Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
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> | |
61a58c6c | 29 | #include <linux/pid_namespace.h> |
1da177e4 LT |
30 | |
31 | #define pid_hashfn(nr) hash_long((unsigned long)nr, pidhash_shift) | |
92476d7f | 32 | static struct hlist_head *pid_hash; |
1da177e4 | 33 | static int pidhash_shift; |
e18b890b | 34 | static struct kmem_cache *pid_cachep; |
1da177e4 LT |
35 | |
36 | int pid_max = PID_MAX_DEFAULT; | |
1da177e4 LT |
37 | |
38 | #define RESERVED_PIDS 300 | |
39 | ||
40 | int pid_max_min = RESERVED_PIDS + 1; | |
41 | int pid_max_max = PID_MAX_LIMIT; | |
42 | ||
1da177e4 LT |
43 | #define BITS_PER_PAGE (PAGE_SIZE*8) |
44 | #define BITS_PER_PAGE_MASK (BITS_PER_PAGE-1) | |
3fbc9648 | 45 | |
61a58c6c SB |
46 | static inline int mk_pid(struct pid_namespace *pid_ns, |
47 | struct pidmap *map, int off) | |
3fbc9648 | 48 | { |
61a58c6c | 49 | return (map - pid_ns->pidmap)*BITS_PER_PAGE + off; |
3fbc9648 SB |
50 | } |
51 | ||
1da177e4 LT |
52 | #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 | */ | |
61a58c6c | 61 | struct pid_namespace init_pid_ns = { |
3fbc9648 SB |
62 | .pidmap = { |
63 | [ 0 ... PIDMAP_ENTRIES-1] = { ATOMIC_INIT(BITS_PER_PAGE), NULL } | |
64 | }, | |
65 | .last_pid = 0 | |
66 | }; | |
1da177e4 | 67 | |
92476d7f EB |
68 | /* |
69 | * Note: disable interrupts while the pidmap_lock is held as an | |
70 | * interrupt might come in and do read_lock(&tasklist_lock). | |
71 | * | |
72 | * If we don't disable interrupts there is a nasty deadlock between | |
73 | * detach_pid()->free_pid() and another cpu that does | |
74 | * spin_lock(&pidmap_lock) followed by an interrupt routine that does | |
75 | * read_lock(&tasklist_lock); | |
76 | * | |
77 | * After we clean up the tasklist_lock and know there are no | |
78 | * irq handlers that take it we can leave the interrupts enabled. | |
79 | * For now it is easier to be safe than to prove it can't happen. | |
80 | */ | |
3fbc9648 | 81 | |
1da177e4 LT |
82 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(pidmap_lock); |
83 | ||
61a58c6c | 84 | static fastcall void free_pidmap(struct pid_namespace *pid_ns, int pid) |
1da177e4 | 85 | { |
61a58c6c | 86 | struct pidmap *map = pid_ns->pidmap + pid / BITS_PER_PAGE; |
1da177e4 LT |
87 | int offset = pid & BITS_PER_PAGE_MASK; |
88 | ||
89 | clear_bit(offset, map->page); | |
90 | atomic_inc(&map->nr_free); | |
91 | } | |
92 | ||
61a58c6c | 93 | static int alloc_pidmap(struct pid_namespace *pid_ns) |
1da177e4 | 94 | { |
61a58c6c | 95 | int i, offset, max_scan, pid, last = pid_ns->last_pid; |
6a1f3b84 | 96 | struct pidmap *map; |
1da177e4 LT |
97 | |
98 | pid = last + 1; | |
99 | if (pid >= pid_max) | |
100 | pid = RESERVED_PIDS; | |
101 | offset = pid & BITS_PER_PAGE_MASK; | |
61a58c6c | 102 | map = &pid_ns->pidmap[pid/BITS_PER_PAGE]; |
1da177e4 LT |
103 | max_scan = (pid_max + BITS_PER_PAGE - 1)/BITS_PER_PAGE - !offset; |
104 | for (i = 0; i <= max_scan; ++i) { | |
105 | if (unlikely(!map->page)) { | |
3fbc9648 | 106 | void *page = kzalloc(PAGE_SIZE, GFP_KERNEL); |
1da177e4 LT |
107 | /* |
108 | * Free the page if someone raced with us | |
109 | * installing it: | |
110 | */ | |
92476d7f | 111 | spin_lock_irq(&pidmap_lock); |
