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 #include <linux/compiler.h>
  77
  78 #include <asm/pgtable.h>
  79 #include <asm/pgalloc.h>
  80 #include <asm/uaccess.h>
  81 #include <asm/mmu_context.h>
  82 #include <asm/cacheflush.h>
  83 #include <asm/tlbflush.h>
  84
  85 #include <trace/events/sched.h>
  86
  87 #define CREATE_TRACE_POINTS
  88 #include <trace/events/task.h>
  89
  90 /*
  91  * Protected counters by write_lock_irq(&tasklist_lock)
  92  */
  93 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  94 int nr_threads;                 /* The idle threads do not count.. */
  95
  96 int max_threads;                /* tunable limit on nr_threads */
  97
  98 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  99
 100 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 101
 102 #ifdef CONFIG_PROVE_RCU
 103 int lockdep_tasklist_lock_is_held(void)
 104 {
 105         return lockdep_is_held(&tasklist_lock);
 106 }
 107 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 108 #endif /* #ifdef CONFIG_PROVE_RCU */
 109
 110 int nr_processes(void)
 111 {
 112         int cpu;
 113         int total = 0;
 114
 115         for_each_possible_cpu(cpu)
 116                 total += per_cpu(process_counts, cpu);
 117
 118         return total;
 119 }
 120
 121 void __weak arch_release_task_struct(struct task_struct *tsk)
 122 {
 123 }
 124
 125 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 126 static struct kmem_cache *task_struct_cachep;
 127
 128 static inline struct task_struct *alloc_task_struct_node(int node)
 129 {
 130         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 131 }
 132
 133 static inline void free_task_struct(struct task_struct *tsk)
 134 {
 135         kmem_cache_free(task_struct_cachep, tsk);
 136 }
 137 #endif
 138
 139 void __weak arch_release_thread_info(struct thread_info *ti)
 140 {
 141 }
 142
 143 #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
 144
 145 /*
 146  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 147  * kmemcache based allocator.
 148  */
 149 # if THREAD_SIZE >= PAGE_SIZE
 150 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 151                                                   int node)
 152 {
 153         struct page *page = alloc_kmem_pages_node(node, THREADINFO_GFP,
 154                                                   THREAD_SIZE_ORDER);
 155
 156         return page ? page_address(page) : NULL;
 157 }
 158
 159 static inline void free_thread_info(struct thread_info *ti)
 160 {
 161         free_kmem_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 162 }
 163 # else
 164 static struct kmem_cache *thread_info_cache;
 165
 166 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 167                                                   int node)
 168 {
 169         return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
 170 }
 171
 172 static void free_thread_info(struct thread_info *ti)
 173 {
 174         kmem_cache_free(thread_info_cache, ti);
 175 }
 176
 177 void thread_info_cache_init(void)
 178 {
 179         thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
 180                                               THREAD_SIZE, 0, NULL);
 181         BUG_ON(thread_info_cache == NULL);
 182 }
 183 # endif
 184 #endif
 185
 186 /* SLAB cache for signal_struct structures (tsk->signal) */
 187 static struct kmem_cache *signal_cachep;
 188
 189 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 190 struct kmem_cache *sighand_cachep;
 191
 192 /* SLAB cache for files_struct structures (tsk->files) */
 193 struct kmem_cache *files_cachep;
 194
 195 /* SLAB cache for fs_struct structures (tsk->fs) */
 196 struct kmem_cache *fs_cachep;
 197
 198 /* SLAB cache for vm_area_struct structures */
 199 struct kmem_cache *vm_area_cachep;
 200
 201 /* SLAB cache for mm_struct structures (tsk->mm) */
 202 static struct kmem_cache *mm_cachep;
 203
 204 static void account_kernel_stack(struct thread_info *ti, int account)
 205 {
 206         struct zone *zone = page_zone(virt_to_page(ti));
 207
 208         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
 209 }
 210
 211 void free_task(struct task_struct *tsk)
 212 {
 213         account_kernel_stack(tsk->stack, -1);
 214         arch_release_thread_info(tsk->stack);
 215         free_thread_info(tsk->stack);
 216         rt_mutex_debug_task_free(tsk);
 217         ftrace_graph_exit_task(tsk);
 218         put_seccomp_filter(tsk);
 219         arch_release_task_struct(tsk);
 220         free_task_struct(tsk);
 221 }
 222 EXPORT_SYMBOL(free_task);
 223
 224 static inline void free_signal_struct(struct signal_struct *sig)
 225 {
 226         taskstats_tgid_free(sig);
 227         sched_autogroup_exit(sig);
 228         kmem_cache_free(signal_cachep, sig);
 229 }
 230
 231 static inline void put_signal_struct(struct signal_struct *sig)
 232 {
 233         if (atomic_dec_and_test(&sig->sigcnt))
 234                 free_signal_struct(sig);
 235 }
 236
 237 void __put_task_struct(struct task_struct *tsk)
 238 {
 239         WARN_ON(!tsk->exit_state);
 240         WARN_ON(atomic_read(&tsk->usage));
 241         WARN_ON(tsk == current);
 242
 243         task_numa_free(tsk);
 244         security_task_free(tsk);
 245         exit_creds(tsk);
 246         delayacct_tsk_free(tsk);
 247         put_signal_struct(tsk->signal);
 248
 249         if (!profile_handoff_task(tsk))
 250                 free_task(tsk);
 251 }
 252 EXPORT_SYMBOL_GPL(__put_task_struct);
 253
 254 void __init __weak arch_task_cache_init(void) { }
 255
 256 void __init fork_init(unsigned long mempages)
 257 {
 258 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 259 #ifndef ARCH_MIN_TASKALIGN
 260 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 261 #endif
 262         /* create a slab on which task_structs can be allocated */
 263         task_struct_cachep =
 264                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 265                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
 266 #endif
 267
 268         /* do the arch specific task caches init */
 269         arch_task_cache_init();
 270
 271         /*
 272          * The default maximum number of threads is set to a safe
 273          * value: the thread structures can take up at most half
 274          * of memory.
