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         struct mempolicy *pol;
 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                 pol = mpol_dup(vma_policy(mpnt));
 404                 retval = PTR_ERR(pol);
 405                 if (IS_ERR(pol))
 406                         goto fail_nomem_policy;
 407                 vma_set_policy(tmp, pol);
 408                 tmp->vm_mm = mm;
 409                 if (anon_vma_fork(tmp, mpnt))
 410                         goto fail_nomem_anon_vma_fork;
 411                 tmp->vm_flags &= ~VM_LOCKED;
 412                 tmp->vm_next = tmp->vm_prev = NULL;
 413                 file = tmp->vm_file;
 414                 if (file) {
 415                         struct inode *inode = file_inode(file);
 416                         struct address_space *mapping = file->f_mapping;
 417
 418                         get_file(file);
 419                         if (tmp->vm_flags & VM_DENYWRITE)
 420                                 atomic_dec(&inode->i_writecount);
 421                         mutex_lock(&mapping->i_mmap_mutex);
 422                         if (tmp->vm_flags & VM_SHARED)
 423                                 mapping->i_mmap_writable++;
 424                         flush_dcache_mmap_lock(mapping);
 425                         /* insert tmp into the share list, just after mpnt */
 426                         if (unlikely(tmp->vm_flags & VM_NONLINEAR))
 427                                 vma_nonlinear_insert(tmp,
 428                                                 &mapping->i_mmap_nonlinear);
 429                         else
 430                                 vma_interval_tree_insert_after(tmp, mpnt,
 431                                                         &mapping->i_mmap);
 432                         flush_dcache_mmap_unlock(mapping);
 433                         mutex_unlock(&mapping->i_mmap_mutex);
 434                 }
 435
 436                 /*
 437                  * Clear hugetlb-related page reserves for children. This only
 438                  * affects MAP_PRIVATE mappings. Faults generated by the child
 439                  * are not guaranteed to succeed, even if read-only
 440                  */
 441                 if (is_vm_hugetlb_page(tmp))
 442                         reset_vma_resv_huge_pages(tmp);
 443
 444                 /*
 445                  * Link in the new vma and copy the page table entries.
 446                  */
 447                 *pprev = tmp;
 448                 pprev = &tmp->vm_next;
 449                 tmp->vm_prev = prev;
 450                 prev = tmp;
 451
 452                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 453                 rb_link = &tmp->vm_rb.rb_right;
 454                 rb_parent = &tmp->vm_rb;
 455
 456                 mm->map_count++;
 457                 retval = copy_page_range(mm, oldmm, mpnt);
 458
 459                 if (tmp->vm_ops && tmp->vm_ops->open)
 460                         tmp->vm_ops->open(tmp);
 461
 462                 if (retval)
 463                         goto out;
 464         }
 465         /* a new mm has just been created */
 466         arch_dup_mmap(oldmm, mm);
 467         retval = 0;
 468 out:
 469         up_write(&mm->mmap_sem);
 470         flush_tlb_mm(oldmm);
 471         up_write(&oldmm->mmap_sem);
 472         uprobe_end_dup_mmap();
 473         return retval;
 474 fail_nomem_anon_vma_fork:
 475         mpol_put(pol);
 476 fail_nomem_policy:
 477         kmem_cache_free(vm_area_cachep, tmp);
 478 fail_nomem:
 479         retval = -ENOMEM;
 480         vm_unacct_memory(charge);
 481         goto out;
 482 }
 483
 484 static inline int mm_alloc_pgd(struct mm_struct *mm)
 485 {
 486         mm->pgd = pgd_alloc(mm);
 487         if (unlikely(!mm->pgd))
 488                 return -ENOMEM;
 489         return 0;
 490 }
 491
 492 static inline void mm_free_pgd(struct mm_struct *mm)
 493 {
 494         pgd_free(mm, mm->pgd);
 495 }
 496 #else
 497 #define dup_mmap(mm, oldmm)     (0)
 498 #define mm_alloc_pgd(mm)        (0)
 499 #define mm_free_pgd(mm)
 500 #endif /* CONFIG_MMU */
 501
 502 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 503
 504 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 505 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 506
 507 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 508
 509 static int __init coredump_filter_setup(char *s)
 510 {
 511         default_dump_filter =
 512                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 513                 MMF_DUMP_FILTER_MASK;
 514         return 1;
 515 }
 516
 517 __setup("coredump_filter=", coredump_filter_setup);
 518
 519 #include <linux/init_task.h>
 520
 521 static void mm_init_aio(struct mm_struct *mm)
 522 {
 523 #ifdef CONFIG_AIO
 524         spin_lock_init(&mm->ioctx_lock);
 525         INIT_HLIST_HEAD(&mm->ioctx_list);
 526 #endif
 527 }
 528
 529 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
 530 {
 531         atomic_set(&mm->mm_users, 1);
 532         atomic_set(&mm->mm_count, 1);
 533         init_rwsem(&mm->mmap_sem);
 534         INIT_LIST_HEAD(&mm->mmlist);
 535         mm->flags = (current->mm) ?
 536                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
 537         mm->core_state = NULL;
 538         mm->nr_ptes = 0;
 539         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 540         spin_lock_init(&mm->page_table_lock);
 541         mm_init_aio(mm);
 542         mm_init_owner(mm, p);
 543
 544         if (likely(!mm_alloc_pgd(mm))) {
 545                 mm->def_flags = 0;
 546                 mmu_notifier_mm_init(mm);
 547                 return mm;
 548         }
 549
 550         free_mm(mm);
 551         return NULL;
 552 }
 553
 554 static void check_mm(struct mm_struct *mm)
 555 {
 556         int i;
 557
 558         for (i = 0; i < NR_MM_COUNTERS; i++) {
 559                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 560
 561                 if (unlikely(x))
 562                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 563                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 564         }
 565
 566 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 567         VM_BUG_ON(mm->pmd_huge_pte);
 568 #endif
 569 }
 570
 571 /*
 572  * Allocate and initialize an mm_struct.
