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