sem.c revision 239521f3
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/* One semaphore structure for each semaphore in the system. */
94struct sem {
95	int	semval;		/* current value */
96	int	sempid;		/* pid of last operation */
97	spinlock_t	lock;	/* spinlock for fine-grained semtimedop */
98	struct list_head pending_alter; /* pending single-sop operations */
99					/* that alter the semaphore */
100	struct list_head pending_const; /* pending single-sop operations */
101					/* that do not alter the semaphore*/
102	time_t	sem_otime;	/* candidate for sem_otime */
103} ____cacheline_aligned_in_smp;
104
105/* One queue for each sleeping process in the system. */
106struct sem_queue {
107	struct list_head	list;	 /* queue of pending operations */
108	struct task_struct	*sleeper; /* this process */
109	struct sem_undo		*undo;	 /* undo structure */
110	int			pid;	 /* process id of requesting process */
111	int			status;	 /* completion status of operation */
112	struct sembuf		*sops;	 /* array of pending operations */
113	int			nsops;	 /* number of operations */
114	int			alter;	 /* does *sops alter the array? */
115};
116
117/* Each task has a list of undo requests. They are executed automatically
118 * when the process exits.
119 */
120struct sem_undo {
121	struct list_head	list_proc;	/* per-process list: *
122						 * all undos from one process
123						 * rcu protected */
124	struct rcu_head		rcu;		/* rcu struct for sem_undo */
125	struct sem_undo_list	*ulp;		/* back ptr to sem_undo_list */
126	struct list_head	list_id;	/* per semaphore array list:
127						 * all undos for one array */
128	int			semid;		/* semaphore set identifier */
129	short			*semadj;	/* array of adjustments */
130						/* one per semaphore */
131};
132
133/* sem_undo_list controls shared access to the list of sem_undo structures
134 * that may be shared among all a CLONE_SYSVSEM task group.
135 */
136struct sem_undo_list {
137	atomic_t		refcnt;
138	spinlock_t		lock;
139	struct list_head	list_proc;
140};
141
142
143#define sem_ids(ns)	((ns)->ids[IPC_SEM_IDS])
144
145#define sem_checkid(sma, semid)	ipc_checkid(&sma->sem_perm, semid)
146
147static int newary(struct ipc_namespace *, struct ipc_params *);
148static void freeary(struct ipc_namespace *, struct kern_ipc_perm *);
149#ifdef CONFIG_PROC_FS
150static int sysvipc_sem_proc_show(struct seq_file *s, void *it);
151#endif
152
153#define SEMMSL_FAST	256 /* 512 bytes on stack */
154#define SEMOPM_FAST	64  /* ~ 372 bytes on stack */
155
156/*
157 * Locking:
158 *	sem_undo.id_next,
159 *	sem_array.complex_count,
160 *	sem_array.pending{_alter,_cont},
161 *	sem_array.sem_undo: global sem_lock() for read/write
162 *	sem_undo.proc_next: only "current" is allowed to read/write that field.
163 *
164 *	sem_array.sem_base[i].pending_{const,alter}:
165 *		global or semaphore sem_lock() for read/write
166 */
167
168#define sc_semmsl	sem_ctls[0]
169#define sc_semmns	sem_ctls[1]
170#define sc_semopm	sem_ctls[2]
171#define sc_semmni	sem_ctls[3]
172
173void sem_init_ns(struct ipc_namespace *ns)
174{
175	ns->sc_semmsl = SEMMSL;
176	ns->sc_semmns = SEMMNS;
177	ns->sc_semopm = SEMOPM;
178	ns->sc_semmni = SEMMNI;
179	ns->used_sems = 0;
180	ipc_init_ids(&ns->ids[IPC_SEM_IDS]);
181}
182
183#ifdef CONFIG_IPC_NS
184void sem_exit_ns(struct ipc_namespace *ns)
185{
186	free_ipcs(ns, &sem_ids(ns), freeary);
187	idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr);
188}
189#endif
190
191void __init sem_init(void)
192{
193	sem_init_ns(&init_ipc_ns);
194	ipc_init_proc_interface("sysvipc/sem",
195				"       key      semid perms      nsems   uid   gid  cuid  cgid      otime      ctime\n",
196				IPC_SEM_IDS, sysvipc_sem_proc_show);
197}
198
199/**
200 * unmerge_queues - unmerge queues, if possible.
201 * @sma: semaphore array
202 *
203 * The function unmerges the wait queues if complex_count is 0.
204 * It must be called prior to dropping the global semaphore array lock.
205 */
206static void unmerge_queues(struct sem_array *sma)
207{
208	struct sem_queue *q, *tq;
209
210	/* complex operations still around? */
211	if (sma->complex_count)
212		return;
213	/*
214	 * We will switch back to simple mode.
215	 * Move all pending operation back into the per-semaphore
216	 * queues.
217	 */
218	list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
219		struct sem *curr;
220		curr = &sma->sem_base[q->sops[0].sem_num];
221
222		list_add_tail(&q->list, &curr->pending_alter);
223	}
224	INIT_LIST_HEAD(&sma->pending_alter);
225}
226
227/**
228 * merge_queues - Merge single semop queues into global queue
229 * @sma: semaphore array
230 *
231 * This function merges all per-semaphore queues into the global queue.
232 * It is necessary to achieve FIFO ordering for the pending single-sop
233 * operations when a multi-semop operation must sleep.
234 * Only the alter operations must be moved, the const operations can stay.
235 */
236static void merge_queues(struct sem_array *sma)
237{
238	int i;
239	for (i = 0; i < sma->sem_nsems; i++) {
240		struct sem *sem = sma->sem_base + i;
241
242		list_splice_init(&sem->pending_alter, &sma->pending_alter);
243	}
244}
245
246static void sem_rcu_free(struct rcu_head *head)
247{
248	struct ipc_rcu *p = container_of(head, struct ipc_rcu, rcu);
249	struct sem_array *sma = ipc_rcu_to_struct(p);
250
251	security_sem_free(sma);
252	ipc_rcu_free(head);
253}
254
255/*
256 * Wait until all currently ongoing simple ops have completed.
257 * Caller must own sem_perm.lock.
258 * New simple ops cannot start, because simple ops first check
259 * that sem_perm.lock is free.
260 * that a) sem_perm.lock is free and b) complex_count is 0.
261 */
262static void sem_wait_array(struct sem_array *sma)
263{
264	int i;
265	struct sem *sem;
266
267	if (sma->complex_count)  {
268		/* The thread that increased sma->complex_count waited on
269		 * all sem->lock locks. Thus we don't need to wait again.
270		 */
271		return;
272	}
273
274	for (i = 0; i < sma->sem_nsems; i++) {
275		sem = sma->sem_base + i;
276		spin_unlock_wait(&sem->lock);
277	}
278}
279
280/*
281 * If the request contains only one semaphore operation, and there are
282 * no complex transactions pending, lock only the semaphore involved.
283 * Otherwise, lock the entire semaphore array, since we either have
284 * multiple semaphores in our own semops, or we need to look at
285 * semaphores from other pending complex operations.
286 */
287static inline int sem_lock(struct sem_array *sma, struct sembuf *sops,
288			      int nsops)
289{
290	struct sem *sem;
291
292	if (nsops != 1) {
293		/* Complex operation - acquire a full lock */
294		ipc_lock_object(&sma->sem_perm);
295
296		/* And wait until all simple ops that are processed
297		 * right now have dropped their locks.
298		 */
299		sem_wait_array(sma);
300		return -1;
301	}
302
303	/*
304	 * Only one semaphore affected - try to optimize locking.
305	 * The rules are:
306	 * - optimized locking is possible if no complex operation
307	 *   is either enqueued or processed right now.
308	 * - The test for enqueued complex ops is simple:
309	 *      sma->complex_count != 0
310	 * - Testing for complex ops that are processed right now is
311	 *   a bit more difficult. Complex ops acquire the full lock
312	 *   and first wait that the running simple ops have completed.
313	 *   (see above)
314	 *   Thus: If we own a simple lock and the global lock is free
315	 *	and complex_count is now 0, then it will stay 0 and
316	 *	thus just locking sem->lock is sufficient.
317	 */
318	sem = sma->sem_base + sops->sem_num;
319
320	if (sma->complex_count == 0) {
321		/*
322		 * It appears that no complex operation is around.
323		 * Acquire the per-semaphore lock.
324		 */
325		spin_lock(&sem->lock);
326
327		/* Then check that the global lock is free */
328		if (!spin_is_locked(&sma->sem_perm.lock)) {
329			/* spin_is_locked() is not a memory barrier */
330			smp_mb();
331
332			/* Now repeat the test of complex_count:
333			 * It can't change anymore until we drop sem->lock.
