sem.c revision b1989a3d
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/ipc/sem.c
4 * Copyright (C) 1992 Krishna Balasubramanian
5 * Copyright (C) 1995 Eric Schenk, Bruno Haible
6 *
7 * /proc/sysvipc/sem support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
8 *
9 * SMP-threaded, sysctl's added
10 * (c) 1999 Manfred Spraul <manfred@colorfullife.com>
11 * Enforced range limit on SEM_UNDO
12 * (c) 2001 Red Hat Inc
13 * Lockless wakeup
14 * (c) 2003 Manfred Spraul <manfred@colorfullife.com>
15 * (c) 2016 Davidlohr Bueso <dave@stgolabs.net>
16 * Further wakeup optimizations, documentation
17 * (c) 2010 Manfred Spraul <manfred@colorfullife.com>
18 *
19 * support for audit of ipc object properties and permission changes
20 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
21 *
22 * namespaces support
23 * OpenVZ, SWsoft Inc.
24 * Pavel Emelianov <xemul@openvz.org>
25 *
26 * Implementation notes: (May 2010)
27 * This file implements System V semaphores.
28 *
29 * User space visible behavior:
30 * - FIFO ordering for semop() operations (just FIFO, not starvation
31 *   protection)
32 * - multiple semaphore operations that alter the same semaphore in
33 *   one semop() are handled.
34 * - sem_ctime (time of last semctl()) is updated in the IPC_SET, SETVAL and
35 *   SETALL calls.
36 * - two Linux specific semctl() commands: SEM_STAT, SEM_INFO.
37 * - undo adjustments at process exit are limited to 0..SEMVMX.
38 * - namespace are supported.
39 * - SEMMSL, SEMMNS, SEMOPM and SEMMNI can be configured at runtime by writing
40 *   to /proc/sys/kernel/sem.
41 * - statistics about the usage are reported in /proc/sysvipc/sem.
42 *
43 * Internals:
44 * - scalability:
45 *   - all global variables are read-mostly.
46 *   - semop() calls and semctl(RMID) are synchronized by RCU.
47 *   - most operations do write operations (actually: spin_lock calls) to
48 *     the per-semaphore array structure.
49 *   Thus: Perfect SMP scaling between independent semaphore arrays.
50 *         If multiple semaphores in one array are used, then cache line
51 *         trashing on the semaphore array spinlock will limit the scaling.
52 * - semncnt and semzcnt are calculated on demand in count_semcnt()
53 * - the task that performs a successful semop() scans the list of all
54 *   sleeping tasks and completes any pending operations that can be fulfilled.
55 *   Semaphores are actively given to waiting tasks (necessary for FIFO).
56 *   (see update_queue())
57 * - To improve the scalability, the actual wake-up calls are performed after
58 *   dropping all locks. (see wake_up_sem_queue_prepare())
59 * - All work is done by the waker, the woken up task does not have to do
60 *   anything - not even acquiring a lock or dropping a refcount.
61 * - A woken up task may not even touch the semaphore array anymore, it may
62 *   have been destroyed already by a semctl(RMID).
63 * - UNDO values are stored in an array (one per process and per
64 *   semaphore array, lazily allocated). For backwards compatibility, multiple
65 *   modes for the UNDO variables are supported (per process, per thread)
66 *   (see copy_semundo, CLONE_SYSVSEM)
67 * - There are two lists of the pending operations: a per-array list
68 *   and per-semaphore list (stored in the array). This allows to achieve FIFO
69 *   ordering without always scanning all pending operations.
70 *   The worst-case behavior is nevertheless O(N^2) for N wakeups.
71 */
72
73#include <linux/compat.h>
74#include <linux/slab.h>
75#include <linux/spinlock.h>
76#include <linux/init.h>
77#include <linux/proc_fs.h>
78#include <linux/time.h>
79#include <linux/security.h>
80#include <linux/syscalls.h>
81#include <linux/audit.h>
82#include <linux/capability.h>
83#include <linux/seq_file.h>
84#include <linux/rwsem.h>
85#include <linux/nsproxy.h>
86#include <linux/ipc_namespace.h>
87#include <linux/sched/wake_q.h>
88#include <linux/nospec.h>
89#include <linux/rhashtable.h>
90
91#include <linux/uaccess.h>
92#include "util.h"
93
94/* One semaphore structure for each semaphore in the system. */
95struct sem {
96	int	semval;		/* current value */
97	/*
98	 * PID of the process that last modified the semaphore. For
99	 * Linux, specifically these are:
100	 *  - semop
101	 *  - semctl, via SETVAL and SETALL.
102	 *  - at task exit when performing undo adjustments (see exit_sem).
103	 */
104	struct pid *sempid;
105	spinlock_t	lock;	/* spinlock for fine-grained semtimedop */
106	struct list_head pending_alter; /* pending single-sop operations */
107					/* that alter the semaphore */
108	struct list_head pending_const; /* pending single-sop operations */
109					/* that do not alter the semaphore*/
110	time64_t	 sem_otime;	/* candidate for sem_otime */
111} ____cacheline_aligned_in_smp;
112
113/* One sem_array data structure for each set of semaphores in the system. */
114struct sem_array {
115	struct kern_ipc_perm	sem_perm;	/* permissions .. see ipc.h */
116	time64_t		sem_ctime;	/* create/last semctl() time */
117	struct list_head	pending_alter;	/* pending operations */
118						/* that alter the array */
119	struct list_head	pending_const;	/* pending complex operations */
120						/* that do not alter semvals */
121	struct list_head	list_id;	/* undo requests on this array */
122	int			sem_nsems;	/* no. of semaphores in array */
123	int			complex_count;	/* pending complex operations */
124	unsigned int		use_global_lock;/* >0: global lock required */
125
126	struct sem		sems[];
127} __randomize_layout;
128
129/* One queue for each sleeping process in the system. */
130struct sem_queue {
131	struct list_head	list;	 /* queue of pending operations */
132	struct task_struct	*sleeper; /* this process */
133	struct sem_undo		*undo;	 /* undo structure */
134	struct pid		*pid;	 /* process id of requesting process */
135	int			status;	 /* completion status of operation */
136	struct sembuf		*sops;	 /* array of pending operations */
137	struct sembuf		*blocking; /* the operation that blocked */
138	int			nsops;	 /* number of operations */
139	bool			alter;	 /* does *sops alter the array? */
140	bool                    dupsop;	 /* sops on more than one sem_num */
141};
142
143/* Each task has a list of undo requests. They are executed automatically
144 * when the process exits.
145 */
146struct sem_undo {
147	struct list_head	list_proc;	/* per-process list: *
148						 * all undos from one process
149						 * rcu protected */
150	struct rcu_head		rcu;		/* rcu struct for sem_undo */
151	struct sem_undo_list	*ulp;		/* back ptr to sem_undo_list */
152	struct list_head	list_id;	/* per semaphore array list:
153						 * all undos for one array */
154	int			semid;		/* semaphore set identifier */
155	short			*semadj;	/* array of adjustments */
156						/* one per semaphore */
157};
158
159/* sem_undo_list controls shared access to the list of sem_undo structures
160 * that may be shared among all a CLONE_SYSVSEM task group.
161 */
162struct sem_undo_list {
163	refcount_t		refcnt;
164	spinlock_t		lock;
165	struct list_head	list_proc;
166};
167
168
169#define sem_ids(ns)	((ns)->ids[IPC_SEM_IDS])
170
171static int newary(struct ipc_namespace *, struct ipc_params *);
172static void freeary(struct ipc_namespace *, struct kern_ipc_perm *);
173#ifdef CONFIG_PROC_FS
174static int sysvipc_sem_proc_show(struct seq_file *s, void *it);
175#endif
176
177#define SEMMSL_FAST	256 /* 512 bytes on stack */
178#define SEMOPM_FAST	64  /* ~ 372 bytes on stack */
179
180/*
181 * Switching from the mode suitable for simple ops
182 * to the mode for complex ops is costly. Therefore:
183 * use some hysteresis
184 */
185#define USE_GLOBAL_LOCK_HYSTERESIS	10
186
187/*
188 * Locking:
189 * a) global sem_lock() for read/write
190 *	sem_undo.id_next,
191 *	sem_array.complex_count,
192 *	sem_array.pending{_alter,_const},
193 *	sem_array.sem_undo
194 *
195 * b) global or semaphore sem_lock() for read/write:
196 *	sem_array.sems[i].pending_{const,alter}:
197 *
198 * c) special:
199 *	sem_undo_list.list_proc:
200 *	* undo_list->lock for write
201 *	* rcu for read
202 *	use_global_lock:
203 *	* global sem_lock() for write
204 *	* either local or global sem_lock() for read.
205 *
206 * Memory ordering:
207 * Most ordering is enforced by using spin_lock() and spin_unlock().
208 *
209 * Exceptions:
210 * 1) use_global_lock: (SEM_BARRIER_1)
211 * Setting it from non-zero to 0 is a RELEASE, this is ensured by
212 * using smp_store_release(): Immediately after setting it to 0,
213 * a simple op can start.
214 * Testing if it is non-zero is an ACQUIRE, this is ensured by using
215 * smp_load_acquire().
216 * Setting it from 0 to non-zero must be ordered with regards to
217 * this smp_load_acquire(), this is guaranteed because the smp_load_acquire()
218 * is inside a spin_lock() and after a write from 0 to non-zero a
219 * spin_lock()+spin_unlock() is done.
220 *
221 * 2) queue.status: (SEM_BARRIER_2)
222 * Initialization is done while holding sem_lock(), so no further barrier is
223 * required.
224 * Setting it to a result code is a RELEASE, this is ensured by both a
225 * smp_store_release() (for case a) and while holding sem_lock()
226 * (for case b).
227 * The ACQUIRE when reading the result code without holding sem_lock() is
228 * achieved by using READ_ONCE() + smp_acquire__after_ctrl_dep().
229 * (case a above).
230 * Reading the result code while holding sem_lock() needs no further barriers,
231 * the locks inside sem_lock() enforce ordering (case b above)
232 *
233 * 3) current->state:
234 * current->state is set to TASK_INTERRUPTIBLE while holding sem_lock().
235 * The wakeup is handled using the wake_q infrastructure. wake_q wakeups may
236 * happen immediately after calling wake_q_add. As wake_q_add_safe() is called
237 * when holding sem_lock(), no further barriers are required.
238 *
239 * See also ipc/mqueue.c for more details on the covered races.
