]>
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> |
820e45db | 30 | #include <linux/init_task.h> |
1da177e4 | 31 | |
8ef047aa PE |
32 | #define pid_hashfn(nr, ns) \ |
33 | hash_long((unsigned long)nr + (unsigned long)ns, pidhash_shift) | |
92476d7f | 34 | static struct hlist_head *pid_hash; |
1da177e4 | 35 | static int pidhash_shift; |
820e45db | 36 | struct pid init_struct_pid = INIT_STRUCT_PID; |
1da177e4 LT |
37 | |
38 | int pid_max = PID_MAX_DEFAULT; | |
1da177e4 LT |
39 | |
40 | #define RESERVED_PIDS 300 | |
41 | ||
42 | int pid_max_min = RESERVED_PIDS + 1; | |
43 | int pid_max_max = PID_MAX_LIMIT; | |
44 | ||
1da177e4 LT |
45 | #define BITS_PER_PAGE (PAGE_SIZE*8) |
46 | #define BITS_PER_PAGE_MASK (BITS_PER_PAGE-1) | |
3fbc9648 | 47 | |
61a58c6c SB |
48 | static inline int mk_pid(struct pid_namespace *pid_ns, |
49 | struct pidmap *map, int off) | |
3fbc9648 | 50 | { |
61a58c6c | 51 | return (map - pid_ns->pidmap)*BITS_PER_PAGE + off; |
3fbc9648 SB |
52 | } |
53 | ||
1da177e4 LT |
54 | #define find_next_offset(map, off) \ |
55 | find_next_zero_bit((map)->page, BITS_PER_PAGE, off) | |
56 | ||
57 | /* | |
58 | * PID-map pages start out as NULL, they get allocated upon | |
59 | * first use and are never deallocated. This way a low pid_max | |
60 | * value does not cause lots of bitmaps to be allocated, but | |
61 | * the scheme scales to up to 4 million PIDs, runtime. | |
62 | */ | |
61a58c6c | 63 | struct pid_namespace init_pid_ns = { |
9a575a92 CLG |
64 | .kref = { |
65 | .refcount = ATOMIC_INIT(2), | |
66 | }, | |
3fbc9648 SB |
67 | .pidmap = { |
68 | [ 0 ... PIDMAP_ENTRIES-1] = { ATOMIC_INIT(BITS_PER_PAGE), NULL } | |
69 | }, | |
84d73786 | 70 | .last_pid = 0, |
faacbfd3 PE |
71 | .level = 0, |
72 | .child_reaper = &init_task, | |
3fbc9648 | 73 | }; |
1da177e4 | 74 | |
b460cbc5 SH |
75 | int is_global_init(struct task_struct *tsk) |
76 | { | |
77 | return tsk == init_pid_ns.child_reaper; | |
78 | } | |
79 | ||
92476d7f EB |
80 | /* |
81 | * Note: disable interrupts while the pidmap_lock is held as an | |
82 | * interrupt might come in and do read_lock(&tasklist_lock). | |
83 | * | |
84 | * If we don't disable interrupts there is a nasty deadlock between | |
85 | * detach_pid()->free_pid() and another cpu that does | |
86 | * spin_lock(&pidmap_lock) followed by an interrupt routine that does | |
87 | * read_lock(&tasklist_lock); | |
88 | * | |
89 | * After we clean up the tasklist_lock and know there are no | |
90 | * irq handlers that take it we can leave the interrupts enabled. | |
91 | * For now it is easier to be safe than to prove it can't happen. | |
92 | */ | |
3fbc9648 | 93 | |
1da177e4 LT |
94 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(pidmap_lock); |
95 | ||
61a58c6c | 96 | static fastcall void free_pidmap(struct pid_namespace *pid_ns, int pid) |
1da177e4 | 97 | { |
