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 #include <linux/signalfd.h>
  70 #include <linux/uprobes.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 #define CREATE_TRACE_POINTS
  82 #include <trace/events/task.h>
  83
  84 /*
  85  * Protected counters by write_lock_irq(&tasklist_lock)
  86  */
  87 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  88 int nr_threads;                 /* The idle threads do not count.. */
  89
  90 int max_threads;                /* tunable limit on nr_threads */
  91
  92 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  93
  94 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  95
  96 #ifdef CONFIG_PROVE_RCU
  97 int lockdep_tasklist_lock_is_held(void)
  98 {
  99         return lockdep_is_held(&tasklist_lock);
 100 }
 101 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 102 #endif /* #ifdef CONFIG_PROVE_RCU */
 103
 104 int nr_processes(void)
 105 {
 106         int cpu;
 107         int total = 0;
 108
 109         for_each_possible_cpu(cpu)
 110                 total += per_cpu(process_counts, cpu);
 111
 112         return total;
 113 }
 114
 115 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 116 # define alloc_task_struct_node(node)           \
 117                 kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node)
 118 # define free_task_struct(tsk)                  \
 119                 kmem_cache_free(task_struct_cachep, (tsk))
 120 static struct kmem_cache *task_struct_cachep;
 121 #endif
 122
 123 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
 124 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 125                                                   int node)
 126 {
 127 #ifdef CONFIG_DEBUG_STACK_USAGE
 128         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
 129 #else
 130         gfp_t mask = GFP_KERNEL;
 131 #endif
 132         struct page *page = alloc_pages_node(node, mask, THREAD_SIZE_ORDER);
 133
 134         return page ? page_address(page) : NULL;
 135 }
 136
 137 static inline void free_thread_info(struct thread_info *ti)
 138 {
 139         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 140 }
 141 #endif
 142
 143 /* SLAB cache for signal_struct structures (tsk->signal) */
 144 static struct kmem_cache *signal_cachep;
 145
 146 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 147 struct kmem_cache *sighand_cachep;
 148
 149 /* SLAB cache for files_struct structures (tsk->files) */
 150 struct kmem_cache *files_cachep;
 151
 152 /* SLAB cache for fs_struct structures (tsk->fs) */
 153 struct kmem_cache *fs_cachep;
 154
 155 /* SLAB cache for vm_area_struct structures */
 156 struct kmem_cache *vm_area_cachep;
 157
 158 /* SLAB cache for mm_struct structures (tsk->mm) */
 159 static struct kmem_cache *mm_cachep;
 160
 161 static void account_kernel_stack(struct thread_info *ti, int account)
 162 {
 163         struct zone *zone = page_zone(virt_to_page(ti));
 164
 165         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
 166 }
 167
 168 void free_task(struct task_struct *tsk)
 169 {
 170         account_kernel_stack(tsk->stack, -1);
 171         free_thread_info(tsk->stack);
 172         rt_mutex_debug_task_free(tsk);
 173         ftrace_graph_exit_task(tsk);
 174         free_task_struct(tsk);
 175 }
 176 EXPORT_SYMBOL(free_task);
 177
 178 static inline void free_signal_struct(struct signal_struct *sig)
 179 {
 180         taskstats_tgid_free(sig);
 181         sched_autogroup_exit(sig);
 182         kmem_cache_free(signal_cachep, sig);
 183 }
 184
 185 static inline void put_signal_struct(struct signal_struct *sig)
 186 {
 187         if (atomic_dec_and_test(&sig->sigcnt))
 188                 free_signal_struct(sig);
 189 }
 190
 191 void __put_task_struct(struct task_struct *tsk)
 192 {
 193         WARN_ON(!tsk->exit_state);
 194         WARN_ON(atomic_read(&tsk->usage));
 195         WARN_ON(tsk == current);
 196
 197         exit_creds(tsk);
 198         delayacct_tsk_free(tsk);
 199         put_signal_struct(tsk->signal);
 200
 201         if (!profile_handoff_task(tsk))
 202                 free_task(tsk);
 203 }
 204 EXPORT_SYMBOL_GPL(__put_task_struct);
 205
 206 /*
 207  * macro override instead of weak attribute alias, to workaround
 208  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
 209  */
 210 #ifndef arch_task_cache_init
 211 #define arch_task_cache_init()
 212 #endif
 213
 214 void __init fork_init(unsigned long mempages)
 215 {
 216 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 217 #ifndef ARCH_MIN_TASKALIGN
 218 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 219 #endif
 220         /* create a slab on which task_structs can be allocated */
 221         task_struct_cachep =
 222                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 223                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
 224 #endif
 225
 226         /* do the arch specific task caches init */
 227         arch_task_cache_init();
 228
 229         /*
 230          * The default maximum number of threads is set to a safe
 231          * value: the thread structures can take up at most half
 232          * of memory.
 233          */
 234         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 235
 236         /*
 237          * we need to allow at least 20 threads to boot a system
 238          */
 239         if (max_threads < 20)
 240                 max_threads = 20;
 241
 242         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 243         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 244         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 245                 init_task.signal->rlim[RLIMIT_NPROC];
 246 }
 247
 248 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
 249                                                struct task_struct *src)
 250 {
 251         *dst = *src;
 252         return 0;
 253 }
 254
 255 static struct task_struct *dup_task_struct(struct task_struct *orig)
 256 {
 257         struct task_struct *tsk;
 258         struct thread_info *ti;
 259         unsigned long *stackend;
 260         int node = tsk_fork_get_node(orig);
 261         int err;
 262
 263         prepare_to_copy(orig);
 264
 265         tsk = alloc_task_struct_node(node);
 266         if (!tsk)
 267                 return NULL;
 268
 269         ti = alloc_thread_info_node(tsk, node);
 270         if (!ti) {
 271                 free_task_struct(tsk);
 272                 return NULL;
 273         }
 274
 275         err = arch_dup_task_struct(tsk, orig);
 276         if (err)
 277                 goto out;
 278
 279         tsk->stack = ti;
 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         /*
 292          * One for us, one for whoever does the "release_task()" (usually
 293          * parent)
 294          */
 295         atomic_set(&tsk->usage, 2);
 296 #ifdef CONFIG_BLK_DEV_IO_TRACE
 297         tsk->btrace_seq = 0;
 298 #endif
 299         tsk->splice_pipe = NULL;
 300
 301         account_kernel_stack(ti, 1);
 302
 303         return tsk;
 304
 305 out:
 306         free_thread_info(ti);
 307         free_task_struct(tsk);
 308         return NULL;
 309 }
 310
 311 #ifdef CONFIG_MMU
 312 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 313 {
 314         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 315         struct rb_node **rb_link, *rb_parent;
 316         int retval;
 317         unsigned long charge;
 318         struct mempolicy *pol;
