blob: cd36a5e0d173062dfbdd557fa7ad5234350af2ed [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -070042#define pid_hashfn(nr, ns) \
43 hash_long((unsigned long)nr + (unsigned long)ns, pidhash_shift)
Eric W. Biederman92476d72006-03-31 02:31:42 -080044static struct hlist_head *pid_hash;
Jan Beulich2c85f512009-09-21 17:03:07 -070045static unsigned int pidhash_shift = 4;
Sukadev Bhattiprolu820e45d2007-05-10 22:23:00 -070046struct pid init_struct_pid = INIT_STRUCT_PID;
Linus Torvalds1da177e2005-04-16 15:20:36 -070047
48int pid_max = PID_MAX_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -070049
50#define RESERVED_PIDS 300
51
52int pid_max_min = RESERVED_PIDS + 1;
53int pid_max_max = PID_MAX_LIMIT;
54
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -080055static inline int mk_pid(struct pid_namespace *pid_ns,
56 struct pidmap *map, int off)
Sukadev Bhattiprolu3fbc9642006-10-02 02:17:24 -070057{
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -080058 return (map - pid_ns->pidmap)*BITS_PER_PAGE + off;
Sukadev Bhattiprolu3fbc9642006-10-02 02:17:24 -070059}
60
Linus Torvalds1da177e2005-04-16 15:20:36 -070061#define find_next_offset(map, off) \
62 find_next_zero_bit((map)->page, BITS_PER_PAGE, off)
63
64/*
65 * PID-map pages start out as NULL, they get allocated upon
66 * first use and are never deallocated. This way a low pid_max
67 * value does not cause lots of bitmaps to be allocated, but
68 * the scheme scales to up to 4 million PIDs, runtime.
69 */
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -080070struct pid_namespace init_pid_ns = {
Cedric Le Goater9a575a92006-12-08 02:37:59 -080071 .kref = {
72 .refcount = ATOMIC_INIT(2),
73 },
Sukadev Bhattiprolu3fbc9642006-10-02 02:17:24 -070074 .pidmap = {
75 [ 0 ... PIDMAP_ENTRIES-1] = { ATOMIC_INIT(BITS_PER_PAGE), NULL }
76 },
Sukadev Bhattiprolu84d73782006-12-08 02:38:01 -080077 .last_pid = 0,
Raphael S. Carvalho8f75af42013-07-03 15:09:02 -070078 .nr_hashed = PIDNS_HASH_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
Sukadev Bhattiprolu3fbc9642006-10-02 02:17:24 -070086};
Pavel Emelyanov198fe212007-10-18 23:40:06 -070087EXPORT_SYMBOL_GPL(init_pid_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -070088
Eric W. Biederman92476d72006-03-31 02:31:42 -080089/*
90 * Note: disable interrupts while the pidmap_lock is held as an
91 * interrupt might come in and do read_lock(&tasklist_lock).
92 *
93 * If we don't disable interrupts there is a nasty deadlock between
94 * detach_pid()->free_pid() and another cpu that does
95 * spin_lock(&pidmap_lock) followed by an interrupt routine that does
96 * read_lock(&tasklist_lock);
97 *
98 * After we clean up the tasklist_lock and know there are no
99 * irq handlers that take it we can leave the interrupts enabled.
100 * For now it is easier to be safe than to prove it can't happen.
101 */
Sukadev Bhattiprolu3fbc9642006-10-02 02:17:24 -0700102
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103static __cacheline_aligned_in_smp DEFINE_SPINLOCK(pidmap_lock);
104
Oleg Nesterovb7127aa2008-04-30 00:54:22 -0700105static void free_pidmap(struct upid *upid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106{
Oleg Nesterovb7127aa2008-04-30 00:54:22 -0700107 int nr = upid->nr;
108 struct pidmap *map = upid->ns->pidmap + nr / BITS_PER_PAGE;
109 int offset = nr & BITS_PER_PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110
111 clear_bit(offset, map->page);
112 atomic_inc(&map->nr_free);
113}
114
Salman5fdee8c2010-08-10 18:03:16 -0700115/*
116 * If we started walking pids at 'base', is 'a' seen before 'b'?
