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/seccomp.h>
  38 #include <linux/swap.h>
  39 #include <linux/syscalls.h>
  40 #include <linux/jiffies.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/proc_fs.h>
  52 #include <linux/profile.h>
  53 #include <linux/rmap.h>
  54 #include <linux/ksm.h>
  55 #include <linux/acct.h>
  56 #include <linux/tsacct_kern.h>
  57 #include <linux/cn_proc.h>
  58 #include <linux/freezer.h>
  59 #include <linux/delayacct.h>
  60 #include <linux/taskstats_kern.h>
  61 #include <linux/random.h>
  62 #include <linux/tty.h>
  63 #include <linux/blkdev.h>
  64 #include <linux/fs_struct.h>
  65 #include <linux/magic.h>
  66 #include <linux/perf_event.h>
  67 #include <linux/posix-timers.h>
  68 #include <linux/user-return-notifier.h>
  69 #include <linux/oom.h>
  70 #include <linux/khugepaged.h>
  71 #include <linux/signalfd.h>
  72 #include <linux/uprobes.h>
  73 #include <linux/aio.h>
  74
  75 #include <asm/pgtable.h>
  76 #include <asm/pgalloc.h>
  77 #include <asm/uaccess.h>
  78 #include <asm/mmu_context.h>
  79 #include <asm/cacheflush.h>
  80 #include <asm/tlbflush.h>
  81
  82 #include <trace/events/sched.h>
  83
  84 #define CREATE_TRACE_POINTS
  85 #include <trace/events/task.h>
  86
  87 /*
  88  * Protected counters by write_lock_irq(&tasklist_lock)
  89  */
  90 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  91 int nr_threads;                 /* The idle threads do not count.. */
  92
  93 int max_threads;                /* tunable limit on nr_threads */
  94
  95 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  96
  97 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  98
  99 #ifdef CONFIG_PROVE_RCU
 100 int lockdep_tasklist_lock_is_held(void)
 101 {
 102         return lockdep_is_held(&tasklist_lock);
 103 }
 104 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 105 #endif /* #ifdef CONFIG_PROVE_RCU */
 106
 107 int nr_processes(void)
 108 {
 109         int cpu;
 110         int total = 0;
 111
 112         for_each_possible_cpu(cpu)
 113                 total += per_cpu(process_counts, cpu);
 114
 115         return total;
 116 }
 117
 118 void __weak arch_release_task_struct(struct task_struct *tsk)
 119 {
 120 }
 121
 122 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 123 static struct kmem_cache *task_struct_cachep;
 124
 125 static inline struct task_struct *alloc_task_struct_node(int node)
 126 {
 127         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 128 }
 129
 130 static inline void free_task_struct(struct task_struct *tsk)
 131 {
 132         kmem_cache_free(task_struct_cachep, tsk);
 133 }
 134 #endif
 135
 136 void __weak arch_release_thread_info(struct thread_info *ti)
 137 {
 138 }
 139
 140 #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
 141
 142 /*
 143  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 144  * kmemcache based allocator.
 145  */
 146 # if THREAD_SIZE >= PAGE_SIZE
 147 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 148                                                   int node)
 149 {
 150         struct page *page = alloc_pages_node(node, THREADINFO_GFP_ACCOUNTED,
 151                                              THREAD_SIZE_ORDER);
 152
 153         return page ? page_address(page) : NULL;
 154 }
 155
 156 static inline void free_thread_info(struct thread_info *ti)
 157 {
 158         free_memcg_kmem_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 159 }
 160 # else
 161 static struct kmem_cache *thread_info_cache;
 162
 163 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 164                                                   int node)
 165 {
 166         return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
 167 }
 168
 169 static void free_thread_info(struct thread_info *ti)
 170 {
 171         kmem_cache_free(thread_info_cache, ti);
 172 }
 173
 174 void thread_info_cache_init(void)
 175 {
 176         thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
 177                                               THREAD_SIZE, 0, NULL);
 178         BUG_ON(thread_info_cache == NULL);
 179 }
 180 # endif
 181 #endif
 182
 183 /* SLAB cache for signal_struct structures (tsk->signal) */
 184 static struct kmem_cache *signal_cachep;
 185
 186 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 187 struct kmem_cache *sighand_cachep;
 188
 189 /* SLAB cache for files_struct structures (tsk->files) */
 190 struct kmem_cache *files_cachep;
 191
 192 /* SLAB cache for fs_struct structures (tsk->fs) */
 193 struct kmem_cache *fs_cachep;
 194
 195 /* SLAB cache for vm_area_struct structures */
 196 struct kmem_cache *vm_area_cachep;
 197
 198 /* SLAB cache for mm_struct structures (tsk->mm) */
 199 static struct kmem_cache *mm_cachep;
 200
 201 static void account_kernel_stack(struct thread_info *ti, int account)
 202 {
 203         struct zone *zone = page_zone(virt_to_page(ti));
 204
 205         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
 206 }
 207
 208 void free_task(struct task_struct *tsk)
 209 {
 210         account_kernel_stack(tsk->stack, -1);
 211         arch_release_thread_info(tsk->stack);
 212         free_thread_info(tsk->stack);
 213         rt_mutex_debug_task_free(tsk);
 214         ftrace_graph_exit_task(tsk);
 215         put_seccomp_filter(tsk);
 216         arch_release_task_struct(tsk);
 217         free_task_struct(tsk);
 218 }
 219 EXPORT_SYMBOL(free_task);
 220
 221 static inline void free_signal_struct(struct signal_struct *sig)
 222 {
 223         taskstats_tgid_free(sig);
 224         sched_autogroup_exit(sig);
 225         kmem_cache_free(signal_cachep, sig);
 226 }
 227
 228 static inline void put_signal_struct(struct signal_struct *sig)
 229 {
 230         if (atomic_dec_and_test(&sig->sigcnt))
 231                 free_signal_struct(sig);
 232 }
 233
 234 void __put_task_struct(struct task_struct *tsk)
 235 {
 236         WARN_ON(!tsk->exit_state);
 237         WARN_ON(atomic_read(&tsk->usage));
 238         WARN_ON(tsk == current);
 239
 240         task_numa_free(tsk);
 241         security_task_free(tsk);
 242         exit_creds(tsk);
 243         delayacct_tsk_free(tsk);
 244         put_signal_struct(tsk->signal);
 245
 246         if (!profile_handoff_task(tsk))
 247                 free_task(tsk);
 248 }
 249 EXPORT_SYMBOL_GPL(__put_task_struct);
 250
 251 void __init __weak arch_task_cache_init(void) { }
 252
 253 void __init fork_init(unsigned long mempages)
 254 {
 255 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 256 #ifndef ARCH_MIN_TASKALIGN
 257 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 258 #endif
 259         /* create a slab on which task_structs can be allocated */
 260         task_struct_cachep =
 261                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 262                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
 263 #endif
 264
 265         /* do the arch specific task caches init */
 266         arch_task_cache_init();
 267
 268         /*
 269          * The default maximum number of threads is set to a safe
 270          * value: the thread structures can take up at most half
 271          * of memory.
