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/export.h>
  18 #include <linux/vmalloc.h>
  19 #include <linux/completion.h>
  20 #include <linux/personality.h>
  21 #include <linux/mempolicy.h>
  22 #include <linux/module.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/futex.h>
  42 #include <linux/compat.h>
  43 #include <linux/kthread.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/ksm.h>
  54 #include <linux/acct.h>
  55 #include <linux/tsacct_kern.h>
  56 #include <linux/cn_proc.h>
  57 #include <linux/freezer.h>
  58 #include <linux/delayacct.h>
  59 #include <linux/taskstats_kern.h>
  60 #include <linux/random.h>
  61 #include <linux/tty.h>
  62 #include <linux/blkdev.h>
  63 #include <linux/fs_struct.h>
  64 #include <linux/magic.h>
  65 #include <linux/perf_event.h>
  66 #include <linux/posix-timers.h>
  67 #include <linux/user-return-notifier.h>
  68 #include <linux/oom.h>
  69 #include <linux/khugepaged.h>
  70
  71 #include <asm/pgtable.h>
  72 #include <asm/pgalloc.h>
  73 #include <asm/uaccess.h>
  74 #include <asm/mmu_context.h>
  75 #include <asm/cacheflush.h>
  76 #include <asm/tlbflush.h>
  77
  78 #include <trace/events/sched.h>
  79
  80 /*
  81  * Protected counters by write_lock_irq(&tasklist_lock)
  82  */
  83 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  84 int nr_threads;                 /* The idle threads do not count.. */
  85
  86 int max_threads;                /* tunable limit on nr_threads */
  87
  88 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  89
  90 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  91
  92 #ifdef CONFIG_PROVE_RCU
  93 int lockdep_tasklist_lock_is_held(void)
  94 {
  95         return lockdep_is_held(&tasklist_lock);
  96 }
  97 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
  98 #endif /* #ifdef CONFIG_PROVE_RCU */
  99
 100 int nr_processes(void)
 101 {
 102         int cpu;
 103         int total = 0;
 104
 105         for_each_possible_cpu(cpu)
 106                 total += per_cpu(process_counts, cpu);
 107
 108         return total;
 109 }
 110
 111 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 112 # define alloc_task_struct_node(node)           \
 113                 kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node)
 114 # define free_task_struct(tsk)                  \
 115                 kmem_cache_free(task_struct_cachep, (tsk))
 116 static struct kmem_cache *task_struct_cachep;
 117 #endif
 118
 119 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
 120 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 121                                                   int node)
 122 {
 123 #ifdef CONFIG_DEBUG_STACK_USAGE
 124         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
 125 #else
 126         gfp_t mask = GFP_KERNEL;
 127 #endif
 128         struct page *page = alloc_pages_node(node, mask, THREAD_SIZE_ORDER);
 129
 130         return page ? page_address(page) : NULL;
 131 }
 132
 133 static inline void free_thread_info(struct thread_info *ti)
 134 {
 135         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 136 }
 137 #endif
 138
 139 /* SLAB cache for signal_struct structures (tsk->signal) */
 140 static struct kmem_cache *signal_cachep;
 141
 142 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 143 struct kmem_cache *sighand_cachep;
 144
 145 /* SLAB cache for files_struct structures (tsk->files) */
 146 struct kmem_cache *files_cachep;
 147
 148 /* SLAB cache for fs_struct structures (tsk->fs) */
 149 struct kmem_cache *fs_cachep;
 150
 151 /* SLAB cache for vm_area_struct structures */
 152 struct kmem_cache *vm_area_cachep;
 153
 154 /* SLAB cache for mm_struct structures (tsk->mm) */
 155 static struct kmem_cache *mm_cachep;
 156
 157 static void account_kernel_stack(struct thread_info *ti, int account)
 158 {
 159         struct zone *zone = page_zone(virt_to_page(ti));
 160
 161         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
 162 }
 163
 164 void free_task(struct task_struct *tsk)
 165 {
 166         prop_local_destroy_single(&tsk->dirties);
 167         account_kernel_stack(tsk->stack, -1);
 168         free_thread_info(tsk->stack);
 169         rt_mutex_debug_task_free(tsk);
 170         ftrace_graph_exit_task(tsk);
 171         free_task_struct(tsk);
 172 }
 173 EXPORT_SYMBOL(free_task);
 174
 175 static inline void free_signal_struct(struct signal_struct *sig)
 176 {
 177         taskstats_tgid_free(sig);
 178         sched_autogroup_exit(sig);
 179         kmem_cache_free(signal_cachep, sig);
 180 }
 181
 182 static inline void put_signal_struct(struct signal_struct *sig)
 183 {
 184         if (atomic_dec_and_test(&sig->sigcnt))
 185                 free_signal_struct(sig);
 186 }
 187
 188 void __put_task_struct(struct task_struct *tsk)
 189 {
 190         WARN_ON(!tsk->exit_state);
 191         WARN_ON(atomic_read(&tsk->usage));
 192         WARN_ON(tsk == current);
 193
 194         exit_creds(tsk);
 195         delayacct_tsk_free(tsk);
 196         put_signal_struct(tsk->signal);
 197
 198         if (!profile_handoff_task(tsk))
 199                 free_task(tsk);
 200 }
 201 EXPORT_SYMBOL_GPL(__put_task_struct);
 202
 203 /*
 204  * macro override instead of weak attribute alias, to workaround
 205  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
 206  */
 207 #ifndef arch_task_cache_init
 208 #define arch_task_cache_init()
 209 #endif
 210
 211 void __init fork_init(unsigned long mempages)
 212 {
 213 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 214 #ifndef ARCH_MIN_TASKALIGN
 215 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 216 #endif
 217         /* create a slab on which task_structs can be allocated */
 218         task_struct_cachep =
 219                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 220                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
 221 #endif
 222
 223         /* do the arch specific task caches init */
 224         arch_task_cache_init();
 225
 226         /*
 227          * The default maximum number of threads is set to a safe
 228          * value: the thread structures can take up at most half
 229          * of memory.
