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/seccomp.h>
  38 #include <linux/swap.h>
  39 #include <linux/syscalls.h>
  40 #include <linux/jiffies.h>
  41 #include <linux/futex.h>
  42 #include <linux/compat.h>
  43 #include <linux/kthread.h>
  44 #include <linux/task_io_accounting_ops.h>
  45 #include <linux/rcupdate.h>
  46 #include <linux/ptrace.h>
  47 #include <linux/mount.h>
  48 #include <linux/audit.h>
  49 #include <linux/memcontrol.h>
  50 #include <linux/ftrace.h>
  51 #include <linux/proc_fs.h>
  52 #include <linux/profile.h>
  53 #include <linux/rmap.h>
  54 #include <linux/ksm.h>
  55 #include <linux/acct.h>
  56 #include <linux/tsacct_kern.h>
  57 #include <linux/cn_proc.h>
  58 #include <linux/freezer.h>
  59 #include <linux/delayacct.h>
  60 #include <linux/taskstats_kern.h>
  61 #include <linux/random.h>
  62 #include <linux/tty.h>
  63 #include <linux/blkdev.h>
  64 #include <linux/fs_struct.h>
  65 #include <linux/magic.h>
  66 #include <linux/perf_event.h>
  67 #include <linux/posix-timers.h>
  68 #include <linux/user-return-notifier.h>
  69 #include <linux/oom.h>
  70 #include <linux/khugepaged.h>
  71 #include <linux/signalfd.h>
  72
  73 #include <asm/pgtable.h>
  74 #include <asm/pgalloc.h>
  75 #include <asm/uaccess.h>
  76 #include <asm/mmu_context.h>
  77 #include <asm/cacheflush.h>
  78 #include <asm/tlbflush.h>
  79
  80 #include <trace/events/sched.h>
  81
  82 #define CREATE_TRACE_POINTS
  83 #include <trace/events/task.h>
  84
  85 /*
  86  * Protected counters by write_lock_irq(&tasklist_lock)
  87  */
  88 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  89 int nr_threads;                 /* The idle threads do not count.. */
  90
  91 int max_threads;                /* tunable limit on nr_threads */
  92
  93 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  94
  95 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  96
  97 #ifdef CONFIG_PROVE_RCU
  98 int lockdep_tasklist_lock_is_held(void)
  99 {
 100         return lockdep_is_held(&tasklist_lock);
 101 }
 102 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 103 #endif /* #ifdef CONFIG_PROVE_RCU */
 104
 105 int nr_processes(void)
 106 {
 107         int cpu;
 108         int total = 0;
 109
 110         for_each_possible_cpu(cpu)
 111                 total += per_cpu(process_counts, cpu);
 112
 113         return total;
 114 }
 115
 116 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 117 # define alloc_task_struct_node(node)           \
 118                 kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node)
 119 # define free_task_struct(tsk)                  \
 120                 kmem_cache_free(task_struct_cachep, (tsk))
 121 static struct kmem_cache *task_struct_cachep;
 122 #endif
 123
 124 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
 125 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 126                                                   int node)
 127 {
 128 #ifdef CONFIG_DEBUG_STACK_USAGE
 129         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
 130 #else
 131         gfp_t mask = GFP_KERNEL;
 132 #endif
 133         struct page *page = alloc_pages_node(node, mask, THREAD_SIZE_ORDER);
 134
 135         return page ? page_address(page) : NULL;
 136 }
 137
 138 static inline void free_thread_info(struct thread_info *ti)
 139 {
 140         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 141 }
 142 #endif
 143
 144 /* SLAB cache for signal_struct structures (tsk->signal) */
 145 static struct kmem_cache *signal_cachep;
 146
 147 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 148 struct kmem_cache *sighand_cachep;
 149
 150 /* SLAB cache for files_struct structures (tsk->files) */
 151 struct kmem_cache *files_cachep;
 152
 153 /* SLAB cache for fs_struct structures (tsk->fs) */
 154 struct kmem_cache *fs_cachep;
 155
 156 /* SLAB cache for vm_area_struct structures */
 157 struct kmem_cache *vm_area_cachep;
 158
 159 /* SLAB cache for mm_struct structures (tsk->mm) */
 160 static struct kmem_cache *mm_cachep;
 161
 162 static void account_kernel_stack(struct thread_info *ti, int account)
 163 {
 164         struct zone *zone = page_zone(virt_to_page(ti));
 165
 166         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
 167 }
 168
 169 void free_task(struct task_struct *tsk)
 170 {
 171         account_kernel_stack(tsk->stack, -1);
 172         free_thread_info(tsk->stack);
 173         rt_mutex_debug_task_free(tsk);
 174         ftrace_graph_exit_task(tsk);
 175         put_seccomp_filter(tsk);
 176         free_task_struct(tsk);
 177 }
 178 EXPORT_SYMBOL(free_task);
 179
 180 static inline void free_signal_struct(struct signal_struct *sig)
 181 {
 182         taskstats_tgid_free(sig);
 183         sched_autogroup_exit(sig);
 184         kmem_cache_free(signal_cachep, sig);
 185 }
 186
 187 static inline void put_signal_struct(struct signal_struct *sig)
 188 {
 189         if (atomic_dec_and_test(&sig->sigcnt))
 190                 free_signal_struct(sig);
 191 }
 192
 193 void __put_task_struct(struct task_struct *tsk)
 194 {
 195         WARN_ON(!tsk->exit_state);
 196         WARN_ON(atomic_read(&tsk->usage));
 197         WARN_ON(tsk == current);
 198
 199         security_task_free(tsk);
 200         exit_creds(tsk);
 201         delayacct_tsk_free(tsk);
 202         put_signal_struct(tsk->signal);
 203
 204         if (!profile_handoff_task(tsk))
 205                 free_task(tsk);
 206 }
 207 EXPORT_SYMBOL_GPL(__put_task_struct);
 208
 209 /*
 210  * macro override instead of weak attribute alias, to workaround
 211  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
 212  */
 213 #ifndef arch_task_cache_init
 214 #define arch_task_cache_init()
 215 #endif
 216
 217 void __init fork_init(unsigned long mempages)
 218 {
 219 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 220 #ifndef ARCH_MIN_TASKALIGN
 221 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 222 #endif
 223         /* create a slab on which task_structs can be allocated */
 224         task_struct_cachep =
 225                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 226                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
 227 #endif
 228
 229         /* do the arch specific task caches init */
 230         arch_task_cache_init();
 231
 232         /*
 233          * The default maximum number of threads is set to a safe
 234          * value: the thread structures can take up at most half
 235          * of memory.
