xref: /openbmc/linux/ipc/sem.c (revision 75f25bd3)
1 /*
2  * linux/ipc/sem.c
3  * Copyright (C) 1992 Krishna Balasubramanian
4  * Copyright (C) 1995 Eric Schenk, Bruno Haible
5  *
6  * /proc/sysvipc/sem support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
7  *
8  * SMP-threaded, sysctl's added
9  * (c) 1999 Manfred Spraul <manfred@colorfullife.com>
10  * Enforced range limit on SEM_UNDO
11  * (c) 2001 Red Hat Inc
12  * Lockless wakeup
13  * (c) 2003 Manfred Spraul <manfred@colorfullife.com>
14  * Further wakeup optimizations, documentation
15  * (c) 2010 Manfred Spraul <manfred@colorfullife.com>
16  *
17  * support for audit of ipc object properties and permission changes
18  * Dustin Kirkland <dustin.kirkland@us.ibm.com>
19  *
20  * namespaces support
21  * OpenVZ, SWsoft Inc.
22  * Pavel Emelianov <xemul@openvz.org>
23  *
24  * Implementation notes: (May 2010)
25  * This file implements System V semaphores.
26  *
27  * User space visible behavior:
28  * - FIFO ordering for semop() operations (just FIFO, not starvation
29  *   protection)
30  * - multiple semaphore operations that alter the same semaphore in
31  *   one semop() are handled.
32  * - sem_ctime (time of last semctl()) is updated in the IPC_SET, SETVAL and
33  *   SETALL calls.
34  * - two Linux specific semctl() commands: SEM_STAT, SEM_INFO.
35  * - undo adjustments at process exit are limited to 0..SEMVMX.
36  * - namespace are supported.
37  * - SEMMSL, SEMMNS, SEMOPM and SEMMNI can be configured at runtine by writing
38  *   to /proc/sys/kernel/sem.
39  * - statistics about the usage are reported in /proc/sysvipc/sem.
40  *
41  * Internals:
42  * - scalability:
43  *   - all global variables are read-mostly.
44  *   - semop() calls and semctl(RMID) are synchronized by RCU.
45  *   - most operations do write operations (actually: spin_lock calls) to
46  *     the per-semaphore array structure.
47  *   Thus: Perfect SMP scaling between independent semaphore arrays.
48  *         If multiple semaphores in one array are used, then cache line
49  *         trashing on the semaphore array spinlock will limit the scaling.
50  * - semncnt and semzcnt are calculated on demand in count_semncnt() and
51  *   count_semzcnt()
52  * - the task that performs a successful semop() scans the list of all
53  *   sleeping tasks and completes any pending operations that can be fulfilled.
54  *   Semaphores are actively given to waiting tasks (necessary for FIFO).
55  *   (see update_queue())
56  * - To improve the scalability, the actual wake-up calls are performed after
57  *   dropping all locks. (see wake_up_sem_queue_prepare(),
58  *   wake_up_sem_queue_do())
59  * - All work is done by the waker, the woken up task does not have to do
60  *   anything - not even acquiring a lock or dropping a refcount.
61  * - A woken up task may not even touch the semaphore array anymore, it may
62  *   have been destroyed already by a semctl(RMID).
63  * - The synchronizations between wake-ups due to a timeout/signal and a
64  *   wake-up due to a completed semaphore operation is achieved by using an
65  *   intermediate state (IN_WAKEUP).
66  * - UNDO values are stored in an array (one per process and per
67  *   semaphore array, lazily allocated). For backwards compatibility, multiple
68  *   modes for the UNDO variables are supported (per process, per thread)
69  *   (see copy_semundo, CLONE_SYSVSEM)
70  * - There are two lists of the pending operations: a per-array list
71  *   and per-semaphore list (stored in the array). This allows to achieve FIFO
72  *   ordering without always scanning all pending operations.
73  *   The worst-case behavior is nevertheless O(N^2) for N wakeups.
74  */
75 
76 #include <linux/slab.h>
77 #include <linux/spinlock.h>
78 #include <linux/init.h>
79 #include <linux/proc_fs.h>
80 #include <linux/time.h>
81 #include <linux/security.h>
82 #include <linux/syscalls.h>
83 #include <linux/audit.h>
84 #include <linux/capability.h>
85 #include <linux/seq_file.h>
86 #include <linux/rwsem.h>
87 #include <linux/nsproxy.h>
88 #include <linux/ipc_namespace.h>
89 
90 #include <asm/uaccess.h>
91 #include "util.h"
92 
93 #define sem_ids(ns)	((ns)->ids[IPC_SEM_IDS])
94 
95 #define sem_unlock(sma)		ipc_unlock(&(sma)->sem_perm)
96 #define sem_checkid(sma, semid)	ipc_checkid(&sma->sem_perm, semid)
97 
98 static int newary(struct ipc_namespace *, struct ipc_params *);
99 static void freeary(struct ipc_namespace *, struct kern_ipc_perm *);
100 #ifdef CONFIG_PROC_FS
101 static int sysvipc_sem_proc_show(struct seq_file *s, void *it);
102 #endif
103 
104 #define SEMMSL_FAST	256 /* 512 bytes on stack */
105 #define SEMOPM_FAST	64  /* ~ 372 bytes on stack */
106 
107 /*
108  * linked list protection:
109  *	sem_undo.id_next,
110  *	sem_array.sem_pending{,last},
111  *	sem_array.sem_undo: sem_lock() for read/write
112  *	sem_undo.proc_next: only "current" is allowed to read/write that field.
113  *
114  */
115 
116 #define sc_semmsl	sem_ctls[0]
117 #define sc_semmns	sem_ctls[1]
118 #define sc_semopm	sem_ctls[2]
119 #define sc_semmni	sem_ctls[3]
120 
121 void sem_init_ns(struct ipc_namespace *ns)
122 {
123 	ns->sc_semmsl = SEMMSL;
124 	ns->sc_semmns = SEMMNS;
125 	ns->sc_semopm = SEMOPM;
126 	ns->sc_semmni = SEMMNI;
127 	ns->used_sems = 0;
128 	ipc_init_ids(&ns->ids[IPC_SEM_IDS]);
129 }
130 
131 #ifdef CONFIG_IPC_NS
132 void sem_exit_ns(struct ipc_namespace *ns)
133 {
134 	free_ipcs(ns, &sem_ids(ns), freeary);
135 	idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr);
136 }
137 #endif
138 
139 void __init sem_init (void)
140 {
141 	sem_init_ns(&init_ipc_ns);
142 	ipc_init_proc_interface("sysvipc/sem",
143 				"       key      semid perms      nsems   uid   gid  cuid  cgid      otime      ctime\n",
144 				IPC_SEM_IDS, sysvipc_sem_proc_show);
145 }
146 
147 /*
148  * sem_lock_(check_) routines are called in the paths where the rw_mutex
149  * is not held.
