kernel/fork.c

   1 /*
   2  *  linux/kernel/fork.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  *  'fork.c' contains the help-routines for the 'fork' system call
   9  * (see also entry.S and others).
  10  * Fork is rather simple, once you get the hang of it, but the memory
  11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  12  */
  13
  14 #include <linux/slab.h>
  15 #include <linux/init.h>
  16 #include <linux/unistd.h>
  17 #include <linux/module.h>
  18 #include <linux/vmalloc.h>
  19 #include <linux/completion.h>
  20 #include <linux/personality.h>
  21 #include <linux/mempolicy.h>
  22 #include <linux/sem.h>
  23 #include <linux/file.h>
  24 #include <linux/fdtable.h>
  25 #include <linux/iocontext.h>
  26 #include <linux/key.h>
  27 #include <linux/binfmts.h>
  28 #include <linux/mman.h>
  29 #include <linux/mmu_notifier.h>
  30 #include <linux/fs.h>
  31 #include <linux/nsproxy.h>
  32 #include <linux/capability.h>
  33 #include <linux/cpu.h>
  34 #include <linux/cgroup.h>
  35 #include <linux/security.h>
  36 #include <linux/hugetlb.h>
  37 #include <linux/swap.h>
  38 #include <linux/syscalls.h>
  39 #include <linux/jiffies.h>
  40 #include <linux/futex.h>
  41 #include <linux/compat.h>
  42 #include <linux/kthread.h>
  43 #include <linux/task_io_accounting_ops.h>
  44 #include <linux/rcupdate.h>
  45 #include <linux/ptrace.h>
  46 #include <linux/mount.h>
  47 #include <linux/audit.h>
  48 #include <linux/memcontrol.h>
  49 #include <linux/ftrace.h>
  50 #include <linux/profile.h>
  51 #include <linux/rmap.h>
  52 #include <linux/ksm.h>
  53 #include <linux/acct.h>
  54 #include <linux/tsacct_kern.h>
  55 #include <linux/cn_proc.h>
  56 #include <linux/freezer.h>
  57 #include <linux/delayacct.h>
  58 #include <linux/taskstats_kern.h>
  59 #include <linux/random.h>
  60 #include <linux/tty.h>
  61 #include <linux/blkdev.h>
  62 #include <linux/fs_struct.h>
  63 #include <linux/magic.h>
  64 #include <linux/perf_event.h>
  65 #include <linux/posix-timers.h>
  66 #include <linux/user-return-notifier.h>
  67 #include <linux/oom.h>
  68 #include <linux/khugepaged.h>
  69 #include <linux/signalfd.h>
  70
  71 #include <asm/pgtable.h>
  72 #include <asm/pgalloc.h>
  73 #include <asm/uaccess.h>
  74 #include <asm/mmu_context.h>
  75 #include <asm/cacheflush.h>
  76 #include <asm/tlbflush.h>
  77
  78 #include <trace/events/sched.h>
  79
  80 #define CREATE_TRACE_POINTS
  81 #include <trace/events/task.h>
  82
  83 /*
  84  * Protected counters by write_lock_irq(&tasklist_lock)
  85  */
  86 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  87 int nr_threads;                 /* The idle threads do not count.. */
  88
  89 int max_threads;                /* tunable limit on nr_threads */
  90
  91 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  92
  93 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  94
  95 #ifdef CONFIG_PROVE_RCU
  96 int lockdep_tasklist_lock_is_held(void)
  97 {
  98         return lockdep_is_held(&tasklist_lock);
  99 }
 100 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 101 #endif /* #ifdef CONFIG_PROVE_RCU */
 102
 103 int nr_processes(void)
 104 {
 105         int cpu;
 106         int total = 0;
 107
 108         for_each_possible_cpu(cpu)
 109                 total += per_cpu(process_counts, cpu);
 110
 111         return total;
 112 }
 113
 114 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 115 # define alloc_task_struct_node(node)           \
 116                 kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node)
 117 # define free_task_struct(tsk)                  \
 118                 kmem_cache_free(task_struct_cachep, (tsk))
 119 static struct kmem_cache *task_struct_cachep;
 120 #endif
 121
 122 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
 123 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 124                                                   int node)
 125 {
 126 #ifdef CONFIG_DEBUG_STACK_USAGE
 127         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
 128 #else
 129         gfp_t mask = GFP_KERNEL;
 130 #endif
 131         struct page *page = alloc_pages_node(node, mask, THREAD_SIZE_ORDER);
 132
 133         return page ? page_address(page) : NULL;
 134 }
 135
 136 static inline void free_thread_info(struct thread_info *ti)
 137 {
 138         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 139 }
 140 #endif
 141
 142 /* SLAB cache for signal_struct structures (tsk->signal) */
 143 static struct kmem_cache *signal_cachep;
 144
 145 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 146 struct kmem_cache *sighand_cachep;
 147
 148 /* SLAB cache for files_struct structures (tsk->files) */
 149 struct kmem_cache *files_cachep;
 150
 151 /* SLAB cache for fs_struct structures (tsk->fs) */
 152 struct kmem_cache *fs_cachep;
 153
 154 /* SLAB cache for vm_area_struct structures */
 155 struct kmem_cache *vm_area_cachep;
 156
 157 /* SLAB cache for mm_struct structures (tsk->mm) */
 158 static struct kmem_cache *mm_cachep;
 159
 160 static void account_kernel_stack(struct thread_info *ti, int account)
 161 {
 162         struct zone *zone = page_zone(virt_to_page(ti));
 163
 164         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
 165 }
 166
 167 void free_task(struct task_struct *tsk)
 168 {
 169         account_kernel_stack(tsk->stack, -1);
 170         free_thread_info(tsk->stack);
 171         rt_mutex_debug_task_free(tsk);
 172         ftrace_graph_exit_task(tsk);
 173         free_task_struct(tsk);
 174 }
 175 EXPORT_SYMBOL(free_task);
 176
 177 static inline void free_signal_struct(struct signal_struct *sig)
 178 {
 179         taskstats_tgid_free(sig);
 180         sched_autogroup_exit(sig);
 181         kmem_cache_free(signal_cachep, sig);
 182 }
 183
 184 static inline void put_signal_struct(struct signal_struct *sig)
 185 {
 186         if (atomic_dec_and_test(&sig->sigcnt))
 187                 free_signal_struct(sig);
 188 }
 189
 190 void __put_task_struct(struct task_struct *tsk)
 191 {
 192         WARN_ON(!tsk->exit_state);
 193         WARN_ON(atomic_read(&tsk->usage));
 194         WARN_ON(tsk == current);
 195
 196         exit_creds(tsk);
 197         delayacct_tsk_free(tsk);
 198         put_signal_struct(tsk->signal);
 199
 200         if (!profile_handoff_task(tsk))
 201                 free_task(tsk);
 202 }
 203 EXPORT_SYMBOL_GPL(__put_task_struct);
 204
 205 /*
 206  * macro override instead of weak attribute alias, to workaround
 207  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
 208  */
 209 #ifndef arch_task_cache_init
 210 #define arch_task_cache_init()
 211 #endif
 212
 213 void __init fork_init(unsigned long mempages)
 214 {
 215 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 216 #ifndef ARCH_MIN_TASKALIGN
 217 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 218 #endif
 219         /* create a slab on which task_structs can be allocated */
 220         task_struct_cachep =
 221                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 222                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
 223 #endif
 224
 225         /* do the arch specific task caches init */
 226         arch_task_cache_init();
 227
 228         /*
 229          * The default maximum number of threads is set to a safe
 230          * value: the thread structures can take up at most half
 231          * of memory.
