]>
Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * Generic pidhash and scalable, time-bounded PID allocator | |
4 | * | |
6d49e352 NYC |
5 | * (C) 2002-2003 Nadia Yvette Chambers, IBM |
6 | * (C) 2004 Nadia Yvette Chambers, Oracle | |
1da177e4 LT |
7 | * (C) 2002-2004 Ingo Molnar, Red Hat |
8 | * | |
9 | * pid-structures are backing objects for tasks sharing a given ID to chain | |
10 | * against. There is very little to them aside from hashing them and | |
11 | * parking tasks using given ID's on a list. | |
12 | * | |
13 | * The hash is always changed with the tasklist_lock write-acquired, | |
14 | * and the hash is only accessed with the tasklist_lock at least | |
15 | * read-acquired, so there's no additional SMP locking needed here. | |
16 | * | |
17 | * We have a list of bitmap pages, which bitmaps represent the PID space. | |
18 | * Allocating and freeing PIDs is completely lockless. The worst-case | |
19 | * allocation scenario when all but one out of 1 million PIDs possible are | |
20 | * allocated already: the scanning of 32 list entries and at most PAGE_SIZE | |
21 | * bytes. The typical fastpath is a single successful setbit. Freeing is O(1). | |
30e49c26 PE |
22 | * |
23 | * Pid namespaces: | |
24 | * (C) 2007 Pavel Emelyanov <xemul@openvz.org>, OpenVZ, SWsoft Inc. | |
25 | * (C) 2007 Sukadev Bhattiprolu <sukadev@us.ibm.com>, IBM | |
26 | * Many thanks to Oleg Nesterov for comments and help | |
27 | * | |
1da177e4 LT |
28 | */ |
29 | ||
30 | #include <linux/mm.h> | |
9984de1a | 31 | #include <linux/export.h> |
1da177e4 LT |
32 | #include <linux/slab.h> |
33 | #include <linux/init.h> | |
82524746 | 34 | #include <linux/rculist.h> |
57c8a661 | 35 | #include <linux/memblock.h> |
61a58c6c | 36 | #include <linux/pid_namespace.h> |
820e45db | 37 | #include <linux/init_task.h> |
3eb07c8c | 38 | #include <linux/syscalls.h> |
0bb80f24 | 39 | #include <linux/proc_ns.h> |
f57e515a | 40 | #include <linux/refcount.h> |
32fcb426 CB |
41 | #include <linux/anon_inodes.h> |
42 | #include <linux/sched/signal.h> | |
29930025 | 43 | #include <linux/sched/task.h> |
95846ecf | 44 | #include <linux/idr.h> |
4969f8a0 | 45 | #include <net/sock.h> |
1da177e4 | 46 | |
e1e871af | 47 | struct pid init_struct_pid = { |
f57e515a | 48 | .count = REFCOUNT_INIT(1), |
e1e871af DH |
49 | .tasks = { |
50 | { .first = NULL }, | |
51 | { .first = NULL }, | |
52 | { .first = NULL }, | |
53 | }, | |
54 | .level = 0, | |
55 | .numbers = { { | |
56 | .nr = 0, | |
57 | .ns = &init_pid_ns, | |
58 | }, } | |
59 | }; | |
1da177e4 LT |
60 | |
61 | int pid_max = PID_MAX_DEFAULT; | |
1da177e4 LT |
62 | |
63 | #define RESERVED_PIDS 300 | |
64 | ||
65 | int pid_max_min = RESERVED_PIDS + 1; | |
66 | int pid_max_max = PID_MAX_LIMIT; | |
67 | ||
1da177e4 LT |
68 | /* |
69 | * PID-map pages start out as NULL, they get allocated upon | |
70 | * first use and are never deallocated. This way a low pid_max | |
71 | * value does not cause lots of bitmaps to be allocated, but | |
72 | * the scheme scales to up to 4 million PIDs, runtime. | |
73 | */ | |
61a58c6c | 74 | struct pid_namespace init_pid_ns = { |
1e24edca | 75 | .kref = KREF_INIT(2), |
