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/ksm.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 <linux/fs_struct.h>
  64 #include <linux/magic.h>
  65 #include <linux/perf_event.h>
  66
  67 #include <asm/pgtable.h>
  68 #include <asm/pgalloc.h>
  69 #include <asm/uaccess.h>
  70 #include <asm/mmu_context.h>
  71 #include <asm/cacheflush.h>
  72 #include <asm/tlbflush.h>
  73
  74 #include <trace/events/sched.h>
  75
  76 /*
  77  * Protected counters by write_lock_irq(&tasklist_lock)
  78  */
  79 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  80 int nr_threads;                 /* The idle threads do not count.. */
  81
  82 int max_threads;                /* tunable limit on nr_threads */
  83
  84 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  85
  86 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  87
  88 int nr_processes(void)
  89 {
  90         int cpu;
  91         int total = 0;
  92
  93         for_each_online_cpu(cpu)
  94                 total += per_cpu(process_counts, cpu);
  95
  96         return total;
  97 }
  98
  99 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 100 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
 101 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
 102 static struct kmem_cache *task_struct_cachep;
 103 #endif
 104
 105 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
 106 static inline struct thread_info *alloc_thread_info(struct task_struct *tsk)
 107 {
 108 #ifdef CONFIG_DEBUG_STACK_USAGE
 109         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
 110 #else
 111         gfp_t mask = GFP_KERNEL;
 112 #endif
 113         return (struct thread_info *)__get_free_pages(mask, THREAD_SIZE_ORDER);
 114 }
 115
 116 static inline void free_thread_info(struct thread_info *ti)
 117 {
 118         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 119 }
 120 #endif
 121
 122 /* SLAB cache for signal_struct structures (tsk->signal) */
 123 static struct kmem_cache *signal_cachep;
 124
 125 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 126 struct kmem_cache *sighand_cachep;
 127
 128 /* SLAB cache for files_struct structures (tsk->files) */
 129 struct kmem_cache *files_cachep;
 130
 131 /* SLAB cache for fs_struct structures (tsk->fs) */
 132 struct kmem_cache *fs_cachep;
 133
 134 /* SLAB cache for vm_area_struct structures */
 135 struct kmem_cache *vm_area_cachep;
 136
 137 /* SLAB cache for mm_struct structures (tsk->mm) */
 138 static struct kmem_cache *mm_cachep;
 139
 140 static void account_kernel_stack(struct thread_info *ti, int account)
 141 {
 142         struct zone *zone = page_zone(virt_to_page(ti));
 143
 144         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
 145 }
 146
 147 void free_task(struct task_struct *tsk)
 148 {
 149         prop_local_destroy_single(&tsk->dirties);
 150         account_kernel_stack(tsk->stack, -1);
 151         free_thread_info(tsk->stack);
 152         rt_mutex_debug_task_free(tsk);
 153         ftrace_graph_exit_task(tsk);
 154         free_task_struct(tsk);
 155 }
 156 EXPORT_SYMBOL(free_task);
 157
 158 void __put_task_struct(struct task_struct *tsk)
 159 {
 160         WARN_ON(!tsk->exit_state);
 161         WARN_ON(atomic_read(&tsk->usage));
 162         WARN_ON(tsk == current);
 163
 164         exit_creds(tsk);
 165         delayacct_tsk_free(tsk);
 166
 167         if (!profile_handoff_task(tsk))
 168                 free_task(tsk);
 169 }
 170
 171 /*
 172  * macro override instead of weak attribute alias, to workaround
 173  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
 174  */
 175 #ifndef arch_task_cache_init
 176 #define arch_task_cache_init()
 177 #endif
 178
 179 void __init fork_init(unsigned long mempages)
 180 {
 181 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 182 #ifndef ARCH_MIN_TASKALIGN
 183 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 184 #endif
 185         /* create a slab on which task_structs can be allocated */
 186         task_struct_cachep =
 187                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 188                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
 189 #endif
 190
 191         /* do the arch specific task caches init */
 192         arch_task_cache_init();
 193
 194         /*
 195          * The default maximum number of threads is set to a safe
 196          * value: the thread structures can take up at most half
 197          * of memory.
 198          */
 199         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 200
 201         /*
 202          * we need to allow at least 20 threads to boot a system
 203          */
 204         if(max_threads < 20)
 205                 max_threads = 20;
 206
 207         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 208         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 209         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 210                 init_task.signal->rlim[RLIMIT_NPROC];
 211 }
 212
 213 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
 214                                                struct task_struct *src)
 215 {
 216         *dst = *src;
 217         return 0;
 218 }
 219
 220 static struct task_struct *dup_task_struct(struct task_struct *orig)
 221 {
 222         struct task_struct *tsk;
 223         struct thread_info *ti;
 224         unsigned long *stackend;
 225
 226         int err;
 227
 228         prepare_to_copy(orig);
 229
 230         tsk = alloc_task_struct();
 231         if (!tsk)
 232                 return NULL;
 233
 234         ti = alloc_thread_info(tsk);
 235         if (!ti) {
 236                 free_task_struct(tsk);
 237                 return NULL;
 238         }
 239
 240         err = arch_dup_task_struct(tsk, orig);
 241         if (err)
 242                 goto out;
 243
 244         tsk->stack = ti;
 245
 246         err = prop_local_init_single(&tsk->dirties);
 247         if (err)
 248                 goto out;
 249
 250         setup_thread_stack(tsk, orig);
 251         stackend = end_of_stack(tsk);
 252         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 253
 254 #ifdef CONFIG_CC_STACKPROTECTOR
 255         tsk->stack_canary = get_random_int();
 256 #endif
 257
 258         /* One for us, one for whoever does the "release_task()" (usually parent) */
 259         atomic_set(&tsk->usage,2);
 260         atomic_set(&tsk->fs_excl, 0);
 261 #ifdef CONFIG_BLK_DEV_IO_TRACE
 262         tsk->btrace_seq = 0;
 263 #endif
 264         tsk->splice_pipe = NULL;
 265
 266         account_kernel_stack(ti, 1);
 267
 268         return tsk;
 269
 270 out:
 271         free_thread_info(ti);
 272         free_task_struct(tsk);
 273         return NULL;
 274 }
 275
 276 #ifdef CONFIG_MMU
 277 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 278 {
 279         struct vm_area_struct *mpnt, *tmp, **pprev;
 280         struct rb_node **rb_link, *rb_parent;
 281         int retval;
 282         unsigned long charge;
 283         struct mempolicy *pol;
 284
 285         down_write(&oldmm->mmap_sem);
 286         flush_cache_dup_mm(oldmm);
 287         /*
 288          * Not linked in yet - no deadlock potential:
 289          */