 230          */
 231         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 232
 233         /*
 234          * we need to allow at least 20 threads to boot a system
 235          */
 236         if (max_threads < 20)
 237                 max_threads = 20;
 238
 239         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 240         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 241         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 242                 init_task.signal->rlim[RLIMIT_NPROC];
 243 }
 244
 245 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
 246                                                struct task_struct *src)
 247 {
 248         *dst = *src;
 249         return 0;
 250 }
 251
 252 static struct task_struct *dup_task_struct(struct task_struct *orig)
 253 {
 254         struct task_struct *tsk;
 255         struct thread_info *ti;
 256         unsigned long *stackend;
 257         int node = tsk_fork_get_node(orig);
 258         int err;
 259
 260         prepare_to_copy(orig);
 261
 262         tsk = alloc_task_struct_node(node);
 263         if (!tsk)
 264                 return NULL;
 265
 266         ti = alloc_thread_info_node(tsk, node);
 267         if (!ti) {
 268                 free_task_struct(tsk);
 269                 return NULL;
 270         }
 271
 272         err = arch_dup_task_struct(tsk, orig);
 273         if (err)
 274                 goto out;
 275
 276         tsk->stack = ti;
 277
 278         err = prop_local_init_single(&tsk->dirties);
 279         if (err)
 280                 goto out;
 281
 282         setup_thread_stack(tsk, orig);
 283         clear_user_return_notifier(tsk);
 284         clear_tsk_need_resched(tsk);
 285         stackend = end_of_stack(tsk);
 286         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 287
 288 #ifdef CONFIG_CC_STACKPROTECTOR
 289         tsk->stack_canary = get_random_int();
 290 #endif
 291
 292         /*
 293          * One for us, one for whoever does the "release_task()" (usually
 294          * parent)
 295          */
 296         atomic_set(&tsk->usage, 2);
 297 #ifdef CONFIG_BLK_DEV_IO_TRACE
 298         tsk->btrace_seq = 0;
 299 #endif
 300         tsk->splice_pipe = NULL;
 301
 302         account_kernel_stack(ti, 1);
 303
 304         return tsk;
 305
 306 out:
 307         free_thread_info(ti);
 308         free_task_struct(tsk);
 309         return NULL;
 310 }
 311
 312 #ifdef CONFIG_MMU
 313 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 314 {
 315         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 316         struct rb_node **rb_link, *rb_parent;
 317         int retval;
 318         unsigned long charge;
 319         struct mempolicy *pol;
 320
 321         down_write(&oldmm->mmap_sem);
 322         flush_cache_dup_mm(oldmm);
 323         /*
 324          * Not linked in yet - no deadlock potential:
 325          */
 326         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 327
 328         mm->locked_vm = 0;
 329         mm->mmap = NULL;
 330         mm->mmap_cache = NULL;
 331         mm->free_area_cache = oldmm->mmap_base;
 332         mm->cached_hole_size = ~0UL;
 333         mm->map_count = 0;
 334         cpumask_clear(mm_cpumask(mm));
 335         mm->mm_rb = RB_ROOT;
 336         rb_link = &mm->mm_rb.rb_node;
 337         rb_parent = NULL;
 338         pprev = &mm->mmap;
 339         retval = ksm_fork(mm, oldmm);
 340         if (retval)
 341                 goto out;
 342         retval = khugepaged_fork(mm, oldmm);
 343         if (retval)
 344                 goto out;
 345
 346         prev = NULL;
 347         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 348                 struct file *file;
 349
 350                 if (mpnt->vm_flags & VM_DONTCOPY) {
 351                         long pages = vma_pages(mpnt);
 352                         mm->total_vm -= pages;
 353                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 354                                                                 -pages);
 355                         continue;
 356                 }
 357                 charge = 0;
 358                 if (mpnt->vm_flags & VM_ACCOUNT) {
 359                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 360                         if (security_vm_enough_memory(len))
 361                                 goto fail_nomem;
 362                         charge = len;
 363                 }
 364                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 365                 if (!tmp)
 366                         goto fail_nomem;
 367                 *tmp = *mpnt;
 368                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 369                 pol = mpol_dup(vma_policy(mpnt));
 370                 retval = PTR_ERR(pol);
 371                 if (IS_ERR(pol))
 372                         goto fail_nomem_policy;
 373                 vma_set_policy(tmp, pol);
 374                 tmp->vm_mm = mm;
 375                 if (anon_vma_fork(tmp, mpnt))
 376                         goto fail_nomem_anon_vma_fork;
 377                 tmp->vm_flags &= ~VM_LOCKED;
 378                 tmp->vm_next = tmp->vm_prev = NULL;
 379                 file = tmp->vm_file;
 380                 if (file) {
 381                         struct inode *inode = file->f_path.dentry->d_inode;
 382                         struct address_space *mapping = file->f_mapping;
 383
 384                         get_file(file);
 385                         if (tmp->vm_flags & VM_DENYWRITE)
 386                                 atomic_dec(&inode->i_writecount);
 387                         mutex_lock(&mapping->i_mmap_mutex);
 388                         if (tmp->vm_flags & VM_SHARED)
 389                                 mapping->i_mmap_writable++;
 390                         flush_dcache_mmap_lock(mapping);
 391                         /* insert tmp into the share list, just after mpnt */
 392                         vma_prio_tree_add(tmp, mpnt);
 393                         flush_dcache_mmap_unlock(mapping);
 394                         mutex_unlock(&mapping->i_mmap_mutex);
 395                 }
 396
 397                 /*
 398                  * Clear hugetlb-related page reserves for children. This only
 399                  * affects MAP_PRIVATE mappings. Faults generated by the child
 400                  * are not guaranteed to succeed, even if read-only
 401                  */
 402                 if (is_vm_hugetlb_page(tmp))
 403                         reset_vma_resv_huge_pages(tmp);
 404
 405                 /*
 406                  * Link in the new vma and copy the page table entries.
 407                  */
 408                 *pprev = tmp;
 409                 pprev = &tmp->vm_next;
 410                 tmp->vm_prev = prev;
 411                 prev = tmp;
 412
 413                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 414                 rb_link = &tmp->vm_rb.rb_right;
 415                 rb_parent = &tmp->vm_rb;
 416
 417                 mm->map_count++;
 418                 retval = copy_page_range(mm, oldmm, mpnt);
 419
 420                 if (tmp->vm_ops && tmp->vm_ops->open)
 421                         tmp->vm_ops->open(tmp);
 422
 423                 if (retval)
 424                         goto out;
 425         }
 426         /* a new mm has just been created */
 427         arch_dup_mmap(oldmm, mm);
 428         retval = 0;
 429 out:
 430         up_write(&mm->mmap_sem);
 431         flush_tlb_mm(oldmm);
 432         up_write(&oldmm->mmap_sem);
 433         return retval;
 434 fail_nomem_anon_vma_fork:
 435         mpol_put(pol);
 436 fail_nomem_policy:
 437         kmem_cache_free(vm_area_cachep, tmp);
 438 fail_nomem:
 439         retval = -ENOMEM;
 440         vm_unacct_memory(charge);
 441         goto out;
 442 }
 443
 444 static inline int mm_alloc_pgd(struct mm_struct *mm)
 445 {
 446         mm->pgd = pgd_alloc(mm);
 447         if (unlikely(!mm->pgd))
 448                 return -ENOMEM;
 449         return 0;
 450 }
 451
 452 static inline void mm_free_pgd(struct mm_struct *mm)
 453 {
 454         pgd_free(mm, mm->pgd);
 455 }
 456 #else
 457 #define dup_mmap(mm, oldmm)     (0)
 458 #define mm_alloc_pgd(mm)        (0)
 459 #define mm_free_pgd(mm)
 460 #endif /* CONFIG_MMU */
 461
 462 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 463
 464 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 465 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 466
 467 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 468
 469 static int __init coredump_filter_setup(char *s)
 470 {
 471         default_dump_filter =
 472                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 473                 MMF_DUMP_FILTER_MASK;
 474         return 1;
 475 }
 476
 477 __setup("coredump_filter=", coredump_filter_setup);
 478
 479 #include <linux/init_task.h>
 480
 481 static void mm_init_aio(struct mm_struct *mm)
 482 {
 483 #ifdef CONFIG_AIO
 484         spin_lock_init(&mm->ioctx_lock);
 485         INIT_HLIST_HEAD(&mm->ioctx_list);
 486 #endif
 487 }
 488
 489 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
 490 {
 491         atomic_set(&mm->mm_users, 1);
 492         atomic_set(&mm->mm_count, 1);
 493         init_rwsem(&mm->mmap_sem);
 494         INIT_LIST_HEAD(&mm->mmlist);
 495         mm->flags = (current->mm) ?
