xref: /openbmc/linux/ipc/sem.c (revision d2999e1b)
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_semcnt()
51  * - the task that performs a successful semop() scans the list of all
52  *   sleeping tasks and completes any pending operations that can be fulfilled.
53  *   Semaphores are actively given to waiting tasks (necessary for FIFO).
54  *   (see update_queue())
55  * - To improve the scalability, the actual wake-up calls are performed after
56  *   dropping all locks. (see wake_up_sem_queue_prepare(),
57  *   wake_up_sem_queue_do())
58  * - All work is done by the waker, the woken up task does not have to do
59  *   anything - not even acquiring a lock or dropping a refcount.
60  * - A woken up task may not even touch the semaphore array anymore, it may
61  *   have been destroyed already by a semctl(RMID).
62  * - The synchronizations between wake-ups due to a timeout/signal and a
63  *   wake-up due to a completed semaphore operation is achieved by using an
64  *   intermediate state (IN_WAKEUP).
65  * - UNDO values are stored in an array (one per process and per
66  *   semaphore array, lazily allocated). For backwards compatibility, multiple
67  *   modes for the UNDO variables are supported (per process, per thread)
68  *   (see copy_semundo, CLONE_SYSVSEM)
69  * - There are two lists of the pending operations: a per-array list
70  *   and per-semaphore list (stored in the array). This allows to achieve FIFO
71  *   ordering without always scanning all pending operations.
72  *   The worst-case behavior is nevertheless O(N^2) for N wakeups.
73  */
74 
75 #include <linux/slab.h>
76 #include <linux/spinlock.h>
77 #include <linux/init.h>
78 #include <linux/proc_fs.h>
79 #include <linux/time.h>
80 #include <linux/security.h>
81 #include <linux/syscalls.h>
82 #include <linux/audit.h>
83 #include <linux/capability.h>
84 #include <linux/seq_file.h>
85 #include <linux/rwsem.h>
86 #include <linux/nsproxy.h>
87 #include <linux/ipc_namespace.h>
88 
89 #include <linux/uaccess.h>
90 #include "util.h"
91 
92 /* One semaphore structure for each semaphore in the system. */
93 struct sem {
94 	int	semval;		/* current value */
95 	int	sempid;		/* pid of last operation */
96 	spinlock_t	lock;	/* spinlock for fine-grained semtimedop */
97 	struct list_head pending_alter; /* pending single-sop operations */
98 					/* that alter the semaphore */
99 	struct list_head pending_const; /* pending single-sop operations */
100 					/* that do not alter the semaphore*/
101 	time_t	sem_otime;	/* candidate for sem_otime */
102 } ____cacheline_aligned_in_smp;
103 
104 /* One queue for each sleeping process in the system. */
105 struct sem_queue {
106 	struct list_head	list;	 /* queue of pending operations */
107 	struct task_struct	*sleeper; /* this process */
108 	struct sem_undo		*undo;	 /* undo structure */
109 	int			pid;	 /* process id of requesting process */
110 	int			status;	 /* completion status of operation */
111 	struct sembuf		*sops;	 /* array of pending operations */
112 	struct sembuf		*blocking; /* the operation that blocked */
113 	int			nsops;	 /* number of operations */
114 	int			alter;	 /* does *sops alter the array? */
115 };
116 
117 /* Each task has a list of undo requests. They are executed automatically
118  * when the process exits.
119  */
120 struct sem_undo {
121 	struct list_head	list_proc;	/* per-process list: *
122 						 * all undos from one process
123 						 * rcu protected */
124 	struct rcu_head		rcu;		/* rcu struct for sem_undo */
125 	struct sem_undo_list	*ulp;		/* back ptr to sem_undo_list */
126 	struct list_head	list_id;	/* per semaphore array list:
127 						 * all undos for one array */
128 	int			semid;		/* semaphore set identifier */
129 	short			*semadj;	/* array of adjustments */
130 						/* one per semaphore */
131 };
132 
133 /* sem_undo_list controls shared access to the list of sem_undo structures
134  * that may be shared among all a CLONE_SYSVSEM task group.
135  */
136 struct sem_undo_list {
137 	atomic_t		refcnt;
138 	spinlock_t		lock;
139 	struct list_head	list_proc;
140 };
141 
142 
143 #define sem_ids(ns)	((ns)->ids[IPC_SEM_IDS])
144 
145 #define sem_checkid(sma, semid)	ipc_checkid(&sma->sem_perm, semid)
146 
147 static int newary(struct ipc_namespace *, struct ipc_params *);
148 static void freeary(struct ipc_namespace *, struct kern_ipc_perm *);
149 #ifdef CONFIG_PROC_FS
150 static int sysvipc_sem_proc_show(struct seq_file *s, void *it);
151 #endif
152 
153 #define SEMMSL_FAST	256 /* 512 bytes on stack */
154 #define SEMOPM_FAST	64  /* ~ 372 bytes on stack */
155 
156 /*
157  * Locking:
158  *	sem_undo.id_next,
159  *	sem_array.complex_count,
160  *	sem_array.pending{_alter,_cont},
161  *	sem_array.sem_undo: global sem_lock() for read/write
162  *	sem_undo.proc_next: only "current" is allowed to read/write that field.
163  *
164  *	sem_array.sem_base[i].pending_{const,alter}:
165  *		global or semaphore sem_lock() for read/write
166  */
167 
168 #define sc_semmsl	sem_ctls[0]
169 #define sc_semmns	sem_ctls[1]
170 #define sc_semopm	sem_ctls[2]
171 #define sc_semmni	sem_ctls[3]
172 
173 void sem_init_ns(struct ipc_namespace *ns)
174 {
175 	ns->sc_semmsl = SEMMSL;
176 	ns->sc_semmns = SEMMNS;
177 	ns->sc_semopm = SEMOPM;
178 	ns->sc_semmni = SEMMNI;
179 	ns->used_sems = 0;
180 	ipc_init_ids(&ns->ids[IPC_SEM_IDS]);
181 }
182 
183 #ifdef CONFIG_IPC_NS
184 void sem_exit_ns(struct ipc_namespace *ns)
185 {
186 	free_ipcs(ns, &sem_ids(ns), freeary);
187 	idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr);
188 }
189 #endif
190 
191 void __init sem_init(void)
192 {
193 	sem_init_ns(&init_ipc_ns);
194 	ipc_init_proc_interface("sysvipc/sem",
195 				"       key      semid perms      nsems   uid   gid  cuid  cgid      otime      ctime\n",
196 				IPC_SEM_IDS, sysvipc_sem_proc_show);
197 }
198 
199 /**
200  * unmerge_queues - unmerge queues, if possible.
201  * @sma: semaphore array
202  *
203  * The function unmerges the wait queues if complex_count is 0.
204  * It must be called prior to dropping the global semaphore array lock.
205  */
206 static void unmerge_queues(struct sem_array *sma)
207 {
208 	struct sem_queue *q, *tq;
209 
210 	/* complex operations still around? */
211 	if (sma->complex_count)
212 		return;
213 	/*
214 	 * We will switch back to simple mode.
215 	 * Move all pending operation back into the per-semaphore
216 	 * queues.
217 	 */
218 	list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
219 		struct sem *curr;
220 		curr = &sma->sem_base[q->sops[0].sem_num];
221 
222 		list_add_tail(&q->list, &curr->pending_alter);
223 	}
224 	INIT_LIST_HEAD(&sma->pending_alter);
225 }
226 
227 /**
228  * merge_queues - merge single semop queues into global queue
229  * @sma: semaphore array
230  *
231  * This function merges all per-semaphore queues into the global queue.
232  * It is necessary to achieve FIFO ordering for the pending single-sop
233  * operations when a multi-semop operation must sleep.
234  * Only the alter operations must be moved, the const operations can stay.
235  */
236 static void merge_queues(struct sem_array *sma)
237 {
238 	int i;
239 	for (i = 0; i < sma->sem_nsems; i++) {
240 		struct sem *sem = sma->sem_base + i;
241 
242 		list_splice_init(&sem->pending_alter, &sma->pending_alter);
243 	}
244 }
245 
246 static void sem_rcu_free(struct rcu_head *head)
247 {
248 	struct ipc_rcu *p = container_of(head, struct ipc_rcu, rcu);
249 	struct sem_array *sma = ipc_rcu_to_struct(p);
250 
251 	security_sem_free(sma);
252 	ipc_rcu_free(head);
253 }
254 
255 /*
256  * Wait until all currently ongoing simple ops have completed.
257  * Caller must own sem_perm.lock.
258  * New simple ops cannot start, because simple ops first check
259  * that sem_perm.lock is free.
260  * that a) sem_perm.lock is free and b) complex_count is 0.
261  */
262 static void sem_wait_array(struct sem_array *sma)
263 {
264 	int i;
265 	struct sem *sem;
266 
267 	if (sma->complex_count)  {
268 		/* The thread that increased sma->complex_count waited on
269 		 * all sem->lock locks. Thus we don't need to wait again.
270 		 */
271 		return;
272 	}
273 
274 	for (i = 0; i < sma->sem_nsems; i++) {
275 		sem = sma->sem_base + i;
276 		spin_unlock_wait(&sem->lock);
277 	}
278 }
279 
280 /*
281  * If the request contains only one semaphore operation, and there are
282  * no complex transactions pending, lock only the semaphore involved.
283  * Otherwise, lock the entire semaphore array, since we either have
284  * multiple semaphores in our own semops, or we need to look at
285  * semaphores from other pending complex operations.
