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/mnt_namespace.h>
  21 #include <linux/personality.h>
  22 #include <linux/mempolicy.h>
  23 #include <linux/sem.h>
  24 #include <linux/file.h>
  25 #include <linux/fdtable.h>
  26 #include <linux/iocontext.h>
  27 #include <linux/key.h>
  28 #include <linux/binfmts.h>
  29 #include <linux/mman.h>
  30 #include <linux/mmu_notifier.h>
  31 #include <linux/fs.h>
  32 #include <linux/nsproxy.h>
  33 #include <linux/capability.h>
  34 #include <linux/cpu.h>
  35 #include <linux/cgroup.h>
  36 #include <linux/security.h>
  37 #include <linux/hugetlb.h>
  38 #include <linux/swap.h>
  39 #include <linux/syscalls.h>
  40 #include <linux/jiffies.h>
  41 #include <linux/tracehook.h>
  42 #include <linux/futex.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/profile.h>
  50 #include <linux/rmap.h>
  51 #include <linux/acct.h>
  52 #include <linux/tsacct_kern.h>
  53 #include <linux/cn_proc.h>
  54 #include <linux/freezer.h>
  55 #include <linux/delayacct.h>
  56 #include <linux/taskstats_kern.h>
  57 #include <linux/random.h>
  58 #include <linux/tty.h>
  59 #include <linux/proc_fs.h>
  60 #include <linux/blkdev.h>
  61 #include <trace/sched.h>
  62
  63 #include <asm/pgtable.h>
  64 #include <asm/pgalloc.h>
  65 #include <asm/uaccess.h>
  66 #include <asm/mmu_context.h>
  67 #include <asm/cacheflush.h>
  68 #include <asm/tlbflush.h>
  69
  70 /*
  71  * Protected counters by write_lock_irq(&tasklist_lock)
  72  */
  73 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  74 int nr_threads;                 /* The idle threads do not count.. */
  75
  76 int max_threads;                /* tunable limit on nr_threads */
  77
  78 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  79
  80 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  81
  82 int nr_processes(void)
  83 {
  84         int cpu;
  85         int total = 0;
  86
  87         for_each_online_cpu(cpu)
  88                 total += per_cpu(process_counts, cpu);
  89
  90         return total;
  91 }
  92
  93 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
  94 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
  95 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
  96 static struct kmem_cache *task_struct_cachep;
  97 #endif
  98
  99 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
 100 static inline struct thread_info *alloc_thread_info(struct task_struct *tsk)
 101 {
 102 #ifdef CONFIG_DEBUG_STACK_USAGE
 103         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
 104 #else
 105         gfp_t mask = GFP_KERNEL;
 106 #endif
 107         return (struct thread_info *)__get_free_pages(mask, THREAD_SIZE_ORDER);
 108 }
 109
 110 static inline void free_thread_info(struct thread_info *ti)
 111 {
 112         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 113 }
 114 #endif
 115
 116 /* SLAB cache for signal_struct structures (tsk->signal) */
 117 static struct kmem_cache *signal_cachep;
 118
 119 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 120 struct kmem_cache *sighand_cachep;
 121
 122 /* SLAB cache for files_struct structures (tsk->files) */
 123 struct kmem_cache *files_cachep;
 124
 125 /* SLAB cache for fs_struct structures (tsk->fs) */
 126 struct kmem_cache *fs_cachep;
 127
 128 /* SLAB cache for vm_area_struct structures */
 129 struct kmem_cache *vm_area_cachep;
 130
 131 /* SLAB cache for mm_struct structures (tsk->mm) */
 132 static struct kmem_cache *mm_cachep;
 133
 134 void free_task(struct task_struct *tsk)
 135 {
 136         prop_local_destroy_single(&tsk->dirties);
 137         free_thread_info(tsk->stack);
 138         rt_mutex_debug_task_free(tsk);
 139         free_task_struct(tsk);
 140 }
 141 EXPORT_SYMBOL(free_task);
 142
 143 void __put_task_struct(struct task_struct *tsk)
 144 {
 145         WARN_ON(!tsk->exit_state);
 146         WARN_ON(atomic_read(&tsk->usage));
 147         WARN_ON(tsk == current);
 148
 149         security_task_free(tsk);
 150         free_uid(tsk->user);
 151         put_group_info(tsk->group_info);
 152         delayacct_tsk_free(tsk);
 153
 154         if (!profile_handoff_task(tsk))
 155                 free_task(tsk);
 156 }
 157
 158 /*
 159  * macro override instead of weak attribute alias, to workaround
 160  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
 161  */
 162 #ifndef arch_task_cache_init
 163 #define arch_task_cache_init()
 164 #endif
 165
 166 void __init fork_init(unsigned long mempages)
 167 {
 168 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 169 #ifndef ARCH_MIN_TASKALIGN
 170 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 171 #endif
 172         /* create a slab on which task_structs can be allocated */
 173         task_struct_cachep =
 174                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 175                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL);
 176 #endif
 177
 178         /* do the arch specific task caches init */
 179         arch_task_cache_init();
 180
 181         /*
 182          * The default maximum number of threads is set to a safe
 183          * value: the thread structures can take up at most half
 184          * of memory.
