sem.c revision 9ad66ae6
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 sem_pending; /* pending single-sop operations */
99};
100
101/* One queue for each sleeping process in the system. */
102struct sem_queue {
103	struct list_head	list;	 /* queue of pending operations */
104	struct task_struct	*sleeper; /* this process */
105	struct sem_undo		*undo;	 /* undo structure */
106	int			pid;	 /* process id of requesting process */
107	int			status;	 /* completion status of operation */
108	struct sembuf		*sops;	 /* array of pending operations */
109	int			nsops;	 /* number of operations */
110	int			alter;	 /* does *sops alter the array? */
111};
112
113/* Each task has a list of undo requests. They are executed automatically
114 * when the process exits.
115 */
116struct sem_undo {
117	struct list_head	list_proc;	/* per-process list: *
118						 * all undos from one process
119						 * rcu protected */
120	struct rcu_head		rcu;		/* rcu struct for sem_undo */
121	struct sem_undo_list	*ulp;		/* back ptr to sem_undo_list */
122	struct list_head	list_id;	/* per semaphore array list:
123						 * all undos for one array */
124	int			semid;		/* semaphore set identifier */
125	short			*semadj;	/* array of adjustments */
126						/* one per semaphore */
127};
128
129/* sem_undo_list controls shared access to the list of sem_undo structures
130 * that may be shared among all a CLONE_SYSVSEM task group.
131 */
132struct sem_undo_list {
133	atomic_t		refcnt;
134	spinlock_t		lock;
135	struct list_head	list_proc;
136};
137
138
139#define sem_ids(ns)	((ns)->ids[IPC_SEM_IDS])
140
141#define sem_checkid(sma, semid)	ipc_checkid(&sma->sem_perm, semid)
142
143static int newary(struct ipc_namespace *, struct ipc_params *);
144static void freeary(struct ipc_namespace *, struct kern_ipc_perm *);
145#ifdef CONFIG_PROC_FS
146static int sysvipc_sem_proc_show(struct seq_file *s, void *it);
147#endif
148
149#define SEMMSL_FAST	256 /* 512 bytes on stack */
150#define SEMOPM_FAST	64  /* ~ 372 bytes on stack */
151
152/*
153 * linked list protection:
154 *	sem_undo.id_next,
155 *	sem_array.sem_pending{,last},
156 *	sem_array.sem_undo: sem_lock() for read/write
157 *	sem_undo.proc_next: only "current" is allowed to read/write that field.
158 *
159 */
160
161#define sc_semmsl	sem_ctls[0]
162#define sc_semmns	sem_ctls[1]
163#define sc_semopm	sem_ctls[2]
164#define sc_semmni	sem_ctls[3]
165
166void sem_init_ns(struct ipc_namespace *ns)
167{
168	ns->sc_semmsl = SEMMSL;
169	ns->sc_semmns = SEMMNS;
170	ns->sc_semopm = SEMOPM;
171	ns->sc_semmni = SEMMNI;
172	ns->used_sems = 0;
173	ipc_init_ids(&ns->ids[IPC_SEM_IDS]);
174}
175
176#ifdef CONFIG_IPC_NS
177void sem_exit_ns(struct ipc_namespace *ns)
178{
179	free_ipcs(ns, &sem_ids(ns), freeary);
180	idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr);
181}
182#endif
183
184void __init sem_init (void)
185{
186	sem_init_ns(&init_ipc_ns);
187	ipc_init_proc_interface("sysvipc/sem",
188				"       key      semid perms      nsems   uid   gid  cuid  cgid      otime      ctime\n",
189				IPC_SEM_IDS, sysvipc_sem_proc_show);
190}
191
192/*
193 * If the request contains only one semaphore operation, and there are
194 * no complex transactions pending, lock only the semaphore involved.
195 * Otherwise, lock the entire semaphore array, since we either have
196 * multiple semaphores in our own semops, or we need to look at
197 * semaphores from other pending complex operations.
198 *
199 * Carefully guard against sma->complex_count changing between zero
200 * and non-zero while we are spinning for the lock. The value of
201 * sma->complex_count cannot change while we are holding the lock,
202 * so sem_unlock should be fine.
203 *
204 * The global lock path checks that all the local locks have been released,
205 * checking each local lock once. This means that the local lock paths
206 * cannot start their critical sections while the global lock is held.
207 */
208static inline int sem_lock(struct sem_array *sma, struct sembuf *sops,
209			      int nsops)
210{
211	int locknum;
212 again:
213	if (nsops == 1 && !sma->complex_count) {
214		struct sem *sem = sma->sem_base + sops->sem_num;
215
216		/* Lock just the semaphore we are interested in. */
217		spin_lock(&sem->lock);
218
219		/*
220		 * If sma->complex_count was set while we were spinning,
221		 * we may need to look at things we did not lock here.
222		 */
223		if (unlikely(sma->complex_count)) {
224			spin_unlock(&sem->lock);
225			goto lock_array;
226		}
227
228		/*
229		 * Another process is holding the global lock on the
230		 * sem_array; we cannot enter our critical section,
231		 * but have to wait for the global lock to be released.
232		 */
233		if (unlikely(spin_is_locked(&sma->sem_perm.lock))) {
234			spin_unlock(&sem->lock);
235			spin_unlock_wait(&sma->sem_perm.lock);
236			goto again;
237		}
238
239		locknum = sops->sem_num;
240	} else {
241		int i;
242		/*
243		 * Lock the semaphore array, and wait for all of the
244		 * individual semaphore locks to go away.  The code
245		 * above ensures no new single-lock holders will enter
246		 * their critical section while the array lock is held.
247		 */
248 lock_array:
249		ipc_lock_object(&sma->sem_perm);
250		for (i = 0; i < sma->sem_nsems; i++) {
251			struct sem *sem = sma->sem_base + i;
252			spin_unlock_wait(&sem->lock);
253		}
254		locknum = -1;
255	}
256	return locknum;
257}
258
259static inline void sem_unlock(struct sem_array *sma, int locknum)
260{
261	if (locknum == -1) {
262		ipc_unlock_object(&sma->sem_perm);
263	} else {
264		struct sem *sem = sma->sem_base + locknum;
265		spin_unlock(&sem->lock);
266	}
267}
268
269/*
270 * sem_lock_(check_) routines are called in the paths where the rw_mutex
271 * is not held.
272 *
273 * The caller holds the RCU read lock.
274 */
275static inline struct sem_array *sem_obtain_lock(struct ipc_namespace *ns,
276			int id, struct sembuf *sops, int nsops, int *locknum)
277{
278	struct kern_ipc_perm *ipcp;
279	struct sem_array *sma;
280
281	ipcp = ipc_obtain_object(&sem_ids(ns), id);
282	if (IS_ERR(ipcp))
283		return ERR_CAST(ipcp);
284
285	sma = container_of(ipcp, struct sem_array, sem_perm);
286	*locknum = sem_lock(sma, sops, nsops);
287
288	/* ipc_rmid() may have already freed the ID while sem_lock
289	 * was spinning: verify that the structure is still valid
290	 */
291	if (!ipcp->deleted)
292		return container_of(ipcp, struct sem_array, sem_perm);
293
294	sem_unlock(sma, *locknum);
295	return ERR_PTR(-EINVAL);
296}
297
298static inline struct sem_array *sem_obtain_object(struct ipc_namespace *ns, int id)
299{
300	struct kern_ipc_perm *ipcp = ipc_obtain_object(&sem_ids(ns), id);
301
302	if (IS_ERR(ipcp))
303		return ERR_CAST(ipcp);
304
305	return container_of(ipcp, struct sem_array, sem_perm);
306}
307
308static inline struct sem_array *sem_obtain_object_check(struct ipc_namespace *ns,
309							int id)
310{
311	struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&sem_ids(ns), id);
312
313	if (IS_ERR(ipcp))
314		return ERR_CAST(ipcp);
315
316	return container_of(ipcp, struct sem_array, sem_perm);
317}
318
319static inline void sem_lock_and_putref(struct sem_array *sma)
320{
321	sem_lock(sma, NULL, -1);
322	ipc_rcu_putref(sma);
323}
324
325static inline void sem_putref(struct sem_array *sma)
326{
327	ipc_rcu_putref(sma);
328}
329
330static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
331{
332	ipc_rmid(&sem_ids(ns), &s->sem_perm);
333}
334
335/*
336 * Lockless wakeup algorithm:
337 * Without the check/retry algorithm a lockless wakeup is possible:
338 * - queue.status is initialized to -EINTR before blocking.
