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 #include <linux/posix-timers.h>
  67 #include <linux/user-return-notifier.h>
  68
  69 #include <asm/pgtable.h>
  70 #include <asm/pgalloc.h>
  71 #include <asm/uaccess.h>
  72 #include <asm/mmu_context.h>
  73 #include <asm/cacheflush.h>
  74 #include <asm/tlbflush.h>
  75
  76 #include <trace/events/sched.h>
  77
  78 /*
  79  * Protected counters by write_lock_irq(&tasklist_lock)
  80  */
  81 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  82 int nr_threads;                 /* The idle threads do not count.. */
  83
  84 int max_threads;                /* tunable limit on nr_threads */
  85
  86 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  87
  88 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  89
  90 #ifdef CONFIG_PROVE_RCU
  91 int lockdep_tasklist_lock_is_held(void)
  92 {
  93         return lockdep_is_held(&tasklist_lock);
  94 }
  95 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
  96 #endif /* #ifdef CONFIG_PROVE_RCU */
  97
  98 int nr_processes(void)
  99 {
 100         int cpu;
 101         int total = 0;
 102
 103         for_each_possible_cpu(cpu)
 104                 total += per_cpu(process_counts, cpu);
 105
 106         return total;
 107 }
 108
 109 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 110 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
 111 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
 112 static struct kmem_cache *task_struct_cachep;
 113 #endif
 114
 115 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
 116 static inline struct thread_info *alloc_thread_info(struct task_struct *tsk)
 117 {
 118 #ifdef CONFIG_DEBUG_STACK_USAGE
 119         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
 120 #else
 121         gfp_t mask = GFP_KERNEL;
 122 #endif
 123         return (struct thread_info *)__get_free_pages(mask, THREAD_SIZE_ORDER);
 124 }
 125
 126 static inline void free_thread_info(struct thread_info *ti)
 127 {
 128         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 129 }
 130 #endif
 131
 132 /* SLAB cache for signal_struct structures (tsk->signal) */
 133 static struct kmem_cache *signal_cachep;
 134
 135 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 136 struct kmem_cache *sighand_cachep;
 137
 138 /* SLAB cache for files_struct structures (tsk->files) */
 139 struct kmem_cache *files_cachep;
 140
 141 /* SLAB cache for fs_struct structures (tsk->fs) */
 142 struct kmem_cache *fs_cachep;
 143
 144 /* SLAB cache for vm_area_struct structures */
 145 struct kmem_cache *vm_area_cachep;
 146
 147 /* SLAB cache for mm_struct structures (tsk->mm) */
 148 static struct kmem_cache *mm_cachep;
 149
 150 /* Notifier list called when a task struct is freed */
 151 static ATOMIC_NOTIFIER_HEAD(task_free_notifier);
 152
 153 static void account_kernel_stack(struct thread_info *ti, int account)
 154 {
 155         struct zone *zone = page_zone(virt_to_page(ti));
 156
 157         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
 158 }
 159
 160 void free_task(struct task_struct *tsk)
 161 {
 162         prop_local_destroy_single(&tsk->dirties);
 163         account_kernel_stack(tsk->stack, -1);
 164         free_thread_info(tsk->stack);
 165         rt_mutex_debug_task_free(tsk);
 166         ftrace_graph_exit_task(tsk);
 167         free_task_struct(tsk);
 168 }
 169 EXPORT_SYMBOL(free_task);
 170
 171 static inline void free_signal_struct(struct signal_struct *sig)
 172 {
 173         taskstats_tgid_free(sig);
 174         kmem_cache_free(signal_cachep, sig);
 175 }
 176
 177 static inline void put_signal_struct(struct signal_struct *sig)
 178 {
 179         if (atomic_dec_and_test(&sig->sigcnt))
 180                 free_signal_struct(sig);
 181 }
 182
 183 int task_free_register(struct notifier_block *n)
 184 {
 185         return atomic_notifier_chain_register(&task_free_notifier, n);
 186 }
 187 EXPORT_SYMBOL(task_free_register);
 188
 189 int task_free_unregister(struct notifier_block *n)
 190 {
 191         return atomic_notifier_chain_unregister(&task_free_notifier, n);
 192 }
 193 EXPORT_SYMBOL(task_free_unregister);
 194
 195 void __put_task_struct(struct task_struct *tsk)
 196 {
 197         WARN_ON(!tsk->exit_state);
 198         WARN_ON(atomic_read(&tsk->usage));
 199         WARN_ON(tsk == current);
 200
 201         exit_creds(tsk);
 202         delayacct_tsk_free(tsk);
 203         put_signal_struct(tsk->signal);
 204
 205         atomic_notifier_call_chain(&task_free_notifier, 0, tsk);
 206         if (!profile_handoff_task(tsk))
 207                 free_task(tsk);
 208 }
 209
 210 /*
 211  * macro override instead of weak attribute alias, to workaround
 212  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
 213  */
 214 #ifndef arch_task_cache_init
 215 #define arch_task_cache_init()
 216 #endif
 217
 218 void __init fork_init(unsigned long mempages)
 219 {
 220 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 221 #ifndef ARCH_MIN_TASKALIGN
 222 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 223 #endif
 224         /* create a slab on which task_structs can be allocated */
 225         task_struct_cachep =
 226                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 227                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
 228 #endif
 229
 230         /* do the arch specific task caches init */
 231         arch_task_cache_init();
 232
 233         /*
 234          * The default maximum number of threads is set to a safe
 235          * value: the thread structures can take up at most half
 236          * of memory.