334			 * Thus: if is now 0, then it will stay 0.
335			 */
336			if (sma->complex_count == 0) {
337				/* fast path successful! */
338				return sops->sem_num;
339			}
340		}
341		spin_unlock(&sem->lock);
342	}
343
344	/* slow path: acquire the full lock */
345	ipc_lock_object(&sma->sem_perm);
346
347	if (sma->complex_count == 0) {
348		/* False alarm:
349		 * There is no complex operation, thus we can switch
350		 * back to the fast path.
351		 */
352		spin_lock(&sem->lock);
353		ipc_unlock_object(&sma->sem_perm);
354		return sops->sem_num;
355	} else {
356		/* Not a false alarm, thus complete the sequence for a
357		 * full lock.
358		 */
359		sem_wait_array(sma);
360		return -1;
361	}
362}
363
364static inline void sem_unlock(struct sem_array *sma, int locknum)
365{
366	if (locknum == -1) {
367		unmerge_queues(sma);
368		ipc_unlock_object(&sma->sem_perm);
369	} else {
370		struct sem *sem = sma->sem_base + locknum;
371		spin_unlock(&sem->lock);
372	}
373}
374
375/*
376 * sem_lock_(check_) routines are called in the paths where the rwsem
377 * is not held.
378 *
379 * The caller holds the RCU read lock.
380 */
381static inline struct sem_array *sem_obtain_lock(struct ipc_namespace *ns,
382			int id, struct sembuf *sops, int nsops, int *locknum)
383{
384	struct kern_ipc_perm *ipcp;
385	struct sem_array *sma;
386
387	ipcp = ipc_obtain_object(&sem_ids(ns), id);
388	if (IS_ERR(ipcp))
389		return ERR_CAST(ipcp);
390
391	sma = container_of(ipcp, struct sem_array, sem_perm);
392	*locknum = sem_lock(sma, sops, nsops);
393
394	/* ipc_rmid() may have already freed the ID while sem_lock
395	 * was spinning: verify that the structure is still valid
396	 */
397	if (ipc_valid_object(ipcp))
398		return container_of(ipcp, struct sem_array, sem_perm);
399
400	sem_unlock(sma, *locknum);
401	return ERR_PTR(-EINVAL);
402}
403
404static inline struct sem_array *sem_obtain_object(struct ipc_namespace *ns, int id)
405{
406	struct kern_ipc_perm *ipcp = ipc_obtain_object(&sem_ids(ns), id);
407
408	if (IS_ERR(ipcp))
409		return ERR_CAST(ipcp);
410
411	return container_of(ipcp, struct sem_array, sem_perm);
412}
413
414static inline struct sem_array *sem_obtain_object_check(struct ipc_namespace *ns,
415							int id)
416{
417	struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&sem_ids(ns), id);
418
419	if (IS_ERR(ipcp))
420		return ERR_CAST(ipcp);
421
422	return container_of(ipcp, struct sem_array, sem_perm);
423}
424
425static inline void sem_lock_and_putref(struct sem_array *sma)
426{
427	sem_lock(sma, NULL, -1);
428	ipc_rcu_putref(sma, ipc_rcu_free);
429}
430
431static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
432{
433	ipc_rmid(&sem_ids(ns), &s->sem_perm);
434}
435
436/*
437 * Lockless wakeup algorithm:
438 * Without the check/retry algorithm a lockless wakeup is possible:
439 * - queue.status is initialized to -EINTR before blocking.
440 * - wakeup is performed by
441 *	* unlinking the queue entry from the pending list
442 *	* setting queue.status to IN_WAKEUP
443 *	  This is the notification for the blocked thread that a
444 *	  result value is imminent.
445 *	* call wake_up_process
446 *	* set queue.status to the final value.
447 * - the previously blocked thread checks queue.status:
448 *	* if it's IN_WAKEUP, then it must wait until the value changes
449 *	* if it's not -EINTR, then the operation was completed by
450 *	  update_queue. semtimedop can return queue.status without
451 *	  performing any operation on the sem array.
452 *	* otherwise it must acquire the spinlock and check what's up.
453 *
454 * The two-stage algorithm is necessary to protect against the following
455 * races:
456 * - if queue.status is set after wake_up_process, then the woken up idle
457 *   thread could race forward and try (and fail) to acquire sma->lock
458 *   before update_queue had a chance to set queue.status
459 * - if queue.status is written before wake_up_process and if the
460 *   blocked process is woken up by a signal between writing
461 *   queue.status and the wake_up_process, then the woken up
462 *   process could return from semtimedop and die by calling
463 *   sys_exit before wake_up_process is called. Then wake_up_process
464 *   will oops, because the task structure is already invalid.
465 *   (yes, this happened on s390 with sysv msg).
466 *
467 */
468#define IN_WAKEUP	1
469
470/**
471 * newary - Create a new semaphore set
472 * @ns: namespace
473 * @params: ptr to the structure that contains key, semflg and nsems
474 *
475 * Called with sem_ids.rwsem held (as a writer)
476 */
477
478static int newary(struct ipc_namespace *ns, struct ipc_params *params)
479{
480	int id;
481	int retval;
482	struct sem_array *sma;
483	int size;
484	key_t key = params->key;
485	int nsems = params->u.nsems;
486	int semflg = params->flg;
487	int i;
488
489	if (!nsems)
490		return -EINVAL;
491	if (ns->used_sems + nsems > ns->sc_semmns)
492		return -ENOSPC;
493
494	size = sizeof(*sma) + nsems * sizeof(struct sem);
495	sma = ipc_rcu_alloc(size);
496	if (!sma) {
497		return -ENOMEM;
498	}
499	memset(sma, 0, size);
500
501	sma->sem_perm.mode = (semflg & S_IRWXUGO);
502	sma->sem_perm.key = key;
503
504	sma->sem_perm.security = NULL;
505	retval = security_sem_alloc(sma);
506	if (retval) {
507		ipc_rcu_putref(sma, ipc_rcu_free);
508		return retval;
509	}
510
511	id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
512	if (id < 0) {
513		ipc_rcu_putref(sma, sem_rcu_free);
514		return id;
515	}
516	ns->used_sems += nsems;
517
518	sma->sem_base = (struct sem *) &sma[1];
519
520	for (i = 0; i < nsems; i++) {
521		INIT_LIST_HEAD(&sma->sem_base[i].pending_alter);
522		INIT_LIST_HEAD(&sma->sem_base[i].pending_const);
523		spin_lock_init(&sma->sem_base[i].lock);
524	}
525
526	sma->complex_count = 0;
527	INIT_LIST_HEAD(&sma->pending_alter);
528	INIT_LIST_HEAD(&sma->pending_const);
529	INIT_LIST_HEAD(&sma->list_id);
530	sma->sem_nsems = nsems;
531	sma->sem_ctime = get_seconds();
532	sem_unlock(sma, -1);
533	rcu_read_unlock();
534
535	return sma->sem_perm.id;
536}
537
538
539/*
540 * Called with sem_ids.rwsem and ipcp locked.
541 */
542static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg)
543{
544	struct sem_array *sma;
545
546	sma = container_of(ipcp, struct sem_array, sem_perm);
547	return security_sem_associate(sma, semflg);
548}
549
550/*
551 * Called with sem_ids.rwsem and ipcp locked.
552 */
553static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
554				struct ipc_params *params)
555{
556	struct sem_array *sma;
557
558	sma = container_of(ipcp, struct sem_array, sem_perm);
559	if (params->u.nsems > sma->sem_nsems)
560		return -EINVAL;
561
562	return 0;
563}
564
565SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg)
566{
567	struct ipc_namespace *ns;
568	struct ipc_ops sem_ops;
569	struct ipc_params sem_params;
570
571	ns = current->nsproxy->ipc_ns;
572
573	if (nsems < 0 || nsems > ns->sc_semmsl)
574		return -EINVAL;
575
576	sem_ops.getnew = newary;
577	sem_ops.associate = sem_security;
578	sem_ops.more_checks = sem_more_checks;
579
580	sem_params.key = key;
581	sem_params.flg = semflg;
582	sem_params.u.nsems = nsems;
583
584	return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
585}
586
587/**
588 * perform_atomic_semop - Perform (if possible) a semaphore operation
589 * @sma: semaphore array
590 * @sops: array with operations that should be checked
591 * @nsops: number of operations
592 * @un: undo array
593 * @pid: pid that did the change
594 *
595 * Returns 0 if the operation was possible.
596 * Returns 1 if the operation is impossible, the caller must sleep.
597 * Negative values are error codes.
