]>
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). | |
30e49c26 PE |
21 | * |
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 | * | |
1da177e4 LT |
27 | */ |
28 | ||
29 | #include <linux/mm.h> | |
30 | #include <linux/module.h> | |
31 | #include <linux/slab.h> | |
32 | #include <linux/init.h> | |
33 | #include <linux/bootmem.h> | |
34 | #include <linux/hash.h> | |
61a58c6c | 35 | #include <linux/pid_namespace.h> |
820e45db | 36 | #include <linux/init_task.h> |
3eb07c8c | 37 | #include <linux/syscalls.h> |
1da177e4 | 38 | |
8ef047aa PE |
39 | #define pid_hashfn(nr, ns) \ |
40 | hash_long((unsigned long)nr + (unsigned long)ns, pidhash_shift) | |
92476d7f | 41 | static struct hlist_head *pid_hash; |
1da177e4 | 42 | static int pidhash_shift; |
820e45db | 43 | struct pid init_struct_pid = INIT_STRUCT_PID; |
c9c5d922 | 44 | static struct kmem_cache *pid_ns_cachep; |
1da177e4 LT |
45 | |
46 | int pid_max = PID_MAX_DEFAULT; | |
1da177e4 LT |
47 | |
48 | #define RESERVED_PIDS 300 | |
49 | ||
50 | int pid_max_min = RESERVED_PIDS + 1; | |
51 | int pid_max_max = PID_MAX_LIMIT; | |
52 | ||
1da177e4 LT |
53 | #define BITS_PER_PAGE (PAGE_SIZE*8) |
54 | #define BITS_PER_PAGE_MASK (BITS_PER_PAGE-1) | |
3fbc9648 | 55 | |
61a58c6c SB |
56 | static inline int mk_pid(struct pid_namespace *pid_ns, |
57 | struct pidmap *map, int off) | |
3fbc9648 | 58 | { |
61a58c6c | 59 | return (map - pid_ns->pidmap)*BITS_PER_PAGE + off; |
3fbc9648 SB |
60 | } |
61 | ||
1da177e4 LT |
62 | #define find_next_offset(map, off) \ |
63 | find_next_zero_bit((map)->page, BITS_PER_PAGE, off) | |
64 | ||
65 | /* | |
66 | * PID-map pages start out as NULL, they get allocated upon | |
67 | * first use and are never deallocated. This way a low pid_max | |
68 | * value does not cause lots of bitmaps to be allocated, but | |
69 | * the scheme scales to up to 4 million PIDs, runtime. | |
70 | */ | |
61a58c6c | 71 | struct pid_namespace init_pid_ns = { |
9a575a92 CLG |
72 | .kref = { |
73 | .refcount = ATOMIC_INIT(2), | |
74 | }, | |
3fbc9648 SB |
75 | .pidmap = { |
76 | [ 0 ... PIDMAP_ENTRIES-1] = { ATOMIC_INIT(BITS_PER_PAGE), NULL } | |
77 | }, | |
84d73786 | 78 | .last_pid = 0, |
faacbfd3 PE |
79 | .level = 0, |
80 | .child_reaper = &init_task, | |
3fbc9648 | 81 | }; |
198fe21b | 82 | EXPORT_SYMBOL_GPL(init_pid_ns); |
1da177e4 | 83 | |
b461cc03 | 84 | int is_container_init(struct task_struct *tsk) |
b460cbc5 | 85 | { |
b461cc03 PE |
86 | int ret = 0; |
87 | struct pid *pid; | |
88 | ||
89 | rcu_read_lock(); | |
90 | pid = task_pid(tsk); | |
91 | if (pid != NULL && pid->numbers[pid->level].nr == 1) | |
92 | ret = 1; | |
93 | rcu_read_unlock(); | |
94 | ||
95 | return ret; | |
b460cbc5 | 96 | } |
b461cc03 | 97 | EXPORT_SYMBOL(is_container_init); |
b460cbc5 | 98 | |
92476d7f EB |
99 | /* |
100 | * Note: disable interrupts while the pidmap_lock is held as an | |