 272          */
 273         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 274
 275         /*
 276          * we need to allow at least 20 threads to boot a system
 277          */
 278         if (max_threads < 20)
 279                 max_threads = 20;
 280
 281         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 282         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 283         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 284                 init_task.signal->rlim[RLIMIT_NPROC];
 285 }
 286
 287 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
 288                                                struct task_struct *src)
 289 {
 290         *dst = *src;
 291         return 0;
 292 }
 293
 294 static struct task_struct *dup_task_struct(struct task_struct *orig)
 295 {
 296         struct task_struct *tsk;
 297         struct thread_info *ti;
 298         unsigned long *stackend;
 299         int node = tsk_fork_get_node(orig);
 300         int err;
 301
 302         tsk = alloc_task_struct_node(node);
 303         if (!tsk)
 304                 return NULL;
 305
 306         ti = alloc_thread_info_node(tsk, node);
 307         if (!ti)
 308                 goto free_tsk;
 309
 310         err = arch_dup_task_struct(tsk, orig);
 311         if (err)
 312                 goto free_ti;
 313
 314         tsk->stack = ti;
 315
 316         setup_thread_stack(tsk, orig);
 317         clear_user_return_notifier(tsk);
 318         clear_tsk_need_resched(tsk);
 319         stackend = end_of_stack(tsk);
 320         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 321
 322 #ifdef CONFIG_CC_STACKPROTECTOR
 323         tsk->stack_canary = get_random_int();
 324 #endif
 325
 326         /*
 327          * One for us, one for whoever does the "release_task()" (usually
 328          * parent)
 329          */
 330         atomic_set(&tsk->usage, 2);
 331 #ifdef CONFIG_BLK_DEV_IO_TRACE
 332         tsk->btrace_seq = 0;
 333 #endif
 334         tsk->splice_pipe = NULL;
 335         tsk->task_frag.page = NULL;
 336
 337         account_kernel_stack(ti, 1);
 338
 339         return tsk;
 340
 341 free_ti:
 342         free_thread_info(ti);
 343 free_tsk:
 344         free_task_struct(tsk);
 345         return NULL;
 346 }
 347
 348 #ifdef CONFIG_MMU
 349 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 350 {
 351         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 352         struct rb_node **rb_link, *rb_parent;
 353         int retval;
 354         unsigned long charge;
 355
 356         uprobe_start_dup_mmap();
 357         down_write(&oldmm->mmap_sem);
 358         flush_cache_dup_mm(oldmm);
 359         uprobe_dup_mmap(oldmm, mm);
 360         /*
 361          * Not linked in yet - no deadlock potential:
 362          */
 363         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 364
 365         mm->locked_vm = 0;
 366         mm->mmap = NULL;
 367         mm->mmap_cache = NULL;
 368         mm->map_count = 0;
 369         cpumask_clear(mm_cpumask(mm));
 370         mm->mm_rb = RB_ROOT;
 371         rb_link = &mm->mm_rb.rb_node;
 372         rb_parent = NULL;
 373         pprev = &mm->mmap;
 374         retval = ksm_fork(mm, oldmm);
 375         if (retval)
 376                 goto out;
 377         retval = khugepaged_fork(mm, oldmm);
 378         if (retval)
 379                 goto out;
 380
 381         prev = NULL;
 382         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 383                 struct file *file;
 384
 385                 if (mpnt->vm_flags & VM_DONTCOPY) {
 386                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 387                                                         -vma_pages(mpnt));
 388                         continue;
 389                 }
 390                 charge = 0;
 391                 if (mpnt->vm_flags & VM_ACCOUNT) {
 392                         unsigned long len = vma_pages(mpnt);
 393
 394                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 395                                 goto fail_nomem;
 396                         charge = len;
 397                 }
 398                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 399                 if (!tmp)
 400                         goto fail_nomem;
 401                 *tmp = *mpnt;
 402                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 403                 retval = vma_dup_policy(mpnt, tmp);
 404                 if (retval)
 405                         goto fail_nomem_policy;
 406                 tmp->vm_mm = mm;
 407                 if (anon_vma_fork(tmp, mpnt))
 408                         goto fail_nomem_anon_vma_fork;
 409                 tmp->vm_flags &= ~VM_LOCKED;
 410                 tmp->vm_next = tmp->vm_prev = NULL;
 411                 file = tmp->vm_file;
 412                 if (file) {
 413                         struct inode *inode = file_inode(file);
 414                         struct address_space *mapping = file->f_mapping;
 415
 416                         get_file(file);
 417                         if (tmp->vm_flags & VM_DENYWRITE)
 418                                 atomic_dec(&inode->i_writecount);
 419                         mutex_lock(&mapping->i_mmap_mutex);
 420                         if (tmp->vm_flags & VM_SHARED)
 421                                 mapping->i_mmap_writable++;
 422                         flush_dcache_mmap_lock(mapping);
 423                         /* insert tmp into the share list, just after mpnt */
 424                         if (unlikely(tmp->vm_flags & VM_NONLINEAR))
 425                                 vma_nonlinear_insert(tmp,
 426                                                 &mapping->i_mmap_nonlinear);
 427                         else
 428                                 vma_interval_tree_insert_after(tmp, mpnt,
 429                                                         &mapping->i_mmap);
 430                         flush_dcache_mmap_unlock(mapping);
 431                         mutex_unlock(&mapping->i_mmap_mutex);
 432                 }
 433
 434                 /*
 435                  * Clear hugetlb-related page reserves for children. This only
 436                  * affects MAP_PRIVATE mappings. Faults generated by the child
 437                  * are not guaranteed to succeed, even if read-only
 438                  */
 439                 if (is_vm_hugetlb_page(tmp))
 440                         reset_vma_resv_huge_pages(tmp);
 441
 442                 /*
 443                  * Link in the new vma and copy the page table entries.
 444                  */
 445                 *pprev = tmp;
 446                 pprev = &tmp->vm_next;
 447                 tmp->vm_prev = prev;
 448                 prev = tmp;
 449
 450                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 451                 rb_link = &tmp->vm_rb.rb_right;
 452                 rb_parent = &tmp->vm_rb;
 453
 454                 mm->map_count++;
 455                 retval = copy_page_range(mm, oldmm, mpnt);
 456
 457                 if (tmp->vm_ops && tmp->vm_ops->open)
 458                         tmp->vm_ops->open(tmp);
 459
 460                 if (retval)
 461                         goto out;
 462         }
 463         /* a new mm has just been created */
 464         arch_dup_mmap(oldmm, mm);
 465         retval = 0;
 466 out:
 467         up_write(&mm->mmap_sem);
 468         flush_tlb_mm(oldmm);
 469         up_write(&oldmm->mmap_sem);
 470         uprobe_end_dup_mmap();
 471         return retval;
 472 fail_nomem_anon_vma_fork:
 473         mpol_put(vma_policy(tmp));
 474 fail_nomem_policy:
 475         kmem_cache_free(vm_area_cachep, tmp);
 476 fail_nomem:
 477         retval = -ENOMEM;
 478         vm_unacct_memory(charge);
 479         goto out;
 480 }
 481
 482 static inline int mm_alloc_pgd(struct mm_struct *mm)
 483 {
 484         mm->pgd = pgd_alloc(mm);
 485         if (unlikely(!mm->pgd))
 486                 return -ENOMEM;
 487         return 0;
 488 }
 489
 490 static inline void mm_free_pgd(struct mm_struct *mm)
 491 {
 492         pgd_free(mm, mm->pgd);
 493 }
 494 #else
 495 #define dup_mmap(mm, oldmm)     (0)
 496 #define mm_alloc_pgd(mm)        (0)
 497 #define mm_free_pgd(mm)
 498 #endif /* CONFIG_MMU */
 499
 500 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 501
 502 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 503 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 504
 505 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 506
 507 static int __init coredump_filter_setup(char *s)
 508 {
 509         default_dump_filter =
 510                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 511                 MMF_DUMP_FILTER_MASK;
 512         return 1;
 513 }
 514
 515 __setup("coredump_filter=", coredump_filter_setup);
 516
 517 #include <linux/init_task.h>
 518
 519 static void mm_init_aio(struct mm_struct *mm)
 520 {
 521 #ifdef CONFIG_AIO
 522         spin_lock_init(&mm->ioctx_lock);
 523         mm->ioctx_table = NULL;
 524 #endif
 525 }
 526
 527 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
 528 {
 529         atomic_set(&mm->mm_users, 1);
 530         atomic_set(&mm->mm_count, 1);
 531         init_rwsem(&mm->mmap_sem);
 532         INIT_LIST_HEAD(&mm->mmlist);
 533         mm->flags = (current->mm) ?
