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/proc_fs.h>
  51 #include <linux/profile.h>
  52 #include <linux/rmap.h>
  53 #include <linux/ksm.h>
  54 #include <linux/acct.h>
  55 #include <linux/tsacct_kern.h>
  56 #include <linux/cn_proc.h>
  57 #include <linux/freezer.h>
  58 #include <linux/delayacct.h>
  59 #include <linux/taskstats_kern.h>
  60 #include <linux/random.h>
  61 #include <linux/tty.h>
  62 #include <linux/blkdev.h>
  63 #include <linux/fs_struct.h>
  64 #include <linux/magic.h>
  65 #include <linux/perf_event.h>
  66 #include <linux/posix-timers.h>
  67 #include <linux/user-return-notifier.h>
  68 #include <linux/oom.h>
  69 #include <linux/khugepaged.h>
  70 #include <linux/signalfd.h>
  71
  72 #include <asm/pgtable.h>
  73 #include <asm/pgalloc.h>
  74 #include <asm/uaccess.h>
  75 #include <asm/mmu_context.h>
  76 #include <asm/cacheflush.h>
  77 #include <asm/tlbflush.h>
  78
  79 #include <trace/events/sched.h>
  80
  81 /*
  82  * Protected counters by write_lock_irq(&tasklist_lock)
  83  */
  84 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  85 int nr_threads;                 /* The idle threads do not count.. */
  86
  87 int max_threads;                /* tunable limit on nr_threads */
  88
  89 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  90
  91 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  92
  93 #ifdef CONFIG_PROVE_RCU
  94 int lockdep_tasklist_lock_is_held(void)
  95 {
  96         return lockdep_is_held(&tasklist_lock);
  97 }
  98 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
  99 #endif /* #ifdef CONFIG_PROVE_RCU */
 100
 101 int nr_processes(void)
 102 {
 103         int cpu;
 104         int total = 0;
 105
 106         for_each_possible_cpu(cpu)
 107                 total += per_cpu(process_counts, cpu);
 108
 109         return total;
 110 }
 111
 112 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 113 # define alloc_task_struct_node(node)           \
 114                 kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node)
 115 # define free_task_struct(tsk)                  \
 116                 kmem_cache_free(task_struct_cachep, (tsk))
 117 static struct kmem_cache *task_struct_cachep;
 118 #endif
 119
 120 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
 121 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 122                                                   int node)
 123 {
 124 #ifdef CONFIG_DEBUG_STACK_USAGE
 125         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
 126 #else
 127         gfp_t mask = GFP_KERNEL;
 128 #endif
 129         struct page *page = alloc_pages_node(node, mask, THREAD_SIZE_ORDER);
 130
 131         return page ? page_address(page) : NULL;
 132 }
 133
 134 static inline void free_thread_info(struct thread_info *ti)
 135 {
 136         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 137 }
 138 #endif
 139
 140 /* SLAB cache for signal_struct structures (tsk->signal) */
 141 static struct kmem_cache *signal_cachep;
 142
 143 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 144 struct kmem_cache *sighand_cachep;
 145
 146 /* SLAB cache for files_struct structures (tsk->files) */
 147 struct kmem_cache *files_cachep;
 148
 149 /* SLAB cache for fs_struct structures (tsk->fs) */
 150 struct kmem_cache *fs_cachep;
 151
 152 /* SLAB cache for vm_area_struct structures */
 153 struct kmem_cache *vm_area_cachep;
 154
 155 /* SLAB cache for mm_struct structures (tsk->mm) */
 156 static struct kmem_cache *mm_cachep;
 157
 158 static void account_kernel_stack(struct thread_info *ti, int account)
 159 {
 160         struct zone *zone = page_zone(virt_to_page(ti));
 161
 162         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
 163 }
 164
 165 void free_task(struct task_struct *tsk)
 166 {
 167         account_kernel_stack(tsk->stack, -1);
 168         free_thread_info(tsk->stack);
 169         rt_mutex_debug_task_free(tsk);
 170         ftrace_graph_exit_task(tsk);
 171         free_task_struct(tsk);
 172 }
 173 EXPORT_SYMBOL(free_task);
 174
 175 static inline void free_signal_struct(struct signal_struct *sig)
 176 {
 177         taskstats_tgid_free(sig);
 178         sched_autogroup_exit(sig);
 179         kmem_cache_free(signal_cachep, sig);
 180 }
 181
 182 static inline void put_signal_struct(struct signal_struct *sig)
 183 {
 184         if (atomic_dec_and_test(&sig->sigcnt))
 185                 free_signal_struct(sig);
 186 }
 187
 188 void __put_task_struct(struct task_struct *tsk)
 189 {
 190         WARN_ON(!tsk->exit_state);
 191         WARN_ON(atomic_read(&tsk->usage));
 192         WARN_ON(tsk == current);
 193
 194         exit_creds(tsk);
 195         delayacct_tsk_free(tsk);
 196         put_signal_struct(tsk->signal);
 197
 198         if (!profile_handoff_task(tsk))
 199                 free_task(tsk);
 200 }
 201 EXPORT_SYMBOL_GPL(__put_task_struct);
 202
 203 /*
 204  * macro override instead of weak attribute alias, to workaround
 205  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
 206  */
 207 #ifndef arch_task_cache_init
 208 #define arch_task_cache_init()
 209 #endif
 210
 211 void __init fork_init(unsigned long mempages)
 212 {
 213 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 214 #ifndef ARCH_MIN_TASKALIGN
 215 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 216 #endif
 217         /* create a slab on which task_structs can be allocated */
 218         task_struct_cachep =
 219                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 220                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
 221 #endif
 222
 223         /* do the arch specific task caches init */
 224         arch_task_cache_init();
 225
 226         /*
 227          * The default maximum number of threads is set to a safe
 228          * value: the thread structures can take up at most half
 229          * of memory.
