sem.c revision 9ae949fa
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 * (c) 2016 Davidlohr Bueso <dave@stgolabs.net>
15 * Further wakeup optimizations, documentation
16 * (c) 2010 Manfred Spraul <manfred@colorfullife.com>
17 *
18 * support for audit of ipc object properties and permission changes
19 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
20 *
21 * namespaces support
22 * OpenVZ, SWsoft Inc.
23 * Pavel Emelianov <xemul@openvz.org>
24 *
25 * Implementation notes: (May 2010)
26 * This file implements System V semaphores.
27 *
28 * User space visible behavior:
29 * - FIFO ordering for semop() operations (just FIFO, not starvation
30 *   protection)
31 * - multiple semaphore operations that alter the same semaphore in
32 *   one semop() are handled.
33 * - sem_ctime (time of last semctl()) is updated in the IPC_SET, SETVAL and
34 *   SETALL calls.
35 * - two Linux specific semctl() commands: SEM_STAT, SEM_INFO.
36 * - undo adjustments at process exit are limited to 0..SEMVMX.
37 * - namespace are supported.
38 * - SEMMSL, SEMMNS, SEMOPM and SEMMNI can be configured at runtine by writing
39 *   to /proc/sys/kernel/sem.
40 * - statistics about the usage are reported in /proc/sysvipc/sem.
41 *
42 * Internals:
43 * - scalability:
44 *   - all global variables are read-mostly.
45 *   - semop() calls and semctl(RMID) are synchronized by RCU.
46 *   - most operations do write operations (actually: spin_lock calls) to
47 *     the per-semaphore array structure.
48 *   Thus: Perfect SMP scaling between independent semaphore arrays.
49 *         If multiple semaphores in one array are used, then cache line
50 *         trashing on the semaphore array spinlock will limit the scaling.
51 * - semncnt and semzcnt are calculated on demand in count_semcnt()
52 * - the task that performs a successful semop() scans the list of all
53 *   sleeping tasks and completes any pending operations that can be fulfilled.
54 *   Semaphores are actively given to waiting tasks (necessary for FIFO).
55 *   (see update_queue())
56 * - To improve the scalability, the actual wake-up calls are performed after
57 *   dropping all locks. (see wake_up_sem_queue_prepare())
58 * - 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 * - UNDO values are stored in an array (one per process and per
63 *   semaphore array, lazily allocated). For backwards compatibility, multiple
64 *   modes for the UNDO variables are supported (per process, per thread)
65 *   (see copy_semundo, CLONE_SYSVSEM)
66 * - There are two lists of the pending operations: a per-array list
67 *   and per-semaphore list (stored in the array). This allows to achieve FIFO
68 *   ordering without always scanning all pending operations.
69 *   The worst-case behavior is nevertheless O(N^2) for N wakeups.
70 */
71
72#include <linux/slab.h>
73#include <linux/spinlock.h>
74#include <linux/init.h>
75#include <linux/proc_fs.h>
76#include <linux/time.h>
77#include <linux/security.h>
78#include <linux/syscalls.h>
79#include <linux/audit.h>
80#include <linux/capability.h>
81#include <linux/seq_file.h>
82#include <linux/rwsem.h>
83#include <linux/nsproxy.h>
84#include <linux/ipc_namespace.h>
85
86#include <linux/uaccess.h>
87#include "util.h"
88
89/* One semaphore structure for each semaphore in the system. */
90struct sem {
91	int	semval;		/* current value */
92	/*
93	 * PID of the process that last modified the semaphore. For
94	 * Linux, specifically these are:
95	 *  - semop
96	 *  - semctl, via SETVAL and SETALL.
97	 *  - at task exit when performing undo adjustments (see exit_sem).
98	 */
99	int	sempid;
100	spinlock_t	lock;	/* spinlock for fine-grained semtimedop */
101	struct list_head pending_alter; /* pending single-sop operations */
102					/* that alter the semaphore */
103	struct list_head pending_const; /* pending single-sop operations */
104					/* that do not alter the semaphore*/
105	time_t	sem_otime;	/* candidate for sem_otime */
106} ____cacheline_aligned_in_smp;
107
108/* One queue for each sleeping process in the system. */
109struct sem_queue {
110	struct list_head	list;	 /* queue of pending operations */
111	struct task_struct	*sleeper; /* this process */
112	struct sem_undo		*undo;	 /* undo structure */
113	int			pid;	 /* process id of requesting process */
114	int			status;	 /* completion status of operation */
115	struct sembuf		*sops;	 /* array of pending operations */
116	struct sembuf		*blocking; /* the operation that blocked */
117	int			nsops;	 /* number of operations */
118	int			alter;	 /* does *sops alter the array? */
119};
120
121/* Each task has a list of undo requests. They are executed automatically
122 * when the process exits.
123 */
124struct sem_undo {
125	struct list_head	list_proc;	/* per-process list: *
126						 * all undos from one process
127						 * rcu protected */
128	struct rcu_head		rcu;		/* rcu struct for sem_undo */
129	struct sem_undo_list	*ulp;		/* back ptr to sem_undo_list */
130	struct list_head	list_id;	/* per semaphore array list:
131						 * all undos for one array */
132	int			semid;		/* semaphore set identifier */
133	short			*semadj;	/* array of adjustments */
134						/* one per semaphore */
135};
136
137/* sem_undo_list controls shared access to the list of sem_undo structures
138 * that may be shared among all a CLONE_SYSVSEM task group.
139 */
140struct sem_undo_list {
141	atomic_t		refcnt;
142	spinlock_t		lock;
143	struct list_head	list_proc;
144};
145
146
147#define sem_ids(ns)	((ns)->ids[IPC_SEM_IDS])
148
149#define sem_checkid(sma, semid)	ipc_checkid(&sma->sem_perm, semid)
150
151static int newary(struct ipc_namespace *, struct ipc_params *);
152static void freeary(struct ipc_namespace *, struct kern_ipc_perm *);
153#ifdef CONFIG_PROC_FS
154static int sysvipc_sem_proc_show(struct seq_file *s, void *it);
155#endif
156
157#define SEMMSL_FAST	256 /* 512 bytes on stack */
158#define SEMOPM_FAST	64  /* ~ 372 bytes on stack */
159
160/*
161 * Locking:
162 * a) global sem_lock() for read/write
163 *	sem_undo.id_next,
164 *	sem_array.complex_count,
165 *	sem_array.complex_mode
166 *	sem_array.pending{_alter,_const},
167 *	sem_array.sem_undo
168 *
169 * b) global or semaphore sem_lock() for read/write:
170 *	sem_array.sem_base[i].pending_{const,alter}:
171 *	sem_array.complex_mode (for read)
172 *
173 * c) special:
174 *	sem_undo_list.list_proc:
175 *	* undo_list->lock for write
176 *	* rcu for read
177 */
178
179#define sc_semmsl	sem_ctls[0]
180#define sc_semmns	sem_ctls[1]
181#define sc_semopm	sem_ctls[2]
182#define sc_semmni	sem_ctls[3]
183
184void sem_init_ns(struct ipc_namespace *ns)
185{
186	ns->sc_semmsl = SEMMSL;
187	ns->sc_semmns = SEMMNS;
188	ns->sc_semopm = SEMOPM;
189	ns->sc_semmni = SEMMNI;
190	ns->used_sems = 0;
191	ipc_init_ids(&ns->ids[IPC_SEM_IDS]);
192}
193
194#ifdef CONFIG_IPC_NS
195void sem_exit_ns(struct ipc_namespace *ns)
196{
197	free_ipcs(ns, &sem_ids(ns), freeary);
198	idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr);
199}
200#endif
201
202void __init sem_init(void)
203{
204	sem_init_ns(&init_ipc_ns);
205	ipc_init_proc_interface("sysvipc/sem",
206				"       key      semid perms      nsems   uid   gid  cuid  cgid      otime      ctime\n",
207				IPC_SEM_IDS, sysvipc_sem_proc_show);
208}
209
210/**
211 * unmerge_queues - unmerge queues, if possible.
212 * @sma: semaphore array
213 *
214 * The function unmerges the wait queues if complex_count is 0.
215 * It must be called prior to dropping the global semaphore array lock.
216 */
217static void unmerge_queues(struct sem_array *sma)
218{
219	struct sem_queue *q, *tq;
220
221	/* complex operations still around? */
222	if (sma->complex_count)
223		return;
224	/*
225	 * We will switch back to simple mode.
226	 * Move all pending operation back into the per-semaphore
227	 * queues.
228	 */
229	list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
230		struct sem *curr;
231		curr = &sma->sem_base[q->sops[0].sem_num];
232
233		list_add_tail(&q->list, &curr->pending_alter);
234	}
235	INIT_LIST_HEAD(&sma->pending_alter);
236}
237
238/**
239 * merge_queues - merge single semop queues into global queue
240 * @sma: semaphore array
241 *
242 * This function merges all per-semaphore queues into the global queue.