 236          */
 237         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 238
 239         /*
 240          * we need to allow at least 20 threads to boot a system
 241          */
 242         if (max_threads < 20)
 243                 max_threads = 20;
 244
 245         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 246         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 247         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 248                 init_task.signal->rlim[RLIMIT_NPROC];
 249 }
 250
 251 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
 252                                                struct task_struct *src)
 253 {
 254         *dst = *src;
 255         return 0;
 256 }
 257
 258 static struct task_struct *dup_task_struct(struct task_struct *orig)
 259 {
 260         struct task_struct *tsk;
 261         struct thread_info *ti;
 262         unsigned long *stackend;
 263         int node = tsk_fork_get_node(orig);
 264         int err;
 265
 266         prepare_to_copy(orig);
 267
 268         tsk = alloc_task_struct_node(node);
 269         if (!tsk)
 270                 return NULL;
 271
 272         ti = alloc_thread_info_node(tsk, node);
 273         if (!ti) {
 274                 free_task_struct(tsk);
 275                 return NULL;
 276         }
 277
 278         err = arch_dup_task_struct(tsk, orig);
 279         if (err)
 280                 goto out;
 281
 282         tsk->stack = ti;
 283
 284         setup_thread_stack(tsk, orig);
 285         clear_user_return_notifier(tsk);
 286         clear_tsk_need_resched(tsk);
 287         stackend = end_of_stack(tsk);
 288         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 289
 290 #ifdef CONFIG_CC_STACKPROTECTOR
 291         tsk->stack_canary = get_random_int();
 292 #endif
 293
 294         /*
 295          * One for us, one for whoever does the "release_task()" (usually
 296          * parent)
 297          */
 298         atomic_set(&tsk->usage, 2);
 299 #ifdef CONFIG_BLK_DEV_IO_TRACE
 300         tsk->btrace_seq = 0;
 301 #endif
 302         tsk->splice_pipe = NULL;
 303
 304         account_kernel_stack(ti, 1);
 305
 306         return tsk;
 307
 308 out:
 309         free_thread_info(ti);
 310         free_task_struct(tsk);
 311         return NULL;
 312 }
 313
 314 #ifdef CONFIG_MMU
 315 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 316 {
 317         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 318         struct rb_node **rb_link, *rb_parent;
 319         int retval;
 320         unsigned long charge;
 321         struct mempolicy *pol;
 322
 323         down_write(&oldmm->mmap_sem);
 324         flush_cache_dup_mm(oldmm);
 325         /*
 326          * Not linked in yet - no deadlock potential:
 327          */
 328         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 329
 330         mm->locked_vm = 0;
 331         mm->mmap = NULL;
 332         mm->mmap_cache = NULL;
 333         mm->free_area_cache = oldmm->mmap_base;
 334         mm->cached_hole_size = ~0UL;
 335         mm->map_count = 0;
 336         cpumask_clear(mm_cpumask(mm));
 337         mm->mm_rb = RB_ROOT;
 338         rb_link = &mm->mm_rb.rb_node;
 339         rb_parent = NULL;
 340         pprev = &mm->mmap;
 341         retval = ksm_fork(mm, oldmm);
 342         if (retval)
 343                 goto out;
 344         retval = khugepaged_fork(mm, oldmm);
 345         if (retval)
 346                 goto out;
 347
 348         prev = NULL;
 349         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 350                 struct file *file;
 351
 352                 if (mpnt->vm_flags & VM_DONTCOPY) {
 353                         long pages = vma_pages(mpnt);
 354                         mm->total_vm -= pages;
 355                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 356                                                                 -pages);
 357                         continue;
 358                 }
 359                 charge = 0;
 360                 if (mpnt->vm_flags & VM_ACCOUNT) {
 361                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 362                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 363                                 goto fail_nomem;
 364                         charge = len;
 365                 }
 366                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 367                 if (!tmp)
 368                         goto fail_nomem;
 369                 *tmp = *mpnt;
 370                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 371                 pol = mpol_dup(vma_policy(mpnt));
 372                 retval = PTR_ERR(pol);
 373                 if (IS_ERR(pol))
 374                         goto fail_nomem_policy;
 375                 vma_set_policy(tmp, pol);
 376                 tmp->vm_mm = mm;
 377                 if (anon_vma_fork(tmp, mpnt))
 378                         goto fail_nomem_anon_vma_fork;
 379                 tmp->vm_flags &= ~VM_LOCKED;
 380                 tmp->vm_next = tmp->vm_prev = NULL;
 381                 file = tmp->vm_file;
 382                 if (file) {
 383                         struct inode *inode = file->f_path.dentry->d_inode;
 384                         struct address_space *mapping = file->f_mapping;
 385
 386                         get_file(file);
 387                         if (tmp->vm_flags & VM_DENYWRITE)
 388                                 atomic_dec(&inode->i_writecount);
 389                         mutex_lock(&mapping->i_mmap_mutex);
 390                         if (tmp->vm_flags & VM_SHARED)
 391                                 mapping->i_mmap_writable++;
 392                         flush_dcache_mmap_lock(mapping);
 393                         /* insert tmp into the share list, just after mpnt */
 394                         vma_prio_tree_add(tmp, mpnt);
 395                         flush_dcache_mmap_unlock(mapping);
 396                         mutex_unlock(&mapping->i_mmap_mutex);
 397                 }
 398
 399                 /*
 400                  * Clear hugetlb-related page reserves for children. This only
 401                  * affects MAP_PRIVATE mappings. Faults generated by the child
 402                  * are not guaranteed to succeed, even if read-only
 403                  */
 404                 if (is_vm_hugetlb_page(tmp))
 405                         reset_vma_resv_huge_pages(tmp);
 406
 407                 /*
 408                  * Link in the new vma and copy the page table entries.
 409                  */
 410                 *pprev = tmp;
 411                 pprev = &tmp->vm_next;
 412                 tmp->vm_prev = prev;
 413                 prev = tmp;
 414
 415                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 416                 rb_link = &tmp->vm_rb.rb_right;
 417                 rb_parent = &tmp->vm_rb;
 418
 419                 mm->map_count++;
 420                 retval = copy_page_range(mm, oldmm, mpnt);
 421
 422                 if (tmp->vm_ops && tmp->vm_ops->open)
 423                         tmp->vm_ops->open(tmp);
 424
 425                 if (retval)
 426                         goto out;
 427         }
 428         /* a new mm has just been created */
 429         arch_dup_mmap(oldmm, mm);
 430         retval = 0;
 431 out:
 432         up_write(&mm->mmap_sem);
 433         flush_tlb_mm(oldmm);
 434         up_write(&oldmm->mmap_sem);
 435         return retval;
 436 fail_nomem_anon_vma_fork:
 437         mpol_put(pol);
 438 fail_nomem_policy:
 439         kmem_cache_free(vm_area_cachep, tmp);
 440 fail_nomem:
 441         retval = -ENOMEM;
 442         vm_unacct_memory(charge);
 443         goto out;
 444 }
 445
 446 static inline int mm_alloc_pgd(struct mm_struct *mm)
 447 {
 448         mm->pgd = pgd_alloc(mm);
 449         if (unlikely(!mm->pgd))
 450                 return -ENOMEM;
 451         return 0;
 452 }
 453
 454 static inline void mm_free_pgd(struct mm_struct *mm)
 455 {
 456         pgd_free(mm, mm->pgd);
 457 }
 458 #else
 459 #define dup_mmap(mm, oldmm)     (0)
 460 #define mm_alloc_pgd(mm)        (0)
 461 #define mm_free_pgd(mm)
 462 #endif /* CONFIG_MMU */
 463
 464 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 465
 466 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 467 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 468
 469 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 470
 471 static int __init coredump_filter_setup(char *s)
 472 {
 473         default_dump_filter =
 474                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 475                 MMF_DUMP_FILTER_MASK;
 476         return 1;
 477 }
 478
 479 __setup("coredump_filter=", coredump_filter_setup);
 480
 481 #include <linux/init_task.h>
 482
 483 static void mm_init_aio(struct mm_struct *mm)
 484 {
 485 #ifdef CONFIG_AIO
 486         spin_lock_init(&mm->ioctx_lock);
 487         INIT_HLIST_HEAD(&mm->ioctx_list);
 488 #endif
 489 }
 490
 491 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
 492 {
 493         atomic_set(&mm->mm_users, 1);
 494         atomic_set(&mm->mm_count, 1);
 495         init_rwsem(&mm->mmap_sem);
 496         INIT_LIST_HEAD(&mm->mmlist);
 497         mm->flags = (current->mm) ?