f6bb2a2c | 76 | .idr = IDR_INIT(init_pid_ns.idr), |
e8cfbc24 | 77 | .pid_allocated = PIDNS_ADDING, |
faacbfd3 PE |
78 | .level = 0, |
79 | .child_reaper = &init_task, | |
49f4d8b9 | 80 | .user_ns = &init_user_ns, |
435d5f4b | 81 | .ns.inum = PROC_PID_INIT_INO, |
33c42940 AV |
82 | #ifdef CONFIG_PID_NS |
83 | .ns.ops = &pidns_operations, | |
84 | #endif | |
3fbc9648 | 85 | }; |
198fe21b | 86 | EXPORT_SYMBOL_GPL(init_pid_ns); |
1da177e4 | 87 | |
92476d7f EB |
88 | /* |
89 | * Note: disable interrupts while the pidmap_lock is held as an | |
90 | * interrupt might come in and do read_lock(&tasklist_lock). | |
91 | * | |
92 | * If we don't disable interrupts there is a nasty deadlock between | |
93 | * detach_pid()->free_pid() and another cpu that does | |
94 | * spin_lock(&pidmap_lock) followed by an interrupt routine that does | |
95 | * read_lock(&tasklist_lock); | |
96 | * | |
97 | * After we clean up the tasklist_lock and know there are no | |
98 | * irq handlers that take it we can leave the interrupts enabled. | |
99 | * For now it is easier to be safe than to prove it can't happen. | |
100 | */ | |
3fbc9648 | 101 | |
1da177e4 LT |
102 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(pidmap_lock); |
103 | ||
7ad5b3a5 | 104 | void put_pid(struct pid *pid) |
92476d7f | 105 | { |
baf8f0f8 PE |
106 | struct pid_namespace *ns; |
107 | ||
92476d7f EB |
108 | if (!pid) |
109 | return; | |
baf8f0f8 | 110 | |
8ef047aa | 111 | ns = pid->numbers[pid->level].ns; |
f57e515a | 112 | if (refcount_dec_and_test(&pid->count)) { |
baf8f0f8 | 113 | kmem_cache_free(ns->pid_cachep, pid); |
b461cc03 | 114 | put_pid_ns(ns); |
8ef047aa | 115 | } |
92476d7f | 116 | } |
bbf73147 | 117 | EXPORT_SYMBOL_GPL(put_pid); |
92476d7f EB |
118 | |
119 | static void delayed_put_pid(struct rcu_head *rhp) | |
120 | { | |
121 | struct pid *pid = container_of(rhp, struct pid, rcu); | |
122 | put_pid(pid); | |
123 | } | |
124 | ||
7ad5b3a5 | 125 | void free_pid(struct pid *pid) |
92476d7f EB |
126 | { |
127 | /* We can be called with write_lock_irq(&tasklist_lock) held */ | |
8ef047aa | 128 | int i; |
92476d7f EB |
129 | unsigned long flags; |
130 | ||
131 | spin_lock_irqsave(&pidmap_lock, flags); | |
0a01f2cc EB |
132 | for (i = 0; i <= pid->level; i++) { |
133 | struct upid *upid = pid->numbers + i; | |
af4b8a83 | 134 | struct pid_namespace *ns = upid->ns; |
e8cfbc24 | 135 | switch (--ns->pid_allocated) { |
a6064885 | 136 | case 2: |
af4b8a83 EB |
137 | case 1: |
138 | /* When all that is left in the pid namespace | |
139 | * is the reaper wake up the reaper. The reaper | |
140 | * may be sleeping in zap_pid_ns_processes(). | |
141 | */ | |
142 | wake_up_process(ns->child_reaper); | |
143 | break; | |
e8cfbc24 | 144 | case PIDNS_ADDING: |
314a8ad0 ON |
145 | /* Handle a fork failure of the first process */ |
146 | WARN_ON(ns->child_reaper); | |
e8cfbc24 | 147 | ns->pid_allocated = 0; |
af4b8a83 | 148 | break; |
5e1182de | 149 | } |
95846ecf GS |
150 | |
151 | idr_remove(&ns->idr, upid->nr); | |
0a01f2cc | 152 | } |
92476d7f EB |
153 | spin_unlock_irqrestore(&pidmap_lock, flags); |
154 | ||
92476d7f EB |
155 | call_rcu(&pid->rcu, delayed_put_pid); |
156 | } | |
157 | ||
49cb2fc4 AR |