 232          */
 233         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 234
 235         /*
 236          * we need to allow at least 20 threads to boot a system
 237          */
 238         if (max_threads < 20)
 239                 max_threads = 20;
 240
 241         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 242         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 243         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 244                 init_task.signal->rlim[RLIMIT_NPROC];
 245 }
 246
 247 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
 248                                                struct task_struct *src)
 249 {
 250         *dst = *src;
 251         return 0;
 252 }
 253
 254 static struct task_struct *dup_task_struct(struct task_struct *orig)
 255 {
 256         struct task_struct *tsk;
 257         struct thread_info *ti;
 258         unsigned long *stackend;
 259         int node = tsk_fork_get_node(orig);
 260         int err;
 261
 262         prepare_to_copy(orig);
 263
 264         tsk = alloc_task_struct_node(node);
 265         if (!tsk)
 266                 return NULL;
 267
 268         ti = alloc_thread_info_node(tsk, node);
 269         if (!ti) {
 270                 free_task_struct(tsk);
 271                 return NULL;
 272         }
 273
 274         err = arch_dup_task_struct(tsk, orig);
 275         if (err)
 276                 goto out;
 277
 278         tsk->stack = ti;
 279
 280         setup_thread_stack(tsk, orig);
 281         clear_user_return_notifier(tsk);
 282         clear_tsk_need_resched(tsk);
 283         stackend = end_of_stack(tsk);
 284         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 285
 286 #ifdef CONFIG_CC_STACKPROTECTOR
 287         tsk->stack_canary = get_random_int();
 288 #endif
 289
 290         /*
 291          * One for us, one for whoever does the "release_task()" (usually
 292          * parent)
 293          */
 294         atomic_set(&tsk->usage, 2);
 295 #ifdef CONFIG_BLK_DEV_IO_TRACE
 296         tsk->btrace_seq = 0;
 297 #endif
 298         tsk->splice_pipe = NULL;
 299
 300         account_kernel_stack(ti, 1);
 301
 302         return tsk;
 303
 304 out:
 305         free_thread_info(ti);
 306         free_task_struct(tsk);
 307         return NULL;
 308 }
 309
 310 #ifdef CONFIG_MMU
 311 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 312 {
 313         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 314         struct rb_node **rb_link, *rb_parent;
 315         int retval;
 316         unsigned long charge;
 317         struct mempolicy *pol;
 318
 319         down_write(&oldmm->mmap_sem);
 320         flush_cache_dup_mm(oldmm);
 321         /*
 322          * Not linked in yet - no deadlock potential:
 323          */
 324         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 325
 326         mm->locked_vm = 0;
 327         mm->mmap = NULL;
 328         mm->mmap_cache = NULL;
 329         mm->free_area_cache = oldmm->mmap_base;
 330         mm->cached_hole_size = ~0UL;
 331         mm->map_count = 0;
 332         cpumask_clear(mm_cpumask(mm));
 333         mm->mm_rb = RB_ROOT;
 334         rb_link = &mm->mm_rb.rb_node;
 335         rb_parent = NULL;
 336         pprev = &mm->mmap;
 337         retval = ksm_fork(mm, oldmm);
 338         if (retval)
 339                 goto out;
 340         retval = khugepaged_fork(mm, oldmm);
 341         if (retval)
 342                 goto out;
 343
 344         prev = NULL;
 345         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 346                 struct file *file;
 347
 348                 if (mpnt->vm_flags & VM_DONTCOPY) {
 349                         long pages = vma_pages(mpnt);
 350                         mm->total_vm -= pages;
 351                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 352                                                                 -pages);
 353                         continue;
 354                 }
 355                 charge = 0;
 356                 if (mpnt->vm_flags & VM_ACCOUNT) {
 357                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 358                         if (security_vm_enough_memory(len))
 359                                 goto fail_nomem;
 360                         charge = len;
 361                 }
 362                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 363                 if (!tmp)
 364                         goto fail_nomem;
 365                 *tmp = *mpnt;
 366                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 367                 pol = mpol_dup(vma_policy(mpnt));
 368                 retval = PTR_ERR(pol);
 369                 if (IS_ERR(pol))
 370                         goto fail_nomem_policy;
 371                 vma_set_policy(tmp, pol);
 372                 tmp->vm_mm = mm;
 373                 if (anon_vma_fork(tmp, mpnt))
 374                         goto fail_nomem_anon_vma_fork;
 375                 tmp->vm_flags &= ~VM_LOCKED;
 376                 tmp->vm_next = tmp->vm_prev = NULL;
 377                 file = tmp->vm_file;
 378                 if (file) {
 379                         struct inode *inode = file->f_path.dentry->d_inode;
 380                         struct address_space *mapping = file->f_mapping;
 381
 382                         get_file(file);
 383                         if (tmp->vm_flags & VM_DENYWRITE)
 384                                 atomic_dec(&inode->i_writecount);
 385                         mutex_lock(&mapping->i_mmap_mutex);
 386                         if (tmp->vm_flags & VM_SHARED)
 387                                 mapping->i_mmap_writable++;
 388                         flush_dcache_mmap_lock(mapping);
 389                         /* insert tmp into the share list, just after mpnt */
 390                         vma_prio_tree_add(tmp, mpnt);
 391                         flush_dcache_mmap_unlock(mapping);
 392                         mutex_unlock(&mapping->i_mmap_mutex);
 393                 }
 394
 395                 /*
 396                  * Clear hugetlb-related page reserves for children. This only
 397                  * affects MAP_PRIVATE mappings. Faults generated by the child
 398                  * are not guaranteed to succeed, even if read-only
 399                  */
 400                 if (is_vm_hugetlb_page(tmp))
 401                         reset_vma_resv_huge_pages(tmp);
 402
 403                 /*
 404                  * Link in the new vma and copy the page table entries.
 405                  */
 406                 *pprev = tmp;
 407                 pprev = &tmp->vm_next;
 408                 tmp->vm_prev = prev;
 409                 prev = tmp;
 410
 411                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 412                 rb_link = &tmp->vm_rb.rb_right;
 413                 rb_parent = &tmp->vm_rb;
 414
 415                 mm->map_count++;
 416                 retval = copy_page_range(mm, oldmm, mpnt);
 417
 418                 if (tmp->vm_ops && tmp->vm_ops->open)
 419                         tmp->vm_ops->open(tmp);
 420
 421                 if (retval)
 422                         goto out;
 423         }
 424         /* a new mm has just been created */
 425         arch_dup_mmap(oldmm, mm);
 426         retval = 0;
 427 out:
 428         up_write(&mm->mmap_sem);
 429         flush_tlb_mm(oldmm);
 430         up_write(&oldmm->mmap_sem);
 431         return retval;
 432 fail_nomem_anon_vma_fork:
 433         mpol_put(pol);
 434 fail_nomem_policy:
 435         kmem_cache_free(vm_area_cachep, tmp);
 436 fail_nomem:
 437         retval = -ENOMEM;
 438         vm_unacct_memory(charge);
 439         goto out;
 440 }
 441
 442 static inline int mm_alloc_pgd(struct mm_struct *mm)
 443 {
 444         mm->pgd = pgd_alloc(mm);
 445         if (unlikely(!mm->pgd))
 446                 return -ENOMEM;
 447         return 0;
 448 }
 449
 450 static inline void mm_free_pgd(struct mm_struct *mm)
 451 {
 452         pgd_free(mm, mm->pgd);
 453 }
 454 #else
 455 #define dup_mmap(mm, oldmm)     (0)
 456 #define mm_alloc_pgd(mm)        (0)
 457 #define mm_free_pgd(mm)
 458 #endif /* CONFIG_MMU */
 459
 460 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 461
 462 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 463 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 464
 465 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 466
 467 static int __init coredump_filter_setup(char *s)
 468 {
 469         default_dump_filter =
 470                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 471                 MMF_DUMP_FILTER_MASK;
 472         return 1;
 473 }
 474
 475 __setup("coredump_filter=", coredump_filter_setup);
 476
 477 #include <linux/init_task.h>
 478
 479 static void mm_init_aio(struct mm_struct *mm)
 480 {
 481 #ifdef CONFIG_AIO
 482         spin_lock_init(&mm->ioctx_lock);
 483         INIT_HLIST_HEAD(&mm->ioctx_list);
 484 #endif
 485 }
 486
 487 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
 488 {
 489         atomic_set(&mm->mm_users, 1);
 490         atomic_set(&mm->mm_count, 1);
 491         init_rwsem(&mm->mmap_sem);
 492         INIT_LIST_HEAD(&mm->mmlist);
 493         mm->flags = (current->mm) ?
