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/personality.h>
  21 #include <linux/mempolicy.h>
  22 #include <linux/sem.h>
  23 #include <linux/file.h>
  24 #include <linux/fdtable.h>
  25 #include <linux/iocontext.h>
  26 #include <linux/key.h>
  27 #include <linux/binfmts.h>
  28 #include <linux/mman.h>
  29 #include <linux/mmu_notifier.h>
  30 #include <linux/fs.h>
  31 #include <linux/nsproxy.h>
  32 #include <linux/capability.h>
  33 #include <linux/cpu.h>
  34 #include <linux/cgroup.h>
  35 #include <linux/security.h>
  36 #include <linux/hugetlb.h>
  37 #include <linux/swap.h>
  38 #include <linux/syscalls.h>
  39 #include <linux/jiffies.h>
  40 #include <linux/tracehook.h>
  41 #include <linux/futex.h>
  42 #include <linux/compat.h>
  43 #include <linux/task_io_accounting_ops.h>
  44 #include <linux/rcupdate.h>
  45 #include <linux/ptrace.h>
  46 #include <linux/mount.h>
  47 #include <linux/audit.h>
  48 #include <linux/memcontrol.h>
  49 #include <linux/ftrace.h>
  50 #include <linux/profile.h>
  51 #include <linux/rmap.h>
  52 #include <linux/acct.h>
  53 #include <linux/tsacct_kern.h>
  54 #include <linux/cn_proc.h>
  55 #include <linux/freezer.h>
  56 #include <linux/delayacct.h>
  57 #include <linux/taskstats_kern.h>
  58 #include <linux/random.h>
  59 #include <linux/tty.h>
  60 #include <linux/proc_fs.h>
  61 #include <linux/blkdev.h>
  62 #include <linux/fs_struct.h>
  63 #include <linux/magic.h>
  64 #include <linux/perf_counter.h>
  65
  66 #include <asm/pgtable.h>
  67 #include <asm/pgalloc.h>
  68 #include <asm/uaccess.h>
  69 #include <asm/mmu_context.h>
  70 #include <asm/cacheflush.h>
  71 #include <asm/tlbflush.h>
  72
  73 #include <trace/events/sched.h>
  74
  75 /*
  76  * Protected counters by write_lock_irq(&tasklist_lock)
  77  */
  78 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  79 int nr_threads;                 /* The idle threads do not count.. */
  80
  81 int max_threads;                /* tunable limit on nr_threads */
  82
  83 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  84
  85 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  86
  87 int nr_processes(void)
  88 {
  89         int cpu;
  90         int total = 0;
  91
  92         for_each_online_cpu(cpu)
  93                 total += per_cpu(process_counts, cpu);
  94
  95         return total;
  96 }
  97
  98 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
  99 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
 100 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
 101 static struct kmem_cache *task_struct_cachep;
 102 #endif
 103
 104 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
 105 static inline struct thread_info *alloc_thread_info(struct task_struct *tsk)
 106 {
 107 #ifdef CONFIG_DEBUG_STACK_USAGE
 108         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
 109 #else
 110         gfp_t mask = GFP_KERNEL;
 111 #endif
 112         return (struct thread_info *)__get_free_pages(mask, THREAD_SIZE_ORDER);
 113 }
 114
 115 static inline void free_thread_info(struct thread_info *ti)
 116 {
 117         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 118 }
 119 #endif
 120
 121 /* SLAB cache for signal_struct structures (tsk->signal) */
 122 static struct kmem_cache *signal_cachep;
 123
 124 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 125 struct kmem_cache *sighand_cachep;
 126
 127 /* SLAB cache for files_struct structures (tsk->files) */
 128 struct kmem_cache *files_cachep;
 129
 130 /* SLAB cache for fs_struct structures (tsk->fs) */
 131 struct kmem_cache *fs_cachep;
 132
 133 /* SLAB cache for vm_area_struct structures */
 134 struct kmem_cache *vm_area_cachep;
 135
 136 /* SLAB cache for mm_struct structures (tsk->mm) */
 137 static struct kmem_cache *mm_cachep;
 138
 139 void free_task(struct task_struct *tsk)
 140 {
 141         prop_local_destroy_single(&tsk->dirties);
 142         free_thread_info(tsk->stack);
 143         rt_mutex_debug_task_free(tsk);
 144         ftrace_graph_exit_task(tsk);
 145         free_task_struct(tsk);
 146 }
 147 EXPORT_SYMBOL(free_task);
 148
 149 void __put_task_struct(struct task_struct *tsk)
 150 {
 151         WARN_ON(!tsk->exit_state);
 152         WARN_ON(atomic_read(&tsk->usage));
 153         WARN_ON(tsk == current);
 154
 155         exit_creds(tsk);
 156         delayacct_tsk_free(tsk);
 157
 158         if (!profile_handoff_task(tsk))
 159                 free_task(tsk);
 160 }
 161
 162 /*
 163  * macro override instead of weak attribute alias, to workaround
 164  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
 165  */
 166 #ifndef arch_task_cache_init
 167 #define arch_task_cache_init()
 168 #endif
 169
 170 void __init fork_init(unsigned long mempages)
 171 {
 172 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 173 #ifndef ARCH_MIN_TASKALIGN
 174 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 175 #endif
 176         /* create a slab on which task_structs can be allocated */
 177         task_struct_cachep =
 178                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 179                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
 180 #endif
 181
 182         /* do the arch specific task caches init */
 183         arch_task_cache_init();
 184
 185         /*
 186          * The default maximum number of threads is set to a safe
 187          * value: the thread structures can take up at most half
 188          * of memory.
