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