 494                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
 495         mm->core_state = NULL;
 496         mm->nr_ptes = 0;
 497         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 498         spin_lock_init(&mm->page_table_lock);
 499         mm->free_area_cache = TASK_UNMAPPED_BASE;
 500         mm->cached_hole_size = ~0UL;
 501         mm_init_aio(mm);
 502         mm_init_owner(mm, p);
 503
 504         if (likely(!mm_alloc_pgd(mm))) {
 505                 mm->def_flags = 0;
 506                 mmu_notifier_mm_init(mm);
 507                 return mm;
 508         }
 509
 510         free_mm(mm);
 511         return NULL;
 512 }
 513
 514 static void check_mm(struct mm_struct *mm)
 515 {
 516         int i;
 517
 518         for (i = 0; i < NR_MM_COUNTERS; i++) {
 519                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 520
 521                 if (unlikely(x))
 522                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 523                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 524         }
 525
 526 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 527         VM_BUG_ON(mm->pmd_huge_pte);
 528 #endif
 529 }
 530
 531 /*
 532  * Allocate and initialize an mm_struct.
 533  */
 534 struct mm_struct *mm_alloc(void)
 535 {
 536         struct mm_struct *mm;
 537
 538         mm = allocate_mm();
 539         if (!mm)
 540                 return NULL;
 541
 542         memset(mm, 0, sizeof(*mm));
 543         mm_init_cpumask(mm);
 544         return mm_init(mm, current);
 545 }
 546
 547 /*
 548  * Called when the last reference to the mm
 549  * is dropped: either by a lazy thread or by
 550  * mmput. Free the page directory and the mm.
 551  */
 552 void __mmdrop(struct mm_struct *mm)
 553 {
 554         BUG_ON(mm == &init_mm);
 555         mm_free_pgd(mm);
 556         destroy_context(mm);
 557         mmu_notifier_mm_destroy(mm);
 558         check_mm(mm);
 559         free_mm(mm);
 560 }
 561 EXPORT_SYMBOL_GPL(__mmdrop);
 562
 563 /*
 564  * Decrement the use count and release all resources for an mm.
 565  */
 566 void mmput(struct mm_struct *mm)
 567 {
 568         might_sleep();
 569
 570         if (atomic_dec_and_test(&mm->mm_users)) {
 571                 exit_aio(mm);
 572                 ksm_exit(mm);
 573                 khugepaged_exit(mm); /* must run before exit_mmap */
 574                 exit_mmap(mm);
 575                 set_mm_exe_file(mm, NULL);
 576                 if (!list_empty(&mm->mmlist)) {
 577                         spin_lock(&mmlist_lock);
 578                         list_del(&mm->mmlist);
 579                         spin_unlock(&mmlist_lock);
 580                 }
 581                 put_swap_token(mm);
 582                 if (mm->binfmt)
 583                         module_put(mm->binfmt->module);
 584                 mmdrop(mm);
 585         }
 586 }
 587 EXPORT_SYMBOL_GPL(mmput);
 588
 589 /*
 590  * We added or removed a vma mapping the executable. The vmas are only mapped
 591  * during exec and are not mapped with the mmap system call.
 592  * Callers must hold down_write() on the mm's mmap_sem for these
 593  */
 594 void added_exe_file_vma(struct mm_struct *mm)
 595 {
 596         mm->num_exe_file_vmas++;
 597 }
 598
 599 void removed_exe_file_vma(struct mm_struct *mm)
 600 {
 601         mm->num_exe_file_vmas--;
 602         if ((mm->num_exe_file_vmas == 0) && mm->exe_file) {
 603                 fput(mm->exe_file);
 604                 mm->exe_file = NULL;
 605         }
 606
 607 }
 608
 609 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 610 {
 611         if (new_exe_file)
 612                 get_file(new_exe_file);
 613         if (mm->exe_file)
 614                 fput(mm->exe_file);
 615         mm->exe_file = new_exe_file;
 616         mm->num_exe_file_vmas = 0;
 617 }
 618
 619 struct file *get_mm_exe_file(struct mm_struct *mm)
 620 {
 621         struct file *exe_file;
 622
 623         /* We need mmap_sem to protect against races with removal of
 624          * VM_EXECUTABLE vmas */
 625         down_read(&mm->mmap_sem);
 626         exe_file = mm->exe_file;
 627         if (exe_file)
 628                 get_file(exe_file);
 629         up_read(&mm->mmap_sem);
 630         return exe_file;
 631 }
 632
 633 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
 634 {
 635         /* It's safe to write the exe_file pointer without exe_file_lock because
 636          * this is called during fork when the task is not yet in /proc */
 637         newmm->exe_file = get_mm_exe_file(oldmm);
 638 }
 639
 640 /**
 641  * get_task_mm - acquire a reference to the task's mm
 642  *
 643  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 644  * this kernel workthread has transiently adopted a user mm with use_mm,
 645  * to do its AIO) is not set and if so returns a reference to it, after
 646  * bumping up the use count.  User must release the mm via mmput()
 647  * after use.  Typically used by /proc and ptrace.
 648  */
 649 struct mm_struct *get_task_mm(struct task_struct *task)
 650 {
 651         struct mm_struct *mm;
 652
 653         task_lock(task);
 654         mm = task->mm;
 655         if (mm) {
 656                 if (task->flags & PF_KTHREAD)
 657                         mm = NULL;
 658                 else
 659                         atomic_inc(&mm->mm_users);
 660         }
 661         task_unlock(task);
 662         return mm;
 663 }
 664 EXPORT_SYMBOL_GPL(get_task_mm);
 665
 666 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 667 {
 668         struct mm_struct *mm;
 669         int err;
 670
 671         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 672         if (err)
 673                 return ERR_PTR(err);
 674
 675         mm = get_task_mm(task);
 676         if (mm && mm != current->mm &&
 677                         !ptrace_may_access(task, mode)) {
 678                 mmput(mm);
 679                 mm = ERR_PTR(-EACCES);
 680         }
 681         mutex_unlock(&task->signal->cred_guard_mutex);
 682
 683         return mm;
 684 }
 685
 686 static void complete_vfork_done(struct task_struct *tsk)
 687 {
 688         struct completion *vfork;
 689
 690         task_lock(tsk);
 691         vfork = tsk->vfork_done;
 692         if (likely(vfork)) {
 693                 tsk->vfork_done = NULL;
 694                 complete(vfork);
 695         }
 696         task_unlock(tsk);
 697 }
 698
 699 static int wait_for_vfork_done(struct task_struct *child,
 700                                 struct completion *vfork)
 701 {
 702         int killed;
 703
 704         freezer_do_not_count();
 705         killed = wait_for_completion_killable(vfork);
 706         freezer_count();
 707
 708         if (killed) {
 709                 task_lock(child);
 710                 child->vfork_done = NULL;
 711                 task_unlock(child);
 712         }
 713
 714         put_task_struct(child);
 715         return killed;
 716 }
 717
 718 /* Please note the differences between mmput and mm_release.