 290         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 291
 292         mm->locked_vm = 0;
 293         mm->mmap = NULL;
 294         mm->mmap_cache = NULL;
 295         mm->free_area_cache = oldmm->mmap_base;
 296         mm->cached_hole_size = ~0UL;
 297         mm->map_count = 0;
 298         cpumask_clear(mm_cpumask(mm));
 299         mm->mm_rb = RB_ROOT;
 300         rb_link = &mm->mm_rb.rb_node;
 301         rb_parent = NULL;
 302         pprev = &mm->mmap;
 303         retval = ksm_fork(mm, oldmm);
 304         if (retval)
 305                 goto out;
 306
 307         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 308                 struct file *file;
 309
 310                 if (mpnt->vm_flags & VM_DONTCOPY) {
 311                         long pages = vma_pages(mpnt);
 312                         mm->total_vm -= pages;
 313                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 314                                                                 -pages);
 315                         continue;
 316                 }
 317                 charge = 0;
 318                 if (mpnt->vm_flags & VM_ACCOUNT) {
 319                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 320                         if (security_vm_enough_memory(len))
 321                                 goto fail_nomem;
 322                         charge = len;
 323                 }
 324                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 325                 if (!tmp)
 326                         goto fail_nomem;
 327                 *tmp = *mpnt;
 328                 pol = mpol_dup(vma_policy(mpnt));
 329                 retval = PTR_ERR(pol);
 330                 if (IS_ERR(pol))
 331                         goto fail_nomem_policy;
 332                 vma_set_policy(tmp, pol);
 333                 tmp->vm_flags &= ~VM_LOCKED;
 334                 tmp->vm_mm = mm;
 335                 tmp->vm_next = NULL;
 336                 anon_vma_link(tmp);
 337                 file = tmp->vm_file;
 338                 if (file) {
 339                         struct inode *inode = file->f_path.dentry->d_inode;
 340                         struct address_space *mapping = file->f_mapping;
 341
 342                         get_file(file);
 343                         if (tmp->vm_flags & VM_DENYWRITE)
 344                                 atomic_dec(&inode->i_writecount);
 345                         spin_lock(&mapping->i_mmap_lock);
 346                         if (tmp->vm_flags & VM_SHARED)
 347                                 mapping->i_mmap_writable++;
 348                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
 349                         flush_dcache_mmap_lock(mapping);
 350                         /* insert tmp into the share list, just after mpnt */
 351                         vma_prio_tree_add(tmp, mpnt);
 352                         flush_dcache_mmap_unlock(mapping);
 353                         spin_unlock(&mapping->i_mmap_lock);
 354                 }
 355
 356                 /*
 357                  * Clear hugetlb-related page reserves for children. This only
 358                  * affects MAP_PRIVATE mappings. Faults generated by the child
 359                  * are not guaranteed to succeed, even if read-only
 360                  */
 361                 if (is_vm_hugetlb_page(tmp))
 362                         reset_vma_resv_huge_pages(tmp);
 363
 364                 /*
 365                  * Link in the new vma and copy the page table entries.
 366                  */
 367                 *pprev = tmp;
 368                 pprev = &tmp->vm_next;
 369
 370                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 371                 rb_link = &tmp->vm_rb.rb_right;
 372                 rb_parent = &tmp->vm_rb;
 373
 374                 mm->map_count++;
 375                 retval = copy_page_range(mm, oldmm, mpnt);
 376
 377                 if (tmp->vm_ops && tmp->vm_ops->open)
 378                         tmp->vm_ops->open(tmp);
 379
 380                 if (retval)
 381                         goto out;
 382         }
 383         /* a new mm has just been created */
 384         arch_dup_mmap(oldmm, mm);
 385         retval = 0;
 386 out:
 387         up_write(&mm->mmap_sem);
 388         flush_tlb_mm(oldmm);
 389         up_write(&oldmm->mmap_sem);
 390         return retval;
 391 fail_nomem_policy:
 392         kmem_cache_free(vm_area_cachep, tmp);
 393 fail_nomem:
 394         retval = -ENOMEM;
 395         vm_unacct_memory(charge);
 396         goto out;
 397 }
 398
 399 static inline int mm_alloc_pgd(struct mm_struct * mm)
 400 {
 401         mm->pgd = pgd_alloc(mm);
 402         if (unlikely(!mm->pgd))
 403                 return -ENOMEM;
 404         return 0;
 405 }
 406
 407 static inline void mm_free_pgd(struct mm_struct * mm)
 408 {
 409         pgd_free(mm, mm->pgd);
 410 }
 411 #else
 412 #define dup_mmap(mm, oldmm)     (0)
 413 #define mm_alloc_pgd(mm)        (0)
 414 #define mm_free_pgd(mm)
 415 #endif /* CONFIG_MMU */
 416
 417 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 418
 419 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 420 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 421
 422 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 423
 424 static int __init coredump_filter_setup(char *s)
 425 {
 426         default_dump_filter =
 427                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 428                 MMF_DUMP_FILTER_MASK;
 429         return 1;
 430 }
 431
 432 __setup("coredump_filter=", coredump_filter_setup);
 433
 434 #include <linux/init_task.h>
 435
 436 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
 437 {
 438         atomic_set(&mm->mm_users, 1);
 439         atomic_set(&mm->mm_count, 1);
 440         init_rwsem(&mm->mmap_sem);
 441         INIT_LIST_HEAD(&mm->mmlist);
 442         mm->flags = (current->mm) ?
 443                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
 444         mm->core_state = NULL;
 445         mm->nr_ptes = 0;
 446         set_mm_counter(mm, file_rss, 0);
 447         set_mm_counter(mm, anon_rss, 0);
 448         spin_lock_init(&mm->page_table_lock);
 449         spin_lock_init(&mm->ioctx_lock);
 450         INIT_HLIST_HEAD(&mm->ioctx_list);
 451         mm->free_area_cache = TASK_UNMAPPED_BASE;
 452         mm->cached_hole_size = ~0UL;
 453         mm_init_owner(mm, p);
 454
 455         if (likely(!mm_alloc_pgd(mm))) {
 456                 mm->def_flags = 0;
 457                 mmu_notifier_mm_init(mm);
 458                 return mm;
 459         }
 460
 461         free_mm(mm);
 462         return NULL;
 463 }
 464
 465 /*
 466  * Allocate and initialize an mm_struct.
 467  */
 468 struct mm_struct * mm_alloc(void)
 469 {
 470         struct mm_struct * mm;
 471
 472         mm = allocate_mm();
 473         if (mm) {
 474                 memset(mm, 0, sizeof(*mm));
 475                 mm = mm_init(mm, current);
 476         }
 477         return mm;
 478 }
 479
 480 /*
 481  * Called when the last reference to the mm
 482  * is dropped: either by a lazy thread or by
 483  * mmput. Free the page directory and the mm.