 237          */
 238         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 239
 240         /*
 241          * we need to allow at least 20 threads to boot a system
 242          */
 243         if(max_threads < 20)
 244                 max_threads = 20;
 245
 246         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 247         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 248         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 249                 init_task.signal->rlim[RLIMIT_NPROC];
 250 }
 251
 252 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
 253                                                struct task_struct *src)
 254 {
 255         *dst = *src;
 256         return 0;
 257 }
 258
 259 static struct task_struct *dup_task_struct(struct task_struct *orig)
 260 {
 261         struct task_struct *tsk;
 262         struct thread_info *ti;
 263         unsigned long *stackend;
 264
 265         int err;
 266
 267         prepare_to_copy(orig);
 268
 269         tsk = alloc_task_struct();
 270         if (!tsk)
 271                 return NULL;
 272
 273         ti = alloc_thread_info(tsk);
 274         if (!ti) {
 275                 free_task_struct(tsk);
 276                 return NULL;
 277         }
 278
 279         err = arch_dup_task_struct(tsk, orig);
 280         if (err)
 281                 goto out;
 282
 283         tsk->stack = ti;
 284
 285         err = prop_local_init_single(&tsk->dirties);
 286         if (err)
 287                 goto out;
 288
 289         setup_thread_stack(tsk, orig);
 290         clear_user_return_notifier(tsk);
 291         stackend = end_of_stack(tsk);
 292         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 293
 294 #ifdef CONFIG_CC_STACKPROTECTOR
 295         tsk->stack_canary = get_random_int();
 296 #endif
 297
 298         /* One for us, one for whoever does the "release_task()" (usually parent) */
 299         atomic_set(&tsk->usage,2);
 300         atomic_set(&tsk->fs_excl, 0);
 301 #ifdef CONFIG_BLK_DEV_IO_TRACE
 302         tsk->btrace_seq = 0;
 303 #endif
 304         tsk->splice_pipe = NULL;
 305
 306         account_kernel_stack(ti, 1);
 307
 308         return tsk;
 309
 310 out:
 311         free_thread_info(ti);
 312         free_task_struct(tsk);
 313         return NULL;
 314 }
 315
 316 #ifdef CONFIG_MMU
 317 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 318 {
 319         struct vm_area_struct *mpnt, *tmp, **pprev;
 320         struct rb_node **rb_link, *rb_parent;
 321         int retval;
 322         unsigned long charge;
 323         struct mempolicy *pol;
 324
 325         down_write(&oldmm->mmap_sem);
 326         flush_cache_dup_mm(oldmm);
 327         /*
 328          * Not linked in yet - no deadlock potential:
 329          */
 330         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 331
 332         mm->locked_vm = 0;
 333         mm->mmap = NULL;
 334         mm->mmap_cache = NULL;
 335         mm->free_area_cache = oldmm->mmap_base;
 336         mm->cached_hole_size = ~0UL;
 337         mm->map_count = 0;
 338         cpumask_clear(mm_cpumask(mm));
 339         mm->mm_rb = RB_ROOT;
 340         rb_link = &mm->mm_rb.rb_node;
 341         rb_parent = NULL;
 342         pprev = &mm->mmap;
 343         retval = ksm_fork(mm, oldmm);
 344         if (retval)
 345                 goto out;
 346
 347         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 348                 struct file *file;
 349
 350                 if (mpnt->vm_flags & VM_DONTCOPY) {
 351                         long pages = vma_pages(mpnt);
 352                         mm->total_vm -= pages;
 353                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 354                                                                 -pages);
 355                         continue;
 356                 }
 357                 charge = 0;
 358                 if (mpnt->vm_flags & VM_ACCOUNT) {
 359                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 360                         if (security_vm_enough_memory(len))
 361                                 goto fail_nomem;
 362                         charge = len;
 363                 }
 364                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 365                 if (!tmp)
 366                         goto fail_nomem;
 367                 *tmp = *mpnt;
 368                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 369                 pol = mpol_dup(vma_policy(mpnt));
 370                 retval = PTR_ERR(pol);
 371                 if (IS_ERR(pol))
 372                         goto fail_nomem_policy;
 373                 vma_set_policy(tmp, pol);
 374                 if (anon_vma_fork(tmp, mpnt))
 375                         goto fail_nomem_anon_vma_fork;
 376                 tmp->vm_flags &= ~VM_LOCKED;
 377                 tmp->vm_mm = mm;
 378                 tmp->vm_next = NULL;
 379                 file = tmp->vm_file;
 380                 if (file) {
 381                         struct inode *inode = file->f_path.dentry->d_inode;
 382                         struct address_space *mapping = file->f_mapping;
 383
 384                         get_file(file);
 385                         if (tmp->vm_flags & VM_DENYWRITE)
 386                                 atomic_dec(&inode->i_writecount);
 387                         spin_lock(&mapping->i_mmap_lock);
 388                         if (tmp->vm_flags & VM_SHARED)
 389                                 mapping->i_mmap_writable++;
 390                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
 391                         flush_dcache_mmap_lock(mapping);
 392                         /* insert tmp into the share list, just after mpnt */
 393                         vma_prio_tree_add(tmp, mpnt);
 394                         flush_dcache_mmap_unlock(mapping);
 395                         spin_unlock(&mapping->i_mmap_lock);
 396                 }
 397
 398                 /*
 399                  * Clear hugetlb-related page reserves for children. This only
 400                  * affects MAP_PRIVATE mappings. Faults generated by the child
 401                  * are not guaranteed to succeed, even if read-only
 402                  */
 403                 if (is_vm_hugetlb_page(tmp))
 404                         reset_vma_resv_huge_pages(tmp);
 405
 406                 /*
 407                  * Link in the new vma and copy the page table entries.
 408                  */
 409                 *pprev = tmp;
 410                 pprev = &tmp->vm_next;
 411
 412                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 413                 rb_link = &tmp->vm_rb.rb_right;
 414                 rb_parent = &tmp->vm_rb;
 415
 416                 mm->map_count++;
 417                 retval = copy_page_range(mm, oldmm, mpnt);
 418
 419                 if (tmp->vm_ops && tmp->vm_ops->open)
 420                         tmp->vm_ops->open(tmp);
 421
 422                 if (retval)
 423                         goto out;
 424         }
 425         /* a new mm has just been created */
 426         arch_dup_mmap(oldmm, mm);
 427         retval = 0;
 428 out:
 429         up_write(&mm->mmap_sem);
 430         flush_tlb_mm(oldmm);
 431         up_write(&oldmm->mmap_sem);
 432         return retval;
 433 fail_nomem_anon_vma_fork:
 434         mpol_put(pol);
 435 fail_nomem_policy:
 436         kmem_cache_free(vm_area_cachep, tmp);
 437 fail_nomem:
 438         retval = -ENOMEM;
 439         vm_unacct_memory(charge);
 440         goto out;
 441 }
 442
 443 static inline int mm_alloc_pgd(struct mm_struct * mm)
 444 {
 445         mm->pgd = pgd_alloc(mm);
 446         if (unlikely(!mm->pgd))
 447                 return -ENOMEM;
 448         return 0;
 449 }
 450
 451 static inline void mm_free_pgd(struct mm_struct * mm)
 452 {
 453         pgd_free(mm, mm->pgd);
 454 }
 455 #else
 456 #define dup_mmap(mm, oldmm)     (0)
 457 #define mm_alloc_pgd(mm)        (0)
 458 #define mm_free_pgd(mm)
 459 #endif /* CONFIG_MMU */
 460
 461 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 462
 463 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 464 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 465
 466 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 467
 468 static int __init coredump_filter_setup(char *s)
 469 {
 470         default_dump_filter =
 471                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 472                 MMF_DUMP_FILTER_MASK;
 473         return 1;
 474 }
 475
 476 __setup("coredump_filter=", coredump_filter_setup);
 477
 478 #include <linux/init_task.h>
 479
 480 static void mm_init_aio(struct mm_struct *mm)
 481 {
 482 #ifdef CONFIG_AIO
 483         spin_lock_init(&mm->ioctx_lock);
 484         INIT_HLIST_HEAD(&mm->ioctx_list);
 485 #endif
 486 }
 487
 488 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
 489 {
 490         atomic_set(&mm->mm_users, 1);
 491         atomic_set(&mm->mm_count, 1);
 492         init_rwsem(&mm->mmap_sem);
 493         INIT_LIST_HEAD(&mm->mmlist);
 494         mm->flags = (current->mm) ?