 719  * mmput is called whenever we stop holding onto a mm_struct,
 720  * error success whatever.
 721  *
 722  * mm_release is called after a mm_struct has been removed
 723  * from the current process.
 724  *
 725  * This difference is important for error handling, when we
 726  * only half set up a mm_struct for a new process and need to restore
 727  * the old one.  Because we mmput the new mm_struct before
 728  * restoring the old one. . .
 729  * Eric Biederman 10 January 1998
 730  */
 731 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 732 {
 733         /* Get rid of any futexes when releasing the mm */
 734 #ifdef CONFIG_FUTEX
 735         if (unlikely(tsk->robust_list)) {
 736                 exit_robust_list(tsk);
 737                 tsk->robust_list = NULL;
 738         }
 739 #ifdef CONFIG_COMPAT
 740         if (unlikely(tsk->compat_robust_list)) {
 741                 compat_exit_robust_list(tsk);
 742                 tsk->compat_robust_list = NULL;
 743         }
 744 #endif
 745         if (unlikely(!list_empty(&tsk->pi_state_list)))
 746                 exit_pi_state_list(tsk);
 747 #endif
 748
 749         /* Get rid of any cached register state */
 750         deactivate_mm(tsk, mm);
 751
 752         if (tsk->vfork_done)
 753                 complete_vfork_done(tsk);
 754
 755         /*
 756          * If we're exiting normally, clear a user-space tid field if
 757          * requested.  We leave this alone when dying by signal, to leave
 758          * the value intact in a core dump, and to save the unnecessary
 759          * trouble, say, a killed vfork parent shouldn't touch this mm.
 760          * Userland only wants this done for a sys_exit.
 761          */
 762         if (tsk->clear_child_tid) {
 763                 if (!(tsk->flags & PF_SIGNALED) &&
 764                     atomic_read(&mm->mm_users) > 1) {
 765                         /*
 766                          * We don't check the error code - if userspace has
 767                          * not set up a proper pointer then tough luck.
 768                          */
 769                         put_user(0, tsk->clear_child_tid);
 770                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 771                                         1, NULL, NULL, 0);
 772                 }
 773                 tsk->clear_child_tid = NULL;
 774         }
 775 }
 776
 777 /*
 778  * Allocate a new mm structure and copy contents from the
 779  * mm structure of the passed in task structure.
 780  */
 781 struct mm_struct *dup_mm(struct task_struct *tsk)
 782 {
 783         struct mm_struct *mm, *oldmm = current->mm;
 784         int err;
 785
 786         if (!oldmm)
 787                 return NULL;
 788
 789         mm = allocate_mm();
 790         if (!mm)
 791                 goto fail_nomem;
 792
 793         memcpy(mm, oldmm, sizeof(*mm));
 794         mm_init_cpumask(mm);
 795
 796         /* Initializing for Swap token stuff */
 797         mm->token_priority = 0;
 798         mm->last_interval = 0;
 799
 800 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 801         mm->pmd_huge_pte = NULL;
 802 #endif
 803
 804         if (!mm_init(mm, tsk))
 805                 goto fail_nomem;
 806
 807         if (init_new_context(tsk, mm))
 808                 goto fail_nocontext;
 809
 810         dup_mm_exe_file(oldmm, mm);
 811
 812         err = dup_mmap(mm, oldmm);
 813         if (err)
 814                 goto free_pt;
 815
 816         mm->hiwater_rss = get_mm_rss(mm);
 817         mm->hiwater_vm = mm->total_vm;
 818
 819         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 820                 goto free_pt;
 821
 822         return mm;
 823
 824 free_pt:
 825         /* don't put binfmt in mmput, we haven't got module yet */
 826         mm->binfmt = NULL;
 827         mmput(mm);
 828
 829 fail_nomem:
 830         return NULL;
 831
 832 fail_nocontext:
 833         /*
 834          * If init_new_context() failed, we cannot use mmput() to free the mm
 835          * because it calls destroy_context()
 836          */
 837         mm_free_pgd(mm);
 838         free_mm(mm);
 839         return NULL;
 840 }
 841
 842 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 843 {
 844         struct mm_struct *mm, *oldmm;
 845         int retval;
 846
 847         tsk->min_flt = tsk->maj_flt = 0;
 848         tsk->nvcsw = tsk->nivcsw = 0;
 849 #ifdef CONFIG_DETECT_HUNG_TASK
 850         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 851 #endif
 852
 853         tsk->mm = NULL;
 854         tsk->active_mm = NULL;
 855
 856         /*
 857          * Are we cloning a kernel thread?
 858          *
 859          * We need to steal a active VM for that..
 860          */
 861         oldmm = current->mm;
 862         if (!oldmm)
 863                 return 0;
 864
 865         if (clone_flags & CLONE_VM) {
 866                 atomic_inc(&oldmm->mm_users);
 867                 mm = oldmm;
 868                 goto good_mm;
 869         }
 870
 871         retval = -ENOMEM;
 872         mm = dup_mm(tsk);
 873         if (!mm)
 874                 goto fail_nomem;
 875
 876 good_mm:
 877         /* Initializing for Swap token stuff */
 878         mm->token_priority = 0;
 879         mm->last_interval = 0;
 880
 881         tsk->mm = mm;
 882         tsk->active_mm = mm;
 883         return 0;
 884
 885 fail_nomem:
 886         return retval;
 887 }
 888
 889 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 890 {
 891         struct fs_struct *fs = current->fs;
 892         if (clone_flags & CLONE_FS) {
 893                 /* tsk->fs is already what we want */
 894                 spin_lock(&fs->lock);
 895                 if (fs->in_exec) {
 896                         spin_unlock(&fs->lock);
 897                         return -EAGAIN;
 898                 }
 899                 fs->users++;
 900                 spin_unlock(&fs->lock);
 901                 return 0;
 902         }
 903         tsk->fs = copy_fs_struct(fs);
 904         if (!tsk->fs)
 905                 return -ENOMEM;
 906         return 0;
 907 }
 908
 909 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
 910 {
 911         struct files_struct *oldf, *newf;
 912         int error = 0;
 913
 914         /*
 915          * A background process may not have any files ...