 275          */
 276         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 277
 278         /*
 279          * we need to allow at least 20 threads to boot a system
 280          */
 281         if (max_threads < 20)
 282                 max_threads = 20;
 283
 284         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 285         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 286         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 287                 init_task.signal->rlim[RLIMIT_NPROC];
 288 }
 289
 290 int __weak arch_dup_task_struct(struct task_struct *dst,
 291                                                struct task_struct *src)
 292 {
 293         *dst = *src;
 294         return 0;
 295 }
 296
 297 static struct task_struct *dup_task_struct(struct task_struct *orig)
 298 {
 299         struct task_struct *tsk;
 300         struct thread_info *ti;
 301         unsigned long *stackend;
 302         int node = tsk_fork_get_node(orig);
 303         int err;
 304
 305         tsk = alloc_task_struct_node(node);
 306         if (!tsk)
 307                 return NULL;
 308
 309         ti = alloc_thread_info_node(tsk, node);
 310         if (!ti)
 311                 goto free_tsk;
 312
 313         err = arch_dup_task_struct(tsk, orig);
 314         if (err)
 315                 goto free_ti;
 316
 317         tsk->stack = ti;
 318
 319         setup_thread_stack(tsk, orig);
 320         clear_user_return_notifier(tsk);
 321         clear_tsk_need_resched(tsk);
 322         stackend = end_of_stack(tsk);
 323         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 324
 325 #ifdef CONFIG_CC_STACKPROTECTOR
 326         tsk->stack_canary = get_random_int();
 327 #endif
 328
 329         /*
 330          * One for us, one for whoever does the "release_task()" (usually
 331          * parent)
 332          */
 333         atomic_set(&tsk->usage, 2);
 334 #ifdef CONFIG_BLK_DEV_IO_TRACE
 335         tsk->btrace_seq = 0;
 336 #endif
 337         tsk->splice_pipe = NULL;
 338         tsk->task_frag.page = NULL;
 339
 340         account_kernel_stack(ti, 1);
 341
 342         return tsk;
 343
 344 free_ti:
 345         free_thread_info(ti);
 346 free_tsk:
 347         free_task_struct(tsk);
 348         return NULL;
 349 }
 350
 351 #ifdef CONFIG_MMU
 352 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 353 {
 354         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 355         struct rb_node **rb_link, *rb_parent;
 356         int retval;
 357         unsigned long charge;
 358
 359         uprobe_start_dup_mmap();
 360         down_write(&oldmm->mmap_sem);
 361         flush_cache_dup_mm(oldmm);
 362         uprobe_dup_mmap(oldmm, mm);
 363         /*
 364          * Not linked in yet - no deadlock potential:
 365          */
 366         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 367
 368         mm->locked_vm = 0;
 369         mm->mmap = NULL;
 370         mm->vmacache_seqnum = 0;
 371         mm->map_count = 0;
 372         cpumask_clear(mm_cpumask(mm));
 373         mm->mm_rb = RB_ROOT;
 374         rb_link = &mm->mm_rb.rb_node;
 375         rb_parent = NULL;
 376         pprev = &mm->mmap;
 377         retval = ksm_fork(mm, oldmm);
 378         if (retval)
 379                 goto out;
 380         retval = khugepaged_fork(mm, oldmm);
 381         if (retval)
 382                 goto out;
 383
 384         prev = NULL;
 385         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 386                 struct file *file;
 387
 388                 if (mpnt->vm_flags & VM_DONTCOPY) {
 389                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 390                                                         -vma_pages(mpnt));
 391                         continue;
 392                 }
 393                 charge = 0;
 394                 if (mpnt->vm_flags & VM_ACCOUNT) {
 395                         unsigned long len = vma_pages(mpnt);
 396
 397                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 398                                 goto fail_nomem;
 399                         charge = len;
 400                 }
 401                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 402                 if (!tmp)
 403                         goto fail_nomem;
 404                 *tmp = *mpnt;
 405                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 406                 retval = vma_dup_policy(mpnt, tmp);
 407                 if (retval)
 408                         goto fail_nomem_policy;
 409                 tmp->vm_mm = mm;
 410                 if (anon_vma_fork(tmp, mpnt))
 411                         goto fail_nomem_anon_vma_fork;
 412                 tmp->vm_flags &= ~VM_LOCKED;
 413                 tmp->vm_next = tmp->vm_prev = NULL;
 414                 file = tmp->vm_file;
 415                 if (file) {
 416                         struct inode *inode = file_inode(file);
 417                         struct address_space *mapping = file->f_mapping;
 418
 419                         get_file(file);
 420                         if (tmp->vm_flags & VM_DENYWRITE)
 421                                 atomic_dec(&inode->i_writecount);
 422                         mutex_lock(&mapping->i_mmap_mutex);
 423                         if (tmp->vm_flags & VM_SHARED)
 424                                 mapping->i_mmap_writable++;
 425                         flush_dcache_mmap_lock(mapping);
 426                         /* insert tmp into the share list, just after mpnt */
 427                         if (unlikely(tmp->vm_flags & VM_NONLINEAR))
 428                                 vma_nonlinear_insert(tmp,
 429                                                 &mapping->i_mmap_nonlinear);
 430                         else
 431                                 vma_interval_tree_insert_after(tmp, mpnt,
 432                                                         &mapping->i_mmap);
 433                         flush_dcache_mmap_unlock(mapping);
 434                         mutex_unlock(&mapping->i_mmap_mutex);
 435                 }
 436
 437                 /*
 438                  * Clear hugetlb-related page reserves for children. This only
 439                  * affects MAP_PRIVATE mappings. Faults generated by the child
 440                  * are not guaranteed to succeed, even if read-only
 441                  */
 442                 if (is_vm_hugetlb_page(tmp))
 443                         reset_vma_resv_huge_pages(tmp);
 444
 445                 /*
 446                  * Link in the new vma and copy the page table entries.