240 */
241
242#define sc_semmsl	sem_ctls[0]
243#define sc_semmns	sem_ctls[1]
244#define sc_semopm	sem_ctls[2]
245#define sc_semmni	sem_ctls[3]
246
247void sem_init_ns(struct ipc_namespace *ns)
248{
249	ns->sc_semmsl = SEMMSL;
250	ns->sc_semmns = SEMMNS;
251	ns->sc_semopm = SEMOPM;
252	ns->sc_semmni = SEMMNI;
253	ns->used_sems = 0;
254	ipc_init_ids(&ns->ids[IPC_SEM_IDS]);
255}
256
257#ifdef CONFIG_IPC_NS
258void sem_exit_ns(struct ipc_namespace *ns)
259{
260	free_ipcs(ns, &sem_ids(ns), freeary);
261	idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr);
262	rhashtable_destroy(&ns->ids[IPC_SEM_IDS].key_ht);
263}
264#endif
265
266void __init sem_init(void)
267{
268	sem_init_ns(&init_ipc_ns);
269	ipc_init_proc_interface("sysvipc/sem",
270				"       key      semid perms      nsems   uid   gid  cuid  cgid      otime      ctime\n",
271				IPC_SEM_IDS, sysvipc_sem_proc_show);
272}
273
274/**
275 * unmerge_queues - unmerge queues, if possible.
276 * @sma: semaphore array
277 *
278 * The function unmerges the wait queues if complex_count is 0.
279 * It must be called prior to dropping the global semaphore array lock.
280 */
281static void unmerge_queues(struct sem_array *sma)
282{
283	struct sem_queue *q, *tq;
284
285	/* complex operations still around? */
286	if (sma->complex_count)
287		return;
288	/*
289	 * We will switch back to simple mode.
290	 * Move all pending operation back into the per-semaphore
291	 * queues.
292	 */
293	list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
294		struct sem *curr;
295		curr = &sma->sems[q->sops[0].sem_num];
296
297		list_add_tail(&q->list, &curr->pending_alter);
298	}
299	INIT_LIST_HEAD(&sma->pending_alter);
300}
301
302/**
303 * merge_queues - merge single semop queues into global queue
304 * @sma: semaphore array
305 *
306 * This function merges all per-semaphore queues into the global queue.
307 * It is necessary to achieve FIFO ordering for the pending single-sop
308 * operations when a multi-semop operation must sleep.
309 * Only the alter operations must be moved, the const operations can stay.
310 */
311static void merge_queues(struct sem_array *sma)
312{
313	int i;
314	for (i = 0; i < sma->sem_nsems; i++) {
315		struct sem *sem = &sma->sems[i];
316
317		list_splice_init(&sem->pending_alter, &sma->pending_alter);
318	}
319}
320
321static void sem_rcu_free(struct rcu_head *head)
322{
323	struct kern_ipc_perm *p = container_of(head, struct kern_ipc_perm, rcu);
324	struct sem_array *sma = container_of(p, struct sem_array, sem_perm);
325
326	security_sem_free(&sma->sem_perm);
327	kvfree(sma);
328}
329
330/*
331 * Enter the mode suitable for non-simple operations:
332 * Caller must own sem_perm.lock.
333 */
334static void complexmode_enter(struct sem_array *sma)
335{
336	int i;
337	struct sem *sem;
338
339	if (sma->use_global_lock > 0)  {
340		/*
341		 * We are already in global lock mode.
342		 * Nothing to do, just reset the
343		 * counter until we return to simple mode.
344		 */
345		sma->use_global_lock = USE_GLOBAL_LOCK_HYSTERESIS;
346		return;
347	}
348	sma->use_global_lock = USE_GLOBAL_LOCK_HYSTERESIS;
349
350	for (i = 0; i < sma->sem_nsems; i++) {
351		sem = &sma->sems[i];
352		spin_lock(&sem->lock);
353		spin_unlock(&sem->lock);
354	}
355}
356
357/*
358 * Try to leave the mode that disallows simple operations:
359 * Caller must own sem_perm.lock.
360 */
361static void complexmode_tryleave(struct sem_array *sma)
362{
363	if (sma->complex_count)  {
364		/* Complex ops are sleeping.
365		 * We must stay in complex mode
366		 */
367		return;
368	}
369	if (sma->use_global_lock == 1) {
370
371		/* See SEM_BARRIER_1 for purpose/pairing */
372		smp_store_release(&sma->use_global_lock, 0);
373	} else {
374		sma->use_global_lock--;
375	}
376}
377
378#define SEM_GLOBAL_LOCK	(-1)
379/*
380 * If the request contains only one semaphore operation, and there are
381 * no complex transactions pending, lock only the semaphore involved.
382 * Otherwise, lock the entire semaphore array, since we either have
383 * multiple semaphores in our own semops, or we need to look at
384 * semaphores from other pending complex operations.
385 */
386static inline int sem_lock(struct sem_array *sma, struct sembuf *sops,
387			      int nsops)
388{
389	struct sem *sem;
390	int idx;
391
392	if (nsops != 1) {
393		/* Complex operation - acquire a full lock */
394		ipc_lock_object(&sma->sem_perm);
395
396		/* Prevent parallel simple ops */
397		complexmode_enter(sma);
398		return SEM_GLOBAL_LOCK;
399	}
400
401	/*
402	 * Only one semaphore affected - try to optimize locking.
403	 * Optimized locking is possible if no complex operation
404	 * is either enqueued or processed right now.
405	 *
406	 * Both facts are tracked by use_global_mode.
407	 */
408	idx = array_index_nospec(sops->sem_num, sma->sem_nsems);
409	sem = &sma->sems[idx];
410
411	/*
412	 * Initial check for use_global_lock. Just an optimization,
413	 * no locking, no memory barrier.
414	 */
415	if (!sma->use_global_lock) {
416		/*
417		 * It appears that no complex operation is around.
418		 * Acquire the per-semaphore lock.
419		 */
420		spin_lock(&sem->lock);
421
422		/* see SEM_BARRIER_1 for purpose/pairing */
423		if (!smp_load_acquire(&sma->use_global_lock)) {
424			/* fast path successful! */
425			return sops->sem_num;
426		}
427		spin_unlock(&sem->lock);
428	}
429
430	/* slow path: acquire the full lock */
431	ipc_lock_object(&sma->sem_perm);
432
433	if (sma->use_global_lock == 0) {
434		/*
435		 * The use_global_lock mode ended while we waited for
436		 * sma->sem_perm.lock. Thus we must switch to locking
437		 * with sem->lock.
438		 * Unlike in the fast path, there is no need to recheck
439		 * sma->use_global_lock after we have acquired sem->lock:
440		 * We own sma->sem_perm.lock, thus use_global_lock cannot
441		 * change.
442		 */
443		spin_lock(&sem->lock);
444
445		ipc_unlock_object(&sma->sem_perm);
446		return sops->sem_num;
447	} else {
448		/*
449		 * Not a false alarm, thus continue to use the global lock
450		 * mode. No need for complexmode_enter(), this was done by
451		 * the caller that has set use_global_mode to non-zero.
452		 */
453		return SEM_GLOBAL_LOCK;
454	}
455}
456
457static inline void sem_unlock(struct sem_array *sma, int locknum)
458{
459	if (locknum == SEM_GLOBAL_LOCK) {
460		unmerge_queues(sma);
461		complexmode_tryleave(sma);
462		ipc_unlock_object(&sma->sem_perm);
463	} else {
464		struct sem *sem = &sma->sems[locknum];
465		spin_unlock(&sem->lock);
466	}
467}
468
469/*
470 * sem_lock_(check_) routines are called in the paths where the rwsem
471 * is not held.
472 *
473 * The caller holds the RCU read lock.
474 */
475static inline struct sem_array *sem_obtain_object(struct ipc_namespace *ns, int id)
476{
477	struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&sem_ids(ns), id);
478
479	if (IS_ERR(ipcp))
480		return ERR_CAST(ipcp);
481
482	return container_of(ipcp, struct sem_array, sem_perm);
483}
484
485static inline struct sem_array *sem_obtain_object_check(struct ipc_namespace *ns,
486							int id)
487{
488	struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&sem_ids(ns), id);
489
490	if (IS_ERR(ipcp))
491		return ERR_CAST(ipcp);
492
493	return container_of(ipcp, struct sem_array, sem_perm);
494}
495
496static inline void sem_lock_and_putref(struct sem_array *sma)
497{
498	sem_lock(sma, NULL, -1);
499	ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
500}
501
502static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
503{
504	ipc_rmid(&sem_ids(ns), &s->sem_perm);
505}
506
507static struct sem_array *sem_alloc(size_t nsems)
508{
509	struct sem_array *sma;
510
511	if (nsems > (INT_MAX - sizeof(*sma)) / sizeof(sma->sems[0]))
512		return NULL;
513
514	sma = kvzalloc(struct_size(sma, sems, nsems), GFP_KERNEL);
515	if (unlikely(!sma))
516		return NULL;
517
518	return sma;
519}
520
521/**
522 * newary - Create a new semaphore set
523 * @ns: namespace
524 * @params: ptr to the structure that contains key, semflg and nsems
525 *
526 * Called with sem_ids.rwsem held (as a writer)
527 */
528static int newary(struct ipc_namespace *ns, struct ipc_params *params)
529{
530	int retval;
531	struct sem_array *sma;
532	key_t key = params->key;
533	int nsems = params->u.nsems;
534	int semflg = params->flg;
535	int i;
536
537	if (!nsems)
538		return -EINVAL;
539	if (ns->used_sems + nsems > ns->sc_semmns)
540		return -ENOSPC;
541
542	sma = sem_alloc(nsems);
543	if (!sma)
544		return -ENOMEM;
545
546	sma->sem_perm.mode = (semflg & S_IRWXUGO);
547	sma->sem_perm.key = key;
548
549	sma->sem_perm.security = NULL;
550	retval = security_sem_alloc(&sma->sem_perm);
551	if (retval) {
552		kvfree(sma);
553		return retval;
554	}
555
556	for (i = 0; i < nsems; i++) {
557		INIT_LIST_HEAD(&sma->sems[i].pending_alter);
558		INIT_LIST_HEAD(&sma->sems[i].pending_const);
559		spin_lock_init(&sma->sems[i].lock);
560	}
561
562	sma->complex_count = 0;
563	sma->use_global_lock = USE_GLOBAL_LOCK_HYSTERESIS;
564	INIT_LIST_HEAD(&sma->pending_alter);
565	INIT_LIST_HEAD(&sma->pending_const);
566	INIT_LIST_HEAD(&sma->list_id);
567	sma->sem_nsems = nsems;
568	sma->sem_ctime = ktime_get_real_seconds();
569
570	/* ipc_addid() locks sma upon success. */
571	retval = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
572	if (retval < 0) {
573		ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
574		return retval;
575	}
576	ns->used_sems += nsems;
577
578	sem_unlock(sma, -1);
579	rcu_read_unlock();
580
581	return sma->sem_perm.id;
582}
583
584
585/*
586 * Called with sem_ids.rwsem and ipcp locked.