150  */
151 static inline struct sem_array *sem_lock(struct ipc_namespace *ns, int id)
152 {
153 	struct kern_ipc_perm *ipcp = ipc_lock(&sem_ids(ns), id);
154 
155 	if (IS_ERR(ipcp))
156 		return (struct sem_array *)ipcp;
157 
158 	return container_of(ipcp, struct sem_array, sem_perm);
159 }
160 
161 static inline struct sem_array *sem_lock_check(struct ipc_namespace *ns,
162 						int id)
163 {
164 	struct kern_ipc_perm *ipcp = ipc_lock_check(&sem_ids(ns), id);
165 
166 	if (IS_ERR(ipcp))
167 		return (struct sem_array *)ipcp;
168 
169 	return container_of(ipcp, struct sem_array, sem_perm);
170 }
171 
172 static inline void sem_lock_and_putref(struct sem_array *sma)
173 {
174 	ipc_lock_by_ptr(&sma->sem_perm);
175 	ipc_rcu_putref(sma);
176 }
177 
178 static inline void sem_getref_and_unlock(struct sem_array *sma)
179 {
180 	ipc_rcu_getref(sma);
181 	ipc_unlock(&(sma)->sem_perm);
182 }
183 
184 static inline void sem_putref(struct sem_array *sma)
185 {
186 	ipc_lock_by_ptr(&sma->sem_perm);
187 	ipc_rcu_putref(sma);
188 	ipc_unlock(&(sma)->sem_perm);
189 }
190 
191 static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
192 {
193 	ipc_rmid(&sem_ids(ns), &s->sem_perm);
194 }
195 
196 /*
197  * Lockless wakeup algorithm:
198  * Without the check/retry algorithm a lockless wakeup is possible:
199  * - queue.status is initialized to -EINTR before blocking.
200  * - wakeup is performed by
201  *	* unlinking the queue entry from sma->sem_pending
202  *	* setting queue.status to IN_WAKEUP
203  *	  This is the notification for the blocked thread that a
204  *	  result value is imminent.
205  *	* call wake_up_process
206  *	* set queue.status to the final value.
207  * - the previously blocked thread checks queue.status:
208  *   	* if it's IN_WAKEUP, then it must wait until the value changes
209  *   	* if it's not -EINTR, then the operation was completed by
210  *   	  update_queue. semtimedop can return queue.status without
211  *   	  performing any operation on the sem array.
212  *   	* otherwise it must acquire the spinlock and check what's up.
213  *
214  * The two-stage algorithm is necessary to protect against the following
215  * races:
216  * - if queue.status is set after wake_up_process, then the woken up idle
217  *   thread could race forward and try (and fail) to acquire sma->lock
218  *   before update_queue had a chance to set queue.status
219  * - if queue.status is written before wake_up_process and if the
220  *   blocked process is woken up by a signal between writing
221  *   queue.status and the wake_up_process, then the woken up
222  *   process could return from semtimedop and die by calling
223  *   sys_exit before wake_up_process is called. Then wake_up_process
224  *   will oops, because the task structure is already invalid.
225  *   (yes, this happened on s390 with sysv msg).
226  *
227  */
228 #define IN_WAKEUP	1
229 
230 /**
231  * newary - Create a new semaphore set
232  * @ns: namespace
233  * @params: ptr to the structure that contains key, semflg and nsems
234  *
235  * Called with sem_ids.rw_mutex held (as a writer)
236  */
237 
238 static int newary(struct ipc_namespace *ns, struct ipc_params *params)
239 {
240 	int id;
241 	int retval;
242 	struct sem_array *sma;
243 	int size;
244 	key_t key = params->key;
245 	int nsems = params->u.nsems;
246 	int semflg = params->flg;
247 	int i;
248 
249 	if (!nsems)
250 		return -EINVAL;
251 	if (ns->used_sems + nsems > ns->sc_semmns)
252 		return -ENOSPC;
253 
254 	size = sizeof (*sma) + nsems * sizeof (struct sem);
255 	sma = ipc_rcu_alloc(size);
256 	if (!sma) {
257 		return -ENOMEM;
258 	}
259 	memset (sma, 0, size);
260 
261 	sma->sem_perm.mode = (semflg & S_IRWXUGO);
262 	sma->sem_perm.key = key;
263 
264 	sma->sem_perm.security = NULL;
265 	retval = security_sem_alloc(sma);
266 	if (retval) {
267 		ipc_rcu_putref(sma);
268 		return retval;
269 	}
270 
271 	id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
272 	if (id < 0) {
273 		security_sem_free(sma);
274 		ipc_rcu_putref(sma);
275 		return id;
276 	}
277 	ns->used_sems += nsems;
278 
279 	sma->sem_base = (struct sem *) &sma[1];
280 
281 	for (i = 0; i < nsems; i++)
282 		INIT_LIST_HEAD(&sma->sem_base[i].sem_pending);
283 
284 	sma->complex_count = 0;
285 	INIT_LIST_HEAD(&sma->sem_pending);
286 	INIT_LIST_HEAD(&sma->list_id);
287 	sma->sem_nsems = nsems;
288 	sma->sem_ctime = get_seconds();
289 	sem_unlock(sma);
290 
291 	return sma->sem_perm.id;
292 }
293 
294 
295 /*
296  * Called with sem_ids.rw_mutex and ipcp locked.
297  */
298 static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg)
299 {
300 	struct sem_array *sma;
301 
302 	sma = container_of(ipcp, struct sem_array, sem_perm);
303 	return security_sem_associate(sma, semflg);
304 }
305 
306 /*
307  * Called with sem_ids.rw_mutex and ipcp locked.
308  */
309 static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
310 				struct ipc_params *params)
311 {
312 	struct sem_array *sma;
313 
314 	sma = container_of(ipcp, struct sem_array, sem_perm);
315 	if (params->u.nsems > sma->sem_nsems)
316 		return -EINVAL;
317 
318 	return 0;
319 }
320 
321 SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg)
322 {
323 	struct ipc_namespace *ns;
324 	struct ipc_ops sem_ops;
325 	struct ipc_params sem_params;
326 
327 	ns = current->nsproxy->ipc_ns;
328 
329 	if (nsems < 0 || nsems > ns->sc_semmsl)
330 		return -EINVAL;
331 
332 	sem_ops.getnew = newary;
333 	sem_ops.associate = sem_security;
334 	sem_ops.more_checks = sem_more_checks;
335 
336 	sem_params.key = key;
337 	sem_params.flg = semflg;
338 	sem_params.u.nsems = nsems;
339 
340 	return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
341 }
342 
343 /*
344  * Determine whether a sequence of semaphore operations would succeed
345  * all at once. Return 0 if yes, 1 if need to sleep, else return error code.
346  */
347 
348 static int try_atomic_semop (struct sem_array * sma, struct sembuf * sops,
349 			     int nsops, struct sem_undo *un, int pid)
350 {
351 	int result, sem_op;
352 	struct sembuf *sop;
353 	struct sem * curr;
354 
355 	for (sop = sops; sop < sops + nsops; sop++) {
356 		curr = sma->sem_base + sop->sem_num;
357 		sem_op = sop->sem_op;
358 		result = curr->semval;
359 
360 		if (!sem_op && result)
361 			goto would_block;
362 
363 		result += sem_op;
364 		if (result < 0)
365 			goto would_block;
366 		if (result > SEMVMX)
367 			goto out_of_range;
368 		if (sop->sem_flg & SEM_UNDO) {
369 			int undo = un->semadj[sop->sem_num] - sem_op;
370 			/*
371 	 		 *	Exceeding the undo range is an error.
