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         mmu_notifier_mm_destroy(mm);
 681         check_mm(mm);
 682         put_user_ns(mm->user_ns);
 683         free_mm(mm);
 684 }
 685 EXPORT_SYMBOL_GPL(__mmdrop);
 686
 687 static void mmdrop_async_fn(struct work_struct *work)
 688 {
 689         struct mm_struct *mm;
 690
 691         mm = container_of(work, struct mm_struct, async_put_work);
 692         __mmdrop(mm);
 693 }
 694
 695 static void mmdrop_async(struct mm_struct *mm)
 696 {
 697         if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
 698                 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
 699                 schedule_work(&mm->async_put_work);
 700         }
 701 }
 702
 703 static inline void free_signal_struct(struct signal_struct *sig)
 704 {
 705         taskstats_tgid_free(sig);
 706         sched_autogroup_exit(sig);
 707         /*
 708          * __mmdrop is not safe to call from softirq context on x86 due to
 709          * pgd_dtor so postpone it to the async context
 710          */
 711         if (sig->oom_mm)
 712                 mmdrop_async(sig->oom_mm);
 713         kmem_cache_free(signal_cachep, sig);
 714 }
 715
 716 static inline void put_signal_struct(struct signal_struct *sig)
 717 {
 718         if (refcount_dec_and_test(&sig->sigcnt))
 719                 free_signal_struct(sig);
 720 }
 721
 722 void __put_task_struct(struct task_struct *tsk)
 723 {
 724         WARN_ON(!tsk->exit_state);
 725         WARN_ON(refcount_read(&tsk->usage));
 726         WARN_ON(tsk == current);
 727
 728         cgroup_free(tsk);
 729         task_numa_free(tsk);
 730         security_task_free(tsk);
 731         exit_creds(tsk);
 732         delayacct_tsk_free(tsk);
 733         put_signal_struct(tsk->signal);
 734
 735         if (!profile_handoff_task(tsk))
 736                 free_task(tsk);
 737 }
 738 EXPORT_SYMBOL_GPL(__put_task_struct);
 739
 740 void __init __weak arch_task_cache_init(void) { }
 741
 742 /*
 743  * set_max_threads
 744  */
 745 static void set_max_threads(unsigned int max_threads_suggested)
 746 {
 747         u64 threads;
 748         unsigned long nr_pages = totalram_pages();
 749
 750         /*
 751          * The number of threads shall be limited such that the thread
 752          * structures may only consume a small part of the available memory.
 753          */
 754         if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
 755                 threads = MAX_THREADS;
 756         else
 757                 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
 758                                     (u64) THREAD_SIZE * 8UL);
 759
 760         if (threads > max_threads_suggested)
 761                 threads = max_threads_suggested;
 762
 763         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 764 }
 765
 766 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 767 /* Initialized by the architecture: */
 768 int arch_task_struct_size __read_mostly;
 769 #endif
 770
 771 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
 772 {
 773         /* Fetch thread_struct whitelist for the architecture. */
 774         arch_thread_struct_whitelist(offset, size);
 775
 776         /*
 777          * Handle zero-sized whitelist or empty thread_struct, otherwise
 778          * adjust offset to position of thread_struct in task_struct.
 779          */
 780         if (unlikely(*size == 0))
 781                 *offset = 0;
 782         else
 783                 *offset += offsetof(struct task_struct, thread);
 784 }
 785
 786 void __init fork_init(void)
 787 {
 788         int i;
 789 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 790 #ifndef ARCH_MIN_TASKALIGN
 791 #define ARCH_MIN_TASKALIGN      0
 792 #endif
 793         int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
 794         unsigned long useroffset, usersize;
 795
 796         /* create a slab on which task_structs can be allocated */
 797         task_struct_whitelist(&useroffset, &usersize);
 798         task_struct_cachep = kmem_cache_create_usercopy("task_struct",
 799                         arch_task_struct_size, align,
 800                         SLAB_PANIC|SLAB_ACCOUNT,
 801                         useroffset, usersize, NULL);
 802 #endif
 803
 804         /* do the arch specific task caches init */
 805         arch_task_cache_init();
 806
 807         set_max_threads(MAX_THREADS);
 808
 809         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 810         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 811         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 812                 init_task.signal->rlim[RLIMIT_NPROC];
 813
 814         for (i = 0; i < UCOUNT_COUNTS; i++) {
 815                 init_user_ns.ucount_max[i] = max_threads/2;
 816         }
 817
 818 #ifdef CONFIG_VMAP_STACK
 819         cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
 820                           NULL, free_vm_stack_cache);
 821 #endif
 822
 823         lockdep_init_task(&init_task);
 824         uprobes_init();
 825 }
 826
 827 int __weak arch_dup_task_struct(struct task_struct *dst,
 828                                                struct task_struct *src)
 829 {
 830         *dst = *src;
 831         return 0;
 832 }
 833
 834 void set_task_stack_end_magic(struct task_struct *tsk)
 835 {
 836         unsigned long *stackend;
 837
 838         stackend = end_of_stack(tsk);
 839         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 840 }
 841
 842 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 843 {
 844         struct task_struct *tsk;
 845         unsigned long *stack;
 846         struct vm_struct *stack_vm_area __maybe_unused;
 847         int err;
 848
 849         if (node == NUMA_NO_NODE)
 850                 node = tsk_fork_get_node(orig);
 851         tsk = alloc_task_struct_node(node);
 852         if (!tsk)
 853                 return NULL;
 854
 855         stack = alloc_thread_stack_node(tsk, node);
 856         if (!stack)
 857                 goto free_tsk;
 858
 859         if (memcg_charge_kernel_stack(tsk))
 860                 goto free_stack;
 861
 862         stack_vm_area = task_stack_vm_area(tsk);
 863
 864         err = arch_dup_task_struct(tsk, orig);
 865
 866         /*
 867          * arch_dup_task_struct() clobbers the stack-related fields.  Make
 868          * sure they're properly initialized before using any stack-related
 869          * functions again.
 870          */
 871         tsk->stack = stack;
 872 #ifdef CONFIG_VMAP_STACK
 873         tsk->stack_vm_area = stack_vm_area;
 874 #endif
 875 #ifdef CONFIG_THREAD_INFO_IN_TASK
 876         refcount_set(&tsk->stack_refcount, 1);
 877 #endif
 878
 879         if (err)
 880                 goto free_stack;
 881
 882 #ifdef CONFIG_SECCOMP
 883         /*
 884          * We must handle setting up seccomp filters once we're under
 885          * the sighand lock in case orig has changed between now and
 886          * then. Until then, filter must be NULL to avoid messing up
 887          * the usage counts on the error path calling free_task.