587 */
588static int sem_more_checks(struct kern_ipc_perm *ipcp, struct ipc_params *params)
589{
590	struct sem_array *sma;
591
592	sma = container_of(ipcp, struct sem_array, sem_perm);
593	if (params->u.nsems > sma->sem_nsems)
594		return -EINVAL;
595
596	return 0;
597}
598
599long ksys_semget(key_t key, int nsems, int semflg)
600{
601	struct ipc_namespace *ns;
602	static const struct ipc_ops sem_ops = {
603		.getnew = newary,
604		.associate = security_sem_associate,
605		.more_checks = sem_more_checks,
606	};
607	struct ipc_params sem_params;
608
609	ns = current->nsproxy->ipc_ns;
610
611	if (nsems < 0 || nsems > ns->sc_semmsl)
612		return -EINVAL;
613
614	sem_params.key = key;
615	sem_params.flg = semflg;
616	sem_params.u.nsems = nsems;
617
618	return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
619}
620
621SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg)
622{
623	return ksys_semget(key, nsems, semflg);
624}
625
626/**
627 * perform_atomic_semop[_slow] - Attempt to perform semaphore
628 *                               operations on a given array.
629 * @sma: semaphore array
630 * @q: struct sem_queue that describes the operation
631 *
632 * Caller blocking are as follows, based the value
633 * indicated by the semaphore operation (sem_op):
634 *
635 *  (1) >0 never blocks.
636 *  (2)  0 (wait-for-zero operation): semval is non-zero.
637 *  (3) <0 attempting to decrement semval to a value smaller than zero.
638 *
639 * Returns 0 if the operation was possible.
640 * Returns 1 if the operation is impossible, the caller must sleep.
641 * Returns <0 for error codes.
642 */
643static int perform_atomic_semop_slow(struct sem_array *sma, struct sem_queue *q)
644{
645	int result, sem_op, nsops;
646	struct pid *pid;
647	struct sembuf *sop;
648	struct sem *curr;
649	struct sembuf *sops;
650	struct sem_undo *un;
651
652	sops = q->sops;
653	nsops = q->nsops;
654	un = q->undo;
655
656	for (sop = sops; sop < sops + nsops; sop++) {
657		int idx = array_index_nospec(sop->sem_num, sma->sem_nsems);
658		curr = &sma->sems[idx];
659		sem_op = sop->sem_op;
660		result = curr->semval;
661
662		if (!sem_op && result)
663			goto would_block;
664
665		result += sem_op;
666		if (result < 0)
667			goto would_block;
668		if (result > SEMVMX)
669			goto out_of_range;
670
671		if (sop->sem_flg & SEM_UNDO) {
672			int undo = un->semadj[sop->sem_num] - sem_op;
673			/* Exceeding the undo range is an error. */
674			if (undo < (-SEMAEM - 1) || undo > SEMAEM)
675				goto out_of_range;
676			un->semadj[sop->sem_num] = undo;
677		}
678
679		curr->semval = result;
680	}
681
682	sop--;
683	pid = q->pid;
684	while (sop >= sops) {
685		ipc_update_pid(&sma->sems[sop->sem_num].sempid, pid);
686		sop--;
687	}
688
689	return 0;
690
691out_of_range:
692	result = -ERANGE;
693	goto undo;
694
695would_block:
696	q->blocking = sop;
697
698	if (sop->sem_flg & IPC_NOWAIT)
699		result = -EAGAIN;
700	else
701		result = 1;
702
703undo:
704	sop--;
705	while (sop >= sops) {
706		sem_op = sop->sem_op;
707		sma->sems[sop->sem_num].semval -= sem_op;
708		if (sop->sem_flg & SEM_UNDO)
709			un->semadj[sop->sem_num] += sem_op;
710		sop--;
711	}
712
713	return result;
714}
715
716static int perform_atomic_semop(struct sem_array *sma, struct sem_queue *q)
717{
718	int result, sem_op, nsops;
719	struct sembuf *sop;
720	struct sem *curr;
721	struct sembuf *sops;
722	struct sem_undo *un;
723
724	sops = q->sops;
725	nsops = q->nsops;
726	un = q->undo;
727
728	if (unlikely(q->dupsop))
729		return perform_atomic_semop_slow(sma, q);
730
731	/*
732	 * We scan the semaphore set twice, first to ensure that the entire
733	 * operation can succeed, therefore avoiding any pointless writes
734	 * to shared memory and having to undo such changes in order to block
735	 * until the operations can go through.
736	 */
737	for (sop = sops; sop < sops + nsops; sop++) {
738		int idx = array_index_nospec(sop->sem_num, sma->sem_nsems);
739
740		curr = &sma->sems[idx];
741		sem_op = sop->sem_op;
742		result = curr->semval;
743
744		if (!sem_op && result)
745			goto would_block; /* wait-for-zero */
746
747		result += sem_op;
748		if (result < 0)
749			goto would_block;
750
751		if (result > SEMVMX)
752			return -ERANGE;
753
754		if (sop->sem_flg & SEM_UNDO) {
755			int undo = un->semadj[sop->sem_num] - sem_op;
756
757			/* Exceeding the undo range is an error. */
758			if (undo < (-SEMAEM - 1) || undo > SEMAEM)
759				return -ERANGE;
760		}
761	}
762
763	for (sop = sops; sop < sops + nsops; sop++) {
764		curr = &sma->sems[sop->sem_num];
765		sem_op = sop->sem_op;
766		result = curr->semval;
767
768		if (sop->sem_flg & SEM_UNDO) {
769			int undo = un->semadj[sop->sem_num] - sem_op;
770
771			un->semadj[sop->sem_num] = undo;
772		}
773		curr->semval += sem_op;
774		ipc_update_pid(&curr->sempid, q->pid);
775	}
776
777	return 0;
778
779would_block:
780	q->blocking = sop;
781	return sop->sem_flg & IPC_NOWAIT ? -EAGAIN : 1;
782}
783
784static inline void wake_up_sem_queue_prepare(struct sem_queue *q, int error,
785					     struct wake_q_head *wake_q)
786{
787	get_task_struct(q->sleeper);
788
789	/* see SEM_BARRIER_2 for purpuse/pairing */
790	smp_store_release(&q->status, error);
791
792	wake_q_add_safe(wake_q, q->sleeper);
793}
794
795static void unlink_queue(struct sem_array *sma, struct sem_queue *q)
796{
797	list_del(&q->list);
798	if (q->nsops > 1)
799		sma->complex_count--;
800}
801
802/** check_restart(sma, q)
803 * @sma: semaphore array
804 * @q: the operation that just completed
805 *
806 * update_queue is O(N^2) when it restarts scanning the whole queue of
807 * waiting operations. Therefore this function checks if the restart is
808 * really necessary. It is called after a previously waiting operation
809 * modified the array.
810 * Note that wait-for-zero operations are handled without restart.
811 */
812static inline int check_restart(struct sem_array *sma, struct sem_queue *q)
813{
814	/* pending complex alter operations are too difficult to analyse */
815	if (!list_empty(&sma->pending_alter))
816		return 1;
817
818	/* we were a sleeping complex operation. Too difficult */
819	if (q->nsops > 1)
820		return 1;
821
822	/* It is impossible that someone waits for the new value:
823	 * - complex operations always restart.
824	 * - wait-for-zero are handled separately.
825	 * - q is a previously sleeping simple operation that
826	 *   altered the array. It must be a decrement, because
827	 *   simple increments never sleep.
828	 * - If there are older (higher priority) decrements
829	 *   in the queue, then they have observed the original
830	 *   semval value and couldn't proceed. The operation
831	 *   decremented to value - thus they won't proceed either.
832	 */
833	return 0;
834}
835
836/**
837 * wake_const_ops - wake up non-alter tasks
838 * @sma: semaphore array.
839 * @semnum: semaphore that was modified.
840 * @wake_q: lockless wake-queue head.
841 *
842 * wake_const_ops must be called after a semaphore in a semaphore array
843 * was set to 0. If complex const operations are pending, wake_const_ops must
844 * be called with semnum = -1, as well as with the number of each modified
845 * semaphore.
846 * The tasks that must be woken up are added to @wake_q. The return code
847 * is stored in q->pid.
848 * The function returns 1 if at least one operation was completed successfully.
849 */
850static int wake_const_ops(struct sem_array *sma, int semnum,
851			  struct wake_q_head *wake_q)
852{
853	struct sem_queue *q, *tmp;
854	struct list_head *pending_list;
855	int semop_completed = 0;
856
857	if (semnum == -1)
858		pending_list = &sma->pending_const;
859	else
860		pending_list = &sma->sems[semnum].pending_const;
861
862	list_for_each_entry_safe(q, tmp, pending_list, list) {
863		int error = perform_atomic_semop(sma, q);
864
865		if (error > 0)
866			continue;
867		/* operation completed, remove from queue & wakeup */
868		unlink_queue(sma, q);
869
870		wake_up_sem_queue_prepare(q, error, wake_q);
871		if (error == 0)
872			semop_completed = 1;
873	}
874
875	return semop_completed;
876}
877
878/**
879 * do_smart_wakeup_zero - wakeup all wait for zero tasks
880 * @sma: semaphore array
881 * @sops: operations that were performed
882 * @nsops: number of operations
883 * @wake_q: lockless wake-queue head
884 *
885 * Checks all required queue for wait-for-zero operations, based
886 * on the actual changes that were performed on the semaphore array.
887 * The function returns 1 if at least one operation was completed successfully.
888 */
889static int do_smart_wakeup_zero(struct sem_array *sma, struct sembuf *sops,
890				int nsops, struct wake_q_head *wake_q)
891{
892	int i;
893	int semop_completed = 0;
894	int got_zero = 0;
895
896	/* first: the per-semaphore queues, if known */
897	if (sops) {
898		for (i = 0; i < nsops; i++) {
899			int num = sops[i].sem_num;
900
901			if (sma->sems[num].semval == 0) {
902				got_zero = 1;
903				semop_completed |= wake_const_ops(sma, num, wake_q);
904			}
905		}
906	} else {
907		/*
908		 * No sops means modified semaphores not known.
909		 * Assume all were changed.
910		 */
911		for (i = 0; i < sma->sem_nsems; i++) {
912			if (sma->sems[i].semval == 0) {
913				got_zero = 1;
914				semop_completed |= wake_const_ops(sma, i, wake_q);
915			}
916		}
917	}
918	/*
919	 * If one of the modified semaphores got 0,
920	 * then check the global queue, too.