 185          */
 186         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 187
 188         /*
 189          * we need to allow at least 20 threads to boot a system
 190          */
 191         if(max_threads < 20)
 192                 max_threads = 20;
 193
 194         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 195         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 196         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 197                 init_task.signal->rlim[RLIMIT_NPROC];
 198 }
 199
 200 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
 201                                                struct task_struct *src)
 202 {
 203         *dst = *src;
 204         return 0;
 205 }
 206
 207 static struct task_struct *dup_task_struct(struct task_struct *orig)
 208 {
 209         struct task_struct *tsk;
 210         struct thread_info *ti;
 211         int err;
 212
 213         prepare_to_copy(orig);
 214
 215         tsk = alloc_task_struct();
 216         if (!tsk)
 217                 return NULL;
 218
 219         ti = alloc_thread_info(tsk);
 220         if (!ti) {
 221                 free_task_struct(tsk);
 222                 return NULL;
 223         }
 224
 225         err = arch_dup_task_struct(tsk, orig);
 226         if (err)
 227                 goto out;
 228
 229         tsk->stack = ti;
 230
 231         err = prop_local_init_single(&tsk->dirties);
 232         if (err)
 233                 goto out;
 234
 235         setup_thread_stack(tsk, orig);
 236
 237 #ifdef CONFIG_CC_STACKPROTECTOR
 238         tsk->stack_canary = get_random_int();
 239 #endif
 240
 241         /* One for us, one for whoever does the "release_task()" (usually parent) */
 242         atomic_set(&tsk->usage,2);
 243         atomic_set(&tsk->fs_excl, 0);
 244 #ifdef CONFIG_BLK_DEV_IO_TRACE
 245         tsk->btrace_seq = 0;
 246 #endif
 247         tsk->splice_pipe = NULL;
 248         return tsk;
 249
 250 out:
 251         free_thread_info(ti);
 252         free_task_struct(tsk);
 253         return NULL;
 254 }
 255
 256 #ifdef CONFIG_MMU
 257 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 258 {
 259         struct vm_area_struct *mpnt, *tmp, **pprev;
 260         struct rb_node **rb_link, *rb_parent;
 261         int retval;
 262         unsigned long charge;
 263         struct mempolicy *pol;
 264
 265         down_write(&oldmm->mmap_sem);
 266         flush_cache_dup_mm(oldmm);
 267         /*
 268          * Not linked in yet - no deadlock potential:
 269          */
 270         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 271
 272         mm->locked_vm = 0;
 273         mm->mmap = NULL;
 274         mm->mmap_cache = NULL;
 275         mm->free_area_cache = oldmm->mmap_base;
 276         mm->cached_hole_size = ~0UL;
 277         mm->map_count = 0;
 278         cpus_clear(mm->cpu_vm_mask);
 279         mm->mm_rb = RB_ROOT;
 280         rb_link = &mm->mm_rb.rb_node;
 281         rb_parent = NULL;
 282         pprev = &mm->mmap;
 283
 284         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 285                 struct file *file;
 286
 287                 if (mpnt->vm_flags & VM_DONTCOPY) {
 288                         long pages = vma_pages(mpnt);
 289                         mm->total_vm -= pages;
 290                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 291                                                                 -pages);
 292                         continue;
 293                 }
 294                 charge = 0;
 295                 if (mpnt->vm_flags & VM_ACCOUNT) {
 296                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 297                         if (security_vm_enough_memory(len))
 298                                 goto fail_nomem;
 299                         charge = len;
 300                 }
 301                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 302                 if (!tmp)
 303                         goto fail_nomem;
 304                 *tmp = *mpnt;
 305                 pol = mpol_dup(vma_policy(mpnt));
 306                 retval = PTR_ERR(pol);
 307                 if (IS_ERR(pol))
 308                         goto fail_nomem_policy;
 309                 vma_set_policy(tmp, pol);
 310                 tmp->vm_flags &= ~VM_LOCKED;
 311                 tmp->vm_mm = mm;
 312                 tmp->vm_next = NULL;
 313                 anon_vma_link(tmp);
 314                 file = tmp->vm_file;
 315                 if (file) {
 316                         struct inode *inode = file->f_path.dentry->d_inode;
 317                         get_file(file);
 318                         if (tmp->vm_flags & VM_DENYWRITE)
 319                                 atomic_dec(&inode->i_writecount);
 320
 321                         /* insert tmp into the share list, just after mpnt */
 322                         spin_lock(&file->f_mapping->i_mmap_lock);
 323                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
 324                         flush_dcache_mmap_lock(file->f_mapping);
 325                         vma_prio_tree_add(tmp, mpnt);
 326                         flush_dcache_mmap_unlock(file->f_mapping);
 327                         spin_unlock(&file->f_mapping->i_mmap_lock);
 328                 }
 329
 330                 /*
 331                  * Clear hugetlb-related page reserves for children. This only
 332                  * affects MAP_PRIVATE mappings. Faults generated by the child
 333                  * are not guaranteed to succeed, even if read-only
 334                  */
 335                 if (is_vm_hugetlb_page(tmp))
 336                         reset_vma_resv_huge_pages(tmp);
 337
 338                 /*
 339                  * Link in the new vma and copy the page table entries.
 340                  */
 341                 *pprev = tmp;
 342                 pprev = &tmp->vm_next;
 343
 344                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 345                 rb_link = &tmp->vm_rb.rb_right;
 346                 rb_parent = &tmp->vm_rb;
 347
 348                 mm->map_count++;
 349                 retval = copy_page_range(mm, oldmm, mpnt);
 350
 351                 if (tmp->vm_ops && tmp->vm_ops->open)
 352                         tmp->vm_ops->open(tmp);
 353
 354                 if (retval)
 355                         goto out;
 356         }
 357         /* a new mm has just been created */
 358         arch_dup_mmap(oldmm, mm);
 359         retval = 0;
 360 out:
 361         up_write(&mm->mmap_sem);
 362         flush_tlb_mm(oldmm);
 363         up_write(&oldmm->mmap_sem);
 364         return retval;
 365 fail_nomem_policy:
 366         kmem_cache_free(vm_area_cachep, tmp);
 367 fail_nomem:
 368         retval = -ENOMEM;
 369         vm_unacct_memory(charge);
 370         goto out;
 371 }
 372
 373 static inline int mm_alloc_pgd(struct mm_struct * mm)
 374 {
 375         mm->pgd = pgd_alloc(mm);
 376         if (unlikely(!mm->pgd))
 377                 return -ENOMEM;
 378         return 0;
 379 }
 380
 381 static inline void mm_free_pgd(struct mm_struct * mm)
 382 {
 383         pgd_free(mm, mm->pgd);
 384 }
 385 #else
 386 #define dup_mmap(mm, oldmm)     (0)
 387 #define mm_alloc_pgd(mm)        (0)
 388 #define mm_free_pgd(mm)
 389 #endif /* CONFIG_MMU */
 390
 391 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 392
 393 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 394 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 395
 396 #include <linux/init_task.h>
 397
 398 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
 399 {
 400         atomic_set(&mm->mm_users, 1);
 401         atomic_set(&mm->mm_count, 1);
 402         init_rwsem(&mm->mmap_sem);
 403         INIT_LIST_HEAD(&mm->mmlist);
 404         mm->flags = (current->mm) ? current->mm->flags
 405                                   : MMF_DUMP_FILTER_DEFAULT;
 406         mm->core_state = NULL;
 407         mm->nr_ptes = 0;
 408         set_mm_counter(mm, file_rss, 0);
 409         set_mm_counter(mm, anon_rss, 0);
 410         spin_lock_init(&mm->page_table_lock);
 411         rwlock_init(&mm->ioctx_list_lock);
 412         mm->ioctx_list = NULL;
 413         mm->free_area_cache = TASK_UNMAPPED_BASE;
 414         mm->cached_hole_size = ~0UL;
 415         mm_init_owner(mm, p);
 416
 417         if (likely(!mm_alloc_pgd(mm))) {
 418                 mm->def_flags = 0;
 419                 mmu_notifier_mm_init(mm);
 420                 return mm;
 421         }
 422
 423         free_mm(mm);
 424         return NULL;
 425 }
 426
 427 /*
 428  * Allocate and initialize an mm_struct.
 429  */
 430 struct mm_struct * mm_alloc(void)
 431 {
 432         struct mm_struct * mm;
 433
 434         mm = allocate_mm();
 435         if (mm) {
 436                 memset(mm, 0, sizeof(*mm));
 437                 mm = mm_init(mm, current);
 438         }
 439         return mm;
 440 }
 441
 442 /*
 443  * Called when the last reference to the mm
 444  * is dropped: either by a lazy thread or by
 445  * mmput. Free the page directory and the mm.