 484  */
 485 void __mmdrop(struct mm_struct *mm)
 486 {
 487         BUG_ON(mm == &init_mm);
 488         mm_free_pgd(mm);
 489         destroy_context(mm);
 490         mmu_notifier_mm_destroy(mm);
 491         free_mm(mm);
 492 }
 493 EXPORT_SYMBOL_GPL(__mmdrop);
 494
 495 /*
 496  * Decrement the use count and release all resources for an mm.
 497  */
 498 void mmput(struct mm_struct *mm)
 499 {
 500         might_sleep();
 501
 502         if (atomic_dec_and_test(&mm->mm_users)) {
 503                 exit_aio(mm);
 504                 ksm_exit(mm);
 505                 exit_mmap(mm);
 506                 set_mm_exe_file(mm, NULL);
 507                 if (!list_empty(&mm->mmlist)) {
 508                         spin_lock(&mmlist_lock);
 509                         list_del(&mm->mmlist);
 510                         spin_unlock(&mmlist_lock);
 511                 }
 512                 put_swap_token(mm);
 513                 mmdrop(mm);
 514         }
 515 }
 516 EXPORT_SYMBOL_GPL(mmput);
 517
 518 /**
 519  * get_task_mm - acquire a reference to the task's mm
 520  *
 521  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 522  * this kernel workthread has transiently adopted a user mm with use_mm,
 523  * to do its AIO) is not set and if so returns a reference to it, after
 524  * bumping up the use count.  User must release the mm via mmput()
 525  * after use.  Typically used by /proc and ptrace.
 526  */
 527 struct mm_struct *get_task_mm(struct task_struct *task)
 528 {
 529         struct mm_struct *mm;
 530
 531         task_lock(task);
 532         mm = task->mm;
 533         if (mm) {
 534                 if (task->flags & PF_KTHREAD)
 535                         mm = NULL;
 536                 else
 537                         atomic_inc(&mm->mm_users);
 538         }
 539         task_unlock(task);
 540         return mm;
 541 }
 542 EXPORT_SYMBOL_GPL(get_task_mm);
 543
 544 /* Please note the differences between mmput and mm_release.
 545  * mmput is called whenever we stop holding onto a mm_struct,
 546  * error success whatever.
 547  *
 548  * mm_release is called after a mm_struct has been removed
 549  * from the current process.
 550  *
 551  * This difference is important for error handling, when we
 552  * only half set up a mm_struct for a new process and need to restore
 553  * the old one.  Because we mmput the new mm_struct before
 554  * restoring the old one. . .
 555  * Eric Biederman 10 January 1998
 556  */
 557 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 558 {
 559         struct completion *vfork_done = tsk->vfork_done;
 560
 561         /* Get rid of any futexes when releasing the mm */
 562 #ifdef CONFIG_FUTEX
 563         if (unlikely(tsk->robust_list))
 564                 exit_robust_list(tsk);
 565 #ifdef CONFIG_COMPAT
 566         if (unlikely(tsk->compat_robust_list))
 567                 compat_exit_robust_list(tsk);
 568 #endif
 569 #endif
 570
 571         /* Get rid of any cached register state */
 572         deactivate_mm(tsk, mm);
 573
 574         /* notify parent sleeping on vfork() */
 575         if (vfork_done) {
 576                 tsk->vfork_done = NULL;
 577                 complete(vfork_done);
 578         }
 579
 580         /*
 581          * If we're exiting normally, clear a user-space tid field if
 582          * requested.  We leave this alone when dying by signal, to leave
 583          * the value intact in a core dump, and to save the unnecessary
 584          * trouble otherwise.  Userland only wants this done for a sys_exit.
 585          */
 586         if (tsk->clear_child_tid) {
 587                 if (!(tsk->flags & PF_SIGNALED) &&
 588                     atomic_read(&mm->mm_users) > 1) {
 589                         /*
 590                          * We don't check the error code - if userspace has
 591                          * not set up a proper pointer then tough luck.
 592                          */
 593                         put_user(0, tsk->clear_child_tid);
 594                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 595                                         1, NULL, NULL, 0);
 596                 }
 597                 tsk->clear_child_tid = NULL;
 598         }
 599 }
 600
 601 /*
 602  * Allocate a new mm structure and copy contents from the
 603  * mm structure of the passed in task structure.
 604  */
 605 struct mm_struct *dup_mm(struct task_struct *tsk)
 606 {
 607         struct mm_struct *mm, *oldmm = current->mm;
 608         int err;
 609
 610         if (!oldmm)
 611                 return NULL;
 612
 613         mm = allocate_mm();
 614         if (!mm)
 615                 goto fail_nomem;
 616
 617         memcpy(mm, oldmm, sizeof(*mm));
 618
 619         /* Initializing for Swap token stuff */
 620         mm->token_priority = 0;
 621         mm->last_interval = 0;
 622
 623         if (!mm_init(mm, tsk))
 624                 goto fail_nomem;
 625
 626         if (init_new_context(tsk, mm))
 627                 goto fail_nocontext;
 628
 629         dup_mm_exe_file(oldmm, mm);
 630
 631         err = dup_mmap(mm, oldmm);
 632         if (err)
 633                 goto free_pt;
 634
 635         mm->hiwater_rss = get_mm_rss(mm);
 636         mm->hiwater_vm = mm->total_vm;
 637
 638         return mm;
 639
 640 free_pt:
 641         mmput(mm);
 642
 643 fail_nomem:
 644         return NULL;
 645
 646 fail_nocontext:
 647         /*
 648          * If init_new_context() failed, we cannot use mmput() to free the mm
 649          * because it calls destroy_context()
 650          */
 651         mm_free_pgd(mm);
 652         free_mm(mm);
 653         return NULL;
 654 }
 655
 656 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 657 {
 658         struct mm_struct * mm, *oldmm;
 659         int retval;
 660
 661         tsk->min_flt = tsk->maj_flt = 0;
 662         tsk->nvcsw = tsk->nivcsw = 0;
 663 #ifdef CONFIG_DETECT_HUNG_TASK
 664         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 665 #endif
 666
 667         tsk->mm = NULL;
 668         tsk->active_mm = NULL;
 669
 670         /*
 671          * Are we cloning a kernel thread?
 672          *
 673          * We need to steal a active VM for that..
