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/slab.h>
  15 #include <linux/init.h>
  16 #include <linux/unistd.h>
  17 #include <linux/module.h>
  18 #include <linux/vmalloc.h>
  19 #include <linux/completion.h>
  20 #include <linux/mnt_namespace.h>
  21 #include <linux/personality.h>
  22 #include <linux/mempolicy.h>
  23 #include <linux/sem.h>
  24 #include <linux/file.h>
  25 #include <linux/fdtable.h>
  26 #include <linux/iocontext.h>
  27 #include <linux/key.h>
  28 #include <linux/binfmts.h>
  29 #include <linux/mman.h>
  30 #include <linux/mmu_notifier.h>
  31 #include <linux/fs.h>
  32 #include <linux/nsproxy.h>
  33 #include <linux/capability.h>
  34 #include <linux/cpu.h>
  35 #include <linux/cgroup.h>
  36 #include <linux/security.h>
  37 #include <linux/hugetlb.h>
  38 #include <linux/swap.h>
  39 #include <linux/syscalls.h>
  40 #include <linux/jiffies.h>
  41 #include <linux/tracehook.h>
  42 #include <linux/futex.h>
  43 #include <linux/compat.h>
  44 #include <linux/task_io_accounting_ops.h>
  45 #include <linux/rcupdate.h>
  46 #include <linux/ptrace.h>
  47 #include <linux/mount.h>
  48 #include <linux/audit.h>
  49 #include <linux/memcontrol.h>
  50 #include <linux/ftrace.h>
  51 #include <linux/profile.h>
  52 #include <linux/rmap.h>
  53 #include <linux/acct.h>
  54 #include <linux/tsacct_kern.h>
  55 #include <linux/cn_proc.h>
  56 #include <linux/freezer.h>
  57 #include <linux/delayacct.h>
  58 #include <linux/taskstats_kern.h>
  59 #include <linux/random.h>
  60 #include <linux/tty.h>
  61 #include <linux/proc_fs.h>
  62 #include <linux/blkdev.h>
  63 #include <trace/sched.h>
  64
  65 #include <asm/pgtable.h>
  66 #include <asm/pgalloc.h>
  67 #include <asm/uaccess.h>
  68 #include <asm/mmu_context.h>
  69 #include <asm/cacheflush.h>
  70 #include <asm/tlbflush.h>
  71
  72 /*
  73  * Protected counters by write_lock_irq(&tasklist_lock)
  74  */
  75 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  76 int nr_threads;                 /* The idle threads do not count.. */
  77
  78 int max_threads;                /* tunable limit on nr_threads */
  79
  80 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  81
  82 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  83
  84 DEFINE_TRACE(sched_process_fork);
  85
  86 int nr_processes(void)
  87 {
  88         int cpu;
  89         int total = 0;
  90
  91         for_each_online_cpu(cpu)
  92                 total += per_cpu(process_counts, cpu);
  93
  94         return total;
  95 }
  96
  97 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
  98 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
  99 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
 100 static struct kmem_cache *task_struct_cachep;
 101 #endif
 102
 103 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
 104 static inline struct thread_info *alloc_thread_info(struct task_struct *tsk)
 105 {
 106 #ifdef CONFIG_DEBUG_STACK_USAGE
 107         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
 108 #else
 109         gfp_t mask = GFP_KERNEL;
 110 #endif
 111         return (struct thread_info *)__get_free_pages(mask, THREAD_SIZE_ORDER);
 112 }
 113
 114 static inline void free_thread_info(struct thread_info *ti)
 115 {
 116         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 117 }
 118 #endif
 119
 120 /* SLAB cache for signal_struct structures (tsk->signal) */
 121 static struct kmem_cache *signal_cachep;
 122
 123 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 124 struct kmem_cache *sighand_cachep;
 125
 126 /* SLAB cache for files_struct structures (tsk->files) */
 127 struct kmem_cache *files_cachep;
 128
 129 /* SLAB cache for fs_struct structures (tsk->fs) */
 130 struct kmem_cache *fs_cachep;
 131
 132 /* SLAB cache for vm_area_struct structures */
 133 struct kmem_cache *vm_area_cachep;
 134
 135 /* SLAB cache for mm_struct structures (tsk->mm) */
 136 static struct kmem_cache *mm_cachep;
 137
 138 void free_task(struct task_struct *tsk)
 139 {
 140         prop_local_destroy_single(&tsk->dirties);
 141         free_thread_info(tsk->stack);
 142         rt_mutex_debug_task_free(tsk);
 143         ftrace_graph_exit_task(tsk);
 144         free_task_struct(tsk);
 145 }
 146 EXPORT_SYMBOL(free_task);
 147
 148 void __put_task_struct(struct task_struct *tsk)
 149 {
 150         WARN_ON(!tsk->exit_state);
 151         WARN_ON(atomic_read(&tsk->usage));
 152         WARN_ON(tsk == current);
 153
 154         security_task_free(tsk);
 155         free_uid(tsk->user);
 156         put_group_info(tsk->group_info);
 157         delayacct_tsk_free(tsk);
 158
 159         if (!profile_handoff_task(tsk))
 160                 free_task(tsk);
 161 }
 162
 163 /*
 164  * macro override instead of weak attribute alias, to workaround
 165  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
 166  */
 167 #ifndef arch_task_cache_init
 168 #define arch_task_cache_init()
 169 #endif
 170
 171 void __init fork_init(unsigned long mempages)
 172 {
 173 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 174 #ifndef ARCH_MIN_TASKALIGN
 175 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 176 #endif
 177         /* create a slab on which task_structs can be allocated */
 178         task_struct_cachep =
 179                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 180                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL);
 181 #endif
 182
 183         /* do the arch specific task caches init */
 184         arch_task_cache_init();
 185
 186         /*
 187          * The default maximum number of threads is set to a safe
 188          * value: the thread structures can take up at most half
 189          * of memory.
