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> |
cb12fd8e | 45 | #include <linux/pidfs.h> |
4969f8a0 | 46 | #include <net/sock.h> |
6da73d15 | 47 | #include <uapi/linux/pidfd.h> |
1da177e4 | 48 | |
e1e871af | 49 | struct pid init_struct_pid = { |
f57e515a | 50 | .count = REFCOUNT_INIT(1), |
e1e871af DH |
51 | .tasks = { |
52 | { .first = NULL }, | |
53 | { .first = NULL }, | |
54 | { .first = NULL }, | |
55 | }, | |
56 | .level = 0, | |
57 | .numbers = { { | |
58 | .nr = 0, | |
59 | .ns = &init_pid_ns, | |
60 | }, } | |
61 | }; | |
1da177e4 LT |
62 | |
63 | int pid_max = PID_MAX_DEFAULT; | |
1da177e4 | 64 | |
1da177e4 LT |
65 | int pid_max_min = RESERVED_PIDS + 1; |
66 | int pid_max_max = PID_MAX_LIMIT; | |
cb12fd8e CB |
67 | /* |
68 | * Pseudo filesystems start inode numbering after one. We use Reserved | |
69 | * PIDs as a natural offset. | |
70 | */ | |
71 | static u64 pidfs_ino = RESERVED_PIDS; | |
1da177e4 | 72 | |
1da177e4 LT |
73 | /* |
74 | * PID-map pages start out as NULL, they get allocated upon | |
75 | * first use and are never deallocated. This way a low pid_max | |
76 | * value does not cause lots of bitmaps to be allocated, but | |
77 | * the scheme scales to up to 4 million PIDs, runtime. | |
78 | */ | |
61a58c6c | 79 | struct pid_namespace init_pid_ns = { |
8eb71d95 | 80 | .ns.count = REFCOUNT_INIT(2), |
f6bb2a2c | 81 | .idr = IDR_INIT(init_pid_ns.idr), |
e8cfbc24 | 82 | .pid_allocated = PIDNS_ADDING, |
faacbfd3 PE |
83 | .level = 0, |
84 | .child_reaper = &init_task, | |
49f4d8b9 | 85 | .user_ns = &init_user_ns, |
435d5f4b | 86 | .ns.inum = PROC_PID_INIT_INO, |
33c42940 AV |
87 | #ifdef CONFIG_PID_NS |
88 | .ns.ops = &pidns_operations, | |
89 | #endif | |
9876cfe8 AS |
90 | #if defined(CONFIG_SYSCTL) && defined(CONFIG_MEMFD_CREATE) |
91 | .memfd_noexec_scope = MEMFD_NOEXEC_SCOPE_EXEC, | |
92 | #endif | |
3fbc9648 | 93 | }; |
198fe21b | 94 | EXPORT_SYMBOL_GPL(init_pid_ns); |
1da177e4 | 95 | |
92476d7f EB |
96 | /* |
97 | * Note: disable interrupts while the pidmap_lock is held as an | |
98 | * interrupt might come in and do read_lock(&tasklist_lock). | |
99 | * | |
100 | * If we don't disable interrupts there is a nasty deadlock between | |
101 | * detach_pid()->free_pid() and another cpu that does | |
102 | * spin_lock(&pidmap_lock) followed by an interrupt routine that does | |
103 | * read_lock(&tasklist_lock); | |
104 | * | |
105 | * After we clean up the tasklist_lock and know there are no | |
106 | * irq handlers that take it we can leave the interrupts enabled. | |
107 | * For now it is easier to be safe than to prove it can't happen. | |
108 | */ | |
3fbc9648 | 109 | |
1da177e4 LT |
110 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(pidmap_lock); |
111 | ||
7ad5b3a5 | 112 | void put_pid(struct pid *pid) |
92476d7f | 113 | { |
baf8f0f8 PE |
114 | struct pid_namespace *ns; |
115 | ||
92476d7f EB |
116 | if (!pid) |
117 | return; | |
baf8f0f8 | 118 | |
8ef047aa | 119 | ns = pid->numbers[pid->level].ns; |
f57e515a | 120 | if (refcount_dec_and_test(&pid->count)) { |
baf8f0f8 | 121 | kmem_cache_free(ns->pid_cachep, pid); |
b461cc03 | 122 | put_pid_ns(ns); |
8ef047aa | 123 | } |
92476d7f | 124 | } |
bbf73147 | 125 | EXPORT_SYMBOL_GPL(put_pid); |
92476d7f EB |
126 | |
127 | static void delayed_put_pid(struct rcu_head *rhp) | |
128 | { | |
