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 NAMED_ARRAY_INDEX(x)    [x] = __stringify(x)
 129
 130 static const char * const resident_page_types[] = {
 131         NAMED_ARRAY_INDEX(MM_FILEPAGES),
 132         NAMED_ARRAY_INDEX(MM_ANONPAGES),
 133         NAMED_ARRAY_INDEX(MM_SWAPENTS),
 134         NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
 135 };
 136
 137 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 138
 139 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 140
 141 #ifdef CONFIG_PROVE_RCU
 142 int lockdep_tasklist_lock_is_held(void)
 143 {
 144         return lockdep_is_held(&tasklist_lock);
 145 }
 146 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 147 #endif /* #ifdef CONFIG_PROVE_RCU */
 148
 149 int nr_processes(void)
 150 {
 151         int cpu;
 152         int total = 0;
 153
 154         for_each_possible_cpu(cpu)
 155                 total += per_cpu(process_counts, cpu);
 156
 157         return total;
 158 }
 159
 160 void __weak arch_release_task_struct(struct task_struct *tsk)
 161 {
 162 }
 163
 164 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 165 static struct kmem_cache *task_struct_cachep;
 166
 167 static inline struct task_struct *alloc_task_struct_node(int node)
 168 {
 169         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 170 }
 171
 172 static inline void free_task_struct(struct task_struct *tsk)
 173 {
 174         kmem_cache_free(task_struct_cachep, tsk);
 175 }
 176 #endif
 177
 178 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
 179
 180 /*
 181  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 182  * kmemcache based allocator.
 183  */
 184 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
 185
 186 #ifdef CONFIG_VMAP_STACK
 187 /*
 188  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
 189  * flush.  Try to minimize the number of calls by caching stacks.
 190  */
 191 #define NR_CACHED_STACKS 2
 192 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
 193
 194 static int free_vm_stack_cache(unsigned int cpu)
 195 {
 196         struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
 197         int i;
 198
 199         for (i = 0; i < NR_CACHED_STACKS; i++) {
 200                 struct vm_struct *vm_stack = cached_vm_stacks[i];
 201
 202                 if (!vm_stack)
 203                         continue;
 204
 205                 vfree(vm_stack->addr);
 206                 cached_vm_stacks[i] = NULL;
 207         }
 208
 209         return 0;
 210 }
 211 #endif
 212
 213 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
 214 {
 215 #ifdef CONFIG_VMAP_STACK
 216         void *stack;
 217         int i;
 218
 219         for (i = 0; i < NR_CACHED_STACKS; i++) {
 220                 struct vm_struct *s;
 221
 222                 s = this_cpu_xchg(cached_stacks[i], NULL);
 223
 224                 if (!s)
 225                         continue;
 226
 227                 /* Clear stale pointers from reused stack. */
 228                 memset(s->addr, 0, THREAD_SIZE);
 229
 230                 tsk->stack_vm_area = s;
 231                 tsk->stack = s->addr;
 232                 return s->addr;
 233         }
 234
 235         /*
 236          * Allocated stacks are cached and later reused by new threads,
 237          * so memcg accounting is performed manually on assigning/releasing
 238          * stacks to tasks. Drop __GFP_ACCOUNT.
 239          */
 240         stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
 241                                      VMALLOC_START, VMALLOC_END,
 242                                      THREADINFO_GFP & ~__GFP_ACCOUNT,
 243                                      PAGE_KERNEL,
 244                                      0, node, __builtin_return_address(0));
 245
 246         /*
 247          * We can't call find_vm_area() in interrupt context, and
 248          * free_thread_stack() can be called in interrupt context,
 249          * so cache the vm_struct.
 250          */
 251         if (stack) {
 252                 tsk->stack_vm_area = find_vm_area(stack);
 253                 tsk->stack = stack;
 254         }
 255         return stack;
 256 #else
 257         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
 258                                              THREAD_SIZE_ORDER);
 259
 260         if (likely(page)) {
 261                 tsk->stack = page_address(page);
 262                 return tsk->stack;
 263         }
 264         return NULL;
 265 #endif
 266 }
 267
 268 static inline void free_thread_stack(struct task_struct *tsk)
 269 {
 270 #ifdef CONFIG_VMAP_STACK
 271         struct vm_struct *vm = task_stack_vm_area(tsk);
 272
 273         if (vm) {
 274                 int i;
 275
 276                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 277                         mod_memcg_page_state(vm->pages[i],
 278                                              MEMCG_KERNEL_STACK_KB,
 279                                              -(int)(PAGE_SIZE / 1024));
 280
 281                         memcg_kmem_uncharge(vm->pages[i], 0);
 282                 }
 283
 284                 for (i = 0; i < NR_CACHED_STACKS; i++) {
 285                         if (this_cpu_cmpxchg(cached_stacks[i],
 286                                         NULL, tsk->stack_vm_area) != NULL)
 287                                 continue;
 288
 289                         return;
 290                 }
 291
 292                 vfree_atomic(tsk->stack);
 293                 return;
 294         }
 295 #endif
 296
 297         __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
 298 }
 299 # else
 300 static struct kmem_cache *thread_stack_cache;
 301
 302 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
 303                                                   int node)
 304 {
 305         unsigned long *stack;
 306         stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
 307         tsk->stack = stack;
 308         return stack;
 309 }
 310
 311 static void free_thread_stack(struct task_struct *tsk)
 312 {
 313         kmem_cache_free(thread_stack_cache, tsk->stack);
 314 }
 315
 316 void thread_stack_cache_init(void)
 317 {
 318         thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
 319                                         THREAD_SIZE, THREAD_SIZE, 0, 0,
 320                                         THREAD_SIZE, NULL);
 321         BUG_ON(thread_stack_cache == NULL);
 322 }
 323 # endif
 324 #endif
 325
 326 /* SLAB cache for signal_struct structures (tsk->signal) */
 327 static struct kmem_cache *signal_cachep;
 328
 329 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 330 struct kmem_cache *sighand_cachep;
 331
 332 /* SLAB cache for files_struct structures (tsk->files) */
 333 struct kmem_cache *files_cachep;
 334
 335 /* SLAB cache for fs_struct structures (tsk->fs) */
 336 struct kmem_cache *fs_cachep;
 337
 338 /* SLAB cache for vm_area_struct structures */
 339 static struct kmem_cache *vm_area_cachep;
 340
 341 /* SLAB cache for mm_struct structures (tsk->mm) */
 342 static struct kmem_cache *mm_cachep;
 343
 344 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
 345 {
 346         struct vm_area_struct *vma;
 347
 348         vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 349         if (vma)
 350                 vma_init(vma, mm);
 351         return vma;
 352 }
 353
 354 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
 355 {
 356         struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 357
 358         if (new) {
 359                 *new = *orig;
 360                 INIT_LIST_HEAD(&new->anon_vma_chain);
 361         }
 362         return new;
 363 }
 364
 365 void vm_area_free(struct vm_area_struct *vma)
 366 {
 367         kmem_cache_free(vm_area_cachep, vma);
 368 }
 369
 370 static void account_kernel_stack(struct task_struct *tsk, int account)
 371 {
 372         void *stack = task_stack_page(tsk);
 373         struct vm_struct *vm = task_stack_vm_area(tsk);
 374
 375         BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
 376
 377         if (vm) {
 378                 int i;
 379
 380                 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
 381
 382                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 383                         mod_zone_page_state(page_zone(vm->pages[i]),
 384                                             NR_KERNEL_STACK_KB,
 385                                             PAGE_SIZE / 1024 * account);
 386                 }
 387         } else {
 388                 /*
 389                  * All stack pages are in the same zone and belong to the
 390                  * same memcg.
 391                  */
 392                 struct page *first_page = virt_to_page(stack);
 393
 394                 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
 395                                     THREAD_SIZE / 1024 * account);
 396
 397                 mod_memcg_page_state(first_page, MEMCG_KERNEL_STACK_KB,
 398                                      account * (THREAD_SIZE / 1024));
 399         }
 400 }
 401
 402 static int memcg_charge_kernel_stack(struct task_struct *tsk)
 403 {
 404 #ifdef CONFIG_VMAP_STACK
 405         struct vm_struct *vm = task_stack_vm_area(tsk);
 406         int ret;
 407
 408         if (vm) {
 409                 int i;
 410
 411                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 412                         /*
 413                          * If memcg_kmem_charge() fails, page->mem_cgroup
 414                          * pointer is NULL, and both memcg_kmem_uncharge()
 415                          * and mod_memcg_page_state() in free_thread_stack()
 416                          * will ignore this page. So it's safe.
 417                          */
 418                         ret = memcg_kmem_charge(vm->pages[i], GFP_KERNEL, 0);
 419                         if (ret)
 420                                 return ret;
 421
 422                         mod_memcg_page_state(vm->pages[i],
 423                                              MEMCG_KERNEL_STACK_KB,
 424                                              PAGE_SIZE / 1024);
 425                 }
 426         }
 427 #endif
 428         return 0;
 429 }
 430
 431 static void release_task_stack(struct task_struct *tsk)
 432 {
 433         if (WARN_ON(tsk->state != TASK_DEAD))
 434                 return;  /* Better to leak the stack than to free prematurely */
 435
 436         account_kernel_stack(tsk, -1);
 437         free_thread_stack(tsk);
 438         tsk->stack = NULL;
 439 #ifdef CONFIG_VMAP_STACK
 440         tsk->stack_vm_area = NULL;
 441 #endif
 442 }
 443
 444 #ifdef CONFIG_THREAD_INFO_IN_TASK
 445 void put_task_stack(struct task_struct *tsk)
 446 {
 447         if (refcount_dec_and_test(&tsk->stack_refcount))
 448                 release_task_stack(tsk);
 449 }
 450 #endif
 451
 452 void free_task(struct task_struct *tsk)
 453 {
 454 #ifndef CONFIG_THREAD_INFO_IN_TASK
 455         /*
 456          * The task is finally done with both the stack and thread_info,
 457          * so free both.
 458          */
 459         release_task_stack(tsk);
 460 #else
 461         /*
 462          * If the task had a separate stack allocation, it should be gone
 463          * by now.
