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/fs.h>
  44 #include <linux/mm.h>
  45 #include <linux/vmacache.h>
  46 #include <linux/nsproxy.h>
  47 #include <linux/capability.h>
  48 #include <linux/cpu.h>
  49 #include <linux/cgroup.h>
  50 #include <linux/security.h>
  51 #include <linux/hugetlb.h>
  52 #include <linux/seccomp.h>
  53 #include <linux/swap.h>
  54 #include <linux/syscalls.h>
  55 #include <linux/jiffies.h>
  56 #include <linux/futex.h>
  57 #include <linux/compat.h>
  58 #include <linux/kthread.h>
  59 #include <linux/task_io_accounting_ops.h>
  60 #include <linux/rcupdate.h>
  61 #include <linux/ptrace.h>
  62 #include <linux/mount.h>
  63 #include <linux/audit.h>
  64 #include <linux/memcontrol.h>
  65 #include <linux/ftrace.h>
  66 #include <linux/proc_fs.h>
  67 #include <linux/profile.h>
  68 #include <linux/rmap.h>
  69 #include <linux/ksm.h>
  70 #include <linux/acct.h>
  71 #include <linux/userfaultfd_k.h>
  72 #include <linux/tsacct_kern.h>
  73 #include <linux/cn_proc.h>
  74 #include <linux/freezer.h>
  75 #include <linux/delayacct.h>
  76 #include <linux/taskstats_kern.h>
  77 #include <linux/random.h>
  78 #include <linux/tty.h>
  79 #include <linux/blkdev.h>
  80 #include <linux/fs_struct.h>
  81 #include <linux/magic.h>
  82 #include <linux/perf_event.h>
  83 #include <linux/posix-timers.h>
  84 #include <linux/user-return-notifier.h>
  85 #include <linux/oom.h>
  86 #include <linux/khugepaged.h>
  87 #include <linux/signalfd.h>
  88 #include <linux/uprobes.h>
  89 #include <linux/aio.h>
  90 #include <linux/compiler.h>
  91 #include <linux/sysctl.h>
  92 #include <linux/kcov.h>
  93 #include <linux/livepatch.h>
  94 #include <linux/thread_info.h>
  95 #include <linux/stackleak.h>
  96
  97 #include <asm/pgtable.h>
  98 #include <asm/pgalloc.h>
  99 #include <linux/uaccess.h>
 100 #include <asm/mmu_context.h>
 101 #include <asm/cacheflush.h>
 102 #include <asm/tlbflush.h>
 103
 104 #include <trace/events/sched.h>
 105
 106 #define CREATE_TRACE_POINTS
 107 #include <trace/events/task.h>
 108
 109 /*
 110  * Minimum number of threads to boot the kernel
 111  */
 112 #define MIN_THREADS 20
 113
 114 /*
 115  * Maximum number of threads
 116  */
 117 #define MAX_THREADS FUTEX_TID_MASK
 118
 119 /*
 120  * Protected counters by write_lock_irq(&tasklist_lock)
 121  */
 122 unsigned long total_forks;      /* Handle normal Linux uptimes. */
 123 int nr_threads;                 /* The idle threads do not count.. */
 124
 125 static int max_threads;         /* tunable limit on nr_threads */
 126
 127 #define NAMED_ARRAY_INDEX(x)    [x] = __stringify(x)
 128
 129 static const char * const resident_page_types[] = {
 130         NAMED_ARRAY_INDEX(MM_FILEPAGES),
 131         NAMED_ARRAY_INDEX(MM_ANONPAGES),
 132         NAMED_ARRAY_INDEX(MM_SWAPENTS),
 133         NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
 134 };
 135
 136 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 137
 138 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 139
 140 #ifdef CONFIG_PROVE_RCU
 141 int lockdep_tasklist_lock_is_held(void)
 142 {
 143         return lockdep_is_held(&tasklist_lock);
 144 }
 145 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 146 #endif /* #ifdef CONFIG_PROVE_RCU */
 147
 148 int nr_processes(void)
 149 {
 150         int cpu;
 151         int total = 0;
 152
 153         for_each_possible_cpu(cpu)
 154                 total += per_cpu(process_counts, cpu);
 155
 156         return total;
 157 }
 158
 159 void __weak arch_release_task_struct(struct task_struct *tsk)
 160 {
 161 }
 162
 163 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 164 static struct kmem_cache *task_struct_cachep;
 165
 166 static inline struct task_struct *alloc_task_struct_node(int node)
 167 {
 168         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 169 }
 170
 171 static inline void free_task_struct(struct task_struct *tsk)
 172 {
 173         kmem_cache_free(task_struct_cachep, tsk);
 174 }
 175 #endif
 176
 177 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
 178
 179 /*
 180  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 181  * kmemcache based allocator.
 182  */
 183 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
 184
 185 #ifdef CONFIG_VMAP_STACK
 186 /*
 187  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
 188  * flush.  Try to minimize the number of calls by caching stacks.
 189  */
 190 #define NR_CACHED_STACKS 2
 191 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
 192
 193 static int free_vm_stack_cache(unsigned int cpu)
 194 {
 195         struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
 196         int i;
 197
 198         for (i = 0; i < NR_CACHED_STACKS; i++) {
 199                 struct vm_struct *vm_stack = cached_vm_stacks[i];
 200
 201                 if (!vm_stack)
 202                         continue;
 203
 204                 vfree(vm_stack->addr);
 205                 cached_vm_stacks[i] = NULL;
 206         }
 207
 208         return 0;
 209 }
 210 #endif
 211
 212 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
 213 {
 214 #ifdef CONFIG_VMAP_STACK
 215         void *stack;
 216         int i;
 217
 218         for (i = 0; i < NR_CACHED_STACKS; i++) {
 219                 struct vm_struct *s;
 220
 221                 s = this_cpu_xchg(cached_stacks[i], NULL);
 222
 223                 if (!s)
 224                         continue;
 225
 226                 /* Clear stale pointers from reused stack. */
 227                 memset(s->addr, 0, THREAD_SIZE);
 228
 229                 tsk->stack_vm_area = s;
 230                 tsk->stack = s->addr;
 231                 return s->addr;
 232         }
 233
 234         /*
 235          * Allocated stacks are cached and later reused by new threads,
 236          * so memcg accounting is performed manually on assigning/releasing
 237          * stacks to tasks. Drop __GFP_ACCOUNT.
 238          */
 239         stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
 240                                      VMALLOC_START, VMALLOC_END,
 241                                      THREADINFO_GFP & ~__GFP_ACCOUNT,
 242                                      PAGE_KERNEL,
 243                                      0, node, __builtin_return_address(0));
 244
 245         /*
 246          * We can't call find_vm_area() in interrupt context, and
 247          * free_thread_stack() can be called in interrupt context,
 248          * so cache the vm_struct.
 249          */
 250         if (stack) {
 251                 tsk->stack_vm_area = find_vm_area(stack);
 252                 tsk->stack = stack;
 253         }
 254         return stack;
 255 #else
 256         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
 257                                              THREAD_SIZE_ORDER);
 258
 259         if (likely(page)) {
 260                 tsk->stack = page_address(page);
 261                 return tsk->stack;
 262         }
 263         return NULL;
 264 #endif
 265 }
 266
 267 static inline void free_thread_stack(struct task_struct *tsk)
 268 {
 269 #ifdef CONFIG_VMAP_STACK
 270         struct vm_struct *vm = task_stack_vm_area(tsk);
 271
 272         if (vm) {
 273                 int i;
 274
 275                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 276                         mod_memcg_page_state(vm->pages[i],
 277                                              MEMCG_KERNEL_STACK_KB,
 278                                              -(int)(PAGE_SIZE / 1024));
 279
 280                         memcg_kmem_uncharge(vm->pages[i], 0);
 281                 }
 282
 283                 for (i = 0; i < NR_CACHED_STACKS; i++) {
 284                         if (this_cpu_cmpxchg(cached_stacks[i],
 285                                         NULL, tsk->stack_vm_area) != NULL)
 286                                 continue;
 287
 288                         return;
 289                 }
 290
 291                 vfree_atomic(tsk->stack);
 292                 return;
 293         }
 294 #endif
 295
 296         __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
 297 }
 298 # else
 299 static struct kmem_cache *thread_stack_cache;
 300
 301 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
 302                                                   int node)
 303 {
 304         unsigned long *stack;
 305         stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
 306         tsk->stack = stack;
 307         return stack;
 308 }
 309
 310 static void free_thread_stack(struct task_struct *tsk)
 311 {
 312         kmem_cache_free(thread_stack_cache, tsk->stack);
 313 }
 314
 315 void thread_stack_cache_init(void)
 316 {
 317         thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
 318                                         THREAD_SIZE, THREAD_SIZE, 0, 0,
 319                                         THREAD_SIZE, NULL);
 320         BUG_ON(thread_stack_cache == NULL);
 321 }
 322 # endif
 323 #endif
 324
 325 /* SLAB cache for signal_struct structures (tsk->signal) */
 326 static struct kmem_cache *signal_cachep;
 327
 328 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 329 struct kmem_cache *sighand_cachep;
 330
 331 /* SLAB cache for files_struct structures (tsk->files) */
 332 struct kmem_cache *files_cachep;
 333
 334 /* SLAB cache for fs_struct structures (tsk->fs) */
 335 struct kmem_cache *fs_cachep;
 336
 337 /* SLAB cache for vm_area_struct structures */
 338 static struct kmem_cache *vm_area_cachep;
 339
 340 /* SLAB cache for mm_struct structures (tsk->mm) */
 341 static struct kmem_cache *mm_cachep;
 342
 343 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
 344 {
 345         struct vm_area_struct *vma;
 346
 347         vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 348         if (vma)
 349                 vma_init(vma, mm);
 350         return vma;
 351 }
 352
 353 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
 354 {
 355         struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 356
 357         if (new) {
 358                 *new = *orig;
 359                 INIT_LIST_HEAD(&new->anon_vma_chain);
 360         }
 361         return new;
 362 }
 363
 364 void vm_area_free(struct vm_area_struct *vma)
 365 {
 366         kmem_cache_free(vm_area_cachep, vma);
 367 }
 368
 369 static void account_kernel_stack(struct task_struct *tsk, int account)
 370 {
 371         void *stack = task_stack_page(tsk);
 372         struct vm_struct *vm = task_stack_vm_area(tsk);
 373
 374         BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
 375
 376         if (vm) {
 377                 int i;
 378
 379                 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
 380
 381                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 382                         mod_zone_page_state(page_zone(vm->pages[i]),
 383                                             NR_KERNEL_STACK_KB,
 384                                             PAGE_SIZE / 1024 * account);
 385                 }
 386         } else {
 387                 /*
 388                  * All stack pages are in the same zone and belong to the
 389                  * same memcg.
 390                  */
 391                 struct page *first_page = virt_to_page(stack);
 392
 393                 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
 394                                     THREAD_SIZE / 1024 * account);
 395
 396                 mod_memcg_page_state(first_page, MEMCG_KERNEL_STACK_KB,
 397                                      account * (THREAD_SIZE / 1024));
 398         }
 399 }
 400
 401 static int memcg_charge_kernel_stack(struct task_struct *tsk)
 402 {
 403 #ifdef CONFIG_VMAP_STACK
 404         struct vm_struct *vm = task_stack_vm_area(tsk);
 405         int ret;
 406
 407         if (vm) {
 408                 int i;
 409
 410                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 411                         /*
 412                          * If memcg_kmem_charge() fails, page->mem_cgroup
 413                          * pointer is NULL, and both memcg_kmem_uncharge()
 414                          * and mod_memcg_page_state() in free_thread_stack()
 415                          * will ignore this page. So it's safe.