286  */
287 static inline int sem_lock(struct sem_array *sma, struct sembuf *sops,
288 			      int nsops)
289 {
290 	struct sem *sem;
291 
292 	if (nsops != 1) {
293 		/* Complex operation - acquire a full lock */
294 		ipc_lock_object(&sma->sem_perm);
295 
296 		/* And wait until all simple ops that are processed
297 		 * right now have dropped their locks.
298 		 */
299 		sem_wait_array(sma);
300 		return -1;
301 	}
302 
303 	/*
304 	 * Only one semaphore affected - try to optimize locking.
305 	 * The rules are:
306 	 * - optimized locking is possible if no complex operation
307 	 *   is either enqueued or processed right now.
308 	 * - The test for enqueued complex ops is simple:
309 	 *      sma->complex_count != 0
310 	 * - Testing for complex ops that are processed right now is
311 	 *   a bit more difficult. Complex ops acquire the full lock
312 	 *   and first wait that the running simple ops have completed.
313 	 *   (see above)
314 	 *   Thus: If we own a simple lock and the global lock is free
315 	 *	and complex_count is now 0, then it will stay 0 and
316 	 *	thus just locking sem->lock is sufficient.
317 	 */
318 	sem = sma->sem_base + sops->sem_num;
319 
320 	if (sma->complex_count == 0) {
321 		/*
322 		 * It appears that no complex operation is around.
323 		 * Acquire the per-semaphore lock.
324 		 */
325 		spin_lock(&sem->lock);
326 
327 		/* Then check that the global lock is free */
328 		if (!spin_is_locked(&sma->sem_perm.lock)) {
329 			/* spin_is_locked() is not a memory barrier */
330 			smp_mb();
331 
332 			/* Now repeat the test of complex_count:
333 			 * It can't change anymore until we drop sem->lock.
334 			 * Thus: if is now 0, then it will stay 0.
335 			 */
336 			if (sma->complex_count == 0) {
337 				/* fast path successful! */
338 				return sops->sem_num;
339 			}
340 		}
341 		spin_unlock(&sem->lock);
342 	}
343 
344 	/* slow path: acquire the full lock */
345 	ipc_lock_object(&sma->sem_perm);
346 
347 	if (sma->complex_count == 0) {
348 		/* False alarm:
349 		 * There is no complex operation, thus we can switch
350 		 * back to the fast path.
351 		 */
352 		spin_lock(&sem->lock);
353 		ipc_unlock_object(&sma->sem_perm);
354 		return sops->sem_num;
355 	} else {
356 		/* Not a false alarm, thus complete the sequence for a
357 		 * full lock.
358 		 */
359 		sem_wait_array(sma);
360 		return -1;
361 	}
362 }
363 
364 static inline void sem_unlock(struct sem_array *sma, int locknum)
365 {
366 	if (locknum == -1) {
367 		unmerge_queues(sma);
368 		ipc_unlock_object(&sma->sem_perm);
369 	} else {
370 		struct sem *sem = sma->sem_base + locknum;
371 		spin_unlock(&sem->lock);
372 	}
373 }
374 
375 /*
376  * sem_lock_(check_) routines are called in the paths where the rwsem
377  * is not held.
378  *
379  * The caller holds the RCU read lock.
380  */
381 static inline struct sem_array *sem_obtain_lock(struct ipc_namespace *ns,
382 			int id, struct sembuf *sops, int nsops, int *locknum)
383 {
384 	struct kern_ipc_perm *ipcp;
385 	struct sem_array *sma;
386 
387 	ipcp = ipc_obtain_object(&sem_ids(ns), id);
388 	if (IS_ERR(ipcp))
389 		return ERR_CAST(ipcp);
390 
391 	sma = container_of(ipcp, struct sem_array, sem_perm);
392 	*locknum = sem_lock(sma, sops, nsops);
393 
394 	/* ipc_rmid() may have already freed the ID while sem_lock
395 	 * was spinning: verify that the structure is still valid
396 	 */
397 	if (ipc_valid_object(ipcp))
398 		return container_of(ipcp, struct sem_array, sem_perm);
399 
400 	sem_unlock(sma, *locknum);
401 	return ERR_PTR(-EINVAL);
402 }
403 
404 static inline struct sem_array *sem_obtain_object(struct ipc_namespace *ns, int id)
405 {
406 	struct kern_ipc_perm *ipcp = ipc_obtain_object(&sem_ids(ns), id);
407 
408 	if (IS_ERR(ipcp))
409 		return ERR_CAST(ipcp);
410 
411 	return container_of(ipcp, struct sem_array, sem_perm);
412 }
413 
414 static inline struct sem_array *sem_obtain_object_check(struct ipc_namespace *ns,
415 							int id)
416 {
417 	struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&sem_ids(ns), id);
418 
419 	if (IS_ERR(ipcp))
420 		return ERR_CAST(ipcp);
421 
422 	return container_of(ipcp, struct sem_array, sem_perm);
423 }
424 
425 static inline void sem_lock_and_putref(struct sem_array *sma)
426 {
427 	sem_lock(sma, NULL, -1);
428 	ipc_rcu_putref(sma, ipc_rcu_free);
429 }
430 
431 static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
432 {
433 	ipc_rmid(&sem_ids(ns), &s->sem_perm);
434 }
435 
436 /*
437  * Lockless wakeup algorithm:
438  * Without the check/retry algorithm a lockless wakeup is possible:
439  * - queue.status is initialized to -EINTR before blocking.
440  * - wakeup is performed by
441  *	* unlinking the queue entry from the pending list
442  *	* setting queue.status to IN_WAKEUP
443  *	  This is the notification for the blocked thread that a
444  *	  result value is imminent.
445  *	* call wake_up_process
446  *	* set queue.status to the final value.
447  * - the previously blocked thread checks queue.status:
448  *	* if it's IN_WAKEUP, then it must wait until the value changes
449  *	* if it's not -EINTR, then the operation was completed by
450  *	  update_queue. semtimedop can return queue.status without
451  *	  performing any operation on the sem array.
452  *	* otherwise it must acquire the spinlock and check what's up.
453  *
454  * The two-stage algorithm is necessary to protect against the following
455  * races:
456  * - if queue.status is set after wake_up_process, then the woken up idle
457  *   thread could race forward and try (and fail) to acquire sma->lock
458  *   before update_queue had a chance to set queue.status
459  * - if queue.status is written before wake_up_process and if the
460  *   blocked process is woken up by a signal between writing
461  *   queue.status and the wake_up_process, then the woken up
462  *   process could return from semtimedop and die by calling
463  *   sys_exit before wake_up_process is called. Then wake_up_process
464  *   will oops, because the task structure is already invalid.
465  *   (yes, this happened on s390 with sysv msg).
466  *
467  */
468 #define IN_WAKEUP	1
469 
470 /**
471  * newary - Create a new semaphore set
472  * @ns: namespace
473  * @params: ptr to the structure that contains key, semflg and nsems
474  *
475  * Called with sem_ids.rwsem held (as a writer)
476  */
477 static int newary(struct ipc_namespace *ns, struct ipc_params *params)
478 {
479 	int id;
480 	int retval;
481 	struct sem_array *sma;
482 	int size;
483 	key_t key = params->key;
484 	int nsems = params->u.nsems;
485 	int semflg = params->flg;
486 	int i;
487 
488 	if (!nsems)
489 		return -EINVAL;
490 	if (ns->used_sems + nsems > ns->sc_semmns)
491 		return -ENOSPC;
492 
493 	size = sizeof(*sma) + nsems * sizeof(struct sem);
494 	sma = ipc_rcu_alloc(size);
495 	if (!sma)
496 		return -ENOMEM;
497 
498 	memset(sma, 0, size);
499 
500 	sma->sem_perm.mode = (semflg & S_IRWXUGO);
501 	sma->sem_perm.key = key;
502 
503 	sma->sem_perm.security = NULL;
504 	retval = security_sem_alloc(sma);
505 	if (retval) {
506 		ipc_rcu_putref(sma, ipc_rcu_free);
507 		return retval;
508 	}
509 
510 	id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
511 	if (id < 0) {
512 		ipc_rcu_putref(sma, sem_rcu_free);
513 		return id;
514 	}
515 	ns->used_sems += nsems;
516 
517 	sma->sem_base = (struct sem *) &sma[1];
518 
519 	for (i = 0; i < nsems; i++) {
520 		INIT_LIST_HEAD(&sma->sem_base[i].pending_alter);
521 		INIT_LIST_HEAD(&sma->sem_base[i].pending_const);
522 		spin_lock_init(&sma->sem_base[i].lock);
523 	}
524 
525 	sma->complex_count = 0;
526 	INIT_LIST_HEAD(&sma->pending_alter);
527 	INIT_LIST_HEAD(&sma->pending_const);
528 	INIT_LIST_HEAD(&sma->list_id);
529 	sma->sem_nsems = nsems;
530 	sma->sem_ctime = get_seconds();
531 	sem_unlock(sma, -1);
532 	rcu_read_unlock();
533 
534 	return sma->sem_perm.id;
535 }
536 
537 
538 /*
539  * Called with sem_ids.rwsem and ipcp locked.
540  */
541 static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg)
542 {
543 	struct sem_array *sma;
544 
545 	sma = container_of(ipcp, struct sem_array, sem_perm);
546 	return security_sem_associate(sma, semflg);
547 }
548 
549 /*
550  * Called with sem_ids.rwsem and ipcp locked.