101 | * interrupt might come in and do read_lock(&tasklist_lock). | |
102 | * | |
103 | * If we don't disable interrupts there is a nasty deadlock between | |
104 | * detach_pid()->free_pid() and another cpu that does | |
105 | * spin_lock(&pidmap_lock) followed by an interrupt routine that does | |
106 | * read_lock(&tasklist_lock); | |
107 | * | |
108 | * After we clean up the tasklist_lock and know there are no | |
109 | * irq handlers that take it we can leave the interrupts enabled. | |
110 | * For now it is easier to be safe than to prove it can't happen. | |
111 | */ | |
3fbc9648 | 112 | |
1da177e4 LT |
113 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(pidmap_lock); |
114 | ||
61a58c6c | 115 | static fastcall void free_pidmap(struct pid_namespace *pid_ns, int pid) |
1da177e4 | 116 | { |
61a58c6c | 117 | struct pidmap *map = pid_ns->pidmap + pid / BITS_PER_PAGE; |
1da177e4 LT |
118 | int offset = pid & BITS_PER_PAGE_MASK; |
119 | ||
120 | clear_bit(offset, map->page); | |
121 | atomic_inc(&map->nr_free); | |
122 | } | |
123 | ||
61a58c6c | 124 | static int alloc_pidmap(struct pid_namespace *pid_ns) |
1da177e4 | 125 | { |
61a58c6c | 126 | int i, offset, max_scan, pid, last = pid_ns->last_pid; |
6a1f3b84 | 127 | struct pidmap *map; |
1da177e4 LT |
128 | |
129 | pid = last + 1; | |
130 | if (pid >= pid_max) | |
131 | pid = RESERVED_PIDS; | |
132 | offset = pid & BITS_PER_PAGE_MASK; | |
61a58c6c | 133 | map = &pid_ns->pidmap[pid/BITS_PER_PAGE]; |
1da177e4 LT |
134 | max_scan = (pid_max + BITS_PER_PAGE - 1)/BITS_PER_PAGE - !offset; |
135 | for (i = 0; i <= max_scan; ++i) { | |
136 | if (unlikely(!map->page)) { | |
3fbc9648 | 137 | void *page = kzalloc(PAGE_SIZE, GFP_KERNEL); |
1da177e4 LT |
138 | /* |
139 | * Free the page if someone raced with us | |
140 | * installing it: | |
141 | */ | |
92476d7f | 142 | spin_lock_irq(&pidmap_lock); |
1da177e4 | 143 | if (map->page) |
3fbc9648 | 144 | kfree(page); |
1da177e4 | 145 | else |
3fbc9648 | 146 | map->page = page; |
92476d7f | 147 | spin_unlock_irq(&pidmap_lock); |
1da177e4 LT |
148 | if (unlikely(!map->page)) |
149 | break; | |
150 | } | |
151 | if (likely(atomic_read(&map->nr_free))) { | |
152 | do { | |
153 | if (!test_and_set_bit(offset, map->page)) { | |
154 | atomic_dec(&map->nr_free); | |
61a58c6c | 155 | pid_ns->last_pid = pid; |
1da177e4 LT |
156 | return pid; |
157 | } | |
158 | offset = find_next_offset(map, offset); | |
61a58c6c | 159 | pid = mk_pid(pid_ns, map, offset); |
1da177e4 LT |
160 | /* |
161 | * find_next_offset() found a bit, the pid from it | |
162 | * is in-bounds, and if we fell back to the last | |
163 | * bitmap block and the final block was the same | |
164 | * as the starting point, pid is before last_pid. | |
165 | */ | |
166 | } while (offset < BITS_PER_PAGE && pid < pid_max && | |
