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/mm.h>
  32 #include <linux/vmacache.h>
  33 #include <linux/nsproxy.h>
  34 #include <linux/capability.h>
  35 #include <linux/cpu.h>
  36 #include <linux/cgroup.h>
  37 #include <linux/security.h>
  38 #include <linux/hugetlb.h>
  39 #include <linux/seccomp.h>
  40 #include <linux/swap.h>
  41 #include <linux/syscalls.h>
  42 #include <linux/jiffies.h>
  43 #include <linux/futex.h>
  44 #include <linux/compat.h>
  45 #include <linux/kthread.h>
  46 #include <linux/task_io_accounting_ops.h>
  47 #include <linux/rcupdate.h>
  48 #include <linux/ptrace.h>
  49 #include <linux/mount.h>
  50 #include <linux/audit.h>
  51 #include <linux/memcontrol.h>
  52 #include <linux/ftrace.h>
  53 #include <linux/proc_fs.h>
  54 #include <linux/profile.h>
  55 #include <linux/rmap.h>
  56 #include <linux/ksm.h>
  57 #include <linux/acct.h>
  58 #include <linux/tsacct_kern.h>
  59 #include <linux/cn_proc.h>
  60 #include <linux/freezer.h>
  61 #include <linux/delayacct.h>
  62 #include <linux/taskstats_kern.h>
  63 #include <linux/random.h>
  64 #include <linux/tty.h>
  65 #include <linux/blkdev.h>
  66 #include <linux/fs_struct.h>
  67 #include <linux/magic.h>
  68 #include <linux/perf_event.h>
  69 #include <linux/posix-timers.h>
  70 #include <linux/user-return-notifier.h>
  71 #include <linux/oom.h>
  72 #include <linux/khugepaged.h>
  73 #include <linux/signalfd.h>
  74 #include <linux/uprobes.h>
  75 #include <linux/aio.h>
  76 #include <linux/compiler.h>
  77 #include <linux/sysctl.h>
  78 #include <linux/kcov.h>
  79
  80 #include <asm/pgtable.h>
  81 #include <asm/pgalloc.h>
  82 #include <asm/uaccess.h>
  83 #include <asm/mmu_context.h>
  84 #include <asm/cacheflush.h>
  85 #include <asm/tlbflush.h>
  86
  87 #include <trace/events/sched.h>
  88
  89 #define CREATE_TRACE_POINTS
  90 #include <trace/events/task.h>
  91
  92 /*
  93  * Minimum number of threads to boot the kernel
  94  */
  95 #define MIN_THREADS 20
  96
  97 /*
  98  * Maximum number of threads
  99  */
 100 #define MAX_THREADS FUTEX_TID_MASK
 101
 102 /*
 103  * Protected counters by write_lock_irq(&tasklist_lock)
 104  */
 105 unsigned long total_forks;      /* Handle normal Linux uptimes. */
 106 int nr_threads;                 /* The idle threads do not count.. */
 107
 108 int max_threads;                /* tunable limit on nr_threads */
 109
 110 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 111
 112 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 113
 114 #ifdef CONFIG_PROVE_RCU
 115 int lockdep_tasklist_lock_is_held(void)
 116 {
 117         return lockdep_is_held(&tasklist_lock);
 118 }
 119 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 120 #endif /* #ifdef CONFIG_PROVE_RCU */
 121
 122 int nr_processes(void)
 123 {
 124         int cpu;
 125         int total = 0;
 126
 127         for_each_possible_cpu(cpu)
 128                 total += per_cpu(process_counts, cpu);
 129
 130         return total;
 131 }
 132
 133 void __weak arch_release_task_struct(struct task_struct *tsk)
 134 {
 135 }
 136
 137 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 138 static struct kmem_cache *task_struct_cachep;
 139
 140 static inline struct task_struct *alloc_task_struct_node(int node)
 141 {
 142         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 143 }
 144
 145 static inline void free_task_struct(struct task_struct *tsk)
 146 {
 147         kmem_cache_free(task_struct_cachep, tsk);
 148 }
 149 #endif
 150
 151 void __weak arch_release_thread_info(struct thread_info *ti)
 152 {
 153 }
 154
 155 #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
 156
 157 /*
 158  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 159  * kmemcache based allocator.
 160  */
 161 # if THREAD_SIZE >= PAGE_SIZE
 162 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 163                                                   int node)
 164 {
 165         struct page *page = alloc_kmem_pages_node(node, THREADINFO_GFP,
 166                                                   THREAD_SIZE_ORDER);
 167
 168         if (page)
 169                 memcg_kmem_update_page_stat(page, MEMCG_KERNEL_STACK,
 170                                             1 << THREAD_SIZE_ORDER);
 171
 172         return page ? page_address(page) : NULL;
 173 }
 174
 175 static inline void free_thread_info(struct thread_info *ti)
 176 {
 177         struct page *page = virt_to_page(ti);
 178
 179         memcg_kmem_update_page_stat(page, MEMCG_KERNEL_STACK,
 180                                     -(1 << THREAD_SIZE_ORDER));
 181         __free_kmem_pages(page, THREAD_SIZE_ORDER);
 182 }
 183 # else
 184 static struct kmem_cache *thread_info_cache;
 185
 186 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 187                                                   int node)
 188 {
 189         return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
 190 }
 191
 192 static void free_thread_info(struct thread_info *ti)
 193 {
 194         kmem_cache_free(thread_info_cache, ti);
 195 }
 196
 197 void thread_info_cache_init(void)
 198 {
 199         thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
 200                                               THREAD_SIZE, 0, NULL);
 201         BUG_ON(thread_info_cache == NULL);
 202 }
 203 # endif
 204 #endif
 205
 206 /* SLAB cache for signal_struct structures (tsk->signal) */
 207 static struct kmem_cache *signal_cachep;
 208
 209 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 210 struct kmem_cache *sighand_cachep;
 211
 212 /* SLAB cache for files_struct structures (tsk->files) */
 213 struct kmem_cache *files_cachep;
 214
 215 /* SLAB cache for fs_struct structures (tsk->fs) */
 216 struct kmem_cache *fs_cachep;
 217
 218 /* SLAB cache for vm_area_struct structures */
 219 struct kmem_cache *vm_area_cachep;
 220
 221 /* SLAB cache for mm_struct structures (tsk->mm) */
 222 static struct kmem_cache *mm_cachep;
 223
 224 static void account_kernel_stack(struct thread_info *ti, int account)
 225 {
 226         struct zone *zone = page_zone(virt_to_page(ti));
 227
 228         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
 229 }
 230
 231 void free_task(struct task_struct *tsk)
 232 {
 233         account_kernel_stack(tsk->stack, -1);
 234         arch_release_thread_info(tsk->stack);
 235         free_thread_info(tsk->stack);
 236         rt_mutex_debug_task_free(tsk);
 237         ftrace_graph_exit_task(tsk);
 238         put_seccomp_filter(tsk);
 239         arch_release_task_struct(tsk);
 240         free_task_struct(tsk);
 241 }
 242 EXPORT_SYMBOL(free_task);
 243
 244 static inline void free_signal_struct(struct signal_struct *sig)
 245 {
 246         taskstats_tgid_free(sig);
 247         sched_autogroup_exit(sig);
 248         kmem_cache_free(signal_cachep, sig);
 249 }
 250
 251 static inline void put_signal_struct(struct signal_struct *sig)
 252 {
 253         if (atomic_dec_and_test(&sig->sigcnt))
 254                 free_signal_struct(sig);
 255 }
 256
 257 void __put_task_struct(struct task_struct *tsk)
 258 {
 259         WARN_ON(!tsk->exit_state);
 260         WARN_ON(atomic_read(&tsk->usage));
 261         WARN_ON(tsk == current);
 262
 263         cgroup_free(tsk);
 264         task_numa_free(tsk);
 265         security_task_free(tsk);
 266         exit_creds(tsk);
 267         delayacct_tsk_free(tsk);
 268         put_signal_struct(tsk->signal);
 269
 270         if (!profile_handoff_task(tsk))
 271                 free_task(tsk);
 272 }
 273 EXPORT_SYMBOL_GPL(__put_task_struct);
 274
 275 void __init __weak arch_task_cache_init(void) { }
 276
 277 /*
 278  * set_max_threads
 279  */
 280 static void set_max_threads(unsigned int max_threads_suggested)
 281 {
 282         u64 threads;
 283
 284         /*
 285          * The number of threads shall be limited such that the thread
 286          * structures may only consume a small part of the available memory.
