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         security_task_free(tsk);
 243         exit_creds(tsk);
 244         delayacct_tsk_free(tsk);
 245         put_signal_struct(tsk->signal);
 246
 247         if (!profile_handoff_task(tsk))
 248                 free_task(tsk);
 249 }
 250 EXPORT_SYMBOL_GPL(__put_task_struct);
 251
 252 void __init __weak arch_task_cache_init(void) { }
 253
 254 void __init fork_init(unsigned long mempages)
 255 {
 256 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 257 #ifndef ARCH_MIN_TASKALIGN
 258 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 259 #endif
 260         /* create a slab on which task_structs can be allocated */
 261         task_struct_cachep =
 262                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 263                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
 264 #endif
 265
 266         /* do the arch specific task caches init */
 267         arch_task_cache_init();
 268
 269         /*
 270          * The default maximum number of threads is set to a safe
 271          * value: the thread structures can take up at most half
 272          * of memory.
 273          */
 274         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 275
 276         /*
 277          * we need to allow at least 20 threads to boot a system
 278          */
 279         if (max_threads < 20)
 280                 max_threads = 20;
 281
 282         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 283         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 284         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 285                 init_task.signal->rlim[RLIMIT_NPROC];
 286 }
 287
 288 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
 289                                                struct task_struct *src)
 290 {
 291         *dst = *src;
 292         return 0;
 293 }
 294
 295 static struct task_struct *dup_task_struct(struct task_struct *orig)
 296 {
 297         struct task_struct *tsk;
 298         struct thread_info *ti;
 299         unsigned long *stackend;
 300         int node = tsk_fork_get_node(orig);
 301         int err;
 302
 303         tsk = alloc_task_struct_node(node);
 304         if (!tsk)
 305                 return NULL;
 306
 307         ti = alloc_thread_info_node(tsk, node);
 308         if (!ti)
 309                 goto free_tsk;
 310
 311         err = arch_dup_task_struct(tsk, orig);
 312         if (err)
 313                 goto free_ti;
 314
 315         tsk->stack = ti;
 316
 317         setup_thread_stack(tsk, orig);
 318         clear_user_return_notifier(tsk);
 319         clear_tsk_need_resched(tsk);
 320         stackend = end_of_stack(tsk);
 321         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 322
 323 #ifdef CONFIG_CC_STACKPROTECTOR
 324         tsk->stack_canary = get_random_int();
 325 #endif
 326
 327         /*
 328          * One for us, one for whoever does the "release_task()" (usually
 329          * parent)
 330          */
 331         atomic_set(&tsk->usage, 2);
 332 #ifdef CONFIG_BLK_DEV_IO_TRACE
 333         tsk->btrace_seq = 0;
 334 #endif
 335         tsk->splice_pipe = NULL;
 336         tsk->task_frag.page = NULL;
 337
 338         account_kernel_stack(ti, 1);
 339
 340         return tsk;
 341
 342 free_ti:
 343         free_thread_info(ti);
 344 free_tsk:
 345         free_task_struct(tsk);
 346         return NULL;
 347 }
 348
 349 #ifdef CONFIG_MMU
 350 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 351 {
 352         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 353         struct rb_node **rb_link, *rb_parent;
 354         int retval;
 355         unsigned long charge;
 356
 357         uprobe_start_dup_mmap();
 358         down_write(&oldmm->mmap_sem);
 359         flush_cache_dup_mm(oldmm);
 360         uprobe_dup_mmap(oldmm, mm);
 361         /*
 362          * Not linked in yet - no deadlock potential:
 363          */
 364         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 365
 366         mm->locked_vm = 0;
 367         mm->mmap = NULL;
 368         mm->vmacache_seqnum = 0;
 369         mm->map_count = 0;
 370         cpumask_clear(mm_cpumask(mm));
 371         mm->mm_rb = RB_ROOT;
 372         rb_link = &mm->mm_rb.rb_node;
 373         rb_parent = NULL;
 374         pprev = &mm->mmap;
 375         retval = ksm_fork(mm, oldmm);
 376         if (retval)
 377                 goto out;
 378         retval = khugepaged_fork(mm, oldmm);
 379         if (retval)
 380                 goto out;
 381
 382         prev = NULL;
 383         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 384                 struct file *file;
 385
 386                 if (mpnt->vm_flags & VM_DONTCOPY) {
 387                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 388                                                         -vma_pages(mpnt));
 389                         continue;
 390                 }
 391                 charge = 0;
 392                 if (mpnt->vm_flags & VM_ACCOUNT) {
 393                         unsigned long len = vma_pages(mpnt);
 394
 395                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 396                                 goto fail_nomem;
 397                         charge = len;
 398                 }
 399                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 400                 if (!tmp)
 401                         goto fail_nomem;
 402                 *tmp = *mpnt;
 403                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 404                 retval = vma_dup_policy(mpnt, tmp);
 405                 if (retval)
 406                         goto fail_nomem_policy;
 407                 tmp->vm_mm = mm;
 408                 if (anon_vma_fork(tmp, mpnt))
 409                         goto fail_nomem_anon_vma_fork;
 410                 tmp->vm_flags &= ~VM_LOCKED;
 411                 tmp->vm_next = tmp->vm_prev = NULL;
 412                 file = tmp->vm_file;
 413                 if (file) {
 414                         struct inode *inode = file_inode(file);
 415                         struct address_space *mapping = file->f_mapping;
 416
 417                         get_file(file);
 418                         if (tmp->vm_flags & VM_DENYWRITE)
 419                                 atomic_dec(&inode->i_writecount);
 420                         mutex_lock(&mapping->i_mmap_mutex);
 421                         if (tmp->vm_flags & VM_SHARED)
 422                                 mapping->i_mmap_writable++;
 423                         flush_dcache_mmap_lock(mapping);
 424                         /* insert tmp into the share list, just after mpnt */
 425                         if (unlikely(tmp->vm_flags & VM_NONLINEAR))
 426                                 vma_nonlinear_insert(tmp,
 427                                                 &mapping->i_mmap_nonlinear);
 428                         else
 429                                 vma_interval_tree_insert_after(tmp, mpnt,
 430                                                         &mapping->i_mmap);
 431                         flush_dcache_mmap_unlock(mapping);
 432                         mutex_unlock(&mapping->i_mmap_mutex);
 433                 }
 434
 435                 /*
 436                  * Clear hugetlb-related page reserves for children. This only
 437                  * affects MAP_PRIVATE mappings. Faults generated by the child
 438                  * are not guaranteed to succeed, even if read-only
 439                  */
 440                 if (is_vm_hugetlb_page(tmp))
 441                         reset_vma_resv_huge_pages(tmp);
 442
 443                 /*
 444                  * Link in the new vma and copy the page table entries.