339 * - wakeup is performed by
340 *	* unlinking the queue entry from sma->sem_pending
341 *	* setting queue.status to IN_WAKEUP
342 *	  This is the notification for the blocked thread that a
343 *	  result value is imminent.
344 *	* call wake_up_process
345 *	* set queue.status to the final value.
346 * - the previously blocked thread checks queue.status:
347 *   	* if it's IN_WAKEUP, then it must wait until the value changes
348 *   	* if it's not -EINTR, then the operation was completed by
349 *   	  update_queue. semtimedop can return queue.status without
350 *   	  performing any operation on the sem array.
351 *   	* otherwise it must acquire the spinlock and check what's up.
352 *
353 * The two-stage algorithm is necessary to protect against the following
354 * races:
355 * - if queue.status is set after wake_up_process, then the woken up idle
356 *   thread could race forward and try (and fail) to acquire sma->lock
357 *   before update_queue had a chance to set queue.status
358 * - if queue.status is written before wake_up_process and if the
359 *   blocked process is woken up by a signal between writing
360 *   queue.status and the wake_up_process, then the woken up
361 *   process could return from semtimedop and die by calling
362 *   sys_exit before wake_up_process is called. Then wake_up_process
363 *   will oops, because the task structure is already invalid.
364 *   (yes, this happened on s390 with sysv msg).
365 *
366 */
367#define IN_WAKEUP	1
368
369/**
370 * newary - Create a new semaphore set
371 * @ns: namespace
372 * @params: ptr to the structure that contains key, semflg and nsems
373 *
374 * Called with sem_ids.rw_mutex held (as a writer)
375 */
376
377static int newary(struct ipc_namespace *ns, struct ipc_params *params)
378{
379	int id;
380	int retval;
381	struct sem_array *sma;
382	int size;
383	key_t key = params->key;
384	int nsems = params->u.nsems;
385	int semflg = params->flg;
386	int i;
387
388	if (!nsems)
389		return -EINVAL;
390	if (ns->used_sems + nsems > ns->sc_semmns)
391		return -ENOSPC;
392
393	size = sizeof (*sma) + nsems * sizeof (struct sem);
394	sma = ipc_rcu_alloc(size);
395	if (!sma) {
396		return -ENOMEM;
397	}
398	memset (sma, 0, size);
399
400	sma->sem_perm.mode = (semflg & S_IRWXUGO);
401	sma->sem_perm.key = key;
402
403	sma->sem_perm.security = NULL;
404	retval = security_sem_alloc(sma);
405	if (retval) {
406		ipc_rcu_putref(sma);
407		return retval;
408	}
409
410	id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
411	if (id < 0) {
412		security_sem_free(sma);
413		ipc_rcu_putref(sma);
414		return id;
415	}
416	ns->used_sems += nsems;
417
418	sma->sem_base = (struct sem *) &sma[1];
419
420	for (i = 0; i < nsems; i++) {
421		INIT_LIST_HEAD(&sma->sem_base[i].sem_pending);
422		spin_lock_init(&sma->sem_base[i].lock);
423	}
424
425	sma->complex_count = 0;
426	INIT_LIST_HEAD(&sma->sem_pending);
427	INIT_LIST_HEAD(&sma->list_id);
428	sma->sem_nsems = nsems;
429	sma->sem_ctime = get_seconds();
430	sem_unlock(sma, -1);
431	rcu_read_unlock();
432
433	return sma->sem_perm.id;
434}
435
436
437/*
438 * Called with sem_ids.rw_mutex and ipcp locked.
439 */
440static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg)
441{
442	struct sem_array *sma;
443
444	sma = container_of(ipcp, struct sem_array, sem_perm);
445	return security_sem_associate(sma, semflg);
446}
447
448/*
449 * Called with sem_ids.rw_mutex and ipcp locked.
450 */
451static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
452				struct ipc_params *params)
453{
454	struct sem_array *sma;
455
456	sma = container_of(ipcp, struct sem_array, sem_perm);
457	if (params->u.nsems > sma->sem_nsems)
458		return -EINVAL;
459
460	return 0;
461}
462
463SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg)
464{
465	struct ipc_namespace *ns;
466	struct ipc_ops sem_ops;
467	struct ipc_params sem_params;
468
469	ns = current->nsproxy->ipc_ns;
470
471	if (nsems < 0 || nsems > ns->sc_semmsl)
472		return -EINVAL;
473
474	sem_ops.getnew = newary;
475	sem_ops.associate = sem_security;
476	sem_ops.more_checks = sem_more_checks;
477
478	sem_params.key = key;
479	sem_params.flg = semflg;
480	sem_params.u.nsems = nsems;
481
482	return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
483}
484
485/*
486 * Determine whether a sequence of semaphore operations would succeed
487 * all at once. Return 0 if yes, 1 if need to sleep, else return error code.
488 */
489
490static int try_atomic_semop (struct sem_array * sma, struct sembuf * sops,
491			     int nsops, struct sem_undo *un, int pid)
492{
493	int result, sem_op;
494	struct sembuf *sop;
495	struct sem * curr;
496
497	for (sop = sops; sop < sops + nsops; sop++) {
498		curr = sma->sem_base + sop->sem_num;
499		sem_op = sop->sem_op;
500		result = curr->semval;
501
502		if (!sem_op && result)
503			goto would_block;
504
505		result += sem_op;
506		if (result < 0)
507			goto would_block;
508		if (result > SEMVMX)
509			goto out_of_range;
510		if (sop->sem_flg & SEM_UNDO) {
511			int undo = un->semadj[sop->sem_num] - sem_op;
512			/*
513	 		 *	Exceeding the undo range is an error.
514			 */
515			if (undo < (-SEMAEM - 1) || undo > SEMAEM)
516				goto out_of_range;
517		}
518		curr->semval = result;
519	}
520
521	sop--;
522	while (sop >= sops) {
523		sma->sem_base[sop->sem_num].sempid = pid;
524		if (sop->sem_flg & SEM_UNDO)
525			un->semadj[sop->sem_num] -= sop->sem_op;
526		sop--;
527	}
528
529	return 0;
530
531out_of_range:
532	result = -ERANGE;
533	goto undo;
534
535would_block:
536	if (sop->sem_flg & IPC_NOWAIT)
537		result = -EAGAIN;
538	else
539		result = 1;
540
541undo:
542	sop--;
543	while (sop >= sops) {
544		sma->sem_base[sop->sem_num].semval -= sop->sem_op;
545		sop--;
546	}
547
548	return result;
549}
550
551/** wake_up_sem_queue_prepare(q, error): Prepare wake-up
552 * @q: queue entry that must be signaled
553 * @error: Error value for the signal
554 *
555 * Prepare the wake-up of the queue entry q.