 573  */
 574 struct mm_struct *mm_alloc(void)
 575 {
 576         struct mm_struct *mm;
 577
 578         mm = allocate_mm();
 579         if (!mm)
 580                 return NULL;
 581
 582         memset(mm, 0, sizeof(*mm));
 583         mm_init_cpumask(mm);
 584         return mm_init(mm, current);
 585 }
 586
 587 /*
 588  * Called when the last reference to the mm
 589  * is dropped: either by a lazy thread or by
 590  * mmput. Free the page directory and the mm.
 591  */
 592 void __mmdrop(struct mm_struct *mm)
 593 {
 594         BUG_ON(mm == &init_mm);
 595         mm_free_pgd(mm);
 596         destroy_context(mm);
 597         mmu_notifier_mm_destroy(mm);
 598         check_mm(mm);
 599         free_mm(mm);
 600 }
 601 EXPORT_SYMBOL_GPL(__mmdrop);
 602
 603 /*
 604  * Decrement the use count and release all resources for an mm.
 605  */
 606 void mmput(struct mm_struct *mm)
 607 {
 608         might_sleep();
 609
 610         if (atomic_dec_and_test(&mm->mm_users)) {
 611                 uprobe_clear_state(mm);
 612                 exit_aio(mm);
 613                 ksm_exit(mm);
 614                 khugepaged_exit(mm); /* must run before exit_mmap */
 615                 exit_mmap(mm);
 616                 set_mm_exe_file(mm, NULL);
 617                 if (!list_empty(&mm->mmlist)) {
 618                         spin_lock(&mmlist_lock);
 619                         list_del(&mm->mmlist);
 620                         spin_unlock(&mmlist_lock);
 621                 }
 622                 if (mm->binfmt)
 623                         module_put(mm->binfmt->module);
 624                 mmdrop(mm);
 625         }
 626 }
 627 EXPORT_SYMBOL_GPL(mmput);
 628
 629 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 630 {
 631         if (new_exe_file)
 632                 get_file(new_exe_file);
 633         if (mm->exe_file)
 634                 fput(mm->exe_file);
 635         mm->exe_file = new_exe_file;
 636 }
 637
 638 struct file *get_mm_exe_file(struct mm_struct *mm)
 639 {
 640         struct file *exe_file;
 641
 642         /* We need mmap_sem to protect against races with removal of exe_file */
 643         down_read(&mm->mmap_sem);
 644         exe_file = mm->exe_file;
 645         if (exe_file)
 646                 get_file(exe_file);
 647         up_read(&mm->mmap_sem);
 648         return exe_file;
 649 }
 650
 651 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
 652 {
 653         /* It's safe to write the exe_file pointer without exe_file_lock because
 654          * this is called during fork when the task is not yet in /proc */
 655         newmm->exe_file = get_mm_exe_file(oldmm);
 656 }
 657
 658 /**
 659  * get_task_mm - acquire a reference to the task's mm
 660  *
 661  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 662  * this kernel workthread has transiently adopted a user mm with use_mm,
 663  * to do its AIO) is not set and if so returns a reference to it, after
 664  * bumping up the use count.  User must release the mm via mmput()
 665  * after use.  Typically used by /proc and ptrace.
 666  */
 667 struct mm_struct *get_task_mm(struct task_struct *task)
 668 {
 669         struct mm_struct *mm;
 670
 671         task_lock(task);
 672         mm = task->mm;
 673         if (mm) {
 674                 if (task->flags & PF_KTHREAD)
 675                         mm = NULL;
 676                 else
 677                         atomic_inc(&mm->mm_users);
 678         }
 679         task_unlock(task);
 680         return mm;
 681 }
 682 EXPORT_SYMBOL_GPL(get_task_mm);
 683
 684 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 685 {
 686         struct mm_struct *mm;
 687         int err;
 688
 689         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 690         if (err)
 691                 return ERR_PTR(err);
 692
 693         mm = get_task_mm(task);
 694         if (mm && mm != current->mm &&
 695                         !ptrace_may_access(task, mode)) {
 696                 mmput(mm);
 697                 mm = ERR_PTR(-EACCES);
 698         }
 699         mutex_unlock(&task->signal->cred_guard_mutex);
 700
 701         return mm;
 702 }
 703
 704 static void complete_vfork_done(struct task_struct *tsk)
 705 {
 706         struct completion *vfork;
 707
 708         task_lock(tsk);
 709         vfork = tsk->vfork_done;
 710         if (likely(vfork)) {
 711                 tsk->vfork_done = NULL;
 712                 complete(vfork);
 713         }
 714         task_unlock(tsk);
 715 }
 716
 717 static int wait_for_vfork_done(struct task_struct *child,
 718                                 struct completion *vfork)
 719 {
 720         int killed;
 721
 722         freezer_do_not_count();
 723         killed = wait_for_completion_killable(vfork);
 724         freezer_count();
 725
 726         if (killed) {
 727                 task_lock(child);
 728                 child->vfork_done = NULL;
 729                 task_unlock(child);
 730         }
 731
 732         put_task_struct(child);
 733         return killed;
 734 }
 735
 736 /* Please note the differences between mmput and mm_release.
 737  * mmput is called whenever we stop holding onto a mm_struct,
 738  * error success whatever.
 739  *
 740  * mm_release is called after a mm_struct has been removed
 741  * from the current process.
 742  *
 743  * This difference is important for error handling, when we
 744  * only half set up a mm_struct for a new process and need to restore
 745  * the old one.  Because we mmput the new mm_struct before
 746  * restoring the old one. . .
