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/sched/autogroup.h>
  16 #include <linux/sched/mm.h>
  17 #include <linux/sched/coredump.h>
  18 #include <linux/sched/user.h>
  19 #include <linux/sched/numa_balancing.h>
  20 #include <linux/sched/stat.h>
  21 #include <linux/sched/task.h>
  22 #include <linux/sched/task_stack.h>
  23 #include <linux/sched/cputime.h>
  24 #include <linux/rtmutex.h>
  25 #include <linux/init.h>
  26 #include <linux/unistd.h>
  27 #include <linux/module.h>
  28 #include <linux/vmalloc.h>
  29 #include <linux/completion.h>
  30 #include <linux/personality.h>
  31 #include <linux/mempolicy.h>
  32 #include <linux/sem.h>
  33 #include <linux/file.h>
  34 #include <linux/fdtable.h>
  35 #include <linux/iocontext.h>
  36 #include <linux/key.h>
  37 #include <linux/binfmts.h>
  38 #include <linux/mman.h>
  39 #include <linux/mmu_notifier.h>
  40 #include <linux/hmm.h>
  41 #include <linux/fs.h>
  42 #include <linux/mm.h>
  43 #include <linux/vmacache.h>
  44 #include <linux/nsproxy.h>
  45 #include <linux/capability.h>
  46 #include <linux/cpu.h>
  47 #include <linux/cgroup.h>
  48 #include <linux/security.h>
  49 #include <linux/hugetlb.h>
  50 #include <linux/seccomp.h>
  51 #include <linux/swap.h>
  52 #include <linux/syscalls.h>
  53 #include <linux/jiffies.h>
  54 #include <linux/futex.h>
  55 #include <linux/compat.h>
  56 #include <linux/kthread.h>
  57 #include <linux/task_io_accounting_ops.h>
  58 #include <linux/rcupdate.h>
  59 #include <linux/ptrace.h>
  60 #include <linux/mount.h>
  61 #include <linux/audit.h>
  62 #include <linux/memcontrol.h>
  63 #include <linux/ftrace.h>
  64 #include <linux/proc_fs.h>
  65 #include <linux/profile.h>
  66 #include <linux/rmap.h>
  67 #include <linux/ksm.h>
  68 #include <linux/acct.h>
  69 #include <linux/userfaultfd_k.h>
  70 #include <linux/tsacct_kern.h>
  71 #include <linux/cn_proc.h>
  72 #include <linux/freezer.h>
  73 #include <linux/delayacct.h>
  74 #include <linux/taskstats_kern.h>
  75 #include <linux/random.h>
  76 #include <linux/tty.h>
  77 #include <linux/blkdev.h>
  78 #include <linux/fs_struct.h>
  79 #include <linux/magic.h>
  80 #include <linux/perf_event.h>
  81 #include <linux/posix-timers.h>
  82 #include <linux/user-return-notifier.h>
  83 #include <linux/oom.h>
  84 #include <linux/khugepaged.h>
  85 #include <linux/signalfd.h>
  86 #include <linux/uprobes.h>
  87 #include <linux/aio.h>
  88 #include <linux/compiler.h>
  89 #include <linux/sysctl.h>
  90 #include <linux/kcov.h>
  91 #include <linux/livepatch.h>
  92 #include <linux/thread_info.h>
  93
  94 #include <asm/pgtable.h>
  95 #include <asm/pgalloc.h>
  96 #include <linux/uaccess.h>
  97 #include <asm/mmu_context.h>
  98 #include <asm/cacheflush.h>
  99 #include <asm/tlbflush.h>
 100
 101 #include <trace/events/sched.h>
 102
 103 #define CREATE_TRACE_POINTS
 104 #include <trace/events/task.h>
 105
 106 /*
 107  * Minimum number of threads to boot the kernel
 108  */
 109 #define MIN_THREADS 20
 110
 111 /*
 112  * Maximum number of threads
 113  */
 114 #define MAX_THREADS FUTEX_TID_MASK
 115
 116 /*
 117  * Protected counters by write_lock_irq(&tasklist_lock)
 118  */
 119 unsigned long total_forks;      /* Handle normal Linux uptimes. */
 120 int nr_threads;                 /* The idle threads do not count.. */
 121
 122 int max_threads;                /* tunable limit on nr_threads */
 123
 124 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 125
 126 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 127
 128 #ifdef CONFIG_PROVE_RCU
 129 int lockdep_tasklist_lock_is_held(void)
 130 {
 131         return lockdep_is_held(&tasklist_lock);
 132 }
 133 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 134 #endif /* #ifdef CONFIG_PROVE_RCU */
 135
 136 int nr_processes(void)
 137 {
 138         int cpu;
 139         int total = 0;
 140
 141         for_each_possible_cpu(cpu)
 142                 total += per_cpu(process_counts, cpu);
 143
 144         return total;
 145 }
 146
 147 void __weak arch_release_task_struct(struct task_struct *tsk)
 148 {
 149 }
 150
 151 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 152 static struct kmem_cache *task_struct_cachep;
 153
 154 static inline struct task_struct *alloc_task_struct_node(int node)
 155 {
 156         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 157 }
 158
 159 static inline void free_task_struct(struct task_struct *tsk)
 160 {
 161         kmem_cache_free(task_struct_cachep, tsk);
 162 }
 163 #endif
 164
 165 void __weak arch_release_thread_stack(unsigned long *stack)
 166 {
 167 }
 168
 169 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
 170
 171 /*
 172  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 173  * kmemcache based allocator.
 174  */
 175 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
 176
 177 #ifdef CONFIG_VMAP_STACK
 178 /*
 179  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
 180  * flush.  Try to minimize the number of calls by caching stacks.
 181  */
 182 #define NR_CACHED_STACKS 2
 183 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
 184
 185 static int free_vm_stack_cache(unsigned int cpu)
 186 {
 187         struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
 188         int i;
 189
 190         for (i = 0; i < NR_CACHED_STACKS; i++) {
 191                 struct vm_struct *vm_stack = cached_vm_stacks[i];
 192
 193                 if (!vm_stack)
 194                         continue;
 195
 196                 vfree(vm_stack->addr);
 197                 cached_vm_stacks[i] = NULL;
 198         }
 199
 200         return 0;
 201 }
 202 #endif
 203
 204 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
 205 {
 206 #ifdef CONFIG_VMAP_STACK
 207         void *stack;
 208         int i;
 209
 210         for (i = 0; i < NR_CACHED_STACKS; i++) {
 211                 struct vm_struct *s;
 212
 213                 s = this_cpu_xchg(cached_stacks[i], NULL);
 214
 215                 if (!s)
 216                         continue;
 217
 218                 /* Clear stale pointers from reused stack. */
 219                 memset(s->addr, 0, THREAD_SIZE);
 220
 221                 tsk->stack_vm_area = s;
 222                 return s->addr;
 223         }
 224
 225         stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
 226                                      VMALLOC_START, VMALLOC_END,
 227                                      THREADINFO_GFP,
 228                                      PAGE_KERNEL,
 229                                      0, node, __builtin_return_address(0));
 230
 231         /*
 232          * We can't call find_vm_area() in interrupt context, and
 233          * free_thread_stack() can be called in interrupt context,
 234          * so cache the vm_struct.
 235          */
 236         if (stack)
 237                 tsk->stack_vm_area = find_vm_area(stack);
 238         return stack;
 239 #else
 240         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
 241                                              THREAD_SIZE_ORDER);
 242
 243         return page ? page_address(page) : NULL;
 244 #endif
 245 }
 246
 247 static inline void free_thread_stack(struct task_struct *tsk)
 248 {
 249 #ifdef CONFIG_VMAP_STACK
 250         if (task_stack_vm_area(tsk)) {
 251                 int i;
 252
 253                 for (i = 0; i < NR_CACHED_STACKS; i++) {
 254                         if (this_cpu_cmpxchg(cached_stacks[i],
 255                                         NULL, tsk->stack_vm_area) != NULL)
 256                                 continue;
 257
 258                         return;
 259                 }
 260
 261                 vfree_atomic(tsk->stack);
 262                 return;
 263         }
 264 #endif
 265
 266         __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
 267 }
 268 # else
 269 static struct kmem_cache *thread_stack_cache;
 270
 271 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
 272                                                   int node)
 273 {
 274         return kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
 275 }
 276
 277 static void free_thread_stack(struct task_struct *tsk)
 278 {
 279         kmem_cache_free(thread_stack_cache, tsk->stack);
 280 }
 281
 282 void thread_stack_cache_init(void)
 283 {
 284         thread_stack_cache = kmem_cache_create("thread_stack", THREAD_SIZE,
 285                                               THREAD_SIZE, 0, NULL);
 286         BUG_ON(thread_stack_cache == NULL);
 287 }
 288 # endif
 289 #endif
 290
 291 /* SLAB cache for signal_struct structures (tsk->signal) */
 292 static struct kmem_cache *signal_cachep;
 293
 294 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 295 struct kmem_cache *sighand_cachep;
 296
 297 /* SLAB cache for files_struct structures (tsk->files) */
 298 struct kmem_cache *files_cachep;
 299
 300 /* SLAB cache for fs_struct structures (tsk->fs) */
 301 struct kmem_cache *fs_cachep;
 302
 303 /* SLAB cache for vm_area_struct structures */
 304 struct kmem_cache *vm_area_cachep;
 305
 306 /* SLAB cache for mm_struct structures (tsk->mm) */
 307 static struct kmem_cache *mm_cachep;
 308
 309 static void account_kernel_stack(struct task_struct *tsk, int account)
 310 {
 311         void *stack = task_stack_page(tsk);
 312         struct vm_struct *vm = task_stack_vm_area(tsk);
 313
 314         BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
 315
 316         if (vm) {
 317                 int i;
 318
 319                 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
 320
 321                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 322                         mod_zone_page_state(page_zone(vm->pages[i]),
 323                                             NR_KERNEL_STACK_KB,
 324                                             PAGE_SIZE / 1024 * account);
 325                 }
 326
 327                 /* All stack pages belong to the same memcg. */
 328                 mod_memcg_page_state(vm->pages[0], MEMCG_KERNEL_STACK_KB,
 329                                      account * (THREAD_SIZE / 1024));
 330         } else {
 331                 /*
 332                  * All stack pages are in the same zone and belong to the
 333                  * same memcg.