598 */
599static int perform_atomic_semop(struct sem_array *sma, struct sembuf *sops,
600			     int nsops, struct sem_undo *un, int pid)
601{
602	int result, sem_op;
603	struct sembuf *sop;
604	struct sem *curr;
605
606	for (sop = sops; sop < sops + nsops; sop++) {
607		curr = sma->sem_base + sop->sem_num;
608		sem_op = sop->sem_op;
609		result = curr->semval;
610
611		if (!sem_op && result)
612			goto would_block;
613
614		result += sem_op;
615		if (result < 0)
616			goto would_block;
617		if (result > SEMVMX)
618			goto out_of_range;
619
620		if (sop->sem_flg & SEM_UNDO) {
621			int undo = un->semadj[sop->sem_num] - sem_op;
622			/* Exceeding the undo range is an error. */
623			if (undo < (-SEMAEM - 1) || undo > SEMAEM)
624				goto out_of_range;
625			un->semadj[sop->sem_num] = undo;
626		}
627
628		curr->semval = result;
629	}
630
631	sop--;
632	while (sop >= sops) {
633		sma->sem_base[sop->sem_num].sempid = pid;
634		sop--;
635	}
636
637	return 0;
638
639out_of_range:
640	result = -ERANGE;
641	goto undo;
642
643would_block:
644	if (sop->sem_flg & IPC_NOWAIT)
645		result = -EAGAIN;
646	else
647		result = 1;
648
649undo:
650	sop--;
651	while (sop >= sops) {
652		sem_op = sop->sem_op;
653		sma->sem_base[sop->sem_num].semval -= sem_op;
654		if (sop->sem_flg & SEM_UNDO)
655			un->semadj[sop->sem_num] += sem_op;
656		sop--;
657	}
658
659	return result;
660}
661
662/** wake_up_sem_queue_prepare(q, error): Prepare wake-up
663 * @q: queue entry that must be signaled
664 * @error: Error value for the signal
665 *
666 * Prepare the wake-up of the queue entry q.
667 */
668static void wake_up_sem_queue_prepare(struct list_head *pt,
669				struct sem_queue *q, int error)
670{
671	if (list_empty(pt)) {
672		/*
673		 * Hold preempt off so that we don't get preempted and have the
674		 * wakee busy-wait until we're scheduled back on.
675		 */
676		preempt_disable();
677	}
678	q->status = IN_WAKEUP;
679	q->pid = error;
680
681	list_add_tail(&q->list, pt);
682}
683
684/**
685 * wake_up_sem_queue_do(pt) - do the actual wake-up
686 * @pt: list of tasks to be woken up
687 *
688 * Do the actual wake-up.
689 * The function is called without any locks held, thus the semaphore array
690 * could be destroyed already and the tasks can disappear as soon as the
691 * status is set to the actual return code.
692 */
693static void wake_up_sem_queue_do(struct list_head *pt)
694{
695	struct sem_queue *q, *t;
696	int did_something;
697
698	did_something = !list_empty(pt);
699	list_for_each_entry_safe(q, t, pt, list) {
700		wake_up_process(q->sleeper);
701		/* q can disappear immediately after writing q->status. */
702		smp_wmb();
703		q->status = q->pid;
704	}
705	if (did_something)
706		preempt_enable();
707}
708
709static void unlink_queue(struct sem_array *sma, struct sem_queue *q)
710{
711	list_del(&q->list);
712	if (q->nsops > 1)
713		sma->complex_count--;
714}
715
716/** check_restart(sma, q)
717 * @sma: semaphore array
718 * @q: the operation that just completed
719 *
720 * update_queue is O(N^2) when it restarts scanning the whole queue of
721 * waiting operations. Therefore this function checks if the restart is
722 * really necessary. It is called after a previously waiting operation
723 * modified the array.
724 * Note that wait-for-zero operations are handled without restart.
725 */
726static int check_restart(struct sem_array *sma, struct sem_queue *q)
727{
728	/* pending complex alter operations are too difficult to analyse */
729	if (!list_empty(&sma->pending_alter))
730		return 1;
731
732	/* we were a sleeping complex operation. Too difficult */
733	if (q->nsops > 1)
734		return 1;
735
736	/* It is impossible that someone waits for the new value:
737	 * - complex operations always restart.
738	 * - wait-for-zero are handled seperately.
739	 * - q is a previously sleeping simple operation that
740	 *   altered the array. It must be a decrement, because
741	 *   simple increments never sleep.
742	 * - If there are older (higher priority) decrements
743	 *   in the queue, then they have observed the original
744	 *   semval value and couldn't proceed. The operation
745	 *   decremented to value - thus they won't proceed either.
746	 */
747	return 0;
748}
749
750/**
751 * wake_const_ops(sma, semnum, pt) - Wake up non-alter tasks
752 * @sma: semaphore array.
753 * @semnum: semaphore that was modified.
754 * @pt: list head for the tasks that must be woken up.
755 *
756 * wake_const_ops must be called after a semaphore in a semaphore array
757 * was set to 0. If complex const operations are pending, wake_const_ops must
758 * be called with semnum = -1, as well as with the number of each modified
759 * semaphore.
760 * The tasks that must be woken up are added to @pt. The return code
761 * is stored in q->pid.
762 * The function returns 1 if at least one operation was completed successfully.
763 */
764static int wake_const_ops(struct sem_array *sma, int semnum,
765				struct list_head *pt)
766{
767	struct sem_queue *q;
768	struct list_head *walk;
769	struct list_head *pending_list;
770	int semop_completed = 0;
771
772	if (semnum == -1)
773		pending_list = &sma->pending_const;
774	else
775		pending_list = &sma->sem_base[semnum].pending_const;
776
777	walk = pending_list->next;
778	while (walk != pending_list) {
779		int error;
780
781		q = container_of(walk, struct sem_queue, list);
782		walk = walk->next;
783
784		error = perform_atomic_semop(sma, q->sops, q->nsops,
785						 q->undo, q->pid);
786
787		if (error <= 0) {
788			/* operation completed, remove from queue & wakeup */
789
790			unlink_queue(sma, q);
791
792			wake_up_sem_queue_prepare(pt, q, error);
793			if (error == 0)
794				semop_completed = 1;
795		}
796	}
797	return semop_completed;
798}
799
800/**
801 * do_smart_wakeup_zero(sma, sops, nsops, pt) - wakeup all wait for zero tasks
802 * @sma: semaphore array
803 * @sops: operations that were performed
804 * @nsops: number of operations
805 * @pt: list head of the tasks that must be woken up.
806 *
807 * do_smart_wakeup_zero() checks all required queue for wait-for-zero
808 * operations, based on the actual changes that were performed on the
809 * semaphore array.
810 * The function returns 1 if at least one operation was completed successfully.
811 */
812static int do_smart_wakeup_zero(struct sem_array *sma, struct sembuf *sops,
813					int nsops, struct list_head *pt)
814{
815	int i;
816	int semop_completed = 0;
817	int got_zero = 0;
818
819	/* first: the per-semaphore queues, if known */
820	if (sops) {
821		for (i = 0; i < nsops; i++) {
822			int num = sops[i].sem_num;
823
824			if (sma->sem_base[num].semval == 0) {
825				got_zero = 1;
826				semop_completed |= wake_const_ops(sma, num, pt);
827			}
828		}
829	} else {
830		/*
831		 * No sops means modified semaphores not known.
832		 * Assume all were changed.
833		 */
834		for (i = 0; i < sma->sem_nsems; i++) {
835			if (sma->sem_base[i].semval == 0) {
836				got_zero = 1;
837				semop_completed |= wake_const_ops(sma, i, pt);
838			}
839		}
840	}
841	/*
842	 * If one of the modified semaphores got 0,
843	 * then check the global queue, too.
844	 */
845	if (got_zero)
846		semop_completed |= wake_const_ops(sma, -1, pt);
847
848	return semop_completed;
849}
850
851
852/**
853 * update_queue(sma, semnum): Look for tasks that can be completed.
854 * @sma: semaphore array.
855 * @semnum: semaphore that was modified.
856 * @pt: list head for the tasks that must be woken up.
857 *
858 * update_queue must be called after a semaphore in a semaphore array
859 * was modified. If multiple semaphores were modified, update_queue must
860 * be called with semnum = -1, as well as with the number of each modified
861 * semaphore.
862 * The tasks that must be woken up are added to @pt. The return code
863 * is stored in q->pid.
864 * The function internally checks if const operations can now succeed.
865 *
866 * The function return 1 if at least one semop was completed successfully.