 230          */
 231         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 232
 233         /*
 234          * we need to allow at least 20 threads to boot a system
 235          */
 236         if (max_threads < 20)
 237                 max_threads = 20;
 238
 239         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 240         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 241         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 242                 init_task.signal->rlim[RLIMIT_NPROC];
 243 }
 244
 245 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
 246                                                struct task_struct *src)
 247 {
 248         *dst = *src;
 249         return 0;
 250 }
 251
 252 static struct task_struct *dup_task_struct(struct task_struct *orig)
 253 {
 254         struct task_struct *tsk;
 255         struct thread_info *ti;
 256         unsigned long *stackend;
 257         int node = tsk_fork_get_node(orig);
 258         int err;
 259
 260         prepare_to_copy(orig);
 261
 262         tsk = alloc_task_struct_node(node);
 263         if (!tsk)
 264                 return NULL;
 265
 266         ti = alloc_thread_info_node(tsk, node);
 267         if (!ti) {
 268                 free_task_struct(tsk);
 269                 return NULL;
 270         }
 271
 272         err = arch_dup_task_struct(tsk, orig);
 273         if (err)
 274                 goto out;
 275
 276         tsk->stack = ti;
 277
 278         setup_thread_stack(tsk, orig);
 279         clear_user_return_notifier(tsk);
 280         clear_tsk_need_resched(tsk);
 281         stackend = end_of_stack(tsk);
 282         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 283
 284 #ifdef CONFIG_CC_STACKPROTECTOR
 285         tsk->stack_canary = get_random_int();
 286 #endif
 287
 288         /*
 289          * One for us, one for whoever does the "release_task()" (usually
 290          * parent)
 291          */
 292         atomic_set(&tsk->usage, 2);
 293 #ifdef CONFIG_BLK_DEV_IO_TRACE
 294         tsk->btrace_seq = 0;
 295 #endif
 296         tsk->splice_pipe = NULL;
 297
 298         account_kernel_stack(ti, 1);
 299
 300         return tsk;
 301
 302 out:
 303         free_thread_info(ti);
 304         free_task_struct(tsk);
 305         return NULL;
 306 }
 307
 308 #ifdef CONFIG_MMU
 309 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 310 {
 311         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 312         struct rb_node **rb_link, *rb_parent;
 313         int retval;
 314         unsigned long charge;
 315         struct mempolicy *pol;
 316
 317         down_write(&oldmm->mmap_sem);
 318         flush_cache_dup_mm(oldmm);
 319         /*
 320          * Not linked in yet - no deadlock potential:
 321          */
 322         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 323
 324         mm->locked_vm = 0;
 325         mm->mmap = NULL;
 326         mm->mmap_cache = NULL;
 327         mm->free_area_cache = oldmm->mmap_base;
 328         mm->cached_hole_size = ~0UL;
 329         mm->map_count = 0;
 330         cpumask_clear(mm_cpumask(mm));
 331         mm->mm_rb = RB_ROOT;
 332         rb_link = &mm->mm_rb.rb_node;
 333         rb_parent = NULL;
 334         pprev = &mm->mmap;
 335         retval = ksm_fork(mm, oldmm);
 336         if (retval)
 337                 goto out;
 338         retval = khugepaged_fork(mm, oldmm);
 339         if (retval)
 340                 goto out;
 341
 342         prev = NULL;
 343         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 344                 struct file *file;
 345
 346                 if (mpnt->vm_flags & VM_DONTCOPY) {
 347                         long pages = vma_pages(mpnt);
 348                         mm->total_vm -= pages;
 349                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 350                                                                 -pages);
 351                         continue;
 352                 }
 353                 charge = 0;
 354                 if (mpnt->vm_flags & VM_ACCOUNT) {
 355                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 356                         if (security_vm_enough_memory(len))
 357                                 goto fail_nomem;
 358                         charge = len;
 359                 }
 360                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 361                 if (!tmp)
 362                         goto fail_nomem;
 363                 *tmp = *mpnt;
 364                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 365                 pol = mpol_dup(vma_policy(mpnt));
 366                 retval = PTR_ERR(pol);
 367                 if (IS_ERR(pol))
 368                         goto fail_nomem_policy;
 369                 vma_set_policy(tmp, pol);
 370                 tmp->vm_mm = mm;
 371                 if (anon_vma_fork(tmp, mpnt))
 372                         goto fail_nomem_anon_vma_fork;
 373                 tmp->vm_flags &= ~VM_LOCKED;
 374                 tmp->vm_next = tmp->vm_prev = NULL;
 375                 file = tmp->vm_file;
 376                 if (file) {
 377                         struct inode *inode = file->f_path.dentry->d_inode;
 378                         struct address_space *mapping = file->f_mapping;
 379
 380                         get_file(file);
 381                         if (tmp->vm_flags & VM_DENYWRITE)
 382                                 atomic_dec(&inode->i_writecount);
 383                         mutex_lock(&mapping->i_mmap_mutex);
 384                         if (tmp->vm_flags & VM_SHARED)
 385                                 mapping->i_mmap_writable++;
 386                         flush_dcache_mmap_lock(mapping);
 387                         /* insert tmp into the share list, just after mpnt */
 388                         vma_prio_tree_add(tmp, mpnt);
 389                         flush_dcache_mmap_unlock(mapping);
 390                         mutex_unlock(&mapping->i_mmap_mutex);
 391                 }
 392
 393                 /*
 394                  * Clear hugetlb-related page reserves for children. This only
 395                  * affects MAP_PRIVATE mappings. Faults generated by the child
 396                  * are not guaranteed to succeed, even if read-only
 397                  */
 398                 if (is_vm_hugetlb_page(tmp))
 399                         reset_vma_resv_huge_pages(tmp);
 400
 401                 /*
 402                  * Link in the new vma and copy the page table entries.
 403                  */
 404                 *pprev = tmp;
 405                 pprev = &tmp->vm_next;
 406                 tmp->vm_prev = prev;
 407                 prev = tmp;
 408
 409                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 410                 rb_link = &tmp->vm_rb.rb_right;
 411                 rb_parent = &tmp->vm_rb;
 412
 413                 mm->map_count++;
 414                 retval = copy_page_range(mm, oldmm, mpnt);
 415
 416                 if (tmp->vm_ops && tmp->vm_ops->open)
 417                         tmp->vm_ops->open(tmp);
 418
 419                 if (retval)
 420                         goto out;
 421         }
 422         /* a new mm has just been created */
 423         arch_dup_mmap(oldmm, mm);
 424         retval = 0;
 425 out:
 426         up_write(&mm->mmap_sem);
 427         flush_tlb_mm(oldmm);
 428         up_write(&oldmm->mmap_sem);
 429         return retval;
 430 fail_nomem_anon_vma_fork:
 431         mpol_put(pol);
 432 fail_nomem_policy:
 433         kmem_cache_free(vm_area_cachep, tmp);
 434 fail_nomem:
 435         retval = -ENOMEM;
 436         vm_unacct_memory(charge);
 437         goto out;
 438 }
 439
 440 static inline int mm_alloc_pgd(struct mm_struct *mm)
 441 {
 442         mm->pgd = pgd_alloc(mm);
 443         if (unlikely(!mm->pgd))
 444                 return -ENOMEM;
 445         return 0;
 446 }
 447
 448 static inline void mm_free_pgd(struct mm_struct *mm)
 449 {
 450         pgd_free(mm, mm->pgd);
 451 }
 452 #else
 453 #define dup_mmap(mm, oldmm)     (0)
 454 #define mm_alloc_pgd(mm)        (0)
 455 #define mm_free_pgd(mm)
 456 #endif /* CONFIG_MMU */
 457
 458 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 459
 460 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 461 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 462
 463 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 464
 465 static int __init coredump_filter_setup(char *s)
 466 {
 467         default_dump_filter =
 468                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 469                 MMF_DUMP_FILTER_MASK;
 470         return 1;
 471 }
 472
 473 __setup("coredump_filter=", coredump_filter_setup);
 474
 475 #include <linux/init_task.h>
 476
 477 static void mm_init_aio(struct mm_struct *mm)
 478 {
 479 #ifdef CONFIG_AIO
 480         spin_lock_init(&mm->ioctx_lock);
 481         INIT_HLIST_HEAD(&mm->ioctx_list);
 482 #endif
 483 }
 484
 485 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
 486 {
 487         atomic_set(&mm->mm_users, 1);
 488         atomic_set(&mm->mm_count, 1);
 489         init_rwsem(&mm->mmap_sem);
 490         INIT_LIST_HEAD(&mm->mmlist);
 491         mm->flags = (current->mm) ?