556 */
557static void wake_up_sem_queue_prepare(struct list_head *pt,
558				struct sem_queue *q, int error)
559{
560	if (list_empty(pt)) {
561		/*
562		 * Hold preempt off so that we don't get preempted and have the
563		 * wakee busy-wait until we're scheduled back on.
564		 */
565		preempt_disable();
566	}
567	q->status = IN_WAKEUP;
568	q->pid = error;
569
570	list_add_tail(&q->list, pt);
571}
572
573/**
574 * wake_up_sem_queue_do(pt) - do the actual wake-up
575 * @pt: list of tasks to be woken up
576 *
577 * Do the actual wake-up.
578 * The function is called without any locks held, thus the semaphore array
579 * could be destroyed already and the tasks can disappear as soon as the
580 * status is set to the actual return code.
581 */
582static void wake_up_sem_queue_do(struct list_head *pt)
583{
584	struct sem_queue *q, *t;
585	int did_something;
586
587	did_something = !list_empty(pt);
588	list_for_each_entry_safe(q, t, pt, list) {
589		wake_up_process(q->sleeper);
590		/* q can disappear immediately after writing q->status. */
591		smp_wmb();
592		q->status = q->pid;
593	}
594	if (did_something)
595		preempt_enable();
596}
597
598static void unlink_queue(struct sem_array *sma, struct sem_queue *q)
599{
600	list_del(&q->list);
601	if (q->nsops > 1)
602		sma->complex_count--;
603}
604
605/** check_restart(sma, q)
606 * @sma: semaphore array
607 * @q: the operation that just completed
608 *
609 * update_queue is O(N^2) when it restarts scanning the whole queue of
610 * waiting operations. Therefore this function checks if the restart is
611 * really necessary. It is called after a previously waiting operation
612 * was completed.
613 */
614static int check_restart(struct sem_array *sma, struct sem_queue *q)
615{
616	struct sem *curr;
617	struct sem_queue *h;
618
619	/* if the operation didn't modify the array, then no restart */
620	if (q->alter == 0)
621		return 0;
622
623	/* pending complex operations are too difficult to analyse */
624	if (sma->complex_count)
625		return 1;
626
627	/* we were a sleeping complex operation. Too difficult */
628	if (q->nsops > 1)
629		return 1;
630
631	curr = sma->sem_base + q->sops[0].sem_num;
632
633	/* No-one waits on this queue */
634	if (list_empty(&curr->sem_pending))
635		return 0;
636
637	/* the new semaphore value */
638	if (curr->semval) {
639		/* It is impossible that someone waits for the new value:
640		 * - q is a previously sleeping simple operation that
641		 *   altered the array. It must be a decrement, because
642		 *   simple increments never sleep.
643		 * - The value is not 0, thus wait-for-zero won't proceed.
644		 * - If there are older (higher priority) decrements
645		 *   in the queue, then they have observed the original
646		 *   semval value and couldn't proceed. The operation
647		 *   decremented to value - thus they won't proceed either.
648		 */
649		BUG_ON(q->sops[0].sem_op >= 0);
650		return 0;
651	}
652	/*
653	 * semval is 0. Check if there are wait-for-zero semops.
654	 * They must be the first entries in the per-semaphore queue
655	 */
656	h = list_first_entry(&curr->sem_pending, struct sem_queue, list);
657	BUG_ON(h->nsops != 1);
658	BUG_ON(h->sops[0].sem_num != q->sops[0].sem_num);
659
660	/* Yes, there is a wait-for-zero semop. Restart */
661	if (h->sops[0].sem_op == 0)
662		return 1;
663
664	/* Again - no-one is waiting for the new value. */
665	return 0;
666}
667
668
669/**
670 * update_queue(sma, semnum): Look for tasks that can be completed.
671 * @sma: semaphore array.
672 * @semnum: semaphore that was modified.
673 * @pt: list head for the tasks that must be woken up.
674 *
675 * update_queue must be called after a semaphore in a semaphore array
676 * was modified. If multiple semaphores were modified, update_queue must
677 * be called with semnum = -1, as well as with the number of each modified
678 * semaphore.
679 * The tasks that must be woken up are added to @pt. The return code
680 * is stored in q->pid.
681 * The function return 1 if at least one semop was completed successfully.
682 */
683static int update_queue(struct sem_array *sma, int semnum, struct list_head *pt)
684{
685	struct sem_queue *q;
686	struct list_head *walk;
687	struct list_head *pending_list;
688	int semop_completed = 0;
689
690	if (semnum == -1)
691		pending_list = &sma->sem_pending;
692	else
693		pending_list = &sma->sem_base[semnum].sem_pending;
694
695again:
696	walk = pending_list->next;
697	while (walk != pending_list) {
698		int error, restart;
699
700		q = container_of(walk, struct sem_queue, list);
701		walk = walk->next;
702
703		/* If we are scanning the single sop, per-semaphore list of
704		 * one semaphore and that semaphore is 0, then it is not
705		 * necessary to scan the "alter" entries: simple increments
706		 * that affect only one entry succeed immediately and cannot
707		 * be in the  per semaphore pending queue, and decrements
708		 * cannot be successful if the value is already 0.
709		 */
710		if (semnum != -1 && sma->sem_base[semnum].semval == 0 &&
711				q->alter)
712			break;
713
714		error = try_atomic_semop(sma, q->sops, q->nsops,
715					 q->undo, q->pid);
716
717		/* Does q->sleeper still need to sleep? */
718		if (error > 0)
719			continue;
720
721		unlink_queue(sma, q);
722
723		if (error) {
724			restart = 0;
725		} else {
726			semop_completed = 1;
727			restart = check_restart(sma, q);
728		}
729
730		wake_up_sem_queue_prepare(pt, q, error);
731		if (restart)
732			goto again;
733	}
734	return semop_completed;
735}
736
737/**
738 * do_smart_update(sma, sops, nsops, otime, pt) - optimized update_queue
739 * @sma: semaphore array
740 * @sops: operations that were performed
741 * @nsops: number of operations
742 * @otime: force setting otime
743 * @pt: list head of the tasks that must be woken up.
744 *
745 * do_smart_update() does the required called to update_queue, based on the
746 * actual changes that were performed on the semaphore array.
747 * Note that the function does not do the actual wake-up: the caller is
748 * responsible for calling wake_up_sem_queue_do(@pt).
749 * It is safe to perform this call after dropping all locks.
750 */
751static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops,
752			int otime, struct list_head *pt)
753{
754	int i;
755	int progress;
756
757	progress = 1;
758retry_global:
759	if (sma->complex_count) {
760		if (update_queue(sma, -1, pt)) {
761			progress = 1;
762			otime = 1;
763			sops = NULL;
764		}
765	}
766	if (!progress)
767		goto done;
768
769	if (!sops) {
770		/* No semops; something special is going on. */
771		for (i = 0; i < sma->sem_nsems; i++) {
772			if (update_queue(sma, i, pt)) {
773				otime = 1;
774				progress = 1;
775			}
776		}
777		goto done_checkretry;
778	}
779
780	/* Check the semaphores that were modified. */
781	for (i = 0; i < nsops; i++) {
782		if (sops[i].sem_op > 0 ||
783			(sops[i].sem_op < 0 &&
784				sma->sem_base[sops[i].sem_num].semval == 0))
785			if (update_queue(sma, sops[i].sem_num, pt)) {
786				otime = 1;
787				progress = 1;
788			}
789	}
790done_checkretry:
791	if (progress) {
792		progress = 0;
793		goto retry_global;
794	}
795done:
796	if (otime)
797		sma->sem_otime = get_seconds();
798}
799
800
801/* The following counts are associated to each semaphore:
802 *   semncnt        number of tasks waiting on semval being nonzero
803 *   semzcnt        number of tasks waiting on semval being zero
804 * This model assumes that a task waits on exactly one semaphore.
