blob: 6500ef956f2f885793833e160891b4f441acd104 [file] [log] [blame]
Thomas Gleixner457c8992019-05-19 13:08:55 +01001// SPDX-License-Identifier: GPL-2.0-only
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
3 * Generic pidhash and scalable, time-bounded PID allocator
4 *
Nadia Yvette Chambers6d49e352012-12-06 10:39:54 +01005 * (C) 2002-2003 Nadia Yvette Chambers, IBM
6 * (C) 2004 Nadia Yvette Chambers, Oracle
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 * (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).
Pavel Emelyanov30e49c22007-10-18 23:40:10 -070022 *
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 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070028 */
29
30#include <linux/mm.h>
Paul Gortmaker9984de12011-05-23 14:51:41 -040031#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include <linux/slab.h>
33#include <linux/init.h>
Franck Bui-Huu82524742008-05-12 21:21:05 +020034#include <linux/rculist.h>
Mike Rapoport57c8a662018-10-30 15:09:49 -070035#include <linux/memblock.h>
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -080036#include <linux/pid_namespace.h>
Sukadev Bhattiprolu820e45d2007-05-10 22:23:00 -070037#include <linux/init_task.h>
Sukadev Bhattiprolu3eb07c82007-10-18 23:40:13 -070038#include <linux/syscalls.h>
David Howells0bb80f22013-04-12 01:50:06 +010039#include <linux/proc_ns.h>
Joel Fernandes (Google)f57e5152019-07-16 16:30:06 -070040#include <linux/refcount.h>
Christian Brauner32fcb422019-05-24 12:43:51 +020041#include <linux/anon_inodes.h>
42#include <linux/sched/signal.h>
Ingo Molnar29930022017-02-08 18:51:36 +010043#include <linux/sched/task.h>
Gargi Sharma95846ec2017-11-17 15:30:30 -080044#include <linux/idr.h>
Kees Cook4969f8a2020-06-09 16:21:38 -070045#include <net/sock.h>
Christian Brauner6da73d12020-09-02 12:21:27 +020046#include <uapi/linux/pidfd.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070047
David Howellse1e871a2018-01-02 15:12:01 +000048struct pid init_struct_pid = {
Joel Fernandes (Google)f57e5152019-07-16 16:30:06 -070049 .count = REFCOUNT_INIT(1),
David Howellse1e871a2018-01-02 15:12:01 +000050 .tasks = {
51 { .first = NULL },
52 { .first = NULL },
53 { .first = NULL },
54 },
55 .level = 0,
56 .numbers = { {
57 .nr = 0,
58 .ns = &init_pid_ns,
59 }, }
60};
Linus Torvalds1da177e2005-04-16 15:20:36 -070061
62int pid_max = PID_MAX_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -070063
64#define RESERVED_PIDS 300
65
66int pid_max_min = RESERVED_PIDS + 1;
67int pid_max_max = PID_MAX_LIMIT;
68
Linus Torvalds1da177e2005-04-16 15:20:36 -070069/*
70 * PID-map pages start out as NULL, they get allocated upon
71 * first use and are never deallocated. This way a low pid_max
72 * value does not cause lots of bitmaps to be allocated, but
73 * the scheme scales to up to 4 million PIDs, runtime.
74 */
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -080075struct pid_namespace init_pid_ns = {
Kirill Tkhai8eb71d92020-08-03 13:16:32 +030076 .ns.count = REFCOUNT_INIT(2),
Matthew Wilcoxf6bb2a22018-04-10 16:36:52 -070077 .idr = IDR_INIT(init_pid_ns.idr),
Gargi Sharmae8cfbc22017-11-17 15:30:34 -080078 .pid_allocated = PIDNS_ADDING,
Pavel Emelyanovfaacbfd2007-10-18 23:40:04 -070079 .level = 0,
80 .child_reaper = &init_task,
Eric W. Biederman49f4d8b2012-08-02 04:25:10 -070081 .user_ns = &init_user_ns,
Al Viro435d5f42014-10-31 22:56:04 -040082 .ns.inum = PROC_PID_INIT_INO,
Al Viro33c42942014-11-01 02:32:53 -040083#ifdef CONFIG_PID_NS
84 .ns.ops = &pidns_operations,
85#endif
Aleksa Sarai9876cfe2023-08-14 18:41:00 +100086#if defined(CONFIG_SYSCTL) && defined(CONFIG_MEMFD_CREATE)
87 .memfd_noexec_scope = MEMFD_NOEXEC_SCOPE_EXEC,
88#endif
Sukadev Bhattiprolu3fbc9642006-10-02 02:17:24 -070089};
Pavel Emelyanov198fe212007-10-18 23:40:06 -070090EXPORT_SYMBOL_GPL(init_pid_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -070091
Eric W. Biederman92476d72006-03-31 02:31:42 -080092/*
93 * Note: disable interrupts while the pidmap_lock is held as an
94 * interrupt might come in and do read_lock(&tasklist_lock).
95 *
96 * If we don't disable interrupts there is a nasty deadlock between
97 * detach_pid()->free_pid() and another cpu that does
98 * spin_lock(&pidmap_lock) followed by an interrupt routine that does
99 * read_lock(&tasklist_lock);
100 *
101 * After we clean up the tasklist_lock and know there are no
102 * irq handlers that take it we can leave the interrupts enabled.