867 */
868static int update_queue(struct sem_array *sma, int semnum, struct list_head *pt)
869{
870	struct sem_queue *q;
871	struct list_head *walk;
872	struct list_head *pending_list;
873	int semop_completed = 0;
874
875	if (semnum == -1)
876		pending_list = &sma->pending_alter;
877	else
878		pending_list = &sma->sem_base[semnum].pending_alter;
879
880again:
881	walk = pending_list->next;
882	while (walk != pending_list) {
883		int error, restart;
884
885		q = container_of(walk, struct sem_queue, list);
886		walk = walk->next;
887
888		/* If we are scanning the single sop, per-semaphore list of
889		 * one semaphore and that semaphore is 0, then it is not
890		 * necessary to scan further: simple increments
891		 * that affect only one entry succeed immediately and cannot
892		 * be in the  per semaphore pending queue, and decrements
893		 * cannot be successful if the value is already 0.
894		 */
895		if (semnum != -1 && sma->sem_base[semnum].semval == 0)
896			break;
897
898		error = perform_atomic_semop(sma, q->sops, q->nsops,
899					 q->undo, q->pid);
900
901		/* Does q->sleeper still need to sleep? */
902		if (error > 0)
903			continue;
904
905		unlink_queue(sma, q);
906
907		if (error) {
908			restart = 0;
909		} else {
910			semop_completed = 1;
911			do_smart_wakeup_zero(sma, q->sops, q->nsops, pt);
912			restart = check_restart(sma, q);
913		}
914
915		wake_up_sem_queue_prepare(pt, q, error);
916		if (restart)
917			goto again;
918	}
919	return semop_completed;
920}
921
922/**
923 * set_semotime(sma, sops) - set sem_otime
924 * @sma: semaphore array
925 * @sops: operations that modified the array, may be NULL
926 *
927 * sem_otime is replicated to avoid cache line trashing.
928 * This function sets one instance to the current time.
929 */
930static void set_semotime(struct sem_array *sma, struct sembuf *sops)
931{
932	if (sops == NULL) {
933		sma->sem_base[0].sem_otime = get_seconds();
934	} else {
935		sma->sem_base[sops[0].sem_num].sem_otime =
936							get_seconds();
937	}
938}
939
940/**
941 * do_smart_update(sma, sops, nsops, otime, pt) - optimized update_queue
942 * @sma: semaphore array
943 * @sops: operations that were performed
944 * @nsops: number of operations
945 * @otime: force setting otime
946 * @pt: list head of the tasks that must be woken up.
947 *
948 * do_smart_update() does the required calls to update_queue and wakeup_zero,
949 * based on the actual changes that were performed on the semaphore array.
950 * Note that the function does not do the actual wake-up: the caller is
951 * responsible for calling wake_up_sem_queue_do(@pt).
952 * It is safe to perform this call after dropping all locks.
953 */
954static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops,
955			int otime, struct list_head *pt)
956{
957	int i;
958
959	otime |= do_smart_wakeup_zero(sma, sops, nsops, pt);
960
961	if (!list_empty(&sma->pending_alter)) {
962		/* semaphore array uses the global queue - just process it. */
963		otime |= update_queue(sma, -1, pt);
964	} else {
965		if (!sops) {
966			/*
967			 * No sops, thus the modified semaphores are not
968			 * known. Check all.
969			 */
970			for (i = 0; i < sma->sem_nsems; i++)
971				otime |= update_queue(sma, i, pt);
972		} else {
973			/*
974			 * Check the semaphores that were increased:
975			 * - No complex ops, thus all sleeping ops are
976			 *   decrease.
977			 * - if we decreased the value, then any sleeping
978			 *   semaphore ops wont be able to run: If the
979			 *   previous value was too small, then the new
980			 *   value will be too small, too.
981			 */
982			for (i = 0; i < nsops; i++) {
983				if (sops[i].sem_op > 0) {
984					otime |= update_queue(sma,
985							sops[i].sem_num, pt);
986				}
987			}
988		}
989	}
990	if (otime)
991		set_semotime(sma, sops);
992}
993
994/* The following counts are associated to each semaphore:
995 *   semncnt        number of tasks waiting on semval being nonzero
996 *   semzcnt        number of tasks waiting on semval being zero
997 * This model assumes that a task waits on exactly one semaphore.
998 * Since semaphore operations are to be performed atomically, tasks actually
999 * wait on a whole sequence of semaphores simultaneously.
1000 * The counts we return here are a rough approximation, but still
1001 * warrant that semncnt+semzcnt>0 if the task is on the pending queue.
1002 */
1003static int count_semncnt(struct sem_array *sma, ushort semnum)
1004{
1005	int semncnt;
1006	struct sem_queue *q;
1007
1008	semncnt = 0;
1009	list_for_each_entry(q, &sma->sem_base[semnum].pending_alter, list) {
1010		struct sembuf *sops = q->sops;
1011		BUG_ON(sops->sem_num != semnum);
1012		if ((sops->sem_op < 0) && !(sops->sem_flg & IPC_NOWAIT))
1013			semncnt++;
1014	}
1015
1016	list_for_each_entry(q, &sma->pending_alter, list) {
1017		struct sembuf *sops = q->sops;
1018		int nsops = q->nsops;
1019		int i;
1020		for (i = 0; i < nsops; i++)
1021			if (sops[i].sem_num == semnum
1022			    && (sops[i].sem_op < 0)
1023			    && !(sops[i].sem_flg & IPC_NOWAIT))
1024				semncnt++;
1025	}
1026	return semncnt;
1027}
1028
1029static int count_semzcnt(struct sem_array *sma, ushort semnum)
1030{
1031	int semzcnt;
1032	struct sem_queue *q;
1033
1034	semzcnt = 0;
1035	list_for_each_entry(q, &sma->sem_base[semnum].pending_const, list) {
1036		struct sembuf *sops = q->sops;
1037		BUG_ON(sops->sem_num != semnum);
1038		if ((sops->sem_op == 0) && !(sops->sem_flg & IPC_NOWAIT))
1039			semzcnt++;
1040	}
1041
1042	list_for_each_entry(q, &sma->pending_const, list) {
1043		struct sembuf *sops = q->sops;
1044		int nsops = q->nsops;
1045		int i;
1046		for (i = 0; i < nsops; i++)
1047			if (sops[i].sem_num == semnum
1048			    && (sops[i].sem_op == 0)
1049			    && !(sops[i].sem_flg & IPC_NOWAIT))
1050				semzcnt++;
1051	}
1052	return semzcnt;
1053}
1054
1055/* Free a semaphore set. freeary() is called with sem_ids.rwsem locked
1056 * as a writer and the spinlock for this semaphore set hold. sem_ids.rwsem
1057 * remains locked on exit.