117 */
118static int pid_before(int base, int a, int b)
119{
120 /*
121 * This is the same as saying
122 *
123 * (a - base + MAXUINT) % MAXUINT < (b - base + MAXUINT) % MAXUINT
124 * and that mapping orders 'a' and 'b' with respect to 'base'.
125 */
126 return (unsigned)(a - base) < (unsigned)(b - base);
127}
128
129/*
Pavel Emelyanovb8f566b2012-01-12 17:20:27 -0800130 * We might be racing with someone else trying to set pid_ns->last_pid
131 * at the pid allocation time (there's also a sysctl for this, but racing
132 * with this one is OK, see comment in kernel/pid_namespace.c about it).
Salman5fdee8c2010-08-10 18:03:16 -0700133 * We want the winner to have the "later" value, because if the
134 * "earlier" value prevails, then a pid may get reused immediately.
135 *
136 * Since pids rollover, it is not sufficient to just pick the bigger
137 * value. We have to consider where we started counting from.
138 *
139 * 'base' is the value of pid_ns->last_pid that we observed when
140 * we started looking for a pid.
141 *
142 * 'pid' is the pid that we eventually found.
143 */
144static void set_last_pid(struct pid_namespace *pid_ns, int base, int pid)
145{
146 int prev;
147 int last_write = base;
148 do {
149 prev = last_write;
150 last_write = cmpxchg(&pid_ns->last_pid, prev, pid);
151 } while ((prev != last_write) && (pid_before(base, last_write, pid)));
152}
153
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -0800154static int alloc_pidmap(struct pid_namespace *pid_ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155{
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -0800156 int i, offset, max_scan, pid, last = pid_ns->last_pid;
Sukadev Bhattiprolu6a1f3b82006-10-02 02:17:20 -0700157 struct pidmap *map;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158
159 pid = last + 1;
160 if (pid >= pid_max)
161 pid = RESERVED_PIDS;
162 offset = pid & BITS_PER_PAGE_MASK;
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -0800163 map = &pid_ns->pidmap[pid/BITS_PER_PAGE];
Oleg Nesterovc52b0b92010-08-10 18:03:17 -0700164 /*
165 * If last_pid points into the middle of the map->page we
166 * want to scan this bitmap block twice, the second time
167 * we start with offset == 0 (or RESERVED_PIDS).
168 */
169 max_scan = DIV_ROUND_UP(pid_max, BITS_PER_PAGE) - !offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 for (i = 0; i <= max_scan; ++i) {
171 if (unlikely(!map->page)) {
Sukadev Bhattiprolu3fbc9642006-10-02 02:17:24 -0700172 void *page = kzalloc(PAGE_SIZE, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 /*
174 * Free the page if someone raced with us
175 * installing it:
176 */
Eric W. Biederman92476d72006-03-31 02:31:42 -0800177 spin_lock_irq(&pidmap_lock);
André Goddard Rosa7be6d992009-12-15 16:47:39 -0800178 if (!map->page) {
Sukadev Bhattiprolu3fbc9642006-10-02 02:17:24 -0700179 map->page = page;
André Goddard Rosa7be6d992009-12-15 16:47:39 -0800180 page = NULL;
181 }
Eric W. Biederman92476d72006-03-31 02:31:42 -0800182 spin_unlock_irq(&pidmap_lock);
André Goddard Rosa7be6d992009-12-15 16:47:39 -0800183 kfree(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184 if (unlikely(!map->page))
185 break;
186 }
187 if (likely(atomic_read(&map->nr_free))) {
Raphael S. Carvalho8db049b2013-04-30 15:28:26 -0700188 for ( ; ; ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 if (!test_and_set_bit(offset, map->page)) {
190 atomic_dec(&map->nr_free);
Salman5fdee8c2010-08-10 18:03:16 -0700191 set_last_pid(pid_ns, last, pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 return pid;
193 }
194 offset = find_next_offset(map, offset);
Raphael S. Carvalho8db049b2013-04-30 15:28:26 -0700195 if (offset >= BITS_PER_PAGE)
196 break;
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -0800197 pid = mk_pid(pid_ns, map, offset);
Raphael S. Carvalho8db049b2013-04-30 15:28:26 -0700198 if (pid >= pid_max)
199 break;
200 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 }
