blob: 020dedbdf066bccbc370cba20be8dc7dbc914629 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Generic pidhash and scalable, time-bounded PID allocator
3 *
Nadia Yvette Chambers6d49e352012-12-06 10:39:54 +01004 * (C) 2002-2003 Nadia Yvette Chambers, IBM
5 * (C) 2004 Nadia Yvette Chambers, Oracle
Linus Torvalds1da177e2005-04-16 15:20:36 -07006 * (C) 2002-2004 Ingo Molnar, Red Hat
7 *
8 * pid-structures are backing objects for tasks sharing a given ID to chain
9 * against. There is very little to them aside from hashing them and
10 * parking tasks using given ID's on a list.
11 *
12 * The hash is always changed with the tasklist_lock write-acquired,
13 * and the hash is only accessed with the tasklist_lock at least
14 * read-acquired, so there's no additional SMP locking needed here.
15 *
16 * We have a list of bitmap pages, which bitmaps represent the PID space.
17 * Allocating and freeing PIDs is completely lockless. The worst-case
18 * allocation scenario when all but one out of 1 million PIDs possible are
19 * allocated already: the scanning of 32 list entries and at most PAGE_SIZE
20 * bytes. The typical fastpath is a single successful setbit. Freeing is O(1).
Pavel Emelyanov30e49c22007-10-18 23:40:10 -070021 *
22 * Pid namespaces:
23 * (C) 2007 Pavel Emelyanov <xemul@openvz.org>, OpenVZ, SWsoft Inc.
24 * (C) 2007 Sukadev Bhattiprolu <sukadev@us.ibm.com>, IBM
25 * Many thanks to Oleg Nesterov for comments and help
26 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070027 */
28
29#include <linux/mm.h>
Paul Gortmaker9984de12011-05-23 14:51:41 -040030#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include <linux/slab.h>
32#include <linux/init.h>
Franck Bui-Huu82524742008-05-12 21:21:05 +020033#include <linux/rculist.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include <linux/bootmem.h>
35#include <linux/hash.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>
Eric W. Biederman0a01f2c2012-08-01 10:33:47 -070040#include <linux/proc_fs.h>
Ingo Molnar29930022017-02-08 18:51:36 +010041#include <linux/sched/task.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -070043#define pid_hashfn(nr, ns) \
44 hash_long((unsigned long)nr + (unsigned long)ns, pidhash_shift)
Eric W. Biederman92476d72006-03-31 02:31:42 -080045static struct hlist_head *pid_hash;
Jan Beulich2c85f512009-09-21 17:03:07 -070046static unsigned int pidhash_shift = 4;
Sukadev Bhattiprolu820e45d2007-05-10 22:23:00 -070047struct pid init_struct_pid = INIT_STRUCT_PID;
Linus Torvalds1da177e2005-04-16 15:20:36 -070048
49int pid_max = PID_MAX_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -070050
51#define RESERVED_PIDS 300
52
53int pid_max_min = RESERVED_PIDS + 1;
54int pid_max_max = PID_MAX_LIMIT;
55
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -080056static inline int mk_pid(struct pid_namespace *pid_ns,
57 struct pidmap *map, int off)
Sukadev Bhattiprolu3fbc9642006-10-02 02:17:24 -070058{
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -080059 return (map - pid_ns->pidmap)*BITS_PER_PAGE + off;
Sukadev Bhattiprolu3fbc9642006-10-02 02:17:24 -070060}
61
Linus Torvalds1da177e2005-04-16 15:20:36 -070062#define find_next_offset(map, off) \
63 find_next_zero_bit((map)->page, BITS_PER_PAGE, off)
64
65/*
66 * PID-map pages start out as NULL, they get allocated upon
67 * first use and are never deallocated. This way a low pid_max
68 * value does not cause lots of bitmaps to be allocated, but
69 * the scheme scales to up to 4 million PIDs, runtime.