 916          */
 917         oldf = current->files;
 918         if (!oldf)
 919                 goto out;
 920
 921         if (clone_flags & CLONE_FILES) {
 922                 atomic_inc(&oldf->count);
 923                 goto out;
 924         }
 925
 926         newf = dup_fd(oldf, &error);
 927         if (!newf)
 928                 goto out;
 929
 930         tsk->files = newf;
 931         error = 0;
 932 out:
 933         return error;
 934 }
 935
 936 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 937 {
 938 #ifdef CONFIG_BLOCK
 939         struct io_context *ioc = current->io_context;
 940         struct io_context *new_ioc;
 941
 942         if (!ioc)
 943                 return 0;
 944         /*
 945          * Share io context with parent, if CLONE_IO is set
 946          */
 947         if (clone_flags & CLONE_IO) {
 948                 tsk->io_context = ioc_task_link(ioc);
 949                 if (unlikely(!tsk->io_context))
 950                         return -ENOMEM;
 951         } else if (ioprio_valid(ioc->ioprio)) {
 952                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
 953                 if (unlikely(!new_ioc))
 954                         return -ENOMEM;
 955
 956                 new_ioc->ioprio = ioc->ioprio;
 957                 put_io_context(new_ioc);
 958         }
 959 #endif
 960         return 0;
 961 }
 962
 963 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 964 {
 965         struct sighand_struct *sig;
 966
 967         if (clone_flags & CLONE_SIGHAND) {
 968                 atomic_inc(&current->sighand->count);
 969                 return 0;
 970         }
 971         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 972         rcu_assign_pointer(tsk->sighand, sig);
 973         if (!sig)
 974                 return -ENOMEM;
 975         atomic_set(&sig->count, 1);
 976         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 977         return 0;
 978 }
 979
 980 void __cleanup_sighand(struct sighand_struct *sighand)
 981 {
 982         if (atomic_dec_and_test(&sighand->count)) {
 983                 signalfd_cleanup(sighand);
 984                 kmem_cache_free(sighand_cachep, sighand);
 985         }
 986 }
 987
 988
 989 /*
 990  * Initialize POSIX timer handling for a thread group.
 991  */
 992 static void posix_cpu_timers_init_group(struct signal_struct *sig)
 993 {
 994         unsigned long cpu_limit;
 995
 996         /* Thread group counters. */
 997         thread_group_cputime_init(sig);
 998
 999         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1000         if (cpu_limit != RLIM_INFINITY) {
1001                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1002                 sig->cputimer.running = 1;
1003         }
1004
1005         /* The timer lists. */
1006         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1007         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1008         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1009 }
1010
1011 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1012 {
1013         struct signal_struct *sig;
1014
1015         if (clone_flags & CLONE_THREAD)
1016                 return 0;
1017
1018         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1019         tsk->signal = sig;
1020         if (!sig)
1021                 return -ENOMEM;
1022
1023         sig->nr_threads = 1;
1024         atomic_set(&sig->live, 1);
1025         atomic_set(&sig->sigcnt, 1);
1026         init_waitqueue_head(&sig->wait_chldexit);
1027         if (clone_flags & CLONE_NEWPID)
1028                 sig->flags |= SIGNAL_UNKILLABLE;
1029         sig->curr_target = tsk;
1030         init_sigpending(&sig->shared_pending);
1031         INIT_LIST_HEAD(&sig->posix_timers);
1032
1033         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1034         sig->real_timer.function = it_real_fn;
1035
1036         task_lock(current->group_leader);
1037         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1038         task_unlock(current->group_leader);
1039
1040         posix_cpu_timers_init_group(sig);
1041
1042         tty_audit_fork(sig);
1043         sched_autogroup_fork(sig);
1044
1045 #ifdef CONFIG_CGROUPS
1046         init_rwsem(&sig->group_rwsem);
1047 #endif
1048
1049         sig->oom_adj = current->signal->oom_adj;
1050         sig->oom_score_adj = current->signal->oom_score_adj;
1051         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1052
1053         mutex_init(&sig->cred_guard_mutex);
1054
1055         return 0;
1056 }
1057
1058 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
1059 {
1060         unsigned long new_flags = p->flags;
1061
1062         new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1063         new_flags |= PF_FORKNOEXEC;
1064         p->flags = new_flags;
1065 }
1066
1067 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1068 {
1069         current->clear_child_tid = tidptr;
1070
1071         return task_pid_vnr(current);
1072 }
1073
1074 static void rt_mutex_init_task(struct task_struct *p)
1075 {
1076         raw_spin_lock_init(&p->pi_lock);
1077 #ifdef CONFIG_RT_MUTEXES
1078         plist_head_init(&p->pi_waiters);
1079         p->pi_blocked_on = NULL;
1080 #endif
1081 }
1082
1083 #ifdef CONFIG_MM_OWNER
1084 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1085 {
1086         mm->owner = p;
1087 }
1088 #endif /* CONFIG_MM_OWNER */
1089
1090 /*
1091  * Initialize POSIX timer handling for a single task.
1092  */
1093 static void posix_cpu_timers_init(struct task_struct *tsk)
1094 {
1095         tsk->cputime_expires.prof_exp = 0;
1096         tsk->cputime_expires.virt_exp = 0;
1097         tsk->cputime_expires.sched_exp = 0;
1098         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1099         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1100         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1101 }
1102
1103 /*
1104  * This creates a new process as a copy of the old one,
1105  * but does not actually start it yet.
1106  *
1107  * It copies the registers, and all the appropriate
1108  * parts of the process environment (as per the clone
1109  * flags). The actual kick-off is left to the caller.
1110  */
1111 static struct task_struct *copy_process(unsigned long clone_flags,
1112                                         unsigned long stack_start,
1113                                         struct pt_regs *regs,
1114                                         unsigned long stack_size,
1115                                         int __user *child_tidptr,
1116                                         struct pid *pid,
1117                                         int trace)
1118 {
1119         int retval;
1120         struct task_struct *p;
1121         int cgroup_callbacks_done = 0;
1122
1123         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1124                 return ERR_PTR(-EINVAL);
1125
1126         /*
1127          * Thread groups must share signals as well, and detached threads
1128          * can only be started up within the thread group.
1129          */
1130         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1131                 return ERR_PTR(-EINVAL);
1132
1133         /*
1134          * Shared signal handlers imply shared VM. By way of the above,
1135          * thread groups also imply shared VM. Blocking this case allows
1136          * for various simplifications in other code.
1137          */
1138         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1139                 return ERR_PTR(-EINVAL);
1140
1141         /*
1142          * Siblings of global init remain as zombies on exit since they are
1143          * not reaped by their parent (swapper). To solve this and to avoid
1144          * multi-rooted process trees, prevent global and container-inits
1145          * from creating siblings.
1146          */
1147         if ((clone_flags & CLONE_PARENT) &&
1148                                 current->signal->flags & SIGNAL_UNKILLABLE)
1149                 return ERR_PTR(-EINVAL);
1150
1151         retval = security_task_create(clone_flags);
1152         if (retval)
1153                 goto fork_out;
1154
1155         retval = -ENOMEM;
1156         p = dup_task_struct(current);
1157         if (!p)
1158                 goto fork_out;
1159
1160         ftrace_graph_init_task(p);
1161
1162         rt_mutex_init_task(p);
1163
1164 #ifdef CONFIG_PROVE_LOCKING
1165         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1166         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1167 #endif
1168         retval = -EAGAIN;
1169         if (atomic_read(&p->real_cred->user->processes) >=
1170                         task_rlimit(p, RLIMIT_NPROC)) {
1171                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1172                     p->real_cred->user != INIT_USER)
1173                         goto bad_fork_free;
1174         }
1175         current->flags &= ~PF_NPROC_EXCEEDED;
1176
1177         retval = copy_creds(p, clone_flags);
1178         if (retval < 0)
1179                 goto bad_fork_free;
1180
1181         /*
1182          * If multiple threads are within copy_process(), then this check
1183          * triggers too late. This doesn't hurt, the check is only there
1184          * to stop root fork bombs.