921	 */
922	if (got_zero)
923		semop_completed |= wake_const_ops(sma, -1, wake_q);
924
925	return semop_completed;
926}
927
928
929/**
930 * update_queue - look for tasks that can be completed.
931 * @sma: semaphore array.
932 * @semnum: semaphore that was modified.
933 * @wake_q: lockless wake-queue head.
934 *
935 * update_queue must be called after a semaphore in a semaphore array
936 * was modified. If multiple semaphores were modified, update_queue must
937 * be called with semnum = -1, as well as with the number of each modified
938 * semaphore.
939 * The tasks that must be woken up are added to @wake_q. The return code
940 * is stored in q->pid.
941 * The function internally checks if const operations can now succeed.
942 *
943 * The function return 1 if at least one semop was completed successfully.
944 */
945static int update_queue(struct sem_array *sma, int semnum, struct wake_q_head *wake_q)
946{
947	struct sem_queue *q, *tmp;
948	struct list_head *pending_list;
949	int semop_completed = 0;
950
951	if (semnum == -1)
952		pending_list = &sma->pending_alter;
953	else
954		pending_list = &sma->sems[semnum].pending_alter;
955
956again:
957	list_for_each_entry_safe(q, tmp, pending_list, list) {
958		int error, restart;
959
960		/* If we are scanning the single sop, per-semaphore list of
961		 * one semaphore and that semaphore is 0, then it is not
962		 * necessary to scan further: simple increments
963		 * that affect only one entry succeed immediately and cannot
964		 * be in the  per semaphore pending queue, and decrements
965		 * cannot be successful if the value is already 0.
966		 */
967		if (semnum != -1 && sma->sems[semnum].semval == 0)
968			break;
969
970		error = perform_atomic_semop(sma, q);
971
972		/* Does q->sleeper still need to sleep? */
973		if (error > 0)
974			continue;
975
976		unlink_queue(sma, q);
977
978		if (error) {
979			restart = 0;
980		} else {
981			semop_completed = 1;
982			do_smart_wakeup_zero(sma, q->sops, q->nsops, wake_q);
983			restart = check_restart(sma, q);
984		}
985
986		wake_up_sem_queue_prepare(q, error, wake_q);
987		if (restart)
988			goto again;
989	}
990	return semop_completed;
991}
992
993/**
994 * set_semotime - set sem_otime
995 * @sma: semaphore array
996 * @sops: operations that modified the array, may be NULL
997 *
998 * sem_otime is replicated to avoid cache line trashing.
999 * This function sets one instance to the current time.
1000 */
1001static void set_semotime(struct sem_array *sma, struct sembuf *sops)
1002{
1003	if (sops == NULL) {
1004		sma->sems[0].sem_otime = ktime_get_real_seconds();
1005	} else {
1006		sma->sems[sops[0].sem_num].sem_otime =
1007						ktime_get_real_seconds();
1008	}
1009}
1010
1011/**
1012 * do_smart_update - optimized update_queue
1013 * @sma: semaphore array
1014 * @sops: operations that were performed
1015 * @nsops: number of operations
1016 * @otime: force setting otime
1017 * @wake_q: lockless wake-queue head
1018 *
1019 * do_smart_update() does the required calls to update_queue and wakeup_zero,
1020 * based on the actual changes that were performed on the semaphore array.
1021 * Note that the function does not do the actual wake-up: the caller is
1022 * responsible for calling wake_up_q().
1023 * It is safe to perform this call after dropping all locks.
1024 */
1025static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops,
1026			    int otime, struct wake_q_head *wake_q)
1027{
1028	int i;
1029
1030	otime |= do_smart_wakeup_zero(sma, sops, nsops, wake_q);
1031
1032	if (!list_empty(&sma->pending_alter)) {
1033		/* semaphore array uses the global queue - just process it. */
1034		otime |= update_queue(sma, -1, wake_q);
1035	} else {
1036		if (!sops) {
1037			/*
1038			 * No sops, thus the modified semaphores are not
1039			 * known. Check all.
1040			 */
1041			for (i = 0; i < sma->sem_nsems; i++)
1042				otime |= update_queue(sma, i, wake_q);
1043		} else {
1044			/*
1045			 * Check the semaphores that were increased:
1046			 * - No complex ops, thus all sleeping ops are
1047			 *   decrease.
1048			 * - if we decreased the value, then any sleeping
1049			 *   semaphore ops won't be able to run: If the
1050			 *   previous value was too small, then the new
1051			 *   value will be too small, too.
1052			 */
1053			for (i = 0; i < nsops; i++) {
1054				if (sops[i].sem_op > 0) {
1055					otime |= update_queue(sma,
1056							      sops[i].sem_num, wake_q);
1057				}
1058			}
1059		}
1060	}
1061	if (otime)
1062		set_semotime(sma, sops);
1063}
1064
1065/*
1066 * check_qop: Test if a queued operation sleeps on the semaphore semnum
1067 */
1068static int check_qop(struct sem_array *sma, int semnum, struct sem_queue *q,
1069			bool count_zero)
1070{
1071	struct sembuf *sop = q->blocking;
1072
1073	/*
1074	 * Linux always (since 0.99.10) reported a task as sleeping on all
1075	 * semaphores. This violates SUS, therefore it was changed to the
1076	 * standard compliant behavior.
1077	 * Give the administrators a chance to notice that an application
1078	 * might misbehave because it relies on the Linux behavior.
1079	 */
1080	pr_info_once("semctl(GETNCNT/GETZCNT) is since 3.16 Single Unix Specification compliant.\n"
1081			"The task %s (%d) triggered the difference, watch for misbehavior.\n",
1082			current->comm, task_pid_nr(current));
1083
1084	if (sop->sem_num != semnum)
1085		return 0;
1086
1087	if (count_zero && sop->sem_op == 0)
1088		return 1;
1089	if (!count_zero && sop->sem_op < 0)
1090		return 1;
1091
1092	return 0;
1093}
1094
1095/* The following counts are associated to each semaphore:
1096 *   semncnt        number of tasks waiting on semval being nonzero
1097 *   semzcnt        number of tasks waiting on semval being zero
1098 *
1099 * Per definition, a task waits only on the semaphore of the first semop
1100 * that cannot proceed, even if additional operation would block, too.
1101 */
1102static int count_semcnt(struct sem_array *sma, ushort semnum,
1103			bool count_zero)
1104{
1105	struct list_head *l;
1106	struct sem_queue *q;
1107	int semcnt;
1108
1109	semcnt = 0;
1110	/* First: check the simple operations. They are easy to evaluate */
1111	if (count_zero)
1112		l = &sma->sems[semnum].pending_const;
1113	else
1114		l = &sma->sems[semnum].pending_alter;
1115
1116	list_for_each_entry(q, l, list) {
1117		/* all task on a per-semaphore list sleep on exactly
1118		 * that semaphore
1119		 */
1120		semcnt++;
1121	}
1122
1123	/* Then: check the complex operations. */
1124	list_for_each_entry(q, &sma->pending_alter, list) {
1125		semcnt += check_qop(sma, semnum, q, count_zero);
1126	}
1127	if (count_zero) {
1128		list_for_each_entry(q, &sma->pending_const, list) {
1129			semcnt += check_qop(sma, semnum, q, count_zero);
1130		}
1131	}
1132	return semcnt;
1133}
1134
1135/* Free a semaphore set. freeary() is called with sem_ids.rwsem locked
1136 * as a writer and the spinlock for this semaphore set hold. sem_ids.rwsem
1137 * remains locked on exit.
1138 */
1139static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
1140{
1141	struct sem_undo *un, *tu;
1142	struct sem_queue *q, *tq;
1143	struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
1144	int i;
1145	DEFINE_WAKE_Q(wake_q);
1146
1147	/* Free the existing undo structures for this semaphore set.  */
1148	ipc_assert_locked_object(&sma->sem_perm);
1149	list_for_each_entry_safe(un, tu, &sma->list_id, list_id) {
1150		list_del(&un->list_id);
1151		spin_lock(&un->ulp->lock);
1152		un->semid = -1;
1153		list_del_rcu(&un->list_proc);
1154		spin_unlock(&un->ulp->lock);
1155		kfree_rcu(un, rcu);
1156	}
1157
1158	/* Wake up all pending processes and let them fail with EIDRM. */
1159	list_for_each_entry_safe(q, tq, &sma->pending_const, list) {
1160		unlink_queue(sma, q);
1161		wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
1162	}
1163
1164	list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
1165		unlink_queue(sma, q);
1166		wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
1167	}
1168	for (i = 0; i < sma->sem_nsems; i++) {
1169		struct sem *sem = &sma->sems[i];
1170		list_for_each_entry_safe(q, tq, &sem->pending_const, list) {
1171			unlink_queue(sma, q);
1172			wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
1173		}
1174		list_for_each_entry_safe(q, tq, &sem->pending_alter, list) {
1175			unlink_queue(sma, q);
1176			wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
1177		}
1178		ipc_update_pid(&sem->sempid, NULL);
1179	}
1180
1181	/* Remove the semaphore set from the IDR */
1182	sem_rmid(ns, sma);
1183	sem_unlock(sma, -1);
1184	rcu_read_unlock();
1185
1186	wake_up_q(&wake_q);
1187	ns->used_sems -= sma->sem_nsems;
1188	ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
1189}
1190
1191static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
1192{
1193	switch (version) {
1194	case IPC_64:
1195		return copy_to_user(buf, in, sizeof(*in));
1196	case IPC_OLD:
1197	    {
1198		struct semid_ds out;
1199
1200		memset(&out, 0, sizeof(out));
1201
1202		ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
1203
1204		out.sem_otime	= in->sem_otime;
1205		out.sem_ctime	= in->sem_ctime;
1206		out.sem_nsems	= in->sem_nsems;
1207
1208		return copy_to_user(buf, &out, sizeof(out));
1209	    }
1210	default:
1211		return -EINVAL;
1212	}
1213}
1214
1215static time64_t get_semotime(struct sem_array *sma)
1216{
1217	int i;
1218	time64_t res;
1219
1220	res = sma->sems[0].sem_otime;
1221	for (i = 1; i < sma->sem_nsems; i++) {