 445                  */
 446                 *pprev = tmp;
 447                 pprev = &tmp->vm_next;
 448                 tmp->vm_prev = prev;
 449                 prev = tmp;
 450
 451                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 452                 rb_link = &tmp->vm_rb.rb_right;
 453                 rb_parent = &tmp->vm_rb;
 454
 455                 mm->map_count++;
 456                 retval = copy_page_range(mm, oldmm, mpnt);
 457
 458                 if (tmp->vm_ops && tmp->vm_ops->open)
 459                         tmp->vm_ops->open(tmp);
 460
 461                 if (retval)
 462                         goto out;
 463         }
 464         /* a new mm has just been created */
 465         arch_dup_mmap(oldmm, mm);
 466         retval = 0;
 467 out:
 468         up_write(&mm->mmap_sem);
 469         flush_tlb_mm(oldmm);
 470         up_write(&oldmm->mmap_sem);
 471         uprobe_end_dup_mmap();
 472         return retval;
 473 fail_nomem_anon_vma_fork:
 474         mpol_put(vma_policy(tmp));
 475 fail_nomem_policy:
 476         kmem_cache_free(vm_area_cachep, tmp);
 477 fail_nomem:
 478         retval = -ENOMEM;
 479         vm_unacct_memory(charge);
 480         goto out;
 481 }
 482
 483 static inline int mm_alloc_pgd(struct mm_struct *mm)
 484 {
 485         mm->pgd = pgd_alloc(mm);
 486         if (unlikely(!mm->pgd))
 487                 return -ENOMEM;
 488         return 0;
 489 }
 490
 491 static inline void mm_free_pgd(struct mm_struct *mm)
 492 {
 493         pgd_free(mm, mm->pgd);
 494 }
 495 #else
 496 #define dup_mmap(mm, oldmm)     (0)
 497 #define mm_alloc_pgd(mm)        (0)
 498 #define mm_free_pgd(mm)
 499 #endif /* CONFIG_MMU */
 500
 501 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 502
 503 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 504 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 505
 506 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 507
 508 static int __init coredump_filter_setup(char *s)
 509 {
 510         default_dump_filter =
 511                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 512                 MMF_DUMP_FILTER_MASK;
 513         return 1;
 514 }
 515
 516 __setup("coredump_filter=", coredump_filter_setup);
 517
 518 #include <linux/init_task.h>
 519
 520 static void mm_init_aio(struct mm_struct *mm)
 521 {
 522 #ifdef CONFIG_AIO
 523         spin_lock_init(&mm->ioctx_lock);
 524         mm->ioctx_table = NULL;
 525 #endif
 526 }
 527
 528 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
 529 {
 530         atomic_set(&mm->mm_users, 1);
 531         atomic_set(&mm->mm_count, 1);
 532         init_rwsem(&mm->mmap_sem);
 533         INIT_LIST_HEAD(&mm->mmlist);
 534         mm->flags = (current->mm) ?
 535                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
 536         mm->core_state = NULL;
 537         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 (likely(!mm_alloc_pgd(mm))) {
 545                 mm->def_flags = 0;
 546                 mmu_notifier_mm_init(mm);
 547                 return mm;
 548         }
 549
 550         free_mm(mm);
 551         return NULL;
 552 }
 553
 554 static void check_mm(struct mm_struct *mm)
 555 {
 556         int i;
 557
 558         for (i = 0; i < NR_MM_COUNTERS; i++) {
 559                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 560
 561                 if (unlikely(x))
 562                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 563                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 564         }
 565
 566 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 567         VM_BUG_ON(mm->pmd_huge_pte);
 568 #endif
 569 }
 570
 571 /*
 572  * Allocate and initialize an mm_struct.
 573  */
 574 struct mm_struct *mm_alloc(void)
 575 {
 576         struct mm_struct *mm;
 577
 578         mm = allocate_mm();
 579         if (!mm)
 580                 return NULL;
 581
 582         memset(mm, 0, sizeof(*mm));
 583         mm_init_cpumask(mm);
 584         return mm_init(mm, current);
 585 }
 586
 587 /*
 588  * Called when the last reference to the mm
 589  * is dropped: either by a lazy thread or by
 590  * mmput. Free the page directory and the mm.
 591  */
 592 void __mmdrop(struct mm_struct *mm)
 593 {
 594         BUG_ON(mm == &init_mm);
 595         mm_free_pgd(mm);
 596         destroy_context(mm);
 597         mmu_notifier_mm_destroy(mm);
 598         check_mm(mm);
 599         free_mm(mm);
 600 }
 601 EXPORT_SYMBOL_GPL(__mmdrop);
 602
 603 /*
 604  * Decrement the use count and release all resources for an mm.
 605  */
 606 void mmput(struct mm_struct *mm)
 607 {
 608         might_sleep();
 609
 610         if (atomic_dec_and_test(&mm->mm_users)) {
 611                 uprobe_clear_state(mm);
 612                 exit_aio(mm);
 613                 ksm_exit(mm);
 614                 khugepaged_exit(mm); /* must run before exit_mmap */
 615                 exit_mmap(mm);
 616                 set_mm_exe_file(mm, NULL);
 617                 if (!list_empty(&mm->mmlist)) {
 618                         spin_lock(&mmlist_lock);
 619                         list_del(&mm->mmlist);
 620                         spin_unlock(&mmlist_lock);
 621                 }
 622                 if (mm->binfmt)
 623                         module_put(mm->binfmt->module);
 624                 mmdrop(mm);
 625         }
 626 }
 627 EXPORT_SYMBOL_GPL(mmput);
 628
 629 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 630 {
 631         if (new_exe_file)
 632                 get_file(new_exe_file);
 633         if (mm->exe_file)
 634                 fput(mm->exe_file);
 635         mm->exe_file = new_exe_file;
 636 }
 637
 638 struct file *get_mm_exe_file(struct mm_struct *mm)
 639 {
 640         struct file *exe_file;
 641
 642         /* We need mmap_sem to protect against races with removal of exe_file */
 643         down_read(&mm->mmap_sem);
 644         exe_file = mm->exe_file;
 645         if (exe_file)
 646                 get_file(exe_file);
 647         up_read(&mm->mmap_sem);
 648         return exe_file;
 649 }
 650
 651 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
 652 {
 653         /* It's safe to write the exe_file pointer without exe_file_lock because
 654          * this is called during fork when the task is not yet in /proc */
 655         newmm->exe_file = get_mm_exe_file(oldmm);
 656 }
 657
 658 /**
 659  * get_task_mm - acquire a reference to the task's mm
 660  *
 661  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 662  * this kernel workthread has transiently adopted a user mm with use_mm,
 663  * to do its AIO) is not set and if so returns a reference to it, after
 664  * bumping up the use count.  User must release the mm via mmput()
 665  * after use.  Typically used by /proc and ptrace.
