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/seccomp.h>
  38 #include <linux/swap.h>
  39 #include <linux/syscalls.h>
  40 #include <linux/jiffies.h>
  41 #include <linux/futex.h>
  42 #include <linux/compat.h>
  43 #include <linux/kthread.h>
  44 #include <linux/task_io_accounting_ops.h>
  45 #include <linux/rcupdate.h>
  46 #include <linux/ptrace.h>
  47 #include <linux/mount.h>
  48 #include <linux/audit.h>
  49 #include <linux/memcontrol.h>
  50 #include <linux/ftrace.h>
  51 #include <linux/profile.h>
  52 #include <linux/rmap.h>
  53 #include <linux/ksm.h>
  54 #include <linux/acct.h>
  55 #include <linux/tsacct_kern.h>
  56 #include <linux/cn_proc.h>
  57 #include <linux/freezer.h>
  58 #include <linux/delayacct.h>
  59 #include <linux/taskstats_kern.h>
  60 #include <linux/random.h>
  61 #include <linux/tty.h>
  62 #include <linux/blkdev.h>
  63 #include <linux/fs_struct.h>
  64 #include <linux/magic.h>
  65 #include <linux/perf_event.h>
  66 #include <linux/posix-timers.h>
  67 #include <linux/user-return-notifier.h>
  68 #include <linux/oom.h>
  69 #include <linux/khugepaged.h>
  70 #include <linux/signalfd.h>
  71
  72 #include <asm/pgtable.h>
  73 #include <asm/pgalloc.h>
  74 #include <asm/uaccess.h>
  75 #include <asm/mmu_context.h>
  76 #include <asm/cacheflush.h>
  77 #include <asm/tlbflush.h>
  78
  79 #include <trace/events/sched.h>
  80
  81 /*
  82  * Protected counters by write_lock_irq(&tasklist_lock)
  83  */
  84 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  85 int nr_threads;                 /* The idle threads do not count.. */
  86
  87 int max_threads;                /* tunable limit on nr_threads */
  88
  89 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  90
  91 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  92
  93 #ifdef CONFIG_PROVE_RCU
  94 int lockdep_tasklist_lock_is_held(void)
  95 {
  96         return lockdep_is_held(&tasklist_lock);
  97 }
  98 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
  99 #endif /* #ifdef CONFIG_PROVE_RCU */
 100
 101 int nr_processes(void)
 102 {
 103         int cpu;
 104         int total = 0;
 105
 106         for_each_possible_cpu(cpu)
 107                 total += per_cpu(process_counts, cpu);
 108
 109         return total;
 110 }
 111
 112 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 113 # define alloc_task_struct_node(node)           \
 114                 kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node)
 115 # define free_task_struct(tsk)                  \
 116                 kmem_cache_free(task_struct_cachep, (tsk))
 117 static struct kmem_cache *task_struct_cachep;
 118 #endif
 119
 120 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
 121 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 122                                                   int node)
 123 {
 124 #ifdef CONFIG_DEBUG_STACK_USAGE
 125         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
 126 #else
 127         gfp_t mask = GFP_KERNEL;
 128 #endif
 129         struct page *page = alloc_pages_node(node, mask, THREAD_SIZE_ORDER);
 130
 131         return page ? page_address(page) : NULL;
 132 }
 133
 134 static inline void free_thread_info(struct thread_info *ti)
 135 {
 136         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 137 }
 138 #endif
 139
 140 /* SLAB cache for signal_struct structures (tsk->signal) */
 141 static struct kmem_cache *signal_cachep;
 142
 143 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 144 struct kmem_cache *sighand_cachep;
 145
 146 /* SLAB cache for files_struct structures (tsk->files) */
 147 struct kmem_cache *files_cachep;
 148
 149 /* SLAB cache for fs_struct structures (tsk->fs) */
 150 struct kmem_cache *fs_cachep;
 151
 152 /* SLAB cache for vm_area_struct structures */
 153 struct kmem_cache *vm_area_cachep;
 154
 155 /* SLAB cache for mm_struct structures (tsk->mm) */
 156 static struct kmem_cache *mm_cachep;
 157
 158 static void account_kernel_stack(struct thread_info *ti, int account)
 159 {
 160         struct zone *zone = page_zone(virt_to_page(ti));
 161
 162         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
 163 }
 164
 165 void free_task(struct task_struct *tsk)
 166 {
 167         account_kernel_stack(tsk->stack, -1);
 168         free_thread_info(tsk->stack);
 169         rt_mutex_debug_task_free(tsk);
 170         ftrace_graph_exit_task(tsk);
 171         put_seccomp_filter(tsk);
 172         free_task_struct(tsk);
 173 }
 174 EXPORT_SYMBOL(free_task);
 175
 176 static inline void free_signal_struct(struct signal_struct *sig)
 177 {
 178         taskstats_tgid_free(sig);
 179         sched_autogroup_exit(sig);
 180         kmem_cache_free(signal_cachep, sig);
 181 }
 182
 183 static inline void put_signal_struct(struct signal_struct *sig)
 184 {
 185         if (atomic_dec_and_test(&sig->sigcnt))
 186                 free_signal_struct(sig);
 187 }
 188
 189 void __put_task_struct(struct task_struct *tsk)
 190 {
 191         WARN_ON(!tsk->exit_state);
 192         WARN_ON(atomic_read(&tsk->usage));
 193         WARN_ON(tsk == current);
 194
 195         security_task_free(tsk);
 196         exit_creds(tsk);
 197         delayacct_tsk_free(tsk);
 198         put_signal_struct(tsk->signal);
 199
 200         if (!profile_handoff_task(tsk))
 201                 free_task(tsk);
 202 }
 203 EXPORT_SYMBOL_GPL(__put_task_struct);
 204
 205 /*
 206  * macro override instead of weak attribute alias, to workaround
 207  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
 208  */
 209 #ifndef arch_task_cache_init
 210 #define arch_task_cache_init()
 211 #endif
 212
 213 void __init fork_init(unsigned long mempages)
 214 {
 215 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 216 #ifndef ARCH_MIN_TASKALIGN
 217 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 218 #endif
 219         /* create a slab on which task_structs can be allocated */
 220         task_struct_cachep =
 221                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 222                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
 223 #endif
 224
 225         /* do the arch specific task caches init */
 226         arch_task_cache_init();
 227
 228         /*
 229          * The default maximum number of threads is set to a safe
 230          * value: the thread structures can take up at most half
 231          * of memory.
