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
 308         /*
 309          * We defer looking at err, because we will need this setup
 310          * for the clean up path to work correctly.
 311          */
 312         tsk->stack = ti;
 313         setup_thread_stack(tsk, orig);
 314
 315         if (err)
 316                 goto out;
 317
 318         clear_user_return_notifier(tsk);
 319         clear_tsk_need_resched(tsk);
 320         stackend = end_of_stack(tsk);
 321         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 322
 323 #ifdef CONFIG_CC_STACKPROTECTOR
 324         tsk->stack_canary = get_random_int();
 325 #endif
 326
 327         /*
 328          * One for us, one for whoever does the "release_task()" (usually
 329          * parent)
 330          */
 331         atomic_set(&tsk->usage, 2);
 332 #ifdef CONFIG_BLK_DEV_IO_TRACE
 333         tsk->btrace_seq = 0;
 334 #endif
 335         tsk->splice_pipe = NULL;
 336
 337         account_kernel_stack(ti, 1);
 338
 339         return tsk;
 340
 341 out:
 342         free_thread_info(ti);
 343         free_task_struct(tsk);
 344         return NULL;
 345 }
 346
 347 #ifdef CONFIG_MMU
 348 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 349 {
 350         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 351         struct rb_node **rb_link, *rb_parent;
 352         int retval;
 353         unsigned long charge;
 354         struct mempolicy *pol;
 355
 356         down_write(&oldmm->mmap_sem);
 357         flush_cache_dup_mm(oldmm);
 358         /*
 359          * Not linked in yet - no deadlock potential:
 360          */
 361         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 362
 363         mm->locked_vm = 0;
 364         mm->mmap = NULL;
 365         mm->mmap_cache = NULL;
 366         mm->free_area_cache = oldmm->mmap_base;
 367         mm->cached_hole_size = ~0UL;
 368         mm->map_count = 0;
 369         cpumask_clear(mm_cpumask(mm));
 370         mm->mm_rb = RB_ROOT;
 371         rb_link = &mm->mm_rb.rb_node;
 372         rb_parent = NULL;
 373         pprev = &mm->mmap;
 374         retval = ksm_fork(mm, oldmm);
 375         if (retval)
 376                 goto out;
 377         retval = khugepaged_fork(mm, oldmm);
 378         if (retval)
 379                 goto out;
 380
 381         prev = NULL;
 382         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 383                 struct file *file;
 384
 385                 if (mpnt->vm_flags & VM_DONTCOPY) {
 386                         long pages = vma_pages(mpnt);
 387                         mm->total_vm -= pages;
 388                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 389                                                                 -pages);
 390                         continue;
 391                 }
 392                 charge = 0;
 393                 if (mpnt->vm_flags & VM_ACCOUNT) {
 394                         unsigned long len;
 395                         len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 396                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 397                                 goto fail_nomem;
 398                         charge = len;
 399                 }
 400                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 401                 if (!tmp)
 402                         goto fail_nomem;
 403                 *tmp = *mpnt;
 404                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 405                 pol = mpol_dup(vma_policy(mpnt));
 406                 retval = PTR_ERR(pol);
 407                 if (IS_ERR(pol))
 408                         goto fail_nomem_policy;
 409                 vma_set_policy(tmp, pol);
 410                 tmp->vm_mm = mm;
 411                 if (anon_vma_fork(tmp, mpnt))
 412                         goto fail_nomem_anon_vma_fork;
 413                 tmp->vm_flags &= ~VM_LOCKED;
 414                 tmp->vm_next = tmp->vm_prev = NULL;
 415                 file = tmp->vm_file;
 416                 if (file) {
 417                         struct inode *inode = file->f_path.dentry->d_inode;
 418                         struct address_space *mapping = file->f_mapping;
 419
 420                         get_file(file);
 421                         if (tmp->vm_flags & VM_DENYWRITE)
 422                                 atomic_dec(&inode->i_writecount);
 423                         mutex_lock(&mapping->i_mmap_mutex);
 424                         if (tmp->vm_flags & VM_SHARED)
 425                                 mapping->i_mmap_writable++;
 426                         flush_dcache_mmap_lock(mapping);
 427                         /* insert tmp into the share list, just after mpnt */
 428                         vma_prio_tree_add(tmp, mpnt);
 429                         flush_dcache_mmap_unlock(mapping);
 430                         mutex_unlock(&mapping->i_mmap_mutex);
 431                 }
 432
 433                 /*
 434                  * Clear hugetlb-related page reserves for children. This only
 435                  * affects MAP_PRIVATE mappings. Faults generated by the child
 436                  * are not guaranteed to succeed, even if read-only
 437                  */
 438                 if (is_vm_hugetlb_page(tmp))
 439                         reset_vma_resv_huge_pages(tmp);
 440
 441                 /*
 442                  * Link in the new vma and copy the page table entries.
