kernel/fork.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *  linux/kernel/fork.c
   4  *
   5  *  Copyright (C) 1991, 1992  Linus Torvalds
   6  */
   7
   8 /*
   9  *  'fork.c' contains the help-routines for the 'fork' system call
  10  * (see also entry.S and others).
  11  * Fork is rather simple, once you get the hang of it, but the memory
  12  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  13  */
  14
  15 #include <linux/anon_inodes.h>
  16 #include <linux/slab.h>
  17 #include <linux/sched/autogroup.h>
  18 #include <linux/sched/mm.h>
  19 #include <linux/sched/coredump.h>
  20 #include <linux/sched/user.h>
  21 #include <linux/sched/numa_balancing.h>
  22 #include <linux/sched/stat.h>
  23 #include <linux/sched/task.h>
  24 #include <linux/sched/task_stack.h>
  25 #include <linux/sched/cputime.h>
  26 #include <linux/seq_file.h>
  27 #include <linux/rtmutex.h>
  28 #include <linux/init.h>
  29 #include <linux/unistd.h>
  30 #include <linux/module.h>
  31 #include <linux/vmalloc.h>
  32 #include <linux/completion.h>
  33 #include <linux/personality.h>
  34 #include <linux/mempolicy.h>
  35 #include <linux/sem.h>
  36 #include <linux/file.h>
  37 #include <linux/fdtable.h>
  38 #include <linux/iocontext.h>
  39 #include <linux/key.h>
  40 #include <linux/binfmts.h>
  41 #include <linux/mman.h>
  42 #include <linux/mmu_notifier.h>
  43 #include <linux/hmm.h>
  44 #include <linux/fs.h>
  45 #include <linux/mm.h>
  46 #include <linux/vmacache.h>
  47 #include <linux/nsproxy.h>
  48 #include <linux/capability.h>
  49 #include <linux/cpu.h>
  50 #include <linux/cgroup.h>
  51 #include <linux/security.h>
  52 #include <linux/hugetlb.h>
  53 #include <linux/seccomp.h>
  54 #include <linux/swap.h>
  55 #include <linux/syscalls.h>
  56 #include <linux/jiffies.h>
  57 #include <linux/futex.h>
  58 #include <linux/compat.h>
  59 #include <linux/kthread.h>
  60 #include <linux/task_io_accounting_ops.h>
  61 #include <linux/rcupdate.h>
  62 #include <linux/ptrace.h>
  63 #include <linux/mount.h>
  64 #include <linux/audit.h>
  65 #include <linux/memcontrol.h>
  66 #include <linux/ftrace.h>
  67 #include <linux/proc_fs.h>
  68 #include <linux/profile.h>
  69 #include <linux/rmap.h>
  70 #include <linux/ksm.h>
  71 #include <linux/acct.h>
  72 #include <linux/userfaultfd_k.h>
  73 #include <linux/tsacct_kern.h>
  74 #include <linux/cn_proc.h>
  75 #include <linux/freezer.h>
  76 #include <linux/delayacct.h>
  77 #include <linux/taskstats_kern.h>
  78 #include <linux/random.h>
  79 #include <linux/tty.h>
  80 #include <linux/blkdev.h>
  81 #include <linux/fs_struct.h>
  82 #include <linux/magic.h>
  83 #include <linux/perf_event.h>
  84 #include <linux/posix-timers.h>
  85 #include <linux/user-return-notifier.h>
  86 #include <linux/oom.h>
  87 #include <linux/khugepaged.h>
  88 #include <linux/signalfd.h>
  89 #include <linux/uprobes.h>
  90 #include <linux/aio.h>
  91 #include <linux/compiler.h>
  92 #include <linux/sysctl.h>
  93 #include <linux/kcov.h>
  94 #include <linux/livepatch.h>
  95 #include <linux/thread_info.h>
  96 #include <linux/stackleak.h>
  97
  98 #include <asm/pgtable.h>
  99 #include <asm/pgalloc.h>
 100 #include <linux/uaccess.h>
 101 #include <asm/mmu_context.h>
 102 #include <asm/cacheflush.h>
 103 #include <asm/tlbflush.h>
 104
 105 #include <trace/events/sched.h>
 106
 107 #define CREATE_TRACE_POINTS
 108 #include <trace/events/task.h>
 109
 110 /*
 111  * Minimum number of threads to boot the kernel
 112  */
 113 #define MIN_THREADS 20
 114
 115 /*
 116  * Maximum number of threads
 117  */
 118 #define MAX_THREADS FUTEX_TID_MASK
 119
 120 /*
 121  * Protected counters by write_lock_irq(&tasklist_lock)
 122  */
 123 unsigned long total_forks;      /* Handle normal Linux uptimes. */
 124 int nr_threads;                 /* The idle threads do not count.. */
 125
 126 static int max_threads;         /* tunable limit on nr_threads */
 127
 128 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 129
 130 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 131
 132 #ifdef CONFIG_PROVE_RCU
 133 int lockdep_tasklist_lock_is_held(void)
 134 {
 135         return lockdep_is_held(&tasklist_lock);
 136 }
 137 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 138 #endif /* #ifdef CONFIG_PROVE_RCU */
 139
 140 int nr_processes(void)
 141 {
 142         int cpu;
 143         int total = 0;
 144
 145         for_each_possible_cpu(cpu)
 146                 total += per_cpu(process_counts, cpu);
 147
 148         return total;
 149 }
 150
 151 void __weak arch_release_task_struct(struct task_struct *tsk)
 152 {
 153 }
 154
 155 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 156 static struct kmem_cache *task_struct_cachep;
 157
 158 static inline struct task_struct *alloc_task_struct_node(int node)
 159 {
 160         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 161 }
 162
 163 static inline void free_task_struct(struct task_struct *tsk)
 164 {
 165         kmem_cache_free(task_struct_cachep, tsk);
 166 }
 167 #endif
 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                 tsk->stack = s->addr;
 223                 return s->addr;
 224         }
 225
 226         /*
 227          * Allocated stacks are cached and later reused by new threads,
 228          * so memcg accounting is performed manually on assigning/releasing
 229          * stacks to tasks. Drop __GFP_ACCOUNT.
 230          */
 231         stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
 232                                      VMALLOC_START, VMALLOC_END,
 233                                      THREADINFO_GFP & ~__GFP_ACCOUNT,
 234                                      PAGE_KERNEL,
 235                                      0, node, __builtin_return_address(0));
 236
 237         /*
 238          * We can't call find_vm_area() in interrupt context, and
 239          * free_thread_stack() can be called in interrupt context,
 240          * so cache the vm_struct.
 241          */
 242         if (stack) {
 243                 tsk->stack_vm_area = find_vm_area(stack);
 244                 tsk->stack = stack;
 245         }
 246         return stack;
 247 #else
 248         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
 249                                              THREAD_SIZE_ORDER);
 250
 251         if (likely(page)) {
 252                 tsk->stack = page_address(page);
 253                 return tsk->stack;
 254         }
 255         return NULL;
 256 #endif
 257 }
 258
 259 static inline void free_thread_stack(struct task_struct *tsk)
 260 {
 261 #ifdef CONFIG_VMAP_STACK
 262         struct vm_struct *vm = task_stack_vm_area(tsk);
 263
 264         if (vm) {
 265                 int i;
 266
 267                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 268                         mod_memcg_page_state(vm->pages[i],
 269                                              MEMCG_KERNEL_STACK_KB,
 270                                              -(int)(PAGE_SIZE / 1024));
 271
 272                         memcg_kmem_uncharge(vm->pages[i], 0);
 273                 }
 274
 275                 for (i = 0; i < NR_CACHED_STACKS; i++) {
 276                         if (this_cpu_cmpxchg(cached_stacks[i],
 277                                         NULL, tsk->stack_vm_area) != NULL)
 278                                 continue;
 279
 280                         return;
 281                 }
 282
 283                 vfree_atomic(tsk->stack);
 284                 return;
 285         }
 286 #endif
 287
 288         __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
 289 }
 290 # else
 291 static struct kmem_cache *thread_stack_cache;
 292
 293 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
 294                                                   int node)
 295 {
 296         unsigned long *stack;
 297         stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
 298         tsk->stack = stack;
 299         return stack;
 300 }
 301
 302 static void free_thread_stack(struct task_struct *tsk)
 303 {
 304         kmem_cache_free(thread_stack_cache, tsk->stack);
 305 }
 306
 307 void thread_stack_cache_init(void)
 308 {
 309         thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
 310                                         THREAD_SIZE, THREAD_SIZE, 0, 0,
 311                                         THREAD_SIZE, NULL);
 312         BUG_ON(thread_stack_cache == NULL);
 313 }
 314 # endif
 315 #endif
 316
 317 /* SLAB cache for signal_struct structures (tsk->signal) */
 318 static struct kmem_cache *signal_cachep;
 319
 320 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 321 struct kmem_cache *sighand_cachep;
 322
 323 /* SLAB cache for files_struct structures (tsk->files) */
 324 struct kmem_cache *files_cachep;
 325
 326 /* SLAB cache for fs_struct structures (tsk->fs) */
 327 struct kmem_cache *fs_cachep;
 328
 329 /* SLAB cache for vm_area_struct structures */
 330 static struct kmem_cache *vm_area_cachep;
 331
 332 /* SLAB cache for mm_struct structures (tsk->mm) */
 333 static struct kmem_cache *mm_cachep;
 334
 335 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
 336 {
 337         struct vm_area_struct *vma;
 338
 339         vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 340         if (vma)
 341                 vma_init(vma, mm);
 342         return vma;
 343 }
 344
 345 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
 346 {
 347         struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 348
 349         if (new) {
 350                 *new = *orig;
 351                 INIT_LIST_HEAD(&new->anon_vma_chain);
 352         }
 353         return new;
 354 }
 355
 356 void vm_area_free(struct vm_area_struct *vma)
 357 {
 358         kmem_cache_free(vm_area_cachep, vma);
 359 }
 360
 361 static void account_kernel_stack(struct task_struct *tsk, int account)
 362 {
 363         void *stack = task_stack_page(tsk);
 364         struct vm_struct *vm = task_stack_vm_area(tsk);
 365
 366         BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
 367
 368         if (vm) {
 369                 int i;
 370
 371                 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
 372
 373                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 374                         mod_zone_page_state(page_zone(vm->pages[i]),
 375                                             NR_KERNEL_STACK_KB,
 376                                             PAGE_SIZE / 1024 * account);
 377                 }
 378         } else {
 379                 /*
 380                  * All stack pages are in the same zone and belong to the
 381                  * same memcg.
 382                  */
 383                 struct page *first_page = virt_to_page(stack);
 384
 385                 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
 386                                     THREAD_SIZE / 1024 * account);
 387
 388                 mod_memcg_page_state(first_page, MEMCG_KERNEL_STACK_KB,
 389                                      account * (THREAD_SIZE / 1024));
 390         }
 391 }
 392
 393 static int memcg_charge_kernel_stack(struct task_struct *tsk)
 394 {
 395 #ifdef CONFIG_VMAP_STACK
 396         struct vm_struct *vm = task_stack_vm_area(tsk);
 397         int ret;
 398
 399         if (vm) {
 400                 int i;
 401
 402                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 403                         /*
 404                          * If memcg_kmem_charge() fails, page->mem_cgroup
 405                          * pointer is NULL, and both memcg_kmem_uncharge()
 406                          * and mod_memcg_page_state() in free_thread_stack()
 407                          * will ignore this page. So it's safe.