 495                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
 496         mm->core_state = NULL;
 497         mm->nr_ptes = 0;
 498         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 499         spin_lock_init(&mm->page_table_lock);
 500         mm->free_area_cache = TASK_UNMAPPED_BASE;
 501         mm->cached_hole_size = ~0UL;
 502         mm_init_aio(mm);
 503         mm_init_owner(mm, p);
 504
 505         if (likely(!mm_alloc_pgd(mm))) {
 506                 mm->def_flags = 0;
 507                 mmu_notifier_mm_init(mm);
 508                 return mm;
 509         }
 510
 511         free_mm(mm);
 512         return NULL;
 513 }
 514
 515 /*
 516  * Allocate and initialize an mm_struct.
 517  */
 518 struct mm_struct * mm_alloc(void)
 519 {
 520         struct mm_struct * mm;
 521
 522         mm = allocate_mm();
 523         if (mm) {
 524                 memset(mm, 0, sizeof(*mm));
 525                 mm = mm_init(mm, current);
 526         }
 527         return mm;
 528 }
 529
 530 /*
 531  * Called when the last reference to the mm
 532  * is dropped: either by a lazy thread or by
 533  * mmput. Free the page directory and the mm.
 534  */
 535 void __mmdrop(struct mm_struct *mm)
 536 {
 537         BUG_ON(mm == &init_mm);
 538         mm_free_pgd(mm);
 539         destroy_context(mm);
 540         mmu_notifier_mm_destroy(mm);
 541         free_mm(mm);
 542 }
 543 EXPORT_SYMBOL_GPL(__mmdrop);
 544
 545 /*
 546  * Decrement the use count and release all resources for an mm.
 547  */
 548 void mmput(struct mm_struct *mm)
 549 {
 550         might_sleep();
 551
 552         if (atomic_dec_and_test(&mm->mm_users)) {
 553                 exit_aio(mm);
 554                 ksm_exit(mm);
 555                 exit_mmap(mm);
 556                 set_mm_exe_file(mm, NULL);
 557                 if (!list_empty(&mm->mmlist)) {
 558                         spin_lock(&mmlist_lock);
 559                         list_del(&mm->mmlist);
 560                         spin_unlock(&mmlist_lock);
 561                 }
 562                 put_swap_token(mm);
 563                 if (mm->binfmt)
 564                         module_put(mm->binfmt->module);
 565                 mmdrop(mm);
 566         }
 567 }
 568 EXPORT_SYMBOL_GPL(mmput);
 569
 570 /**
 571  * get_task_mm - acquire a reference to the task's mm
 572  *
 573  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 574  * this kernel workthread has transiently adopted a user mm with use_mm,
 575  * to do its AIO) is not set and if so returns a reference to it, after
 576  * bumping up the use count.  User must release the mm via mmput()
 577  * after use.  Typically used by /proc and ptrace.
 578  */
 579 struct mm_struct *get_task_mm(struct task_struct *task)
 580 {
 581         struct mm_struct *mm;
 582
 583         task_lock(task);
 584         mm = task->mm;
 585         if (mm) {
 586                 if (task->flags & PF_KTHREAD)
 587                         mm = NULL;
 588                 else
 589                         atomic_inc(&mm->mm_users);
 590         }
 591         task_unlock(task);
 592         return mm;
 593 }
 594 EXPORT_SYMBOL_GPL(get_task_mm);
 595
 596 /* Please note the differences between mmput and mm_release.
 597  * mmput is called whenever we stop holding onto a mm_struct,
 598  * error success whatever.
 599  *
 600  * mm_release is called after a mm_struct has been removed
 601  * from the current process.
 602  *
 603  * This difference is important for error handling, when we
 604  * only half set up a mm_struct for a new process and need to restore
 605  * the old one.  Because we mmput the new mm_struct before
 606  * restoring the old one. . .
 607  * Eric Biederman 10 January 1998
 608  */
 609 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 610 {
 611         struct completion *vfork_done = tsk->vfork_done;
 612
 613         /* Get rid of any futexes when releasing the mm */
 614 #ifdef CONFIG_FUTEX
 615         if (unlikely(tsk->robust_list)) {
 616                 exit_robust_list(tsk);
 617                 tsk->robust_list = NULL;
 618         }
 619 #ifdef CONFIG_COMPAT
 620         if (unlikely(tsk->compat_robust_list)) {
 621                 compat_exit_robust_list(tsk);
 622                 tsk->compat_robust_list = NULL;
 623         }
 624 #endif
 625         if (unlikely(!list_empty(&tsk->pi_state_list)))
 626                 exit_pi_state_list(tsk);
 627 #endif
 628
 629         /* Get rid of any cached register state */
 630         deactivate_mm(tsk, mm);
 631
 632         /* notify parent sleeping on vfork() */
 633         if (vfork_done) {
 634                 tsk->vfork_done = NULL;
 635                 complete(vfork_done);
 636         }
 637
 638         /*
 639          * If we're exiting normally, clear a user-space tid field if
 640          * requested.  We leave this alone when dying by signal, to leave
 641          * the value intact in a core dump, and to save the unnecessary
 642          * trouble otherwise.  Userland only wants this done for a sys_exit.
 643          */
 644         if (tsk->clear_child_tid) {
 645                 if (!(tsk->flags & PF_SIGNALED) &&
 646                     atomic_read(&mm->mm_users) > 1) {
 647                         /*
 648                          * We don't check the error code - if userspace has
 649                          * not set up a proper pointer then tough luck.
 650                          */
 651                         put_user(0, tsk->clear_child_tid);
 652                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 653                                         1, NULL, NULL, 0);
 654                 }
 655                 tsk->clear_child_tid = NULL;
 656         }
 657 }
 658
 659 /*
 660  * Allocate a new mm structure and copy contents from the
 661  * mm structure of the passed in task structure.
 662  */
 663 struct mm_struct *dup_mm(struct task_struct *tsk)
 664 {
 665         struct mm_struct *mm, *oldmm = current->mm;
 666         int err;
 667
 668         if (!oldmm)
 669                 return NULL;
 670
 671         mm = allocate_mm();
 672         if (!mm)
 673                 goto fail_nomem;
 674
 675         memcpy(mm, oldmm, sizeof(*mm));
 676
 677         /* Initializing for Swap token stuff */
 678         mm->token_priority = 0;
 679         mm->last_interval = 0;
 680
 681         if (!mm_init(mm, tsk))
 682                 goto fail_nomem;
 683
 684         if (init_new_context(tsk, mm))
 685                 goto fail_nocontext;
 686
 687         dup_mm_exe_file(oldmm, mm);
 688
 689         err = dup_mmap(mm, oldmm);
 690         if (err)
 691                 goto free_pt;
 692
 693         mm->hiwater_rss = get_mm_rss(mm);
 694         mm->hiwater_vm = mm->total_vm;
 695
 696         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 697                 goto free_pt;
 698
 699         return mm;
 700
 701 free_pt:
 702         /* don't put binfmt in mmput, we haven't got module yet */
 703         mm->binfmt = NULL;
 704         mmput(mm);
 705
 706 fail_nomem:
 707         return NULL;
 708
 709 fail_nocontext:
 710         /*
 711          * If init_new_context() failed, we cannot use mmput() to free the mm
 712          * because it calls destroy_context()
 713          */
 714         mm_free_pgd(mm);
 715         free_mm(mm);
 716         return NULL;
 717 }
 718
 719 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 720 {
 721         struct mm_struct * mm, *oldmm;
 722         int retval;
 723
 724         tsk->min_flt = tsk->maj_flt = 0;
 725         tsk->nvcsw = tsk->nivcsw = 0;
 726 #ifdef CONFIG_DETECT_HUNG_TASK
 727         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 728 #endif
 729
 730         tsk->mm = NULL;
 731         tsk->active_mm = NULL;
 732
 733         /*
 734          * Are we cloning a kernel thread?