 496                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
 497         mm->core_state = NULL;
 498         mm->nr_ptes = 0;
 499         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 500         spin_lock_init(&mm->page_table_lock);
 501         mm->free_area_cache = TASK_UNMAPPED_BASE;
 502         mm->cached_hole_size = ~0UL;
 503         mm_init_aio(mm);
 504         mm_init_owner(mm, p);
 505         atomic_set(&mm->oom_disable_count, 0);
 506
 507         if (likely(!mm_alloc_pgd(mm))) {
 508                 mm->def_flags = 0;
 509                 mmu_notifier_mm_init(mm);
 510                 return mm;
 511         }
 512
 513         free_mm(mm);
 514         return NULL;
 515 }
 516
 517 /*
 518  * Allocate and initialize an mm_struct.
 519  */
 520 struct mm_struct *mm_alloc(void)
 521 {
 522         struct mm_struct *mm;
 523
 524         mm = allocate_mm();
 525         if (!mm)
 526                 return NULL;
 527
 528         memset(mm, 0, sizeof(*mm));
 529         mm_init_cpumask(mm);
 530         return mm_init(mm, current);
 531 }
 532
 533 /*
 534  * Called when the last reference to the mm
 535  * is dropped: either by a lazy thread or by
 536  * mmput. Free the page directory and the mm.
 537  */
 538 void __mmdrop(struct mm_struct *mm)
 539 {
 540         BUG_ON(mm == &init_mm);
 541         mm_free_pgd(mm);
 542         destroy_context(mm);
 543         mmu_notifier_mm_destroy(mm);
 544 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 545         VM_BUG_ON(mm->pmd_huge_pte);
 546 #endif
 547         free_mm(mm);
 548 }
 549 EXPORT_SYMBOL_GPL(__mmdrop);
 550
 551 /*
 552  * Decrement the use count and release all resources for an mm.
 553  */
 554 void mmput(struct mm_struct *mm)
 555 {
 556         might_sleep();
 557
 558         if (atomic_dec_and_test(&mm->mm_users)) {
 559                 exit_aio(mm);
 560                 ksm_exit(mm);
 561                 khugepaged_exit(mm); /* must run before exit_mmap */
 562                 exit_mmap(mm);
 563                 set_mm_exe_file(mm, NULL);
 564                 if (!list_empty(&mm->mmlist)) {
 565                         spin_lock(&mmlist_lock);
 566                         list_del(&mm->mmlist);
 567                         spin_unlock(&mmlist_lock);
 568                 }
 569                 put_swap_token(mm);
 570                 if (mm->binfmt)
 571                         module_put(mm->binfmt->module);
 572                 mmdrop(mm);
 573         }
 574 }
 575 EXPORT_SYMBOL_GPL(mmput);
 576
 577 /*
 578  * We added or removed a vma mapping the executable. The vmas are only mapped
 579  * during exec and are not mapped with the mmap system call.
 580  * Callers must hold down_write() on the mm's mmap_sem for these
 581  */
 582 void added_exe_file_vma(struct mm_struct *mm)
 583 {
 584         mm->num_exe_file_vmas++;
 585 }
 586
 587 void removed_exe_file_vma(struct mm_struct *mm)
 588 {
 589         mm->num_exe_file_vmas--;
 590         if ((mm->num_exe_file_vmas == 0) && mm->exe_file) {
 591                 fput(mm->exe_file);
 592                 mm->exe_file = NULL;
 593         }
 594
 595 }
 596
 597 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 598 {
 599         if (new_exe_file)
 600                 get_file(new_exe_file);
 601         if (mm->exe_file)
 602                 fput(mm->exe_file);
 603         mm->exe_file = new_exe_file;
 604         mm->num_exe_file_vmas = 0;
 605 }
 606
 607 struct file *get_mm_exe_file(struct mm_struct *mm)
 608 {
 609         struct file *exe_file;
 610
 611         /* We need mmap_sem to protect against races with removal of
 612          * VM_EXECUTABLE vmas */
 613         down_read(&mm->mmap_sem);
 614         exe_file = mm->exe_file;
 615         if (exe_file)
 616                 get_file(exe_file);
 617         up_read(&mm->mmap_sem);
 618         return exe_file;
 619 }
 620
 621 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
 622 {
 623         /* It's safe to write the exe_file pointer without exe_file_lock because
 624          * this is called during fork when the task is not yet in /proc */
 625         newmm->exe_file = get_mm_exe_file(oldmm);
 626 }
 627
 628 /**
 629  * get_task_mm - acquire a reference to the task's mm
 630  *
 631  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 632  * this kernel workthread has transiently adopted a user mm with use_mm,
 633  * to do its AIO) is not set and if so returns a reference to it, after
 634  * bumping up the use count.  User must release the mm via mmput()
 635  * after use.  Typically used by /proc and ptrace.
 636  */
 637 struct mm_struct *get_task_mm(struct task_struct *task)
 638 {
 639         struct mm_struct *mm;
 640
 641         task_lock(task);
 642         mm = task->mm;
 643         if (mm) {
 644                 if (task->flags & PF_KTHREAD)
 645                         mm = NULL;
 646                 else
 647                         atomic_inc(&mm->mm_users);
 648         }
 649         task_unlock(task);
 650         return mm;
 651 }
 652 EXPORT_SYMBOL_GPL(get_task_mm);
 653
 654 static void complete_vfork_done(struct task_struct *tsk)
 655 {
 656         struct completion *vfork;
 657
 658         task_lock(tsk);
 659         vfork = tsk->vfork_done;
 660         if (likely(vfork)) {
 661                 tsk->vfork_done = NULL;
 662                 complete(vfork);
 663         }
 664         task_unlock(tsk);
 665 }
 666
 667 static int wait_for_vfork_done(struct task_struct *child,
 668                                 struct completion *vfork)
 669 {
 670         int killed;
 671
 672         freezer_do_not_count();
 673         killed = wait_for_completion_killable(vfork);
 674         freezer_count();
 675
 676         if (killed) {
 677                 task_lock(child);
 678                 child->vfork_done = NULL;
 679                 task_unlock(child);
 680         }
 681
 682         put_task_struct(child);
 683         return killed;
 684 }
 685
 686 /* Please note the differences between mmput and mm_release.
 687  * mmput is called whenever we stop holding onto a mm_struct,
 688  * error success whatever.
 689  *
 690  * mm_release is called after a mm_struct has been removed
 691  * from the current process.
 692  *
 693  * This difference is important for error handling, when we
 694  * only half set up a mm_struct for a new process and need to restore
 695  * the old one.  Because we mmput the new mm_struct before
 696  * restoring the old one. . .