 534                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
 535         mm->core_state = NULL;
 536         atomic_long_set(&mm->nr_ptes, 0);
 537         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 538         spin_lock_init(&mm->page_table_lock);
 539         mm_init_aio(mm);
 540         mm_init_owner(mm, p);
 541         clear_tlb_flush_pending(mm);
 542
 543         if (likely(!mm_alloc_pgd(mm))) {
 544                 mm->def_flags = 0;
 545                 mmu_notifier_mm_init(mm);
 546                 return mm;
 547         }
 548
 549         free_mm(mm);
 550         return NULL;
 551 }
 552
 553 static void check_mm(struct mm_struct *mm)
 554 {
 555         int i;
 556
 557         for (i = 0; i < NR_MM_COUNTERS; i++) {
 558                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 559
 560                 if (unlikely(x))
 561                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 562                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 563         }
 564
 565 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 566         VM_BUG_ON(mm->pmd_huge_pte);
 567 #endif
 568 }
 569
 570 /*
 571  * Allocate and initialize an mm_struct.
 572  */
 573 struct mm_struct *mm_alloc(void)
 574 {
 575         struct mm_struct *mm;
 576
 577         mm = allocate_mm();
 578         if (!mm)
 579                 return NULL;
 580
 581         memset(mm, 0, sizeof(*mm));
 582         mm_init_cpumask(mm);
 583         return mm_init(mm, current);
 584 }
 585
 586 /*
 587  * Called when the last reference to the mm
 588  * is dropped: either by a lazy thread or by
 589  * mmput. Free the page directory and the mm.
 590  */
 591 void __mmdrop(struct mm_struct *mm)
 592 {
 593         BUG_ON(mm == &init_mm);
 594         mm_free_pgd(mm);
 595         destroy_context(mm);
 596         mmu_notifier_mm_destroy(mm);
 597         check_mm(mm);
 598         free_mm(mm);
 599 }
 600 EXPORT_SYMBOL_GPL(__mmdrop);
 601
 602 /*
 603  * Decrement the use count and release all resources for an mm.
 604  */
 605 void mmput(struct mm_struct *mm)
 606 {
 607         might_sleep();
 608
 609         if (atomic_dec_and_test(&mm->mm_users)) {
 610                 uprobe_clear_state(mm);
 611                 exit_aio(mm);
 612                 ksm_exit(mm);
 613                 khugepaged_exit(mm); /* must run before exit_mmap */
 614                 exit_mmap(mm);
 615                 set_mm_exe_file(mm, NULL);
 616                 if (!list_empty(&mm->mmlist)) {
 617                         spin_lock(&mmlist_lock);
 618                         list_del(&mm->mmlist);
 619                         spin_unlock(&mmlist_lock);
 620                 }
 621                 if (mm->binfmt)
 622                         module_put(mm->binfmt->module);
 623                 mmdrop(mm);
 624         }
 625 }
 626 EXPORT_SYMBOL_GPL(mmput);
 627
 628 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 629 {
 630         if (new_exe_file)
 631                 get_file(new_exe_file);
 632         if (mm->exe_file)
 633                 fput(mm->exe_file);
 634         mm->exe_file = new_exe_file;
 635 }
 636
 637 struct file *get_mm_exe_file(struct mm_struct *mm)
 638 {
 639         struct file *exe_file;
 640
 641         /* We need mmap_sem to protect against races with removal of exe_file */
 642         down_read(&mm->mmap_sem);
 643         exe_file = mm->exe_file;
 644         if (exe_file)
 645                 get_file(exe_file);
 646         up_read(&mm->mmap_sem);
 647         return exe_file;
 648 }
 649
 650 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
 651 {
 652         /* It's safe to write the exe_file pointer without exe_file_lock because
 653          * this is called during fork when the task is not yet in /proc */
 654         newmm->exe_file = get_mm_exe_file(oldmm);
 655 }
 656
 657 /**
 658  * get_task_mm - acquire a reference to the task's mm
 659  *
 660  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 661  * this kernel workthread has transiently adopted a user mm with use_mm,
 662  * to do its AIO) is not set and if so returns a reference to it, after
 663  * bumping up the use count.  User must release the mm via mmput()
 664  * after use.  Typically used by /proc and ptrace.
 665  */
 666 struct mm_struct *get_task_mm(struct task_struct *task)
 667 {
 668         struct mm_struct *mm;
 669
 670         task_lock(task);
 671         mm = task->mm;
 672         if (mm) {
 673                 if (task->flags & PF_KTHREAD)
 674                         mm = NULL;
 675                 else
 676                         atomic_inc(&mm->mm_users);
 677         }
 678         task_unlock(task);
 679         return mm;
 680 }
 681 EXPORT_SYMBOL_GPL(get_task_mm);
 682
 683 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 684 {
 685         struct mm_struct *mm;
 686         int err;
 687
 688         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 689         if (err)
 690                 return ERR_PTR(err);
 691
 692         mm = get_task_mm(task);
 693         if (mm && mm != current->mm &&
 694                         !ptrace_may_access(task, mode)) {
 695                 mmput(mm);
 696                 mm = ERR_PTR(-EACCES);
 697         }
 698         mutex_unlock(&task->signal->cred_guard_mutex);
 699
 700         return mm;
 701 }
 702
 703 static void complete_vfork_done(struct task_struct *tsk)
 704 {
 705         struct completion *vfork;
 706
 707         task_lock(tsk);
 708         vfork = tsk->vfork_done;
 709         if (likely(vfork)) {
 710                 tsk->vfork_done = NULL;
 711                 complete(vfork);
 712         }
 713         task_unlock(tsk);
 714 }
 715
 716 static int wait_for_vfork_done(struct task_struct *child,
 717                                 struct completion *vfork)
 718 {
 719         int killed;
 720
 721         freezer_do_not_count();
 722         killed = wait_for_completion_killable(vfork);
 723         freezer_count();
 724
 725         if (killed) {
 726                 task_lock(child);
 727                 child->vfork_done = NULL;
 728                 task_unlock(child);
 729         }
 730
 731         put_task_struct(child);
 732         return killed;
 733 }
 734
 735 /* Please note the differences between mmput and mm_release.
 736  * mmput is called whenever we stop holding onto a mm_struct,
 737  * error success whatever.
 738  *
 739  * mm_release is called after a mm_struct has been removed
 740  * from the current process.
 741  *
 742  * This difference is important for error handling, when we
 743  * only half set up a mm_struct for a new process and need to restore
 744  * the old one.  Because we mmput the new mm_struct before
 745  * restoring the old one. . .