158 | struct pid *alloc_pid(struct pid_namespace *ns, pid_t *set_tid, |
159 | size_t set_tid_size) | |
92476d7f EB |
160 | { |
161 | struct pid *pid; | |
162 | enum pid_type type; | |
8ef047aa PE |
163 | int i, nr; |
164 | struct pid_namespace *tmp; | |
198fe21b | 165 | struct upid *upid; |
35f71bc0 | 166 | int retval = -ENOMEM; |
92476d7f | 167 | |
49cb2fc4 AR |
168 | /* |
169 | * set_tid_size contains the size of the set_tid array. Starting at | |
170 | * the most nested currently active PID namespace it tells alloc_pid() | |
171 | * which PID to set for a process in that most nested PID namespace | |
172 | * up to set_tid_size PID namespaces. It does not have to set the PID | |
173 | * for a process in all nested PID namespaces but set_tid_size must | |
174 | * never be greater than the current ns->level + 1. | |
175 | */ | |
176 | if (set_tid_size > ns->level + 1) | |
177 | return ERR_PTR(-EINVAL); | |
178 | ||
baf8f0f8 | 179 | pid = kmem_cache_alloc(ns->pid_cachep, GFP_KERNEL); |
92476d7f | 180 | if (!pid) |
35f71bc0 | 181 | return ERR_PTR(retval); |
92476d7f | 182 | |
8ef047aa | 183 | tmp = ns; |
0a01f2cc | 184 | pid->level = ns->level; |
95846ecf | 185 | |
8ef047aa | 186 | for (i = ns->level; i >= 0; i--) { |
49cb2fc4 AR |
187 | int tid = 0; |
188 | ||
189 | if (set_tid_size) { | |
190 | tid = set_tid[ns->level - i]; | |
191 | ||
192 | retval = -EINVAL; | |
193 | if (tid < 1 || tid >= pid_max) | |
194 | goto out_free; | |
195 | /* | |
196 | * Also fail if a PID != 1 is requested and | |
197 | * no PID 1 exists. | |
198 | */ | |
199 | if (tid != 1 && !tmp->child_reaper) | |
200 | goto out_free; | |
201 | retval = -EPERM; | |
1caef81d | 202 | if (!checkpoint_restore_ns_capable(tmp->user_ns)) |
49cb2fc4 AR |
203 | goto out_free; |
204 | set_tid_size--; | |
205 | } | |
95846ecf GS |
206 | |
207 | idr_preload(GFP_KERNEL); | |
208 | spin_lock_irq(&pidmap_lock); | |
209 | ||
49cb2fc4 AR |
210 | if (tid) { |
211 | nr = idr_alloc(&tmp->idr, NULL, tid, | |
212 | tid + 1, GFP_ATOMIC); | |
213 | /* | |
214 | * If ENOSPC is returned it means that the PID is | |
215 | * alreay in use. Return EEXIST in that case. | |
216 | */ | |
217 | if (nr == -ENOSPC) | |
218 | nr = -EEXIST; | |
219 | } else { | |
220 | int pid_min = 1; | |
221 | /* | |
222 | * init really needs pid 1, but after reaching the | |
223 | * maximum wrap back to RESERVED_PIDS | |
224 | */ | |
225 | if (idr_get_cursor(&tmp->idr) > RESERVED_PIDS) | |
226 | pid_min = RESERVED_PIDS; | |
227 | ||
228 | /* | |
229 | * Store a null pointer so find_pid_ns does not find | |
230 | * a partially initialized PID (see below). | |
231 | */ | |
232 | nr = idr_alloc_cyclic(&tmp->idr, NULL, pid_min, | |
233 | pid_max, GFP_ATOMIC); | |
234 | } | |
95846ecf GS |
235 | spin_unlock_irq(&pidmap_lock); |
236 | idr_preload_end(); | |
237 | ||
287980e4 | 238 | if (nr < 0) { |
f83606f5 | 239 | retval = (nr == -ENOSPC) ? -EAGAIN : nr; |
8ef047aa | 240 | goto out_free; |
35f71bc0 | 241 | } |
92476d7f | 242 | |
8ef047aa PE |
243 | pid->numbers[i].nr = nr; |
244 | pid->numbers[i].ns = tmp; | |
245 | tmp = tmp->parent; | |
246 | } | |
247 | ||
10dab84c CB |
248 | /* |