1058 */
1059static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
1060{
1061	struct sem_undo *un, *tu;
1062	struct sem_queue *q, *tq;
1063	struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
1064	struct list_head tasks;
1065	int i;
1066
1067	/* Free the existing undo structures for this semaphore set.  */
1068	ipc_assert_locked_object(&sma->sem_perm);
1069	list_for_each_entry_safe(un, tu, &sma->list_id, list_id) {
1070		list_del(&un->list_id);
1071		spin_lock(&un->ulp->lock);
1072		un->semid = -1;
1073		list_del_rcu(&un->list_proc);
1074		spin_unlock(&un->ulp->lock);
1075		kfree_rcu(un, rcu);
1076	}
1077
1078	/* Wake up all pending processes and let them fail with EIDRM. */
1079	INIT_LIST_HEAD(&tasks);
1080	list_for_each_entry_safe(q, tq, &sma->pending_const, list) {
1081		unlink_queue(sma, q);
1082		wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1083	}
1084
1085	list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
1086		unlink_queue(sma, q);
1087		wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1088	}
1089	for (i = 0; i < sma->sem_nsems; i++) {
1090		struct sem *sem = sma->sem_base + i;
1091		list_for_each_entry_safe(q, tq, &sem->pending_const, list) {
1092			unlink_queue(sma, q);
1093			wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1094		}
1095		list_for_each_entry_safe(q, tq, &sem->pending_alter, list) {
1096			unlink_queue(sma, q);
1097			wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1098		}
1099	}
1100
1101	/* Remove the semaphore set from the IDR */
1102	sem_rmid(ns, sma);
1103	sem_unlock(sma, -1);
1104	rcu_read_unlock();
1105
1106	wake_up_sem_queue_do(&tasks);
1107	ns->used_sems -= sma->sem_nsems;
1108	ipc_rcu_putref(sma, sem_rcu_free);
1109}
1110
1111static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
1112{
1113	switch (version) {
1114	case IPC_64:
1115		return copy_to_user(buf, in, sizeof(*in));
1116	case IPC_OLD:
1117	    {
1118		struct semid_ds out;
1119
1120		memset(&out, 0, sizeof(out));
1121
1122		ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
1123
1124		out.sem_otime	= in->sem_otime;
1125		out.sem_ctime	= in->sem_ctime;
1126		out.sem_nsems	= in->sem_nsems;
1127
1128		return copy_to_user(buf, &out, sizeof(out));
1129	    }
1130	default:
1131		return -EINVAL;
1132	}
1133}
1134
1135static time_t get_semotime(struct sem_array *sma)
1136{
1137	int i;
1138	time_t res;
1139
1140	res = sma->sem_base[0].sem_otime;
1141	for (i = 1; i < sma->sem_nsems; i++) {
1142		time_t to = sma->sem_base[i].sem_otime;
1143
1144		if (to > res)
1145			res = to;
1146	}
1147	return res;
1148}
1149
1150static int semctl_nolock(struct ipc_namespace *ns, int semid,
1151			 int cmd, int version, void __user *p)
1152{
1153	int err;
1154	struct sem_array *sma;
1155
1156	switch (cmd) {
1157	case IPC_INFO:
1158	case SEM_INFO:
1159	{
1160		struct seminfo seminfo;
1161		int max_id;
1162
1163		err = security_sem_semctl(NULL, cmd);
1164		if (err)
1165			return err;
1166
1167		memset(&seminfo, 0, sizeof(seminfo));
1168		seminfo.semmni = ns->sc_semmni;
1169		seminfo.semmns = ns->sc_semmns;
1170		seminfo.semmsl = ns->sc_semmsl;
1171		seminfo.semopm = ns->sc_semopm;
1172		seminfo.semvmx = SEMVMX;
1173		seminfo.semmnu = SEMMNU;
1174		seminfo.semmap = SEMMAP;
1175		seminfo.semume = SEMUME;
1176		down_read(&sem_ids(ns).rwsem);
1177		if (cmd == SEM_INFO) {
1178			seminfo.semusz = sem_ids(ns).in_use;
1179			seminfo.semaem = ns->used_sems;
1180		} else {
1181			seminfo.semusz = SEMUSZ;
1182			seminfo.semaem = SEMAEM;
1183		}
1184		max_id = ipc_get_maxid(&sem_ids(ns));
1185		up_read(&sem_ids(ns).rwsem);
1186		if (copy_to_user(p, &seminfo, sizeof(struct seminfo)))
1187			return -EFAULT;
1188		return (max_id < 0) ? 0 : max_id;
1189	}
1190	case IPC_STAT:
1191	case SEM_STAT:
1192	{
1193		struct semid64_ds tbuf;
1194		int id = 0;
1195
1196		memset(&tbuf, 0, sizeof(tbuf));
1197
1198		rcu_read_lock();
1199		if (cmd == SEM_STAT) {
1200			sma = sem_obtain_object(ns, semid);
1201			if (IS_ERR(sma)) {
1202				err = PTR_ERR(sma);
1203				goto out_unlock;
1204			}
1205			id = sma->sem_perm.id;
1206		} else {
1207			sma = sem_obtain_object_check(ns, semid);
1208			if (IS_ERR(sma)) {
1209				err = PTR_ERR(sma);
1210				goto out_unlock;
1211			}
1212		}
1213
1214		err = -EACCES;
1215		if (ipcperms(ns, &sma->sem_perm, S_IRUGO))
1216			goto out_unlock;
1217
1218		err = security_sem_semctl(sma, cmd);
1219		if (err)
1220			goto out_unlock;
1221
1222		kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
1223		tbuf.sem_otime = get_semotime(sma);
1224		tbuf.sem_ctime = sma->sem_ctime;
1225		tbuf.sem_nsems = sma->sem_nsems;
1226		rcu_read_unlock();
1227		if (copy_semid_to_user(p, &tbuf, version))
1228			return -EFAULT;
1229		return id;
1230	}
1231	default:
1232		return -EINVAL;
1233	}
1234out_unlock:
1235	rcu_read_unlock();
1236	return err;
1237}
1238
1239static int semctl_setval(struct ipc_namespace *ns, int semid, int semnum,
1240		unsigned long arg)
1241{
1242	struct sem_undo *un;
1243	struct sem_array *sma;
1244	struct sem *curr;
1245	int err;
1246	struct list_head tasks;
1247	int val;
1248#if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN)
1249	/* big-endian 64bit */
1250	val = arg >> 32;
1251#else
1252	/* 32bit or little-endian 64bit */
1253	val = arg;
1254#endif
1255
1256	if (val > SEMVMX || val < 0)
1257		return -ERANGE;
1258
1259	INIT_LIST_HEAD(&tasks);
1260
1261	rcu_read_lock();
1262	sma = sem_obtain_object_check(ns, semid);
1263	if (IS_ERR(sma)) {
1264		rcu_read_unlock();
1265		return PTR_ERR(sma);
1266	}
1267
1268	if (semnum < 0 || semnum >= sma->sem_nsems) {
1269		rcu_read_unlock();
1270		return -EINVAL;
1271	}
1272
1273
1274	if (ipcperms(ns, &sma->sem_perm, S_IWUGO)) {
1275		rcu_read_unlock();
1276		return -EACCES;
1277	}
1278
1279	err = security_sem_semctl(sma, SETVAL);
1280	if (err) {
1281		rcu_read_unlock();
1282		return -EACCES;
1283	}
1284
1285	sem_lock(sma, NULL, -1);
1286
1287	if (!ipc_valid_object(&sma->sem_perm)) {
1288		sem_unlock(sma, -1);
1289		rcu_read_unlock();
1290		return -EIDRM;
1291	}
1292
1293	curr = &sma->sem_base[semnum];
1294
1295	ipc_assert_locked_object(&sma->sem_perm);
1296	list_for_each_entry(un, &sma->list_id, list_id)
1297		un->semadj[semnum] = 0;
1298
1299	curr->semval = val;
1300	curr->sempid = task_tgid_vnr(current);
1301	sma->sem_ctime = get_seconds();
1302	/* maybe some queued-up processes were waiting for this */
1303	do_smart_update(sma, NULL, 0, 0, &tasks);
1304	sem_unlock(sma, -1);
1305	rcu_read_unlock();
1306	wake_up_sem_queue_do(&tasks);
1307	return 0;
1308}
1309
1310static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
1311		int cmd, void __user *p)
1312{
1313	struct sem_array *sma;
1314	struct sem *curr;
1315	int err, nsems;
1316	ushort fast_sem_io[SEMMSL_FAST];
1317	ushort *sem_io = fast_sem_io;
1318	struct list_head tasks;
1319
1320	INIT_LIST_HEAD(&tasks);
1321
1322	rcu_read_lock();
1323	sma = sem_obtain_object_check(ns, semid);
1324	if (IS_ERR(sma)) {
1325		rcu_read_unlock();
1326		return PTR_ERR(sma);
1327	}
1328
1329	nsems = sma->sem_nsems;
1330
1331	err = -EACCES;
1332	if (ipcperms(ns, &sma->sem_perm, cmd == SETALL ? S_IWUGO : S_IRUGO))
1333		goto out_rcu_wakeup;
1334
1335	err = security_sem_semctl(sma, cmd);
1336	if (err)
1337		goto out_rcu_wakeup;
1338
1339	err = -EACCES;
1340	switch (cmd) {
1341	case GETALL:
1342	{
1343		ushort __user *array = p;
1344		int i;
1345
1346		sem_lock(sma, NULL, -1);
1347		if (!ipc_valid_object(&sma->sem_perm)) {
1348			err = -EIDRM;
1349			goto out_unlock;
1350		}
1351		if (nsems > SEMMSL_FAST) {
1352			if (!ipc_rcu_getref(sma)) {
1353				err = -EIDRM;
1354				goto out_unlock;
1355			}
1356			sem_unlock(sma, -1);
1357			rcu_read_unlock();
1358			sem_io = ipc_alloc(sizeof(ushort)*nsems);
1359			if (sem_io == NULL) {
1360				ipc_rcu_putref(sma, ipc_rcu_free);