551  */
552 static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
553 				struct ipc_params *params)
554 {
555 	struct sem_array *sma;
556 
557 	sma = container_of(ipcp, struct sem_array, sem_perm);
558 	if (params->u.nsems > sma->sem_nsems)
559 		return -EINVAL;
560 
561 	return 0;
562 }
563 
564 SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg)
565 {
566 	struct ipc_namespace *ns;
567 	static const struct ipc_ops sem_ops = {
568 		.getnew = newary,
569 		.associate = sem_security,
570 		.more_checks = sem_more_checks,
571 	};
572 	struct ipc_params sem_params;
573 
574 	ns = current->nsproxy->ipc_ns;
575 
576 	if (nsems < 0 || nsems > ns->sc_semmsl)
577 		return -EINVAL;
578 
579 	sem_params.key = key;
580 	sem_params.flg = semflg;
581 	sem_params.u.nsems = nsems;
582 
583 	return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
584 }
585 
586 /**
587  * perform_atomic_semop - Perform (if possible) a semaphore operation
588  * @sma: semaphore array
589  * @q: struct sem_queue that describes the operation
590  *
591  * Returns 0 if the operation was possible.
592  * Returns 1 if the operation is impossible, the caller must sleep.
593  * Negative values are error codes.
594  */
595 static int perform_atomic_semop(struct sem_array *sma, struct sem_queue *q)
596 {
597 	int result, sem_op, nsops, pid;
598 	struct sembuf *sop;
599 	struct sem *curr;
600 	struct sembuf *sops;
601 	struct sem_undo *un;
602 
603 	sops = q->sops;
604 	nsops = q->nsops;
605 	un = q->undo;
606 
607 	for (sop = sops; sop < sops + nsops; sop++) {
608 		curr = sma->sem_base + sop->sem_num;
609 		sem_op = sop->sem_op;
610 		result = curr->semval;
611 
612 		if (!sem_op && result)
613 			goto would_block;
614 
615 		result += sem_op;
616 		if (result < 0)
617 			goto would_block;
618 		if (result > SEMVMX)
619 			goto out_of_range;
620 
621 		if (sop->sem_flg & SEM_UNDO) {
622 			int undo = un->semadj[sop->sem_num] - sem_op;
623 			/* Exceeding the undo range is an error. */
624 			if (undo < (-SEMAEM - 1) || undo > SEMAEM)
625 				goto out_of_range;
626 			un->semadj[sop->sem_num] = undo;
627 		}
628 
629 		curr->semval = result;
630 	}
631 
632 	sop--;
633 	pid = q->pid;
634 	while (sop >= sops) {
635 		sma->sem_base[sop->sem_num].sempid = pid;
636 		sop--;
637 	}
638 
639 	return 0;
640 
641 out_of_range:
642 	result = -ERANGE;
643 	goto undo;
644 
645 would_block:
646 	q->blocking = sop;
647 
648 	if (sop->sem_flg & IPC_NOWAIT)
649 		result = -EAGAIN;
650 	else
651 		result = 1;
652 
653 undo:
654 	sop--;
655 	while (sop >= sops) {
656 		sem_op = sop->sem_op;
657 		sma->sem_base[sop->sem_num].semval -= sem_op;
658 		if (sop->sem_flg & SEM_UNDO)
659 			un->semadj[sop->sem_num] += sem_op;
660 		sop--;
661 	}
662 
663 	return result;
664 }
665 
666 /** wake_up_sem_queue_prepare(q, error): Prepare wake-up
667  * @q: queue entry that must be signaled
668  * @error: Error value for the signal
669  *
670  * Prepare the wake-up of the queue entry q.
671  */
672 static void wake_up_sem_queue_prepare(struct list_head *pt,
673 				struct sem_queue *q, int error)
674 {
675 	if (list_empty(pt)) {
676 		/*
677 		 * Hold preempt off so that we don't get preempted and have the
678 		 * wakee busy-wait until we're scheduled back on.
679 		 */
680 		preempt_disable();
681 	}
682 	q->status = IN_WAKEUP;
683 	q->pid = error;
684 
685 	list_add_tail(&q->list, pt);
686 }
687 
688 /**
689  * wake_up_sem_queue_do - do the actual wake-up
690  * @pt: list of tasks to be woken up
691  *
692  * Do the actual wake-up.
693  * The function is called without any locks held, thus the semaphore array
694  * could be destroyed already and the tasks can disappear as soon as the
695  * status is set to the actual return code.
696  */
697 static void wake_up_sem_queue_do(struct list_head *pt)
698 {
699 	struct sem_queue *q, *t;
700 	int did_something;
701 
702 	did_something = !list_empty(pt);
703 	list_for_each_entry_safe(q, t, pt, list) {
704 		wake_up_process(q->sleeper);
705 		/* q can disappear immediately after writing q->status. */
706 		smp_wmb();
707 		q->status = q->pid;
708 	}
709 	if (did_something)
710 		preempt_enable();
711 }
712 
713 static void unlink_queue(struct sem_array *sma, struct sem_queue *q)
714 {
715 	list_del(&q->list);
716 	if (q->nsops > 1)
717 		sma->complex_count--;
718 }
719 
720 /** check_restart(sma, q)
721  * @sma: semaphore array
722  * @q: the operation that just completed
723  *
724  * update_queue is O(N^2) when it restarts scanning the whole queue of
725  * waiting operations. Therefore this function checks if the restart is
726  * really necessary. It is called after a previously waiting operation
727  * modified the array.
728  * Note that wait-for-zero operations are handled without restart.
729  */
730 static int check_restart(struct sem_array *sma, struct sem_queue *q)
731 {
732 	/* pending complex alter operations are too difficult to analyse */
733 	if (!list_empty(&sma->pending_alter))
734 		return 1;
735 
736 	/* we were a sleeping complex operation. Too difficult */
737 	if (q->nsops > 1)
738 		return 1;
739 
740 	/* It is impossible that someone waits for the new value:
741 	 * - complex operations always restart.
742 	 * - wait-for-zero are handled seperately.
743 	 * - q is a previously sleeping simple operation that
744 	 *   altered the array. It must be a decrement, because
745 	 *   simple increments never sleep.
746 	 * - If there are older (higher priority) decrements
747 	 *   in the queue, then they have observed the original
748 	 *   semval value and couldn't proceed. The operation
749 	 *   decremented to value - thus they won't proceed either.
750 	 */
751 	return 0;
752 }
753 
754 /**
755  * wake_const_ops - wake up non-alter tasks
756  * @sma: semaphore array.
757  * @semnum: semaphore that was modified.
758  * @pt: list head for the tasks that must be woken up.
759  *
760  * wake_const_ops must be called after a semaphore in a semaphore array
761  * was set to 0. If complex const operations are pending, wake_const_ops must
762  * be called with semnum = -1, as well as with the number of each modified
763  * semaphore.
764  * The tasks that must be woken up are added to @pt. The return code
765  * is stored in q->pid.
766  * The function returns 1 if at least one operation was completed successfully.
767  */
768 static int wake_const_ops(struct sem_array *sma, int semnum,
769 				struct list_head *pt)
770 {
771 	struct sem_queue *q;
772 	struct list_head *walk;
773 	struct list_head *pending_list;
774 	int semop_completed = 0;
775 
776 	if (semnum == -1)
777 		pending_list = &sma->pending_const;
778 	else
779 		pending_list = &sma->sem_base[semnum].pending_const;
780 
781 	walk = pending_list->next;
782 	while (walk != pending_list) {
783 		int error;
784 
785 		q = container_of(walk, struct sem_queue, list);
786 		walk = walk->next;
787 
788 		error = perform_atomic_semop(sma, q);
789 
790 		if (error <= 0) {
791 			/* operation completed, remove from queue & wakeup */
792 
793 			unlink_queue(sma, q);
794 
795 			wake_up_sem_queue_prepare(pt, q, error);
796 			if (error == 0)
797 				semop_completed = 1;
798 		}
799 	}
800 	return semop_completed;
801 }
802 
803 /**
804  * do_smart_wakeup_zero - wakeup all wait for zero tasks
805  * @sma: semaphore array
806  * @sops: operations that were performed
807  * @nsops: number of operations
808  * @pt: list head of the tasks that must be woken up.
809  *
810  * Checks all required queue for wait-for-zero operations, based
811  * on the actual changes that were performed on the semaphore array.
812  * The function returns 1 if at least one operation was completed successfully.
813  */
814 static int do_smart_wakeup_zero(struct sem_array *sma, struct sembuf *sops,
815 					int nsops, struct list_head *pt)
816 {
817 	int i;
818 	int semop_completed = 0;
819 	int got_zero = 0;
820 
821 	/* first: the per-semaphore queues, if known */
822 	if (sops) {
823 		for (i = 0; i < nsops; i++) {
824 			int num = sops[i].sem_num;
825 
826 			if (sma->sem_base[num].semval == 0) {
827 				got_zero = 1;
828 				semop_completed |= wake_const_ops(sma, num, pt);
829 			}
830 		}
831 	} else {
832 		/*
833 		 * No sops means modified semaphores not known.
834 		 * Assume all were changed.
835 		 */
836 		for (i = 0; i < sma->sem_nsems; i++) {
837 			if (sma->sem_base[i].semval == 0) {
838 				got_zero = 1;
839 				semop_completed |= wake_const_ops(sma, i, pt);
840 			}
841 		}
842 	}
843 	/*
844 	 * If one of the modified semaphores got 0,
845 	 * then check the global queue, too.