 666  */
 667 struct mm_struct *get_task_mm(struct task_struct *task)
 668 {
 669         struct mm_struct *mm;
 670
 671         task_lock(task);
 672         mm = task->mm;
 673         if (mm) {
 674                 if (task->flags & PF_KTHREAD)
 675                         mm = NULL;
 676                 else
 677                         atomic_inc(&mm->mm_users);
 678         }
 679         task_unlock(task);
 680         return mm;
 681 }
 682 EXPORT_SYMBOL_GPL(get_task_mm);
 683
 684 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 685 {
 686         struct mm_struct *mm;
 687         int err;
 688
 689         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 690         if (err)
 691                 return ERR_PTR(err);
 692
 693         mm = get_task_mm(task);
 694         if (mm && mm != current->mm &&
 695                         !ptrace_may_access(task, mode)) {
 696                 mmput(mm);
 697                 mm = ERR_PTR(-EACCES);
 698         }
 699         mutex_unlock(&task->signal->cred_guard_mutex);
 700
 701         return mm;
 702 }
 703
 704 static void complete_vfork_done(struct task_struct *tsk)
 705 {
 706         struct completion *vfork;
 707
 708         task_lock(tsk);
 709         vfork = tsk->vfork_done;
 710         if (likely(vfork)) {
 711                 tsk->vfork_done = NULL;
 712                 complete(vfork);
 713         }
 714         task_unlock(tsk);
 715 }
 716
 717 static int wait_for_vfork_done(struct task_struct *child,
 718                                 struct completion *vfork)
 719 {
 720         int killed;
 721
 722         freezer_do_not_count();
 723         killed = wait_for_completion_killable(vfork);
 724         freezer_count();
 725
 726         if (killed) {
 727                 task_lock(child);
 728                 child->vfork_done = NULL;
 729                 task_unlock(child);
 730         }
 731
 732         put_task_struct(child);
 733         return killed;
 734 }
 735
 736 /* Please note the differences between mmput and mm_release.
 737  * mmput is called whenever we stop holding onto a mm_struct,
 738  * error success whatever.
 739  *
 740  * mm_release is called after a mm_struct has been removed
 741  * from the current process.
 742  *
 743  * This difference is important for error handling, when we
 744  * only half set up a mm_struct for a new process and need to restore
 745  * the old one.  Because we mmput the new mm_struct before
 746  * restoring the old one. . .
 747  * Eric Biederman 10 January 1998
 748  */
 749 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 750 {
 751         /* Get rid of any futexes when releasing the mm */
 752 #ifdef CONFIG_FUTEX
 753         if (unlikely(tsk->robust_list)) {
 754                 exit_robust_list(tsk);
 755                 tsk->robust_list = NULL;
 756         }
 757 #ifdef CONFIG_COMPAT
 758         if (unlikely(tsk->compat_robust_list)) {
 759                 compat_exit_robust_list(tsk);
 760                 tsk->compat_robust_list = NULL;
 761         }
 762 #endif
 763         if (unlikely(!list_empty(&tsk->pi_state_list)))
 764                 exit_pi_state_list(tsk);
 765 #endif
 766
 767         uprobe_free_utask(tsk);
 768
 769         /* Get rid of any cached register state */
 770         deactivate_mm(tsk, mm);
 771
 772         /*
 773          * Signal userspace if we're not exiting with a core dump
 774          * because we want to leave the value intact for debugging
 775          * purposes.
 776          */
 777         if (tsk->clear_child_tid) {
 778                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
 779                     atomic_read(&mm->mm_users) > 1) {
 780                         /*
 781                          * We don't check the error code - if userspace has
 782                          * not set up a proper pointer then tough luck.
 783                          */
 784                         put_user(0, tsk->clear_child_tid);
 785                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 786                                         1, NULL, NULL, 0);
 787                 }
 788                 tsk->clear_child_tid = NULL;
 789         }
 790
 791         /*
 792          * All done, finally we can wake up parent and return this mm to him.
 793          * Also kthread_stop() uses this completion for synchronization.
 794          */
 795         if (tsk->vfork_done)
 796                 complete_vfork_done(tsk);
 797 }
 798
 799 /*
 800  * Allocate a new mm structure and copy contents from the
 801  * mm structure of the passed in task structure.
 802  */
 803 struct mm_struct *dup_mm(struct task_struct *tsk)
 804 {
 805         struct mm_struct *mm, *oldmm = current->mm;
 806         int err;
 807
 808         if (!oldmm)
 809                 return NULL;
 810
 811         mm = allocate_mm();
 812         if (!mm)
 813                 goto fail_nomem;
 814
 815         memcpy(mm, oldmm, sizeof(*mm));
 816         mm_init_cpumask(mm);
 817
 818 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 819         mm->pmd_huge_pte = NULL;
 820 #endif
 821 #ifdef CONFIG_NUMA_BALANCING
 822         mm->first_nid = NUMA_PTE_SCAN_INIT;
 823 #endif
 824         if (!mm_init(mm, tsk))
 825                 goto fail_nomem;
 826
 827         if (init_new_context(tsk, mm))
 828                 goto fail_nocontext;
 829
 830         dup_mm_exe_file(oldmm, mm);
 831
 832         err = dup_mmap(mm, oldmm);
 833         if (err)
 834                 goto free_pt;
 835
 836         mm->hiwater_rss = get_mm_rss(mm);
 837         mm->hiwater_vm = mm->total_vm;
 838
 839         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 840                 goto free_pt;
 841
 842         return mm;
 843
 844 free_pt:
 845         /* don't put binfmt in mmput, we haven't got module yet */
 846         mm->binfmt = NULL;
 847         mmput(mm);
 848
 849 fail_nomem:
 850         return NULL;
 851
 852 fail_nocontext:
 853         /*
 854          * If init_new_context() failed, we cannot use mmput() to free the mm
 855          * because it calls destroy_context()
 856          */
 857         mm_free_pgd(mm);
 858         free_mm(mm);
 859         return NULL;
 860 }
 861
 862 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 863 {
 864         struct mm_struct *mm, *oldmm;
 865         int retval;
 866
 867         tsk->min_flt = tsk->maj_flt = 0;
 868         tsk->nvcsw = tsk->nivcsw = 0;
 869 #ifdef CONFIG_DETECT_HUNG_TASK
 870         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 871 #endif
 872
 873         tsk->mm = NULL;
 874         tsk->active_mm = NULL;
 875
 876         /*
 877          * Are we cloning a kernel thread?
 878          *
 879          * We need to steal a active VM for that..
