Back to home page

LXR

 
 

    


0001 /*
0002  *  linux/kernel/fork.c
0003  *
0004  *  Copyright (C) 1991, 1992  Linus Torvalds
0005  */
0006 
0007 /*
0008  *  'fork.c' contains the help-routines for the 'fork' system call
0009  * (see also entry.S and others).
0010  * Fork is rather simple, once you get the hang of it, but the memory
0011  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
0012  */
0013 
0014 #include <linux/slab.h>
0015 #include <linux/init.h>
0016 #include <linux/unistd.h>
0017 #include <linux/module.h>
0018 #include <linux/vmalloc.h>
0019 #include <linux/completion.h>
0020 #include <linux/personality.h>
0021 #include <linux/mempolicy.h>
0022 #include <linux/sem.h>
0023 #include <linux/file.h>
0024 #include <linux/fdtable.h>
0025 #include <linux/iocontext.h>
0026 #include <linux/key.h>
0027 #include <linux/binfmts.h>
0028 #include <linux/mman.h>
0029 #include <linux/mmu_notifier.h>
0030 #include <linux/fs.h>
0031 #include <linux/mm.h>
0032 #include <linux/vmacache.h>
0033 #include <linux/nsproxy.h>
0034 #include <linux/capability.h>
0035 #include <linux/cpu.h>
0036 #include <linux/cgroup.h>
0037 #include <linux/security.h>
0038 #include <linux/hugetlb.h>
0039 #include <linux/seccomp.h>
0040 #include <linux/swap.h>
0041 #include <linux/syscalls.h>
0042 #include <linux/jiffies.h>
0043 #include <linux/futex.h>
0044 #include <linux/compat.h>
0045 #include <linux/kthread.h>
0046 #include <linux/task_io_accounting_ops.h>
0047 #include <linux/rcupdate.h>
0048 #include <linux/ptrace.h>
0049 #include <linux/mount.h>
0050 #include <linux/audit.h>
0051 #include <linux/memcontrol.h>
0052 #include <linux/ftrace.h>
0053 #include <linux/proc_fs.h>
0054 #include <linux/profile.h>
0055 #include <linux/rmap.h>
0056 #include <linux/ksm.h>
0057 #include <linux/acct.h>
0058 #include <linux/tsacct_kern.h>
0059 #include <linux/cn_proc.h>
0060 #include <linux/freezer.h>
0061 #include <linux/delayacct.h>
0062 #include <linux/taskstats_kern.h>
0063 #include <linux/random.h>
0064 #include <linux/tty.h>
0065 #include <linux/blkdev.h>
0066 #include <linux/fs_struct.h>
0067 #include <linux/magic.h>
0068 #include <linux/perf_event.h>
0069 #include <linux/posix-timers.h>
0070 #include <linux/user-return-notifier.h>
0071 #include <linux/oom.h>
0072 #include <linux/khugepaged.h>
0073 #include <linux/signalfd.h>
0074 #include <linux/uprobes.h>
0075 #include <linux/aio.h>
0076 #include <linux/compiler.h>
0077 #include <linux/sysctl.h>
0078 #include <linux/kcov.h>
0079 
0080 #include <asm/pgtable.h>
0081 #include <asm/pgalloc.h>
0082 #include <linux/uaccess.h>
0083 #include <asm/mmu_context.h>
0084 #include <asm/cacheflush.h>
0085 #include <asm/tlbflush.h>
0086 
0087 #include <trace/events/sched.h>
0088 
0089 #define CREATE_TRACE_POINTS
0090 #include <trace/events/task.h>
0091 
0092 /*
0093  * Minimum number of threads to boot the kernel
0094  */
0095 #define MIN_THREADS 20
0096 
0097 /*
0098  * Maximum number of threads
0099  */
0100 #define MAX_THREADS FUTEX_TID_MASK
0101 
0102 /*
0103  * Protected counters by write_lock_irq(&tasklist_lock)
0104  */
0105 unsigned long total_forks;  /* Handle normal Linux uptimes. */
0106 int nr_threads;         /* The idle threads do not count.. */
0107 
0108 int max_threads;        /* tunable limit on nr_threads */
0109 
0110 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
0111 
0112 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
0113 
0114 #ifdef CONFIG_PROVE_RCU
0115 int lockdep_tasklist_lock_is_held(void)
0116 {
0117     return lockdep_is_held(&tasklist_lock);
0118 }
0119 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
0120 #endif /* #ifdef CONFIG_PROVE_RCU */
0121 
0122 int nr_processes(void)
0123 {
0124     int cpu;
0125     int total = 0;
0126 
0127     for_each_possible_cpu(cpu)
0128         total += per_cpu(process_counts, cpu);
0129 
0130     return total;
0131 }
0132 
0133 void __weak arch_release_task_struct(struct task_struct *tsk)
0134 {
0135 }
0136 
0137 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
0138 static struct kmem_cache *task_struct_cachep;
0139 
0140 static inline struct task_struct *alloc_task_struct_node(int node)
0141 {
0142     return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
0143 }
0144 
0145 static inline void free_task_struct(struct task_struct *tsk)
0146 {
0147     kmem_cache_free(task_struct_cachep, tsk);
0148 }
0149 #endif
0150 
0151 void __weak arch_release_thread_stack(unsigned long *stack)
0152 {
0153 }
0154 
0155 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
0156 
0157 /*
0158  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
0159  * kmemcache based allocator.
0160  */
0161 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
0162 
0163 #ifdef CONFIG_VMAP_STACK
0164 /*
0165  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
0166  * flush.  Try to minimize the number of calls by caching stacks.
0167  */
0168 #define NR_CACHED_STACKS 2
0169 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
0170 #endif
0171 
0172 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
0173 {
0174 #ifdef CONFIG_VMAP_STACK
0175     void *stack;
0176     int i;
0177 
0178     local_irq_disable();
0179     for (i = 0; i < NR_CACHED_STACKS; i++) {
0180         struct vm_struct *s = this_cpu_read(cached_stacks[i]);
0181 
0182         if (!s)
0183             continue;
0184         this_cpu_write(cached_stacks[i], NULL);
0185 
0186         tsk->stack_vm_area = s;
0187         local_irq_enable();
0188         return s->addr;
0189     }
0190     local_irq_enable();
0191 
0192     stack = __vmalloc_node_range(THREAD_SIZE, THREAD_SIZE,
0193                      VMALLOC_START, VMALLOC_END,
0194                      THREADINFO_GFP | __GFP_HIGHMEM,
0195                      PAGE_KERNEL,
0196                      0, node, __builtin_return_address(0));
0197 
0198     /*
0199      * We can't call find_vm_area() in interrupt context, and
0200      * free_thread_stack() can be called in interrupt context,
0201      * so cache the vm_struct.
0202      */
0203     if (stack)
0204         tsk->stack_vm_area = find_vm_area(stack);
0205     return stack;
0206 #else
0207     struct page *page = alloc_pages_node(node, THREADINFO_GFP,
0208                          THREAD_SIZE_ORDER);
0209 
0210     return page ? page_address(page) : NULL;
0211 #endif
0212 }
0213 
0214 static inline void free_thread_stack(struct task_struct *tsk)
0215 {
0216 #ifdef CONFIG_VMAP_STACK
0217     if (task_stack_vm_area(tsk)) {
0218         unsigned long flags;
0219         int i;
0220 
0221         local_irq_save(flags);
0222         for (i = 0; i < NR_CACHED_STACKS; i++) {
0223             if (this_cpu_read(cached_stacks[i]))
0224                 continue;
0225 
0226             this_cpu_write(cached_stacks[i], tsk->stack_vm_area);
0227             local_irq_restore(flags);
0228             return;
0229         }
0230         local_irq_restore(flags);
0231 
0232         vfree_atomic(tsk->stack);
0233         return;
0234     }
0235 #endif
0236 
0237     __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
0238 }
0239 # else
0240 static struct kmem_cache *thread_stack_cache;
0241 
0242 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
0243                           int node)
0244 {
0245     return kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
0246 }
0247 
0248 static void free_thread_stack(struct task_struct *tsk)
0249 {
0250     kmem_cache_free(thread_stack_cache, tsk->stack);
0251 }
0252 
0253 void thread_stack_cache_init(void)
0254 {
0255     thread_stack_cache = kmem_cache_create("thread_stack", THREAD_SIZE,
0256                           THREAD_SIZE, 0, NULL);
0257     BUG_ON(thread_stack_cache == NULL);
0258 }
0259 # endif
0260 #endif
0261 
0262 /* SLAB cache for signal_struct structures (tsk->signal) */
0263 static struct kmem_cache *signal_cachep;
0264 
0265 /* SLAB cache for sighand_struct structures (tsk->sighand) */
0266 struct kmem_cache *sighand_cachep;
0267 
0268 /* SLAB cache for files_struct structures (tsk->files) */
0269 struct kmem_cache *files_cachep;
0270 
0271 /* SLAB cache for fs_struct structures (tsk->fs) */
0272 struct kmem_cache *fs_cachep;
0273 
0274 /* SLAB cache for vm_area_struct structures */
0275 struct kmem_cache *vm_area_cachep;
0276 
0277 /* SLAB cache for mm_struct structures (tsk->mm) */
0278 static struct kmem_cache *mm_cachep;
0279 
0280 static void account_kernel_stack(struct task_struct *tsk, int account)
0281 {
0282     void *stack = task_stack_page(tsk);
0283     struct vm_struct *vm = task_stack_vm_area(tsk);
0284 
0285     BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
0286 
0287     if (vm) {
0288         int i;
0289 
0290         BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
0291 
0292         for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
0293             mod_zone_page_state(page_zone(vm->pages[i]),
0294                         NR_KERNEL_STACK_KB,
0295                         PAGE_SIZE / 1024 * account);
0296         }
0297 
0298         /* All stack pages belong to the same memcg. */
0299         memcg_kmem_update_page_stat(vm->pages[0], MEMCG_KERNEL_STACK_KB,
0300                         account * (THREAD_SIZE / 1024));
0301     } else {
0302         /*
0303          * All stack pages are in the same zone and belong to the
0304          * same memcg.
