~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/kernel/fork.c

Version: ~ [ linux-5.15-rc1 ] ~ [ linux-5.14.5 ] ~ [ linux-5.13.18 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.66 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.147 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.206 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.246 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.282 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.283 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  *  linux/kernel/fork.c
  3  *
  4  *  Copyright (C) 1991, 1992  Linus Torvalds
  5  */
  6 
  7 /*
  8  *  'fork.c' contains the help-routines for the 'fork' system call
  9  * (see also entry.S and others).
 10  * Fork is rather simple, once you get the hang of it, but the memory
 11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
 12  */
 13 
 14 #include <linux/slab.h>
 15 #include <linux/sched/autogroup.h>
 16 #include <linux/sched/mm.h>
 17 #include <linux/sched/coredump.h>
 18 #include <linux/sched/user.h>
 19 #include <linux/sched/numa_balancing.h>
 20 #include <linux/sched/stat.h>
 21 #include <linux/sched/task.h>
 22 #include <linux/sched/task_stack.h>
 23 #include <linux/sched/cputime.h>
 24 #include <linux/rtmutex.h>
 25 #include <linux/init.h>
 26 #include <linux/unistd.h>
 27 #include <linux/module.h>
 28 #include <linux/vmalloc.h>
 29 #include <linux/completion.h>
 30 #include <linux/personality.h>
 31 #include <linux/mempolicy.h>
 32 #include <linux/sem.h>
 33 #include <linux/file.h>
 34 #include <linux/fdtable.h>
 35 #include <linux/iocontext.h>
 36 #include <linux/key.h>
 37 #include <linux/binfmts.h>
 38 #include <linux/mman.h>
 39 #include <linux/mmu_notifier.h>
 40 #include <linux/hmm.h>
 41 #include <linux/fs.h>
 42 #include <linux/mm.h>
 43 #include <linux/vmacache.h>
 44 #include <linux/nsproxy.h>
 45 #include <linux/capability.h>
 46 #include <linux/cpu.h>
 47 #include <linux/cgroup.h>
 48 #include <linux/security.h>
 49 #include <linux/hugetlb.h>
 50 #include <linux/seccomp.h>
 51 #include <linux/swap.h>
 52 #include <linux/syscalls.h>
 53 #include <linux/jiffies.h>
 54 #include <linux/futex.h>
 55 #include <linux/compat.h>
 56 #include <linux/kthread.h>
 57 #include <linux/task_io_accounting_ops.h>
 58 #include <linux/rcupdate.h>
 59 #include <linux/ptrace.h>
 60 #include <linux/mount.h>
 61 #include <linux/audit.h>
 62 #include <linux/memcontrol.h>
 63 #include <linux/ftrace.h>
 64 #include <linux/proc_fs.h>
 65 #include <linux/profile.h>
 66 #include <linux/rmap.h>
 67 #include <linux/ksm.h>
 68 #include <linux/acct.h>
 69 #include <linux/userfaultfd_k.h>
 70 #include <linux/tsacct_kern.h>
 71 #include <linux/cn_proc.h>
 72 #include <linux/freezer.h>
 73 #include <linux/delayacct.h>
 74 #include <linux/taskstats_kern.h>
 75 #include <linux/random.h>
 76 #include <linux/tty.h>
 77 #include <linux/blkdev.h>
 78 #include <linux/fs_struct.h>
 79 #include <linux/magic.h>
 80 #include <linux/sched/mm.h>
 81 #include <linux/perf_event.h>
 82 #include <linux/posix-timers.h>
 83 #include <linux/user-return-notifier.h>
 84 #include <linux/oom.h>
 85 #include <linux/khugepaged.h>
 86 #include <linux/signalfd.h>
 87 #include <linux/uprobes.h>
 88 #include <linux/aio.h>
 89 #include <linux/compiler.h>
 90 #include <linux/sysctl.h>
 91 #include <linux/kcov.h>
 92 #include <linux/livepatch.h>
 93 #include <linux/thread_info.h>
 94 
 95 #include <asm/pgtable.h>
 96 #include <asm/pgalloc.h>
 97 #include <linux/uaccess.h>
 98 #include <asm/mmu_context.h>
 99 #include <asm/cacheflush.h>
100 #include <asm/tlbflush.h>
101 
102 #include <trace/events/sched.h>
103 
104 #define CREATE_TRACE_POINTS
105 #include <trace/events/task.h>
106 
107 /*
108  * Minimum number of threads to boot the kernel
109  */
110 #define MIN_THREADS 20
111 
112 /*
113  * Maximum number of threads
114  */
115 #define MAX_THREADS FUTEX_TID_MASK
116 
117 /*
118  * Protected counters by write_lock_irq(&tasklist_lock)
119  */
120 unsigned long total_forks;      /* Handle normal Linux uptimes. */
121 int nr_threads;                 /* The idle threads do not count.. */
122 
123 int max_threads;                /* tunable limit on nr_threads */
124 
125 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
126 
127 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
128 
129 #ifdef CONFIG_PROVE_RCU
130 int lockdep_tasklist_lock_is_held(void)
131 {
132         return lockdep_is_held(&tasklist_lock);
133 }
134 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
135 #endif /* #ifdef CONFIG_PROVE_RCU */
136 
137 int nr_processes(void)
138 {
139         int cpu;
140         int total = 0;
141 
142         for_each_possible_cpu(cpu)
143                 total += per_cpu(process_counts, cpu);
144 
145         return total;
146 }
147 
148 void __weak arch_release_task_struct(struct task_struct *tsk)
149 {
150 }
151 
152 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
153 static struct kmem_cache *task_struct_cachep;
154 
155 static inline struct task_struct *alloc_task_struct_node(int node)
156 {
157         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
158 }
159 
160 static inline void free_task_struct(struct task_struct *tsk)
161 {
162         kmem_cache_free(task_struct_cachep, tsk);
163 }
164 #endif
165 
166 void __weak arch_release_thread_stack(unsigned long *stack)
167 {
168 }
169 
170 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
171 
172 /*
173  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
174  * kmemcache based allocator.
175  */
176 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
177 
178 #ifdef CONFIG_VMAP_STACK
179 /*
180  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
181  * flush.  Try to minimize the number of calls by caching stacks.
182  */
183 #define NR_CACHED_STACKS 2
184 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
185 
186 static int free_vm_stack_cache(unsigned int cpu)
187 {
188         struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
189         int i;
190 
191         for (i = 0; i < NR_CACHED_STACKS; i++) {
192                 struct vm_struct *vm_stack = cached_vm_stacks[i];
193 
194                 if (!vm_stack)
195                         continue;
196 
197                 vfree(vm_stack->addr);
198                 cached_vm_stacks[i] = NULL;
199         }
200 
201         return 0;
202 }
203 #endif
204 
205 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
206 {
207 #ifdef CONFIG_VMAP_STACK
208         void *stack;
209         int i;
210 
211         for (i = 0; i < NR_CACHED_STACKS; i++) {
212                 struct vm_struct *s;
213 
214                 s = this_cpu_xchg(cached_stacks[i], NULL);
215 
216                 if (!s)
217                         continue;
218 
219                 /* Clear stale pointers from reused stack. */
220                 memset(s->addr, 0, THREAD_SIZE);
221 
222                 tsk->stack_vm_area = s;
223                 return s->addr;
224         }
225 
226         stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
227                                      VMALLOC_START, VMALLOC_END,
228                                      THREADINFO_GFP,
229                                      PAGE_KERNEL,
230                                      0, node, __builtin_return_address(0));
231 
232         /*
233          * We can't call find_vm_area() in interrupt context, and
234          * free_thread_stack() can be called in interrupt context,
235          * so cache the vm_struct.
236          */
237         if (stack)
238                 tsk->stack_vm_area = find_vm_area(stack);
239         return stack;
240 #else
241         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
242                                              THREAD_SIZE_ORDER);
243 
244         return page ? page_address(page) : NULL;
245 #endif
246 }
247 
248 static inline void free_thread_stack(struct task_struct *tsk)
249 {
250 #ifdef CONFIG_VMAP_STACK
251         if (task_stack_vm_area(tsk)) {
252                 int i;
253 
254                 for (i = 0; i < NR_CACHED_STACKS; i++) {
255                         if (this_cpu_cmpxchg(cached_stacks[i],
256                                         NULL, tsk->stack_vm_area) != NULL)
257                                 continue;
258 
259                         return;
260                 }
261 
262                 vfree_atomic(tsk->stack);
263                 return;
264         }
265 #endif
266 
267         __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
268 }
269 # else
270 static struct kmem_cache *thread_stack_cache;
271 
272 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
273                                                   int node)
274 {
275         return kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
276 }
277 
278 static void free_thread_stack(struct task_struct *tsk)
279 {
280         kmem_cache_free(thread_stack_cache, tsk->stack);
281 }
282 
283 void thread_stack_cache_init(void)
284 {
285         thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
286                                         THREAD_SIZE, THREAD_SIZE, 0, 0,
287                                         THREAD_SIZE, NULL);
288         BUG_ON(thread_stack_cache == NULL);
289 }
290 # endif
291 #endif
292 
293 /* SLAB cache for signal_struct structures (tsk->signal) */
294 static struct kmem_cache *signal_cachep;
295 
296 /* SLAB cache for sighand_struct structures (tsk->sighand) */
297 struct kmem_cache *sighand_cachep;
298 
299 /* SLAB cache for files_struct structures (tsk->files) */
300 struct kmem_cache *files_cachep;
301 
302 /* SLAB cache for fs_struct structures (tsk->fs) */
303 struct kmem_cache *fs_cachep;
304 
305 /* SLAB cache for vm_area_struct structures */
306 static struct kmem_cache *vm_area_cachep;
307 
308 /* SLAB cache for mm_struct structures (tsk->mm) */
309 static struct kmem_cache *mm_cachep;
310 
311 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
312 {
313         struct vm_area_struct *vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
314 
315         if (vma)
316                 vma_init(vma, mm);
317         return vma;
318 }
319 
320 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
321 {
322         struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
323 
324         if (new) {
325                 *new = *orig;
326                 INIT_LIST_HEAD(&new->anon_vma_chain);
327         }
328         return new;
329 }
330 
331 void vm_area_free(struct vm_area_struct *vma)
332 {
333         kmem_cache_free(vm_area_cachep, vma);
334 }
335 
336 static void account_kernel_stack(struct task_struct *tsk, int account)
337 {
338         void *stack = task_stack_page(tsk);
339         struct vm_struct *vm = task_stack_vm_area(tsk);
340 
341         BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
342 
343         if (vm) {
344                 int i;
345 
346                 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
347 
348                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
349                         mod_zone_page_state(page_zone(vm->pages[i]),
350                                             NR_KERNEL_STACK_KB,
351                                             PAGE_SIZE / 1024 * account);
352                 }
353 
354                 /* All stack pages belong to the same memcg. */
355                 mod_memcg_page_state(vm->pages[0], MEMCG_KERNEL_STACK_KB,
356                                      account * (THREAD_SIZE / 1024));
357         } else {
358                 /*
359                  * All stack pages are in the same zone and belong to the
360                  * same memcg.