372 			 */
373 			if (undo < (-SEMAEM - 1) || undo > SEMAEM)
374 				goto out_of_range;
375 		}
376 		curr->semval = result;
377 	}
378 
379 	sop--;
380 	while (sop >= sops) {
381 		sma->sem_base[sop->sem_num].sempid = pid;
382 		if (sop->sem_flg & SEM_UNDO)
383 			un->semadj[sop->sem_num] -= sop->sem_op;
384 		sop--;
385 	}
386 
387 	return 0;
388 
389 out_of_range:
390 	result = -ERANGE;
391 	goto undo;
392 
393 would_block:
394 	if (sop->sem_flg & IPC_NOWAIT)
395 		result = -EAGAIN;
396 	else
397 		result = 1;
398 
399 undo:
400 	sop--;
401 	while (sop >= sops) {
402 		sma->sem_base[sop->sem_num].semval -= sop->sem_op;
403 		sop--;
404 	}
405 
406 	return result;
407 }
408 
409 /** wake_up_sem_queue_prepare(q, error): Prepare wake-up
410  * @q: queue entry that must be signaled
411  * @error: Error value for the signal
412  *
413  * Prepare the wake-up of the queue entry q.
414  */
415 static void wake_up_sem_queue_prepare(struct list_head *pt,
416 				struct sem_queue *q, int error)
417 {
418 	if (list_empty(pt)) {
419 		/*
420 		 * Hold preempt off so that we don't get preempted and have the
421 		 * wakee busy-wait until we're scheduled back on.
422 		 */
423 		preempt_disable();
424 	}
425 	q->status = IN_WAKEUP;
426 	q->pid = error;
427 
428 	list_add_tail(&q->simple_list, pt);
429 }
430 
431 /**
432  * wake_up_sem_queue_do(pt) - do the actual wake-up
433  * @pt: list of tasks to be woken up
434  *
435  * Do the actual wake-up.
436  * The function is called without any locks held, thus the semaphore array
437  * could be destroyed already and the tasks can disappear as soon as the
438  * status is set to the actual return code.
439  */
440 static void wake_up_sem_queue_do(struct list_head *pt)
441 {
442 	struct sem_queue *q, *t;
443 	int did_something;
444 
445 	did_something = !list_empty(pt);
446 	list_for_each_entry_safe(q, t, pt, simple_list) {
447 		wake_up_process(q->sleeper);
448 		/* q can disappear immediately after writing q->status. */
449 		smp_wmb();
450 		q->status = q->pid;
451 	}
452 	if (did_something)
453 		preempt_enable();
454 }
455 
456 static void unlink_queue(struct sem_array *sma, struct sem_queue *q)
457 {
458 	list_del(&q->list);
459 	if (q->nsops == 1)
460 		list_del(&q->simple_list);
461 	else
462 		sma->complex_count--;
463 }
464 
465 /** check_restart(sma, q)
466  * @sma: semaphore array
467  * @q: the operation that just completed
468  *
469  * update_queue is O(N^2) when it restarts scanning the whole queue of
470  * waiting operations. Therefore this function checks if the restart is
471  * really necessary. It is called after a previously waiting operation
472  * was completed.
473  */
474 static int check_restart(struct sem_array *sma, struct sem_queue *q)
475 {
476 	struct sem *curr;
477 	struct sem_queue *h;
478 
479 	/* if the operation didn't modify the array, then no restart */
480 	if (q->alter == 0)
481 		return 0;
482 
483 	/* pending complex operations are too difficult to analyse */
484 	if (sma->complex_count)
485 		return 1;
486 
487 	/* we were a sleeping complex operation. Too difficult */
488 	if (q->nsops > 1)
489 		return 1;
490 
491 	curr = sma->sem_base + q->sops[0].sem_num;
492 
493 	/* No-one waits on this queue */
494 	if (list_empty(&curr->sem_pending))
495 		return 0;
496 
497 	/* the new semaphore value */
498 	if (curr->semval) {
499 		/* It is impossible that someone waits for the new value:
500 		 * - q is a previously sleeping simple operation that
501 		 *   altered the array. It must be a decrement, because
502 		 *   simple increments never sleep.
503 		 * - The value is not 0, thus wait-for-zero won't proceed.
504 		 * - If there are older (higher priority) decrements
505 		 *   in the queue, then they have observed the original
506 		 *   semval value and couldn't proceed. The operation
507 		 *   decremented to value - thus they won't proceed either.
508 		 */
509 		BUG_ON(q->sops[0].sem_op >= 0);
510 		return 0;
511 	}
512 	/*
513 	 * semval is 0. Check if there are wait-for-zero semops.
514 	 * They must be the first entries in the per-semaphore simple queue
515 	 */
516 	h = list_first_entry(&curr->sem_pending, struct sem_queue, simple_list);
517 	BUG_ON(h->nsops != 1);
518 	BUG_ON(h->sops[0].sem_num != q->sops[0].sem_num);
519 
520 	/* Yes, there is a wait-for-zero semop. Restart */
521 	if (h->sops[0].sem_op == 0)
522 		return 1;
523 
524 	/* Again - no-one is waiting for the new value. */
525 	return 0;
526 }
527 
528 
529 /**
530  * update_queue(sma, semnum): Look for tasks that can be completed.
531  * @sma: semaphore array.
532  * @semnum: semaphore that was modified.
533  * @pt: list head for the tasks that must be woken up.
534  *
535  * update_queue must be called after a semaphore in a semaphore array
536  * was modified. If multiple semaphore were modified, then @semnum
537  * must be set to -1.
538  * The tasks that must be woken up are added to @pt. The return code
539  * is stored in q->pid.
540  * The function return 1 if at least one semop was completed successfully.
541  */
542 static int update_queue(struct sem_array *sma, int semnum, struct list_head *pt)
543 {
544 	struct sem_queue *q;
545 	struct list_head *walk;
546 	struct list_head *pending_list;
547 	int offset;
548 	int semop_completed = 0;
549 
550 	/* if there are complex operations around, then knowing the semaphore
551 	 * that was modified doesn't help us. Assume that multiple semaphores
552 	 * were modified.
553 	 */
554 	if (sma->complex_count)
555 		semnum = -1;
556 
557 	if (semnum == -1) {
558 		pending_list = &sma->sem_pending;
559 		offset = offsetof(struct sem_queue, list);
560 	} else {
561 		pending_list = &sma->sem_base[semnum].sem_pending;
562 		offset = offsetof(struct sem_queue, simple_list);
563 	}
564 
565 again:
566 	walk = pending_list->next;
567 	while (walk != pending_list) {
568 		int error, restart;
569 
570 		q = (struct sem_queue *)((char *)walk - offset);
571 		walk = walk->next;
572 
573 		/* If we are scanning the single sop, per-semaphore list of
574 		 * one semaphore and that semaphore is 0, then it is not
575 		 * necessary to scan the "alter" entries: simple increments
576 		 * that affect only one entry succeed immediately and cannot
577 		 * be in the  per semaphore pending queue, and decrements
578 		 * cannot be successful if the value is already 0.
