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