70 */
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -080071struct pid_namespace init_pid_ns = {
Peter Zijlstra1e24edc2016-11-14 17:12:23 +010072 .kref = KREF_INIT(2),
Sukadev Bhattiprolu3fbc9642006-10-02 02:17:24 -070073 .pidmap = {
74 [ 0 ... PIDMAP_ENTRIES-1] = { ATOMIC_INIT(BITS_PER_PAGE), NULL }
75 },
Sukadev Bhattiprolu84d73782006-12-08 02:38:01 -080076 .last_pid = 0,
Raphael S. Carvalho8f75af42013-07-03 15:09:02 -070077 .nr_hashed = PIDNS_HASH_ADDING,
Pavel Emelyanovfaacbfd2007-10-18 23:40:04 -070078 .level = 0,
79 .child_reaper = &init_task,
Eric W. Biederman49f4d8b2012-08-02 04:25:10 -070080 .user_ns = &init_user_ns,
Al Viro435d5f42014-10-31 22:56:04 -040081 .ns.inum = PROC_PID_INIT_INO,
Al Viro33c42942014-11-01 02:32:53 -040082#ifdef CONFIG_PID_NS
83 .ns.ops = &pidns_operations,
84#endif
Sukadev Bhattiprolu3fbc9642006-10-02 02:17:24 -070085};
Pavel Emelyanov198fe212007-10-18 23:40:06 -070086EXPORT_SYMBOL_GPL(init_pid_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -070087
Eric W. Biederman92476d72006-03-31 02:31:42 -080088/*
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 */
Sukadev Bhattiprolu3fbc9642006-10-02 02:17:24 -0700101
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102static __cacheline_aligned_in_smp DEFINE_SPINLOCK(pidmap_lock);
103
Oleg Nesterovb7127aa2008-04-30 00:54:22 -0700104static void free_pidmap(struct upid *upid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105{
Oleg Nesterovb7127aa2008-04-30 00:54:22 -0700106 int nr = upid->nr;
107 struct pidmap *map = upid->ns->pidmap + nr / BITS_PER_PAGE;
108 int offset = nr & BITS_PER_PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109
110 clear_bit(offset, map->page);
111 atomic_inc(&map->nr_free);
112}
113
Salman5fdee8c2010-08-10 18:03:16 -0700114/*
115 * If we started walking pids at 'base', is 'a' seen before 'b'?
116 */
117static int pid_before(int base, int a, int b)
118{
119 /*
120 * This is the same as saying
121 *
122 * (a - base + MAXUINT) % MAXUINT < (b - base + MAXUINT) % MAXUINT
123 * and that mapping orders 'a' and 'b' with respect to 'base'.
124 */
125 return (unsigned)(a - base) < (unsigned)(b - base);
126}
127
128/*
Pavel Emelyanovb8f566b2012-01-12 17:20:27 -0800129 * We might be racing with someone else trying to set pid_ns->last_pid
130 * at the pid allocation time (there's also a sysctl for this, but racing
131 * with this one is OK, see comment in kernel/pid_namespace.c about it).
Salman5fdee8c2010-08-10 18:03:16 -0700132 * We want the winner to have the "later" value, because if the
133 * "earlier" value prevails, then a pid may get reused immediately.
134 *
135 * Since pids rollover, it is not sufficient to just pick the bigger
136 * value. We have to consider where we started counting from.
137 *
138 * 'base' is the value of pid_ns->last_pid that we observed when
139 * we started looking for a pid.
140 *
141 * 'pid' is the pid that we eventually found.
142 */
143static void set_last_pid(struct pid_namespace *pid_ns, int base, int pid)
144{
145 int prev;
146 int last_write = base;
147 do {
148 prev = last_write;
149 last_write = cmpxchg(&pid_ns->last_pid, prev, pid);
150 } while ((prev != last_write) && (pid_before(base, last_write, pid)));
151}
152
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -0800153static int alloc_pidmap(struct pid_namespace *pid_ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154{
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -0800155 int i, offset, max_scan, pid, last = pid_ns->last_pid;
Sukadev Bhattiprolu6a1f3b82006-10-02 02:17:20 -0700156 struct pidmap *map;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157
158 pid = last + 1;
159 if (pid >= pid_max)
160 pid = RESERVED_PIDS;
161 offset = pid & BITS_PER_PAGE_MASK;
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -0800162 map = &pid_ns->pidmap[pid/BITS_PER_PAGE];
Oleg Nesterovc52b0b92010-08-10 18:03:17 -0700163 /*
164 * If last_pid points into the middle of the map->page we
165 * want to scan this bitmap block twice, the second time
166 * we start with offset == 0 (or RESERVED_PIDS).
