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/tracehook.h>
  41 #include <linux/futex.h>
  42 #include <linux/compat.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/proc_fs.h>
  62 #include <linux/blkdev.h>
  63 #include <linux/fs_struct.h>
  64 #include <linux/magic.h>
  65 #include <linux/perf_event.h>
  66 #include <linux/posix-timers.h>
  67 #include <linux/user-return-notifier.h>
  68 #include <linux/oom.h>
  69
  70 #include <asm/pgtable.h>
  71 #include <asm/pgalloc.h>
  72 #include <asm/uaccess.h>
  73 #include <asm/mmu_context.h>
  74 #include <asm/cacheflush.h>
  75 #include <asm/tlbflush.h>
  76
  77 #include <trace/events/sched.h>
  78
  79 /*
  80  * Protected counters by write_lock_irq(&tasklist_lock)
  81  */
  82 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  83 int nr_threads;                 /* The idle threads do not count.. */
  84
  85 int max_threads;                /* tunable limit on nr_threads */
  86
  87 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  88
  89 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  90
  91 #ifdef CONFIG_PROVE_RCU
  92 int lockdep_tasklist_lock_is_held(void)
  93 {
  94         return lockdep_is_held(&tasklist_lock);
  95 }
  96 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
  97 #endif /* #ifdef CONFIG_PROVE_RCU */
  98
  99 int nr_processes(void)
 100 {
 101         int cpu;
 102         int total = 0;
 103
 104         for_each_possible_cpu(cpu)
 105                 total += per_cpu(process_counts, cpu);
 106
 107         return total;
 108 }
 109
 110 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 111 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
 112 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
 113 static struct kmem_cache *task_struct_cachep;
 114 #endif
 115
 116 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
 117 static inline struct thread_info *alloc_thread_info(struct task_struct *tsk)
 118 {
 119 #ifdef CONFIG_DEBUG_STACK_USAGE
 120         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
 121 #else
 122         gfp_t mask = GFP_KERNEL;
 123 #endif
 124         return (struct thread_info *)__get_free_pages(mask, THREAD_SIZE_ORDER);
 125 }
 126
 127 static inline void free_thread_info(struct thread_info *ti)
 128 {
 129         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 130 }
 131 #endif
 132
 133 /* SLAB cache for signal_struct structures (tsk->signal) */
 134 static struct kmem_cache *signal_cachep;
 135
 136 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 137 struct kmem_cache *sighand_cachep;
 138
 139 /* SLAB cache for files_struct structures (tsk->files) */
 140 struct kmem_cache *files_cachep;
 141
 142 /* SLAB cache for fs_struct structures (tsk->fs) */
 143 struct kmem_cache *fs_cachep;
 144
 145 /* SLAB cache for vm_area_struct structures */
 146 struct kmem_cache *vm_area_cachep;
 147
 148 /* SLAB cache for mm_struct structures (tsk->mm) */
 149 static struct kmem_cache *mm_cachep;
 150
 151 static void account_kernel_stack(struct thread_info *ti, int account)
 152 {
 153         struct zone *zone = page_zone(virt_to_page(ti));
 154
 155         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
 156 }
 157
 158 void free_task(struct task_struct *tsk)
 159 {
 160         prop_local_destroy_single(&tsk->dirties);
 161         account_kernel_stack(tsk->stack, -1);
 162         free_thread_info(tsk->stack);
 163         rt_mutex_debug_task_free(tsk);
 164         ftrace_graph_exit_task(tsk);
 165         free_task_struct(tsk);
 166 }
 167 EXPORT_SYMBOL(free_task);
 168
 169 static inline void free_signal_struct(struct signal_struct *sig)
 170 {
 171         taskstats_tgid_free(sig);
 172         kmem_cache_free(signal_cachep, sig);
 173 }
 174
 175 static inline void put_signal_struct(struct signal_struct *sig)
 176 {
 177         if (atomic_dec_and_test(&sig->sigcnt))
 178                 free_signal_struct(sig);
 179 }
 180
 181 void __put_task_struct(struct task_struct *tsk)
 182 {
 183         WARN_ON(!tsk->exit_state);
 184         WARN_ON(atomic_read(&tsk->usage));
 185         WARN_ON(tsk == current);
 186
 187         exit_creds(tsk);
 188         delayacct_tsk_free(tsk);
 189         put_signal_struct(tsk->signal);
 190
 191         if (!profile_handoff_task(tsk))
 192                 free_task(tsk);
 193 }
 194
 195 /*
 196  * macro override instead of weak attribute alias, to workaround
 197  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
 198  */
 199 #ifndef arch_task_cache_init
 200 #define arch_task_cache_init()
 201 #endif
 202
 203 void __init fork_init(unsigned long mempages)
 204 {
 205 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 206 #ifndef ARCH_MIN_TASKALIGN
 207 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 208 #endif
 209         /* create a slab on which task_structs can be allocated */
 210         task_struct_cachep =
 211                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 212                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
 213 #endif
 214
 215         /* do the arch specific task caches init */
 216         arch_task_cache_init();
 217
 218         /*
 219          * The default maximum number of threads is set to a safe
 220          * value: the thread structures can take up at most half
 221          * of memory.