 287          */
 288         if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
 289                 threads = MAX_THREADS;
 290         else
 291                 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
 292                                     (u64) THREAD_SIZE * 8UL);
 293
 294         if (threads > max_threads_suggested)
 295                 threads = max_threads_suggested;
 296
 297         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 298 }
 299
 300 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 301 /* Initialized by the architecture: */
 302 int arch_task_struct_size __read_mostly;
 303 #endif
 304
 305 void __init fork_init(void)
 306 {
 307 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 308 #ifndef ARCH_MIN_TASKALIGN
 309 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 310 #endif
 311         /* create a slab on which task_structs can be allocated */
 312         task_struct_cachep = kmem_cache_create("task_struct",
 313                         arch_task_struct_size, ARCH_MIN_TASKALIGN,
 314                         SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT, NULL);
 315 #endif
 316
 317         /* do the arch specific task caches init */
 318         arch_task_cache_init();
 319
 320         set_max_threads(MAX_THREADS);
 321
 322         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 323         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 324         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 325                 init_task.signal->rlim[RLIMIT_NPROC];
 326 }
 327
 328 int __weak arch_dup_task_struct(struct task_struct *dst,
 329                                                struct task_struct *src)
 330 {
 331         *dst = *src;
 332         return 0;
 333 }
 334
 335 void set_task_stack_end_magic(struct task_struct *tsk)
 336 {
 337         unsigned long *stackend;
 338
 339         stackend = end_of_stack(tsk);
 340         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 341 }
 342
 343 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 344 {
 345         struct task_struct *tsk;
 346         struct thread_info *ti;
 347         int err;
 348
 349         if (node == NUMA_NO_NODE)
 350                 node = tsk_fork_get_node(orig);
 351         tsk = alloc_task_struct_node(node);
 352         if (!tsk)
 353                 return NULL;
 354
 355         ti = alloc_thread_info_node(tsk, node);
 356         if (!ti)
 357                 goto free_tsk;
 358
 359         err = arch_dup_task_struct(tsk, orig);
 360         if (err)
 361                 goto free_ti;
 362
 363         tsk->stack = ti;
 364 #ifdef CONFIG_SECCOMP
 365         /*
 366          * We must handle setting up seccomp filters once we're under
 367          * the sighand lock in case orig has changed between now and
 368          * then. Until then, filter must be NULL to avoid messing up
 369          * the usage counts on the error path calling free_task.
 370          */
 371         tsk->seccomp.filter = NULL;
 372 #endif
 373
 374         setup_thread_stack(tsk, orig);
 375         clear_user_return_notifier(tsk);
 376         clear_tsk_need_resched(tsk);
 377         set_task_stack_end_magic(tsk);
 378
 379 #ifdef CONFIG_CC_STACKPROTECTOR
 380         tsk->stack_canary = get_random_int();
 381 #endif
 382
 383         /*
 384          * One for us, one for whoever does the "release_task()" (usually
 385          * parent)
 386          */
 387         atomic_set(&tsk->usage, 2);
 388 #ifdef CONFIG_BLK_DEV_IO_TRACE
 389         tsk->btrace_seq = 0;
 390 #endif
 391         tsk->splice_pipe = NULL;
 392         tsk->task_frag.page = NULL;
 393         tsk->wake_q.next = NULL;
 394
 395         account_kernel_stack(ti, 1);
 396
 397         kcov_task_init(tsk);
 398
 399         return tsk;
 400
 401 free_ti:
 402         free_thread_info(ti);
 403 free_tsk:
 404         free_task_struct(tsk);
 405         return NULL;
 406 }
 407
 408 #ifdef CONFIG_MMU
 409 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 410 {
 411         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 412         struct rb_node **rb_link, *rb_parent;
 413         int retval;
 414         unsigned long charge;
 415
 416         uprobe_start_dup_mmap();
 417         if (down_write_killable(&oldmm->mmap_sem)) {
 418                 retval = -EINTR;
 419                 goto fail_uprobe_end;
 420         }
 421         flush_cache_dup_mm(oldmm);
 422         uprobe_dup_mmap(oldmm, mm);
 423         /*
 424          * Not linked in yet - no deadlock potential:
 425          */
 426         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 427
 428         /* No ordering required: file already has been exposed. */
 429         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 430
 431         mm->total_vm = oldmm->total_vm;
 432         mm->data_vm = oldmm->data_vm;
 433         mm->exec_vm = oldmm->exec_vm;
 434         mm->stack_vm = oldmm->stack_vm;
 435
 436         rb_link = &mm->mm_rb.rb_node;
 437         rb_parent = NULL;
 438         pprev = &mm->mmap;
 439         retval = ksm_fork(mm, oldmm);
 440         if (retval)
 441                 goto out;
 442         retval = khugepaged_fork(mm, oldmm);
 443         if (retval)
 444                 goto out;
 445
 446         prev = NULL;
 447         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 448                 struct file *file;
 449
 450                 if (mpnt->vm_flags & VM_DONTCOPY) {
 451                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 452                         continue;
 453                 }
 454                 charge = 0;
 455                 if (mpnt->vm_flags & VM_ACCOUNT) {
 456                         unsigned long len = vma_pages(mpnt);
 457
 458                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 459                                 goto fail_nomem;
 460                         charge = len;
 461                 }
 462                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 463                 if (!tmp)
 464                         goto fail_nomem;
 465                 *tmp = *mpnt;
 466                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 467                 retval = vma_dup_policy(mpnt, tmp);
 468                 if (retval)
 469                         goto fail_nomem_policy;
 470                 tmp->vm_mm = mm;
 471                 if (anon_vma_fork(tmp, mpnt))
 472                         goto fail_nomem_anon_vma_fork;
 473                 tmp->vm_flags &=
 474                         ~(VM_LOCKED|VM_LOCKONFAULT|VM_UFFD_MISSING|VM_UFFD_WP);
 475                 tmp->vm_next = tmp->vm_prev = NULL;
 476                 tmp->vm_userfaultfd_ctx = NULL_VM_UFFD_CTX;
 477                 file = tmp->vm_file;
 478                 if (file) {
 479                         struct inode *inode = file_inode(file);
 480                         struct address_space *mapping = file->f_mapping;
 481
 482                         get_file(file);
 483                         if (tmp->vm_flags & VM_DENYWRITE)
 484                                 atomic_dec(&inode->i_writecount);
 485                         i_mmap_lock_write(mapping);
 486                         if (tmp->vm_flags & VM_SHARED)
 487                                 atomic_inc(&mapping->i_mmap_writable);
 488                         flush_dcache_mmap_lock(mapping);
 489                         /* insert tmp into the share list, just after mpnt */
 490                         vma_interval_tree_insert_after(tmp, mpnt,
 491                                         &mapping->i_mmap);
 492                         flush_dcache_mmap_unlock(mapping);
 493                         i_mmap_unlock_write(mapping);
 494                 }
 495
 496                 /*
 497                  * Clear hugetlb-related page reserves for children. This only
 498                  * affects MAP_PRIVATE mappings. Faults generated by the child
 499                  * are not guaranteed to succeed, even if read-only
 500                  */
 501                 if (is_vm_hugetlb_page(tmp))
 502                         reset_vma_resv_huge_pages(tmp);
 503
 504                 /*
 505                  * Link in the new vma and copy the page table entries.