129 | struct pid *pid = container_of(rhp, struct pid, rcu); | |
130 | put_pid(pid); | |
131 | } | |
132 | ||
7ad5b3a5 | 133 | void free_pid(struct pid *pid) |
92476d7f EB |
134 | { |
135 | /* We can be called with write_lock_irq(&tasklist_lock) held */ | |
8ef047aa | 136 | int i; |
92476d7f EB |
137 | unsigned long flags; |
138 | ||
139 | spin_lock_irqsave(&pidmap_lock, flags); | |
0a01f2cc EB |
140 | for (i = 0; i <= pid->level; i++) { |
141 | struct upid *upid = pid->numbers + i; | |
af4b8a83 | 142 | struct pid_namespace *ns = upid->ns; |
e8cfbc24 | 143 | switch (--ns->pid_allocated) { |
a6064885 | 144 | case 2: |
af4b8a83 EB |
145 | case 1: |
146 | /* When all that is left in the pid namespace | |
147 | * is the reaper wake up the reaper. The reaper | |
148 | * may be sleeping in zap_pid_ns_processes(). | |
149 | */ | |
150 | wake_up_process(ns->child_reaper); | |
151 | break; | |
e8cfbc24 | 152 | case PIDNS_ADDING: |
314a8ad0 ON |
153 | /* Handle a fork failure of the first process */ |
154 | WARN_ON(ns->child_reaper); | |
e8cfbc24 | 155 | ns->pid_allocated = 0; |
af4b8a83 | 156 | break; |
5e1182de | 157 | } |
95846ecf GS |
158 | |
159 | idr_remove(&ns->idr, upid->nr); | |
0a01f2cc | 160 | } |
92476d7f EB |
161 | spin_unlock_irqrestore(&pidmap_lock, flags); |
162 | ||
92476d7f EB |
163 | call_rcu(&pid->rcu, delayed_put_pid); |
164 | } | |
165 | ||
49cb2fc4 AR |
166 | struct pid *alloc_pid(struct pid_namespace *ns, pid_t *set_tid, |
167 | size_t set_tid_size) | |
92476d7f EB |
168 | { |
169 | struct pid *pid; | |
170 | enum pid_type type; | |
8ef047aa PE |
171 | int i, nr; |
172 | struct pid_namespace *tmp; | |
198fe21b | 173 | struct upid *upid; |
35f71bc0 | 174 | int retval = -ENOMEM; |
92476d7f | 175 | |
49cb2fc4 AR |
176 | /* |
177 | * set_tid_size contains the size of the set_tid array. Starting at | |
178 | * the most nested currently active PID namespace it tells alloc_pid() | |
179 | * which PID to set for a process in that most nested PID namespace | |
180 | * up to set_tid_size PID namespaces. It does not have to set the PID | |
181 | * for a process in all nested PID namespaces but set_tid_size must | |
182 | * never be greater than the current ns->level + 1. | |
183 | */ | |
184 | if (set_tid_size > ns->level + 1) | |
185 | return ERR_PTR(-EINVAL); | |
186 | ||
baf8f0f8 | 187 | pid = kmem_cache_alloc(ns->pid_cachep, GFP_KERNEL); |
92476d7f | 188 | if (!pid) |
35f71bc0 | 189 | return ERR_PTR(retval); |
92476d7f | 190 | |
8ef047aa | 191 | tmp = ns; |
0a01f2cc | 192 | pid->level = ns->level; |
95846ecf | 193 | |
8ef047aa | 194 | for (i = ns->level; i >= 0; i--) { |
49cb2fc4 AR |
195 | int tid = 0; |
196 | ||
197 | if (set_tid_size) { | |
198 | tid = set_tid[ns->level - i]; | |
199 | ||
200 | retval = -EINVAL; | |
201 | if (tid < 1 || tid >= pid_max) | |
202 | goto out_free; | |
203 | /* | |
204 | * Also fail if a PID != 1 is requested and | |
205 | * no PID 1 exists. | |
206 | */ | |
207 | if (tid != 1 && !tmp->child_reaper) | |
208 | goto out_free; | |
209 | retval = -EPERM; | |
1caef81d | 210 | if (!checkpoint_restore_ns_capable(tmp->user_ns)) |
49cb2fc4 AR |
211 | goto out_free; |
212 | set_tid_size--; | |
213 | } | |
95846ecf GS |
214 | |
215 | idr_preload(GFP_KERNEL); | |
216 | spin_lock_irq(&pidmap_lock); | |
217 | ||
49cb2fc4 AR |
218 | if (tid) { |
