1 // SPDX-License-Identifier: GPL-2.0-only <<
2 /* 1 /*
3 * Generic pidhash and scalable, time-bounded 2 * Generic pidhash and scalable, time-bounded PID allocator
4 * 3 *
5 * (C) 2002-2003 Nadia Yvette Chambers, IBM !! 4 * (C) 2002 William Irwin, IBM
6 * (C) 2004 Nadia Yvette Chambers, Oracle !! 5 * (C) 2002 Ingo Molnar, Red Hat
7 * (C) 2002-2004 Ingo Molnar, Red Hat <<
8 * 6 *
9 * pid-structures are backing objects for task 7 * pid-structures are backing objects for tasks sharing a given ID to chain
10 * against. There is very little to them aside 8 * against. There is very little to them aside from hashing them and
11 * parking tasks using given ID's on a list. 9 * parking tasks using given ID's on a list.
12 * 10 *
13 * The hash is always changed with the tasklis 11 * The hash is always changed with the tasklist_lock write-acquired,
14 * and the hash is only accessed with the task 12 * and the hash is only accessed with the tasklist_lock at least
15 * read-acquired, so there's no additional SMP 13 * read-acquired, so there's no additional SMP locking needed here.
16 * 14 *
17 * We have a list of bitmap pages, which bitma 15 * We have a list of bitmap pages, which bitmaps represent the PID space.
18 * Allocating and freeing PIDs is completely l 16 * Allocating and freeing PIDs is completely lockless. The worst-case
19 * allocation scenario when all but one out of 17 * allocation scenario when all but one out of 1 million PIDs possible are
20 * allocated already: the scanning of 32 list 18 * allocated already: the scanning of 32 list entries and at most PAGE_SIZE
21 * bytes. The typical fastpath is a single suc 19 * bytes. The typical fastpath is a single successful setbit. Freeing is O(1).
22 * <<
23 * Pid namespaces: <<
24 * (C) 2007 Pavel Emelyanov <xemul@openvz.o <<
25 * (C) 2007 Sukadev Bhattiprolu <sukadev@us <<
26 * Many thanks to Oleg Nesterov for commen <<
27 * <<
28 */ 20 */
29 21
30 #include <linux/mm.h> 22 #include <linux/mm.h>
31 #include <linux/export.h> !! 23 #include <linux/module.h>
32 #include <linux/slab.h> 24 #include <linux/slab.h>
33 #include <linux/init.h> 25 #include <linux/init.h>
34 #include <linux/rculist.h> !! 26 #include <linux/bootmem.h>
35 #include <linux/memblock.h> !! 27 #include <linux/hash.h>
36 #include <linux/pid_namespace.h> !! 28
37 #include <linux/init_task.h> !! 29 #define pid_hashfn(nr) hash_long((unsigned long)nr, pidhash_shift)
38 #include <linux/syscalls.h> !! 30 static struct list_head *pid_hash[PIDTYPE_MAX];
39 #include <linux/proc_ns.h> !! 31 static int pidhash_shift;
40 #include <linux/refcount.h> <<
41 #include <linux/anon_inodes.h> <<
42 #include <linux/sched/signal.h> <<
43 #include <linux/sched/task.h> <<
44 #include <linux/idr.h> <<
45 #include <linux/pidfs.h> <<
46 #include <net/sock.h> <<
47 #include <uapi/linux/pidfd.h> <<
48 <<
49 struct pid init_struct_pid = { <<
50 .count = REFCOUNT_INIT(1), <<
51 .tasks = { <<
52 { .first = NULL }, <<
53 { .first = NULL }, <<
54 { .first = NULL }, <<
55 }, <<
56 .level = 0, <<
57 .numbers = { { <<
58 .nr = 0, <<
59 .ns = &init_pid_ns <<
60 }, } <<
61 }; <<
62 32
63 int pid_max = PID_MAX_DEFAULT; 33 int pid_max = PID_MAX_DEFAULT;
>> 34 int last_pid;
64 35
65 int pid_max_min = RESERVED_PIDS + 1; !! 36 #define RESERVED_PIDS 300
66 int pid_max_max = PID_MAX_LIMIT; !! 37
67 /* !! 38 #define PIDMAP_ENTRIES (PID_MAX_LIMIT/PAGE_SIZE/8)
68 * Pseudo filesystems start inode numbering af !! 39 #define BITS_PER_PAGE (PAGE_SIZE*8)
69 * PIDs as a natural offset. !! 40 #define BITS_PER_PAGE_MASK (BITS_PER_PAGE-1)
70 */ <<
71 static u64 pidfs_ino = RESERVED_PIDS; <<
72 41
73 /* 42 /*
74 * PID-map pages start out as NULL, they get a 43 * PID-map pages start out as NULL, they get allocated upon
75 * first use and are never deallocated. This w 44 * first use and are never deallocated. This way a low pid_max
76 * value does not cause lots of bitmaps to be 45 * value does not cause lots of bitmaps to be allocated, but
77 * the scheme scales to up to 4 million PIDs, 46 * the scheme scales to up to 4 million PIDs, runtime.
78 */ 47 */
79 struct pid_namespace init_pid_ns = { !! 48 typedef struct pidmap {
80 .ns.count = REFCOUNT_INIT(2), !! 49 atomic_t nr_free;
81 .idr = IDR_INIT(init_pid_ns.idr), !! 50 void *page;
82 .pid_allocated = PIDNS_ADDING, !! 51 } pidmap_t;
83 .level = 0, <<
84 .child_reaper = &init_task, <<
85 .user_ns = &init_user_ns, <<
86 .ns.inum = PROC_PID_INIT_INO, <<
87 #ifdef CONFIG_PID_NS <<
88 .ns.ops = &pidns_operations, <<
89 #endif <<
90 #if defined(CONFIG_SYSCTL) && defined(CONFIG_M <<
91 .memfd_noexec_scope = MEMFD_NOEXEC_SCO <<
92 #endif <<
93 }; <<
94 EXPORT_SYMBOL_GPL(init_pid_ns); <<
95 52
96 /* !! 53 static pidmap_t pidmap_array[PIDMAP_ENTRIES] =
97 * Note: disable interrupts while the pidmap_l !! 54 { [ 0 ... PIDMAP_ENTRIES-1 ] = { ATOMIC_INIT(BITS_PER_PAGE), NULL } };
98 * interrupt might come in and do read_lock(&t <<
99 * <<
100 * If we don't disable interrupts there is a n <<
101 * detach_pid()->free_pid() and another cpu th <<
102 * spin_lock(&pidmap_lock) followed by an inte <<
103 * read_lock(&tasklist_lock); <<
104 * <<
105 * After we clean up the tasklist_lock and kno <<
106 * irq handlers that take it we can leave the <<
107 * For now it is easier to be safe than to pro <<
108 */ <<
109 55
110 static __cacheline_aligned_in_smp DEFINE_SPIN !! 56 static pidmap_t *map_limit = pidmap_array + PIDMAP_ENTRIES;
111 57
112 void put_pid(struct pid *pid) !! 58 static spinlock_t pidmap_lock __cacheline_aligned_in_smp = SPIN_LOCK_UNLOCKED;
113 { <<
114 struct pid_namespace *ns; <<
115 59
116 if (!pid) !! 60 inline void free_pidmap(int pid)
117 return; <<
118 <<
119 ns = pid->numbers[pid->level].ns; <<
120 if (refcount_dec_and_test(&pid->count) <<
121 kmem_cache_free(ns->pid_cachep <<
122 put_pid_ns(ns); <<
123 } <<
124 } <<
125 EXPORT_SYMBOL_GPL(put_pid); <<
126 <<
127 static void delayed_put_pid(struct rcu_head *r <<
128 { <<
129 struct pid *pid = container_of(rhp, st <<
130 put_pid(pid); <<
131 } <<
132 <<
133 void free_pid(struct pid *pid) <<
134 { 61 {
135 /* We can be called with write_lock_ir !! 62 pidmap_t *map = pidmap_array + pid / BITS_PER_PAGE;
136 int i; !! 63 int offset = pid & BITS_PER_PAGE_MASK;
137 unsigned long flags; <<
138 <<
139 spin_lock_irqsave(&pidmap_lock, flags) <<
140 for (i = 0; i <= pid->level; i++) { <<
141 struct upid *upid = pid->numbe <<
142 struct pid_namespace *ns = upi <<
143 switch (--ns->pid_allocated) { <<
144 case 2: <<
145 case 1: <<
146 /* When all that is le <<
147 * is the reaper wake <<
148 * may be sleeping in <<
149 */ <<
150 wake_up_process(ns->ch <<
151 break; <<
152 case PIDNS_ADDING: <<
153 /* Handle a fork failu <<
154 WARN_ON(ns->child_reap <<
155 ns->pid_allocated = 0; <<
156 break; <<
157 } <<
158 64
159 idr_remove(&ns->idr, upid->nr) !! 65 clear_bit(offset, map->page);
160 } !! 66 atomic_inc(&map->nr_free);
161 spin_unlock_irqrestore(&pidmap_lock, f <<
162 <<
163 call_rcu(&pid->rcu, delayed_put_pid); <<
164 } 67 }
165 68
166 struct pid *alloc_pid(struct pid_namespace *ns !! 69 /*
167 size_t set_tid_size) !! 70 * Here we search for the next map that has free bits left.
>> 71 * Normally the next map has free PIDs.
>> 72 */
>> 73 static inline pidmap_t *next_free_map(pidmap_t *map, int *max_steps)
168 { 74 {
169 struct pid *pid; !! 75 while (--*max_steps) {
170 enum pid_type type; !! 76 if (++map == map_limit)
171 int i, nr; !! 77 map = pidmap_array;
172 struct pid_namespace *tmp; !! 78 if (unlikely(!map->page)) {
173 struct upid *upid; !! 79 unsigned long page = get_zeroed_page(GFP_KERNEL);
174 int retval = -ENOMEM; <<
175 <<
176 /* <<
177 * set_tid_size contains the size of t <<
178 * the most nested currently active PI <<
179 * which PID to set for a process in t <<
180 * up to set_tid_size PID namespaces. <<
181 * for a process in all nested PID nam <<
182 * never be greater than the current n <<
183 */ <<
184 if (set_tid_size > ns->level + 1) <<
185 return ERR_PTR(-EINVAL); <<
186 <<
187 pid = kmem_cache_alloc(ns->pid_cachep, <<
188 if (!pid) <<
189 return ERR_PTR(retval); <<
190 <<
191 tmp = ns; <<
192 pid->level = ns->level; <<
193 <<
194 for (i = ns->level; i >= 0; i--) { <<
195 int tid = 0; <<
196 <<
197 if (set_tid_size) { <<
198 tid = set_tid[ns->leve <<
199 <<
200 retval = -EINVAL; <<
201 if (tid < 1 || tid >= <<
202 goto out_free; <<
203 /* <<
204 * Also fail if a PID <<
205 * no PID 1 exists. <<
206 */ <<
207 if (tid != 1 && !tmp-> <<
208 goto out_free; <<
209 retval = -EPERM; <<
210 if (!checkpoint_restor <<
211 goto out_free; <<
212 set_tid_size--; <<
213 } <<
214 <<
215 idr_preload(GFP_KERNEL); <<
216 spin_lock_irq(&pidmap_lock); <<
217 <<
218 if (tid) { <<
219 nr = idr_alloc(&tmp->i <<
220 tid + 1 <<
221 /* 80 /*
222 * If ENOSPC is return !! 81 * Free the page if someone raced with us
223 * alreay in use. Retu !! 82 * installing it:
224 */ 83 */
225 if (nr == -ENOSPC) !! 84 spin_lock(&pidmap_lock);
226 nr = -EEXIST; !! 85 if (map->page)
227 } else { !! 86 free_page(page);
228 int pid_min = 1; !! 87 else
229 /* !! 88 map->page = (void *)page;
230 * init really needs p !! 89 spin_unlock(&pidmap_lock);
231 * maximum wrap back t <<
232 */ <<
233 if (idr_get_cursor(&tm <<
234 pid_min = RESE <<
235 <<
236 /* <<
237 * Store a null pointe <<
238 * a partially initial <<
239 */ <<
240 nr = idr_alloc_cyclic( <<
241 <<
242 } <<
243 spin_unlock_irq(&pidmap_lock); <<
244 idr_preload_end(); <<
245 90
246 if (nr < 0) { !! 91 if (!map->page)
247 retval = (nr == -ENOSP !! 92 break;
248 goto out_free; <<
249 } 93 }
250 !! 94 if (atomic_read(&map->nr_free))
251 pid->numbers[i].nr = nr; !! 95 return map;
252 pid->numbers[i].ns = tmp; <<
253 tmp = tmp->parent; <<
254 } 96 }
255 !! 97 return NULL;
256 /* <<
257 * ENOMEM is not the most obvious choi <<
258 * where the child subreaper has alrea <<
259 * namespace denies the creation of an <<
260 * is what we have exposed to userspac <<
261 * documented behavior for pid namespa <<
262 * change it even if there were an err <<
263 */ <<
264 retval = -ENOMEM; <<
265 <<
266 get_pid_ns(ns); <<
267 refcount_set(&pid->count, 1); <<
268 spin_lock_init(&pid->lock); <<
269 for (type = 0; type < PIDTYPE_MAX; ++t <<
270 INIT_HLIST_HEAD(&pid->tasks[ty <<
271 <<
272 init_waitqueue_head(&pid->wait_pidfd); <<
273 INIT_HLIST_HEAD(&pid->inodes); <<
274 <<
275 upid = pid->numbers + ns->level; <<
276 spin_lock_irq(&pidmap_lock); <<
277 if (!(ns->pid_allocated & PIDNS_ADDING <<
278 goto out_unlock; <<
279 pid->stashed = NULL; <<
280 pid->ino = ++pidfs_ino; <<
281 for ( ; upid >= pid->numbers; --upid) <<
282 /* Make the PID visible to fin <<
283 idr_replace(&upid->ns->idr, pi <<
284 upid->ns->pid_allocated++; <<
285 } <<
286 spin_unlock_irq(&pidmap_lock); <<
287 <<
288 return pid; <<
289 <<
290 out_unlock: <<
291 spin_unlock_irq(&pidmap_lock); <<
292 put_pid_ns(ns); <<
293 <<
294 out_free: <<
295 spin_lock_irq(&pidmap_lock); <<
296 while (++i <= ns->level) { <<
297 upid = pid->numbers + i; <<
298 idr_remove(&upid->ns->idr, upi <<
299 } <<
300 <<
301 /* On failure to allocate the first pi <<
302 if (ns->pid_allocated == PIDNS_ADDING) <<
303 idr_set_cursor(&ns->idr, 0); <<
304 <<
305 spin_unlock_irq(&pidmap_lock); <<
306 <<
307 kmem_cache_free(ns->pid_cachep, pid); <<
308 return ERR_PTR(retval); <<
309 } <<
310 <<
311 void disable_pid_allocation(struct pid_namespa <<
312 { <<
313 spin_lock_irq(&pidmap_lock); <<
314 ns->pid_allocated &= ~PIDNS_ADDING; <<
315 spin_unlock_irq(&pidmap_lock); <<
316 } 98 }
317 99
318 struct pid *find_pid_ns(int nr, struct pid_nam !! 100 int alloc_pidmap(void)
319 { 101 {
320 return idr_find(&ns->idr, nr); !! 102 int pid, offset, max_steps = PIDMAP_ENTRIES + 1;
321 } !! 103 pidmap_t *map;
322 EXPORT_SYMBOL_GPL(find_pid_ns); <<
323 104
324 struct pid *find_vpid(int nr) !! 105 pid = last_pid + 1;
325 { !! 106 if (pid >= pid_max)
326 return find_pid_ns(nr, task_active_pid !! 107 pid = RESERVED_PIDS;
327 } <<
328 EXPORT_SYMBOL_GPL(find_vpid); <<
329 108
330 static struct pid **task_pid_ptr(struct task_s !! 109 offset = pid & BITS_PER_PAGE_MASK;
331 { !! 110 map = pidmap_array + pid / BITS_PER_PAGE;
332 return (type == PIDTYPE_PID) ? <<
333 &task->thread_pid : <<
334 &task->signal->pids[type]; <<
335 } <<
336 111
337 /* !! 112 if (likely(map->page && !test_and_set_bit(offset, map->page))) {
338 * attach_pid() must be called with the taskli !! 113 /*
339 */ !! 114 * There is a small window for last_pid updates to race,
340 void attach_pid(struct task_struct *task, enum !! 115 * but in that case the next allocation will go into the
341 { !! 116 * slowpath and that fixes things up.
342 struct pid *pid = *task_pid_ptr(task, !! 117 */
343 hlist_add_head_rcu(&task->pid_links[ty !! 118 return_pid:
344 } !! 119 atomic_dec(&map->nr_free);
345 !! 120 last_pid = pid;
346 static void __change_pid(struct task_struct *t !! 121 return pid;
347 struct pid *new) <<
348 { <<
349 struct pid **pid_ptr = task_pid_ptr(ta <<
350 struct pid *pid; <<
351 int tmp; <<
352 <<
353 pid = *pid_ptr; <<
354 <<
355 hlist_del_rcu(&task->pid_links[type]); <<
356 *pid_ptr = new; <<
357 <<
358 if (type == PIDTYPE_PID) { <<
359 WARN_ON_ONCE(pid_has_task(pid, <<
360 wake_up_all(&pid->wait_pidfd); <<
361 } 122 }
362 !! 123
363 for (tmp = PIDTYPE_MAX; --tmp >= 0; ) !! 124 if (!offset || !atomic_read(&map->nr_free)) {
364 if (pid_has_task(pid, tmp)) !! 125 next_map:
365 return; !! 126 map = next_free_map(map, &max_steps);
366 !! 127 if (!map)
367 free_pid(pid); !! 128 goto failure;
368 } !! 129 offset = 0;
369 <<
370 void detach_pid(struct task_struct *task, enum <<
371 { <<
372 __change_pid(task, type, NULL); <<
373 } <<
374 <<
375 void change_pid(struct task_struct *task, enum <<
376 struct pid *pid) <<
377 { <<
378 __change_pid(task, type, pid); <<
379 attach_pid(task, type); <<
380 } <<
381 <<
382 void exchange_tids(struct task_struct *left, s <<
383 { <<
384 struct pid *pid1 = left->thread_pid; <<
385 struct pid *pid2 = right->thread_pid; <<
386 struct hlist_head *head1 = &pid1->task <<
387 struct hlist_head *head2 = &pid2->task <<
388 <<
389 /* Swap the single entry tid lists */ <<
390 hlists_swap_heads_rcu(head1, head2); <<
391 <<
392 /* Swap the per task_struct pid */ <<
393 rcu_assign_pointer(left->thread_pid, p <<
394 rcu_assign_pointer(right->thread_pid, <<
395 <<
396 /* Swap the cached value */ <<
397 WRITE_ONCE(left->pid, pid_nr(pid2)); <<
398 WRITE_ONCE(right->pid, pid_nr(pid1)); <<
399 } <<
400 <<
401 /* transfer_pid is an optimization of attach_p <<
402 void transfer_pid(struct task_struct *old, str <<
403 enum pid_type type) <<
404 { <<
405 WARN_ON_ONCE(type == PIDTYPE_PID); <<
406 hlist_replace_rcu(&old->pid_links[type <<
407 } <<
408 <<
409 struct task_struct *pid_task(struct pid *pid, <<
410 { <<
411 struct task_struct *result = NULL; <<
412 if (pid) { <<
413 struct hlist_node *first; <<
414 first = rcu_dereference_check( <<
415 <<
416 if (first) <<
417 result = hlist_entry(f <<
418 } 130 }
419 return result; !! 131 /*
420 } !! 132 * Find the next zero bit:
421 EXPORT_SYMBOL(pid_task); !! 133 */
422 !! 134 scan_more:
423 /* !! 135 offset = find_next_zero_bit(map->page, BITS_PER_PAGE, offset);
424 * Must be called under rcu_read_lock(). !! 136 if (offset >= BITS_PER_PAGE)
425 */ !! 137 goto next_map;
426 struct task_struct *find_task_by_pid_ns(pid_t !! 138 if (test_and_set_bit(offset, map->page))
427 { !! 139 goto scan_more;
428 RCU_LOCKDEP_WARN(!rcu_read_lock_held() <<
429 "find_task_by_pid_ns( <<
430 return pid_task(find_pid_ns(nr, ns), P <<
431 } <<
432 <<
433 struct task_struct *find_task_by_vpid(pid_t vn <<
434 { <<
435 return find_task_by_pid_ns(vnr, task_a <<
436 } <<
437 <<
438 struct task_struct *find_get_task_by_vpid(pid_ <<
439 { <<
440 struct task_struct *task; <<
441 <<
442 rcu_read_lock(); <<
443 task = find_task_by_vpid(nr); <<
444 if (task) <<
445 get_task_struct(task); <<
446 rcu_read_unlock(); <<
447 140
448 return task; !! 141 /* we got the PID: */
449 } !! 142 pid = (map - pidmap_array) * BITS_PER_PAGE + offset;
>> 143 goto return_pid;
450 144
451 struct pid *get_task_pid(struct task_struct *t !! 145 failure:
452 { !! 146 return -1;
453 struct pid *pid; <<
454 rcu_read_lock(); <<
455 pid = get_pid(rcu_dereference(*task_pi <<
456 rcu_read_unlock(); <<
457 return pid; <<
458 } <<
459 EXPORT_SYMBOL_GPL(get_task_pid); <<
460 <<
461 struct task_struct *get_pid_task(struct pid *p <<
462 { <<
463 struct task_struct *result; <<
464 rcu_read_lock(); <<
465 result = pid_task(pid, type); <<
466 if (result) <<
467 get_task_struct(result); <<
468 rcu_read_unlock(); <<
469 return result; <<
470 } 147 }
471 EXPORT_SYMBOL_GPL(get_pid_task); <<
472 148
473 struct pid *find_get_pid(pid_t nr) !! 149 inline struct pid *find_pid(enum pid_type type, int nr)
474 { 150 {
>> 151 struct list_head *elem, *bucket = &pid_hash[type][pid_hashfn(nr)];
475 struct pid *pid; 152 struct pid *pid;
476 153
477 rcu_read_lock(); !! 154 __list_for_each(elem, bucket) {
478 pid = get_pid(find_vpid(nr)); !! 155 pid = list_entry(elem, struct pid, hash_chain);
479 rcu_read_unlock(); !! 156 if (pid->nr == nr)
480 !! 157 return pid;
481 return pid; <<
482 } <<
483 EXPORT_SYMBOL_GPL(find_get_pid); <<
484 <<
485 pid_t pid_nr_ns(struct pid *pid, struct pid_na <<
486 { <<
487 struct upid *upid; <<
488 pid_t nr = 0; <<
489 <<
490 if (pid && ns->level <= pid->level) { <<
491 upid = &pid->numbers[ns->level <<
492 if (upid->ns == ns) <<
493 nr = upid->nr; <<
494 } 158 }
495 return nr; !! 159 return NULL;
496 } <<
497 EXPORT_SYMBOL_GPL(pid_nr_ns); <<
498 <<
499 pid_t pid_vnr(struct pid *pid) <<
500 { <<
501 return pid_nr_ns(pid, task_active_pid_ <<
502 } <<
503 EXPORT_SYMBOL_GPL(pid_vnr); <<
504 <<
505 pid_t __task_pid_nr_ns(struct task_struct *tas <<
506 struct pid_namespace * <<
507 { <<
508 pid_t nr = 0; <<
509 <<
510 rcu_read_lock(); <<
511 if (!ns) <<
512 ns = task_active_pid_ns(curren <<
513 nr = pid_nr_ns(rcu_dereference(*task_p <<
514 rcu_read_unlock(); <<
515 <<
516 return nr; <<
517 } <<
518 EXPORT_SYMBOL(__task_pid_nr_ns); <<
519 <<
520 struct pid_namespace *task_active_pid_ns(struc <<
521 { <<
522 return ns_of_pid(task_pid(tsk)); <<
523 } 160 }
524 EXPORT_SYMBOL_GPL(task_active_pid_ns); <<
525 161
526 /* !! 162 void link_pid(task_t *task, struct pid_link *link, struct pid *pid)
527 * Used by proc to find the first pid that is <<
528 * <<
529 * If there is a pid at nr this function is ex <<
530 */ <<
531 struct pid *find_ge_pid(int nr, struct pid_nam <<
532 { 163 {
533 return idr_get_next(&ns->idr, &nr); !! 164 atomic_inc(&pid->count);
>> 165 list_add_tail(&link->pid_chain, &pid->task_list);
>> 166 link->pidptr = pid;
534 } 167 }
535 EXPORT_SYMBOL_GPL(find_ge_pid); <<
536 168
537 struct pid *pidfd_get_pid(unsigned int fd, uns !! 169 int attach_pid(task_t *task, enum pid_type type, int nr)
538 { 170 {
539 struct fd f; !! 171 struct pid *pid = find_pid(type, nr);
540 struct pid *pid; <<
541 172
542 f = fdget(fd); !! 173 if (pid)
543 if (!fd_file(f)) !! 174 atomic_inc(&pid->count);
544 return ERR_PTR(-EBADF); !! 175 else {
545 !! 176 pid = &task->pids[type].pid;
546 pid = pidfd_pid(fd_file(f)); !! 177 pid->nr = nr;
547 if (!IS_ERR(pid)) { !! 178 atomic_set(&pid->count, 1);
548 get_pid(pid); !! 179 INIT_LIST_HEAD(&pid->task_list);
549 *flags = fd_file(f)->f_flags; !! 180 pid->task = task;
>> 181 get_task_struct(task);
>> 182 list_add(&pid->hash_chain, &pid_hash[type][pid_hashfn(nr)]);
550 } 183 }
>> 184 list_add_tail(&task->pids[type].pid_chain, &pid->task_list);
>> 185 task->pids[type].pidptr = pid;
551 186
552 fdput(f); !! 187 return 0;
553 return pid; <<
554 } 188 }
555 189
556 /** !! 190 static inline int __detach_pid(task_t *task, enum pid_type type)
557 * pidfd_get_task() - Get the task associated <<
558 * <<
559 * @pidfd: pidfd for which to get the task <<
560 * @flags: flags associated with this pidfd <<
561 * <<
562 * Return the task associated with @pidfd. The <<
563 * the returned task. The caller is responsibl <<
564 * <<
565 * Return: On success, the task_struct associa <<
566 * On error, a negative errno number w <<
567 */ <<
568 struct task_struct *pidfd_get_task(int pidfd, <<
569 { 191 {
570 unsigned int f_flags; !! 192 struct pid_link *link = task->pids + type;
571 struct pid *pid; !! 193 struct pid *pid = link->pidptr;
572 struct task_struct *task; !! 194 int nr;
573 195
574 pid = pidfd_get_pid(pidfd, &f_flags); !! 196 list_del(&link->pid_chain);
575 if (IS_ERR(pid)) !! 197 if (!atomic_dec_and_test(&pid->count))
576 return ERR_CAST(pid); !! 198 return 0;
577 199
578 task = get_pid_task(pid, PIDTYPE_TGID) !! 200 nr = pid->nr;
579 put_pid(pid); !! 201 list_del(&pid->hash_chain);
580 if (!task) !! 202 put_task_struct(pid->task);
581 return ERR_PTR(-ESRCH); <<
582 203
583 *flags = f_flags; !! 204 return nr;
584 return task; <<
585 } 205 }
586 206
587 /** !! 207 static void _detach_pid(task_t *task, enum pid_type type)
588 * pidfd_create() - Create a new pid file desc <<
589 * <<
590 * @pid: struct pid that the pidfd will refe <<
591 * @flags: flags to pass <<
592 * <<
593 * This creates a new pid file descriptor with <<
594 * <<
595 * Note, that this function can only be called <<
596 * been unshared to avoid leaking the pidfd to <<
597 * <<
598 * This symbol should not be explicitly export <<
599 * <<
600 * Return: On success, a cloexec pidfd is retu <<
601 * On error, a negative errno number w <<
602 */ <<
603 static int pidfd_create(struct pid *pid, unsig <<
604 { 208 {
605 int pidfd; !! 209 __detach_pid(task, type);
606 struct file *pidfd_file; <<
607 <<
608 pidfd = pidfd_prepare(pid, flags, &pid <<
609 if (pidfd < 0) <<
610 return pidfd; <<
611 <<
612 fd_install(pidfd, pidfd_file); <<
613 return pidfd; <<
614 } 210 }
615 211