0305          */
0306         struct page *first_page = virt_to_page(stack);
0307 
0308         mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
0309                     THREAD_SIZE / 1024 * account);
0310 
0311         memcg_kmem_update_page_stat(first_page, MEMCG_KERNEL_STACK_KB,
0312                         account * (THREAD_SIZE / 1024));
0313     }
0314 }
0315 
0316 static void release_task_stack(struct task_struct *tsk)
0317 {
0318     if (WARN_ON(tsk->state != TASK_DEAD))
0319         return;  /* Better to leak the stack than to free prematurely */
0320 
0321     account_kernel_stack(tsk, -1);
0322     arch_release_thread_stack(tsk->stack);
0323     free_thread_stack(tsk);
0324     tsk->stack = NULL;
0325 #ifdef CONFIG_VMAP_STACK
0326     tsk->stack_vm_area = NULL;
0327 #endif
0328 }
0329 
0330 #ifdef CONFIG_THREAD_INFO_IN_TASK
0331 void put_task_stack(struct task_struct *tsk)
0332 {
0333     if (atomic_dec_and_test(&tsk->stack_refcount))
0334         release_task_stack(tsk);
0335 }
0336 #endif
0337 
0338 void free_task(struct task_struct *tsk)
0339 {
0340 #ifndef CONFIG_THREAD_INFO_IN_TASK
0341     /*
0342      * The task is finally done with both the stack and thread_info,
0343      * so free both.
0344      */
0345     release_task_stack(tsk);
0346 #else
0347     /*
0348      * If the task had a separate stack allocation, it should be gone
0349      * by now.
0350      */
0351     WARN_ON_ONCE(atomic_read(&tsk->stack_refcount) != 0);
0352 #endif
0353     rt_mutex_debug_task_free(tsk);
0354     ftrace_graph_exit_task(tsk);
0355     put_seccomp_filter(tsk);
0356     arch_release_task_struct(tsk);
0357     if (tsk->flags & PF_KTHREAD)
0358         free_kthread_struct(tsk);
0359     free_task_struct(tsk);
0360 }
0361 EXPORT_SYMBOL(free_task);
0362 
0363 static inline void free_signal_struct(struct signal_struct *sig)
0364 {
0365     taskstats_tgid_free(sig);
0366     sched_autogroup_exit(sig);
0367     /*
0368      * __mmdrop is not safe to call from softirq context on x86 due to
0369      * pgd_dtor so postpone it to the async context
0370      */
0371     if (sig->oom_mm)
0372         mmdrop_async(sig->oom_mm);
0373     kmem_cache_free(signal_cachep, sig);
0374 }
0375 
0376 static inline void put_signal_struct(struct signal_struct *sig)
0377 {
0378     if (atomic_dec_and_test(&sig->sigcnt))
0379         free_signal_struct(sig);
0380 }
0381 
0382 void __put_task_struct(struct task_struct *tsk)
0383 {
0384     WARN_ON(!tsk->exit_state);
0385     WARN_ON(atomic_read(&tsk->usage));
0386     WARN_ON(tsk == current);
0387 
0388     cgroup_free(tsk);
0389     task_numa_free(tsk);
0390     security_task_free(tsk);
0391     exit_creds(tsk);
0392     delayacct_tsk_free(tsk);
0393     put_signal_struct(tsk->signal);
0394 
0395     if (!profile_handoff_task(tsk))
0396         free_task(tsk);
0397 }
0398 EXPORT_SYMBOL_GPL(__put_task_struct);
0399 
0400 void __init __weak arch_task_cache_init(void) { }
0401 
0402 /*
0403  * set_max_threads
0404  */
0405 static void set_max_threads(unsigned int max_threads_suggested)
0406 {
0407     u64 threads;
0408 
0409     /*
0410      * The number of threads shall be limited such that the thread
0411      * structures may only consume a small part of the available memory.
0412      */
0413     if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
0414         threads = MAX_THREADS;
0415     else
0416         threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
0417                     (u64) THREAD_SIZE * 8UL);
0418 
0419     if (threads > max_threads_suggested)
0420         threads = max_threads_suggested;
0421 
0422     max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
0423 }
0424 
0425 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
0426 /* Initialized by the architecture: */
0427 int arch_task_struct_size __read_mostly;
0428 #endif
0429 
0430 void __init fork_init(void)
0431 {
0432     int i;
0433 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
0434 #ifndef ARCH_MIN_TASKALIGN
0435 #define ARCH_MIN_TASKALIGN  L1_CACHE_BYTES
0436 #endif
0437     /* create a slab on which task_structs can be allocated */
0438     task_struct_cachep = kmem_cache_create("task_struct",
0439             arch_task_struct_size, ARCH_MIN_TASKALIGN,
0440             SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT, NULL);
0441 #endif
0442 
0443     /* do the arch specific task caches init */
0444     arch_task_cache_init();
0445 
0446     set_max_threads(MAX_THREADS);
0447 
0448     init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
0449     init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
0450     init_task.signal->rlim[RLIMIT_SIGPENDING] =
0451         init_task.signal->rlim[RLIMIT_NPROC];
0452 
0453     for (i = 0; i < UCOUNT_COUNTS; i++) {
0454         init_user_ns.ucount_max[i] = max_threads/2;
0455     }
0456 }
0457 
0458 int __weak arch_dup_task_struct(struct task_struct *dst,
0459                            struct task_struct *src)
0460 {
0461     *dst = *src;
0462     return 0;
0463 }
0464 
0465 void set_task_stack_end_magic(struct task_struct *tsk)
0466 {
0467     unsigned long *stackend;
0468 
0469     stackend = end_of_stack(tsk);
0470     *stackend = STACK_END_MAGIC;    /* for overflow detection */
0471 }
0472 
0473 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
0474 {
0475     struct task_struct *tsk;
0476     unsigned long *stack;
0477     struct vm_struct *stack_vm_area;
0478     int err;
0479 
0480     if (node == NUMA_NO_NODE)
0481         node = tsk_fork_get_node(orig);
0482     tsk = alloc_task_struct_node(node);
0483     if (!tsk)
0484         return NULL;
0485 
0486     stack = alloc_thread_stack_node(tsk, node);
0487     if (!stack)
0488         goto free_tsk;
0489 
0490     stack_vm_area = task_stack_vm_area(tsk);
0491 
0492     err = arch_dup_task_struct(tsk, orig);
0493 
0494     /*
0495      * arch_dup_task_struct() clobbers the stack-related fields.  Make
0496      * sure they're properly initialized before using any stack-related
0497      * functions again.
0498      */
0499     tsk->stack = stack;
0500 #ifdef CONFIG_VMAP_STACK
0501     tsk->stack_vm_area = stack_vm_area;
0502 #endif
0503 #ifdef CONFIG_THREAD_INFO_IN_TASK
0504     atomic_set(&tsk->stack_refcount, 1);
0505 #endif
0506 
0507     if (err)
0508         goto free_stack;
0509 
0510 #ifdef CONFIG_SECCOMP
0511     /*
0512      * We must handle setting up seccomp filters once we're under
0513      * the sighand lock in case orig has changed between now and
0514      * then. Until then, filter must be NULL to avoid messing up
0515      * the usage counts on the error path calling free_task.