 492                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
 493         mm->core_state = NULL;
 494         mm->nr_ptes = 0;
 495         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 496         spin_lock_init(&mm->page_table_lock);
 497         mm->free_area_cache = TASK_UNMAPPED_BASE;
 498         mm->cached_hole_size = ~0UL;
 499         mm_init_aio(mm);
 500         mm_init_owner(mm, p);
 501
 502         if (likely(!mm_alloc_pgd(mm))) {
 503                 mm->def_flags = 0;
 504                 mmu_notifier_mm_init(mm);
 505                 return mm;
 506         }
 507
 508         free_mm(mm);
 509         return NULL;
 510 }
 511
 512 /*
 513  * Allocate and initialize an mm_struct.
 514  */
 515 struct mm_struct *mm_alloc(void)
 516 {
 517         struct mm_struct *mm;
 518
 519         mm = allocate_mm();
 520         if (!mm)
 521                 return NULL;
 522
 523         memset(mm, 0, sizeof(*mm));
 524         mm_init_cpumask(mm);
 525         return mm_init(mm, current);
 526 }
 527
 528 /*
 529  * Called when the last reference to the mm
 530  * is dropped: either by a lazy thread or by
 531  * mmput. Free the page directory and the mm.
 532  */
 533 void __mmdrop(struct mm_struct *mm)
 534 {
 535         BUG_ON(mm == &init_mm);
 536         mm_free_pgd(mm);
 537         destroy_context(mm);
 538         mmu_notifier_mm_destroy(mm);
 539 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 540         VM_BUG_ON(mm->pmd_huge_pte);
 541 #endif
 542         free_mm(mm);
 543 }
 544 EXPORT_SYMBOL_GPL(__mmdrop);
 545
 546 /*
 547  * Decrement the use count and release all resources for an mm.
 548  */
 549 void mmput(struct mm_struct *mm)
 550 {
 551         might_sleep();
 552
 553         if (atomic_dec_and_test(&mm->mm_users)) {
 554                 exit_aio(mm);
 555                 ksm_exit(mm);
 556                 khugepaged_exit(mm); /* must run before exit_mmap */
 557                 exit_mmap(mm);
 558                 set_mm_exe_file(mm, NULL);
 559                 if (!list_empty(&mm->mmlist)) {
 560                         spin_lock(&mmlist_lock);
 561                         list_del(&mm->mmlist);
 562                         spin_unlock(&mmlist_lock);
 563                 }
 564                 put_swap_token(mm);
 565                 if (mm->binfmt)
 566                         module_put(mm->binfmt->module);
 567                 mmdrop(mm);
 568         }
 569 }
 570 EXPORT_SYMBOL_GPL(mmput);
 571
 572 /*
 573  * We added or removed a vma mapping the executable. The vmas are only mapped
 574  * during exec and are not mapped with the mmap system call.
 575  * Callers must hold down_write() on the mm's mmap_sem for these
 576  */
 577 void added_exe_file_vma(struct mm_struct *mm)
 578 {
 579         mm->num_exe_file_vmas++;
 580 }
 581
 582 void removed_exe_file_vma(struct mm_struct *mm)
 583 {
 584         mm->num_exe_file_vmas--;
 585         if ((mm->num_exe_file_vmas == 0) && mm->exe_file) {
 586                 fput(mm->exe_file);
 587                 mm->exe_file = NULL;
 588         }
 589
 590 }
 591
 592 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 593 {
 594         if (new_exe_file)
 595                 get_file(new_exe_file);
 596         if (mm->exe_file)
 597                 fput(mm->exe_file);
 598         mm->exe_file = new_exe_file;
 599         mm->num_exe_file_vmas = 0;
 600 }
 601
 602 struct file *get_mm_exe_file(struct mm_struct *mm)
 603 {
 604         struct file *exe_file;
 605
 606         /* We need mmap_sem to protect against races with removal of
 607          * VM_EXECUTABLE vmas */
 608         down_read(&mm->mmap_sem);
 609         exe_file = mm->exe_file;
 610         if (exe_file)
 611                 get_file(exe_file);
 612         up_read(&mm->mmap_sem);
 613         return exe_file;
 614 }
 615
 616 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
 617 {
 618         /* It's safe to write the exe_file pointer without exe_file_lock because
 619          * this is called during fork when the task is not yet in /proc */
 620         newmm->exe_file = get_mm_exe_file(oldmm);
 621 }
 622
 623 /**
 624  * get_task_mm - acquire a reference to the task's mm
 625  *
 626  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 627  * this kernel workthread has transiently adopted a user mm with use_mm,
 628  * to do its AIO) is not set and if so returns a reference to it, after
 629  * bumping up the use count.  User must release the mm via mmput()
 630  * after use.  Typically used by /proc and ptrace.
 631  */
 632 struct mm_struct *get_task_mm(struct task_struct *task)
 633 {
 634         struct mm_struct *mm;
 635
 636         task_lock(task);
 637         mm = task->mm;
 638         if (mm) {
 639                 if (task->flags & PF_KTHREAD)
 640                         mm = NULL;
 641                 else
 642                         atomic_inc(&mm->mm_users);
 643         }
 644         task_unlock(task);
 645         return mm;
 646 }
 647 EXPORT_SYMBOL_GPL(get_task_mm);
 648
 649 /* Please note the differences between mmput and mm_release.
 650  * mmput is called whenever we stop holding onto a mm_struct,
 651  * error success whatever.
 652  *
 653  * mm_release is called after a mm_struct has been removed
 654  * from the current process.
 655  *
 656  * This difference is important for error handling, when we
 657  * only half set up a mm_struct for a new process and need to restore
 658  * the old one.  Because we mmput the new mm_struct before
 659  * restoring the old one. . .
 660  * Eric Biederman 10 January 1998
 661  */
 662 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 663 {
 664         struct completion *vfork_done = tsk->vfork_done;
 665
 666         /* Get rid of any futexes when releasing the mm */
 667 #ifdef CONFIG_FUTEX
 668         if (unlikely(tsk->robust_list)) {
 669                 exit_robust_list(tsk);
 670                 tsk->robust_list = NULL;
 671         }
 672 #ifdef CONFIG_COMPAT
 673         if (unlikely(tsk->compat_robust_list)) {
 674                 compat_exit_robust_list(tsk);
 675                 tsk->compat_robust_list = NULL;
 676         }
 677 #endif
 678         if (unlikely(!list_empty(&tsk->pi_state_list)))
 679                 exit_pi_state_list(tsk);
 680 #endif
 681
 682         /* Get rid of any cached register state */
 683         deactivate_mm(tsk, mm);
 684
 685         /* notify parent sleeping on vfork() */
 686         if (vfork_done) {
 687                 tsk->vfork_done = NULL;
 688                 complete(vfork_done);
 689         }
 690
 691         /*
 692          * If we're exiting normally, clear a user-space tid field if
 693          * requested.  We leave this alone when dying by signal, to leave
 694          * the value intact in a core dump, and to save the unnecessary
 695          * trouble otherwise.  Userland only wants this done for a sys_exit.