219 | nr = idr_alloc(&tmp->idr, NULL, tid, | |
220 | tid + 1, GFP_ATOMIC); | |
221 | /* | |
222 | * If ENOSPC is returned it means that the PID is | |
223 | * alreay in use. Return EEXIST in that case. | |
224 | */ | |
225 | if (nr == -ENOSPC) | |
226 | nr = -EEXIST; | |
227 | } else { | |
228 | int pid_min = 1; | |
229 | /* | |
230 | * init really needs pid 1, but after reaching the | |
231 | * maximum wrap back to RESERVED_PIDS | |
232 | */ | |
233 | if (idr_get_cursor(&tmp->idr) > RESERVED_PIDS) | |
234 | pid_min = RESERVED_PIDS; | |
235 | ||
236 | /* | |
237 | * Store a null pointer so find_pid_ns does not find | |
238 | * a partially initialized PID (see below). | |
239 | */ | |
240 | nr = idr_alloc_cyclic(&tmp->idr, NULL, pid_min, | |
241 | pid_max, GFP_ATOMIC); | |
242 | } | |
95846ecf GS |
243 | spin_unlock_irq(&pidmap_lock); |
244 | idr_preload_end(); | |
245 | ||
287980e4 | 246 | if (nr < 0) { |
f83606f5 | 247 | retval = (nr == -ENOSPC) ? -EAGAIN : nr; |
8ef047aa | 248 | goto out_free; |
35f71bc0 | 249 | } |
92476d7f | 250 | |
8ef047aa PE |
251 | pid->numbers[i].nr = nr; |
252 | pid->numbers[i].ns = tmp; | |
253 | tmp = tmp->parent; | |
254 | } | |
255 | ||
10dab84c CB |
256 | /* |
257 | * ENOMEM is not the most obvious choice especially for the case | |
258 | * where the child subreaper has already exited and the pid | |
259 | * namespace denies the creation of any new processes. But ENOMEM | |
260 | * is what we have exposed to userspace for a long time and it is | |
261 | * documented behavior for pid namespaces. So we can't easily | |
262 | * change it even if there were an error code better suited. | |
263 | */ | |
b26ebfe1 CM |
264 | retval = -ENOMEM; |
265 | ||
b461cc03 | 266 | get_pid_ns(ns); |
f57e515a | 267 | refcount_set(&pid->count, 1); |
63f818f4 | 268 | spin_lock_init(&pid->lock); |
92476d7f EB |
269 | for (type = 0; type < PIDTYPE_MAX; ++type) |
270 | INIT_HLIST_HEAD(&pid->tasks[type]); | |
271 | ||
b53b0b9d | 272 | init_waitqueue_head(&pid->wait_pidfd); |
7bc3e6e5 | 273 | INIT_HLIST_HEAD(&pid->inodes); |
b53b0b9d | 274 | |
417e3152 | 275 | upid = pid->numbers + ns->level; |
92476d7f | 276 | spin_lock_irq(&pidmap_lock); |
e8cfbc24 | 277 | if (!(ns->pid_allocated & PIDNS_ADDING)) |
5e1182de | 278 | goto out_unlock; |
b28ddcc3 | 279 | pid->stashed = NULL; |
cb12fd8e | 280 | pid->ino = ++pidfs_ino; |
0a01f2cc | 281 | for ( ; upid >= pid->numbers; --upid) { |
95846ecf GS |
282 | /* Make the PID visible to find_pid_ns. */ |
283 | idr_replace(&upid->ns->idr, pid, upid->nr); | |
e8cfbc24 | 284 | upid->ns->pid_allocated++; |
0a01f2cc | 285 | } |
92476d7f EB |
286 | spin_unlock_irq(&pidmap_lock); |
287 | ||
92476d7f EB |
288 | return pid; |
289 | ||
5e1182de | 290 | out_unlock: |
6e666884 | 291 | spin_unlock_irq(&pidmap_lock); |
24c037eb ON |
292 | put_pid_ns(ns); |
293 | ||
92476d7f | 294 | out_free: |
95846ecf | 295 | spin_lock_irq(&pidmap_lock); |
1a80dade MW |
296 | while (++i <= ns->level) { |
297 | upid = pid->numbers + i; | |
298 | idr_remove(&upid->ns->idr, upid->nr); | |
299 | } | |
95846ecf | 300 | |
c0ee5549 EB |
301 | /* On failure to allocate the first pid, reset the state */ |
302 | if (ns->pid_allocated == PIDNS_ADDING) | |
303 | idr_set_cursor(&ns->idr, 0); | |
304 | ||
95846ecf | 305 | spin_unlock_irq(&pidmap_lock); |
8ef047aa | 306 | |