1da177e4 | 112 | if (map->page) |
3fbc9648 | 113 | kfree(page); |
1da177e4 | 114 | else |
3fbc9648 | 115 | map->page = page; |
92476d7f | 116 | spin_unlock_irq(&pidmap_lock); |
1da177e4 LT |
117 | if (unlikely(!map->page)) |
118 | break; | |
119 | } | |
120 | if (likely(atomic_read(&map->nr_free))) { | |
121 | do { | |
122 | if (!test_and_set_bit(offset, map->page)) { | |
123 | atomic_dec(&map->nr_free); | |
61a58c6c | 124 | pid_ns->last_pid = pid; |
1da177e4 LT |
125 | return pid; |
126 | } | |
127 | offset = find_next_offset(map, offset); | |
61a58c6c | 128 | pid = mk_pid(pid_ns, map, offset); |
1da177e4 LT |
129 | /* |
130 | * find_next_offset() found a bit, the pid from it | |
131 | * is in-bounds, and if we fell back to the last | |
132 | * bitmap block and the final block was the same | |
133 | * as the starting point, pid is before last_pid. | |
134 | */ | |
135 | } while (offset < BITS_PER_PAGE && pid < pid_max && | |
136 | (i != max_scan || pid < last || | |
137 | !((last+1) & BITS_PER_PAGE_MASK))); | |
138 | } | |
61a58c6c | 139 | if (map < &pid_ns->pidmap[(pid_max-1)/BITS_PER_PAGE]) { |
1da177e4 LT |
140 | ++map; |
141 | offset = 0; | |
142 | } else { | |
61a58c6c | 143 | map = &pid_ns->pidmap[0]; |
1da177e4 LT |
144 | offset = RESERVED_PIDS; |
145 | if (unlikely(last == offset)) | |
146 | break; | |
147 | } | |
61a58c6c | 148 | pid = mk_pid(pid_ns, map, offset); |
1da177e4 LT |
149 | } |
150 | return -1; | |
151 | } | |
152 | ||
61a58c6c | 153 | static int next_pidmap(struct pid_namespace *pid_ns, int last) |
0804ef4b EB |
154 | { |
155 | int offset; | |
f40f50d3 | 156 | struct pidmap *map, *end; |
0804ef4b EB |
157 | |
158 | offset = (last + 1) & BITS_PER_PAGE_MASK; | |
61a58c6c SB |
159 | map = &pid_ns->pidmap[(last + 1)/BITS_PER_PAGE]; |
160 | end = &pid_ns->pidmap[PIDMAP_ENTRIES]; | |
f40f50d3 | 161 | for (; map < end; map++, offset = 0) { |
0804ef4b EB |
162 | if (unlikely(!map->page)) |
163 | continue; | |
164 | offset = find_next_bit((map)->page, BITS_PER_PAGE, offset); | |
165 | if (offset < BITS_PER_PAGE) | |
61a58c6c | 166 | return mk_pid(pid_ns, map, offset); |
0804ef4b EB |
167 | } |
168 | return -1; | |
169 | } | |
170 | ||
92476d7f EB |
171 | fastcall void put_pid(struct pid *pid) |
172 | { | |
173 | if (!pid) | |
174 | return; | |
175 | if ((atomic_read(&pid->count) == 1) || | |
176 | atomic_dec_and_test(&pid->count)) | |
177 | kmem_cache_free(pid_cachep, pid); | |
178 | } | |
bbf73147 | 179 | EXPORT_SYMBOL_GPL(put_pid); |
92476d7f EB |
180 | |
181 | static void delayed_put_pid(struct rcu_head *rhp) | |
182 | { | |
183 | struct pid *pid = container_of(rhp, struct pid, rcu); | |
184 | put_pid(pid); | |
185 | } | |
186 | ||
187 | fastcall void free_pid(struct pid *pid) | |
188 | { | |
189 | /* We can be called with write_lock_irq(&tasklist_lock) held */ | |
190 | unsigned long flags; | |
191 | ||
192 | spin_lock_irqsave(&pidmap_lock, flags); | |
193 | hlist_del_rcu(&pid->pid_chain); | |
194 | spin_unlock_irqrestore(&pidmap_lock, flags); | |
195 | ||
61a58c6c | 196 | free_pidmap(&init_pid_ns, pid->nr); |
92476d7f EB |
197 | call_rcu(&pid->rcu, delayed_put_pid); |
198 | } | |
199 | ||
200 | struct pid *alloc_pid(void) | |
201 | { | |
202 | struct pid *pid; | |
203 | enum pid_type type; | |
