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/compat.h>
  44 #include <linux/task_io_accounting_ops.h>
  45 #include <linux/rcupdate.h>
  46 #include <linux/ptrace.h>
  47 #include <linux/mount.h>
  48 #include <linux/audit.h>
  49 #include <linux/memcontrol.h>
  50 #include <linux/ftrace.h>
  51 #include <linux/profile.h>
  52 #include <linux/rmap.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 <trace/sched.h>
  64
  65 #include <asm/pgtable.h>
  66 #include <asm/pgalloc.h>
  67 #include <asm/uaccess.h>
  68 #include <asm/mmu_context.h>
  69 #include <asm/cacheflush.h>
  70 #include <asm/tlbflush.h>
  71
  72 /*
  73  * Protected counters by write_lock_irq(&tasklist_lock)
  74  */
  75 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  76 int nr_threads;                 /* The idle threads do not count.. */
  77
  78 int max_threads;                /* tunable limit on nr_threads */
  79
  80 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  81
  82 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  83
  84 DEFINE_TRACE(sched_process_fork);
  85
  86 int nr_processes(void)
  87 {
  88         int cpu;
  89         int total = 0;
  90
  91         for_each_online_cpu(cpu)
  92                 total += per_cpu(process_counts, cpu);
  93
  94         return total;
  95 }
  96
  97 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
  98 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
  99 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
 100 static struct kmem_cache *task_struct_cachep;
 101 #endif
 102
 103 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
 104 static inline struct thread_info *alloc_thread_info(struct task_struct *tsk)
 105 {
 106 #ifdef CONFIG_DEBUG_STACK_USAGE
 107         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
 108 #else
 109         gfp_t mask = GFP_KERNEL;
 110 #endif
 111         return (struct thread_info *)__get_free_pages(mask, THREAD_SIZE_ORDER);
 112 }
 113
 114 static inline void free_thread_info(struct thread_info *ti)
 115 {
 116         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 117 }
 118 #endif
 119
 120 /* SLAB cache for signal_struct structures (tsk->signal) */
 121 static struct kmem_cache *signal_cachep;
 122
 123 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 124 struct kmem_cache *sighand_cachep;
 125
 126 /* SLAB cache for files_struct structures (tsk->files) */
 127 struct kmem_cache *files_cachep;
 128
 129 /* SLAB cache for fs_struct structures (tsk->fs) */
 130 struct kmem_cache *fs_cachep;
 131
 132 /* SLAB cache for vm_area_struct structures */
 133 struct kmem_cache *vm_area_cachep;
 134
 135 /* SLAB cache for mm_struct structures (tsk->mm) */
 136 static struct kmem_cache *mm_cachep;
 137
 138 void free_task(struct task_struct *tsk)
 139 {
 140         prop_local_destroy_single(&tsk->dirties);
 141         free_thread_info(tsk->stack);
 142         rt_mutex_debug_task_free(tsk);
 143         ftrace_graph_exit_task(tsk);
 144         free_task_struct(tsk);
 145 }
 146 EXPORT_SYMBOL(free_task);
 147
 148 void __put_task_struct(struct task_struct *tsk)
 149 {
 150         WARN_ON(!tsk->exit_state);
 151         WARN_ON(atomic_read(&tsk->usage));
 152         WARN_ON(tsk == current);
 153
 154         put_cred(tsk->real_cred);
 155         put_cred(tsk->cred);
 156         delayacct_tsk_free(tsk);
 157
 158         if (!profile_handoff_task(tsk))
 159                 free_task(tsk);
 160 }
 161
 162 /*
 163  * macro override instead of weak attribute alias, to workaround
 164  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
 165  */
 166 #ifndef arch_task_cache_init
 167 #define arch_task_cache_init()
 168 #endif
 169
 170 void __init fork_init(unsigned long mempages)
 171 {
 172 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 173 #ifndef ARCH_MIN_TASKALIGN
 174 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 175 #endif
 176         /* create a slab on which task_structs can be allocated */
 177         task_struct_cachep =
 178                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 179                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL);
 180 #endif
 181
 182         /* do the arch specific task caches init */
 183         arch_task_cache_init();
 184
 185         /*
 186          * The default maximum number of threads is set to a safe
 187          * value: the thread structures can take up at most half
 188          * of memory.
 189          */
 190         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 191
 192         /*
 193          * we need to allow at least 20 threads to boot a system
 194          */
 195         if(max_threads < 20)
 196                 max_threads = 20;
 197
 198         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 199         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 200         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 201                 init_task.signal->rlim[RLIMIT_NPROC];
 202 }
 203
 204 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
 205                                                struct task_struct *src)
 206 {
 207         *dst = *src;
 208         return 0;
 209 }
 210
 211 static struct task_struct *dup_task_struct(struct task_struct *orig)
 212 {
 213         struct task_struct *tsk;
 214         struct thread_info *ti;
 215         int err;
 216
 217         prepare_to_copy(orig);
 218
 219         tsk = alloc_task_struct();
 220         if (!tsk)
 221                 return NULL;
 222
 223         ti = alloc_thread_info(tsk);
 224         if (!ti) {
 225                 free_task_struct(tsk);
 226                 return NULL;
 227         }
 228
 229         err = arch_dup_task_struct(tsk, orig);
 230         if (err)
 231                 goto out;
 232
 233         tsk->stack = ti;
 234
 235         err = prop_local_init_single(&tsk->dirties);
 236         if (err)
 237                 goto out;
 238
 239         setup_thread_stack(tsk, orig);
 240
 241 #ifdef CONFIG_CC_STACKPROTECTOR
 242         tsk->stack_canary = get_random_int();
 243 #endif
 244
 245         /* One for us, one for whoever does the "release_task()" (usually parent) */
 246         atomic_set(&tsk->usage,2);
 247         atomic_set(&tsk->fs_excl, 0);
 248 #ifdef CONFIG_BLK_DEV_IO_TRACE
 249         tsk->btrace_seq = 0;
 250 #endif
 251         tsk->splice_pipe = NULL;
 252         return tsk;
 253
 254 out:
 255         free_thread_info(ti);
 256         free_task_struct(tsk);
 257         return NULL;
 258 }
 259
 260 #ifdef CONFIG_MMU
 261 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 262 {
 263         struct vm_area_struct *mpnt, *tmp, **pprev;
 264         struct rb_node **rb_link, *rb_parent;
 265         int retval;
 266         unsigned long charge;
 267         struct mempolicy *pol;
 268
 269         down_write(&oldmm->mmap_sem);
 270         flush_cache_dup_mm(oldmm);
 271         /*
 272          * Not linked in yet - no deadlock potential:
 273          */