baf8f0f8 | 307 | kmem_cache_free(ns->pid_cachep, pid); |
35f71bc0 | 308 | return ERR_PTR(retval); |
92476d7f EB |
309 | } |
310 | ||
c876ad76 EB |
311 | void disable_pid_allocation(struct pid_namespace *ns) |
312 | { | |
313 | spin_lock_irq(&pidmap_lock); | |
e8cfbc24 | 314 | ns->pid_allocated &= ~PIDNS_ADDING; |
c876ad76 EB |
315 | spin_unlock_irq(&pidmap_lock); |
316 | } | |
317 | ||
7ad5b3a5 | 318 | struct pid *find_pid_ns(int nr, struct pid_namespace *ns) |
1da177e4 | 319 | { |
e8cfbc24 | 320 | return idr_find(&ns->idr, nr); |
1da177e4 | 321 | } |
198fe21b | 322 | EXPORT_SYMBOL_GPL(find_pid_ns); |
1da177e4 | 323 | |
8990571e PE |
324 | struct pid *find_vpid(int nr) |
325 | { | |
17cf22c3 | 326 | return find_pid_ns(nr, task_active_pid_ns(current)); |
8990571e PE |
327 | } |
328 | EXPORT_SYMBOL_GPL(find_vpid); | |
329 | ||
2c470475 EB |
330 | static struct pid **task_pid_ptr(struct task_struct *task, enum pid_type type) |
331 | { | |
332 | return (type == PIDTYPE_PID) ? | |
333 | &task->thread_pid : | |
2c470475 EB |
334 | &task->signal->pids[type]; |
335 | } | |
336 | ||
e713d0da SB |
337 | /* |
338 | * attach_pid() must be called with the tasklist_lock write-held. | |
339 | */ | |
81907739 | 340 | void attach_pid(struct task_struct *task, enum pid_type type) |
1da177e4 | 341 | { |
2c470475 EB |
342 | struct pid *pid = *task_pid_ptr(task, type); |
343 | hlist_add_head_rcu(&task->pid_links[type], &pid->tasks[type]); | |
1da177e4 LT |
344 | } |
345 | ||
24336eae ON |
346 | static void __change_pid(struct task_struct *task, enum pid_type type, |
347 | struct pid *new) | |
1da177e4 | 348 | { |
2c470475 | 349 | struct pid **pid_ptr = task_pid_ptr(task, type); |
92476d7f EB |
350 | struct pid *pid; |
351 | int tmp; | |
1da177e4 | 352 | |
2c470475 | 353 | pid = *pid_ptr; |
1da177e4 | 354 | |
2c470475 EB |
355 | hlist_del_rcu(&task->pid_links[type]); |
356 | *pid_ptr = new; | |
1da177e4 | 357 | |
43f0df54 ON |
358 | if (type == PIDTYPE_PID) { |
359 | WARN_ON_ONCE(pid_has_task(pid, PIDTYPE_PID)); | |
360 | wake_up_all(&pid->wait_pidfd); | |
361 | } | |
362 | ||
92476d7f | 363 | for (tmp = PIDTYPE_MAX; --tmp >= 0; ) |
1d416a11 | 364 | if (pid_has_task(pid, tmp)) |
92476d7f | 365 | return; |
1da177e4 | 366 | |
92476d7f | 367 | free_pid(pid); |
1da177e4 LT |
368 | } |
369 | ||
24336eae ON |
370 | void detach_pid(struct task_struct *task, enum pid_type type) |
371 | { | |
372 | __change_pid(task, type, NULL); | |
373 | } | |
374 | ||
375 | void change_pid(struct task_struct *task, enum pid_type type, | |
376 | struct pid *pid) | |
377 | { | |
378 | __change_pid(task, type, pid); | |
81907739 | 379 | attach_pid(task, type); |
24336eae ON |
380 | } |
381 | ||
6b03d130 EB |
382 | void exchange_tids(struct task_struct *left, struct task_struct *right) |
383 | { | |
384 | struct pid *pid1 = left->thread_pid; | |
385 | struct pid *pid2 = right->thread_pid; | |
386 | struct hlist_head *head1 = &pid1->tasks[PIDTYPE_PID]; | |
387 | struct hlist_head *head2 = &pid2->tasks[PIDTYPE_PID]; | |
388 | ||
389 | /* Swap the single entry tid lists */ | |
390 | hlists_swap_heads_rcu(head1, head2); | |
391 | ||
392 | /* Swap the per task_struct pid */ | |
393 | rcu_assign_pointer(left->thread_pid, pid2); | |
394 | rcu_assign_pointer(right->thread_pid, pid1); | |
395 | ||
396 | /* Swap the cached value */ | |