 746  * Eric Biederman 10 January 1998
 747  */
 748 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 749 {
 750         /* Get rid of any futexes when releasing the mm */
 751 #ifdef CONFIG_FUTEX
 752         if (unlikely(tsk->robust_list)) {
 753                 exit_robust_list(tsk);
 754                 tsk->robust_list = NULL;
 755         }
 756 #ifdef CONFIG_COMPAT
 757         if (unlikely(tsk->compat_robust_list)) {
 758                 compat_exit_robust_list(tsk);
 759                 tsk->compat_robust_list = NULL;
 760         }
 761 #endif
 762         if (unlikely(!list_empty(&tsk->pi_state_list)))
 763                 exit_pi_state_list(tsk);
 764 #endif
 765
 766         uprobe_free_utask(tsk);
 767
 768         /* Get rid of any cached register state */
 769         deactivate_mm(tsk, mm);
 770
 771         /*
 772          * If we're exiting normally, clear a user-space tid field if
 773          * requested.  We leave this alone when dying by signal, to leave
 774          * the value intact in a core dump, and to save the unnecessary
 775          * trouble, say, a killed vfork parent shouldn't touch this mm.
 776          * Userland only wants this done for a sys_exit.
 777          */
 778         if (tsk->clear_child_tid) {
 779                 if (!(tsk->flags & PF_SIGNALED) &&
 780                     atomic_read(&mm->mm_users) > 1) {
 781                         /*
 782                          * We don't check the error code - if userspace has
 783                          * not set up a proper pointer then tough luck.
 784                          */
 785                         put_user(0, tsk->clear_child_tid);
 786                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 787                                         1, NULL, NULL, 0);
 788                 }
 789                 tsk->clear_child_tid = NULL;
 790         }
 791
 792         /*
 793          * All done, finally we can wake up parent and return this mm to him.
 794          * Also kthread_stop() uses this completion for synchronization.
 795          */
 796         if (tsk->vfork_done)
 797                 complete_vfork_done(tsk);
 798 }
 799
 800 /*
 801  * Allocate a new mm structure and copy contents from the
 802  * mm structure of the passed in task structure.
 803  */
 804 static struct mm_struct *dup_mm(struct task_struct *tsk)
 805 {
 806         struct mm_struct *mm, *oldmm = current->mm;
 807         int err;
 808
 809         mm = allocate_mm();
 810         if (!mm)
 811                 goto fail_nomem;
 812
 813         memcpy(mm, oldmm, sizeof(*mm));
 814         mm_init_cpumask(mm);
 815
 816 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 817         mm->pmd_huge_pte = NULL;
 818 #endif
 819         if (!mm_init(mm, tsk))
 820                 goto fail_nomem;
 821
 822         if (init_new_context(tsk, mm))
 823                 goto fail_nocontext;
 824
 825         dup_mm_exe_file(oldmm, mm);
 826
 827         err = dup_mmap(mm, oldmm);
 828         if (err)
 829                 goto free_pt;
 830
 831         mm->hiwater_rss = get_mm_rss(mm);
 832         mm->hiwater_vm = mm->total_vm;
 833
 834         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 835                 goto free_pt;
 836
 837         return mm;
 838
 839 free_pt:
 840         /* don't put binfmt in mmput, we haven't got module yet */
 841         mm->binfmt = NULL;
 842         mmput(mm);
 843
 844 fail_nomem:
 845         return NULL;
 846
 847 fail_nocontext:
 848         /*
 849          * If init_new_context() failed, we cannot use mmput() to free the mm
 850          * because it calls destroy_context()
 851          */
 852         mm_free_pgd(mm);
 853         free_mm(mm);
 854         return NULL;
 855 }
 856
 857 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 858 {
 859         struct mm_struct *mm, *oldmm;
 860         int retval;
 861
 862         tsk->min_flt = tsk->maj_flt = 0;
 863         tsk->nvcsw = tsk->nivcsw = 0;
 864 #ifdef CONFIG_DETECT_HUNG_TASK
 865         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 866 #endif
 867
 868         tsk->mm = NULL;
 869         tsk->active_mm = NULL;
 870
 871         /*
 872          * Are we cloning a kernel thread?
 873          *
 874          * We need to steal a active VM for that..
 875          */
 876         oldmm = current->mm;
 877         if (!oldmm)
 878                 return 0;
 879
 880         if (clone_flags & CLONE_VM) {
 881                 atomic_inc(&oldmm->mm_users);
 882                 mm = oldmm;
 883                 goto good_mm;
 884         }
 885
 886         retval = -ENOMEM;
 887         mm = dup_mm(tsk);
 888         if (!mm)
 889                 goto fail_nomem;
 890
 891 good_mm:
 892         tsk->mm = mm;
 893         tsk->active_mm = mm;
 894         return 0;
 895
 896 fail_nomem:
 897         return retval;
 898 }
 899
 900 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 901 {
 902         struct fs_struct *fs = current->fs;
 903         if (clone_flags & CLONE_FS) {
 904                 /* tsk->fs is already what we want */
 905                 spin_lock(&fs->lock);
 906                 if (fs->in_exec) {
 907                         spin_unlock(&fs->lock);
 908                         return -EAGAIN;
 909                 }
 910                 fs->users++;
 911                 spin_unlock(&fs->lock);
 912                 return 0;
 913         }
 914         tsk->fs = copy_fs_struct(fs);
 915         if (!tsk->fs)
 916                 return -ENOMEM;
 917         return 0;
 918 }
 919
 920 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
 921 {
 922         struct files_struct *oldf, *newf;
 923         int error = 0;
 924
 925         /*
 926          * A background process may not have any files ...
 927          */
 928         oldf = current->files;
 929         if (!oldf)
 930                 goto out;
 931
 932         if (clone_flags & CLONE_FILES) {
 933                 atomic_inc(&oldf->count);
 934                 goto out;
 935         }
 936
 937         newf = dup_fd(oldf, &error);
 938         if (!newf)
 939                 goto out;
 940
 941         tsk->files = newf;
 942         error = 0;
 943 out:
 944         return error;
 945 }
 946
 947 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 948 {
 949 #ifdef CONFIG_BLOCK
 950         struct io_context *ioc = current->io_context;
 951         struct io_context *new_ioc;
 952
 953         if (!ioc)
 954                 return 0;
 955         /*
 956          * Share io context with parent, if CLONE_IO is set
 957          */
 958         if (clone_flags & CLONE_IO) {
 959                 ioc_task_link(ioc);
 960                 tsk->io_context = ioc;
 961         } else if (ioprio_valid(ioc->ioprio)) {
 962                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
 963                 if (unlikely(!new_ioc))
 964                         return -ENOMEM;
 965
 966                 new_ioc->ioprio = ioc->ioprio;
 967                 put_io_context(new_ioc);
 968         }
 969 #endif
 970         return 0;
 971 }
 972
 973 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 974 {
 975         struct sighand_struct *sig;
 976
 977         if (clone_flags & CLONE_SIGHAND) {
 978                 atomic_inc(&current->sighand->count);
 979                 return 0;
 980         }
 981         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 982         rcu_assign_pointer(tsk->sighand, sig);
 983         if (!sig)
 984                 return -ENOMEM;
 985         atomic_set(&sig->count, 1);
 986         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 987         return 0;
 988 }
 989
 990 void __cleanup_sighand(struct sighand_struct *sighand)
 991 {
 992         if (atomic_dec_and_test(&sighand->count)) {
 993                 signalfd_cleanup(sighand);
 994                 kmem_cache_free(sighand_cachep, sighand);
 995         }
 996 }
 997
 998
 999 /*
1000  * Initialize POSIX timer handling for a thread group.
