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