 888          */
 889         tsk->seccomp.filter = NULL;
 890 #endif
 891
 892         setup_thread_stack(tsk, orig);
 893         clear_user_return_notifier(tsk);
 894         clear_tsk_need_resched(tsk);
 895         set_task_stack_end_magic(tsk);
 896
 897 #ifdef CONFIG_STACKPROTECTOR
 898         tsk->stack_canary = get_random_canary();
 899 #endif
 900
 901         /*
 902          * One for us, one for whoever does the "release_task()" (usually
 903          * parent)
 904          */
 905         refcount_set(&tsk->usage, 2);
 906 #ifdef CONFIG_BLK_DEV_IO_TRACE
 907         tsk->btrace_seq = 0;
 908 #endif
 909         tsk->splice_pipe = NULL;
 910         tsk->task_frag.page = NULL;
 911         tsk->wake_q.next = NULL;
 912
 913         account_kernel_stack(tsk, 1);
 914
 915         kcov_task_init(tsk);
 916
 917 #ifdef CONFIG_FAULT_INJECTION
 918         tsk->fail_nth = 0;
 919 #endif
 920
 921 #ifdef CONFIG_BLK_CGROUP
 922         tsk->throttle_queue = NULL;
 923         tsk->use_memdelay = 0;
 924 #endif
 925
 926 #ifdef CONFIG_MEMCG
 927         tsk->active_memcg = NULL;
 928 #endif
 929         return tsk;
 930
 931 free_stack:
 932         free_thread_stack(tsk);
 933 free_tsk:
 934         free_task_struct(tsk);
 935         return NULL;
 936 }
 937
 938 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 939
 940 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 941
 942 static int __init coredump_filter_setup(char *s)
 943 {
 944         default_dump_filter =
 945                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 946                 MMF_DUMP_FILTER_MASK;
 947         return 1;
 948 }
 949
 950 __setup("coredump_filter=", coredump_filter_setup);
 951
 952 #include <linux/init_task.h>
 953
 954 static void mm_init_aio(struct mm_struct *mm)
 955 {
 956 #ifdef CONFIG_AIO
 957         spin_lock_init(&mm->ioctx_lock);
 958         mm->ioctx_table = NULL;
 959 #endif
 960 }
 961
 962 static __always_inline void mm_clear_owner(struct mm_struct *mm,
 963                                            struct task_struct *p)
 964 {
 965 #ifdef CONFIG_MEMCG
 966         if (mm->owner == p)
 967                 WRITE_ONCE(mm->owner, NULL);
 968 #endif
 969 }
 970
 971 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 972 {
 973 #ifdef CONFIG_MEMCG
 974         mm->owner = p;
 975 #endif
 976 }
 977
 978 static void mm_init_uprobes_state(struct mm_struct *mm)
 979 {
 980 #ifdef CONFIG_UPROBES
 981         mm->uprobes_state.xol_area = NULL;
 982 #endif
 983 }
 984
 985 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
 986         struct user_namespace *user_ns)
 987 {
 988         mm->mmap = NULL;
 989         mm->mm_rb = RB_ROOT;
 990         mm->vmacache_seqnum = 0;
 991         atomic_set(&mm->mm_users, 1);
 992         atomic_set(&mm->mm_count, 1);
 993         init_rwsem(&mm->mmap_sem);
 994         INIT_LIST_HEAD(&mm->mmlist);
 995         mm->core_state = NULL;
 996         mm_pgtables_bytes_init(mm);
 997         mm->map_count = 0;
 998         mm->locked_vm = 0;
 999         atomic64_set(&mm->pinned_vm, 0);
1000         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1001         spin_lock_init(&mm->page_table_lock);
1002         spin_lock_init(&mm->arg_lock);
1003         mm_init_cpumask(mm);
1004         mm_init_aio(mm);
1005         mm_init_owner(mm, p);
1006         RCU_INIT_POINTER(mm->exe_file, NULL);
1007         mmu_notifier_mm_init(mm);
1008         hmm_mm_init(mm);
1009         init_tlb_flush_pending(mm);
1010 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1011         mm->pmd_huge_pte = NULL;
1012 #endif
1013         mm_init_uprobes_state(mm);
1014
1015         if (current->mm) {
1016                 mm->flags = current->mm->flags & MMF_INIT_MASK;
1017                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1018         } else {
1019                 mm->flags = default_dump_filter;
1020                 mm->def_flags = 0;
1021         }
1022
1023         if (mm_alloc_pgd(mm))
1024                 goto fail_nopgd;
1025
1026         if (init_new_context(p, mm))
1027                 goto fail_nocontext;
1028
1029         mm->user_ns = get_user_ns(user_ns);
1030         return mm;
1031
1032 fail_nocontext:
1033         mm_free_pgd(mm);
1034 fail_nopgd:
1035         free_mm(mm);
1036         return NULL;
1037 }
1038
1039 /*
1040  * Allocate and initialize an mm_struct.
1041  */
1042 struct mm_struct *mm_alloc(void)
1043 {
1044         struct mm_struct *mm;
1045
1046         mm = allocate_mm();
1047         if (!mm)
1048                 return NULL;
1049
1050         memset(mm, 0, sizeof(*mm));
1051         return mm_init(mm, current, current_user_ns());
1052 }
1053
1054 static inline void __mmput(struct mm_struct *mm)
1055 {
1056         VM_BUG_ON(atomic_read(&mm->mm_users));
1057
1058         uprobe_clear_state(mm);
1059         exit_aio(mm);
1060         ksm_exit(mm);
1061         khugepaged_exit(mm); /* must run before exit_mmap */
1062         exit_mmap(mm);
1063         mm_put_huge_zero_page(mm);
1064         set_mm_exe_file(mm, NULL);
1065         if (!list_empty(&mm->mmlist)) {
1066                 spin_lock(&mmlist_lock);
1067                 list_del(&mm->mmlist);
1068                 spin_unlock(&mmlist_lock);
1069         }
1070         if (mm->binfmt)
1071                 module_put(mm->binfmt->module);
1072         mmdrop(mm);
1073 }
1074
1075 /*
1076  * Decrement the use count and release all resources for an mm.
1077  */
1078 void mmput(struct mm_struct *mm)
1079 {
1080         might_sleep();
1081
1082         if (atomic_dec_and_test(&mm->mm_users))
1083                 __mmput(mm);
1084 }
1085 EXPORT_SYMBOL_GPL(mmput);
1086
1087 #ifdef CONFIG_MMU
1088 static void mmput_async_fn(struct work_struct *work)
1089 {
1090         struct mm_struct *mm = container_of(work, struct mm_struct,
1091                                             async_put_work);
1092
1093         __mmput(mm);
1094 }
1095
1096 void mmput_async(struct mm_struct *mm)
1097 {
1098         if (atomic_dec_and_test(&mm->mm_users)) {
1099                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1100                 schedule_work(&mm->async_put_work);
1101         }
1102 }
1103 #endif
1104
1105 /**
1106  * set_mm_exe_file - change a reference to the mm's executable file
1107  *
1108  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1109  *
1110  * Main users are mmput() and sys_execve(). Callers prevent concurrent
1111  * invocations: in mmput() nobody alive left, in execve task is single
1112  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1113  * mm->exe_file, but does so without using set_mm_exe_file() in order
1114  * to do avoid the need for any locks.
1115  */
1116 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1117 {
1118         struct file *old_exe_file;
1119
1120         /*
1121          * It is safe to dereference the exe_file without RCU as
1122          * this function is only called if nobody else can access
1123          * this mm -- see comment above for justification.
1124          */
1125         old_exe_file = rcu_dereference_raw(mm->exe_file);
1126
1127         if (new_exe_file)
1128                 get_file(new_exe_file);
1129         rcu_assign_pointer(mm->exe_file, new_exe_file);
1130         if (old_exe_file)
1131                 fput(old_exe_file);
1132 }
1133
1134 /**
1135  * get_mm_exe_file - acquire a reference to the mm's executable file
1136  *
1137  * Returns %NULL if mm has no associated executable file.
1138  * User must release file via fput().
1139  */
1140 struct file *get_mm_exe_file(struct mm_struct *mm)
1141 {
1142         struct file *exe_file;
1143
1144         rcu_read_lock();
1145         exe_file = rcu_dereference(mm->exe_file);
1146         if (exe_file && !get_file_rcu(exe_file))
1147                 exe_file = NULL;
1148         rcu_read_unlock();
1149         return exe_file;
1150 }
1151 EXPORT_SYMBOL(get_mm_exe_file);
1152
1153 /**
1154  * get_task_exe_file - acquire a reference to the task's executable file
1155  *
1156  * Returns %NULL if task's mm (if any) has no associated executable file or
1157  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1158  * User must release file via fput().
1159  */
1160 struct file *get_task_exe_file(struct task_struct *task)
1161 {
1162         struct file *exe_file = NULL;
1163         struct mm_struct *mm;
1164
1165         task_lock(task);
1166         mm = task->mm;
1167         if (mm) {
1168                 if (!(task->flags & PF_KTHREAD))
1169                         exe_file = get_mm_exe_file(mm);
1170         }
1171         task_unlock(task);
1172         return exe_file;
1173 }
1174 EXPORT_SYMBOL(get_task_exe_file);
1175
1176 /**
1177  * get_task_mm - acquire a reference to the task's mm
1178  *
1179  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1180  * this kernel workthread has transiently adopted a user mm with use_mm,
1181  * to do its AIO) is not set and if so returns a reference to it, after
1182  * bumping up the use count.  User must release the mm via mmput()
1183  * after use.  Typically used by /proc and ptrace.