 880          */
 881         oldmm = current->mm;
 882         if (!oldmm)
 883                 return 0;
 884
 885         /* initialize the new vmacache entries */
 886         vmacache_flush(tsk);
 887
 888         if (clone_flags & CLONE_VM) {
 889                 atomic_inc(&oldmm->mm_users);
 890                 mm = oldmm;
 891                 goto good_mm;
 892         }
 893
 894         retval = -ENOMEM;
 895         mm = dup_mm(tsk);
 896         if (!mm)
 897                 goto fail_nomem;
 898
 899 good_mm:
 900         tsk->mm = mm;
 901         tsk->active_mm = mm;
 902         return 0;
 903
 904 fail_nomem:
 905         return retval;
 906 }
 907
 908 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 909 {
 910         struct fs_struct *fs = current->fs;
 911         if (clone_flags & CLONE_FS) {
 912                 /* tsk->fs is already what we want */
 913                 spin_lock(&fs->lock);
 914                 if (fs->in_exec) {
 915                         spin_unlock(&fs->lock);
 916                         return -EAGAIN;
 917                 }
 918                 fs->users++;
 919                 spin_unlock(&fs->lock);
 920                 return 0;
 921         }
 922         tsk->fs = copy_fs_struct(fs);
 923         if (!tsk->fs)
 924                 return -ENOMEM;
 925         return 0;
 926 }
 927
 928 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
 929 {
 930         struct files_struct *oldf, *newf;
 931         int error = 0;
 932
 933         /*
 934          * A background process may not have any files ...
 935          */
 936         oldf = current->files;
 937         if (!oldf)
 938                 goto out;
 939
 940         if (clone_flags & CLONE_FILES) {
 941                 atomic_inc(&oldf->count);
 942                 goto out;
 943         }
 944
 945         newf = dup_fd(oldf, &error);
 946         if (!newf)
 947                 goto out;
 948
 949         tsk->files = newf;
 950         error = 0;
 951 out:
 952         return error;
 953 }
 954
 955 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 956 {
 957 #ifdef CONFIG_BLOCK
 958         struct io_context *ioc = current->io_context;
 959         struct io_context *new_ioc;
 960
 961         if (!ioc)
 962                 return 0;
 963         /*
 964          * Share io context with parent, if CLONE_IO is set
 965          */
 966         if (clone_flags & CLONE_IO) {
 967                 ioc_task_link(ioc);
 968                 tsk->io_context = ioc;
 969         } else if (ioprio_valid(ioc->ioprio)) {
 970                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
 971                 if (unlikely(!new_ioc))
 972                         return -ENOMEM;
 973
 974                 new_ioc->ioprio = ioc->ioprio;
 975                 put_io_context(new_ioc);
 976         }
 977 #endif
 978         return 0;
 979 }
 980
 981 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 982 {
 983         struct sighand_struct *sig;
 984
 985         if (clone_flags & CLONE_SIGHAND) {
 986                 atomic_inc(&current->sighand->count);
 987                 return 0;
 988         }
 989         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 990         rcu_assign_pointer(tsk->sighand, sig);
 991         if (!sig)
 992                 return -ENOMEM;
 993         atomic_set(&sig->count, 1);
 994         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 995         return 0;
 996 }
 997
 998 void __cleanup_sighand(struct sighand_struct *sighand)
 999 {
1000         if (atomic_dec_and_test(&sighand->count)) {
1001                 signalfd_cleanup(sighand);
1002                 kmem_cache_free(sighand_cachep, sighand);
1003         }
1004 }
1005
1006
1007 /*
1008  * Initialize POSIX timer handling for a thread group.
1009  */
1010 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1011 {
1012         unsigned long cpu_limit;
1013
1014         /* Thread group counters. */
1015         thread_group_cputime_init(sig);
1016
1017         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1018         if (cpu_limit != RLIM_INFINITY) {
1019                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1020                 sig->cputimer.running = 1;
1021         }
1022
1023         /* The timer lists. */
1024         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1025         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1026         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1027 }
1028
1029 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1030 {
1031         struct signal_struct *sig;
1032
1033         if (clone_flags & CLONE_THREAD)
1034                 return 0;
1035
1036         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1037         tsk->signal = sig;
1038         if (!sig)
1039                 return -ENOMEM;
1040
1041         sig->nr_threads = 1;
1042         atomic_set(&sig->live, 1);
1043         atomic_set(&sig->sigcnt, 1);
1044
1045         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1046         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1047         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1048
1049         init_waitqueue_head(&sig->wait_chldexit);
1050         sig->curr_target = tsk;
1051         init_sigpending(&sig->shared_pending);
1052         INIT_LIST_HEAD(&sig->posix_timers);
1053
1054         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1055         sig->real_timer.function = it_real_fn;
1056
1057         task_lock(current->group_leader);
1058         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1059         task_unlock(current->group_leader);
1060
1061         posix_cpu_timers_init_group(sig);
1062
1063         tty_audit_fork(sig);
1064         sched_autogroup_fork(sig);
1065
1066 #ifdef CONFIG_CGROUPS
1067         init_rwsem(&sig->group_rwsem);
1068 #endif
1069
1070         sig->oom_score_adj = current->signal->oom_score_adj;
1071         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1072
1073         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1074                                    current->signal->is_child_subreaper;
1075
1076         mutex_init(&sig->cred_guard_mutex);
1077
1078         return 0;
1079 }
1080
1081 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
1082 {
1083         unsigned long new_flags = p->flags;
1084
1085         new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1086         new_flags |= PF_FORKNOEXEC;
1087         p->flags = new_flags;
1088 }
1089
1090 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1091 {
1092         current->clear_child_tid = tidptr;
1093
1094         return task_pid_vnr(current);
1095 }
1096
1097 static void rt_mutex_init_task(struct task_struct *p)
1098 {
1099         raw_spin_lock_init(&p->pi_lock);
1100 #ifdef CONFIG_RT_MUTEXES
1101         plist_head_init(&p->pi_waiters);
1102         p->pi_blocked_on = NULL;
1103 #endif
1104 }
1105
1106 #ifdef CONFIG_MM_OWNER
1107 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1108 {
1109         mm->owner = p;
1110 }
1111 #endif /* CONFIG_MM_OWNER */
1112
1113 /*
1114  * Initialize POSIX timer handling for a single task.
1115  */
1116 static void posix_cpu_timers_init(struct task_struct *tsk)
1117 {
1118         tsk->cputime_expires.prof_exp = 0;
1119         tsk->cputime_expires.virt_exp = 0;
1120         tsk->cputime_expires.sched_exp = 0;
1121         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1122         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1123         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1124 }
1125
1126 static inline void
1127 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1128 {
1129          task->pids[type].pid = pid;
1130 }
1131
1132 /*
1133  * This creates a new process as a copy of the old one,
1134  * but does not actually start it yet.
1135  *
1136  * It copies the registers, and all the appropriate
1137  * parts of the process environment (as per the clone
1138  * flags). The actual kick-off is left to the caller.
1139  */
1140 static struct task_struct *copy_process(unsigned long clone_flags,
1141                                         unsigned long stack_start,
1142                                         unsigned long stack_size,
1143                                         int __user *child_tidptr,
1144                                         struct pid *pid,
1145                                         int trace)
1146 {
1147         int retval;
1148         struct task_struct *p;
1149
1150         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1151                 return ERR_PTR(-EINVAL);
1152
1153         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1154                 return ERR_PTR(-EINVAL);
1155
1156         /*
1157          * Thread groups must share signals as well, and detached threads
1158          * can only be started up within the thread group.