 443                  */
 444                 *pprev = tmp;
 445                 pprev = &tmp->vm_next;
 446                 tmp->vm_prev = prev;
 447                 prev = tmp;
 448
 449                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 450                 rb_link = &tmp->vm_rb.rb_right;
 451                 rb_parent = &tmp->vm_rb;
 452
 453                 mm->map_count++;
 454                 retval = copy_page_range(mm, oldmm, mpnt);
 455
 456                 if (tmp->vm_ops && tmp->vm_ops->open)
 457                         tmp->vm_ops->open(tmp);
 458
 459                 if (retval)
 460                         goto out;
 461
 462                 if (file && uprobe_mmap(tmp))
 463                         goto out;
 464         }
 465         /* a new mm has just been created */
 466         arch_dup_mmap(oldmm, mm);
 467         retval = 0;
 468 out:
 469         up_write(&mm->mmap_sem);
 470         flush_tlb_mm(oldmm);
 471         up_write(&oldmm->mmap_sem);
 472         return retval;
 473 fail_nomem_anon_vma_fork:
 474         mpol_put(pol);
 475 fail_nomem_policy:
 476         kmem_cache_free(vm_area_cachep, tmp);
 477 fail_nomem:
 478         retval = -ENOMEM;
 479         vm_unacct_memory(charge);
 480         goto out;
 481 }
 482
 483 static inline int mm_alloc_pgd(struct mm_struct *mm)
 484 {
 485         mm->pgd = pgd_alloc(mm);
 486         if (unlikely(!mm->pgd))
 487                 return -ENOMEM;
 488         return 0;
 489 }
 490
 491 static inline void mm_free_pgd(struct mm_struct *mm)
 492 {
 493         pgd_free(mm, mm->pgd);
 494 }
 495 #else
 496 #define dup_mmap(mm, oldmm)     (0)
 497 #define mm_alloc_pgd(mm)        (0)
 498 #define mm_free_pgd(mm)
 499 #endif /* CONFIG_MMU */
 500
 501 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 502
 503 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 504 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 505
 506 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 507
 508 static int __init coredump_filter_setup(char *s)
 509 {
 510         default_dump_filter =
 511                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 512                 MMF_DUMP_FILTER_MASK;
 513         return 1;
 514 }
 515
 516 __setup("coredump_filter=", coredump_filter_setup);
 517
 518 #include <linux/init_task.h>
 519
 520 static void mm_init_aio(struct mm_struct *mm)
 521 {
 522 #ifdef CONFIG_AIO
 523         spin_lock_init(&mm->ioctx_lock);
 524         INIT_HLIST_HEAD(&mm->ioctx_list);
 525 #endif
 526 }
 527
 528 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
 529 {
 530         atomic_set(&mm->mm_users, 1);
 531         atomic_set(&mm->mm_count, 1);
 532         init_rwsem(&mm->mmap_sem);
 533         INIT_LIST_HEAD(&mm->mmlist);
 534         mm->flags = (current->mm) ?
 535                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
 536         mm->core_state = NULL;
 537         mm->nr_ptes = 0;
 538         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 539         spin_lock_init(&mm->page_table_lock);
 540         mm->free_area_cache = TASK_UNMAPPED_BASE;
 541         mm->cached_hole_size = ~0UL;
 542         mm_init_aio(mm);
 543         mm_init_owner(mm, p);
 544
 545         if (likely(!mm_alloc_pgd(mm))) {
 546                 mm->def_flags = 0;
 547                 mmu_notifier_mm_init(mm);
 548                 return mm;
 549         }
 550
 551         free_mm(mm);
 552         return NULL;
 553 }
 554
 555 static void check_mm(struct mm_struct *mm)
 556 {
 557         int i;
 558
 559         for (i = 0; i < NR_MM_COUNTERS; i++) {
 560                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 561
 562                 if (unlikely(x))
 563                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 564                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 565         }
 566
 567 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 568         VM_BUG_ON(mm->pmd_huge_pte);
 569 #endif
 570 }
 571
 572 /*
 573  * Allocate and initialize an mm_struct.
 574  */
 575 struct mm_struct *mm_alloc(void)
 576 {
 577         struct mm_struct *mm;
 578
 579         mm = allocate_mm();
 580         if (!mm)
 581                 return NULL;
 582
 583         memset(mm, 0, sizeof(*mm));
 584         mm_init_cpumask(mm);
 585         return mm_init(mm, current);
 586 }
 587
 588 /*
 589  * Called when the last reference to the mm
 590  * is dropped: either by a lazy thread or by
 591  * mmput. Free the page directory and the mm.