 498                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
 499         mm->core_state = NULL;
 500         mm->nr_ptes = 0;
 501         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 502         spin_lock_init(&mm->page_table_lock);
 503         mm->free_area_cache = TASK_UNMAPPED_BASE;
 504         mm->cached_hole_size = ~0UL;
 505         mm_init_aio(mm);
 506         mm_init_owner(mm, p);
 507
 508         if (likely(!mm_alloc_pgd(mm))) {
 509                 mm->def_flags = 0;
 510                 mmu_notifier_mm_init(mm);
 511                 return mm;
 512         }
 513
 514         free_mm(mm);
 515         return NULL;
 516 }
 517
 518 static void check_mm(struct mm_struct *mm)
 519 {
 520         int i;
 521
 522         for (i = 0; i < NR_MM_COUNTERS; i++) {
 523                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 524
 525                 if (unlikely(x))
 526                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 527                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 528         }
 529
 530 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 531         VM_BUG_ON(mm->pmd_huge_pte);
 532 #endif
 533 }
 534
 535 /*
 536  * Allocate and initialize an mm_struct.
 537  */
 538 struct mm_struct *mm_alloc(void)
 539 {
 540         struct mm_struct *mm;
 541
 542         mm = allocate_mm();
 543         if (!mm)
 544                 return NULL;
 545
 546         memset(mm, 0, sizeof(*mm));
 547         mm_init_cpumask(mm);
 548         return mm_init(mm, current);
 549 }
 550
 551 /*
 552  * Called when the last reference to the mm
 553  * is dropped: either by a lazy thread or by
 554  * mmput. Free the page directory and the mm.
 555  */
 556 void __mmdrop(struct mm_struct *mm)
 557 {
 558         BUG_ON(mm == &init_mm);
 559         mm_free_pgd(mm);
 560         destroy_context(mm);
 561         mmu_notifier_mm_destroy(mm);
 562         check_mm(mm);
 563         free_mm(mm);
 564 }
 565 EXPORT_SYMBOL_GPL(__mmdrop);
 566
 567 /*
 568  * Decrement the use count and release all resources for an mm.
 569  */
 570 void mmput(struct mm_struct *mm)
 571 {
 572         might_sleep();
 573
 574         if (atomic_dec_and_test(&mm->mm_users)) {
 575                 exit_aio(mm);
 576                 ksm_exit(mm);
 577                 khugepaged_exit(mm); /* must run before exit_mmap */
 578                 exit_mmap(mm);
 579                 set_mm_exe_file(mm, NULL);
 580                 if (!list_empty(&mm->mmlist)) {
 581                         spin_lock(&mmlist_lock);
 582                         list_del(&mm->mmlist);
 583                         spin_unlock(&mmlist_lock);
 584                 }
 585                 put_swap_token(mm);
 586                 if (mm->binfmt)
 587                         module_put(mm->binfmt->module);
 588                 mmdrop(mm);
 589         }
 590 }
 591 EXPORT_SYMBOL_GPL(mmput);
 592
 593 /*
 594  * We added or removed a vma mapping the executable. The vmas are only mapped
 595  * during exec and are not mapped with the mmap system call.
 596  * Callers must hold down_write() on the mm's mmap_sem for these
 597  */
 598 void added_exe_file_vma(struct mm_struct *mm)
 599 {
 600         mm->num_exe_file_vmas++;
 601 }
 602
 603 void removed_exe_file_vma(struct mm_struct *mm)
 604 {
 605         mm->num_exe_file_vmas--;
 606         if ((mm->num_exe_file_vmas == 0) && mm->exe_file) {
 607                 fput(mm->exe_file);
 608                 mm->exe_file = NULL;
 609         }
 610
 611 }
 612
 613 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 614 {
 615         if (new_exe_file)
 616                 get_file(new_exe_file);
 617         if (mm->exe_file)
 618                 fput(mm->exe_file);
 619         mm->exe_file = new_exe_file;
 620         mm->num_exe_file_vmas = 0;
 621 }
 622
 623 struct file *get_mm_exe_file(struct mm_struct *mm)
 624 {
 625         struct file *exe_file;
 626
 627         /* We need mmap_sem to protect against races with removal of
 628          * VM_EXECUTABLE vmas */
 629         down_read(&mm->mmap_sem);
 630         exe_file = mm->exe_file;
 631         if (exe_file)
 632                 get_file(exe_file);
 633         up_read(&mm->mmap_sem);
 634         return exe_file;
 635 }
 636
 637 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
 638 {
 639         /* It's safe to write the exe_file pointer without exe_file_lock because
 640          * this is called during fork when the task is not yet in /proc */
 641         newmm->exe_file = get_mm_exe_file(oldmm);
 642 }
 643
 644 /**
 645  * get_task_mm - acquire a reference to the task's mm
 646  *
 647  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 648  * this kernel workthread has transiently adopted a user mm with use_mm,
 649  * to do its AIO) is not set and if so returns a reference to it, after
 650  * bumping up the use count.  User must release the mm via mmput()
 651  * after use.  Typically used by /proc and ptrace.
 652  */
 653 struct mm_struct *get_task_mm(struct task_struct *task)
 654 {
 655         struct mm_struct *mm;
 656
 657         task_lock(task);
 658         mm = task->mm;
 659         if (mm) {
 660                 if (task->flags & PF_KTHREAD)
 661                         mm = NULL;
 662                 else
 663                         atomic_inc(&mm->mm_users);
 664         }
 665         task_unlock(task);
 666         return mm;
 667 }
 668 EXPORT_SYMBOL_GPL(get_task_mm);
 669
 670 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 671 {
 672         struct mm_struct *mm;
 673         int err;
 674
 675         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 676         if (err)
 677                 return ERR_PTR(err);
 678
 679         mm = get_task_mm(task);
 680         if (mm && mm != current->mm &&
 681                         !ptrace_may_access(task, mode)) {
 682                 mmput(mm);
 683                 mm = ERR_PTR(-EACCES);
 684         }
 685         mutex_unlock(&task->signal->cred_guard_mutex);
 686
 687         return mm;
 688 }
 689
 690 static void complete_vfork_done(struct task_struct *tsk)
 691 {
 692         struct completion *vfork;
 693
 694         task_lock(tsk);
 695         vfork = tsk->vfork_done;
 696         if (likely(vfork)) {
 697                 tsk->vfork_done = NULL;
 698                 complete(vfork);
 699         }
 700         task_unlock(tsk);
 701 }
 702
 703 static int wait_for_vfork_done(struct task_struct *child,
 704                                 struct completion *vfork)
 705 {
 706         int killed;
 707
 708         freezer_do_not_count();
 709         killed = wait_for_completion_killable(vfork);
 710         freezer_count();
 711
 712         if (killed) {
 713                 task_lock(child);
 714                 child->vfork_done = NULL;
 715                 task_unlock(child);
 716         }
 717
 718         put_task_struct(child);
 719         return killed;
 720 }
 721
 722 /* Please note the differences between mmput and mm_release.
