kernel/fork.c

   1 /*
   2  *  linux/kernel/fork.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  *  'fork.c' contains the help-routines for the 'fork' system call
   9  * (see also entry.S and others).
  10  * Fork is rather simple, once you get the hang of it, but the memory
  11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  12  */
  13
  14 #include <linux/config.h>
  15 #include <linux/slab.h>
  16 #include <linux/init.h>
  17 #include <linux/unistd.h>
  18 #include <linux/smp_lock.h>
  19 #include <linux/module.h>
  20 #include <linux/vmalloc.h>
  21 #include <linux/completion.h>
  22 #include <linux/namespace.h>
  23 #include <linux/personality.h>
  24 #include <linux/mempolicy.h>
  25 #include <linux/sem.h>
  26 #include <linux/file.h>
  27 #include <linux/key.h>
  28 #include <linux/binfmts.h>
  29 #include <linux/mman.h>
  30 #include <linux/fs.h>
  31 #include <linux/cpu.h>
  32 #include <linux/security.h>
  33 #include <linux/swap.h>
  34 #include <linux/syscalls.h>
  35 #include <linux/jiffies.h>
  36 #include <linux/futex.h>
  37 #include <linux/ptrace.h>
  38 #include <linux/mount.h>
  39 #include <linux/audit.h>
  40 #include <linux/profile.h>
  41 #include <linux/rmap.h>
  42 #include <linux/vs_network.h>
  43 #include <linux/vs_limit.h>
  44 #include <linux/vs_memory.h>
  45
  46 #include <asm/pgtable.h>
  47 #include <asm/pgalloc.h>
  48 #include <asm/uaccess.h>
  49 #include <asm/mmu_context.h>
  50 #include <asm/cacheflush.h>
  51 #include <asm/tlbflush.h>
  52
  53 /* The idle threads do not count..
  54  * Protected by write_lock_irq(&tasklist_lock)
  55  */
  56 int nr_threads;
  57
  58 int max_threads;
  59 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  60
  61 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  62
  63 rwlock_t tasklist_lock __cacheline_aligned = RW_LOCK_UNLOCKED;  /* outer */
  64
  65 EXPORT_SYMBOL(tasklist_lock);
  66
  67 int nr_processes(void)
  68 {
  69         int cpu;
  70         int total = 0;
  71
  72         for_each_online_cpu(cpu)
  73                 total += per_cpu(process_counts, cpu);
  74
  75         return total;
  76 }
  77
  78 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
  79 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
  80 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
  81 static kmem_cache_t *task_struct_cachep;
  82 #endif
  83
  84 void free_task(struct task_struct *tsk)
  85 {
  86         free_thread_info(tsk->thread_info);
  87         clr_vx_info(&tsk->vx_info);
  88         clr_nx_info(&tsk->nx_info);
  89         free_task_struct(tsk);
  90 }
  91 EXPORT_SYMBOL(free_task);
  92
  93 void __put_task_struct(struct task_struct *tsk)
  94 {
  95         WARN_ON(!(tsk->exit_state & (EXIT_DEAD | EXIT_ZOMBIE)));
  96         WARN_ON(atomic_read(&tsk->usage));
  97         WARN_ON(tsk == current);
  98
  99         if (unlikely(tsk->audit_context))
 100                 audit_free(tsk);
 101         security_task_free(tsk);
 102         free_uid(tsk->user);
 103         put_group_info(tsk->group_info);
 104
 105         if (!profile_handoff_task(tsk))
 106                 free_task(tsk);
 107 }
 108
 109 void __init fork_init(unsigned long mempages)
 110 {
 111 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 112 #ifndef ARCH_MIN_TASKALIGN
 113 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 114 #endif
 115         /* create a slab on which task_structs can be allocated */
 116         task_struct_cachep =
 117                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 118                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL);
 119 #endif
 120
 121         /*
 122          * The default maximum number of threads is set to a safe
 123          * value: the thread structures can take up at most half
 124          * of memory.