361                  */
362                 struct page *first_page = virt_to_page(stack);
363 
364                 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
365                                     THREAD_SIZE / 1024 * account);
366 
367                 mod_memcg_page_state(first_page, MEMCG_KERNEL_STACK_KB,
368                                      account * (THREAD_SIZE / 1024));
369         }
370 }
371 
372 static void release_task_stack(struct task_struct *tsk)
373 {
374         if (WARN_ON(tsk->state != TASK_DEAD))
375                 return;  /* Better to leak the stack than to free prematurely */
376 
377         account_kernel_stack(tsk, -1);
378         arch_release_thread_stack(tsk->stack);
379         free_thread_stack(tsk);
380         tsk->stack = NULL;
381 #ifdef CONFIG_VMAP_STACK
382         tsk->stack_vm_area = NULL;
383 #endif
384 }
385 
386 #ifdef CONFIG_THREAD_INFO_IN_TASK
387 void put_task_stack(struct task_struct *tsk)
388 {
389         if (atomic_dec_and_test(&tsk->stack_refcount))
390                 release_task_stack(tsk);
391 }
392 #endif
393 
394 void free_task(struct task_struct *tsk)
395 {
396 #ifndef CONFIG_THREAD_INFO_IN_TASK
397         /*
398          * The task is finally done with both the stack and thread_info,
399          * so free both.
400          */
401         release_task_stack(tsk);
402 #else
403         /*
404          * If the task had a separate stack allocation, it should be gone
405          * by now.
406          */
407         WARN_ON_ONCE(atomic_read(&tsk->stack_refcount) != 0);
408 #endif
409         rt_mutex_debug_task_free(tsk);
410         ftrace_graph_exit_task(tsk);
411         put_seccomp_filter(tsk);
412         arch_release_task_struct(tsk);
413         if (tsk->flags & PF_KTHREAD)
414                 free_kthread_struct(tsk);
415         free_task_struct(tsk);
416 }
417 EXPORT_SYMBOL(free_task);
418 
419 #ifdef CONFIG_MMU
420 static __latent_entropy int dup_mmap(struct mm_struct *mm,
421                                         struct mm_struct *oldmm)
422 {
423         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
424         struct rb_node **rb_link, *rb_parent;
425         int retval;
426         unsigned long charge;
427         LIST_HEAD(uf);
428 
429         uprobe_start_dup_mmap();
430         if (down_write_killable(&oldmm->mmap_sem)) {
431                 retval = -EINTR;
432                 goto fail_uprobe_end;
433         }
434         flush_cache_dup_mm(oldmm);
435         uprobe_dup_mmap(oldmm, mm);
436         /*
437          * Not linked in yet - no deadlock potential:
438          */
439         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
440 
441         /* No ordering required: file already has been exposed. */
442         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
443 
444         mm->total_vm = oldmm->total_vm;
445         mm->data_vm = oldmm->data_vm;
446         mm->exec_vm = oldmm->exec_vm;
447         mm->stack_vm = oldmm->stack_vm;
448 
449         rb_link = &mm->mm_rb.rb_node;
450         rb_parent = NULL;
451         pprev = &mm->mmap;
452         retval = ksm_fork(mm, oldmm);
453         if (retval)
454                 goto out;
455         retval = khugepaged_fork(mm, oldmm);
456         if (retval)
457                 goto out;
458 
459         prev = NULL;
460         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
461                 struct file *file;
462 
463                 if (mpnt->vm_flags & VM_DONTCOPY) {
464                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
465                         continue;
466                 }
467                 charge = 0;
468                 /*
469                  * Don't duplicate many vmas if we've been oom-killed (for
470                  * example)
471                  */
472                 if (fatal_signal_pending(current)) {
473                         retval = -EINTR;
474                         goto out;
475                 }
476                 if (mpnt->vm_flags & VM_ACCOUNT) {
477                         unsigned long len = vma_pages(mpnt);
478 
479                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
480                                 goto fail_nomem;
481                         charge = len;
482                 }
483                 tmp = vm_area_dup(mpnt);
484                 if (!tmp)
485                         goto fail_nomem;
486                 retval = vma_dup_policy(mpnt, tmp);
487                 if (retval)
488                         goto fail_nomem_policy;
489                 tmp->vm_mm = mm;
490                 retval = dup_userfaultfd(tmp, &uf);
491                 if (retval)
492                         goto fail_nomem_anon_vma_fork;
493                 if (tmp->vm_flags & VM_WIPEONFORK) {
494                         /* VM_WIPEONFORK gets a clean slate in the child. */
495                         tmp->anon_vma = NULL;
496                         if (anon_vma_prepare(tmp))
497                                 goto fail_nomem_anon_vma_fork;
498                 } else if (anon_vma_fork(tmp, mpnt))
499                         goto fail_nomem_anon_vma_fork;
500                 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
501                 tmp->vm_next = tmp->vm_prev = NULL;
502                 file = tmp->vm_file;
503                 if (file) {
504                         struct inode *inode = file_inode(file);
505                         struct address_space *mapping = file->f_mapping;
506 
507                         get_file(file);
508                         if (tmp->vm_flags & VM_DENYWRITE)
509                                 atomic_dec(&inode->i_writecount);
510                         i_mmap_lock_write(mapping);
511                         if (tmp->vm_flags & VM_SHARED)
512                                 atomic_inc(&mapping->i_mmap_writable);
513                         flush_dcache_mmap_lock(mapping);
514                         /* insert tmp into the share list, just after mpnt */
515                         vma_interval_tree_insert_after(tmp, mpnt,
516                                         &mapping->i_mmap);
517                         flush_dcache_mmap_unlock(mapping);
518                         i_mmap_unlock_write(mapping);
519                 }
520 
521                 /*
522                  * Clear hugetlb-related page reserves for children. This only
523                  * affects MAP_PRIVATE mappings. Faults generated by the child
524                  * are not guaranteed to succeed, even if read-only
525                  */
526                 if (is_vm_hugetlb_page(tmp))
527                         reset_vma_resv_huge_pages(tmp);
528 
529                 /*
530                  * Link in the new vma and copy the page table entries.
531                  */
532                 *pprev = tmp;
533                 pprev = &tmp->vm_next;
534                 tmp->vm_prev = prev;
535                 prev = tmp;
536 
537                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
538                 rb_link = &tmp->vm_rb.rb_right;
539                 rb_parent = &tmp->vm_rb;
540 
541                 mm->map_count++;
542                 if (!(tmp->vm_flags & VM_WIPEONFORK))
543                         retval = copy_page_range(mm, oldmm, mpnt);
544 
545                 if (tmp->vm_ops && tmp->vm_ops->open)
546                         tmp->vm_ops->open(tmp);
547 
548                 if (retval)
549                         goto out;
550         }
551         /* a new mm has just been created */
552         retval = arch_dup_mmap(oldmm, mm);
553 out:
554         up_write(&mm->mmap_sem);
555         flush_tlb_mm(oldmm);
556         up_write(&oldmm->mmap_sem);
557         dup_userfaultfd_complete(&uf);
558 fail_uprobe_end:
559         uprobe_end_dup_mmap();
560         return retval;
561 fail_nomem_anon_vma_fork:
562         mpol_put(vma_policy(tmp));
563 fail_nomem_policy:
564         vm_area_free(tmp);
565 fail_nomem:
566         retval = -ENOMEM;
567         vm_unacct_memory(charge);
568         goto out;
569 }
570 
571 static inline int mm_alloc_pgd(struct mm_struct *mm)
572 {
573         mm->pgd = pgd_alloc(mm);
574         if (unlikely(!mm->pgd))
575                 return -ENOMEM;
576         return 0;
577 }
578 
579 static inline void mm_free_pgd(struct mm_struct *mm)
580 {
581         pgd_free(mm, mm->pgd);
582 }
583 #else
584 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
585 {
586         down_write(&oldmm->mmap_sem);
587         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
588         up_write(&oldmm->mmap_sem);
589         return 0;
590 }
591 #define mm_alloc_pgd(mm)        (0)
592 #define mm_free_pgd(mm)
593 #endif /* CONFIG_MMU */
594 
595 static void check_mm(struct mm_struct *mm)
596 {
597         int i;
598 
599         for (i = 0; i < NR_MM_COUNTERS; i++) {
600                 long x = atomic_long_read(&mm->rss_stat.count[i]);
601 
602                 if (unlikely(x))
603                         printk(KERN_ALERT "BUG: Bad rss-counter state "
604                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
605         }
606 
607         if (mm_pgtables_bytes(mm))
608                 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
609                                 mm_pgtables_bytes(mm));
610 
611 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
612         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
613 #endif
614 }
615 
616 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
617 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
618 
619 /*
620  * Called when the last reference to the mm
621  * is dropped: either by a lazy thread or by
622  * mmput. Free the page directory and the mm.