 723  * mmput is called whenever we stop holding onto a mm_struct,
 724  * error success whatever.
 725  *
 726  * mm_release is called after a mm_struct has been removed
 727  * from the current process.
 728  *
 729  * This difference is important for error handling, when we
 730  * only half set up a mm_struct for a new process and need to restore
 731  * the old one.  Because we mmput the new mm_struct before
 732  * restoring the old one. . .
 733  * Eric Biederman 10 January 1998
 734  */
 735 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 736 {
 737         /* Get rid of any futexes when releasing the mm */
 738 #ifdef CONFIG_FUTEX
 739         if (unlikely(tsk->robust_list)) {
 740                 exit_robust_list(tsk);
 741                 tsk->robust_list = NULL;
 742         }
 743 #ifdef CONFIG_COMPAT
 744         if (unlikely(tsk->compat_robust_list)) {
 745                 compat_exit_robust_list(tsk);
 746                 tsk->compat_robust_list = NULL;
 747         }
 748 #endif
 749         if (unlikely(!list_empty(&tsk->pi_state_list)))
 750                 exit_pi_state_list(tsk);
 751 #endif
 752
 753         /* Get rid of any cached register state */
 754         deactivate_mm(tsk, mm);
 755
 756         if (tsk->vfork_done)
 757                 complete_vfork_done(tsk);
 758
 759         /*
 760          * If we're exiting normally, clear a user-space tid field if
 761          * requested.  We leave this alone when dying by signal, to leave
 762          * the value intact in a core dump, and to save the unnecessary
 763          * trouble, say, a killed vfork parent shouldn't touch this mm.
 764          * Userland only wants this done for a sys_exit.
 765          */
 766         if (tsk->clear_child_tid) {
 767                 if (!(tsk->flags & PF_SIGNALED) &&
 768                     atomic_read(&mm->mm_users) > 1) {
 769                         /*
 770                          * We don't check the error code - if userspace has
 771                          * not set up a proper pointer then tough luck.
 772                          */
 773                         put_user(0, tsk->clear_child_tid);
 774                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 775                                         1, NULL, NULL, 0);
 776                 }
 777                 tsk->clear_child_tid = NULL;
 778         }
 779 }
 780
 781 /*
 782  * Allocate a new mm structure and copy contents from the
 783  * mm structure of the passed in task structure.
 784  */
 785 struct mm_struct *dup_mm(struct task_struct *tsk)
 786 {
 787         struct mm_struct *mm, *oldmm = current->mm;
 788         int err;
 789
 790         if (!oldmm)
 791                 return NULL;
 792
 793         mm = allocate_mm();
 794         if (!mm)
 795                 goto fail_nomem;
 796
 797         memcpy(mm, oldmm, sizeof(*mm));
 798         mm_init_cpumask(mm);
 799
 800         /* Initializing for Swap token stuff */
 801         mm->token_priority = 0;
 802         mm->last_interval = 0;
 803
 804 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 805         mm->pmd_huge_pte = NULL;
 806 #endif
 807
 808         if (!mm_init(mm, tsk))
 809                 goto fail_nomem;
 810
 811         if (init_new_context(tsk, mm))
 812                 goto fail_nocontext;
 813
 814         dup_mm_exe_file(oldmm, mm);
 815
 816         err = dup_mmap(mm, oldmm);
 817         if (err)
 818                 goto free_pt;
 819
 820         mm->hiwater_rss = get_mm_rss(mm);
 821         mm->hiwater_vm = mm->total_vm;
 822
 823         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 824                 goto free_pt;
 825
 826         return mm;
 827
 828 free_pt:
 829         /* don't put binfmt in mmput, we haven't got module yet */
 830         mm->binfmt = NULL;
 831         mmput(mm);
 832
 833 fail_nomem:
 834         return NULL;
 835
 836 fail_nocontext:
 837         /*
 838          * If init_new_context() failed, we cannot use mmput() to free the mm
 839          * because it calls destroy_context()
 840          */
 841         mm_free_pgd(mm);
 842         free_mm(mm);
 843         return NULL;
 844 }
 845
 846 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 847 {
 848         struct mm_struct *mm, *oldmm;
 849         int retval;
 850
 851         tsk->min_flt = tsk->maj_flt = 0;
 852         tsk->nvcsw = tsk->nivcsw = 0;
 853 #ifdef CONFIG_DETECT_HUNG_TASK
 854         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 855 #endif
 856
 857         tsk->mm = NULL;
 858         tsk->active_mm = NULL;
 859
 860         /*
 861          * Are we cloning a kernel thread?
 862          *
 863          * We need to steal a active VM for that..
 864          */
 865         oldmm = current->mm;
 866         if (!oldmm)
 867                 return 0;
 868
 869         if (clone_flags & CLONE_VM) {
 870                 atomic_inc(&oldmm->mm_users);
 871                 mm = oldmm;
 872                 goto good_mm;
 873         }
 874
 875         retval = -ENOMEM;
 876         mm = dup_mm(tsk);
 877         if (!mm)
 878                 goto fail_nomem;
 879
 880 good_mm:
 881         /* Initializing for Swap token stuff */
 882         mm->token_priority = 0;
 883         mm->last_interval = 0;
 884
 885         tsk->mm = mm;
 886         tsk->active_mm = mm;
 887         return 0;
 888
 889 fail_nomem:
 890         return retval;
 891 }
 892
 893 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 894 {
 895         struct fs_struct *fs = current->fs;
 896         if (clone_flags & CLONE_FS) {
 897                 /* tsk->fs is already what we want */
 898                 spin_lock(&fs->lock);
 899                 if (fs->in_exec) {
 900                         spin_unlock(&fs->lock);
 901                         return -EAGAIN;
 902                 }
 903                 fs->users++;
 904                 spin_unlock(&fs->lock);
 905                 return 0;
 906         }
 907         tsk->fs = copy_fs_struct(fs);
 908         if (!tsk->fs)
 909                 return -ENOMEM;
 910         return 0;
 911 }
 912
 913 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
 914 {
 915         struct files_struct *oldf, *newf;
 916         int error = 0;
 917
 918         /*
 919          * A background process may not have any files ...
