kernel/fork.c

   1 /*
   2  *  linux/kernel/fork.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  *  'fork.c' contains the help-routines for the 'fork' system call
   9  * (see also entry.S and others).
  10  * Fork is rather simple, once you get the hang of it, but the memory
  11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  12  */
  13
  14 #include <linux/slab.h>
  15 #include <linux/init.h>
  16 #include <linux/unistd.h>
  17 #include <linux/module.h>
  18 #include <linux/vmalloc.h>
  19 #include <linux/completion.h>
  20 #include <linux/personality.h>
  21 #include <linux/mempolicy.h>
  22 #include <linux/sem.h>
  23 #include <linux/file.h>
  24 #include <linux/fdtable.h>
  25 #include <linux/iocontext.h>
  26 #include <linux/key.h>
  27 #include <linux/binfmts.h>
  28 #include <linux/mman.h>
  29 #include <linux/mmu_notifier.h>
  30 #include <linux/fs.h>
  31 #include <linux/nsproxy.h>
  32 #include <linux/capability.h>
  33 #include <linux/cpu.h>
  34 #include <linux/cgroup.h>
  35 #include <linux/security.h>
  36 #include <linux/hugetlb.h>
  37 #include <linux/swap.h>
  38 #include <linux/syscalls.h>
  39 #include <linux/jiffies.h>
  40 #include <linux/futex.h>
  41 #include <linux/compat.h>
  42 #include <linux/kthread.h>
  43 #include <linux/task_io_accounting_ops.h>
  44 #include <linux/rcupdate.h>
  45 #include <linux/ptrace.h>
  46 #include <linux/mount.h>
  47 #include <linux/audit.h>
  48 #include <linux/memcontrol.h>
  49 #include <linux/ftrace.h>
  50 #include <linux/profile.h>
  51 #include <linux/rmap.h>
  52 #include <linux/ksm.h>
  53 #include <linux/acct.h>
  54 #include <linux/tsacct_kern.h>
  55 #include <linux/cn_proc.h>
  56 #include <linux/freezer.h>
  57 #include <linux/delayacct.h>
  58 #include <linux/taskstats_kern.h>
  59 #include <linux/random.h>
  60 #include <linux/tty.h>
  61 #include <linux/blkdev.h>
  62 #include <linux/fs_struct.h>
  63 #include <linux/magic.h>
  64 #include <linux/perf_event.h>
  65 #include <linux/posix-timers.h>
  66 #include <linux/user-return-notifier.h>
  67 #include <linux/oom.h>
  68 #include <linux/khugepaged.h>
  69
  70 #include <asm/pgtable.h>
  71 #include <asm/pgalloc.h>
  72 #include <asm/uaccess.h>
  73 #include <asm/mmu_context.h>
  74 #include <asm/cacheflush.h>
  75 #include <asm/tlbflush.h>
  76
  77 #include <trace/events/sched.h>
  78
  79 /*
  80  * Protected counters by write_lock_irq(&tasklist_lock)
  81  */
  82 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  83 int nr_threads;                 /* The idle threads do not count.. */
  84
  85 int max_threads;                /* tunable limit on nr_threads */
  86
  87 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  88
  89 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  90
  91 #ifdef CONFIG_PROVE_RCU
  92 int lockdep_tasklist_lock_is_held(void)
  93 {
  94         return lockdep_is_held(&tasklist_lock);
  95 }
  96 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
  97 #endif /* #ifdef CONFIG_PROVE_RCU */
  98
  99 int nr_processes(void)
 100 {
 101         int cpu;
 102         int total = 0;
 103
 104         for_each_possible_cpu(cpu)
 105                 total += per_cpu(process_counts, cpu);
 106
 107         return total;
 108 }
 109
 110 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 111 # define alloc_task_struct_node(node)           \
 112                 kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node)
 113 # define free_task_struct(tsk)                  \
 114                 kmem_cache_free(task_struct_cachep, (tsk))
 115 static struct kmem_cache *task_struct_cachep;
 116 #endif
 117
 118 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
 119 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 120                                                   int node)
 121 {
 122 #ifdef CONFIG_DEBUG_STACK_USAGE
 123         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
 124 #else
 125         gfp_t mask = GFP_KERNEL;
 126 #endif
 127         struct page *page = alloc_pages_node(node, mask, THREAD_SIZE_ORDER);
 128
 129         return page ? page_address(page) : NULL;
 130 }
 131
 132 static inline void free_thread_info(struct thread_info *ti)
 133 {
 134         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 135 }
 136 #endif
 137
 138 /* SLAB cache for signal_struct structures (tsk->signal) */
 139 static struct kmem_cache *signal_cachep;
 140
 141 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 142 struct kmem_cache *sighand_cachep;
 143
 144 /* SLAB cache for files_struct structures (tsk->files) */
 145 struct kmem_cache *files_cachep;
 146
 147 /* SLAB cache for fs_struct structures (tsk->fs) */
 148 struct kmem_cache *fs_cachep;
 149
 150 /* SLAB cache for vm_area_struct structures */
 151 struct kmem_cache *vm_area_cachep;
 152
 153 /* SLAB cache for mm_struct structures (tsk->mm) */
 154 static struct kmem_cache *mm_cachep;
 155
 156 static void account_kernel_stack(struct thread_info *ti, int account)
 157 {
 158         struct zone *zone = page_zone(virt_to_page(ti));
 159
 160         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
 161 }
 162
 163 void free_task(struct task_struct *tsk)
 164 {
 165         prop_local_destroy_single(&tsk->dirties);
 166         account_kernel_stack(tsk->stack, -1);
 167         free_thread_info(tsk->stack);
 168         rt_mutex_debug_task_free(tsk);
 169         ftrace_graph_exit_task(tsk);
 170         free_task_struct(tsk);
 171 }
 172 EXPORT_SYMBOL(free_task);
 173
 174 static inline void free_signal_struct(struct signal_struct *sig)
 175 {
 176         taskstats_tgid_free(sig);
 177         sched_autogroup_exit(sig);
 178         kmem_cache_free(signal_cachep, sig);
 179 }
 180
 181 static inline void put_signal_struct(struct signal_struct *sig)
 182 {
 183         if (atomic_dec_and_test(&sig->sigcnt))
 184                 free_signal_struct(sig);
 185 }
 186
 187 void __put_task_struct(struct task_struct *tsk)
 188 {
 189         WARN_ON(!tsk->exit_state);
 190         WARN_ON(atomic_read(&tsk->usage));
 191         WARN_ON(tsk == current);
 192
 193         exit_creds(tsk);
 194         delayacct_tsk_free(tsk);
 195         put_signal_struct(tsk->signal);
 196
 197         if (!profile_handoff_task(tsk))
 198                 free_task(tsk);
 199 }
 200 EXPORT_SYMBOL_GPL(__put_task_struct);
 201
 202 /*
 203  * macro override instead of weak attribute alias, to workaround
 204  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
 205  */
 206 #ifndef arch_task_cache_init
 207 #define arch_task_cache_init()
 208 #endif
 209
 210 void __init fork_init(unsigned long mempages)
 211 {
 212 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 213 #ifndef ARCH_MIN_TASKALIGN
 214 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 215 #endif
 216         /* create a slab on which task_structs can be allocated */
 217         task_struct_cachep =
 218                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 219                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
 220 #endif
 221
 222         /* do the arch specific task caches init */
 223         arch_task_cache_init();
 224
 225         /*
 226          * The default maximum number of threads is set to a safe
 227          * value: the thread structures can take up at most half
 228          * of memory.