 464          */
 465         WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
 466 #endif
 467         rt_mutex_debug_task_free(tsk);
 468         ftrace_graph_exit_task(tsk);
 469         put_seccomp_filter(tsk);
 470         arch_release_task_struct(tsk);
 471         if (tsk->flags & PF_KTHREAD)
 472                 free_kthread_struct(tsk);
 473         free_task_struct(tsk);
 474 }
 475 EXPORT_SYMBOL(free_task);
 476
 477 #ifdef CONFIG_MMU
 478 static __latent_entropy int dup_mmap(struct mm_struct *mm,
 479                                         struct mm_struct *oldmm)
 480 {
 481         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 482         struct rb_node **rb_link, *rb_parent;
 483         int retval;
 484         unsigned long charge;
 485         LIST_HEAD(uf);
 486
 487         uprobe_start_dup_mmap();
 488         if (down_write_killable(&oldmm->mmap_sem)) {
 489                 retval = -EINTR;
 490                 goto fail_uprobe_end;
 491         }
 492         flush_cache_dup_mm(oldmm);
 493         uprobe_dup_mmap(oldmm, mm);
 494         /*
 495          * Not linked in yet - no deadlock potential:
 496          */
 497         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 498
 499         /* No ordering required: file already has been exposed. */
 500         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 501
 502         mm->total_vm = oldmm->total_vm;
 503         mm->data_vm = oldmm->data_vm;
 504         mm->exec_vm = oldmm->exec_vm;
 505         mm->stack_vm = oldmm->stack_vm;
 506
 507         rb_link = &mm->mm_rb.rb_node;
 508         rb_parent = NULL;
 509         pprev = &mm->mmap;
 510         retval = ksm_fork(mm, oldmm);
 511         if (retval)
 512                 goto out;
 513         retval = khugepaged_fork(mm, oldmm);
 514         if (retval)
 515                 goto out;
 516
 517         prev = NULL;
 518         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 519                 struct file *file;
 520
 521                 if (mpnt->vm_flags & VM_DONTCOPY) {
 522                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 523                         continue;
 524                 }
 525                 charge = 0;
 526                 /*
 527                  * Don't duplicate many vmas if we've been oom-killed (for
 528                  * example)
 529                  */
 530                 if (fatal_signal_pending(current)) {
 531                         retval = -EINTR;
 532                         goto out;
 533                 }
 534                 if (mpnt->vm_flags & VM_ACCOUNT) {
 535                         unsigned long len = vma_pages(mpnt);
 536
 537                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 538                                 goto fail_nomem;
 539                         charge = len;
 540                 }
 541                 tmp = vm_area_dup(mpnt);
 542                 if (!tmp)
 543                         goto fail_nomem;
 544                 retval = vma_dup_policy(mpnt, tmp);
 545                 if (retval)
 546                         goto fail_nomem_policy;
 547                 tmp->vm_mm = mm;
 548                 retval = dup_userfaultfd(tmp, &uf);
 549                 if (retval)
 550                         goto fail_nomem_anon_vma_fork;
 551                 if (tmp->vm_flags & VM_WIPEONFORK) {
 552                         /* VM_WIPEONFORK gets a clean slate in the child. */
 553                         tmp->anon_vma = NULL;
 554                         if (anon_vma_prepare(tmp))
 555                                 goto fail_nomem_anon_vma_fork;
 556                 } else if (anon_vma_fork(tmp, mpnt))
 557                         goto fail_nomem_anon_vma_fork;
 558                 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
 559                 tmp->vm_next = tmp->vm_prev = NULL;
 560                 file = tmp->vm_file;
 561                 if (file) {
 562                         struct inode *inode = file_inode(file);
 563                         struct address_space *mapping = file->f_mapping;
 564
 565                         get_file(file);
 566                         if (tmp->vm_flags & VM_DENYWRITE)
 567                                 atomic_dec(&inode->i_writecount);
 568                         i_mmap_lock_write(mapping);
 569                         if (tmp->vm_flags & VM_SHARED)
 570                                 atomic_inc(&mapping->i_mmap_writable);
 571                         flush_dcache_mmap_lock(mapping);
 572                         /* insert tmp into the share list, just after mpnt */
 573                         vma_interval_tree_insert_after(tmp, mpnt,
 574                                         &mapping->i_mmap);
 575                         flush_dcache_mmap_unlock(mapping);
 576                         i_mmap_unlock_write(mapping);
 577                 }
 578
 579                 /*
 580                  * Clear hugetlb-related page reserves for children. This only
 581                  * affects MAP_PRIVATE mappings. Faults generated by the child
 582                  * are not guaranteed to succeed, even if read-only
 583                  */
 584                 if (is_vm_hugetlb_page(tmp))
 585                         reset_vma_resv_huge_pages(tmp);
 586
 587                 /*
 588                  * Link in the new vma and copy the page table entries.
 589                  */
 590                 *pprev = tmp;
 591                 pprev = &tmp->vm_next;
 592                 tmp->vm_prev = prev;
 593                 prev = tmp;
 594
 595                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 596                 rb_link = &tmp->vm_rb.rb_right;
 597                 rb_parent = &tmp->vm_rb;
 598
 599                 mm->map_count++;
 600                 if (!(tmp->vm_flags & VM_WIPEONFORK))
 601                         retval = copy_page_range(mm, oldmm, mpnt);
 602
 603                 if (tmp->vm_ops && tmp->vm_ops->open)
 604                         tmp->vm_ops->open(tmp);
 605
 606                 if (retval)
 607                         goto out;
 608         }
 609         /* a new mm has just been created */
 610         retval = arch_dup_mmap(oldmm, mm);
 611 out:
 612         up_write(&mm->mmap_sem);
 613         flush_tlb_mm(oldmm);
 614         up_write(&oldmm->mmap_sem);
 615         dup_userfaultfd_complete(&uf);
 616 fail_uprobe_end:
 617         uprobe_end_dup_mmap();
 618         return retval;
 619 fail_nomem_anon_vma_fork:
 620         mpol_put(vma_policy(tmp));
 621 fail_nomem_policy:
 622         vm_area_free(tmp);
 623 fail_nomem:
 624         retval = -ENOMEM;
 625         vm_unacct_memory(charge);
 626         goto out;
 627 }
 628
 629 static inline int mm_alloc_pgd(struct mm_struct *mm)
 630 {
 631         mm->pgd = pgd_alloc(mm);
 632         if (unlikely(!mm->pgd))
 633                 return -ENOMEM;
 634         return 0;
 635 }
 636
 637 static inline void mm_free_pgd(struct mm_struct *mm)
 638 {
 639         pgd_free(mm, mm->pgd);
 640 }
 641 #else
 642 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 643 {
 644         down_write(&oldmm->mmap_sem);
 645         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 646         up_write(&oldmm->mmap_sem);
 647         return 0;
 648 }
 649 #define mm_alloc_pgd(mm)        (0)
 650 #define mm_free_pgd(mm)
 651 #endif /* CONFIG_MMU */
 652
 653 static void check_mm(struct mm_struct *mm)
 654 {
 655         int i;
 656
 657         BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
 658                          "Please make sure 'struct resident_page_types[]' is updated as well");
 659
 660         for (i = 0; i < NR_MM_COUNTERS; i++) {
 661                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 662
 663                 if (unlikely(x))
 664                         pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
 665                                  mm, resident_page_types[i], x);
 666         }
 667
 668         if (mm_pgtables_bytes(mm))
 669                 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
 670                                 mm_pgtables_bytes(mm));
 671
 672 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 673         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 674 #endif
 675 }
 676
 677 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 678 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 679
 680 /*
 681  * Called when the last reference to the mm
 682  * is dropped: either by a lazy thread or by
 683  * mmput. Free the page directory and the mm.
 684  */
 685 void __mmdrop(struct mm_struct *mm)
 686 {
 687         BUG_ON(mm == &init_mm);
 688         WARN_ON_ONCE(mm == current->mm);
 689         WARN_ON_ONCE(mm == current->active_mm);
 690         mm_free_pgd(mm);
 691         destroy_context(mm);
 692         mmu_notifier_mm_destroy(mm);
 693         check_mm(mm);
 694         put_user_ns(mm->user_ns);
 695         free_mm(mm);
 696 }
 697 EXPORT_SYMBOL_GPL(__mmdrop);
 698
 699 static void mmdrop_async_fn(struct work_struct *work)
 700 {
 701         struct mm_struct *mm;
 702
 703         mm = container_of(work, struct mm_struct, async_put_work);
 704         __mmdrop(mm);
 705 }
 706
 707 static void mmdrop_async(struct mm_struct *mm)
 708 {
 709         if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
 710                 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
 711                 schedule_work(&mm->async_put_work);
 712         }
 713 }
 714
 715 static inline void free_signal_struct(struct signal_struct *sig)
 716 {
 717         taskstats_tgid_free(sig);
 718         sched_autogroup_exit(sig);
 719         /*
 720          * __mmdrop is not safe to call from softirq context on x86 due to
 721          * pgd_dtor so postpone it to the async context
 722          */
 723         if (sig->oom_mm)
 724                 mmdrop_async(sig->oom_mm);
 725         kmem_cache_free(signal_cachep, sig);
 726 }
 727
 728 static inline void put_signal_struct(struct signal_struct *sig)
 729 {
 730         if (refcount_dec_and_test(&sig->sigcnt))
 731                 free_signal_struct(sig);
 732 }
 733
 734 void __put_task_struct(struct task_struct *tsk)
 735 {
 736         WARN_ON(!tsk->exit_state);
 737         WARN_ON(refcount_read(&tsk->usage));
 738         WARN_ON(tsk == current);
 739
 740         cgroup_free(tsk);
 741         task_numa_free(tsk, true);
 742         security_task_free(tsk);
 743         exit_creds(tsk);
 744         delayacct_tsk_free(tsk);
 745         put_signal_struct(tsk->signal);
 746
 747         if (!profile_handoff_task(tsk))
 748                 free_task(tsk);
 749 }
 750 EXPORT_SYMBOL_GPL(__put_task_struct);
 751
 752 void __init __weak arch_task_cache_init(void) { }
 753
 754 /*
 755  * set_max_threads
 756  */
 757 static void set_max_threads(unsigned int max_threads_suggested)
 758 {
 759         u64 threads;
 760         unsigned long nr_pages = totalram_pages();
 761
 762         /*
 763          * The number of threads shall be limited such that the thread
 764          * structures may only consume a small part of the available memory.
 765          */
 766         if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
 767                 threads = MAX_THREADS;
 768         else
 769                 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
 770                                     (u64) THREAD_SIZE * 8UL);
 771
 772         if (threads > max_threads_suggested)
 773                 threads = max_threads_suggested;
 774
 775         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 776 }
 777
 778 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 779 /* Initialized by the architecture: */
 780 int arch_task_struct_size __read_mostly;
 781 #endif
 782
 783 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 784 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
 785 {
 786         /* Fetch thread_struct whitelist for the architecture. */
 787         arch_thread_struct_whitelist(offset, size);
 788
 789         /*
 790          * Handle zero-sized whitelist or empty thread_struct, otherwise
 791          * adjust offset to position of thread_struct in task_struct.
 792          */
 793         if (unlikely(*size == 0))
 794                 *offset = 0;
 795         else
 796                 *offset += offsetof(struct task_struct, thread);
 797 }
 798 #endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
 799
 800 void __init fork_init(void)
 801 {
 802         int i;
 803 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 804 #ifndef ARCH_MIN_TASKALIGN
 805 #define ARCH_MIN_TASKALIGN      0
 806 #endif
 807         int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
 808         unsigned long useroffset, usersize;
 809
 810         /* create a slab on which task_structs can be allocated */
 811         task_struct_whitelist(&useroffset, &usersize);
 812         task_struct_cachep = kmem_cache_create_usercopy("task_struct",
 813                         arch_task_struct_size, align,
 814                         SLAB_PANIC|SLAB_ACCOUNT,
 815                         useroffset, usersize, NULL);
 816 #endif
 817
 818         /* do the arch specific task caches init */
 819         arch_task_cache_init();
 820
 821         set_max_threads(MAX_THREADS);
 822
 823         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 824         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 825         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 826                 init_task.signal->rlim[RLIMIT_NPROC];
 827
 828         for (i = 0; i < UCOUNT_COUNTS; i++) {
 829                 init_user_ns.ucount_max[i] = max_threads/2;
 830         }
 831
 832 #ifdef CONFIG_VMAP_STACK
 833         cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
 834                           NULL, free_vm_stack_cache);
 835 #endif
 836
 837         lockdep_init_task(&init_task);
 838         uprobes_init();
 839 }
 840
 841 int __weak arch_dup_task_struct(struct task_struct *dst,
 842                                                struct task_struct *src)
 843 {
 844         *dst = *src;
 845         return 0;
 846 }
 847
 848 void set_task_stack_end_magic(struct task_struct *tsk)
 849 {
 850         unsigned long *stackend;
 851
 852         stackend = end_of_stack(tsk);
 853         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 854 }
 855
 856 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 857 {
 858         struct task_struct *tsk;
 859         unsigned long *stack;
 860         struct vm_struct *stack_vm_area __maybe_unused;
 861         int err;
 862
 863         if (node == NUMA_NO_NODE)
 864                 node = tsk_fork_get_node(orig);
 865         tsk = alloc_task_struct_node(node);
 866         if (!tsk)
 867                 return NULL;
 868
 869         stack = alloc_thread_stack_node(tsk, node);
 870         if (!stack)
 871                 goto free_tsk;
 872
 873         if (memcg_charge_kernel_stack(tsk))
 874                 goto free_stack;
 875
 876         stack_vm_area = task_stack_vm_area(tsk);
 877
 878         err = arch_dup_task_struct(tsk, orig);
 879
 880         /*
 881          * arch_dup_task_struct() clobbers the stack-related fields.  Make
 882          * sure they're properly initialized before using any stack-related
 883          * functions again.