 189          */
 190         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 191
 192         /*
 193          * we need to allow at least 20 threads to boot a system
 194          */
 195         if(max_threads < 20)
 196                 max_threads = 20;
 197
 198         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 199         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 200         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 201                 init_task.signal->rlim[RLIMIT_NPROC];
 202 }
 203
 204 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
 205                                                struct task_struct *src)
 206 {
 207         *dst = *src;
 208         return 0;
 209 }
 210
 211 static struct task_struct *dup_task_struct(struct task_struct *orig)
 212 {
 213         struct task_struct *tsk;
 214         struct thread_info *ti;
 215         unsigned long *stackend;
 216
 217         int err;
 218
 219         prepare_to_copy(orig);
 220
 221         tsk = alloc_task_struct();
 222         if (!tsk)
 223                 return NULL;
 224
 225         ti = alloc_thread_info(tsk);
 226         if (!ti) {
 227                 free_task_struct(tsk);
 228                 return NULL;
 229         }
 230
 231         err = arch_dup_task_struct(tsk, orig);
 232         if (err)
 233                 goto out;
 234
 235         tsk->stack = ti;
 236
 237         err = prop_local_init_single(&tsk->dirties);
 238         if (err)
 239                 goto out;
 240
 241         setup_thread_stack(tsk, orig);
 242         stackend = end_of_stack(tsk);
 243         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 244
 245 #ifdef CONFIG_CC_STACKPROTECTOR
 246         tsk->stack_canary = get_random_int();
 247 #endif
 248
 249         /* One for us, one for whoever does the "release_task()" (usually parent) */
 250         atomic_set(&tsk->usage,2);
 251         atomic_set(&tsk->fs_excl, 0);
 252 #ifdef CONFIG_BLK_DEV_IO_TRACE
 253         tsk->btrace_seq = 0;
 254 #endif
 255         tsk->splice_pipe = NULL;
 256         return tsk;
 257
 258 out:
 259         free_thread_info(ti);
 260         free_task_struct(tsk);
 261         return NULL;
 262 }
 263
 264 #ifdef CONFIG_MMU
 265 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 266 {
 267         struct vm_area_struct *mpnt, *tmp, **pprev;
 268         struct rb_node **rb_link, *rb_parent;
 269         int retval;
 270         unsigned long charge;
 271         struct mempolicy *pol;
 272
 273         down_write(&oldmm->mmap_sem);
 274         flush_cache_dup_mm(oldmm);
 275         /*
 276          * Not linked in yet - no deadlock potential:
 277          */
 278         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 279
 280         mm->locked_vm = 0;
 281         mm->mmap = NULL;
 282         mm->mmap_cache = NULL;
 283         mm->free_area_cache = oldmm->mmap_base;
 284         mm->cached_hole_size = ~0UL;
 285         mm->map_count = 0;
 286         cpumask_clear(mm_cpumask(mm));
 287         mm->mm_rb = RB_ROOT;
 288         rb_link = &mm->mm_rb.rb_node;
 289         rb_parent = NULL;
 290         pprev = &mm->mmap;
 291
 292         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 293                 struct file *file;
 294
 295                 if (mpnt->vm_flags & VM_DONTCOPY) {
 296                         long pages = vma_pages(mpnt);
 297                         mm->total_vm -= pages;
 298                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 299                                                                 -pages);
 300                         continue;
 301                 }
 302                 charge = 0;
 303                 if (mpnt->vm_flags & VM_ACCOUNT) {
 304                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 305                         if (security_vm_enough_memory(len))
 306                                 goto fail_nomem;
 307                         charge = len;
 308                 }
 309                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 310                 if (!tmp)
 311                         goto fail_nomem;
 312                 *tmp = *mpnt;
 313                 pol = mpol_dup(vma_policy(mpnt));
 314                 retval = PTR_ERR(pol);
 315                 if (IS_ERR(pol))
 316                         goto fail_nomem_policy;
 317                 vma_set_policy(tmp, pol);
 318                 tmp->vm_flags &= ~VM_LOCKED;
 319                 tmp->vm_mm = mm;
 320                 tmp->vm_next = NULL;
 321                 anon_vma_link(tmp);
 322                 file = tmp->vm_file;
 323                 if (file) {
 324                         struct inode *inode = file->f_path.dentry->d_inode;
 325                         struct address_space *mapping = file->f_mapping;
 326
 327                         get_file(file);
 328                         if (tmp->vm_flags & VM_DENYWRITE)
 329                                 atomic_dec(&inode->i_writecount);
 330                         spin_lock(&mapping->i_mmap_lock);
 331                         if (tmp->vm_flags & VM_SHARED)
 332                                 mapping->i_mmap_writable++;
 333                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
 334                         flush_dcache_mmap_lock(mapping);
 335                         /* insert tmp into the share list, just after mpnt */
 336                         vma_prio_tree_add(tmp, mpnt);
 337                         flush_dcache_mmap_unlock(mapping);
 338                         spin_unlock(&mapping->i_mmap_lock);
 339                 }
 340
 341                 /*
 342                  * Clear hugetlb-related page reserves for children. This only
 343                  * affects MAP_PRIVATE mappings. Faults generated by the child
 344                  * are not guaranteed to succeed, even if read-only
 345                  */
 346                 if (is_vm_hugetlb_page(tmp))
 347                         reset_vma_resv_huge_pages(tmp);
 348
 349                 /*
 350                  * Link in the new vma and copy the page table entries.
 351                  */
 352                 *pprev = tmp;
 353                 pprev = &tmp->vm_next;
 354
 355                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 356                 rb_link = &tmp->vm_rb.rb_right;
 357                 rb_parent = &tmp->vm_rb;
 358
 359                 mm->map_count++;
 360                 retval = copy_page_range(mm, oldmm, mpnt);
 361
 362                 if (tmp->vm_ops && tmp->vm_ops->open)
 363                         tmp->vm_ops->open(tmp);
 364
 365                 if (retval)
 366                         goto out;
 367         }
 368         /* a new mm has just been created */
 369         arch_dup_mmap(oldmm, mm);
 370         retval = 0;
 371 out:
 372         up_write(&mm->mmap_sem);
 373         flush_tlb_mm(oldmm);
 374         up_write(&oldmm->mmap_sem);
 375         return retval;
 376 fail_nomem_policy:
 377         kmem_cache_free(vm_area_cachep, tmp);
 378 fail_nomem:
 379         retval = -ENOMEM;
 380         vm_unacct_memory(charge);
 381         goto out;
 382 }
 383
 384 static inline int mm_alloc_pgd(struct mm_struct * mm)
 385 {
 386         mm->pgd = pgd_alloc(mm);
 387         if (unlikely(!mm->pgd))
 388                 return -ENOMEM;
 389         return 0;
 390 }
 391
 392 static inline void mm_free_pgd(struct mm_struct * mm)
 393 {
 394         pgd_free(mm, mm->pgd);
 395 }
 396 #else
 397 #define dup_mmap(mm, oldmm)     (0)
 398 #define mm_alloc_pgd(mm)        (0)
 399 #define mm_free_pgd(mm)
 400 #endif /* CONFIG_MMU */
 401
 402 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 403
 404 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 405 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 406
 407 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 408
 409 static int __init coredump_filter_setup(char *s)
 410 {
 411         default_dump_filter =
 412                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 413                 MMF_DUMP_FILTER_MASK;
 414         return 1;
 415 }
 416
 417 __setup("coredump_filter=", coredump_filter_setup);
 418
 419 #include <linux/init_task.h>
 420
 421 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
 422 {
 423         atomic_set(&mm->mm_users, 1);
 424         atomic_set(&mm->mm_count, 1);
 425         init_rwsem(&mm->mmap_sem);
 426         INIT_LIST_HEAD(&mm->mmlist);
 427         mm->flags = (current->mm) ? current->mm->flags : default_dump_filter;
 428         mm->core_state = NULL;
 429         mm->nr_ptes = 0;
 430         set_mm_counter(mm, file_rss, 0);
 431         set_mm_counter(mm, anon_rss, 0);
 432         spin_lock_init(&mm->page_table_lock);
 433         spin_lock_init(&mm->ioctx_lock);
 434         INIT_HLIST_HEAD(&mm->ioctx_list);
 435         mm->free_area_cache = TASK_UNMAPPED_BASE;
 436         mm->cached_hole_size = ~0UL;
 437         mm_init_owner(mm, p);
 438
 439         if (likely(!mm_alloc_pgd(mm))) {
 440                 mm->def_flags = 0;
 441                 mmu_notifier_mm_init(mm);
 442                 return mm;
 443         }
 444
 445         free_mm(mm);
 446         return NULL;
 447 }
 448
 449 /*
 450  * Allocate and initialize an mm_struct.