0516      */
0517     tsk->seccomp.filter = NULL;
0518 #endif
0519 
0520     setup_thread_stack(tsk, orig);
0521     clear_user_return_notifier(tsk);
0522     clear_tsk_need_resched(tsk);
0523     set_task_stack_end_magic(tsk);
0524 
0525 #ifdef CONFIG_CC_STACKPROTECTOR
0526     tsk->stack_canary = get_random_int();
0527 #endif
0528 
0529     /*
0530      * One for us, one for whoever does the "release_task()" (usually
0531      * parent)
0532      */
0533     atomic_set(&tsk->usage, 2);
0534 #ifdef CONFIG_BLK_DEV_IO_TRACE
0535     tsk->btrace_seq = 0;
0536 #endif
0537     tsk->splice_pipe = NULL;
0538     tsk->task_frag.page = NULL;
0539     tsk->wake_q.next = NULL;
0540 
0541     account_kernel_stack(tsk, 1);
0542 
0543     kcov_task_init(tsk);
0544 
0545     return tsk;
0546 
0547 free_stack:
0548     free_thread_stack(tsk);
0549 free_tsk:
0550     free_task_struct(tsk);
0551     return NULL;
0552 }
0553 
0554 #ifdef CONFIG_MMU
0555 static __latent_entropy int dup_mmap(struct mm_struct *mm,
0556                     struct mm_struct *oldmm)
0557 {
0558     struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
0559     struct rb_node **rb_link, *rb_parent;
0560     int retval;
0561     unsigned long charge;
0562 
0563     uprobe_start_dup_mmap();
0564     if (down_write_killable(&oldmm->mmap_sem)) {
0565         retval = -EINTR;
0566         goto fail_uprobe_end;
0567     }
0568     flush_cache_dup_mm(oldmm);
0569     uprobe_dup_mmap(oldmm, mm);
0570     /*
0571      * Not linked in yet - no deadlock potential:
0572      */
0573     down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
0574 
0575     /* No ordering required: file already has been exposed. */
0576     RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
0577 
0578     mm->total_vm = oldmm->total_vm;
0579     mm->data_vm = oldmm->data_vm;
0580     mm->exec_vm = oldmm->exec_vm;
0581     mm->stack_vm = oldmm->stack_vm;
0582 
0583     rb_link = &mm->mm_rb.rb_node;
0584     rb_parent = NULL;
0585     pprev = &mm->mmap;
0586     retval = ksm_fork(mm, oldmm);
0587     if (retval)
0588         goto out;
0589     retval = khugepaged_fork(mm, oldmm);
0590     if (retval)
0591         goto out;
0592 
0593     prev = NULL;
0594     for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
0595         struct file *file;
0596 
0597         if (mpnt->vm_flags & VM_DONTCOPY) {
0598             vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
0599             continue;
0600         }
0601         charge = 0;
0602         if (mpnt->vm_flags & VM_ACCOUNT) {
0603             unsigned long len = vma_pages(mpnt);
0604 
0605             if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
0606                 goto fail_nomem;
0607             charge = len;
0608         }
0609         tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
0610         if (!tmp)
0611             goto fail_nomem;
0612         *tmp = *mpnt;
0613         INIT_LIST_HEAD(&tmp->anon_vma_chain);
0614         retval = vma_dup_policy(mpnt, tmp);
0615         if (retval)
0616             goto fail_nomem_policy;
0617         tmp->vm_mm = mm;
0618         if (anon_vma_fork(tmp, mpnt))
0619             goto fail_nomem_anon_vma_fork;
0620         tmp->vm_flags &=
0621             ~(VM_LOCKED|VM_LOCKONFAULT|VM_UFFD_MISSING|VM_UFFD_WP);
0622         tmp->vm_next = tmp->vm_prev = NULL;
0623         tmp->vm_userfaultfd_ctx = NULL_VM_UFFD_CTX;
0624         file = tmp->vm_file;
0625         if (file) {
0626             struct inode *inode = file_inode(file);
0627             struct address_space *mapping = file->f_mapping;
0628 
0629             get_file(file);
0630             if (tmp->vm_flags & VM_DENYWRITE)
0631                 atomic_dec(&inode->i_writecount);
0632             i_mmap_lock_write(mapping);
0633             if (tmp->vm_flags & VM_SHARED)
0634                 atomic_inc(&mapping->i_mmap_writable);
0635             flush_dcache_mmap_lock(mapping);
0636             /* insert tmp into the share list, just after mpnt */
0637             vma_interval_tree_insert_after(tmp, mpnt,
0638                     &mapping->i_mmap);
0639             flush_dcache_mmap_unlock(mapping);
0640             i_mmap_unlock_write(mapping);
0641         }
0642 
0643         /*
0644          * Clear hugetlb-related page reserves for children. This only
0645          * affects MAP_PRIVATE mappings. Faults generated by the child
0646          * are not guaranteed to succeed, even if read-only
0647          */
0648         if (is_vm_hugetlb_page(tmp))
0649             reset_vma_resv_huge_pages(tmp);
0650 
0651         /*
0652          * Link in the new vma and copy the page table entries.
0653          */
0654         *pprev = tmp;
0655         pprev = &tmp->vm_next;
0656         tmp->vm_prev = prev;
0657         prev = tmp;
0658 
0659         __vma_link_rb(mm, tmp, rb_link, rb_parent);
0660         rb_link = &tmp->vm_rb.rb_right;
0661         rb_parent = &tmp->vm_rb;
0662 
0663         mm->map_count++;
0664         retval = copy_page_range(mm, oldmm, mpnt);
0665 
0666         if (tmp->vm_ops && tmp->vm_ops->open)
0667             tmp->vm_ops->open(tmp);
0668 
0669         if (retval)
0670             goto out;
0671     }
0672     /* a new mm has just been created */
0673     arch_dup_mmap(oldmm, mm);
0674     retval = 0;
0675 out:
0676     up_write(&mm->mmap_sem);
0677     flush_tlb_mm(oldmm);
0678     up_write(&oldmm->mmap_sem);
0679 fail_uprobe_end:
0680     uprobe_end_dup_mmap();
0681     return retval;
0682 fail_nomem_anon_vma_fork:
0683     mpol_put(vma_policy(tmp));
0684 fail_nomem_policy:
0685     kmem_cache_free(vm_area_cachep, tmp);
0686 fail_nomem:
0687     retval = -ENOMEM;
0688     vm_unacct_memory(charge);
0689     goto out;
0690 }
0691 
0692 static inline int mm_alloc_pgd(struct mm_struct *mm)
0693 {
0694     mm->pgd = pgd_alloc(mm);
0695     if (unlikely(!mm->pgd))
0696         return -ENOMEM;
0697     return 0;
0698 }
0699 
0700 static inline void mm_free_pgd(struct mm_struct *mm)
0701 {
0702     pgd_free(mm, mm->pgd);
0703 }
0704 #else
0705 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
0706 {
0707     down_write(&oldmm->mmap_sem);
0708     RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
0709     up_write(&oldmm->mmap_sem);
0710     return 0;
0711 }
0712 #define mm_alloc_pgd(mm)    (0)
0713 #define mm_free_pgd(mm)
0714 #endif /* CONFIG_MMU */
0715 
0716 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
0717 
0718 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
0719 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
0720 
0721 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
0722 
0723 static int __init coredump_filter_setup(char *s)
0724 {
0725     default_dump_filter =
0726         (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
0727         MMF_DUMP_FILTER_MASK;
0728     return 1;
0729 }
0730 
0731 __setup("coredump_filter=", coredump_filter_setup);
0732 
0733 #include <linux/init_task.h>
0734 
0735 static void mm_init_aio(struct mm_struct *mm)
0736 {
0737 #ifdef CONFIG_AIO
0738     spin_lock_init(&mm->ioctx_lock);
0739     mm->ioctx_table = NULL;
0740 #endif
0741 }
0742 
0743 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
0744 {
0745 #ifdef CONFIG_MEMCG
0746     mm->owner = p;
0747 #endif
0748 }
0749 
0750 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
0751     struct user_namespace *user_ns)
0752 {
0753     mm->mmap = NULL;
0754     mm->mm_rb = RB_ROOT;
0755     mm->vmacache_seqnum = 0;
0756     atomic_set(&mm->mm_users, 1);
0757     atomic_set(&mm->mm_count, 1);
0758     init_rwsem(&mm->mmap_sem);
0759     INIT_LIST_HEAD(&mm->mmlist);
0760     mm->core_state = NULL;
0761     atomic_long_set(&mm->nr_ptes, 0);
0762     mm_nr_pmds_init(mm);
0763     mm->map_count = 0;
0764     mm->locked_vm = 0;
0765     mm->pinned_vm = 0;
0766     memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
0767     spin_lock_init(&mm->page_table_lock);
0768     mm_init_cpumask(mm);
0769     mm_init_aio(mm);
0770     mm_init_owner(mm, p);
0771     mmu_notifier_mm_init(mm);
0772     clear_tlb_flush_pending(mm);
0773 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
0774     mm->pmd_huge_pte = NULL;
0775 #endif
0776 
0777     if (current->mm) {
0778         mm->flags = current->mm->flags & MMF_INIT_MASK;
0779         mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
0780     } else {
0781         mm->flags = default_dump_filter;
0782         mm->def_flags = 0;
0783     }
0784 
0785     if (mm_alloc_pgd(mm))
0786         goto fail_nopgd;
0787 
0788     if (init_new_context(p, mm))
0789         goto fail_nocontext;
0790 
0791     mm->user_ns = get_user_ns(user_ns);
0792     return mm;
0793 
0794 fail_nocontext:
0795     mm_free_pgd(mm);
0796 fail_nopgd:
0797     free_mm(mm);
0798     return NULL;
0799 }
0800 
0801 static void check_mm(struct mm_struct *mm)
0802 {
0803     int i;
0804 
0805     for (i = 0; i < NR_MM_COUNTERS; i++) {
0806         long x = atomic_long_read(&mm->rss_stat.count[i]);
0807 
0808         if (unlikely(x))
0809             printk(KERN_ALERT "BUG: Bad rss-counter state "
0810                       "mm:%p idx:%d val:%ld\n", mm, i, x);
0811     }
0812 
0813     if (atomic_long_read(&mm->nr_ptes))
0814         pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
0815                 atomic_long_read(&mm->nr_ptes));
0816     if (mm_nr_pmds(mm))
0817         pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
0818                 mm_nr_pmds(mm));
0819 
0820 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
0821     VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
0822 #endif
0823 }
0824 
0825 /*
0826  * Allocate and initialize an mm_struct.
0827  */
0828 struct mm_struct *mm_alloc(void)
0829 {
0830     struct mm_struct *mm;
0831 
0832     mm = allocate_mm();
0833     if (!mm)
0834         return NULL;
0835 
0836     memset(mm, 0, sizeof(*mm));
0837     return mm_init(mm, current, current_user_ns());
0838 }
0839 
0840 /*
0841  * Called when the last reference to the mm
0842  * is dropped: either by a lazy thread or by
0843  * mmput. Free the page directory and the mm.