 506                  */
 507                 *pprev = tmp;
 508                 pprev = &tmp->vm_next;
 509                 tmp->vm_prev = prev;
 510                 prev = tmp;
 511
 512                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 513                 rb_link = &tmp->vm_rb.rb_right;
 514                 rb_parent = &tmp->vm_rb;
 515
 516                 mm->map_count++;
 517                 retval = copy_page_range(mm, oldmm, mpnt);
 518
 519                 if (tmp->vm_ops && tmp->vm_ops->open)
 520                         tmp->vm_ops->open(tmp);
 521
 522                 if (retval)
 523                         goto out;
 524         }
 525         /* a new mm has just been created */
 526         arch_dup_mmap(oldmm, mm);
 527         retval = 0;
 528 out:
 529         up_write(&mm->mmap_sem);
 530         flush_tlb_mm(oldmm);
 531         up_write(&oldmm->mmap_sem);
 532 fail_uprobe_end:
 533         uprobe_end_dup_mmap();
 534         return retval;
 535 fail_nomem_anon_vma_fork:
 536         mpol_put(vma_policy(tmp));
 537 fail_nomem_policy:
 538         kmem_cache_free(vm_area_cachep, tmp);
 539 fail_nomem:
 540         retval = -ENOMEM;
 541         vm_unacct_memory(charge);
 542         goto out;
 543 }
 544
 545 static inline int mm_alloc_pgd(struct mm_struct *mm)
 546 {
 547         mm->pgd = pgd_alloc(mm);
 548         if (unlikely(!mm->pgd))
 549                 return -ENOMEM;
 550         return 0;
 551 }
 552
 553 static inline void mm_free_pgd(struct mm_struct *mm)
 554 {
 555         pgd_free(mm, mm->pgd);
 556 }
 557 #else
 558 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 559 {
 560         down_write(&oldmm->mmap_sem);
 561         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 562         up_write(&oldmm->mmap_sem);
 563         return 0;
 564 }
 565 #define mm_alloc_pgd(mm)        (0)
 566 #define mm_free_pgd(mm)
 567 #endif /* CONFIG_MMU */
 568
 569 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 570
 571 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 572 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 573
 574 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 575
 576 static int __init coredump_filter_setup(char *s)
 577 {
 578         default_dump_filter =
 579                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 580                 MMF_DUMP_FILTER_MASK;
 581         return 1;
 582 }
 583
 584 __setup("coredump_filter=", coredump_filter_setup);
 585
 586 #include <linux/init_task.h>
 587
 588 static void mm_init_aio(struct mm_struct *mm)
 589 {
 590 #ifdef CONFIG_AIO
 591         spin_lock_init(&mm->ioctx_lock);
 592         mm->ioctx_table = NULL;
 593 #endif
 594 }
 595
 596 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 597 {
 598 #ifdef CONFIG_MEMCG
 599         mm->owner = p;
 600 #endif
 601 }
 602
 603 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
 604 {
 605         mm->mmap = NULL;
 606         mm->mm_rb = RB_ROOT;
 607         mm->vmacache_seqnum = 0;
 608         atomic_set(&mm->mm_users, 1);
 609         atomic_set(&mm->mm_count, 1);
 610         init_rwsem(&mm->mmap_sem);
 611         INIT_LIST_HEAD(&mm->mmlist);
 612         mm->core_state = NULL;
 613         atomic_long_set(&mm->nr_ptes, 0);
 614         mm_nr_pmds_init(mm);
 615         mm->map_count = 0;
 616         mm->locked_vm = 0;
 617         mm->pinned_vm = 0;
 618         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 619         spin_lock_init(&mm->page_table_lock);
 620         mm_init_cpumask(mm);
 621         mm_init_aio(mm);
 622         mm_init_owner(mm, p);
 623         mmu_notifier_mm_init(mm);
 624         clear_tlb_flush_pending(mm);
 625 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 626         mm->pmd_huge_pte = NULL;
 627 #endif
 628
 629         if (current->mm) {
 630                 mm->flags = current->mm->flags & MMF_INIT_MASK;
 631                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
 632         } else {
 633                 mm->flags = default_dump_filter;
 634                 mm->def_flags = 0;
 635         }
 636
 637         if (mm_alloc_pgd(mm))
 638                 goto fail_nopgd;
 639
 640         if (init_new_context(p, mm))
 641                 goto fail_nocontext;
 642
 643         return mm;
 644
 645 fail_nocontext:
 646         mm_free_pgd(mm);
 647 fail_nopgd:
 648         free_mm(mm);
 649         return NULL;
 650 }
 651
 652 static void check_mm(struct mm_struct *mm)
 653 {
 654         int i;
 655
 656         for (i = 0; i < NR_MM_COUNTERS; i++) {
 657                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 658
 659                 if (unlikely(x))
 660                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 661                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 662         }
 663
 664         if (atomic_long_read(&mm->nr_ptes))
 665                 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
 666                                 atomic_long_read(&mm->nr_ptes));
 667         if (mm_nr_pmds(mm))
 668                 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
 669                                 mm_nr_pmds(mm));
 670
 671 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 672         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 673 #endif
 674 }
 675
 676 /*
 677  * Allocate and initialize an mm_struct.
 678  */
 679 struct mm_struct *mm_alloc(void)
 680 {
 681         struct mm_struct *mm;
 682
 683         mm = allocate_mm();
 684         if (!mm)
 685                 return NULL;
 686
 687         memset(mm, 0, sizeof(*mm));
 688         return mm_init(mm, current);
 689 }
 690
 691 /*
 692  * Called when the last reference to the mm
 693  * is dropped: either by a lazy thread or by
 694  * mmput. Free the page directory and the mm.
 695  */
 696 void __mmdrop(struct mm_struct *mm)
 697 {
 698         BUG_ON(mm == &init_mm);
 699         mm_free_pgd(mm);
 700         destroy_context(mm);
 701         mmu_notifier_mm_destroy(mm);
 702         check_mm(mm);
 703         free_mm(mm);
 704 }
 705 EXPORT_SYMBOL_GPL(__mmdrop);
 706
 707 static inline void __mmput(struct mm_struct *mm)
 708 {
 709         VM_BUG_ON(atomic_read(&mm->mm_users));
 710
 711         uprobe_clear_state(mm);
 712         exit_aio(mm);
 713         ksm_exit(mm);
 714         khugepaged_exit(mm); /* must run before exit_mmap */
 715         exit_mmap(mm);
 716         set_mm_exe_file(mm, NULL);
 717         if (!list_empty(&mm->mmlist)) {
 718                 spin_lock(&mmlist_lock);
 719                 list_del(&mm->mmlist);
 720                 spin_unlock(&mmlist_lock);
 721         }
 722         if (mm->binfmt)
 723                 module_put(mm->binfmt->module);
 724         mmdrop(mm);
 725 }
 726
 727 /*
 728  * Decrement the use count and release all resources for an mm.
 729  */
 730 void mmput(struct mm_struct *mm)
 731 {
 732         might_sleep();
 733
 734         if (atomic_dec_and_test(&mm->mm_users))
 735                 __mmput(mm);
 736 }
 737 EXPORT_SYMBOL_GPL(mmput);
 738
 739 #ifdef CONFIG_MMU
 740 static void mmput_async_fn(struct work_struct *work)
 741 {
 742         struct mm_struct *mm = container_of(work, struct mm_struct, async_put_work);
 743         __mmput(mm);
 744 }
 745
 746 void mmput_async(struct mm_struct *mm)
 747 {
 748         if (atomic_dec_and_test(&mm->mm_users)) {
 749                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
 750                 schedule_work(&mm->async_put_work);
 751         }
 752 }
 753 #endif
 754
 755 /**
 756  * set_mm_exe_file - change a reference to the mm's executable file
 757  *
 758  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
 759  *
 760  * Main users are mmput() and sys_execve(). Callers prevent concurrent
 761  * invocations: in mmput() nobody alive left, in execve task is single
 762  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
 763  * mm->exe_file, but does so without using set_mm_exe_file() in order
 764  * to do avoid the need for any locks.
 765  */
 766 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 767 {
 768         struct file *old_exe_file;
 769
 770         /*
 771          * It is safe to dereference the exe_file without RCU as
 772          * this function is only called if nobody else can access
 773          * this mm -- see comment above for justification.
 774          */
 775         old_exe_file = rcu_dereference_raw(mm->exe_file);
 776
 777         if (new_exe_file)
 778                 get_file(new_exe_file);
 779         rcu_assign_pointer(mm->exe_file, new_exe_file);
 780         if (old_exe_file)
 781                 fput(old_exe_file);
 782 }
 783
 784 /**
 785  * get_mm_exe_file - acquire a reference to the mm's executable file
 786  *
 787  * Returns %NULL if mm has no associated executable file.
 788  * User must release file via fput().
 789  */
 790 struct file *get_mm_exe_file(struct mm_struct *mm)
 791 {
 792         struct file *exe_file;
 793
 794         rcu_read_lock();
 795         exe_file = rcu_dereference(mm->exe_file);
 796         if (exe_file && !get_file_rcu(exe_file))
 797                 exe_file = NULL;
 798         rcu_read_unlock();
 799         return exe_file;
 800 }
 801 EXPORT_SYMBOL(get_mm_exe_file);
 802
 803 /**
 804  * get_task_mm - acquire a reference to the task's mm
 805  *
 806  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 807  * this kernel workthread has transiently adopted a user mm with use_mm,
 808  * to do its AIO) is not set and if so returns a reference to it, after
 809  * bumping up the use count.  User must release the mm via mmput()
 810  * after use.  Typically used by /proc and ptrace.