 884          */
 885         tsk->stack = stack;
 886 #ifdef CONFIG_VMAP_STACK
 887         tsk->stack_vm_area = stack_vm_area;
 888 #endif
 889 #ifdef CONFIG_THREAD_INFO_IN_TASK
 890         refcount_set(&tsk->stack_refcount, 1);
 891 #endif
 892
 893         if (err)
 894                 goto free_stack;
 895
 896 #ifdef CONFIG_SECCOMP
 897         /*
 898          * We must handle setting up seccomp filters once we're under
 899          * the sighand lock in case orig has changed between now and
 900          * then. Until then, filter must be NULL to avoid messing up
 901          * the usage counts on the error path calling free_task.
 902          */
 903         tsk->seccomp.filter = NULL;
 904 #endif
 905
 906         setup_thread_stack(tsk, orig);
 907         clear_user_return_notifier(tsk);
 908         clear_tsk_need_resched(tsk);
 909         set_task_stack_end_magic(tsk);
 910
 911 #ifdef CONFIG_STACKPROTECTOR
 912         tsk->stack_canary = get_random_canary();
 913 #endif
 914         if (orig->cpus_ptr == &orig->cpus_mask)
 915                 tsk->cpus_ptr = &tsk->cpus_mask;
 916
 917         /*
 918          * One for the user space visible state that goes away when reaped.
 919          * One for the scheduler.
 920          */
 921         refcount_set(&tsk->rcu_users, 2);
 922         /* One for the rcu users */
 923         refcount_set(&tsk->usage, 1);
 924 #ifdef CONFIG_BLK_DEV_IO_TRACE
 925         tsk->btrace_seq = 0;
 926 #endif
 927         tsk->splice_pipe = NULL;
 928         tsk->task_frag.page = NULL;
 929         tsk->wake_q.next = NULL;
 930
 931         account_kernel_stack(tsk, 1);
 932
 933         kcov_task_init(tsk);
 934
 935 #ifdef CONFIG_FAULT_INJECTION
 936         tsk->fail_nth = 0;
 937 #endif
 938
 939 #ifdef CONFIG_BLK_CGROUP
 940         tsk->throttle_queue = NULL;
 941         tsk->use_memdelay = 0;
 942 #endif
 943
 944 #ifdef CONFIG_MEMCG
 945         tsk->active_memcg = NULL;
 946 #endif
 947         return tsk;
 948
 949 free_stack:
 950         free_thread_stack(tsk);
 951 free_tsk:
 952         free_task_struct(tsk);
 953         return NULL;
 954 }
 955
 956 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 957
 958 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 959
 960 static int __init coredump_filter_setup(char *s)
 961 {
 962         default_dump_filter =
 963                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 964                 MMF_DUMP_FILTER_MASK;
 965         return 1;
 966 }
 967
 968 __setup("coredump_filter=", coredump_filter_setup);
 969
 970 #include <linux/init_task.h>
 971
 972 static void mm_init_aio(struct mm_struct *mm)
 973 {
 974 #ifdef CONFIG_AIO
 975         spin_lock_init(&mm->ioctx_lock);
 976         mm->ioctx_table = NULL;
 977 #endif
 978 }
 979
 980 static __always_inline void mm_clear_owner(struct mm_struct *mm,
 981                                            struct task_struct *p)
 982 {
 983 #ifdef CONFIG_MEMCG
 984         if (mm->owner == p)
 985                 WRITE_ONCE(mm->owner, NULL);
 986 #endif
 987 }
 988
 989 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 990 {
 991 #ifdef CONFIG_MEMCG
 992         mm->owner = p;
 993 #endif
 994 }
 995
 996 static void mm_init_uprobes_state(struct mm_struct *mm)
 997 {
 998 #ifdef CONFIG_UPROBES
 999         mm->uprobes_state.xol_area = NULL;
1000 #endif
1001 }
1002
1003 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1004         struct user_namespace *user_ns)
1005 {
1006         mm->mmap = NULL;
1007         mm->mm_rb = RB_ROOT;
1008         mm->vmacache_seqnum = 0;
1009         atomic_set(&mm->mm_users, 1);
1010         atomic_set(&mm->mm_count, 1);
1011         init_rwsem(&mm->mmap_sem);
1012         INIT_LIST_HEAD(&mm->mmlist);
1013         mm->core_state = NULL;
1014         mm_pgtables_bytes_init(mm);
1015         mm->map_count = 0;
1016         mm->locked_vm = 0;
1017         atomic64_set(&mm->pinned_vm, 0);
1018         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1019         spin_lock_init(&mm->page_table_lock);
1020         spin_lock_init(&mm->arg_lock);
1021         mm_init_cpumask(mm);
1022         mm_init_aio(mm);
1023         mm_init_owner(mm, p);
1024         RCU_INIT_POINTER(mm->exe_file, NULL);
1025         mmu_notifier_mm_init(mm);
1026         init_tlb_flush_pending(mm);
1027 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1028         mm->pmd_huge_pte = NULL;
1029 #endif
1030         mm_init_uprobes_state(mm);
1031
1032         if (current->mm) {
1033                 mm->flags = current->mm->flags & MMF_INIT_MASK;
1034                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1035         } else {
1036                 mm->flags = default_dump_filter;
1037                 mm->def_flags = 0;
1038         }
1039
1040         if (mm_alloc_pgd(mm))
1041                 goto fail_nopgd;
1042
1043         if (init_new_context(p, mm))
1044                 goto fail_nocontext;
1045
1046         mm->user_ns = get_user_ns(user_ns);
1047         return mm;
1048
1049 fail_nocontext:
1050         mm_free_pgd(mm);
1051 fail_nopgd:
1052         free_mm(mm);
1053         return NULL;
1054 }
1055
1056 /*
1057  * Allocate and initialize an mm_struct.
1058  */
1059 struct mm_struct *mm_alloc(void)
1060 {
1061         struct mm_struct *mm;
1062
1063         mm = allocate_mm();
1064         if (!mm)
1065                 return NULL;
1066
1067         memset(mm, 0, sizeof(*mm));
1068         return mm_init(mm, current, current_user_ns());
1069 }
1070
1071 static inline void __mmput(struct mm_struct *mm)
1072 {
1073         VM_BUG_ON(atomic_read(&mm->mm_users));
1074
1075         uprobe_clear_state(mm);
1076         exit_aio(mm);
1077         ksm_exit(mm);
1078         khugepaged_exit(mm); /* must run before exit_mmap */
1079         exit_mmap(mm);
1080         mm_put_huge_zero_page(mm);
1081         set_mm_exe_file(mm, NULL);
1082         if (!list_empty(&mm->mmlist)) {
1083                 spin_lock(&mmlist_lock);
1084                 list_del(&mm->mmlist);
1085                 spin_unlock(&mmlist_lock);
1086         }
1087         if (mm->binfmt)
1088                 module_put(mm->binfmt->module);
1089         mmdrop(mm);
1090 }
1091
1092 /*
1093  * Decrement the use count and release all resources for an mm.
1094  */
1095 void mmput(struct mm_struct *mm)
1096 {
1097         might_sleep();
1098
1099         if (atomic_dec_and_test(&mm->mm_users))
1100                 __mmput(mm);
1101 }
1102 EXPORT_SYMBOL_GPL(mmput);
1103
1104 #ifdef CONFIG_MMU
1105 static void mmput_async_fn(struct work_struct *work)
1106 {
1107         struct mm_struct *mm = container_of(work, struct mm_struct,
1108                                             async_put_work);
1109
1110         __mmput(mm);
1111 }
1112
1113 void mmput_async(struct mm_struct *mm)
1114 {
1115         if (atomic_dec_and_test(&mm->mm_users)) {
1116                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1117                 schedule_work(&mm->async_put_work);
1118         }
1119 }
1120 #endif
1121
1122 /**
1123  * set_mm_exe_file - change a reference to the mm's executable file
1124  *
1125  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1126  *
1127  * Main users are mmput() and sys_execve(). Callers prevent concurrent
1128  * invocations: in mmput() nobody alive left, in execve task is single
1129  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1130  * mm->exe_file, but does so without using set_mm_exe_file() in order
1131  * to do avoid the need for any locks.
1132  */
1133 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1134 {
1135         struct file *old_exe_file;
1136
1137         /*
1138          * It is safe to dereference the exe_file without RCU as
1139          * this function is only called if nobody else can access
1140          * this mm -- see comment above for justification.
1141          */
1142         old_exe_file = rcu_dereference_raw(mm->exe_file);
1143
1144         if (new_exe_file)
1145                 get_file(new_exe_file);
1146         rcu_assign_pointer(mm->exe_file, new_exe_file);
1147         if (old_exe_file)
1148                 fput(old_exe_file);
1149 }
1150
1151 /**
1152  * get_mm_exe_file - acquire a reference to the mm's executable file
1153  *
1154  * Returns %NULL if mm has no associated executable file.
1155  * User must release file via fput().
1156  */
1157 struct file *get_mm_exe_file(struct mm_struct *mm)
1158 {
1159         struct file *exe_file;
1160
1161         rcu_read_lock();
1162         exe_file = rcu_dereference(mm->exe_file);
1163         if (exe_file && !get_file_rcu(exe_file))
1164                 exe_file = NULL;
1165         rcu_read_unlock();
1166         return exe_file;
1167 }
1168 EXPORT_SYMBOL(get_mm_exe_file);
1169
1170 /**
1171  * get_task_exe_file - acquire a reference to the task's executable file
1172  *
1173  * Returns %NULL if task's mm (if any) has no associated executable file or
1174  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1175  * User must release file via fput().
1176  */
1177 struct file *get_task_exe_file(struct task_struct *task)
1178 {
1179         struct file *exe_file = NULL;
1180         struct mm_struct *mm;
1181
1182         task_lock(task);
1183         mm = task->mm;
1184         if (mm) {
1185                 if (!(task->flags & PF_KTHREAD))
1186                         exe_file = get_mm_exe_file(mm);
1187         }
1188         task_unlock(task);
1189         return exe_file;
1190 }
1191 EXPORT_SYMBOL(get_task_exe_file);
1192
1193 /**
1194  * get_task_mm - acquire a reference to the task's mm
1195  *
1196  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1197  * this kernel workthread has transiently adopted a user mm with use_mm,
1198  * to do its AIO) is not set and if so returns a reference to it, after
1199  * bumping up the use count.  User must release the mm via mmput()
1200  * after use.  Typically used by /proc and ptrace.
