kernel/fork.c

   1 /*
   2  *  linux/kernel/fork.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  *  'fork.c' contains the help-routines for the 'fork' system call
   9  * (see also entry.S and others).
  10  * Fork is rather simple, once you get the hang of it, but the memory
  11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  12  */
  13
  14 #include <linux/slab.h>
  15 #include <linux/init.h>
  16 #include <linux/unistd.h>
  17 #include <linux/module.h>
  18 #include <linux/vmalloc.h>
  19 #include <linux/completion.h>
  20 #include <linux/personality.h>
  21 #include <linux/mempolicy.h>
  22 #include <linux/sem.h>
  23 #include <linux/file.h>
  24 #include <linux/fdtable.h>
  25 #include <linux/iocontext.h>
  26 #include <linux/key.h>
  27 #include <linux/binfmts.h>
  28 #include <linux/mman.h>
  29 #include <linux/mmu_notifier.h>
  30 #include <linux/fs.h>
  31 #include <linux/nsproxy.h>
  32 #include <linux/capability.h>
  33 #include <linux/cpu.h>
  34 #include <linux/cgroup.h>
  35 #include <linux/security.h>
  36 #include <linux/hugetlb.h>
  37 #include <linux/swap.h>
  38 #include <linux/syscalls.h>
  39 #include <linux/jiffies.h>
  40 #include <linux/tracehook.h>
  41 #include <linux/futex.h>
  42 #include <linux/compat.h>
  43 #include <linux/task_io_accounting_ops.h>
  44 #include <linux/rcupdate.h>
  45 #include <linux/ptrace.h>
  46 #include <linux/mount.h>
  47 #include <linux/audit.h>
  48 #include <linux/memcontrol.h>
  49 #include <linux/ftrace.h>
  50 #include <linux/profile.h>
  51 #include <linux/rmap.h>
  52 #include <linux/ksm.h>
  53 #include <linux/acct.h>
  54 #include <linux/tsacct_kern.h>
  55 #include <linux/cn_proc.h>
  56 #include <linux/freezer.h>
  57 #include <linux/delayacct.h>
  58 #include <linux/taskstats_kern.h>
  59 #include <linux/random.h>
  60 #include <linux/tty.h>
  61 #include <linux/proc_fs.h>
  62 #include <linux/blkdev.h>
  63 #include <linux/fs_struct.h>
  64 #include <linux/magic.h>
  65 #include <linux/perf_event.h>
  66 #include <linux/posix-timers.h>
  67
  68 #include <asm/pgtable.h>
  69 #include <asm/pgalloc.h>
  70 #include <asm/uaccess.h>
  71 #include <asm/mmu_context.h>
  72 #include <asm/cacheflush.h>
  73 #include <asm/tlbflush.h>
  74
  75 #include <trace/events/sched.h>
  76
  77 /*
  78  * Protected counters by write_lock_irq(&tasklist_lock)
  79  */
  80 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  81 int nr_threads;                 /* The idle threads do not count.. */
  82
  83 int max_threads;                /* tunable limit on nr_threads */
  84
  85 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  86
  87 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  88
  89 int nr_processes(void)
  90 {
  91         int cpu;
  92         int total = 0;
  93
  94         for_each_possible_cpu(cpu)
  95                 total += per_cpu(process_counts, cpu);
  96
  97         return total;
  98 }
  99
 100 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 101 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
 102 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
 103 static struct kmem_cache *task_struct_cachep;
 104 #endif
 105
 106 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
 107 static inline struct thread_info *alloc_thread_info(struct task_struct *tsk)
 108 {
 109 #ifdef CONFIG_DEBUG_STACK_USAGE
 110         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
 111 #else
 112         gfp_t mask = GFP_KERNEL;
 113 #endif
 114         return (struct thread_info *)__get_free_pages(mask, THREAD_SIZE_ORDER);
 115 }
 116
 117 static inline void free_thread_info(struct thread_info *ti)
 118 {
 119         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 120 }
 121 #endif
 122
 123 /* SLAB cache for signal_struct structures (tsk->signal) */
 124 static struct kmem_cache *signal_cachep;
 125
 126 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 127 struct kmem_cache *sighand_cachep;
 128
 129 /* SLAB cache for files_struct structures (tsk->files) */
 130 struct kmem_cache *files_cachep;
 131
 132 /* SLAB cache for fs_struct structures (tsk->fs) */
 133 struct kmem_cache *fs_cachep;
 134
 135 /* SLAB cache for vm_area_struct structures */
 136 struct kmem_cache *vm_area_cachep;
 137
 138 /* SLAB cache for mm_struct structures (tsk->mm) */
 139 static struct kmem_cache *mm_cachep;
 140
 141 static void account_kernel_stack(struct thread_info *ti, int account)
 142 {
 143         struct zone *zone = page_zone(virt_to_page(ti));
 144
 145         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
 146 }
 147
 148 void free_task(struct task_struct *tsk)
 149 {
 150         prop_local_destroy_single(&tsk->dirties);
 151         account_kernel_stack(tsk->stack, -1);
 152         free_thread_info(tsk->stack);
 153         rt_mutex_debug_task_free(tsk);
 154         ftrace_graph_exit_task(tsk);
 155         free_task_struct(tsk);
 156 }
 157 EXPORT_SYMBOL(free_task);
 158
 159 void __put_task_struct(struct task_struct *tsk)
 160 {
 161         WARN_ON(!tsk->exit_state);
 162         WARN_ON(atomic_read(&tsk->usage));
 163         WARN_ON(tsk == current);
 164
 165         exit_creds(tsk);
 166         delayacct_tsk_free(tsk);
 167
 168         if (!profile_handoff_task(tsk))
 169                 free_task(tsk);
 170 }
 171
 172 /*
 173  * macro override instead of weak attribute alias, to workaround
 174  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
 175  */
 176 #ifndef arch_task_cache_init
 177 #define arch_task_cache_init()
 178 #endif
 179
 180 void __init fork_init(unsigned long mempages)
 181 {
 182 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 183 #ifndef ARCH_MIN_TASKALIGN
 184 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 185 #endif
 186         /* create a slab on which task_structs can be allocated */
 187         task_struct_cachep =
 188                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 189                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
 190 #endif
 191
 192         /* do the arch specific task caches init */
 193         arch_task_cache_init();
 194
 195         /*
 196          * The default maximum number of threads is set to a safe
 197          * value: the thread structures can take up at most half
 198          * of memory.