 811  */
 812 struct mm_struct *get_task_mm(struct task_struct *task)
 813 {
 814         struct mm_struct *mm;
 815
 816         task_lock(task);
 817         mm = task->mm;
 818         if (mm) {
 819                 if (task->flags & PF_KTHREAD)
 820                         mm = NULL;
 821                 else
 822                         atomic_inc(&mm->mm_users);
 823         }
 824         task_unlock(task);
 825         return mm;
 826 }
 827 EXPORT_SYMBOL_GPL(get_task_mm);
 828
 829 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 830 {
 831         struct mm_struct *mm;
 832         int err;
 833
 834         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 835         if (err)
 836                 return ERR_PTR(err);
 837
 838         mm = get_task_mm(task);
 839         if (mm && mm != current->mm &&
 840                         !ptrace_may_access(task, mode)) {
 841                 mmput(mm);
 842                 mm = ERR_PTR(-EACCES);
 843         }
 844         mutex_unlock(&task->signal->cred_guard_mutex);
 845
 846         return mm;
 847 }
 848
 849 static void complete_vfork_done(struct task_struct *tsk)
 850 {
 851         struct completion *vfork;
 852
 853         task_lock(tsk);
 854         vfork = tsk->vfork_done;
 855         if (likely(vfork)) {
 856                 tsk->vfork_done = NULL;
 857                 complete(vfork);
 858         }
 859         task_unlock(tsk);
 860 }
 861
 862 static int wait_for_vfork_done(struct task_struct *child,
 863                                 struct completion *vfork)
 864 {
 865         int killed;
 866
 867         freezer_do_not_count();
 868         killed = wait_for_completion_killable(vfork);
 869         freezer_count();
 870
 871         if (killed) {
 872                 task_lock(child);
 873                 child->vfork_done = NULL;
 874                 task_unlock(child);
 875         }
 876
 877         put_task_struct(child);
 878         return killed;
 879 }
 880
 881 /* Please note the differences between mmput and mm_release.
 882  * mmput is called whenever we stop holding onto a mm_struct,
 883  * error success whatever.
 884  *
 885  * mm_release is called after a mm_struct has been removed
 886  * from the current process.
 887  *
 888  * This difference is important for error handling, when we
 889  * only half set up a mm_struct for a new process and need to restore
 890  * the old one.  Because we mmput the new mm_struct before
 891  * restoring the old one. . .
 892  * Eric Biederman 10 January 1998
 893  */
 894 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 895 {
 896         /* Get rid of any futexes when releasing the mm */
 897 #ifdef CONFIG_FUTEX
 898         if (unlikely(tsk->robust_list)) {
 899                 exit_robust_list(tsk);
 900                 tsk->robust_list = NULL;
 901         }
 902 #ifdef CONFIG_COMPAT
 903         if (unlikely(tsk->compat_robust_list)) {
 904                 compat_exit_robust_list(tsk);
 905                 tsk->compat_robust_list = NULL;
 906         }
 907 #endif
 908         if (unlikely(!list_empty(&tsk->pi_state_list)))
 909                 exit_pi_state_list(tsk);
 910 #endif
 911
 912         uprobe_free_utask(tsk);
 913
 914         /* Get rid of any cached register state */
 915         deactivate_mm(tsk, mm);
 916
 917         /*
 918          * If we're exiting normally, clear a user-space tid field if
 919          * requested.  We leave this alone when dying by signal, to leave
 920          * the value intact in a core dump, and to save the unnecessary
 921          * trouble, say, a killed vfork parent shouldn't touch this mm.
 922          * Userland only wants this done for a sys_exit.
 923          */
 924         if (tsk->clear_child_tid) {
 925                 if (!(tsk->flags & PF_SIGNALED) &&
 926                     atomic_read(&mm->mm_users) > 1) {
 927                         /*
 928                          * We don't check the error code - if userspace has
 929                          * not set up a proper pointer then tough luck.
 930                          */
 931                         put_user(0, tsk->clear_child_tid);
 932                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 933                                         1, NULL, NULL, 0);
 934                 }
 935                 tsk->clear_child_tid = NULL;
 936         }
 937
 938         /*
 939          * All done, finally we can wake up parent and return this mm to him.
 940          * Also kthread_stop() uses this completion for synchronization.
 941          */
 942         if (tsk->vfork_done)
 943                 complete_vfork_done(tsk);
 944 }
 945
 946 /*
 947  * Allocate a new mm structure and copy contents from the
 948  * mm structure of the passed in task structure.
 949  */
 950 static struct mm_struct *dup_mm(struct task_struct *tsk)
 951 {
 952         struct mm_struct *mm, *oldmm = current->mm;
 953         int err;
 954
 955         mm = allocate_mm();
 956         if (!mm)
 957                 goto fail_nomem;
 958
 959         memcpy(mm, oldmm, sizeof(*mm));
 960
 961         if (!mm_init(mm, tsk))
 962                 goto fail_nomem;
 963
 964         err = dup_mmap(mm, oldmm);
 965         if (err)
 966                 goto free_pt;
 967
 968         mm->hiwater_rss = get_mm_rss(mm);
 969         mm->hiwater_vm = mm->total_vm;
 970
 971         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 972                 goto free_pt;
 973
 974         return mm;
 975
 976 free_pt:
 977         /* don't put binfmt in mmput, we haven't got module yet */
 978         mm->binfmt = NULL;
 979         mmput(mm);
 980
 981 fail_nomem:
 982         return NULL;
 983 }
 984
 985 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 986 {
 987         struct mm_struct *mm, *oldmm;
 988         int retval;
 989
 990         tsk->min_flt = tsk->maj_flt = 0;
 991         tsk->nvcsw = tsk->nivcsw = 0;
 992 #ifdef CONFIG_DETECT_HUNG_TASK
 993         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 994 #endif
 995
 996         tsk->mm = NULL;
 997         tsk->active_mm = NULL;
 998
 999         /*
1000          * Are we cloning a kernel thread?
1001          *
1002          * We need to steal a active VM for that..
1003          */
1004         oldmm = current->mm;
1005         if (!oldmm)
1006                 return 0;
1007
1008         /* initialize the new vmacache entries */
1009         vmacache_flush(tsk);
1010
1011         if (clone_flags & CLONE_VM) {
1012                 atomic_inc(&oldmm->mm_users);
1013                 mm = oldmm;
1014                 goto good_mm;
1015         }
1016
1017         retval = -ENOMEM;
1018         mm = dup_mm(tsk);
1019         if (!mm)
1020                 goto fail_nomem;
1021
1022 good_mm:
1023         tsk->mm = mm;
1024         tsk->active_mm = mm;
1025         return 0;
1026
1027 fail_nomem:
1028         return retval;
1029 }
1030
1031 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1032 {
1033         struct fs_struct *fs = current->fs;
1034         if (clone_flags & CLONE_FS) {
1035                 /* tsk->fs is already what we want */
1036                 spin_lock(&fs->lock);
1037                 if (fs->in_exec) {
1038                         spin_unlock(&fs->lock);
1039                         return -EAGAIN;
1040                 }
1041                 fs->users++;
1042                 spin_unlock(&fs->lock);
1043                 return 0;
1044         }
1045         tsk->fs = copy_fs_struct(fs);
1046         if (!tsk->fs)
1047                 return -ENOMEM;
1048         return 0;
1049 }
1050
1051 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1052 {
1053         struct files_struct *oldf, *newf;
1054         int error = 0;
1055
1056         /*
1057          * A background process may not have any files ...
1058          */
1059         oldf = current->files;
1060         if (!oldf)
1061                 goto out;
1062
1063         if (clone_flags & CLONE_FILES) {
1064                 atomic_inc(&oldf->count);
1065                 goto out;
1066         }
1067
1068         newf = dup_fd(oldf, &error);
1069         if (!newf)
1070                 goto out;
1071
1072         tsk->files = newf;
1073         error = 0;
1074 out:
1075         return error;
1076 }
1077
1078 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1079 {
1080 #ifdef CONFIG_BLOCK
1081         struct io_context *ioc = current->io_context;
1082         struct io_context *new_ioc;
1083
1084         if (!ioc)
1085                 return 0;
1086         /*
1087          * Share io context with parent, if CLONE_IO is set
1088          */
1089         if (clone_flags & CLONE_IO) {
1090                 ioc_task_link(ioc);
1091                 tsk->io_context = ioc;
1092         } else if (ioprio_valid(ioc->ioprio)) {
1093                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1094                 if (unlikely(!new_ioc))
1095                         return -ENOMEM;
1096
1097                 new_ioc->ioprio = ioc->ioprio;
1098                 put_io_context(new_ioc);
1099         }
1100 #endif
1101         return 0;
1102 }
1103
1104 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1105 {
1106         struct sighand_struct *sig;
1107
1108         if (clone_flags & CLONE_SIGHAND) {
1109                 atomic_inc(&current->sighand->count);
1110                 return 0;
1111         }
1112         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1113         rcu_assign_pointer(tsk->sighand, sig);
1114         if (!sig)
1115                 return -ENOMEM;
1116
1117         atomic_set(&sig->count, 1);
1118         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1119         return 0;
1120 }
1121
1122 void __cleanup_sighand(struct sighand_struct *sighand)
1123 {
1124         if (atomic_dec_and_test(&sighand->count)) {
1125                 signalfd_cleanup(sighand);
1126                 /*
1127                  * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1128                  * without an RCU grace period, see __lock_task_sighand().