 334                  */
 335                 struct page *first_page = virt_to_page(stack);
 336
 337                 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
 338                                     THREAD_SIZE / 1024 * account);
 339
 340                 mod_memcg_page_state(first_page, MEMCG_KERNEL_STACK_KB,
 341                                      account * (THREAD_SIZE / 1024));
 342         }
 343 }
 344
 345 static void release_task_stack(struct task_struct *tsk)
 346 {
 347         if (WARN_ON(tsk->state != TASK_DEAD))
 348                 return;  /* Better to leak the stack than to free prematurely */
 349
 350         account_kernel_stack(tsk, -1);
 351         arch_release_thread_stack(tsk->stack);
 352         free_thread_stack(tsk);
 353         tsk->stack = NULL;
 354 #ifdef CONFIG_VMAP_STACK
 355         tsk->stack_vm_area = NULL;
 356 #endif
 357 }
 358
 359 #ifdef CONFIG_THREAD_INFO_IN_TASK
 360 void put_task_stack(struct task_struct *tsk)
 361 {
 362         if (atomic_dec_and_test(&tsk->stack_refcount))
 363                 release_task_stack(tsk);
 364 }
 365 #endif
 366
 367 void free_task(struct task_struct *tsk)
 368 {
 369 #ifndef CONFIG_THREAD_INFO_IN_TASK
 370         /*
 371          * The task is finally done with both the stack and thread_info,
 372          * so free both.
 373          */
 374         release_task_stack(tsk);
 375 #else
 376         /*
 377          * If the task had a separate stack allocation, it should be gone
 378          * by now.
 379          */
 380         WARN_ON_ONCE(atomic_read(&tsk->stack_refcount) != 0);
 381 #endif
 382         rt_mutex_debug_task_free(tsk);
 383         ftrace_graph_exit_task(tsk);
 384         put_seccomp_filter(tsk);
 385         arch_release_task_struct(tsk);
 386         if (tsk->flags & PF_KTHREAD)
 387                 free_kthread_struct(tsk);
 388         free_task_struct(tsk);
 389 }
 390 EXPORT_SYMBOL(free_task);
 391
 392 static inline void free_signal_struct(struct signal_struct *sig)
 393 {
 394         taskstats_tgid_free(sig);
 395         sched_autogroup_exit(sig);
 396         /*
 397          * __mmdrop is not safe to call from softirq context on x86 due to
 398          * pgd_dtor so postpone it to the async context
 399          */
 400         if (sig->oom_mm)
 401                 mmdrop_async(sig->oom_mm);
 402         kmem_cache_free(signal_cachep, sig);
 403 }
 404
 405 static inline void put_signal_struct(struct signal_struct *sig)
 406 {
 407         if (atomic_dec_and_test(&sig->sigcnt))
 408                 free_signal_struct(sig);
 409 }
 410
 411 void __put_task_struct(struct task_struct *tsk)
 412 {
 413         WARN_ON(!tsk->exit_state);
 414         WARN_ON(atomic_read(&tsk->usage));
 415         WARN_ON(tsk == current);
 416
 417         cgroup_free(tsk);
 418         task_numa_free(tsk);
 419         security_task_free(tsk);
 420         exit_creds(tsk);
 421         delayacct_tsk_free(tsk);
 422         put_signal_struct(tsk->signal);
 423
 424         if (!profile_handoff_task(tsk))
 425                 free_task(tsk);
 426 }
 427 EXPORT_SYMBOL_GPL(__put_task_struct);
 428
 429 void __init __weak arch_task_cache_init(void) { }
 430
 431 /*
 432  * set_max_threads
 433  */
 434 static void set_max_threads(unsigned int max_threads_suggested)
 435 {
 436         u64 threads;
 437
 438         /*
 439          * The number of threads shall be limited such that the thread
 440          * structures may only consume a small part of the available memory.
 441          */
 442         if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
 443                 threads = MAX_THREADS;
 444         else
 445                 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
 446                                     (u64) THREAD_SIZE * 8UL);
 447
 448         if (threads > max_threads_suggested)
 449                 threads = max_threads_suggested;
 450
 451         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 452 }
 453
 454 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 455 /* Initialized by the architecture: */
 456 int arch_task_struct_size __read_mostly;
 457 #endif
 458
 459 void __init fork_init(void)
 460 {
 461         int i;
 462 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 463 #ifndef ARCH_MIN_TASKALIGN
 464 #define ARCH_MIN_TASKALIGN      0
 465 #endif
 466         int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
 467
 468         /* create a slab on which task_structs can be allocated */
 469         task_struct_cachep = kmem_cache_create("task_struct",
 470                         arch_task_struct_size, align,
 471                         SLAB_PANIC|SLAB_ACCOUNT, NULL);
 472 #endif
 473
 474         /* do the arch specific task caches init */
 475         arch_task_cache_init();
 476
 477         set_max_threads(MAX_THREADS);
 478
 479         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 480         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 481         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 482                 init_task.signal->rlim[RLIMIT_NPROC];
 483
 484         for (i = 0; i < UCOUNT_COUNTS; i++) {
 485                 init_user_ns.ucount_max[i] = max_threads/2;
 486         }
 487
 488 #ifdef CONFIG_VMAP_STACK
 489         cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
 490                           NULL, free_vm_stack_cache);
 491 #endif
 492
 493         lockdep_init_task(&init_task);
 494 }
 495
 496 int __weak arch_dup_task_struct(struct task_struct *dst,
 497                                                struct task_struct *src)
 498 {
 499         *dst = *src;
 500         return 0;
 501 }
 502
 503 void set_task_stack_end_magic(struct task_struct *tsk)
 504 {
 505         unsigned long *stackend;
 506
 507         stackend = end_of_stack(tsk);
 508         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 509 }
 510
 511 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 512 {
 513         struct task_struct *tsk;
 514         unsigned long *stack;
 515         struct vm_struct *stack_vm_area;
 516         int err;
 517
 518         if (node == NUMA_NO_NODE)
 519                 node = tsk_fork_get_node(orig);
 520         tsk = alloc_task_struct_node(node);
 521         if (!tsk)
 522                 return NULL;
 523
 524         stack = alloc_thread_stack_node(tsk, node);
 525         if (!stack)
 526                 goto free_tsk;
 527
 528         stack_vm_area = task_stack_vm_area(tsk);
 529
 530         err = arch_dup_task_struct(tsk, orig);
 531
 532         /*
 533          * arch_dup_task_struct() clobbers the stack-related fields.  Make
 534          * sure they're properly initialized before using any stack-related
 535          * functions again.
 536          */
 537         tsk->stack = stack;
 538 #ifdef CONFIG_VMAP_STACK
 539         tsk->stack_vm_area = stack_vm_area;
 540 #endif
 541 #ifdef CONFIG_THREAD_INFO_IN_TASK
 542         atomic_set(&tsk->stack_refcount, 1);
 543 #endif
 544
 545         if (err)
 546                 goto free_stack;
 547
 548 #ifdef CONFIG_SECCOMP
 549         /*
 550          * We must handle setting up seccomp filters once we're under
 551          * the sighand lock in case orig has changed between now and
 552          * then. Until then, filter must be NULL to avoid messing up
 553          * the usage counts on the error path calling free_task.