103 * For now it is easier to be safe than to prove it can't happen.
104 */
Sukadev Bhattiprolu3fbc9642006-10-02 02:17:24 -0700105
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106static __cacheline_aligned_in_smp DEFINE_SPINLOCK(pidmap_lock);
107
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -0800108void put_pid(struct pid *pid)
Eric W. Biederman92476d72006-03-31 02:31:42 -0800109{
Pavel Emelianovbaf8f0f2007-10-18 23:39:48 -0700110 struct pid_namespace *ns;
111
Eric W. Biederman92476d72006-03-31 02:31:42 -0800112 if (!pid)
113 return;
Pavel Emelianovbaf8f0f2007-10-18 23:39:48 -0700114
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -0700115 ns = pid->numbers[pid->level].ns;
Joel Fernandes (Google)f57e5152019-07-16 16:30:06 -0700116 if (refcount_dec_and_test(&pid->count)) {
Pavel Emelianovbaf8f0f2007-10-18 23:39:48 -0700117 kmem_cache_free(ns->pid_cachep, pid);
Pavel Emelyanovb461cc02007-10-18 23:40:09 -0700118 put_pid_ns(ns);
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -0700119 }
Eric W. Biederman92476d72006-03-31 02:31:42 -0800120}
Eric W. Biedermanbbf73142006-10-02 02:17:11 -0700121EXPORT_SYMBOL_GPL(put_pid);
Eric W. Biederman92476d72006-03-31 02:31:42 -0800122
123static void delayed_put_pid(struct rcu_head *rhp)
124{
125 struct pid *pid = container_of(rhp, struct pid, rcu);
126 put_pid(pid);
127}
128
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -0800129void free_pid(struct pid *pid)
Eric W. Biederman92476d72006-03-31 02:31:42 -0800130{
131 /* We can be called with write_lock_irq(&tasklist_lock) held */
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -0700132 int i;
Eric W. Biederman92476d72006-03-31 02:31:42 -0800133 unsigned long flags;
134
135 spin_lock_irqsave(&pidmap_lock, flags);
Eric W. Biederman0a01f2c2012-08-01 10:33:47 -0700136 for (i = 0; i <= pid->level; i++) {
137 struct upid *upid = pid->numbers + i;
Eric W. Biedermanaf4b8a82012-08-01 15:03:42 -0700138 struct pid_namespace *ns = upid->ns;
Gargi Sharmae8cfbc22017-11-17 15:30:34 -0800139 switch (--ns->pid_allocated) {
Eric W. Biedermana6064882013-08-29 13:56:50 -0700140 case 2:
Eric W. Biedermanaf4b8a82012-08-01 15:03:42 -0700141 case 1:
142 /* When all that is left in the pid namespace
143 * is the reaper wake up the reaper. The reaper
144 * may be sleeping in zap_pid_ns_processes().
145 */
146 wake_up_process(ns->child_reaper);
147 break;
Gargi Sharmae8cfbc22017-11-17 15:30:34 -0800148 case PIDNS_ADDING:
Oleg Nesterov314a8ad2013-09-30 13:45:27 -0700149 /* Handle a fork failure of the first process */
150 WARN_ON(ns->child_reaper);
Gargi Sharmae8cfbc22017-11-17 15:30:34 -0800151 ns->pid_allocated = 0;
Eric W. Biedermanaf4b8a82012-08-01 15:03:42 -0700152 break;
Eric W. Biederman5e1182deb2010-07-12 18:50:25 -0700153 }
Gargi Sharma95846ec2017-11-17 15:30:30 -0800154
155 idr_remove(&ns->idr, upid->nr);
Eric W. Biederman0a01f2c2012-08-01 10:33:47 -0700156 }
Eric W. Biederman92476d72006-03-31 02:31:42 -0800157 spin_unlock_irqrestore(&pidmap_lock, flags);
158
Eric W. Biederman92476d72006-03-31 02:31:42 -0800159 call_rcu(&pid->rcu, delayed_put_pid);
160}
161
Adrian Reber49cb2fc2019-11-15 13:36:20 +0100162struct pid *alloc_pid(struct pid_namespace *ns, pid_t *set_tid,
163 size_t set_tid_size)
Eric W. Biederman92476d72006-03-31 02:31:42 -0800164{
165 struct pid *pid;
166 enum pid_type type;
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -0700167 int i, nr;
168 struct pid_namespace *tmp;
Pavel Emelyanov198fe212007-10-18 23:40:06 -0700169 struct upid *upid;
Michal Hocko35f71bc2015-04-16 12:47:38 -0700170 int retval = -ENOMEM;
Eric W. Biederman92476d72006-03-31 02:31:42 -0800171
Adrian Reber49cb2fc2019-11-15 13:36:20 +0100172 /*
173 * set_tid_size contains the size of the set_tid array. Starting at
174 * the most nested currently active PID namespace it tells alloc_pid()
175 * which PID to set for a process in that most nested PID namespace
176 * up to set_tid_size PID namespaces. It does not have to set the PID
177 * for a process in all nested PID namespaces but set_tid_size must
178 * never be greater than the current ns->level + 1.
