sem.c revision c728b9c8
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		spin_lock(&sma->sem_perm.lock);
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		spin_unlock(&sma->sem_perm.lock);
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
756	if (sma->complex_count || sops == NULL) {
757		if (update_queue(sma, -1, pt))
758			otime = 1;
759	}
760
761	if (!sops) {
762		/* No semops; something special is going on. */
763		for (i = 0; i < sma->sem_nsems; i++) {
764			if (update_queue(sma, i, pt))
765				otime = 1;
766		}
767		goto done;
768	}
769
770	/* Check the semaphores that were modified. */
771	for (i = 0; i < nsops; i++) {
772		if (sops[i].sem_op > 0 ||
773			(sops[i].sem_op < 0 &&
774				sma->sem_base[sops[i].sem_num].semval == 0))
775			if (update_queue(sma, sops[i].sem_num, pt))
776				otime = 1;
777	}
778done:
779	if (otime)
780		sma->sem_otime = get_seconds();
781}
782
783
784/* The following counts are associated to each semaphore:
785 *   semncnt        number of tasks waiting on semval being nonzero
786 *   semzcnt        number of tasks waiting on semval being zero
787 * This model assumes that a task waits on exactly one semaphore.
788 * Since semaphore operations are to be performed atomically, tasks actually
789 * wait on a whole sequence of semaphores simultaneously.
790 * The counts we return here are a rough approximation, but still
791 * warrant that semncnt+semzcnt>0 if the task is on the pending queue.
792 */
793static int count_semncnt (struct sem_array * sma, ushort semnum)
794{
795	int semncnt;
796	struct sem_queue * q;
797
798	semncnt = 0;
799	list_for_each_entry(q, &sma->sem_pending, list) {
800		struct sembuf * sops = q->sops;
801		int nsops = q->nsops;
802		int i;
803		for (i = 0; i < nsops; i++)
804			if (sops[i].sem_num == semnum
805			    && (sops[i].sem_op < 0)
806			    && !(sops[i].sem_flg & IPC_NOWAIT))
807				semncnt++;
808	}
809	return semncnt;
810}
811
812static int count_semzcnt (struct sem_array * sma, ushort semnum)
813{
814	int semzcnt;
815	struct sem_queue * q;
816
817	semzcnt = 0;
818	list_for_each_entry(q, &sma->sem_pending, list) {
819		struct sembuf * sops = q->sops;
820		int nsops = q->nsops;
821		int i;
822		for (i = 0; i < nsops; i++)
823			if (sops[i].sem_num == semnum
824			    && (sops[i].sem_op == 0)
825			    && !(sops[i].sem_flg & IPC_NOWAIT))
826				semzcnt++;
827	}
828	return semzcnt;
829}
830
831/* Free a semaphore set. freeary() is called with sem_ids.rw_mutex locked
832 * as a writer and the spinlock for this semaphore set hold. sem_ids.rw_mutex
833 * remains locked on exit.
834 */
835static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
836{
837	struct sem_undo *un, *tu;
838	struct sem_queue *q, *tq;
839	struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
840	struct list_head tasks;
841	int i;
842
843	/* Free the existing undo structures for this semaphore set.  */
844	assert_spin_locked(&sma->sem_perm.lock);
845	list_for_each_entry_safe(un, tu, &sma->list_id, list_id) {
846		list_del(&un->list_id);
847		spin_lock(&un->ulp->lock);
848		un->semid = -1;
849		list_del_rcu(&un->list_proc);
850		spin_unlock(&un->ulp->lock);
851		kfree_rcu(un, rcu);
852	}
853
854	/* Wake up all pending processes and let them fail with EIDRM. */
855	INIT_LIST_HEAD(&tasks);
856	list_for_each_entry_safe(q, tq, &sma->sem_pending, list) {
857		unlink_queue(sma, q);
858		wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
859	}
860	for (i = 0; i < sma->sem_nsems; i++) {
861		struct sem *sem = sma->sem_base + i;
862		list_for_each_entry_safe(q, tq, &sem->sem_pending, list) {
863			unlink_queue(sma, q);
864			wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
865		}
866	}
867
868	/* Remove the semaphore set from the IDR */
869	sem_rmid(ns, sma);
870	sem_unlock(sma, -1);
871	rcu_read_unlock();