 199          */
 200         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 201
 202         /*
 203          * we need to allow at least 20 threads to boot a system
 204          */
 205         if(max_threads < 20)
 206                 max_threads = 20;
 207
 208         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 209         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 210         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 211                 init_task.signal->rlim[RLIMIT_NPROC];
 212 }
 213
 214 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
 215                                                struct task_struct *src)
 216 {
 217         *dst = *src;
 218         return 0;
 219 }
 220
 221 static struct task_struct *dup_task_struct(struct task_struct *orig)
 222 {
 223         struct task_struct *tsk;
 224         struct thread_info *ti;
 225         unsigned long *stackend;
 226
 227         int err;
 228
 229         prepare_to_copy(orig);
 230
 231         tsk = alloc_task_struct();
 232         if (!tsk)
 233                 return NULL;
 234
 235         ti = alloc_thread_info(tsk);
 236         if (!ti) {
 237                 free_task_struct(tsk);
 238                 return NULL;
 239         }
 240
 241         err = arch_dup_task_struct(tsk, orig);
 242         if (err)
 243                 goto out;
 244
 245         tsk->stack = ti;
 246
 247         err = prop_local_init_single(&tsk->dirties);
 248         if (err)
 249                 goto out;
 250
 251         setup_thread_stack(tsk, orig);
 252         stackend = end_of_stack(tsk);
 253         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 254
 255 #ifdef CONFIG_CC_STACKPROTECTOR
 256         tsk->stack_canary = get_random_int();
 257 #endif
 258
 259         /* One for us, one for whoever does the "release_task()" (usually parent) */
 260         atomic_set(&tsk->usage,2);
 261         atomic_set(&tsk->fs_excl, 0);
 262 #ifdef CONFIG_BLK_DEV_IO_TRACE
 263         tsk->btrace_seq = 0;
 264 #endif
 265         tsk->splice_pipe = NULL;
 266
 267         account_kernel_stack(ti, 1);
 268
 269         return tsk;
 270
 271 out:
 272         free_thread_info(ti);
 273         free_task_struct(tsk);
 274         return NULL;
 275 }
 276
 277 #ifdef CONFIG_MMU
 278 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 279 {
 280         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 281         struct rb_node **rb_link, *rb_parent;
 282         int retval;
 283         unsigned long charge;
 284         struct mempolicy *pol;
 285
 286         down_write(&oldmm->mmap_sem);
 287         flush_cache_dup_mm(oldmm);
 288         /*
 289          * Not linked in yet - no deadlock potential:
 290          */
 291         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 292
 293         mm->locked_vm = 0;
 294         mm->mmap = NULL;
 295         mm->mmap_cache = NULL;
 296         mm->free_area_cache = oldmm->mmap_base;
 297         mm->cached_hole_size = ~0UL;
 298         mm->map_count = 0;
 299         cpumask_clear(mm_cpumask(mm));
 300         mm->mm_rb = RB_ROOT;
 301         rb_link = &mm->mm_rb.rb_node;
 302         rb_parent = NULL;
 303         pprev = &mm->mmap;
 304         retval = ksm_fork(mm, oldmm);
 305         if (retval)
 306                 goto out;
 307
 308         prev = NULL;
 309         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 310                 struct file *file;
 311
 312                 if (mpnt->vm_flags & VM_DONTCOPY) {
 313                         long pages = vma_pages(mpnt);
 314                         mm->total_vm -= pages;
 315                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 316                                                                 -pages);
 317                         continue;
 318                 }
 319                 charge = 0;
 320                 if (mpnt->vm_flags & VM_ACCOUNT) {
 321                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 322                         if (security_vm_enough_memory(len))
 323                                 goto fail_nomem;
 324                         charge = len;
 325                 }
 326                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 327                 if (!tmp)
 328                         goto fail_nomem;
 329                 *tmp = *mpnt;
 330                 pol = mpol_dup(vma_policy(mpnt));
 331                 retval = PTR_ERR(pol);
 332                 if (IS_ERR(pol))
 333                         goto fail_nomem_policy;
 334                 vma_set_policy(tmp, pol);
 335                 tmp->vm_flags &= ~VM_LOCKED;
 336                 tmp->vm_mm = mm;
 337                 tmp->vm_next = tmp->vm_prev = NULL;
 338                 anon_vma_link(tmp);
 339                 file = tmp->vm_file;
 340                 if (file) {
 341                         struct inode *inode = file->f_path.dentry->d_inode;
 342                         struct address_space *mapping = file->f_mapping;
 343
 344                         get_file(file);
 345                         if (tmp->vm_flags & VM_DENYWRITE)
 346                                 atomic_dec(&inode->i_writecount);
 347                         spin_lock(&mapping->i_mmap_lock);
 348                         if (tmp->vm_flags & VM_SHARED)
 349                                 mapping->i_mmap_writable++;
 350                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
 351                         flush_dcache_mmap_lock(mapping);
 352                         /* insert tmp into the share list, just after mpnt */
 353                         vma_prio_tree_add(tmp, mpnt);
 354                         flush_dcache_mmap_unlock(mapping);
 355                         spin_unlock(&mapping->i_mmap_lock);
 356                 }
 357
 358                 /*
 359                  * Clear hugetlb-related page reserves for children. This only
 360                  * affects MAP_PRIVATE mappings. Faults generated by the child
 361                  * are not guaranteed to succeed, even if read-only
 362                  */
 363                 if (is_vm_hugetlb_page(tmp))
 364                         reset_vma_resv_huge_pages(tmp);
 365
 366                 /*
 367                  * Link in the new vma and copy the page table entries.
 368                  */
 369                 *pprev = tmp;
 370                 pprev = &tmp->vm_next;
 371                 tmp->vm_prev = prev;
 372                 prev = tmp;
 373
 374                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 375                 rb_link = &tmp->vm_rb.rb_right;
 376                 rb_parent = &tmp->vm_rb;
 377
 378                 mm->map_count++;
 379                 retval = copy_page_range(mm, oldmm, mpnt);
 380
 381                 if (tmp->vm_ops && tmp->vm_ops->open)
 382                         tmp->vm_ops->open(tmp);
 383
 384                 if (retval)
 385                         goto out;
 386         }
 387         /* a new mm has just been created */
 388         arch_dup_mmap(oldmm, mm);
 389         retval = 0;
 390 out:
 391         up_write(&mm->mmap_sem);
 392         flush_tlb_mm(oldmm);
 393         up_write(&oldmm->mmap_sem);
 394         return retval;
 395 fail_nomem_policy:
 396         kmem_cache_free(vm_area_cachep, tmp);
 397 fail_nomem:
 398         retval = -ENOMEM;
 399         vm_unacct_memory(charge);
 400         goto out;
 401 }
 402
 403 static inline int mm_alloc_pgd(struct mm_struct * mm)
 404 {
 405         mm->pgd = pgd_alloc(mm);
 406         if (unlikely(!mm->pgd))
 407                 return -ENOMEM;
 408         return 0;
 409 }
 410
 411 static inline void mm_free_pgd(struct mm_struct * mm)
 412 {
 413         pgd_free(mm, mm->pgd);
 414 }
 415 #else
 416 #define dup_mmap(mm, oldmm)     (0)
 417 #define mm_alloc_pgd(mm)        (0)
 418 #define mm_free_pgd(mm)
 419 #endif /* CONFIG_MMU */
 420
 421 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 422
 423 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 424 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 425
 426 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 427
 428 static int __init coredump_filter_setup(char *s)
 429 {
 430         default_dump_filter =
 431                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 432                 MMF_DUMP_FILTER_MASK;
 433         return 1;
 434 }
 435
 436 __setup("coredump_filter=", coredump_filter_setup);
 437
 438 #include <linux/init_task.h>
 439
 440 static void mm_init_aio(struct mm_struct *mm)
 441 {
 442 #ifdef CONFIG_AIO
 443         spin_lock_init(&mm->ioctx_lock);
 444         INIT_HLIST_HEAD(&mm->ioctx_list);
 445 #endif
 446 }
 447
 448 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
 449 {
 450         atomic_set(&mm->mm_users, 1);
 451         atomic_set(&mm->mm_count, 1);
 452         init_rwsem(&mm->mmap_sem);
 453         INIT_LIST_HEAD(&mm->mmlist);
 454         mm->flags = (current->mm) ?