167 */
168 max_scan = DIV_ROUND_UP(pid_max, BITS_PER_PAGE) - !offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169 for (i = 0; i <= max_scan; ++i) {
170 if (unlikely(!map->page)) {
Sukadev Bhattiprolu3fbc9642006-10-02 02:17:24 -0700171 void *page = kzalloc(PAGE_SIZE, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 /*
173 * Free the page if someone raced with us
174 * installing it:
175 */
Eric W. Biederman92476d72006-03-31 02:31:42 -0800176 spin_lock_irq(&pidmap_lock);
André Goddard Rosa7be6d992009-12-15 16:47:39 -0800177 if (!map->page) {
Sukadev Bhattiprolu3fbc9642006-10-02 02:17:24 -0700178 map->page = page;
André Goddard Rosa7be6d992009-12-15 16:47:39 -0800179 page = NULL;
180 }
Eric W. Biederman92476d72006-03-31 02:31:42 -0800181 spin_unlock_irq(&pidmap_lock);
André Goddard Rosa7be6d992009-12-15 16:47:39 -0800182 kfree(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183 if (unlikely(!map->page))
Michal Hocko35f71bc2015-04-16 12:47:38 -0700184 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 }
186 if (likely(atomic_read(&map->nr_free))) {
Raphael S. Carvalho8db049b2013-04-30 15:28:26 -0700187 for ( ; ; ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188 if (!test_and_set_bit(offset, map->page)) {
189 atomic_dec(&map->nr_free);
Salman5fdee8c2010-08-10 18:03:16 -0700190 set_last_pid(pid_ns, last, pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 return pid;
192 }
193 offset = find_next_offset(map, offset);
Raphael S. Carvalho8db049b2013-04-30 15:28:26 -0700194 if (offset >= BITS_PER_PAGE)
195 break;
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -0800196 pid = mk_pid(pid_ns, map, offset);
Raphael S. Carvalho8db049b2013-04-30 15:28:26 -0700197 if (pid >= pid_max)
198 break;
199 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 }
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -0800201 if (map < &pid_ns->pidmap[(pid_max-1)/BITS_PER_PAGE]) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 ++map;
203 offset = 0;
204 } else {
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -0800205 map = &pid_ns->pidmap[0];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 offset = RESERVED_PIDS;
207 if (unlikely(last == offset))
208 break;
209 }
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -0800210 pid = mk_pid(pid_ns, map, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211 }
Michal Hocko35f71bc2015-04-16 12:47:38 -0700212 return -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213}
214
Linus Torvaldsc78193e2011-04-18 10:35:30 -0700215int next_pidmap(struct pid_namespace *pid_ns, unsigned int last)
Eric W. Biederman0804ef42006-10-02 02:17:04 -0700216{
217 int offset;
Eric W. Biedermanf40f50d2006-10-02 02:17:25 -0700218 struct pidmap *map, *end;
Eric W. Biederman0804ef42006-10-02 02:17:04 -0700219
Linus Torvaldsc78193e2011-04-18 10:35:30 -0700220 if (last >= PID_MAX_LIMIT)
221 return -1;
222
Eric W. Biederman0804ef42006-10-02 02:17:04 -0700223 offset = (last + 1) & BITS_PER_PAGE_MASK;
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -0800224 map = &pid_ns->pidmap[(last + 1)/BITS_PER_PAGE];
225 end = &pid_ns->pidmap[PIDMAP_ENTRIES];
Eric W. Biedermanf40f50d2006-10-02 02:17:25 -0700226 for (; map < end; map++, offset = 0) {
Eric W. Biederman0804ef42006-10-02 02:17:04 -0700227 if (unlikely(!map->page))
228 continue;
229 offset = find_next_bit((map)->page, BITS_PER_PAGE, offset);
230 if (offset < BITS_PER_PAGE)
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -0800231 return mk_pid(pid_ns, map, offset);
Eric W. Biederman0804ef42006-10-02 02:17:04 -0700232 }
233 return -1;
234}
235
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -0800236void put_pid(struct pid *pid)
Eric W. Biederman92476d72006-03-31 02:31:42 -0800237{
Pavel Emelianovbaf8f0f2007-10-18 23:39:48 -0700238 struct pid_namespace *ns;
239
Eric W. Biederman92476d72006-03-31 02:31:42 -0800240 if (!pid)
241 return;
Pavel Emelianovbaf8f0f2007-10-18 23:39:48 -0700242