 735          *
 736          * We need to steal a active VM for that..
 737          */
 738         oldmm = current->mm;
 739         if (!oldmm)
 740                 return 0;
 741
 742         if (clone_flags & CLONE_VM) {
 743                 atomic_inc(&oldmm->mm_users);
 744                 mm = oldmm;
 745                 goto good_mm;
 746         }
 747
 748         retval = -ENOMEM;
 749         mm = dup_mm(tsk);
 750         if (!mm)
 751                 goto fail_nomem;
 752
 753 good_mm:
 754         /* Initializing for Swap token stuff */
 755         mm->token_priority = 0;
 756         mm->last_interval = 0;
 757
 758         tsk->mm = mm;
 759         tsk->active_mm = mm;
 760         return 0;
 761
 762 fail_nomem:
 763         return retval;
 764 }
 765
 766 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 767 {
 768         struct fs_struct *fs = current->fs;
 769         if (clone_flags & CLONE_FS) {
 770                 /* tsk->fs is already what we want */
 771                 write_lock(&fs->lock);
 772                 if (fs->in_exec) {
 773                         write_unlock(&fs->lock);
 774                         return -EAGAIN;
 775                 }
 776                 fs->users++;
 777                 write_unlock(&fs->lock);
 778                 return 0;
 779         }
 780         tsk->fs = copy_fs_struct(fs);
 781         if (!tsk->fs)
 782                 return -ENOMEM;
 783         return 0;
 784 }
 785
 786 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 787 {
 788         struct files_struct *oldf, *newf;
 789         int error = 0;
 790
 791         /*
 792          * A background process may not have any files ...
 793          */
 794         oldf = current->files;
 795         if (!oldf)
 796                 goto out;
 797
 798         if (clone_flags & CLONE_FILES) {
 799                 atomic_inc(&oldf->count);
 800                 goto out;
 801         }
 802
 803         newf = dup_fd(oldf, &error);
 804         if (!newf)
 805                 goto out;
 806
 807         tsk->files = newf;
 808         error = 0;
 809 out:
 810         return error;
 811 }
 812
 813 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 814 {
 815 #ifdef CONFIG_BLOCK
 816         struct io_context *ioc = current->io_context;
 817
 818         if (!ioc)
 819                 return 0;
 820         /*
 821          * Share io context with parent, if CLONE_IO is set
 822          */
 823         if (clone_flags & CLONE_IO) {
 824                 tsk->io_context = ioc_task_link(ioc);
 825                 if (unlikely(!tsk->io_context))
 826                         return -ENOMEM;
 827         } else if (ioprio_valid(ioc->ioprio)) {
 828                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
 829                 if (unlikely(!tsk->io_context))
 830                         return -ENOMEM;
 831
 832                 tsk->io_context->ioprio = ioc->ioprio;
 833         }
 834 #endif
 835         return 0;
 836 }
 837
 838 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 839 {
 840         struct sighand_struct *sig;
 841
 842         if (clone_flags & CLONE_SIGHAND) {
 843                 atomic_inc(&current->sighand->count);
 844                 return 0;
 845         }
 846         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 847         rcu_assign_pointer(tsk->sighand, sig);
 848         if (!sig)
 849                 return -ENOMEM;
 850         atomic_set(&sig->count, 1);
 851         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 852         return 0;
 853 }
 854
 855 void __cleanup_sighand(struct sighand_struct *sighand)
 856 {
 857         if (atomic_dec_and_test(&sighand->count))
 858                 kmem_cache_free(sighand_cachep, sighand);
 859 }
 860
 861
 862 /*
 863  * Initialize POSIX timer handling for a thread group.
 864  */
 865 static void posix_cpu_timers_init_group(struct signal_struct *sig)
 866 {
 867         unsigned long cpu_limit;
 868
 869         /* Thread group counters. */
 870         thread_group_cputime_init(sig);
 871
 872         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
 873         if (cpu_limit != RLIM_INFINITY) {
 874                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
 875                 sig->cputimer.running = 1;
 876         }
 877
 878         /* The timer lists. */
 879         INIT_LIST_HEAD(&sig->cpu_timers[0]);
 880         INIT_LIST_HEAD(&sig->cpu_timers[1]);
 881         INIT_LIST_HEAD(&sig->cpu_timers[2]);
 882 }
 883
 884 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
 885 {
 886         struct signal_struct *sig;
 887
 888         if (clone_flags & CLONE_THREAD)
 889                 return 0;
 890
 891         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
 892         tsk->signal = sig;
 893         if (!sig)
 894                 return -ENOMEM;
 895
 896         sig->nr_threads = 1;
 897         atomic_set(&sig->live, 1);
 898         atomic_set(&sig->sigcnt, 1);
 899         init_waitqueue_head(&sig->wait_chldexit);
 900         if (clone_flags & CLONE_NEWPID)
 901                 sig->flags |= SIGNAL_UNKILLABLE;
 902         sig->curr_target = tsk;
 903         init_sigpending(&sig->shared_pending);
 904         INIT_LIST_HEAD(&sig->posix_timers);
 905
 906         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 907         sig->real_timer.function = it_real_fn;
 908
 909         task_lock(current->group_leader);
 910         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
 911         task_unlock(current->group_leader);
 912
 913         posix_cpu_timers_init_group(sig);
 914
 915         tty_audit_fork(sig);
 916
 917         sig->oom_adj = current->signal->oom_adj;
 918
 919         return 0;
 920 }
 921
 922 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
 923 {
 924         unsigned long new_flags = p->flags;
 925
 926         new_flags &= ~PF_SUPERPRIV;
 927         new_flags |= PF_FORKNOEXEC;
 928         new_flags |= PF_STARTING;
 929         p->flags = new_flags;
 930         clear_freeze_flag(p);
 931 }
 932
 933 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
 934 {
 935         current->clear_child_tid = tidptr;
 936
 937         return task_pid_vnr(current);
 938 }
 939
 940 static void rt_mutex_init_task(struct task_struct *p)
 941 {
 942         raw_spin_lock_init(&p->pi_lock);
 943 #ifdef CONFIG_RT_MUTEXES
 944         plist_head_init_raw(&p->pi_waiters, &p->pi_lock);
 945         p->pi_blocked_on = NULL;
 946 #endif
 947 }
 948
 949 #ifdef CONFIG_MM_OWNER
 950 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 951 {
 952         mm->owner = p;
 953 }
 954 #endif /* CONFIG_MM_OWNER */
 955
 956 /*
 957  * Initialize POSIX timer handling for a single task.