623  */
624 void __mmdrop(struct mm_struct *mm)
625 {
626         BUG_ON(mm == &init_mm);
627         WARN_ON_ONCE(mm == current->mm);
628         WARN_ON_ONCE(mm == current->active_mm);
629         mm_free_pgd(mm);
630         destroy_context(mm);
631         hmm_mm_destroy(mm);
632         mmu_notifier_mm_destroy(mm);
633         check_mm(mm);
634         put_user_ns(mm->user_ns);
635         free_mm(mm);
636 }
637 EXPORT_SYMBOL_GPL(__mmdrop);
638 
639 static void mmdrop_async_fn(struct work_struct *work)
640 {
641         struct mm_struct *mm;
642 
643         mm = container_of(work, struct mm_struct, async_put_work);
644         __mmdrop(mm);
645 }
646 
647 static void mmdrop_async(struct mm_struct *mm)
648 {
649         if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
650                 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
651                 schedule_work(&mm->async_put_work);
652         }
653 }
654 
655 static inline void free_signal_struct(struct signal_struct *sig)
656 {
657         taskstats_tgid_free(sig);
658         sched_autogroup_exit(sig);
659         /*
660          * __mmdrop is not safe to call from softirq context on x86 due to
661          * pgd_dtor so postpone it to the async context
662          */
663         if (sig->oom_mm)
664                 mmdrop_async(sig->oom_mm);
665         kmem_cache_free(signal_cachep, sig);
666 }
667 
668 static inline void put_signal_struct(struct signal_struct *sig)
669 {
670         if (atomic_dec_and_test(&sig->sigcnt))
671                 free_signal_struct(sig);
672 }
673 
674 void __put_task_struct(struct task_struct *tsk)
675 {
676         WARN_ON(!tsk->exit_state);
677         WARN_ON(atomic_read(&tsk->usage));
678         WARN_ON(tsk == current);
679 
680         cgroup_free(tsk);
681         task_numa_free(tsk);
682         security_task_free(tsk);
683         exit_creds(tsk);
684         delayacct_tsk_free(tsk);
685         put_signal_struct(tsk->signal);
686 
687         if (!profile_handoff_task(tsk))
688                 free_task(tsk);
689 }
690 EXPORT_SYMBOL_GPL(__put_task_struct);
691 
692 void __init __weak arch_task_cache_init(void) { }
693 
694 /*
695  * set_max_threads
696  */
697 static void set_max_threads(unsigned int max_threads_suggested)
698 {
699         u64 threads;
700 
701         /*
702          * The number of threads shall be limited such that the thread
703          * structures may only consume a small part of the available memory.
704          */
705         if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
706                 threads = MAX_THREADS;
707         else
708                 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
709                                     (u64) THREAD_SIZE * 8UL);
710 
711         if (threads > max_threads_suggested)
712                 threads = max_threads_suggested;
713 
714         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
715 }
716 
717 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
718 /* Initialized by the architecture: */
719 int arch_task_struct_size __read_mostly;
720 #endif
721 
722 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
723 {
724         /* Fetch thread_struct whitelist for the architecture. */
725         arch_thread_struct_whitelist(offset, size);
726 
727         /*
728          * Handle zero-sized whitelist or empty thread_struct, otherwise
729          * adjust offset to position of thread_struct in task_struct.
730          */
731         if (unlikely(*size == 0))
732                 *offset = 0;
733         else
734                 *offset += offsetof(struct task_struct, thread);
735 }
736 
737 void __init fork_init(void)
738 {
739         int i;
740 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
741 #ifndef ARCH_MIN_TASKALIGN
742 #define ARCH_MIN_TASKALIGN      0
743 #endif
744         int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
745         unsigned long useroffset, usersize;
746 
747         /* create a slab on which task_structs can be allocated */
748         task_struct_whitelist(&useroffset, &usersize);
749         task_struct_cachep = kmem_cache_create_usercopy("task_struct",
750                         arch_task_struct_size, align,
751                         SLAB_PANIC|SLAB_ACCOUNT,
752                         useroffset, usersize, NULL);
753 #endif
754 
755         /* do the arch specific task caches init */
756         arch_task_cache_init();
757 
758         set_max_threads(MAX_THREADS);
759 
760         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
761         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
762         init_task.signal->rlim[RLIMIT_SIGPENDING] =
763                 init_task.signal->rlim[RLIMIT_NPROC];
764 
765         for (i = 0; i < UCOUNT_COUNTS; i++) {
766                 init_user_ns.ucount_max[i] = max_threads/2;
767         }
768 
769 #ifdef CONFIG_VMAP_STACK
770         cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
771                           NULL, free_vm_stack_cache);
772 #endif
773 
774         lockdep_init_task(&init_task);
775 }
776 
777 int __weak arch_dup_task_struct(struct task_struct *dst,
778                                                struct task_struct *src)
779 {
780         *dst = *src;
781         return 0;
782 }
783 
784 void set_task_stack_end_magic(struct task_struct *tsk)
785 {
786         unsigned long *stackend;
787 
788         stackend = end_of_stack(tsk);
789         *stackend = STACK_END_MAGIC;    /* for overflow detection */
790 }
791 
792 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
793 {
794         struct task_struct *tsk;
795         unsigned long *stack;
796         struct vm_struct *stack_vm_area;
797         int err;
798 
799         if (node == NUMA_NO_NODE)
800                 node = tsk_fork_get_node(orig);
801         tsk = alloc_task_struct_node(node);
802         if (!tsk)
803                 return NULL;
804 
805         stack = alloc_thread_stack_node(tsk, node);
806         if (!stack)
807                 goto free_tsk;
808 
809         stack_vm_area = task_stack_vm_area(tsk);
810 
811         err = arch_dup_task_struct(tsk, orig);
812 
813         /*
814          * arch_dup_task_struct() clobbers the stack-related fields.  Make
815          * sure they're properly initialized before using any stack-related
816          * functions again.
817          */
818         tsk->stack = stack;
819 #ifdef CONFIG_VMAP_STACK
820         tsk->stack_vm_area = stack_vm_area;
821 #endif
822 #ifdef CONFIG_THREAD_INFO_IN_TASK
823         atomic_set(&tsk->stack_refcount, 1);
824 #endif
825 
826         if (err)
827                 goto free_stack;
828 
829 #ifdef CONFIG_SECCOMP
830         /*
831          * We must handle setting up seccomp filters once we're under
832          * the sighand lock in case orig has changed between now and
833          * then. Until then, filter must be NULL to avoid messing up
834          * the usage counts on the error path calling free_task.
835          */
836         tsk->seccomp.filter = NULL;
837 #endif
838 
839         setup_thread_stack(tsk, orig);
840         clear_user_return_notifier(tsk);
841         clear_tsk_need_resched(tsk);
842         set_task_stack_end_magic(tsk);
843 
844 #ifdef CONFIG_STACKPROTECTOR
845         tsk->stack_canary = get_random_canary();
846 #endif
847 
848         /*
849          * One for us, one for whoever does the "release_task()" (usually
850          * parent)
851          */
852         atomic_set(&tsk->usage, 2);
853 #ifdef CONFIG_BLK_DEV_IO_TRACE
854         tsk->btrace_seq = 0;
855 #endif
856         tsk->splice_pipe = NULL;
857         tsk->task_frag.page = NULL;
858         tsk->wake_q.next = NULL;
859 
860         account_kernel_stack(tsk, 1);
861 
862         kcov_task_init(tsk);
863 
864 #ifdef CONFIG_FAULT_INJECTION
865         tsk->fail_nth = 0;
866 #endif
867 
868         return tsk;
869 
870 free_stack:
871         free_thread_stack(tsk);
872 free_tsk:
873         free_task_struct(tsk);
874         return NULL;
875 }
876 
877 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
878 
879 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
880 
881 static int __init coredump_filter_setup(char *s)
882 {
883         default_dump_filter =
884                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
885                 MMF_DUMP_FILTER_MASK;
886         return 1;
887 }
888 
889 __setup("coredump_filter=", coredump_filter_setup);
890 
891 #include <linux/init_task.h>
892 
893 static void mm_init_aio(struct mm_struct *mm)
894 {
895 #ifdef CONFIG_AIO
896         spin_lock_init(&mm->ioctx_lock);
897         mm->ioctx_table = NULL;
898 #endif
899 }
900 
901 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
902 {
903 #ifdef CONFIG_MEMCG
904         mm->owner = p;
905 #endif
906 }
907 
908 static void mm_init_uprobes_state(struct mm_struct *mm)
909 {
910 #ifdef CONFIG_UPROBES
911         mm->uprobes_state.xol_area = NULL;
912 #endif
913 }
914 
915 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
916         struct user_namespace *user_ns)
917 {
918         mm->mmap = NULL;
919         mm->mm_rb = RB_ROOT;
920         mm->vmacache_seqnum = 0;
921         atomic_set(&mm->mm_users, 1);
922         atomic_set(&mm->mm_count, 1);
923         init_rwsem(&mm->mmap_sem);
924         INIT_LIST_HEAD(&mm->mmlist);
925         mm->core_state = NULL;
926         mm_pgtables_bytes_init(mm);
927         mm->map_count = 0;
928         mm->locked_vm = 0;
929         mm->pinned_vm = 0;
930         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
931         spin_lock_init(&mm->page_table_lock);
932         spin_lock_init(&mm->arg_lock);
933         mm_init_cpumask(mm);
934         mm_init_aio(mm);
935         mm_init_owner(mm, p);
936         RCU_INIT_POINTER(mm->exe_file, NULL);
937         mmu_notifier_mm_init(mm);
938         hmm_mm_init(mm);
939         init_tlb_flush_pending(mm);
940 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
941         mm->pmd_huge_pte = NULL;
942 #endif
943         mm_init_uprobes_state(mm);
944 
945         if (current->mm) {
946                 mm->flags = current->mm->flags & MMF_INIT_MASK;
947                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
948         } else {
949                 mm->flags = default_dump_filter;
950                 mm->def_flags = 0;
951         }
952 
953         if (mm_alloc_pgd(mm))
954                 goto fail_nopgd;
955 
956         if (init_new_context(p, mm))
957                 goto fail_nocontext;
958 
959         mm->user_ns = get_user_ns(user_ns);
960         return mm;
961 
962 fail_nocontext:
963         mm_free_pgd(mm);
964 fail_nopgd:
965         free_mm(mm);
966         return NULL;
967 }
968 
969 /*
970  * Allocate and initialize an mm_struct.
