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