 408                          */
 409                         ret = memcg_kmem_charge(vm->pages[i], GFP_KERNEL, 0);
 410                         if (ret)
 411                                 return ret;
 412
 413                         mod_memcg_page_state(vm->pages[i],
 414                                              MEMCG_KERNEL_STACK_KB,
 415                                              PAGE_SIZE / 1024);
 416                 }
 417         }
 418 #endif
 419         return 0;
 420 }
 421
 422 static void release_task_stack(struct task_struct *tsk)
 423 {
 424         if (WARN_ON(tsk->state != TASK_DEAD))
 425                 return;  /* Better to leak the stack than to free prematurely */
 426
 427         account_kernel_stack(tsk, -1);
 428         free_thread_stack(tsk);
 429         tsk->stack = NULL;
 430 #ifdef CONFIG_VMAP_STACK
 431         tsk->stack_vm_area = NULL;
 432 #endif
 433 }
 434
 435 #ifdef CONFIG_THREAD_INFO_IN_TASK
 436 void put_task_stack(struct task_struct *tsk)
 437 {
 438         if (refcount_dec_and_test(&tsk->stack_refcount))
 439                 release_task_stack(tsk);
 440 }
 441 #endif
 442
 443 void free_task(struct task_struct *tsk)
 444 {
 445 #ifndef CONFIG_THREAD_INFO_IN_TASK
 446         /*
 447          * The task is finally done with both the stack and thread_info,
 448          * so free both.
 449          */
 450         release_task_stack(tsk);
 451 #else
 452         /*
 453          * If the task had a separate stack allocation, it should be gone
 454          * by now.
 455          */
 456         WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
 457 #endif
 458         rt_mutex_debug_task_free(tsk);
 459         ftrace_graph_exit_task(tsk);
 460         put_seccomp_filter(tsk);
 461         arch_release_task_struct(tsk);
 462         if (tsk->flags & PF_KTHREAD)
 463                 free_kthread_struct(tsk);
 464         free_task_struct(tsk);
 465 }
 466 EXPORT_SYMBOL(free_task);
 467
 468 #ifdef CONFIG_MMU
 469 static __latent_entropy int dup_mmap(struct mm_struct *mm,
 470                                         struct mm_struct *oldmm)
 471 {
 472         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 473         struct rb_node **rb_link, *rb_parent;
 474         int retval;
 475         unsigned long charge;
 476         LIST_HEAD(uf);
 477
 478         uprobe_start_dup_mmap();
 479         if (down_write_killable(&oldmm->mmap_sem)) {
 480                 retval = -EINTR;
 481                 goto fail_uprobe_end;
 482         }
 483         flush_cache_dup_mm(oldmm);
 484         uprobe_dup_mmap(oldmm, mm);
 485         /*
 486          * Not linked in yet - no deadlock potential:
 487          */
 488         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 489
 490         /* No ordering required: file already has been exposed. */
 491         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 492
 493         mm->total_vm = oldmm->total_vm;
 494         mm->data_vm = oldmm->data_vm;
 495         mm->exec_vm = oldmm->exec_vm;
 496         mm->stack_vm = oldmm->stack_vm;
 497
 498         rb_link = &mm->mm_rb.rb_node;
 499         rb_parent = NULL;
 500         pprev = &mm->mmap;
 501         retval = ksm_fork(mm, oldmm);
 502         if (retval)
 503                 goto out;
 504         retval = khugepaged_fork(mm, oldmm);
 505         if (retval)
 506                 goto out;
 507
 508         prev = NULL;
 509         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 510                 struct file *file;
 511
 512                 if (mpnt->vm_flags & VM_DONTCOPY) {
 513                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 514                         continue;
 515                 }
 516                 charge = 0;
 517                 /*
 518                  * Don't duplicate many vmas if we've been oom-killed (for
 519                  * example)
 520                  */
 521                 if (fatal_signal_pending(current)) {
 522                         retval = -EINTR;
 523                         goto out;
 524                 }
 525                 if (mpnt->vm_flags & VM_ACCOUNT) {
 526                         unsigned long len = vma_pages(mpnt);
 527
 528                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 529                                 goto fail_nomem;
 530                         charge = len;
 531                 }
 532                 tmp = vm_area_dup(mpnt);
 533                 if (!tmp)
 534                         goto fail_nomem;
 535                 retval = vma_dup_policy(mpnt, tmp);
 536                 if (retval)
 537                         goto fail_nomem_policy;
 538                 tmp->vm_mm = mm;
 539                 retval = dup_userfaultfd(tmp, &uf);
 540                 if (retval)
 541                         goto fail_nomem_anon_vma_fork;
 542                 if (tmp->vm_flags & VM_WIPEONFORK) {
 543                         /* VM_WIPEONFORK gets a clean slate in the child. */
 544                         tmp->anon_vma = NULL;
 545                         if (anon_vma_prepare(tmp))
 546                                 goto fail_nomem_anon_vma_fork;
 547                 } else if (anon_vma_fork(tmp, mpnt))
 548                         goto fail_nomem_anon_vma_fork;
 549                 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
 550                 tmp->vm_next = tmp->vm_prev = NULL;
 551                 file = tmp->vm_file;
 552                 if (file) {
 553                         struct inode *inode = file_inode(file);
 554                         struct address_space *mapping = file->f_mapping;
 555
 556                         get_file(file);
 557                         if (tmp->vm_flags & VM_DENYWRITE)
 558                                 atomic_dec(&inode->i_writecount);
 559                         i_mmap_lock_write(mapping);
 560                         if (tmp->vm_flags & VM_SHARED)
 561                                 atomic_inc(&mapping->i_mmap_writable);
 562                         flush_dcache_mmap_lock(mapping);
 563                         /* insert tmp into the share list, just after mpnt */
 564                         vma_interval_tree_insert_after(tmp, mpnt,
 565                                         &mapping->i_mmap);
 566                         flush_dcache_mmap_unlock(mapping);
 567                         i_mmap_unlock_write(mapping);
 568                 }
 569
 570                 /*
 571                  * Clear hugetlb-related page reserves for children. This only
 572                  * affects MAP_PRIVATE mappings. Faults generated by the child
 573                  * are not guaranteed to succeed, even if read-only
 574                  */
 575                 if (is_vm_hugetlb_page(tmp))
 576                         reset_vma_resv_huge_pages(tmp);
 577
 578                 /*
 579                  * Link in the new vma and copy the page table entries.
 580                  */
 581                 *pprev = tmp;
 582                 pprev = &tmp->vm_next;
 583                 tmp->vm_prev = prev;
 584                 prev = tmp;
 585
 586                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 587                 rb_link = &tmp->vm_rb.rb_right;
 588                 rb_parent = &tmp->vm_rb;
 589
 590                 mm->map_count++;
 591                 if (!(tmp->vm_flags & VM_WIPEONFORK))
 592                         retval = copy_page_range(mm, oldmm, mpnt);
 593
 594                 if (tmp->vm_ops && tmp->vm_ops->open)
 595                         tmp->vm_ops->open(tmp);
 596
 597                 if (retval)
 598                         goto out;
 599         }
 600         /* a new mm has just been created */
 601         retval = arch_dup_mmap(oldmm, mm);
 602 out:
 603         up_write(&mm->mmap_sem);
 604         flush_tlb_mm(oldmm);
 605         up_write(&oldmm->mmap_sem);
 606         dup_userfaultfd_complete(&uf);
 607 fail_uprobe_end:
 608         uprobe_end_dup_mmap();
 609         return retval;
 610 fail_nomem_anon_vma_fork:
 611         mpol_put(vma_policy(tmp));
 612 fail_nomem_policy:
 613         vm_area_free(tmp);
 614 fail_nomem:
 615         retval = -ENOMEM;
 616         vm_unacct_memory(charge);
 617         goto out;
 618 }
 619
 620 static inline int mm_alloc_pgd(struct mm_struct *mm)
 621 {
 622         mm->pgd = pgd_alloc(mm);
 623         if (unlikely(!mm->pgd))
 624                 return -ENOMEM;
 625         return 0;
 626 }
 627
 628 static inline void mm_free_pgd(struct mm_struct *mm)
 629 {
 630         pgd_free(mm, mm->pgd);
 631 }
 632 #else
 633 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 634 {
 635         down_write(&oldmm->mmap_sem);
 636         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 637         up_write(&oldmm->mmap_sem);
 638         return 0;
 639 }
 640 #define mm_alloc_pgd(mm)        (0)
 641 #define mm_free_pgd(mm)
 642 #endif /* CONFIG_MMU */
 643
 644 static void check_mm(struct mm_struct *mm)
 645 {
 646         int i;
 647
 648         for (i = 0; i < NR_MM_COUNTERS; i++) {
 649                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 650
 651                 if (unlikely(x))
 652                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 653                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 654         }
 655
 656         if (mm_pgtables_bytes(mm))
 657                 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
 658                                 mm_pgtables_bytes(mm));
 659
 660 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 661         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 662 #endif
 663 }
 664
 665 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 666 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 667
 668 /*
 669  * Called when the last reference to the mm
 670  * is dropped: either by a lazy thread or by
 671  * mmput. Free the page directory and the mm.
 672  */
 673 void __mmdrop(struct mm_struct *mm)
 674 {
 675         BUG_ON(mm == &init_mm);
 676         WARN_ON_ONCE(mm == current->mm);
 677         WARN_ON_ONCE(mm == current->active_mm);
 678         mm_free_pgd(mm);
 679         destroy_context(mm);
 680         hmm_mm_destroy(mm);
 681         mmu_notifier_mm_destroy(mm);
 682         check_mm(mm);
 683         put_user_ns(mm->user_ns);
 684         free_mm(mm);
 685 }
 686 EXPORT_SYMBOL_GPL(__mmdrop);
 687
 688 static void mmdrop_async_fn(struct work_struct *work)
 689 {
 690         struct mm_struct *mm;
 691
 692         mm = container_of(work, struct mm_struct, async_put_work);
 693         __mmdrop(mm);
 694 }
 695
 696 static void mmdrop_async(struct mm_struct *mm)
 697 {
 698         if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
 699                 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
 700                 schedule_work(&mm->async_put_work);
 701         }
 702 }
 703
 704 static inline void free_signal_struct(struct signal_struct *sig)
 705 {
 706         taskstats_tgid_free(sig);
 707         sched_autogroup_exit(sig);
 708         /*
 709          * __mmdrop is not safe to call from softirq context on x86 due to
 710          * pgd_dtor so postpone it to the async context
 711          */
 712         if (sig->oom_mm)
 713                 mmdrop_async(sig->oom_mm);
 714         kmem_cache_free(signal_cachep, sig);
 715 }
 716
 717 static inline void put_signal_struct(struct signal_struct *sig)
 718 {
 719         if (refcount_dec_and_test(&sig->sigcnt))
 720                 free_signal_struct(sig);
 721 }
 722
 723 void __put_task_struct(struct task_struct *tsk)
 724 {
 725         WARN_ON(!tsk->exit_state);
 726         WARN_ON(refcount_read(&tsk->usage));
 727         WARN_ON(tsk == current);
 728
 729         cgroup_free(tsk);
 730         task_numa_free(tsk);
 731         security_task_free(tsk);
 732         exit_creds(tsk);
 733         delayacct_tsk_free(tsk);
 734         put_signal_struct(tsk->signal);
 735
 736         if (!profile_handoff_task(tsk))
 737                 free_task(tsk);
 738 }
 739 EXPORT_SYMBOL_GPL(__put_task_struct);
 740
 741 void __init __weak arch_task_cache_init(void) { }
 742
 743 /*
 744  * set_max_threads
 745  */
 746 static void set_max_threads(unsigned int max_threads_suggested)
 747 {
 748         u64 threads;
 749         unsigned long nr_pages = totalram_pages();
 750
 751         /*
 752          * The number of threads shall be limited such that the thread
 753          * structures may only consume a small part of the available memory.