167 | (i != max_scan || pid < last || | |
168 | !((last+1) & BITS_PER_PAGE_MASK))); | |
169 | } | |
61a58c6c | 170 | if (map < &pid_ns->pidmap[(pid_max-1)/BITS_PER_PAGE]) { |
1da177e4 LT |
171 | ++map; |
172 | offset = 0; | |
173 | } else { | |
61a58c6c | 174 | map = &pid_ns->pidmap[0]; |
1da177e4 LT |
175 | offset = RESERVED_PIDS; |
176 | if (unlikely(last == offset)) | |
177 | break; | |
178 | } | |
61a58c6c | 179 | pid = mk_pid(pid_ns, map, offset); |
1da177e4 LT |
180 | } |
181 | return -1; | |
182 | } | |
183 | ||
61a58c6c | 184 | static int next_pidmap(struct pid_namespace *pid_ns, int last) |
0804ef4b EB |
185 | { |
186 | int offset; | |
f40f50d3 | 187 | struct pidmap *map, *end; |
0804ef4b EB |
188 | |
189 | offset = (last + 1) & BITS_PER_PAGE_MASK; | |
61a58c6c SB |
190 | map = &pid_ns->pidmap[(last + 1)/BITS_PER_PAGE]; |
191 | end = &pid_ns->pidmap[PIDMAP_ENTRIES]; | |
f40f50d3 | 192 | for (; map < end; map++, offset = 0) { |
0804ef4b EB |
193 | if (unlikely(!map->page)) |
194 | continue; | |
195 | offset = find_next_bit((map)->page, BITS_PER_PAGE, offset); | |
196 | if (offset < BITS_PER_PAGE) | |
61a58c6c | 197 | return mk_pid(pid_ns, map, offset); |
0804ef4b EB |
198 | } |
199 | return -1; | |
200 | } | |
201 | ||
92476d7f EB |
202 | fastcall void put_pid(struct pid *pid) |
203 | { | |
baf8f0f8 PE |
204 | struct pid_namespace *ns; |
205 | ||
92476d7f EB |
206 | if (!pid) |
207 | return; | |
baf8f0f8 | 208 | |
8ef047aa | 209 | ns = pid->numbers[pid->level].ns; |
92476d7f | 210 | if ((atomic_read(&pid->count) == 1) || |
8ef047aa | 211 | atomic_dec_and_test(&pid->count)) { |
baf8f0f8 | 212 | kmem_cache_free(ns->pid_cachep, pid); |
b461cc03 | 213 | put_pid_ns(ns); |
8ef047aa | 214 | } |
92476d7f | 215 | } |
bbf73147 | 216 | EXPORT_SYMBOL_GPL(put_pid); |
92476d7f EB |
217 | |
218 | static void delayed_put_pid(struct rcu_head *rhp) | |
219 | { | |
220 | struct pid *pid = container_of(rhp, struct pid, rcu); | |
221 | put_pid(pid); | |
222 | } | |
223 | ||
224 | fastcall void free_pid(struct pid *pid) | |
225 | { | |
226 | /* We can be called with write_lock_irq(&tasklist_lock) held */ | |
8ef047aa | 227 | int i; |
92476d7f EB |
228 | unsigned long flags; |
229 | ||
230 | spin_lock_irqsave(&pidmap_lock, flags); | |
198fe21b PE |
231 | for (i = 0; i <= pid->level; i++) |
232 | hlist_del_rcu(&pid->numbers[i].pid_chain); | |
92476d7f EB |
233 | spin_unlock_irqrestore(&pidmap_lock, flags); |
234 | ||
8ef047aa PE |
235 | for (i = 0; i <= pid->level; i++) |
236 | free_pidmap(pid->numbers[i].ns, pid->numbers[i].nr); | |
237 | ||
92476d7f EB |
238 | call_rcu(&pid->rcu, delayed_put_pid); |
239 | } | |
240 | ||
8ef047aa | 241 | struct pid *alloc_pid(struct pid_namespace *ns) |
92476d7f EB |
242 | { |
243 | struct pid *pid; | |
244 | enum pid_type type; | |
8ef047aa PE |
245 | int i, nr; |