61a58c6c | 98 | struct pidmap *map = pid_ns->pidmap + pid / BITS_PER_PAGE; |
1da177e4 LT |
99 | int offset = pid & BITS_PER_PAGE_MASK; |
100 | ||
101 | clear_bit(offset, map->page); | |
102 | atomic_inc(&map->nr_free); | |
103 | } | |
104 | ||
61a58c6c | 105 | static int alloc_pidmap(struct pid_namespace *pid_ns) |
1da177e4 | 106 | { |
61a58c6c | 107 | int i, offset, max_scan, pid, last = pid_ns->last_pid; |
6a1f3b84 | 108 | struct pidmap *map; |
1da177e4 LT |
109 | |
110 | pid = last + 1; | |
111 | if (pid >= pid_max) | |
112 | pid = RESERVED_PIDS; | |
113 | offset = pid & BITS_PER_PAGE_MASK; | |
61a58c6c | 114 | map = &pid_ns->pidmap[pid/BITS_PER_PAGE]; |
1da177e4 LT |
115 | max_scan = (pid_max + BITS_PER_PAGE - 1)/BITS_PER_PAGE - !offset; |
116 | for (i = 0; i <= max_scan; ++i) { | |
117 | if (unlikely(!map->page)) { | |
3fbc9648 | 118 | void *page = kzalloc(PAGE_SIZE, GFP_KERNEL); |
1da177e4 LT |
119 | /* |
120 | * Free the page if someone raced with us | |
121 | * installing it: | |
122 | */ | |
92476d7f | 123 | spin_lock_irq(&pidmap_lock); |
1da177e4 | 124 | if (map->page) |
3fbc9648 | 125 | kfree(page); |
1da177e4 | 126 | else |
3fbc9648 | 127 | map->page = page; |
92476d7f | 128 | spin_unlock_irq(&pidmap_lock); |
1da177e4 LT |
129 | if (unlikely(!map->page)) |
130 | break; | |
131 | } | |
132 | if (likely(atomic_read(&map->nr_free))) { | |
133 | do { | |
134 | if (!test_and_set_bit(offset, map->page)) { | |
135 | atomic_dec(&map->nr_free); | |
61a58c6c | 136 | pid_ns->last_pid = pid; |
1da177e4 LT |
137 | return pid; |
138 | } | |
139 | offset = find_next_offset(map, offset); | |
61a58c6c | 140 | pid = mk_pid(pid_ns, map, offset); |
1da177e4 LT |
141 | /* |
142 | * find_next_offset() found a bit, the pid from it | |
143 | * is in-bounds, and if we fell back to the last | |
144 | * bitmap block and the final block was the same | |
145 | * as the starting point, pid is before last_pid. | |
146 | */ | |
147 | } while (offset < BITS_PER_PAGE && pid < pid_max && | |
148 | (i != max_scan || pid < last || | |
149 | !((last+1) & BITS_PER_PAGE_MASK))); | |
150 | } | |
61a58c6c | 151 | if (map < &pid_ns->pidmap[(pid_max-1)/BITS_PER_PAGE]) { |
1da177e4 LT |
152 | ++map; |
153 | offset = 0; | |
154 | } else { | |
61a58c6c | 155 | map = &pid_ns->pidmap[0]; |
1da177e4 LT |
156 | offset = RESERVED_PIDS; |
157 | if (unlikely(last == offset)) | |
158 | break; | |
159 | } | |
61a58c6c | 160 | pid = mk_pid(pid_ns, map, offset); |
1da177e4 LT |
161 | } |
162 | return -1; | |
163 | } | |
164 | ||
61a58c6c | 165 | static int next_pidmap(struct pid_namespace *pid_ns, int last) |
0804ef4b EB |
166 | { |
167 | int offset; | |
f40f50d3 | 168 | struct pidmap *map, *end; |
0804ef4b EB |
169 | |
170 | offset = (last + 1) & BITS_PER_PAGE_MASK; | |
61a58c6c SB |
171 | map = &pid_ns->pidmap[(last + 1)/BITS_PER_PAGE]; |
172 | end = &pid_ns->pidmap[PIDMAP_ENTRIES]; | |