 674          */
 675         oldmm = current->mm;
 676         if (!oldmm)
 677                 return 0;
 678
 679         if (clone_flags & CLONE_VM) {
 680                 atomic_inc(&oldmm->mm_users);
 681                 mm = oldmm;
 682                 goto good_mm;
 683         }
 684
 685         retval = -ENOMEM;
 686         mm = dup_mm(tsk);
 687         if (!mm)
 688                 goto fail_nomem;
 689
 690 good_mm:
 691         /* Initializing for Swap token stuff */
 692         mm->token_priority = 0;
 693         mm->last_interval = 0;
 694
 695         tsk->mm = mm;
 696         tsk->active_mm = mm;
 697         return 0;
 698
 699 fail_nomem:
 700         return retval;
 701 }
 702
 703 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 704 {
 705         struct fs_struct *fs = current->fs;
 706         if (clone_flags & CLONE_FS) {
 707                 /* tsk->fs is already what we want */
 708                 write_lock(&fs->lock);
 709                 if (fs->in_exec) {
 710                         write_unlock(&fs->lock);
 711                         return -EAGAIN;
 712                 }
 713                 fs->users++;
 714                 write_unlock(&fs->lock);
 715                 return 0;
 716         }
 717         tsk->fs = copy_fs_struct(fs);
 718         if (!tsk->fs)
 719                 return -ENOMEM;
 720         return 0;
 721 }
 722
 723 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 724 {
 725         struct files_struct *oldf, *newf;
 726         int error = 0;
 727
 728         /*
 729          * A background process may not have any files ...
 730          */
 731         oldf = current->files;
 732         if (!oldf)
 733                 goto out;
 734
 735         if (clone_flags & CLONE_FILES) {
 736                 atomic_inc(&oldf->count);
 737                 goto out;
 738         }
 739
 740         newf = dup_fd(oldf, &error);
 741         if (!newf)
 742                 goto out;
 743
 744         tsk->files = newf;
 745         error = 0;
 746 out:
 747         return error;
 748 }
 749
 750 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 751 {
 752 #ifdef CONFIG_BLOCK
 753         struct io_context *ioc = current->io_context;
 754
 755         if (!ioc)
 756                 return 0;
 757         /*
 758          * Share io context with parent, if CLONE_IO is set
 759          */
 760         if (clone_flags & CLONE_IO) {
 761                 tsk->io_context = ioc_task_link(ioc);
 762                 if (unlikely(!tsk->io_context))
 763                         return -ENOMEM;
 764         } else if (ioprio_valid(ioc->ioprio)) {
 765                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
 766                 if (unlikely(!tsk->io_context))
 767                         return -ENOMEM;
 768
 769                 tsk->io_context->ioprio = ioc->ioprio;
 770         }
 771 #endif
 772         return 0;
 773 }
 774
 775 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 776 {
 777         struct sighand_struct *sig;
 778
 779         if (clone_flags & CLONE_SIGHAND) {
 780                 atomic_inc(&current->sighand->count);
 781                 return 0;
 782         }
 783         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 784         rcu_assign_pointer(tsk->sighand, sig);
 785         if (!sig)
 786                 return -ENOMEM;
 787         atomic_set(&sig->count, 1);
 788         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 789         return 0;
 790 }
 791
 792 void __cleanup_sighand(struct sighand_struct *sighand)
 793 {
 794         if (atomic_dec_and_test(&sighand->count))
 795                 kmem_cache_free(sighand_cachep, sighand);
 796 }
 797
 798
 799 /*
 800  * Initialize POSIX timer handling for a thread group.
 801  */
 802 static void posix_cpu_timers_init_group(struct signal_struct *sig)
 803 {
 804         /* Thread group counters. */
 805         thread_group_cputime_init(sig);
 806
 807         /* Expiration times and increments. */
 808         sig->it_virt_expires = cputime_zero;
 809         sig->it_virt_incr = cputime_zero;
 810         sig->it_prof_expires = cputime_zero;
 811         sig->it_prof_incr = cputime_zero;
 812
 813         /* Cached expiration times. */
 814         sig->cputime_expires.prof_exp = cputime_zero;
 815         sig->cputime_expires.virt_exp = cputime_zero;
 816         sig->cputime_expires.sched_exp = 0;
 817
 818         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
 819                 sig->cputime_expires.prof_exp =
 820                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
 821                 sig->cputimer.running = 1;
 822         }
 823
 824         /* The timer lists. */
 825         INIT_LIST_HEAD(&sig->cpu_timers[0]);
 826         INIT_LIST_HEAD(&sig->cpu_timers[1]);
 827         INIT_LIST_HEAD(&sig->cpu_timers[2]);
 828 }
 829
 830 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
 831 {
 832         struct signal_struct *sig;
 833
 834         if (clone_flags & CLONE_THREAD)
 835                 return 0;
 836
 837         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 838         tsk->signal = sig;
 839         if (!sig)
 840                 return -ENOMEM;
 841
 842         atomic_set(&sig->count, 1);
 843         atomic_set(&sig->live, 1);
 844         init_waitqueue_head(&sig->wait_chldexit);
 845         sig->flags = 0;
 846         if (clone_flags & CLONE_NEWPID)
 847                 sig->flags |= SIGNAL_UNKILLABLE;
 848         sig->group_exit_code = 0;
 849         sig->group_exit_task = NULL;
 850         sig->group_stop_count = 0;
 851         sig->curr_target = tsk;
 852         init_sigpending(&sig->shared_pending);
 853         INIT_LIST_HEAD(&sig->posix_timers);
 854
 855         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 856         sig->it_real_incr.tv64 = 0;
 857         sig->real_timer.function = it_real_fn;
 858
 859         sig->leader = 0;        /* session leadership doesn't inherit */
 860         sig->tty_old_pgrp = NULL;
 861         sig->tty = NULL;
 862
 863         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
 864         sig->gtime = cputime_zero;
 865         sig->cgtime = cputime_zero;
 866         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
 867         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
 868         sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
 869         sig->maxrss = sig->cmaxrss = 0;
 870         task_io_accounting_init(&sig->ioac);
 871         sig->sum_sched_runtime = 0;
 872         taskstats_tgid_init(sig);
 873
 874         task_lock(current->group_leader);
 875         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
 876         task_unlock(current->group_leader);
 877
 878         posix_cpu_timers_init_group(sig);
 879
 880         acct_init_pacct(&sig->pacct);
 881
 882         tty_audit_fork(sig);
 883
 884         sig->oom_adj = current->signal->oom_adj;
 885
 886         return 0;
 887 }
 888
 889 void __cleanup_signal(struct signal_struct *sig)
 890 {
 891         thread_group_cputime_free(sig);
 892         tty_kref_put(sig->tty);
 893         kmem_cache_free(signal_cachep, sig);
 894 }
 895
 896 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
 897 {
 898         unsigned long new_flags = p->flags;
 899
 900         new_flags &= ~PF_SUPERPRIV;
 901         new_flags |= PF_FORKNOEXEC;
 902         new_flags |= PF_STARTING;
 903         p->flags = new_flags;
 904         clear_freeze_flag(p);
 905 }
 906
 907 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
 908 {
 909         current->clear_child_tid = tidptr;
 910
 911         return task_pid_vnr(current);
 912 }
 913
 914 static void rt_mutex_init_task(struct task_struct *p)
 915 {
 916         spin_lock_init(&p->pi_lock);
 917 #ifdef CONFIG_RT_MUTEXES
 918         plist_head_init(&p->pi_waiters, &p->pi_lock);
 919         p->pi_blocked_on = NULL;
 920 #endif
 921 }
 922
 923 #ifdef CONFIG_MM_OWNER
 924 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 925 {
 926         mm->owner = p;
 927 }
 928 #endif /* CONFIG_MM_OWNER */
 929
 930 /*
 931  * Initialize POSIX timer handling for a single task.
