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