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -0700243 ns = pid->numbers[pid->level].ns;
Eric W. Biederman92476d72006-03-31 02:31:42 -0800244 if ((atomic_read(&pid->count) == 1) ||
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -0700245 atomic_dec_and_test(&pid->count)) {
Pavel Emelianovbaf8f0f2007-10-18 23:39:48 -0700246 kmem_cache_free(ns->pid_cachep, pid);
Pavel Emelyanovb461cc02007-10-18 23:40:09 -0700247 put_pid_ns(ns);
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -0700248 }
Eric W. Biederman92476d72006-03-31 02:31:42 -0800249}
Eric W. Biedermanbbf73142006-10-02 02:17:11 -0700250EXPORT_SYMBOL_GPL(put_pid);
Eric W. Biederman92476d72006-03-31 02:31:42 -0800251
252static void delayed_put_pid(struct rcu_head *rhp)
253{
254 struct pid *pid = container_of(rhp, struct pid, rcu);
255 put_pid(pid);
256}
257
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -0800258void free_pid(struct pid *pid)
Eric W. Biederman92476d72006-03-31 02:31:42 -0800259{
260 /* We can be called with write_lock_irq(&tasklist_lock) held */
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -0700261 int i;
Eric W. Biederman92476d72006-03-31 02:31:42 -0800262 unsigned long flags;
263
264 spin_lock_irqsave(&pidmap_lock, flags);
Eric W. Biederman0a01f2c2012-08-01 10:33:47 -0700265 for (i = 0; i <= pid->level; i++) {
266 struct upid *upid = pid->numbers + i;
Eric W. Biedermanaf4b8a82012-08-01 15:03:42 -0700267 struct pid_namespace *ns = upid->ns;
Eric W. Biederman0a01f2c2012-08-01 10:33:47 -0700268 hlist_del_rcu(&upid->pid_chain);
Eric W. Biedermanaf4b8a82012-08-01 15:03:42 -0700269 switch(--ns->nr_hashed) {
Eric W. Biedermana6064882013-08-29 13:56:50 -0700270 case 2:
Eric W. Biedermanaf4b8a82012-08-01 15:03:42 -0700271 case 1:
272 /* When all that is left in the pid namespace
273 * is the reaper wake up the reaper. The reaper
274 * may be sleeping in zap_pid_ns_processes().
275 */
276 wake_up_process(ns->child_reaper);
277 break;
Oleg Nesterov314a8ad2013-09-30 13:45:27 -0700278 case PIDNS_HASH_ADDING:
279 /* Handle a fork failure of the first process */
280 WARN_ON(ns->child_reaper);
281 ns->nr_hashed = 0;
282 /* fall through */
Eric W. Biedermanaf4b8a82012-08-01 15:03:42 -0700283 case 0:
Eric W. Biedermanaf4b8a82012-08-01 15:03:42 -0700284 schedule_work(&ns->proc_work);
285 break;
Eric W. Biederman5e1182deb2010-07-12 18:50:25 -0700286 }
Eric W. Biederman0a01f2c2012-08-01 10:33:47 -0700287 }
Eric W. Biederman92476d72006-03-31 02:31:42 -0800288 spin_unlock_irqrestore(&pidmap_lock, flags);
289
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -0700290 for (i = 0; i <= pid->level; i++)
Oleg Nesterovb7127aa2008-04-30 00:54:22 -0700291 free_pidmap(pid->numbers + i);
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -0700292
Eric W. Biederman92476d72006-03-31 02:31:42 -0800293 call_rcu(&pid->rcu, delayed_put_pid);
294}
295
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -0700296struct pid *alloc_pid(struct pid_namespace *ns)
Eric W. Biederman92476d72006-03-31 02:31:42 -0800297{
298 struct pid *pid;
299 enum pid_type type;
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -0700300 int i, nr;
301 struct pid_namespace *tmp;
Pavel Emelyanov198fe212007-10-18 23:40:06 -0700302 struct upid *upid;
Michal Hocko35f71bc2015-04-16 12:47:38 -0700303 int retval = -ENOMEM;
Eric W. Biederman92476d72006-03-31 02:31:42 -0800304
Pavel Emelianovbaf8f0f2007-10-18 23:39:48 -0700305 pid = kmem_cache_alloc(ns->pid_cachep, GFP_KERNEL);
Eric W. Biederman92476d72006-03-31 02:31:42 -0800306 if (!pid)
Michal Hocko35f71bc2015-04-16 12:47:38 -0700307 return ERR_PTR(retval);
Eric W. Biederman92476d72006-03-31 02:31:42 -0800308
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -0700309 tmp = ns;
Eric W. Biederman0a01f2c2012-08-01 10:33:47 -0700310 pid->level = ns->level;
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -0700311 for (i = ns->level; i >= 0; i--) {
312 nr = alloc_pidmap(tmp);
Arnd Bergmann287980e2016-05-27 23:23:25 +0200313 if (nr < 0) {