1361				return -ENOMEM;
1362			}
1363
1364			rcu_read_lock();
1365			sem_lock_and_putref(sma);
1366			if (!ipc_valid_object(&sma->sem_perm)) {
1367				err = -EIDRM;
1368				goto out_unlock;
1369			}
1370		}
1371		for (i = 0; i < sma->sem_nsems; i++)
1372			sem_io[i] = sma->sem_base[i].semval;
1373		sem_unlock(sma, -1);
1374		rcu_read_unlock();
1375		err = 0;
1376		if (copy_to_user(array, sem_io, nsems*sizeof(ushort)))
1377			err = -EFAULT;
1378		goto out_free;
1379	}
1380	case SETALL:
1381	{
1382		int i;
1383		struct sem_undo *un;
1384
1385		if (!ipc_rcu_getref(sma)) {
1386			err = -EIDRM;
1387			goto out_rcu_wakeup;
1388		}
1389		rcu_read_unlock();
1390
1391		if (nsems > SEMMSL_FAST) {
1392			sem_io = ipc_alloc(sizeof(ushort)*nsems);
1393			if (sem_io == NULL) {
1394				ipc_rcu_putref(sma, ipc_rcu_free);
1395				return -ENOMEM;
1396			}
1397		}
1398
1399		if (copy_from_user(sem_io, p, nsems*sizeof(ushort))) {
1400			ipc_rcu_putref(sma, ipc_rcu_free);
1401			err = -EFAULT;
1402			goto out_free;
1403		}
1404
1405		for (i = 0; i < nsems; i++) {
1406			if (sem_io[i] > SEMVMX) {
1407				ipc_rcu_putref(sma, ipc_rcu_free);
1408				err = -ERANGE;
1409				goto out_free;
1410			}
1411		}
1412		rcu_read_lock();
1413		sem_lock_and_putref(sma);
1414		if (!ipc_valid_object(&sma->sem_perm)) {
1415			err = -EIDRM;
1416			goto out_unlock;
1417		}
1418
1419		for (i = 0; i < nsems; i++)
1420			sma->sem_base[i].semval = sem_io[i];
1421
1422		ipc_assert_locked_object(&sma->sem_perm);
1423		list_for_each_entry(un, &sma->list_id, list_id) {
1424			for (i = 0; i < nsems; i++)
1425				un->semadj[i] = 0;
1426		}
1427		sma->sem_ctime = get_seconds();
1428		/* maybe some queued-up processes were waiting for this */
1429		do_smart_update(sma, NULL, 0, 0, &tasks);
1430		err = 0;
1431		goto out_unlock;
1432	}
1433	/* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */
1434	}
1435	err = -EINVAL;
1436	if (semnum < 0 || semnum >= nsems)
1437		goto out_rcu_wakeup;
1438
1439	sem_lock(sma, NULL, -1);
1440	if (!ipc_valid_object(&sma->sem_perm)) {
1441		err = -EIDRM;
1442		goto out_unlock;
1443	}
1444	curr = &sma->sem_base[semnum];
1445
1446	switch (cmd) {
1447	case GETVAL:
1448		err = curr->semval;
1449		goto out_unlock;
1450	case GETPID:
1451		err = curr->sempid;
1452		goto out_unlock;
1453	case GETNCNT:
1454		err = count_semncnt(sma, semnum);
1455		goto out_unlock;
1456	case GETZCNT:
1457		err = count_semzcnt(sma, semnum);
1458		goto out_unlock;
1459	}
1460
1461out_unlock:
1462	sem_unlock(sma, -1);
1463out_rcu_wakeup:
1464	rcu_read_unlock();
1465	wake_up_sem_queue_do(&tasks);
1466out_free:
1467	if (sem_io != fast_sem_io)
1468		ipc_free(sem_io, sizeof(ushort)*nsems);
1469	return err;
1470}
1471
1472static inline unsigned long
1473copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
1474{
1475	switch (version) {
1476	case IPC_64:
1477		if (copy_from_user(out, buf, sizeof(*out)))
1478			return -EFAULT;
1479		return 0;
1480	case IPC_OLD:
1481	    {
1482		struct semid_ds tbuf_old;
1483
1484		if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
1485			return -EFAULT;
1486
1487		out->sem_perm.uid	= tbuf_old.sem_perm.uid;
1488		out->sem_perm.gid	= tbuf_old.sem_perm.gid;
1489		out->sem_perm.mode	= tbuf_old.sem_perm.mode;
1490
1491		return 0;
1492	    }
1493	default:
1494		return -EINVAL;
1495	}
1496}
1497
1498/*
1499 * This function handles some semctl commands which require the rwsem
1500 * to be held in write mode.
1501 * NOTE: no locks must be held, the rwsem is taken inside this function.
1502 */
1503static int semctl_down(struct ipc_namespace *ns, int semid,
1504		       int cmd, int version, void __user *p)
1505{
1506	struct sem_array *sma;
1507	int err;
1508	struct semid64_ds semid64;
1509	struct kern_ipc_perm *ipcp;
1510
1511	if (cmd == IPC_SET) {
1512		if (copy_semid_from_user(&semid64, p, version))
1513			return -EFAULT;
1514	}
1515
1516	down_write(&sem_ids(ns).rwsem);
1517	rcu_read_lock();
1518
1519	ipcp = ipcctl_pre_down_nolock(ns, &sem_ids(ns), semid, cmd,
1520				      &semid64.sem_perm, 0);
1521	if (IS_ERR(ipcp)) {
1522		err = PTR_ERR(ipcp);
1523		goto out_unlock1;
1524	}
1525
1526	sma = container_of(ipcp, struct sem_array, sem_perm);
1527
1528	err = security_sem_semctl(sma, cmd);
1529	if (err)
1530		goto out_unlock1;
1531
1532	switch (cmd) {
1533	case IPC_RMID:
1534		sem_lock(sma, NULL, -1);
1535		/* freeary unlocks the ipc object and rcu */
1536		freeary(ns, ipcp);
1537		goto out_up;
1538	case IPC_SET:
1539		sem_lock(sma, NULL, -1);
1540		err = ipc_update_perm(&semid64.sem_perm, ipcp);
1541		if (err)
1542			goto out_unlock0;
1543		sma->sem_ctime = get_seconds();
1544		break;
1545	default:
1546		err = -EINVAL;
1547		goto out_unlock1;
1548	}
1549
1550out_unlock0:
1551	sem_unlock(sma, -1);
1552out_unlock1:
1553	rcu_read_unlock();
1554out_up:
1555	up_write(&sem_ids(ns).rwsem);
1556	return err;
1557}
1558
1559SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, unsigned long, arg)
1560{
1561	int version;
1562	struct ipc_namespace *ns;
1563	void __user *p = (void __user *)arg;
1564
1565	if (semid < 0)
1566		return -EINVAL;
1567
1568	version = ipc_parse_version(&cmd);
1569	ns = current->nsproxy->ipc_ns;
1570
1571	switch (cmd) {
1572	case IPC_INFO:
1573	case SEM_INFO:
1574	case IPC_STAT:
1575	case SEM_STAT:
1576		return semctl_nolock(ns, semid, cmd, version, p);
1577	case GETALL:
1578	case GETVAL:
1579	case GETPID:
1580	case GETNCNT:
1581	case GETZCNT:
1582	case SETALL:
1583		return semctl_main(ns, semid, semnum, cmd, p);
1584	case SETVAL:
1585		return semctl_setval(ns, semid, semnum, arg);
1586	case IPC_RMID:
1587	case IPC_SET:
1588		return semctl_down(ns, semid, cmd, version, p);
1589	default:
1590		return -EINVAL;
1591	}
1592}
1593
1594/* If the task doesn't already have a undo_list, then allocate one
1595 * here.  We guarantee there is only one thread using this undo list,
1596 * and current is THE ONE
1597 *
1598 * If this allocation and assignment succeeds, but later
1599 * portions of this code fail, there is no need to free the sem_undo_list.
1600 * Just let it stay associated with the task, and it'll be freed later
1601 * at exit time.
1602 *
1603 * This can block, so callers must hold no locks.
1604 */
1605static inline int get_undo_list(struct sem_undo_list **undo_listp)
1606{
1607	struct sem_undo_list *undo_list;
1608
1609	undo_list = current->sysvsem.undo_list;
1610	if (!undo_list) {
1611		undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
1612		if (undo_list == NULL)
1613			return -ENOMEM;
1614		spin_lock_init(&undo_list->lock);
1615		atomic_set(&undo_list->refcnt, 1);
1616		INIT_LIST_HEAD(&undo_list->list_proc);
1617
1618		current->sysvsem.undo_list = undo_list;
1619	}
1620	*undo_listp = undo_list;
1621	return 0;
1622}
1623
1624static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid)
1625{
1626	struct sem_undo *un;
1627
1628	list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) {
1629		if (un->semid == semid)
1630			return un;
1631	}
1632	return NULL;
1633}
1634
1635static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1636{
1637	struct sem_undo *un;
1638
1639	assert_spin_locked(&ulp->lock);
1640
1641	un = __lookup_undo(ulp, semid);
1642	if (un) {
1643		list_del_rcu(&un->list_proc);
1644		list_add_rcu(&un->list_proc, &ulp->list_proc);
1645	}
1646	return un;
1647}
1648
1649/**
1650 * find_alloc_undo - Lookup (and if not present create) undo array
1651 * @ns: namespace
1652 * @semid: semaphore array id
1653 *
1654 * The function looks up (and if not present creates) the undo structure.
1655 * The size of the undo structure depends on the size of the semaphore
1656 * array, thus the alloc path is not that straightforward.
1657 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1658 * performs a rcu_read_lock().