 696          */
 697         if (tsk->clear_child_tid) {
 698                 if (!(tsk->flags & PF_SIGNALED) &&
 699                     atomic_read(&mm->mm_users) > 1) {
 700                         /*
 701                          * We don't check the error code - if userspace has
 702                          * not set up a proper pointer then tough luck.
 703                          */
 704                         put_user(0, tsk->clear_child_tid);
 705                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 706                                         1, NULL, NULL, 0);
 707                 }
 708                 tsk->clear_child_tid = NULL;
 709         }
 710 }
 711
 712 /*
 713  * Allocate a new mm structure and copy contents from the
 714  * mm structure of the passed in task structure.
 715  */
 716 struct mm_struct *dup_mm(struct task_struct *tsk)
 717 {
 718         struct mm_struct *mm, *oldmm = current->mm;
 719         int err;
 720
 721         if (!oldmm)
 722                 return NULL;
 723
 724         mm = allocate_mm();
 725         if (!mm)
 726                 goto fail_nomem;
 727
 728         memcpy(mm, oldmm, sizeof(*mm));
 729         mm_init_cpumask(mm);
 730
 731         /* Initializing for Swap token stuff */
 732         mm->token_priority = 0;
 733         mm->last_interval = 0;
 734
 735 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 736         mm->pmd_huge_pte = NULL;
 737 #endif
 738
 739         if (!mm_init(mm, tsk))
 740                 goto fail_nomem;
 741
 742         if (init_new_context(tsk, mm))
 743                 goto fail_nocontext;
 744
 745         dup_mm_exe_file(oldmm, mm);
 746
 747         err = dup_mmap(mm, oldmm);
 748         if (err)
 749                 goto free_pt;
 750
 751         mm->hiwater_rss = get_mm_rss(mm);
 752         mm->hiwater_vm = mm->total_vm;
 753
 754         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 755                 goto free_pt;
 756
 757         return mm;
 758
 759 free_pt:
 760         /* don't put binfmt in mmput, we haven't got module yet */
 761         mm->binfmt = NULL;
 762         mmput(mm);
 763
 764 fail_nomem:
 765         return NULL;
 766
 767 fail_nocontext:
 768         /*
 769          * If init_new_context() failed, we cannot use mmput() to free the mm
 770          * because it calls destroy_context()
 771          */
 772         mm_free_pgd(mm);
 773         free_mm(mm);
 774         return NULL;
 775 }
 776
 777 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 778 {
 779         struct mm_struct *mm, *oldmm;
 780         int retval;
 781
 782         tsk->min_flt = tsk->maj_flt = 0;
 783         tsk->nvcsw = tsk->nivcsw = 0;
 784 #ifdef CONFIG_DETECT_HUNG_TASK
 785         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 786 #endif
 787
 788         tsk->mm = NULL;
 789         tsk->active_mm = NULL;
 790
 791         /*
 792          * Are we cloning a kernel thread?
 793          *
 794          * We need to steal a active VM for that..
 795          */
 796         oldmm = current->mm;
 797         if (!oldmm)
 798                 return 0;
 799
 800         if (clone_flags & CLONE_VM) {
 801                 atomic_inc(&oldmm->mm_users);
 802                 mm = oldmm;
 803                 goto good_mm;
 804         }
 805
 806         retval = -ENOMEM;
 807         mm = dup_mm(tsk);
 808         if (!mm)
 809                 goto fail_nomem;
 810
 811 good_mm:
 812         /* Initializing for Swap token stuff */
 813         mm->token_priority = 0;
 814         mm->last_interval = 0;
 815
 816         tsk->mm = mm;
 817         tsk->active_mm = mm;
 818         return 0;
 819
 820 fail_nomem:
 821         return retval;
 822 }
 823
 824 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 825 {
 826         struct fs_struct *fs = current->fs;
 827         if (clone_flags & CLONE_FS) {
 828                 /* tsk->fs is already what we want */
 829                 spin_lock(&fs->lock);
 830                 if (fs->in_exec) {
 831                         spin_unlock(&fs->lock);
 832                         return -EAGAIN;
 833                 }
 834                 fs->users++;
 835                 spin_unlock(&fs->lock);
 836                 return 0;
 837         }
 838         tsk->fs = copy_fs_struct(fs);
 839         if (!tsk->fs)
 840                 return -ENOMEM;
 841         return 0;
 842 }
 843
 844 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
 845 {
 846         struct files_struct *oldf, *newf;
 847         int error = 0;
 848
 849         /*
 850          * A background process may not have any files ...
 851          */
 852         oldf = current->files;
 853         if (!oldf)
 854                 goto out;
 855
 856         if (clone_flags & CLONE_FILES) {
 857                 atomic_inc(&oldf->count);
 858                 goto out;
 859         }
 860
 861         newf = dup_fd(oldf, &error);
 862         if (!newf)
 863                 goto out;
 864
 865         tsk->files = newf;
 866         error = 0;
 867 out:
 868         return error;
 869 }
 870
 871 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 872 {
 873 #ifdef CONFIG_BLOCK
 874         struct io_context *ioc = current->io_context;
 875
 876         if (!ioc)
 877                 return 0;
 878         /*
 879          * Share io context with parent, if CLONE_IO is set
 880          */
 881         if (clone_flags & CLONE_IO) {
 882                 tsk->io_context = ioc_task_link(ioc);
 883                 if (unlikely(!tsk->io_context))
 884                         return -ENOMEM;
 885         } else if (ioprio_valid(ioc->ioprio)) {
 886                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
 887                 if (unlikely(!tsk->io_context))
 888                         return -ENOMEM;
 889
 890                 tsk->io_context->ioprio = ioc->ioprio;
 891         }
 892 #endif
 893         return 0;
 894 }
 895
 896 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 897 {
 898         struct sighand_struct *sig;
 899
 900         if (clone_flags & CLONE_SIGHAND) {
 901                 atomic_inc(&current->sighand->count);
 902                 return 0;
 903         }
 904         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 905         rcu_assign_pointer(tsk->sighand, sig);
 906         if (!sig)
 907                 return -ENOMEM;
 908         atomic_set(&sig->count, 1);
 909         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 910         return 0;
 911 }
 912
 913 void __cleanup_sighand(struct sighand_struct *sighand)
 914 {
 915         if (atomic_dec_and_test(&sighand->count)) {
 916                 signalfd_cleanup(sighand);
 917                 kmem_cache_free(sighand_cachep, sighand);
 918         }
 919 }
 920
 921
 922 /*
 923  * Initialize POSIX timer handling for a thread group.