 229          */
 230         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 231
 232         /*
 233          * we need to allow at least 20 threads to boot a system
 234          */
 235         if (max_threads < 20)
 236                 max_threads = 20;
 237
 238         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 239         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 240         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 241                 init_task.signal->rlim[RLIMIT_NPROC];
 242 }
 243
 244 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
 245                                                struct task_struct *src)
 246 {
 247         *dst = *src;
 248         return 0;
 249 }
 250
 251 static struct task_struct *dup_task_struct(struct task_struct *orig)
 252 {
 253         struct task_struct *tsk;
 254         struct thread_info *ti;
 255         unsigned long *stackend;
 256         int node = tsk_fork_get_node(orig);
 257         int err;
 258
 259         prepare_to_copy(orig);
 260
 261         tsk = alloc_task_struct_node(node);
 262         if (!tsk)
 263                 return NULL;
 264
 265         ti = alloc_thread_info_node(tsk, node);
 266         if (!ti) {
 267                 free_task_struct(tsk);
 268                 return NULL;
 269         }
 270
 271         err = arch_dup_task_struct(tsk, orig);
 272         if (err)
 273                 goto out;
 274
 275         tsk->stack = ti;
 276
 277         err = prop_local_init_single(&tsk->dirties);
 278         if (err)
 279                 goto out;
 280
 281         setup_thread_stack(tsk, orig);
 282         clear_user_return_notifier(tsk);
 283         clear_tsk_need_resched(tsk);
 284         stackend = end_of_stack(tsk);
 285         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 286
 287 #ifdef CONFIG_CC_STACKPROTECTOR
 288         tsk->stack_canary = get_random_int();
 289 #endif
 290
 291         /*
 292          * One for us, one for whoever does the "release_task()" (usually
 293          * parent)
 294          */
 295         atomic_set(&tsk->usage, 2);
 296 #ifdef CONFIG_BLK_DEV_IO_TRACE
 297         tsk->btrace_seq = 0;
 298 #endif
 299         tsk->splice_pipe = NULL;
 300
 301         account_kernel_stack(ti, 1);
 302
 303         return tsk;
 304
 305 out:
 306         free_thread_info(ti);
 307         free_task_struct(tsk);
 308         return NULL;
 309 }
 310
 311 #ifdef CONFIG_MMU
 312 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 313 {
 314         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 315         struct rb_node **rb_link, *rb_parent;
 316         int retval;
 317         unsigned long charge;
 318         struct mempolicy *pol;
 319
 320         down_write(&oldmm->mmap_sem);
 321         flush_cache_dup_mm(oldmm);
 322         /*
 323          * Not linked in yet - no deadlock potential:
 324          */
 325         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 326
 327         mm->locked_vm = 0;
 328         mm->mmap = NULL;
 329         mm->mmap_cache = NULL;
 330         mm->free_area_cache = oldmm->mmap_base;
 331         mm->cached_hole_size = ~0UL;
 332         mm->map_count = 0;
 333         cpumask_clear(mm_cpumask(mm));
 334         mm->mm_rb = RB_ROOT;
 335         rb_link = &mm->mm_rb.rb_node;
 336         rb_parent = NULL;
 337         pprev = &mm->mmap;
 338         retval = ksm_fork(mm, oldmm);
 339         if (retval)
 340                 goto out;
 341         retval = khugepaged_fork(mm, oldmm);
 342         if (retval)
 343                 goto out;
 344
 345         prev = NULL;
 346         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 347                 struct file *file;
 348
 349                 if (mpnt->vm_flags & VM_DONTCOPY) {
 350                         long pages = vma_pages(mpnt);
 351                         mm->total_vm -= pages;
 352                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 353                                                                 -pages);
 354                         continue;
 355                 }
 356                 charge = 0;
 357                 if (mpnt->vm_flags & VM_ACCOUNT) {
 358                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 359                         if (security_vm_enough_memory(len))
 360                                 goto fail_nomem;
 361                         charge = len;
 362                 }
 363                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 364                 if (!tmp)
 365                         goto fail_nomem;
 366                 *tmp = *mpnt;
 367                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 368                 pol = mpol_dup(vma_policy(mpnt));
 369                 retval = PTR_ERR(pol);
 370                 if (IS_ERR(pol))
 371                         goto fail_nomem_policy;
 372                 vma_set_policy(tmp, pol);
 373                 tmp->vm_mm = mm;
 374                 if (anon_vma_fork(tmp, mpnt))
 375                         goto fail_nomem_anon_vma_fork;
 376                 tmp->vm_flags &= ~VM_LOCKED;
 377                 tmp->vm_next = tmp->vm_prev = NULL;
 378                 file = tmp->vm_file;
 379                 if (file) {
 380                         struct inode *inode = file->f_path.dentry->d_inode;
 381                         struct address_space *mapping = file->f_mapping;
 382
 383                         get_file(file);
 384                         if (tmp->vm_flags & VM_DENYWRITE)
 385                                 atomic_dec(&inode->i_writecount);
 386                         mutex_lock(&mapping->i_mmap_mutex);
 387                         if (tmp->vm_flags & VM_SHARED)
 388                                 mapping->i_mmap_writable++;
 389                         flush_dcache_mmap_lock(mapping);
 390                         /* insert tmp into the share list, just after mpnt */
 391                         vma_prio_tree_add(tmp, mpnt);
 392                         flush_dcache_mmap_unlock(mapping);
 393                         mutex_unlock(&mapping->i_mmap_mutex);
 394                 }
 395
 396                 /*
 397                  * Clear hugetlb-related page reserves for children. This only
 398                  * affects MAP_PRIVATE mappings. Faults generated by the child
 399                  * are not guaranteed to succeed, even if read-only
 400                  */
 401                 if (is_vm_hugetlb_page(tmp))
 402                         reset_vma_resv_huge_pages(tmp);
 403
 404                 /*
 405                  * Link in the new vma and copy the page table entries.