243 * It is necessary to achieve FIFO ordering for the pending single-sop
244 * operations when a multi-semop operation must sleep.
245 * Only the alter operations must be moved, the const operations can stay.
246 */
247static void merge_queues(struct sem_array *sma)
248{
249	int i;
250	for (i = 0; i < sma->sem_nsems; i++) {
251		struct sem *sem = sma->sem_base + i;
252
253		list_splice_init(&sem->pending_alter, &sma->pending_alter);
254	}
255}
256
257static void sem_rcu_free(struct rcu_head *head)
258{
259	struct ipc_rcu *p = container_of(head, struct ipc_rcu, rcu);
260	struct sem_array *sma = ipc_rcu_to_struct(p);
261
262	security_sem_free(sma);
263	ipc_rcu_free(head);
264}
265
266/*
267 * Enter the mode suitable for non-simple operations:
268 * Caller must own sem_perm.lock.
269 */
270static void complexmode_enter(struct sem_array *sma)
271{
272	int i;
273	struct sem *sem;
274
275	if (sma->complex_mode)  {
276		/* We are already in complex_mode. Nothing to do */
277		return;
278	}
279
280	/* We need a full barrier after seting complex_mode:
281	 * The write to complex_mode must be visible
282	 * before we read the first sem->lock spinlock state.
283	 */
284	smp_store_mb(sma->complex_mode, true);
285
286	for (i = 0; i < sma->sem_nsems; i++) {
287		sem = sma->sem_base + i;
288		spin_unlock_wait(&sem->lock);
289	}
290	/*
291	 * spin_unlock_wait() is not a memory barriers, it is only a
292	 * control barrier. The code must pair with spin_unlock(&sem->lock),
293	 * thus just the control barrier is insufficient.
294	 *
295	 * smp_rmb() is sufficient, as writes cannot pass the control barrier.
296	 */
297	smp_rmb();
298}
299
300/*
301 * Try to leave the mode that disallows simple operations:
302 * Caller must own sem_perm.lock.
303 */
304static void complexmode_tryleave(struct sem_array *sma)
305{
306	if (sma->complex_count)  {
307		/* Complex ops are sleeping.
308		 * We must stay in complex mode
309		 */
310		return;
311	}
312	/*
313	 * Immediately after setting complex_mode to false,
314	 * a simple op can start. Thus: all memory writes
315	 * performed by the current operation must be visible
316	 * before we set complex_mode to false.
317	 */
318	smp_store_release(&sma->complex_mode, false);
319}
320
321#define SEM_GLOBAL_LOCK	(-1)
322/*
323 * If the request contains only one semaphore operation, and there are
324 * no complex transactions pending, lock only the semaphore involved.
325 * Otherwise, lock the entire semaphore array, since we either have
326 * multiple semaphores in our own semops, or we need to look at
327 * semaphores from other pending complex operations.
328 */
329static inline int sem_lock(struct sem_array *sma, struct sembuf *sops,
330			      int nsops)
331{
332	struct sem *sem;
333
334	if (nsops != 1) {
335		/* Complex operation - acquire a full lock */
336		ipc_lock_object(&sma->sem_perm);
337
338		/* Prevent parallel simple ops */
339		complexmode_enter(sma);
340		return SEM_GLOBAL_LOCK;
341	}
342
343	/*
344	 * Only one semaphore affected - try to optimize locking.
345	 * Optimized locking is possible if no complex operation
346	 * is either enqueued or processed right now.
347	 *
348	 * Both facts are tracked by complex_mode.
349	 */
350	sem = sma->sem_base + sops->sem_num;
351
352	/*
353	 * Initial check for complex_mode. Just an optimization,
354	 * no locking, no memory barrier.
355	 */
356	if (!sma->complex_mode) {
357		/*
358		 * It appears that no complex operation is around.
359		 * Acquire the per-semaphore lock.
360		 */
361		spin_lock(&sem->lock);
362
363		/*
364		 * See 51d7d5205d33
365		 * ("powerpc: Add smp_mb() to arch_spin_is_locked()"):
366		 * A full barrier is required: the write of sem->lock
367		 * must be visible before the read is executed
368		 */
369		smp_mb();
370
371		if (!smp_load_acquire(&sma->complex_mode)) {
372			/* fast path successful! */
373			return sops->sem_num;
374		}
375		spin_unlock(&sem->lock);
376	}
377
378	/* slow path: acquire the full lock */
379	ipc_lock_object(&sma->sem_perm);
380
381	if (sma->complex_count == 0) {
382		/* False alarm:
383		 * There is no complex operation, thus we can switch
384		 * back to the fast path.
385		 */
386		spin_lock(&sem->lock);
387		ipc_unlock_object(&sma->sem_perm);
388		return sops->sem_num;
389	} else {
390		/* Not a false alarm, thus complete the sequence for a
391		 * full lock.
392		 */
393		complexmode_enter(sma);
394		return SEM_GLOBAL_LOCK;
395	}
396}
397
398static inline void sem_unlock(struct sem_array *sma, int locknum)
399{
400	if (locknum == SEM_GLOBAL_LOCK) {
401		unmerge_queues(sma);
402		complexmode_tryleave(sma);
403		ipc_unlock_object(&sma->sem_perm);
404	} else {
405		struct sem *sem = sma->sem_base + locknum;
406		spin_unlock(&sem->lock);
407	}
408}
409
410/*
411 * sem_lock_(check_) routines are called in the paths where the rwsem
412 * is not held.
413 *
414 * The caller holds the RCU read lock.
415 */
416static inline struct sem_array *sem_obtain_lock(struct ipc_namespace *ns,
417			int id, struct sembuf *sops, int nsops, int *locknum)
418{
419	struct kern_ipc_perm *ipcp;
420	struct sem_array *sma;
421
422	ipcp = ipc_obtain_object_idr(&sem_ids(ns), id);
423	if (IS_ERR(ipcp))
424		return ERR_CAST(ipcp);
425
426	sma = container_of(ipcp, struct sem_array, sem_perm);
427	*locknum = sem_lock(sma, sops, nsops);
428
429	/* ipc_rmid() may have already freed the ID while sem_lock
430	 * was spinning: verify that the structure is still valid
431	 */
432	if (ipc_valid_object(ipcp))
433		return container_of(ipcp, struct sem_array, sem_perm);
434
435	sem_unlock(sma, *locknum);
436	return ERR_PTR(-EINVAL);
437}
438
439static inline struct sem_array *sem_obtain_object(struct ipc_namespace *ns, int id)
440{
441	struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&sem_ids(ns), id);
442
443	if (IS_ERR(ipcp))
444		return ERR_CAST(ipcp);
445
446	return container_of(ipcp, struct sem_array, sem_perm);
447}
448
449static inline struct sem_array *sem_obtain_object_check(struct ipc_namespace *ns,
450							int id)
451{
452	struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&sem_ids(ns), id);
453
454	if (IS_ERR(ipcp))
455		return ERR_CAST(ipcp);
456
457	return container_of(ipcp, struct sem_array, sem_perm);
458}
459
460static inline void sem_lock_and_putref(struct sem_array *sma)
461{
462	sem_lock(sma, NULL, -1);
463	ipc_rcu_putref(sma, sem_rcu_free);
464}
465
466static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
467{
468	ipc_rmid(&sem_ids(ns), &s->sem_perm);
469}
470
471/**
472 * newary - Create a new semaphore set
473 * @ns: namespace
474 * @params: ptr to the structure that contains key, semflg and nsems
475 *
476 * Called with sem_ids.rwsem held (as a writer)
477 */
478static int newary(struct ipc_namespace *ns, struct ipc_params *params)
479{
480	int id;
481	int retval;
482	struct sem_array *sma;
483	int size;
484	key_t key = params->key;
485	int nsems = params->u.nsems;
486	int semflg = params->flg;
487	int i;
488
489	if (!nsems)
490		return -EINVAL;
491	if (ns->used_sems + nsems > ns->sc_semmns)
492		return -ENOSPC;
493
494	size = sizeof(*sma) + nsems * sizeof(struct sem);
495	sma = ipc_rcu_alloc(size);
496	if (!sma)
497		return -ENOMEM;
498
499	memset(sma, 0, size);
500
501	sma->sem_perm.mode = (semflg & S_IRWXUGO);
502	sma->sem_perm.key = key;
503
504	sma->sem_perm.security = NULL;
505	retval = security_sem_alloc(sma);
506	if (retval) {
507		ipc_rcu_putref(sma, ipc_rcu_free);
508		return retval;
509	}
510
511	sma->sem_base = (struct sem *) &sma[1];
512
513	for (i = 0; i < nsems; i++) {
514		INIT_LIST_HEAD(&sma->sem_base[i].pending_alter);
515		INIT_LIST_HEAD(&sma->sem_base[i].pending_const);
516		spin_lock_init(&sma->sem_base[i].lock);
517	}
518
519	sma->complex_count = 0;
520	sma->complex_mode = true; /* dropped by sem_unlock below */
521	INIT_LIST_HEAD(&sma->pending_alter);
522	INIT_LIST_HEAD(&sma->pending_const);
523	INIT_LIST_HEAD(&sma->list_id);
524	sma->sem_nsems = nsems;
525	sma->sem_ctime = get_seconds();
526
527	id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
528	if (id < 0) {
529		ipc_rcu_putref(sma, sem_rcu_free);
530		return id;
531	}
532	ns->used_sems += nsems;
533
534	sem_unlock(sma, -1);
535	rcu_read_unlock();
536
537	return sma->sem_perm.id;
538}
539
540
541/*
542 * Called with sem_ids.rwsem and ipcp locked.