 592  */
 593 void __mmdrop(struct mm_struct *mm)
 594 {
 595         BUG_ON(mm == &init_mm);
 596         mm_free_pgd(mm);
 597         destroy_context(mm);
 598         mmu_notifier_mm_destroy(mm);
 599         check_mm(mm);
 600         free_mm(mm);
 601 }
 602 EXPORT_SYMBOL_GPL(__mmdrop);
 603
 604 /*
 605  * Decrement the use count and release all resources for an mm.
 606  */
 607 void mmput(struct mm_struct *mm)
 608 {
 609         might_sleep();
 610
 611         if (atomic_dec_and_test(&mm->mm_users)) {
 612                 uprobe_clear_state(mm);
 613                 exit_aio(mm);
 614                 ksm_exit(mm);
 615                 khugepaged_exit(mm); /* must run before exit_mmap */
 616                 exit_mmap(mm);
 617                 set_mm_exe_file(mm, NULL);
 618                 if (!list_empty(&mm->mmlist)) {
 619                         spin_lock(&mmlist_lock);
 620                         list_del(&mm->mmlist);
 621                         spin_unlock(&mmlist_lock);
 622                 }
 623                 if (mm->binfmt)
 624                         module_put(mm->binfmt->module);
 625                 mmdrop(mm);
 626         }
 627 }
 628 EXPORT_SYMBOL_GPL(mmput);
 629
 630 /*
 631  * We added or removed a vma mapping the executable. The vmas are only mapped
 632  * during exec and are not mapped with the mmap system call.
 633  * Callers must hold down_write() on the mm's mmap_sem for these
 634  */
 635 void added_exe_file_vma(struct mm_struct *mm)
 636 {
 637         mm->num_exe_file_vmas++;
 638 }
 639
 640 void removed_exe_file_vma(struct mm_struct *mm)
 641 {
 642         mm->num_exe_file_vmas--;
 643         if ((mm->num_exe_file_vmas == 0) && mm->exe_file) {
 644                 fput(mm->exe_file);
 645                 mm->exe_file = NULL;
 646         }
 647
 648 }
 649
 650 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 651 {
 652         if (new_exe_file)
 653                 get_file(new_exe_file);
 654         if (mm->exe_file)
 655                 fput(mm->exe_file);
 656         mm->exe_file = new_exe_file;
 657         mm->num_exe_file_vmas = 0;
 658 }
 659
 660 struct file *get_mm_exe_file(struct mm_struct *mm)
 661 {
 662         struct file *exe_file;
 663
 664         /* We need mmap_sem to protect against races with removal of
 665          * VM_EXECUTABLE vmas */
 666         down_read(&mm->mmap_sem);
 667         exe_file = mm->exe_file;
 668         if (exe_file)
 669                 get_file(exe_file);
 670         up_read(&mm->mmap_sem);
 671         return exe_file;
 672 }
 673
 674 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
 675 {
 676         /* It's safe to write the exe_file pointer without exe_file_lock because
 677          * this is called during fork when the task is not yet in /proc */
 678         newmm->exe_file = get_mm_exe_file(oldmm);
 679 }
 680
 681 /**
 682  * get_task_mm - acquire a reference to the task's mm
 683  *
 684  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 685  * this kernel workthread has transiently adopted a user mm with use_mm,
 686  * to do its AIO) is not set and if so returns a reference to it, after
 687  * bumping up the use count.  User must release the mm via mmput()
 688  * after use.  Typically used by /proc and ptrace.
 689  */
 690 struct mm_struct *get_task_mm(struct task_struct *task)
 691 {
 692         struct mm_struct *mm;
 693
 694         task_lock(task);
 695         mm = task->mm;
 696         if (mm) {
 697                 if (task->flags & PF_KTHREAD)
 698                         mm = NULL;
 699                 else
 700                         atomic_inc(&mm->mm_users);
 701         }
 702         task_unlock(task);
 703         return mm;
 704 }
 705 EXPORT_SYMBOL_GPL(get_task_mm);
 706
 707 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 708 {
 709         struct mm_struct *mm;
 710         int err;
 711
 712         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 713         if (err)
 714                 return ERR_PTR(err);
 715
 716         mm = get_task_mm(task);
 717         if (mm && mm != current->mm &&
 718                         !ptrace_may_access(task, mode)) {
 719                 mmput(mm);
 720                 mm = ERR_PTR(-EACCES);
 721         }
 722         mutex_unlock(&task->signal->cred_guard_mutex);
 723
 724         return mm;
 725 }
 726
 727 static void complete_vfork_done(struct task_struct *tsk)
 728 {
 729         struct completion *vfork;
 730
 731         task_lock(tsk);
 732         vfork = tsk->vfork_done;
 733         if (likely(vfork)) {
 734                 tsk->vfork_done = NULL;
 735                 complete(vfork);
 736         }
 737         task_unlock(tsk);
 738 }
 739
 740 static int wait_for_vfork_done(struct task_struct *child,
 741                                 struct completion *vfork)
 742 {
 743         int killed;
 744
 745         freezer_do_not_count();
 746         killed = wait_for_completion_killable(vfork);
 747         freezer_count();
 748
 749         if (killed) {
 750                 task_lock(child);
 751                 child->vfork_done = NULL;
 752                 task_unlock(child);
 753         }
 754
 755         put_task_struct(child);
 756         return killed;
 757 }
 758
 759 /* Please note the differences between mmput and mm_release.
