kernel/fork.c

   1 /*
   2  *  linux/kernel/fork.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  *  'fork.c' contains the help-routines for the 'fork' system call
   9  * (see also entry.S and others).
  10  * Fork is rather simple, once you get the hang of it, but the memory
  11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  12  */
  13
  14 #include <linux/slab.h>
  15 #include <linux/init.h>
  16 #include <linux/unistd.h>
  17 #include <linux/module.h>
  18 #include <linux/vmalloc.h>
  19 #include <linux/completion.h>
  20 #include <linux/personality.h>
  21 #include <linux/mempolicy.h>
  22 #include <linux/sem.h>
  23 #include <linux/file.h>
  24 #include <linux/fdtable.h>
  25 #include <linux/iocontext.h>
  26 #include <linux/key.h>
  27 #include <linux/binfmts.h>
  28 #include <linux/mman.h>
  29 #include <linux/mmu_notifier.h>
  30 #include <linux/fs.h>
  31 #include <linux/nsproxy.h>
  32 #include <linux/capability.h>
  33 #include <linux/cpu.h>
  34 #include <linux/cgroup.h>
  35 #include <linux/security.h>
  36 #include <linux/hugetlb.h>
  37 #include <linux/seccomp.h>
  38 #include <linux/swap.h>
  39 #include <linux/syscalls.h>
  40 #include <linux/jiffies.h>
  41 #include <linux/futex.h>
  42 #include <linux/compat.h>
  43 #include <linux/kthread.h>
  44 #include <linux/task_io_accounting_ops.h>
  45 #include <linux/rcupdate.h>
  46 #include <linux/ptrace.h>
  47 #include <linux/mount.h>
  48 #include <linux/audit.h>
  49 #include <linux/memcontrol.h>
  50 #include <linux/ftrace.h>
  51 #include <linux/proc_fs.h>
  52 #include <linux/profile.h>
  53 #include <linux/rmap.h>
  54 #include <linux/ksm.h>
  55 #include <linux/acct.h>
  56 #include <linux/tsacct_kern.h>
  57 #include <linux/cn_proc.h>
  58 #include <linux/freezer.h>
  59 #include <linux/delayacct.h>
  60 #include <linux/taskstats_kern.h>
  61 #include <linux/random.h>
  62 #include <linux/tty.h>
  63 #include <linux/blkdev.h>
  64 #include <linux/fs_struct.h>
  65 #include <linux/magic.h>
  66 #include <linux/perf_event.h>
  67 #include <linux/posix-timers.h>
  68 #include <linux/user-return-notifier.h>
  69 #include <linux/oom.h>
  70 #include <linux/khugepaged.h>
  71 #include <linux/signalfd.h>
  72 #include <linux/uprobes.h>
  73 #include <linux/aio.h>
  74
  75 #include <asm/pgtable.h>
  76 #include <asm/pgalloc.h>
  77 #include <asm/uaccess.h>
  78 #include <asm/mmu_context.h>
  79 #include <asm/cacheflush.h>
  80 #include <asm/tlbflush.h>
  81
  82 #include <trace/events/sched.h>
  83
  84 #define CREATE_TRACE_POINTS
  85 #include <trace/events/task.h>
  86
  87 /*
  88  * Protected counters by write_lock_irq(&tasklist_lock)
  89  */
  90 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  91 int nr_threads;                 /* The idle threads do not count.. */
  92
  93 int max_threads;                /* tunable limit on nr_threads */
  94
  95 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  96
  97 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  98
  99 #ifdef CONFIG_PROVE_RCU
 100 int lockdep_tasklist_lock_is_held(void)
 101 {
 102         return lockdep_is_held(&tasklist_lock);
 103 }
 104 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 105 #endif /* #ifdef CONFIG_PROVE_RCU */
 106
 107 int nr_processes(void)
 108 {
 109         int cpu;
 110         int total = 0;
 111
 112         for_each_possible_cpu(cpu)
 113                 total += per_cpu(process_counts, cpu);
 114
 115         return total;
 116 }
 117
 118 void __weak arch_release_task_struct(struct task_struct *tsk)
 119 {
 120 }
 121
 122 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 123 static struct kmem_cache *task_struct_cachep;
 124
 125 static inline struct task_struct *alloc_task_struct_node(int node)
 126 {
 127         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 128 }
 129
 130 static inline void free_task_struct(struct task_struct *tsk)
 131 {
 132         kmem_cache_free(task_struct_cachep, tsk);
 133 }
 134 #endif
 135
 136 void __weak arch_release_thread_info(struct thread_info *ti)
 137 {
 138 }
 139
 140 #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
 141
 142 /*
 143  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 144  * kmemcache based allocator.
 145  */
 146 # if THREAD_SIZE >= PAGE_SIZE
 147 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 148                                                   int node)
 149 {
 150         struct page *page = alloc_pages_node(node, THREADINFO_GFP_ACCOUNTED,
 151                                              THREAD_SIZE_ORDER);
 152
 153         return page ? page_address(page) : NULL;
 154 }
 155
 156 static inline void free_thread_info(struct thread_info *ti)
 157 {
 158         free_memcg_kmem_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 159 }
 160 # else
 161 static struct kmem_cache *thread_info_cache;
 162
 163 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 164                                                   int node)
 165 {
 166         return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
 167 }
 168
 169 static void free_thread_info(struct thread_info *ti)
 170 {
 171         kmem_cache_free(thread_info_cache, ti);
 172 }
 173
 174 void thread_info_cache_init(void)
 175 {
 176         thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
 177                                               THREAD_SIZE, 0, NULL);
 178         BUG_ON(thread_info_cache == NULL);
 179 }
 180 # endif
 181 #endif
 182
 183 /* SLAB cache for signal_struct structures (tsk->signal) */
 184 static struct kmem_cache *signal_cachep;
 185
 186 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 187 struct kmem_cache *sighand_cachep;
 188
 189 /* SLAB cache for files_struct structures (tsk->files) */
 190 struct kmem_cache *files_cachep;
 191
 192 /* SLAB cache for fs_struct structures (tsk->fs) */
 193 struct kmem_cache *fs_cachep;
 194
 195 /* SLAB cache for vm_area_struct structures */
 196 struct kmem_cache *vm_area_cachep;
 197
 198 /* SLAB cache for mm_struct structures (tsk->mm) */
 199 static struct kmem_cache *mm_cachep;
 200
 201 static void account_kernel_stack(struct thread_info *ti, int account)
 202 {
 203         struct zone *zone = page_zone(virt_to_page(ti));
 204
 205         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
 206 }
 207
 208 void free_task(struct task_struct *tsk)
 209 {
 210         account_kernel_stack(tsk->stack, -1);
 211         arch_release_thread_info(tsk->stack);
 212         free_thread_info(tsk->stack);
 213         rt_mutex_debug_task_free(tsk);
 214         ftrace_graph_exit_task(tsk);
 215         put_seccomp_filter(tsk);
 216         arch_release_task_struct(tsk);
 217         free_task_struct(tsk);
 218 }
 219 EXPORT_SYMBOL(free_task);
 220
 221 static inline void free_signal_struct(struct signal_struct *sig)
 222 {
 223         taskstats_tgid_free(sig);
 224         sched_autogroup_exit(sig);
 225         kmem_cache_free(signal_cachep, sig);
 226 }
 227
 228 static inline void put_signal_struct(struct signal_struct *sig)
 229 {
 230         if (atomic_dec_and_test(&sig->sigcnt))
 231                 free_signal_struct(sig);
 232 }
 233
 234 void __put_task_struct(struct task_struct *tsk)
 235 {
 236         WARN_ON(!tsk->exit_state);
 237         WARN_ON(atomic_read(&tsk->usage));
 238         WARN_ON(tsk == current);
 239
 240         task_numa_free(tsk);
 241         security_task_free(tsk);
 242         exit_creds(tsk);
 243         delayacct_tsk_free(tsk);
 244         put_signal_struct(tsk->signal);
 245
 246         if (!profile_handoff_task(tsk))
 247                 free_task(tsk);
 248 }
 249 EXPORT_SYMBOL_GPL(__put_task_struct);
 250
 251 void __init __weak arch_task_cache_init(void) { }
 252
 253 void __init fork_init(unsigned long mempages)
 254 {
 255 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 256 #ifndef ARCH_MIN_TASKALIGN
 257 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 258 #endif
 259         /* create a slab on which task_structs can be allocated */
 260         task_struct_cachep =
 261                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 262                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
 263 #endif
 264
 265         /* do the arch specific task caches init */
 266         arch_task_cache_init();
 267
 268         /*
 269          * The default maximum number of threads is set to a safe
 270          * value: the thread structures can take up at most half
 271          * of memory.
