kernel/fork.c

   1 /*
   2  *  linux/kernel/fork.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  *  'fork.c' contains the help-routines for the 'fork' system call
   9  * (see also entry.S and others).
  10  * Fork is rather simple, once you get the hang of it, but the memory
  11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  12  */
  13
  14 #include <linux/slab.h>
  15 #include <linux/init.h>
  16 #include <linux/unistd.h>
  17 #include <linux/module.h>
  18 #include <linux/vmalloc.h>
  19 #include <linux/completion.h>
  20 #include <linux/personality.h>
  21 #include <linux/mempolicy.h>
  22 #include <linux/sem.h>
  23 #include <linux/file.h>
  24 #include <linux/fdtable.h>
  25 #include <linux/iocontext.h>
  26 #include <linux/key.h>
  27 #include <linux/binfmts.h>
  28 #include <linux/mman.h>
  29 #include <linux/mmu_notifier.h>
  30 #include <linux/fs.h>
  31 #include <linux/nsproxy.h>
  32 #include <linux/capability.h>
  33 #include <linux/cpu.h>
  34 #include <linux/cgroup.h>
  35 #include <linux/security.h>
  36 #include <linux/hugetlb.h>
  37 #include <linux/swap.h>
  38 #include <linux/syscalls.h>
  39 #include <linux/jiffies.h>
  40 #include <linux/futex.h>
  41 #include <linux/compat.h>
  42 #include <linux/kthread.h>
  43 #include <linux/task_io_accounting_ops.h>
  44 #include <linux/rcupdate.h>
  45 #include <linux/ptrace.h>
  46 #include <linux/mount.h>
  47 #include <linux/audit.h>
  48 #include <linux/memcontrol.h>
  49 #include <linux/ftrace.h>
  50 #include <linux/profile.h>
  51 #include <linux/rmap.h>
  52 #include <linux/ksm.h>
  53 #include <linux/acct.h>
  54 #include <linux/tsacct_kern.h>
  55 #include <linux/cn_proc.h>
  56 #include <linux/freezer.h>
  57 #include <linux/delayacct.h>
  58 #include <linux/taskstats_kern.h>
  59 #include <linux/random.h>
  60 #include <linux/tty.h>
  61 #include <linux/blkdev.h>
  62 #include <linux/fs_struct.h>
  63 #include <linux/magic.h>
  64 #include <linux/perf_event.h>
  65 #include <linux/posix-timers.h>
  66 #include <linux/user-return-notifier.h>
  67 #include <linux/oom.h>
  68 #include <linux/khugepaged.h>
  69
  70 #include <asm/pgtable.h>
  71 #include <asm/pgalloc.h>
  72 #include <asm/uaccess.h>
  73 #include <asm/mmu_context.h>
  74 #include <asm/cacheflush.h>
  75 #include <asm/tlbflush.h>
  76
  77 #include <trace/events/sched.h>
  78
  79 #define CREATE_TRACE_POINTS
  80 #include <trace/events/task.h>
  81
  82 /*
  83  * Protected counters by write_lock_irq(&tasklist_lock)
  84  */
  85 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  86 int nr_threads;                 /* The idle threads do not count.. */
  87
  88 int max_threads;                /* tunable limit on nr_threads */
  89
  90 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  91
  92 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  93
  94 #ifdef CONFIG_PROVE_RCU
  95 int lockdep_tasklist_lock_is_held(void)
  96 {
  97         return lockdep_is_held(&tasklist_lock);
  98 }
  99 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 100 #endif /* #ifdef CONFIG_PROVE_RCU */
 101
 102 int nr_processes(void)
 103 {
 104         int cpu;
 105         int total = 0;
 106
 107         for_each_possible_cpu(cpu)
 108                 total += per_cpu(process_counts, cpu);
 109
 110         return total;
 111 }
 112
 113 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 114 # define alloc_task_struct_node(node)           \
 115                 kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node)
 116 # define free_task_struct(tsk)                  \
 117                 kmem_cache_free(task_struct_cachep, (tsk))
 118 static struct kmem_cache *task_struct_cachep;
 119 #endif
 120
 121 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
 122 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 123                                                   int node)
 124 {
 125 #ifdef CONFIG_DEBUG_STACK_USAGE
 126         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
 127 #else
 128         gfp_t mask = GFP_KERNEL;
 129 #endif
 130         struct page *page = alloc_pages_node(node, mask, THREAD_SIZE_ORDER);
 131
 132         return page ? page_address(page) : NULL;
 133 }
 134
 135 static inline void free_thread_info(struct thread_info *ti)
 136 {
 137         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 138 }
 139 #endif
 140
 141 /* SLAB cache for signal_struct structures (tsk->signal) */
 142 static struct kmem_cache *signal_cachep;
 143
 144 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 145 struct kmem_cache *sighand_cachep;
 146
 147 /* SLAB cache for files_struct structures (tsk->files) */
 148 struct kmem_cache *files_cachep;
 149
 150 /* SLAB cache for fs_struct structures (tsk->fs) */
 151 struct kmem_cache *fs_cachep;
 152
 153 /* SLAB cache for vm_area_struct structures */
 154 struct kmem_cache *vm_area_cachep;
 155
 156 /* SLAB cache for mm_struct structures (tsk->mm) */
 157 static struct kmem_cache *mm_cachep;
 158
 159 static void account_kernel_stack(struct thread_info *ti, int account)
 160 {
 161         struct zone *zone = page_zone(virt_to_page(ti));
 162
 163         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
 164 }
 165
 166 void free_task(struct task_struct *tsk)
 167 {
 168         account_kernel_stack(tsk->stack, -1);
 169         free_thread_info(tsk->stack);
 170         rt_mutex_debug_task_free(tsk);
 171         ftrace_graph_exit_task(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         exit_creds(tsk);
 196         delayacct_tsk_free(tsk);
 197         put_signal_struct(tsk->signal);
 198
 199         if (!profile_handoff_task(tsk))
 200                 free_task(tsk);
 201 }
 202 EXPORT_SYMBOL_GPL(__put_task_struct);
 203
 204 /*
 205  * macro override instead of weak attribute alias, to workaround
 206  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
 207  */
 208 #ifndef arch_task_cache_init
 209 #define arch_task_cache_init()
 210 #endif
 211
 212 void __init fork_init(unsigned long mempages)
 213 {
 214 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 215 #ifndef ARCH_MIN_TASKALIGN
 216 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 217 #endif
 218         /* create a slab on which task_structs can be allocated */
 219         task_struct_cachep =
 220                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 221                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
 222 #endif
 223
 224         /* do the arch specific task caches init */
 225         arch_task_cache_init();
 226
 227         /*
 228          * The default maximum number of threads is set to a safe
 229          * value: the thread structures can take up at most half
 230          * of memory.
