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