 272          */
 273         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 274
 275         /*
 276          * we need to allow at least 20 threads to boot a system
 277          */
 278         if (max_threads < 20)
 279                 max_threads = 20;
 280
 281         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 282         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 283         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 284                 init_task.signal->rlim[RLIMIT_NPROC];
 285 }
 286
 287 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
 288                                                struct task_struct *src)
 289 {
 290         *dst = *src;
 291         return 0;
 292 }
 293
 294 static struct task_struct *dup_task_struct(struct task_struct *orig)
 295 {
 296         struct task_struct *tsk;
 297         struct thread_info *ti;
 298         unsigned long *stackend;
 299         int node = tsk_fork_get_node(orig);
 300         int err;
 301
 302         tsk = alloc_task_struct_node(node);
 303         if (!tsk)
 304                 return NULL;
 305
 306         ti = alloc_thread_info_node(tsk, node);
 307         if (!ti)
 308                 goto free_tsk;
 309
 310         err = arch_dup_task_struct(tsk, orig);
 311         if (err)
 312                 goto free_ti;
 313
 314         tsk->stack = ti;
 315
 316         setup_thread_stack(tsk, orig);
 317         clear_user_return_notifier(tsk);
 318         clear_tsk_need_resched(tsk);
 319         stackend = end_of_stack(tsk);
 320         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 321
 322 #ifdef CONFIG_CC_STACKPROTECTOR
 323         tsk->stack_canary = get_random_int();
 324 #endif
 325
 326         /*
 327          * One for us, one for whoever does the "release_task()" (usually
 328          * parent)
 329          */
 330         atomic_set(&tsk->usage, 2);
 331 #ifdef CONFIG_BLK_DEV_IO_TRACE
 332         tsk->btrace_seq = 0;
 333 #endif
 334         tsk->splice_pipe = NULL;
 335         tsk->task_frag.page = NULL;
 336
 337         account_kernel_stack(ti, 1);
 338
 339         return tsk;
 340
 341 free_ti:
 342         free_thread_info(ti);
 343 free_tsk:
 344         free_task_struct(tsk);
 345         return NULL;
 346 }
 347
 348 #ifdef CONFIG_MMU
 349 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 350 {
 351         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 352         struct rb_node **rb_link, *rb_parent;
 353         int retval;
 354         unsigned long charge;
 355
 356         uprobe_start_dup_mmap();
 357         down_write(&oldmm->mmap_sem);
 358         flush_cache_dup_mm(oldmm);
 359         uprobe_dup_mmap(oldmm, mm);
 360         /*
 361          * Not linked in yet - no deadlock potential:
 362          */
 363         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 364
 365         mm->locked_vm = 0;
 366         mm->mmap = NULL;
 367         mm->mmap_cache = NULL;
 368         mm->map_count = 0;
 369         cpumask_clear(mm_cpumask(mm));
 370         mm->mm_rb = RB_ROOT;
 371         rb_link = &mm->mm_rb.rb_node;
 372         rb_parent = NULL;
 373         pprev = &mm->mmap;
 374         retval = ksm_fork(mm, oldmm);
 375         if (retval)
 376                 goto out;
 377         retval = khugepaged_fork(mm, oldmm);
 378         if (retval)
 379                 goto out;
 380
 381         prev = NULL;
 382         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 383                 struct file *file;
 384
 385                 if (mpnt->vm_flags & VM_DONTCOPY) {
 386                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 387                                                         -vma_pages(mpnt));
 388                         continue;
 389                 }
 390                 charge = 0;
 391                 if (mpnt->vm_flags & VM_ACCOUNT) {
 392                         unsigned long len = vma_pages(mpnt);
 393
 394                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 395                                 goto fail_nomem;
 396                         charge = len;
 397                 }
 398                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 399                 if (!tmp)
 400                         goto fail_nomem;
 401                 *tmp = *mpnt;
 402                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 403                 retval = vma_dup_policy(mpnt, tmp);
 404                 if (retval)
 405                         goto fail_nomem_policy;
 406                 tmp->vm_mm = mm;
 407                 if (anon_vma_fork(tmp, mpnt))
 408                         goto fail_nomem_anon_vma_fork;
 409                 tmp->vm_flags &= ~VM_LOCKED;
 410                 tmp->vm_next = tmp->vm_prev = NULL;
 411                 file = tmp->vm_file;
 412                 if (file) {
 413                         struct inode *inode = file_inode(file);
 414                         struct address_space *mapping = file->f_mapping;
 415
 416                         get_file(file);
 417                         if (tmp->vm_flags & VM_DENYWRITE)
 418                                 atomic_dec(&inode->i_writecount);
 419                         mutex_lock(&mapping->i_mmap_mutex);
 420                         if (tmp->vm_flags & VM_SHARED)
 421                                 mapping->i_mmap_writable++;
 422                         flush_dcache_mmap_lock(mapping);
 423                         /* insert tmp into the share list, just after mpnt */
 424                         if (unlikely(tmp->vm_flags & VM_NONLINEAR))
 425                                 vma_nonlinear_insert(tmp,
 426                                                 &mapping->i_mmap_nonlinear);
 427                         else
 428                                 vma_interval_tree_insert_after(tmp, mpnt,
 429                                                         &mapping->i_mmap);
 430                         flush_dcache_mmap_unlock(mapping);
 431                         mutex_unlock(&mapping->i_mmap_mutex);
 432                 }
 433
 434                 /*
 435                  * Clear hugetlb-related page reserves for children. This only
 436                  * affects MAP_PRIVATE mappings. Faults generated by the child
 437                  * are not guaranteed to succeed, even if read-only
 438                  */
 439                 if (is_vm_hugetlb_page(tmp))
 440                         reset_vma_resv_huge_pages(tmp);
 441
 442                 /*
 443                  * Link in the new vma and copy the page table entries.