1201  */
1202 struct mm_struct *get_task_mm(struct task_struct *task)
1203 {
1204         struct mm_struct *mm;
1205
1206         task_lock(task);
1207         mm = task->mm;
1208         if (mm) {
1209                 if (task->flags & PF_KTHREAD)
1210                         mm = NULL;
1211                 else
1212                         mmget(mm);
1213         }
1214         task_unlock(task);
1215         return mm;
1216 }
1217 EXPORT_SYMBOL_GPL(get_task_mm);
1218
1219 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1220 {
1221         struct mm_struct *mm;
1222         int err;
1223
1224         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
1225         if (err)
1226                 return ERR_PTR(err);
1227
1228         mm = get_task_mm(task);
1229         if (mm && mm != current->mm &&
1230                         !ptrace_may_access(task, mode)) {
1231                 mmput(mm);
1232                 mm = ERR_PTR(-EACCES);
1233         }
1234         mutex_unlock(&task->signal->cred_guard_mutex);
1235
1236         return mm;
1237 }
1238
1239 static void complete_vfork_done(struct task_struct *tsk)
1240 {
1241         struct completion *vfork;
1242
1243         task_lock(tsk);
1244         vfork = tsk->vfork_done;
1245         if (likely(vfork)) {
1246                 tsk->vfork_done = NULL;
1247                 complete(vfork);
1248         }
1249         task_unlock(tsk);
1250 }
1251
1252 static int wait_for_vfork_done(struct task_struct *child,
1253                                 struct completion *vfork)
1254 {
1255         int killed;
1256
1257         freezer_do_not_count();
1258         cgroup_enter_frozen();
1259         killed = wait_for_completion_killable(vfork);
1260         cgroup_leave_frozen(false);
1261         freezer_count();
1262
1263         if (killed) {
1264                 task_lock(child);
1265                 child->vfork_done = NULL;
1266                 task_unlock(child);
1267         }
1268
1269         put_task_struct(child);
1270         return killed;
1271 }
1272
1273 /* Please note the differences between mmput and mm_release.
1274  * mmput is called whenever we stop holding onto a mm_struct,
1275  * error success whatever.
1276  *
1277  * mm_release is called after a mm_struct has been removed
1278  * from the current process.
1279  *
1280  * This difference is important for error handling, when we
1281  * only half set up a mm_struct for a new process and need to restore
1282  * the old one.  Because we mmput the new mm_struct before
1283  * restoring the old one. . .
1284  * Eric Biederman 10 January 1998
1285  */
1286 static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1287 {
1288         uprobe_free_utask(tsk);
1289
1290         /* Get rid of any cached register state */
1291         deactivate_mm(tsk, mm);
1292
1293         /*
1294          * Signal userspace if we're not exiting with a core dump
1295          * because we want to leave the value intact for debugging
1296          * purposes.
1297          */
1298         if (tsk->clear_child_tid) {
1299                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1300                     atomic_read(&mm->mm_users) > 1) {
1301                         /*
1302                          * We don't check the error code - if userspace has
1303                          * not set up a proper pointer then tough luck.
1304                          */
1305                         put_user(0, tsk->clear_child_tid);
1306                         do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1307                                         1, NULL, NULL, 0, 0);
1308                 }
1309                 tsk->clear_child_tid = NULL;
1310         }
1311
1312         /*
1313          * All done, finally we can wake up parent and return this mm to him.
1314          * Also kthread_stop() uses this completion for synchronization.
1315          */
1316         if (tsk->vfork_done)
1317                 complete_vfork_done(tsk);
1318 }
1319
1320 void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1321 {
1322         futex_exit_release(tsk);
1323         mm_release(tsk, mm);
1324 }
1325
1326 void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1327 {
1328         futex_exec_release(tsk);
1329         mm_release(tsk, mm);
1330 }
1331
1332 /**
1333  * dup_mm() - duplicates an existing mm structure
1334  * @tsk: the task_struct with which the new mm will be associated.
1335  * @oldmm: the mm to duplicate.
1336  *
1337  * Allocates a new mm structure and duplicates the provided @oldmm structure
1338  * content into it.
1339  *
1340  * Return: the duplicated mm or NULL on failure.
1341  */
1342 static struct mm_struct *dup_mm(struct task_struct *tsk,
1343                                 struct mm_struct *oldmm)
1344 {
1345         struct mm_struct *mm;
1346         int err;
1347
1348         mm = allocate_mm();
1349         if (!mm)
1350                 goto fail_nomem;
1351
1352         memcpy(mm, oldmm, sizeof(*mm));
1353
1354         if (!mm_init(mm, tsk, mm->user_ns))
1355                 goto fail_nomem;
1356
1357         err = dup_mmap(mm, oldmm);
1358         if (err)
1359                 goto free_pt;
1360
1361         mm->hiwater_rss = get_mm_rss(mm);
1362         mm->hiwater_vm = mm->total_vm;
1363
1364         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1365                 goto free_pt;
1366
1367         return mm;
1368
1369 free_pt:
1370         /* don't put binfmt in mmput, we haven't got module yet */
1371         mm->binfmt = NULL;
1372         mm_init_owner(mm, NULL);
1373         mmput(mm);
1374
1375 fail_nomem:
1376         return NULL;
1377 }
1378
1379 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1380 {
1381         struct mm_struct *mm, *oldmm;
1382         int retval;
1383
1384         tsk->min_flt = tsk->maj_flt = 0;
1385         tsk->nvcsw = tsk->nivcsw = 0;
1386 #ifdef CONFIG_DETECT_HUNG_TASK
1387         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1388         tsk->last_switch_time = 0;
1389 #endif
1390
1391         tsk->mm = NULL;
1392         tsk->active_mm = NULL;
1393
1394         /*
1395          * Are we cloning a kernel thread?
1396          *
1397          * We need to steal a active VM for that..
1398          */
1399         oldmm = current->mm;
1400         if (!oldmm)
1401                 return 0;
1402
1403         /* initialize the new vmacache entries */
1404         vmacache_flush(tsk);
1405
1406         if (clone_flags & CLONE_VM) {
1407                 mmget(oldmm);
1408                 mm = oldmm;
1409                 goto good_mm;
1410         }
1411
1412         retval = -ENOMEM;
1413         mm = dup_mm(tsk, current->mm);
1414         if (!mm)
1415                 goto fail_nomem;
1416
1417 good_mm:
1418         tsk->mm = mm;
1419         tsk->active_mm = mm;
1420         return 0;
1421
1422 fail_nomem:
1423         return retval;
1424 }
1425
1426 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1427 {
1428         struct fs_struct *fs = current->fs;
1429         if (clone_flags & CLONE_FS) {
1430                 /* tsk->fs is already what we want */
1431                 spin_lock(&fs->lock);
1432                 if (fs->in_exec) {
1433                         spin_unlock(&fs->lock);
1434                         return -EAGAIN;
1435                 }
1436                 fs->users++;
1437                 spin_unlock(&fs->lock);
1438                 return 0;
1439         }
1440         tsk->fs = copy_fs_struct(fs);
1441         if (!tsk->fs)
1442                 return -ENOMEM;
1443         return 0;
1444 }
1445
1446 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1447 {
1448         struct files_struct *oldf, *newf;
1449         int error = 0;
1450
1451         /*
1452          * A background process may not have any files ...
1453          */
1454         oldf = current->files;
1455         if (!oldf)
1456                 goto out;
1457
1458         if (clone_flags & CLONE_FILES) {
1459                 atomic_inc(&oldf->count);
1460                 goto out;
1461         }
1462
1463         newf = dup_fd(oldf, &error);
1464         if (!newf)
1465                 goto out;
1466
1467         tsk->files = newf;
1468         error = 0;
1469 out:
1470         return error;
1471 }
1472
1473 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1474 {
1475 #ifdef CONFIG_BLOCK
1476         struct io_context *ioc = current->io_context;
1477         struct io_context *new_ioc;
1478
1479         if (!ioc)
1480                 return 0;
1481         /*
1482          * Share io context with parent, if CLONE_IO is set
1483          */
1484         if (clone_flags & CLONE_IO) {
1485                 ioc_task_link(ioc);
1486                 tsk->io_context = ioc;
1487         } else if (ioprio_valid(ioc->ioprio)) {
1488                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1489                 if (unlikely(!new_ioc))
1490                         return -ENOMEM;
1491
1492                 new_ioc->ioprio = ioc->ioprio;
1493                 put_io_context(new_ioc);
1494         }
1495 #endif
1496         return 0;
1497 }
1498
1499 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1500 {
1501         struct sighand_struct *sig;
1502
1503         if (clone_flags & CLONE_SIGHAND) {
1504                 refcount_inc(&current->sighand->count);
1505                 return 0;
1506         }
1507         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1508         rcu_assign_pointer(tsk->sighand, sig);
1509         if (!sig)
1510                 return -ENOMEM;
1511
1512         refcount_set(&sig->count, 1);
1513         spin_lock_irq(&current->sighand->siglock);
1514         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1515         spin_unlock_irq(&current->sighand->siglock);
1516
1517         /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1518         if (clone_flags & CLONE_CLEAR_SIGHAND)
1519                 flush_signal_handlers(tsk, 0);
1520
1521         return 0;
1522 }
1523
1524 void __cleanup_sighand(struct sighand_struct *sighand)
1525 {
1526         if (refcount_dec_and_test(&sighand->count)) {
1527                 signalfd_cleanup(sighand);
1528                 /*
1529                  * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1530                  * without an RCU grace period, see __lock_task_sighand().
1531                  */
1532                 kmem_cache_free(sighand_cachep, sighand);
1533         }
1534 }
1535
1536 /*
1537  * Initialize POSIX timer handling for a thread group.
1538  */
1539 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1540 {
1541         struct posix_cputimers *pct = &sig->posix_cputimers;
1542         unsigned long cpu_limit;
1543
1544         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1545         posix_cputimers_group_init(pct, cpu_limit);
1546 }
1547
1548 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1549 {
1550         struct signal_struct *sig;
1551
1552         if (clone_flags & CLONE_THREAD)
1553                 return 0;
1554
1555         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1556         tsk->signal = sig;
1557         if (!sig)
1558                 return -ENOMEM;
1559
1560         sig->nr_threads = 1;
1561         atomic_set(&sig->live, 1);
1562         refcount_set(&sig->sigcnt, 1);
1563
1564         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1565         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1566         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1567
1568         init_waitqueue_head(&sig->wait_chldexit);
1569         sig->curr_target = tsk;
1570         init_sigpending(&sig->shared_pending);
1571         INIT_HLIST_HEAD(&sig->multiprocess);
1572         seqlock_init(&sig->stats_lock);
1573         prev_cputime_init(&sig->prev_cputime);
1574
1575 #ifdef CONFIG_POSIX_TIMERS
1576         INIT_LIST_HEAD(&sig->posix_timers);
1577         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1578         sig->real_timer.function = it_real_fn;
1579 #endif
1580
1581         task_lock(current->group_leader);
1582         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1583         task_unlock(current->group_leader);
1584
1585         posix_cpu_timers_init_group(sig);
1586
1587         tty_audit_fork(sig);
1588         sched_autogroup_fork(sig);
1589
1590         sig->oom_score_adj = current->signal->oom_score_adj;
1591         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1592
1593         mutex_init(&sig->cred_guard_mutex);
1594
1595         return 0;
1596 }
1597
1598 static void copy_seccomp(struct task_struct *p)
1599 {
1600 #ifdef CONFIG_SECCOMP
1601         /*
1602          * Must be called with sighand->lock held, which is common to
1603          * all threads in the group. Holding cred_guard_mutex is not
1604          * needed because this new task is not yet running and cannot
1605          * be racing exec.
1606          */
1607         assert_spin_locked(&current->sighand->siglock);
1608
1609         /* Ref-count the new filter user, and assign it. */
1610         get_seccomp_filter(current);
1611         p->seccomp = current->seccomp;
1612
1613         /*
1614          * Explicitly enable no_new_privs here in case it got set
1615          * between the task_struct being duplicated and holding the
1616          * sighand lock. The seccomp state and nnp must be in sync.
1617          */
1618         if (task_no_new_privs(current))
1619                 task_set_no_new_privs(p);
1620
1621         /*
1622          * If the parent gained a seccomp mode after copying thread
1623          * flags and between before we held the sighand lock, we have
1624          * to manually enable the seccomp thread flag here.