872
873	wake_up_sem_queue_do(&tasks);
874	ns->used_sems -= sma->sem_nsems;
875	security_sem_free(sma);
876	ipc_rcu_putref(sma);
877}
878
879static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
880{
881	switch(version) {
882	case IPC_64:
883		return copy_to_user(buf, in, sizeof(*in));
884	case IPC_OLD:
885	    {
886		struct semid_ds out;
887
888		memset(&out, 0, sizeof(out));
889
890		ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
891
892		out.sem_otime	= in->sem_otime;
893		out.sem_ctime	= in->sem_ctime;
894		out.sem_nsems	= in->sem_nsems;
895
896		return copy_to_user(buf, &out, sizeof(out));
897	    }
898	default:
899		return -EINVAL;
900	}
901}
902
903static int semctl_nolock(struct ipc_namespace *ns, int semid,
904			 int cmd, int version, void __user *p)
905{
906	int err;
907	struct sem_array *sma;
908
909	switch(cmd) {
910	case IPC_INFO:
911	case SEM_INFO:
912	{
913		struct seminfo seminfo;
914		int max_id;
915
916		err = security_sem_semctl(NULL, cmd);
917		if (err)
918			return err;
919
920		memset(&seminfo,0,sizeof(seminfo));
921		seminfo.semmni = ns->sc_semmni;
922		seminfo.semmns = ns->sc_semmns;
923		seminfo.semmsl = ns->sc_semmsl;
924		seminfo.semopm = ns->sc_semopm;
925		seminfo.semvmx = SEMVMX;
926		seminfo.semmnu = SEMMNU;
927		seminfo.semmap = SEMMAP;
928		seminfo.semume = SEMUME;
929		down_read(&sem_ids(ns).rw_mutex);
930		if (cmd == SEM_INFO) {
931			seminfo.semusz = sem_ids(ns).in_use;
932			seminfo.semaem = ns->used_sems;
933		} else {
934			seminfo.semusz = SEMUSZ;
935			seminfo.semaem = SEMAEM;
936		}
937		max_id = ipc_get_maxid(&sem_ids(ns));
938		up_read(&sem_ids(ns).rw_mutex);
939		if (copy_to_user(p, &seminfo, sizeof(struct seminfo)))
940			return -EFAULT;
941		return (max_id < 0) ? 0: max_id;
942	}
943	case IPC_STAT:
944	case SEM_STAT:
945	{
946		struct semid64_ds tbuf;
947		int id = 0;
948
949		memset(&tbuf, 0, sizeof(tbuf));
950
951		if (cmd == SEM_STAT) {
952			rcu_read_lock();
953			sma = sem_obtain_object(ns, semid);
954			if (IS_ERR(sma)) {
955				err = PTR_ERR(sma);
956				goto out_unlock;
957			}
958			id = sma->sem_perm.id;
959		} else {
960			rcu_read_lock();
961			sma = sem_obtain_object_check(ns, semid);
962			if (IS_ERR(sma)) {
963				err = PTR_ERR(sma);
964				goto out_unlock;
965			}
966		}
967
968		err = -EACCES;
969		if (ipcperms(ns, &sma->sem_perm, S_IRUGO))
970			goto out_unlock;
971
972		err = security_sem_semctl(sma, cmd);
973		if (err)
974			goto out_unlock;
975
976		kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
977		tbuf.sem_otime  = sma->sem_otime;
978		tbuf.sem_ctime  = sma->sem_ctime;
979		tbuf.sem_nsems  = sma->sem_nsems;
980		rcu_read_unlock();
981		if (copy_semid_to_user(p, &tbuf, version))
982			return -EFAULT;
983		return id;
984	}
985	default:
986		return -EINVAL;
987	}
988out_unlock:
989	rcu_read_unlock();
990	return err;
991}
992
993static int semctl_setval(struct ipc_namespace *ns, int semid, int semnum,
994		unsigned long arg)
995{
996	struct sem_undo *un;
997	struct sem_array *sma;
998	struct sem* curr;
999	int err;
1000	struct list_head tasks;
1001	int val;
1002#if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN)
1003	/* big-endian 64bit */
1004	val = arg >> 32;
1005#else
1006	/* 32bit or little-endian 64bit */
1007	val = arg;
1008#endif
1009
1010	if (val > SEMVMX || val < 0)
1011		return -ERANGE;
1012
1013	INIT_LIST_HEAD(&tasks);
1014
1015	rcu_read_lock();
1016	sma = sem_obtain_object_check(ns, semid);
1017	if (IS_ERR(sma)) {
1018		rcu_read_unlock();
1019		return PTR_ERR(sma);
1020	}
1021
1022	if (semnum < 0 || semnum >= sma->sem_nsems) {
1023		rcu_read_unlock();
1024		return -EINVAL;
1025	}
1026
1027
1028	if (ipcperms(ns, &sma->sem_perm, S_IWUGO)) {
1029		rcu_read_unlock();
1030		return -EACCES;
1031	}
1032
1033	err = security_sem_semctl(sma, SETVAL);
1034	if (err) {