179 */
180 if (set_tid_size > ns->level + 1)
181 return ERR_PTR(-EINVAL);
182
Pavel Emelianovbaf8f0f2007-10-18 23:39:48 -0700183 pid = kmem_cache_alloc(ns->pid_cachep, GFP_KERNEL);
Eric W. Biederman92476d72006-03-31 02:31:42 -0800184 if (!pid)
Michal Hocko35f71bc2015-04-16 12:47:38 -0700185 return ERR_PTR(retval);
Eric W. Biederman92476d72006-03-31 02:31:42 -0800186
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -0700187 tmp = ns;
Eric W. Biederman0a01f2c2012-08-01 10:33:47 -0700188 pid->level = ns->level;
Gargi Sharma95846ec2017-11-17 15:30:30 -0800189
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -0700190 for (i = ns->level; i >= 0; i--) {
Adrian Reber49cb2fc2019-11-15 13:36:20 +0100191 int tid = 0;
192
193 if (set_tid_size) {
194 tid = set_tid[ns->level - i];
195
196 retval = -EINVAL;
197 if (tid < 1 || tid >= pid_max)
198 goto out_free;
199 /*
200 * Also fail if a PID != 1 is requested and
201 * no PID 1 exists.
202 */
203 if (tid != 1 && !tmp->child_reaper)
204 goto out_free;
205 retval = -EPERM;
Adrian Reber1caef812020-07-19 12:04:12 +0200206 if (!checkpoint_restore_ns_capable(tmp->user_ns))
Adrian Reber49cb2fc2019-11-15 13:36:20 +0100207 goto out_free;
208 set_tid_size--;
209 }
Gargi Sharma95846ec2017-11-17 15:30:30 -0800210
211 idr_preload(GFP_KERNEL);
212 spin_lock_irq(&pidmap_lock);
213
Adrian Reber49cb2fc2019-11-15 13:36:20 +0100214 if (tid) {
215 nr = idr_alloc(&tmp->idr, NULL, tid,
216 tid + 1, GFP_ATOMIC);
217 /*
218 * If ENOSPC is returned it means that the PID is
219 * alreay in use. Return EEXIST in that case.
220 */
221 if (nr == -ENOSPC)
222 nr = -EEXIST;
223 } else {
224 int pid_min = 1;
225 /*
226 * init really needs pid 1, but after reaching the
227 * maximum wrap back to RESERVED_PIDS
228 */
229 if (idr_get_cursor(&tmp->idr) > RESERVED_PIDS)
230 pid_min = RESERVED_PIDS;
Gargi Sharma95846ec2017-11-17 15:30:30 -0800231
Adrian Reber49cb2fc2019-11-15 13:36:20 +0100232 /*
233 * Store a null pointer so find_pid_ns does not find
234 * a partially initialized PID (see below).
235 */
236 nr = idr_alloc_cyclic(&tmp->idr, NULL, pid_min,
237 pid_max, GFP_ATOMIC);
238 }
Gargi Sharma95846ec2017-11-17 15:30:30 -0800239 spin_unlock_irq(&pidmap_lock);
240 idr_preload_end();
241
Arnd Bergmann287980e2016-05-27 23:23:25 +0200242 if (nr < 0) {
KJ Tsanaktsidisf83606f2018-09-20 12:22:25 -0700243 retval = (nr == -ENOSPC) ? -EAGAIN : nr;
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -0700244 goto out_free;
Michal Hocko35f71bc2015-04-16 12:47:38 -0700245 }
Eric W. Biederman92476d72006-03-31 02:31:42 -0800246
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -0700247 pid->numbers[i].nr = nr;
248 pid->numbers[i].ns = tmp;
249 tmp = tmp->parent;
250 }
251
Christian Brauner10dab842020-03-08 14:29:17 +0100252 /*
253 * ENOMEM is not the most obvious choice especially for the case
254 * where the child subreaper has already exited and the pid
255 * namespace denies the creation of any new processes. But ENOMEM
256 * is what we have exposed to userspace for a long time and it is
257 * documented behavior for pid namespaces. So we can't easily
258 * change it even if there were an error code better suited.