 444                  */
 445                 *pprev = tmp;
 446                 pprev = &tmp->vm_next;
 447                 tmp->vm_prev = prev;
 448                 prev = tmp;
 449
 450                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 451                 rb_link = &tmp->vm_rb.rb_right;
 452                 rb_parent = &tmp->vm_rb;
 453
 454                 mm->map_count++;
 455                 retval = copy_page_range(mm, oldmm, mpnt);
 456
 457                 if (tmp->vm_ops && tmp->vm_ops->open)
 458                         tmp->vm_ops->open(tmp);
 459
 460                 if (retval)
 461                         goto out;
 462         }
 463         /* a new mm has just been created */
 464         arch_dup_mmap(oldmm, mm);
 465         retval = 0;
 466 out:
 467         up_write(&mm->mmap_sem);
 468         flush_tlb_mm(oldmm);
 469         up_write(&oldmm->mmap_sem);
 470         uprobe_end_dup_mmap();
 471         return retval;
 472 fail_nomem_anon_vma_fork:
 473         mpol_put(vma_policy(tmp));
 474 fail_nomem_policy:
 475         kmem_cache_free(vm_area_cachep, tmp);
 476 fail_nomem:
 477         retval = -ENOMEM;
 478         vm_unacct_memory(charge);
 479         goto out;
 480 }
 481
 482 static inline int mm_alloc_pgd(struct mm_struct *mm)
 483 {
 484         mm->pgd = pgd_alloc(mm);
 485         if (unlikely(!mm->pgd))
 486                 return -ENOMEM;
 487         return 0;
 488 }
 489
 490 static inline void mm_free_pgd(struct mm_struct *mm)
 491 {
 492         pgd_free(mm, mm->pgd);
 493 }
 494 #else
 495 #define dup_mmap(mm, oldmm)     (0)
 496 #define mm_alloc_pgd(mm)        (0)
 497 #define mm_free_pgd(mm)
 498 #endif /* CONFIG_MMU */
 499
 500 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 501
 502 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 503 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 504
 505 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 506
 507 static int __init coredump_filter_setup(char *s)
 508 {
 509         default_dump_filter =
 510                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 511                 MMF_DUMP_FILTER_MASK;
 512         return 1;
 513 }
 514
 515 __setup("coredump_filter=", coredump_filter_setup);
 516
 517 #include <linux/init_task.h>
 518
 519 static void mm_init_aio(struct mm_struct *mm)
 520 {
 521 #ifdef CONFIG_AIO
 522         spin_lock_init(&mm->ioctx_lock);
 523         mm->ioctx_table = NULL;
 524 #endif
 525 }
 526
 527 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
 528 {
 529         atomic_set(&mm->mm_users, 1);
 530         atomic_set(&mm->mm_count, 1);
 531         init_rwsem(&mm->mmap_sem);
 532         INIT_LIST_HEAD(&mm->mmlist);
 533         mm->core_state = NULL;
 534         atomic_long_set(&mm->nr_ptes, 0);
 535         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 536         spin_lock_init(&mm->page_table_lock);
 537         mm_init_aio(mm);
 538         mm_init_owner(mm, p);
 539         clear_tlb_flush_pending(mm);
 540
 541         if (current->mm) {
 542                 mm->flags = current->mm->flags & MMF_INIT_MASK;
 543                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
 544         } else {
 545                 mm->flags = default_dump_filter;
 546                 mm->def_flags = 0;
 547         }
 548
 549         if (likely(!mm_alloc_pgd(mm))) {
 550                 mmu_notifier_mm_init(mm);
 551                 return mm;
 552         }
 553
 554         free_mm(mm);
 555         return NULL;
 556 }
 557
 558 static void check_mm(struct mm_struct *mm)
 559 {
 560         int i;
 561
 562         for (i = 0; i < NR_MM_COUNTERS; i++) {
 563                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 564
 565                 if (unlikely(x))
 566                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 567                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 568         }
 569
 570 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 571         VM_BUG_ON(mm->pmd_huge_pte);
 572 #endif
 573 }
 574
 575 /*
 576  * Allocate and initialize an mm_struct.
 577  */
 578 struct mm_struct *mm_alloc(void)
 579 {
 580         struct mm_struct *mm;
 581
 582         mm = allocate_mm();
 583         if (!mm)
 584                 return NULL;
 585
 586         memset(mm, 0, sizeof(*mm));
 587         mm_init_cpumask(mm);
 588         return mm_init(mm, current);
 589 }
 590
 591 /*
 592  * Called when the last reference to the mm
 593  * is dropped: either by a lazy thread or by
 594  * mmput. Free the page directory and the mm.
 595  */
 596 void __mmdrop(struct mm_struct *mm)
 597 {
 598         BUG_ON(mm == &init_mm);
 599         mm_free_pgd(mm);
 600         destroy_context(mm);
 601         mmu_notifier_mm_destroy(mm);
 602         check_mm(mm);
 603         free_mm(mm);
 604 }
 605 EXPORT_SYMBOL_GPL(__mmdrop);
 606
 607 /*
 608  * Decrement the use count and release all resources for an mm.
 609  */
 610 void mmput(struct mm_struct *mm)
 611 {
 612         might_sleep();
 613
 614         if (atomic_dec_and_test(&mm->mm_users)) {
 615                 uprobe_clear_state(mm);
 616                 exit_aio(mm);
 617                 ksm_exit(mm);
 618                 khugepaged_exit(mm); /* must run before exit_mmap */
 619                 exit_mmap(mm);
 620                 set_mm_exe_file(mm, NULL);
 621                 if (!list_empty(&mm->mmlist)) {
 622                         spin_lock(&mmlist_lock);
 623                         list_del(&mm->mmlist);
 624                         spin_unlock(&mmlist_lock);
 625                 }
 626                 if (mm->binfmt)
 627                         module_put(mm->binfmt->module);
 628                 mmdrop(mm);
 629         }
 630 }
 631 EXPORT_SYMBOL_GPL(mmput);
 632
 633 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 634 {
 635         if (new_exe_file)
 636                 get_file(new_exe_file);
 637         if (mm->exe_file)
 638                 fput(mm->exe_file);
 639         mm->exe_file = new_exe_file;
 640 }
 641
 642 struct file *get_mm_exe_file(struct mm_struct *mm)
 643 {
 644         struct file *exe_file;
 645
 646         /* We need mmap_sem to protect against races with removal of exe_file */
 647         down_read(&mm->mmap_sem);
 648         exe_file = mm->exe_file;
 649         if (exe_file)
 650                 get_file(exe_file);
 651         up_read(&mm->mmap_sem);
 652         return exe_file;
 653 }
 654
 655 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
 656 {
 657         /* It's safe to write the exe_file pointer without exe_file_lock because
 658          * this is called during fork when the task is not yet in /proc */
 659         newmm->exe_file = get_mm_exe_file(oldmm);
 660 }
 661
 662 /**
 663  * get_task_mm - acquire a reference to the task's mm
 664  *
 665  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 666  * this kernel workthread has transiently adopted a user mm with use_mm,
 667  * to do its AIO) is not set and if so returns a reference to it, after
 668  * bumping up the use count.  User must release the mm via mmput()
 669  * after use.  Typically used by /proc and ptrace.
 670  */
 671 struct mm_struct *get_task_mm(struct task_struct *task)
 672 {
 673         struct mm_struct *mm;
 674
 675         task_lock(task);
 676         mm = task->mm;
 677         if (mm) {
 678                 if (task->flags & PF_KTHREAD)
 679                         mm = NULL;
 680                 else
 681                         atomic_inc(&mm->mm_users);
 682         }
 683         task_unlock(task);
 684         return mm;
 685 }
 686 EXPORT_SYMBOL_GPL(get_task_mm);
 687
 688 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 689 {
 690         struct mm_struct *mm;
 691         int err;
 692
 693         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 694         if (err)
 695                 return ERR_PTR(err);
 696
 697         mm = get_task_mm(task);
 698         if (mm && mm != current->mm &&
 699                         !ptrace_may_access(task, mode)) {
 700                 mmput(mm);
 701                 mm = ERR_PTR(-EACCES);
 702         }
 703         mutex_unlock(&task->signal->cred_guard_mutex);
 704
 705         return mm;
 706 }
 707
 708 static void complete_vfork_done(struct task_struct *tsk)
 709 {
 710         struct completion *vfork;
 711
 712         task_lock(tsk);
 713         vfork = tsk->vfork_done;
 714         if (likely(vfork)) {
 715                 tsk->vfork_done = NULL;
 716                 complete(vfork);
 717         }
 718         task_unlock(tsk);
 719 }
 720
 721 static int wait_for_vfork_done(struct task_struct *child,
 722                                 struct completion *vfork)
 723 {
 724         int killed;
 725
 726         freezer_do_not_count();
 727         killed = wait_for_completion_killable(vfork);
 728         freezer_count();
 729
 730         if (killed) {
 731                 task_lock(child);
 732                 child->vfork_done = NULL;
 733                 task_unlock(child);
 734         }
 735
 736         put_task_struct(child);
 737         return killed;
 738 }
 739
 740 /* Please note the differences between mmput and mm_release.