 125          */
 126         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 127
 128         /*
 129          * we need to allow at least 20 threads to boot a system
 130          */
 131         if(max_threads < 20)
 132                 max_threads = 20;
 133
 134         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 135         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 136 }
 137
 138 static struct task_struct *dup_task_struct(struct task_struct *orig)
 139 {
 140         struct task_struct *tsk;
 141         struct thread_info *ti;
 142
 143         prepare_to_copy(orig);
 144
 145         tsk = alloc_task_struct();
 146         if (!tsk)
 147                 return NULL;
 148
 149         ti = alloc_thread_info(tsk);
 150         if (!ti) {
 151                 free_task_struct(tsk);
 152                 return NULL;
 153         }
 154
 155         *ti = *orig->thread_info;
 156         *tsk = *orig;
 157         tsk->thread_info = ti;
 158         ti->task = tsk;
 159
 160         /* One for us, one for whoever does the "release_task()" (usually parent) */
 161         atomic_set(&tsk->usage,2);
 162         return tsk;
 163 }
 164
 165 #ifdef CONFIG_MMU
 166 static inline int dup_mmap(struct mm_struct * mm, struct mm_struct * oldmm)
 167 {
 168         struct vm_area_struct * mpnt, *tmp, **pprev;
 169         struct rb_node **rb_link, *rb_parent;
 170         int retval;
 171         unsigned long charge;
 172         struct mempolicy *pol;
 173
 174         down_write(&oldmm->mmap_sem);
 175         flush_cache_mm(current->mm);
 176         mm->locked_vm = 0;
 177         mm->mmap = NULL;
 178         mm->mmap_cache = NULL;
 179         mm->free_area_cache = oldmm->mmap_base;
 180         mm->map_count = 0;
 181         mm->rss = 0;
 182         mm->anon_rss = 0;
 183         cpus_clear(mm->cpu_vm_mask);
 184         mm->mm_rb = RB_ROOT;
 185         rb_link = &mm->mm_rb.rb_node;
 186         rb_parent = NULL;
 187         pprev = &mm->mmap;
 188
 189         for (mpnt = current->mm->mmap ; mpnt ; mpnt = mpnt->vm_next) {
 190                 struct file *file;
 191
 192                 if (mpnt->vm_flags & VM_DONTCOPY) {
 193                         __vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 194                                                         -vma_pages(mpnt));
 195                         continue;
 196                 }
 197                 charge = 0;
 198                 if (mpnt->vm_flags & VM_ACCOUNT) {
 199                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 200                         if (security_vm_enough_memory(len))
 201                                 goto fail_nomem;
 202                         charge = len;
 203                 }
 204                 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
 205                 if (!tmp)
 206                         goto fail_nomem;
 207                 *tmp = *mpnt;
 208                 pol = mpol_copy(vma_policy(mpnt));
 209                 retval = PTR_ERR(pol);
 210                 if (IS_ERR(pol))
 211                         goto fail_nomem_policy;
 212                 vma_set_policy(tmp, pol);
 213                 tmp->vm_flags &= ~VM_LOCKED;
 214                 tmp->vm_mm = mm;
 215                 tmp->vm_next = NULL;
 216                 anon_vma_link(tmp);
 217                 file = tmp->vm_file;
 218                 if (file) {
 219                         struct inode *inode = file->f_dentry->d_inode;
 220                         get_file(file);
 221                         if (tmp->vm_flags & VM_DENYWRITE)
 222                                 atomic_dec(&inode->i_writecount);
 223
 224                         /* insert tmp into the share list, just after mpnt */
 225                         spin_lock(&file->f_mapping->i_mmap_lock);
 226                         flush_dcache_mmap_lock(file->f_mapping);
 227                         vma_prio_tree_add(tmp, mpnt);
 228                         flush_dcache_mmap_unlock(file->f_mapping);
 229                         spin_unlock(&file->f_mapping->i_mmap_lock);
 230                 }
 231
 232                 /*
 233                  * Link in the new vma and copy the page table entries:
 234                  * link in first so that swapoff can see swap entries,
 235                  * and try_to_unmap_one's find_vma find the new vma.
 236                  */
 237                 spin_lock(&mm->page_table_lock);
 238                 *pprev = tmp;
 239                 pprev = &tmp->vm_next;
 240
 241                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 242                 rb_link = &tmp->vm_rb.rb_right;
 243                 rb_parent = &tmp->vm_rb;
 244
 245                 mm->map_count++;
 246                 retval = copy_page_range(mm, current->mm, tmp);
 247                 spin_unlock(&mm->page_table_lock);
 248
 249                 if (tmp->vm_ops && tmp->vm_ops->open)
 250                         tmp->vm_ops->open(tmp);
 251
 252                 if (retval)
 253                         goto out;
 254         }
 255         retval = 0;
 256
 257 out:
 258         flush_tlb_mm(current->mm);
 259         up_write(&oldmm->mmap_sem);
 260         return retval;
 261 fail_nomem_policy:
 262         kmem_cache_free(vm_area_cachep, tmp);
 263 fail_nomem:
 264         retval = -ENOMEM;
 265         vm_unacct_memory(charge);
 266         goto out;
 267 }
 268
 269 static inline int mm_alloc_pgd(struct mm_struct * mm)
 270 {
 271         mm->pgd = pgd_alloc(mm);
 272         if (unlikely(!mm->pgd))
 273                 return -ENOMEM;
 274         return 0;
 275 }
 276
 277 static inline void mm_free_pgd(struct mm_struct * mm)
 278 {
 279         pgd_free(mm->pgd);
 280 }
 281 #else
 282 #define dup_mmap(mm, oldmm)     (0)
 283 #define mm_alloc_pgd(mm)        (0)
 284 #define mm_free_pgd(mm)
 285 #endif /* CONFIG_MMU */
 286
 287 spinlock_t mmlist_lock __cacheline_aligned_in_smp = SPIN_LOCK_UNLOCKED;
 288
 289 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
 290 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 291
 292 #include <linux/init_task.h>
 293
 294 static struct mm_struct * mm_init(struct mm_struct * mm)
 295 {
 296         atomic_set(&mm->mm_users, 1);
 297         atomic_set(&mm->mm_count, 1);
 298         init_rwsem(&mm->mmap_sem);
 299         INIT_LIST_HEAD(&mm->mmlist);
 300         mm->core_waiters = 0;
 301         mm->nr_ptes = 0;
 302         spin_lock_init(&mm->page_table_lock);
 303         rwlock_init(&mm->ioctx_list_lock);
 304         mm->ioctx_list = NULL;
 305         mm->default_kioctx = (struct kioctx)INIT_KIOCTX(mm->default_kioctx, *mm);
 306         mm->free_area_cache = TASK_UNMAPPED_BASE;
 307
 308         if (likely(!mm_alloc_pgd(mm))) {
 309                 mm->def_flags = 0;
 310                 set_vx_info(&mm->mm_vx_info, current->vx_info);
 311                 return mm;
 312         }
 313         free_mm(mm);
 314         return NULL;
 315 }
 316
 317 /*
 318  * Allocate and initialize an mm_struct.