 406                  */
 407                 *pprev = tmp;
 408                 pprev = &tmp->vm_next;
 409                 tmp->vm_prev = prev;
 410                 prev = tmp;
 411
 412                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 413                 rb_link = &tmp->vm_rb.rb_right;
 414                 rb_parent = &tmp->vm_rb;
 415
 416                 mm->map_count++;
 417                 retval = copy_page_range(mm, oldmm, mpnt);
 418
 419                 if (tmp->vm_ops && tmp->vm_ops->open)
 420                         tmp->vm_ops->open(tmp);
 421
 422                 if (retval)
 423                         goto out;
 424         }
 425         /* a new mm has just been created */
 426         arch_dup_mmap(oldmm, mm);
 427         retval = 0;
 428 out:
 429         up_write(&mm->mmap_sem);
 430         flush_tlb_mm(oldmm);
 431         up_write(&oldmm->mmap_sem);
 432         return retval;
 433 fail_nomem_anon_vma_fork:
 434         mpol_put(pol);
 435 fail_nomem_policy:
 436         kmem_cache_free(vm_area_cachep, tmp);
 437 fail_nomem:
 438         retval = -ENOMEM;
 439         vm_unacct_memory(charge);
 440         goto out;
 441 }
 442
 443 static inline int mm_alloc_pgd(struct mm_struct *mm)
 444 {
 445         mm->pgd = pgd_alloc(mm);
 446         if (unlikely(!mm->pgd))
 447                 return -ENOMEM;
 448         return 0;
 449 }
 450
 451 static inline void mm_free_pgd(struct mm_struct *mm)
 452 {
 453         pgd_free(mm, mm->pgd);
 454 }
 455 #else
 456 #define dup_mmap(mm, oldmm)     (0)
 457 #define mm_alloc_pgd(mm)        (0)
 458 #define mm_free_pgd(mm)
 459 #endif /* CONFIG_MMU */
 460
 461 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 462
 463 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 464 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 465
 466 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 467
 468 static int __init coredump_filter_setup(char *s)
 469 {
 470         default_dump_filter =
 471                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 472                 MMF_DUMP_FILTER_MASK;
 473         return 1;
 474 }
 475
 476 __setup("coredump_filter=", coredump_filter_setup);
 477
 478 #include <linux/init_task.h>
 479
 480 static void mm_init_aio(struct mm_struct *mm)
 481 {
 482 #ifdef CONFIG_AIO
 483         spin_lock_init(&mm->ioctx_lock);
 484         INIT_HLIST_HEAD(&mm->ioctx_list);
 485 #endif
 486 }
 487
 488 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
 489 {
 490         atomic_set(&mm->mm_users, 1);
 491         atomic_set(&mm->mm_count, 1);
 492         init_rwsem(&mm->mmap_sem);
 493         INIT_LIST_HEAD(&mm->mmlist);
 494         mm->flags = (current->mm) ?
 495                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
 496         mm->core_state = NULL;
 497         mm->nr_ptes = 0;
 498         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 499         spin_lock_init(&mm->page_table_lock);
 500         mm->free_area_cache = TASK_UNMAPPED_BASE;
 501         mm->cached_hole_size = ~0UL;
 502         mm_init_aio(mm);
 503         mm_init_owner(mm, p);
 504
 505         if (likely(!mm_alloc_pgd(mm))) {
 506                 mm->def_flags = 0;
 507                 mmu_notifier_mm_init(mm);
 508                 return mm;
 509         }
 510
 511         free_mm(mm);
 512         return NULL;
 513 }
 514
 515 /*
 516  * Allocate and initialize an mm_struct.
 517  */
 518 struct mm_struct *mm_alloc(void)
 519 {
 520         struct mm_struct *mm;
 521
 522         mm = allocate_mm();
 523         if (!mm)
 524                 return NULL;
 525
 526         memset(mm, 0, sizeof(*mm));
 527         mm_init_cpumask(mm);
 528         return mm_init(mm, current);
 529 }
 530
 531 /*
 532  * Called when the last reference to the mm
 533  * is dropped: either by a lazy thread or by
 534  * mmput. Free the page directory and the mm.
 535  */
 536 void __mmdrop(struct mm_struct *mm)
 537 {
 538         BUG_ON(mm == &init_mm);
 539         mm_free_pgd(mm);
 540         destroy_context(mm);
 541         mmu_notifier_mm_destroy(mm);
 542 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 543         VM_BUG_ON(mm->pmd_huge_pte);
 544 #endif
 545         free_mm(mm);
 546 }
 547 EXPORT_SYMBOL_GPL(__mmdrop);
 548
 549 /*
 550  * Decrement the use count and release all resources for an mm.
 551  */
 552 void mmput(struct mm_struct *mm)
 553 {
 554         might_sleep();
 555
 556         if (atomic_dec_and_test(&mm->mm_users)) {
 557                 exit_aio(mm);
 558                 ksm_exit(mm);
 559                 khugepaged_exit(mm); /* must run before exit_mmap */
 560                 exit_mmap(mm);
 561                 set_mm_exe_file(mm, NULL);
 562                 if (!list_empty(&mm->mmlist)) {
 563                         spin_lock(&mmlist_lock);
 564                         list_del(&mm->mmlist);
 565                         spin_unlock(&mmlist_lock);
 566                 }
 567                 put_swap_token(mm);
 568                 if (mm->binfmt)
 569                         module_put(mm->binfmt->module);
 570                 mmdrop(mm);
 571         }
 572 }
 573 EXPORT_SYMBOL_GPL(mmput);
 574
 575 /*
 576  * We added or removed a vma mapping the executable. The vmas are only mapped
 577  * during exec and are not mapped with the mmap system call.
 578  * Callers must hold down_write() on the mm's mmap_sem for these
 579  */
 580 void added_exe_file_vma(struct mm_struct *mm)
 581 {
 582         mm->num_exe_file_vmas++;
 583 }
 584
 585 void removed_exe_file_vma(struct mm_struct *mm)
 586 {
 587         mm->num_exe_file_vmas--;
 588         if ((mm->num_exe_file_vmas == 0) && mm->exe_file) {
 589                 fput(mm->exe_file);
 590                 mm->exe_file = NULL;
 591         }
 592
 593 }
 594
 595 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 596 {
 597         if (new_exe_file)
 598                 get_file(new_exe_file);
 599         if (mm->exe_file)
 600                 fput(mm->exe_file);
 601         mm->exe_file = new_exe_file;
 602         mm->num_exe_file_vmas = 0;
 603 }
 604
 605 struct file *get_mm_exe_file(struct mm_struct *mm)
 606 {
 607         struct file *exe_file;
 608
 609         /* We need mmap_sem to protect against races with removal of
 610          * VM_EXECUTABLE vmas */
 611         down_read(&mm->mmap_sem);
 612         exe_file = mm->exe_file;
 613         if (exe_file)
 614                 get_file(exe_file);
 615         up_read(&mm->mmap_sem);
 616         return exe_file;
 617 }
 618
 619 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
 620 {
 621         /* It's safe to write the exe_file pointer without exe_file_lock because
 622          * this is called during fork when the task is not yet in /proc */
 623         newmm->exe_file = get_mm_exe_file(oldmm);
 624 }
 625
 626 /**
 627  * get_task_mm - acquire a reference to the task's mm
 628  *
 629  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 630  * this kernel workthread has transiently adopted a user mm with use_mm,
 631  * to do its AIO) is not set and if so returns a reference to it, after
 632  * bumping up the use count.  User must release the mm via mmput()
 633  * after use.  Typically used by /proc and ptrace.
 634  */
 635 struct mm_struct *get_task_mm(struct task_struct *task)
 636 {
 637         struct mm_struct *mm;
 638
 639         task_lock(task);
 640         mm = task->mm;
 641         if (mm) {
 642                 if (task->flags & PF_KTHREAD)
 643                         mm = NULL;
 644                 else
 645                         atomic_inc(&mm->mm_users);
 646         }
 647         task_unlock(task);
 648         return mm;
 649 }
 650 EXPORT_SYMBOL_GPL(get_task_mm);
 651
 652 /* Please note the differences between mmput and mm_release.
 653  * mmput is called whenever we stop holding onto a mm_struct,
 654  * error success whatever.
 655  *
 656  * mm_release is called after a mm_struct has been removed
 657  * from the current process.