0844  */
0845 void __mmdrop(struct mm_struct *mm)
0846 {
0847     BUG_ON(mm == &init_mm);
0848     mm_free_pgd(mm);
0849     destroy_context(mm);
0850     mmu_notifier_mm_destroy(mm);
0851     check_mm(mm);
0852     put_user_ns(mm->user_ns);
0853     free_mm(mm);
0854 }
0855 EXPORT_SYMBOL_GPL(__mmdrop);
0856 
0857 static inline void __mmput(struct mm_struct *mm)
0858 {
0859     VM_BUG_ON(atomic_read(&mm->mm_users));
0860 
0861     uprobe_clear_state(mm);
0862     exit_aio(mm);
0863     ksm_exit(mm);
0864     khugepaged_exit(mm); /* must run before exit_mmap */
0865     exit_mmap(mm);
0866     mm_put_huge_zero_page(mm);
0867     set_mm_exe_file(mm, NULL);
0868     if (!list_empty(&mm->mmlist)) {
0869         spin_lock(&mmlist_lock);
0870         list_del(&mm->mmlist);
0871         spin_unlock(&mmlist_lock);
0872     }
0873     if (mm->binfmt)
0874         module_put(mm->binfmt->module);
0875     set_bit(MMF_OOM_SKIP, &mm->flags);
0876     mmdrop(mm);
0877 }
0878 
0879 /*
0880  * Decrement the use count and release all resources for an mm.
0881  */
0882 void mmput(struct mm_struct *mm)
0883 {
0884     might_sleep();
0885 
0886     if (atomic_dec_and_test(&mm->mm_users))
0887         __mmput(mm);
0888 }
0889 EXPORT_SYMBOL_GPL(mmput);
0890 
0891 #ifdef CONFIG_MMU
0892 static void mmput_async_fn(struct work_struct *work)
0893 {
0894     struct mm_struct *mm = container_of(work, struct mm_struct, async_put_work);
0895     __mmput(mm);
0896 }
0897 
0898 void mmput_async(struct mm_struct *mm)
0899 {
0900     if (atomic_dec_and_test(&mm->mm_users)) {
0901         INIT_WORK(&mm->async_put_work, mmput_async_fn);
0902         schedule_work(&mm->async_put_work);
0903     }
0904 }
0905 #endif
0906 
0907 /**
0908  * set_mm_exe_file - change a reference to the mm's executable file
0909  *
0910  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
0911  *
0912  * Main users are mmput() and sys_execve(). Callers prevent concurrent
0913  * invocations: in mmput() nobody alive left, in execve task is single
0914  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
0915  * mm->exe_file, but does so without using set_mm_exe_file() in order
0916  * to do avoid the need for any locks.
0917  */
0918 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
0919 {
0920     struct file *old_exe_file;
0921 
0922     /*
0923      * It is safe to dereference the exe_file without RCU as
0924      * this function is only called if nobody else can access
0925      * this mm -- see comment above for justification.
0926      */
0927     old_exe_file = rcu_dereference_raw(mm->exe_file);
0928 
0929     if (new_exe_file)
0930         get_file(new_exe_file);
0931     rcu_assign_pointer(mm->exe_file, new_exe_file);
0932     if (old_exe_file)
0933         fput(old_exe_file);
0934 }
0935 
0936 /**
0937  * get_mm_exe_file - acquire a reference to the mm's executable file
0938  *
0939  * Returns %NULL if mm has no associated executable file.
0940  * User must release file via fput().
0941  */
0942 struct file *get_mm_exe_file(struct mm_struct *mm)
0943 {
0944     struct file *exe_file;
0945 
0946     rcu_read_lock();
0947     exe_file = rcu_dereference(mm->exe_file);
0948     if (exe_file && !get_file_rcu(exe_file))
0949         exe_file = NULL;
0950     rcu_read_unlock();
0951     return exe_file;
0952 }
0953 EXPORT_SYMBOL(get_mm_exe_file);
0954 
0955 /**
0956  * get_task_exe_file - acquire a reference to the task's executable file
0957  *
0958  * Returns %NULL if task's mm (if any) has no associated executable file or
0959  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
0960  * User must release file via fput().
0961  */
0962 struct file *get_task_exe_file(struct task_struct *task)
0963 {
0964     struct file *exe_file = NULL;
0965     struct mm_struct *mm;
0966 
0967     task_lock(task);
0968     mm = task->mm;
0969     if (mm) {
0970         if (!(task->flags & PF_KTHREAD))
0971             exe_file = get_mm_exe_file(mm);
0972     }
0973     task_unlock(task);
0974     return exe_file;
0975 }
0976 EXPORT_SYMBOL(get_task_exe_file);
0977 
0978 /**
0979  * get_task_mm - acquire a reference to the task's mm
0980  *
0981  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
0982  * this kernel workthread has transiently adopted a user mm with use_mm,
0983  * to do its AIO) is not set and if so returns a reference to it, after
0984  * bumping up the use count.  User must release the mm via mmput()
0985  * after use.  Typically used by /proc and ptrace.
0986  */
0987 struct mm_struct *get_task_mm(struct task_struct *task)
0988 {
0989     struct mm_struct *mm;
0990 
0991     task_lock(task);
0992     mm = task->mm;
0993     if (mm) {
0994         if (task->flags & PF_KTHREAD)
0995             mm = NULL;
0996         else
0997             atomic_inc(&mm->mm_users);
0998     }
0999     task_unlock(task);
1000     return mm;
1001 }
1002 EXPORT_SYMBOL_GPL(get_task_mm);
1003 
1004 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1005 {
1006     struct mm_struct *mm;
1007     int err;
1008 
1009     err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
1010     if (err)
1011         return ERR_PTR(err);
1012 
1013     mm = get_task_mm(task);
1014     if (mm && mm != current->mm &&
1015             !ptrace_may_access(task, mode)) {
1016         mmput(mm);
1017         mm = ERR_PTR(-EACCES);
1018     }
1019     mutex_unlock(&task->signal->cred_guard_mutex);
1020 
1021     return mm;
1022 }
1023 
1024 static void complete_vfork_done(struct task_struct *tsk)
1025 {
1026     struct completion *vfork;
1027 
1028     task_lock(tsk);
1029     vfork = tsk->vfork_done;
1030     if (likely(vfork)) {
1031         tsk->vfork_done = NULL;
1032         complete(vfork);
1033     }
1034     task_unlock(tsk);
1035 }
1036 
1037 static int wait_for_vfork_done(struct task_struct *child,
1038                 struct completion *vfork)
1039 {
1040     int killed;
1041 
1042     freezer_do_not_count();
1043     killed = wait_for_completion_killable(vfork);
1044     freezer_count();
1045 
1046     if (killed) {
1047         task_lock(child);
1048         child->vfork_done = NULL;
1049         task_unlock(child);
1050     }
1051 
1052     put_task_struct(child);
1053     return killed;
1054 }
1055 
1056 /* Please note the differences between mmput and mm_release.
1057  * mmput is called whenever we stop holding onto a mm_struct,
1058  * error success whatever.
1059  *
1060  * mm_release is called after a mm_struct has been removed
1061  * from the current process.
1062  *
1063  * This difference is important for error handling, when we
1064  * only half set up a mm_struct for a new process and need to restore
1065  * the old one.  Because we mmput the new mm_struct before
1066  * restoring the old one. . .
1067  * Eric Biederman 10 January 1998
1068  */
1069 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1070 {
1071     /* Get rid of any futexes when releasing the mm */
1072 #ifdef CONFIG_FUTEX
1073     if (unlikely(tsk->robust_list)) {
1074         exit_robust_list(tsk);
1075         tsk->robust_list = NULL;
1076     }
1077 #ifdef CONFIG_COMPAT
1078     if (unlikely(tsk->compat_robust_list)) {
1079         compat_exit_robust_list(tsk);
1080         tsk->compat_robust_list = NULL;
1081     }
1082 #endif
1083     if (unlikely(!list_empty(&tsk->pi_state_list)))
1084         exit_pi_state_list(tsk);
1085 #endif
1086 
1087     uprobe_free_utask(tsk);
1088 
1089     /* Get rid of any cached register state */
1090     deactivate_mm(tsk, mm);
1091 
1092     /*
1093      * Signal userspace if we're not exiting with a core dump
1094      * because we want to leave the value intact for debugging
1095      * purposes.
1096      */
1097     if (tsk->clear_child_tid) {
1098         if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1099             atomic_read(&mm->mm_users) > 1) {
1100             /*
1101              * We don't check the error code - if userspace has
1102              * not set up a proper pointer then tough luck.
1103              */
1104             put_user(0, tsk->clear_child_tid);
1105             sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
1106                     1, NULL, NULL, 0);
1107         }
1108         tsk->clear_child_tid = NULL;
1109     }
1110 
1111     /*
1112      * All done, finally we can wake up parent and return this mm to him.
1113      * Also kthread_stop() uses this completion for synchronization.
1114      */
1115     if (tsk->vfork_done)
1116         complete_vfork_done(tsk);
1117 }
1118 
1119 /*
1120  * Allocate a new mm structure and copy contents from the
1121  * mm structure of the passed in task structure.
1122  */
1123 static struct mm_struct *dup_mm(struct task_struct *tsk)
1124 {
1125     struct mm_struct *mm, *oldmm = current->mm;
1126     int err;
1127 
1128     mm = allocate_mm();
1129     if (!mm)
1130         goto fail_nomem;
1131 
1132     memcpy(mm, oldmm, sizeof(*mm));
1133 
1134     if (!mm_init(mm, tsk, mm->user_ns))
1135         goto fail_nomem;
1136 
1137     err = dup_mmap(mm, oldmm);
1138     if (err)
1139         goto free_pt;
1140 
1141     mm->hiwater_rss = get_mm_rss(mm);
1142     mm->hiwater_vm = mm->total_vm;
1143 
1144     if (mm->binfmt && !try_module_get(mm->binfmt->module))
1145         goto free_pt;
1146 
1147     return mm;
1148 
1149 free_pt:
1150     /* don't put binfmt in mmput, we haven't got module yet */
1151     mm->binfmt = NULL;
1152     mmput(mm);
1153 
1154 fail_nomem:
1155     return NULL;
1156 }
1157 
1158 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1159 {
1160     struct mm_struct *mm, *oldmm;
1161     int retval;
1162 
1163     tsk->min_flt = tsk->maj_flt = 0;
1164     tsk->nvcsw = tsk->nivcsw = 0;
1165 #ifdef CONFIG_DETECT_HUNG_TASK
1166     tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1167 #endif
1168 
1169     tsk->mm = NULL;
1170     tsk->active_mm = NULL;
1171 
1172     /*
1173      * Are we cloning a kernel thread?