 319  */
 320 struct mm_struct * mm_alloc(void)
 321 {
 322         struct mm_struct * mm;
 323
 324         mm = allocate_mm();
 325         if (mm) {
 326                 memset(mm, 0, sizeof(*mm));
 327                 mm = mm_init(mm);
 328         }
 329         return mm;
 330 }
 331
 332 /*
 333  * Called when the last reference to the mm
 334  * is dropped: either by a lazy thread or by
 335  * mmput. Free the page directory and the mm.
 336  */
 337 void fastcall __mmdrop(struct mm_struct *mm)
 338 {
 339         BUG_ON(mm == &init_mm);
 340         mm_free_pgd(mm);
 341         destroy_context(mm);
 342         clr_vx_info(&mm->mm_vx_info);
 343         free_mm(mm);
 344 }
 345
 346 /*
 347  * Decrement the use count and release all resources for an mm.
 348  */
 349 void mmput(struct mm_struct *mm)
 350 {
 351         if (atomic_dec_and_test(&mm->mm_users)) {
 352                 exit_aio(mm);
 353                 exit_mmap(mm);
 354                 if (!list_empty(&mm->mmlist)) {
 355                         spin_lock(&mmlist_lock);
 356                         list_del(&mm->mmlist);
 357                         spin_unlock(&mmlist_lock);
 358                 }
 359                 put_swap_token(mm);
 360                 mmdrop(mm);
 361         }
 362 }
 363 EXPORT_SYMBOL_GPL(mmput);
 364
 365 /**
 366  * get_task_mm - acquire a reference to the task's mm
 367  *
 368  * Returns %NULL if the task has no mm.  Checks PF_BORROWED_MM (meaning
 369  * this kernel workthread has transiently adopted a user mm with use_mm,
 370  * to do its AIO) is not set and if so returns a reference to it, after
 371  * bumping up the use count.  User must release the mm via mmput()
 372  * after use.  Typically used by /proc and ptrace.
 373  */
 374 struct mm_struct *get_task_mm(struct task_struct *task)
 375 {
 376         struct mm_struct *mm;
 377
 378         task_lock(task);
 379         mm = task->mm;
 380         if (mm) {
 381                 if (task->flags & PF_BORROWED_MM)
 382                         mm = NULL;
 383                 else
 384                         atomic_inc(&mm->mm_users);
 385         }
 386         task_unlock(task);
 387         return mm;
 388 }
 389 EXPORT_SYMBOL_GPL(get_task_mm);
 390
 391 /* Please note the differences between mmput and mm_release.
 392  * mmput is called whenever we stop holding onto a mm_struct,
 393  * error success whatever.
 394  *
 395  * mm_release is called after a mm_struct has been removed
 396  * from the current process.
 397  *
 398  * This difference is important for error handling, when we
 399  * only half set up a mm_struct for a new process and need to restore
 400  * the old one.  Because we mmput the new mm_struct before
 401  * restoring the old one. . .
 402  * Eric Biederman 10 January 1998
 403  */
 404 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 405 {
 406         struct completion *vfork_done = tsk->vfork_done;
 407
 408         /* Get rid of any cached register state */
 409         deactivate_mm(tsk, mm);
 410
 411         /* notify parent sleeping on vfork() */
 412         if (vfork_done) {
 413                 tsk->vfork_done = NULL;
 414                 complete(vfork_done);
 415         }
 416         if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
 417                 u32 __user * tidptr = tsk->clear_child_tid;
 418                 tsk->clear_child_tid = NULL;
 419
 420                 /*
 421                  * We don't check the error code - if userspace has
 422                  * not set up a proper pointer then tough luck.
 423                  */
 424                 put_user(0, tidptr);
 425                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
 426         }
 427 }
 428
 429 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 430 {
 431         struct mm_struct * mm, *oldmm;
 432         int retval;
 433
 434         tsk->min_flt = tsk->maj_flt = 0;
 435         tsk->nvcsw = tsk->nivcsw = 0;
 436
 437         tsk->mm = NULL;
 438         tsk->active_mm = NULL;
 439
 440         /*
 441          * Are we cloning a kernel thread?
 442          *
 443          * We need to steal a active VM for that..
 444          */
 445         oldmm = current->mm;
 446         if (!oldmm)
 447                 return 0;
 448
 449         if (clone_flags & CLONE_VM) {
 450                 atomic_inc(&oldmm->mm_users);
 451                 mm = oldmm;
 452                 /*
 453                  * There are cases where the PTL is held to ensure no
 454                  * new threads start up in user mode using an mm, which
 455                  * allows optimizing out ipis; the tlb_gather_mmu code
 456                  * is an example.