1001  */
1002 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1003 {
1004         unsigned long cpu_limit;
1005
1006         /* Thread group counters. */
1007         thread_group_cputime_init(sig);
1008
1009         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1010         if (cpu_limit != RLIM_INFINITY) {
1011                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1012                 sig->cputimer.running = 1;
1013         }
1014
1015         /* The timer lists. */
1016         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1017         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1018         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1019 }
1020
1021 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1022 {
1023         struct signal_struct *sig;
1024
1025         if (clone_flags & CLONE_THREAD)
1026                 return 0;
1027
1028         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1029         tsk->signal = sig;
1030         if (!sig)
1031                 return -ENOMEM;
1032
1033         sig->nr_threads = 1;
1034         atomic_set(&sig->live, 1);
1035         atomic_set(&sig->sigcnt, 1);
1036
1037         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1038         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1039         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1040
1041         init_waitqueue_head(&sig->wait_chldexit);
1042         sig->curr_target = tsk;
1043         init_sigpending(&sig->shared_pending);
1044         INIT_LIST_HEAD(&sig->posix_timers);
1045
1046         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1047         sig->real_timer.function = it_real_fn;
1048
1049         task_lock(current->group_leader);
1050         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1051         task_unlock(current->group_leader);
1052
1053         posix_cpu_timers_init_group(sig);
1054
1055         tty_audit_fork(sig);
1056         sched_autogroup_fork(sig);
1057
1058 #ifdef CONFIG_CGROUPS
1059         init_rwsem(&sig->group_rwsem);
1060 #endif
1061
1062         sig->oom_score_adj = current->signal->oom_score_adj;
1063         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1064
1065         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1066                                    current->signal->is_child_subreaper;
1067
1068         mutex_init(&sig->cred_guard_mutex);
1069
1070         return 0;
1071 }
1072
1073 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
1074 {
1075         unsigned long new_flags = p->flags;
1076
1077         new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1078         new_flags |= PF_FORKNOEXEC;
1079         p->flags = new_flags;
1080 }
1081
1082 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1083 {
1084         current->clear_child_tid = tidptr;
1085
1086         return task_pid_vnr(current);
1087 }
1088
1089 static void rt_mutex_init_task(struct task_struct *p)
1090 {
1091         raw_spin_lock_init(&p->pi_lock);
1092 #ifdef CONFIG_RT_MUTEXES
1093         p->pi_waiters = RB_ROOT;
1094         p->pi_waiters_leftmost = NULL;
1095         p->pi_blocked_on = NULL;
1096         p->pi_top_task = NULL;
1097 #endif
1098 }
1099
1100 #ifdef CONFIG_MM_OWNER
1101 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1102 {
1103         mm->owner = p;
1104 }
1105 #endif /* CONFIG_MM_OWNER */
1106
1107 /*
1108  * Initialize POSIX timer handling for a single task.
1109  */
1110 static void posix_cpu_timers_init(struct task_struct *tsk)
1111 {
1112         tsk->cputime_expires.prof_exp = 0;
1113         tsk->cputime_expires.virt_exp = 0;
1114         tsk->cputime_expires.sched_exp = 0;
1115         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1116         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1117         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1118 }
1119
1120 static inline void
1121 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1122 {
1123          task->pids[type].pid = pid;
1124 }
1125
1126 /*
1127  * This creates a new process as a copy of the old one,
1128  * but does not actually start it yet.
1129  *
1130  * It copies the registers, and all the appropriate
1131  * parts of the process environment (as per the clone
1132  * flags). The actual kick-off is left to the caller.
1133  */
1134 static struct task_struct *copy_process(unsigned long clone_flags,
1135                                         unsigned long stack_start,
1136                                         unsigned long stack_size,
1137                                         int __user *child_tidptr,
1138                                         struct pid *pid,
1139                                         int trace)
1140 {
1141         int retval;
1142         struct task_struct *p;
1143
1144         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1145                 return ERR_PTR(-EINVAL);
1146
1147         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1148                 return ERR_PTR(-EINVAL);
1149
1150         /*
1151          * Thread groups must share signals as well, and detached threads
1152          * can only be started up within the thread group.
1153          */
1154         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1155                 return ERR_PTR(-EINVAL);
1156
1157         /*
1158          * Shared signal handlers imply shared VM. By way of the above,
1159          * thread groups also imply shared VM. Blocking this case allows
1160          * for various simplifications in other code.
1161          */
1162         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1163                 return ERR_PTR(-EINVAL);
1164
1165         /*
1166          * Siblings of global init remain as zombies on exit since they are
1167          * not reaped by their parent (swapper). To solve this and to avoid
1168          * multi-rooted process trees, prevent global and container-inits
1169          * from creating siblings.
1170          */
1171         if ((clone_flags & CLONE_PARENT) &&
1172                                 current->signal->flags & SIGNAL_UNKILLABLE)
1173                 return ERR_PTR(-EINVAL);
1174
1175         /*
1176          * If the new process will be in a different pid or user namespace
1177          * do not allow it to share a thread group or signal handlers or
1178          * parent with the forking task.
1179          */
1180         if (clone_flags & CLONE_SIGHAND) {
1181                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1182                     (task_active_pid_ns(current) !=
1183                                 current->nsproxy->pid_ns_for_children))
1184                         return ERR_PTR(-EINVAL);
1185         }
1186
1187         retval = security_task_create(clone_flags);
1188         if (retval)
1189                 goto fork_out;
1190
1191         retval = -ENOMEM;
1192         p = dup_task_struct(current);
1193         if (!p)
1194                 goto fork_out;
1195
1196         ftrace_graph_init_task(p);
1197         get_seccomp_filter(p);
1198
1199         rt_mutex_init_task(p);
1200
1201 #ifdef CONFIG_PROVE_LOCKING
1202         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1203         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1204 #endif
1205         retval = -EAGAIN;
1206         if (atomic_read(&p->real_cred->user->processes) >=
1207                         task_rlimit(p, RLIMIT_NPROC)) {
1208                 if (p->real_cred->user != INIT_USER &&
1209                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1210                         goto bad_fork_free;
1211         }
1212         current->flags &= ~PF_NPROC_EXCEEDED;
1213
1214         retval = copy_creds(p, clone_flags);
1215         if (retval < 0)
1216                 goto bad_fork_free;
1217
1218         /*
1219          * If multiple threads are within copy_process(), then this check
1220          * triggers too late. This doesn't hurt, the check is only there
1221          * to stop root fork bombs.