616 /** !! 212 void detach_pid(task_t *task, enum pid_type type)
617 * sys_pidfd_open() - Open new pid file descri <<
618 * <<
619 * @pid: pid for which to retrieve a pidfd <<
620 * @flags: flags to pass <<
621 * <<
622 * This creates a new pid file descriptor with <<
623 * the task identified by @pid. Without PIDFD_ <<
624 * must be a thread-group leader. <<
625 * <<
626 * Return: On success, a cloexec pidfd is retu <<
627 * On error, a negative errno number w <<
628 */ <<
629 SYSCALL_DEFINE2(pidfd_open, pid_t, pid, unsign <<
630 { 213 {
631 int fd; !! 214 int nr = __detach_pid(task, type);
632 struct pid *p; <<
633 <<
634 if (flags & ~(PIDFD_NONBLOCK | PIDFD_T <<
635 return -EINVAL; <<
636 <<
637 if (pid <= 0) <<
638 return -EINVAL; <<
639 215
640 p = find_get_pid(pid); !! 216 if (!nr)
641 if (!p) !! 217 return;
642 return -ESRCH; <<
643 <<
644 fd = pidfd_create(p, flags); <<
645 218
646 put_pid(p); !! 219 for (type = 0; type < PIDTYPE_MAX; ++type)
647 return fd; !! 220 if (find_pid(type, nr))
>> 221 return;
>> 222 free_pidmap(nr);
648 } 223 }
649 224
650 void __init pid_idr_init(void) !! 225 task_t *find_task_by_pid(int nr)
651 { 226 {
652 /* Verify no one has done anything sil !! 227 struct pid *pid = find_pid(PIDTYPE_PID, nr);
653 BUILD_BUG_ON(PID_MAX_LIMIT >= PIDNS_AD <<
654 <<
655 /* bump default and minimum pid_max ba <<
656 pid_max = min(pid_max_max, max_t(int, <<
657 PIDS_PER_CPU_D <<
658 pid_max_min = max_t(int, pid_max_min, <<
659 PIDS_PER_CPU_M <<
660 pr_info("pid_max: default: %u minimum: <<
661 <<
662 idr_init(&init_pid_ns.idr); <<
663 228
664 init_pid_ns.pid_cachep = kmem_cache_cr !! 229 if (!pid)
665 struct_size_t(struct p !! 230 return NULL;
666 __alignof__(struct pid !! 231 return pid_task(pid->task_list.next, PIDTYPE_PID);
667 SLAB_HWCACHE_ALIGN | S <<
668 NULL); <<
669 } 232 }
670 233
671 static struct file *__pidfd_fget(struct task_s !! 234 EXPORT_SYMBOL(find_task_by_pid);
672 { <<
673 struct file *file; <<
674 int ret; <<
675 <<
676 ret = down_read_killable(&task->signal <<
677 if (ret) <<
678 return ERR_PTR(ret); <<
679 235
680 if (ptrace_may_access(task, PTRACE_MOD !! 236 /*
681 file = fget_task(task, fd); !! 237 * This function switches the PIDs if a non-leader thread calls
682 else !! 238 * sys_execve() - this must be done without releasing the PID.
683 file = ERR_PTR(-EPERM); !! 239 * (which a detach_pid() would eventually do.)
684 !! 240 */
685 up_read(&task->signal->exec_update_loc !! 241 void switch_exec_pids(task_t *leader, task_t *thread)
>> 242 {
>> 243 _detach_pid(leader, PIDTYPE_PID);
>> 244 _detach_pid(leader, PIDTYPE_TGID);
>> 245 _detach_pid(leader, PIDTYPE_PGID);
>> 246 _detach_pid(leader, PIDTYPE_SID);
>> 247
>> 248 _detach_pid(thread, PIDTYPE_PID);
>> 249 _detach_pid(thread, PIDTYPE_TGID);
>> 250
>> 251 leader->pid = leader->tgid = thread->pid;
>> 252 thread->pid = thread->tgid;
>> 253
>> 254 attach_pid(thread, PIDTYPE_PID, thread->pid);
>> 255 attach_pid(thread, PIDTYPE_TGID, thread->tgid);
>> 256 attach_pid(thread, PIDTYPE_PGID, leader->__pgrp);
>> 257 attach_pid(thread, PIDTYPE_SID, thread->session);
>> 258 list_add_tail(&thread->tasks, &init_task.tasks);
>> 259
>> 260 attach_pid(leader, PIDTYPE_PID, leader->pid);
>> 261 attach_pid(leader, PIDTYPE_TGID, leader->tgid);
>> 262 attach_pid(leader, PIDTYPE_PGID, leader->__pgrp);
>> 263 attach_pid(leader, PIDTYPE_SID, leader->session);
>> 264 }
686 265
687 if (!file) { !! 266 /*
688 /* !! 267 * The pid hash table is scaled according to the amount of memory in the
689 * It is possible that the tar !! 268 * machine. From a minimum of 16 slots up to 4096 slots at one gigabyte or
690 * either: !! 269 * more.