 319
 320         down_write(&oldmm->mmap_sem);
 321         flush_cache_dup_mm(oldmm);
 322         /*
 323          * Not linked in yet - no deadlock potential:
 324          */
 325         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 326
 327         mm->locked_vm = 0;
 328         mm->mmap = NULL;
 329         mm->mmap_cache = NULL;
 330         mm->free_area_cache = oldmm->mmap_base;
 331         mm->cached_hole_size = ~0UL;
 332         mm->map_count = 0;
 333         cpumask_clear(mm_cpumask(mm));
 334         mm->mm_rb = RB_ROOT;
 335         rb_link = &mm->mm_rb.rb_node;
 336         rb_parent = NULL;
 337         pprev = &mm->mmap;
 338         retval = ksm_fork(mm, oldmm);
 339         if (retval)
 340                 goto out;
 341         retval = khugepaged_fork(mm, oldmm);
 342         if (retval)
 343                 goto out;
 344
 345         prev = NULL;
 346         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 347                 struct file *file;
 348
 349                 if (mpnt->vm_flags & VM_DONTCOPY) {
 350                         long pages = vma_pages(mpnt);
 351                         mm->total_vm -= pages;
 352                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 353                                                                 -pages);
 354                         continue;
 355                 }
 356                 charge = 0;
 357                 if (mpnt->vm_flags & VM_ACCOUNT) {
 358                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 359                         if (security_vm_enough_memory(len))
 360                                 goto fail_nomem;
 361                         charge = len;
 362                 }
 363                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 364                 if (!tmp)
 365                         goto fail_nomem;
 366                 *tmp = *mpnt;
 367                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 368                 pol = mpol_dup(vma_policy(mpnt));
 369                 retval = PTR_ERR(pol);
 370                 if (IS_ERR(pol))
 371                         goto fail_nomem_policy;
 372                 vma_set_policy(tmp, pol);
 373                 tmp->vm_mm = mm;
 374                 if (anon_vma_fork(tmp, mpnt))
 375                         goto fail_nomem_anon_vma_fork;
 376                 tmp->vm_flags &= ~VM_LOCKED;
 377                 tmp->vm_next = tmp->vm_prev = NULL;
 378                 file = tmp->vm_file;
 379                 if (file) {
 380                         struct inode *inode = file->f_path.dentry->d_inode;
 381                         struct address_space *mapping = file->f_mapping;
 382
 383                         get_file(file);
 384                         if (tmp->vm_flags & VM_DENYWRITE)
 385                                 atomic_dec(&inode->i_writecount);
 386                         mutex_lock(&mapping->i_mmap_mutex);
 387                         if (tmp->vm_flags & VM_SHARED)
 388                                 mapping->i_mmap_writable++;
 389                         flush_dcache_mmap_lock(mapping);
 390                         /* insert tmp into the share list, just after mpnt */
 391                         vma_prio_tree_add(tmp, mpnt);
 392                         flush_dcache_mmap_unlock(mapping);
 393                         mutex_unlock(&mapping->i_mmap_mutex);
 394                 }
 395
 396                 /*
 397                  * Clear hugetlb-related page reserves for children. This only
 398                  * affects MAP_PRIVATE mappings. Faults generated by the child
 399                  * are not guaranteed to succeed, even if read-only
 400                  */
 401                 if (is_vm_hugetlb_page(tmp))
 402                         reset_vma_resv_huge_pages(tmp);
 403
 404                 /*
 405                  * Link in the new vma and copy the page table entries.
 406                  */
 407                 *pprev = tmp;
 408                 pprev = &tmp->vm_next;
 409                 tmp->vm_prev = prev;
 410                 prev = tmp;
 411
 412                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 413                 rb_link = &tmp->vm_rb.rb_right;
 414                 rb_parent = &tmp->vm_rb;
 415
 416                 mm->map_count++;
 417                 retval = copy_page_range(mm, oldmm, mpnt);
 418
 419                 if (tmp->vm_ops && tmp->vm_ops->open)
 420                         tmp->vm_ops->open(tmp);
 421
 422                 if (retval)
 423                         goto out;
 424
 425                 if (file && uprobe_mmap(tmp))
 426                         goto out;
 427         }
 428         /* a new mm has just been created */
 429         arch_dup_mmap(oldmm, mm);
 430         retval = 0;
 431 out:
 432         up_write(&mm->mmap_sem);
 433         flush_tlb_mm(oldmm);
 434         up_write(&oldmm->mmap_sem);
 435         return retval;
 436 fail_nomem_anon_vma_fork:
 437         mpol_put(pol);
 438 fail_nomem_policy:
 439         kmem_cache_free(vm_area_cachep, tmp);
 440 fail_nomem:
 441         retval = -ENOMEM;
 442         vm_unacct_memory(charge);
 443         goto out;
 444 }
 445
 446 static inline int mm_alloc_pgd(struct mm_struct *mm)
 447 {
 448         mm->pgd = pgd_alloc(mm);
 449         if (unlikely(!mm->pgd))
 450                 return -ENOMEM;
 451         return 0;
 452 }
 453
 454 static inline void mm_free_pgd(struct mm_struct *mm)
 455 {
 456         pgd_free(mm, mm->pgd);
 457 }
 458 #else
 459 #define dup_mmap(mm, oldmm)     (0)
 460 #define mm_alloc_pgd(mm)        (0)
 461 #define mm_free_pgd(mm)
 462 #endif /* CONFIG_MMU */
 463
 464 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 465
 466 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 467 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 468
 469 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 470
 471 static int __init coredump_filter_setup(char *s)
 472 {
 473         default_dump_filter =
 474                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 475                 MMF_DUMP_FILTER_MASK;
 476         return 1;
 477 }
 478
 479 __setup("coredump_filter=", coredump_filter_setup);
 480
 481 #include <linux/init_task.h>
 482
 483 static void mm_init_aio(struct mm_struct *mm)
 484 {
 485 #ifdef CONFIG_AIO
 486         spin_lock_init(&mm->ioctx_lock);
 487         INIT_HLIST_HEAD(&mm->ioctx_list);
 488 #endif
 489 }
 490
 491 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
 492 {
 493         atomic_set(&mm->mm_users, 1);
 494         atomic_set(&mm->mm_count, 1);
 495         init_rwsem(&mm->mmap_sem);
 496         INIT_LIST_HEAD(&mm->mmlist);
 497         mm->flags = (current->mm) ?
 498                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
 499         mm->core_state = NULL;
 500         mm->nr_ptes = 0;
 501         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 502         spin_lock_init(&mm->page_table_lock);
 503         mm->free_area_cache = TASK_UNMAPPED_BASE;
 504         mm->cached_hole_size = ~0UL;
 505         mm_init_aio(mm);
 506         mm_init_owner(mm, p);
 507
 508         if (likely(!mm_alloc_pgd(mm))) {
 509                 mm->def_flags = 0;
 510                 mmu_notifier_mm_init(mm);
 511                 return mm;
 512         }
 513
 514         free_mm(mm);
 515         return NULL;
 516 }
 517
 518 /*
 519  * Allocate and initialize an mm_struct.
