xref: /openbmc/linux/ipc/sem.c (revision 22246614)
1 /*
2  * linux/ipc/sem.c
3  * Copyright (C) 1992 Krishna Balasubramanian
4  * Copyright (C) 1995 Eric Schenk, Bruno Haible
5  *
6  * IMPLEMENTATION NOTES ON CODE REWRITE (Eric Schenk, January 1995):
7  * This code underwent a massive rewrite in order to solve some problems
8  * with the original code. In particular the original code failed to
9  * wake up processes that were waiting for semval to go to 0 if the
10  * value went to 0 and was then incremented rapidly enough. In solving
11  * this problem I have also modified the implementation so that it
12  * processes pending operations in a FIFO manner, thus give a guarantee
13  * that processes waiting for a lock on the semaphore won't starve
14  * unless another locking process fails to unlock.
15  * In addition the following two changes in behavior have been introduced:
16  * - The original implementation of semop returned the value
17  *   last semaphore element examined on success. This does not
18  *   match the manual page specifications, and effectively
19  *   allows the user to read the semaphore even if they do not
20  *   have read permissions. The implementation now returns 0
21  *   on success as stated in the manual page.
22  * - There is some confusion over whether the set of undo adjustments
23  *   to be performed at exit should be done in an atomic manner.
24  *   That is, if we are attempting to decrement the semval should we queue
25  *   up and wait until we can do so legally?
26  *   The original implementation attempted to do this.
27  *   The current implementation does not do so. This is because I don't
28  *   think it is the right thing (TM) to do, and because I couldn't
29  *   see a clean way to get the old behavior with the new design.
30  *   The POSIX standard and SVID should be consulted to determine
31  *   what behavior is mandated.
32  *
33  * Further notes on refinement (Christoph Rohland, December 1998):
34  * - The POSIX standard says, that the undo adjustments simply should
35  *   redo. So the current implementation is o.K.
36  * - The previous code had two flaws:
37  *   1) It actively gave the semaphore to the next waiting process
38  *      sleeping on the semaphore. Since this process did not have the
39  *      cpu this led to many unnecessary context switches and bad
40  *      performance. Now we only check which process should be able to
41  *      get the semaphore and if this process wants to reduce some
42  *      semaphore value we simply wake it up without doing the
43  *      operation. So it has to try to get it later. Thus e.g. the
44  *      running process may reacquire the semaphore during the current
45  *      time slice. If it only waits for zero or increases the semaphore,
46  *      we do the operation in advance and wake it up.
47  *   2) It did not wake up all zero waiting processes. We try to do
48  *      better but only get the semops right which only wait for zero or
49  *      increase. If there are decrement operations in the operations
50  *      array we do the same as before.
51  *
52  * With the incarnation of O(1) scheduler, it becomes unnecessary to perform
53  * check/retry algorithm for waking up blocked processes as the new scheduler
54  * is better at handling thread switch than the old one.
55  *
56  * /proc/sysvipc/sem support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
57  *
58  * SMP-threaded, sysctl's added
59  * (c) 1999 Manfred Spraul <manfred@colorfullife.com>
60  * Enforced range limit on SEM_UNDO
61  * (c) 2001 Red Hat Inc <alan@redhat.com>
62  * Lockless wakeup
63  * (c) 2003 Manfred Spraul <manfred@colorfullife.com>
64  *
65  * support for audit of ipc object properties and permission changes
66  * Dustin Kirkland <dustin.kirkland@us.ibm.com>
67  *
68  * namespaces support
69  * OpenVZ, SWsoft Inc.
70  * Pavel Emelianov <xemul@openvz.org>
71  */
72 
73 #include <linux/slab.h>
74 #include <linux/spinlock.h>
75 #include <linux/init.h>
76 #include <linux/proc_fs.h>
77 #include <linux/time.h>
78 #include <linux/security.h>
79 #include <linux/syscalls.h>
80 #include <linux/audit.h>
81 #include <linux/capability.h>
82 #include <linux/seq_file.h>
83 #include <linux/rwsem.h>
84 #include <linux/nsproxy.h>
85 #include <linux/ipc_namespace.h>
86 
87 #include <asm/uaccess.h>
88 #include "util.h"
89 
90 #define sem_ids(ns)	((ns)->ids[IPC_SEM_IDS])
91 
92 #define sem_unlock(sma)		ipc_unlock(&(sma)->sem_perm)
93 #define sem_checkid(sma, semid)	ipc_checkid(&sma->sem_perm, semid)
94 
95 static int newary(struct ipc_namespace *, struct ipc_params *);
96 static void freeary(struct ipc_namespace *, struct kern_ipc_perm *);
97 #ifdef CONFIG_PROC_FS
98 static int sysvipc_sem_proc_show(struct seq_file *s, void *it);
99 #endif
100 
101 #define SEMMSL_FAST	256 /* 512 bytes on stack */
102 #define SEMOPM_FAST	64  /* ~ 372 bytes on stack */
103 
104 /*
105  * linked list protection:
106  *	sem_undo.id_next,
107  *	sem_array.sem_pending{,last},
108  *	sem_array.sem_undo: sem_lock() for read/write
109  *	sem_undo.proc_next: only "current" is allowed to read/write that field.
110  *
111  */
112 
113 #define sc_semmsl	sem_ctls[0]
114 #define sc_semmns	sem_ctls[1]
115 #define sc_semopm	sem_ctls[2]
116 #define sc_semmni	sem_ctls[3]
117 
118 void sem_init_ns(struct ipc_namespace *ns)
119 {
120 	ns->sc_semmsl = SEMMSL;
121 	ns->sc_semmns = SEMMNS;
122 	ns->sc_semopm = SEMOPM;
123 	ns->sc_semmni = SEMMNI;
124 	ns->used_sems = 0;
125 	ipc_init_ids(&ns->ids[IPC_SEM_IDS]);
126 }
127 
128 #ifdef CONFIG_IPC_NS
129 void sem_exit_ns(struct ipc_namespace *ns)
130 {
131 	free_ipcs(ns, &sem_ids(ns), freeary);
132 }
133 #endif
134 
135 void __init sem_init (void)
136 {
137 	sem_init_ns(&init_ipc_ns);
138 	ipc_init_proc_interface("sysvipc/sem",
139 				"       key      semid perms      nsems   uid   gid  cuid  cgid      otime      ctime\n",
140 				IPC_SEM_IDS, sysvipc_sem_proc_show);
141 }
142 
143 /*
144  * sem_lock_(check_) routines are called in the paths where the rw_mutex
145  * is not held.