 455                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
 456         mm->core_state = NULL;
 457         mm->nr_ptes = 0;
 458         set_mm_counter(mm, file_rss, 0);
 459         set_mm_counter(mm, anon_rss, 0);
 460         spin_lock_init(&mm->page_table_lock);
 461         mm->free_area_cache = TASK_UNMAPPED_BASE;
 462         mm->cached_hole_size = ~0UL;
 463         mm_init_aio(mm);
 464         mm_init_owner(mm, p);
 465
 466         if (likely(!mm_alloc_pgd(mm))) {
 467                 mm->def_flags = 0;
 468                 mmu_notifier_mm_init(mm);
 469                 return mm;
 470         }
 471
 472         free_mm(mm);
 473         return NULL;
 474 }
 475
 476 /*
 477  * Allocate and initialize an mm_struct.
 478  */
 479 struct mm_struct * mm_alloc(void)
 480 {
 481         struct mm_struct * mm;
 482
 483         mm = allocate_mm();
 484         if (mm) {
 485                 memset(mm, 0, sizeof(*mm));
 486                 mm = mm_init(mm, current);
 487         }
 488         return mm;
 489 }
 490
 491 /*
 492  * Called when the last reference to the mm
 493  * is dropped: either by a lazy thread or by
 494  * mmput. Free the page directory and the mm.
 495  */
 496 void __mmdrop(struct mm_struct *mm)
 497 {
 498         BUG_ON(mm == &init_mm);
 499         mm_free_pgd(mm);
 500         destroy_context(mm);
 501         mmu_notifier_mm_destroy(mm);
 502         free_mm(mm);
 503 }
 504 EXPORT_SYMBOL_GPL(__mmdrop);
 505
 506 /*
 507  * Decrement the use count and release all resources for an mm.
 508  */
 509 void mmput(struct mm_struct *mm)
 510 {
 511         might_sleep();
 512
 513         if (atomic_dec_and_test(&mm->mm_users)) {
 514                 exit_aio(mm);
 515                 ksm_exit(mm);
 516                 exit_mmap(mm);
 517                 set_mm_exe_file(mm, NULL);
 518                 if (!list_empty(&mm->mmlist)) {
 519                         spin_lock(&mmlist_lock);
 520                         list_del(&mm->mmlist);
 521                         spin_unlock(&mmlist_lock);
 522                 }
 523                 put_swap_token(mm);
 524                 if (mm->binfmt)
 525                         module_put(mm->binfmt->module);
 526                 mmdrop(mm);
 527         }
 528 }
 529 EXPORT_SYMBOL_GPL(mmput);
 530
 531 /**
 532  * get_task_mm - acquire a reference to the task's mm
 533  *
 534  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 535  * this kernel workthread has transiently adopted a user mm with use_mm,
 536  * to do its AIO) is not set and if so returns a reference to it, after
 537  * bumping up the use count.  User must release the mm via mmput()
 538  * after use.  Typically used by /proc and ptrace.
 539  */
 540 struct mm_struct *get_task_mm(struct task_struct *task)
 541 {
 542         struct mm_struct *mm;
 543
 544         task_lock(task);
 545         mm = task->mm;
 546         if (mm) {
 547                 if (task->flags & PF_KTHREAD)
 548                         mm = NULL;
 549                 else
 550                         atomic_inc(&mm->mm_users);
 551         }
 552         task_unlock(task);
 553         return mm;
 554 }
 555 EXPORT_SYMBOL_GPL(get_task_mm);
 556
 557 /* Please note the differences between mmput and mm_release.
 558  * mmput is called whenever we stop holding onto a mm_struct,
 559  * error success whatever.
 560  *
 561  * mm_release is called after a mm_struct has been removed
 562  * from the current process.
 563  *
 564  * This difference is important for error handling, when we
 565  * only half set up a mm_struct for a new process and need to restore
 566  * the old one.  Because we mmput the new mm_struct before
 567  * restoring the old one. . .
 568  * Eric Biederman 10 January 1998
 569  */
 570 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 571 {
 572         struct completion *vfork_done = tsk->vfork_done;
 573
 574         /* Get rid of any futexes when releasing the mm */
 575 #ifdef CONFIG_FUTEX
 576         if (unlikely(tsk->robust_list)) {
 577                 exit_robust_list(tsk);
 578                 tsk->robust_list = NULL;
 579         }
 580 #ifdef CONFIG_COMPAT
 581         if (unlikely(tsk->compat_robust_list)) {
 582                 compat_exit_robust_list(tsk);
 583                 tsk->compat_robust_list = NULL;
 584         }
 585 #endif
 586         if (unlikely(!list_empty(&tsk->pi_state_list)))
 587                 exit_pi_state_list(tsk);
 588 #endif
 589
 590         /* Get rid of any cached register state */
 591         deactivate_mm(tsk, mm);
 592
 593         /* notify parent sleeping on vfork() */
 594         if (vfork_done) {
 595                 tsk->vfork_done = NULL;
 596                 complete(vfork_done);
 597         }
 598
 599         /*
 600          * If we're exiting normally, clear a user-space tid field if
 601          * requested.  We leave this alone when dying by signal, to leave
 602          * the value intact in a core dump, and to save the unnecessary
 603          * trouble otherwise.  Userland only wants this done for a sys_exit.
 604          */
 605         if (tsk->clear_child_tid) {
 606                 if (!(tsk->flags & PF_SIGNALED) &&
 607                     atomic_read(&mm->mm_users) > 1) {
 608                         /*
 609                          * We don't check the error code - if userspace has
 610                          * not set up a proper pointer then tough luck.
 611                          */
 612                         put_user(0, tsk->clear_child_tid);
 613                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 614                                         1, NULL, NULL, 0);
 615                 }
 616                 tsk->clear_child_tid = NULL;
 617         }
 618 }
 619
 620 /*
 621  * Allocate a new mm structure and copy contents from the
 622  * mm structure of the passed in task structure.