1625          */
1626         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1627                 set_tsk_thread_flag(p, TIF_SECCOMP);
1628 #endif
1629 }
1630
1631 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1632 {
1633         current->clear_child_tid = tidptr;
1634
1635         return task_pid_vnr(current);
1636 }
1637
1638 static void rt_mutex_init_task(struct task_struct *p)
1639 {
1640         raw_spin_lock_init(&p->pi_lock);
1641 #ifdef CONFIG_RT_MUTEXES
1642         p->pi_waiters = RB_ROOT_CACHED;
1643         p->pi_top_task = NULL;
1644         p->pi_blocked_on = NULL;
1645 #endif
1646 }
1647
1648 static inline void init_task_pid_links(struct task_struct *task)
1649 {
1650         enum pid_type type;
1651
1652         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1653                 INIT_HLIST_NODE(&task->pid_links[type]);
1654         }
1655 }
1656
1657 static inline void
1658 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1659 {
1660         if (type == PIDTYPE_PID)
1661                 task->thread_pid = pid;
1662         else
1663                 task->signal->pids[type] = pid;
1664 }
1665
1666 static inline void rcu_copy_process(struct task_struct *p)
1667 {
1668 #ifdef CONFIG_PREEMPT_RCU
1669         p->rcu_read_lock_nesting = 0;
1670         p->rcu_read_unlock_special.s = 0;
1671         p->rcu_blocked_node = NULL;
1672         INIT_LIST_HEAD(&p->rcu_node_entry);
1673 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1674 #ifdef CONFIG_TASKS_RCU
1675         p->rcu_tasks_holdout = false;
1676         INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1677         p->rcu_tasks_idle_cpu = -1;
1678 #endif /* #ifdef CONFIG_TASKS_RCU */
1679 }
1680
1681 struct pid *pidfd_pid(const struct file *file)
1682 {
1683         if (file->f_op == &pidfd_fops)
1684                 return file->private_data;
1685
1686         return ERR_PTR(-EBADF);
1687 }
1688
1689 static int pidfd_release(struct inode *inode, struct file *file)
1690 {
1691         struct pid *pid = file->private_data;
1692
1693         file->private_data = NULL;
1694         put_pid(pid);
1695         return 0;
1696 }
1697
1698 #ifdef CONFIG_PROC_FS
1699 /**
1700  * pidfd_show_fdinfo - print information about a pidfd
1701  * @m: proc fdinfo file
1702  * @f: file referencing a pidfd
1703  *
1704  * Pid:
1705  * This function will print the pid that a given pidfd refers to in the
1706  * pid namespace of the procfs instance.
1707  * If the pid namespace of the process is not a descendant of the pid
1708  * namespace of the procfs instance 0 will be shown as its pid. This is
1709  * similar to calling getppid() on a process whose parent is outside of
1710  * its pid namespace.
1711  *
1712  * NSpid:
1713  * If pid namespaces are supported then this function will also print
1714  * the pid of a given pidfd refers to for all descendant pid namespaces
1715  * starting from the current pid namespace of the instance, i.e. the
1716  * Pid field and the first entry in the NSpid field will be identical.
1717  * If the pid namespace of the process is not a descendant of the pid
1718  * namespace of the procfs instance 0 will be shown as its first NSpid
1719  * entry and no others will be shown.
1720  * Note that this differs from the Pid and NSpid fields in
1721  * /proc/<pid>/status where Pid and NSpid are always shown relative to
1722  * the  pid namespace of the procfs instance. The difference becomes
1723  * obvious when sending around a pidfd between pid namespaces from a
1724  * different branch of the tree, i.e. where no ancestoral relation is
1725  * present between the pid namespaces:
1726  * - create two new pid namespaces ns1 and ns2 in the initial pid
1727  *   namespace (also take care to create new mount namespaces in the
1728  *   new pid namespace and mount procfs)
1729  * - create a process with a pidfd in ns1
1730  * - send pidfd from ns1 to ns2
1731  * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
1732  *   have exactly one entry, which is 0
1733  */
1734 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1735 {
1736         struct pid *pid = f->private_data;
1737         struct pid_namespace *ns;
1738         pid_t nr = -1;
1739
1740         if (likely(pid_has_task(pid, PIDTYPE_PID))) {
1741                 ns = proc_pid_ns(file_inode(m->file));
1742                 nr = pid_nr_ns(pid, ns);
1743         }
1744
1745         seq_put_decimal_ll(m, "Pid:\t", nr);
1746
1747 #ifdef CONFIG_PID_NS
1748         seq_put_decimal_ll(m, "\nNSpid:\t", nr);
1749         if (nr > 0) {
1750                 int i;
1751
1752                 /* If nr is non-zero it means that 'pid' is valid and that
1753                  * ns, i.e. the pid namespace associated with the procfs
1754                  * instance, is in the pid namespace hierarchy of pid.
1755                  * Start at one below the already printed level.
1756                  */
1757                 for (i = ns->level + 1; i <= pid->level; i++)
1758                         seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
1759         }
1760 #endif
1761         seq_putc(m, '\n');
1762 }
1763 #endif
1764
1765 /*
1766  * Poll support for process exit notification.
1767  */
1768 static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
1769 {
1770         struct task_struct *task;
1771         struct pid *pid = file->private_data;
1772         __poll_t poll_flags = 0;
1773
1774         poll_wait(file, &pid->wait_pidfd, pts);
1775
1776         rcu_read_lock();
1777         task = pid_task(pid, PIDTYPE_PID);
1778         /*
1779          * Inform pollers only when the whole thread group exits.
1780          * If the thread group leader exits before all other threads in the
1781          * group, then poll(2) should block, similar to the wait(2) family.
1782          */
1783         if (!task || (task->exit_state && thread_group_empty(task)))
1784                 poll_flags = EPOLLIN | EPOLLRDNORM;
1785         rcu_read_unlock();
1786
1787         return poll_flags;
1788 }
1789
1790 const struct file_operations pidfd_fops = {
1791         .release = pidfd_release,
1792         .poll = pidfd_poll,
1793 #ifdef CONFIG_PROC_FS
1794         .show_fdinfo = pidfd_show_fdinfo,
1795 #endif
1796 };
1797
1798 static void __delayed_free_task(struct rcu_head *rhp)
1799 {
1800         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1801
1802         free_task(tsk);
1803 }
1804
1805 static __always_inline void delayed_free_task(struct task_struct *tsk)
1806 {
1807         if (IS_ENABLED(CONFIG_MEMCG))
1808                 call_rcu(&tsk->rcu, __delayed_free_task);
1809         else
1810                 free_task(tsk);
1811 }
1812
1813 /*
1814  * This creates a new process as a copy of the old one,
1815  * but does not actually start it yet.
1816  *
1817  * It copies the registers, and all the appropriate
1818  * parts of the process environment (as per the clone
1819  * flags). The actual kick-off is left to the caller.
1820  */
1821 static __latent_entropy struct task_struct *copy_process(
1822                                         struct pid *pid,
1823                                         int trace,
1824                                         int node,
1825                                         struct kernel_clone_args *args)
1826 {
1827         int pidfd = -1, retval;
1828         struct task_struct *p;
1829         struct multiprocess_signals delayed;
1830         struct file *pidfile = NULL;
1831         u64 clone_flags = args->flags;
1832
1833         /*
1834          * Don't allow sharing the root directory with processes in a different
1835          * namespace
1836          */
1837         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1838                 return ERR_PTR(-EINVAL);
1839
1840         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1841                 return ERR_PTR(-EINVAL);
1842
1843         /*
1844          * Thread groups must share signals as well, and detached threads
1845          * can only be started up within the thread group.
1846          */
1847         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1848                 return ERR_PTR(-EINVAL);
1849
1850         /*
1851          * Shared signal handlers imply shared VM. By way of the above,
1852          * thread groups also imply shared VM. Blocking this case allows
1853          * for various simplifications in other code.
1854          */
1855         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1856                 return ERR_PTR(-EINVAL);
1857
1858         /*
1859          * Siblings of global init remain as zombies on exit since they are
1860          * not reaped by their parent (swapper). To solve this and to avoid
1861          * multi-rooted process trees, prevent global and container-inits
1862          * from creating siblings.
1863          */
1864         if ((clone_flags & CLONE_PARENT) &&
1865                                 current->signal->flags & SIGNAL_UNKILLABLE)
1866                 return ERR_PTR(-EINVAL);
1867
1868         /*
1869          * If the new process will be in a different pid or user namespace
1870          * do not allow it to share a thread group with the forking task.
1871          */
1872         if (clone_flags & CLONE_THREAD) {
1873                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1874                     (task_active_pid_ns(current) !=
1875                                 current->nsproxy->pid_ns_for_children))
1876                         return ERR_PTR(-EINVAL);
1877         }
1878
1879         if (clone_flags & CLONE_PIDFD) {
1880                 /*
1881                  * - CLONE_DETACHED is blocked so that we can potentially
1882                  *   reuse it later for CLONE_PIDFD.
1883                  * - CLONE_THREAD is blocked until someone really needs it.
1884                  */
1885                 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
1886                         return ERR_PTR(-EINVAL);
1887         }
1888
1889         /*
1890          * Force any signals received before this point to be delivered
1891          * before the fork happens.  Collect up signals sent to multiple
1892          * processes that happen during the fork and delay them so that
1893          * they appear to happen after the fork.
1894          */
1895         sigemptyset(&delayed.signal);
1896         INIT_HLIST_NODE(&delayed.node);
1897
1898         spin_lock_irq(&current->sighand->siglock);
1899         if (!(clone_flags & CLONE_THREAD))
1900                 hlist_add_head(&delayed.node, &current->signal->multiprocess);
1901         recalc_sigpending();
1902         spin_unlock_irq(&current->sighand->siglock);
1903         retval = -ERESTARTNOINTR;
1904         if (signal_pending(current))
1905                 goto fork_out;
1906
1907         retval = -ENOMEM;
1908         p = dup_task_struct(current, node);
1909         if (!p)
1910                 goto fork_out;
1911
1912         /*
1913          * This _must_ happen before we call free_task(), i.e. before we jump
1914          * to any of the bad_fork_* labels. This is to avoid freeing
1915          * p->set_child_tid which is (ab)used as a kthread's data pointer for
1916          * kernel threads (PF_KTHREAD).
1917          */
1918         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
1919         /*
1920          * Clear TID on mm_release()?
1921          */
1922         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
1923
1924         ftrace_graph_init_task(p);
1925
1926         rt_mutex_init_task(p);
1927
1928 #ifdef CONFIG_PROVE_LOCKING
1929         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1930         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1931 #endif
1932         retval = -EAGAIN;
1933         if (atomic_read(&p->real_cred->user->processes) >=
1934                         task_rlimit(p, RLIMIT_NPROC)) {
1935                 if (p->real_cred->user != INIT_USER &&
1936                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1937                         goto bad_fork_free;
1938         }
1939         current->flags &= ~PF_NPROC_EXCEEDED;
1940
1941         retval = copy_creds(p, clone_flags);
1942         if (retval < 0)
1943                 goto bad_fork_free;
1944
1945         /*
1946          * If multiple threads are within copy_process(), then this check
1947          * triggers too late. This doesn't hurt, the check is only there
1948          * to stop root fork bombs.