 741  * mmput is called whenever we stop holding onto a mm_struct,
 742  * error success whatever.
 743  *
 744  * mm_release is called after a mm_struct has been removed
 745  * from the current process.
 746  *
 747  * This difference is important for error handling, when we
 748  * only half set up a mm_struct for a new process and need to restore
 749  * the old one.  Because we mmput the new mm_struct before
 750  * restoring the old one. . .
 751  * Eric Biederman 10 January 1998
 752  */
 753 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 754 {
 755         /* Get rid of any futexes when releasing the mm */
 756 #ifdef CONFIG_FUTEX
 757         if (unlikely(tsk->robust_list)) {
 758                 exit_robust_list(tsk);
 759                 tsk->robust_list = NULL;
 760         }
 761 #ifdef CONFIG_COMPAT
 762         if (unlikely(tsk->compat_robust_list)) {
 763                 compat_exit_robust_list(tsk);
 764                 tsk->compat_robust_list = NULL;
 765         }
 766 #endif
 767         if (unlikely(!list_empty(&tsk->pi_state_list)))
 768                 exit_pi_state_list(tsk);
 769 #endif
 770
 771         uprobe_free_utask(tsk);
 772
 773         /* Get rid of any cached register state */
 774         deactivate_mm(tsk, mm);
 775
 776         /*
 777          * If we're exiting normally, clear a user-space tid field if
 778          * requested.  We leave this alone when dying by signal, to leave
 779          * the value intact in a core dump, and to save the unnecessary
 780          * trouble, say, a killed vfork parent shouldn't touch this mm.
 781          * Userland only wants this done for a sys_exit.
 782          */
 783         if (tsk->clear_child_tid) {
 784                 if (!(tsk->flags & PF_SIGNALED) &&
 785                     atomic_read(&mm->mm_users) > 1) {
 786                         /*
 787                          * We don't check the error code - if userspace has
 788                          * not set up a proper pointer then tough luck.
 789                          */
 790                         put_user(0, tsk->clear_child_tid);
 791                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 792                                         1, NULL, NULL, 0);
 793                 }
 794                 tsk->clear_child_tid = NULL;
 795         }
 796
 797         /*
 798          * All done, finally we can wake up parent and return this mm to him.
 799          * Also kthread_stop() uses this completion for synchronization.
 800          */
 801         if (tsk->vfork_done)
 802                 complete_vfork_done(tsk);
 803 }
 804
 805 /*
 806  * Allocate a new mm structure and copy contents from the
 807  * mm structure of the passed in task structure.
 808  */
 809 static struct mm_struct *dup_mm(struct task_struct *tsk)
 810 {
 811         struct mm_struct *mm, *oldmm = current->mm;
 812         int err;
 813
 814         mm = allocate_mm();
 815         if (!mm)
 816                 goto fail_nomem;
 817
 818         memcpy(mm, oldmm, sizeof(*mm));
 819         mm_init_cpumask(mm);
 820
 821 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 822         mm->pmd_huge_pte = NULL;
 823 #endif
 824         if (!mm_init(mm, tsk))
 825                 goto fail_nomem;
 826
 827         if (init_new_context(tsk, mm))
 828                 goto fail_nocontext;
 829
 830         dup_mm_exe_file(oldmm, mm);
 831
 832         err = dup_mmap(mm, oldmm);
 833         if (err)
 834                 goto free_pt;
 835
 836         mm->hiwater_rss = get_mm_rss(mm);
 837         mm->hiwater_vm = mm->total_vm;
 838
 839         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 840                 goto free_pt;
 841
 842         return mm;
 843
 844 free_pt:
 845         /* don't put binfmt in mmput, we haven't got module yet */
 846         mm->binfmt = NULL;
 847         mmput(mm);
 848
 849 fail_nomem:
 850         return NULL;
 851
 852 fail_nocontext:
 853         /*
 854          * If init_new_context() failed, we cannot use mmput() to free the mm
 855          * because it calls destroy_context()
 856          */
 857         mm_free_pgd(mm);
 858         free_mm(mm);
 859         return NULL;
 860 }
 861
 862 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 863 {
 864         struct mm_struct *mm, *oldmm;
 865         int retval;
 866
 867         tsk->min_flt = tsk->maj_flt = 0;
 868         tsk->nvcsw = tsk->nivcsw = 0;
 869 #ifdef CONFIG_DETECT_HUNG_TASK
 870         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 871 #endif
 872
 873         tsk->mm = NULL;
 874         tsk->active_mm = NULL;
 875
 876         /*
 877          * Are we cloning a kernel thread?
 878          *
 879          * We need to steal a active VM for that..
 880          */
 881         oldmm = current->mm;
 882         if (!oldmm)
 883                 return 0;
 884
 885         if (clone_flags & CLONE_VM) {
 886                 atomic_inc(&oldmm->mm_users);
 887                 mm = oldmm;
 888                 goto good_mm;
 889         }
 890
 891         retval = -ENOMEM;
 892         mm = dup_mm(tsk);
 893         if (!mm)
 894                 goto fail_nomem;
 895
 896 good_mm:
 897         tsk->mm = mm;
 898         tsk->active_mm = mm;
 899         return 0;
 900
 901 fail_nomem:
 902         return retval;
 903 }
 904
 905 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 906 {
 907         struct fs_struct *fs = current->fs;
 908         if (clone_flags & CLONE_FS) {
 909                 /* tsk->fs is already what we want */
 910                 spin_lock(&fs->lock);
 911                 if (fs->in_exec) {
 912                         spin_unlock(&fs->lock);
 913                         return -EAGAIN;
 914                 }
 915                 fs->users++;
 916                 spin_unlock(&fs->lock);
 917                 return 0;
 918         }
 919         tsk->fs = copy_fs_struct(fs);
 920         if (!tsk->fs)
 921                 return -ENOMEM;
 922         return 0;
 923 }
 924
 925 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
 926 {
 927         struct files_struct *oldf, *newf;
 928         int error = 0;
 929
 930         /*
 931          * A background process may not have any files ...
 932          */
 933         oldf = current->files;
 934         if (!oldf)
 935                 goto out;
 936
 937         if (clone_flags & CLONE_FILES) {
 938                 atomic_inc(&oldf->count);
 939                 goto out;
 940         }
 941
 942         newf = dup_fd(oldf, &error);
 943         if (!newf)
 944                 goto out;
 945
 946         tsk->files = newf;
 947         error = 0;
 948 out:
 949         return error;
 950 }
 951
 952 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 953 {
 954 #ifdef CONFIG_BLOCK
 955         struct io_context *ioc = current->io_context;
 956         struct io_context *new_ioc;
 957
 958         if (!ioc)
 959                 return 0;
 960         /*
 961          * Share io context with parent, if CLONE_IO is set
 962          */
 963         if (clone_flags & CLONE_IO) {
 964                 ioc_task_link(ioc);
 965                 tsk->io_context = ioc;
 966         } else if (ioprio_valid(ioc->ioprio)) {
 967                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
 968                 if (unlikely(!new_ioc))
 969                         return -ENOMEM;
 970
 971                 new_ioc->ioprio = ioc->ioprio;
 972                 put_io_context(new_ioc);
 973         }
 974 #endif
 975         return 0;
 976 }
 977
 978 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 979 {
 980         struct sighand_struct *sig;
 981
 982         if (clone_flags & CLONE_SIGHAND) {
 983                 atomic_inc(&current->sighand->count);
 984                 return 0;
 985         }
 986         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 987         rcu_assign_pointer(tsk->sighand, sig);
 988         if (!sig)
 989                 return -ENOMEM;
 990         atomic_set(&sig->count, 1);
 991         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 992         return 0;
 993 }
 994
 995 void __cleanup_sighand(struct sighand_struct *sighand)
 996 {
 997         if (atomic_dec_and_test(&sighand->count)) {
 998                 signalfd_cleanup(sighand);
 999                 kmem_cache_free(sighand_cachep, sighand);
1000         }
1001 }
1002
1003
1004 /*
1005  * Initialize POSIX timer handling for a thread group.
