blob: 047dc62646389fd5abdf5fc63cb8db963a1a5b60 [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>
Eric W. Biederman0a01f2c2012-08-01 10:33:47 -070039#include <linux/proc_fs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070040
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -070041#define pid_hashfn(nr, ns) \
42 hash_long((unsigned long)nr + (unsigned long)ns, pidhash_shift)
Eric W. Biederman92476d72006-03-31 02:31:42 -080043static struct hlist_head *pid_hash;
Jan Beulich2c85f512009-09-21 17:03:07 -070044static unsigned int pidhash_shift = 4;
Sukadev Bhattiprolu820e45d2007-05-10 22:23:00 -070045struct pid init_struct_pid = INIT_STRUCT_PID;
Linus Torvalds1da177e2005-04-16 15:20:36 -070046
47int pid_max = PID_MAX_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -070048
49#define RESERVED_PIDS 300
50
51int pid_max_min = RESERVED_PIDS + 1;
52int pid_max_max = PID_MAX_LIMIT;
53
Linus Torvalds1da177e2005-04-16 15:20:36 -070054#define BITS_PER_PAGE (PAGE_SIZE*8)
55#define BITS_PER_PAGE_MASK (BITS_PER_PAGE-1)
Sukadev Bhattiprolu3fbc9642006-10-02 02:17:24 -070056
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -080057static inline int mk_pid(struct pid_namespace *pid_ns,
58 struct pidmap *map, int off)
Sukadev Bhattiprolu3fbc9642006-10-02 02:17:24 -070059{
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -080060 return (map - pid_ns->pidmap)*BITS_PER_PAGE + off;
Sukadev Bhattiprolu3fbc9642006-10-02 02:17:24 -070061}
62
Linus Torvalds1da177e2005-04-16 15:20:36 -070063#define find_next_offset(map, off) \
64 find_next_zero_bit((map)->page, BITS_PER_PAGE, off)
65
66/*
67 * PID-map pages start out as NULL, they get allocated upon
68 * first use and are never deallocated. This way a low pid_max
69 * value does not cause lots of bitmaps to be allocated, but
70 * the scheme scales to up to 4 million PIDs, runtime.
71 */
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -080072struct pid_namespace init_pid_ns = {
Cedric Le Goater9a575a92006-12-08 02:37:59 -080073 .kref = {
74 .refcount = ATOMIC_INIT(2),
75 },
Sukadev Bhattiprolu3fbc9642006-10-02 02:17:24 -070076 .pidmap = {
77 [ 0 ... PIDMAP_ENTRIES-1] = { ATOMIC_INIT(BITS_PER_PAGE), NULL }
78 },
Sukadev Bhattiprolu84d73782006-12-08 02:38:01 -080079 .last_pid = 0,
Pavel Emelyanovfaacbfd2007-10-18 23:40:04 -070080 .level = 0,
81 .child_reaper = &init_task,
Eric W. Biederman49f4d8b2012-08-02 04:25:10 -070082 .user_ns = &init_user_ns,
Eric W. Biederman98f842e2011-06-15 10:21:48 -070083 .proc_inum = PROC_PID_INIT_INO,
Sukadev Bhattiprolu3fbc9642006-10-02 02:17:24 -070084};
Pavel Emelyanov198fe212007-10-18 23:40:06 -070085EXPORT_SYMBOL_GPL(init_pid_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -070086
Eric W. Biederman92476d72006-03-31 02:31:42 -080087/*
88 * Note: disable interrupts while the pidmap_lock is held as an
89 * interrupt might come in and do read_lock(&tasklist_lock).
90 *
91 * If we don't disable interrupts there is a nasty deadlock between
92 * detach_pid()->free_pid() and another cpu that does
93 * spin_lock(&pidmap_lock) followed by an interrupt routine that does
94 * read_lock(&tasklist_lock);
95 *
96 * After we clean up the tasklist_lock and know there are no
97 * irq handlers that take it we can leave the interrupts enabled.
98 * For now it is easier to be safe than to prove it can't happen.