 231          */
 232         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 233
 234         /*
 235          * we need to allow at least 20 threads to boot a system
 236          */
 237         if (max_threads < 20)
 238                 max_threads = 20;
 239
 240         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 241         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 242         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 243                 init_task.signal->rlim[RLIMIT_NPROC];
 244 }
 245
 246 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
 247                                                struct task_struct *src)
 248 {
 249         *dst = *src;
 250         return 0;
 251 }
 252
 253 static struct task_struct *dup_task_struct(struct task_struct *orig)
 254 {
 255         struct task_struct *tsk;
 256         struct thread_info *ti;
 257         unsigned long *stackend;
 258         int node = tsk_fork_get_node(orig);
 259         int err;
 260
 261         prepare_to_copy(orig);
 262
 263         tsk = alloc_task_struct_node(node);
 264         if (!tsk)
 265                 return NULL;
 266
 267         ti = alloc_thread_info_node(tsk, node);
 268         if (!ti) {
 269                 free_task_struct(tsk);
 270                 return NULL;
 271         }
 272
 273         err = arch_dup_task_struct(tsk, orig);
 274         if (err)
 275                 goto out;
 276
 277         tsk->stack = ti;
 278
 279         setup_thread_stack(tsk, orig);
 280         clear_user_return_notifier(tsk);
 281         clear_tsk_need_resched(tsk);
 282         stackend = end_of_stack(tsk);
 283         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 284
 285 #ifdef CONFIG_CC_STACKPROTECTOR
 286         tsk->stack_canary = get_random_int();
 287 #endif
 288
 289         /*
 290          * One for us, one for whoever does the "release_task()" (usually
 291          * parent)
 292          */
 293         atomic_set(&tsk->usage, 2);
 294 #ifdef CONFIG_BLK_DEV_IO_TRACE
 295         tsk->btrace_seq = 0;
 296 #endif
 297         tsk->splice_pipe = NULL;
 298
 299         account_kernel_stack(ti, 1);
 300
 301         return tsk;
 302
 303 out:
 304         free_thread_info(ti);
 305         free_task_struct(tsk);
 306         return NULL;
 307 }
 308
 309 #ifdef CONFIG_MMU
 310 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 311 {
 312         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 313         struct rb_node **rb_link, *rb_parent;
 314         int retval;
 315         unsigned long charge;
 316         struct mempolicy *pol;
 317
 318         down_write(&oldmm->mmap_sem);
 319         flush_cache_dup_mm(oldmm);
 320         /*
 321          * Not linked in yet - no deadlock potential:
 322          */
 323         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 324
 325         mm->locked_vm = 0;
 326         mm->mmap = NULL;
 327         mm->mmap_cache = NULL;
 328         mm->free_area_cache = oldmm->mmap_base;
 329         mm->cached_hole_size = ~0UL;
 330         mm->map_count = 0;
 331         cpumask_clear(mm_cpumask(mm));
 332         mm->mm_rb = RB_ROOT;
 333         rb_link = &mm->mm_rb.rb_node;
 334         rb_parent = NULL;
 335         pprev = &mm->mmap;
 336         retval = ksm_fork(mm, oldmm);
 337         if (retval)
 338                 goto out;
 339         retval = khugepaged_fork(mm, oldmm);
 340         if (retval)
 341                 goto out;
 342
 343         prev = NULL;
 344         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 345                 struct file *file;
 346
 347                 if (mpnt->vm_flags & VM_DONTCOPY) {
 348                         long pages = vma_pages(mpnt);
 349                         mm->total_vm -= pages;
 350                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 351                                                                 -pages);
 352                         continue;
 353                 }
 354                 charge = 0;
 355                 if (mpnt->vm_flags & VM_ACCOUNT) {
 356                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 357                         if (security_vm_enough_memory(len))
 358                                 goto fail_nomem;
 359                         charge = len;
 360                 }
 361                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 362                 if (!tmp)
 363                         goto fail_nomem;
 364                 *tmp = *mpnt;
 365                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 366                 pol = mpol_dup(vma_policy(mpnt));
 367                 retval = PTR_ERR(pol);
 368                 if (IS_ERR(pol))
 369                         goto fail_nomem_policy;
 370                 vma_set_policy(tmp, pol);
 371                 tmp->vm_mm = mm;
 372                 if (anon_vma_fork(tmp, mpnt))
 373                         goto fail_nomem_anon_vma_fork;
 374                 tmp->vm_flags &= ~VM_LOCKED;
 375                 tmp->vm_next = tmp->vm_prev = NULL;
 376                 file = tmp->vm_file;
 377                 if (file) {
 378                         struct inode *inode = file->f_path.dentry->d_inode;
 379                         struct address_space *mapping = file->f_mapping;
 380
 381                         get_file(file);
 382                         if (tmp->vm_flags & VM_DENYWRITE)
 383                                 atomic_dec(&inode->i_writecount);
 384                         mutex_lock(&mapping->i_mmap_mutex);
 385                         if (tmp->vm_flags & VM_SHARED)
 386                                 mapping->i_mmap_writable++;
 387                         flush_dcache_mmap_lock(mapping);
 388                         /* insert tmp into the share list, just after mpnt */
 389                         vma_prio_tree_add(tmp, mpnt);
 390                         flush_dcache_mmap_unlock(mapping);
 391                         mutex_unlock(&mapping->i_mmap_mutex);
 392                 }
 393
 394                 /*
 395                  * Clear hugetlb-related page reserves for children. This only
 396                  * affects MAP_PRIVATE mappings. Faults generated by the child
 397                  * are not guaranteed to succeed, even if read-only
 398                  */
 399                 if (is_vm_hugetlb_page(tmp))
 400                         reset_vma_resv_huge_pages(tmp);
 401
 402                 /*
 403                  * Link in the new vma and copy the page table entries.
 404                  */
 405                 *pprev = tmp;
 406                 pprev = &tmp->vm_next;
 407                 tmp->vm_prev = prev;
 408                 prev = tmp;
 409
 410                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 411                 rb_link = &tmp->vm_rb.rb_right;
 412                 rb_parent = &tmp->vm_rb;
 413
 414                 mm->map_count++;
 415                 retval = copy_page_range(mm, oldmm, mpnt);
 416
 417                 if (tmp->vm_ops && tmp->vm_ops->open)
 418                         tmp->vm_ops->open(tmp);
 419
 420                 if (retval)
 421                         goto out;
 422         }
 423         /* a new mm has just been created */
 424         arch_dup_mmap(oldmm, mm);
 425         retval = 0;
 426 out:
 427         up_write(&mm->mmap_sem);
 428         flush_tlb_mm(oldmm);
 429         up_write(&oldmm->mmap_sem);
 430         return retval;
 431 fail_nomem_anon_vma_fork:
 432         mpol_put(pol);
 433 fail_nomem_policy:
 434         kmem_cache_free(vm_area_cachep, tmp);
 435 fail_nomem:
 436         retval = -ENOMEM;
 437         vm_unacct_memory(charge);
 438         goto out;
 439 }
 440
 441 static inline int mm_alloc_pgd(struct mm_struct *mm)
 442 {
 443         mm->pgd = pgd_alloc(mm);
 444         if (unlikely(!mm->pgd))
 445                 return -ENOMEM;
 446         return 0;
 447 }
 448
 449 static inline void mm_free_pgd(struct mm_struct *mm)
 450 {
 451         pgd_free(mm, mm->pgd);
 452 }
 453 #else
 454 #define dup_mmap(mm, oldmm)     (0)
 455 #define mm_alloc_pgd(mm)        (0)
 456 #define mm_free_pgd(mm)
 457 #endif /* CONFIG_MMU */
 458
 459 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 460
 461 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 462 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 463
 464 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 465
 466 static int __init coredump_filter_setup(char *s)
 467 {
 468         default_dump_filter =
 469                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 470                 MMF_DUMP_FILTER_MASK;
 471         return 1;
 472 }
 473
 474 __setup("coredump_filter=", coredump_filter_setup);
 475
 476 #include <linux/init_task.h>
 477
 478 static void mm_init_aio(struct mm_struct *mm)
 479 {
 480 #ifdef CONFIG_AIO
 481         spin_lock_init(&mm->ioctx_lock);
 482         INIT_HLIST_HEAD(&mm->ioctx_list);
 483 #endif
 484 }
 485
 486 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
 487 {
 488         atomic_set(&mm->mm_users, 1);
 489         atomic_set(&mm->mm_count, 1);
 490         init_rwsem(&mm->mmap_sem);
 491         INIT_LIST_HEAD(&mm->mmlist);
 492         mm->flags = (current->mm) ?