846 	 */
847 	if (got_zero)
848 		semop_completed |= wake_const_ops(sma, -1, pt);
849 
850 	return semop_completed;
851 }
852 
853 
854 /**
855  * update_queue - look for tasks that can be completed.
856  * @sma: semaphore array.
857  * @semnum: semaphore that was modified.
858  * @pt: list head for the tasks that must be woken up.
859  *
860  * update_queue must be called after a semaphore in a semaphore array
861  * was modified. If multiple semaphores were modified, update_queue must
862  * be called with semnum = -1, as well as with the number of each modified
863  * semaphore.
864  * The tasks that must be woken up are added to @pt. The return code
865  * is stored in q->pid.
866  * The function internally checks if const operations can now succeed.
867  *
868  * The function return 1 if at least one semop was completed successfully.
869  */
870 static int update_queue(struct sem_array *sma, int semnum, struct list_head *pt)
871 {
872 	struct sem_queue *q;
873 	struct list_head *walk;
874 	struct list_head *pending_list;
875 	int semop_completed = 0;
876 
877 	if (semnum == -1)
878 		pending_list = &sma->pending_alter;
879 	else
880 		pending_list = &sma->sem_base[semnum].pending_alter;
881 
882 again:
883 	walk = pending_list->next;
884 	while (walk != pending_list) {
885 		int error, restart;
886 
887 		q = container_of(walk, struct sem_queue, list);
888 		walk = walk->next;
889 
890 		/* If we are scanning the single sop, per-semaphore list of
891 		 * one semaphore and that semaphore is 0, then it is not
892 		 * necessary to scan further: simple increments
893 		 * that affect only one entry succeed immediately and cannot
894 		 * be in the  per semaphore pending queue, and decrements
895 		 * cannot be successful if the value is already 0.
896 		 */
897 		if (semnum != -1 && sma->sem_base[semnum].semval == 0)
898 			break;
899 
900 		error = perform_atomic_semop(sma, q);
901 
902 		/* Does q->sleeper still need to sleep? */
903 		if (error > 0)
904 			continue;
905 
906 		unlink_queue(sma, q);
907 
908 		if (error) {
909 			restart = 0;
910 		} else {
911 			semop_completed = 1;
912 			do_smart_wakeup_zero(sma, q->sops, q->nsops, pt);
913 			restart = check_restart(sma, q);
914 		}
915 
916 		wake_up_sem_queue_prepare(pt, q, error);
917 		if (restart)
918 			goto again;
919 	}
920 	return semop_completed;
921 }
922 
923 /**
924  * set_semotime - set sem_otime
925  * @sma: semaphore array
926  * @sops: operations that modified the array, may be NULL
927  *
928  * sem_otime is replicated to avoid cache line trashing.
929  * This function sets one instance to the current time.
930  */
931 static void set_semotime(struct sem_array *sma, struct sembuf *sops)
932 {
933 	if (sops == NULL) {
934 		sma->sem_base[0].sem_otime = get_seconds();
935 	} else {
936 		sma->sem_base[sops[0].sem_num].sem_otime =
937 							get_seconds();
938 	}
939 }
940 
941 /**
942  * do_smart_update - optimized update_queue
943  * @sma: semaphore array
944  * @sops: operations that were performed
945  * @nsops: number of operations
946  * @otime: force setting otime
947  * @pt: list head of the tasks that must be woken up.
948  *
949  * do_smart_update() does the required calls to update_queue and wakeup_zero,
950  * based on the actual changes that were performed on the semaphore array.
951  * Note that the function does not do the actual wake-up: the caller is
952  * responsible for calling wake_up_sem_queue_do(@pt).
953  * It is safe to perform this call after dropping all locks.
954  */
955 static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops,
956 			int otime, struct list_head *pt)
957 {
958 	int i;
959 
960 	otime |= do_smart_wakeup_zero(sma, sops, nsops, pt);
961 
962 	if (!list_empty(&sma->pending_alter)) {
963 		/* semaphore array uses the global queue - just process it. */
964 		otime |= update_queue(sma, -1, pt);
965 	} else {
966 		if (!sops) {
967 			/*
968 			 * No sops, thus the modified semaphores are not
969 			 * known. Check all.
970 			 */
971 			for (i = 0; i < sma->sem_nsems; i++)
972 				otime |= update_queue(sma, i, pt);
973 		} else {
974 			/*
975 			 * Check the semaphores that were increased:
976 			 * - No complex ops, thus all sleeping ops are
977 			 *   decrease.
978 			 * - if we decreased the value, then any sleeping
979 			 *   semaphore ops wont be able to run: If the
980 			 *   previous value was too small, then the new
981 			 *   value will be too small, too.
982 			 */
983 			for (i = 0; i < nsops; i++) {
984 				if (sops[i].sem_op > 0) {
985 					otime |= update_queue(sma,
986 							sops[i].sem_num, pt);
987 				}
988 			}
989 		}
990 	}
991 	if (otime)
992 		set_semotime(sma, sops);
993 }
994 
995 /*
996  * check_qop: Test if a queued operation sleeps on the semaphore semnum
997  */
998 static int check_qop(struct sem_array *sma, int semnum, struct sem_queue *q,
999 			bool count_zero)
1000 {
1001 	struct sembuf *sop = q->blocking;
1002 
1003 	/*
1004 	 * Linux always (since 0.99.10) reported a task as sleeping on all
1005 	 * semaphores. This violates SUS, therefore it was changed to the
1006 	 * standard compliant behavior.
1007 	 * Give the administrators a chance to notice that an application
1008 	 * might misbehave because it relies on the Linux behavior.
1009 	 */
1010 	pr_info_once("semctl(GETNCNT/GETZCNT) is since 3.16 Single Unix Specification compliant.\n"
1011 			"The task %s (%d) triggered the difference, watch for misbehavior.\n",
1012 			current->comm, task_pid_nr(current));
1013 
1014 	if (sop->sem_num != semnum)
1015 		return 0;
1016 
1017 	if (count_zero && sop->sem_op == 0)
1018 		return 1;
1019 	if (!count_zero && sop->sem_op < 0)
1020 		return 1;
1021 
1022 	return 0;
1023 }
1024 
1025 /* The following counts are associated to each semaphore:
1026  *   semncnt        number of tasks waiting on semval being nonzero
1027  *   semzcnt        number of tasks waiting on semval being zero
1028  *
1029  * Per definition, a task waits only on the semaphore of the first semop
1030  * that cannot proceed, even if additional operation would block, too.
1031  */
1032 static int count_semcnt(struct sem_array *sma, ushort semnum,
1033 			bool count_zero)
1034 {
1035 	struct list_head *l;
1036 	struct sem_queue *q;
1037 	int semcnt;
1038 
1039 	semcnt = 0;
1040 	/* First: check the simple operations. They are easy to evaluate */
1041 	if (count_zero)
1042 		l = &sma->sem_base[semnum].pending_const;
1043 	else
1044 		l = &sma->sem_base[semnum].pending_alter;
1045 
1046 	list_for_each_entry(q, l, list) {
1047 		/* all task on a per-semaphore list sleep on exactly
1048 		 * that semaphore
1049 		 */
1050 		semcnt++;
1051 	}
1052 
1053 	/* Then: check the complex operations. */
1054 	list_for_each_entry(q, &sma->pending_alter, list) {
1055 		semcnt += check_qop(sma, semnum, q, count_zero);
1056 	}
1057 	if (count_zero) {
1058 		list_for_each_entry(q, &sma->pending_const, list) {
1059 			semcnt += check_qop(sma, semnum, q, count_zero);
1060 		}
1061 	}
1062 	return semcnt;
1063 }
1064 
1065 /* Free a semaphore set. freeary() is called with sem_ids.rwsem locked
1066  * as a writer and the spinlock for this semaphore set hold. sem_ids.rwsem
1067  * remains locked on exit.