 958  */
 959 static void posix_cpu_timers_init(struct task_struct *tsk)
 960 {
 961         tsk->cputime_expires.prof_exp = cputime_zero;
 962         tsk->cputime_expires.virt_exp = cputime_zero;
 963         tsk->cputime_expires.sched_exp = 0;
 964         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
 965         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
 966         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
 967 }
 968
 969 /*
 970  * This creates a new process as a copy of the old one,
 971  * but does not actually start it yet.
 972  *
 973  * It copies the registers, and all the appropriate
 974  * parts of the process environment (as per the clone
 975  * flags). The actual kick-off is left to the caller.
 976  */
 977 static struct task_struct *copy_process(unsigned long clone_flags,
 978                                         unsigned long stack_start,
 979                                         struct pt_regs *regs,
 980                                         unsigned long stack_size,
 981                                         int __user *child_tidptr,
 982                                         struct pid *pid,
 983                                         int trace)
 984 {
 985         int retval;
 986         struct task_struct *p;
 987         int cgroup_callbacks_done = 0;
 988
 989         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 990                 return ERR_PTR(-EINVAL);
 991
 992         /*
 993          * Thread groups must share signals as well, and detached threads
 994          * can only be started up within the thread group.
 995          */
 996         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
 997                 return ERR_PTR(-EINVAL);
 998
 999         /*
1000          * Shared signal handlers imply shared VM. By way of the above,
1001          * thread groups also imply shared VM. Blocking this case allows
1002          * for various simplifications in other code.
1003          */
1004         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1005                 return ERR_PTR(-EINVAL);
1006
1007         /*
1008          * Siblings of global init remain as zombies on exit since they are
1009          * not reaped by their parent (swapper). To solve this and to avoid
1010          * multi-rooted process trees, prevent global and container-inits
1011          * from creating siblings.
1012          */
1013         if ((clone_flags & CLONE_PARENT) &&
1014                                 current->signal->flags & SIGNAL_UNKILLABLE)
1015                 return ERR_PTR(-EINVAL);
1016
1017         retval = security_task_create(clone_flags);
1018         if (retval)
1019                 goto fork_out;
1020
1021         retval = -ENOMEM;
1022         p = dup_task_struct(current);
1023         if (!p)
1024                 goto fork_out;
1025
1026         ftrace_graph_init_task(p);
1027
1028         rt_mutex_init_task(p);
1029
1030 #ifdef CONFIG_PROVE_LOCKING
1031         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1032         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1033 #endif
1034         retval = -EAGAIN;
1035         if (atomic_read(&p->real_cred->user->processes) >=
1036                         task_rlimit(p, RLIMIT_NPROC)) {
1037                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1038                     p->real_cred->user != INIT_USER)
1039                         goto bad_fork_free;
1040         }
1041
1042         retval = copy_creds(p, clone_flags);
1043         if (retval < 0)
1044                 goto bad_fork_free;
1045
1046         /*
1047          * If multiple threads are within copy_process(), then this check
1048          * triggers too late. This doesn't hurt, the check is only there
1049          * to stop root fork bombs.
1050          */
1051         retval = -EAGAIN;
1052         if (nr_threads >= max_threads)
1053                 goto bad_fork_cleanup_count;
1054
1055         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1056                 goto bad_fork_cleanup_count;
1057
1058         p->did_exec = 0;
1059         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1060         copy_flags(clone_flags, p);
1061         INIT_LIST_HEAD(&p->children);
1062         INIT_LIST_HEAD(&p->sibling);
1063         rcu_copy_process(p);
1064         p->vfork_done = NULL;
1065         spin_lock_init(&p->alloc_lock);
1066
1067         init_sigpending(&p->pending);
1068
1069         p->utime = cputime_zero;
1070         p->stime = cputime_zero;
1071         p->gtime = cputime_zero;
1072         p->utimescaled = cputime_zero;
1073         p->stimescaled = cputime_zero;
1074 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1075         p->prev_utime = cputime_zero;
1076         p->prev_stime = cputime_zero;
1077 #endif
1078 #if defined(SPLIT_RSS_COUNTING)
1079         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1080 #endif
1081
1082         p->default_timer_slack_ns = current->timer_slack_ns;
1083
1084         task_io_accounting_init(&p->ioac);
1085         acct_clear_integrals(p);
1086
1087         posix_cpu_timers_init(p);
1088
1089         p->lock_depth = -1;             /* -1 = no lock */
1090         do_posix_clock_monotonic_gettime(&p->start_time);
1091         p->real_start_time = p->start_time;
1092         monotonic_to_bootbased(&p->real_start_time);
1093         p->io_context = NULL;
1094         p->audit_context = NULL;
1095         cgroup_fork(p);
1096 #ifdef CONFIG_NUMA
1097         p->mempolicy = mpol_dup(p->mempolicy);
1098         if (IS_ERR(p->mempolicy)) {
1099                 retval = PTR_ERR(p->mempolicy);
1100                 p->mempolicy = NULL;
1101                 goto bad_fork_cleanup_cgroup;
1102         }
1103         mpol_fix_fork_child_flag(p);
1104 #endif
1105 #ifdef CONFIG_TRACE_IRQFLAGS
1106         p->irq_events = 0;
1107 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1108         p->hardirqs_enabled = 1;
1109 #else
1110         p->hardirqs_enabled = 0;
1111 #endif
1112         p->hardirq_enable_ip = 0;
1113         p->hardirq_enable_event = 0;
1114         p->hardirq_disable_ip = _THIS_IP_;
1115         p->hardirq_disable_event = 0;
1116         p->softirqs_enabled = 1;
1117         p->softirq_enable_ip = _THIS_IP_;
1118         p->softirq_enable_event = 0;
1119         p->softirq_disable_ip = 0;
1120         p->softirq_disable_event = 0;
1121         p->hardirq_context = 0;
1122         p->softirq_context = 0;
1123 #endif
1124 #ifdef CONFIG_LOCKDEP
1125         p->lockdep_depth = 0; /* no locks held yet */
1126         p->curr_chain_key = 0;
1127         p->lockdep_recursion = 0;
1128 #endif
1129
1130 #ifdef CONFIG_DEBUG_MUTEXES
1131         p->blocked_on = NULL; /* not blocked yet */
1132 #endif
1133 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1134         p->memcg_batch.do_batch = 0;
1135         p->memcg_batch.memcg = NULL;
1136 #endif
1137
1138         /* Perform scheduler related setup. Assign this task to a CPU. */
1139         sched_fork(p, clone_flags);
1140
1141         retval = perf_event_init_task(p);
1142         if (retval)
1143                 goto bad_fork_cleanup_policy;
1144
1145         if ((retval = audit_alloc(p)))
1146                 goto bad_fork_cleanup_policy;
1147         /* copy all the process information */
1148         if ((retval = copy_semundo(clone_flags, p)))
1149                 goto bad_fork_cleanup_audit;
1150         if ((retval = copy_files(clone_flags, p)))
1151                 goto bad_fork_cleanup_semundo;
1152         if ((retval = copy_fs(clone_flags, p)))
1153                 goto bad_fork_cleanup_files;
1154         if ((retval = copy_sighand(clone_flags, p)))
1155                 goto bad_fork_cleanup_fs;
1156         if ((retval = copy_signal(clone_flags, p)))
1157                 goto bad_fork_cleanup_sighand;
1158         if ((retval = copy_mm(clone_flags, p)))
1159                 goto bad_fork_cleanup_signal;
1160         if ((retval = copy_namespaces(clone_flags, p)))
1161                 goto bad_fork_cleanup_mm;
1162         if ((retval = copy_io(clone_flags, p)))
