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_vfsmnt);
 108                 *dentry = dget(vma->vm_file->f_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 __attribute__((weak)) const char *arch_vma_name(struct vm_area_struct *vma)
 136 {
 137         return NULL;
 138 }
 139
 140 static int show_map_internal(struct seq_file *m, void *v, struct mem_size_stats *mss)
 141 {
 142         struct proc_maps_private *priv = m->private;
 143         struct task_struct *task = priv->task;
 144 #ifdef __i386__
 145         struct mm_struct *tmm = get_task_mm(task);
 146 #endif
 147         struct vm_area_struct *vma = v;
 148         struct mm_struct *mm = vma->vm_mm;
 149         struct file *file = vma->vm_file;
 150         int flags = vma->vm_flags;
 151         unsigned long ino = 0;
 152         dev_t dev = 0;
 153         int len;
 154
 155         if (file) {
 156                 struct inode *inode = vma->vm_file->f_dentry->d_inode;
 157                 dev = inode->i_sb->s_dev;
 158                 ino = inode->i_ino;
 159         }
 160
 161         seq_printf(m, "%08lx-%08lx %c%c%c%c %08lx %02x:%02x %lu %n",
 162                         vma->vm_start,
 163                         vma->vm_end,
 164                         flags & VM_READ ? 'r' : '-',
 165                         flags & VM_WRITE ? 'w' : '-',
 166                         (flags & VM_EXEC
 167 #ifdef __i386__
 168                                 || (!nx_enabled && tmm &&
 169                                 (vma->vm_start < tmm->context.exec_limit))
 170 #endif
 171                         )
 172                                 ? 'x' : '-',
 173                         flags & VM_MAYSHARE ? 's' : 'p',
 174                         vma->vm_pgoff << PAGE_SHIFT,
 175                         MAJOR(dev), MINOR(dev), ino, &len);
 176 #ifdef __i386__
 177         if (tmm)
 178                 mmput(tmm);
 179 #endif
 180
 181         /*
 182          * Print the dentry name for named mappings, and a
 183          * special [heap] marker for the heap:
 184          */
 185         if (file) {
 186                 pad_len_spaces(m, len);
 187                 seq_path(m, file->f_vfsmnt, file->f_dentry, "\n");
 188         } else {
 189                 const char *name = arch_vma_name(vma);
 190                 if (!name) {
 191                         if (mm) {
 192                                 if (vma->vm_start <= mm->start_brk &&
 193                                                 vma->vm_end >= mm->brk) {
 194                                         name = "[heap]";
 195                                 } else if (vma->vm_start <= mm->start_stack &&
 196                                            vma->vm_end >= mm->start_stack) {
 197                                         name = "[stack]";
 198                                 }
 199                         } else {
 200                                 name = "[vdso]";
 201                         }
 202                 }
 203                 if (name) {
 204                         pad_len_spaces(m, len);
 205                         seq_puts(m, name);
 206                 }
 207         }
 208         seq_putc(m, '\n');
 209
 210         if (mss)
 211                 seq_printf(m,
 212                            "Size:          %8lu kB\n"
 213                            "Rss:           %8lu kB\n"
 214                            "Shared_Clean:  %8lu kB\n"
 215                            "Shared_Dirty:  %8lu kB\n"
 216                            "Private_Clean: %8lu kB\n"
 217                            "Private_Dirty: %8lu kB\n",
 218                            (vma->vm_end - vma->vm_start) >> 10,
 219                            mss->resident >> 10,
 220                            mss->shared_clean  >> 10,
 221                            mss->shared_dirty  >> 10,
 222                            mss->private_clean >> 10,
 223                            mss->private_dirty >> 10);
 224
 225         if (m->count < m->size)  /* vma is copied successfully */
 226                 m->version = (vma != get_gate_vma(task))? vma->vm_start: 0;
 227         return 0;
 228 }
 229
 230 static int show_map(struct seq_file *m, void *v)
 231 {
 232         return show_map_internal(m, v, NULL);
 233 }
 234
 235 static void smaps_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
 236                                 unsigned long addr, unsigned long end,
 237                                 struct mem_size_stats *mss)
 238 {
 239         pte_t *pte, ptent;
 240         spinlock_t *ptl;
 241         struct page *page;
 242
 243         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 244         do {
 245                 ptent = *pte;