1068  */
1069 static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
1070 {
1071 	struct sem_undo *un, *tu;
1072 	struct sem_queue *q, *tq;
1073 	struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
1074 	struct list_head tasks;
1075 	int i;
1076 
1077 	/* Free the existing undo structures for this semaphore set.  */
1078 	ipc_assert_locked_object(&sma->sem_perm);
1079 	list_for_each_entry_safe(un, tu, &sma->list_id, list_id) {
1080 		list_del(&un->list_id);
1081 		spin_lock(&un->ulp->lock);
1082 		un->semid = -1;
1083 		list_del_rcu(&un->list_proc);
1084 		spin_unlock(&un->ulp->lock);
1085 		kfree_rcu(un, rcu);
1086 	}
1087 
1088 	/* Wake up all pending processes and let them fail with EIDRM. */
1089 	INIT_LIST_HEAD(&tasks);
1090 	list_for_each_entry_safe(q, tq, &sma->pending_const, list) {
1091 		unlink_queue(sma, q);
1092 		wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1093 	}
1094 
1095 	list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
1096 		unlink_queue(sma, q);
1097 		wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1098 	}
1099 	for (i = 0; i < sma->sem_nsems; i++) {
1100 		struct sem *sem = sma->sem_base + i;
1101 		list_for_each_entry_safe(q, tq, &sem->pending_const, list) {
1102 			unlink_queue(sma, q);
1103 			wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1104 		}
1105 		list_for_each_entry_safe(q, tq, &sem->pending_alter, list) {
1106 			unlink_queue(sma, q);
1107 			wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1108 		}
1109 	}
1110 
1111 	/* Remove the semaphore set from the IDR */
1112 	sem_rmid(ns, sma);
1113 	sem_unlock(sma, -1);
1114 	rcu_read_unlock();
1115 
1116 	wake_up_sem_queue_do(&tasks);
1117 	ns->used_sems -= sma->sem_nsems;
1118 	ipc_rcu_putref(sma, sem_rcu_free);
1119 }
1120 
1121 static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
1122 {
1123 	switch (version) {
1124 	case IPC_64:
1125 		return copy_to_user(buf, in, sizeof(*in));
1126 	case IPC_OLD:
1127 	    {
1128 		struct semid_ds out;
1129 
1130 		memset(&out, 0, sizeof(out));
1131 
1132 		ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
1133 
1134 		out.sem_otime	= in->sem_otime;
1135 		out.sem_ctime	= in->sem_ctime;
1136 		out.sem_nsems	= in->sem_nsems;
1137 
1138 		return copy_to_user(buf, &out, sizeof(out));
1139 	    }
1140 	default:
1141 		return -EINVAL;
1142 	}
1143 }
1144 
1145 static time_t get_semotime(struct sem_array *sma)
1146 {
1147 	int i;
1148 	time_t res;
1149 
1150 	res = sma->sem_base[0].sem_otime;
1151 	for (i = 1; i < sma->sem_nsems; i++) {
1152 		time_t to = sma->sem_base[i].sem_otime;
1153 
1154 		if (to > res)
1155 			res = to;
1156 	}
1157 	return res;
1158 }
1159 
1160 static int semctl_nolock(struct ipc_namespace *ns, int semid,
1161 			 int cmd, int version, void __user *p)
1162 {
1163 	int err;
1164 	struct sem_array *sma;
1165 
1166 	switch (cmd) {
1167 	case IPC_INFO:
1168 	case SEM_INFO:
1169 	{
1170 		struct seminfo seminfo;
1171 		int max_id;
1172 
1173 		err = security_sem_semctl(NULL, cmd);
1174 		if (err)
1175 			return err;
1176 
1177 		memset(&seminfo, 0, sizeof(seminfo));
1178 		seminfo.semmni = ns->sc_semmni;
1179 		seminfo.semmns = ns->sc_semmns;
1180 		seminfo.semmsl = ns->sc_semmsl;
1181 		seminfo.semopm = ns->sc_semopm;
1182 		seminfo.semvmx = SEMVMX;
1183 		seminfo.semmnu = SEMMNU;
1184 		seminfo.semmap = SEMMAP;
1185 		seminfo.semume = SEMUME;
1186 		down_read(&sem_ids(ns).rwsem);
1187 		if (cmd == SEM_INFO) {
1188 			seminfo.semusz = sem_ids(ns).in_use;
1189 			seminfo.semaem = ns->used_sems;
1190 		} else {
1191 			seminfo.semusz = SEMUSZ;
1192 			seminfo.semaem = SEMAEM;
1193 		}
1194 		max_id = ipc_get_maxid(&sem_ids(ns));
1195 		up_read(&sem_ids(ns).rwsem);
1196 		if (copy_to_user(p, &seminfo, sizeof(struct seminfo)))
1197 			return -EFAULT;
1198 		return (max_id < 0) ? 0 : max_id;
1199 	}
1200 	case IPC_STAT:
1201 	case SEM_STAT:
1202 	{
1203 		struct semid64_ds tbuf;
1204 		int id = 0;
1205 
1206 		memset(&tbuf, 0, sizeof(tbuf));
1207 
1208 		rcu_read_lock();
1209 		if (cmd == SEM_STAT) {
1210 			sma = sem_obtain_object(ns, semid);
1211 			if (IS_ERR(sma)) {
1212 				err = PTR_ERR(sma);
1213 				goto out_unlock;
1214 			}
1215 			id = sma->sem_perm.id;
1216 		} else {
1217 			sma = sem_obtain_object_check(ns, semid);
1218 			if (IS_ERR(sma)) {
1219 				err = PTR_ERR(sma);
1220 				goto out_unlock;
1221 			}
1222 		}
1223 
1224 		err = -EACCES;
1225 		if (ipcperms(ns, &sma->sem_perm, S_IRUGO))
1226 			goto out_unlock;
1227 
1228 		err = security_sem_semctl(sma, cmd);
1229 		if (err)
1230 			goto out_unlock;
1231 
1232 		kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
1233 		tbuf.sem_otime = get_semotime(sma);
1234 		tbuf.sem_ctime = sma->sem_ctime;
1235 		tbuf.sem_nsems = sma->sem_nsems;
1236 		rcu_read_unlock();
1237 		if (copy_semid_to_user(p, &tbuf, version))
1238 			return -EFAULT;
1239 		return id;
1240 	}
1241 	default:
1242 		return -EINVAL;
1243 	}
1244 out_unlock:
1245 	rcu_read_unlock();
1246 	return err;
1247 }
1248 
1249 static int semctl_setval(struct ipc_namespace *ns, int semid, int semnum,
1250 		unsigned long arg)
1251 {
1252 	struct sem_undo *un;
1253 	struct sem_array *sma;
1254 	struct sem *curr;
1255 	int err;
1256 	struct list_head tasks;
1257 	int val;
1258 #if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN)
1259 	/* big-endian 64bit */
1260 	val = arg >> 32;
1261 #else
1262 	/* 32bit or little-endian 64bit */
1263 	val = arg;
1264 #endif
1265 
1266 	if (val > SEMVMX || val < 0)
1267 		return -ERANGE;
1268 
1269 	INIT_LIST_HEAD(&tasks);
1270 
1271 	rcu_read_lock();
1272 	sma = sem_obtain_object_check(ns, semid);
1273 	if (IS_ERR(sma)) {
1274 		rcu_read_unlock();
1275 		return PTR_ERR(sma);
1276 	}
1277 
1278 	if (semnum < 0 || semnum >= sma->sem_nsems) {
1279 		rcu_read_unlock();
1280 		return -EINVAL;
1281 	}
1282 
1283 
1284 	if (ipcperms(ns, &sma->sem_perm, S_IWUGO)) {
1285 		rcu_read_unlock();
1286 		return -EACCES;
1287 	}
1288 
1289 	err = security_sem_semctl(sma, SETVAL);
1290 	if (err) {
1291 		rcu_read_unlock();
1292 		return -EACCES;
1293 	}
1294 
1295 	sem_lock(sma, NULL, -1);
1296 
1297 	if (!ipc_valid_object(&sma->sem_perm)) {
1298 		sem_unlock(sma, -1);
1299 		rcu_read_unlock();
1300 		return -EIDRM;
1301 	}
1302 
1303 	curr = &sma->sem_base[semnum];
1304 
1305 	ipc_assert_locked_object(&sma->sem_perm);
1306 	list_for_each_entry(un, &sma->list_id, list_id)
1307 		un->semadj[semnum] = 0;
1308 
1309 	curr->semval = val;
1310 	curr->sempid = task_tgid_vnr(current);
1311 	sma->sem_ctime = get_seconds();
1312 	/* maybe some queued-up processes were waiting for this */
1313 	do_smart_update(sma, NULL, 0, 0, &tasks);
1314 	sem_unlock(sma, -1);
1315 	rcu_read_unlock();
1316 	wake_up_sem_queue_do(&tasks);
1317 	return 0;
1318 }
1319 
1320 static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
1321 		int cmd, void __user *p)
1322 {
1323 	struct sem_array *sma;
1324 	struct sem *curr;
1325 	int err, nsems;
1326 	ushort fast_sem_io[SEMMSL_FAST];
1327 	ushort *sem_io = fast_sem_io;
1328 	struct list_head tasks;
1329 
1330 	INIT_LIST_HEAD(&tasks);
1331 
1332 	rcu_read_lock();
1333 	sma = sem_obtain_object_check(ns, semid);
1334 	if (IS_ERR(sma)) {
1335 		rcu_read_unlock();
1336 		return PTR_ERR(sma);
1337 	}
1338 
1339 	nsems = sma->sem_nsems;
1340 
1341 	err = -EACCES;
1342 	if (ipcperms(ns, &sma->sem_perm, cmd == SETALL ? S_IWUGO : S_IRUGO))
1343 		goto out_rcu_wakeup;
1344 
1345 	err = security_sem_semctl(sma, cmd);
1346 	if (err)
1347 		goto out_rcu_wakeup;
1348 
1349 	err = -EACCES;
1350 	switch (cmd) {
1351 	case GETALL:
1352 	{
1353 		ushort __user *array = p;
1354 		int i;
1355 
1356 		sem_lock(sma, NULL, -1);
1357 		if (!ipc_valid_object(&sma->sem_perm)) {
1358 			err = -EIDRM;
1359 			goto out_unlock;
1360 		}
1361 		if (nsems > SEMMSL_FAST) {
1362 			if (!ipc_rcu_getref(sma)) {
1363 				err = -EIDRM;
1364 				goto out_unlock;
1365 			}
1366 			sem_unlock(sma, -1);