1159          */
1160         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1161                 return ERR_PTR(-EINVAL);
1162
1163         /*
1164          * Shared signal handlers imply shared VM. By way of the above,
1165          * thread groups also imply shared VM. Blocking this case allows
1166          * for various simplifications in other code.
1167          */
1168         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1169                 return ERR_PTR(-EINVAL);
1170
1171         /*
1172          * Siblings of global init remain as zombies on exit since they are
1173          * not reaped by their parent (swapper). To solve this and to avoid
1174          * multi-rooted process trees, prevent global and container-inits
1175          * from creating siblings.
1176          */
1177         if ((clone_flags & CLONE_PARENT) &&
1178                                 current->signal->flags & SIGNAL_UNKILLABLE)
1179                 return ERR_PTR(-EINVAL);
1180
1181         /*
1182          * If the new process will be in a different pid or user namespace
1183          * do not allow it to share a thread group or signal handlers or
1184          * parent with the forking task.
1185          */
1186         if (clone_flags & CLONE_SIGHAND) {
1187                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1188                     (task_active_pid_ns(current) !=
1189                                 current->nsproxy->pid_ns_for_children))
1190                         return ERR_PTR(-EINVAL);
1191         }
1192
1193         retval = security_task_create(clone_flags);
1194         if (retval)
1195                 goto fork_out;
1196
1197         retval = -ENOMEM;
1198         p = dup_task_struct(current);
1199         if (!p)
1200                 goto fork_out;
1201
1202         ftrace_graph_init_task(p);
1203         get_seccomp_filter(p);
1204
1205         rt_mutex_init_task(p);
1206
1207 #ifdef CONFIG_PROVE_LOCKING
1208         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1209         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1210 #endif
1211         retval = -EAGAIN;
1212         if (atomic_read(&p->real_cred->user->processes) >=
1213                         task_rlimit(p, RLIMIT_NPROC)) {
1214                 if (p->real_cred->user != INIT_USER &&
1215                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1216                         goto bad_fork_free;
1217         }
1218         current->flags &= ~PF_NPROC_EXCEEDED;
1219
1220         retval = copy_creds(p, clone_flags);
1221         if (retval < 0)
1222                 goto bad_fork_free;
1223
1224         /*
1225          * If multiple threads are within copy_process(), then this check
1226          * triggers too late. This doesn't hurt, the check is only there
1227          * to stop root fork bombs.
1228          */
1229         retval = -EAGAIN;
1230         if (nr_threads >= max_threads)
1231                 goto bad_fork_cleanup_count;
1232
1233         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1234                 goto bad_fork_cleanup_count;
1235
1236         p->did_exec = 0;
1237         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1238         copy_flags(clone_flags, p);
1239         INIT_LIST_HEAD(&p->children);
1240         INIT_LIST_HEAD(&p->sibling);
1241         rcu_copy_process(p);
1242         p->vfork_done = NULL;
1243         spin_lock_init(&p->alloc_lock);
1244
1245         init_sigpending(&p->pending);
1246
1247         p->utime = p->stime = p->gtime = 0;
1248         p->utimescaled = p->stimescaled = 0;
1249 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
1250         p->prev_cputime.utime = p->prev_cputime.stime = 0;
1251 #endif
1252 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1253         seqlock_init(&p->vtime_seqlock);
1254         p->vtime_snap = 0;
1255         p->vtime_snap_whence = VTIME_SLEEPING;
1256 #endif
1257
1258 #if defined(SPLIT_RSS_COUNTING)
1259         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1260 #endif
1261
1262         p->default_timer_slack_ns = current->timer_slack_ns;
1263
1264         task_io_accounting_init(&p->ioac);
1265         acct_clear_integrals(p);
1266
1267         posix_cpu_timers_init(p);
1268
1269         do_posix_clock_monotonic_gettime(&p->start_time);
1270         p->real_start_time = p->start_time;
1271         monotonic_to_bootbased(&p->real_start_time);
1272         p->io_context = NULL;
1273         p->audit_context = NULL;
1274         if (clone_flags & CLONE_THREAD)
1275                 threadgroup_change_begin(current);
1276         cgroup_fork(p);
1277 #ifdef CONFIG_NUMA
1278         p->mempolicy = mpol_dup(p->mempolicy);
1279         if (IS_ERR(p->mempolicy)) {
1280                 retval = PTR_ERR(p->mempolicy);
1281                 p->mempolicy = NULL;
1282                 goto bad_fork_cleanup_cgroup;
1283         }
1284         mpol_fix_fork_child_flag(p);
1285 #endif
1286 #ifdef CONFIG_CPUSETS
1287         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1288         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1289         seqcount_init(&p->mems_allowed_seq);
1290 #endif
1291 #ifdef CONFIG_TRACE_IRQFLAGS
1292         p->irq_events = 0;
1293         p->hardirqs_enabled = 0;
1294         p->hardirq_enable_ip = 0;
1295         p->hardirq_enable_event = 0;
1296         p->hardirq_disable_ip = _THIS_IP_;
1297         p->hardirq_disable_event = 0;
1298         p->softirqs_enabled = 1;
1299         p->softirq_enable_ip = _THIS_IP_;
1300         p->softirq_enable_event = 0;
1301         p->softirq_disable_ip = 0;
1302         p->softirq_disable_event = 0;
1303         p->hardirq_context = 0;
1304         p->softirq_context = 0;
1305 #endif
1306 #ifdef CONFIG_LOCKDEP
1307         p->lockdep_depth = 0; /* no locks held yet */
1308         p->curr_chain_key = 0;
1309         p->lockdep_recursion = 0;
1310 #endif
1311
1312 #ifdef CONFIG_DEBUG_MUTEXES
1313         p->blocked_on = NULL; /* not blocked yet */
1314 #endif
1315 #ifdef CONFIG_MEMCG
1316         p->memcg_batch.do_batch = 0;
1317         p->memcg_batch.memcg = NULL;
1318 #endif
1319 #ifdef CONFIG_BCACHE
1320         p->sequential_io        = 0;
1321         p->sequential_io_avg    = 0;
1322 #endif
1323
1324         /* Perform scheduler related setup. Assign this task to a CPU. */
1325         sched_fork(p);
1326
1327         retval = perf_event_init_task(p);
1328         if (retval)
1329                 goto bad_fork_cleanup_policy;
1330         retval = audit_alloc(p);
1331         if (retval)
1332                 goto bad_fork_cleanup_perf;
1333         /* copy all the process information */
1334         retval = copy_semundo(clone_flags, p);
1335         if (retval)
1336                 goto bad_fork_cleanup_audit;
1337         retval = copy_files(clone_flags, p);
1338         if (retval)
1339                 goto bad_fork_cleanup_semundo;
1340         retval = copy_fs(clone_flags, p);
1341         if (retval)
1342                 goto bad_fork_cleanup_files;
1343         retval = copy_sighand(clone_flags, p);
1344         if (retval)
1345                 goto bad_fork_cleanup_fs;
1346         retval = copy_signal(clone_flags, p);
1347         if (retval)
1348                 goto bad_fork_cleanup_sighand;
1349         retval = copy_mm(clone_flags, p);
1350         if (retval)
1351                 goto bad_fork_cleanup_signal;
1352         retval = copy_namespaces(clone_flags, p);
1353         if (retval)
1354                 goto bad_fork_cleanup_mm;
1355         retval = copy_io(clone_flags, p);
1356         if (retval)
1357                 goto bad_fork_cleanup_namespaces;
1358         retval = copy_thread(clone_flags, stack_start, stack_size, p);
1359         if (retval)
1360                 goto bad_fork_cleanup_io;
1361
1362         if (pid != &init_struct_pid) {
1363                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1364                 if (IS_ERR(pid)) {
1365                         retval = PTR_ERR(pid);
1366                         goto bad_fork_cleanup_io;
1367                 }
1368         }
1369
1370         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1371         /*
1372          * Clear TID on mm_release()?