 274         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 275
 276         mm->locked_vm = 0;
 277         mm->mmap = NULL;
 278         mm->mmap_cache = NULL;
 279         mm->free_area_cache = oldmm->mmap_base;
 280         mm->cached_hole_size = ~0UL;
 281         mm->map_count = 0;
 282         cpus_clear(mm->cpu_vm_mask);
 283         mm->mm_rb = RB_ROOT;
 284         rb_link = &mm->mm_rb.rb_node;
 285         rb_parent = NULL;
 286         pprev = &mm->mmap;
 287
 288         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 289                 struct file *file;
 290
 291                 if (mpnt->vm_flags & VM_DONTCOPY) {
 292                         long pages = vma_pages(mpnt);
 293                         mm->total_vm -= pages;
 294                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 295                                                                 -pages);
 296                         continue;
 297                 }
 298                 charge = 0;
 299                 if (mpnt->vm_flags & VM_ACCOUNT) {
 300                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 301                         if (security_vm_enough_memory(len))
 302                                 goto fail_nomem;
 303                         charge = len;
 304                 }
 305                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 306                 if (!tmp)
 307                         goto fail_nomem;
 308                 *tmp = *mpnt;
 309                 pol = mpol_dup(vma_policy(mpnt));
 310                 retval = PTR_ERR(pol);
 311                 if (IS_ERR(pol))
 312                         goto fail_nomem_policy;
 313                 vma_set_policy(tmp, pol);
 314                 tmp->vm_flags &= ~VM_LOCKED;
 315                 tmp->vm_mm = mm;
 316                 tmp->vm_next = NULL;
 317                 anon_vma_link(tmp);
 318                 file = tmp->vm_file;
 319                 if (file) {
 320                         struct inode *inode = file->f_path.dentry->d_inode;
 321                         struct address_space *mapping = file->f_mapping;
 322
 323                         get_file(file);
 324                         if (tmp->vm_flags & VM_DENYWRITE)
 325                                 atomic_dec(&inode->i_writecount);
 326                         spin_lock(&mapping->i_mmap_lock);
 327                         if (tmp->vm_flags & VM_SHARED)
 328                                 mapping->i_mmap_writable++;
 329                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
 330                         flush_dcache_mmap_lock(mapping);
 331                         /* insert tmp into the share list, just after mpnt */
 332                         vma_prio_tree_add(tmp, mpnt);
 333                         flush_dcache_mmap_unlock(mapping);
 334                         spin_unlock(&mapping->i_mmap_lock);
 335                 }
 336
 337                 /*
 338                  * Clear hugetlb-related page reserves for children. This only
 339                  * affects MAP_PRIVATE mappings. Faults generated by the child
 340                  * are not guaranteed to succeed, even if read-only
 341                  */
 342                 if (is_vm_hugetlb_page(tmp))
 343                         reset_vma_resv_huge_pages(tmp);
 344
 345                 /*
 346                  * Link in the new vma and copy the page table entries.
 347                  */
 348                 *pprev = tmp;
 349                 pprev = &tmp->vm_next;
 350
 351                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 352                 rb_link = &tmp->vm_rb.rb_right;
 353                 rb_parent = &tmp->vm_rb;
 354
 355                 mm->map_count++;
 356                 retval = copy_page_range(mm, oldmm, mpnt);
 357
 358                 if (tmp->vm_ops && tmp->vm_ops->open)
 359                         tmp->vm_ops->open(tmp);
 360
 361                 if (retval)
 362                         goto out;
 363         }
 364         /* a new mm has just been created */
 365         arch_dup_mmap(oldmm, mm);
 366         retval = 0;
 367 out:
 368         up_write(&mm->mmap_sem);
 369         flush_tlb_mm(oldmm);
 370         up_write(&oldmm->mmap_sem);
 371         return retval;
 372 fail_nomem_policy:
 373         kmem_cache_free(vm_area_cachep, tmp);
 374 fail_nomem:
 375         retval = -ENOMEM;
 376         vm_unacct_memory(charge);
 377         goto out;
 378 }
 379
 380 static inline int mm_alloc_pgd(struct mm_struct * mm)
 381 {
 382         mm->pgd = pgd_alloc(mm);
 383         if (unlikely(!mm->pgd))
 384                 return -ENOMEM;
 385         return 0;
 386 }
 387
 388 static inline void mm_free_pgd(struct mm_struct * mm)
 389 {
 390         pgd_free(mm, mm->pgd);
 391 }
 392 #else
 393 #define dup_mmap(mm, oldmm)     (0)
 394 #define mm_alloc_pgd(mm)        (0)
 395 #define mm_free_pgd(mm)
 396 #endif /* CONFIG_MMU */
 397
 398 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 399
 400 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 401 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 402
 403 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 404
 405 static int __init coredump_filter_setup(char *s)
 406 {
 407         default_dump_filter =
 408                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 409                 MMF_DUMP_FILTER_MASK;
 410         return 1;
 411 }
 412
 413 __setup("coredump_filter=", coredump_filter_setup);
 414
 415 #include <linux/init_task.h>
 416
 417 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
 418 {
 419         atomic_set(&mm->mm_users, 1);
 420         atomic_set(&mm->mm_count, 1);
 421         init_rwsem(&mm->mmap_sem);
 422         INIT_LIST_HEAD(&mm->mmlist);
 423         mm->flags = (current->mm) ? current->mm->flags : default_dump_filter;
 424         mm->core_state = NULL;
 425         mm->nr_ptes = 0;
 426         set_mm_counter(mm, file_rss, 0);
 427         set_mm_counter(mm, anon_rss, 0);
 428         spin_lock_init(&mm->page_table_lock);
 429         spin_lock_init(&mm->ioctx_lock);
 430         INIT_HLIST_HEAD(&mm->ioctx_list);
 431         mm->free_area_cache = TASK_UNMAPPED_BASE;
 432         mm->cached_hole_size = ~0UL;
 433         mm_init_owner(mm, p);
 434
 435         if (likely(!mm_alloc_pgd(mm))) {
 436                 mm->def_flags = 0;
 437                 mmu_notifier_mm_init(mm);
 438                 return mm;
 439         }
 440
 441         free_mm(mm);
 442         return NULL;
 443 }
 444
 445 /*
 446  * Allocate and initialize an mm_struct.
 447  */
 448 struct mm_struct * mm_alloc(void)
 449 {
 450         struct mm_struct * mm;
 451
 452         mm = allocate_mm();
 453         if (mm) {
 454                 memset(mm, 0, sizeof(*mm));
 455                 mm = mm_init(mm, current);
 456         }
 457         return mm;
 458 }
 459
 460 /*
 461  * Called when the last reference to the mm
 462  * is dropped: either by a lazy thread or by
 463  * mmput. Free the page directory and the mm.