 222          */
 223         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 224
 225         /*
 226          * we need to allow at least 20 threads to boot a system
 227          */
 228         if(max_threads < 20)
 229                 max_threads = 20;
 230
 231         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 232         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 233         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 234                 init_task.signal->rlim[RLIMIT_NPROC];
 235 }
 236
 237 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
 238                                                struct task_struct *src)
 239 {
 240         *dst = *src;
 241         return 0;
 242 }
 243
 244 static struct task_struct *dup_task_struct(struct task_struct *orig)
 245 {
 246         struct task_struct *tsk;
 247         struct thread_info *ti;
 248         unsigned long *stackend;
 249
 250         int err;
 251
 252         prepare_to_copy(orig);
 253
 254         tsk = alloc_task_struct();
 255         if (!tsk)
 256                 return NULL;
 257
 258         ti = alloc_thread_info(tsk);
 259         if (!ti) {
 260                 free_task_struct(tsk);
 261                 return NULL;
 262         }
 263
 264         err = arch_dup_task_struct(tsk, orig);
 265         if (err)
 266                 goto out;
 267
 268         tsk->stack = ti;
 269
 270         err = prop_local_init_single(&tsk->dirties);
 271         if (err)
 272                 goto out;
 273
 274         setup_thread_stack(tsk, orig);
 275         clear_user_return_notifier(tsk);
 276         clear_tsk_need_resched(tsk);
 277         stackend = end_of_stack(tsk);
 278         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 279
 280 #ifdef CONFIG_CC_STACKPROTECTOR
 281         tsk->stack_canary = get_random_int();
 282 #endif
 283
 284         /* One for us, one for whoever does the "release_task()" (usually parent) */
 285         atomic_set(&tsk->usage,2);
 286         atomic_set(&tsk->fs_excl, 0);
 287 #ifdef CONFIG_BLK_DEV_IO_TRACE
 288         tsk->btrace_seq = 0;
 289 #endif
 290         tsk->splice_pipe = NULL;
 291
 292         account_kernel_stack(ti, 1);
 293
 294         return tsk;
 295
 296 out:
 297         free_thread_info(ti);
 298         free_task_struct(tsk);
 299         return NULL;
 300 }
 301
 302 #ifdef CONFIG_MMU
 303 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 304 {
 305         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 306         struct rb_node **rb_link, *rb_parent;
 307         int retval;
 308         unsigned long charge;
 309         struct mempolicy *pol;
 310
 311         down_write(&oldmm->mmap_sem);
 312         flush_cache_dup_mm(oldmm);
 313         /*
 314          * Not linked in yet - no deadlock potential:
 315          */
 316         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 317
 318         mm->locked_vm = 0;
 319         mm->mmap = NULL;
 320         mm->mmap_cache = NULL;
 321         mm->free_area_cache = oldmm->mmap_base;
 322         mm->cached_hole_size = ~0UL;
 323         mm->map_count = 0;
 324         cpumask_clear(mm_cpumask(mm));
 325         mm->mm_rb = RB_ROOT;
 326         rb_link = &mm->mm_rb.rb_node;
 327         rb_parent = NULL;
 328         pprev = &mm->mmap;
 329         retval = ksm_fork(mm, oldmm);
 330         if (retval)
 331                 goto out;
 332
 333         prev = NULL;
 334         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 335                 struct file *file;
 336
 337                 if (mpnt->vm_flags & VM_DONTCOPY) {
 338                         long pages = vma_pages(mpnt);
 339                         mm->total_vm -= pages;
 340                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 341                                                                 -pages);
 342                         continue;
 343                 }
 344                 charge = 0;
 345                 if (mpnt->vm_flags & VM_ACCOUNT) {
 346                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 347                         if (security_vm_enough_memory(len))
 348                                 goto fail_nomem;
 349                         charge = len;
 350                 }
 351                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 352                 if (!tmp)
 353                         goto fail_nomem;
 354                 *tmp = *mpnt;
 355                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 356                 pol = mpol_dup(vma_policy(mpnt));
 357                 retval = PTR_ERR(pol);
 358                 if (IS_ERR(pol))
 359                         goto fail_nomem_policy;
 360                 vma_set_policy(tmp, pol);
 361                 tmp->vm_mm = mm;
 362                 if (anon_vma_fork(tmp, mpnt))
 363                         goto fail_nomem_anon_vma_fork;
 364                 tmp->vm_flags &= ~VM_LOCKED;
 365                 tmp->vm_next = tmp->vm_prev = NULL;
 366                 file = tmp->vm_file;
 367                 if (file) {
 368                         struct inode *inode = file->f_path.dentry->d_inode;
 369                         struct address_space *mapping = file->f_mapping;
 370
 371                         get_file(file);
 372                         if (tmp->vm_flags & VM_DENYWRITE)
 373                                 atomic_dec(&inode->i_writecount);
 374                         spin_lock(&mapping->i_mmap_lock);
 375                         if (tmp->vm_flags & VM_SHARED)
 376                                 mapping->i_mmap_writable++;
 377                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
 378                         flush_dcache_mmap_lock(mapping);
 379                         /* insert tmp into the share list, just after mpnt */
 380                         vma_prio_tree_add(tmp, mpnt);
 381                         flush_dcache_mmap_unlock(mapping);
 382                         spin_unlock(&mapping->i_mmap_lock);
 383                 }
 384
 385                 /*
 386                  * Clear hugetlb-related page reserves for children. This only
 387                  * affects MAP_PRIVATE mappings. Faults generated by the child
 388                  * are not guaranteed to succeed, even if read-only
 389                  */
 390                 if (is_vm_hugetlb_page(tmp))
 391                         reset_vma_resv_huge_pages(tmp);
 392
 393                 /*
 394                  * Link in the new vma and copy the page table entries.
 395                  */
 396                 *pprev = tmp;
 397                 pprev = &tmp->vm_next;
 398                 tmp->vm_prev = prev;
 399                 prev = tmp;
 400
 401                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 402                 rb_link = &tmp->vm_rb.rb_right;
 403                 rb_parent = &tmp->vm_rb;
 404
 405                 mm->map_count++;
 406                 retval = copy_page_range(mm, oldmm, mpnt);
 407
 408                 if (tmp->vm_ops && tmp->vm_ops->open)
 409                         tmp->vm_ops->open(tmp);
 410
 411                 if (retval)
 412                         goto out;
 413         }
 414         /* a new mm has just been created */
 415         arch_dup_mmap(oldmm, mm);
 416         retval = 0;
 417 out:
 418         up_write(&mm->mmap_sem);
 419         flush_tlb_mm(oldmm);
 420         up_write(&oldmm->mmap_sem);
 421         return retval;
 422 fail_nomem_anon_vma_fork:
 423         mpol_put(pol);
 424 fail_nomem_policy:
 425         kmem_cache_free(vm_area_cachep, tmp);
 426 fail_nomem:
 427         retval = -ENOMEM;
 428         vm_unacct_memory(charge);
 429         goto out;
 430 }
 431
 432 static inline int mm_alloc_pgd(struct mm_struct * mm)
 433 {
 434         mm->pgd = pgd_alloc(mm);
 435         if (unlikely(!mm->pgd))
 436                 return -ENOMEM;
 437         return 0;
 438 }
 439
 440 static inline void mm_free_pgd(struct mm_struct * mm)
 441 {
 442         pgd_free(mm, mm->pgd);
 443 }
 444 #else
 445 #define dup_mmap(mm, oldmm)     (0)
 446 #define mm_alloc_pgd(mm)        (0)
 447 #define mm_free_pgd(mm)
 448 #endif /* CONFIG_MMU */
 449
 450 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 451
 452 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 453 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 454
 455 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 456
 457 static int __init coredump_filter_setup(char *s)
 458 {
 459         default_dump_filter =
 460                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 461                 MMF_DUMP_FILTER_MASK;
 462         return 1;
 463 }
 464
 465 __setup("coredump_filter=", coredump_filter_setup);
 466
 467 #include <linux/init_task.h>
 468
 469 static void mm_init_aio(struct mm_struct *mm)
 470 {
 471 #ifdef CONFIG_AIO
 472         spin_lock_init(&mm->ioctx_lock);
 473         INIT_HLIST_HEAD(&mm->ioctx_list);
 474 #endif
 475 }
 476
 477 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
 478 {
 479         atomic_set(&mm->mm_users, 1);
 480         atomic_set(&mm->mm_count, 1);
 481         init_rwsem(&mm->mmap_sem);
 482         INIT_LIST_HEAD(&mm->mmlist);
 483         mm->flags = (current->mm) ?