 623  */
 624 struct mm_struct *dup_mm(struct task_struct *tsk)
 625 {
 626         struct mm_struct *mm, *oldmm = current->mm;
 627         int err;
 628
 629         if (!oldmm)
 630                 return NULL;
 631
 632         mm = allocate_mm();
 633         if (!mm)
 634                 goto fail_nomem;
 635
 636         memcpy(mm, oldmm, sizeof(*mm));
 637
 638         /* Initializing for Swap token stuff */
 639         mm->token_priority = 0;
 640         mm->last_interval = 0;
 641
 642         if (!mm_init(mm, tsk))
 643                 goto fail_nomem;
 644
 645         if (init_new_context(tsk, mm))
 646                 goto fail_nocontext;
 647
 648         dup_mm_exe_file(oldmm, mm);
 649
 650         err = dup_mmap(mm, oldmm);
 651         if (err)
 652                 goto free_pt;
 653
 654         mm->hiwater_rss = get_mm_rss(mm);
 655         mm->hiwater_vm = mm->total_vm;
 656
 657         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 658                 goto free_pt;
 659
 660         return mm;
 661
 662 free_pt:
 663         /* don't put binfmt in mmput, we haven't got module yet */
 664         mm->binfmt = NULL;
 665         mmput(mm);
 666
 667 fail_nomem:
 668         return NULL;
 669
 670 fail_nocontext:
 671         /*
 672          * If init_new_context() failed, we cannot use mmput() to free the mm
 673          * because it calls destroy_context()
 674          */
 675         mm_free_pgd(mm);
 676         free_mm(mm);
 677         return NULL;
 678 }
 679
 680 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 681 {
 682         struct mm_struct * mm, *oldmm;
 683         int retval;
 684
 685         tsk->min_flt = tsk->maj_flt = 0;
 686         tsk->nvcsw = tsk->nivcsw = 0;
 687 #ifdef CONFIG_DETECT_HUNG_TASK
 688         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 689 #endif
 690
 691         tsk->mm = NULL;
 692         tsk->active_mm = NULL;
 693
 694         /*
 695          * Are we cloning a kernel thread?
 696          *
 697          * We need to steal a active VM for that..
 698          */
 699         oldmm = current->mm;
 700         if (!oldmm)
 701                 return 0;
 702
 703         if (clone_flags & CLONE_VM) {
 704                 atomic_inc(&oldmm->mm_users);
 705                 mm = oldmm;
 706                 goto good_mm;
 707         }
 708
 709         retval = -ENOMEM;
 710         mm = dup_mm(tsk);
 711         if (!mm)
 712                 goto fail_nomem;
 713
 714 good_mm:
 715         /* Initializing for Swap token stuff */
 716         mm->token_priority = 0;
 717         mm->last_interval = 0;
 718
 719         tsk->mm = mm;
 720         tsk->active_mm = mm;
 721         return 0;
 722
 723 fail_nomem:
 724         return retval;
 725 }
 726
 727 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 728 {
 729         struct fs_struct *fs = current->fs;
 730         if (clone_flags & CLONE_FS) {
 731                 /* tsk->fs is already what we want */
 732                 write_lock(&fs->lock);
 733                 if (fs->in_exec) {
 734                         write_unlock(&fs->lock);
 735                         return -EAGAIN;
 736                 }
 737                 fs->users++;
 738                 write_unlock(&fs->lock);
 739                 return 0;
 740         }
 741         tsk->fs = copy_fs_struct(fs);
 742         if (!tsk->fs)
 743                 return -ENOMEM;
 744         return 0;
 745 }
 746
 747 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 748 {
 749         struct files_struct *oldf, *newf;
 750         int error = 0;
 751
 752         /*
 753          * A background process may not have any files ...
 754          */
 755         oldf = current->files;
 756         if (!oldf)
 757                 goto out;
 758
 759         if (clone_flags & CLONE_FILES) {
 760                 atomic_inc(&oldf->count);
 761                 goto out;
 762         }
 763
 764         newf = dup_fd(oldf, &error);
 765         if (!newf)
 766                 goto out;
 767
 768         tsk->files = newf;
 769         error = 0;
 770 out:
 771         return error;
 772 }
 773
 774 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 775 {
 776 #ifdef CONFIG_BLOCK
 777         struct io_context *ioc = current->io_context;
 778
 779         if (!ioc)
 780                 return 0;
 781         /*
 782          * Share io context with parent, if CLONE_IO is set
 783          */
 784         if (clone_flags & CLONE_IO) {
 785                 tsk->io_context = ioc_task_link(ioc);
 786                 if (unlikely(!tsk->io_context))
 787                         return -ENOMEM;
 788         } else if (ioprio_valid(ioc->ioprio)) {
 789                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
 790                 if (unlikely(!tsk->io_context))
 791                         return -ENOMEM;
 792
 793                 tsk->io_context->ioprio = ioc->ioprio;
 794         }
 795 #endif
 796         return 0;
 797 }
 798
 799 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 800 {
 801         struct sighand_struct *sig;
 802
 803         if (clone_flags & CLONE_SIGHAND) {
 804                 atomic_inc(&current->sighand->count);
 805                 return 0;
 806         }
 807         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 808         rcu_assign_pointer(tsk->sighand, sig);
 809         if (!sig)
 810                 return -ENOMEM;
 811         atomic_set(&sig->count, 1);
 812         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 813         return 0;
 814 }
 815
 816 void __cleanup_sighand(struct sighand_struct *sighand)
 817 {
 818         if (atomic_dec_and_test(&sighand->count))
 819                 kmem_cache_free(sighand_cachep, sighand);
 820 }
 821
 822
 823 /*
 824  * Initialize POSIX timer handling for a thread group.
 825  */
 826 static void posix_cpu_timers_init_group(struct signal_struct *sig)
 827 {
 828         /* Thread group counters. */
 829         thread_group_cputime_init(sig);
 830
 831         /* Expiration times and increments. */
 832         sig->it[CPUCLOCK_PROF].expires = cputime_zero;
 833         sig->it[CPUCLOCK_PROF].incr = cputime_zero;
 834         sig->it[CPUCLOCK_VIRT].expires = cputime_zero;
 835         sig->it[CPUCLOCK_VIRT].incr = cputime_zero;
 836
 837         /* Cached expiration times. */
 838         sig->cputime_expires.prof_exp = cputime_zero;
 839         sig->cputime_expires.virt_exp = cputime_zero;
 840         sig->cputime_expires.sched_exp = 0;
 841
 842         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
 843                 sig->cputime_expires.prof_exp =
 844                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
 845                 sig->cputimer.running = 1;
 846         }
 847
 848         /* The timer lists. */
 849         INIT_LIST_HEAD(&sig->cpu_timers[0]);
 850         INIT_LIST_HEAD(&sig->cpu_timers[1]);
 851         INIT_LIST_HEAD(&sig->cpu_timers[2]);
 852 }
 853
 854 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
 855 {
 856         struct signal_struct *sig;
 857
 858         if (clone_flags & CLONE_THREAD)
 859                 return 0;
 860
 861         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 862         tsk->signal = sig;
 863         if (!sig)
 864                 return -ENOMEM;
 865
 866         atomic_set(&sig->count, 1);
 867         atomic_set(&sig->live, 1);
 868         init_waitqueue_head(&sig->wait_chldexit);
 869         sig->flags = 0;
 870         if (clone_flags & CLONE_NEWPID)
 871                 sig->flags |= SIGNAL_UNKILLABLE;
 872         sig->group_exit_code = 0;
 873         sig->group_exit_task = NULL;
 874         sig->group_stop_count = 0;
 875         sig->curr_target = tsk;
 876         init_sigpending(&sig->shared_pending);
 877         INIT_LIST_HEAD(&sig->posix_timers);
 878
 879         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 880         sig->it_real_incr.tv64 = 0;
 881         sig->real_timer.function = it_real_fn;
 882
 883         sig->leader = 0;        /* session leadership doesn't inherit */
 884         sig->tty_old_pgrp = NULL;
 885         sig->tty = NULL;
 886
 887         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
 888         sig->gtime = cputime_zero;
 889         sig->cgtime = cputime_zero;
 890 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
 891         sig->prev_utime = sig->prev_stime = cputime_zero;
 892 #endif
 893         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
 894         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
 895         sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
 896         sig->maxrss = sig->cmaxrss = 0;
 897         task_io_accounting_init(&sig->ioac);
 898         sig->sum_sched_runtime = 0;
 899         taskstats_tgid_init(sig);
 900
 901         task_lock(current->group_leader);
 902         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
 903         task_unlock(current->group_leader);
 904
 905         posix_cpu_timers_init_group(sig);
 906
 907         acct_init_pacct(&sig->pacct);
 908
 909         tty_audit_fork(sig);
 910
 911         sig->oom_adj = current->signal->oom_adj;
 912
 913         return 0;
 914 }
 915
 916 void __cleanup_signal(struct signal_struct *sig)
 917 {
 918         thread_group_cputime_free(sig);
 919         tty_kref_put(sig->tty);
 920         kmem_cache_free(signal_cachep, sig);
 921 }
 922
 923 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
 924 {
 925         unsigned long new_flags = p->flags;
 926
 927         new_flags &= ~PF_SUPERPRIV;
 928         new_flags |= PF_FORKNOEXEC;
 929         new_flags |= PF_STARTING;
 930         p->flags = new_flags;
 931         clear_freeze_flag(p);
 932 }
 933
 934 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
 935 {
 936         current->clear_child_tid = tidptr;
 937
 938         return task_pid_vnr(current);
 939 }
 940
 941 static void rt_mutex_init_task(struct task_struct *p)
 942 {
 943         spin_lock_init(&p->pi_lock);
 944 #ifdef CONFIG_RT_MUTEXES
 945         plist_head_init(&p->pi_waiters, &p->pi_lock);
 946         p->pi_blocked_on = NULL;
 947 #endif
 948 }
 949
 950 #ifdef CONFIG_MM_OWNER
 951 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 952 {
 953         mm->owner = p;
 954 }
 955 #endif /* CONFIG_MM_OWNER */
 956
 957 /*
 958  * Initialize POSIX timer handling for a single task.
