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