 457                  */
 458                 spin_unlock_wait(&oldmm->page_table_lock);
 459                 goto good_mm;
 460         }
 461
 462         retval = -ENOMEM;
 463         mm = allocate_mm();
 464         if (!mm)
 465                 goto fail_nomem;
 466
 467         /* Copy the current MM stuff.. */
 468         memcpy(mm, oldmm, sizeof(*mm));
 469         mm->mm_vx_info = NULL;
 470         if (!mm_init(mm))
 471                 goto fail_nomem;
 472
 473         if (init_new_context(tsk,mm))
 474                 goto fail_nocontext;
 475
 476         retval = dup_mmap(mm, oldmm);
 477         if (retval)
 478                 goto free_pt;
 479
 480 good_mm:
 481         tsk->mm = mm;
 482         tsk->active_mm = mm;
 483         return 0;
 484
 485 free_pt:
 486         mmput(mm);
 487 fail_nomem:
 488         return retval;
 489
 490 fail_nocontext:
 491         /*
 492          * If init_new_context() failed, we cannot use mmput() to free the mm
 493          * because it calls destroy_context()
 494          */
 495         mm_free_pgd(mm);
 496         free_mm(mm);
 497         return retval;
 498 }
 499
 500 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
 501 {
 502         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
 503         /* We don't need to lock fs - think why ;-) */
 504         if (fs) {
 505                 atomic_set(&fs->count, 1);
 506                 rwlock_init(&fs->lock);
 507                 fs->umask = old->umask;
 508                 read_lock(&old->lock);
 509                 fs->rootmnt = mntget(old->rootmnt);
 510                 fs->root = dget(old->root);
 511                 fs->pwdmnt = mntget(old->pwdmnt);
 512                 fs->pwd = dget(old->pwd);
 513                 if (old->altroot) {
 514                         fs->altrootmnt = mntget(old->altrootmnt);
 515                         fs->altroot = dget(old->altroot);
 516                 } else {
 517                         fs->altrootmnt = NULL;
 518                         fs->altroot = NULL;
 519                 }
 520                 read_unlock(&old->lock);
 521         }
 522         return fs;
 523 }
 524
 525 struct fs_struct *copy_fs_struct(struct fs_struct *old)
 526 {
 527         return __copy_fs_struct(old);
 528 }
 529
 530 EXPORT_SYMBOL_GPL(copy_fs_struct);
 531
 532 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
 533 {
 534         if (clone_flags & CLONE_FS) {
 535                 atomic_inc(&current->fs->count);
 536                 return 0;
 537         }
 538         tsk->fs = __copy_fs_struct(current->fs);
 539         if (!tsk->fs)
 540                 return -ENOMEM;
 541         return 0;
 542 }
 543
 544 static int count_open_files(struct files_struct *files, int size)
 545 {
 546         int i;
 547
 548         /* Find the last open fd */
 549         for (i = size/(8*sizeof(long)); i > 0; ) {
 550                 if (files->open_fds->fds_bits[--i])
 551                         break;
 552         }
 553         i = (i+1) * 8 * sizeof(long);
 554         return i;
 555 }
 556
 557 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 558 {
 559         struct files_struct *oldf, *newf;
 560         struct file **old_fds, **new_fds;
 561         int open_files, nfds, size, i, error = 0;
 562
 563         /*
 564          * A background process may not have any files ...
 565          */
 566         oldf = current->files;
 567         if (!oldf)
 568                 goto out;
 569
 570         if (clone_flags & CLONE_FILES) {
 571                 atomic_inc(&oldf->count);
 572                 goto out;
 573         }
 574
 575         /*
 576          * Note: we may be using current for both targets (See exec.c)
 577          * This works because we cache current->files (old) as oldf. Don't
 578          * break this.
 579          */
 580         tsk->files = NULL;
 581         error = -ENOMEM;
 582         newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
 583         if (!newf)
 584                 goto out;
 585
 586         atomic_set(&newf->count, 1);
 587
 588         spin_lock_init(&newf->file_lock);
 589         newf->next_fd       = 0;
 590         newf->max_fds       = NR_OPEN_DEFAULT;
 591         newf->max_fdset     = __FD_SETSIZE;
 592         newf->close_on_exec = &newf->close_on_exec_init;
 593         newf->open_fds      = &newf->open_fds_init;
 594         newf->fd            = &newf->fd_array[0];
 595
 596         /* We don't yet have the oldf readlock, but even if the old
 597            fdset gets grown now, we'll only copy up to "size" fds */
 598         size = oldf->max_fdset;
 599         if (size > __FD_SETSIZE) {
 600                 newf->max_fdset = 0;
 601                 spin_lock(&newf->file_lock);
 602                 error = expand_fdset(newf, size-1);
 603                 spin_unlock(&newf->file_lock);
 604                 if (error)
 605                         goto out_release;
 606         }
 607         spin_lock(&oldf->file_lock);
 608
 609         open_files = count_open_files(oldf, size);
 610
 611         /*
 612          * Check whether we need to allocate a larger fd array.
 613          * Note: we're not a clone task, so the open count won't
 614          * change.
 615          */
 616         nfds = NR_OPEN_DEFAULT;
 617         if (open_files > nfds) {
 618                 spin_unlock(&oldf->file_lock);
 619                 newf->max_fds = 0;
 620                 spin_lock(&newf->file_lock);
 621                 error = expand_fd_array(newf, open_files-1);
 622                 spin_unlock(&newf->file_lock);
 623                 if (error)
 624                         goto out_release;
 625                 nfds = newf->max_fds;
 626                 spin_lock(&oldf->file_lock);
 627         }
 628
 629         old_fds = oldf->fd;
 630         new_fds = newf->fd;
 631
 632         memcpy(newf->open_fds->fds_bits, oldf->open_fds->fds_bits, open_files/8);
 633         memcpy(newf->close_on_exec->fds_bits, oldf->close_on_exec->fds_bits, open_files/8);
 634
 635         for (i = open_files; i != 0; i--) {
 636                 struct file *f = *old_fds++;
 637                 if (f) {
 638                         get_file(f);
 639                 } else {
 640                         /*
 641                          * The fd may be claimed in the fd bitmap but not yet
 642                          * instantiated in the files array if a sibling thread
 643                          * is partway through open().  So make sure that this
 644                          * fd is available to the new process.