 658  *
 659  * This difference is important for error handling, when we
 660  * only half set up a mm_struct for a new process and need to restore
 661  * the old one.  Because we mmput the new mm_struct before
 662  * restoring the old one. . .
 663  * Eric Biederman 10 January 1998
 664  */
 665 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 666 {
 667         struct completion *vfork_done = tsk->vfork_done;
 668
 669         /* Get rid of any futexes when releasing the mm */
 670 #ifdef CONFIG_FUTEX
 671         if (unlikely(tsk->robust_list)) {
 672                 exit_robust_list(tsk);
 673                 tsk->robust_list = NULL;
 674         }
 675 #ifdef CONFIG_COMPAT
 676         if (unlikely(tsk->compat_robust_list)) {
 677                 compat_exit_robust_list(tsk);
 678                 tsk->compat_robust_list = NULL;
 679         }
 680 #endif
 681         if (unlikely(!list_empty(&tsk->pi_state_list)))
 682                 exit_pi_state_list(tsk);
 683 #endif
 684
 685         /* Get rid of any cached register state */
 686         deactivate_mm(tsk, mm);
 687
 688         /* notify parent sleeping on vfork() */
 689         if (vfork_done) {
 690                 tsk->vfork_done = NULL;
 691                 complete(vfork_done);
 692         }
 693
 694         /*
 695          * If we're exiting normally, clear a user-space tid field if
 696          * requested.  We leave this alone when dying by signal, to leave
 697          * the value intact in a core dump, and to save the unnecessary
 698          * trouble otherwise.  Userland only wants this done for a sys_exit.
 699          */
 700         if (tsk->clear_child_tid) {
 701                 if (!(tsk->flags & PF_SIGNALED) &&
 702                     atomic_read(&mm->mm_users) > 1) {
 703                         /*
 704                          * We don't check the error code - if userspace has
 705                          * not set up a proper pointer then tough luck.
 706                          */
 707                         put_user(0, tsk->clear_child_tid);
 708                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 709                                         1, NULL, NULL, 0);
 710                 }
 711                 tsk->clear_child_tid = NULL;
 712         }
 713 }
 714
 715 /*
 716  * Allocate a new mm structure and copy contents from the
 717  * mm structure of the passed in task structure.
 718  */
 719 struct mm_struct *dup_mm(struct task_struct *tsk)
 720 {
 721         struct mm_struct *mm, *oldmm = current->mm;
 722         int err;
 723
 724         if (!oldmm)
 725                 return NULL;
 726
 727         mm = allocate_mm();
 728         if (!mm)
 729                 goto fail_nomem;
 730
 731         memcpy(mm, oldmm, sizeof(*mm));
 732         mm_init_cpumask(mm);
 733
 734         /* Initializing for Swap token stuff */
 735         mm->token_priority = 0;
 736         mm->last_interval = 0;
 737
 738 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 739         mm->pmd_huge_pte = NULL;
 740 #endif
 741
 742         if (!mm_init(mm, tsk))
 743                 goto fail_nomem;
 744
 745         if (init_new_context(tsk, mm))
 746                 goto fail_nocontext;
 747
 748         dup_mm_exe_file(oldmm, mm);
 749
 750         err = dup_mmap(mm, oldmm);
 751         if (err)
 752                 goto free_pt;
 753
 754         mm->hiwater_rss = get_mm_rss(mm);
 755         mm->hiwater_vm = mm->total_vm;
 756
 757         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 758                 goto free_pt;
 759
 760         return mm;
 761
 762 free_pt:
 763         /* don't put binfmt in mmput, we haven't got module yet */
 764         mm->binfmt = NULL;
 765         mmput(mm);
 766
 767 fail_nomem:
 768         return NULL;
 769
 770 fail_nocontext:
 771         /*
 772          * If init_new_context() failed, we cannot use mmput() to free the mm
 773          * because it calls destroy_context()
 774          */
 775         mm_free_pgd(mm);
 776         free_mm(mm);
 777         return NULL;
 778 }
 779
 780 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 781 {
 782         struct mm_struct *mm, *oldmm;
 783         int retval;
 784
 785         tsk->min_flt = tsk->maj_flt = 0;
 786         tsk->nvcsw = tsk->nivcsw = 0;
 787 #ifdef CONFIG_DETECT_HUNG_TASK
 788         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 789 #endif
 790
 791         tsk->mm = NULL;
 792         tsk->active_mm = NULL;
 793
 794         /*
 795          * Are we cloning a kernel thread?
 796          *
 797          * We need to steal a active VM for that..
 798          */
 799         oldmm = current->mm;
 800         if (!oldmm)
 801                 return 0;
 802
 803         if (clone_flags & CLONE_VM) {
 804                 atomic_inc(&oldmm->mm_users);
 805                 mm = oldmm;
 806                 goto good_mm;
 807         }
 808
 809         retval = -ENOMEM;
 810         mm = dup_mm(tsk);
 811         if (!mm)
 812                 goto fail_nomem;
 813
 814 good_mm:
 815         /* Initializing for Swap token stuff */
 816         mm->token_priority = 0;
 817         mm->last_interval = 0;
 818
 819         tsk->mm = mm;
 820         tsk->active_mm = mm;
 821         return 0;
 822
 823 fail_nomem:
 824         return retval;
 825 }
 826
 827 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 828 {
 829         struct fs_struct *fs = current->fs;
 830         if (clone_flags & CLONE_FS) {
 831                 /* tsk->fs is already what we want */
 832                 spin_lock(&fs->lock);
 833                 if (fs->in_exec) {
 834                         spin_unlock(&fs->lock);
 835                         return -EAGAIN;
 836                 }
 837                 fs->users++;
 838                 spin_unlock(&fs->lock);
 839                 return 0;
 840         }
 841         tsk->fs = copy_fs_struct(fs);
 842         if (!tsk->fs)
 843                 return -ENOMEM;
 844         return 0;
 845 }
 846
 847 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
 848 {
 849         struct files_struct *oldf, *newf;
 850         int error = 0;
 851
 852         /*
 853          * A background process may not have any files ...
 854          */
 855         oldf = current->files;
 856         if (!oldf)
 857                 goto out;
 858
 859         if (clone_flags & CLONE_FILES) {
 860                 atomic_inc(&oldf->count);
 861                 goto out;
 862         }
 863
 864         newf = dup_fd(oldf, &error);
 865         if (!newf)
 866                 goto out;
 867
 868         tsk->files = newf;
 869         error = 0;
 870 out:
 871         return error;
 872 }
 873
 874 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 875 {
 876 #ifdef CONFIG_BLOCK
 877         struct io_context *ioc = current->io_context;
 878
 879         if (!ioc)
 880                 return 0;
 881         /*
 882          * Share io context with parent, if CLONE_IO is set
 883          */
 884         if (clone_flags & CLONE_IO) {
 885                 tsk->io_context = ioc_task_link(ioc);
 886                 if (unlikely(!tsk->io_context))
 887                         return -ENOMEM;
 888         } else if (ioprio_valid(ioc->ioprio)) {
 889                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
 890                 if (unlikely(!tsk->io_context))
 891                         return -ENOMEM;
 892
 893                 tsk->io_context->ioprio = ioc->ioprio;
 894         }
 895 #endif
 896         return 0;
 897 }
 898
 899 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 900 {
 901         struct sighand_struct *sig;
 902
 903         if (clone_flags & CLONE_SIGHAND) {
 904                 atomic_inc(&current->sighand->count);
 905                 return 0;
 906         }
 907         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 908         rcu_assign_pointer(tsk->sighand, sig);
 909         if (!sig)
 910                 return -ENOMEM;
 911         atomic_set(&sig->count, 1);
 912         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 913         return 0;
 914 }
 915
 916 void __cleanup_sighand(struct sighand_struct *sighand)
 917 {
 918         if (atomic_dec_and_test(&sighand->count))
 919                 kmem_cache_free(sighand_cachep, sighand);
 920 }
 921
 922
 923 /*
 924  * Initialize POSIX timer handling for a thread group.
