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