99 */
Sukadev Bhattiprolu3fbc9642006-10-02 02:17:24 -0700100
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101static __cacheline_aligned_in_smp DEFINE_SPINLOCK(pidmap_lock);
102
Oleg Nesterovb7127aa2008-04-30 00:54:22 -0700103static void free_pidmap(struct upid *upid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104{
Oleg Nesterovb7127aa2008-04-30 00:54:22 -0700105 int nr = upid->nr;
106 struct pidmap *map = upid->ns->pidmap + nr / BITS_PER_PAGE;
107 int offset = nr & BITS_PER_PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108
109 clear_bit(offset, map->page);
110 atomic_inc(&map->nr_free);
111}
112
Salman5fdee8c2010-08-10 18:03:16 -0700113/*
114 * If we started walking pids at 'base', is 'a' seen before 'b'?
115 */
116static int pid_before(int base, int a, int b)
117{
118 /*
119 * This is the same as saying
120 *
121 * (a - base + MAXUINT) % MAXUINT < (b - base + MAXUINT) % MAXUINT
122 * and that mapping orders 'a' and 'b' with respect to 'base'.
123 */
124 return (unsigned)(a - base) < (unsigned)(b - base);
125}
126
127/*
Pavel Emelyanovb8f566b2012-01-12 17:20:27 -0800128 * We might be racing with someone else trying to set pid_ns->last_pid
129 * at the pid allocation time (there's also a sysctl for this, but racing
130 * with this one is OK, see comment in kernel/pid_namespace.c about it).
Salman5fdee8c2010-08-10 18:03:16 -0700131 * We want the winner to have the "later" value, because if the
132 * "earlier" value prevails, then a pid may get reused immediately.
133 *
134 * Since pids rollover, it is not sufficient to just pick the bigger
135 * value. We have to consider where we started counting from.
136 *
137 * 'base' is the value of pid_ns->last_pid that we observed when
138 * we started looking for a pid.
139 *
140 * 'pid' is the pid that we eventually found.
141 */
142static void set_last_pid(struct pid_namespace *pid_ns, int base, int pid)
143{
144 int prev;
145 int last_write = base;
146 do {
147 prev = last_write;
148 last_write = cmpxchg(&pid_ns->last_pid, prev, pid);
149 } while ((prev != last_write) && (pid_before(base, last_write, pid)));
150}
151
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -0800152static int alloc_pidmap(struct pid_namespace *pid_ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153{
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -0800154 int i, offset, max_scan, pid, last = pid_ns->last_pid;
Sukadev Bhattiprolu6a1f3b82006-10-02 02:17:20 -0700155 struct pidmap *map;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156
157 pid = last + 1;
158 if (pid >= pid_max)
159 pid = RESERVED_PIDS;
160 offset = pid & BITS_PER_PAGE_MASK;
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -0800161 map = &pid_ns->pidmap[pid/BITS_PER_PAGE];
Oleg Nesterovc52b0b92010-08-10 18:03:17 -0700162 /*
163 * If last_pid points into the middle of the map->page we
164 * want to scan this bitmap block twice, the second time
165 * we start with offset == 0 (or RESERVED_PIDS).