 959  */
 960 static void posix_cpu_timers_init(struct task_struct *tsk)
 961 {
 962         tsk->cputime_expires.prof_exp = cputime_zero;
 963         tsk->cputime_expires.virt_exp = cputime_zero;
 964         tsk->cputime_expires.sched_exp = 0;
 965         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
 966         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
 967         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
 968 }
 969
 970 /*
 971  * This creates a new process as a copy of the old one,
 972  * but does not actually start it yet.
 973  *
 974  * It copies the registers, and all the appropriate
 975  * parts of the process environment (as per the clone
 976  * flags). The actual kick-off is left to the caller.
 977  */
 978 static struct task_struct *copy_process(unsigned long clone_flags,
 979                                         unsigned long stack_start,
 980                                         struct pt_regs *regs,
 981                                         unsigned long stack_size,
 982                                         int __user *child_tidptr,
 983                                         struct pid *pid,
 984                                         int trace)
 985 {
 986         int retval;
 987         struct task_struct *p;
 988         int cgroup_callbacks_done = 0;
 989
 990         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 991                 return ERR_PTR(-EINVAL);
 992
 993         /*
 994          * Thread groups must share signals as well, and detached threads
 995          * can only be started up within the thread group.
 996          */
 997         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
 998                 return ERR_PTR(-EINVAL);
 999
1000         /*
1001          * Shared signal handlers imply shared VM. By way of the above,
1002          * thread groups also imply shared VM. Blocking this case allows
1003          * for various simplifications in other code.
1004          */
1005         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1006                 return ERR_PTR(-EINVAL);
1007
1008         /*
1009          * Siblings of global init remain as zombies on exit since they are
1010          * not reaped by their parent (swapper). To solve this and to avoid
1011          * multi-rooted process trees, prevent global and container-inits
1012          * from creating siblings.
1013          */
1014         if ((clone_flags & CLONE_PARENT) &&
1015                                 current->signal->flags & SIGNAL_UNKILLABLE)
1016                 return ERR_PTR(-EINVAL);
1017
1018         retval = security_task_create(clone_flags);
1019         if (retval)
1020                 goto fork_out;
1021
1022         retval = -ENOMEM;
1023         p = dup_task_struct(current);
1024         if (!p)
1025                 goto fork_out;
1026
1027         ftrace_graph_init_task(p);
1028
1029         rt_mutex_init_task(p);
1030
1031 #ifdef CONFIG_PROVE_LOCKING
1032         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1033         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1034 #endif
1035         retval = -EAGAIN;
1036         if (atomic_read(&p->real_cred->user->processes) >=
1037                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
1038                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1039                     p->real_cred->user != INIT_USER)
1040                         goto bad_fork_free;
1041         }
1042
1043         retval = copy_creds(p, clone_flags);
1044         if (retval < 0)
1045                 goto bad_fork_free;
1046
1047         /*
1048          * If multiple threads are within copy_process(), then this check
1049          * triggers too late. This doesn't hurt, the check is only there
1050          * to stop root fork bombs.
1051          */
1052         retval = -EAGAIN;
1053         if (nr_threads >= max_threads)
1054                 goto bad_fork_cleanup_count;
1055
1056         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1057                 goto bad_fork_cleanup_count;
1058
1059         p->did_exec = 0;
1060         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1061         copy_flags(clone_flags, p);
1062         INIT_LIST_HEAD(&p->children);
1063         INIT_LIST_HEAD(&p->sibling);
1064         rcu_copy_process(p);
1065         p->vfork_done = NULL;
1066         spin_lock_init(&p->alloc_lock);
1067
1068         init_sigpending(&p->pending);
1069
1070         p->utime = cputime_zero;
1071         p->stime = cputime_zero;
1072         p->gtime = cputime_zero;
1073         p->utimescaled = cputime_zero;
1074         p->stimescaled = cputime_zero;
1075         p->prev_utime = cputime_zero;
1076         p->prev_stime = cputime_zero;
1077
1078         p->default_timer_slack_ns = current->timer_slack_ns;
1079
1080         task_io_accounting_init(&p->ioac);
1081         acct_clear_integrals(p);
1082
1083         posix_cpu_timers_init(p);
1084
1085         p->lock_depth = -1;             /* -1 = no lock */
1086         do_posix_clock_monotonic_gettime(&p->start_time);
1087         p->real_start_time = p->start_time;
1088         monotonic_to_bootbased(&p->real_start_time);
1089         p->io_context = NULL;
1090         p->audit_context = NULL;
1091         cgroup_fork(p);
1092 #ifdef CONFIG_NUMA
1093         p->mempolicy = mpol_dup(p->mempolicy);
1094         if (IS_ERR(p->mempolicy)) {
1095                 retval = PTR_ERR(p->mempolicy);
1096                 p->mempolicy = NULL;
1097                 goto bad_fork_cleanup_cgroup;
1098         }
1099         mpol_fix_fork_child_flag(p);
1100 #endif
1101 #ifdef CONFIG_TRACE_IRQFLAGS
1102         p->irq_events = 0;
1103 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1104         p->hardirqs_enabled = 1;
1105 #else