 920          */
 921         oldf = current->files;
 922         if (!oldf)
 923                 goto out;
 924
 925         if (clone_flags & CLONE_FILES) {
 926                 atomic_inc(&oldf->count);
 927                 goto out;
 928         }
 929
 930         newf = dup_fd(oldf, &error);
 931         if (!newf)
 932                 goto out;
 933
 934         tsk->files = newf;
 935         error = 0;
 936 out:
 937         return error;
 938 }
 939
 940 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 941 {
 942 #ifdef CONFIG_BLOCK
 943         struct io_context *ioc = current->io_context;
 944         struct io_context *new_ioc;
 945
 946         if (!ioc)
 947                 return 0;
 948         /*
 949          * Share io context with parent, if CLONE_IO is set
 950          */
 951         if (clone_flags & CLONE_IO) {
 952                 tsk->io_context = ioc_task_link(ioc);
 953                 if (unlikely(!tsk->io_context))
 954                         return -ENOMEM;
 955         } else if (ioprio_valid(ioc->ioprio)) {
 956                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
 957                 if (unlikely(!new_ioc))
 958                         return -ENOMEM;
 959
 960                 new_ioc->ioprio = ioc->ioprio;
 961                 put_io_context(new_ioc);
 962         }
 963 #endif
 964         return 0;
 965 }
 966
 967 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 968 {
 969         struct sighand_struct *sig;
 970
 971         if (clone_flags & CLONE_SIGHAND) {
 972                 atomic_inc(&current->sighand->count);
 973                 return 0;
 974         }
 975         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 976         rcu_assign_pointer(tsk->sighand, sig);
 977         if (!sig)
 978                 return -ENOMEM;
 979         atomic_set(&sig->count, 1);
 980         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 981         return 0;
 982 }
 983
 984 void __cleanup_sighand(struct sighand_struct *sighand)
 985 {
 986         if (atomic_dec_and_test(&sighand->count)) {
 987                 signalfd_cleanup(sighand);
 988                 kmem_cache_free(sighand_cachep, sighand);
 989         }
 990 }
 991
 992
 993 /*
 994  * Initialize POSIX timer handling for a thread group.
 995  */
 996 static void posix_cpu_timers_init_group(struct signal_struct *sig)
 997 {
 998         unsigned long cpu_limit;
 999
1000         /* Thread group counters. */
1001         thread_group_cputime_init(sig);
1002
1003         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1004         if (cpu_limit != RLIM_INFINITY) {
1005                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1006                 sig->cputimer.running = 1;
1007         }
1008
1009         /* The timer lists. */
1010         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1011         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1012         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1013 }
1014
1015 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1016 {
1017         struct signal_struct *sig;
1018
1019         if (clone_flags & CLONE_THREAD)
1020                 return 0;
1021
1022         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1023         tsk->signal = sig;
1024         if (!sig)
1025                 return -ENOMEM;
1026
1027         sig->nr_threads = 1;
1028         atomic_set(&sig->live, 1);
1029         atomic_set(&sig->sigcnt, 1);
1030         init_waitqueue_head(&sig->wait_chldexit);
1031         if (clone_flags & CLONE_NEWPID)
1032                 sig->flags |= SIGNAL_UNKILLABLE;
1033         sig->curr_target = tsk;
1034         init_sigpending(&sig->shared_pending);
1035         INIT_LIST_HEAD(&sig->posix_timers);
1036
1037         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1038         sig->real_timer.function = it_real_fn;
1039
1040         task_lock(current->group_leader);
1041         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1042         task_unlock(current->group_leader);
1043
1044         posix_cpu_timers_init_group(sig);
1045
1046         tty_audit_fork(sig);
1047         sched_autogroup_fork(sig);
1048
1049 #ifdef CONFIG_CGROUPS
1050         init_rwsem(&sig->group_rwsem);
1051 #endif
1052
1053         sig->oom_adj = current->signal->oom_adj;
1054         sig->oom_score_adj = current->signal->oom_score_adj;
1055         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1056
1057         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1058                                    current->signal->is_child_subreaper;
1059
1060         mutex_init(&sig->cred_guard_mutex);
1061
1062         return 0;
1063 }
1064
1065 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
1066 {
1067         unsigned long new_flags = p->flags;
1068
1069         new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1070         new_flags |= PF_FORKNOEXEC;
1071         p->flags = new_flags;
1072 }
1073
1074 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1075 {
1076         current->clear_child_tid = tidptr;
1077
1078         return task_pid_vnr(current);
1079 }
1080
1081 static void rt_mutex_init_task(struct task_struct *p)
1082 {
1083         raw_spin_lock_init(&p->pi_lock);
1084 #ifdef CONFIG_RT_MUTEXES
1085         plist_head_init(&p->pi_waiters);
1086         p->pi_blocked_on = NULL;
1087 #endif
1088 }
1089
1090 #ifdef CONFIG_MM_OWNER
1091 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1092 {
1093         mm->owner = p;
1094 }
1095 #endif /* CONFIG_MM_OWNER */
1096
1097 /*
1098  * Initialize POSIX timer handling for a single task.
1099  */
1100 static void posix_cpu_timers_init(struct task_struct *tsk)
1101 {
1102         tsk->cputime_expires.prof_exp = 0;
1103         tsk->cputime_expires.virt_exp = 0;
1104         tsk->cputime_expires.sched_exp = 0;
1105         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1106         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1107         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1108 }
1109
1110 /*
1111  * This creates a new process as a copy of the old one,
1112  * but does not actually start it yet.
1113  *
1114  * It copies the registers, and all the appropriate
1115  * parts of the process environment (as per the clone
1116  * flags). The actual kick-off is left to the caller.
1117  */
1118 static struct task_struct *copy_process(unsigned long clone_flags,
1119                                         unsigned long stack_start,
1120                                         struct pt_regs *regs,
1121                                         unsigned long stack_size,
1122                                         int __user *child_tidptr,
1123                                         struct pid *pid,
1124                                         int trace)
1125 {
1126         int retval;
1127         struct task_struct *p;
1128         int cgroup_callbacks_done = 0;
1129
1130         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1131                 return ERR_PTR(-EINVAL);
1132
1133         /*
1134          * Thread groups must share signals as well, and detached threads
1135          * can only be started up within the thread group.
1136          */
1137         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1138                 return ERR_PTR(-EINVAL);
1139
1140         /*
1141          * Shared signal handlers imply shared VM. By way of the above,
1142          * thread groups also imply shared VM. Blocking this case allows
1143          * for various simplifications in other code.
1144          */
1145         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1146                 return ERR_PTR(-EINVAL);
1147
1148         /*
1149          * Siblings of global init remain as zombies on exit since they are
1150          * not reaped by their parent (swapper). To solve this and to avoid
1151          * multi-rooted process trees, prevent global and container-inits
1152          * from creating siblings.
1153          */
1154         if ((clone_flags & CLONE_PARENT) &&
1155                                 current->signal->flags & SIGNAL_UNKILLABLE)
1156                 return ERR_PTR(-EINVAL);
1157
1158         retval = security_task_create(clone_flags);
1159         if (retval)
1160                 goto fork_out;
1161
1162         retval = -ENOMEM;
1163         p = dup_task_struct(current);
1164         if (!p)
1165                 goto fork_out;
1166
1167         ftrace_graph_init_task(p);
1168         get_seccomp_filter(p);
1169
1170         rt_mutex_init_task(p);
1171
1172 #ifdef CONFIG_PROVE_LOCKING
1173         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1174         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1175 #endif
1176         retval = -EAGAIN;
1177         if (atomic_read(&p->real_cred->user->processes) >=
1178                         task_rlimit(p, RLIMIT_NPROC)) {
1179                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1180                     p->real_cred->user != INIT_USER)
1181                         goto bad_fork_free;
1182         }
1183         current->flags &= ~PF_NPROC_EXCEEDED;
1184
1185         retval = copy_creds(p, clone_flags);
1186         if (retval < 0)
1187                 goto bad_fork_free;
1188
1189         /*
1190          * If multiple threads are within copy_process(), then this check
1191          * triggers too late. This doesn't hurt, the check is only there
1192          * to stop root fork bombs.