 747  * Eric Biederman 10 January 1998
 748  */
 749 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 750 {
 751         /* Get rid of any futexes when releasing the mm */
 752 #ifdef CONFIG_FUTEX
 753         if (unlikely(tsk->robust_list)) {
 754                 exit_robust_list(tsk);
 755                 tsk->robust_list = NULL;
 756         }
 757 #ifdef CONFIG_COMPAT
 758         if (unlikely(tsk->compat_robust_list)) {
 759                 compat_exit_robust_list(tsk);
 760                 tsk->compat_robust_list = NULL;
 761         }
 762 #endif
 763         if (unlikely(!list_empty(&tsk->pi_state_list)))
 764                 exit_pi_state_list(tsk);
 765 #endif
 766
 767         uprobe_free_utask(tsk);
 768
 769         /* Get rid of any cached register state */
 770         deactivate_mm(tsk, mm);
 771
 772         /*
 773          * If we're exiting normally, clear a user-space tid field if
 774          * requested.  We leave this alone when dying by signal, to leave
 775          * the value intact in a core dump, and to save the unnecessary
 776          * trouble, say, a killed vfork parent shouldn't touch this mm.
 777          * Userland only wants this done for a sys_exit.
 778          */
 779         if (tsk->clear_child_tid) {
 780                 if (!(tsk->flags & PF_SIGNALED) &&
 781                     atomic_read(&mm->mm_users) > 1) {
 782                         /*
 783                          * We don't check the error code - if userspace has
 784                          * not set up a proper pointer then tough luck.
 785                          */
 786                         put_user(0, tsk->clear_child_tid);
 787                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 788                                         1, NULL, NULL, 0);
 789                 }
 790                 tsk->clear_child_tid = NULL;
 791         }
 792
 793         /*
 794          * All done, finally we can wake up parent and return this mm to him.
 795          * Also kthread_stop() uses this completion for synchronization.
 796          */
 797         if (tsk->vfork_done)
 798                 complete_vfork_done(tsk);
 799 }
 800
 801 /*
 802  * Allocate a new mm structure and copy contents from the
 803  * mm structure of the passed in task structure.
 804  */
 805 struct mm_struct *dup_mm(struct task_struct *tsk)
 806 {
 807         struct mm_struct *mm, *oldmm = current->mm;
 808         int err;
 809
 810         if (!oldmm)
 811                 return NULL;
 812
 813         mm = allocate_mm();
 814         if (!mm)
 815                 goto fail_nomem;
 816
 817         memcpy(mm, oldmm, sizeof(*mm));
 818         mm_init_cpumask(mm);
 819
 820 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 821         mm->pmd_huge_pte = NULL;
 822 #endif
 823 #ifdef CONFIG_NUMA_BALANCING
 824         mm->first_nid = NUMA_PTE_SCAN_INIT;
 825 #endif
 826         if (!mm_init(mm, tsk))
 827                 goto fail_nomem;
 828
 829         if (init_new_context(tsk, mm))
 830                 goto fail_nocontext;
 831
 832         dup_mm_exe_file(oldmm, mm);
 833
 834         err = dup_mmap(mm, oldmm);
 835         if (err)
 836                 goto free_pt;
 837
 838         mm->hiwater_rss = get_mm_rss(mm);
 839         mm->hiwater_vm = mm->total_vm;
 840
 841         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 842                 goto free_pt;
 843
 844         return mm;
 845
 846 free_pt:
 847         /* don't put binfmt in mmput, we haven't got module yet */
 848         mm->binfmt = NULL;
 849         mmput(mm);
 850
 851 fail_nomem:
 852         return NULL;
 853
 854 fail_nocontext:
 855         /*
 856          * If init_new_context() failed, we cannot use mmput() to free the mm
 857          * because it calls destroy_context()
 858          */
 859         mm_free_pgd(mm);
 860         free_mm(mm);
 861         return NULL;
 862 }
 863
 864 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 865 {
 866         struct mm_struct *mm, *oldmm;
 867         int retval;
 868
 869         tsk->min_flt = tsk->maj_flt = 0;
 870         tsk->nvcsw = tsk->nivcsw = 0;
 871 #ifdef CONFIG_DETECT_HUNG_TASK
 872         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 873 #endif
 874
 875         tsk->mm = NULL;
 876         tsk->active_mm = NULL;
 877
 878         /*
 879          * Are we cloning a kernel thread?
 880          *
 881          * We need to steal a active VM for that..
 882          */
 883         oldmm = current->mm;
 884         if (!oldmm)
 885                 return 0;
 886
 887         if (clone_flags & CLONE_VM) {
 888                 atomic_inc(&oldmm->mm_users);
 889                 mm = oldmm;
 890                 goto good_mm;
 891         }
 892
 893         retval = -ENOMEM;
 894         mm = dup_mm(tsk);
 895         if (!mm)
 896                 goto fail_nomem;
 897
 898 good_mm:
 899         tsk->mm = mm;
 900         tsk->active_mm = mm;
 901         return 0;
 902
 903 fail_nomem:
 904         return retval;
 905 }
 906
 907 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 908 {
 909         struct fs_struct *fs = current->fs;
 910         if (clone_flags & CLONE_FS) {
 911                 /* tsk->fs is already what we want */
 912                 spin_lock(&fs->lock);
 913                 if (fs->in_exec) {
 914                         spin_unlock(&fs->lock);
 915                         return -EAGAIN;
 916                 }
 917                 fs->users++;
 918                 spin_unlock(&fs->lock);
 919                 return 0;
 920         }
 921         tsk->fs = copy_fs_struct(fs);
 922         if (!tsk->fs)
 923                 return -ENOMEM;
 924         return 0;
 925 }
 926
 927 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
 928 {
 929         struct files_struct *oldf, *newf;
 930         int error = 0;
 931
 932         /*
 933          * A background process may not have any files ...