146  */
147 static inline struct sem_array *sem_lock(struct ipc_namespace *ns, int id)
148 {
149 	struct kern_ipc_perm *ipcp = ipc_lock(&sem_ids(ns), id);
150 
151 	if (IS_ERR(ipcp))
152 		return (struct sem_array *)ipcp;
153 
154 	return container_of(ipcp, struct sem_array, sem_perm);
155 }
156 
157 static inline struct sem_array *sem_lock_check(struct ipc_namespace *ns,
158 						int id)
159 {
160 	struct kern_ipc_perm *ipcp = ipc_lock_check(&sem_ids(ns), id);
161 
162 	if (IS_ERR(ipcp))
163 		return (struct sem_array *)ipcp;
164 
165 	return container_of(ipcp, struct sem_array, sem_perm);
166 }
167 
168 static inline void sem_lock_and_putref(struct sem_array *sma)
169 {
170 	ipc_lock_by_ptr(&sma->sem_perm);
171 	ipc_rcu_putref(sma);
172 }
173 
174 static inline void sem_getref_and_unlock(struct sem_array *sma)
175 {
176 	ipc_rcu_getref(sma);
177 	ipc_unlock(&(sma)->sem_perm);
178 }
179 
180 static inline void sem_putref(struct sem_array *sma)
181 {
182 	ipc_lock_by_ptr(&sma->sem_perm);
183 	ipc_rcu_putref(sma);
184 	ipc_unlock(&(sma)->sem_perm);
185 }
186 
187 static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
188 {
189 	ipc_rmid(&sem_ids(ns), &s->sem_perm);
190 }
191 
192 /*
193  * Lockless wakeup algorithm:
194  * Without the check/retry algorithm a lockless wakeup is possible:
195  * - queue.status is initialized to -EINTR before blocking.
196  * - wakeup is performed by
197  *	* unlinking the queue entry from sma->sem_pending
198  *	* setting queue.status to IN_WAKEUP
199  *	  This is the notification for the blocked thread that a
200  *	  result value is imminent.
201  *	* call wake_up_process
202  *	* set queue.status to the final value.
203  * - the previously blocked thread checks queue.status:
204  *   	* if it's IN_WAKEUP, then it must wait until the value changes
205  *   	* if it's not -EINTR, then the operation was completed by
206  *   	  update_queue. semtimedop can return queue.status without
207  *   	  performing any operation on the sem array.
208  *   	* otherwise it must acquire the spinlock and check what's up.
209  *
210  * The two-stage algorithm is necessary to protect against the following
211  * races:
212  * - if queue.status is set after wake_up_process, then the woken up idle
213  *   thread could race forward and try (and fail) to acquire sma->lock
214  *   before update_queue had a chance to set queue.status
215  * - if queue.status is written before wake_up_process and if the
216  *   blocked process is woken up by a signal between writing
217  *   queue.status and the wake_up_process, then the woken up
218  *   process could return from semtimedop and die by calling
219  *   sys_exit before wake_up_process is called. Then wake_up_process
220  *   will oops, because the task structure is already invalid.
221  *   (yes, this happened on s390 with sysv msg).
222  *
223  */
224 #define IN_WAKEUP	1
225 
226 /**
227  * newary - Create a new semaphore set
228  * @ns: namespace
229  * @params: ptr to the structure that contains key, semflg and nsems
230  *
231  * Called with sem_ids.rw_mutex held (as a writer)
232  */
233 
234 static int newary(struct ipc_namespace *ns, struct ipc_params *params)
235 {
236 	int id;
237 	int retval;
238 	struct sem_array *sma;
239 	int size;
240 	key_t key = params->key;
241 	int nsems = params->u.nsems;
242 	int semflg = params->flg;
243 
244 	if (!nsems)
245 		return -EINVAL;
246 	if (ns->used_sems + nsems > ns->sc_semmns)
247 		return -ENOSPC;
248 
249 	size = sizeof (*sma) + nsems * sizeof (struct sem);
250 	sma = ipc_rcu_alloc(size);
251 	if (!sma) {
252 		return -ENOMEM;
253 	}
254 	memset (sma, 0, size);
255 
256 	sma->sem_perm.mode = (semflg & S_IRWXUGO);
257 	sma->sem_perm.key = key;
258 
259 	sma->sem_perm.security = NULL;
260 	retval = security_sem_alloc(sma);
261 	if (retval) {
262 		ipc_rcu_putref(sma);
263 		return retval;
264 	}
265 
266 	id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
267 	if (id < 0) {
268 		security_sem_free(sma);
269 		ipc_rcu_putref(sma);
270 		return id;
271 	}
272 	ns->used_sems += nsems;
273 
274 	sma->sem_base = (struct sem *) &sma[1];
275 	/* sma->sem_pending = NULL; */
276 	sma->sem_pending_last = &sma->sem_pending;
277 	/* sma->undo = NULL; */
278 	sma->sem_nsems = nsems;
279 	sma->sem_ctime = get_seconds();
280 	sem_unlock(sma);
281 
282 	return sma->sem_perm.id;
283 }
284 
285 
286 /*
287  * Called with sem_ids.rw_mutex and ipcp locked.
288  */
289 static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg)
290 {
291 	struct sem_array *sma;
292 
293 	sma = container_of(ipcp, struct sem_array, sem_perm);
294 	return security_sem_associate(sma, semflg);
295 }
296 
297 /*
298  * Called with sem_ids.rw_mutex and ipcp locked.