1193          */
1194         retval = -EAGAIN;
1195         if (nr_threads >= max_threads)
1196                 goto bad_fork_cleanup_count;
1197
1198         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1199                 goto bad_fork_cleanup_count;
1200
1201         p->did_exec = 0;
1202         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1203         copy_flags(clone_flags, p);
1204         INIT_LIST_HEAD(&p->children);
1205         INIT_LIST_HEAD(&p->sibling);
1206         rcu_copy_process(p);
1207         p->vfork_done = NULL;
1208         spin_lock_init(&p->alloc_lock);
1209
1210         init_sigpending(&p->pending);
1211
1212         p->utime = p->stime = p->gtime = 0;
1213         p->utimescaled = p->stimescaled = 0;
1214 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1215         p->prev_utime = p->prev_stime = 0;
1216 #endif
1217 #if defined(SPLIT_RSS_COUNTING)
1218         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1219 #endif
1220
1221         p->default_timer_slack_ns = current->timer_slack_ns;
1222
1223         task_io_accounting_init(&p->ioac);
1224         acct_clear_integrals(p);
1225
1226         posix_cpu_timers_init(p);
1227
1228         do_posix_clock_monotonic_gettime(&p->start_time);
1229         p->real_start_time = p->start_time;
1230         monotonic_to_bootbased(&p->real_start_time);
1231         p->io_context = NULL;
1232         p->audit_context = NULL;
1233         if (clone_flags & CLONE_THREAD)
1234                 threadgroup_change_begin(current);
1235         cgroup_fork(p);
1236 #ifdef CONFIG_NUMA
1237         p->mempolicy = mpol_dup(p->mempolicy);
1238         if (IS_ERR(p->mempolicy)) {
1239                 retval = PTR_ERR(p->mempolicy);
1240                 p->mempolicy = NULL;
1241                 goto bad_fork_cleanup_cgroup;
1242         }
1243         mpol_fix_fork_child_flag(p);
1244 #endif
1245 #ifdef CONFIG_CPUSETS
1246         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1247         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1248         seqcount_init(&p->mems_allowed_seq);
1249 #endif
1250 #ifdef CONFIG_TRACE_IRQFLAGS
1251         p->irq_events = 0;
1252 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1253         p->hardirqs_enabled = 1;
1254 #else
1255         p->hardirqs_enabled = 0;
1256 #endif
1257         p->hardirq_enable_ip = 0;
1258         p->hardirq_enable_event = 0;
1259         p->hardirq_disable_ip = _THIS_IP_;
1260         p->hardirq_disable_event = 0;
1261         p->softirqs_enabled = 1;
1262         p->softirq_enable_ip = _THIS_IP_;
1263         p->softirq_enable_event = 0;
1264         p->softirq_disable_ip = 0;
1265         p->softirq_disable_event = 0;
1266         p->hardirq_context = 0;
1267         p->softirq_context = 0;
1268 #endif
1269 #ifdef CONFIG_LOCKDEP
1270         p->lockdep_depth = 0; /* no locks held yet */
1271         p->curr_chain_key = 0;
1272         p->lockdep_recursion = 0;
1273 #endif
1274
1275 #ifdef CONFIG_DEBUG_MUTEXES
1276         p->blocked_on = NULL; /* not blocked yet */
1277 #endif
1278 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1279         p->memcg_batch.do_batch = 0;
1280         p->memcg_batch.memcg = NULL;
1281 #endif
1282
1283         /* Perform scheduler related setup. Assign this task to a CPU. */
1284         sched_fork(p);
1285
1286         retval = perf_event_init_task(p);
1287         if (retval)
1288                 goto bad_fork_cleanup_policy;
1289         retval = audit_alloc(p);
1290         if (retval)
1291                 goto bad_fork_cleanup_policy;
1292         /* copy all the process information */
1293         retval = copy_semundo(clone_flags, p);
1294         if (retval)
1295                 goto bad_fork_cleanup_audit;
1296         retval = copy_files(clone_flags, p);
1297         if (retval)
1298                 goto bad_fork_cleanup_semundo;
1299         retval = copy_fs(clone_flags, p);
1300         if (retval)
1301                 goto bad_fork_cleanup_files;
1302         retval = copy_sighand(clone_flags, p);
1303         if (retval)
1304                 goto bad_fork_cleanup_fs;
1305         retval = copy_signal(clone_flags, p);
1306         if (retval)
1307                 goto bad_fork_cleanup_sighand;
1308         retval = copy_mm(clone_flags, p);
1309         if (retval)
1310                 goto bad_fork_cleanup_signal;
1311         retval = copy_namespaces(clone_flags, p);
1312         if (retval)
1313                 goto bad_fork_cleanup_mm;
1314         retval = copy_io(clone_flags, p);
1315         if (retval)
1316                 goto bad_fork_cleanup_namespaces;
1317         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1318         if (retval)
1319                 goto bad_fork_cleanup_io;
1320
1321         if (pid != &init_struct_pid) {
1322                 retval = -ENOMEM;
1323                 pid = alloc_pid(p->nsproxy->pid_ns);
1324                 if (!pid)
1325                         goto bad_fork_cleanup_io;
1326         }
1327
1328         p->pid = pid_nr(pid);
1329         p->tgid = p->pid;
1330         if (clone_flags & CLONE_THREAD)
1331                 p->tgid = current->tgid;
1332
1333         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1334         /*
1335          * Clear TID on mm_release()?
1336          */
1337         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1338 #ifdef CONFIG_BLOCK
1339         p->plug = NULL;
1340 #endif
1341 #ifdef CONFIG_FUTEX
1342         p->robust_list = NULL;
1343 #ifdef CONFIG_COMPAT
1344         p->compat_robust_list = NULL;
1345 #endif
1346         INIT_LIST_HEAD(&p->pi_state_list);
1347         p->pi_state_cache = NULL;
1348 #endif
1349         /*
1350          * sigaltstack should be cleared when sharing the same VM
1351          */
1352         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1353                 p->sas_ss_sp = p->sas_ss_size = 0;
1354
1355         /*
1356          * Syscall tracing and stepping should be turned off in the
1357          * child regardless of CLONE_PTRACE.
1358          */
1359         user_disable_single_step(p);
1360         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1361 #ifdef TIF_SYSCALL_EMU
1362         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1363 #endif
1364         clear_all_latency_tracing(p);
1365
1366         /* ok, now we should be set up.. */
1367         if (clone_flags & CLONE_THREAD)
1368                 p->exit_signal = -1;
1369         else if (clone_flags & CLONE_PARENT)
1370                 p->exit_signal = current->group_leader->exit_signal;
1371         else
1372                 p->exit_signal = (clone_flags & CSIGNAL);
1373
1374         p->pdeath_signal = 0;
1375         p->exit_state = 0;
1376
1377         p->nr_dirtied = 0;
1378         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1379         p->dirty_paused_when = 0;
1380
1381         /*
1382          * Ok, make it visible to the rest of the system.
1383          * We dont wake it up yet.
1384          */
1385         p->group_leader = p;
1386         INIT_LIST_HEAD(&p->thread_group);
1387
1388         /* Now that the task is set up, run cgroup callbacks if
1389          * necessary. We need to run them before the task is visible
1390          * on the tasklist. */
1391         cgroup_fork_callbacks(p);
1392         cgroup_callbacks_done = 1;
1393
1394         /* Need tasklist lock for parent etc handling! */
1395         write_lock_irq(&tasklist_lock);
1396
1397         /* CLONE_PARENT re-uses the old parent */
1398         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1399                 p->real_parent = current->real_parent;
1400                 p->parent_exec_id = current->parent_exec_id;
1401         } else {
1402                 p->real_parent = current;
1403                 p->parent_exec_id = current->self_exec_id;
1404         }
1405
1406         spin_lock(&current->sighand->siglock);
1407
1408         /*
1409          * Process group and session signals need to be delivered to just the
1410          * parent before the fork or both the parent and the child after the
1411          * fork. Restart if a signal comes in before we add the new process to
1412          * it's process group.