1659 */
1660static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
1661{
1662	struct sem_array *sma;
1663	struct sem_undo_list *ulp;
1664	struct sem_undo *un, *new;
1665	int nsems, error;
1666
1667	error = get_undo_list(&ulp);
1668	if (error)
1669		return ERR_PTR(error);
1670
1671	rcu_read_lock();
1672	spin_lock(&ulp->lock);
1673	un = lookup_undo(ulp, semid);
1674	spin_unlock(&ulp->lock);
1675	if (likely(un != NULL))
1676		goto out;
1677
1678	/* no undo structure around - allocate one. */
1679	/* step 1: figure out the size of the semaphore array */
1680	sma = sem_obtain_object_check(ns, semid);
1681	if (IS_ERR(sma)) {
1682		rcu_read_unlock();
1683		return ERR_CAST(sma);
1684	}
1685
1686	nsems = sma->sem_nsems;
1687	if (!ipc_rcu_getref(sma)) {
1688		rcu_read_unlock();
1689		un = ERR_PTR(-EIDRM);
1690		goto out;
1691	}
1692	rcu_read_unlock();
1693
1694	/* step 2: allocate new undo structure */
1695	new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1696	if (!new) {
1697		ipc_rcu_putref(sma, ipc_rcu_free);
1698		return ERR_PTR(-ENOMEM);
1699	}
1700
1701	/* step 3: Acquire the lock on semaphore array */
1702	rcu_read_lock();
1703	sem_lock_and_putref(sma);
1704	if (!ipc_valid_object(&sma->sem_perm)) {
1705		sem_unlock(sma, -1);
1706		rcu_read_unlock();
1707		kfree(new);
1708		un = ERR_PTR(-EIDRM);
1709		goto out;
1710	}
1711	spin_lock(&ulp->lock);
1712
1713	/*
1714	 * step 4: check for races: did someone else allocate the undo struct?
1715	 */
1716	un = lookup_undo(ulp, semid);
1717	if (un) {
1718		kfree(new);
1719		goto success;
1720	}
1721	/* step 5: initialize & link new undo structure */
1722	new->semadj = (short *) &new[1];
1723	new->ulp = ulp;
1724	new->semid = semid;
1725	assert_spin_locked(&ulp->lock);
1726	list_add_rcu(&new->list_proc, &ulp->list_proc);
1727	ipc_assert_locked_object(&sma->sem_perm);
1728	list_add(&new->list_id, &sma->list_id);
1729	un = new;
1730
1731success:
1732	spin_unlock(&ulp->lock);
1733	sem_unlock(sma, -1);
1734out:
1735	return un;
1736}
1737
1738
1739/**
1740 * get_queue_result - Retrieve the result code from sem_queue
1741 * @q: Pointer to queue structure
1742 *
1743 * Retrieve the return code from the pending queue. If IN_WAKEUP is found in
1744 * q->status, then we must loop until the value is replaced with the final
1745 * value: This may happen if a task is woken up by an unrelated event (e.g.
1746 * signal) and in parallel the task is woken up by another task because it got
1747 * the requested semaphores.
1748 *
1749 * The function can be called with or without holding the semaphore spinlock.
1750 */
1751static int get_queue_result(struct sem_queue *q)
1752{
1753	int error;
1754
1755	error = q->status;
1756	while (unlikely(error == IN_WAKEUP)) {
1757		cpu_relax();
1758		error = q->status;
1759	}
1760
1761	return error;
1762}
1763
1764SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
1765		unsigned, nsops, const struct timespec __user *, timeout)
1766{
1767	int error = -EINVAL;
1768	struct sem_array *sma;
1769	struct sembuf fast_sops[SEMOPM_FAST];
1770	struct sembuf *sops = fast_sops, *sop;
1771	struct sem_undo *un;
1772	int undos = 0, alter = 0, max, locknum;
1773	struct sem_queue queue;
1774	unsigned long jiffies_left = 0;
1775	struct ipc_namespace *ns;
1776	struct list_head tasks;
1777
1778	ns = current->nsproxy->ipc_ns;
1779
1780	if (nsops < 1 || semid < 0)
1781		return -EINVAL;
1782	if (nsops > ns->sc_semopm)
1783		return -E2BIG;
1784	if (nsops > SEMOPM_FAST) {
1785		sops = kmalloc(sizeof(*sops)*nsops, GFP_KERNEL);
1786		if (sops == NULL)
1787			return -ENOMEM;
1788	}
1789	if (copy_from_user(sops, tsops, nsops * sizeof(*tsops))) {
1790		error =  -EFAULT;
1791		goto out_free;
1792	}
1793	if (timeout) {
1794		struct timespec _timeout;
1795		if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1796			error = -EFAULT;
1797			goto out_free;
1798		}
1799		if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1800			_timeout.tv_nsec >= 1000000000L) {
1801			error = -EINVAL;
1802			goto out_free;
1803		}
1804		jiffies_left = timespec_to_jiffies(&_timeout);
1805	}
1806	max = 0;
1807	for (sop = sops; sop < sops + nsops; sop++) {
1808		if (sop->sem_num >= max)
1809			max = sop->sem_num;
1810		if (sop->sem_flg & SEM_UNDO)
1811			undos = 1;
1812		if (sop->sem_op != 0)
1813			alter = 1;
1814	}
1815
1816	INIT_LIST_HEAD(&tasks);
1817
1818	if (undos) {
1819		/* On success, find_alloc_undo takes the rcu_read_lock */
1820		un = find_alloc_undo(ns, semid);
1821		if (IS_ERR(un)) {
1822			error = PTR_ERR(un);
1823			goto out_free;
1824		}
1825	} else {
1826		un = NULL;
1827		rcu_read_lock();
1828	}
1829
1830	sma = sem_obtain_object_check(ns, semid);
1831	if (IS_ERR(sma)) {
1832		rcu_read_unlock();
1833		error = PTR_ERR(sma);
1834		goto out_free;
1835	}
1836
1837	error = -EFBIG;
1838	if (max >= sma->sem_nsems)
1839		goto out_rcu_wakeup;
1840
1841	error = -EACCES;
1842	if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO))
1843		goto out_rcu_wakeup;
1844
1845	error = security_sem_semop(sma, sops, nsops, alter);
1846	if (error)
1847		goto out_rcu_wakeup;
1848
1849	error = -EIDRM;
1850	locknum = sem_lock(sma, sops, nsops);
1851	/*
1852	 * We eventually might perform the following check in a lockless
1853	 * fashion, considering ipc_valid_object() locking constraints.
1854	 * If nsops == 1 and there is no contention for sem_perm.lock, then
1855	 * only a per-semaphore lock is held and it's OK to proceed with the
1856	 * check below. More details on the fine grained locking scheme
1857	 * entangled here and why it's RMID race safe on comments at sem_lock()
1858	 */
1859	if (!ipc_valid_object(&sma->sem_perm))
1860		goto out_unlock_free;
1861	/*
1862	 * semid identifiers are not unique - find_alloc_undo may have
1863	 * allocated an undo structure, it was invalidated by an RMID
1864	 * and now a new array with received the same id. Check and fail.
1865	 * This case can be detected checking un->semid. The existence of
1866	 * "un" itself is guaranteed by rcu.
1867	 */
1868	if (un && un->semid == -1)
1869		goto out_unlock_free;
1870
1871	error = perform_atomic_semop(sma, sops, nsops, un,
1872					task_tgid_vnr(current));
1873	if (error == 0) {
1874		/* If the operation was successful, then do
1875		 * the required updates.
1876		 */
1877		if (alter)
1878			do_smart_update(sma, sops, nsops, 1, &tasks);
1879		else
1880			set_semotime(sma, sops);
1881	}
1882	if (error <= 0)
1883		goto out_unlock_free;
1884
1885	/* We need to sleep on this operation, so we put the current
1886	 * task into the pending queue and go to sleep.
1887	 */
1888
1889	queue.sops = sops;
1890	queue.nsops = nsops;
1891	queue.undo = un;
1892	queue.pid = task_tgid_vnr(current);
1893	queue.alter = alter;
1894
1895	if (nsops == 1) {
1896		struct sem *curr;
1897		curr = &sma->sem_base[sops->sem_num];
1898
1899		if (alter) {
1900			if (sma->complex_count) {
1901				list_add_tail(&queue.list,
1902						&sma->pending_alter);
1903			} else {
1904
1905				list_add_tail(&queue.list,
1906						&curr->pending_alter);
1907			}
1908		} else {
1909			list_add_tail(&queue.list, &curr->pending_const);
1910		}
1911	} else {
1912		if (!sma->complex_count)
1913			merge_queues(sma);
1914
1915		if (alter)
1916			list_add_tail(&queue.list, &sma->pending_alter);
1917		else
1918			list_add_tail(&queue.list, &sma->pending_const);
1919
1920		sma->complex_count++;
1921	}
1922
1923	queue.status = -EINTR;
1924	queue.sleeper = current;
1925
1926sleep_again:
1927	current->state = TASK_INTERRUPTIBLE;
1928	sem_unlock(sma, locknum);
1929	rcu_read_unlock();
1930
1931	if (timeout)
1932		jiffies_left = schedule_timeout(jiffies_left);
1933	else
1934		schedule();
1935
1936	error = get_queue_result(&queue);
1937
1938	if (error != -EINTR) {
1939		/* fast path: update_queue already obtained all requested
1940		 * resources.