 232          */
 233         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 234
 235         /*
 236          * we need to allow at least 20 threads to boot a system
 237          */
 238         if (max_threads < 20)
 239                 max_threads = 20;
 240
 241         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 242         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 243         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 244                 init_task.signal->rlim[RLIMIT_NPROC];
 245 }
 246
 247 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
 248                                                struct task_struct *src)
 249 {
 250         *dst = *src;
 251         return 0;
 252 }
 253
 254 static struct task_struct *dup_task_struct(struct task_struct *orig)
 255 {
 256         struct task_struct *tsk;
 257         struct thread_info *ti;
 258         unsigned long *stackend;
 259         int node = tsk_fork_get_node(orig);
 260         int err;
 261
 262         prepare_to_copy(orig);
 263
 264         tsk = alloc_task_struct_node(node);
 265         if (!tsk)
 266                 return NULL;
 267
 268         ti = alloc_thread_info_node(tsk, node);
 269         if (!ti) {
 270                 free_task_struct(tsk);
 271                 return NULL;
 272         }
 273
 274         err = arch_dup_task_struct(tsk, orig);
 275         if (err)
 276                 goto out;
 277
 278         tsk->stack = ti;
 279
 280         setup_thread_stack(tsk, orig);
 281         clear_user_return_notifier(tsk);
 282         clear_tsk_need_resched(tsk);
 283         stackend = end_of_stack(tsk);
 284         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 285
 286 #ifdef CONFIG_CC_STACKPROTECTOR
 287         tsk->stack_canary = get_random_int();
 288 #endif
 289
 290         /*
 291          * One for us, one for whoever does the "release_task()" (usually
 292          * parent)
 293          */
 294         atomic_set(&tsk->usage, 2);
 295 #ifdef CONFIG_BLK_DEV_IO_TRACE
 296         tsk->btrace_seq = 0;
 297 #endif
 298         tsk->splice_pipe = NULL;
 299
 300         account_kernel_stack(ti, 1);
 301
 302         return tsk;
 303
 304 out:
 305         free_thread_info(ti);
 306         free_task_struct(tsk);
 307         return NULL;
 308 }
 309
 310 #ifdef CONFIG_MMU
 311 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 312 {
 313         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 314         struct rb_node **rb_link, *rb_parent;
 315         int retval;
 316         unsigned long charge;
 317         struct mempolicy *pol;
 318
 319         down_write(&oldmm->mmap_sem);
 320         flush_cache_dup_mm(oldmm);
 321         /*
 322          * Not linked in yet - no deadlock potential:
 323          */
 324         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 325
 326         mm->locked_vm = 0;
 327         mm->mmap = NULL;
 328         mm->mmap_cache = NULL;
 329         mm->free_area_cache = oldmm->mmap_base;
 330         mm->cached_hole_size = ~0UL;
 331         mm->map_count = 0;
 332         cpumask_clear(mm_cpumask(mm));
 333         mm->mm_rb = RB_ROOT;
 334         rb_link = &mm->mm_rb.rb_node;
 335         rb_parent = NULL;
 336         pprev = &mm->mmap;
 337         retval = ksm_fork(mm, oldmm);
 338         if (retval)
 339                 goto out;
 340         retval = khugepaged_fork(mm, oldmm);
 341         if (retval)
 342                 goto out;
 343
 344         prev = NULL;
 345         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 346                 struct file *file;
 347
 348                 if (mpnt->vm_flags & VM_DONTCOPY) {
 349                         long pages = vma_pages(mpnt);
 350                         mm->total_vm -= pages;
 351                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 352                                                                 -pages);
 353                         continue;
 354                 }
 355                 charge = 0;
 356                 if (mpnt->vm_flags & VM_ACCOUNT) {
 357                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 358                         if (security_vm_enough_memory(len))
 359                                 goto fail_nomem;
 360                         charge = len;
 361                 }
 362                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 363                 if (!tmp)
 364                         goto fail_nomem;
 365                 *tmp = *mpnt;
 366                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 367                 pol = mpol_dup(vma_policy(mpnt));