259 */
Corey Minyardb26ebfe2020-03-06 11:23:14 -0600260 retval = -ENOMEM;
261
Pavel Emelyanovb461cc02007-10-18 23:40:09 -0700262 get_pid_ns(ns);
Joel Fernandes (Google)f57e5152019-07-16 16:30:06 -0700263 refcount_set(&pid->count, 1);
Eric W. Biederman63f818f2020-04-07 09:43:04 -0500264 spin_lock_init(&pid->lock);
Eric W. Biederman92476d72006-03-31 02:31:42 -0800265 for (type = 0; type < PIDTYPE_MAX; ++type)
266 INIT_HLIST_HEAD(&pid->tasks[type]);
267
Joel Fernandes (Google)b53b0b92019-04-30 12:21:53 -0400268 init_waitqueue_head(&pid->wait_pidfd);
Eric W. Biederman7bc3e6e2020-02-19 18:22:26 -0600269 INIT_HLIST_HEAD(&pid->inodes);
Joel Fernandes (Google)b53b0b92019-04-30 12:21:53 -0400270
André Goddard Rosa417e3152009-12-15 16:47:40 -0800271 upid = pid->numbers + ns->level;
Eric W. Biederman92476d72006-03-31 02:31:42 -0800272 spin_lock_irq(&pidmap_lock);
Gargi Sharmae8cfbc22017-11-17 15:30:34 -0800273 if (!(ns->pid_allocated & PIDNS_ADDING))
Eric W. Biederman5e1182deb2010-07-12 18:50:25 -0700274 goto out_unlock;
Eric W. Biederman0a01f2c2012-08-01 10:33:47 -0700275 for ( ; upid >= pid->numbers; --upid) {
Gargi Sharma95846ec2017-11-17 15:30:30 -0800276 /* Make the PID visible to find_pid_ns. */
277 idr_replace(&upid->ns->idr, pid, upid->nr);
Gargi Sharmae8cfbc22017-11-17 15:30:34 -0800278 upid->ns->pid_allocated++;
Eric W. Biederman0a01f2c2012-08-01 10:33:47 -0700279 }
Eric W. Biederman92476d72006-03-31 02:31:42 -0800280 spin_unlock_irq(&pidmap_lock);
281
Eric W. Biederman92476d72006-03-31 02:31:42 -0800282 return pid;
283
Eric W. Biederman5e1182deb2010-07-12 18:50:25 -0700284out_unlock:
Eric W. Biederman6e666882013-02-12 13:46:23 -0800285 spin_unlock_irq(&pidmap_lock);
Oleg Nesterov24c037e2014-12-10 15:55:25 -0800286 put_pid_ns(ns);
287
Eric W. Biederman92476d72006-03-31 02:31:42 -0800288out_free:
Gargi Sharma95846ec2017-11-17 15:30:30 -0800289 spin_lock_irq(&pidmap_lock);
Matthew Wilcox1a80dad2018-12-28 07:22:26 -0800290 while (++i <= ns->level) {
291 upid = pid->numbers + i;
292 idr_remove(&upid->ns->idr, upid->nr);
293 }
Gargi Sharma95846ec2017-11-17 15:30:30 -0800294
Eric W. Biedermanc0ee5542017-12-22 12:37:43 -0600295 /* On failure to allocate the first pid, reset the state */
296 if (ns->pid_allocated == PIDNS_ADDING)
297 idr_set_cursor(&ns->idr, 0);
298
Gargi Sharma95846ec2017-11-17 15:30:30 -0800299 spin_unlock_irq(&pidmap_lock);
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -0700300
Pavel Emelianovbaf8f0f2007-10-18 23:39:48 -0700301 kmem_cache_free(ns->pid_cachep, pid);
Michal Hocko35f71bc2015-04-16 12:47:38 -0700302 return ERR_PTR(retval);
Eric W. Biederman92476d72006-03-31 02:31:42 -0800303}
304
Eric W. Biedermanc876ad762012-12-21 20:27:12 -0800305void disable_pid_allocation(struct pid_namespace *ns)
306{
307 spin_lock_irq(&pidmap_lock);
Gargi Sharmae8cfbc22017-11-17 15:30:34 -0800308 ns->pid_allocated &= ~PIDNS_ADDING;
Eric W. Biedermanc876ad762012-12-21 20:27:12 -0800309 spin_unlock_irq(&pidmap_lock);
310}
311
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -0800312struct pid *find_pid_ns(int nr, struct pid_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313{
Gargi Sharmae8cfbc22017-11-17 15:30:34 -0800314 return idr_find(&ns->idr, nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315}
Pavel Emelyanov198fe212007-10-18 23:40:06 -0700316EXPORT_SYMBOL_GPL(find_pid_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317
Pavel Emelyanov89905712007-10-18 23:40:19 -0700318struct pid *find_vpid(int nr)
319{
Eric W. Biederman17cf22c2010-03-02 14:51:53 -0800320 return find_pid_ns(nr, task_active_pid_ns(current));
Pavel Emelyanov89905712007-10-18 23:40:19 -0700321}
322EXPORT_SYMBOL_GPL(find_vpid);
323
Eric W. Biederman2c470472017-09-26 13:06:43 -0500324static struct pid **task_pid_ptr(struct task_struct *task, enum pid_type type)
325{
326 return (type == PIDTYPE_PID) ?
327 &task->thread_pid :
Eric W. Biederman2c470472017-09-26 13:06:43 -0500328 &task->signal->pids[type];
329}
330
Sukadev Bhattiprolue713d0d2007-05-10 22:22:58 -0700331/*
332 * attach_pid() must be called with the tasklist_lock write-held.