543 */
544static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg)
545{
546	struct sem_array *sma;
547
548	sma = container_of(ipcp, struct sem_array, sem_perm);
549	return security_sem_associate(sma, semflg);
550}
551
552/*
553 * Called with sem_ids.rwsem and ipcp locked.
554 */
555static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
556				struct ipc_params *params)
557{
558	struct sem_array *sma;
559
560	sma = container_of(ipcp, struct sem_array, sem_perm);
561	if (params->u.nsems > sma->sem_nsems)
562		return -EINVAL;
563
564	return 0;
565}
566
567SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg)
568{
569	struct ipc_namespace *ns;
570	static const struct ipc_ops sem_ops = {
571		.getnew = newary,
572		.associate = sem_security,
573		.more_checks = sem_more_checks,
574	};
575	struct ipc_params sem_params;
576
577	ns = current->nsproxy->ipc_ns;
578
579	if (nsems < 0 || nsems > ns->sc_semmsl)
580		return -EINVAL;
581
582	sem_params.key = key;
583	sem_params.flg = semflg;
584	sem_params.u.nsems = nsems;
585
586	return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
587}
588
589/**
590 * perform_atomic_semop - Perform (if possible) a semaphore operation
591 * @sma: semaphore array
592 * @q: struct sem_queue that describes the operation
593 *
594 * Returns 0 if the operation was possible.
595 * Returns 1 if the operation is impossible, the caller must sleep.
596 * Negative values are error codes.
597 */
598static int perform_atomic_semop(struct sem_array *sma, struct sem_queue *q)
599{
600	int result, sem_op, nsops, pid;
601	struct sembuf *sop;
602	struct sem *curr;
603	struct sembuf *sops;
604	struct sem_undo *un;
605
606	sops = q->sops;
607	nsops = q->nsops;
608	un = q->undo;
609
610	for (sop = sops; sop < sops + nsops; sop++) {
611		curr = sma->sem_base + sop->sem_num;
612		sem_op = sop->sem_op;
613		result = curr->semval;
614
615		if (!sem_op && result)
616			goto would_block;
617
618		result += sem_op;
619		if (result < 0)
620			goto would_block;
621		if (result > SEMVMX)
622			goto out_of_range;
623
624		if (sop->sem_flg & SEM_UNDO) {
625			int undo = un->semadj[sop->sem_num] - sem_op;
626			/* Exceeding the undo range is an error. */
627			if (undo < (-SEMAEM - 1) || undo > SEMAEM)
628				goto out_of_range;
629			un->semadj[sop->sem_num] = undo;
630		}
631
632		curr->semval = result;
633	}
634
635	sop--;
636	pid = q->pid;
637	while (sop >= sops) {
638		sma->sem_base[sop->sem_num].sempid = pid;
639		sop--;
640	}
641
642	return 0;
643
644out_of_range:
645	result = -ERANGE;
646	goto undo;
647
648would_block:
649	q->blocking = sop;
650
651	if (sop->sem_flg & IPC_NOWAIT)
652		result = -EAGAIN;
653	else
654		result = 1;
655
656undo:
657	sop--;
658	while (sop >= sops) {
659		sem_op = sop->sem_op;
660		sma->sem_base[sop->sem_num].semval -= sem_op;
661		if (sop->sem_flg & SEM_UNDO)
662			un->semadj[sop->sem_num] += sem_op;
663		sop--;
664	}
665
666	return result;
667}
668
669static inline void wake_up_sem_queue_prepare(struct sem_queue *q, int error,
670					     struct wake_q_head *wake_q)
671{
672	wake_q_add(wake_q, q->sleeper);
673	/*
674	 * Rely on the above implicit barrier, such that we can
675	 * ensure that we hold reference to the task before setting
676	 * q->status. Otherwise we could race with do_exit if the
677	 * task is awoken by an external event before calling
678	 * wake_up_process().
679	 */
680	WRITE_ONCE(q->status, error);
681}
682
683static void unlink_queue(struct sem_array *sma, struct sem_queue *q)
684{
685	list_del(&q->list);
686	if (q->nsops > 1)
687		sma->complex_count--;
688}
689
690/** check_restart(sma, q)
691 * @sma: semaphore array
692 * @q: the operation that just completed
693 *
694 * update_queue is O(N^2) when it restarts scanning the whole queue of
695 * waiting operations. Therefore this function checks if the restart is
696 * really necessary. It is called after a previously waiting operation
697 * modified the array.
698 * Note that wait-for-zero operations are handled without restart.
699 */
700static int check_restart(struct sem_array *sma, struct sem_queue *q)
701{
702	/* pending complex alter operations are too difficult to analyse */
703	if (!list_empty(&sma->pending_alter))
704		return 1;
705
706	/* we were a sleeping complex operation. Too difficult */
707	if (q->nsops > 1)
708		return 1;
709
710	/* It is impossible that someone waits for the new value:
711	 * - complex operations always restart.
712	 * - wait-for-zero are handled seperately.
713	 * - q is a previously sleeping simple operation that
714	 *   altered the array. It must be a decrement, because
715	 *   simple increments never sleep.
716	 * - If there are older (higher priority) decrements
717	 *   in the queue, then they have observed the original
718	 *   semval value and couldn't proceed. The operation
719	 *   decremented to value - thus they won't proceed either.
720	 */
721	return 0;
722}
723
724/**
725 * wake_const_ops - wake up non-alter tasks
726 * @sma: semaphore array.
727 * @semnum: semaphore that was modified.
728 * @wake_q: lockless wake-queue head.
729 *
730 * wake_const_ops must be called after a semaphore in a semaphore array
731 * was set to 0. If complex const operations are pending, wake_const_ops must
732 * be called with semnum = -1, as well as with the number of each modified
733 * semaphore.
734 * The tasks that must be woken up are added to @wake_q. The return code
735 * is stored in q->pid.
736 * The function returns 1 if at least one operation was completed successfully.
737 */
738static int wake_const_ops(struct sem_array *sma, int semnum,
739			  struct wake_q_head *wake_q)
740{
741	struct sem_queue *q;
742	struct list_head *walk;
743	struct list_head *pending_list;
744	int semop_completed = 0;
745
746	if (semnum == -1)
747		pending_list = &sma->pending_const;
748	else
749		pending_list = &sma->sem_base[semnum].pending_const;
750
751	walk = pending_list->next;
752	while (walk != pending_list) {
753		int error;
754
755		q = container_of(walk, struct sem_queue, list);
756		walk = walk->next;
757
758		error = perform_atomic_semop(sma, q);
759
760		if (error <= 0) {
761			/* operation completed, remove from queue & wakeup */
762
763			unlink_queue(sma, q);
764
765			wake_up_sem_queue_prepare(q, error, wake_q);
766			if (error == 0)
767				semop_completed = 1;
768		}
769	}
770	return semop_completed;
771}
772
773/**
774 * do_smart_wakeup_zero - wakeup all wait for zero tasks
775 * @sma: semaphore array
776 * @sops: operations that were performed
777 * @nsops: number of operations
778 * @wake_q: lockless wake-queue head
779 *
780 * Checks all required queue for wait-for-zero operations, based
781 * on the actual changes that were performed on the semaphore array.
782 * The function returns 1 if at least one operation was completed successfully.
783 */
784static int do_smart_wakeup_zero(struct sem_array *sma, struct sembuf *sops,
785				int nsops, struct wake_q_head *wake_q)
786{
787	int i;
788	int semop_completed = 0;
789	int got_zero = 0;
790
791	/* first: the per-semaphore queues, if known */
792	if (sops) {
793		for (i = 0; i < nsops; i++) {
794			int num = sops[i].sem_num;
795
796			if (sma->sem_base[num].semval == 0) {
797				got_zero = 1;
798				semop_completed |= wake_const_ops(sma, num, wake_q);
799			}
800		}
801	} else {
802		/*
803		 * No sops means modified semaphores not known.
804		 * Assume all were changed.
805		 */
806		for (i = 0; i < sma->sem_nsems; i++) {
807			if (sma->sem_base[i].semval == 0) {
808				got_zero = 1;
809				semop_completed |= wake_const_ops(sma, i, wake_q);
810			}
811		}
812	}
813	/*
814	 * If one of the modified semaphores got 0,
815	 * then check the global queue, too.