299  */
300 static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
301 				struct ipc_params *params)
302 {
303 	struct sem_array *sma;
304 
305 	sma = container_of(ipcp, struct sem_array, sem_perm);
306 	if (params->u.nsems > sma->sem_nsems)
307 		return -EINVAL;
308 
309 	return 0;
310 }
311 
312 asmlinkage long sys_semget(key_t key, int nsems, int semflg)
313 {
314 	struct ipc_namespace *ns;
315 	struct ipc_ops sem_ops;
316 	struct ipc_params sem_params;
317 
318 	ns = current->nsproxy->ipc_ns;
319 
320 	if (nsems < 0 || nsems > ns->sc_semmsl)
321 		return -EINVAL;
322 
323 	sem_ops.getnew = newary;
324 	sem_ops.associate = sem_security;
325 	sem_ops.more_checks = sem_more_checks;
326 
327 	sem_params.key = key;
328 	sem_params.flg = semflg;
329 	sem_params.u.nsems = nsems;
330 
331 	return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
332 }
333 
334 /* Manage the doubly linked list sma->sem_pending as a FIFO:
335  * insert new queue elements at the tail sma->sem_pending_last.
336  */
337 static inline void append_to_queue (struct sem_array * sma,
338 				    struct sem_queue * q)
339 {
340 	*(q->prev = sma->sem_pending_last) = q;
341 	*(sma->sem_pending_last = &q->next) = NULL;
342 }
343 
344 static inline void prepend_to_queue (struct sem_array * sma,
345 				     struct sem_queue * q)
346 {
347 	q->next = sma->sem_pending;
348 	*(q->prev = &sma->sem_pending) = q;
349 	if (q->next)
350 		q->next->prev = &q->next;
351 	else /* sma->sem_pending_last == &sma->sem_pending */
352 		sma->sem_pending_last = &q->next;
353 }
354 
355 static inline void remove_from_queue (struct sem_array * sma,
356 				      struct sem_queue * q)
357 {
358 	*(q->prev) = q->next;
359 	if (q->next)
360 		q->next->prev = q->prev;
361 	else /* sma->sem_pending_last == &q->next */
362 		sma->sem_pending_last = q->prev;
363 	q->prev = NULL; /* mark as removed */
364 }
365 
366 /*
367  * Determine whether a sequence of semaphore operations would succeed
368  * all at once. Return 0 if yes, 1 if need to sleep, else return error code.
369  */
370 
371 static int try_atomic_semop (struct sem_array * sma, struct sembuf * sops,
372 			     int nsops, struct sem_undo *un, int pid)
373 {
374 	int result, sem_op;
375 	struct sembuf *sop;
376 	struct sem * curr;
377 
378 	for (sop = sops; sop < sops + nsops; sop++) {
379 		curr = sma->sem_base + sop->sem_num;
380 		sem_op = sop->sem_op;
381 		result = curr->semval;
382 
383 		if (!sem_op && result)
384 			goto would_block;
385 
386 		result += sem_op;
387 		if (result < 0)
388 			goto would_block;
389 		if (result > SEMVMX)
390 			goto out_of_range;
391 		if (sop->sem_flg & SEM_UNDO) {
392 			int undo = un->semadj[sop->sem_num] - sem_op;
393 			/*
394 	 		 *	Exceeding the undo range is an error.
395 			 */
396 			if (undo < (-SEMAEM - 1) || undo > SEMAEM)
397 				goto out_of_range;
398 		}
399 		curr->semval = result;
400 	}
401 
402 	sop--;
403 	while (sop >= sops) {
404 		sma->sem_base[sop->sem_num].sempid = pid;
405 		if (sop->sem_flg & SEM_UNDO)
406 			un->semadj[sop->sem_num] -= sop->sem_op;
407 		sop--;
408 	}
409 
410 	sma->sem_otime = get_seconds();
411 	return 0;
412 
413 out_of_range:
414 	result = -ERANGE;
415 	goto undo;
416 
417 would_block:
418 	if (sop->sem_flg & IPC_NOWAIT)
419 		result = -EAGAIN;
420 	else
421 		result = 1;
422 
423 undo:
424 	sop--;
425 	while (sop >= sops) {
426 		sma->sem_base[sop->sem_num].semval -= sop->sem_op;
427 		sop--;
428 	}
429 
430 	return result;
431 }
432 
433 /* Go through the pending queue for the indicated semaphore
434  * looking for tasks that can be completed.
435  */
436 static void update_queue (struct sem_array * sma)
437 {
438 	int error;
439 	struct sem_queue * q;
440 
441 	q = sma->sem_pending;
442 	while(q) {
443 		error = try_atomic_semop(sma, q->sops, q->nsops,
444 					 q->undo, q->pid);
445 
446 		/* Does q->sleeper still need to sleep? */
447 		if (error <= 0) {
448 			struct sem_queue *n;
449 			remove_from_queue(sma,q);
450 			q->status = IN_WAKEUP;
451 			/*
452 			 * Continue scanning. The next operation
453 			 * that must be checked depends on the type of the
454 			 * completed operation:
455 			 * - if the operation modified the array, then
456 			 *   restart from the head of the queue and
457 			 *   check for threads that might be waiting
458 			 *   for semaphore values to become 0.
459 			 * - if the operation didn't modify the array,
460 			 *   then just continue.
461 			 */
462 			if (q->alter)
463 				n = sma->sem_pending;
464 			else
465 				n = q->next;
466 			wake_up_process(q->sleeper);
467 			/* hands-off: q will disappear immediately after
468 			 * writing q->status.
469 			 */
470 			smp_wmb();
471 			q->status = error;
472 			q = n;
473 		} else {
474 			q = q->next;
475 		}
476 	}
477 }
478 
479 /* The following counts are associated to each semaphore:
480  *   semncnt        number of tasks waiting on semval being nonzero
481  *   semzcnt        number of tasks waiting on semval being zero
482  * This model assumes that a task waits on exactly one semaphore.