 464  */
 465 void __mmdrop(struct mm_struct *mm)
 466 {
 467         BUG_ON(mm == &init_mm);
 468         mm_free_pgd(mm);
 469         destroy_context(mm);
 470         mmu_notifier_mm_destroy(mm);
 471         free_mm(mm);
 472 }
 473 EXPORT_SYMBOL_GPL(__mmdrop);
 474
 475 /*
 476  * Decrement the use count and release all resources for an mm.
 477  */
 478 void mmput(struct mm_struct *mm)
 479 {
 480         might_sleep();
 481
 482         if (atomic_dec_and_test(&mm->mm_users)) {
 483                 exit_aio(mm);
 484                 exit_mmap(mm);
 485                 set_mm_exe_file(mm, NULL);
 486                 if (!list_empty(&mm->mmlist)) {
 487                         spin_lock(&mmlist_lock);
 488                         list_del(&mm->mmlist);
 489                         spin_unlock(&mmlist_lock);
 490                 }
 491                 put_swap_token(mm);
 492                 mmdrop(mm);
 493         }
 494 }
 495 EXPORT_SYMBOL_GPL(mmput);
 496
 497 /**
 498  * get_task_mm - acquire a reference to the task's mm
 499  *
 500  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 501  * this kernel workthread has transiently adopted a user mm with use_mm,
 502  * to do its AIO) is not set and if so returns a reference to it, after
 503  * bumping up the use count.  User must release the mm via mmput()
 504  * after use.  Typically used by /proc and ptrace.
 505  */
 506 struct mm_struct *get_task_mm(struct task_struct *task)
 507 {
 508         struct mm_struct *mm;
 509
 510         task_lock(task);
 511         mm = task->mm;
 512         if (mm) {
 513                 if (task->flags & PF_KTHREAD)
 514                         mm = NULL;
 515                 else
 516                         atomic_inc(&mm->mm_users);
 517         }
 518         task_unlock(task);
 519         return mm;
 520 }
 521 EXPORT_SYMBOL_GPL(get_task_mm);
 522
 523 /* Please note the differences between mmput and mm_release.
 524  * mmput is called whenever we stop holding onto a mm_struct,
 525  * error success whatever.
 526  *
 527  * mm_release is called after a mm_struct has been removed
 528  * from the current process.
 529  *
 530  * This difference is important for error handling, when we
 531  * only half set up a mm_struct for a new process and need to restore
 532  * the old one.  Because we mmput the new mm_struct before
 533  * restoring the old one. . .
 534  * Eric Biederman 10 January 1998
 535  */
 536 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 537 {
 538         struct completion *vfork_done = tsk->vfork_done;
 539
 540         /* Get rid of any futexes when releasing the mm */
 541 #ifdef CONFIG_FUTEX
 542         if (unlikely(tsk->robust_list))
 543                 exit_robust_list(tsk);
 544 #ifdef CONFIG_COMPAT
 545         if (unlikely(tsk->compat_robust_list))
 546                 compat_exit_robust_list(tsk);
 547 #endif
 548 #endif
 549
 550         /* Get rid of any cached register state */
 551         deactivate_mm(tsk, mm);
 552
 553         /* notify parent sleeping on vfork() */
 554         if (vfork_done) {
 555                 tsk->vfork_done = NULL;
 556                 complete(vfork_done);
 557         }
 558
 559         /*
 560          * If we're exiting normally, clear a user-space tid field if
 561          * requested.  We leave this alone when dying by signal, to leave
 562          * the value intact in a core dump, and to save the unnecessary
 563          * trouble otherwise.  Userland only wants this done for a sys_exit.
 564          */
 565         if (tsk->clear_child_tid
 566             && !(tsk->flags & PF_SIGNALED)
 567             && atomic_read(&mm->mm_users) > 1) {
 568                 u32 __user * tidptr = tsk->clear_child_tid;
 569                 tsk->clear_child_tid = NULL;
 570
 571                 /*
 572                  * We don't check the error code - if userspace has
 573                  * not set up a proper pointer then tough luck.
 574                  */
 575                 put_user(0, tidptr);
 576                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
 577         }
 578 }
 579
 580 /*
 581  * Allocate a new mm structure and copy contents from the
 582  * mm structure of the passed in task structure.
 583  */
 584 struct mm_struct *dup_mm(struct task_struct *tsk)
 585 {
 586         struct mm_struct *mm, *oldmm = current->mm;
 587         int err;
 588
 589         if (!oldmm)
 590                 return NULL;
 591
 592         mm = allocate_mm();
 593         if (!mm)
 594                 goto fail_nomem;
 595
 596         memcpy(mm, oldmm, sizeof(*mm));
 597
 598         /* Initializing for Swap token stuff */
 599         mm->token_priority = 0;
 600         mm->last_interval = 0;
 601
 602         if (!mm_init(mm, tsk))
 603                 goto fail_nomem;
 604
 605         if (init_new_context(tsk, mm))
 606                 goto fail_nocontext;
 607
 608         dup_mm_exe_file(oldmm, mm);
 609
 610         err = dup_mmap(mm, oldmm);
 611         if (err)
 612                 goto free_pt;
 613
 614         mm->hiwater_rss = get_mm_rss(mm);
 615         mm->hiwater_vm = mm->total_vm;
 616
 617         return mm;
 618
 619 free_pt:
 620         mmput(mm);
 621
 622 fail_nomem:
 623         return NULL;
 624
 625 fail_nocontext:
 626         /*
 627          * If init_new_context() failed, we cannot use mmput() to free the mm
 628          * because it calls destroy_context()
 629          */
 630         mm_free_pgd(mm);
 631         free_mm(mm);
 632         return NULL;
 633 }
 634
 635 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 636 {
 637         struct mm_struct * mm, *oldmm;
 638         int retval;
 639
 640         tsk->min_flt = tsk->maj_flt = 0;
 641         tsk->nvcsw = tsk->nivcsw = 0;
 642
 643         tsk->mm = NULL;
 644         tsk->active_mm = NULL;
 645
 646         /*
 647          * Are we cloning a kernel thread?
 648          *
 649          * We need to steal a active VM for that..