 925  */
 926 static void posix_cpu_timers_init_group(struct signal_struct *sig)
 927 {
 928         unsigned long cpu_limit;
 929
 930         /* Thread group counters. */
 931         thread_group_cputime_init(sig);
 932
 933         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
 934         if (cpu_limit != RLIM_INFINITY) {
 935                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
 936                 sig->cputimer.running = 1;
 937         }
 938
 939         /* The timer lists. */
 940         INIT_LIST_HEAD(&sig->cpu_timers[0]);
 941         INIT_LIST_HEAD(&sig->cpu_timers[1]);
 942         INIT_LIST_HEAD(&sig->cpu_timers[2]);
 943 }
 944
 945 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
 946 {
 947         struct signal_struct *sig;
 948
 949         if (clone_flags & CLONE_THREAD)
 950                 return 0;
 951
 952         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
 953         tsk->signal = sig;
 954         if (!sig)
 955                 return -ENOMEM;
 956
 957         sig->nr_threads = 1;
 958         atomic_set(&sig->live, 1);
 959         atomic_set(&sig->sigcnt, 1);
 960         init_waitqueue_head(&sig->wait_chldexit);
 961         if (clone_flags & CLONE_NEWPID)
 962                 sig->flags |= SIGNAL_UNKILLABLE;
 963         sig->curr_target = tsk;
 964         init_sigpending(&sig->shared_pending);
 965         INIT_LIST_HEAD(&sig->posix_timers);
 966
 967         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 968         sig->real_timer.function = it_real_fn;
 969
 970         task_lock(current->group_leader);
 971         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
 972         task_unlock(current->group_leader);
 973
 974         posix_cpu_timers_init_group(sig);
 975
 976         tty_audit_fork(sig);
 977         sched_autogroup_fork(sig);
 978
 979 #ifdef CONFIG_CGROUPS
 980         init_rwsem(&sig->threadgroup_fork_lock);
 981 #endif
 982
 983         sig->oom_adj = current->signal->oom_adj;
 984         sig->oom_score_adj = current->signal->oom_score_adj;
 985         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
 986
 987         mutex_init(&sig->cred_guard_mutex);
 988
 989         return 0;
 990 }
 991
 992 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
 993 {
 994         unsigned long new_flags = p->flags;
 995
 996         new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
 997         new_flags |= PF_FORKNOEXEC;
 998         new_flags |= PF_STARTING;
 999         p->flags = new_flags;
1000         clear_freeze_flag(p);
1001 }
1002
1003 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1004 {
1005         current->clear_child_tid = tidptr;
1006
1007         return task_pid_vnr(current);
1008 }
1009
1010 static void rt_mutex_init_task(struct task_struct *p)
1011 {
1012         raw_spin_lock_init(&p->pi_lock);
1013 #ifdef CONFIG_RT_MUTEXES
1014         plist_head_init(&p->pi_waiters);
1015         p->pi_blocked_on = NULL;
1016 #endif
1017 }
1018
1019 #ifdef CONFIG_MM_OWNER
1020 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1021 {
1022         mm->owner = p;
1023 }
1024 #endif /* CONFIG_MM_OWNER */
1025
1026 /*
1027  * Initialize POSIX timer handling for a single task.
1028  */
1029 static void posix_cpu_timers_init(struct task_struct *tsk)
1030 {
1031         tsk->cputime_expires.prof_exp = cputime_zero;
1032         tsk->cputime_expires.virt_exp = cputime_zero;
1033         tsk->cputime_expires.sched_exp = 0;
1034         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1035         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1036         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1037 }
1038
1039 /*
1040  * This creates a new process as a copy of the old one,
1041  * but does not actually start it yet.
1042  *
1043  * It copies the registers, and all the appropriate
1044  * parts of the process environment (as per the clone
1045  * flags). The actual kick-off is left to the caller.
1046  */
1047 static struct task_struct *copy_process(unsigned long clone_flags,
1048                                         unsigned long stack_start,
1049                                         struct pt_regs *regs,
1050                                         unsigned long stack_size,
1051                                         int __user *child_tidptr,
1052                                         struct pid *pid,
1053                                         int trace)
1054 {
1055         int retval;
1056         struct task_struct *p;
1057         int cgroup_callbacks_done = 0;
1058
1059         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1060                 return ERR_PTR(-EINVAL);
1061
1062         /*
1063          * Thread groups must share signals as well, and detached threads
1064          * can only be started up within the thread group.
1065          */
1066         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1067                 return ERR_PTR(-EINVAL);
1068
1069         /*
1070          * Shared signal handlers imply shared VM. By way of the above,
1071          * thread groups also imply shared VM. Blocking this case allows
1072          * for various simplifications in other code.
1073          */
1074         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1075                 return ERR_PTR(-EINVAL);
1076
1077         /*
1078          * Siblings of global init remain as zombies on exit since they are
1079          * not reaped by their parent (swapper). To solve this and to avoid
1080          * multi-rooted process trees, prevent global and container-inits
1081          * from creating siblings.
1082          */
1083         if ((clone_flags & CLONE_PARENT) &&
1084                                 current->signal->flags & SIGNAL_UNKILLABLE)
1085                 return ERR_PTR(-EINVAL);
1086
1087         retval = security_task_create(clone_flags);
1088         if (retval)
1089                 goto fork_out;
1090
1091         retval = -ENOMEM;
1092         p = dup_task_struct(current);
1093         if (!p)
1094                 goto fork_out;
1095
1096         ftrace_graph_init_task(p);
1097
1098         rt_mutex_init_task(p);
1099
1100 #ifdef CONFIG_PROVE_LOCKING
1101         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1102         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1103 #endif
1104         retval = -EAGAIN;
1105         if (atomic_read(&p->real_cred->user->processes) >=
1106                         task_rlimit(p, RLIMIT_NPROC)) {
1107                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1108                     p->real_cred->user != INIT_USER)
1109                         goto bad_fork_free;
1110         }
1111         current->flags &= ~PF_NPROC_EXCEEDED;
1112
1113         retval = copy_creds(p, clone_flags);
1114         if (retval < 0)
1115                 goto bad_fork_free;
1116
1117         /*
1118          * If multiple threads are within copy_process(), then this check
1119          * triggers too late. This doesn't hurt, the check is only there
1120          * to stop root fork bombs.
