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