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -0800202 if (map < &pid_ns->pidmap[(pid_max-1)/BITS_PER_PAGE]) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 ++map;
204 offset = 0;
205 } else {
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -0800206 map = &pid_ns->pidmap[0];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 offset = RESERVED_PIDS;
208 if (unlikely(last == offset))
209 break;
210 }
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -0800211 pid = mk_pid(pid_ns, map, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 }
213 return -1;
214}
215
Linus Torvaldsc78193e2011-04-18 10:35:30 -0700216int next_pidmap(struct pid_namespace *pid_ns, unsigned int last)
Eric W. Biederman0804ef42006-10-02 02:17:04 -0700217{
218 int offset;
Eric W. Biedermanf40f50d2006-10-02 02:17:25 -0700219 struct pidmap *map, *end;
Eric W. Biederman0804ef42006-10-02 02:17:04 -0700220
Linus Torvaldsc78193e2011-04-18 10:35:30 -0700221 if (last >= PID_MAX_LIMIT)
222 return -1;
223
Eric W. Biederman0804ef42006-10-02 02:17:04 -0700224 offset = (last + 1) & BITS_PER_PAGE_MASK;
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -0800225 map = &pid_ns->pidmap[(last + 1)/BITS_PER_PAGE];
226 end = &pid_ns->pidmap[PIDMAP_ENTRIES];
Eric W. Biedermanf40f50d2006-10-02 02:17:25 -0700227 for (; map < end; map++, offset = 0) {
Eric W. Biederman0804ef42006-10-02 02:17:04 -0700228 if (unlikely(!map->page))
229 continue;
230 offset = find_next_bit((map)->page, BITS_PER_PAGE, offset);
231 if (offset < BITS_PER_PAGE)
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -0800232 return mk_pid(pid_ns, map, offset);
Eric W. Biederman0804ef42006-10-02 02:17:04 -0700233 }
234 return -1;
235}
236
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -0800237void put_pid(struct pid *pid)
Eric W. Biederman92476d72006-03-31 02:31:42 -0800238{
Pavel Emelianovbaf8f0f2007-10-18 23:39:48 -0700239 struct pid_namespace *ns;
240
Eric W. Biederman92476d72006-03-31 02:31:42 -0800241 if (!pid)
242 return;
Pavel Emelianovbaf8f0f2007-10-18 23:39:48 -0700243
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -0700244 ns = pid->numbers[pid->level].ns;
Eric W. Biederman92476d72006-03-31 02:31:42 -0800245 if ((atomic_read(&pid->count) == 1) ||
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -0700246 atomic_dec_and_test(&pid->count)) {
Pavel Emelianovbaf8f0f2007-10-18 23:39:48 -0700247 kmem_cache_free(ns->pid_cachep, pid);
Pavel Emelyanovb461cc02007-10-18 23:40:09 -0700248 put_pid_ns(ns);
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -0700249 }
Eric W. Biederman92476d72006-03-31 02:31:42 -0800250}
Eric W. Biedermanbbf73142006-10-02 02:17:11 -0700251EXPORT_SYMBOL_GPL(put_pid);
Eric W. Biederman92476d72006-03-31 02:31:42 -0800252
253static void delayed_put_pid(struct rcu_head *rhp)
254{
255 struct pid *pid = container_of(rhp, struct pid, rcu);
256 put_pid(pid);
257}
258
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -0800259void free_pid(struct pid *pid)
Eric W. Biederman92476d72006-03-31 02:31:42 -0800260{
261 /* We can be called with write_lock_irq(&tasklist_lock) held */
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -0700262 int i;
Eric W. Biederman92476d72006-03-31 02:31:42 -0800263 unsigned long flags;
264
265 spin_lock_irqsave(&pidmap_lock, flags);
Eric W. Biederman0a01f2c2012-08-01 10:33:47 -0700266 for (i = 0; i <= pid->level; i++) {
267 struct upid *upid = pid->numbers + i;
Eric W. Biedermanaf4b8a82012-08-01 15:03:42 -0700268 struct pid_namespace *ns = upid->ns;
Eric W. Biederman0a01f2c2012-08-01 10:33:47 -0700269 hlist_del_rcu(&upid->pid_chain);
Eric W. Biedermanaf4b8a82012-08-01 15:03:42 -0700270 switch(--ns->nr_hashed) {
Eric W. Biedermana6064882013-08-29 13:56:50 -0700271 case 2:
Eric W. Biedermanaf4b8a82012-08-01 15:03:42 -0700272 case 1:
273 /* When all that is left in the pid namespace
274 * is the reaper wake up the reaper. The reaper
275 * may be sleeping in zap_pid_ns_processes().