1222          */
1223         retval = -EAGAIN;
1224         if (nr_threads >= max_threads)
1225                 goto bad_fork_cleanup_count;
1226
1227         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1228                 goto bad_fork_cleanup_count;
1229
1230         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1231         copy_flags(clone_flags, p);
1232         INIT_LIST_HEAD(&p->children);
1233         INIT_LIST_HEAD(&p->sibling);
1234         rcu_copy_process(p);
1235         p->vfork_done = NULL;
1236         spin_lock_init(&p->alloc_lock);
1237
1238         init_sigpending(&p->pending);
1239
1240         p->utime = p->stime = p->gtime = 0;
1241         p->utimescaled = p->stimescaled = 0;
1242 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
1243         p->prev_cputime.utime = p->prev_cputime.stime = 0;
1244 #endif
1245 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1246         seqlock_init(&p->vtime_seqlock);
1247         p->vtime_snap = 0;
1248         p->vtime_snap_whence = VTIME_SLEEPING;
1249 #endif
1250
1251 #if defined(SPLIT_RSS_COUNTING)
1252         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1253 #endif
1254
1255         p->default_timer_slack_ns = current->timer_slack_ns;
1256
1257         task_io_accounting_init(&p->ioac);
1258         acct_clear_integrals(p);
1259
1260         posix_cpu_timers_init(p);
1261
1262         do_posix_clock_monotonic_gettime(&p->start_time);
1263         p->real_start_time = p->start_time;
1264         monotonic_to_bootbased(&p->real_start_time);
1265         p->io_context = NULL;
1266         p->audit_context = NULL;
1267         if (clone_flags & CLONE_THREAD)
1268                 threadgroup_change_begin(current);
1269         cgroup_fork(p);
1270 #ifdef CONFIG_NUMA
1271         p->mempolicy = mpol_dup(p->mempolicy);
1272         if (IS_ERR(p->mempolicy)) {
1273                 retval = PTR_ERR(p->mempolicy);
1274                 p->mempolicy = NULL;
1275                 goto bad_fork_cleanup_cgroup;
1276         }
1277         mpol_fix_fork_child_flag(p);
1278 #endif
1279 #ifdef CONFIG_CPUSETS
1280         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1281         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1282         seqcount_init(&p->mems_allowed_seq);
1283 #endif
1284 #ifdef CONFIG_TRACE_IRQFLAGS
1285         p->irq_events = 0;
1286         p->hardirqs_enabled = 0;
1287         p->hardirq_enable_ip = 0;
1288         p->hardirq_enable_event = 0;
1289         p->hardirq_disable_ip = _THIS_IP_;
1290         p->hardirq_disable_event = 0;
1291         p->softirqs_enabled = 1;
1292         p->softirq_enable_ip = _THIS_IP_;
1293         p->softirq_enable_event = 0;
1294         p->softirq_disable_ip = 0;
1295         p->softirq_disable_event = 0;
1296         p->hardirq_context = 0;
1297         p->softirq_context = 0;
1298 #endif
1299 #ifdef CONFIG_LOCKDEP
1300         p->lockdep_depth = 0; /* no locks held yet */
1301         p->curr_chain_key = 0;
1302         p->lockdep_recursion = 0;
1303 #endif
1304
1305 #ifdef CONFIG_DEBUG_MUTEXES
1306         p->blocked_on = NULL; /* not blocked yet */
1307 #endif
1308 #ifdef CONFIG_MEMCG
1309         p->memcg_batch.do_batch = 0;
1310         p->memcg_batch.memcg = NULL;
1311 #endif
1312 #ifdef CONFIG_BCACHE
1313         p->sequential_io        = 0;
1314         p->sequential_io_avg    = 0;
1315 #endif
1316
1317         /* Perform scheduler related setup. Assign this task to a CPU. */
1318         retval = sched_fork(clone_flags, p);
1319         if (retval)
1320                 goto bad_fork_cleanup_policy;
1321
1322         retval = perf_event_init_task(p);
1323         if (retval)
1324                 goto bad_fork_cleanup_policy;
1325         retval = audit_alloc(p);
1326         if (retval)
1327                 goto bad_fork_cleanup_policy;
1328         /* copy all the process information */
1329         retval = copy_semundo(clone_flags, p);
1330         if (retval)
1331                 goto bad_fork_cleanup_audit;
1332         retval = copy_files(clone_flags, p);
1333         if (retval)
1334                 goto bad_fork_cleanup_semundo;
1335         retval = copy_fs(clone_flags, p);
1336         if (retval)
1337                 goto bad_fork_cleanup_files;
1338         retval = copy_sighand(clone_flags, p);
1339         if (retval)
1340                 goto bad_fork_cleanup_fs;
1341         retval = copy_signal(clone_flags, p);
1342         if (retval)
1343                 goto bad_fork_cleanup_sighand;
1344         retval = copy_mm(clone_flags, p);
1345         if (retval)
1346                 goto bad_fork_cleanup_signal;
1347         retval = copy_namespaces(clone_flags, p);
1348         if (retval)
1349                 goto bad_fork_cleanup_mm;
1350         retval = copy_io(clone_flags, p);
1351         if (retval)
1352                 goto bad_fork_cleanup_namespaces;
1353         retval = copy_thread(clone_flags, stack_start, stack_size, p);
1354         if (retval)
1355                 goto bad_fork_cleanup_io;
1356
1357         if (pid != &init_struct_pid) {
1358                 retval = -ENOMEM;
1359                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1360                 if (!pid)
1361                         goto bad_fork_cleanup_io;
1362         }
1363
1364         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1365         /*
1366          * Clear TID on mm_release()?
1367          */
1368         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1369 #ifdef CONFIG_BLOCK
1370         p->plug = NULL;
1371 #endif
1372 #ifdef CONFIG_FUTEX
1373         p->robust_list = NULL;
1374 #ifdef CONFIG_COMPAT
1375         p->compat_robust_list = NULL;
1376 #endif
1377         INIT_LIST_HEAD(&p->pi_state_list);
1378         p->pi_state_cache = NULL;
1379 #endif
1380         /*
1381          * sigaltstack should be cleared when sharing the same VM
1382          */
1383         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1384                 p->sas_ss_sp = p->sas_ss_size = 0;
1385
1386         /*
1387          * Syscall tracing and stepping should be turned off in the
1388          * child regardless of CLONE_PTRACE.
1389          */
1390         user_disable_single_step(p);
1391         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1392 #ifdef TIF_SYSCALL_EMU
1393         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1394 #endif
1395         clear_all_latency_tracing(p);
1396
1397         /* ok, now we should be set up.. */
1398         p->pid = pid_nr(pid);
1399         if (clone_flags & CLONE_THREAD) {
1400                 p->exit_signal = -1;
1401                 p->group_leader = current->group_leader;
1402                 p->tgid = current->tgid;
1403         } else {
1404                 if (clone_flags & CLONE_PARENT)
1405                         p->exit_signal = current->group_leader->exit_signal;
1406                 else
1407                         p->exit_signal = (clone_flags & CSIGNAL);
1408                 p->group_leader = p;
1409                 p->tgid = p->pid;
1410         }
1411
1412         p->nr_dirtied = 0;
1413         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1414         p->dirty_paused_when = 0;
1415
1416         p->pdeath_signal = 0;
1417         INIT_LIST_HEAD(&p->thread_group);
1418         p->task_works = NULL;
1419
1420         /*
1421          * Make it visible to the rest of the system, but dont wake it up yet.
1422          * Need tasklist lock for parent etc handling!
1423          */
1424         write_lock_irq(&tasklist_lock);
1425
1426         /* CLONE_PARENT re-uses the old parent */
1427         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1428                 p->real_parent = current->real_parent;
1429                 p->parent_exec_id = current->parent_exec_id;
1430         } else {
1431                 p->real_parent = current;
1432                 p->parent_exec_id = current->self_exec_id;
1433         }
1434
1435         spin_lock(&current->sighand->siglock);
1436
1437         /*
1438          * Process group and session signals need to be delivered to just the
1439          * parent before the fork or both the parent and the child after the
1440          * fork. Restart if a signal comes in before we add the new process to
1441          * it's process group.
1442          * A fatal signal pending means that current will exit, so the new
1443          * thread can't slip out of an OOM kill (or normal SIGKILL).