246 | struct pid_namespace *tmp; | |
198fe21b | 247 | struct upid *upid; |
92476d7f | 248 | |
baf8f0f8 | 249 | pid = kmem_cache_alloc(ns->pid_cachep, GFP_KERNEL); |
92476d7f EB |
250 | if (!pid) |
251 | goto out; | |
252 | ||
8ef047aa PE |
253 | tmp = ns; |
254 | for (i = ns->level; i >= 0; i--) { | |
255 | nr = alloc_pidmap(tmp); | |
256 | if (nr < 0) | |
257 | goto out_free; | |
92476d7f | 258 | |
8ef047aa PE |
259 | pid->numbers[i].nr = nr; |
260 | pid->numbers[i].ns = tmp; | |
261 | tmp = tmp->parent; | |
262 | } | |
263 | ||
b461cc03 | 264 | get_pid_ns(ns); |
8ef047aa | 265 | pid->level = ns->level; |
92476d7f | 266 | atomic_set(&pid->count, 1); |
92476d7f EB |
267 | for (type = 0; type < PIDTYPE_MAX; ++type) |
268 | INIT_HLIST_HEAD(&pid->tasks[type]); | |
269 | ||
270 | spin_lock_irq(&pidmap_lock); | |
198fe21b PE |
271 | for (i = ns->level; i >= 0; i--) { |
272 | upid = &pid->numbers[i]; | |
273 | hlist_add_head_rcu(&upid->pid_chain, | |
274 | &pid_hash[pid_hashfn(upid->nr, upid->ns)]); | |
275 | } | |
92476d7f EB |
276 | spin_unlock_irq(&pidmap_lock); |
277 | ||
278 | out: | |
279 | return pid; | |
280 | ||
281 | out_free: | |
8ef047aa PE |
282 | for (i++; i <= ns->level; i++) |
283 | free_pidmap(pid->numbers[i].ns, pid->numbers[i].nr); | |
284 | ||
baf8f0f8 | 285 | kmem_cache_free(ns->pid_cachep, pid); |
92476d7f EB |
286 | pid = NULL; |
287 | goto out; | |
288 | } | |
289 | ||
198fe21b | 290 | struct pid * fastcall find_pid_ns(int nr, struct pid_namespace *ns) |
1da177e4 LT |
291 | { |
292 | struct hlist_node *elem; | |
198fe21b PE |
293 | struct upid *pnr; |
294 | ||
295 | hlist_for_each_entry_rcu(pnr, elem, | |
296 | &pid_hash[pid_hashfn(nr, ns)], pid_chain) | |
297 | if (pnr->nr == nr && pnr->ns == ns) | |
298 | return container_of(pnr, struct pid, | |
299 | numbers[ns->level]); | |
1da177e4 | 300 | |
1da177e4 LT |
301 | return NULL; |
302 | } | |
198fe21b | 303 | EXPORT_SYMBOL_GPL(find_pid_ns); |
1da177e4 | 304 | |
e713d0da SB |
305 | /* |
306 | * attach_pid() must be called with the tasklist_lock write-held. | |
307 | */ | |
308 | int fastcall attach_pid(struct task_struct *task, enum pid_type type, | |
309 | struct pid *pid) | |
1da177e4 | 310 | { |
92476d7f | 311 | struct pid_link *link; |
92476d7f | 312 | |
92476d7f | 313 | link = &task->pids[type]; |
e713d0da | 314 | link->pid = pid; |
92476d7f | 315 | hlist_add_head_rcu(&link->node, &pid->tasks[type]); |
1da177e4 LT |
316 | |
317 | return 0; | |
318 | } | |
319 | ||
36c8b586 | 320 | void fastcall detach_pid(struct task_struct *task, enum pid_type type) |
1da177e4 | 321 | { |
92476d7f EB |
322 | struct pid_link *link; |
323 | struct pid *pid; | |
324 | int tmp; | |
1da177e4 | 325 | |
92476d7f EB |
326 | link = &task->pids[type]; |
327 | pid = link->pid; | |
1da177e4 | 328 | |
92476d7f EB |
329 | hlist_del_rcu(&link->node); |
330 | link->pid = NULL; | |