816	 */
817	if (got_zero)
818		semop_completed |= wake_const_ops(sma, -1, wake_q);
819
820	return semop_completed;
821}
822
823
824/**
825 * update_queue - look for tasks that can be completed.
826 * @sma: semaphore array.
827 * @semnum: semaphore that was modified.
828 * @wake_q: lockless wake-queue head.
829 *
830 * update_queue must be called after a semaphore in a semaphore array
831 * was modified. If multiple semaphores were modified, update_queue must
832 * be called with semnum = -1, as well as with the number of each modified
833 * semaphore.
834 * The tasks that must be woken up are added to @wake_q. The return code
835 * is stored in q->pid.
836 * The function internally checks if const operations can now succeed.
837 *
838 * The function return 1 if at least one semop was completed successfully.
839 */
840static int update_queue(struct sem_array *sma, int semnum, struct wake_q_head *wake_q)
841{
842	struct sem_queue *q;
843	struct list_head *walk;
844	struct list_head *pending_list;
845	int semop_completed = 0;
846
847	if (semnum == -1)
848		pending_list = &sma->pending_alter;
849	else
850		pending_list = &sma->sem_base[semnum].pending_alter;
851
852again:
853	walk = pending_list->next;
854	while (walk != pending_list) {
855		int error, restart;
856
857		q = container_of(walk, struct sem_queue, list);
858		walk = walk->next;
859
860		/* If we are scanning the single sop, per-semaphore list of
861		 * one semaphore and that semaphore is 0, then it is not
862		 * necessary to scan further: simple increments
863		 * that affect only one entry succeed immediately and cannot
864		 * be in the  per semaphore pending queue, and decrements
865		 * cannot be successful if the value is already 0.
866		 */
867		if (semnum != -1 && sma->sem_base[semnum].semval == 0)
868			break;
869
870		error = perform_atomic_semop(sma, q);
871
872		/* Does q->sleeper still need to sleep? */
873		if (error > 0)
874			continue;
875
876		unlink_queue(sma, q);
877
878		if (error) {
879			restart = 0;
880		} else {
881			semop_completed = 1;
882			do_smart_wakeup_zero(sma, q->sops, q->nsops, wake_q);
883			restart = check_restart(sma, q);
884		}
885
886		wake_up_sem_queue_prepare(q, error, wake_q);
887		if (restart)
888			goto again;
889	}
890	return semop_completed;
891}
892
893/**
894 * set_semotime - set sem_otime
895 * @sma: semaphore array
896 * @sops: operations that modified the array, may be NULL
897 *
898 * sem_otime is replicated to avoid cache line trashing.
899 * This function sets one instance to the current time.
900 */
901static void set_semotime(struct sem_array *sma, struct sembuf *sops)
902{
903	if (sops == NULL) {
904		sma->sem_base[0].sem_otime = get_seconds();
905	} else {
906		sma->sem_base[sops[0].sem_num].sem_otime =
907							get_seconds();
908	}
909}
910
911/**
912 * do_smart_update - optimized update_queue
913 * @sma: semaphore array
914 * @sops: operations that were performed
915 * @nsops: number of operations
916 * @otime: force setting otime
917 * @wake_q: lockless wake-queue head
918 *
919 * do_smart_update() does the required calls to update_queue and wakeup_zero,
920 * based on the actual changes that were performed on the semaphore array.
921 * Note that the function does not do the actual wake-up: the caller is
922 * responsible for calling wake_up_q().
923 * It is safe to perform this call after dropping all locks.
924 */
925static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops,
926			    int otime, struct wake_q_head *wake_q)
927{
928	int i;
929
930	otime |= do_smart_wakeup_zero(sma, sops, nsops, wake_q);
931
932	if (!list_empty(&sma->pending_alter)) {
933		/* semaphore array uses the global queue - just process it. */
934		otime |= update_queue(sma, -1, wake_q);
935	} else {
936		if (!sops) {
937			/*
938			 * No sops, thus the modified semaphores are not
939			 * known. Check all.
940			 */
941			for (i = 0; i < sma->sem_nsems; i++)
942				otime |= update_queue(sma, i, wake_q);
943		} else {
944			/*
945			 * Check the semaphores that were increased:
946			 * - No complex ops, thus all sleeping ops are
947			 *   decrease.
948			 * - if we decreased the value, then any sleeping
949			 *   semaphore ops wont be able to run: If the
950			 *   previous value was too small, then the new
951			 *   value will be too small, too.
952			 */
953			for (i = 0; i < nsops; i++) {
954				if (sops[i].sem_op > 0) {
955					otime |= update_queue(sma,
956							      sops[i].sem_num, wake_q);
957				}
958			}
959		}
960	}
961	if (otime)
962		set_semotime(sma, sops);
963}
964
965/*
966 * check_qop: Test if a queued operation sleeps on the semaphore semnum
967 */
968static int check_qop(struct sem_array *sma, int semnum, struct sem_queue *q,
969			bool count_zero)
970{
971	struct sembuf *sop = q->blocking;
972
973	/*
974	 * Linux always (since 0.99.10) reported a task as sleeping on all
975	 * semaphores. This violates SUS, therefore it was changed to the
976	 * standard compliant behavior.
977	 * Give the administrators a chance to notice that an application
978	 * might misbehave because it relies on the Linux behavior.
979	 */
980	pr_info_once("semctl(GETNCNT/GETZCNT) is since 3.16 Single Unix Specification compliant.\n"
981			"The task %s (%d) triggered the difference, watch for misbehavior.\n",
982			current->comm, task_pid_nr(current));
983
984	if (sop->sem_num != semnum)
985		return 0;
986
987	if (count_zero && sop->sem_op == 0)
988		return 1;
989	if (!count_zero && sop->sem_op < 0)
990		return 1;
991
992	return 0;
993}
994
995/* The following counts are associated to each semaphore:
996 *   semncnt        number of tasks waiting on semval being nonzero
997 *   semzcnt        number of tasks waiting on semval being zero
998 *
999 * Per definition, a task waits only on the semaphore of the first semop
1000 * that cannot proceed, even if additional operation would block, too.
1001 */
1002static int count_semcnt(struct sem_array *sma, ushort semnum,
1003			bool count_zero)
1004{
1005	struct list_head *l;
1006	struct sem_queue *q;
1007	int semcnt;
1008
1009	semcnt = 0;
1010	/* First: check the simple operations. They are easy to evaluate */
1011	if (count_zero)
1012		l = &sma->sem_base[semnum].pending_const;
1013	else
1014		l = &sma->sem_base[semnum].pending_alter;
1015
1016	list_for_each_entry(q, l, list) {
1017		/* all task on a per-semaphore list sleep on exactly
1018		 * that semaphore
1019		 */
1020		semcnt++;
1021	}
1022
1023	/* Then: check the complex operations. */
1024	list_for_each_entry(q, &sma->pending_alter, list) {
1025		semcnt += check_qop(sma, semnum, q, count_zero);
1026	}
1027	if (count_zero) {
1028		list_for_each_entry(q, &sma->pending_const, list) {
1029			semcnt += check_qop(sma, semnum, q, count_zero);
1030		}
1031	}
1032	return semcnt;
1033}
1034
1035/* Free a semaphore set. freeary() is called with sem_ids.rwsem locked
1036 * as a writer and the spinlock for this semaphore set hold. sem_ids.rwsem
1037 * remains locked on exit.