1106         p->hardirqs_enabled = 0;
1107 #endif
1108         p->hardirq_enable_ip = 0;
1109         p->hardirq_enable_event = 0;
1110         p->hardirq_disable_ip = _THIS_IP_;
1111         p->hardirq_disable_event = 0;
1112         p->softirqs_enabled = 1;
1113         p->softirq_enable_ip = _THIS_IP_;
1114         p->softirq_enable_event = 0;
1115         p->softirq_disable_ip = 0;
1116         p->softirq_disable_event = 0;
1117         p->hardirq_context = 0;
1118         p->softirq_context = 0;
1119 #endif
1120 #ifdef CONFIG_LOCKDEP
1121         p->lockdep_depth = 0; /* no locks held yet */
1122         p->curr_chain_key = 0;
1123         p->lockdep_recursion = 0;
1124 #endif
1125
1126 #ifdef CONFIG_DEBUG_MUTEXES
1127         p->blocked_on = NULL; /* not blocked yet */
1128 #endif
1129
1130         p->bts = NULL;
1131
1132         /* Perform scheduler related setup. Assign this task to a CPU. */
1133         sched_fork(p, clone_flags);
1134
1135         retval = perf_event_init_task(p);
1136         if (retval)
1137                 goto bad_fork_cleanup_policy;
1138
1139         if ((retval = audit_alloc(p)))
1140                 goto bad_fork_cleanup_policy;
1141         /* copy all the process information */
1142         if ((retval = copy_semundo(clone_flags, p)))
1143                 goto bad_fork_cleanup_audit;
1144         if ((retval = copy_files(clone_flags, p)))
1145                 goto bad_fork_cleanup_semundo;
1146         if ((retval = copy_fs(clone_flags, p)))
1147                 goto bad_fork_cleanup_files;
1148         if ((retval = copy_sighand(clone_flags, p)))
1149                 goto bad_fork_cleanup_fs;
1150         if ((retval = copy_signal(clone_flags, p)))
1151                 goto bad_fork_cleanup_sighand;
1152         if ((retval = copy_mm(clone_flags, p)))
1153                 goto bad_fork_cleanup_signal;
1154         if ((retval = copy_namespaces(clone_flags, p)))
1155                 goto bad_fork_cleanup_mm;
1156         if ((retval = copy_io(clone_flags, p)))
1157                 goto bad_fork_cleanup_namespaces;
1158         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1159         if (retval)
1160                 goto bad_fork_cleanup_io;
1161
1162         if (pid != &init_struct_pid) {
1163                 retval = -ENOMEM;
1164                 pid = alloc_pid(p->nsproxy->pid_ns);
1165                 if (!pid)
1166                         goto bad_fork_cleanup_io;
1167
1168                 if (clone_flags & CLONE_NEWPID) {
1169                         retval = pid_ns_prepare_proc(p->nsproxy->pid_ns);
1170                         if (retval < 0)
1171                                 goto bad_fork_free_pid;
1172                 }
1173         }
1174
1175         p->pid = pid_nr(pid);
1176         p->tgid = p->pid;
1177         if (clone_flags & CLONE_THREAD)
1178                 p->tgid = current->tgid;
1179
1180         if (current->nsproxy != p->nsproxy) {
1181                 retval = ns_cgroup_clone(p, pid);
1182                 if (retval)
1183                         goto bad_fork_free_pid;
1184         }
1185
1186         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1187         /*
1188          * Clear TID on mm_release()?
1189          */
1190         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1191 #ifdef CONFIG_FUTEX
1192         p->robust_list = NULL;
1193 #ifdef CONFIG_COMPAT
1194         p->compat_robust_list = NULL;
1195 #endif
1196         INIT_LIST_HEAD(&p->pi_state_list);
1197         p->pi_state_cache = NULL;
1198 #endif
1199         /*
1200          * sigaltstack should be cleared when sharing the same VM
1201          */
1202         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1203                 p->sas_ss_sp = p->sas_ss_size = 0;
1204
1205         /*
1206          * Syscall tracing should be turned off in the child regardless
1207          * of CLONE_PTRACE.
1208          */
1209         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1210 #ifdef TIF_SYSCALL_EMU
1211         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1212 #endif
1213         clear_all_latency_tracing(p);
1214
1215         /* ok, now we should be set up.. */
1216         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1217         p->pdeath_signal = 0;
1218         p->exit_state = 0;
1219
1220         /*
1221          * Ok, make it visible to the rest of the system.
1222          * We dont wake it up yet.
1223          */
1224         p->group_leader = p;
1225         INIT_LIST_HEAD(&p->thread_group);
1226
1227         /* Now that the task is set up, run cgroup callbacks if
1228          * necessary. We need to run them before the task is visible
1229          * on the tasklist. */
1230         cgroup_fork_callbacks(p);
1231         cgroup_callbacks_done = 1;
1232
1233         /* Need tasklist lock for parent etc handling! */
1234         write_lock_irq(&tasklist_lock);
1235
1236         /* CLONE_PARENT re-uses the old parent */
1237         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1238                 p->real_parent = current->real_parent;
1239                 p->parent_exec_id = current->parent_exec_id;
1240         } else {
1241                 p->real_parent = current;
1242                 p->parent_exec_id = current->self_exec_id;
1243         }
1244
1245         spin_lock(&current->sighand->siglock);
1246
1247         /*
1248          * Process group and session signals need to be delivered to just the
1249          * parent before the fork or both the parent and the child after the
1250          * fork. Restart if a signal comes in before we add the new process to
1251          * it's process group.
1252          * A fatal signal pending means that current will exit, so the new
1253          * thread can't slip out of an OOM kill (or normal SIGKILL).