1da177e4 | 331 | |
92476d7f EB |
332 | for (tmp = PIDTYPE_MAX; --tmp >= 0; ) |
333 | if (!hlist_empty(&pid->tasks[tmp])) | |
334 | return; | |
1da177e4 | 335 | |
92476d7f | 336 | free_pid(pid); |
1da177e4 LT |
337 | } |
338 | ||
c18258c6 EB |
339 | /* transfer_pid is an optimization of attach_pid(new), detach_pid(old) */ |
340 | void fastcall transfer_pid(struct task_struct *old, struct task_struct *new, | |
341 | enum pid_type type) | |
342 | { | |
343 | new->pids[type].pid = old->pids[type].pid; | |
344 | hlist_replace_rcu(&old->pids[type].node, &new->pids[type].node); | |
345 | old->pids[type].pid = NULL; | |
346 | } | |
347 | ||
92476d7f | 348 | struct task_struct * fastcall pid_task(struct pid *pid, enum pid_type type) |
1da177e4 | 349 | { |
92476d7f EB |
350 | struct task_struct *result = NULL; |
351 | if (pid) { | |
352 | struct hlist_node *first; | |
353 | first = rcu_dereference(pid->tasks[type].first); | |
354 | if (first) | |
355 | result = hlist_entry(first, struct task_struct, pids[(type)].node); | |
356 | } | |
357 | return result; | |
358 | } | |
1da177e4 | 359 | |
92476d7f EB |
360 | /* |
361 | * Must be called under rcu_read_lock() or with tasklist_lock read-held. | |
362 | */ | |
198fe21b PE |
363 | struct task_struct *find_task_by_pid_type_ns(int type, int nr, |
364 | struct pid_namespace *ns) | |
92476d7f | 365 | { |
198fe21b | 366 | return pid_task(find_pid_ns(nr, ns), type); |
92476d7f | 367 | } |
1da177e4 | 368 | |
198fe21b | 369 | EXPORT_SYMBOL(find_task_by_pid_type_ns); |
1da177e4 | 370 | |
228ebcbe PE |
371 | struct task_struct *find_task_by_pid(pid_t nr) |
372 | { | |
373 | return find_task_by_pid_type_ns(PIDTYPE_PID, nr, &init_pid_ns); | |
374 | } | |
375 | EXPORT_SYMBOL(find_task_by_pid); | |
376 | ||
377 | struct task_struct *find_task_by_vpid(pid_t vnr) | |
378 | { | |
379 | return find_task_by_pid_type_ns(PIDTYPE_PID, vnr, | |
380 | current->nsproxy->pid_ns); | |
381 | } | |
382 | EXPORT_SYMBOL(find_task_by_vpid); | |
383 | ||
384 | struct task_struct *find_task_by_pid_ns(pid_t nr, struct pid_namespace *ns) | |
385 | { | |
386 | return find_task_by_pid_type_ns(PIDTYPE_PID, nr, ns); | |
387 | } | |
388 | EXPORT_SYMBOL(find_task_by_pid_ns); | |
389 | ||
1a657f78 ON |
390 | struct pid *get_task_pid(struct task_struct *task, enum pid_type type) |
391 | { | |
392 | struct pid *pid; | |
393 | rcu_read_lock(); | |
394 | pid = get_pid(task->pids[type].pid); | |
395 | rcu_read_unlock(); | |
396 | return pid; | |
397 | } | |
398 | ||
92476d7f EB |
399 | struct task_struct *fastcall get_pid_task(struct pid *pid, enum pid_type type) |
400 | { | |
401 | struct task_struct *result; | |
402 | rcu_read_lock(); | |
403 | result = pid_task(pid, type); | |
404 | if (result) | |
405 | get_task_struct(result); | |
406 | rcu_read_unlock(); | |
407 | return result; | |
1da177e4 LT |