1184  */
1185 struct mm_struct *get_task_mm(struct task_struct *task)
1186 {
1187         struct mm_struct *mm;
1188
1189         task_lock(task);
1190         mm = task->mm;
1191         if (mm) {
1192                 if (task->flags & PF_KTHREAD)
1193                         mm = NULL;
1194                 else
1195                         mmget(mm);
1196         }
1197         task_unlock(task);
1198         return mm;
1199 }
1200 EXPORT_SYMBOL_GPL(get_task_mm);
1201
1202 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1203 {
1204         struct mm_struct *mm;
1205         int err;
1206
1207         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
1208         if (err)
1209                 return ERR_PTR(err);
1210
1211         mm = get_task_mm(task);
1212         if (mm && mm != current->mm &&
1213                         !ptrace_may_access(task, mode)) {
1214                 mmput(mm);
1215                 mm = ERR_PTR(-EACCES);
1216         }
1217         mutex_unlock(&task->signal->cred_guard_mutex);
1218
1219         return mm;
1220 }
1221
1222 static void complete_vfork_done(struct task_struct *tsk)
1223 {
1224         struct completion *vfork;
1225
1226         task_lock(tsk);
1227         vfork = tsk->vfork_done;
1228         if (likely(vfork)) {
1229                 tsk->vfork_done = NULL;
1230                 complete(vfork);
1231         }
1232         task_unlock(tsk);
1233 }
1234
1235 static int wait_for_vfork_done(struct task_struct *child,
1236                                 struct completion *vfork)
1237 {
1238         int killed;
1239
1240         freezer_do_not_count();
1241         cgroup_enter_frozen();
1242         killed = wait_for_completion_killable(vfork);
1243         cgroup_leave_frozen(false);
1244         freezer_count();
1245
1246         if (killed) {
1247                 task_lock(child);
1248                 child->vfork_done = NULL;
1249                 task_unlock(child);
1250         }
1251
1252         put_task_struct(child);
1253         return killed;
1254 }
1255
1256 /* Please note the differences between mmput and mm_release.
1257  * mmput is called whenever we stop holding onto a mm_struct,
1258  * error success whatever.
1259  *
1260  * mm_release is called after a mm_struct has been removed
1261  * from the current process.
1262  *
1263  * This difference is important for error handling, when we
1264  * only half set up a mm_struct for a new process and need to restore
1265  * the old one.  Because we mmput the new mm_struct before
1266  * restoring the old one. . .
1267  * Eric Biederman 10 January 1998
1268  */
1269 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1270 {
1271         /* Get rid of any futexes when releasing the mm */
1272 #ifdef CONFIG_FUTEX
1273         if (unlikely(tsk->robust_list)) {
1274                 exit_robust_list(tsk);
1275                 tsk->robust_list = NULL;
1276         }
1277 #ifdef CONFIG_COMPAT
1278         if (unlikely(tsk->compat_robust_list)) {
1279                 compat_exit_robust_list(tsk);
1280                 tsk->compat_robust_list = NULL;
1281         }
1282 #endif
1283         if (unlikely(!list_empty(&tsk->pi_state_list)))
1284                 exit_pi_state_list(tsk);
1285 #endif
1286
1287         uprobe_free_utask(tsk);
1288
1289         /* Get rid of any cached register state */
1290         deactivate_mm(tsk, mm);
1291
1292         /*
1293          * Signal userspace if we're not exiting with a core dump
1294          * because we want to leave the value intact for debugging
1295          * purposes.
1296          */
1297         if (tsk->clear_child_tid) {
1298                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1299                     atomic_read(&mm->mm_users) > 1) {
1300                         /*
1301                          * We don't check the error code - if userspace has
1302                          * not set up a proper pointer then tough luck.
1303                          */
1304                         put_user(0, tsk->clear_child_tid);
1305                         do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1306                                         1, NULL, NULL, 0, 0);
1307                 }
1308                 tsk->clear_child_tid = NULL;
1309         }
1310
1311         /*
1312          * All done, finally we can wake up parent and return this mm to him.
1313          * Also kthread_stop() uses this completion for synchronization.
1314          */
1315         if (tsk->vfork_done)
1316                 complete_vfork_done(tsk);
1317 }
1318
1319 /**
1320  * dup_mm() - duplicates an existing mm structure
1321  * @tsk: the task_struct with which the new mm will be associated.
1322  * @oldmm: the mm to duplicate.
1323  *
1324  * Allocates a new mm structure and duplicates the provided @oldmm structure
1325  * content into it.
1326  *
1327  * Return: the duplicated mm or NULL on failure.
1328  */
1329 static struct mm_struct *dup_mm(struct task_struct *tsk,
1330                                 struct mm_struct *oldmm)
1331 {
1332         struct mm_struct *mm;
1333         int err;
1334
1335         mm = allocate_mm();
1336         if (!mm)
1337                 goto fail_nomem;
1338
1339         memcpy(mm, oldmm, sizeof(*mm));
1340
1341         if (!mm_init(mm, tsk, mm->user_ns))
1342                 goto fail_nomem;
1343
1344         err = dup_mmap(mm, oldmm);
1345         if (err)
1346                 goto free_pt;
1347
1348         mm->hiwater_rss = get_mm_rss(mm);
1349         mm->hiwater_vm = mm->total_vm;
1350
1351         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1352                 goto free_pt;
1353
1354         return mm;
1355
1356 free_pt:
1357         /* don't put binfmt in mmput, we haven't got module yet */
1358         mm->binfmt = NULL;
1359         mm_init_owner(mm, NULL);
1360         mmput(mm);
1361
1362 fail_nomem:
1363         return NULL;
1364 }
1365
1366 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1367 {
1368         struct mm_struct *mm, *oldmm;
1369         int retval;
1370
1371         tsk->min_flt = tsk->maj_flt = 0;
1372         tsk->nvcsw = tsk->nivcsw = 0;
1373 #ifdef CONFIG_DETECT_HUNG_TASK
1374         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1375         tsk->last_switch_time = 0;
1376 #endif
1377
1378         tsk->mm = NULL;
1379         tsk->active_mm = NULL;
1380
1381         /*
1382          * Are we cloning a kernel thread?
1383          *
1384          * We need to steal a active VM for that..
1385          */
1386         oldmm = current->mm;
1387         if (!oldmm)
1388                 return 0;
1389
1390         /* initialize the new vmacache entries */
1391         vmacache_flush(tsk);
1392
1393         if (clone_flags & CLONE_VM) {
1394                 mmget(oldmm);
1395                 mm = oldmm;
1396                 goto good_mm;
1397         }
1398
1399         retval = -ENOMEM;
1400         mm = dup_mm(tsk, current->mm);
1401         if (!mm)
1402                 goto fail_nomem;
1403
1404 good_mm:
1405         tsk->mm = mm;
1406         tsk->active_mm = mm;
1407         return 0;
1408
1409 fail_nomem:
1410         return retval;
1411 }
1412
1413 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1414 {
1415         struct fs_struct *fs = current->fs;
1416         if (clone_flags & CLONE_FS) {
1417                 /* tsk->fs is already what we want */
1418                 spin_lock(&fs->lock);
1419                 if (fs->in_exec) {
1420                         spin_unlock(&fs->lock);
1421                         return -EAGAIN;
1422                 }
1423                 fs->users++;
1424                 spin_unlock(&fs->lock);
1425                 return 0;
1426         }
1427         tsk->fs = copy_fs_struct(fs);
1428         if (!tsk->fs)
1429                 return -ENOMEM;
1430         return 0;
1431 }
1432
1433 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1434 {
1435         struct files_struct *oldf, *newf;
1436         int error = 0;
1437
1438         /*
1439          * A background process may not have any files ...