 447                  */
 448                 *pprev = tmp;
 449                 pprev = &tmp->vm_next;
 450                 tmp->vm_prev = prev;
 451                 prev = tmp;
 452
 453                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 454                 rb_link = &tmp->vm_rb.rb_right;
 455                 rb_parent = &tmp->vm_rb;
 456
 457                 mm->map_count++;
 458                 retval = copy_page_range(mm, oldmm, mpnt);
 459
 460                 if (tmp->vm_ops && tmp->vm_ops->open)
 461                         tmp->vm_ops->open(tmp);
 462
 463                 if (retval)
 464                         goto out;
 465         }
 466         /* a new mm has just been created */
 467         arch_dup_mmap(oldmm, mm);
 468         retval = 0;
 469 out:
 470         up_write(&mm->mmap_sem);
 471         flush_tlb_mm(oldmm);
 472         up_write(&oldmm->mmap_sem);
 473         uprobe_end_dup_mmap();
 474         return retval;
 475 fail_nomem_anon_vma_fork:
 476         mpol_put(vma_policy(tmp));
 477 fail_nomem_policy:
 478         kmem_cache_free(vm_area_cachep, tmp);
 479 fail_nomem:
 480         retval = -ENOMEM;
 481         vm_unacct_memory(charge);
 482         goto out;
 483 }
 484
 485 static inline int mm_alloc_pgd(struct mm_struct *mm)
 486 {
 487         mm->pgd = pgd_alloc(mm);
 488         if (unlikely(!mm->pgd))
 489                 return -ENOMEM;
 490         return 0;
 491 }
 492
 493 static inline void mm_free_pgd(struct mm_struct *mm)
 494 {
 495         pgd_free(mm, mm->pgd);
 496 }
 497 #else
 498 #define dup_mmap(mm, oldmm)     (0)
 499 #define mm_alloc_pgd(mm)        (0)
 500 #define mm_free_pgd(mm)
 501 #endif /* CONFIG_MMU */
 502
 503 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 504
 505 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 506 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 507
 508 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 509
 510 static int __init coredump_filter_setup(char *s)
 511 {
 512         default_dump_filter =
 513                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 514                 MMF_DUMP_FILTER_MASK;
 515         return 1;
 516 }
 517
 518 __setup("coredump_filter=", coredump_filter_setup);
 519
 520 #include <linux/init_task.h>
 521
 522 static void mm_init_aio(struct mm_struct *mm)
 523 {
 524 #ifdef CONFIG_AIO
 525         spin_lock_init(&mm->ioctx_lock);
 526         mm->ioctx_table = NULL;
 527 #endif
 528 }
 529
 530 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
 531 {
 532         atomic_set(&mm->mm_users, 1);
 533         atomic_set(&mm->mm_count, 1);
 534         init_rwsem(&mm->mmap_sem);
 535         INIT_LIST_HEAD(&mm->mmlist);
 536         mm->core_state = NULL;
 537         atomic_long_set(&mm->nr_ptes, 0);
 538         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 539         spin_lock_init(&mm->page_table_lock);
 540         mm_init_aio(mm);
 541         mm_init_owner(mm, p);
 542         clear_tlb_flush_pending(mm);
 543
 544         if (current->mm) {
 545                 mm->flags = current->mm->flags & MMF_INIT_MASK;
 546                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
 547         } else {
 548                 mm->flags = default_dump_filter;
 549                 mm->def_flags = 0;
 550         }
 551
 552         if (likely(!mm_alloc_pgd(mm))) {
 553                 mmu_notifier_mm_init(mm);
 554                 return mm;
 555         }
 556
 557         free_mm(mm);
 558         return NULL;
 559 }
 560
 561 static void check_mm(struct mm_struct *mm)
 562 {
 563         int i;
 564
 565         for (i = 0; i < NR_MM_COUNTERS; i++) {
 566                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 567
 568                 if (unlikely(x))
 569                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 570                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 571         }
 572
 573 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 574         VM_BUG_ON(mm->pmd_huge_pte);
 575 #endif
 576 }
 577
 578 /*
 579  * Allocate and initialize an mm_struct.
 580  */
 581 struct mm_struct *mm_alloc(void)
 582 {
 583         struct mm_struct *mm;
 584
 585         mm = allocate_mm();
 586         if (!mm)
 587                 return NULL;
 588
 589         memset(mm, 0, sizeof(*mm));
 590         mm_init_cpumask(mm);
 591         return mm_init(mm, current);
 592 }
 593
 594 /*
 595  * Called when the last reference to the mm
 596  * is dropped: either by a lazy thread or by
 597  * mmput. Free the page directory and the mm.
 598  */
 599 void __mmdrop(struct mm_struct *mm)
 600 {
 601         BUG_ON(mm == &init_mm);
 602         mm_free_pgd(mm);
 603         destroy_context(mm);
 604         mmu_notifier_mm_destroy(mm);
 605         check_mm(mm);
 606         free_mm(mm);
 607 }
 608 EXPORT_SYMBOL_GPL(__mmdrop);
 609
 610 /*
 611  * Decrement the use count and release all resources for an mm.
 612  */
 613 void mmput(struct mm_struct *mm)
 614 {
 615         might_sleep();
 616
 617         if (atomic_dec_and_test(&mm->mm_users)) {
 618                 uprobe_clear_state(mm);
 619                 exit_aio(mm);
 620                 ksm_exit(mm);
 621                 khugepaged_exit(mm); /* must run before exit_mmap */
 622                 exit_mmap(mm);
 623                 set_mm_exe_file(mm, NULL);
 624                 if (!list_empty(&mm->mmlist)) {
 625                         spin_lock(&mmlist_lock);
 626                         list_del(&mm->mmlist);
 627                         spin_unlock(&mmlist_lock);
 628                 }
 629                 if (mm->binfmt)
 630                         module_put(mm->binfmt->module);
 631                 mmdrop(mm);
 632         }
 633 }
 634 EXPORT_SYMBOL_GPL(mmput);
 635
 636 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 637 {
 638         if (new_exe_file)
 639                 get_file(new_exe_file);
 640         if (mm->exe_file)
 641                 fput(mm->exe_file);
 642         mm->exe_file = new_exe_file;
 643 }
 644
 645 struct file *get_mm_exe_file(struct mm_struct *mm)
 646 {
 647         struct file *exe_file;
 648
 649         /* We need mmap_sem to protect against races with removal of exe_file */
 650         down_read(&mm->mmap_sem);
 651         exe_file = mm->exe_file;
 652         if (exe_file)
 653                 get_file(exe_file);
 654         up_read(&mm->mmap_sem);
 655         return exe_file;
 656 }
 657
 658 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
 659 {
 660         /* It's safe to write the exe_file pointer without exe_file_lock because
 661          * this is called during fork when the task is not yet in /proc */
 662         newmm->exe_file = get_mm_exe_file(oldmm);
 663 }
 664
 665 /**
 666  * get_task_mm - acquire a reference to the task's mm
 667  *
 668  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 669  * this kernel workthread has transiently adopted a user mm with use_mm,
 670  * to do its AIO) is not set and if so returns a reference to it, after
 671  * bumping up the use count.  User must release the mm via mmput()
 672  * after use.  Typically used by /proc and ptrace.