805 * Since semaphore operations are to be performed atomically, tasks actually
806 * wait on a whole sequence of semaphores simultaneously.
807 * The counts we return here are a rough approximation, but still
808 * warrant that semncnt+semzcnt>0 if the task is on the pending queue.
809 */
810static int count_semncnt (struct sem_array * sma, ushort semnum)
811{
812	int semncnt;
813	struct sem_queue * q;
814
815	semncnt = 0;
816	list_for_each_entry(q, &sma->sem_base[semnum].sem_pending, list) {
817		struct sembuf * sops = q->sops;
818		BUG_ON(sops->sem_num != semnum);
819		if ((sops->sem_op < 0) && !(sops->sem_flg & IPC_NOWAIT))
820			semncnt++;
821	}
822
823	list_for_each_entry(q, &sma->sem_pending, list) {
824		struct sembuf * sops = q->sops;
825		int nsops = q->nsops;
826		int i;
827		for (i = 0; i < nsops; i++)
828			if (sops[i].sem_num == semnum
829			    && (sops[i].sem_op < 0)
830			    && !(sops[i].sem_flg & IPC_NOWAIT))
831				semncnt++;
832	}
833	return semncnt;
834}
835
836static int count_semzcnt (struct sem_array * sma, ushort semnum)
837{
838	int semzcnt;
839	struct sem_queue * q;
840
841	semzcnt = 0;
842	list_for_each_entry(q, &sma->sem_base[semnum].sem_pending, list) {
843		struct sembuf * sops = q->sops;
844		BUG_ON(sops->sem_num != semnum);
845		if ((sops->sem_op == 0) && !(sops->sem_flg & IPC_NOWAIT))
846			semzcnt++;
847	}
848
849	list_for_each_entry(q, &sma->sem_pending, list) {
850		struct sembuf * sops = q->sops;
851		int nsops = q->nsops;
852		int i;
853		for (i = 0; i < nsops; i++)
854			if (sops[i].sem_num == semnum
855			    && (sops[i].sem_op == 0)
856			    && !(sops[i].sem_flg & IPC_NOWAIT))
857				semzcnt++;
858	}
859	return semzcnt;
860}
861
862/* Free a semaphore set. freeary() is called with sem_ids.rw_mutex locked
863 * as a writer and the spinlock for this semaphore set hold. sem_ids.rw_mutex
864 * remains locked on exit.
865 */
866static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
867{
868	struct sem_undo *un, *tu;
869	struct sem_queue *q, *tq;
870	struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
871	struct list_head tasks;
872	int i;
873
874	/* Free the existing undo structures for this semaphore set.  */
875	ipc_assert_locked_object(&sma->sem_perm);
876	list_for_each_entry_safe(un, tu, &sma->list_id, list_id) {
877		list_del(&un->list_id);
878		spin_lock(&un->ulp->lock);
879		un->semid = -1;
880		list_del_rcu(&un->list_proc);
881		spin_unlock(&un->ulp->lock);
882		kfree_rcu(un, rcu);
883	}
884
885	/* Wake up all pending processes and let them fail with EIDRM. */
886	INIT_LIST_HEAD(&tasks);
887	list_for_each_entry_safe(q, tq, &sma->sem_pending, list) {
888		unlink_queue(sma, q);
889		wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
890	}
891	for (i = 0; i < sma->sem_nsems; i++) {
892		struct sem *sem = sma->sem_base + i;
893		list_for_each_entry_safe(q, tq, &sem->sem_pending, list) {
894			unlink_queue(sma, q);
895			wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
896		}
897	}
898
899	/* Remove the semaphore set from the IDR */
900	sem_rmid(ns, sma);
901	sem_unlock(sma, -1);
902	rcu_read_unlock();
903
904	wake_up_sem_queue_do(&tasks);
905	ns->used_sems -= sma->sem_nsems;
906	security_sem_free(sma);
907	ipc_rcu_putref(sma);
908}
909
910static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
911{
912	switch(version) {
913	case IPC_64:
914		return copy_to_user(buf, in, sizeof(*in));
915	case IPC_OLD:
916	    {
917		struct semid_ds out;
918
919		memset(&out, 0, sizeof(out));
920
921		ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
922
923		out.sem_otime	= in->sem_otime;
924		out.sem_ctime	= in->sem_ctime;
925		out.sem_nsems	= in->sem_nsems;
926
927		return copy_to_user(buf, &out, sizeof(out));
928	    }
929	default:
930		return -EINVAL;
931	}
932}
933
934static int semctl_nolock(struct ipc_namespace *ns, int semid,
935			 int cmd, int version, void __user *p)
936{
937	int err;
938	struct sem_array *sma;
939
940	switch(cmd) {
941	case IPC_INFO:
942	case SEM_INFO:
943	{
944		struct seminfo seminfo;
945		int max_id;
946
947		err = security_sem_semctl(NULL, cmd);
948		if (err)
949			return err;
950
951		memset(&seminfo,0,sizeof(seminfo));
952		seminfo.semmni = ns->sc_semmni;
953		seminfo.semmns = ns->sc_semmns;
954		seminfo.semmsl = ns->sc_semmsl;
955		seminfo.semopm = ns->sc_semopm;
956		seminfo.semvmx = SEMVMX;
957		seminfo.semmnu = SEMMNU;
958		seminfo.semmap = SEMMAP;
959		seminfo.semume = SEMUME;
960		down_read(&sem_ids(ns).rw_mutex);
961		if (cmd == SEM_INFO) {
962			seminfo.semusz = sem_ids(ns).in_use;
963			seminfo.semaem = ns->used_sems;
964		} else {
965			seminfo.semusz = SEMUSZ;
966			seminfo.semaem = SEMAEM;
967		}
968		max_id = ipc_get_maxid(&sem_ids(ns));
969		up_read(&sem_ids(ns).rw_mutex);
970		if (copy_to_user(p, &seminfo, sizeof(struct seminfo)))
971			return -EFAULT;
972		return (max_id < 0) ? 0: max_id;
973	}
974	case IPC_STAT:
975	case SEM_STAT:
976	{
977		struct semid64_ds tbuf;
978		int id = 0;
979
980		memset(&tbuf, 0, sizeof(tbuf));
981
982		rcu_read_lock();
983		if (cmd == SEM_STAT) {
984			sma = sem_obtain_object(ns, semid);
985			if (IS_ERR(sma)) {
986				err = PTR_ERR(sma);
987				goto out_unlock;
988			}
989			id = sma->sem_perm.id;
990		} else {
991			sma = sem_obtain_object_check(ns, semid);
992			if (IS_ERR(sma)) {
993				err = PTR_ERR(sma);
994				goto out_unlock;
995			}
996		}
997
998		err = -EACCES;
999		if (ipcperms(ns, &sma->sem_perm, S_IRUGO))
1000			goto out_unlock;
1001
1002		err = security_sem_semctl(sma, cmd);
1003		if (err)
1004			goto out_unlock;
1005
1006		kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
1007		tbuf.sem_otime  = sma->sem_otime;
1008		tbuf.sem_ctime  = sma->sem_ctime;
1009		tbuf.sem_nsems  = sma->sem_nsems;
1010		rcu_read_unlock();
1011		if (copy_semid_to_user(p, &tbuf, version))
1012			return -EFAULT;
1013		return id;
1014	}
1015	default:
1016		return -EINVAL;
1017	}
1018out_unlock:
1019	rcu_read_unlock();
1020	return err;
1021}
1022
1023static int semctl_setval(struct ipc_namespace *ns, int semid, int semnum,