408 | } |
409 | ||
92476d7f | 410 | struct pid *find_get_pid(pid_t nr) |
1da177e4 LT |
411 | { |
412 | struct pid *pid; | |
413 | ||
92476d7f | 414 | rcu_read_lock(); |
198fe21b | 415 | pid = get_pid(find_vpid(nr)); |
92476d7f | 416 | rcu_read_unlock(); |
1da177e4 | 417 | |
92476d7f | 418 | return pid; |
1da177e4 LT |
419 | } |
420 | ||
7af57294 PE |
421 | pid_t pid_nr_ns(struct pid *pid, struct pid_namespace *ns) |
422 | { | |
423 | struct upid *upid; | |
424 | pid_t nr = 0; | |
425 | ||
426 | if (pid && ns->level <= pid->level) { | |
427 | upid = &pid->numbers[ns->level]; | |
428 | if (upid->ns == ns) | |
429 | nr = upid->nr; | |
430 | } | |
431 | return nr; | |
432 | } | |
433 | ||
0804ef4b EB |
434 | /* |
435 | * Used by proc to find the first pid that is greater then or equal to nr. | |
436 | * | |
437 | * If there is a pid at nr this function is exactly the same as find_pid. | |
438 | */ | |
198fe21b | 439 | struct pid *find_ge_pid(int nr, struct pid_namespace *ns) |
0804ef4b EB |
440 | { |
441 | struct pid *pid; | |
442 | ||
443 | do { | |
198fe21b | 444 | pid = find_pid_ns(nr, ns); |
0804ef4b EB |
445 | if (pid) |
446 | break; | |
198fe21b | 447 | nr = next_pidmap(ns, nr); |
0804ef4b EB |
448 | } while (nr > 0); |
449 | ||
450 | return pid; | |
451 | } | |
bbf73147 | 452 | EXPORT_SYMBOL_GPL(find_get_pid); |
0804ef4b | 453 | |
baf8f0f8 PE |
454 | struct pid_cache { |
455 | int nr_ids; | |
456 | char name[16]; | |
457 | struct kmem_cache *cachep; | |
458 | struct list_head list; | |
459 | }; | |
460 | ||
461 | static LIST_HEAD(pid_caches_lh); | |
462 | static DEFINE_MUTEX(pid_caches_mutex); | |
463 | ||
464 | /* | |
465 | * creates the kmem cache to allocate pids from. | |
466 | * @nr_ids: the number of numerical ids this pid will have to carry | |
467 | */ | |
468 | ||
469 | static struct kmem_cache *create_pid_cachep(int nr_ids) | |
470 | { | |
471 | struct pid_cache *pcache; | |
472 | struct kmem_cache *cachep; | |
473 | ||
474 | mutex_lock(&pid_caches_mutex); | |
475 | list_for_each_entry (pcache, &pid_caches_lh, list) | |
476 | if (pcache->nr_ids == nr_ids) | |
477 | goto out; | |
478 | ||
479 | pcache = kmalloc(sizeof(struct pid_cache), GFP_KERNEL); | |
480 | if (pcache == NULL) | |
481 | goto err_alloc; | |
482 | ||
483 | snprintf(pcache->name, sizeof(pcache->name), "pid_%d", nr_ids); | |
484 | cachep = kmem_cache_create(pcache->name, | |
30e49c26 PE |
485 | sizeof(struct pid) + (nr_ids - 1) * sizeof(struct upid), |
486 | 0, SLAB_HWCACHE_ALIGN, NULL); | |
baf8f0f8 PE |
487 | if (cachep == NULL) |
488 | goto err_cachep; | |
489 | ||
490 | pcache->nr_ids = nr_ids; | |
491 | pcache->cachep = cachep; | |
492 | list_add(&pcache->list, &pid_caches_lh); | |
493 | out: | |
494 | mutex_unlock(&pid_caches_mutex); | |
495 | return pcache->cachep; | |
496 | ||
497 | err_cachep: | |