 446  */
 447 void __mmdrop(struct mm_struct *mm)
 448 {
 449         BUG_ON(mm == &init_mm);
 450         mm_free_pgd(mm);
 451         destroy_context(mm);
 452         mmu_notifier_mm_destroy(mm);
 453         free_mm(mm);
 454 }
 455 EXPORT_SYMBOL_GPL(__mmdrop);
 456
 457 /*
 458  * Decrement the use count and release all resources for an mm.
 459  */
 460 void mmput(struct mm_struct *mm)
 461 {
 462         might_sleep();
 463
 464         if (atomic_dec_and_test(&mm->mm_users)) {
 465                 exit_aio(mm);
 466                 exit_mmap(mm);
 467                 set_mm_exe_file(mm, NULL);
 468                 if (!list_empty(&mm->mmlist)) {
 469                         spin_lock(&mmlist_lock);
 470                         list_del(&mm->mmlist);
 471                         spin_unlock(&mmlist_lock);
 472                 }
 473                 put_swap_token(mm);
 474                 mmdrop(mm);
 475         }
 476 }
 477 EXPORT_SYMBOL_GPL(mmput);
 478
 479 /**
 480  * get_task_mm - acquire a reference to the task's mm
 481  *
 482  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 483  * this kernel workthread has transiently adopted a user mm with use_mm,
 484  * to do its AIO) is not set and if so returns a reference to it, after
 485  * bumping up the use count.  User must release the mm via mmput()
 486  * after use.  Typically used by /proc and ptrace.
 487  */
 488 struct mm_struct *get_task_mm(struct task_struct *task)
 489 {
 490         struct mm_struct *mm;
 491
 492         task_lock(task);
 493         mm = task->mm;
 494         if (mm) {
 495                 if (task->flags & PF_KTHREAD)
 496                         mm = NULL;
 497                 else
 498                         atomic_inc(&mm->mm_users);
 499         }
 500         task_unlock(task);
 501         return mm;
 502 }
 503 EXPORT_SYMBOL_GPL(get_task_mm);
 504
 505 /* Please note the differences between mmput and mm_release.
 506  * mmput is called whenever we stop holding onto a mm_struct,
 507  * error success whatever.
 508  *
 509  * mm_release is called after a mm_struct has been removed
 510  * from the current process.
 511  *
 512  * This difference is important for error handling, when we
 513  * only half set up a mm_struct for a new process and need to restore
 514  * the old one.  Because we mmput the new mm_struct before
 515  * restoring the old one. . .
 516  * Eric Biederman 10 January 1998
 517  */
 518 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 519 {
 520         struct completion *vfork_done = tsk->vfork_done;
 521
 522         /* Get rid of any cached register state */
 523         deactivate_mm(tsk, mm);
 524
 525         /* notify parent sleeping on vfork() */
 526         if (vfork_done) {
 527                 tsk->vfork_done = NULL;
 528                 complete(vfork_done);
 529         }
 530
 531         /*
 532          * If we're exiting normally, clear a user-space tid field if
 533          * requested.  We leave this alone when dying by signal, to leave
 534          * the value intact in a core dump, and to save the unnecessary
 535          * trouble otherwise.  Userland only wants this done for a sys_exit.
 536          */
 537         if (tsk->clear_child_tid
 538             && !(tsk->flags & PF_SIGNALED)
 539             && atomic_read(&mm->mm_users) > 1) {
 540                 u32 __user * tidptr = tsk->clear_child_tid;
 541                 tsk->clear_child_tid = NULL;
 542
 543                 /*
 544                  * We don't check the error code - if userspace has
 545                  * not set up a proper pointer then tough luck.
 546                  */
 547                 put_user(0, tidptr);
 548                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
 549         }
 550 }
 551
 552 /*
 553  * Allocate a new mm structure and copy contents from the
 554  * mm structure of the passed in task structure.
 555  */
 556 struct mm_struct *dup_mm(struct task_struct *tsk)
 557 {
 558         struct mm_struct *mm, *oldmm = current->mm;
 559         int err;
 560
 561         if (!oldmm)
 562                 return NULL;
 563
 564         mm = allocate_mm();
 565         if (!mm)
 566                 goto fail_nomem;
 567
 568         memcpy(mm, oldmm, sizeof(*mm));
 569
 570         /* Initializing for Swap token stuff */
 571         mm->token_priority = 0;
 572         mm->last_interval = 0;
 573
 574         if (!mm_init(mm, tsk))
 575                 goto fail_nomem;
 576
 577         if (init_new_context(tsk, mm))
 578                 goto fail_nocontext;
 579
 580         dup_mm_exe_file(oldmm, mm);
 581
 582         err = dup_mmap(mm, oldmm);
 583         if (err)
 584                 goto free_pt;
 585
 586         mm->hiwater_rss = get_mm_rss(mm);
 587         mm->hiwater_vm = mm->total_vm;
 588
 589         return mm;
 590
 591 free_pt:
 592         mmput(mm);
 593
 594 fail_nomem:
 595         return NULL;
 596
 597 fail_nocontext:
 598         /*
 599          * If init_new_context() failed, we cannot use mmput() to free the mm
 600          * because it calls destroy_context()
 601          */
 602         mm_free_pgd(mm);
 603         free_mm(mm);
 604         return NULL;
 605 }
 606
 607 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 608 {
 609         struct mm_struct * mm, *oldmm;
 610         int retval;
 611
 612         tsk->min_flt = tsk->maj_flt = 0;
 613         tsk->nvcsw = tsk->nivcsw = 0;
 614
 615         tsk->mm = NULL;
 616         tsk->active_mm = NULL;
 617
 618         /*
 619          * Are we cloning a kernel thread?
 620          *
 621          * We need to steal a active VM for that..
 622          */
 623         oldmm = current->mm;
 624         if (!oldmm)
 625                 return 0;
 626
 627         if (clone_flags & CLONE_VM) {
 628                 atomic_inc(&oldmm->mm_users);
 629                 mm = oldmm;
 630                 goto good_mm;
 631         }
 632
 633         retval = -ENOMEM;
 634         mm = dup_mm(tsk);
 635         if (!mm)
 636                 goto fail_nomem;
 637
 638 good_mm:
 639         /* Initializing for Swap token stuff */
 640         mm->token_priority = 0;
 641         mm->last_interval = 0;
 642
 643         tsk->mm = mm;
 644         tsk->active_mm = mm;
 645         return 0;
 646
 647 fail_nomem:
 648         return retval;
 649 }
 650
 651 static struct fs_struct *__copy_fs_struct(struct fs_struct *old)
 652 {
 653         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
 654         /* We don't need to lock fs - think why ;-) */
 655         if (fs) {
 656                 atomic_set(&fs->count, 1);
 657                 rwlock_init(&fs->lock);
 658                 fs->umask = old->umask;
 659                 read_lock(&old->lock);
 660                 fs->root = old->root;
 661                 path_get(&old->root);
 662                 fs->pwd = old->pwd;
 663                 path_get(&old->pwd);
 664                 read_unlock(&old->lock);
 665         }
 666         return fs;
 667 }
 668
 669 struct fs_struct *copy_fs_struct(struct fs_struct *old)
 670 {
 671         return __copy_fs_struct(old);
 672 }
 673
 674 EXPORT_SYMBOL_GPL(copy_fs_struct);
 675
 676 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 677 {
 678         if (clone_flags & CLONE_FS) {
 679                 atomic_inc(&current->fs->count);
 680                 return 0;
 681         }
 682         tsk->fs = __copy_fs_struct(current->fs);
 683         if (!tsk->fs)
 684                 return -ENOMEM;
 685         return 0;
 686 }
 687
 688 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 689 {
 690         struct files_struct *oldf, *newf;
 691         int error = 0;
 692
 693         /*
 694          * A background process may not have any files ...