 451  */
 452 struct mm_struct * mm_alloc(void)
 453 {
 454         struct mm_struct * mm;
 455
 456         mm = allocate_mm();
 457         if (mm) {
 458                 memset(mm, 0, sizeof(*mm));
 459                 mm = mm_init(mm, current);
 460         }
 461         return mm;
 462 }
 463
 464 /*
 465  * Called when the last reference to the mm
 466  * is dropped: either by a lazy thread or by
 467  * mmput. Free the page directory and the mm.
 468  */
 469 void __mmdrop(struct mm_struct *mm)
 470 {
 471         BUG_ON(mm == &init_mm);
 472         mm_free_pgd(mm);
 473         destroy_context(mm);
 474         mmu_notifier_mm_destroy(mm);
 475         free_mm(mm);
 476 }
 477 EXPORT_SYMBOL_GPL(__mmdrop);
 478
 479 /*
 480  * Decrement the use count and release all resources for an mm.
 481  */
 482 void mmput(struct mm_struct *mm)
 483 {
 484         might_sleep();
 485
 486         if (atomic_dec_and_test(&mm->mm_users)) {
 487                 exit_aio(mm);
 488                 exit_mmap(mm);
 489                 set_mm_exe_file(mm, NULL);
 490                 if (!list_empty(&mm->mmlist)) {
 491                         spin_lock(&mmlist_lock);
 492                         list_del(&mm->mmlist);
 493                         spin_unlock(&mmlist_lock);
 494                 }
 495                 put_swap_token(mm);
 496                 mmdrop(mm);
 497         }
 498 }
 499 EXPORT_SYMBOL_GPL(mmput);
 500
 501 /**
 502  * get_task_mm - acquire a reference to the task's mm
 503  *
 504  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 505  * this kernel workthread has transiently adopted a user mm with use_mm,
 506  * to do its AIO) is not set and if so returns a reference to it, after
 507  * bumping up the use count.  User must release the mm via mmput()
 508  * after use.  Typically used by /proc and ptrace.
 509  */
 510 struct mm_struct *get_task_mm(struct task_struct *task)
 511 {
 512         struct mm_struct *mm;
 513
 514         task_lock(task);
 515         mm = task->mm;
 516         if (mm) {
 517                 if (task->flags & PF_KTHREAD)
 518                         mm = NULL;
 519                 else
 520                         atomic_inc(&mm->mm_users);
 521         }
 522         task_unlock(task);
 523         return mm;
 524 }
 525 EXPORT_SYMBOL_GPL(get_task_mm);
 526
 527 /* Please note the differences between mmput and mm_release.
 528  * mmput is called whenever we stop holding onto a mm_struct,
 529  * error success whatever.
 530  *
 531  * mm_release is called after a mm_struct has been removed
 532  * from the current process.
 533  *
 534  * This difference is important for error handling, when we
 535  * only half set up a mm_struct for a new process and need to restore
 536  * the old one.  Because we mmput the new mm_struct before
 537  * restoring the old one. . .
 538  * Eric Biederman 10 January 1998
 539  */
 540 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 541 {
 542         struct completion *vfork_done = tsk->vfork_done;
 543
 544         /* Get rid of any futexes when releasing the mm */
 545 #ifdef CONFIG_FUTEX
 546         if (unlikely(tsk->robust_list))
 547                 exit_robust_list(tsk);
 548 #ifdef CONFIG_COMPAT
 549         if (unlikely(tsk->compat_robust_list))
 550                 compat_exit_robust_list(tsk);
 551 #endif
 552 #endif
 553
 554         /* Get rid of any cached register state */
 555         deactivate_mm(tsk, mm);
 556
 557         /* notify parent sleeping on vfork() */
 558         if (vfork_done) {
 559                 tsk->vfork_done = NULL;
 560                 complete(vfork_done);
 561         }
 562
 563         /*
 564          * If we're exiting normally, clear a user-space tid field if
 565          * requested.  We leave this alone when dying by signal, to leave
 566          * the value intact in a core dump, and to save the unnecessary
 567          * trouble otherwise.  Userland only wants this done for a sys_exit.
 568          */
 569         if (tsk->clear_child_tid) {
 570                 if (!(tsk->flags & PF_SIGNALED) &&
 571                     atomic_read(&mm->mm_users) > 1) {
 572                         /*
 573                          * We don't check the error code - if userspace has
 574                          * not set up a proper pointer then tough luck.
 575                          */
 576                         put_user(0, tsk->clear_child_tid);
 577                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 578                                         1, NULL, NULL, 0);
 579                 }
 580                 tsk->clear_child_tid = NULL;
 581         }
 582 }
 583
 584 /*
 585  * Allocate a new mm structure and copy contents from the
 586  * mm structure of the passed in task structure.
 587  */
 588 struct mm_struct *dup_mm(struct task_struct *tsk)
 589 {
 590         struct mm_struct *mm, *oldmm = current->mm;
 591         int err;
 592
 593         if (!oldmm)
 594                 return NULL;
 595
 596         mm = allocate_mm();
 597         if (!mm)
 598                 goto fail_nomem;
 599
 600         memcpy(mm, oldmm, sizeof(*mm));
 601
 602         /* Initializing for Swap token stuff */
 603         mm->token_priority = 0;
 604         mm->last_interval = 0;
 605
 606         if (!mm_init(mm, tsk))
 607                 goto fail_nomem;
 608
 609         if (init_new_context(tsk, mm))
 610                 goto fail_nocontext;
 611
 612         dup_mm_exe_file(oldmm, mm);
 613
 614         err = dup_mmap(mm, oldmm);
 615         if (err)
 616                 goto free_pt;
 617
 618         mm->hiwater_rss = get_mm_rss(mm);
 619         mm->hiwater_vm = mm->total_vm;
 620
 621         return mm;
 622
 623 free_pt:
 624         mmput(mm);
 625
 626 fail_nomem:
 627         return NULL;
 628
 629 fail_nocontext:
 630         /*
 631          * If init_new_context() failed, we cannot use mmput() to free the mm
 632          * because it calls destroy_context()
 633          */
 634         mm_free_pgd(mm);
 635         free_mm(mm);
 636         return NULL;
 637 }
 638
 639 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 640 {
 641         struct mm_struct * mm, *oldmm;
 642         int retval;
 643
 644         tsk->min_flt = tsk->maj_flt = 0;
 645         tsk->nvcsw = tsk->nivcsw = 0;
 646 #ifdef CONFIG_DETECT_HUNG_TASK
 647         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 648 #endif
 649
 650         tsk->mm = NULL;
 651         tsk->active_mm = NULL;
 652
 653         /*
 654          * Are we cloning a kernel thread?