 924  */
 925 static void posix_cpu_timers_init_group(struct signal_struct *sig)
 926 {
 927         unsigned long cpu_limit;
 928
 929         /* Thread group counters. */
 930         thread_group_cputime_init(sig);
 931
 932         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
 933         if (cpu_limit != RLIM_INFINITY) {
 934                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
 935                 sig->cputimer.running = 1;
 936         }
 937
 938         /* The timer lists. */
 939         INIT_LIST_HEAD(&sig->cpu_timers[0]);
 940         INIT_LIST_HEAD(&sig->cpu_timers[1]);
 941         INIT_LIST_HEAD(&sig->cpu_timers[2]);
 942 }
 943
 944 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
 945 {
 946         struct signal_struct *sig;
 947
 948         if (clone_flags & CLONE_THREAD)
 949                 return 0;
 950
 951         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
 952         tsk->signal = sig;
 953         if (!sig)
 954                 return -ENOMEM;
 955
 956         sig->nr_threads = 1;
 957         atomic_set(&sig->live, 1);
 958         atomic_set(&sig->sigcnt, 1);
 959         init_waitqueue_head(&sig->wait_chldexit);
 960         if (clone_flags & CLONE_NEWPID)
 961                 sig->flags |= SIGNAL_UNKILLABLE;
 962         sig->curr_target = tsk;
 963         init_sigpending(&sig->shared_pending);
 964         INIT_LIST_HEAD(&sig->posix_timers);
 965
 966         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 967         sig->real_timer.function = it_real_fn;
 968
 969         task_lock(current->group_leader);
 970         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
 971         task_unlock(current->group_leader);
 972
 973         posix_cpu_timers_init_group(sig);
 974
 975         tty_audit_fork(sig);
 976         sched_autogroup_fork(sig);
 977
 978 #ifdef CONFIG_CGROUPS
 979         init_rwsem(&sig->threadgroup_fork_lock);
 980 #endif
 981
 982         sig->oom_adj = current->signal->oom_adj;
 983         sig->oom_score_adj = current->signal->oom_score_adj;
 984         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
 985
 986         mutex_init(&sig->cred_guard_mutex);
 987
 988         return 0;
 989 }
 990
 991 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
 992 {
 993         unsigned long new_flags = p->flags;
 994
 995         new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
 996         new_flags |= PF_FORKNOEXEC;
 997         new_flags |= PF_STARTING;
 998         p->flags = new_flags;
 999         clear_freeze_flag(p);
1000 }
1001
1002 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1003 {
1004         current->clear_child_tid = tidptr;
1005
1006         return task_pid_vnr(current);
1007 }
1008
1009 static void rt_mutex_init_task(struct task_struct *p)
1010 {
1011         raw_spin_lock_init(&p->pi_lock);
1012 #ifdef CONFIG_RT_MUTEXES
1013         plist_head_init(&p->pi_waiters);
1014         p->pi_blocked_on = NULL;
1015 #endif
1016 }
1017
1018 #ifdef CONFIG_MM_OWNER
1019 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1020 {
1021         mm->owner = p;
1022 }
1023 #endif /* CONFIG_MM_OWNER */
1024
1025 /*
1026  * Initialize POSIX timer handling for a single task.
1027  */
1028 static void posix_cpu_timers_init(struct task_struct *tsk)
1029 {
1030         tsk->cputime_expires.prof_exp = cputime_zero;
1031         tsk->cputime_expires.virt_exp = cputime_zero;
1032         tsk->cputime_expires.sched_exp = 0;
1033         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1034         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1035         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1036 }
1037
1038 /*
1039  * This creates a new process as a copy of the old one,
1040  * but does not actually start it yet.
1041  *
1042  * It copies the registers, and all the appropriate
1043  * parts of the process environment (as per the clone
1044  * flags). The actual kick-off is left to the caller.
1045  */
1046 static struct task_struct *copy_process(unsigned long clone_flags,
1047                                         unsigned long stack_start,
1048                                         struct pt_regs *regs,
1049                                         unsigned long stack_size,
1050                                         int __user *child_tidptr,
1051                                         struct pid *pid,
1052                                         int trace)
1053 {
1054         int retval;
1055         struct task_struct *p;
1056         int cgroup_callbacks_done = 0;
1057
1058         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1059                 return ERR_PTR(-EINVAL);
1060
1061         /*
1062          * Thread groups must share signals as well, and detached threads
1063          * can only be started up within the thread group.
1064          */
1065         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1066                 return ERR_PTR(-EINVAL);
1067
1068         /*
1069          * Shared signal handlers imply shared VM. By way of the above,
1070          * thread groups also imply shared VM. Blocking this case allows
1071          * for various simplifications in other code.
1072          */
1073         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1074                 return ERR_PTR(-EINVAL);
1075
1076         /*
1077          * Siblings of global init remain as zombies on exit since they are
1078          * not reaped by their parent (swapper). To solve this and to avoid
1079          * multi-rooted process trees, prevent global and container-inits
1080          * from creating siblings.
1081          */
1082         if ((clone_flags & CLONE_PARENT) &&
1083                                 current->signal->flags & SIGNAL_UNKILLABLE)
1084                 return ERR_PTR(-EINVAL);
1085
1086         retval = security_task_create(clone_flags);
1087         if (retval)
1088                 goto fork_out;
1089
1090         retval = -ENOMEM;
1091         p = dup_task_struct(current);
1092         if (!p)
1093                 goto fork_out;
1094
1095         ftrace_graph_init_task(p);
1096
1097         rt_mutex_init_task(p);
1098
1099 #ifdef CONFIG_PROVE_LOCKING
1100         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1101         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1102 #endif
1103         retval = -EAGAIN;
1104         if (atomic_read(&p->real_cred->user->processes) >=
1105                         task_rlimit(p, RLIMIT_NPROC)) {
1106                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1107                     p->real_cred->user != INIT_USER)
1108                         goto bad_fork_free;
1109         }
1110         current->flags &= ~PF_NPROC_EXCEEDED;
1111
1112         retval = copy_creds(p, clone_flags);
1113         if (retval < 0)
1114                 goto bad_fork_free;
1115
1116         /*
1117          * If multiple threads are within copy_process(), then this check
1118          * triggers too late. This doesn't hurt, the check is only there
1119          * to stop root fork bombs.