 190          */
 191         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 192
 193         /*
 194          * we need to allow at least 20 threads to boot a system
 195          */
 196         if(max_threads < 20)
 197                 max_threads = 20;
 198
 199         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 200         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 201         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 202                 init_task.signal->rlim[RLIMIT_NPROC];
 203 }
 204
 205 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
 206                                                struct task_struct *src)
 207 {
 208         *dst = *src;
 209         return 0;
 210 }
 211
 212 static struct task_struct *dup_task_struct(struct task_struct *orig)
 213 {
 214         struct task_struct *tsk;
 215         struct thread_info *ti;
 216         int err;
 217
 218         prepare_to_copy(orig);
 219
 220         tsk = alloc_task_struct();
 221         if (!tsk)
 222                 return NULL;
 223
 224         ti = alloc_thread_info(tsk);
 225         if (!ti) {
 226                 free_task_struct(tsk);
 227                 return NULL;
 228         }
 229
 230         err = arch_dup_task_struct(tsk, orig);
 231         if (err)
 232                 goto out;
 233
 234         tsk->stack = ti;
 235
 236         err = prop_local_init_single(&tsk->dirties);
 237         if (err)
 238                 goto out;
 239
 240         setup_thread_stack(tsk, orig);
 241
 242 #ifdef CONFIG_CC_STACKPROTECTOR
 243         tsk->stack_canary = get_random_int();
 244 #endif
 245
 246         /* One for us, one for whoever does the "release_task()" (usually parent) */
 247         atomic_set(&tsk->usage,2);
 248         atomic_set(&tsk->fs_excl, 0);
 249 #ifdef CONFIG_BLK_DEV_IO_TRACE
 250         tsk->btrace_seq = 0;
 251 #endif
 252         tsk->splice_pipe = NULL;
 253         return tsk;
 254
 255 out:
 256         free_thread_info(ti);
 257         free_task_struct(tsk);
 258         return NULL;
 259 }
 260
 261 #ifdef CONFIG_MMU
 262 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 263 {
 264         struct vm_area_struct *mpnt, *tmp, **pprev;
 265         struct rb_node **rb_link, *rb_parent;
 266         int retval;
 267         unsigned long charge;
 268         struct mempolicy *pol;
 269
 270         down_write(&oldmm->mmap_sem);
 271         flush_cache_dup_mm(oldmm);
 272         /*
 273          * Not linked in yet - no deadlock potential:
 274          */
 275         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 276
 277         mm->locked_vm = 0;
 278         mm->mmap = NULL;
 279         mm->mmap_cache = NULL;
 280         mm->free_area_cache = oldmm->mmap_base;
 281         mm->cached_hole_size = ~0UL;
 282         mm->map_count = 0;
 283         cpus_clear(mm->cpu_vm_mask);
 284         mm->mm_rb = RB_ROOT;
 285         rb_link = &mm->mm_rb.rb_node;
 286         rb_parent = NULL;
 287         pprev = &mm->mmap;
 288
 289         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 290                 struct file *file;
 291
 292                 if (mpnt->vm_flags & VM_DONTCOPY) {
 293                         long pages = vma_pages(mpnt);
 294                         mm->total_vm -= pages;
 295                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 296                                                                 -pages);
 297                         continue;
 298                 }
 299                 charge = 0;
 300                 if (mpnt->vm_flags & VM_ACCOUNT) {
 301                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 302                         if (security_vm_enough_memory(len))
 303                                 goto fail_nomem;
 304                         charge = len;
 305                 }
 306                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 307                 if (!tmp)
 308                         goto fail_nomem;
 309                 *tmp = *mpnt;
 310                 pol = mpol_dup(vma_policy(mpnt));
 311                 retval = PTR_ERR(pol);
 312                 if (IS_ERR(pol))
 313                         goto fail_nomem_policy;
 314                 vma_set_policy(tmp, pol);
 315                 tmp->vm_flags &= ~VM_LOCKED;
 316                 tmp->vm_mm = mm;
 317                 tmp->vm_next = NULL;
 318                 anon_vma_link(tmp);
 319                 file = tmp->vm_file;
 320                 if (file) {
 321                         struct inode *inode = file->f_path.dentry->d_inode;
 322                         get_file(file);
 323                         if (tmp->vm_flags & VM_DENYWRITE)
 324                                 atomic_dec(&inode->i_writecount);
 325
 326                         /* insert tmp into the share list, just after mpnt */
 327                         spin_lock(&file->f_mapping->i_mmap_lock);
 328                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
 329                         flush_dcache_mmap_lock(file->f_mapping);
 330                         vma_prio_tree_add(tmp, mpnt);
 331                         flush_dcache_mmap_unlock(file->f_mapping);
 332                         spin_unlock(&file->f_mapping->i_mmap_lock);
 333                 }
 334
 335                 /*
 336                  * Clear hugetlb-related page reserves for children. This only
 337                  * affects MAP_PRIVATE mappings. Faults generated by the child
 338                  * are not guaranteed to succeed, even if read-only
 339                  */
 340                 if (is_vm_hugetlb_page(tmp))
 341                         reset_vma_resv_huge_pages(tmp);
 342
 343                 /*
 344                  * Link in the new vma and copy the page table entries.
 345                  */
 346                 *pprev = tmp;
 347                 pprev = &tmp->vm_next;
 348
 349                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 350                 rb_link = &tmp->vm_rb.rb_right;
 351                 rb_parent = &tmp->vm_rb;
 352
 353                 mm->map_count++;
 354                 retval = copy_page_range(mm, oldmm, mpnt);
 355
 356                 if (tmp->vm_ops && tmp->vm_ops->open)
 357                         tmp->vm_ops->open(tmp);
 358
 359                 if (retval)
 360                         goto out;
 361         }
 362         /* a new mm has just been created */
 363         arch_dup_mmap(oldmm, mm);
 364         retval = 0;
 365 out:
 366         up_write(&mm->mmap_sem);
 367         flush_tlb_mm(oldmm);
 368         up_write(&oldmm->mmap_sem);
 369         return retval;
 370 fail_nomem_policy:
 371         kmem_cache_free(vm_area_cachep, tmp);
 372 fail_nomem:
 373         retval = -ENOMEM;
 374         vm_unacct_memory(charge);
 375         goto out;
 376 }
 377
 378 static inline int mm_alloc_pgd(struct mm_struct * mm)
 379 {
 380         mm->pgd = pgd_alloc(mm);
 381         if (unlikely(!mm->pgd))
 382                 return -ENOMEM;
 383         return 0;
 384 }
 385
 386 static inline void mm_free_pgd(struct mm_struct * mm)
 387 {
 388         pgd_free(mm, mm->pgd);
 389 }
 390 #else
 391 #define dup_mmap(mm, oldmm)     (0)
 392 #define mm_alloc_pgd(mm)        (0)
 393 #define mm_free_pgd(mm)
 394 #endif /* CONFIG_MMU */
 395
 396 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 397
 398 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 399 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 400
 401 #include <linux/init_task.h>
 402
 403 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
 404 {
 405         atomic_set(&mm->mm_users, 1);
 406         atomic_set(&mm->mm_count, 1);
 407         init_rwsem(&mm->mmap_sem);
 408         INIT_LIST_HEAD(&mm->mmlist);
 409         mm->flags = (current->mm) ? current->mm->flags
 410                                   : MMF_DUMP_FILTER_DEFAULT;
 411         mm->core_state = NULL;
 412         mm->nr_ptes = 0;
 413         set_mm_counter(mm, file_rss, 0);
 414         set_mm_counter(mm, anon_rss, 0);
 415         spin_lock_init(&mm->page_table_lock);
 416         rwlock_init(&mm->ioctx_list_lock);
 417         mm->ioctx_list = NULL;
 418         mm->free_area_cache = TASK_UNMAPPED_BASE;
 419         mm->cached_hole_size = ~0UL;
 420         mm_init_owner(mm, p);
 421
 422         if (likely(!mm_alloc_pgd(mm))) {
 423                 mm->def_flags = 0;
 424                 mmu_notifier_mm_init(mm);
 425                 return mm;
 426         }
 427
 428         free_mm(mm);
 429         return NULL;
 430 }
 431
 432 /*
 433  * Allocate and initialize an mm_struct.