397 | WRITE_ONCE(left->pid, pid_nr(pid2)); | |
398 | WRITE_ONCE(right->pid, pid_nr(pid1)); | |
399 | } | |
400 | ||
c18258c6 | 401 | /* transfer_pid is an optimization of attach_pid(new), detach_pid(old) */ |
7ad5b3a5 | 402 | void transfer_pid(struct task_struct *old, struct task_struct *new, |
c18258c6 EB |
403 | enum pid_type type) |
404 | { | |
a1c6d543 | 405 | WARN_ON_ONCE(type == PIDTYPE_PID); |
2c470475 | 406 | hlist_replace_rcu(&old->pid_links[type], &new->pid_links[type]); |
c18258c6 EB |
407 | } |
408 | ||
7ad5b3a5 | 409 | struct task_struct *pid_task(struct pid *pid, enum pid_type type) |
1da177e4 | 410 | { |
92476d7f EB |
411 | struct task_struct *result = NULL; |
412 | if (pid) { | |
413 | struct hlist_node *first; | |
67bdbffd | 414 | first = rcu_dereference_check(hlist_first_rcu(&pid->tasks[type]), |
db1466b3 | 415 | lockdep_tasklist_lock_is_held()); |
92476d7f | 416 | if (first) |
2c470475 | 417 | result = hlist_entry(first, struct task_struct, pid_links[(type)]); |
92476d7f EB |
418 | } |
419 | return result; | |
420 | } | |
eccba068 | 421 | EXPORT_SYMBOL(pid_task); |
1da177e4 | 422 | |
92476d7f | 423 | /* |
9728e5d6 | 424 | * Must be called under rcu_read_lock(). |
92476d7f | 425 | */ |
17f98dcf | 426 | struct task_struct *find_task_by_pid_ns(pid_t nr, struct pid_namespace *ns) |
92476d7f | 427 | { |
f78f5b90 PM |
428 | RCU_LOCKDEP_WARN(!rcu_read_lock_held(), |
429 | "find_task_by_pid_ns() needs rcu_read_lock() protection"); | |
17f98dcf | 430 | return pid_task(find_pid_ns(nr, ns), PIDTYPE_PID); |
92476d7f | 431 | } |
1da177e4 | 432 | |
228ebcbe PE |
433 | struct task_struct *find_task_by_vpid(pid_t vnr) |
434 | { | |
17cf22c3 | 435 | return find_task_by_pid_ns(vnr, task_active_pid_ns(current)); |
228ebcbe | 436 | } |
228ebcbe | 437 | |
2ee08260 MR |
438 | struct task_struct *find_get_task_by_vpid(pid_t nr) |
439 | { | |
440 | struct task_struct *task; | |
441 | ||
442 | rcu_read_lock(); | |
443 | task = find_task_by_vpid(nr); | |
444 | if (task) | |
445 | get_task_struct(task); | |
446 | rcu_read_unlock(); | |
447 | ||
448 | return task; | |
449 | } | |
450 | ||
1a657f78 ON |
451 | struct pid *get_task_pid(struct task_struct *task, enum pid_type type) |
452 | { | |
453 | struct pid *pid; | |
454 | rcu_read_lock(); | |
2c470475 | 455 | pid = get_pid(rcu_dereference(*task_pid_ptr(task, type))); |
1a657f78 ON |
456 | rcu_read_unlock(); |
457 | return pid; | |
458 | } | |
77c100c8 | 459 | EXPORT_SYMBOL_GPL(get_task_pid); |
1a657f78 | 460 | |
7ad5b3a5 | 461 | struct task_struct *get_pid_task(struct pid *pid, enum pid_type type) |
92476d7f EB |
462 | { |
463 | struct task_struct *result; | |
464 | rcu_read_lock(); | |
465 | result = pid_task(pid, type); | |
466 | if (result) | |
467 | get_task_struct(result); | |
468 | rcu_read_unlock(); | |
469 | return result; | |
1da177e4 | 470 | } |
77c100c8 | 471 | EXPORT_SYMBOL_GPL(get_pid_task); |
1da177e4 | 472 | |
92476d7f | 473 | struct pid *find_get_pid(pid_t nr) |
1da177e4 LT |
474 | { |
475 | struct pid *pid; | |
476 | ||
92476d7f | 477 | rcu_read_lock(); |
198fe21b | 478 | pid = get_pid(find_vpid(nr)); |
92476d7f | 479 | rcu_read_unlock(); |
1da177e4 | 480 | |
92476d7f | 481 | return pid; |
1da177e4 | 482 | } |
339caf2a | 483 | EXPORT_SYMBOL_GPL(find_get_pid); |
1da177e4 | 484 | |
7af57294 PE |