1129                  */
1130                 kmem_cache_free(sighand_cachep, sighand);
1131         }
1132 }
1133
1134 /*
1135  * Initialize POSIX timer handling for a thread group.
1136  */
1137 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1138 {
1139         unsigned long cpu_limit;
1140
1141         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1142         if (cpu_limit != RLIM_INFINITY) {
1143                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1144                 sig->cputimer.running = true;
1145         }
1146
1147         /* The timer lists. */
1148         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1149         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1150         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1151 }
1152
1153 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1154 {
1155         struct signal_struct *sig;
1156
1157         if (clone_flags & CLONE_THREAD)
1158                 return 0;
1159
1160         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1161         tsk->signal = sig;
1162         if (!sig)
1163                 return -ENOMEM;
1164
1165         sig->nr_threads = 1;
1166         atomic_set(&sig->live, 1);
1167         atomic_set(&sig->sigcnt, 1);
1168
1169         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1170         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1171         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1172
1173         init_waitqueue_head(&sig->wait_chldexit);
1174         sig->curr_target = tsk;
1175         init_sigpending(&sig->shared_pending);
1176         INIT_LIST_HEAD(&sig->posix_timers);
1177         seqlock_init(&sig->stats_lock);
1178         prev_cputime_init(&sig->prev_cputime);
1179
1180         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1181         sig->real_timer.function = it_real_fn;
1182
1183         task_lock(current->group_leader);
1184         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1185         task_unlock(current->group_leader);
1186
1187         posix_cpu_timers_init_group(sig);
1188
1189         tty_audit_fork(sig);
1190         sched_autogroup_fork(sig);
1191
1192         sig->oom_score_adj = current->signal->oom_score_adj;
1193         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1194
1195         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1196                                    current->signal->is_child_subreaper;
1197
1198         mutex_init(&sig->cred_guard_mutex);
1199
1200         return 0;
1201 }
1202
1203 static void copy_seccomp(struct task_struct *p)
1204 {
1205 #ifdef CONFIG_SECCOMP
1206         /*
1207          * Must be called with sighand->lock held, which is common to
1208          * all threads in the group. Holding cred_guard_mutex is not
1209          * needed because this new task is not yet running and cannot
1210          * be racing exec.
1211          */
1212         assert_spin_locked(&current->sighand->siglock);
1213
1214         /* Ref-count the new filter user, and assign it. */
1215         get_seccomp_filter(current);
1216         p->seccomp = current->seccomp;
1217
1218         /*
1219          * Explicitly enable no_new_privs here in case it got set
1220          * between the task_struct being duplicated and holding the
1221          * sighand lock. The seccomp state and nnp must be in sync.
1222          */
1223         if (task_no_new_privs(current))
1224                 task_set_no_new_privs(p);
1225
1226         /*
1227          * If the parent gained a seccomp mode after copying thread
1228          * flags and between before we held the sighand lock, we have
1229          * to manually enable the seccomp thread flag here.
1230          */
1231         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1232                 set_tsk_thread_flag(p, TIF_SECCOMP);
1233 #endif
1234 }
1235
1236 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1237 {
1238         current->clear_child_tid = tidptr;
1239
1240         return task_pid_vnr(current);
1241 }
1242
1243 static void rt_mutex_init_task(struct task_struct *p)
1244 {
1245         raw_spin_lock_init(&p->pi_lock);
1246 #ifdef CONFIG_RT_MUTEXES
1247         p->pi_waiters = RB_ROOT;
1248         p->pi_waiters_leftmost = NULL;
1249         p->pi_blocked_on = NULL;
1250 #endif
1251 }
1252
1253 /*
1254  * Initialize POSIX timer handling for a single task.
1255  */
1256 static void posix_cpu_timers_init(struct task_struct *tsk)
1257 {
1258         tsk->cputime_expires.prof_exp = 0;
1259         tsk->cputime_expires.virt_exp = 0;
1260         tsk->cputime_expires.sched_exp = 0;
1261         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1262         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1263         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1264 }
1265
1266 static inline void
1267 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1268 {
1269          task->pids[type].pid = pid;
1270 }
1271
1272 /*
1273  * This creates a new process as a copy of the old one,
1274  * but does not actually start it yet.
1275  *
1276  * It copies the registers, and all the appropriate
1277  * parts of the process environment (as per the clone
1278  * flags). The actual kick-off is left to the caller.
1279  */
1280 static struct task_struct *copy_process(unsigned long clone_flags,
1281                                         unsigned long stack_start,
1282                                         unsigned long stack_size,
1283                                         int __user *child_tidptr,
1284                                         struct pid *pid,
1285                                         int trace,
1286                                         unsigned long tls,
1287                                         int node)
1288 {
1289         int retval;
1290         struct task_struct *p;
1291
1292         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1293                 return ERR_PTR(-EINVAL);
1294
1295         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1296                 return ERR_PTR(-EINVAL);
1297
1298         /*
1299          * Thread groups must share signals as well, and detached threads
1300          * can only be started up within the thread group.
1301          */
1302         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1303                 return ERR_PTR(-EINVAL);
1304
1305         /*
1306          * Shared signal handlers imply shared VM. By way of the above,
1307          * thread groups also imply shared VM. Blocking this case allows
1308          * for various simplifications in other code.
1309          */
1310         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1311                 return ERR_PTR(-EINVAL);
1312
1313         /*
1314          * Siblings of global init remain as zombies on exit since they are
1315          * not reaped by their parent (swapper). To solve this and to avoid
1316          * multi-rooted process trees, prevent global and container-inits
1317          * from creating siblings.
1318          */
1319         if ((clone_flags & CLONE_PARENT) &&
1320                                 current->signal->flags & SIGNAL_UNKILLABLE)
1321                 return ERR_PTR(-EINVAL);
1322
1323         /*
1324          * If the new process will be in a different pid or user namespace
1325          * do not allow it to share a thread group with the forking task.
1326          */
1327         if (clone_flags & CLONE_THREAD) {
1328                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1329                     (task_active_pid_ns(current) !=
1330                                 current->nsproxy->pid_ns_for_children))
1331                         return ERR_PTR(-EINVAL);
1332         }
1333
1334         retval = security_task_create(clone_flags);
1335         if (retval)
1336                 goto fork_out;
1337
1338         retval = -ENOMEM;
1339         p = dup_task_struct(current, node);
1340         if (!p)
1341                 goto fork_out;
1342
1343         ftrace_graph_init_task(p);
1344
1345         rt_mutex_init_task(p);
1346
1347 #ifdef CONFIG_PROVE_LOCKING
1348         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1349         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1350 #endif
1351         retval = -EAGAIN;
1352         if (atomic_read(&p->real_cred->user->processes) >=
1353                         task_rlimit(p, RLIMIT_NPROC)) {
1354                 if (p->real_cred->user != INIT_USER &&
1355                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1356                         goto bad_fork_free;
1357         }
1358         current->flags &= ~PF_NPROC_EXCEEDED;
1359
1360         retval = copy_creds(p, clone_flags);
1361         if (retval < 0)
1362                 goto bad_fork_free;
1363
1364         /*
1365          * If multiple threads are within copy_process(), then this check
1366          * triggers too late. This doesn't hurt, the check is only there
1367          * to stop root fork bombs.