1163                 goto bad_fork_cleanup_namespaces;
1164         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1165         if (retval)
1166                 goto bad_fork_cleanup_io;
1167
1168         if (pid != &init_struct_pid) {
1169                 retval = -ENOMEM;
1170                 pid = alloc_pid(p->nsproxy->pid_ns);
1171                 if (!pid)
1172                         goto bad_fork_cleanup_io;
1173
1174                 if (clone_flags & CLONE_NEWPID) {
1175                         retval = pid_ns_prepare_proc(p->nsproxy->pid_ns);
1176                         if (retval < 0)
1177                                 goto bad_fork_free_pid;
1178                 }
1179         }
1180
1181         p->pid = pid_nr(pid);
1182         p->tgid = p->pid;
1183         if (clone_flags & CLONE_THREAD)
1184                 p->tgid = current->tgid;
1185
1186         if (current->nsproxy != p->nsproxy) {
1187                 retval = ns_cgroup_clone(p, pid);
1188                 if (retval)
1189                         goto bad_fork_free_pid;
1190         }
1191
1192         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1193         /*
1194          * Clear TID on mm_release()?
1195          */
1196         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1197 #ifdef CONFIG_FUTEX
1198         p->robust_list = NULL;
1199 #ifdef CONFIG_COMPAT
1200         p->compat_robust_list = NULL;
1201 #endif
1202         INIT_LIST_HEAD(&p->pi_state_list);
1203         p->pi_state_cache = NULL;
1204 #endif
1205         /*
1206          * sigaltstack should be cleared when sharing the same VM
1207          */
1208         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1209                 p->sas_ss_sp = p->sas_ss_size = 0;
1210
1211         /*
1212          * Syscall tracing and stepping should be turned off in the
1213          * child regardless of CLONE_PTRACE.
1214          */
1215         user_disable_single_step(p);
1216         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1217 #ifdef TIF_SYSCALL_EMU
1218         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1219 #endif
1220         clear_all_latency_tracing(p);
1221
1222         /* ok, now we should be set up.. */
1223         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1224         p->pdeath_signal = 0;
1225         p->exit_state = 0;
1226
1227         /*
1228          * Ok, make it visible to the rest of the system.
1229          * We dont wake it up yet.
1230          */
1231         p->group_leader = p;
1232         INIT_LIST_HEAD(&p->thread_group);
1233
1234         /* Now that the task is set up, run cgroup callbacks if
1235          * necessary. We need to run them before the task is visible
1236          * on the tasklist. */
1237         cgroup_fork_callbacks(p);
1238         cgroup_callbacks_done = 1;
1239
1240         /* Need tasklist lock for parent etc handling! */
1241         write_lock_irq(&tasklist_lock);
1242
1243         /* CLONE_PARENT re-uses the old parent */
1244         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1245                 p->real_parent = current->real_parent;
1246                 p->parent_exec_id = current->parent_exec_id;
1247         } else {
1248                 p->real_parent = current;
1249                 p->parent_exec_id = current->self_exec_id;
1250         }
1251
1252         spin_lock(&current->sighand->siglock);
1253
1254         /*
1255          * Process group and session signals need to be delivered to just the
1256          * parent before the fork or both the parent and the child after the
1257          * fork. Restart if a signal comes in before we add the new process to
1258          * it's process group.
1259          * A fatal signal pending means that current will exit, so the new
1260          * thread can't slip out of an OOM kill (or normal SIGKILL).
1261          */
1262         recalc_sigpending();
1263         if (signal_pending(current)) {
1264                 spin_unlock(&current->sighand->siglock);
1265                 write_unlock_irq(&tasklist_lock);
1266                 retval = -ERESTARTNOINTR;
1267                 goto bad_fork_free_pid;
1268         }
1269
1270         if (clone_flags & CLONE_THREAD) {
1271                 current->signal->nr_threads++;
1272                 atomic_inc(&current->signal->live);
1273                 atomic_inc(&current->signal->sigcnt);
1274                 p->group_leader = current->group_leader;
1275                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1276         }
1277
1278         if (likely(p->pid)) {
1279                 tracehook_finish_clone(p, clone_flags, trace);
1280
1281                 if (thread_group_leader(p)) {
1282                         if (clone_flags & CLONE_NEWPID)
1283                                 p->nsproxy->pid_ns->child_reaper = p;
1284
1285                         p->signal->leader_pid = pid;
1286                         p->signal->tty = tty_kref_get(current->signal->tty);
1287                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1288                         attach_pid(p, PIDTYPE_SID, task_session(current));
1289                         list_add_tail(&p->sibling, &p->real_parent->children);
1290                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1291                         __get_cpu_var(process_counts)++;
1292                 }
1293                 attach_pid(p, PIDTYPE_PID, pid);
1294                 nr_threads++;
1295         }
1296
1297         total_forks++;
1298         spin_unlock(&current->sighand->siglock);
1299         write_unlock_irq(&tasklist_lock);
1300         proc_fork_connector(p);
1301         cgroup_post_fork(p);
1302         perf_event_fork(p);
1303         return p;
1304
1305 bad_fork_free_pid:
1306         if (pid != &init_struct_pid)
1307                 free_pid(pid);
1308 bad_fork_cleanup_io:
1309         if (p->io_context)
1310                 exit_io_context(p);
1311 bad_fork_cleanup_namespaces:
1312         exit_task_namespaces(p);
1313 bad_fork_cleanup_mm:
1314         if (p->mm)
1315                 mmput(p->mm);
1316 bad_fork_cleanup_signal:
1317         if (!(clone_flags & CLONE_THREAD))
1318                 free_signal_struct(p->signal);
1319 bad_fork_cleanup_sighand:
1320         __cleanup_sighand(p->sighand);
1321 bad_fork_cleanup_fs:
1322         exit_fs(p); /* blocking */
1323 bad_fork_cleanup_files:
1324         exit_files(p); /* blocking */
1325 bad_fork_cleanup_semundo:
1326         exit_sem(p);
1327 bad_fork_cleanup_audit:
1328         audit_free(p);
1329 bad_fork_cleanup_policy:
1330         perf_event_free_task(p);
1331 #ifdef CONFIG_NUMA
1332         mpol_put(p->mempolicy);
1333 bad_fork_cleanup_cgroup:
1334 #endif
1335         cgroup_exit(p, cgroup_callbacks_done);
1336         delayacct_tsk_free(p);
1337         module_put(task_thread_info(p)->exec_domain->module);
1338 bad_fork_cleanup_count:
1339         atomic_dec(&p->cred->user->processes);
1340         exit_creds(p);
1341 bad_fork_free:
1342         free_task(p);
1343 fork_out:
1344         return ERR_PTR(retval);
1345 }
1346
1347 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1348 {
1349         memset(regs, 0, sizeof(struct pt_regs));
1350         return regs;
1351 }
1352
1353 static inline void init_idle_pids(struct pid_link *links)
1354 {
1355         enum pid_type type;
1356
1357         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1358                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1359                 links[type].pid = &init_struct_pid;
1360         }
1361 }
1362
1363 struct task_struct * __cpuinit fork_idle(int cpu)
1364 {
1365         struct task_struct *task;
1366         struct pt_regs regs;
1367
1368         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1369                             &init_struct_pid, 0);
1370         if (!IS_ERR(task)) {
1371                 init_idle_pids(task->pids);
1372                 init_idle(task, cpu);
1373         }
1374
1375         return task;
1376 }
1377
1378 /*
1379  *  Ok, this is the main fork-routine.