 484                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
 485         mm->core_state = NULL;
 486         mm->nr_ptes = 0;
 487         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 488         spin_lock_init(&mm->page_table_lock);
 489         mm->free_area_cache = TASK_UNMAPPED_BASE;
 490         mm->cached_hole_size = ~0UL;
 491         mm_init_aio(mm);
 492         mm_init_owner(mm, p);
 493         atomic_set(&mm->oom_disable_count, 0);
 494
 495         if (likely(!mm_alloc_pgd(mm))) {
 496                 mm->def_flags = 0;
 497                 mmu_notifier_mm_init(mm);
 498                 return mm;
 499         }
 500
 501         free_mm(mm);
 502         return NULL;
 503 }
 504
 505 /*
 506  * Allocate and initialize an mm_struct.
 507  */
 508 struct mm_struct * mm_alloc(void)
 509 {
 510         struct mm_struct * mm;
 511
 512         mm = allocate_mm();
 513         if (mm) {
 514                 memset(mm, 0, sizeof(*mm));
 515                 mm = mm_init(mm, current);
 516         }
 517         return mm;
 518 }
 519
 520 /*
 521  * Called when the last reference to the mm
 522  * is dropped: either by a lazy thread or by
 523  * mmput. Free the page directory and the mm.
 524  */
 525 void __mmdrop(struct mm_struct *mm)
 526 {
 527         BUG_ON(mm == &init_mm);
 528         mm_free_pgd(mm);
 529         destroy_context(mm);
 530         mmu_notifier_mm_destroy(mm);
 531         free_mm(mm);
 532 }
 533 EXPORT_SYMBOL_GPL(__mmdrop);
 534
 535 /*
 536  * Decrement the use count and release all resources for an mm.
 537  */
 538 void mmput(struct mm_struct *mm)
 539 {
 540         might_sleep();
 541
 542         if (atomic_dec_and_test(&mm->mm_users)) {
 543                 exit_aio(mm);
 544                 ksm_exit(mm);
 545                 exit_mmap(mm);
 546                 set_mm_exe_file(mm, NULL);
 547                 if (!list_empty(&mm->mmlist)) {
 548                         spin_lock(&mmlist_lock);
 549                         list_del(&mm->mmlist);
 550                         spin_unlock(&mmlist_lock);
 551                 }
 552                 put_swap_token(mm);
 553                 if (mm->binfmt)
 554                         module_put(mm->binfmt->module);
 555                 mmdrop(mm);
 556         }
 557 }
 558 EXPORT_SYMBOL_GPL(mmput);
 559
 560 /**
 561  * get_task_mm - acquire a reference to the task's mm
 562  *
 563  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 564  * this kernel workthread has transiently adopted a user mm with use_mm,
 565  * to do its AIO) is not set and if so returns a reference to it, after
 566  * bumping up the use count.  User must release the mm via mmput()
 567  * after use.  Typically used by /proc and ptrace.
 568  */
 569 struct mm_struct *get_task_mm(struct task_struct *task)
 570 {
 571         struct mm_struct *mm;
 572
 573         task_lock(task);
 574         mm = task->mm;
 575         if (mm) {
 576                 if (task->flags & PF_KTHREAD)
 577                         mm = NULL;
 578                 else
 579                         atomic_inc(&mm->mm_users);
 580         }
 581         task_unlock(task);
 582         return mm;
 583 }
 584 EXPORT_SYMBOL_GPL(get_task_mm);
 585
 586 /* Please note the differences between mmput and mm_release.
 587  * mmput is called whenever we stop holding onto a mm_struct,
 588  * error success whatever.
 589  *
 590  * mm_release is called after a mm_struct has been removed
 591  * from the current process.
 592  *
 593  * This difference is important for error handling, when we
 594  * only half set up a mm_struct for a new process and need to restore
 595  * the old one.  Because we mmput the new mm_struct before
 596  * restoring the old one. . .
 597  * Eric Biederman 10 January 1998
 598  */
 599 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 600 {
 601         struct completion *vfork_done = tsk->vfork_done;
 602
 603         /* Get rid of any futexes when releasing the mm */
 604 #ifdef CONFIG_FUTEX
 605         if (unlikely(tsk->robust_list)) {
 606                 exit_robust_list(tsk);
 607                 tsk->robust_list = NULL;
 608         }
 609 #ifdef CONFIG_COMPAT
 610         if (unlikely(tsk->compat_robust_list)) {
 611                 compat_exit_robust_list(tsk);
 612                 tsk->compat_robust_list = NULL;
 613         }
 614 #endif
 615         if (unlikely(!list_empty(&tsk->pi_state_list)))
 616                 exit_pi_state_list(tsk);
 617 #endif
 618
 619         /* Get rid of any cached register state */
 620         deactivate_mm(tsk, mm);
 621
 622         /* notify parent sleeping on vfork() */
 623         if (vfork_done) {
 624                 tsk->vfork_done = NULL;
 625                 complete(vfork_done);
 626         }
 627
 628         /*
 629          * If we're exiting normally, clear a user-space tid field if
 630          * requested.  We leave this alone when dying by signal, to leave
 631          * the value intact in a core dump, and to save the unnecessary
 632          * trouble otherwise.  Userland only wants this done for a sys_exit.
 633          */
 634         if (tsk->clear_child_tid) {
 635                 if (!(tsk->flags & PF_SIGNALED) &&
 636                     atomic_read(&mm->mm_users) > 1) {
 637                         /*
 638                          * We don't check the error code - if userspace has
 639                          * not set up a proper pointer then tough luck.
 640                          */
 641                         put_user(0, tsk->clear_child_tid);
 642                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 643                                         1, NULL, NULL, 0);
 644                 }
 645                 tsk->clear_child_tid = NULL;
 646         }
 647 }
 648
 649 /*
 650  * Allocate a new mm structure and copy contents from the
 651  * mm structure of the passed in task structure.
 652  */
 653 struct mm_struct *dup_mm(struct task_struct *tsk)
 654 {
 655         struct mm_struct *mm, *oldmm = current->mm;
 656         int err;
 657
 658         if (!oldmm)
 659                 return NULL;
 660
 661         mm = allocate_mm();
 662         if (!mm)
 663                 goto fail_nomem;
 664
 665         memcpy(mm, oldmm, sizeof(*mm));
 666
 667         /* Initializing for Swap token stuff */
 668         mm->token_priority = 0;
 669         mm->last_interval = 0;
 670
 671         if (!mm_init(mm, tsk))
 672                 goto fail_nomem;
 673
 674         if (init_new_context(tsk, mm))
 675                 goto fail_nocontext;
 676
 677         dup_mm_exe_file(oldmm, mm);
 678
 679         err = dup_mmap(mm, oldmm);
 680         if (err)
 681                 goto free_pt;
 682
 683         mm->hiwater_rss = get_mm_rss(mm);
 684         mm->hiwater_vm = mm->total_vm;
 685
 686         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 687                 goto free_pt;
 688
 689         return mm;
 690
 691 free_pt:
 692         /* don't put binfmt in mmput, we haven't got module yet */
 693         mm->binfmt = NULL;
 694         mmput(mm);
 695
 696 fail_nomem:
 697         return NULL;
 698
 699 fail_nocontext:
 700         /*
 701          * If init_new_context() failed, we cannot use mmput() to free the mm
 702          * because it calls destroy_context()
 703          */
 704         mm_free_pgd(mm);
 705         free_mm(mm);
 706         return NULL;
 707 }
 708
 709 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 710 {
 711         struct mm_struct * mm, *oldmm;
 712         int retval;
 713
 714         tsk->min_flt = tsk->maj_flt = 0;
 715         tsk->nvcsw = tsk->nivcsw = 0;
 716 #ifdef CONFIG_DETECT_HUNG_TASK
 717         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 718 #endif
 719
 720         tsk->mm = NULL;
 721         tsk->active_mm = NULL;
 722
 723         /*
 724          * Are we cloning a kernel thread?