1038 */
1039static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
1040{
1041	struct sem_undo *un, *tu;
1042	struct sem_queue *q, *tq;
1043	struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
1044	int i;
1045	DEFINE_WAKE_Q(wake_q);
1046
1047	/* Free the existing undo structures for this semaphore set.  */
1048	ipc_assert_locked_object(&sma->sem_perm);
1049	list_for_each_entry_safe(un, tu, &sma->list_id, list_id) {
1050		list_del(&un->list_id);
1051		spin_lock(&un->ulp->lock);
1052		un->semid = -1;
1053		list_del_rcu(&un->list_proc);
1054		spin_unlock(&un->ulp->lock);
1055		kfree_rcu(un, rcu);
1056	}
1057
1058	/* Wake up all pending processes and let them fail with EIDRM. */
1059	list_for_each_entry_safe(q, tq, &sma->pending_const, list) {
1060		unlink_queue(sma, q);
1061		wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
1062	}
1063
1064	list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
1065		unlink_queue(sma, q);
1066		wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
1067	}
1068	for (i = 0; i < sma->sem_nsems; i++) {
1069		struct sem *sem = sma->sem_base + i;
1070		list_for_each_entry_safe(q, tq, &sem->pending_const, list) {
1071			unlink_queue(sma, q);
1072			wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
1073		}
1074		list_for_each_entry_safe(q, tq, &sem->pending_alter, list) {
1075			unlink_queue(sma, q);
1076			wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
1077		}
1078	}
1079
1080	/* Remove the semaphore set from the IDR */
1081	sem_rmid(ns, sma);
1082	sem_unlock(sma, -1);
1083	rcu_read_unlock();
1084
1085	wake_up_q(&wake_q);
1086	ns->used_sems -= sma->sem_nsems;
1087	ipc_rcu_putref(sma, sem_rcu_free);
1088}
1089
1090static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
1091{
1092	switch (version) {
1093	case IPC_64:
1094		return copy_to_user(buf, in, sizeof(*in));
1095	case IPC_OLD:
1096	    {
1097		struct semid_ds out;
1098
1099		memset(&out, 0, sizeof(out));
1100
1101		ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
1102
1103		out.sem_otime	= in->sem_otime;
1104		out.sem_ctime	= in->sem_ctime;
1105		out.sem_nsems	= in->sem_nsems;
1106
1107		return copy_to_user(buf, &out, sizeof(out));
1108	    }
1109	default:
1110		return -EINVAL;
1111	}
1112}
1113
1114static time_t get_semotime(struct sem_array *sma)
1115{
1116	int i;
1117	time_t res;
1118
1119	res = sma->sem_base[0].sem_otime;
1120	for (i = 1; i < sma->sem_nsems; i++) {
1121		time_t to = sma->sem_base[i].sem_otime;
1122
1123		if (to > res)
1124			res = to;
1125	}
1126	return res;
1127}
1128
1129static int semctl_nolock(struct ipc_namespace *ns, int semid,
1130			 int cmd, int version, void __user *p)
1131{
1132	int err;
1133	struct sem_array *sma;
1134
1135	switch (cmd) {
1136	case IPC_INFO:
1137	case SEM_INFO:
1138	{
1139		struct seminfo seminfo;
1140		int max_id;
1141
1142		err = security_sem_semctl(NULL, cmd);
1143		if (err)
1144			return err;
1145
1146		memset(&seminfo, 0, sizeof(seminfo));
1147		seminfo.semmni = ns->sc_semmni;
1148		seminfo.semmns = ns->sc_semmns;
1149		seminfo.semmsl = ns->sc_semmsl;
1150		seminfo.semopm = ns->sc_semopm;
1151		seminfo.semvmx = SEMVMX;
1152		seminfo.semmnu = SEMMNU;
1153		seminfo.semmap = SEMMAP;
1154		seminfo.semume = SEMUME;
1155		down_read(&sem_ids(ns).rwsem);
1156		if (cmd == SEM_INFO) {
1157			seminfo.semusz = sem_ids(ns).in_use;
1158			seminfo.semaem = ns->used_sems;
1159		} else {
1160			seminfo.semusz = SEMUSZ;
1161			seminfo.semaem = SEMAEM;
1162		}
1163		max_id = ipc_get_maxid(&sem_ids(ns));
1164		up_read(&sem_ids(ns).rwsem);
1165		if (copy_to_user(p, &seminfo, sizeof(struct seminfo)))
1166			return -EFAULT;
1167		return (max_id < 0) ? 0 : max_id;
1168	}
1169	case IPC_STAT:
1170	case SEM_STAT:
1171	{
1172		struct semid64_ds tbuf;
1173		int id = 0;
1174
1175		memset(&tbuf, 0, sizeof(tbuf));
1176
1177		rcu_read_lock();
1178		if (cmd == SEM_STAT) {
1179			sma = sem_obtain_object(ns, semid);
1180			if (IS_ERR(sma)) {
1181				err = PTR_ERR(sma);
1182				goto out_unlock;
1183			}
1184			id = sma->sem_perm.id;
1185		} else {
1186			sma = sem_obtain_object_check(ns, semid);
1187			if (IS_ERR(sma)) {
1188				err = PTR_ERR(sma);
1189				goto out_unlock;
1190			}
1191		}
1192
1193		err = -EACCES;
1194		if (ipcperms(ns, &sma->sem_perm, S_IRUGO))
1195			goto out_unlock;
1196
1197		err = security_sem_semctl(sma, cmd);
1198		if (err)
1199			goto out_unlock;
1200
1201		kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
1202		tbuf.sem_otime = get_semotime(sma);
1203		tbuf.sem_ctime = sma->sem_ctime;
1204		tbuf.sem_nsems = sma->sem_nsems;
1205		rcu_read_unlock();
1206		if (copy_semid_to_user(p, &tbuf, version))
1207			return -EFAULT;
1208		return id;
1209	}
1210	default:
1211		return -EINVAL;
1212	}
1213out_unlock:
1214	rcu_read_unlock();
1215	return err;
1216}
1217
1218static int semctl_setval(struct ipc_namespace *ns, int semid, int semnum,
1219		unsigned long arg)
1220{
1221	struct sem_undo *un;
1222	struct sem_array *sma;
1223	struct sem *curr;
1224	int err, val;
1225	DEFINE_WAKE_Q(wake_q);
1226
1227#if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN)
1228	/* big-endian 64bit */
1229	val = arg >> 32;
1230#else
1231	/* 32bit or little-endian 64bit */
1232	val = arg;
1233#endif
1234
1235	if (val > SEMVMX || val < 0)
1236		return -ERANGE;
1237
1238	rcu_read_lock();
1239	sma = sem_obtain_object_check(ns, semid);
1240	if (IS_ERR(sma)) {
1241		rcu_read_unlock();
1242		return PTR_ERR(sma);
1243	}
1244
1245	if (semnum < 0 || semnum >= sma->sem_nsems) {
1246		rcu_read_unlock();
1247		return -EINVAL;
1248	}
1249
1250
1251	if (ipcperms(ns, &sma->sem_perm, S_IWUGO)) {
1252		rcu_read_unlock();
1253		return -EACCES;
1254	}
1255
1256	err = security_sem_semctl(sma, SETVAL);
1257	if (err) {
1258		rcu_read_unlock();
1259		return -EACCES;
1260	}
1261
1262	sem_lock(sma, NULL, -1);
1263
1264	if (!ipc_valid_object(&sma->sem_perm)) {
1265		sem_unlock(sma, -1);
1266		rcu_read_unlock();
1267		return -EIDRM;
1268	}
1269
1270	curr = &sma->sem_base[semnum];
1271
1272	ipc_assert_locked_object(&sma->sem_perm);
1273	list_for_each_entry(un, &sma->list_id, list_id)
1274		un->semadj[semnum] = 0;
1275
1276	curr->semval = val;
1277	curr->sempid = task_tgid_vnr(current);
1278	sma->sem_ctime = get_seconds();
1279	/* maybe some queued-up processes were waiting for this */
1280	do_smart_update(sma, NULL, 0, 0, &wake_q);
1281	sem_unlock(sma, -1);
1282	rcu_read_unlock();
1283	wake_up_q(&wake_q);
1284	return 0;
1285}
1286
1287static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
1288		int cmd, void __user *p)
1289{
1290	struct sem_array *sma;
1291	struct sem *curr;
1292	int err, nsems;
1293	ushort fast_sem_io[SEMMSL_FAST];
1294	ushort *sem_io = fast_sem_io;
1295	DEFINE_WAKE_Q(wake_q);
1296
1297	rcu_read_lock();
1298	sma = sem_obtain_object_check(ns, semid);
1299	if (IS_ERR(sma)) {
1300		rcu_read_unlock();
1301		return PTR_ERR(sma);
1302	}
1303
1304	nsems = sma->sem_nsems;
1305
1306	err = -EACCES;
1307	if (ipcperms(ns, &sma->sem_perm, cmd == SETALL ? S_IWUGO : S_IRUGO))
1308		goto out_rcu_wakeup;
1309
1310	err = security_sem_semctl(sma, cmd);
1311	if (err)
1312		goto out_rcu_wakeup;
1313
1314	err = -EACCES;
1315	switch (cmd) {
1316	case GETALL:
1317	{
1318		ushort __user *array = p;
1319		int i;
1320
1321		sem_lock(sma, NULL, -1);
1322		if (!ipc_valid_object(&sma->sem_perm)) {
1323			err = -EIDRM;
1324			goto out_unlock;
1325		}
1326		if (nsems > SEMMSL_FAST) {
1327			if (!ipc_rcu_getref(sma)) {
1328				err = -EIDRM;
1329				goto out_unlock;
1330			}
1331			sem_unlock(sma, -1);
1332			rcu_read_unlock();
1333			sem_io = ipc_alloc(sizeof(ushort)*nsems);
1334			if (sem_io == NULL) {