 246                 if (!pte_present(ptent))
 247                         continue;
 248
 249                 mss->resident += PAGE_SIZE;
 250
 251                 page = vm_normal_page(vma, addr, ptent);
 252                 if (!page)
 253                         continue;
 254
 255                 if (page_mapcount(page) >= 2) {
 256                         if (pte_dirty(ptent))
 257                                 mss->shared_dirty += PAGE_SIZE;
 258                         else
 259                                 mss->shared_clean += PAGE_SIZE;
 260                 } else {
 261                         if (pte_dirty(ptent))
 262                                 mss->private_dirty += PAGE_SIZE;
 263                         else
 264                                 mss->private_clean += PAGE_SIZE;
 265                 }
 266         } while (pte++, addr += PAGE_SIZE, addr != end);
 267         pte_unmap_unlock(pte - 1, ptl);
 268         cond_resched();
 269 }
 270
 271 static inline void smaps_pmd_range(struct vm_area_struct *vma, pud_t *pud,
 272                                 unsigned long addr, unsigned long end,
 273                                 struct mem_size_stats *mss)
 274 {
 275         pmd_t *pmd;
 276         unsigned long next;
 277
 278         pmd = pmd_offset(pud, addr);
 279         do {
 280                 next = pmd_addr_end(addr, end);
 281                 if (pmd_none_or_clear_bad(pmd))
 282                         continue;
 283                 smaps_pte_range(vma, pmd, addr, next, mss);
 284         } while (pmd++, addr = next, addr != end);
 285 }
 286
 287 static inline void smaps_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
 288                                 unsigned long addr, unsigned long end,
 289                                 struct mem_size_stats *mss)
 290 {
 291         pud_t *pud;
 292         unsigned long next;
 293
 294         pud = pud_offset(pgd, addr);
 295         do {
 296                 next = pud_addr_end(addr, end);
 297                 if (pud_none_or_clear_bad(pud))
 298                         continue;
 299                 smaps_pmd_range(vma, pud, addr, next, mss);
 300         } while (pud++, addr = next, addr != end);
 301 }
 302
 303 static inline void smaps_pgd_range(struct vm_area_struct *vma,
 304                                 unsigned long addr, unsigned long end,
 305                                 struct mem_size_stats *mss)
 306 {
 307         pgd_t *pgd;
 308         unsigned long next;
 309
 310         pgd = pgd_offset(vma->vm_mm, addr);
 311         do {
 312                 next = pgd_addr_end(addr, end);
 313                 if (pgd_none_or_clear_bad(pgd))
 314                         continue;
 315                 smaps_pud_range(vma, pgd, addr, next, mss);
 316         } while (pgd++, addr = next, addr != end);
 317 }
 318
 319 static int show_smap(struct seq_file *m, void *v)
 320 {
 321         struct vm_area_struct *vma = v;
 322         struct mem_size_stats mss;
 323
 324         memset(&mss, 0, sizeof mss);
 325         if (vma->vm_mm && !is_vm_hugetlb_page(vma))
 326                 smaps_pgd_range(vma, vma->vm_start, vma->vm_end, &mss);
 327         return show_map_internal(m, v, &mss);
 328 }
 329
 330 static void *m_start(struct seq_file *m, loff_t *pos)
 331 {
 332         struct proc_maps_private *priv = m->private;
 333         unsigned long last_addr = m->version;
 334         struct mm_struct *mm;
 335         struct vm_area_struct *vma, *tail_vma = NULL;
 336         loff_t l = *pos;
 337
 338         /* Clear the per syscall fields in priv */
 339         priv->task = NULL;
 340         priv->tail_vma = NULL;
 341
 342         /*
 343          * We remember last_addr rather than next_addr to hit with
 344          * mmap_cache most of the time. We have zero last_addr at
 345          * the beginning and also after lseek. We will have -1 last_addr
 346          * after the end of the vmas.
 347          */
 348
 349         if (last_addr == -1UL)
 350                 return NULL;
 351
 352         priv->task = get_pid_task(priv->pid, PIDTYPE_PID);
 353         if (!priv->task)
 354                 return NULL;
 355
 356         mm = mm_for_maps(priv->task);
 357         if (!mm)
 358                 return NULL;
 359
 360         priv->tail_vma = tail_vma = get_gate_vma(priv->task);
 361
 362         /* Start with last addr hint */
 363         if (last_addr && (vma = find_vma(mm, last_addr))) {
 364                 vma = vma->vm_next;
 365                 goto out;
 366         }
 367
 368         /*
 369          * Check the vma index is within the range and do
 370          * sequential scan until m_index.