1035		rcu_read_unlock();
1036		return -EACCES;
1037	}
1038
1039	sem_lock(sma, NULL, -1);
1040
1041	curr = &sma->sem_base[semnum];
1042
1043	assert_spin_locked(&sma->sem_perm.lock);
1044	list_for_each_entry(un, &sma->list_id, list_id)
1045		un->semadj[semnum] = 0;
1046
1047	curr->semval = val;
1048	curr->sempid = task_tgid_vnr(current);
1049	sma->sem_ctime = get_seconds();
1050	/* maybe some queued-up processes were waiting for this */
1051	do_smart_update(sma, NULL, 0, 0, &tasks);
1052	sem_unlock(sma, -1);
1053	rcu_read_unlock();
1054	wake_up_sem_queue_do(&tasks);
1055	return 0;
1056}
1057
1058static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
1059		int cmd, void __user *p)
1060{
1061	struct sem_array *sma;
1062	struct sem* curr;
1063	int err, nsems;
1064	ushort fast_sem_io[SEMMSL_FAST];
1065	ushort* sem_io = fast_sem_io;
1066	struct list_head tasks;
1067
1068	INIT_LIST_HEAD(&tasks);
1069
1070	rcu_read_lock();
1071	sma = sem_obtain_object_check(ns, semid);
1072	if (IS_ERR(sma)) {
1073		rcu_read_unlock();
1074		return PTR_ERR(sma);
1075	}
1076
1077	nsems = sma->sem_nsems;
1078
1079	err = -EACCES;
1080	if (ipcperms(ns, &sma->sem_perm, cmd == SETALL ? S_IWUGO : S_IRUGO))
1081		goto out_rcu_wakeup;
1082
1083	err = security_sem_semctl(sma, cmd);
1084	if (err)
1085		goto out_rcu_wakeup;
1086
1087	err = -EACCES;
1088	switch (cmd) {
1089	case GETALL:
1090	{
1091		ushort __user *array = p;
1092		int i;
1093
1094		sem_lock(sma, NULL, -1);
1095		if(nsems > SEMMSL_FAST) {
1096			if (!ipc_rcu_getref(sma)) {
1097				sem_unlock(sma, -1);
1098				rcu_read_unlock();
1099				err = -EIDRM;
1100				goto out_free;
1101			}
1102			sem_unlock(sma, -1);
1103			rcu_read_unlock();
1104			sem_io = ipc_alloc(sizeof(ushort)*nsems);
1105			if(sem_io == NULL) {
1106				sem_putref(sma);
1107				return -ENOMEM;
1108			}
1109
1110			rcu_read_lock();
1111			sem_lock_and_putref(sma);
1112			if (sma->sem_perm.deleted) {
1113				sem_unlock(sma, -1);
1114				rcu_read_unlock();
1115				err = -EIDRM;
1116				goto out_free;
1117			}
1118		}
1119		for (i = 0; i < sma->sem_nsems; i++)
1120			sem_io[i] = sma->sem_base[i].semval;
1121		sem_unlock(sma, -1);
1122		rcu_read_unlock();
1123		err = 0;
1124		if(copy_to_user(array, sem_io, nsems*sizeof(ushort)))
1125			err = -EFAULT;
1126		goto out_free;
1127	}
1128	case SETALL:
1129	{
1130		int i;
1131		struct sem_undo *un;
1132
1133		if (!ipc_rcu_getref(sma)) {
1134			rcu_read_unlock();
1135			return -EIDRM;
1136		}
1137		rcu_read_unlock();
1138
1139		if(nsems > SEMMSL_FAST) {
1140			sem_io = ipc_alloc(sizeof(ushort)*nsems);
1141			if(sem_io == NULL) {
1142				sem_putref(sma);
1143				return -ENOMEM;
1144			}
1145		}
1146
1147		if (copy_from_user (sem_io, p, nsems*sizeof(ushort))) {
1148			sem_putref(sma);
1149			err = -EFAULT;
1150			goto out_free;
1151		}
1152
1153		for (i = 0; i < nsems; i++) {
1154			if (sem_io[i] > SEMVMX) {
1155				sem_putref(sma);
1156				err = -ERANGE;
1157				goto out_free;
1158			}
1159		}
1160		rcu_read_lock();
1161		sem_lock_and_putref(sma);
1162		if (sma->sem_perm.deleted) {
1163			sem_unlock(sma, -1);
1164			rcu_read_unlock();
1165			err = -EIDRM;
1166			goto out_free;
1167		}
1168
1169		for (i = 0; i < nsems; i++)
1170			sma->sem_base[i].semval = sem_io[i];
1171
1172		assert_spin_locked(&sma->sem_perm.lock);
1173		list_for_each_entry(un, &sma->list_id, list_id) {
1174			for (i = 0; i < nsems; i++)
1175				un->semadj[i] = 0;
1176		}
1177		sma->sem_ctime = get_seconds();
1178		/* maybe some queued-up processes were waiting for this */
1179		do_smart_update(sma, NULL, 0, 0, &tasks);
1180		err = 0;
1181		goto out_unlock;
1182	}
1183	/* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */
1184	}
1185	err = -EINVAL;
1186	if (semnum < 0 || semnum >= nsems)
1187		goto out_rcu_wakeup;
1188
1189	sem_lock(sma, NULL, -1);
