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