1024		unsigned long arg)
1025{
1026	struct sem_undo *un;
1027	struct sem_array *sma;
1028	struct sem* curr;
1029	int err;
1030	struct list_head tasks;
1031	int val;
1032#if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN)
1033	/* big-endian 64bit */
1034	val = arg >> 32;
1035#else
1036	/* 32bit or little-endian 64bit */
1037	val = arg;
1038#endif
1039
1040	if (val > SEMVMX || val < 0)
1041		return -ERANGE;
1042
1043	INIT_LIST_HEAD(&tasks);
1044
1045	rcu_read_lock();
1046	sma = sem_obtain_object_check(ns, semid);
1047	if (IS_ERR(sma)) {
1048		rcu_read_unlock();
1049		return PTR_ERR(sma);
1050	}
1051
1052	if (semnum < 0 || semnum >= sma->sem_nsems) {
1053		rcu_read_unlock();
1054		return -EINVAL;
1055	}
1056
1057
1058	if (ipcperms(ns, &sma->sem_perm, S_IWUGO)) {
1059		rcu_read_unlock();
1060		return -EACCES;
1061	}
1062
1063	err = security_sem_semctl(sma, SETVAL);
1064	if (err) {
1065		rcu_read_unlock();
1066		return -EACCES;
1067	}
1068
1069	sem_lock(sma, NULL, -1);
1070
1071	curr = &sma->sem_base[semnum];
1072
1073	ipc_assert_locked_object(&sma->sem_perm);
1074	list_for_each_entry(un, &sma->list_id, list_id)
1075		un->semadj[semnum] = 0;
1076
1077	curr->semval = val;
1078	curr->sempid = task_tgid_vnr(current);
1079	sma->sem_ctime = get_seconds();
1080	/* maybe some queued-up processes were waiting for this */
1081	do_smart_update(sma, NULL, 0, 0, &tasks);
1082	sem_unlock(sma, -1);
1083	rcu_read_unlock();
1084	wake_up_sem_queue_do(&tasks);
1085	return 0;
1086}
1087
1088static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
1089		int cmd, void __user *p)
1090{
1091	struct sem_array *sma;
1092	struct sem* curr;
1093	int err, nsems;
1094	ushort fast_sem_io[SEMMSL_FAST];
1095	ushort* sem_io = fast_sem_io;
1096	struct list_head tasks;
1097
1098	INIT_LIST_HEAD(&tasks);
1099
1100	rcu_read_lock();
1101	sma = sem_obtain_object_check(ns, semid);
1102	if (IS_ERR(sma)) {
1103		rcu_read_unlock();
1104		return PTR_ERR(sma);
1105	}
1106
1107	nsems = sma->sem_nsems;
1108
1109	err = -EACCES;
1110	if (ipcperms(ns, &sma->sem_perm, cmd == SETALL ? S_IWUGO : S_IRUGO))
1111		goto out_rcu_wakeup;
1112
1113	err = security_sem_semctl(sma, cmd);
1114	if (err)
1115		goto out_rcu_wakeup;
1116
1117	err = -EACCES;
1118	switch (cmd) {
1119	case GETALL:
1120	{
1121		ushort __user *array = p;
1122		int i;
1123
1124		sem_lock(sma, NULL, -1);
1125		if(nsems > SEMMSL_FAST) {
1126			if (!ipc_rcu_getref(sma)) {
1127				sem_unlock(sma, -1);
1128				rcu_read_unlock();
1129				err = -EIDRM;
1130				goto out_free;
1131			}
1132			sem_unlock(sma, -1);
1133			rcu_read_unlock();
1134			sem_io = ipc_alloc(sizeof(ushort)*nsems);
1135			if(sem_io == NULL) {
1136				sem_putref(sma);
1137				return -ENOMEM;
1138			}
1139
1140			rcu_read_lock();
1141			sem_lock_and_putref(sma);
1142			if (sma->sem_perm.deleted) {
1143				sem_unlock(sma, -1);
1144				rcu_read_unlock();
1145				err = -EIDRM;
1146				goto out_free;
1147			}
1148		}
1149		for (i = 0; i < sma->sem_nsems; i++)
1150			sem_io[i] = sma->sem_base[i].semval;
1151		sem_unlock(sma, -1);
1152		rcu_read_unlock();
1153		err = 0;
1154		if(copy_to_user(array, sem_io, nsems*sizeof(ushort)))
1155			err = -EFAULT;
1156		goto out_free;
1157	}
1158	case SETALL:
1159	{
1160		int i;
1161		struct sem_undo *un;
1162
1163		if (!ipc_rcu_getref(sma)) {
1164			rcu_read_unlock();
1165			return -EIDRM;
1166		}
1167		rcu_read_unlock();
1168
1169		if(nsems > SEMMSL_FAST) {
1170			sem_io = ipc_alloc(sizeof(ushort)*nsems);
1171			if(sem_io == NULL) {
1172				sem_putref(sma);
1173				return -ENOMEM;
1174			}
1175		}
1176
1177		if (copy_from_user (sem_io, p, nsems*sizeof(ushort))) {
1178			sem_putref(sma);
1179			err = -EFAULT;
1180			goto out_free;
1181		}
1182
1183		for (i = 0; i < nsems; i++) {
1184			if (sem_io[i] > SEMVMX) {
1185				sem_putref(sma);
1186				err = -ERANGE;
1187				goto out_free;
1188			}
1189		}
1190		rcu_read_lock();
1191		sem_lock_and_putref(sma);
1192		if (sma->sem_perm.deleted) {
1193			sem_unlock(sma, -1);
1194			rcu_read_unlock();
1195			err = -EIDRM;
1196			goto out_free;
1197		}
1198
1199		for (i = 0; i < nsems; i++)
1200			sma->sem_base[i].semval = sem_io[i];
1201
1202		ipc_assert_locked_object(&sma->sem_perm);
1203		list_for_each_entry(un, &sma->list_id, list_id) {
1204			for (i = 0; i < nsems; i++)
1205				un->semadj[i] = 0;
1206		}
1207		sma->sem_ctime = get_seconds();
1208		/* maybe some queued-up processes were waiting for this */
1209		do_smart_update(sma, NULL, 0, 0, &tasks);
1210		err = 0;
1211		goto out_unlock;
1212	}
1213	/* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */
1214	}
1215	err = -EINVAL;
1216	if (semnum < 0 || semnum >= nsems)
1217		goto out_rcu_wakeup;
1218
1219	sem_lock(sma, NULL, -1);
1220	curr = &sma->sem_base[semnum];
1221
1222	switch (cmd) {
1223	case GETVAL:
1224		err = curr->semval;
1225		goto out_unlock;
1226	case GETPID:
1227		err = curr->sempid;
1228		goto out_unlock;
1229	case GETNCNT:
1230		err = count_semncnt(sma,semnum);
1231		goto out_unlock;
1232	case GETZCNT:
1233		err = count_semzcnt(sma,semnum);
1234		goto out_unlock;
1235	}
1236
1237out_unlock:
1238	sem_unlock(sma, -1);
1239out_rcu_wakeup:
1240	rcu_read_unlock();
1241	wake_up_sem_queue_do(&tasks);
1242out_free:
1243	if(sem_io != fast_sem_io)
1244		ipc_free(sem_io, sizeof(ushort)*nsems);
1245	return err;
1246}
1247
1248static inline unsigned long
1249copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
1250{
1251	switch(version) {
1252	case IPC_64:
1253		if (copy_from_user(out, buf, sizeof(*out)))
1254			return -EFAULT;
1255		return 0;
1256	case IPC_OLD:
1257	    {
1258		struct semid_ds tbuf_old;
1259
1260		if(copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
1261			return -EFAULT;
1262
1263		out->sem_perm.uid	= tbuf_old.sem_perm.uid;
1264		out->sem_perm.gid	= tbuf_old.sem_perm.gid;
1265		out->sem_perm.mode	= tbuf_old.sem_perm.mode;
1266
1267		return 0;
1268	    }
1269	default:
1270		return -EINVAL;
1271	}
1272}
1273
1274/*
1275 * This function handles some semctl commands which require the rw_mutex
1276 * to be held in write mode.