1222		time64_t to = sma->sems[i].sem_otime;
1223
1224		if (to > res)
1225			res = to;
1226	}
1227	return res;
1228}
1229
1230static int semctl_stat(struct ipc_namespace *ns, int semid,
1231			 int cmd, struct semid64_ds *semid64)
1232{
1233	struct sem_array *sma;
1234	time64_t semotime;
1235	int err;
1236
1237	memset(semid64, 0, sizeof(*semid64));
1238
1239	rcu_read_lock();
1240	if (cmd == SEM_STAT || cmd == SEM_STAT_ANY) {
1241		sma = sem_obtain_object(ns, semid);
1242		if (IS_ERR(sma)) {
1243			err = PTR_ERR(sma);
1244			goto out_unlock;
1245		}
1246	} else { /* IPC_STAT */
1247		sma = sem_obtain_object_check(ns, semid);
1248		if (IS_ERR(sma)) {
1249			err = PTR_ERR(sma);
1250			goto out_unlock;
1251		}
1252	}
1253
1254	/* see comment for SHM_STAT_ANY */
1255	if (cmd == SEM_STAT_ANY)
1256		audit_ipc_obj(&sma->sem_perm);
1257	else {
1258		err = -EACCES;
1259		if (ipcperms(ns, &sma->sem_perm, S_IRUGO))
1260			goto out_unlock;
1261	}
1262
1263	err = security_sem_semctl(&sma->sem_perm, cmd);
1264	if (err)
1265		goto out_unlock;
1266
1267	ipc_lock_object(&sma->sem_perm);
1268
1269	if (!ipc_valid_object(&sma->sem_perm)) {
1270		ipc_unlock_object(&sma->sem_perm);
1271		err = -EIDRM;
1272		goto out_unlock;
1273	}
1274
1275	kernel_to_ipc64_perm(&sma->sem_perm, &semid64->sem_perm);
1276	semotime = get_semotime(sma);
1277	semid64->sem_otime = semotime;
1278	semid64->sem_ctime = sma->sem_ctime;
1279#ifndef CONFIG_64BIT
1280	semid64->sem_otime_high = semotime >> 32;
1281	semid64->sem_ctime_high = sma->sem_ctime >> 32;
1282#endif
1283	semid64->sem_nsems = sma->sem_nsems;
1284
1285	if (cmd == IPC_STAT) {
1286		/*
1287		 * As defined in SUS:
1288		 * Return 0 on success
1289		 */
1290		err = 0;
1291	} else {
1292		/*
1293		 * SEM_STAT and SEM_STAT_ANY (both Linux specific)
1294		 * Return the full id, including the sequence number
1295		 */
1296		err = sma->sem_perm.id;
1297	}
1298	ipc_unlock_object(&sma->sem_perm);
1299out_unlock:
1300	rcu_read_unlock();
1301	return err;
1302}
1303
1304static int semctl_info(struct ipc_namespace *ns, int semid,
1305			 int cmd, void __user *p)
1306{
1307	struct seminfo seminfo;
1308	int max_idx;
1309	int err;
1310
1311	err = security_sem_semctl(NULL, cmd);
1312	if (err)
1313		return err;
1314
1315	memset(&seminfo, 0, sizeof(seminfo));
1316	seminfo.semmni = ns->sc_semmni;
1317	seminfo.semmns = ns->sc_semmns;
1318	seminfo.semmsl = ns->sc_semmsl;
1319	seminfo.semopm = ns->sc_semopm;
1320	seminfo.semvmx = SEMVMX;
1321	seminfo.semmnu = SEMMNU;
1322	seminfo.semmap = SEMMAP;
1323	seminfo.semume = SEMUME;
1324	down_read(&sem_ids(ns).rwsem);
1325	if (cmd == SEM_INFO) {
1326		seminfo.semusz = sem_ids(ns).in_use;
1327		seminfo.semaem = ns->used_sems;
1328	} else {
1329		seminfo.semusz = SEMUSZ;
1330		seminfo.semaem = SEMAEM;
1331	}
1332	max_idx = ipc_get_maxidx(&sem_ids(ns));
1333	up_read(&sem_ids(ns).rwsem);
1334	if (copy_to_user(p, &seminfo, sizeof(struct seminfo)))
1335		return -EFAULT;
1336	return (max_idx < 0) ? 0 : max_idx;
1337}
1338
1339static int semctl_setval(struct ipc_namespace *ns, int semid, int semnum,
1340		int val)
1341{
1342	struct sem_undo *un;
1343	struct sem_array *sma;
1344	struct sem *curr;
1345	int err;
1346	DEFINE_WAKE_Q(wake_q);
1347
1348	if (val > SEMVMX || val < 0)
1349		return -ERANGE;
1350
1351	rcu_read_lock();
1352	sma = sem_obtain_object_check(ns, semid);
1353	if (IS_ERR(sma)) {
1354		rcu_read_unlock();
1355		return PTR_ERR(sma);
1356	}
1357
1358	if (semnum < 0 || semnum >= sma->sem_nsems) {
1359		rcu_read_unlock();
1360		return -EINVAL;
1361	}
1362
1363
1364	if (ipcperms(ns, &sma->sem_perm, S_IWUGO)) {
1365		rcu_read_unlock();
1366		return -EACCES;
1367	}
1368
1369	err = security_sem_semctl(&sma->sem_perm, SETVAL);
1370	if (err) {
1371		rcu_read_unlock();
1372		return -EACCES;
1373	}
1374
1375	sem_lock(sma, NULL, -1);
1376
1377	if (!ipc_valid_object(&sma->sem_perm)) {
1378		sem_unlock(sma, -1);
1379		rcu_read_unlock();
1380		return -EIDRM;
1381	}
1382
1383	semnum = array_index_nospec(semnum, sma->sem_nsems);
1384	curr = &sma->sems[semnum];
1385
1386	ipc_assert_locked_object(&sma->sem_perm);
1387	list_for_each_entry(un, &sma->list_id, list_id)
1388		un->semadj[semnum] = 0;
1389
1390	curr->semval = val;
1391	ipc_update_pid(&curr->sempid, task_tgid(current));
1392	sma->sem_ctime = ktime_get_real_seconds();
1393	/* maybe some queued-up processes were waiting for this */
1394	do_smart_update(sma, NULL, 0, 0, &wake_q);
1395	sem_unlock(sma, -1);
1396	rcu_read_unlock();
1397	wake_up_q(&wake_q);
1398	return 0;
1399}
1400
1401static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
1402		int cmd, void __user *p)
1403{
1404	struct sem_array *sma;
1405	struct sem *curr;
1406	int err, nsems;
1407	ushort fast_sem_io[SEMMSL_FAST];
1408	ushort *sem_io = fast_sem_io;
1409	DEFINE_WAKE_Q(wake_q);
1410
1411	rcu_read_lock();
1412	sma = sem_obtain_object_check(ns, semid);
1413	if (IS_ERR(sma)) {
1414		rcu_read_unlock();
1415		return PTR_ERR(sma);
1416	}
1417
1418	nsems = sma->sem_nsems;
1419
1420	err = -EACCES;
1421	if (ipcperms(ns, &sma->sem_perm, cmd == SETALL ? S_IWUGO : S_IRUGO))
1422		goto out_rcu_wakeup;
1423
1424	err = security_sem_semctl(&sma->sem_perm, cmd);
1425	if (err)
1426		goto out_rcu_wakeup;
1427
1428	err = -EACCES;
1429	switch (cmd) {
1430	case GETALL:
1431	{
1432		ushort __user *array = p;
1433		int i;
1434
1435		sem_lock(sma, NULL, -1);
1436		if (!ipc_valid_object(&sma->sem_perm)) {
1437			err = -EIDRM;
1438			goto out_unlock;
1439		}
1440		if (nsems > SEMMSL_FAST) {
1441			if (!ipc_rcu_getref(&sma->sem_perm)) {
1442				err = -EIDRM;
1443				goto out_unlock;
1444			}
1445			sem_unlock(sma, -1);
1446			rcu_read_unlock();
1447			sem_io = kvmalloc_array(nsems, sizeof(ushort),
1448						GFP_KERNEL);
1449			if (sem_io == NULL) {
1450				ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
1451				return -ENOMEM;
1452			}
1453
1454			rcu_read_lock();
1455			sem_lock_and_putref(sma);
1456			if (!ipc_valid_object(&sma->sem_perm)) {
1457				err = -EIDRM;
1458				goto out_unlock;
1459			}
1460		}
1461		for (i = 0; i < sma->sem_nsems; i++)
1462			sem_io[i] = sma->sems[i].semval;
1463		sem_unlock(sma, -1);
1464		rcu_read_unlock();
1465		err = 0;
1466		if (copy_to_user(array, sem_io, nsems*sizeof(ushort)))
1467			err = -EFAULT;
1468		goto out_free;
1469	}
1470	case SETALL:
1471	{
1472		int i;
1473		struct sem_undo *un;
1474
1475		if (!ipc_rcu_getref(&sma->sem_perm)) {
1476			err = -EIDRM;
1477			goto out_rcu_wakeup;
1478		}
1479		rcu_read_unlock();
1480
1481		if (nsems > SEMMSL_FAST) {
1482			sem_io = kvmalloc_array(nsems, sizeof(ushort),
1483						GFP_KERNEL);
1484			if (sem_io == NULL) {
1485				ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
1486				return -ENOMEM;
1487			}
1488		}
1489
1490		if (copy_from_user(sem_io, p, nsems*sizeof(ushort))) {
1491			ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
1492			err = -EFAULT;
1493			goto out_free;
1494		}
1495
1496		for (i = 0; i < nsems; i++) {
1497			if (sem_io[i] > SEMVMX) {
1498				ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
1499				err = -ERANGE;
1500				goto out_free;
1501			}
1502		}
1503		rcu_read_lock();
1504		sem_lock_and_putref(sma);
1505		if (!ipc_valid_object(&sma->sem_perm)) {
1506			err = -EIDRM;
1507			goto out_unlock;
1508		}
1509
1510		for (i = 0; i < nsems; i++) {
1511			sma->sems[i].semval = sem_io[i];
1512			ipc_update_pid(&sma->sems[i].sempid, task_tgid(current));
1513		}
1514
1515		ipc_assert_locked_object(&sma->sem_perm);
1516		list_for_each_entry(un, &sma->list_id, list_id) {
1517			for (i = 0; i < nsems; i++)
1518				un->semadj[i] = 0;
1519		}
1520		sma->sem_ctime = ktime_get_real_seconds();
1521		/* maybe some queued-up processes were waiting for this */
1522		do_smart_update(sma, NULL, 0, 0, &wake_q);
1523		err = 0;
1524		goto out_unlock;
1525	}
1526	/* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */
1527	}
1528	err = -EINVAL;
1529	if (semnum < 0 || semnum >= nsems)
1530		goto out_rcu_wakeup;
1531
1532	sem_lock(sma, NULL, -1);
1533	if (!ipc_valid_object(&sma->sem_perm)) {
1534		err = -EIDRM;
1535		goto out_unlock;
1536	}
1537
1538	semnum = array_index_nospec(semnum, nsems);
1539	curr = &sma->sems[semnum];
1540
1541	switch (cmd) {
1542	case GETVAL:
1543		err = curr->semval;
1544		goto out_unlock;
1545	case GETPID:
1546		err = pid_vnr(curr->sempid);
1547		goto out_unlock;
1548	case GETNCNT:
1549		err = count_semcnt(sma, semnum, 0);
1550		goto out_unlock;
1551	case GETZCNT:
1552		err = count_semcnt(sma, semnum, 1);
1553		goto out_unlock;
1554	}
1555
1556out_unlock:
1557	sem_unlock(sma, -1);
1558out_rcu_wakeup:
1559	rcu_read_unlock();
1560	wake_up_q(&wake_q);
1561out_free:
1562	if (sem_io != fast_sem_io)
1563		kvfree(sem_io);
1564	return err;
1565}
1566
1567static inline unsigned long
1568copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
1569{
1570	switch (version) {
1571	case IPC_64:
1572		if (copy_from_user(out, buf, sizeof(*out)))
1573			return -EFAULT;
1574		return 0;
1575	case IPC_OLD:
1576	    {
1577		struct semid_ds tbuf_old;
1578
1579		if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
1580			return -EFAULT;
1581
1582		out->sem_perm.uid	= tbuf_old.sem_perm.uid;
1583		out->sem_perm.gid	= tbuf_old.sem_perm.gid;
1584		out->sem_perm.mode	= tbuf_old.sem_perm.mode;
1585
1586		return 0;
1587	    }
1588	default:
1589		return -EINVAL;
1590	}
1591}
1592
1593/*
1594 * This function handles some semctl commands which require the rwsem
1595 * to be held in write mode.
