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         struct vm_area_struct *vma;
 139         unsigned long resident;
 140         unsigned long shared_clean;
 141         unsigned long shared_dirty;
 142         unsigned long private_clean;
 143         unsigned long private_dirty;
 144         unsigned long referenced;
 145         u64 pss;
 146 };
 147
 148 static int show_map(struct seq_file *m, void *v)
 149 {
 150         struct proc_maps_private *priv = m->private;
 151         struct task_struct *task = priv->task;
 152         struct vm_area_struct *vma = v;
 153         struct mm_struct *mm = vma->vm_mm;
 154         struct file *file = vma->vm_file;
 155         int flags = vma->vm_flags;
 156         unsigned long ino = 0;
 157         dev_t dev = 0;
 158         int len;
 159
 160         if (maps_protect && !ptrace_may_attach(task))
 161                 return -EACCES;
 162
 163         if (file) {
 164                 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
 165                 dev = inode->i_sb->s_dev;
 166                 ino = inode->i_ino;
 167         }
 168
 169         seq_printf(m, "%08lx-%08lx %c%c%c%c %08lx %02x:%02x %lu %n",
 170                         vma->vm_start,
 171                         vma->vm_end,
 172                         flags & VM_READ ? 'r' : '-',
 173                         flags & VM_WRITE ? 'w' : '-',
 174                         flags & VM_EXEC ? 'x' : '-',
 175                         flags & VM_MAYSHARE ? 's' : 'p',
 176                         vma->vm_pgoff << PAGE_SHIFT,
 177                         MAJOR(dev), MINOR(dev), ino, &len);
 178
 179         /*
 180          * Print the dentry name for named mappings, and a
 181          * special [heap] marker for the heap:
 182          */
 183         if (file) {
 184                 pad_len_spaces(m, len);
 185                 seq_path(m, file->f_path.mnt, file->f_path.dentry, "\n");
 186         } else {
 187                 const char *name = arch_vma_name(vma);
 188                 if (!name) {
 189                         if (mm) {
 190                                 if (vma->vm_start <= mm->start_brk &&
 191                                                 vma->vm_end >= mm->brk) {
 192                                         name = "[heap]";
 193                                 } else if (vma->vm_start <= mm->start_stack &&
 194                                            vma->vm_end >= mm->start_stack) {
 195                                         name = "[stack]";
 196                                 }
 197                         } else {
 198                                 name = "[vdso]";
 199                         }
 200                 }
 201                 if (name) {
 202                         pad_len_spaces(m, len);
 203                         seq_puts(m, name);
 204                 }
 205         }
 206         seq_putc(m, '\n');
 207
 208         if (m->count < m->size)  /* vma is copied successfully */
 209                 m->version = (vma != get_gate_vma(task))? vma->vm_start: 0;
 210         return 0;
 211 }
 212
 213 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 214                            void *private)
 215 {
 216         struct mem_size_stats *mss = private;
 217         struct vm_area_struct *vma = mss->vma;
 218         pte_t *pte, ptent;
 219         spinlock_t *ptl;
 220         struct page *page;
 221         int mapcount;
 222
 223         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 224         for (; addr != end; pte++, addr += PAGE_SIZE) {
 225                 ptent = *pte;
 226                 if (!pte_present(ptent))
 227                         continue;
 228
 229                 mss->resident += PAGE_SIZE;
 230
 231                 page = vm_normal_page(vma, addr, ptent);
 232                 if (!page)
 233                         continue;
 234
 235                 /* Accumulate the size in pages that have been accessed. */
 236                 if (pte_young(ptent) || PageReferenced(page))
 237                         mss->referenced += PAGE_SIZE;
 238                 mapcount = page_mapcount(page);
 239                 if (mapcount >= 2) {
 240                         if (pte_dirty(ptent))
 241                                 mss->shared_dirty += PAGE_SIZE;
 242                         else
 243                                 mss->shared_clean += PAGE_SIZE;
 244                         mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount;
 245                 } else {
 246                         if (pte_dirty(ptent))
 247                                 mss->private_dirty += PAGE_SIZE;
 248                         else
 249                                 mss->private_clean += PAGE_SIZE;
 250                         mss->pss += (PAGE_SIZE << PSS_SHIFT);
 251                 }
 252         }
 253         pte_unmap_unlock(pte - 1, ptl);
 254         cond_resched();
 255         return 0;
 256 }
 257
 258 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
 259                                 unsigned long end, void *private)
 260 {
 261         struct vm_area_struct *vma = private;
 262         pte_t *pte, ptent;
 263         spinlock_t *ptl;
 264         struct page *page;
 265