 434  */
 435 struct mm_struct * mm_alloc(void)
 436 {
 437         struct mm_struct * mm;
 438
 439         mm = allocate_mm();
 440         if (mm) {
 441                 memset(mm, 0, sizeof(*mm));
 442                 mm = mm_init(mm, current);
 443         }
 444         return mm;
 445 }
 446
 447 /*
 448  * Called when the last reference to the mm
 449  * is dropped: either by a lazy thread or by
 450  * mmput. Free the page directory and the mm.
 451  */
 452 void __mmdrop(struct mm_struct *mm)
 453 {
 454         BUG_ON(mm == &init_mm);
 455         mm_free_pgd(mm);
 456         destroy_context(mm);
 457         mmu_notifier_mm_destroy(mm);
 458         free_mm(mm);
 459 }
 460 EXPORT_SYMBOL_GPL(__mmdrop);
 461
 462 /*
 463  * Decrement the use count and release all resources for an mm.
 464  */
 465 void mmput(struct mm_struct *mm)
 466 {
 467         might_sleep();
 468
 469         if (atomic_dec_and_test(&mm->mm_users)) {
 470                 exit_aio(mm);
 471                 exit_mmap(mm);
 472                 set_mm_exe_file(mm, NULL);
 473                 if (!list_empty(&mm->mmlist)) {
 474                         spin_lock(&mmlist_lock);
 475                         list_del(&mm->mmlist);
 476                         spin_unlock(&mmlist_lock);
 477                 }
 478                 put_swap_token(mm);
 479                 mmdrop(mm);
 480         }
 481 }
 482 EXPORT_SYMBOL_GPL(mmput);
 483
 484 /**
 485  * get_task_mm - acquire a reference to the task's mm
 486  *
 487  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 488  * this kernel workthread has transiently adopted a user mm with use_mm,
 489  * to do its AIO) is not set and if so returns a reference to it, after
 490  * bumping up the use count.  User must release the mm via mmput()
 491  * after use.  Typically used by /proc and ptrace.
 492  */
 493 struct mm_struct *get_task_mm(struct task_struct *task)
 494 {
 495         struct mm_struct *mm;
 496
 497         task_lock(task);
 498         mm = task->mm;
 499         if (mm) {
 500                 if (task->flags & PF_KTHREAD)
 501                         mm = NULL;
 502                 else
 503                         atomic_inc(&mm->mm_users);
 504         }
 505         task_unlock(task);
 506         return mm;
 507 }
 508 EXPORT_SYMBOL_GPL(get_task_mm);
 509
 510 /* Please note the differences between mmput and mm_release.
 511  * mmput is called whenever we stop holding onto a mm_struct,
 512  * error success whatever.
 513  *
 514  * mm_release is called after a mm_struct has been removed
 515  * from the current process.
 516  *
 517  * This difference is important for error handling, when we
 518  * only half set up a mm_struct for a new process and need to restore
 519  * the old one.  Because we mmput the new mm_struct before
 520  * restoring the old one. . .
 521  * Eric Biederman 10 January 1998
 522  */
 523 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 524 {
 525         struct completion *vfork_done = tsk->vfork_done;
 526
 527         /* Get rid of any futexes when releasing the mm */
 528 #ifdef CONFIG_FUTEX
 529         if (unlikely(tsk->robust_list))
 530                 exit_robust_list(tsk);
 531 #ifdef CONFIG_COMPAT
 532         if (unlikely(tsk->compat_robust_list))
 533                 compat_exit_robust_list(tsk);
 534 #endif
 535 #endif
 536
 537         /* Get rid of any cached register state */
 538         deactivate_mm(tsk, mm);
 539
 540         /* notify parent sleeping on vfork() */
 541         if (vfork_done) {
 542                 tsk->vfork_done = NULL;
 543                 complete(vfork_done);
 544         }
 545
 546         /*
 547          * If we're exiting normally, clear a user-space tid field if
 548          * requested.  We leave this alone when dying by signal, to leave
 549          * the value intact in a core dump, and to save the unnecessary
 550          * trouble otherwise.  Userland only wants this done for a sys_exit.
 551          */
 552         if (tsk->clear_child_tid
 553             && !(tsk->flags & PF_SIGNALED)
 554             && atomic_read(&mm->mm_users) > 1) {
 555                 u32 __user * tidptr = tsk->clear_child_tid;
 556                 tsk->clear_child_tid = NULL;
 557
 558                 /*
 559                  * We don't check the error code - if userspace has
 560                  * not set up a proper pointer then tough luck.
 561                  */
 562                 put_user(0, tidptr);
 563                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
 564         }
 565 }
 566
 567 /*
 568  * Allocate a new mm structure and copy contents from the
 569  * mm structure of the passed in task structure.
 570  */
 571 struct mm_struct *dup_mm(struct task_struct *tsk)
 572 {
 573         struct mm_struct *mm, *oldmm = current->mm;
 574         int err;
 575
 576         if (!oldmm)
 577                 return NULL;
 578
 579         mm = allocate_mm();
 580         if (!mm)
 581                 goto fail_nomem;
 582
 583         memcpy(mm, oldmm, sizeof(*mm));
 584
 585         /* Initializing for Swap token stuff */
 586         mm->token_priority = 0;
 587         mm->last_interval = 0;
 588
 589         if (!mm_init(mm, tsk))
 590                 goto fail_nomem;
 591
 592         if (init_new_context(tsk, mm))
 593                 goto fail_nocontext;
 594
 595         dup_mm_exe_file(oldmm, mm);
 596
 597         err = dup_mmap(mm, oldmm);
 598         if (err)
 599                 goto free_pt;
 600
 601         mm->hiwater_rss = get_mm_rss(mm);
 602         mm->hiwater_vm = mm->total_vm;
 603
 604         return mm;
 605
 606 free_pt:
 607         mmput(mm);
 608
 609 fail_nomem:
 610         return NULL;
 611
 612 fail_nocontext:
 613         /*
 614          * If init_new_context() failed, we cannot use mmput() to free the mm
 615          * because it calls destroy_context()
 616          */
 617         mm_free_pgd(mm);
 618         free_mm(mm);
 619         return NULL;
 620 }
 621
 622 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 623 {
 624         struct mm_struct * mm, *oldmm;
 625         int retval;
 626
 627         tsk->min_flt = tsk->maj_flt = 0;
 628         tsk->nvcsw = tsk->nivcsw = 0;
 629
 630         tsk->mm = NULL;
 631         tsk->active_mm = NULL;
 632
 633         /*
 634          * Are we cloning a kernel thread?
 635          *
 636          * We need to steal a active VM for that..