1596 * NOTE: no locks must be held, the rwsem is taken inside this function.
1597 */
1598static int semctl_down(struct ipc_namespace *ns, int semid,
1599		       int cmd, struct semid64_ds *semid64)
1600{
1601	struct sem_array *sma;
1602	int err;
1603	struct kern_ipc_perm *ipcp;
1604
1605	down_write(&sem_ids(ns).rwsem);
1606	rcu_read_lock();
1607
1608	ipcp = ipcctl_obtain_check(ns, &sem_ids(ns), semid, cmd,
1609				      &semid64->sem_perm, 0);
1610	if (IS_ERR(ipcp)) {
1611		err = PTR_ERR(ipcp);
1612		goto out_unlock1;
1613	}
1614
1615	sma = container_of(ipcp, struct sem_array, sem_perm);
1616
1617	err = security_sem_semctl(&sma->sem_perm, cmd);
1618	if (err)
1619		goto out_unlock1;
1620
1621	switch (cmd) {
1622	case IPC_RMID:
1623		sem_lock(sma, NULL, -1);
1624		/* freeary unlocks the ipc object and rcu */
1625		freeary(ns, ipcp);
1626		goto out_up;
1627	case IPC_SET:
1628		sem_lock(sma, NULL, -1);
1629		err = ipc_update_perm(&semid64->sem_perm, ipcp);
1630		if (err)
1631			goto out_unlock0;
1632		sma->sem_ctime = ktime_get_real_seconds();
1633		break;
1634	default:
1635		err = -EINVAL;
1636		goto out_unlock1;
1637	}
1638
1639out_unlock0:
1640	sem_unlock(sma, -1);
1641out_unlock1:
1642	rcu_read_unlock();
1643out_up:
1644	up_write(&sem_ids(ns).rwsem);
1645	return err;
1646}
1647
1648static long ksys_semctl(int semid, int semnum, int cmd, unsigned long arg, int version)
1649{
1650	struct ipc_namespace *ns;
1651	void __user *p = (void __user *)arg;
1652	struct semid64_ds semid64;
1653	int err;
1654
1655	if (semid < 0)
1656		return -EINVAL;
1657
1658	ns = current->nsproxy->ipc_ns;
1659
1660	switch (cmd) {
1661	case IPC_INFO:
1662	case SEM_INFO:
1663		return semctl_info(ns, semid, cmd, p);
1664	case IPC_STAT:
1665	case SEM_STAT:
1666	case SEM_STAT_ANY:
1667		err = semctl_stat(ns, semid, cmd, &semid64);
1668		if (err < 0)
1669			return err;
1670		if (copy_semid_to_user(p, &semid64, version))
1671			err = -EFAULT;
1672		return err;
1673	case GETALL:
1674	case GETVAL:
1675	case GETPID:
1676	case GETNCNT:
1677	case GETZCNT:
1678	case SETALL:
1679		return semctl_main(ns, semid, semnum, cmd, p);
1680	case SETVAL: {
1681		int val;
1682#if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN)
1683		/* big-endian 64bit */
1684		val = arg >> 32;
1685#else
1686		/* 32bit or little-endian 64bit */
1687		val = arg;
1688#endif
1689		return semctl_setval(ns, semid, semnum, val);
1690	}
1691	case IPC_SET:
1692		if (copy_semid_from_user(&semid64, p, version))
1693			return -EFAULT;
1694		fallthrough;
1695	case IPC_RMID:
1696		return semctl_down(ns, semid, cmd, &semid64);
1697	default:
1698		return -EINVAL;
1699	}
1700}
1701
1702SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, unsigned long, arg)
1703{
1704	return ksys_semctl(semid, semnum, cmd, arg, IPC_64);
1705}
1706
1707#ifdef CONFIG_ARCH_WANT_IPC_PARSE_VERSION
1708long ksys_old_semctl(int semid, int semnum, int cmd, unsigned long arg)
1709{
1710	int version = ipc_parse_version(&cmd);
1711
1712	return ksys_semctl(semid, semnum, cmd, arg, version);
1713}
1714
1715SYSCALL_DEFINE4(old_semctl, int, semid, int, semnum, int, cmd, unsigned long, arg)
1716{
1717	return ksys_old_semctl(semid, semnum, cmd, arg);
1718}
1719#endif
1720
1721#ifdef CONFIG_COMPAT
1722
1723struct compat_semid_ds {
1724	struct compat_ipc_perm sem_perm;
1725	old_time32_t sem_otime;
1726	old_time32_t sem_ctime;
1727	compat_uptr_t sem_base;
1728	compat_uptr_t sem_pending;
1729	compat_uptr_t sem_pending_last;
1730	compat_uptr_t undo;
1731	unsigned short sem_nsems;
1732};
1733
1734static int copy_compat_semid_from_user(struct semid64_ds *out, void __user *buf,
1735					int version)
1736{
1737	memset(out, 0, sizeof(*out));
1738	if (version == IPC_64) {
1739		struct compat_semid64_ds __user *p = buf;
1740		return get_compat_ipc64_perm(&out->sem_perm, &p->sem_perm);
1741	} else {
1742		struct compat_semid_ds __user *p = buf;
1743		return get_compat_ipc_perm(&out->sem_perm, &p->sem_perm);
1744	}
1745}
1746
1747static int copy_compat_semid_to_user(void __user *buf, struct semid64_ds *in,
1748					int version)
1749{
1750	if (version == IPC_64) {
1751		struct compat_semid64_ds v;
1752		memset(&v, 0, sizeof(v));
1753		to_compat_ipc64_perm(&v.sem_perm, &in->sem_perm);
1754		v.sem_otime	 = lower_32_bits(in->sem_otime);
1755		v.sem_otime_high = upper_32_bits(in->sem_otime);
1756		v.sem_ctime	 = lower_32_bits(in->sem_ctime);
1757		v.sem_ctime_high = upper_32_bits(in->sem_ctime);
1758		v.sem_nsems = in->sem_nsems;
1759		return copy_to_user(buf, &v, sizeof(v));
1760	} else {
1761		struct compat_semid_ds v;
1762		memset(&v, 0, sizeof(v));
1763		to_compat_ipc_perm(&v.sem_perm, &in->sem_perm);
1764		v.sem_otime = in->sem_otime;
1765		v.sem_ctime = in->sem_ctime;
1766		v.sem_nsems = in->sem_nsems;
1767		return copy_to_user(buf, &v, sizeof(v));
1768	}
1769}
1770
1771static long compat_ksys_semctl(int semid, int semnum, int cmd, int arg, int version)
1772{
1773	void __user *p = compat_ptr(arg);
1774	struct ipc_namespace *ns;
1775	struct semid64_ds semid64;
1776	int err;
1777
1778	ns = current->nsproxy->ipc_ns;
1779
1780	if (semid < 0)
1781		return -EINVAL;
1782
1783	switch (cmd & (~IPC_64)) {
1784	case IPC_INFO:
1785	case SEM_INFO:
1786		return semctl_info(ns, semid, cmd, p);
1787	case IPC_STAT:
1788	case SEM_STAT:
1789	case SEM_STAT_ANY:
1790		err = semctl_stat(ns, semid, cmd, &semid64);
1791		if (err < 0)
1792			return err;
1793		if (copy_compat_semid_to_user(p, &semid64, version))
1794			err = -EFAULT;
1795		return err;
1796	case GETVAL:
1797	case GETPID:
1798	case GETNCNT:
1799	case GETZCNT:
1800	case GETALL:
1801	case SETALL:
1802		return semctl_main(ns, semid, semnum, cmd, p);
1803	case SETVAL:
1804		return semctl_setval(ns, semid, semnum, arg);
1805	case IPC_SET:
1806		if (copy_compat_semid_from_user(&semid64, p, version))
1807			return -EFAULT;
1808		fallthrough;
1809	case IPC_RMID:
1810		return semctl_down(ns, semid, cmd, &semid64);
1811	default:
1812		return -EINVAL;
1813	}
1814}
1815
1816COMPAT_SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, int, arg)
1817{
1818	return compat_ksys_semctl(semid, semnum, cmd, arg, IPC_64);
1819}
1820
1821#ifdef CONFIG_ARCH_WANT_COMPAT_IPC_PARSE_VERSION
1822long compat_ksys_old_semctl(int semid, int semnum, int cmd, int arg)
1823{
1824	int version = compat_ipc_parse_version(&cmd);
1825
1826	return compat_ksys_semctl(semid, semnum, cmd, arg, version);
1827}
1828
1829COMPAT_SYSCALL_DEFINE4(old_semctl, int, semid, int, semnum, int, cmd, int, arg)
1830{
1831	return compat_ksys_old_semctl(semid, semnum, cmd, arg);
1832}
1833#endif
1834#endif
1835
1836/* If the task doesn't already have a undo_list, then allocate one
1837 * here.  We guarantee there is only one thread using this undo list,
1838 * and current is THE ONE
1839 *
1840 * If this allocation and assignment succeeds, but later
1841 * portions of this code fail, there is no need to free the sem_undo_list.
1842 * Just let it stay associated with the task, and it'll be freed later
1843 * at exit time.
1844 *
1845 * This can block, so callers must hold no locks.