166 */
167 max_scan = DIV_ROUND_UP(pid_max, BITS_PER_PAGE) - !offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168 for (i = 0; i <= max_scan; ++i) {
169 if (unlikely(!map->page)) {
Sukadev Bhattiprolu3fbc9642006-10-02 02:17:24 -0700170 void *page = kzalloc(PAGE_SIZE, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 /*
172 * Free the page if someone raced with us
173 * installing it:
174 */
Eric W. Biederman92476d72006-03-31 02:31:42 -0800175 spin_lock_irq(&pidmap_lock);
André Goddard Rosa7be6d992009-12-15 16:47:39 -0800176 if (!map->page) {
Sukadev Bhattiprolu3fbc9642006-10-02 02:17:24 -0700177 map->page = page;
André Goddard Rosa7be6d992009-12-15 16:47:39 -0800178 page = NULL;
179 }
Eric W. Biederman92476d72006-03-31 02:31:42 -0800180 spin_unlock_irq(&pidmap_lock);
André Goddard Rosa7be6d992009-12-15 16:47:39 -0800181 kfree(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182 if (unlikely(!map->page))
183 break;
184 }
185 if (likely(atomic_read(&map->nr_free))) {
186 do {
187 if (!test_and_set_bit(offset, map->page)) {
188 atomic_dec(&map->nr_free);
Salman5fdee8c2010-08-10 18:03:16 -0700189 set_last_pid(pid_ns, last, pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 return pid;
191 }
192 offset = find_next_offset(map, offset);
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -0800193 pid = mk_pid(pid_ns, map, offset);
Oleg Nesterovc52b0b92010-08-10 18:03:17 -0700194 } while (offset < BITS_PER_PAGE && pid < pid_max);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 }
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -0800196 if (map < &pid_ns->pidmap[(pid_max-1)/BITS_PER_PAGE]) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 ++map;
198 offset = 0;
199 } else {
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -0800200 map = &pid_ns->pidmap[0];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 offset = RESERVED_PIDS;
202 if (unlikely(last == offset))
203 break;
204 }
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -0800205 pid = mk_pid(pid_ns, map, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 }
207 return -1;
208}
209
Linus Torvaldsc78193e2011-04-18 10:35:30 -0700210int next_pidmap(struct pid_namespace *pid_ns, unsigned int last)
Eric W. Biederman0804ef42006-10-02 02:17:04 -0700211{
212 int offset;
Eric W. Biedermanf40f50d2006-10-02 02:17:25 -0700213 struct pidmap *map, *end;
Eric W. Biederman0804ef42006-10-02 02:17:04 -0700214
Linus Torvaldsc78193e2011-04-18 10:35:30 -0700215 if (last >= PID_MAX_LIMIT)
216 return -1;
217
Eric W. Biederman0804ef42006-10-02 02:17:04 -0700218 offset = (last + 1) & BITS_PER_PAGE_MASK;
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -0800219 map = &pid_ns->pidmap[(last + 1)/BITS_PER_PAGE];
220 end = &pid_ns->pidmap[PIDMAP_ENTRIES];
Eric W. Biedermanf40f50d2006-10-02 02:17:25 -0700221 for (; map < end; map++, offset = 0) {
Eric W. Biederman0804ef42006-10-02 02:17:04 -0700222 if (unlikely(!map->page))
223 continue;
224 offset = find_next_bit((map)->page, BITS_PER_PAGE, offset);
225 if (offset < BITS_PER_PAGE)
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -0800226 return mk_pid(pid_ns, map, offset);
Eric W. Biederman0804ef42006-10-02 02:17:04 -0700227 }
228 return -1;
229}
230
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -0800231void put_pid(struct pid *pid)
Eric W. Biederman92476d72006-03-31 02:31:42 -0800232{
Pavel Emelianovbaf8f0f2007-10-18 23:39:48 -0700233 struct pid_namespace *ns;
234
Eric W. Biederman92476d72006-03-31 02:31:42 -0800235 if (!pid)
236 return;
Pavel Emelianovbaf8f0f2007-10-18 23:39:48 -0700237
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -0700238 ns = pid->numbers[pid->level].ns;
Eric W. Biederman92476d72006-03-31 02:31:42 -0800239 if ((atomic_read(&pid->count) == 1) ||
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -0700240 atomic_dec_and_test(&pid->count)) {
Pavel Emelianovbaf8f0f2007-10-18 23:39:48 -0700241 kmem_cache_free(ns->pid_cachep, pid);
Pavel Emelyanovb461cc02007-10-18 23:40:09 -0700242 put_pid_ns(ns);
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -0700243 }
Eric W. Biederman92476d72006-03-31 02:31:42 -0800244}
Eric W. Biedermanbbf73142006-10-02 02:17:11 -0700245EXPORT_SYMBOL_GPL(put_pid);
Eric W. Biederman92476d72006-03-31 02:31:42 -0800246
247static void delayed_put_pid(struct rcu_head *rhp)
248{
249 struct pid *pid = container_of(rhp, struct pid, rcu);
250 put_pid(pid);
251}
252
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -0800253void free_pid(struct pid *pid)
Eric W. Biederman92476d72006-03-31 02:31:42 -0800254{
255 /* We can be called with write_lock_irq(&tasklist_lock) held */
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -0700256 int i;
Eric W. Biederman92476d72006-03-31 02:31:42 -0800257 unsigned long flags;
258
259 spin_lock_irqsave(&pidmap_lock, flags);
Eric W. Biederman0a01f2c2012-08-01 10:33:47 -0700260 for (i = 0; i <= pid->level; i++) {
261 struct upid *upid = pid->numbers + i;
Eric W. Biedermanaf4b8a82012-08-01 15:03:42 -0700262 struct pid_namespace *ns = upid->ns;
Eric W. Biederman0a01f2c2012-08-01 10:33:47 -0700263 hlist_del_rcu(&upid->pid_chain);
Eric W. Biedermanaf4b8a82012-08-01 15:03:42 -0700264 switch(--ns->nr_hashed) {
265 case 1:
266 /* When all that is left in the pid namespace
267 * is the reaper wake up the reaper. The reaper
268 * may be sleeping in zap_pid_ns_processes().
269 */
270 wake_up_process(ns->child_reaper);
271 break;
272 case 0:
Eric W. Biedermanaf4b8a82012-08-01 15:03:42 -0700273 schedule_work(&ns->proc_work);
274 break;
Eric W. Biederman5e1182deb2010-07-12 18:50:25 -0700275 }
Eric W. Biederman0a01f2c2012-08-01 10:33:47 -0700276 }
Eric W. Biederman92476d72006-03-31 02:31:42 -0800277 spin_unlock_irqrestore(&pidmap_lock, flags);
278
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -0700279 for (i = 0; i <= pid->level; i++)
Oleg Nesterovb7127aa2008-04-30 00:54:22 -0700280 free_pidmap(pid->numbers + i);
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -0700281
Eric W. Biederman92476d72006-03-31 02:31:42 -0800282 call_rcu(&pid->rcu, delayed_put_pid);
283}
284
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -0700285struct pid *alloc_pid(struct pid_namespace *ns)
Eric W. Biederman92476d72006-03-31 02:31:42 -0800286{
287 struct pid *pid;
288 enum pid_type type;
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -0700289 int i, nr;
290 struct pid_namespace *tmp;
Pavel Emelyanov198fe212007-10-18 23:40:06 -0700291 struct upid *upid;
Eric W. Biederman92476d72006-03-31 02:31:42 -0800292
Pavel Emelianovbaf8f0f2007-10-18 23:39:48 -0700293 pid = kmem_cache_alloc(ns->pid_cachep, GFP_KERNEL);
Eric W. Biederman92476d72006-03-31 02:31:42 -0800294 if (!pid)
295 goto out;
296
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -0700297 tmp = ns;
Eric W. Biederman0a01f2c2012-08-01 10:33:47 -0700298 pid->level = ns->level;
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -0700299 for (i = ns->level; i >= 0; i--) {
300 nr = alloc_pidmap(tmp);
301 if (nr < 0)
302 goto out_free;
Eric W. Biederman92476d72006-03-31 02:31:42 -0800303
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -0700304 pid->numbers[i].nr = nr;
305 pid->numbers[i].ns = tmp;
306 tmp = tmp->parent;
307 }
308
Eric W. Biederman0a01f2c2012-08-01 10:33:47 -0700309 if (unlikely(is_child_reaper(pid))) {
310 if (pid_ns_prepare_proc(ns))
311 goto out_free;
312 }
313
Pavel Emelyanovb461cc02007-10-18 23:40:09 -0700314 get_pid_ns(ns);
Eric W. Biederman92476d72006-03-31 02:31:42 -0800315 atomic_set(&pid->count, 1);
Eric W. Biederman92476d72006-03-31 02:31:42 -0800316 for (type = 0; type < PIDTYPE_MAX; ++type)
317 INIT_HLIST_HEAD(&pid->tasks[type]);
318
André Goddard Rosa417e3152009-12-15 16:47:40 -0800319 upid = pid->numbers + ns->level;
Eric W. Biederman92476d72006-03-31 02:31:42 -0800320 spin_lock_irq(&pidmap_lock);
Eric W. Biedermanc876ad762012-12-21 20:27:12 -0800321 if (!(ns->nr_hashed & PIDNS_HASH_ADDING))
Eric W. Biederman5e1182deb2010-07-12 18:50:25 -0700322 goto out_unlock;
Eric W. Biederman0a01f2c2012-08-01 10:33:47 -0700323 for ( ; upid >= pid->numbers; --upid) {
Pavel Emelyanov198fe212007-10-18 23:40:06 -0700324 hlist_add_head_rcu(&upid->pid_chain,
325 &pid_hash[pid_hashfn(upid->nr, upid->ns)]);
Eric W. Biederman0a01f2c2012-08-01 10:33:47 -0700326 upid->ns->nr_hashed++;
327 }
Eric W. Biederman92476d72006-03-31 02:31:42 -0800328 spin_unlock_irq(&pidmap_lock);
329
330out:
331 return pid;
332
Eric W. Biederman5e1182deb2010-07-12 18:50:25 -0700333out_unlock:
Eric W. Biederman6e666882013-02-12 13:46:23 -0800334 spin_unlock_irq(&pidmap_lock);
Eric W. Biederman92476d72006-03-31 02:31:42 -0800335out_free:
Oleg Nesterovb7127aa2008-04-30 00:54:22 -0700336 while (++i <= ns->level)
337 free_pidmap(pid->numbers + i);
Pavel Emelyanov8ef047a2007-10-18 23:40:05 -0700338
Pavel Emelianovbaf8f0f2007-10-18 23:39:48 -0700339 kmem_cache_free(ns->pid_cachep, pid);
Eric W. Biederman92476d72006-03-31 02:31:42 -0800340 pid = NULL;
341 goto out;
342}
343
Eric W. Biedermanc876ad762012-12-21 20:27:12 -0800344void disable_pid_allocation(struct pid_namespace *ns)
345{
346 spin_lock_irq(&pidmap_lock);
347 ns->nr_hashed &= ~PIDNS_HASH_ADDING;
348 spin_unlock_irq(&pidmap_lock);
349}
350
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -0800351struct pid *find_pid_ns(int nr, struct pid_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352{
Pavel Emelyanov198fe212007-10-18 23:40:06 -0700353 struct upid *pnr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354
Sasha Levinb67bfe02013-02-27 17:06:00 -0800355 hlist_for_each_entry_rcu(pnr,
Pavel Emelyanov198fe212007-10-18 23:40:06 -0700356 &pid_hash[pid_hashfn(nr, ns)], pid_chain)
357 if (pnr->nr == nr && pnr->ns == ns)
358 return container_of(pnr, struct pid,
359 numbers[ns->level]);
360
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 return NULL;
362}
Pavel Emelyanov198fe212007-10-18 23:40:06 -0700363EXPORT_SYMBOL_GPL(find_pid_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364
Pavel Emelyanov89905712007-10-18 23:40:19 -0700365struct pid *find_vpid(int nr)
366{
Eric W. Biederman17cf22c2010-03-02 14:51:53 -0800367 return find_pid_ns(nr, task_active_pid_ns(current));
Pavel Emelyanov89905712007-10-18 23:40:19 -0700368}
369EXPORT_SYMBOL_GPL(find_vpid);
370
Sukadev Bhattiprolue713d0d2007-05-10 22:22:58 -0700371/*
372 * attach_pid() must be called with the tasklist_lock write-held.
373 */
Oleg Nesterov24336ea2008-04-30 00:54:26 -0700374void attach_pid(struct task_struct *task, enum pid_type type,
Sukadev Bhattiprolue713d0d2007-05-10 22:22:58 -0700375 struct pid *pid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376{
Eric W. Biederman92476d72006-03-31 02:31:42 -0800377 struct pid_link *link;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378
Eric W. Biederman92476d72006-03-31 02:31:42 -0800379 link = &task->pids[type];
Sukadev Bhattiprolue713d0d2007-05-10 22:22:58 -0700380 link->pid = pid;
Eric W. Biederman92476d72006-03-31 02:31:42 -0800381 hlist_add_head_rcu(&link->node, &pid->tasks[type]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382}
383
Oleg Nesterov24336ea2008-04-30 00:54:26 -0700384static void __change_pid(struct task_struct *task, enum pid_type type,
385 struct pid *new)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386{
Eric W. Biederman92476d72006-03-31 02:31:42 -0800387 struct pid_link *link;
388 struct pid *pid;
389 int tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390
Eric W. Biederman92476d72006-03-31 02:31:42 -0800391 link = &task->pids[type];
392 pid = link->pid;
393
394 hlist_del_rcu(&link->node);
Oleg Nesterov24336ea2008-04-30 00:54:26 -0700395 link->pid = new;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396
397 for (tmp = PIDTYPE_MAX; --tmp >= 0; )
Eric W. Biederman92476d72006-03-31 02:31:42 -0800398 if (!hlist_empty(&pid->tasks[tmp]))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 return;
400
Eric W. Biederman92476d72006-03-31 02:31:42 -0800401 free_pid(pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402}
403
Oleg Nesterov24336ea2008-04-30 00:54:26 -0700404void detach_pid(struct task_struct *task, enum pid_type type)
405{
406 __change_pid(task, type, NULL);
407}
408
409void change_pid(struct task_struct *task, enum pid_type type,
410 struct pid *pid)
411{
412 __change_pid(task, type, pid);
413 attach_pid(task, type, pid);
414}
415
Eric W. Biedermanc18258c2006-09-27 01:51:06 -0700416/* transfer_pid is an optimization of attach_pid(new), detach_pid(old) */
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -0800417void transfer_pid(struct task_struct *old, struct task_struct *new,
Eric W. Biedermanc18258c2006-09-27 01:51:06 -0700418 enum pid_type type)
419{
420 new->pids[type].pid = old->pids[type].pid;
421 hlist_replace_rcu(&old->pids[type].node, &new->pids[type].node);
Eric W. Biedermanc18258c2006-09-27 01:51:06 -0700422}
423
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -0800424struct task_struct *pid_task(struct pid *pid, enum pid_type type)
Eric W. Biederman92476d72006-03-31 02:31:42 -0800425{
426 struct task_struct *result = NULL;
427 if (pid) {
428 struct hlist_node *first;
Arnd Bergmann67bdbff2010-02-25 16:55:13 +0100429 first = rcu_dereference_check(hlist_first_rcu(&pid->tasks[type]),
Paul E. McKenneydb1466b2010-03-03 07:46:56 -0800430 lockdep_tasklist_lock_is_held());
Eric W. Biederman92476d72006-03-31 02:31:42 -0800431 if (first)
432 result = hlist_entry(first, struct task_struct, pids[(type)].node);
433 }
434 return result;
435}
Pavel Emelyanoveccba062008-02-07 00:13:21 -0800436EXPORT_SYMBOL(pid_task);
Eric W. Biederman92476d72006-03-31 02:31:42 -0800437
438/*
Tetsuo Handa9728e5d2010-03-05 13:42:56 -0800439 * Must be called under rcu_read_lock().
Eric W. Biederman92476d72006-03-31 02:31:42 -0800440 */
Christoph Hellwig17f98dc2009-06-17 16:27:51 -0700441struct task_struct *find_task_by_pid_ns(pid_t nr, struct pid_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442{
Paul E. McKenneyb3fbab02011-05-24 08:31:09 -0700443 rcu_lockdep_assert(rcu_read_lock_held(),
444 "find_task_by_pid_ns() needs rcu_read_lock()"
445 " protection");
Christoph Hellwig17f98dc2009-06-17 16:27:51 -0700446 return pid_task(find_pid_ns(nr, ns), PIDTYPE_PID);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447}
448
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -0700449struct task_struct *find_task_by_vpid(pid_t vnr)
450{
Eric W. Biederman17cf22c2010-03-02 14:51:53 -0800451 return find_task_by_pid_ns(vnr, task_active_pid_ns(current));
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -0700452}
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -0700453
Oleg Nesterov1a657f782006-10-02 02:18:59 -0700454struct pid *get_task_pid(struct task_struct *task, enum pid_type type)
455{
456 struct pid *pid;
457 rcu_read_lock();
Oleg Nesterov2ae448e2009-04-02 16:58:36 -0700458 if (type != PIDTYPE_PID)
459 task = task->group_leader;
Oleg Nesterov1a657f782006-10-02 02:18:59 -0700460 pid = get_pid(task->pids[type].pid);
461 rcu_read_unlock();
462 return pid;
463}
Rik van Riel77c100c2011-02-01 09:51:46 -0500464EXPORT_SYMBOL_GPL(get_task_pid);
Oleg Nesterov1a657f782006-10-02 02:18:59 -0700465
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -0800466struct task_struct *get_pid_task(struct pid *pid, enum pid_type type)
Eric W. Biederman92476d72006-03-31 02:31:42 -0800467{
468 struct task_struct *result;
469 rcu_read_lock();
470 result = pid_task(pid, type);
471 if (result)
472 get_task_struct(result);
473 rcu_read_unlock();
474 return result;
475}
Rik van Riel77c100c2011-02-01 09:51:46 -0500476EXPORT_SYMBOL_GPL(get_pid_task);
Eric W. Biederman92476d72006-03-31 02:31:42 -0800477
478struct pid *find_get_pid(pid_t nr)
479{
480 struct pid *pid;
481
482 rcu_read_lock();
Pavel Emelyanov198fe212007-10-18 23:40:06 -0700483 pid = get_pid(find_vpid(nr));
Eric W. Biederman92476d72006-03-31 02:31:42 -0800484 rcu_read_unlock();
485
486 return pid;
487}
David Sterba339caf22008-07-25 01:48:31 -0700488EXPORT_SYMBOL_GPL(find_get_pid);
Eric W. Biederman92476d72006-03-31 02:31:42 -0800489
Pavel Emelyanov7af57292007-10-18 23:40:06 -0700490pid_t pid_nr_ns(struct pid *pid, struct pid_namespace *ns)
491{
492 struct upid *upid;
493 pid_t nr = 0;
494
495 if (pid && ns->level <= pid->level) {
496 upid = &pid->numbers[ns->level];
497 if (upid->ns == ns)
498 nr = upid->nr;
499 }
500 return nr;
501}
Eric W. Biederman4f82f452012-05-24 10:37:59 -0600502EXPORT_SYMBOL_GPL(pid_nr_ns);
Pavel Emelyanov7af57292007-10-18 23:40:06 -0700503
Eric W. Biederman44c4e1b2008-02-08 04:19:15 -0800504pid_t pid_vnr(struct pid *pid)
505{
Eric W. Biederman17cf22c2010-03-02 14:51:53 -0800506 return pid_nr_ns(pid, task_active_pid_ns(current));
Eric W. Biederman44c4e1b2008-02-08 04:19:15 -0800507}
508EXPORT_SYMBOL_GPL(pid_vnr);
509
Oleg Nesterov52ee2df2009-04-02 16:58:38 -0700510pid_t __task_pid_nr_ns(struct task_struct *task, enum pid_type type,
511 struct pid_namespace *ns)
Pavel Emelyanov2f2a3a42007-10-18 23:40:19 -0700512{
Oleg Nesterov52ee2df2009-04-02 16:58:38 -0700513 pid_t nr = 0;
514
515 rcu_read_lock();
516 if (!ns)
Eric W. Biederman17cf22c2010-03-02 14:51:53 -0800517 ns = task_active_pid_ns(current);
Oleg Nesterov52ee2df2009-04-02 16:58:38 -0700518 if (likely(pid_alive(task))) {
519 if (type != PIDTYPE_PID)
520 task = task->group_leader;
521 nr = pid_nr_ns(task->pids[type].pid, ns);
522 }
523 rcu_read_unlock();
524
525 return nr;
Pavel Emelyanov2f2a3a42007-10-18 23:40:19 -0700526}
Oleg Nesterov52ee2df2009-04-02 16:58:38 -0700527EXPORT_SYMBOL(__task_pid_nr_ns);
Pavel Emelyanov2f2a3a42007-10-18 23:40:19 -0700528
529pid_t task_tgid_nr_ns(struct task_struct *tsk, struct pid_namespace *ns)
530{
531 return pid_nr_ns(task_tgid(tsk), ns);
532}
533EXPORT_SYMBOL(task_tgid_nr_ns);
534
Eric W. Biederman61bce0f2009-01-07 18:08:49 -0800535struct pid_namespace *task_active_pid_ns(struct task_struct *tsk)
536{
537 return ns_of_pid(task_pid(tsk));
538}
539EXPORT_SYMBOL_GPL(task_active_pid_ns);
540
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541/*
Frederik Schwarzer025dfda2008-10-16 19:02:37 +0200542 * Used by proc to find the first pid that is greater than or equal to nr.
Eric W. Biederman0804ef42006-10-02 02:17:04 -0700543 *
Pavel Emelyanove49859e2008-07-25 01:48:36 -0700544 * 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 -0700545 */
Pavel Emelyanov198fe212007-10-18 23:40:06 -0700546struct pid *find_ge_pid(int nr, struct pid_namespace *ns)
Eric W. Biederman0804ef42006-10-02 02:17:04 -0700547{
548 struct pid *pid;
549
550 do {
Pavel Emelyanov198fe212007-10-18 23:40:06 -0700551 pid = find_pid_ns(nr, ns);
Eric W. Biederman0804ef42006-10-02 02:17:04 -0700552 if (pid)
553 break;
Pavel Emelyanov198fe212007-10-18 23:40:06 -0700554 nr = next_pidmap(ns, nr);
Eric W. Biederman0804ef42006-10-02 02:17:04 -0700555 } while (nr > 0);
556
557 return pid;
558}
559
560/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 * The pid hash table is scaled according to the amount of memory in the
562 * machine. From a minimum of 16 slots up to 4096 slots at one gigabyte or
563 * more.
564 */
565void __init pidhash_init(void)
566{
Dimitri Sivanich074b8512012-02-08 12:39:07 -0800567 unsigned int i, pidhash_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568
Jan Beulich2c85f512009-09-21 17:03:07 -0700569 pid_hash = alloc_large_system_hash("PID", sizeof(*pid_hash), 0, 18,
570 HASH_EARLY | HASH_SMALL,
Tim Bird31fe62b2012-05-23 13:33:35 +0000571 &pidhash_shift, NULL,
572 0, 4096);
Dimitri Sivanich074b8512012-02-08 12:39:07 -0800573 pidhash_size = 1U << pidhash_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574
Eric W. Biederman92476d72006-03-31 02:31:42 -0800575 for (i = 0; i < pidhash_size; i++)
576 INIT_HLIST_HEAD(&pid_hash[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577}
578
579void __init pidmap_init(void)
580{
Eric W. Biedermanc876ad762012-12-21 20:27:12 -0800581 /* Veryify no one has done anything silly */
582 BUILD_BUG_ON(PID_MAX_LIMIT >= PIDNS_HASH_ADDING);
583
Hedi Berriche72680a12010-05-26 14:44:06 -0700584 /* bump default and minimum pid_max based on number of cpus */
585 pid_max = min(pid_max_max, max_t(int, pid_max,
586 PIDS_PER_CPU_DEFAULT * num_possible_cpus()));
587 pid_max_min = max_t(int, pid_max_min,
588 PIDS_PER_CPU_MIN * num_possible_cpus());
589 pr_info("pid_max: default: %u minimum: %u\n", pid_max, pid_max_min);
590
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -0800591 init_pid_ns.pidmap[0].page = kzalloc(PAGE_SIZE, GFP_KERNEL);
Oleg Nesterov73b9ebf2006-03-28 16:11:07 -0800592 /* Reserve PID 0. We never call free_pidmap(0) */
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -0800593 set_bit(0, init_pid_ns.pidmap[0].page);
594 atomic_dec(&init_pid_ns.pidmap[0].nr_free);
Eric W. Biedermanc876ad762012-12-21 20:27:12 -0800595 init_pid_ns.nr_hashed = PIDNS_HASH_ADDING;
Eric W. Biederman92476d72006-03-31 02:31:42 -0800596
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800597 init_pid_ns.pid_cachep = KMEM_CACHE(pid,
598 SLAB_HWCACHE_ALIGN | SLAB_PANIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599}