 554          */
 555         tsk->seccomp.filter = NULL;
 556 #endif
 557
 558         setup_thread_stack(tsk, orig);
 559         clear_user_return_notifier(tsk);
 560         clear_tsk_need_resched(tsk);
 561         set_task_stack_end_magic(tsk);
 562
 563 #ifdef CONFIG_CC_STACKPROTECTOR
 564         tsk->stack_canary = get_random_canary();
 565 #endif
 566
 567         /*
 568          * One for us, one for whoever does the "release_task()" (usually
 569          * parent)
 570          */
 571         atomic_set(&tsk->usage, 2);
 572 #ifdef CONFIG_BLK_DEV_IO_TRACE
 573         tsk->btrace_seq = 0;
 574 #endif
 575         tsk->splice_pipe = NULL;
 576         tsk->task_frag.page = NULL;
 577         tsk->wake_q.next = NULL;
 578
 579         account_kernel_stack(tsk, 1);
 580
 581         kcov_task_init(tsk);
 582
 583 #ifdef CONFIG_FAULT_INJECTION
 584         tsk->fail_nth = 0;
 585 #endif
 586
 587         return tsk;
 588
 589 free_stack:
 590         free_thread_stack(tsk);
 591 free_tsk:
 592         free_task_struct(tsk);
 593         return NULL;
 594 }
 595
 596 #ifdef CONFIG_MMU
 597 static __latent_entropy int dup_mmap(struct mm_struct *mm,
 598                                         struct mm_struct *oldmm)
 599 {
 600         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 601         struct rb_node **rb_link, *rb_parent;
 602         int retval;
 603         unsigned long charge;
 604         LIST_HEAD(uf);
 605
 606         uprobe_start_dup_mmap();
 607         if (down_write_killable(&oldmm->mmap_sem)) {
 608                 retval = -EINTR;
 609                 goto fail_uprobe_end;
 610         }
 611         flush_cache_dup_mm(oldmm);
 612         uprobe_dup_mmap(oldmm, mm);
 613         /*
 614          * Not linked in yet - no deadlock potential:
 615          */
 616         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 617
 618         /* No ordering required: file already has been exposed. */
 619         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 620
 621         mm->total_vm = oldmm->total_vm;
 622         mm->data_vm = oldmm->data_vm;
 623         mm->exec_vm = oldmm->exec_vm;
 624         mm->stack_vm = oldmm->stack_vm;
 625
 626         rb_link = &mm->mm_rb.rb_node;
 627         rb_parent = NULL;
 628         pprev = &mm->mmap;
 629         retval = ksm_fork(mm, oldmm);
 630         if (retval)
 631                 goto out;
 632         retval = khugepaged_fork(mm, oldmm);
 633         if (retval)
 634                 goto out;
 635
 636         prev = NULL;
 637         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 638                 struct file *file;
 639
 640                 if (mpnt->vm_flags & VM_DONTCOPY) {
 641                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 642                         continue;
 643                 }
 644                 charge = 0;
 645                 if (mpnt->vm_flags & VM_ACCOUNT) {
 646                         unsigned long len = vma_pages(mpnt);
 647
 648                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 649                                 goto fail_nomem;
 650                         charge = len;
 651                 }
 652                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 653                 if (!tmp)
 654                         goto fail_nomem;
 655                 *tmp = *mpnt;
 656                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 657                 retval = vma_dup_policy(mpnt, tmp);
 658                 if (retval)
 659                         goto fail_nomem_policy;
 660                 tmp->vm_mm = mm;
 661                 retval = dup_userfaultfd(tmp, &uf);
 662                 if (retval)
 663                         goto fail_nomem_anon_vma_fork;
 664                 if (tmp->vm_flags & VM_WIPEONFORK) {
 665                         /* VM_WIPEONFORK gets a clean slate in the child. */
 666                         tmp->anon_vma = NULL;
 667                         if (anon_vma_prepare(tmp))
 668                                 goto fail_nomem_anon_vma_fork;
 669                 } else if (anon_vma_fork(tmp, mpnt))
 670                         goto fail_nomem_anon_vma_fork;
 671                 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
 672                 tmp->vm_next = tmp->vm_prev = NULL;
 673                 file = tmp->vm_file;
 674                 if (file) {
 675                         struct inode *inode = file_inode(file);
 676                         struct address_space *mapping = file->f_mapping;
 677
 678                         get_file(file);
 679                         if (tmp->vm_flags & VM_DENYWRITE)
 680                                 atomic_dec(&inode->i_writecount);
 681                         i_mmap_lock_write(mapping);
 682                         if (tmp->vm_flags & VM_SHARED)
 683                                 atomic_inc(&mapping->i_mmap_writable);
 684                         flush_dcache_mmap_lock(mapping);
 685                         /* insert tmp into the share list, just after mpnt */
 686                         vma_interval_tree_insert_after(tmp, mpnt,
 687                                         &mapping->i_mmap);
 688                         flush_dcache_mmap_unlock(mapping);
 689                         i_mmap_unlock_write(mapping);
 690                 }
 691
 692                 /*
 693                  * Clear hugetlb-related page reserves for children. This only
 694                  * affects MAP_PRIVATE mappings. Faults generated by the child
 695                  * are not guaranteed to succeed, even if read-only
 696                  */
 697                 if (is_vm_hugetlb_page(tmp))
 698                         reset_vma_resv_huge_pages(tmp);
 699
 700                 /*
 701                  * Link in the new vma and copy the page table entries.
 702                  */
 703                 *pprev = tmp;
 704                 pprev = &tmp->vm_next;
 705                 tmp->vm_prev = prev;
 706                 prev = tmp;
 707
 708                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 709                 rb_link = &tmp->vm_rb.rb_right;
 710                 rb_parent = &tmp->vm_rb;
 711
 712                 mm->map_count++;
 713                 if (!(tmp->vm_flags & VM_WIPEONFORK))
 714                         retval = copy_page_range(mm, oldmm, mpnt);
 715
 716                 if (tmp->vm_ops && tmp->vm_ops->open)
 717                         tmp->vm_ops->open(tmp);
 718
 719                 if (retval)
 720                         goto out;
 721         }
 722         /* a new mm has just been created */
 723         retval = arch_dup_mmap(oldmm, mm);
 724 out:
 725         up_write(&mm->mmap_sem);
 726         flush_tlb_mm(oldmm);
 727         up_write(&oldmm->mmap_sem);
 728         dup_userfaultfd_complete(&uf);
 729 fail_uprobe_end:
 730         uprobe_end_dup_mmap();
 731         return retval;
 732 fail_nomem_anon_vma_fork:
 733         mpol_put(vma_policy(tmp));
 734 fail_nomem_policy:
 735         kmem_cache_free(vm_area_cachep, tmp);
 736 fail_nomem:
 737         retval = -ENOMEM;
 738         vm_unacct_memory(charge);
 739         goto out;
 740 }
 741
 742 static inline int mm_alloc_pgd(struct mm_struct *mm)
 743 {
 744         mm->pgd = pgd_alloc(mm);
 745         if (unlikely(!mm->pgd))
 746                 return -ENOMEM;
 747         return 0;
 748 }
 749
 750 static inline void mm_free_pgd(struct mm_struct *mm)
 751 {
 752         pgd_free(mm, mm->pgd);
 753 }
 754 #else
 755 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 756 {
 757         down_write(&oldmm->mmap_sem);
 758         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 759         up_write(&oldmm->mmap_sem);
 760         return 0;
 761 }
 762 #define mm_alloc_pgd(mm)        (0)
 763 #define mm_free_pgd(mm)
 764 #endif /* CONFIG_MMU */
 765
 766 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 767
 768 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 769 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 770
 771 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 772
 773 static int __init coredump_filter_setup(char *s)
 774 {
 775         default_dump_filter =
 776                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 777                 MMF_DUMP_FILTER_MASK;
 778         return 1;
 779 }
 780
 781 __setup("coredump_filter=", coredump_filter_setup);
 782
 783 #include <linux/init_task.h>
 784
 785 static void mm_init_aio(struct mm_struct *mm)
 786 {
 787 #ifdef CONFIG_AIO
 788         spin_lock_init(&mm->ioctx_lock);
 789         mm->ioctx_table = NULL;
 790 #endif
 791 }
 792
 793 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 794 {
 795 #ifdef CONFIG_MEMCG
 796         mm->owner = p;
 797 #endif
 798 }
 799
 800 static void mm_init_uprobes_state(struct mm_struct *mm)
 801 {
 802 #ifdef CONFIG_UPROBES
 803         mm->uprobes_state.xol_area = NULL;
 804 #endif
 805 }
 806
 807 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
 808         struct user_namespace *user_ns)
 809 {
 810         mm->mmap = NULL;
 811         mm->mm_rb = RB_ROOT;
 812         mm->vmacache_seqnum = 0;
 813         atomic_set(&mm->mm_users, 1);
 814         atomic_set(&mm->mm_count, 1);
 815         init_rwsem(&mm->mmap_sem);
 816         INIT_LIST_HEAD(&mm->mmlist);
 817         mm->core_state = NULL;
 818         atomic_long_set(&mm->nr_ptes, 0);
 819         mm_nr_pmds_init(mm);
 820         mm->map_count = 0;
 821         mm->locked_vm = 0;
 822         mm->pinned_vm = 0;
 823         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 824         spin_lock_init(&mm->page_table_lock);
 825         mm_init_cpumask(mm);
 826         mm_init_aio(mm);
 827         mm_init_owner(mm, p);
 828         RCU_INIT_POINTER(mm->exe_file, NULL);
 829         mmu_notifier_mm_init(mm);
 830         hmm_mm_init(mm);
 831         init_tlb_flush_pending(mm);
 832 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 833         mm->pmd_huge_pte = NULL;
 834 #endif
 835         mm_init_uprobes_state(mm);
 836
 837         if (current->mm) {
 838                 mm->flags = current->mm->flags & MMF_INIT_MASK;
 839                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
 840         } else {
 841                 mm->flags = default_dump_filter;
 842                 mm->def_flags = 0;
 843         }
 844
 845         if (mm_alloc_pgd(mm))
 846                 goto fail_nopgd;
 847
 848         if (init_new_context(p, mm))
 849                 goto fail_nocontext;
 850
 851         mm->user_ns = get_user_ns(user_ns);
 852         return mm;
 853
 854 fail_nocontext:
 855         mm_free_pgd(mm);
 856 fail_nopgd:
 857         free_mm(mm);
 858         return NULL;
 859 }
 860
 861 static void check_mm(struct mm_struct *mm)
 862 {
 863         int i;
 864
 865         for (i = 0; i < NR_MM_COUNTERS; i++) {
 866                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 867
 868                 if (unlikely(x))
 869                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 870                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 871         }
 872
 873         if (atomic_long_read(&mm->nr_ptes))
 874                 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
 875                                 atomic_long_read(&mm->nr_ptes));
 876         if (mm_nr_pmds(mm))
 877                 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
 878                                 mm_nr_pmds(mm));
 879
 880 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 881         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 882 #endif
 883 }
 884
 885 /*
 886  * Allocate and initialize an mm_struct.
 887  */
 888 struct mm_struct *mm_alloc(void)
 889 {
 890         struct mm_struct *mm;
 891
 892         mm = allocate_mm();
 893         if (!mm)
 894                 return NULL;
 895
 896         memset(mm, 0, sizeof(*mm));
 897         return mm_init(mm, current, current_user_ns());
 898 }
 899
 900 /*
 901  * Called when the last reference to the mm
 902  * is dropped: either by a lazy thread or by
 903  * mmput. Free the page directory and the mm.
 904  */
 905 void __mmdrop(struct mm_struct *mm)
 906 {
 907         BUG_ON(mm == &init_mm);
 908         mm_free_pgd(mm);
 909         destroy_context(mm);
 910         hmm_mm_destroy(mm);
 911         mmu_notifier_mm_destroy(mm);
 912         check_mm(mm);
 913         put_user_ns(mm->user_ns);
 914         free_mm(mm);
 915 }
 916 EXPORT_SYMBOL_GPL(__mmdrop);
 917
 918 static inline void __mmput(struct mm_struct *mm)
 919 {
 920         VM_BUG_ON(atomic_read(&mm->mm_users));
 921
 922         uprobe_clear_state(mm);
 923         exit_aio(mm);
 924         ksm_exit(mm);
 925         khugepaged_exit(mm); /* must run before exit_mmap */
 926         exit_mmap(mm);
 927         mm_put_huge_zero_page(mm);
 928         set_mm_exe_file(mm, NULL);
 929         if (!list_empty(&mm->mmlist)) {
 930                 spin_lock(&mmlist_lock);
 931                 list_del(&mm->mmlist);
 932                 spin_unlock(&mmlist_lock);
 933         }
 934         if (mm->binfmt)
 935                 module_put(mm->binfmt->module);
 936         mmdrop(mm);
 937 }
 938
 939 /*
 940  * Decrement the use count and release all resources for an mm.