 416                          */
 417                         ret = memcg_kmem_charge(vm->pages[i], GFP_KERNEL, 0);
 418                         if (ret)
 419                                 return ret;
 420
 421                         mod_memcg_page_state(vm->pages[i],
 422                                              MEMCG_KERNEL_STACK_KB,
 423                                              PAGE_SIZE / 1024);
 424                 }
 425         }
 426 #endif
 427         return 0;
 428 }
 429
 430 static void release_task_stack(struct task_struct *tsk)
 431 {
 432         if (WARN_ON(tsk->state != TASK_DEAD))
 433                 return;  /* Better to leak the stack than to free prematurely */
 434
 435         account_kernel_stack(tsk, -1);
 436         free_thread_stack(tsk);
 437         tsk->stack = NULL;
 438 #ifdef CONFIG_VMAP_STACK
 439         tsk->stack_vm_area = NULL;
 440 #endif
 441 }
 442
 443 #ifdef CONFIG_THREAD_INFO_IN_TASK
 444 void put_task_stack(struct task_struct *tsk)
 445 {
 446         if (refcount_dec_and_test(&tsk->stack_refcount))
 447                 release_task_stack(tsk);
 448 }
 449 #endif
 450
 451 void free_task(struct task_struct *tsk)
 452 {
 453 #ifndef CONFIG_THREAD_INFO_IN_TASK
 454         /*
 455          * The task is finally done with both the stack and thread_info,
 456          * so free both.
 457          */
 458         release_task_stack(tsk);
 459 #else
 460         /*
 461          * If the task had a separate stack allocation, it should be gone
 462          * by now.
 463          */
 464         WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
 465 #endif
 466         rt_mutex_debug_task_free(tsk);
 467         ftrace_graph_exit_task(tsk);
 468         put_seccomp_filter(tsk);
 469         arch_release_task_struct(tsk);
 470         if (tsk->flags & PF_KTHREAD)
 471                 free_kthread_struct(tsk);
 472         free_task_struct(tsk);
 473 }
 474 EXPORT_SYMBOL(free_task);
 475
 476 #ifdef CONFIG_MMU
 477 static __latent_entropy int dup_mmap(struct mm_struct *mm,
 478                                         struct mm_struct *oldmm)
 479 {
 480         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 481         struct rb_node **rb_link, *rb_parent;
 482         int retval;
 483         unsigned long charge;
 484         LIST_HEAD(uf);
 485
 486         uprobe_start_dup_mmap();
 487         if (down_write_killable(&oldmm->mmap_sem)) {
 488                 retval = -EINTR;
 489                 goto fail_uprobe_end;
 490         }
 491         flush_cache_dup_mm(oldmm);
 492         uprobe_dup_mmap(oldmm, mm);
 493         /*
 494          * Not linked in yet - no deadlock potential:
 495          */
 496         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 497
 498         /* No ordering required: file already has been exposed. */
 499         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 500
 501         mm->total_vm = oldmm->total_vm;
 502         mm->data_vm = oldmm->data_vm;
 503         mm->exec_vm = oldmm->exec_vm;
 504         mm->stack_vm = oldmm->stack_vm;
 505
 506         rb_link = &mm->mm_rb.rb_node;
 507         rb_parent = NULL;
 508         pprev = &mm->mmap;
 509         retval = ksm_fork(mm, oldmm);
 510         if (retval)
 511                 goto out;
 512         retval = khugepaged_fork(mm, oldmm);
 513         if (retval)
 514                 goto out;
 515
 516         prev = NULL;
 517         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 518                 struct file *file;
 519
 520                 if (mpnt->vm_flags & VM_DONTCOPY) {
 521                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 522                         continue;
 523                 }
 524                 charge = 0;
 525                 /*
 526                  * Don't duplicate many vmas if we've been oom-killed (for
 527                  * example)
 528                  */
 529                 if (fatal_signal_pending(current)) {
 530                         retval = -EINTR;
 531                         goto out;
 532                 }
 533                 if (mpnt->vm_flags & VM_ACCOUNT) {
 534                         unsigned long len = vma_pages(mpnt);
 535
 536                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 537                                 goto fail_nomem;
 538                         charge = len;
 539                 }
 540                 tmp = vm_area_dup(mpnt);
 541                 if (!tmp)
 542                         goto fail_nomem;
 543                 retval = vma_dup_policy(mpnt, tmp);
 544                 if (retval)
 545                         goto fail_nomem_policy;
 546                 tmp->vm_mm = mm;
 547                 retval = dup_userfaultfd(tmp, &uf);
 548                 if (retval)
 549                         goto fail_nomem_anon_vma_fork;
 550                 if (tmp->vm_flags & VM_WIPEONFORK) {
 551                         /* VM_WIPEONFORK gets a clean slate in the child. */
 552                         tmp->anon_vma = NULL;
 553                         if (anon_vma_prepare(tmp))
 554                                 goto fail_nomem_anon_vma_fork;
 555                 } else if (anon_vma_fork(tmp, mpnt))
 556                         goto fail_nomem_anon_vma_fork;
 557                 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
 558                 tmp->vm_next = tmp->vm_prev = NULL;
 559                 file = tmp->vm_file;
 560                 if (file) {
 561                         struct inode *inode = file_inode(file);
 562                         struct address_space *mapping = file->f_mapping;
 563
 564                         get_file(file);
 565                         if (tmp->vm_flags & VM_DENYWRITE)
 566                                 atomic_dec(&inode->i_writecount);
 567                         i_mmap_lock_write(mapping);
 568                         if (tmp->vm_flags & VM_SHARED)
 569                                 atomic_inc(&mapping->i_mmap_writable);
 570                         flush_dcache_mmap_lock(mapping);
 571                         /* insert tmp into the share list, just after mpnt */
 572                         vma_interval_tree_insert_after(tmp, mpnt,
 573                                         &mapping->i_mmap);
 574                         flush_dcache_mmap_unlock(mapping);
 575                         i_mmap_unlock_write(mapping);
 576                 }
 577
 578                 /*
 579                  * Clear hugetlb-related page reserves for children. This only
 580                  * affects MAP_PRIVATE mappings. Faults generated by the child
 581                  * are not guaranteed to succeed, even if read-only
 582                  */
 583                 if (is_vm_hugetlb_page(tmp))
 584                         reset_vma_resv_huge_pages(tmp);
 585
 586                 /*
 587                  * Link in the new vma and copy the page table entries.
 588                  */
 589                 *pprev = tmp;
 590                 pprev = &tmp->vm_next;
 591                 tmp->vm_prev = prev;
 592                 prev = tmp;
 593
 594                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 595                 rb_link = &tmp->vm_rb.rb_right;
 596                 rb_parent = &tmp->vm_rb;
 597
 598                 mm->map_count++;
 599                 if (!(tmp->vm_flags & VM_WIPEONFORK))
 600                         retval = copy_page_range(mm, oldmm, mpnt);
 601
 602                 if (tmp->vm_ops && tmp->vm_ops->open)
 603                         tmp->vm_ops->open(tmp);
 604
 605                 if (retval)
 606                         goto out;
 607         }
 608         /* a new mm has just been created */
 609         retval = arch_dup_mmap(oldmm, mm);
 610 out:
 611         up_write(&mm->mmap_sem);
 612         flush_tlb_mm(oldmm);
 613         up_write(&oldmm->mmap_sem);
 614         dup_userfaultfd_complete(&uf);
 615 fail_uprobe_end:
 616         uprobe_end_dup_mmap();
 617         return retval;
 618 fail_nomem_anon_vma_fork:
 619         mpol_put(vma_policy(tmp));
 620 fail_nomem_policy:
 621         vm_area_free(tmp);
 622 fail_nomem:
 623         retval = -ENOMEM;
 624         vm_unacct_memory(charge);
 625         goto out;
 626 }
 627
 628 static inline int mm_alloc_pgd(struct mm_struct *mm)
 629 {
 630         mm->pgd = pgd_alloc(mm);
 631         if (unlikely(!mm->pgd))
 632                 return -ENOMEM;
 633         return 0;
 634 }
 635
 636 static inline void mm_free_pgd(struct mm_struct *mm)
 637 {
 638         pgd_free(mm, mm->pgd);
 639 }
 640 #else
 641 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 642 {
 643         down_write(&oldmm->mmap_sem);
 644         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 645         up_write(&oldmm->mmap_sem);
 646         return 0;
 647 }
 648 #define mm_alloc_pgd(mm)        (0)
 649 #define mm_free_pgd(mm)
 650 #endif /* CONFIG_MMU */
 651
 652 static void check_mm(struct mm_struct *mm)
 653 {
 654         int i;
 655
 656         BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
 657                          "Please make sure 'struct resident_page_types[]' is updated as well");
 658
 659         for (i = 0; i < NR_MM_COUNTERS; i++) {
 660                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 661
 662                 if (unlikely(x))
 663                         pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
 664                                  mm, resident_page_types[i], x);
 665         }
 666
 667         if (mm_pgtables_bytes(mm))
 668                 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
 669                                 mm_pgtables_bytes(mm));
 670
 671 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 672         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 673 #endif
 674 }
 675
 676 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 677 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 678
 679 /*
 680  * Called when the last reference to the mm
 681  * is dropped: either by a lazy thread or by
 682  * mmput. Free the page directory and the mm.
 683  */
 684 void __mmdrop(struct mm_struct *mm)
 685 {
 686         BUG_ON(mm == &init_mm);
 687         WARN_ON_ONCE(mm == current->mm);
 688         WARN_ON_ONCE(mm == current->active_mm);
 689         mm_free_pgd(mm);
 690         destroy_context(mm);
 691         mmu_notifier_mm_destroy(mm);
 692         check_mm(mm);
 693         put_user_ns(mm->user_ns);
 694         free_mm(mm);
 695 }
 696 EXPORT_SYMBOL_GPL(__mmdrop);
 697
 698 static void mmdrop_async_fn(struct work_struct *work)
 699 {
 700         struct mm_struct *mm;
 701
 702         mm = container_of(work, struct mm_struct, async_put_work);
 703         __mmdrop(mm);
 704 }
 705
 706 static void mmdrop_async(struct mm_struct *mm)
 707 {
 708         if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
 709                 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
 710                 schedule_work(&mm->async_put_work);
 711         }
 712 }
 713
 714 static inline void free_signal_struct(struct signal_struct *sig)
 715 {
 716         taskstats_tgid_free(sig);
 717         sched_autogroup_exit(sig);
 718         /*
 719          * __mmdrop is not safe to call from softirq context on x86 due to
 720          * pgd_dtor so postpone it to the async context
 721          */
 722         if (sig->oom_mm)
 723                 mmdrop_async(sig->oom_mm);
 724         kmem_cache_free(signal_cachep, sig);
 725 }
 726
 727 static inline void put_signal_struct(struct signal_struct *sig)
 728 {
 729         if (refcount_dec_and_test(&sig->sigcnt))
 730                 free_signal_struct(sig);
 731 }
 732
 733 void __put_task_struct(struct task_struct *tsk)
 734 {
 735         WARN_ON(!tsk->exit_state);
 736         WARN_ON(refcount_read(&tsk->usage));
 737         WARN_ON(tsk == current);
 738
 739         cgroup_free(tsk);
 740         task_numa_free(tsk, true);
 741         security_task_free(tsk);
 742         exit_creds(tsk);
 743         delayacct_tsk_free(tsk);
 744         put_signal_struct(tsk->signal);
 745
 746         if (!profile_handoff_task(tsk))
 747                 free_task(tsk);
 748 }
 749 EXPORT_SYMBOL_GPL(__put_task_struct);
 750
 751 void __init __weak arch_task_cache_init(void) { }
 752
 753 /*
 754  * set_max_threads
 755  */
 756 static void set_max_threads(unsigned int max_threads_suggested)
 757 {
 758         u64 threads;
 759         unsigned long nr_pages = totalram_pages();
 760
 761         /*
 762          * The number of threads shall be limited such that the thread
 763          * structures may only consume a small part of the available memory.
 764          */
 765         if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
 766                 threads = MAX_THREADS;
 767         else
 768                 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
 769                                     (u64) THREAD_SIZE * 8UL);
 770
 771         if (threads > max_threads_suggested)
 772                 threads = max_threads_suggested;
 773
 774         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 775 }
 776
 777 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 778 /* Initialized by the architecture: */
 779 int arch_task_struct_size __read_mostly;
 780 #endif
 781
 782 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 783 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
 784 {
 785         /* Fetch thread_struct whitelist for the architecture. */
 786         arch_thread_struct_whitelist(offset, size);
 787
 788         /*
 789          * Handle zero-sized whitelist or empty thread_struct, otherwise
 790          * adjust offset to position of thread_struct in task_struct.