1277 * NOTE: no locks must be held, the rw_mutex is taken inside this function.
1278 */
1279static int semctl_down(struct ipc_namespace *ns, int semid,
1280		       int cmd, int version, void __user *p)
1281{
1282	struct sem_array *sma;
1283	int err;
1284	struct semid64_ds semid64;
1285	struct kern_ipc_perm *ipcp;
1286
1287	if(cmd == IPC_SET) {
1288		if (copy_semid_from_user(&semid64, p, version))
1289			return -EFAULT;
1290	}
1291
1292	down_write(&sem_ids(ns).rw_mutex);
1293	rcu_read_lock();
1294
1295	ipcp = ipcctl_pre_down_nolock(ns, &sem_ids(ns), semid, cmd,
1296				      &semid64.sem_perm, 0);
1297	if (IS_ERR(ipcp)) {
1298		err = PTR_ERR(ipcp);
1299		goto out_unlock1;
1300	}
1301
1302	sma = container_of(ipcp, struct sem_array, sem_perm);
1303
1304	err = security_sem_semctl(sma, cmd);
1305	if (err)
1306		goto out_unlock1;
1307
1308	switch (cmd) {
1309	case IPC_RMID:
1310		sem_lock(sma, NULL, -1);
1311		/* freeary unlocks the ipc object and rcu */
1312		freeary(ns, ipcp);
1313		goto out_up;
1314	case IPC_SET:
1315		sem_lock(sma, NULL, -1);
1316		err = ipc_update_perm(&semid64.sem_perm, ipcp);
1317		if (err)
1318			goto out_unlock0;
1319		sma->sem_ctime = get_seconds();
1320		break;
1321	default:
1322		err = -EINVAL;
1323		goto out_unlock1;
1324	}
1325
1326out_unlock0:
1327	sem_unlock(sma, -1);
1328out_unlock1:
1329	rcu_read_unlock();
1330out_up:
1331	up_write(&sem_ids(ns).rw_mutex);
1332	return err;
1333}
1334
1335SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, unsigned long, arg)
1336{
1337	int version;
1338	struct ipc_namespace *ns;
1339	void __user *p = (void __user *)arg;
1340
1341	if (semid < 0)
1342		return -EINVAL;
1343
1344	version = ipc_parse_version(&cmd);
1345	ns = current->nsproxy->ipc_ns;
1346
1347	switch(cmd) {
1348	case IPC_INFO:
1349	case SEM_INFO:
1350	case IPC_STAT:
1351	case SEM_STAT:
1352		return semctl_nolock(ns, semid, cmd, version, p);
1353	case GETALL:
1354	case GETVAL:
1355	case GETPID:
1356	case GETNCNT:
1357	case GETZCNT:
1358	case SETALL:
1359		return semctl_main(ns, semid, semnum, cmd, p);
1360	case SETVAL:
1361		return semctl_setval(ns, semid, semnum, arg);
1362	case IPC_RMID:
1363	case IPC_SET:
1364		return semctl_down(ns, semid, cmd, version, p);
1365	default:
1366		return -EINVAL;
1367	}
1368}
1369
1370/* If the task doesn't already have a undo_list, then allocate one
1371 * here.  We guarantee there is only one thread using this undo list,
1372 * and current is THE ONE
1373 *
1374 * If this allocation and assignment succeeds, but later
1375 * portions of this code fail, there is no need to free the sem_undo_list.
1376 * Just let it stay associated with the task, and it'll be freed later
1377 * at exit time.
1378 *
1379 * This can block, so callers must hold no locks.
1380 */
1381static inline int get_undo_list(struct sem_undo_list **undo_listp)
1382{
1383	struct sem_undo_list *undo_list;
1384
1385	undo_list = current->sysvsem.undo_list;
1386	if (!undo_list) {
1387		undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
1388		if (undo_list == NULL)
1389			return -ENOMEM;
1390		spin_lock_init(&undo_list->lock);
1391		atomic_set(&undo_list->refcnt, 1);
1392		INIT_LIST_HEAD(&undo_list->list_proc);
1393
1394		current->sysvsem.undo_list = undo_list;
1395	}
1396	*undo_listp = undo_list;
1397	return 0;
1398}
1399
1400static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid)
1401{
1402	struct sem_undo *un;
1403
1404	list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) {
1405		if (un->semid == semid)
1406			return un;
1407	}
1408	return NULL;
1409}
1410
1411static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1412{
1413	struct sem_undo *un;
1414
1415  	assert_spin_locked(&ulp->lock);
1416
1417	un = __lookup_undo(ulp, semid);
1418	if (un) {
1419		list_del_rcu(&un->list_proc);
1420		list_add_rcu(&un->list_proc, &ulp->list_proc);
1421	}
1422	return un;
1423}
1424
1425/**
1426 * find_alloc_undo - Lookup (and if not present create) undo array
1427 * @ns: namespace
1428 * @semid: semaphore array id
1429 *
1430 * The function looks up (and if not present creates) the undo structure.
1431 * The size of the undo structure depends on the size of the semaphore
1432 * array, thus the alloc path is not that straightforward.
1433 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1434 * performs a rcu_read_lock().