1174      *
1175      * We need to steal a active VM for that..
1176      */
1177     oldmm = current->mm;
1178     if (!oldmm)
1179         return 0;
1180 
1181     /* initialize the new vmacache entries */
1182     vmacache_flush(tsk);
1183 
1184     if (clone_flags & CLONE_VM) {
1185         atomic_inc(&oldmm->mm_users);
1186         mm = oldmm;
1187         goto good_mm;
1188     }
1189 
1190     retval = -ENOMEM;
1191     mm = dup_mm(tsk);
1192     if (!mm)
1193         goto fail_nomem;
1194 
1195 good_mm:
1196     tsk->mm = mm;
1197     tsk->active_mm = mm;
1198     return 0;
1199 
1200 fail_nomem:
1201     return retval;
1202 }
1203 
1204 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1205 {
1206     struct fs_struct *fs = current->fs;
1207     if (clone_flags & CLONE_FS) {
1208         /* tsk->fs is already what we want */
1209         spin_lock(&fs->lock);
1210         if (fs->in_exec) {
1211             spin_unlock(&fs->lock);
1212             return -EAGAIN;
1213         }
1214         fs->users++;
1215         spin_unlock(&fs->lock);
1216         return 0;
1217     }
1218     tsk->fs = copy_fs_struct(fs);
1219     if (!tsk->fs)
1220         return -ENOMEM;
1221     return 0;
1222 }
1223 
1224 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1225 {
1226     struct files_struct *oldf, *newf;
1227     int error = 0;
1228 
1229     /*
1230      * A background process may not have any files ...
1231      */
1232     oldf = current->files;
1233     if (!oldf)
1234         goto out;
1235 
1236     if (clone_flags & CLONE_FILES) {
1237         atomic_inc(&oldf->count);
1238         goto out;
1239     }
1240 
1241     newf = dup_fd(oldf, &error);
1242     if (!newf)
1243         goto out;
1244 
1245     tsk->files = newf;
1246     error = 0;
1247 out:
1248     return error;
1249 }
1250 
1251 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1252 {
1253 #ifdef CONFIG_BLOCK
1254     struct io_context *ioc = current->io_context;
1255     struct io_context *new_ioc;
1256 
1257     if (!ioc)
1258         return 0;
1259     /*
1260      * Share io context with parent, if CLONE_IO is set
1261      */
1262     if (clone_flags & CLONE_IO) {
1263         ioc_task_link(ioc);
1264         tsk->io_context = ioc;
1265     } else if (ioprio_valid(ioc->ioprio)) {
1266         new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1267         if (unlikely(!new_ioc))
1268             return -ENOMEM;
1269 
1270         new_ioc->ioprio = ioc->ioprio;
1271         put_io_context(new_ioc);
1272     }
1273 #endif
1274     return 0;
1275 }
1276 
1277 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1278 {
1279     struct sighand_struct *sig;
1280 
1281     if (clone_flags & CLONE_SIGHAND) {
1282         atomic_inc(&current->sighand->count);
1283         return 0;
1284     }
1285     sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1286     rcu_assign_pointer(tsk->sighand, sig);
1287     if (!sig)
1288         return -ENOMEM;
1289 
1290     atomic_set(&sig->count, 1);
1291     memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1292     return 0;
1293 }
1294 
1295 void __cleanup_sighand(struct sighand_struct *sighand)
1296 {
1297     if (atomic_dec_and_test(&sighand->count)) {
1298         signalfd_cleanup(sighand);
1299         /*
1300          * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1301          * without an RCU grace period, see __lock_task_sighand().
1302          */
1303         kmem_cache_free(sighand_cachep, sighand);
1304     }
1305 }
1306 
1307 /*
1308  * Initialize POSIX timer handling for a thread group.
1309  */
1310 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1311 {
1312     unsigned long cpu_limit;
1313 
1314     cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1315     if (cpu_limit != RLIM_INFINITY) {
1316         sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1317         sig->cputimer.running = true;
1318     }
1319 
1320     /* The timer lists. */
1321     INIT_LIST_HEAD(&sig->cpu_timers[0]);
1322     INIT_LIST_HEAD(&sig->cpu_timers[1]);
1323     INIT_LIST_HEAD(&sig->cpu_timers[2]);
1324 }
1325 
1326 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1327 {
1328     struct signal_struct *sig;
1329 
1330     if (clone_flags & CLONE_THREAD)
1331         return 0;
1332 
1333     sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1334     tsk->signal = sig;
1335     if (!sig)
1336         return -ENOMEM;
1337 
1338     sig->nr_threads = 1;
1339     atomic_set(&sig->live, 1);
1340     atomic_set(&sig->sigcnt, 1);
1341 
1342     /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1343     sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1344     tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1345 
1346     init_waitqueue_head(&sig->wait_chldexit);
1347     sig->curr_target = tsk;
1348     init_sigpending(&sig->shared_pending);
1349     INIT_LIST_HEAD(&sig->posix_timers);
1350     seqlock_init(&sig->stats_lock);
1351     prev_cputime_init(&sig->prev_cputime);
1352 
1353 #ifdef CONFIG_POSIX_TIMERS
1354     hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1355     sig->real_timer.function = it_real_fn;
1356 #endif
1357 
1358     task_lock(current->group_leader);
1359     memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1360     task_unlock(current->group_leader);
1361 
1362     posix_cpu_timers_init_group(sig);
1363 
1364     tty_audit_fork(sig);
1365     sched_autogroup_fork(sig);
1366 
1367     sig->oom_score_adj = current->signal->oom_score_adj;
1368     sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1369 
1370     sig->has_child_subreaper = current->signal->has_child_subreaper ||
1371                    current->signal->is_child_subreaper;
1372 
1373     mutex_init(&sig->cred_guard_mutex);
1374 
1375     return 0;
1376 }
1377 
1378 static void copy_seccomp(struct task_struct *p)
1379 {
1380 #ifdef CONFIG_SECCOMP
1381     /*
1382      * Must be called with sighand->lock held, which is common to
1383      * all threads in the group. Holding cred_guard_mutex is not
1384      * needed because this new task is not yet running and cannot
1385      * be racing exec.
1386      */
1387     assert_spin_locked(&current->sighand->siglock);
1388 
1389     /* Ref-count the new filter user, and assign it. */
1390     get_seccomp_filter(current);
1391     p->seccomp = current->seccomp;
1392 
1393     /*
1394      * Explicitly enable no_new_privs here in case it got set
1395      * between the task_struct being duplicated and holding the
1396      * sighand lock. The seccomp state and nnp must be in sync.
1397      */
1398     if (task_no_new_privs(current))
1399         task_set_no_new_privs(p);
1400 
1401     /*
1402      * If the parent gained a seccomp mode after copying thread
1403      * flags and between before we held the sighand lock, we have
1404      * to manually enable the seccomp thread flag here.
1405      */
1406     if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1407         set_tsk_thread_flag(p, TIF_SECCOMP);
1408 #endif
1409 }
1410 
1411 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1412 {
1413     current->clear_child_tid = tidptr;
1414 
1415     return task_pid_vnr(current);
1416 }
1417 
1418 static void rt_mutex_init_task(struct task_struct *p)
1419 {
1420     raw_spin_lock_init(&p->pi_lock);
1421 #ifdef CONFIG_RT_MUTEXES
1422     p->pi_waiters = RB_ROOT;
1423     p->pi_waiters_leftmost = NULL;
1424     p->pi_blocked_on = NULL;
1425 #endif
1426 }
1427 
1428 /*
1429  * Initialize POSIX timer handling for a single task.
1430  */
1431 static void posix_cpu_timers_init(struct task_struct *tsk)
1432 {
1433     tsk->cputime_expires.prof_exp = 0;
1434     tsk->cputime_expires.virt_exp = 0;
1435     tsk->cputime_expires.sched_exp = 0;
1436     INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1437     INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1438     INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1439 }
1440 
1441 static inline void
1442 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1443 {
1444      task->pids[type].pid = pid;
1445 }
1446 
1447 /*
1448  * This creates a new process as a copy of the old one,
1449  * but does not actually start it yet.
1450  *
1451  * It copies the registers, and all the appropriate
1452  * parts of the process environment (as per the clone
1453  * flags). The actual kick-off is left to the caller.
1454  */
1455 static __latent_entropy struct task_struct *copy_process(
1456                     unsigned long clone_flags,
1457                     unsigned long stack_start,
1458                     unsigned long stack_size,
1459                     int __user *child_tidptr,
1460                     struct pid *pid,
1461                     int trace,
1462                     unsigned long tls,
1463                     int node)
1464 {
1465     int retval;
1466     struct task_struct *p;
1467 
1468     if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1469         return ERR_PTR(-EINVAL);
1470 
1471     if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1472         return ERR_PTR(-EINVAL);
1473 
1474     /*
1475      * Thread groups must share signals as well, and detached threads
1476      * can only be started up within the thread group.