 754          */
 755         if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
 756                 threads = MAX_THREADS;
 757         else
 758                 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
 759                                     (u64) THREAD_SIZE * 8UL);
 760
 761         if (threads > max_threads_suggested)
 762                 threads = max_threads_suggested;
 763
 764         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 765 }
 766
 767 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 768 /* Initialized by the architecture: */
 769 int arch_task_struct_size __read_mostly;
 770 #endif
 771
 772 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
 773 {
 774         /* Fetch thread_struct whitelist for the architecture. */
 775         arch_thread_struct_whitelist(offset, size);
 776
 777         /*
 778          * Handle zero-sized whitelist or empty thread_struct, otherwise
 779          * adjust offset to position of thread_struct in task_struct.
 780          */
 781         if (unlikely(*size == 0))
 782                 *offset = 0;
 783         else
 784                 *offset += offsetof(struct task_struct, thread);
 785 }
 786
 787 void __init fork_init(void)
 788 {
 789         int i;
 790 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 791 #ifndef ARCH_MIN_TASKALIGN
 792 #define ARCH_MIN_TASKALIGN      0
 793 #endif
 794         int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
 795         unsigned long useroffset, usersize;
 796
 797         /* create a slab on which task_structs can be allocated */
 798         task_struct_whitelist(&useroffset, &usersize);
 799         task_struct_cachep = kmem_cache_create_usercopy("task_struct",
 800                         arch_task_struct_size, align,
 801                         SLAB_PANIC|SLAB_ACCOUNT,
 802                         useroffset, usersize, NULL);
 803 #endif
 804
 805         /* do the arch specific task caches init */
 806         arch_task_cache_init();
 807
 808         set_max_threads(MAX_THREADS);
 809
 810         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 811         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 812         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 813                 init_task.signal->rlim[RLIMIT_NPROC];
 814
 815         for (i = 0; i < UCOUNT_COUNTS; i++) {
 816                 init_user_ns.ucount_max[i] = max_threads/2;
 817         }
 818
 819 #ifdef CONFIG_VMAP_STACK
 820         cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
 821                           NULL, free_vm_stack_cache);
 822 #endif
 823
 824         lockdep_init_task(&init_task);
 825         uprobes_init();
 826 }
 827
 828 int __weak arch_dup_task_struct(struct task_struct *dst,
 829                                                struct task_struct *src)
 830 {
 831         *dst = *src;
 832         return 0;
 833 }
 834
 835 void set_task_stack_end_magic(struct task_struct *tsk)
 836 {
 837         unsigned long *stackend;
 838
 839         stackend = end_of_stack(tsk);
 840         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 841 }
 842
 843 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 844 {
 845         struct task_struct *tsk;
 846         unsigned long *stack;
 847         struct vm_struct *stack_vm_area __maybe_unused;
 848         int err;
 849
 850         if (node == NUMA_NO_NODE)
 851                 node = tsk_fork_get_node(orig);
 852         tsk = alloc_task_struct_node(node);
 853         if (!tsk)
 854                 return NULL;
 855
 856         stack = alloc_thread_stack_node(tsk, node);
 857         if (!stack)
 858                 goto free_tsk;
 859
 860         if (memcg_charge_kernel_stack(tsk))
 861                 goto free_stack;
 862
 863         stack_vm_area = task_stack_vm_area(tsk);
 864
 865         err = arch_dup_task_struct(tsk, orig);
 866
 867         /*
 868          * arch_dup_task_struct() clobbers the stack-related fields.  Make
 869          * sure they're properly initialized before using any stack-related
 870          * functions again.
 871          */
 872         tsk->stack = stack;
 873 #ifdef CONFIG_VMAP_STACK
 874         tsk->stack_vm_area = stack_vm_area;
 875 #endif
 876 #ifdef CONFIG_THREAD_INFO_IN_TASK
 877         refcount_set(&tsk->stack_refcount, 1);
 878 #endif
 879
 880         if (err)
 881                 goto free_stack;
 882
 883 #ifdef CONFIG_SECCOMP
 884         /*
 885          * We must handle setting up seccomp filters once we're under
 886          * the sighand lock in case orig has changed between now and
 887          * then. Until then, filter must be NULL to avoid messing up
 888          * the usage counts on the error path calling free_task.
 889          */
 890         tsk->seccomp.filter = NULL;
 891 #endif
 892
 893         setup_thread_stack(tsk, orig);
 894         clear_user_return_notifier(tsk);
 895         clear_tsk_need_resched(tsk);
 896         set_task_stack_end_magic(tsk);
 897
 898 #ifdef CONFIG_STACKPROTECTOR
 899         tsk->stack_canary = get_random_canary();
 900 #endif
 901         if (orig->cpus_ptr == &orig->cpus_mask)
 902                 tsk->cpus_ptr = &tsk->cpus_mask;
 903
 904         /*
 905          * One for us, one for whoever does the "release_task()" (usually
 906          * parent)
 907          */
 908         refcount_set(&tsk->usage, 2);
 909 #ifdef CONFIG_BLK_DEV_IO_TRACE
 910         tsk->btrace_seq = 0;
 911 #endif
 912         tsk->splice_pipe = NULL;
 913         tsk->task_frag.page = NULL;
 914         tsk->wake_q.next = NULL;
 915
 916         account_kernel_stack(tsk, 1);
 917
 918         kcov_task_init(tsk);
 919
 920 #ifdef CONFIG_FAULT_INJECTION
 921         tsk->fail_nth = 0;
 922 #endif
 923
 924 #ifdef CONFIG_BLK_CGROUP
 925         tsk->throttle_queue = NULL;
 926         tsk->use_memdelay = 0;
 927 #endif
 928
 929 #ifdef CONFIG_MEMCG
 930         tsk->active_memcg = NULL;
 931 #endif
 932         return tsk;
 933
 934 free_stack:
 935         free_thread_stack(tsk);
 936 free_tsk:
 937         free_task_struct(tsk);
 938         return NULL;
 939 }
 940
 941 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 942
 943 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 944
 945 static int __init coredump_filter_setup(char *s)
 946 {
 947         default_dump_filter =
 948                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 949                 MMF_DUMP_FILTER_MASK;
 950         return 1;
 951 }
 952
 953 __setup("coredump_filter=", coredump_filter_setup);
 954
 955 #include <linux/init_task.h>
 956
 957 static void mm_init_aio(struct mm_struct *mm)
 958 {
 959 #ifdef CONFIG_AIO
 960         spin_lock_init(&mm->ioctx_lock);
 961         mm->ioctx_table = NULL;
 962 #endif
 963 }
 964
 965 static __always_inline void mm_clear_owner(struct mm_struct *mm,
 966                                            struct task_struct *p)
 967 {
 968 #ifdef CONFIG_MEMCG
 969         if (mm->owner == p)
 970                 WRITE_ONCE(mm->owner, NULL);
 971 #endif
 972 }
 973
 974 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 975 {
 976 #ifdef CONFIG_MEMCG
 977         mm->owner = p;
 978 #endif
 979 }
 980
 981 static void mm_init_uprobes_state(struct mm_struct *mm)
 982 {
 983 #ifdef CONFIG_UPROBES
 984         mm->uprobes_state.xol_area = NULL;
 985 #endif
 986 }
 987
 988 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
 989         struct user_namespace *user_ns)
 990 {
 991         mm->mmap = NULL;
 992         mm->mm_rb = RB_ROOT;
 993         mm->vmacache_seqnum = 0;
 994         atomic_set(&mm->mm_users, 1);
 995         atomic_set(&mm->mm_count, 1);
 996         init_rwsem(&mm->mmap_sem);
 997         INIT_LIST_HEAD(&mm->mmlist);
 998         mm->core_state = NULL;
 999         mm_pgtables_bytes_init(mm);
1000         mm->map_count = 0;
1001         mm->locked_vm = 0;
1002         atomic64_set(&mm->pinned_vm, 0);
1003         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1004         spin_lock_init(&mm->page_table_lock);
1005         spin_lock_init(&mm->arg_lock);
1006         mm_init_cpumask(mm);
1007         mm_init_aio(mm);
1008         mm_init_owner(mm, p);
1009         RCU_INIT_POINTER(mm->exe_file, NULL);
1010         mmu_notifier_mm_init(mm);
1011         hmm_mm_init(mm);
1012         init_tlb_flush_pending(mm);
1013 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1014         mm->pmd_huge_pte = NULL;
1015 #endif
1016         mm_init_uprobes_state(mm);
1017
1018         if (current->mm) {
1019                 mm->flags = current->mm->flags & MMF_INIT_MASK;
1020                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1021         } else {
1022                 mm->flags = default_dump_filter;
1023                 mm->def_flags = 0;
1024         }
1025
1026         if (mm_alloc_pgd(mm))
1027                 goto fail_nopgd;
1028
1029         if (init_new_context(p, mm))
1030                 goto fail_nocontext;
1031
1032         mm->user_ns = get_user_ns(user_ns);
1033         return mm;
1034
1035 fail_nocontext:
1036         mm_free_pgd(mm);
1037 fail_nopgd:
1038         free_mm(mm);
1039         return NULL;
1040 }
1041
1042 /*
1043  * Allocate and initialize an mm_struct.
1044  */
1045 struct mm_struct *mm_alloc(void)
1046 {
1047         struct mm_struct *mm;
1048
1049         mm = allocate_mm();
1050         if (!mm)
1051                 return NULL;
1052
1053         memset(mm, 0, sizeof(*mm));
1054         return mm_init(mm, current, current_user_ns());
1055 }
1056
1057 static inline void __mmput(struct mm_struct *mm)
1058 {
1059         VM_BUG_ON(atomic_read(&mm->mm_users));
1060
1061         uprobe_clear_state(mm);
1062         exit_aio(mm);
1063         ksm_exit(mm);
1064         khugepaged_exit(mm); /* must run before exit_mmap */
1065         exit_mmap(mm);
1066         mm_put_huge_zero_page(mm);
1067         set_mm_exe_file(mm, NULL);
1068         if (!list_empty(&mm->mmlist)) {
1069                 spin_lock(&mmlist_lock);
1070                 list_del(&mm->mmlist);
1071                 spin_unlock(&mmlist_lock);
1072         }
1073         if (mm->binfmt)
1074                 module_put(mm->binfmt->module);
1075         mmdrop(mm);
1076 }
1077
1078 /*
1079  * Decrement the use count and release all resources for an mm.