204 | int nr = -1; | |
205 | ||
206 | pid = kmem_cache_alloc(pid_cachep, GFP_KERNEL); | |
207 | if (!pid) | |
208 | goto out; | |
209 | ||
61a58c6c | 210 | nr = alloc_pidmap(&init_pid_ns); |
92476d7f EB |
211 | if (nr < 0) |
212 | goto out_free; | |
213 | ||
214 | atomic_set(&pid->count, 1); | |
215 | pid->nr = nr; | |
216 | for (type = 0; type < PIDTYPE_MAX; ++type) | |
217 | INIT_HLIST_HEAD(&pid->tasks[type]); | |
218 | ||
219 | spin_lock_irq(&pidmap_lock); | |
220 | hlist_add_head_rcu(&pid->pid_chain, &pid_hash[pid_hashfn(pid->nr)]); | |
221 | spin_unlock_irq(&pidmap_lock); | |
222 | ||
223 | out: | |
224 | return pid; | |
225 | ||
226 | out_free: | |
227 | kmem_cache_free(pid_cachep, pid); | |
228 | pid = NULL; | |
229 | goto out; | |
230 | } | |
231 | ||
232 | struct pid * fastcall find_pid(int nr) | |
1da177e4 LT |
233 | { |
234 | struct hlist_node *elem; | |
235 | struct pid *pid; | |
236 | ||
e56d0903 | 237 | hlist_for_each_entry_rcu(pid, elem, |
92476d7f | 238 | &pid_hash[pid_hashfn(nr)], pid_chain) { |
1da177e4 LT |
239 | if (pid->nr == nr) |
240 | return pid; | |
241 | } | |
242 | return NULL; | |
243 | } | |
bbf73147 | 244 | EXPORT_SYMBOL_GPL(find_pid); |
1da177e4 | 245 | |
36c8b586 | 246 | int fastcall attach_pid(struct task_struct *task, enum pid_type type, int nr) |
1da177e4 | 247 | { |
92476d7f EB |
248 | struct pid_link *link; |
249 | struct pid *pid; | |
250 | ||
92476d7f EB |
251 | link = &task->pids[type]; |
252 | link->pid = pid = find_pid(nr); | |
253 | hlist_add_head_rcu(&link->node, &pid->tasks[type]); | |
1da177e4 LT |
254 | |
255 | return 0; | |
256 | } | |
257 | ||
36c8b586 | 258 | void fastcall detach_pid(struct task_struct *task, enum pid_type type) |
1da177e4 | 259 | { |
92476d7f EB |
260 | struct pid_link *link; |
261 | struct pid *pid; | |
262 | int tmp; | |
1da177e4 | 263 | |
92476d7f EB |
264 | link = &task->pids[type]; |
265 | pid = link->pid; | |
1da177e4 | 266 | |
92476d7f EB |
267 | hlist_del_rcu(&link->node); |
268 | link->pid = NULL; | |
1da177e4 | 269 | |
92476d7f EB |
270 | for (tmp = PIDTYPE_MAX; --tmp >= 0; ) |
271 | if (!hlist_empty(&pid->tasks[tmp])) | |
272 | return; | |
1da177e4 | 273 | |
92476d7f | 274 | free_pid(pid); |
1da177e4 LT |
275 | } |
276 | ||
c18258c6 EB |
277 | /* transfer_pid is an optimization of attach_pid(new), detach_pid(old) */ |
278 | void fastcall transfer_pid(struct task_struct *old, struct task_struct *new, | |
279 | enum pid_type type) | |
280 | { | |
281 | new->pids[type].pid = old->pids[type].pid; | |
282 | hlist_replace_rcu(&old->pids[type].node, &new->pids[type].node); | |
283 | old->pids[type].pid = NULL; | |
284 | } | |
285 | ||
92476d7f | 286 | struct task_struct * fastcall pid_task(struct pid *pid, enum pid_type type) |
1da177e4 | 287 | { |
92476d7f EB |
288 | struct task_struct *result = NULL; |
289 | if (pid) { | |
290 | struct hlist_node *first; | |
291 | first = rcu_dereference(pid->tasks[type].first); | |
292 | if (first) | |
293 | result = hlist_entry(first, struct task_struct, pids[(type)].node); | |
294 | } | |
295 | return result; | |
296 | } | |
1da177e4 | 297 | |
92476d7f EB |
298 | /* |
299 | * Must be called under rcu_read_lock() or with tasklist_lock read-held. | |
300 | */ | |
36c8b586 | 301 | struct task_struct *find_task_by_pid_type(int type, int nr) |