 655          *
 656          * We need to steal a active VM for that..
 657          */
 658         oldmm = current->mm;
 659         if (!oldmm)
 660                 return 0;
 661
 662         if (clone_flags & CLONE_VM) {
 663                 atomic_inc(&oldmm->mm_users);
 664                 mm = oldmm;
 665                 goto good_mm;
 666         }
 667
 668         retval = -ENOMEM;
 669         mm = dup_mm(tsk);
 670         if (!mm)
 671                 goto fail_nomem;
 672
 673 good_mm:
 674         /* Initializing for Swap token stuff */
 675         mm->token_priority = 0;
 676         mm->last_interval = 0;
 677
 678         tsk->mm = mm;
 679         tsk->active_mm = mm;
 680         return 0;
 681
 682 fail_nomem:
 683         return retval;
 684 }
 685
 686 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 687 {
 688         struct fs_struct *fs = current->fs;
 689         if (clone_flags & CLONE_FS) {
 690                 /* tsk->fs is already what we want */
 691                 write_lock(&fs->lock);
 692                 if (fs->in_exec) {
 693                         write_unlock(&fs->lock);
 694                         return -EAGAIN;
 695                 }
 696                 fs->users++;
 697                 write_unlock(&fs->lock);
 698                 return 0;
 699         }
 700         tsk->fs = copy_fs_struct(fs);
 701         if (!tsk->fs)
 702                 return -ENOMEM;
 703         return 0;
 704 }
 705
 706 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 707 {
 708         struct files_struct *oldf, *newf;
 709         int error = 0;
 710
 711         /*
 712          * A background process may not have any files ...
 713          */
 714         oldf = current->files;
 715         if (!oldf)
 716                 goto out;
 717
 718         if (clone_flags & CLONE_FILES) {
 719                 atomic_inc(&oldf->count);
 720                 goto out;
 721         }
 722
 723         newf = dup_fd(oldf, &error);
 724         if (!newf)
 725                 goto out;
 726
 727         tsk->files = newf;
 728         error = 0;
 729 out:
 730         return error;
 731 }
 732
 733 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 734 {
 735 #ifdef CONFIG_BLOCK
 736         struct io_context *ioc = current->io_context;
 737
 738         if (!ioc)
 739                 return 0;
 740         /*
 741          * Share io context with parent, if CLONE_IO is set
 742          */
 743         if (clone_flags & CLONE_IO) {
 744                 tsk->io_context = ioc_task_link(ioc);
 745                 if (unlikely(!tsk->io_context))
 746                         return -ENOMEM;
 747         } else if (ioprio_valid(ioc->ioprio)) {
 748                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
 749                 if (unlikely(!tsk->io_context))
 750                         return -ENOMEM;
 751
 752                 tsk->io_context->ioprio = ioc->ioprio;
 753         }
 754 #endif
 755         return 0;
 756 }
 757
 758 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 759 {
 760         struct sighand_struct *sig;
 761
 762         if (clone_flags & CLONE_SIGHAND) {
 763                 atomic_inc(&current->sighand->count);
 764                 return 0;
 765         }
 766         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 767         rcu_assign_pointer(tsk->sighand, sig);
 768         if (!sig)
 769                 return -ENOMEM;
 770         atomic_set(&sig->count, 1);
 771         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 772         return 0;
 773 }
 774
 775 void __cleanup_sighand(struct sighand_struct *sighand)
 776 {
 777         if (atomic_dec_and_test(&sighand->count))
 778                 kmem_cache_free(sighand_cachep, sighand);
 779 }
 780
 781
 782 /*
 783  * Initialize POSIX timer handling for a thread group.
 784  */
 785 static void posix_cpu_timers_init_group(struct signal_struct *sig)
 786 {
 787         /* Thread group counters. */
 788         thread_group_cputime_init(sig);
 789
 790         /* Expiration times and increments. */
 791         sig->it_virt_expires = cputime_zero;
 792         sig->it_virt_incr = cputime_zero;
 793         sig->it_prof_expires = cputime_zero;
 794         sig->it_prof_incr = cputime_zero;
 795
 796         /* Cached expiration times. */
 797         sig->cputime_expires.prof_exp = cputime_zero;
 798         sig->cputime_expires.virt_exp = cputime_zero;
 799         sig->cputime_expires.sched_exp = 0;
 800
 801         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
 802                 sig->cputime_expires.prof_exp =
 803                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
 804                 sig->cputimer.running = 1;
 805         }
 806
 807         /* The timer lists. */
 808         INIT_LIST_HEAD(&sig->cpu_timers[0]);
 809         INIT_LIST_HEAD(&sig->cpu_timers[1]);
 810         INIT_LIST_HEAD(&sig->cpu_timers[2]);
 811 }
 812
 813 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
 814 {
 815         struct signal_struct *sig;
 816
 817         if (clone_flags & CLONE_THREAD)
 818                 return 0;
 819
 820         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 821         tsk->signal = sig;
 822         if (!sig)
 823                 return -ENOMEM;
 824
 825         atomic_set(&sig->count, 1);
 826         atomic_set(&sig->live, 1);
 827         init_waitqueue_head(&sig->wait_chldexit);
 828         sig->flags = 0;
 829         if (clone_flags & CLONE_NEWPID)
 830                 sig->flags |= SIGNAL_UNKILLABLE;
 831         sig->group_exit_code = 0;
 832         sig->group_exit_task = NULL;
 833         sig->group_stop_count = 0;
 834         sig->curr_target = tsk;
 835         init_sigpending(&sig->shared_pending);
 836         INIT_LIST_HEAD(&sig->posix_timers);
 837
 838         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 839         sig->it_real_incr.tv64 = 0;
 840         sig->real_timer.function = it_real_fn;
 841
 842         sig->leader = 0;        /* session leadership doesn't inherit */
 843         sig->tty_old_pgrp = NULL;
 844         sig->tty = NULL;
 845
 846         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
 847         sig->gtime = cputime_zero;
 848         sig->cgtime = cputime_zero;
 849         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
 850         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
 851         sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
 852         task_io_accounting_init(&sig->ioac);
 853         sig->sum_sched_runtime = 0;
 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         tty_kref_put(sig->tty);
 873         kmem_cache_free(signal_cachep, sig);
 874 }
 875
 876 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
 877 {
 878         unsigned long new_flags = p->flags;
 879
 880         new_flags &= ~PF_SUPERPRIV;
 881         new_flags |= PF_FORKNOEXEC;
 882         new_flags |= PF_STARTING;
 883         p->flags = new_flags;
 884         clear_freeze_flag(p);
 885 }
 886
 887 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
 888 {
 889         current->clear_child_tid = tidptr;
 890
 891         return task_pid_vnr(current);
 892 }
 893
 894 static void rt_mutex_init_task(struct task_struct *p)
 895 {
 896         spin_lock_init(&p->pi_lock);
 897 #ifdef CONFIG_RT_MUTEXES
 898         plist_head_init(&p->pi_waiters, &p->pi_lock);
 899         p->pi_blocked_on = NULL;
 900 #endif
 901 }
 902
 903 #ifdef CONFIG_MM_OWNER
 904 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 905 {
 906         mm->owner = p;
 907 }
 908 #endif /* CONFIG_MM_OWNER */
 909
 910 /*
 911  * Initialize POSIX timer handling for a single task.