1368          */
1369         retval = -EAGAIN;
1370         if (nr_threads >= max_threads)
1371                 goto bad_fork_cleanup_count;
1372
1373         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1374         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1375         p->flags |= PF_FORKNOEXEC;
1376         INIT_LIST_HEAD(&p->children);
1377         INIT_LIST_HEAD(&p->sibling);
1378         rcu_copy_process(p);
1379         p->vfork_done = NULL;
1380         spin_lock_init(&p->alloc_lock);
1381
1382         init_sigpending(&p->pending);
1383
1384         p->utime = p->stime = p->gtime = 0;
1385         p->utimescaled = p->stimescaled = 0;
1386         prev_cputime_init(&p->prev_cputime);
1387
1388 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1389         seqcount_init(&p->vtime_seqcount);
1390         p->vtime_snap = 0;
1391         p->vtime_snap_whence = VTIME_INACTIVE;
1392 #endif
1393
1394 #if defined(SPLIT_RSS_COUNTING)
1395         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1396 #endif
1397
1398         p->default_timer_slack_ns = current->timer_slack_ns;
1399
1400         task_io_accounting_init(&p->ioac);
1401         acct_clear_integrals(p);
1402
1403         posix_cpu_timers_init(p);
1404
1405         p->start_time = ktime_get_ns();
1406         p->real_start_time = ktime_get_boot_ns();
1407         p->io_context = NULL;
1408         p->audit_context = NULL;
1409         threadgroup_change_begin(current);
1410         cgroup_fork(p);
1411 #ifdef CONFIG_NUMA
1412         p->mempolicy = mpol_dup(p->mempolicy);
1413         if (IS_ERR(p->mempolicy)) {
1414                 retval = PTR_ERR(p->mempolicy);
1415                 p->mempolicy = NULL;
1416                 goto bad_fork_cleanup_threadgroup_lock;
1417         }
1418 #endif
1419 #ifdef CONFIG_CPUSETS
1420         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1421         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1422         seqcount_init(&p->mems_allowed_seq);
1423 #endif
1424 #ifdef CONFIG_TRACE_IRQFLAGS
1425         p->irq_events = 0;
1426         p->hardirqs_enabled = 0;
1427         p->hardirq_enable_ip = 0;
1428         p->hardirq_enable_event = 0;
1429         p->hardirq_disable_ip = _THIS_IP_;
1430         p->hardirq_disable_event = 0;
1431         p->softirqs_enabled = 1;
1432         p->softirq_enable_ip = _THIS_IP_;
1433         p->softirq_enable_event = 0;
1434         p->softirq_disable_ip = 0;
1435         p->softirq_disable_event = 0;
1436         p->hardirq_context = 0;
1437         p->softirq_context = 0;
1438 #endif
1439
1440         p->pagefault_disabled = 0;
1441
1442 #ifdef CONFIG_LOCKDEP
1443         p->lockdep_depth = 0; /* no locks held yet */
1444         p->curr_chain_key = 0;
1445         p->lockdep_recursion = 0;
1446 #endif
1447
1448 #ifdef CONFIG_DEBUG_MUTEXES
1449         p->blocked_on = NULL; /* not blocked yet */
1450 #endif
1451 #ifdef CONFIG_BCACHE
1452         p->sequential_io        = 0;
1453         p->sequential_io_avg    = 0;
1454 #endif
1455
1456         /* Perform scheduler related setup. Assign this task to a CPU. */
1457         retval = sched_fork(clone_flags, p);
1458         if (retval)
1459                 goto bad_fork_cleanup_policy;
1460
1461         retval = perf_event_init_task(p);
1462         if (retval)
1463                 goto bad_fork_cleanup_policy;
1464         retval = audit_alloc(p);
1465         if (retval)
1466                 goto bad_fork_cleanup_perf;
1467         /* copy all the process information */
1468         shm_init_task(p);
1469         retval = copy_semundo(clone_flags, p);
1470         if (retval)
1471                 goto bad_fork_cleanup_audit;
1472         retval = copy_files(clone_flags, p);
1473         if (retval)
1474                 goto bad_fork_cleanup_semundo;
1475         retval = copy_fs(clone_flags, p);
1476         if (retval)
1477                 goto bad_fork_cleanup_files;
1478         retval = copy_sighand(clone_flags, p);
1479         if (retval)
1480                 goto bad_fork_cleanup_fs;
1481         retval = copy_signal(clone_flags, p);
1482         if (retval)
1483                 goto bad_fork_cleanup_sighand;
1484         retval = copy_mm(clone_flags, p);
1485         if (retval)
1486                 goto bad_fork_cleanup_signal;
1487         retval = copy_namespaces(clone_flags, p);
1488         if (retval)
1489                 goto bad_fork_cleanup_mm;
1490         retval = copy_io(clone_flags, p);
1491         if (retval)
1492                 goto bad_fork_cleanup_namespaces;
1493         retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1494         if (retval)
1495                 goto bad_fork_cleanup_io;
1496
1497         if (pid != &init_struct_pid) {
1498                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1499                 if (IS_ERR(pid)) {
1500                         retval = PTR_ERR(pid);
1501                         goto bad_fork_cleanup_thread;
1502                 }
1503         }
1504
1505         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1506         /*
1507          * Clear TID on mm_release()?
1508          */
1509         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1510 #ifdef CONFIG_BLOCK
1511         p->plug = NULL;
1512 #endif
1513 #ifdef CONFIG_FUTEX
1514         p->robust_list = NULL;
1515 #ifdef CONFIG_COMPAT
1516         p->compat_robust_list = NULL;
1517 #endif
1518         INIT_LIST_HEAD(&p->pi_state_list);
1519         p->pi_state_cache = NULL;
1520 #endif
1521         /*
1522          * sigaltstack should be cleared when sharing the same VM
1523          */
1524         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1525                 sas_ss_reset(p);
1526
1527         /*
1528          * Syscall tracing and stepping should be turned off in the
1529          * child regardless of CLONE_PTRACE.
1530          */
1531         user_disable_single_step(p);
1532         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1533 #ifdef TIF_SYSCALL_EMU
1534         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1535 #endif
1536         clear_all_latency_tracing(p);
1537
1538         /* ok, now we should be set up.. */
1539         p->pid = pid_nr(pid);
1540         if (clone_flags & CLONE_THREAD) {
1541                 p->exit_signal = -1;
1542                 p->group_leader = current->group_leader;
1543                 p->tgid = current->tgid;
1544         } else {
1545                 if (clone_flags & CLONE_PARENT)
1546                         p->exit_signal = current->group_leader->exit_signal;
1547                 else
1548                         p->exit_signal = (clone_flags & CSIGNAL);
1549                 p->group_leader = p;
1550                 p->tgid = p->pid;
1551         }
1552
1553         p->nr_dirtied = 0;
1554         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1555         p->dirty_paused_when = 0;
1556
1557         p->pdeath_signal = 0;
1558         INIT_LIST_HEAD(&p->thread_group);
1559         p->task_works = NULL;
1560
1561         /*
1562          * Ensure that the cgroup subsystem policies allow the new process to be
1563          * forked. It should be noted the the new process's css_set can be changed
1564          * between here and cgroup_post_fork() if an organisation operation is in
1565          * progress.
1566          */
1567         retval = cgroup_can_fork(p);
1568         if (retval)
1569                 goto bad_fork_free_pid;
1570
1571         /*
1572          * Make it visible to the rest of the system, but dont wake it up yet.
1573          * Need tasklist lock for parent etc handling!
1574          */
1575         write_lock_irq(&tasklist_lock);
1576
1577         /* CLONE_PARENT re-uses the old parent */
1578         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1579                 p->real_parent = current->real_parent;
1580                 p->parent_exec_id = current->parent_exec_id;
1581         } else {
1582                 p->real_parent = current;
1583                 p->parent_exec_id = current->self_exec_id;
1584         }
1585
1586         spin_lock(&current->sighand->siglock);
1587
1588         /*
1589          * Copy seccomp details explicitly here, in case they were changed
1590          * before holding sighand lock.
1591          */
1592         copy_seccomp(p);
1593
1594         /*
1595          * Process group and session signals need to be delivered to just the
1596          * parent before the fork or both the parent and the child after the
1597          * fork. Restart if a signal comes in before we add the new process to
1598          * it's process group.
1599          * A fatal signal pending means that current will exit, so the new
1600          * thread can't slip out of an OOM kill (or normal SIGKILL).