1367 			rcu_read_unlock();
1368 			sem_io = ipc_alloc(sizeof(ushort)*nsems);
1369 			if (sem_io == NULL) {
1370 				ipc_rcu_putref(sma, ipc_rcu_free);
1371 				return -ENOMEM;
1372 			}
1373 
1374 			rcu_read_lock();
1375 			sem_lock_and_putref(sma);
1376 			if (!ipc_valid_object(&sma->sem_perm)) {
1377 				err = -EIDRM;
1378 				goto out_unlock;
1379 			}
1380 		}
1381 		for (i = 0; i < sma->sem_nsems; i++)
1382 			sem_io[i] = sma->sem_base[i].semval;
1383 		sem_unlock(sma, -1);
1384 		rcu_read_unlock();
1385 		err = 0;
1386 		if (copy_to_user(array, sem_io, nsems*sizeof(ushort)))
1387 			err = -EFAULT;
1388 		goto out_free;
1389 	}
1390 	case SETALL:
1391 	{
1392 		int i;
1393 		struct sem_undo *un;
1394 
1395 		if (!ipc_rcu_getref(sma)) {
1396 			err = -EIDRM;
1397 			goto out_rcu_wakeup;
1398 		}
1399 		rcu_read_unlock();
1400 
1401 		if (nsems > SEMMSL_FAST) {
1402 			sem_io = ipc_alloc(sizeof(ushort)*nsems);
1403 			if (sem_io == NULL) {
1404 				ipc_rcu_putref(sma, ipc_rcu_free);
1405 				return -ENOMEM;
1406 			}
1407 		}
1408 
1409 		if (copy_from_user(sem_io, p, nsems*sizeof(ushort))) {
1410 			ipc_rcu_putref(sma, ipc_rcu_free);
1411 			err = -EFAULT;
1412 			goto out_free;
1413 		}
1414 
1415 		for (i = 0; i < nsems; i++) {
1416 			if (sem_io[i] > SEMVMX) {
1417 				ipc_rcu_putref(sma, ipc_rcu_free);
1418 				err = -ERANGE;
1419 				goto out_free;
1420 			}
1421 		}
1422 		rcu_read_lock();
1423 		sem_lock_and_putref(sma);
1424 		if (!ipc_valid_object(&sma->sem_perm)) {
1425 			err = -EIDRM;
1426 			goto out_unlock;
1427 		}
1428 
1429 		for (i = 0; i < nsems; i++)
1430 			sma->sem_base[i].semval = sem_io[i];
1431 
1432 		ipc_assert_locked_object(&sma->sem_perm);
1433 		list_for_each_entry(un, &sma->list_id, list_id) {
1434 			for (i = 0; i < nsems; i++)
1435 				un->semadj[i] = 0;
1436 		}
1437 		sma->sem_ctime = get_seconds();
1438 		/* maybe some queued-up processes were waiting for this */
1439 		do_smart_update(sma, NULL, 0, 0, &tasks);
1440 		err = 0;
1441 		goto out_unlock;
1442 	}
1443 	/* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */
1444 	}
1445 	err = -EINVAL;
1446 	if (semnum < 0 || semnum >= nsems)
1447 		goto out_rcu_wakeup;
1448 
1449 	sem_lock(sma, NULL, -1);
1450 	if (!ipc_valid_object(&sma->sem_perm)) {
1451 		err = -EIDRM;
1452 		goto out_unlock;
1453 	}
1454 	curr = &sma->sem_base[semnum];
1455 
1456 	switch (cmd) {
1457 	case GETVAL:
1458 		err = curr->semval;
1459 		goto out_unlock;
1460 	case GETPID:
1461 		err = curr->sempid;
1462 		goto out_unlock;
1463 	case GETNCNT:
1464 		err = count_semcnt(sma, semnum, 0);
1465 		goto out_unlock;
1466 	case GETZCNT:
1467 		err = count_semcnt(sma, semnum, 1);
1468 		goto out_unlock;
1469 	}
1470 
1471 out_unlock:
1472 	sem_unlock(sma, -1);
1473 out_rcu_wakeup:
1474 	rcu_read_unlock();
1475 	wake_up_sem_queue_do(&tasks);
1476 out_free:
1477 	if (sem_io != fast_sem_io)
1478 		ipc_free(sem_io, sizeof(ushort)*nsems);
1479 	return err;
1480 }
1481 
1482 static inline unsigned long
1483 copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
1484 {
1485 	switch (version) {
1486 	case IPC_64:
1487 		if (copy_from_user(out, buf, sizeof(*out)))
1488 			return -EFAULT;
1489 		return 0;
1490 	case IPC_OLD:
1491 	    {
1492 		struct semid_ds tbuf_old;
1493 
1494 		if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
1495 			return -EFAULT;
1496 
1497 		out->sem_perm.uid	= tbuf_old.sem_perm.uid;
1498 		out->sem_perm.gid	= tbuf_old.sem_perm.gid;
1499 		out->sem_perm.mode	= tbuf_old.sem_perm.mode;
1500 
1501 		return 0;
1502 	    }
1503 	default:
1504 		return -EINVAL;
1505 	}
1506 }
1507 
1508 /*
1509  * This function handles some semctl commands which require the rwsem
1510  * to be held in write mode.
1511  * NOTE: no locks must be held, the rwsem is taken inside this function.
1512  */
1513 static int semctl_down(struct ipc_namespace *ns, int semid,
1514 		       int cmd, int version, void __user *p)
1515 {
1516 	struct sem_array *sma;
1517 	int err;
1518 	struct semid64_ds semid64;
1519 	struct kern_ipc_perm *ipcp;
1520 
1521 	if (cmd == IPC_SET) {
1522 		if (copy_semid_from_user(&semid64, p, version))
1523 			return -EFAULT;
1524 	}
1525 
1526 	down_write(&sem_ids(ns).rwsem);
1527 	rcu_read_lock();
1528 
1529 	ipcp = ipcctl_pre_down_nolock(ns, &sem_ids(ns), semid, cmd,
1530 				      &semid64.sem_perm, 0);
1531 	if (IS_ERR(ipcp)) {
1532 		err = PTR_ERR(ipcp);
1533 		goto out_unlock1;
1534 	}
1535 
1536 	sma = container_of(ipcp, struct sem_array, sem_perm);
1537 
1538 	err = security_sem_semctl(sma, cmd);
1539 	if (err)
1540 		goto out_unlock1;
1541 
1542 	switch (cmd) {
1543 	case IPC_RMID:
1544 		sem_lock(sma, NULL, -1);
1545 		/* freeary unlocks the ipc object and rcu */
1546 		freeary(ns, ipcp);
1547 		goto out_up;
1548 	case IPC_SET:
1549 		sem_lock(sma, NULL, -1);
1550 		err = ipc_update_perm(&semid64.sem_perm, ipcp);
1551 		if (err)
1552 			goto out_unlock0;
1553 		sma->sem_ctime = get_seconds();
1554 		break;
1555 	default:
1556 		err = -EINVAL;
1557 		goto out_unlock1;
1558 	}
1559 
1560 out_unlock0:
1561 	sem_unlock(sma, -1);
1562 out_unlock1:
1563 	rcu_read_unlock();
1564 out_up:
1565 	up_write(&sem_ids(ns).rwsem);
1566 	return err;
1567 }
1568 
1569 SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, unsigned long, arg)
1570 {
1571 	int version;
1572 	struct ipc_namespace *ns;
1573 	void __user *p = (void __user *)arg;
1574 
1575 	if (semid < 0)
1576 		return -EINVAL;
1577 
1578 	version = ipc_parse_version(&cmd);
1579 	ns = current->nsproxy->ipc_ns;
1580 
1581 	switch (cmd) {
1582 	case IPC_INFO:
1583 	case SEM_INFO:
1584 	case IPC_STAT:
1585 	case SEM_STAT:
1586 		return semctl_nolock(ns, semid, cmd, version, p);
1587 	case GETALL:
1588 	case GETVAL:
1589 	case GETPID:
1590 	case GETNCNT:
1591 	case GETZCNT:
1592 	case SETALL:
1593 		return semctl_main(ns, semid, semnum, cmd, p);
1594 	case SETVAL:
1595 		return semctl_setval(ns, semid, semnum, arg);
1596 	case IPC_RMID:
1597 	case IPC_SET:
1598 		return semctl_down(ns, semid, cmd, version, p);
1599 	default:
1600 		return -EINVAL;
1601 	}
1602 }
1603 
1604 /* If the task doesn't already have a undo_list, then allocate one
1605  * here.  We guarantee there is only one thread using this undo list,
1606  * and current is THE ONE
1607  *
1608  * If this allocation and assignment succeeds, but later
1609  * portions of this code fail, there is no need to free the sem_undo_list.
1610  * Just let it stay associated with the task, and it'll be freed later
1611  * at exit time.
1612  *
1613  * This can block, so callers must hold no locks.
1614  */
1615 static inline int get_undo_list(struct sem_undo_list **undo_listp)
1616 {
1617 	struct sem_undo_list *undo_list;
1618 
1619 	undo_list = current->sysvsem.undo_list;
1620 	if (!undo_list) {
1621 		undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
1622 		if (undo_list == NULL)
1623 			return -ENOMEM;
1624 		spin_lock_init(&undo_list->lock);
1625 		atomic_set(&undo_list->refcnt, 1);
1626 		INIT_LIST_HEAD(&undo_list->list_proc);
1627 
1628 		current->sysvsem.undo_list = undo_list;
1629 	}
1630 	*undo_listp = undo_list;
1631 	return 0;
1632 }
1633 
1634 static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid)
1635 {
1636 	struct sem_undo *un;
1637 
1638 	list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) {
1639 		if (un->semid == semid)
1640 			return un;
1641 	}
1642 	return NULL;
1643 }
1644 
1645 static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1646 {
1647 	struct sem_undo *un;
1648 
1649 	assert_spin_locked(&ulp->lock);
1650 
1651 	un = __lookup_undo(ulp, semid);
1652 	if (un) {
1653 		list_del_rcu(&un->list_proc);
1654 		list_add_rcu(&un->list_proc, &ulp->list_proc);
1655 	}
1656 	return un;
1657 }
1658 
1659 /**
1660  * find_alloc_undo - lookup (and if not present create) undo array
1661  * @ns: namespace
1662  * @semid: semaphore array id
1663  *
1664  * The function looks up (and if not present creates) the undo structure.