1080  */
1081 void mmput(struct mm_struct *mm)
1082 {
1083         might_sleep();
1084
1085         if (atomic_dec_and_test(&mm->mm_users))
1086                 __mmput(mm);
1087 }
1088 EXPORT_SYMBOL_GPL(mmput);
1089
1090 #ifdef CONFIG_MMU
1091 static void mmput_async_fn(struct work_struct *work)
1092 {
1093         struct mm_struct *mm = container_of(work, struct mm_struct,
1094                                             async_put_work);
1095
1096         __mmput(mm);
1097 }
1098
1099 void mmput_async(struct mm_struct *mm)
1100 {
1101         if (atomic_dec_and_test(&mm->mm_users)) {
1102                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1103                 schedule_work(&mm->async_put_work);
1104         }
1105 }
1106 #endif
1107
1108 /**
1109  * set_mm_exe_file - change a reference to the mm's executable file
1110  *
1111  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1112  *
1113  * Main users are mmput() and sys_execve(). Callers prevent concurrent
1114  * invocations: in mmput() nobody alive left, in execve task is single
1115  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1116  * mm->exe_file, but does so without using set_mm_exe_file() in order
1117  * to do avoid the need for any locks.
1118  */
1119 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1120 {
1121         struct file *old_exe_file;
1122
1123         /*
1124          * It is safe to dereference the exe_file without RCU as
1125          * this function is only called if nobody else can access
1126          * this mm -- see comment above for justification.
1127          */
1128         old_exe_file = rcu_dereference_raw(mm->exe_file);
1129
1130         if (new_exe_file)
1131                 get_file(new_exe_file);
1132         rcu_assign_pointer(mm->exe_file, new_exe_file);
1133         if (old_exe_file)
1134                 fput(old_exe_file);
1135 }
1136
1137 /**
1138  * get_mm_exe_file - acquire a reference to the mm's executable file
1139  *
1140  * Returns %NULL if mm has no associated executable file.
1141  * User must release file via fput().
1142  */
1143 struct file *get_mm_exe_file(struct mm_struct *mm)
1144 {
1145         struct file *exe_file;
1146
1147         rcu_read_lock();
1148         exe_file = rcu_dereference(mm->exe_file);
1149         if (exe_file && !get_file_rcu(exe_file))
1150                 exe_file = NULL;
1151         rcu_read_unlock();
1152         return exe_file;
1153 }
1154 EXPORT_SYMBOL(get_mm_exe_file);
1155
1156 /**
1157  * get_task_exe_file - acquire a reference to the task's executable file
1158  *
1159  * Returns %NULL if task's mm (if any) has no associated executable file or
1160  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1161  * User must release file via fput().
1162  */
1163 struct file *get_task_exe_file(struct task_struct *task)
1164 {
1165         struct file *exe_file = NULL;
1166         struct mm_struct *mm;
1167
1168         task_lock(task);
1169         mm = task->mm;
1170         if (mm) {
1171                 if (!(task->flags & PF_KTHREAD))
1172                         exe_file = get_mm_exe_file(mm);
1173         }
1174         task_unlock(task);
1175         return exe_file;
1176 }
1177 EXPORT_SYMBOL(get_task_exe_file);
1178
1179 /**
1180  * get_task_mm - acquire a reference to the task's mm
1181  *
1182  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1183  * this kernel workthread has transiently adopted a user mm with use_mm,
1184  * to do its AIO) is not set and if so returns a reference to it, after
1185  * bumping up the use count.  User must release the mm via mmput()
1186  * after use.  Typically used by /proc and ptrace.
1187  */
1188 struct mm_struct *get_task_mm(struct task_struct *task)
1189 {
1190         struct mm_struct *mm;
1191
1192         task_lock(task);
1193         mm = task->mm;
1194         if (mm) {
1195                 if (task->flags & PF_KTHREAD)
1196                         mm = NULL;
1197                 else
1198                         mmget(mm);
1199         }
1200         task_unlock(task);
1201         return mm;
1202 }
1203 EXPORT_SYMBOL_GPL(get_task_mm);
1204
1205 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1206 {
1207         struct mm_struct *mm;
1208         int err;
1209
1210         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
1211         if (err)
1212                 return ERR_PTR(err);
1213
1214         mm = get_task_mm(task);
1215         if (mm && mm != current->mm &&
1216                         !ptrace_may_access(task, mode)) {
1217                 mmput(mm);
1218                 mm = ERR_PTR(-EACCES);
1219         }
1220         mutex_unlock(&task->signal->cred_guard_mutex);
1221
1222         return mm;
1223 }
1224
1225 static void complete_vfork_done(struct task_struct *tsk)
1226 {
1227         struct completion *vfork;
1228
1229         task_lock(tsk);
1230         vfork = tsk->vfork_done;
1231         if (likely(vfork)) {
1232                 tsk->vfork_done = NULL;
1233                 complete(vfork);
1234         }
1235         task_unlock(tsk);
1236 }
1237
1238 static int wait_for_vfork_done(struct task_struct *child,
1239                                 struct completion *vfork)
1240 {
1241         int killed;
1242
1243         freezer_do_not_count();
1244         cgroup_enter_frozen();
1245         killed = wait_for_completion_killable(vfork);
1246         cgroup_leave_frozen(false);
1247         freezer_count();
1248
1249         if (killed) {
1250                 task_lock(child);
1251                 child->vfork_done = NULL;
1252                 task_unlock(child);
1253         }
1254
1255         put_task_struct(child);
1256         return killed;
1257 }
1258
1259 /* Please note the differences between mmput and mm_release.
1260  * mmput is called whenever we stop holding onto a mm_struct,
1261  * error success whatever.
1262  *
1263  * mm_release is called after a mm_struct has been removed
1264  * from the current process.
1265  *
1266  * This difference is important for error handling, when we
1267  * only half set up a mm_struct for a new process and need to restore
1268  * the old one.  Because we mmput the new mm_struct before
1269  * restoring the old one. . .
1270  * Eric Biederman 10 January 1998
1271  */
1272 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1273 {
1274         /* Get rid of any futexes when releasing the mm */
1275 #ifdef CONFIG_FUTEX
1276         if (unlikely(tsk->robust_list)) {
1277                 exit_robust_list(tsk);
1278                 tsk->robust_list = NULL;
1279         }
1280 #ifdef CONFIG_COMPAT
1281         if (unlikely(tsk->compat_robust_list)) {
1282                 compat_exit_robust_list(tsk);
1283                 tsk->compat_robust_list = NULL;
1284         }
1285 #endif
1286         if (unlikely(!list_empty(&tsk->pi_state_list)))
1287                 exit_pi_state_list(tsk);
1288 #endif
1289
1290         uprobe_free_utask(tsk);
1291
1292         /* Get rid of any cached register state */
1293         deactivate_mm(tsk, mm);
1294
1295         /*
1296          * Signal userspace if we're not exiting with a core dump
1297          * because we want to leave the value intact for debugging
1298          * purposes.
1299          */
1300         if (tsk->clear_child_tid) {
1301                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1302                     atomic_read(&mm->mm_users) > 1) {
1303                         /*
1304                          * We don't check the error code - if userspace has
1305                          * not set up a proper pointer then tough luck.
1306                          */
1307                         put_user(0, tsk->clear_child_tid);
1308                         do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1309                                         1, NULL, NULL, 0, 0);
1310                 }
1311                 tsk->clear_child_tid = NULL;
1312         }
1313
1314         /*
1315          * All done, finally we can wake up parent and return this mm to him.
1316          * Also kthread_stop() uses this completion for synchronization.
1317          */
1318         if (tsk->vfork_done)
1319                 complete_vfork_done(tsk);
1320 }
1321
1322 /**
1323  * dup_mm() - duplicates an existing mm structure
1324  * @tsk: the task_struct with which the new mm will be associated.
1325  * @oldmm: the mm to duplicate.
1326  *
1327  * Allocates a new mm structure and duplicates the provided @oldmm structure
1328  * content into it.
1329  *
1330  * Return: the duplicated mm or NULL on failure.
1331  */
1332 static struct mm_struct *dup_mm(struct task_struct *tsk,
1333                                 struct mm_struct *oldmm)
1334 {
1335         struct mm_struct *mm;
1336         int err;
1337
1338         mm = allocate_mm();
1339         if (!mm)
1340                 goto fail_nomem;
1341
1342         memcpy(mm, oldmm, sizeof(*mm));
1343
1344         if (!mm_init(mm, tsk, mm->user_ns))
1345                 goto fail_nomem;
1346
1347         err = dup_mmap(mm, oldmm);
1348         if (err)
1349                 goto free_pt;
1350
1351         mm->hiwater_rss = get_mm_rss(mm);
1352         mm->hiwater_vm = mm->total_vm;
1353
1354         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1355                 goto free_pt;
1356
1357         return mm;
1358
1359 free_pt:
1360         /* don't put binfmt in mmput, we haven't got module yet */
1361         mm->binfmt = NULL;
1362         mm_init_owner(mm, NULL);
1363         mmput(mm);
1364
1365 fail_nomem:
1366         return NULL;
1367 }
1368
1369 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1370 {
1371         struct mm_struct *mm, *oldmm;
1372         int retval;
1373
1374         tsk->min_flt = tsk->maj_flt = 0;
1375         tsk->nvcsw = tsk->nivcsw = 0;
1376 #ifdef CONFIG_DETECT_HUNG_TASK
1377         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1378         tsk->last_switch_time = 0;
1379 #endif
1380
1381         tsk->mm = NULL;
1382         tsk->active_mm = NULL;
1383
1384         /*
1385          * Are we cloning a kernel thread?
1386          *
1387          * We need to steal a active VM for that..
1388          */
1389         oldmm = current->mm;
1390         if (!oldmm)
1391                 return 0;
1392
1393         /* initialize the new vmacache entries */
1394         vmacache_flush(tsk);
1395
1396         if (clone_flags & CLONE_VM) {
1397                 mmget(oldmm);
1398                 mm = oldmm;
1399                 goto good_mm;
1400         }
1401
1402         retval = -ENOMEM;
1403         mm = dup_mm(tsk, current->mm);
1404         if (!mm)
1405                 goto fail_nomem;
1406
1407 good_mm:
1408         tsk->mm = mm;
1409         tsk->active_mm = mm;
1410         return 0;
1411
1412 fail_nomem:
1413         return retval;
1414 }
1415
1416 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1417 {
1418         struct fs_struct *fs = current->fs;
1419         if (clone_flags & CLONE_FS) {
1420                 /* tsk->fs is already what we want */
1421                 spin_lock(&fs->lock);
1422                 if (fs->in_exec) {
1423                         spin_unlock(&fs->lock);
1424                         return -EAGAIN;
1425                 }
1426                 fs->users++;
1427                 spin_unlock(&fs->lock);
1428                 return 0;
1429         }
1430         tsk->fs = copy_fs_struct(fs);
1431         if (!tsk->fs)
1432                 return -ENOMEM;
1433         return 0;
1434 }
1435
1436 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1437 {
1438         struct files_struct *oldf, *newf;
1439         int error = 0;
1440
1441         /*
1442          * A background process may not have any files ...