 637          */
 638         oldmm = current->mm;
 639         if (!oldmm)
 640                 return 0;
 641
 642         if (clone_flags & CLONE_VM) {
 643                 atomic_inc(&oldmm->mm_users);
 644                 mm = oldmm;
 645                 goto good_mm;
 646         }
 647
 648         retval = -ENOMEM;
 649         mm = dup_mm(tsk);
 650         if (!mm)
 651                 goto fail_nomem;
 652
 653 good_mm:
 654         /* Initializing for Swap token stuff */
 655         mm->token_priority = 0;
 656         mm->last_interval = 0;
 657
 658         tsk->mm = mm;
 659         tsk->active_mm = mm;
 660         return 0;
 661
 662 fail_nomem:
 663         return retval;
 664 }
 665
 666 static struct fs_struct *__copy_fs_struct(struct fs_struct *old)
 667 {
 668         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
 669         /* We don't need to lock fs - think why ;-) */
 670         if (fs) {
 671                 atomic_set(&fs->count, 1);
 672                 rwlock_init(&fs->lock);
 673                 fs->umask = old->umask;
 674                 read_lock(&old->lock);
 675                 fs->root = old->root;
 676                 path_get(&old->root);
 677                 fs->pwd = old->pwd;
 678                 path_get(&old->pwd);
 679                 read_unlock(&old->lock);
 680         }
 681         return fs;
 682 }
 683
 684 struct fs_struct *copy_fs_struct(struct fs_struct *old)
 685 {
 686         return __copy_fs_struct(old);
 687 }
 688
 689 EXPORT_SYMBOL_GPL(copy_fs_struct);
 690
 691 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 692 {
 693         if (clone_flags & CLONE_FS) {
 694                 atomic_inc(&current->fs->count);
 695                 return 0;
 696         }
 697         tsk->fs = __copy_fs_struct(current->fs);
 698         if (!tsk->fs)
 699                 return -ENOMEM;
 700         return 0;
 701 }
 702
 703 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 704 {
 705         struct files_struct *oldf, *newf;
 706         int error = 0;
 707
 708         /*
 709          * A background process may not have any files ...
 710          */
 711         oldf = current->files;
 712         if (!oldf)
 713                 goto out;
 714
 715         if (clone_flags & CLONE_FILES) {
 716                 atomic_inc(&oldf->count);
 717                 goto out;
 718         }
 719
 720         newf = dup_fd(oldf, &error);
 721         if (!newf)
 722                 goto out;
 723
 724         tsk->files = newf;
 725         error = 0;
 726 out:
 727         return error;
 728 }
 729
 730 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 731 {
 732 #ifdef CONFIG_BLOCK
 733         struct io_context *ioc = current->io_context;
 734
 735         if (!ioc)
 736                 return 0;
 737         /*
 738          * Share io context with parent, if CLONE_IO is set
 739          */
 740         if (clone_flags & CLONE_IO) {
 741                 tsk->io_context = ioc_task_link(ioc);
 742                 if (unlikely(!tsk->io_context))
 743                         return -ENOMEM;
 744         } else if (ioprio_valid(ioc->ioprio)) {
 745                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
 746                 if (unlikely(!tsk->io_context))
 747                         return -ENOMEM;
 748
 749                 tsk->io_context->ioprio = ioc->ioprio;
 750         }
 751 #endif
 752         return 0;
 753 }
 754
 755 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 756 {
 757         struct sighand_struct *sig;
 758
 759         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
 760                 atomic_inc(&current->sighand->count);
 761                 return 0;
 762         }
 763         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 764         rcu_assign_pointer(tsk->sighand, sig);
 765         if (!sig)
 766                 return -ENOMEM;
 767         atomic_set(&sig->count, 1);
 768         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 769         return 0;
 770 }
 771
 772 void __cleanup_sighand(struct sighand_struct *sighand)
 773 {
 774         if (atomic_dec_and_test(&sighand->count))
 775                 kmem_cache_free(sighand_cachep, sighand);
 776 }
 777
 778
 779 /*
 780  * Initialize POSIX timer handling for a thread group.
 781  */
 782 static void posix_cpu_timers_init_group(struct signal_struct *sig)
 783 {
 784         /* Thread group counters. */
 785         thread_group_cputime_init(sig);
 786
 787         /* Expiration times and increments. */
 788         sig->it_virt_expires = cputime_zero;
 789         sig->it_virt_incr = cputime_zero;
 790         sig->it_prof_expires = cputime_zero;
 791         sig->it_prof_incr = cputime_zero;
 792
 793         /* Cached expiration times. */
 794         sig->cputime_expires.prof_exp = cputime_zero;
 795         sig->cputime_expires.virt_exp = cputime_zero;
 796         sig->cputime_expires.sched_exp = 0;
 797
 798         /* The timer lists. */
 799         INIT_LIST_HEAD(&sig->cpu_timers[0]);
 800         INIT_LIST_HEAD(&sig->cpu_timers[1]);
 801         INIT_LIST_HEAD(&sig->cpu_timers[2]);
 802 }
 803
 804 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
 805 {
 806         struct signal_struct *sig;
 807         int ret;
 808
 809         if (clone_flags & CLONE_THREAD) {
 810                 ret = thread_group_cputime_clone_thread(current);
 811                 if (likely(!ret)) {
 812                         atomic_inc(&current->signal->count);
 813                         atomic_inc(&current->signal->live);
 814                 }
 815                 return ret;
 816         }
 817         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 818         tsk->signal = sig;
 819         if (!sig)
 820                 return -ENOMEM;
 821
 822         ret = copy_thread_group_keys(tsk);
 823         if (ret < 0) {
 824                 kmem_cache_free(signal_cachep, sig);
 825                 return ret;
 826         }
 827
 828         atomic_set(&sig->count, 1);
 829         atomic_set(&sig->live, 1);
 830         init_waitqueue_head(&sig->wait_chldexit);
 831         sig->flags = 0;
 832         sig->group_exit_code = 0;
 833         sig->group_exit_task = NULL;
 834         sig->group_stop_count = 0;
 835         sig->curr_target = tsk;
 836         init_sigpending(&sig->shared_pending);
 837         INIT_LIST_HEAD(&sig->posix_timers);
 838
 839         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 840         sig->it_real_incr.tv64 = 0;
 841         sig->real_timer.function = it_real_fn;
 842
 843         sig->leader = 0;        /* session leadership doesn't inherit */
 844         sig->tty_old_pgrp = NULL;
 845         sig->tty = NULL;
 846
 847         sig->cutime = sig->cstime = cputime_zero;
 848         sig->gtime = cputime_zero;
 849         sig->cgtime = cputime_zero;
 850         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
 851         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
 852         sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
 853         task_io_accounting_init(&sig->ioac);
 854         taskstats_tgid_init(sig);
 855
 856         task_lock(current->group_leader);
 857         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
 858         task_unlock(current->group_leader);
 859
 860         posix_cpu_timers_init_group(sig);
 861
 862         acct_init_pacct(&sig->pacct);
 863
 864         tty_audit_fork(sig);
 865
 866         return 0;
 867 }
 868
 869 void __cleanup_signal(struct signal_struct *sig)
 870 {
 871         thread_group_cputime_free(sig);
 872         exit_thread_group_keys(sig);
 873         tty_kref_put(sig->tty);
 874         kmem_cache_free(signal_cachep, sig);
 875 }
 876
 877 static void cleanup_signal(struct task_struct *tsk)
 878 {
 879         struct signal_struct *sig = tsk->signal;
 880
 881         atomic_dec(&sig->live);
 882
 883         if (atomic_dec_and_test(&sig->count))
 884                 __cleanup_signal(sig);
 885 }
 886
 887 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
 888 {
 889         unsigned long new_flags = p->flags;
 890
 891         new_flags &= ~PF_SUPERPRIV;
 892         new_flags |= PF_FORKNOEXEC;
 893         new_flags |= PF_STARTING;
 894         p->flags = new_flags;
 895         clear_freeze_flag(p);
 896 }
 897
 898 asmlinkage long sys_set_tid_address(int __user *tidptr)
 899 {
 900         current->clear_child_tid = tidptr;
 901
 902         return task_pid_vnr(current);
 903 }
 904
 905 static void rt_mutex_init_task(struct task_struct *p)
 906 {
 907         spin_lock_init(&p->pi_lock);
 908 #ifdef CONFIG_RT_MUTEXES
 909         plist_head_init(&p->pi_waiters, &p->pi_lock);
 910         p->pi_blocked_on = NULL;
 911 #endif
 912 }
 913
 914 #ifdef CONFIG_MM_OWNER
 915 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 916 {
 917         mm->owner = p;
 918 }
 919 #endif /* CONFIG_MM_OWNER */
 920
 921 /*
 922  * Initialize POSIX timer handling for a single task.