 695          */
 696         oldf = current->files;
 697         if (!oldf)
 698                 goto out;
 699
 700         if (clone_flags & CLONE_FILES) {
 701                 atomic_inc(&oldf->count);
 702                 goto out;
 703         }
 704
 705         newf = dup_fd(oldf, &error);
 706         if (!newf)
 707                 goto out;
 708
 709         tsk->files = newf;
 710         error = 0;
 711 out:
 712         return error;
 713 }
 714
 715 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 716 {
 717 #ifdef CONFIG_BLOCK
 718         struct io_context *ioc = current->io_context;
 719
 720         if (!ioc)
 721                 return 0;
 722         /*
 723          * Share io context with parent, if CLONE_IO is set
 724          */
 725         if (clone_flags & CLONE_IO) {
 726                 tsk->io_context = ioc_task_link(ioc);
 727                 if (unlikely(!tsk->io_context))
 728                         return -ENOMEM;
 729         } else if (ioprio_valid(ioc->ioprio)) {
 730                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
 731                 if (unlikely(!tsk->io_context))
 732                         return -ENOMEM;
 733
 734                 tsk->io_context->ioprio = ioc->ioprio;
 735         }
 736 #endif
 737         return 0;
 738 }
 739
 740 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 741 {
 742         struct sighand_struct *sig;
 743
 744         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
 745                 atomic_inc(&current->sighand->count);
 746                 return 0;
 747         }
 748         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 749         rcu_assign_pointer(tsk->sighand, sig);
 750         if (!sig)
 751                 return -ENOMEM;
 752         atomic_set(&sig->count, 1);
 753         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 754         return 0;
 755 }
 756
 757 void __cleanup_sighand(struct sighand_struct *sighand)
 758 {
 759         if (atomic_dec_and_test(&sighand->count))
 760                 kmem_cache_free(sighand_cachep, sighand);
 761 }
 762
 763 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
 764 {
 765         struct signal_struct *sig;
 766         int ret;
 767
 768         if (clone_flags & CLONE_THREAD) {
 769                 atomic_inc(&current->signal->count);
 770                 atomic_inc(&current->signal->live);
 771                 return 0;
 772         }
 773         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 774         tsk->signal = sig;
 775         if (!sig)
 776                 return -ENOMEM;
 777
 778         ret = copy_thread_group_keys(tsk);
 779         if (ret < 0) {
 780                 kmem_cache_free(signal_cachep, sig);
 781                 return ret;
 782         }
 783
 784         atomic_set(&sig->count, 1);
 785         atomic_set(&sig->live, 1);
 786         init_waitqueue_head(&sig->wait_chldexit);
 787         sig->flags = 0;
 788         sig->group_exit_code = 0;
 789         sig->group_exit_task = NULL;
 790         sig->group_stop_count = 0;
 791         sig->curr_target = tsk;
 792         init_sigpending(&sig->shared_pending);
 793         INIT_LIST_HEAD(&sig->posix_timers);
 794
 795         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 796         sig->it_real_incr.tv64 = 0;
 797         sig->real_timer.function = it_real_fn;
 798
 799         sig->it_virt_expires = cputime_zero;
 800         sig->it_virt_incr = cputime_zero;
 801         sig->it_prof_expires = cputime_zero;
 802         sig->it_prof_incr = cputime_zero;
 803
 804         sig->leader = 0;        /* session leadership doesn't inherit */
 805         sig->tty_old_pgrp = NULL;
 806         sig->tty = NULL;
 807
 808         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
 809         sig->gtime = cputime_zero;
 810         sig->cgtime = cputime_zero;
 811         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
 812         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
 813         sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
 814         task_io_accounting_init(&sig->ioac);
 815         sig->sum_sched_runtime = 0;
 816         INIT_LIST_HEAD(&sig->cpu_timers[0]);
 817         INIT_LIST_HEAD(&sig->cpu_timers[1]);
 818         INIT_LIST_HEAD(&sig->cpu_timers[2]);
 819         taskstats_tgid_init(sig);
 820
 821         task_lock(current->group_leader);
 822         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
 823         task_unlock(current->group_leader);
 824
 825         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
 826                 /*
 827                  * New sole thread in the process gets an expiry time
 828                  * of the whole CPU time limit.
 829                  */
 830                 tsk->it_prof_expires =
 831                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
 832         }
 833         acct_init_pacct(&sig->pacct);
 834
 835         tty_audit_fork(sig);
 836
 837         return 0;
 838 }
 839
 840 void __cleanup_signal(struct signal_struct *sig)
 841 {
 842         exit_thread_group_keys(sig);
 843         tty_kref_put(sig->tty);
 844         kmem_cache_free(signal_cachep, sig);
 845 }
 846
 847 static void cleanup_signal(struct task_struct *tsk)
 848 {
 849         struct signal_struct *sig = tsk->signal;
 850
 851         atomic_dec(&sig->live);
 852
 853         if (atomic_dec_and_test(&sig->count))
 854                 __cleanup_signal(sig);
 855 }
 856
 857 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
 858 {
 859         unsigned long new_flags = p->flags;
 860
 861         new_flags &= ~PF_SUPERPRIV;
 862         new_flags |= PF_FORKNOEXEC;
 863         new_flags |= PF_STARTING;
 864         p->flags = new_flags;
 865         clear_freeze_flag(p);
 866 }
 867
 868 asmlinkage long sys_set_tid_address(int __user *tidptr)
 869 {
 870         current->clear_child_tid = tidptr;
 871
 872         return task_pid_vnr(current);
 873 }
 874
 875 static void rt_mutex_init_task(struct task_struct *p)
 876 {
 877         spin_lock_init(&p->pi_lock);
 878 #ifdef CONFIG_RT_MUTEXES
 879         plist_head_init(&p->pi_waiters, &p->pi_lock);
 880         p->pi_blocked_on = NULL;
 881 #endif
 882 }
 883
 884 #ifdef CONFIG_MM_OWNER
 885 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 886 {
 887         mm->owner = p;
 888 }
 889 #endif /* CONFIG_MM_OWNER */
 890
 891 /*
 892  * This creates a new process as a copy of the old one,
 893  * but does not actually start it yet.
 894  *
 895  * It copies the registers, and all the appropriate
 896  * parts of the process environment (as per the clone
 897  * flags). The actual kick-off is left to the caller.