 791          */
 792         if (unlikely(*size == 0))
 793                 *offset = 0;
 794         else
 795                 *offset += offsetof(struct task_struct, thread);
 796 }
 797 #endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
 798
 799 void __init fork_init(void)
 800 {
 801         int i;
 802 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 803 #ifndef ARCH_MIN_TASKALIGN
 804 #define ARCH_MIN_TASKALIGN      0
 805 #endif
 806         int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
 807         unsigned long useroffset, usersize;
 808
 809         /* create a slab on which task_structs can be allocated */
 810         task_struct_whitelist(&useroffset, &usersize);
 811         task_struct_cachep = kmem_cache_create_usercopy("task_struct",
 812                         arch_task_struct_size, align,
 813                         SLAB_PANIC|SLAB_ACCOUNT,
 814                         useroffset, usersize, NULL);
 815 #endif
 816
 817         /* do the arch specific task caches init */
 818         arch_task_cache_init();
 819
 820         set_max_threads(MAX_THREADS);
 821
 822         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 823         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 824         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 825                 init_task.signal->rlim[RLIMIT_NPROC];
 826
 827         for (i = 0; i < UCOUNT_COUNTS; i++) {
 828                 init_user_ns.ucount_max[i] = max_threads/2;
 829         }
 830
 831 #ifdef CONFIG_VMAP_STACK
 832         cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
 833                           NULL, free_vm_stack_cache);
 834 #endif
 835
 836         lockdep_init_task(&init_task);
 837         uprobes_init();
 838 }
 839
 840 int __weak arch_dup_task_struct(struct task_struct *dst,
 841                                                struct task_struct *src)
 842 {
 843         *dst = *src;
 844         return 0;
 845 }
 846
 847 void set_task_stack_end_magic(struct task_struct *tsk)
 848 {
 849         unsigned long *stackend;
 850
 851         stackend = end_of_stack(tsk);
 852         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 853 }
 854
 855 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 856 {
 857         struct task_struct *tsk;
 858         unsigned long *stack;
 859         struct vm_struct *stack_vm_area __maybe_unused;
 860         int err;
 861
 862         if (node == NUMA_NO_NODE)
 863                 node = tsk_fork_get_node(orig);
 864         tsk = alloc_task_struct_node(node);
 865         if (!tsk)
 866                 return NULL;
 867
 868         stack = alloc_thread_stack_node(tsk, node);
 869         if (!stack)
 870                 goto free_tsk;
 871
 872         if (memcg_charge_kernel_stack(tsk))
 873                 goto free_stack;
 874
 875         stack_vm_area = task_stack_vm_area(tsk);
 876
 877         err = arch_dup_task_struct(tsk, orig);
 878
 879         /*
 880          * arch_dup_task_struct() clobbers the stack-related fields.  Make
 881          * sure they're properly initialized before using any stack-related
 882          * functions again.
 883          */
 884         tsk->stack = stack;
 885 #ifdef CONFIG_VMAP_STACK
 886         tsk->stack_vm_area = stack_vm_area;
 887 #endif
 888 #ifdef CONFIG_THREAD_INFO_IN_TASK
 889         refcount_set(&tsk->stack_refcount, 1);
 890 #endif
 891
 892         if (err)
 893                 goto free_stack;
 894
 895 #ifdef CONFIG_SECCOMP
 896         /*
 897          * We must handle setting up seccomp filters once we're under
 898          * the sighand lock in case orig has changed between now and
 899          * then. Until then, filter must be NULL to avoid messing up
 900          * the usage counts on the error path calling free_task.
 901          */
 902         tsk->seccomp.filter = NULL;
 903 #endif
 904
 905         setup_thread_stack(tsk, orig);
 906         clear_user_return_notifier(tsk);
 907         clear_tsk_need_resched(tsk);
 908         set_task_stack_end_magic(tsk);
 909
 910 #ifdef CONFIG_STACKPROTECTOR
 911         tsk->stack_canary = get_random_canary();
 912 #endif
 913         if (orig->cpus_ptr == &orig->cpus_mask)
 914                 tsk->cpus_ptr = &tsk->cpus_mask;
 915
 916         /*
 917          * One for the user space visible state that goes away when reaped.
 918          * One for the scheduler.
 919          */
 920         refcount_set(&tsk->rcu_users, 2);
 921         /* One for the rcu users */
 922         refcount_set(&tsk->usage, 1);
 923 #ifdef CONFIG_BLK_DEV_IO_TRACE
 924         tsk->btrace_seq = 0;
 925 #endif
 926         tsk->splice_pipe = NULL;
 927         tsk->task_frag.page = NULL;
 928         tsk->wake_q.next = NULL;
 929
 930         account_kernel_stack(tsk, 1);
 931
 932         kcov_task_init(tsk);
 933
 934 #ifdef CONFIG_FAULT_INJECTION
 935         tsk->fail_nth = 0;
 936 #endif
 937
 938 #ifdef CONFIG_BLK_CGROUP
 939         tsk->throttle_queue = NULL;
 940         tsk->use_memdelay = 0;
 941 #endif
 942
 943 #ifdef CONFIG_MEMCG
 944         tsk->active_memcg = NULL;
 945 #endif
 946         return tsk;
 947
 948 free_stack:
 949         free_thread_stack(tsk);
 950 free_tsk:
 951         free_task_struct(tsk);
 952         return NULL;
 953 }
 954
 955 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 956
 957 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 958
 959 static int __init coredump_filter_setup(char *s)
 960 {
 961         default_dump_filter =
 962                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 963                 MMF_DUMP_FILTER_MASK;
 964         return 1;
 965 }
 966
 967 __setup("coredump_filter=", coredump_filter_setup);
 968
 969 #include <linux/init_task.h>
 970
 971 static void mm_init_aio(struct mm_struct *mm)
 972 {
 973 #ifdef CONFIG_AIO
 974         spin_lock_init(&mm->ioctx_lock);
 975         mm->ioctx_table = NULL;
 976 #endif
 977 }
 978
 979 static __always_inline void mm_clear_owner(struct mm_struct *mm,
 980                                            struct task_struct *p)
 981 {
 982 #ifdef CONFIG_MEMCG
 983         if (mm->owner == p)
 984                 WRITE_ONCE(mm->owner, NULL);
 985 #endif
 986 }
 987
 988 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 989 {
 990 #ifdef CONFIG_MEMCG
 991         mm->owner = p;
 992 #endif
 993 }
 994
 995 static void mm_init_uprobes_state(struct mm_struct *mm)
 996 {
 997 #ifdef CONFIG_UPROBES
 998         mm->uprobes_state.xol_area = NULL;
 999 #endif
1000 }
1001
1002 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1003         struct user_namespace *user_ns)
1004 {
1005         mm->mmap = NULL;
1006         mm->mm_rb = RB_ROOT;
1007         mm->vmacache_seqnum = 0;
1008         atomic_set(&mm->mm_users, 1);
1009         atomic_set(&mm->mm_count, 1);
1010         init_rwsem(&mm->mmap_sem);
1011         INIT_LIST_HEAD(&mm->mmlist);
1012         mm->core_state = NULL;
1013         mm_pgtables_bytes_init(mm);
1014         mm->map_count = 0;
1015         mm->locked_vm = 0;
1016         atomic64_set(&mm->pinned_vm, 0);
1017         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1018         spin_lock_init(&mm->page_table_lock);
1019         spin_lock_init(&mm->arg_lock);
1020         mm_init_cpumask(mm);
1021         mm_init_aio(mm);
1022         mm_init_owner(mm, p);
1023         RCU_INIT_POINTER(mm->exe_file, NULL);
1024         mmu_notifier_mm_init(mm);
1025         init_tlb_flush_pending(mm);
1026 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1027         mm->pmd_huge_pte = NULL;
1028 #endif
1029         mm_init_uprobes_state(mm);
1030
1031         if (current->mm) {
1032                 mm->flags = current->mm->flags & MMF_INIT_MASK;
1033                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1034         } else {
1035                 mm->flags = default_dump_filter;
1036                 mm->def_flags = 0;
1037         }
1038
1039         if (mm_alloc_pgd(mm))
1040                 goto fail_nopgd;
1041
1042         if (init_new_context(p, mm))
1043                 goto fail_nocontext;
1044
1045         mm->user_ns = get_user_ns(user_ns);
1046         return mm;
1047
1048 fail_nocontext:
1049         mm_free_pgd(mm);
1050 fail_nopgd:
1051         free_mm(mm);
1052         return NULL;
1053 }
1054
1055 /*
1056  * Allocate and initialize an mm_struct.
1057  */
1058 struct mm_struct *mm_alloc(void)
1059 {
1060         struct mm_struct *mm;
1061
1062         mm = allocate_mm();
1063         if (!mm)
1064                 return NULL;
1065
1066         memset(mm, 0, sizeof(*mm));
1067         return mm_init(mm, current, current_user_ns());
1068 }
1069
1070 static inline void __mmput(struct mm_struct *mm)
1071 {
1072         VM_BUG_ON(atomic_read(&mm->mm_users));
1073
1074         uprobe_clear_state(mm);
1075         exit_aio(mm);
1076         ksm_exit(mm);
1077         khugepaged_exit(mm); /* must run before exit_mmap */
1078         exit_mmap(mm);
1079         mm_put_huge_zero_page(mm);
1080         set_mm_exe_file(mm, NULL);
1081         if (!list_empty(&mm->mmlist)) {
1082                 spin_lock(&mmlist_lock);
1083                 list_del(&mm->mmlist);
1084                 spin_unlock(&mmlist_lock);
1085         }
1086         if (mm->binfmt)
1087                 module_put(mm->binfmt->module);
1088         mmdrop(mm);
1089 }
1090
1091 /*
1092  * Decrement the use count and release all resources for an mm.
1093  */
1094 void mmput(struct mm_struct *mm)
1095 {
1096         might_sleep();
1097
1098         if (atomic_dec_and_test(&mm->mm_users))
1099                 __mmput(mm);
1100 }
1101 EXPORT_SYMBOL_GPL(mmput);
1102
1103 #ifdef CONFIG_MMU
1104 static void mmput_async_fn(struct work_struct *work)
1105 {
1106         struct mm_struct *mm = container_of(work, struct mm_struct,
1107                                             async_put_work);
1108
1109         __mmput(mm);
1110 }
1111
1112 void mmput_async(struct mm_struct *mm)
1113 {
1114         if (atomic_dec_and_test(&mm->mm_users)) {
1115                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1116                 schedule_work(&mm->async_put_work);
1117         }
1118 }
1119 #endif
1120
1121 /**
1122  * set_mm_exe_file - change a reference to the mm's executable file
1123  *
1124  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1125  *
1126  * Main users are mmput() and sys_execve(). Callers prevent concurrent
1127  * invocations: in mmput() nobody alive left, in execve task is single
1128  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1129  * mm->exe_file, but does so without using set_mm_exe_file() in order
1130  * to do avoid the need for any locks.
1131  */
1132 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1133 {
1134         struct file *old_exe_file;
1135
1136         /*
1137          * It is safe to dereference the exe_file without RCU as
1138          * this function is only called if nobody else can access
1139          * this mm -- see comment above for justification.
1140          */
1141         old_exe_file = rcu_dereference_raw(mm->exe_file);
1142
1143         if (new_exe_file)
1144                 get_file(new_exe_file);
1145         rcu_assign_pointer(mm->exe_file, new_exe_file);
1146         if (old_exe_file)
1147                 fput(old_exe_file);
1148 }
1149
1150 /**
1151  * get_mm_exe_file - acquire a reference to the mm's executable file
1152  *
1153  * Returns %NULL if mm has no associated executable file.
1154  * User must release file via fput().
1155  */
1156 struct file *get_mm_exe_file(struct mm_struct *mm)
1157 {
1158         struct file *exe_file;
1159
1160         rcu_read_lock();
1161         exe_file = rcu_dereference(mm->exe_file);
1162         if (exe_file && !get_file_rcu(exe_file))
1163                 exe_file = NULL;
1164         rcu_read_unlock();
1165         return exe_file;
1166 }
1167 EXPORT_SYMBOL(get_mm_exe_file);
1168
1169 /**
1170  * get_task_exe_file - acquire a reference to the task's executable file
1171  *
1172  * Returns %NULL if task's mm (if any) has no associated executable file or
1173  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1174  * User must release file via fput().