971  */
972 struct mm_struct *mm_alloc(void)
973 {
974         struct mm_struct *mm;
975 
976         mm = allocate_mm();
977         if (!mm)
978                 return NULL;
979 
980         memset(mm, 0, sizeof(*mm));
981         return mm_init(mm, current, current_user_ns());
982 }
983 
984 static inline void __mmput(struct mm_struct *mm)
985 {
986         VM_BUG_ON(atomic_read(&mm->mm_users));
987 
988         uprobe_clear_state(mm);
989         exit_aio(mm);
990         ksm_exit(mm);
991         khugepaged_exit(mm); /* must run before exit_mmap */
992         exit_mmap(mm);
993         mm_put_huge_zero_page(mm);
994         set_mm_exe_file(mm, NULL);
995         if (!list_empty(&mm->mmlist)) {
996                 spin_lock(&mmlist_lock);
997                 list_del(&mm->mmlist);
998                 spin_unlock(&mmlist_lock);
999         }
1000         if (mm->binfmt)
1001                 module_put(mm->binfmt->module);
1002         mmdrop(mm);
1003 }
1004 
1005 /*
1006  * Decrement the use count and release all resources for an mm.
1007  */
1008 void mmput(struct mm_struct *mm)
1009 {
1010         might_sleep();
1011 
1012         if (atomic_dec_and_test(&mm->mm_users))
1013                 __mmput(mm);
1014 }
1015 EXPORT_SYMBOL_GPL(mmput);
1016 
1017 #ifdef CONFIG_MMU
1018 static void mmput_async_fn(struct work_struct *work)
1019 {
1020         struct mm_struct *mm = container_of(work, struct mm_struct,
1021                                             async_put_work);
1022 
1023         __mmput(mm);
1024 }
1025 
1026 void mmput_async(struct mm_struct *mm)
1027 {
1028         if (atomic_dec_and_test(&mm->mm_users)) {
1029                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1030                 schedule_work(&mm->async_put_work);
1031         }
1032 }
1033 #endif
1034 
1035 /**
1036  * set_mm_exe_file - change a reference to the mm's executable file
1037  *
1038  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1039  *
1040  * Main users are mmput() and sys_execve(). Callers prevent concurrent
1041  * invocations: in mmput() nobody alive left, in execve task is single
1042  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1043  * mm->exe_file, but does so without using set_mm_exe_file() in order
1044  * to do avoid the need for any locks.
1045  */
1046 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1047 {
1048         struct file *old_exe_file;
1049 
1050         /*
1051          * It is safe to dereference the exe_file without RCU as
1052          * this function is only called if nobody else can access
1053          * this mm -- see comment above for justification.
1054          */
1055         old_exe_file = rcu_dereference_raw(mm->exe_file);
1056 
1057         if (new_exe_file)
1058                 get_file(new_exe_file);
1059         rcu_assign_pointer(mm->exe_file, new_exe_file);
1060         if (old_exe_file)
1061                 fput(old_exe_file);
1062 }
1063 
1064 /**
1065  * get_mm_exe_file - acquire a reference to the mm's executable file
1066  *
1067  * Returns %NULL if mm has no associated executable file.
1068  * User must release file via fput().
1069  */
1070 struct file *get_mm_exe_file(struct mm_struct *mm)
1071 {
1072         struct file *exe_file;
1073 
1074         rcu_read_lock();
1075         exe_file = rcu_dereference(mm->exe_file);
1076         if (exe_file && !get_file_rcu(exe_file))
1077                 exe_file = NULL;
1078         rcu_read_unlock();
1079         return exe_file;
1080 }
1081 EXPORT_SYMBOL(get_mm_exe_file);
1082 
1083 /**
1084  * get_task_exe_file - acquire a reference to the task's executable file
1085  *
1086  * Returns %NULL if task's mm (if any) has no associated executable file or
1087  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1088  * User must release file via fput().
1089  */
1090 struct file *get_task_exe_file(struct task_struct *task)
1091 {
1092         struct file *exe_file = NULL;
1093         struct mm_struct *mm;
1094 
1095         task_lock(task);
1096         mm = task->mm;
1097         if (mm) {
1098                 if (!(task->flags & PF_KTHREAD))
1099                         exe_file = get_mm_exe_file(mm);
1100         }
1101         task_unlock(task);
1102         return exe_file;
1103 }
1104 EXPORT_SYMBOL(get_task_exe_file);
1105 
1106 /**
1107  * get_task_mm - acquire a reference to the task's mm
1108  *
1109  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1110  * this kernel workthread has transiently adopted a user mm with use_mm,
1111  * to do its AIO) is not set and if so returns a reference to it, after
1112  * bumping up the use count.  User must release the mm via mmput()
1113  * after use.  Typically used by /proc and ptrace.
1114  */
1115 struct mm_struct *get_task_mm(struct task_struct *task)
1116 {
1117         struct mm_struct *mm;
1118 
1119         task_lock(task);
1120         mm = task->mm;
1121         if (mm) {
1122                 if (task->flags & PF_KTHREAD)
1123                         mm = NULL;
1124                 else
1125                         mmget(mm);
1126         }
1127         task_unlock(task);
1128         return mm;
1129 }
1130 EXPORT_SYMBOL_GPL(get_task_mm);
1131 
1132 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1133 {
1134         struct mm_struct *mm;
1135         int err;
1136 
1137         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
1138         if (err)
1139                 return ERR_PTR(err);
1140 
1141         mm = get_task_mm(task);
1142         if (mm && mm != current->mm &&
1143                         !ptrace_may_access(task, mode)) {
1144                 mmput(mm);
1145                 mm = ERR_PTR(-EACCES);
1146         }
1147         mutex_unlock(&task->signal->cred_guard_mutex);
1148 
1149         return mm;
1150 }
1151 
1152 static void complete_vfork_done(struct task_struct *tsk)
1153 {
1154         struct completion *vfork;
1155 
1156         task_lock(tsk);
1157         vfork = tsk->vfork_done;
1158         if (likely(vfork)) {
1159                 tsk->vfork_done = NULL;
1160                 complete(vfork);
1161         }
1162         task_unlock(tsk);
1163 }
1164 
1165 static int wait_for_vfork_done(struct task_struct *child,
1166                                 struct completion *vfork)
1167 {
1168         int killed;
1169 
1170         freezer_do_not_count();
1171         killed = wait_for_completion_killable(vfork);
1172         freezer_count();
1173 
1174         if (killed) {
1175                 task_lock(child);
1176                 child->vfork_done = NULL;
1177                 task_unlock(child);
1178         }
1179 
1180         put_task_struct(child);
1181         return killed;
1182 }
1183 
1184 /* Please note the differences between mmput and mm_release.
1185  * mmput is called whenever we stop holding onto a mm_struct,
1186  * error success whatever.
1187  *
1188  * mm_release is called after a mm_struct has been removed
1189  * from the current process.
1190  *
1191  * This difference is important for error handling, when we
1192  * only half set up a mm_struct for a new process and need to restore
1193  * the old one.  Because we mmput the new mm_struct before
1194  * restoring the old one. . .
1195  * Eric Biederman 10 January 1998
1196  */
1197 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1198 {
1199         /* Get rid of any futexes when releasing the mm */
1200 #ifdef CONFIG_FUTEX
1201         if (unlikely(tsk->robust_list)) {
1202                 exit_robust_list(tsk);
1203                 tsk->robust_list = NULL;
1204         }
1205 #ifdef CONFIG_COMPAT
1206         if (unlikely(tsk->compat_robust_list)) {
1207                 compat_exit_robust_list(tsk);
1208                 tsk->compat_robust_list = NULL;
1209         }
1210 #endif
1211         if (unlikely(!list_empty(&tsk->pi_state_list)))
1212                 exit_pi_state_list(tsk);
1213 #endif
1214 
1215         uprobe_free_utask(tsk);
1216 
1217         /* Get rid of any cached register state */
1218         deactivate_mm(tsk, mm);
1219 
1220         /*
1221          * Signal userspace if we're not exiting with a core dump
1222          * because we want to leave the value intact for debugging
1223          * purposes.
1224          */
1225         if (tsk->clear_child_tid) {
1226                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1227                     atomic_read(&mm->mm_users) > 1) {
1228                         /*
1229                          * We don't check the error code - if userspace has
1230                          * not set up a proper pointer then tough luck.
1231                          */
1232                         put_user(0, tsk->clear_child_tid);
1233                         do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1234                                         1, NULL, NULL, 0, 0);
1235                 }
1236                 tsk->clear_child_tid = NULL;
1237         }
1238 
1239         /*
1240          * All done, finally we can wake up parent and return this mm to him.
1241          * Also kthread_stop() uses this completion for synchronization.
1242          */
1243         if (tsk->vfork_done)
1244                 complete_vfork_done(tsk);
1245 }
1246 
1247 /*
1248  * Allocate a new mm structure and copy contents from the
1249  * mm structure of the passed in task structure.
1250  */
1251 static struct mm_struct *dup_mm(struct task_struct *tsk)
1252 {
1253         struct mm_struct *mm, *oldmm = current->mm;
1254         int err;
1255 
1256         mm = allocate_mm();
1257         if (!mm)
1258                 goto fail_nomem;
1259 
1260         memcpy(mm, oldmm, sizeof(*mm));
1261 
1262         if (!mm_init(mm, tsk, mm->user_ns))
1263                 goto fail_nomem;
1264 
1265         err = dup_mmap(mm, oldmm);
1266         if (err)
1267                 goto free_pt;
1268 
1269         mm->hiwater_rss = get_mm_rss(mm);
1270         mm->hiwater_vm = mm->total_vm;
1271 
1272         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1273                 goto free_pt;
1274 
1275         return mm;
1276 
1277 free_pt:
1278         /* don't put binfmt in mmput, we haven't got module yet */
1279         mm->binfmt = NULL;
1280         mmput(mm);
1281 
1282 fail_nomem:
1283         return NULL;
1284 }
1285 
1286 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1287 {
1288         struct mm_struct *mm, *oldmm;
1289         int retval;
1290 
1291         tsk->min_flt = tsk->maj_flt = 0;
1292         tsk->nvcsw = tsk->nivcsw = 0;
1293 #ifdef CONFIG_DETECT_HUNG_TASK
1294         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1295 #endif
1296 
1297         tsk->mm = NULL;
1298         tsk->active_mm = NULL;
1299 
1300         /*
1301          * Are we cloning a kernel thread?