 520  */
 521 struct mm_struct *mm_alloc(void)
 522 {
 523         struct mm_struct *mm;
 524
 525         mm = allocate_mm();
 526         if (!mm)
 527                 return NULL;
 528
 529         memset(mm, 0, sizeof(*mm));
 530         mm_init_cpumask(mm);
 531         return mm_init(mm, current);
 532 }
 533
 534 /*
 535  * Called when the last reference to the mm
 536  * is dropped: either by a lazy thread or by
 537  * mmput. Free the page directory and the mm.
 538  */
 539 void __mmdrop(struct mm_struct *mm)
 540 {
 541         BUG_ON(mm == &init_mm);
 542         mm_free_pgd(mm);
 543         destroy_context(mm);
 544         mmu_notifier_mm_destroy(mm);
 545 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 546         VM_BUG_ON(mm->pmd_huge_pte);
 547 #endif
 548         free_mm(mm);
 549 }
 550 EXPORT_SYMBOL_GPL(__mmdrop);
 551
 552 /*
 553  * Decrement the use count and release all resources for an mm.
 554  */
 555 void mmput(struct mm_struct *mm)
 556 {
 557         might_sleep();
 558
 559         if (atomic_dec_and_test(&mm->mm_users)) {
 560                 uprobe_clear_state(mm);
 561                 exit_aio(mm);
 562                 ksm_exit(mm);
 563                 khugepaged_exit(mm); /* must run before exit_mmap */
 564                 exit_mmap(mm);
 565                 set_mm_exe_file(mm, NULL);
 566                 if (!list_empty(&mm->mmlist)) {
 567                         spin_lock(&mmlist_lock);
 568                         list_del(&mm->mmlist);
 569                         spin_unlock(&mmlist_lock);
 570                 }
 571                 put_swap_token(mm);
 572                 if (mm->binfmt)
 573                         module_put(mm->binfmt->module);
 574                 mmdrop(mm);
 575         }
 576 }
 577 EXPORT_SYMBOL_GPL(mmput);
 578
 579 /*
 580  * We added or removed a vma mapping the executable. The vmas are only mapped
 581  * during exec and are not mapped with the mmap system call.
 582  * Callers must hold down_write() on the mm's mmap_sem for these
 583  */
 584 void added_exe_file_vma(struct mm_struct *mm)
 585 {
 586         mm->num_exe_file_vmas++;
 587 }
 588
 589 void removed_exe_file_vma(struct mm_struct *mm)
 590 {
 591         mm->num_exe_file_vmas--;
 592         if ((mm->num_exe_file_vmas == 0) && mm->exe_file) {
 593                 fput(mm->exe_file);
 594                 mm->exe_file = NULL;
 595         }
 596
 597 }
 598
 599 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 600 {
 601         if (new_exe_file)
 602                 get_file(new_exe_file);
 603         if (mm->exe_file)
 604                 fput(mm->exe_file);
 605         mm->exe_file = new_exe_file;
 606         mm->num_exe_file_vmas = 0;
 607 }
 608
 609 struct file *get_mm_exe_file(struct mm_struct *mm)
 610 {
 611         struct file *exe_file;
 612
 613         /* We need mmap_sem to protect against races with removal of
 614          * VM_EXECUTABLE vmas */
 615         down_read(&mm->mmap_sem);
 616         exe_file = mm->exe_file;
 617         if (exe_file)
 618                 get_file(exe_file);
 619         up_read(&mm->mmap_sem);
 620         return exe_file;
 621 }
 622
 623 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
 624 {
 625         /* It's safe to write the exe_file pointer without exe_file_lock because
 626          * this is called during fork when the task is not yet in /proc */
 627         newmm->exe_file = get_mm_exe_file(oldmm);
 628 }
 629
 630 /**
 631  * get_task_mm - acquire a reference to the task's mm
 632  *
 633  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 634  * this kernel workthread has transiently adopted a user mm with use_mm,
 635  * to do its AIO) is not set and if so returns a reference to it, after
 636  * bumping up the use count.  User must release the mm via mmput()
 637  * after use.  Typically used by /proc and ptrace.
 638  */
 639 struct mm_struct *get_task_mm(struct task_struct *task)
 640 {
 641         struct mm_struct *mm;
 642
 643         task_lock(task);
 644         mm = task->mm;
 645         if (mm) {
 646                 if (task->flags & PF_KTHREAD)
 647                         mm = NULL;
 648                 else
 649                         atomic_inc(&mm->mm_users);
 650         }
 651         task_unlock(task);
 652         return mm;
 653 }
 654 EXPORT_SYMBOL_GPL(get_task_mm);
 655
 656 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 657 {
 658         struct mm_struct *mm;
 659         int err;
 660
 661         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 662         if (err)
 663                 return ERR_PTR(err);
 664
 665         mm = get_task_mm(task);
 666         if (mm && mm != current->mm &&
 667                         !ptrace_may_access(task, mode)) {
 668                 mmput(mm);
 669                 mm = ERR_PTR(-EACCES);
 670         }
 671         mutex_unlock(&task->signal->cred_guard_mutex);
 672
 673         return mm;
 674 }
 675
 676 /* Please note the differences between mmput and mm_release.
 677  * mmput is called whenever we stop holding onto a mm_struct,
 678  * error success whatever.
 679  *
 680  * mm_release is called after a mm_struct has been removed
 681  * from the current process.
 682  *
 683  * This difference is important for error handling, when we
 684  * only half set up a mm_struct for a new process and need to restore
 685  * the old one.  Because we mmput the new mm_struct before
 686  * restoring the old one. . .
 687  * Eric Biederman 10 January 1998
 688  */
 689 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 690 {
 691         struct completion *vfork_done = tsk->vfork_done;
 692
 693         /* Get rid of any futexes when releasing the mm */
 694 #ifdef CONFIG_FUTEX
 695         if (unlikely(tsk->robust_list)) {
 696                 exit_robust_list(tsk);
 697                 tsk->robust_list = NULL;
 698         }
 699 #ifdef CONFIG_COMPAT
 700         if (unlikely(tsk->compat_robust_list)) {
 701                 compat_exit_robust_list(tsk);
 702                 tsk->compat_robust_list = NULL;
 703         }
 704 #endif
 705         if (unlikely(!list_empty(&tsk->pi_state_list)))
 706                 exit_pi_state_list(tsk);
 707 #endif
 708
 709         uprobe_free_utask(tsk);
 710
 711         /* Get rid of any cached register state */
 712         deactivate_mm(tsk, mm);
 713
 714         /* notify parent sleeping on vfork() */
 715         if (vfork_done) {
 716                 tsk->vfork_done = NULL;
 717                 complete(vfork_done);
 718         }
 719
 720         /*
 721          * If we're exiting normally, clear a user-space tid field if
 722          * requested.  We leave this alone when dying by signal, to leave
 723          * the value intact in a core dump, and to save the unnecessary
 724          * trouble otherwise.  Userland only wants this done for a sys_exit.