 932  */
 933 static void posix_cpu_timers_init(struct task_struct *tsk)
 934 {
 935         tsk->cputime_expires.prof_exp = cputime_zero;
 936         tsk->cputime_expires.virt_exp = cputime_zero;
 937         tsk->cputime_expires.sched_exp = 0;
 938         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
 939         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
 940         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
 941 }
 942
 943 /*
 944  * This creates a new process as a copy of the old one,
 945  * but does not actually start it yet.
 946  *
 947  * It copies the registers, and all the appropriate
 948  * parts of the process environment (as per the clone
 949  * flags). The actual kick-off is left to the caller.
 950  */
 951 static struct task_struct *copy_process(unsigned long clone_flags,
 952                                         unsigned long stack_start,
 953                                         struct pt_regs *regs,
 954                                         unsigned long stack_size,
 955                                         int __user *child_tidptr,
 956                                         struct pid *pid,
 957                                         int trace)
 958 {
 959         int retval;
 960         struct task_struct *p;
 961         int cgroup_callbacks_done = 0;
 962
 963         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 964                 return ERR_PTR(-EINVAL);
 965
 966         /*
 967          * Thread groups must share signals as well, and detached threads
 968          * can only be started up within the thread group.
 969          */
 970         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
 971                 return ERR_PTR(-EINVAL);
 972
 973         /*
 974          * Shared signal handlers imply shared VM. By way of the above,
 975          * thread groups also imply shared VM. Blocking this case allows
 976          * for various simplifications in other code.
 977          */
 978         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
 979                 return ERR_PTR(-EINVAL);
 980
 981         retval = security_task_create(clone_flags);
 982         if (retval)
 983                 goto fork_out;
 984
 985         retval = -ENOMEM;
 986         p = dup_task_struct(current);
 987         if (!p)
 988                 goto fork_out;
 989
 990         ftrace_graph_init_task(p);
 991
 992         rt_mutex_init_task(p);
 993
 994 #ifdef CONFIG_PROVE_LOCKING
 995         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
 996         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
 997 #endif
 998         retval = -EAGAIN;
 999         if (atomic_read(&p->real_cred->user->processes) >=
1000                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
1001                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1002                     p->real_cred->user != INIT_USER)
1003                         goto bad_fork_free;
1004         }
1005
1006         retval = copy_creds(p, clone_flags);
1007         if (retval < 0)
1008                 goto bad_fork_free;
1009
1010         /*
1011          * If multiple threads are within copy_process(), then this check
1012          * triggers too late. This doesn't hurt, the check is only there
1013          * to stop root fork bombs.
1014          */
1015         retval = -EAGAIN;
1016         if (nr_threads >= max_threads)
1017                 goto bad_fork_cleanup_count;
1018
1019         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1020                 goto bad_fork_cleanup_count;
1021
1022         if (p->binfmt && !try_module_get(p->binfmt->module))
1023                 goto bad_fork_cleanup_put_domain;
1024
1025         p->did_exec = 0;
1026         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1027         copy_flags(clone_flags, p);
1028         INIT_LIST_HEAD(&p->children);
1029         INIT_LIST_HEAD(&p->sibling);
1030         rcu_copy_process(p);
1031         p->vfork_done = NULL;
1032         spin_lock_init(&p->alloc_lock);
1033
1034         init_sigpending(&p->pending);
1035
1036         p->utime = cputime_zero;
1037         p->stime = cputime_zero;
1038         p->gtime = cputime_zero;
1039         p->utimescaled = cputime_zero;
1040         p->stimescaled = cputime_zero;
1041         p->prev_utime = cputime_zero;
1042         p->prev_stime = cputime_zero;
1043
1044         p->default_timer_slack_ns = current->timer_slack_ns;
1045
1046         task_io_accounting_init(&p->ioac);
1047         acct_clear_integrals(p);
1048
1049         posix_cpu_timers_init(p);
1050
1051         p->lock_depth = -1;             /* -1 = no lock */
1052         do_posix_clock_monotonic_gettime(&p->start_time);
1053         p->real_start_time = p->start_time;
1054         monotonic_to_bootbased(&p->real_start_time);
1055         p->io_context = NULL;
1056         p->audit_context = NULL;
1057         cgroup_fork(p);
1058 #ifdef CONFIG_NUMA
1059         p->mempolicy = mpol_dup(p->mempolicy);
1060         if (IS_ERR(p->mempolicy)) {
1061                 retval = PTR_ERR(p->mempolicy);
1062                 p->mempolicy = NULL;
1063                 goto bad_fork_cleanup_cgroup;
1064         }
1065         mpol_fix_fork_child_flag(p);
1066 #endif
1067 #ifdef CONFIG_TRACE_IRQFLAGS
1068         p->irq_events = 0;
1069 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1070         p->hardirqs_enabled = 1;
1071 #else
1072         p->hardirqs_enabled = 0;
1073 #endif
1074         p->hardirq_enable_ip = 0;
1075         p->hardirq_enable_event = 0;
1076         p->hardirq_disable_ip = _THIS_IP_;
1077         p->hardirq_disable_event = 0;
1078         p->softirqs_enabled = 1;
1079         p->softirq_enable_ip = _THIS_IP_;
1080         p->softirq_enable_event = 0;
1081         p->softirq_disable_ip = 0;
1082         p->softirq_disable_event = 0;
1083         p->hardirq_context = 0;
1084         p->softirq_context = 0;
1085 #endif
1086 #ifdef CONFIG_LOCKDEP
1087         p->lockdep_depth = 0; /* no locks held yet */
1088         p->curr_chain_key = 0;
1089         p->lockdep_recursion = 0;
1090 #endif
1091
1092 #ifdef CONFIG_DEBUG_MUTEXES
1093         p->blocked_on = NULL; /* not blocked yet */
1094 #endif
1095
1096         p->bts = NULL;
1097
1098         p->stack_start = stack_start;
1099
1100         /* Perform scheduler related setup. Assign this task to a CPU. */
1101         sched_fork(p, clone_flags);
1102
1103         retval = perf_event_init_task(p);
1104         if (retval)
1105                 goto bad_fork_cleanup_policy;
1106
1107         if ((retval = audit_alloc(p)))
1108                 goto bad_fork_cleanup_policy;
1109         /* copy all the process information */
1110         if ((retval = copy_semundo(clone_flags, p)))
1111                 goto bad_fork_cleanup_audit;
1112         if ((retval = copy_files(clone_flags, p)))
1113                 goto bad_fork_cleanup_semundo;
1114         if ((retval = copy_fs(clone_flags, p)))
1115                 goto bad_fork_cleanup_files;
1116         if ((retval = copy_sighand(clone_flags, p)))
1117                 goto bad_fork_cleanup_fs;
1118         if ((retval = copy_signal(clone_flags, p)))
1119                 goto bad_fork_cleanup_sighand;
1120         if ((retval = copy_mm(clone_flags, p)))
1121                 goto bad_fork_cleanup_signal;
1122         if ((retval = copy_namespaces(clone_flags, p)))
1123                 goto bad_fork_cleanup_mm;
1124         if ((retval = copy_io(clone_flags, p)))
1125                 goto bad_fork_cleanup_namespaces;
1126         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1127         if (retval)
1128                 goto bad_fork_cleanup_io;
1129
1130         if (pid != &init_struct_pid) {
1131                 retval = -ENOMEM;
1132                 pid = alloc_pid(p->nsproxy->pid_ns);
1133                 if (!pid)
1134                         goto bad_fork_cleanup_io;
1135
1136                 if (clone_flags & CLONE_NEWPID) {
1137                         retval = pid_ns_prepare_proc(p->nsproxy->pid_ns);
1138                         if (retval < 0)
1139                                 goto bad_fork_free_pid;
1140                 }
1141         }
1142
1143         p->pid = pid_nr(pid);
1144         p->tgid = p->pid;
1145         if (clone_flags & CLONE_THREAD)
1146                 p->tgid = current->tgid;
1147
1148         if (current->nsproxy != p->nsproxy) {
1149                 retval = ns_cgroup_clone(p, pid);
1150                 if (retval)
1151                         goto bad_fork_free_pid;
1152         }
1153
1154         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1155         /*
1156          * Clear TID on mm_release()?