 368                 retval = PTR_ERR(pol);
 369                 if (IS_ERR(pol))
 370                         goto fail_nomem_policy;
 371                 vma_set_policy(tmp, pol);
 372                 tmp->vm_mm = mm;
 373                 if (anon_vma_fork(tmp, mpnt))
 374                         goto fail_nomem_anon_vma_fork;
 375                 tmp->vm_flags &= ~VM_LOCKED;
 376                 tmp->vm_next = tmp->vm_prev = NULL;
 377                 file = tmp->vm_file;
 378                 if (file) {
 379                         struct inode *inode = file->f_path.dentry->d_inode;
 380                         struct address_space *mapping = file->f_mapping;
 381
 382                         get_file(file);
 383                         if (tmp->vm_flags & VM_DENYWRITE)
 384                                 atomic_dec(&inode->i_writecount);
 385                         mutex_lock(&mapping->i_mmap_mutex);
 386                         if (tmp->vm_flags & VM_SHARED)
 387                                 mapping->i_mmap_writable++;
 388                         flush_dcache_mmap_lock(mapping);
 389                         /* insert tmp into the share list, just after mpnt */
 390                         vma_prio_tree_add(tmp, mpnt);
 391                         flush_dcache_mmap_unlock(mapping);
 392                         mutex_unlock(&mapping->i_mmap_mutex);
 393                 }
 394
 395                 /*
 396                  * Clear hugetlb-related page reserves for children. This only
 397                  * affects MAP_PRIVATE mappings. Faults generated by the child
 398                  * are not guaranteed to succeed, even if read-only
 399                  */
 400                 if (is_vm_hugetlb_page(tmp))
 401                         reset_vma_resv_huge_pages(tmp);
 402
 403                 /*
 404                  * Link in the new vma and copy the page table entries.
 405                  */
 406                 *pprev = tmp;
 407                 pprev = &tmp->vm_next;
 408                 tmp->vm_prev = prev;
 409                 prev = tmp;
 410
 411                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 412                 rb_link = &tmp->vm_rb.rb_right;
 413                 rb_parent = &tmp->vm_rb;
 414
 415                 mm->map_count++;
 416                 retval = copy_page_range(mm, oldmm, mpnt);
 417
 418                 if (tmp->vm_ops && tmp->vm_ops->open)
 419                         tmp->vm_ops->open(tmp);
 420
 421                 if (retval)
 422                         goto out;
 423         }
 424         /* a new mm has just been created */
 425         arch_dup_mmap(oldmm, mm);
 426         retval = 0;
 427 out:
 428         up_write(&mm->mmap_sem);
 429         flush_tlb_mm(oldmm);
 430         up_write(&oldmm->mmap_sem);
 431         return retval;
 432 fail_nomem_anon_vma_fork:
 433         mpol_put(pol);
 434 fail_nomem_policy:
 435         kmem_cache_free(vm_area_cachep, tmp);
 436 fail_nomem:
 437         retval = -ENOMEM;
 438         vm_unacct_memory(charge);
 439         goto out;
 440 }
 441
 442 static inline int mm_alloc_pgd(struct mm_struct *mm)
 443 {
 444         mm->pgd = pgd_alloc(mm);
 445         if (unlikely(!mm->pgd))
 446                 return -ENOMEM;
 447         return 0;
 448 }
 449
 450 static inline void mm_free_pgd(struct mm_struct *mm)
 451 {
 452         pgd_free(mm, mm->pgd);
 453 }
 454 #else
 455 #define dup_mmap(mm, oldmm)     (0)
 456 #define mm_alloc_pgd(mm)        (0)
 457 #define mm_free_pgd(mm)
 458 #endif /* CONFIG_MMU */
 459
 460 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 461
 462 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 463 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 464
 465 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 466
 467 static int __init coredump_filter_setup(char *s)
 468 {
 469         default_dump_filter =
 470                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 471                 MMF_DUMP_FILTER_MASK;
 472         return 1;
 473 }
 474
 475 __setup("coredump_filter=", coredump_filter_setup);
 476
 477 #include <linux/init_task.h>
 478
 479 static void mm_init_aio(struct mm_struct *mm)
 480 {
 481 #ifdef CONFIG_AIO
 482         spin_lock_init(&mm->ioctx_lock);
 483         INIT_HLIST_HEAD(&mm->ioctx_list);
 484 #endif
 485 }
 486
 487 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
 488 {
 489         atomic_set(&mm->mm_users, 1);
 490         atomic_set(&mm->mm_count, 1);
 491         init_rwsem(&mm->mmap_sem);
 492         INIT_LIST_HEAD(&mm->mmlist);
 493         mm->flags = (current->mm) ?
 494                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
 495         mm->core_state = NULL;
 496         mm->nr_ptes = 0;
 497         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 498         spin_lock_init(&mm->page_table_lock);
 499         mm->free_area_cache = TASK_UNMAPPED_BASE;
 500         mm->cached_hole_size = ~0UL;
 501         mm_init_aio(mm);
 502         mm_init_owner(mm, p);
 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
 818         tsk->mm = mm;
 819         tsk->active_mm = mm;
 820         return 0;
 821
 822 fail_nomem:
 823         return retval;
 824 }
 825
 826 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 827 {
 828         struct fs_struct *fs = current->fs;
 829         if (clone_flags & CLONE_FS) {
 830                 /* tsk->fs is already what we want */
 831                 spin_lock(&fs->lock);
 832                 if (fs->in_exec) {
 833                         spin_unlock(&fs->lock);
 834                         return -EAGAIN;
 835                 }
 836                 fs->users++;
 837                 spin_unlock(&fs->lock);
 838                 return 0;
 839         }
 840         tsk->fs = copy_fs_struct(fs);
 841         if (!tsk->fs)
 842                 return -ENOMEM;
 843         return 0;
 844 }
 845
 846 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
 847 {
 848         struct files_struct *oldf, *newf;
 849         int error = 0;
 850
 851         /*
 852          * A background process may not have any files ...
 853          */
 854         oldf = current->files;
 855         if (!oldf)
 856                 goto out;
 857
 858         if (clone_flags & CLONE_FILES) {
 859                 atomic_inc(&oldf->count);
 860                 goto out;
 861         }
 862
 863         newf = dup_fd(oldf, &error);
 864         if (!newf)
 865                 goto out;
 866
 867         tsk->files = newf;
 868         error = 0;
 869 out:
 870         return error;
 871 }
 872
 873 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 874 {
 875 #ifdef CONFIG_BLOCK
 876         struct io_context *ioc = current->io_context;
 877
 878         if (!ioc)
 879                 return 0;
 880         /*
 881          * Share io context with parent, if CLONE_IO is set
 882          */
 883         if (clone_flags & CLONE_IO) {
 884                 tsk->io_context = ioc_task_link(ioc);
 885                 if (unlikely(!tsk->io_context))
 886                         return -ENOMEM;
 887         } else if (ioprio_valid(ioc->ioprio)) {
 888                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
 889                 if (unlikely(!tsk->io_context))
 890                         return -ENOMEM;
 891
 892                 tsk->io_context->ioprio = ioc->ioprio;
 893         }
 894 #endif
 895         return 0;
 896 }
 897
 898 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 899 {
 900         struct sighand_struct *sig;
 901
 902         if (clone_flags & CLONE_SIGHAND) {
 903                 atomic_inc(&current->sighand->count);
 904                 return 0;
 905         }
 906         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 907         rcu_assign_pointer(tsk->sighand, sig);
 908         if (!sig)
 909                 return -ENOMEM;
 910         atomic_set(&sig->count, 1);
 911         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 912         return 0;
 913 }
 914
 915 void __cleanup_sighand(struct sighand_struct *sighand)
 916 {
 917         if (atomic_dec_and_test(&sighand->count)) {
 918                 signalfd_cleanup(sighand);
 919                 kmem_cache_free(sighand_cachep, sighand);
 920         }
 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         get_seccomp_filter(p);
1099
1100         rt_mutex_init_task(p);
1101
1102 #ifdef CONFIG_PROVE_LOCKING
1103         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1104         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1105 #endif
1106         retval = -EAGAIN;
1107         if (atomic_read(&p->real_cred->user->processes) >=
1108                         task_rlimit(p, RLIMIT_NPROC)) {
1109                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1110                     p->real_cred->user != INIT_USER)
1111                         goto bad_fork_free;
1112         }
1113         current->flags &= ~PF_NPROC_EXCEEDED;
1114
1115         retval = copy_creds(p, clone_flags);
1116         if (retval < 0)
1117                 goto bad_fork_free;
1118
1119         /*
1120          * If multiple threads are within copy_process(), then this check
1121          * triggers too late. This doesn't hurt, the check is only there
1122          * to stop root fork bombs.
1123          */
1124         retval = -EAGAIN;
1125         if (nr_threads >= max_threads)
1126                 goto bad_fork_cleanup_count;
1127
1128         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1129                 goto bad_fork_cleanup_count;
1130
1131         p->did_exec = 0;
1132         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1133         copy_flags(clone_flags, p);
1134         INIT_LIST_HEAD(&p->children);
1135         INIT_LIST_HEAD(&p->sibling);
1136         rcu_copy_process(p);
1137         p->vfork_done = NULL;
1138         spin_lock_init(&p->alloc_lock);
1139
1140         init_sigpending(&p->pending);
1141
1142         p->utime = cputime_zero;
1143         p->stime = cputime_zero;
1144         p->gtime = cputime_zero;
1145         p->utimescaled = cputime_zero;
1146         p->stimescaled = cputime_zero;
1147 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1148         p->prev_utime = cputime_zero;
1149         p->prev_stime = cputime_zero;
1150 #endif
1151 #if defined(SPLIT_RSS_COUNTING)
1152         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1153 #endif
1154
1155         p->default_timer_slack_ns = current->timer_slack_ns;
1156
1157         task_io_accounting_init(&p->ioac);
1158         acct_clear_integrals(p);
1159
1160         posix_cpu_timers_init(p);
1161
1162         do_posix_clock_monotonic_gettime(&p->start_time);
1163         p->real_start_time = p->start_time;
1164         monotonic_to_bootbased(&p->real_start_time);
1165         p->io_context = NULL;
1166         p->audit_context = NULL;
1167         if (clone_flags & CLONE_THREAD)
1168                 threadgroup_fork_read_lock(current);
1169         cgroup_fork(p);
1170 #ifdef CONFIG_NUMA
1171         p->mempolicy = mpol_dup(p->mempolicy);
1172         if (IS_ERR(p->mempolicy)) {
1173                 retval = PTR_ERR(p->mempolicy);
1174                 p->mempolicy = NULL;
1175                 goto bad_fork_cleanup_cgroup;
1176         }
1177         mpol_fix_fork_child_flag(p);
1178 #endif
1179 #ifdef CONFIG_CPUSETS
1180         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1181         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1182 #endif
1183 #ifdef CONFIG_TRACE_IRQFLAGS
1184         p->irq_events = 0;
1185 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1186         p->hardirqs_enabled = 1;
1187 #else
1188         p->hardirqs_enabled = 0;
1189 #endif
1190         p->hardirq_enable_ip = 0;
1191         p->hardirq_enable_event = 0;
1192         p->hardirq_disable_ip = _THIS_IP_;
1193         p->hardirq_disable_event = 0;
1194         p->softirqs_enabled = 1;
1195         p->softirq_enable_ip = _THIS_IP_;
1196         p->softirq_enable_event = 0;
1197         p->softirq_disable_ip = 0;
1198         p->softirq_disable_event = 0;
1199         p->hardirq_context = 0;
1200         p->softirq_context = 0;
1201 #endif
1202 #ifdef CONFIG_LOCKDEP
1203         p->lockdep_depth = 0; /* no locks held yet */
1204         p->curr_chain_key = 0;
1205         p->lockdep_recursion = 0;
1206 #endif
1207
1208 #ifdef CONFIG_DEBUG_MUTEXES
1209         p->blocked_on = NULL; /* not blocked yet */
1210 #endif
1211 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1212         p->memcg_batch.do_batch = 0;
1213         p->memcg_batch.memcg = NULL;
1214 #endif
1215
1216         /* Perform scheduler related setup. Assign this task to a CPU. */
1217         sched_fork(p);
1218
1219         retval = perf_event_init_task(p);
1220         if (retval)
1221                 goto bad_fork_cleanup_policy;
1222         retval = audit_alloc(p);
1223         if (retval)
1224                 goto bad_fork_cleanup_policy;
1225         /* copy all the process information */
1226         retval = copy_semundo(clone_flags, p);
1227         if (retval)
1228                 goto bad_fork_cleanup_audit;
1229         retval = copy_files(clone_flags, p);
1230         if (retval)
1231                 goto bad_fork_cleanup_semundo;
1232         retval = copy_fs(clone_flags, p);
1233         if (retval)
1234                 goto bad_fork_cleanup_files;
1235         retval = copy_sighand(clone_flags, p);
1236         if (retval)
1237                 goto bad_fork_cleanup_fs;
1238         retval = copy_signal(clone_flags, p);
1239         if (retval)
1240                 goto bad_fork_cleanup_sighand;
1241         retval = copy_mm(clone_flags, p);
1242         if (retval)
1243                 goto bad_fork_cleanup_signal;
1244         retval = copy_namespaces(clone_flags, p);
1245         if (retval)
1246                 goto bad_fork_cleanup_mm;
1247         retval = copy_io(clone_flags, p);
1248         if (retval)
1249                 goto bad_fork_cleanup_namespaces;
1250         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1251         if (retval)
1252                 goto bad_fork_cleanup_io;
1253
1254         if (pid != &init_struct_pid) {
1255                 retval = -ENOMEM;
1256                 pid = alloc_pid(p->nsproxy->pid_ns);
1257                 if (!pid)
1258                         goto bad_fork_cleanup_io;
1259         }
1260
1261         p->pid = pid_nr(pid);
1262         p->tgid = p->pid;
1263         if (clone_flags & CLONE_THREAD)
1264                 p->tgid = current->tgid;
1265
1266         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1267         /*
1268          * Clear TID on mm_release()?
1269          */
1270         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1271 #ifdef CONFIG_BLOCK
1272         p->plug = NULL;
1273 #endif
1274 #ifdef CONFIG_FUTEX
1275         p->robust_list = NULL;
1276 #ifdef CONFIG_COMPAT
1277         p->compat_robust_list = NULL;
1278 #endif
1279         INIT_LIST_HEAD(&p->pi_state_list);
1280         p->pi_state_cache = NULL;
1281 #endif
1282         /*
1283          * sigaltstack should be cleared when sharing the same VM
1284          */
1285         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1286                 p->sas_ss_sp = p->sas_ss_size = 0;
1287
1288         /*
1289          * Syscall tracing and stepping should be turned off in the
1290          * child regardless of CLONE_PTRACE.
1291          */
1292         user_disable_single_step(p);
1293         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1294 #ifdef TIF_SYSCALL_EMU
1295         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1296 #endif
1297         clear_all_latency_tracing(p);
1298
1299         /* ok, now we should be set up.. */
1300         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1301         p->pdeath_signal = 0;
1302         p->exit_state = 0;
1303
1304         p->nr_dirtied = 0;
1305         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1306
1307         /*
1308          * Ok, make it visible to the rest of the system.
1309          * We dont wake it up yet.
1310          */
1311         p->group_leader = p;
1312         INIT_LIST_HEAD(&p->thread_group);
1313
1314         /* Now that the task is set up, run cgroup callbacks if
1315          * necessary. We need to run them before the task is visible
1316          * on the tasklist. */
1317         cgroup_fork_callbacks(p);
1318         cgroup_callbacks_done = 1;
1319
1320         /* Need tasklist lock for parent etc handling! */
1321         write_lock_irq(&tasklist_lock);
1322
1323         /* CLONE_PARENT re-uses the old parent */
1324         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1325                 p->real_parent = current->real_parent;
1326                 p->parent_exec_id = current->parent_exec_id;
1327         } else {
1328                 p->real_parent = current;
1329                 p->parent_exec_id = current->self_exec_id;
1330         }
1331
1332         spin_lock(&current->sighand->siglock);
1333
1334         /*
1335          * Process group and session signals need to be delivered to just the
1336          * parent before the fork or both the parent and the child after the
1337          * fork. Restart if a signal comes in before we add the new process to
1338          * it's process group.
1339          * A fatal signal pending means that current will exit, so the new
1340          * thread can't slip out of an OOM kill (or normal SIGKILL).
1341         */
1342         recalc_sigpending();
1343         if (signal_pending(current)) {
1344                 spin_unlock(&current->sighand->siglock);
1345                 write_unlock_irq(&tasklist_lock);
1346                 retval = -ERESTARTNOINTR;
1347                 goto bad_fork_free_pid;
1348         }
1349
1350         if (clone_flags & CLONE_THREAD) {
1351                 current->signal->nr_threads++;
1352                 atomic_inc(&current->signal->live);
1353                 atomic_inc(&current->signal->sigcnt);
1354                 p->group_leader = current->group_leader;
1355                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1356         }
1357
1358         if (likely(p->pid)) {
1359                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1360
1361                 if (thread_group_leader(p)) {
1362                         if (is_child_reaper(pid))
1363                                 p->nsproxy->pid_ns->child_reaper = p;
1364
1365                         p->signal->leader_pid = pid;
1366                         p->signal->tty = tty_kref_get(current->signal->tty);
1367                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1368                         attach_pid(p, PIDTYPE_SID, task_session(current));
1369                         list_add_tail(&p->sibling, &p->real_parent->children);
1370                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1371                         __this_cpu_inc(process_counts);
1372                 }
1373                 attach_pid(p, PIDTYPE_PID, pid);
1374                 nr_threads++;
1375         }
1376
1377         total_forks++;
1378         spin_unlock(&current->sighand->siglock);
1379         write_unlock_irq(&tasklist_lock);
1380         proc_fork_connector(p);
1381         cgroup_post_fork(p);
1382         if (clone_flags & CLONE_THREAD)
1383                 threadgroup_fork_read_unlock(current);
1384         perf_event_fork(p);
1385         return p;
1386
1387 bad_fork_free_pid:
1388         if (pid != &init_struct_pid)
1389                 free_pid(pid);
1390 bad_fork_cleanup_io:
1391         if (p->io_context)
1392                 exit_io_context(p);
1393 bad_fork_cleanup_namespaces:
1394         exit_task_namespaces(p);
1395 bad_fork_cleanup_mm:
1396         if (p->mm)
1397                 mmput(p->mm);
1398 bad_fork_cleanup_signal:
1399         if (!(clone_flags & CLONE_THREAD))
1400                 free_signal_struct(p->signal);
1401 bad_fork_cleanup_sighand:
1402         __cleanup_sighand(p->sighand);
1403 bad_fork_cleanup_fs:
1404         exit_fs(p); /* blocking */
1405 bad_fork_cleanup_files:
1406         exit_files(p); /* blocking */
1407 bad_fork_cleanup_semundo:
1408         exit_sem(p);
1409 bad_fork_cleanup_audit:
1410         audit_free(p);
1411 bad_fork_cleanup_policy:
1412         perf_event_free_task(p);
1413 #ifdef CONFIG_NUMA
1414         mpol_put(p->mempolicy);
1415 bad_fork_cleanup_cgroup:
1416 #endif
1417         if (clone_flags & CLONE_THREAD)
1418                 threadgroup_fork_read_unlock(current);
1419         cgroup_exit(p, cgroup_callbacks_done);
1420         delayacct_tsk_free(p);
1421         module_put(task_thread_info(p)->exec_domain->module);
1422 bad_fork_cleanup_count:
1423         atomic_dec(&p->cred->user->processes);
1424         exit_creds(p);
1425 bad_fork_free:
1426         free_task(p);
1427 fork_out:
1428         return ERR_PTR(retval);
1429 }
1430
1431 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1432 {
1433         memset(regs, 0, sizeof(struct pt_regs));
1434         return regs;
1435 }
1436
1437 static inline void init_idle_pids(struct pid_link *links)
1438 {
1439         enum pid_type type;
1440
1441         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1442                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1443                 links[type].pid = &init_struct_pid;
1444         }
1445 }
1446
1447 struct task_struct * __cpuinit fork_idle(int cpu)
1448 {
1449         struct task_struct *task;
1450         struct pt_regs regs;
1451
1452         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1453                             &init_struct_pid, 0);
1454         if (!IS_ERR(task)) {
1455                 init_idle_pids(task->pids);
1456                 init_idle(task, cpu);
1457         }
1458
1459         return task;
1460 }
1461
1462 /*
1463  *  Ok, this is the main fork-routine.
1464  *
1465  * It copies the process, and if successful kick-starts
1466  * it and waits for it to finish using the VM if required.
1467  */
1468 long do_fork(unsigned long clone_flags,
1469               unsigned long stack_start,
1470               struct pt_regs *regs,
1471               unsigned long stack_size,
1472               int __user *parent_tidptr,
1473               int __user *child_tidptr)
1474 {
1475         struct task_struct *p;
1476         int trace = 0;
1477         long nr;
1478
1479         /*
1480          * Do some preliminary argument and permissions checking before we
1481          * actually start allocating stuff
1482          */
1483         if (clone_flags & CLONE_NEWUSER) {
1484                 if (clone_flags & CLONE_THREAD)
1485                         return -EINVAL;
1486                 /* hopefully this check will go away when userns support is
1487                  * complete
1488                  */
1489                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1490                                 !capable(CAP_SETGID))
1491                         return -EPERM;
1492         }
1493
1494         /*
1495          * Determine whether and which event to report to ptracer.  When
1496          * called from kernel_thread or CLONE_UNTRACED is explicitly
1497          * requested, no event is reported; otherwise, report if the event
1498          * for the type of forking is enabled.
1499          */
1500         if (likely(user_mode(regs)) && !(clone_flags & CLONE_UNTRACED)) {
1501                 if (clone_flags & CLONE_VFORK)
1502                         trace = PTRACE_EVENT_VFORK;
1503                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1504                         trace = PTRACE_EVENT_CLONE;
1505                 else
1506                         trace = PTRACE_EVENT_FORK;
1507
1508                 if (likely(!ptrace_event_enabled(current, trace)))
1509                         trace = 0;
1510         }
1511
1512         p = copy_process(clone_flags, stack_start, regs, stack_size,
1513                          child_tidptr, NULL, trace);
1514         /*
1515          * Do this prior waking up the new thread - the thread pointer
1516          * might get invalid after that point, if the thread exits quickly.
1517          */
1518         if (!IS_ERR(p)) {
1519                 struct completion vfork;
1520
1521                 trace_sched_process_fork(current, p);
1522
1523                 nr = task_pid_vnr(p);
1524
1525                 if (clone_flags & CLONE_PARENT_SETTID)
1526                         put_user(nr, parent_tidptr);
1527
1528                 if (clone_flags & CLONE_VFORK) {
1529                         p->vfork_done = &vfork;
1530                         init_completion(&vfork);
1531                 }
1532
1533                 audit_finish_fork(p);
1534
1535                 /*
1536                  * We set PF_STARTING at creation in case tracing wants to
1537                  * use this to distinguish a fully live task from one that
1538                  * hasn't finished SIGSTOP raising yet.  Now we clear it
1539                  * and set the child going.
1540                  */
1541                 p->flags &= ~PF_STARTING;
1542
1543                 wake_up_new_task(p);
1544
1545                 /* forking complete and child started to run, tell ptracer */
1546                 if (unlikely(trace))
1547                         ptrace_event(trace, nr);
1548
1549                 if (clone_flags & CLONE_VFORK) {
1550                         freezer_do_not_count();
1551                         wait_for_completion(&vfork);
1552                         freezer_count();
1553                         ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1554                 }
1555         } else {
1556                 nr = PTR_ERR(p);
1557         }
1558         return nr;
1559 }
1560
1561 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1562 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1563 #endif
1564
1565 static void sighand_ctor(void *data)
1566 {
1567         struct sighand_struct *sighand = data;
1568
1569         spin_lock_init(&sighand->siglock);
1570         init_waitqueue_head(&sighand->signalfd_wqh);
1571 }
1572
1573 void __init proc_caches_init(void)
1574 {
1575         sighand_cachep = kmem_cache_create("sighand_cache",
1576                         sizeof(struct sighand_struct), 0,
1577                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1578                         SLAB_NOTRACK, sighand_ctor);
1579         signal_cachep = kmem_cache_create("signal_cache",
1580                         sizeof(struct signal_struct), 0,
1581                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1582         files_cachep = kmem_cache_create("files_cache",
1583                         sizeof(struct files_struct), 0,
1584                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1585         fs_cachep = kmem_cache_create("fs_cache",
1586                         sizeof(struct fs_struct), 0,
1587                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1588         /*
1589          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1590          * whole struct cpumask for the OFFSTACK case. We could change
1591          * this to *only* allocate as much of it as required by the
1592          * maximum number of CPU's we can ever have.  The cpumask_allocation
1593          * is at the end of the structure, exactly for that reason.
1594          */
1595         mm_cachep = kmem_cache_create("mm_struct",
1596                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1597                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1598         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1599         mmap_init();
1600         nsproxy_cache_init();
1601 }
1602
1603 /*
1604  * Check constraints on flags passed to the unshare system call.
1605  */
1606 static int check_unshare_flags(unsigned long unshare_flags)
1607 {
1608         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1609                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1610                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1611                 return -EINVAL;
1612         /*
1613          * Not implemented, but pretend it works if there is nothing to
1614          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1615          * needs to unshare vm.
1616          */
1617         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1618                 /* FIXME: get_task_mm() increments ->mm_users */
1619                 if (atomic_read(&current->mm->mm_users) > 1)
1620                         return -EINVAL;
1621         }
1622
1623         return 0;
1624 }
1625
1626 /*
1627  * Unshare the filesystem structure if it is being shared
1628  */
1629 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1630 {
1631         struct fs_struct *fs = current->fs;
1632
1633         if (!(unshare_flags & CLONE_FS) || !fs)
1634                 return 0;
1635
1636         /* don't need lock here; in the worst case we'll do useless copy */
1637         if (fs->users == 1)
1638                 return 0;
1639
1640         *new_fsp = copy_fs_struct(fs);
1641         if (!*new_fsp)
1642                 return -ENOMEM;
1643
1644         return 0;
1645 }
1646
1647 /*
1648  * Unshare file descriptor table if it is being shared
1649  */
1650 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1651 {
1652         struct files_struct *fd = current->files;
1653         int error = 0;
1654
1655         if ((unshare_flags & CLONE_FILES) &&
1656             (fd && atomic_read(&fd->count) > 1)) {
1657                 *new_fdp = dup_fd(fd, &error);
1658                 if (!*new_fdp)
1659                         return error;
1660         }
1661
1662         return 0;
1663 }
1664
1665 /*
1666  * unshare allows a process to 'unshare' part of the process
1667  * context which was originally shared using clone.  copy_*
1668  * functions used by do_fork() cannot be used here directly
1669  * because they modify an inactive task_struct that is being
1670  * constructed. Here we are modifying the current, active,
1671  * task_struct.
1672  */
1673 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1674 {
1675         struct fs_struct *fs, *new_fs = NULL;
1676         struct files_struct *fd, *new_fd = NULL;
1677         struct nsproxy *new_nsproxy = NULL;
1678         int do_sysvsem = 0;
1679         int err;
1680
1681         err = check_unshare_flags(unshare_flags);
1682         if (err)
1683                 goto bad_unshare_out;
1684
1685         /*
1686          * If unsharing namespace, must also unshare filesystem information.
1687          */
1688         if (unshare_flags & CLONE_NEWNS)
1689                 unshare_flags |= CLONE_FS;
1690         /*
1691          * CLONE_NEWIPC must also detach from the undolist: after switching
1692          * to a new ipc namespace, the semaphore arrays from the old
1693          * namespace are unreachable.
1694          */
1695         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1696                 do_sysvsem = 1;
1697         err = unshare_fs(unshare_flags, &new_fs);
1698         if (err)
1699                 goto bad_unshare_out;
1700         err = unshare_fd(unshare_flags, &new_fd);
1701         if (err)
1702                 goto bad_unshare_cleanup_fs;
1703         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy, new_fs);
1704         if (err)
1705                 goto bad_unshare_cleanup_fd;
1706
1707         if (new_fs || new_fd || do_sysvsem || new_nsproxy) {
1708                 if (do_sysvsem) {
1709                         /*
1710                          * CLONE_SYSVSEM is equivalent to sys_exit().
1711                          */
1712                         exit_sem(current);
1713                 }
1714
1715                 if (new_nsproxy) {
1716                         switch_task_namespaces(current, new_nsproxy);
1717                         new_nsproxy = NULL;
1718                 }
1719
1720                 task_lock(current);
1721
1722                 if (new_fs) {
1723                         fs = current->fs;
1724                         spin_lock(&fs->lock);
1725                         current->fs = new_fs;
1726                         if (--fs->users)
1727                                 new_fs = NULL;
1728                         else
1729                                 new_fs = fs;
1730                         spin_unlock(&fs->lock);
1731                 }
1732
1733                 if (new_fd) {
1734                         fd = current->files;
1735                         current->files = new_fd;
1736                         new_fd = fd;
1737                 }
1738
1739                 task_unlock(current);
1740         }
1741
1742         if (new_nsproxy)
1743                 put_nsproxy(new_nsproxy);
1744
1745 bad_unshare_cleanup_fd:
1746         if (new_fd)
1747                 put_files_struct(new_fd);
1748
1749 bad_unshare_cleanup_fs:
1750         if (new_fs)
1751                 free_fs_struct(new_fs);
1752
1753 bad_unshare_out:
1754         return err;
1755 }
1756
1757 /*
1758  *      Helper to unshare the files of the current task.
1759  *      We don't want to expose copy_files internals to
1760  *      the exec layer of the kernel.
1761  */
1762
1763 int unshare_files(struct files_struct **displaced)
1764 {
1765         struct task_struct *task = current;
1766         struct files_struct *copy = NULL;
1767         int error;
1768
1769         error = unshare_fd(CLONE_FILES, &copy);
1770         if (error || !copy) {
1771                 *displaced = NULL;
1772                 return error;
1773         }
1774         *displaced = task->files;
1775         task_lock(task);
1776         task->files = copy;
1777         task_unlock(task);
1778         return 0;
1779 }