1175  */
1176 struct file *get_task_exe_file(struct task_struct *task)
1177 {
1178         struct file *exe_file = NULL;
1179         struct mm_struct *mm;
1180
1181         task_lock(task);
1182         mm = task->mm;
1183         if (mm) {
1184                 if (!(task->flags & PF_KTHREAD))
1185                         exe_file = get_mm_exe_file(mm);
1186         }
1187         task_unlock(task);
1188         return exe_file;
1189 }
1190 EXPORT_SYMBOL(get_task_exe_file);
1191
1192 /**
1193  * get_task_mm - acquire a reference to the task's mm
1194  *
1195  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1196  * this kernel workthread has transiently adopted a user mm with use_mm,
1197  * to do its AIO) is not set and if so returns a reference to it, after
1198  * bumping up the use count.  User must release the mm via mmput()
1199  * after use.  Typically used by /proc and ptrace.
1200  */
1201 struct mm_struct *get_task_mm(struct task_struct *task)
1202 {
1203         struct mm_struct *mm;
1204
1205         task_lock(task);
1206         mm = task->mm;
1207         if (mm) {
1208                 if (task->flags & PF_KTHREAD)
1209                         mm = NULL;
1210                 else
1211                         mmget(mm);
1212         }
1213         task_unlock(task);
1214         return mm;
1215 }
1216 EXPORT_SYMBOL_GPL(get_task_mm);
1217
1218 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1219 {
1220         struct mm_struct *mm;
1221         int err;
1222
1223         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
1224         if (err)
1225                 return ERR_PTR(err);
1226
1227         mm = get_task_mm(task);
1228         if (mm && mm != current->mm &&
1229                         !ptrace_may_access(task, mode)) {
1230                 mmput(mm);
1231                 mm = ERR_PTR(-EACCES);
1232         }
1233         mutex_unlock(&task->signal->cred_guard_mutex);
1234
1235         return mm;
1236 }
1237
1238 static void complete_vfork_done(struct task_struct *tsk)
1239 {
1240         struct completion *vfork;
1241
1242         task_lock(tsk);
1243         vfork = tsk->vfork_done;
1244         if (likely(vfork)) {
1245                 tsk->vfork_done = NULL;
1246                 complete(vfork);
1247         }
1248         task_unlock(tsk);
1249 }
1250
1251 static int wait_for_vfork_done(struct task_struct *child,
1252                                 struct completion *vfork)
1253 {
1254         int killed;
1255
1256         freezer_do_not_count();
1257         cgroup_enter_frozen();
1258         killed = wait_for_completion_killable(vfork);
1259         cgroup_leave_frozen(false);
1260         freezer_count();
1261
1262         if (killed) {
1263                 task_lock(child);
1264                 child->vfork_done = NULL;
1265                 task_unlock(child);
1266         }
1267
1268         put_task_struct(child);
1269         return killed;
1270 }
1271
1272 /* Please note the differences between mmput and mm_release.
1273  * mmput is called whenever we stop holding onto a mm_struct,
1274  * error success whatever.
1275  *
1276  * mm_release is called after a mm_struct has been removed
1277  * from the current process.
1278  *
1279  * This difference is important for error handling, when we
1280  * only half set up a mm_struct for a new process and need to restore
1281  * the old one.  Because we mmput the new mm_struct before
1282  * restoring the old one. . .
1283  * Eric Biederman 10 January 1998
1284  */
1285 static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1286 {
1287         uprobe_free_utask(tsk);
1288
1289         /* Get rid of any cached register state */
1290         deactivate_mm(tsk, mm);
1291
1292         /*
1293          * Signal userspace if we're not exiting with a core dump
1294          * because we want to leave the value intact for debugging
1295          * purposes.
1296          */
1297         if (tsk->clear_child_tid) {
1298                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1299                     atomic_read(&mm->mm_users) > 1) {
1300                         /*
1301                          * We don't check the error code - if userspace has
1302                          * not set up a proper pointer then tough luck.
1303                          */
1304                         put_user(0, tsk->clear_child_tid);
1305                         do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1306                                         1, NULL, NULL, 0, 0);
1307                 }
1308                 tsk->clear_child_tid = NULL;
1309         }
1310
1311         /*
1312          * All done, finally we can wake up parent and return this mm to him.
1313          * Also kthread_stop() uses this completion for synchronization.
1314          */
1315         if (tsk->vfork_done)
1316                 complete_vfork_done(tsk);
1317 }
1318
1319 void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1320 {
1321         futex_exit_release(tsk);
1322         mm_release(tsk, mm);
1323 }
1324
1325 void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1326 {
1327         futex_exec_release(tsk);
1328         mm_release(tsk, mm);
1329 }
1330
1331 /**
1332  * dup_mm() - duplicates an existing mm structure
1333  * @tsk: the task_struct with which the new mm will be associated.
1334  * @oldmm: the mm to duplicate.
1335  *
1336  * Allocates a new mm structure and duplicates the provided @oldmm structure
1337  * content into it.
1338  *
1339  * Return: the duplicated mm or NULL on failure.
1340  */
1341 static struct mm_struct *dup_mm(struct task_struct *tsk,
1342                                 struct mm_struct *oldmm)
1343 {
1344         struct mm_struct *mm;
1345         int err;
1346
1347         mm = allocate_mm();
1348         if (!mm)
1349                 goto fail_nomem;
1350
1351         memcpy(mm, oldmm, sizeof(*mm));
1352
1353         if (!mm_init(mm, tsk, mm->user_ns))
1354                 goto fail_nomem;
1355
1356         err = dup_mmap(mm, oldmm);
1357         if (err)
1358                 goto free_pt;
1359
1360         mm->hiwater_rss = get_mm_rss(mm);
1361         mm->hiwater_vm = mm->total_vm;
1362
1363         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1364                 goto free_pt;
1365
1366         return mm;
1367
1368 free_pt:
1369         /* don't put binfmt in mmput, we haven't got module yet */
1370         mm->binfmt = NULL;
1371         mm_init_owner(mm, NULL);
1372         mmput(mm);
1373
1374 fail_nomem:
1375         return NULL;
1376 }
1377
1378 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1379 {
1380         struct mm_struct *mm, *oldmm;
1381         int retval;
1382
1383         tsk->min_flt = tsk->maj_flt = 0;
1384         tsk->nvcsw = tsk->nivcsw = 0;
1385 #ifdef CONFIG_DETECT_HUNG_TASK
1386         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1387         tsk->last_switch_time = 0;
1388 #endif
1389
1390         tsk->mm = NULL;
1391         tsk->active_mm = NULL;
1392
1393         /*
1394          * Are we cloning a kernel thread?
1395          *
1396          * We need to steal a active VM for that..
1397          */
1398         oldmm = current->mm;
1399         if (!oldmm)
1400                 return 0;
1401
1402         /* initialize the new vmacache entries */
1403         vmacache_flush(tsk);
1404
1405         if (clone_flags & CLONE_VM) {
1406                 mmget(oldmm);
1407                 mm = oldmm;
1408                 goto good_mm;
1409         }
1410
1411         retval = -ENOMEM;
1412         mm = dup_mm(tsk, current->mm);
1413         if (!mm)
1414                 goto fail_nomem;
1415
1416 good_mm:
1417         tsk->mm = mm;
1418         tsk->active_mm = mm;
1419         return 0;
1420
1421 fail_nomem:
1422         return retval;
1423 }
1424
1425 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1426 {
1427         struct fs_struct *fs = current->fs;
1428         if (clone_flags & CLONE_FS) {
1429                 /* tsk->fs is already what we want */
1430                 spin_lock(&fs->lock);
1431                 if (fs->in_exec) {
1432                         spin_unlock(&fs->lock);
1433                         return -EAGAIN;
1434                 }
1435                 fs->users++;
1436                 spin_unlock(&fs->lock);
1437                 return 0;
1438         }
1439         tsk->fs = copy_fs_struct(fs);
1440         if (!tsk->fs)
1441                 return -ENOMEM;
1442         return 0;
1443 }
1444
1445 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1446 {
1447         struct files_struct *oldf, *newf;
1448         int error = 0;
1449
1450         /*
1451          * A background process may not have any files ...
1452          */
1453         oldf = current->files;
1454         if (!oldf)
1455                 goto out;
1456
1457         if (clone_flags & CLONE_FILES) {
1458                 atomic_inc(&oldf->count);
1459                 goto out;
1460         }
1461
1462         newf = dup_fd(oldf, &error);
1463         if (!newf)
1464                 goto out;
1465
1466         tsk->files = newf;
1467         error = 0;
1468 out:
1469         return error;
1470 }
1471
1472 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1473 {
1474 #ifdef CONFIG_BLOCK
1475         struct io_context *ioc = current->io_context;
1476         struct io_context *new_ioc;
1477
1478         if (!ioc)
1479                 return 0;
1480         /*
1481          * Share io context with parent, if CLONE_IO is set
1482          */
1483         if (clone_flags & CLONE_IO) {
1484                 ioc_task_link(ioc);
1485                 tsk->io_context = ioc;
1486         } else if (ioprio_valid(ioc->ioprio)) {
1487                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1488                 if (unlikely(!new_ioc))
1489                         return -ENOMEM;
1490
1491                 new_ioc->ioprio = ioc->ioprio;
1492                 put_io_context(new_ioc);
1493         }
1494 #endif
1495         return 0;
1496 }
1497
1498 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1499 {
1500         struct sighand_struct *sig;
1501
1502         if (clone_flags & CLONE_SIGHAND) {
1503                 refcount_inc(&current->sighand->count);
1504                 return 0;
1505         }
1506         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1507         rcu_assign_pointer(tsk->sighand, sig);
1508         if (!sig)
1509                 return -ENOMEM;
1510
1511         refcount_set(&sig->count, 1);
1512         spin_lock_irq(&current->sighand->siglock);
1513         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1514         spin_unlock_irq(&current->sighand->siglock);
1515
1516         /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1517         if (clone_flags & CLONE_CLEAR_SIGHAND)
1518                 flush_signal_handlers(tsk, 0);
1519
1520         return 0;
1521 }
1522
1523 void __cleanup_sighand(struct sighand_struct *sighand)
1524 {
1525         if (refcount_dec_and_test(&sighand->count)) {
1526                 signalfd_cleanup(sighand);
1527                 /*
1528                  * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1529                  * without an RCU grace period, see __lock_task_sighand().
1530                  */
1531                 kmem_cache_free(sighand_cachep, sighand);
1532         }
1533 }
1534
1535 /*
1536  * Initialize POSIX timer handling for a thread group.
1537  */
1538 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1539 {
1540         struct posix_cputimers *pct = &sig->posix_cputimers;
1541         unsigned long cpu_limit;
1542
1543         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1544         posix_cputimers_group_init(pct, cpu_limit);
1545 }
1546
1547 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1548 {
1549         struct signal_struct *sig;
1550
1551         if (clone_flags & CLONE_THREAD)
1552                 return 0;
1553
1554         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1555         tsk->signal = sig;
1556         if (!sig)
1557                 return -ENOMEM;
1558
1559         sig->nr_threads = 1;
1560         atomic_set(&sig->live, 1);
1561         refcount_set(&sig->sigcnt, 1);
1562
1563         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1564         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1565         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1566
1567         init_waitqueue_head(&sig->wait_chldexit);
1568         sig->curr_target = tsk;
1569         init_sigpending(&sig->shared_pending);
1570         INIT_HLIST_HEAD(&sig->multiprocess);
1571         seqlock_init(&sig->stats_lock);
1572         prev_cputime_init(&sig->prev_cputime);
1573
1574 #ifdef CONFIG_POSIX_TIMERS
1575         INIT_LIST_HEAD(&sig->posix_timers);
1576         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1577         sig->real_timer.function = it_real_fn;
1578 #endif
1579
1580         task_lock(current->group_leader);
1581         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1582         task_unlock(current->group_leader);
1583
1584         posix_cpu_timers_init_group(sig);
1585
1586         tty_audit_fork(sig);
1587         sched_autogroup_fork(sig);
1588
1589         sig->oom_score_adj = current->signal->oom_score_adj;
1590         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1591
1592         mutex_init(&sig->cred_guard_mutex);
1593
1594         return 0;
1595 }
1596
1597 static void copy_seccomp(struct task_struct *p)
1598 {
1599 #ifdef CONFIG_SECCOMP
1600         /*
1601          * Must be called with sighand->lock held, which is common to
1602          * all threads in the group. Holding cred_guard_mutex is not
1603          * needed because this new task is not yet running and cannot
1604          * be racing exec.