 934          */
 935         oldf = current->files;
 936         if (!oldf)
 937                 goto out;
 938
 939         if (clone_flags & CLONE_FILES) {
 940                 atomic_inc(&oldf->count);
 941                 goto out;
 942         }
 943
 944         newf = dup_fd(oldf, &error);
 945         if (!newf)
 946                 goto out;
 947
 948         tsk->files = newf;
 949         error = 0;
 950 out:
 951         return error;
 952 }
 953
 954 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 955 {
 956 #ifdef CONFIG_BLOCK
 957         struct io_context *ioc = current->io_context;
 958         struct io_context *new_ioc;
 959
 960         if (!ioc)
 961                 return 0;
 962         /*
 963          * Share io context with parent, if CLONE_IO is set
 964          */
 965         if (clone_flags & CLONE_IO) {
 966                 ioc_task_link(ioc);
 967                 tsk->io_context = ioc;
 968         } else if (ioprio_valid(ioc->ioprio)) {
 969                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
 970                 if (unlikely(!new_ioc))
 971                         return -ENOMEM;
 972
 973                 new_ioc->ioprio = ioc->ioprio;
 974                 put_io_context(new_ioc);
 975         }
 976 #endif
 977         return 0;
 978 }
 979
 980 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 981 {
 982         struct sighand_struct *sig;
 983
 984         if (clone_flags & CLONE_SIGHAND) {
 985                 atomic_inc(&current->sighand->count);
 986                 return 0;
 987         }
 988         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 989         rcu_assign_pointer(tsk->sighand, sig);
 990         if (!sig)
 991                 return -ENOMEM;
 992         atomic_set(&sig->count, 1);
 993         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 994         return 0;
 995 }
 996
 997 void __cleanup_sighand(struct sighand_struct *sighand)
 998 {
 999         if (atomic_dec_and_test(&sighand->count)) {
1000                 signalfd_cleanup(sighand);
1001                 kmem_cache_free(sighand_cachep, sighand);
1002         }
1003 }
1004
1005
1006 /*
1007  * Initialize POSIX timer handling for a thread group.
1008  */
1009 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1010 {
1011         unsigned long cpu_limit;
1012
1013         /* Thread group counters. */
1014         thread_group_cputime_init(sig);
1015
1016         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1017         if (cpu_limit != RLIM_INFINITY) {
1018                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1019                 sig->cputimer.running = 1;
1020         }
1021
1022         /* The timer lists. */
1023         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1024         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1025         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1026 }
1027
1028 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1029 {
1030         struct signal_struct *sig;
1031
1032         if (clone_flags & CLONE_THREAD)
1033                 return 0;
1034
1035         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1036         tsk->signal = sig;
1037         if (!sig)
1038                 return -ENOMEM;
1039
1040         sig->nr_threads = 1;
1041         atomic_set(&sig->live, 1);
1042         atomic_set(&sig->sigcnt, 1);
1043         init_waitqueue_head(&sig->wait_chldexit);
1044         sig->curr_target = tsk;
1045         init_sigpending(&sig->shared_pending);
1046         INIT_LIST_HEAD(&sig->posix_timers);
1047
1048         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1049         sig->real_timer.function = it_real_fn;
1050
1051         task_lock(current->group_leader);
1052         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1053         task_unlock(current->group_leader);
1054
1055         posix_cpu_timers_init_group(sig);
1056
1057         tty_audit_fork(sig);
1058         sched_autogroup_fork(sig);
1059
1060 #ifdef CONFIG_CGROUPS
1061         init_rwsem(&sig->group_rwsem);
1062 #endif
1063
1064         sig->oom_score_adj = current->signal->oom_score_adj;
1065         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1066
1067         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1068                                    current->signal->is_child_subreaper;
1069
1070         mutex_init(&sig->cred_guard_mutex);
1071
1072         return 0;
1073 }
1074
1075 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
1076 {
1077         unsigned long new_flags = p->flags;
1078
1079         new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1080         new_flags |= PF_FORKNOEXEC;
1081         p->flags = new_flags;
1082 }
1083
1084 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1085 {
1086         current->clear_child_tid = tidptr;
1087
1088         return task_pid_vnr(current);
1089 }
1090
1091 static void rt_mutex_init_task(struct task_struct *p)
1092 {
1093         raw_spin_lock_init(&p->pi_lock);
1094 #ifdef CONFIG_RT_MUTEXES
1095         plist_head_init(&p->pi_waiters);
1096         p->pi_blocked_on = 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 namespace
1177          * don't allow the creation of threads.
1178          */
1179         if ((clone_flags & (CLONE_VM|CLONE_NEWPID)) &&
1180             (task_active_pid_ns(current) !=
1181              current->nsproxy->pid_ns_for_children))
1182                 return ERR_PTR(-EINVAL);
1183
1184         retval = security_task_create(clone_flags);
1185         if (retval)
1186                 goto fork_out;
1187
1188         retval = -ENOMEM;
1189         p = dup_task_struct(current);
1190         if (!p)
1191                 goto fork_out;
1192
1193         ftrace_graph_init_task(p);
1194         get_seccomp_filter(p);
1195
1196         rt_mutex_init_task(p);
1197
1198 #ifdef CONFIG_PROVE_LOCKING
1199         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1200         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1201 #endif
1202         retval = -EAGAIN;
1203         if (atomic_read(&p->real_cred->user->processes) >=
1204                         task_rlimit(p, RLIMIT_NPROC)) {
1205                 if (p->real_cred->user != INIT_USER &&
1206                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1207                         goto bad_fork_free;
1208         }
1209         current->flags &= ~PF_NPROC_EXCEEDED;
1210
1211         retval = copy_creds(p, clone_flags);
1212         if (retval < 0)
1213                 goto bad_fork_free;
1214
1215         /*
1216          * If multiple threads are within copy_process(), then this check
1217          * triggers too late. This doesn't hurt, the check is only there
1218          * to stop root fork bombs.
