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/pagemap.h>
   7 #include <linux/mempolicy.h>
   8
   9 #include <asm/elf.h>
  10 #include <asm/uaccess.h>
  11 #include <asm/tlbflush.h>
  12 #include "internal.h"
  13
  14 char *task_mem(struct mm_struct *mm, char *buffer)
  15 {
  16         unsigned long data, text, lib;
  17         unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
  18
  19         /*
  20          * Note: to minimize their overhead, mm maintains hiwater_vm and
  21          * hiwater_rss only when about to *lower* total_vm or rss.  Any
  22          * collector of these hiwater stats must therefore get total_vm
  23          * and rss too, which will usually be the higher.  Barriers? not
  24          * worth the effort, such snapshots can always be inconsistent.
  25          */
  26         hiwater_vm = total_vm = mm->total_vm;
  27         if (hiwater_vm < mm->hiwater_vm)
  28                 hiwater_vm = mm->hiwater_vm;
  29         hiwater_rss = total_rss = get_mm_rss(mm);
  30         if (hiwater_rss < mm->hiwater_rss)
  31                 hiwater_rss = mm->hiwater_rss;
  32
  33         data = mm->total_vm - mm->shared_vm - mm->stack_vm;
  34         text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
  35         lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
  36         buffer += sprintf(buffer,
  37                 "VmPeak:\t%8lu kB\n"
  38                 "VmSize:\t%8lu kB\n"
  39                 "VmLck:\t%8lu kB\n"
  40                 "VmHWM:\t%8lu kB\n"
  41                 "VmRSS:\t%8lu kB\n"
  42                 "VmData:\t%8lu kB\n"
  43                 "VmStk:\t%8lu kB\n"
  44                 "VmExe:\t%8lu kB\n"
  45                 "VmLib:\t%8lu kB\n"
  46                 "VmPTE:\t%8lu kB\n"
  47                 "StaBrk:\t%08lx kB\n"
  48                 "Brk:\t%08lx kB\n"
  49                 "StaStk:\t%08lx kB\n"
  50                 ,
  51                 hiwater_vm << (PAGE_SHIFT-10),
  52                 (total_vm - mm->reserved_vm) << (PAGE_SHIFT-10),
  53                 mm->locked_vm << (PAGE_SHIFT-10),
  54                 hiwater_rss << (PAGE_SHIFT-10),
  55                 total_rss << (PAGE_SHIFT-10),
  56                 data << (PAGE_SHIFT-10),
  57                 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
  58                 (PTRS_PER_PTE*sizeof(pte_t)*mm->nr_ptes) >> 10,
  59                 mm->start_brk, mm->brk, mm->start_stack);
  60 #ifdef __i386__
  61         if (!nx_enabled)
  62                 buffer += sprintf(buffer,
  63                         "ExecLim:\t%08lx\n", mm->context.exec_limit);
  64 #endif
  65         return buffer;
  66 }
  67
  68 unsigned long task_vsize(struct mm_struct *mm)
  69 {
  70         return PAGE_SIZE * mm->total_vm;
  71 }
  72
  73 int task_statm(struct mm_struct *mm, int *shared, int *text,
  74                int *data, int *resident)
  75 {
  76         *shared = get_mm_counter(mm, file_rss);
  77         *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
  78                                                                 >> PAGE_SHIFT;
  79         *data = mm->total_vm - mm->shared_vm;
  80         *resident = *shared + get_mm_counter(mm, anon_rss);
  81         return mm->total_vm;
  82 }
  83
  84 int proc_exe_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt)
  85 {
  86         struct vm_area_struct * vma;
  87         int result = -ENOENT;
  88         struct task_struct *task = get_proc_task(inode);
  89         struct mm_struct * mm = NULL;
  90
  91         if (task) {
  92                 mm = get_task_mm(task);
  93                 put_task_struct(task);
  94         }
  95         if (!mm)
  96                 goto out;
  97         down_read(&mm->mmap_sem);
  98
  99         vma = mm->mmap;
 100         while (vma) {
 101                 if ((vma->vm_flags & VM_EXECUTABLE) && vma->vm_file)
 102                         break;
 103                 vma = vma->vm_next;
 104         }
 105
 106         if (vma) {
 107                 *mnt = mntget(vma->vm_file->f_path.mnt);
 108                 *dentry = dget(vma->vm_file->f_path.dentry);
 109                 result = 0;
 110         }
 111
 112         up_read(&mm->mmap_sem);
 113         mmput(mm);
 114 out:
 115         return result;
 116 }
 117
 118 static void pad_len_spaces(struct seq_file *m, int len)
 119 {
 120         len = 25 + sizeof(void*) * 6 - len;
 121         if (len < 1)
 122                 len = 1;
 123         seq_printf(m, "%*c", len, ' ');
 124 }
 125
 126 struct mem_size_stats
 127 {
 128         unsigned long resident;
 129         unsigned long shared_clean;