1190	curr = &sma->sem_base[semnum];
1191
1192	switch (cmd) {
1193	case GETVAL:
1194		err = curr->semval;
1195		goto out_unlock;
1196	case GETPID:
1197		err = curr->sempid;
1198		goto out_unlock;
1199	case GETNCNT:
1200		err = count_semncnt(sma,semnum);
1201		goto out_unlock;
1202	case GETZCNT:
1203		err = count_semzcnt(sma,semnum);
1204		goto out_unlock;
1205	}
1206
1207out_unlock:
1208	sem_unlock(sma, -1);
1209out_rcu_wakeup:
1210	rcu_read_unlock();
1211	wake_up_sem_queue_do(&tasks);
1212out_free:
1213	if(sem_io != fast_sem_io)
1214		ipc_free(sem_io, sizeof(ushort)*nsems);
1215	return err;
1216}
1217
1218static inline unsigned long
1219copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
1220{
1221	switch(version) {
1222	case IPC_64:
1223		if (copy_from_user(out, buf, sizeof(*out)))
1224			return -EFAULT;
1225		return 0;
1226	case IPC_OLD:
1227	    {
1228		struct semid_ds tbuf_old;
1229
1230		if(copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
1231			return -EFAULT;
1232
1233		out->sem_perm.uid	= tbuf_old.sem_perm.uid;
1234		out->sem_perm.gid	= tbuf_old.sem_perm.gid;
1235		out->sem_perm.mode	= tbuf_old.sem_perm.mode;
1236
1237		return 0;
1238	    }
1239	default:
1240		return -EINVAL;
1241	}
1242}
1243
1244/*
1245 * This function handles some semctl commands which require the rw_mutex
1246 * to be held in write mode.
1247 * NOTE: no locks must be held, the rw_mutex is taken inside this function.
1248 */
1249static int semctl_down(struct ipc_namespace *ns, int semid,
1250		       int cmd, int version, void __user *p)
1251{
1252	struct sem_array *sma;
1253	int err;
1254	struct semid64_ds semid64;
1255	struct kern_ipc_perm *ipcp;
1256
1257	if(cmd == IPC_SET) {
1258		if (copy_semid_from_user(&semid64, p, version))
1259			return -EFAULT;
1260	}
1261
1262	ipcp = ipcctl_pre_down_nolock(ns, &sem_ids(ns), semid, cmd,
1263				      &semid64.sem_perm, 0);
1264	if (IS_ERR(ipcp))
1265		return PTR_ERR(ipcp);
1266
1267	sma = container_of(ipcp, struct sem_array, sem_perm);
1268
1269	err = security_sem_semctl(sma, cmd);
1270	if (err) {
1271		rcu_read_unlock();
1272		goto out_up;
1273	}
1274
1275	switch(cmd){
1276	case IPC_RMID:
1277		sem_lock(sma, NULL, -1);
1278		freeary(ns, ipcp);
1279		goto out_up;
1280	case IPC_SET:
1281		sem_lock(sma, NULL, -1);
1282		err = ipc_update_perm(&semid64.sem_perm, ipcp);
1283		if (err)
1284			goto out_unlock;
1285		sma->sem_ctime = get_seconds();
1286		break;
1287	default:
1288		rcu_read_unlock();
1289		err = -EINVAL;
1290		goto out_up;
1291	}
1292
1293out_unlock:
1294	sem_unlock(sma, -1);
1295	rcu_read_unlock();
1296out_up:
1297	up_write(&sem_ids(ns).rw_mutex);
1298	return err;
1299}
1300
1301SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, unsigned long, arg)
1302{
1303	int version;
1304	struct ipc_namespace *ns;
1305	void __user *p = (void __user *)arg;
1306
1307	if (semid < 0)
1308		return -EINVAL;
1309
1310	version = ipc_parse_version(&cmd);
1311	ns = current->nsproxy->ipc_ns;
1312
1313	switch(cmd) {
1314	case IPC_INFO:
1315	case SEM_INFO:
1316	case IPC_STAT:
1317	case SEM_STAT:
1318		return semctl_nolock(ns, semid, cmd, version, p);
1319	case GETALL:
1320	case GETVAL:
1321	case GETPID:
1322	case GETNCNT:
1323	case GETZCNT:
1324	case SETALL:
1325		return semctl_main(ns, semid, semnum, cmd, p);
1326	case SETVAL:
1327		return semctl_setval(ns, semid, semnum, arg);
1328	case IPC_RMID:
1329	case IPC_SET:
1330		return semctl_down(ns, semid, cmd, version, p);
1331	default:
1332		return -EINVAL;
1333	}
1334}
1335
1336/* If the task doesn't already have a undo_list, then allocate one
1337 * here.  We guarantee there is only one thread using this undo list,
1338 * and current is THE ONE
1339 *
1340 * If this allocation and assignment succeeds, but later
1341 * portions of this code fail, there is no need to free the sem_undo_list.