1121          */
1122         retval = -EAGAIN;
1123         if (nr_threads >= max_threads)
1124                 goto bad_fork_cleanup_count;
1125
1126         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1127                 goto bad_fork_cleanup_count;
1128
1129         p->did_exec = 0;
1130         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1131         copy_flags(clone_flags, p);
1132         INIT_LIST_HEAD(&p->children);
1133         INIT_LIST_HEAD(&p->sibling);
1134         rcu_copy_process(p);
1135         p->vfork_done = NULL;
1136         spin_lock_init(&p->alloc_lock);
1137
1138         init_sigpending(&p->pending);
1139
1140         p->utime = cputime_zero;
1141         p->stime = cputime_zero;
1142         p->gtime = cputime_zero;
1143         p->utimescaled = cputime_zero;
1144         p->stimescaled = cputime_zero;
1145 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1146         p->prev_utime = cputime_zero;
1147         p->prev_stime = cputime_zero;
1148 #endif
1149 #if defined(SPLIT_RSS_COUNTING)
1150         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1151 #endif
1152
1153         p->default_timer_slack_ns = current->timer_slack_ns;
1154
1155         task_io_accounting_init(&p->ioac);
1156         acct_clear_integrals(p);
1157
1158         posix_cpu_timers_init(p);
1159
1160         do_posix_clock_monotonic_gettime(&p->start_time);
1161         p->real_start_time = p->start_time;
1162         monotonic_to_bootbased(&p->real_start_time);
1163         p->io_context = NULL;
1164         p->audit_context = NULL;
1165         if (clone_flags & CLONE_THREAD)
1166                 threadgroup_fork_read_lock(current);
1167         cgroup_fork(p);
1168 #ifdef CONFIG_NUMA
1169         p->mempolicy = mpol_dup(p->mempolicy);
1170         if (IS_ERR(p->mempolicy)) {
1171                 retval = PTR_ERR(p->mempolicy);
1172                 p->mempolicy = NULL;
1173                 goto bad_fork_cleanup_cgroup;
1174         }
1175         mpol_fix_fork_child_flag(p);
1176 #endif
1177 #ifdef CONFIG_CPUSETS
1178         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1179         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1180 #endif
1181 #ifdef CONFIG_TRACE_IRQFLAGS
1182         p->irq_events = 0;
1183 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1184         p->hardirqs_enabled = 1;
1185 #else
1186         p->hardirqs_enabled = 0;
1187 #endif
1188         p->hardirq_enable_ip = 0;
1189         p->hardirq_enable_event = 0;
1190         p->hardirq_disable_ip = _THIS_IP_;
1191         p->hardirq_disable_event = 0;
1192         p->softirqs_enabled = 1;
1193         p->softirq_enable_ip = _THIS_IP_;
1194         p->softirq_enable_event = 0;
1195         p->softirq_disable_ip = 0;
1196         p->softirq_disable_event = 0;
1197         p->hardirq_context = 0;
1198         p->softirq_context = 0;
1199 #endif
1200 #ifdef CONFIG_LOCKDEP
1201         p->lockdep_depth = 0; /* no locks held yet */
1202         p->curr_chain_key = 0;
1203         p->lockdep_recursion = 0;
1204 #endif
1205
1206 #ifdef CONFIG_DEBUG_MUTEXES
1207         p->blocked_on = NULL; /* not blocked yet */
1208 #endif
1209 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1210         p->memcg_batch.do_batch = 0;
1211         p->memcg_batch.memcg = NULL;
1212 #endif
1213
1214         /* Perform scheduler related setup. Assign this task to a CPU. */
1215         sched_fork(p);
1216
1217         retval = perf_event_init_task(p);
1218         if (retval)
1219                 goto bad_fork_cleanup_policy;
1220         retval = audit_alloc(p);
1221         if (retval)
1222                 goto bad_fork_cleanup_policy;
1223         /* copy all the process information */
1224         retval = copy_semundo(clone_flags, p);
1225         if (retval)
1226                 goto bad_fork_cleanup_audit;
1227         retval = copy_files(clone_flags, p);
1228         if (retval)
1229                 goto bad_fork_cleanup_semundo;
1230         retval = copy_fs(clone_flags, p);
1231         if (retval)
1232                 goto bad_fork_cleanup_files;
1233         retval = copy_sighand(clone_flags, p);
1234         if (retval)
1235                 goto bad_fork_cleanup_fs;
1236         retval = copy_signal(clone_flags, p);
1237         if (retval)
1238                 goto bad_fork_cleanup_sighand;
1239         retval = copy_mm(clone_flags, p);
1240         if (retval)
1241                 goto bad_fork_cleanup_signal;
1242         retval = copy_namespaces(clone_flags, p);
1243         if (retval)
1244                 goto bad_fork_cleanup_mm;
1245         retval = copy_io(clone_flags, p);
1246         if (retval)
1247                 goto bad_fork_cleanup_namespaces;
1248         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1249         if (retval)
1250                 goto bad_fork_cleanup_io;
1251
1252         if (pid != &init_struct_pid) {
1253                 retval = -ENOMEM;
1254                 pid = alloc_pid(p->nsproxy->pid_ns);
1255                 if (!pid)
1256                         goto bad_fork_cleanup_io;
1257         }
1258
1259         p->pid = pid_nr(pid);
1260         p->tgid = p->pid;
1261         if (clone_flags & CLONE_THREAD)
1262                 p->tgid = current->tgid;
1263
1264         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1265         /*
1266          * Clear TID on mm_release()?
1267          */
1268         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1269 #ifdef CONFIG_BLOCK
1270         p->plug = NULL;
1271 #endif
1272 #ifdef CONFIG_FUTEX
1273         p->robust_list = NULL;
1274 #ifdef CONFIG_COMPAT
1275         p->compat_robust_list = NULL;
1276 #endif
1277         INIT_LIST_HEAD(&p->pi_state_list);
1278         p->pi_state_cache = NULL;
1279 #endif
1280         /*
1281          * sigaltstack should be cleared when sharing the same VM
1282          */
1283         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1284                 p->sas_ss_sp = p->sas_ss_size = 0;
1285
1286         /*
1287          * Syscall tracing and stepping should be turned off in the
1288          * child regardless of CLONE_PTRACE.
1289          */
1290         user_disable_single_step(p);
1291         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1292 #ifdef TIF_SYSCALL_EMU
1293         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1294 #endif
1295         clear_all_latency_tracing(p);
1296
1297         /* ok, now we should be set up.. */
1298         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1299         p->pdeath_signal = 0;
1300         p->exit_state = 0;
1301
1302         p->nr_dirtied = 0;
1303         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1304
1305         /*
1306          * Ok, make it visible to the rest of the system.
1307          * We dont wake it up yet.
1308          */
1309         p->group_leader = p;
1310         INIT_LIST_HEAD(&p->thread_group);
1311
1312         /* Now that the task is set up, run cgroup callbacks if
1313          * necessary. We need to run them before the task is visible
1314          * on the tasklist. */
1315         cgroup_fork_callbacks(p);
1316         cgroup_callbacks_done = 1;
1317
1318         /* Need tasklist lock for parent etc handling! */
1319         write_lock_irq(&tasklist_lock);
1320
1321         /* CLONE_PARENT re-uses the old parent */
1322         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1323                 p->real_parent = current->real_parent;
1324                 p->parent_exec_id = current->parent_exec_id;
1325         } else {
1326                 p->real_parent = current;
1327                 p->parent_exec_id = current->self_exec_id;
1328         }
1329
1330         spin_lock(&current->sighand->siglock);
1331
1332         /*
1333          * Process group and session signals need to be delivered to just the
1334          * parent before the fork or both the parent and the child after the
1335          * fork. Restart if a signal comes in before we add the new process to
1336          * it's process group.