579 		 */
580 		if (semnum != -1 && sma->sem_base[semnum].semval == 0 &&
581 				q->alter)
582 			break;
583 
584 		error = try_atomic_semop(sma, q->sops, q->nsops,
585 					 q->undo, q->pid);
586 
587 		/* Does q->sleeper still need to sleep? */
588 		if (error > 0)
589 			continue;
590 
591 		unlink_queue(sma, q);
592 
593 		if (error) {
594 			restart = 0;
595 		} else {
596 			semop_completed = 1;
597 			restart = check_restart(sma, q);
598 		}
599 
600 		wake_up_sem_queue_prepare(pt, q, error);
601 		if (restart)
602 			goto again;
603 	}
604 	return semop_completed;
605 }
606 
607 /**
608  * do_smart_update(sma, sops, nsops, otime, pt) - optimized update_queue
609  * @sma: semaphore array
610  * @sops: operations that were performed
611  * @nsops: number of operations
612  * @otime: force setting otime
613  * @pt: list head of the tasks that must be woken up.
614  *
615  * do_smart_update() does the required called to update_queue, based on the
616  * actual changes that were performed on the semaphore array.
617  * Note that the function does not do the actual wake-up: the caller is
618  * responsible for calling wake_up_sem_queue_do(@pt).
619  * It is safe to perform this call after dropping all locks.
620  */
621 static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops,
622 			int otime, struct list_head *pt)
623 {
624 	int i;
625 
626 	if (sma->complex_count || sops == NULL) {
627 		if (update_queue(sma, -1, pt))
628 			otime = 1;
629 		goto done;
630 	}
631 
632 	for (i = 0; i < nsops; i++) {
633 		if (sops[i].sem_op > 0 ||
634 			(sops[i].sem_op < 0 &&
635 				sma->sem_base[sops[i].sem_num].semval == 0))
636 			if (update_queue(sma, sops[i].sem_num, pt))
637 				otime = 1;
638 	}
639 done:
640 	if (otime)
641 		sma->sem_otime = get_seconds();
642 }
643 
644 
645 /* The following counts are associated to each semaphore:
646  *   semncnt        number of tasks waiting on semval being nonzero
647  *   semzcnt        number of tasks waiting on semval being zero
648  * This model assumes that a task waits on exactly one semaphore.
649  * Since semaphore operations are to be performed atomically, tasks actually
650  * wait on a whole sequence of semaphores simultaneously.
651  * The counts we return here are a rough approximation, but still
652  * warrant that semncnt+semzcnt>0 if the task is on the pending queue.
653  */
654 static int count_semncnt (struct sem_array * sma, ushort semnum)
655 {
656 	int semncnt;
657 	struct sem_queue * q;
658 
659 	semncnt = 0;
660 	list_for_each_entry(q, &sma->sem_pending, list) {
661 		struct sembuf * sops = q->sops;
662 		int nsops = q->nsops;
663 		int i;
664 		for (i = 0; i < nsops; i++)
665 			if (sops[i].sem_num == semnum
666 			    && (sops[i].sem_op < 0)
667 			    && !(sops[i].sem_flg & IPC_NOWAIT))
668 				semncnt++;
669 	}
670 	return semncnt;
671 }
672 
673 static int count_semzcnt (struct sem_array * sma, ushort semnum)
674 {
675 	int semzcnt;
676 	struct sem_queue * q;
677 
678 	semzcnt = 0;
679 	list_for_each_entry(q, &sma->sem_pending, list) {
680 		struct sembuf * sops = q->sops;
681 		int nsops = q->nsops;
682 		int i;
683 		for (i = 0; i < nsops; i++)
684 			if (sops[i].sem_num == semnum
685 			    && (sops[i].sem_op == 0)
686 			    && !(sops[i].sem_flg & IPC_NOWAIT))
687 				semzcnt++;
688 	}
689 	return semzcnt;
690 }
691 
692 /* Free a semaphore set. freeary() is called with sem_ids.rw_mutex locked
693  * as a writer and the spinlock for this semaphore set hold. sem_ids.rw_mutex
694  * remains locked on exit.
695  */
696 static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
697 {
698 	struct sem_undo *un, *tu;
699 	struct sem_queue *q, *tq;
700 	struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
701 	struct list_head tasks;
702 
703 	/* Free the existing undo structures for this semaphore set.  */
704 	assert_spin_locked(&sma->sem_perm.lock);
705 	list_for_each_entry_safe(un, tu, &sma->list_id, list_id) {
706 		list_del(&un->list_id);
707 		spin_lock(&un->ulp->lock);
708 		un->semid = -1;
709 		list_del_rcu(&un->list_proc);
710 		spin_unlock(&un->ulp->lock);
711 		kfree_rcu(un, rcu);
712 	}
713 
714 	/* Wake up all pending processes and let them fail with EIDRM. */
715 	INIT_LIST_HEAD(&tasks);
716 	list_for_each_entry_safe(q, tq, &sma->sem_pending, list) {
717 		unlink_queue(sma, q);
718 		wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
719 	}
720 
721 	/* Remove the semaphore set from the IDR */
722 	sem_rmid(ns, sma);
723 	sem_unlock(sma);
724 
725 	wake_up_sem_queue_do(&tasks);
726 	ns->used_sems -= sma->sem_nsems;
727 	security_sem_free(sma);
728 	ipc_rcu_putref(sma);
729 }
730 
731 static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
732 {
733 	switch(version) {
734 	case IPC_64:
735 		return copy_to_user(buf, in, sizeof(*in));
736 	case IPC_OLD:
737 	    {
738 		struct semid_ds out;
739 
740 		memset(&out, 0, sizeof(out));
741 
742 		ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
743 
744 		out.sem_otime	= in->sem_otime;
745 		out.sem_ctime	= in->sem_ctime;
746 		out.sem_nsems	= in->sem_nsems;
747 
748 		return copy_to_user(buf, &out, sizeof(out));
749 	    }
750 	default:
751 		return -EINVAL;
752 	}
753 }
754 
755 static int semctl_nolock(struct ipc_namespace *ns, int semid,
756 			 int cmd, int version, union semun arg)
757 {
758 	int err;
759 	struct sem_array *sma;
760 
761 	switch(cmd) {
762 	case IPC_INFO:
763 	case SEM_INFO:
764 	{
765 		struct seminfo seminfo;
766 		int max_id;
767 
768 		err = security_sem_semctl(NULL, cmd);
769 		if (err)
770 			return err;
771 
772 		memset(&seminfo,0,sizeof(seminfo));
773 		seminfo.semmni = ns->sc_semmni;
774 		seminfo.semmns = ns->sc_semmns;
775 		seminfo.semmsl = ns->sc_semmsl;
776 		seminfo.semopm = ns->sc_semopm;
777 		seminfo.semvmx = SEMVMX;
778 		seminfo.semmnu = SEMMNU;
779 		seminfo.semmap = SEMMAP;
780 		seminfo.semume = SEMUME;
781 		down_read(&sem_ids(ns).rw_mutex);
782 		if (cmd == SEM_INFO) {
783 			seminfo.semusz = sem_ids(ns).in_use;