1254          */
1255         recalc_sigpending();
1256         if (signal_pending(current)) {
1257                 spin_unlock(&current->sighand->siglock);
1258                 write_unlock_irq(&tasklist_lock);
1259                 retval = -ERESTARTNOINTR;
1260                 goto bad_fork_free_pid;
1261         }
1262
1263         if (clone_flags & CLONE_THREAD) {
1264                 atomic_inc(&current->signal->count);
1265                 atomic_inc(&current->signal->live);
1266                 p->group_leader = current->group_leader;
1267                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1268         }
1269
1270         if (likely(p->pid)) {
1271                 list_add_tail(&p->sibling, &p->real_parent->children);
1272                 tracehook_finish_clone(p, clone_flags, trace);
1273
1274                 if (thread_group_leader(p)) {
1275                         if (clone_flags & CLONE_NEWPID)
1276                                 p->nsproxy->pid_ns->child_reaper = p;
1277
1278                         p->signal->leader_pid = pid;
1279                         tty_kref_put(p->signal->tty);
1280                         p->signal->tty = tty_kref_get(current->signal->tty);
1281                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1282                         attach_pid(p, PIDTYPE_SID, task_session(current));
1283                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1284                         __get_cpu_var(process_counts)++;
1285                 }
1286                 attach_pid(p, PIDTYPE_PID, pid);
1287                 nr_threads++;
1288         }
1289
1290         total_forks++;
1291         spin_unlock(&current->sighand->siglock);
1292         write_unlock_irq(&tasklist_lock);
1293         proc_fork_connector(p);
1294         cgroup_post_fork(p);
1295         perf_event_fork(p);
1296         return p;
1297
1298 bad_fork_free_pid:
1299         if (pid != &init_struct_pid)
1300                 free_pid(pid);
1301 bad_fork_cleanup_io:
1302         put_io_context(p->io_context);
1303 bad_fork_cleanup_namespaces:
1304         exit_task_namespaces(p);
1305 bad_fork_cleanup_mm:
1306         if (p->mm)
1307                 mmput(p->mm);
1308 bad_fork_cleanup_signal:
1309         if (!(clone_flags & CLONE_THREAD))
1310                 __cleanup_signal(p->signal);
1311 bad_fork_cleanup_sighand:
1312         __cleanup_sighand(p->sighand);
1313 bad_fork_cleanup_fs:
1314         exit_fs(p); /* blocking */
1315 bad_fork_cleanup_files:
1316         exit_files(p); /* blocking */
1317 bad_fork_cleanup_semundo:
1318         exit_sem(p);
1319 bad_fork_cleanup_audit:
1320         audit_free(p);
1321 bad_fork_cleanup_policy:
1322         perf_event_free_task(p);
1323 #ifdef CONFIG_NUMA
1324         mpol_put(p->mempolicy);
1325 bad_fork_cleanup_cgroup:
1326 #endif
1327         cgroup_exit(p, cgroup_callbacks_done);
1328         delayacct_tsk_free(p);
1329         module_put(task_thread_info(p)->exec_domain->module);
1330 bad_fork_cleanup_count:
1331         atomic_dec(&p->cred->user->processes);
1332         exit_creds(p);
1333 bad_fork_free:
1334         free_task(p);
1335 fork_out:
1336         return ERR_PTR(retval);
1337 }
1338
1339 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1340 {
1341         memset(regs, 0, sizeof(struct pt_regs));
1342         return regs;
1343 }
1344
1345 struct task_struct * __cpuinit fork_idle(int cpu)
1346 {
1347         struct task_struct *task;
1348         struct pt_regs regs;
1349
1350         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1351                             &init_struct_pid, 0);
1352         if (!IS_ERR(task))
1353                 init_idle(task, cpu);
1354
1355         return task;
1356 }
1357
1358 /*
1359  *  Ok, this is the main fork-routine.
1360  *
1361  * It copies the process, and if successful kick-starts
1362  * it and waits for it to finish using the VM if required.
1363  */
1364 long do_fork(unsigned long clone_flags,
1365               unsigned long stack_start,
1366               struct pt_regs *regs,
1367               unsigned long stack_size,
1368               int __user *parent_tidptr,
1369               int __user *child_tidptr)
1370 {
1371         struct task_struct *p;
1372         int trace = 0;
1373         long nr;
1374
1375         /*
1376          * Do some preliminary argument and permissions checking before we
1377          * actually start allocating stuff
1378          */
1379         if (clone_flags & CLONE_NEWUSER) {
1380                 if (clone_flags & CLONE_THREAD)
1381                         return -EINVAL;
1382                 /* hopefully this check will go away when userns support is
1383                  * complete
1384                  */
1385                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1386                                 !capable(CAP_SETGID))
1387                         return -EPERM;
1388         }
1389
1390         /*
1391          * We hope to recycle these flags after 2.6.26
1392          */
1393         if (unlikely(clone_flags & CLONE_STOPPED)) {
1394                 static int __read_mostly count = 100;
1395
1396                 if (count > 0 && printk_ratelimit()) {
1397                         char comm[TASK_COMM_LEN];
1398
1399                         count--;
1400                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1401                                         "clone flags 0x%lx\n",
1402                                 get_task_comm(comm, current),
1403                                 clone_flags & CLONE_STOPPED);
1404                 }
1405         }
1406
1407         /*
1408          * When called from kernel_thread, don't do user tracing stuff.
1409          */
1410         if (likely(user_mode(regs)))
1411                 trace = tracehook_prepare_clone(clone_flags);
1412
1413         p = copy_process(clone_flags, stack_start, regs, stack_size,
1414                          child_tidptr, NULL, trace);
1415         /*
1416          * Do this prior waking up the new thread - the thread pointer
1417          * might get invalid after that point, if the thread exits quickly.
1418          */
1419         if (!IS_ERR(p)) {
1420                 struct completion vfork;
1421
1422                 trace_sched_process_fork(current, p);
1423
1424                 nr = task_pid_vnr(p);
1425
1426                 if (clone_flags & CLONE_PARENT_SETTID)
1427                         put_user(nr, parent_tidptr);
1428
1429                 if (clone_flags & CLONE_VFORK) {
1430                         p->vfork_done = &vfork;
1431                         init_completion(&vfork);
1432                 }
1433
1434                 audit_finish_fork(p);
1435                 tracehook_report_clone(regs, clone_flags, nr, p);
1436
1437                 /*
1438                  * We set PF_STARTING at creation in case tracing wants to
1439                  * use this to distinguish a fully live task from one that
1440                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1441                  * clear it and set the child going.
1442                  */
1443                 p->flags &= ~PF_STARTING;
1444
1445                 if (unlikely(clone_flags & CLONE_STOPPED)) {
1446                         /*
1447                          * We'll start up with an immediate SIGSTOP.
1448                          */
1449                         sigaddset(&p->pending.signal, SIGSTOP);
1450                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1451                         __set_task_state(p, TASK_STOPPED);
1452                 } else {
1453                         wake_up_new_task(p, clone_flags);
1454                 }
1455
1456                 tracehook_report_clone_complete(trace, regs,
1457                                                 clone_flags, nr, p);
1458
1459                 if (clone_flags & CLONE_VFORK) {
1460                         freezer_do_not_count();
1461                         wait_for_completion(&vfork);
1462                         freezer_count();
1463                         tracehook_report_vfork_done(p, nr);
1464                 }
1465         } else {
1466                 nr = PTR_ERR(p);
1467         }
1468         return nr;
1469 }
1470
1471 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1472 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1473 #endif
1474
1475 static void sighand_ctor(void *data)
1476 {
1477         struct sighand_struct *sighand = data;
1478
1479         spin_lock_init(&sighand->siglock);
1480         init_waitqueue_head(&sighand->signalfd_wqh);
1481 }
1482
1483 void __init proc_caches_init(void)
1484 {
1485         sighand_cachep = kmem_cache_create("sighand_cache",
1486                         sizeof(struct sighand_struct), 0,
1487                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1488                         SLAB_NOTRACK, sighand_ctor);
1489         signal_cachep = kmem_cache_create("signal_cache",
1490                         sizeof(struct signal_struct), 0,
1491                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1492         files_cachep = kmem_cache_create("files_cache",
1493                         sizeof(struct files_struct), 0,
1494                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1495         fs_cachep = kmem_cache_create("fs_cache",
1496                         sizeof(struct fs_struct), 0,
1497                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1498         mm_cachep = kmem_cache_create("mm_struct",
1499                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1500                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1501         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1502         mmap_init();
1503 }
1504
1505 /*
1506  * Check constraints on flags passed to the unshare system call and
1507  * force unsharing of additional process context as appropriate.
1508  */
1509 static void check_unshare_flags(unsigned long *flags_ptr)
1510 {
1511         /*
1512          * If unsharing a thread from a thread group, must also
1513          * unshare vm.
1514          */
1515         if (*flags_ptr & CLONE_THREAD)
1516                 *flags_ptr |= CLONE_VM;
1517
1518         /*
1519          * If unsharing vm, must also unshare signal handlers.