 645                          */
 646                         FD_CLR(open_files - i, newf->open_fds);
 647                 }
 648                 *new_fds++ = f;
 649         }
 650         spin_unlock(&oldf->file_lock);
 651
 652         /* compute the remainder to be cleared */
 653         size = (newf->max_fds - open_files) * sizeof(struct file *);
 654
 655         /* This is long word aligned thus could use a optimized version */
 656         memset(new_fds, 0, size);
 657
 658         if (newf->max_fdset > open_files) {
 659                 int left = (newf->max_fdset-open_files)/8;
 660                 int start = open_files / (8 * sizeof(unsigned long));
 661
 662                 memset(&newf->open_fds->fds_bits[start], 0, left);
 663                 memset(&newf->close_on_exec->fds_bits[start], 0, left);
 664         }
 665
 666         tsk->files = newf;
 667         error = 0;
 668 out:
 669         return error;
 670
 671 out_release:
 672         free_fdset (newf->close_on_exec, newf->max_fdset);
 673         free_fdset (newf->open_fds, newf->max_fdset);
 674         kmem_cache_free(files_cachep, newf);
 675         goto out;
 676 }
 677
 678 /*
 679  *      Helper to unshare the files of the current task.
 680  *      We don't want to expose copy_files internals to
 681  *      the exec layer of the kernel.
 682  */
 683
 684 int unshare_files(void)
 685 {
 686         struct files_struct *files  = current->files;
 687         int rc;
 688
 689         if(!files)
 690                 BUG();
 691
 692         /* This can race but the race causes us to copy when we don't
 693            need to and drop the copy */
 694         if(atomic_read(&files->count) == 1)
 695         {
 696                 atomic_inc(&files->count);
 697                 return 0;
 698         }
 699         rc = copy_files(0, current);
 700         if(rc)
 701                 current->files = files;
 702         return rc;
 703 }
 704
 705 EXPORT_SYMBOL(unshare_files);
 706
 707 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
 708 {
 709         struct sighand_struct *sig;
 710
 711         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
 712                 atomic_inc(&current->sighand->count);
 713                 return 0;
 714         }
 715         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 716         tsk->sighand = sig;
 717         if (!sig)
 718                 return -ENOMEM;
 719         spin_lock_init(&sig->siglock);
 720         atomic_set(&sig->count, 1);
 721         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 722         return 0;
 723 }
 724
 725 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
 726 {
 727         struct signal_struct *sig;
 728
 729         if (clone_flags & CLONE_THREAD) {
 730                 atomic_inc(&current->signal->count);
 731                 atomic_inc(&current->signal->live);
 732                 return 0;
 733         }
 734         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 735         tsk->signal = sig;
 736         if (!sig)
 737                 return -ENOMEM;
 738         atomic_set(&sig->count, 1);
 739         atomic_set(&sig->live, 1);
 740         sig->group_exit = 0;
 741         sig->group_exit_code = 0;
 742         sig->group_exit_task = NULL;
 743         sig->group_stop_count = 0;
 744         sig->stop_state = 0;
 745         sig->curr_target = NULL;
 746         init_sigpending(&sig->shared_pending);
 747         INIT_LIST_HEAD(&sig->posix_timers);
 748
 749         sig->tty = current->signal->tty;
 750         sig->pgrp = process_group(current);
 751         sig->session = current->signal->session;
 752         sig->leader = 0;        /* session leadership doesn't inherit */
 753         sig->tty_old_pgrp = 0;
 754
 755         sig->utime = sig->stime = sig->cutime = sig->cstime = 0;
 756         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
 757         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
 758
 759         task_lock(current->group_leader);
 760         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
 761         task_unlock(current->group_leader);
 762
 763         return 0;
 764 }
 765
 766 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
 767 {
 768         unsigned long new_flags = p->flags;
 769
 770         new_flags &= ~PF_SUPERPRIV;
 771         new_flags |= PF_FORKNOEXEC;
 772         if (!(clone_flags & CLONE_PTRACE))
 773                 p->ptrace = 0;
 774         p->flags = new_flags;
 775 }
 776
 777 asmlinkage long sys_set_tid_address(int __user *tidptr)
 778 {
 779         current->clear_child_tid = tidptr;
 780
 781         return current->pid;
 782 }
 783
 784 /*
 785  * This creates a new process as a copy of the old one,
 786  * but does not actually start it yet.
 787  *
 788  * It copies the registers, and all the appropriate
 789  * parts of the process environment (as per the clone
 790  * flags). The actual kick-off is left to the caller.
 791  */
 792 static task_t *copy_process(unsigned long clone_flags,
 793                                  unsigned long stack_start,
 794                                  struct pt_regs *regs,
 795                                  unsigned long stack_size,
 796                                  int __user *parent_tidptr,
 797                                  int __user *child_tidptr,
 798                                  int pid)
 799 {
 800         int retval;
 801         struct task_struct *p = NULL;
 802         struct vx_info *vxi;
 803
 804         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 805                 return ERR_PTR(-EINVAL);
 806
 807         /*
 808          * Thread groups must share signals as well, and detached threads
 809          * can only be started up within the thread group.
 810          */
 811         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
 812                 return ERR_PTR(-EINVAL);
 813
 814         /*
 815          * Shared signal handlers imply shared VM. By way of the above,
 816          * thread groups also imply shared VM. Blocking this case allows
 817          * for various simplifications in other code.