1302          *
1303          * We need to steal a active VM for that..
1304          */
1305         oldmm = current->mm;
1306         if (!oldmm)
1307                 return 0;
1308 
1309         /* initialize the new vmacache entries */
1310         vmacache_flush(tsk);
1311 
1312         if (clone_flags & CLONE_VM) {
1313                 mmget(oldmm);
1314                 mm = oldmm;
1315                 goto good_mm;
1316         }
1317 
1318         retval = -ENOMEM;
1319         mm = dup_mm(tsk);
1320         if (!mm)
1321                 goto fail_nomem;
1322 
1323 good_mm:
1324         tsk->mm = mm;
1325         tsk->active_mm = mm;
1326         return 0;
1327 
1328 fail_nomem:
1329         return retval;
1330 }
1331 
1332 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1333 {
1334         struct fs_struct *fs = current->fs;
1335         if (clone_flags & CLONE_FS) {
1336                 /* tsk->fs is already what we want */
1337                 spin_lock(&fs->lock);
1338                 if (fs->in_exec) {
1339                         spin_unlock(&fs->lock);
1340                         return -EAGAIN;
1341                 }
1342                 fs->users++;
1343                 spin_unlock(&fs->lock);
1344                 return 0;
1345         }
1346         tsk->fs = copy_fs_struct(fs);
1347         if (!tsk->fs)
1348                 return -ENOMEM;
1349         return 0;
1350 }
1351 
1352 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1353 {
1354         struct files_struct *oldf, *newf;
1355         int error = 0;
1356 
1357         /*
1358          * A background process may not have any files ...
1359          */
1360         oldf = current->files;
1361         if (!oldf)
1362                 goto out;
1363 
1364         if (clone_flags & CLONE_FILES) {
1365                 atomic_inc(&oldf->count);
1366                 goto out;
1367         }
1368 
1369         newf = dup_fd(oldf, &error);
1370         if (!newf)
1371                 goto out;
1372 
1373         tsk->files = newf;
1374         error = 0;
1375 out:
1376         return error;
1377 }
1378 
1379 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1380 {
1381 #ifdef CONFIG_BLOCK
1382         struct io_context *ioc = current->io_context;
1383         struct io_context *new_ioc;
1384 
1385         if (!ioc)
1386                 return 0;
1387         /*
1388          * Share io context with parent, if CLONE_IO is set
1389          */
1390         if (clone_flags & CLONE_IO) {
1391                 ioc_task_link(ioc);
1392                 tsk->io_context = ioc;
1393         } else if (ioprio_valid(ioc->ioprio)) {
1394                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1395                 if (unlikely(!new_ioc))
1396                         return -ENOMEM;
1397 
1398                 new_ioc->ioprio = ioc->ioprio;
1399                 put_io_context(new_ioc);
1400         }
1401 #endif
1402         return 0;
1403 }
1404 
1405 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1406 {
1407         struct sighand_struct *sig;
1408 
1409         if (clone_flags & CLONE_SIGHAND) {
1410                 atomic_inc(&current->sighand->count);
1411                 return 0;
1412         }
1413         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1414         rcu_assign_pointer(tsk->sighand, sig);
1415         if (!sig)
1416                 return -ENOMEM;
1417 
1418         atomic_set(&sig->count, 1);
1419         spin_lock_irq(&current->sighand->siglock);
1420         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1421         spin_unlock_irq(&current->sighand->siglock);
1422         return 0;
1423 }
1424 
1425 void __cleanup_sighand(struct sighand_struct *sighand)
1426 {
1427         if (atomic_dec_and_test(&sighand->count)) {
1428                 signalfd_cleanup(sighand);
1429                 /*
1430                  * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1431                  * without an RCU grace period, see __lock_task_sighand().
1432                  */
1433                 kmem_cache_free(sighand_cachep, sighand);
1434         }
1435 }
1436 
1437 #ifdef CONFIG_POSIX_TIMERS
1438 /*
1439  * Initialize POSIX timer handling for a thread group.
1440  */
1441 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1442 {
1443         unsigned long cpu_limit;
1444 
1445         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1446         if (cpu_limit != RLIM_INFINITY) {
1447                 sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1448                 sig->cputimer.running = true;
1449         }
1450 
1451         /* The timer lists. */
1452         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1453         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1454         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1455 }
1456 #else
1457 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1458 #endif
1459 
1460 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1461 {
1462         struct signal_struct *sig;
1463 
1464         if (clone_flags & CLONE_THREAD)
1465                 return 0;
1466 
1467         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1468         tsk->signal = sig;
1469         if (!sig)
1470                 return -ENOMEM;
1471 
1472         sig->nr_threads = 1;
1473         atomic_set(&sig->live, 1);
1474         atomic_set(&sig->sigcnt, 1);
1475 
1476         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1477         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1478         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1479 
1480         init_waitqueue_head(&sig->wait_chldexit);
1481         sig->curr_target = tsk;
1482         init_sigpending(&sig->shared_pending);
1483         seqlock_init(&sig->stats_lock);
1484         prev_cputime_init(&sig->prev_cputime);
1485 
1486 #ifdef CONFIG_POSIX_TIMERS
1487         INIT_LIST_HEAD(&sig->posix_timers);
1488         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1489         sig->real_timer.function = it_real_fn;
1490 #endif
1491 
1492         task_lock(current->group_leader);
1493         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1494         task_unlock(current->group_leader);
1495 
1496         posix_cpu_timers_init_group(sig);
1497 
1498         tty_audit_fork(sig);
1499         sched_autogroup_fork(sig);
1500 
1501         sig->oom_score_adj = current->signal->oom_score_adj;
1502         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1503 
1504         mutex_init(&sig->cred_guard_mutex);
1505 
1506         return 0;
1507 }
1508 
1509 static void copy_seccomp(struct task_struct *p)
1510 {
1511 #ifdef CONFIG_SECCOMP
1512         /*
1513          * Must be called with sighand->lock held, which is common to
1514          * all threads in the group. Holding cred_guard_mutex is not
1515          * needed because this new task is not yet running and cannot
1516          * be racing exec.
1517          */
1518         assert_spin_locked(&current->sighand->siglock);
1519 
1520         /* Ref-count the new filter user, and assign it. */
1521         get_seccomp_filter(current);
1522         p->seccomp = current->seccomp;
1523 
1524         /*
1525          * Explicitly enable no_new_privs here in case it got set
1526          * between the task_struct being duplicated and holding the
1527          * sighand lock. The seccomp state and nnp must be in sync.
1528          */
1529         if (task_no_new_privs(current))
1530                 task_set_no_new_privs(p);
1531 
1532         /*
1533          * If the parent gained a seccomp mode after copying thread
1534          * flags and between before we held the sighand lock, we have
1535          * to manually enable the seccomp thread flag here.
1536          */
1537         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1538                 set_tsk_thread_flag(p, TIF_SECCOMP);
1539 #endif
1540 }
1541 
1542 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1543 {
1544         current->clear_child_tid = tidptr;
1545 
1546         return task_pid_vnr(current);
1547 }
1548 
1549 static void rt_mutex_init_task(struct task_struct *p)
1550 {
1551         raw_spin_lock_init(&p->pi_lock);
1552 #ifdef CONFIG_RT_MUTEXES
1553         p->pi_waiters = RB_ROOT_CACHED;
1554         p->pi_top_task = NULL;
1555         p->pi_blocked_on = NULL;
1556 #endif
1557 }
1558 
1559 #ifdef CONFIG_POSIX_TIMERS
1560 /*
1561  * Initialize POSIX timer handling for a single task.
1562  */
1563 static void posix_cpu_timers_init(struct task_struct *tsk)
1564 {
1565         tsk->cputime_expires.prof_exp = 0;
1566         tsk->cputime_expires.virt_exp = 0;
1567         tsk->cputime_expires.sched_exp = 0;
1568         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1569         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1570         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1571 }
1572 #else
1573 static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1574 #endif
1575 
1576 static inline void
1577 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1578 {
1579          task->pids[type].pid = pid;
1580 }
1581 
1582 static inline void rcu_copy_process(struct task_struct *p)
1583 {
1584 #ifdef CONFIG_PREEMPT_RCU
1585         p->rcu_read_lock_nesting = 0;
1586         p->rcu_read_unlock_special.s = 0;
1587         p->rcu_blocked_node = NULL;
1588         INIT_LIST_HEAD(&p->rcu_node_entry);
1589 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1590 #ifdef CONFIG_TASKS_RCU
1591         p->rcu_tasks_holdout = false;
1592         INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1593         p->rcu_tasks_idle_cpu = -1;
1594 #endif /* #ifdef CONFIG_TASKS_RCU */
1595 }
1596 
1597 /*
1598  * This creates a new process as a copy of the old one,
1599  * but does not actually start it yet.
1600  *
1601  * It copies the registers, and all the appropriate
1602  * parts of the process environment (as per the clone
1603  * flags). The actual kick-off is left to the caller.
1604  */
1605 static __latent_entropy struct task_struct *copy_process(
1606                                         unsigned long clone_flags,
1607                                         unsigned long stack_start,
1608                                         unsigned long stack_size,
1609                                         int __user *child_tidptr,
1610                                         struct pid *pid,
1611                                         int trace,
1612                                         unsigned long tls,
1613                                         int node)
1614 {
1615         int retval;
1616         struct task_struct *p;
1617 
1618         /*
1619          * Don't allow sharing the root directory with processes in a different
1620          * namespace
1621          */
1622         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1623                 return ERR_PTR(-EINVAL);
1624 
1625         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1626                 return ERR_PTR(-EINVAL);
1627 
1628         /*
1629          * Thread groups must share signals as well, and detached threads
1630          * can only be started up within the thread group.
1631          */
1632         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1633                 return ERR_PTR(-EINVAL);
1634 
1635         /*
1636          * Shared signal handlers imply shared VM. By way of the above,
1637          * thread groups also imply shared VM. Blocking this case allows
1638          * for various simplifications in other code.
1639          */
1640         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1641                 return ERR_PTR(-EINVAL);
1642 
1643         /*
1644          * Siblings of global init remain as zombies on exit since they are
1645          * not reaped by their parent (swapper). To solve this and to avoid
1646          * multi-rooted process trees, prevent global and container-inits
1647          * from creating siblings.
