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) 2002-2003 Nadia Yvette Chambers, IBM
6 * (C) 2004 Nadia Yvette Chambers, Oracle 5 * (C) 2004 Nadia Yvette Chambers, Oracle
7 * (C) 2002-2004 Ingo Molnar, Red Hat 6 * (C) 2002-2004 Ingo Molnar, Red Hat
8 * 7 *
9 * pid-structures are backing objects for task 8 * pid-structures are backing objects for tasks sharing a given ID to chain
10 * against. There is very little to them aside 9 * against. There is very little to them aside from hashing them and
11 * parking tasks using given ID's on a list. 10 * parking tasks using given ID's on a list.
12 * 11 *
13 * The hash is always changed with the tasklis 12 * The hash is always changed with the tasklist_lock write-acquired,
14 * and the hash is only accessed with the task 13 * and the hash is only accessed with the tasklist_lock at least
15 * read-acquired, so there's no additional SMP 14 * read-acquired, so there's no additional SMP locking needed here.
16 * 15 *
17 * We have a list of bitmap pages, which bitma 16 * We have a list of bitmap pages, which bitmaps represent the PID space.
18 * Allocating and freeing PIDs is completely l 17 * Allocating and freeing PIDs is completely lockless. The worst-case
19 * allocation scenario when all but one out of 18 * allocation scenario when all but one out of 1 million PIDs possible are
20 * allocated already: the scanning of 32 list 19 * allocated already: the scanning of 32 list entries and at most PAGE_SIZE
21 * bytes. The typical fastpath is a single suc 20 * bytes. The typical fastpath is a single successful setbit. Freeing is O(1).
22 * 21 *
23 * Pid namespaces: 22 * Pid namespaces:
24 * (C) 2007 Pavel Emelyanov <xemul@openvz.o 23 * (C) 2007 Pavel Emelyanov <xemul@openvz.org>, OpenVZ, SWsoft Inc.
25 * (C) 2007 Sukadev Bhattiprolu <sukadev@us 24 * (C) 2007 Sukadev Bhattiprolu <sukadev@us.ibm.com>, IBM
26 * Many thanks to Oleg Nesterov for commen 25 * Many thanks to Oleg Nesterov for comments and help
27 * 26 *
28 */ 27 */
29 28
30 #include <linux/mm.h> 29 #include <linux/mm.h>
31 #include <linux/export.h> 30 #include <linux/export.h>
32 #include <linux/slab.h> 31 #include <linux/slab.h>
33 #include <linux/init.h> 32 #include <linux/init.h>
34 #include <linux/rculist.h> 33 #include <linux/rculist.h>
35 #include <linux/memblock.h> !! 34 #include <linux/bootmem.h>
>> 35 #include <linux/hash.h>
36 #include <linux/pid_namespace.h> 36 #include <linux/pid_namespace.h>
37 #include <linux/init_task.h> 37 #include <linux/init_task.h>
38 #include <linux/syscalls.h> 38 #include <linux/syscalls.h>
39 #include <linux/proc_ns.h> 39 #include <linux/proc_ns.h>
40 #include <linux/refcount.h> !! 40 #include <linux/proc_fs.h>
41 #include <linux/anon_inodes.h> <<
42 #include <linux/sched/signal.h> <<
43 #include <linux/sched/task.h> 41 #include <linux/sched/task.h>
44 #include <linux/idr.h> 42 #include <linux/idr.h>
45 #include <linux/pidfs.h> <<
46 #include <net/sock.h> <<
47 #include <uapi/linux/pidfd.h> <<
48 43
49 struct pid init_struct_pid = { 44 struct pid init_struct_pid = {
50 .count = REFCOUNT_INIT(1), !! 45 .count = ATOMIC_INIT(1),
51 .tasks = { 46 .tasks = {
52 { .first = NULL }, 47 { .first = NULL },
53 { .first = NULL }, 48 { .first = NULL },
54 { .first = NULL }, 49 { .first = NULL },
55 }, 50 },
56 .level = 0, 51 .level = 0,
57 .numbers = { { 52 .numbers = { {
58 .nr = 0, 53 .nr = 0,
59 .ns = &init_pid_ns 54 .ns = &init_pid_ns,
60 }, } 55 }, }
61 }; 56 };
62 57
63 int pid_max = PID_MAX_DEFAULT; 58 int pid_max = PID_MAX_DEFAULT;
64 59
>> 60 #define RESERVED_PIDS 300
>> 61
65 int pid_max_min = RESERVED_PIDS + 1; 62 int pid_max_min = RESERVED_PIDS + 1;
66 int pid_max_max = PID_MAX_LIMIT; 63 int pid_max_max = PID_MAX_LIMIT;
67 /* <<
68 * Pseudo filesystems start inode numbering af <<
69 * PIDs as a natural offset. <<
70 */ <<
71 static u64 pidfs_ino = RESERVED_PIDS; <<
72 64
73 /* 65 /*
74 * PID-map pages start out as NULL, they get a 66 * PID-map pages start out as NULL, they get allocated upon
75 * first use and are never deallocated. This w 67 * first use and are never deallocated. This way a low pid_max
76 * value does not cause lots of bitmaps to be 68 * value does not cause lots of bitmaps to be allocated, but
77 * the scheme scales to up to 4 million PIDs, 69 * the scheme scales to up to 4 million PIDs, runtime.