 266         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 267         for (; addr != end; pte++, addr += PAGE_SIZE) {
 268                 ptent = *pte;
 269                 if (!pte_present(ptent))
 270                         continue;
 271
 272                 page = vm_normal_page(vma, addr, ptent);
 273                 if (!page)
 274                         continue;
 275
 276                 /* Clear accessed and referenced bits. */
 277                 ptep_test_and_clear_young(vma, addr, pte);
 278                 ClearPageReferenced(page);
 279         }
 280         pte_unmap_unlock(pte - 1, ptl);
 281         cond_resched();
 282         return 0;
 283 }
 284
 285 static struct mm_walk smaps_walk = { .pmd_entry = smaps_pte_range };
 286
 287 static int show_smap(struct seq_file *m, void *v)
 288 {
 289         struct vm_area_struct *vma = v;
 290         struct mem_size_stats mss;
 291         int ret;
 292
 293         memset(&mss, 0, sizeof mss);
 294         mss.vma = vma;
 295         if (vma->vm_mm && !is_vm_hugetlb_page(vma))
 296                 walk_page_range(vma->vm_mm, vma->vm_start, vma->vm_end,
 297                                 &smaps_walk, &mss);
 298
 299         ret = show_map(m, v);
 300         if (ret)
 301                 return ret;
 302
 303         seq_printf(m,
 304                    "Size:           %8lu kB\n"
 305                    "Rss:            %8lu kB\n"
 306                    "Pss:            %8lu kB\n"
 307                    "Shared_Clean:   %8lu kB\n"
 308                    "Shared_Dirty:   %8lu kB\n"
 309                    "Private_Clean:  %8lu kB\n"
 310                    "Private_Dirty:  %8lu kB\n"
 311                    "Referenced:     %8lu kB\n",
 312                    (vma->vm_end - vma->vm_start) >> 10,
 313                    mss.resident >> 10,
 314                    (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
 315                    mss.shared_clean  >> 10,
 316                    mss.shared_dirty  >> 10,
 317                    mss.private_clean >> 10,
 318                    mss.private_dirty >> 10,
 319                    mss.referenced >> 10);
 320
 321         return ret;
 322 }
 323
 324 static struct mm_walk clear_refs_walk = { .pmd_entry = clear_refs_pte_range };
 325
 326 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
 327                                 size_t count, loff_t *ppos)
 328 {
 329         struct task_struct *task;
 330         char buffer[PROC_NUMBUF], *end;
 331         struct mm_struct *mm;
 332         struct vm_area_struct *vma;
 333
 334         memset(buffer, 0, sizeof(buffer));
 335         if (count > sizeof(buffer) - 1)
 336                 count = sizeof(buffer) - 1;
 337         if (copy_from_user(buffer, buf, count))
 338                 return -EFAULT;
 339         if (!simple_strtol(buffer, &end, 0))
 340                 return -EINVAL;
 341         if (*end == '\n')
 342                 end++;
 343         task = get_proc_task(file->f_path.dentry->d_inode);
 344         if (!task)
 345                 return -ESRCH;
 346         mm = get_task_mm(task);
 347         if (mm) {
 348                 down_read(&mm->mmap_sem);
 349                 for (vma = mm->mmap; vma; vma = vma->vm_next)
 350                         if (!is_vm_hugetlb_page(vma))
 351                                 walk_page_range(mm, vma->vm_start, vma->vm_end,
 352                                                 &clear_refs_walk, vma);
 353                 flush_tlb_mm(mm);
 354                 up_read(&mm->mmap_sem);
 355                 mmput(mm);
 356         }
 357         put_task_struct(task);
 358         if (end - buffer == 0)
 359                 return -EIO;
 360         return end - buffer;
 361 }
 362
 363 const struct file_operations proc_clear_refs_operations = {
 364         .write          = clear_refs_write,
 365 };
 366
 367 static void *m_start(struct seq_file *m, loff_t *pos)
 368 {
 369         struct proc_maps_private *priv = m->private;
 370         unsigned long last_addr = m->version;
 371         struct mm_struct *mm;
 372         struct vm_area_struct *vma, *tail_vma = NULL;
 373         loff_t l = *pos;
 374
 375         /* Clear the per syscall fields in priv */
 376         priv->task = NULL;
 377         priv->tail_vma = NULL;
 378
 379         /*
 380          * We remember last_addr rather than next_addr to hit with
 381          * mmap_cache most of the time. We have zero last_addr at
 382          * the beginning and also after lseek. We will have -1 last_addr
 383          * after the end of the vmas.
 384          */
 385
 386         if (last_addr == -1UL)
 387                 return NULL;
 388
 389         priv->task = get_pid_task(priv->pid, PIDTYPE_PID);
 390         if (!priv->task)
 391                 return NULL;
 392
 393         mm = mm_for_maps(priv->task);
 394         if (!mm)
 395                 return NULL;
 396
 397         priv->tail_vma = tail_vma = get_gate_vma(priv->task);
 398
 399         /* Start with last addr hint */
 400         if (last_addr && (vma = find_vma(mm, last_addr))) {
 401                 vma = vma->vm_next;
 402                 goto out;
 403         }
 404
 405         /*
 406          * Check the vma index is within the range and do
 407          * sequential scan until m_index.