1443          */
1444         oldf = current->files;
1445         if (!oldf)
1446                 goto out;
1447
1448         if (clone_flags & CLONE_FILES) {
1449                 atomic_inc(&oldf->count);
1450                 goto out;
1451         }
1452
1453         newf = dup_fd(oldf, &error);
1454         if (!newf)
1455                 goto out;
1456
1457         tsk->files = newf;
1458         error = 0;
1459 out:
1460         return error;
1461 }
1462
1463 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1464 {
1465 #ifdef CONFIG_BLOCK
1466         struct io_context *ioc = current->io_context;
1467         struct io_context *new_ioc;
1468
1469         if (!ioc)
1470                 return 0;
1471         /*
1472          * Share io context with parent, if CLONE_IO is set
1473          */
1474         if (clone_flags & CLONE_IO) {
1475                 ioc_task_link(ioc);
1476                 tsk->io_context = ioc;
1477         } else if (ioprio_valid(ioc->ioprio)) {
1478                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1479                 if (unlikely(!new_ioc))
1480                         return -ENOMEM;
1481
1482                 new_ioc->ioprio = ioc->ioprio;
1483                 put_io_context(new_ioc);
1484         }
1485 #endif
1486         return 0;
1487 }
1488
1489 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1490 {
1491         struct sighand_struct *sig;
1492
1493         if (clone_flags & CLONE_SIGHAND) {
1494                 refcount_inc(&current->sighand->count);
1495                 return 0;
1496         }
1497         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1498         rcu_assign_pointer(tsk->sighand, sig);
1499         if (!sig)
1500                 return -ENOMEM;
1501
1502         refcount_set(&sig->count, 1);
1503         spin_lock_irq(&current->sighand->siglock);
1504         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1505         spin_unlock_irq(&current->sighand->siglock);
1506         return 0;
1507 }
1508
1509 void __cleanup_sighand(struct sighand_struct *sighand)
1510 {
1511         if (refcount_dec_and_test(&sighand->count)) {
1512                 signalfd_cleanup(sighand);
1513                 /*
1514                  * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1515                  * without an RCU grace period, see __lock_task_sighand().
1516                  */
1517                 kmem_cache_free(sighand_cachep, sighand);
1518         }
1519 }
1520
1521 #ifdef CONFIG_POSIX_TIMERS
1522 /*
1523  * Initialize POSIX timer handling for a thread group.
1524  */
1525 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1526 {
1527         unsigned long cpu_limit;
1528
1529         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1530         if (cpu_limit != RLIM_INFINITY) {
1531                 sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1532                 sig->cputimer.running = true;
1533         }
1534
1535         /* The timer lists. */
1536         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1537         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1538         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1539 }
1540 #else
1541 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1542 #endif
1543
1544 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1545 {
1546         struct signal_struct *sig;
1547
1548         if (clone_flags & CLONE_THREAD)
1549                 return 0;
1550
1551         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1552         tsk->signal = sig;
1553         if (!sig)
1554                 return -ENOMEM;
1555
1556         sig->nr_threads = 1;
1557         atomic_set(&sig->live, 1);
1558         refcount_set(&sig->sigcnt, 1);
1559
1560         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1561         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1562         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1563
1564         init_waitqueue_head(&sig->wait_chldexit);
1565         sig->curr_target = tsk;
1566         init_sigpending(&sig->shared_pending);
1567         INIT_HLIST_HEAD(&sig->multiprocess);
1568         seqlock_init(&sig->stats_lock);
1569         prev_cputime_init(&sig->prev_cputime);
1570
1571 #ifdef CONFIG_POSIX_TIMERS
1572         INIT_LIST_HEAD(&sig->posix_timers);
1573         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1574         sig->real_timer.function = it_real_fn;
1575 #endif
1576
1577         task_lock(current->group_leader);
1578         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1579         task_unlock(current->group_leader);
1580
1581         posix_cpu_timers_init_group(sig);
1582
1583         tty_audit_fork(sig);
1584         sched_autogroup_fork(sig);
1585
1586         sig->oom_score_adj = current->signal->oom_score_adj;
1587         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1588
1589         mutex_init(&sig->cred_guard_mutex);
1590
1591         return 0;
1592 }
1593
1594 static void copy_seccomp(struct task_struct *p)
1595 {
1596 #ifdef CONFIG_SECCOMP
1597         /*
1598          * Must be called with sighand->lock held, which is common to
1599          * all threads in the group. Holding cred_guard_mutex is not
1600          * needed because this new task is not yet running and cannot
1601          * be racing exec.
1602          */
1603         assert_spin_locked(&current->sighand->siglock);
1604
1605         /* Ref-count the new filter user, and assign it. */
1606         get_seccomp_filter(current);
1607         p->seccomp = current->seccomp;
1608
1609         /*
1610          * Explicitly enable no_new_privs here in case it got set
1611          * between the task_struct being duplicated and holding the
1612          * sighand lock. The seccomp state and nnp must be in sync.
1613          */
1614         if (task_no_new_privs(current))
1615                 task_set_no_new_privs(p);
1616
1617         /*
1618          * If the parent gained a seccomp mode after copying thread
1619          * flags and between before we held the sighand lock, we have
1620          * to manually enable the seccomp thread flag here.
1621          */
1622         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1623                 set_tsk_thread_flag(p, TIF_SECCOMP);
1624 #endif
1625 }
1626
1627 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1628 {
1629         current->clear_child_tid = tidptr;
1630
1631         return task_pid_vnr(current);
1632 }
1633
1634 static void rt_mutex_init_task(struct task_struct *p)
1635 {
1636         raw_spin_lock_init(&p->pi_lock);
1637 #ifdef CONFIG_RT_MUTEXES
1638         p->pi_waiters = RB_ROOT_CACHED;
1639         p->pi_top_task = NULL;
1640         p->pi_blocked_on = NULL;
1641 #endif
1642 }
1643
1644 #ifdef CONFIG_POSIX_TIMERS
1645 /*
1646  * Initialize POSIX timer handling for a single task.
1647  */
1648 static void posix_cpu_timers_init(struct task_struct *tsk)
1649 {
1650         tsk->cputime_expires.prof_exp = 0;
1651         tsk->cputime_expires.virt_exp = 0;
1652         tsk->cputime_expires.sched_exp = 0;
1653         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1654         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1655         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1656 }
1657 #else
1658 static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1659 #endif
1660
1661 static inline void init_task_pid_links(struct task_struct *task)
1662 {
1663         enum pid_type type;
1664
1665         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1666                 INIT_HLIST_NODE(&task->pid_links[type]);
1667         }
1668 }
1669
1670 static inline void
1671 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1672 {
1673         if (type == PIDTYPE_PID)
1674                 task->thread_pid = pid;
1675         else
1676                 task->signal->pids[type] = pid;
1677 }
1678
1679 static inline void rcu_copy_process(struct task_struct *p)
1680 {
1681 #ifdef CONFIG_PREEMPT_RCU
1682         p->rcu_read_lock_nesting = 0;
1683         p->rcu_read_unlock_special.s = 0;
1684         p->rcu_blocked_node = NULL;
1685         INIT_LIST_HEAD(&p->rcu_node_entry);
1686 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1687 #ifdef CONFIG_TASKS_RCU
1688         p->rcu_tasks_holdout = false;
1689         INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1690         p->rcu_tasks_idle_cpu = -1;
1691 #endif /* #ifdef CONFIG_TASKS_RCU */
1692 }
1693
1694 static int pidfd_release(struct inode *inode, struct file *file)
1695 {
1696         struct pid *pid = file->private_data;
1697
1698         file->private_data = NULL;
1699         put_pid(pid);
1700         return 0;
1701 }
1702
1703 #ifdef CONFIG_PROC_FS
1704 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1705 {
1706         struct pid_namespace *ns = proc_pid_ns(file_inode(m->file));
1707         struct pid *pid = f->private_data;
1708
1709         seq_put_decimal_ull(m, "Pid:\t", pid_nr_ns(pid, ns));
1710         seq_putc(m, '\n');
1711 }
1712 #endif
1713
1714 const struct file_operations pidfd_fops = {
1715         .release = pidfd_release,
1716 #ifdef CONFIG_PROC_FS
1717         .show_fdinfo = pidfd_show_fdinfo,
1718 #endif
1719 };
1720
1721 static void __delayed_free_task(struct rcu_head *rhp)
1722 {
1723         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1724
1725         free_task(tsk);
1726 }
1727
1728 static __always_inline void delayed_free_task(struct task_struct *tsk)
1729 {
1730         if (IS_ENABLED(CONFIG_MEMCG))
1731                 call_rcu(&tsk->rcu, __delayed_free_task);
1732         else
1733                 free_task(tsk);
1734 }
1735
1736 /*
1737  * This creates a new process as a copy of the old one,
1738  * but does not actually start it yet.
1739  *
1740  * It copies the registers, and all the appropriate
1741  * parts of the process environment (as per the clone
1742  * flags). The actual kick-off is left to the caller.
1743  */
1744 static __latent_entropy struct task_struct *copy_process(
1745                                         unsigned long clone_flags,
1746                                         unsigned long stack_start,
1747                                         unsigned long stack_size,
1748                                         int __user *parent_tidptr,
1749                                         int __user *child_tidptr,
1750                                         struct pid *pid,
1751                                         int trace,
1752                                         unsigned long tls,
1753                                         int node)
1754 {
1755         int pidfd = -1, retval;
1756         struct task_struct *p;
1757         struct multiprocess_signals delayed;
1758         struct file *pidfile = NULL;
1759
1760         /*
1761          * Don't allow sharing the root directory with processes in a different
1762          * namespace
1763          */
1764         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1765                 return ERR_PTR(-EINVAL);
1766
1767         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1768                 return ERR_PTR(-EINVAL);
1769
1770         /*
1771          * Thread groups must share signals as well, and detached threads
1772          * can only be started up within the thread group.
1773          */
1774         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1775                 return ERR_PTR(-EINVAL);
1776
1777         /*
1778          * Shared signal handlers imply shared VM. By way of the above,
1779          * thread groups also imply shared VM. Blocking this case allows
1780          * for various simplifications in other code.
1781          */
1782         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1783                 return ERR_PTR(-EINVAL);
1784
1785         /*
1786          * Siblings of global init remain as zombies on exit since they are
1787          * not reaped by their parent (swapper). To solve this and to avoid
1788          * multi-rooted process trees, prevent global and container-inits
1789          * from creating siblings.
1790          */
1791         if ((clone_flags & CLONE_PARENT) &&
1792                                 current->signal->flags & SIGNAL_UNKILLABLE)
1793                 return ERR_PTR(-EINVAL);
1794
1795         /*
1796          * If the new process will be in a different pid or user namespace
1797          * do not allow it to share a thread group with the forking task.