1440          */
1441         oldf = current->files;
1442         if (!oldf)
1443                 goto out;
1444
1445         if (clone_flags & CLONE_FILES) {
1446                 atomic_inc(&oldf->count);
1447                 goto out;
1448         }
1449
1450         newf = dup_fd(oldf, &error);
1451         if (!newf)
1452                 goto out;
1453
1454         tsk->files = newf;
1455         error = 0;
1456 out:
1457         return error;
1458 }
1459
1460 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1461 {
1462 #ifdef CONFIG_BLOCK
1463         struct io_context *ioc = current->io_context;
1464         struct io_context *new_ioc;
1465
1466         if (!ioc)
1467                 return 0;
1468         /*
1469          * Share io context with parent, if CLONE_IO is set
1470          */
1471         if (clone_flags & CLONE_IO) {
1472                 ioc_task_link(ioc);
1473                 tsk->io_context = ioc;
1474         } else if (ioprio_valid(ioc->ioprio)) {
1475                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1476                 if (unlikely(!new_ioc))
1477                         return -ENOMEM;
1478
1479                 new_ioc->ioprio = ioc->ioprio;
1480                 put_io_context(new_ioc);
1481         }
1482 #endif
1483         return 0;
1484 }
1485
1486 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1487 {
1488         struct sighand_struct *sig;
1489
1490         if (clone_flags & CLONE_SIGHAND) {
1491                 refcount_inc(&current->sighand->count);
1492                 return 0;
1493         }
1494         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1495         rcu_assign_pointer(tsk->sighand, sig);
1496         if (!sig)
1497                 return -ENOMEM;
1498
1499         refcount_set(&sig->count, 1);
1500         spin_lock_irq(&current->sighand->siglock);
1501         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1502         spin_unlock_irq(&current->sighand->siglock);
1503         return 0;
1504 }
1505
1506 void __cleanup_sighand(struct sighand_struct *sighand)
1507 {
1508         if (refcount_dec_and_test(&sighand->count)) {
1509                 signalfd_cleanup(sighand);
1510                 /*
1511                  * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1512                  * without an RCU grace period, see __lock_task_sighand().
1513                  */
1514                 kmem_cache_free(sighand_cachep, sighand);
1515         }
1516 }
1517
1518 #ifdef CONFIG_POSIX_TIMERS
1519 /*
1520  * Initialize POSIX timer handling for a thread group.
1521  */
1522 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1523 {
1524         unsigned long cpu_limit;
1525
1526         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1527         if (cpu_limit != RLIM_INFINITY) {
1528                 sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1529                 sig->cputimer.running = true;
1530         }
1531
1532         /* The timer lists. */
1533         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1534         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1535         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1536 }
1537 #else
1538 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1539 #endif
1540
1541 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1542 {
1543         struct signal_struct *sig;
1544
1545         if (clone_flags & CLONE_THREAD)
1546                 return 0;
1547
1548         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1549         tsk->signal = sig;
1550         if (!sig)
1551                 return -ENOMEM;
1552
1553         sig->nr_threads = 1;
1554         atomic_set(&sig->live, 1);
1555         refcount_set(&sig->sigcnt, 1);
1556
1557         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1558         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1559         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1560
1561         init_waitqueue_head(&sig->wait_chldexit);
1562         sig->curr_target = tsk;
1563         init_sigpending(&sig->shared_pending);
1564         INIT_HLIST_HEAD(&sig->multiprocess);
1565         seqlock_init(&sig->stats_lock);
1566         prev_cputime_init(&sig->prev_cputime);
1567
1568 #ifdef CONFIG_POSIX_TIMERS
1569         INIT_LIST_HEAD(&sig->posix_timers);
1570         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1571         sig->real_timer.function = it_real_fn;
1572 #endif
1573
1574         task_lock(current->group_leader);
1575         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1576         task_unlock(current->group_leader);
1577
1578         posix_cpu_timers_init_group(sig);
1579
1580         tty_audit_fork(sig);
1581         sched_autogroup_fork(sig);
1582
1583         sig->oom_score_adj = current->signal->oom_score_adj;
1584         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1585
1586         mutex_init(&sig->cred_guard_mutex);
1587
1588         return 0;
1589 }
1590
1591 static void copy_seccomp(struct task_struct *p)
1592 {
1593 #ifdef CONFIG_SECCOMP
1594         /*
1595          * Must be called with sighand->lock held, which is common to
1596          * all threads in the group. Holding cred_guard_mutex is not
1597          * needed because this new task is not yet running and cannot
1598          * be racing exec.
1599          */
1600         assert_spin_locked(&current->sighand->siglock);
1601
1602         /* Ref-count the new filter user, and assign it. */
1603         get_seccomp_filter(current);
1604         p->seccomp = current->seccomp;
1605
1606         /*
1607          * Explicitly enable no_new_privs here in case it got set
1608          * between the task_struct being duplicated and holding the
1609          * sighand lock. The seccomp state and nnp must be in sync.
1610          */
1611         if (task_no_new_privs(current))
1612                 task_set_no_new_privs(p);
1613
1614         /*
1615          * If the parent gained a seccomp mode after copying thread
1616          * flags and between before we held the sighand lock, we have
1617          * to manually enable the seccomp thread flag here.
1618          */
1619         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1620                 set_tsk_thread_flag(p, TIF_SECCOMP);
1621 #endif
1622 }
1623
1624 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1625 {
1626         current->clear_child_tid = tidptr;
1627
1628         return task_pid_vnr(current);
1629 }
1630
1631 static void rt_mutex_init_task(struct task_struct *p)
1632 {
1633         raw_spin_lock_init(&p->pi_lock);
1634 #ifdef CONFIG_RT_MUTEXES
1635         p->pi_waiters = RB_ROOT_CACHED;
1636         p->pi_top_task = NULL;
1637         p->pi_blocked_on = NULL;
1638 #endif
1639 }
1640
1641 #ifdef CONFIG_POSIX_TIMERS
1642 /*
1643  * Initialize POSIX timer handling for a single task.
1644  */
1645 static void posix_cpu_timers_init(struct task_struct *tsk)
1646 {
1647         tsk->cputime_expires.prof_exp = 0;
1648         tsk->cputime_expires.virt_exp = 0;
1649         tsk->cputime_expires.sched_exp = 0;
1650         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1651         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1652         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1653 }
1654 #else
1655 static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1656 #endif
1657
1658 static inline void init_task_pid_links(struct task_struct *task)
1659 {
1660         enum pid_type type;
1661
1662         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1663                 INIT_HLIST_NODE(&task->pid_links[type]);
1664         }
1665 }
1666
1667 static inline void
1668 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1669 {
1670         if (type == PIDTYPE_PID)
1671                 task->thread_pid = pid;
1672         else
1673                 task->signal->pids[type] = pid;
1674 }
1675
1676 static inline void rcu_copy_process(struct task_struct *p)
1677 {
1678 #ifdef CONFIG_PREEMPT_RCU
1679         p->rcu_read_lock_nesting = 0;
1680         p->rcu_read_unlock_special.s = 0;
1681         p->rcu_blocked_node = NULL;
1682         INIT_LIST_HEAD(&p->rcu_node_entry);
1683 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1684 #ifdef CONFIG_TASKS_RCU
1685         p->rcu_tasks_holdout = false;
1686         INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1687         p->rcu_tasks_idle_cpu = -1;
1688 #endif /* #ifdef CONFIG_TASKS_RCU */
1689 }
1690
1691 static int pidfd_release(struct inode *inode, struct file *file)
1692 {
1693         struct pid *pid = file->private_data;
1694
1695         file->private_data = NULL;
1696         put_pid(pid);
1697         return 0;
1698 }
1699
1700 #ifdef CONFIG_PROC_FS
1701 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1702 {
1703         struct pid_namespace *ns = proc_pid_ns(file_inode(m->file));
1704         struct pid *pid = f->private_data;
1705
1706         seq_put_decimal_ull(m, "Pid:\t", pid_nr_ns(pid, ns));
1707         seq_putc(m, '\n');
1708 }
1709 #endif
1710
1711 const struct file_operations pidfd_fops = {
1712         .release = pidfd_release,
1713 #ifdef CONFIG_PROC_FS
1714         .show_fdinfo = pidfd_show_fdinfo,
1715 #endif
1716 };
1717
1718 static void __delayed_free_task(struct rcu_head *rhp)
1719 {
1720         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1721
1722         free_task(tsk);
1723 }
1724
1725 static __always_inline void delayed_free_task(struct task_struct *tsk)
1726 {
1727         if (IS_ENABLED(CONFIG_MEMCG))
1728                 call_rcu(&tsk->rcu, __delayed_free_task);
1729         else
1730                 free_task(tsk);
1731 }
1732
1733 /*
1734  * This creates a new process as a copy of the old one,
1735  * but does not actually start it yet.