1477      */
1478     if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1479         return ERR_PTR(-EINVAL);
1480 
1481     /*
1482      * Shared signal handlers imply shared VM. By way of the above,
1483      * thread groups also imply shared VM. Blocking this case allows
1484      * for various simplifications in other code.
1485      */
1486     if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1487         return ERR_PTR(-EINVAL);
1488 
1489     /*
1490      * Siblings of global init remain as zombies on exit since they are
1491      * not reaped by their parent (swapper). To solve this and to avoid
1492      * multi-rooted process trees, prevent global and container-inits
1493      * from creating siblings.
1494      */
1495     if ((clone_flags & CLONE_PARENT) &&
1496                 current->signal->flags & SIGNAL_UNKILLABLE)
1497         return ERR_PTR(-EINVAL);
1498 
1499     /*
1500      * If the new process will be in a different pid or user namespace
1501      * do not allow it to share a thread group with the forking task.
1502      */
1503     if (clone_flags & CLONE_THREAD) {
1504         if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1505             (task_active_pid_ns(current) !=
1506                 current->nsproxy->pid_ns_for_children))
1507             return ERR_PTR(-EINVAL);
1508     }
1509 
1510     retval = security_task_create(clone_flags);
1511     if (retval)
1512         goto fork_out;
1513 
1514     retval = -ENOMEM;
1515     p = dup_task_struct(current, node);
1516     if (!p)
1517         goto fork_out;
1518 
1519     ftrace_graph_init_task(p);
1520 
1521     rt_mutex_init_task(p);
1522 
1523 #ifdef CONFIG_PROVE_LOCKING
1524     DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1525     DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1526 #endif
1527     retval = -EAGAIN;
1528     if (atomic_read(&p->real_cred->user->processes) >=
1529             task_rlimit(p, RLIMIT_NPROC)) {
1530         if (p->real_cred->user != INIT_USER &&
1531             !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1532             goto bad_fork_free;
1533     }
1534     current->flags &= ~PF_NPROC_EXCEEDED;
1535 
1536     retval = copy_creds(p, clone_flags);
1537     if (retval < 0)
1538         goto bad_fork_free;
1539 
1540     /*
1541      * If multiple threads are within copy_process(), then this check
1542      * triggers too late. This doesn't hurt, the check is only there
1543      * to stop root fork bombs.
1544      */
1545     retval = -EAGAIN;
1546     if (nr_threads >= max_threads)
1547         goto bad_fork_cleanup_count;
1548 
1549     delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1550     p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1551     p->flags |= PF_FORKNOEXEC;
1552     INIT_LIST_HEAD(&p->children);
1553     INIT_LIST_HEAD(&p->sibling);
1554     rcu_copy_process(p);
1555     p->vfork_done = NULL;
1556     spin_lock_init(&p->alloc_lock);
1557 
1558     init_sigpending(&p->pending);
1559 
1560     p->utime = p->stime = p->gtime = 0;
1561 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1562     p->utimescaled = p->stimescaled = 0;
1563 #endif
1564     prev_cputime_init(&p->prev_cputime);
1565 
1566 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1567     seqcount_init(&p->vtime_seqcount);
1568     p->vtime_snap = 0;
1569     p->vtime_snap_whence = VTIME_INACTIVE;
1570 #endif
1571 
1572 #if defined(SPLIT_RSS_COUNTING)
1573     memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1574 #endif
1575 
1576     p->default_timer_slack_ns = current->timer_slack_ns;
1577 
1578     task_io_accounting_init(&p->ioac);
1579     acct_clear_integrals(p);
1580 
1581     posix_cpu_timers_init(p);
1582 
1583     p->start_time = ktime_get_ns();
1584     p->real_start_time = ktime_get_boot_ns();
1585     p->io_context = NULL;
1586     p->audit_context = NULL;
1587     cgroup_fork(p);
1588 #ifdef CONFIG_NUMA
1589     p->mempolicy = mpol_dup(p->mempolicy);
1590     if (IS_ERR(p->mempolicy)) {
1591         retval = PTR_ERR(p->mempolicy);
1592         p->mempolicy = NULL;
1593         goto bad_fork_cleanup_threadgroup_lock;
1594     }
1595 #endif
1596 #ifdef CONFIG_CPUSETS
1597     p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1598     p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1599     seqcount_init(&p->mems_allowed_seq);
1600 #endif
1601 #ifdef CONFIG_TRACE_IRQFLAGS
1602     p->irq_events = 0;
1603     p->hardirqs_enabled = 0;
1604     p->hardirq_enable_ip = 0;
1605     p->hardirq_enable_event = 0;
1606     p->hardirq_disable_ip = _THIS_IP_;
1607     p->hardirq_disable_event = 0;
1608     p->softirqs_enabled = 1;
1609     p->softirq_enable_ip = _THIS_IP_;
1610     p->softirq_enable_event = 0;
1611     p->softirq_disable_ip = 0;
1612     p->softirq_disable_event = 0;
1613     p->hardirq_context = 0;
1614     p->softirq_context = 0;
1615 #endif
1616 
1617     p->pagefault_disabled = 0;
1618 
1619 #ifdef CONFIG_LOCKDEP
1620     p->lockdep_depth = 0; /* no locks held yet */
1621     p->curr_chain_key = 0;
1622     p->lockdep_recursion = 0;
1623 #endif
1624 
1625 #ifdef CONFIG_DEBUG_MUTEXES
1626     p->blocked_on = NULL; /* not blocked yet */
1627 #endif
1628 #ifdef CONFIG_BCACHE
1629     p->sequential_io    = 0;
1630     p->sequential_io_avg    = 0;
1631 #endif
1632 
1633     /* Perform scheduler related setup. Assign this task to a CPU. */
1634     retval = sched_fork(clone_flags, p);
1635     if (retval)
1636         goto bad_fork_cleanup_policy;
1637 
1638     retval = perf_event_init_task(p);
1639     if (retval)
1640         goto bad_fork_cleanup_policy;
1641     retval = audit_alloc(p);
1642     if (retval)
1643         goto bad_fork_cleanup_perf;
1644     /* copy all the process information */
1645     shm_init_task(p);
1646     retval = copy_semundo(clone_flags, p);
1647     if (retval)
1648         goto bad_fork_cleanup_audit;
1649     retval = copy_files(clone_flags, p);
1650     if (retval)
1651         goto bad_fork_cleanup_semundo;
1652     retval = copy_fs(clone_flags, p);
1653     if (retval)
1654         goto bad_fork_cleanup_files;
1655     retval = copy_sighand(clone_flags, p);
1656     if (retval)
1657         goto bad_fork_cleanup_fs;
1658     retval = copy_signal(clone_flags, p);
1659     if (retval)
1660         goto bad_fork_cleanup_sighand;
1661     retval = copy_mm(clone_flags, p);
1662     if (retval)
1663         goto bad_fork_cleanup_signal;
1664     retval = copy_namespaces(clone_flags, p);
1665     if (retval)
1666         goto bad_fork_cleanup_mm;
1667     retval = copy_io(clone_flags, p);
1668     if (retval)
1669         goto bad_fork_cleanup_namespaces;
1670     retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1671     if (retval)
1672         goto bad_fork_cleanup_io;
1673 
1674     if (pid != &init_struct_pid) {
1675         pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1676         if (IS_ERR(pid)) {
1677             retval = PTR_ERR(pid);
1678             goto bad_fork_cleanup_thread;
1679         }
1680     }
1681 
1682     p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1683     /*
1684      * Clear TID on mm_release()?
1685      */
1686     p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1687 #ifdef CONFIG_BLOCK
1688     p->plug = NULL;
1689 #endif
1690 #ifdef CONFIG_FUTEX
1691     p->robust_list = NULL;
1692 #ifdef CONFIG_COMPAT
1693     p->compat_robust_list = NULL;
1694 #endif
1695     INIT_LIST_HEAD(&p->pi_state_list);
1696     p->pi_state_cache = NULL;
1697 #endif
1698     /*
1699      * sigaltstack should be cleared when sharing the same VM
1700      */
1701     if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1702         sas_ss_reset(p);
1703 
1704     /*
1705      * Syscall tracing and stepping should be turned off in the
1706      * child regardless of CLONE_PTRACE.
1707      */
1708     user_disable_single_step(p);
1709     clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1710 #ifdef TIF_SYSCALL_EMU
1711     clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1712 #endif
1713     clear_all_latency_tracing(p);
1714 
1715     /* ok, now we should be set up.. */
1716     p->pid = pid_nr(pid);
1717     if (clone_flags & CLONE_THREAD) {
1718         p->exit_signal = -1;
1719         p->group_leader = current->group_leader;
1720         p->tgid = current->tgid;
1721     } else {
1722         if (clone_flags & CLONE_PARENT)
1723             p->exit_signal = current->group_leader->exit_signal;
1724         else
1725             p->exit_signal = (clone_flags & CSIGNAL);
1726         p->group_leader = p;
1727         p->tgid = p->pid;
1728     }
1729 
1730     p->nr_dirtied = 0;
1731     p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1732     p->dirty_paused_when = 0;
1733 
1734     p->pdeath_signal = 0;
1735     INIT_LIST_HEAD(&p->thread_group);
1736     p->task_works = NULL;
1737 
1738     threadgroup_change_begin(current);
1739     /*
1740      * Ensure that the cgroup subsystem policies allow the new process to be
1741      * forked. It should be noted the the new process's css_set can be changed
1742      * between here and cgroup_post_fork() if an organisation operation is in
1743      * progress.
1744      */
1745     retval = cgroup_can_fork(p);
1746     if (retval)
1747         goto bad_fork_free_pid;
1748 
1749     /*
1750      * Make it visible to the rest of the system, but dont wake it up yet.