 923  */
 924 static void posix_cpu_timers_init(struct task_struct *tsk)
 925 {
 926         tsk->cputime_expires.prof_exp = cputime_zero;
 927         tsk->cputime_expires.virt_exp = cputime_zero;
 928         tsk->cputime_expires.sched_exp = 0;
 929         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
 930         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
 931         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
 932 }
 933
 934 /*
 935  * This creates a new process as a copy of the old one,
 936  * but does not actually start it yet.
 937  *
 938  * It copies the registers, and all the appropriate
 939  * parts of the process environment (as per the clone
 940  * flags). The actual kick-off is left to the caller.
 941  */
 942 static struct task_struct *copy_process(unsigned long clone_flags,
 943                                         unsigned long stack_start,
 944                                         struct pt_regs *regs,
 945                                         unsigned long stack_size,
 946                                         int __user *child_tidptr,
 947                                         struct pid *pid,
 948                                         int trace)
 949 {
 950         int retval;
 951         struct task_struct *p;
 952         int cgroup_callbacks_done = 0;
 953
 954         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 955                 return ERR_PTR(-EINVAL);
 956
 957         /*
 958          * Thread groups must share signals as well, and detached threads
 959          * can only be started up within the thread group.
 960          */
 961         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
 962                 return ERR_PTR(-EINVAL);
 963
 964         /*
 965          * Shared signal handlers imply shared VM. By way of the above,
 966          * thread groups also imply shared VM. Blocking this case allows
 967          * for various simplifications in other code.
 968          */
 969         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
 970                 return ERR_PTR(-EINVAL);
 971
 972         retval = security_task_create(clone_flags);
 973         if (retval)
 974                 goto fork_out;
 975
 976         retval = -ENOMEM;
 977         p = dup_task_struct(current);
 978         if (!p)
 979                 goto fork_out;
 980
 981         rt_mutex_init_task(p);
 982
 983 #ifdef CONFIG_PROVE_LOCKING
 984         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
 985         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
 986 #endif
 987         retval = -EAGAIN;
 988         if (atomic_read(&p->user->processes) >=
 989                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
 990                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
 991                     p->user != current->nsproxy->user_ns->root_user)
 992                         goto bad_fork_free;
 993         }
 994
 995         atomic_inc(&p->user->__count);
 996         atomic_inc(&p->user->processes);
 997         get_group_info(p->group_info);
 998
 999         /*
1000          * If multiple threads are within copy_process(), then this check
1001          * triggers too late. This doesn't hurt, the check is only there
1002          * to stop root fork bombs.
1003          */
1004         if (nr_threads >= max_threads)
1005                 goto bad_fork_cleanup_count;
1006
1007         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1008                 goto bad_fork_cleanup_count;
1009
1010         if (p->binfmt && !try_module_get(p->binfmt->module))
1011                 goto bad_fork_cleanup_put_domain;
1012
1013         p->did_exec = 0;
1014         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1015         copy_flags(clone_flags, p);
1016         INIT_LIST_HEAD(&p->children);
1017         INIT_LIST_HEAD(&p->sibling);
1018 #ifdef CONFIG_PREEMPT_RCU
1019         p->rcu_read_lock_nesting = 0;
1020         p->rcu_flipctr_idx = 0;
1021 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1022         p->vfork_done = NULL;
1023         spin_lock_init(&p->alloc_lock);
1024
1025         clear_tsk_thread_flag(p, TIF_SIGPENDING);
1026         init_sigpending(&p->pending);
1027
1028         p->utime = cputime_zero;
1029         p->stime = cputime_zero;
1030         p->gtime = cputime_zero;
1031         p->utimescaled = cputime_zero;
1032         p->stimescaled = cputime_zero;
1033         p->prev_utime = cputime_zero;
1034         p->prev_stime = cputime_zero;
1035
1036         p->default_timer_slack_ns = current->timer_slack_ns;
1037
1038 #ifdef CONFIG_DETECT_SOFTLOCKUP
1039         p->last_switch_count = 0;
1040         p->last_switch_timestamp = 0;
1041 #endif
1042
1043         task_io_accounting_init(&p->ioac);
1044         acct_clear_integrals(p);
1045
1046         posix_cpu_timers_init(p);
1047
1048         p->lock_depth = -1;             /* -1 = no lock */
1049         do_posix_clock_monotonic_gettime(&p->start_time);
1050         p->real_start_time = p->start_time;
1051         monotonic_to_bootbased(&p->real_start_time);
1052 #ifdef CONFIG_SECURITY
1053         p->security = NULL;
1054 #endif
1055         p->cap_bset = current->cap_bset;
1056         p->io_context = NULL;
1057         p->audit_context = NULL;
1058         cgroup_fork(p);
1059 #ifdef CONFIG_NUMA
1060         p->mempolicy = mpol_dup(p->mempolicy);
1061         if (IS_ERR(p->mempolicy)) {
1062                 retval = PTR_ERR(p->mempolicy);
1063                 p->mempolicy = NULL;
1064                 goto bad_fork_cleanup_cgroup;
1065         }
1066         mpol_fix_fork_child_flag(p);
1067 #endif
1068 #ifdef CONFIG_TRACE_IRQFLAGS
1069         p->irq_events = 0;
1070 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1071         p->hardirqs_enabled = 1;
1072 #else
1073         p->hardirqs_enabled = 0;
1074 #endif
1075         p->hardirq_enable_ip = 0;
1076         p->hardirq_enable_event = 0;
1077         p->hardirq_disable_ip = _THIS_IP_;
1078         p->hardirq_disable_event = 0;
1079         p->softirqs_enabled = 1;
1080         p->softirq_enable_ip = _THIS_IP_;
1081         p->softirq_enable_event = 0;
1082         p->softirq_disable_ip = 0;
1083         p->softirq_disable_event = 0;