1736  *
1737  * It copies the registers, and all the appropriate
1738  * parts of the process environment (as per the clone
1739  * flags). The actual kick-off is left to the caller.
1740  */
1741 static __latent_entropy struct task_struct *copy_process(
1742                                         unsigned long clone_flags,
1743                                         unsigned long stack_start,
1744                                         unsigned long stack_size,
1745                                         int __user *parent_tidptr,
1746                                         int __user *child_tidptr,
1747                                         struct pid *pid,
1748                                         int trace,
1749                                         unsigned long tls,
1750                                         int node)
1751 {
1752         int pidfd = -1, retval;
1753         struct task_struct *p;
1754         struct multiprocess_signals delayed;
1755         struct file *pidfile = NULL;
1756
1757         /*
1758          * Don't allow sharing the root directory with processes in a different
1759          * namespace
1760          */
1761         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1762                 return ERR_PTR(-EINVAL);
1763
1764         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1765                 return ERR_PTR(-EINVAL);
1766
1767         /*
1768          * Thread groups must share signals as well, and detached threads
1769          * can only be started up within the thread group.
1770          */
1771         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1772                 return ERR_PTR(-EINVAL);
1773
1774         /*
1775          * Shared signal handlers imply shared VM. By way of the above,
1776          * thread groups also imply shared VM. Blocking this case allows
1777          * for various simplifications in other code.
1778          */
1779         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1780                 return ERR_PTR(-EINVAL);
1781
1782         /*
1783          * Siblings of global init remain as zombies on exit since they are
1784          * not reaped by their parent (swapper). To solve this and to avoid
1785          * multi-rooted process trees, prevent global and container-inits
1786          * from creating siblings.
1787          */
1788         if ((clone_flags & CLONE_PARENT) &&
1789                                 current->signal->flags & SIGNAL_UNKILLABLE)
1790                 return ERR_PTR(-EINVAL);
1791
1792         /*
1793          * If the new process will be in a different pid or user namespace
1794          * do not allow it to share a thread group with the forking task.
1795          */
1796         if (clone_flags & CLONE_THREAD) {
1797                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1798                     (task_active_pid_ns(current) !=
1799                                 current->nsproxy->pid_ns_for_children))
1800                         return ERR_PTR(-EINVAL);
1801         }
1802
1803         if (clone_flags & CLONE_PIDFD) {
1804                 /*
1805                  * - CLONE_PARENT_SETTID is useless for pidfds and also
1806                  *   parent_tidptr is used to return pidfds.
1807                  * - CLONE_DETACHED is blocked so that we can potentially
1808                  *   reuse it later for CLONE_PIDFD.
1809                  * - CLONE_THREAD is blocked until someone really needs it.
1810                  */
1811                 if (clone_flags &
1812                     (CLONE_DETACHED | CLONE_PARENT_SETTID | CLONE_THREAD))
1813                         return ERR_PTR(-EINVAL);
1814         }
1815
1816         /*
1817          * Force any signals received before this point to be delivered
1818          * before the fork happens.  Collect up signals sent to multiple
1819          * processes that happen during the fork and delay them so that
1820          * they appear to happen after the fork.
1821          */
1822         sigemptyset(&delayed.signal);
1823         INIT_HLIST_NODE(&delayed.node);
1824
1825         spin_lock_irq(&current->sighand->siglock);
1826         if (!(clone_flags & CLONE_THREAD))
1827                 hlist_add_head(&delayed.node, &current->signal->multiprocess);
1828         recalc_sigpending();
1829         spin_unlock_irq(&current->sighand->siglock);
1830         retval = -ERESTARTNOINTR;
1831         if (signal_pending(current))
1832                 goto fork_out;
1833
1834         retval = -ENOMEM;
1835         p = dup_task_struct(current, node);
1836         if (!p)
1837                 goto fork_out;
1838
1839         /*
1840          * This _must_ happen before we call free_task(), i.e. before we jump
1841          * to any of the bad_fork_* labels. This is to avoid freeing
1842          * p->set_child_tid which is (ab)used as a kthread's data pointer for
1843          * kernel threads (PF_KTHREAD).
1844          */
1845         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1846         /*
1847          * Clear TID on mm_release()?
1848          */
1849         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1850
1851         ftrace_graph_init_task(p);
1852
1853         rt_mutex_init_task(p);
1854
1855 #ifdef CONFIG_PROVE_LOCKING
1856         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1857         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1858 #endif
1859         retval = -EAGAIN;
1860         if (atomic_read(&p->real_cred->user->processes) >=
1861                         task_rlimit(p, RLIMIT_NPROC)) {
1862                 if (p->real_cred->user != INIT_USER &&
1863                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1864                         goto bad_fork_free;
1865         }
1866         current->flags &= ~PF_NPROC_EXCEEDED;
1867
1868         retval = copy_creds(p, clone_flags);
1869         if (retval < 0)
1870                 goto bad_fork_free;
1871
1872         /*
1873          * If multiple threads are within copy_process(), then this check
1874          * triggers too late. This doesn't hurt, the check is only there
1875          * to stop root fork bombs.
1876          */
1877         retval = -EAGAIN;
1878         if (nr_threads >= max_threads)
1879                 goto bad_fork_cleanup_count;
1880
1881         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1882         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1883         p->flags |= PF_FORKNOEXEC;
1884         INIT_LIST_HEAD(&p->children);
1885         INIT_LIST_HEAD(&p->sibling);
1886         rcu_copy_process(p);
1887         p->vfork_done = NULL;
1888         spin_lock_init(&p->alloc_lock);
1889
1890         init_sigpending(&p->pending);
1891
1892         p->utime = p->stime = p->gtime = 0;
1893 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1894         p->utimescaled = p->stimescaled = 0;
1895 #endif
1896         prev_cputime_init(&p->prev_cputime);
1897
1898 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1899         seqcount_init(&p->vtime.seqcount);
1900         p->vtime.starttime = 0;
1901         p->vtime.state = VTIME_INACTIVE;
1902 #endif
1903
1904 #if defined(SPLIT_RSS_COUNTING)
1905         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1906 #endif
1907
1908         p->default_timer_slack_ns = current->timer_slack_ns;
1909
1910 #ifdef CONFIG_PSI
1911         p->psi_flags = 0;
1912 #endif
1913
1914         task_io_accounting_init(&p->ioac);
1915         acct_clear_integrals(p);
1916
1917         posix_cpu_timers_init(p);
1918
1919         p->io_context = NULL;
1920         audit_set_context(p, NULL);
1921         cgroup_fork(p);
1922 #ifdef CONFIG_NUMA
1923         p->mempolicy = mpol_dup(p->mempolicy);
1924         if (IS_ERR(p->mempolicy)) {
1925                 retval = PTR_ERR(p->mempolicy);
1926                 p->mempolicy = NULL;
1927                 goto bad_fork_cleanup_threadgroup_lock;
1928         }
1929 #endif
1930 #ifdef CONFIG_CPUSETS
1931         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1932         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1933         seqcount_init(&p->mems_allowed_seq);
1934 #endif
1935 #ifdef CONFIG_TRACE_IRQFLAGS
1936         p->irq_events = 0;
1937         p->hardirqs_enabled = 0;
1938         p->hardirq_enable_ip = 0;
1939         p->hardirq_enable_event = 0;
1940         p->hardirq_disable_ip = _THIS_IP_;
1941         p->hardirq_disable_event = 0;
1942         p->softirqs_enabled = 1;
1943         p->softirq_enable_ip = _THIS_IP_;
1944         p->softirq_enable_event = 0;
1945         p->softirq_disable_ip = 0;
1946         p->softirq_disable_event = 0;
1947         p->hardirq_context = 0;
1948         p->softirq_context = 0;
1949 #endif
1950
1951         p->pagefault_disabled = 0;
1952
1953 #ifdef CONFIG_LOCKDEP
1954         p->lockdep_depth = 0; /* no locks held yet */
1955         p->curr_chain_key = 0;
1956         p->lockdep_recursion = 0;
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                         goto bad_fork_free_pid;
2039                 }
2040                 get_pid(pid);   /* held by pidfile now */
2041
2042                 retval = put_user(pidfd, parent_tidptr);
2043                 if (retval)
2044                         goto bad_fork_put_pidfd;
2045         }
2046
2047 #ifdef CONFIG_BLOCK
2048         p->plug = NULL;
2049 #endif
2050 #ifdef CONFIG_FUTEX
2051         p->robust_list = NULL;
2052 #ifdef CONFIG_COMPAT
2053         p->compat_robust_list = NULL;
2054 #endif
2055         INIT_LIST_HEAD(&p->pi_state_list);
2056         p->pi_state_cache = NULL;
2057 #endif
2058         /*
2059          * sigaltstack should be cleared when sharing the same VM
2060          */
2061         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2062                 sas_ss_reset(p);
2063
2064         /*
2065          * Syscall tracing and stepping should be turned off in the
2066          * child regardless of CLONE_PTRACE.