 725          *
 726          * We need to steal a active VM for that..
 727          */
 728         oldmm = current->mm;
 729         if (!oldmm)
 730                 return 0;
 731
 732         if (clone_flags & CLONE_VM) {
 733                 atomic_inc(&oldmm->mm_users);
 734                 mm = oldmm;
 735                 goto good_mm;
 736         }
 737
 738         retval = -ENOMEM;
 739         mm = dup_mm(tsk);
 740         if (!mm)
 741                 goto fail_nomem;
 742
 743 good_mm:
 744         /* Initializing for Swap token stuff */
 745         mm->token_priority = 0;
 746         mm->last_interval = 0;
 747         if (tsk->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
 748                 atomic_inc(&mm->oom_disable_count);
 749
 750         tsk->mm = mm;
 751         tsk->active_mm = mm;
 752         return 0;
 753
 754 fail_nomem:
 755         return retval;
 756 }
 757
 758 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 759 {
 760         struct fs_struct *fs = current->fs;
 761         if (clone_flags & CLONE_FS) {
 762                 /* tsk->fs is already what we want */
 763                 spin_lock(&fs->lock);
 764                 if (fs->in_exec) {
 765                         spin_unlock(&fs->lock);
 766                         return -EAGAIN;
 767                 }
 768                 fs->users++;
 769                 spin_unlock(&fs->lock);
 770                 return 0;
 771         }
 772         tsk->fs = copy_fs_struct(fs);
 773         if (!tsk->fs)
 774                 return -ENOMEM;
 775         return 0;
 776 }
 777
 778 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 779 {
 780         struct files_struct *oldf, *newf;
 781         int error = 0;
 782
 783         /*
 784          * A background process may not have any files ...
 785          */
 786         oldf = current->files;
 787         if (!oldf)
 788                 goto out;
 789
 790         if (clone_flags & CLONE_FILES) {
 791                 atomic_inc(&oldf->count);
 792                 goto out;
 793         }
 794
 795         newf = dup_fd(oldf, &error);
 796         if (!newf)
 797                 goto out;
 798
 799         tsk->files = newf;
 800         error = 0;
 801 out:
 802         return error;
 803 }
 804
 805 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 806 {
 807 #ifdef CONFIG_BLOCK
 808         struct io_context *ioc = current->io_context;
 809
 810         if (!ioc)
 811                 return 0;
 812         /*
 813          * Share io context with parent, if CLONE_IO is set
 814          */
 815         if (clone_flags & CLONE_IO) {
 816                 tsk->io_context = ioc_task_link(ioc);
 817                 if (unlikely(!tsk->io_context))
 818                         return -ENOMEM;
 819         } else if (ioprio_valid(ioc->ioprio)) {
 820                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
 821                 if (unlikely(!tsk->io_context))
 822                         return -ENOMEM;
 823
 824                 tsk->io_context->ioprio = ioc->ioprio;
 825         }
 826 #endif
 827         return 0;
 828 }
 829
 830 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 831 {
 832         struct sighand_struct *sig;
 833
 834         if (clone_flags & CLONE_SIGHAND) {
 835                 atomic_inc(&current->sighand->count);
 836                 return 0;
 837         }
 838         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 839         rcu_assign_pointer(tsk->sighand, sig);
 840         if (!sig)
 841                 return -ENOMEM;
 842         atomic_set(&sig->count, 1);
 843         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 844         return 0;
 845 }
 846
 847 void __cleanup_sighand(struct sighand_struct *sighand)
 848 {
 849         if (atomic_dec_and_test(&sighand->count))
 850                 kmem_cache_free(sighand_cachep, sighand);
 851 }
 852
 853
 854 /*
 855  * Initialize POSIX timer handling for a thread group.
 856  */
 857 static void posix_cpu_timers_init_group(struct signal_struct *sig)
 858 {
 859         unsigned long cpu_limit;
 860
 861         /* Thread group counters. */
 862         thread_group_cputime_init(sig);
 863
 864         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
 865         if (cpu_limit != RLIM_INFINITY) {
 866                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
 867                 sig->cputimer.running = 1;
 868         }
 869
 870         /* The timer lists. */
 871         INIT_LIST_HEAD(&sig->cpu_timers[0]);
 872         INIT_LIST_HEAD(&sig->cpu_timers[1]);
 873         INIT_LIST_HEAD(&sig->cpu_timers[2]);
 874 }
 875
 876 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
 877 {
 878         struct signal_struct *sig;
 879
 880         if (clone_flags & CLONE_THREAD)
 881                 return 0;
 882
 883         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
 884         tsk->signal = sig;
 885         if (!sig)
 886                 return -ENOMEM;
 887
 888         sig->nr_threads = 1;
 889         atomic_set(&sig->live, 1);
 890         atomic_set(&sig->sigcnt, 1);
 891         init_waitqueue_head(&sig->wait_chldexit);
 892         if (clone_flags & CLONE_NEWPID)
 893                 sig->flags |= SIGNAL_UNKILLABLE;
 894         sig->curr_target = tsk;
 895         init_sigpending(&sig->shared_pending);
 896         INIT_LIST_HEAD(&sig->posix_timers);
 897
 898         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 899         sig->real_timer.function = it_real_fn;
 900
 901         task_lock(current->group_leader);
 902         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
 903         task_unlock(current->group_leader);
 904
 905         posix_cpu_timers_init_group(sig);
 906
 907         tty_audit_fork(sig);
 908
 909         sig->oom_adj = current->signal->oom_adj;
 910         sig->oom_score_adj = current->signal->oom_score_adj;
 911
 912         mutex_init(&sig->cred_guard_mutex);
 913
 914         return 0;
 915 }
 916
 917 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
 918 {
 919         unsigned long new_flags = p->flags;
 920
 921         new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
 922         new_flags |= PF_FORKNOEXEC;
 923         new_flags |= PF_STARTING;
 924         p->flags = new_flags;
 925         clear_freeze_flag(p);
 926 }
 927
 928 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
 929 {
 930         current->clear_child_tid = tidptr;
 931
 932         return task_pid_vnr(current);
 933 }
 934
 935 static void rt_mutex_init_task(struct task_struct *p)
 936 {
 937         raw_spin_lock_init(&p->pi_lock);
 938 #ifdef CONFIG_RT_MUTEXES
 939         plist_head_init_raw(&p->pi_waiters, &p->pi_lock);
 940         p->pi_blocked_on = NULL;
 941 #endif
 942 }
 943
 944 #ifdef CONFIG_MM_OWNER
 945 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 946 {
 947         mm->owner = p;
 948 }
 949 #endif /* CONFIG_MM_OWNER */
 950
 951 /*
 952  * Initialize POSIX timer handling for a single task.
