kernel/fork.c

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