485 | pid_t pid_nr_ns(struct pid *pid, struct pid_namespace *ns) |
486 | { | |
487 | struct upid *upid; | |
488 | pid_t nr = 0; | |
489 | ||
490 | if (pid && ns->level <= pid->level) { | |
491 | upid = &pid->numbers[ns->level]; | |
492 | if (upid->ns == ns) | |
493 | nr = upid->nr; | |
494 | } | |
495 | return nr; | |
496 | } | |
4f82f457 | 497 | EXPORT_SYMBOL_GPL(pid_nr_ns); |
7af57294 | 498 | |
44c4e1b2 EB |
499 | pid_t pid_vnr(struct pid *pid) |
500 | { | |
17cf22c3 | 501 | return pid_nr_ns(pid, task_active_pid_ns(current)); |
44c4e1b2 EB |
502 | } |
503 | EXPORT_SYMBOL_GPL(pid_vnr); | |
504 | ||
52ee2dfd ON |
505 | pid_t __task_pid_nr_ns(struct task_struct *task, enum pid_type type, |
506 | struct pid_namespace *ns) | |
2f2a3a46 | 507 | { |
52ee2dfd ON |
508 | pid_t nr = 0; |
509 | ||
510 | rcu_read_lock(); | |
511 | if (!ns) | |
17cf22c3 | 512 | ns = task_active_pid_ns(current); |
1dd694a1 | 513 | nr = pid_nr_ns(rcu_dereference(*task_pid_ptr(task, type)), ns); |
52ee2dfd ON |
514 | rcu_read_unlock(); |
515 | ||
516 | return nr; | |
2f2a3a46 | 517 | } |
52ee2dfd | 518 | EXPORT_SYMBOL(__task_pid_nr_ns); |
2f2a3a46 | 519 | |
61bce0f1 EB |
520 | struct pid_namespace *task_active_pid_ns(struct task_struct *tsk) |
521 | { | |
522 | return ns_of_pid(task_pid(tsk)); | |
523 | } | |
524 | EXPORT_SYMBOL_GPL(task_active_pid_ns); | |
525 | ||
0804ef4b | 526 | /* |
025dfdaf | 527 | * Used by proc to find the first pid that is greater than or equal to nr. |
0804ef4b | 528 | * |
e49859e7 | 529 | * If there is a pid at nr this function is exactly the same as find_pid_ns. |
0804ef4b | 530 | */ |
198fe21b | 531 | struct pid *find_ge_pid(int nr, struct pid_namespace *ns) |
0804ef4b | 532 | { |
95846ecf | 533 | return idr_get_next(&ns->idr, &nr); |
0804ef4b | 534 | } |
4480c27c | 535 | EXPORT_SYMBOL_GPL(find_ge_pid); |
0804ef4b | 536 | |
1aa92cd3 MK |
537 | struct pid *pidfd_get_pid(unsigned int fd, unsigned int *flags) |
538 | { | |
539 | struct fd f; | |
540 | struct pid *pid; | |
541 | ||
542 | f = fdget(fd); | |
543 | if (!f.file) | |
544 | return ERR_PTR(-EBADF); | |
545 | ||
546 | pid = pidfd_pid(f.file); | |
547 | if (!IS_ERR(pid)) { | |
548 | get_pid(pid); | |
549 | *flags = f.file->f_flags; | |
550 | } | |
551 | ||
552 | fdput(f); | |
553 | return pid; | |
554 | } | |
555 | ||
e9bdcdbf CB |
556 | /** |
557 | * pidfd_get_task() - Get the task associated with a pidfd | |
558 | * | |
559 | * @pidfd: pidfd for which to get the task | |
560 | * @flags: flags associated with this pidfd | |
561 | * | |
562 | * Return the task associated with @pidfd. The function takes a reference on | |
563 | * the returned task. The caller is responsible for releasing that reference. | |
564 | * | |
e9bdcdbf CB |
565 | * Return: On success, the task_struct associated with the pidfd. |
566 | * On error, a negative errno number will be returned. | |
567 | */ | |
568 | struct task_struct *pidfd_get_task(int pidfd, unsigned int *flags) | |
569 | { | |
570 | unsigned int f_flags; | |
571 | struct pid *pid; | |
572 | struct task_struct *task; | |
573 | ||
574 | pid = pidfd_get_pid(pidfd, &f_flags); | |
575 | if (IS_ERR(pid)) | |
576 | return ERR_CAST(pid); | |
577 | ||
578 | task = get_pid_task(pid, PIDTYPE_TGID); | |
579 | put_pid(pid); | |
580 | if (!task) | |
581 | return ERR_PTR(-ESRCH); | |
582 | ||
583 | *flags = f_flags; | |