1342 * Just let it stay associated with the task, and it'll be freed later
1343 * at exit time.
1344 *
1345 * This can block, so callers must hold no locks.
1346 */
1347static inline int get_undo_list(struct sem_undo_list **undo_listp)
1348{
1349	struct sem_undo_list *undo_list;
1350
1351	undo_list = current->sysvsem.undo_list;
1352	if (!undo_list) {
1353		undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
1354		if (undo_list == NULL)
1355			return -ENOMEM;
1356		spin_lock_init(&undo_list->lock);
1357		atomic_set(&undo_list->refcnt, 1);
1358		INIT_LIST_HEAD(&undo_list->list_proc);
1359
1360		current->sysvsem.undo_list = undo_list;
1361	}
1362	*undo_listp = undo_list;
1363	return 0;
1364}
1365
1366static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid)
1367{
1368	struct sem_undo *un;
1369
1370	list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) {
1371		if (un->semid == semid)
1372			return un;
1373	}
1374	return NULL;
1375}
1376
1377static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1378{
1379	struct sem_undo *un;
1380
1381  	assert_spin_locked(&ulp->lock);
1382
1383	un = __lookup_undo(ulp, semid);
1384	if (un) {
1385		list_del_rcu(&un->list_proc);
1386		list_add_rcu(&un->list_proc, &ulp->list_proc);
1387	}
1388	return un;
1389}
1390
1391/**
1392 * find_alloc_undo - Lookup (and if not present create) undo array
1393 * @ns: namespace
1394 * @semid: semaphore array id
1395 *
1396 * The function looks up (and if not present creates) the undo structure.
1397 * The size of the undo structure depends on the size of the semaphore
1398 * array, thus the alloc path is not that straightforward.
1399 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1400 * performs a rcu_read_lock().
1401 */
1402static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
1403{
1404	struct sem_array *sma;
1405	struct sem_undo_list *ulp;
1406	struct sem_undo *un, *new;
1407	int nsems, error;
1408
1409	error = get_undo_list(&ulp);
1410	if (error)
1411		return ERR_PTR(error);
1412
1413	rcu_read_lock();
1414	spin_lock(&ulp->lock);
1415	un = lookup_undo(ulp, semid);
1416	spin_unlock(&ulp->lock);
1417	if (likely(un!=NULL))
1418		goto out;
1419
1420	/* no undo structure around - allocate one. */
1421	/* step 1: figure out the size of the semaphore array */
1422	sma = sem_obtain_object_check(ns, semid);
1423	if (IS_ERR(sma)) {
1424		rcu_read_unlock();
1425		return ERR_CAST(sma);
1426	}
1427
1428	nsems = sma->sem_nsems;
1429	if (!ipc_rcu_getref(sma)) {
1430		rcu_read_unlock();
1431		un = ERR_PTR(-EIDRM);
1432		goto out;
1433	}
1434	rcu_read_unlock();
1435
1436	/* step 2: allocate new undo structure */
1437	new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1438	if (!new) {
1439		sem_putref(sma);
1440		return ERR_PTR(-ENOMEM);
1441	}
1442
1443	/* step 3: Acquire the lock on semaphore array */
1444	rcu_read_lock();
1445	sem_lock_and_putref(sma);
1446	if (sma->sem_perm.deleted) {
1447		sem_unlock(sma, -1);
1448		rcu_read_unlock();
1449		kfree(new);
1450		un = ERR_PTR(-EIDRM);
1451		goto out;
1452	}
1453	spin_lock(&ulp->lock);
1454
1455	/*
1456	 * step 4: check for races: did someone else allocate the undo struct?