1373          */
1374         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1375 #ifdef CONFIG_BLOCK
1376         p->plug = NULL;
1377 #endif
1378 #ifdef CONFIG_FUTEX
1379         p->robust_list = NULL;
1380 #ifdef CONFIG_COMPAT
1381         p->compat_robust_list = NULL;
1382 #endif
1383         INIT_LIST_HEAD(&p->pi_state_list);
1384         p->pi_state_cache = NULL;
1385 #endif
1386         uprobe_copy_process(p);
1387         /*
1388          * sigaltstack should be cleared when sharing the same VM
1389          */
1390         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1391                 p->sas_ss_sp = p->sas_ss_size = 0;
1392
1393         /*
1394          * Syscall tracing and stepping should be turned off in the
1395          * child regardless of CLONE_PTRACE.
1396          */
1397         user_disable_single_step(p);
1398         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1399 #ifdef TIF_SYSCALL_EMU
1400         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1401 #endif
1402         clear_all_latency_tracing(p);
1403
1404         /* ok, now we should be set up.. */
1405         p->pid = pid_nr(pid);
1406         if (clone_flags & CLONE_THREAD) {
1407                 p->exit_signal = -1;
1408                 p->group_leader = current->group_leader;
1409                 p->tgid = current->tgid;
1410         } else {
1411                 if (clone_flags & CLONE_PARENT)
1412                         p->exit_signal = current->group_leader->exit_signal;
1413                 else
1414                         p->exit_signal = (clone_flags & CSIGNAL);
1415                 p->group_leader = p;
1416                 p->tgid = p->pid;
1417         }
1418
1419         p->pdeath_signal = 0;
1420         p->exit_state = 0;
1421
1422         p->nr_dirtied = 0;
1423         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1424         p->dirty_paused_when = 0;
1425
1426         INIT_LIST_HEAD(&p->thread_group);
1427         p->task_works = NULL;
1428
1429         /*
1430          * Make it visible to the rest of the system, but dont wake it up yet.
1431          * Need tasklist lock for parent etc handling!
1432          */
1433         write_lock_irq(&tasklist_lock);
1434
1435         /* CLONE_PARENT re-uses the old parent */
1436         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1437                 p->real_parent = current->real_parent;
1438                 p->parent_exec_id = current->parent_exec_id;
1439         } else {
1440                 p->real_parent = current;
1441                 p->parent_exec_id = current->self_exec_id;
1442         }
1443
1444         spin_lock(&current->sighand->siglock);
1445
1446         /*
1447          * Process group and session signals need to be delivered to just the
1448          * parent before the fork or both the parent and the child after the
1449          * fork. Restart if a signal comes in before we add the new process to
1450          * it's process group.
1451          * A fatal signal pending means that current will exit, so the new
1452          * thread can't slip out of an OOM kill (or normal SIGKILL).
1453         */
1454         recalc_sigpending();
1455         if (signal_pending(current)) {
1456                 spin_unlock(&current->sighand->siglock);
1457                 write_unlock_irq(&tasklist_lock);
1458                 retval = -ERESTARTNOINTR;
1459                 goto bad_fork_free_pid;
1460         }
1461
1462         if (likely(p->pid)) {
1463                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1464
1465                 init_task_pid(p, PIDTYPE_PID, pid);
1466                 if (thread_group_leader(p)) {
1467                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1468                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1469
1470                         if (is_child_reaper(pid)) {
1471                                 ns_of_pid(pid)->child_reaper = p;
1472                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1473                         }
1474
1475                         p->signal->leader_pid = pid;
1476                         p->signal->tty = tty_kref_get(current->signal->tty);
1477                         list_add_tail(&p->sibling, &p->real_parent->children);
1478                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1479                         attach_pid(p, PIDTYPE_PGID);
1480                         attach_pid(p, PIDTYPE_SID);
1481                         __this_cpu_inc(process_counts);
1482                 } else {
1483                         current->signal->nr_threads++;
1484                         atomic_inc(&current->signal->live);
1485                         atomic_inc(&current->signal->sigcnt);
1486                         list_add_tail_rcu(&p->thread_group,
1487                                           &p->group_leader->thread_group);
1488                         list_add_tail_rcu(&p->thread_node,
1489                                           &p->signal->thread_head);
1490                 }
1491                 attach_pid(p, PIDTYPE_PID);
1492                 nr_threads++;
1493         }
1494
1495         total_forks++;
1496         spin_unlock(&current->sighand->siglock);
1497         syscall_tracepoint_update(p);
1498         write_unlock_irq(&tasklist_lock);
1499
1500         proc_fork_connector(p);
1501         cgroup_post_fork(p);
1502         if (clone_flags & CLONE_THREAD)
1503                 threadgroup_change_end(current);
1504         perf_event_fork(p);
1505
1506         trace_task_newtask(p, clone_flags);
1507
1508         return p;
1509
1510 bad_fork_free_pid:
1511         if (pid != &init_struct_pid)
1512                 free_pid(pid);
1513 bad_fork_cleanup_io:
1514         if (p->io_context)
1515                 exit_io_context(p);
1516 bad_fork_cleanup_namespaces:
1517         exit_task_namespaces(p);
1518 bad_fork_cleanup_mm:
1519         if (p->mm)
1520                 mmput(p->mm);
1521 bad_fork_cleanup_signal:
1522         if (!(clone_flags & CLONE_THREAD))
1523                 free_signal_struct(p->signal);
1524 bad_fork_cleanup_sighand:
1525         __cleanup_sighand(p->sighand);
1526 bad_fork_cleanup_fs:
1527         exit_fs(p); /* blocking */
1528 bad_fork_cleanup_files:
1529         exit_files(p); /* blocking */
1530 bad_fork_cleanup_semundo:
1531         exit_sem(p);
1532 bad_fork_cleanup_audit:
1533         audit_free(p);
1534 bad_fork_cleanup_perf:
1535         perf_event_free_task(p);
1536 bad_fork_cleanup_policy:
1537 #ifdef CONFIG_NUMA
1538         mpol_put(p->mempolicy);
1539 bad_fork_cleanup_cgroup:
1540 #endif
1541         if (clone_flags & CLONE_THREAD)
1542                 threadgroup_change_end(current);
1543         cgroup_exit(p, 0);
1544         delayacct_tsk_free(p);
1545         module_put(task_thread_info(p)->exec_domain->module);
1546 bad_fork_cleanup_count:
1547         atomic_dec(&p->cred->user->processes);
1548         exit_creds(p);
1549 bad_fork_free:
1550         free_task(p);
1551 fork_out:
1552         return ERR_PTR(retval);
1553 }
1554
1555 static inline void init_idle_pids(struct pid_link *links)
1556 {
1557         enum pid_type type;
1558
1559         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1560                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1561                 links[type].pid = &init_struct_pid;
1562         }
1563 }
1564
1565 struct task_struct *fork_idle(int cpu)
1566 {
1567         struct task_struct *task;
1568         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0);
1569         if (!IS_ERR(task)) {
1570                 init_idle_pids(task->pids);
1571                 init_idle(task, cpu);
1572         }
1573
1574         return task;
1575 }
1576
1577 /*
1578  *  Ok, this is the main fork-routine.