1335				ipc_rcu_putref(sma, sem_rcu_free);
1336				return -ENOMEM;
1337			}
1338
1339			rcu_read_lock();
1340			sem_lock_and_putref(sma);
1341			if (!ipc_valid_object(&sma->sem_perm)) {
1342				err = -EIDRM;
1343				goto out_unlock;
1344			}
1345		}
1346		for (i = 0; i < sma->sem_nsems; i++)
1347			sem_io[i] = sma->sem_base[i].semval;
1348		sem_unlock(sma, -1);
1349		rcu_read_unlock();
1350		err = 0;
1351		if (copy_to_user(array, sem_io, nsems*sizeof(ushort)))
1352			err = -EFAULT;
1353		goto out_free;
1354	}
1355	case SETALL:
1356	{
1357		int i;
1358		struct sem_undo *un;
1359
1360		if (!ipc_rcu_getref(sma)) {
1361			err = -EIDRM;
1362			goto out_rcu_wakeup;
1363		}
1364		rcu_read_unlock();
1365
1366		if (nsems > SEMMSL_FAST) {
1367			sem_io = ipc_alloc(sizeof(ushort)*nsems);
1368			if (sem_io == NULL) {
1369				ipc_rcu_putref(sma, sem_rcu_free);
1370				return -ENOMEM;
1371			}
1372		}
1373
1374		if (copy_from_user(sem_io, p, nsems*sizeof(ushort))) {
1375			ipc_rcu_putref(sma, sem_rcu_free);
1376			err = -EFAULT;
1377			goto out_free;
1378		}
1379
1380		for (i = 0; i < nsems; i++) {
1381			if (sem_io[i] > SEMVMX) {
1382				ipc_rcu_putref(sma, sem_rcu_free);
1383				err = -ERANGE;
1384				goto out_free;
1385			}
1386		}
1387		rcu_read_lock();
1388		sem_lock_and_putref(sma);
1389		if (!ipc_valid_object(&sma->sem_perm)) {
1390			err = -EIDRM;
1391			goto out_unlock;
1392		}
1393
1394		for (i = 0; i < nsems; i++) {
1395			sma->sem_base[i].semval = sem_io[i];
1396			sma->sem_base[i].sempid = task_tgid_vnr(current);
1397		}
1398
1399		ipc_assert_locked_object(&sma->sem_perm);
1400		list_for_each_entry(un, &sma->list_id, list_id) {
1401			for (i = 0; i < nsems; i++)
1402				un->semadj[i] = 0;
1403		}
1404		sma->sem_ctime = get_seconds();
1405		/* maybe some queued-up processes were waiting for this */
1406		do_smart_update(sma, NULL, 0, 0, &wake_q);
1407		err = 0;
1408		goto out_unlock;
1409	}
1410	/* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */
1411	}
1412	err = -EINVAL;
1413	if (semnum < 0 || semnum >= nsems)
1414		goto out_rcu_wakeup;
1415
1416	sem_lock(sma, NULL, -1);
1417	if (!ipc_valid_object(&sma->sem_perm)) {
1418		err = -EIDRM;
1419		goto out_unlock;
1420	}
1421	curr = &sma->sem_base[semnum];
1422
1423	switch (cmd) {
1424	case GETVAL:
1425		err = curr->semval;
1426		goto out_unlock;
1427	case GETPID:
1428		err = curr->sempid;
1429		goto out_unlock;
1430	case GETNCNT:
1431		err = count_semcnt(sma, semnum, 0);
1432		goto out_unlock;
1433	case GETZCNT:
1434		err = count_semcnt(sma, semnum, 1);
1435		goto out_unlock;
1436	}
1437
1438out_unlock:
1439	sem_unlock(sma, -1);
1440out_rcu_wakeup:
1441	rcu_read_unlock();
1442	wake_up_q(&wake_q);
1443out_free:
1444	if (sem_io != fast_sem_io)
1445		ipc_free(sem_io);
1446	return err;
1447}
1448
1449static inline unsigned long
1450copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
1451{
1452	switch (version) {
1453	case IPC_64:
1454		if (copy_from_user(out, buf, sizeof(*out)))
1455			return -EFAULT;
1456		return 0;
1457	case IPC_OLD:
1458	    {
1459		struct semid_ds tbuf_old;
1460
1461		if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
1462			return -EFAULT;
1463
1464		out->sem_perm.uid	= tbuf_old.sem_perm.uid;
1465		out->sem_perm.gid	= tbuf_old.sem_perm.gid;
1466		out->sem_perm.mode	= tbuf_old.sem_perm.mode;
1467
1468		return 0;
1469	    }
1470	default:
1471		return -EINVAL;
1472	}
1473}
1474
1475/*
1476 * This function handles some semctl commands which require the rwsem
1477 * to be held in write mode.
1478 * NOTE: no locks must be held, the rwsem is taken inside this function.
1479 */
1480static int semctl_down(struct ipc_namespace *ns, int semid,
1481		       int cmd, int version, void __user *p)
1482{
1483	struct sem_array *sma;
1484	int err;
1485	struct semid64_ds semid64;
1486	struct kern_ipc_perm *ipcp;
1487
1488	if (cmd == IPC_SET) {
1489		if (copy_semid_from_user(&semid64, p, version))
1490			return -EFAULT;
1491	}
1492
1493	down_write(&sem_ids(ns).rwsem);
1494	rcu_read_lock();
1495
1496	ipcp = ipcctl_pre_down_nolock(ns, &sem_ids(ns), semid, cmd,
1497				      &semid64.sem_perm, 0);
1498	if (IS_ERR(ipcp)) {
1499		err = PTR_ERR(ipcp);
1500		goto out_unlock1;
1501	}
1502
1503	sma = container_of(ipcp, struct sem_array, sem_perm);
1504
1505	err = security_sem_semctl(sma, cmd);
1506	if (err)
1507		goto out_unlock1;
1508
1509	switch (cmd) {
1510	case IPC_RMID:
1511		sem_lock(sma, NULL, -1);
1512		/* freeary unlocks the ipc object and rcu */
1513		freeary(ns, ipcp);
1514		goto out_up;
1515	case IPC_SET:
1516		sem_lock(sma, NULL, -1);
1517		err = ipc_update_perm(&semid64.sem_perm, ipcp);
1518		if (err)
1519			goto out_unlock0;
1520		sma->sem_ctime = get_seconds();
1521		break;
1522	default:
1523		err = -EINVAL;
1524		goto out_unlock1;
1525	}
1526
1527out_unlock0:
1528	sem_unlock(sma, -1);
1529out_unlock1:
1530	rcu_read_unlock();
1531out_up:
1532	up_write(&sem_ids(ns).rwsem);
1533	return err;
1534}
1535
1536SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, unsigned long, arg)
1537{
1538	int version;
1539	struct ipc_namespace *ns;
1540	void __user *p = (void __user *)arg;
1541
1542	if (semid < 0)
1543		return -EINVAL;
1544
1545	version = ipc_parse_version(&cmd);
1546	ns = current->nsproxy->ipc_ns;
1547
1548	switch (cmd) {
1549	case IPC_INFO:
1550	case SEM_INFO:
1551	case IPC_STAT:
1552	case SEM_STAT:
1553		return semctl_nolock(ns, semid, cmd, version, p);
1554	case GETALL:
1555	case GETVAL:
1556	case GETPID:
1557	case GETNCNT:
1558	case GETZCNT:
1559	case SETALL:
1560		return semctl_main(ns, semid, semnum, cmd, p);
1561	case SETVAL:
1562		return semctl_setval(ns, semid, semnum, arg);
1563	case IPC_RMID:
1564	case IPC_SET:
1565		return semctl_down(ns, semid, cmd, version, p);
1566	default:
1567		return -EINVAL;
1568	}
1569}
1570
1571/* If the task doesn't already have a undo_list, then allocate one
1572 * here.  We guarantee there is only one thread using this undo list,
1573 * and current is THE ONE
1574 *
1575 * If this allocation and assignment succeeds, but later
1576 * portions of this code fail, there is no need to free the sem_undo_list.
1577 * Just let it stay associated with the task, and it'll be freed later
1578 * at exit time.
1579 *
1580 * This can block, so callers must hold no locks.
1581 */
1582static inline int get_undo_list(struct sem_undo_list **undo_listp)
1583{
1584	struct sem_undo_list *undo_list;
1585
1586	undo_list = current->sysvsem.undo_list;
1587	if (!undo_list) {
1588		undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
1589		if (undo_list == NULL)
1590			return -ENOMEM;
1591		spin_lock_init(&undo_list->lock);
1592		atomic_set(&undo_list->refcnt, 1);
1593		INIT_LIST_HEAD(&undo_list->list_proc);
1594
1595		current->sysvsem.undo_list = undo_list;
1596	}
1597	*undo_listp = undo_list;
1598	return 0;
1599}
1600
1601static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid)
1602{
1603	struct sem_undo *un;
1604
1605	list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) {
1606		if (un->semid == semid)
1607			return un;
1608	}
1609	return NULL;
1610}
1611
1612static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1613{
1614	struct sem_undo *un;
1615
1616	assert_spin_locked(&ulp->lock);
1617
1618	un = __lookup_undo(ulp, semid);
1619	if (un) {
1620		list_del_rcu(&un->list_proc);
1621		list_add_rcu(&un->list_proc, &ulp->list_proc);
1622	}
1623	return un;
1624}
1625
1626/**
1627 * find_alloc_undo - lookup (and if not present create) undo array
1628 * @ns: namespace
1629 * @semid: semaphore array id
1630 *
1631 * The function looks up (and if not present creates) the undo structure.