 493                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
 494         mm->core_state = NULL;
 495         mm->nr_ptes = 0;
 496         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 497         spin_lock_init(&mm->page_table_lock);
 498         mm->free_area_cache = TASK_UNMAPPED_BASE;
 499         mm->cached_hole_size = ~0UL;
 500         mm_init_aio(mm);
 501         mm_init_owner(mm, p);
 502
 503         if (likely(!mm_alloc_pgd(mm))) {
 504                 mm->def_flags = 0;
 505                 mmu_notifier_mm_init(mm);
 506                 return mm;
 507         }
 508
 509         free_mm(mm);
 510         return NULL;
 511 }
 512
 513 /*
 514  * Allocate and initialize an mm_struct.
 515  */
 516 struct mm_struct *mm_alloc(void)
 517 {
 518         struct mm_struct *mm;
 519
 520         mm = allocate_mm();
 521         if (!mm)
 522                 return NULL;
 523
 524         memset(mm, 0, sizeof(*mm));
 525         mm_init_cpumask(mm);
 526         return mm_init(mm, current);
 527 }
 528
 529 /*
 530  * Called when the last reference to the mm
 531  * is dropped: either by a lazy thread or by
 532  * mmput. Free the page directory and the mm.
 533  */
 534 void __mmdrop(struct mm_struct *mm)
 535 {
 536         BUG_ON(mm == &init_mm);
 537         mm_free_pgd(mm);
 538         destroy_context(mm);
 539         mmu_notifier_mm_destroy(mm);
 540 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 541         VM_BUG_ON(mm->pmd_huge_pte);
 542 #endif
 543         free_mm(mm);
 544 }
 545 EXPORT_SYMBOL_GPL(__mmdrop);
 546
 547 /*
 548  * Decrement the use count and release all resources for an mm.
 549  */
 550 void mmput(struct mm_struct *mm)
 551 {
 552         might_sleep();
 553
 554         if (atomic_dec_and_test(&mm->mm_users)) {
 555                 exit_aio(mm);
 556                 ksm_exit(mm);
 557                 khugepaged_exit(mm); /* must run before exit_mmap */
 558                 exit_mmap(mm);
 559                 set_mm_exe_file(mm, NULL);
 560                 if (!list_empty(&mm->mmlist)) {
 561                         spin_lock(&mmlist_lock);
 562                         list_del(&mm->mmlist);
 563                         spin_unlock(&mmlist_lock);
 564                 }
 565                 put_swap_token(mm);
 566                 if (mm->binfmt)
 567                         module_put(mm->binfmt->module);
 568                 mmdrop(mm);
 569         }
 570 }
 571 EXPORT_SYMBOL_GPL(mmput);
 572
 573 /*
 574  * We added or removed a vma mapping the executable. The vmas are only mapped
 575  * during exec and are not mapped with the mmap system call.
 576  * Callers must hold down_write() on the mm's mmap_sem for these
 577  */
 578 void added_exe_file_vma(struct mm_struct *mm)
 579 {
 580         mm->num_exe_file_vmas++;
 581 }
 582
 583 void removed_exe_file_vma(struct mm_struct *mm)
 584 {
 585         mm->num_exe_file_vmas--;
 586         if ((mm->num_exe_file_vmas == 0) && mm->exe_file) {
 587                 fput(mm->exe_file);
 588                 mm->exe_file = NULL;
 589         }
 590
 591 }
 592
 593 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 594 {
 595         if (new_exe_file)
 596                 get_file(new_exe_file);
 597         if (mm->exe_file)
 598                 fput(mm->exe_file);
 599         mm->exe_file = new_exe_file;
 600         mm->num_exe_file_vmas = 0;
 601 }
 602
 603 struct file *get_mm_exe_file(struct mm_struct *mm)
 604 {
 605         struct file *exe_file;
 606
 607         /* We need mmap_sem to protect against races with removal of
 608          * VM_EXECUTABLE vmas */
 609         down_read(&mm->mmap_sem);
 610         exe_file = mm->exe_file;
 611         if (exe_file)
 612                 get_file(exe_file);
 613         up_read(&mm->mmap_sem);
 614         return exe_file;
 615 }
 616
 617 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
 618 {
 619         /* It's safe to write the exe_file pointer without exe_file_lock because
 620          * this is called during fork when the task is not yet in /proc */
 621         newmm->exe_file = get_mm_exe_file(oldmm);
 622 }
 623
 624 /**
 625  * get_task_mm - acquire a reference to the task's mm
 626  *
 627  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 628  * this kernel workthread has transiently adopted a user mm with use_mm,
 629  * to do its AIO) is not set and if so returns a reference to it, after
 630  * bumping up the use count.  User must release the mm via mmput()
 631  * after use.  Typically used by /proc and ptrace.
 632  */
 633 struct mm_struct *get_task_mm(struct task_struct *task)
 634 {
 635         struct mm_struct *mm;
 636
 637         task_lock(task);
 638         mm = task->mm;
 639         if (mm) {
 640                 if (task->flags & PF_KTHREAD)
 641                         mm = NULL;
 642                 else
 643                         atomic_inc(&mm->mm_users);
 644         }
 645         task_unlock(task);
 646         return mm;
 647 }
 648 EXPORT_SYMBOL_GPL(get_task_mm);
 649
 650 /* Please note the differences between mmput and mm_release.
 651  * mmput is called whenever we stop holding onto a mm_struct,
 652  * error success whatever.
 653  *
 654  * mm_release is called after a mm_struct has been removed
 655  * from the current process.
 656  *
 657  * This difference is important for error handling, when we
 658  * only half set up a mm_struct for a new process and need to restore
 659  * the old one.  Because we mmput the new mm_struct before
 660  * restoring the old one. . .