1380  *
1381  * It copies the process, and if successful kick-starts
1382  * it and waits for it to finish using the VM if required.
1383  */
1384 long do_fork(unsigned long clone_flags,
1385               unsigned long stack_start,
1386               struct pt_regs *regs,
1387               unsigned long stack_size,
1388               int __user *parent_tidptr,
1389               int __user *child_tidptr)
1390 {
1391         struct task_struct *p;
1392         int trace = 0;
1393         long nr;
1394
1395         /*
1396          * Do some preliminary argument and permissions checking before we
1397          * actually start allocating stuff
1398          */
1399         if (clone_flags & CLONE_NEWUSER) {
1400                 if (clone_flags & CLONE_THREAD)
1401                         return -EINVAL;
1402                 /* hopefully this check will go away when userns support is
1403                  * complete
1404                  */
1405                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1406                                 !capable(CAP_SETGID))
1407                         return -EPERM;
1408         }
1409
1410         /*
1411          * We hope to recycle these flags after 2.6.26
1412          */
1413         if (unlikely(clone_flags & CLONE_STOPPED)) {
1414                 static int __read_mostly count = 100;
1415
1416                 if (count > 0 && printk_ratelimit()) {
1417                         char comm[TASK_COMM_LEN];
1418
1419                         count--;
1420                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1421                                         "clone flags 0x%lx\n",
1422                                 get_task_comm(comm, current),
1423                                 clone_flags & CLONE_STOPPED);
1424                 }
1425         }
1426
1427         /*
1428          * When called from kernel_thread, don't do user tracing stuff.
1429          */
1430         if (likely(user_mode(regs)))
1431                 trace = tracehook_prepare_clone(clone_flags);
1432
1433         p = copy_process(clone_flags, stack_start, regs, stack_size,
1434                          child_tidptr, NULL, trace);
1435         /*
1436          * Do this prior waking up the new thread - the thread pointer
1437          * might get invalid after that point, if the thread exits quickly.
1438          */
1439         if (!IS_ERR(p)) {
1440                 struct completion vfork;
1441
1442                 trace_sched_process_fork(current, p);
1443
1444                 nr = task_pid_vnr(p);
1445
1446                 if (clone_flags & CLONE_PARENT_SETTID)
1447                         put_user(nr, parent_tidptr);
1448
1449                 if (clone_flags & CLONE_VFORK) {
1450                         p->vfork_done = &vfork;
1451                         init_completion(&vfork);
1452                 }
1453
1454                 audit_finish_fork(p);
1455                 tracehook_report_clone(regs, clone_flags, nr, p);
1456
1457                 /*
1458                  * We set PF_STARTING at creation in case tracing wants to
1459                  * use this to distinguish a fully live task from one that
1460                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1461                  * clear it and set the child going.
1462                  */
1463                 p->flags &= ~PF_STARTING;
1464
1465                 if (unlikely(clone_flags & CLONE_STOPPED)) {
1466                         /*
1467                          * We'll start up with an immediate SIGSTOP.
1468                          */
1469                         sigaddset(&p->pending.signal, SIGSTOP);
1470                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1471                         __set_task_state(p, TASK_STOPPED);
1472                 } else {
1473                         wake_up_new_task(p, clone_flags);
1474                 }
1475
1476                 tracehook_report_clone_complete(trace, regs,
1477                                                 clone_flags, nr, p);
1478
1479                 if (clone_flags & CLONE_VFORK) {
1480                         freezer_do_not_count();
1481                         wait_for_completion(&vfork);
1482                         freezer_count();
1483                         tracehook_report_vfork_done(p, nr);
1484                 }
1485         } else {
1486                 nr = PTR_ERR(p);
1487         }
1488         return nr;
1489 }
1490
1491 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1492 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1493 #endif
1494
1495 static void sighand_ctor(void *data)
1496 {
1497         struct sighand_struct *sighand = data;
1498
1499         spin_lock_init(&sighand->siglock);
1500         init_waitqueue_head(&sighand->signalfd_wqh);
1501 }
1502
1503 void __init proc_caches_init(void)
1504 {
1505         sighand_cachep = kmem_cache_create("sighand_cache",
1506                         sizeof(struct sighand_struct), 0,
1507                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1508                         SLAB_NOTRACK, sighand_ctor);
1509         signal_cachep = kmem_cache_create("signal_cache",
1510                         sizeof(struct signal_struct), 0,
1511                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1512         files_cachep = kmem_cache_create("files_cache",
1513                         sizeof(struct files_struct), 0,
1514                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1515         fs_cachep = kmem_cache_create("fs_cache",
1516                         sizeof(struct fs_struct), 0,
1517                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1518         mm_cachep = kmem_cache_create("mm_struct",
1519                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1520                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1521         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1522         mmap_init();
1523 }
1524
1525 /*
1526  * Check constraints on flags passed to the unshare system call and
1527  * force unsharing of additional process context as appropriate.
1528  */
1529 static void check_unshare_flags(unsigned long *flags_ptr)
1530 {
1531         /*
1532          * If unsharing a thread from a thread group, must also
1533          * unshare vm.