1601         */
1602         recalc_sigpending();
1603         if (signal_pending(current)) {
1604                 spin_unlock(&current->sighand->siglock);
1605                 write_unlock_irq(&tasklist_lock);
1606                 retval = -ERESTARTNOINTR;
1607                 goto bad_fork_cancel_cgroup;
1608         }
1609
1610         if (likely(p->pid)) {
1611                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1612
1613                 init_task_pid(p, PIDTYPE_PID, pid);
1614                 if (thread_group_leader(p)) {
1615                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1616                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1617
1618                         if (is_child_reaper(pid)) {
1619                                 ns_of_pid(pid)->child_reaper = p;
1620                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1621                         }
1622
1623                         p->signal->leader_pid = pid;
1624                         p->signal->tty = tty_kref_get(current->signal->tty);
1625                         list_add_tail(&p->sibling, &p->real_parent->children);
1626                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1627                         attach_pid(p, PIDTYPE_PGID);
1628                         attach_pid(p, PIDTYPE_SID);
1629                         __this_cpu_inc(process_counts);
1630                 } else {
1631                         current->signal->nr_threads++;
1632                         atomic_inc(&current->signal->live);
1633                         atomic_inc(&current->signal->sigcnt);
1634                         list_add_tail_rcu(&p->thread_group,
1635                                           &p->group_leader->thread_group);
1636                         list_add_tail_rcu(&p->thread_node,
1637                                           &p->signal->thread_head);
1638                 }
1639                 attach_pid(p, PIDTYPE_PID);
1640                 nr_threads++;
1641         }
1642
1643         total_forks++;
1644         spin_unlock(&current->sighand->siglock);
1645         syscall_tracepoint_update(p);
1646         write_unlock_irq(&tasklist_lock);
1647
1648         proc_fork_connector(p);
1649         cgroup_post_fork(p);
1650         threadgroup_change_end(current);
1651         perf_event_fork(p);
1652
1653         trace_task_newtask(p, clone_flags);
1654         uprobe_copy_process(p, clone_flags);
1655
1656         return p;
1657
1658 bad_fork_cancel_cgroup:
1659         cgroup_cancel_fork(p);
1660 bad_fork_free_pid:
1661         if (pid != &init_struct_pid)
1662                 free_pid(pid);
1663 bad_fork_cleanup_thread:
1664         exit_thread(p);
1665 bad_fork_cleanup_io:
1666         if (p->io_context)
1667                 exit_io_context(p);
1668 bad_fork_cleanup_namespaces:
1669         exit_task_namespaces(p);
1670 bad_fork_cleanup_mm:
1671         if (p->mm)
1672                 mmput(p->mm);
1673 bad_fork_cleanup_signal:
1674         if (!(clone_flags & CLONE_THREAD))
1675                 free_signal_struct(p->signal);
1676 bad_fork_cleanup_sighand:
1677         __cleanup_sighand(p->sighand);
1678 bad_fork_cleanup_fs:
1679         exit_fs(p); /* blocking */
1680 bad_fork_cleanup_files:
1681         exit_files(p); /* blocking */
1682 bad_fork_cleanup_semundo:
1683         exit_sem(p);
1684 bad_fork_cleanup_audit:
1685         audit_free(p);
1686 bad_fork_cleanup_perf:
1687         perf_event_free_task(p);
1688 bad_fork_cleanup_policy:
1689 #ifdef CONFIG_NUMA
1690         mpol_put(p->mempolicy);
1691 bad_fork_cleanup_threadgroup_lock:
1692 #endif
1693         threadgroup_change_end(current);
1694         delayacct_tsk_free(p);
1695 bad_fork_cleanup_count:
1696         atomic_dec(&p->cred->user->processes);
1697         exit_creds(p);
1698 bad_fork_free:
1699         free_task(p);
1700 fork_out:
1701         return ERR_PTR(retval);
1702 }
1703
1704 static inline void init_idle_pids(struct pid_link *links)
1705 {
1706         enum pid_type type;
1707
1708         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1709                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1710                 links[type].pid = &init_struct_pid;
1711         }
1712 }
1713
1714 struct task_struct *fork_idle(int cpu)
1715 {
1716         struct task_struct *task;
1717         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
1718                             cpu_to_node(cpu));
1719         if (!IS_ERR(task)) {
1720                 init_idle_pids(task->pids);
1721                 init_idle(task, cpu);
1722         }
1723
1724         return task;
1725 }
1726
1727 /*
1728  *  Ok, this is the main fork-routine.
1729  *
1730  * It copies the process, and if successful kick-starts
1731  * it and waits for it to finish using the VM if required.
1732  */
1733 long _do_fork(unsigned long clone_flags,
1734               unsigned long stack_start,
1735               unsigned long stack_size,
1736               int __user *parent_tidptr,
1737               int __user *child_tidptr,
1738               unsigned long tls)
1739 {
1740         struct task_struct *p;
1741         int trace = 0;
1742         long nr;
1743
1744         /*
1745          * Determine whether and which event to report to ptracer.  When
1746          * called from kernel_thread or CLONE_UNTRACED is explicitly
1747          * requested, no event is reported; otherwise, report if the event
1748          * for the type of forking is enabled.
1749          */
1750         if (!(clone_flags & CLONE_UNTRACED)) {
1751                 if (clone_flags & CLONE_VFORK)
1752                         trace = PTRACE_EVENT_VFORK;
1753                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1754                         trace = PTRACE_EVENT_CLONE;
1755                 else
1756                         trace = PTRACE_EVENT_FORK;
1757
1758                 if (likely(!ptrace_event_enabled(current, trace)))
1759                         trace = 0;
1760         }
1761
1762         p = copy_process(clone_flags, stack_start, stack_size,
1763                          child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
1764         /*
1765          * Do this prior waking up the new thread - the thread pointer
1766          * might get invalid after that point, if the thread exits quickly.
1767          */
1768         if (!IS_ERR(p)) {
1769                 struct completion vfork;
1770                 struct pid *pid;
1771
1772                 trace_sched_process_fork(current, p);
1773
1774                 pid = get_task_pid(p, PIDTYPE_PID);
1775                 nr = pid_vnr(pid);
1776
1777                 if (clone_flags & CLONE_PARENT_SETTID)
1778                         put_user(nr, parent_tidptr);
1779
1780                 if (clone_flags & CLONE_VFORK) {
1781                         p->vfork_done = &vfork;
1782                         init_completion(&vfork);
1783                         get_task_struct(p);
1784                 }
1785
1786                 wake_up_new_task(p);
1787
1788                 /* forking complete and child started to run, tell ptracer */
1789                 if (unlikely(trace))
1790                         ptrace_event_pid(trace, pid);
1791
1792                 if (clone_flags & CLONE_VFORK) {
1793                         if (!wait_for_vfork_done(p, &vfork))
1794                                 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1795                 }
1796
1797                 put_pid(pid);
1798         } else {
1799                 nr = PTR_ERR(p);
1800         }
1801         return nr;
1802 }
1803
1804 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
1805 /* For compatibility with architectures that call do_fork directly rather than
1806  * using the syscall entry points below. */
1807 long do_fork(unsigned long clone_flags,
1808               unsigned long stack_start,
1809               unsigned long stack_size,
1810               int __user *parent_tidptr,
1811               int __user *child_tidptr)
1812 {
1813         return _do_fork(clone_flags, stack_start, stack_size,
1814                         parent_tidptr, child_tidptr, 0);
1815 }
1816 #endif
1817
1818 /*
1819  * Create a kernel thread.
1820  */
1821 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1822 {
1823         return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1824                 (unsigned long)arg, NULL, NULL, 0);
1825 }
1826
1827 #ifdef __ARCH_WANT_SYS_FORK
1828 SYSCALL_DEFINE0(fork)
1829 {
1830 #ifdef CONFIG_MMU
1831         return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
1832 #else
1833         /* can not support in nommu mode */
1834         return -EINVAL;
1835 #endif
1836 }
1837 #endif
1838
1839 #ifdef __ARCH_WANT_SYS_VFORK
1840 SYSCALL_DEFINE0(vfork)
1841 {
1842         return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1843                         0, NULL, NULL, 0);
1844 }
1845 #endif
1846
1847 #ifdef __ARCH_WANT_SYS_CLONE
1848 #ifdef CONFIG_CLONE_BACKWARDS
1849 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1850                  int __user *, parent_tidptr,
1851                  unsigned long, tls,
1852                  int __user *, child_tidptr)
1853 #elif defined(CONFIG_CLONE_BACKWARDS2)
1854 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1855                  int __user *, parent_tidptr,
1856                  int __user *, child_tidptr,
1857                  unsigned long, tls)
1858 #elif defined(CONFIG_CLONE_BACKWARDS3)
1859 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1860                 int, stack_size,
1861                 int __user *, parent_tidptr,
1862                 int __user *, child_tidptr,
1863                 unsigned long, tls)
1864 #else
1865 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1866                  int __user *, parent_tidptr,
1867                  int __user *, child_tidptr,
1868                  unsigned long, tls)
1869 #endif
1870 {
1871         return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
1872 }
1873 #endif
1874
1875 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1876 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1877 #endif
1878
1879 static void sighand_ctor(void *data)
1880 {
1881         struct sighand_struct *sighand = data;
1882
1883         spin_lock_init(&sighand->siglock);
1884         init_waitqueue_head(&sighand->signalfd_wqh);
1885 }
1886
1887 void __init proc_caches_init(void)
1888 {
1889         sighand_cachep = kmem_cache_create("sighand_cache",
1890                         sizeof(struct sighand_struct), 0,
1891                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1892                         SLAB_NOTRACK|SLAB_ACCOUNT, sighand_ctor);
1893         signal_cachep = kmem_cache_create("signal_cache",
1894                         sizeof(struct signal_struct), 0,
1895                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
1896                         NULL);
1897         files_cachep = kmem_cache_create("files_cache",
1898                         sizeof(struct files_struct), 0,
1899                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
1900                         NULL);
1901         fs_cachep = kmem_cache_create("fs_cache",
1902                         sizeof(struct fs_struct), 0,
1903                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
1904                         NULL);
1905         /*
1906          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1907          * whole struct cpumask for the OFFSTACK case. We could change
1908          * this to *only* allocate as much of it as required by the
1909          * maximum number of CPU's we can ever have.  The cpumask_allocation
1910          * is at the end of the structure, exactly for that reason.