 760  * mmput is called whenever we stop holding onto a mm_struct,
 761  * error success whatever.
 762  *
 763  * mm_release is called after a mm_struct has been removed
 764  * from the current process.
 765  *
 766  * This difference is important for error handling, when we
 767  * only half set up a mm_struct for a new process and need to restore
 768  * the old one.  Because we mmput the new mm_struct before
 769  * restoring the old one. . .
 770  * Eric Biederman 10 January 1998
 771  */
 772 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 773 {
 774         /* Get rid of any futexes when releasing the mm */
 775 #ifdef CONFIG_FUTEX
 776         if (unlikely(tsk->robust_list)) {
 777                 exit_robust_list(tsk);
 778                 tsk->robust_list = NULL;
 779         }
 780 #ifdef CONFIG_COMPAT
 781         if (unlikely(tsk->compat_robust_list)) {
 782                 compat_exit_robust_list(tsk);
 783                 tsk->compat_robust_list = NULL;
 784         }
 785 #endif
 786         if (unlikely(!list_empty(&tsk->pi_state_list)))
 787                 exit_pi_state_list(tsk);
 788 #endif
 789
 790         uprobe_free_utask(tsk);
 791
 792         /* Get rid of any cached register state */
 793         deactivate_mm(tsk, mm);
 794
 795         /*
 796          * If we're exiting normally, clear a user-space tid field if
 797          * requested.  We leave this alone when dying by signal, to leave
 798          * the value intact in a core dump, and to save the unnecessary
 799          * trouble, say, a killed vfork parent shouldn't touch this mm.
 800          * Userland only wants this done for a sys_exit.
 801          */
 802         if (tsk->clear_child_tid) {
 803                 if (!(tsk->flags & PF_SIGNALED) &&
 804                     atomic_read(&mm->mm_users) > 1) {
 805                         /*
 806                          * We don't check the error code - if userspace has
 807                          * not set up a proper pointer then tough luck.
 808                          */
 809                         put_user(0, tsk->clear_child_tid);
 810                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 811                                         1, NULL, NULL, 0);
 812                 }
 813                 tsk->clear_child_tid = NULL;
 814         }
 815
 816         /*
 817          * All done, finally we can wake up parent and return this mm to him.
 818          * Also kthread_stop() uses this completion for synchronization.
 819          */
 820         if (tsk->vfork_done)
 821                 complete_vfork_done(tsk);
 822 }
 823
 824 /*
 825  * Allocate a new mm structure and copy contents from the
 826  * mm structure of the passed in task structure.
 827  */
 828 struct mm_struct *dup_mm(struct task_struct *tsk)
 829 {
 830         struct mm_struct *mm, *oldmm = current->mm;
 831         int err;
 832
 833         if (!oldmm)
 834                 return NULL;
 835
 836         mm = allocate_mm();
 837         if (!mm)
 838                 goto fail_nomem;
 839
 840         memcpy(mm, oldmm, sizeof(*mm));
 841         mm_init_cpumask(mm);
 842
 843 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 844         mm->pmd_huge_pte = NULL;
 845 #endif
 846         uprobe_reset_state(mm);
 847
 848         if (!mm_init(mm, tsk))
 849                 goto fail_nomem;
 850
 851         if (init_new_context(tsk, mm))
 852                 goto fail_nocontext;
 853
 854         dup_mm_exe_file(oldmm, mm);
 855
 856         err = dup_mmap(mm, oldmm);
 857         if (err)
 858                 goto free_pt;
 859
 860         mm->hiwater_rss = get_mm_rss(mm);
 861         mm->hiwater_vm = mm->total_vm;
 862
 863         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 864                 goto free_pt;
 865
 866         return mm;
 867
 868 free_pt:
 869         /* don't put binfmt in mmput, we haven't got module yet */
 870         mm->binfmt = NULL;
 871         mmput(mm);
 872
 873 fail_nomem:
 874         return NULL;
 875
 876 fail_nocontext:
 877         /*
 878          * If init_new_context() failed, we cannot use mmput() to free the mm
 879          * because it calls destroy_context()
 880          */
 881         mm_free_pgd(mm);
 882         free_mm(mm);
 883         return NULL;
 884 }
 885
 886 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 887 {
 888         struct mm_struct *mm, *oldmm;
 889         int retval;
 890
 891         tsk->min_flt = tsk->maj_flt = 0;
 892         tsk->nvcsw = tsk->nivcsw = 0;
 893 #ifdef CONFIG_DETECT_HUNG_TASK
 894         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 895 #endif
 896
 897         tsk->mm = NULL;
 898         tsk->active_mm = NULL;
 899
 900         /*
 901          * Are we cloning a kernel thread?