1219          */
1220         retval = -EAGAIN;
1221         if (nr_threads >= max_threads)
1222                 goto bad_fork_cleanup_count;
1223
1224         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1225                 goto bad_fork_cleanup_count;
1226
1227         p->did_exec = 0;
1228         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1229         copy_flags(clone_flags, p);
1230         INIT_LIST_HEAD(&p->children);
1231         INIT_LIST_HEAD(&p->sibling);
1232         rcu_copy_process(p);
1233         p->vfork_done = NULL;
1234         spin_lock_init(&p->alloc_lock);
1235
1236         init_sigpending(&p->pending);
1237
1238         p->utime = p->stime = p->gtime = 0;
1239         p->utimescaled = p->stimescaled = 0;
1240 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
1241         p->prev_cputime.utime = p->prev_cputime.stime = 0;
1242 #endif
1243 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1244         seqlock_init(&p->vtime_seqlock);
1245         p->vtime_snap = 0;
1246         p->vtime_snap_whence = VTIME_SLEEPING;
1247 #endif
1248
1249 #if defined(SPLIT_RSS_COUNTING)
1250         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1251 #endif
1252
1253         p->default_timer_slack_ns = current->timer_slack_ns;
1254
1255         task_io_accounting_init(&p->ioac);
1256         acct_clear_integrals(p);
1257
1258         posix_cpu_timers_init(p);
1259
1260         do_posix_clock_monotonic_gettime(&p->start_time);
1261         p->real_start_time = p->start_time;
1262         monotonic_to_bootbased(&p->real_start_time);
1263         p->io_context = NULL;
1264         p->audit_context = NULL;
1265         if (clone_flags & CLONE_THREAD)
1266                 threadgroup_change_begin(current);
1267         cgroup_fork(p);
1268 #ifdef CONFIG_NUMA
1269         p->mempolicy = mpol_dup(p->mempolicy);
1270         if (IS_ERR(p->mempolicy)) {
1271                 retval = PTR_ERR(p->mempolicy);
1272                 p->mempolicy = NULL;
1273                 goto bad_fork_cleanup_cgroup;
1274         }
1275         mpol_fix_fork_child_flag(p);
1276 #endif
1277 #ifdef CONFIG_CPUSETS
1278         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1279         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1280         seqcount_init(&p->mems_allowed_seq);
1281 #endif
1282 #ifdef CONFIG_TRACE_IRQFLAGS
1283         p->irq_events = 0;
1284         p->hardirqs_enabled = 0;
1285         p->hardirq_enable_ip = 0;
1286         p->hardirq_enable_event = 0;
1287         p->hardirq_disable_ip = _THIS_IP_;
1288         p->hardirq_disable_event = 0;
1289         p->softirqs_enabled = 1;
1290         p->softirq_enable_ip = _THIS_IP_;
1291         p->softirq_enable_event = 0;
1292         p->softirq_disable_ip = 0;
1293         p->softirq_disable_event = 0;
1294         p->hardirq_context = 0;
1295         p->softirq_context = 0;
1296 #endif
1297 #ifdef CONFIG_LOCKDEP
1298         p->lockdep_depth = 0; /* no locks held yet */
1299         p->curr_chain_key = 0;
1300         p->lockdep_recursion = 0;
1301 #endif
1302
1303 #ifdef CONFIG_DEBUG_MUTEXES
1304         p->blocked_on = NULL; /* not blocked yet */
1305 #endif
1306 #ifdef CONFIG_MEMCG
1307         p->memcg_batch.do_batch = 0;
1308         p->memcg_batch.memcg = NULL;
1309 #endif
1310 #ifdef CONFIG_BCACHE
1311         p->sequential_io        = 0;
1312         p->sequential_io_avg    = 0;
1313 #endif
1314
1315         /* Perform scheduler related setup. Assign this task to a CPU. */
1316         sched_fork(p);
1317
1318         retval = perf_event_init_task(p);
1319         if (retval)
1320                 goto bad_fork_cleanup_policy;
1321         retval = audit_alloc(p);
1322         if (retval)
1323                 goto bad_fork_cleanup_policy;
1324         /* copy all the process information */
1325         retval = copy_semundo(clone_flags, p);
1326         if (retval)
1327                 goto bad_fork_cleanup_audit;
1328         retval = copy_files(clone_flags, p);
1329         if (retval)
1330                 goto bad_fork_cleanup_semundo;
1331         retval = copy_fs(clone_flags, p);
1332         if (retval)
1333                 goto bad_fork_cleanup_files;
1334         retval = copy_sighand(clone_flags, p);
1335         if (retval)
1336                 goto bad_fork_cleanup_fs;
1337         retval = copy_signal(clone_flags, p);
1338         if (retval)
1339                 goto bad_fork_cleanup_sighand;
1340         retval = copy_mm(clone_flags, p);
1341         if (retval)
1342                 goto bad_fork_cleanup_signal;
1343         retval = copy_namespaces(clone_flags, p);
1344         if (retval)
1345                 goto bad_fork_cleanup_mm;
1346         retval = copy_io(clone_flags, p);
1347         if (retval)
1348                 goto bad_fork_cleanup_namespaces;
1349         retval = copy_thread(clone_flags, stack_start, stack_size, p);
1350         if (retval)
1351                 goto bad_fork_cleanup_io;
1352
1353         if (pid != &init_struct_pid) {
1354                 retval = -ENOMEM;
1355                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1356                 if (!pid)
1357                         goto bad_fork_cleanup_io;
1358         }
1359
1360         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1361         /*
1362          * Clear TID on mm_release()?
1363          */
1364         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1365 #ifdef CONFIG_BLOCK
1366         p->plug = NULL;
1367 #endif
1368 #ifdef CONFIG_FUTEX
1369         p->robust_list = NULL;
1370 #ifdef CONFIG_COMPAT
1371         p->compat_robust_list = NULL;
1372 #endif
1373         INIT_LIST_HEAD(&p->pi_state_list);
1374         p->pi_state_cache = NULL;
1375 #endif
1376         uprobe_copy_process(p);
1377         /*
1378          * sigaltstack should be cleared when sharing the same VM
1379          */
1380         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1381                 p->sas_ss_sp = p->sas_ss_size = 0;
1382
1383         /*
1384          * Syscall tracing and stepping should be turned off in the
1385          * child regardless of CLONE_PTRACE.
