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