 408          */
 409         vma = NULL;
 410         if ((unsigned long)l < mm->map_count) {
 411                 vma = mm->mmap;
 412                 while (l-- && vma)
 413                         vma = vma->vm_next;
 414                 goto out;
 415         }
 416
 417         if (l != mm->map_count)
 418                 tail_vma = NULL; /* After gate vma */
 419
 420 out:
 421         if (vma)
 422                 return vma;
 423
 424         /* End of vmas has been reached */
 425         m->version = (tail_vma != NULL)? 0: -1UL;
 426         up_read(&mm->mmap_sem);
 427         mmput(mm);
 428         return tail_vma;
 429 }
 430
 431 static void vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma)
 432 {
 433         if (vma && vma != priv->tail_vma) {
 434                 struct mm_struct *mm = vma->vm_mm;
 435                 up_read(&mm->mmap_sem);
 436                 mmput(mm);
 437         }
 438 }
 439
 440 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
 441 {
 442         struct proc_maps_private *priv = m->private;
 443         struct vm_area_struct *vma = v;
 444         struct vm_area_struct *tail_vma = priv->tail_vma;
 445
 446         (*pos)++;
 447         if (vma && (vma != tail_vma) && vma->vm_next)
 448                 return vma->vm_next;
 449         vma_stop(priv, vma);
 450         return (vma != tail_vma)? tail_vma: NULL;
 451 }
 452
 453 static void m_stop(struct seq_file *m, void *v)
 454 {
 455         struct proc_maps_private *priv = m->private;
 456         struct vm_area_struct *vma = v;
 457
 458         vma_stop(priv, vma);
 459         if (priv->task)
 460                 put_task_struct(priv->task);
 461 }
 462
 463 static struct seq_operations proc_pid_maps_op = {
 464         .start  = m_start,
 465         .next   = m_next,
 466         .stop   = m_stop,
 467         .show   = show_map
 468 };
 469
 470 static struct seq_operations proc_pid_smaps_op = {
 471         .start  = m_start,
 472         .next   = m_next,
 473         .stop   = m_stop,
 474         .show   = show_smap
 475 };
 476
 477 static int do_maps_open(struct inode *inode, struct file *file,
 478                         struct seq_operations *ops)
 479 {
 480         struct proc_maps_private *priv;
 481         int ret = -ENOMEM;
 482         priv = kzalloc(sizeof(*priv), GFP_KERNEL);
 483         if (priv) {
 484                 priv->pid = proc_pid(inode);
 485                 ret = seq_open(file, ops);
 486                 if (!ret) {
 487                         struct seq_file *m = file->private_data;
 488                         m->private = priv;
 489                 } else {
 490                         kfree(priv);
 491                 }
 492         }
 493         return ret;
 494 }
 495
 496 static int maps_open(struct inode *inode, struct file *file)
 497 {
 498         return do_maps_open(inode, file, &proc_pid_maps_op);
 499 }
 500
 501 const struct file_operations proc_maps_operations = {
 502         .open           = maps_open,
 503         .read           = seq_read,
 504         .llseek         = seq_lseek,
 505         .release        = seq_release_private,
 506 };
 507
 508 #ifdef CONFIG_NUMA
 509 extern int show_numa_map(struct seq_file *m, void *v);
 510
 511 static int show_numa_map_checked(struct seq_file *m, void *v)
 512 {
 513         struct proc_maps_private *priv = m->private;
 514         struct task_struct *task = priv->task;
 515
 516         if (maps_protect && !ptrace_may_attach(task))
 517                 return -EACCES;
 518
 519         return show_numa_map(m, v);
 520 }
 521
 522 static struct seq_operations proc_pid_numa_maps_op = {
 523         .start  = m_start,
 524         .next   = m_next,
 525         .stop   = m_stop,
 526         .show   = show_numa_map_checked
 527 };
 528
 529 static int numa_maps_open(struct inode *inode, struct file *file)
 530 {
 531         return do_maps_open(inode, file, &proc_pid_numa_maps_op);
 532 }
 533
 534 const struct file_operations proc_numa_maps_operations = {
 535         .open           = numa_maps_open,
 536         .read           = seq_read,
 537         .llseek         = seq_lseek,
 538         .release        = seq_release_private,
 539 };
 540 #endif
 541
 542 static int smaps_open(struct inode *inode, struct file *file)
 543 {
 544         return do_maps_open(inode, file, &proc_pid_smaps_op);
 545 }
 546
 547 const struct file_operations proc_smaps_operations = {
 548         .open           = smaps_open,
 549         .read           = seq_read,
 550         .llseek         = seq_lseek,
 551         .release        = seq_release_private,
 552 };