 912  */
 913 static void posix_cpu_timers_init(struct task_struct *tsk)
 914 {
 915         tsk->cputime_expires.prof_exp = cputime_zero;
 916         tsk->cputime_expires.virt_exp = cputime_zero;
 917         tsk->cputime_expires.sched_exp = 0;
 918         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
 919         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
 920         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
 921 }
 922
 923 /*
 924  * This creates a new process as a copy of the old one,
 925  * but does not actually start it yet.
 926  *
 927  * It copies the registers, and all the appropriate
 928  * parts of the process environment (as per the clone
 929  * flags). The actual kick-off is left to the caller.
 930  */
 931 static struct task_struct *copy_process(unsigned long clone_flags,
 932                                         unsigned long stack_start,
 933                                         struct pt_regs *regs,
 934                                         unsigned long stack_size,
 935                                         int __user *child_tidptr,
 936                                         struct pid *pid,
 937                                         int trace)
 938 {
 939         int retval;
 940         struct task_struct *p;
 941         int cgroup_callbacks_done = 0;
 942
 943         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 944                 return ERR_PTR(-EINVAL);
 945
 946         /*
 947          * Thread groups must share signals as well, and detached threads
 948          * can only be started up within the thread group.
 949          */
 950         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
 951                 return ERR_PTR(-EINVAL);
 952
 953         /*
 954          * Shared signal handlers imply shared VM. By way of the above,
 955          * thread groups also imply shared VM. Blocking this case allows
 956          * for various simplifications in other code.
 957          */
 958         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
 959                 return ERR_PTR(-EINVAL);
 960
 961         retval = security_task_create(clone_flags);
 962         if (retval)
 963                 goto fork_out;
 964
 965         retval = -ENOMEM;
 966         p = dup_task_struct(current);
 967         if (!p)
 968                 goto fork_out;
 969
 970         ftrace_graph_init_task(p);
 971
 972         rt_mutex_init_task(p);
 973
 974 #ifdef CONFIG_PROVE_LOCKING
 975         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
 976         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
 977 #endif
 978         retval = -EAGAIN;
 979         if (atomic_read(&p->real_cred->user->processes) >=
 980                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
 981                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
 982                     p->real_cred->user != INIT_USER)
 983                         goto bad_fork_free;
 984         }
 985
 986         retval = copy_creds(p, clone_flags);
 987         if (retval < 0)
 988                 goto bad_fork_free;
 989
 990         /*
 991          * If multiple threads are within copy_process(), then this check
 992          * triggers too late. This doesn't hurt, the check is only there
 993          * to stop root fork bombs.
 994          */
 995         retval = -EAGAIN;
 996         if (nr_threads >= max_threads)
 997                 goto bad_fork_cleanup_count;
 998
 999         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1000                 goto bad_fork_cleanup_count;
1001
1002         if (p->binfmt && !try_module_get(p->binfmt->module))
1003                 goto bad_fork_cleanup_put_domain;
1004
1005         p->did_exec = 0;
1006         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1007         copy_flags(clone_flags, p);
1008         INIT_LIST_HEAD(&p->children);
1009         INIT_LIST_HEAD(&p->sibling);
1010 #ifdef CONFIG_PREEMPT_RCU
1011         p->rcu_read_lock_nesting = 0;
1012         p->rcu_flipctr_idx = 0;
1013 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1014         p->vfork_done = NULL;
1015         spin_lock_init(&p->alloc_lock);
1016
1017         init_sigpending(&p->pending);
1018
1019         p->utime = cputime_zero;
1020         p->stime = cputime_zero;
1021         p->gtime = cputime_zero;
1022         p->utimescaled = cputime_zero;
1023         p->stimescaled = cputime_zero;
1024         p->prev_utime = cputime_zero;
1025         p->prev_stime = cputime_zero;
1026
1027         p->default_timer_slack_ns = current->timer_slack_ns;
1028
1029         task_io_accounting_init(&p->ioac);
1030         acct_clear_integrals(p);
1031
1032         posix_cpu_timers_init(p);
1033
1034         p->lock_depth = -1;             /* -1 = no lock */
1035         do_posix_clock_monotonic_gettime(&p->start_time);
1036         p->real_start_time = p->start_time;
1037         monotonic_to_bootbased(&p->real_start_time);
1038         p->io_context = NULL;
1039         p->audit_context = NULL;
1040         cgroup_fork(p);
1041 #ifdef CONFIG_NUMA
1042         p->mempolicy = mpol_dup(p->mempolicy);
1043         if (IS_ERR(p->mempolicy)) {
1044                 retval = PTR_ERR(p->mempolicy);
1045                 p->mempolicy = NULL;
1046                 goto bad_fork_cleanup_cgroup;
1047         }
1048         mpol_fix_fork_child_flag(p);
1049 #endif
1050 #ifdef CONFIG_TRACE_IRQFLAGS
1051         p->irq_events = 0;
1052 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1053         p->hardirqs_enabled = 1;
1054 #else
1055         p->hardirqs_enabled = 0;
1056 #endif
1057         p->hardirq_enable_ip = 0;
1058         p->hardirq_enable_event = 0;
1059         p->hardirq_disable_ip = _THIS_IP_;
1060         p->hardirq_disable_event = 0;
1061         p->softirqs_enabled = 1;
1062         p->softirq_enable_ip = _THIS_IP_;
1063         p->softirq_enable_event = 0;
1064         p->softirq_disable_ip = 0;
1065         p->softirq_disable_event = 0;
1066         p->hardirq_context = 0;
1067         p->softirq_context = 0;
1068 #endif
1069 #ifdef CONFIG_LOCKDEP
1070         p->lockdep_depth = 0; /* no locks held yet */
1071         p->curr_chain_key = 0;
1072         p->lockdep_recursion = 0;
1073 #endif
1074
1075 #ifdef CONFIG_DEBUG_MUTEXES
1076         p->blocked_on = NULL; /* not blocked yet */
1077 #endif
1078
1079         p->bts = NULL;
1080
1081         /* Perform scheduler related setup. Assign this task to a CPU. */
1082         sched_fork(p, clone_flags);
1083
1084         retval = perf_counter_init_task(p);
1085         if (retval)
1086                 goto bad_fork_cleanup_policy;
1087
1088         if ((retval = audit_alloc(p)))
1089                 goto bad_fork_cleanup_policy;
1090         /* copy all the process information */
1091         if ((retval = copy_semundo(clone_flags, p)))
1092                 goto bad_fork_cleanup_audit;
1093         if ((retval = copy_files(clone_flags, p)))
1094                 goto bad_fork_cleanup_semundo;
1095         if ((retval = copy_fs(clone_flags, p)))
1096                 goto bad_fork_cleanup_files;
1097         if ((retval = copy_sighand(clone_flags, p)))
1098                 goto bad_fork_cleanup_fs;
1099         if ((retval = copy_signal(clone_flags, p)))
1100                 goto bad_fork_cleanup_sighand;
1101         if ((retval = copy_mm(clone_flags, p)))
1102                 goto bad_fork_cleanup_signal;
1103         if ((retval = copy_namespaces(clone_flags, p)))
1104                 goto bad_fork_cleanup_mm;
1105         if ((retval = copy_io(clone_flags, p)))
1106                 goto bad_fork_cleanup_namespaces;
1107         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1108         if (retval)
1109                 goto bad_fork_cleanup_io;
1110
1111         if (pid != &init_struct_pid) {
1112                 retval = -ENOMEM;
1113                 pid = alloc_pid(p->nsproxy->pid_ns);
1114                 if (!pid)
1115                         goto bad_fork_cleanup_io;
1116
1117                 if (clone_flags & CLONE_NEWPID) {
1118                         retval = pid_ns_prepare_proc(p->nsproxy->pid_ns);
1119                         if (retval < 0)
1120                                 goto bad_fork_free_pid;
1121                 }
1122         }
1123
1124         p->pid = pid_nr(pid);
1125         p->tgid = p->pid;
1126         if (clone_flags & CLONE_THREAD)
1127                 p->tgid = current->tgid;
1128
1129         if (current->nsproxy != p->nsproxy) {
1130                 retval = ns_cgroup_clone(p, pid);
1131                 if (retval)
1132                         goto bad_fork_free_pid;
1133         }
1134
1135         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1136         /*
1137          * Clear TID on mm_release()?