1457	 */
1458	un = lookup_undo(ulp, semid);
1459	if (un) {
1460		kfree(new);
1461		goto success;
1462	}
1463	/* step 5: initialize & link new undo structure */
1464	new->semadj = (short *) &new[1];
1465	new->ulp = ulp;
1466	new->semid = semid;
1467	assert_spin_locked(&ulp->lock);
1468	list_add_rcu(&new->list_proc, &ulp->list_proc);
1469	assert_spin_locked(&sma->sem_perm.lock);
1470	list_add(&new->list_id, &sma->list_id);
1471	un = new;
1472
1473success:
1474	spin_unlock(&ulp->lock);
1475	sem_unlock(sma, -1);
1476out:
1477	return un;
1478}
1479
1480
1481/**
1482 * get_queue_result - Retrieve the result code from sem_queue
1483 * @q: Pointer to queue structure
1484 *
1485 * Retrieve the return code from the pending queue. If IN_WAKEUP is found in
1486 * q->status, then we must loop until the value is replaced with the final
1487 * value: This may happen if a task is woken up by an unrelated event (e.g.
1488 * signal) and in parallel the task is woken up by another task because it got
1489 * the requested semaphores.
1490 *
1491 * The function can be called with or without holding the semaphore spinlock.
1492 */
1493static int get_queue_result(struct sem_queue *q)
1494{
1495	int error;
1496
1497	error = q->status;
1498	while (unlikely(error == IN_WAKEUP)) {
1499		cpu_relax();
1500		error = q->status;
1501	}
1502
1503	return error;
1504}
1505
1506
1507SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
1508		unsigned, nsops, const struct timespec __user *, timeout)
1509{
1510	int error = -EINVAL;
1511	struct sem_array *sma;
1512	struct sembuf fast_sops[SEMOPM_FAST];
1513	struct sembuf* sops = fast_sops, *sop;
1514	struct sem_undo *un;
1515	int undos = 0, alter = 0, max, locknum;
1516	struct sem_queue queue;
1517	unsigned long jiffies_left = 0;
1518	struct ipc_namespace *ns;
1519	struct list_head tasks;
1520
1521	ns = current->nsproxy->ipc_ns;
1522
1523	if (nsops < 1 || semid < 0)
1524		return -EINVAL;
1525	if (nsops > ns->sc_semopm)
1526		return -E2BIG;
1527	if(nsops > SEMOPM_FAST) {
1528		sops = kmalloc(sizeof(*sops)*nsops,GFP_KERNEL);
1529		if(sops==NULL)
1530			return -ENOMEM;
1531	}
1532	if (copy_from_user (sops, tsops, nsops * sizeof(*tsops))) {
1533		error=-EFAULT;
1534		goto out_free;
1535	}
1536	if (timeout) {
1537		struct timespec _timeout;
1538		if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1539			error = -EFAULT;
1540			goto out_free;
1541		}
1542		if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1543			_timeout.tv_nsec >= 1000000000L) {
1544			error = -EINVAL;
1545			goto out_free;
1546		}
1547		jiffies_left = timespec_to_jiffies(&_timeout);
1548	}
1549	max = 0;
1550	for (sop = sops; sop < sops + nsops; sop++) {
1551		if (sop->sem_num >= max)
1552			max = sop->sem_num;
1553		if (sop->sem_flg & SEM_UNDO)
1554			undos = 1;
1555		if (sop->sem_op != 0)
1556			alter = 1;
1557	}
1558
1559	INIT_LIST_HEAD(&tasks);
1560
1561	if (undos) {
1562		/* On success, find_alloc_undo takes the rcu_read_lock */
1563		un = find_alloc_undo(ns, semid);
1564		if (IS_ERR(un)) {
1565			error = PTR_ERR(un);
1566			goto out_free;
1567		}
1568	} else {
1569		un = NULL;
1570		rcu_read_lock();
1571	}
1572
1573	sma = sem_obtain_object_check(ns, semid);
1574	if (IS_ERR(sma)) {
1575		rcu_read_unlock();
1576		error = PTR_ERR(sma);
1577		goto out_free;
1578	}
1579
1580	error = -EFBIG;
1581	if (max >= sma->sem_nsems)
1582		goto out_rcu_wakeup;
1583
1584	error = -EACCES;
1585	if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO))
1586		goto out_rcu_wakeup;
1587
1588	error = security_sem_semop(sma, sops, nsops, alter);
1589	if (error)
1590		goto out_rcu_wakeup;
1591
1592	/*
1593	 * semid identifiers are not unique - find_alloc_undo may have
1594	 * allocated an undo structure, it was invalidated by an RMID
1595	 * and now a new array with received the same id. Check and fail.
1596	 * This case can be detected checking un->semid. The existence of
1597	 * "un" itself is guaranteed by rcu.