2067          */
2068         user_disable_single_step(p);
2069         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
2070 #ifdef TIF_SYSCALL_EMU
2071         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
2072 #endif
2073         clear_tsk_latency_tracing(p);
2074
2075         /* ok, now we should be set up.. */
2076         p->pid = pid_nr(pid);
2077         if (clone_flags & CLONE_THREAD) {
2078                 p->exit_signal = -1;
2079                 p->group_leader = current->group_leader;
2080                 p->tgid = current->tgid;
2081         } else {
2082                 if (clone_flags & CLONE_PARENT)
2083                         p->exit_signal = current->group_leader->exit_signal;
2084                 else
2085                         p->exit_signal = (clone_flags & CSIGNAL);
2086                 p->group_leader = p;
2087                 p->tgid = p->pid;
2088         }
2089
2090         p->nr_dirtied = 0;
2091         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2092         p->dirty_paused_when = 0;
2093
2094         p->pdeath_signal = 0;
2095         INIT_LIST_HEAD(&p->thread_group);
2096         p->task_works = NULL;
2097
2098         cgroup_threadgroup_change_begin(current);
2099         /*
2100          * Ensure that the cgroup subsystem policies allow the new process to be
2101          * forked. It should be noted the the new process's css_set can be changed
2102          * between here and cgroup_post_fork() if an organisation operation is in
2103          * progress.
2104          */
2105         retval = cgroup_can_fork(p);
2106         if (retval)
2107                 goto bad_fork_cgroup_threadgroup_change_end;
2108
2109         /*
2110          * From this point on we must avoid any synchronous user-space
2111          * communication until we take the tasklist-lock. In particular, we do
2112          * not want user-space to be able to predict the process start-time by
2113          * stalling fork(2) after we recorded the start_time but before it is
2114          * visible to the system.
2115          */
2116
2117         p->start_time = ktime_get_ns();
2118         p->real_start_time = ktime_get_boot_ns();
2119
2120         /*
2121          * Make it visible to the rest of the system, but dont wake it up yet.
2122          * Need tasklist lock for parent etc handling!
2123          */
2124         write_lock_irq(&tasklist_lock);
2125
2126         /* CLONE_PARENT re-uses the old parent */
2127         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2128                 p->real_parent = current->real_parent;
2129                 p->parent_exec_id = current->parent_exec_id;
2130         } else {
2131                 p->real_parent = current;
2132                 p->parent_exec_id = current->self_exec_id;
2133         }
2134
2135         klp_copy_process(p);
2136
2137         spin_lock(&current->sighand->siglock);
2138
2139         /*
2140          * Copy seccomp details explicitly here, in case they were changed
2141          * before holding sighand lock.
2142          */
2143         copy_seccomp(p);
2144
2145         rseq_fork(p, clone_flags);
2146
2147         /* Don't start children in a dying pid namespace */
2148         if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2149                 retval = -ENOMEM;
2150                 goto bad_fork_cancel_cgroup;
2151         }
2152
2153         /* Let kill terminate clone/fork in the middle */
2154         if (fatal_signal_pending(current)) {
2155                 retval = -EINTR;
2156                 goto bad_fork_cancel_cgroup;
2157         }
2158
2159         /* past the last point of failure */
2160         if (pidfile)
2161                 fd_install(pidfd, pidfile);
2162
2163         init_task_pid_links(p);
2164         if (likely(p->pid)) {
2165                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2166
2167                 init_task_pid(p, PIDTYPE_PID, pid);
2168                 if (thread_group_leader(p)) {
2169                         init_task_pid(p, PIDTYPE_TGID, pid);
2170                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2171                         init_task_pid(p, PIDTYPE_SID, task_session(current));
2172
2173                         if (is_child_reaper(pid)) {
2174                                 ns_of_pid(pid)->child_reaper = p;
2175                                 p->signal->flags |= SIGNAL_UNKILLABLE;
2176                         }
2177                         p->signal->shared_pending.signal = delayed.signal;
2178                         p->signal->tty = tty_kref_get(current->signal->tty);
2179                         /*
2180                          * Inherit has_child_subreaper flag under the same
2181                          * tasklist_lock with adding child to the process tree
2182                          * for propagate_has_child_subreaper optimization.
2183                          */
2184                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2185                                                          p->real_parent->signal->is_child_subreaper;
2186                         list_add_tail(&p->sibling, &p->real_parent->children);
2187                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
2188                         attach_pid(p, PIDTYPE_TGID);
2189                         attach_pid(p, PIDTYPE_PGID);
2190                         attach_pid(p, PIDTYPE_SID);
2191                         __this_cpu_inc(process_counts);
2192                 } else {
2193                         current->signal->nr_threads++;
2194                         atomic_inc(&current->signal->live);
2195                         refcount_inc(&current->signal->sigcnt);
2196                         task_join_group_stop(p);
2197                         list_add_tail_rcu(&p->thread_group,
2198                                           &p->group_leader->thread_group);
2199                         list_add_tail_rcu(&p->thread_node,
2200                                           &p->signal->thread_head);
2201                 }
2202                 attach_pid(p, PIDTYPE_PID);
2203                 nr_threads++;
2204         }
2205         total_forks++;
2206         hlist_del_init(&delayed.node);
2207         spin_unlock(&current->sighand->siglock);
2208         syscall_tracepoint_update(p);
2209         write_unlock_irq(&tasklist_lock);
2210
2211         proc_fork_connector(p);
2212         cgroup_post_fork(p);
2213         cgroup_threadgroup_change_end(current);
2214         perf_event_fork(p);
2215
2216         trace_task_newtask(p, clone_flags);
2217         uprobe_copy_process(p, clone_flags);
2218
2219         return p;
2220
2221 bad_fork_cancel_cgroup:
2222         spin_unlock(&current->sighand->siglock);
2223         write_unlock_irq(&tasklist_lock);
2224         cgroup_cancel_fork(p);
2225 bad_fork_cgroup_threadgroup_change_end:
2226         cgroup_threadgroup_change_end(current);
2227 bad_fork_put_pidfd:
2228         if (clone_flags & CLONE_PIDFD) {
2229                 fput(pidfile);
2230                 put_unused_fd(pidfd);
2231         }
2232 bad_fork_free_pid:
2233         if (pid != &init_struct_pid)
2234                 free_pid(pid);
2235 bad_fork_cleanup_thread:
2236         exit_thread(p);
2237 bad_fork_cleanup_io:
2238         if (p->io_context)
2239                 exit_io_context(p);
2240 bad_fork_cleanup_namespaces:
2241         exit_task_namespaces(p);
2242 bad_fork_cleanup_mm:
2243         if (p->mm) {
2244                 mm_clear_owner(p->mm, p);
2245                 mmput(p->mm);
2246         }
2247 bad_fork_cleanup_signal:
2248         if (!(clone_flags & CLONE_THREAD))
2249                 free_signal_struct(p->signal);
2250 bad_fork_cleanup_sighand:
2251         __cleanup_sighand(p->sighand);
2252 bad_fork_cleanup_fs:
2253         exit_fs(p); /* blocking */
2254 bad_fork_cleanup_files:
2255         exit_files(p); /* blocking */
2256 bad_fork_cleanup_semundo:
2257         exit_sem(p);
2258 bad_fork_cleanup_security:
2259         security_task_free(p);
2260 bad_fork_cleanup_audit:
2261         audit_free(p);
2262 bad_fork_cleanup_perf:
2263         perf_event_free_task(p);
2264 bad_fork_cleanup_policy:
2265         lockdep_free_task(p);
2266 #ifdef CONFIG_NUMA
2267         mpol_put(p->mempolicy);
2268 bad_fork_cleanup_threadgroup_lock:
2269 #endif
2270         delayacct_tsk_free(p);
2271 bad_fork_cleanup_count:
2272         atomic_dec(&p->cred->user->processes);
2273         exit_creds(p);
2274 bad_fork_free:
2275         p->state = TASK_DEAD;
2276         put_task_stack(p);
2277         delayed_free_task(p);
2278 fork_out:
2279         spin_lock_irq(&current->sighand->siglock);
2280         hlist_del_init(&delayed.node);
2281         spin_unlock_irq(&current->sighand->siglock);
2282         return ERR_PTR(retval);
2283 }
2284
2285 static inline void init_idle_pids(struct task_struct *idle)
2286 {
2287         enum pid_type type;
2288
2289         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2290                 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2291                 init_task_pid(idle, type, &init_struct_pid);
2292         }
2293 }
2294
2295 struct task_struct *fork_idle(int cpu)
2296 {
2297         struct task_struct *task;
2298         task = copy_process(CLONE_VM, 0, 0, NULL, NULL, &init_struct_pid, 0, 0,
2299                             cpu_to_node(cpu));
2300         if (!IS_ERR(task)) {
2301                 init_idle_pids(task);
2302                 init_idle(task, cpu);
2303         }
2304
2305         return task;
2306 }
2307
2308 struct mm_struct *copy_init_mm(void)
2309 {
2310         return dup_mm(NULL, &init_mm);
2311 }
2312
2313 /*
2314  *  Ok, this is the main fork-routine.