1444         */
1445         recalc_sigpending();
1446         if (signal_pending(current)) {
1447                 spin_unlock(&current->sighand->siglock);
1448                 write_unlock_irq(&tasklist_lock);
1449                 retval = -ERESTARTNOINTR;
1450                 goto bad_fork_free_pid;
1451         }
1452
1453         if (likely(p->pid)) {
1454                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1455
1456                 init_task_pid(p, PIDTYPE_PID, pid);
1457                 if (thread_group_leader(p)) {
1458                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1459                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1460
1461                         if (is_child_reaper(pid)) {
1462                                 ns_of_pid(pid)->child_reaper = p;
1463                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1464                         }
1465
1466                         p->signal->leader_pid = pid;
1467                         p->signal->tty = tty_kref_get(current->signal->tty);
1468                         list_add_tail(&p->sibling, &p->real_parent->children);
1469                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1470                         attach_pid(p, PIDTYPE_PGID);
1471                         attach_pid(p, PIDTYPE_SID);
1472                         __this_cpu_inc(process_counts);
1473                 } else {
1474                         current->signal->nr_threads++;
1475                         atomic_inc(&current->signal->live);
1476                         atomic_inc(&current->signal->sigcnt);
1477                         list_add_tail_rcu(&p->thread_group,
1478                                           &p->group_leader->thread_group);
1479                         list_add_tail_rcu(&p->thread_node,
1480                                           &p->signal->thread_head);
1481                 }
1482                 attach_pid(p, PIDTYPE_PID);
1483                 nr_threads++;
1484         }
1485
1486         total_forks++;
1487         spin_unlock(&current->sighand->siglock);
1488         write_unlock_irq(&tasklist_lock);
1489         proc_fork_connector(p);
1490         cgroup_post_fork(p);
1491         if (clone_flags & CLONE_THREAD)
1492                 threadgroup_change_end(current);
1493         perf_event_fork(p);
1494
1495         trace_task_newtask(p, clone_flags);
1496         uprobe_copy_process(p, clone_flags);
1497
1498         return p;
1499
1500 bad_fork_free_pid:
1501         if (pid != &init_struct_pid)
1502                 free_pid(pid);
1503 bad_fork_cleanup_io:
1504         if (p->io_context)
1505                 exit_io_context(p);
1506 bad_fork_cleanup_namespaces:
1507         exit_task_namespaces(p);
1508 bad_fork_cleanup_mm:
1509         if (p->mm)
1510                 mmput(p->mm);
1511 bad_fork_cleanup_signal:
1512         if (!(clone_flags & CLONE_THREAD))
1513                 free_signal_struct(p->signal);
1514 bad_fork_cleanup_sighand:
1515         __cleanup_sighand(p->sighand);
1516 bad_fork_cleanup_fs:
1517         exit_fs(p); /* blocking */
1518 bad_fork_cleanup_files:
1519         exit_files(p); /* blocking */
1520 bad_fork_cleanup_semundo:
1521         exit_sem(p);
1522 bad_fork_cleanup_audit:
1523         audit_free(p);
1524 bad_fork_cleanup_policy:
1525         perf_event_free_task(p);
1526 #ifdef CONFIG_NUMA
1527         mpol_put(p->mempolicy);
1528 bad_fork_cleanup_cgroup:
1529 #endif
1530         if (clone_flags & CLONE_THREAD)
1531                 threadgroup_change_end(current);
1532         cgroup_exit(p, 0);
1533         delayacct_tsk_free(p);
1534         module_put(task_thread_info(p)->exec_domain->module);
1535 bad_fork_cleanup_count:
1536         atomic_dec(&p->cred->user->processes);
1537         exit_creds(p);
1538 bad_fork_free:
1539         free_task(p);
1540 fork_out:
1541         return ERR_PTR(retval);
1542 }
1543
1544 static inline void init_idle_pids(struct pid_link *links)
1545 {
1546         enum pid_type type;
1547
1548         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1549                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1550                 links[type].pid = &init_struct_pid;
1551         }
1552 }
1553
1554 struct task_struct *fork_idle(int cpu)
1555 {
1556         struct task_struct *task;
1557         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0);
1558         if (!IS_ERR(task)) {
1559                 init_idle_pids(task->pids);
1560                 init_idle(task, cpu);
1561         }
1562
1563         return task;
1564 }
1565
1566 /*
1567  *  Ok, this is the main fork-routine.
1568  *
1569  * It copies the process, and if successful kick-starts
1570  * it and waits for it to finish using the VM if required.
1571  */
1572 long do_fork(unsigned long clone_flags,
1573               unsigned long stack_start,
1574               unsigned long stack_size,
1575               int __user *parent_tidptr,
1576               int __user *child_tidptr)
1577 {
1578         struct task_struct *p;
1579         int trace = 0;
1580         long nr;
1581
1582         /*
1583          * Determine whether and which event to report to ptracer.  When
1584          * called from kernel_thread or CLONE_UNTRACED is explicitly
1585          * requested, no event is reported; otherwise, report if the event
1586          * for the type of forking is enabled.
1587          */
1588         if (!(clone_flags & CLONE_UNTRACED)) {
1589                 if (clone_flags & CLONE_VFORK)
1590                         trace = PTRACE_EVENT_VFORK;
1591                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1592                         trace = PTRACE_EVENT_CLONE;
1593                 else
1594                         trace = PTRACE_EVENT_FORK;
1595
1596                 if (likely(!ptrace_event_enabled(current, trace)))
1597                         trace = 0;
1598         }
1599
1600         p = copy_process(clone_flags, stack_start, stack_size,
1601                          child_tidptr, NULL, trace);
1602         /*
1603          * Do this prior waking up the new thread - the thread pointer
1604          * might get invalid after that point, if the thread exits quickly.
1605          */
1606         if (!IS_ERR(p)) {
1607                 struct completion vfork;
1608
1609                 trace_sched_process_fork(current, p);
1610
1611                 nr = task_pid_vnr(p);
1612
1613                 if (clone_flags & CLONE_PARENT_SETTID)
1614                         put_user(nr, parent_tidptr);
1615
1616                 if (clone_flags & CLONE_VFORK) {
1617                         p->vfork_done = &vfork;
1618                         init_completion(&vfork);
1619                         get_task_struct(p);
1620                 }
1621
1622                 wake_up_new_task(p);
1623
1624                 /* forking complete and child started to run, tell ptracer */
1625                 if (unlikely(trace))
1626                         ptrace_event(trace, nr);
1627
1628                 if (clone_flags & CLONE_VFORK) {
1629                         if (!wait_for_vfork_done(p, &vfork))
1630                                 ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1631                 }
1632         } else {
1633                 nr = PTR_ERR(p);
1634         }
1635         return nr;
1636 }
1637
1638 /*
1639  * Create a kernel thread.
1640  */
1641 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1642 {
1643         return do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1644                 (unsigned long)arg, NULL, NULL);
1645 }
1646
1647 #ifdef __ARCH_WANT_SYS_FORK
1648 SYSCALL_DEFINE0(fork)
1649 {
1650 #ifdef CONFIG_MMU
1651         return do_fork(SIGCHLD, 0, 0, NULL, NULL);
1652 #else
1653         /* can not support in nommu mode */
1654         return -EINVAL;
1655 #endif
1656 }
1657 #endif
1658
1659 #ifdef __ARCH_WANT_SYS_VFORK
1660 SYSCALL_DEFINE0(vfork)
1661 {
1662         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1663                         0, NULL, NULL);
1664 }
1665 #endif
1666
1667 #ifdef __ARCH_WANT_SYS_CLONE
1668 #ifdef CONFIG_CLONE_BACKWARDS
1669 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1670                  int __user *, parent_tidptr,
1671                  int, tls_val,
1672                  int __user *, child_tidptr)
1673 #elif defined(CONFIG_CLONE_BACKWARDS2)
1674 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1675                  int __user *, parent_tidptr,
1676                  int __user *, child_tidptr,
1677                  int, tls_val)
1678 #elif defined(CONFIG_CLONE_BACKWARDS3)
1679 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1680                 int, stack_size,
1681                 int __user *, parent_tidptr,
1682                 int __user *, child_tidptr,
1683                 int, tls_val)
1684 #else
1685 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1686                  int __user *, parent_tidptr,
1687                  int __user *, child_tidptr,
1688                  int, tls_val)
1689 #endif
1690 {
1691         return do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr);
1692 }
1693 #endif
1694
1695 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1696 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1697 #endif
1698
1699 static void sighand_ctor(void *data)
1700 {
1701         struct sighand_struct *sighand = data;
1702
1703         spin_lock_init(&sighand->siglock);
1704         init_waitqueue_head(&sighand->signalfd_wqh);
1705 }
1706
1707 void __init proc_caches_init(void)
1708 {
1709         sighand_cachep = kmem_cache_create("sighand_cache",
1710                         sizeof(struct sighand_struct), 0,
1711                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1712                         SLAB_NOTRACK, sighand_ctor);
1713         signal_cachep = kmem_cache_create("signal_cache",
1714                         sizeof(struct signal_struct), 0,
1715                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1716         files_cachep = kmem_cache_create("files_cache",
1717                         sizeof(struct files_struct), 0,
1718                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1719         fs_cachep = kmem_cache_create("fs_cache",
1720                         sizeof(struct fs_struct), 0,
1721                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1722         /*
1723          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1724          * whole struct cpumask for the OFFSTACK case. We could change
1725          * this to *only* allocate as much of it as required by the
1726          * maximum number of CPU's we can ever have.  The cpumask_allocation
1727          * is at the end of the structure, exactly for that reason.