1435 */
1436static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
1437{
1438	struct sem_array *sma;
1439	struct sem_undo_list *ulp;
1440	struct sem_undo *un, *new;
1441	int nsems, error;
1442
1443	error = get_undo_list(&ulp);
1444	if (error)
1445		return ERR_PTR(error);
1446
1447	rcu_read_lock();
1448	spin_lock(&ulp->lock);
1449	un = lookup_undo(ulp, semid);
1450	spin_unlock(&ulp->lock);
1451	if (likely(un!=NULL))
1452		goto out;
1453
1454	/* no undo structure around - allocate one. */
1455	/* step 1: figure out the size of the semaphore array */
1456	sma = sem_obtain_object_check(ns, semid);
1457	if (IS_ERR(sma)) {
1458		rcu_read_unlock();
1459		return ERR_CAST(sma);
1460	}
1461
1462	nsems = sma->sem_nsems;
1463	if (!ipc_rcu_getref(sma)) {
1464		rcu_read_unlock();
1465		un = ERR_PTR(-EIDRM);
1466		goto out;
1467	}
1468	rcu_read_unlock();
1469
1470	/* step 2: allocate new undo structure */
1471	new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1472	if (!new) {
1473		sem_putref(sma);
1474		return ERR_PTR(-ENOMEM);
1475	}
1476
1477	/* step 3: Acquire the lock on semaphore array */
1478	rcu_read_lock();
1479	sem_lock_and_putref(sma);
1480	if (sma->sem_perm.deleted) {
1481		sem_unlock(sma, -1);
1482		rcu_read_unlock();
1483		kfree(new);
1484		un = ERR_PTR(-EIDRM);
1485		goto out;
1486	}
1487	spin_lock(&ulp->lock);
1488
1489	/*
1490	 * step 4: check for races: did someone else allocate the undo struct?
1491	 */
1492	un = lookup_undo(ulp, semid);
1493	if (un) {
1494		kfree(new);
1495		goto success;
1496	}
1497	/* step 5: initialize & link new undo structure */
1498	new->semadj = (short *) &new[1];
1499	new->ulp = ulp;
1500	new->semid = semid;
1501	assert_spin_locked(&ulp->lock);
1502	list_add_rcu(&new->list_proc, &ulp->list_proc);
1503	ipc_assert_locked_object(&sma->sem_perm);
1504	list_add(&new->list_id, &sma->list_id);
1505	un = new;
1506
1507success:
1508	spin_unlock(&ulp->lock);
1509	sem_unlock(sma, -1);
1510out:
1511	return un;
1512}
1513
1514
1515/**
1516 * get_queue_result - Retrieve the result code from sem_queue
1517 * @q: Pointer to queue structure
1518 *
1519 * Retrieve the return code from the pending queue. If IN_WAKEUP is found in
1520 * q->status, then we must loop until the value is replaced with the final
1521 * value: This may happen if a task is woken up by an unrelated event (e.g.
1522 * signal) and in parallel the task is woken up by another task because it got
1523 * the requested semaphores.
1524 *
1525 * The function can be called with or without holding the semaphore spinlock.
1526 */
1527static int get_queue_result(struct sem_queue *q)
1528{
1529	int error;
1530
1531	error = q->status;
1532	while (unlikely(error == IN_WAKEUP)) {
1533		cpu_relax();
1534		error = q->status;
1535	}
1536
1537	return error;
1538}
1539
1540
1541SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
1542		unsigned, nsops, const struct timespec __user *, timeout)
1543{
1544	int error = -EINVAL;
1545	struct sem_array *sma;
1546	struct sembuf fast_sops[SEMOPM_FAST];
1547	struct sembuf* sops = fast_sops, *sop;
1548	struct sem_undo *un;
1549	int undos = 0, alter = 0, max, locknum;
1550	struct sem_queue queue;
1551	unsigned long jiffies_left = 0;
1552	struct ipc_namespace *ns;
1553	struct list_head tasks;
1554
1555	ns = current->nsproxy->ipc_ns;
1556
1557	if (nsops < 1 || semid < 0)
1558		return -EINVAL;
1559	if (nsops > ns->sc_semopm)
1560		return -E2BIG;
1561	if(nsops > SEMOPM_FAST) {
1562		sops = kmalloc(sizeof(*sops)*nsops,GFP_KERNEL);
1563		if(sops==NULL)
1564			return -ENOMEM;
1565	}
1566	if (copy_from_user (sops, tsops, nsops * sizeof(*tsops))) {
1567		error=-EFAULT;
1568		goto out_free;
1569	}
1570	if (timeout) {
1571		struct timespec _timeout;
1572		if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1573			error = -EFAULT;
1574			goto out_free;
1575		}
1576		if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1577			_timeout.tv_nsec >= 1000000000L) {
1578			error = -EINVAL;
1579			goto out_free;
1580		}
1581		jiffies_left = timespec_to_jiffies(&_timeout);
1582	}
1583	max = 0;
1584	for (sop = sops; sop < sops + nsops; sop++) {
1585		if (sop->sem_num >= max)
1586			max = sop->sem_num;
1587		if (sop->sem_flg & SEM_UNDO)
1588			undos = 1;
1589		if (sop->sem_op != 0)
1590			alter = 1;
1591	}
1592
1593	INIT_LIST_HEAD(&tasks);
1594
1595	if (undos) {
1596		/* On success, find_alloc_undo takes the rcu_read_lock */
1597		un = find_alloc_undo(ns, semid);
1598		if (IS_ERR(un)) {
1599			error = PTR_ERR(un);
1600			goto out_free;
1601		}
1602	} else {
1603		un = NULL;
1604		rcu_read_lock();
1605	}
1606
1607	sma = sem_obtain_object_check(ns, semid);
1608	if (IS_ERR(sma)) {
1609		rcu_read_unlock();
1610		error = PTR_ERR(sma);
1611		goto out_free;
1612	}
1613
1614	error = -EFBIG;
1615	if (max >= sma->sem_nsems)
1616		goto out_rcu_wakeup;
1617
1618	error = -EACCES;
1619	if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO))
1620		goto out_rcu_wakeup;
1621
1622	error = security_sem_semop(sma, sops, nsops, alter);
1623	if (error)
1624		goto out_rcu_wakeup;
1625
1626	/*
1627	 * semid identifiers are not unique - find_alloc_undo may have
1628	 * allocated an undo structure, it was invalidated by an RMID
1629	 * and now a new array with received the same id. Check and fail.
1630	 * This case can be detected checking un->semid. The existence of
1631	 * "un" itself is guaranteed by rcu.
1632	 */
1633	error = -EIDRM;
1634	locknum = sem_lock(sma, sops, nsops);
1635	if (un && un->semid == -1)
1636		goto out_unlock_free;
1637
1638	error = try_atomic_semop (sma, sops, nsops, un, task_tgid_vnr(current));
1639	if (error <= 0) {
1640		if (alter && error == 0)
1641			do_smart_update(sma, sops, nsops, 1, &tasks);
1642
1643		goto out_unlock_free;
1644	}
1645
1646	/* We need to sleep on this operation, so we put the current
1647	 * task into the pending queue and go to sleep.
1648	 */
1649
1650	queue.sops = sops;
1651	queue.nsops = nsops;
1652	queue.undo = un;
1653	queue.pid = task_tgid_vnr(current);
1654	queue.alter = alter;
1655
1656	if (nsops == 1) {
1657		struct sem *curr;
1658		curr = &sma->sem_base[sops->sem_num];
1659
1660		if (alter)
1661			list_add_tail(&queue.list, &curr->sem_pending);
1662		else
1663			list_add(&queue.list, &curr->sem_pending);
1664	} else {
1665		if (alter)
1666			list_add_tail(&queue.list, &sma->sem_pending);
1667		else
1668			list_add(&queue.list, &sma->sem_pending);
1669		sma->complex_count++;
1670	}
1671
1672	queue.status = -EINTR;
1673	queue.sleeper = current;
1674
1675sleep_again:
1676	current->state = TASK_INTERRUPTIBLE;
1677	sem_unlock(sma, locknum);
1678	rcu_read_unlock();
1679
1680	if (timeout)
1681		jiffies_left = schedule_timeout(jiffies_left);
1682	else
1683		schedule();
1684
1685	error = get_queue_result(&queue);
1686
1687	if (error != -EINTR) {
1688		/* fast path: update_queue already obtained all requested
1689		 * resources.