784 			seminfo.semaem = ns->used_sems;
785 		} else {
786 			seminfo.semusz = SEMUSZ;
787 			seminfo.semaem = SEMAEM;
788 		}
789 		max_id = ipc_get_maxid(&sem_ids(ns));
790 		up_read(&sem_ids(ns).rw_mutex);
791 		if (copy_to_user (arg.__buf, &seminfo, sizeof(struct seminfo)))
792 			return -EFAULT;
793 		return (max_id < 0) ? 0: max_id;
794 	}
795 	case IPC_STAT:
796 	case SEM_STAT:
797 	{
798 		struct semid64_ds tbuf;
799 		int id;
800 
801 		if (cmd == SEM_STAT) {
802 			sma = sem_lock(ns, semid);
803 			if (IS_ERR(sma))
804 				return PTR_ERR(sma);
805 			id = sma->sem_perm.id;
806 		} else {
807 			sma = sem_lock_check(ns, semid);
808 			if (IS_ERR(sma))
809 				return PTR_ERR(sma);
810 			id = 0;
811 		}
812 
813 		err = -EACCES;
814 		if (ipcperms(ns, &sma->sem_perm, S_IRUGO))
815 			goto out_unlock;
816 
817 		err = security_sem_semctl(sma, cmd);
818 		if (err)
819 			goto out_unlock;
820 
821 		memset(&tbuf, 0, sizeof(tbuf));
822 
823 		kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
824 		tbuf.sem_otime  = sma->sem_otime;
825 		tbuf.sem_ctime  = sma->sem_ctime;
826 		tbuf.sem_nsems  = sma->sem_nsems;
827 		sem_unlock(sma);
828 		if (copy_semid_to_user (arg.buf, &tbuf, version))
829 			return -EFAULT;
830 		return id;
831 	}
832 	default:
833 		return -EINVAL;
834 	}
835 out_unlock:
836 	sem_unlock(sma);
837 	return err;
838 }
839 
840 static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
841 		int cmd, int version, union semun arg)
842 {
843 	struct sem_array *sma;
844 	struct sem* curr;
845 	int err;
846 	ushort fast_sem_io[SEMMSL_FAST];
847 	ushort* sem_io = fast_sem_io;
848 	int nsems;
849 	struct list_head tasks;
850 
851 	sma = sem_lock_check(ns, semid);
852 	if (IS_ERR(sma))
853 		return PTR_ERR(sma);
854 
855 	INIT_LIST_HEAD(&tasks);
856 	nsems = sma->sem_nsems;
857 
858 	err = -EACCES;
859 	if (ipcperms(ns, &sma->sem_perm,
860 			(cmd == SETVAL || cmd == SETALL) ? S_IWUGO : S_IRUGO))
861 		goto out_unlock;
862 
863 	err = security_sem_semctl(sma, cmd);
864 	if (err)
865 		goto out_unlock;
866 
867 	err = -EACCES;
868 	switch (cmd) {
869 	case GETALL:
870 	{
871 		ushort __user *array = arg.array;
872 		int i;
873 
874 		if(nsems > SEMMSL_FAST) {
875 			sem_getref_and_unlock(sma);
876 
877 			sem_io = ipc_alloc(sizeof(ushort)*nsems);
878 			if(sem_io == NULL) {
879 				sem_putref(sma);
880 				return -ENOMEM;
881 			}
882 
883 			sem_lock_and_putref(sma);
884 			if (sma->sem_perm.deleted) {
885 				sem_unlock(sma);
886 				err = -EIDRM;
887 				goto out_free;
888 			}
889 		}
890 
891 		for (i = 0; i < sma->sem_nsems; i++)
892 			sem_io[i] = sma->sem_base[i].semval;
893 		sem_unlock(sma);
894 		err = 0;
895 		if(copy_to_user(array, sem_io, nsems*sizeof(ushort)))
896 			err = -EFAULT;
897 		goto out_free;
898 	}
899 	case SETALL:
900 	{
901 		int i;
902 		struct sem_undo *un;
903 
904 		sem_getref_and_unlock(sma);
905 
906 		if(nsems > SEMMSL_FAST) {
907 			sem_io = ipc_alloc(sizeof(ushort)*nsems);
908 			if(sem_io == NULL) {
909 				sem_putref(sma);
910 				return -ENOMEM;
911 			}
912 		}
913 
914 		if (copy_from_user (sem_io, arg.array, nsems*sizeof(ushort))) {
915 			sem_putref(sma);
916 			err = -EFAULT;
917 			goto out_free;
918 		}
919 
920 		for (i = 0; i < nsems; i++) {
921 			if (sem_io[i] > SEMVMX) {
922 				sem_putref(sma);
923 				err = -ERANGE;
924 				goto out_free;
925 			}
926 		}
927 		sem_lock_and_putref(sma);
928 		if (sma->sem_perm.deleted) {
929 			sem_unlock(sma);
930 			err = -EIDRM;
931 			goto out_free;
932 		}
933 
934 		for (i = 0; i < nsems; i++)
935 			sma->sem_base[i].semval = sem_io[i];
936 
937 		assert_spin_locked(&sma->sem_perm.lock);
938 		list_for_each_entry(un, &sma->list_id, list_id) {
939 			for (i = 0; i < nsems; i++)
940 				un->semadj[i] = 0;
941 		}
942 		sma->sem_ctime = get_seconds();
943 		/* maybe some queued-up processes were waiting for this */
944 		do_smart_update(sma, NULL, 0, 0, &tasks);
945 		err = 0;
946 		goto out_unlock;
947 	}
948 	/* GETVAL, GETPID, GETNCTN, GETZCNT, SETVAL: fall-through */
949 	}
950 	err = -EINVAL;
951 	if(semnum < 0 || semnum >= nsems)
952 		goto out_unlock;
953 
954 	curr = &sma->sem_base[semnum];
955 
956 	switch (cmd) {
957 	case GETVAL:
958 		err = curr->semval;
959 		goto out_unlock;
960 	case GETPID:
961 		err = curr->sempid;
962 		goto out_unlock;
963 	case GETNCNT:
964 		err = count_semncnt(sma,semnum);
965 		goto out_unlock;
966 	case GETZCNT:
967 		err = count_semzcnt(sma,semnum);
968 		goto out_unlock;
969 	case SETVAL:
970 	{
971 		int val = arg.val;
972 		struct sem_undo *un;
973 
974 		err = -ERANGE;
975 		if (val > SEMVMX || val < 0)
976 			goto out_unlock;
977 
978 		assert_spin_locked(&sma->sem_perm.lock);
979 		list_for_each_entry(un, &sma->list_id, list_id)
980 			un->semadj[semnum] = 0;
981 
982 		curr->semval = val;
983 		curr->sempid = task_tgid_vnr(current);
984 		sma->sem_ctime = get_seconds();
985 		/* maybe some queued-up processes were waiting for this */
986 		do_smart_update(sma, NULL, 0, 0, &tasks);
987 		err = 0;
988 		goto out_unlock;
989 	}
990 	}
991 out_unlock:
992 	sem_unlock(sma);
993 	wake_up_sem_queue_do(&tasks);
994 
995 out_free:
996 	if(sem_io != fast_sem_io)
997 		ipc_free(sem_io, sizeof(ushort)*nsems);
998 	return err;
999 }
1000 
1001 static inline unsigned long
1002 copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
1003 {
1004 	switch(version) {
1005 	case IPC_64:
1006 		if (copy_from_user(out, buf, sizeof(*out)))
1007 			return -EFAULT;
1008 		return 0;
1009 	case IPC_OLD:
1010 	    {
1011 		struct semid_ds tbuf_old;
1012 
1013 		if(copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
1014 			return -EFAULT;
1015 
1016 		out->sem_perm.uid	= tbuf_old.sem_perm.uid;
1017 		out->sem_perm.gid	= tbuf_old.sem_perm.gid;
1018 		out->sem_perm.mode	= tbuf_old.sem_perm.mode;
1019 
1020 		return 0;
1021 	    }
1022 	default:
1023 		return -EINVAL;
1024 	}
1025 }
1026 
1027 /*
1028  * This function handles some semctl commands which require the rw_mutex
1029  * to be held in write mode.