 130         unsigned long shared_dirty;
 131         unsigned long private_clean;
 132         unsigned long private_dirty;
 133 };
 134
 135 static int show_map_internal(struct seq_file *m, void *v, struct mem_size_stats *mss)
 136 {
 137         struct proc_maps_private *priv = m->private;
 138         struct task_struct *task = priv->task;
 139 #ifdef __i386__
 140         struct mm_struct *tmm = get_task_mm(task);
 141 #endif
 142         struct vm_area_struct *vma = v;
 143         struct mm_struct *mm = vma->vm_mm;
 144         struct file *file = vma->vm_file;
 145         int flags = vma->vm_flags;
 146         unsigned long ino = 0;
 147         dev_t dev = 0;
 148         int len;
 149
 150         if (file) {
 151                 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
 152                 dev = inode->i_sb->s_dev;
 153                 ino = inode->i_ino;
 154         }
 155
 156         seq_printf(m, "%08lx-%08lx %c%c%c%c %08lx %02x:%02x %lu %n",
 157                         vma->vm_start,
 158                         vma->vm_end,
 159                         flags & VM_READ ? 'r' : '-',
 160                         flags & VM_WRITE ? 'w' : '-',
 161                         (flags & VM_EXEC
 162 #ifdef __i386__
 163                                 || (!nx_enabled && tmm &&
 164                                 (vma->vm_start < tmm->context.exec_limit))
 165 #endif
 166                         )
 167                                 ? 'x' : '-',
 168                         flags & VM_MAYSHARE ? 's' : 'p',
 169                         vma->vm_pgoff << PAGE_SHIFT,
 170                         MAJOR(dev), MINOR(dev), ino, &len);
 171 #ifdef __i386__
 172         if (tmm)
 173                 mmput(tmm);
 174 #endif
 175
 176         /*
 177          * Print the dentry name for named mappings, and a
 178          * special [heap] marker for the heap:
 179          */
 180         if (file) {
 181                 pad_len_spaces(m, len);
 182                 seq_path(m, file->f_path.mnt, file->f_path.dentry, "\n");
 183         } else {
 184                 const char *name = arch_vma_name(vma);
 185                 if (!name) {
 186                         if (mm) {
 187                                 if (vma->vm_start <= mm->start_brk &&
 188                                                 vma->vm_end >= mm->brk) {
 189                                         name = "[heap]";
 190                                 } else if (vma->vm_start <= mm->start_stack &&
 191                                            vma->vm_end >= mm->start_stack) {
 192                                         name = "[stack]";
 193                                 }
 194                         } else {
 195                                 name = "[vdso]";
 196                         }
 197                 }
 198                 if (name) {
 199                         pad_len_spaces(m, len);
 200                         seq_puts(m, name);
 201                 }
 202         }
 203         seq_putc(m, '\n');
 204
 205         if (mss)
 206                 seq_printf(m,
 207                            "Size:          %8lu kB\n"
 208                            "Rss:           %8lu kB\n"
 209                            "Shared_Clean:  %8lu kB\n"
 210                            "Shared_Dirty:  %8lu kB\n"
 211                            "Private_Clean: %8lu kB\n"
 212                            "Private_Dirty: %8lu kB\n",
 213                            (vma->vm_end - vma->vm_start) >> 10,
 214                            mss->resident >> 10,
 215                            mss->shared_clean  >> 10,
 216                            mss->shared_dirty  >> 10,
 217                            mss->private_clean >> 10,
 218                            mss->private_dirty >> 10);
 219
 220         if (m->count < m->size)  /* vma is copied successfully */
 221                 m->version = (vma != get_gate_vma(task))? vma->vm_start: 0;
 222         return 0;
 223 }
 224
 225 static int show_map(struct seq_file *m, void *v)
 226 {
 227         return show_map_internal(m, v, NULL);
 228 }
 229
 230 static void smaps_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
 231                                 unsigned long addr, unsigned long end,
 232                                 struct mem_size_stats *mss)
 233 {
 234         pte_t *pte, ptent;
 235         spinlock_t *ptl;
 236         struct page *page;
 237
 238         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 239         do {
 240                 ptent = *pte;
 241                 if (!pte_present(ptent))