1846 */
1847static inline int get_undo_list(struct sem_undo_list **undo_listp)
1848{
1849	struct sem_undo_list *undo_list;
1850
1851	undo_list = current->sysvsem.undo_list;
1852	if (!undo_list) {
1853		undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
1854		if (undo_list == NULL)
1855			return -ENOMEM;
1856		spin_lock_init(&undo_list->lock);
1857		refcount_set(&undo_list->refcnt, 1);
1858		INIT_LIST_HEAD(&undo_list->list_proc);
1859
1860		current->sysvsem.undo_list = undo_list;
1861	}
1862	*undo_listp = undo_list;
1863	return 0;
1864}
1865
1866static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid)
1867{
1868	struct sem_undo *un;
1869
1870	list_for_each_entry_rcu(un, &ulp->list_proc, list_proc,
1871				spin_is_locked(&ulp->lock)) {
1872		if (un->semid == semid)
1873			return un;
1874	}
1875	return NULL;
1876}
1877
1878static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1879{
1880	struct sem_undo *un;
1881
1882	assert_spin_locked(&ulp->lock);
1883
1884	un = __lookup_undo(ulp, semid);
1885	if (un) {
1886		list_del_rcu(&un->list_proc);
1887		list_add_rcu(&un->list_proc, &ulp->list_proc);
1888	}
1889	return un;
1890}
1891
1892/**
1893 * find_alloc_undo - lookup (and if not present create) undo array
1894 * @ns: namespace
1895 * @semid: semaphore array id
1896 *
1897 * The function looks up (and if not present creates) the undo structure.
1898 * The size of the undo structure depends on the size of the semaphore
1899 * array, thus the alloc path is not that straightforward.
1900 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1901 * performs a rcu_read_lock().
1902 */
1903static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
1904{
1905	struct sem_array *sma;
1906	struct sem_undo_list *ulp;
1907	struct sem_undo *un, *new;
1908	int nsems, error;
1909
1910	error = get_undo_list(&ulp);
1911	if (error)
1912		return ERR_PTR(error);
1913
1914	rcu_read_lock();
1915	spin_lock(&ulp->lock);
1916	un = lookup_undo(ulp, semid);
1917	spin_unlock(&ulp->lock);
1918	if (likely(un != NULL))
1919		goto out;
1920
1921	/* no undo structure around - allocate one. */
1922	/* step 1: figure out the size of the semaphore array */
1923	sma = sem_obtain_object_check(ns, semid);
1924	if (IS_ERR(sma)) {
1925		rcu_read_unlock();
1926		return ERR_CAST(sma);
1927	}
1928
1929	nsems = sma->sem_nsems;
1930	if (!ipc_rcu_getref(&sma->sem_perm)) {
1931		rcu_read_unlock();
1932		un = ERR_PTR(-EIDRM);
1933		goto out;
1934	}
1935	rcu_read_unlock();
1936
1937	/* step 2: allocate new undo structure */
1938	new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1939	if (!new) {
1940		ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
1941		return ERR_PTR(-ENOMEM);
1942	}
1943
1944	/* step 3: Acquire the lock on semaphore array */
1945	rcu_read_lock();
1946	sem_lock_and_putref(sma);
1947	if (!ipc_valid_object(&sma->sem_perm)) {
1948		sem_unlock(sma, -1);
1949		rcu_read_unlock();
1950		kfree(new);
1951		un = ERR_PTR(-EIDRM);
1952		goto out;
1953	}
1954	spin_lock(&ulp->lock);
1955
1956	/*
1957	 * step 4: check for races: did someone else allocate the undo struct?
1958	 */
1959	un = lookup_undo(ulp, semid);
1960	if (un) {
1961		kfree(new);
1962		goto success;
1963	}
1964	/* step 5: initialize & link new undo structure */
1965	new->semadj = (short *) &new[1];
1966	new->ulp = ulp;
1967	new->semid = semid;
1968	assert_spin_locked(&ulp->lock);
1969	list_add_rcu(&new->list_proc, &ulp->list_proc);
1970	ipc_assert_locked_object(&sma->sem_perm);
1971	list_add(&new->list_id, &sma->list_id);
1972	un = new;
1973
1974success:
1975	spin_unlock(&ulp->lock);
1976	sem_unlock(sma, -1);
1977out:
1978	return un;
1979}
1980
1981static long do_semtimedop(int semid, struct sembuf __user *tsops,
1982		unsigned nsops, const struct timespec64 *timeout)
1983{
1984	int error = -EINVAL;
1985	struct sem_array *sma;
1986	struct sembuf fast_sops[SEMOPM_FAST];
1987	struct sembuf *sops = fast_sops, *sop;
1988	struct sem_undo *un;
1989	int max, locknum;
1990	bool undos = false, alter = false, dupsop = false;
1991	struct sem_queue queue;
1992	unsigned long dup = 0, jiffies_left = 0;
1993	struct ipc_namespace *ns;
1994
1995	ns = current->nsproxy->ipc_ns;
1996
1997	if (nsops < 1 || semid < 0)
1998		return -EINVAL;
1999	if (nsops > ns->sc_semopm)
2000		return -E2BIG;
2001	if (nsops > SEMOPM_FAST) {
2002		sops = kvmalloc_array(nsops, sizeof(*sops), GFP_KERNEL);
2003		if (sops == NULL)
2004			return -ENOMEM;
2005	}
2006
2007	if (copy_from_user(sops, tsops, nsops * sizeof(*tsops))) {
2008		error =  -EFAULT;
2009		goto out_free;
2010	}
2011
2012	if (timeout) {
2013		if (timeout->tv_sec < 0 || timeout->tv_nsec < 0 ||
2014			timeout->tv_nsec >= 1000000000L) {
2015			error = -EINVAL;
2016			goto out_free;
2017		}
2018		jiffies_left = timespec64_to_jiffies(timeout);
2019	}
2020
2021	max = 0;
2022	for (sop = sops; sop < sops + nsops; sop++) {
2023		unsigned long mask = 1ULL << ((sop->sem_num) % BITS_PER_LONG);
2024
2025		if (sop->sem_num >= max)
2026			max = sop->sem_num;
2027		if (sop->sem_flg & SEM_UNDO)
2028			undos = true;
2029		if (dup & mask) {
2030			/*
2031			 * There was a previous alter access that appears
2032			 * to have accessed the same semaphore, thus use
2033			 * the dupsop logic. "appears", because the detection
2034			 * can only check % BITS_PER_LONG.
2035			 */
2036			dupsop = true;
2037		}
2038		if (sop->sem_op != 0) {
2039			alter = true;
2040			dup |= mask;
2041		}
2042	}
2043
2044	if (undos) {
2045		/* On success, find_alloc_undo takes the rcu_read_lock */
2046		un = find_alloc_undo(ns, semid);
2047		if (IS_ERR(un)) {
2048			error = PTR_ERR(un);
2049			goto out_free;
2050		}
2051	} else {
2052		un = NULL;
2053		rcu_read_lock();
2054	}
2055
2056	sma = sem_obtain_object_check(ns, semid);
2057	if (IS_ERR(sma)) {
2058		rcu_read_unlock();
2059		error = PTR_ERR(sma);
2060		goto out_free;
2061	}
2062
2063	error = -EFBIG;
2064	if (max >= sma->sem_nsems) {
2065		rcu_read_unlock();
2066		goto out_free;
2067	}
2068
2069	error = -EACCES;
2070	if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO)) {
2071		rcu_read_unlock();
2072		goto out_free;
2073	}
2074
2075	error = security_sem_semop(&sma->sem_perm, sops, nsops, alter);
2076	if (error) {
2077		rcu_read_unlock();
2078		goto out_free;
2079	}
2080
2081	error = -EIDRM;
2082	locknum = sem_lock(sma, sops, nsops);
2083	/*
2084	 * We eventually might perform the following check in a lockless
2085	 * fashion, considering ipc_valid_object() locking constraints.
2086	 * If nsops == 1 and there is no contention for sem_perm.lock, then
2087	 * only a per-semaphore lock is held and it's OK to proceed with the
2088	 * check below. More details on the fine grained locking scheme
2089	 * entangled here and why it's RMID race safe on comments at sem_lock()
2090	 */
2091	if (!ipc_valid_object(&sma->sem_perm))
2092		goto out_unlock_free;
2093	/*
2094	 * semid identifiers are not unique - find_alloc_undo may have
2095	 * allocated an undo structure, it was invalidated by an RMID
2096	 * and now a new array with received the same id. Check and fail.
2097	 * This case can be detected checking un->semid. The existence of
2098	 * "un" itself is guaranteed by rcu.
2099	 */
2100	if (un && un->semid == -1)
2101		goto out_unlock_free;
2102
2103	queue.sops = sops;
2104	queue.nsops = nsops;
2105	queue.undo = un;
2106	queue.pid = task_tgid(current);
2107	queue.alter = alter;
2108	queue.dupsop = dupsop;
2109
2110	error = perform_atomic_semop(sma, &queue);
2111	if (error == 0) { /* non-blocking successful path */
2112		DEFINE_WAKE_Q(wake_q);
2113
2114		/*
2115		 * If the operation was successful, then do
2116		 * the required updates.
2117		 */
2118		if (alter)
2119			do_smart_update(sma, sops, nsops, 1, &wake_q);
2120		else
2121			set_semotime(sma, sops);
2122
2123		sem_unlock(sma, locknum);
2124		rcu_read_unlock();
2125		wake_up_q(&wake_q);
2126
2127		goto out_free;
2128	}
2129	if (error < 0) /* non-blocking error path */
2130		goto out_unlock_free;
2131
2132	/*
2133	 * We need to sleep on this operation, so we put the current
2134	 * task into the pending queue and go to sleep.
2135	 */
2136	if (nsops == 1) {
2137		struct sem *curr;
2138		int idx = array_index_nospec(sops->sem_num, sma->sem_nsems);
2139		curr = &sma->sems[idx];
2140
2141		if (alter) {
2142			if (sma->complex_count) {
2143				list_add_tail(&queue.list,
2144						&sma->pending_alter);
2145			} else {
2146
2147				list_add_tail(&queue.list,
2148						&curr->pending_alter);
2149			}
2150		} else {
2151			list_add_tail(&queue.list, &curr->pending_const);
2152		}
2153	} else {
2154		if (!sma->complex_count)
2155			merge_queues(sma);
2156
2157		if (alter)
2158			list_add_tail(&queue.list, &sma->pending_alter);
2159		else
2160			list_add_tail(&queue.list, &sma->pending_const);
2161
2162		sma->complex_count++;
2163	}
2164
2165	do {
2166		/* memory ordering ensured by the lock in sem_lock() */
2167		WRITE_ONCE(queue.status, -EINTR);
2168		queue.sleeper = current;
2169
2170		/* memory ordering is ensured by the lock in sem_lock() */
2171		__set_current_state(TASK_INTERRUPTIBLE);
2172		sem_unlock(sma, locknum);
2173		rcu_read_unlock();
2174
2175		if (timeout)
2176			jiffies_left = schedule_timeout(jiffies_left);
2177		else
2178			schedule();
2179
2180		/*
2181		 * fastpath: the semop has completed, either successfully or
2182		 * not, from the syscall pov, is quite irrelevant to us at this
2183		 * point; we're done.