1534          */
1535         if (*flags_ptr & CLONE_THREAD)
1536                 *flags_ptr |= CLONE_VM;
1537
1538         /*
1539          * If unsharing vm, must also unshare signal handlers.
1540          */
1541         if (*flags_ptr & CLONE_VM)
1542                 *flags_ptr |= CLONE_SIGHAND;
1543
1544         /*
1545          * If unsharing namespace, must also unshare filesystem information.
1546          */
1547         if (*flags_ptr & CLONE_NEWNS)
1548                 *flags_ptr |= CLONE_FS;
1549 }
1550
1551 /*
1552  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1553  */
1554 static int unshare_thread(unsigned long unshare_flags)
1555 {
1556         if (unshare_flags & CLONE_THREAD)
1557                 return -EINVAL;
1558
1559         return 0;
1560 }
1561
1562 /*
1563  * Unshare the filesystem structure if it is being shared
1564  */
1565 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1566 {
1567         struct fs_struct *fs = current->fs;
1568
1569         if (!(unshare_flags & CLONE_FS) || !fs)
1570                 return 0;
1571
1572         /* don't need lock here; in the worst case we'll do useless copy */
1573         if (fs->users == 1)
1574                 return 0;
1575
1576         *new_fsp = copy_fs_struct(fs);
1577         if (!*new_fsp)
1578                 return -ENOMEM;
1579
1580         return 0;
1581 }
1582
1583 /*
1584  * Unsharing of sighand is not supported yet
1585  */
1586 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1587 {
1588         struct sighand_struct *sigh = current->sighand;
1589
1590         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1591                 return -EINVAL;
1592         else
1593                 return 0;
1594 }
1595
1596 /*
1597  * Unshare vm if it is being shared
1598  */
1599 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1600 {
1601         struct mm_struct *mm = current->mm;
1602
1603         if ((unshare_flags & CLONE_VM) &&
1604             (mm && atomic_read(&mm->mm_users) > 1)) {
1605                 return -EINVAL;
1606         }
1607
1608         return 0;
1609 }
1610
1611 /*
1612  * Unshare file descriptor table if it is being shared
1613  */
1614 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1615 {
1616         struct files_struct *fd = current->files;
1617         int error = 0;
1618
1619         if ((unshare_flags & CLONE_FILES) &&
1620             (fd && atomic_read(&fd->count) > 1)) {
1621                 *new_fdp = dup_fd(fd, &error);
1622                 if (!*new_fdp)
1623                         return error;
1624         }
1625
1626         return 0;
1627 }
1628
1629 /*
1630  * unshare allows a process to 'unshare' part of the process
1631  * context which was originally shared using clone.  copy_*
1632  * functions used by do_fork() cannot be used here directly
1633  * because they modify an inactive task_struct that is being
1634  * constructed. Here we are modifying the current, active,
1635  * task_struct.
1636  */
1637 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1638 {
1639         int err = 0;
1640         struct fs_struct *fs, *new_fs = NULL;
1641         struct sighand_struct *new_sigh = NULL;
1642         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1643         struct files_struct *fd, *new_fd = NULL;
1644         struct nsproxy *new_nsproxy = NULL;
1645         int do_sysvsem = 0;
1646
1647         check_unshare_flags(&unshare_flags);
1648
1649         /* Return -EINVAL for all unsupported flags */
1650         err = -EINVAL;
1651         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1652                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1653                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1654                 goto bad_unshare_out;
1655
1656         /*
1657          * CLONE_NEWIPC must also detach from the undolist: after switching
1658          * to a new ipc namespace, the semaphore arrays from the old
1659          * namespace are unreachable.
1660          */
1661         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1662                 do_sysvsem = 1;
1663         if ((err = unshare_thread(unshare_flags)))
1664                 goto bad_unshare_out;
1665         if ((err = unshare_fs(unshare_flags, &new_fs)))
1666                 goto bad_unshare_cleanup_thread;
1667         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1668                 goto bad_unshare_cleanup_fs;
1669         if ((err = unshare_vm(unshare_flags, &new_mm)))
1670                 goto bad_unshare_cleanup_sigh;
1671         if ((err = unshare_fd(unshare_flags, &new_fd)))
1672                 goto bad_unshare_cleanup_vm;
1673         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1674                         new_fs)))
1675                 goto bad_unshare_cleanup_fd;
1676
1677         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1678                 if (do_sysvsem) {
1679                         /*
1680                          * CLONE_SYSVSEM is equivalent to sys_exit().
1681                          */
1682                         exit_sem(current);
1683                 }
1684
1685                 if (new_nsproxy) {
1686                         switch_task_namespaces(current, new_nsproxy);
1687                         new_nsproxy = NULL;
1688                 }
1689
1690                 task_lock(current);
1691
1692                 if (new_fs) {
1693                         fs = current->fs;
1694                         write_lock(&fs->lock);
1695                         current->fs = new_fs;
1696                         if (--fs->users)
1697                                 new_fs = NULL;
1698                         else
1699                                 new_fs = fs;
1700                         write_unlock(&fs->lock);
1701                 }
1702
1703                 if (new_mm) {
1704                         mm = current->mm;
1705                         active_mm = current->active_mm;
1706                         current->mm = new_mm;
1707                         current->active_mm = new_mm;
1708                         activate_mm(active_mm, new_mm);
1709                         new_mm = mm;
1710                 }
1711
1712                 if (new_fd) {
1713                         fd = current->files;
1714                         current->files = new_fd;
1715                         new_fd = fd;
1716                 }
1717
1718                 task_unlock(current);
1719         }
1720
1721         if (new_nsproxy)
1722                 put_nsproxy(new_nsproxy);
1723
1724 bad_unshare_cleanup_fd:
1725         if (new_fd)
1726                 put_files_struct(new_fd);
1727
1728 bad_unshare_cleanup_vm:
1729         if (new_mm)
1730                 mmput(new_mm);
1731
1732 bad_unshare_cleanup_sigh:
1733         if (new_sigh)
1734                 if (atomic_dec_and_test(&new_sigh->count))
1735                         kmem_cache_free(sighand_cachep, new_sigh);
1736
1737 bad_unshare_cleanup_fs:
1738         if (new_fs)
1739                 free_fs_struct(new_fs);
1740
1741 bad_unshare_cleanup_thread:
1742 bad_unshare_out:
1743         return err;
1744 }
1745
1746 /*
1747  *      Helper to unshare the files of the current task.
1748  *      We don't want to expose copy_files internals to
1749  *      the exec layer of the kernel.
1750  */
1751
1752 int unshare_files(struct files_struct **displaced)
1753 {
1754         struct task_struct *task = current;
1755         struct files_struct *copy = NULL;
1756         int error;
1757
1758         error = unshare_fd(CLONE_FILES, &copy);
1759         if (error || !copy) {
1760                 *displaced = NULL;
1761                 return error;
1762         }
1763         *displaced = task->files;
1764         task_lock(task);
1765         task->files = copy;
1766         task_unlock(task);
1767         return 0;
1768 }