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