483  * Since semaphore operations are to be performed atomically, tasks actually
484  * wait on a whole sequence of semaphores simultaneously.
485  * The counts we return here are a rough approximation, but still
486  * warrant that semncnt+semzcnt>0 if the task is on the pending queue.
487  */
488 static int count_semncnt (struct sem_array * sma, ushort semnum)
489 {
490 	int semncnt;
491 	struct sem_queue * q;
492 
493 	semncnt = 0;
494 	for (q = sma->sem_pending; q; q = q->next) {
495 		struct sembuf * sops = q->sops;
496 		int nsops = q->nsops;
497 		int i;
498 		for (i = 0; i < nsops; i++)
499 			if (sops[i].sem_num == semnum
500 			    && (sops[i].sem_op < 0)
501 			    && !(sops[i].sem_flg & IPC_NOWAIT))
502 				semncnt++;
503 	}
504 	return semncnt;
505 }
506 static int count_semzcnt (struct sem_array * sma, ushort semnum)
507 {
508 	int semzcnt;
509 	struct sem_queue * q;
510 
511 	semzcnt = 0;
512 	for (q = sma->sem_pending; q; q = q->next) {
513 		struct sembuf * sops = q->sops;
514 		int nsops = q->nsops;
515 		int i;
516 		for (i = 0; i < nsops; i++)
517 			if (sops[i].sem_num == semnum
518 			    && (sops[i].sem_op == 0)
519 			    && !(sops[i].sem_flg & IPC_NOWAIT))
520 				semzcnt++;
521 	}
522 	return semzcnt;
523 }
524 
525 /* Free a semaphore set. freeary() is called with sem_ids.rw_mutex locked
526  * as a writer and the spinlock for this semaphore set hold. sem_ids.rw_mutex
527  * remains locked on exit.
528  */
529 static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
530 {
531 	struct sem_undo *un;
532 	struct sem_queue *q;
533 	struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
534 
535 	/* Invalidate the existing undo structures for this semaphore set.
536 	 * (They will be freed without any further action in exit_sem()
537 	 * or during the next semop.)
538 	 */
539 	for (un = sma->undo; un; un = un->id_next)
540 		un->semid = -1;
541 
542 	/* Wake up all pending processes and let them fail with EIDRM. */
543 	q = sma->sem_pending;
544 	while(q) {
545 		struct sem_queue *n;
546 		/* lazy remove_from_queue: we are killing the whole queue */
547 		q->prev = NULL;
548 		n = q->next;
549 		q->status = IN_WAKEUP;
550 		wake_up_process(q->sleeper); /* doesn't sleep */
551 		smp_wmb();
552 		q->status = -EIDRM;	/* hands-off q */
553 		q = n;
554 	}
555 
556 	/* Remove the semaphore set from the IDR */
557 	sem_rmid(ns, sma);
558 	sem_unlock(sma);
559 
560 	ns->used_sems -= sma->sem_nsems;
561 	security_sem_free(sma);
562 	ipc_rcu_putref(sma);
563 }
564 
565 static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
566 {
567 	switch(version) {
568 	case IPC_64:
569 		return copy_to_user(buf, in, sizeof(*in));
570 	case IPC_OLD:
571 	    {
572 		struct semid_ds out;
573 
574 		ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
575 
576 		out.sem_otime	= in->sem_otime;
577 		out.sem_ctime	= in->sem_ctime;
578 		out.sem_nsems	= in->sem_nsems;
579 
580 		return copy_to_user(buf, &out, sizeof(out));
581 	    }
582 	default:
583 		return -EINVAL;
584 	}
585 }
586 
587 static int semctl_nolock(struct ipc_namespace *ns, int semid,
588 			 int cmd, int version, union semun arg)
589 {
590 	int err = -EINVAL;
591 	struct sem_array *sma;
592 
593 	switch(cmd) {
594 	case IPC_INFO:
595 	case SEM_INFO:
596 	{
597 		struct seminfo seminfo;
598 		int max_id;
599 
600 		err = security_sem_semctl(NULL, cmd);
601 		if (err)
602 			return err;
603 
604 		memset(&seminfo,0,sizeof(seminfo));
605 		seminfo.semmni = ns->sc_semmni;
606 		seminfo.semmns = ns->sc_semmns;
607 		seminfo.semmsl = ns->sc_semmsl;
608 		seminfo.semopm = ns->sc_semopm;
609 		seminfo.semvmx = SEMVMX;
610 		seminfo.semmnu = SEMMNU;
611 		seminfo.semmap = SEMMAP;
612 		seminfo.semume = SEMUME;
613 		down_read(&sem_ids(ns).rw_mutex);
614 		if (cmd == SEM_INFO) {
615 			seminfo.semusz = sem_ids(ns).in_use;
616 			seminfo.semaem = ns->used_sems;
617 		} else {
618 			seminfo.semusz = SEMUSZ;
619 			seminfo.semaem = SEMAEM;
620 		}
621 		max_id = ipc_get_maxid(&sem_ids(ns));
622 		up_read(&sem_ids(ns).rw_mutex);
623 		if (copy_to_user (arg.__buf, &seminfo, sizeof(struct seminfo)))
624 			return -EFAULT;
625 		return (max_id < 0) ? 0: max_id;
626 	}
627 	case IPC_STAT:
628 	case SEM_STAT:
629 	{
630 		struct semid64_ds tbuf;
631 		int id;
632 
633 		if (cmd == SEM_STAT) {
634 			sma = sem_lock(ns, semid);
635 			if (IS_ERR(sma))
636 				return PTR_ERR(sma);
637 			id = sma->sem_perm.id;
638 		} else {
639 			sma = sem_lock_check(ns, semid);
640 			if (IS_ERR(sma))
641 				return PTR_ERR(sma);
642 			id = 0;
643 		}
644 
645 		err = -EACCES;
646 		if (ipcperms (&sma->sem_perm, S_IRUGO))
647 			goto out_unlock;
648 
649 		err = security_sem_semctl(sma, cmd);
650 		if (err)
651 			goto out_unlock;
652 
653 		memset(&tbuf, 0, sizeof(tbuf));