1138          */
1139         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1140 #ifdef CONFIG_FUTEX
1141         p->robust_list = NULL;
1142 #ifdef CONFIG_COMPAT
1143         p->compat_robust_list = NULL;
1144 #endif
1145         INIT_LIST_HEAD(&p->pi_state_list);
1146         p->pi_state_cache = NULL;
1147 #endif
1148         /*
1149          * sigaltstack should be cleared when sharing the same VM
1150          */
1151         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1152                 p->sas_ss_sp = p->sas_ss_size = 0;
1153
1154         /*
1155          * Syscall tracing should be turned off in the child regardless
1156          * of CLONE_PTRACE.
1157          */
1158         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1159 #ifdef TIF_SYSCALL_EMU
1160         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1161 #endif
1162         clear_all_latency_tracing(p);
1163
1164         /* ok, now we should be set up.. */
1165         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1166         p->pdeath_signal = 0;
1167         p->exit_state = 0;
1168
1169         /*
1170          * Ok, make it visible to the rest of the system.
1171          * We dont wake it up yet.
1172          */
1173         p->group_leader = p;
1174         INIT_LIST_HEAD(&p->thread_group);
1175
1176         /* Now that the task is set up, run cgroup callbacks if
1177          * necessary. We need to run them before the task is visible
1178          * on the tasklist. */
1179         cgroup_fork_callbacks(p);
1180         cgroup_callbacks_done = 1;
1181
1182         /* Need tasklist lock for parent etc handling! */
1183         write_lock_irq(&tasklist_lock);
1184
1185         /*
1186          * The task hasn't been attached yet, so its cpus_allowed mask will
1187          * not be changed, nor will its assigned CPU.
1188          *
1189          * The cpus_allowed mask of the parent may have changed after it was
1190          * copied first time - so re-copy it here, then check the child's CPU
1191          * to ensure it is on a valid CPU (and if not, just force it back to
1192          * parent's CPU). This avoids alot of nasty races.
1193          */
1194         p->cpus_allowed = current->cpus_allowed;
1195         p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1196         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1197                         !cpu_online(task_cpu(p))))
1198                 set_task_cpu(p, smp_processor_id());
1199
1200         /* CLONE_PARENT re-uses the old parent */
1201         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1202                 p->real_parent = current->real_parent;
1203                 p->parent_exec_id = current->parent_exec_id;
1204         } else {
1205                 p->real_parent = current;
1206                 p->parent_exec_id = current->self_exec_id;
1207         }
1208
1209         spin_lock(&current->sighand->siglock);
1210
1211         /*
1212          * Process group and session signals need to be delivered to just the
1213          * parent before the fork or both the parent and the child after the
1214          * fork. Restart if a signal comes in before we add the new process to
1215          * it's process group.
1216          * A fatal signal pending means that current will exit, so the new
1217          * thread can't slip out of an OOM kill (or normal SIGKILL).
1218          */
1219         recalc_sigpending();
1220         if (signal_pending(current)) {
1221                 spin_unlock(&current->sighand->siglock);
1222                 write_unlock_irq(&tasklist_lock);
1223                 retval = -ERESTARTNOINTR;
1224                 goto bad_fork_free_pid;
1225         }
1226
1227         if (clone_flags & CLONE_THREAD) {
1228                 atomic_inc(&current->signal->count);
1229                 atomic_inc(&current->signal->live);
1230                 p->group_leader = current->group_leader;
1231                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1232         }
1233
1234         if (likely(p->pid)) {
1235                 list_add_tail(&p->sibling, &p->real_parent->children);
1236                 tracehook_finish_clone(p, clone_flags, trace);
1237
1238                 if (thread_group_leader(p)) {
1239                         if (clone_flags & CLONE_NEWPID)
1240                                 p->nsproxy->pid_ns->child_reaper = p;
1241
1242                         p->signal->leader_pid = pid;
1243                         tty_kref_put(p->signal->tty);
1244                         p->signal->tty = tty_kref_get(current->signal->tty);
1245                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1246                         attach_pid(p, PIDTYPE_SID, task_session(current));
1247                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1248                         __get_cpu_var(process_counts)++;
1249                 }
1250                 attach_pid(p, PIDTYPE_PID, pid);
1251                 nr_threads++;
1252         }
1253
1254         total_forks++;
1255         spin_unlock(&current->sighand->siglock);
1256         write_unlock_irq(&tasklist_lock);
1257         proc_fork_connector(p);
1258         cgroup_post_fork(p);
1259         perf_counter_fork(p);
1260         return p;
1261
1262 bad_fork_free_pid:
1263         if (pid != &init_struct_pid)
1264                 free_pid(pid);
1265 bad_fork_cleanup_io:
1266         put_io_context(p->io_context);
1267 bad_fork_cleanup_namespaces:
1268         exit_task_namespaces(p);
1269 bad_fork_cleanup_mm:
1270         if (p->mm)
1271                 mmput(p->mm);
1272 bad_fork_cleanup_signal:
1273         if (!(clone_flags & CLONE_THREAD))
1274                 __cleanup_signal(p->signal);
1275 bad_fork_cleanup_sighand:
1276         __cleanup_sighand(p->sighand);
1277 bad_fork_cleanup_fs:
1278         exit_fs(p); /* blocking */
1279 bad_fork_cleanup_files:
1280         exit_files(p); /* blocking */
1281 bad_fork_cleanup_semundo:
1282         exit_sem(p);
1283 bad_fork_cleanup_audit:
1284         audit_free(p);
1285 bad_fork_cleanup_policy:
1286         perf_counter_free_task(p);
1287 #ifdef CONFIG_NUMA
1288         mpol_put(p->mempolicy);
1289 bad_fork_cleanup_cgroup:
1290 #endif
1291         cgroup_exit(p, cgroup_callbacks_done);
1292         delayacct_tsk_free(p);
1293         if (p->binfmt)
1294                 module_put(p->binfmt->module);
1295 bad_fork_cleanup_put_domain:
1296         module_put(task_thread_info(p)->exec_domain->module);
1297 bad_fork_cleanup_count:
1298         atomic_dec(&p->cred->user->processes);
1299         exit_creds(p);
1300 bad_fork_free:
1301         free_task(p);
1302 fork_out:
1303         return ERR_PTR(retval);
1304 }
1305
1306 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1307 {
1308         memset(regs, 0, sizeof(struct pt_regs));
1309         return regs;
1310 }
1311
1312 struct task_struct * __cpuinit fork_idle(int cpu)
1313 {
1314         struct task_struct *task;
1315         struct pt_regs regs;
1316
1317         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1318                             &init_struct_pid, 0);
1319         if (!IS_ERR(task))
1320                 init_idle(task, cpu);
1321
1322         return task;
1323 }
1324
1325 /*
1326  *  Ok, this is the main fork-routine.