2184		 *
2185		 * We _do_ care, nonetheless, about being awoken by a signal or
2186		 * spuriously.  The queue.status is checked again in the
2187		 * slowpath (aka after taking sem_lock), such that we can detect
2188		 * scenarios where we were awakened externally, during the
2189		 * window between wake_q_add() and wake_up_q().
2190		 */
2191		error = READ_ONCE(queue.status);
2192		if (error != -EINTR) {
2193			/* see SEM_BARRIER_2 for purpose/pairing */
2194			smp_acquire__after_ctrl_dep();
2195			goto out_free;
2196		}
2197
2198		rcu_read_lock();
2199		locknum = sem_lock(sma, sops, nsops);
2200
2201		if (!ipc_valid_object(&sma->sem_perm))
2202			goto out_unlock_free;
2203
2204		/*
2205		 * No necessity for any barrier: We are protect by sem_lock()
2206		 */
2207		error = READ_ONCE(queue.status);
2208
2209		/*
2210		 * If queue.status != -EINTR we are woken up by another process.
2211		 * Leave without unlink_queue(), but with sem_unlock().
2212		 */
2213		if (error != -EINTR)
2214			goto out_unlock_free;
2215
2216		/*
2217		 * If an interrupt occurred we have to clean up the queue.
2218		 */
2219		if (timeout && jiffies_left == 0)
2220			error = -EAGAIN;
2221	} while (error == -EINTR && !signal_pending(current)); /* spurious */
2222
2223	unlink_queue(sma, &queue);
2224
2225out_unlock_free:
2226	sem_unlock(sma, locknum);
2227	rcu_read_unlock();
2228out_free:
2229	if (sops != fast_sops)
2230		kvfree(sops);
2231	return error;
2232}
2233
2234long ksys_semtimedop(int semid, struct sembuf __user *tsops,
2235		     unsigned int nsops, const struct __kernel_timespec __user *timeout)
2236{
2237	if (timeout) {
2238		struct timespec64 ts;
2239		if (get_timespec64(&ts, timeout))
2240			return -EFAULT;
2241		return do_semtimedop(semid, tsops, nsops, &ts);
2242	}
2243	return do_semtimedop(semid, tsops, nsops, NULL);
2244}
2245
2246SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
2247		unsigned int, nsops, const struct __kernel_timespec __user *, timeout)
2248{
2249	return ksys_semtimedop(semid, tsops, nsops, timeout);
2250}
2251
2252#ifdef CONFIG_COMPAT_32BIT_TIME
2253long compat_ksys_semtimedop(int semid, struct sembuf __user *tsems,
2254			    unsigned int nsops,
2255			    const struct old_timespec32 __user *timeout)
2256{
2257	if (timeout) {
2258		struct timespec64 ts;
2259		if (get_old_timespec32(&ts, timeout))
2260			return -EFAULT;
2261		return do_semtimedop(semid, tsems, nsops, &ts);
2262	}
2263	return do_semtimedop(semid, tsems, nsops, NULL);
2264}
2265
2266SYSCALL_DEFINE4(semtimedop_time32, int, semid, struct sembuf __user *, tsems,
2267		       unsigned int, nsops,
2268		       const struct old_timespec32 __user *, timeout)
2269{
2270	return compat_ksys_semtimedop(semid, tsems, nsops, timeout);
2271}
2272#endif
2273
2274SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
2275		unsigned, nsops)
2276{
2277	return do_semtimedop(semid, tsops, nsops, NULL);
2278}
2279
2280/* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
2281 * parent and child tasks.
2282 */
2283
2284int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
2285{
2286	struct sem_undo_list *undo_list;
2287	int error;
2288
2289	if (clone_flags & CLONE_SYSVSEM) {
2290		error = get_undo_list(&undo_list);
2291		if (error)
2292			return error;
2293		refcount_inc(&undo_list->refcnt);
2294		tsk->sysvsem.undo_list = undo_list;
2295	} else
2296		tsk->sysvsem.undo_list = NULL;
2297
2298	return 0;
2299}
2300
2301/*
2302 * add semadj values to semaphores, free undo structures.
2303 * undo structures are not freed when semaphore arrays are destroyed
2304 * so some of them may be out of date.
2305 * IMPLEMENTATION NOTE: There is some confusion over whether the
2306 * set of adjustments that needs to be done should be done in an atomic
2307 * manner or not. That is, if we are attempting to decrement the semval
2308 * should we queue up and wait until we can do so legally?
2309 * The original implementation attempted to do this (queue and wait).
2310 * The current implementation does not do so. The POSIX standard
2311 * and SVID should be consulted to determine what behavior is mandated.
2312 */
2313void exit_sem(struct task_struct *tsk)
2314{
2315	struct sem_undo_list *ulp;
2316
2317	ulp = tsk->sysvsem.undo_list;
2318	if (!ulp)
2319		return;
2320	tsk->sysvsem.undo_list = NULL;
2321
2322	if (!refcount_dec_and_test(&ulp->refcnt))
2323		return;
2324
2325	for (;;) {
2326		struct sem_array *sma;
2327		struct sem_undo *un;
2328		int semid, i;
2329		DEFINE_WAKE_Q(wake_q);
2330
2331		cond_resched();
2332
2333		rcu_read_lock();
2334		un = list_entry_rcu(ulp->list_proc.next,
2335				    struct sem_undo, list_proc);
2336		if (&un->list_proc == &ulp->list_proc) {
2337			/*
2338			 * We must wait for freeary() before freeing this ulp,
2339			 * in case we raced with last sem_undo. There is a small
2340			 * possibility where we exit while freeary() didn't
2341			 * finish unlocking sem_undo_list.
2342			 */
2343			spin_lock(&ulp->lock);
2344			spin_unlock(&ulp->lock);
2345			rcu_read_unlock();
2346			break;
2347		}
2348		spin_lock(&ulp->lock);
2349		semid = un->semid;
2350		spin_unlock(&ulp->lock);
2351
2352		/* exit_sem raced with IPC_RMID, nothing to do */
2353		if (semid == -1) {
2354			rcu_read_unlock();
2355			continue;
2356		}
2357
2358		sma = sem_obtain_object_check(tsk->nsproxy->ipc_ns, semid);
2359		/* exit_sem raced with IPC_RMID, nothing to do */
2360		if (IS_ERR(sma)) {
2361			rcu_read_unlock();
2362			continue;
2363		}
2364
2365		sem_lock(sma, NULL, -1);
2366		/* exit_sem raced with IPC_RMID, nothing to do */
2367		if (!ipc_valid_object(&sma->sem_perm)) {
2368			sem_unlock(sma, -1);
2369			rcu_read_unlock();
2370			continue;
2371		}
2372		un = __lookup_undo(ulp, semid);
2373		if (un == NULL) {
2374			/* exit_sem raced with IPC_RMID+semget() that created
2375			 * exactly the same semid. Nothing to do.
2376			 */
2377			sem_unlock(sma, -1);
2378			rcu_read_unlock();
2379			continue;
2380		}
2381
2382		/* remove un from the linked lists */
2383		ipc_assert_locked_object(&sma->sem_perm);
2384		list_del(&un->list_id);
2385
2386		spin_lock(&ulp->lock);
2387		list_del_rcu(&un->list_proc);
2388		spin_unlock(&ulp->lock);
2389
2390		/* perform adjustments registered in un */
2391		for (i = 0; i < sma->sem_nsems; i++) {
2392			struct sem *semaphore = &sma->sems[i];
2393			if (un->semadj[i]) {
2394				semaphore->semval += un->semadj[i];
2395				/*
2396				 * Range checks of the new semaphore value,
2397				 * not defined by sus:
2398				 * - Some unices ignore the undo entirely
2399				 *   (e.g. HP UX 11i 11.22, Tru64 V5.1)
2400				 * - some cap the value (e.g. FreeBSD caps
2401				 *   at 0, but doesn't enforce SEMVMX)
2402				 *
2403				 * Linux caps the semaphore value, both at 0
2404				 * and at SEMVMX.
2405				 *
2406				 *	Manfred <manfred@colorfullife.com>
2407				 */
2408				if (semaphore->semval < 0)
2409					semaphore->semval = 0;
2410				if (semaphore->semval > SEMVMX)
2411					semaphore->semval = SEMVMX;
2412				ipc_update_pid(&semaphore->sempid, task_tgid(current));
2413			}
2414		}
2415		/* maybe some queued-up processes were waiting for this */
2416		do_smart_update(sma, NULL, 0, 1, &wake_q);
2417		sem_unlock(sma, -1);
2418		rcu_read_unlock();
2419		wake_up_q(&wake_q);
2420
2421		kfree_rcu(un, rcu);
2422	}
2423	kfree(ulp);
2424}
2425
2426#ifdef CONFIG_PROC_FS
2427static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
2428{
2429	struct user_namespace *user_ns = seq_user_ns(s);
2430	struct kern_ipc_perm *ipcp = it;
2431	struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
2432	time64_t sem_otime;
2433
2434	/*
2435	 * The proc interface isn't aware of sem_lock(), it calls
2436	 * ipc_lock_object() directly (in sysvipc_find_ipc).
2437	 * In order to stay compatible with sem_lock(), we must
2438	 * enter / leave complex_mode.
2439	 */
2440	complexmode_enter(sma);
2441
2442	sem_otime = get_semotime(sma);
2443
2444	seq_printf(s,
2445		   "%10d %10d  %4o %10u %5u %5u %5u %5u %10llu %10llu\n",
2446		   sma->sem_perm.key,
2447		   sma->sem_perm.id,
2448		   sma->sem_perm.mode,
2449		   sma->sem_nsems,
2450		   from_kuid_munged(user_ns, sma->sem_perm.uid),
2451		   from_kgid_munged(user_ns, sma->sem_perm.gid),
2452		   from_kuid_munged(user_ns, sma->sem_perm.cuid),
2453		   from_kgid_munged(user_ns, sma->sem_perm.cgid),
2454		   sem_otime,
2455		   sma->sem_ctime);
2456
2457	complexmode_tryleave(sma);
2458
2459	return 0;
2460}
2461#endif
2462