1120          */
1121         retval = -EAGAIN;
1122         if (nr_threads >= max_threads)
1123                 goto bad_fork_cleanup_count;
1124
1125         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1126                 goto bad_fork_cleanup_count;
1127
1128         p->did_exec = 0;
1129         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1130         copy_flags(clone_flags, p);
1131         INIT_LIST_HEAD(&p->children);
1132         INIT_LIST_HEAD(&p->sibling);
1133         rcu_copy_process(p);
1134         p->vfork_done = NULL;
1135         spin_lock_init(&p->alloc_lock);
1136
1137         init_sigpending(&p->pending);
1138
1139         p->utime = cputime_zero;
1140         p->stime = cputime_zero;
1141         p->gtime = cputime_zero;
1142         p->utimescaled = cputime_zero;
1143         p->stimescaled = cputime_zero;
1144 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1145         p->prev_utime = cputime_zero;
1146         p->prev_stime = cputime_zero;
1147 #endif
1148 #if defined(SPLIT_RSS_COUNTING)
1149         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1150 #endif
1151
1152         p->default_timer_slack_ns = current->timer_slack_ns;
1153
1154         task_io_accounting_init(&p->ioac);
1155         acct_clear_integrals(p);
1156
1157         posix_cpu_timers_init(p);
1158
1159         do_posix_clock_monotonic_gettime(&p->start_time);
1160         p->real_start_time = p->start_time;
1161         monotonic_to_bootbased(&p->real_start_time);
1162         p->io_context = NULL;
1163         p->audit_context = NULL;
1164         if (clone_flags & CLONE_THREAD)
1165                 threadgroup_fork_read_lock(current);
1166         cgroup_fork(p);
1167 #ifdef CONFIG_NUMA
1168         p->mempolicy = mpol_dup(p->mempolicy);
1169         if (IS_ERR(p->mempolicy)) {
1170                 retval = PTR_ERR(p->mempolicy);
1171                 p->mempolicy = NULL;
1172                 goto bad_fork_cleanup_cgroup;
1173         }
1174         mpol_fix_fork_child_flag(p);
1175 #endif
1176 #ifdef CONFIG_CPUSETS
1177         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1178         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1179 #endif
1180 #ifdef CONFIG_TRACE_IRQFLAGS
1181         p->irq_events = 0;
1182 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1183         p->hardirqs_enabled = 1;
1184 #else
1185         p->hardirqs_enabled = 0;
1186 #endif
1187         p->hardirq_enable_ip = 0;
1188         p->hardirq_enable_event = 0;
1189         p->hardirq_disable_ip = _THIS_IP_;
1190         p->hardirq_disable_event = 0;
1191         p->softirqs_enabled = 1;
1192         p->softirq_enable_ip = _THIS_IP_;
1193         p->softirq_enable_event = 0;
1194         p->softirq_disable_ip = 0;
1195         p->softirq_disable_event = 0;
1196         p->hardirq_context = 0;
1197         p->softirq_context = 0;
1198 #endif
1199 #ifdef CONFIG_LOCKDEP
1200         p->lockdep_depth = 0; /* no locks held yet */
1201         p->curr_chain_key = 0;
1202         p->lockdep_recursion = 0;
1203 #endif
1204
1205 #ifdef CONFIG_DEBUG_MUTEXES
1206         p->blocked_on = NULL; /* not blocked yet */
1207 #endif
1208 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1209         p->memcg_batch.do_batch = 0;
1210         p->memcg_batch.memcg = NULL;
1211 #endif
1212
1213         /* Perform scheduler related setup. Assign this task to a CPU. */
1214         sched_fork(p);
1215
1216         retval = perf_event_init_task(p);
1217         if (retval)
1218                 goto bad_fork_cleanup_policy;
1219         retval = audit_alloc(p);
1220         if (retval)
1221                 goto bad_fork_cleanup_policy;
1222         /* copy all the process information */
1223         retval = copy_semundo(clone_flags, p);
1224         if (retval)
1225                 goto bad_fork_cleanup_audit;
1226         retval = copy_files(clone_flags, p);
1227         if (retval)
1228                 goto bad_fork_cleanup_semundo;
1229         retval = copy_fs(clone_flags, p);
1230         if (retval)
1231                 goto bad_fork_cleanup_files;
1232         retval = copy_sighand(clone_flags, p);
1233         if (retval)
1234                 goto bad_fork_cleanup_fs;
1235         retval = copy_signal(clone_flags, p);
1236         if (retval)
1237                 goto bad_fork_cleanup_sighand;
1238         retval = copy_mm(clone_flags, p);
1239         if (retval)
1240                 goto bad_fork_cleanup_signal;
1241         retval = copy_namespaces(clone_flags, p);
1242         if (retval)
1243                 goto bad_fork_cleanup_mm;
1244         retval = copy_io(clone_flags, p);
1245         if (retval)
1246                 goto bad_fork_cleanup_namespaces;
1247         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1248         if (retval)
1249                 goto bad_fork_cleanup_io;
1250
1251         if (pid != &init_struct_pid) {
1252                 retval = -ENOMEM;
1253                 pid = alloc_pid(p->nsproxy->pid_ns);
1254                 if (!pid)
1255                         goto bad_fork_cleanup_io;
1256         }
1257
1258         p->pid = pid_nr(pid);
1259         p->tgid = p->pid;
1260         if (clone_flags & CLONE_THREAD)
1261                 p->tgid = current->tgid;
1262
1263         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1264         /*
1265          * Clear TID on mm_release()?
1266          */
1267         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1268 #ifdef CONFIG_BLOCK
1269         p->plug = NULL;
1270 #endif
1271 #ifdef CONFIG_FUTEX
1272         p->robust_list = NULL;
1273 #ifdef CONFIG_COMPAT
1274         p->compat_robust_list = NULL;
1275 #endif
1276         INIT_LIST_HEAD(&p->pi_state_list);
1277         p->pi_state_cache = NULL;
1278 #endif
1279         /*
1280          * sigaltstack should be cleared when sharing the same VM
1281          */
1282         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1283                 p->sas_ss_sp = p->sas_ss_size = 0;
1284
1285         /*
1286          * Syscall tracing and stepping should be turned off in the
1287          * child regardless of CLONE_PTRACE.
1288          */
1289         user_disable_single_step(p);
1290         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1291 #ifdef TIF_SYSCALL_EMU
1292         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1293 #endif
1294         clear_all_latency_tracing(p);
1295
1296         /* ok, now we should be set up.. */
1297         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1298         p->pdeath_signal = 0;
1299         p->exit_state = 0;
1300
1301         p->nr_dirtied = 0;
1302         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1303
1304         /*
1305          * Ok, make it visible to the rest of the system.
1306          * We dont wake it up yet.
1307          */
1308         p->group_leader = p;
1309         INIT_LIST_HEAD(&p->thread_group);
1310
1311         /* Now that the task is set up, run cgroup callbacks if
1312          * necessary. We need to run them before the task is visible
1313          * on the tasklist. */
1314         cgroup_fork_callbacks(p);
1315         cgroup_callbacks_done = 1;
1316
1317         /* Need tasklist lock for parent etc handling! */
1318         write_lock_irq(&tasklist_lock);
1319
1320         /* CLONE_PARENT re-uses the old parent */
1321         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1322                 p->real_parent = current->real_parent;
1323                 p->parent_exec_id = current->parent_exec_id;
1324         } else {
1325                 p->real_parent = current;
1326                 p->parent_exec_id = current->self_exec_id;
1327         }
1328
1329         spin_lock(&current->sighand->siglock);
1330
1331         /*
1332          * Process group and session signals need to be delivered to just the
1333          * parent before the fork or both the parent and the child after the
1334          * fork. Restart if a signal comes in before we add the new process to
1335          * it's process group.
1336          * A fatal signal pending means that current will exit, so the new
1337          * thread can't slip out of an OOM kill (or normal SIGKILL).