1605          */
1606         assert_spin_locked(&current->sighand->siglock);
1607
1608         /* Ref-count the new filter user, and assign it. */
1609         get_seccomp_filter(current);
1610         p->seccomp = current->seccomp;
1611
1612         /*
1613          * Explicitly enable no_new_privs here in case it got set
1614          * between the task_struct being duplicated and holding the
1615          * sighand lock. The seccomp state and nnp must be in sync.
1616          */
1617         if (task_no_new_privs(current))
1618                 task_set_no_new_privs(p);
1619
1620         /*
1621          * If the parent gained a seccomp mode after copying thread
1622          * flags and between before we held the sighand lock, we have
1623          * to manually enable the seccomp thread flag here.
1624          */
1625         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1626                 set_tsk_thread_flag(p, TIF_SECCOMP);
1627 #endif
1628 }
1629
1630 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1631 {
1632         current->clear_child_tid = tidptr;
1633
1634         return task_pid_vnr(current);
1635 }
1636
1637 static void rt_mutex_init_task(struct task_struct *p)
1638 {
1639         raw_spin_lock_init(&p->pi_lock);
1640 #ifdef CONFIG_RT_MUTEXES
1641         p->pi_waiters = RB_ROOT_CACHED;
1642         p->pi_top_task = NULL;
1643         p->pi_blocked_on = NULL;
1644 #endif
1645 }
1646
1647 static inline void init_task_pid_links(struct task_struct *task)
1648 {
1649         enum pid_type type;
1650
1651         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1652                 INIT_HLIST_NODE(&task->pid_links[type]);
1653         }
1654 }
1655
1656 static inline void
1657 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1658 {
1659         if (type == PIDTYPE_PID)
1660                 task->thread_pid = pid;
1661         else
1662                 task->signal->pids[type] = pid;
1663 }
1664
1665 static inline void rcu_copy_process(struct task_struct *p)
1666 {
1667 #ifdef CONFIG_PREEMPT_RCU
1668         p->rcu_read_lock_nesting = 0;
1669         p->rcu_read_unlock_special.s = 0;
1670         p->rcu_blocked_node = NULL;
1671         INIT_LIST_HEAD(&p->rcu_node_entry);
1672 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1673 #ifdef CONFIG_TASKS_RCU
1674         p->rcu_tasks_holdout = false;
1675         INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1676         p->rcu_tasks_idle_cpu = -1;
1677 #endif /* #ifdef CONFIG_TASKS_RCU */
1678 }
1679
1680 struct pid *pidfd_pid(const struct file *file)
1681 {
1682         if (file->f_op == &pidfd_fops)
1683                 return file->private_data;
1684
1685         return ERR_PTR(-EBADF);
1686 }
1687
1688 static int pidfd_release(struct inode *inode, struct file *file)
1689 {
1690         struct pid *pid = file->private_data;
1691
1692         file->private_data = NULL;
1693         put_pid(pid);
1694         return 0;
1695 }
1696
1697 #ifdef CONFIG_PROC_FS
1698 /**
1699  * pidfd_show_fdinfo - print information about a pidfd
1700  * @m: proc fdinfo file
1701  * @f: file referencing a pidfd
1702  *
1703  * Pid:
1704  * This function will print the pid that a given pidfd refers to in the
1705  * pid namespace of the procfs instance.
1706  * If the pid namespace of the process is not a descendant of the pid
1707  * namespace of the procfs instance 0 will be shown as its pid. This is
1708  * similar to calling getppid() on a process whose parent is outside of
1709  * its pid namespace.
1710  *
1711  * NSpid:
1712  * If pid namespaces are supported then this function will also print
1713  * the pid of a given pidfd refers to for all descendant pid namespaces
1714  * starting from the current pid namespace of the instance, i.e. the
1715  * Pid field and the first entry in the NSpid field will be identical.
1716  * If the pid namespace of the process is not a descendant of the pid
1717  * namespace of the procfs instance 0 will be shown as its first NSpid
1718  * entry and no others will be shown.
1719  * Note that this differs from the Pid and NSpid fields in
1720  * /proc/<pid>/status where Pid and NSpid are always shown relative to
1721  * the  pid namespace of the procfs instance. The difference becomes
1722  * obvious when sending around a pidfd between pid namespaces from a
1723  * different branch of the tree, i.e. where no ancestoral relation is
1724  * present between the pid namespaces:
1725  * - create two new pid namespaces ns1 and ns2 in the initial pid
1726  *   namespace (also take care to create new mount namespaces in the
1727  *   new pid namespace and mount procfs)
1728  * - create a process with a pidfd in ns1
1729  * - send pidfd from ns1 to ns2
1730  * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
1731  *   have exactly one entry, which is 0
1732  */
1733 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1734 {
1735         struct pid *pid = f->private_data;
1736         struct pid_namespace *ns;
1737         pid_t nr = -1;
1738
1739         if (likely(pid_has_task(pid, PIDTYPE_PID))) {
1740                 ns = proc_pid_ns(file_inode(m->file));
1741                 nr = pid_nr_ns(pid, ns);
1742         }
1743
1744         seq_put_decimal_ll(m, "Pid:\t", nr);
1745
1746 #ifdef CONFIG_PID_NS
1747         seq_put_decimal_ll(m, "\nNSpid:\t", nr);
1748         if (nr > 0) {
1749                 int i;
1750
1751                 /* If nr is non-zero it means that 'pid' is valid and that
1752                  * ns, i.e. the pid namespace associated with the procfs
1753                  * instance, is in the pid namespace hierarchy of pid.
1754                  * Start at one below the already printed level.
1755                  */
1756                 for (i = ns->level + 1; i <= pid->level; i++)
1757                         seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
1758         }
1759 #endif
1760         seq_putc(m, '\n');
1761 }
1762 #endif
1763
1764 /*
1765  * Poll support for process exit notification.
1766  */
1767 static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
1768 {
1769         struct task_struct *task;
1770         struct pid *pid = file->private_data;
1771         __poll_t poll_flags = 0;
1772
1773         poll_wait(file, &pid->wait_pidfd, pts);
1774
1775         rcu_read_lock();
1776         task = pid_task(pid, PIDTYPE_PID);
1777         /*
1778          * Inform pollers only when the whole thread group exits.
1779          * If the thread group leader exits before all other threads in the
1780          * group, then poll(2) should block, similar to the wait(2) family.
1781          */
1782         if (!task || (task->exit_state && thread_group_empty(task)))
1783                 poll_flags = EPOLLIN | EPOLLRDNORM;
1784         rcu_read_unlock();
1785
1786         return poll_flags;
1787 }
1788
1789 const struct file_operations pidfd_fops = {
1790         .release = pidfd_release,
1791         .poll = pidfd_poll,
1792 #ifdef CONFIG_PROC_FS
1793         .show_fdinfo = pidfd_show_fdinfo,
1794 #endif
1795 };
1796
1797 static void __delayed_free_task(struct rcu_head *rhp)
1798 {
1799         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1800
1801         free_task(tsk);
1802 }
1803
1804 static __always_inline void delayed_free_task(struct task_struct *tsk)
1805 {
1806         if (IS_ENABLED(CONFIG_MEMCG))
1807                 call_rcu(&tsk->rcu, __delayed_free_task);
1808         else
1809                 free_task(tsk);
1810 }
1811
1812 /*
1813  * This creates a new process as a copy of the old one,
1814  * but does not actually start it yet.
1815  *
1816  * It copies the registers, and all the appropriate
1817  * parts of the process environment (as per the clone
1818  * flags). The actual kick-off is left to the caller.
1819  */
1820 static __latent_entropy struct task_struct *copy_process(
1821                                         struct pid *pid,
1822                                         int trace,
1823                                         int node,
1824                                         struct kernel_clone_args *args)
1825 {
1826         int pidfd = -1, retval;
1827         struct task_struct *p;
1828         struct multiprocess_signals delayed;
1829         struct file *pidfile = NULL;
1830         u64 clone_flags = args->flags;
1831
1832         /*
1833          * Don't allow sharing the root directory with processes in a different
1834          * namespace
1835          */
1836         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1837                 return ERR_PTR(-EINVAL);
1838
1839         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1840                 return ERR_PTR(-EINVAL);
1841
1842         /*
1843          * Thread groups must share signals as well, and detached threads
1844          * can only be started up within the thread group.
1845          */
1846         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1847                 return ERR_PTR(-EINVAL);
1848
1849         /*
1850          * Shared signal handlers imply shared VM. By way of the above,
1851          * thread groups also imply shared VM. Blocking this case allows
1852          * for various simplifications in other code.
1853          */
1854         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1855                 return ERR_PTR(-EINVAL);
1856
1857         /*
1858          * Siblings of global init remain as zombies on exit since they are
1859          * not reaped by their parent (swapper). To solve this and to avoid
1860          * multi-rooted process trees, prevent global and container-inits
1861          * from creating siblings.
1862          */
1863         if ((clone_flags & CLONE_PARENT) &&
1864                                 current->signal->flags & SIGNAL_UNKILLABLE)
1865                 return ERR_PTR(-EINVAL);
1866
1867         /*
1868          * If the new process will be in a different pid or user namespace
1869          * do not allow it to share a thread group with the forking task.
1870          */
1871         if (clone_flags & CLONE_THREAD) {
1872                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1873                     (task_active_pid_ns(current) !=
1874                                 current->nsproxy->pid_ns_for_children))
1875                         return ERR_PTR(-EINVAL);
1876         }
1877
1878         if (clone_flags & CLONE_PIDFD) {
1879                 /*
1880                  * - CLONE_DETACHED is blocked so that we can potentially
1881                  *   reuse it later for CLONE_PIDFD.
1882                  * - CLONE_THREAD is blocked until someone really needs it.
1883                  */
1884                 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
1885                         return ERR_PTR(-EINVAL);
1886         }
1887
1888         /*
1889          * Force any signals received before this point to be delivered
1890          * before the fork happens.  Collect up signals sent to multiple
1891          * processes that happen during the fork and delay them so that
1892          * they appear to happen after the fork.
1893          */
1894         sigemptyset(&delayed.signal);
1895         INIT_HLIST_NODE(&delayed.node);
1896
1897         spin_lock_irq(&current->sighand->siglock);
1898         if (!(clone_flags & CLONE_THREAD))
1899                 hlist_add_head(&delayed.node, &current->signal->multiprocess);
1900         recalc_sigpending();
1901         spin_unlock_irq(&current->sighand->siglock);
1902         retval = -ERESTARTNOINTR;
1903         if (signal_pending(current))
1904                 goto fork_out;
1905
1906         retval = -ENOMEM;
1907         p = dup_task_struct(current, node);
1908         if (!p)
1909                 goto fork_out;
1910
1911         /*
1912          * This _must_ happen before we call free_task(), i.e. before we jump
1913          * to any of the bad_fork_* labels. This is to avoid freeing
1914          * p->set_child_tid which is (ab)used as a kthread's data pointer for
1915          * kernel threads (PF_KTHREAD).
1916          */
1917         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
1918         /*
1919          * Clear TID on mm_release()?