1665  * The size of the undo structure depends on the size of the semaphore
1666  * array, thus the alloc path is not that straightforward.
1667  * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1668  * performs a rcu_read_lock().
1669  */
1670 static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
1671 {
1672 	struct sem_array *sma;
1673 	struct sem_undo_list *ulp;
1674 	struct sem_undo *un, *new;
1675 	int nsems, error;
1676 
1677 	error = get_undo_list(&ulp);
1678 	if (error)
1679 		return ERR_PTR(error);
1680 
1681 	rcu_read_lock();
1682 	spin_lock(&ulp->lock);
1683 	un = lookup_undo(ulp, semid);
1684 	spin_unlock(&ulp->lock);
1685 	if (likely(un != NULL))
1686 		goto out;
1687 
1688 	/* no undo structure around - allocate one. */
1689 	/* step 1: figure out the size of the semaphore array */
1690 	sma = sem_obtain_object_check(ns, semid);
1691 	if (IS_ERR(sma)) {
1692 		rcu_read_unlock();
1693 		return ERR_CAST(sma);
1694 	}
1695 
1696 	nsems = sma->sem_nsems;
1697 	if (!ipc_rcu_getref(sma)) {
1698 		rcu_read_unlock();
1699 		un = ERR_PTR(-EIDRM);
1700 		goto out;
1701 	}
1702 	rcu_read_unlock();
1703 
1704 	/* step 2: allocate new undo structure */
1705 	new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1706 	if (!new) {
1707 		ipc_rcu_putref(sma, ipc_rcu_free);
1708 		return ERR_PTR(-ENOMEM);
1709 	}
1710 
1711 	/* step 3: Acquire the lock on semaphore array */
1712 	rcu_read_lock();
1713 	sem_lock_and_putref(sma);
1714 	if (!ipc_valid_object(&sma->sem_perm)) {
1715 		sem_unlock(sma, -1);
1716 		rcu_read_unlock();
1717 		kfree(new);
1718 		un = ERR_PTR(-EIDRM);
1719 		goto out;
1720 	}
1721 	spin_lock(&ulp->lock);
1722 
1723 	/*
1724 	 * step 4: check for races: did someone else allocate the undo struct?
1725 	 */
1726 	un = lookup_undo(ulp, semid);
1727 	if (un) {
1728 		kfree(new);
1729 		goto success;
1730 	}
1731 	/* step 5: initialize & link new undo structure */
1732 	new->semadj = (short *) &new[1];
1733 	new->ulp = ulp;
1734 	new->semid = semid;
1735 	assert_spin_locked(&ulp->lock);
1736 	list_add_rcu(&new->list_proc, &ulp->list_proc);
1737 	ipc_assert_locked_object(&sma->sem_perm);
1738 	list_add(&new->list_id, &sma->list_id);
1739 	un = new;
1740 
1741 success:
1742 	spin_unlock(&ulp->lock);
1743 	sem_unlock(sma, -1);
1744 out:
1745 	return un;
1746 }
1747 
1748 
1749 /**
1750  * get_queue_result - retrieve the result code from sem_queue
1751  * @q: Pointer to queue structure
1752  *
1753  * Retrieve the return code from the pending queue. If IN_WAKEUP is found in
1754  * q->status, then we must loop until the value is replaced with the final
1755  * value: This may happen if a task is woken up by an unrelated event (e.g.
1756  * signal) and in parallel the task is woken up by another task because it got
1757  * the requested semaphores.
1758  *
1759  * The function can be called with or without holding the semaphore spinlock.
1760  */
1761 static int get_queue_result(struct sem_queue *q)
1762 {
1763 	int error;
1764 
1765 	error = q->status;
1766 	while (unlikely(error == IN_WAKEUP)) {
1767 		cpu_relax();
1768 		error = q->status;
1769 	}
1770 
1771 	return error;
1772 }
1773 
1774 SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
1775 		unsigned, nsops, const struct timespec __user *, timeout)
1776 {
1777 	int error = -EINVAL;
1778 	struct sem_array *sma;
1779 	struct sembuf fast_sops[SEMOPM_FAST];
1780 	struct sembuf *sops = fast_sops, *sop;
1781 	struct sem_undo *un;
1782 	int undos = 0, alter = 0, max, locknum;
1783 	struct sem_queue queue;
1784 	unsigned long jiffies_left = 0;
1785 	struct ipc_namespace *ns;
1786 	struct list_head tasks;
1787 
1788 	ns = current->nsproxy->ipc_ns;
1789 
1790 	if (nsops < 1 || semid < 0)
1791 		return -EINVAL;
1792 	if (nsops > ns->sc_semopm)
1793 		return -E2BIG;
1794 	if (nsops > SEMOPM_FAST) {
1795 		sops = kmalloc(sizeof(*sops)*nsops, GFP_KERNEL);
1796 		if (sops == NULL)
1797 			return -ENOMEM;
1798 	}
1799 	if (copy_from_user(sops, tsops, nsops * sizeof(*tsops))) {
1800 		error =  -EFAULT;
1801 		goto out_free;
1802 	}
1803 	if (timeout) {
1804 		struct timespec _timeout;
1805 		if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1806 			error = -EFAULT;
1807 			goto out_free;
1808 		}
1809 		if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1810 			_timeout.tv_nsec >= 1000000000L) {
1811 			error = -EINVAL;
1812 			goto out_free;
1813 		}
1814 		jiffies_left = timespec_to_jiffies(&_timeout);
1815 	}
1816 	max = 0;
1817 	for (sop = sops; sop < sops + nsops; sop++) {
1818 		if (sop->sem_num >= max)
1819 			max = sop->sem_num;
1820 		if (sop->sem_flg & SEM_UNDO)
1821 			undos = 1;
1822 		if (sop->sem_op != 0)
1823 			alter = 1;
1824 	}
1825 
1826 	INIT_LIST_HEAD(&tasks);
1827 
1828 	if (undos) {
1829 		/* On success, find_alloc_undo takes the rcu_read_lock */
1830 		un = find_alloc_undo(ns, semid);
1831 		if (IS_ERR(un)) {
1832 			error = PTR_ERR(un);
1833 			goto out_free;
1834 		}
1835 	} else {
1836 		un = NULL;
1837 		rcu_read_lock();
1838 	}
1839 
1840 	sma = sem_obtain_object_check(ns, semid);
1841 	if (IS_ERR(sma)) {
1842 		rcu_read_unlock();
1843 		error = PTR_ERR(sma);
1844 		goto out_free;
1845 	}
1846 
1847 	error = -EFBIG;
1848 	if (max >= sma->sem_nsems)
1849 		goto out_rcu_wakeup;
1850 
1851 	error = -EACCES;
1852 	if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO))
1853 		goto out_rcu_wakeup;
1854 
1855 	error = security_sem_semop(sma, sops, nsops, alter);
1856 	if (error)
1857 		goto out_rcu_wakeup;
1858 
1859 	error = -EIDRM;
1860 	locknum = sem_lock(sma, sops, nsops);
1861 	/*
1862 	 * We eventually might perform the following check in a lockless
1863 	 * fashion, considering ipc_valid_object() locking constraints.
1864 	 * If nsops == 1 and there is no contention for sem_perm.lock, then
1865 	 * only a per-semaphore lock is held and it's OK to proceed with the
1866 	 * check below. More details on the fine grained locking scheme
1867 	 * entangled here and why it's RMID race safe on comments at sem_lock()
1868 	 */
1869 	if (!ipc_valid_object(&sma->sem_perm))
1870 		goto out_unlock_free;
1871 	/*
1872 	 * semid identifiers are not unique - find_alloc_undo may have
1873 	 * allocated an undo structure, it was invalidated by an RMID
1874 	 * and now a new array with received the same id. Check and fail.
1875 	 * This case can be detected checking un->semid. The existence of
1876 	 * "un" itself is guaranteed by rcu.
1877 	 */
1878 	if (un && un->semid == -1)
1879 		goto out_unlock_free;
1880 
1881 	queue.sops = sops;
1882 	queue.nsops = nsops;
1883 	queue.undo = un;
1884 	queue.pid = task_tgid_vnr(current);
1885 	queue.alter = alter;
1886 
1887 	error = perform_atomic_semop(sma, &queue);
1888 	if (error == 0) {
1889 		/* If the operation was successful, then do
1890 		 * the required updates.
1891 		 */
1892 		if (alter)
1893 			do_smart_update(sma, sops, nsops, 1, &tasks);
1894 		else
1895 			set_semotime(sma, sops);
1896 	}
1897 	if (error <= 0)
1898 		goto out_unlock_free;
1899 
1900 	/* We need to sleep on this operation, so we put the current
1901 	 * task into the pending queue and go to sleep.