 818          */
 819         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
 820                 return ERR_PTR(-EINVAL);
 821
 822         retval = security_task_create(clone_flags);
 823         if (retval)
 824                 goto fork_out;
 825
 826         retval = -ENOMEM;
 827         p = dup_task_struct(current);
 828         if (!p)
 829                 goto fork_out;
 830         p->tux_info = NULL;
 831
 832         p->vx_info = NULL;
 833         set_vx_info(&p->vx_info, current->vx_info);
 834         p->nx_info = NULL;
 835         set_nx_info(&p->nx_info, current->nx_info);
 836
 837         /* check vserver memory */
 838         if (p->mm && !(clone_flags & CLONE_VM)) {
 839                 if (vx_vmpages_avail(p->mm, p->mm->total_vm))
 840                         vx_pages_add(p->mm->mm_vx_info, RLIMIT_AS, p->mm->total_vm);
 841                 else
 842                         goto bad_fork_free;
 843         }
 844         if (p->mm && vx_flags(VXF_FORK_RSS, 0)) {
 845                 if (!vx_rsspages_avail(p->mm, p->mm->rss))
 846                         goto bad_fork_cleanup_vm;
 847         }
 848
 849         p->vx_info = NULL;
 850         set_vx_info(&p->vx_info, current->vx_info);
 851         p->nx_info = NULL;
 852         set_nx_info(&p->nx_info, current->nx_info);
 853
 854         /* check vserver memory */
 855         if (p->mm && !(clone_flags & CLONE_VM)) {
 856                 if (vx_vmpages_avail(p->mm, p->mm->total_vm))
 857                         vx_pages_add(p->mm->mm_vx_info, RLIMIT_AS, p->mm->total_vm);
 858                 else
 859                         goto bad_fork_free;
 860         }
 861         if (p->mm && vx_flags(VXF_FORK_RSS, 0)) {
 862                 if (!vx_rsspages_avail(p->mm, p->mm->rss))
 863                         goto bad_fork_cleanup_vm;
 864         }
 865
 866         retval = -EAGAIN;
 867         if (!vx_nproc_avail(1))
 868                 goto bad_fork_cleanup_vm;
 869
 870         if (atomic_read(&p->user->processes) >=
 871                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
 872                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
 873                                 p->user != &root_user)
 874                         goto bad_fork_cleanup_vm;
 875         }
 876
 877         atomic_inc(&p->user->__count);
 878         atomic_inc(&p->user->processes);
 879         get_group_info(p->group_info);
 880
 881         /*
 882          * If multiple threads are within copy_process(), then this check
 883          * triggers too late. This doesn't hurt, the check is only there
 884          * to stop root fork bombs.
 885          */
 886         if (nr_threads >= max_threads)
 887                 goto bad_fork_cleanup_count;
 888
 889         if (!try_module_get(p->thread_info->exec_domain->module))
 890                 goto bad_fork_cleanup_count;
 891
 892         if (p->binfmt && !try_module_get(p->binfmt->module))
 893                 goto bad_fork_cleanup_put_domain;
 894
 895         p->did_exec = 0;
 896         copy_flags(clone_flags, p);
 897         p->pid = pid;
 898         retval = -EFAULT;
 899         if (clone_flags & CLONE_PARENT_SETTID)
 900                 if (put_user(p->pid, parent_tidptr))
 901                         goto bad_fork_cleanup;
 902
 903         p->proc_dentry = NULL;
 904
 905         INIT_LIST_HEAD(&p->children);
 906         INIT_LIST_HEAD(&p->sibling);
 907         init_waitqueue_head(&p->wait_chldexit);
 908         p->vfork_done = NULL;
 909         spin_lock_init(&p->alloc_lock);
 910         spin_lock_init(&p->proc_lock);
 911
 912         clear_tsk_thread_flag(p, TIF_SIGPENDING);
 913         init_sigpending(&p->pending);
 914
 915         p->it_real_value = p->it_virt_value = p->it_prof_value = 0;
 916         p->it_real_incr = p->it_virt_incr = p->it_prof_incr = 0;
 917         init_timer(&p->real_timer);
 918         p->real_timer.data = (unsigned long) p;
 919
 920         p->utime = p->stime = 0;
 921         p->lock_depth = -1;             /* -1 = no lock */
 922         do_posix_clock_monotonic_gettime(&p->start_time);
 923         p->security = NULL;
 924         p->io_context = NULL;
 925         p->io_wait = NULL;
 926         p->audit_context = NULL;
 927 #ifdef CONFIG_NUMA
 928         p->mempolicy = mpol_copy(p->mempolicy);
 929         if (IS_ERR(p->mempolicy)) {
 930                 retval = PTR_ERR(p->mempolicy);
 931                 p->mempolicy = NULL;
 932                 goto bad_fork_cleanup;
 933         }
 934 #endif
 935
 936         p->tgid = p->pid;
 937         if (clone_flags & CLONE_THREAD)
 938                 p->tgid = current->tgid;
 939
 940         if ((retval = security_task_alloc(p)))
 941                 goto bad_fork_cleanup_policy;
 942         if ((retval = audit_alloc(p)))
 943                 goto bad_fork_cleanup_security;
 944         /* copy all the process information */
 945         if ((retval = copy_semundo(clone_flags, p)))
 946                 goto bad_fork_cleanup_audit;
 947         if ((retval = copy_files(clone_flags, p)))
 948                 goto bad_fork_cleanup_semundo;
 949         if ((retval = copy_fs(clone_flags, p)))
 950                 goto bad_fork_cleanup_files;
 951         if ((retval = copy_sighand(clone_flags, p)))
 952                 goto bad_fork_cleanup_fs;
 953         if ((retval = copy_signal(clone_flags, p)))
 954                 goto bad_fork_cleanup_sighand;
 955         if ((retval = copy_mm(clone_flags, p)))
 956                 goto bad_fork_cleanup_signal;
 957         if ((retval = copy_keys(clone_flags, p)))
 958                 goto bad_fork_cleanup_mm;
 959         if ((retval = copy_namespace(clone_flags, p)))
 960                 goto bad_fork_cleanup_keys;
 961         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
 962         if (retval)
 963                 goto bad_fork_cleanup_namespace;
 964
 965         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
 966         /*
 967          * Clear TID on mm_release()?