1030  * NOTE: no locks must be held, the rw_mutex is taken inside this function.
1031  */
1032 static int semctl_down(struct ipc_namespace *ns, int semid,
1033 		       int cmd, int version, union semun arg)
1034 {
1035 	struct sem_array *sma;
1036 	int err;
1037 	struct semid64_ds semid64;
1038 	struct kern_ipc_perm *ipcp;
1039 
1040 	if(cmd == IPC_SET) {
1041 		if (copy_semid_from_user(&semid64, arg.buf, version))
1042 			return -EFAULT;
1043 	}
1044 
1045 	ipcp = ipcctl_pre_down(ns, &sem_ids(ns), semid, cmd,
1046 			       &semid64.sem_perm, 0);
1047 	if (IS_ERR(ipcp))
1048 		return PTR_ERR(ipcp);
1049 
1050 	sma = container_of(ipcp, struct sem_array, sem_perm);
1051 
1052 	err = security_sem_semctl(sma, cmd);
1053 	if (err)
1054 		goto out_unlock;
1055 
1056 	switch(cmd){
1057 	case IPC_RMID:
1058 		freeary(ns, ipcp);
1059 		goto out_up;
1060 	case IPC_SET:
1061 		ipc_update_perm(&semid64.sem_perm, ipcp);
1062 		sma->sem_ctime = get_seconds();
1063 		break;
1064 	default:
1065 		err = -EINVAL;
1066 	}
1067 
1068 out_unlock:
1069 	sem_unlock(sma);
1070 out_up:
1071 	up_write(&sem_ids(ns).rw_mutex);
1072 	return err;
1073 }
1074 
1075 SYSCALL_DEFINE(semctl)(int semid, int semnum, int cmd, union semun arg)
1076 {
1077 	int err = -EINVAL;
1078 	int version;
1079 	struct ipc_namespace *ns;
1080 
1081 	if (semid < 0)
1082 		return -EINVAL;
1083 
1084 	version = ipc_parse_version(&cmd);
1085 	ns = current->nsproxy->ipc_ns;
1086 
1087 	switch(cmd) {
1088 	case IPC_INFO:
1089 	case SEM_INFO:
1090 	case IPC_STAT:
1091 	case SEM_STAT:
1092 		err = semctl_nolock(ns, semid, cmd, version, arg);
1093 		return err;
1094 	case GETALL:
1095 	case GETVAL:
1096 	case GETPID:
1097 	case GETNCNT:
1098 	case GETZCNT:
1099 	case SETVAL:
1100 	case SETALL:
1101 		err = semctl_main(ns,semid,semnum,cmd,version,arg);
1102 		return err;
1103 	case IPC_RMID:
1104 	case IPC_SET:
1105 		err = semctl_down(ns, semid, cmd, version, arg);
1106 		return err;
1107 	default:
1108 		return -EINVAL;
1109 	}
1110 }
1111 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
1112 asmlinkage long SyS_semctl(int semid, int semnum, int cmd, union semun arg)
1113 {
1114 	return SYSC_semctl((int) semid, (int) semnum, (int) cmd, arg);
1115 }
1116 SYSCALL_ALIAS(sys_semctl, SyS_semctl);
1117 #endif
1118 
1119 /* If the task doesn't already have a undo_list, then allocate one
1120  * here.  We guarantee there is only one thread using this undo list,
1121  * and current is THE ONE
1122  *
1123  * If this allocation and assignment succeeds, but later
1124  * portions of this code fail, there is no need to free the sem_undo_list.
1125  * Just let it stay associated with the task, and it'll be freed later
1126  * at exit time.
1127  *
1128  * This can block, so callers must hold no locks.
1129  */
1130 static inline int get_undo_list(struct sem_undo_list **undo_listp)
1131 {
1132 	struct sem_undo_list *undo_list;
1133 
1134 	undo_list = current->sysvsem.undo_list;
1135 	if (!undo_list) {
1136 		undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
1137 		if (undo_list == NULL)
1138 			return -ENOMEM;
1139 		spin_lock_init(&undo_list->lock);
1140 		atomic_set(&undo_list->refcnt, 1);
1141 		INIT_LIST_HEAD(&undo_list->list_proc);
1142 
1143 		current->sysvsem.undo_list = undo_list;
1144 	}
1145 	*undo_listp = undo_list;
1146 	return 0;
1147 }
1148 
1149 static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid)
1150 {
1151 	struct sem_undo *un;
1152 
1153 	list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) {
1154 		if (un->semid == semid)
1155 			return un;
1156 	}
1157 	return NULL;
1158 }
1159 
1160 static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1161 {
1162 	struct sem_undo *un;
1163 
1164   	assert_spin_locked(&ulp->lock);
1165 
1166 	un = __lookup_undo(ulp, semid);
1167 	if (un) {
1168 		list_del_rcu(&un->list_proc);
1169 		list_add_rcu(&un->list_proc, &ulp->list_proc);
1170 	}
1171 	return un;
1172 }
1173 
1174 /**
1175  * find_alloc_undo - Lookup (and if not present create) undo array
1176  * @ns: namespace
1177  * @semid: semaphore array id
1178  *
1179  * The function looks up (and if not present creates) the undo structure.
1180  * The size of the undo structure depends on the size of the semaphore
1181  * array, thus the alloc path is not that straightforward.
1182  * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1183  * performs a rcu_read_lock().
1184  */
1185 static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
1186 {
1187 	struct sem_array *sma;
1188 	struct sem_undo_list *ulp;
1189 	struct sem_undo *un, *new;
1190 	int nsems;
1191 	int error;
1192 
1193 	error = get_undo_list(&ulp);
1194 	if (error)
1195 		return ERR_PTR(error);
1196 
1197 	rcu_read_lock();
1198 	spin_lock(&ulp->lock);
1199 	un = lookup_undo(ulp, semid);
1200 	spin_unlock(&ulp->lock);
1201 	if (likely(un!=NULL))
1202 		goto out;
1203 	rcu_read_unlock();
1204 
1205 	/* no undo structure around - allocate one. */
1206 	/* step 1: figure out the size of the semaphore array */
1207 	sma = sem_lock_check(ns, semid);
1208 	if (IS_ERR(sma))
1209 		return ERR_CAST(sma);
1210 
1211 	nsems = sma->sem_nsems;
1212 	sem_getref_and_unlock(sma);
1213 
1214 	/* step 2: allocate new undo structure */
1215 	new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1216 	if (!new) {
1217 		sem_putref(sma);
1218 		return ERR_PTR(-ENOMEM);
1219 	}
1220 
1221 	/* step 3: Acquire the lock on semaphore array */
1222 	sem_lock_and_putref(sma);
1223 	if (sma->sem_perm.deleted) {
1224 		sem_unlock(sma);
1225 		kfree(new);
1226 		un = ERR_PTR(-EIDRM);
1227 		goto out;
1228 	}
1229 	spin_lock(&ulp->lock);
1230 
1231 	/*
1232 	 * step 4: check for races: did someone else allocate the undo struct?