 725          */
 726         if (tsk->clear_child_tid) {
 727                 if (!(tsk->flags & PF_SIGNALED) &&
 728                     atomic_read(&mm->mm_users) > 1) {
 729                         /*
 730                          * We don't check the error code - if userspace has
 731                          * not set up a proper pointer then tough luck.
 732                          */
 733                         put_user(0, tsk->clear_child_tid);
 734                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 735                                         1, NULL, NULL, 0);
 736                 }
 737                 tsk->clear_child_tid = NULL;
 738         }
 739 }
 740
 741 /*
 742  * Allocate a new mm structure and copy contents from the
 743  * mm structure of the passed in task structure.
 744  */
 745 struct mm_struct *dup_mm(struct task_struct *tsk)
 746 {
 747         struct mm_struct *mm, *oldmm = current->mm;
 748         int err;
 749
 750         if (!oldmm)
 751                 return NULL;
 752
 753         mm = allocate_mm();
 754         if (!mm)
 755                 goto fail_nomem;
 756
 757         memcpy(mm, oldmm, sizeof(*mm));
 758         mm_init_cpumask(mm);
 759
 760         /* Initializing for Swap token stuff */
 761         mm->token_priority = 0;
 762         mm->last_interval = 0;
 763
 764 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 765         mm->pmd_huge_pte = NULL;
 766 #endif
 767         uprobe_reset_state(mm);
 768
 769         if (!mm_init(mm, tsk))
 770                 goto fail_nomem;
 771
 772         if (init_new_context(tsk, mm))
 773                 goto fail_nocontext;
 774
 775         dup_mm_exe_file(oldmm, mm);
 776
 777         err = dup_mmap(mm, oldmm);
 778         if (err)
 779                 goto free_pt;
 780
 781         mm->hiwater_rss = get_mm_rss(mm);
 782         mm->hiwater_vm = mm->total_vm;
 783
 784         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 785                 goto free_pt;
 786
 787         return mm;
 788
 789 free_pt:
 790         /* don't put binfmt in mmput, we haven't got module yet */
 791         mm->binfmt = NULL;
 792         mmput(mm);
 793
 794 fail_nomem:
 795         return NULL;
 796
 797 fail_nocontext:
 798         /*
 799          * If init_new_context() failed, we cannot use mmput() to free the mm
 800          * because it calls destroy_context()
 801          */
 802         mm_free_pgd(mm);
 803         free_mm(mm);
 804         return NULL;
 805 }
 806
 807 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 808 {
 809         struct mm_struct *mm, *oldmm;
 810         int retval;
 811
 812         tsk->min_flt = tsk->maj_flt = 0;
 813         tsk->nvcsw = tsk->nivcsw = 0;
 814 #ifdef CONFIG_DETECT_HUNG_TASK
 815         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 816 #endif
 817
 818         tsk->mm = NULL;
 819         tsk->active_mm = NULL;
 820
 821         /*
 822          * Are we cloning a kernel thread?
 823          *
 824          * We need to steal a active VM for that..
 825          */
 826         oldmm = current->mm;
 827         if (!oldmm)
 828                 return 0;
 829
 830         if (clone_flags & CLONE_VM) {
 831                 atomic_inc(&oldmm->mm_users);
 832                 mm = oldmm;
 833                 goto good_mm;
 834         }
 835
 836         retval = -ENOMEM;
 837         mm = dup_mm(tsk);
 838         if (!mm)
 839                 goto fail_nomem;
 840
 841 good_mm:
 842         /* Initializing for Swap token stuff */
 843         mm->token_priority = 0;
 844         mm->last_interval = 0;
 845
 846         tsk->mm = mm;
 847         tsk->active_mm = mm;
 848         return 0;
 849
 850 fail_nomem:
 851         return retval;
 852 }
 853
 854 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 855 {
 856         struct fs_struct *fs = current->fs;
 857         if (clone_flags & CLONE_FS) {
 858                 /* tsk->fs is already what we want */
 859                 spin_lock(&fs->lock);
 860                 if (fs->in_exec) {
 861                         spin_unlock(&fs->lock);
 862                         return -EAGAIN;
 863                 }
 864                 fs->users++;
 865                 spin_unlock(&fs->lock);
 866                 return 0;
 867         }
 868         tsk->fs = copy_fs_struct(fs);
 869         if (!tsk->fs)
 870                 return -ENOMEM;
 871         return 0;
 872 }
 873
 874 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
 875 {
 876         struct files_struct *oldf, *newf;
 877         int error = 0;
 878
 879         /*
 880          * A background process may not have any files ...
 881          */
 882         oldf = current->files;
 883         if (!oldf)
 884                 goto out;
 885
 886         if (clone_flags & CLONE_FILES) {
 887                 atomic_inc(&oldf->count);
 888                 goto out;
 889         }
 890
 891         newf = dup_fd(oldf, &error);
 892         if (!newf)
 893                 goto out;
 894
 895         tsk->files = newf;
 896         error = 0;
 897 out:
 898         return error;
 899 }
 900
 901 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 902 {
 903 #ifdef CONFIG_BLOCK
 904         struct io_context *ioc = current->io_context;
 905         struct io_context *new_ioc;
 906
 907         if (!ioc)
 908                 return 0;
 909         /*
 910          * Share io context with parent, if CLONE_IO is set
 911          */
 912         if (clone_flags & CLONE_IO) {
 913                 tsk->io_context = ioc_task_link(ioc);
 914                 if (unlikely(!tsk->io_context))
 915                         return -ENOMEM;
 916         } else if (ioprio_valid(ioc->ioprio)) {
 917                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
 918                 if (unlikely(!new_ioc))
 919                         return -ENOMEM;
 920
 921                 new_ioc->ioprio = ioc->ioprio;
 922                 put_io_context(new_ioc);
 923         }
 924 #endif
 925         return 0;
 926 }
 927
 928 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 929 {
 930         struct sighand_struct *sig;
 931
 932         if (clone_flags & CLONE_SIGHAND) {
 933                 atomic_inc(&current->sighand->count);
 934                 return 0;
 935         }
 936         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 937         rcu_assign_pointer(tsk->sighand, sig);
 938         if (!sig)
 939                 return -ENOMEM;
 940         atomic_set(&sig->count, 1);
 941         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 942         return 0;
 943 }
 944
 945 void __cleanup_sighand(struct sighand_struct *sighand)
 946 {
 947         if (atomic_dec_and_test(&sighand->count)) {
 948                 signalfd_cleanup(sighand);
 949                 kmem_cache_free(sighand_cachep, sighand);
 950         }
 951 }
 952
 953
 954 /*
 955  * Initialize POSIX timer handling for a thread group.
 956  */
 957 static void posix_cpu_timers_init_group(struct signal_struct *sig)
 958 {
 959         unsigned long cpu_limit;
 960
 961         /* Thread group counters. */
 962         thread_group_cputime_init(sig);
 963
 964         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
 965         if (cpu_limit != RLIM_INFINITY) {
 966                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
 967                 sig->cputimer.running = 1;
 968         }
 969
 970         /* The timer lists. */
 971         INIT_LIST_HEAD(&sig->cpu_timers[0]);
 972         INIT_LIST_HEAD(&sig->cpu_timers[1]);
 973         INIT_LIST_HEAD(&sig->cpu_timers[2]);
 974 }
 975
 976 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
 977 {
 978         struct signal_struct *sig;
 979
 980         if (clone_flags & CLONE_THREAD)
 981                 return 0;
 982
 983         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
 984         tsk->signal = sig;
 985         if (!sig)
 986                 return -ENOMEM;
 987
 988         sig->nr_threads = 1;
 989         atomic_set(&sig->live, 1);
 990         atomic_set(&sig->sigcnt, 1);
 991         init_waitqueue_head(&sig->wait_chldexit);
 992         if (clone_flags & CLONE_NEWPID)
 993                 sig->flags |= SIGNAL_UNKILLABLE;
 994         sig->curr_target = tsk;
 995         init_sigpending(&sig->shared_pending);
 996         INIT_LIST_HEAD(&sig->posix_timers);
 997
 998         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 999         sig->real_timer.function = it_real_fn;
1000
1001         task_lock(current->group_leader);
1002         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1003         task_unlock(current->group_leader);
1004
1005         posix_cpu_timers_init_group(sig);
1006
1007         tty_audit_fork(sig);
1008         sched_autogroup_fork(sig);
1009
1010 #ifdef CONFIG_CGROUPS
1011         init_rwsem(&sig->group_rwsem);
1012 #endif
1013
1014         sig->oom_adj = current->signal->oom_adj;
1015         sig->oom_score_adj = current->signal->oom_score_adj;
1016         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1017
1018         mutex_init(&sig->cred_guard_mutex);
1019
1020         return 0;
1021 }
1022
1023 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
1024 {
1025         unsigned long new_flags = p->flags;
1026
1027         new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1028         new_flags |= PF_FORKNOEXEC;
1029         new_flags |= PF_STARTING;
1030         p->flags = new_flags;
1031 }
1032
1033 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1034 {
1035         current->clear_child_tid = tidptr;
1036
1037         return task_pid_vnr(current);
1038 }
1039
1040 static void rt_mutex_init_task(struct task_struct *p)
1041 {
1042         raw_spin_lock_init(&p->pi_lock);
1043 #ifdef CONFIG_RT_MUTEXES
1044         plist_head_init(&p->pi_waiters);
1045         p->pi_blocked_on = NULL;
1046 #endif
1047 }
1048
1049 #ifdef CONFIG_MM_OWNER
1050 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1051 {
1052         mm->owner = p;
1053 }
1054 #endif /* CONFIG_MM_OWNER */
1055
1056 /*
1057  * Initialize POSIX timer handling for a single task.
1058  */
1059 static void posix_cpu_timers_init(struct task_struct *tsk)
1060 {
1061         tsk->cputime_expires.prof_exp = 0;
1062         tsk->cputime_expires.virt_exp = 0;
1063         tsk->cputime_expires.sched_exp = 0;
1064         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1065         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1066         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1067 }
1068
1069 /*
1070  * This creates a new process as a copy of the old one,
1071  * but does not actually start it yet.
1072  *
1073  * It copies the registers, and all the appropriate
1074  * parts of the process environment (as per the clone
1075  * flags). The actual kick-off is left to the caller.
1076  */
1077 static struct task_struct *copy_process(unsigned long clone_flags,
1078                                         unsigned long stack_start,
1079                                         struct pt_regs *regs,
1080                                         unsigned long stack_size,
1081                                         int __user *child_tidptr,
1082                                         struct pid *pid,
1083                                         int trace)
1084 {
1085         int retval;
1086         struct task_struct *p;
1087         int cgroup_callbacks_done = 0;
1088
1089         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1090                 return ERR_PTR(-EINVAL);
1091
1092         /*
1093          * Thread groups must share signals as well, and detached threads
1094          * can only be started up within the thread group.
1095          */
1096         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1097                 return ERR_PTR(-EINVAL);
1098
1099         /*
1100          * Shared signal handlers imply shared VM. By way of the above,
1101          * thread groups also imply shared VM. Blocking this case allows
1102          * for various simplifications in other code.
1103          */
1104         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1105                 return ERR_PTR(-EINVAL);
1106
1107         /*
1108          * Siblings of global init remain as zombies on exit since they are
1109          * not reaped by their parent (swapper). To solve this and to avoid
1110          * multi-rooted process trees, prevent global and container-inits
1111          * from creating siblings.
1112          */
1113         if ((clone_flags & CLONE_PARENT) &&
1114                                 current->signal->flags & SIGNAL_UNKILLABLE)
1115                 return ERR_PTR(-EINVAL);
1116
1117         retval = security_task_create(clone_flags);
1118         if (retval)
1119                 goto fork_out;
1120
1121         retval = -ENOMEM;
1122         p = dup_task_struct(current);
1123         if (!p)
1124                 goto fork_out;
1125
1126         ftrace_graph_init_task(p);
1127
1128         rt_mutex_init_task(p);
1129
1130 #ifdef CONFIG_PROVE_LOCKING
1131         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1132         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1133 #endif
1134         retval = -EAGAIN;
1135         if (atomic_read(&p->real_cred->user->processes) >=
1136                         task_rlimit(p, RLIMIT_NPROC)) {
1137                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1138                     p->real_cred->user != INIT_USER)
1139                         goto bad_fork_free;
1140         }
1141         current->flags &= ~PF_NPROC_EXCEEDED;
1142
1143         retval = copy_creds(p, clone_flags);
1144         if (retval < 0)
1145                 goto bad_fork_free;
1146
1147         /*
1148          * If multiple threads are within copy_process(), then this check
1149          * triggers too late. This doesn't hurt, the check is only there
1150          * to stop root fork bombs.
1151          */
1152         retval = -EAGAIN;
1153         if (nr_threads >= max_threads)
1154                 goto bad_fork_cleanup_count;
1155
1156         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1157                 goto bad_fork_cleanup_count;
1158
1159         p->did_exec = 0;
1160         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1161         copy_flags(clone_flags, p);
1162         INIT_LIST_HEAD(&p->children);
1163         INIT_LIST_HEAD(&p->sibling);
1164         rcu_copy_process(p);
1165         p->vfork_done = NULL;
1166         spin_lock_init(&p->alloc_lock);
1167
1168         init_sigpending(&p->pending);
1169
1170         p->utime = p->stime = p->gtime = 0;
1171         p->utimescaled = p->stimescaled = 0;
1172 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1173         p->prev_utime = p->prev_stime = 0;
1174 #endif
1175 #if defined(SPLIT_RSS_COUNTING)
1176         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1177 #endif
1178
1179         p->default_timer_slack_ns = current->timer_slack_ns;
1180
1181         task_io_accounting_init(&p->ioac);
1182         acct_clear_integrals(p);
1183
1184         posix_cpu_timers_init(p);
1185
1186         do_posix_clock_monotonic_gettime(&p->start_time);
1187         p->real_start_time = p->start_time;
1188         monotonic_to_bootbased(&p->real_start_time);
1189         p->io_context = NULL;
1190         p->audit_context = NULL;
1191         if (clone_flags & CLONE_THREAD)
1192                 threadgroup_change_begin(current);
1193         cgroup_fork(p);
1194 #ifdef CONFIG_NUMA
1195         p->mempolicy = mpol_dup(p->mempolicy);
1196         if (IS_ERR(p->mempolicy)) {
1197                 retval = PTR_ERR(p->mempolicy);
1198                 p->mempolicy = NULL;
1199                 goto bad_fork_cleanup_cgroup;
1200         }
1201         mpol_fix_fork_child_flag(p);
1202 #endif
1203 #ifdef CONFIG_CPUSETS
1204         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1205         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1206 #endif
1207 #ifdef CONFIG_TRACE_IRQFLAGS
1208         p->irq_events = 0;
1209 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1210         p->hardirqs_enabled = 1;
1211 #else
1212         p->hardirqs_enabled = 0;
1213 #endif
1214         p->hardirq_enable_ip = 0;
1215         p->hardirq_enable_event = 0;
1216         p->hardirq_disable_ip = _THIS_IP_;
1217         p->hardirq_disable_event = 0;
1218         p->softirqs_enabled = 1;
1219         p->softirq_enable_ip = _THIS_IP_;
1220         p->softirq_enable_event = 0;
1221         p->softirq_disable_ip = 0;
1222         p->softirq_disable_event = 0;
1223         p->hardirq_context = 0;
1224         p->softirq_context = 0;
1225 #endif
1226 #ifdef CONFIG_LOCKDEP
1227         p->lockdep_depth = 0; /* no locks held yet */
1228         p->curr_chain_key = 0;
1229         p->lockdep_recursion = 0;
1230 #endif
1231
1232 #ifdef CONFIG_DEBUG_MUTEXES
1233         p->blocked_on = NULL; /* not blocked yet */
1234 #endif
1235 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1236         p->memcg_batch.do_batch = 0;
1237         p->memcg_batch.memcg = NULL;
1238 #endif
1239
1240         /* Perform scheduler related setup. Assign this task to a CPU. */
1241         sched_fork(p);
1242
1243         retval = perf_event_init_task(p);
1244         if (retval)
1245                 goto bad_fork_cleanup_policy;
1246         retval = audit_alloc(p);
1247         if (retval)
1248                 goto bad_fork_cleanup_policy;
1249         /* copy all the process information */
1250         retval = copy_semundo(clone_flags, p);
1251         if (retval)
1252                 goto bad_fork_cleanup_audit;
1253         retval = copy_files(clone_flags, p);
1254         if (retval)
1255                 goto bad_fork_cleanup_semundo;
1256         retval = copy_fs(clone_flags, p);
1257         if (retval)
1258                 goto bad_fork_cleanup_files;
1259         retval = copy_sighand(clone_flags, p);
1260         if (retval)
1261                 goto bad_fork_cleanup_fs;
1262         retval = copy_signal(clone_flags, p);
1263         if (retval)
1264                 goto bad_fork_cleanup_sighand;
1265         retval = copy_mm(clone_flags, p);
1266         if (retval)
1267                 goto bad_fork_cleanup_signal;
1268         retval = copy_namespaces(clone_flags, p);
1269         if (retval)
1270                 goto bad_fork_cleanup_mm;
1271         retval = copy_io(clone_flags, p);
1272         if (retval)
1273                 goto bad_fork_cleanup_namespaces;
1274         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1275         if (retval)
1276                 goto bad_fork_cleanup_io;
1277
1278         if (pid != &init_struct_pid) {
1279                 retval = -ENOMEM;
1280                 pid = alloc_pid(p->nsproxy->pid_ns);
1281                 if (!pid)
1282                         goto bad_fork_cleanup_io;
1283         }
1284
1285         p->pid = pid_nr(pid);
1286         p->tgid = p->pid;
1287         if (clone_flags & CLONE_THREAD)
1288                 p->tgid = current->tgid;
1289
1290         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1291         /*
1292          * Clear TID on mm_release()?
1293          */
1294         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1295 #ifdef CONFIG_BLOCK
1296         p->plug = NULL;
1297 #endif
1298 #ifdef CONFIG_FUTEX
1299         p->robust_list = NULL;
1300 #ifdef CONFIG_COMPAT
1301         p->compat_robust_list = NULL;
1302 #endif
1303         INIT_LIST_HEAD(&p->pi_state_list);
1304         p->pi_state_cache = NULL;
1305 #endif
1306         uprobe_copy_process(p);
1307         /*
1308          * sigaltstack should be cleared when sharing the same VM
1309          */
1310         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1311                 p->sas_ss_sp = p->sas_ss_size = 0;
1312
1313         /*
1314          * Syscall tracing and stepping should be turned off in the
1315          * child regardless of CLONE_PTRACE.
1316          */
1317         user_disable_single_step(p);
1318         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1319 #ifdef TIF_SYSCALL_EMU
1320         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1321 #endif
1322         clear_all_latency_tracing(p);
1323
1324         /* ok, now we should be set up.. */
1325         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1326         p->pdeath_signal = 0;
1327         p->exit_state = 0;
1328
1329         p->nr_dirtied = 0;
1330         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1331         p->dirty_paused_when = 0;
1332
1333         /*
1334          * Ok, make it visible to the rest of the system.
1335          * We dont wake it up yet.
1336          */
1337         p->group_leader = p;
1338         INIT_LIST_HEAD(&p->thread_group);
1339
1340         /* Now that the task is set up, run cgroup callbacks if
1341          * necessary. We need to run them before the task is visible
1342          * on the tasklist. */
1343         cgroup_fork_callbacks(p);
1344         cgroup_callbacks_done = 1;
1345
1346         /* Need tasklist lock for parent etc handling! */
1347         write_lock_irq(&tasklist_lock);
1348
1349         /* CLONE_PARENT re-uses the old parent */
1350         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1351                 p->real_parent = current->real_parent;
1352                 p->parent_exec_id = current->parent_exec_id;
1353         } else {
1354                 p->real_parent = current;
1355                 p->parent_exec_id = current->self_exec_id;
1356         }
1357
1358         spin_lock(&current->sighand->siglock);
1359
1360         /*
1361          * Process group and session signals need to be delivered to just the
1362          * parent before the fork or both the parent and the child after the
1363          * fork. Restart if a signal comes in before we add the new process to
1364          * it's process group.
1365          * A fatal signal pending means that current will exit, so the new
1366          * thread can't slip out of an OOM kill (or normal SIGKILL).
1367         */
1368         recalc_sigpending();
1369         if (signal_pending(current)) {
1370                 spin_unlock(&current->sighand->siglock);
1371                 write_unlock_irq(&tasklist_lock);
1372                 retval = -ERESTARTNOINTR;
1373                 goto bad_fork_free_pid;
1374         }
1375
1376         if (clone_flags & CLONE_THREAD) {
1377                 current->signal->nr_threads++;
1378                 atomic_inc(&current->signal->live);
1379                 atomic_inc(&current->signal->sigcnt);
1380                 p->group_leader = current->group_leader;
1381                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1382         }
1383
1384         if (likely(p->pid)) {
1385                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1386
1387                 if (thread_group_leader(p)) {
1388                         if (is_child_reaper(pid))
1389                                 p->nsproxy->pid_ns->child_reaper = p;
1390
1391                         p->signal->leader_pid = pid;
1392                         p->signal->tty = tty_kref_get(current->signal->tty);
1393                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1394                         attach_pid(p, PIDTYPE_SID, task_session(current));
1395                         list_add_tail(&p->sibling, &p->real_parent->children);
1396                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1397                         __this_cpu_inc(process_counts);
1398                 }
1399                 attach_pid(p, PIDTYPE_PID, pid);
1400                 nr_threads++;
1401         }
1402
1403         total_forks++;
1404         spin_unlock(&current->sighand->siglock);
1405         write_unlock_irq(&tasklist_lock);
1406         proc_fork_connector(p);
1407         cgroup_post_fork(p);
1408         if (clone_flags & CLONE_THREAD)
1409                 threadgroup_change_end(current);
1410         perf_event_fork(p);
1411
1412         trace_task_newtask(p, clone_flags);
1413
1414         return p;
1415
1416 bad_fork_free_pid:
1417         if (pid != &init_struct_pid)
1418                 free_pid(pid);
1419 bad_fork_cleanup_io:
1420         if (p->io_context)
1421                 exit_io_context(p);
1422 bad_fork_cleanup_namespaces:
1423         exit_task_namespaces(p);
1424 bad_fork_cleanup_mm:
1425         if (p->mm)
1426                 mmput(p->mm);
1427 bad_fork_cleanup_signal:
1428         if (!(clone_flags & CLONE_THREAD))
1429                 free_signal_struct(p->signal);
1430 bad_fork_cleanup_sighand:
1431         __cleanup_sighand(p->sighand);
1432 bad_fork_cleanup_fs:
1433         exit_fs(p); /* blocking */
1434 bad_fork_cleanup_files:
1435         exit_files(p); /* blocking */
1436 bad_fork_cleanup_semundo:
1437         exit_sem(p);
1438 bad_fork_cleanup_audit:
1439         audit_free(p);
1440 bad_fork_cleanup_policy:
1441         perf_event_free_task(p);
1442 #ifdef CONFIG_NUMA
1443         mpol_put(p->mempolicy);
1444 bad_fork_cleanup_cgroup:
1445 #endif
1446         if (clone_flags & CLONE_THREAD)
1447                 threadgroup_change_end(current);
1448         cgroup_exit(p, cgroup_callbacks_done);
1449         delayacct_tsk_free(p);
1450         module_put(task_thread_info(p)->exec_domain->module);
1451 bad_fork_cleanup_count:
1452         atomic_dec(&p->cred->user->processes);
1453         exit_creds(p);
1454 bad_fork_free:
1455         free_task(p);
1456 fork_out:
1457         return ERR_PTR(retval);
1458 }
1459
1460 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1461 {
1462         memset(regs, 0, sizeof(struct pt_regs));
1463         return regs;
1464 }
1465
1466 static inline void init_idle_pids(struct pid_link *links)
1467 {
1468         enum pid_type type;
1469
1470         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1471                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1472                 links[type].pid = &init_struct_pid;
1473         }
1474 }
1475
1476 struct task_struct * __cpuinit fork_idle(int cpu)
1477 {
1478         struct task_struct *task;
1479         struct pt_regs regs;
1480
1481         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1482                             &init_struct_pid, 0);
1483         if (!IS_ERR(task)) {
1484                 init_idle_pids(task->pids);
1485                 init_idle(task, cpu);
1486         }
1487
1488         return task;
1489 }
1490
1491 /*
1492  *  Ok, this is the main fork-routine.
1493  *
1494  * It copies the process, and if successful kick-starts
1495  * it and waits for it to finish using the VM if required.
1496  */
1497 long do_fork(unsigned long clone_flags,
1498               unsigned long stack_start,
1499               struct pt_regs *regs,
1500               unsigned long stack_size,
1501               int __user *parent_tidptr,
1502               int __user *child_tidptr)
1503 {
1504         struct task_struct *p;
1505         int trace = 0;
1506         long nr;
1507
1508         /*
1509          * Do some preliminary argument and permissions checking before we
1510          * actually start allocating stuff
1511          */
1512         if (clone_flags & CLONE_NEWUSER) {
1513                 if (clone_flags & CLONE_THREAD)
1514                         return -EINVAL;
1515                 /* hopefully this check will go away when userns support is
1516                  * complete
1517                  */
1518                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1519                                 !capable(CAP_SETGID))
1520                         return -EPERM;
1521         }
1522
1523         /*
1524          * Determine whether and which event to report to ptracer.  When
1525          * called from kernel_thread or CLONE_UNTRACED is explicitly
1526          * requested, no event is reported; otherwise, report if the event
1527          * for the type of forking is enabled.
1528          */
1529         if (likely(user_mode(regs)) && !(clone_flags & CLONE_UNTRACED)) {
1530                 if (clone_flags & CLONE_VFORK)
1531                         trace = PTRACE_EVENT_VFORK;
1532                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1533                         trace = PTRACE_EVENT_CLONE;
1534                 else
1535                         trace = PTRACE_EVENT_FORK;
1536
1537                 if (likely(!ptrace_event_enabled(current, trace)))
1538                         trace = 0;
1539         }
1540
1541         p = copy_process(clone_flags, stack_start, regs, stack_size,
1542                          child_tidptr, NULL, trace);
1543         /*
1544          * Do this prior waking up the new thread - the thread pointer
1545          * might get invalid after that point, if the thread exits quickly.
1546          */
1547         if (!IS_ERR(p)) {
1548                 struct completion vfork;
1549
1550                 trace_sched_process_fork(current, p);
1551
1552                 nr = task_pid_vnr(p);
1553
1554                 if (clone_flags & CLONE_PARENT_SETTID)
1555                         put_user(nr, parent_tidptr);
1556
1557                 if (clone_flags & CLONE_VFORK) {
1558                         p->vfork_done = &vfork;
1559                         init_completion(&vfork);
1560                 }
1561
1562                 /*
1563                  * We set PF_STARTING at creation in case tracing wants to
1564                  * use this to distinguish a fully live task from one that
1565                  * hasn't finished SIGSTOP raising yet.  Now we clear it
1566                  * and set the child going.