 697  * Eric Biederman 10 January 1998
 698  */
 699 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 700 {
 701         /* Get rid of any futexes when releasing the mm */
 702 #ifdef CONFIG_FUTEX
 703         if (unlikely(tsk->robust_list)) {
 704                 exit_robust_list(tsk);
 705                 tsk->robust_list = NULL;
 706         }
 707 #ifdef CONFIG_COMPAT
 708         if (unlikely(tsk->compat_robust_list)) {
 709                 compat_exit_robust_list(tsk);
 710                 tsk->compat_robust_list = NULL;
 711         }
 712 #endif
 713         if (unlikely(!list_empty(&tsk->pi_state_list)))
 714                 exit_pi_state_list(tsk);
 715 #endif
 716
 717         /* Get rid of any cached register state */
 718         deactivate_mm(tsk, mm);
 719
 720         if (tsk->vfork_done)
 721                 complete_vfork_done(tsk);
 722
 723         /*
 724          * If we're exiting normally, clear a user-space tid field if
 725          * requested.  We leave this alone when dying by signal, to leave
 726          * the value intact in a core dump, and to save the unnecessary
 727          * trouble, say, a killed vfork parent shouldn't touch this mm.
 728          * Userland only wants this done for a sys_exit.
 729          */
 730         if (tsk->clear_child_tid) {
 731                 if (!(tsk->flags & PF_SIGNALED) &&
 732                     atomic_read(&mm->mm_users) > 1) {
 733                         /*
 734                          * We don't check the error code - if userspace has
 735                          * not set up a proper pointer then tough luck.
 736                          */
 737                         put_user(0, tsk->clear_child_tid);
 738                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 739                                         1, NULL, NULL, 0);
 740                 }
 741                 tsk->clear_child_tid = NULL;
 742         }
 743 }
 744
 745 /*
 746  * Allocate a new mm structure and copy contents from the
 747  * mm structure of the passed in task structure.
 748  */
 749 struct mm_struct *dup_mm(struct task_struct *tsk)
 750 {
 751         struct mm_struct *mm, *oldmm = current->mm;
 752         int err;
 753
 754         if (!oldmm)
 755                 return NULL;
 756
 757         mm = allocate_mm();
 758         if (!mm)
 759                 goto fail_nomem;
 760
 761         memcpy(mm, oldmm, sizeof(*mm));
 762         mm_init_cpumask(mm);
 763
 764         /* Initializing for Swap token stuff */
 765         mm->token_priority = 0;
 766         mm->last_interval = 0;
 767
 768 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 769         mm->pmd_huge_pte = NULL;
 770 #endif
 771
 772         if (!mm_init(mm, tsk))
 773                 goto fail_nomem;
 774
 775         if (init_new_context(tsk, mm))
 776                 goto fail_nocontext;
 777
 778         dup_mm_exe_file(oldmm, mm);
 779
 780         err = dup_mmap(mm, oldmm);
 781         if (err)
 782                 goto free_pt;
 783
 784         mm->hiwater_rss = get_mm_rss(mm);
 785         mm->hiwater_vm = mm->total_vm;
 786
 787         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 788                 goto free_pt;
 789
 790         return mm;
 791
 792 free_pt:
 793         /* don't put binfmt in mmput, we haven't got module yet */
 794         mm->binfmt = NULL;
 795         mmput(mm);
 796
 797 fail_nomem:
 798         return NULL;
 799
 800 fail_nocontext:
 801         /*
 802          * If init_new_context() failed, we cannot use mmput() to free the mm
 803          * because it calls destroy_context()
 804          */
 805         mm_free_pgd(mm);
 806         free_mm(mm);
 807         return NULL;
 808 }
 809
 810 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 811 {
 812         struct mm_struct *mm, *oldmm;
 813         int retval;
 814
 815         tsk->min_flt = tsk->maj_flt = 0;
 816         tsk->nvcsw = tsk->nivcsw = 0;
 817 #ifdef CONFIG_DETECT_HUNG_TASK
 818         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 819 #endif
 820
 821         tsk->mm = NULL;
 822         tsk->active_mm = NULL;
 823
 824         /*
 825          * Are we cloning a kernel thread?
 826          *
 827          * We need to steal a active VM for that..
 828          */
 829         oldmm = current->mm;
 830         if (!oldmm)
 831                 return 0;
 832
 833         if (clone_flags & CLONE_VM) {
 834                 atomic_inc(&oldmm->mm_users);
 835                 mm = oldmm;
 836                 goto good_mm;
 837         }
 838
 839         retval = -ENOMEM;
 840         mm = dup_mm(tsk);
 841         if (!mm)
 842                 goto fail_nomem;
 843
 844 good_mm:
 845         /* Initializing for Swap token stuff */
 846         mm->token_priority = 0;
 847         mm->last_interval = 0;
 848         if (tsk->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
 849                 atomic_inc(&mm->oom_disable_count);
 850
 851         tsk->mm = mm;
 852         tsk->active_mm = mm;
 853         return 0;
 854
 855 fail_nomem:
 856         return retval;
 857 }
 858
 859 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 860 {
 861         struct fs_struct *fs = current->fs;
 862         if (clone_flags & CLONE_FS) {
 863                 /* tsk->fs is already what we want */
 864                 spin_lock(&fs->lock);
 865                 if (fs->in_exec) {
 866                         spin_unlock(&fs->lock);
 867                         return -EAGAIN;
 868                 }
 869                 fs->users++;
 870                 spin_unlock(&fs->lock);
 871                 return 0;
 872         }
 873         tsk->fs = copy_fs_struct(fs);
 874         if (!tsk->fs)
 875                 return -ENOMEM;
 876         return 0;
 877 }
 878
 879 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
 880 {
 881         struct files_struct *oldf, *newf;
 882         int error = 0;
 883
 884         /*
 885          * A background process may not have any files ...
 886          */
 887         oldf = current->files;
 888         if (!oldf)
 889                 goto out;
 890
 891         if (clone_flags & CLONE_FILES) {
 892                 atomic_inc(&oldf->count);
 893                 goto out;
 894         }
 895
 896         newf = dup_fd(oldf, &error);
 897         if (!newf)
 898                 goto out;
 899
 900         tsk->files = newf;
 901         error = 0;
 902 out:
 903         return error;
 904 }
 905
 906 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 907 {
 908 #ifdef CONFIG_BLOCK
 909         struct io_context *ioc = current->io_context;
 910
 911         if (!ioc)
 912                 return 0;
 913         /*
 914          * Share io context with parent, if CLONE_IO is set
 915          */
 916         if (clone_flags & CLONE_IO) {
 917                 tsk->io_context = ioc_task_link(ioc);
 918                 if (unlikely(!tsk->io_context))
 919                         return -ENOMEM;
 920         } else if (ioprio_valid(ioc->ioprio)) {
 921                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
 922                 if (unlikely(!tsk->io_context))
 923                         return -ENOMEM;
 924
 925                 tsk->io_context->ioprio = ioc->ioprio;
 926         }
 927 #endif
 928         return 0;
 929 }
 930
 931 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 932 {
 933         struct sighand_struct *sig;
 934
 935         if (clone_flags & CLONE_SIGHAND) {
 936                 atomic_inc(&current->sighand->count);
 937                 return 0;
 938         }
 939         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 940         rcu_assign_pointer(tsk->sighand, sig);
 941         if (!sig)
 942                 return -ENOMEM;
 943         atomic_set(&sig->count, 1);
 944         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 945         return 0;
 946 }
 947
 948 void __cleanup_sighand(struct sighand_struct *sighand)
 949 {
 950         if (atomic_dec_and_test(&sighand->count))
 951                 kmem_cache_free(sighand_cachep, sighand);
 952 }
 953
 954
 955 /*
 956  * Initialize POSIX timer handling for a thread group.