 242                         continue;
 243
 244                 mss->resident += PAGE_SIZE;
 245
 246                 page = vm_normal_page(vma, addr, ptent);
 247                 if (!page)
 248                         continue;
 249
 250                 if (page_mapcount(page) >= 2) {
 251                         if (pte_dirty(ptent))
 252                                 mss->shared_dirty += PAGE_SIZE;
 253                         else
 254                                 mss->shared_clean += PAGE_SIZE;
 255                 } else {
 256                         if (pte_dirty(ptent))
 257                                 mss->private_dirty += PAGE_SIZE;
 258                         else
 259                                 mss->private_clean += PAGE_SIZE;
 260                 }
 261         } while (pte++, addr += PAGE_SIZE, addr != end);
 262         pte_unmap_unlock(pte - 1, ptl);
 263         cond_resched();
 264 }
 265
 266 static inline void smaps_pmd_range(struct vm_area_struct *vma, pud_t *pud,
 267                                 unsigned long addr, unsigned long end,
 268                                 struct mem_size_stats *mss)
 269 {
 270         pmd_t *pmd;
 271         unsigned long next;
 272
 273         pmd = pmd_offset(pud, addr);
 274         do {
 275                 next = pmd_addr_end(addr, end);
 276                 if (pmd_none_or_clear_bad(pmd))
 277                         continue;
 278                 smaps_pte_range(vma, pmd, addr, next, mss);
 279         } while (pmd++, addr = next, addr != end);
 280 }
 281
 282 static inline void smaps_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
 283                                 unsigned long addr, unsigned long end,
 284                                 struct mem_size_stats *mss)
 285 {
 286         pud_t *pud;
 287         unsigned long next;
 288
 289         pud = pud_offset(pgd, addr);
 290         do {
 291                 next = pud_addr_end(addr, end);
 292                 if (pud_none_or_clear_bad(pud))
 293                         continue;
 294                 smaps_pmd_range(vma, pud, addr, next, mss);
 295         } while (pud++, addr = next, addr != end);
 296 }
 297
 298 static inline void smaps_pgd_range(struct vm_area_struct *vma,
 299                                 unsigned long addr, unsigned long end,
 300                                 struct mem_size_stats *mss)
 301 {
 302         pgd_t *pgd;
 303         unsigned long next;
 304
 305         pgd = pgd_offset(vma->vm_mm, addr);
 306         do {
 307                 next = pgd_addr_end(addr, end);
 308                 if (pgd_none_or_clear_bad(pgd))
 309                         continue;
 310                 smaps_pud_range(vma, pgd, addr, next, mss);
 311         } while (pgd++, addr = next, addr != end);
 312 }
 313
 314 static int show_smap(struct seq_file *m, void *v)
 315 {
 316         struct vm_area_struct *vma = v;
 317         struct mem_size_stats mss;
 318
 319         memset(&mss, 0, sizeof mss);
 320         if (vma->vm_mm && !is_vm_hugetlb_page(vma))
 321                 smaps_pgd_range(vma, vma->vm_start, vma->vm_end, &mss);
 322         return show_map_internal(m, v, &mss);
 323 }
 324
 325 static void *m_start(struct seq_file *m, loff_t *pos)
 326 {
 327         struct proc_maps_private *priv = m->private;
 328         unsigned long last_addr = m->version;
 329         struct mm_struct *mm;
 330         struct vm_area_struct *vma, *tail_vma = NULL;
 331         loff_t l = *pos;
 332
 333         /* Clear the per syscall fields in priv */
 334         priv->task = NULL;
 335         priv->tail_vma = NULL;
 336
 337         /*
 338          * We remember last_addr rather than next_addr to hit with
 339          * mmap_cache most of the time. We have zero last_addr at
 340          * the beginning and also after lseek. We will have -1 last_addr
 341          * after the end of the vmas.
 342          */
 343
 344         if (last_addr == -1UL)
 345                 return NULL;
 346
 347         priv->task = get_pid_task(priv->pid, PIDTYPE_PID);
 348         if (!priv->task)
 349                 return NULL;
 350
 351         mm = mm_for_maps(priv->task);
 352         if (!mm)
 353                 return NULL;
 354
 355         priv->tail_vma = tail_vma = get_gate_vma(priv->task);
 356
 357         /* Start with last addr hint */
 358         if (last_addr && (vma = find_vma(mm, last_addr))) {
 359                 vma = vma->vm_next;
 360                 goto out;
 361         }
 362
 363         /*
 364          * Check the vma index is within the range and do
 365          * sequential scan until m_index.