1941		 * Perform a smp_mb(): User space could assume that semop()
1942		 * is a memory barrier: Without the mb(), the cpu could
1943		 * speculatively read in user space stale data that was
1944		 * overwritten by the previous owner of the semaphore.
1945		 */
1946		smp_mb();
1947
1948		goto out_free;
1949	}
1950
1951	rcu_read_lock();
1952	sma = sem_obtain_lock(ns, semid, sops, nsops, &locknum);
1953
1954	/*
1955	 * Wait until it's guaranteed that no wakeup_sem_queue_do() is ongoing.
1956	 */
1957	error = get_queue_result(&queue);
1958
1959	/*
1960	 * Array removed? If yes, leave without sem_unlock().
1961	 */
1962	if (IS_ERR(sma)) {
1963		rcu_read_unlock();
1964		goto out_free;
1965	}
1966
1967
1968	/*
1969	 * If queue.status != -EINTR we are woken up by another process.
1970	 * Leave without unlink_queue(), but with sem_unlock().
1971	 */
1972
1973	if (error != -EINTR) {
1974		goto out_unlock_free;
1975	}
1976
1977	/*
1978	 * If an interrupt occurred we have to clean up the queue
1979	 */
1980	if (timeout && jiffies_left == 0)
1981		error = -EAGAIN;
1982
1983	/*
1984	 * If the wakeup was spurious, just retry
1985	 */
1986	if (error == -EINTR && !signal_pending(current))
1987		goto sleep_again;
1988
1989	unlink_queue(sma, &queue);
1990
1991out_unlock_free:
1992	sem_unlock(sma, locknum);
1993out_rcu_wakeup:
1994	rcu_read_unlock();
1995	wake_up_sem_queue_do(&tasks);
1996out_free:
1997	if (sops != fast_sops)
1998		kfree(sops);
1999	return error;
2000}
2001
2002SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
2003		unsigned, nsops)
2004{
2005	return sys_semtimedop(semid, tsops, nsops, NULL);
2006}
2007
2008/* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
2009 * parent and child tasks.
2010 */
2011
2012int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
2013{
2014	struct sem_undo_list *undo_list;
2015	int error;
2016
2017	if (clone_flags & CLONE_SYSVSEM) {
2018		error = get_undo_list(&undo_list);
2019		if (error)
2020			return error;
2021		atomic_inc(&undo_list->refcnt);
2022		tsk->sysvsem.undo_list = undo_list;
2023	} else
2024		tsk->sysvsem.undo_list = NULL;
2025
2026	return 0;
2027}
2028
2029/*
2030 * add semadj values to semaphores, free undo structures.
2031 * undo structures are not freed when semaphore arrays are destroyed
2032 * so some of them may be out of date.
2033 * IMPLEMENTATION NOTE: There is some confusion over whether the
2034 * set of adjustments that needs to be done should be done in an atomic
2035 * manner or not. That is, if we are attempting to decrement the semval
2036 * should we queue up and wait until we can do so legally?
2037 * The original implementation attempted to do this (queue and wait).
2038 * The current implementation does not do so. The POSIX standard
2039 * and SVID should be consulted to determine what behavior is mandated.
2040 */
2041void exit_sem(struct task_struct *tsk)
2042{
2043	struct sem_undo_list *ulp;
2044
2045	ulp = tsk->sysvsem.undo_list;
2046	if (!ulp)
2047		return;
2048	tsk->sysvsem.undo_list = NULL;
2049
2050	if (!atomic_dec_and_test(&ulp->refcnt))
2051		return;
2052
2053	for (;;) {
2054		struct sem_array *sma;
2055		struct sem_undo *un;
2056		struct list_head tasks;
2057		int semid, i;
2058
2059		rcu_read_lock();
2060		un = list_entry_rcu(ulp->list_proc.next,
2061				    struct sem_undo, list_proc);
2062		if (&un->list_proc == &ulp->list_proc)
2063			semid = -1;
2064		 else
2065			semid = un->semid;
2066
2067		if (semid == -1) {
2068			rcu_read_unlock();
2069			break;
2070		}
2071
2072		sma = sem_obtain_object_check(tsk->nsproxy->ipc_ns, un->semid);
2073		/* exit_sem raced with IPC_RMID, nothing to do */
2074		if (IS_ERR(sma)) {
2075			rcu_read_unlock();
2076			continue;
2077		}
2078
2079		sem_lock(sma, NULL, -1);
2080		/* exit_sem raced with IPC_RMID, nothing to do */
2081		if (!ipc_valid_object(&sma->sem_perm)) {
2082			sem_unlock(sma, -1);
2083			rcu_read_unlock();
2084			continue;
2085		}
2086		un = __lookup_undo(ulp, semid);
2087		if (un == NULL) {
2088			/* exit_sem raced with IPC_RMID+semget() that created
2089			 * exactly the same semid. Nothing to do.
2090			 */
2091			sem_unlock(sma, -1);
2092			rcu_read_unlock();
2093			continue;
2094		}
2095
2096		/* remove un from the linked lists */
2097		ipc_assert_locked_object(&sma->sem_perm);
2098		list_del(&un->list_id);
2099
2100		spin_lock(&ulp->lock);
2101		list_del_rcu(&un->list_proc);
2102		spin_unlock(&ulp->lock);
2103
2104		/* perform adjustments registered in un */
2105		for (i = 0; i < sma->sem_nsems; i++) {
2106			struct sem *semaphore = &sma->sem_base[i];
2107			if (un->semadj[i]) {
2108				semaphore->semval += un->semadj[i];
2109				/*
2110				 * Range checks of the new semaphore value,
2111				 * not defined by sus:
2112				 * - Some unices ignore the undo entirely
2113				 *   (e.g. HP UX 11i 11.22, Tru64 V5.1)
2114				 * - some cap the value (e.g. FreeBSD caps
2115				 *   at 0, but doesn't enforce SEMVMX)
2116				 *
2117				 * Linux caps the semaphore value, both at 0
2118				 * and at SEMVMX.
2119				 *
2120				 *	Manfred <manfred@colorfullife.com>
2121				 */
2122				if (semaphore->semval < 0)
2123					semaphore->semval = 0;
2124				if (semaphore->semval > SEMVMX)
2125					semaphore->semval = SEMVMX;
2126				semaphore->sempid = task_tgid_vnr(current);
2127			}
2128		}
2129		/* maybe some queued-up processes were waiting for this */
2130		INIT_LIST_HEAD(&tasks);
2131		do_smart_update(sma, NULL, 0, 1, &tasks);
2132		sem_unlock(sma, -1);
2133		rcu_read_unlock();
2134		wake_up_sem_queue_do(&tasks);
2135
2136		kfree_rcu(un, rcu);
2137	}
2138	kfree(ulp);
2139}
2140
2141#ifdef CONFIG_PROC_FS
2142static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
2143{
2144	struct user_namespace *user_ns = seq_user_ns(s);
2145	struct sem_array *sma = it;
2146	time_t sem_otime;
2147
2148	/*
2149	 * The proc interface isn't aware of sem_lock(), it calls
2150	 * ipc_lock_object() directly (in sysvipc_find_ipc).
2151	 * In order to stay compatible with sem_lock(), we must wait until
2152	 * all simple semop() calls have left their critical regions.
2153	 */
2154	sem_wait_array(sma);
2155
2156	sem_otime = get_semotime(sma);
2157
2158	return seq_printf(s,
2159			  "%10d %10d  %4o %10u %5u %5u %5u %5u %10lu %10lu\n",
2160			  sma->sem_perm.key,
2161			  sma->sem_perm.id,
2162			  sma->sem_perm.mode,
2163			  sma->sem_nsems,
2164			  from_kuid_munged(user_ns, sma->sem_perm.uid),
2165			  from_kgid_munged(user_ns, sma->sem_perm.gid),
2166			  from_kuid_munged(user_ns, sma->sem_perm.cuid),
2167			  from_kgid_munged(user_ns, sma->sem_perm.cgid),
2168			  sem_otime,
2169			  sma->sem_ctime);
2170}
2171#endif
2172