1598	 */
1599	error = -EIDRM;
1600	locknum = sem_lock(sma, sops, nsops);
1601	if (un && un->semid == -1)
1602		goto out_unlock_free;
1603
1604	error = try_atomic_semop (sma, sops, nsops, un, task_tgid_vnr(current));
1605	if (error <= 0) {
1606		if (alter && error == 0)
1607			do_smart_update(sma, sops, nsops, 1, &tasks);
1608
1609		goto out_unlock_free;
1610	}
1611
1612	/* We need to sleep on this operation, so we put the current
1613	 * task into the pending queue and go to sleep.
1614	 */
1615
1616	queue.sops = sops;
1617	queue.nsops = nsops;
1618	queue.undo = un;
1619	queue.pid = task_tgid_vnr(current);
1620	queue.alter = alter;
1621
1622	if (nsops == 1) {
1623		struct sem *curr;
1624		curr = &sma->sem_base[sops->sem_num];
1625
1626		if (alter)
1627			list_add_tail(&queue.list, &curr->sem_pending);
1628		else
1629			list_add(&queue.list, &curr->sem_pending);
1630	} else {
1631		if (alter)
1632			list_add_tail(&queue.list, &sma->sem_pending);
1633		else
1634			list_add(&queue.list, &sma->sem_pending);
1635		sma->complex_count++;
1636	}
1637
1638	queue.status = -EINTR;
1639	queue.sleeper = current;
1640
1641sleep_again:
1642	current->state = TASK_INTERRUPTIBLE;
1643	sem_unlock(sma, locknum);
1644	rcu_read_unlock();
1645
1646	if (timeout)
1647		jiffies_left = schedule_timeout(jiffies_left);
1648	else
1649		schedule();
1650
1651	error = get_queue_result(&queue);
1652
1653	if (error != -EINTR) {
1654		/* fast path: update_queue already obtained all requested
1655		 * resources.
1656		 * Perform a smp_mb(): User space could assume that semop()
1657		 * is a memory barrier: Without the mb(), the cpu could
1658		 * speculatively read in user space stale data that was
1659		 * overwritten by the previous owner of the semaphore.
1660		 */
1661		smp_mb();
1662
1663		goto out_free;
1664	}
1665
1666	rcu_read_lock();
1667	sma = sem_obtain_lock(ns, semid, sops, nsops, &locknum);
1668
1669	/*
1670	 * Wait until it's guaranteed that no wakeup_sem_queue_do() is ongoing.
1671	 */
1672	error = get_queue_result(&queue);
1673
1674	/*
1675	 * Array removed? If yes, leave without sem_unlock().
1676	 */
1677	if (IS_ERR(sma)) {
1678		rcu_read_unlock();
1679		goto out_free;
1680	}
1681
1682
1683	/*
1684	 * If queue.status != -EINTR we are woken up by another process.
1685	 * Leave without unlink_queue(), but with sem_unlock().
1686	 */
1687
1688	if (error != -EINTR) {
1689		goto out_unlock_free;
1690	}
1691
1692	/*
1693	 * If an interrupt occurred we have to clean up the queue
1694	 */
1695	if (timeout && jiffies_left == 0)
1696		error = -EAGAIN;
1697
1698	/*
1699	 * If the wakeup was spurious, just retry
1700	 */
1701	if (error == -EINTR && !signal_pending(current))
1702		goto sleep_again;
1703
1704	unlink_queue(sma, &queue);
1705
1706out_unlock_free:
1707	sem_unlock(sma, locknum);
1708out_rcu_wakeup:
1709	rcu_read_unlock();
1710	wake_up_sem_queue_do(&tasks);
1711out_free:
1712	if(sops != fast_sops)
1713		kfree(sops);
1714	return error;
1715}
1716
1717SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
1718		unsigned, nsops)
1719{
1720	return sys_semtimedop(semid, tsops, nsops, NULL);
1721}
1722
1723/* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
1724 * parent and child tasks.
1725 */
1726
1727int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
1728{
1729	struct sem_undo_list *undo_list;
1730	int error;
1731
1732	if (clone_flags & CLONE_SYSVSEM) {
1733		error = get_undo_list(&undo_list);
1734		if (error)
1735			return error;
1736		atomic_inc(&undo_list->refcnt);
1737		tsk->sysvsem.undo_list = undo_list;
1738	} else
1739		tsk->sysvsem.undo_list = NULL;
1740
1741	return 0;
1742}
1743
1744/*
1745 * add semadj values to semaphores, free undo structures.
1746 * undo structures are not freed when semaphore arrays are destroyed
1747 * so some of them may be out of date.