1751      * Need tasklist lock for parent etc handling!
1752      */
1753     write_lock_irq(&tasklist_lock);
1754 
1755     /* CLONE_PARENT re-uses the old parent */
1756     if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1757         p->real_parent = current->real_parent;
1758         p->parent_exec_id = current->parent_exec_id;
1759     } else {
1760         p->real_parent = current;
1761         p->parent_exec_id = current->self_exec_id;
1762     }
1763 
1764     spin_lock(&current->sighand->siglock);
1765 
1766     /*
1767      * Copy seccomp details explicitly here, in case they were changed
1768      * before holding sighand lock.
1769      */
1770     copy_seccomp(p);
1771 
1772     /*
1773      * Process group and session signals need to be delivered to just the
1774      * parent before the fork or both the parent and the child after the
1775      * fork. Restart if a signal comes in before we add the new process to
1776      * it's process group.
1777      * A fatal signal pending means that current will exit, so the new
1778      * thread can't slip out of an OOM kill (or normal SIGKILL).
1779     */
1780     recalc_sigpending();
1781     if (signal_pending(current)) {
1782         spin_unlock(&current->sighand->siglock);
1783         write_unlock_irq(&tasklist_lock);
1784         retval = -ERESTARTNOINTR;
1785         goto bad_fork_cancel_cgroup;
1786     }
1787 
1788     if (likely(p->pid)) {
1789         ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1790 
1791         init_task_pid(p, PIDTYPE_PID, pid);
1792         if (thread_group_leader(p)) {
1793             init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1794             init_task_pid(p, PIDTYPE_SID, task_session(current));
1795 
1796             if (is_child_reaper(pid)) {
1797                 ns_of_pid(pid)->child_reaper = p;
1798                 p->signal->flags |= SIGNAL_UNKILLABLE;
1799             }
1800 
1801             p->signal->leader_pid = pid;
1802             p->signal->tty = tty_kref_get(current->signal->tty);
1803             list_add_tail(&p->sibling, &p->real_parent->children);
1804             list_add_tail_rcu(&p->tasks, &init_task.tasks);
1805             attach_pid(p, PIDTYPE_PGID);
1806             attach_pid(p, PIDTYPE_SID);
1807             __this_cpu_inc(process_counts);
1808         } else {
1809             current->signal->nr_threads++;
1810             atomic_inc(&current->signal->live);
1811             atomic_inc(&current->signal->sigcnt);
1812             list_add_tail_rcu(&p->thread_group,
1813                       &p->group_leader->thread_group);
1814             list_add_tail_rcu(&p->thread_node,
1815                       &p->signal->thread_head);
1816         }
1817         attach_pid(p, PIDTYPE_PID);
1818         nr_threads++;
1819     }
1820 
1821     total_forks++;
1822     spin_unlock(&current->sighand->siglock);
1823     syscall_tracepoint_update(p);
1824     write_unlock_irq(&tasklist_lock);
1825 
1826     proc_fork_connector(p);
1827     cgroup_post_fork(p);
1828     threadgroup_change_end(current);
1829     perf_event_fork(p);
1830 
1831     trace_task_newtask(p, clone_flags);
1832     uprobe_copy_process(p, clone_flags);
1833 
1834     return p;
1835 
1836 bad_fork_cancel_cgroup:
1837     cgroup_cancel_fork(p);
1838 bad_fork_free_pid:
1839     threadgroup_change_end(current);
1840     if (pid != &init_struct_pid)
1841         free_pid(pid);
1842 bad_fork_cleanup_thread:
1843     exit_thread(p);
1844 bad_fork_cleanup_io:
1845     if (p->io_context)
1846         exit_io_context(p);
1847 bad_fork_cleanup_namespaces:
1848     exit_task_namespaces(p);
1849 bad_fork_cleanup_mm:
1850     if (p->mm)
1851         mmput(p->mm);
1852 bad_fork_cleanup_signal:
1853     if (!(clone_flags & CLONE_THREAD))
1854         free_signal_struct(p->signal);
1855 bad_fork_cleanup_sighand:
1856     __cleanup_sighand(p->sighand);
1857 bad_fork_cleanup_fs:
1858     exit_fs(p); /* blocking */
1859 bad_fork_cleanup_files:
1860     exit_files(p); /* blocking */
1861 bad_fork_cleanup_semundo:
1862     exit_sem(p);
1863 bad_fork_cleanup_audit:
1864     audit_free(p);
1865 bad_fork_cleanup_perf:
1866     perf_event_free_task(p);
1867 bad_fork_cleanup_policy:
1868 #ifdef CONFIG_NUMA
1869     mpol_put(p->mempolicy);
1870 bad_fork_cleanup_threadgroup_lock:
1871 #endif
1872     delayacct_tsk_free(p);
1873 bad_fork_cleanup_count:
1874     atomic_dec(&p->cred->user->processes);
1875     exit_creds(p);
1876 bad_fork_free:
1877     p->state = TASK_DEAD;
1878     put_task_stack(p);
1879     free_task(p);
1880 fork_out:
1881     return ERR_PTR(retval);
1882 }
1883 
1884 static inline void init_idle_pids(struct pid_link *links)
1885 {
1886     enum pid_type type;
1887 
1888     for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1889         INIT_HLIST_NODE(&links[type].node); /* not really needed */
1890         links[type].pid = &init_struct_pid;
1891     }
1892 }
1893 
1894 struct task_struct *fork_idle(int cpu)
1895 {
1896     struct task_struct *task;
1897     task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
1898                 cpu_to_node(cpu));
1899     if (!IS_ERR(task)) {
1900         init_idle_pids(task->pids);
1901         init_idle(task, cpu);
1902     }
1903 
1904     return task;
1905 }
1906 
1907 /*
1908  *  Ok, this is the main fork-routine.
1909  *
1910  * It copies the process, and if successful kick-starts
1911  * it and waits for it to finish using the VM if required.
1912  */
1913 long _do_fork(unsigned long clone_flags,
1914           unsigned long stack_start,
1915           unsigned long stack_size,
1916           int __user *parent_tidptr,
1917           int __user *child_tidptr,
1918           unsigned long tls)
1919 {
1920     struct task_struct *p;
1921     int trace = 0;
1922     long nr;
1923 
1924     /*
1925      * Determine whether and which event to report to ptracer.  When
1926      * called from kernel_thread or CLONE_UNTRACED is explicitly
1927      * requested, no event is reported; otherwise, report if the event
1928      * for the type of forking is enabled.
1929      */
1930     if (!(clone_flags & CLONE_UNTRACED)) {
1931         if (clone_flags & CLONE_VFORK)
1932             trace = PTRACE_EVENT_VFORK;
1933         else if ((clone_flags & CSIGNAL) != SIGCHLD)
1934             trace = PTRACE_EVENT_CLONE;
1935         else
1936             trace = PTRACE_EVENT_FORK;
1937 
1938         if (likely(!ptrace_event_enabled(current, trace)))
1939             trace = 0;
1940     }
1941 
1942     p = copy_process(clone_flags, stack_start, stack_size,
1943              child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
1944     add_latent_entropy();
1945     /*
1946      * Do this prior waking up the new thread - the thread pointer
1947      * might get invalid after that point, if the thread exits quickly.
1948      */
1949     if (!IS_ERR(p)) {
1950         struct completion vfork;
1951         struct pid *pid;
1952 
1953         trace_sched_process_fork(current, p);
1954 
1955         pid = get_task_pid(p, PIDTYPE_PID);
1956         nr = pid_vnr(pid);
1957 
1958         if (clone_flags & CLONE_PARENT_SETTID)
1959             put_user(nr, parent_tidptr);
1960 
1961         if (clone_flags & CLONE_VFORK) {
1962             p->vfork_done = &vfork;
1963             init_completion(&vfork);
1964             get_task_struct(p);
1965         }
1966 
1967         wake_up_new_task(p);
1968 
1969         /* forking complete and child started to run, tell ptracer */
1970         if (unlikely(trace))
1971             ptrace_event_pid(trace, pid);
1972 
1973         if (clone_flags & CLONE_VFORK) {
1974             if (!wait_for_vfork_done(p, &vfork))
1975                 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1976         }
1977 
1978         put_pid(pid);
1979     } else {
1980         nr = PTR_ERR(p);
1981     }
1982     return nr;
1983 }
1984 
1985 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
1986 /* For compatibility with architectures that call do_fork directly rather than
1987  * using the syscall entry points below. */
1988 long do_fork(unsigned long clone_flags,
1989           unsigned long stack_start,
1990           unsigned long stack_size,
1991           int __user *parent_tidptr,
1992           int __user *child_tidptr)
1993 {
1994     return _do_fork(clone_flags, stack_start, stack_size,
1995             parent_tidptr, child_tidptr, 0);
1996 }
1997 #endif
1998 
1999 /*
2000  * Create a kernel thread.