1632 * The size of the undo structure depends on the size of the semaphore
1633 * array, thus the alloc path is not that straightforward.
1634 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1635 * performs a rcu_read_lock().
1636 */
1637static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
1638{
1639	struct sem_array *sma;
1640	struct sem_undo_list *ulp;
1641	struct sem_undo *un, *new;
1642	int nsems, error;
1643
1644	error = get_undo_list(&ulp);
1645	if (error)
1646		return ERR_PTR(error);
1647
1648	rcu_read_lock();
1649	spin_lock(&ulp->lock);
1650	un = lookup_undo(ulp, semid);
1651	spin_unlock(&ulp->lock);
1652	if (likely(un != NULL))
1653		goto out;
1654
1655	/* no undo structure around - allocate one. */
1656	/* step 1: figure out the size of the semaphore array */
1657	sma = sem_obtain_object_check(ns, semid);
1658	if (IS_ERR(sma)) {
1659		rcu_read_unlock();
1660		return ERR_CAST(sma);
1661	}
1662
1663	nsems = sma->sem_nsems;
1664	if (!ipc_rcu_getref(sma)) {
1665		rcu_read_unlock();
1666		un = ERR_PTR(-EIDRM);
1667		goto out;
1668	}
1669	rcu_read_unlock();
1670
1671	/* step 2: allocate new undo structure */
1672	new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1673	if (!new) {
1674		ipc_rcu_putref(sma, sem_rcu_free);
1675		return ERR_PTR(-ENOMEM);
1676	}
1677
1678	/* step 3: Acquire the lock on semaphore array */
1679	rcu_read_lock();
1680	sem_lock_and_putref(sma);
1681	if (!ipc_valid_object(&sma->sem_perm)) {
1682		sem_unlock(sma, -1);
1683		rcu_read_unlock();
1684		kfree(new);
1685		un = ERR_PTR(-EIDRM);
1686		goto out;
1687	}
1688	spin_lock(&ulp->lock);
1689
1690	/*
1691	 * step 4: check for races: did someone else allocate the undo struct?
1692	 */
1693	un = lookup_undo(ulp, semid);
1694	if (un) {
1695		kfree(new);
1696		goto success;
1697	}
1698	/* step 5: initialize & link new undo structure */
1699	new->semadj = (short *) &new[1];
1700	new->ulp = ulp;
1701	new->semid = semid;
1702	assert_spin_locked(&ulp->lock);
1703	list_add_rcu(&new->list_proc, &ulp->list_proc);
1704	ipc_assert_locked_object(&sma->sem_perm);
1705	list_add(&new->list_id, &sma->list_id);
1706	un = new;
1707
1708success:
1709	spin_unlock(&ulp->lock);
1710	sem_unlock(sma, -1);
1711out:
1712	return un;
1713}
1714
1715SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
1716		unsigned, nsops, const struct timespec __user *, timeout)
1717{
1718	int error = -EINVAL;
1719	struct sem_array *sma;
1720	struct sembuf fast_sops[SEMOPM_FAST];
1721	struct sembuf *sops = fast_sops, *sop;
1722	struct sem_undo *un;
1723	int undos = 0, alter = 0, max, locknum;
1724	struct sem_queue queue;
1725	unsigned long jiffies_left = 0;
1726	struct ipc_namespace *ns;
1727
1728	ns = current->nsproxy->ipc_ns;
1729
1730	if (nsops < 1 || semid < 0)
1731		return -EINVAL;
1732	if (nsops > ns->sc_semopm)
1733		return -E2BIG;
1734	if (nsops > SEMOPM_FAST) {
1735		sops = kmalloc(sizeof(*sops)*nsops, GFP_KERNEL);
1736		if (sops == NULL)
1737			return -ENOMEM;
1738	}
1739	if (copy_from_user(sops, tsops, nsops * sizeof(*tsops))) {
1740		error =  -EFAULT;
1741		goto out_free;
1742	}
1743	if (timeout) {
1744		struct timespec _timeout;
1745		if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1746			error = -EFAULT;
1747			goto out_free;
1748		}
1749		if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1750			_timeout.tv_nsec >= 1000000000L) {
1751			error = -EINVAL;
1752			goto out_free;
1753		}
1754		jiffies_left = timespec_to_jiffies(&_timeout);
1755	}
1756	max = 0;
1757	for (sop = sops; sop < sops + nsops; sop++) {
1758		if (sop->sem_num >= max)
1759			max = sop->sem_num;
1760		if (sop->sem_flg & SEM_UNDO)
1761			undos = 1;
1762		if (sop->sem_op != 0)
1763			alter = 1;
1764	}
1765
1766
1767	if (undos) {
1768		/* On success, find_alloc_undo takes the rcu_read_lock */
1769		un = find_alloc_undo(ns, semid);
1770		if (IS_ERR(un)) {
1771			error = PTR_ERR(un);
1772			goto out_free;
1773		}
1774	} else {
1775		un = NULL;
1776		rcu_read_lock();
1777	}
1778
1779	sma = sem_obtain_object_check(ns, semid);
1780	if (IS_ERR(sma)) {
1781		rcu_read_unlock();
1782		error = PTR_ERR(sma);
1783		goto out_free;
1784	}
1785
1786	error = -EFBIG;
1787	if (max >= sma->sem_nsems) {
1788		rcu_read_unlock();
1789		goto out_free;
1790	}
1791
1792	error = -EACCES;
1793	if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO)) {
1794		rcu_read_unlock();
1795		goto out_free;
1796	}
1797
1798	error = security_sem_semop(sma, sops, nsops, alter);
1799	if (error) {
1800		rcu_read_unlock();
1801		goto out_free;
1802	}
1803
1804	error = -EIDRM;
1805	locknum = sem_lock(sma, sops, nsops);
1806	/*
1807	 * We eventually might perform the following check in a lockless
1808	 * fashion, considering ipc_valid_object() locking constraints.
1809	 * If nsops == 1 and there is no contention for sem_perm.lock, then
1810	 * only a per-semaphore lock is held and it's OK to proceed with the
1811	 * check below. More details on the fine grained locking scheme
1812	 * entangled here and why it's RMID race safe on comments at sem_lock()
1813	 */
1814	if (!ipc_valid_object(&sma->sem_perm))
1815		goto out_unlock_free;
1816	/*
1817	 * semid identifiers are not unique - find_alloc_undo may have
1818	 * allocated an undo structure, it was invalidated by an RMID
1819	 * and now a new array with received the same id. Check and fail.
1820	 * This case can be detected checking un->semid. The existence of
1821	 * "un" itself is guaranteed by rcu.
1822	 */
1823	if (un && un->semid == -1)
1824		goto out_unlock_free;
1825
1826	queue.sops = sops;
1827	queue.nsops = nsops;
1828	queue.undo = un;
1829	queue.pid = task_tgid_vnr(current);
1830	queue.alter = alter;
1831
1832	error = perform_atomic_semop(sma, &queue);
1833	if (error == 0) { /* non-blocking succesfull path */
1834		DEFINE_WAKE_Q(wake_q);
1835
1836		/*
1837		 * If the operation was successful, then do
1838		 * the required updates.
1839		 */
1840		if (alter)
1841			do_smart_update(sma, sops, nsops, 1, &wake_q);
1842		else
1843			set_semotime(sma, sops);
1844
1845		sem_unlock(sma, locknum);
1846		rcu_read_unlock();
1847		wake_up_q(&wake_q);
1848
1849		goto out_free;
1850	}
1851	if (error < 0) /* non-blocking error path */
1852		goto out_unlock_free;
1853
1854	/*
1855	 * We need to sleep on this operation, so we put the current
1856	 * task into the pending queue and go to sleep.
1857	 */
1858	if (nsops == 1) {
1859		struct sem *curr;
1860		curr = &sma->sem_base[sops->sem_num];
1861
1862		if (alter) {
1863			if (sma->complex_count) {
1864				list_add_tail(&queue.list,
1865						&sma->pending_alter);
1866			} else {
1867
1868				list_add_tail(&queue.list,
1869						&curr->pending_alter);
1870			}
1871		} else {
1872			list_add_tail(&queue.list, &curr->pending_const);
1873		}
1874	} else {
1875		if (!sma->complex_count)
1876			merge_queues(sma);
1877
1878		if (alter)
1879			list_add_tail(&queue.list, &sma->pending_alter);
1880		else
1881			list_add_tail(&queue.list, &sma->pending_const);
1882
1883		sma->complex_count++;
1884	}
1885
1886sleep_again:
1887	queue.status = -EINTR;
1888	queue.sleeper = current;
1889
1890	__set_current_state(TASK_INTERRUPTIBLE);
1891	sem_unlock(sma, locknum);
1892	rcu_read_unlock();
1893
1894	if (timeout)
1895		jiffies_left = schedule_timeout(jiffies_left);
1896	else
1897		schedule();
1898
1899	/*
1900	 * fastpath: the semop has completed, either successfully or not, from
1901	 * the syscall pov, is quite irrelevant to us at this point; we're done.