Michal Hocko35f71bc2015-04-16 12:47:38 -0700314 retval = nr;
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -0700315 goto out_free;
Michal Hocko35f71bc2015-04-16 12:47:38 -0700316 }
Eric W. Biederman92476d72006-03-31 02:31:42 -0800317
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -0700318 pid->numbers[i].nr = nr;
319 pid->numbers[i].ns = tmp;
320 tmp = tmp->parent;
321 }
322
Eric W. Biederman0a01f2c2012-08-01 10:33:47 -0700323 if (unlikely(is_child_reaper(pid))) {
Kirill Tkhai8896c232017-05-08 15:56:34 -0700324 if (pid_ns_prepare_proc(ns)) {
325 disable_pid_allocation(ns);
Eric W. Biederman0a01f2c2012-08-01 10:33:47 -0700326 goto out_free;
Kirill Tkhai8896c232017-05-08 15:56:34 -0700327 }
Eric W. Biederman0a01f2c2012-08-01 10:33:47 -0700328 }
329
Pavel Emelyanovb461cc02007-10-18 23:40:09 -0700330 get_pid_ns(ns);
Eric W. Biederman92476d72006-03-31 02:31:42 -0800331 atomic_set(&pid->count, 1);
Eric W. Biederman92476d72006-03-31 02:31:42 -0800332 for (type = 0; type < PIDTYPE_MAX; ++type)
333 INIT_HLIST_HEAD(&pid->tasks[type]);
334
André Goddard Rosa417e3152009-12-15 16:47:40 -0800335 upid = pid->numbers + ns->level;
Eric W. Biederman92476d72006-03-31 02:31:42 -0800336 spin_lock_irq(&pidmap_lock);
Eric W. Biedermanc876ad762012-12-21 20:27:12 -0800337 if (!(ns->nr_hashed & PIDNS_HASH_ADDING))
Eric W. Biederman5e1182deb2010-07-12 18:50:25 -0700338 goto out_unlock;
Eric W. Biederman0a01f2c2012-08-01 10:33:47 -0700339 for ( ; upid >= pid->numbers; --upid) {
Pavel Emelyanov198fe212007-10-18 23:40:06 -0700340 hlist_add_head_rcu(&upid->pid_chain,
341 &pid_hash[pid_hashfn(upid->nr, upid->ns)]);
Eric W. Biederman0a01f2c2012-08-01 10:33:47 -0700342 upid->ns->nr_hashed++;
343 }
Eric W. Biederman92476d72006-03-31 02:31:42 -0800344 spin_unlock_irq(&pidmap_lock);
345
Eric W. Biederman92476d72006-03-31 02:31:42 -0800346 return pid;
347
Eric W. Biederman5e1182deb2010-07-12 18:50:25 -0700348out_unlock:
Eric W. Biederman6e666882013-02-12 13:46:23 -0800349 spin_unlock_irq(&pidmap_lock);
Oleg Nesterov24c037e2014-12-10 15:55:25 -0800350 put_pid_ns(ns);
351
Eric W. Biederman92476d72006-03-31 02:31:42 -0800352out_free:
Oleg Nesterovb7127aa2008-04-30 00:54:22 -0700353 while (++i <= ns->level)
354 free_pidmap(pid->numbers + i);
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -0700355
Pavel Emelianovbaf8f0f2007-10-18 23:39:48 -0700356 kmem_cache_free(ns->pid_cachep, pid);
Michal Hocko35f71bc2015-04-16 12:47:38 -0700357 return ERR_PTR(retval);
Eric W. Biederman92476d72006-03-31 02:31:42 -0800358}
359
Eric W. Biedermanc876ad762012-12-21 20:27:12 -0800360void disable_pid_allocation(struct pid_namespace *ns)
361{
362 spin_lock_irq(&pidmap_lock);
363 ns->nr_hashed &= ~PIDNS_HASH_ADDING;
364 spin_unlock_irq(&pidmap_lock);
365}
366
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -0800367struct pid *find_pid_ns(int nr, struct pid_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368{
Pavel Emelyanov198fe212007-10-18 23:40:06 -0700369 struct upid *pnr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370
Sasha Levinb67bfe02013-02-27 17:06:00 -0800371 hlist_for_each_entry_rcu(pnr,
Pavel Emelyanov198fe212007-10-18 23:40:06 -0700372 &pid_hash[pid_hashfn(nr, ns)], pid_chain)
373 if (pnr->nr == nr && pnr->ns == ns)
374 return container_of(pnr, struct pid,
375 numbers[ns->level]);
376
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377 return NULL;
378}
Pavel Emelyanov198fe212007-10-18 23:40:06 -0700379EXPORT_SYMBOL_GPL(find_pid_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380
Pavel Emelyanov89905712007-10-18 23:40:19 -0700381struct pid *find_vpid(int nr)
382{
Eric W. Biederman17cf22c2010-03-02 14:51:53 -0800383 return find_pid_ns(nr, task_active_pid_ns(current));
Pavel Emelyanov89905712007-10-18 23:40:19 -0700384}
385EXPORT_SYMBOL_GPL(find_vpid);
386
Sukadev Bhattiprolue713d0d2007-05-10 22:22:58 -0700387/*
388 * attach_pid() must be called with the tasklist_lock write-held.