1648          */
1649         if ((clone_flags & CLONE_PARENT) &&
1650                                 current->signal->flags & SIGNAL_UNKILLABLE)
1651                 return ERR_PTR(-EINVAL);
1652 
1653         /*
1654          * If the new process will be in a different pid or user namespace
1655          * do not allow it to share a thread group with the forking task.
1656          */
1657         if (clone_flags & CLONE_THREAD) {
1658                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1659                     (task_active_pid_ns(current) !=
1660                                 current->nsproxy->pid_ns_for_children))
1661                         return ERR_PTR(-EINVAL);
1662         }
1663 
1664         retval = -ENOMEM;
1665         p = dup_task_struct(current, node);
1666         if (!p)
1667                 goto fork_out;
1668 
1669         /*
1670          * This _must_ happen before we call free_task(), i.e. before we jump
1671          * to any of the bad_fork_* labels. This is to avoid freeing
1672          * p->set_child_tid which is (ab)used as a kthread's data pointer for
1673          * kernel threads (PF_KTHREAD).
1674          */
1675         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1676         /*
1677          * Clear TID on mm_release()?
1678          */
1679         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1680 
1681         ftrace_graph_init_task(p);
1682 
1683         rt_mutex_init_task(p);
1684 
1685 #ifdef CONFIG_PROVE_LOCKING
1686         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1687         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1688 #endif
1689         retval = -EAGAIN;
1690         if (atomic_read(&p->real_cred->user->processes) >=
1691                         task_rlimit(p, RLIMIT_NPROC)) {
1692                 if (p->real_cred->user != INIT_USER &&
1693                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1694                         goto bad_fork_free;
1695         }
1696         current->flags &= ~PF_NPROC_EXCEEDED;
1697 
1698         retval = copy_creds(p, clone_flags);
1699         if (retval < 0)
1700                 goto bad_fork_free;
1701 
1702         /*
1703          * If multiple threads are within copy_process(), then this check
1704          * triggers too late. This doesn't hurt, the check is only there
1705          * to stop root fork bombs.
1706          */
1707         retval = -EAGAIN;
1708         if (nr_threads >= max_threads)
1709                 goto bad_fork_cleanup_count;
1710 
1711         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1712         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1713         p->flags |= PF_FORKNOEXEC;
1714         INIT_LIST_HEAD(&p->children);
1715         INIT_LIST_HEAD(&p->sibling);
1716         rcu_copy_process(p);
1717         p->vfork_done = NULL;
1718         spin_lock_init(&p->alloc_lock);
1719 
1720         init_sigpending(&p->pending);
1721 
1722         p->utime = p->stime = p->gtime = 0;
1723 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1724         p->utimescaled = p->stimescaled = 0;
1725 #endif
1726         prev_cputime_init(&p->prev_cputime);
1727 
1728 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1729         seqcount_init(&p->vtime.seqcount);
1730         p->vtime.starttime = 0;
1731         p->vtime.state = VTIME_INACTIVE;
1732 #endif
1733 
1734 #if defined(SPLIT_RSS_COUNTING)
1735         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1736 #endif
1737 
1738         p->default_timer_slack_ns = current->timer_slack_ns;
1739 
1740         task_io_accounting_init(&p->ioac);
1741         acct_clear_integrals(p);
1742 
1743         posix_cpu_timers_init(p);
1744 
1745         p->start_time = ktime_get_ns();
1746         p->real_start_time = ktime_get_boot_ns();
1747         p->io_context = NULL;
1748         audit_set_context(p, NULL);
1749         cgroup_fork(p);
1750 #ifdef CONFIG_NUMA
1751         p->mempolicy = mpol_dup(p->mempolicy);
1752         if (IS_ERR(p->mempolicy)) {
1753                 retval = PTR_ERR(p->mempolicy);
1754                 p->mempolicy = NULL;
1755                 goto bad_fork_cleanup_threadgroup_lock;
1756         }
1757 #endif
1758 #ifdef CONFIG_CPUSETS
1759         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1760         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1761         seqcount_init(&p->mems_allowed_seq);
1762 #endif
1763 #ifdef CONFIG_TRACE_IRQFLAGS
1764         p->irq_events = 0;
1765         p->hardirqs_enabled = 0;
1766         p->hardirq_enable_ip = 0;
1767         p->hardirq_enable_event = 0;
1768         p->hardirq_disable_ip = _THIS_IP_;
1769         p->hardirq_disable_event = 0;
1770         p->softirqs_enabled = 1;
1771         p->softirq_enable_ip = _THIS_IP_;
1772         p->softirq_enable_event = 0;
1773         p->softirq_disable_ip = 0;
1774         p->softirq_disable_event = 0;
1775         p->hardirq_context = 0;
1776         p->softirq_context = 0;
1777 #endif
1778 
1779         p->pagefault_disabled = 0;
1780 
1781 #ifdef CONFIG_LOCKDEP
1782         p->lockdep_depth = 0; /* no locks held yet */
1783         p->curr_chain_key = 0;
1784         p->lockdep_recursion = 0;
1785         lockdep_init_task(p);
1786 #endif
1787 
1788 #ifdef CONFIG_DEBUG_MUTEXES
1789         p->blocked_on = NULL; /* not blocked yet */
1790 #endif
1791 #ifdef CONFIG_BCACHE
1792         p->sequential_io        = 0;
1793         p->sequential_io_avg    = 0;
1794 #endif
1795 
1796         /* Perform scheduler related setup. Assign this task to a CPU. */
1797         retval = sched_fork(clone_flags, p);
1798         if (retval)
1799                 goto bad_fork_cleanup_policy;
1800 
1801         retval = perf_event_init_task(p);
1802         if (retval)
1803                 goto bad_fork_cleanup_policy;
1804         retval = audit_alloc(p);
1805         if (retval)
1806                 goto bad_fork_cleanup_perf;
1807         /* copy all the process information */
1808         shm_init_task(p);
1809         retval = security_task_alloc(p, clone_flags);
1810         if (retval)
1811                 goto bad_fork_cleanup_audit;
1812         retval = copy_semundo(clone_flags, p);
1813         if (retval)
1814                 goto bad_fork_cleanup_security;
1815         retval = copy_files(clone_flags, p);
1816         if (retval)
1817                 goto bad_fork_cleanup_semundo;
1818         retval = copy_fs(clone_flags, p);
1819         if (retval)
1820                 goto bad_fork_cleanup_files;
1821         retval = copy_sighand(clone_flags, p);
1822         if (retval)
1823                 goto bad_fork_cleanup_fs;
1824         retval = copy_signal(clone_flags, p);
1825         if (retval)
1826                 goto bad_fork_cleanup_sighand;
1827         retval = copy_mm(clone_flags, p);
1828         if (retval)
1829                 goto bad_fork_cleanup_signal;
1830         retval = copy_namespaces(clone_flags, p);
1831         if (retval)
1832                 goto bad_fork_cleanup_mm;
1833         retval = copy_io(clone_flags, p);
1834         if (retval)
1835                 goto bad_fork_cleanup_namespaces;
1836         retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1837         if (retval)
1838                 goto bad_fork_cleanup_io;
1839 
1840         if (pid != &init_struct_pid) {
1841                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1842                 if (IS_ERR(pid)) {
1843                         retval = PTR_ERR(pid);
1844                         goto bad_fork_cleanup_thread;
1845                 }
1846         }
1847 
1848 #ifdef CONFIG_BLOCK
1849         p->plug = NULL;
1850 #endif
1851 #ifdef CONFIG_FUTEX
1852         p->robust_list = NULL;
1853 #ifdef CONFIG_COMPAT
1854         p->compat_robust_list = NULL;
1855 #endif
1856         INIT_LIST_HEAD(&p->pi_state_list);
1857         p->pi_state_cache = NULL;
1858 #endif
1859         /*
1860          * sigaltstack should be cleared when sharing the same VM
1861          */
1862         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1863                 sas_ss_reset(p);
1864 
1865         /*
1866          * Syscall tracing and stepping should be turned off in the
1867          * child regardless of CLONE_PTRACE.
1868          */
1869         user_disable_single_step(p);
1870         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1871 #ifdef TIF_SYSCALL_EMU
1872         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1873 #endif
1874         clear_all_latency_tracing(p);
1875 
1876         /* ok, now we should be set up.. */
1877         p->pid = pid_nr(pid);
1878         if (clone_flags & CLONE_THREAD) {
1879                 p->exit_signal = -1;
1880                 p->group_leader = current->group_leader;
1881                 p->tgid = current->tgid;
1882         } else {
1883                 if (clone_flags & CLONE_PARENT)
1884                         p->exit_signal = current->group_leader->exit_signal;
1885                 else
1886                         p->exit_signal = (clone_flags & CSIGNAL);
1887                 p->group_leader = p;
1888                 p->tgid = p->pid;
1889         }
1890 
1891         p->nr_dirtied = 0;
1892         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1893         p->dirty_paused_when = 0;
1894 
1895         p->pdeath_signal = 0;
1896         INIT_LIST_HEAD(&p->thread_group);
1897         p->task_works = NULL;
1898 
1899         cgroup_threadgroup_change_begin(current);
1900         /*
1901          * Ensure that the cgroup subsystem policies allow the new process to be
1902          * forked. It should be noted the the new process's css_set can be changed
1903          * between here and cgroup_post_fork() if an organisation operation is in
1904          * progress.
1905          */
1906         retval = cgroup_can_fork(p);
1907         if (retval)
1908                 goto bad_fork_free_pid;
1909 
1910         /*
1911          * Make it visible to the rest of the system, but dont wake it up yet.
1912          * Need tasklist lock for parent etc handling!
1913          */
1914         write_lock_irq(&tasklist_lock);
1915 
1916         /* CLONE_PARENT re-uses the old parent */
1917         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1918                 p->real_parent = current->real_parent;
1919                 p->parent_exec_id = current->parent_exec_id;
1920         } else {
1921                 p->real_parent = current;
1922                 p->parent_exec_id = current->self_exec_id;
1923         }
1924 
1925         klp_copy_process(p);
1926 
1927         spin_lock(&current->sighand->siglock);
1928 
1929         /*
1930          * Copy seccomp details explicitly here, in case they were changed
1931          * before holding sighand lock.