 650          */
 651         oldmm = current->mm;
 652         if (!oldmm)
 653                 return 0;
 654
 655         if (clone_flags & CLONE_VM) {
 656                 atomic_inc(&oldmm->mm_users);
 657                 mm = oldmm;
 658                 goto good_mm;
 659         }
 660
 661         retval = -ENOMEM;
 662         mm = dup_mm(tsk);
 663         if (!mm)
 664                 goto fail_nomem;
 665
 666 good_mm:
 667         /* Initializing for Swap token stuff */
 668         mm->token_priority = 0;
 669         mm->last_interval = 0;
 670
 671         tsk->mm = mm;
 672         tsk->active_mm = mm;
 673         return 0;
 674
 675 fail_nomem:
 676         return retval;
 677 }
 678
 679 static struct fs_struct *__copy_fs_struct(struct fs_struct *old)
 680 {
 681         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
 682         /* We don't need to lock fs - think why ;-) */
 683         if (fs) {
 684                 atomic_set(&fs->count, 1);
 685                 rwlock_init(&fs->lock);
 686                 fs->umask = old->umask;
 687                 read_lock(&old->lock);
 688                 fs->root = old->root;
 689                 path_get(&old->root);
 690                 fs->pwd = old->pwd;
 691                 path_get(&old->pwd);
 692                 read_unlock(&old->lock);
 693         }
 694         return fs;
 695 }
 696
 697 struct fs_struct *copy_fs_struct(struct fs_struct *old)
 698 {
 699         return __copy_fs_struct(old);
 700 }
 701
 702 EXPORT_SYMBOL_GPL(copy_fs_struct);
 703
 704 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 705 {
 706         if (clone_flags & CLONE_FS) {
 707                 atomic_inc(&current->fs->count);
 708                 return 0;
 709         }
 710         tsk->fs = __copy_fs_struct(current->fs);
 711         if (!tsk->fs)
 712                 return -ENOMEM;
 713         return 0;
 714 }
 715
 716 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 717 {
 718         struct files_struct *oldf, *newf;
 719         int error = 0;
 720
 721         /*
 722          * A background process may not have any files ...
 723          */
 724         oldf = current->files;
 725         if (!oldf)
 726                 goto out;
 727
 728         if (clone_flags & CLONE_FILES) {
 729                 atomic_inc(&oldf->count);
 730                 goto out;
 731         }
 732
 733         newf = dup_fd(oldf, &error);
 734         if (!newf)
 735                 goto out;
 736
 737         tsk->files = newf;
 738         error = 0;
 739 out:
 740         return error;
 741 }
 742
 743 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 744 {
 745 #ifdef CONFIG_BLOCK
 746         struct io_context *ioc = current->io_context;
 747
 748         if (!ioc)
 749                 return 0;
 750         /*
 751          * Share io context with parent, if CLONE_IO is set
 752          */
 753         if (clone_flags & CLONE_IO) {
 754                 tsk->io_context = ioc_task_link(ioc);
 755                 if (unlikely(!tsk->io_context))
 756                         return -ENOMEM;
 757         } else if (ioprio_valid(ioc->ioprio)) {
 758                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
 759                 if (unlikely(!tsk->io_context))
 760                         return -ENOMEM;
 761
 762                 tsk->io_context->ioprio = ioc->ioprio;
 763         }
 764 #endif
 765         return 0;
 766 }
 767
 768 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 769 {
 770         struct sighand_struct *sig;
 771
 772         if (clone_flags & CLONE_SIGHAND) {
 773                 atomic_inc(&current->sighand->count);
 774                 return 0;
 775         }
 776         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 777         rcu_assign_pointer(tsk->sighand, sig);
 778         if (!sig)
 779                 return -ENOMEM;
 780         atomic_set(&sig->count, 1);
 781         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 782         return 0;
 783 }
 784
 785 void __cleanup_sighand(struct sighand_struct *sighand)
 786 {
 787         if (atomic_dec_and_test(&sighand->count))
 788                 kmem_cache_free(sighand_cachep, sighand);
 789 }
 790
 791
 792 /*
 793  * Initialize POSIX timer handling for a thread group.
 794  */
 795 static void posix_cpu_timers_init_group(struct signal_struct *sig)
 796 {
 797         /* Thread group counters. */
 798         thread_group_cputime_init(sig);
 799
 800         /* Expiration times and increments. */
 801         sig->it_virt_expires = cputime_zero;
 802         sig->it_virt_incr = cputime_zero;
 803         sig->it_prof_expires = cputime_zero;
 804         sig->it_prof_incr = cputime_zero;
 805
 806         /* Cached expiration times. */
 807         sig->cputime_expires.prof_exp = cputime_zero;
 808         sig->cputime_expires.virt_exp = cputime_zero;
 809         sig->cputime_expires.sched_exp = 0;
 810
 811         /* The timer lists. */
 812         INIT_LIST_HEAD(&sig->cpu_timers[0]);
 813         INIT_LIST_HEAD(&sig->cpu_timers[1]);
 814         INIT_LIST_HEAD(&sig->cpu_timers[2]);
 815 }
 816
 817 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
 818 {
 819         struct signal_struct *sig;
 820
 821         if (clone_flags & CLONE_THREAD) {
 822                 atomic_inc(&current->signal->count);
 823                 atomic_inc(&current->signal->live);
 824                 return 0;
 825         }
 826         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 827
 828         if (sig)
 829                 posix_cpu_timers_init_group(sig);
 830
 831         tsk->signal = sig;
 832         if (!sig)
 833                 return -ENOMEM;
 834
 835         atomic_set(&sig->count, 1);
 836         atomic_set(&sig->live, 1);
 837         init_waitqueue_head(&sig->wait_chldexit);
 838         sig->flags = 0;
 839         sig->group_exit_code = 0;
 840         sig->group_exit_task = NULL;
 841         sig->group_stop_count = 0;
 842         sig->curr_target = tsk;
 843         init_sigpending(&sig->shared_pending);
 844         INIT_LIST_HEAD(&sig->posix_timers);
 845
 846         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 847         sig->it_real_incr.tv64 = 0;
 848         sig->real_timer.function = it_real_fn;
 849
 850         sig->leader = 0;        /* session leadership doesn't inherit */
 851         sig->tty_old_pgrp = NULL;
 852         sig->tty = NULL;
 853
 854         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
 855         sig->gtime = cputime_zero;
 856         sig->cgtime = cputime_zero;
 857         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
 858         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
 859         sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
 860         task_io_accounting_init(&sig->ioac);
 861         sig->sum_sched_runtime = 0;
 862         taskstats_tgid_init(sig);
 863
 864         task_lock(current->group_leader);
 865         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
 866         task_unlock(current->group_leader);
 867
 868         acct_init_pacct(&sig->pacct);
 869
 870         tty_audit_fork(sig);
 871
 872         return 0;
 873 }
 874
 875 void __cleanup_signal(struct signal_struct *sig)
 876 {
 877         thread_group_cputime_free(sig);
 878         tty_kref_put(sig->tty);
 879         kmem_cache_free(signal_cachep, sig);
 880 }
 881
 882 static void cleanup_signal(struct task_struct *tsk)
 883 {
 884         struct signal_struct *sig = tsk->signal;
 885
 886         atomic_dec(&sig->live);
 887
 888         if (atomic_dec_and_test(&sig->count))
 889                 __cleanup_signal(sig);
 890 }
 891
 892 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
 893 {
 894         unsigned long new_flags = p->flags;
 895
 896         new_flags &= ~PF_SUPERPRIV;
 897         new_flags |= PF_FORKNOEXEC;
 898         new_flags |= PF_STARTING;
 899         p->flags = new_flags;
 900         clear_freeze_flag(p);
 901 }
 902
 903 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
 904 {
 905         current->clear_child_tid = tidptr;
 906
 907         return task_pid_vnr(current);
 908 }
 909
 910 static void rt_mutex_init_task(struct task_struct *p)
 911 {
 912         spin_lock_init(&p->pi_lock);
 913 #ifdef CONFIG_RT_MUTEXES
 914         plist_head_init(&p->pi_waiters, &p->pi_lock);
 915         p->pi_blocked_on = NULL;
 916 #endif
 917 }
 918
 919 #ifdef CONFIG_MM_OWNER
 920 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 921 {
 922         mm->owner = p;
 923 }
 924 #endif /* CONFIG_MM_OWNER */
 925
 926 /*
 927  * Initialize POSIX timer handling for a single task.