654 
655 		kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
656 		tbuf.sem_otime  = sma->sem_otime;
657 		tbuf.sem_ctime  = sma->sem_ctime;
658 		tbuf.sem_nsems  = sma->sem_nsems;
659 		sem_unlock(sma);
660 		if (copy_semid_to_user (arg.buf, &tbuf, version))
661 			return -EFAULT;
662 		return id;
663 	}
664 	default:
665 		return -EINVAL;
666 	}
667 	return err;
668 out_unlock:
669 	sem_unlock(sma);
670 	return err;
671 }
672 
673 static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
674 		int cmd, int version, union semun arg)
675 {
676 	struct sem_array *sma;
677 	struct sem* curr;
678 	int err;
679 	ushort fast_sem_io[SEMMSL_FAST];
680 	ushort* sem_io = fast_sem_io;
681 	int nsems;
682 
683 	sma = sem_lock_check(ns, semid);
684 	if (IS_ERR(sma))
685 		return PTR_ERR(sma);
686 
687 	nsems = sma->sem_nsems;
688 
689 	err = -EACCES;
690 	if (ipcperms (&sma->sem_perm, (cmd==SETVAL||cmd==SETALL)?S_IWUGO:S_IRUGO))
691 		goto out_unlock;
692 
693 	err = security_sem_semctl(sma, cmd);
694 	if (err)
695 		goto out_unlock;
696 
697 	err = -EACCES;
698 	switch (cmd) {
699 	case GETALL:
700 	{
701 		ushort __user *array = arg.array;
702 		int i;
703 
704 		if(nsems > SEMMSL_FAST) {
705 			sem_getref_and_unlock(sma);
706 
707 			sem_io = ipc_alloc(sizeof(ushort)*nsems);
708 			if(sem_io == NULL) {
709 				sem_putref(sma);
710 				return -ENOMEM;
711 			}
712 
713 			sem_lock_and_putref(sma);
714 			if (sma->sem_perm.deleted) {
715 				sem_unlock(sma);
716 				err = -EIDRM;
717 				goto out_free;
718 			}
719 		}
720 
721 		for (i = 0; i < sma->sem_nsems; i++)
722 			sem_io[i] = sma->sem_base[i].semval;
723 		sem_unlock(sma);
724 		err = 0;
725 		if(copy_to_user(array, sem_io, nsems*sizeof(ushort)))
726 			err = -EFAULT;
727 		goto out_free;
728 	}
729 	case SETALL:
730 	{
731 		int i;
732 		struct sem_undo *un;
733 
734 		sem_getref_and_unlock(sma);
735 
736 		if(nsems > SEMMSL_FAST) {
737 			sem_io = ipc_alloc(sizeof(ushort)*nsems);
738 			if(sem_io == NULL) {
739 				sem_putref(sma);
740 				return -ENOMEM;
741 			}
742 		}
743 
744 		if (copy_from_user (sem_io, arg.array, nsems*sizeof(ushort))) {
745 			sem_putref(sma);
746 			err = -EFAULT;
747 			goto out_free;
748 		}
749 
750 		for (i = 0; i < nsems; i++) {
751 			if (sem_io[i] > SEMVMX) {
752 				sem_putref(sma);
753 				err = -ERANGE;
754 				goto out_free;
755 			}
756 		}
757 		sem_lock_and_putref(sma);
758 		if (sma->sem_perm.deleted) {
759 			sem_unlock(sma);
760 			err = -EIDRM;
761 			goto out_free;
762 		}
763 
764 		for (i = 0; i < nsems; i++)
765 			sma->sem_base[i].semval = sem_io[i];
766 		for (un = sma->undo; un; un = un->id_next)
767 			for (i = 0; i < nsems; i++)
768 				un->semadj[i] = 0;
769 		sma->sem_ctime = get_seconds();
770 		/* maybe some queued-up processes were waiting for this */
771 		update_queue(sma);
772 		err = 0;
773 		goto out_unlock;
774 	}
775 	/* GETVAL, GETPID, GETNCTN, GETZCNT, SETVAL: fall-through */
776 	}
777 	err = -EINVAL;
778 	if(semnum < 0 || semnum >= nsems)
779 		goto out_unlock;
780 
781 	curr = &sma->sem_base[semnum];
782 
783 	switch (cmd) {
784 	case GETVAL:
785 		err = curr->semval;
786 		goto out_unlock;
787 	case GETPID:
788 		err = curr->sempid;
789 		goto out_unlock;
790 	case GETNCNT:
791 		err = count_semncnt(sma,semnum);
792 		goto out_unlock;
793 	case GETZCNT:
794 		err = count_semzcnt(sma,semnum);
795 		goto out_unlock;
796 	case SETVAL:
797 	{
798 		int val = arg.val;
799 		struct sem_undo *un;
800 		err = -ERANGE;
801 		if (val > SEMVMX || val < 0)
802 			goto out_unlock;
803 
804 		for (un = sma->undo; un; un = un->id_next)
805 			un->semadj[semnum] = 0;
806 		curr->semval = val;
807 		curr->sempid = task_tgid_vnr(current);
808 		sma->sem_ctime = get_seconds();
809 		/* maybe some queued-up processes were waiting for this */
810 		update_queue(sma);
811 		err = 0;
812 		goto out_unlock;
813 	}
814 	}
815 out_unlock:
816 	sem_unlock(sma);
817 out_free:
818 	if(sem_io != fast_sem_io)
819 		ipc_free(sem_io, sizeof(ushort)*nsems);
820 	return err;
821 }
822 
823 static inline unsigned long
824 copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
825 {
826 	switch(version) {
827 	case IPC_64:
828 		if (copy_from_user(out, buf, sizeof(*out)))
829 			return -EFAULT;
830 		return 0;
831 	case IPC_OLD:
832 	    {
833 		struct semid_ds tbuf_old;
834 
835 		if(copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
836 			return -EFAULT;
837 
838 		out->sem_perm.uid	= tbuf_old.sem_perm.uid;
839 		out->sem_perm.gid	= tbuf_old.sem_perm.gid;
840 		out->sem_perm.mode	= tbuf_old.sem_perm.mode;
841 
842 		return 0;
843 	    }
844 	default:
845 		return -EINVAL;
846 	}
847 }
848 
849 /*
850  * This function handles some semctl commands which require the rw_mutex
851  * to be held in write mode.