1798          */
1799         if (clone_flags & CLONE_THREAD) {
1800                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1801                     (task_active_pid_ns(current) !=
1802                                 current->nsproxy->pid_ns_for_children))
1803                         return ERR_PTR(-EINVAL);
1804         }
1805
1806         if (clone_flags & CLONE_PIDFD) {
1807                 /*
1808                  * - CLONE_PARENT_SETTID is useless for pidfds and also
1809                  *   parent_tidptr is used to return pidfds.
1810                  * - CLONE_DETACHED is blocked so that we can potentially
1811                  *   reuse it later for CLONE_PIDFD.
1812                  * - CLONE_THREAD is blocked until someone really needs it.
1813                  */
1814                 if (clone_flags &
1815                     (CLONE_DETACHED | CLONE_PARENT_SETTID | CLONE_THREAD))
1816                         return ERR_PTR(-EINVAL);
1817         }
1818
1819         /*
1820          * Force any signals received before this point to be delivered
1821          * before the fork happens.  Collect up signals sent to multiple
1822          * processes that happen during the fork and delay them so that
1823          * they appear to happen after the fork.
1824          */
1825         sigemptyset(&delayed.signal);
1826         INIT_HLIST_NODE(&delayed.node);
1827
1828         spin_lock_irq(&current->sighand->siglock);
1829         if (!(clone_flags & CLONE_THREAD))
1830                 hlist_add_head(&delayed.node, &current->signal->multiprocess);
1831         recalc_sigpending();
1832         spin_unlock_irq(&current->sighand->siglock);
1833         retval = -ERESTARTNOINTR;
1834         if (signal_pending(current))
1835                 goto fork_out;
1836
1837         retval = -ENOMEM;
1838         p = dup_task_struct(current, node);
1839         if (!p)
1840                 goto fork_out;
1841
1842         /*
1843          * This _must_ happen before we call free_task(), i.e. before we jump
1844          * to any of the bad_fork_* labels. This is to avoid freeing
1845          * p->set_child_tid which is (ab)used as a kthread's data pointer for
1846          * kernel threads (PF_KTHREAD).
1847          */
1848         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1849         /*
1850          * Clear TID on mm_release()?
1851          */
1852         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1853
1854         ftrace_graph_init_task(p);
1855
1856         rt_mutex_init_task(p);
1857
1858 #ifdef CONFIG_PROVE_LOCKING
1859         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1860         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1861 #endif
1862         retval = -EAGAIN;
1863         if (atomic_read(&p->real_cred->user->processes) >=
1864                         task_rlimit(p, RLIMIT_NPROC)) {
1865                 if (p->real_cred->user != INIT_USER &&
1866                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1867                         goto bad_fork_free;
1868         }
1869         current->flags &= ~PF_NPROC_EXCEEDED;
1870
1871         retval = copy_creds(p, clone_flags);
1872         if (retval < 0)
1873                 goto bad_fork_free;
1874
1875         /*
1876          * If multiple threads are within copy_process(), then this check
1877          * triggers too late. This doesn't hurt, the check is only there
1878          * to stop root fork bombs.
1879          */
1880         retval = -EAGAIN;
1881         if (nr_threads >= max_threads)
1882                 goto bad_fork_cleanup_count;
1883
1884         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1885         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1886         p->flags |= PF_FORKNOEXEC;
1887         INIT_LIST_HEAD(&p->children);
1888         INIT_LIST_HEAD(&p->sibling);
1889         rcu_copy_process(p);
1890         p->vfork_done = NULL;
1891         spin_lock_init(&p->alloc_lock);
1892
1893         init_sigpending(&p->pending);
1894
1895         p->utime = p->stime = p->gtime = 0;
1896 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1897         p->utimescaled = p->stimescaled = 0;
1898 #endif
1899         prev_cputime_init(&p->prev_cputime);
1900
1901 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1902         seqcount_init(&p->vtime.seqcount);
1903         p->vtime.starttime = 0;
1904         p->vtime.state = VTIME_INACTIVE;
1905 #endif
1906
1907 #if defined(SPLIT_RSS_COUNTING)
1908         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1909 #endif
1910
1911         p->default_timer_slack_ns = current->timer_slack_ns;
1912
1913 #ifdef CONFIG_PSI
1914         p->psi_flags = 0;
1915 #endif
1916
1917         task_io_accounting_init(&p->ioac);
1918         acct_clear_integrals(p);
1919
1920         posix_cpu_timers_init(p);
1921
1922         p->io_context = NULL;
1923         audit_set_context(p, NULL);
1924         cgroup_fork(p);
1925 #ifdef CONFIG_NUMA
1926         p->mempolicy = mpol_dup(p->mempolicy);
1927         if (IS_ERR(p->mempolicy)) {
1928                 retval = PTR_ERR(p->mempolicy);
1929                 p->mempolicy = NULL;
1930                 goto bad_fork_cleanup_threadgroup_lock;
1931         }
1932 #endif
1933 #ifdef CONFIG_CPUSETS
1934         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1935         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1936         seqcount_init(&p->mems_allowed_seq);
1937 #endif
1938 #ifdef CONFIG_TRACE_IRQFLAGS
1939         p->irq_events = 0;
1940         p->hardirqs_enabled = 0;
1941         p->hardirq_enable_ip = 0;
1942         p->hardirq_enable_event = 0;
1943         p->hardirq_disable_ip = _THIS_IP_;
1944         p->hardirq_disable_event = 0;
1945         p->softirqs_enabled = 1;
1946         p->softirq_enable_ip = _THIS_IP_;
1947         p->softirq_enable_event = 0;
1948         p->softirq_disable_ip = 0;
1949         p->softirq_disable_event = 0;
1950         p->hardirq_context = 0;
1951         p->softirq_context = 0;
1952 #endif
1953
1954         p->pagefault_disabled = 0;
1955
1956 #ifdef CONFIG_LOCKDEP
1957         lockdep_init_task(p);
1958 #endif
1959
1960 #ifdef CONFIG_DEBUG_MUTEXES
1961         p->blocked_on = NULL; /* not blocked yet */
1962 #endif
1963 #ifdef CONFIG_BCACHE
1964         p->sequential_io        = 0;
1965         p->sequential_io_avg    = 0;
1966 #endif
1967
1968         /* Perform scheduler related setup. Assign this task to a CPU. */
1969         retval = sched_fork(clone_flags, p);
1970         if (retval)
1971                 goto bad_fork_cleanup_policy;
1972
1973         retval = perf_event_init_task(p);
1974         if (retval)
1975                 goto bad_fork_cleanup_policy;
1976         retval = audit_alloc(p);
1977         if (retval)
1978                 goto bad_fork_cleanup_perf;
1979         /* copy all the process information */
1980         shm_init_task(p);
1981         retval = security_task_alloc(p, clone_flags);
1982         if (retval)
1983                 goto bad_fork_cleanup_audit;
1984         retval = copy_semundo(clone_flags, p);
1985         if (retval)
1986                 goto bad_fork_cleanup_security;
1987         retval = copy_files(clone_flags, p);
1988         if (retval)
1989                 goto bad_fork_cleanup_semundo;
1990         retval = copy_fs(clone_flags, p);
1991         if (retval)
1992                 goto bad_fork_cleanup_files;
1993         retval = copy_sighand(clone_flags, p);
1994         if (retval)
1995                 goto bad_fork_cleanup_fs;
1996         retval = copy_signal(clone_flags, p);
1997         if (retval)
1998                 goto bad_fork_cleanup_sighand;
1999         retval = copy_mm(clone_flags, p);
2000         if (retval)
2001                 goto bad_fork_cleanup_signal;
2002         retval = copy_namespaces(clone_flags, p);
2003         if (retval)
2004                 goto bad_fork_cleanup_mm;
2005         retval = copy_io(clone_flags, p);
2006         if (retval)
2007                 goto bad_fork_cleanup_namespaces;
2008         retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
2009         if (retval)
2010                 goto bad_fork_cleanup_io;
2011
2012         stackleak_task_init(p);
2013
2014         if (pid != &init_struct_pid) {
2015                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
2016                 if (IS_ERR(pid)) {
2017                         retval = PTR_ERR(pid);
2018                         goto bad_fork_cleanup_thread;
2019                 }
2020         }
2021
2022         /*
2023          * This has to happen after we've potentially unshared the file
2024          * descriptor table (so that the pidfd doesn't leak into the child
2025          * if the fd table isn't shared).
2026          */
2027         if (clone_flags & CLONE_PIDFD) {
2028                 retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2029                 if (retval < 0)
2030                         goto bad_fork_free_pid;
2031
2032                 pidfd = retval;
2033
2034                 pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2035                                               O_RDWR | O_CLOEXEC);
2036                 if (IS_ERR(pidfile)) {
2037                         put_unused_fd(pidfd);
2038                         retval = PTR_ERR(pidfile);
2039                         goto bad_fork_free_pid;
2040                 }
2041                 get_pid(pid);   /* held by pidfile now */
2042
2043                 retval = put_user(pidfd, parent_tidptr);
2044                 if (retval)
2045                         goto bad_fork_put_pidfd;
2046         }
2047
2048 #ifdef CONFIG_BLOCK
2049         p->plug = NULL;
2050 #endif
2051 #ifdef CONFIG_FUTEX
2052         p->robust_list = NULL;
2053 #ifdef CONFIG_COMPAT
2054         p->compat_robust_list = NULL;
2055 #endif
2056         INIT_LIST_HEAD(&p->pi_state_list);
2057         p->pi_state_cache = NULL;
2058 #endif
2059         /*
2060          * sigaltstack should be cleared when sharing the same VM
2061          */
2062         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2063                 sas_ss_reset(p);
2064
2065         /*
2066          * Syscall tracing and stepping should be turned off in the
2067          * child regardless of CLONE_PTRACE.
2068          */
2069         user_disable_single_step(p);
2070         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
2071 #ifdef TIF_SYSCALL_EMU
2072         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
2073 #endif
2074         clear_tsk_latency_tracing(p);
2075
2076         /* ok, now we should be set up.. */
2077         p->pid = pid_nr(pid);
2078         if (clone_flags & CLONE_THREAD) {
2079                 p->exit_signal = -1;
2080                 p->group_leader = current->group_leader;
2081                 p->tgid = current->tgid;
2082         } else {
2083                 if (clone_flags & CLONE_PARENT)
2084                         p->exit_signal = current->group_leader->exit_signal;
2085                 else
2086                         p->exit_signal = (clone_flags & CSIGNAL);
2087                 p->group_leader = p;
2088                 p->tgid = p->pid;
2089         }
2090
2091         p->nr_dirtied = 0;
2092         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2093         p->dirty_paused_when = 0;
2094
2095         p->pdeath_signal = 0;
2096         INIT_LIST_HEAD(&p->thread_group);
2097         p->task_works = NULL;
2098
2099         cgroup_threadgroup_change_begin(current);
2100         /*
2101          * Ensure that the cgroup subsystem policies allow the new process to be
2102          * forked. It should be noted the the new process's css_set can be changed
2103          * between here and cgroup_post_fork() if an organisation operation is in
2104          * progress.