1006  */
1007 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1008 {
1009         unsigned long cpu_limit;
1010
1011         /* Thread group counters. */
1012         thread_group_cputime_init(sig);
1013
1014         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1015         if (cpu_limit != RLIM_INFINITY) {
1016                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1017                 sig->cputimer.running = 1;
1018         }
1019
1020         /* The timer lists. */
1021         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1022         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1023         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1024 }
1025
1026 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1027 {
1028         struct signal_struct *sig;
1029
1030         if (clone_flags & CLONE_THREAD)
1031                 return 0;
1032
1033         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1034         tsk->signal = sig;
1035         if (!sig)
1036                 return -ENOMEM;
1037
1038         sig->nr_threads = 1;
1039         atomic_set(&sig->live, 1);
1040         atomic_set(&sig->sigcnt, 1);
1041
1042         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1043         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1044         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1045
1046         init_waitqueue_head(&sig->wait_chldexit);
1047         sig->curr_target = tsk;
1048         init_sigpending(&sig->shared_pending);
1049         INIT_LIST_HEAD(&sig->posix_timers);
1050
1051         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1052         sig->real_timer.function = it_real_fn;
1053
1054         task_lock(current->group_leader);
1055         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1056         task_unlock(current->group_leader);
1057
1058         posix_cpu_timers_init_group(sig);
1059
1060         tty_audit_fork(sig);
1061         sched_autogroup_fork(sig);
1062
1063 #ifdef CONFIG_CGROUPS
1064         init_rwsem(&sig->group_rwsem);
1065 #endif
1066
1067         sig->oom_score_adj = current->signal->oom_score_adj;
1068         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1069
1070         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1071                                    current->signal->is_child_subreaper;
1072
1073         mutex_init(&sig->cred_guard_mutex);
1074
1075         return 0;
1076 }
1077
1078 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
1079 {
1080         unsigned long new_flags = p->flags;
1081
1082         new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1083         new_flags |= PF_FORKNOEXEC;
1084         p->flags = new_flags;
1085 }
1086
1087 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1088 {
1089         current->clear_child_tid = tidptr;
1090
1091         return task_pid_vnr(current);
1092 }
1093
1094 static void rt_mutex_init_task(struct task_struct *p)
1095 {
1096         raw_spin_lock_init(&p->pi_lock);
1097 #ifdef CONFIG_RT_MUTEXES
1098         p->pi_waiters = RB_ROOT;
1099         p->pi_waiters_leftmost = NULL;
1100         p->pi_blocked_on = NULL;
1101         p->pi_top_task = NULL;
1102 #endif
1103 }
1104
1105 #ifdef CONFIG_MM_OWNER
1106 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1107 {
1108         mm->owner = p;
1109 }
1110 #endif /* CONFIG_MM_OWNER */
1111
1112 /*
1113  * Initialize POSIX timer handling for a single task.
1114  */
1115 static void posix_cpu_timers_init(struct task_struct *tsk)
1116 {
1117         tsk->cputime_expires.prof_exp = 0;
1118         tsk->cputime_expires.virt_exp = 0;
1119         tsk->cputime_expires.sched_exp = 0;
1120         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1121         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1122         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1123 }
1124
1125 static inline void
1126 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1127 {
1128          task->pids[type].pid = pid;
1129 }
1130
1131 /*
1132  * This creates a new process as a copy of the old one,
1133  * but does not actually start it yet.
1134  *
1135  * It copies the registers, and all the appropriate
1136  * parts of the process environment (as per the clone
1137  * flags). The actual kick-off is left to the caller.
1138  */
1139 static struct task_struct *copy_process(unsigned long clone_flags,
1140                                         unsigned long stack_start,
1141                                         unsigned long stack_size,
1142                                         int __user *child_tidptr,
1143                                         struct pid *pid,
1144                                         int trace)
1145 {
1146         int retval;
1147         struct task_struct *p;
1148
1149         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1150                 return ERR_PTR(-EINVAL);
1151
1152         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1153                 return ERR_PTR(-EINVAL);
1154
1155         /*
1156          * Thread groups must share signals as well, and detached threads
1157          * can only be started up within the thread group.
1158          */
1159         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1160                 return ERR_PTR(-EINVAL);
1161
1162         /*
1163          * Shared signal handlers imply shared VM. By way of the above,
1164          * thread groups also imply shared VM. Blocking this case allows
1165          * for various simplifications in other code.
1166          */
1167         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1168                 return ERR_PTR(-EINVAL);
1169
1170         /*
1171          * Siblings of global init remain as zombies on exit since they are
1172          * not reaped by their parent (swapper). To solve this and to avoid
1173          * multi-rooted process trees, prevent global and container-inits
1174          * from creating siblings.
1175          */
1176         if ((clone_flags & CLONE_PARENT) &&
1177                                 current->signal->flags & SIGNAL_UNKILLABLE)
1178                 return ERR_PTR(-EINVAL);
1179
1180         /*
1181          * If the new process will be in a different pid or user namespace
1182          * do not allow it to share a thread group or signal handlers or
1183          * parent with the forking task.
1184          */
1185         if (clone_flags & CLONE_SIGHAND) {
1186                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1187                     (task_active_pid_ns(current) !=
1188                                 current->nsproxy->pid_ns_for_children))
1189                         return ERR_PTR(-EINVAL);
1190         }
1191
1192         retval = security_task_create(clone_flags);
1193         if (retval)
1194                 goto fork_out;
1195
1196         retval = -ENOMEM;
1197         p = dup_task_struct(current);
1198         if (!p)
1199                 goto fork_out;
1200
1201         ftrace_graph_init_task(p);
1202         get_seccomp_filter(p);
1203
1204         rt_mutex_init_task(p);
1205
1206 #ifdef CONFIG_PROVE_LOCKING
1207         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1208         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1209 #endif
1210         retval = -EAGAIN;
1211         if (atomic_read(&p->real_cred->user->processes) >=
1212                         task_rlimit(p, RLIMIT_NPROC)) {
1213                 if (p->real_cred->user != INIT_USER &&
1214                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1215                         goto bad_fork_free;
1216         }
1217         current->flags &= ~PF_NPROC_EXCEEDED;
1218
1219         retval = copy_creds(p, clone_flags);
1220         if (retval < 0)
1221                 goto bad_fork_free;
1222
1223         /*
1224          * If multiple threads are within copy_process(), then this check
1225          * triggers too late. This doesn't hurt, the check is only there
1226          * to stop root fork bombs.