852  * NOTE: no locks must be held, the rw_mutex is taken inside this function.
853  */
854 static int semctl_down(struct ipc_namespace *ns, int semid,
855 		       int cmd, int version, union semun arg)
856 {
857 	struct sem_array *sma;
858 	int err;
859 	struct semid64_ds semid64;
860 	struct kern_ipc_perm *ipcp;
861 
862 	if(cmd == IPC_SET) {
863 		if (copy_semid_from_user(&semid64, arg.buf, version))
864 			return -EFAULT;
865 	}
866 
867 	ipcp = ipcctl_pre_down(&sem_ids(ns), semid, cmd, &semid64.sem_perm, 0);
868 	if (IS_ERR(ipcp))
869 		return PTR_ERR(ipcp);
870 
871 	sma = container_of(ipcp, struct sem_array, sem_perm);
872 
873 	err = security_sem_semctl(sma, cmd);
874 	if (err)
875 		goto out_unlock;
876 
877 	switch(cmd){
878 	case IPC_RMID:
879 		freeary(ns, ipcp);
880 		goto out_up;
881 	case IPC_SET:
882 		ipc_update_perm(&semid64.sem_perm, ipcp);
883 		sma->sem_ctime = get_seconds();
884 		break;
885 	default:
886 		err = -EINVAL;
887 	}
888 
889 out_unlock:
890 	sem_unlock(sma);
891 out_up:
892 	up_write(&sem_ids(ns).rw_mutex);
893 	return err;
894 }
895 
896 asmlinkage long sys_semctl (int semid, int semnum, int cmd, union semun arg)
897 {
898 	int err = -EINVAL;
899 	int version;
900 	struct ipc_namespace *ns;
901 
902 	if (semid < 0)
903 		return -EINVAL;
904 
905 	version = ipc_parse_version(&cmd);
906 	ns = current->nsproxy->ipc_ns;
907 
908 	switch(cmd) {
909 	case IPC_INFO:
910 	case SEM_INFO:
911 	case IPC_STAT:
912 	case SEM_STAT:
913 		err = semctl_nolock(ns, semid, cmd, version, arg);
914 		return err;
915 	case GETALL:
916 	case GETVAL:
917 	case GETPID:
918 	case GETNCNT:
919 	case GETZCNT:
920 	case SETVAL:
921 	case SETALL:
922 		err = semctl_main(ns,semid,semnum,cmd,version,arg);
923 		return err;
924 	case IPC_RMID:
925 	case IPC_SET:
926 		err = semctl_down(ns, semid, cmd, version, arg);
927 		return err;
928 	default:
929 		return -EINVAL;
930 	}
931 }
932 
933 /* If the task doesn't already have a undo_list, then allocate one
934  * here.  We guarantee there is only one thread using this undo list,
935  * and current is THE ONE
936  *
937  * If this allocation and assignment succeeds, but later
938  * portions of this code fail, there is no need to free the sem_undo_list.
939  * Just let it stay associated with the task, and it'll be freed later
940  * at exit time.
941  *
942  * This can block, so callers must hold no locks.
943  */
944 static inline int get_undo_list(struct sem_undo_list **undo_listp)
945 {
946 	struct sem_undo_list *undo_list;
947 
948 	undo_list = current->sysvsem.undo_list;
949 	if (!undo_list) {
950 		undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
951 		if (undo_list == NULL)
952 			return -ENOMEM;
953 		spin_lock_init(&undo_list->lock);
954 		atomic_set(&undo_list->refcnt, 1);
955 		current->sysvsem.undo_list = undo_list;
956 	}
957 	*undo_listp = undo_list;
958 	return 0;
959 }
960 
961 static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
962 {
963 	struct sem_undo **last, *un;
964 
965 	last = &ulp->proc_list;
966 	un = *last;
967 	while(un != NULL) {
968 		if(un->semid==semid)
969 			break;
970 		if(un->semid==-1) {
971 			*last=un->proc_next;
972 			kfree(un);
973 		} else {
974 			last=&un->proc_next;
975 		}
976 		un=*last;
977 	}
978 	return un;
979 }
980 
981 static struct sem_undo *find_undo(struct ipc_namespace *ns, int semid)
982 {
983 	struct sem_array *sma;
984 	struct sem_undo_list *ulp;
985 	struct sem_undo *un, *new;
986 	int nsems;
987 	int error;
988 
989 	error = get_undo_list(&ulp);
990 	if (error)
991 		return ERR_PTR(error);
992 
993 	spin_lock(&ulp->lock);
994 	un = lookup_undo(ulp, semid);
995 	spin_unlock(&ulp->lock);
996 	if (likely(un!=NULL))
997 		goto out;
998 
999 	/* no undo structure around - allocate one. */
1000 	sma = sem_lock_check(ns, semid);
1001 	if (IS_ERR(sma))
1002 		return ERR_PTR(PTR_ERR(sma));
1003 
1004 	nsems = sma->sem_nsems;
1005 	sem_getref_and_unlock(sma);
1006 
1007 	new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1008 	if (!new) {
1009 		sem_putref(sma);
1010 		return ERR_PTR(-ENOMEM);
1011 	}
1012 	new->semadj = (short *) &new[1];
1013 	new->semid = semid;
1014 
1015 	spin_lock(&ulp->lock);
1016 	un = lookup_undo(ulp, semid);
1017 	if (un) {
1018 		spin_unlock(&ulp->lock);
1019 		kfree(new);
1020 		sem_putref(sma);
1021 		goto out;
1022 	}
1023 	sem_lock_and_putref(sma);
1024 	if (sma->sem_perm.deleted) {
1025 		sem_unlock(sma);
1026 		spin_unlock(&ulp->lock);
1027 		kfree(new);
1028 		un = ERR_PTR(-EIDRM);
1029 		goto out;
1030 	}
1031 	new->proc_next = ulp->proc_list;
1032 	ulp->proc_list = new;
1033 	new->id_next = sma->undo;
1034 	sma->undo = new;
1035 	sem_unlock(sma);