 661  * Eric Biederman 10 January 1998
 662  */
 663 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 664 {
 665         struct completion *vfork_done = tsk->vfork_done;
 666
 667         /* Get rid of any futexes when releasing the mm */
 668 #ifdef CONFIG_FUTEX
 669         if (unlikely(tsk->robust_list)) {
 670                 exit_robust_list(tsk);
 671                 tsk->robust_list = NULL;
 672         }
 673 #ifdef CONFIG_COMPAT
 674         if (unlikely(tsk->compat_robust_list)) {
 675                 compat_exit_robust_list(tsk);
 676                 tsk->compat_robust_list = NULL;
 677         }
 678 #endif
 679         if (unlikely(!list_empty(&tsk->pi_state_list)))
 680                 exit_pi_state_list(tsk);
 681 #endif
 682
 683         /* Get rid of any cached register state */
 684         deactivate_mm(tsk, mm);
 685
 686         /* notify parent sleeping on vfork() */
 687         if (vfork_done) {
 688                 tsk->vfork_done = NULL;
 689                 complete(vfork_done);
 690         }
 691
 692         /*
 693          * If we're exiting normally, clear a user-space tid field if
 694          * requested.  We leave this alone when dying by signal, to leave
 695          * the value intact in a core dump, and to save the unnecessary
 696          * trouble otherwise.  Userland only wants this done for a sys_exit.
 697          */
 698         if (tsk->clear_child_tid) {
 699                 if (!(tsk->flags & PF_SIGNALED) &&
 700                     atomic_read(&mm->mm_users) > 1) {
 701                         /*
 702                          * We don't check the error code - if userspace has
 703                          * not set up a proper pointer then tough luck.
 704                          */
 705                         put_user(0, tsk->clear_child_tid);
 706                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 707                                         1, NULL, NULL, 0);
 708                 }
 709                 tsk->clear_child_tid = NULL;
 710         }
 711 }
 712
 713 /*
 714  * Allocate a new mm structure and copy contents from the
 715  * mm structure of the passed in task structure.
 716  */
 717 struct mm_struct *dup_mm(struct task_struct *tsk)
 718 {
 719         struct mm_struct *mm, *oldmm = current->mm;
 720         int err;
 721
 722         if (!oldmm)
 723                 return NULL;
 724
 725         mm = allocate_mm();
 726         if (!mm)
 727                 goto fail_nomem;
 728
 729         memcpy(mm, oldmm, sizeof(*mm));
 730         mm_init_cpumask(mm);
 731
 732         /* Initializing for Swap token stuff */
 733         mm->token_priority = 0;
 734         mm->last_interval = 0;
 735
 736 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 737         mm->pmd_huge_pte = NULL;
 738 #endif
 739
 740         if (!mm_init(mm, tsk))
 741                 goto fail_nomem;
 742
 743         if (init_new_context(tsk, mm))
 744                 goto fail_nocontext;
 745
 746         dup_mm_exe_file(oldmm, mm);
 747
 748         err = dup_mmap(mm, oldmm);
 749         if (err)
 750                 goto free_pt;
 751
 752         mm->hiwater_rss = get_mm_rss(mm);
 753         mm->hiwater_vm = mm->total_vm;
 754
 755         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 756                 goto free_pt;
 757
 758         return mm;
 759
 760 free_pt:
 761         /* don't put binfmt in mmput, we haven't got module yet */
 762         mm->binfmt = NULL;
 763         mmput(mm);
 764
 765 fail_nomem:
 766         return NULL;
 767
 768 fail_nocontext:
 769         /*
 770          * If init_new_context() failed, we cannot use mmput() to free the mm
 771          * because it calls destroy_context()
 772          */
 773         mm_free_pgd(mm);
 774         free_mm(mm);
 775         return NULL;
 776 }
 777
 778 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 779 {
 780         struct mm_struct *mm, *oldmm;
 781         int retval;
 782
 783         tsk->min_flt = tsk->maj_flt = 0;
 784         tsk->nvcsw = tsk->nivcsw = 0;
 785 #ifdef CONFIG_DETECT_HUNG_TASK
 786         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 787 #endif
 788
 789         tsk->mm = NULL;
 790         tsk->active_mm = NULL;
 791
 792         /*
 793          * Are we cloning a kernel thread?
 794          *
 795          * We need to steal a active VM for that..
 796          */
 797         oldmm = current->mm;
 798         if (!oldmm)
 799                 return 0;
 800
 801         if (clone_flags & CLONE_VM) {
 802                 atomic_inc(&oldmm->mm_users);
 803                 mm = oldmm;
 804                 goto good_mm;
 805         }
 806
 807         retval = -ENOMEM;
 808         mm = dup_mm(tsk);
 809         if (!mm)
 810                 goto fail_nomem;
 811
 812 good_mm:
 813         /* Initializing for Swap token stuff */
 814         mm->token_priority = 0;
 815         mm->last_interval = 0;
 816
 817         tsk->mm = mm;
 818         tsk->active_mm = mm;
 819         return 0;
 820
 821 fail_nomem:
 822         return retval;
 823 }
 824
 825 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 826 {
 827         struct fs_struct *fs = current->fs;
 828         if (clone_flags & CLONE_FS) {
 829                 /* tsk->fs is already what we want */
 830                 spin_lock(&fs->lock);
 831                 if (fs->in_exec) {
 832                         spin_unlock(&fs->lock);
 833                         return -EAGAIN;
 834                 }
 835                 fs->users++;
 836                 spin_unlock(&fs->lock);
 837                 return 0;
 838         }
 839         tsk->fs = copy_fs_struct(fs);
 840         if (!tsk->fs)
 841                 return -ENOMEM;
 842         return 0;
 843 }
 844
 845 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
 846 {
 847         struct files_struct *oldf, *newf;
 848         int error = 0;
 849
 850         /*
 851          * A background process may not have any files ...
 852          */
 853         oldf = current->files;
 854         if (!oldf)
 855                 goto out;
 856
 857         if (clone_flags & CLONE_FILES) {
 858                 atomic_inc(&oldf->count);
 859                 goto out;
 860         }
 861
 862         newf = dup_fd(oldf, &error);
 863         if (!newf)
 864                 goto out;
 865
 866         tsk->files = newf;
 867         error = 0;
 868 out:
 869         return error;
 870 }
 871
 872 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 873 {
 874 #ifdef CONFIG_BLOCK
 875         struct io_context *ioc = current->io_context;
 876         struct io_context *new_ioc;
 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                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
 889                 if (unlikely(!new_ioc))
 890                         return -ENOMEM;
 891
 892                 new_ioc->ioprio = ioc->ioprio;
 893                 put_io_context(new_ioc, NULL);
 894         }
 895 #endif
 896         return 0;
 897 }
 898
 899 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 900 {
 901         struct sighand_struct *sig;
 902
 903         if (clone_flags & CLONE_SIGHAND) {
 904                 atomic_inc(&current->sighand->count);
 905                 return 0;
 906         }
 907         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 908         rcu_assign_pointer(tsk->sighand, sig);
 909         if (!sig)
 910                 return -ENOMEM;
 911         atomic_set(&sig->count, 1);
 912         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 913         return 0;
 914 }
 915
 916 void __cleanup_sighand(struct sighand_struct *sighand)
 917 {
 918         if (atomic_dec_and_test(&sighand->count))
 919                 kmem_cache_free(sighand_cachep, sighand);
 920 }
 921
 922
 923 /*
 924  * Initialize POSIX timer handling for a thread group.