f40f50d3 | 173 | for (; map < end; map++, offset = 0) { |
0804ef4b EB |
174 | if (unlikely(!map->page)) |
175 | continue; | |
176 | offset = find_next_bit((map)->page, BITS_PER_PAGE, offset); | |
177 | if (offset < BITS_PER_PAGE) | |
61a58c6c | 178 | return mk_pid(pid_ns, map, offset); |
0804ef4b EB |
179 | } |
180 | return -1; | |
181 | } | |
182 | ||
92476d7f EB |
183 | fastcall void put_pid(struct pid *pid) |
184 | { | |
baf8f0f8 PE |
185 | struct pid_namespace *ns; |
186 | ||
92476d7f EB |
187 | if (!pid) |
188 | return; | |
baf8f0f8 | 189 | |
8ef047aa | 190 | ns = pid->numbers[pid->level].ns; |
92476d7f | 191 | if ((atomic_read(&pid->count) == 1) || |
8ef047aa | 192 | atomic_dec_and_test(&pid->count)) { |
baf8f0f8 | 193 | kmem_cache_free(ns->pid_cachep, pid); |
8ef047aa PE |
194 | if (ns != &init_pid_ns) |
195 | put_pid_ns(ns); | |
196 | } | |
92476d7f | 197 | } |
bbf73147 | 198 | EXPORT_SYMBOL_GPL(put_pid); |
92476d7f EB |
199 | |
200 | static void delayed_put_pid(struct rcu_head *rhp) | |
201 | { | |
202 | struct pid *pid = container_of(rhp, struct pid, rcu); | |
203 | put_pid(pid); | |
204 | } | |
205 | ||
206 | fastcall void free_pid(struct pid *pid) | |
207 | { | |
208 | /* We can be called with write_lock_irq(&tasklist_lock) held */ | |
8ef047aa | 209 | int i; |
92476d7f EB |
210 | unsigned long flags; |
211 | ||
212 | spin_lock_irqsave(&pidmap_lock, flags); | |
213 | hlist_del_rcu(&pid->pid_chain); | |
214 | spin_unlock_irqrestore(&pidmap_lock, flags); | |
215 | ||
8ef047aa PE |
216 | for (i = 0; i <= pid->level; i++) |
217 | free_pidmap(pid->numbers[i].ns, pid->numbers[i].nr); | |
218 | ||
92476d7f EB |
219 | call_rcu(&pid->rcu, delayed_put_pid); |
220 | } | |
221 | ||
8ef047aa | 222 | struct pid *alloc_pid(struct pid_namespace *ns) |
92476d7f EB |
223 | { |
224 | struct pid *pid; | |
225 | enum pid_type type; | |
8ef047aa PE |
226 | int i, nr; |
227 | struct pid_namespace *tmp; | |
92476d7f | 228 | |
baf8f0f8 | 229 | pid = kmem_cache_alloc(ns->pid_cachep, GFP_KERNEL); |
92476d7f EB |
230 | if (!pid) |
231 | goto out; | |
232 | ||
8ef047aa PE |
233 | tmp = ns; |
234 | for (i = ns->level; i >= 0; i--) { | |
235 | nr = alloc_pidmap(tmp); | |
236 | if (nr < 0) | |
237 | goto out_free; | |
92476d7f | 238 | |
8ef047aa PE |
239 | pid->numbers[i].nr = nr; |
240 | pid->numbers[i].ns = tmp; | |
241 | tmp = tmp->parent; | |
242 | } | |
243 | ||
244 | if (ns != &init_pid_ns) | |
245 | get_pid_ns(ns); | |
246 | ||
247 | pid->level = ns->level; | |
248 | pid->nr = pid->numbers[0].nr; | |
92476d7f | 249 | atomic_set(&pid->count, 1); |
92476d7f EB |
250 | for (type = 0; type < PIDTYPE_MAX; ++type) |
251 | INIT_HLIST_HEAD(&pid->tasks[type]); | |
252 | ||
253 | spin_lock_irq(&pidmap_lock); | |
8ef047aa | 254 | hlist_add_head_rcu(&pid->pid_chain, &pid_hash[pid_hashfn(pid->nr, ns)]); |
92476d7f EB |
255 | spin_unlock_irq(&pidmap_lock); |
256 | ||
257 | out: | |