 928  */
 929 static void posix_cpu_timers_init(struct task_struct *tsk)
 930 {
 931         tsk->cputime_expires.prof_exp = cputime_zero;
 932         tsk->cputime_expires.virt_exp = cputime_zero;
 933         tsk->cputime_expires.sched_exp = 0;
 934         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
 935         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
 936         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
 937 }
 938
 939 /*
 940  * This creates a new process as a copy of the old one,
 941  * but does not actually start it yet.
 942  *
 943  * It copies the registers, and all the appropriate
 944  * parts of the process environment (as per the clone
 945  * flags). The actual kick-off is left to the caller.
 946  */
 947 static struct task_struct *copy_process(unsigned long clone_flags,
 948                                         unsigned long stack_start,
 949                                         struct pt_regs *regs,
 950                                         unsigned long stack_size,
 951                                         int __user *child_tidptr,
 952                                         struct pid *pid,
 953                                         int trace)
 954 {
 955         int retval;
 956         struct task_struct *p;
 957         int cgroup_callbacks_done = 0;
 958
 959         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 960                 return ERR_PTR(-EINVAL);
 961
 962         /*
 963          * Thread groups must share signals as well, and detached threads
 964          * can only be started up within the thread group.
 965          */
 966         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
 967                 return ERR_PTR(-EINVAL);
 968
 969         /*
 970          * Shared signal handlers imply shared VM. By way of the above,
 971          * thread groups also imply shared VM. Blocking this case allows
 972          * for various simplifications in other code.
 973          */
 974         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
 975                 return ERR_PTR(-EINVAL);
 976
 977         retval = security_task_create(clone_flags);
 978         if (retval)
 979                 goto fork_out;
 980
 981         retval = -ENOMEM;
 982         p = dup_task_struct(current);
 983         if (!p)
 984                 goto fork_out;
 985
 986         rt_mutex_init_task(p);
 987
 988 #ifdef CONFIG_PROVE_LOCKING
 989         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
 990         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
 991 #endif
 992         retval = -EAGAIN;
 993         if (atomic_read(&p->real_cred->user->processes) >=
 994                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
 995                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
 996                     p->real_cred->user != INIT_USER)
 997                         goto bad_fork_free;
 998         }
 999
1000         retval = copy_creds(p, clone_flags);
1001         if (retval < 0)
1002                 goto bad_fork_free;
1003
1004         /*
1005          * If multiple threads are within copy_process(), then this check
1006          * triggers too late. This doesn't hurt, the check is only there
1007          * to stop root fork bombs.
1008          */
1009         retval = -EAGAIN;
1010         if (nr_threads >= max_threads)
1011                 goto bad_fork_cleanup_count;
1012
1013         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1014                 goto bad_fork_cleanup_count;
1015
1016         if (p->binfmt && !try_module_get(p->binfmt->module))
1017                 goto bad_fork_cleanup_put_domain;
1018
1019         p->did_exec = 0;
1020         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1021         copy_flags(clone_flags, p);
1022         INIT_LIST_HEAD(&p->children);
1023         INIT_LIST_HEAD(&p->sibling);
1024 #ifdef CONFIG_PREEMPT_RCU
1025         p->rcu_read_lock_nesting = 0;
1026         p->rcu_flipctr_idx = 0;
1027 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1028         p->vfork_done = NULL;
1029         spin_lock_init(&p->alloc_lock);
1030
1031         clear_tsk_thread_flag(p, TIF_SIGPENDING);
1032         init_sigpending(&p->pending);
1033
1034         p->utime = cputime_zero;
1035         p->stime = cputime_zero;
1036         p->gtime = cputime_zero;
1037         p->utimescaled = cputime_zero;
1038         p->stimescaled = cputime_zero;
1039         p->prev_utime = cputime_zero;
1040         p->prev_stime = cputime_zero;
1041
1042         p->default_timer_slack_ns = current->timer_slack_ns;
1043
1044 #ifdef CONFIG_DETECT_SOFTLOCKUP
1045         p->last_switch_count = 0;
1046         p->last_switch_timestamp = 0;
1047 #endif
1048
1049         task_io_accounting_init(&p->ioac);
1050         acct_clear_integrals(p);
1051
1052         posix_cpu_timers_init(p);
1053
1054         p->lock_depth = -1;             /* -1 = no lock */
1055         do_posix_clock_monotonic_gettime(&p->start_time);
1056         p->real_start_time = p->start_time;
1057         monotonic_to_bootbased(&p->real_start_time);
1058         p->io_context = NULL;
1059         p->audit_context = NULL;
1060         cgroup_fork(p);
1061 #ifdef CONFIG_NUMA
1062         p->mempolicy = mpol_dup(p->mempolicy);
1063         if (IS_ERR(p->mempolicy)) {
1064                 retval = PTR_ERR(p->mempolicy);
1065                 p->mempolicy = NULL;
1066                 goto bad_fork_cleanup_cgroup;
1067         }
1068         mpol_fix_fork_child_flag(p);
1069 #endif
1070 #ifdef CONFIG_TRACE_IRQFLAGS
1071         p->irq_events = 0;
1072 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1073         p->hardirqs_enabled = 1;
1074 #else
1075         p->hardirqs_enabled = 0;
1076 #endif
1077         p->hardirq_enable_ip = 0;
1078         p->hardirq_enable_event = 0;
1079         p->hardirq_disable_ip = _THIS_IP_;
1080         p->hardirq_disable_event = 0;
1081         p->softirqs_enabled = 1;
1082         p->softirq_enable_ip = _THIS_IP_;
1083         p->softirq_enable_event = 0;
1084         p->softirq_disable_ip = 0;
1085         p->softirq_disable_event = 0;
1086         p->hardirq_context = 0;