1728          */
1729         mm_cachep = kmem_cache_create("mm_struct",
1730                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1731                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1732         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1733         mmap_init();
1734         nsproxy_cache_init();
1735 }
1736
1737 /*
1738  * Check constraints on flags passed to the unshare system call.
1739  */
1740 static int check_unshare_flags(unsigned long unshare_flags)
1741 {
1742         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1743                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1744                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1745                                 CLONE_NEWUSER|CLONE_NEWPID))
1746                 return -EINVAL;
1747         /*
1748          * Not implemented, but pretend it works if there is nothing to
1749          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1750          * needs to unshare vm.
1751          */
1752         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1753                 /* FIXME: get_task_mm() increments ->mm_users */
1754                 if (atomic_read(&current->mm->mm_users) > 1)
1755                         return -EINVAL;
1756         }
1757
1758         return 0;
1759 }
1760
1761 /*
1762  * Unshare the filesystem structure if it is being shared
1763  */
1764 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1765 {
1766         struct fs_struct *fs = current->fs;
1767
1768         if (!(unshare_flags & CLONE_FS) || !fs)
1769                 return 0;
1770
1771         /* don't need lock here; in the worst case we'll do useless copy */
1772         if (fs->users == 1)
1773                 return 0;
1774
1775         *new_fsp = copy_fs_struct(fs);
1776         if (!*new_fsp)
1777                 return -ENOMEM;
1778
1779         return 0;
1780 }
1781
1782 /*
1783  * Unshare file descriptor table if it is being shared
1784  */
1785 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1786 {
1787         struct files_struct *fd = current->files;
1788         int error = 0;
1789
1790         if ((unshare_flags & CLONE_FILES) &&
1791             (fd && atomic_read(&fd->count) > 1)) {
1792                 *new_fdp = dup_fd(fd, &error);
1793                 if (!*new_fdp)
1794                         return error;
1795         }
1796
1797         return 0;
1798 }
1799
1800 /*
1801  * unshare allows a process to 'unshare' part of the process
1802  * context which was originally shared using clone.  copy_*
1803  * functions used by do_fork() cannot be used here directly
1804  * because they modify an inactive task_struct that is being
1805  * constructed. Here we are modifying the current, active,
1806  * task_struct.
1807  */
1808 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1809 {
1810         struct fs_struct *fs, *new_fs = NULL;
1811         struct files_struct *fd, *new_fd = NULL;
1812         struct cred *new_cred = NULL;
1813         struct nsproxy *new_nsproxy = NULL;
1814         int do_sysvsem = 0;
1815         int err;
1816
1817         /*
1818          * If unsharing a user namespace must also unshare the thread.
1819          */
1820         if (unshare_flags & CLONE_NEWUSER)
1821                 unshare_flags |= CLONE_THREAD | CLONE_FS;
1822         /*
1823          * If unsharing a thread from a thread group, must also unshare vm.
1824          */
1825         if (unshare_flags & CLONE_THREAD)
1826                 unshare_flags |= CLONE_VM;
1827         /*
1828          * If unsharing vm, must also unshare signal handlers.
1829          */
1830         if (unshare_flags & CLONE_VM)
1831                 unshare_flags |= CLONE_SIGHAND;
1832         /*
1833          * If unsharing namespace, must also unshare filesystem information.
1834          */
1835         if (unshare_flags & CLONE_NEWNS)
1836                 unshare_flags |= CLONE_FS;
1837
1838         err = check_unshare_flags(unshare_flags);
1839         if (err)
1840                 goto bad_unshare_out;
1841         /*
1842          * CLONE_NEWIPC must also detach from the undolist: after switching
1843          * to a new ipc namespace, the semaphore arrays from the old
1844          * namespace are unreachable.
1845          */
1846         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1847                 do_sysvsem = 1;
1848         err = unshare_fs(unshare_flags, &new_fs);
1849         if (err)
1850                 goto bad_unshare_out;
1851         err = unshare_fd(unshare_flags, &new_fd);
1852         if (err)
1853                 goto bad_unshare_cleanup_fs;
1854         err = unshare_userns(unshare_flags, &new_cred);
1855         if (err)
1856                 goto bad_unshare_cleanup_fd;
1857         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1858                                          new_cred, new_fs);
1859         if (err)
1860                 goto bad_unshare_cleanup_cred;
1861
1862         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
1863                 if (do_sysvsem) {
1864                         /*
1865                          * CLONE_SYSVSEM is equivalent to sys_exit().
1866                          */
1867                         exit_sem(current);
1868                 }
1869
1870                 if (new_nsproxy)
1871                         switch_task_namespaces(current, new_nsproxy);
1872
1873                 task_lock(current);
1874
1875                 if (new_fs) {
1876                         fs = current->fs;
1877                         spin_lock(&fs->lock);
1878                         current->fs = new_fs;
1879                         if (--fs->users)
1880                                 new_fs = NULL;
1881                         else
1882                                 new_fs = fs;
1883                         spin_unlock(&fs->lock);
1884                 }
1885
1886                 if (new_fd) {
1887                         fd = current->files;
1888                         current->files = new_fd;
1889                         new_fd = fd;
1890                 }
1891
1892                 task_unlock(current);
1893
1894                 if (new_cred) {
1895                         /* Install the new user namespace */
1896                         commit_creds(new_cred);
1897                         new_cred = NULL;
1898                 }
1899         }
1900
1901 bad_unshare_cleanup_cred:
1902         if (new_cred)
1903                 put_cred(new_cred);
1904 bad_unshare_cleanup_fd:
1905         if (new_fd)
1906                 put_files_struct(new_fd);
1907
1908 bad_unshare_cleanup_fs:
1909         if (new_fs)
1910                 free_fs_struct(new_fs);
1911
1912 bad_unshare_out:
1913         return err;
1914 }
1915
1916 /*
1917  *      Helper to unshare the files of the current task.
1918  *      We don't want to expose copy_files internals to
1919  *      the exec layer of the kernel.
1920  */
1921
1922 int unshare_files(struct files_struct **displaced)
1923 {
1924         struct task_struct *task = current;
1925         struct files_struct *copy = NULL;
1926         int error;
1927
1928         error = unshare_fd(CLONE_FILES, &copy);
1929         if (error || !copy) {
1930                 *displaced = NULL;
1931                 return error;
1932         }
1933         *displaced = task->files;
1934         task_lock(task);
1935         task->files = copy;
1936         task_unlock(task);
1937         return 0;
1938 }