1227          */
1228         retval = -EAGAIN;
1229         if (nr_threads >= max_threads)
1230                 goto bad_fork_cleanup_count;
1231
1232         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1233                 goto bad_fork_cleanup_count;
1234
1235         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1236         copy_flags(clone_flags, p);
1237         INIT_LIST_HEAD(&p->children);
1238         INIT_LIST_HEAD(&p->sibling);
1239         rcu_copy_process(p);
1240         p->vfork_done = NULL;
1241         spin_lock_init(&p->alloc_lock);
1242
1243         init_sigpending(&p->pending);
1244
1245         p->utime = p->stime = p->gtime = 0;
1246         p->utimescaled = p->stimescaled = 0;
1247 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
1248         p->prev_cputime.utime = p->prev_cputime.stime = 0;
1249 #endif
1250 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1251         seqlock_init(&p->vtime_seqlock);
1252         p->vtime_snap = 0;
1253         p->vtime_snap_whence = VTIME_SLEEPING;
1254 #endif
1255
1256 #if defined(SPLIT_RSS_COUNTING)
1257         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1258 #endif
1259
1260         p->default_timer_slack_ns = current->timer_slack_ns;
1261
1262         task_io_accounting_init(&p->ioac);
1263         acct_clear_integrals(p);
1264
1265         posix_cpu_timers_init(p);
1266
1267         do_posix_clock_monotonic_gettime(&p->start_time);
1268         p->real_start_time = p->start_time;
1269         monotonic_to_bootbased(&p->real_start_time);
1270         p->io_context = NULL;
1271         p->audit_context = NULL;
1272         if (clone_flags & CLONE_THREAD)
1273                 threadgroup_change_begin(current);
1274         cgroup_fork(p);
1275 #ifdef CONFIG_NUMA
1276         p->mempolicy = mpol_dup(p->mempolicy);
1277         if (IS_ERR(p->mempolicy)) {
1278                 retval = PTR_ERR(p->mempolicy);
1279                 p->mempolicy = NULL;
1280                 goto bad_fork_cleanup_threadgroup_lock;
1281         }
1282         mpol_fix_fork_child_flag(p);
1283 #endif
1284 #ifdef CONFIG_CPUSETS
1285         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1286         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1287         seqcount_init(&p->mems_allowed_seq);
1288 #endif
1289 #ifdef CONFIG_TRACE_IRQFLAGS
1290         p->irq_events = 0;
1291         p->hardirqs_enabled = 0;
1292         p->hardirq_enable_ip = 0;
1293         p->hardirq_enable_event = 0;
1294         p->hardirq_disable_ip = _THIS_IP_;
1295         p->hardirq_disable_event = 0;
1296         p->softirqs_enabled = 1;
1297         p->softirq_enable_ip = _THIS_IP_;
1298         p->softirq_enable_event = 0;
1299         p->softirq_disable_ip = 0;
1300         p->softirq_disable_event = 0;
1301         p->hardirq_context = 0;
1302         p->softirq_context = 0;
1303 #endif
1304 #ifdef CONFIG_LOCKDEP
1305         p->lockdep_depth = 0; /* no locks held yet */
1306         p->curr_chain_key = 0;
1307         p->lockdep_recursion = 0;
1308 #endif
1309
1310 #ifdef CONFIG_DEBUG_MUTEXES
1311         p->blocked_on = NULL; /* not blocked yet */
1312 #endif
1313 #ifdef CONFIG_MEMCG
1314         p->memcg_batch.do_batch = 0;
1315         p->memcg_batch.memcg = NULL;
1316 #endif
1317 #ifdef CONFIG_BCACHE
1318         p->sequential_io        = 0;
1319         p->sequential_io_avg    = 0;
1320 #endif
1321
1322         /* Perform scheduler related setup. Assign this task to a CPU. */
1323         retval = sched_fork(clone_flags, p);
1324         if (retval)
1325                 goto bad_fork_cleanup_policy;
1326
1327         retval = perf_event_init_task(p);
1328         if (retval)
1329                 goto bad_fork_cleanup_policy;
1330         retval = audit_alloc(p);
1331         if (retval)
1332                 goto bad_fork_cleanup_policy;
1333         /* copy all the process information */
1334         retval = copy_semundo(clone_flags, p);
1335         if (retval)
1336                 goto bad_fork_cleanup_audit;
1337         retval = copy_files(clone_flags, p);
1338         if (retval)
1339                 goto bad_fork_cleanup_semundo;
1340         retval = copy_fs(clone_flags, p);
1341         if (retval)
1342                 goto bad_fork_cleanup_files;
1343         retval = copy_sighand(clone_flags, p);
1344         if (retval)
1345                 goto bad_fork_cleanup_fs;
1346         retval = copy_signal(clone_flags, p);
1347         if (retval)
1348                 goto bad_fork_cleanup_sighand;
1349         retval = copy_mm(clone_flags, p);
1350         if (retval)
1351                 goto bad_fork_cleanup_signal;
1352         retval = copy_namespaces(clone_flags, p);
1353         if (retval)
1354                 goto bad_fork_cleanup_mm;
1355         retval = copy_io(clone_flags, p);
1356         if (retval)
1357                 goto bad_fork_cleanup_namespaces;
1358         retval = copy_thread(clone_flags, stack_start, stack_size, p);
1359         if (retval)
1360                 goto bad_fork_cleanup_io;
1361
1362         if (pid != &init_struct_pid) {
1363                 retval = -ENOMEM;
1364                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1365                 if (!pid)
1366                         goto bad_fork_cleanup_io;
1367         }
1368
1369         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1370         /*
1371          * Clear TID on mm_release()?
1372          */
1373         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1374 #ifdef CONFIG_BLOCK
1375         p->plug = NULL;
1376 #endif
1377 #ifdef CONFIG_FUTEX
1378         p->robust_list = NULL;
1379 #ifdef CONFIG_COMPAT
1380         p->compat_robust_list = NULL;
1381 #endif
1382         INIT_LIST_HEAD(&p->pi_state_list);
1383         p->pi_state_cache = NULL;
1384 #endif
1385         /*
1386          * sigaltstack should be cleared when sharing the same VM
1387          */
1388         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1389                 p->sas_ss_sp = p->sas_ss_size = 0;
1390
1391         /*
1392          * Syscall tracing and stepping should be turned off in the
1393          * child regardless of CLONE_PTRACE.
1394          */
1395         user_disable_single_step(p);
1396         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1397 #ifdef TIF_SYSCALL_EMU
1398         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1399 #endif
1400         clear_all_latency_tracing(p);
1401
1402         /* ok, now we should be set up.. */
1403         p->pid = pid_nr(pid);
1404         if (clone_flags & CLONE_THREAD) {
1405                 p->exit_signal = -1;
1406                 p->group_leader = current->group_leader;
1407                 p->tgid = current->tgid;
1408         } else {
1409                 if (clone_flags & CLONE_PARENT)
1410                         p->exit_signal = current->group_leader->exit_signal;
1411                 else
1412                         p->exit_signal = (clone_flags & CSIGNAL);
1413                 p->group_leader = p;
1414                 p->tgid = p->pid;
1415         }
1416
1417         p->nr_dirtied = 0;
1418         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1419         p->dirty_paused_when = 0;
1420
1421         p->pdeath_signal = 0;
1422         INIT_LIST_HEAD(&p->thread_group);
1423         p->task_works = NULL;
1424
1425         /*
1426          * Make it visible to the rest of the system, but dont wake it up yet.
1427          * Need tasklist lock for parent etc handling!
1428          */
1429         write_lock_irq(&tasklist_lock);
1430
1431         /* CLONE_PARENT re-uses the old parent */
1432         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1433                 p->real_parent = current->real_parent;
1434                 p->parent_exec_id = current->parent_exec_id;
1435         } else {
1436                 p->real_parent = current;
1437                 p->parent_exec_id = current->self_exec_id;
1438         }
1439
1440         spin_lock(&current->sighand->siglock);
1441
1442         /*
1443          * Process group and session signals need to be delivered to just the
1444          * parent before the fork or both the parent and the child after the
1445          * fork. Restart if a signal comes in before we add the new process to
1446          * it's process group.
1447          * A fatal signal pending means that current will exit, so the new
1448          * thread can't slip out of an OOM kill (or normal SIGKILL).