 953  */
 954 static void posix_cpu_timers_init(struct task_struct *tsk)
 955 {
 956         tsk->cputime_expires.prof_exp = cputime_zero;
 957         tsk->cputime_expires.virt_exp = cputime_zero;
 958         tsk->cputime_expires.sched_exp = 0;
 959         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
 960         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
 961         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
 962 }
 963
 964 /*
 965  * This creates a new process as a copy of the old one,
 966  * but does not actually start it yet.
 967  *
 968  * It copies the registers, and all the appropriate
 969  * parts of the process environment (as per the clone
 970  * flags). The actual kick-off is left to the caller.
 971  */
 972 static struct task_struct *copy_process(unsigned long clone_flags,
 973                                         unsigned long stack_start,
 974                                         struct pt_regs *regs,
 975                                         unsigned long stack_size,
 976                                         int __user *child_tidptr,
 977                                         struct pid *pid,
 978                                         int trace)
 979 {
 980         int retval;
 981         struct task_struct *p;
 982         int cgroup_callbacks_done = 0;
 983
 984         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 985                 return ERR_PTR(-EINVAL);
 986
 987         /*
 988          * Thread groups must share signals as well, and detached threads
 989          * can only be started up within the thread group.
 990          */
 991         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
 992                 return ERR_PTR(-EINVAL);
 993
 994         /*
 995          * Shared signal handlers imply shared VM. By way of the above,
 996          * thread groups also imply shared VM. Blocking this case allows
 997          * for various simplifications in other code.
 998          */
 999         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1000                 return ERR_PTR(-EINVAL);
1001
1002         /*
1003          * Siblings of global init remain as zombies on exit since they are
1004          * not reaped by their parent (swapper). To solve this and to avoid
1005          * multi-rooted process trees, prevent global and container-inits
1006          * from creating siblings.
1007          */
1008         if ((clone_flags & CLONE_PARENT) &&
1009                                 current->signal->flags & SIGNAL_UNKILLABLE)
1010                 return ERR_PTR(-EINVAL);
1011
1012         retval = security_task_create(clone_flags);
1013         if (retval)
1014                 goto fork_out;
1015
1016         retval = -ENOMEM;
1017         p = dup_task_struct(current);
1018         if (!p)
1019                 goto fork_out;
1020
1021         ftrace_graph_init_task(p);
1022
1023         rt_mutex_init_task(p);
1024
1025 #ifdef CONFIG_PROVE_LOCKING
1026         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1027         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1028 #endif
1029         retval = -EAGAIN;
1030         if (atomic_read(&p->real_cred->user->processes) >=
1031                         task_rlimit(p, RLIMIT_NPROC)) {
1032                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1033                     p->real_cred->user != INIT_USER)
1034                         goto bad_fork_free;
1035         }
1036
1037         retval = copy_creds(p, clone_flags);
1038         if (retval < 0)
1039                 goto bad_fork_free;
1040
1041         /*
1042          * If multiple threads are within copy_process(), then this check
1043          * triggers too late. This doesn't hurt, the check is only there
1044          * to stop root fork bombs.
1045          */
1046         retval = -EAGAIN;
1047         if (nr_threads >= max_threads)
1048                 goto bad_fork_cleanup_count;
1049
1050         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1051                 goto bad_fork_cleanup_count;
1052
1053         p->did_exec = 0;
1054         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1055         copy_flags(clone_flags, p);
1056         INIT_LIST_HEAD(&p->children);
1057         INIT_LIST_HEAD(&p->sibling);
1058         rcu_copy_process(p);
1059         p->vfork_done = NULL;
1060         spin_lock_init(&p->alloc_lock);
1061
1062         init_sigpending(&p->pending);
1063
1064         p->utime = cputime_zero;
1065         p->stime = cputime_zero;
1066         p->gtime = cputime_zero;
1067         p->utimescaled = cputime_zero;
1068         p->stimescaled = cputime_zero;
1069 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1070         p->prev_utime = cputime_zero;
1071         p->prev_stime = cputime_zero;
1072 #endif
1073 #if defined(SPLIT_RSS_COUNTING)
1074         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1075 #endif
1076
1077         p->default_timer_slack_ns = current->timer_slack_ns;
1078
1079         task_io_accounting_init(&p->ioac);
1080         acct_clear_integrals(p);
1081
1082         posix_cpu_timers_init(p);
1083
1084         p->lock_depth = -1;             /* -1 = no lock */
1085         do_posix_clock_monotonic_gettime(&p->start_time);
1086         p->real_start_time = p->start_time;
1087         monotonic_to_bootbased(&p->real_start_time);
1088         p->io_context = NULL;
1089         p->audit_context = NULL;
1090         cgroup_fork(p);
1091 #ifdef CONFIG_NUMA
1092         p->mempolicy = mpol_dup(p->mempolicy);
1093         if (IS_ERR(p->mempolicy)) {
1094                 retval = PTR_ERR(p->mempolicy);
1095                 p->mempolicy = NULL;
1096                 goto bad_fork_cleanup_cgroup;
1097         }
1098         mpol_fix_fork_child_flag(p);
1099 #endif
1100 #ifdef CONFIG_TRACE_IRQFLAGS
1101         p->irq_events = 0;
1102 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1103         p->hardirqs_enabled = 1;
1104 #else
1105         p->hardirqs_enabled = 0;
1106 #endif
1107         p->hardirq_enable_ip = 0;
1108         p->hardirq_enable_event = 0;
1109         p->hardirq_disable_ip = _THIS_IP_;
1110         p->hardirq_disable_event = 0;
1111         p->softirqs_enabled = 1;
1112         p->softirq_enable_ip = _THIS_IP_;
1113         p->softirq_enable_event = 0;
1114         p->softirq_disable_ip = 0;
1115         p->softirq_disable_event = 0;
1116         p->hardirq_context = 0;
1117         p->softirq_context = 0;
1118 #endif
1119 #ifdef CONFIG_LOCKDEP
1120         p->lockdep_depth = 0; /* no locks held yet */
1121         p->curr_chain_key = 0;
1122         p->lockdep_recursion = 0;
1123 #endif
1124
1125 #ifdef CONFIG_DEBUG_MUTEXES
1126         p->blocked_on = NULL; /* not blocked yet */
1127 #endif
1128 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1129         p->memcg_batch.do_batch = 0;
1130         p->memcg_batch.memcg = NULL;
1131 #endif
1132
1133         /* Perform scheduler related setup. Assign this task to a CPU. */
1134         sched_fork(p, clone_flags);
1135
1136         retval = perf_event_init_task(p);
1137         if (retval)
1138                 goto bad_fork_cleanup_policy;
1139
1140         if ((retval = audit_alloc(p)))
1141                 goto bad_fork_cleanup_policy;
1142         /* copy all the process information */
1143         if ((retval = copy_semundo(clone_flags, p)))
1144                 goto bad_fork_cleanup_audit;
1145         if ((retval = copy_files(clone_flags, p)))
1146                 goto bad_fork_cleanup_semundo;
1147         if ((retval = copy_fs(clone_flags, p)))
1148                 goto bad_fork_cleanup_files;
1149         if ((retval = copy_sighand(clone_flags, p)))
1150                 goto bad_fork_cleanup_fs;
1151         if ((retval = copy_signal(clone_flags, p)))
1152                 goto bad_fork_cleanup_sighand;
1153         if ((retval = copy_mm(clone_flags, p)))
1154                 goto bad_fork_cleanup_signal;
1155         if ((retval = copy_namespaces(clone_flags, p)))
1156                 goto bad_fork_cleanup_mm;
1157         if ((retval = copy_io(clone_flags, p)))
1158                 goto bad_fork_cleanup_namespaces;