 957  */
 958 static void posix_cpu_timers_init_group(struct signal_struct *sig)
 959 {
 960         unsigned long cpu_limit;
 961
 962         /* Thread group counters. */
 963         thread_group_cputime_init(sig);
 964
 965         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
 966         if (cpu_limit != RLIM_INFINITY) {
 967                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
 968                 sig->cputimer.running = 1;
 969         }
 970
 971         /* The timer lists. */
 972         INIT_LIST_HEAD(&sig->cpu_timers[0]);
 973         INIT_LIST_HEAD(&sig->cpu_timers[1]);
 974         INIT_LIST_HEAD(&sig->cpu_timers[2]);
 975 }
 976
 977 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
 978 {
 979         struct signal_struct *sig;
 980
 981         if (clone_flags & CLONE_THREAD)
 982                 return 0;
 983
 984         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
 985         tsk->signal = sig;
 986         if (!sig)
 987                 return -ENOMEM;
 988
 989         sig->nr_threads = 1;
 990         atomic_set(&sig->live, 1);
 991         atomic_set(&sig->sigcnt, 1);
 992         init_waitqueue_head(&sig->wait_chldexit);
 993         if (clone_flags & CLONE_NEWPID)
 994                 sig->flags |= SIGNAL_UNKILLABLE;
 995         sig->curr_target = tsk;
 996         init_sigpending(&sig->shared_pending);
 997         INIT_LIST_HEAD(&sig->posix_timers);
 998
 999         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1000         sig->real_timer.function = it_real_fn;
1001
1002         task_lock(current->group_leader);
1003         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1004         task_unlock(current->group_leader);
1005
1006         posix_cpu_timers_init_group(sig);
1007
1008         tty_audit_fork(sig);
1009         sched_autogroup_fork(sig);
1010
1011 #ifdef CONFIG_CGROUPS
1012         init_rwsem(&sig->threadgroup_fork_lock);
1013 #endif
1014
1015         sig->oom_adj = current->signal->oom_adj;
1016         sig->oom_score_adj = current->signal->oom_score_adj;
1017         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1018
1019         mutex_init(&sig->cred_guard_mutex);
1020
1021         return 0;
1022 }
1023
1024 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
1025 {
1026         unsigned long new_flags = p->flags;
1027
1028         new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1029         new_flags |= PF_FORKNOEXEC;
1030         p->flags = new_flags;
1031 }
1032
1033 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1034 {
1035         current->clear_child_tid = tidptr;
1036
1037         return task_pid_vnr(current);
1038 }
1039
1040 static void rt_mutex_init_task(struct task_struct *p)
1041 {
1042         raw_spin_lock_init(&p->pi_lock);
1043 #ifdef CONFIG_RT_MUTEXES
1044         plist_head_init(&p->pi_waiters);
1045         p->pi_blocked_on = NULL;
1046 #endif
1047 }
1048
1049 #ifdef CONFIG_MM_OWNER
1050 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1051 {
1052         mm->owner = p;
1053 }
1054 #endif /* CONFIG_MM_OWNER */
1055
1056 /*
1057  * Initialize POSIX timer handling for a single task.
1058  */
1059 static void posix_cpu_timers_init(struct task_struct *tsk)
1060 {
1061         tsk->cputime_expires.prof_exp = cputime_zero;
1062         tsk->cputime_expires.virt_exp = cputime_zero;
1063         tsk->cputime_expires.sched_exp = 0;
1064         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1065         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1066         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1067 }
1068
1069 /*
1070  * This creates a new process as a copy of the old one,
1071  * but does not actually start it yet.
1072  *
1073  * It copies the registers, and all the appropriate
1074  * parts of the process environment (as per the clone
1075  * flags). The actual kick-off is left to the caller.
1076  */
1077 static struct task_struct *copy_process(unsigned long clone_flags,
1078                                         unsigned long stack_start,
1079                                         struct pt_regs *regs,
1080                                         unsigned long stack_size,
1081                                         int __user *child_tidptr,
1082                                         struct pid *pid,
1083                                         int trace)
1084 {
1085         int retval;
1086         struct task_struct *p;
1087         int cgroup_callbacks_done = 0;
1088
1089         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1090                 return ERR_PTR(-EINVAL);
1091
1092         /*
1093          * Thread groups must share signals as well, and detached threads
1094          * can only be started up within the thread group.
1095          */
1096         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1097                 return ERR_PTR(-EINVAL);
1098
1099         /*
1100          * Shared signal handlers imply shared VM. By way of the above,
1101          * thread groups also imply shared VM. Blocking this case allows
1102          * for various simplifications in other code.
1103          */
1104         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1105                 return ERR_PTR(-EINVAL);
1106
1107         /*
1108          * Siblings of global init remain as zombies on exit since they are
1109          * not reaped by their parent (swapper). To solve this and to avoid
1110          * multi-rooted process trees, prevent global and container-inits
1111          * from creating siblings.
1112          */
1113         if ((clone_flags & CLONE_PARENT) &&
1114                                 current->signal->flags & SIGNAL_UNKILLABLE)
1115                 return ERR_PTR(-EINVAL);
1116
1117         retval = security_task_create(clone_flags);
1118         if (retval)
1119                 goto fork_out;
1120
1121         retval = -ENOMEM;
1122         p = dup_task_struct(current);
1123         if (!p)
1124                 goto fork_out;
1125
1126         ftrace_graph_init_task(p);
1127
1128         rt_mutex_init_task(p);
1129
1130 #ifdef CONFIG_PROVE_LOCKING
1131         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1132         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1133 #endif
1134         retval = -EAGAIN;
1135         if (atomic_read(&p->real_cred->user->processes) >=
1136                         task_rlimit(p, RLIMIT_NPROC)) {
1137                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1138                     p->real_cred->user != INIT_USER)
1139                         goto bad_fork_free;
1140         }
1141         current->flags &= ~PF_NPROC_EXCEEDED;
1142
1143         retval = copy_creds(p, clone_flags);
1144         if (retval < 0)
1145                 goto bad_fork_free;
1146
1147         /*
1148          * If multiple threads are within copy_process(), then this check
1149          * triggers too late. This doesn't hurt, the check is only there
1150          * to stop root fork bombs.