1084         p->hardirq_context = 0;
1085         p->softirq_context = 0;
1086 #endif
1087 #ifdef CONFIG_LOCKDEP
1088         p->lockdep_depth = 0; /* no locks held yet */
1089         p->curr_chain_key = 0;
1090         p->lockdep_recursion = 0;
1091 #endif
1092
1093 #ifdef CONFIG_DEBUG_MUTEXES
1094         p->blocked_on = NULL; /* not blocked yet */
1095 #endif
1096
1097         /* Perform scheduler related setup. Assign this task to a CPU. */
1098         sched_fork(p, clone_flags);
1099
1100         if ((retval = security_task_alloc(p)))
1101                 goto bad_fork_cleanup_policy;
1102         if ((retval = audit_alloc(p)))
1103                 goto bad_fork_cleanup_security;
1104         /* copy all the process information */
1105         if ((retval = copy_semundo(clone_flags, p)))
1106                 goto bad_fork_cleanup_audit;
1107         if ((retval = copy_files(clone_flags, p)))
1108                 goto bad_fork_cleanup_semundo;
1109         if ((retval = copy_fs(clone_flags, p)))
1110                 goto bad_fork_cleanup_files;
1111         if ((retval = copy_sighand(clone_flags, p)))
1112                 goto bad_fork_cleanup_fs;
1113         if ((retval = copy_signal(clone_flags, p)))
1114                 goto bad_fork_cleanup_sighand;
1115         if ((retval = copy_mm(clone_flags, p)))
1116                 goto bad_fork_cleanup_signal;
1117         if ((retval = copy_keys(clone_flags, p)))
1118                 goto bad_fork_cleanup_mm;
1119         if ((retval = copy_namespaces(clone_flags, p)))
1120                 goto bad_fork_cleanup_keys;
1121         if ((retval = copy_io(clone_flags, p)))
1122                 goto bad_fork_cleanup_namespaces;
1123         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1124         if (retval)
1125                 goto bad_fork_cleanup_io;
1126
1127         if (pid != &init_struct_pid) {
1128                 retval = -ENOMEM;
1129                 pid = alloc_pid(task_active_pid_ns(p));
1130                 if (!pid)
1131                         goto bad_fork_cleanup_io;
1132
1133                 if (clone_flags & CLONE_NEWPID) {
1134                         retval = pid_ns_prepare_proc(task_active_pid_ns(p));
1135                         if (retval < 0)
1136                                 goto bad_fork_free_pid;
1137                 }
1138         }
1139
1140         p->pid = pid_nr(pid);
1141         p->tgid = p->pid;
1142         if (clone_flags & CLONE_THREAD)
1143                 p->tgid = current->tgid;
1144
1145         if (current->nsproxy != p->nsproxy) {
1146                 retval = ns_cgroup_clone(p, pid);
1147                 if (retval)
1148                         goto bad_fork_free_pid;
1149         }
1150
1151         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1152         /*
1153          * Clear TID on mm_release()?
1154          */
1155         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1156 #ifdef CONFIG_FUTEX
1157         p->robust_list = NULL;
1158 #ifdef CONFIG_COMPAT
1159         p->compat_robust_list = NULL;
1160 #endif
1161         INIT_LIST_HEAD(&p->pi_state_list);
1162         p->pi_state_cache = NULL;
1163 #endif
1164         /*
1165          * sigaltstack should be cleared when sharing the same VM
1166          */
1167         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1168                 p->sas_ss_sp = p->sas_ss_size = 0;
1169
1170         /*
1171          * Syscall tracing should be turned off in the child regardless
1172          * of CLONE_PTRACE.
1173          */
1174         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1175 #ifdef TIF_SYSCALL_EMU
1176         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1177 #endif
1178         clear_all_latency_tracing(p);
1179
1180         /* Our parent execution domain becomes current domain
1181            These must match for thread signalling to apply */
1182         p->parent_exec_id = p->self_exec_id;
1183
1184         /* ok, now we should be set up.. */
1185         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1186         p->pdeath_signal = 0;
1187         p->exit_state = 0;
1188
1189         /*
1190          * Ok, make it visible to the rest of the system.
1191          * We dont wake it up yet.
1192          */
1193         p->group_leader = p;
1194         INIT_LIST_HEAD(&p->thread_group);
1195
1196         /* Now that the task is set up, run cgroup callbacks if
1197          * necessary. We need to run them before the task is visible
1198          * on the tasklist. */
1199         cgroup_fork_callbacks(p);
1200         cgroup_callbacks_done = 1;
1201
1202         /* Need tasklist lock for parent etc handling! */
1203         write_lock_irq(&tasklist_lock);
1204
1205         /*
1206          * The task hasn't been attached yet, so its cpus_allowed mask will
1207          * not be changed, nor will its assigned CPU.
1208          *
1209          * The cpus_allowed mask of the parent may have changed after it was
1210          * copied first time - so re-copy it here, then check the child's CPU
1211          * to ensure it is on a valid CPU (and if not, just force it back to
1212          * parent's CPU). This avoids alot of nasty races.
1213          */
1214         p->cpus_allowed = current->cpus_allowed;
1215         p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1216         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1217                         !cpu_online(task_cpu(p))))
1218                 set_task_cpu(p, smp_processor_id());
1219
1220         /* CLONE_PARENT re-uses the old parent */
1221         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1222                 p->real_parent = current->real_parent;
1223         else
1224                 p->real_parent = current;
1225
1226         spin_lock(&current->sighand->siglock);
1227
1228         /*
1229          * Process group and session signals need to be delivered to just the
1230          * parent before the fork or both the parent and the child after the
1231          * fork. Restart if a signal comes in before we add the new process to
1232          * it's process group.
1233          * A fatal signal pending means that current will exit, so the new
1234          * thread can't slip out of an OOM kill (or normal SIGKILL).