1920          */
1921         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
1922
1923         ftrace_graph_init_task(p);
1924
1925         rt_mutex_init_task(p);
1926
1927 #ifdef CONFIG_PROVE_LOCKING
1928         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1929         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1930 #endif
1931         retval = -EAGAIN;
1932         if (atomic_read(&p->real_cred->user->processes) >=
1933                         task_rlimit(p, RLIMIT_NPROC)) {
1934                 if (p->real_cred->user != INIT_USER &&
1935                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1936                         goto bad_fork_free;
1937         }
1938         current->flags &= ~PF_NPROC_EXCEEDED;
1939
1940         retval = copy_creds(p, clone_flags);
1941         if (retval < 0)
1942                 goto bad_fork_free;
1943
1944         /*
1945          * If multiple threads are within copy_process(), then this check
1946          * triggers too late. This doesn't hurt, the check is only there
1947          * to stop root fork bombs.
1948          */
1949         retval = -EAGAIN;
1950         if (nr_threads >= max_threads)
1951                 goto bad_fork_cleanup_count;
1952
1953         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1954         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1955         p->flags |= PF_FORKNOEXEC;
1956         INIT_LIST_HEAD(&p->children);
1957         INIT_LIST_HEAD(&p->sibling);
1958         rcu_copy_process(p);
1959         p->vfork_done = NULL;
1960         spin_lock_init(&p->alloc_lock);
1961
1962         init_sigpending(&p->pending);
1963
1964         p->utime = p->stime = p->gtime = 0;
1965 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1966         p->utimescaled = p->stimescaled = 0;
1967 #endif
1968         prev_cputime_init(&p->prev_cputime);
1969
1970 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1971         seqcount_init(&p->vtime.seqcount);
1972         p->vtime.starttime = 0;
1973         p->vtime.state = VTIME_INACTIVE;
1974 #endif
1975
1976 #if defined(SPLIT_RSS_COUNTING)
1977         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1978 #endif
1979
1980         p->default_timer_slack_ns = current->timer_slack_ns;
1981
1982 #ifdef CONFIG_PSI
1983         p->psi_flags = 0;
1984 #endif
1985
1986         task_io_accounting_init(&p->ioac);
1987         acct_clear_integrals(p);
1988
1989         posix_cputimers_init(&p->posix_cputimers);
1990
1991         p->io_context = NULL;
1992         audit_set_context(p, NULL);
1993         cgroup_fork(p);
1994 #ifdef CONFIG_NUMA
1995         p->mempolicy = mpol_dup(p->mempolicy);
1996         if (IS_ERR(p->mempolicy)) {
1997                 retval = PTR_ERR(p->mempolicy);
1998                 p->mempolicy = NULL;
1999                 goto bad_fork_cleanup_threadgroup_lock;
2000         }
2001 #endif
2002 #ifdef CONFIG_CPUSETS
2003         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2004         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2005         seqcount_init(&p->mems_allowed_seq);
2006 #endif
2007 #ifdef CONFIG_TRACE_IRQFLAGS
2008         p->irq_events = 0;
2009         p->hardirqs_enabled = 0;
2010         p->hardirq_enable_ip = 0;
2011         p->hardirq_enable_event = 0;
2012         p->hardirq_disable_ip = _THIS_IP_;
2013         p->hardirq_disable_event = 0;
2014         p->softirqs_enabled = 1;
2015         p->softirq_enable_ip = _THIS_IP_;
2016         p->softirq_enable_event = 0;
2017         p->softirq_disable_ip = 0;
2018         p->softirq_disable_event = 0;
2019         p->hardirq_context = 0;
2020         p->softirq_context = 0;
2021 #endif
2022
2023         p->pagefault_disabled = 0;
2024
2025 #ifdef CONFIG_LOCKDEP
2026         lockdep_init_task(p);
2027 #endif
2028
2029 #ifdef CONFIG_DEBUG_MUTEXES
2030         p->blocked_on = NULL; /* not blocked yet */
2031 #endif
2032 #ifdef CONFIG_BCACHE
2033         p->sequential_io        = 0;
2034         p->sequential_io_avg    = 0;
2035 #endif
2036
2037         /* Perform scheduler related setup. Assign this task to a CPU. */
2038         retval = sched_fork(clone_flags, p);
2039         if (retval)
2040                 goto bad_fork_cleanup_policy;
2041
2042         retval = perf_event_init_task(p);
2043         if (retval)
2044                 goto bad_fork_cleanup_policy;
2045         retval = audit_alloc(p);
2046         if (retval)
2047                 goto bad_fork_cleanup_perf;
2048         /* copy all the process information */
2049         shm_init_task(p);
2050         retval = security_task_alloc(p, clone_flags);
2051         if (retval)
2052                 goto bad_fork_cleanup_audit;
2053         retval = copy_semundo(clone_flags, p);
2054         if (retval)
2055                 goto bad_fork_cleanup_security;
2056         retval = copy_files(clone_flags, p);
2057         if (retval)
2058                 goto bad_fork_cleanup_semundo;
2059         retval = copy_fs(clone_flags, p);
2060         if (retval)
2061                 goto bad_fork_cleanup_files;
2062         retval = copy_sighand(clone_flags, p);
2063         if (retval)
2064                 goto bad_fork_cleanup_fs;
2065         retval = copy_signal(clone_flags, p);
2066         if (retval)
2067                 goto bad_fork_cleanup_sighand;
2068         retval = copy_mm(clone_flags, p);
2069         if (retval)
2070                 goto bad_fork_cleanup_signal;
2071         retval = copy_namespaces(clone_flags, p);
2072         if (retval)
2073                 goto bad_fork_cleanup_mm;
2074         retval = copy_io(clone_flags, p);
2075         if (retval)
2076                 goto bad_fork_cleanup_namespaces;
2077         retval = copy_thread_tls(clone_flags, args->stack, args->stack_size, p,
2078                                  args->tls);
2079         if (retval)
2080                 goto bad_fork_cleanup_io;
2081
2082         stackleak_task_init(p);
2083
2084         if (pid != &init_struct_pid) {
2085                 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2086                                 args->set_tid_size);
2087                 if (IS_ERR(pid)) {
2088                         retval = PTR_ERR(pid);
2089                         goto bad_fork_cleanup_thread;
2090                 }
2091         }
2092
2093         /*
2094          * This has to happen after we've potentially unshared the file
2095          * descriptor table (so that the pidfd doesn't leak into the child
2096          * if the fd table isn't shared).
2097          */
2098         if (clone_flags & CLONE_PIDFD) {
2099                 retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2100                 if (retval < 0)
2101                         goto bad_fork_free_pid;
2102
2103                 pidfd = retval;
2104
2105                 pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2106                                               O_RDWR | O_CLOEXEC);
2107                 if (IS_ERR(pidfile)) {
2108                         put_unused_fd(pidfd);
2109                         retval = PTR_ERR(pidfile);
2110                         goto bad_fork_free_pid;
2111                 }
2112                 get_pid(pid);   /* held by pidfile now */
2113
2114                 retval = put_user(pidfd, args->pidfd);
2115                 if (retval)
2116                         goto bad_fork_put_pidfd;
2117         }
2118
2119 #ifdef CONFIG_BLOCK
2120         p->plug = NULL;
2121 #endif
2122         futex_init_task(p);
2123
2124         /*
2125          * sigaltstack should be cleared when sharing the same VM
2126          */
2127         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2128                 sas_ss_reset(p);
2129
2130         /*
2131          * Syscall tracing and stepping should be turned off in the
2132          * child regardless of CLONE_PTRACE.
2133          */
2134         user_disable_single_step(p);
2135         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
2136 #ifdef TIF_SYSCALL_EMU
2137         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
2138 #endif
2139         clear_tsk_latency_tracing(p);
2140
2141         /* ok, now we should be set up.. */
2142         p->pid = pid_nr(pid);
2143         if (clone_flags & CLONE_THREAD) {
2144                 p->exit_signal = -1;
2145                 p->group_leader = current->group_leader;
2146                 p->tgid = current->tgid;
2147         } else {
2148                 if (clone_flags & CLONE_PARENT)
2149                         p->exit_signal = current->group_leader->exit_signal;
2150                 else
2151                         p->exit_signal = args->exit_signal;
2152                 p->group_leader = p;
2153                 p->tgid = p->pid;
2154         }
2155
2156         p->nr_dirtied = 0;
2157         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2158         p->dirty_paused_when = 0;
2159
2160         p->pdeath_signal = 0;
2161         INIT_LIST_HEAD(&p->thread_group);
2162         p->task_works = NULL;
2163
2164         cgroup_threadgroup_change_begin(current);
2165         /*
2166          * Ensure that the cgroup subsystem policies allow the new process to be
2167          * forked. It should be noted the the new process's css_set can be changed
2168          * between here and cgroup_post_fork() if an organisation operation is in
2169          * progress.
2170          */
2171         retval = cgroup_can_fork(p);
2172         if (retval)
2173                 goto bad_fork_cgroup_threadgroup_change_end;
2174
2175         /*
2176          * From this point on we must avoid any synchronous user-space
2177          * communication until we take the tasklist-lock. In particular, we do
2178          * not want user-space to be able to predict the process start-time by
2179          * stalling fork(2) after we recorded the start_time but before it is
2180          * visible to the system.
2181          */
2182
2183         p->start_time = ktime_get_ns();
2184         p->real_start_time = ktime_get_boottime_ns();
2185
2186         /*
2187          * Make it visible to the rest of the system, but dont wake it up yet.
2188          * Need tasklist lock for parent etc handling!
2189          */
2190         write_lock_irq(&tasklist_lock);
2191
2192         /* CLONE_PARENT re-uses the old parent */
2193         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2194                 p->real_parent = current->real_parent;
2195                 p->parent_exec_id = current->parent_exec_id;
2196         } else {
2197                 p->real_parent = current;
2198                 p->parent_exec_id = current->self_exec_id;
2199         }
2200
2201         klp_copy_process(p);
2202
2203         spin_lock(&current->sighand->siglock);
2204
2205         /*
2206          * Copy seccomp details explicitly here, in case they were changed
2207          * before holding sighand lock.
2208          */
2209         copy_seccomp(p);
2210
2211         rseq_fork(p, clone_flags);
2212
2213         /* Don't start children in a dying pid namespace */
2214         if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2215                 retval = -ENOMEM;
2216                 goto bad_fork_cancel_cgroup;
2217         }
2218
2219         /* Let kill terminate clone/fork in the middle */
2220         if (fatal_signal_pending(current)) {
2221                 retval = -EINTR;
2222                 goto bad_fork_cancel_cgroup;
2223         }
2224
2225         /* past the last point of failure */
2226         if (pidfile)
2227                 fd_install(pidfd, pidfile);
2228
2229         init_task_pid_links(p);
2230         if (likely(p->pid)) {
2231                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2232
2233                 init_task_pid(p, PIDTYPE_PID, pid);
2234                 if (thread_group_leader(p)) {
2235                         init_task_pid(p, PIDTYPE_TGID, pid);
2236                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2237                         init_task_pid(p, PIDTYPE_SID, task_session(current));
2238
2239                         if (is_child_reaper(pid)) {
2240                                 ns_of_pid(pid)->child_reaper = p;
2241                                 p->signal->flags |= SIGNAL_UNKILLABLE;
2242                         }
2243                         p->signal->shared_pending.signal = delayed.signal;
2244                         p->signal->tty = tty_kref_get(current->signal->tty);
2245                         /*
2246                          * Inherit has_child_subreaper flag under the same
2247                          * tasklist_lock with adding child to the process tree
2248                          * for propagate_has_child_subreaper optimization.