1327  *
1328  * It copies the process, and if successful kick-starts
1329  * it and waits for it to finish using the VM if required.
1330  */
1331 long do_fork(unsigned long clone_flags,
1332               unsigned long stack_start,
1333               struct pt_regs *regs,
1334               unsigned long stack_size,
1335               int __user *parent_tidptr,
1336               int __user *child_tidptr)
1337 {
1338         struct task_struct *p;
1339         int trace = 0;
1340         long nr;
1341
1342         /*
1343          * Do some preliminary argument and permissions checking before we
1344          * actually start allocating stuff
1345          */
1346         if (clone_flags & CLONE_NEWUSER) {
1347                 if (clone_flags & CLONE_THREAD)
1348                         return -EINVAL;
1349                 /* hopefully this check will go away when userns support is
1350                  * complete
1351                  */
1352                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1353                                 !capable(CAP_SETGID))
1354                         return -EPERM;
1355         }
1356
1357         /*
1358          * We hope to recycle these flags after 2.6.26
1359          */
1360         if (unlikely(clone_flags & CLONE_STOPPED)) {
1361                 static int __read_mostly count = 100;
1362
1363                 if (count > 0 && printk_ratelimit()) {
1364                         char comm[TASK_COMM_LEN];
1365
1366                         count--;
1367                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1368                                         "clone flags 0x%lx\n",
1369                                 get_task_comm(comm, current),
1370                                 clone_flags & CLONE_STOPPED);
1371                 }
1372         }
1373
1374         /*
1375          * When called from kernel_thread, don't do user tracing stuff.
1376          */
1377         if (likely(user_mode(regs)))
1378                 trace = tracehook_prepare_clone(clone_flags);
1379
1380         p = copy_process(clone_flags, stack_start, regs, stack_size,
1381                          child_tidptr, NULL, trace);
1382         /*
1383          * Do this prior waking up the new thread - the thread pointer
1384          * might get invalid after that point, if the thread exits quickly.
1385          */
1386         if (!IS_ERR(p)) {
1387                 struct completion vfork;
1388
1389                 trace_sched_process_fork(current, p);
1390
1391                 nr = task_pid_vnr(p);
1392
1393                 if (clone_flags & CLONE_PARENT_SETTID)
1394                         put_user(nr, parent_tidptr);
1395
1396                 if (clone_flags & CLONE_VFORK) {
1397                         p->vfork_done = &vfork;
1398                         init_completion(&vfork);
1399                 }
1400
1401                 audit_finish_fork(p);
1402                 tracehook_report_clone(regs, clone_flags, nr, p);
1403
1404                 /*
1405                  * We set PF_STARTING at creation in case tracing wants to
1406                  * use this to distinguish a fully live task from one that
1407                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1408                  * clear it and set the child going.
1409                  */
1410                 p->flags &= ~PF_STARTING;
1411
1412                 if (unlikely(clone_flags & CLONE_STOPPED)) {
1413                         /*
1414                          * We'll start up with an immediate SIGSTOP.
1415                          */
1416                         sigaddset(&p->pending.signal, SIGSTOP);
1417                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1418                         __set_task_state(p, TASK_STOPPED);
1419                 } else {
1420                         wake_up_new_task(p, clone_flags);
1421                 }
1422
1423                 tracehook_report_clone_complete(trace, regs,
1424                                                 clone_flags, nr, p);
1425
1426                 if (clone_flags & CLONE_VFORK) {
1427                         freezer_do_not_count();
1428                         wait_for_completion(&vfork);
1429                         freezer_count();
1430                         tracehook_report_vfork_done(p, nr);
1431                 }
1432         } else {
1433                 nr = PTR_ERR(p);
1434         }
1435         return nr;
1436 }
1437
1438 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1439 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1440 #endif
1441
1442 static void sighand_ctor(void *data)
1443 {
1444         struct sighand_struct *sighand = data;
1445
1446         spin_lock_init(&sighand->siglock);
1447         init_waitqueue_head(&sighand->signalfd_wqh);
1448 }
1449
1450 void __init proc_caches_init(void)
1451 {
1452         sighand_cachep = kmem_cache_create("sighand_cache",
1453                         sizeof(struct sighand_struct), 0,
1454                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1455                         SLAB_NOTRACK, sighand_ctor);
1456         signal_cachep = kmem_cache_create("signal_cache",
1457                         sizeof(struct signal_struct), 0,
1458                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1459         files_cachep = kmem_cache_create("files_cache",
1460                         sizeof(struct files_struct), 0,
1461                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1462         fs_cachep = kmem_cache_create("fs_cache",
1463                         sizeof(struct fs_struct), 0,
1464                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1465         mm_cachep = kmem_cache_create("mm_struct",
1466                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1467                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1468         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1469         mmap_init();
1470 }
1471
1472 /*
1473  * Check constraints on flags passed to the unshare system call and
1474  * force unsharing of additional process context as appropriate.
1475  */
1476 static void check_unshare_flags(unsigned long *flags_ptr)
1477 {
1478         /*
1479          * If unsharing a thread from a thread group, must also
1480          * unshare vm.
1481          */
1482         if (*flags_ptr & CLONE_THREAD)
1483                 *flags_ptr |= CLONE_VM;
1484
1485         /*
1486          * If unsharing vm, must also unshare signal handlers.