389 */
Oleg Nesterov81907732013-07-03 15:08:31 -0700390void attach_pid(struct task_struct *task, enum pid_type type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391{
Oleg Nesterov81907732013-07-03 15:08:31 -0700392 struct pid_link *link = &task->pids[type];
393 hlist_add_head_rcu(&link->node, &link->pid->tasks[type]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394}
395
Oleg Nesterov24336ea2008-04-30 00:54:26 -0700396static void __change_pid(struct task_struct *task, enum pid_type type,
397 struct pid *new)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398{
Eric W. Biederman92476d72006-03-31 02:31:42 -0800399 struct pid_link *link;
400 struct pid *pid;
401 int tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402
Eric W. Biederman92476d72006-03-31 02:31:42 -0800403 link = &task->pids[type];
404 pid = link->pid;
405
406 hlist_del_rcu(&link->node);
Oleg Nesterov24336ea2008-04-30 00:54:26 -0700407 link->pid = new;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408
409 for (tmp = PIDTYPE_MAX; --tmp >= 0; )
Eric W. Biederman92476d72006-03-31 02:31:42 -0800410 if (!hlist_empty(&pid->tasks[tmp]))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 return;
412
Eric W. Biederman92476d72006-03-31 02:31:42 -0800413 free_pid(pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414}
415
Oleg Nesterov24336ea2008-04-30 00:54:26 -0700416void detach_pid(struct task_struct *task, enum pid_type type)
417{
418 __change_pid(task, type, NULL);
419}
420
421void change_pid(struct task_struct *task, enum pid_type type,
422 struct pid *pid)
423{
424 __change_pid(task, type, pid);
Oleg Nesterov81907732013-07-03 15:08:31 -0700425 attach_pid(task, type);
Oleg Nesterov24336ea2008-04-30 00:54:26 -0700426}
427
Eric W. Biedermanc18258c2006-09-27 01:51:06 -0700428/* transfer_pid is an optimization of attach_pid(new), detach_pid(old) */
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -0800429void transfer_pid(struct task_struct *old, struct task_struct *new,
Eric W. Biedermanc18258c2006-09-27 01:51:06 -0700430 enum pid_type type)
431{
432 new->pids[type].pid = old->pids[type].pid;
433 hlist_replace_rcu(&old->pids[type].node, &new->pids[type].node);
Eric W. Biedermanc18258c2006-09-27 01:51:06 -0700434}
435
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -0800436struct task_struct *pid_task(struct pid *pid, enum pid_type type)
Eric W. Biederman92476d72006-03-31 02:31:42 -0800437{
438 struct task_struct *result = NULL;
439 if (pid) {
440 struct hlist_node *first;
Arnd Bergmann67bdbff2010-02-25 16:55:13 +0100441 first = rcu_dereference_check(hlist_first_rcu(&pid->tasks[type]),
Paul E. McKenneydb1466b2010-03-03 07:46:56 -0800442 lockdep_tasklist_lock_is_held());
Eric W. Biederman92476d72006-03-31 02:31:42 -0800443 if (first)
444 result = hlist_entry(first, struct task_struct, pids[(type)].node);
445 }
446 return result;
447}
Pavel Emelyanoveccba062008-02-07 00:13:21 -0800448EXPORT_SYMBOL(pid_task);
Eric W. Biederman92476d72006-03-31 02:31:42 -0800449
450/*
Tetsuo Handa9728e5d2010-03-05 13:42:56 -0800451 * Must be called under rcu_read_lock().