 366          */
 367         vma = NULL;
 368         if ((unsigned long)l < mm->map_count) {
 369                 vma = mm->mmap;
 370                 while (l-- && vma)
 371                         vma = vma->vm_next;
 372                 goto out;
 373         }
 374
 375         if (l != mm->map_count)
 376                 tail_vma = NULL; /* After gate vma */
 377
 378 out:
 379         if (vma)
 380                 return vma;
 381
 382         /* End of vmas has been reached */
 383         m->version = (tail_vma != NULL)? 0: -1UL;
 384         up_read(&mm->mmap_sem);
 385         mmput(mm);
 386         return tail_vma;
 387 }
 388
 389 static void vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma)
 390 {
 391         if (vma && vma != priv->tail_vma) {
 392                 struct mm_struct *mm = vma->vm_mm;
 393                 up_read(&mm->mmap_sem);
 394                 mmput(mm);
 395         }
 396 }
 397
 398 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
 399 {
 400         struct proc_maps_private *priv = m->private;
 401         struct vm_area_struct *vma = v;
 402         struct vm_area_struct *tail_vma = priv->tail_vma;
 403
 404         (*pos)++;
 405         if (vma && (vma != tail_vma) && vma->vm_next)
 406                 return vma->vm_next;
 407         vma_stop(priv, vma);
 408         return (vma != tail_vma)? tail_vma: NULL;
 409 }
 410
 411 static void m_stop(struct seq_file *m, void *v)
 412 {
 413         struct proc_maps_private *priv = m->private;
 414         struct vm_area_struct *vma = v;
 415
 416         vma_stop(priv, vma);
 417         if (priv->task)
 418                 put_task_struct(priv->task);
 419 }
 420
 421 static struct seq_operations proc_pid_maps_op = {
 422         .start  = m_start,
 423         .next   = m_next,
 424         .stop   = m_stop,
 425         .show   = show_map
 426 };
 427
 428 static struct seq_operations proc_pid_smaps_op = {
 429         .start  = m_start,
 430         .next   = m_next,
 431         .stop   = m_stop,
 432         .show   = show_smap
 433 };
 434
 435 static int do_maps_open(struct inode *inode, struct file *file,
 436                         struct seq_operations *ops)
 437 {
 438         struct proc_maps_private *priv;
 439         int ret = -ENOMEM;
 440         priv = kzalloc(sizeof(*priv), GFP_KERNEL);
 441         if (priv) {
 442                 priv->pid = proc_pid(inode);
 443                 ret = seq_open(file, ops);
 444                 if (!ret) {
 445                         struct seq_file *m = file->private_data;
 446                         m->private = priv;
 447                 } else {
 448                         kfree(priv);
 449                 }
 450         }
 451         return ret;
 452 }
 453
 454 static int maps_open(struct inode *inode, struct file *file)
 455 {
 456         return do_maps_open(inode, file, &proc_pid_maps_op);
 457 }
 458
 459 struct file_operations proc_maps_operations = {
 460         .open           = maps_open,
 461         .read           = seq_read,
 462         .llseek         = seq_lseek,
 463         .release        = seq_release_private,
 464 };
 465
 466 #ifdef CONFIG_NUMA
 467 extern int show_numa_map(struct seq_file *m, void *v);
 468
 469 static struct seq_operations proc_pid_numa_maps_op = {
 470         .start  = m_start,
 471         .next   = m_next,
 472         .stop   = m_stop,
 473         .show   = show_numa_map
 474 };
 475
 476 static int numa_maps_open(struct inode *inode, struct file *file)
 477 {
 478         return do_maps_open(inode, file, &proc_pid_numa_maps_op);
 479 }
 480
 481 struct file_operations proc_numa_maps_operations = {
 482         .open           = numa_maps_open,
 483         .read           = seq_read,
 484         .llseek         = seq_lseek,
 485         .release        = seq_release_private,
 486 };
 487 #endif
 488
 489 static int smaps_open(struct inode *inode, struct file *file)
 490 {
 491         return do_maps_open(inode, file, &proc_pid_smaps_op);
 492 }
 493
 494 struct file_operations proc_smaps_operations = {
 495         .open           = smaps_open,
 496         .read           = seq_read,
 497         .llseek         = seq_lseek,
 498         .release        = seq_release_private,
 499 };