1151          */
1152         retval = -EAGAIN;
1153         if (nr_threads >= max_threads)
1154                 goto bad_fork_cleanup_count;
1155
1156         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1157                 goto bad_fork_cleanup_count;
1158
1159         p->did_exec = 0;
1160         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1161         copy_flags(clone_flags, p);
1162         INIT_LIST_HEAD(&p->children);
1163         INIT_LIST_HEAD(&p->sibling);
1164         rcu_copy_process(p);
1165         p->vfork_done = NULL;
1166         spin_lock_init(&p->alloc_lock);
1167
1168         init_sigpending(&p->pending);
1169
1170         p->utime = cputime_zero;
1171         p->stime = cputime_zero;
1172         p->gtime = cputime_zero;
1173         p->utimescaled = cputime_zero;
1174         p->stimescaled = cputime_zero;
1175 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1176         p->prev_utime = cputime_zero;
1177         p->prev_stime = cputime_zero;
1178 #endif
1179 #if defined(SPLIT_RSS_COUNTING)
1180         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1181 #endif
1182
1183         p->default_timer_slack_ns = current->timer_slack_ns;
1184
1185         task_io_accounting_init(&p->ioac);
1186         acct_clear_integrals(p);
1187
1188         posix_cpu_timers_init(p);
1189
1190         do_posix_clock_monotonic_gettime(&p->start_time);
1191         p->real_start_time = p->start_time;
1192         monotonic_to_bootbased(&p->real_start_time);
1193         p->io_context = NULL;
1194         p->audit_context = NULL;
1195         if (clone_flags & CLONE_THREAD)
1196                 threadgroup_fork_read_lock(current);
1197         cgroup_fork(p);
1198 #ifdef CONFIG_NUMA
1199         p->mempolicy = mpol_dup(p->mempolicy);
1200         if (IS_ERR(p->mempolicy)) {
1201                 retval = PTR_ERR(p->mempolicy);
1202                 p->mempolicy = NULL;
1203                 goto bad_fork_cleanup_cgroup;
1204         }
1205         mpol_fix_fork_child_flag(p);
1206 #endif
1207 #ifdef CONFIG_CPUSETS
1208         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1209         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1210 #endif
1211 #ifdef CONFIG_TRACE_IRQFLAGS
1212         p->irq_events = 0;
1213 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1214         p->hardirqs_enabled = 1;
1215 #else
1216         p->hardirqs_enabled = 0;
1217 #endif
1218         p->hardirq_enable_ip = 0;
1219         p->hardirq_enable_event = 0;
1220         p->hardirq_disable_ip = _THIS_IP_;
1221         p->hardirq_disable_event = 0;
1222         p->softirqs_enabled = 1;
1223         p->softirq_enable_ip = _THIS_IP_;
1224         p->softirq_enable_event = 0;
1225         p->softirq_disable_ip = 0;
1226         p->softirq_disable_event = 0;
1227         p->hardirq_context = 0;
1228         p->softirq_context = 0;
1229 #endif
1230 #ifdef CONFIG_LOCKDEP
1231         p->lockdep_depth = 0; /* no locks held yet */
1232         p->curr_chain_key = 0;
1233         p->lockdep_recursion = 0;
1234 #endif
1235
1236 #ifdef CONFIG_DEBUG_MUTEXES
1237         p->blocked_on = NULL; /* not blocked yet */
1238 #endif
1239 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1240         p->memcg_batch.do_batch = 0;
1241         p->memcg_batch.memcg = NULL;
1242 #endif
1243
1244         /* Perform scheduler related setup. Assign this task to a CPU. */
1245         sched_fork(p);
1246
1247         retval = perf_event_init_task(p);
1248         if (retval)
1249                 goto bad_fork_cleanup_policy;
1250         retval = audit_alloc(p);
1251         if (retval)
1252                 goto bad_fork_cleanup_policy;
1253         /* copy all the process information */
1254         retval = copy_semundo(clone_flags, p);
1255         if (retval)
1256                 goto bad_fork_cleanup_audit;
1257         retval = copy_files(clone_flags, p);
1258         if (retval)
1259                 goto bad_fork_cleanup_semundo;
1260         retval = copy_fs(clone_flags, p);
1261         if (retval)
1262                 goto bad_fork_cleanup_files;
1263         retval = copy_sighand(clone_flags, p);
1264         if (retval)
1265                 goto bad_fork_cleanup_fs;
1266         retval = copy_signal(clone_flags, p);
1267         if (retval)
1268                 goto bad_fork_cleanup_sighand;
1269         retval = copy_mm(clone_flags, p);
1270         if (retval)
1271                 goto bad_fork_cleanup_signal;
1272         retval = copy_namespaces(clone_flags, p);
1273         if (retval)
1274                 goto bad_fork_cleanup_mm;
1275         retval = copy_io(clone_flags, p);
1276         if (retval)
1277                 goto bad_fork_cleanup_namespaces;
1278         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1279         if (retval)
1280                 goto bad_fork_cleanup_io;
1281
1282         if (pid != &init_struct_pid) {
1283                 retval = -ENOMEM;
1284                 pid = alloc_pid(p->nsproxy->pid_ns);
1285                 if (!pid)
1286                         goto bad_fork_cleanup_io;
1287         }
1288
1289         p->pid = pid_nr(pid);
1290         p->tgid = p->pid;
1291         if (clone_flags & CLONE_THREAD)
1292                 p->tgid = current->tgid;
1293
1294         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1295         /*
1296          * Clear TID on mm_release()?
1297          */
1298         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1299 #ifdef CONFIG_BLOCK
1300         p->plug = NULL;
1301 #endif
1302 #ifdef CONFIG_FUTEX
1303         p->robust_list = NULL;
1304 #ifdef CONFIG_COMPAT
1305         p->compat_robust_list = NULL;
1306 #endif
1307         INIT_LIST_HEAD(&p->pi_state_list);
1308         p->pi_state_cache = NULL;
1309 #endif
1310         /*
1311          * sigaltstack should be cleared when sharing the same VM
1312          */
1313         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1314                 p->sas_ss_sp = p->sas_ss_size = 0;
1315
1316         /*
1317          * Syscall tracing and stepping should be turned off in the
1318          * child regardless of CLONE_PTRACE.
1319          */
1320         user_disable_single_step(p);
1321         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1322 #ifdef TIF_SYSCALL_EMU
1323         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1324 #endif
1325         clear_all_latency_tracing(p);
1326
1327         /* ok, now we should be set up.. */
1328         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1329         p->pdeath_signal = 0;
1330         p->exit_state = 0;
1331
1332         /*
1333          * Ok, make it visible to the rest of the system.
1334          * We dont wake it up yet.
1335          */
1336         p->group_leader = p;
1337         INIT_LIST_HEAD(&p->thread_group);
1338
1339         /* Now that the task is set up, run cgroup callbacks if
1340          * necessary. We need to run them before the task is visible
1341          * on the tasklist. */
1342         cgroup_fork_callbacks(p);
1343         cgroup_callbacks_done = 1;
1344
1345         /* Need tasklist lock for parent etc handling! */
1346         write_lock_irq(&tasklist_lock);
1347
1348         /* CLONE_PARENT re-uses the old parent */
1349         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1350                 p->real_parent = current->real_parent;
1351                 p->parent_exec_id = current->parent_exec_id;
1352         } else {
1353                 p->real_parent = current;
1354                 p->parent_exec_id = current->self_exec_id;
1355         }
1356
1357         spin_lock(&current->sighand->siglock);
1358
1359         /*
1360          * Process group and session signals need to be delivered to just the
1361          * parent before the fork or both the parent and the child after the
1362          * fork. Restart if a signal comes in before we add the new process to
1363          * it's process group.