 968          */
 969         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
 970
 971         /*
 972          * Syscall tracing should be turned off in the child regardless
 973          * of CLONE_PTRACE.
 974          */
 975         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
 976
 977         /* Our parent execution domain becomes current domain
 978            These must match for thread signalling to apply */
 979
 980         p->parent_exec_id = p->self_exec_id;
 981
 982         /* ok, now we should be set up.. */
 983         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
 984         p->pdeath_signal = 0;
 985         p->exit_state = 0;
 986
 987         /* Perform scheduler related setup */
 988         sched_fork(p);
 989
 990         /*
 991          * Ok, make it visible to the rest of the system.
 992          * We dont wake it up yet.
 993          */
 994         p->group_leader = p;
 995         INIT_LIST_HEAD(&p->ptrace_children);
 996         INIT_LIST_HEAD(&p->ptrace_list);
 997
 998         /* Need tasklist lock for parent etc handling! */
 999         write_lock_irq(&tasklist_lock);
1000
1001         /*
1002          * The task hasn't been attached yet, so cpus_allowed mask cannot
1003          * have changed. The cpus_allowed mask of the parent may have
1004          * changed after it was copied first time, and it may then move to
1005          * another CPU - so we re-copy it here and set the child's CPU to
1006          * the parent's CPU. This avoids alot of nasty races.
1007          */
1008         p->cpus_allowed = current->cpus_allowed;
1009         set_task_cpu(p, smp_processor_id());
1010
1011         /*
1012          * Check for pending SIGKILL! The new thread should not be allowed
1013          * to slip out of an OOM kill. (or normal SIGKILL.)
1014          */
1015         if (sigismember(&current->pending.signal, SIGKILL)) {
1016                 write_unlock_irq(&tasklist_lock);
1017                 retval = -EINTR;
1018                 goto bad_fork_cleanup_namespace;
1019         }
1020
1021         /* CLONE_PARENT re-uses the old parent */
1022         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1023                 p->real_parent = current->real_parent;
1024         else
1025                 p->real_parent = current;
1026         p->parent = p->real_parent;
1027
1028         if (clone_flags & CLONE_THREAD) {
1029                 spin_lock(&current->sighand->siglock);
1030                 /*
1031                  * Important: if an exit-all has been started then
1032                  * do not create this new thread - the whole thread
1033                  * group is supposed to exit anyway.
1034                  */
1035                 if (current->signal->group_exit) {
1036                         spin_unlock(&current->sighand->siglock);
1037                         write_unlock_irq(&tasklist_lock);
1038                         retval = -EAGAIN;
1039                         goto bad_fork_cleanup_namespace;
1040                 }
1041                 p->group_leader = current->group_leader;
1042
1043                 if (current->signal->group_stop_count > 0) {
1044                         /*
1045                          * There is an all-stop in progress for the group.
1046                          * We ourselves will stop as soon as we check signals.
1047                          * Make the new thread part of that group stop too.
1048                          */
1049                         current->signal->group_stop_count++;
1050                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1051                 }
1052
1053                 spin_unlock(&current->sighand->siglock);
1054         }
1055
1056         SET_LINKS(p);
1057         if (unlikely(p->ptrace & PT_PTRACED))
1058                 __ptrace_link(p, current->parent);
1059
1060         attach_pid(p, PIDTYPE_PID, p->pid);
1061         attach_pid(p, PIDTYPE_TGID, p->tgid);
1062         if (thread_group_leader(p)) {
1063                 attach_pid(p, PIDTYPE_PGID, process_group(p));
1064                 attach_pid(p, PIDTYPE_SID, p->signal->session);
1065                 if (p->pid)
1066                         __get_cpu_var(process_counts)++;
1067         }
1068
1069         p->ioprio = current->ioprio;
1070         nr_threads++;
1071         /* p is copy of current */
1072         vxi = p->vx_info;
1073         if (vxi) {
1074                 atomic_inc(&vxi->cvirt.nr_threads);
1075                 vx_nproc_inc(p);
1076         }
1077         write_unlock_irq(&tasklist_lock);
1078         retval = 0;
1079
1080 fork_out:
1081         if (retval)
1082                 return ERR_PTR(retval);
1083         return p;
1084
1085 bad_fork_cleanup_namespace:
1086         exit_namespace(p);
1087 bad_fork_cleanup_keys:
1088         exit_keys(p);
1089 bad_fork_cleanup_mm:
1090         if (p->mm)
1091                 mmput(p->mm);
1092 bad_fork_cleanup_signal:
1093         exit_signal(p);
1094 bad_fork_cleanup_sighand:
1095         exit_sighand(p);
1096 bad_fork_cleanup_fs:
1097         exit_fs(p); /* blocking */
1098 bad_fork_cleanup_files:
1099         exit_files(p); /* blocking */
1100 bad_fork_cleanup_semundo:
1101         exit_sem(p);
1102 bad_fork_cleanup_audit:
1103         audit_free(p);
1104 bad_fork_cleanup_security:
1105         security_task_free(p);
1106 bad_fork_cleanup_policy:
1107 #ifdef CONFIG_NUMA
1108         mpol_free(p->mempolicy);
1109 #endif
1110 bad_fork_cleanup:
1111         if (p->binfmt)
1112                 module_put(p->binfmt->module);
1113 bad_fork_cleanup_put_domain:
1114         module_put(p->thread_info->exec_domain->module);
1115 bad_fork_cleanup_count:
1116         put_group_info(p->group_info);
1117         atomic_dec(&p->user->processes);
1118         free_uid(p->user);