1911          */
1912         mm_cachep = kmem_cache_create("mm_struct",
1913                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1914                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
1915                         NULL);
1916         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
1917         mmap_init();
1918         nsproxy_cache_init();
1919 }
1920
1921 /*
1922  * Check constraints on flags passed to the unshare system call.
1923  */
1924 static int check_unshare_flags(unsigned long unshare_flags)
1925 {
1926         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1927                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1928                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1929                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
1930                 return -EINVAL;
1931         /*
1932          * Not implemented, but pretend it works if there is nothing
1933          * to unshare.  Note that unsharing the address space or the
1934          * signal handlers also need to unshare the signal queues (aka
1935          * CLONE_THREAD).
1936          */
1937         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1938                 if (!thread_group_empty(current))
1939                         return -EINVAL;
1940         }
1941         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
1942                 if (atomic_read(&current->sighand->count) > 1)
1943                         return -EINVAL;
1944         }
1945         if (unshare_flags & CLONE_VM) {
1946                 if (!current_is_single_threaded())
1947                         return -EINVAL;
1948         }
1949
1950         return 0;
1951 }
1952
1953 /*
1954  * Unshare the filesystem structure if it is being shared
1955  */
1956 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1957 {
1958         struct fs_struct *fs = current->fs;
1959
1960         if (!(unshare_flags & CLONE_FS) || !fs)
1961                 return 0;
1962
1963         /* don't need lock here; in the worst case we'll do useless copy */
1964         if (fs->users == 1)
1965                 return 0;
1966
1967         *new_fsp = copy_fs_struct(fs);
1968         if (!*new_fsp)
1969                 return -ENOMEM;
1970
1971         return 0;
1972 }
1973
1974 /*
1975  * Unshare file descriptor table if it is being shared
1976  */
1977 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1978 {
1979         struct files_struct *fd = current->files;
1980         int error = 0;
1981
1982         if ((unshare_flags & CLONE_FILES) &&
1983             (fd && atomic_read(&fd->count) > 1)) {
1984                 *new_fdp = dup_fd(fd, &error);
1985                 if (!*new_fdp)
1986                         return error;
1987         }
1988
1989         return 0;
1990 }
1991
1992 /*
1993  * unshare allows a process to 'unshare' part of the process
1994  * context which was originally shared using clone.  copy_*
1995  * functions used by do_fork() cannot be used here directly
1996  * because they modify an inactive task_struct that is being
1997  * constructed. Here we are modifying the current, active,
1998  * task_struct.
1999  */
2000 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2001 {
2002         struct fs_struct *fs, *new_fs = NULL;
2003         struct files_struct *fd, *new_fd = NULL;
2004         struct cred *new_cred = NULL;
2005         struct nsproxy *new_nsproxy = NULL;
2006         int do_sysvsem = 0;
2007         int err;
2008
2009         /*
2010          * If unsharing a user namespace must also unshare the thread group
2011          * and unshare the filesystem root and working directories.
2012          */
2013         if (unshare_flags & CLONE_NEWUSER)
2014                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2015         /*
2016          * If unsharing vm, must also unshare signal handlers.
2017          */
2018         if (unshare_flags & CLONE_VM)
2019                 unshare_flags |= CLONE_SIGHAND;
2020         /*
2021          * If unsharing a signal handlers, must also unshare the signal queues.
2022          */
2023         if (unshare_flags & CLONE_SIGHAND)
2024                 unshare_flags |= CLONE_THREAD;
2025         /*
2026          * If unsharing namespace, must also unshare filesystem information.
2027          */
2028         if (unshare_flags & CLONE_NEWNS)
2029                 unshare_flags |= CLONE_FS;
2030
2031         err = check_unshare_flags(unshare_flags);
2032         if (err)
2033                 goto bad_unshare_out;
2034         /*
2035          * CLONE_NEWIPC must also detach from the undolist: after switching
2036          * to a new ipc namespace, the semaphore arrays from the old
2037          * namespace are unreachable.
2038          */
2039         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2040                 do_sysvsem = 1;
2041         err = unshare_fs(unshare_flags, &new_fs);
2042         if (err)
2043                 goto bad_unshare_out;
2044         err = unshare_fd(unshare_flags, &new_fd);
2045         if (err)
2046                 goto bad_unshare_cleanup_fs;
2047         err = unshare_userns(unshare_flags, &new_cred);
2048         if (err)
2049                 goto bad_unshare_cleanup_fd;
2050         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2051                                          new_cred, new_fs);
2052         if (err)
2053                 goto bad_unshare_cleanup_cred;
2054
2055         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2056                 if (do_sysvsem) {
2057                         /*
2058                          * CLONE_SYSVSEM is equivalent to sys_exit().
2059                          */
2060                         exit_sem(current);
2061                 }
2062                 if (unshare_flags & CLONE_NEWIPC) {
2063                         /* Orphan segments in old ns (see sem above). */
2064                         exit_shm(current);
2065                         shm_init_task(current);
2066                 }
2067
2068                 if (new_nsproxy)
2069                         switch_task_namespaces(current, new_nsproxy);
2070
2071                 task_lock(current);
2072
2073                 if (new_fs) {
2074                         fs = current->fs;
2075                         spin_lock(&fs->lock);
2076                         current->fs = new_fs;
2077                         if (--fs->users)
2078                                 new_fs = NULL;
2079                         else
2080                                 new_fs = fs;
2081                         spin_unlock(&fs->lock);
2082                 }
2083
2084                 if (new_fd) {
2085                         fd = current->files;
2086                         current->files = new_fd;
2087                         new_fd = fd;
2088                 }
2089
2090                 task_unlock(current);
2091
2092                 if (new_cred) {
2093                         /* Install the new user namespace */
2094                         commit_creds(new_cred);
2095                         new_cred = NULL;
2096                 }
2097         }
2098
2099 bad_unshare_cleanup_cred:
2100         if (new_cred)
2101                 put_cred(new_cred);
2102 bad_unshare_cleanup_fd:
2103         if (new_fd)
2104                 put_files_struct(new_fd);
2105
2106 bad_unshare_cleanup_fs:
2107         if (new_fs)
2108                 free_fs_struct(new_fs);
2109
2110 bad_unshare_out:
2111         return err;
2112 }
2113
2114 /*
2115  *      Helper to unshare the files of the current task.
2116  *      We don't want to expose copy_files internals to
2117  *      the exec layer of the kernel.
2118  */
2119
2120 int unshare_files(struct files_struct **displaced)
2121 {
2122         struct task_struct *task = current;
2123         struct files_struct *copy = NULL;
2124         int error;
2125
2126         error = unshare_fd(CLONE_FILES, &copy);
2127         if (error || !copy) {
2128                 *displaced = NULL;
2129                 return error;
2130         }
2131         *displaced = task->files;
2132         task_lock(task);
2133         task->files = copy;
2134         task_unlock(task);
2135         return 0;
2136 }
2137
2138 int sysctl_max_threads(struct ctl_table *table, int write,
2139                        void __user *buffer, size_t *lenp, loff_t *ppos)
2140 {
2141         struct ctl_table t;
2142         int ret;
2143         int threads = max_threads;
2144         int min = MIN_THREADS;
2145         int max = MAX_THREADS;
2146
2147         t = *table;
2148         t.data = &threads;
2149         t.extra1 = &min;
2150         t.extra2 = &max;
2151
2152         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2153         if (ret || !write)
2154                 return ret;
2155
2156         set_max_threads(threads);
2157
2158         return 0;
2159 }