 902          *
 903          * We need to steal a active VM for that..
 904          */
 905         oldmm = current->mm;
 906         if (!oldmm)
 907                 return 0;
 908
 909         if (clone_flags & CLONE_VM) {
 910                 atomic_inc(&oldmm->mm_users);
 911                 mm = oldmm;
 912                 goto good_mm;
 913         }
 914
 915         retval = -ENOMEM;
 916         mm = dup_mm(tsk);
 917         if (!mm)
 918                 goto fail_nomem;
 919
 920 good_mm:
 921         tsk->mm = mm;
 922         tsk->active_mm = mm;
 923         return 0;
 924
 925 fail_nomem:
 926         return retval;
 927 }
 928
 929 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 930 {
 931         struct fs_struct *fs = current->fs;
 932         if (clone_flags & CLONE_FS) {
 933                 /* tsk->fs is already what we want */
 934                 spin_lock(&fs->lock);
 935                 if (fs->in_exec) {
 936                         spin_unlock(&fs->lock);
 937                         return -EAGAIN;
 938                 }
 939                 fs->users++;
 940                 spin_unlock(&fs->lock);
 941                 return 0;
 942         }
 943         tsk->fs = copy_fs_struct(fs);
 944         if (!tsk->fs)
 945                 return -ENOMEM;
 946         return 0;
 947 }
 948
 949 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
 950 {
 951         struct files_struct *oldf, *newf;
 952         int error = 0;
 953
 954         /*
 955          * A background process may not have any files ...
 956          */
 957         oldf = current->files;
 958         if (!oldf)
 959                 goto out;
 960
 961         if (clone_flags & CLONE_FILES) {
 962                 atomic_inc(&oldf->count);
 963                 goto out;
 964         }
 965
 966         newf = dup_fd(oldf, &error);
 967         if (!newf)
 968                 goto out;
 969
 970         tsk->files = newf;
 971         error = 0;
 972 out:
 973         return error;
 974 }
 975
 976 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 977 {
 978 #ifdef CONFIG_BLOCK
 979         struct io_context *ioc = current->io_context;
 980         struct io_context *new_ioc;
 981
 982         if (!ioc)
 983                 return 0;
 984         /*
 985          * Share io context with parent, if CLONE_IO is set
 986          */
 987         if (clone_flags & CLONE_IO) {
 988                 ioc_task_link(ioc);
 989                 tsk->io_context = ioc;
 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         p->task_works = NULL;
1424
1425         /* Now that the task is set up, run cgroup callbacks if
1426          * necessary. We need to run them before the task is visible
1427          * on the tasklist. */
1428         cgroup_fork_callbacks(p);
1429         cgroup_callbacks_done = 1;
1430
1431         /* Need tasklist lock for parent etc handling! */
1432         write_lock_irq(&tasklist_lock);
1433
1434         /* CLONE_PARENT re-uses the old parent */
1435         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1436                 p->real_parent = current->real_parent;
1437                 p->parent_exec_id = current->parent_exec_id;
1438         } else {
1439                 p->real_parent = current;
1440                 p->parent_exec_id = current->self_exec_id;
1441         }
1442
1443         spin_lock(&current->sighand->siglock);
1444
1445         /*
1446          * Process group and session signals need to be delivered to just the
1447          * parent before the fork or both the parent and the child after the
1448          * fork. Restart if a signal comes in before we add the new process to
1449          * it's process group.
1450          * A fatal signal pending means that current will exit, so the new
1451          * thread can't slip out of an OOM kill (or normal SIGKILL).