249 | * ENOMEM is not the most obvious choice especially for the case | |
250 | * where the child subreaper has already exited and the pid | |
251 | * namespace denies the creation of any new processes. But ENOMEM | |
252 | * is what we have exposed to userspace for a long time and it is | |
253 | * documented behavior for pid namespaces. So we can't easily | |
254 | * change it even if there were an error code better suited. | |
255 | */ | |
b26ebfe1 CM |
256 | retval = -ENOMEM; |
257 | ||
b461cc03 | 258 | get_pid_ns(ns); |
f57e515a | 259 | refcount_set(&pid->count, 1); |
63f818f4 | 260 | spin_lock_init(&pid->lock); |
92476d7f EB |
261 | for (type = 0; type < PIDTYPE_MAX; ++type) |
262 | INIT_HLIST_HEAD(&pid->tasks[type]); | |
263 | ||
b53b0b9d | 264 | init_waitqueue_head(&pid->wait_pidfd); |
7bc3e6e5 | 265 | INIT_HLIST_HEAD(&pid->inodes); |
b53b0b9d | 266 | |
417e3152 | 267 | upid = pid->numbers + ns->level; |
92476d7f | 268 | spin_lock_irq(&pidmap_lock); |
e8cfbc24 | 269 | if (!(ns->pid_allocated & PIDNS_ADDING)) |
5e1182de | 270 | goto out_unlock; |
0a01f2cc | 271 | for ( ; upid >= pid->numbers; --upid) { |
95846ecf GS |
272 | /* Make the PID visible to find_pid_ns. */ |
273 | idr_replace(&upid->ns->idr, pid, upid->nr); | |
e8cfbc24 | 274 | upid->ns->pid_allocated++; |
0a01f2cc | 275 | } |
92476d7f EB |
276 | spin_unlock_irq(&pidmap_lock); |
277 | ||
92476d7f EB |
278 | return pid; |
279 | ||
5e1182de | 280 | out_unlock: |
6e666884 | 281 | spin_unlock_irq(&pidmap_lock); |
24c037eb ON |
282 | put_pid_ns(ns); |
283 | ||
92476d7f | 284 | out_free: |
95846ecf | 285 | spin_lock_irq(&pidmap_lock); |
1a80dade MW |
286 | while (++i <= ns->level) { |
287 | upid = pid->numbers + i; | |
288 | idr_remove(&upid->ns->idr, upid->nr); | |
289 | } | |
95846ecf | 290 | |
c0ee5549 EB |
291 | /* On failure to allocate the first pid, reset the state */ |
292 | if (ns->pid_allocated == PIDNS_ADDING) | |
293 | idr_set_cursor(&ns->idr, 0); | |
294 | ||
95846ecf | 295 | spin_unlock_irq(&pidmap_lock); |
8ef047aa | 296 | |
baf8f0f8 | 297 | kmem_cache_free(ns->pid_cachep, pid); |
35f71bc0 | 298 | return ERR_PTR(retval); |
92476d7f EB |
299 | } |
300 | ||
c876ad76 EB |
301 | void disable_pid_allocation(struct pid_namespace *ns) |
302 | { | |
303 | spin_lock_irq(&pidmap_lock); | |
e8cfbc24 | 304 | ns->pid_allocated &= ~PIDNS_ADDING; |
c876ad76 EB |
305 | spin_unlock_irq(&pidmap_lock); |
306 | } | |
307 | ||
7ad5b3a5 | 308 | struct pid *find_pid_ns(int nr, struct pid_namespace *ns) |
1da177e4 | 309 | { |
e8cfbc24 | 310 | return idr_find(&ns->idr, nr); |
1da177e4 | 311 | } |
198fe21b | 312 | EXPORT_SYMBOL_GPL(find_pid_ns); |
1da177e4 | 313 | |
8990571e PE |
314 | struct pid *find_vpid(int nr) |
315 | { | |
17cf22c3 | 316 | return find_pid_ns(nr, task_active_pid_ns(current)); |
8990571e PE |
317 | } |
318 | EXPORT_SYMBOL_GPL(find_vpid); | |
319 | ||
2c470475 EB |
320 | static struct pid **task_pid_ptr(struct task_struct *task, enum pid_type type) |
321 | { | |
322 | return (type == PIDTYPE_PID) ? | |
323 | &task->thread_pid : | |
2c470475 EB |
324 | &task->signal->pids[type]; |
325 | } | |
326 | ||
e713d0da SB |
327 | /* |
328 | * attach_pid() must be called with the tasklist_lock write-held. | |