276 */
277 wake_up_process(ns->child_reaper);
278 break;
Oleg Nesterov314a8ad2013-09-30 13:45:27 -0700279 case PIDNS_HASH_ADDING:
280 /* Handle a fork failure of the first process */
281 WARN_ON(ns->child_reaper);
282 ns->nr_hashed = 0;
283 /* fall through */
Eric W. Biedermanaf4b8a82012-08-01 15:03:42 -0700284 case 0:
Eric W. Biedermanaf4b8a82012-08-01 15:03:42 -0700285 schedule_work(&ns->proc_work);
286 break;
Eric W. Biederman5e1182deb2010-07-12 18:50:25 -0700287 }
Eric W. Biederman0a01f2c2012-08-01 10:33:47 -0700288 }
Eric W. Biederman92476d72006-03-31 02:31:42 -0800289 spin_unlock_irqrestore(&pidmap_lock, flags);
290
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -0700291 for (i = 0; i <= pid->level; i++)
Oleg Nesterovb7127aa2008-04-30 00:54:22 -0700292 free_pidmap(pid->numbers + i);
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -0700293
Eric W. Biederman92476d72006-03-31 02:31:42 -0800294 call_rcu(&pid->rcu, delayed_put_pid);
295}
296
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -0700297struct pid *alloc_pid(struct pid_namespace *ns)
Eric W. Biederman92476d72006-03-31 02:31:42 -0800298{
299 struct pid *pid;
300 enum pid_type type;
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -0700301 int i, nr;
302 struct pid_namespace *tmp;
Pavel Emelyanov198fe212007-10-18 23:40:06 -0700303 struct upid *upid;
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)
307 goto out;
308
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);
313 if (nr < 0)
314 goto out_free;
Eric W. Biederman92476d72006-03-31 02:31:42 -0800315
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -0700316 pid->numbers[i].nr = nr;
317 pid->numbers[i].ns = tmp;
318 tmp = tmp->parent;
319 }
320
Eric W. Biederman0a01f2c2012-08-01 10:33:47 -0700321 if (unlikely(is_child_reaper(pid))) {
322 if (pid_ns_prepare_proc(ns))
323 goto out_free;
324 }
325
Pavel Emelyanovb461cc02007-10-18 23:40:09 -0700326 get_pid_ns(ns);
Eric W. Biederman92476d72006-03-31 02:31:42 -0800327 atomic_set(&pid->count, 1);
Eric W. Biederman92476d72006-03-31 02:31:42 -0800328 for (type = 0; type < PIDTYPE_MAX; ++type)
329 INIT_HLIST_HEAD(&pid->tasks[type]);
330
André Goddard Rosa417e3152009-12-15 16:47:40 -0800331 upid = pid->numbers + ns->level;
Eric W. Biederman92476d72006-03-31 02:31:42 -0800332 spin_lock_irq(&pidmap_lock);
Eric W. Biedermanc876ad762012-12-21 20:27:12 -0800333 if (!(ns->nr_hashed & PIDNS_HASH_ADDING))
Eric W. Biederman5e1182deb2010-07-12 18:50:25 -0700334 goto out_unlock;