1949          */
1950         retval = -EAGAIN;
1951         if (nr_threads >= max_threads)
1952                 goto bad_fork_cleanup_count;
1953
1954         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1955         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1956         p->flags |= PF_FORKNOEXEC;
1957         INIT_LIST_HEAD(&p->children);
1958         INIT_LIST_HEAD(&p->sibling);
1959         rcu_copy_process(p);
1960         p->vfork_done = NULL;
1961         spin_lock_init(&p->alloc_lock);
1962
1963         init_sigpending(&p->pending);
1964
1965         p->utime = p->stime = p->gtime = 0;
1966 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1967         p->utimescaled = p->stimescaled = 0;
1968 #endif
1969         prev_cputime_init(&p->prev_cputime);
1970
1971 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1972         seqcount_init(&p->vtime.seqcount);
1973         p->vtime.starttime = 0;
1974         p->vtime.state = VTIME_INACTIVE;
1975 #endif
1976
1977 #if defined(SPLIT_RSS_COUNTING)
1978         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1979 #endif
1980
1981         p->default_timer_slack_ns = current->timer_slack_ns;
1982
1983 #ifdef CONFIG_PSI
1984         p->psi_flags = 0;
1985 #endif
1986
1987         task_io_accounting_init(&p->ioac);
1988         acct_clear_integrals(p);
1989
1990         posix_cputimers_init(&p->posix_cputimers);
1991
1992         p->io_context = NULL;
1993         audit_set_context(p, NULL);
1994         cgroup_fork(p);
1995 #ifdef CONFIG_NUMA
1996         p->mempolicy = mpol_dup(p->mempolicy);
1997         if (IS_ERR(p->mempolicy)) {
1998                 retval = PTR_ERR(p->mempolicy);
1999                 p->mempolicy = NULL;
2000                 goto bad_fork_cleanup_threadgroup_lock;
2001         }
2002 #endif
2003 #ifdef CONFIG_CPUSETS
2004         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2005         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2006         seqcount_init(&p->mems_allowed_seq);
2007 #endif
2008 #ifdef CONFIG_TRACE_IRQFLAGS
2009         p->irq_events = 0;
2010         p->hardirqs_enabled = 0;
2011         p->hardirq_enable_ip = 0;
2012         p->hardirq_enable_event = 0;
2013         p->hardirq_disable_ip = _THIS_IP_;
2014         p->hardirq_disable_event = 0;
2015         p->softirqs_enabled = 1;
2016         p->softirq_enable_ip = _THIS_IP_;
2017         p->softirq_enable_event = 0;
2018         p->softirq_disable_ip = 0;
2019         p->softirq_disable_event = 0;
2020         p->hardirq_context = 0;
2021         p->softirq_context = 0;
2022 #endif
2023
2024         p->pagefault_disabled = 0;
2025
2026 #ifdef CONFIG_LOCKDEP
2027         lockdep_init_task(p);
2028 #endif
2029
2030 #ifdef CONFIG_DEBUG_MUTEXES
2031         p->blocked_on = NULL; /* not blocked yet */
2032 #endif
2033 #ifdef CONFIG_BCACHE
2034         p->sequential_io        = 0;
2035         p->sequential_io_avg    = 0;
2036 #endif
2037
2038         /* Perform scheduler related setup. Assign this task to a CPU. */
2039         retval = sched_fork(clone_flags, p);
2040         if (retval)
2041                 goto bad_fork_cleanup_policy;
2042
2043         retval = perf_event_init_task(p);
2044         if (retval)
2045                 goto bad_fork_cleanup_policy;
2046         retval = audit_alloc(p);
2047         if (retval)
2048                 goto bad_fork_cleanup_perf;
2049         /* copy all the process information */
2050         shm_init_task(p);
2051         retval = security_task_alloc(p, clone_flags);
2052         if (retval)
2053                 goto bad_fork_cleanup_audit;
2054         retval = copy_semundo(clone_flags, p);
2055         if (retval)
2056                 goto bad_fork_cleanup_security;
2057         retval = copy_files(clone_flags, p);
2058         if (retval)
2059                 goto bad_fork_cleanup_semundo;
2060         retval = copy_fs(clone_flags, p);
2061         if (retval)
2062                 goto bad_fork_cleanup_files;
2063         retval = copy_sighand(clone_flags, p);
2064         if (retval)
2065                 goto bad_fork_cleanup_fs;
2066         retval = copy_signal(clone_flags, p);
2067         if (retval)
2068                 goto bad_fork_cleanup_sighand;
2069         retval = copy_mm(clone_flags, p);
2070         if (retval)
2071                 goto bad_fork_cleanup_signal;
2072         retval = copy_namespaces(clone_flags, p);
2073         if (retval)
2074                 goto bad_fork_cleanup_mm;
2075         retval = copy_io(clone_flags, p);
2076         if (retval)
2077                 goto bad_fork_cleanup_namespaces;
2078         retval = copy_thread_tls(clone_flags, args->stack, args->stack_size, p,
2079                                  args->tls);
2080         if (retval)
2081                 goto bad_fork_cleanup_io;
2082
2083         stackleak_task_init(p);
2084
2085         if (pid != &init_struct_pid) {
2086                 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2087                                 args->set_tid_size);
2088                 if (IS_ERR(pid)) {
2089                         retval = PTR_ERR(pid);
2090                         goto bad_fork_cleanup_thread;
2091                 }
2092         }
2093
2094         /*
2095          * This has to happen after we've potentially unshared the file
2096          * descriptor table (so that the pidfd doesn't leak into the child
2097          * if the fd table isn't shared).
2098          */
2099         if (clone_flags & CLONE_PIDFD) {
2100                 retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2101                 if (retval < 0)
2102                         goto bad_fork_free_pid;
2103
2104                 pidfd = retval;
2105
2106                 pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2107                                               O_RDWR | O_CLOEXEC);
2108                 if (IS_ERR(pidfile)) {
2109                         put_unused_fd(pidfd);
2110                         retval = PTR_ERR(pidfile);
2111                         goto bad_fork_free_pid;
2112                 }
2113                 get_pid(pid);   /* held by pidfile now */
2114
2115                 retval = put_user(pidfd, args->pidfd);
2116                 if (retval)
2117                         goto bad_fork_put_pidfd;
2118         }
2119
2120 #ifdef CONFIG_BLOCK
2121         p->plug = NULL;
2122 #endif
2123         futex_init_task(p);
2124
2125         /*
2126          * sigaltstack should be cleared when sharing the same VM
2127          */
2128         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2129                 sas_ss_reset(p);
2130
2131         /*
2132          * Syscall tracing and stepping should be turned off in the
2133          * child regardless of CLONE_PTRACE.
2134          */
2135         user_disable_single_step(p);
2136         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
2137 #ifdef TIF_SYSCALL_EMU
2138         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
2139 #endif
2140         clear_tsk_latency_tracing(p);
2141
2142         /* ok, now we should be set up.. */
2143         p->pid = pid_nr(pid);
2144         if (clone_flags & CLONE_THREAD) {
2145                 p->exit_signal = -1;
2146                 p->group_leader = current->group_leader;
2147                 p->tgid = current->tgid;
2148         } else {
2149                 if (clone_flags & CLONE_PARENT)
2150                         p->exit_signal = current->group_leader->exit_signal;
2151                 else
2152                         p->exit_signal = args->exit_signal;
2153                 p->group_leader = p;
2154                 p->tgid = p->pid;
2155         }
2156
2157         p->nr_dirtied = 0;
2158         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2159         p->dirty_paused_when = 0;
2160
2161         p->pdeath_signal = 0;
2162         INIT_LIST_HEAD(&p->thread_group);
2163         p->task_works = NULL;
2164
2165         cgroup_threadgroup_change_begin(current);
2166         /*
2167          * Ensure that the cgroup subsystem policies allow the new process to be
2168          * forked. It should be noted the the new process's css_set can be changed
2169          * between here and cgroup_post_fork() if an organisation operation is in
2170          * progress.
2171          */
2172         retval = cgroup_can_fork(p);
2173         if (retval)
2174                 goto bad_fork_cgroup_threadgroup_change_end;
2175
2176         /*
2177          * From this point on we must avoid any synchronous user-space
2178          * communication until we take the tasklist-lock. In particular, we do
2179          * not want user-space to be able to predict the process start-time by
2180          * stalling fork(2) after we recorded the start_time but before it is
2181          * visible to the system.
2182          */
2183
2184         p->start_time = ktime_get_ns();
2185         p->real_start_time = ktime_get_boottime_ns();
2186
2187         /*
2188          * Make it visible to the rest of the system, but dont wake it up yet.
2189          * Need tasklist lock for parent etc handling!
2190          */
2191         write_lock_irq(&tasklist_lock);
2192
2193         /* CLONE_PARENT re-uses the old parent */
2194         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2195                 p->real_parent = current->real_parent;
2196                 p->parent_exec_id = current->parent_exec_id;
2197         } else {
2198                 p->real_parent = current;
2199                 p->parent_exec_id = current->self_exec_id;
2200         }
2201
2202         klp_copy_process(p);
2203
2204         spin_lock(&current->sighand->siglock);
2205
2206         /*
2207          * Copy seccomp details explicitly here, in case they were changed
2208          * before holding sighand lock.
2209          */
2210         copy_seccomp(p);
2211
2212         rseq_fork(p, clone_flags);
2213
2214         /* Don't start children in a dying pid namespace */
2215         if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2216                 retval = -ENOMEM;
2217                 goto bad_fork_cancel_cgroup;
2218         }
2219
2220         /* Let kill terminate clone/fork in the middle */
2221         if (fatal_signal_pending(current)) {
2222                 retval = -EINTR;
2223                 goto bad_fork_cancel_cgroup;
2224         }
2225
2226         /* past the last point of failure */
2227         if (pidfile)
2228                 fd_install(pidfd, pidfile);
2229
2230         init_task_pid_links(p);
2231         if (likely(p->pid)) {
2232                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2233
2234                 init_task_pid(p, PIDTYPE_PID, pid);
2235                 if (thread_group_leader(p)) {
2236                         init_task_pid(p, PIDTYPE_TGID, pid);
2237                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2238                         init_task_pid(p, PIDTYPE_SID, task_session(current));
2239
2240                         if (is_child_reaper(pid)) {
2241                                 ns_of_pid(pid)->child_reaper = p;
2242                                 p->signal->flags |= SIGNAL_UNKILLABLE;
2243                         }
2244                         p->signal->shared_pending.signal = delayed.signal;
2245                         p->signal->tty = tty_kref_get(current->signal->tty);
2246                         /*
2247                          * Inherit has_child_subreaper flag under the same
2248                          * tasklist_lock with adding child to the process tree
2249                          * for propagate_has_child_subreaper optimization.
