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