 925  */
 926 static void posix_cpu_timers_init_group(struct signal_struct *sig)
 927 {
 928         unsigned long cpu_limit;
 929
 930         /* Thread group counters. */
 931         thread_group_cputime_init(sig);
 932
 933         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
 934         if (cpu_limit != RLIM_INFINITY) {
 935                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
 936                 sig->cputimer.running = 1;
 937         }
 938
 939         /* The timer lists. */
 940         INIT_LIST_HEAD(&sig->cpu_timers[0]);
 941         INIT_LIST_HEAD(&sig->cpu_timers[1]);
 942         INIT_LIST_HEAD(&sig->cpu_timers[2]);
 943 }
 944
 945 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
 946 {
 947         struct signal_struct *sig;
 948
 949         if (clone_flags & CLONE_THREAD)
 950                 return 0;
 951
 952         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
 953         tsk->signal = sig;
 954         if (!sig)
 955                 return -ENOMEM;
 956
 957         sig->nr_threads = 1;
 958         atomic_set(&sig->live, 1);
 959         atomic_set(&sig->sigcnt, 1);
 960         init_waitqueue_head(&sig->wait_chldexit);
 961         if (clone_flags & CLONE_NEWPID)
 962                 sig->flags |= SIGNAL_UNKILLABLE;
 963         sig->curr_target = tsk;
 964         init_sigpending(&sig->shared_pending);
 965         INIT_LIST_HEAD(&sig->posix_timers);
 966
 967         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 968         sig->real_timer.function = it_real_fn;
 969
 970         task_lock(current->group_leader);
 971         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
 972         task_unlock(current->group_leader);
 973
 974         posix_cpu_timers_init_group(sig);
 975
 976         tty_audit_fork(sig);
 977         sched_autogroup_fork(sig);
 978
 979 #ifdef CONFIG_CGROUPS
 980         init_rwsem(&sig->group_rwsem);
 981 #endif
 982
 983         sig->oom_adj = current->signal->oom_adj;
 984         sig->oom_score_adj = current->signal->oom_score_adj;
 985         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
 986
 987         mutex_init(&sig->cred_guard_mutex);
 988
 989         return 0;
 990 }
 991
 992 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
 993 {
 994         unsigned long new_flags = p->flags;
 995
 996         new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
 997         new_flags |= PF_FORKNOEXEC;
 998         new_flags |= PF_STARTING;
 999         p->flags = new_flags;
1000 }
1001
1002 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1003 {
1004         current->clear_child_tid = tidptr;
1005
1006         return task_pid_vnr(current);
1007 }
1008
1009 static void rt_mutex_init_task(struct task_struct *p)
1010 {
1011         raw_spin_lock_init(&p->pi_lock);
1012 #ifdef CONFIG_RT_MUTEXES
1013         plist_head_init(&p->pi_waiters);
1014         p->pi_blocked_on = NULL;
1015 #endif
1016 }
1017
1018 #ifdef CONFIG_MM_OWNER
1019 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1020 {
1021         mm->owner = p;
1022 }
1023 #endif /* CONFIG_MM_OWNER */
1024
1025 /*
1026  * Initialize POSIX timer handling for a single task.
1027  */
1028 static void posix_cpu_timers_init(struct task_struct *tsk)
1029 {
1030         tsk->cputime_expires.prof_exp = 0;
1031         tsk->cputime_expires.virt_exp = 0;
1032         tsk->cputime_expires.sched_exp = 0;
1033         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1034         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1035         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1036 }
1037
1038 /*
1039  * This creates a new process as a copy of the old one,
1040  * but does not actually start it yet.
1041  *
1042  * It copies the registers, and all the appropriate
1043  * parts of the process environment (as per the clone
1044  * flags). The actual kick-off is left to the caller.
1045  */
1046 static struct task_struct *copy_process(unsigned long clone_flags,
1047                                         unsigned long stack_start,
1048                                         struct pt_regs *regs,
1049                                         unsigned long stack_size,
1050                                         int __user *child_tidptr,
1051                                         struct pid *pid,
1052                                         int trace)
1053 {
1054         int retval;
1055         struct task_struct *p;
1056         int cgroup_callbacks_done = 0;
1057
1058         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1059                 return ERR_PTR(-EINVAL);
1060
1061         /*
1062          * Thread groups must share signals as well, and detached threads
1063          * can only be started up within the thread group.
1064          */
1065         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1066                 return ERR_PTR(-EINVAL);
1067
1068         /*
1069          * Shared signal handlers imply shared VM. By way of the above,
1070          * thread groups also imply shared VM. Blocking this case allows
1071          * for various simplifications in other code.
1072          */
1073         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1074                 return ERR_PTR(-EINVAL);
1075
1076         /*
1077          * Siblings of global init remain as zombies on exit since they are
1078          * not reaped by their parent (swapper). To solve this and to avoid
1079          * multi-rooted process trees, prevent global and container-inits
1080          * from creating siblings.
1081          */
1082         if ((clone_flags & CLONE_PARENT) &&
1083                                 current->signal->flags & SIGNAL_UNKILLABLE)
1084                 return ERR_PTR(-EINVAL);
1085
1086         retval = security_task_create(clone_flags);
1087         if (retval)
1088                 goto fork_out;
1089
1090         retval = -ENOMEM;
1091         p = dup_task_struct(current);
1092         if (!p)
1093                 goto fork_out;
1094
1095         ftrace_graph_init_task(p);
1096
1097         rt_mutex_init_task(p);
1098
1099 #ifdef CONFIG_PROVE_LOCKING
1100         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1101         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1102 #endif
1103         retval = -EAGAIN;
1104         if (atomic_read(&p->real_cred->user->processes) >=
1105                         task_rlimit(p, RLIMIT_NPROC)) {
1106                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1107                     p->real_cred->user != INIT_USER)
1108                         goto bad_fork_free;
1109         }
1110         current->flags &= ~PF_NPROC_EXCEEDED;
1111
1112         retval = copy_creds(p, clone_flags);
1113         if (retval < 0)
1114                 goto bad_fork_free;
1115
1116         /*
1117          * If multiple threads are within copy_process(), then this check
1118          * triggers too late. This doesn't hurt, the check is only there
1119          * to stop root fork bombs.