1579  *
1580  * It copies the process, and if successful kick-starts
1581  * it and waits for it to finish using the VM if required.
1582  */
1583 long do_fork(unsigned long clone_flags,
1584               unsigned long stack_start,
1585               unsigned long stack_size,
1586               int __user *parent_tidptr,
1587               int __user *child_tidptr)
1588 {
1589         struct task_struct *p;
1590         int trace = 0;
1591         long nr;
1592
1593         /*
1594          * Determine whether and which event to report to ptracer.  When
1595          * called from kernel_thread or CLONE_UNTRACED is explicitly
1596          * requested, no event is reported; otherwise, report if the event
1597          * for the type of forking is enabled.
1598          */
1599         if (!(clone_flags & CLONE_UNTRACED)) {
1600                 if (clone_flags & CLONE_VFORK)
1601                         trace = PTRACE_EVENT_VFORK;
1602                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1603                         trace = PTRACE_EVENT_CLONE;
1604                 else
1605                         trace = PTRACE_EVENT_FORK;
1606
1607                 if (likely(!ptrace_event_enabled(current, trace)))
1608                         trace = 0;
1609         }
1610
1611         p = copy_process(clone_flags, stack_start, stack_size,
1612                          child_tidptr, NULL, trace);
1613         /*
1614          * Do this prior waking up the new thread - the thread pointer
1615          * might get invalid after that point, if the thread exits quickly.
1616          */
1617         if (!IS_ERR(p)) {
1618                 struct completion vfork;
1619                 struct pid *pid;
1620
1621                 trace_sched_process_fork(current, p);
1622
1623                 pid = get_task_pid(p, PIDTYPE_PID);
1624                 nr = pid_vnr(pid);
1625
1626                 if (clone_flags & CLONE_PARENT_SETTID)
1627                         put_user(nr, parent_tidptr);
1628
1629                 if (clone_flags & CLONE_VFORK) {
1630                         p->vfork_done = &vfork;
1631                         init_completion(&vfork);
1632                         get_task_struct(p);
1633                 }
1634
1635                 wake_up_new_task(p);
1636
1637                 /* forking complete and child started to run, tell ptracer */
1638                 if (unlikely(trace))
1639                         ptrace_event_pid(trace, pid);
1640
1641                 if (clone_flags & CLONE_VFORK) {
1642                         if (!wait_for_vfork_done(p, &vfork))
1643                                 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1644                 }
1645
1646                 put_pid(pid);
1647         } else {
1648                 nr = PTR_ERR(p);
1649         }
1650         return nr;
1651 }
1652
1653 /*
1654  * Create a kernel thread.
1655  */
1656 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1657 {
1658         return do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1659                 (unsigned long)arg, NULL, NULL);
1660 }
1661
1662 #ifdef __ARCH_WANT_SYS_FORK
1663 SYSCALL_DEFINE0(fork)
1664 {
1665 #ifdef CONFIG_MMU
1666         return do_fork(SIGCHLD, 0, 0, NULL, NULL);
1667 #else
1668         /* can not support in nommu mode */
1669         return(-EINVAL);
1670 #endif
1671 }
1672 #endif
1673
1674 #ifdef __ARCH_WANT_SYS_VFORK
1675 SYSCALL_DEFINE0(vfork)
1676 {
1677         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1678                         0, NULL, NULL);
1679 }
1680 #endif
1681
1682 #ifdef __ARCH_WANT_SYS_CLONE
1683 #ifdef CONFIG_CLONE_BACKWARDS
1684 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1685                  int __user *, parent_tidptr,
1686                  int, tls_val,
1687                  int __user *, child_tidptr)
1688 #elif defined(CONFIG_CLONE_BACKWARDS2)
1689 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1690                  int __user *, parent_tidptr,
1691                  int __user *, child_tidptr,
1692                  int, tls_val)
1693 #elif defined(CONFIG_CLONE_BACKWARDS3)
1694 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1695                 int, stack_size,
1696                 int __user *, parent_tidptr,
1697                 int __user *, child_tidptr,
1698                 int, tls_val)
1699 #else
1700 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1701                  int __user *, parent_tidptr,
1702                  int __user *, child_tidptr,
1703                  int, tls_val)
1704 #endif
1705 {
1706         return do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr);
1707 }
1708 #endif
1709
1710 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1711 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1712 #endif
1713
1714 static void sighand_ctor(void *data)
1715 {
1716         struct sighand_struct *sighand = data;
1717
1718         spin_lock_init(&sighand->siglock);
1719         init_waitqueue_head(&sighand->signalfd_wqh);
1720 }
1721
1722 void __init proc_caches_init(void)
1723 {
1724         sighand_cachep = kmem_cache_create("sighand_cache",
1725                         sizeof(struct sighand_struct), 0,
1726                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1727                         SLAB_NOTRACK, sighand_ctor);
1728         signal_cachep = kmem_cache_create("signal_cache",
1729                         sizeof(struct signal_struct), 0,
1730                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1731         files_cachep = kmem_cache_create("files_cache",
1732                         sizeof(struct files_struct), 0,
1733                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1734         fs_cachep = kmem_cache_create("fs_cache",
1735                         sizeof(struct fs_struct), 0,
1736                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1737         /*
1738          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1739          * whole struct cpumask for the OFFSTACK case. We could change
1740          * this to *only* allocate as much of it as required by the
1741          * maximum number of CPU's we can ever have.  The cpumask_allocation
1742          * is at the end of the structure, exactly for that reason.