 673  */
 674 struct mm_struct *get_task_mm(struct task_struct *task)
 675 {
 676         struct mm_struct *mm;
 677
 678         task_lock(task);
 679         mm = task->mm;
 680         if (mm) {
 681                 if (task->flags & PF_KTHREAD)
 682                         mm = NULL;
 683                 else
 684                         atomic_inc(&mm->mm_users);
 685         }
 686         task_unlock(task);
 687         return mm;
 688 }
 689 EXPORT_SYMBOL_GPL(get_task_mm);
 690
 691 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 692 {
 693         struct mm_struct *mm;
 694         int err;
 695
 696         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 697         if (err)
 698                 return ERR_PTR(err);
 699
 700         mm = get_task_mm(task);
 701         if (mm && mm != current->mm &&
 702                         !ptrace_may_access(task, mode)) {
 703                 mmput(mm);
 704                 mm = ERR_PTR(-EACCES);
 705         }
 706         mutex_unlock(&task->signal->cred_guard_mutex);
 707
 708         return mm;
 709 }
 710
 711 static void complete_vfork_done(struct task_struct *tsk)
 712 {
 713         struct completion *vfork;
 714
 715         task_lock(tsk);
 716         vfork = tsk->vfork_done;
 717         if (likely(vfork)) {
 718                 tsk->vfork_done = NULL;
 719                 complete(vfork);
 720         }
 721         task_unlock(tsk);
 722 }
 723
 724 static int wait_for_vfork_done(struct task_struct *child,
 725                                 struct completion *vfork)
 726 {
 727         int killed;
 728
 729         freezer_do_not_count();
 730         killed = wait_for_completion_killable(vfork);
 731         freezer_count();
 732
 733         if (killed) {
 734                 task_lock(child);
 735                 child->vfork_done = NULL;
 736                 task_unlock(child);
 737         }
 738
 739         put_task_struct(child);
 740         return killed;
 741 }
 742
 743 /* Please note the differences between mmput and mm_release.
 744  * mmput is called whenever we stop holding onto a mm_struct,
 745  * error success whatever.
 746  *
 747  * mm_release is called after a mm_struct has been removed
 748  * from the current process.
 749  *
 750  * This difference is important for error handling, when we
 751  * only half set up a mm_struct for a new process and need to restore
 752  * the old one.  Because we mmput the new mm_struct before
 753  * restoring the old one. . .
 754  * Eric Biederman 10 January 1998
 755  */
 756 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 757 {
 758         /* Get rid of any futexes when releasing the mm */
 759 #ifdef CONFIG_FUTEX
 760         if (unlikely(tsk->robust_list)) {
 761                 exit_robust_list(tsk);
 762                 tsk->robust_list = NULL;
 763         }
 764 #ifdef CONFIG_COMPAT
 765         if (unlikely(tsk->compat_robust_list)) {
 766                 compat_exit_robust_list(tsk);
 767                 tsk->compat_robust_list = NULL;
 768         }
 769 #endif
 770         if (unlikely(!list_empty(&tsk->pi_state_list)))
 771                 exit_pi_state_list(tsk);
 772 #endif
 773
 774         uprobe_free_utask(tsk);
 775
 776         /* Get rid of any cached register state */
 777         deactivate_mm(tsk, mm);
 778
 779         /*
 780          * If we're exiting normally, clear a user-space tid field if
 781          * requested.  We leave this alone when dying by signal, to leave
 782          * the value intact in a core dump, and to save the unnecessary
 783          * trouble, say, a killed vfork parent shouldn't touch this mm.
 784          * Userland only wants this done for a sys_exit.
 785          */
 786         if (tsk->clear_child_tid) {
 787                 if (!(tsk->flags & PF_SIGNALED) &&
 788                     atomic_read(&mm->mm_users) > 1) {
 789                         /*
 790                          * We don't check the error code - if userspace has
 791                          * not set up a proper pointer then tough luck.
 792                          */
 793                         put_user(0, tsk->clear_child_tid);
 794                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 795                                         1, NULL, NULL, 0);
 796                 }
 797                 tsk->clear_child_tid = NULL;
 798         }
 799
 800         /*
 801          * All done, finally we can wake up parent and return this mm to him.
 802          * Also kthread_stop() uses this completion for synchronization.
 803          */
 804         if (tsk->vfork_done)
 805                 complete_vfork_done(tsk);
 806 }
 807
 808 /*
 809  * Allocate a new mm structure and copy contents from the
 810  * mm structure of the passed in task structure.
 811  */
 812 static struct mm_struct *dup_mm(struct task_struct *tsk)
 813 {
 814         struct mm_struct *mm, *oldmm = current->mm;
 815         int err;
 816
 817         mm = allocate_mm();
 818         if (!mm)
 819                 goto fail_nomem;
 820
 821         memcpy(mm, oldmm, sizeof(*mm));
 822         mm_init_cpumask(mm);
 823
 824 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 825         mm->pmd_huge_pte = NULL;
 826 #endif
 827         if (!mm_init(mm, tsk))
 828                 goto fail_nomem;
 829
 830         if (init_new_context(tsk, mm))
 831                 goto fail_nocontext;
 832
 833         dup_mm_exe_file(oldmm, mm);
 834
 835         err = dup_mmap(mm, oldmm);
 836         if (err)
 837                 goto free_pt;
 838
 839         mm->hiwater_rss = get_mm_rss(mm);
 840         mm->hiwater_vm = mm->total_vm;
 841
 842         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 843                 goto free_pt;
 844
 845         return mm;
 846
 847 free_pt:
 848         /* don't put binfmt in mmput, we haven't got module yet */
 849         mm->binfmt = NULL;
 850         mmput(mm);
 851
 852 fail_nomem:
 853         return NULL;
 854
 855 fail_nocontext:
 856         /*
 857          * If init_new_context() failed, we cannot use mmput() to free the mm
 858          * because it calls destroy_context()
 859          */
 860         mm_free_pgd(mm);
 861         free_mm(mm);
 862         return NULL;
 863 }
 864
 865 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 866 {
 867         struct mm_struct *mm, *oldmm;
 868         int retval;
 869
 870         tsk->min_flt = tsk->maj_flt = 0;
 871         tsk->nvcsw = tsk->nivcsw = 0;
 872 #ifdef CONFIG_DETECT_HUNG_TASK
 873         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 874 #endif
 875
 876         tsk->mm = NULL;
 877         tsk->active_mm = NULL;
 878
 879         /*
 880          * Are we cloning a kernel thread?
 881          *
 882          * We need to steal a active VM for that..