1413          * A fatal signal pending means that current will exit, so the new
1414          * thread can't slip out of an OOM kill (or normal SIGKILL).
1415         */
1416         recalc_sigpending();
1417         if (signal_pending(current)) {
1418                 spin_unlock(&current->sighand->siglock);
1419                 write_unlock_irq(&tasklist_lock);
1420                 retval = -ERESTARTNOINTR;
1421                 goto bad_fork_free_pid;
1422         }
1423
1424         if (clone_flags & CLONE_THREAD) {
1425                 current->signal->nr_threads++;
1426                 atomic_inc(&current->signal->live);
1427                 atomic_inc(&current->signal->sigcnt);
1428                 p->group_leader = current->group_leader;
1429                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1430         }
1431
1432         if (likely(p->pid)) {
1433                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1434
1435                 if (thread_group_leader(p)) {
1436                         if (is_child_reaper(pid))
1437                                 p->nsproxy->pid_ns->child_reaper = p;
1438
1439                         p->signal->leader_pid = pid;
1440                         p->signal->tty = tty_kref_get(current->signal->tty);
1441                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1442                         attach_pid(p, PIDTYPE_SID, task_session(current));
1443                         list_add_tail(&p->sibling, &p->real_parent->children);
1444                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1445                         __this_cpu_inc(process_counts);
1446                 }
1447                 attach_pid(p, PIDTYPE_PID, pid);
1448                 nr_threads++;
1449         }
1450
1451         total_forks++;
1452         spin_unlock(&current->sighand->siglock);
1453         write_unlock_irq(&tasklist_lock);
1454         proc_fork_connector(p);
1455         cgroup_post_fork(p);
1456         if (clone_flags & CLONE_THREAD)
1457                 threadgroup_change_end(current);
1458         perf_event_fork(p);
1459
1460         trace_task_newtask(p, clone_flags);
1461
1462         return p;
1463
1464 bad_fork_free_pid:
1465         if (pid != &init_struct_pid)
1466                 free_pid(pid);
1467 bad_fork_cleanup_io:
1468         if (p->io_context)
1469                 exit_io_context(p);
1470 bad_fork_cleanup_namespaces:
1471         if (unlikely(clone_flags & CLONE_NEWPID))
1472                 pid_ns_release_proc(p->nsproxy->pid_ns);
1473         exit_task_namespaces(p);
1474 bad_fork_cleanup_mm:
1475         if (p->mm)
1476                 mmput(p->mm);
1477 bad_fork_cleanup_signal:
1478         if (!(clone_flags & CLONE_THREAD))
1479                 free_signal_struct(p->signal);
1480 bad_fork_cleanup_sighand:
1481         __cleanup_sighand(p->sighand);
1482 bad_fork_cleanup_fs:
1483         exit_fs(p); /* blocking */
1484 bad_fork_cleanup_files:
1485         exit_files(p); /* blocking */
1486 bad_fork_cleanup_semundo:
1487         exit_sem(p);
1488 bad_fork_cleanup_audit:
1489         audit_free(p);
1490 bad_fork_cleanup_policy:
1491         perf_event_free_task(p);
1492 #ifdef CONFIG_NUMA
1493         mpol_put(p->mempolicy);
1494 bad_fork_cleanup_cgroup:
1495 #endif
1496         if (clone_flags & CLONE_THREAD)
1497                 threadgroup_change_end(current);
1498         cgroup_exit(p, cgroup_callbacks_done);
1499         delayacct_tsk_free(p);
1500         module_put(task_thread_info(p)->exec_domain->module);
1501 bad_fork_cleanup_count:
1502         atomic_dec(&p->cred->user->processes);
1503         exit_creds(p);
1504 bad_fork_free:
1505         free_task(p);
1506 fork_out:
1507         return ERR_PTR(retval);
1508 }
1509
1510 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1511 {
1512         memset(regs, 0, sizeof(struct pt_regs));
1513         return regs;
1514 }
1515
1516 static inline void init_idle_pids(struct pid_link *links)
1517 {
1518         enum pid_type type;
1519
1520         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1521                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1522                 links[type].pid = &init_struct_pid;
1523         }
1524 }
1525
1526 struct task_struct * __cpuinit fork_idle(int cpu)
1527 {
1528         struct task_struct *task;
1529         struct pt_regs regs;
1530
1531         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1532                             &init_struct_pid, 0);
1533         if (!IS_ERR(task)) {
1534                 init_idle_pids(task->pids);
1535                 init_idle(task, cpu);
1536         }
1537
1538         return task;
1539 }
1540
1541 /*
1542  *  Ok, this is the main fork-routine.
1543  *
1544  * It copies the process, and if successful kick-starts
1545  * it and waits for it to finish using the VM if required.
1546  */
1547 long do_fork(unsigned long clone_flags,
1548               unsigned long stack_start,
1549               struct pt_regs *regs,
1550               unsigned long stack_size,
1551               int __user *parent_tidptr,
1552               int __user *child_tidptr)
1553 {
1554         struct task_struct *p;
1555         int trace = 0;
1556         long nr;
1557
1558         /*
1559          * Do some preliminary argument and permissions checking before we
1560          * actually start allocating stuff
1561          */
1562         if (clone_flags & CLONE_NEWUSER) {
1563                 if (clone_flags & CLONE_THREAD)
1564                         return -EINVAL;
1565                 /* hopefully this check will go away when userns support is
1566                  * complete
1567                  */
1568                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1569                                 !capable(CAP_SETGID))
1570                         return -EPERM;
1571         }
1572
1573         /*
1574          * Determine whether and which event to report to ptracer.  When
1575          * called from kernel_thread or CLONE_UNTRACED is explicitly
1576          * requested, no event is reported; otherwise, report if the event
1577          * for the type of forking is enabled.
1578          */
1579         if (likely(user_mode(regs)) && !(clone_flags & CLONE_UNTRACED)) {
1580                 if (clone_flags & CLONE_VFORK)
1581                         trace = PTRACE_EVENT_VFORK;
1582                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1583                         trace = PTRACE_EVENT_CLONE;
1584                 else
1585                         trace = PTRACE_EVENT_FORK;
1586
1587                 if (likely(!ptrace_event_enabled(current, trace)))
1588                         trace = 0;
1589         }
1590
1591         p = copy_process(clone_flags, stack_start, regs, stack_size,
1592                          child_tidptr, NULL, trace);
1593         /*
1594          * Do this prior waking up the new thread - the thread pointer
1595          * might get invalid after that point, if the thread exits quickly.
