UL)
 420                 return NULL;
 421
 422         priv->task = get_pid_task(priv->pid, PIDTYPE_PID);
 423         if (!priv->task)
 424                 return NULL;
 425
 426         mm = mm_for_maps(priv->task);
 427         if (!mm)
 428                 return NULL;
 429
 430         priv->tail_vma = tail_vma = get_gate_vma(priv->task);
 431
 432         /* Start with last addr hint */
 433         if (last_addr && (vma = find_vma(mm, last_addr))) {
 434                 vma = vma->vm_next;
 435                 goto out;
 436         }
 437
 438         /*
 439          * Check the vma index is within the range and do
 440          * sequential scan until m_index.
 441          */
 442         vma = NULL;
 443         if ((unsigned long)l < mm->map_count) {
 444                 vma = mm->mmap;
 445                 while (l-- && vma)
 446                         vma = vma->vm_next;
 447                 goto out;
 448         }
 449
 450         if (l != mm->map_count)
 451                 tail_vma = NULL; /* After gate vma */
 452
 453 out:
 454         if (vma)
 455                 return vma;
 456
 457         /* End of vmas has been reached */
 458         m->version = (tail_vma != NULL)? 0: -1UL;
 459         up_read(&mm->mmap_sem);
 460         mmput(mm);
 461         return tail_vma;
 462 }
 463
 464 static void vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma)
 465 {
 466         if (vma && vma != priv->tail_vma) {
 467                 struct mm_struct *mm = vma->vm_mm;
 468                 up_read(&mm->mmap_sem);
 469                 mmput(mm);
 470         }
 471 }
 472
 473 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
 474 {
 475         struct proc_maps_private *priv = m->private;
 476         struct vm_area_struct *vma = v;
 477         struct vm_area_struct *tail_vma = priv->tail_vma;
 478
 479         (*pos)++;
 480         if (vma && (vma != tail_vma) && vma->vm_next)
 481                 return vma->vm_next;
 482         vma_stop(priv, vma);
 483         return (vma != tail_vma)? tail_vma: NULL;
 484 }
 485
 486 static void m_stop(struct seq_file *m, void *v)
 487 {
 488         struct proc_maps_private *priv = m->private;
 489         struct vm_area_struct *vma = v;
 490
 491         vma_stop(priv, vma);
 492         if (priv->task)
 493                 put_task_struct(priv->task);
 494 }
 495
 496 static struct seq_operations proc_pid_maps_op = {
 497         .start  = m_start,
 498         .next   = m_next,
 499         .stop   = m_stop,
 500         .show   = show_map
 501 };
 502
 503 static struct seq_operations proc_pid_smaps_op = {
 504         .start  = m_start,
 505         .next   = m_next,
 506         .stop   = m_stop,
 507         .show   = show_smap
 508 };
 509
 510 static int do_maps_open(struct inode *inode, struct file *file,
 511                         struct seq_operations *ops)
 512 {
 513         struct proc_maps_private *priv;
 514         int ret = -ENOMEM;
 515         priv = kzalloc(sizeof(*priv), GFP_KERNEL);
 516         if (priv) {
 517                 priv->pid = proc_pid(inode);
 518                 ret = seq_open(file, ops);
 519                 if (!ret) {
 520                         struct seq_file *m = file->private_data;
 521                         m->private = priv;
 522                 } else {
 523                         kfree(priv);
 524                 }
 525         }
 526         return ret;
 527 }
 528
 529 static int maps_open(struct inode *inode, struct file *file)
 530 {
 531         return do_maps_open(inode, file, &proc_pid_maps_op);
 532 }
 533
 534 const struct file_operations proc_maps_operations = {
 535         .open           = maps_open,
 536         .read           = seq_read,
 537         .llseek         = seq_lseek,
 538         .release        = seq_release_private,
 539 };
 540
 541 #ifdef CONFIG_NUMA
 542 extern int show_numa_map(struct seq_file *m, void *v);
 543
 544 static int show_numa_map_checked(struct seq_file *m, void *v)
 545 {
 546         struct proc_maps_private *priv = m->private;
 547         struct task_struct *task = priv->task;
 548
 549         if (maps_protect && !ptrace_may_attach(task))
 550                 return -EACCES;
 551
 552         return show_numa_map(m, v);
 553 }
 554
 555 static struct seq_operations proc_pid_numa_maps_op = {
 556         .start  = m_start,
 557         .next   = m_next,
 558         .stop   = m_stop,
 559         .show   = show_numa_map_checked
 560 };
 561
 562 static int numa_maps_open(struct inode *inode, struct file *file)
 563 {
 564         return do_maps_open(inode, file, &proc_pid_numa_maps_op);
 565 }
 566
 567 const struct file_operations proc_numa_maps_operations = {
 568         .open           = numa_maps_open,
 569         .read           = seq_read,
 570         .llseek         = seq_lseek,
 571         .release        = seq_release_private,
 572 };
 573 #endif
 574
 575 static int smaps_open(struct inode *inode, struct file *file)
 576 {
 577         return do_maps_open(inode, file, &proc_pid_smaps_op);
 578 }
 579
 580 const struct file_operations proc_smaps_operations = {
 581         .open           = smaps_open,
 582         .read           = seq_read,
 583         .llseek         = seq_lseek,
 584         .release        = seq_release_private,
 585 };