1120          */
1121         retval = -EAGAIN;
1122         if (nr_threads >= max_threads)
1123                 goto bad_fork_cleanup_count;
1124
1125         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1126                 goto bad_fork_cleanup_count;
1127
1128         p->did_exec = 0;
1129         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1130         copy_flags(clone_flags, p);
1131         INIT_LIST_HEAD(&p->children);
1132         INIT_LIST_HEAD(&p->sibling);
1133         rcu_copy_process(p);
1134         p->vfork_done = NULL;
1135         spin_lock_init(&p->alloc_lock);
1136
1137         init_sigpending(&p->pending);
1138
1139         p->utime = p->stime = p->gtime = 0;
1140         p->utimescaled = p->stimescaled = 0;
1141 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1142         p->prev_utime = p->prev_stime = 0;
1143 #endif
1144 #if defined(SPLIT_RSS_COUNTING)
1145         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1146 #endif
1147
1148         p->default_timer_slack_ns = current->timer_slack_ns;
1149
1150         task_io_accounting_init(&p->ioac);
1151         acct_clear_integrals(p);
1152
1153         posix_cpu_timers_init(p);
1154
1155         do_posix_clock_monotonic_gettime(&p->start_time);
1156         p->real_start_time = p->start_time;
1157         monotonic_to_bootbased(&p->real_start_time);
1158         p->io_context = NULL;
1159         p->audit_context = NULL;
1160         if (clone_flags & CLONE_THREAD)
1161                 threadgroup_change_begin(current);
1162         cgroup_fork(p);
1163 #ifdef CONFIG_NUMA
1164         p->mempolicy = mpol_dup(p->mempolicy);
1165         if (IS_ERR(p->mempolicy)) {
1166                 retval = PTR_ERR(p->mempolicy);
1167                 p->mempolicy = NULL;
1168                 goto bad_fork_cleanup_cgroup;
1169         }
1170         mpol_fix_fork_child_flag(p);
1171 #endif
1172 #ifdef CONFIG_CPUSETS
1173         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1174         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1175 #endif
1176 #ifdef CONFIG_TRACE_IRQFLAGS
1177         p->irq_events = 0;
1178 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1179         p->hardirqs_enabled = 1;
1180 #else
1181         p->hardirqs_enabled = 0;
1182 #endif
1183         p->hardirq_enable_ip = 0;
1184         p->hardirq_enable_event = 0;
1185         p->hardirq_disable_ip = _THIS_IP_;
1186         p->hardirq_disable_event = 0;
1187         p->softirqs_enabled = 1;
1188         p->softirq_enable_ip = _THIS_IP_;
1189         p->softirq_enable_event = 0;
1190         p->softirq_disable_ip = 0;
1191         p->softirq_disable_event = 0;
1192         p->hardirq_context = 0;
1193         p->softirq_context = 0;
1194 #endif
1195 #ifdef CONFIG_LOCKDEP
1196         p->lockdep_depth = 0; /* no locks held yet */
1197         p->curr_chain_key = 0;
1198         p->lockdep_recursion = 0;
1199 #endif
1200
1201 #ifdef CONFIG_DEBUG_MUTEXES
1202         p->blocked_on = NULL; /* not blocked yet */
1203 #endif
1204 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1205         p->memcg_batch.do_batch = 0;
1206         p->memcg_batch.memcg = NULL;
1207 #endif
1208
1209         /* Perform scheduler related setup. Assign this task to a CPU. */
1210         sched_fork(p);
1211
1212         retval = perf_event_init_task(p);
1213         if (retval)
1214                 goto bad_fork_cleanup_policy;
1215         retval = audit_alloc(p);
1216         if (retval)
1217                 goto bad_fork_cleanup_policy;
1218         /* copy all the process information */
1219         retval = copy_semundo(clone_flags, p);
1220         if (retval)
1221                 goto bad_fork_cleanup_audit;
1222         retval = copy_files(clone_flags, p);
1223         if (retval)
1224                 goto bad_fork_cleanup_semundo;
1225         retval = copy_fs(clone_flags, p);
1226         if (retval)
1227                 goto bad_fork_cleanup_files;
1228         retval = copy_sighand(clone_flags, p);
1229         if (retval)
1230                 goto bad_fork_cleanup_fs;
1231         retval = copy_signal(clone_flags, p);
1232         if (retval)
1233                 goto bad_fork_cleanup_sighand;
1234         retval = copy_mm(clone_flags, p);
1235         if (retval)
1236                 goto bad_fork_cleanup_signal;
1237         retval = copy_namespaces(clone_flags, p);
1238         if (retval)
1239                 goto bad_fork_cleanup_mm;
1240         retval = copy_io(clone_flags, p);
1241         if (retval)
1242                 goto bad_fork_cleanup_namespaces;
1243         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1244         if (retval)
1245                 goto bad_fork_cleanup_io;
1246
1247         if (pid != &init_struct_pid) {
1248                 retval = -ENOMEM;
1249                 pid = alloc_pid(p->nsproxy->pid_ns);
1250                 if (!pid)
1251                         goto bad_fork_cleanup_io;
1252         }
1253
1254         p->pid = pid_nr(pid);
1255         p->tgid = p->pid;
1256         if (clone_flags & CLONE_THREAD)
1257                 p->tgid = current->tgid;
1258
1259         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1260         /*
1261          * Clear TID on mm_release()?
1262          */
1263         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1264 #ifdef CONFIG_BLOCK
1265         p->plug = NULL;
1266 #endif
1267 #ifdef CONFIG_FUTEX
1268         p->robust_list = NULL;
1269 #ifdef CONFIG_COMPAT
1270         p->compat_robust_list = NULL;
1271 #endif
1272         INIT_LIST_HEAD(&p->pi_state_list);
1273         p->pi_state_cache = NULL;
1274 #endif
1275         /*
1276          * sigaltstack should be cleared when sharing the same VM
1277          */
1278         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1279                 p->sas_ss_sp = p->sas_ss_size = 0;
1280
1281         /*
1282          * Syscall tracing and stepping should be turned off in the
1283          * child regardless of CLONE_PTRACE.
1284          */
1285         user_disable_single_step(p);
1286         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1287 #ifdef TIF_SYSCALL_EMU
1288         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1289 #endif
1290         clear_all_latency_tracing(p);
1291
1292         /* ok, now we should be set up.. */
1293         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1294         p->pdeath_signal = 0;
1295         p->exit_state = 0;
1296
1297         p->nr_dirtied = 0;
1298         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1299         p->dirty_paused_when = 0;
1300
1301         /*
1302          * Ok, make it visible to the rest of the system.
1303          * We dont wake it up yet.
1304          */
1305         p->group_leader = p;
1306         INIT_LIST_HEAD(&p->thread_group);
1307
1308         /* Now that the task is set up, run cgroup callbacks if
1309          * necessary. We need to run them before the task is visible
1310          * on the tasklist. */
1311         cgroup_fork_callbacks(p);
1312         cgroup_callbacks_done = 1;
1313
1314         /* Need tasklist lock for parent etc handling! */
1315         write_lock_irq(&tasklist_lock);
1316
1317         /* CLONE_PARENT re-uses the old parent */
1318         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1319                 p->real_parent = current->real_parent;
1320                 p->parent_exec_id = current->parent_exec_id;
1321         } else {
1322                 p->real_parent = current;
1323                 p->parent_exec_id = current->self_exec_id;
1324         }
1325
1326         spin_lock(&current->sighand->siglock);
1327
1328         /*
1329          * Process group and session signals need to be delivered to just the
1330          * parent before the fork or both the parent and the child after the
1331          * fork. Restart if a signal comes in before we add the new process to
1332          * it's process group.
1333          * A fatal signal pending means that current will exit, so the new
1334          * thread can't slip out of an OOM kill (or normal SIGKILL).