584 | return task; | |
585 | } | |
586 | ||
32fcb426 CB |
587 | /** |
588 | * pidfd_create() - Create a new pid file descriptor. | |
589 | * | |
6da73d15 CB |
590 | * @pid: struct pid that the pidfd will reference |
591 | * @flags: flags to pass | |
32fcb426 CB |
592 | * |
593 | * This creates a new pid file descriptor with the O_CLOEXEC flag set. | |
594 | * | |
595 | * Note, that this function can only be called after the fd table has | |
596 | * been unshared to avoid leaking the pidfd to the new process. | |
597 | * | |
c576e0fc MB |
598 | * This symbol should not be explicitly exported to loadable modules. |
599 | * | |
32fcb426 CB |
600 | * Return: On success, a cloexec pidfd is returned. |
601 | * On error, a negative errno number will be returned. | |
602 | */ | |
cdefbf23 | 603 | static int pidfd_create(struct pid *pid, unsigned int flags) |
32fcb426 | 604 | { |
6ae930d9 CB |
605 | int pidfd; |
606 | struct file *pidfd_file; | |
32fcb426 | 607 | |
6ae930d9 CB |
608 | pidfd = pidfd_prepare(pid, flags, &pidfd_file); |
609 | if (pidfd < 0) | |
610 | return pidfd; | |
490b9ba8 | 611 | |
6ae930d9 CB |
612 | fd_install(pidfd, pidfd_file); |
613 | return pidfd; | |
32fcb426 CB |
614 | } |
615 | ||
616 | /** | |
0c7752d5 | 617 | * sys_pidfd_open() - Open new pid file descriptor. |
32fcb426 CB |
618 | * |
619 | * @pid: pid for which to retrieve a pidfd | |
620 | * @flags: flags to pass | |
621 | * | |
622 | * This creates a new pid file descriptor with the O_CLOEXEC flag set for | |
64bef697 ON |
623 | * the task identified by @pid. Without PIDFD_THREAD flag the target task |
624 | * must be a thread-group leader. | |
32fcb426 CB |
625 | * |
626 | * Return: On success, a cloexec pidfd is returned. | |
627 | * On error, a negative errno number will be returned. | |
628 | */ | |
629 | SYSCALL_DEFINE2(pidfd_open, pid_t, pid, unsigned int, flags) | |
630 | { | |
1e1d0f0b | 631 | int fd; |
32fcb426 CB |
632 | struct pid *p; |
633 | ||
64bef697 | 634 | if (flags & ~(PIDFD_NONBLOCK | PIDFD_THREAD)) |
32fcb426 CB |
635 | return -EINVAL; |
636 | ||
637 | if (pid <= 0) | |
638 | return -EINVAL; | |
639 | ||
640 | p = find_get_pid(pid); | |
641 | if (!p) | |
642 | return -ESRCH; | |
643 | ||
490b9ba8 | 644 | fd = pidfd_create(p, flags); |
32fcb426 | 645 | |
32fcb426 CB |
646 | put_pid(p); |
647 | return fd; | |
648 | } | |
649 | ||
95846ecf | 650 | void __init pid_idr_init(void) |
1da177e4 | 651 | { |
840d6fe7 | 652 | /* Verify no one has done anything silly: */ |
e8cfbc24 | 653 | BUILD_BUG_ON(PID_MAX_LIMIT >= PIDNS_ADDING); |
c876ad76 | 654 | |
72680a19 HB |
655 | /* bump default and minimum pid_max based on number of cpus */ |
656 | pid_max = min(pid_max_max, max_t(int, pid_max, | |
657 | PIDS_PER_CPU_DEFAULT * num_possible_cpus())); | |
658 | pid_max_min = max_t(int, pid_max_min, | |
659 | PIDS_PER_CPU_MIN * num_possible_cpus()); | |
660 | pr_info("pid_max: default: %u minimum: %u\n", pid_max, pid_max_min); | |
661 | ||
95846ecf | 662 | idr_init(&init_pid_ns.idr); |
92476d7f | 663 | |
b69f0aeb | 664 | init_pid_ns.pid_cachep = kmem_cache_create("pid", |
dd546618 | 665 | struct_size_t(struct pid, numbers, 1), |
b69f0aeb KC |
666 | __alignof__(struct pid), |
667 | SLAB_HWCACHE_ALIGN | SLAB_PANIC | SLAB_ACCOUNT, | |
668 | NULL); | |
1da177e4 | 669 | } |
8649c322 SD |
670 | |