1157          */
1158         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1159 #ifdef CONFIG_FUTEX
1160         p->robust_list = NULL;
1161 #ifdef CONFIG_COMPAT
1162         p->compat_robust_list = NULL;
1163 #endif
1164         INIT_LIST_HEAD(&p->pi_state_list);
1165         p->pi_state_cache = NULL;
1166 #endif
1167         /*
1168          * sigaltstack should be cleared when sharing the same VM
1169          */
1170         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1171                 p->sas_ss_sp = p->sas_ss_size = 0;
1172
1173         /*
1174          * Syscall tracing should be turned off in the child regardless
1175          * of CLONE_PTRACE.
1176          */
1177         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1178 #ifdef TIF_SYSCALL_EMU
1179         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1180 #endif
1181         clear_all_latency_tracing(p);
1182
1183         /* ok, now we should be set up.. */
1184         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1185         p->pdeath_signal = 0;
1186         p->exit_state = 0;
1187
1188         /*
1189          * Ok, make it visible to the rest of the system.
1190          * We dont wake it up yet.
1191          */
1192         p->group_leader = p;
1193         INIT_LIST_HEAD(&p->thread_group);
1194
1195         /* Now that the task is set up, run cgroup callbacks if
1196          * necessary. We need to run them before the task is visible
1197          * on the tasklist. */
1198         cgroup_fork_callbacks(p);
1199         cgroup_callbacks_done = 1;
1200
1201         /* Need tasklist lock for parent etc handling! */
1202         write_lock_irq(&tasklist_lock);
1203
1204         /*
1205          * The task hasn't been attached yet, so its cpus_allowed mask will
1206          * not be changed, nor will its assigned CPU.
1207          *
1208          * The cpus_allowed mask of the parent may have changed after it was
1209          * copied first time - so re-copy it here, then check the child's CPU
1210          * to ensure it is on a valid CPU (and if not, just force it back to
1211          * parent's CPU). This avoids alot of nasty races.
1212          */
1213         p->cpus_allowed = current->cpus_allowed;
1214         p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1215         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1216                         !cpu_online(task_cpu(p))))
1217                 set_task_cpu(p, smp_processor_id());
1218
1219         /* CLONE_PARENT re-uses the old parent */
1220         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1221                 p->real_parent = current->real_parent;
1222                 p->parent_exec_id = current->parent_exec_id;
1223         } else {
1224                 p->real_parent = current;
1225                 p->parent_exec_id = current->self_exec_id;
1226         }
1227
1228         spin_lock(&current->sighand->siglock);
1229
1230         /*
1231          * Process group and session signals need to be delivered to just the
1232          * parent before the fork or both the parent and the child after the
1233          * fork. Restart if a signal comes in before we add the new process to
1234          * it's process group.
1235          * A fatal signal pending means that current will exit, so the new
1236          * thread can't slip out of an OOM kill (or normal SIGKILL).
1237          */
1238         recalc_sigpending();
1239         if (signal_pending(current)) {
1240                 spin_unlock(&current->sighand->siglock);
1241                 write_unlock_irq(&tasklist_lock);
1242                 retval = -ERESTARTNOINTR;
1243                 goto bad_fork_free_pid;
1244         }
1245
1246         if (clone_flags & CLONE_THREAD) {
1247                 atomic_inc(&current->signal->count);
1248                 atomic_inc(&current->signal->live);
1249                 p->group_leader = current->group_leader;
1250                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1251         }
1252
1253         if (likely(p->pid)) {
1254                 list_add_tail(&p->sibling, &p->real_parent->children);
1255                 tracehook_finish_clone(p, clone_flags, trace);
1256
1257                 if (thread_group_leader(p)) {
1258                         if (clone_flags & CLONE_NEWPID)
1259                                 p->nsproxy->pid_ns->child_reaper = p;
1260
1261                         p->signal->leader_pid = pid;
1262                         tty_kref_put(p->signal->tty);
1263                         p->signal->tty = tty_kref_get(current->signal->tty);
1264                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1265                         attach_pid(p, PIDTYPE_SID, task_session(current));
1266                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1267                         __get_cpu_var(process_counts)++;
1268                 }
1269                 attach_pid(p, PIDTYPE_PID, pid);
1270                 nr_threads++;
1271         }
1272
1273         total_forks++;
1274         spin_unlock(&current->sighand->siglock);
1275         write_unlock_irq(&tasklist_lock);
1276         proc_fork_connector(p);
1277         cgroup_post_fork(p);
1278         perf_event_fork(p);
1279         return p;
1280
1281 bad_fork_free_pid:
1282         if (pid != &init_struct_pid)
1283                 free_pid(pid);
1284 bad_fork_cleanup_io:
1285         put_io_context(p->io_context);
1286 bad_fork_cleanup_namespaces:
1287         exit_task_namespaces(p);
1288 bad_fork_cleanup_mm:
1289         if (p->mm)
1290                 mmput(p->mm);
1291 bad_fork_cleanup_signal:
1292         if (!(clone_flags & CLONE_THREAD))
1293                 __cleanup_signal(p->signal);
1294 bad_fork_cleanup_sighand:
1295         __cleanup_sighand(p->sighand);
1296 bad_fork_cleanup_fs:
1297         exit_fs(p); /* blocking */
1298 bad_fork_cleanup_files:
1299         exit_files(p); /* blocking */
1300 bad_fork_cleanup_semundo:
1301         exit_sem(p);
1302 bad_fork_cleanup_audit:
1303         audit_free(p);
1304 bad_fork_cleanup_policy:
1305         perf_event_free_task(p);
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         atomic_dec(&p->cred->user->processes);
1318         exit_creds(p);
1319 bad_fork_free:
1320         free_task(p);
1321 fork_out:
1322         return ERR_PTR(retval);
1323 }
1324
1325 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1326 {
1327         memset(regs, 0, sizeof(struct pt_regs));
1328         return regs;
1329 }
1330
1331 struct task_struct * __cpuinit fork_idle(int cpu)
1332 {
1333         struct task_struct *task;
1334         struct pt_regs regs;
1335
1336         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1337                             &init_struct_pid, 0);
1338         if (!IS_ERR(task))
1339                 init_idle(task, cpu);
1340
1341         return task;
1342 }
1343
1344 /*
1345  *  Ok, this is the main fork-routine.