2249                          */
2250                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2251                                                          p->real_parent->signal->is_child_subreaper;
2252                         list_add_tail(&p->sibling, &p->real_parent->children);
2253                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
2254                         attach_pid(p, PIDTYPE_TGID);
2255                         attach_pid(p, PIDTYPE_PGID);
2256                         attach_pid(p, PIDTYPE_SID);
2257                         __this_cpu_inc(process_counts);
2258                 } else {
2259                         current->signal->nr_threads++;
2260                         atomic_inc(&current->signal->live);
2261                         refcount_inc(&current->signal->sigcnt);
2262                         task_join_group_stop(p);
2263                         list_add_tail_rcu(&p->thread_group,
2264                                           &p->group_leader->thread_group);
2265                         list_add_tail_rcu(&p->thread_node,
2266                                           &p->signal->thread_head);
2267                 }
2268                 attach_pid(p, PIDTYPE_PID);
2269                 nr_threads++;
2270         }
2271         total_forks++;
2272         hlist_del_init(&delayed.node);
2273         spin_unlock(&current->sighand->siglock);
2274         syscall_tracepoint_update(p);
2275         write_unlock_irq(&tasklist_lock);
2276
2277         proc_fork_connector(p);
2278         cgroup_post_fork(p);
2279         cgroup_threadgroup_change_end(current);
2280         perf_event_fork(p);
2281
2282         trace_task_newtask(p, clone_flags);
2283         uprobe_copy_process(p, clone_flags);
2284
2285         return p;
2286
2287 bad_fork_cancel_cgroup:
2288         spin_unlock(&current->sighand->siglock);
2289         write_unlock_irq(&tasklist_lock);
2290         cgroup_cancel_fork(p);
2291 bad_fork_cgroup_threadgroup_change_end:
2292         cgroup_threadgroup_change_end(current);
2293 bad_fork_put_pidfd:
2294         if (clone_flags & CLONE_PIDFD) {
2295                 fput(pidfile);
2296                 put_unused_fd(pidfd);
2297         }
2298 bad_fork_free_pid:
2299         if (pid != &init_struct_pid)
2300                 free_pid(pid);
2301 bad_fork_cleanup_thread:
2302         exit_thread(p);
2303 bad_fork_cleanup_io:
2304         if (p->io_context)
2305                 exit_io_context(p);
2306 bad_fork_cleanup_namespaces:
2307         exit_task_namespaces(p);
2308 bad_fork_cleanup_mm:
2309         if (p->mm) {
2310                 mm_clear_owner(p->mm, p);
2311                 mmput(p->mm);
2312         }
2313 bad_fork_cleanup_signal:
2314         if (!(clone_flags & CLONE_THREAD))
2315                 free_signal_struct(p->signal);
2316 bad_fork_cleanup_sighand:
2317         __cleanup_sighand(p->sighand);
2318 bad_fork_cleanup_fs:
2319         exit_fs(p); /* blocking */
2320 bad_fork_cleanup_files:
2321         exit_files(p); /* blocking */
2322 bad_fork_cleanup_semundo:
2323         exit_sem(p);
2324 bad_fork_cleanup_security:
2325         security_task_free(p);
2326 bad_fork_cleanup_audit:
2327         audit_free(p);
2328 bad_fork_cleanup_perf:
2329         perf_event_free_task(p);
2330 bad_fork_cleanup_policy:
2331         lockdep_free_task(p);
2332 #ifdef CONFIG_NUMA
2333         mpol_put(p->mempolicy);
2334 bad_fork_cleanup_threadgroup_lock:
2335 #endif
2336         delayacct_tsk_free(p);
2337 bad_fork_cleanup_count:
2338         atomic_dec(&p->cred->user->processes);
2339         exit_creds(p);
2340 bad_fork_free:
2341         p->state = TASK_DEAD;
2342         put_task_stack(p);
2343         delayed_free_task(p);
2344 fork_out:
2345         spin_lock_irq(&current->sighand->siglock);
2346         hlist_del_init(&delayed.node);
2347         spin_unlock_irq(&current->sighand->siglock);
2348         return ERR_PTR(retval);
2349 }
2350
2351 static inline void init_idle_pids(struct task_struct *idle)
2352 {
2353         enum pid_type type;
2354
2355         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2356                 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2357                 init_task_pid(idle, type, &init_struct_pid);
2358         }
2359 }
2360
2361 struct task_struct *fork_idle(int cpu)
2362 {
2363         struct task_struct *task;
2364         struct kernel_clone_args args = {
2365                 .flags = CLONE_VM,
2366         };
2367
2368         task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2369         if (!IS_ERR(task)) {
2370                 init_idle_pids(task);
2371                 init_idle(task, cpu);
2372         }
2373
2374         return task;
2375 }
2376
2377 struct mm_struct *copy_init_mm(void)
2378 {
2379         return dup_mm(NULL, &init_mm);
2380 }
2381
2382 /*
2383  *  Ok, this is the main fork-routine.
2384  *
2385  * It copies the process, and if successful kick-starts
2386  * it and waits for it to finish using the VM if required.
2387  *
2388  * args->exit_signal is expected to be checked for sanity by the caller.
2389  */
2390 long _do_fork(struct kernel_clone_args *args)
2391 {
2392         u64 clone_flags = args->flags;
2393         struct completion vfork;
2394         struct pid *pid;
2395         struct task_struct *p;
2396         int trace = 0;
2397         long nr;
2398
2399         /*
2400          * Determine whether and which event to report to ptracer.  When
2401          * called from kernel_thread or CLONE_UNTRACED is explicitly
2402          * requested, no event is reported; otherwise, report if the event
2403          * for the type of forking is enabled.
2404          */
2405         if (!(clone_flags & CLONE_UNTRACED)) {
2406                 if (clone_flags & CLONE_VFORK)
2407                         trace = PTRACE_EVENT_VFORK;
2408                 else if (args->exit_signal != SIGCHLD)
2409                         trace = PTRACE_EVENT_CLONE;
2410                 else
2411                         trace = PTRACE_EVENT_FORK;
2412
2413                 if (likely(!ptrace_event_enabled(current, trace)))
2414                         trace = 0;
2415         }
2416
2417         p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2418         add_latent_entropy();
2419
2420         if (IS_ERR(p))
2421                 return PTR_ERR(p);
2422
2423         /*
2424          * Do this prior waking up the new thread - the thread pointer
2425          * might get invalid after that point, if the thread exits quickly.
2426          */
2427         trace_sched_process_fork(current, p);
2428
2429         pid = get_task_pid(p, PIDTYPE_PID);
2430         nr = pid_vnr(pid);
2431
2432         if (clone_flags & CLONE_PARENT_SETTID)
2433                 put_user(nr, args->parent_tid);
2434
2435         if (clone_flags & CLONE_VFORK) {
2436                 p->vfork_done = &vfork;
2437                 init_completion(&vfork);
2438                 get_task_struct(p);
2439         }
2440
2441         wake_up_new_task(p);
2442
2443         /* forking complete and child started to run, tell ptracer */
2444         if (unlikely(trace))
2445                 ptrace_event_pid(trace, pid);
2446
2447         if (clone_flags & CLONE_VFORK) {
2448                 if (!wait_for_vfork_done(p, &vfork))
2449                         ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2450         }
2451
2452         put_pid(pid);
2453         return nr;
2454 }
2455
2456 bool legacy_clone_args_valid(const struct kernel_clone_args *kargs)
2457 {
2458         /* clone(CLONE_PIDFD) uses parent_tidptr to return a pidfd */
2459         if ((kargs->flags & CLONE_PIDFD) &&
2460             (kargs->flags & CLONE_PARENT_SETTID))
2461                 return false;
2462
2463         return true;
2464 }
2465
2466 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2467 /* For compatibility with architectures that call do_fork directly rather than
2468  * using the syscall entry points below. */
2469 long do_fork(unsigned long clone_flags,
2470               unsigned long stack_start,
2471               unsigned long stack_size,
2472               int __user *parent_tidptr,
2473               int __user *child_tidptr)
2474 {
2475         struct kernel_clone_args args = {
2476                 .flags          = (clone_flags & ~CSIGNAL),
2477                 .pidfd          = parent_tidptr,
2478                 .child_tid      = child_tidptr,
2479                 .parent_tid     = parent_tidptr,
2480                 .exit_signal    = (clone_flags & CSIGNAL),
2481                 .stack          = stack_start,
2482                 .stack_size     = stack_size,
2483         };
2484
2485         if (!legacy_clone_args_valid(&args))
2486                 return -EINVAL;
2487
2488         return _do_fork(&args);
2489 }
2490 #endif
2491
2492 /*
2493  * Create a kernel thread.
2494  */
2495 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2496 {
2497         struct kernel_clone_args args = {
2498                 .flags          = ((flags | CLONE_VM | CLONE_UNTRACED) & ~CSIGNAL),
2499                 .exit_signal    = (flags & CSIGNAL),
2500                 .stack          = (unsigned long)fn,
2501                 .stack_size     = (unsigned long)arg,
2502         };
2503
2504         return _do_fork(&args);
2505 }
2506
2507 #ifdef __ARCH_WANT_SYS_FORK
2508 SYSCALL_DEFINE0(fork)
2509 {
2510 #ifdef CONFIG_MMU
2511         struct kernel_clone_args args = {
2512                 .exit_signal = SIGCHLD,
2513         };
2514
2515         return _do_fork(&args);
2516 #else
2517         /* can not support in nommu mode */
2518         return -EINVAL;
2519 #endif
2520 }
2521 #endif
2522
2523 #ifdef __ARCH_WANT_SYS_VFORK
2524 SYSCALL_DEFINE0(vfork)
2525 {
2526         struct kernel_clone_args args = {
2527                 .flags          = CLONE_VFORK | CLONE_VM,
2528                 .exit_signal    = SIGCHLD,
2529         };
2530
2531         return _do_fork(&args);
2532 }
2533 #endif
2534
2535 #ifdef __ARCH_WANT_SYS_CLONE
2536 #ifdef CONFIG_CLONE_BACKWARDS
2537 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2538                  int __user *, parent_tidptr,
2539                  unsigned long, tls,
2540                  int __user *, child_tidptr)
2541 #elif defined(CONFIG_CLONE_BACKWARDS2)
2542 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2543                  int __user *, parent_tidptr,
2544                  int __user *, child_tidptr,
2545                  unsigned long, tls)
2546 #elif defined(CONFIG_CLONE_BACKWARDS3)
2547 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2548                 int, stack_size,
2549                 int __user *, parent_tidptr,
2550                 int __user *, child_tidptr,
2551                 unsigned long, tls)
2552 #else
2553 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2554                  int __user *, parent_tidptr,
2555                  int __user *, child_tidptr,
2556                  unsigned long, tls)
2557 #endif
2558 {
2559         struct kernel_clone_args args = {
2560                 .flags          = (clone_flags & ~CSIGNAL),
2561                 .pidfd          = parent_tidptr,
2562                 .child_tid      = child_tidptr,
2563                 .parent_tid     = parent_tidptr,
2564                 .exit_signal    = (clone_flags & CSIGNAL),
2565                 .stack          = newsp,
2566                 .tls            = tls,
2567         };
2568
2569         if (!legacy_clone_args_valid(&args))
2570                 return -EINVAL;
2571
2572         return _do_fork(&args);
2573 }
2574 #endif
2575
2576 #ifdef __ARCH_WANT_SYS_CLONE3
2577 noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2578                                               struct clone_args __user *uargs,
2579                                               size_t usize)
2580 {
2581         int err;
2582         struct clone_args args;
2583         pid_t *kset_tid = kargs->set_tid;
2584
2585         if (unlikely(usize > PAGE_SIZE))
2586                 return -E2BIG;
2587         if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2588                 return -EINVAL;
2589
2590         err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2591         if (err)
2592                 return err;
2593
2594         if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2595                 return -EINVAL;
2596
2597         if (unlikely(!args.set_tid && args.set_tid_size > 0))
2598                 return -EINVAL;
2599
2600         if (unlikely(args.set_tid && args.set_tid_size == 0))
2601                 return -EINVAL;
2602
2603         /*
2604          * Verify that higher 32bits of exit_signal are unset and that
2605          * it is a valid signal
2606          */
2607         if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2608                      !valid_signal(args.exit_signal)))
2609                 return -EINVAL;
2610
2611         *kargs = (struct kernel_clone_args){
2612                 .flags          = args.flags,
2613                 .pidfd          = u64_to_user_ptr(args.pidfd),
2614                 .child_tid      = u64_to_user_ptr(args.child_tid),
2615                 .parent_tid     = u64_to_user_ptr(args.parent_tid),
2616                 .exit_signal    = args.exit_signal,
2617                 .stack          = args.stack,
2618                 .stack_size     = args.stack_size,
2619                 .tls            = args.tls,
2620                 .set_tid_size   = args.set_tid_size,
2621         };
2622
2623         if (args.set_tid &&
2624                 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
2625                         (kargs->set_tid_size * sizeof(pid_t))))
2626                 return -EFAULT;
2627
2628         kargs->set_tid = kset_tid;
2629
2630         return 0;
2631 }
2632
2633 /**
2634  * clone3_stack_valid - check and prepare stack
2635  * @kargs: kernel clone args
2636  *
2637  * Verify that the stack arguments userspace gave us are sane.