Eric W. Biederman0a01f2c2012-08-01 10:33:47 -0700335 for ( ; upid >= pid->numbers; --upid) {
Pavel Emelyanov198fe212007-10-18 23:40:06 -0700336 hlist_add_head_rcu(&upid->pid_chain,
337 &pid_hash[pid_hashfn(upid->nr, upid->ns)]);
Eric W. Biederman0a01f2c2012-08-01 10:33:47 -0700338 upid->ns->nr_hashed++;
339 }
Eric W. Biederman92476d72006-03-31 02:31:42 -0800340 spin_unlock_irq(&pidmap_lock);
341
342out:
343 return pid;
344
Eric W. Biederman5e1182deb2010-07-12 18:50:25 -0700345out_unlock:
Eric W. Biederman6e666882013-02-12 13:46:23 -0800346 spin_unlock_irq(&pidmap_lock);
Oleg Nesterov24c037e2014-12-10 15:55:25 -0800347 put_pid_ns(ns);
348
Eric W. Biederman92476d72006-03-31 02:31:42 -0800349out_free:
Oleg Nesterovb7127aa2008-04-30 00:54:22 -0700350 while (++i <= ns->level)
351 free_pidmap(pid->numbers + i);
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -0700352
Pavel Emelianovbaf8f0f2007-10-18 23:39:48 -0700353 kmem_cache_free(ns->pid_cachep, pid);
Eric W. Biederman92476d72006-03-31 02:31:42 -0800354 pid = NULL;
355 goto out;
356}
357
Eric W. Biedermanc876ad762012-12-21 20:27:12 -0800358void disable_pid_allocation(struct pid_namespace *ns)
359{
360 spin_lock_irq(&pidmap_lock);
361 ns->nr_hashed &= ~PIDNS_HASH_ADDING;
362 spin_unlock_irq(&pidmap_lock);
363}
364
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -0800365struct pid *find_pid_ns(int nr, struct pid_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366{
Pavel Emelyanov198fe212007-10-18 23:40:06 -0700367 struct upid *pnr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368
Sasha Levinb67bfe02013-02-27 17:06:00 -0800369 hlist_for_each_entry_rcu(pnr,
Pavel Emelyanov198fe212007-10-18 23:40:06 -0700370 &pid_hash[pid_hashfn(nr, ns)], pid_chain)
371 if (pnr->nr == nr && pnr->ns == ns)
372 return container_of(pnr, struct pid,
373 numbers[ns->level]);
374
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 return NULL;
376}
Pavel Emelyanov198fe212007-10-18 23:40:06 -0700377EXPORT_SYMBOL_GPL(find_pid_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378
Pavel Emelyanov89905712007-10-18 23:40:19 -0700379struct pid *find_vpid(int nr)
380{
Eric W. Biederman17cf22c2010-03-02 14:51:53 -0800381 return find_pid_ns(nr, task_active_pid_ns(current));
Pavel Emelyanov89905712007-10-18 23:40:19 -0700382}
383EXPORT_SYMBOL_GPL(find_vpid);
384
Sukadev Bhattiprolue713d0d2007-05-10 22:22:58 -0700385/*
386 * attach_pid() must be called with the tasklist_lock write-held.