1338         */
1339         recalc_sigpending();
1340         if (signal_pending(current)) {
1341                 spin_unlock(&current->sighand->siglock);
1342                 write_unlock_irq(&tasklist_lock);
1343                 retval = -ERESTARTNOINTR;
1344                 goto bad_fork_free_pid;
1345         }
1346
1347         if (clone_flags & CLONE_THREAD) {
1348                 current->signal->nr_threads++;
1349                 atomic_inc(&current->signal->live);
1350                 atomic_inc(&current->signal->sigcnt);
1351                 p->group_leader = current->group_leader;
1352                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1353         }
1354
1355         if (likely(p->pid)) {
1356                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1357
1358                 if (thread_group_leader(p)) {
1359                         if (is_child_reaper(pid))
1360                                 p->nsproxy->pid_ns->child_reaper = p;
1361
1362                         p->signal->leader_pid = pid;
1363                         p->signal->tty = tty_kref_get(current->signal->tty);
1364                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1365                         attach_pid(p, PIDTYPE_SID, task_session(current));
1366                         list_add_tail(&p->sibling, &p->real_parent->children);
1367                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1368                         __this_cpu_inc(process_counts);
1369                 }
1370                 attach_pid(p, PIDTYPE_PID, pid);
1371                 nr_threads++;
1372         }
1373
1374         total_forks++;
1375         spin_unlock(&current->sighand->siglock);
1376         write_unlock_irq(&tasklist_lock);
1377         proc_fork_connector(p);
1378         cgroup_post_fork(p);
1379         if (clone_flags & CLONE_THREAD)
1380                 threadgroup_fork_read_unlock(current);
1381         perf_event_fork(p);
1382         return p;
1383
1384 bad_fork_free_pid:
1385         if (pid != &init_struct_pid)
1386                 free_pid(pid);
1387 bad_fork_cleanup_io:
1388         if (p->io_context)
1389                 exit_io_context(p);
1390 bad_fork_cleanup_namespaces:
1391         if (unlikely(clone_flags & CLONE_NEWPID))
1392                 pid_ns_release_proc(p->nsproxy->pid_ns);
1393         exit_task_namespaces(p);
1394 bad_fork_cleanup_mm:
1395         if (p->mm)
1396                 mmput(p->mm);
1397 bad_fork_cleanup_signal:
1398         if (!(clone_flags & CLONE_THREAD))
1399                 free_signal_struct(p->signal);
1400 bad_fork_cleanup_sighand:
1401         __cleanup_sighand(p->sighand);
1402 bad_fork_cleanup_fs:
1403         exit_fs(p); /* blocking */
1404 bad_fork_cleanup_files:
1405         exit_files(p); /* blocking */
1406 bad_fork_cleanup_semundo:
1407         exit_sem(p);
1408 bad_fork_cleanup_audit:
1409         audit_free(p);
1410 bad_fork_cleanup_policy:
1411         perf_event_free_task(p);
1412 #ifdef CONFIG_NUMA
1413         mpol_put(p->mempolicy);
1414 bad_fork_cleanup_cgroup:
1415 #endif
1416         if (clone_flags & CLONE_THREAD)
1417                 threadgroup_fork_read_unlock(current);
1418         cgroup_exit(p, cgroup_callbacks_done);
1419         delayacct_tsk_free(p);
1420         module_put(task_thread_info(p)->exec_domain->module);
1421 bad_fork_cleanup_count:
1422         atomic_dec(&p->cred->user->processes);
1423         exit_creds(p);
1424 bad_fork_free:
1425         free_task(p);
1426 fork_out:
1427         return ERR_PTR(retval);
1428 }
1429
1430 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1431 {
1432         memset(regs, 0, sizeof(struct pt_regs));
1433         return regs;
1434 }
1435
1436 static inline void init_idle_pids(struct pid_link *links)
1437 {
1438         enum pid_type type;
1439
1440         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1441                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1442                 links[type].pid = &init_struct_pid;
1443         }
1444 }
1445
1446 struct task_struct * __cpuinit fork_idle(int cpu)
1447 {
1448         struct task_struct *task;
1449         struct pt_regs regs;
1450
1451         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1452                             &init_struct_pid, 0);
1453         if (!IS_ERR(task)) {
1454                 init_idle_pids(task->pids);
1455                 init_idle(task, cpu);
1456         }
1457
1458         return task;
1459 }
1460
1461 /*
1462  *  Ok, this is the main fork-routine.
1463  *
1464  * It copies the process, and if successful kick-starts
1465  * it and waits for it to finish using the VM if required.
1466  */
1467 long do_fork(unsigned long clone_flags,
1468               unsigned long stack_start,
1469               struct pt_regs *regs,
1470               unsigned long stack_size,
1471               int __user *parent_tidptr,
1472               int __user *child_tidptr)
1473 {
1474         struct task_struct *p;
1475         int trace = 0;
1476         long nr;
1477
1478         /*
1479          * Do some preliminary argument and permissions checking before we
1480          * actually start allocating stuff
1481          */
1482         if (clone_flags & CLONE_NEWUSER) {
1483                 if (clone_flags & CLONE_THREAD)
1484                         return -EINVAL;
1485                 /* hopefully this check will go away when userns support is
1486                  * complete
1487                  */
1488                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1489                                 !capable(CAP_SETGID))
1490                         return -EPERM;
1491         }
1492
1493         /*
1494          * Determine whether and which event to report to ptracer.  When
1495          * called from kernel_thread or CLONE_UNTRACED is explicitly
1496          * requested, no event is reported; otherwise, report if the event
1497          * for the type of forking is enabled.
1498          */
1499         if (likely(user_mode(regs)) && !(clone_flags & CLONE_UNTRACED)) {
1500                 if (clone_flags & CLONE_VFORK)
1501                         trace = PTRACE_EVENT_VFORK;
1502                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1503                         trace = PTRACE_EVENT_CLONE;
1504                 else
1505                         trace = PTRACE_EVENT_FORK;
1506
1507                 if (likely(!ptrace_event_enabled(current, trace)))
1508                         trace = 0;
1509         }
1510
1511         p = copy_process(clone_flags, stack_start, regs, stack_size,
1512                          child_tidptr, NULL, trace);
1513         /*
1514          * Do this prior waking up the new thread - the thread pointer
1515          * might get invalid after that point, if the thread exits quickly.
1516          */
1517         if (!IS_ERR(p)) {
1518                 struct completion vfork;
1519
1520                 trace_sched_process_fork(current, p);
1521
1522                 nr = task_pid_vnr(p);
1523
1524                 if (clone_flags & CLONE_PARENT_SETTID)
1525                         put_user(nr, parent_tidptr);
1526
1527                 if (clone_flags & CLONE_VFORK) {
1528                         p->vfork_done = &vfork;
1529                         init_completion(&vfork);
1530                 }
1531
1532                 audit_finish_fork(p);
1533
1534                 /*
1535                  * We set PF_STARTING at creation in case tracing wants to
1536                  * use this to distinguish a fully live task from one that
1537                  * hasn't finished SIGSTOP raising yet.  Now we clear it
1538                  * and set the child going.