1159         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1160         if (retval)
1161                 goto bad_fork_cleanup_io;
1162
1163         if (pid != &init_struct_pid) {
1164                 retval = -ENOMEM;
1165                 pid = alloc_pid(p->nsproxy->pid_ns);
1166                 if (!pid)
1167                         goto bad_fork_cleanup_io;
1168
1169                 if (clone_flags & CLONE_NEWPID) {
1170                         retval = pid_ns_prepare_proc(p->nsproxy->pid_ns);
1171                         if (retval < 0)
1172                                 goto bad_fork_free_pid;
1173                 }
1174         }
1175
1176         p->pid = pid_nr(pid);
1177         p->tgid = p->pid;
1178         if (clone_flags & CLONE_THREAD)
1179                 p->tgid = current->tgid;
1180
1181         if (current->nsproxy != p->nsproxy) {
1182                 retval = ns_cgroup_clone(p, pid);
1183                 if (retval)
1184                         goto bad_fork_free_pid;
1185         }
1186
1187         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1188         /*
1189          * Clear TID on mm_release()?
1190          */
1191         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1192 #ifdef CONFIG_FUTEX
1193         p->robust_list = NULL;
1194 #ifdef CONFIG_COMPAT
1195         p->compat_robust_list = NULL;
1196 #endif
1197         INIT_LIST_HEAD(&p->pi_state_list);
1198         p->pi_state_cache = NULL;
1199 #endif
1200         /*
1201          * sigaltstack should be cleared when sharing the same VM
1202          */
1203         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1204                 p->sas_ss_sp = p->sas_ss_size = 0;
1205
1206         /*
1207          * Syscall tracing and stepping should be turned off in the
1208          * child regardless of CLONE_PTRACE.
1209          */
1210         user_disable_single_step(p);
1211         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1212 #ifdef TIF_SYSCALL_EMU
1213         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1214 #endif
1215         clear_all_latency_tracing(p);
1216
1217         /* ok, now we should be set up.. */
1218         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1219         p->pdeath_signal = 0;
1220         p->exit_state = 0;
1221
1222         /*
1223          * Ok, make it visible to the rest of the system.
1224          * We dont wake it up yet.
1225          */
1226         p->group_leader = p;
1227         INIT_LIST_HEAD(&p->thread_group);
1228
1229         /* Now that the task is set up, run cgroup callbacks if
1230          * necessary. We need to run them before the task is visible
1231          * on the tasklist. */
1232         cgroup_fork_callbacks(p);
1233         cgroup_callbacks_done = 1;
1234
1235         /* Need tasklist lock for parent etc handling! */
1236         write_lock_irq(&tasklist_lock);
1237
1238         /* CLONE_PARENT re-uses the old parent */
1239         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1240                 p->real_parent = current->real_parent;
1241                 p->parent_exec_id = current->parent_exec_id;
1242         } else {
1243                 p->real_parent = current;
1244                 p->parent_exec_id = current->self_exec_id;
1245         }
1246
1247         spin_lock(&current->sighand->siglock);
1248
1249         /*
1250          * Process group and session signals need to be delivered to just the
1251          * parent before the fork or both the parent and the child after the
1252          * fork. Restart if a signal comes in before we add the new process to
1253          * it's process group.
1254          * A fatal signal pending means that current will exit, so the new
1255          * thread can't slip out of an OOM kill (or normal SIGKILL).
1256          */
1257         recalc_sigpending();
1258         if (signal_pending(current)) {
1259                 spin_unlock(&current->sighand->siglock);
1260                 write_unlock_irq(&tasklist_lock);
1261                 retval = -ERESTARTNOINTR;
1262                 goto bad_fork_free_pid;
1263         }
1264
1265         if (clone_flags & CLONE_THREAD) {
1266                 current->signal->nr_threads++;
1267                 atomic_inc(&current->signal->live);
1268                 atomic_inc(&current->signal->sigcnt);
1269                 p->group_leader = current->group_leader;
1270                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1271         }
1272
1273         if (likely(p->pid)) {
1274                 tracehook_finish_clone(p, clone_flags, trace);
1275
1276                 if (thread_group_leader(p)) {
1277                         if (clone_flags & CLONE_NEWPID)
1278                                 p->nsproxy->pid_ns->child_reaper = p;
1279
1280                         p->signal->leader_pid = pid;
1281                         p->signal->tty = tty_kref_get(current->signal->tty);
1282                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1283                         attach_pid(p, PIDTYPE_SID, task_session(current));
1284                         list_add_tail(&p->sibling, &p->real_parent->children);
1285                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1286                         __get_cpu_var(process_counts)++;
1287                 }
1288                 attach_pid(p, PIDTYPE_PID, pid);
1289                 nr_threads++;
1290         }
1291
1292         total_forks++;
1293         spin_unlock(&current->sighand->siglock);
1294         write_unlock_irq(&tasklist_lock);
1295         proc_fork_connector(p);
1296         cgroup_post_fork(p);
1297         perf_event_fork(p);
1298         return p;
1299
1300 bad_fork_free_pid:
1301         if (pid != &init_struct_pid)
1302                 free_pid(pid);
1303 bad_fork_cleanup_io:
1304         if (p->io_context)
1305                 exit_io_context(p);
1306 bad_fork_cleanup_namespaces:
1307         exit_task_namespaces(p);
1308 bad_fork_cleanup_mm:
1309         if (p->mm) {
1310                 task_lock(p);
1311                 if (p->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1312                         atomic_dec(&p->mm->oom_disable_count);
1313                 task_unlock(p);
1314                 mmput(p->mm);
1315         }
1316 bad_fork_cleanup_signal:
1317         if (!(clone_flags & CLONE_THREAD))
1318                 free_signal_struct(p->signal);
1319 bad_fork_cleanup_sighand:
1320         __cleanup_sighand(p->sighand);
1321 bad_fork_cleanup_fs:
1322         exit_fs(p); /* blocking */
1323 bad_fork_cleanup_files:
1324         exit_files(p); /* blocking */
1325 bad_fork_cleanup_semundo:
1326         exit_sem(p);
1327 bad_fork_cleanup_audit:
1328         audit_free(p);
1329 bad_fork_cleanup_policy:
1330         perf_event_free_task(p);
1331 #ifdef CONFIG_NUMA
1332         mpol_put(p->mempolicy);
1333 bad_fork_cleanup_cgroup:
1334 #endif
1335         cgroup_exit(p, cgroup_callbacks_done);
1336         delayacct_tsk_free(p);
1337         module_put(task_thread_info(p)->exec_domain->module);
1338 bad_fork_cleanup_count:
1339         atomic_dec(&p->cred->user->processes);
1340         exit_creds(p);
1341 bad_fork_free:
1342         free_task(p);
1343 fork_out:
1344         return ERR_PTR(retval);
1345 }
1346
1347 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1348 {
1349         memset(regs, 0, sizeof(struct pt_regs));
1350         return regs;
1351 }
1352
1353 static inline void init_idle_pids(struct pid_link *links)
1354 {
1355         enum pid_type type;
1356
1357         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1358                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1359                 links[type].pid = &init_struct_pid;
1360         }
1361 }
1362
1363 struct task_struct * __cpuinit fork_idle(int cpu)
1364 {
1365         struct task_struct *task;
1366         struct pt_regs regs;
1367
1368         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1369                             &init_struct_pid, 0);
1370         if (!IS_ERR(task)) {
1371                 init_idle_pids(task->pids);
1372                 init_idle(task, cpu);
1373         }
1374
1375         return task;
1376 }
1377
1378 /*
1379  *  Ok, this is the main fork-routine.