333 */
Oleg Nesterov81907732013-07-03 15:08:31 -0700334void attach_pid(struct task_struct *task, enum pid_type type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335{
Eric W. Biederman2c470472017-09-26 13:06:43 -0500336 struct pid *pid = *task_pid_ptr(task, type);
337 hlist_add_head_rcu(&task->pid_links[type], &pid->tasks[type]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338}
339
Oleg Nesterov24336eae2008-04-30 00:54:26 -0700340static void __change_pid(struct task_struct *task, enum pid_type type,
341 struct pid *new)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342{
Eric W. Biederman2c470472017-09-26 13:06:43 -0500343 struct pid **pid_ptr = task_pid_ptr(task, type);
Eric W. Biederman92476d72006-03-31 02:31:42 -0800344 struct pid *pid;
345 int tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346
Eric W. Biederman2c470472017-09-26 13:06:43 -0500347 pid = *pid_ptr;
Eric W. Biederman92476d72006-03-31 02:31:42 -0800348
Eric W. Biederman2c470472017-09-26 13:06:43 -0500349 hlist_del_rcu(&task->pid_links[type]);
350 *pid_ptr = new;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351
352 for (tmp = PIDTYPE_MAX; --tmp >= 0; )
Christian Brauner1d416a12019-10-17 12:18:30 +0200353 if (pid_has_task(pid, tmp))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 return;
355
Eric W. Biederman92476d72006-03-31 02:31:42 -0800356 free_pid(pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357}
358
Oleg Nesterov24336eae2008-04-30 00:54:26 -0700359void detach_pid(struct task_struct *task, enum pid_type type)
360{
361 __change_pid(task, type, NULL);
362}
363
364void change_pid(struct task_struct *task, enum pid_type type,
365 struct pid *pid)
366{
367 __change_pid(task, type, pid);
Oleg Nesterov81907732013-07-03 15:08:31 -0700368 attach_pid(task, type);
Oleg Nesterov24336eae2008-04-30 00:54:26 -0700369}
370
Eric W. Biederman6b03d132020-04-19 06:35:02 -0500371void exchange_tids(struct task_struct *left, struct task_struct *right)
372{
373 struct pid *pid1 = left->thread_pid;
374 struct pid *pid2 = right->thread_pid;
375 struct hlist_head *head1 = &pid1->tasks[PIDTYPE_PID];
376 struct hlist_head *head2 = &pid2->tasks[PIDTYPE_PID];
377
378 /* Swap the single entry tid lists */
379 hlists_swap_heads_rcu(head1, head2);
380
381 /* Swap the per task_struct pid */
382 rcu_assign_pointer(left->thread_pid, pid2);
383 rcu_assign_pointer(right->thread_pid, pid1);
384
385 /* Swap the cached value */
386 WRITE_ONCE(left->pid, pid_nr(pid2));
387 WRITE_ONCE(right->pid, pid_nr(pid1));
388}
389
Eric W. Biedermanc18258c2006-09-27 01:51:06 -0700390/* transfer_pid is an optimization of attach_pid(new), detach_pid(old) */
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -0800391void transfer_pid(struct task_struct *old, struct task_struct *new,
Eric W. Biedermanc18258c2006-09-27 01:51:06 -0700392 enum pid_type type)
393{
Eric W. Biederman2c470472017-09-26 13:06:43 -0500394 if (type == PIDTYPE_PID)
395 new->thread_pid = old->thread_pid;
396 hlist_replace_rcu(&old->pid_links[type], &new->pid_links[type]);
Eric W. Biedermanc18258c2006-09-27 01:51:06 -0700397}
398
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -0800399struct task_struct *pid_task(struct pid *pid, enum pid_type type)
Eric W. Biederman92476d72006-03-31 02:31:42 -0800400{
401 struct task_struct *result = NULL;
402 if (pid) {
403 struct hlist_node *first;
Arnd Bergmann67bdbff2010-02-25 16:55:13 +0100404 first = rcu_dereference_check(hlist_first_rcu(&pid->tasks[type]),
Paul E. McKenneydb1466b2010-03-03 07:46:56 -0800405 lockdep_tasklist_lock_is_held());
Eric W. Biederman92476d72006-03-31 02:31:42 -0800406 if (first)
Eric W. Biederman2c470472017-09-26 13:06:43 -0500407 result = hlist_entry(first, struct task_struct, pid_links[(type)]);
Eric W. Biederman92476d72006-03-31 02:31:42 -0800408 }
409 return result;
410}
Pavel Emelyanoveccba062008-02-07 00:13:21 -0800411EXPORT_SYMBOL(pid_task);
Eric W. Biederman92476d72006-03-31 02:31:42 -0800412
413/*
Tetsuo Handa9728e5d2010-03-05 13:42:56 -0800414 * Must be called under rcu_read_lock().