1386          */
1387         user_disable_single_step(p);
1388         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1389 #ifdef TIF_SYSCALL_EMU
1390         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1391 #endif
1392         clear_all_latency_tracing(p);
1393
1394         /* ok, now we should be set up.. */
1395         p->pid = pid_nr(pid);
1396         if (clone_flags & CLONE_THREAD) {
1397                 p->exit_signal = -1;
1398                 p->group_leader = current->group_leader;
1399                 p->tgid = current->tgid;
1400         } else {
1401                 if (clone_flags & CLONE_PARENT)
1402                         p->exit_signal = current->group_leader->exit_signal;
1403                 else
1404                         p->exit_signal = (clone_flags & CSIGNAL);
1405                 p->group_leader = p;
1406                 p->tgid = p->pid;
1407         }
1408
1409         p->pdeath_signal = 0;
1410         p->exit_state = 0;
1411
1412         p->nr_dirtied = 0;
1413         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1414         p->dirty_paused_when = 0;
1415
1416         INIT_LIST_HEAD(&p->thread_group);
1417         p->task_works = NULL;
1418
1419         /*
1420          * Make it visible to the rest of the system, but dont wake it up yet.
1421          * Need tasklist lock for parent etc handling!
1422          */
1423         write_lock_irq(&tasklist_lock);
1424
1425         /* CLONE_PARENT re-uses the old parent */
1426         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1427                 p->real_parent = current->real_parent;
1428                 p->parent_exec_id = current->parent_exec_id;
1429         } else {
1430                 p->real_parent = current;
1431                 p->parent_exec_id = current->self_exec_id;
1432         }
1433
1434         spin_lock(&current->sighand->siglock);
1435
1436         /*
1437          * Process group and session signals need to be delivered to just the
1438          * parent before the fork or both the parent and the child after the
1439          * fork. Restart if a signal comes in before we add the new process to
1440          * it's process group.
1441          * A fatal signal pending means that current will exit, so the new
1442          * thread can't slip out of an OOM kill (or normal SIGKILL).
1443         */
1444         recalc_sigpending();
1445         if (signal_pending(current)) {
1446                 spin_unlock(&current->sighand->siglock);
1447                 write_unlock_irq(&tasklist_lock);
1448                 retval = -ERESTARTNOINTR;
1449                 goto bad_fork_free_pid;
1450         }
1451
1452         if (likely(p->pid)) {
1453                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1454
1455                 init_task_pid(p, PIDTYPE_PID, pid);
1456                 if (thread_group_leader(p)) {
1457                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1458                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1459
1460                         if (is_child_reaper(pid)) {
1461                                 ns_of_pid(pid)->child_reaper = p;
1462                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1463                         }
1464
1465                         p->signal->leader_pid = pid;
1466                         p->signal->tty = tty_kref_get(current->signal->tty);
1467                         list_add_tail(&p->sibling, &p->real_parent->children);
1468                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1469                         attach_pid(p, PIDTYPE_PGID);
1470                         attach_pid(p, PIDTYPE_SID);
1471                         __this_cpu_inc(process_counts);
1472                 } else {
1473                         current->signal->nr_threads++;
1474                         atomic_inc(&current->signal->live);
1475                         atomic_inc(&current->signal->sigcnt);
1476                         list_add_tail_rcu(&p->thread_group,
1477                                           &p->group_leader->thread_group);
1478                 }
1479                 attach_pid(p, PIDTYPE_PID);
1480                 nr_threads++;
1481         }
1482
1483         total_forks++;
1484         spin_unlock(&current->sighand->siglock);
1485         write_unlock_irq(&tasklist_lock);
1486         proc_fork_connector(p);
1487         cgroup_post_fork(p);
1488         if (clone_flags & CLONE_THREAD)
1489                 threadgroup_change_end(current);
1490         perf_event_fork(p);
1491
1492         trace_task_newtask(p, clone_flags);
1493
1494         return p;
1495
1496 bad_fork_free_pid:
1497         if (pid != &init_struct_pid)
1498                 free_pid(pid);
1499 bad_fork_cleanup_io:
1500         if (p->io_context)
1501                 exit_io_context(p);
1502 bad_fork_cleanup_namespaces:
1503         exit_task_namespaces(p);
1504 bad_fork_cleanup_mm:
1505         if (p->mm)
1506                 mmput(p->mm);
1507 bad_fork_cleanup_signal:
1508         if (!(clone_flags & CLONE_THREAD))
1509                 free_signal_struct(p->signal);
1510 bad_fork_cleanup_sighand:
1511         __cleanup_sighand(p->sighand);
1512 bad_fork_cleanup_fs:
1513         exit_fs(p); /* blocking */
1514 bad_fork_cleanup_files:
1515         exit_files(p); /* blocking */
1516 bad_fork_cleanup_semundo:
1517         exit_sem(p);
1518 bad_fork_cleanup_audit:
1519         audit_free(p);
1520 bad_fork_cleanup_policy:
1521         perf_event_free_task(p);
1522 #ifdef CONFIG_NUMA
1523         mpol_put(p->mempolicy);
1524 bad_fork_cleanup_cgroup:
1525 #endif
1526         if (clone_flags & CLONE_THREAD)
1527                 threadgroup_change_end(current);
1528         cgroup_exit(p, 0);
1529         delayacct_tsk_free(p);
1530         module_put(task_thread_info(p)->exec_domain->module);
1531 bad_fork_cleanup_count:
1532         atomic_dec(&p->cred->user->processes);
1533         exit_creds(p);
1534 bad_fork_free:
1535         free_task(p);
1536 fork_out:
1537         return ERR_PTR(retval);
1538 }
1539
1540 static inline void init_idle_pids(struct pid_link *links)
1541 {
1542         enum pid_type type;
1543
1544         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1545                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1546                 links[type].pid = &init_struct_pid;
1547         }
1548 }
1549
1550 struct task_struct *fork_idle(int cpu)
1551 {
1552         struct task_struct *task;
1553         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0);
1554         if (!IS_ERR(task)) {
1555                 init_idle_pids(task->pids);
1556                 init_idle(task, cpu);
1557         }
1558
1559         return task;
1560 }
1561
1562 /*
1563  *  Ok, this is the main fork-routine.
1564  *
1565  * It copies the process, and if successful kick-starts
1566  * it and waits for it to finish using the VM if required.