1932          */
1933         copy_seccomp(p);
1934 
1935         rseq_fork(p, clone_flags);
1936 
1937         /*
1938          * Process group and session signals need to be delivered to just the
1939          * parent before the fork or both the parent and the child after the
1940          * fork. Restart if a signal comes in before we add the new process to
1941          * it's process group.
1942          * A fatal signal pending means that current will exit, so the new
1943          * thread can't slip out of an OOM kill (or normal SIGKILL).
1944         */
1945         recalc_sigpending();
1946         if (signal_pending(current)) {
1947                 retval = -ERESTARTNOINTR;
1948                 goto bad_fork_cancel_cgroup;
1949         }
1950         if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
1951                 retval = -ENOMEM;
1952                 goto bad_fork_cancel_cgroup;
1953         }
1954 
1955         if (likely(p->pid)) {
1956                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1957 
1958                 init_task_pid(p, PIDTYPE_PID, pid);
1959                 if (thread_group_leader(p)) {
1960                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1961                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1962 
1963                         if (is_child_reaper(pid)) {
1964                                 ns_of_pid(pid)->child_reaper = p;
1965                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1966                         }
1967 
1968                         p->signal->leader_pid = pid;
1969                         p->signal->tty = tty_kref_get(current->signal->tty);
1970                         /*
1971                          * Inherit has_child_subreaper flag under the same
1972                          * tasklist_lock with adding child to the process tree
1973                          * for propagate_has_child_subreaper optimization.
1974                          */
1975                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
1976                                                          p->real_parent->signal->is_child_subreaper;
1977                         list_add_tail(&p->sibling, &p->real_parent->children);
1978                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1979                         attach_pid(p, PIDTYPE_PGID);
1980                         attach_pid(p, PIDTYPE_SID);
1981                         __this_cpu_inc(process_counts);
1982                 } else {
1983                         current->signal->nr_threads++;
1984                         atomic_inc(&current->signal->live);
1985                         atomic_inc(&current->signal->sigcnt);
1986                         list_add_tail_rcu(&p->thread_group,
1987                                           &p->group_leader->thread_group);
1988                         list_add_tail_rcu(&p->thread_node,
1989                                           &p->signal->thread_head);
1990                 }
1991                 attach_pid(p, PIDTYPE_PID);
1992                 nr_threads++;
1993         }
1994 
1995         total_forks++;
1996         spin_unlock(&current->sighand->siglock);
1997         syscall_tracepoint_update(p);
1998         write_unlock_irq(&tasklist_lock);
1999 
2000         proc_fork_connector(p);
2001         cgroup_post_fork(p);
2002         cgroup_threadgroup_change_end(current);
2003         perf_event_fork(p);
2004 
2005         trace_task_newtask(p, clone_flags);
2006         uprobe_copy_process(p, clone_flags);
2007 
2008         return p;
2009 
2010 bad_fork_cancel_cgroup:
2011         spin_unlock(&current->sighand->siglock);
2012         write_unlock_irq(&tasklist_lock);
2013         cgroup_cancel_fork(p);
2014 bad_fork_free_pid:
2015         cgroup_threadgroup_change_end(current);
2016         if (pid != &init_struct_pid)
2017                 free_pid(pid);
2018 bad_fork_cleanup_thread:
2019         exit_thread(p);
2020 bad_fork_cleanup_io:
2021         if (p->io_context)
2022                 exit_io_context(p);
2023 bad_fork_cleanup_namespaces:
2024         exit_task_namespaces(p);
2025 bad_fork_cleanup_mm:
2026         if (p->mm)
2027                 mmput(p->mm);
2028 bad_fork_cleanup_signal:
2029         if (!(clone_flags & CLONE_THREAD))
2030                 free_signal_struct(p->signal);
2031 bad_fork_cleanup_sighand:
2032         __cleanup_sighand(p->sighand);
2033 bad_fork_cleanup_fs:
2034         exit_fs(p); /* blocking */
2035 bad_fork_cleanup_files:
2036         exit_files(p); /* blocking */
2037 bad_fork_cleanup_semundo:
2038         exit_sem(p);
2039 bad_fork_cleanup_security:
2040         security_task_free(p);
2041 bad_fork_cleanup_audit:
2042         audit_free(p);
2043 bad_fork_cleanup_perf:
2044         perf_event_free_task(p);
2045 bad_fork_cleanup_policy:
2046         lockdep_free_task(p);
2047 #ifdef CONFIG_NUMA
2048         mpol_put(p->mempolicy);
2049 bad_fork_cleanup_threadgroup_lock:
2050 #endif
2051         delayacct_tsk_free(p);
2052 bad_fork_cleanup_count:
2053         atomic_dec(&p->cred->user->processes);
2054         exit_creds(p);
2055 bad_fork_free:
2056         p->state = TASK_DEAD;
2057         put_task_stack(p);
2058         free_task(p);
2059 fork_out:
2060         return ERR_PTR(retval);
2061 }
2062 
2063 static inline void init_idle_pids(struct pid_link *links)
2064 {
2065         enum pid_type type;
2066 
2067         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2068                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
2069                 links[type].pid = &init_struct_pid;
2070         }
2071 }
2072 
2073 struct task_struct *fork_idle(int cpu)
2074 {
2075         struct task_struct *task;
2076         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
2077                             cpu_to_node(cpu));
2078         if (!IS_ERR(task)) {
2079                 init_idle_pids(task->pids);
2080                 init_idle(task, cpu);
2081         }
2082 
2083         return task;
2084 }
2085 
2086 /*
2087  *  Ok, this is the main fork-routine.
2088  *
2089  * It copies the process, and if successful kick-starts
2090  * it and waits for it to finish using the VM if required.
2091  */
2092 long _do_fork(unsigned long clone_flags,
2093               unsigned long stack_start,
2094               unsigned long stack_size,
2095               int __user *parent_tidptr,
2096               int __user *child_tidptr,
2097               unsigned long tls)
2098 {
2099         struct completion vfork;
2100         struct pid *pid;
2101         struct task_struct *p;
2102         int trace = 0;
2103         long nr;
2104 
2105         /*
2106          * Determine whether and which event to report to ptracer.  When
2107          * called from kernel_thread or CLONE_UNTRACED is explicitly
2108          * requested, no event is reported; otherwise, report if the event
2109          * for the type of forking is enabled.
2110          */
2111         if (!(clone_flags & CLONE_UNTRACED)) {
2112                 if (clone_flags & CLONE_VFORK)
2113                         trace = PTRACE_EVENT_VFORK;
2114                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
2115                         trace = PTRACE_EVENT_CLONE;
2116                 else
2117                         trace = PTRACE_EVENT_FORK;
2118 
2119                 if (likely(!ptrace_event_enabled(current, trace)))
2120                         trace = 0;
2121         }
2122 
2123         p = copy_process(clone_flags, stack_start, stack_size,
2124                          child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
2125         add_latent_entropy();
2126 
2127         if (IS_ERR(p))
2128                 return PTR_ERR(p);
2129 
2130         /*
2131          * Do this prior waking up the new thread - the thread pointer
2132          * might get invalid after that point, if the thread exits quickly.
2133          */
2134         trace_sched_process_fork(current, p);
2135 
2136         pid = get_task_pid(p, PIDTYPE_PID);
2137         nr = pid_vnr(pid);
2138 
2139         if (clone_flags & CLONE_PARENT_SETTID)
2140                 put_user(nr, parent_tidptr);
2141 
2142         if (clone_flags & CLONE_VFORK) {
2143                 p->vfork_done = &vfork;
2144                 init_completion(&vfork);
2145                 get_task_struct(p);
2146         }
2147 
2148         wake_up_new_task(p);
2149 
2150         /* forking complete and child started to run, tell ptracer */
2151         if (unlikely(trace))
2152                 ptrace_event_pid(trace, pid);
2153 
2154         if (clone_flags & CLONE_VFORK) {
2155                 if (!wait_for_vfork_done(p, &vfork))
2156                         ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2157         }
2158 
2159         put_pid(pid);
2160         return nr;
2161 }
2162 
2163 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2164 /* For compatibility with architectures that call do_fork directly rather than
2165  * using the syscall entry points below. */
2166 long do_fork(unsigned long clone_flags,
2167               unsigned long stack_start,
2168               unsigned long stack_size,
2169               int __user *parent_tidptr,
2170               int __user *child_tidptr)
2171 {
2172         return _do_fork(clone_flags, stack_start, stack_size,
2173                         parent_tidptr, child_tidptr, 0);
2174 }
2175 #endif
2176 
2177 /*
2178  * Create a kernel thread.