387 */
Oleg Nesterov81907732013-07-03 15:08:31 -0700388void attach_pid(struct task_struct *task, enum pid_type type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389{
Oleg Nesterov81907732013-07-03 15:08:31 -0700390 struct pid_link *link = &task->pids[type];
391 hlist_add_head_rcu(&link->node, &link->pid->tasks[type]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392}
393
Oleg Nesterov24336ea2008-04-30 00:54:26 -0700394static void __change_pid(struct task_struct *task, enum pid_type type,
395 struct pid *new)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396{
Eric W. Biederman92476d72006-03-31 02:31:42 -0800397 struct pid_link *link;
398 struct pid *pid;
399 int tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400
Eric W. Biederman92476d72006-03-31 02:31:42 -0800401 link = &task->pids[type];
402 pid = link->pid;
403
404 hlist_del_rcu(&link->node);
Oleg Nesterov24336ea2008-04-30 00:54:26 -0700405 link->pid = new;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406
407 for (tmp = PIDTYPE_MAX; --tmp >= 0; )
Eric W. Biederman92476d72006-03-31 02:31:42 -0800408 if (!hlist_empty(&pid->tasks[tmp]))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 return;
410
Eric W. Biederman92476d72006-03-31 02:31:42 -0800411 free_pid(pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412}
413
Oleg Nesterov24336ea2008-04-30 00:54:26 -0700414void detach_pid(struct task_struct *task, enum pid_type type)
415{
416 __change_pid(task, type, NULL);
417}
418
419void change_pid(struct task_struct *task, enum pid_type type,
420 struct pid *pid)
421{
422 __change_pid(task, type, pid);
Oleg Nesterov81907732013-07-03 15:08:31 -0700423 attach_pid(task, type);
Oleg Nesterov24336ea2008-04-30 00:54:26 -0700424}
425
Eric W. Biedermanc18258c2006-09-27 01:51:06 -0700426/* transfer_pid is an optimization of attach_pid(new), detach_pid(old) */
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -0800427void transfer_pid(struct task_struct *old, struct task_struct *new,
Eric W. Biedermanc18258c2006-09-27 01:51:06 -0700428 enum pid_type type)
429{
430 new->pids[type].pid = old->pids[type].pid;
431 hlist_replace_rcu(&old->pids[type].node, &new->pids[type].node);
Eric W. Biedermanc18258c2006-09-27 01:51:06 -0700432}
433
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -0800434struct task_struct *pid_task(struct pid *pid, enum pid_type type)
Eric W. Biederman92476d72006-03-31 02:31:42 -0800435{
436 struct task_struct *result = NULL;
437 if (pid) {
438 struct hlist_node *first;
Arnd Bergmann67bdbff2010-02-25 16:55:13 +0100439 first = rcu_dereference_check(hlist_first_rcu(&pid->tasks[type]),
Paul E. McKenneydb1466b2010-03-03 07:46:56 -0800440 lockdep_tasklist_lock_is_held());
Eric W. Biederman92476d72006-03-31 02:31:42 -0800441 if (first)
442 result = hlist_entry(first, struct task_struct, pids[(type)].node);
443 }
444 return result;
445}
Pavel Emelyanoveccba062008-02-07 00:13:21 -0800446EXPORT_SYMBOL(pid_task);
Eric W. Biederman92476d72006-03-31 02:31:42 -0800447
448/*
Tetsuo Handa9728e5d2010-03-05 13:42:56 -0800449 * Must be called under rcu_read_lock().
Eric W. Biederman92476d72006-03-31 02:31:42 -0800450 */
Christoph Hellwig17f98dc2009-06-17 16:27:51 -0700451struct task_struct *find_task_by_pid_ns(pid_t nr, struct pid_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452{
Paul E. McKenneyb3fbab02011-05-24 08:31:09 -0700453 rcu_lockdep_assert(rcu_read_lock_held(),
454 "find_task_by_pid_ns() needs rcu_read_lock()"
455 " protection");
Christoph Hellwig17f98dc2009-06-17 16:27:51 -0700456 return pid_task(find_pid_ns(nr, ns), PIDTYPE_PID);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457}
458
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -0700459struct task_struct *find_task_by_vpid(pid_t vnr)