1185          */
1186         retval = -EAGAIN;
1187         if (nr_threads >= max_threads)
1188                 goto bad_fork_cleanup_count;
1189
1190         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1191                 goto bad_fork_cleanup_count;
1192
1193         p->did_exec = 0;
1194         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1195         copy_flags(clone_flags, p);
1196         INIT_LIST_HEAD(&p->children);
1197         INIT_LIST_HEAD(&p->sibling);
1198         rcu_copy_process(p);
1199         p->vfork_done = NULL;
1200         spin_lock_init(&p->alloc_lock);
1201
1202         init_sigpending(&p->pending);
1203
1204         p->utime = p->stime = p->gtime = 0;
1205         p->utimescaled = p->stimescaled = 0;
1206 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1207         p->prev_utime = p->prev_stime = 0;
1208 #endif
1209 #if defined(SPLIT_RSS_COUNTING)
1210         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1211 #endif
1212
1213         p->default_timer_slack_ns = current->timer_slack_ns;
1214
1215         task_io_accounting_init(&p->ioac);
1216         acct_clear_integrals(p);
1217
1218         posix_cpu_timers_init(p);
1219
1220         do_posix_clock_monotonic_gettime(&p->start_time);
1221         p->real_start_time = p->start_time;
1222         monotonic_to_bootbased(&p->real_start_time);
1223         p->io_context = NULL;
1224         p->audit_context = NULL;
1225         if (clone_flags & CLONE_THREAD)
1226                 threadgroup_change_begin(current);
1227         cgroup_fork(p);
1228 #ifdef CONFIG_NUMA
1229         p->mempolicy = mpol_dup(p->mempolicy);
1230         if (IS_ERR(p->mempolicy)) {
1231                 retval = PTR_ERR(p->mempolicy);
1232                 p->mempolicy = NULL;
1233                 goto bad_fork_cleanup_cgroup;
1234         }
1235         mpol_fix_fork_child_flag(p);
1236 #endif
1237 #ifdef CONFIG_CPUSETS
1238         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1239         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1240         seqcount_init(&p->mems_allowed_seq);
1241 #endif
1242 #ifdef CONFIG_TRACE_IRQFLAGS
1243         p->irq_events = 0;
1244 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1245         p->hardirqs_enabled = 1;
1246 #else
1247         p->hardirqs_enabled = 0;
1248 #endif
1249         p->hardirq_enable_ip = 0;
1250         p->hardirq_enable_event = 0;
1251         p->hardirq_disable_ip = _THIS_IP_;
1252         p->hardirq_disable_event = 0;
1253         p->softirqs_enabled = 1;
1254         p->softirq_enable_ip = _THIS_IP_;
1255         p->softirq_enable_event = 0;
1256         p->softirq_disable_ip = 0;
1257         p->softirq_disable_event = 0;
1258         p->hardirq_context = 0;
1259         p->softirq_context = 0;
1260 #endif
1261 #ifdef CONFIG_LOCKDEP
1262         p->lockdep_depth = 0; /* no locks held yet */
1263         p->curr_chain_key = 0;
1264         p->lockdep_recursion = 0;
1265 #endif
1266
1267 #ifdef CONFIG_DEBUG_MUTEXES
1268         p->blocked_on = NULL; /* not blocked yet */
1269 #endif
1270 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1271         p->memcg_batch.do_batch = 0;
1272         p->memcg_batch.memcg = NULL;
1273 #endif
1274
1275         /* Perform scheduler related setup. Assign this task to a CPU. */
1276         sched_fork(p);
1277
1278         retval = perf_event_init_task(p);
1279         if (retval)
1280                 goto bad_fork_cleanup_policy;
1281         retval = audit_alloc(p);
1282         if (retval)
1283                 goto bad_fork_cleanup_policy;
1284         /* copy all the process information */
1285         retval = copy_semundo(clone_flags, p);
1286         if (retval)
1287                 goto bad_fork_cleanup_audit;
1288         retval = copy_files(clone_flags, p);
1289         if (retval)
1290                 goto bad_fork_cleanup_semundo;
1291         retval = copy_fs(clone_flags, p);
1292         if (retval)
1293                 goto bad_fork_cleanup_files;
1294         retval = copy_sighand(clone_flags, p);
1295         if (retval)
1296                 goto bad_fork_cleanup_fs;
1297         retval = copy_signal(clone_flags, p);
1298         if (retval)
1299                 goto bad_fork_cleanup_sighand;
1300         retval = copy_mm(clone_flags, p);
1301         if (retval)
1302                 goto bad_fork_cleanup_signal;
1303         retval = copy_namespaces(clone_flags, p);
1304         if (retval)
1305                 goto bad_fork_cleanup_mm;
1306         retval = copy_io(clone_flags, p);
1307         if (retval)
1308                 goto bad_fork_cleanup_namespaces;
1309         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1310         if (retval)
1311                 goto bad_fork_cleanup_io;
1312
1313         if (pid != &init_struct_pid) {
1314                 retval = -ENOMEM;
1315                 pid = alloc_pid(p->nsproxy->pid_ns);
1316                 if (!pid)
1317                         goto bad_fork_cleanup_io;
1318         }
1319
1320         p->pid = pid_nr(pid);
1321         p->tgid = p->pid;
1322         if (clone_flags & CLONE_THREAD)
1323                 p->tgid = current->tgid;
1324
1325         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1326         /*
1327          * Clear TID on mm_release()?
1328          */
1329         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1330 #ifdef CONFIG_BLOCK
1331         p->plug = NULL;
1332 #endif
1333 #ifdef CONFIG_FUTEX
1334         p->robust_list = NULL;
1335 #ifdef CONFIG_COMPAT
1336         p->compat_robust_list = NULL;
1337 #endif
1338         INIT_LIST_HEAD(&p->pi_state_list);
1339         p->pi_state_cache = NULL;
1340 #endif
1341         /*
1342          * sigaltstack should be cleared when sharing the same VM
1343          */
1344         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1345                 p->sas_ss_sp = p->sas_ss_size = 0;
1346
1347         /*
1348          * Syscall tracing and stepping should be turned off in the
1349          * child regardless of CLONE_PTRACE.
1350          */
1351         user_disable_single_step(p);
1352         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1353 #ifdef TIF_SYSCALL_EMU
1354         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1355 #endif
1356         clear_all_latency_tracing(p);
1357
1358         /* ok, now we should be set up.. */
1359         if (clone_flags & CLONE_THREAD)
1360                 p->exit_signal = -1;
1361         else if (clone_flags & CLONE_PARENT)
1362                 p->exit_signal = current->group_leader->exit_signal;
1363         else
1364                 p->exit_signal = (clone_flags & CSIGNAL);
1365
1366         p->pdeath_signal = 0;
1367         p->exit_state = 0;
1368
1369         p->nr_dirtied = 0;
1370         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1371         p->dirty_paused_when = 0;
1372
1373         /*
1374          * Ok, make it visible to the rest of the system.
1375          * We dont wake it up yet.