2105          */
2106         retval = cgroup_can_fork(p);
2107         if (retval)
2108                 goto bad_fork_cgroup_threadgroup_change_end;
2109
2110         /*
2111          * From this point on we must avoid any synchronous user-space
2112          * communication until we take the tasklist-lock. In particular, we do
2113          * not want user-space to be able to predict the process start-time by
2114          * stalling fork(2) after we recorded the start_time but before it is
2115          * visible to the system.
2116          */
2117
2118         p->start_time = ktime_get_ns();
2119         p->real_start_time = ktime_get_boottime_ns();
2120
2121         /*
2122          * Make it visible to the rest of the system, but dont wake it up yet.
2123          * Need tasklist lock for parent etc handling!
2124          */
2125         write_lock_irq(&tasklist_lock);
2126
2127         /* CLONE_PARENT re-uses the old parent */
2128         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2129                 p->real_parent = current->real_parent;
2130                 p->parent_exec_id = current->parent_exec_id;
2131         } else {
2132                 p->real_parent = current;
2133                 p->parent_exec_id = current->self_exec_id;
2134         }
2135
2136         klp_copy_process(p);
2137
2138         spin_lock(&current->sighand->siglock);
2139
2140         /*
2141          * Copy seccomp details explicitly here, in case they were changed
2142          * before holding sighand lock.
2143          */
2144         copy_seccomp(p);
2145
2146         rseq_fork(p, clone_flags);
2147
2148         /* Don't start children in a dying pid namespace */
2149         if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2150                 retval = -ENOMEM;
2151                 goto bad_fork_cancel_cgroup;
2152         }
2153
2154         /* Let kill terminate clone/fork in the middle */
2155         if (fatal_signal_pending(current)) {
2156                 retval = -EINTR;
2157                 goto bad_fork_cancel_cgroup;
2158         }
2159
2160         /* past the last point of failure */
2161         if (pidfile)
2162                 fd_install(pidfd, pidfile);
2163
2164         init_task_pid_links(p);
2165         if (likely(p->pid)) {
2166                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2167
2168                 init_task_pid(p, PIDTYPE_PID, pid);
2169                 if (thread_group_leader(p)) {
2170                         init_task_pid(p, PIDTYPE_TGID, pid);
2171                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2172                         init_task_pid(p, PIDTYPE_SID, task_session(current));
2173
2174                         if (is_child_reaper(pid)) {
2175                                 ns_of_pid(pid)->child_reaper = p;
2176                                 p->signal->flags |= SIGNAL_UNKILLABLE;
2177                         }
2178                         p->signal->shared_pending.signal = delayed.signal;
2179                         p->signal->tty = tty_kref_get(current->signal->tty);
2180                         /*
2181                          * Inherit has_child_subreaper flag under the same
2182                          * tasklist_lock with adding child to the process tree
2183                          * for propagate_has_child_subreaper optimization.
2184                          */
2185                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2186                                                          p->real_parent->signal->is_child_subreaper;
2187                         list_add_tail(&p->sibling, &p->real_parent->children);
2188                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
2189                         attach_pid(p, PIDTYPE_TGID);
2190                         attach_pid(p, PIDTYPE_PGID);
2191                         attach_pid(p, PIDTYPE_SID);
2192                         __this_cpu_inc(process_counts);
2193                 } else {
2194                         current->signal->nr_threads++;
2195                         atomic_inc(&current->signal->live);
2196                         refcount_inc(&current->signal->sigcnt);
2197                         task_join_group_stop(p);
2198                         list_add_tail_rcu(&p->thread_group,
2199                                           &p->group_leader->thread_group);
2200                         list_add_tail_rcu(&p->thread_node,
2201                                           &p->signal->thread_head);
2202                 }
2203                 attach_pid(p, PIDTYPE_PID);
2204                 nr_threads++;
2205         }
2206         total_forks++;
2207         hlist_del_init(&delayed.node);
2208         spin_unlock(&current->sighand->siglock);
2209         syscall_tracepoint_update(p);
2210         write_unlock_irq(&tasklist_lock);
2211
2212         proc_fork_connector(p);
2213         cgroup_post_fork(p);
2214         cgroup_threadgroup_change_end(current);
2215         perf_event_fork(p);
2216
2217         trace_task_newtask(p, clone_flags);
2218         uprobe_copy_process(p, clone_flags);
2219
2220         return p;
2221
2222 bad_fork_cancel_cgroup:
2223         spin_unlock(&current->sighand->siglock);
2224         write_unlock_irq(&tasklist_lock);
2225         cgroup_cancel_fork(p);
2226 bad_fork_cgroup_threadgroup_change_end:
2227         cgroup_threadgroup_change_end(current);
2228 bad_fork_put_pidfd:
2229         if (clone_flags & CLONE_PIDFD) {
2230                 fput(pidfile);
2231                 put_unused_fd(pidfd);
2232         }
2233 bad_fork_free_pid:
2234         if (pid != &init_struct_pid)
2235                 free_pid(pid);
2236 bad_fork_cleanup_thread:
2237         exit_thread(p);
2238 bad_fork_cleanup_io:
2239         if (p->io_context)
2240                 exit_io_context(p);
2241 bad_fork_cleanup_namespaces:
2242         exit_task_namespaces(p);
2243 bad_fork_cleanup_mm:
2244         if (p->mm) {
2245                 mm_clear_owner(p->mm, p);
2246                 mmput(p->mm);
2247         }
2248 bad_fork_cleanup_signal:
2249         if (!(clone_flags & CLONE_THREAD))
2250                 free_signal_struct(p->signal);
2251 bad_fork_cleanup_sighand:
2252         __cleanup_sighand(p->sighand);
2253 bad_fork_cleanup_fs:
2254         exit_fs(p); /* blocking */
2255 bad_fork_cleanup_files:
2256         exit_files(p); /* blocking */
2257 bad_fork_cleanup_semundo:
2258         exit_sem(p);
2259 bad_fork_cleanup_security:
2260         security_task_free(p);
2261 bad_fork_cleanup_audit:
2262         audit_free(p);
2263 bad_fork_cleanup_perf:
2264         perf_event_free_task(p);
2265 bad_fork_cleanup_policy:
2266         lockdep_free_task(p);
2267 #ifdef CONFIG_NUMA
2268         mpol_put(p->mempolicy);
2269 bad_fork_cleanup_threadgroup_lock:
2270 #endif
2271         delayacct_tsk_free(p);
2272 bad_fork_cleanup_count:
2273         atomic_dec(&p->cred->user->processes);
2274         exit_creds(p);
2275 bad_fork_free:
2276         p->state = TASK_DEAD;
2277         put_task_stack(p);
2278         delayed_free_task(p);
2279 fork_out:
2280         spin_lock_irq(&current->sighand->siglock);
2281         hlist_del_init(&delayed.node);
2282         spin_unlock_irq(&current->sighand->siglock);
2283         return ERR_PTR(retval);
2284 }
2285
2286 static inline void init_idle_pids(struct task_struct *idle)
2287 {
2288         enum pid_type type;
2289
2290         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2291                 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2292                 init_task_pid(idle, type, &init_struct_pid);
2293         }
2294 }
2295
2296 struct task_struct *fork_idle(int cpu)
2297 {
2298         struct task_struct *task;
2299         task = copy_process(CLONE_VM, 0, 0, NULL, NULL, &init_struct_pid, 0, 0,
2300                             cpu_to_node(cpu));
2301         if (!IS_ERR(task)) {
2302                 init_idle_pids(task);
2303                 init_idle(task, cpu);
2304         }
2305
2306         return task;
2307 }
2308
2309 struct mm_struct *copy_init_mm(void)
2310 {
2311         return dup_mm(NULL, &init_mm);
2312 }
2313
2314 /*
2315  *  Ok, this is the main fork-routine.
2316  *
2317  * It copies the process, and if successful kick-starts
2318  * it and waits for it to finish using the VM if required.
2319  */
2320 long _do_fork(unsigned long clone_flags,
2321               unsigned long stack_start,
2322               unsigned long stack_size,
2323               int __user *parent_tidptr,
2324               int __user *child_tidptr,
2325               unsigned long tls)
2326 {
2327         struct completion vfork;
2328         struct pid *pid;
2329         struct task_struct *p;
2330         int trace = 0;
2331         long nr;
2332
2333         /*
2334          * Determine whether and which event to report to ptracer.  When
2335          * called from kernel_thread or CLONE_UNTRACED is explicitly
2336          * requested, no event is reported; otherwise, report if the event
2337          * for the type of forking is enabled.
2338          */
2339         if (!(clone_flags & CLONE_UNTRACED)) {
2340                 if (clone_flags & CLONE_VFORK)
2341                         trace = PTRACE_EVENT_VFORK;
2342                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
2343                         trace = PTRACE_EVENT_CLONE;
2344                 else
2345                         trace = PTRACE_EVENT_FORK;
2346
2347                 if (likely(!ptrace_event_enabled(current, trace)))
2348                         trace = 0;
2349         }
2350
2351         p = copy_process(clone_flags, stack_start, stack_size, parent_tidptr,
2352                          child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
2353         add_latent_entropy();
2354
2355         if (IS_ERR(p))
2356                 return PTR_ERR(p);
2357
2358         /*
2359          * Do this prior waking up the new thread - the thread pointer
2360          * might get invalid after that point, if the thread exits quickly.
2361          */
2362         trace_sched_process_fork(current, p);
2363
2364         pid = get_task_pid(p, PIDTYPE_PID);
2365         nr = pid_vnr(pid);
2366
2367         if (clone_flags & CLONE_PARENT_SETTID)
2368                 put_user(nr, parent_tidptr);
2369
2370         if (clone_flags & CLONE_VFORK) {
2371                 p->vfork_done = &vfork;
2372                 init_completion(&vfork);
2373                 get_task_struct(p);
2374         }
2375
2376         wake_up_new_task(p);
2377
2378         /* forking complete and child started to run, tell ptracer */
2379         if (unlikely(trace))
2380                 ptrace_event_pid(trace, pid);
2381
2382         if (clone_flags & CLONE_VFORK) {
2383                 if (!wait_for_vfork_done(p, &vfork))
2384                         ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2385         }
2386
2387         put_pid(pid);
2388         return nr;
2389 }
2390
2391 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2392 /* For compatibility with architectures that call do_fork directly rather than
2393  * using the syscall entry points below. */
2394 long do_fork(unsigned long clone_flags,
2395               unsigned long stack_start,
2396               unsigned long stack_size,
2397               int __user *parent_tidptr,
2398               int __user *child_tidptr)
2399 {
2400         return _do_fork(clone_flags, stack_start, stack_size,
2401                         parent_tidptr, child_tidptr, 0);
2402 }
2403 #endif
2404
2405 /*
2406  * Create a kernel thread.