691 * 1. before exit_signals(), w !! 270 */
692 * 2. before exit_files() take !! 271 void __init pidhash_init(void)
693 * 3. after exit_files() relea !! 272 {
694 * this has PF_EXITING, sin !! 273 int i, j, pidhash_size;
695 * __pidfd_fget() returns E !! 274 unsigned long megabytes = max_pfn >> (20 - PAGE_SHIFT);
696 * In case 3 we get EBADF, but !! 275
697 * the task is currently exiti !! 276 pidhash_shift = max(4, fls(megabytes * 4));
698 * struct, so we fix it up. !! 277 pidhash_shift = min(12, pidhash_shift);
699 */ !! 278 pidhash_size = 1 << pidhash_shift;
700 if (task->flags & PF_EXITING) !! 279
701 file = ERR_PTR(-ESRCH) !! 280 printk("PID hash table entries: %d (order %d: %Zd bytes)\n",
702 else !! 281 pidhash_size, pidhash_shift,
703 file = ERR_PTR(-EBADF) !! 282 pidhash_size * sizeof(struct list_head));
>> 283
>> 284 for (i = 0; i < PIDTYPE_MAX; i++) {
>> 285 pid_hash[i] = alloc_bootmem(pidhash_size *
>> 286 sizeof(struct list_head));
>> 287 if (!pid_hash[i])
>> 288 panic("Could not alloc pidhash!\n");
>> 289 for (j = 0; j < pidhash_size; j++)
>> 290 INIT_LIST_HEAD(&pid_hash[i][j]);
704 } 291 }
705 <<
706 return file; <<
707 } 292 }
708 293
709 static int pidfd_getfd(struct pid *pid, int fd !! 294 void __init pidmap_init(void)
710 { 295 {
711 struct task_struct *task; !! 296 int i;
712 struct file *file; <<
713 int ret; <<
714 <<
715 task = get_pid_task(pid, PIDTYPE_PID); <<
716 if (!task) <<
717 return -ESRCH; <<
718 <<
719 file = __pidfd_fget(task, fd); <<
720 put_task_struct(task); <<
721 if (IS_ERR(file)) <<
722 return PTR_ERR(file); <<
723 <<
724 ret = receive_fd(file, NULL, O_CLOEXEC <<
725 fput(file); <<
726 <<
727 return ret; <<
728 } <<
729 297
730 /** !! 298 pidmap_array->page = (void *)get_zeroed_page(GFP_KERNEL);
731 * sys_pidfd_getfd() - Get a file descriptor f !! 299 set_bit(0, pidmap_array->page);
732 * !! 300 atomic_dec(&pidmap_array->nr_free);
733 * @pidfd: the pidfd file descriptor of t <<
734 * @fd: the file descriptor number to <<
735 * @flags: flags on how to get the fd (re <<
736 * <<
737 * This syscall gets a copy of a file descript <<
738 * based on the pidfd, and file descriptor num <<
739 * the calling process has the ability to ptra <<
740 * by the pidfd. The process which is having i <<
741 * is otherwise unaffected. <<
742 * <<
743 * Return: On success, a cloexec file descript <<
744 * On error, a negative errno number w <<
745 */ <<
746 SYSCALL_DEFINE3(pidfd_getfd, int, pidfd, int, <<
747 unsigned int, flags) <<
748 { <<
749 struct pid *pid; <<
750 struct fd f; <<
751 int ret; <<
752 301
753 /* flags is currently unused - make su !! 302 /*
754 if (flags) !! 303 * Allocate PID 0, and hash it via all PID types:
755 return -EINVAL; !! 304 */
756 <<
757 f = fdget(pidfd); <<
758 if (!fd_file(f)) <<
759 return -EBADF; <<
760 <<
761 pid = pidfd_pid(fd_file(f)); <<
762 if (IS_ERR(pid)) <<
763 ret = PTR_ERR(pid); <<
764 else <<
765 ret = pidfd_getfd(pid, fd); <<
766 305
767 fdput(f); !! 306 for (i = 0; i < PIDTYPE_MAX; i++)
768 return ret; !! 307 attach_pid(current, i, 0);
769 } 308 }
770 309
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.