1902	 *
1903	 * We _do_ care, nonetheless, about being awoken by a signal or
1904	 * spuriously.  The queue.status is checked again in the slowpath (aka
1905	 * after taking sem_lock), such that we can detect scenarios where we
1906	 * were awakened externally, during the window between wake_q_add() and
1907	 * wake_up_q().
1908	 */
1909	error = READ_ONCE(queue.status);
1910	if (error != -EINTR) {
1911		/*
1912		 * User space could assume that semop() is a memory barrier:
1913		 * Without the mb(), the cpu could speculatively read in user
1914		 * space stale data that was overwritten by the previous owner
1915		 * of the semaphore.
1916		 */
1917		smp_mb();
1918		goto out_free;
1919	}
1920
1921	rcu_read_lock();
1922	sma = sem_obtain_lock(ns, semid, sops, nsops, &locknum);
1923	error = READ_ONCE(queue.status);
1924
1925	/*
1926	 * Array removed? If yes, leave without sem_unlock().
1927	 */
1928	if (IS_ERR(sma)) {
1929		rcu_read_unlock();
1930		goto out_free;
1931	}
1932
1933	/*
1934	 * If queue.status != -EINTR we are woken up by another process.
1935	 * Leave without unlink_queue(), but with sem_unlock().
1936	 */
1937	if (error != -EINTR)
1938		goto out_unlock_free;
1939
1940	/*
1941	 * If an interrupt occurred we have to clean up the queue.
1942	 */
1943	if (timeout && jiffies_left == 0)
1944		error = -EAGAIN;
1945
1946	/*
1947	 * If the wakeup was spurious, just retry.
1948	 */
1949	if (error == -EINTR && !signal_pending(current))
1950		goto sleep_again;
1951
1952	unlink_queue(sma, &queue);
1953
1954out_unlock_free:
1955	sem_unlock(sma, locknum);
1956	rcu_read_unlock();
1957out_free:
1958	if (sops != fast_sops)
1959		kfree(sops);
1960	return error;
1961}
1962
1963SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
1964		unsigned, nsops)
1965{
1966	return sys_semtimedop(semid, tsops, nsops, NULL);
1967}
1968
1969/* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
1970 * parent and child tasks.
1971 */
1972
1973int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
1974{
1975	struct sem_undo_list *undo_list;
1976	int error;
1977
1978	if (clone_flags & CLONE_SYSVSEM) {
1979		error = get_undo_list(&undo_list);
1980		if (error)
1981			return error;
1982		atomic_inc(&undo_list->refcnt);
1983		tsk->sysvsem.undo_list = undo_list;
1984	} else
1985		tsk->sysvsem.undo_list = NULL;
1986
1987	return 0;
1988}
1989
1990/*
1991 * add semadj values to semaphores, free undo structures.
1992 * undo structures are not freed when semaphore arrays are destroyed
1993 * so some of them may be out of date.
1994 * IMPLEMENTATION NOTE: There is some confusion over whether the
1995 * set of adjustments that needs to be done should be done in an atomic
1996 * manner or not. That is, if we are attempting to decrement the semval
1997 * should we queue up and wait until we can do so legally?
1998 * The original implementation attempted to do this (queue and wait).
1999 * The current implementation does not do so. The POSIX standard
2000 * and SVID should be consulted to determine what behavior is mandated.
2001 */
2002void exit_sem(struct task_struct *tsk)
2003{
2004	struct sem_undo_list *ulp;
2005
2006	ulp = tsk->sysvsem.undo_list;
2007	if (!ulp)
2008		return;
2009	tsk->sysvsem.undo_list = NULL;
2010
2011	if (!atomic_dec_and_test(&ulp->refcnt))
2012		return;
2013
2014	for (;;) {
2015		struct sem_array *sma;
2016		struct sem_undo *un;
2017		int semid, i;
2018		DEFINE_WAKE_Q(wake_q);
2019
2020		cond_resched();
2021
2022		rcu_read_lock();
2023		un = list_entry_rcu(ulp->list_proc.next,
2024				    struct sem_undo, list_proc);
2025		if (&un->list_proc == &ulp->list_proc) {
2026			/*
2027			 * We must wait for freeary() before freeing this ulp,
2028			 * in case we raced with last sem_undo. There is a small
2029			 * possibility where we exit while freeary() didn't
2030			 * finish unlocking sem_undo_list.
2031			 */
2032			spin_unlock_wait(&ulp->lock);
2033			rcu_read_unlock();
2034			break;
2035		}
2036		spin_lock(&ulp->lock);
2037		semid = un->semid;
2038		spin_unlock(&ulp->lock);
2039
2040		/* exit_sem raced with IPC_RMID, nothing to do */
2041		if (semid == -1) {
2042			rcu_read_unlock();
2043			continue;
2044		}
2045
2046		sma = sem_obtain_object_check(tsk->nsproxy->ipc_ns, semid);
2047		/* exit_sem raced with IPC_RMID, nothing to do */
2048		if (IS_ERR(sma)) {
2049			rcu_read_unlock();
2050			continue;
2051		}
2052
2053		sem_lock(sma, NULL, -1);
2054		/* exit_sem raced with IPC_RMID, nothing to do */
2055		if (!ipc_valid_object(&sma->sem_perm)) {
2056			sem_unlock(sma, -1);
2057			rcu_read_unlock();
2058			continue;
2059		}
2060		un = __lookup_undo(ulp, semid);
2061		if (un == NULL) {
2062			/* exit_sem raced with IPC_RMID+semget() that created
2063			 * exactly the same semid. Nothing to do.
2064			 */
2065			sem_unlock(sma, -1);
2066			rcu_read_unlock();
2067			continue;
2068		}
2069
2070		/* remove un from the linked lists */
2071		ipc_assert_locked_object(&sma->sem_perm);
2072		list_del(&un->list_id);
2073
2074		/* we are the last process using this ulp, acquiring ulp->lock
2075		 * isn't required. Besides that, we are also protected against
2076		 * IPC_RMID as we hold sma->sem_perm lock now
2077		 */
2078		list_del_rcu(&un->list_proc);
2079
2080		/* perform adjustments registered in un */
2081		for (i = 0; i < sma->sem_nsems; i++) {
2082			struct sem *semaphore = &sma->sem_base[i];
2083			if (un->semadj[i]) {
2084				semaphore->semval += un->semadj[i];
2085				/*
2086				 * Range checks of the new semaphore value,
2087				 * not defined by sus:
2088				 * - Some unices ignore the undo entirely
2089				 *   (e.g. HP UX 11i 11.22, Tru64 V5.1)
2090				 * - some cap the value (e.g. FreeBSD caps
2091				 *   at 0, but doesn't enforce SEMVMX)
2092				 *
2093				 * Linux caps the semaphore value, both at 0
2094				 * and at SEMVMX.
2095				 *
2096				 *	Manfred <manfred@colorfullife.com>
2097				 */
2098				if (semaphore->semval < 0)
2099					semaphore->semval = 0;
2100				if (semaphore->semval > SEMVMX)
2101					semaphore->semval = SEMVMX;
2102				semaphore->sempid = task_tgid_vnr(current);
2103			}
2104		}
2105		/* maybe some queued-up processes were waiting for this */
2106		do_smart_update(sma, NULL, 0, 1, &wake_q);
2107		sem_unlock(sma, -1);
2108		rcu_read_unlock();
2109		wake_up_q(&wake_q);
2110
2111		kfree_rcu(un, rcu);
2112	}
2113	kfree(ulp);
2114}
2115
2116#ifdef CONFIG_PROC_FS
2117static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
2118{
2119	struct user_namespace *user_ns = seq_user_ns(s);
2120	struct sem_array *sma = it;
2121	time_t sem_otime;
2122
2123	/*
2124	 * The proc interface isn't aware of sem_lock(), it calls
2125	 * ipc_lock_object() directly (in sysvipc_find_ipc).
2126	 * In order to stay compatible with sem_lock(), we must
2127	 * enter / leave complex_mode.
2128	 */
2129	complexmode_enter(sma);
2130
2131	sem_otime = get_semotime(sma);
2132
2133	seq_printf(s,
2134		   "%10d %10d  %4o %10u %5u %5u %5u %5u %10lu %10lu\n",
2135		   sma->sem_perm.key,
2136		   sma->sem_perm.id,
2137		   sma->sem_perm.mode,
2138		   sma->sem_nsems,
2139		   from_kuid_munged(user_ns, sma->sem_perm.uid),
2140		   from_kgid_munged(user_ns, sma->sem_perm.gid),
2141		   from_kuid_munged(user_ns, sma->sem_perm.cuid),
2142		   from_kgid_munged(user_ns, sma->sem_perm.cgid),
2143		   sem_otime,
2144		   sma->sem_ctime);
2145
2146	complexmode_tryleave(sma);
2147
2148	return 0;
2149}
2150#endif
2151