 898  */
 899 static struct task_struct *copy_process(unsigned long clone_flags,
 900                                         unsigned long stack_start,
 901                                         struct pt_regs *regs,
 902                                         unsigned long stack_size,
 903                                         int __user *child_tidptr,
 904                                         struct pid *pid,
 905                                         int trace)
 906 {
 907         int retval;
 908         struct task_struct *p;
 909         int cgroup_callbacks_done = 0;
 910
 911         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 912                 return ERR_PTR(-EINVAL);
 913
 914         /*
 915          * Thread groups must share signals as well, and detached threads
 916          * can only be started up within the thread group.
 917          */
 918         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
 919                 return ERR_PTR(-EINVAL);
 920
 921         /*
 922          * Shared signal handlers imply shared VM. By way of the above,
 923          * thread groups also imply shared VM. Blocking this case allows
 924          * for various simplifications in other code.
 925          */
 926         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
 927                 return ERR_PTR(-EINVAL);
 928
 929         retval = security_task_create(clone_flags);
 930         if (retval)
 931                 goto fork_out;
 932
 933         retval = -ENOMEM;
 934         p = dup_task_struct(current);
 935         if (!p)
 936                 goto fork_out;
 937
 938         rt_mutex_init_task(p);
 939
 940 #ifdef CONFIG_PROVE_LOCKING
 941         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
 942         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
 943 #endif
 944         retval = -EAGAIN;
 945         if (atomic_read(&p->user->processes) >=
 946                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
 947                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
 948                     p->user != current->nsproxy->user_ns->root_user)
 949                         goto bad_fork_free;
 950         }
 951
 952         atomic_inc(&p->user->__count);
 953         atomic_inc(&p->user->processes);
 954         get_group_info(p->group_info);
 955
 956         /*
 957          * If multiple threads are within copy_process(), then this check
 958          * triggers too late. This doesn't hurt, the check is only there
 959          * to stop root fork bombs.
 960          */
 961         if (nr_threads >= max_threads)
 962                 goto bad_fork_cleanup_count;
 963
 964         if (!try_module_get(task_thread_info(p)->exec_domain->module))
 965                 goto bad_fork_cleanup_count;
 966
 967         if (p->binfmt && !try_module_get(p->binfmt->module))
 968                 goto bad_fork_cleanup_put_domain;
 969
 970         p->did_exec = 0;
 971         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
 972         copy_flags(clone_flags, p);
 973         INIT_LIST_HEAD(&p->children);
 974         INIT_LIST_HEAD(&p->sibling);
 975 #ifdef CONFIG_PREEMPT_RCU
 976         p->rcu_read_lock_nesting = 0;
 977         p->rcu_flipctr_idx = 0;
 978 #endif /* #ifdef CONFIG_PREEMPT_RCU */
 979         p->vfork_done = NULL;
 980         spin_lock_init(&p->alloc_lock);
 981
 982         clear_tsk_thread_flag(p, TIF_SIGPENDING);
 983         init_sigpending(&p->pending);
 984
 985         p->utime = cputime_zero;
 986         p->stime = cputime_zero;
 987         p->gtime = cputime_zero;
 988         p->utimescaled = cputime_zero;
 989         p->stimescaled = cputime_zero;
 990         p->prev_utime = cputime_zero;
 991         p->prev_stime = cputime_zero;
 992
 993 #ifdef CONFIG_DETECT_SOFTLOCKUP
 994         p->last_switch_count = 0;
 995         p->last_switch_timestamp = 0;
 996 #endif
 997
 998         task_io_accounting_init(&p->ioac);
 999         acct_clear_integrals(p);
1000
1001         p->it_virt_expires = cputime_zero;
1002         p->it_prof_expires = cputime_zero;
1003         p->it_sched_expires = 0;
1004         INIT_LIST_HEAD(&p->cpu_timers[0]);
1005         INIT_LIST_HEAD(&p->cpu_timers[1]);
1006         INIT_LIST_HEAD(&p->cpu_timers[2]);
1007
1008         p->lock_depth = -1;             /* -1 = no lock */
1009         do_posix_clock_monotonic_gettime(&p->start_time);
1010         p->real_start_time = p->start_time;
1011         monotonic_to_bootbased(&p->real_start_time);
1012 #ifdef CONFIG_SECURITY
1013         p->security = NULL;
1014 #endif
1015         p->cap_bset = current->cap_bset;
1016         p->io_context = NULL;
1017         p->audit_context = NULL;
1018         cgroup_fork(p);
1019 #ifdef CONFIG_NUMA
1020         p->mempolicy = mpol_dup(p->mempolicy);
1021         if (IS_ERR(p->mempolicy)) {
1022                 retval = PTR_ERR(p->mempolicy);
1023                 p->mempolicy = NULL;
1024                 goto bad_fork_cleanup_cgroup;
1025         }
1026         mpol_fix_fork_child_flag(p);
1027 #endif
1028 #ifdef CONFIG_TRACE_IRQFLAGS
1029         p->irq_events = 0;
1030 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1031         p->hardirqs_enabled = 1;
1032 #else
1033         p->hardirqs_enabled = 0;
1034 #endif
1035         p->hardirq_enable_ip = 0;
1036         p->hardirq_enable_event = 0;
1037         p->hardirq_disable_ip = _THIS_IP_;
1038         p->hardirq_disable_event = 0;
1039         p->softirqs_enabled = 1;
1040         p->softirq_enable_ip = _THIS_IP_;
1041         p->softirq_enable_event = 0;
1042         p->softirq_disable_ip = 0;
1043         p->softirq_disable_event = 0;
1044         p->hardirq_context = 0;
1045         p->softirq_context = 0;
1046 #endif
1047 #ifdef CONFIG_LOCKDEP
1048         p->lockdep_depth = 0; /* no locks held yet */
1049         p->curr_chain_key = 0;
1050         p->lockdep_recursion = 0;
1051 #endif
1052
1053 #ifdef CONFIG_DEBUG_MUTEXES
1054         p->blocked_on = NULL; /* not blocked yet */
1055 #endif
1056
1057         /* Perform scheduler related setup. Assign this task to a CPU. */
1058         sched_fork(p, clone_flags);
1059
1060         if ((retval = security_task_alloc(p)))
1061                 goto bad_fork_cleanup_policy;
1062         if ((retval = audit_alloc(p)))
1063                 goto bad_fork_cleanup_security;
1064         /* copy all the process information */
1065         if ((retval = copy_semundo(clone_flags, p)))
1066                 goto bad_fork_cleanup_audit;
1067         if ((retval = copy_files(clone_flags, p)))
1068                 goto bad_fork_cleanup_semundo;
1069         if ((retval = copy_fs(clone_flags, p)))
1070                 goto bad_fork_cleanup_files;
1071         if ((retval = copy_sighand(clone_flags, p)))
1072                 goto bad_fork_cleanup_fs;
1073         if ((retval = copy_signal(clone_flags, p)))
1074                 goto bad_fork_cleanup_sighand;
1075         if ((retval = copy_mm(clone_flags, p)))
1076                 goto bad_fork_cleanup_signal;
1077         if ((retval = copy_keys(clone_flags, p)))
1078                 goto bad_fork_cleanup_mm;
1079         if ((retval = copy_namespaces(clone_flags, p)))
1080                 goto bad_fork_cleanup_keys;
1081         if ((retval = copy_io(clone_flags, p)))
1082                 goto bad_fork_cleanup_namespaces;
1083         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1084         if (retval)
1085                 goto bad_fork_cleanup_io;
1086
1087         if (pid != &init_struct_pid) {
1088                 retval = -ENOMEM;
1089                 pid = alloc_pid(task_active_pid_ns(p));
1090                 if (!pid)
1091                         goto bad_fork_cleanup_io;
1092
1093                 if (clone_flags & CLONE_NEWPID) {
1094                         retval = pid_ns_prepare_proc(task_active_pid_ns(p));
1095                         if (retval < 0)
1096                                 goto bad_fork_free_pid;
1097                 }
1098         }
1099
1100         p->pid = pid_nr(pid);
1101         p->tgid = p->pid;
1102         if (clone_flags & CLONE_THREAD)
1103                 p->tgid = current->tgid;
1104
1105         if (current->nsproxy != p->nsproxy) {
1106                 retval = ns_cgroup_clone(p, pid);
1107                 if (retval)
1108                         goto bad_fork_free_pid;
1109         }
1110
1111         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1112         /*
1113          * Clear TID on mm_release()?