1452         */
1453         recalc_sigpending();
1454         if (signal_pending(current)) {
1455                 spin_unlock(&current->sighand->siglock);
1456                 write_unlock_irq(&tasklist_lock);
1457                 retval = -ERESTARTNOINTR;
1458                 goto bad_fork_free_pid;
1459         }
1460
1461         if (clone_flags & CLONE_THREAD) {
1462                 current->signal->nr_threads++;
1463                 atomic_inc(&current->signal->live);
1464                 atomic_inc(&current->signal->sigcnt);
1465                 p->group_leader = current->group_leader;
1466                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1467         }
1468
1469         if (likely(p->pid)) {
1470                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1471
1472                 if (thread_group_leader(p)) {
1473                         if (is_child_reaper(pid))
1474                                 p->nsproxy->pid_ns->child_reaper = p;
1475
1476                         p->signal->leader_pid = pid;
1477                         p->signal->tty = tty_kref_get(current->signal->tty);
1478                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1479                         attach_pid(p, PIDTYPE_SID, task_session(current));
1480                         list_add_tail(&p->sibling, &p->real_parent->children);
1481                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1482                         __this_cpu_inc(process_counts);
1483                 }
1484                 attach_pid(p, PIDTYPE_PID, pid);
1485                 nr_threads++;
1486         }
1487
1488         total_forks++;
1489         spin_unlock(&current->sighand->siglock);
1490         write_unlock_irq(&tasklist_lock);
1491         proc_fork_connector(p);
1492         cgroup_post_fork(p);
1493         if (clone_flags & CLONE_THREAD)
1494                 threadgroup_change_end(current);
1495         perf_event_fork(p);
1496
1497         trace_task_newtask(p, clone_flags);
1498
1499         return p;
1500
1501 bad_fork_free_pid:
1502         if (pid != &init_struct_pid)
1503                 free_pid(pid);
1504 bad_fork_cleanup_io:
1505         if (p->io_context)
1506                 exit_io_context(p);
1507 bad_fork_cleanup_namespaces:
1508         if (unlikely(clone_flags & CLONE_NEWPID))
1509                 pid_ns_release_proc(p->nsproxy->pid_ns);
1510         exit_task_namespaces(p);
1511 bad_fork_cleanup_mm:
1512         if (p->mm)
1513                 mmput(p->mm);
1514 bad_fork_cleanup_signal:
1515         if (!(clone_flags & CLONE_THREAD))
1516                 free_signal_struct(p->signal);
1517 bad_fork_cleanup_sighand:
1518         __cleanup_sighand(p->sighand);
1519 bad_fork_cleanup_fs:
1520         exit_fs(p); /* blocking */
1521 bad_fork_cleanup_files:
1522         exit_files(p); /* blocking */
1523 bad_fork_cleanup_semundo:
1524         exit_sem(p);
1525 bad_fork_cleanup_audit:
1526         audit_free(p);
1527 bad_fork_cleanup_policy:
1528         perf_event_free_task(p);
1529 #ifdef CONFIG_NUMA
1530         mpol_put(p->mempolicy);
1531 bad_fork_cleanup_cgroup:
1532 #endif
1533         if (clone_flags & CLONE_THREAD)
1534                 threadgroup_change_end(current);
1535         cgroup_exit(p, cgroup_callbacks_done);
1536         delayacct_tsk_free(p);
1537         module_put(task_thread_info(p)->exec_domain->module);
1538 bad_fork_cleanup_count:
1539         atomic_dec(&p->cred->user->processes);
1540         exit_creds(p);
1541 bad_fork_free:
1542         free_task(p);
1543 fork_out:
1544         return ERR_PTR(retval);
1545 }
1546
1547 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1548 {
1549         memset(regs, 0, sizeof(struct pt_regs));
1550         return regs;
1551 }
1552
1553 static inline void init_idle_pids(struct pid_link *links)
1554 {
1555         enum pid_type type;
1556
1557         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1558                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1559                 links[type].pid = &init_struct_pid;
1560         }
1561 }
1562
1563 struct task_struct * __cpuinit fork_idle(int cpu)
1564 {
1565         struct task_struct *task;
1566         struct pt_regs regs;
1567
1568         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1569                             &init_struct_pid, 0);
1570         if (!IS_ERR(task)) {
1571                 init_idle_pids(task->pids);
1572                 init_idle(task, cpu);
1573         }
1574
1575         return task;
1576 }
1577
1578 /*
1579  *  Ok, this is the main fork-routine.
1580  *
1581  * It copies the process, and if successful kick-starts
1582  * it and waits for it to finish using the VM if required.