 941  */
 942 void mmput(struct mm_struct *mm)
 943 {
 944         might_sleep();
 945
 946         if (atomic_dec_and_test(&mm->mm_users))
 947                 __mmput(mm);
 948 }
 949 EXPORT_SYMBOL_GPL(mmput);
 950
 951 #ifdef CONFIG_MMU
 952 static void mmput_async_fn(struct work_struct *work)
 953 {
 954         struct mm_struct *mm = container_of(work, struct mm_struct,
 955                                             async_put_work);
 956
 957         __mmput(mm);
 958 }
 959
 960 void mmput_async(struct mm_struct *mm)
 961 {
 962         if (atomic_dec_and_test(&mm->mm_users)) {
 963                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
 964                 schedule_work(&mm->async_put_work);
 965         }
 966 }
 967 #endif
 968
 969 /**
 970  * set_mm_exe_file - change a reference to the mm's executable file
 971  *
 972  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
 973  *
 974  * Main users are mmput() and sys_execve(). Callers prevent concurrent
 975  * invocations: in mmput() nobody alive left, in execve task is single
 976  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
 977  * mm->exe_file, but does so without using set_mm_exe_file() in order
 978  * to do avoid the need for any locks.
 979  */
 980 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 981 {
 982         struct file *old_exe_file;
 983
 984         /*
 985          * It is safe to dereference the exe_file without RCU as
 986          * this function is only called if nobody else can access
 987          * this mm -- see comment above for justification.
 988          */
 989         old_exe_file = rcu_dereference_raw(mm->exe_file);
 990
 991         if (new_exe_file)
 992                 get_file(new_exe_file);
 993         rcu_assign_pointer(mm->exe_file, new_exe_file);
 994         if (old_exe_file)
 995                 fput(old_exe_file);
 996 }
 997
 998 /**
 999  * get_mm_exe_file - acquire a reference to the mm's executable file
1000  *
1001  * Returns %NULL if mm has no associated executable file.
1002  * User must release file via fput().
1003  */
1004 struct file *get_mm_exe_file(struct mm_struct *mm)
1005 {
1006         struct file *exe_file;
1007
1008         rcu_read_lock();
1009         exe_file = rcu_dereference(mm->exe_file);
1010         if (exe_file && !get_file_rcu(exe_file))
1011                 exe_file = NULL;
1012         rcu_read_unlock();
1013         return exe_file;
1014 }
1015 EXPORT_SYMBOL(get_mm_exe_file);
1016
1017 /**
1018  * get_task_exe_file - acquire a reference to the task's executable file
1019  *
1020  * Returns %NULL if task's mm (if any) has no associated executable file or
1021  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1022  * User must release file via fput().
1023  */
1024 struct file *get_task_exe_file(struct task_struct *task)
1025 {
1026         struct file *exe_file = NULL;
1027         struct mm_struct *mm;
1028
1029         task_lock(task);
1030         mm = task->mm;
1031         if (mm) {
1032                 if (!(task->flags & PF_KTHREAD))
1033                         exe_file = get_mm_exe_file(mm);
1034         }
1035         task_unlock(task);
1036         return exe_file;
1037 }
1038 EXPORT_SYMBOL(get_task_exe_file);
1039
1040 /**
1041  * get_task_mm - acquire a reference to the task's mm
1042  *
1043  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1044  * this kernel workthread has transiently adopted a user mm with use_mm,
1045  * to do its AIO) is not set and if so returns a reference to it, after
1046  * bumping up the use count.  User must release the mm via mmput()
1047  * after use.  Typically used by /proc and ptrace.
1048  */
1049 struct mm_struct *get_task_mm(struct task_struct *task)
1050 {
1051         struct mm_struct *mm;
1052
1053         task_lock(task);
1054         mm = task->mm;
1055         if (mm) {
1056                 if (task->flags & PF_KTHREAD)
1057                         mm = NULL;
1058                 else
1059                         mmget(mm);
1060         }
1061         task_unlock(task);
1062         return mm;
1063 }
1064 EXPORT_SYMBOL_GPL(get_task_mm);
1065
1066 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1067 {
1068         struct mm_struct *mm;
1069         int err;
1070
1071         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
1072         if (err)
1073                 return ERR_PTR(err);
1074
1075         mm = get_task_mm(task);
1076         if (mm && mm != current->mm &&
1077                         !ptrace_may_access(task, mode)) {
1078                 mmput(mm);
1079                 mm = ERR_PTR(-EACCES);
1080         }
1081         mutex_unlock(&task->signal->cred_guard_mutex);
1082
1083         return mm;
1084 }
1085
1086 static void complete_vfork_done(struct task_struct *tsk)
1087 {
1088         struct completion *vfork;
1089
1090         task_lock(tsk);
1091         vfork = tsk->vfork_done;
1092         if (likely(vfork)) {
1093                 tsk->vfork_done = NULL;
1094                 complete(vfork);
1095         }
1096         task_unlock(tsk);
1097 }
1098
1099 static int wait_for_vfork_done(struct task_struct *child,
1100                                 struct completion *vfork)
1101 {
1102         int killed;
1103
1104         freezer_do_not_count();
1105         killed = wait_for_completion_killable(vfork);
1106         freezer_count();
1107
1108         if (killed) {
1109                 task_lock(child);
1110                 child->vfork_done = NULL;
1111                 task_unlock(child);
1112         }
1113
1114         put_task_struct(child);
1115         return killed;
1116 }
1117
1118 /* Please note the differences between mmput and mm_release.
1119  * mmput is called whenever we stop holding onto a mm_struct,
1120  * error success whatever.
1121  *
1122  * mm_release is called after a mm_struct has been removed
1123  * from the current process.
1124  *
1125  * This difference is important for error handling, when we
1126  * only half set up a mm_struct for a new process and need to restore
1127  * the old one.  Because we mmput the new mm_struct before
1128  * restoring the old one. . .
1129  * Eric Biederman 10 January 1998
1130  */
1131 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1132 {
1133         /* Get rid of any futexes when releasing the mm */
1134 #ifdef CONFIG_FUTEX
1135         if (unlikely(tsk->robust_list)) {
1136                 exit_robust_list(tsk);
1137                 tsk->robust_list = NULL;
1138         }
1139 #ifdef CONFIG_COMPAT
1140         if (unlikely(tsk->compat_robust_list)) {
1141                 compat_exit_robust_list(tsk);
1142                 tsk->compat_robust_list = NULL;
1143         }
1144 #endif
1145         if (unlikely(!list_empty(&tsk->pi_state_list)))
1146                 exit_pi_state_list(tsk);
1147 #endif
1148
1149         uprobe_free_utask(tsk);
1150
1151         /* Get rid of any cached register state */
1152         deactivate_mm(tsk, mm);
1153
1154         /*
1155          * Signal userspace if we're not exiting with a core dump
1156          * because we want to leave the value intact for debugging
1157          * purposes.
1158          */
1159         if (tsk->clear_child_tid) {
1160                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1161                     atomic_read(&mm->mm_users) > 1) {
1162                         /*
1163                          * We don't check the error code - if userspace has
1164                          * not set up a proper pointer then tough luck.
1165                          */
1166                         put_user(0, tsk->clear_child_tid);
1167                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
1168                                         1, NULL, NULL, 0);
1169                 }
1170                 tsk->clear_child_tid = NULL;
1171         }
1172
1173         /*
1174          * All done, finally we can wake up parent and return this mm to him.
1175          * Also kthread_stop() uses this completion for synchronization.
1176          */
1177         if (tsk->vfork_done)
1178                 complete_vfork_done(tsk);
1179 }
1180
1181 /*
1182  * Allocate a new mm structure and copy contents from the
1183  * mm structure of the passed in task structure.
1184  */
1185 static struct mm_struct *dup_mm(struct task_struct *tsk)
1186 {
1187         struct mm_struct *mm, *oldmm = current->mm;
1188         int err;
1189
1190         mm = allocate_mm();
1191         if (!mm)
1192                 goto fail_nomem;
1193
1194         memcpy(mm, oldmm, sizeof(*mm));
1195
1196         if (!mm_init(mm, tsk, mm->user_ns))
1197                 goto fail_nomem;
1198
1199         err = dup_mmap(mm, oldmm);
1200         if (err)
1201                 goto free_pt;
1202
1203         mm->hiwater_rss = get_mm_rss(mm);
1204         mm->hiwater_vm = mm->total_vm;
1205
1206         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1207                 goto free_pt;
1208
1209         return mm;
1210
1211 free_pt:
1212         /* don't put binfmt in mmput, we haven't got module yet */
1213         mm->binfmt = NULL;
1214         mmput(mm);
1215
1216 fail_nomem:
1217         return NULL;
1218 }
1219
1220 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1221 {
1222         struct mm_struct *mm, *oldmm;
1223         int retval;
1224
1225         tsk->min_flt = tsk->maj_flt = 0;
1226         tsk->nvcsw = tsk->nivcsw = 0;
1227 #ifdef CONFIG_DETECT_HUNG_TASK
1228         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1229 #endif
1230
1231         tsk->mm = NULL;
1232         tsk->active_mm = NULL;
1233
1234         /*
1235          * Are we cloning a kernel thread?
1236          *
1237          * We need to steal a active VM for that..