460{
Eric W. Biederman17cf22c2010-03-02 14:51:53 -0800461 return find_task_by_pid_ns(vnr, task_active_pid_ns(current));
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -0700462}
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -0700463
Oleg Nesterov1a657f782006-10-02 02:18:59 -0700464struct pid *get_task_pid(struct task_struct *task, enum pid_type type)
465{
466 struct pid *pid;
467 rcu_read_lock();
Oleg Nesterov2ae448e2009-04-02 16:58:36 -0700468 if (type != PIDTYPE_PID)
469 task = task->group_leader;
Oleg Nesterov1a657f782006-10-02 02:18:59 -0700470 pid = get_pid(task->pids[type].pid);
471 rcu_read_unlock();
472 return pid;
473}
Rik van Riel77c100c2011-02-01 09:51:46 -0500474EXPORT_SYMBOL_GPL(get_task_pid);
Oleg Nesterov1a657f782006-10-02 02:18:59 -0700475
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -0800476struct task_struct *get_pid_task(struct pid *pid, enum pid_type type)
Eric W. Biederman92476d72006-03-31 02:31:42 -0800477{
478 struct task_struct *result;
479 rcu_read_lock();
480 result = pid_task(pid, type);
481 if (result)
482 get_task_struct(result);
483 rcu_read_unlock();
484 return result;
485}
Rik van Riel77c100c2011-02-01 09:51:46 -0500486EXPORT_SYMBOL_GPL(get_pid_task);
Eric W. Biederman92476d72006-03-31 02:31:42 -0800487
488struct pid *find_get_pid(pid_t nr)
489{
490 struct pid *pid;
491
492 rcu_read_lock();
Pavel Emelyanov198fe212007-10-18 23:40:06 -0700493 pid = get_pid(find_vpid(nr));
Eric W. Biederman92476d72006-03-31 02:31:42 -0800494 rcu_read_unlock();
495
496 return pid;
497}
David Sterba339caf22008-07-25 01:48:31 -0700498EXPORT_SYMBOL_GPL(find_get_pid);
Eric W. Biederman92476d72006-03-31 02:31:42 -0800499
Pavel Emelyanov7af57292007-10-18 23:40:06 -0700500pid_t pid_nr_ns(struct pid *pid, struct pid_namespace *ns)
501{
502 struct upid *upid;
503 pid_t nr = 0;
504
505 if (pid && ns->level <= pid->level) {
506 upid = &pid->numbers[ns->level];
507 if (upid->ns == ns)
508 nr = upid->nr;
509 }
510 return nr;
511}
Eric W. Biederman4f82f452012-05-24 10:37:59 -0600512EXPORT_SYMBOL_GPL(pid_nr_ns);
Pavel Emelyanov7af57292007-10-18 23:40:06 -0700513
Eric W. Biederman44c4e1b2008-02-08 04:19:15 -0800514pid_t pid_vnr(struct pid *pid)
515{
Eric W. Biederman17cf22c2010-03-02 14:51:53 -0800516 return pid_nr_ns(pid, task_active_pid_ns(current));
Eric W. Biederman44c4e1b2008-02-08 04:19:15 -0800517}
518EXPORT_SYMBOL_GPL(pid_vnr);
519
Oleg Nesterov52ee2df2009-04-02 16:58:38 -0700520pid_t __task_pid_nr_ns(struct task_struct *task, enum pid_type type,
521 struct pid_namespace *ns)
Pavel Emelyanov2f2a3a42007-10-18 23:40:19 -0700522{
Oleg Nesterov52ee2df2009-04-02 16:58:38 -0700523 pid_t nr = 0;
524
525 rcu_read_lock();
526 if (!ns)
Eric W. Biederman17cf22c2010-03-02 14:51:53 -0800527 ns = task_active_pid_ns(current);
Oleg Nesterov52ee2df2009-04-02 16:58:38 -0700528 if (likely(pid_alive(task))) {
529 if (type != PIDTYPE_PID)
530 task = task->group_leader;
531 nr = pid_nr_ns(task->pids[type].pid, ns);
532 }
533 rcu_read_unlock();
534
535 return nr;
Pavel Emelyanov2f2a3a42007-10-18 23:40:19 -0700536}
Oleg Nesterov52ee2df2009-04-02 16:58:38 -0700537EXPORT_SYMBOL(__task_pid_nr_ns);
Pavel Emelyanov2f2a3a42007-10-18 23:40:19 -0700538
539pid_t task_tgid_nr_ns(struct task_struct *tsk, struct pid_namespace *ns)
540{
541 return pid_nr_ns(task_tgid(tsk), ns);
542}
543EXPORT_SYMBOL(task_tgid_nr_ns);
544
Eric W. Biederman61bce0f2009-01-07 18:08:49 -0800545struct pid_namespace *task_active_pid_ns(struct task_struct *tsk)
546{
547 return ns_of_pid(task_pid(tsk));
548}
549EXPORT_SYMBOL_GPL(task_active_pid_ns);
550
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551/*
Frederik Schwarzer025dfda2008-10-16 19:02:37 +0200552 * Used by proc to find the first pid that is greater than or equal to nr.