498 | kfree(pcache); | |
499 | err_alloc: | |
500 | mutex_unlock(&pid_caches_mutex); | |
501 | return NULL; | |
502 | } | |
503 | ||
30e49c26 PE |
504 | static struct pid_namespace *create_pid_namespace(int level) |
505 | { | |
506 | struct pid_namespace *ns; | |
507 | int i; | |
508 | ||
c9c5d922 | 509 | ns = kmem_cache_alloc(pid_ns_cachep, GFP_KERNEL); |
30e49c26 PE |
510 | if (ns == NULL) |
511 | goto out; | |
512 | ||
513 | ns->pidmap[0].page = kzalloc(PAGE_SIZE, GFP_KERNEL); | |
514 | if (!ns->pidmap[0].page) | |
515 | goto out_free; | |
516 | ||
517 | ns->pid_cachep = create_pid_cachep(level + 1); | |
518 | if (ns->pid_cachep == NULL) | |
519 | goto out_free_map; | |
520 | ||
521 | kref_init(&ns->kref); | |
522 | ns->last_pid = 0; | |
523 | ns->child_reaper = NULL; | |
524 | ns->level = level; | |
525 | ||
526 | set_bit(0, ns->pidmap[0].page); | |
527 | atomic_set(&ns->pidmap[0].nr_free, BITS_PER_PAGE - 1); | |
528 | ||
529 | for (i = 1; i < PIDMAP_ENTRIES; i++) { | |
530 | ns->pidmap[i].page = 0; | |
531 | atomic_set(&ns->pidmap[i].nr_free, BITS_PER_PAGE); | |
532 | } | |
533 | ||
534 | return ns; | |
535 | ||
536 | out_free_map: | |
537 | kfree(ns->pidmap[0].page); | |
538 | out_free: | |
c9c5d922 | 539 | kmem_cache_free(pid_ns_cachep, ns); |
30e49c26 PE |
540 | out: |
541 | return ERR_PTR(-ENOMEM); | |
542 | } | |
543 | ||
544 | static void destroy_pid_namespace(struct pid_namespace *ns) | |
545 | { | |
546 | int i; | |
547 | ||
548 | for (i = 0; i < PIDMAP_ENTRIES; i++) | |
549 | kfree(ns->pidmap[i].page); | |
c9c5d922 | 550 | kmem_cache_free(pid_ns_cachep, ns); |
30e49c26 PE |
551 | } |
552 | ||
213dd266 | 553 | struct pid_namespace *copy_pid_ns(unsigned long flags, struct pid_namespace *old_ns) |
9a575a92 | 554 | { |
30e49c26 PE |
555 | struct pid_namespace *new_ns; |
556 | ||
e3222c4e | 557 | BUG_ON(!old_ns); |
30e49c26 PE |
558 | new_ns = get_pid_ns(old_ns); |
559 | if (!(flags & CLONE_NEWPID)) | |
560 | goto out; | |
561 | ||
562 | new_ns = ERR_PTR(-EINVAL); | |
563 | if (flags & CLONE_THREAD) | |
564 | goto out_put; | |
565 | ||
566 | new_ns = create_pid_namespace(old_ns->level + 1); | |
567 | if (!IS_ERR(new_ns)) | |
568 | new_ns->parent = get_pid_ns(old_ns); | |
569 | ||
570 | out_put: | |
571 | put_pid_ns(old_ns); | |
572 | out: | |
573 | return new_ns; | |
9a575a92 CLG |
574 | } |
575 | ||
576 | void free_pid_ns(struct kref *kref) | |
577 | { | |
30e49c26 | 578 | struct pid_namespace *ns, *parent; |
9a575a92 CLG |
579 | |
580 | ns = container_of(kref, struct pid_namespace, kref); | |
30e49c26 PE |
581 | |
582 | parent = ns->parent; | |
583 | destroy_pid_namespace(ns); | |
584 | ||
585 | if (parent != NULL) | |
586 | put_pid_ns(parent); | |
9a575a92 CLG |
587 | } |
588 | ||
3eb07c8c SB |
589 | void zap_pid_ns_processes(struct pid_namespace *pid_ns) |
590 | { | |
591 | int nr; | |