1337          * A fatal signal pending means that current will exit, so the new
1338          * thread can't slip out of an OOM kill (or normal SIGKILL).
1339         */
1340         recalc_sigpending();
1341         if (signal_pending(current)) {
1342                 spin_unlock(&current->sighand->siglock);
1343                 write_unlock_irq(&tasklist_lock);
1344                 retval = -ERESTARTNOINTR;
1345                 goto bad_fork_free_pid;
1346         }
1347
1348         if (clone_flags & CLONE_THREAD) {
1349                 current->signal->nr_threads++;
1350                 atomic_inc(&current->signal->live);
1351                 atomic_inc(&current->signal->sigcnt);
1352                 p->group_leader = current->group_leader;
1353                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1354         }
1355
1356         if (likely(p->pid)) {
1357                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1358
1359                 if (thread_group_leader(p)) {
1360                         if (is_child_reaper(pid))
1361                                 p->nsproxy->pid_ns->child_reaper = p;
1362
1363                         p->signal->leader_pid = pid;
1364                         p->signal->tty = tty_kref_get(current->signal->tty);
1365                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1366                         attach_pid(p, PIDTYPE_SID, task_session(current));
1367                         list_add_tail(&p->sibling, &p->real_parent->children);
1368                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1369                         __this_cpu_inc(process_counts);
1370                 }
1371                 attach_pid(p, PIDTYPE_PID, pid);
1372                 nr_threads++;
1373         }
1374
1375         total_forks++;
1376         spin_unlock(&current->sighand->siglock);
1377         write_unlock_irq(&tasklist_lock);
1378         proc_fork_connector(p);
1379         cgroup_post_fork(p);
1380         if (clone_flags & CLONE_THREAD)
1381                 threadgroup_fork_read_unlock(current);
1382         perf_event_fork(p);
1383         return p;
1384
1385 bad_fork_free_pid:
1386         if (pid != &init_struct_pid)
1387                 free_pid(pid);
1388 bad_fork_cleanup_io:
1389         if (p->io_context)
1390                 exit_io_context(p);
1391 bad_fork_cleanup_namespaces:
1392         exit_task_namespaces(p);
1393 bad_fork_cleanup_mm:
1394         if (p->mm)
1395                 mmput(p->mm);
1396 bad_fork_cleanup_signal:
1397         if (!(clone_flags & CLONE_THREAD))
1398                 free_signal_struct(p->signal);
1399 bad_fork_cleanup_sighand:
1400         __cleanup_sighand(p->sighand);
1401 bad_fork_cleanup_fs:
1402         exit_fs(p); /* blocking */
1403 bad_fork_cleanup_files:
1404         exit_files(p); /* blocking */
1405 bad_fork_cleanup_semundo:
1406         exit_sem(p);
1407 bad_fork_cleanup_audit:
1408         audit_free(p);
1409 bad_fork_cleanup_policy:
1410         perf_event_free_task(p);
1411 #ifdef CONFIG_NUMA
1412         mpol_put(p->mempolicy);
1413 bad_fork_cleanup_cgroup:
1414 #endif
1415         if (clone_flags & CLONE_THREAD)
1416                 threadgroup_fork_read_unlock(current);
1417         cgroup_exit(p, cgroup_callbacks_done);
1418         delayacct_tsk_free(p);
1419         module_put(task_thread_info(p)->exec_domain->module);
1420 bad_fork_cleanup_count:
1421         atomic_dec(&p->cred->user->processes);
1422         exit_creds(p);
1423 bad_fork_free:
1424         free_task(p);
1425 fork_out:
1426         return ERR_PTR(retval);
1427 }
1428
1429 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1430 {
1431         memset(regs, 0, sizeof(struct pt_regs));
1432         return regs;
1433 }
1434
1435 static inline void init_idle_pids(struct pid_link *links)
1436 {
1437         enum pid_type type;
1438
1439         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1440                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1441                 links[type].pid = &init_struct_pid;
1442         }
1443 }
1444
1445 struct task_struct * __cpuinit fork_idle(int cpu)
1446 {
1447         struct task_struct *task;
1448         struct pt_regs regs;
1449
1450         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1451                             &init_struct_pid, 0);
1452         if (!IS_ERR(task)) {
1453                 init_idle_pids(task->pids);
1454                 init_idle(task, cpu);
1455         }
1456
1457         return task;
1458 }
1459
1460 /*
1461  *  Ok, this is the main fork-routine.
1462  *
1463  * It copies the process, and if successful kick-starts
1464  * it and waits for it to finish using the VM if required.
1465  */
1466 long do_fork(unsigned long clone_flags,
1467               unsigned long stack_start,
1468               struct pt_regs *regs,
1469               unsigned long stack_size,
1470               int __user *parent_tidptr,
1471               int __user *child_tidptr)
1472 {
1473         struct task_struct *p;
1474         int trace = 0;
1475         long nr;
1476
1477         /*
1478          * Do some preliminary argument and permissions checking before we
1479          * actually start allocating stuff
1480          */
1481         if (clone_flags & CLONE_NEWUSER) {
1482                 if (clone_flags & CLONE_THREAD)
1483                         return -EINVAL;
1484                 /* hopefully this check will go away when userns support is
1485                  * complete
1486                  */
1487                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1488                                 !capable(CAP_SETGID))
1489                         return -EPERM;
1490         }
1491
1492         /*
1493          * Determine whether and which event to report to ptracer.  When
1494          * called from kernel_thread or CLONE_UNTRACED is explicitly
1495          * requested, no event is reported; otherwise, report if the event
1496          * for the type of forking is enabled.
1497          */
1498         if (likely(user_mode(regs)) && !(clone_flags & CLONE_UNTRACED)) {
1499                 if (clone_flags & CLONE_VFORK)
1500                         trace = PTRACE_EVENT_VFORK;
1501                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1502                         trace = PTRACE_EVENT_CLONE;
1503                 else
1504                         trace = PTRACE_EVENT_FORK;
1505
1506                 if (likely(!ptrace_event_enabled(current, trace)))
1507                         trace = 0;
1508         }
1509
1510         p = copy_process(clone_flags, stack_start, regs, stack_size,
1511                          child_tidptr, NULL, trace);
1512         /*
1513          * Do this prior waking up the new thread - the thread pointer
1514          * might get invalid after that point, if the thread exits quickly.
1515          */
1516         if (!IS_ERR(p)) {
1517                 struct completion vfork;
1518
1519                 trace_sched_process_fork(current, p);
1520
1521                 nr = task_pid_vnr(p);
1522
1523                 if (clone_flags & CLONE_PARENT_SETTID)
1524                         put_user(nr, parent_tidptr);
1525
1526                 if (clone_flags & CLONE_VFORK) {
1527                         p->vfork_done = &vfork;
1528                         init_completion(&vfork);
1529                 }
1530
1531                 audit_finish_fork(p);
1532
1533                 /*
1534                  * We set PF_STARTING at creation in case tracing wants to
1535                  * use this to distinguish a fully live task from one that
1536                  * hasn't finished SIGSTOP raising yet.  Now we clear it
1537                  * and set the child going.