2407  */
2408 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2409 {
2410         return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2411                 (unsigned long)arg, NULL, NULL, 0);
2412 }
2413
2414 #ifdef __ARCH_WANT_SYS_FORK
2415 SYSCALL_DEFINE0(fork)
2416 {
2417 #ifdef CONFIG_MMU
2418         return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2419 #else
2420         /* can not support in nommu mode */
2421         return -EINVAL;
2422 #endif
2423 }
2424 #endif
2425
2426 #ifdef __ARCH_WANT_SYS_VFORK
2427 SYSCALL_DEFINE0(vfork)
2428 {
2429         return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2430                         0, NULL, NULL, 0);
2431 }
2432 #endif
2433
2434 #ifdef __ARCH_WANT_SYS_CLONE
2435 #ifdef CONFIG_CLONE_BACKWARDS
2436 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2437                  int __user *, parent_tidptr,
2438                  unsigned long, tls,
2439                  int __user *, child_tidptr)
2440 #elif defined(CONFIG_CLONE_BACKWARDS2)
2441 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2442                  int __user *, parent_tidptr,
2443                  int __user *, child_tidptr,
2444                  unsigned long, tls)
2445 #elif defined(CONFIG_CLONE_BACKWARDS3)
2446 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2447                 int, stack_size,
2448                 int __user *, parent_tidptr,
2449                 int __user *, child_tidptr,
2450                 unsigned long, tls)
2451 #else
2452 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2453                  int __user *, parent_tidptr,
2454                  int __user *, child_tidptr,
2455                  unsigned long, tls)
2456 #endif
2457 {
2458         return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2459 }
2460 #endif
2461
2462 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2463 {
2464         struct task_struct *leader, *parent, *child;
2465         int res;
2466
2467         read_lock(&tasklist_lock);
2468         leader = top = top->group_leader;
2469 down:
2470         for_each_thread(leader, parent) {
2471                 list_for_each_entry(child, &parent->children, sibling) {
2472                         res = visitor(child, data);
2473                         if (res) {
2474                                 if (res < 0)
2475                                         goto out;
2476                                 leader = child;
2477                                 goto down;
2478                         }
2479 up:
2480                         ;
2481                 }
2482         }
2483
2484         if (leader != top) {
2485                 child = leader;
2486                 parent = child->real_parent;
2487                 leader = parent->group_leader;
2488                 goto up;
2489         }
2490 out:
2491         read_unlock(&tasklist_lock);
2492 }
2493
2494 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2495 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2496 #endif
2497
2498 static void sighand_ctor(void *data)
2499 {
2500         struct sighand_struct *sighand = data;
2501
2502         spin_lock_init(&sighand->siglock);
2503         init_waitqueue_head(&sighand->signalfd_wqh);
2504 }
2505
2506 void __init proc_caches_init(void)
2507 {
2508         unsigned int mm_size;
2509
2510         sighand_cachep = kmem_cache_create("sighand_cache",
2511                         sizeof(struct sighand_struct), 0,
2512                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2513                         SLAB_ACCOUNT, sighand_ctor);
2514         signal_cachep = kmem_cache_create("signal_cache",
2515                         sizeof(struct signal_struct), 0,
2516                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2517                         NULL);
2518         files_cachep = kmem_cache_create("files_cache",
2519                         sizeof(struct files_struct), 0,
2520                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2521                         NULL);
2522         fs_cachep = kmem_cache_create("fs_cache",
2523                         sizeof(struct fs_struct), 0,
2524                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2525                         NULL);
2526
2527         /*
2528          * The mm_cpumask is located at the end of mm_struct, and is
2529          * dynamically sized based on the maximum CPU number this system
2530          * can have, taking hotplug into account (nr_cpu_ids).
2531          */
2532         mm_size = sizeof(struct mm_struct) + cpumask_size();
2533
2534         mm_cachep = kmem_cache_create_usercopy("mm_struct",
2535                         mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2536                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2537                         offsetof(struct mm_struct, saved_auxv),
2538                         sizeof_field(struct mm_struct, saved_auxv),
2539                         NULL);
2540         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2541         mmap_init();
2542         nsproxy_cache_init();
2543 }
2544
2545 /*
2546  * Check constraints on flags passed to the unshare system call.
2547  */
2548 static int check_unshare_flags(unsigned long unshare_flags)
2549 {
2550         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2551                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2552                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2553                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2554                 return -EINVAL;
2555         /*
2556          * Not implemented, but pretend it works if there is nothing
2557          * to unshare.  Note that unsharing the address space or the
2558          * signal handlers also need to unshare the signal queues (aka
2559          * CLONE_THREAD).
2560          */
2561         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2562                 if (!thread_group_empty(current))
2563                         return -EINVAL;
2564         }
2565         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2566                 if (refcount_read(&current->sighand->count) > 1)
2567                         return -EINVAL;
2568         }
2569         if (unshare_flags & CLONE_VM) {
2570                 if (!current_is_single_threaded())
2571                         return -EINVAL;
2572         }
2573
2574         return 0;
2575 }
2576
2577 /*
2578  * Unshare the filesystem structure if it is being shared
2579  */
2580 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2581 {
2582         struct fs_struct *fs = current->fs;
2583
2584         if (!(unshare_flags & CLONE_FS) || !fs)
2585                 return 0;
2586
2587         /* don't need lock here; in the worst case we'll do useless copy */
2588         if (fs->users == 1)
2589                 return 0;
2590
2591         *new_fsp = copy_fs_struct(fs);
2592         if (!*new_fsp)
2593                 return -ENOMEM;
2594
2595         return 0;
2596 }
2597
2598 /*
2599  * Unshare file descriptor table if it is being shared
2600  */
2601 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2602 {
2603         struct files_struct *fd = current->files;
2604         int error = 0;
2605
2606         if ((unshare_flags & CLONE_FILES) &&
2607             (fd && atomic_read(&fd->count) > 1)) {
2608                 *new_fdp = dup_fd(fd, &error);
2609                 if (!*new_fdp)
2610                         return error;
2611         }
2612
2613         return 0;
2614 }
2615
2616 /*
2617  * unshare allows a process to 'unshare' part of the process
2618  * context which was originally shared using clone.  copy_*
2619  * functions used by do_fork() cannot be used here directly
2620  * because they modify an inactive task_struct that is being
2621  * constructed. Here we are modifying the current, active,
2622  * task_struct.
2623  */
2624 int ksys_unshare(unsigned long unshare_flags)
2625 {
2626         struct fs_struct *fs, *new_fs = NULL;
2627         struct files_struct *fd, *new_fd = NULL;
2628         struct cred *new_cred = NULL;
2629         struct nsproxy *new_nsproxy = NULL;
2630         int do_sysvsem = 0;
2631         int err;
2632
2633         /*
2634          * If unsharing a user namespace must also unshare the thread group
2635          * and unshare the filesystem root and working directories.
2636          */
2637         if (unshare_flags & CLONE_NEWUSER)
2638                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2639         /*
2640          * If unsharing vm, must also unshare signal handlers.
2641          */
2642         if (unshare_flags & CLONE_VM)
2643                 unshare_flags |= CLONE_SIGHAND;
2644         /*
2645          * If unsharing a signal handlers, must also unshare the signal queues.
2646          */
2647         if (unshare_flags & CLONE_SIGHAND)
2648                 unshare_flags |= CLONE_THREAD;
2649         /*
2650          * If unsharing namespace, must also unshare filesystem information.
2651          */
2652         if (unshare_flags & CLONE_NEWNS)
2653                 unshare_flags |= CLONE_FS;
2654
2655         err = check_unshare_flags(unshare_flags);
2656         if (err)
2657                 goto bad_unshare_out;
2658         /*
2659          * CLONE_NEWIPC must also detach from the undolist: after switching
2660          * to a new ipc namespace, the semaphore arrays from the old
2661          * namespace are unreachable.
2662          */
2663         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2664                 do_sysvsem = 1;
2665         err = unshare_fs(unshare_flags, &new_fs);
2666         if (err)
2667                 goto bad_unshare_out;
2668         err = unshare_fd(unshare_flags, &new_fd);
2669         if (err)
2670                 goto bad_unshare_cleanup_fs;
2671         err = unshare_userns(unshare_flags, &new_cred);
2672         if (err)
2673                 goto bad_unshare_cleanup_fd;
2674         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2675                                          new_cred, new_fs);
2676         if (err)
2677                 goto bad_unshare_cleanup_cred;
2678
2679         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2680                 if (do_sysvsem) {
2681                         /*
2682                          * CLONE_SYSVSEM is equivalent to sys_exit().
2683                          */
2684                         exit_sem(current);
2685                 }
2686                 if (unshare_flags & CLONE_NEWIPC) {
2687                         /* Orphan segments in old ns (see sem above). */
2688                         exit_shm(current);
2689                         shm_init_task(current);
2690                 }
2691
2692                 if (new_nsproxy)
2693                         switch_task_namespaces(current, new_nsproxy);
2694
2695                 task_lock(current);
2696
2697                 if (new_fs) {
2698                         fs = current->fs;
2699                         spin_lock(&fs->lock);
2700                         current->fs = new_fs;
2701                         if (--fs->users)
2702                                 new_fs = NULL;
2703                         else
2704                                 new_fs = fs;
2705                         spin_unlock(&fs->lock);
2706                 }
2707
2708                 if (new_fd) {
2709                         fd = current->files;
2710                         current->files = new_fd;
2711                         new_fd = fd;
2712                 }
2713
2714                 task_unlock(current);
2715
2716                 if (new_cred) {
2717                         /* Install the new user namespace */
2718                         commit_creds(new_cred);
2719                         new_cred = NULL;
2720                 }
2721         }
2722
2723         perf_event_namespaces(current);
2724
2725 bad_unshare_cleanup_cred:
2726         if (new_cred)
2727                 put_cred(new_cred);
2728 bad_unshare_cleanup_fd:
2729         if (new_fd)
2730                 put_files_struct(new_fd);
2731
2732 bad_unshare_cleanup_fs:
2733         if (new_fs)
2734                 free_fs_struct(new_fs);
2735
2736 bad_unshare_out:
2737         return err;
2738 }
2739
2740 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2741 {
2742         return ksys_unshare(unshare_flags);
2743 }
2744
2745 /*
2746  *      Helper to unshare the files of the current task.
2747  *      We don't want to expose copy_files internals to
2748  *      the exec layer of the kernel.
2749  */
2750
2751 int unshare_files(struct files_struct **displaced)
2752 {
2753         struct task_struct *task = current;
2754         struct files_struct *copy = NULL;
2755         int error;
2756
2757         error = unshare_fd(CLONE_FILES, &copy);
2758         if (error || !copy) {
2759                 *displaced = NULL;
2760                 return error;
2761         }
2762         *displaced = task->files;
2763         task_lock(task);
2764         task->files = copy;
2765         task_unlock(task);
2766         return 0;
2767 }
2768
2769 int sysctl_max_threads(struct ctl_table *table, int write,
2770                        void __user *buffer, size_t *lenp, loff_t *ppos)
2771 {
2772         struct ctl_table t;
2773         int ret;
2774         int threads = max_threads;
2775         int min = MIN_THREADS;
2776         int max = MAX_THREADS;
2777
2778         t = *table;
2779         t.data = &threads;
2780         t.extra1 = &min;
2781         t.extra2 = &max;
2782
2783         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2784         if (ret || !write)
2785                 return ret;
2786
2787         set_max_threads(threads);
2788
2789         return 0;
2790 }