2638  * In addition, set the stack direction for userspace since it's easy for us to
2639  * determine.
2640  */
2641 static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
2642 {
2643         if (kargs->stack == 0) {
2644                 if (kargs->stack_size > 0)
2645                         return false;
2646         } else {
2647                 if (kargs->stack_size == 0)
2648                         return false;
2649
2650                 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
2651                         return false;
2652
2653 #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2654                 kargs->stack += kargs->stack_size;
2655 #endif
2656         }
2657
2658         return true;
2659 }
2660
2661 static bool clone3_args_valid(struct kernel_clone_args *kargs)
2662 {
2663         /* Verify that no unknown flags are passed along. */
2664         if (kargs->flags & ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND))
2665                 return false;
2666
2667         /*
2668          * - make the CLONE_DETACHED bit reuseable for clone3
2669          * - make the CSIGNAL bits reuseable for clone3
2670          */
2671         if (kargs->flags & (CLONE_DETACHED | CSIGNAL))
2672                 return false;
2673
2674         if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
2675             (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
2676                 return false;
2677
2678         if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
2679             kargs->exit_signal)
2680                 return false;
2681
2682         if (!clone3_stack_valid(kargs))
2683                 return false;
2684
2685         return true;
2686 }
2687
2688 /**
2689  * clone3 - create a new process with specific properties
2690  * @uargs: argument structure
2691  * @size:  size of @uargs
2692  *
2693  * clone3() is the extensible successor to clone()/clone2().
2694  * It takes a struct as argument that is versioned by its size.
2695  *
2696  * Return: On success, a positive PID for the child process.
2697  *         On error, a negative errno number.
2698  */
2699 SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
2700 {
2701         int err;
2702
2703         struct kernel_clone_args kargs;
2704         pid_t set_tid[MAX_PID_NS_LEVEL];
2705
2706         kargs.set_tid = set_tid;
2707
2708         err = copy_clone_args_from_user(&kargs, uargs, size);
2709         if (err)
2710                 return err;
2711
2712         if (!clone3_args_valid(&kargs))
2713                 return -EINVAL;
2714
2715         return _do_fork(&kargs);
2716 }
2717 #endif
2718
2719 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2720 {
2721         struct task_struct *leader, *parent, *child;
2722         int res;
2723
2724         read_lock(&tasklist_lock);
2725         leader = top = top->group_leader;
2726 down:
2727         for_each_thread(leader, parent) {
2728                 list_for_each_entry(child, &parent->children, sibling) {
2729                         res = visitor(child, data);
2730                         if (res) {
2731                                 if (res < 0)
2732                                         goto out;
2733                                 leader = child;
2734                                 goto down;
2735                         }
2736 up:
2737                         ;
2738                 }
2739         }
2740
2741         if (leader != top) {
2742                 child = leader;
2743                 parent = child->real_parent;
2744                 leader = parent->group_leader;
2745                 goto up;
2746         }
2747 out:
2748         read_unlock(&tasklist_lock);
2749 }
2750
2751 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2752 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2753 #endif
2754
2755 static void sighand_ctor(void *data)
2756 {
2757         struct sighand_struct *sighand = data;
2758
2759         spin_lock_init(&sighand->siglock);
2760         init_waitqueue_head(&sighand->signalfd_wqh);
2761 }
2762
2763 void __init proc_caches_init(void)
2764 {
2765         unsigned int mm_size;
2766
2767         sighand_cachep = kmem_cache_create("sighand_cache",
2768                         sizeof(struct sighand_struct), 0,
2769                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2770                         SLAB_ACCOUNT, sighand_ctor);
2771         signal_cachep = kmem_cache_create("signal_cache",
2772                         sizeof(struct signal_struct), 0,
2773                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2774                         NULL);
2775         files_cachep = kmem_cache_create("files_cache",
2776                         sizeof(struct files_struct), 0,
2777                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2778                         NULL);
2779         fs_cachep = kmem_cache_create("fs_cache",
2780                         sizeof(struct fs_struct), 0,
2781                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2782                         NULL);
2783
2784         /*
2785          * The mm_cpumask is located at the end of mm_struct, and is
2786          * dynamically sized based on the maximum CPU number this system
2787          * can have, taking hotplug into account (nr_cpu_ids).
2788          */
2789         mm_size = sizeof(struct mm_struct) + cpumask_size();
2790
2791         mm_cachep = kmem_cache_create_usercopy("mm_struct",
2792                         mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2793                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2794                         offsetof(struct mm_struct, saved_auxv),
2795                         sizeof_field(struct mm_struct, saved_auxv),
2796                         NULL);
2797         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2798         mmap_init();
2799         nsproxy_cache_init();
2800 }
2801
2802 /*
2803  * Check constraints on flags passed to the unshare system call.
2804  */
2805 static int check_unshare_flags(unsigned long unshare_flags)
2806 {
2807         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2808                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2809                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2810                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2811                 return -EINVAL;
2812         /*
2813          * Not implemented, but pretend it works if there is nothing
2814          * to unshare.  Note that unsharing the address space or the
2815          * signal handlers also need to unshare the signal queues (aka
2816          * CLONE_THREAD).
2817          */
2818         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2819                 if (!thread_group_empty(current))
2820                         return -EINVAL;
2821         }
2822         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2823                 if (refcount_read(&current->sighand->count) > 1)
2824                         return -EINVAL;
2825         }
2826         if (unshare_flags & CLONE_VM) {
2827                 if (!current_is_single_threaded())
2828                         return -EINVAL;
2829         }
2830
2831         return 0;
2832 }
2833
2834 /*
2835  * Unshare the filesystem structure if it is being shared
2836  */
2837 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2838 {
2839         struct fs_struct *fs = current->fs;
2840
2841         if (!(unshare_flags & CLONE_FS) || !fs)
2842                 return 0;
2843
2844         /* don't need lock here; in the worst case we'll do useless copy */
2845         if (fs->users == 1)
2846                 return 0;
2847
2848         *new_fsp = copy_fs_struct(fs);
2849         if (!*new_fsp)
2850                 return -ENOMEM;
2851
2852         return 0;
2853 }
2854
2855 /*
2856  * Unshare file descriptor table if it is being shared
2857  */
2858 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2859 {
2860         struct files_struct *fd = current->files;
2861         int error = 0;
2862
2863         if ((unshare_flags & CLONE_FILES) &&
2864             (fd && atomic_read(&fd->count) > 1)) {
2865                 *new_fdp = dup_fd(fd, &error);
2866                 if (!*new_fdp)
2867                         return error;
2868         }
2869
2870         return 0;
2871 }
2872
2873 /*
2874  * unshare allows a process to 'unshare' part of the process
2875  * context which was originally shared using clone.  copy_*
2876  * functions used by do_fork() cannot be used here directly
2877  * because they modify an inactive task_struct that is being
2878  * constructed. Here we are modifying the current, active,
2879  * task_struct.
2880  */
2881 int ksys_unshare(unsigned long unshare_flags)
2882 {
2883         struct fs_struct *fs, *new_fs = NULL;
2884         struct files_struct *fd, *new_fd = NULL;
2885         struct cred *new_cred = NULL;
2886         struct nsproxy *new_nsproxy = NULL;
2887         int do_sysvsem = 0;
2888         int err;
2889
2890         /*
2891          * If unsharing a user namespace must also unshare the thread group
2892          * and unshare the filesystem root and working directories.
2893          */
2894         if (unshare_flags & CLONE_NEWUSER)
2895                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2896         /*
2897          * If unsharing vm, must also unshare signal handlers.
2898          */
2899         if (unshare_flags & CLONE_VM)
2900                 unshare_flags |= CLONE_SIGHAND;
2901         /*
2902          * If unsharing a signal handlers, must also unshare the signal queues.
2903          */
2904         if (unshare_flags & CLONE_SIGHAND)
2905                 unshare_flags |= CLONE_THREAD;
2906         /*
2907          * If unsharing namespace, must also unshare filesystem information.
2908          */
2909         if (unshare_flags & CLONE_NEWNS)
2910                 unshare_flags |= CLONE_FS;
2911
2912         err = check_unshare_flags(unshare_flags);
2913         if (err)
2914                 goto bad_unshare_out;
2915         /*
2916          * CLONE_NEWIPC must also detach from the undolist: after switching
2917          * to a new ipc namespace, the semaphore arrays from the old
2918          * namespace are unreachable.
2919          */
2920         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2921                 do_sysvsem = 1;
2922         err = unshare_fs(unshare_flags, &new_fs);
2923         if (err)
2924                 goto bad_unshare_out;
2925         err = unshare_fd(unshare_flags, &new_fd);
2926         if (err)
2927                 goto bad_unshare_cleanup_fs;
2928         err = unshare_userns(unshare_flags, &new_cred);
2929         if (err)
2930                 goto bad_unshare_cleanup_fd;
2931         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2932                                          new_cred, new_fs);
2933         if (err)
2934                 goto bad_unshare_cleanup_cred;
2935
2936         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2937                 if (do_sysvsem) {
2938                         /*
2939                          * CLONE_SYSVSEM is equivalent to sys_exit().
2940                          */
2941                         exit_sem(current);
2942                 }
2943                 if (unshare_flags & CLONE_NEWIPC) {
2944                         /* Orphan segments in old ns (see sem above). */
2945                         exit_shm(current);
2946                         shm_init_task(current);
2947                 }
2948
2949                 if (new_nsproxy)
2950                         switch_task_namespaces(current, new_nsproxy);
2951
2952                 task_lock(current);
2953
2954                 if (new_fs) {
2955                         fs = current->fs;
2956                         spin_lock(&fs->lock);
2957                         current->fs = new_fs;
2958                         if (--fs->users)
2959                                 new_fs = NULL;
2960                         else
2961                                 new_fs = fs;
2962                         spin_unlock(&fs->lock);
2963                 }
2964
2965                 if (new_fd) {
2966                         fd = current->files;
2967                         current->files = new_fd;
2968                         new_fd = fd;
2969                 }
2970
2971                 task_unlock(current);
2972
2973                 if (new_cred) {
2974                         /* Install the new user namespace */
2975                         commit_creds(new_cred);
2976                         new_cred = NULL;
2977                 }
2978         }
2979
2980         perf_event_namespaces(current);
2981
2982 bad_unshare_cleanup_cred:
2983         if (new_cred)
2984                 put_cred(new_cred);
2985 bad_unshare_cleanup_fd:
2986         if (new_fd)
2987                 put_files_struct(new_fd);
2988
2989 bad_unshare_cleanup_fs:
2990         if (new_fs)
2991                 free_fs_struct(new_fs);
2992
2993 bad_unshare_out:
2994         return err;
2995 }
2996
2997 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2998 {
2999         return ksys_unshare(unshare_flags);
3000 }
3001
3002 /*
3003  *      Helper to unshare the files of the current task.
3004  *      We don't want to expose copy_files internals to
3005  *      the exec layer of the kernel.
3006  */
3007
3008 int unshare_files(struct files_struct **displaced)
3009 {
3010         struct task_struct *task = current;
3011         struct files_struct *copy = NULL;
3012         int error;
3013
3014         error = unshare_fd(CLONE_FILES, &copy);
3015         if (error || !copy) {
3016                 *displaced = NULL;
3017                 return error;
3018         }
3019         *displaced = task->files;
3020         task_lock(task);
3021         task->files = copy;
3022         task_unlock(task);
3023         return 0;
3024 }
3025
3026 int sysctl_max_threads(struct ctl_table *table, int write,
3027                        void __user *buffer, size_t *lenp, loff_t *ppos)
3028 {
3029         struct ctl_table t;
3030         int ret;
3031         int threads = max_threads;
3032         int min = 1;
3033         int max = MAX_THREADS;
3034
3035         t = *table;
3036         t.data = &threads;
3037         t.extra1 = &min;
3038         t.extra2 = &max;
3039
3040         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3041         if (ret || !write)
3042                 return ret;
3043
3044         max_threads = threads;
3045
3046         return 0;
3047 }