1380  *
1381  * It copies the process, and if successful kick-starts
1382  * it and waits for it to finish using the VM if required.
1383  */
1384 long do_fork(unsigned long clone_flags,
1385               unsigned long stack_start,
1386               struct pt_regs *regs,
1387               unsigned long stack_size,
1388               int __user *parent_tidptr,
1389               int __user *child_tidptr)
1390 {
1391         struct task_struct *p;
1392         int trace = 0;
1393         long nr;
1394
1395         /*
1396          * Do some preliminary argument and permissions checking before we
1397          * actually start allocating stuff
1398          */
1399         if (clone_flags & CLONE_NEWUSER) {
1400                 if (clone_flags & CLONE_THREAD)
1401                         return -EINVAL;
1402                 /* hopefully this check will go away when userns support is
1403                  * complete
1404                  */
1405                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1406                                 !capable(CAP_SETGID))
1407                         return -EPERM;
1408         }
1409
1410         /*
1411          * We hope to recycle these flags after 2.6.26
1412          */
1413         if (unlikely(clone_flags & CLONE_STOPPED)) {
1414                 static int __read_mostly count = 100;
1415
1416                 if (count > 0 && printk_ratelimit()) {
1417                         char comm[TASK_COMM_LEN];
1418
1419                         count--;
1420                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1421                                         "clone flags 0x%lx\n",
1422                                 get_task_comm(comm, current),
1423                                 clone_flags & CLONE_STOPPED);
1424                 }
1425         }
1426
1427         /*
1428          * When called from kernel_thread, don't do user tracing stuff.
1429          */
1430         if (likely(user_mode(regs)))
1431                 trace = tracehook_prepare_clone(clone_flags);
1432
1433         p = copy_process(clone_flags, stack_start, regs, stack_size,
1434                          child_tidptr, NULL, trace);
1435         /*
1436          * Do this prior waking up the new thread - the thread pointer
1437          * might get invalid after that point, if the thread exits quickly.
1438          */
1439         if (!IS_ERR(p)) {
1440                 struct completion vfork;
1441
1442                 trace_sched_process_fork(current, p);
1443
1444                 nr = task_pid_vnr(p);
1445
1446                 if (clone_flags & CLONE_PARENT_SETTID)
1447                         put_user(nr, parent_tidptr);
1448
1449                 if (clone_flags & CLONE_VFORK) {
1450                         p->vfork_done = &vfork;
1451                         init_completion(&vfork);
1452                 }
1453
1454                 audit_finish_fork(p);
1455                 tracehook_report_clone(regs, clone_flags, nr, p);
1456
1457                 /*
1458                  * We set PF_STARTING at creation in case tracing wants to
1459                  * use this to distinguish a fully live task from one that
1460                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1461                  * clear it and set the child going.
1462                  */
1463                 p->flags &= ~PF_STARTING;
1464
1465                 if (unlikely(clone_flags & CLONE_STOPPED)) {
1466                         /*
1467                          * We'll start up with an immediate SIGSTOP.
1468                          */
1469                         sigaddset(&p->pending.signal, SIGSTOP);
1470                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1471                         __set_task_state(p, TASK_STOPPED);
1472                 } else {
1473                         wake_up_new_task(p, clone_flags);
1474                 }
1475
1476                 tracehook_report_clone_complete(trace, regs,
1477                                                 clone_flags, nr, p);
1478
1479                 if (clone_flags & CLONE_VFORK) {
1480                         freezer_do_not_count();
1481                         wait_for_completion(&vfork);
1482                         freezer_count();
1483                         tracehook_report_vfork_done(p, nr);
1484                 }
1485         } else {
1486                 nr = PTR_ERR(p);
1487         }
1488         return nr;
1489 }
1490
1491 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1492 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1493 #endif
1494
1495 static void sighand_ctor(void *data)
1496 {
1497         struct sighand_struct *sighand = data;
1498
1499         spin_lock_init(&sighand->siglock);
1500         init_waitqueue_head(&sighand->signalfd_wqh);
1501 }
1502
1503 void __init proc_caches_init(void)
1504 {
1505         sighand_cachep = kmem_cache_create("sighand_cache",
1506                         sizeof(struct sighand_struct), 0,
1507                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1508                         SLAB_NOTRACK, sighand_ctor);
1509         signal_cachep = kmem_cache_create("signal_cache",
1510                         sizeof(struct signal_struct), 0,
1511                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1512         files_cachep = kmem_cache_create("files_cache",
1513                         sizeof(struct files_struct), 0,
1514                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1515         fs_cachep = kmem_cache_create("fs_cache",
1516                         sizeof(struct fs_struct), 0,
1517                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1518         mm_cachep = kmem_cache_create("mm_struct",
1519                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1520                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1521         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1522         mmap_init();
1523 }
1524
1525 /*
1526  * Check constraints on flags passed to the unshare system call and
1527  * force unsharing of additional process context as appropriate.
1528  */
1529 static void check_unshare_flags(unsigned long *flags_ptr)
1530 {
1531         /*
1532          * If unsharing a thread from a thread group, must also
1533          * unshare vm.
1534          */
1535         if (*flags_ptr & CLONE_THREAD)
1536                 *flags_ptr |= CLONE_VM;
1537
1538         /*
1539          * If unsharing vm, must also unshare signal handlers.