78 */ 70 */
79 struct pid_namespace init_pid_ns = { 71 struct pid_namespace init_pid_ns = {
80 .ns.count = REFCOUNT_INIT(2), !! 72 .kref = KREF_INIT(2),
81 .idr = IDR_INIT(init_pid_ns.idr), 73 .idr = IDR_INIT(init_pid_ns.idr),
82 .pid_allocated = PIDNS_ADDING, 74 .pid_allocated = PIDNS_ADDING,
83 .level = 0, 75 .level = 0,
84 .child_reaper = &init_task, 76 .child_reaper = &init_task,
85 .user_ns = &init_user_ns, 77 .user_ns = &init_user_ns,
86 .ns.inum = PROC_PID_INIT_INO, 78 .ns.inum = PROC_PID_INIT_INO,
87 #ifdef CONFIG_PID_NS 79 #ifdef CONFIG_PID_NS
88 .ns.ops = &pidns_operations, 80 .ns.ops = &pidns_operations,
89 #endif 81 #endif
90 #if defined(CONFIG_SYSCTL) && defined(CONFIG_M <<
91 .memfd_noexec_scope = MEMFD_NOEXEC_SCO <<
92 #endif <<
93 }; 82 };
94 EXPORT_SYMBOL_GPL(init_pid_ns); 83 EXPORT_SYMBOL_GPL(init_pid_ns);
95 84
96 /* 85 /*
97 * Note: disable interrupts while the pidmap_l 86 * Note: disable interrupts while the pidmap_lock is held as an
98 * interrupt might come in and do read_lock(&t 87 * interrupt might come in and do read_lock(&tasklist_lock).
99 * 88 *
100 * If we don't disable interrupts there is a n 89 * If we don't disable interrupts there is a nasty deadlock between
101 * detach_pid()->free_pid() and another cpu th 90 * detach_pid()->free_pid() and another cpu that does
102 * spin_lock(&pidmap_lock) followed by an inte 91 * spin_lock(&pidmap_lock) followed by an interrupt routine that does
103 * read_lock(&tasklist_lock); 92 * read_lock(&tasklist_lock);
104 * 93 *
105 * After we clean up the tasklist_lock and kno 94 * After we clean up the tasklist_lock and know there are no
106 * irq handlers that take it we can leave the 95 * irq handlers that take it we can leave the interrupts enabled.
107 * For now it is easier to be safe than to pro 96 * For now it is easier to be safe than to prove it can't happen.
