Linux Audio

Check our new training course

Loading...
v3.5.6
 
   1/*
   2 * linux/ipc/sem.c
   3 * Copyright (C) 1992 Krishna Balasubramanian
   4 * Copyright (C) 1995 Eric Schenk, Bruno Haible
   5 *
   6 * /proc/sysvipc/sem support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
   7 *
   8 * SMP-threaded, sysctl's added
   9 * (c) 1999 Manfred Spraul <manfred@colorfullife.com>
  10 * Enforced range limit on SEM_UNDO
  11 * (c) 2001 Red Hat Inc
  12 * Lockless wakeup
  13 * (c) 2003 Manfred Spraul <manfred@colorfullife.com>
 
  14 * Further wakeup optimizations, documentation
  15 * (c) 2010 Manfred Spraul <manfred@colorfullife.com>
  16 *
  17 * support for audit of ipc object properties and permission changes
  18 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
  19 *
  20 * namespaces support
  21 * OpenVZ, SWsoft Inc.
  22 * Pavel Emelianov <xemul@openvz.org>
  23 *
  24 * Implementation notes: (May 2010)
  25 * This file implements System V semaphores.
  26 *
  27 * User space visible behavior:
  28 * - FIFO ordering for semop() operations (just FIFO, not starvation
  29 *   protection)
  30 * - multiple semaphore operations that alter the same semaphore in
  31 *   one semop() are handled.
  32 * - sem_ctime (time of last semctl()) is updated in the IPC_SET, SETVAL and
  33 *   SETALL calls.
  34 * - two Linux specific semctl() commands: SEM_STAT, SEM_INFO.
  35 * - undo adjustments at process exit are limited to 0..SEMVMX.
  36 * - namespace are supported.
  37 * - SEMMSL, SEMMNS, SEMOPM and SEMMNI can be configured at runtine by writing
  38 *   to /proc/sys/kernel/sem.
  39 * - statistics about the usage are reported in /proc/sysvipc/sem.
  40 *
  41 * Internals:
  42 * - scalability:
  43 *   - all global variables are read-mostly.
  44 *   - semop() calls and semctl(RMID) are synchronized by RCU.
  45 *   - most operations do write operations (actually: spin_lock calls) to
  46 *     the per-semaphore array structure.
  47 *   Thus: Perfect SMP scaling between independent semaphore arrays.
  48 *         If multiple semaphores in one array are used, then cache line
  49 *         trashing on the semaphore array spinlock will limit the scaling.
  50 * - semncnt and semzcnt are calculated on demand in count_semncnt() and
  51 *   count_semzcnt()
  52 * - the task that performs a successful semop() scans the list of all
  53 *   sleeping tasks and completes any pending operations that can be fulfilled.
  54 *   Semaphores are actively given to waiting tasks (necessary for FIFO).
  55 *   (see update_queue())
  56 * - To improve the scalability, the actual wake-up calls are performed after
  57 *   dropping all locks. (see wake_up_sem_queue_prepare(),
  58 *   wake_up_sem_queue_do())
  59 * - All work is done by the waker, the woken up task does not have to do
  60 *   anything - not even acquiring a lock or dropping a refcount.
  61 * - A woken up task may not even touch the semaphore array anymore, it may
  62 *   have been destroyed already by a semctl(RMID).
  63 * - The synchronizations between wake-ups due to a timeout/signal and a
  64 *   wake-up due to a completed semaphore operation is achieved by using an
  65 *   intermediate state (IN_WAKEUP).
  66 * - UNDO values are stored in an array (one per process and per
  67 *   semaphore array, lazily allocated). For backwards compatibility, multiple
  68 *   modes for the UNDO variables are supported (per process, per thread)
  69 *   (see copy_semundo, CLONE_SYSVSEM)
  70 * - There are two lists of the pending operations: a per-array list
  71 *   and per-semaphore list (stored in the array). This allows to achieve FIFO
  72 *   ordering without always scanning all pending operations.
  73 *   The worst-case behavior is nevertheless O(N^2) for N wakeups.
  74 */
  75
 
  76#include <linux/slab.h>
  77#include <linux/spinlock.h>
  78#include <linux/init.h>
  79#include <linux/proc_fs.h>
  80#include <linux/time.h>
  81#include <linux/security.h>
  82#include <linux/syscalls.h>
  83#include <linux/audit.h>
  84#include <linux/capability.h>
  85#include <linux/seq_file.h>
  86#include <linux/rwsem.h>
  87#include <linux/nsproxy.h>
  88#include <linux/ipc_namespace.h>
 
 
 
  89
  90#include <asm/uaccess.h>
  91#include "util.h"
  92
  93/* One semaphore structure for each semaphore in the system. */
  94struct sem {
  95	int	semval;		/* current value */
  96	int	sempid;		/* pid of last operation */
  97	struct list_head sem_pending; /* pending single-sop operations */
  98};
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  99
 100/* One queue for each sleeping process in the system. */
 101struct sem_queue {
 102	struct list_head	simple_list; /* queue of pending operations */
 103	struct list_head	list;	 /* queue of pending operations */
 104	struct task_struct	*sleeper; /* this process */
 105	struct sem_undo		*undo;	 /* undo structure */
 106	int			pid;	 /* process id of requesting process */
 107	int			status;	 /* completion status of operation */
 108	struct sembuf		*sops;	 /* array of pending operations */
 
 109	int			nsops;	 /* number of operations */
 110	int			alter;	 /* does *sops alter the array? */
 
 111};
 112
 113/* Each task has a list of undo requests. They are executed automatically
 114 * when the process exits.
 115 */
 116struct sem_undo {
 117	struct list_head	list_proc;	/* per-process list: *
 118						 * all undos from one process
 119						 * rcu protected */
 120	struct rcu_head		rcu;		/* rcu struct for sem_undo */
 121	struct sem_undo_list	*ulp;		/* back ptr to sem_undo_list */
 122	struct list_head	list_id;	/* per semaphore array list:
 123						 * all undos for one array */
 124	int			semid;		/* semaphore set identifier */
 125	short			*semadj;	/* array of adjustments */
 126						/* one per semaphore */
 127};
 128
 129/* sem_undo_list controls shared access to the list of sem_undo structures
 130 * that may be shared among all a CLONE_SYSVSEM task group.
 131 */
 132struct sem_undo_list {
 133	atomic_t		refcnt;
 134	spinlock_t		lock;
 135	struct list_head	list_proc;
 136};
 137
 138
 139#define sem_ids(ns)	((ns)->ids[IPC_SEM_IDS])
 140
 141#define sem_unlock(sma)		ipc_unlock(&(sma)->sem_perm)
 142#define sem_checkid(sma, semid)	ipc_checkid(&sma->sem_perm, semid)
 143
 144static int newary(struct ipc_namespace *, struct ipc_params *);
 145static void freeary(struct ipc_namespace *, struct kern_ipc_perm *);
 146#ifdef CONFIG_PROC_FS
 147static int sysvipc_sem_proc_show(struct seq_file *s, void *it);
 148#endif
 149
 150#define SEMMSL_FAST	256 /* 512 bytes on stack */
 151#define SEMOPM_FAST	64  /* ~ 372 bytes on stack */
 152
 153/*
 154 * linked list protection:
 
 
 
 
 
 
 
 
 155 *	sem_undo.id_next,
 156 *	sem_array.sem_pending{,last},
 157 *	sem_array.sem_undo: sem_lock() for read/write
 158 *	sem_undo.proc_next: only "current" is allowed to read/write that field.
 159 *	
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 160 */
 161
 162#define sc_semmsl	sem_ctls[0]
 163#define sc_semmns	sem_ctls[1]
 164#define sc_semopm	sem_ctls[2]
 165#define sc_semmni	sem_ctls[3]
 166
 167void sem_init_ns(struct ipc_namespace *ns)
 168{
 169	ns->sc_semmsl = SEMMSL;
 170	ns->sc_semmns = SEMMNS;
 171	ns->sc_semopm = SEMOPM;
 172	ns->sc_semmni = SEMMNI;
 173	ns->used_sems = 0;
 174	ipc_init_ids(&ns->ids[IPC_SEM_IDS]);
 175}
 176
 177#ifdef CONFIG_IPC_NS
 178void sem_exit_ns(struct ipc_namespace *ns)
 179{
 180	free_ipcs(ns, &sem_ids(ns), freeary);
 181	idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr);
 
 182}
 183#endif
 184
 185void __init sem_init (void)
 186{
 187	sem_init_ns(&init_ipc_ns);
 188	ipc_init_proc_interface("sysvipc/sem",
 189				"       key      semid perms      nsems   uid   gid  cuid  cgid      otime      ctime\n",
 190				IPC_SEM_IDS, sysvipc_sem_proc_show);
 191}
 192
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 193/*
 194 * sem_lock_(check_) routines are called in the paths where the rw_mutex
 195 * is not held.
 
 
 196 */
 197static inline struct sem_array *sem_lock(struct ipc_namespace *ns, int id)
 198{
 199	struct kern_ipc_perm *ipcp = ipc_lock(&sem_ids(ns), id);
 200
 201	if (IS_ERR(ipcp))
 202		return (struct sem_array *)ipcp;
 203
 204	return container_of(ipcp, struct sem_array, sem_perm);
 205}
 206
 207static inline struct sem_array *sem_lock_check(struct ipc_namespace *ns,
 208						int id)
 209{
 210	struct kern_ipc_perm *ipcp = ipc_lock_check(&sem_ids(ns), id);
 211
 212	if (IS_ERR(ipcp))
 213		return (struct sem_array *)ipcp;
 214
 215	return container_of(ipcp, struct sem_array, sem_perm);
 216}
 217
 218static inline void sem_lock_and_putref(struct sem_array *sma)
 219{
 220	ipc_lock_by_ptr(&sma->sem_perm);
 221	ipc_rcu_putref(sma);
 222}
 223
 224static inline void sem_getref_and_unlock(struct sem_array *sma)
 225{
 226	ipc_rcu_getref(sma);
 227	ipc_unlock(&(sma)->sem_perm);
 228}
 229
 230static inline void sem_putref(struct sem_array *sma)
 231{
 232	ipc_lock_by_ptr(&sma->sem_perm);
 233	ipc_rcu_putref(sma);
 234	ipc_unlock(&(sma)->sem_perm);
 235}
 236
 237static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
 238{
 239	ipc_rmid(&sem_ids(ns), &s->sem_perm);
 240}
 241
 242/*
 243 * Lockless wakeup algorithm:
 244 * Without the check/retry algorithm a lockless wakeup is possible:
 245 * - queue.status is initialized to -EINTR before blocking.
 246 * - wakeup is performed by
 247 *	* unlinking the queue entry from sma->sem_pending
 248 *	* setting queue.status to IN_WAKEUP
 249 *	  This is the notification for the blocked thread that a
 250 *	  result value is imminent.
 251 *	* call wake_up_process
 252 *	* set queue.status to the final value.
 253 * - the previously blocked thread checks queue.status:
 254 *   	* if it's IN_WAKEUP, then it must wait until the value changes
 255 *   	* if it's not -EINTR, then the operation was completed by
 256 *   	  update_queue. semtimedop can return queue.status without
 257 *   	  performing any operation on the sem array.
 258 *   	* otherwise it must acquire the spinlock and check what's up.
 259 *
 260 * The two-stage algorithm is necessary to protect against the following
 261 * races:
 262 * - if queue.status is set after wake_up_process, then the woken up idle
 263 *   thread could race forward and try (and fail) to acquire sma->lock
 264 *   before update_queue had a chance to set queue.status
 265 * - if queue.status is written before wake_up_process and if the
 266 *   blocked process is woken up by a signal between writing
 267 *   queue.status and the wake_up_process, then the woken up
 268 *   process could return from semtimedop and die by calling
 269 *   sys_exit before wake_up_process is called. Then wake_up_process
 270 *   will oops, because the task structure is already invalid.
 271 *   (yes, this happened on s390 with sysv msg).
 272 *
 273 */
 274#define IN_WAKEUP	1
 275
 276/**
 277 * newary - Create a new semaphore set
 278 * @ns: namespace
 279 * @params: ptr to the structure that contains key, semflg and nsems
 280 *
 281 * Called with sem_ids.rw_mutex held (as a writer)
 282 */
 283
 284static int newary(struct ipc_namespace *ns, struct ipc_params *params)
 285{
 286	int id;
 287	int retval;
 288	struct sem_array *sma;
 289	int size;
 290	key_t key = params->key;
 291	int nsems = params->u.nsems;
 292	int semflg = params->flg;
 293	int i;
 294
 295	if (!nsems)
 296		return -EINVAL;
 297	if (ns->used_sems + nsems > ns->sc_semmns)
 298		return -ENOSPC;
 299
 300	size = sizeof (*sma) + nsems * sizeof (struct sem);
 301	sma = ipc_rcu_alloc(size);
 302	if (!sma) {
 303		return -ENOMEM;
 304	}
 305	memset (sma, 0, size);
 306
 307	sma->sem_perm.mode = (semflg & S_IRWXUGO);
 308	sma->sem_perm.key = key;
 309
 310	sma->sem_perm.security = NULL;
 311	retval = security_sem_alloc(sma);
 312	if (retval) {
 313		ipc_rcu_putref(sma);
 314		return retval;
 315	}
 316
 317	id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
 318	if (id < 0) {
 319		security_sem_free(sma);
 320		ipc_rcu_putref(sma);
 321		return id;
 322	}
 323	ns->used_sems += nsems;
 324
 325	sma->sem_base = (struct sem *) &sma[1];
 326
 327	for (i = 0; i < nsems; i++)
 328		INIT_LIST_HEAD(&sma->sem_base[i].sem_pending);
 329
 330	sma->complex_count = 0;
 331	INIT_LIST_HEAD(&sma->sem_pending);
 
 
 332	INIT_LIST_HEAD(&sma->list_id);
 333	sma->sem_nsems = nsems;
 334	sma->sem_ctime = get_seconds();
 335	sem_unlock(sma);
 336
 337	return sma->sem_perm.id;
 338}
 339
 
 
 
 
 
 
 
 340
 341/*
 342 * Called with sem_ids.rw_mutex and ipcp locked.
 343 */
 344static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg)
 345{
 346	struct sem_array *sma;
 347
 348	sma = container_of(ipcp, struct sem_array, sem_perm);
 349	return security_sem_associate(sma, semflg);
 350}
 351
 
 352/*
 353 * Called with sem_ids.rw_mutex and ipcp locked.
 354 */
 355static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
 356				struct ipc_params *params)
 357{
 358	struct sem_array *sma;
 359
 360	sma = container_of(ipcp, struct sem_array, sem_perm);
 361	if (params->u.nsems > sma->sem_nsems)
 362		return -EINVAL;
 363
 364	return 0;
 365}
 366
 367SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg)
 368{
 369	struct ipc_namespace *ns;
 370	struct ipc_ops sem_ops;
 
 
 
 
 371	struct ipc_params sem_params;
 372
 373	ns = current->nsproxy->ipc_ns;
 374
 375	if (nsems < 0 || nsems > ns->sc_semmsl)
 376		return -EINVAL;
 377
 378	sem_ops.getnew = newary;
 379	sem_ops.associate = sem_security;
 380	sem_ops.more_checks = sem_more_checks;
 381
 382	sem_params.key = key;
 383	sem_params.flg = semflg;
 384	sem_params.u.nsems = nsems;
 385
 386	return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
 387}
 388
 389/*
 390 * Determine whether a sequence of semaphore operations would succeed
 391 * all at once. Return 0 if yes, 1 if need to sleep, else return error code.
 392 */
 393
 394static int try_atomic_semop (struct sem_array * sma, struct sembuf * sops,
 395			     int nsops, struct sem_undo *un, int pid)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 396{
 397	int result, sem_op;
 
 398	struct sembuf *sop;
 399	struct sem * curr;
 
 
 
 
 
 
 400
 401	for (sop = sops; sop < sops + nsops; sop++) {
 402		curr = sma->sem_base + sop->sem_num;
 
 403		sem_op = sop->sem_op;
 404		result = curr->semval;
 405  
 406		if (!sem_op && result)
 407			goto would_block;
 408
 409		result += sem_op;
 410		if (result < 0)
 411			goto would_block;
 412		if (result > SEMVMX)
 413			goto out_of_range;
 
 414		if (sop->sem_flg & SEM_UNDO) {
 415			int undo = un->semadj[sop->sem_num] - sem_op;
 416			/*
 417	 		 *	Exceeding the undo range is an error.
 418			 */
 419			if (undo < (-SEMAEM - 1) || undo > SEMAEM)
 420				goto out_of_range;
 
 421		}
 
 422		curr->semval = result;
 423	}
 424
 425	sop--;
 
 426	while (sop >= sops) {
 427		sma->sem_base[sop->sem_num].sempid = pid;
 428		if (sop->sem_flg & SEM_UNDO)
 429			un->semadj[sop->sem_num] -= sop->sem_op;
 430		sop--;
 431	}
 432	
 433	return 0;
 434
 435out_of_range:
 436	result = -ERANGE;
 437	goto undo;
 438
 439would_block:
 
 
 440	if (sop->sem_flg & IPC_NOWAIT)
 441		result = -EAGAIN;
 442	else
 443		result = 1;
 444
 445undo:
 446	sop--;
 447	while (sop >= sops) {
 448		sma->sem_base[sop->sem_num].semval -= sop->sem_op;
 
 
 
 449		sop--;
 450	}
 451
 452	return result;
 453}
 454
 455/** wake_up_sem_queue_prepare(q, error): Prepare wake-up
 456 * @q: queue entry that must be signaled
 457 * @error: Error value for the signal
 458 *
 459 * Prepare the wake-up of the queue entry q.
 460 */
 461static void wake_up_sem_queue_prepare(struct list_head *pt,
 462				struct sem_queue *q, int error)
 463{
 464	if (list_empty(pt)) {
 465		/*
 466		 * Hold preempt off so that we don't get preempted and have the
 467		 * wakee busy-wait until we're scheduled back on.
 468		 */
 469		preempt_disable();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 470	}
 471	q->status = IN_WAKEUP;
 472	q->pid = error;
 473
 474	list_add_tail(&q->simple_list, pt);
 475}
 
 
 476
 477/**
 478 * wake_up_sem_queue_do(pt) - do the actual wake-up
 479 * @pt: list of tasks to be woken up
 480 *
 481 * Do the actual wake-up.
 482 * The function is called without any locks held, thus the semaphore array
 483 * could be destroyed already and the tasks can disappear as soon as the
 484 * status is set to the actual return code.
 485 */
 486static void wake_up_sem_queue_do(struct list_head *pt)
 487{
 488	struct sem_queue *q, *t;
 489	int did_something;
 490
 491	did_something = !list_empty(pt);
 492	list_for_each_entry_safe(q, t, pt, simple_list) {
 493		wake_up_process(q->sleeper);
 494		/* q can disappear immediately after writing q->status. */
 495		smp_wmb();
 496		q->status = q->pid;
 497	}
 498	if (did_something)
 499		preempt_enable();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 500}
 501
 502static void unlink_queue(struct sem_array *sma, struct sem_queue *q)
 503{
 504	list_del(&q->list);
 505	if (q->nsops == 1)
 506		list_del(&q->simple_list);
 507	else
 508		sma->complex_count--;
 509}
 510
 511/** check_restart(sma, q)
 512 * @sma: semaphore array
 513 * @q: the operation that just completed
 514 *
 515 * update_queue is O(N^2) when it restarts scanning the whole queue of
 516 * waiting operations. Therefore this function checks if the restart is
 517 * really necessary. It is called after a previously waiting operation
 518 * was completed.
 
 519 */
 520static int check_restart(struct sem_array *sma, struct sem_queue *q)
 521{
 522	struct sem *curr;
 523	struct sem_queue *h;
 524
 525	/* if the operation didn't modify the array, then no restart */
 526	if (q->alter == 0)
 527		return 0;
 528
 529	/* pending complex operations are too difficult to analyse */
 530	if (sma->complex_count)
 531		return 1;
 532
 533	/* we were a sleeping complex operation. Too difficult */
 534	if (q->nsops > 1)
 535		return 1;
 536
 537	curr = sma->sem_base + q->sops[0].sem_num;
 
 
 
 
 
 
 
 
 
 
 
 
 538
 539	/* No-one waits on this queue */
 540	if (list_empty(&curr->sem_pending))
 541		return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 542
 543	/* the new semaphore value */
 544	if (curr->semval) {
 545		/* It is impossible that someone waits for the new value:
 546		 * - q is a previously sleeping simple operation that
 547		 *   altered the array. It must be a decrement, because
 548		 *   simple increments never sleep.
 549		 * - The value is not 0, thus wait-for-zero won't proceed.
 550		 * - If there are older (higher priority) decrements
 551		 *   in the queue, then they have observed the original
 552		 *   semval value and couldn't proceed. The operation
 553		 *   decremented to value - thus they won't proceed either.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 554		 */
 555		BUG_ON(q->sops[0].sem_op >= 0);
 556		return 0;
 
 
 
 
 557	}
 558	/*
 559	 * semval is 0. Check if there are wait-for-zero semops.
 560	 * They must be the first entries in the per-semaphore simple queue
 561	 */
 562	h = list_first_entry(&curr->sem_pending, struct sem_queue, simple_list);
 563	BUG_ON(h->nsops != 1);
 564	BUG_ON(h->sops[0].sem_num != q->sops[0].sem_num);
 565
 566	/* Yes, there is a wait-for-zero semop. Restart */
 567	if (h->sops[0].sem_op == 0)
 568		return 1;
 569
 570	/* Again - no-one is waiting for the new value. */
 571	return 0;
 572}
 573
 574
 575/**
 576 * update_queue(sma, semnum): Look for tasks that can be completed.
 577 * @sma: semaphore array.
 578 * @semnum: semaphore that was modified.
 579 * @pt: list head for the tasks that must be woken up.
 580 *
 581 * update_queue must be called after a semaphore in a semaphore array
 582 * was modified. If multiple semaphore were modified, then @semnum
 583 * must be set to -1.
 584 * The tasks that must be woken up are added to @pt. The return code
 
 585 * is stored in q->pid.
 
 
 586 * The function return 1 if at least one semop was completed successfully.
 587 */
 588static int update_queue(struct sem_array *sma, int semnum, struct list_head *pt)
 589{
 590	struct sem_queue *q;
 591	struct list_head *walk;
 592	struct list_head *pending_list;
 593	int offset;
 594	int semop_completed = 0;
 595
 596	/* if there are complex operations around, then knowing the semaphore
 597	 * that was modified doesn't help us. Assume that multiple semaphores
 598	 * were modified.
 599	 */
 600	if (sma->complex_count)
 601		semnum = -1;
 602
 603	if (semnum == -1) {
 604		pending_list = &sma->sem_pending;
 605		offset = offsetof(struct sem_queue, list);
 606	} else {
 607		pending_list = &sma->sem_base[semnum].sem_pending;
 608		offset = offsetof(struct sem_queue, simple_list);
 609	}
 610
 611again:
 612	walk = pending_list->next;
 613	while (walk != pending_list) {
 614		int error, restart;
 615
 616		q = (struct sem_queue *)((char *)walk - offset);
 617		walk = walk->next;
 618
 619		/* If we are scanning the single sop, per-semaphore list of
 620		 * one semaphore and that semaphore is 0, then it is not
 621		 * necessary to scan the "alter" entries: simple increments
 622		 * that affect only one entry succeed immediately and cannot
 623		 * be in the  per semaphore pending queue, and decrements
 624		 * cannot be successful if the value is already 0.
 625		 */
 626		if (semnum != -1 && sma->sem_base[semnum].semval == 0 &&
 627				q->alter)
 628			break;
 629
 630		error = try_atomic_semop(sma, q->sops, q->nsops,
 631					 q->undo, q->pid);
 632
 633		/* Does q->sleeper still need to sleep? */
 634		if (error > 0)
 635			continue;
 636
 637		unlink_queue(sma, q);
 638
 639		if (error) {
 640			restart = 0;
 641		} else {
 642			semop_completed = 1;
 
 643			restart = check_restart(sma, q);
 644		}
 645
 646		wake_up_sem_queue_prepare(pt, q, error);
 647		if (restart)
 648			goto again;
 649	}
 650	return semop_completed;
 651}
 652
 653/**
 654 * do_smart_update(sma, sops, nsops, otime, pt) - optimized update_queue
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 655 * @sma: semaphore array
 656 * @sops: operations that were performed
 657 * @nsops: number of operations
 658 * @otime: force setting otime
 659 * @pt: list head of the tasks that must be woken up.
 660 *
 661 * do_smart_update() does the required called to update_queue, based on the
 662 * actual changes that were performed on the semaphore array.
 663 * Note that the function does not do the actual wake-up: the caller is
 664 * responsible for calling wake_up_sem_queue_do(@pt).
 665 * It is safe to perform this call after dropping all locks.
 666 */
 667static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops,
 668			int otime, struct list_head *pt)
 669{
 670	int i;
 671
 672	if (sma->complex_count || sops == NULL) {
 673		if (update_queue(sma, -1, pt))
 674			otime = 1;
 675		goto done;
 676	}
 677
 678	for (i = 0; i < nsops; i++) {
 679		if (sops[i].sem_op > 0 ||
 680			(sops[i].sem_op < 0 &&
 681				sma->sem_base[sops[i].sem_num].semval == 0))
 682			if (update_queue(sma, sops[i].sem_num, pt))
 683				otime = 1;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 684	}
 685done:
 686	if (otime)
 687		sma->sem_otime = get_seconds();
 688}
 689
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 690
 691/* The following counts are associated to each semaphore:
 692 *   semncnt        number of tasks waiting on semval being nonzero
 693 *   semzcnt        number of tasks waiting on semval being zero
 694 * This model assumes that a task waits on exactly one semaphore.
 695 * Since semaphore operations are to be performed atomically, tasks actually
 696 * wait on a whole sequence of semaphores simultaneously.
 697 * The counts we return here are a rough approximation, but still
 698 * warrant that semncnt+semzcnt>0 if the task is on the pending queue.
 699 */
 700static int count_semncnt (struct sem_array * sma, ushort semnum)
 701{
 702	int semncnt;
 703	struct sem_queue * q;
 704
 705	semncnt = 0;
 706	list_for_each_entry(q, &sma->sem_pending, list) {
 707		struct sembuf * sops = q->sops;
 708		int nsops = q->nsops;
 709		int i;
 710		for (i = 0; i < nsops; i++)
 711			if (sops[i].sem_num == semnum
 712			    && (sops[i].sem_op < 0)
 713			    && !(sops[i].sem_flg & IPC_NOWAIT))
 714				semncnt++;
 715	}
 716	return semncnt;
 717}
 718
 719static int count_semzcnt (struct sem_array * sma, ushort semnum)
 720{
 721	int semzcnt;
 722	struct sem_queue * q;
 
 723
 724	semzcnt = 0;
 725	list_for_each_entry(q, &sma->sem_pending, list) {
 726		struct sembuf * sops = q->sops;
 727		int nsops = q->nsops;
 728		int i;
 729		for (i = 0; i < nsops; i++)
 730			if (sops[i].sem_num == semnum
 731			    && (sops[i].sem_op == 0)
 732			    && !(sops[i].sem_flg & IPC_NOWAIT))
 733				semzcnt++;
 
 
 734	}
 735	return semzcnt;
 
 
 
 
 
 
 
 
 
 
 736}
 737
 738/* Free a semaphore set. freeary() is called with sem_ids.rw_mutex locked
 739 * as a writer and the spinlock for this semaphore set hold. sem_ids.rw_mutex
 740 * remains locked on exit.
 741 */
 742static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
 743{
 744	struct sem_undo *un, *tu;
 745	struct sem_queue *q, *tq;
 746	struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
 747	struct list_head tasks;
 
 748
 749	/* Free the existing undo structures for this semaphore set.  */
 750	assert_spin_locked(&sma->sem_perm.lock);
 751	list_for_each_entry_safe(un, tu, &sma->list_id, list_id) {
 752		list_del(&un->list_id);
 753		spin_lock(&un->ulp->lock);
 754		un->semid = -1;
 755		list_del_rcu(&un->list_proc);
 756		spin_unlock(&un->ulp->lock);
 757		kfree_rcu(un, rcu);
 758	}
 759
 760	/* Wake up all pending processes and let them fail with EIDRM. */
 761	INIT_LIST_HEAD(&tasks);
 762	list_for_each_entry_safe(q, tq, &sma->sem_pending, list) {
 763		unlink_queue(sma, q);
 764		wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 765	}
 766
 767	/* Remove the semaphore set from the IDR */
 768	sem_rmid(ns, sma);
 769	sem_unlock(sma);
 
 770
 771	wake_up_sem_queue_do(&tasks);
 772	ns->used_sems -= sma->sem_nsems;
 773	security_sem_free(sma);
 774	ipc_rcu_putref(sma);
 775}
 776
 777static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
 778{
 779	switch(version) {
 780	case IPC_64:
 781		return copy_to_user(buf, in, sizeof(*in));
 782	case IPC_OLD:
 783	    {
 784		struct semid_ds out;
 785
 786		memset(&out, 0, sizeof(out));
 787
 788		ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
 789
 790		out.sem_otime	= in->sem_otime;
 791		out.sem_ctime	= in->sem_ctime;
 792		out.sem_nsems	= in->sem_nsems;
 793
 794		return copy_to_user(buf, &out, sizeof(out));
 795	    }
 796	default:
 797		return -EINVAL;
 798	}
 799}
 800
 801static int semctl_nolock(struct ipc_namespace *ns, int semid,
 802			 int cmd, int version, union semun arg)
 803{
 804	int err;
 805	struct sem_array *sma;
 806
 807	switch(cmd) {
 808	case IPC_INFO:
 809	case SEM_INFO:
 810	{
 811		struct seminfo seminfo;
 812		int max_id;
 813
 814		err = security_sem_semctl(NULL, cmd);
 815		if (err)
 816			return err;
 817		
 818		memset(&seminfo,0,sizeof(seminfo));
 819		seminfo.semmni = ns->sc_semmni;
 820		seminfo.semmns = ns->sc_semmns;
 821		seminfo.semmsl = ns->sc_semmsl;
 822		seminfo.semopm = ns->sc_semopm;
 823		seminfo.semvmx = SEMVMX;
 824		seminfo.semmnu = SEMMNU;
 825		seminfo.semmap = SEMMAP;
 826		seminfo.semume = SEMUME;
 827		down_read(&sem_ids(ns).rw_mutex);
 828		if (cmd == SEM_INFO) {
 829			seminfo.semusz = sem_ids(ns).in_use;
 830			seminfo.semaem = ns->used_sems;
 831		} else {
 832			seminfo.semusz = SEMUSZ;
 833			seminfo.semaem = SEMAEM;
 834		}
 835		max_id = ipc_get_maxid(&sem_ids(ns));
 836		up_read(&sem_ids(ns).rw_mutex);
 837		if (copy_to_user (arg.__buf, &seminfo, sizeof(struct seminfo))) 
 838			return -EFAULT;
 839		return (max_id < 0) ? 0: max_id;
 840	}
 841	case IPC_STAT:
 842	case SEM_STAT:
 843	{
 844		struct semid64_ds tbuf;
 845		int id;
 846
 847		if (cmd == SEM_STAT) {
 848			sma = sem_lock(ns, semid);
 849			if (IS_ERR(sma))
 850				return PTR_ERR(sma);
 851			id = sma->sem_perm.id;
 852		} else {
 853			sma = sem_lock_check(ns, semid);
 854			if (IS_ERR(sma))
 855				return PTR_ERR(sma);
 856			id = 0;
 
 
 
 
 
 
 
 
 
 
 
 857		}
 
 858
 
 
 
 
 859		err = -EACCES;
 860		if (ipcperms(ns, &sma->sem_perm, S_IRUGO))
 861			goto out_unlock;
 
 862
 863		err = security_sem_semctl(sma, cmd);
 864		if (err)
 865			goto out_unlock;
 866
 867		memset(&tbuf, 0, sizeof(tbuf));
 868
 869		kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
 870		tbuf.sem_otime  = sma->sem_otime;
 871		tbuf.sem_ctime  = sma->sem_ctime;
 872		tbuf.sem_nsems  = sma->sem_nsems;
 873		sem_unlock(sma);
 874		if (copy_semid_to_user (arg.buf, &tbuf, version))
 875			return -EFAULT;
 876		return id;
 877	}
 878	default:
 879		return -EINVAL;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 880	}
 
 881out_unlock:
 882	sem_unlock(sma);
 883	return err;
 884}
 885
 886static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
 887		int cmd, int version, union semun arg)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 888{
 
 889	struct sem_array *sma;
 890	struct sem* curr;
 891	int err;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 892	ushort fast_sem_io[SEMMSL_FAST];
 893	ushort* sem_io = fast_sem_io;
 894	int nsems;
 895	struct list_head tasks;
 896
 897	sma = sem_lock_check(ns, semid);
 898	if (IS_ERR(sma))
 
 
 899		return PTR_ERR(sma);
 
 900
 901	INIT_LIST_HEAD(&tasks);
 902	nsems = sma->sem_nsems;
 903
 904	err = -EACCES;
 905	if (ipcperms(ns, &sma->sem_perm,
 906			(cmd == SETVAL || cmd == SETALL) ? S_IWUGO : S_IRUGO))
 907		goto out_unlock;
 908
 909	err = security_sem_semctl(sma, cmd);
 910	if (err)
 911		goto out_unlock;
 912
 913	err = -EACCES;
 914	switch (cmd) {
 915	case GETALL:
 916	{
 917		ushort __user *array = arg.array;
 918		int i;
 919
 920		if(nsems > SEMMSL_FAST) {
 921			sem_getref_and_unlock(sma);
 922
 923			sem_io = ipc_alloc(sizeof(ushort)*nsems);
 924			if(sem_io == NULL) {
 925				sem_putref(sma);
 
 
 
 
 
 
 
 
 
 
 926				return -ENOMEM;
 927			}
 928
 
 929			sem_lock_and_putref(sma);
 930			if (sma->sem_perm.deleted) {
 931				sem_unlock(sma);
 932				err = -EIDRM;
 933				goto out_free;
 934			}
 935		}
 936
 937		for (i = 0; i < sma->sem_nsems; i++)
 938			sem_io[i] = sma->sem_base[i].semval;
 939		sem_unlock(sma);
 
 940		err = 0;
 941		if(copy_to_user(array, sem_io, nsems*sizeof(ushort)))
 942			err = -EFAULT;
 943		goto out_free;
 944	}
 945	case SETALL:
 946	{
 947		int i;
 948		struct sem_undo *un;
 949
 950		sem_getref_and_unlock(sma);
 
 
 
 
 951
 952		if(nsems > SEMMSL_FAST) {
 953			sem_io = ipc_alloc(sizeof(ushort)*nsems);
 954			if(sem_io == NULL) {
 955				sem_putref(sma);
 
 956				return -ENOMEM;
 957			}
 958		}
 959
 960		if (copy_from_user (sem_io, arg.array, nsems*sizeof(ushort))) {
 961			sem_putref(sma);
 962			err = -EFAULT;
 963			goto out_free;
 964		}
 965
 966		for (i = 0; i < nsems; i++) {
 967			if (sem_io[i] > SEMVMX) {
 968				sem_putref(sma);
 969				err = -ERANGE;
 970				goto out_free;
 971			}
 972		}
 
 973		sem_lock_and_putref(sma);
 974		if (sma->sem_perm.deleted) {
 975			sem_unlock(sma);
 976			err = -EIDRM;
 977			goto out_free;
 978		}
 979
 980		for (i = 0; i < nsems; i++)
 981			sma->sem_base[i].semval = sem_io[i];
 
 
 982
 983		assert_spin_locked(&sma->sem_perm.lock);
 984		list_for_each_entry(un, &sma->list_id, list_id) {
 985			for (i = 0; i < nsems; i++)
 986				un->semadj[i] = 0;
 987		}
 988		sma->sem_ctime = get_seconds();
 989		/* maybe some queued-up processes were waiting for this */
 990		do_smart_update(sma, NULL, 0, 0, &tasks);
 991		err = 0;
 992		goto out_unlock;
 993	}
 994	/* GETVAL, GETPID, GETNCTN, GETZCNT, SETVAL: fall-through */
 995	}
 996	err = -EINVAL;
 997	if(semnum < 0 || semnum >= nsems)
 
 
 
 
 
 998		goto out_unlock;
 
 999
1000	curr = &sma->sem_base[semnum];
 
1001
1002	switch (cmd) {
1003	case GETVAL:
1004		err = curr->semval;
1005		goto out_unlock;
1006	case GETPID:
1007		err = curr->sempid;
1008		goto out_unlock;
1009	case GETNCNT:
1010		err = count_semncnt(sma,semnum);
1011		goto out_unlock;
1012	case GETZCNT:
1013		err = count_semzcnt(sma,semnum);
1014		goto out_unlock;
1015	case SETVAL:
1016	{
1017		int val = arg.val;
1018		struct sem_undo *un;
1019
1020		err = -ERANGE;
1021		if (val > SEMVMX || val < 0)
1022			goto out_unlock;
1023
1024		assert_spin_locked(&sma->sem_perm.lock);
1025		list_for_each_entry(un, &sma->list_id, list_id)
1026			un->semadj[semnum] = 0;
1027
1028		curr->semval = val;
1029		curr->sempid = task_tgid_vnr(current);
1030		sma->sem_ctime = get_seconds();
1031		/* maybe some queued-up processes were waiting for this */
1032		do_smart_update(sma, NULL, 0, 0, &tasks);
1033		err = 0;
1034		goto out_unlock;
1035	}
1036	}
1037out_unlock:
1038	sem_unlock(sma);
1039	wake_up_sem_queue_do(&tasks);
1040
 
 
 
 
 
1041out_free:
1042	if(sem_io != fast_sem_io)
1043		ipc_free(sem_io, sizeof(ushort)*nsems);
1044	return err;
1045}
1046
1047static inline unsigned long
1048copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
1049{
1050	switch(version) {
1051	case IPC_64:
1052		if (copy_from_user(out, buf, sizeof(*out)))
1053			return -EFAULT;
1054		return 0;
1055	case IPC_OLD:
1056	    {
1057		struct semid_ds tbuf_old;
1058
1059		if(copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
1060			return -EFAULT;
1061
1062		out->sem_perm.uid	= tbuf_old.sem_perm.uid;
1063		out->sem_perm.gid	= tbuf_old.sem_perm.gid;
1064		out->sem_perm.mode	= tbuf_old.sem_perm.mode;
1065
1066		return 0;
1067	    }
1068	default:
1069		return -EINVAL;
1070	}
1071}
1072
1073/*
1074 * This function handles some semctl commands which require the rw_mutex
1075 * to be held in write mode.
1076 * NOTE: no locks must be held, the rw_mutex is taken inside this function.
1077 */
1078static int semctl_down(struct ipc_namespace *ns, int semid,
1079		       int cmd, int version, union semun arg)
1080{
1081	struct sem_array *sma;
1082	int err;
1083	struct semid64_ds semid64;
1084	struct kern_ipc_perm *ipcp;
1085
1086	if(cmd == IPC_SET) {
1087		if (copy_semid_from_user(&semid64, arg.buf, version))
1088			return -EFAULT;
1089	}
1090
1091	ipcp = ipcctl_pre_down(ns, &sem_ids(ns), semid, cmd,
1092			       &semid64.sem_perm, 0);
1093	if (IS_ERR(ipcp))
1094		return PTR_ERR(ipcp);
 
 
1095
1096	sma = container_of(ipcp, struct sem_array, sem_perm);
1097
1098	err = security_sem_semctl(sma, cmd);
1099	if (err)
1100		goto out_unlock;
1101
1102	switch(cmd){
1103	case IPC_RMID:
 
 
1104		freeary(ns, ipcp);
1105		goto out_up;
1106	case IPC_SET:
1107		ipc_update_perm(&semid64.sem_perm, ipcp);
1108		sma->sem_ctime = get_seconds();
 
 
 
1109		break;
1110	default:
1111		err = -EINVAL;
 
1112	}
1113
1114out_unlock:
1115	sem_unlock(sma);
 
 
1116out_up:
1117	up_write(&sem_ids(ns).rw_mutex);
1118	return err;
1119}
1120
1121SYSCALL_DEFINE(semctl)(int semid, int semnum, int cmd, union semun arg)
1122{
1123	int err = -EINVAL;
1124	int version;
1125	struct ipc_namespace *ns;
 
 
 
1126
1127	if (semid < 0)
1128		return -EINVAL;
1129
1130	version = ipc_parse_version(&cmd);
1131	ns = current->nsproxy->ipc_ns;
1132
1133	switch(cmd) {
1134	case IPC_INFO:
1135	case SEM_INFO:
 
1136	case IPC_STAT:
1137	case SEM_STAT:
1138		err = semctl_nolock(ns, semid, cmd, version, arg);
 
 
 
 
 
1139		return err;
1140	case GETALL:
1141	case GETVAL:
1142	case GETPID:
1143	case GETNCNT:
1144	case GETZCNT:
1145	case SETVAL:
1146	case SETALL:
1147		err = semctl_main(ns,semid,semnum,cmd,version,arg);
1148		return err;
1149	case IPC_RMID:
 
 
 
 
 
 
 
 
 
1150	case IPC_SET:
1151		err = semctl_down(ns, semid, cmd, version, arg);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1152		return err;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1153	default:
1154		return -EINVAL;
1155	}
1156}
1157#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
1158asmlinkage long SyS_semctl(int semid, int semnum, int cmd, union semun arg)
 
 
 
 
 
 
 
 
 
 
 
 
 
1159{
1160	return SYSC_semctl((int) semid, (int) semnum, (int) cmd, arg);
1161}
1162SYSCALL_ALIAS(sys_semctl, SyS_semctl);
1163#endif
1164
1165/* If the task doesn't already have a undo_list, then allocate one
1166 * here.  We guarantee there is only one thread using this undo list,
1167 * and current is THE ONE
1168 *
1169 * If this allocation and assignment succeeds, but later
1170 * portions of this code fail, there is no need to free the sem_undo_list.
1171 * Just let it stay associated with the task, and it'll be freed later
1172 * at exit time.
1173 *
1174 * This can block, so callers must hold no locks.
1175 */
1176static inline int get_undo_list(struct sem_undo_list **undo_listp)
1177{
1178	struct sem_undo_list *undo_list;
1179
1180	undo_list = current->sysvsem.undo_list;
1181	if (!undo_list) {
1182		undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
1183		if (undo_list == NULL)
1184			return -ENOMEM;
1185		spin_lock_init(&undo_list->lock);
1186		atomic_set(&undo_list->refcnt, 1);
1187		INIT_LIST_HEAD(&undo_list->list_proc);
1188
1189		current->sysvsem.undo_list = undo_list;
1190	}
1191	*undo_listp = undo_list;
1192	return 0;
1193}
1194
1195static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid)
1196{
1197	struct sem_undo *un;
1198
1199	list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) {
 
1200		if (un->semid == semid)
1201			return un;
1202	}
1203	return NULL;
1204}
1205
1206static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1207{
1208	struct sem_undo *un;
1209
1210  	assert_spin_locked(&ulp->lock);
1211
1212	un = __lookup_undo(ulp, semid);
1213	if (un) {
1214		list_del_rcu(&un->list_proc);
1215		list_add_rcu(&un->list_proc, &ulp->list_proc);
1216	}
1217	return un;
1218}
1219
1220/**
1221 * find_alloc_undo - Lookup (and if not present create) undo array
1222 * @ns: namespace
1223 * @semid: semaphore array id
1224 *
1225 * The function looks up (and if not present creates) the undo structure.
1226 * The size of the undo structure depends on the size of the semaphore
1227 * array, thus the alloc path is not that straightforward.
1228 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1229 * performs a rcu_read_lock().
1230 */
1231static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
1232{
1233	struct sem_array *sma;
1234	struct sem_undo_list *ulp;
1235	struct sem_undo *un, *new;
1236	int nsems;
1237	int error;
1238
1239	error = get_undo_list(&ulp);
1240	if (error)
1241		return ERR_PTR(error);
1242
1243	rcu_read_lock();
1244	spin_lock(&ulp->lock);
1245	un = lookup_undo(ulp, semid);
1246	spin_unlock(&ulp->lock);
1247	if (likely(un!=NULL))
1248		goto out;
1249	rcu_read_unlock();
1250
1251	/* no undo structure around - allocate one. */
1252	/* step 1: figure out the size of the semaphore array */
1253	sma = sem_lock_check(ns, semid);
1254	if (IS_ERR(sma))
 
1255		return ERR_CAST(sma);
 
1256
1257	nsems = sma->sem_nsems;
1258	sem_getref_and_unlock(sma);
 
 
 
 
 
1259
1260	/* step 2: allocate new undo structure */
1261	new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
 
1262	if (!new) {
1263		sem_putref(sma);
1264		return ERR_PTR(-ENOMEM);
1265	}
1266
1267	/* step 3: Acquire the lock on semaphore array */
 
1268	sem_lock_and_putref(sma);
1269	if (sma->sem_perm.deleted) {
1270		sem_unlock(sma);
1271		kfree(new);
 
1272		un = ERR_PTR(-EIDRM);
1273		goto out;
1274	}
1275	spin_lock(&ulp->lock);
1276
1277	/*
1278	 * step 4: check for races: did someone else allocate the undo struct?
1279	 */
1280	un = lookup_undo(ulp, semid);
1281	if (un) {
1282		kfree(new);
1283		goto success;
1284	}
1285	/* step 5: initialize & link new undo structure */
1286	new->semadj = (short *) &new[1];
1287	new->ulp = ulp;
1288	new->semid = semid;
1289	assert_spin_locked(&ulp->lock);
1290	list_add_rcu(&new->list_proc, &ulp->list_proc);
1291	assert_spin_locked(&sma->sem_perm.lock);
1292	list_add(&new->list_id, &sma->list_id);
1293	un = new;
1294
1295success:
1296	spin_unlock(&ulp->lock);
1297	rcu_read_lock();
1298	sem_unlock(sma);
1299out:
1300	return un;
1301}
1302
1303
1304/**
1305 * get_queue_result - Retrieve the result code from sem_queue
1306 * @q: Pointer to queue structure
1307 *
1308 * Retrieve the return code from the pending queue. If IN_WAKEUP is found in
1309 * q->status, then we must loop until the value is replaced with the final
1310 * value: This may happen if a task is woken up by an unrelated event (e.g.
1311 * signal) and in parallel the task is woken up by another task because it got
1312 * the requested semaphores.
1313 *
1314 * The function can be called with or without holding the semaphore spinlock.
1315 */
1316static int get_queue_result(struct sem_queue *q)
1317{
1318	int error;
1319
1320	error = q->status;
1321	while (unlikely(error == IN_WAKEUP)) {
1322		cpu_relax();
1323		error = q->status;
1324	}
1325
1326	return error;
1327}
1328
1329
1330SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
1331		unsigned, nsops, const struct timespec __user *, timeout)
1332{
1333	int error = -EINVAL;
1334	struct sem_array *sma;
1335	struct sembuf fast_sops[SEMOPM_FAST];
1336	struct sembuf* sops = fast_sops, *sop;
1337	struct sem_undo *un;
1338	int undos = 0, alter = 0, max;
 
1339	struct sem_queue queue;
1340	unsigned long jiffies_left = 0;
1341	struct ipc_namespace *ns;
1342	struct list_head tasks;
1343
1344	ns = current->nsproxy->ipc_ns;
1345
1346	if (nsops < 1 || semid < 0)
1347		return -EINVAL;
1348	if (nsops > ns->sc_semopm)
1349		return -E2BIG;
1350	if(nsops > SEMOPM_FAST) {
1351		sops = kmalloc(sizeof(*sops)*nsops,GFP_KERNEL);
1352		if(sops==NULL)
1353			return -ENOMEM;
1354	}
1355	if (copy_from_user (sops, tsops, nsops * sizeof(*tsops))) {
1356		error=-EFAULT;
 
1357		goto out_free;
1358	}
 
1359	if (timeout) {
1360		struct timespec _timeout;
1361		if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1362			error = -EFAULT;
1363			goto out_free;
1364		}
1365		if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1366			_timeout.tv_nsec >= 1000000000L) {
1367			error = -EINVAL;
1368			goto out_free;
1369		}
1370		jiffies_left = timespec_to_jiffies(&_timeout);
1371	}
 
1372	max = 0;
1373	for (sop = sops; sop < sops + nsops; sop++) {
 
 
1374		if (sop->sem_num >= max)
1375			max = sop->sem_num;
1376		if (sop->sem_flg & SEM_UNDO)
1377			undos = 1;
1378		if (sop->sem_op != 0)
1379			alter = 1;
 
 
 
 
 
 
 
 
 
 
 
1380	}
1381
1382	if (undos) {
 
1383		un = find_alloc_undo(ns, semid);
1384		if (IS_ERR(un)) {
1385			error = PTR_ERR(un);
1386			goto out_free;
1387		}
1388	} else
1389		un = NULL;
 
 
1390
1391	INIT_LIST_HEAD(&tasks);
1392
1393	sma = sem_lock_check(ns, semid);
1394	if (IS_ERR(sma)) {
1395		if (un)
1396			rcu_read_unlock();
1397		error = PTR_ERR(sma);
1398		goto out_free;
1399	}
1400
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1401	/*
1402	 * semid identifiers are not unique - find_alloc_undo may have
1403	 * allocated an undo structure, it was invalidated by an RMID
1404	 * and now a new array with received the same id. Check and fail.
1405	 * This case can be detected checking un->semid. The existence of
1406	 * "un" itself is guaranteed by rcu.
1407	 */
1408	error = -EIDRM;
1409	if (un) {
1410		if (un->semid == -1) {
1411			rcu_read_unlock();
1412			goto out_unlock_free;
1413		} else {
1414			/*
1415			 * rcu lock can be released, "un" cannot disappear:
1416			 * - sem_lock is acquired, thus IPC_RMID is
1417			 *   impossible.
1418			 * - exit_sem is impossible, it always operates on
1419			 *   current (or a dead task).
1420			 */
1421
1422			rcu_read_unlock();
1423		}
1424	}
1425
1426	error = -EFBIG;
1427	if (max >= sma->sem_nsems)
1428		goto out_unlock_free;
1429
1430	error = -EACCES;
1431	if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO))
1432		goto out_unlock_free;
1433
1434	error = security_sem_semop(sma, sops, nsops, alter);
1435	if (error)
1436		goto out_unlock_free;
1437
1438	error = try_atomic_semop (sma, sops, nsops, un, task_tgid_vnr(current));
1439	if (error <= 0) {
1440		if (alter && error == 0)
1441			do_smart_update(sma, sops, nsops, 1, &tasks);
1442
1443		goto out_unlock_free;
1444	}
1445
1446	/* We need to sleep on this operation, so we put the current
1447	 * task into the pending queue and go to sleep.
1448	 */
1449		
1450	queue.sops = sops;
1451	queue.nsops = nsops;
1452	queue.undo = un;
1453	queue.pid = task_tgid_vnr(current);
1454	queue.alter = alter;
1455	if (alter)
1456		list_add_tail(&queue.list, &sma->sem_pending);
1457	else
1458		list_add(&queue.list, &sma->sem_pending);
1459
1460	if (nsops == 1) {
1461		struct sem *curr;
1462		curr = &sma->sem_base[sops->sem_num];
1463
 
 
 
 
1464		if (alter)
1465			list_add_tail(&queue.simple_list, &curr->sem_pending);
1466		else
1467			list_add(&queue.simple_list, &curr->sem_pending);
1468	} else {
1469		INIT_LIST_HEAD(&queue.simple_list);
1470		sma->complex_count++;
1471	}
1472
1473	queue.status = -EINTR;
1474	queue.sleeper = current;
 
1475
1476sleep_again:
1477	current->state = TASK_INTERRUPTIBLE;
1478	sem_unlock(sma);
 
1479
1480	if (timeout)
1481		jiffies_left = schedule_timeout(jiffies_left);
1482	else
1483		schedule();
 
 
 
 
1484
1485	error = get_queue_result(&queue);
 
 
 
 
1486
1487	if (error != -EINTR) {
1488		/* fast path: update_queue already obtained all requested
1489		 * resources.
1490		 * Perform a smp_mb(): User space could assume that semop()
1491		 * is a memory barrier: Without the mb(), the cpu could
1492		 * speculatively read in user space stale data that was
1493		 * overwritten by the previous owner of the semaphore.
1494		 */
1495		smp_mb();
1496
1497		goto out_free;
 
 
 
 
 
1498	}
1499
1500	sma = sem_lock(ns, semid);
 
 
 
 
 
 
 
 
1501
1502	/*
1503	 * Wait until it's guaranteed that no wakeup_sem_queue_do() is ongoing.
1504	 */
1505	error = get_queue_result(&queue);
1506
1507	/*
1508	 * Array removed? If yes, leave without sem_unlock().
1509	 */
1510	if (IS_ERR(sma)) {
1511		goto out_free;
1512	}
 
 
 
 
 
 
 
 
 
 
 
1513
 
 
1514
1515	/*
1516	 * If queue.status != -EINTR we are woken up by another process.
1517	 * Leave without unlink_queue(), but with sem_unlock().
1518	 */
1519
1520	if (error != -EINTR) {
1521		goto out_unlock_free;
1522	}
 
1523
1524	/*
1525	 * If an interrupt occurred we have to clean up the queue
1526	 */
1527	if (timeout && jiffies_left == 0)
1528		error = -EAGAIN;
 
1529
1530	/*
1531	 * If the wakeup was spurious, just retry
1532	 */
1533	if (error == -EINTR && !signal_pending(current))
1534		goto sleep_again;
 
1535
1536	unlink_queue(sma, &queue);
1537
1538out_unlock_free:
1539	sem_unlock(sma);
1540
1541	wake_up_sem_queue_do(&tasks);
1542out_free:
1543	if(sops != fast_sops)
1544		kfree(sops);
1545	return error;
1546}
1547
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1548SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
1549		unsigned, nsops)
1550{
1551	return sys_semtimedop(semid, tsops, nsops, NULL);
1552}
1553
1554/* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
1555 * parent and child tasks.
1556 */
1557
1558int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
1559{
1560	struct sem_undo_list *undo_list;
1561	int error;
1562
1563	if (clone_flags & CLONE_SYSVSEM) {
1564		error = get_undo_list(&undo_list);
1565		if (error)
1566			return error;
1567		atomic_inc(&undo_list->refcnt);
1568		tsk->sysvsem.undo_list = undo_list;
1569	} else 
1570		tsk->sysvsem.undo_list = NULL;
1571
1572	return 0;
1573}
1574
1575/*
1576 * add semadj values to semaphores, free undo structures.
1577 * undo structures are not freed when semaphore arrays are destroyed
1578 * so some of them may be out of date.
1579 * IMPLEMENTATION NOTE: There is some confusion over whether the
1580 * set of adjustments that needs to be done should be done in an atomic
1581 * manner or not. That is, if we are attempting to decrement the semval
1582 * should we queue up and wait until we can do so legally?
1583 * The original implementation attempted to do this (queue and wait).
1584 * The current implementation does not do so. The POSIX standard
1585 * and SVID should be consulted to determine what behavior is mandated.
1586 */
1587void exit_sem(struct task_struct *tsk)
1588{
1589	struct sem_undo_list *ulp;
1590
1591	ulp = tsk->sysvsem.undo_list;
1592	if (!ulp)
1593		return;
1594	tsk->sysvsem.undo_list = NULL;
1595
1596	if (!atomic_dec_and_test(&ulp->refcnt))
1597		return;
1598
1599	for (;;) {
1600		struct sem_array *sma;
1601		struct sem_undo *un;
1602		struct list_head tasks;
1603		int semid;
1604		int i;
 
1605
1606		rcu_read_lock();
1607		un = list_entry_rcu(ulp->list_proc.next,
1608				    struct sem_undo, list_proc);
1609		if (&un->list_proc == &ulp->list_proc)
1610			semid = -1;
1611		 else
1612			semid = un->semid;
1613		rcu_read_unlock();
1614
1615		if (semid == -1)
 
 
 
1616			break;
 
 
 
 
1617
1618		sma = sem_lock_check(tsk->nsproxy->ipc_ns, un->semid);
 
 
 
 
1619
 
1620		/* exit_sem raced with IPC_RMID, nothing to do */
1621		if (IS_ERR(sma))
 
1622			continue;
 
1623
 
 
 
 
 
 
 
1624		un = __lookup_undo(ulp, semid);
1625		if (un == NULL) {
1626			/* exit_sem raced with IPC_RMID+semget() that created
1627			 * exactly the same semid. Nothing to do.
1628			 */
1629			sem_unlock(sma);
 
1630			continue;
1631		}
1632
1633		/* remove un from the linked lists */
1634		assert_spin_locked(&sma->sem_perm.lock);
1635		list_del(&un->list_id);
1636
1637		spin_lock(&ulp->lock);
1638		list_del_rcu(&un->list_proc);
1639		spin_unlock(&ulp->lock);
1640
1641		/* perform adjustments registered in un */
1642		for (i = 0; i < sma->sem_nsems; i++) {
1643			struct sem * semaphore = &sma->sem_base[i];
1644			if (un->semadj[i]) {
1645				semaphore->semval += un->semadj[i];
1646				/*
1647				 * Range checks of the new semaphore value,
1648				 * not defined by sus:
1649				 * - Some unices ignore the undo entirely
1650				 *   (e.g. HP UX 11i 11.22, Tru64 V5.1)
1651				 * - some cap the value (e.g. FreeBSD caps
1652				 *   at 0, but doesn't enforce SEMVMX)
1653				 *
1654				 * Linux caps the semaphore value, both at 0
1655				 * and at SEMVMX.
1656				 *
1657				 * 	Manfred <manfred@colorfullife.com>
1658				 */
1659				if (semaphore->semval < 0)
1660					semaphore->semval = 0;
1661				if (semaphore->semval > SEMVMX)
1662					semaphore->semval = SEMVMX;
1663				semaphore->sempid = task_tgid_vnr(current);
1664			}
1665		}
1666		/* maybe some queued-up processes were waiting for this */
1667		INIT_LIST_HEAD(&tasks);
1668		do_smart_update(sma, NULL, 0, 1, &tasks);
1669		sem_unlock(sma);
1670		wake_up_sem_queue_do(&tasks);
1671
1672		kfree_rcu(un, rcu);
1673	}
1674	kfree(ulp);
1675}
1676
1677#ifdef CONFIG_PROC_FS
1678static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
1679{
1680	struct sem_array *sma = it;
 
 
 