1567  */
1568 long do_fork(unsigned long clone_flags,
1569               unsigned long stack_start,
1570               unsigned long stack_size,
1571               int __user *parent_tidptr,
1572               int __user *child_tidptr)
1573 {
1574         struct task_struct *p;
1575         int trace = 0;
1576         long nr;
1577
1578         /*
1579          * Do some preliminary argument and permissions checking before we
1580          * actually start allocating stuff
1581          */
1582         if (clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) {
1583                 if (clone_flags & (CLONE_THREAD|CLONE_PARENT))
1584                         return -EINVAL;
1585         }
1586
1587         /*
1588          * Determine whether and which event to report to ptracer.  When
1589          * called from kernel_thread or CLONE_UNTRACED is explicitly
1590          * requested, no event is reported; otherwise, report if the event
1591          * for the type of forking is enabled.
1592          */
1593         if (!(clone_flags & CLONE_UNTRACED)) {
1594                 if (clone_flags & CLONE_VFORK)
1595                         trace = PTRACE_EVENT_VFORK;
1596                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1597                         trace = PTRACE_EVENT_CLONE;
1598                 else
1599                         trace = PTRACE_EVENT_FORK;
1600
1601                 if (likely(!ptrace_event_enabled(current, trace)))
1602                         trace = 0;
1603         }
1604
1605         p = copy_process(clone_flags, stack_start, stack_size,
1606                          child_tidptr, NULL, trace);
1607         /*
1608          * Do this prior waking up the new thread - the thread pointer
1609          * might get invalid after that point, if the thread exits quickly.
1610          */
1611         if (!IS_ERR(p)) {
1612                 struct completion vfork;
1613
1614                 trace_sched_process_fork(current, p);
1615
1616                 nr = task_pid_vnr(p);
1617
1618                 if (clone_flags & CLONE_PARENT_SETTID)
1619                         put_user(nr, parent_tidptr);
1620
1621                 if (clone_flags & CLONE_VFORK) {
1622                         p->vfork_done = &vfork;
1623                         init_completion(&vfork);
1624                         get_task_struct(p);
1625                 }
1626
1627                 wake_up_new_task(p);
1628
1629                 /* forking complete and child started to run, tell ptracer */
1630                 if (unlikely(trace))
1631                         ptrace_event(trace, nr);
1632
1633                 if (clone_flags & CLONE_VFORK) {
1634                         if (!wait_for_vfork_done(p, &vfork))
1635                                 ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1636                 }
1637         } else {
1638                 nr = PTR_ERR(p);
1639         }
1640         return nr;
1641 }
1642
1643 /*
1644  * Create a kernel thread.
1645  */
1646 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1647 {
1648         return do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1649                 (unsigned long)arg, NULL, NULL);
1650 }
1651
1652 #ifdef __ARCH_WANT_SYS_FORK
1653 SYSCALL_DEFINE0(fork)
1654 {
1655 #ifdef CONFIG_MMU
1656         return do_fork(SIGCHLD, 0, 0, NULL, NULL);
1657 #else
1658         /* can not support in nommu mode */
1659         return(-EINVAL);
1660 #endif
1661 }
1662 #endif
1663
1664 #ifdef __ARCH_WANT_SYS_VFORK
1665 SYSCALL_DEFINE0(vfork)
1666 {
1667         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1668                         0, NULL, NULL);
1669 }
1670 #endif
1671
1672 #ifdef __ARCH_WANT_SYS_CLONE
1673 #ifdef CONFIG_CLONE_BACKWARDS
1674 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1675                  int __user *, parent_tidptr,
1676                  int, tls_val,
1677                  int __user *, child_tidptr)
1678 #elif defined(CONFIG_CLONE_BACKWARDS2)
1679 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1680                  int __user *, parent_tidptr,
1681                  int __user *, child_tidptr,
1682                  int, tls_val)
1683 #elif defined(CONFIG_CLONE_BACKWARDS3)
1684 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1685                 int, stack_size,
1686                 int __user *, parent_tidptr,
1687                 int __user *, child_tidptr,
1688                 int, tls_val)
1689 #else
1690 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1691                  int __user *, parent_tidptr,
1692                  int __user *, child_tidptr,
1693                  int, tls_val)
1694 #endif
1695 {
1696         return do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr);
1697 }
1698 #endif
1699
1700 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1701 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1702 #endif
1703
1704 static void sighand_ctor(void *data)
1705 {
1706         struct sighand_struct *sighand = data;
1707
1708         spin_lock_init(&sighand->siglock);
1709         init_waitqueue_head(&sighand->signalfd_wqh);
1710 }
1711
1712 void __init proc_caches_init(void)
1713 {
1714         sighand_cachep = kmem_cache_create("sighand_cache",
1715                         sizeof(struct sighand_struct), 0,
1716                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1717                         SLAB_NOTRACK, sighand_ctor);
1718         signal_cachep = kmem_cache_create("signal_cache",
1719                         sizeof(struct signal_struct), 0,
1720                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1721         files_cachep = kmem_cache_create("files_cache",
1722                         sizeof(struct files_struct), 0,
1723                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1724         fs_cachep = kmem_cache_create("fs_cache",
1725                         sizeof(struct fs_struct), 0,
1726                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1727         /*
1728          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1729          * whole struct cpumask for the OFFSTACK case. We could change
1730          * this to *only* allocate as much of it as required by the
1731          * maximum number of CPU's we can ever have.  The cpumask_allocation
1732          * is at the end of the structure, exactly for that reason.
1733          */
1734         mm_cachep = kmem_cache_create("mm_struct",
1735                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1736                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1737         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1738         mmap_init();
1739         nsproxy_cache_init();
1740 }
1741
1742 /*
1743  * Check constraints on flags passed to the unshare system call.
1744  */
1745 static int check_unshare_flags(unsigned long unshare_flags)
1746 {
1747         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1748                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1749                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1750                                 CLONE_NEWUSER|CLONE_NEWPID))
1751                 return -EINVAL;
1752         /*
1753          * Not implemented, but pretend it works if there is nothing to
1754          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1755          * needs to unshare vm.