2250                          */
2251                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2252                                                          p->real_parent->signal->is_child_subreaper;
2253                         list_add_tail(&p->sibling, &p->real_parent->children);
2254                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
2255                         attach_pid(p, PIDTYPE_TGID);
2256                         attach_pid(p, PIDTYPE_PGID);
2257                         attach_pid(p, PIDTYPE_SID);
2258                         __this_cpu_inc(process_counts);
2259                 } else {
2260                         current->signal->nr_threads++;
2261                         atomic_inc(&current->signal->live);
2262                         refcount_inc(&current->signal->sigcnt);
2263                         task_join_group_stop(p);
2264                         list_add_tail_rcu(&p->thread_group,
2265                                           &p->group_leader->thread_group);
2266                         list_add_tail_rcu(&p->thread_node,
2267                                           &p->signal->thread_head);
2268                 }
2269                 attach_pid(p, PIDTYPE_PID);
2270                 nr_threads++;
2271         }
2272         total_forks++;
2273         hlist_del_init(&delayed.node);
2274         spin_unlock(&current->sighand->siglock);
2275         syscall_tracepoint_update(p);
2276         write_unlock_irq(&tasklist_lock);
2277
2278         proc_fork_connector(p);
2279         cgroup_post_fork(p);
2280         cgroup_threadgroup_change_end(current);
2281         perf_event_fork(p);
2282
2283         trace_task_newtask(p, clone_flags);
2284         uprobe_copy_process(p, clone_flags);
2285
2286         return p;
2287
2288 bad_fork_cancel_cgroup:
2289         spin_unlock(&current->sighand->siglock);
2290         write_unlock_irq(&tasklist_lock);
2291         cgroup_cancel_fork(p);
2292 bad_fork_cgroup_threadgroup_change_end:
2293         cgroup_threadgroup_change_end(current);
2294 bad_fork_put_pidfd:
2295         if (clone_flags & CLONE_PIDFD) {
2296                 fput(pidfile);
2297                 put_unused_fd(pidfd);
2298         }
2299 bad_fork_free_pid:
2300         if (pid != &init_struct_pid)
2301                 free_pid(pid);
2302 bad_fork_cleanup_thread:
2303         exit_thread(p);
2304 bad_fork_cleanup_io:
2305         if (p->io_context)
2306                 exit_io_context(p);
2307 bad_fork_cleanup_namespaces:
2308         exit_task_namespaces(p);
2309 bad_fork_cleanup_mm:
2310         if (p->mm) {
2311                 mm_clear_owner(p->mm, p);
2312                 mmput(p->mm);
2313         }
2314 bad_fork_cleanup_signal:
2315         if (!(clone_flags & CLONE_THREAD))
2316                 free_signal_struct(p->signal);
2317 bad_fork_cleanup_sighand:
2318         __cleanup_sighand(p->sighand);
2319 bad_fork_cleanup_fs:
2320         exit_fs(p); /* blocking */
2321 bad_fork_cleanup_files:
2322         exit_files(p); /* blocking */
2323 bad_fork_cleanup_semundo:
2324         exit_sem(p);
2325 bad_fork_cleanup_security:
2326         security_task_free(p);
2327 bad_fork_cleanup_audit:
2328         audit_free(p);
2329 bad_fork_cleanup_perf:
2330         perf_event_free_task(p);
2331 bad_fork_cleanup_policy:
2332         lockdep_free_task(p);
2333 #ifdef CONFIG_NUMA
2334         mpol_put(p->mempolicy);
2335 bad_fork_cleanup_threadgroup_lock:
2336 #endif
2337         delayacct_tsk_free(p);
2338 bad_fork_cleanup_count:
2339         atomic_dec(&p->cred->user->processes);
2340         exit_creds(p);
2341 bad_fork_free:
2342         p->state = TASK_DEAD;
2343         put_task_stack(p);
2344         delayed_free_task(p);
2345 fork_out:
2346         spin_lock_irq(&current->sighand->siglock);
2347         hlist_del_init(&delayed.node);
2348         spin_unlock_irq(&current->sighand->siglock);
2349         return ERR_PTR(retval);
2350 }
2351
2352 static inline void init_idle_pids(struct task_struct *idle)
2353 {
2354         enum pid_type type;
2355
2356         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2357                 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2358                 init_task_pid(idle, type, &init_struct_pid);
2359         }
2360 }
2361
2362 struct task_struct *fork_idle(int cpu)
2363 {
2364         struct task_struct *task;
2365         struct kernel_clone_args args = {
2366                 .flags = CLONE_VM,
2367         };
2368
2369         task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2370         if (!IS_ERR(task)) {
2371                 init_idle_pids(task);
2372                 init_idle(task, cpu);
2373         }
2374
2375         return task;
2376 }
2377
2378 struct mm_struct *copy_init_mm(void)
2379 {
2380         return dup_mm(NULL, &init_mm);
2381 }
2382
2383 /*
2384  *  Ok, this is the main fork-routine.
2385  *
2386  * It copies the process, and if successful kick-starts
2387  * it and waits for it to finish using the VM if required.
2388  *
2389  * args->exit_signal is expected to be checked for sanity by the caller.
2390  */
2391 long _do_fork(struct kernel_clone_args *args)
2392 {
2393         u64 clone_flags = args->flags;
2394         struct completion vfork;
2395         struct pid *pid;
2396         struct task_struct *p;
2397         int trace = 0;
2398         long nr;
2399
2400         /*
2401          * Determine whether and which event to report to ptracer.  When
2402          * called from kernel_thread or CLONE_UNTRACED is explicitly
2403          * requested, no event is reported; otherwise, report if the event
2404          * for the type of forking is enabled.
2405          */
2406         if (!(clone_flags & CLONE_UNTRACED)) {
2407                 if (clone_flags & CLONE_VFORK)
2408                         trace = PTRACE_EVENT_VFORK;
2409                 else if (args->exit_signal != SIGCHLD)
2410                         trace = PTRACE_EVENT_CLONE;
2411                 else
2412                         trace = PTRACE_EVENT_FORK;
2413
2414                 if (likely(!ptrace_event_enabled(current, trace)))
2415                         trace = 0;
2416         }
2417
2418         p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2419         add_latent_entropy();
2420
2421         if (IS_ERR(p))
2422                 return PTR_ERR(p);
2423
2424         /*
2425          * Do this prior waking up the new thread - the thread pointer
2426          * might get invalid after that point, if the thread exits quickly.
2427          */
2428         trace_sched_process_fork(current, p);
2429
2430         pid = get_task_pid(p, PIDTYPE_PID);
2431         nr = pid_vnr(pid);
2432
2433         if (clone_flags & CLONE_PARENT_SETTID)
2434                 put_user(nr, args->parent_tid);
2435
2436         if (clone_flags & CLONE_VFORK) {
2437                 p->vfork_done = &vfork;
2438                 init_completion(&vfork);
2439                 get_task_struct(p);
2440         }
2441
2442         wake_up_new_task(p);
2443
2444         /* forking complete and child started to run, tell ptracer */
2445         if (unlikely(trace))
2446                 ptrace_event_pid(trace, pid);
2447
2448         if (clone_flags & CLONE_VFORK) {
2449                 if (!wait_for_vfork_done(p, &vfork))
2450                         ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2451         }
2452
2453         put_pid(pid);
2454         return nr;
2455 }
2456
2457 bool legacy_clone_args_valid(const struct kernel_clone_args *kargs)
2458 {
2459         /* clone(CLONE_PIDFD) uses parent_tidptr to return a pidfd */
2460         if ((kargs->flags & CLONE_PIDFD) &&
2461             (kargs->flags & CLONE_PARENT_SETTID))
2462                 return false;
2463
2464         return true;
2465 }
2466
2467 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2468 /* For compatibility with architectures that call do_fork directly rather than
2469  * using the syscall entry points below. */
2470 long do_fork(unsigned long clone_flags,
2471               unsigned long stack_start,
2472               unsigned long stack_size,
2473               int __user *parent_tidptr,
2474               int __user *child_tidptr)
2475 {
2476         struct kernel_clone_args args = {
2477                 .flags          = (clone_flags & ~CSIGNAL),
2478                 .pidfd          = parent_tidptr,
2479                 .child_tid      = child_tidptr,
2480                 .parent_tid     = parent_tidptr,
2481                 .exit_signal    = (clone_flags & CSIGNAL),
2482                 .stack          = stack_start,
2483                 .stack_size     = stack_size,
2484         };
2485
2486         if (!legacy_clone_args_valid(&args))
2487                 return -EINVAL;
2488
2489         return _do_fork(&args);
2490 }
2491 #endif
2492
2493 /*
2494  * Create a kernel thread.
2495  */
2496 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2497 {
2498         struct kernel_clone_args args = {
2499                 .flags          = ((flags | CLONE_VM | CLONE_UNTRACED) & ~CSIGNAL),
2500                 .exit_signal    = (flags & CSIGNAL),
2501                 .stack          = (unsigned long)fn,
2502                 .stack_size     = (unsigned long)arg,
2503         };
2504
2505         return _do_fork(&args);
2506 }
2507
2508 #ifdef __ARCH_WANT_SYS_FORK
2509 SYSCALL_DEFINE0(fork)
2510 {
2511 #ifdef CONFIG_MMU
2512         struct kernel_clone_args args = {
2513                 .exit_signal = SIGCHLD,
2514         };
2515
2516         return _do_fork(&args);
2517 #else
2518         /* can not support in nommu mode */
2519         return -EINVAL;
2520 #endif
2521 }
2522 #endif
2523
2524 #ifdef __ARCH_WANT_SYS_VFORK
2525 SYSCALL_DEFINE0(vfork)
2526 {
2527         struct kernel_clone_args args = {
2528                 .flags          = CLONE_VFORK | CLONE_VM,
2529                 .exit_signal    = SIGCHLD,
2530         };
2531
2532         return _do_fork(&args);
2533 }
2534 #endif
2535
2536 #ifdef __ARCH_WANT_SYS_CLONE
2537 #ifdef CONFIG_CLONE_BACKWARDS
2538 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2539                  int __user *, parent_tidptr,
2540                  unsigned long, tls,
2541                  int __user *, child_tidptr)
2542 #elif defined(CONFIG_CLONE_BACKWARDS2)
2543 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2544                  int __user *, parent_tidptr,
2545                  int __user *, child_tidptr,
2546                  unsigned long, tls)
2547 #elif defined(CONFIG_CLONE_BACKWARDS3)
2548 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2549                 int, stack_size,
2550                 int __user *, parent_tidptr,
2551                 int __user *, child_tidptr,
2552                 unsigned long, tls)
2553 #else
2554 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2555                  int __user *, parent_tidptr,
2556                  int __user *, child_tidptr,
2557                  unsigned long, tls)
2558 #endif
2559 {
2560         struct kernel_clone_args args = {
2561                 .flags          = (clone_flags & ~CSIGNAL),
2562                 .pidfd          = parent_tidptr,
2563                 .child_tid      = child_tidptr,
2564                 .parent_tid     = parent_tidptr,
2565                 .exit_signal    = (clone_flags & CSIGNAL),
2566                 .stack          = newsp,
2567                 .tls            = tls,
2568         };
2569
2570         if (!legacy_clone_args_valid(&args))
2571                 return -EINVAL;
2572
2573         return _do_fork(&args);
2574 }
2575 #endif
2576
2577 #ifdef __ARCH_WANT_SYS_CLONE3
2578 noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2579                                               struct clone_args __user *uargs,
2580                                               size_t usize)
2581 {
2582         int err;
2583         struct clone_args args;
2584         pid_t *kset_tid = kargs->set_tid;
2585
2586         if (unlikely(usize > PAGE_SIZE))
2587                 return -E2BIG;
2588         if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2589                 return -EINVAL;
2590
2591         err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2592         if (err)
2593                 return err;
2594
2595         if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2596                 return -EINVAL;
2597
2598         if (unlikely(!args.set_tid && args.set_tid_size > 0))
2599                 return -EINVAL;
2600
2601         if (unlikely(args.set_tid && args.set_tid_size == 0))
2602                 return -EINVAL;
2603
2604         /*
2605          * Verify that higher 32bits of exit_signal are unset and that
2606          * it is a valid signal
2607          */
2608         if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2609                      !valid_signal(args.exit_signal)))
2610                 return -EINVAL;
2611
2612         *kargs = (struct kernel_clone_args){
2613                 .flags          = args.flags,
2614                 .pidfd          = u64_to_user_ptr(args.pidfd),
2615                 .child_tid      = u64_to_user_ptr(args.child_tid),
2616                 .parent_tid     = u64_to_user_ptr(args.parent_tid),
2617                 .exit_signal    = args.exit_signal,
2618                 .stack          = args.stack,
2619                 .stack_size     = args.stack_size,
2620                 .tls            = args.tls,
2621                 .set_tid_size   = args.set_tid_size,
2622         };
2623
2624         if (args.set_tid &&
2625                 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
2626                         (kargs->set_tid_size * sizeof(pid_t))))
2627                 return -EFAULT;
2628
2629         kargs->set_tid = kset_tid;
2630
2631         return 0;
2632 }
2633
2634 /**
2635  * clone3_stack_valid - check and prepare stack
2636  * @kargs: kernel clone args
2637  *
2638  * Verify that the stack arguments userspace gave us are sane.
2639  * In addition, set the stack direction for userspace since it's easy for us to
2640  * determine.