1520          */
1521         if (*flags_ptr & CLONE_VM)
1522                 *flags_ptr |= CLONE_SIGHAND;
1523
1524         /*
1525          * If unsharing signal handlers and the task was created
1526          * using CLONE_THREAD, then must unshare the thread
1527          */
1528         if ((*flags_ptr & CLONE_SIGHAND) &&
1529             (atomic_read(&current->signal->count) > 1))
1530                 *flags_ptr |= CLONE_THREAD;
1531
1532         /*
1533          * If unsharing namespace, must also unshare filesystem information.
1534          */
1535         if (*flags_ptr & CLONE_NEWNS)
1536                 *flags_ptr |= CLONE_FS;
1537 }
1538
1539 /*
1540  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1541  */
1542 static int unshare_thread(unsigned long unshare_flags)
1543 {
1544         if (unshare_flags & CLONE_THREAD)
1545                 return -EINVAL;
1546
1547         return 0;
1548 }
1549
1550 /*
1551  * Unshare the filesystem structure if it is being shared
1552  */
1553 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1554 {
1555         struct fs_struct *fs = current->fs;
1556
1557         if (!(unshare_flags & CLONE_FS) || !fs)
1558                 return 0;
1559
1560         /* don't need lock here; in the worst case we'll do useless copy */
1561         if (fs->users == 1)
1562                 return 0;
1563
1564         *new_fsp = copy_fs_struct(fs);
1565         if (!*new_fsp)
1566                 return -ENOMEM;
1567
1568         return 0;
1569 }
1570
1571 /*
1572  * Unsharing of sighand is not supported yet
1573  */
1574 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1575 {
1576         struct sighand_struct *sigh = current->sighand;
1577
1578         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1579                 return -EINVAL;
1580         else
1581                 return 0;
1582 }
1583
1584 /*
1585  * Unshare vm if it is being shared
1586  */
1587 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1588 {
1589         struct mm_struct *mm = current->mm;
1590
1591         if ((unshare_flags & CLONE_VM) &&
1592             (mm && atomic_read(&mm->mm_users) > 1)) {
1593                 return -EINVAL;
1594         }
1595
1596         return 0;
1597 }
1598
1599 /*
1600  * Unshare file descriptor table if it is being shared
1601  */
1602 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1603 {
1604         struct files_struct *fd = current->files;
1605         int error = 0;
1606
1607         if ((unshare_flags & CLONE_FILES) &&
1608             (fd && atomic_read(&fd->count) > 1)) {
1609                 *new_fdp = dup_fd(fd, &error);
1610                 if (!*new_fdp)
1611                         return error;
1612         }
1613
1614         return 0;
1615 }
1616
1617 /*
1618  * unshare allows a process to 'unshare' part of the process
1619  * context which was originally shared using clone.  copy_*
1620  * functions used by do_fork() cannot be used here directly
1621  * because they modify an inactive task_struct that is being
1622  * constructed. Here we are modifying the current, active,
1623  * task_struct.
1624  */
1625 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1626 {
1627         int err = 0;
1628         struct fs_struct *fs, *new_fs = NULL;
1629         struct sighand_struct *new_sigh = NULL;
1630         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1631         struct files_struct *fd, *new_fd = NULL;
1632         struct nsproxy *new_nsproxy = NULL;
1633         int do_sysvsem = 0;
1634
1635         check_unshare_flags(&unshare_flags);
1636
1637         /* Return -EINVAL for all unsupported flags */
1638         err = -EINVAL;
1639         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1640                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1641                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1642                 goto bad_unshare_out;
1643
1644         /*
1645          * CLONE_NEWIPC must also detach from the undolist: after switching
1646          * to a new ipc namespace, the semaphore arrays from the old
1647          * namespace are unreachable.
1648          */
1649         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1650                 do_sysvsem = 1;
1651         if ((err = unshare_thread(unshare_flags)))
1652                 goto bad_unshare_out;
1653         if ((err = unshare_fs(unshare_flags, &new_fs)))
1654                 goto bad_unshare_cleanup_thread;
1655         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1656                 goto bad_unshare_cleanup_fs;
1657         if ((err = unshare_vm(unshare_flags, &new_mm)))
1658                 goto bad_unshare_cleanup_sigh;
1659         if ((err = unshare_fd(unshare_flags, &new_fd)))
1660                 goto bad_unshare_cleanup_vm;
1661         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1662                         new_fs)))
1663                 goto bad_unshare_cleanup_fd;
1664
1665         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1666                 if (do_sysvsem) {
1667                         /*
1668                          * CLONE_SYSVSEM is equivalent to sys_exit().
1669                          */
1670                         exit_sem(current);
1671                 }
1672
1673                 if (new_nsproxy) {
1674                         switch_task_namespaces(current, new_nsproxy);
1675                         new_nsproxy = NULL;
1676                 }
1677
1678                 task_lock(current);
1679
1680                 if (new_fs) {
1681                         fs = current->fs;
1682                         write_lock(&fs->lock);
1683                         current->fs = new_fs;
1684                         if (--fs->users)
1685                                 new_fs = NULL;
1686                         else
1687                                 new_fs = fs;
1688                         write_unlock(&fs->lock);
1689                 }
1690
1691                 if (new_mm) {
1692                         mm = current->mm;
1693                         active_mm = current->active_mm;
1694                         current->mm = new_mm;
1695                         current->active_mm = new_mm;
1696                         activate_mm(active_mm, new_mm);
1697                         new_mm = mm;
1698                 }
1699
1700                 if (new_fd) {
1701                         fd = current->files;
1702                         current->files = new_fd;
1703                         new_fd = fd;
1704                 }
1705
1706                 task_unlock(current);
1707         }
1708
1709         if (new_nsproxy)
1710                 put_nsproxy(new_nsproxy);
1711
1712 bad_unshare_cleanup_fd:
1713         if (new_fd)
1714                 put_files_struct(new_fd);
1715
1716 bad_unshare_cleanup_vm:
1717         if (new_mm)
1718                 mmput(new_mm);
1719
1720 bad_unshare_cleanup_sigh:
1721         if (new_sigh)
1722                 if (atomic_dec_and_test(&new_sigh->count))
1723                         kmem_cache_free(sighand_cachep, new_sigh);
1724
1725 bad_unshare_cleanup_fs:
1726         if (new_fs)
1727                 free_fs_struct(new_fs);
1728
1729 bad_unshare_cleanup_thread:
1730 bad_unshare_out:
1731         return err;
1732 }
1733
1734 /*
1735  *      Helper to unshare the files of the current task.
1736  *      We don't want to expose copy_files internals to
1737  *      the exec layer of the kernel.
1738  */
1739
1740 int unshare_files(struct files_struct **displaced)
1741 {
1742         struct task_struct *task = current;
1743         struct files_struct *copy = NULL;
1744         int error;
1745
1746         error = unshare_fd(CLONE_FILES, &copy);
1747         if (error || !copy) {
1748                 *displaced = NULL;
1749                 return error;
1750         }
1751         *displaced = task->files;
1752         task_lock(task);
1753         task->files = copy;
1754         task_unlock(task);
1755         return 0;
1756 }