2179  */
2180 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2181 {
2182         return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2183                 (unsigned long)arg, NULL, NULL, 0);
2184 }
2185 
2186 #ifdef __ARCH_WANT_SYS_FORK
2187 SYSCALL_DEFINE0(fork)
2188 {
2189 #ifdef CONFIG_MMU
2190         return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2191 #else
2192         /* can not support in nommu mode */
2193         return -EINVAL;
2194 #endif
2195 }
2196 #endif
2197 
2198 #ifdef __ARCH_WANT_SYS_VFORK
2199 SYSCALL_DEFINE0(vfork)
2200 {
2201         return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2202                         0, NULL, NULL, 0);
2203 }
2204 #endif
2205 
2206 #ifdef __ARCH_WANT_SYS_CLONE
2207 #ifdef CONFIG_CLONE_BACKWARDS
2208 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2209                  int __user *, parent_tidptr,
2210                  unsigned long, tls,
2211                  int __user *, child_tidptr)
2212 #elif defined(CONFIG_CLONE_BACKWARDS2)
2213 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2214                  int __user *, parent_tidptr,
2215                  int __user *, child_tidptr,
2216                  unsigned long, tls)
2217 #elif defined(CONFIG_CLONE_BACKWARDS3)
2218 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2219                 int, stack_size,
2220                 int __user *, parent_tidptr,
2221                 int __user *, child_tidptr,
2222                 unsigned long, tls)
2223 #else
2224 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2225                  int __user *, parent_tidptr,
2226                  int __user *, child_tidptr,
2227                  unsigned long, tls)
2228 #endif
2229 {
2230         return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2231 }
2232 #endif
2233 
2234 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2235 {
2236         struct task_struct *leader, *parent, *child;
2237         int res;
2238 
2239         read_lock(&tasklist_lock);
2240         leader = top = top->group_leader;
2241 down:
2242         for_each_thread(leader, parent) {
2243                 list_for_each_entry(child, &parent->children, sibling) {
2244                         res = visitor(child, data);
2245                         if (res) {
2246                                 if (res < 0)
2247                                         goto out;
2248                                 leader = child;
2249                                 goto down;
2250                         }
2251 up:
2252                         ;
2253                 }
2254         }
2255 
2256         if (leader != top) {
2257                 child = leader;
2258                 parent = child->real_parent;
2259                 leader = parent->group_leader;
2260                 goto up;
2261         }
2262 out:
2263         read_unlock(&tasklist_lock);
2264 }
2265 
2266 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2267 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2268 #endif
2269 
2270 static void sighand_ctor(void *data)
2271 {
2272         struct sighand_struct *sighand = data;
2273 
2274         spin_lock_init(&sighand->siglock);
2275         init_waitqueue_head(&sighand->signalfd_wqh);
2276 }
2277 
2278 void __init proc_caches_init(void)
2279 {
2280         sighand_cachep = kmem_cache_create("sighand_cache",
2281                         sizeof(struct sighand_struct), 0,
2282                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2283                         SLAB_ACCOUNT, sighand_ctor);
2284         signal_cachep = kmem_cache_create("signal_cache",
2285                         sizeof(struct signal_struct), 0,
2286                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2287                         NULL);
2288         files_cachep = kmem_cache_create("files_cache",
2289                         sizeof(struct files_struct), 0,
2290                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2291                         NULL);
2292         fs_cachep = kmem_cache_create("fs_cache",
2293                         sizeof(struct fs_struct), 0,
2294                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2295                         NULL);
2296         /*
2297          * FIXME! The "sizeof(struct mm_struct)" currently includes the
2298          * whole struct cpumask for the OFFSTACK case. We could change
2299          * this to *only* allocate as much of it as required by the
2300          * maximum number of CPU's we can ever have.  The cpumask_allocation
2301          * is at the end of the structure, exactly for that reason.
2302          */
2303         mm_cachep = kmem_cache_create_usercopy("mm_struct",
2304                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
2305                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2306                         offsetof(struct mm_struct, saved_auxv),
2307                         sizeof_field(struct mm_struct, saved_auxv),
2308                         NULL);
2309         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2310         mmap_init();
2311         nsproxy_cache_init();
2312 }
2313 
2314 /*
2315  * Check constraints on flags passed to the unshare system call.
2316  */
2317 static int check_unshare_flags(unsigned long unshare_flags)
2318 {
2319         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2320                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2321                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2322                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2323                 return -EINVAL;
2324         /*
2325          * Not implemented, but pretend it works if there is nothing
2326          * to unshare.  Note that unsharing the address space or the
2327          * signal handlers also need to unshare the signal queues (aka
2328          * CLONE_THREAD).
2329          */
2330         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2331                 if (!thread_group_empty(current))
2332                         return -EINVAL;
2333         }
2334         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2335                 if (atomic_read(&current->sighand->count) > 1)
2336                         return -EINVAL;
2337         }
2338         if (unshare_flags & CLONE_VM) {
2339                 if (!current_is_single_threaded())
2340                         return -EINVAL;
2341         }
2342 
2343         return 0;
2344 }
2345 
2346 /*
2347  * Unshare the filesystem structure if it is being shared
2348  */
2349 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2350 {
2351         struct fs_struct *fs = current->fs;
2352 
2353         if (!(unshare_flags & CLONE_FS) || !fs)
2354                 return 0;
2355 
2356         /* don't need lock here; in the worst case we'll do useless copy */
2357         if (fs->users == 1)
2358                 return 0;
2359 
2360         *new_fsp = copy_fs_struct(fs);
2361         if (!*new_fsp)
2362                 return -ENOMEM;
2363 
2364         return 0;
2365 }
2366 
2367 /*
2368  * Unshare file descriptor table if it is being shared
2369  */
2370 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2371 {
2372         struct files_struct *fd = current->files;
2373         int error = 0;
2374 
2375         if ((unshare_flags & CLONE_FILES) &&
2376             (fd && atomic_read(&fd->count) > 1)) {
2377                 *new_fdp = dup_fd(fd, &error);
2378                 if (!*new_fdp)
2379                         return error;
2380         }
2381 
2382         return 0;
2383 }
2384 
2385 /*
2386  * unshare allows a process to 'unshare' part of the process
2387  * context which was originally shared using clone.  copy_*
2388  * functions used by do_fork() cannot be used here directly
2389  * because they modify an inactive task_struct that is being
2390  * constructed. Here we are modifying the current, active,
2391  * task_struct.
2392  */
2393 int ksys_unshare(unsigned long unshare_flags)
2394 {
2395         struct fs_struct *fs, *new_fs = NULL;
2396         struct files_struct *fd, *new_fd = NULL;
2397         struct cred *new_cred = NULL;
2398         struct nsproxy *new_nsproxy = NULL;
2399         int do_sysvsem = 0;
2400         int err;
2401 
2402         /*
2403          * If unsharing a user namespace must also unshare the thread group
2404          * and unshare the filesystem root and working directories.
2405          */
2406         if (unshare_flags & CLONE_NEWUSER)
2407                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2408         /*
2409          * If unsharing vm, must also unshare signal handlers.
2410          */
2411         if (unshare_flags & CLONE_VM)
2412                 unshare_flags |= CLONE_SIGHAND;
2413         /*
2414          * If unsharing a signal handlers, must also unshare the signal queues.
2415          */
2416         if (unshare_flags & CLONE_SIGHAND)
2417                 unshare_flags |= CLONE_THREAD;
2418         /*
2419          * If unsharing namespace, must also unshare filesystem information.
2420          */
2421         if (unshare_flags & CLONE_NEWNS)
2422                 unshare_flags |= CLONE_FS;
2423 
2424         err = check_unshare_flags(unshare_flags);
2425         if (err)
2426                 goto bad_unshare_out;
2427         /*
2428          * CLONE_NEWIPC must also detach from the undolist: after switching
2429          * to a new ipc namespace, the semaphore arrays from the old
2430          * namespace are unreachable.
2431          */
2432         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2433                 do_sysvsem = 1;
2434         err = unshare_fs(unshare_flags, &new_fs);
2435         if (err)
2436                 goto bad_unshare_out;
2437         err = unshare_fd(unshare_flags, &new_fd);
2438         if (err)
2439                 goto bad_unshare_cleanup_fs;
2440         err = unshare_userns(unshare_flags, &new_cred);
2441         if (err)
2442                 goto bad_unshare_cleanup_fd;
2443         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2444                                          new_cred, new_fs);
2445         if (err)
2446                 goto bad_unshare_cleanup_cred;
2447 
2448         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2449                 if (do_sysvsem) {
2450                         /*
2451                          * CLONE_SYSVSEM is equivalent to sys_exit().
2452                          */
2453                         exit_sem(current);
2454                 }
2455                 if (unshare_flags & CLONE_NEWIPC) {
2456                         /* Orphan segments in old ns (see sem above). */
2457                         exit_shm(current);
2458                         shm_init_task(current);
2459                 }
2460 
2461                 if (new_nsproxy)
2462                         switch_task_namespaces(current, new_nsproxy);
2463 
2464                 task_lock(current);
2465 
2466                 if (new_fs) {
2467                         fs = current->fs;
2468                         spin_lock(&fs->lock);
2469                         current->fs = new_fs;
2470                         if (--fs->users)
2471                                 new_fs = NULL;
2472                         else
2473                                 new_fs = fs;
2474                         spin_unlock(&fs->lock);
2475                 }
2476 
2477                 if (new_fd) {
2478                         fd = current->files;
2479                         current->files = new_fd;
2480                         new_fd = fd;
2481                 }
2482 
2483                 task_unlock(current);
2484 
2485                 if (new_cred) {
2486                         /* Install the new user namespace */
2487                         commit_creds(new_cred);
2488                         new_cred = NULL;
2489                 }
2490         }
2491 
2492         perf_event_namespaces(current);
2493 
2494 bad_unshare_cleanup_cred:
2495         if (new_cred)
2496                 put_cred(new_cred);
2497 bad_unshare_cleanup_fd:
2498         if (new_fd)
2499                 put_files_struct(new_fd);
2500 
2501 bad_unshare_cleanup_fs:
2502         if (new_fs)
2503                 free_fs_struct(new_fs);
2504 
2505 bad_unshare_out:
2506         return err;
2507 }
2508 
2509 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2510 {
2511         return ksys_unshare(unshare_flags);
2512 }
2513 
2514 /*
2515  *      Helper to unshare the files of the current task.
2516  *      We don't want to expose copy_files internals to
2517  *      the exec layer of the kernel.
2518  */
2519 
2520 int unshare_files(struct files_struct **displaced)
2521 {
2522         struct task_struct *task = current;
2523         struct files_struct *copy = NULL;
2524         int error;
2525 
2526         error = unshare_fd(CLONE_FILES, &copy);
2527         if (error || !copy) {
2528                 *displaced = NULL;
2529                 return error;
2530         }
2531         *displaced = task->files;
2532         task_lock(task);
2533         task->files = copy;
2534         task_unlock(task);
2535         return 0;
2536 }
2537 
2538 int sysctl_max_threads(struct ctl_table *table, int write,
2539                        void __user *buffer, size_t *lenp, loff_t *ppos)
2540 {
2541         struct ctl_table t;
2542         int ret;
2543         int threads = max_threads;
2544         int min = MIN_THREADS;
2545         int max = MAX_THREADS;
2546 
2547         t = *table;
2548         t.data = &threads;
2549         t.extra1 = &min;
2550         t.extra2 = &max;
2551 
2552         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2553         if (ret || !write)
2554                 return ret;
2555 
2556         set_max_threads(threads);
2557 
2558         return 0;
2559 }
2560 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp