blob: d76f59326bd448724b5f0236d3ea604571bd3d29 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Generic pidhash and scalable, time-bounded PID allocator
3 *
4 * (C) 2002-2003 William Irwin, IBM
5 * (C) 2004 William Irwin, Oracle
6 * (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).
21 */
22
23#include <linux/mm.h>
24#include <linux/module.h>
25#include <linux/slab.h>
26#include <linux/init.h>
27#include <linux/bootmem.h>
28#include <linux/hash.h>
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -080029#include <linux/pid_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030
31#define pid_hashfn(nr) hash_long((unsigned long)nr, pidhash_shift)
Eric W. Biederman92476d72006-03-31 02:31:42 -080032static struct hlist_head *pid_hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -070033static int pidhash_shift;
Christoph Lametere18b8902006-12-06 20:33:20 -080034static struct kmem_cache *pid_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070035
36int pid_max = PID_MAX_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
38#define RESERVED_PIDS 300
39
40int pid_max_min = RESERVED_PIDS + 1;
41int pid_max_max = PID_MAX_LIMIT;
42
Linus Torvalds1da177e2005-04-16 15:20:36 -070043#define BITS_PER_PAGE (PAGE_SIZE*8)
44#define BITS_PER_PAGE_MASK (BITS_PER_PAGE-1)
Sukadev Bhattiprolu3fbc9642006-10-02 02:17:24 -070045
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -080046static inline int mk_pid(struct pid_namespace *pid_ns,
47 struct pidmap *map, int off)
Sukadev Bhattiprolu3fbc9642006-10-02 02:17:24 -070048{
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -080049 return (map - pid_ns->pidmap)*BITS_PER_PAGE + off;
Sukadev Bhattiprolu3fbc9642006-10-02 02:17:24 -070050}
51
Linus Torvalds1da177e2005-04-16 15:20:36 -070052#define find_next_offset(map, off) \
53 find_next_zero_bit((map)->page, BITS_PER_PAGE, off)
54
55/*
56 * PID-map pages start out as NULL, they get allocated upon
57 * first use and are never deallocated. This way a low pid_max
58 * value does not cause lots of bitmaps to be allocated, but
59 * the scheme scales to up to 4 million PIDs, runtime.
60 */
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -080061struct pid_namespace init_pid_ns = {
Cedric Le Goater9a575a92006-12-08 02:37:59 -080062 .kref = {
63 .refcount = ATOMIC_INIT(2),
64 },
Sukadev Bhattiprolu3fbc9642006-10-02 02:17:24 -070065 .pidmap = {
66 [ 0 ... PIDMAP_ENTRIES-1] = { ATOMIC_INIT(BITS_PER_PAGE), NULL }
67 },
Sukadev Bhattiprolu84d73782006-12-08 02:38:01 -080068 .last_pid = 0,
69 .child_reaper = &init_task
Sukadev Bhattiprolu3fbc9642006-10-02 02:17:24 -070070};
Linus Torvalds1da177e2005-04-16 15:20:36 -070071
Eric W. Biederman92476d72006-03-31 02:31:42 -080072/*
73 * Note: disable interrupts while the pidmap_lock is held as an
74 * interrupt might come in and do read_lock(&tasklist_lock).
75 *
76 * If we don't disable interrupts there is a nasty deadlock between
77 * detach_pid()->free_pid() and another cpu that does
78 * spin_lock(&pidmap_lock) followed by an interrupt routine that does
79 * read_lock(&tasklist_lock);
80 *
81 * After we clean up the tasklist_lock and know there are no
82 * irq handlers that take it we can leave the interrupts enabled.
83 * For now it is easier to be safe than to prove it can't happen.
84 */
Sukadev Bhattiprolu3fbc9642006-10-02 02:17:24 -070085
Linus Torvalds1da177e2005-04-16 15:20:36 -070086static __cacheline_aligned_in_smp DEFINE_SPINLOCK(pidmap_lock);
87
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -080088static fastcall void free_pidmap(struct pid_namespace *pid_ns, int pid)
Linus Torvalds1da177e2005-04-16 15:20:36 -070089{
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -080090 struct pidmap *map = pid_ns->pidmap + pid / BITS_PER_PAGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -070091 int offset = pid & BITS_PER_PAGE_MASK;
92
93 clear_bit(offset, map->page);
94 atomic_inc(&map->nr_free);
95}
96
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -080097static int alloc_pidmap(struct pid_namespace *pid_ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -070098{
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -080099 int i, offset, max_scan, pid, last = pid_ns->last_pid;
Sukadev Bhattiprolu6a1f3b82006-10-02 02:17:20 -0700100 struct pidmap *map;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101
102 pid = last + 1;
103 if (pid >= pid_max)
104 pid = RESERVED_PIDS;
105 offset = pid & BITS_PER_PAGE_MASK;
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -0800106 map = &pid_ns->pidmap[pid/BITS_PER_PAGE];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 max_scan = (pid_max + BITS_PER_PAGE - 1)/BITS_PER_PAGE - !offset;
108 for (i = 0; i <= max_scan; ++i) {
109 if (unlikely(!map->page)) {
Sukadev Bhattiprolu3fbc9642006-10-02 02:17:24 -0700110 void *page = kzalloc(PAGE_SIZE, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 /*
112 * Free the page if someone raced with us
113 * installing it:
114 */
Eric W. Biederman92476d72006-03-31 02:31:42 -0800115 spin_lock_irq(&pidmap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116 if (map->page)
Sukadev Bhattiprolu3fbc9642006-10-02 02:17:24 -0700117 kfree(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118 else
Sukadev Bhattiprolu3fbc9642006-10-02 02:17:24 -0700119 map->page = page;
Eric W. Biederman92476d72006-03-31 02:31:42 -0800120 spin_unlock_irq(&pidmap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121 if (unlikely(!map->page))
122 break;
123 }
124 if (likely(atomic_read(&map->nr_free))) {
125 do {
126 if (!test_and_set_bit(offset, map->page)) {
127 atomic_dec(&map->nr_free);
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -0800128 pid_ns->last_pid = pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129 return pid;
130 }
131 offset = find_next_offset(map, offset);
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -0800132 pid = mk_pid(pid_ns, map, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133 /*
134 * find_next_offset() found a bit, the pid from it
135 * is in-bounds, and if we fell back to the last
136 * bitmap block and the final block was the same
137 * as the starting point, pid is before last_pid.
138 */
139 } while (offset < BITS_PER_PAGE && pid < pid_max &&
140 (i != max_scan || pid < last ||
141 !((last+1) & BITS_PER_PAGE_MASK)));
142 }
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -0800143 if (map < &pid_ns->pidmap[(pid_max-1)/BITS_PER_PAGE]) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 ++map;
145 offset = 0;
146 } else {
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -0800147 map = &pid_ns->pidmap[0];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148 offset = RESERVED_PIDS;
149 if (unlikely(last == offset))
150 break;
151 }
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -0800152 pid = mk_pid(pid_ns, map, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153 }
154 return -1;
155}
156
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -0800157static int next_pidmap(struct pid_namespace *pid_ns, int last)
Eric W. Biederman0804ef42006-10-02 02:17:04 -0700158{
159 int offset;
Eric W. Biedermanf40f50d2006-10-02 02:17:25 -0700160 struct pidmap *map, *end;
Eric W. Biederman0804ef42006-10-02 02:17:04 -0700161
162 offset = (last + 1) & BITS_PER_PAGE_MASK;
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -0800163 map = &pid_ns->pidmap[(last + 1)/BITS_PER_PAGE];
164 end = &pid_ns->pidmap[PIDMAP_ENTRIES];
Eric W. Biedermanf40f50d2006-10-02 02:17:25 -0700165 for (; map < end; map++, offset = 0) {
Eric W. Biederman0804ef42006-10-02 02:17:04 -0700166 if (unlikely(!map->page))
167 continue;
168 offset = find_next_bit((map)->page, BITS_PER_PAGE, offset);
169 if (offset < BITS_PER_PAGE)
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -0800170 return mk_pid(pid_ns, map, offset);
Eric W. Biederman0804ef42006-10-02 02:17:04 -0700171 }
172 return -1;
173}
174
Eric W. Biederman92476d72006-03-31 02:31:42 -0800175fastcall void put_pid(struct pid *pid)
176{
177 if (!pid)
178 return;
179 if ((atomic_read(&pid->count) == 1) ||
180 atomic_dec_and_test(&pid->count))
181 kmem_cache_free(pid_cachep, pid);
182}
Eric W. Biedermanbbf73142006-10-02 02:17:11 -0700183EXPORT_SYMBOL_GPL(put_pid);
Eric W. Biederman92476d72006-03-31 02:31:42 -0800184
185static void delayed_put_pid(struct rcu_head *rhp)
186{
187 struct pid *pid = container_of(rhp, struct pid, rcu);
188 put_pid(pid);
189}
190
191fastcall void free_pid(struct pid *pid)
192{
193 /* We can be called with write_lock_irq(&tasklist_lock) held */
194 unsigned long flags;
195
196 spin_lock_irqsave(&pidmap_lock, flags);
197 hlist_del_rcu(&pid->pid_chain);
198 spin_unlock_irqrestore(&pidmap_lock, flags);
199
Serge E. Hallyn0f245282007-01-30 15:28:23 -0600200 free_pidmap(&init_pid_ns, pid->nr);
Eric W. Biederman92476d72006-03-31 02:31:42 -0800201 call_rcu(&pid->rcu, delayed_put_pid);
202}
203
204struct pid *alloc_pid(void)
205{
206 struct pid *pid;
207 enum pid_type type;
208 int nr = -1;
209
210 pid = kmem_cache_alloc(pid_cachep, GFP_KERNEL);
211 if (!pid)
212 goto out;
213
Cedric Le Goater6cc1b222006-12-08 02:38:00 -0800214 nr = alloc_pidmap(current->nsproxy->pid_ns);
Eric W. Biederman92476d72006-03-31 02:31:42 -0800215 if (nr < 0)
216 goto out_free;
217
218 atomic_set(&pid->count, 1);
219 pid->nr = nr;
220 for (type = 0; type < PIDTYPE_MAX; ++type)
221 INIT_HLIST_HEAD(&pid->tasks[type]);
222
223 spin_lock_irq(&pidmap_lock);
224 hlist_add_head_rcu(&pid->pid_chain, &pid_hash[pid_hashfn(pid->nr)]);
225 spin_unlock_irq(&pidmap_lock);
226
227out:
228 return pid;
229
230out_free:
231 kmem_cache_free(pid_cachep, pid);
232 pid = NULL;
233 goto out;
234}
235
236struct pid * fastcall find_pid(int nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237{
238 struct hlist_node *elem;
239 struct pid *pid;
240
Ingo Molnare56d0902006-01-08 01:01:37 -0800241 hlist_for_each_entry_rcu(pid, elem,
Eric W. Biederman92476d72006-03-31 02:31:42 -0800242 &pid_hash[pid_hashfn(nr)], pid_chain) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243 if (pid->nr == nr)
244 return pid;
245 }
246 return NULL;
247}
Eric W. Biedermanbbf73142006-10-02 02:17:11 -0700248EXPORT_SYMBOL_GPL(find_pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249
Sukadev Bhattiprolue713d0d2007-05-10 22:22:58 -0700250/*
251 * attach_pid() must be called with the tasklist_lock write-held.
252 */
253int fastcall attach_pid(struct task_struct *task, enum pid_type type,
254 struct pid *pid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255{
Eric W. Biederman92476d72006-03-31 02:31:42 -0800256 struct pid_link *link;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257
Eric W. Biederman92476d72006-03-31 02:31:42 -0800258 link = &task->pids[type];
Sukadev Bhattiprolue713d0d2007-05-10 22:22:58 -0700259 link->pid = pid;
Eric W. Biederman92476d72006-03-31 02:31:42 -0800260 hlist_add_head_rcu(&link->node, &pid->tasks[type]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261
262 return 0;
263}
264
Ingo Molnar36c8b582006-07-03 00:25:41 -0700265void fastcall detach_pid(struct task_struct *task, enum pid_type type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266{
Eric W. Biederman92476d72006-03-31 02:31:42 -0800267 struct pid_link *link;
268 struct pid *pid;
269 int tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270
Eric W. Biederman92476d72006-03-31 02:31:42 -0800271 link = &task->pids[type];
272 pid = link->pid;
273
274 hlist_del_rcu(&link->node);
275 link->pid = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276
277 for (tmp = PIDTYPE_MAX; --tmp >= 0; )
Eric W. Biederman92476d72006-03-31 02:31:42 -0800278 if (!hlist_empty(&pid->tasks[tmp]))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279 return;
280
Eric W. Biederman92476d72006-03-31 02:31:42 -0800281 free_pid(pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282}
283
Eric W. Biedermanc18258c2006-09-27 01:51:06 -0700284/* transfer_pid is an optimization of attach_pid(new), detach_pid(old) */
285void fastcall transfer_pid(struct task_struct *old, struct task_struct *new,
286 enum pid_type type)
287{
288 new->pids[type].pid = old->pids[type].pid;
289 hlist_replace_rcu(&old->pids[type].node, &new->pids[type].node);
290 old->pids[type].pid = NULL;
291}
292
Eric W. Biederman92476d72006-03-31 02:31:42 -0800293struct task_struct * fastcall pid_task(struct pid *pid, enum pid_type type)
294{
295 struct task_struct *result = NULL;
296 if (pid) {
297 struct hlist_node *first;
298 first = rcu_dereference(pid->tasks[type].first);
299 if (first)
300 result = hlist_entry(first, struct task_struct, pids[(type)].node);
301 }
302 return result;
303}
304
305/*
306 * Must be called under rcu_read_lock() or with tasklist_lock read-held.
307 */
Ingo Molnar36c8b582006-07-03 00:25:41 -0700308struct task_struct *find_task_by_pid_type(int type, int nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309{
Eric W. Biederman92476d72006-03-31 02:31:42 -0800310 return pid_task(find_pid(nr), type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311}
312
313EXPORT_SYMBOL(find_task_by_pid_type);
314
Oleg Nesterov1a657f782006-10-02 02:18:59 -0700315struct pid *get_task_pid(struct task_struct *task, enum pid_type type)
316{
317 struct pid *pid;
318 rcu_read_lock();
319 pid = get_pid(task->pids[type].pid);
320 rcu_read_unlock();
321 return pid;
322}
323
Eric W. Biederman92476d72006-03-31 02:31:42 -0800324struct task_struct *fastcall get_pid_task(struct pid *pid, enum pid_type type)
325{
326 struct task_struct *result;
327 rcu_read_lock();
328 result = pid_task(pid, type);
329 if (result)
330 get_task_struct(result);
331 rcu_read_unlock();
332 return result;
333}
334
335struct pid *find_get_pid(pid_t nr)
336{
337 struct pid *pid;
338
339 rcu_read_lock();
340 pid = get_pid(find_pid(nr));
341 rcu_read_unlock();
342
343 return pid;
344}
345
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346/*
Eric W. Biederman0804ef42006-10-02 02:17:04 -0700347 * Used by proc to find the first pid that is greater then or equal to nr.
348 *
349 * If there is a pid at nr this function is exactly the same as find_pid.
350 */
351struct pid *find_ge_pid(int nr)
352{
353 struct pid *pid;
354
355 do {
356 pid = find_pid(nr);
357 if (pid)
358 break;
Cedric Le Goater6cc1b222006-12-08 02:38:00 -0800359 nr = next_pidmap(current->nsproxy->pid_ns, nr);
Eric W. Biederman0804ef42006-10-02 02:17:04 -0700360 } while (nr > 0);
361
362 return pid;
363}
Eric W. Biedermanbbf73142006-10-02 02:17:11 -0700364EXPORT_SYMBOL_GPL(find_get_pid);
Eric W. Biederman0804ef42006-10-02 02:17:04 -0700365
Badari Pulavartye3222c42007-05-08 00:25:21 -0700366struct pid_namespace *copy_pid_ns(int flags, struct pid_namespace *old_ns)
Cedric Le Goater9a575a92006-12-08 02:37:59 -0800367{
Badari Pulavartye3222c42007-05-08 00:25:21 -0700368 BUG_ON(!old_ns);
Cedric Le Goater9a575a92006-12-08 02:37:59 -0800369 get_pid_ns(old_ns);
Badari Pulavartye3222c42007-05-08 00:25:21 -0700370 return old_ns;
Cedric Le Goater9a575a92006-12-08 02:37:59 -0800371}
372
373void free_pid_ns(struct kref *kref)
374{
375 struct pid_namespace *ns;
376
377 ns = container_of(kref, struct pid_namespace, kref);
378 kfree(ns);
379}
380
Eric W. Biederman0804ef42006-10-02 02:17:04 -0700381/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 * The pid hash table is scaled according to the amount of memory in the
383 * machine. From a minimum of 16 slots up to 4096 slots at one gigabyte or
384 * more.
385 */
386void __init pidhash_init(void)
387{
Eric W. Biederman92476d72006-03-31 02:31:42 -0800388 int i, pidhash_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 unsigned long megabytes = nr_kernel_pages >> (20 - PAGE_SHIFT);
390
391 pidhash_shift = max(4, fls(megabytes * 4));
392 pidhash_shift = min(12, pidhash_shift);
393 pidhash_size = 1 << pidhash_shift;
394
395 printk("PID hash table entries: %d (order: %d, %Zd bytes)\n",
396 pidhash_size, pidhash_shift,
Eric W. Biederman92476d72006-03-31 02:31:42 -0800397 pidhash_size * sizeof(struct hlist_head));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398
Eric W. Biederman92476d72006-03-31 02:31:42 -0800399 pid_hash = alloc_bootmem(pidhash_size * sizeof(*(pid_hash)));
400 if (!pid_hash)
401 panic("Could not alloc pidhash!\n");
402 for (i = 0; i < pidhash_size; i++)
403 INIT_HLIST_HEAD(&pid_hash[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404}
405
406void __init pidmap_init(void)
407{
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -0800408 init_pid_ns.pidmap[0].page = kzalloc(PAGE_SIZE, GFP_KERNEL);
Oleg Nesterov73b9ebf2006-03-28 16:11:07 -0800409 /* Reserve PID 0. We never call free_pidmap(0) */
Sukadev Bhattiprolu61a58c62006-12-08 02:37:58 -0800410 set_bit(0, init_pid_ns.pidmap[0].page);
411 atomic_dec(&init_pid_ns.pidmap[0].nr_free);
Eric W. Biederman92476d72006-03-31 02:31:42 -0800412
Christoph Lameter0a31bd52007-05-06 14:49:57 -0700413 pid_cachep = KMEM_CACHE(pid, SLAB_PANIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414}