1538                  */
1539                 p->flags &= ~PF_STARTING;
1540
1541                 wake_up_new_task(p);
1542
1543                 /* forking complete and child started to run, tell ptracer */
1544                 if (unlikely(trace))
1545                         ptrace_event(trace, nr);
1546
1547                 if (clone_flags & CLONE_VFORK) {
1548                         freezer_do_not_count();
1549                         wait_for_completion(&vfork);
1550                         freezer_count();
1551                         ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1552                 }
1553         } else {
1554                 nr = PTR_ERR(p);
1555         }
1556         return nr;
1557 }
1558
1559 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1560 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1561 #endif
1562
1563 static void sighand_ctor(void *data)
1564 {
1565         struct sighand_struct *sighand = data;
1566
1567         spin_lock_init(&sighand->siglock);
1568         init_waitqueue_head(&sighand->signalfd_wqh);
1569 }
1570
1571 void __init proc_caches_init(void)
1572 {
1573         sighand_cachep = kmem_cache_create("sighand_cache",
1574                         sizeof(struct sighand_struct), 0,
1575                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1576                         SLAB_NOTRACK, sighand_ctor);
1577         signal_cachep = kmem_cache_create("signal_cache",
1578                         sizeof(struct signal_struct), 0,
1579                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1580         files_cachep = kmem_cache_create("files_cache",
1581                         sizeof(struct files_struct), 0,
1582                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1583         fs_cachep = kmem_cache_create("fs_cache",
1584                         sizeof(struct fs_struct), 0,
1585                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1586         /*
1587          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1588          * whole struct cpumask for the OFFSTACK case. We could change
1589          * this to *only* allocate as much of it as required by the
1590          * maximum number of CPU's we can ever have.  The cpumask_allocation
1591          * is at the end of the structure, exactly for that reason.
1592          */
1593         mm_cachep = kmem_cache_create("mm_struct",
1594                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1595                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1596         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1597         mmap_init();
1598         nsproxy_cache_init();
1599 }
1600
1601 /*
1602  * Check constraints on flags passed to the unshare system call.
1603  */
1604 static int check_unshare_flags(unsigned long unshare_flags)
1605 {
1606         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1607                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1608                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1609                 return -EINVAL;
1610         /*
1611          * Not implemented, but pretend it works if there is nothing to
1612          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1613          * needs to unshare vm.
1614          */
1615         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1616                 /* FIXME: get_task_mm() increments ->mm_users */
1617                 if (atomic_read(&current->mm->mm_users) > 1)
1618                         return -EINVAL;
1619         }
1620
1621         return 0;
1622 }
1623
1624 /*
1625  * Unshare the filesystem structure if it is being shared
1626  */
1627 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1628 {
1629         struct fs_struct *fs = current->fs;
1630
1631         if (!(unshare_flags & CLONE_FS) || !fs)
1632                 return 0;
1633
1634         /* don't need lock here; in the worst case we'll do useless copy */
1635         if (fs->users == 1)
1636                 return 0;
1637
1638         *new_fsp = copy_fs_struct(fs);
1639         if (!*new_fsp)
1640                 return -ENOMEM;
1641
1642         return 0;
1643 }
1644
1645 /*
1646  * Unshare file descriptor table if it is being shared
1647  */
1648 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1649 {
1650         struct files_struct *fd = current->files;
1651         int error = 0;
1652
1653         if ((unshare_flags & CLONE_FILES) &&
1654             (fd && atomic_read(&fd->count) > 1)) {
1655                 *new_fdp = dup_fd(fd, &error);
1656                 if (!*new_fdp)
1657                         return error;
1658         }
1659
1660         return 0;
1661 }
1662
1663 /*
1664  * unshare allows a process to 'unshare' part of the process
1665  * context which was originally shared using clone.  copy_*
1666  * functions used by do_fork() cannot be used here directly
1667  * because they modify an inactive task_struct that is being
1668  * constructed. Here we are modifying the current, active,
1669  * task_struct.
1670  */
1671 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1672 {
1673         struct fs_struct *fs, *new_fs = NULL;
1674         struct files_struct *fd, *new_fd = NULL;
1675         struct nsproxy *new_nsproxy = NULL;
1676         int do_sysvsem = 0;
1677         int err;
1678
1679         err = check_unshare_flags(unshare_flags);
1680         if (err)
1681                 goto bad_unshare_out;
1682
1683         /*
1684          * If unsharing namespace, must also unshare filesystem information.
1685          */
1686         if (unshare_flags & CLONE_NEWNS)
1687                 unshare_flags |= CLONE_FS;
1688         /*
1689          * CLONE_NEWIPC must also detach from the undolist: after switching
1690          * to a new ipc namespace, the semaphore arrays from the old
1691          * namespace are unreachable.
1692          */
1693         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1694                 do_sysvsem = 1;
1695         err = unshare_fs(unshare_flags, &new_fs);
1696         if (err)
1697                 goto bad_unshare_out;
1698         err = unshare_fd(unshare_flags, &new_fd);
1699         if (err)
1700                 goto bad_unshare_cleanup_fs;
1701         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy, new_fs);
1702         if (err)
1703                 goto bad_unshare_cleanup_fd;
1704
1705         if (new_fs || new_fd || do_sysvsem || new_nsproxy) {
1706                 if (do_sysvsem) {
1707                         /*
1708                          * CLONE_SYSVSEM is equivalent to sys_exit().
1709                          */
1710                         exit_sem(current);
1711                 }
1712
1713                 if (new_nsproxy) {
1714                         switch_task_namespaces(current, new_nsproxy);
1715                         new_nsproxy = NULL;
1716                 }
1717
1718                 task_lock(current);
1719
1720                 if (new_fs) {
1721                         fs = current->fs;
1722                         spin_lock(&fs->lock);
1723                         current->fs = new_fs;
1724                         if (--fs->users)
1725                                 new_fs = NULL;
1726                         else
1727                                 new_fs = fs;
1728                         spin_unlock(&fs->lock);
1729                 }
1730
1731                 if (new_fd) {
1732                         fd = current->files;
1733                         current->files = new_fd;
1734                         new_fd = fd;
1735                 }
1736
1737                 task_unlock(current);
1738         }
1739
1740         if (new_nsproxy)
1741                 put_nsproxy(new_nsproxy);
1742
1743 bad_unshare_cleanup_fd:
1744         if (new_fd)
1745                 put_files_struct(new_fd);
1746
1747 bad_unshare_cleanup_fs:
1748         if (new_fs)
1749                 free_fs_struct(new_fs);
1750
1751 bad_unshare_out:
1752         return err;
1753 }
1754
1755 /*
1756  *      Helper to unshare the files of the current task.
1757  *      We don't want to expose copy_files internals to
1758  *      the exec layer of the kernel.
1759  */
1760
1761 int unshare_files(struct files_struct **displaced)
1762 {
1763         struct task_struct *task = current;
1764         struct files_struct *copy = NULL;
1765         int error;
1766
1767         error = unshare_fd(CLONE_FILES, &copy);
1768         if (error || !copy) {
1769                 *displaced = NULL;
1770                 return error;
1771         }
1772         *displaced = task->files;
1773         task_lock(task);
1774         task->files = copy;
1775         task_unlock(task);
1776         return 0;
1777 }