Eric W. Biederman92476d72006-03-31 02:31:42 -0800452 */
Christoph Hellwig17f98dc2009-06-17 16:27:51 -0700453struct task_struct *find_task_by_pid_ns(pid_t nr, struct pid_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454{
Paul E. McKenneyf78f5b92015-06-18 15:50:02 -0700455 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
456 "find_task_by_pid_ns() needs rcu_read_lock() protection");
Christoph Hellwig17f98dc2009-06-17 16:27:51 -0700457 return pid_task(find_pid_ns(nr, ns), PIDTYPE_PID);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458}
459
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -0700460struct task_struct *find_task_by_vpid(pid_t vnr)
461{
Eric W. Biederman17cf22c2010-03-02 14:51:53 -0800462 return find_task_by_pid_ns(vnr, task_active_pid_ns(current));
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -0700463}
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -0700464
Oleg Nesterov1a657f782006-10-02 02:18:59 -0700465struct pid *get_task_pid(struct task_struct *task, enum pid_type type)
466{
467 struct pid *pid;
468 rcu_read_lock();
Oleg Nesterov2ae448e2009-04-02 16:58:36 -0700469 if (type != PIDTYPE_PID)
470 task = task->group_leader;
Eric Dumazet81b1a832015-11-24 11:39:54 -0800471 pid = get_pid(rcu_dereference(task->pids[type].pid));
Oleg Nesterov1a657f782006-10-02 02:18:59 -0700472 rcu_read_unlock();
473 return pid;
474}
Rik van Riel77c100c2011-02-01 09:51:46 -0500475EXPORT_SYMBOL_GPL(get_task_pid);
Oleg Nesterov1a657f782006-10-02 02:18:59 -0700476
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -0800477struct task_struct *get_pid_task(struct pid *pid, enum pid_type type)
Eric W. Biederman92476d72006-03-31 02:31:42 -0800478{
479 struct task_struct *result;
480 rcu_read_lock();
481 result = pid_task(pid, type);
482 if (result)
483 get_task_struct(result);
484 rcu_read_unlock();
485 return result;
486}
Rik van Riel77c100c2011-02-01 09:51:46 -0500487EXPORT_SYMBOL_GPL(get_pid_task);
Eric W. Biederman92476d72006-03-31 02:31:42 -0800488
489struct pid *find_get_pid(pid_t nr)
490{
491 struct pid *pid;
492
493 rcu_read_lock();
Pavel Emelyanov198fe212007-10-18 23:40:06 -0700494 pid = get_pid(find_vpid(nr));
Eric W. Biederman92476d72006-03-31 02:31:42 -0800495 rcu_read_unlock();
496
497 return pid;
498}
David Sterba339caf22008-07-25 01:48:31 -0700499EXPORT_SYMBOL_GPL(find_get_pid);
Eric W. Biederman92476d72006-03-31 02:31:42 -0800500
Pavel Emelyanov7af57292007-10-18 23:40:06 -0700501pid_t pid_nr_ns(struct pid *pid, struct pid_namespace *ns)
502{
503 struct upid *upid;
504 pid_t nr = 0;
505
506 if (pid && ns->level <= pid->level) {
507 upid = &pid->numbers[ns->level];
508 if (upid->ns == ns)
509 nr = upid->nr;
510 }
511 return nr;
512}
Eric W. Biederman4f82f452012-05-24 10:37:59 -0600513EXPORT_SYMBOL_GPL(pid_nr_ns);
Pavel Emelyanov7af57292007-10-18 23:40:06 -0700514
Eric W. Biederman44c4e1b2008-02-08 04:19:15 -0800515pid_t pid_vnr(struct pid *pid)
516{
Eric W. Biederman17cf22c2010-03-02 14:51:53 -0800517 return pid_nr_ns(pid, task_active_pid_ns(current));
Eric W. Biederman44c4e1b2008-02-08 04:19:15 -0800518}
519EXPORT_SYMBOL_GPL(pid_vnr);
520
Oleg Nesterov52ee2df2009-04-02 16:58:38 -0700521pid_t __task_pid_nr_ns(struct task_struct *task, enum pid_type type,
522 struct pid_namespace *ns)
Pavel Emelyanov2f2a3a42007-10-18 23:40:19 -0700523{
Oleg Nesterov52ee2df2009-04-02 16:58:38 -0700524 pid_t nr = 0;
525
526 rcu_read_lock();
527 if (!ns)
Eric W. Biederman17cf22c2010-03-02 14:51:53 -0800528 ns = task_active_pid_ns(current);
Oleg Nesterov52ee2df2009-04-02 16:58:38 -0700529 if (likely(pid_alive(task))) {
Oleg Nesterovdd1c1f22017-08-21 17:35:02 +0200530 if (type != PIDTYPE_PID) {
531 if (type == __PIDTYPE_TGID)
532 type = PIDTYPE_PID;
Oleg Nesterov52ee2df2009-04-02 16:58:38 -0700533 task = task->group_leader;
Oleg Nesterovdd1c1f22017-08-21 17:35:02 +0200534 }
Eric Dumazet81b1a832015-11-24 11:39:54 -0800535 nr = pid_nr_ns(rcu_dereference(task->pids[type].pid), ns);
Oleg Nesterov52ee2df2009-04-02 16:58:38 -0700536 }
537 rcu_read_unlock();
538
539 return nr;
Pavel Emelyanov2f2a3a42007-10-18 23:40:19 -0700540}
Oleg Nesterov52ee2df2009-04-02 16:58:38 -0700541EXPORT_SYMBOL(__task_pid_nr_ns);
Pavel Emelyanov2f2a3a42007-10-18 23:40:19 -0700542
Eric W. Biederman61bce0f2009-01-07 18:08:49 -0800543struct pid_namespace *task_active_pid_ns(struct task_struct *tsk)
544{
545 return ns_of_pid(task_pid(tsk));
546}
547EXPORT_SYMBOL_GPL(task_active_pid_ns);
548
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549/*
Frederik Schwarzer025dfda2008-10-16 19:02:37 +0200550 * Used by proc to find the first pid that is greater than or equal to nr.