1238          */
1239         oldmm = current->mm;
1240         if (!oldmm)
1241                 return 0;
1242
1243         /* initialize the new vmacache entries */
1244         vmacache_flush(tsk);
1245
1246         if (clone_flags & CLONE_VM) {
1247                 mmget(oldmm);
1248                 mm = oldmm;
1249                 goto good_mm;
1250         }
1251
1252         retval = -ENOMEM;
1253         mm = dup_mm(tsk);
1254         if (!mm)
1255                 goto fail_nomem;
1256
1257 good_mm:
1258         tsk->mm = mm;
1259         tsk->active_mm = mm;
1260         return 0;
1261
1262 fail_nomem:
1263         return retval;
1264 }
1265
1266 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1267 {
1268         struct fs_struct *fs = current->fs;
1269         if (clone_flags & CLONE_FS) {
1270                 /* tsk->fs is already what we want */
1271                 spin_lock(&fs->lock);
1272                 if (fs->in_exec) {
1273                         spin_unlock(&fs->lock);
1274                         return -EAGAIN;
1275                 }
1276                 fs->users++;
1277                 spin_unlock(&fs->lock);
1278                 return 0;
1279         }
1280         tsk->fs = copy_fs_struct(fs);
1281         if (!tsk->fs)
1282                 return -ENOMEM;
1283         return 0;
1284 }
1285
1286 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1287 {
1288         struct files_struct *oldf, *newf;
1289         int error = 0;
1290
1291         /*
1292          * A background process may not have any files ...
1293          */
1294         oldf = current->files;
1295         if (!oldf)
1296                 goto out;
1297
1298         if (clone_flags & CLONE_FILES) {
1299                 atomic_inc(&oldf->count);
1300                 goto out;
1301         }
1302
1303         newf = dup_fd(oldf, &error);
1304         if (!newf)
1305                 goto out;
1306
1307         tsk->files = newf;
1308         error = 0;
1309 out:
1310         return error;
1311 }
1312
1313 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1314 {
1315 #ifdef CONFIG_BLOCK
1316         struct io_context *ioc = current->io_context;
1317         struct io_context *new_ioc;
1318
1319         if (!ioc)
1320                 return 0;
1321         /*
1322          * Share io context with parent, if CLONE_IO is set
1323          */
1324         if (clone_flags & CLONE_IO) {
1325                 ioc_task_link(ioc);
1326                 tsk->io_context = ioc;
1327         } else if (ioprio_valid(ioc->ioprio)) {
1328                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1329                 if (unlikely(!new_ioc))
1330                         return -ENOMEM;
1331
1332                 new_ioc->ioprio = ioc->ioprio;
1333                 put_io_context(new_ioc);
1334         }
1335 #endif
1336         return 0;
1337 }
1338
1339 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1340 {
1341         struct sighand_struct *sig;
1342
1343         if (clone_flags & CLONE_SIGHAND) {
1344                 atomic_inc(&current->sighand->count);
1345                 return 0;
1346         }
1347         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1348         rcu_assign_pointer(tsk->sighand, sig);
1349         if (!sig)
1350                 return -ENOMEM;
1351
1352         atomic_set(&sig->count, 1);
1353         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1354         return 0;
1355 }
1356
1357 void __cleanup_sighand(struct sighand_struct *sighand)
1358 {
1359         if (atomic_dec_and_test(&sighand->count)) {
1360                 signalfd_cleanup(sighand);
1361                 /*
1362                  * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1363                  * without an RCU grace period, see __lock_task_sighand().
1364                  */
1365                 kmem_cache_free(sighand_cachep, sighand);
1366         }
1367 }
1368
1369 #ifdef CONFIG_POSIX_TIMERS
1370 /*
1371  * Initialize POSIX timer handling for a thread group.
1372  */
1373 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1374 {
1375         unsigned long cpu_limit;
1376
1377         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1378         if (cpu_limit != RLIM_INFINITY) {
1379                 sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1380                 sig->cputimer.running = true;
1381         }
1382
1383         /* The timer lists. */
1384         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1385         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1386         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1387 }
1388 #else
1389 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1390 #endif
1391
1392 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1393 {
1394         struct signal_struct *sig;
1395
1396         if (clone_flags & CLONE_THREAD)
1397                 return 0;
1398
1399         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1400         tsk->signal = sig;
1401         if (!sig)
1402                 return -ENOMEM;
1403
1404         sig->nr_threads = 1;
1405         atomic_set(&sig->live, 1);
1406         atomic_set(&sig->sigcnt, 1);
1407
1408         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1409         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1410         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1411
1412         init_waitqueue_head(&sig->wait_chldexit);
1413         sig->curr_target = tsk;
1414         init_sigpending(&sig->shared_pending);
1415         seqlock_init(&sig->stats_lock);
1416         prev_cputime_init(&sig->prev_cputime);
1417
1418 #ifdef CONFIG_POSIX_TIMERS
1419         INIT_LIST_HEAD(&sig->posix_timers);
1420         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1421         sig->real_timer.function = it_real_fn;
1422 #endif
1423
1424         task_lock(current->group_leader);
1425         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1426         task_unlock(current->group_leader);
1427
1428         posix_cpu_timers_init_group(sig);
1429
1430         tty_audit_fork(sig);
1431         sched_autogroup_fork(sig);
1432
1433         sig->oom_score_adj = current->signal->oom_score_adj;
1434         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1435
1436         mutex_init(&sig->cred_guard_mutex);
1437
1438         return 0;
1439 }
1440
1441 static void copy_seccomp(struct task_struct *p)
1442 {
1443 #ifdef CONFIG_SECCOMP
1444         /*
1445          * Must be called with sighand->lock held, which is common to
1446          * all threads in the group. Holding cred_guard_mutex is not
1447          * needed because this new task is not yet running and cannot
1448          * be racing exec.
1449          */
1450         assert_spin_locked(&current->sighand->siglock);
1451
1452         /* Ref-count the new filter user, and assign it. */
1453         get_seccomp_filter(current);
1454         p->seccomp = current->seccomp;
1455
1456         /*
1457          * Explicitly enable no_new_privs here in case it got set
1458          * between the task_struct being duplicated and holding the
1459          * sighand lock. The seccomp state and nnp must be in sync.
1460          */
1461         if (task_no_new_privs(current))
1462                 task_set_no_new_privs(p);
1463
1464         /*
1465          * If the parent gained a seccomp mode after copying thread
1466          * flags and between before we held the sighand lock, we have
1467          * to manually enable the seccomp thread flag here.
1468          */
1469         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1470                 set_tsk_thread_flag(p, TIF_SECCOMP);
1471 #endif
1472 }
1473
1474 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1475 {
1476         current->clear_child_tid = tidptr;
1477
1478         return task_pid_vnr(current);
1479 }
1480
1481 static void rt_mutex_init_task(struct task_struct *p)
1482 {
1483         raw_spin_lock_init(&p->pi_lock);
1484 #ifdef CONFIG_RT_MUTEXES
1485         p->pi_waiters = RB_ROOT_CACHED;
1486         p->pi_top_task = NULL;
1487         p->pi_blocked_on = NULL;
1488 #endif
1489 }
1490
1491 #ifdef CONFIG_POSIX_TIMERS
1492 /*
1493  * Initialize POSIX timer handling for a single task.
1494  */
1495 static void posix_cpu_timers_init(struct task_struct *tsk)
1496 {
1497         tsk->cputime_expires.prof_exp = 0;
1498         tsk->cputime_expires.virt_exp = 0;
1499         tsk->cputime_expires.sched_exp = 0;
1500         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1501         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1502         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1503 }
1504 #else
1505 static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1506 #endif
1507
1508 static inline void
1509 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1510 {
1511          task->pids[type].pid = pid;
1512 }
1513
1514 static inline void rcu_copy_process(struct task_struct *p)
1515 {
1516 #ifdef CONFIG_PREEMPT_RCU
1517         p->rcu_read_lock_nesting = 0;
1518         p->rcu_read_unlock_special.s = 0;
1519         p->rcu_blocked_node = NULL;
1520         INIT_LIST_HEAD(&p->rcu_node_entry);
1521 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1522 #ifdef CONFIG_TASKS_RCU
1523         p->rcu_tasks_holdout = false;
1524         INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1525         p->rcu_tasks_idle_cpu = -1;
1526 #endif /* #ifdef CONFIG_TASKS_RCU */
1527 }
1528
1529 /*
1530  * This creates a new process as a copy of the old one,
1531  * but does not actually start it yet.
1532  *
1533  * It copies the registers, and all the appropriate
1534  * parts of the process environment (as per the clone
1535  * flags). The actual kick-off is left to the caller.
1536  */
1537 static __latent_entropy struct task_struct *copy_process(
1538                                         unsigned long clone_flags,
1539                                         unsigned long stack_start,
1540                                         unsigned long stack_size,
1541                                         int __user *child_tidptr,
1542                                         struct pid *pid,
1543                                         int trace,
1544                                         unsigned long tls,
1545                                         int node)
1546 {
1547         int retval;
1548         struct task_struct *p;
1549
1550         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1551                 return ERR_PTR(-EINVAL);
1552
1553         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1554                 return ERR_PTR(-EINVAL);
1555
1556         /*
1557          * Thread groups must share signals as well, and detached threads
1558          * can only be started up within the thread group.
1559          */
1560         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1561                 return ERR_PTR(-EINVAL);
1562
1563         /*
1564          * Shared signal handlers imply shared VM. By way of the above,
1565          * thread groups also imply shared VM. Blocking this case allows
1566          * for various simplifications in other code.
1567          */
1568         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1569                 return ERR_PTR(-EINVAL);
1570
1571         /*
1572          * Siblings of global init remain as zombies on exit since they are
1573          * not reaped by their parent (swapper). To solve this and to avoid
1574          * multi-rooted process trees, prevent global and container-inits
1575          * from creating siblings.