1346  *
1347  * It copies the process, and if successful kick-starts
1348  * it and waits for it to finish using the VM if required.
1349  */
1350 long do_fork(unsigned long clone_flags,
1351               unsigned long stack_start,
1352               struct pt_regs *regs,
1353               unsigned long stack_size,
1354               int __user *parent_tidptr,
1355               int __user *child_tidptr)
1356 {
1357         struct task_struct *p;
1358         int trace = 0;
1359         long nr;
1360
1361         /*
1362          * Do some preliminary argument and permissions checking before we
1363          * actually start allocating stuff
1364          */
1365         if (clone_flags & CLONE_NEWUSER) {
1366                 if (clone_flags & CLONE_THREAD)
1367                         return -EINVAL;
1368                 /* hopefully this check will go away when userns support is
1369                  * complete
1370                  */
1371                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1372                                 !capable(CAP_SETGID))
1373                         return -EPERM;
1374         }
1375
1376         /*
1377          * We hope to recycle these flags after 2.6.26
1378          */
1379         if (unlikely(clone_flags & CLONE_STOPPED)) {
1380                 static int __read_mostly count = 100;
1381
1382                 if (count > 0 && printk_ratelimit()) {
1383                         char comm[TASK_COMM_LEN];
1384
1385                         count--;
1386                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1387                                         "clone flags 0x%lx\n",
1388                                 get_task_comm(comm, current),
1389                                 clone_flags & CLONE_STOPPED);
1390                 }
1391         }
1392
1393         /*
1394          * When called from kernel_thread, don't do user tracing stuff.
1395          */
1396         if (likely(user_mode(regs)))
1397                 trace = tracehook_prepare_clone(clone_flags);
1398
1399         p = copy_process(clone_flags, stack_start, regs, stack_size,
1400                          child_tidptr, NULL, trace);
1401         /*
1402          * Do this prior waking up the new thread - the thread pointer
1403          * might get invalid after that point, if the thread exits quickly.
1404          */
1405         if (!IS_ERR(p)) {
1406                 struct completion vfork;
1407
1408                 trace_sched_process_fork(current, p);
1409
1410                 nr = task_pid_vnr(p);
1411
1412                 if (clone_flags & CLONE_PARENT_SETTID)
1413                         put_user(nr, parent_tidptr);
1414
1415                 if (clone_flags & CLONE_VFORK) {
1416                         p->vfork_done = &vfork;
1417                         init_completion(&vfork);
1418                 }
1419
1420                 audit_finish_fork(p);
1421                 tracehook_report_clone(regs, clone_flags, nr, p);
1422
1423                 /*
1424                  * We set PF_STARTING at creation in case tracing wants to
1425                  * use this to distinguish a fully live task from one that
1426                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1427                  * clear it and set the child going.
1428                  */
1429                 p->flags &= ~PF_STARTING;
1430
1431                 if (unlikely(clone_flags & CLONE_STOPPED)) {
1432                         /*
1433                          * We'll start up with an immediate SIGSTOP.
1434                          */
1435                         sigaddset(&p->pending.signal, SIGSTOP);
1436                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1437                         __set_task_state(p, TASK_STOPPED);
1438                 } else {
1439                         wake_up_new_task(p, clone_flags);
1440                 }
1441
1442                 tracehook_report_clone_complete(trace, regs,
1443                                                 clone_flags, nr, p);
1444
1445                 if (clone_flags & CLONE_VFORK) {
1446                         freezer_do_not_count();
1447                         wait_for_completion(&vfork);
1448                         freezer_count();
1449                         tracehook_report_vfork_done(p, nr);
1450                 }
1451         } else {
1452                 nr = PTR_ERR(p);
1453         }
1454         return nr;
1455 }
1456
1457 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1458 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1459 #endif
1460
1461 static void sighand_ctor(void *data)
1462 {
1463         struct sighand_struct *sighand = data;
1464
1465         spin_lock_init(&sighand->siglock);
1466         init_waitqueue_head(&sighand->signalfd_wqh);
1467 }
1468
1469 void __init proc_caches_init(void)
1470 {
1471         sighand_cachep = kmem_cache_create("sighand_cache",
1472                         sizeof(struct sighand_struct), 0,
1473                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1474                         SLAB_NOTRACK, sighand_ctor);
1475         signal_cachep = kmem_cache_create("signal_cache",
1476                         sizeof(struct signal_struct), 0,
1477                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1478         files_cachep = kmem_cache_create("files_cache",
1479                         sizeof(struct files_struct), 0,
1480                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1481         fs_cachep = kmem_cache_create("fs_cache",
1482                         sizeof(struct fs_struct), 0,
1483                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1484         mm_cachep = kmem_cache_create("mm_struct",
1485                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1486                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1487         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1488         mmap_init();
1489 }
1490
1491 /*
1492  * Check constraints on flags passed to the unshare system call and
1493  * force unsharing of additional process context as appropriate.
1494  */
1495 static void check_unshare_flags(unsigned long *flags_ptr)
1496 {
1497         /*
1498          * If unsharing a thread from a thread group, must also
1499          * unshare vm.
1500          */
1501         if (*flags_ptr & CLONE_THREAD)
1502                 *flags_ptr |= CLONE_VM;
1503
1504         /*
1505          * If unsharing vm, must also unshare signal handlers.
1506          */
1507         if (*flags_ptr & CLONE_VM)
1508                 *flags_ptr |= CLONE_SIGHAND;
1509
1510         /*
1511          * If unsharing signal handlers and the task was created
1512          * using CLONE_THREAD, then must unshare the thread
1513          */
1514         if ((*flags_ptr & CLONE_SIGHAND) &&
1515             (atomic_read(&current->signal->count) > 1))
1516                 *flags_ptr |= CLONE_THREAD;
1517
1518         /*
1519          * If unsharing namespace, must also unshare filesystem information.