258 | return pid; | |
259 | ||
260 | out_free: | |
8ef047aa PE |
261 | for (i++; i <= ns->level; i++) |
262 | free_pidmap(pid->numbers[i].ns, pid->numbers[i].nr); | |
263 | ||
baf8f0f8 | 264 | kmem_cache_free(ns->pid_cachep, pid); |
92476d7f EB |
265 | pid = NULL; |
266 | goto out; | |
267 | } | |
268 | ||
269 | struct pid * fastcall find_pid(int nr) | |
1da177e4 LT |
270 | { |
271 | struct hlist_node *elem; | |
272 | struct pid *pid; | |
273 | ||
e56d0903 | 274 | hlist_for_each_entry_rcu(pid, elem, |
8ef047aa | 275 | &pid_hash[pid_hashfn(nr, &init_pid_ns)], pid_chain) { |
1da177e4 LT |
276 | if (pid->nr == nr) |
277 | return pid; | |
278 | } | |
279 | return NULL; | |
280 | } | |
bbf73147 | 281 | EXPORT_SYMBOL_GPL(find_pid); |
1da177e4 | 282 | |
e713d0da SB |
283 | /* |
284 | * attach_pid() must be called with the tasklist_lock write-held. | |
285 | */ | |
286 | int fastcall attach_pid(struct task_struct *task, enum pid_type type, | |
287 | struct pid *pid) | |
1da177e4 | 288 | { |
92476d7f | 289 | struct pid_link *link; |
92476d7f | 290 | |
92476d7f | 291 | link = &task->pids[type]; |
e713d0da | 292 | link->pid = pid; |
92476d7f | 293 | hlist_add_head_rcu(&link->node, &pid->tasks[type]); |
1da177e4 LT |
294 | |
295 | return 0; | |
296 | } | |
297 | ||
36c8b586 | 298 | void fastcall detach_pid(struct task_struct *task, enum pid_type type) |
1da177e4 | 299 | { |
92476d7f EB |
300 | struct pid_link *link; |
301 | struct pid *pid; | |
302 | int tmp; | |
1da177e4 | 303 | |
92476d7f EB |
304 | link = &task->pids[type]; |
305 | pid = link->pid; | |
1da177e4 | 306 | |
92476d7f EB |
307 | hlist_del_rcu(&link->node); |
308 | link->pid = NULL; | |
1da177e4 | 309 | |
92476d7f EB |
310 | for (tmp = PIDTYPE_MAX; --tmp >= 0; ) |
311 | if (!hlist_empty(&pid->tasks[tmp])) | |
312 | return; | |
1da177e4 | 313 | |
92476d7f | 314 | free_pid(pid); |
1da177e4 LT |
315 | } |
316 | ||
c18258c6 EB |
317 | /* transfer_pid is an optimization of attach_pid(new), detach_pid(old) */ |
318 | void fastcall transfer_pid(struct task_struct *old, struct task_struct *new, | |
319 | enum pid_type type) | |
320 | { | |
321 | new->pids[type].pid = old->pids[type].pid; | |
322 | hlist_replace_rcu(&old->pids[type].node, &new->pids[type].node); | |
323 | old->pids[type].pid = NULL; | |
324 | } | |
325 | ||
92476d7f | 326 | struct task_struct * fastcall pid_task(struct pid *pid, enum pid_type type) |
1da177e4 | 327 | { |
92476d7f EB |
328 | struct task_struct *result = NULL; |
329 | if (pid) { | |
330 | struct hlist_node *first; | |
331 | first = rcu_dereference(pid->tasks[type].first); | |
332 | if (first) | |
333 | result = hlist_entry(first, struct task_struct, pids[(type)].node); | |
334 | } | |
335 | return result; | |
336 | } | |
1da177e4 | 337 | |
92476d7f EB |
338 | /* |
339 | * Must be called under rcu_read_lock() or with tasklist_lock read-held. | |
340 | */ | |