1743          */
1744         mm_cachep = kmem_cache_create("mm_struct",
1745                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1746                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1747         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1748         mmap_init();
1749         nsproxy_cache_init();
1750 }
1751
1752 /*
1753  * Check constraints on flags passed to the unshare system call.
1754  */
1755 static int check_unshare_flags(unsigned long unshare_flags)
1756 {
1757         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1758                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1759                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1760                                 CLONE_NEWUSER|CLONE_NEWPID))
1761                 return -EINVAL;
1762         /*
1763          * Not implemented, but pretend it works if there is nothing
1764          * to unshare.  Note that unsharing the address space or the
1765          * signal handlers also need to unshare the signal queues (aka
1766          * CLONE_THREAD).
1767          */
1768         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1769                 if (!thread_group_empty(current))
1770                         return -EINVAL;
1771         }
1772         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
1773                 if (atomic_read(&current->sighand->count) > 1)
1774                         return -EINVAL;
1775         }
1776         if (unshare_flags & CLONE_VM) {
1777                 if (!current_is_single_threaded())
1778                         return -EINVAL;
1779         }
1780
1781         return 0;
1782 }
1783
1784 /*
1785  * Unshare the filesystem structure if it is being shared
1786  */
1787 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1788 {
1789         struct fs_struct *fs = current->fs;
1790
1791         if (!(unshare_flags & CLONE_FS) || !fs)
1792                 return 0;
1793
1794         /* don't need lock here; in the worst case we'll do useless copy */
1795         if (fs->users == 1)
1796                 return 0;
1797
1798         *new_fsp = copy_fs_struct(fs);
1799         if (!*new_fsp)
1800                 return -ENOMEM;
1801
1802         return 0;
1803 }
1804
1805 /*
1806  * Unshare file descriptor table if it is being shared
1807  */
1808 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1809 {
1810         struct files_struct *fd = current->files;
1811         int error = 0;
1812
1813         if ((unshare_flags & CLONE_FILES) &&
1814             (fd && atomic_read(&fd->count) > 1)) {
1815                 *new_fdp = dup_fd(fd, &error);
1816                 if (!*new_fdp)
1817                         return error;
1818         }
1819
1820         return 0;
1821 }
1822
1823 /*
1824  * unshare allows a process to 'unshare' part of the process
1825  * context which was originally shared using clone.  copy_*
1826  * functions used by do_fork() cannot be used here directly
1827  * because they modify an inactive task_struct that is being
1828  * constructed. Here we are modifying the current, active,
1829  * task_struct.
1830  */
1831 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1832 {
1833         struct fs_struct *fs, *new_fs = NULL;
1834         struct files_struct *fd, *new_fd = NULL;
1835         struct cred *new_cred = NULL;
1836         struct nsproxy *new_nsproxy = NULL;
1837         int do_sysvsem = 0;
1838         int err;
1839
1840         /*
1841          * If unsharing a user namespace must also unshare the thread.
1842          */
1843         if (unshare_flags & CLONE_NEWUSER)
1844                 unshare_flags |= CLONE_THREAD | CLONE_FS;
1845         /*
1846          * If unsharing vm, must also unshare signal handlers.
1847          */
1848         if (unshare_flags & CLONE_VM)
1849                 unshare_flags |= CLONE_SIGHAND;
1850         /*
1851          * If unsharing a signal handlers, must also unshare the signal queues.
1852          */
1853         if (unshare_flags & CLONE_SIGHAND)
1854                 unshare_flags |= CLONE_THREAD;
1855         /*
1856          * If unsharing namespace, must also unshare filesystem information.
1857          */
1858         if (unshare_flags & CLONE_NEWNS)
1859                 unshare_flags |= CLONE_FS;
1860
1861         err = check_unshare_flags(unshare_flags);
1862         if (err)
1863                 goto bad_unshare_out;
1864         /*
1865          * CLONE_NEWIPC must also detach from the undolist: after switching
1866          * to a new ipc namespace, the semaphore arrays from the old
1867          * namespace are unreachable.
1868          */
1869         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1870                 do_sysvsem = 1;
1871         err = unshare_fs(unshare_flags, &new_fs);
1872         if (err)
1873                 goto bad_unshare_out;
1874         err = unshare_fd(unshare_flags, &new_fd);
1875         if (err)
1876                 goto bad_unshare_cleanup_fs;
1877         err = unshare_userns(unshare_flags, &new_cred);
1878         if (err)
1879                 goto bad_unshare_cleanup_fd;
1880         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1881                                          new_cred, new_fs);
1882         if (err)
1883                 goto bad_unshare_cleanup_cred;
1884
1885         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
1886                 if (do_sysvsem) {
1887                         /*
1888                          * CLONE_SYSVSEM is equivalent to sys_exit().
1889                          */
1890                         exit_sem(current);
1891                 }
1892
1893                 if (new_nsproxy)
1894                         switch_task_namespaces(current, new_nsproxy);
1895
1896                 task_lock(current);
1897
1898                 if (new_fs) {
1899                         fs = current->fs;
1900                         spin_lock(&fs->lock);
1901                         current->fs = new_fs;
1902                         if (--fs->users)
1903                                 new_fs = NULL;
1904                         else
1905                                 new_fs = fs;
1906                         spin_unlock(&fs->lock);
1907                 }
1908
1909                 if (new_fd) {
1910                         fd = current->files;
1911                         current->files = new_fd;
1912                         new_fd = fd;
1913                 }
1914
1915                 task_unlock(current);
1916
1917                 if (new_cred) {
1918                         /* Install the new user namespace */
1919                         commit_creds(new_cred);
1920                         new_cred = NULL;
1921                 }
1922         }
1923
1924 bad_unshare_cleanup_cred:
1925         if (new_cred)
1926                 put_cred(new_cred);
1927 bad_unshare_cleanup_fd:
1928         if (new_fd)
1929                 put_files_struct(new_fd);
1930
1931 bad_unshare_cleanup_fs:
1932         if (new_fs)
1933                 free_fs_struct(new_fs);
1934
1935 bad_unshare_out:
1936         return err;
1937 }
1938
1939 /*
1940  *      Helper to unshare the files of the current task.
1941  *      We don't want to expose copy_files internals to
1942  *      the exec layer of the kernel.
1943  */
1944
1945 int unshare_files(struct files_struct **displaced)
1946 {
1947         struct task_struct *task = current;
1948         struct files_struct *copy = NULL;
1949         int error;
1950
1951         error = unshare_fd(CLONE_FILES, &copy);
1952         if (error || !copy) {
1953                 *displaced = NULL;
1954                 return error;
1955         }
1956         *displaced = task->files;
1957         task_lock(task);
1958         task->files = copy;
1959         task_unlock(task);
1960         return 0;
1961 }