Eric W. Biederman92476d72006-03-31 02:31:42 -0800415 */
Christoph Hellwig17f98dc2009-06-17 16:27:51 -0700416struct task_struct *find_task_by_pid_ns(pid_t nr, struct pid_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417{
Paul E. McKenneyf78f5b92015-06-18 15:50:02 -0700418 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
419 "find_task_by_pid_ns() needs rcu_read_lock() protection");
Christoph Hellwig17f98dc2009-06-17 16:27:51 -0700420 return pid_task(find_pid_ns(nr, ns), PIDTYPE_PID);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421}
422
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -0700423struct task_struct *find_task_by_vpid(pid_t vnr)
424{
Eric W. Biederman17cf22c2010-03-02 14:51:53 -0800425 return find_task_by_pid_ns(vnr, task_active_pid_ns(current));
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -0700426}
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -0700427
Mike Rapoport2ee08262018-02-06 15:40:17 -0800428struct task_struct *find_get_task_by_vpid(pid_t nr)
429{
430 struct task_struct *task;
431
432 rcu_read_lock();
433 task = find_task_by_vpid(nr);
434 if (task)
435 get_task_struct(task);
436 rcu_read_unlock();
437
438 return task;
439}
440
Oleg Nesterov1a657f782006-10-02 02:18:59 -0700441struct pid *get_task_pid(struct task_struct *task, enum pid_type type)
442{
443 struct pid *pid;
444 rcu_read_lock();
Eric W. Biederman2c470472017-09-26 13:06:43 -0500445 pid = get_pid(rcu_dereference(*task_pid_ptr(task, type)));
Oleg Nesterov1a657f782006-10-02 02:18:59 -0700446 rcu_read_unlock();
447 return pid;
448}
Rik van Riel77c100c2011-02-01 09:51:46 -0500449EXPORT_SYMBOL_GPL(get_task_pid);
Oleg Nesterov1a657f782006-10-02 02:18:59 -0700450
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -0800451struct task_struct *get_pid_task(struct pid *pid, enum pid_type type)
Eric W. Biederman92476d72006-03-31 02:31:42 -0800452{
453 struct task_struct *result;
454 rcu_read_lock();
455 result = pid_task(pid, type);
456 if (result)
457 get_task_struct(result);
458 rcu_read_unlock();
459 return result;
460}
Rik van Riel77c100c2011-02-01 09:51:46 -0500461EXPORT_SYMBOL_GPL(get_pid_task);
Eric W. Biederman92476d72006-03-31 02:31:42 -0800462
463struct pid *find_get_pid(pid_t nr)
464{
465 struct pid *pid;
466
467 rcu_read_lock();
Pavel Emelyanov198fe212007-10-18 23:40:06 -0700468 pid = get_pid(find_vpid(nr));
Eric W. Biederman92476d72006-03-31 02:31:42 -0800469 rcu_read_unlock();
470
471 return pid;
472}
David Sterba339caf22008-07-25 01:48:31 -0700473EXPORT_SYMBOL_GPL(find_get_pid);
Eric W. Biederman92476d72006-03-31 02:31:42 -0800474
Pavel Emelyanov7af57292007-10-18 23:40:06 -0700475pid_t pid_nr_ns(struct pid *pid, struct pid_namespace *ns)
476{
477 struct upid *upid;
478 pid_t nr = 0;
479
480 if (pid && ns->level <= pid->level) {
481 upid = &pid->numbers[ns->level];
482 if (upid->ns == ns)
483 nr = upid->nr;
484 }
485 return nr;
486}
Eric W. Biederman4f82f452012-05-24 10:37:59 -0600487EXPORT_SYMBOL_GPL(pid_nr_ns);
Pavel Emelyanov7af57292007-10-18 23:40:06 -0700488
Eric W. Biederman44c4e1b2008-02-08 04:19:15 -0800489pid_t pid_vnr(struct pid *pid)
490{
Eric W. Biederman17cf22c2010-03-02 14:51:53 -0800491 return pid_nr_ns(pid, task_active_pid_ns(current));
Eric W. Biederman44c4e1b2008-02-08 04:19:15 -0800492}
493EXPORT_SYMBOL_GPL(pid_vnr);
494
Oleg Nesterov52ee2df2009-04-02 16:58:38 -0700495pid_t __task_pid_nr_ns(struct task_struct *task, enum pid_type type,
496 struct pid_namespace *ns)
Pavel Emelyanov2f2a3a42007-10-18 23:40:19 -0700497{
Oleg Nesterov52ee2df2009-04-02 16:58:38 -0700498 pid_t nr = 0;
499
500 rcu_read_lock();
501 if (!ns)
Eric W. Biederman17cf22c2010-03-02 14:51:53 -0800502 ns = task_active_pid_ns(current);
Oleg Nesterov1dd694a2020-04-21 12:19:04 +0200503 nr = pid_nr_ns(rcu_dereference(*task_pid_ptr(task, type)), ns);
Oleg Nesterov52ee2df2009-04-02 16:58:38 -0700504 rcu_read_unlock();
505
506 return nr;
Pavel Emelyanov2f2a3a42007-10-18 23:40:19 -0700507}
Oleg Nesterov52ee2df2009-04-02 16:58:38 -0700508EXPORT_SYMBOL(__task_pid_nr_ns);
Pavel Emelyanov2f2a3a42007-10-18 23:40:19 -0700509
Eric W. Biederman61bce0f2009-01-07 18:08:49 -0800510struct pid_namespace *task_active_pid_ns(struct task_struct *tsk)
511{
512 return ns_of_pid(task_pid(tsk));
513}
514EXPORT_SYMBOL_GPL(task_active_pid_ns);
515
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516/*
Frederik Schwarzer025dfda2008-10-16 19:02:37 +0200517 * Used by proc to find the first pid that is greater than or equal to nr.
Eric W. Biederman0804ef42006-10-02 02:17:04 -0700518 *
Pavel Emelyanove49859e2008-07-25 01:48:36 -0700519 * If there is a pid at nr this function is exactly the same as find_pid_ns.
