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