 371          */
 372         vma = NULL;
 373         if ((unsigned long)l < mm->map_count) {
 374                 vma = mm->mmap;
 375                 while (l-- && vma)
 376                         vma = vma->vm_next;
 377                 goto out;
 378         }
 379
 380         if (l != mm->map_count)
 381                 tail_vma = NULL; /* After gate vma */
 382
 383 out:
 384         if (vma)
 385                 return vma;
 386
 387         /* End of vmas has been reached */
 388         m->version = (tail_vma != NULL)? 0: -1UL;
 389         up_read(&mm->mmap_sem);
 390         mmput(mm);
 391         return tail_vma;
 392 }
 393
 394 static void vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma)
 395 {
 396         if (vma && vma != priv->tail_vma) {
 397                 struct mm_struct *mm = vma->vm_mm;
 398                 up_read(&mm->mmap_sem);
 399                 mmput(mm);
 400         }
 401 }
 402
 403 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
 404 {
 405         struct proc_maps_private *priv = m->private;
 406         struct vm_area_struct *vma = v;
 407         struct vm_area_struct *tail_vma = priv->tail_vma;
 408
 409         (*pos)++;
 410         if (vma && (vma != tail_vma) && vma->vm_next)
 411                 return vma->vm_next;
 412         vma_stop(priv, vma);
 413         return (vma != tail_vma)? tail_vma: NULL;
 414 }
 415
 416 static void m_stop(struct seq_file *m, void *v)
 417 {
 418         struct proc_maps_private *priv = m->private;
 419         struct vm_area_struct *vma = v;
 420
 421         vma_stop(priv, vma);
 422         if (priv->task)
 423                 put_task_struct(priv->task);
 424 }
 425
 426 static struct seq_operations proc_pid_maps_op = {
 427         .start  = m_start,
 428         .next   = m_next,
 429         .stop   = m_stop,
 430         .show   = show_map
 431 };
 432
 433 static struct seq_operations proc_pid_smaps_op = {
 434         .start  = m_start,
 435         .next   = m_next,
 436         .stop   = m_stop,
 437         .show   = show_smap
 438 };
 439
 440 static int do_maps_open(struct inode *inode, struct file *file,
 441                         struct seq_operations *ops)
 442 {
 443         struct proc_maps_private *priv;
 444         int ret = -ENOMEM;
 445         priv = kzalloc(sizeof(*priv), GFP_KERNEL);
 446         if (priv) {
 447                 priv->pid = proc_pid(inode);
 448                 ret = seq_open(file, ops);
 449                 if (!ret) {
 450                         struct seq_file *m = file->private_data;
 451                         m->private = priv;
 452                 } else {
 453                         kfree(priv);
 454                 }
 455         }
 456         return ret;
 457 }
 458
 459 static int maps_open(struct inode *inode, struct file *file)
 460 {
 461         return do_maps_open(inode, file, &proc_pid_maps_op);
 462 }
 463
 464 struct file_operations proc_maps_operations = {
 465         .open           = maps_open,
 466         .read           = seq_read,
 467         .llseek         = seq_lseek,
 468         .release        = seq_release_private,
 469 };
 470
 471 #ifdef CONFIG_NUMA
 472 extern int show_numa_map(struct seq_file *m, void *v);
 473
 474 static struct seq_operations proc_pid_numa_maps_op = {
 475         .start  = m_start,
 476         .next   = m_next,
 477         .stop   = m_stop,
 478         .show   = show_numa_map
 479 };
 480
 481 static int numa_maps_open(struct inode *inode, struct file *file)
 482 {
 483         return do_maps_open(inode, file, &proc_pid_numa_maps_op);
 484 }
 485
 486 struct file_operations proc_numa_maps_operations = {
 487         .open           = numa_maps_open,
 488         .read           = seq_read,
 489         .llseek         = seq_lseek,
 490         .release        = seq_release_private,
 491 };
 492 #endif
 493
 494 static int smaps_open(struct inode *inode, struct file *file)
 495 {
 496         return do_maps_open(inode, file, &proc_pid_smaps_op);
 497 }
 498
 499 struct file_operations proc_smaps_operations = {
 500         .open           = smaps_open,
 501         .read           = seq_read,
 502         .llseek         = seq_lseek,
 503         .release        = seq_release_private,
 504 };