1364          * A fatal signal pending means that current will exit, so the new
1365          * thread can't slip out of an OOM kill (or normal SIGKILL).
1366         */
1367         recalc_sigpending();
1368         if (signal_pending(current)) {
1369                 spin_unlock(&current->sighand->siglock);
1370                 write_unlock_irq(&tasklist_lock);
1371                 retval = -ERESTARTNOINTR;
1372                 goto bad_fork_free_pid;
1373         }
1374
1375         if (clone_flags & CLONE_THREAD) {
1376                 current->signal->nr_threads++;
1377                 atomic_inc(&current->signal->live);
1378                 atomic_inc(&current->signal->sigcnt);
1379                 p->group_leader = current->group_leader;
1380                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1381         }
1382
1383         if (likely(p->pid)) {
1384                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1385
1386                 if (thread_group_leader(p)) {
1387                         if (is_child_reaper(pid))
1388                                 p->nsproxy->pid_ns->child_reaper = p;
1389
1390                         p->signal->leader_pid = pid;
1391                         p->signal->tty = tty_kref_get(current->signal->tty);
1392                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1393                         attach_pid(p, PIDTYPE_SID, task_session(current));
1394                         list_add_tail(&p->sibling, &p->real_parent->children);
1395                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1396                         __this_cpu_inc(process_counts);
1397                 }
1398                 attach_pid(p, PIDTYPE_PID, pid);
1399                 nr_threads++;
1400         }
1401
1402         total_forks++;
1403         spin_unlock(&current->sighand->siglock);
1404         write_unlock_irq(&tasklist_lock);
1405         proc_fork_connector(p);
1406         cgroup_post_fork(p);
1407         if (clone_flags & CLONE_THREAD)
1408                 threadgroup_fork_read_unlock(current);
1409         perf_event_fork(p);
1410         return p;
1411
1412 bad_fork_free_pid:
1413         if (pid != &init_struct_pid)
1414                 free_pid(pid);
1415 bad_fork_cleanup_io:
1416         if (p->io_context)
1417                 exit_io_context(p);
1418 bad_fork_cleanup_namespaces:
1419         exit_task_namespaces(p);
1420 bad_fork_cleanup_mm:
1421         if (p->mm) {
1422                 task_lock(p);
1423                 if (p->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1424                         atomic_dec(&p->mm->oom_disable_count);
1425                 task_unlock(p);
1426                 mmput(p->mm);
1427         }
1428 bad_fork_cleanup_signal:
1429         if (!(clone_flags & CLONE_THREAD))
1430                 free_signal_struct(p->signal);
1431 bad_fork_cleanup_sighand:
1432         __cleanup_sighand(p->sighand);
1433 bad_fork_cleanup_fs:
1434         exit_fs(p); /* blocking */
1435 bad_fork_cleanup_files:
1436         exit_files(p); /* blocking */
1437 bad_fork_cleanup_semundo:
1438         exit_sem(p);
1439 bad_fork_cleanup_audit:
1440         audit_free(p);
1441 bad_fork_cleanup_policy:
1442         perf_event_free_task(p);
1443 #ifdef CONFIG_NUMA
1444         mpol_put(p->mempolicy);
1445 bad_fork_cleanup_cgroup:
1446 #endif
1447         if (clone_flags & CLONE_THREAD)
1448                 threadgroup_fork_read_unlock(current);
1449         cgroup_exit(p, cgroup_callbacks_done);
1450         delayacct_tsk_free(p);
1451         module_put(task_thread_info(p)->exec_domain->module);
1452 bad_fork_cleanup_count:
1453         atomic_dec(&p->cred->user->processes);
1454         exit_creds(p);
1455 bad_fork_free:
1456         free_task(p);
1457 fork_out:
1458         return ERR_PTR(retval);
1459 }
1460
1461 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1462 {
1463         memset(regs, 0, sizeof(struct pt_regs));
1464         return regs;
1465 }
1466
1467 static inline void init_idle_pids(struct pid_link *links)
1468 {
1469         enum pid_type type;
1470
1471         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1472                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1473                 links[type].pid = &init_struct_pid;
1474         }
1475 }
1476
1477 struct task_struct * __cpuinit fork_idle(int cpu)
1478 {
1479         struct task_struct *task;
1480         struct pt_regs regs;
1481
1482         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1483                             &init_struct_pid, 0);
1484         if (!IS_ERR(task)) {
1485                 init_idle_pids(task->pids);
1486                 init_idle(task, cpu);
1487         }
1488
1489         return task;
1490 }
1491
1492 /*
1493  *  Ok, this is the main fork-routine.
1494  *
1495  * It copies the process, and if successful kick-starts
1496  * it and waits for it to finish using the VM if required.
1497  */
1498 long do_fork(unsigned long clone_flags,
1499               unsigned long stack_start,
1500               struct pt_regs *regs,
1501               unsigned long stack_size,
1502               int __user *parent_tidptr,
1503               int __user *child_tidptr)
1504 {
1505         struct task_struct *p;
1506         int trace = 0;
1507         long nr;
1508
1509         /*
1510          * Do some preliminary argument and permissions checking before we
1511          * actually start allocating stuff
1512          */
1513         if (clone_flags & CLONE_NEWUSER) {
1514                 if (clone_flags & CLONE_THREAD)
1515                         return -EINVAL;
1516                 /* hopefully this check will go away when userns support is
1517                  * complete
1518                  */
1519                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1520                                 !capable(CAP_SETGID))
1521                         return -EPERM;
1522         }
1523
1524         /*
1525          * Determine whether and which event to report to ptracer.  When
1526          * called from kernel_thread or CLONE_UNTRACED is explicitly
1527          * requested, no event is reported; otherwise, report if the event
1528          * for the type of forking is enabled.
1529          */
1530         if (likely(user_mode(regs)) && !(clone_flags & CLONE_UNTRACED)) {
1531                 if (clone_flags & CLONE_VFORK)
1532                         trace = PTRACE_EVENT_VFORK;
1533                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1534                         trace = PTRACE_EVENT_CLONE;
1535                 else
1536                         trace = PTRACE_EVENT_FORK;
1537
1538                 if (likely(!ptrace_event_enabled(current, trace)))
1539                         trace = 0;
1540         }
1541
1542         p = copy_process(clone_flags, stack_start, regs, stack_size,
1543                          child_tidptr, NULL, trace);
1544         /*
1545          * Do this prior waking up the new thread - the thread pointer
1546          * might get invalid after that point, if the thread exits quickly.