1087         p->softirq_context = 0;
1088 #endif
1089 #ifdef CONFIG_LOCKDEP
1090         p->lockdep_depth = 0; /* no locks held yet */
1091         p->curr_chain_key = 0;
1092         p->lockdep_recursion = 0;
1093 #endif
1094
1095 #ifdef CONFIG_DEBUG_MUTEXES
1096         p->blocked_on = NULL; /* not blocked yet */
1097 #endif
1098         if (unlikely(current->ptrace))
1099                 ptrace_fork(p, clone_flags);
1100
1101         /* Perform scheduler related setup. Assign this task to a CPU. */
1102         sched_fork(p, clone_flags);
1103
1104         if ((retval = audit_alloc(p)))
1105                 goto bad_fork_cleanup_policy;
1106         /* copy all the process information */
1107         if ((retval = copy_semundo(clone_flags, p)))
1108                 goto bad_fork_cleanup_audit;
1109         if ((retval = copy_files(clone_flags, p)))
1110                 goto bad_fork_cleanup_semundo;
1111         if ((retval = copy_fs(clone_flags, p)))
1112                 goto bad_fork_cleanup_files;
1113         if ((retval = copy_sighand(clone_flags, p)))
1114                 goto bad_fork_cleanup_fs;
1115         if ((retval = copy_signal(clone_flags, p)))
1116                 goto bad_fork_cleanup_sighand;
1117         if ((retval = copy_mm(clone_flags, p)))
1118                 goto bad_fork_cleanup_signal;
1119         if ((retval = copy_namespaces(clone_flags, p)))
1120                 goto bad_fork_cleanup_mm;
1121         if ((retval = copy_io(clone_flags, p)))
1122                 goto bad_fork_cleanup_namespaces;
1123         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1124         if (retval)
1125                 goto bad_fork_cleanup_io;
1126
1127         if (pid != &init_struct_pid) {
1128                 retval = -ENOMEM;
1129                 pid = alloc_pid(p->nsproxy->pid_ns);
1130                 if (!pid)
1131                         goto bad_fork_cleanup_io;
1132
1133                 if (clone_flags & CLONE_NEWPID) {
1134                         retval = pid_ns_prepare_proc(p->nsproxy->pid_ns);
1135                         if (retval < 0)
1136                                 goto bad_fork_free_pid;
1137                 }
1138         }
1139
1140         ftrace_graph_init_task(p);
1141
1142         p->pid = pid_nr(pid);
1143         p->tgid = p->pid;
1144         if (clone_flags & CLONE_THREAD)
1145                 p->tgid = current->tgid;
1146
1147         if (current->nsproxy != p->nsproxy) {
1148                 retval = ns_cgroup_clone(p, pid);
1149                 if (retval)
1150                         goto bad_fork_free_graph;
1151         }
1152
1153         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1154         /*
1155          * Clear TID on mm_release()?
1156          */
1157         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1158 #ifdef CONFIG_FUTEX
1159         p->robust_list = NULL;
1160 #ifdef CONFIG_COMPAT
1161         p->compat_robust_list = NULL;
1162 #endif
1163         INIT_LIST_HEAD(&p->pi_state_list);
1164         p->pi_state_cache = NULL;
1165 #endif
1166         /*
1167          * sigaltstack should be cleared when sharing the same VM
1168          */
1169         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1170                 p->sas_ss_sp = p->sas_ss_size = 0;
1171
1172         /*
1173          * Syscall tracing should be turned off in the child regardless
1174          * of CLONE_PTRACE.
1175          */
1176         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1177 #ifdef TIF_SYSCALL_EMU
1178         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1179 #endif
1180         clear_all_latency_tracing(p);
1181
1182         /* Our parent execution domain becomes current domain
1183            These must match for thread signalling to apply */
1184         p->parent_exec_id = p->self_exec_id;
1185
1186         /* ok, now we should be set up.. */
1187         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1188         p->pdeath_signal = 0;
1189         p->exit_state = 0;
1190
1191         /*
1192          * Ok, make it visible to the rest of the system.
1193          * We dont wake it up yet.
1194          */
1195         p->group_leader = p;
1196         INIT_LIST_HEAD(&p->thread_group);
1197
1198         /* Now that the task is set up, run cgroup callbacks if
1199          * necessary. We need to run them before the task is visible
1200          * on the tasklist. */
1201         cgroup_fork_callbacks(p);
1202         cgroup_callbacks_done = 1;
1203
1204         /* Need tasklist lock for parent etc handling! */
1205         write_lock_irq(&tasklist_lock);
1206
1207         /*
1208          * The task hasn't been attached yet, so its cpus_allowed mask will
1209          * not be changed, nor will its assigned CPU.
1210          *
1211          * The cpus_allowed mask of the parent may have changed after it was
1212          * copied first time - so re-copy it here, then check the child's CPU
1213          * to ensure it is on a valid CPU (and if not, just force it back to
1214          * parent's CPU). This avoids alot of nasty races.
1215          */
1216         p->cpus_allowed = current->cpus_allowed;
1217         p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1218         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1219                         !cpu_online(task_cpu(p))))
1220                 set_task_cpu(p, smp_processor_id());
1221
1222         /* CLONE_PARENT re-uses the old parent */
1223         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1224                 p->real_parent = current->real_parent;
1225         else
1226                 p->real_parent = current;
1227
1228         spin_lock(&current->sighand->siglock);
1229
1230         /*
1231          * Process group and session signals need to be delivered to just the
1232          * parent before the fork or both the parent and the child after the
1233          * fork. Restart if a signal comes in before we add the new process to
1234          * it's process group.
1235          * A fatal signal pending means that current will exit, so the new
1236          * thread can't slip out of an OOM kill (or normal SIGKILL).