1487          */
1488         if (*flags_ptr & CLONE_VM)
1489                 *flags_ptr |= CLONE_SIGHAND;
1490
1491         /*
1492          * If unsharing signal handlers and the task was created
1493          * using CLONE_THREAD, then must unshare the thread
1494          */
1495         if ((*flags_ptr & CLONE_SIGHAND) &&
1496             (atomic_read(&current->signal->count) > 1))
1497                 *flags_ptr |= CLONE_THREAD;
1498
1499         /*
1500          * If unsharing namespace, must also unshare filesystem information.
1501          */
1502         if (*flags_ptr & CLONE_NEWNS)
1503                 *flags_ptr |= CLONE_FS;
1504 }
1505
1506 /*
1507  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1508  */
1509 static int unshare_thread(unsigned long unshare_flags)
1510 {
1511         if (unshare_flags & CLONE_THREAD)
1512                 return -EINVAL;
1513
1514         return 0;
1515 }
1516
1517 /*
1518  * Unshare the filesystem structure if it is being shared
1519  */
1520 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1521 {
1522         struct fs_struct *fs = current->fs;
1523
1524         if (!(unshare_flags & CLONE_FS) || !fs)
1525                 return 0;
1526
1527         /* don't need lock here; in the worst case we'll do useless copy */
1528         if (fs->users == 1)
1529                 return 0;
1530
1531         *new_fsp = copy_fs_struct(fs);
1532         if (!*new_fsp)
1533                 return -ENOMEM;
1534
1535         return 0;
1536 }
1537
1538 /*
1539  * Unsharing of sighand is not supported yet
1540  */
1541 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1542 {
1543         struct sighand_struct *sigh = current->sighand;
1544
1545         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1546                 return -EINVAL;
1547         else
1548                 return 0;
1549 }
1550
1551 /*
1552  * Unshare vm if it is being shared
1553  */
1554 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1555 {
1556         struct mm_struct *mm = current->mm;
1557
1558         if ((unshare_flags & CLONE_VM) &&
1559             (mm && atomic_read(&mm->mm_users) > 1)) {
1560                 return -EINVAL;
1561         }
1562
1563         return 0;
1564 }
1565
1566 /*
1567  * Unshare file descriptor table if it is being shared
1568  */
1569 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1570 {
1571         struct files_struct *fd = current->files;
1572         int error = 0;
1573
1574         if ((unshare_flags & CLONE_FILES) &&
1575             (fd && atomic_read(&fd->count) > 1)) {
1576                 *new_fdp = dup_fd(fd, &error);
1577                 if (!*new_fdp)
1578                         return error;
1579         }
1580
1581         return 0;
1582 }
1583
1584 /*
1585  * unshare allows a process to 'unshare' part of the process
1586  * context which was originally shared using clone.  copy_*
1587  * functions used by do_fork() cannot be used here directly
1588  * because they modify an inactive task_struct that is being
1589  * constructed. Here we are modifying the current, active,
1590  * task_struct.
1591  */
1592 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1593 {
1594         int err = 0;
1595         struct fs_struct *fs, *new_fs = NULL;
1596         struct sighand_struct *new_sigh = NULL;
1597         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1598         struct files_struct *fd, *new_fd = NULL;
1599         struct nsproxy *new_nsproxy = NULL;
1600         int do_sysvsem = 0;
1601
1602         check_unshare_flags(&unshare_flags);
1603
1604         /* Return -EINVAL for all unsupported flags */
1605         err = -EINVAL;
1606         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1607                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1608                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1609                 goto bad_unshare_out;
1610
1611         /*
1612          * CLONE_NEWIPC must also detach from the undolist: after switching
1613          * to a new ipc namespace, the semaphore arrays from the old
1614          * namespace are unreachable.
1615          */
1616         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1617                 do_sysvsem = 1;
1618         if ((err = unshare_thread(unshare_flags)))
1619                 goto bad_unshare_out;
1620         if ((err = unshare_fs(unshare_flags, &new_fs)))
1621                 goto bad_unshare_cleanup_thread;
1622         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1623                 goto bad_unshare_cleanup_fs;
1624         if ((err = unshare_vm(unshare_flags, &new_mm)))
1625                 goto bad_unshare_cleanup_sigh;
1626         if ((err = unshare_fd(unshare_flags, &new_fd)))
1627                 goto bad_unshare_cleanup_vm;
1628         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1629                         new_fs)))
1630                 goto bad_unshare_cleanup_fd;
1631
1632         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1633                 if (do_sysvsem) {
1634                         /*
1635                          * CLONE_SYSVSEM is equivalent to sys_exit().
1636                          */
1637                         exit_sem(current);
1638                 }
1639
1640                 if (new_nsproxy) {
1641                         switch_task_namespaces(current, new_nsproxy);
1642                         new_nsproxy = NULL;
1643                 }
1644
1645                 task_lock(current);
1646
1647                 if (new_fs) {
1648                         fs = current->fs;
1649                         write_lock(&fs->lock);
1650                         current->fs = new_fs;
1651                         if (--fs->users)
1652                                 new_fs = NULL;
1653                         else
1654                                 new_fs = fs;
1655                         write_unlock(&fs->lock);
1656                 }
1657
1658                 if (new_mm) {
1659                         mm = current->mm;
1660                         active_mm = current->active_mm;
1661                         current->mm = new_mm;
1662                         current->active_mm = new_mm;
1663                         activate_mm(active_mm, new_mm);
1664                         new_mm = mm;
1665                 }
1666
1667                 if (new_fd) {
1668                         fd = current->files;
1669                         current->files = new_fd;
1670                         new_fd = fd;
1671                 }
1672
1673                 task_unlock(current);
1674         }
1675
1676         if (new_nsproxy)
1677                 put_nsproxy(new_nsproxy);
1678
1679 bad_unshare_cleanup_fd:
1680         if (new_fd)
1681                 put_files_struct(new_fd);
1682
1683 bad_unshare_cleanup_vm:
1684         if (new_mm)
1685                 mmput(new_mm);
1686
1687 bad_unshare_cleanup_sigh:
1688         if (new_sigh)
1689                 if (atomic_dec_and_test(&new_sigh->count))
1690                         kmem_cache_free(sighand_cachep, new_sigh);
1691
1692 bad_unshare_cleanup_fs:
1693         if (new_fs)
1694                 free_fs_struct(new_fs);
1695
1696 bad_unshare_cleanup_thread:
1697 bad_unshare_out:
1698         return err;
1699 }
1700
1701 /*
1702  *      Helper to unshare the files of the current task.
1703  *      We don't want to expose copy_files internals to
1704  *      the exec layer of the kernel.
1705  */
1706
1707 int unshare_files(struct files_struct **displaced)
1708 {
1709         struct task_struct *task = current;
1710         struct files_struct *copy = NULL;
1711         int error;
1712
1713         error = unshare_fd(CLONE_FILES, &copy);
1714         if (error || !copy) {
1715                 *displaced = NULL;
1716                 return error;
1717         }
1718         *displaced = task->files;
1719         task_lock(task);
1720         task->files = copy;
1721         task_unlock(task);
1722         return 0;
1723 }