Eric W. Biederman0804ef42006-10-02 02:17:04 -0700553 *
Pavel Emelyanove49859e2008-07-25 01:48:36 -0700554 * 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 -0700555 */
Pavel Emelyanov198fe212007-10-18 23:40:06 -0700556struct pid *find_ge_pid(int nr, struct pid_namespace *ns)
Eric W. Biederman0804ef42006-10-02 02:17:04 -0700557{
558 struct pid *pid;
559
560 do {
Pavel Emelyanov198fe212007-10-18 23:40:06 -0700561 pid = find_pid_ns(nr, ns);
Eric W. Biederman0804ef42006-10-02 02:17:04 -0700562 if (pid)
563 break;
Pavel Emelyanov198fe212007-10-18 23:40:06 -0700564 nr = next_pidmap(ns, nr);
Eric W. Biederman0804ef42006-10-02 02:17:04 -0700565 } while (nr > 0);
566
567 return pid;
568}
569
570/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 * The pid hash table is scaled according to the amount of memory in the
572 * machine. From a minimum of 16 slots up to 4096 slots at one gigabyte or
573 * more.
574 */
575void __init pidhash_init(void)
576{
Dimitri Sivanich074b8512012-02-08 12:39:07 -0800577 unsigned int i, pidhash_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578
Jan Beulich2c85f512009-09-21 17:03:07 -0700579 pid_hash = alloc_large_system_hash("PID", sizeof(*pid_hash), 0, 18,
580 HASH_EARLY | HASH_SMALL,
Tim Bird31fe62b2012-05-23 13:33:35 +0000581 &pidhash_shift, NULL,
582 0, 4096);
Dimitri Sivanich074b8512012-02-08 12:39:07 -0800583 pidhash_size = 1U << pidhash_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584
Eric W. Biederman92476d72006-03-31 02:31:42 -0800585 for (i = 0; i < pidhash_size; i++)
586 INIT_HLIST_HEAD(&pid_hash[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587}
588
589void __init pidmap_init(void)
590{
Eric W. Biedermanc876ad762012-12-21 20:27:12 -0800591 /* Veryify no one has done anything silly */
592 BUILD_BUG_ON(PID_MAX_LIMIT >= PIDNS_HASH_ADDING);
593
Hedi Berriche72680a12010-05-26 14:44:06 -0700594 /* bump default and minimum pid_max based on number of cpus */
595 pid_max = min(pid_max_max, max_t(int, pid_max,
596 PIDS_PER_CPU_DEFAULT * num_possible_cpus()));
597 pid_max_min = max_t(int, pid_max_min,
598 PIDS_PER_CPU_MIN * num_possible_cpus());
599 pr_info("pid_max: default: %u minimum: %u\n", pid_max, pid_max_min);
600
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -0800601 init_pid_ns.pidmap[0].page = kzalloc(PAGE_SIZE, GFP_KERNEL);
Oleg Nesterov73b9ebf2006-03-28 16:11:07 -0800602 /* Reserve PID 0. We never call free_pidmap(0) */
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -0800603 set_bit(0, init_pid_ns.pidmap[0].page);
604 atomic_dec(&init_pid_ns.pidmap[0].nr_free);
Eric W. Biederman92476d72006-03-31 02:31:42 -0800605
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800606 init_pid_ns.pid_cachep = KMEM_CACHE(pid,
607 SLAB_HWCACHE_ALIGN | SLAB_PANIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608}