Eric W. Biederman0804ef42006-10-02 02:17:04 -0700551 *
Pavel Emelyanove49859e2008-07-25 01:48:36 -0700552 * 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 -0700553 */
Pavel Emelyanov198fe212007-10-18 23:40:06 -0700554struct pid *find_ge_pid(int nr, struct pid_namespace *ns)
Eric W. Biederman0804ef42006-10-02 02:17:04 -0700555{
556 struct pid *pid;
557
558 do {
Pavel Emelyanov198fe212007-10-18 23:40:06 -0700559 pid = find_pid_ns(nr, ns);
Eric W. Biederman0804ef42006-10-02 02:17:04 -0700560 if (pid)
561 break;
Pavel Emelyanov198fe212007-10-18 23:40:06 -0700562 nr = next_pidmap(ns, nr);
Eric W. Biederman0804ef42006-10-02 02:17:04 -0700563 } while (nr > 0);
564
565 return pid;
566}
567
568/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 * The pid hash table is scaled according to the amount of memory in the
570 * machine. From a minimum of 16 slots up to 4096 slots at one gigabyte or
571 * more.
572 */
573void __init pidhash_init(void)
574{
Jan Beulich2c85f512009-09-21 17:03:07 -0700575 pid_hash = alloc_large_system_hash("PID", sizeof(*pid_hash), 0, 18,
Pavel Tatashin3d375d72017-07-06 15:39:11 -0700576 HASH_EARLY | HASH_SMALL | HASH_ZERO,
Tim Bird31fe62b2012-05-23 13:33:35 +0000577 &pidhash_shift, NULL,
578 0, 4096);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579}
580
581void __init pidmap_init(void)
582{
Zhen Lei840d6fe2016-01-30 10:04:17 +0800583 /* Verify no one has done anything silly: */
Eric W. Biedermanc876ad762012-12-21 20:27:12 -0800584 BUILD_BUG_ON(PID_MAX_LIMIT >= PIDNS_HASH_ADDING);
585
Hedi Berriche72680a12010-05-26 14:44:06 -0700586 /* bump default and minimum pid_max based on number of cpus */
587 pid_max = min(pid_max_max, max_t(int, pid_max,
588 PIDS_PER_CPU_DEFAULT * num_possible_cpus()));
589 pid_max_min = max_t(int, pid_max_min,
590 PIDS_PER_CPU_MIN * num_possible_cpus());
591 pr_info("pid_max: default: %u minimum: %u\n", pid_max, pid_max_min);
592
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -0800593 init_pid_ns.pidmap[0].page = kzalloc(PAGE_SIZE, GFP_KERNEL);
Oleg Nesterov73b9ebf2006-03-28 16:11:07 -0800594 /* Reserve PID 0. We never call free_pidmap(0) */
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -0800595 set_bit(0, init_pid_ns.pidmap[0].page);
596 atomic_dec(&init_pid_ns.pidmap[0].nr_free);
Eric W. Biederman92476d72006-03-31 02:31:42 -0800597
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800598 init_pid_ns.pid_cachep = KMEM_CACHE(pid,
Vladimir Davydov5d097052016-01-14 15:18:21 -0800599 SLAB_HWCACHE_ALIGN | SLAB_PANIC | SLAB_ACCOUNT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600}