1233 	 */
1234 	un = lookup_undo(ulp, semid);
1235 	if (un) {
1236 		kfree(new);
1237 		goto success;
1238 	}
1239 	/* step 5: initialize & link new undo structure */
1240 	new->semadj = (short *) &new[1];
1241 	new->ulp = ulp;
1242 	new->semid = semid;
1243 	assert_spin_locked(&ulp->lock);
1244 	list_add_rcu(&new->list_proc, &ulp->list_proc);
1245 	assert_spin_locked(&sma->sem_perm.lock);
1246 	list_add(&new->list_id, &sma->list_id);
1247 	un = new;
1248 
1249 success:
1250 	spin_unlock(&ulp->lock);
1251 	rcu_read_lock();
1252 	sem_unlock(sma);
1253 out:
1254 	return un;
1255 }
1256 
1257 
1258 /**
1259  * get_queue_result - Retrieve the result code from sem_queue
1260  * @q: Pointer to queue structure
1261  *
1262  * Retrieve the return code from the pending queue. If IN_WAKEUP is found in
1263  * q->status, then we must loop until the value is replaced with the final
1264  * value: This may happen if a task is woken up by an unrelated event (e.g.
1265  * signal) and in parallel the task is woken up by another task because it got
1266  * the requested semaphores.
1267  *
1268  * The function can be called with or without holding the semaphore spinlock.
1269  */
1270 static int get_queue_result(struct sem_queue *q)
1271 {
1272 	int error;
1273 
1274 	error = q->status;
1275 	while (unlikely(error == IN_WAKEUP)) {
1276 		cpu_relax();
1277 		error = q->status;
1278 	}
1279 
1280 	return error;
1281 }
1282 
1283 
1284 SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
1285 		unsigned, nsops, const struct timespec __user *, timeout)
1286 {
1287 	int error = -EINVAL;
1288 	struct sem_array *sma;
1289 	struct sembuf fast_sops[SEMOPM_FAST];
1290 	struct sembuf* sops = fast_sops, *sop;
1291 	struct sem_undo *un;
1292 	int undos = 0, alter = 0, max;
1293 	struct sem_queue queue;
1294 	unsigned long jiffies_left = 0;
1295 	struct ipc_namespace *ns;
1296 	struct list_head tasks;
1297 
1298 	ns = current->nsproxy->ipc_ns;
1299 
1300 	if (nsops < 1 || semid < 0)
1301 		return -EINVAL;
1302 	if (nsops > ns->sc_semopm)
1303 		return -E2BIG;
1304 	if(nsops > SEMOPM_FAST) {
1305 		sops = kmalloc(sizeof(*sops)*nsops,GFP_KERNEL);
1306 		if(sops==NULL)
1307 			return -ENOMEM;
1308 	}
1309 	if (copy_from_user (sops, tsops, nsops * sizeof(*tsops))) {
1310 		error=-EFAULT;
1311 		goto out_free;
1312 	}
1313 	if (timeout) {
1314 		struct timespec _timeout;
1315 		if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1316 			error = -EFAULT;
1317 			goto out_free;
1318 		}
1319 		if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1320 			_timeout.tv_nsec >= 1000000000L) {
1321 			error = -EINVAL;
1322 			goto out_free;
1323 		}
1324 		jiffies_left = timespec_to_jiffies(&_timeout);
1325 	}
1326 	max = 0;
1327 	for (sop = sops; sop < sops + nsops; sop++) {
1328 		if (sop->sem_num >= max)
1329 			max = sop->sem_num;
1330 		if (sop->sem_flg & SEM_UNDO)
1331 			undos = 1;
1332 		if (sop->sem_op != 0)
1333 			alter = 1;
1334 	}
1335 
1336 	if (undos) {
1337 		un = find_alloc_undo(ns, semid);
1338 		if (IS_ERR(un)) {
1339 			error = PTR_ERR(un);
1340 			goto out_free;
1341 		}
1342 	} else
1343 		un = NULL;
1344 
1345 	INIT_LIST_HEAD(&tasks);
1346 
1347 	sma = sem_lock_check(ns, semid);
1348 	if (IS_ERR(sma)) {
1349 		if (un)
1350 			rcu_read_unlock();
1351 		error = PTR_ERR(sma);
1352 		goto out_free;
1353 	}
1354 
1355 	/*
1356 	 * semid identifiers are not unique - find_alloc_undo may have
1357 	 * allocated an undo structure, it was invalidated by an RMID
1358 	 * and now a new array with received the same id. Check and fail.
1359 	 * This case can be detected checking un->semid. The existence of
1360 	 * "un" itself is guaranteed by rcu.
1361 	 */
1362 	error = -EIDRM;
1363 	if (un) {
1364 		if (un->semid == -1) {
1365 			rcu_read_unlock();
1366 			goto out_unlock_free;
1367 		} else {
1368 			/*
1369 			 * rcu lock can be released, "un" cannot disappear:
1370 			 * - sem_lock is acquired, thus IPC_RMID is
1371 			 *   impossible.
1372 			 * - exit_sem is impossible, it always operates on
1373 			 *   current (or a dead task).
1374 			 */
1375 
1376 			rcu_read_unlock();
1377 		}
1378 	}
1379 
1380 	error = -EFBIG;
1381 	if (max >= sma->sem_nsems)
1382 		goto out_unlock_free;
1383 
1384 	error = -EACCES;
1385 	if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO))
1386 		goto out_unlock_free;
1387 
1388 	error = security_sem_semop(sma, sops, nsops, alter);
1389 	if (error)
1390 		goto out_unlock_free;
1391 
1392 	error = try_atomic_semop (sma, sops, nsops, un, task_tgid_vnr(current));
1393 	if (error <= 0) {
1394 		if (alter && error == 0)
1395 			do_smart_update(sma, sops, nsops, 1, &tasks);
1396 
1397 		goto out_unlock_free;
1398 	}
1399 
1400 	/* We need to sleep on this operation, so we put the current
1401 	 * task into the pending queue and go to sleep.
1402 	 */
1403 
1404 	queue.sops = sops;
1405 	queue.nsops = nsops;
1406 	queue.undo = un;
1407 	queue.pid = task_tgid_vnr(current);
1408 	queue.alter = alter;
1409 	if (alter)
1410 		list_add_tail(&queue.list, &sma->sem_pending);
1411 	else
1412 		list_add(&queue.list, &sma->sem_pending);
1413 
1414 	if (nsops == 1) {
1415 		struct sem *curr;
1416 		curr = &sma->sem_base[sops->sem_num];
1417 
1418 		if (alter)
1419 			list_add_tail(&queue.simple_list, &curr->sem_pending);
1420 		else
1421 			list_add(&queue.simple_list, &curr->sem_pending);
1422 	} else {
1423 		INIT_LIST_HEAD(&queue.simple_list);
1424 		sma->complex_count++;
1425 	}
1426 
1427 	queue.status = -EINTR;
1428 	queue.sleeper = current;
1429 	current->state = TASK_INTERRUPTIBLE;
1430 	sem_unlock(sma);
1431 
1432 	if (timeout)
1433 		jiffies_left = schedule_timeout(jiffies_left);
1434 	else
1435 		schedule();
1436 
1437 	error = get_queue_result(&queue);
1438 
1439 	if (error != -EINTR) {
1440 		/* fast path: update_queue already obtained all requested
1441 		 * resources.