1583  */
1584 long do_fork(unsigned long clone_flags,
1585               unsigned long stack_start,
1586               struct pt_regs *regs,
1587               unsigned long stack_size,
1588               int __user *parent_tidptr,
1589               int __user *child_tidptr)
1590 {
1591         struct task_struct *p;
1592         int trace = 0;
1593         long nr;
1594
1595         /*
1596          * Do some preliminary argument and permissions checking before we
1597          * actually start allocating stuff
1598          */
1599         if (clone_flags & CLONE_NEWUSER) {
1600                 if (clone_flags & CLONE_THREAD)
1601                         return -EINVAL;
1602                 /* hopefully this check will go away when userns support is
1603                  * complete
1604                  */
1605                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1606                                 !capable(CAP_SETGID))
1607                         return -EPERM;
1608         }
1609
1610         /*
1611          * Determine whether and which event to report to ptracer.  When
1612          * called from kernel_thread or CLONE_UNTRACED is explicitly
1613          * requested, no event is reported; otherwise, report if the event
1614          * for the type of forking is enabled.
1615          */
1616         if (likely(user_mode(regs)) && !(clone_flags & CLONE_UNTRACED)) {
1617                 if (clone_flags & CLONE_VFORK)
1618                         trace = PTRACE_EVENT_VFORK;
1619                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1620                         trace = PTRACE_EVENT_CLONE;
1621                 else
1622                         trace = PTRACE_EVENT_FORK;
1623
1624                 if (likely(!ptrace_event_enabled(current, trace)))
1625                         trace = 0;
1626         }
1627
1628         p = copy_process(clone_flags, stack_start, regs, stack_size,
1629                          child_tidptr, NULL, trace);
1630         /*
1631          * Do this prior waking up the new thread - the thread pointer
1632          * might get invalid after that point, if the thread exits quickly.
1633          */
1634         if (!IS_ERR(p)) {
1635                 struct completion vfork;
1636
1637                 trace_sched_process_fork(current, p);
1638
1639                 nr = task_pid_vnr(p);
1640
1641                 if (clone_flags & CLONE_PARENT_SETTID)
1642                         put_user(nr, parent_tidptr);
1643
1644                 if (clone_flags & CLONE_VFORK) {
1645                         p->vfork_done = &vfork;
1646                         init_completion(&vfork);
1647                         get_task_struct(p);
1648                 }
1649
1650                 wake_up_new_task(p);
1651
1652                 /* forking complete and child started to run, tell ptracer */
1653                 if (unlikely(trace))
1654                         ptrace_event(trace, nr);
1655
1656                 if (clone_flags & CLONE_VFORK) {
1657                         if (!wait_for_vfork_done(p, &vfork))
1658                                 ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1659                 }
1660         } else {
1661                 nr = PTR_ERR(p);
1662         }
1663         return nr;
1664 }
1665
1666 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1667 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1668 #endif
1669
1670 static void sighand_ctor(void *data)
1671 {
1672         struct sighand_struct *sighand = data;
1673
1674         spin_lock_init(&sighand->siglock);
1675         init_waitqueue_head(&sighand->signalfd_wqh);
1676 }
1677
1678 void __init proc_caches_init(void)
1679 {
1680         sighand_cachep = kmem_cache_create("sighand_cache",
1681                         sizeof(struct sighand_struct), 0,
1682                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1683                         SLAB_NOTRACK, sighand_ctor);
1684         signal_cachep = kmem_cache_create("signal_cache",
1685                         sizeof(struct signal_struct), 0,
1686                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1687         files_cachep = kmem_cache_create("files_cache",
1688                         sizeof(struct files_struct), 0,
1689                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1690         fs_cachep = kmem_cache_create("fs_cache",
1691                         sizeof(struct fs_struct), 0,
1692                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1693         /*
1694          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1695          * whole struct cpumask for the OFFSTACK case. We could change
1696          * this to *only* allocate as much of it as required by the
1697          * maximum number of CPU's we can ever have.  The cpumask_allocation
1698          * is at the end of the structure, exactly for that reason.
1699          */
1700         mm_cachep = kmem_cache_create("mm_struct",
1701                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1702                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1703         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1704         mmap_init();
1705         nsproxy_cache_init();
1706 }
1707
1708 /*
1709  * Check constraints on flags passed to the unshare system call.
1710  */
1711 static int check_unshare_flags(unsigned long unshare_flags)
1712 {
1713         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1714                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1715                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1716                 return -EINVAL;
1717         /*
1718          * Not implemented, but pretend it works if there is nothing to
1719          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1720          * needs to unshare vm.