1748 * IMPLEMENTATION NOTE: There is some confusion over whether the
1749 * set of adjustments that needs to be done should be done in an atomic
1750 * manner or not. That is, if we are attempting to decrement the semval
1751 * should we queue up and wait until we can do so legally?
1752 * The original implementation attempted to do this (queue and wait).
1753 * The current implementation does not do so. The POSIX standard
1754 * and SVID should be consulted to determine what behavior is mandated.
1755 */
1756void exit_sem(struct task_struct *tsk)
1757{
1758	struct sem_undo_list *ulp;
1759
1760	ulp = tsk->sysvsem.undo_list;
1761	if (!ulp)
1762		return;
1763	tsk->sysvsem.undo_list = NULL;
1764
1765	if (!atomic_dec_and_test(&ulp->refcnt))
1766		return;
1767
1768	for (;;) {
1769		struct sem_array *sma;
1770		struct sem_undo *un;
1771		struct list_head tasks;
1772		int semid, i;
1773
1774		rcu_read_lock();
1775		un = list_entry_rcu(ulp->list_proc.next,
1776				    struct sem_undo, list_proc);
1777		if (&un->list_proc == &ulp->list_proc)
1778			semid = -1;
1779		 else
1780			semid = un->semid;
1781
1782		if (semid == -1) {
1783			rcu_read_unlock();
1784			break;
1785		}
1786
1787		sma = sem_obtain_object_check(tsk->nsproxy->ipc_ns, un->semid);
1788		/* exit_sem raced with IPC_RMID, nothing to do */
1789		if (IS_ERR(sma)) {
1790			rcu_read_unlock();
1791			continue;
1792		}
1793
1794		sem_lock(sma, NULL, -1);
1795		un = __lookup_undo(ulp, semid);
1796		if (un == NULL) {
1797			/* exit_sem raced with IPC_RMID+semget() that created
1798			 * exactly the same semid. Nothing to do.
1799			 */
1800			sem_unlock(sma, -1);
1801			rcu_read_unlock();
1802			continue;
1803		}
1804
1805		/* remove un from the linked lists */
1806		assert_spin_locked(&sma->sem_perm.lock);
1807		list_del(&un->list_id);
1808
1809		spin_lock(&ulp->lock);
1810		list_del_rcu(&un->list_proc);
1811		spin_unlock(&ulp->lock);
1812
1813		/* perform adjustments registered in un */
1814		for (i = 0; i < sma->sem_nsems; i++) {
1815			struct sem * semaphore = &sma->sem_base[i];
1816			if (un->semadj[i]) {
1817				semaphore->semval += un->semadj[i];
1818				/*
1819				 * Range checks of the new semaphore value,
1820				 * not defined by sus:
1821				 * - Some unices ignore the undo entirely
1822				 *   (e.g. HP UX 11i 11.22, Tru64 V5.1)
1823				 * - some cap the value (e.g. FreeBSD caps
1824				 *   at 0, but doesn't enforce SEMVMX)
1825				 *
1826				 * Linux caps the semaphore value, both at 0
1827				 * and at SEMVMX.
1828				 *
1829				 * 	Manfred <manfred@colorfullife.com>
1830				 */
1831				if (semaphore->semval < 0)
1832					semaphore->semval = 0;
1833				if (semaphore->semval > SEMVMX)
1834					semaphore->semval = SEMVMX;
1835				semaphore->sempid = task_tgid_vnr(current);
1836			}
1837		}
1838		/* maybe some queued-up processes were waiting for this */
1839		INIT_LIST_HEAD(&tasks);
1840		do_smart_update(sma, NULL, 0, 1, &tasks);
1841		sem_unlock(sma, -1);
1842		rcu_read_unlock();
1843		wake_up_sem_queue_do(&tasks);
1844
1845		kfree_rcu(un, rcu);
1846	}
1847	kfree(ulp);
1848}
1849
1850#ifdef CONFIG_PROC_FS
1851static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
1852{
1853	struct user_namespace *user_ns = seq_user_ns(s);
1854	struct sem_array *sma = it;
1855
1856	return seq_printf(s,
1857			  "%10d %10d  %4o %10u %5u %5u %5u %5u %10lu %10lu\n",
1858			  sma->sem_perm.key,
1859			  sma->sem_perm.id,
1860			  sma->sem_perm.mode,
1861			  sma->sem_nsems,
1862			  from_kuid_munged(user_ns, sma->sem_perm.uid),
1863			  from_kgid_munged(user_ns, sma->sem_perm.gid),
1864			  from_kuid_munged(user_ns, sma->sem_perm.cuid),
1865			  from_kgid_munged(user_ns, sma->sem_perm.cgid),
1866			  sma->sem_otime,
1867			  sma->sem_ctime);
1868}
1869#endif
1870