2641  */
2642 static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
2643 {
2644         if (kargs->stack == 0) {
2645                 if (kargs->stack_size > 0)
2646                         return false;
2647         } else {
2648                 if (kargs->stack_size == 0)
2649                         return false;
2650
2651                 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
2652                         return false;
2653
2654 #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2655                 kargs->stack += kargs->stack_size;
2656 #endif
2657         }
2658
2659         return true;
2660 }
2661
2662 static bool clone3_args_valid(struct kernel_clone_args *kargs)
2663 {
2664         /* Verify that no unknown flags are passed along. */
2665         if (kargs->flags & ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND))
2666                 return false;
2667
2668         /*
2669          * - make the CLONE_DETACHED bit reuseable for clone3
2670          * - make the CSIGNAL bits reuseable for clone3
2671          */
2672         if (kargs->flags & (CLONE_DETACHED | CSIGNAL))
2673                 return false;
2674
2675         if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
2676             (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
2677                 return false;
2678
2679         if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
2680             kargs->exit_signal)
2681                 return false;
2682
2683         if (!clone3_stack_valid(kargs))
2684                 return false;
2685
2686         return true;
2687 }
2688
2689 /**
2690  * clone3 - create a new process with specific properties
2691  * @uargs: argument structure
2692  * @size:  size of @uargs
2693  *
2694  * clone3() is the extensible successor to clone()/clone2().
2695  * It takes a struct as argument that is versioned by its size.
2696  *
2697  * Return: On success, a positive PID for the child process.
2698  *         On error, a negative errno number.
2699  */
2700 SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
2701 {
2702         int err;
2703
2704         struct kernel_clone_args kargs;
2705         pid_t set_tid[MAX_PID_NS_LEVEL];
2706
2707         kargs.set_tid = set_tid;
2708
2709         err = copy_clone_args_from_user(&kargs, uargs, size);
2710         if (err)
2711                 return err;
2712
2713         if (!clone3_args_valid(&kargs))
2714                 return -EINVAL;
2715
2716         return _do_fork(&kargs);
2717 }
2718 #endif
2719
2720 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2721 {
2722         struct task_struct *leader, *parent, *child;
2723         int res;
2724
2725         read_lock(&tasklist_lock);
2726         leader = top = top->group_leader;
2727 down:
2728         for_each_thread(leader, parent) {
2729                 list_for_each_entry(child, &parent->children, sibling) {
2730                         res = visitor(child, data);
2731                         if (res) {
2732                                 if (res < 0)
2733                                         goto out;
2734                                 leader = child;
2735                                 goto down;
2736                         }
2737 up:
2738                         ;
2739                 }
2740         }
2741
2742         if (leader != top) {
2743                 child = leader;
2744                 parent = child->real_parent;
2745                 leader = parent->group_leader;
2746                 goto up;
2747         }
2748 out:
2749         read_unlock(&tasklist_lock);
2750 }
2751
2752 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2753 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2754 #endif
2755
2756 static void sighand_ctor(void *data)
2757 {
2758         struct sighand_struct *sighand = data;
2759
2760         spin_lock_init(&sighand->siglock);
2761         init_waitqueue_head(&sighand->signalfd_wqh);
2762 }
2763
2764 void __init proc_caches_init(void)
2765 {
2766         unsigned int mm_size;
2767
2768         sighand_cachep = kmem_cache_create("sighand_cache",
2769                         sizeof(struct sighand_struct), 0,
2770                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2771                         SLAB_ACCOUNT, sighand_ctor);
2772         signal_cachep = kmem_cache_create("signal_cache",
2773                         sizeof(struct signal_struct), 0,
2774                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2775                         NULL);
2776         files_cachep = kmem_cache_create("files_cache",
2777                         sizeof(struct files_struct), 0,
2778                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2779                         NULL);
2780         fs_cachep = kmem_cache_create("fs_cache",
2781                         sizeof(struct fs_struct), 0,
2782                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2783                         NULL);
2784
2785         /*
2786          * The mm_cpumask is located at the end of mm_struct, and is
2787          * dynamically sized based on the maximum CPU number this system
2788          * can have, taking hotplug into account (nr_cpu_ids).
2789          */
2790         mm_size = sizeof(struct mm_struct) + cpumask_size();
2791
2792         mm_cachep = kmem_cache_create_usercopy("mm_struct",
2793                         mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2794                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2795                         offsetof(struct mm_struct, saved_auxv),
2796                         sizeof_field(struct mm_struct, saved_auxv),
2797                         NULL);
2798         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2799         mmap_init();
2800         nsproxy_cache_init();
2801 }
2802
2803 /*
2804  * Check constraints on flags passed to the unshare system call.
2805  */
2806 static int check_unshare_flags(unsigned long unshare_flags)
2807 {
2808         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2809                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2810                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2811                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2812                 return -EINVAL;
2813         /*
2814          * Not implemented, but pretend it works if there is nothing
2815          * to unshare.  Note that unsharing the address space or the
2816          * signal handlers also need to unshare the signal queues (aka
2817          * CLONE_THREAD).
2818          */
2819         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2820                 if (!thread_group_empty(current))
2821                         return -EINVAL;
2822         }
2823         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2824                 if (refcount_read(&current->sighand->count) > 1)
2825                         return -EINVAL;
2826         }
2827         if (unshare_flags & CLONE_VM) {
2828                 if (!current_is_single_threaded())
2829                         return -EINVAL;
2830         }
2831
2832         return 0;
2833 }
2834
2835 /*
2836  * Unshare the filesystem structure if it is being shared
2837  */
2838 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2839 {
2840         struct fs_struct *fs = current->fs;
2841
2842         if (!(unshare_flags & CLONE_FS) || !fs)
2843                 return 0;
2844
2845         /* don't need lock here; in the worst case we'll do useless copy */
2846         if (fs->users == 1)
2847                 return 0;
2848
2849         *new_fsp = copy_fs_struct(fs);
2850         if (!*new_fsp)
2851                 return -ENOMEM;
2852
2853         return 0;
2854 }
2855
2856 /*
2857  * Unshare file descriptor table if it is being shared
2858  */
2859 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2860 {
2861         struct files_struct *fd = current->files;
2862         int error = 0;
2863
2864         if ((unshare_flags & CLONE_FILES) &&
2865             (fd && atomic_read(&fd->count) > 1)) {
2866                 *new_fdp = dup_fd(fd, &error);
2867                 if (!*new_fdp)
2868                         return error;
2869         }
2870
2871         return 0;
2872 }
2873
2874 /*
2875  * unshare allows a process to 'unshare' part of the process
2876  * context which was originally shared using clone.  copy_*
2877  * functions used by do_fork() cannot be used here directly
2878  * because they modify an inactive task_struct that is being
2879  * constructed. Here we are modifying the current, active,
2880  * task_struct.
2881  */
2882 int ksys_unshare(unsigned long unshare_flags)
2883 {
2884         struct fs_struct *fs, *new_fs = NULL;
2885         struct files_struct *fd, *new_fd = NULL;
2886         struct cred *new_cred = NULL;
2887         struct nsproxy *new_nsproxy = NULL;
2888         int do_sysvsem = 0;
2889         int err;
2890
2891         /*
2892          * If unsharing a user namespace must also unshare the thread group
2893          * and unshare the filesystem root and working directories.
2894          */
2895         if (unshare_flags & CLONE_NEWUSER)
2896                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2897         /*
2898          * If unsharing vm, must also unshare signal handlers.
2899          */
2900         if (unshare_flags & CLONE_VM)
2901                 unshare_flags |= CLONE_SIGHAND;
2902         /*
2903          * If unsharing a signal handlers, must also unshare the signal queues.
2904          */
2905         if (unshare_flags & CLONE_SIGHAND)
2906                 unshare_flags |= CLONE_THREAD;
2907         /*
2908          * If unsharing namespace, must also unshare filesystem information.
2909          */
2910         if (unshare_flags & CLONE_NEWNS)
2911                 unshare_flags |= CLONE_FS;
2912
2913         err = check_unshare_flags(unshare_flags);
2914         if (err)
2915                 goto bad_unshare_out;
2916         /*
2917          * CLONE_NEWIPC must also detach from the undolist: after switching
2918          * to a new ipc namespace, the semaphore arrays from the old
2919          * namespace are unreachable.
2920          */
2921         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2922                 do_sysvsem = 1;
2923         err = unshare_fs(unshare_flags, &new_fs);
2924         if (err)
2925                 goto bad_unshare_out;
2926         err = unshare_fd(unshare_flags, &new_fd);
2927         if (err)
2928                 goto bad_unshare_cleanup_fs;
2929         err = unshare_userns(unshare_flags, &new_cred);
2930         if (err)
2931                 goto bad_unshare_cleanup_fd;
2932         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2933                                          new_cred, new_fs);
2934         if (err)
2935                 goto bad_unshare_cleanup_cred;
2936
2937         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2938                 if (do_sysvsem) {
2939                         /*
2940                          * CLONE_SYSVSEM is equivalent to sys_exit().
2941                          */
2942                         exit_sem(current);
2943                 }
2944                 if (unshare_flags & CLONE_NEWIPC) {
2945                         /* Orphan segments in old ns (see sem above). */
2946                         exit_shm(current);
2947                         shm_init_task(current);
2948                 }
2949
2950                 if (new_nsproxy)
2951                         switch_task_namespaces(current, new_nsproxy);
2952
2953                 task_lock(current);
2954
2955                 if (new_fs) {
2956                         fs = current->fs;
2957                         spin_lock(&fs->lock);
2958                         current->fs = new_fs;
2959                         if (--fs->users)
2960                                 new_fs = NULL;
2961                         else
2962                                 new_fs = fs;
2963                         spin_unlock(&fs->lock);
2964                 }
2965
2966                 if (new_fd) {
2967                         fd = current->files;
2968                         current->files = new_fd;
2969                         new_fd = fd;
2970                 }
2971
2972                 task_unlock(current);
2973
2974                 if (new_cred) {
2975                         /* Install the new user namespace */
2976                         commit_creds(new_cred);
2977                         new_cred = NULL;
2978                 }
2979         }
2980
2981         perf_event_namespaces(current);
2982
2983 bad_unshare_cleanup_cred:
2984         if (new_cred)
2985                 put_cred(new_cred);
2986 bad_unshare_cleanup_fd:
2987         if (new_fd)
2988                 put_files_struct(new_fd);
2989
2990 bad_unshare_cleanup_fs:
2991         if (new_fs)
2992                 free_fs_struct(new_fs);
2993
2994 bad_unshare_out:
2995         return err;
2996 }
2997
2998 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2999 {
3000         return ksys_unshare(unshare_flags);
3001 }
3002
3003 /*
3004  *      Helper to unshare the files of the current task.
3005  *      We don't want to expose copy_files internals to
3006  *      the exec layer of the kernel.
3007  */
3008
3009 int unshare_files(struct files_struct **displaced)
3010 {
3011         struct task_struct *task = current;
3012         struct files_struct *copy = NULL;
3013         int error;
3014
3015         error = unshare_fd(CLONE_FILES, &copy);
3016         if (error || !copy) {
3017                 *displaced = NULL;
3018                 return error;
3019         }
3020         *displaced = task->files;
3021         task_lock(task);
3022         task->files = copy;
3023         task_unlock(task);
3024         return 0;
3025 }
3026
3027 int sysctl_max_threads(struct ctl_table *table, int write,
3028                        void __user *buffer, size_t *lenp, loff_t *ppos)
3029 {
3030         struct ctl_table t;
3031         int ret;
3032         int threads = max_threads;
3033         int min = 1;
3034         int max = MAX_THREADS;
3035
3036         t = *table;
3037         t.data = &threads;
3038         t.extra1 = &min;
3039         t.extra2 = &max;
3040
3041         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3042         if (ret || !write)
3043                 return ret;
3044
3045         max_threads = threads;
3046
3047         return 0;
3048 }