Eric W. Biederman0804ef42006-10-02 02:17:04 -0700520 */
Pavel Emelyanov198fe212007-10-18 23:40:06 -0700521struct pid *find_ge_pid(int nr, struct pid_namespace *ns)
Eric W. Biederman0804ef42006-10-02 02:17:04 -0700522{
Gargi Sharma95846ec2017-11-17 15:30:30 -0800523 return idr_get_next(&ns->idr, &nr);
Eric W. Biederman0804ef42006-10-02 02:17:04 -0700524}
Andreas Gruenbacher4480c272022-06-08 16:22:55 +0200525EXPORT_SYMBOL_GPL(find_ge_pid);
Eric W. Biederman0804ef42006-10-02 02:17:04 -0700526
Minchan Kim1aa92cd2020-10-17 16:14:54 -0700527struct pid *pidfd_get_pid(unsigned int fd, unsigned int *flags)
528{
529 struct fd f;
530 struct pid *pid;
531
532 f = fdget(fd);
533 if (!f.file)
534 return ERR_PTR(-EBADF);
535
536 pid = pidfd_pid(f.file);
537 if (!IS_ERR(pid)) {
538 get_pid(pid);
539 *flags = f.file->f_flags;
540 }
541
542 fdput(f);
543 return pid;
544}
545
Christian Brauner32fcb422019-05-24 12:43:51 +0200546/**
Christian Braunere9bdcdb2021-10-11 15:32:44 +0200547 * pidfd_get_task() - Get the task associated with a pidfd
548 *
549 * @pidfd: pidfd for which to get the task
550 * @flags: flags associated with this pidfd
551 *
552 * Return the task associated with @pidfd. The function takes a reference on
553 * the returned task. The caller is responsible for releasing that reference.
554 *
555 * Currently, the process identified by @pidfd is always a thread-group leader.
556 * This restriction currently exists for all aspects of pidfds including pidfd
557 * creation (CLONE_PIDFD cannot be used with CLONE_THREAD) and pidfd polling
558 * (only supports thread group leaders).
559 *
560 * Return: On success, the task_struct associated with the pidfd.
561 * On error, a negative errno number will be returned.
562 */
563struct task_struct *pidfd_get_task(int pidfd, unsigned int *flags)
564{
565 unsigned int f_flags;
566 struct pid *pid;
567 struct task_struct *task;
568
569 pid = pidfd_get_pid(pidfd, &f_flags);
570 if (IS_ERR(pid))
571 return ERR_CAST(pid);
572
573 task = get_pid_task(pid, PIDTYPE_TGID);
574 put_pid(pid);
575 if (!task)
576 return ERR_PTR(-ESRCH);
577
578 *flags = f_flags;
579 return task;
580}
581
582/**
Christian Brauner32fcb422019-05-24 12:43:51 +0200583 * pidfd_create() - Create a new pid file descriptor.
584 *
Christian Brauner6da73d12020-09-02 12:21:27 +0200585 * @pid: struct pid that the pidfd will reference
586 * @flags: flags to pass
Christian Brauner32fcb422019-05-24 12:43:51 +0200587 *
588 * This creates a new pid file descriptor with the O_CLOEXEC flag set.
589 *
590 * Note, that this function can only be called after the fd table has
591 * been unshared to avoid leaking the pidfd to the new process.
592 *
Matthew Bobrowskic576e0f2021-08-08 15:24:33 +1000593 * This symbol should not be explicitly exported to loadable modules.
594 *
Christian Brauner32fcb422019-05-24 12:43:51 +0200595 * Return: On success, a cloexec pidfd is returned.
596 * On error, a negative errno number will be returned.
597 */
Matthew Bobrowskic576e0f2021-08-08 15:24:33 +1000598int pidfd_create(struct pid *pid, unsigned int flags)
Christian Brauner32fcb422019-05-24 12:43:51 +0200599{
Christian Brauner6ae930d2023-03-27 20:22:51 +0200600 int pidfd;
601 struct file *pidfd_file;
Christian Brauner32fcb422019-05-24 12:43:51 +0200602
Christian Brauner6ae930d2023-03-27 20:22:51 +0200603 pidfd = pidfd_prepare(pid, flags, &pidfd_file);
604 if (pidfd < 0)
605 return pidfd;
Matthew Bobrowski490b9ba2021-08-08 15:25:05 +1000606
Christian Brauner6ae930d2023-03-27 20:22:51 +0200607 fd_install(pidfd, pidfd_file);
608 return pidfd;
Christian Brauner32fcb422019-05-24 12:43:51 +0200609}
610
611/**
Randy Dunlap0c7752d2023-09-11 23:08:22 -0700612 * sys_pidfd_open() - Open new pid file descriptor.
Christian Brauner32fcb422019-05-24 12:43:51 +0200613 *
614 * @pid: pid for which to retrieve a pidfd
615 * @flags: flags to pass
616 *
617 * This creates a new pid file descriptor with the O_CLOEXEC flag set for
618 * the process identified by @pid. Currently, the process identified by
619 * @pid must be a thread-group leader. This restriction currently exists
620 * for all aspects of pidfds including pidfd creation (CLONE_PIDFD cannot
621 * be used with CLONE_THREAD) and pidfd polling (only supports thread group
622 * leaders).
623 *
624 * Return: On success, a cloexec pidfd is returned.
625 * On error, a negative errno number will be returned.