329 | */ | |
81907739 | 330 | void attach_pid(struct task_struct *task, enum pid_type type) |
1da177e4 | 331 | { |
2c470475 EB |
332 | struct pid *pid = *task_pid_ptr(task, type); |
333 | hlist_add_head_rcu(&task->pid_links[type], &pid->tasks[type]); | |
1da177e4 LT |
334 | } |
335 | ||
24336eae ON |
336 | static void __change_pid(struct task_struct *task, enum pid_type type, |
337 | struct pid *new) | |
1da177e4 | 338 | { |
2c470475 | 339 | struct pid **pid_ptr = task_pid_ptr(task, type); |
92476d7f EB |
340 | struct pid *pid; |
341 | int tmp; | |
1da177e4 | 342 | |
2c470475 | 343 | pid = *pid_ptr; |
1da177e4 | 344 | |
2c470475 EB |
345 | hlist_del_rcu(&task->pid_links[type]); |
346 | *pid_ptr = new; | |
1da177e4 | 347 | |
92476d7f | 348 | for (tmp = PIDTYPE_MAX; --tmp >= 0; ) |
1d416a11 | 349 | if (pid_has_task(pid, tmp)) |
92476d7f | 350 | return; |
1da177e4 | 351 | |
92476d7f | 352 | free_pid(pid); |
1da177e4 LT |
353 | } |
354 | ||
24336eae ON |
355 | void detach_pid(struct task_struct *task, enum pid_type type) |
356 | { | |
357 | __change_pid(task, type, NULL); | |
358 | } | |
359 | ||
360 | void change_pid(struct task_struct *task, enum pid_type type, | |
361 | struct pid *pid) | |
362 | { | |
363 | __change_pid(task, type, pid); | |
81907739 | 364 | attach_pid(task, type); |
24336eae ON |
365 | } |
366 | ||
6b03d130 EB |
367 | void exchange_tids(struct task_struct *left, struct task_struct *right) |
368 | { | |
369 | struct pid *pid1 = left->thread_pid; | |
370 | struct pid *pid2 = right->thread_pid; | |
371 | struct hlist_head *head1 = &pid1->tasks[PIDTYPE_PID]; | |
372 | struct hlist_head *head2 = &pid2->tasks[PIDTYPE_PID]; | |
373 | ||
374 | /* Swap the single entry tid lists */ | |
375 | hlists_swap_heads_rcu(head1, head2); | |
376 | ||
377 | /* Swap the per task_struct pid */ | |
378 | rcu_assign_pointer(left->thread_pid, pid2); | |
379 | rcu_assign_pointer(right->thread_pid, pid1); | |
380 | ||
381 | /* Swap the cached value */ | |
382 | WRITE_ONCE(left->pid, pid_nr(pid2)); | |
383 | WRITE_ONCE(right->pid, pid_nr(pid1)); | |
384 | } | |
385 | ||
c18258c6 | 386 | /* transfer_pid is an optimization of attach_pid(new), detach_pid(old) */ |
7ad5b3a5 | 387 | void transfer_pid(struct task_struct *old, struct task_struct *new, |
c18258c6 EB |
388 | enum pid_type type) |
389 | { | |
2c470475 EB |
390 | if (type == PIDTYPE_PID) |
391 | new->thread_pid = old->thread_pid; | |
392 | hlist_replace_rcu(&old->pid_links[type], &new->pid_links[type]); | |
c18258c6 EB |
393 | } |
394 | ||
7ad5b3a5 | 395 | struct task_struct *pid_task(struct pid *pid, enum pid_type type) |
1da177e4 | 396 | { |
92476d7f EB |
397 | struct task_struct *result = NULL; |
398 | if (pid) { | |
399 | struct hlist_node *first; | |
67bdbffd | 400 | first = rcu_dereference_check(hlist_first_rcu(&pid->tasks[type]), |
db1466b3 | 401 | lockdep_tasklist_lock_is_held()); |
92476d7f | 402 | if (first) |
2c470475 | 403 | result = hlist_entry(first, struct task_struct, pid_links[(type)]); |
92476d7f EB |
404 | } |
405 | return result; | |
406 | } | |
eccba068 | 407 | EXPORT_SYMBOL(pid_task); |
1da177e4 | 408 | |
92476d7f | 409 | /* |