 883          */
 884         oldmm = current->mm;
 885         if (!oldmm)
 886                 return 0;
 887
 888         /* initialize the new vmacache entries */
 889         vmacache_flush(tsk);
 890
 891         if (clone_flags & CLONE_VM) {
 892                 atomic_inc(&oldmm->mm_users);
 893                 mm = oldmm;
 894                 goto good_mm;
 895         }
 896
 897         retval = -ENOMEM;
 898         mm = dup_mm(tsk);
 899         if (!mm)
 900                 goto fail_nomem;
 901
 902 good_mm:
 903         tsk->mm = mm;
 904         tsk->active_mm = mm;
 905         return 0;
 906
 907 fail_nomem:
 908         return retval;
 909 }
 910
 911 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 912 {
 913         struct fs_struct *fs = current->fs;
 914         if (clone_flags & CLONE_FS) {
 915                 /* tsk->fs is already what we want */
 916                 spin_lock(&fs->lock);
 917                 if (fs->in_exec) {
 918                         spin_unlock(&fs->lock);
 919                         return -EAGAIN;
 920                 }
 921                 fs->users++;
 922                 spin_unlock(&fs->lock);
 923                 return 0;
 924         }
 925         tsk->fs = copy_fs_struct(fs);
 926         if (!tsk->fs)
 927                 return -ENOMEM;
 928         return 0;
 929 }
 930
 931 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
 932 {
 933         struct files_struct *oldf, *newf;
 934         int error = 0;
 935
 936         /*
 937          * A background process may not have any files ...
 938          */
 939         oldf = current->files;
 940         if (!oldf)
 941                 goto out;
 942
 943         if (clone_flags & CLONE_FILES) {
 944                 atomic_inc(&oldf->count);
 945                 goto out;
 946         }
 947
 948         newf = dup_fd(oldf, &error);
 949         if (!newf)
 950                 goto out;
 951
 952         tsk->files = newf;
 953         error = 0;
 954 out:
 955         return error;
 956 }
 957
 958 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 959 {
 960 #ifdef CONFIG_BLOCK
 961         struct io_context *ioc = current->io_context;
 962         struct io_context *new_ioc;
 963
 964         if (!ioc)
 965                 return 0;
 966         /*
 967          * Share io context with parent, if CLONE_IO is set
 968          */
 969         if (clone_flags & CLONE_IO) {
 970                 ioc_task_link(ioc);
 971                 tsk->io_context = ioc;
 972         } else if (ioprio_valid(ioc->ioprio)) {
 973                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
 974                 if (unlikely(!new_ioc))
 975                         return -ENOMEM;
 976
 977                 new_ioc->ioprio = ioc->ioprio;
 978                 put_io_context(new_ioc);
 979         }
 980 #endif
 981         return 0;
 982 }
 983
 984 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 985 {
 986         struct sighand_struct *sig;
 987
 988         if (clone_flags & CLONE_SIGHAND) {
 989                 atomic_inc(&current->sighand->count);
 990                 return 0;
 991         }
 992         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 993         rcu_assign_pointer(tsk->sighand, sig);
 994         if (!sig)
 995                 return -ENOMEM;
 996         atomic_set(&sig->count, 1);
 997         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 998         return 0;
 999 }
1000
1001 void __cleanup_sighand(struct sighand_struct *sighand)
1002 {
1003         if (atomic_dec_and_test(&sighand->count)) {
1004                 signalfd_cleanup(sighand);
1005                 kmem_cache_free(sighand_cachep, sighand);
1006         }
1007 }
1008
1009
1010 /*
1011  * Initialize POSIX timer handling for a thread group.
1012  */
1013 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1014 {
1015         unsigned long cpu_limit;
1016
1017         /* Thread group counters. */
1018         thread_group_cputime_init(sig);
1019
1020         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1021         if (cpu_limit != RLIM_INFINITY) {
1022                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1023                 sig->cputimer.running = 1;
1024         }
1025
1026         /* The timer lists. */
1027         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1028         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1029         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1030 }
1031
1032 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1033 {
1034         struct signal_struct *sig;
1035
1036         if (clone_flags & CLONE_THREAD)
1037                 return 0;
1038
1039         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1040         tsk->signal = sig;
1041         if (!sig)
1042                 return -ENOMEM;
1043
1044         sig->nr_threads = 1;
1045         atomic_set(&sig->live, 1);
1046         atomic_set(&sig->sigcnt, 1);
1047
1048         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1049         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1050         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1051
1052         init_waitqueue_head(&sig->wait_chldexit);
1053         sig->curr_target = tsk;
1054         init_sigpending(&sig->shared_pending);
1055         INIT_LIST_HEAD(&sig->posix_timers);
1056
1057         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1058         sig->real_timer.function = it_real_fn;
1059
1060         task_lock(current->group_leader);
1061         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1062         task_unlock(current->group_leader);
1063
1064         posix_cpu_timers_init_group(sig);
1065
1066         tty_audit_fork(sig);
1067         sched_autogroup_fork(sig);
1068
1069 #ifdef CONFIG_CGROUPS
1070         init_rwsem(&sig->group_rwsem);
1071 #endif
1072
1073         sig->oom_score_adj = current->signal->oom_score_adj;
1074         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1075
1076         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1077                                    current->signal->is_child_subreaper;
1078
1079         mutex_init(&sig->cred_guard_mutex);
1080
1081         return 0;
1082 }
1083
1084 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1085 {
1086         current->clear_child_tid = tidptr;
1087
1088         return task_pid_vnr(current);
1089 }
1090
1091 static void rt_mutex_init_task(struct task_struct *p)
1092 {
1093         raw_spin_lock_init(&p->pi_lock);
1094 #ifdef CONFIG_RT_MUTEXES
1095         p->pi_waiters = RB_ROOT;
1096         p->pi_waiters_leftmost = NULL;
1097         p->pi_blocked_on = NULL;
1098         p->pi_top_task = NULL;
1099 #endif
1100 }
1101
1102 #ifdef CONFIG_MEMCG
1103 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1104 {
1105         mm->owner = p;
1106 }
1107 #endif /* CONFIG_MEMCG */
1108
1109 /*
1110  * Initialize POSIX timer handling for a single task.
1111  */
1112 static void posix_cpu_timers_init(struct task_struct *tsk)
1113 {
1114         tsk->cputime_expires.prof_exp = 0;
1115         tsk->cputime_expires.virt_exp = 0;
1116         tsk->cputime_expires.sched_exp = 0;
1117         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1118         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1119         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1120 }
1121
1122 static inline void
1123 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1124 {
1125          task->pids[type].pid = pid;
1126 }
1127
1128 /*
1129  * This creates a new process as a copy of the old one,
1130  * but does not actually start it yet.
1131  *
1132  * It copies the registers, and all the appropriate
1133  * parts of the process environment (as per the clone
1134  * flags). The actual kick-off is left to the caller.