2001  */
2002 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2003 {
2004     return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2005         (unsigned long)arg, NULL, NULL, 0);
2006 }
2007 
2008 #ifdef __ARCH_WANT_SYS_FORK
2009 SYSCALL_DEFINE0(fork)
2010 {
2011 #ifdef CONFIG_MMU
2012     return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2013 #else
2014     /* can not support in nommu mode */
2015     return -EINVAL;
2016 #endif
2017 }
2018 #endif
2019 
2020 #ifdef __ARCH_WANT_SYS_VFORK
2021 SYSCALL_DEFINE0(vfork)
2022 {
2023     return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2024             0, NULL, NULL, 0);
2025 }
2026 #endif
2027 
2028 #ifdef __ARCH_WANT_SYS_CLONE
2029 #ifdef CONFIG_CLONE_BACKWARDS
2030 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2031          int __user *, parent_tidptr,
2032          unsigned long, tls,
2033          int __user *, child_tidptr)
2034 #elif defined(CONFIG_CLONE_BACKWARDS2)
2035 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2036          int __user *, parent_tidptr,
2037          int __user *, child_tidptr,
2038          unsigned long, tls)
2039 #elif defined(CONFIG_CLONE_BACKWARDS3)
2040 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2041         int, stack_size,
2042         int __user *, parent_tidptr,
2043         int __user *, child_tidptr,
2044         unsigned long, tls)
2045 #else
2046 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2047          int __user *, parent_tidptr,
2048          int __user *, child_tidptr,
2049          unsigned long, tls)
2050 #endif
2051 {
2052     return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2053 }
2054 #endif
2055 
2056 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2057 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2058 #endif
2059 
2060 static void sighand_ctor(void *data)
2061 {
2062     struct sighand_struct *sighand = data;
2063 
2064     spin_lock_init(&sighand->siglock);
2065     init_waitqueue_head(&sighand->signalfd_wqh);
2066 }
2067 
2068 void __init proc_caches_init(void)
2069 {
2070     sighand_cachep = kmem_cache_create("sighand_cache",
2071             sizeof(struct sighand_struct), 0,
2072             SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
2073             SLAB_NOTRACK|SLAB_ACCOUNT, sighand_ctor);
2074     signal_cachep = kmem_cache_create("signal_cache",
2075             sizeof(struct signal_struct), 0,
2076             SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2077             NULL);
2078     files_cachep = kmem_cache_create("files_cache",
2079             sizeof(struct files_struct), 0,
2080             SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2081             NULL);
2082     fs_cachep = kmem_cache_create("fs_cache",
2083             sizeof(struct fs_struct), 0,
2084             SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2085             NULL);
2086     /*
2087      * FIXME! The "sizeof(struct mm_struct)" currently includes the
2088      * whole struct cpumask for the OFFSTACK case. We could change
2089      * this to *only* allocate as much of it as required by the
2090      * maximum number of CPU's we can ever have.  The cpumask_allocation
2091      * is at the end of the structure, exactly for that reason.
2092      */
2093     mm_cachep = kmem_cache_create("mm_struct",
2094             sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
2095             SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2096             NULL);
2097     vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2098     mmap_init();
2099     nsproxy_cache_init();
2100 }
2101 
2102 /*
2103  * Check constraints on flags passed to the unshare system call.
2104  */
2105 static int check_unshare_flags(unsigned long unshare_flags)
2106 {
2107     if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2108                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2109                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2110                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2111         return -EINVAL;
2112     /*
2113      * Not implemented, but pretend it works if there is nothing
2114      * to unshare.  Note that unsharing the address space or the
2115      * signal handlers also need to unshare the signal queues (aka
2116      * CLONE_THREAD).
2117      */
2118     if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2119         if (!thread_group_empty(current))
2120             return -EINVAL;
2121     }
2122     if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2123         if (atomic_read(&current->sighand->count) > 1)
2124             return -EINVAL;
2125     }
2126     if (unshare_flags & CLONE_VM) {
2127         if (!current_is_single_threaded())
2128             return -EINVAL;
2129     }
2130 
2131     return 0;
2132 }
2133 
2134 /*
2135  * Unshare the filesystem structure if it is being shared
2136  */
2137 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2138 {
2139     struct fs_struct *fs = current->fs;
2140 
2141     if (!(unshare_flags & CLONE_FS) || !fs)
2142         return 0;
2143 
2144     /* don't need lock here; in the worst case we'll do useless copy */
2145     if (fs->users == 1)
2146         return 0;
2147 
2148     *new_fsp = copy_fs_struct(fs);
2149     if (!*new_fsp)
2150         return -ENOMEM;
2151 
2152     return 0;
2153 }
2154 
2155 /*
2156  * Unshare file descriptor table if it is being shared
2157  */
2158 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2159 {
2160     struct files_struct *fd = current->files;
2161     int error = 0;
2162 
2163     if ((unshare_flags & CLONE_FILES) &&
2164         (fd && atomic_read(&fd->count) > 1)) {
2165         *new_fdp = dup_fd(fd, &error);
2166         if (!*new_fdp)
2167             return error;
2168     }
2169 
2170     return 0;
2171 }
2172 
2173 /*
2174  * unshare allows a process to 'unshare' part of the process
2175  * context which was originally shared using clone.  copy_*
2176  * functions used by do_fork() cannot be used here directly
2177  * because they modify an inactive task_struct that is being
2178  * constructed. Here we are modifying the current, active,
2179  * task_struct.
2180  */
2181 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2182 {
2183     struct fs_struct *fs, *new_fs = NULL;
2184     struct files_struct *fd, *new_fd = NULL;
2185     struct cred *new_cred = NULL;
2186     struct nsproxy *new_nsproxy = NULL;
2187     int do_sysvsem = 0;
2188     int err;
2189 
2190     /*
2191      * If unsharing a user namespace must also unshare the thread group
2192      * and unshare the filesystem root and working directories.
2193      */
2194     if (unshare_flags & CLONE_NEWUSER)
2195         unshare_flags |= CLONE_THREAD | CLONE_FS;
2196     /*
2197      * If unsharing vm, must also unshare signal handlers.
2198      */
2199     if (unshare_flags & CLONE_VM)
2200         unshare_flags |= CLONE_SIGHAND;
2201     /*
2202      * If unsharing a signal handlers, must also unshare the signal queues.
2203      */
2204     if (unshare_flags & CLONE_SIGHAND)
2205         unshare_flags |= CLONE_THREAD;
2206     /*
2207      * If unsharing namespace, must also unshare filesystem information.
2208      */
2209     if (unshare_flags & CLONE_NEWNS)
2210         unshare_flags |= CLONE_FS;
2211 
2212     err = check_unshare_flags(unshare_flags);
2213     if (err)
2214         goto bad_unshare_out;
2215     /*
2216      * CLONE_NEWIPC must also detach from the undolist: after switching
2217      * to a new ipc namespace, the semaphore arrays from the old
2218      * namespace are unreachable.
2219      */
2220     if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2221         do_sysvsem = 1;
2222     err = unshare_fs(unshare_flags, &new_fs);
2223     if (err)
2224         goto bad_unshare_out;
2225     err = unshare_fd(unshare_flags, &new_fd);
2226     if (err)
2227         goto bad_unshare_cleanup_fs;
2228     err = unshare_userns(unshare_flags, &new_cred);
2229     if (err)
2230         goto bad_unshare_cleanup_fd;
2231     err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2232                      new_cred, new_fs);
2233     if (err)
2234         goto bad_unshare_cleanup_cred;
2235 
2236     if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2237         if (do_sysvsem) {
2238             /*
2239              * CLONE_SYSVSEM is equivalent to sys_exit().
2240              */
2241             exit_sem(current);
2242         }
2243         if (unshare_flags & CLONE_NEWIPC) {
2244             /* Orphan segments in old ns (see sem above). */
2245             exit_shm(current);
2246             shm_init_task(current);
2247         }
2248 
2249         if (new_nsproxy)
2250             switch_task_namespaces(current, new_nsproxy);
2251 
2252         task_lock(current);
2253 
2254         if (new_fs) {
2255             fs = current->fs;
2256             spin_lock(&fs->lock);
2257             current->fs = new_fs;
2258             if (--fs->users)
2259                 new_fs = NULL;
2260             else
2261                 new_fs = fs;
2262             spin_unlock(&fs->lock);
2263         }
2264 
2265         if (new_fd) {
2266             fd = current->files;
2267             current->files = new_fd;
2268             new_fd = fd;
2269         }
2270 
2271         task_unlock(current);
2272 
2273         if (new_cred) {
2274             /* Install the new user namespace */
2275             commit_creds(new_cred);
2276             new_cred = NULL;
2277         }
2278     }
2279 
2280 bad_unshare_cleanup_cred:
2281     if (new_cred)
2282         put_cred(new_cred);
2283 bad_unshare_cleanup_fd:
2284     if (new_fd)
2285         put_files_struct(new_fd);
2286 
2287 bad_unshare_cleanup_fs:
2288     if (new_fs)
2289         free_fs_struct(new_fs);
2290 
2291 bad_unshare_out:
2292     return err;
2293 }
2294 
2295 /*
2296  *  Helper to unshare the files of the current task.
2297  *  We don't want to expose copy_files internals to
2298  *  the exec layer of the kernel.
2299  */
2300 
2301 int unshare_files(struct files_struct **displaced)
2302 {
2303     struct task_struct *task = current;
2304     struct files_struct *copy = NULL;
2305     int error;
2306 
2307     error = unshare_fd(CLONE_FILES, &copy);
2308     if (error || !copy) {
2309         *displaced = NULL;
2310         return error;
2311     }
2312     *displaced = task->files;
2313     task_lock(task);
2314     task->files = copy;
2315     task_unlock(task);
2316     return 0;
2317 }
2318 
2319 int sysctl_max_threads(struct ctl_table *table, int write,
2320                void __user *buffer, size_t *lenp, loff_t *ppos)
2321 {
2322     struct ctl_table t;
2323     int ret;
2324     int threads = max_threads;
2325     int min = MIN_THREADS;
2326     int max = MAX_THREADS;
2327 
2328     t = *table;
2329     t.data = &threads;
2330     t.extra1 = &min;
2331     t.extra2 = &max;
2332 
2333     ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2334     if (ret || !write)
2335         return ret;
2336 
2337     set_max_threads(threads);
2338 
2339     return 0;
2340 }