1119 bad_fork_cleanup_vm:
1120         if (p->mm && !(clone_flags & CLONE_VM))
1121                 vx_pages_sub(p->mm->mm_vx_info, RLIMIT_AS, p->mm->total_vm);
1122 bad_fork_free:
1123         free_task(p);
1124         goto fork_out;
1125 }
1126
1127 struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1128 {
1129         memset(regs, 0, sizeof(struct pt_regs));
1130         return regs;
1131 }
1132
1133 task_t * __devinit fork_idle(int cpu)
1134 {
1135         task_t *task;
1136         struct pt_regs regs;
1137
1138         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL, NULL, 0);
1139         if (!task)
1140                 return ERR_PTR(-ENOMEM);
1141         init_idle(task, cpu);
1142         unhash_process(task);
1143         return task;
1144 }
1145
1146 static inline int fork_traceflag (unsigned clone_flags)
1147 {
1148         if (clone_flags & CLONE_UNTRACED)
1149                 return 0;
1150         else if (clone_flags & CLONE_VFORK) {
1151                 if (current->ptrace & PT_TRACE_VFORK)
1152                         return PTRACE_EVENT_VFORK;
1153         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1154                 if (current->ptrace & PT_TRACE_CLONE)
1155                         return PTRACE_EVENT_CLONE;
1156         } else if (current->ptrace & PT_TRACE_FORK)
1157                 return PTRACE_EVENT_FORK;
1158
1159         return 0;
1160 }
1161
1162 /*
1163  *  Ok, this is the main fork-routine.
1164  *
1165  * It copies the process, and if successful kick-starts
1166  * it and waits for it to finish using the VM if required.
1167  */
1168 long do_fork(unsigned long clone_flags,
1169               unsigned long stack_start,
1170               struct pt_regs *regs,
1171               unsigned long stack_size,
1172               int __user *parent_tidptr,
1173               int __user *child_tidptr)
1174 {
1175         struct task_struct *p;
1176         int trace = 0;
1177         long pid = alloc_pidmap();
1178
1179         if (pid < 0)
1180                 return -EAGAIN;
1181         if (unlikely(current->ptrace)) {
1182                 trace = fork_traceflag (clone_flags);
1183                 if (trace)
1184                         clone_flags |= CLONE_PTRACE;
1185         }
1186
1187         p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, pid);
1188
1189         /*
1190          * Do this prior waking up the new thread - the thread pointer
1191          * might get invalid after that point, if the thread exits quickly.
1192          */
1193         if (!IS_ERR(p)) {
1194                 struct completion vfork;
1195
1196                 if (clone_flags & CLONE_VFORK) {
1197                         p->vfork_done = &vfork;
1198                         init_completion(&vfork);
1199                 }
1200
1201                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1202                         /*
1203                          * We'll start up with an immediate SIGSTOP.
1204                          */
1205                         sigaddset(&p->pending.signal, SIGSTOP);
1206                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1207                 }
1208
1209                 if (!(clone_flags & CLONE_STOPPED))
1210                         wake_up_new_task(p, clone_flags);
1211                 else
1212                         p->state = TASK_STOPPED;
1213                 ++total_forks;
1214
1215                 if (unlikely (trace)) {
1216                         current->ptrace_message = pid;
1217                         ptrace_notify ((trace << 8) | SIGTRAP);
1218                 }
1219
1220                 if (clone_flags & CLONE_VFORK) {
1221                         wait_for_completion(&vfork);
1222                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE))
1223                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1224                 }
1225         } else {
1226                 free_pidmap(pid);
1227                 pid = PTR_ERR(p);
1228         }
1229         return pid;
1230 }
1231
1232 /* SLAB cache for signal_struct structures (tsk->signal) */
1233 kmem_cache_t *signal_cachep;
1234
1235 /* SLAB cache for sighand_struct structures (tsk->sighand) */
1236 kmem_cache_t *sighand_cachep;
1237
1238 /* SLAB cache for files_struct structures (tsk->files) */
1239 kmem_cache_t *files_cachep;
1240
1241 /* SLAB cache for fs_struct structures (tsk->fs) */
1242 kmem_cache_t *fs_cachep;
1243
1244 /* SLAB cache for vm_area_struct structures */
1245 kmem_cache_t *vm_area_cachep;
1246
1247 /* SLAB cache for mm_struct structures (tsk->mm) */
1248 kmem_cache_t *mm_cachep;
1249
1250 void __init proc_caches_init(void)
1251 {
1252         sighand_cachep = kmem_cache_create("sighand_cache",
1253                         sizeof(struct sighand_struct), 0,
1254                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1255         signal_cachep = kmem_cache_create("signal_cache",
1256                         sizeof(struct signal_struct), 0,
1257                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1258         files_cachep = kmem_cache_create("files_cache",
1259                         sizeof(struct files_struct), 0,
1260                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1261         fs_cachep = kmem_cache_create("fs_cache",
1262                         sizeof(struct fs_struct), 0,
1263                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1264         vm_area_cachep = kmem_cache_create("vm_area_struct",
1265                         sizeof(struct vm_area_struct), 0,
1266                         SLAB_PANIC, NULL, NULL);
1267         mm_cachep = kmem_cache_create("mm_struct",
1268                         sizeof(struct mm_struct), 0,
1269                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1270 }