1235          */
1236         recalc_sigpending();
1237         if (signal_pending(current)) {
1238                 spin_unlock(&current->sighand->siglock);
1239                 write_unlock_irq(&tasklist_lock);
1240                 retval = -ERESTARTNOINTR;
1241                 goto bad_fork_free_pid;
1242         }
1243
1244         if (clone_flags & CLONE_THREAD) {
1245                 p->group_leader = current->group_leader;
1246                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1247         }
1248
1249         if (likely(p->pid)) {
1250                 list_add_tail(&p->sibling, &p->real_parent->children);
1251                 tracehook_finish_clone(p, clone_flags, trace);
1252
1253                 if (thread_group_leader(p)) {
1254                         if (clone_flags & CLONE_NEWPID)
1255                                 p->nsproxy->pid_ns->child_reaper = p;
1256
1257                         p->signal->leader_pid = pid;
1258                         tty_kref_put(p->signal->tty);
1259                         p->signal->tty = tty_kref_get(current->signal->tty);
1260                         set_task_pgrp(p, task_pgrp_nr(current));
1261                         set_task_session(p, task_session_nr(current));
1262                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1263                         attach_pid(p, PIDTYPE_SID, task_session(current));
1264                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1265                         __get_cpu_var(process_counts)++;
1266                 }
1267                 attach_pid(p, PIDTYPE_PID, pid);
1268                 nr_threads++;
1269         }
1270
1271         total_forks++;
1272         spin_unlock(&current->sighand->siglock);
1273         write_unlock_irq(&tasklist_lock);
1274         ftrace_graph_init_task(p);
1275         proc_fork_connector(p);
1276         cgroup_post_fork(p);
1277         return p;
1278
1279 bad_fork_free_pid:
1280         if (pid != &init_struct_pid)
1281                 free_pid(pid);
1282 bad_fork_cleanup_io:
1283         put_io_context(p->io_context);
1284 bad_fork_cleanup_namespaces:
1285         exit_task_namespaces(p);
1286 bad_fork_cleanup_keys:
1287         exit_keys(p);
1288 bad_fork_cleanup_mm:
1289         if (p->mm)
1290                 mmput(p->mm);
1291 bad_fork_cleanup_signal:
1292         cleanup_signal(p);
1293 bad_fork_cleanup_sighand:
1294         __cleanup_sighand(p->sighand);
1295 bad_fork_cleanup_fs:
1296         exit_fs(p); /* blocking */
1297 bad_fork_cleanup_files:
1298         exit_files(p); /* blocking */
1299 bad_fork_cleanup_semundo:
1300         exit_sem(p);
1301 bad_fork_cleanup_audit:
1302         audit_free(p);
1303 bad_fork_cleanup_security:
1304         security_task_free(p);
1305 bad_fork_cleanup_policy:
1306 #ifdef CONFIG_NUMA
1307         mpol_put(p->mempolicy);
1308 bad_fork_cleanup_cgroup:
1309 #endif
1310         cgroup_exit(p, cgroup_callbacks_done);
1311         delayacct_tsk_free(p);
1312         if (p->binfmt)
1313                 module_put(p->binfmt->module);
1314 bad_fork_cleanup_put_domain:
1315         module_put(task_thread_info(p)->exec_domain->module);
1316 bad_fork_cleanup_count:
1317         put_group_info(p->group_info);
1318         atomic_dec(&p->user->processes);
1319         free_uid(p->user);
1320 bad_fork_free:
1321         free_task(p);
1322 fork_out:
1323         return ERR_PTR(retval);
1324 }
1325
1326 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1327 {
1328         memset(regs, 0, sizeof(struct pt_regs));
1329         return regs;
1330 }
1331
1332 struct task_struct * __cpuinit fork_idle(int cpu)
1333 {
1334         struct task_struct *task;
1335         struct pt_regs regs;
1336
1337         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1338                             &init_struct_pid, 0);
1339         if (!IS_ERR(task))
1340                 init_idle(task, cpu);
1341
1342         return task;
1343 }
1344
1345 /*
1346  *  Ok, this is the main fork-routine.
1347  *
1348  * It copies the process, and if successful kick-starts
1349  * it and waits for it to finish using the VM if required.
1350  */
1351 long do_fork(unsigned long clone_flags,
1352               unsigned long stack_start,
1353               struct pt_regs *regs,
1354               unsigned long stack_size,
1355               int __user *parent_tidptr,
1356               int __user *child_tidptr)
1357 {
1358         struct task_struct *p;
1359         int trace = 0;
1360         long nr;
1361
1362         /*
1363          * We hope to recycle these flags after 2.6.26
1364          */
1365         if (unlikely(clone_flags & CLONE_STOPPED)) {
1366                 static int __read_mostly count = 100;
1367
1368                 if (count > 0 && printk_ratelimit()) {
1369                         char comm[TASK_COMM_LEN];
1370
1371                         count--;
1372                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1373                                         "clone flags 0x%lx\n",
1374                                 get_task_comm(comm, current),
1375                                 clone_flags & CLONE_STOPPED);
1376                 }
1377         }
1378
1379         /*
1380          * When called from kernel_thread, don't do user tracing stuff.
1381          */
1382         if (likely(user_mode(regs)))
1383                 trace = tracehook_prepare_clone(clone_flags);
1384
1385         p = copy_process(clone_flags, stack_start, regs, stack_size,
1386                          child_tidptr, NULL, trace);
1387         /*
1388          * Do this prior waking up the new thread - the thread pointer
1389          * might get invalid after that point, if the thread exits quickly.
1390          */
1391         if (!IS_ERR(p)) {
1392                 struct completion vfork;
1393
1394                 trace_sched_process_fork(current, p);
1395
1396                 nr = task_pid_vnr(p);
1397
1398                 if (clone_flags & CLONE_PARENT_SETTID)
1399                         put_user(nr, parent_tidptr);
1400
1401                 if (clone_flags & CLONE_VFORK) {
1402                         p->vfork_done = &vfork;
1403                         init_completion(&vfork);
1404                 }
1405
1406                 tracehook_report_clone(trace, regs, clone_flags, nr, p);
1407
1408                 /*
1409                  * We set PF_STARTING at creation in case tracing wants to
1410                  * use this to distinguish a fully live task from one that
1411                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1412                  * clear it and set the child going.
1413                  */
1414                 p->flags &= ~PF_STARTING;
1415
1416                 if (unlikely(clone_flags & CLONE_STOPPED)) {
1417                         /*
1418                          * We'll start up with an immediate SIGSTOP.
1419                          */
1420                         sigaddset(&p->pending.signal, SIGSTOP);
1421                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1422                         __set_task_state(p, TASK_STOPPED);
1423                 } else {
1424                         wake_up_new_task(p, clone_flags);
1425                 }
1426
1427                 tracehook_report_clone_complete(trace, regs,
1428                                                 clone_flags, nr, p);
1429
1430                 if (clone_flags & CLONE_VFORK) {
1431                         freezer_do_not_count();
1432                         wait_for_completion(&vfork);
1433                         freezer_count();
1434                         tracehook_report_vfork_done(p, nr);
1435                 }
1436         } else {
1437                 nr = PTR_ERR(p);
1438         }
1439         return nr;
1440 }
1441
1442 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1443 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1444 #endif
1445
1446 static void sighand_ctor(void *data)
1447 {
1448         struct sighand_struct *sighand = data;
1449
1450         spin_lock_init(&sighand->siglock);
1451         init_waitqueue_head(&sighand->signalfd_wqh);
1452 }
1453
1454 void __init proc_caches_init(void)
1455 {
1456         sighand_cachep = kmem_cache_create("sighand_cache",
1457                         sizeof(struct sighand_struct), 0,
1458                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1459                         sighand_ctor);
1460         signal_cachep = kmem_cache_create("signal_cache",
1461                         sizeof(struct signal_struct), 0,
1462                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1463         files_cachep = kmem_cache_create("files_cache",
1464                         sizeof(struct files_struct), 0,
1465                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1466         fs_cachep = kmem_cache_create("fs_cache",
1467                         sizeof(struct fs_struct), 0,
1468                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1469         vm_area_cachep = kmem_cache_create("vm_area_struct",
1470                         sizeof(struct vm_area_struct), 0,
1471                         SLAB_PANIC, NULL);
1472         mm_cachep = kmem_cache_create("mm_struct",
1473                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1474                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1475 }
1476
1477 /*
1478  * Check constraints on flags passed to the unshare system call and
1479  * force unsharing of additional process context as appropriate.
1480  */
1481 static void check_unshare_flags(unsigned long *flags_ptr)
1482 {
1483         /*
1484          * If unsharing a thread from a thread group, must also
1485          * unshare vm.