1576          */
1577         if ((clone_flags & CLONE_PARENT) &&
1578                                 current->signal->flags & SIGNAL_UNKILLABLE)
1579                 return ERR_PTR(-EINVAL);
1580
1581         /*
1582          * If the new process will be in a different pid or user namespace
1583          * do not allow it to share a thread group with the forking task.
1584          */
1585         if (clone_flags & CLONE_THREAD) {
1586                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1587                     (task_active_pid_ns(current) !=
1588                                 current->nsproxy->pid_ns_for_children))
1589                         return ERR_PTR(-EINVAL);
1590         }
1591
1592         retval = -ENOMEM;
1593         p = dup_task_struct(current, node);
1594         if (!p)
1595                 goto fork_out;
1596
1597         /*
1598          * This _must_ happen before we call free_task(), i.e. before we jump
1599          * to any of the bad_fork_* labels. This is to avoid freeing
1600          * p->set_child_tid which is (ab)used as a kthread's data pointer for
1601          * kernel threads (PF_KTHREAD).
1602          */
1603         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1604         /*
1605          * Clear TID on mm_release()?
1606          */
1607         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1608
1609         ftrace_graph_init_task(p);
1610
1611         rt_mutex_init_task(p);
1612
1613 #ifdef CONFIG_PROVE_LOCKING
1614         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1615         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1616 #endif
1617         retval = -EAGAIN;
1618         if (atomic_read(&p->real_cred->user->processes) >=
1619                         task_rlimit(p, RLIMIT_NPROC)) {
1620                 if (p->real_cred->user != INIT_USER &&
1621                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1622                         goto bad_fork_free;
1623         }
1624         current->flags &= ~PF_NPROC_EXCEEDED;
1625
1626         retval = copy_creds(p, clone_flags);
1627         if (retval < 0)
1628                 goto bad_fork_free;
1629
1630         /*
1631          * If multiple threads are within copy_process(), then this check
1632          * triggers too late. This doesn't hurt, the check is only there
1633          * to stop root fork bombs.
1634          */
1635         retval = -EAGAIN;
1636         if (nr_threads >= max_threads)
1637                 goto bad_fork_cleanup_count;
1638
1639         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1640         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1641         p->flags |= PF_FORKNOEXEC;
1642         INIT_LIST_HEAD(&p->children);
1643         INIT_LIST_HEAD(&p->sibling);
1644         rcu_copy_process(p);
1645         p->vfork_done = NULL;
1646         spin_lock_init(&p->alloc_lock);
1647
1648         init_sigpending(&p->pending);
1649
1650         p->utime = p->stime = p->gtime = 0;
1651 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1652         p->utimescaled = p->stimescaled = 0;
1653 #endif
1654         prev_cputime_init(&p->prev_cputime);
1655
1656 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1657         seqcount_init(&p->vtime.seqcount);
1658         p->vtime.starttime = 0;
1659         p->vtime.state = VTIME_INACTIVE;
1660 #endif
1661
1662 #if defined(SPLIT_RSS_COUNTING)
1663         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1664 #endif
1665
1666         p->default_timer_slack_ns = current->timer_slack_ns;
1667
1668         task_io_accounting_init(&p->ioac);
1669         acct_clear_integrals(p);
1670
1671         posix_cpu_timers_init(p);
1672
1673         p->start_time = ktime_get_ns();
1674         p->real_start_time = ktime_get_boot_ns();
1675         p->io_context = NULL;
1676         p->audit_context = NULL;
1677         cgroup_fork(p);
1678 #ifdef CONFIG_NUMA
1679         p->mempolicy = mpol_dup(p->mempolicy);
1680         if (IS_ERR(p->mempolicy)) {
1681                 retval = PTR_ERR(p->mempolicy);
1682                 p->mempolicy = NULL;
1683                 goto bad_fork_cleanup_threadgroup_lock;
1684         }
1685 #endif
1686 #ifdef CONFIG_CPUSETS
1687         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1688         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1689         seqcount_init(&p->mems_allowed_seq);
1690 #endif
1691 #ifdef CONFIG_TRACE_IRQFLAGS
1692         p->irq_events = 0;
1693         p->hardirqs_enabled = 0;
1694         p->hardirq_enable_ip = 0;
1695         p->hardirq_enable_event = 0;
1696         p->hardirq_disable_ip = _THIS_IP_;
1697         p->hardirq_disable_event = 0;
1698         p->softirqs_enabled = 1;
1699         p->softirq_enable_ip = _THIS_IP_;
1700         p->softirq_enable_event = 0;
1701         p->softirq_disable_ip = 0;
1702         p->softirq_disable_event = 0;
1703         p->hardirq_context = 0;
1704         p->softirq_context = 0;
1705 #endif
1706
1707         p->pagefault_disabled = 0;
1708
1709 #ifdef CONFIG_LOCKDEP
1710         p->lockdep_depth = 0; /* no locks held yet */
1711         p->curr_chain_key = 0;
1712         p->lockdep_recursion = 0;
1713         lockdep_init_task(p);
1714 #endif
1715
1716 #ifdef CONFIG_DEBUG_MUTEXES
1717         p->blocked_on = NULL; /* not blocked yet */
1718 #endif
1719 #ifdef CONFIG_BCACHE
1720         p->sequential_io        = 0;
1721         p->sequential_io_avg    = 0;
1722 #endif
1723
1724         /* Perform scheduler related setup. Assign this task to a CPU. */
1725         retval = sched_fork(clone_flags, p);
1726         if (retval)
1727                 goto bad_fork_cleanup_policy;
1728
1729         retval = perf_event_init_task(p);
1730         if (retval)
1731                 goto bad_fork_cleanup_policy;
1732         retval = audit_alloc(p);
1733         if (retval)
1734                 goto bad_fork_cleanup_perf;
1735         /* copy all the process information */
1736         shm_init_task(p);
1737         retval = security_task_alloc(p, clone_flags);
1738         if (retval)
1739                 goto bad_fork_cleanup_audit;
1740         retval = copy_semundo(clone_flags, p);
1741         if (retval)
1742                 goto bad_fork_cleanup_security;
1743         retval = copy_files(clone_flags, p);
1744         if (retval)
1745                 goto bad_fork_cleanup_semundo;
1746         retval = copy_fs(clone_flags, p);
1747         if (retval)
1748                 goto bad_fork_cleanup_files;
1749         retval = copy_sighand(clone_flags, p);
1750         if (retval)
1751                 goto bad_fork_cleanup_fs;
1752         retval = copy_signal(clone_flags, p);
1753         if (retval)
1754                 goto bad_fork_cleanup_sighand;
1755         retval = copy_mm(clone_flags, p);
1756         if (retval)
1757                 goto bad_fork_cleanup_signal;
1758         retval = copy_namespaces(clone_flags, p);
1759         if (retval)
1760                 goto bad_fork_cleanup_mm;
1761         retval = copy_io(clone_flags, p);
1762         if (retval)
1763                 goto bad_fork_cleanup_namespaces;
1764         retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1765         if (retval)
1766                 goto bad_fork_cleanup_io;
1767
1768         if (pid != &init_struct_pid) {
1769                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1770                 if (IS_ERR(pid)) {
1771                         retval = PTR_ERR(pid);
1772                         goto bad_fork_cleanup_thread;
1773                 }
1774         }
1775
1776 #ifdef CONFIG_BLOCK
1777         p->plug = NULL;
1778 #endif
1779 #ifdef CONFIG_FUTEX
1780         p->robust_list = NULL;
1781 #ifdef CONFIG_COMPAT
1782         p->compat_robust_list = NULL;
1783 #endif
1784         INIT_LIST_HEAD(&p->pi_state_list);
1785         p->pi_state_cache = NULL;
1786 #endif
1787         /*
1788          * sigaltstack should be cleared when sharing the same VM
1789          */
1790         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1791                 sas_ss_reset(p);
1792
1793         /*
1794          * Syscall tracing and stepping should be turned off in the
1795          * child regardless of CLONE_PTRACE.
1796          */
1797         user_disable_single_step(p);
1798         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1799 #ifdef TIF_SYSCALL_EMU
1800         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1801 #endif
1802         clear_all_latency_tracing(p);
1803
1804         /* ok, now we should be set up.. */
1805         p->pid = pid_nr(pid);
1806         if (clone_flags & CLONE_THREAD) {
1807                 p->exit_signal = -1;
1808                 p->group_leader = current->group_leader;
1809                 p->tgid = current->tgid;
1810         } else {
1811                 if (clone_flags & CLONE_PARENT)
1812                         p->exit_signal = current->group_leader->exit_signal;
1813                 else
1814                         p->exit_signal = (clone_flags & CSIGNAL);
1815                 p->group_leader = p;
1816                 p->tgid = p->pid;
1817         }
1818
1819         p->nr_dirtied = 0;
1820         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1821         p->dirty_paused_when = 0;
1822
1823         p->pdeath_signal = 0;
1824         INIT_LIST_HEAD(&p->thread_group);
1825         p->task_works = NULL;
1826
1827         cgroup_threadgroup_change_begin(current);
1828         /*
1829          * Ensure that the cgroup subsystem policies allow the new process to be
1830          * forked. It should be noted the the new process's css_set can be changed
1831          * between here and cgroup_post_fork() if an organisation operation is in
1832          * progress.
1833          */
1834         retval = cgroup_can_fork(p);
1835         if (retval)
1836                 goto bad_fork_free_pid;
1837
1838         /*
1839          * Make it visible to the rest of the system, but dont wake it up yet.
1840          * Need tasklist lock for parent etc handling!
1841          */
1842         write_lock_irq(&tasklist_lock);
1843
1844         /* CLONE_PARENT re-uses the old parent */
1845         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1846                 p->real_parent = current->real_parent;
1847                 p->parent_exec_id = current->parent_exec_id;
1848         } else {
1849                 p->real_parent = current;
1850                 p->parent_exec_id = current->self_exec_id;
1851         }
1852
1853         klp_copy_process(p);
1854
1855         spin_lock(&current->sighand->siglock);
1856
1857         /*
1858          * Copy seccomp details explicitly here, in case they were changed
1859          * before holding sighand lock.