9728e5d6 | 410 | * Must be called under rcu_read_lock(). |
92476d7f | 411 | */ |
17f98dcf | 412 | struct task_struct *find_task_by_pid_ns(pid_t nr, struct pid_namespace *ns) |
92476d7f | 413 | { |
f78f5b90 PM |
414 | RCU_LOCKDEP_WARN(!rcu_read_lock_held(), |
415 | "find_task_by_pid_ns() needs rcu_read_lock() protection"); | |
17f98dcf | 416 | return pid_task(find_pid_ns(nr, ns), PIDTYPE_PID); |
92476d7f | 417 | } |
1da177e4 | 418 | |
228ebcbe PE |
419 | struct task_struct *find_task_by_vpid(pid_t vnr) |
420 | { | |
17cf22c3 | 421 | return find_task_by_pid_ns(vnr, task_active_pid_ns(current)); |
228ebcbe | 422 | } |
228ebcbe | 423 | |
2ee08260 MR |
424 | struct task_struct *find_get_task_by_vpid(pid_t nr) |
425 | { | |
426 | struct task_struct *task; | |
427 | ||
428 | rcu_read_lock(); | |
429 | task = find_task_by_vpid(nr); | |
430 | if (task) | |
431 | get_task_struct(task); | |
432 | rcu_read_unlock(); | |
433 | ||
434 | return task; | |
435 | } | |
436 | ||
1a657f78 ON |
437 | struct pid *get_task_pid(struct task_struct *task, enum pid_type type) |
438 | { | |
439 | struct pid *pid; | |
440 | rcu_read_lock(); | |
2c470475 | 441 | pid = get_pid(rcu_dereference(*task_pid_ptr(task, type))); |
1a657f78 ON |
442 | rcu_read_unlock(); |
443 | return pid; | |
444 | } | |
77c100c8 | 445 | EXPORT_SYMBOL_GPL(get_task_pid); |
1a657f78 | 446 | |
7ad5b3a5 | 447 | struct task_struct *get_pid_task(struct pid *pid, enum pid_type type) |
92476d7f EB |
448 | { |
449 | struct task_struct *result; | |
450 | rcu_read_lock(); | |
451 | result = pid_task(pid, type); | |
452 | if (result) | |
453 | get_task_struct(result); | |
454 | rcu_read_unlock(); | |
455 | return result; | |
1da177e4 | 456 | } |
77c100c8 | 457 | EXPORT_SYMBOL_GPL(get_pid_task); |
1da177e4 | 458 | |
92476d7f | 459 | struct pid *find_get_pid(pid_t nr) |
1da177e4 LT |
460 | { |
461 | struct pid *pid; | |
462 | ||
92476d7f | 463 | rcu_read_lock(); |
198fe21b | 464 | pid = get_pid(find_vpid(nr)); |
92476d7f | 465 | rcu_read_unlock(); |
1da177e4 | 466 | |
92476d7f | 467 | return pid; |
1da177e4 | 468 | } |
339caf2a | 469 | EXPORT_SYMBOL_GPL(find_get_pid); |
1da177e4 | 470 | |
7af57294 PE |
471 | pid_t pid_nr_ns(struct pid *pid, struct pid_namespace *ns) |
472 | { | |
473 | struct upid *upid; | |
474 | pid_t nr = 0; | |
475 | ||
476 | if (pid && ns->level <= pid->level) { | |
477 | upid = &pid->numbers[ns->level]; | |
478 | if (upid->ns == ns) | |
479 | nr = upid->nr; | |
480 | } | |
481 | return nr; | |
482 | } | |
4f82f457 | 483 | EXPORT_SYMBOL_GPL(pid_nr_ns); |
7af57294 | 484 | |
44c4e1b2 EB |
485 | pid_t pid_vnr(struct pid *pid) |
486 | { | |
17cf22c3 | 487 | return pid_nr_ns(pid, task_active_pid_ns(current)); |
44c4e1b2 EB |
488 | } |
489 | EXPORT_SYMBOL_GPL(pid_vnr); | |
490 | ||
52ee2dfd ON |
491 | pid_t __task_pid_nr_ns(struct task_struct *task, enum pid_type type, |
492 | struct pid_namespace *ns) | |
2f2a3a46 | 493 | { |
52ee2dfd ON |
494 | pid_t nr = 0; |
495 | ||
496 | rcu_read_lock(); | |
497 | if (!ns) | |
17cf22c3 | 498 | ns = task_active_pid_ns(current); |
1dd694a1 | 499 | nr = pid_nr_ns(rcu_dereference(*task_pid_ptr(task, type)), ns); |
52ee2dfd ON |
500 | rcu_read_unlock(); |
501 | ||