592 | int rc; | |
593 | ||
594 | /* | |
595 | * The last thread in the cgroup-init thread group is terminating. | |
596 | * Find remaining pid_ts in the namespace, signal and wait for them | |
597 | * to exit. | |
598 | * | |
599 | * Note: This signals each threads in the namespace - even those that | |
600 | * belong to the same thread group, To avoid this, we would have | |
601 | * to walk the entire tasklist looking a processes in this | |
602 | * namespace, but that could be unnecessarily expensive if the | |
603 | * pid namespace has just a few processes. Or we need to | |
604 | * maintain a tasklist for each pid namespace. | |
605 | * | |
606 | */ | |
607 | read_lock(&tasklist_lock); | |
608 | nr = next_pidmap(pid_ns, 1); | |
609 | while (nr > 0) { | |
610 | kill_proc_info(SIGKILL, SEND_SIG_PRIV, nr); | |
611 | nr = next_pidmap(pid_ns, nr); | |
612 | } | |
613 | read_unlock(&tasklist_lock); | |
614 | ||
615 | do { | |
616 | clear_thread_flag(TIF_SIGPENDING); | |
617 | rc = sys_wait4(-1, NULL, __WALL, NULL); | |
618 | } while (rc != -ECHILD); | |
619 | ||
620 | ||
621 | /* Child reaper for the pid namespace is going away */ | |
622 | pid_ns->child_reaper = NULL; | |
623 | return; | |
624 | } | |
625 | ||
1da177e4 LT |
626 | /* |
627 | * The pid hash table is scaled according to the amount of memory in the | |
628 | * machine. From a minimum of 16 slots up to 4096 slots at one gigabyte or | |
629 | * more. | |
630 | */ | |
631 | void __init pidhash_init(void) | |
632 | { | |
92476d7f | 633 | int i, pidhash_size; |
1da177e4 LT |
634 | unsigned long megabytes = nr_kernel_pages >> (20 - PAGE_SHIFT); |
635 | ||
636 | pidhash_shift = max(4, fls(megabytes * 4)); | |
637 | pidhash_shift = min(12, pidhash_shift); | |
638 | pidhash_size = 1 << pidhash_shift; | |
639 | ||
640 | printk("PID hash table entries: %d (order: %d, %Zd bytes)\n", | |
641 | pidhash_size, pidhash_shift, | |
92476d7f EB |
642 | pidhash_size * sizeof(struct hlist_head)); |
643 | ||
644 | pid_hash = alloc_bootmem(pidhash_size * sizeof(*(pid_hash))); | |
645 | if (!pid_hash) | |
646 | panic("Could not alloc pidhash!\n"); | |
647 | for (i = 0; i < pidhash_size; i++) | |
648 | INIT_HLIST_HEAD(&pid_hash[i]); | |
1da177e4 LT |
649 | } |
650 | ||
651 | void __init pidmap_init(void) | |
652 | { | |
61a58c6c | 653 | init_pid_ns.pidmap[0].page = kzalloc(PAGE_SIZE, GFP_KERNEL); |
73b9ebfe | 654 | /* Reserve PID 0. We never call free_pidmap(0) */ |
61a58c6c SB |
655 | set_bit(0, init_pid_ns.pidmap[0].page); |
656 | atomic_dec(&init_pid_ns.pidmap[0].nr_free); | |
92476d7f | 657 | |
baf8f0f8 PE |
658 | init_pid_ns.pid_cachep = create_pid_cachep(1); |
659 | if (init_pid_ns.pid_cachep == NULL) | |
660 | panic("Can't create pid_1 cachep\n"); | |
c9c5d922 SB |
661 | |
662 | pid_ns_cachep = KMEM_CACHE(pid_namespace, SLAB_PANIC); | |
1da177e4 | 663 | } |