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