502 | return nr; | |
2f2a3a46 | 503 | } |
52ee2dfd | 504 | EXPORT_SYMBOL(__task_pid_nr_ns); |
2f2a3a46 | 505 | |
61bce0f1 EB |
506 | struct pid_namespace *task_active_pid_ns(struct task_struct *tsk) |
507 | { | |
508 | return ns_of_pid(task_pid(tsk)); | |
509 | } | |
510 | EXPORT_SYMBOL_GPL(task_active_pid_ns); | |
511 | ||
0804ef4b | 512 | /* |
025dfdaf | 513 | * Used by proc to find the first pid that is greater than or equal to nr. |
0804ef4b | 514 | * |
e49859e7 | 515 | * If there is a pid at nr this function is exactly the same as find_pid_ns. |
0804ef4b | 516 | */ |
198fe21b | 517 | struct pid *find_ge_pid(int nr, struct pid_namespace *ns) |
0804ef4b | 518 | { |
95846ecf | 519 | return idr_get_next(&ns->idr, &nr); |
0804ef4b EB |
520 | } |
521 | ||
32fcb426 CB |
522 | /** |
523 | * pidfd_create() - Create a new pid file descriptor. | |
524 | * | |
525 | * @pid: struct pid that the pidfd will reference | |
526 | * | |
527 | * This creates a new pid file descriptor with the O_CLOEXEC flag set. | |
528 | * | |
529 | * Note, that this function can only be called after the fd table has | |
530 | * been unshared to avoid leaking the pidfd to the new process. | |
531 | * | |
532 | * Return: On success, a cloexec pidfd is returned. | |
533 | * On error, a negative errno number will be returned. | |
534 | */ | |
535 | static int pidfd_create(struct pid *pid) | |
536 | { | |
537 | int fd; | |
538 | ||
539 | fd = anon_inode_getfd("[pidfd]", &pidfd_fops, get_pid(pid), | |
540 | O_RDWR | O_CLOEXEC); | |
541 | if (fd < 0) | |
542 | put_pid(pid); | |
543 | ||
544 | return fd; | |
545 | } | |
546 | ||
547 | /** | |
548 | * pidfd_open() - Open new pid file descriptor. | |
549 | * | |
550 | * @pid: pid for which to retrieve a pidfd | |
551 | * @flags: flags to pass | |
552 | * | |
553 | * This creates a new pid file descriptor with the O_CLOEXEC flag set for | |
554 | * the process identified by @pid. Currently, the process identified by | |
555 | * @pid must be a thread-group leader. This restriction currently exists | |
556 | * for all aspects of pidfds including pidfd creation (CLONE_PIDFD cannot | |
557 | * be used with CLONE_THREAD) and pidfd polling (only supports thread group | |
558 | * leaders). | |
559 | * | |
560 | * Return: On success, a cloexec pidfd is returned. | |
561 | * On error, a negative errno number will be returned. | |
562 | */ | |
563 | SYSCALL_DEFINE2(pidfd_open, pid_t, pid, unsigned int, flags) | |
564 | { | |
1e1d0f0b | 565 | int fd; |
32fcb426 CB |
566 | struct pid *p; |
567 | ||
568 | if (flags) | |
569 | return -EINVAL; | |
570 | ||
571 | if (pid <= 0) | |
572 | return -EINVAL; | |
573 | ||
574 | p = find_get_pid(pid); | |
575 | if (!p) | |
576 | return -ESRCH; | |
577 | ||
1e1d0f0b CB |
578 | if (pid_has_task(p, PIDTYPE_TGID)) |
579 | fd = pidfd_create(p); | |
580 | else | |
581 | fd = -EINVAL; | |
32fcb426 | 582 | |
32fcb426 CB |
583 | put_pid(p); |
584 | return fd; | |
585 | } | |
586 | ||
95846ecf | 587 | void __init pid_idr_init(void) |
1da177e4 | 588 | { |
840d6fe7 | 589 | /* Verify no one has done anything silly: */ |
e8cfbc24 | 590 | BUILD_BUG_ON(PID_MAX_LIMIT >= PIDNS_ADDING); |
c876ad76 | 591 | |