1690		 * Perform a smp_mb(): User space could assume that semop()
1691		 * is a memory barrier: Without the mb(), the cpu could
1692		 * speculatively read in user space stale data that was
1693		 * overwritten by the previous owner of the semaphore.
1694		 */
1695		smp_mb();
1696
1697		goto out_free;
1698	}
1699
1700	rcu_read_lock();
1701	sma = sem_obtain_lock(ns, semid, sops, nsops, &locknum);
1702
1703	/*
1704	 * Wait until it's guaranteed that no wakeup_sem_queue_do() is ongoing.
1705	 */
1706	error = get_queue_result(&queue);
1707
1708	/*
1709	 * Array removed? If yes, leave without sem_unlock().
1710	 */
1711	if (IS_ERR(sma)) {
1712		rcu_read_unlock();
1713		goto out_free;
1714	}
1715
1716
1717	/*
1718	 * If queue.status != -EINTR we are woken up by another process.
1719	 * Leave without unlink_queue(), but with sem_unlock().
1720	 */
1721
1722	if (error != -EINTR) {
1723		goto out_unlock_free;
1724	}
1725
1726	/*
1727	 * If an interrupt occurred we have to clean up the queue
1728	 */
1729	if (timeout && jiffies_left == 0)
1730		error = -EAGAIN;
1731
1732	/*
1733	 * If the wakeup was spurious, just retry
1734	 */
1735	if (error == -EINTR && !signal_pending(current))
1736		goto sleep_again;
1737
1738	unlink_queue(sma, &queue);
1739
1740out_unlock_free:
1741	sem_unlock(sma, locknum);
1742out_rcu_wakeup:
1743	rcu_read_unlock();
1744	wake_up_sem_queue_do(&tasks);
1745out_free:
1746	if(sops != fast_sops)
1747		kfree(sops);
1748	return error;
1749}
1750
1751SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
1752		unsigned, nsops)
1753{
1754	return sys_semtimedop(semid, tsops, nsops, NULL);
1755}
1756
1757/* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
1758 * parent and child tasks.
1759 */
1760
1761int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
1762{
1763	struct sem_undo_list *undo_list;
1764	int error;
1765
1766	if (clone_flags & CLONE_SYSVSEM) {
1767		error = get_undo_list(&undo_list);
1768		if (error)
1769			return error;
1770		atomic_inc(&undo_list->refcnt);
1771		tsk->sysvsem.undo_list = undo_list;
1772	} else
1773		tsk->sysvsem.undo_list = NULL;
1774
1775	return 0;
1776}
1777
1778/*
1779 * add semadj values to semaphores, free undo structures.
1780 * undo structures are not freed when semaphore arrays are destroyed
1781 * so some of them may be out of date.
1782 * IMPLEMENTATION NOTE: There is some confusion over whether the
1783 * set of adjustments that needs to be done should be done in an atomic
1784 * manner or not. That is, if we are attempting to decrement the semval
1785 * should we queue up and wait until we can do so legally?
1786 * The original implementation attempted to do this (queue and wait).
1787 * The current implementation does not do so. The POSIX standard
1788 * and SVID should be consulted to determine what behavior is mandated.
1789 */
1790void exit_sem(struct task_struct *tsk)
1791{
1792	struct sem_undo_list *ulp;
1793
1794	ulp = tsk->sysvsem.undo_list;
1795	if (!ulp)
1796		return;
1797	tsk->sysvsem.undo_list = NULL;
1798
1799	if (!atomic_dec_and_test(&ulp->refcnt))
1800		return;
1801
1802	for (;;) {
1803		struct sem_array *sma;
1804		struct sem_undo *un;
1805		struct list_head tasks;
1806		int semid, i;
1807
1808		rcu_read_lock();
1809		un = list_entry_rcu(ulp->list_proc.next,
1810				    struct sem_undo, list_proc);
1811		if (&un->list_proc == &ulp->list_proc)
1812			semid = -1;
1813		 else
1814			semid = un->semid;
1815
1816		if (semid == -1) {
1817			rcu_read_unlock();
1818			break;
1819		}
1820
1821		sma = sem_obtain_object_check(tsk->nsproxy->ipc_ns, un->semid);
1822		/* exit_sem raced with IPC_RMID, nothing to do */
1823		if (IS_ERR(sma)) {
1824			rcu_read_unlock();
1825			continue;
1826		}
1827
1828		sem_lock(sma, NULL, -1);
1829		un = __lookup_undo(ulp, semid);
1830		if (un == NULL) {
1831			/* exit_sem raced with IPC_RMID+semget() that created
1832			 * exactly the same semid. Nothing to do.
1833			 */
1834			sem_unlock(sma, -1);
1835			rcu_read_unlock();
1836			continue;
1837		}
1838
1839		/* remove un from the linked lists */
1840		ipc_assert_locked_object(&sma->sem_perm);
1841		list_del(&un->list_id);
1842
1843		spin_lock(&ulp->lock);
1844		list_del_rcu(&un->list_proc);
1845		spin_unlock(&ulp->lock);
1846
1847		/* perform adjustments registered in un */
1848		for (i = 0; i < sma->sem_nsems; i++) {
1849			struct sem * semaphore = &sma->sem_base[i];
1850			if (un->semadj[i]) {
1851				semaphore->semval += un->semadj[i];
1852				/*
1853				 * Range checks of the new semaphore value,
1854				 * not defined by sus:
1855				 * - Some unices ignore the undo entirely
1856				 *   (e.g. HP UX 11i 11.22, Tru64 V5.1)
1857				 * - some cap the value (e.g. FreeBSD caps
1858				 *   at 0, but doesn't enforce SEMVMX)
1859				 *
1860				 * Linux caps the semaphore value, both at 0
1861				 * and at SEMVMX.
1862				 *
1863				 * 	Manfred <manfred@colorfullife.com>
1864				 */
1865				if (semaphore->semval < 0)
1866					semaphore->semval = 0;
1867				if (semaphore->semval > SEMVMX)
1868					semaphore->semval = SEMVMX;
1869				semaphore->sempid = task_tgid_vnr(current);
1870			}
1871		}
1872		/* maybe some queued-up processes were waiting for this */
1873		INIT_LIST_HEAD(&tasks);
1874		do_smart_update(sma, NULL, 0, 1, &tasks);
1875		sem_unlock(sma, -1);
1876		rcu_read_unlock();
1877		wake_up_sem_queue_do(&tasks);
1878
1879		kfree_rcu(un, rcu);
1880	}
1881	kfree(ulp);
1882}
1883
1884#ifdef CONFIG_PROC_FS
1885static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
1886{
1887	struct user_namespace *user_ns = seq_user_ns(s);
1888	struct sem_array *sma = it;
1889
1890	return seq_printf(s,
1891			  "%10d %10d  %4o %10u %5u %5u %5u %5u %10lu %10lu\n",
1892			  sma->sem_perm.key,
1893			  sma->sem_perm.id,
1894			  sma->sem_perm.mode,
1895			  sma->sem_nsems,
1896			  from_kuid_munged(user_ns, sma->sem_perm.uid),
1897			  from_kgid_munged(user_ns, sma->sem_perm.gid),
1898			  from_kuid_munged(user_ns, sma->sem_perm.cuid),
1899			  from_kgid_munged(user_ns, sma->sem_perm.cgid),
1900			  sma->sem_otime,
1901			  sma->sem_ctime);
1902}
1903#endif
1904