1449         */
1450         recalc_sigpending();
1451         if (signal_pending(current)) {
1452                 spin_unlock(&current->sighand->siglock);
1453                 write_unlock_irq(&tasklist_lock);
1454                 retval = -ERESTARTNOINTR;
1455                 goto bad_fork_free_pid;
1456         }
1457
1458         if (likely(p->pid)) {
1459                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1460
1461                 init_task_pid(p, PIDTYPE_PID, pid);
1462                 if (thread_group_leader(p)) {
1463                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1464                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1465
1466                         if (is_child_reaper(pid)) {
1467                                 ns_of_pid(pid)->child_reaper = p;
1468                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1469                         }
1470
1471                         p->signal->leader_pid = pid;
1472                         p->signal->tty = tty_kref_get(current->signal->tty);
1473                         list_add_tail(&p->sibling, &p->real_parent->children);
1474                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1475                         attach_pid(p, PIDTYPE_PGID);
1476                         attach_pid(p, PIDTYPE_SID);
1477                         __this_cpu_inc(process_counts);
1478                 } else {
1479                         current->signal->nr_threads++;
1480                         atomic_inc(&current->signal->live);
1481                         atomic_inc(&current->signal->sigcnt);
1482                         list_add_tail_rcu(&p->thread_group,
1483                                           &p->group_leader->thread_group);
1484                         list_add_tail_rcu(&p->thread_node,
1485                                           &p->signal->thread_head);
1486                 }
1487                 attach_pid(p, PIDTYPE_PID);
1488                 nr_threads++;
1489         }
1490
1491         total_forks++;
1492         spin_unlock(&current->sighand->siglock);
1493         write_unlock_irq(&tasklist_lock);
1494         proc_fork_connector(p);
1495         cgroup_post_fork(p);
1496         if (clone_flags & CLONE_THREAD)
1497                 threadgroup_change_end(current);
1498         perf_event_fork(p);
1499
1500         trace_task_newtask(p, clone_flags);
1501         uprobe_copy_process(p, clone_flags);
1502
1503         return p;
1504
1505 bad_fork_free_pid:
1506         if (pid != &init_struct_pid)
1507                 free_pid(pid);
1508 bad_fork_cleanup_io:
1509         if (p->io_context)
1510                 exit_io_context(p);
1511 bad_fork_cleanup_namespaces:
1512         exit_task_namespaces(p);
1513 bad_fork_cleanup_mm:
1514         if (p->mm)
1515                 mmput(p->mm);
1516 bad_fork_cleanup_signal:
1517         if (!(clone_flags & CLONE_THREAD))
1518                 free_signal_struct(p->signal);
1519 bad_fork_cleanup_sighand:
1520         __cleanup_sighand(p->sighand);
1521 bad_fork_cleanup_fs:
1522         exit_fs(p); /* blocking */
1523 bad_fork_cleanup_files:
1524         exit_files(p); /* blocking */
1525 bad_fork_cleanup_semundo:
1526         exit_sem(p);
1527 bad_fork_cleanup_audit:
1528         audit_free(p);
1529 bad_fork_cleanup_policy:
1530         perf_event_free_task(p);
1531 #ifdef CONFIG_NUMA
1532         mpol_put(p->mempolicy);
1533 bad_fork_cleanup_threadgroup_lock:
1534 #endif
1535         if (clone_flags & CLONE_THREAD)
1536                 threadgroup_change_end(current);
1537         delayacct_tsk_free(p);
1538         module_put(task_thread_info(p)->exec_domain->module);
1539 bad_fork_cleanup_count:
1540         atomic_dec(&p->cred->user->processes);
1541         exit_creds(p);
1542 bad_fork_free:
1543         free_task(p);
1544 fork_out:
1545         return ERR_PTR(retval);
1546 }
1547
1548 static inline void init_idle_pids(struct pid_link *links)
1549 {
1550         enum pid_type type;
1551
1552         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1553                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1554                 links[type].pid = &init_struct_pid;
1555         }
1556 }
1557
1558 struct task_struct *fork_idle(int cpu)
1559 {
1560         struct task_struct *task;
1561         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0);
1562         if (!IS_ERR(task)) {
1563                 init_idle_pids(task->pids);
1564                 init_idle(task, cpu);
1565         }
1566
1567         return task;
1568 }
1569
1570 /*
1571  *  Ok, this is the main fork-routine.
1572  *
1573  * It copies the process, and if successful kick-starts
1574  * it and waits for it to finish using the VM if required.
1575  */
1576 long do_fork(unsigned long clone_flags,
1577               unsigned long stack_start,
1578               unsigned long stack_size,
1579               int __user *parent_tidptr,
1580               int __user *child_tidptr)
1581 {
1582         struct task_struct *p;
1583         int trace = 0;
1584         long nr;
1585
1586         /*
1587          * Determine whether and which event to report to ptracer.  When
1588          * called from kernel_thread or CLONE_UNTRACED is explicitly
1589          * requested, no event is reported; otherwise, report if the event
1590          * for the type of forking is enabled.
1591          */
1592         if (!(clone_flags & CLONE_UNTRACED)) {
1593                 if (clone_flags & CLONE_VFORK)
1594                         trace = PTRACE_EVENT_VFORK;
1595                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1596                         trace = PTRACE_EVENT_CLONE;
1597                 else
1598                         trace = PTRACE_EVENT_FORK;
1599
1600                 if (likely(!ptrace_event_enabled(current, trace)))
1601                         trace = 0;
1602         }
1603
1604         p = copy_process(clone_flags, stack_start, stack_size,
1605                          child_tidptr, NULL, trace);
1606         /*
1607          * Do this prior waking up the new thread - the thread pointer
1608          * might get invalid after that point, if the thread exits quickly.
1609          */
1610         if (!IS_ERR(p)) {
1611                 struct completion vfork;
1612
1613                 trace_sched_process_fork(current, p);
1614
1615                 nr = task_pid_vnr(p);
1616
1617                 if (clone_flags & CLONE_PARENT_SETTID)
1618                         put_user(nr, parent_tidptr);
1619
1620                 if (clone_flags & CLONE_VFORK) {
1621                         p->vfork_done = &vfork;
1622                         init_completion(&vfork);
1623                         get_task_struct(p);
1624                 }
1625
1626                 wake_up_new_task(p);
1627
1628                 /* forking complete and child started to run, tell ptracer */
1629                 if (unlikely(trace))
1630                         ptrace_event(trace, nr);
1631
1632                 if (clone_flags & CLONE_VFORK) {
1633                         if (!wait_for_vfork_done(p, &vfork))
1634                                 ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1635                 }
1636         } else {
1637                 nr = PTR_ERR(p);
1638         }
1639         return nr;
1640 }
1641
1642 /*
1643  * Create a kernel thread.
1644  */
1645 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1646 {
1647         return do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1648                 (unsigned long)arg, NULL, NULL);
1649 }
1650
1651 #ifdef __ARCH_WANT_SYS_FORK
1652 SYSCALL_DEFINE0(fork)
1653 {
1654 #ifdef CONFIG_MMU
1655         return do_fork(SIGCHLD, 0, 0, NULL, NULL);
1656 #else
1657         /* can not support in nommu mode */
1658         return -EINVAL;
1659 #endif
1660 }
1661 #endif
1662
1663 #ifdef __ARCH_WANT_SYS_VFORK
1664 SYSCALL_DEFINE0(vfork)
1665 {
1666         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1667                         0, NULL, NULL);
1668 }
1669 #endif
1670
1671 #ifdef __ARCH_WANT_SYS_CLONE
1672 #ifdef CONFIG_CLONE_BACKWARDS
1673 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1674                  int __user *, parent_tidptr,
1675                  int, tls_val,
1676                  int __user *, child_tidptr)
1677 #elif defined(CONFIG_CLONE_BACKWARDS2)
1678 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1679                  int __user *, parent_tidptr,
1680                  int __user *, child_tidptr,
1681                  int, tls_val)
1682 #elif defined(CONFIG_CLONE_BACKWARDS3)
1683 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1684                 int, stack_size,
1685                 int __user *, parent_tidptr,
1686                 int __user *, child_tidptr,
1687                 int, tls_val)
1688 #else
1689 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1690                  int __user *, parent_tidptr,
1691                  int __user *, child_tidptr,
1692                  int, tls_val)
1693 #endif
1694 {
1695         return do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr);
1696 }
1697 #endif
1698
1699 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1700 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1701 #endif
1702
1703 static void sighand_ctor(void *data)
1704 {
1705         struct sighand_struct *sighand = data;
1706
1707         spin_lock_init(&sighand->siglock);
1708         init_waitqueue_head(&sighand->signalfd_wqh);
1709 }
1710
1711 void __init proc_caches_init(void)
1712 {
1713         sighand_cachep = kmem_cache_create("sighand_cache",
1714                         sizeof(struct sighand_struct), 0,
1715                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1716                         SLAB_NOTRACK, sighand_ctor);
1717         signal_cachep = kmem_cache_create("signal_cache",
1718                         sizeof(struct signal_struct), 0,
1719                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1720         files_cachep = kmem_cache_create("files_cache",
1721                         sizeof(struct files_struct), 0,
1722                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1723         fs_cachep = kmem_cache_create("fs_cache",
1724                         sizeof(struct fs_struct), 0,
1725                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1726         /*
1727          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1728          * whole struct cpumask for the OFFSTACK case. We could change
1729          * this to *only* allocate as much of it as required by the
1730          * maximum number of CPU's we can ever have.  The cpumask_allocation
1731          * is at the end of the structure, exactly for that reason.
