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