1376          */
1377         p->group_leader = p;
1378         INIT_LIST_HEAD(&p->thread_group);
1379
1380         /* Now that the task is set up, run cgroup callbacks if
1381          * necessary. We need to run them before the task is visible
1382          * on the tasklist. */
1383         cgroup_fork_callbacks(p);
1384         cgroup_callbacks_done = 1;
1385
1386         /* Need tasklist lock for parent etc handling! */
1387         write_lock_irq(&tasklist_lock);
1388
1389         /* CLONE_PARENT re-uses the old parent */
1390         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1391                 p->real_parent = current->real_parent;
1392                 p->parent_exec_id = current->parent_exec_id;
1393         } else {
1394                 p->real_parent = current;
1395                 p->parent_exec_id = current->self_exec_id;
1396         }
1397
1398         spin_lock(&current->sighand->siglock);
1399
1400         /*
1401          * Process group and session signals need to be delivered to just the
1402          * parent before the fork or both the parent and the child after the
1403          * fork. Restart if a signal comes in before we add the new process to
1404          * it's process group.
1405          * A fatal signal pending means that current will exit, so the new
1406          * thread can't slip out of an OOM kill (or normal SIGKILL).
1407         */
1408         recalc_sigpending();
1409         if (signal_pending(current)) {
1410                 spin_unlock(&current->sighand->siglock);
1411                 write_unlock_irq(&tasklist_lock);
1412                 retval = -ERESTARTNOINTR;
1413                 goto bad_fork_free_pid;
1414         }
1415
1416         if (clone_flags & CLONE_THREAD) {
1417                 current->signal->nr_threads++;
1418                 atomic_inc(&current->signal->live);
1419                 atomic_inc(&current->signal->sigcnt);
1420                 p->group_leader = current->group_leader;
1421                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1422         }
1423
1424         if (likely(p->pid)) {
1425                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1426
1427                 if (thread_group_leader(p)) {
1428                         if (is_child_reaper(pid))
1429                                 p->nsproxy->pid_ns->child_reaper = p;
1430
1431                         p->signal->leader_pid = pid;
1432                         p->signal->tty = tty_kref_get(current->signal->tty);
1433                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1434                         attach_pid(p, PIDTYPE_SID, task_session(current));
1435                         list_add_tail(&p->sibling, &p->real_parent->children);
1436                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1437                         __this_cpu_inc(process_counts);
1438                 }
1439                 attach_pid(p, PIDTYPE_PID, pid);
1440                 nr_threads++;
1441         }
1442
1443         total_forks++;
1444         spin_unlock(&current->sighand->siglock);
1445         write_unlock_irq(&tasklist_lock);
1446         proc_fork_connector(p);
1447         cgroup_post_fork(p);
1448         if (clone_flags & CLONE_THREAD)
1449                 threadgroup_change_end(current);
1450         perf_event_fork(p);
1451
1452         trace_task_newtask(p, clone_flags);
1453
1454         return p;
1455
1456 bad_fork_free_pid:
1457         if (pid != &init_struct_pid)
1458                 free_pid(pid);
1459 bad_fork_cleanup_io:
1460         if (p->io_context)
1461                 exit_io_context(p);
1462 bad_fork_cleanup_namespaces:
1463         exit_task_namespaces(p);
1464 bad_fork_cleanup_mm:
1465         if (p->mm)
1466                 mmput(p->mm);
1467 bad_fork_cleanup_signal:
1468         if (!(clone_flags & CLONE_THREAD))
1469                 free_signal_struct(p->signal);
1470 bad_fork_cleanup_sighand:
1471         __cleanup_sighand(p->sighand);
1472 bad_fork_cleanup_fs:
1473         exit_fs(p); /* blocking */
1474 bad_fork_cleanup_files:
1475         exit_files(p); /* blocking */
1476 bad_fork_cleanup_semundo:
1477         exit_sem(p);
1478 bad_fork_cleanup_audit:
1479         audit_free(p);
1480 bad_fork_cleanup_policy:
1481         perf_event_free_task(p);
1482 #ifdef CONFIG_NUMA
1483         mpol_put(p->mempolicy);
1484 bad_fork_cleanup_cgroup:
1485 #endif
1486         if (clone_flags & CLONE_THREAD)
1487                 threadgroup_change_end(current);
1488         cgroup_exit(p, cgroup_callbacks_done);
1489         delayacct_tsk_free(p);
1490         module_put(task_thread_info(p)->exec_domain->module);
1491 bad_fork_cleanup_count:
1492         atomic_dec(&p->cred->user->processes);
1493         exit_creds(p);
1494 bad_fork_free:
1495         free_task(p);
1496 fork_out:
1497         return ERR_PTR(retval);
1498 }
1499
1500 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1501 {
1502         memset(regs, 0, sizeof(struct pt_regs));
1503         return regs;
1504 }
1505
1506 static inline void init_idle_pids(struct pid_link *links)
1507 {
1508         enum pid_type type;
1509
1510         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1511                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1512                 links[type].pid = &init_struct_pid;
1513         }
1514 }
1515
1516 struct task_struct * __cpuinit fork_idle(int cpu)
1517 {
1518         struct task_struct *task;
1519         struct pt_regs regs;
1520
1521         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1522                             &init_struct_pid, 0);
1523         if (!IS_ERR(task)) {
1524                 init_idle_pids(task->pids);
1525                 init_idle(task, cpu);
1526         }
1527
1528         return task;
1529 }
1530
1531 /*
1532  *  Ok, this is the main fork-routine.
1533  *
1534  * It copies the process, and if successful kick-starts
1535  * it and waits for it to finish using the VM if required.
1536  */
1537 long do_fork(unsigned long clone_flags,
1538               unsigned long stack_start,
1539               struct pt_regs *regs,
1540               unsigned long stack_size,
1541               int __user *parent_tidptr,
1542               int __user *child_tidptr)
1543 {
1544         struct task_struct *p;
1545         int trace = 0;
1546         long nr;
1547
1548         /*
1549          * Do some preliminary argument and permissions checking before we
1550          * actually start allocating stuff
1551          */
1552         if (clone_flags & CLONE_NEWUSER) {
1553                 if (clone_flags & CLONE_THREAD)
1554                         return -EINVAL;
1555                 /* hopefully this check will go away when userns support is
1556                  * complete
1557                  */
1558                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1559                                 !capable(CAP_SETGID))
1560                         return -EPERM;
1561         }
1562
1563         /*
1564          * Determine whether and which event to report to ptracer.  When
1565          * called from kernel_thread or CLONE_UNTRACED is explicitly
1566          * requested, no event is reported; otherwise, report if the event
1567          * for the type of forking is enabled.
1568          */
1569         if (likely(user_mode(regs)) && !(clone_flags & CLONE_UNTRACED)) {
1570                 if (clone_flags & CLONE_VFORK)
1571                         trace = PTRACE_EVENT_VFORK;
1572                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1573                         trace = PTRACE_EVENT_CLONE;
1574                 else
1575                         trace = PTRACE_EVENT_FORK;
1576
1577                 if (likely(!ptrace_event_enabled(current, trace)))
1578                         trace = 0;
1579         }
1580
1581         p = copy_process(clone_flags, stack_start, regs, stack_size,
1582                          child_tidptr, NULL, trace);
1583         /*
1584          * Do this prior waking up the new thread - the thread pointer
1585          * might get invalid after that point, if the thread exits quickly.