1860          */
1861         copy_seccomp(p);
1862
1863         /*
1864          * Process group and session signals need to be delivered to just the
1865          * parent before the fork or both the parent and the child after the
1866          * fork. Restart if a signal comes in before we add the new process to
1867          * it's process group.
1868          * A fatal signal pending means that current will exit, so the new
1869          * thread can't slip out of an OOM kill (or normal SIGKILL).
1870         */
1871         recalc_sigpending();
1872         if (signal_pending(current)) {
1873                 retval = -ERESTARTNOINTR;
1874                 goto bad_fork_cancel_cgroup;
1875         }
1876         if (unlikely(!(ns_of_pid(pid)->nr_hashed & PIDNS_HASH_ADDING))) {
1877                 retval = -ENOMEM;
1878                 goto bad_fork_cancel_cgroup;
1879         }
1880
1881         if (likely(p->pid)) {
1882                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1883
1884                 init_task_pid(p, PIDTYPE_PID, pid);
1885                 if (thread_group_leader(p)) {
1886                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1887                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1888
1889                         if (is_child_reaper(pid)) {
1890                                 ns_of_pid(pid)->child_reaper = p;
1891                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1892                         }
1893
1894                         p->signal->leader_pid = pid;
1895                         p->signal->tty = tty_kref_get(current->signal->tty);
1896                         /*
1897                          * Inherit has_child_subreaper flag under the same
1898                          * tasklist_lock with adding child to the process tree
1899                          * for propagate_has_child_subreaper optimization.
1900                          */
1901                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
1902                                                          p->real_parent->signal->is_child_subreaper;
1903                         list_add_tail(&p->sibling, &p->real_parent->children);
1904                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1905                         attach_pid(p, PIDTYPE_PGID);
1906                         attach_pid(p, PIDTYPE_SID);
1907                         __this_cpu_inc(process_counts);
1908                 } else {
1909                         current->signal->nr_threads++;
1910                         atomic_inc(&current->signal->live);
1911                         atomic_inc(&current->signal->sigcnt);
1912                         list_add_tail_rcu(&p->thread_group,
1913                                           &p->group_leader->thread_group);
1914                         list_add_tail_rcu(&p->thread_node,
1915                                           &p->signal->thread_head);
1916                 }
1917                 attach_pid(p, PIDTYPE_PID);
1918                 nr_threads++;
1919         }
1920
1921         total_forks++;
1922         spin_unlock(&current->sighand->siglock);
1923         syscall_tracepoint_update(p);
1924         write_unlock_irq(&tasklist_lock);
1925
1926         proc_fork_connector(p);
1927         cgroup_post_fork(p);
1928         cgroup_threadgroup_change_end(current);
1929         perf_event_fork(p);
1930
1931         trace_task_newtask(p, clone_flags);
1932         uprobe_copy_process(p, clone_flags);
1933
1934         return p;
1935
1936 bad_fork_cancel_cgroup:
1937         spin_unlock(&current->sighand->siglock);
1938         write_unlock_irq(&tasklist_lock);
1939         cgroup_cancel_fork(p);
1940 bad_fork_free_pid:
1941         cgroup_threadgroup_change_end(current);
1942         if (pid != &init_struct_pid)
1943                 free_pid(pid);
1944 bad_fork_cleanup_thread:
1945         exit_thread(p);
1946 bad_fork_cleanup_io:
1947         if (p->io_context)
1948                 exit_io_context(p);
1949 bad_fork_cleanup_namespaces:
1950         exit_task_namespaces(p);
1951 bad_fork_cleanup_mm:
1952         if (p->mm)
1953                 mmput(p->mm);
1954 bad_fork_cleanup_signal:
1955         if (!(clone_flags & CLONE_THREAD))
1956                 free_signal_struct(p->signal);
1957 bad_fork_cleanup_sighand:
1958         __cleanup_sighand(p->sighand);
1959 bad_fork_cleanup_fs:
1960         exit_fs(p); /* blocking */
1961 bad_fork_cleanup_files:
1962         exit_files(p); /* blocking */
1963 bad_fork_cleanup_semundo:
1964         exit_sem(p);
1965 bad_fork_cleanup_security:
1966         security_task_free(p);
1967 bad_fork_cleanup_audit:
1968         audit_free(p);
1969 bad_fork_cleanup_perf:
1970         perf_event_free_task(p);
1971 bad_fork_cleanup_policy:
1972         lockdep_free_task(p);
1973 #ifdef CONFIG_NUMA
1974         mpol_put(p->mempolicy);
1975 bad_fork_cleanup_threadgroup_lock:
1976 #endif
1977         delayacct_tsk_free(p);
1978 bad_fork_cleanup_count:
1979         atomic_dec(&p->cred->user->processes);
1980         exit_creds(p);
1981 bad_fork_free:
1982         p->state = TASK_DEAD;
1983         put_task_stack(p);
1984         free_task(p);
1985 fork_out:
1986         return ERR_PTR(retval);
1987 }
1988
1989 static inline void init_idle_pids(struct pid_link *links)
1990 {
1991         enum pid_type type;
1992
1993         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1994                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1995                 links[type].pid = &init_struct_pid;
1996         }
1997 }
1998
1999 struct task_struct *fork_idle(int cpu)
2000 {
2001         struct task_struct *task;
2002         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
2003                             cpu_to_node(cpu));
2004         if (!IS_ERR(task)) {
2005                 init_idle_pids(task->pids);
2006                 init_idle(task, cpu);
2007         }
2008
2009         return task;
2010 }
2011
2012 /*
2013  *  Ok, this is the main fork-routine.
2014  *
2015  * It copies the process, and if successful kick-starts
2016  * it and waits for it to finish using the VM if required.
2017  */
2018 long _do_fork(unsigned long clone_flags,
2019               unsigned long stack_start,
2020               unsigned long stack_size,
2021               int __user *parent_tidptr,
2022               int __user *child_tidptr,
2023               unsigned long tls)
2024 {
2025         struct task_struct *p;
2026         int trace = 0;
2027         long nr;
2028
2029         /*
2030          * Determine whether and which event to report to ptracer.  When
2031          * called from kernel_thread or CLONE_UNTRACED is explicitly
2032          * requested, no event is reported; otherwise, report if the event
2033          * for the type of forking is enabled.
2034          */
2035         if (!(clone_flags & CLONE_UNTRACED)) {
2036                 if (clone_flags & CLONE_VFORK)
2037                         trace = PTRACE_EVENT_VFORK;
2038                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
2039                         trace = PTRACE_EVENT_CLONE;
2040                 else
2041                         trace = PTRACE_EVENT_FORK;
2042
2043                 if (likely(!ptrace_event_enabled(current, trace)))
2044                         trace = 0;
2045         }
2046
2047         p = copy_process(clone_flags, stack_start, stack_size,
2048                          child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
2049         add_latent_entropy();
2050         /*
2051          * Do this prior waking up the new thread - the thread pointer
2052          * might get invalid after that point, if the thread exits quickly.
2053          */
2054         if (!IS_ERR(p)) {
2055                 struct completion vfork;
2056                 struct pid *pid;
2057
2058                 trace_sched_process_fork(current, p);
2059
2060                 pid = get_task_pid(p, PIDTYPE_PID);
2061                 nr = pid_vnr(pid);
2062
2063                 if (clone_flags & CLONE_PARENT_SETTID)
2064                         put_user(nr, parent_tidptr);
2065
2066                 if (clone_flags & CLONE_VFORK) {
2067                         p->vfork_done = &vfork;
2068                         init_completion(&vfork);
2069                         get_task_struct(p);
2070                 }
2071
2072                 wake_up_new_task(p);
2073
2074                 /* forking complete and child started to run, tell ptracer */
2075                 if (unlikely(trace))
2076                         ptrace_event_pid(trace, pid);
2077
2078                 if (clone_flags & CLONE_VFORK) {
2079                         if (!wait_for_vfork_done(p, &vfork))
2080                                 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2081                 }
2082
2083                 put_pid(pid);
2084         } else {
2085                 nr = PTR_ERR(p);
2086         }
2087         return nr;
2088 }
2089
2090 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2091 /* For compatibility with architectures that call do_fork directly rather than
2092  * using the syscall entry points below. */
2093 long do_fork(unsigned long clone_flags,
2094               unsigned long stack_start,
2095               unsigned long stack_size,
2096               int __user *parent_tidptr,
2097               int __user *child_tidptr)
2098 {
2099         return _do_fork(clone_flags, stack_start, stack_size,
2100                         parent_tidptr, child_tidptr, 0);
2101 }
2102 #endif
2103
2104 /*
2105  * Create a kernel thread.