1540          */
1541         if (*flags_ptr & CLONE_VM)
1542                 *flags_ptr |= CLONE_SIGHAND;
1543
1544         /*
1545          * If unsharing namespace, must also unshare filesystem information.
1546          */
1547         if (*flags_ptr & CLONE_NEWNS)
1548                 *flags_ptr |= CLONE_FS;
1549 }
1550
1551 /*
1552  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1553  */
1554 static int unshare_thread(unsigned long unshare_flags)
1555 {
1556         if (unshare_flags & CLONE_THREAD)
1557                 return -EINVAL;
1558
1559         return 0;
1560 }
1561
1562 /*
1563  * Unshare the filesystem structure if it is being shared
1564  */
1565 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1566 {
1567         struct fs_struct *fs = current->fs;
1568
1569         if (!(unshare_flags & CLONE_FS) || !fs)
1570                 return 0;
1571
1572         /* don't need lock here; in the worst case we'll do useless copy */
1573         if (fs->users == 1)
1574                 return 0;
1575
1576         *new_fsp = copy_fs_struct(fs);
1577         if (!*new_fsp)
1578                 return -ENOMEM;
1579
1580         return 0;
1581 }
1582
1583 /*
1584  * Unsharing of sighand is not supported yet
1585  */
1586 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1587 {
1588         struct sighand_struct *sigh = current->sighand;
1589
1590         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1591                 return -EINVAL;
1592         else
1593                 return 0;
1594 }
1595
1596 /*
1597  * Unshare vm if it is being shared
1598  */
1599 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1600 {
1601         struct mm_struct *mm = current->mm;
1602
1603         if ((unshare_flags & CLONE_VM) &&
1604             (mm && atomic_read(&mm->mm_users) > 1)) {
1605                 return -EINVAL;
1606         }
1607
1608         return 0;
1609 }
1610
1611 /*
1612  * Unshare file descriptor table if it is being shared
1613  */
1614 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1615 {
1616         struct files_struct *fd = current->files;
1617         int error = 0;
1618
1619         if ((unshare_flags & CLONE_FILES) &&
1620             (fd && atomic_read(&fd->count) > 1)) {
1621                 *new_fdp = dup_fd(fd, &error);
1622                 if (!*new_fdp)
1623                         return error;
1624         }
1625
1626         return 0;
1627 }
1628
1629 /*
1630  * unshare allows a process to 'unshare' part of the process
1631  * context which was originally shared using clone.  copy_*
1632  * functions used by do_fork() cannot be used here directly
1633  * because they modify an inactive task_struct that is being
1634  * constructed. Here we are modifying the current, active,
1635  * task_struct.
1636  */
1637 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1638 {
1639         int err = 0;
1640         struct fs_struct *fs, *new_fs = NULL;
1641         struct sighand_struct *new_sigh = NULL;
1642         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1643         struct files_struct *fd, *new_fd = NULL;
1644         struct nsproxy *new_nsproxy = NULL;
1645         int do_sysvsem = 0;
1646
1647         check_unshare_flags(&unshare_flags);
1648
1649         /* Return -EINVAL for all unsupported flags */
1650         err = -EINVAL;
1651         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1652                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1653                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1654                 goto bad_unshare_out;
1655
1656         /*
1657          * CLONE_NEWIPC must also detach from the undolist: after switching
1658          * to a new ipc namespace, the semaphore arrays from the old
1659          * namespace are unreachable.
1660          */
1661         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1662                 do_sysvsem = 1;
1663         if ((err = unshare_thread(unshare_flags)))
1664                 goto bad_unshare_out;
1665         if ((err = unshare_fs(unshare_flags, &new_fs)))
1666                 goto bad_unshare_cleanup_thread;
1667         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1668                 goto bad_unshare_cleanup_fs;
1669         if ((err = unshare_vm(unshare_flags, &new_mm)))
1670                 goto bad_unshare_cleanup_sigh;
1671         if ((err = unshare_fd(unshare_flags, &new_fd)))
1672                 goto bad_unshare_cleanup_vm;
1673         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1674                         new_fs)))
1675                 goto bad_unshare_cleanup_fd;
1676
1677         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1678                 if (do_sysvsem) {
1679                         /*
1680                          * CLONE_SYSVSEM is equivalent to sys_exit().
1681                          */
1682                         exit_sem(current);
1683                 }
1684
1685                 if (new_nsproxy) {
1686                         switch_task_namespaces(current, new_nsproxy);
1687                         new_nsproxy = NULL;
1688                 }
1689
1690                 task_lock(current);
1691
1692                 if (new_fs) {
1693                         fs = current->fs;
1694                         spin_lock(&fs->lock);
1695                         current->fs = new_fs;
1696                         if (--fs->users)
1697                                 new_fs = NULL;
1698                         else
1699                                 new_fs = fs;
1700                         spin_unlock(&fs->lock);
1701                 }
1702
1703                 if (new_mm) {
1704                         mm = current->mm;
1705                         active_mm = current->active_mm;
1706                         current->mm = new_mm;
1707                         current->active_mm = new_mm;
1708                         if (current->signal->oom_score_adj == OOM_SCORE_ADJ_MIN) {
1709                                 atomic_dec(&mm->oom_disable_count);
1710                                 atomic_inc(&new_mm->oom_disable_count);
1711                         }
1712                         activate_mm(active_mm, new_mm);
1713                         new_mm = mm;
1714                 }
1715
1716                 if (new_fd) {
1717                         fd = current->files;
1718                         current->files = new_fd;
1719                         new_fd = fd;
1720                 }
1721
1722                 task_unlock(current);
1723         }
1724
1725         if (new_nsproxy)
1726                 put_nsproxy(new_nsproxy);
1727
1728 bad_unshare_cleanup_fd:
1729         if (new_fd)
1730                 put_files_struct(new_fd);
1731
1732 bad_unshare_cleanup_vm:
1733         if (new_mm)
1734                 mmput(new_mm);
1735
1736 bad_unshare_cleanup_sigh:
1737         if (new_sigh)
1738                 if (atomic_dec_and_test(&new_sigh->count))
1739                         kmem_cache_free(sighand_cachep, new_sigh);
1740
1741 bad_unshare_cleanup_fs:
1742         if (new_fs)
1743                 free_fs_struct(new_fs);
1744
1745 bad_unshare_cleanup_thread:
1746 bad_unshare_out:
1747         return err;
1748 }
1749
1750 /*
1751  *      Helper to unshare the files of the current task.
1752  *      We don't want to expose copy_files internals to
1753  *      the exec layer of the kernel.
1754  */
1755
1756 int unshare_files(struct files_struct **displaced)
1757 {
1758         struct task_struct *task = current;
1759         struct files_struct *copy = NULL;
1760         int error;
1761
1762         error = unshare_fd(CLONE_FILES, &copy);
1763         if (error || !copy) {
1764                 *displaced = NULL;
1765                 return error;
1766         }
1767         *displaced = task->files;
1768         task_lock(task);
1769         task->files = copy;
1770         task_unlock(task);
1771         return 0;
1772 }