1135  */
1136 static struct task_struct *copy_process(unsigned long clone_flags,
1137                                         unsigned long stack_start,
1138                                         unsigned long stack_size,
1139                                         int __user *child_tidptr,
1140                                         struct pid *pid,
1141                                         int trace)
1142 {
1143         int retval;
1144         struct task_struct *p;
1145
1146         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1147                 return ERR_PTR(-EINVAL);
1148
1149         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1150                 return ERR_PTR(-EINVAL);
1151
1152         /*
1153          * Thread groups must share signals as well, and detached threads
1154          * can only be started up within the thread group.
1155          */
1156         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1157                 return ERR_PTR(-EINVAL);
1158
1159         /*
1160          * Shared signal handlers imply shared VM. By way of the above,
1161          * thread groups also imply shared VM. Blocking this case allows
1162          * for various simplifications in other code.
1163          */
1164         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1165                 return ERR_PTR(-EINVAL);
1166
1167         /*
1168          * Siblings of global init remain as zombies on exit since they are
1169          * not reaped by their parent (swapper). To solve this and to avoid
1170          * multi-rooted process trees, prevent global and container-inits
1171          * from creating siblings.
1172          */
1173         if ((clone_flags & CLONE_PARENT) &&
1174                                 current->signal->flags & SIGNAL_UNKILLABLE)
1175                 return ERR_PTR(-EINVAL);
1176
1177         /*
1178          * If the new process will be in a different pid or user namespace
1179          * do not allow it to share a thread group or signal handlers or
1180          * parent with the forking task.
1181          */
1182         if (clone_flags & CLONE_SIGHAND) {
1183                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1184                     (task_active_pid_ns(current) !=
1185                                 current->nsproxy->pid_ns_for_children))
1186                         return ERR_PTR(-EINVAL);
1187         }
1188
1189         retval = security_task_create(clone_flags);
1190         if (retval)
1191                 goto fork_out;
1192
1193         retval = -ENOMEM;
1194         p = dup_task_struct(current);
1195         if (!p)
1196                 goto fork_out;
1197
1198         ftrace_graph_init_task(p);
1199         get_seccomp_filter(p);
1200
1201         rt_mutex_init_task(p);
1202
1203 #ifdef CONFIG_PROVE_LOCKING
1204         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1205         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1206 #endif
1207         retval = -EAGAIN;
1208         if (atomic_read(&p->real_cred->user->processes) >=
1209                         task_rlimit(p, RLIMIT_NPROC)) {
1210                 if (p->real_cred->user != INIT_USER &&
1211                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1212                         goto bad_fork_free;
1213         }
1214         current->flags &= ~PF_NPROC_EXCEEDED;
1215
1216         retval = copy_creds(p, clone_flags);
1217         if (retval < 0)
1218                 goto bad_fork_free;
1219
1220         /*
1221          * If multiple threads are within copy_process(), then this check
1222          * triggers too late. This doesn't hurt, the check is only there
1223          * to stop root fork bombs.
1224          */
1225         retval = -EAGAIN;
1226         if (nr_threads >= max_threads)
1227                 goto bad_fork_cleanup_count;
1228
1229         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1230                 goto bad_fork_cleanup_count;
1231
1232         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1233         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1234         p->flags |= PF_FORKNOEXEC;
1235         INIT_LIST_HEAD(&p->children);
1236         INIT_LIST_HEAD(&p->sibling);
1237         rcu_copy_process(p);
1238         p->vfork_done = NULL;
1239         spin_lock_init(&p->alloc_lock);
1240
1241         init_sigpending(&p->pending);
1242
1243         p->utime = p->stime = p->gtime = 0;
1244         p->utimescaled = p->stimescaled = 0;
1245 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
1246         p->prev_cputime.utime = p->prev_cputime.stime = 0;
1247 #endif
1248 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1249         seqlock_init(&p->vtime_seqlock);
1250         p->vtime_snap = 0;
1251         p->vtime_snap_whence = VTIME_SLEEPING;
1252 #endif
1253
1254 #if defined(SPLIT_RSS_COUNTING)
1255         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1256 #endif
1257
1258         p->default_timer_slack_ns = current->timer_slack_ns;
1259
1260         task_io_accounting_init(&p->ioac);
1261         acct_clear_integrals(p);
1262
1263         posix_cpu_timers_init(p);
1264
1265         do_posix_clock_monotonic_gettime(&p->start_time);
1266         p->real_start_time = p->start_time;
1267         monotonic_to_bootbased(&p->real_start_time);
1268         p->io_context = NULL;
1269         p->audit_context = NULL;
1270         if (clone_flags & CLONE_THREAD)
1271                 threadgroup_change_begin(current);
1272         cgroup_fork(p);
1273 #ifdef CONFIG_NUMA
1274         p->mempolicy = mpol_dup(p->mempolicy);
1275         if (IS_ERR(p->mempolicy)) {
1276                 retval = PTR_ERR(p->mempolicy);
1277                 p->mempolicy = NULL;
1278                 goto bad_fork_cleanup_threadgroup_lock;
1279         }
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         syscall_tracepoint_update(p);
1491         write_unlock_irq(&tasklist_lock);
1492
1493         proc_fork_connector(p);
1494         cgroup_post_fork(p);
1495         if (clone_flags & CLONE_THREAD)
1496                 threadgroup_change_end(current);
1497         perf_event_fork(p);
1498
1499         trace_task_newtask(p, clone_flags);
1500         uprobe_copy_process(p, clone_flags);
1501
1502         return p;
1503
1504 bad_fork_free_pid:
1505         if (pid != &init_struct_pid)
1506                 free_pid(pid);
1507 bad_fork_cleanup_io:
1508         if (p->io_context)
1509                 exit_io_context(p);
1510 bad_fork_cleanup_namespaces:
1511         exit_task_namespaces(p);
1512 bad_fork_cleanup_mm:
1513         if (p->mm)
1514                 mmput(p->mm);
1515 bad_fork_cleanup_signal:
1516         if (!(clone_flags & CLONE_THREAD))
1517                 free_signal_struct(p->signal);
1518 bad_fork_cleanup_sighand:
1519         __cleanup_sighand(p->sighand);
1520 bad_fork_cleanup_fs:
1521         exit_fs(p); /* blocking */
1522 bad_fork_cleanup_files:
1523         exit_files(p); /* blocking */
1524 bad_fork_cleanup_semundo:
1525         exit_sem(p);
1526 bad_fork_cleanup_audit:
1527         audit_free(p);
1528 bad_fork_cleanup_policy:
1529         perf_event_free_task(p);
1530 #ifdef CONFIG_NUMA
1531         mpol_put(p->mempolicy);
1532 bad_fork_cleanup_threadgroup_lock:
1533 #endif
1534         if (clone_flags & CLONE_THREAD)
1535                 threadgroup_change_end(current);
1536         delayacct_tsk_free(p);
1537         module_put(task_thread_info(p)->exec_domain->module);
1538 bad_fork_cleanup_count:
1539         atomic_dec(&p->cred->user->processes);
1540         exit_creds(p);
1541 bad_fork_free:
1542         free_task(p);
1543 fork_out:
1544         return ERR_PTR(retval);
1545 }
1546
1547 static inline void init_idle_pids(struct pid_link *links)
1548 {
1549         enum pid_type type;
1550
1551         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1552                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1553                 links[type].pid = &init_struct_pid;
1554         }
1555 }
1556
1557 struct task_struct *fork_idle(int cpu)
1558 {
1559         struct task_struct *task;
1560         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0);
1561         if (!IS_ERR(task)) {
1562                 init_idle_pids(task->pids);
1563                 init_idle(task, cpu);
1564         }
1565
1566         return task;
1567 }
1568
1569 /*
1570  *  Ok, this is the main fork-routine.