1237          */
1238         recalc_sigpending();
1239         if (signal_pending(current)) {
1240                 spin_unlock(&current->sighand->siglock);
1241                 write_unlock_irq(&tasklist_lock);
1242                 retval = -ERESTARTNOINTR;
1243                 goto bad_fork_free_graph;
1244         }
1245
1246         if (clone_flags & CLONE_THREAD) {
1247                 p->group_leader = current->group_leader;
1248                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1249         }
1250
1251         if (likely(p->pid)) {
1252                 list_add_tail(&p->sibling, &p->real_parent->children);
1253                 tracehook_finish_clone(p, clone_flags, trace);
1254
1255                 if (thread_group_leader(p)) {
1256                         if (clone_flags & CLONE_NEWPID)
1257                                 p->nsproxy->pid_ns->child_reaper = p;
1258
1259                         p->signal->leader_pid = pid;
1260                         tty_kref_put(p->signal->tty);
1261                         p->signal->tty = tty_kref_get(current->signal->tty);
1262                         set_task_pgrp(p, task_pgrp_nr(current));
1263                         set_task_session(p, task_session_nr(current));
1264                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1265                         attach_pid(p, PIDTYPE_SID, task_session(current));
1266                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1267                         __get_cpu_var(process_counts)++;
1268                 }
1269                 attach_pid(p, PIDTYPE_PID, pid);
1270                 nr_threads++;
1271         }
1272
1273         total_forks++;
1274         spin_unlock(&current->sighand->siglock);
1275         write_unlock_irq(&tasklist_lock);
1276         proc_fork_connector(p);
1277         cgroup_post_fork(p);
1278         return p;
1279
1280 bad_fork_free_graph:
1281         ftrace_graph_exit_task(p);
1282 bad_fork_free_pid:
1283         if (pid != &init_struct_pid)
1284                 free_pid(pid);
1285 bad_fork_cleanup_io:
1286         put_io_context(p->io_context);
1287 bad_fork_cleanup_namespaces:
1288         exit_task_namespaces(p);
1289 bad_fork_cleanup_mm:
1290         if (p->mm)
1291                 mmput(p->mm);
1292 bad_fork_cleanup_signal:
1293         cleanup_signal(p);
1294 bad_fork_cleanup_sighand:
1295         __cleanup_sighand(p->sighand);
1296 bad_fork_cleanup_fs:
1297         exit_fs(p); /* blocking */
1298 bad_fork_cleanup_files:
1299         exit_files(p); /* blocking */
1300 bad_fork_cleanup_semundo:
1301         exit_sem(p);
1302 bad_fork_cleanup_audit:
1303         audit_free(p);
1304 bad_fork_cleanup_policy:
1305 #ifdef CONFIG_NUMA
1306         mpol_put(p->mempolicy);
1307 bad_fork_cleanup_cgroup:
1308 #endif
1309         cgroup_exit(p, cgroup_callbacks_done);
1310         delayacct_tsk_free(p);
1311         if (p->binfmt)
1312                 module_put(p->binfmt->module);
1313 bad_fork_cleanup_put_domain:
1314         module_put(task_thread_info(p)->exec_domain->module);
1315 bad_fork_cleanup_count:
1316         atomic_dec(&p->cred->user->processes);
1317         put_cred(p->real_cred);
1318         put_cred(p->cred);
1319 bad_fork_free:
1320         free_task(p);
1321 fork_out:
1322         return ERR_PTR(retval);
1323 }
1324
1325 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1326 {
1327         memset(regs, 0, sizeof(struct pt_regs));
1328         return regs;
1329 }
1330
1331 struct task_struct * __cpuinit fork_idle(int cpu)
1332 {
1333         struct task_struct *task;
1334         struct pt_regs regs;
1335
1336         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1337                             &init_struct_pid, 0);
1338         if (!IS_ERR(task))
1339                 init_idle(task, cpu);
1340
1341         return task;
1342 }
1343
1344 /*
1345  *  Ok, this is the main fork-routine.
1346  *
1347  * It copies the process, and if successful kick-starts
1348  * it and waits for it to finish using the VM if required.
1349  */
1350 long do_fork(unsigned long clone_flags,
1351               unsigned long stack_start,
1352               struct pt_regs *regs,
1353               unsigned long stack_size,
1354               int __user *parent_tidptr,
1355               int __user *child_tidptr)
1356 {
1357         struct task_struct *p;
1358         int trace = 0;
1359         long nr;
1360
1361         /*
1362          * Do some preliminary argument and permissions checking before we
1363          * actually start allocating stuff
1364          */
1365         if (clone_flags & CLONE_NEWUSER) {
1366                 if (clone_flags & CLONE_THREAD)
1367                         return -EINVAL;
1368                 /* hopefully this check will go away when userns support is
1369                  * complete
1370                  */
1371                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1372                                 !capable(CAP_SETGID))
1373                         return -EPERM;
1374         }
1375
1376         /*
1377          * We hope to recycle these flags after 2.6.26
1378          */
1379         if (unlikely(clone_flags & CLONE_STOPPED)) {
1380                 static int __read_mostly count = 100;
1381
1382                 if (count > 0 && printk_ratelimit()) {
1383                         char comm[TASK_COMM_LEN];
1384
1385                         count--;
1386                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1387                                         "clone flags 0x%lx\n",
1388                                 get_task_comm(comm, current),
1389                                 clone_flags & CLONE_STOPPED);
1390                 }
1391         }
1392
1393         /*
1394          * When called from kernel_thread, don't do user tracing stuff.
1395          */
1396         if (likely(user_mode(regs)))
1397                 trace = tracehook_prepare_clone(clone_flags);
1398
1399         p = copy_process(clone_flags, stack_start, regs, stack_size,
1400                          child_tidptr, NULL, trace);
1401         /*
1402          * Do this prior waking up the new thread - the thread pointer
1403          * might get invalid after that point, if the thread exits quickly.
1404          */
1405         if (!IS_ERR(p)) {
1406                 struct completion vfork;
1407
1408                 trace_sched_process_fork(current, p);
1409
1410                 nr = task_pid_vnr(p);
1411
1412                 if (clone_flags & CLONE_PARENT_SETTID)
1413                         put_user(nr, parent_tidptr);
1414
1415                 if (clone_flags & CLONE_VFORK) {
1416                         p->vfork_done = &vfork;
1417                         init_completion(&vfork);
1418                 }
1419
1420                 audit_finish_fork(p);
1421                 tracehook_report_clone(trace, regs, clone_flags, nr, p);
1422
1423                 /*
1424                  * We set PF_STARTING at creation in case tracing wants to
1425                  * use this to distinguish a fully live task from one that
1426                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1427                  * clear it and set the child going.
1428                  */
1429                 p->flags &= ~PF_STARTING;
1430
1431                 if (unlikely(clone_flags & CLONE_STOPPED)) {
1432                         /*
1433                          * We'll start up with an immediate SIGSTOP.