1902 	 */
1903 
1904 	if (nsops == 1) {
1905 		struct sem *curr;
1906 		curr = &sma->sem_base[sops->sem_num];
1907 
1908 		if (alter) {
1909 			if (sma->complex_count) {
1910 				list_add_tail(&queue.list,
1911 						&sma->pending_alter);
1912 			} else {
1913 
1914 				list_add_tail(&queue.list,
1915 						&curr->pending_alter);
1916 			}
1917 		} else {
1918 			list_add_tail(&queue.list, &curr->pending_const);
1919 		}
1920 	} else {
1921 		if (!sma->complex_count)
1922 			merge_queues(sma);
1923 
1924 		if (alter)
1925 			list_add_tail(&queue.list, &sma->pending_alter);
1926 		else
1927 			list_add_tail(&queue.list, &sma->pending_const);
1928 
1929 		sma->complex_count++;
1930 	}
1931 
1932 	queue.status = -EINTR;
1933 	queue.sleeper = current;
1934 
1935 sleep_again:
1936 	current->state = TASK_INTERRUPTIBLE;
1937 	sem_unlock(sma, locknum);
1938 	rcu_read_unlock();
1939 
1940 	if (timeout)
1941 		jiffies_left = schedule_timeout(jiffies_left);
1942 	else
1943 		schedule();
1944 
1945 	error = get_queue_result(&queue);
1946 
1947 	if (error != -EINTR) {
1948 		/* fast path: update_queue already obtained all requested
1949 		 * resources.
1950 		 * Perform a smp_mb(): User space could assume that semop()
1951 		 * is a memory barrier: Without the mb(), the cpu could
1952 		 * speculatively read in user space stale data that was
1953 		 * overwritten by the previous owner of the semaphore.
1954 		 */
1955 		smp_mb();
1956 
1957 		goto out_free;
1958 	}
1959 
1960 	rcu_read_lock();
1961 	sma = sem_obtain_lock(ns, semid, sops, nsops, &locknum);
1962 
1963 	/*
1964 	 * Wait until it's guaranteed that no wakeup_sem_queue_do() is ongoing.
1965 	 */
1966 	error = get_queue_result(&queue);
1967 
1968 	/*
1969 	 * Array removed? If yes, leave without sem_unlock().
1970 	 */
1971 	if (IS_ERR(sma)) {
1972 		rcu_read_unlock();
1973 		goto out_free;
1974 	}
1975 
1976 
1977 	/*
1978 	 * If queue.status != -EINTR we are woken up by another process.
1979 	 * Leave without unlink_queue(), but with sem_unlock().
1980 	 */
1981 	if (error != -EINTR)
1982 		goto out_unlock_free;
1983 
1984 	/*
1985 	 * If an interrupt occurred we have to clean up the queue
1986 	 */
1987 	if (timeout && jiffies_left == 0)
1988 		error = -EAGAIN;
1989 
1990 	/*
1991 	 * If the wakeup was spurious, just retry
1992 	 */
1993 	if (error == -EINTR && !signal_pending(current))
1994 		goto sleep_again;
1995 
1996 	unlink_queue(sma, &queue);
1997 
1998 out_unlock_free:
1999 	sem_unlock(sma, locknum);
2000 out_rcu_wakeup:
2001 	rcu_read_unlock();
2002 	wake_up_sem_queue_do(&tasks);
2003 out_free:
2004 	if (sops != fast_sops)
2005 		kfree(sops);
2006 	return error;
2007 }
2008 
2009 SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
2010 		unsigned, nsops)
2011 {
2012 	return sys_semtimedop(semid, tsops, nsops, NULL);
2013 }
2014 
2015 /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
2016  * parent and child tasks.
2017  */
2018 
2019 int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
2020 {
2021 	struct sem_undo_list *undo_list;
2022 	int error;
2023 
2024 	if (clone_flags & CLONE_SYSVSEM) {
2025 		error = get_undo_list(&undo_list);
2026 		if (error)
2027 			return error;
2028 		atomic_inc(&undo_list->refcnt);
2029 		tsk->sysvsem.undo_list = undo_list;
2030 	} else
2031 		tsk->sysvsem.undo_list = NULL;
2032 
2033 	return 0;
2034 }
2035 
2036 /*
2037  * add semadj values to semaphores, free undo structures.
2038  * undo structures are not freed when semaphore arrays are destroyed
2039  * so some of them may be out of date.
2040  * IMPLEMENTATION NOTE: There is some confusion over whether the
2041  * set of adjustments that needs to be done should be done in an atomic
2042  * manner or not. That is, if we are attempting to decrement the semval
2043  * should we queue up and wait until we can do so legally?
2044  * The original implementation attempted to do this (queue and wait).
2045  * The current implementation does not do so. The POSIX standard
2046  * and SVID should be consulted to determine what behavior is mandated.
2047  */
2048 void exit_sem(struct task_struct *tsk)
2049 {
2050 	struct sem_undo_list *ulp;
2051 
2052 	ulp = tsk->sysvsem.undo_list;
2053 	if (!ulp)
2054 		return;
2055 	tsk->sysvsem.undo_list = NULL;
2056 
2057 	if (!atomic_dec_and_test(&ulp->refcnt))
2058 		return;
2059 
2060 	for (;;) {
2061 		struct sem_array *sma;
2062 		struct sem_undo *un;
2063 		struct list_head tasks;
2064 		int semid, i;
2065 
2066 		rcu_read_lock();
2067 		un = list_entry_rcu(ulp->list_proc.next,
2068 				    struct sem_undo, list_proc);
2069 		if (&un->list_proc == &ulp->list_proc)
2070 			semid = -1;
2071 		 else
2072 			semid = un->semid;
2073 
2074 		if (semid == -1) {
2075 			rcu_read_unlock();
2076 			break;
2077 		}
2078 
2079 		sma = sem_obtain_object_check(tsk->nsproxy->ipc_ns, un->semid);
2080 		/* exit_sem raced with IPC_RMID, nothing to do */
2081 		if (IS_ERR(sma)) {
2082 			rcu_read_unlock();
2083 			continue;
2084 		}
2085 
2086 		sem_lock(sma, NULL, -1);
2087 		/* exit_sem raced with IPC_RMID, nothing to do */
2088 		if (!ipc_valid_object(&sma->sem_perm)) {
2089 			sem_unlock(sma, -1);
2090 			rcu_read_unlock();
2091 			continue;
2092 		}
2093 		un = __lookup_undo(ulp, semid);
2094 		if (un == NULL) {
2095 			/* exit_sem raced with IPC_RMID+semget() that created
2096 			 * exactly the same semid. Nothing to do.
2097 			 */
2098 			sem_unlock(sma, -1);
2099 			rcu_read_unlock();
2100 			continue;
2101 		}
2102 
2103 		/* remove un from the linked lists */
2104 		ipc_assert_locked_object(&sma->sem_perm);
2105 		list_del(&un->list_id);
2106 
2107 		spin_lock(&ulp->lock);
2108 		list_del_rcu(&un->list_proc);
2109 		spin_unlock(&ulp->lock);
2110 
2111 		/* perform adjustments registered in un */
2112 		for (i = 0; i < sma->sem_nsems; i++) {
2113 			struct sem *semaphore = &sma->sem_base[i];
2114 			if (un->semadj[i]) {
2115 				semaphore->semval += un->semadj[i];
2116 				/*
2117 				 * Range checks of the new semaphore value,
2118 				 * not defined by sus:
2119 				 * - Some unices ignore the undo entirely
2120 				 *   (e.g. HP UX 11i 11.22, Tru64 V5.1)
2121 				 * - some cap the value (e.g. FreeBSD caps
2122 				 *   at 0, but doesn't enforce SEMVMX)
2123 				 *
2124 				 * Linux caps the semaphore value, both at 0
2125 				 * and at SEMVMX.
2126 				 *
2127 				 *	Manfred <manfred@colorfullife.com>
2128 				 */
2129 				if (semaphore->semval < 0)
2130 					semaphore->semval = 0;
2131 				if (semaphore->semval > SEMVMX)
2132 					semaphore->semval = SEMVMX;
2133 				semaphore->sempid = task_tgid_vnr(current);
2134 			}
2135 		}
2136 		/* maybe some queued-up processes were waiting for this */
2137 		INIT_LIST_HEAD(&tasks);
2138 		do_smart_update(sma, NULL, 0, 1, &tasks);
2139 		sem_unlock(sma, -1);
2140 		rcu_read_unlock();
2141 		wake_up_sem_queue_do(&tasks);
2142 
2143 		kfree_rcu(un, rcu);
2144 	}
2145 	kfree(ulp);
2146 }
2147 
2148 #ifdef CONFIG_PROC_FS
2149 static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
2150 {
2151 	struct user_namespace *user_ns = seq_user_ns(s);
2152 	struct sem_array *sma = it;
2153 	time_t sem_otime;
2154 
2155 	/*
2156 	 * The proc interface isn't aware of sem_lock(), it calls
2157 	 * ipc_lock_object() directly (in sysvipc_find_ipc).
2158 	 * In order to stay compatible with sem_lock(), we must wait until
2159 	 * all simple semop() calls have left their critical regions.
2160 	 */
2161 	sem_wait_array(sma);
2162 
2163 	sem_otime = get_semotime(sma);
2164 
2165 	return seq_printf(s,
2166 			  "%10d %10d  %4o %10u %5u %5u %5u %5u %10lu %10lu\n",
2167 			  sma->sem_perm.key,
2168 			  sma->sem_perm.id,
2169 			  sma->sem_perm.mode,
2170 			  sma->sem_nsems,
2171 			  from_kuid_munged(user_ns, sma->sem_perm.uid),
2172 			  from_kgid_munged(user_ns, sma->sem_perm.gid),
2173 			  from_kuid_munged(user_ns, sma->sem_perm.cuid),
2174 			  from_kgid_munged(user_ns, sma->sem_perm.cgid),
2175 			  sem_otime,
2176 			  sma->sem_ctime);
2177 }
2178 #endif
2179