1571  *
1572  * It copies the process, and if successful kick-starts
1573  * it and waits for it to finish using the VM if required.
1574  */
1575 long do_fork(unsigned long clone_flags,
1576               unsigned long stack_start,
1577               unsigned long stack_size,
1578               int __user *parent_tidptr,
1579               int __user *child_tidptr)
1580 {
1581         struct task_struct *p;
1582         int trace = 0;
1583         long nr;
1584
1585         /*
1586          * Determine whether and which event to report to ptracer.  When
1587          * called from kernel_thread or CLONE_UNTRACED is explicitly
1588          * requested, no event is reported; otherwise, report if the event
1589          * for the type of forking is enabled.
1590          */
1591         if (!(clone_flags & CLONE_UNTRACED)) {
1592                 if (clone_flags & CLONE_VFORK)
1593                         trace = PTRACE_EVENT_VFORK;
1594                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1595                         trace = PTRACE_EVENT_CLONE;
1596                 else
1597                         trace = PTRACE_EVENT_FORK;
1598
1599                 if (likely(!ptrace_event_enabled(current, trace)))
1600                         trace = 0;
1601         }
1602
1603         p = copy_process(clone_flags, stack_start, stack_size,
1604                          child_tidptr, NULL, trace);
1605         /*
1606          * Do this prior waking up the new thread - the thread pointer
1607          * might get invalid after that point, if the thread exits quickly.
1608          */
1609         if (!IS_ERR(p)) {
1610                 struct completion vfork;
1611                 struct pid *pid;
1612
1613                 trace_sched_process_fork(current, p);
1614
1615                 pid = get_task_pid(p, PIDTYPE_PID);
1616                 nr = pid_vnr(pid);
1617
1618                 if (clone_flags & CLONE_PARENT_SETTID)
1619                         put_user(nr, parent_tidptr);
1620
1621                 if (clone_flags & CLONE_VFORK) {
1622                         p->vfork_done = &vfork;
1623                         init_completion(&vfork);
1624                         get_task_struct(p);
1625                 }
1626
1627                 wake_up_new_task(p);
1628
1629                 /* forking complete and child started to run, tell ptracer */
1630                 if (unlikely(trace))
1631                         ptrace_event_pid(trace, pid);
1632
1633                 if (clone_flags & CLONE_VFORK) {
1634                         if (!wait_for_vfork_done(p, &vfork))
1635                                 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1636                 }
1637
1638                 put_pid(pid);
1639         } else {
1640                 nr = PTR_ERR(p);
1641         }
1642         return nr;
1643 }
1644
1645 /*
1646  * Create a kernel thread.
1647  */
1648 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1649 {
1650         return do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1651                 (unsigned long)arg, NULL, NULL);
1652 }
1653
1654 #ifdef __ARCH_WANT_SYS_FORK
1655 SYSCALL_DEFINE0(fork)
1656 {
1657 #ifdef CONFIG_MMU
1658         return do_fork(SIGCHLD, 0, 0, NULL, NULL);
1659 #else
1660         /* can not support in nommu mode */
1661         return -EINVAL;
1662 #endif
1663 }
1664 #endif
1665
1666 #ifdef __ARCH_WANT_SYS_VFORK
1667 SYSCALL_DEFINE0(vfork)
1668 {
1669         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1670                         0, NULL, NULL);
1671 }
1672 #endif
1673
1674 #ifdef __ARCH_WANT_SYS_CLONE
1675 #ifdef CONFIG_CLONE_BACKWARDS
1676 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1677                  int __user *, parent_tidptr,
1678                  int, tls_val,
1679                  int __user *, child_tidptr)
1680 #elif defined(CONFIG_CLONE_BACKWARDS2)
1681 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1682                  int __user *, parent_tidptr,
1683                  int __user *, child_tidptr,
1684                  int, tls_val)
1685 #elif defined(CONFIG_CLONE_BACKWARDS3)
1686 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1687                 int, stack_size,
1688                 int __user *, parent_tidptr,
1689                 int __user *, child_tidptr,
1690                 int, tls_val)
1691 #else
1692 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1693                  int __user *, parent_tidptr,
1694                  int __user *, child_tidptr,
1695                  int, tls_val)
1696 #endif
1697 {
1698         return do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr);
1699 }
1700 #endif
1701
1702 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1703 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1704 #endif
1705
1706 static void sighand_ctor(void *data)
1707 {
1708         struct sighand_struct *sighand = data;
1709
1710         spin_lock_init(&sighand->siglock);
1711         init_waitqueue_head(&sighand->signalfd_wqh);
1712 }
1713
1714 void __init proc_caches_init(void)
1715 {
1716         sighand_cachep = kmem_cache_create("sighand_cache",
1717                         sizeof(struct sighand_struct), 0,
1718                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1719                         SLAB_NOTRACK, sighand_ctor);
1720         signal_cachep = kmem_cache_create("signal_cache",
1721                         sizeof(struct signal_struct), 0,
1722                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1723         files_cachep = kmem_cache_create("files_cache",
1724                         sizeof(struct files_struct), 0,
1725                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1726         fs_cachep = kmem_cache_create("fs_cache",
1727                         sizeof(struct fs_struct), 0,
1728                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1729         /*
1730          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1731          * whole struct cpumask for the OFFSTACK case. We could change
1732          * this to *only* allocate as much of it as required by the
1733          * maximum number of CPU's we can ever have.  The cpumask_allocation
1734          * is at the end of the structure, exactly for that reason.