626 */
627SYSCALL_DEFINE2(pidfd_open, pid_t, pid, unsigned int, flags)
628{
Christian Brauner1e1d0f02019-10-17 12:18:32 +0200629 int fd;
Christian Brauner32fcb422019-05-24 12:43:51 +0200630 struct pid *p;
631
Christian Brauner6da73d12020-09-02 12:21:27 +0200632 if (flags & ~PIDFD_NONBLOCK)
Christian Brauner32fcb422019-05-24 12:43:51 +0200633 return -EINVAL;
634
635 if (pid <= 0)
636 return -EINVAL;
637
638 p = find_get_pid(pid);
639 if (!p)
640 return -ESRCH;
641
Matthew Bobrowski490b9ba2021-08-08 15:25:05 +1000642 fd = pidfd_create(p, flags);
Christian Brauner32fcb422019-05-24 12:43:51 +0200643
Christian Brauner32fcb422019-05-24 12:43:51 +0200644 put_pid(p);
645 return fd;
646}
647
Gargi Sharma95846ec2017-11-17 15:30:30 -0800648void __init pid_idr_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649{
Zhen Lei840d6fe2016-01-30 10:04:17 +0800650 /* Verify no one has done anything silly: */
Gargi Sharmae8cfbc22017-11-17 15:30:34 -0800651 BUILD_BUG_ON(PID_MAX_LIMIT >= PIDNS_ADDING);
Eric W. Biedermanc876ad762012-12-21 20:27:12 -0800652
Hedi Berriche72680a12010-05-26 14:44:06 -0700653 /* bump default and minimum pid_max based on number of cpus */
654 pid_max = min(pid_max_max, max_t(int, pid_max,
655 PIDS_PER_CPU_DEFAULT * num_possible_cpus()));
656 pid_max_min = max_t(int, pid_max_min,
657 PIDS_PER_CPU_MIN * num_possible_cpus());
658 pr_info("pid_max: default: %u minimum: %u\n", pid_max, pid_max_min);
659
Gargi Sharma95846ec2017-11-17 15:30:30 -0800660 idr_init(&init_pid_ns.idr);
Eric W. Biederman92476d72006-03-31 02:31:42 -0800661
Kees Cookb69f0ae2023-06-30 09:46:17 +0200662 init_pid_ns.pid_cachep = kmem_cache_create("pid",
Christian Braunerdd546612023-07-01 08:44:44 +0200663 struct_size_t(struct pid, numbers, 1),
Kees Cookb69f0ae2023-06-30 09:46:17 +0200664 __alignof__(struct pid),
665 SLAB_HWCACHE_ALIGN | SLAB_PANIC | SLAB_ACCOUNT,
666 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667}
Sargun Dhillon8649c322020-01-07 09:59:25 -0800668
669static struct file *__pidfd_fget(struct task_struct *task, int fd)
670{
671 struct file *file;
672 int ret;
673
Eric W. Biedermanf7cfd872020-12-03 14:12:00 -0600674 ret = down_read_killable(&task->signal->exec_update_lock);
Sargun Dhillon8649c322020-01-07 09:59:25 -0800675 if (ret)
676 return ERR_PTR(ret);
677
678 if (ptrace_may_access(task, PTRACE_MODE_ATTACH_REALCREDS))
679 file = fget_task(task, fd);
680 else
681 file = ERR_PTR(-EPERM);
682
Eric W. Biedermanf7cfd872020-12-03 14:12:00 -0600683 up_read(&task->signal->exec_update_lock);
Sargun Dhillon8649c322020-01-07 09:59:25 -0800684
685 return file ?: ERR_PTR(-EBADF);
686}
687
688static int pidfd_getfd(struct pid *pid, int fd)
689{
690 struct task_struct *task;
691 struct file *file;
692 int ret;
693
694 task = get_pid_task(pid, PIDTYPE_PID);
695 if (!task)
696 return -ESRCH;
697
698 file = __pidfd_fget(task, fd);
699 put_task_struct(task);
700 if (IS_ERR(file))
701 return PTR_ERR(file);
702
Kees Cook910d2f12020-06-09 16:21:38 -0700703 ret = receive_fd(file, O_CLOEXEC);
704 fput(file);
Sargun Dhillon8649c322020-01-07 09:59:25 -0800705
706 return ret;
707}
708
709/**
710 * sys_pidfd_getfd() - Get a file descriptor from another process
711 *
712 * @pidfd: the pidfd file descriptor of the process
713 * @fd: the file descriptor number to get
714 * @flags: flags on how to get the fd (reserved)
715 *
716 * This syscall gets a copy of a file descriptor from another process
717 * based on the pidfd, and file descriptor number. It requires that
718 * the calling process has the ability to ptrace the process represented
719 * by the pidfd. The process which is having its file descriptor copied
720 * is otherwise unaffected.
721 *
722 * Return: On success, a cloexec file descriptor is returned.
723 * On error, a negative errno number will be returned.
724 */
725SYSCALL_DEFINE3(pidfd_getfd, int, pidfd, int, fd,
726 unsigned int, flags)
727{
728 struct pid *pid;
729 struct fd f;
730 int ret;
731
732 /* flags is currently unused - make sure it's unset */
733 if (flags)
734 return -EINVAL;
735
736 f = fdget(pidfd);
737 if (!f.file)
738 return -EBADF;
739
740 pid = pidfd_pid(f.file);
741 if (IS_ERR(pid))
742 ret = PTR_ERR(pid);
743 else
744 ret = pidfd_getfd(pid, fd);
745
746 fdput(f);
747 return ret;
748}