1596          */
1597         if (!IS_ERR(p)) {
1598                 struct completion vfork;
1599
1600                 trace_sched_process_fork(current, p);
1601
1602                 nr = task_pid_vnr(p);
1603
1604                 if (clone_flags & CLONE_PARENT_SETTID)
1605                         put_user(nr, parent_tidptr);
1606
1607                 if (clone_flags & CLONE_VFORK) {
1608                         p->vfork_done = &vfork;
1609                         init_completion(&vfork);
1610                         get_task_struct(p);
1611                 }
1612
1613                 wake_up_new_task(p);
1614
1615                 /* forking complete and child started to run, tell ptracer */
1616                 if (unlikely(trace))
1617                         ptrace_event(trace, nr);
1618
1619                 if (clone_flags & CLONE_VFORK) {
1620                         if (!wait_for_vfork_done(p, &vfork))
1621                                 ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1622                 }
1623         } else {
1624                 nr = PTR_ERR(p);
1625         }
1626         return nr;
1627 }
1628
1629 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1630 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1631 #endif
1632
1633 static void sighand_ctor(void *data)
1634 {
1635         struct sighand_struct *sighand = data;
1636
1637         spin_lock_init(&sighand->siglock);
1638         init_waitqueue_head(&sighand->signalfd_wqh);
1639 }
1640
1641 void __init proc_caches_init(void)
1642 {
1643         sighand_cachep = kmem_cache_create("sighand_cache",
1644                         sizeof(struct sighand_struct), 0,
1645                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1646                         SLAB_NOTRACK, sighand_ctor);
1647         signal_cachep = kmem_cache_create("signal_cache",
1648                         sizeof(struct signal_struct), 0,
1649                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1650         files_cachep = kmem_cache_create("files_cache",
1651                         sizeof(struct files_struct), 0,
1652                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1653         fs_cachep = kmem_cache_create("fs_cache",
1654                         sizeof(struct fs_struct), 0,
1655                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1656         /*
1657          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1658          * whole struct cpumask for the OFFSTACK case. We could change
1659          * this to *only* allocate as much of it as required by the
1660          * maximum number of CPU's we can ever have.  The cpumask_allocation
1661          * is at the end of the structure, exactly for that reason.
1662          */
1663         mm_cachep = kmem_cache_create("mm_struct",
1664                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1665                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1666         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1667         mmap_init();
1668         nsproxy_cache_init();
1669 }
1670
1671 /*
1672  * Check constraints on flags passed to the unshare system call.
1673  */
1674 static int check_unshare_flags(unsigned long unshare_flags)
1675 {
1676         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1677                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1678                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1679                 return -EINVAL;
1680         /*
1681          * Not implemented, but pretend it works if there is nothing to
1682          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1683          * needs to unshare vm.
1684          */
1685         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1686                 /* FIXME: get_task_mm() increments ->mm_users */
1687                 if (atomic_read(&current->mm->mm_users) > 1)
1688                         return -EINVAL;
1689         }
1690
1691         return 0;
1692 }
1693
1694 /*
1695  * Unshare the filesystem structure if it is being shared
1696  */
1697 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1698 {
1699         struct fs_struct *fs = current->fs;
1700
1701         if (!(unshare_flags & CLONE_FS) || !fs)
1702                 return 0;
1703
1704         /* don't need lock here; in the worst case we'll do useless copy */
1705         if (fs->users == 1)
1706                 return 0;
1707
1708         *new_fsp = copy_fs_struct(fs);
1709         if (!*new_fsp)
1710                 return -ENOMEM;
1711
1712         return 0;
1713 }
1714
1715 /*
1716  * Unshare file descriptor table if it is being shared
1717  */
1718 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1719 {
1720         struct files_struct *fd = current->files;
1721         int error = 0;
1722
1723         if ((unshare_flags & CLONE_FILES) &&
1724             (fd && atomic_read(&fd->count) > 1)) {
1725                 *new_fdp = dup_fd(fd, &error);
1726                 if (!*new_fdp)
1727                         return error;
1728         }
1729
1730         return 0;
1731 }
1732
1733 /*
1734  * unshare allows a process to 'unshare' part of the process
1735  * context which was originally shared using clone.  copy_*
1736  * functions used by do_fork() cannot be used here directly
1737  * because they modify an inactive task_struct that is being
1738  * constructed. Here we are modifying the current, active,
1739  * task_struct.
1740  */
1741 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1742 {
1743         struct fs_struct *fs, *new_fs = NULL;
1744         struct files_struct *fd, *new_fd = NULL;
1745         struct nsproxy *new_nsproxy = NULL;
1746         int do_sysvsem = 0;
1747         int err;
1748
1749         err = check_unshare_flags(unshare_flags);
1750         if (err)
1751                 goto bad_unshare_out;
1752
1753         /*
1754          * If unsharing namespace, must also unshare filesystem information.
1755          */
1756         if (unshare_flags & CLONE_NEWNS)
1757                 unshare_flags |= CLONE_FS;
1758         /*
1759          * CLONE_NEWIPC must also detach from the undolist: after switching
1760          * to a new ipc namespace, the semaphore arrays from the old
1761          * namespace are unreachable.
1762          */
1763         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1764                 do_sysvsem = 1;
1765         err = unshare_fs(unshare_flags, &new_fs);
1766         if (err)
1767                 goto bad_unshare_out;
1768         err = unshare_fd(unshare_flags, &new_fd);
1769         if (err)
1770                 goto bad_unshare_cleanup_fs;
1771         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy, new_fs);
1772         if (err)
1773                 goto bad_unshare_cleanup_fd;
1774
1775         if (new_fs || new_fd || do_sysvsem || new_nsproxy) {
1776                 if (do_sysvsem) {
1777                         /*
1778                          * CLONE_SYSVSEM is equivalent to sys_exit().
1779                          */
1780                         exit_sem(current);
1781                 }
1782
1783                 if (new_nsproxy) {
1784                         switch_task_namespaces(current, new_nsproxy);
1785                         new_nsproxy = NULL;
1786                 }
1787
1788                 task_lock(current);
1789
1790                 if (new_fs) {
1791                         fs = current->fs;
1792                         spin_lock(&fs->lock);
1793                         current->fs = new_fs;
1794                         if (--fs->users)
1795                                 new_fs = NULL;
1796                         else
1797                                 new_fs = fs;
1798                         spin_unlock(&fs->lock);
1799                 }
1800
1801                 if (new_fd) {
1802                         fd = current->files;
1803                         current->files = new_fd;
1804                         new_fd = fd;
1805                 }
1806
1807                 task_unlock(current);
1808         }
1809
1810         if (new_nsproxy)
1811                 put_nsproxy(new_nsproxy);
1812
1813 bad_unshare_cleanup_fd:
1814         if (new_fd)
1815                 put_files_struct(new_fd);
1816
1817 bad_unshare_cleanup_fs:
1818         if (new_fs)
1819                 free_fs_struct(new_fs);
1820
1821 bad_unshare_out:
1822         return err;
1823 }
1824
1825 /*
1826  *      Helper to unshare the files of the current task.
1827  *      We don't want to expose copy_files internals to
1828  *      the exec layer of the kernel.
1829  */
1830
1831 int unshare_files(struct files_struct **displaced)
1832 {
1833         struct task_struct *task = current;
1834         struct files_struct *copy = NULL;
1835         int error;
1836
1837         error = unshare_fd(CLONE_FILES, &copy);
1838         if (error || !copy) {
1839                 *displaced = NULL;
1840                 return error;
1841         }
1842         *displaced = task->files;
1843         task_lock(task);
1844         task->files = copy;
1845         task_unlock(task);
1846         return 0;
1847 }