1539                  */
1540                 p->flags &= ~PF_STARTING;
1541
1542                 wake_up_new_task(p);
1543
1544                 /* forking complete and child started to run, tell ptracer */
1545                 if (unlikely(trace))
1546                         ptrace_event(trace, nr);
1547
1548                 if (clone_flags & CLONE_VFORK) {
1549                         freezer_do_not_count();
1550                         wait_for_completion(&vfork);
1551                         freezer_count();
1552                         ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1553                 }
1554         } else {
1555                 nr = PTR_ERR(p);
1556         }
1557         return nr;
1558 }
1559
1560 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1561 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1562 #endif
1563
1564 static void sighand_ctor(void *data)
1565 {
1566         struct sighand_struct *sighand = data;
1567
1568         spin_lock_init(&sighand->siglock);
1569         init_waitqueue_head(&sighand->signalfd_wqh);
1570 }
1571
1572 void __init proc_caches_init(void)
1573 {
1574         sighand_cachep = kmem_cache_create("sighand_cache",
1575                         sizeof(struct sighand_struct), 0,
1576                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1577                         SLAB_NOTRACK, sighand_ctor);
1578         signal_cachep = kmem_cache_create("signal_cache",
1579                         sizeof(struct signal_struct), 0,
1580                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1581         files_cachep = kmem_cache_create("files_cache",
1582                         sizeof(struct files_struct), 0,
1583                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1584         fs_cachep = kmem_cache_create("fs_cache",
1585                         sizeof(struct fs_struct), 0,
1586                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1587         /*
1588          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1589          * whole struct cpumask for the OFFSTACK case. We could change
1590          * this to *only* allocate as much of it as required by the
1591          * maximum number of CPU's we can ever have.  The cpumask_allocation
1592          * is at the end of the structure, exactly for that reason.
1593          */
1594         mm_cachep = kmem_cache_create("mm_struct",
1595                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1596                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1597         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1598         mmap_init();
1599         nsproxy_cache_init();
1600 }
1601
1602 /*
1603  * Check constraints on flags passed to the unshare system call.
1604  */
1605 static int check_unshare_flags(unsigned long unshare_flags)
1606 {
1607         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1608                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1609                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1610                 return -EINVAL;
1611         /*
1612          * Not implemented, but pretend it works if there is nothing to
1613          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1614          * needs to unshare vm.
1615          */
1616         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1617                 /* FIXME: get_task_mm() increments ->mm_users */
1618                 if (atomic_read(&current->mm->mm_users) > 1)
1619                         return -EINVAL;
1620         }
1621
1622         return 0;
1623 }
1624
1625 /*
1626  * Unshare the filesystem structure if it is being shared
1627  */
1628 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1629 {
1630         struct fs_struct *fs = current->fs;
1631
1632         if (!(unshare_flags & CLONE_FS) || !fs)
1633                 return 0;
1634
1635         /* don't need lock here; in the worst case we'll do useless copy */
1636         if (fs->users == 1)
1637                 return 0;
1638
1639         *new_fsp = copy_fs_struct(fs);
1640         if (!*new_fsp)
1641                 return -ENOMEM;
1642
1643         return 0;
1644 }
1645
1646 /*
1647  * Unshare file descriptor table if it is being shared
1648  */
1649 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1650 {
1651         struct files_struct *fd = current->files;
1652         int error = 0;
1653
1654         if ((unshare_flags & CLONE_FILES) &&
1655             (fd && atomic_read(&fd->count) > 1)) {
1656                 *new_fdp = dup_fd(fd, &error);
1657                 if (!*new_fdp)
1658                         return error;
1659         }
1660
1661         return 0;
1662 }
1663
1664 /*
1665  * unshare allows a process to 'unshare' part of the process
1666  * context which was originally shared using clone.  copy_*
1667  * functions used by do_fork() cannot be used here directly
1668  * because they modify an inactive task_struct that is being
1669  * constructed. Here we are modifying the current, active,
1670  * task_struct.
1671  */
1672 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1673 {
1674         struct fs_struct *fs, *new_fs = NULL;
1675         struct files_struct *fd, *new_fd = NULL;
1676         struct nsproxy *new_nsproxy = NULL;
1677         int do_sysvsem = 0;
1678         int err;
1679
1680         err = check_unshare_flags(unshare_flags);
1681         if (err)
1682                 goto bad_unshare_out;
1683
1684         /*
1685          * If unsharing namespace, must also unshare filesystem information.
1686          */
1687         if (unshare_flags & CLONE_NEWNS)
1688                 unshare_flags |= CLONE_FS;
1689         /*
1690          * CLONE_NEWIPC must also detach from the undolist: after switching
1691          * to a new ipc namespace, the semaphore arrays from the old
1692          * namespace are unreachable.
1693          */
1694         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1695                 do_sysvsem = 1;
1696         err = unshare_fs(unshare_flags, &new_fs);
1697         if (err)
1698                 goto bad_unshare_out;
1699         err = unshare_fd(unshare_flags, &new_fd);
1700         if (err)
1701                 goto bad_unshare_cleanup_fs;
1702         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy, new_fs);
1703         if (err)
1704                 goto bad_unshare_cleanup_fd;
1705
1706         if (new_fs || new_fd || do_sysvsem || new_nsproxy) {
1707                 if (do_sysvsem) {
1708                         /*
1709                          * CLONE_SYSVSEM is equivalent to sys_exit().
1710                          */
1711                         exit_sem(current);
1712                 }
1713
1714                 if (new_nsproxy) {
1715                         switch_task_namespaces(current, new_nsproxy);
1716                         new_nsproxy = NULL;
1717                 }
1718
1719                 task_lock(current);
1720
1721                 if (new_fs) {
1722                         fs = current->fs;
1723                         spin_lock(&fs->lock);
1724                         current->fs = new_fs;
1725                         if (--fs->users)
1726                                 new_fs = NULL;
1727                         else
1728                                 new_fs = fs;
1729                         spin_unlock(&fs->lock);
1730                 }
1731
1732                 if (new_fd) {
1733                         fd = current->files;
1734                         current->files = new_fd;
1735                         new_fd = fd;
1736                 }
1737
1738                 task_unlock(current);
1739         }
1740
1741         if (new_nsproxy)
1742                 put_nsproxy(new_nsproxy);
1743
1744 bad_unshare_cleanup_fd:
1745         if (new_fd)
1746                 put_files_struct(new_fd);
1747
1748 bad_unshare_cleanup_fs:
1749         if (new_fs)
1750                 free_fs_struct(new_fs);
1751
1752 bad_unshare_out:
1753         return err;
1754 }
1755
1756 /*
1757  *      Helper to unshare the files of the current task.
1758  *      We don't want to expose copy_files internals to
1759  *      the exec layer of the kernel.
1760  */
1761
1762 int unshare_files(struct files_struct **displaced)
1763 {
1764         struct task_struct *task = current;
1765         struct files_struct *copy = NULL;
1766         int error;
1767
1768         error = unshare_fd(CLONE_FILES, &copy);
1769         if (error || !copy) {
1770                 *displaced = NULL;
1771                 return error;
1772         }
1773         *displaced = task->files;
1774         task_lock(task);
1775         task->files = copy;
1776         task_unlock(task);
1777         return 0;
1778 }