kernel/fork.c

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