2315  *
2316  * It copies the process, and if successful kick-starts
2317  * it and waits for it to finish using the VM if required.
2318  */
2319 long _do_fork(unsigned long clone_flags,
2320               unsigned long stack_start,
2321               unsigned long stack_size,
2322               int __user *parent_tidptr,
2323               int __user *child_tidptr,
2324               unsigned long tls)
2325 {
2326         struct completion vfork;
2327         struct pid *pid;
2328         struct task_struct *p;
2329         int trace = 0;
2330         long nr;
2331
2332         /*
2333          * Determine whether and which event to report to ptracer.  When
2334          * called from kernel_thread or CLONE_UNTRACED is explicitly
2335          * requested, no event is reported; otherwise, report if the event
2336          * for the type of forking is enabled.
2337          */
2338         if (!(clone_flags & CLONE_UNTRACED)) {
2339                 if (clone_flags & CLONE_VFORK)
2340                         trace = PTRACE_EVENT_VFORK;
2341                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
2342                         trace = PTRACE_EVENT_CLONE;
2343                 else
2344                         trace = PTRACE_EVENT_FORK;
2345
2346                 if (likely(!ptrace_event_enabled(current, trace)))
2347                         trace = 0;
2348         }
2349
2350         p = copy_process(clone_flags, stack_start, stack_size, parent_tidptr,
2351                          child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
2352         add_latent_entropy();
2353
2354         if (IS_ERR(p))
2355                 return PTR_ERR(p);
2356
2357         /*
2358          * Do this prior waking up the new thread - the thread pointer
2359          * might get invalid after that point, if the thread exits quickly.
2360          */
2361         trace_sched_process_fork(current, p);
2362
2363         pid = get_task_pid(p, PIDTYPE_PID);
2364         nr = pid_vnr(pid);
2365
2366         if (clone_flags & CLONE_PARENT_SETTID)
2367                 put_user(nr, parent_tidptr);
2368
2369         if (clone_flags & CLONE_VFORK) {
2370                 p->vfork_done = &vfork;
2371                 init_completion(&vfork);
2372                 get_task_struct(p);
2373         }
2374
2375         wake_up_new_task(p);
2376
2377         /* forking complete and child started to run, tell ptracer */
2378         if (unlikely(trace))
2379                 ptrace_event_pid(trace, pid);
2380
2381         if (clone_flags & CLONE_VFORK) {
2382                 if (!wait_for_vfork_done(p, &vfork))
2383                         ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2384         }
2385
2386         put_pid(pid);
2387         return nr;
2388 }
2389
2390 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2391 /* For compatibility with architectures that call do_fork directly rather than
2392  * using the syscall entry points below. */
2393 long do_fork(unsigned long clone_flags,
2394               unsigned long stack_start,
2395               unsigned long stack_size,
2396               int __user *parent_tidptr,
2397               int __user *child_tidptr)
2398 {
2399         return _do_fork(clone_flags, stack_start, stack_size,
2400                         parent_tidptr, child_tidptr, 0);
2401 }
2402 #endif
2403
2404 /*
2405  * Create a kernel thread.
2406  */
2407 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2408 {
2409         return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2410                 (unsigned long)arg, NULL, NULL, 0);
2411 }
2412
2413 #ifdef __ARCH_WANT_SYS_FORK
2414 SYSCALL_DEFINE0(fork)
2415 {
2416 #ifdef CONFIG_MMU
2417         return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2418 #else
2419         /* can not support in nommu mode */
2420         return -EINVAL;
2421 #endif
2422 }
2423 #endif
2424
2425 #ifdef __ARCH_WANT_SYS_VFORK
2426 SYSCALL_DEFINE0(vfork)
2427 {
2428         return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2429                         0, NULL, NULL, 0);
2430 }
2431 #endif
2432
2433 #ifdef __ARCH_WANT_SYS_CLONE
2434 #ifdef CONFIG_CLONE_BACKWARDS
2435 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2436                  int __user *, parent_tidptr,
2437                  unsigned long, tls,
2438                  int __user *, child_tidptr)
2439 #elif defined(CONFIG_CLONE_BACKWARDS2)
2440 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2441                  int __user *, parent_tidptr,
2442                  int __user *, child_tidptr,
2443                  unsigned long, tls)
2444 #elif defined(CONFIG_CLONE_BACKWARDS3)
2445 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2446                 int, stack_size,
2447                 int __user *, parent_tidptr,
2448                 int __user *, child_tidptr,
2449                 unsigned long, tls)
2450 #else
2451 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2452                  int __user *, parent_tidptr,
2453                  int __user *, child_tidptr,
2454                  unsigned long, tls)
2455 #endif
2456 {
2457         return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2458 }
2459 #endif
2460
2461 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2462 {
2463         struct task_struct *leader, *parent, *child;
2464         int res;
2465
2466         read_lock(&tasklist_lock);
2467         leader = top = top->group_leader;
2468 down:
2469         for_each_thread(leader, parent) {
2470                 list_for_each_entry(child, &parent->children, sibling) {
2471                         res = visitor(child, data);
2472                         if (res) {
2473                                 if (res < 0)
2474                                         goto out;
2475                                 leader = child;
2476                                 goto down;
2477                         }
2478 up:
2479                         ;
2480                 }
2481         }
2482
2483         if (leader != top) {
2484                 child = leader;
2485                 parent = child->real_parent;
2486                 leader = parent->group_leader;
2487                 goto up;
2488         }
2489 out:
2490         read_unlock(&tasklist_lock);
2491 }
2492
2493 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2494 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2495 #endif
2496
2497 static void sighand_ctor(void *data)
2498 {
2499         struct sighand_struct *sighand = data;
2500
2501         spin_lock_init(&sighand->siglock);
2502         init_waitqueue_head(&sighand->signalfd_wqh);
2503 }
2504
2505 void __init proc_caches_init(void)
2506 {
2507         unsigned int mm_size;
2508
2509         sighand_cachep = kmem_cache_create("sighand_cache",
2510                         sizeof(struct sighand_struct), 0,
2511                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2512                         SLAB_ACCOUNT, sighand_ctor);
2513         signal_cachep = kmem_cache_create("signal_cache",
2514                         sizeof(struct signal_struct), 0,
2515                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2516                         NULL);
2517         files_cachep = kmem_cache_create("files_cache",
2518                         sizeof(struct files_struct), 0,
2519                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2520                         NULL);
2521         fs_cachep = kmem_cache_create("fs_cache",
2522                         sizeof(struct fs_struct), 0,
2523                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2524                         NULL);
2525
2526         /*
2527          * The mm_cpumask is located at the end of mm_struct, and is
2528          * dynamically sized based on the maximum CPU number this system
2529          * can have, taking hotplug into account (nr_cpu_ids).