1486          */
1487         if (*flags_ptr & CLONE_THREAD)
1488                 *flags_ptr |= CLONE_VM;
1489
1490         /*
1491          * If unsharing vm, must also unshare signal handlers.
1492          */
1493         if (*flags_ptr & CLONE_VM)
1494                 *flags_ptr |= CLONE_SIGHAND;
1495
1496         /*
1497          * If unsharing signal handlers and the task was created
1498          * using CLONE_THREAD, then must unshare the thread
1499          */
1500         if ((*flags_ptr & CLONE_SIGHAND) &&
1501             (atomic_read(&current->signal->count) > 1))
1502                 *flags_ptr |= CLONE_THREAD;
1503
1504         /*
1505          * If unsharing namespace, must also unshare filesystem information.
1506          */
1507         if (*flags_ptr & CLONE_NEWNS)
1508                 *flags_ptr |= CLONE_FS;
1509 }
1510
1511 /*
1512  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1513  */
1514 static int unshare_thread(unsigned long unshare_flags)
1515 {
1516         if (unshare_flags & CLONE_THREAD)
1517                 return -EINVAL;
1518
1519         return 0;
1520 }
1521
1522 /*
1523  * Unshare the filesystem structure if it is being shared
1524  */
1525 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1526 {
1527         struct fs_struct *fs = current->fs;
1528
1529         if ((unshare_flags & CLONE_FS) &&
1530             (fs && atomic_read(&fs->count) > 1)) {
1531                 *new_fsp = __copy_fs_struct(current->fs);
1532                 if (!*new_fsp)
1533                         return -ENOMEM;
1534         }
1535
1536         return 0;
1537 }
1538
1539 /*
1540  * Unsharing of sighand is not supported yet
1541  */
1542 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1543 {
1544         struct sighand_struct *sigh = current->sighand;
1545
1546         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1547                 return -EINVAL;
1548         else
1549                 return 0;
1550 }
1551
1552 /*
1553  * Unshare vm if it is being shared
1554  */
1555 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1556 {
1557         struct mm_struct *mm = current->mm;
1558
1559         if ((unshare_flags & CLONE_VM) &&
1560             (mm && atomic_read(&mm->mm_users) > 1)) {
1561                 return -EINVAL;
1562         }
1563
1564         return 0;
1565 }
1566
1567 /*
1568  * Unshare file descriptor table if it is being shared
1569  */
1570 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1571 {
1572         struct files_struct *fd = current->files;
1573         int error = 0;
1574
1575         if ((unshare_flags & CLONE_FILES) &&
1576             (fd && atomic_read(&fd->count) > 1)) {
1577                 *new_fdp = dup_fd(fd, &error);
1578                 if (!*new_fdp)
1579                         return error;
1580         }
1581
1582         return 0;
1583 }
1584
1585 /*
1586  * unshare allows a process to 'unshare' part of the process
1587  * context which was originally shared using clone.  copy_*
1588  * functions used by do_fork() cannot be used here directly
1589  * because they modify an inactive task_struct that is being
1590  * constructed. Here we are modifying the current, active,
1591  * task_struct.
1592  */
1593 asmlinkage long sys_unshare(unsigned long unshare_flags)
1594 {
1595         int err = 0;
1596         struct fs_struct *fs, *new_fs = NULL;
1597         struct sighand_struct *new_sigh = NULL;
1598         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1599         struct files_struct *fd, *new_fd = NULL;
1600         struct nsproxy *new_nsproxy = NULL;
1601         int do_sysvsem = 0;
1602
1603         check_unshare_flags(&unshare_flags);
1604
1605         /* Return -EINVAL for all unsupported flags */
1606         err = -EINVAL;
1607         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1608                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1609                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWUSER|
1610                                 CLONE_NEWNET))
1611                 goto bad_unshare_out;
1612
1613         /*
1614          * CLONE_NEWIPC must also detach from the undolist: after switching
1615          * to a new ipc namespace, the semaphore arrays from the old
1616          * namespace are unreachable.
1617          */
1618         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1619                 do_sysvsem = 1;
1620         if ((err = unshare_thread(unshare_flags)))
1621                 goto bad_unshare_out;
1622         if ((err = unshare_fs(unshare_flags, &new_fs)))
1623                 goto bad_unshare_cleanup_thread;
1624         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1625                 goto bad_unshare_cleanup_fs;
1626         if ((err = unshare_vm(unshare_flags, &new_mm)))
1627                 goto bad_unshare_cleanup_sigh;
1628         if ((err = unshare_fd(unshare_flags, &new_fd)))
1629                 goto bad_unshare_cleanup_vm;
1630         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1631                         new_fs)))
1632                 goto bad_unshare_cleanup_fd;
1633
1634         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1635                 if (do_sysvsem) {
1636                         /*
1637                          * CLONE_SYSVSEM is equivalent to sys_exit().
1638                          */
1639                         exit_sem(current);
1640                 }
1641
1642                 if (new_nsproxy) {
1643                         switch_task_namespaces(current, new_nsproxy);
1644                         new_nsproxy = NULL;
1645                 }
1646
1647                 task_lock(current);
1648
1649                 if (new_fs) {
1650                         fs = current->fs;
1651                         current->fs = new_fs;
1652                         new_fs = fs;
1653                 }
1654
1655                 if (new_mm) {
1656                         mm = current->mm;
1657                         active_mm = current->active_mm;
1658                         current->mm = new_mm;
1659                         current->active_mm = new_mm;
1660                         activate_mm(active_mm, new_mm);
1661                         new_mm = mm;
1662                 }
1663
1664                 if (new_fd) {
1665                         fd = current->files;
1666                         current->files = new_fd;
1667                         new_fd = fd;
1668                 }
1669
1670                 task_unlock(current);
1671         }
1672
1673         if (new_nsproxy)
1674                 put_nsproxy(new_nsproxy);
1675
1676 bad_unshare_cleanup_fd:
1677         if (new_fd)
1678                 put_files_struct(new_fd);
1679
1680 bad_unshare_cleanup_vm:
1681         if (new_mm)
1682                 mmput(new_mm);
1683
1684 bad_unshare_cleanup_sigh:
1685         if (new_sigh)
1686                 if (atomic_dec_and_test(&new_sigh->count))
1687                         kmem_cache_free(sighand_cachep, new_sigh);
1688
1689 bad_unshare_cleanup_fs:
1690         if (new_fs)
1691                 put_fs_struct(new_fs);
1692
1693 bad_unshare_cleanup_thread:
1694 bad_unshare_out:
1695         return err;
1696 }
1697
1698 /*
1699  *      Helper to unshare the files of the current task.
1700  *      We don't want to expose copy_files internals to
1701  *      the exec layer of the kernel.
1702  */
1703
1704 int unshare_files(struct files_struct **displaced)
1705 {
1706         struct task_struct *task = current;
1707         struct files_struct *copy = NULL;
1708         int error;
1709
1710         error = unshare_fd(CLONE_FILES, &copy);
1711         if (error || !copy) {
1712                 *displaced = NULL;
1713                 return error;
1714         }
1715         *displaced = task->files;
1716         task_lock(task);
1717         task->files = copy;
1718         task_unlock(task);
1719         return 0;
1720 }