1735          */
1736         mm_cachep = kmem_cache_create("mm_struct",
1737                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1738                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1739         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1740         mmap_init();
1741         nsproxy_cache_init();
1742 }
1743
1744 /*
1745  * Check constraints on flags passed to the unshare system call.
1746  */
1747 static int check_unshare_flags(unsigned long unshare_flags)
1748 {
1749         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1750                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1751                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1752                                 CLONE_NEWUSER|CLONE_NEWPID))
1753                 return -EINVAL;
1754         /*
1755          * Not implemented, but pretend it works if there is nothing to
1756          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1757          * needs to unshare vm.
1758          */
1759         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1760                 /* FIXME: get_task_mm() increments ->mm_users */
1761                 if (atomic_read(&current->mm->mm_users) > 1)
1762                         return -EINVAL;
1763         }
1764
1765         return 0;
1766 }
1767
1768 /*
1769  * Unshare the filesystem structure if it is being shared
1770  */
1771 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1772 {
1773         struct fs_struct *fs = current->fs;
1774
1775         if (!(unshare_flags & CLONE_FS) || !fs)
1776                 return 0;
1777
1778         /* don't need lock here; in the worst case we'll do useless copy */
1779         if (fs->users == 1)
1780                 return 0;
1781
1782         *new_fsp = copy_fs_struct(fs);
1783         if (!*new_fsp)
1784                 return -ENOMEM;
1785
1786         return 0;
1787 }
1788
1789 /*
1790  * Unshare file descriptor table if it is being shared
1791  */
1792 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1793 {
1794         struct files_struct *fd = current->files;
1795         int error = 0;
1796
1797         if ((unshare_flags & CLONE_FILES) &&
1798             (fd && atomic_read(&fd->count) > 1)) {
1799                 *new_fdp = dup_fd(fd, &error);
1800                 if (!*new_fdp)
1801                         return error;
1802         }
1803
1804         return 0;
1805 }
1806
1807 /*
1808  * unshare allows a process to 'unshare' part of the process
1809  * context which was originally shared using clone.  copy_*
1810  * functions used by do_fork() cannot be used here directly
1811  * because they modify an inactive task_struct that is being
1812  * constructed. Here we are modifying the current, active,
1813  * task_struct.
1814  */
1815 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1816 {
1817         struct fs_struct *fs, *new_fs = NULL;
1818         struct files_struct *fd, *new_fd = NULL;
1819         struct cred *new_cred = NULL;
1820         struct nsproxy *new_nsproxy = NULL;
1821         int do_sysvsem = 0;
1822         int err;
1823
1824         /*
1825          * If unsharing a user namespace must also unshare the thread.
1826          */
1827         if (unshare_flags & CLONE_NEWUSER)
1828                 unshare_flags |= CLONE_THREAD | CLONE_FS;
1829         /*
1830          * If unsharing a thread from a thread group, must also unshare vm.
1831          */
1832         if (unshare_flags & CLONE_THREAD)
1833                 unshare_flags |= CLONE_VM;
1834         /*
1835          * If unsharing vm, must also unshare signal handlers.
1836          */
1837         if (unshare_flags & CLONE_VM)
1838                 unshare_flags |= CLONE_SIGHAND;
1839         /*
1840          * If unsharing namespace, must also unshare filesystem information.
1841          */
1842         if (unshare_flags & CLONE_NEWNS)
1843                 unshare_flags |= CLONE_FS;
1844
1845         err = check_unshare_flags(unshare_flags);
1846         if (err)
1847                 goto bad_unshare_out;
1848         /*
1849          * CLONE_NEWIPC must also detach from the undolist: after switching
1850          * to a new ipc namespace, the semaphore arrays from the old
1851          * namespace are unreachable.
1852          */
1853         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1854                 do_sysvsem = 1;
1855         err = unshare_fs(unshare_flags, &new_fs);
1856         if (err)
1857                 goto bad_unshare_out;
1858         err = unshare_fd(unshare_flags, &new_fd);
1859         if (err)
1860                 goto bad_unshare_cleanup_fs;
1861         err = unshare_userns(unshare_flags, &new_cred);
1862         if (err)
1863                 goto bad_unshare_cleanup_fd;
1864         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1865                                          new_cred, new_fs);
1866         if (err)
1867                 goto bad_unshare_cleanup_cred;
1868
1869         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
1870                 if (do_sysvsem) {
1871                         /*
1872                          * CLONE_SYSVSEM is equivalent to sys_exit().
1873                          */
1874                         exit_sem(current);
1875                 }
1876
1877                 if (new_nsproxy)
1878                         switch_task_namespaces(current, new_nsproxy);
1879
1880                 task_lock(current);
1881
1882                 if (new_fs) {
1883                         fs = current->fs;
1884                         spin_lock(&fs->lock);
1885                         current->fs = new_fs;
1886                         if (--fs->users)
1887                                 new_fs = NULL;
1888                         else
1889                                 new_fs = fs;
1890                         spin_unlock(&fs->lock);
1891                 }
1892
1893                 if (new_fd) {
1894                         fd = current->files;
1895                         current->files = new_fd;
1896                         new_fd = fd;
1897                 }
1898
1899                 task_unlock(current);
1900
1901                 if (new_cred) {
1902                         /* Install the new user namespace */
1903                         commit_creds(new_cred);
1904                         new_cred = NULL;
1905                 }
1906         }
1907
1908 bad_unshare_cleanup_cred:
1909         if (new_cred)
1910                 put_cred(new_cred);
1911 bad_unshare_cleanup_fd:
1912         if (new_fd)
1913                 put_files_struct(new_fd);
1914
1915 bad_unshare_cleanup_fs:
1916         if (new_fs)
1917                 free_fs_struct(new_fs);
1918
1919 bad_unshare_out:
1920         return err;
1921 }
1922
1923 /*
1924  *      Helper to unshare the files of the current task.
1925  *      We don't want to expose copy_files internals to
1926  *      the exec layer of the kernel.
1927  */
1928
1929 int unshare_files(struct files_struct **displaced)
1930 {
1931         struct task_struct *task = current;
1932         struct files_struct *copy = NULL;
1933         int error;
1934
1935         error = unshare_fd(CLONE_FILES, &copy);
1936         if (error || !copy) {
1937                 *displaced = NULL;
1938                 return error;
1939         }
1940         *displaced = task->files;
1941         task_lock(task);
1942         task->files = copy;
1943         task_unlock(task);
1944         return 0;
1945 }