1721          */
1722         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1723                 /* FIXME: get_task_mm() increments ->mm_users */
1724                 if (atomic_read(&current->mm->mm_users) > 1)
1725                         return -EINVAL;
1726         }
1727
1728         return 0;
1729 }
1730
1731 /*
1732  * Unshare the filesystem structure if it is being shared
1733  */
1734 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1735 {
1736         struct fs_struct *fs = current->fs;
1737
1738         if (!(unshare_flags & CLONE_FS) || !fs)
1739                 return 0;
1740
1741         /* don't need lock here; in the worst case we'll do useless copy */
1742         if (fs->users == 1)
1743                 return 0;
1744
1745         *new_fsp = copy_fs_struct(fs);
1746         if (!*new_fsp)
1747                 return -ENOMEM;
1748
1749         return 0;
1750 }
1751
1752 /*
1753  * Unshare file descriptor table if it is being shared
1754  */
1755 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1756 {
1757         struct files_struct *fd = current->files;
1758         int error = 0;
1759
1760         if ((unshare_flags & CLONE_FILES) &&
1761             (fd && atomic_read(&fd->count) > 1)) {
1762                 *new_fdp = dup_fd(fd, &error);
1763                 if (!*new_fdp)
1764                         return error;
1765         }
1766
1767         return 0;
1768 }
1769
1770 /*
1771  * unshare allows a process to 'unshare' part of the process
1772  * context which was originally shared using clone.  copy_*
1773  * functions used by do_fork() cannot be used here directly
1774  * because they modify an inactive task_struct that is being
1775  * constructed. Here we are modifying the current, active,
1776  * task_struct.
1777  */
1778 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1779 {
1780         struct fs_struct *fs, *new_fs = NULL;
1781         struct files_struct *fd, *new_fd = NULL;
1782         struct nsproxy *new_nsproxy = NULL;
1783         int do_sysvsem = 0;
1784         int err;
1785
1786         err = check_unshare_flags(unshare_flags);
1787         if (err)
1788                 goto bad_unshare_out;
1789
1790         /*
1791          * If unsharing namespace, must also unshare filesystem information.
1792          */
1793         if (unshare_flags & CLONE_NEWNS)
1794                 unshare_flags |= CLONE_FS;
1795         /*
1796          * CLONE_NEWIPC must also detach from the undolist: after switching
1797          * to a new ipc namespace, the semaphore arrays from the old
1798          * namespace are unreachable.
1799          */
1800         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1801                 do_sysvsem = 1;
1802         err = unshare_fs(unshare_flags, &new_fs);
1803         if (err)
1804                 goto bad_unshare_out;
1805         err = unshare_fd(unshare_flags, &new_fd);
1806         if (err)
1807                 goto bad_unshare_cleanup_fs;
1808         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy, new_fs);
1809         if (err)
1810                 goto bad_unshare_cleanup_fd;
1811
1812         if (new_fs || new_fd || do_sysvsem || new_nsproxy) {
1813                 if (do_sysvsem) {
1814                         /*
1815                          * CLONE_SYSVSEM is equivalent to sys_exit().
1816                          */
1817                         exit_sem(current);
1818                 }
1819
1820                 if (new_nsproxy) {
1821                         switch_task_namespaces(current, new_nsproxy);
1822                         new_nsproxy = NULL;
1823                 }
1824
1825                 task_lock(current);
1826
1827                 if (new_fs) {
1828                         fs = current->fs;
1829                         spin_lock(&fs->lock);
1830                         current->fs = new_fs;
1831                         if (--fs->users)
1832                                 new_fs = NULL;
1833                         else
1834                                 new_fs = fs;
1835                         spin_unlock(&fs->lock);
1836                 }
1837
1838                 if (new_fd) {
1839                         fd = current->files;
1840                         current->files = new_fd;
1841                         new_fd = fd;
1842                 }
1843
1844                 task_unlock(current);
1845         }
1846
1847         if (new_nsproxy)
1848                 put_nsproxy(new_nsproxy);
1849
1850 bad_unshare_cleanup_fd:
1851         if (new_fd)
1852                 put_files_struct(new_fd);
1853
1854 bad_unshare_cleanup_fs:
1855         if (new_fs)
1856                 free_fs_struct(new_fs);
1857
1858 bad_unshare_out:
1859         return err;
1860 }
1861
1862 /*
1863  *      Helper to unshare the files of the current task.
1864  *      We don't want to expose copy_files internals to
1865  *      the exec layer of the kernel.
1866  */
1867
1868 int unshare_files(struct files_struct **displaced)
1869 {
1870         struct task_struct *task = current;
1871         struct files_struct *copy = NULL;
1872         int error;
1873
1874         error = unshare_fd(CLONE_FILES, &copy);
1875         if (error || !copy) {
1876                 *displaced = NULL;
1877                 return error;
1878         }
1879         *displaced = task->files;
1880         task_lock(task);
1881         task->files = copy;
1882         task_unlock(task);
1883         return 0;
1884 }