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