1520          */
1521         if (*flags_ptr & CLONE_NEWNS)
1522                 *flags_ptr |= CLONE_FS;
1523 }
1524
1525 /*
1526  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1527  */
1528 static int unshare_thread(unsigned long unshare_flags)
1529 {
1530         if (unshare_flags & CLONE_THREAD)
1531                 return -EINVAL;
1532
1533         return 0;
1534 }
1535
1536 /*
1537  * Unshare the filesystem structure if it is being shared
1538  */
1539 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1540 {
1541         struct fs_struct *fs = current->fs;
1542
1543         if (!(unshare_flags & CLONE_FS) || !fs)
1544                 return 0;
1545
1546         /* don't need lock here; in the worst case we'll do useless copy */
1547         if (fs->users == 1)
1548                 return 0;
1549
1550         *new_fsp = copy_fs_struct(fs);
1551         if (!*new_fsp)
1552                 return -ENOMEM;
1553
1554         return 0;
1555 }
1556
1557 /*
1558  * Unsharing of sighand is not supported yet
1559  */
1560 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1561 {
1562         struct sighand_struct *sigh = current->sighand;
1563
1564         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1565                 return -EINVAL;
1566         else
1567                 return 0;
1568 }
1569
1570 /*
1571  * Unshare vm if it is being shared
1572  */
1573 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1574 {
1575         struct mm_struct *mm = current->mm;
1576
1577         if ((unshare_flags & CLONE_VM) &&
1578             (mm && atomic_read(&mm->mm_users) > 1)) {
1579                 return -EINVAL;
1580         }
1581
1582         return 0;
1583 }
1584
1585 /*
1586  * Unshare file descriptor table if it is being shared
1587  */
1588 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1589 {
1590         struct files_struct *fd = current->files;
1591         int error = 0;
1592
1593         if ((unshare_flags & CLONE_FILES) &&
1594             (fd && atomic_read(&fd->count) > 1)) {
1595                 *new_fdp = dup_fd(fd, &error);
1596                 if (!*new_fdp)
1597                         return error;
1598         }
1599
1600         return 0;
1601 }
1602
1603 /*
1604  * unshare allows a process to 'unshare' part of the process
1605  * context which was originally shared using clone.  copy_*
1606  * functions used by do_fork() cannot be used here directly
1607  * because they modify an inactive task_struct that is being
1608  * constructed. Here we are modifying the current, active,
1609  * task_struct.
1610  */
1611 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1612 {
1613         int err = 0;
1614         struct fs_struct *fs, *new_fs = NULL;
1615         struct sighand_struct *new_sigh = NULL;
1616         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1617         struct files_struct *fd, *new_fd = NULL;
1618         struct nsproxy *new_nsproxy = NULL;
1619         int do_sysvsem = 0;
1620
1621         check_unshare_flags(&unshare_flags);
1622
1623         /* Return -EINVAL for all unsupported flags */
1624         err = -EINVAL;
1625         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1626                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1627                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1628                 goto bad_unshare_out;
1629
1630         /*
1631          * CLONE_NEWIPC must also detach from the undolist: after switching
1632          * to a new ipc namespace, the semaphore arrays from the old
1633          * namespace are unreachable.
1634          */
1635         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1636                 do_sysvsem = 1;
1637         if ((err = unshare_thread(unshare_flags)))
1638                 goto bad_unshare_out;
1639         if ((err = unshare_fs(unshare_flags, &new_fs)))
1640                 goto bad_unshare_cleanup_thread;
1641         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1642                 goto bad_unshare_cleanup_fs;
1643         if ((err = unshare_vm(unshare_flags, &new_mm)))
1644                 goto bad_unshare_cleanup_sigh;
1645         if ((err = unshare_fd(unshare_flags, &new_fd)))
1646                 goto bad_unshare_cleanup_vm;
1647         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1648                         new_fs)))
1649                 goto bad_unshare_cleanup_fd;
1650
1651         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1652                 if (do_sysvsem) {
1653                         /*
1654                          * CLONE_SYSVSEM is equivalent to sys_exit().
1655                          */
1656                         exit_sem(current);
1657                 }
1658
1659                 if (new_nsproxy) {
1660                         switch_task_namespaces(current, new_nsproxy);
1661                         new_nsproxy = NULL;
1662                 }
1663
1664                 task_lock(current);
1665
1666                 if (new_fs) {
1667                         fs = current->fs;
1668                         write_lock(&fs->lock);
1669                         current->fs = new_fs;
1670                         if (--fs->users)
1671                                 new_fs = NULL;
1672                         else
1673                                 new_fs = fs;
1674                         write_unlock(&fs->lock);
1675                 }
1676
1677                 if (new_mm) {
1678                         mm = current->mm;
1679                         active_mm = current->active_mm;
1680                         current->mm = new_mm;
1681                         current->active_mm = new_mm;
1682                         activate_mm(active_mm, new_mm);
1683                         new_mm = mm;
1684                 }
1685
1686                 if (new_fd) {
1687                         fd = current->files;
1688                         current->files = new_fd;
1689                         new_fd = fd;
1690                 }
1691
1692                 task_unlock(current);
1693         }
1694
1695         if (new_nsproxy)
1696                 put_nsproxy(new_nsproxy);
1697
1698 bad_unshare_cleanup_fd:
1699         if (new_fd)
1700                 put_files_struct(new_fd);
1701
1702 bad_unshare_cleanup_vm:
1703         if (new_mm)
1704                 mmput(new_mm);
1705
1706 bad_unshare_cleanup_sigh:
1707         if (new_sigh)
1708                 if (atomic_dec_and_test(&new_sigh->count))
1709                         kmem_cache_free(sighand_cachep, new_sigh);
1710
1711 bad_unshare_cleanup_fs:
1712         if (new_fs)
1713                 free_fs_struct(new_fs);
1714
1715 bad_unshare_cleanup_thread:
1716 bad_unshare_out:
1717         return err;
1718 }
1719
1720 /*
1721  *      Helper to unshare the files of the current task.
1722  *      We don't want to expose copy_files internals to
1723  *      the exec layer of the kernel.
1724  */
1725
1726 int unshare_files(struct files_struct **displaced)
1727 {
1728         struct task_struct *task = current;
1729         struct files_struct *copy = NULL;
1730         int error;
1731
1732         error = unshare_fd(CLONE_FILES, &copy);
1733         if (error || !copy) {
1734                 *displaced = NULL;
1735                 return error;
1736         }
1737         *displaced = task->files;
1738         task_lock(task);
1739         task->files = copy;
1740         task_unlock(task);
1741         return 0;
1742 }