2106  */
2107 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2108 {
2109         return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2110                 (unsigned long)arg, NULL, NULL, 0);
2111 }
2112
2113 #ifdef __ARCH_WANT_SYS_FORK
2114 SYSCALL_DEFINE0(fork)
2115 {
2116 #ifdef CONFIG_MMU
2117         return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2118 #else
2119         /* can not support in nommu mode */
2120         return -EINVAL;
2121 #endif
2122 }
2123 #endif
2124
2125 #ifdef __ARCH_WANT_SYS_VFORK
2126 SYSCALL_DEFINE0(vfork)
2127 {
2128         return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2129                         0, NULL, NULL, 0);
2130 }
2131 #endif
2132
2133 #ifdef __ARCH_WANT_SYS_CLONE
2134 #ifdef CONFIG_CLONE_BACKWARDS
2135 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2136                  int __user *, parent_tidptr,
2137                  unsigned long, tls,
2138                  int __user *, child_tidptr)
2139 #elif defined(CONFIG_CLONE_BACKWARDS2)
2140 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2141                  int __user *, parent_tidptr,
2142                  int __user *, child_tidptr,
2143                  unsigned long, tls)
2144 #elif defined(CONFIG_CLONE_BACKWARDS3)
2145 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2146                 int, stack_size,
2147                 int __user *, parent_tidptr,
2148                 int __user *, child_tidptr,
2149                 unsigned long, tls)
2150 #else
2151 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2152                  int __user *, parent_tidptr,
2153                  int __user *, child_tidptr,
2154                  unsigned long, tls)
2155 #endif
2156 {
2157         return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2158 }
2159 #endif
2160
2161 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2162 {
2163         struct task_struct *leader, *parent, *child;
2164         int res;
2165
2166         read_lock(&tasklist_lock);
2167         leader = top = top->group_leader;
2168 down:
2169         for_each_thread(leader, parent) {
2170                 list_for_each_entry(child, &parent->children, sibling) {
2171                         res = visitor(child, data);
2172                         if (res) {
2173                                 if (res < 0)
2174                                         goto out;
2175                                 leader = child;
2176                                 goto down;
2177                         }
2178 up:
2179                         ;
2180                 }
2181         }
2182
2183         if (leader != top) {
2184                 child = leader;
2185                 parent = child->real_parent;
2186                 leader = parent->group_leader;
2187                 goto up;
2188         }
2189 out:
2190         read_unlock(&tasklist_lock);
2191 }
2192
2193 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2194 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2195 #endif
2196
2197 static void sighand_ctor(void *data)
2198 {
2199         struct sighand_struct *sighand = data;
2200
2201         spin_lock_init(&sighand->siglock);
2202         init_waitqueue_head(&sighand->signalfd_wqh);
2203 }
2204
2205 void __init proc_caches_init(void)
2206 {
2207         sighand_cachep = kmem_cache_create("sighand_cache",
2208                         sizeof(struct sighand_struct), 0,
2209                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2210                         SLAB_ACCOUNT, sighand_ctor);
2211         signal_cachep = kmem_cache_create("signal_cache",
2212                         sizeof(struct signal_struct), 0,
2213                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2214                         NULL);
2215         files_cachep = kmem_cache_create("files_cache",
2216                         sizeof(struct files_struct), 0,
2217                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2218                         NULL);
2219         fs_cachep = kmem_cache_create("fs_cache",
2220                         sizeof(struct fs_struct), 0,
2221                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2222                         NULL);
2223         /*
2224          * FIXME! The "sizeof(struct mm_struct)" currently includes the
2225          * whole struct cpumask for the OFFSTACK case. We could change
2226          * this to *only* allocate as much of it as required by the
2227          * maximum number of CPU's we can ever have.  The cpumask_allocation
2228          * is at the end of the structure, exactly for that reason.
2229          */
2230         mm_cachep = kmem_cache_create("mm_struct",
2231                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
2232                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2233                         NULL);
2234         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2235         mmap_init();
2236         nsproxy_cache_init();
2237 }
2238
2239 /*
2240  * Check constraints on flags passed to the unshare system call.
2241  */
2242 static int check_unshare_flags(unsigned long unshare_flags)
2243 {
2244         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2245                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2246                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2247                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2248                 return -EINVAL;
2249         /*
2250          * Not implemented, but pretend it works if there is nothing
2251          * to unshare.  Note that unsharing the address space or the
2252          * signal handlers also need to unshare the signal queues (aka
2253          * CLONE_THREAD).
2254          */
2255         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2256                 if (!thread_group_empty(current))
2257                         return -EINVAL;
2258         }
2259         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2260                 if (atomic_read(&current->sighand->count) > 1)
2261                         return -EINVAL;
2262         }
2263         if (unshare_flags & CLONE_VM) {
2264                 if (!current_is_single_threaded())
2265                         return -EINVAL;
2266         }
2267
2268         return 0;
2269 }
2270
2271 /*
2272  * Unshare the filesystem structure if it is being shared
2273  */
2274 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2275 {
2276         struct fs_struct *fs = current->fs;
2277
2278         if (!(unshare_flags & CLONE_FS) || !fs)
2279                 return 0;
2280
2281         /* don't need lock here; in the worst case we'll do useless copy */
2282         if (fs->users == 1)
2283                 return 0;
2284
2285         *new_fsp = copy_fs_struct(fs);
2286         if (!*new_fsp)
2287                 return -ENOMEM;
2288
2289         return 0;
2290 }
2291
2292 /*
2293  * Unshare file descriptor table if it is being shared
2294  */
2295 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2296 {
2297         struct files_struct *fd = current->files;
2298         int error = 0;
2299
2300         if ((unshare_flags & CLONE_FILES) &&
2301             (fd && atomic_read(&fd->count) > 1)) {
2302                 *new_fdp = dup_fd(fd, &error);
2303                 if (!*new_fdp)
2304                         return error;
2305         }
2306
2307         return 0;
2308 }
2309
2310 /*
2311  * unshare allows a process to 'unshare' part of the process
2312  * context which was originally shared using clone.  copy_*
2313  * functions used by do_fork() cannot be used here directly
2314  * because they modify an inactive task_struct that is being
2315  * constructed. Here we are modifying the current, active,
2316  * task_struct.
2317  */
2318 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2319 {
2320         struct fs_struct *fs, *new_fs = NULL;
2321         struct files_struct *fd, *new_fd = NULL;
2322         struct cred *new_cred = NULL;
2323         struct nsproxy *new_nsproxy = NULL;
2324         int do_sysvsem = 0;
2325         int err;
2326
2327         /*
2328          * If unsharing a user namespace must also unshare the thread group
2329          * and unshare the filesystem root and working directories.
2330          */
2331         if (unshare_flags & CLONE_NEWUSER)
2332                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2333         /*
2334          * If unsharing vm, must also unshare signal handlers.
2335          */
2336         if (unshare_flags & CLONE_VM)
2337                 unshare_flags |= CLONE_SIGHAND;
2338         /*
2339          * If unsharing a signal handlers, must also unshare the signal queues.
2340          */
2341         if (unshare_flags & CLONE_SIGHAND)
2342                 unshare_flags |= CLONE_THREAD;
2343         /*
2344          * If unsharing namespace, must also unshare filesystem information.
2345          */
2346         if (unshare_flags & CLONE_NEWNS)
2347                 unshare_flags |= CLONE_FS;
2348
2349         err = check_unshare_flags(unshare_flags);
2350         if (err)
2351                 goto bad_unshare_out;
2352         /*
2353          * CLONE_NEWIPC must also detach from the undolist: after switching
2354          * to a new ipc namespace, the semaphore arrays from the old
2355          * namespace are unreachable.
2356          */
2357         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2358                 do_sysvsem = 1;
2359         err = unshare_fs(unshare_flags, &new_fs);
2360         if (err)
2361                 goto bad_unshare_out;
2362         err = unshare_fd(unshare_flags, &new_fd);
2363         if (err)
2364                 goto bad_unshare_cleanup_fs;
2365         err = unshare_userns(unshare_flags, &new_cred);
2366         if (err)
2367                 goto bad_unshare_cleanup_fd;
2368         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2369                                          new_cred, new_fs);
2370         if (err)
2371                 goto bad_unshare_cleanup_cred;
2372
2373         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2374                 if (do_sysvsem) {
2375                         /*
2376                          * CLONE_SYSVSEM is equivalent to sys_exit().
2377                          */
2378                         exit_sem(current);
2379                 }
2380                 if (unshare_flags & CLONE_NEWIPC) {
2381                         /* Orphan segments in old ns (see sem above). */
2382                         exit_shm(current);
2383                         shm_init_task(current);
2384                 }
2385
2386                 if (new_nsproxy)
2387                         switch_task_namespaces(current, new_nsproxy);
2388
2389                 task_lock(current);
2390
2391                 if (new_fs) {
2392                         fs = current->fs;
2393                         spin_lock(&fs->lock);
2394                         current->fs = new_fs;
2395                         if (--fs->users)
2396                                 new_fs = NULL;
2397                         else
2398                                 new_fs = fs;
2399                         spin_unlock(&fs->lock);
2400                 }
2401
2402                 if (new_fd) {
2403                         fd = current->files;
2404                         current->files = new_fd;
2405                         new_fd = fd;
2406                 }
2407
2408                 task_unlock(current);
2409
2410                 if (new_cred) {
2411                         /* Install the new user namespace */
2412                         commit_creds(new_cred);
2413                         new_cred = NULL;
2414                 }
2415         }
2416
2417         perf_event_namespaces(current);
2418
2419 bad_unshare_cleanup_cred:
2420         if (new_cred)
2421                 put_cred(new_cred);
2422 bad_unshare_cleanup_fd:
2423         if (new_fd)
2424                 put_files_struct(new_fd);
2425
2426 bad_unshare_cleanup_fs:
2427         if (new_fs)
2428                 free_fs_struct(new_fs);
2429
2430 bad_unshare_out:
2431         return err;
2432 }
2433
2434 /*
2435  *      Helper to unshare the files of the current task.
2436  *      We don't want to expose copy_files internals to
2437  *      the exec layer of the kernel.
2438  */
2439
2440 int unshare_files(struct files_struct **displaced)
2441 {
2442         struct task_struct *task = current;
2443         struct files_struct *copy = NULL;
2444         int error;
2445
2446         error = unshare_fd(CLONE_FILES, &copy);
2447         if (error || !copy) {
2448                 *displaced = NULL;
2449                 return error;
2450         }
2451         *displaced = task->files;
2452         task_lock(task);
2453         task->files = copy;
2454         task_unlock(task);
2455         return 0;
2456 }
2457
2458 int sysctl_max_threads(struct ctl_table *table, int write,
2459                        void __user *buffer, size_t *lenp, loff_t *ppos)
2460 {
2461         struct ctl_table t;
2462         int ret;
2463         int threads = max_threads;
2464         int min = MIN_THREADS;
2465         int max = MAX_THREADS;
2466
2467         t = *table;
2468         t.data = &threads;
2469         t.extra1 = &min;
2470         t.extra2 = &max;
2471
2472         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2473         if (ret || !write)
2474                 return ret;
2475
2476         set_max_threads(threads);
2477
2478         return 0;
2479 }