92476d7f EB |
302 | { |
303 | return pid_task(find_pid(nr), type); | |
304 | } | |
1da177e4 | 305 | |
92476d7f | 306 | EXPORT_SYMBOL(find_task_by_pid_type); |
1da177e4 | 307 | |
1a657f78 ON |
308 | struct pid *get_task_pid(struct task_struct *task, enum pid_type type) |
309 | { | |
310 | struct pid *pid; | |
311 | rcu_read_lock(); | |
312 | pid = get_pid(task->pids[type].pid); | |
313 | rcu_read_unlock(); | |
314 | return pid; | |
315 | } | |
316 | ||
92476d7f EB |
317 | struct task_struct *fastcall get_pid_task(struct pid *pid, enum pid_type type) |
318 | { | |
319 | struct task_struct *result; | |
320 | rcu_read_lock(); | |
321 | result = pid_task(pid, type); | |
322 | if (result) | |
323 | get_task_struct(result); | |
324 | rcu_read_unlock(); | |
325 | return result; | |
1da177e4 LT |
326 | } |
327 | ||
92476d7f | 328 | struct pid *find_get_pid(pid_t nr) |
1da177e4 LT |
329 | { |
330 | struct pid *pid; | |
331 | ||
92476d7f EB |
332 | rcu_read_lock(); |
333 | pid = get_pid(find_pid(nr)); | |
334 | rcu_read_unlock(); | |
1da177e4 | 335 | |
92476d7f | 336 | return pid; |
1da177e4 LT |
337 | } |
338 | ||
0804ef4b EB |
339 | /* |
340 | * Used by proc to find the first pid that is greater then or equal to nr. | |
341 | * | |
342 | * If there is a pid at nr this function is exactly the same as find_pid. | |
343 | */ | |
344 | struct pid *find_ge_pid(int nr) | |
345 | { | |
346 | struct pid *pid; | |
347 | ||
348 | do { | |
349 | pid = find_pid(nr); | |
350 | if (pid) | |
351 | break; | |
61a58c6c | 352 | nr = next_pidmap(&init_pid_ns, nr); |
0804ef4b EB |
353 | } while (nr > 0); |
354 | ||
355 | return pid; | |
356 | } | |
bbf73147 | 357 | EXPORT_SYMBOL_GPL(find_get_pid); |
0804ef4b | 358 | |
1da177e4 LT |
359 | /* |
360 | * The pid hash table is scaled according to the amount of memory in the | |
361 | * machine. From a minimum of 16 slots up to 4096 slots at one gigabyte or | |
362 | * more. | |
363 | */ | |
364 | void __init pidhash_init(void) | |
365 | { | |
92476d7f | 366 | int i, pidhash_size; |
1da177e4 LT |
367 | unsigned long megabytes = nr_kernel_pages >> (20 - PAGE_SHIFT); |
368 | ||
369 | pidhash_shift = max(4, fls(megabytes * 4)); | |
370 | pidhash_shift = min(12, pidhash_shift); | |
371 | pidhash_size = 1 << pidhash_shift; | |
372 | ||
373 | printk("PID hash table entries: %d (order: %d, %Zd bytes)\n", | |
374 | pidhash_size, pidhash_shift, | |
92476d7f EB |
375 | pidhash_size * sizeof(struct hlist_head)); |
376 | ||
377 | pid_hash = alloc_bootmem(pidhash_size * sizeof(*(pid_hash))); | |
378 | if (!pid_hash) | |
379 | panic("Could not alloc pidhash!\n"); | |
380 | for (i = 0; i < pidhash_size; i++) | |
381 | INIT_HLIST_HEAD(&pid_hash[i]); | |
1da177e4 LT |
382 | } |
383 | ||
384 | void __init pidmap_init(void) | |
385 | { | |
61a58c6c | 386 | init_pid_ns.pidmap[0].page = kzalloc(PAGE_SIZE, GFP_KERNEL); |
73b9ebfe | 387 | /* Reserve PID 0. We never call free_pidmap(0) */ |
61a58c6c SB |
388 | set_bit(0, init_pid_ns.pidmap[0].page); |
389 | atomic_dec(&init_pid_ns.pidmap[0].nr_free); | |
92476d7f EB |
390 | |
391 | pid_cachep = kmem_cache_create("pid", sizeof(struct pid), | |
392 | __alignof__(struct pid), | |
393 | SLAB_PANIC, NULL, NULL); | |
1da177e4 | 394 | } |