108 */ 97 */
109 98
110 static __cacheline_aligned_in_smp DEFINE_SPIN 99 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(pidmap_lock);
111 100
112 void put_pid(struct pid *pid) 101 void put_pid(struct pid *pid)
113 { 102 {
114 struct pid_namespace *ns; 103 struct pid_namespace *ns;
115 104
116 if (!pid) 105 if (!pid)
117 return; 106 return;
118 107
119 ns = pid->numbers[pid->level].ns; 108 ns = pid->numbers[pid->level].ns;
120 if (refcount_dec_and_test(&pid->count) !! 109 if ((atomic_read(&pid->count) == 1) ||
>> 110 atomic_dec_and_test(&pid->count)) {
121 kmem_cache_free(ns->pid_cachep 111 kmem_cache_free(ns->pid_cachep, pid);
122 put_pid_ns(ns); 112 put_pid_ns(ns);
123 } 113 }
124 } 114 }
125 EXPORT_SYMBOL_GPL(put_pid); 115 EXPORT_SYMBOL_GPL(put_pid);
126 116
127 static void delayed_put_pid(struct rcu_head *r 117 static void delayed_put_pid(struct rcu_head *rhp)
128 { 118 {
129 struct pid *pid = container_of(rhp, st 119 struct pid *pid = container_of(rhp, struct pid, rcu);
130 put_pid(pid); 120 put_pid(pid);
131 } 121 }
132 122
133 void free_pid(struct pid *pid) 123 void free_pid(struct pid *pid)
134 { 124 {
135 /* We can be called with write_lock_ir 125 /* We can be called with write_lock_irq(&tasklist_lock) held */
136 int i; 126 int i;
137 unsigned long flags; 127 unsigned long flags;
138 128
139 spin_lock_irqsave(&pidmap_lock, flags) 129 spin_lock_irqsave(&pidmap_lock, flags);
140 for (i = 0; i <= pid->level; i++) { 130 for (i = 0; i <= pid->level; i++) {
141 struct upid *upid = pid->numbe 131 struct upid *upid = pid->numbers + i;
142 struct pid_namespace *ns = upi 132 struct pid_namespace *ns = upid->ns;
143 switch (--ns->pid_allocated) { 133 switch (--ns->pid_allocated) {
144 case 2: 134 case 2:
145 case 1: 135 case 1:
146 /* When all that is le 136 /* When all that is left in the pid namespace
147 * is the reaper wake 137 * is the reaper wake up the reaper. The reaper
148 * may be sleeping in 138 * may be sleeping in zap_pid_ns_processes().
149 */ 139 */
150 wake_up_process(ns->ch 140 wake_up_process(ns->child_reaper);
151 break; 141 break;
152 case PIDNS_ADDING: 142 case PIDNS_ADDING:
153 /* Handle a fork failu 143 /* Handle a fork failure of the first process */
154 WARN_ON(ns->child_reap 144 WARN_ON(ns->child_reaper);
155 ns->pid_allocated = 0; 145 ns->pid_allocated = 0;
>> 146 /* fall through */
>> 147 case 0:
>> 148 schedule_work(&ns->proc_work);
156 break; 149 break;
157 } 150 }
158 151
159 idr_remove(&ns->idr, upid->nr) 152 idr_remove(&ns->idr, upid->nr);
160 } 153 }
161 spin_unlock_irqrestore(&pidmap_lock, f 154 spin_unlock_irqrestore(&pidmap_lock, flags);
162 155
163 call_rcu(&pid->rcu, delayed_put_pid); 156 call_rcu(&pid->rcu, delayed_put_pid);
164 } 157 }
165 158
166 struct pid *alloc_pid(struct pid_namespace *ns !! 159 struct pid *alloc_pid(struct pid_namespace *ns)
167 size_t set_tid_size) <<
168 { 160 {
169 struct pid *pid; 161 struct pid *pid;
170 enum pid_type type; 162 enum pid_type type;
171 int i, nr; 163 int i, nr;
172 struct pid_namespace *tmp; 164 struct pid_namespace *tmp;
173 struct upid *upid; 165 struct upid *upid;
174 int retval = -ENOMEM; 166 int retval = -ENOMEM;
175 167
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, 168 pid = kmem_cache_alloc(ns->pid_cachep, GFP_KERNEL);
188 if (!pid) 169 if (!pid)
189 return ERR_PTR(retval); 170 return ERR_PTR(retval);
190 171
191 tmp = ns; 172 tmp = ns;
192 pid->level = ns->level; 173 pid->level = ns->level;
193 174
194 for (i = ns->level; i >= 0; i--) { 175 for (i = ns->level; i >= 0; i--) {
195 int tid = 0; !! 176 int pid_min = 1;
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 } <<
214 177
215 idr_preload(GFP_KERNEL); 178 idr_preload(GFP_KERNEL);
216 spin_lock_irq(&pidmap_lock); 179 spin_lock_irq(&pidmap_lock);
217 180
218 if (tid) { !! 181 /*
219 nr = idr_alloc(&tmp->i !! 182 * init really needs pid 1, but after reaching the maximum
220 tid + 1 !! 183 * wrap back to RESERVED_PIDS
221 /* !! 184 */
222 * If ENOSPC is return !! 185 if (idr_get_cursor(&tmp->idr) > RESERVED_PIDS)
223 * alreay in use. Retu !! 186 pid_min = RESERVED_PIDS;
224 */ <<
225 if (nr == -ENOSPC) <<
226 nr = -EEXIST; <<
227 } else { <<
228 int pid_min = 1; <<
229 /* <<
230 * init really needs p <<
231 * maximum wrap back t <<
232 */ <<
233 if (idr_get_cursor(&tm <<
234 pid_min = RESE <<
235 187
236 /* !! 188 /*
237 * Store a null pointe !! 189 * Store a null pointer so find_pid_ns does not find
238 * a partially initial !! 190 * a partially initialized PID (see below).