1442 		 * Perform a smp_mb(): User space could assume that semop()
1443 		 * is a memory barrier: Without the mb(), the cpu could
1444 		 * speculatively read in user space stale data that was
1445 		 * overwritten by the previous owner of the semaphore.
1446 		 */
1447 		smp_mb();
1448 
1449 		goto out_free;
1450 	}
1451 
1452 	sma = sem_lock(ns, semid);
1453 
1454 	/*
1455 	 * Wait until it's guaranteed that no wakeup_sem_queue_do() is ongoing.
1456 	 */
1457 	error = get_queue_result(&queue);
1458 
1459 	/*
1460 	 * Array removed? If yes, leave without sem_unlock().
1461 	 */
1462 	if (IS_ERR(sma)) {
1463 		error = -EIDRM;
1464 		goto out_free;
1465 	}
1466 
1467 
1468 	/*
1469 	 * If queue.status != -EINTR we are woken up by another process.
1470 	 * Leave without unlink_queue(), but with sem_unlock().
1471 	 */
1472 
1473 	if (error != -EINTR) {
1474 		goto out_unlock_free;
1475 	}
1476 
1477 	/*
1478 	 * If an interrupt occurred we have to clean up the queue
1479 	 */
1480 	if (timeout && jiffies_left == 0)
1481 		error = -EAGAIN;
1482 	unlink_queue(sma, &queue);
1483 
1484 out_unlock_free:
1485 	sem_unlock(sma);
1486 
1487 	wake_up_sem_queue_do(&tasks);
1488 out_free:
1489 	if(sops != fast_sops)
1490 		kfree(sops);
1491 	return error;
1492 }
1493 
1494 SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
1495 		unsigned, nsops)
1496 {
1497 	return sys_semtimedop(semid, tsops, nsops, NULL);
1498 }
1499 
1500 /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
1501  * parent and child tasks.
1502  */
1503 
1504 int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
1505 {
1506 	struct sem_undo_list *undo_list;
1507 	int error;
1508 
1509 	if (clone_flags & CLONE_SYSVSEM) {
1510 		error = get_undo_list(&undo_list);
1511 		if (error)
1512 			return error;
1513 		atomic_inc(&undo_list->refcnt);
1514 		tsk->sysvsem.undo_list = undo_list;
1515 	} else
1516 		tsk->sysvsem.undo_list = NULL;
1517 
1518 	return 0;
1519 }
1520 
1521 /*
1522  * add semadj values to semaphores, free undo structures.
1523  * undo structures are not freed when semaphore arrays are destroyed
1524  * so some of them may be out of date.
1525  * IMPLEMENTATION NOTE: There is some confusion over whether the
1526  * set of adjustments that needs to be done should be done in an atomic
1527  * manner or not. That is, if we are attempting to decrement the semval
1528  * should we queue up and wait until we can do so legally?
1529  * The original implementation attempted to do this (queue and wait).
1530  * The current implementation does not do so. The POSIX standard
1531  * and SVID should be consulted to determine what behavior is mandated.
1532  */
1533 void exit_sem(struct task_struct *tsk)
1534 {
1535 	struct sem_undo_list *ulp;
1536 
1537 	ulp = tsk->sysvsem.undo_list;
1538 	if (!ulp)
1539 		return;
1540 	tsk->sysvsem.undo_list = NULL;
1541 
1542 	if (!atomic_dec_and_test(&ulp->refcnt))
1543 		return;
1544 
1545 	for (;;) {
1546 		struct sem_array *sma;
1547 		struct sem_undo *un;
1548 		struct list_head tasks;
1549 		int semid;
1550 		int i;
1551 
1552 		rcu_read_lock();
1553 		un = list_entry_rcu(ulp->list_proc.next,
1554 				    struct sem_undo, list_proc);
1555 		if (&un->list_proc == &ulp->list_proc)
1556 			semid = -1;
1557 		 else
1558 			semid = un->semid;
1559 		rcu_read_unlock();
1560 
1561 		if (semid == -1)
1562 			break;
1563 
1564 		sma = sem_lock_check(tsk->nsproxy->ipc_ns, un->semid);
1565 
1566 		/* exit_sem raced with IPC_RMID, nothing to do */
1567 		if (IS_ERR(sma))
1568 			continue;
1569 
1570 		un = __lookup_undo(ulp, semid);
1571 		if (un == NULL) {
1572 			/* exit_sem raced with IPC_RMID+semget() that created
1573 			 * exactly the same semid. Nothing to do.
1574 			 */
1575 			sem_unlock(sma);
1576 			continue;
1577 		}
1578 
1579 		/* remove un from the linked lists */
1580 		assert_spin_locked(&sma->sem_perm.lock);
1581 		list_del(&un->list_id);
1582 
1583 		spin_lock(&ulp->lock);
1584 		list_del_rcu(&un->list_proc);
1585 		spin_unlock(&ulp->lock);
1586 
1587 		/* perform adjustments registered in un */
1588 		for (i = 0; i < sma->sem_nsems; i++) {
1589 			struct sem * semaphore = &sma->sem_base[i];
1590 			if (un->semadj[i]) {
1591 				semaphore->semval += un->semadj[i];
1592 				/*
1593 				 * Range checks of the new semaphore value,
1594 				 * not defined by sus:
1595 				 * - Some unices ignore the undo entirely
1596 				 *   (e.g. HP UX 11i 11.22, Tru64 V5.1)
1597 				 * - some cap the value (e.g. FreeBSD caps
1598 				 *   at 0, but doesn't enforce SEMVMX)
1599 				 *
1600 				 * Linux caps the semaphore value, both at 0
1601 				 * and at SEMVMX.
1602 				 *
1603 				 * 	Manfred <manfred@colorfullife.com>
1604 				 */
1605 				if (semaphore->semval < 0)
1606 					semaphore->semval = 0;
1607 				if (semaphore->semval > SEMVMX)
1608 					semaphore->semval = SEMVMX;
1609 				semaphore->sempid = task_tgid_vnr(current);
1610 			}
1611 		}
1612 		/* maybe some queued-up processes were waiting for this */
1613 		INIT_LIST_HEAD(&tasks);
1614 		do_smart_update(sma, NULL, 0, 1, &tasks);
1615 		sem_unlock(sma);
1616 		wake_up_sem_queue_do(&tasks);
1617 
1618 		kfree_rcu(un, rcu);
1619 	}
1620 	kfree(ulp);
1621 }
1622 
1623 #ifdef CONFIG_PROC_FS
1624 static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
1625 {
1626 	struct sem_array *sma = it;
1627 
1628 	return seq_printf(s,
1629 			  "%10d %10d  %4o %10u %5u %5u %5u %5u %10lu %10lu\n",
1630 			  sma->sem_perm.key,
1631 			  sma->sem_perm.id,
1632 			  sma->sem_perm.mode,
1633 			  sma->sem_nsems,
1634 			  sma->sem_perm.uid,
1635 			  sma->sem_perm.gid,
1636 			  sma->sem_perm.cuid,
1637 			  sma->sem_perm.cgid,
1638 			  sma->sem_otime,
1639 			  sma->sem_ctime);
1640 }
1641 #endif
1642