1036 	un = new;
1037 	spin_unlock(&ulp->lock);
1038 out:
1039 	return un;
1040 }
1041 
1042 asmlinkage long sys_semtimedop(int semid, struct sembuf __user *tsops,
1043 			unsigned nsops, const struct timespec __user *timeout)
1044 {
1045 	int error = -EINVAL;
1046 	struct sem_array *sma;
1047 	struct sembuf fast_sops[SEMOPM_FAST];
1048 	struct sembuf* sops = fast_sops, *sop;
1049 	struct sem_undo *un;
1050 	int undos = 0, alter = 0, max;
1051 	struct sem_queue queue;
1052 	unsigned long jiffies_left = 0;
1053 	struct ipc_namespace *ns;
1054 
1055 	ns = current->nsproxy->ipc_ns;
1056 
1057 	if (nsops < 1 || semid < 0)
1058 		return -EINVAL;
1059 	if (nsops > ns->sc_semopm)
1060 		return -E2BIG;
1061 	if(nsops > SEMOPM_FAST) {
1062 		sops = kmalloc(sizeof(*sops)*nsops,GFP_KERNEL);
1063 		if(sops==NULL)
1064 			return -ENOMEM;
1065 	}
1066 	if (copy_from_user (sops, tsops, nsops * sizeof(*tsops))) {
1067 		error=-EFAULT;
1068 		goto out_free;
1069 	}
1070 	if (timeout) {
1071 		struct timespec _timeout;
1072 		if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1073 			error = -EFAULT;
1074 			goto out_free;
1075 		}
1076 		if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1077 			_timeout.tv_nsec >= 1000000000L) {
1078 			error = -EINVAL;
1079 			goto out_free;
1080 		}
1081 		jiffies_left = timespec_to_jiffies(&_timeout);
1082 	}
1083 	max = 0;
1084 	for (sop = sops; sop < sops + nsops; sop++) {
1085 		if (sop->sem_num >= max)
1086 			max = sop->sem_num;
1087 		if (sop->sem_flg & SEM_UNDO)
1088 			undos = 1;
1089 		if (sop->sem_op != 0)
1090 			alter = 1;
1091 	}
1092 
1093 retry_undos:
1094 	if (undos) {
1095 		un = find_undo(ns, semid);
1096 		if (IS_ERR(un)) {
1097 			error = PTR_ERR(un);
1098 			goto out_free;
1099 		}
1100 	} else
1101 		un = NULL;
1102 
1103 	sma = sem_lock_check(ns, semid);
1104 	if (IS_ERR(sma)) {
1105 		error = PTR_ERR(sma);
1106 		goto out_free;
1107 	}
1108 
1109 	/*
1110 	 * semid identifiers are not unique - find_undo may have
1111 	 * allocated an undo structure, it was invalidated by an RMID
1112 	 * and now a new array with received the same id. Check and retry.
1113 	 */
1114 	if (un && un->semid == -1) {
1115 		sem_unlock(sma);
1116 		goto retry_undos;
1117 	}
1118 	error = -EFBIG;
1119 	if (max >= sma->sem_nsems)
1120 		goto out_unlock_free;
1121 
1122 	error = -EACCES;
1123 	if (ipcperms(&sma->sem_perm, alter ? S_IWUGO : S_IRUGO))
1124 		goto out_unlock_free;
1125 
1126 	error = security_sem_semop(sma, sops, nsops, alter);
1127 	if (error)
1128 		goto out_unlock_free;
1129 
1130 	error = try_atomic_semop (sma, sops, nsops, un, task_tgid_vnr(current));
1131 	if (error <= 0) {
1132 		if (alter && error == 0)
1133 			update_queue (sma);
1134 		goto out_unlock_free;
1135 	}
1136 
1137 	/* We need to sleep on this operation, so we put the current
1138 	 * task into the pending queue and go to sleep.
1139 	 */
1140 
1141 	queue.sma = sma;
1142 	queue.sops = sops;
1143 	queue.nsops = nsops;
1144 	queue.undo = un;
1145 	queue.pid = task_tgid_vnr(current);
1146 	queue.id = semid;
1147 	queue.alter = alter;
1148 	if (alter)
1149 		append_to_queue(sma ,&queue);
1150 	else
1151 		prepend_to_queue(sma ,&queue);
1152 
1153 	queue.status = -EINTR;
1154 	queue.sleeper = current;
1155 	current->state = TASK_INTERRUPTIBLE;
1156 	sem_unlock(sma);
1157 
1158 	if (timeout)
1159 		jiffies_left = schedule_timeout(jiffies_left);
1160 	else
1161 		schedule();
1162 
1163 	error = queue.status;
1164 	while(unlikely(error == IN_WAKEUP)) {
1165 		cpu_relax();
1166 		error = queue.status;
1167 	}
1168 
1169 	if (error != -EINTR) {
1170 		/* fast path: update_queue already obtained all requested
1171 		 * resources */
1172 		goto out_free;
1173 	}
1174 
1175 	sma = sem_lock(ns, semid);
1176 	if (IS_ERR(sma)) {
1177 		BUG_ON(queue.prev != NULL);
1178 		error = -EIDRM;
1179 		goto out_free;
1180 	}
1181 
1182 	/*
1183 	 * If queue.status != -EINTR we are woken up by another process
1184 	 */
1185 	error = queue.status;
1186 	if (error != -EINTR) {
1187 		goto out_unlock_free;
1188 	}
1189 
1190 	/*
1191 	 * If an interrupt occurred we have to clean up the queue
1192 	 */
1193 	if (timeout && jiffies_left == 0)
1194 		error = -EAGAIN;
1195 	remove_from_queue(sma,&queue);
1196 	goto out_unlock_free;
1197 
1198 out_unlock_free:
1199 	sem_unlock(sma);
1200 out_free:
1201 	if(sops != fast_sops)
1202 		kfree(sops);
1203 	return error;
1204 }
1205 
1206 asmlinkage long sys_semop (int semid, struct sembuf __user *tsops, unsigned nsops)
1207 {
1208 	return sys_semtimedop(semid, tsops, nsops, NULL);
1209 }
1210 
1211 /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
1212  * parent and child tasks.