2530          */
2531         mm_size = sizeof(struct mm_struct) + cpumask_size();
2532
2533         mm_cachep = kmem_cache_create_usercopy("mm_struct",
2534                         mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2535                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2536                         offsetof(struct mm_struct, saved_auxv),
2537                         sizeof_field(struct mm_struct, saved_auxv),
2538                         NULL);
2539         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2540         mmap_init();
2541         nsproxy_cache_init();
2542 }
2543
2544 /*
2545  * Check constraints on flags passed to the unshare system call.
2546  */
2547 static int check_unshare_flags(unsigned long unshare_flags)
2548 {
2549         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2550                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2551                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2552                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2553                 return -EINVAL;
2554         /*
2555          * Not implemented, but pretend it works if there is nothing
2556          * to unshare.  Note that unsharing the address space or the
2557          * signal handlers also need to unshare the signal queues (aka
2558          * CLONE_THREAD).
2559          */
2560         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2561                 if (!thread_group_empty(current))
2562                         return -EINVAL;
2563         }
2564         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2565                 if (refcount_read(&current->sighand->count) > 1)
2566                         return -EINVAL;
2567         }
2568         if (unshare_flags & CLONE_VM) {
2569                 if (!current_is_single_threaded())
2570                         return -EINVAL;
2571         }
2572
2573         return 0;
2574 }
2575
2576 /*
2577  * Unshare the filesystem structure if it is being shared
2578  */
2579 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2580 {
2581         struct fs_struct *fs = current->fs;
2582
2583         if (!(unshare_flags & CLONE_FS) || !fs)
2584                 return 0;
2585
2586         /* don't need lock here; in the worst case we'll do useless copy */
2587         if (fs->users == 1)
2588                 return 0;
2589
2590         *new_fsp = copy_fs_struct(fs);
2591         if (!*new_fsp)
2592                 return -ENOMEM;
2593
2594         return 0;
2595 }
2596
2597 /*
2598  * Unshare file descriptor table if it is being shared
2599  */
2600 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2601 {
2602         struct files_struct *fd = current->files;
2603         int error = 0;
2604
2605         if ((unshare_flags & CLONE_FILES) &&
2606             (fd && atomic_read(&fd->count) > 1)) {
2607                 *new_fdp = dup_fd(fd, &error);
2608                 if (!*new_fdp)
2609                         return error;
2610         }
2611
2612         return 0;
2613 }
2614
2615 /*
2616  * unshare allows a process to 'unshare' part of the process
2617  * context which was originally shared using clone.  copy_*
2618  * functions used by do_fork() cannot be used here directly
2619  * because they modify an inactive task_struct that is being
2620  * constructed. Here we are modifying the current, active,
2621  * task_struct.
2622  */
2623 int ksys_unshare(unsigned long unshare_flags)
2624 {
2625         struct fs_struct *fs, *new_fs = NULL;
2626         struct files_struct *fd, *new_fd = NULL;
2627         struct cred *new_cred = NULL;
2628         struct nsproxy *new_nsproxy = NULL;
2629         int do_sysvsem = 0;
2630         int err;
2631
2632         /*
2633          * If unsharing a user namespace must also unshare the thread group
2634          * and unshare the filesystem root and working directories.
2635          */
2636         if (unshare_flags & CLONE_NEWUSER)
2637                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2638         /*
2639          * If unsharing vm, must also unshare signal handlers.
2640          */
2641         if (unshare_flags & CLONE_VM)
2642                 unshare_flags |= CLONE_SIGHAND;
2643         /*
2644          * If unsharing a signal handlers, must also unshare the signal queues.
2645          */
2646         if (unshare_flags & CLONE_SIGHAND)
2647                 unshare_flags |= CLONE_THREAD;
2648         /*
2649          * If unsharing namespace, must also unshare filesystem information.
2650          */
2651         if (unshare_flags & CLONE_NEWNS)
2652                 unshare_flags |= CLONE_FS;
2653
2654         err = check_unshare_flags(unshare_flags);
2655         if (err)
2656                 goto bad_unshare_out;
2657         /*
2658          * CLONE_NEWIPC must also detach from the undolist: after switching
2659          * to a new ipc namespace, the semaphore arrays from the old
2660          * namespace are unreachable.
2661          */
2662         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2663                 do_sysvsem = 1;
2664         err = unshare_fs(unshare_flags, &new_fs);
2665         if (err)
2666                 goto bad_unshare_out;
2667         err = unshare_fd(unshare_flags, &new_fd);
2668         if (err)
2669                 goto bad_unshare_cleanup_fs;
2670         err = unshare_userns(unshare_flags, &new_cred);
2671         if (err)
2672                 goto bad_unshare_cleanup_fd;
2673         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2674                                          new_cred, new_fs);
2675         if (err)
2676                 goto bad_unshare_cleanup_cred;
2677
2678         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2679                 if (do_sysvsem) {
2680                         /*
2681                          * CLONE_SYSVSEM is equivalent to sys_exit().
2682                          */
2683                         exit_sem(current);
2684                 }
2685                 if (unshare_flags & CLONE_NEWIPC) {
2686                         /* Orphan segments in old ns (see sem above). */
2687                         exit_shm(current);
2688                         shm_init_task(current);
2689                 }
2690
2691                 if (new_nsproxy)
2692                         switch_task_namespaces(current, new_nsproxy);
2693
2694                 task_lock(current);
2695
2696                 if (new_fs) {
2697                         fs = current->fs;
2698                         spin_lock(&fs->lock);
2699                         current->fs = new_fs;
2700                         if (--fs->users)
2701                                 new_fs = NULL;
2702                         else
2703                                 new_fs = fs;
2704                         spin_unlock(&fs->lock);
2705                 }
2706
2707                 if (new_fd) {
2708                         fd = current->files;
2709                         current->files = new_fd;
2710                         new_fd = fd;
2711                 }
2712
2713                 task_unlock(current);
2714
2715                 if (new_cred) {
2716                         /* Install the new user namespace */
2717                         commit_creds(new_cred);
2718                         new_cred = NULL;
2719                 }
2720         }
2721
2722         perf_event_namespaces(current);
2723
2724 bad_unshare_cleanup_cred:
2725         if (new_cred)
2726                 put_cred(new_cred);
2727 bad_unshare_cleanup_fd:
2728         if (new_fd)
2729                 put_files_struct(new_fd);
2730
2731 bad_unshare_cleanup_fs:
2732         if (new_fs)
2733                 free_fs_struct(new_fs);
2734
2735 bad_unshare_out:
2736         return err;
2737 }
2738
2739 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2740 {
2741         return ksys_unshare(unshare_flags);
2742 }
2743
2744 /*
2745  *      Helper to unshare the files of the current task.
2746  *      We don't want to expose copy_files internals to
2747  *      the exec layer of the kernel.
2748  */
2749
2750 int unshare_files(struct files_struct **displaced)
2751 {
2752         struct task_struct *task = current;
2753         struct files_struct *copy = NULL;
2754         int error;
2755
2756         error = unshare_fd(CLONE_FILES, &copy);
2757         if (error || !copy) {
2758                 *displaced = NULL;
2759                 return error;
2760         }
2761         *displaced = task->files;
2762         task_lock(task);
2763         task->files = copy;
2764         task_unlock(task);
2765         return 0;
2766 }
2767
2768 int sysctl_max_threads(struct ctl_table *table, int write,
2769                        void __user *buffer, size_t *lenp, loff_t *ppos)
2770 {
2771         struct ctl_table t;
2772         int ret;
2773         int threads = max_threads;
2774         int min = MIN_THREADS;
2775         int max = MAX_THREADS;
2776
2777         t = *table;
2778         t.data = &threads;
2779         t.extra1 = &min;
2780         t.extra2 = &max;
2781
2782         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2783         if (ret || !write)
2784                 return ret;
2785
2786         set_max_threads(threads);
2787
2788         return 0;
2789 }