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