239 */ !! 191 */
240 nr = idr_alloc_cyclic( !! 192 nr = idr_alloc_cyclic(&tmp->idr, NULL, pid_min,
241 !! 193 pid_max, GFP_ATOMIC);
242 } <<
243 spin_unlock_irq(&pidmap_lock); 194 spin_unlock_irq(&pidmap_lock);
244 idr_preload_end(); 195 idr_preload_end();
245 196
246 if (nr < 0) { 197 if (nr < 0) {
247 retval = (nr == -ENOSP 198 retval = (nr == -ENOSPC) ? -EAGAIN : nr;
248 goto out_free; 199 goto out_free;
249 } 200 }
250 201
251 pid->numbers[i].nr = nr; 202 pid->numbers[i].nr = nr;
252 pid->numbers[i].ns = tmp; 203 pid->numbers[i].ns = tmp;
253 tmp = tmp->parent; 204 tmp = tmp->parent;
254 } 205 }
255 206
256 /* !! 207 if (unlikely(is_child_reaper(pid))) {
257 * ENOMEM is not the most obvious choi !! 208 if (pid_ns_prepare_proc(ns))
258 * where the child subreaper has alrea !! 209 goto out_free;
259 * namespace denies the creation of an !! 210 }
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; <<
265 211
266 get_pid_ns(ns); 212 get_pid_ns(ns);
267 refcount_set(&pid->count, 1); !! 213 atomic_set(&pid->count, 1);
268 spin_lock_init(&pid->lock); <<
269 for (type = 0; type < PIDTYPE_MAX; ++t 214 for (type = 0; type < PIDTYPE_MAX; ++type)
270 INIT_HLIST_HEAD(&pid->tasks[ty 215 INIT_HLIST_HEAD(&pid->tasks[type]);
271 216
272 init_waitqueue_head(&pid->wait_pidfd); <<
273 INIT_HLIST_HEAD(&pid->inodes); <<
274 <<
275 upid = pid->numbers + ns->level; 217 upid = pid->numbers + ns->level;
276 spin_lock_irq(&pidmap_lock); 218 spin_lock_irq(&pidmap_lock);
277 if (!(ns->pid_allocated & PIDNS_ADDING 219 if (!(ns->pid_allocated & PIDNS_ADDING))
278 goto out_unlock; 220 goto out_unlock;
279 pid->stashed = NULL; <<
280 pid->ino = ++pidfs_ino; <<
281 for ( ; upid >= pid->numbers; --upid) 221 for ( ; upid >= pid->numbers; --upid) {
282 /* Make the PID visible to fin 222 /* Make the PID visible to find_pid_ns. */
283 idr_replace(&upid->ns->idr, pi 223 idr_replace(&upid->ns->idr, pid, upid->nr);
284 upid->ns->pid_allocated++; 224 upid->ns->pid_allocated++;
285 } 225 }
286 spin_unlock_irq(&pidmap_lock); 226 spin_unlock_irq(&pidmap_lock);
287 227
288 return pid; 228 return pid;
289 229
290 out_unlock: 230 out_unlock:
291 spin_unlock_irq(&pidmap_lock); 231 spin_unlock_irq(&pidmap_lock);
292 put_pid_ns(ns); 232 put_pid_ns(ns);
293 233
294 out_free: 234 out_free:
295 spin_lock_irq(&pidmap_lock); 235 spin_lock_irq(&pidmap_lock);
296 while (++i <= ns->level) { 236 while (++i <= ns->level) {
297 upid = pid->numbers + i; 237 upid = pid->numbers + i;
298 idr_remove(&upid->ns->idr, upi 238 idr_remove(&upid->ns->idr, upid->nr);
299 } 239 }
300 240
301 /* On failure to allocate the first pi 241 /* On failure to allocate the first pid, reset the state */
302 if (ns->pid_allocated == PIDNS_ADDING) 242 if (ns->pid_allocated == PIDNS_ADDING)
303 idr_set_cursor(&ns->idr, 0); 243 idr_set_cursor(&ns->idr, 0);
304 244
305 spin_unlock_irq(&pidmap_lock); 245 spin_unlock_irq(&pidmap_lock);
306 246
307 kmem_cache_free(ns->pid_cachep, pid); 247 kmem_cache_free(ns->pid_cachep, pid);
308 return ERR_PTR(retval); 248 return ERR_PTR(retval);
309 } 249 }
310 250
311 void disable_pid_allocation(struct pid_namespa 251 void disable_pid_allocation(struct pid_namespace *ns)
312 { 252 {
313 spin_lock_irq(&pidmap_lock); 253 spin_lock_irq(&pidmap_lock);
314 ns->pid_allocated &= ~PIDNS_ADDING; 254 ns->pid_allocated &= ~PIDNS_ADDING;
315 spin_unlock_irq(&pidmap_lock); 255 spin_unlock_irq(&pidmap_lock);
316 } 256 }
317 257
318 struct pid *find_pid_ns(int nr, struct pid_nam 258 struct pid *find_pid_ns(int nr, struct pid_namespace *ns)
319 { 259 {
320 return idr_find(&ns->idr, nr); 260 return idr_find(&ns->idr, nr);
321 } 261 }
322 EXPORT_SYMBOL_GPL(find_pid_ns); 262 EXPORT_SYMBOL_GPL(find_pid_ns);
323 263
324 struct pid *find_vpid(int nr) 264 struct pid *find_vpid(int nr)
325 { 265 {
326 return find_pid_ns(nr, task_active_pid 266 return find_pid_ns(nr, task_active_pid_ns(current));
327 } 267 }
328 EXPORT_SYMBOL_GPL(find_vpid); 268 EXPORT_SYMBOL_GPL(find_vpid);
329 269
330 static struct pid **task_pid_ptr(struct task_s 270 static struct pid **task_pid_ptr(struct task_struct *task, enum pid_type type)
331 { 271 {
332 return (type == PIDTYPE_PID) ? 272 return (type == PIDTYPE_PID) ?
333 &task->thread_pid : 273 &task->thread_pid :
334 &task->signal->pids[type]; 274 &task->signal->pids[type];
335 } 275 }
336 276
337 /* 277 /*
338 * attach_pid() must be called with the taskli 278 * attach_pid() must be called with the tasklist_lock write-held.
339 */ 279 */
340 void attach_pid(struct task_struct *task, enum 280 void attach_pid(struct task_struct *task, enum pid_type type)
341 { 281 {
342 struct pid *pid = *task_pid_ptr(task, 282 struct pid *pid = *task_pid_ptr(task, type);
343 hlist_add_head_rcu(&task->pid_links[ty 283 hlist_add_head_rcu(&task->pid_links[type], &pid->tasks[type]);
344 } 284 }
345 285
346 static void __change_pid(struct task_struct *t 286 static void __change_pid(struct task_struct *task, enum pid_type type,
347 struct pid *new) 287 struct pid *new)
348 { 288 {
349 struct pid **pid_ptr = task_pid_ptr(ta 289 struct pid **pid_ptr = task_pid_ptr(task, type);
350 struct pid *pid; 290 struct pid *pid;
351 int tmp; 291 int tmp;
352 292
353 pid = *pid_ptr; 293 pid = *pid_ptr;
354 294
355 hlist_del_rcu(&task->pid_links[type]); 295 hlist_del_rcu(&task->pid_links[type]);
356 *pid_ptr = new; 296 *pid_ptr = new;
357 297
358 if (type == PIDTYPE_PID) { <<
359 WARN_ON_ONCE(pid_has_task(pid, <<
360 wake_up_all(&pid->wait_pidfd); <<
361 } <<
362 <<
363 for (tmp = PIDTYPE_MAX; --tmp >= 0; ) 298 for (tmp = PIDTYPE_MAX; --tmp >= 0; )
364 if (pid_has_task(pid, tmp)) !! 299 if (!hlist_empty(&pid->tasks[tmp]))
365 return; 300 return;
366 301
367 free_pid(pid); 302 free_pid(pid);
368 } 303 }
369 304
370 void detach_pid(struct task_struct *task, enum 305 void detach_pid(struct task_struct *task, enum pid_type type)
371 { 306 {
372 __change_pid(task, type, NULL); 307 __change_pid(task, type, NULL);
373 } 308 }
374 309
375 void change_pid(struct task_struct *task, enum 310 void change_pid(struct task_struct *task, enum pid_type type,
376 struct pid *pid) 311 struct pid *pid)
377 { 312 {
378 __change_pid(task, type, pid); 313 __change_pid(task, type, pid);
379 attach_pid(task, type); 314 attach_pid(task, type);
380 } 315 }
381 316
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 317 /* transfer_pid is an optimization of attach_pid(new), detach_pid(old) */
402 void transfer_pid(struct task_struct *old, str 318 void transfer_pid(struct task_struct *old, struct task_struct *new,
403 enum pid_type type) 319 enum pid_type type)
404 { 320 {
405 WARN_ON_ONCE(type == PIDTYPE_PID); !! 321 if (type == PIDTYPE_PID)
>> 322 new->thread_pid = old->thread_pid;
406 hlist_replace_rcu(&old->pid_links[type 323 hlist_replace_rcu(&old->pid_links[type], &new->pid_links[type]);
407 } 324 }
408 325
409 struct task_struct *pid_task(struct pid *pid, 326 struct task_struct *pid_task(struct pid *pid, enum pid_type type)
410 { 327 {
411 struct task_struct *result = NULL; 328 struct task_struct *result = NULL;
412 if (pid) { 329 if (pid) {
413 struct hlist_node *first; 330 struct hlist_node *first;
414 first = rcu_dereference_check( 331 first = rcu_dereference_check(hlist_first_rcu(&pid->tasks[type]),
415 332 lockdep_tasklist_lock_is_held());
416 if (first) 333 if (first)
417 result = hlist_entry(f 334 result = hlist_entry(first, struct task_struct, pid_links[(type)]);
418 } 335 }
419 return result; 336 return result;
420 } 337 }
421 EXPORT_SYMBOL(pid_task); 338 EXPORT_SYMBOL(pid_task);
422 339
423 /* 340 /*
424 * Must be called under rcu_read_lock(). 341 * Must be called under rcu_read_lock().
425 */ 342 */
426 struct task_struct *find_task_by_pid_ns(pid_t 343 struct task_struct *find_task_by_pid_ns(pid_t nr, struct pid_namespace *ns)
427 { 344 {
428 RCU_LOCKDEP_WARN(!rcu_read_lock_held() 345 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
429 "find_task_by_pid_ns( 346 "find_task_by_pid_ns() needs rcu_read_lock() protection");
430 return pid_task(find_pid_ns(nr, ns), P 347 return pid_task(find_pid_ns(nr, ns), PIDTYPE_PID);
431 } 348 }
432 349
433 struct task_struct *find_task_by_vpid(pid_t vn 350 struct task_struct *find_task_by_vpid(pid_t vnr)
434 { 351 {
435 return find_task_by_pid_ns(vnr, task_a 352 return find_task_by_pid_ns(vnr, task_active_pid_ns(current));
436 } 353 }
437 354
438 struct task_struct *find_get_task_by_vpid(pid_ 355 struct task_struct *find_get_task_by_vpid(pid_t nr)
439 { 356 {
440 struct task_struct *task; 357 struct task_struct *task;
441 358
442 rcu_read_lock(); 359 rcu_read_lock();
443 task = find_task_by_vpid(nr); 360 task = find_task_by_vpid(nr);
444 if (task) 361 if (task)
445 get_task_struct(task); 362 get_task_struct(task);
446 rcu_read_unlock(); 363 rcu_read_unlock();
447 364
448 return task; 365 return task;
449 } 366 }
450 367
451 struct pid *get_task_pid(struct task_struct *t 368 struct pid *get_task_pid(struct task_struct *task, enum pid_type type)
452 { 369 {
453 struct pid *pid; 370 struct pid *pid;
454 rcu_read_lock(); 371 rcu_read_lock();
455 pid = get_pid(rcu_dereference(*task_pi 372 pid = get_pid(rcu_dereference(*task_pid_ptr(task, type)));
456 rcu_read_unlock(); 373 rcu_read_unlock();
457 return pid; 374 return pid;
458 } 375 }
459 EXPORT_SYMBOL_GPL(get_task_pid); 376 EXPORT_SYMBOL_GPL(get_task_pid);
460 377
461 struct task_struct *get_pid_task(struct pid *p 378 struct task_struct *get_pid_task(struct pid *pid, enum pid_type type)
462 { 379 {
463 struct task_struct *result; 380 struct task_struct *result;
464 rcu_read_lock(); 381 rcu_read_lock();
465 result = pid_task(pid, type); 382 result = pid_task(pid, type);
466 if (result) 383 if (result)
467 get_task_struct(result); 384 get_task_struct(result);
468 rcu_read_unlock(); 385 rcu_read_unlock();
469 return result; 386 return result;
470 } 387 }
471 EXPORT_SYMBOL_GPL(get_pid_task); 388 EXPORT_SYMBOL_GPL(get_pid_task);
472 389
473 struct pid *find_get_pid(pid_t nr) 390 struct pid *find_get_pid(pid_t nr)
474 { 391 {
475 struct pid *pid; 392 struct pid *pid;
476 393
477 rcu_read_lock(); 394 rcu_read_lock();
478 pid = get_pid(find_vpid(nr)); 395 pid = get_pid(find_vpid(nr));
479 rcu_read_unlock(); 396 rcu_read_unlock();
480 397
481 return pid; 398 return pid;
482 } 399 }
483 EXPORT_SYMBOL_GPL(find_get_pid); 400 EXPORT_SYMBOL_GPL(find_get_pid);
484 401
485 pid_t pid_nr_ns(struct pid *pid, struct pid_na 402 pid_t pid_nr_ns(struct pid *pid, struct pid_namespace *ns)
486 { 403 {
487 struct upid *upid; 404 struct upid *upid;
488 pid_t nr = 0; 405 pid_t nr = 0;
489 406
490 if (pid && ns->level <= pid->level) { 407 if (pid && ns->level <= pid->level) {
491 upid = &pid->numbers[ns->level 408 upid = &pid->numbers[ns->level];
492 if (upid->ns == ns) 409 if (upid->ns == ns)
493 nr = upid->nr; 410 nr = upid->nr;
494 } 411 }
495 return nr; 412 return nr;
496 } 413 }
497 EXPORT_SYMBOL_GPL(pid_nr_ns); 414 EXPORT_SYMBOL_GPL(pid_nr_ns);
498 415
499 pid_t pid_vnr(struct pid *pid) 416 pid_t pid_vnr(struct pid *pid)
500 { 417 {
501 return pid_nr_ns(pid, task_active_pid_ 418 return pid_nr_ns(pid, task_active_pid_ns(current));
502 } 419 }
503 EXPORT_SYMBOL_GPL(pid_vnr); 420 EXPORT_SYMBOL_GPL(pid_vnr);
504 421
505 pid_t __task_pid_nr_ns(struct task_struct *tas 422 pid_t __task_pid_nr_ns(struct task_struct *task, enum pid_type type,
506 struct pid_namespace * 423 struct pid_namespace *ns)
507 { 424 {
508 pid_t nr = 0; 425 pid_t nr = 0;
509 426
510 rcu_read_lock(); 427 rcu_read_lock();
511 if (!ns) 428 if (!ns)
512 ns = task_active_pid_ns(curren 429 ns = task_active_pid_ns(current);
513 nr = pid_nr_ns(rcu_dereference(*task_p !! 430 if (likely(pid_alive(task)))
>> 431 nr = pid_nr_ns(rcu_dereference(*task_pid_ptr(task, type)), ns);
514 rcu_read_unlock(); 432 rcu_read_unlock();
515 433
516 return nr; 434 return nr;
517 } 435 }
518 EXPORT_SYMBOL(__task_pid_nr_ns); 436 EXPORT_SYMBOL(__task_pid_nr_ns);
519 437
520 struct pid_namespace *task_active_pid_ns(struc 438 struct pid_namespace *task_active_pid_ns(struct task_struct *tsk)
521 { 439 {
522 return ns_of_pid(task_pid(tsk)); 440 return ns_of_pid(task_pid(tsk));
523 } 441 }
524 EXPORT_SYMBOL_GPL(task_active_pid_ns); 442 EXPORT_SYMBOL_GPL(task_active_pid_ns);
525 443
526 /* 444 /*
527 * Used by proc to find the first pid that is 445 * Used by proc to find the first pid that is greater than or equal to nr.
528 * 446 *
529 * If there is a pid at nr this function is ex 447 * If there is a pid at nr this function is exactly the same as find_pid_ns.
530 */ 448 */
531 struct pid *find_ge_pid(int nr, struct pid_nam 449 struct pid *find_ge_pid(int nr, struct pid_namespace *ns)
532 { 450 {
533 return idr_get_next(&ns->idr, &nr); 451 return idr_get_next(&ns->idr, &nr);
534 } 452 }
535 EXPORT_SYMBOL_GPL(find_ge_pid); <<
536 <<
537 struct pid *pidfd_get_pid(unsigned int fd, uns <<
538 { <<
539 struct fd f; <<
540 struct pid *pid; <<
541 <<
542 f = fdget(fd); <<
543 if (!fd_file(f)) <<
544 return ERR_PTR(-EBADF); <<
545 <<
546 pid = pidfd_pid(fd_file(f)); <<
547 if (!IS_ERR(pid)) { <<
548 get_pid(pid); <<
549 *flags = fd_file(f)->f_flags; <<
550 } <<
551 <<
552 fdput(f); <<
553 return pid; <<
554 } <<
555 <<
556 /** <<
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 { <<
570 unsigned int f_flags; <<
571 struct pid *pid; <<
572 struct task_struct *task; <<
573 <<
574 pid = pidfd_get_pid(pidfd, &f_flags); <<
575 if (IS_ERR(pid)) <<
576 return ERR_CAST(pid); <<
577 <<
578 task = get_pid_task(pid, PIDTYPE_TGID) <<
579 put_pid(pid); <<
580 if (!task) <<
581 return ERR_PTR(-ESRCH); <<
582 <<
583 *flags = f_flags; <<
584 return task; <<
585 } <<
586 <<
587 /** <<
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 { <<
605 int pidfd; <<
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 } <<
615 <<
616 /** <<
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 { <<
631 int fd; <<
632 struct pid *p; <<
633 <<
634 if (flags & ~(PIDFD_NONBLOCK | PIDFD_T <<
635 return -EINVAL; <<
636 <<
637 if (pid <= 0) <<
638 return -EINVAL; <<
639 <<
640 p = find_get_pid(pid); <<
641 if (!p) <<
642 return -ESRCH; <<
643 <<
644 fd = pidfd_create(p, flags); <<
645 <<
646 put_pid(p); <<
647 return fd; <<
648 } <<
649 453
650 void __init pid_idr_init(void) 454 void __init pid_idr_init(void)
651 { 455 {
652 /* Verify no one has done anything sil 456 /* Verify no one has done anything silly: */
653 BUILD_BUG_ON(PID_MAX_LIMIT >= PIDNS_AD 457 BUILD_BUG_ON(PID_MAX_LIMIT >= PIDNS_ADDING);
654 458
655 /* bump default and minimum pid_max ba 459 /* bump default and minimum pid_max based on number of cpus */
656 pid_max = min(pid_max_max, max_t(int, 460 pid_max = min(pid_max_max, max_t(int, pid_max,
657 PIDS_PER_CPU_D 461 PIDS_PER_CPU_DEFAULT * num_possible_cpus()));
658 pid_max_min = max_t(int, pid_max_min, 462 pid_max_min = max_t(int, pid_max_min,
659 PIDS_PER_CPU_M 463 PIDS_PER_CPU_MIN * num_possible_cpus());
660 pr_info("pid_max: default: %u minimum: 464 pr_info("pid_max: default: %u minimum: %u\n", pid_max, pid_max_min);
661 465
662 idr_init(&init_pid_ns.idr); 466 idr_init(&init_pid_ns.idr);
663 467
664 init_pid_ns.pid_cachep = kmem_cache_cr !! 468 init_pid_ns.pid_cachep = KMEM_CACHE(pid,
665 struct_size_t(struct p !! 469 SLAB_HWCACHE_ALIGN | SLAB_PANIC | SLAB_ACCOUNT);
666 __alignof__(struct pid <<
667 SLAB_HWCACHE_ALIGN | S <<
668 NULL); <<
669 } <<
670 <<
671 static struct file *__pidfd_fget(struct task_s <<
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 <<
680 if (ptrace_may_access(task, PTRACE_MOD <<
681 file = fget_task(task, fd); <<
682 else <<
683 file = ERR_PTR(-EPERM); <<
684 <<
685 up_read(&task->signal->exec_update_loc <<
686 <<
687 if (!file) { <<
688 /* <<
689 * It is possible that the tar <<
690 * either: <<
691 * 1. before exit_signals(), w <<
692 * 2. before exit_files() take <<
693 * 3. after exit_files() relea <<
694 * this has PF_EXITING, sin <<
695 * __pidfd_fget() returns E <<
696 * In case 3 we get EBADF, but <<
697 * the task is currently exiti <<
698 * struct, so we fix it up. <<
699 */ <<
700 if (task->flags & PF_EXITING) <<
701 file = ERR_PTR(-ESRCH) <<
702 else <<
703 file = ERR_PTR(-EBADF) <<
704 } <<
705 <<
706 return file; <<
707 } <<
708 <<
709 static int pidfd_getfd(struct pid *pid, int fd <<
710 { <<
711 struct task_struct *task; <<
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 <<
730 /** <<
731 * sys_pidfd_getfd() - Get a file descriptor f <<
732 * <<
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 <<
753 /* flags is currently unused - make su <<
754 if (flags) <<
755 return -EINVAL; <<
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 <<
767 fdput(f); <<
768 return ret; <<
769 } 470 }
770 471
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.