1335         */
1336         recalc_sigpending();
1337         if (signal_pending(current)) {
1338                 spin_unlock(&current->sighand->siglock);
1339                 write_unlock_irq(&tasklist_lock);
1340                 retval = -ERESTARTNOINTR;
1341                 goto bad_fork_free_pid;
1342         }
1343
1344         if (clone_flags & CLONE_THREAD) {
1345                 current->signal->nr_threads++;
1346                 atomic_inc(&current->signal->live);
1347                 atomic_inc(&current->signal->sigcnt);
1348                 p->group_leader = current->group_leader;
1349                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1350         }
1351
1352         if (likely(p->pid)) {
1353                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1354
1355                 if (thread_group_leader(p)) {
1356                         if (is_child_reaper(pid))
1357                                 p->nsproxy->pid_ns->child_reaper = p;
1358
1359                         p->signal->leader_pid = pid;
1360                         p->signal->tty = tty_kref_get(current->signal->tty);
1361                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1362                         attach_pid(p, PIDTYPE_SID, task_session(current));
1363                         list_add_tail(&p->sibling, &p->real_parent->children);
1364                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1365                         __this_cpu_inc(process_counts);
1366                 }
1367                 attach_pid(p, PIDTYPE_PID, pid);
1368                 nr_threads++;
1369         }
1370
1371         total_forks++;
1372         spin_unlock(&current->sighand->siglock);
1373         write_unlock_irq(&tasklist_lock);
1374         proc_fork_connector(p);
1375         cgroup_post_fork(p);
1376         if (clone_flags & CLONE_THREAD)
1377                 threadgroup_change_end(current);
1378         perf_event_fork(p);
1379
1380         trace_task_newtask(p, clone_flags);
1381
1382         return p;
1383
1384 bad_fork_free_pid:
1385         if (pid != &init_struct_pid)
1386                 free_pid(pid);
1387 bad_fork_cleanup_io:
1388         if (p->io_context)
1389                 exit_io_context(p);
1390 bad_fork_cleanup_namespaces:
1391         exit_task_namespaces(p);
1392 bad_fork_cleanup_mm:
1393         if (p->mm)
1394                 mmput(p->mm);
1395 bad_fork_cleanup_signal:
1396         if (!(clone_flags & CLONE_THREAD))
1397                 free_signal_struct(p->signal);
1398 bad_fork_cleanup_sighand:
1399         __cleanup_sighand(p->sighand);
1400 bad_fork_cleanup_fs:
1401         exit_fs(p); /* blocking */
1402 bad_fork_cleanup_files:
1403         exit_files(p); /* blocking */
1404 bad_fork_cleanup_semundo:
1405         exit_sem(p);
1406 bad_fork_cleanup_audit:
1407         audit_free(p);
1408 bad_fork_cleanup_policy:
1409         perf_event_free_task(p);
1410 #ifdef CONFIG_NUMA
1411         mpol_put(p->mempolicy);
1412 bad_fork_cleanup_cgroup:
1413 #endif
1414         if (clone_flags & CLONE_THREAD)
1415                 threadgroup_change_end(current);
1416         cgroup_exit(p, cgroup_callbacks_done);
1417         delayacct_tsk_free(p);
1418         module_put(task_thread_info(p)->exec_domain->module);
1419 bad_fork_cleanup_count:
1420         atomic_dec(&p->cred->user->processes);
1421         exit_creds(p);
1422 bad_fork_free:
1423         free_task(p);
1424 fork_out:
1425         return ERR_PTR(retval);
1426 }
1427
1428 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1429 {
1430         memset(regs, 0, sizeof(struct pt_regs));
1431         return regs;
1432 }
1433
1434 static inline void init_idle_pids(struct pid_link *links)
1435 {
1436         enum pid_type type;
1437
1438         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1439                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1440                 links[type].pid = &init_struct_pid;
1441         }
1442 }
1443
1444 struct task_struct * __cpuinit fork_idle(int cpu)
1445 {
1446         struct task_struct *task;
1447         struct pt_regs regs;
1448
1449         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1450                             &init_struct_pid, 0);
1451         if (!IS_ERR(task)) {
1452                 init_idle_pids(task->pids);
1453                 init_idle(task, cpu);
1454         }
1455
1456         return task;
1457 }
1458
1459 /*
1460  *  Ok, this is the main fork-routine.
1461  *
1462  * It copies the process, and if successful kick-starts
1463  * it and waits for it to finish using the VM if required.
1464  */
1465 long do_fork(unsigned long clone_flags,
1466               unsigned long stack_start,
1467               struct pt_regs *regs,
1468               unsigned long stack_size,
1469               int __user *parent_tidptr,
1470               int __user *child_tidptr)
1471 {
1472         struct task_struct *p;
1473         int trace = 0;
1474         long nr;
1475
1476         /*
1477          * Do some preliminary argument and permissions checking before we
1478          * actually start allocating stuff
1479          */
1480         if (clone_flags & CLONE_NEWUSER) {
1481                 if (clone_flags & CLONE_THREAD)
1482                         return -EINVAL;
1483                 /* hopefully this check will go away when userns support is
1484                  * complete
1485                  */
1486                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1487                                 !capable(CAP_SETGID))
1488                         return -EPERM;
1489         }
1490
1491         /*
1492          * Determine whether and which event to report to ptracer.  When
1493          * called from kernel_thread or CLONE_UNTRACED is explicitly
1494          * requested, no event is reported; otherwise, report if the event
1495          * for the type of forking is enabled.
1496          */
1497         if (likely(user_mode(regs)) && !(clone_flags & CLONE_UNTRACED)) {
1498                 if (clone_flags & CLONE_VFORK)
1499                         trace = PTRACE_EVENT_VFORK;
1500                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1501                         trace = PTRACE_EVENT_CLONE;
1502                 else
1503                         trace = PTRACE_EVENT_FORK;
1504
1505                 if (likely(!ptrace_event_enabled(current, trace)))
1506                         trace = 0;
1507         }
1508
1509         p = copy_process(clone_flags, stack_start, regs, stack_size,
1510                          child_tidptr, NULL, trace);
1511         /*
1512          * Do this prior waking up the new thread - the thread pointer
1513          * might get invalid after that point, if the thread exits quickly.
1514          */
1515         if (!IS_ERR(p)) {
1516                 struct completion vfork;
1517
1518                 trace_sched_process_fork(current, p);
1519
1520                 nr = task_pid_vnr(p);
1521
1522                 if (clone_flags & CLONE_PARENT_SETTID)
1523                         put_user(nr, parent_tidptr);
1524
1525                 if (clone_flags & CLONE_VFORK) {
1526                         p->vfork_done = &vfork;
1527                         init_completion(&vfork);
1528                 }
1529
1530                 /*
1531                  * We set PF_STARTING at creation in case tracing wants to
1532                  * use this to distinguish a fully live task from one that
1533                  * hasn't finished SIGSTOP raising yet.  Now we clear it
1534                  * and set the child going.
1535                  */
1536                 p->flags &= ~PF_STARTING;
1537
1538                 wake_up_new_task(p);
1539
1540                 /* forking complete and child started to run, tell ptracer */
1541                 if (unlikely(trace))
1542                         ptrace_event(trace, nr);
1543
1544                 if (clone_flags & CLONE_VFORK) {
1545                         freezer_do_not_count();
1546                         wait_for_completion(&vfork);
1547                         freezer_count();
1548                         ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1549                 }
1550         } else {
1551                 nr = PTR_ERR(p);
1552         }
1553         return nr;
1554 }
1555
1556 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1557 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1558 #endif
1559
1560 static void sighand_ctor(void *data)
1561 {
1562         struct sighand_struct *sighand = data;
1563
1564         spin_lock_init(&sighand->siglock);
1565         init_waitqueue_head(&sighand->signalfd_wqh);
1566 }
1567
1568 void __init proc_caches_init(void)
1569 {
1570         sighand_cachep = kmem_cache_create("sighand_cache",
1571                         sizeof(struct sighand_struct), 0,
1572                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1573                         SLAB_NOTRACK, sighand_ctor);
1574         signal_cachep = kmem_cache_create("signal_cache",
1575                         sizeof(struct signal_struct), 0,
1576                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1577         files_cachep = kmem_cache_create("files_cache",
1578                         sizeof(struct files_struct), 0,
1579                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1580         fs_cachep = kmem_cache_create("fs_cache",
1581                         sizeof(struct fs_struct), 0,
1582                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1583         /*
1584          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1585          * whole struct cpumask for the OFFSTACK case. We could change
1586          * this to *only* allocate as much of it as required by the
1587          * maximum number of CPU's we can ever have.  The cpumask_allocation
1588          * is at the end of the structure, exactly for that reason.
1589          */
1590         mm_cachep = kmem_cache_create("mm_struct",
1591                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1592                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1593         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1594         mmap_init();
1595         nsproxy_cache_init();
1596 }
1597
1598 /*
1599  * Check constraints on flags passed to the unshare system call.
1600  */
1601 static int check_unshare_flags(unsigned long unshare_flags)
1602 {
1603         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1604                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1605                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1606                 return -EINVAL;
1607         /*
1608          * Not implemented, but pretend it works if there is nothing to
1609          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1610          * needs to unshare vm.
1611          */
1612         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1613                 /* FIXME: get_task_mm() increments ->mm_users */
1614                 if (atomic_read(&current->mm->mm_users) > 1)
1615                         return -EINVAL;
1616         }
1617
1618         return 0;
1619 }
1620
1621 /*
1622  * Unshare the filesystem structure if it is being shared
1623  */
1624 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1625 {
1626         struct fs_struct *fs = current->fs;
1627
1628         if (!(unshare_flags & CLONE_FS) || !fs)
1629                 return 0;
1630
1631         /* don't need lock here; in the worst case we'll do useless copy */
1632         if (fs->users == 1)
1633                 return 0;
1634
1635         *new_fsp = copy_fs_struct(fs);
1636         if (!*new_fsp)
1637                 return -ENOMEM;
1638
1639         return 0;
1640 }
1641
1642 /*
1643  * Unshare file descriptor table if it is being shared
1644  */
1645 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1646 {
1647         struct files_struct *fd = current->files;
1648         int error = 0;
1649
1650         if ((unshare_flags & CLONE_FILES) &&
1651             (fd && atomic_read(&fd->count) > 1)) {
1652                 *new_fdp = dup_fd(fd, &error);
1653                 if (!*new_fdp)
1654                         return error;
1655         }
1656
1657         return 0;
1658 }
1659
1660 /*
1661  * unshare allows a process to 'unshare' part of the process
1662  * context which was originally shared using clone.  copy_*
1663  * functions used by do_fork() cannot be used here directly
1664  * because they modify an inactive task_struct that is being
1665  * constructed. Here we are modifying the current, active,
1666  * task_struct.
1667  */
1668 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1669 {
1670         struct fs_struct *fs, *new_fs = NULL;
1671         struct files_struct *fd, *new_fd = NULL;
1672         struct nsproxy *new_nsproxy = NULL;
1673         int do_sysvsem = 0;
1674         int err;
1675
1676         err = check_unshare_flags(unshare_flags);
1677         if (err)
1678                 goto bad_unshare_out;
1679
1680         /*
1681          * If unsharing namespace, must also unshare filesystem information.
1682          */
1683         if (unshare_flags & CLONE_NEWNS)
1684                 unshare_flags |= CLONE_FS;
1685         /*
1686          * CLONE_NEWIPC must also detach from the undolist: after switching
1687          * to a new ipc namespace, the semaphore arrays from the old
1688          * namespace are unreachable.
1689          */
1690         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1691                 do_sysvsem = 1;
1692         err = unshare_fs(unshare_flags, &new_fs);
1693         if (err)
1694                 goto bad_unshare_out;
1695         err = unshare_fd(unshare_flags, &new_fd);
1696         if (err)
1697                 goto bad_unshare_cleanup_fs;
1698         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy, new_fs);
1699         if (err)
1700                 goto bad_unshare_cleanup_fd;
1701
1702         if (new_fs || new_fd || do_sysvsem || new_nsproxy) {
1703                 if (do_sysvsem) {
1704                         /*
1705                          * CLONE_SYSVSEM is equivalent to sys_exit().
1706                          */
1707                         exit_sem(current);
1708                 }
1709
1710                 if (new_nsproxy) {
1711                         switch_task_namespaces(current, new_nsproxy);
1712                         new_nsproxy = NULL;
1713                 }
1714
1715                 task_lock(current);
1716
1717                 if (new_fs) {
1718                         fs = current->fs;
1719                         spin_lock(&fs->lock);
1720                         current->fs = new_fs;
1721                         if (--fs->users)
1722                                 new_fs = NULL;
1723                         else
1724                                 new_fs = fs;
1725                         spin_unlock(&fs->lock);
1726                 }
1727
1728                 if (new_fd) {
1729                         fd = current->files;
1730                         current->files = new_fd;
1731                         new_fd = fd;
1732                 }
1733
1734                 task_unlock(current);
1735         }
1736
1737         if (new_nsproxy)
1738                 put_nsproxy(new_nsproxy);
1739
1740 bad_unshare_cleanup_fd:
1741         if (new_fd)
1742                 put_files_struct(new_fd);
1743
1744 bad_unshare_cleanup_fs:
1745         if (new_fs)
1746                 free_fs_struct(new_fs);
1747
1748 bad_unshare_out:
1749         return err;
1750 }
1751
1752 /*
1753  *      Helper to unshare the files of the current task.
1754  *      We don't want to expose copy_files internals to
1755  *      the exec layer of the kernel.
1756  */
1757
1758 int unshare_files(struct files_struct **displaced)
1759 {
1760         struct task_struct *task = current;
1761         struct files_struct *copy = NULL;
1762         int error;
1763
1764         error = unshare_fd(CLONE_FILES, &copy);
1765         if (error || !copy) {
1766                 *displaced = NULL;
1767                 return error;
1768         }
1769         *displaced = task->files;
1770         task_lock(task);
1771         task->files = copy;
1772         task_unlock(task);
1773         return 0;
1774 }