72680a19 HB |
592 | /* bump default and minimum pid_max based on number of cpus */ |
593 | pid_max = min(pid_max_max, max_t(int, pid_max, | |
594 | PIDS_PER_CPU_DEFAULT * num_possible_cpus())); | |
595 | pid_max_min = max_t(int, pid_max_min, | |
596 | PIDS_PER_CPU_MIN * num_possible_cpus()); | |
597 | pr_info("pid_max: default: %u minimum: %u\n", pid_max, pid_max_min); | |
598 | ||
95846ecf | 599 | idr_init(&init_pid_ns.idr); |
92476d7f | 600 | |
74bd59bb | 601 | init_pid_ns.pid_cachep = KMEM_CACHE(pid, |
5d097056 | 602 | SLAB_HWCACHE_ALIGN | SLAB_PANIC | SLAB_ACCOUNT); |
1da177e4 | 603 | } |
8649c322 SD |
604 | |
605 | static struct file *__pidfd_fget(struct task_struct *task, int fd) | |
606 | { | |
607 | struct file *file; | |
608 | int ret; | |
609 | ||
501f9328 | 610 | ret = mutex_lock_killable(&task->signal->exec_update_mutex); |
8649c322 SD |
611 | if (ret) |
612 | return ERR_PTR(ret); | |
613 | ||
614 | if (ptrace_may_access(task, PTRACE_MODE_ATTACH_REALCREDS)) | |
615 | file = fget_task(task, fd); | |
616 | else | |
617 | file = ERR_PTR(-EPERM); | |
618 | ||
501f9328 | 619 | mutex_unlock(&task->signal->exec_update_mutex); |
8649c322 SD |
620 | |
621 | return file ?: ERR_PTR(-EBADF); | |
622 | } | |
623 | ||
624 | static int pidfd_getfd(struct pid *pid, int fd) | |
625 | { | |
626 | struct task_struct *task; | |
627 | struct file *file; | |
628 | int ret; | |
629 | ||
630 | task = get_pid_task(pid, PIDTYPE_PID); | |
631 | if (!task) | |
632 | return -ESRCH; | |
633 | ||
634 | file = __pidfd_fget(task, fd); | |
635 | put_task_struct(task); | |
636 | if (IS_ERR(file)) | |
637 | return PTR_ERR(file); | |
638 | ||
910d2f16 KC |
639 | ret = receive_fd(file, O_CLOEXEC); |
640 | fput(file); | |
8649c322 SD |
641 | |
642 | return ret; | |
643 | } | |
644 | ||
645 | /** | |
646 | * sys_pidfd_getfd() - Get a file descriptor from another process | |
647 | * | |
648 | * @pidfd: the pidfd file descriptor of the process | |
649 | * @fd: the file descriptor number to get | |
650 | * @flags: flags on how to get the fd (reserved) | |
651 | * | |
652 | * This syscall gets a copy of a file descriptor from another process | |
653 | * based on the pidfd, and file descriptor number. It requires that | |
654 | * the calling process has the ability to ptrace the process represented | |
655 | * by the pidfd. The process which is having its file descriptor copied | |
656 | * is otherwise unaffected. | |
657 | * | |
658 | * Return: On success, a cloexec file descriptor is returned. | |
659 | * On error, a negative errno number will be returned. | |
660 | */ | |
661 | SYSCALL_DEFINE3(pidfd_getfd, int, pidfd, int, fd, | |
662 | unsigned int, flags) | |
663 | { | |
664 | struct pid *pid; | |
665 | struct fd f; | |
666 | int ret; | |
667 | ||
668 | /* flags is currently unused - make sure it's unset */ | |
669 | if (flags) | |
670 | return -EINVAL; | |
671 | ||
672 | f = fdget(pidfd); | |
673 | if (!f.file) | |
674 | return -EBADF; | |
675 | ||
676 | pid = pidfd_pid(f.file); | |
677 | if (IS_ERR(pid)) | |
678 | ret = PTR_ERR(pid); | |
679 | else | |
680 | ret = pidfd_getfd(pid, fd); | |
681 | ||
682 | fdput(f); | |
683 | return ret; | |
684 | } |