1567                  */
1568                 p->flags &= ~PF_STARTING;
1569
1570                 wake_up_new_task(p);
1571
1572                 /* forking complete and child started to run, tell ptracer */
1573                 if (unlikely(trace))
1574                         ptrace_event(trace, nr);
1575
1576                 if (clone_flags & CLONE_VFORK) {
1577                         freezer_do_not_count();
1578                         wait_for_completion(&vfork);
1579                         freezer_count();
1580                         ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1581                 }
1582         } else {
1583                 nr = PTR_ERR(p);
1584         }
1585         return nr;
1586 }
1587
1588 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1589 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1590 #endif
1591
1592 static void sighand_ctor(void *data)
1593 {
1594         struct sighand_struct *sighand = data;
1595
1596         spin_lock_init(&sighand->siglock);
1597         init_waitqueue_head(&sighand->signalfd_wqh);
1598 }
1599
1600 void __init proc_caches_init(void)
1601 {
1602         sighand_cachep = kmem_cache_create("sighand_cache",
1603                         sizeof(struct sighand_struct), 0,
1604                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1605                         SLAB_NOTRACK, sighand_ctor);
1606         signal_cachep = kmem_cache_create("signal_cache",
1607                         sizeof(struct signal_struct), 0,
1608                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1609         files_cachep = kmem_cache_create("files_cache",
1610                         sizeof(struct files_struct), 0,
1611                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1612         fs_cachep = kmem_cache_create("fs_cache",
1613                         sizeof(struct fs_struct), 0,
1614                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1615         /*
1616          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1617          * whole struct cpumask for the OFFSTACK case. We could change
1618          * this to *only* allocate as much of it as required by the
1619          * maximum number of CPU's we can ever have.  The cpumask_allocation
1620          * is at the end of the structure, exactly for that reason.
1621          */
1622         mm_cachep = kmem_cache_create("mm_struct",
1623                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1624                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1625         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1626         mmap_init();
1627         nsproxy_cache_init();
1628 }
1629
1630 /*
1631  * Check constraints on flags passed to the unshare system call.
1632  */
1633 static int check_unshare_flags(unsigned long unshare_flags)
1634 {
1635         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1636                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1637                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1638                 return -EINVAL;
1639         /*
1640          * Not implemented, but pretend it works if there is nothing to
1641          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1642          * needs to unshare vm.
1643          */
1644         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1645                 /* FIXME: get_task_mm() increments ->mm_users */
1646                 if (atomic_read(&current->mm->mm_users) > 1)
1647                         return -EINVAL;
1648         }
1649
1650         return 0;
1651 }
1652
1653 /*
1654  * Unshare the filesystem structure if it is being shared
1655  */
1656 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1657 {
1658         struct fs_struct *fs = current->fs;
1659
1660         if (!(unshare_flags & CLONE_FS) || !fs)
1661                 return 0;
1662
1663         /* don't need lock here; in the worst case we'll do useless copy */
1664         if (fs->users == 1)
1665                 return 0;
1666
1667         *new_fsp = copy_fs_struct(fs);
1668         if (!*new_fsp)
1669                 return -ENOMEM;
1670
1671         return 0;
1672 }
1673
1674 /*
1675  * Unshare file descriptor table if it is being shared
1676  */
1677 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1678 {
1679         struct files_struct *fd = current->files;
1680         int error = 0;
1681
1682         if ((unshare_flags & CLONE_FILES) &&
1683             (fd && atomic_read(&fd->count) > 1)) {
1684                 *new_fdp = dup_fd(fd, &error);
1685                 if (!*new_fdp)
1686                         return error;
1687         }
1688
1689         return 0;
1690 }
1691
1692 /*
1693  * unshare allows a process to 'unshare' part of the process
1694  * context which was originally shared using clone.  copy_*
1695  * functions used by do_fork() cannot be used here directly
1696  * because they modify an inactive task_struct that is being
1697  * constructed. Here we are modifying the current, active,
1698  * task_struct.
1699  */
1700 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1701 {
1702         struct fs_struct *fs, *new_fs = NULL;
1703         struct files_struct *fd, *new_fd = NULL;
1704         struct nsproxy *new_nsproxy = NULL;
1705         int do_sysvsem = 0;
1706         int err;
1707
1708         err = check_unshare_flags(unshare_flags);
1709         if (err)
1710                 goto bad_unshare_out;
1711
1712         /*
1713          * If unsharing namespace, must also unshare filesystem information.
1714          */
1715         if (unshare_flags & CLONE_NEWNS)
1716                 unshare_flags |= CLONE_FS;
1717         /*
1718          * CLONE_NEWIPC must also detach from the undolist: after switching
1719          * to a new ipc namespace, the semaphore arrays from the old
1720          * namespace are unreachable.
1721          */
1722         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1723                 do_sysvsem = 1;
1724         err = unshare_fs(unshare_flags, &new_fs);
1725         if (err)
1726                 goto bad_unshare_out;
1727         err = unshare_fd(unshare_flags, &new_fd);
1728         if (err)
1729                 goto bad_unshare_cleanup_fs;
1730         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy, new_fs);
1731         if (err)
1732                 goto bad_unshare_cleanup_fd;
1733
1734         if (new_fs || new_fd || do_sysvsem || new_nsproxy) {
1735                 if (do_sysvsem) {
1736                         /*
1737                          * CLONE_SYSVSEM is equivalent to sys_exit().
1738                          */
1739                         exit_sem(current);
1740                 }
1741
1742                 if (new_nsproxy) {
1743                         switch_task_namespaces(current, new_nsproxy);
1744                         new_nsproxy = NULL;
1745                 }
1746
1747                 task_lock(current);
1748
1749                 if (new_fs) {
1750                         fs = current->fs;
1751                         spin_lock(&fs->lock);
1752                         current->fs = new_fs;
1753                         if (--fs->users)
1754                                 new_fs = NULL;
1755                         else
1756                                 new_fs = fs;
1757                         spin_unlock(&fs->lock);
1758                 }
1759
1760                 if (new_fd) {
1761                         fd = current->files;
1762                         current->files = new_fd;
1763                         new_fd = fd;
1764                 }
1765
1766                 task_unlock(current);
1767         }
1768
1769         if (new_nsproxy)
1770                 put_nsproxy(new_nsproxy);
1771
1772 bad_unshare_cleanup_fd:
1773         if (new_fd)
1774                 put_files_struct(new_fd);
1775
1776 bad_unshare_cleanup_fs:
1777         if (new_fs)
1778                 free_fs_struct(new_fs);
1779
1780 bad_unshare_out:
1781         return err;
1782 }
1783
1784 /*
1785  *      Helper to unshare the files of the current task.
1786  *      We don't want to expose copy_files internals to
1787  *      the exec layer of the kernel.
1788  */
1789
1790 int unshare_files(struct files_struct **displaced)
1791 {
1792         struct task_struct *task = current;
1793         struct files_struct *copy = NULL;
1794         int error;
1795
1796         error = unshare_fd(CLONE_FILES, &copy);
1797         if (error || !copy) {
1798                 *displaced = NULL;
1799                 return error;
1800         }
1801         *displaced = task->files;
1802         task_lock(task);
1803         task->files = copy;
1804         task_unlock(task);
1805         return 0;
1806 }