1114          */
1115         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1116 #ifdef CONFIG_FUTEX
1117         p->robust_list = NULL;
1118 #ifdef CONFIG_COMPAT
1119         p->compat_robust_list = NULL;
1120 #endif
1121         INIT_LIST_HEAD(&p->pi_state_list);
1122         p->pi_state_cache = NULL;
1123 #endif
1124         /*
1125          * sigaltstack should be cleared when sharing the same VM
1126          */
1127         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1128                 p->sas_ss_sp = p->sas_ss_size = 0;
1129
1130         /*
1131          * Syscall tracing should be turned off in the child regardless
1132          * of CLONE_PTRACE.
1133          */
1134         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1135 #ifdef TIF_SYSCALL_EMU
1136         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1137 #endif
1138         clear_all_latency_tracing(p);
1139
1140         /* Our parent execution domain becomes current domain
1141            These must match for thread signalling to apply */
1142         p->parent_exec_id = p->self_exec_id;
1143
1144         /* ok, now we should be set up.. */
1145         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1146         p->pdeath_signal = 0;
1147         p->exit_state = 0;
1148
1149         /*
1150          * Ok, make it visible to the rest of the system.
1151          * We dont wake it up yet.
1152          */
1153         p->group_leader = p;
1154         INIT_LIST_HEAD(&p->thread_group);
1155
1156         /* Now that the task is set up, run cgroup callbacks if
1157          * necessary. We need to run them before the task is visible
1158          * on the tasklist. */
1159         cgroup_fork_callbacks(p);
1160         cgroup_callbacks_done = 1;
1161
1162         /* Need tasklist lock for parent etc handling! */
1163         write_lock_irq(&tasklist_lock);
1164
1165         /*
1166          * The task hasn't been attached yet, so its cpus_allowed mask will
1167          * not be changed, nor will its assigned CPU.
1168          *
1169          * The cpus_allowed mask of the parent may have changed after it was
1170          * copied first time - so re-copy it here, then check the child's CPU
1171          * to ensure it is on a valid CPU (and if not, just force it back to
1172          * parent's CPU). This avoids alot of nasty races.
1173          */
1174         p->cpus_allowed = current->cpus_allowed;
1175         p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1176         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1177                         !cpu_online(task_cpu(p))))
1178                 set_task_cpu(p, smp_processor_id());
1179
1180         /* CLONE_PARENT re-uses the old parent */
1181         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1182                 p->real_parent = current->real_parent;
1183         else
1184                 p->real_parent = current;
1185
1186         spin_lock(&current->sighand->siglock);
1187
1188         /*
1189          * Process group and session signals need to be delivered to just the
1190          * parent before the fork or both the parent and the child after the
1191          * fork. Restart if a signal comes in before we add the new process to
1192          * it's process group.
1193          * A fatal signal pending means that current will exit, so the new
1194          * thread can't slip out of an OOM kill (or normal SIGKILL).
1195          */
1196         recalc_sigpending();
1197         if (signal_pending(current)) {
1198                 spin_unlock(&current->sighand->siglock);
1199                 write_unlock_irq(&tasklist_lock);
1200                 retval = -ERESTARTNOINTR;
1201                 goto bad_fork_free_pid;
1202         }
1203
1204         if (clone_flags & CLONE_THREAD) {
1205                 p->group_leader = current->group_leader;
1206                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1207
1208                 if (!cputime_eq(current->signal->it_virt_expires,
1209                                 cputime_zero) ||
1210                     !cputime_eq(current->signal->it_prof_expires,
1211                                 cputime_zero) ||
1212                     current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1213                     !list_empty(&current->signal->cpu_timers[0]) ||
1214                     !list_empty(&current->signal->cpu_timers[1]) ||
1215                     !list_empty(&current->signal->cpu_timers[2])) {
1216                         /*
1217                          * Have child wake up on its first tick to check
1218                          * for process CPU timers.
1219                          */
1220                         p->it_prof_expires = jiffies_to_cputime(1);
1221                 }
1222         }
1223
1224         if (likely(p->pid)) {
1225                 list_add_tail(&p->sibling, &p->real_parent->children);
1226                 tracehook_finish_clone(p, clone_flags, trace);
1227
1228                 if (thread_group_leader(p)) {
1229                         if (clone_flags & CLONE_NEWPID)
1230                                 p->nsproxy->pid_ns->child_reaper = p;
1231
1232                         p->signal->leader_pid = pid;
1233                         tty_kref_put(p->signal->tty);
1234                         p->signal->tty = tty_kref_get(current->signal->tty);
1235                         set_task_pgrp(p, task_pgrp_nr(current));
1236                         set_task_session(p, task_session_nr(current));
1237                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1238                         attach_pid(p, PIDTYPE_SID, task_session(current));
1239                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1240                         __get_cpu_var(process_counts)++;
1241                 }
1242                 attach_pid(p, PIDTYPE_PID, pid);
1243                 nr_threads++;
1244         }
1245
1246         total_forks++;
1247         spin_unlock(&current->sighand->siglock);
1248         write_unlock_irq(&tasklist_lock);
1249         proc_fork_connector(p);
1250         cgroup_post_fork(p);
1251         return p;
1252
1253 bad_fork_free_pid:
1254         if (pid != &init_struct_pid)
1255                 free_pid(pid);
1256 bad_fork_cleanup_io:
1257         put_io_context(p->io_context);
1258 bad_fork_cleanup_namespaces:
1259         exit_task_namespaces(p);
1260 bad_fork_cleanup_keys:
1261         exit_keys(p);
1262 bad_fork_cleanup_mm:
1263         if (p->mm)
1264                 mmput(p->mm);
1265 bad_fork_cleanup_signal:
1266         cleanup_signal(p);
1267 bad_fork_cleanup_sighand:
1268         __cleanup_sighand(p->sighand);
1269 bad_fork_cleanup_fs:
1270         exit_fs(p); /* blocking */
1271 bad_fork_cleanup_files:
1272         exit_files(p); /* blocking */
1273 bad_fork_cleanup_semundo:
1274         exit_sem(p);
1275 bad_fork_cleanup_audit:
1276         audit_free(p);
1277 bad_fork_cleanup_security:
1278         security_task_free(p);
1279 bad_fork_cleanup_policy:
1280 #ifdef CONFIG_NUMA
1281         mpol_put(p->mempolicy);
1282 bad_fork_cleanup_cgroup:
1283 #endif
1284         cgroup_exit(p, cgroup_callbacks_done);
1285         delayacct_tsk_free(p);
1286         if (p->binfmt)
1287                 module_put(p->binfmt->module);
1288 bad_fork_cleanup_put_domain:
1289         module_put(task_thread_info(p)->exec_domain->module);
1290 bad_fork_cleanup_count:
1291         put_group_info(p->group_info);
1292         atomic_dec(&p->user->processes);
1293         free_uid(p->user);
1294 bad_fork_free:
1295         free_task(p);
1296 fork_out:
1297         return ERR_PTR(retval);
1298 }
1299
1300 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1301 {
1302         memset(regs, 0, sizeof(struct pt_regs));
1303         return regs;
1304 }
1305
1306 struct task_struct * __cpuinit fork_idle(int cpu)
1307 {
1308         struct task_struct *task;
1309         struct pt_regs regs;
1310
1311         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1312                             &init_struct_pid, 0);
1313         if (!IS_ERR(task))
1314                 init_idle(task, cpu);
1315
1316         return task;
1317 }
1318
1319 /*
1320  *  Ok, this is the main fork-routine.
1321  *
1322  * It copies the process, and if successful kick-starts
1323  * it and waits for it to finish using the VM if required.
1324  */
1325 long do_fork(unsigned long clone_flags,
1326               unsigned long stack_start,
1327               struct pt_regs *regs,
1328               unsigned long stack_size,
1329               int __user *parent_tidptr,
1330               int __user *child_tidptr)
1331 {
1332         struct task_struct *p;
1333         int trace = 0;
1334         long nr;
1335
1336         /*
1337          * We hope to recycle these flags after 2.6.26
1338          */
1339         if (unlikely(clone_flags & CLONE_STOPPED)) {
1340                 static int __read_mostly count = 100;
1341
1342                 if (count > 0 && printk_ratelimit()) {
1343                         char comm[TASK_COMM_LEN];
1344
1345                         count--;
1346                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1347                                         "clone flags 0x%lx\n",
1348                                 get_task_comm(comm, current),
1349                                 clone_flags & CLONE_STOPPED);
1350                 }
1351         }
1352
1353         /*
1354          * When called from kernel_thread, don't do user tracing stuff.
1355          */
1356         if (likely(user_mode(regs)))
1357                 trace = tracehook_prepare_clone(clone_flags);
1358
1359         p = copy_process(clone_flags, stack_start, regs, stack_size,
1360                          child_tidptr, NULL, trace);
1361         /*
1362          * Do this prior waking up the new thread - the thread pointer
1363          * might get invalid after that point, if the thread exits quickly.
1364          */
1365         if (!IS_ERR(p)) {
1366                 struct completion vfork;
1367
1368                 trace_sched_process_fork(current, p);
1369
1370                 nr = task_pid_vnr(p);
1371
1372                 if (clone_flags & CLONE_PARENT_SETTID)
1373                         put_user(nr, parent_tidptr);
1374
1375                 if (clone_flags & CLONE_VFORK) {
1376                         p->vfork_done = &vfork;
1377                         init_completion(&vfork);
1378                 }
1379
1380                 tracehook_report_clone(trace, regs, clone_flags, nr, p);
1381
1382                 /*
1383                  * We set PF_STARTING at creation in case tracing wants to
1384                  * use this to distinguish a fully live task from one that
1385                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1386                  * clear it and set the child going.
1387                  */
1388                 p->flags &= ~PF_STARTING;
1389
1390                 if (unlikely(clone_flags & CLONE_STOPPED)) {
1391                         /*
1392                          * We'll start up with an immediate SIGSTOP.
1393                          */
1394                         sigaddset(&p->pending.signal, SIGSTOP);
1395                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1396                         __set_task_state(p, TASK_STOPPED);
1397                 } else {
1398                         wake_up_new_task(p, clone_flags);
1399                 }
1400
1401                 tracehook_report_clone_complete(trace, regs,
1402                                                 clone_flags, nr, p);
1403
1404                 if (clone_flags & CLONE_VFORK) {
1405                         freezer_do_not_count();
1406                         wait_for_completion(&vfork);
1407                         freezer_count();
1408                         tracehook_report_vfork_done(p, nr);
1409                 }
1410         } else {
1411                 nr = PTR_ERR(p);
1412         }
1413         return nr;
1414 }
1415
1416 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1417 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1418 #endif
1419
1420 static void sighand_ctor(void *data)
1421 {
1422         struct sighand_struct *sighand = data;
1423
1424         spin_lock_init(&sighand->siglock);
1425         init_waitqueue_head(&sighand->signalfd_wqh);
1426 }
1427
1428 void __init proc_caches_init(void)
1429 {
1430         sighand_cachep = kmem_cache_create("sighand_cache",
1431                         sizeof(struct sighand_struct), 0,
1432                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1433                         sighand_ctor);
1434         signal_cachep = kmem_cache_create("signal_cache",
1435                         sizeof(struct signal_struct), 0,
1436                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1437         files_cachep = kmem_cache_create("files_cache",
1438                         sizeof(struct files_struct), 0,
1439                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1440         fs_cachep = kmem_cache_create("fs_cache",
1441                         sizeof(struct fs_struct), 0,
1442                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1443         vm_area_cachep = kmem_cache_create("vm_area_struct",
1444                         sizeof(struct vm_area_struct), 0,
1445                         SLAB_PANIC, NULL);
1446         mm_cachep = kmem_cache_create("mm_struct",
1447                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1448                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1449 }
1450
1451 /*
1452  * Check constraints on flags passed to the unshare system call and
1453  * force unsharing of additional process context as appropriate.
1454  */
1455 static void check_unshare_flags(unsigned long *flags_ptr)
1456 {
1457         /*
1458          * If unsharing a thread from a thread group, must also
1459          * unshare vm.
1460          */
1461         if (*flags_ptr & CLONE_THREAD)
1462                 *flags_ptr |= CLONE_VM;
1463
1464         /*
1465          * If unsharing vm, must also unshare signal handlers.
1466          */
1467         if (*flags_ptr & CLONE_VM)
1468                 *flags_ptr |= CLONE_SIGHAND;
1469
1470         /*
1471          * If unsharing signal handlers and the task was created
1472          * using CLONE_THREAD, then must unshare the thread
1473          */
1474         if ((*flags_ptr & CLONE_SIGHAND) &&
1475             (atomic_read(&current->signal->count) > 1))
1476                 *flags_ptr |= CLONE_THREAD;
1477
1478         /*
1479          * If unsharing namespace, must also unshare filesystem information.
1480          */
1481         if (*flags_ptr & CLONE_NEWNS)
1482                 *flags_ptr |= CLONE_FS;
1483 }
1484
1485 /*
1486  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1487  */
1488 static int unshare_thread(unsigned long unshare_flags)
1489 {
1490         if (unshare_flags & CLONE_THREAD)
1491                 return -EINVAL;
1492
1493         return 0;
1494 }
1495
1496 /*
1497  * Unshare the filesystem structure if it is being shared
1498  */
1499 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1500 {
1501         struct fs_struct *fs = current->fs;
1502
1503         if ((unshare_flags & CLONE_FS) &&
1504             (fs && atomic_read(&fs->count) > 1)) {
1505                 *new_fsp = __copy_fs_struct(current->fs);
1506                 if (!*new_fsp)
1507                         return -ENOMEM;
1508         }
1509
1510         return 0;
1511 }
1512
1513 /*
1514  * Unsharing of sighand is not supported yet
1515  */
1516 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1517 {
1518         struct sighand_struct *sigh = current->sighand;
1519
1520         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1521                 return -EINVAL;
1522         else
1523                 return 0;
1524 }
1525
1526 /*
1527  * Unshare vm if it is being shared
1528  */
1529 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1530 {
1531         struct mm_struct *mm = current->mm;
1532
1533         if ((unshare_flags & CLONE_VM) &&
1534             (mm && atomic_read(&mm->mm_users) > 1)) {
1535                 return -EINVAL;
1536         }
1537
1538         return 0;
1539 }
1540
1541 /*
1542  * Unshare file descriptor table if it is being shared
1543  */
1544 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1545 {
1546         struct files_struct *fd = current->files;
1547         int error = 0;
1548
1549         if ((unshare_flags & CLONE_FILES) &&
1550             (fd && atomic_read(&fd->count) > 1)) {
1551                 *new_fdp = dup_fd(fd, &error);
1552                 if (!*new_fdp)
1553                         return error;
1554         }
1555
1556         return 0;
1557 }
1558
1559 /*
1560  * unshare allows a process to 'unshare' part of the process
1561  * context which was originally shared using clone.  copy_*
1562  * functions used by do_fork() cannot be used here directly
1563  * because they modify an inactive task_struct that is being
1564  * constructed. Here we are modifying the current, active,
1565  * task_struct.
1566  */
1567 asmlinkage long sys_unshare(unsigned long unshare_flags)
1568 {
1569         int err = 0;
1570         struct fs_struct *fs, *new_fs = NULL;
1571         struct sighand_struct *new_sigh = NULL;
1572         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1573         struct files_struct *fd, *new_fd = NULL;
1574         struct nsproxy *new_nsproxy = NULL;
1575         int do_sysvsem = 0;
1576
1577         check_unshare_flags(&unshare_flags);
1578
1579         /* Return -EINVAL for all unsupported flags */
1580         err = -EINVAL;
1581         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1582                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1583                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWUSER|
1584                                 CLONE_NEWNET))
1585                 goto bad_unshare_out;
1586
1587         /*
1588          * CLONE_NEWIPC must also detach from the undolist: after switching
1589          * to a new ipc namespace, the semaphore arrays from the old
1590          * namespace are unreachable.
1591          */
1592         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1593                 do_sysvsem = 1;
1594         if ((err = unshare_thread(unshare_flags)))
1595                 goto bad_unshare_out;
1596         if ((err = unshare_fs(unshare_flags, &new_fs)))
1597                 goto bad_unshare_cleanup_thread;
1598         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1599                 goto bad_unshare_cleanup_fs;
1600         if ((err = unshare_vm(unshare_flags, &new_mm)))
1601                 goto bad_unshare_cleanup_sigh;
1602         if ((err = unshare_fd(unshare_flags, &new_fd)))
1603                 goto bad_unshare_cleanup_vm;
1604         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1605                         new_fs)))
1606                 goto bad_unshare_cleanup_fd;
1607
1608         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1609                 if (do_sysvsem) {
1610                         /*
1611                          * CLONE_SYSVSEM is equivalent to sys_exit().
1612                          */
1613                         exit_sem(current);
1614                 }
1615
1616                 if (new_nsproxy) {
1617                         switch_task_namespaces(current, new_nsproxy);
1618                         new_nsproxy = NULL;
1619                 }
1620
1621                 task_lock(current);
1622
1623                 if (new_fs) {
1624                         fs = current->fs;
1625                         current->fs = new_fs;
1626                         new_fs = fs;
1627                 }
1628
1629                 if (new_mm) {
1630                         mm = current->mm;
1631                         active_mm = current->active_mm;
1632                         current->mm = new_mm;
1633                         current->active_mm = new_mm;
1634                         activate_mm(active_mm, new_mm);
1635                         new_mm = mm;
1636                 }
1637
1638                 if (new_fd) {
1639                         fd = current->files;
1640                         current->files = new_fd;
1641                         new_fd = fd;
1642                 }
1643
1644                 task_unlock(current);
1645         }
1646
1647         if (new_nsproxy)
1648                 put_nsproxy(new_nsproxy);
1649
1650 bad_unshare_cleanup_fd:
1651         if (new_fd)
1652                 put_files_struct(new_fd);
1653
1654 bad_unshare_cleanup_vm:
1655         if (new_mm)
1656                 mmput(new_mm);
1657
1658 bad_unshare_cleanup_sigh:
1659         if (new_sigh)
1660                 if (atomic_dec_and_test(&new_sigh->count))
1661                         kmem_cache_free(sighand_cachep, new_sigh);
1662
1663 bad_unshare_cleanup_fs:
1664         if (new_fs)
1665                 put_fs_struct(new_fs);
1666
1667 bad_unshare_cleanup_thread:
1668 bad_unshare_out:
1669         return err;
1670 }
1671
1672 /*
1673  *      Helper to unshare the files of the current task.
1674  *      We don't want to expose copy_files internals to
1675  *      the exec layer of the kernel.
1676  */
1677
1678 int unshare_files(struct files_struct **displaced)
1679 {
1680         struct task_struct *task = current;
1681         struct files_struct *copy = NULL;
1682         int error;
1683
1684         error = unshare_fd(CLONE_FILES, &copy);
1685         if (error || !copy) {
1686                 *displaced = NULL;
1687                 return error;
1688         }
1689         *displaced = task->files;
1690         task_lock(task);
1691         task->files = copy;
1692         task_unlock(task);
1693         return 0;
1694 }