36c8b586 | 341 | struct task_struct *find_task_by_pid_type(int type, int nr) |
92476d7f EB |
342 | { |
343 | return pid_task(find_pid(nr), type); | |
344 | } | |
1da177e4 | 345 | |
92476d7f | 346 | EXPORT_SYMBOL(find_task_by_pid_type); |
1da177e4 | 347 | |
1a657f78 ON |
348 | struct pid *get_task_pid(struct task_struct *task, enum pid_type type) |
349 | { | |
350 | struct pid *pid; | |
351 | rcu_read_lock(); | |
352 | pid = get_pid(task->pids[type].pid); | |
353 | rcu_read_unlock(); | |
354 | return pid; | |
355 | } | |
356 | ||
92476d7f EB |
357 | struct task_struct *fastcall get_pid_task(struct pid *pid, enum pid_type type) |
358 | { | |
359 | struct task_struct *result; | |
360 | rcu_read_lock(); | |
361 | result = pid_task(pid, type); | |
362 | if (result) | |
363 | get_task_struct(result); | |
364 | rcu_read_unlock(); | |
365 | return result; | |
1da177e4 LT |
366 | } |
367 | ||
92476d7f | 368 | struct pid *find_get_pid(pid_t nr) |
1da177e4 LT |
369 | { |
370 | struct pid *pid; | |
371 | ||
92476d7f EB |
372 | rcu_read_lock(); |
373 | pid = get_pid(find_pid(nr)); | |
374 | rcu_read_unlock(); | |
1da177e4 | 375 | |
92476d7f | 376 | return pid; |
1da177e4 LT |
377 | } |
378 | ||
7af57294 PE |
379 | pid_t pid_nr_ns(struct pid *pid, struct pid_namespace *ns) |
380 | { | |
381 | struct upid *upid; | |
382 | pid_t nr = 0; | |
383 | ||
384 | if (pid && ns->level <= pid->level) { | |
385 | upid = &pid->numbers[ns->level]; | |
386 | if (upid->ns == ns) | |
387 | nr = upid->nr; | |
388 | } | |
389 | return nr; | |
390 | } | |
391 | ||
0804ef4b EB |
392 | /* |
393 | * Used by proc to find the first pid that is greater then or equal to nr. | |
394 | * | |
395 | * If there is a pid at nr this function is exactly the same as find_pid. | |
396 | */ | |
397 | struct pid *find_ge_pid(int nr) | |
398 | { | |
399 | struct pid *pid; | |
400 | ||
401 | do { | |
402 | pid = find_pid(nr); | |
403 | if (pid) | |
404 | break; | |
2894d650 | 405 | nr = next_pidmap(task_active_pid_ns(current), nr); |
0804ef4b EB |
406 | } while (nr > 0); |
407 | ||
408 | return pid; | |
409 | } | |
bbf73147 | 410 | EXPORT_SYMBOL_GPL(find_get_pid); |
0804ef4b | 411 | |
baf8f0f8 PE |
412 | struct pid_cache { |
413 | int nr_ids; | |
414 | char name[16]; | |
415 | struct kmem_cache *cachep; | |
416 | struct list_head list; | |
417 | }; | |
418 | ||
419 | static LIST_HEAD(pid_caches_lh); | |
420 | static DEFINE_MUTEX(pid_caches_mutex); | |
421 | ||
422 | /* | |
423 | * creates the kmem cache to allocate pids from. | |
424 | * @nr_ids: the number of numerical ids this pid will have to carry | |
425 | */ | |
426 | ||
427 | static struct kmem_cache *create_pid_cachep(int nr_ids) | |
428 | { | |
429 | struct pid_cache *pcache; | |
430 | struct kmem_cache *cachep; | |
431 | ||
432 | mutex_lock(&pid_caches_mutex); | |
433 | list_for_each_entry (pcache, &pid_caches_lh, list) | |
434 | if (pcache->nr_ids == nr_ids) | |
435 | goto out; | |
436 | ||
437 | pcache = kmalloc(sizeof(struct pid_cache), GFP_KERNEL); | |
438 | if (pcache == NULL) | |
439 | goto err_alloc; | |
440 | ||
441 | snprintf(pcache->name, sizeof(pcache->name), "pid_%d", nr_ids); | |
442 | cachep = kmem_cache_create(pcache->name, | |
443 | /* FIXME add numerical ids here */ | |
444 | sizeof(struct pid), 0, SLAB_HWCACHE_ALIGN, NULL); | |
445 | if (cachep == NULL) | |
446 | goto err_cachep; | |
447 | ||
448 | pcache->nr_ids = nr_ids; | |
449 | pcache->cachep = cachep; | |
450 | list_add(&pcache->list, &pid_caches_lh); | |
451 | out: | |
452 | mutex_unlock(&pid_caches_mutex); | |
453 | return pcache->cachep; | |
454 | ||
455 | err_cachep: | |
456 | kfree(pcache); | |
457 | err_alloc: | |
458 | mutex_unlock(&pid_caches_mutex); | |
459 | return NULL; | |
460 | } | |
461 | ||
213dd266 | 462 | struct pid_namespace *copy_pid_ns(unsigned long flags, struct pid_namespace *old_ns) |
9a575a92 | 463 | { |
e3222c4e | 464 | BUG_ON(!old_ns); |
9a575a92 | 465 | get_pid_ns(old_ns); |
e3222c4e | 466 | return old_ns; |
9a575a92 CLG |
467 | } |
468 | ||
469 | void free_pid_ns(struct kref *kref) | |
470 | { | |
471 | struct pid_namespace *ns; | |
472 | ||
473 | ns = container_of(kref, struct pid_namespace, kref); | |
474 | kfree(ns); | |
475 | } | |
476 | ||
1da177e4 LT |
477 | /* |
478 | * The pid hash table is scaled according to the amount of memory in the | |
479 | * machine. From a minimum of 16 slots up to 4096 slots at one gigabyte or | |
480 | * more. | |
481 | */ | |
482 | void __init pidhash_init(void) | |
483 | { | |
92476d7f | 484 | int i, pidhash_size; |
1da177e4 LT |
485 | unsigned long megabytes = nr_kernel_pages >> (20 - PAGE_SHIFT); |
486 | ||
487 | pidhash_shift = max(4, fls(megabytes * 4)); | |
488 | pidhash_shift = min(12, pidhash_shift); | |
489 | pidhash_size = 1 << pidhash_shift; | |
490 | ||
491 | printk("PID hash table entries: %d (order: %d, %Zd bytes)\n", | |
492 | pidhash_size, pidhash_shift, | |
92476d7f EB |
493 | pidhash_size * sizeof(struct hlist_head)); |
494 | ||
495 | pid_hash = alloc_bootmem(pidhash_size * sizeof(*(pid_hash))); | |
496 | if (!pid_hash) | |
497 | panic("Could not alloc pidhash!\n"); | |
498 | for (i = 0; i < pidhash_size; i++) | |
499 | INIT_HLIST_HEAD(&pid_hash[i]); | |
1da177e4 LT |
500 | } |
501 | ||
502 | void __init pidmap_init(void) | |
503 | { | |
61a58c6c | 504 | init_pid_ns.pidmap[0].page = kzalloc(PAGE_SIZE, GFP_KERNEL); |
73b9ebfe | 505 | /* Reserve PID 0. We never call free_pidmap(0) */ |
61a58c6c SB |
506 | set_bit(0, init_pid_ns.pidmap[0].page); |
507 | atomic_dec(&init_pid_ns.pidmap[0].nr_free); | |
92476d7f | 508 | |
baf8f0f8 PE |
509 | init_pid_ns.pid_cachep = create_pid_cachep(1); |
510 | if (init_pid_ns.pid_cachep == NULL) | |
511 | panic("Can't create pid_1 cachep\n"); | |
1da177e4 | 512 | } |