1586          */
1587         if (!IS_ERR(p)) {
1588                 struct completion vfork;
1589
1590                 trace_sched_process_fork(current, p);
1591
1592                 nr = task_pid_vnr(p);
1593
1594                 if (clone_flags & CLONE_PARENT_SETTID)
1595                         put_user(nr, parent_tidptr);
1596
1597                 if (clone_flags & CLONE_VFORK) {
1598                         p->vfork_done = &vfork;
1599                         init_completion(&vfork);
1600                         get_task_struct(p);
1601                 }
1602
1603                 wake_up_new_task(p);
1604
1605                 /* forking complete and child started to run, tell ptracer */
1606                 if (unlikely(trace))
1607                         ptrace_event(trace, nr);
1608
1609                 if (clone_flags & CLONE_VFORK) {
1610                         if (!wait_for_vfork_done(p, &vfork))
1611                                 ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1612                 }
1613         } else {
1614                 nr = PTR_ERR(p);
1615         }
1616         return nr;
1617 }
1618
1619 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1620 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1621 #endif
1622
1623 static void sighand_ctor(void *data)
1624 {
1625         struct sighand_struct *sighand = data;
1626
1627         spin_lock_init(&sighand->siglock);
1628         init_waitqueue_head(&sighand->signalfd_wqh);
1629 }
1630
1631 void __init proc_caches_init(void)
1632 {
1633         sighand_cachep = kmem_cache_create("sighand_cache",
1634                         sizeof(struct sighand_struct), 0,
1635                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1636                         SLAB_NOTRACK, sighand_ctor);
1637         signal_cachep = kmem_cache_create("signal_cache",
1638                         sizeof(struct signal_struct), 0,
1639                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1640         files_cachep = kmem_cache_create("files_cache",
1641                         sizeof(struct files_struct), 0,
1642                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1643         fs_cachep = kmem_cache_create("fs_cache",
1644                         sizeof(struct fs_struct), 0,
1645                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1646         /*
1647          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1648          * whole struct cpumask for the OFFSTACK case. We could change
1649          * this to *only* allocate as much of it as required by the
1650          * maximum number of CPU's we can ever have.  The cpumask_allocation
1651          * is at the end of the structure, exactly for that reason.
1652          */
1653         mm_cachep = kmem_cache_create("mm_struct",
1654                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1655                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1656         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1657         mmap_init();
1658         nsproxy_cache_init();
1659 }
1660
1661 /*
1662  * Check constraints on flags passed to the unshare system call.
1663  */
1664 static int check_unshare_flags(unsigned long unshare_flags)
1665 {
1666         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1667                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1668                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1669                 return -EINVAL;
1670         /*
1671          * Not implemented, but pretend it works if there is nothing to
1672          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1673          * needs to unshare vm.
1674          */
1675         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1676                 /* FIXME: get_task_mm() increments ->mm_users */
1677                 if (atomic_read(&current->mm->mm_users) > 1)
1678                         return -EINVAL;
1679         }
1680
1681         return 0;
1682 }
1683
1684 /*
1685  * Unshare the filesystem structure if it is being shared
1686  */
1687 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1688 {
1689         struct fs_struct *fs = current->fs;
1690
1691         if (!(unshare_flags & CLONE_FS) || !fs)
1692                 return 0;
1693
1694         /* don't need lock here; in the worst case we'll do useless copy */
1695         if (fs->users == 1)
1696                 return 0;
1697
1698         *new_fsp = copy_fs_struct(fs);
1699         if (!*new_fsp)
1700                 return -ENOMEM;
1701
1702         return 0;
1703 }
1704
1705 /*
1706  * Unshare file descriptor table if it is being shared
1707  */
1708 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1709 {
1710         struct files_struct *fd = current->files;
1711         int error = 0;
1712
1713         if ((unshare_flags & CLONE_FILES) &&
1714             (fd && atomic_read(&fd->count) > 1)) {
1715                 *new_fdp = dup_fd(fd, &error);
1716                 if (!*new_fdp)
1717                         return error;
1718         }
1719
1720         return 0;
1721 }
1722
1723 /*
1724  * unshare allows a process to 'unshare' part of the process
1725  * context which was originally shared using clone.  copy_*
1726  * functions used by do_fork() cannot be used here directly
1727  * because they modify an inactive task_struct that is being
1728  * constructed. Here we are modifying the current, active,
1729  * task_struct.
1730  */
1731 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1732 {
1733         struct fs_struct *fs, *new_fs = NULL;
1734         struct files_struct *fd, *new_fd = NULL;
1735         struct nsproxy *new_nsproxy = NULL;
1736         int do_sysvsem = 0;
1737         int err;
1738
1739         err = check_unshare_flags(unshare_flags);
1740         if (err)
1741                 goto bad_unshare_out;
1742
1743         /*
1744          * If unsharing namespace, must also unshare filesystem information.
1745          */
1746         if (unshare_flags & CLONE_NEWNS)
1747                 unshare_flags |= CLONE_FS;
1748         /*
1749          * CLONE_NEWIPC must also detach from the undolist: after switching
1750          * to a new ipc namespace, the semaphore arrays from the old
1751          * namespace are unreachable.
1752          */
1753         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1754                 do_sysvsem = 1;
1755         err = unshare_fs(unshare_flags, &new_fs);
1756         if (err)
1757                 goto bad_unshare_out;
1758         err = unshare_fd(unshare_flags, &new_fd);
1759         if (err)
1760                 goto bad_unshare_cleanup_fs;
1761         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy, new_fs);
1762         if (err)
1763                 goto bad_unshare_cleanup_fd;
1764
1765         if (new_fs || new_fd || do_sysvsem || new_nsproxy) {
1766                 if (do_sysvsem) {
1767                         /*
1768                          * CLONE_SYSVSEM is equivalent to sys_exit().
1769                          */
1770                         exit_sem(current);
1771                 }
1772
1773                 if (new_nsproxy) {
1774                         switch_task_namespaces(current, new_nsproxy);
1775                         new_nsproxy = NULL;
1776                 }
1777
1778                 task_lock(current);
1779
1780                 if (new_fs) {
1781                         fs = current->fs;
1782                         spin_lock(&fs->lock);
1783                         current->fs = new_fs;
1784                         if (--fs->users)
1785                                 new_fs = NULL;
1786                         else
1787                                 new_fs = fs;
1788                         spin_unlock(&fs->lock);
1789                 }
1790
1791                 if (new_fd) {
1792                         fd = current->files;
1793                         current->files = new_fd;
1794                         new_fd = fd;
1795                 }
1796
1797                 task_unlock(current);
1798         }
1799
1800         if (new_nsproxy)
1801                 put_nsproxy(new_nsproxy);
1802
1803 bad_unshare_cleanup_fd:
1804         if (new_fd)
1805                 put_files_struct(new_fd);
1806
1807 bad_unshare_cleanup_fs:
1808         if (new_fs)
1809                 free_fs_struct(new_fs);
1810
1811 bad_unshare_out:
1812         return err;
1813 }
1814
1815 /*
1816  *      Helper to unshare the files of the current task.
1817  *      We don't want to expose copy_files internals to
1818  *      the exec layer of the kernel.
1819  */
1820
1821 int unshare_files(struct files_struct **displaced)
1822 {
1823         struct task_struct *task = current;
1824         struct files_struct *copy = NULL;
1825         int error;
1826
1827         error = unshare_fd(CLONE_FILES, &copy);
1828         if (error || !copy) {
1829                 *displaced = NULL;
1830                 return error;
1831         }
1832         *displaced = task->files;
1833         task_lock(task);
1834         task->files = copy;
1835         task_unlock(task);
1836         return 0;
1837 }