1213  */
1214 
1215 int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
1216 {
1217 	struct sem_undo_list *undo_list;
1218 	int error;
1219 
1220 	if (clone_flags & CLONE_SYSVSEM) {
1221 		error = get_undo_list(&undo_list);
1222 		if (error)
1223 			return error;
1224 		atomic_inc(&undo_list->refcnt);
1225 		tsk->sysvsem.undo_list = undo_list;
1226 	} else
1227 		tsk->sysvsem.undo_list = NULL;
1228 
1229 	return 0;
1230 }
1231 
1232 /*
1233  * add semadj values to semaphores, free undo structures.
1234  * undo structures are not freed when semaphore arrays are destroyed
1235  * so some of them may be out of date.
1236  * IMPLEMENTATION NOTE: There is some confusion over whether the
1237  * set of adjustments that needs to be done should be done in an atomic
1238  * manner or not. That is, if we are attempting to decrement the semval
1239  * should we queue up and wait until we can do so legally?
1240  * The original implementation attempted to do this (queue and wait).
1241  * The current implementation does not do so. The POSIX standard
1242  * and SVID should be consulted to determine what behavior is mandated.
1243  */
1244 void exit_sem(struct task_struct *tsk)
1245 {
1246 	struct sem_undo_list *undo_list;
1247 	struct sem_undo *u, **up;
1248 	struct ipc_namespace *ns;
1249 
1250 	undo_list = tsk->sysvsem.undo_list;
1251 	if (!undo_list)
1252 		return;
1253 	tsk->sysvsem.undo_list = NULL;
1254 
1255 	if (!atomic_dec_and_test(&undo_list->refcnt))
1256 		return;
1257 
1258 	ns = tsk->nsproxy->ipc_ns;
1259 	/* There's no need to hold the semundo list lock, as current
1260          * is the last task exiting for this undo list.
1261 	 */
1262 	for (up = &undo_list->proc_list; (u = *up); *up = u->proc_next, kfree(u)) {
1263 		struct sem_array *sma;
1264 		int nsems, i;
1265 		struct sem_undo *un, **unp;
1266 		int semid;
1267 
1268 		semid = u->semid;
1269 
1270 		if(semid == -1)
1271 			continue;
1272 		sma = sem_lock(ns, semid);
1273 		if (IS_ERR(sma))
1274 			continue;
1275 
1276 		if (u->semid == -1)
1277 			goto next_entry;
1278 
1279 		BUG_ON(sem_checkid(sma, u->semid));
1280 
1281 		/* remove u from the sma->undo list */
1282 		for (unp = &sma->undo; (un = *unp); unp = &un->id_next) {
1283 			if (u == un)
1284 				goto found;
1285 		}
1286 		printk ("exit_sem undo list error id=%d\n", u->semid);
1287 		goto next_entry;
1288 found:
1289 		*unp = un->id_next;
1290 		/* perform adjustments registered in u */
1291 		nsems = sma->sem_nsems;
1292 		for (i = 0; i < nsems; i++) {
1293 			struct sem * semaphore = &sma->sem_base[i];
1294 			if (u->semadj[i]) {
1295 				semaphore->semval += u->semadj[i];
1296 				/*
1297 				 * Range checks of the new semaphore value,
1298 				 * not defined by sus:
1299 				 * - Some unices ignore the undo entirely
1300 				 *   (e.g. HP UX 11i 11.22, Tru64 V5.1)
1301 				 * - some cap the value (e.g. FreeBSD caps
1302 				 *   at 0, but doesn't enforce SEMVMX)
1303 				 *
1304 				 * Linux caps the semaphore value, both at 0
1305 				 * and at SEMVMX.
1306 				 *
1307 				 * 	Manfred <manfred@colorfullife.com>
1308 				 */
1309 				if (semaphore->semval < 0)
1310 					semaphore->semval = 0;
1311 				if (semaphore->semval > SEMVMX)
1312 					semaphore->semval = SEMVMX;
1313 				semaphore->sempid = task_tgid_vnr(current);
1314 			}
1315 		}
1316 		sma->sem_otime = get_seconds();
1317 		/* maybe some queued-up processes were waiting for this */
1318 		update_queue(sma);
1319 next_entry:
1320 		sem_unlock(sma);
1321 	}
1322 	kfree(undo_list);
1323 }
1324 
1325 #ifdef CONFIG_PROC_FS
1326 static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
1327 {
1328 	struct sem_array *sma = it;
1329 
1330 	return seq_printf(s,
1331 			  "%10d %10d  %4o %10lu %5u %5u %5u %5u %10lu %10lu\n",
1332 			  sma->sem_perm.key,
1333 			  sma->sem_perm.id,
1334 			  sma->sem_perm.mode,
1335 			  sma->sem_nsems,
1336 			  sma->sem_perm.uid,
1337 			  sma->sem_perm.gid,
1338 			  sma->sem_perm.cuid,
1339 			  sma->sem_perm.cgid,
1340 			  sma->sem_otime,
1341 			  sma->sem_ctime);
1342 }
1343 #endif
1344