1547          */
1548         if (!IS_ERR(p)) {
1549                 struct completion vfork;
1550
1551                 trace_sched_process_fork(current, p);
1552
1553                 nr = task_pid_vnr(p);
1554
1555                 if (clone_flags & CLONE_PARENT_SETTID)
1556                         put_user(nr, parent_tidptr);
1557
1558                 if (clone_flags & CLONE_VFORK) {
1559                         p->vfork_done = &vfork;
1560                         init_completion(&vfork);
1561                         get_task_struct(p);
1562                 }
1563
1564                 audit_finish_fork(p);
1565                 wake_up_new_task(p);
1566
1567                 /* forking complete and child started to run, tell ptracer */
1568                 if (unlikely(trace))
1569                         ptrace_event(trace, nr);
1570
1571                 if (clone_flags & CLONE_VFORK) {
1572                         if (!wait_for_vfork_done(p, &vfork))
1573                                 ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1574                 }
1575         } else {
1576                 nr = PTR_ERR(p);
1577         }
1578         return nr;
1579 }
1580
1581 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1582 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1583 #endif
1584
1585 static void sighand_ctor(void *data)
1586 {
1587         struct sighand_struct *sighand = data;
1588
1589         spin_lock_init(&sighand->siglock);
1590         init_waitqueue_head(&sighand->signalfd_wqh);
1591 }
1592
1593 void __init proc_caches_init(void)
1594 {
1595         sighand_cachep = kmem_cache_create("sighand_cache",
1596                         sizeof(struct sighand_struct), 0,
1597                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1598                         SLAB_NOTRACK, sighand_ctor);
1599         signal_cachep = kmem_cache_create("signal_cache",
1600                         sizeof(struct signal_struct), 0,
1601                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1602         files_cachep = kmem_cache_create("files_cache",
1603                         sizeof(struct files_struct), 0,
1604                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1605         fs_cachep = kmem_cache_create("fs_cache",
1606                         sizeof(struct fs_struct), 0,
1607                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1608         /*
1609          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1610          * whole struct cpumask for the OFFSTACK case. We could change
1611          * this to *only* allocate as much of it as required by the
1612          * maximum number of CPU's we can ever have.  The cpumask_allocation
1613          * is at the end of the structure, exactly for that reason.
1614          */
1615         mm_cachep = kmem_cache_create("mm_struct",
1616                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1617                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1618         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1619         mmap_init();
1620         nsproxy_cache_init();
1621 }
1622
1623 /*
1624  * Check constraints on flags passed to the unshare system call.
1625  */
1626 static int check_unshare_flags(unsigned long unshare_flags)
1627 {
1628         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1629                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1630                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1631                 return -EINVAL;
1632         /*
1633          * Not implemented, but pretend it works if there is nothing to
1634          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1635          * needs to unshare vm.
1636          */
1637         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1638                 /* FIXME: get_task_mm() increments ->mm_users */
1639                 if (atomic_read(&current->mm->mm_users) > 1)
1640                         return -EINVAL;
1641         }
1642
1643         return 0;
1644 }
1645
1646 /*
1647  * Unshare the filesystem structure if it is being shared
1648  */
1649 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1650 {
1651         struct fs_struct *fs = current->fs;
1652
1653         if (!(unshare_flags & CLONE_FS) || !fs)
1654                 return 0;
1655
1656         /* don't need lock here; in the worst case we'll do useless copy */
1657         if (fs->users == 1)
1658                 return 0;
1659
1660         *new_fsp = copy_fs_struct(fs);
1661         if (!*new_fsp)
1662                 return -ENOMEM;
1663
1664         return 0;
1665 }
1666
1667 /*
1668  * Unshare file descriptor table if it is being shared
1669  */
1670 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1671 {
1672         struct files_struct *fd = current->files;
1673         int error = 0;
1674
1675         if ((unshare_flags & CLONE_FILES) &&
1676             (fd && atomic_read(&fd->count) > 1)) {
1677                 *new_fdp = dup_fd(fd, &error);
1678                 if (!*new_fdp)
1679                         return error;
1680         }
1681
1682         return 0;
1683 }
1684
1685 /*
1686  * unshare allows a process to 'unshare' part of the process
1687  * context which was originally shared using clone.  copy_*
1688  * functions used by do_fork() cannot be used here directly
1689  * because they modify an inactive task_struct that is being
1690  * constructed. Here we are modifying the current, active,
1691  * task_struct.
1692  */
1693 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1694 {
1695         struct fs_struct *fs, *new_fs = NULL;
1696         struct files_struct *fd, *new_fd = NULL;
1697         struct nsproxy *new_nsproxy = NULL;
1698         int do_sysvsem = 0;
1699         int err;
1700
1701         err = check_unshare_flags(unshare_flags);
1702         if (err)
1703                 goto bad_unshare_out;
1704
1705         /*
1706          * If unsharing namespace, must also unshare filesystem information.
1707          */
1708         if (unshare_flags & CLONE_NEWNS)
1709                 unshare_flags |= CLONE_FS;
1710         /*
1711          * CLONE_NEWIPC must also detach from the undolist: after switching
1712          * to a new ipc namespace, the semaphore arrays from the old
1713          * namespace are unreachable.
1714          */
1715         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1716                 do_sysvsem = 1;
1717         err = unshare_fs(unshare_flags, &new_fs);
1718         if (err)
1719                 goto bad_unshare_out;
1720         err = unshare_fd(unshare_flags, &new_fd);
1721         if (err)
1722                 goto bad_unshare_cleanup_fs;
1723         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy, new_fs);
1724         if (err)
1725                 goto bad_unshare_cleanup_fd;
1726
1727         if (new_fs || new_fd || do_sysvsem || new_nsproxy) {
1728                 if (do_sysvsem) {
1729                         /*
1730                          * CLONE_SYSVSEM is equivalent to sys_exit().
1731                          */
1732                         exit_sem(current);
1733                 }
1734
1735                 if (new_nsproxy) {
1736                         switch_task_namespaces(current, new_nsproxy);
1737                         new_nsproxy = NULL;
1738                 }
1739
1740                 task_lock(current);
1741
1742                 if (new_fs) {
1743                         fs = current->fs;
1744                         spin_lock(&fs->lock);
1745                         current->fs = new_fs;
1746                         if (--fs->users)
1747                                 new_fs = NULL;
1748                         else
1749                                 new_fs = fs;
1750                         spin_unlock(&fs->lock);
1751                 }
1752
1753                 if (new_fd) {
1754                         fd = current->files;
1755                         current->files = new_fd;
1756                         new_fd = fd;
1757                 }
1758
1759                 task_unlock(current);
1760         }
1761
1762         if (new_nsproxy)
1763                 put_nsproxy(new_nsproxy);
1764
1765 bad_unshare_cleanup_fd:
1766         if (new_fd)
1767                 put_files_struct(new_fd);
1768
1769 bad_unshare_cleanup_fs:
1770         if (new_fs)
1771                 free_fs_struct(new_fs);
1772
1773 bad_unshare_out:
1774         return err;
1775 }
1776
1777 /*
1778  *      Helper to unshare the files of the current task.
1779  *      We don't want to expose copy_files internals to
1780  *      the exec layer of the kernel.
1781  */
1782
1783 int unshare_files(struct files_struct **displaced)
1784 {
1785         struct task_struct *task = current;
1786         struct files_struct *copy = NULL;
1787         int error;
1788
1789         error = unshare_fd(CLONE_FILES, &copy);
1790         if (error || !copy) {
1791                 *displaced = NULL;
1792                 return error;
1793         }
1794         *displaced = task->files;
1795         task_lock(task);
1796         task->files = copy;
1797         task_unlock(task);
1798         return 0;
1799 }