1756          */
1757         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1758                 /* FIXME: get_task_mm() increments ->mm_users */
1759                 if (atomic_read(&current->mm->mm_users) > 1)
1760                         return -EINVAL;
1761         }
1762
1763         return 0;
1764 }
1765
1766 /*
1767  * Unshare the filesystem structure if it is being shared
1768  */
1769 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1770 {
1771         struct fs_struct *fs = current->fs;
1772
1773         if (!(unshare_flags & CLONE_FS) || !fs)
1774                 return 0;
1775
1776         /* don't need lock here; in the worst case we'll do useless copy */
1777         if (fs->users == 1)
1778                 return 0;
1779
1780         *new_fsp = copy_fs_struct(fs);
1781         if (!*new_fsp)
1782                 return -ENOMEM;
1783
1784         return 0;
1785 }
1786
1787 /*
1788  * Unshare file descriptor table if it is being shared
1789  */
1790 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1791 {
1792         struct files_struct *fd = current->files;
1793         int error = 0;
1794
1795         if ((unshare_flags & CLONE_FILES) &&
1796             (fd && atomic_read(&fd->count) > 1)) {
1797                 *new_fdp = dup_fd(fd, &error);
1798                 if (!*new_fdp)
1799                         return error;
1800         }
1801
1802         return 0;
1803 }
1804
1805 /*
1806  * unshare allows a process to 'unshare' part of the process
1807  * context which was originally shared using clone.  copy_*
1808  * functions used by do_fork() cannot be used here directly
1809  * because they modify an inactive task_struct that is being
1810  * constructed. Here we are modifying the current, active,
1811  * task_struct.
1812  */
1813 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1814 {
1815         struct fs_struct *fs, *new_fs = NULL;
1816         struct files_struct *fd, *new_fd = NULL;
1817         struct cred *new_cred = NULL;
1818         struct nsproxy *new_nsproxy = NULL;
1819         int do_sysvsem = 0;
1820         int err;
1821
1822         /*
1823          * If unsharing a user namespace must also unshare the thread.
1824          */
1825         if (unshare_flags & CLONE_NEWUSER)
1826                 unshare_flags |= CLONE_THREAD | CLONE_FS;
1827         /*
1828          * If unsharing a thread from a thread group, must also unshare vm.
1829          */
1830         if (unshare_flags & CLONE_THREAD)
1831                 unshare_flags |= CLONE_VM;
1832         /*
1833          * If unsharing vm, must also unshare signal handlers.
1834          */
1835         if (unshare_flags & CLONE_VM)
1836                 unshare_flags |= CLONE_SIGHAND;
1837         /*
1838          * If unsharing namespace, must also unshare filesystem information.
1839          */
1840         if (unshare_flags & CLONE_NEWNS)
1841                 unshare_flags |= CLONE_FS;
1842
1843         err = check_unshare_flags(unshare_flags);
1844         if (err)
1845                 goto bad_unshare_out;
1846         /*
1847          * CLONE_NEWIPC must also detach from the undolist: after switching
1848          * to a new ipc namespace, the semaphore arrays from the old
1849          * namespace are unreachable.
1850          */
1851         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1852                 do_sysvsem = 1;
1853         err = unshare_fs(unshare_flags, &new_fs);
1854         if (err)
1855                 goto bad_unshare_out;
1856         err = unshare_fd(unshare_flags, &new_fd);
1857         if (err)
1858                 goto bad_unshare_cleanup_fs;
1859         err = unshare_userns(unshare_flags, &new_cred);
1860         if (err)
1861                 goto bad_unshare_cleanup_fd;
1862         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1863                                          new_cred, new_fs);
1864         if (err)
1865                 goto bad_unshare_cleanup_cred;
1866
1867         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
1868                 if (do_sysvsem) {
1869                         /*
1870                          * CLONE_SYSVSEM is equivalent to sys_exit().
1871                          */
1872                         exit_sem(current);
1873                 }
1874
1875                 if (new_nsproxy)
1876                         switch_task_namespaces(current, new_nsproxy);
1877
1878                 task_lock(current);
1879
1880                 if (new_fs) {
1881                         fs = current->fs;
1882                         spin_lock(&fs->lock);
1883                         current->fs = new_fs;
1884                         if (--fs->users)
1885                                 new_fs = NULL;
1886                         else
1887                                 new_fs = fs;
1888                         spin_unlock(&fs->lock);
1889                 }
1890
1891                 if (new_fd) {
1892                         fd = current->files;
1893                         current->files = new_fd;
1894                         new_fd = fd;
1895                 }
1896
1897                 task_unlock(current);
1898
1899                 if (new_cred) {
1900                         /* Install the new user namespace */
1901                         commit_creds(new_cred);
1902                         new_cred = NULL;
1903                 }
1904         }
1905
1906 bad_unshare_cleanup_cred:
1907         if (new_cred)
1908                 put_cred(new_cred);
1909 bad_unshare_cleanup_fd:
1910         if (new_fd)
1911                 put_files_struct(new_fd);
1912
1913 bad_unshare_cleanup_fs:
1914         if (new_fs)
1915                 free_fs_struct(new_fs);
1916
1917 bad_unshare_out:
1918         return err;
1919 }
1920
1921 /*
1922  *      Helper to unshare the files of the current task.
1923  *      We don't want to expose copy_files internals to
1924  *      the exec layer of the kernel.
1925  */
1926
1927 int unshare_files(struct files_struct **displaced)
1928 {
1929         struct task_struct *task = current;
1930         struct files_struct *copy = NULL;
1931         int error;
1932
1933         error = unshare_fd(CLONE_FILES, &copy);
1934         if (error || !copy) {
1935                 *displaced = NULL;
1936                 return error;
1937         }
1938         *displaced = task->files;
1939         task_lock(task);
1940         task->files = copy;
1941         task_unlock(task);
1942         return 0;
1943 }