1434                          */
1435                         sigaddset(&p->pending.signal, SIGSTOP);
1436                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1437                         __set_task_state(p, TASK_STOPPED);
1438                 } else {
1439                         wake_up_new_task(p, clone_flags);
1440                 }
1441
1442                 tracehook_report_clone_complete(trace, regs,
1443                                                 clone_flags, nr, p);
1444
1445                 if (clone_flags & CLONE_VFORK) {
1446                         freezer_do_not_count();
1447                         wait_for_completion(&vfork);
1448                         freezer_count();
1449                         tracehook_report_vfork_done(p, nr);
1450                 }
1451         } else {
1452                 nr = PTR_ERR(p);
1453         }
1454         return nr;
1455 }
1456
1457 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1458 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1459 #endif
1460
1461 static void sighand_ctor(void *data)
1462 {
1463         struct sighand_struct *sighand = data;
1464
1465         spin_lock_init(&sighand->siglock);
1466         init_waitqueue_head(&sighand->signalfd_wqh);
1467 }
1468
1469 void __init proc_caches_init(void)
1470 {
1471         sighand_cachep = kmem_cache_create("sighand_cache",
1472                         sizeof(struct sighand_struct), 0,
1473                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1474                         sighand_ctor);
1475         signal_cachep = kmem_cache_create("signal_cache",
1476                         sizeof(struct signal_struct), 0,
1477                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1478         files_cachep = kmem_cache_create("files_cache",
1479                         sizeof(struct files_struct), 0,
1480                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1481         fs_cachep = kmem_cache_create("fs_cache",
1482                         sizeof(struct fs_struct), 0,
1483                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1484         mm_cachep = kmem_cache_create("mm_struct",
1485                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1486                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
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) &&
1543             (fs && atomic_read(&fs->count) > 1)) {
1544                 *new_fsp = __copy_fs_struct(current->fs);
1545                 if (!*new_fsp)
1546                         return -ENOMEM;
1547         }
1548
1549         return 0;
1550 }
1551
1552 /*
1553  * Unsharing of sighand is not supported yet
1554  */
1555 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1556 {
1557         struct sighand_struct *sigh = current->sighand;
1558
1559         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1560                 return -EINVAL;
1561         else
1562                 return 0;
1563 }
1564
1565 /*
1566  * Unshare vm if it is being shared
1567  */
1568 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1569 {
1570         struct mm_struct *mm = current->mm;
1571
1572         if ((unshare_flags & CLONE_VM) &&
1573             (mm && atomic_read(&mm->mm_users) > 1)) {
1574                 return -EINVAL;
1575         }
1576
1577         return 0;
1578 }
1579
1580 /*
1581  * Unshare file descriptor table if it is being shared
1582  */
1583 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1584 {
1585         struct files_struct *fd = current->files;
1586         int error = 0;
1587
1588         if ((unshare_flags & CLONE_FILES) &&
1589             (fd && atomic_read(&fd->count) > 1)) {
1590                 *new_fdp = dup_fd(fd, &error);
1591                 if (!*new_fdp)
1592                         return error;
1593         }
1594
1595         return 0;
1596 }
1597
1598 /*
1599  * unshare allows a process to 'unshare' part of the process
1600  * context which was originally shared using clone.  copy_*
1601  * functions used by do_fork() cannot be used here directly
1602  * because they modify an inactive task_struct that is being
1603  * constructed. Here we are modifying the current, active,
1604  * task_struct.
1605  */
1606 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1607 {
1608         int err = 0;
1609         struct fs_struct *fs, *new_fs = NULL;
1610         struct sighand_struct *new_sigh = NULL;
1611         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1612         struct files_struct *fd, *new_fd = NULL;
1613         struct nsproxy *new_nsproxy = NULL;
1614         int do_sysvsem = 0;
1615
1616         check_unshare_flags(&unshare_flags);
1617
1618         /* Return -EINVAL for all unsupported flags */
1619         err = -EINVAL;
1620         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1621                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1622                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1623                 goto bad_unshare_out;
1624
1625         /*
1626          * CLONE_NEWIPC must also detach from the undolist: after switching
1627          * to a new ipc namespace, the semaphore arrays from the old
1628          * namespace are unreachable.
1629          */
1630         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1631                 do_sysvsem = 1;
1632         if ((err = unshare_thread(unshare_flags)))
1633                 goto bad_unshare_out;
1634         if ((err = unshare_fs(unshare_flags, &new_fs)))
1635                 goto bad_unshare_cleanup_thread;
1636         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1637                 goto bad_unshare_cleanup_fs;
1638         if ((err = unshare_vm(unshare_flags, &new_mm)))
1639                 goto bad_unshare_cleanup_sigh;
1640         if ((err = unshare_fd(unshare_flags, &new_fd)))
1641                 goto bad_unshare_cleanup_vm;
1642         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1643                         new_fs)))
1644                 goto bad_unshare_cleanup_fd;
1645
1646         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1647                 if (do_sysvsem) {
1648                         /*
1649                          * CLONE_SYSVSEM is equivalent to sys_exit().
1650                          */
1651                         exit_sem(current);
1652                 }
1653
1654                 if (new_nsproxy) {
1655                         switch_task_namespaces(current, new_nsproxy);
1656                         new_nsproxy = NULL;
1657                 }
1658
1659                 task_lock(current);
1660
1661                 if (new_fs) {
1662                         fs = current->fs;
1663                         current->fs = new_fs;
1664                         new_fs = fs;
1665                 }
1666
1667                 if (new_mm) {
1668                         mm = current->mm;
1669                         active_mm = current->active_mm;
1670                         current->mm = new_mm;
1671                         current->active_mm = new_mm;
1672                         activate_mm(active_mm, new_mm);
1673                         new_mm = mm;
1674                 }
1675
1676                 if (new_fd) {
1677                         fd = current->files;
1678                         current->files = new_fd;
1679                         new_fd = fd;
1680                 }
1681
1682                 task_unlock(current);
1683         }
1684
1685         if (new_nsproxy)
1686                 put_nsproxy(new_nsproxy);
1687
1688 bad_unshare_cleanup_fd:
1689         if (new_fd)
1690                 put_files_struct(new_fd);
1691
1692 bad_unshare_cleanup_vm:
1693         if (new_mm)
1694                 mmput(new_mm);
1695
1696 bad_unshare_cleanup_sigh:
1697         if (new_sigh)
1698                 if (atomic_dec_and_test(&new_sigh->count))
1699                         kmem_cache_free(sighand_cachep, new_sigh);
1700
1701 bad_unshare_cleanup_fs:
1702         if (new_fs)
1703                 put_fs_struct(new_fs);
1704
1705 bad_unshare_cleanup_thread:
1706 bad_unshare_out:
1707         return err;
1708 }
1709
1710 /*
1711  *      Helper to unshare the files of the current task.
1712  *      We don't want to expose copy_files internals to
1713  *      the exec layer of the kernel.
1714  */
1715
1716 int unshare_files(struct files_struct **displaced)
1717 {
1718         struct task_struct *task = current;
1719         struct files_struct *copy = NULL;
1720         int error;
1721
1722         error = unshare_fd(CLONE_FILES, &copy);
1723         if (error || !copy) {
1724                 *displaced = NULL;
1725                 return error;
1726         }
1727         *displaced = task->files;
1728         task_lock(task);
1729         task->files = copy;
1730         task_unlock(task);
1731         return 0;
1732 }