1732          */
1733         mm_cachep = kmem_cache_create("mm_struct",
1734                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1735                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1736         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1737         mmap_init();
1738         nsproxy_cache_init();
1739 }
1740
1741 /*
1742  * Check constraints on flags passed to the unshare system call.
1743  */
1744 static int check_unshare_flags(unsigned long unshare_flags)
1745 {
1746         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1747                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1748                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1749                                 CLONE_NEWUSER|CLONE_NEWPID))
1750                 return -EINVAL;
1751         /*
1752          * Not implemented, but pretend it works if there is nothing to
1753          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1754          * needs to unshare vm.
1755          */
1756         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1757                 /* FIXME: get_task_mm() increments ->mm_users */
1758                 if (atomic_read(&current->mm->mm_users) > 1)
1759                         return -EINVAL;
1760         }
1761
1762         return 0;
1763 }
1764
1765 /*
1766  * Unshare the filesystem structure if it is being shared
1767  */
1768 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1769 {
1770         struct fs_struct *fs = current->fs;
1771
1772         if (!(unshare_flags & CLONE_FS) || !fs)
1773                 return 0;
1774
1775         /* don't need lock here; in the worst case we'll do useless copy */
1776         if (fs->users == 1)
1777                 return 0;
1778
1779         *new_fsp = copy_fs_struct(fs);
1780         if (!*new_fsp)
1781                 return -ENOMEM;
1782
1783         return 0;
1784 }
1785
1786 /*
1787  * Unshare file descriptor table if it is being shared
1788  */
1789 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1790 {
1791         struct files_struct *fd = current->files;
1792         int error = 0;
1793
1794         if ((unshare_flags & CLONE_FILES) &&
1795             (fd && atomic_read(&fd->count) > 1)) {
1796                 *new_fdp = dup_fd(fd, &error);
1797                 if (!*new_fdp)
1798                         return error;
1799         }
1800
1801         return 0;
1802 }
1803
1804 /*
1805  * unshare allows a process to 'unshare' part of the process
1806  * context which was originally shared using clone.  copy_*
1807  * functions used by do_fork() cannot be used here directly
1808  * because they modify an inactive task_struct that is being
1809  * constructed. Here we are modifying the current, active,
1810  * task_struct.
1811  */
1812 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1813 {
1814         struct fs_struct *fs, *new_fs = NULL;
1815         struct files_struct *fd, *new_fd = NULL;
1816         struct cred *new_cred = NULL;
1817         struct nsproxy *new_nsproxy = NULL;
1818         int do_sysvsem = 0;
1819         int err;
1820
1821         /*
1822          * If unsharing a user namespace must also unshare the thread.
1823          */
1824         if (unshare_flags & CLONE_NEWUSER)
1825                 unshare_flags |= CLONE_THREAD | CLONE_FS;
1826         /*
1827          * If unsharing a thread from a thread group, must also unshare vm.
1828          */
1829         if (unshare_flags & CLONE_THREAD)
1830                 unshare_flags |= CLONE_VM;
1831         /*
1832          * If unsharing vm, must also unshare signal handlers.
1833          */
1834         if (unshare_flags & CLONE_VM)
1835                 unshare_flags |= CLONE_SIGHAND;
1836         /*
1837          * If unsharing namespace, must also unshare filesystem information.
1838          */
1839         if (unshare_flags & CLONE_NEWNS)
1840                 unshare_flags |= CLONE_FS;
1841
1842         err = check_unshare_flags(unshare_flags);
1843         if (err)
1844                 goto bad_unshare_out;
1845         /*
1846          * CLONE_NEWIPC must also detach from the undolist: after switching
1847          * to a new ipc namespace, the semaphore arrays from the old
1848          * namespace are unreachable.
1849          */
1850         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1851                 do_sysvsem = 1;
1852         err = unshare_fs(unshare_flags, &new_fs);
1853         if (err)
1854                 goto bad_unshare_out;
1855         err = unshare_fd(unshare_flags, &new_fd);
1856         if (err)
1857                 goto bad_unshare_cleanup_fs;
1858         err = unshare_userns(unshare_flags, &new_cred);
1859         if (err)
1860                 goto bad_unshare_cleanup_fd;
1861         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1862                                          new_cred, new_fs);
1863         if (err)
1864                 goto bad_unshare_cleanup_cred;
1865
1866         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
1867                 if (do_sysvsem) {
1868                         /*
1869                          * CLONE_SYSVSEM is equivalent to sys_exit().
1870                          */
1871                         exit_sem(current);
1872                 }
1873
1874                 if (new_nsproxy)
1875                         switch_task_namespaces(current, new_nsproxy);
1876
1877                 task_lock(current);
1878
1879                 if (new_fs) {
1880                         fs = current->fs;
1881                         spin_lock(&fs->lock);
1882                         current->fs = new_fs;
1883                         if (--fs->users)
1884                                 new_fs = NULL;
1885                         else
1886                                 new_fs = fs;
1887                         spin_unlock(&fs->lock);
1888                 }
1889
1890                 if (new_fd) {
1891                         fd = current->files;
1892                         current->files = new_fd;
1893                         new_fd = fd;
1894                 }
1895
1896                 task_unlock(current);
1897
1898                 if (new_cred) {
1899                         /* Install the new user namespace */
1900                         commit_creds(new_cred);
1901                         new_cred = NULL;
1902                 }
1903         }
1904
1905 bad_unshare_cleanup_cred:
1906         if (new_cred)
1907                 put_cred(new_cred);
1908 bad_unshare_cleanup_fd:
1909         if (new_fd)
1910                 put_files_struct(new_fd);
1911
1912 bad_unshare_cleanup_fs:
1913         if (new_fs)
1914                 free_fs_struct(new_fs);
1915
1916 bad_unshare_out:
1917         return err;
1918 }
1919
1920 /*
1921  *      Helper to unshare the files of the current task.
1922  *      We don't want to expose copy_files internals to
1923  *      the exec layer of the kernel.
1924  */
1925
1926 int unshare_files(struct files_struct **displaced)
1927 {
1928         struct task_struct *task = current;
1929         struct files_struct *copy = NULL;
1930         int error;
1931
1932         error = unshare_fd(CLONE_FILES, &copy);
1933         if (error || !copy) {
1934                 *displaced = NULL;
1935                 return error;
1936         }
1937         *displaced = task->files;
1938         task_lock(task);
1939         task->files = copy;
1940         task_unlock(task);
1941         return 0;
1942 }