671 | static struct file *__pidfd_fget(struct task_struct *task, int fd) | |
672 | { | |
673 | struct file *file; | |
674 | int ret; | |
675 | ||
f7cfd871 | 676 | ret = down_read_killable(&task->signal->exec_update_lock); |
8649c322 SD |
677 | if (ret) |
678 | return ERR_PTR(ret); | |
679 | ||
680 | if (ptrace_may_access(task, PTRACE_MODE_ATTACH_REALCREDS)) | |
681 | file = fget_task(task, fd); | |
682 | else | |
683 | file = ERR_PTR(-EPERM); | |
684 | ||
f7cfd871 | 685 | up_read(&task->signal->exec_update_lock); |
8649c322 | 686 | |
0c9bd6bc TA |
687 | if (!file) { |
688 | /* | |
689 | * It is possible that the target thread is exiting; it can be | |
690 | * either: | |
691 | * 1. before exit_signals(), which gives a real fd | |
692 | * 2. before exit_files() takes the task_lock() gives a real fd | |
693 | * 3. after exit_files() releases task_lock(), ->files is NULL; | |
694 | * this has PF_EXITING, since it was set in exit_signals(), | |
695 | * __pidfd_fget() returns EBADF. | |
696 | * In case 3 we get EBADF, but that really means ESRCH, since | |
697 | * the task is currently exiting and has freed its files | |
698 | * struct, so we fix it up. | |
699 | */ | |
700 | if (task->flags & PF_EXITING) | |
701 | file = ERR_PTR(-ESRCH); | |
702 | else | |
703 | file = ERR_PTR(-EBADF); | |
704 | } | |
705 | ||
706 | return file; | |
8649c322 SD |
707 | } |
708 | ||
709 | static int pidfd_getfd(struct pid *pid, int fd) | |
710 | { | |
711 | struct task_struct *task; | |
712 | struct file *file; | |
713 | int ret; | |
714 | ||
715 | task = get_pid_task(pid, PIDTYPE_PID); | |
716 | if (!task) | |
717 | return -ESRCH; | |
718 | ||
719 | file = __pidfd_fget(task, fd); | |
720 | put_task_struct(task); | |
721 | if (IS_ERR(file)) | |
722 | return PTR_ERR(file); | |
723 | ||
4e94ddfe | 724 | ret = receive_fd(file, NULL, O_CLOEXEC); |
910d2f16 | 725 | fput(file); |
8649c322 SD |
726 | |
727 | return ret; | |
728 | } | |
729 | ||
730 | /** | |
731 | * sys_pidfd_getfd() - Get a file descriptor from another process | |
732 | * | |
733 | * @pidfd: the pidfd file descriptor of the process | |
734 | * @fd: the file descriptor number to get | |
735 | * @flags: flags on how to get the fd (reserved) | |
736 | * | |
737 | * This syscall gets a copy of a file descriptor from another process | |
738 | * based on the pidfd, and file descriptor number. It requires that | |
739 | * the calling process has the ability to ptrace the process represented | |
740 | * by the pidfd. The process which is having its file descriptor copied | |
741 | * is otherwise unaffected. | |
742 | * | |
743 | * Return: On success, a cloexec file descriptor is returned. | |
744 | * On error, a negative errno number will be returned. | |
745 | */ | |
746 | SYSCALL_DEFINE3(pidfd_getfd, int, pidfd, int, fd, | |
747 | unsigned int, flags) | |
748 | { | |
749 | struct pid *pid; | |
750 | struct fd f; | |
751 | int ret; | |
752 | ||
753 | /* flags is currently unused - make sure it's unset */ | |
754 | if (flags) | |
755 | return -EINVAL; | |
756 | ||
757 | f = fdget(pidfd); | |
758 | if (!f.file) | |
759 | return -EBADF; | |
760 | ||
761 | pid = pidfd_pid(f.file); | |
762 | if (IS_ERR(pid)) | |
763 | ret = PTR_ERR(pid); | |
764 | else | |
765 | ret = pidfd_getfd(pid, fd); | |
766 | ||
767 | fdput(f); | |
768 | return ret; | |
769 | } |