1681
1682	return seq_printf(s,
1683			  "%10d %10d  %4o %10u %5u %5u %5u %5u %10lu %10lu\n",
1684			  sma->sem_perm.key,
1685			  sma->sem_perm.id,
1686			  sma->sem_perm.mode,
1687			  sma->sem_nsems,
1688			  sma->sem_perm.uid,
1689			  sma->sem_perm.gid,
1690			  sma->sem_perm.cuid,
1691			  sma->sem_perm.cgid,
1692			  sma->sem_otime,
1693			  sma->sem_ctime);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1694}
1695#endif
v5.14.15
   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 * To prevent the compiler/cpu temporarily writing 0 to use_global_lock,
 221 * READ_ONCE()/WRITE_ONCE() is used.
 222 *
 223 * 2) queue.status: (SEM_BARRIER_2)
 224 * Initialization is done while holding sem_lock(), so no further barrier is
 225 * required.
 226 * Setting it to a result code is a RELEASE, this is ensured by both a
 227 * smp_store_release() (for case a) and while holding sem_lock()
 228 * (for case b).
 229 * The ACQUIRE when reading the result code without holding sem_lock() is
 230 * achieved by using READ_ONCE() + smp_acquire__after_ctrl_dep().
 231 * (case a above).
 232 * Reading the result code while holding sem_lock() needs no further barriers,
 233 * the locks inside sem_lock() enforce ordering (case b above)
 234 *
 235 * 3) current->state:
 236 * current->state is set to TASK_INTERRUPTIBLE while holding sem_lock().
 237 * The wakeup is handled using the wake_q infrastructure. wake_q wakeups may
 238 * happen immediately after calling wake_q_add. As wake_q_add_safe() is called
 239 * when holding sem_lock(), no further barriers are required.
 240 *
 241 * See also ipc/mqueue.c for more details on the covered races.
 242 */
 243
 244#define sc_semmsl	sem_ctls[0]
 245#define sc_semmns	sem_ctls[1]
 246#define sc_semopm	sem_ctls[2]
 247#define sc_semmni	sem_ctls[3]
 248
 249void sem_init_ns(struct ipc_namespace *ns)
 250{
 251	ns->sc_semmsl = SEMMSL;
 252	ns->sc_semmns = SEMMNS;
 253	ns->sc_semopm = SEMOPM;
 254	ns->sc_semmni = SEMMNI;
 255	ns->used_sems = 0;
 256	ipc_init_ids(&ns->ids[IPC_SEM_IDS]);
 257}
 258
 259#ifdef CONFIG_IPC_NS
 260void sem_exit_ns(struct ipc_namespace *ns)
 261{
 262	free_ipcs(ns, &sem_ids(ns), freeary);
 263	idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr);
 264	rhashtable_destroy(&ns->ids[IPC_SEM_IDS].key_ht);
 265}
 266#endif
 267
 268void __init sem_init(void)
 269{
 270	sem_init_ns(&init_ipc_ns);
 271	ipc_init_proc_interface("sysvipc/sem",
 272				"       key      semid perms      nsems   uid   gid  cuid  cgid      otime      ctime\n",
 273				IPC_SEM_IDS, sysvipc_sem_proc_show);
 274}
 275
 276/**
 277 * unmerge_queues - unmerge queues, if possible.
 278 * @sma: semaphore array
 279 *
 280 * The function unmerges the wait queues if complex_count is 0.
 281 * It must be called prior to dropping the global semaphore array lock.
 282 */
 283static void unmerge_queues(struct sem_array *sma)
 284{
 285	struct sem_queue *q, *tq;
 286
 287	/* complex operations still around? */
 288	if (sma->complex_count)
 289		return;
 290	/*
 291	 * We will switch back to simple mode.
 292	 * Move all pending operation back into the per-semaphore
 293	 * queues.
 294	 */
 295	list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
 296		struct sem *curr;
 297		curr = &sma->sems[q->sops[0].sem_num];
 298
 299		list_add_tail(&q->list, &curr->pending_alter);
 300	}
 301	INIT_LIST_HEAD(&sma->pending_alter);
 302}
 303
 304/**
 305 * merge_queues - merge single semop queues into global queue
 306 * @sma: semaphore array
 307 *
 308 * This function merges all per-semaphore queues into the global queue.
 309 * It is necessary to achieve FIFO ordering for the pending single-sop
 310 * operations when a multi-semop operation must sleep.
 311 * Only the alter operations must be moved, the const operations can stay.
 312 */
 313static void merge_queues(struct sem_array *sma)
 314{
 315	int i;
 316	for (i = 0; i < sma->sem_nsems; i++) {
 317		struct sem *sem = &sma->sems[i];
 318
 319		list_splice_init(&sem->pending_alter, &sma->pending_alter);
 320	}
 321}
 322
 323static void sem_rcu_free(struct rcu_head *head)
 324{
 325	struct kern_ipc_perm *p = container_of(head, struct kern_ipc_perm, rcu);
 326	struct sem_array *sma = container_of(p, struct sem_array, sem_perm);
 327
 328	security_sem_free(&sma->sem_perm);
 329	kvfree(sma);
 330}
 331
 332/*
 333 * Enter the mode suitable for non-simple operations:
 334 * Caller must own sem_perm.lock.
 335 */
 336static void complexmode_enter(struct sem_array *sma)
 337{
 338	int i;
 339	struct sem *sem;
 340
 341	if (sma->use_global_lock > 0)  {
 342		/*
 343		 * We are already in global lock mode.
 344		 * Nothing to do, just reset the
 345		 * counter until we return to simple mode.
 346		 */
 347		WRITE_ONCE(sma->use_global_lock, USE_GLOBAL_LOCK_HYSTERESIS);
 348		return;
 349	}
 350	WRITE_ONCE(sma->use_global_lock, USE_GLOBAL_LOCK_HYSTERESIS);
 351
 352	for (i = 0; i < sma->sem_nsems; i++) {
 353		sem = &sma->sems[i];
 354		spin_lock(&sem->lock);
 355		spin_unlock(&sem->lock);
 356	}
 357}
 358
 359/*
 360 * Try to leave the mode that disallows simple operations:
 361 * Caller must own sem_perm.lock.
 362 */
 363static void complexmode_tryleave(struct sem_array *sma)
 364{
 365	if (sma->complex_count)  {
 366		/* Complex ops are sleeping.
 367		 * We must stay in complex mode
 368		 */
 369		return;
 370	}
 371	if (sma->use_global_lock == 1) {
 372
 373		/* See SEM_BARRIER_1 for purpose/pairing */
 374		smp_store_release(&sma->use_global_lock, 0);
 375	} else {
 376		WRITE_ONCE(sma->use_global_lock,
 377				sma->use_global_lock-1);
 378	}
 379}
 380
 381#define SEM_GLOBAL_LOCK	(-1)
 382/*
 383 * If the request contains only one semaphore operation, and there are
 384 * no complex transactions pending, lock only the semaphore involved.
 385 * Otherwise, lock the entire semaphore array, since we either have
 386 * multiple semaphores in our own semops, or we need to look at
 387 * semaphores from other pending complex operations.
 388 */
 389static inline int sem_lock(struct sem_array *sma, struct sembuf *sops,
 390			      int nsops)
 391{
 392	struct sem *sem;
 393	int idx;
 394
 395	if (nsops != 1) {
 396		/* Complex operation - acquire a full lock */
 397		ipc_lock_object(&sma->sem_perm);
 398
 399		/* Prevent parallel simple ops */
 400		complexmode_enter(sma);
 401		return SEM_GLOBAL_LOCK;
 402	}
 403
 404	/*
 405	 * Only one semaphore affected - try to optimize locking.
 406	 * Optimized locking is possible if no complex operation
 407	 * is either enqueued or processed right now.
 408	 *
 409	 * Both facts are tracked by use_global_mode.
 410	 */
 411	idx = array_index_nospec(sops->sem_num, sma->sem_nsems);
 412	sem = &sma->sems[idx];
 413
 414	/*
 415	 * Initial check for use_global_lock. Just an optimization,
 416	 * no locking, no memory barrier.
 417	 */
 418	if (!READ_ONCE(sma->use_global_lock)) {
 419		/*
 420		 * It appears that no complex operation is around.
 421		 * Acquire the per-semaphore lock.
 422		 */
 423		spin_lock(&sem->lock);
 424
 425		/* see SEM_BARRIER_1 for purpose/pairing */
 426		if (!smp_load_acquire(&sma->use_global_lock)) {
 427			/* fast path successful! */
 428			return sops->sem_num;
 429		}
 430		spin_unlock(&sem->lock);
 431	}
 432
 433	/* slow path: acquire the full lock */
 434	ipc_lock_object(&sma->sem_perm);
 435
 436	if (sma->use_global_lock == 0) {
 437		/*
 438		 * The use_global_lock mode ended while we waited for
 439		 * sma->sem_perm.lock. Thus we must switch to locking
 440		 * with sem->lock.
 441		 * Unlike in the fast path, there is no need to recheck
 442		 * sma->use_global_lock after we have acquired sem->lock:
 443		 * We own sma->sem_perm.lock, thus use_global_lock cannot
 444		 * change.
 445		 */
 446		spin_lock(&sem->lock);
 447
 448		ipc_unlock_object(&sma->sem_perm);
 449		return sops->sem_num;
 450	} else {
 451		/*
 452		 * Not a false alarm, thus continue to use the global lock
 453		 * mode. No need for complexmode_enter(), this was done by
 454		 * the caller that has set use_global_mode to non-zero.
 455		 */
 456		return SEM_GLOBAL_LOCK;
 457	}
 458}
 459
 460static inline void sem_unlock(struct sem_array *sma, int locknum)
 461{
 462	if (locknum == SEM_GLOBAL_LOCK) {
 463		unmerge_queues(sma);
 464		complexmode_tryleave(sma);
 465		ipc_unlock_object(&sma->sem_perm);
 466	} else {
 467		struct sem *sem = &sma->sems[locknum];
 468		spin_unlock(&sem->lock);
 469	}
 470}
 471
 472/*
 473 * sem_lock_(check_) routines are called in the paths where the rwsem
 474 * is not held.
 475 *
 476 * The caller holds the RCU read lock.
 477 */
 478static inline struct sem_array *sem_obtain_object(struct ipc_namespace *ns, int id)
 479{
 480	struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&sem_ids(ns), id);
 481
 482	if (IS_ERR(ipcp))
 483		return ERR_CAST(ipcp);
 484
 485	return container_of(ipcp, struct sem_array, sem_perm);
 486}
 487
 488static inline struct sem_array *sem_obtain_object_check(struct ipc_namespace *ns,
 489							int id)
 490{
 491	struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&sem_ids(ns), id);
 492
 493	if (IS_ERR(ipcp))
 494		return ERR_CAST(ipcp);
 495
 496	return container_of(ipcp, struct sem_array, sem_perm);
 497}
 498
 499static inline void sem_lock_and_putref(struct sem_array *sma)
 500{
 501	sem_lock(sma, NULL, -1);
 502	ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
 503}
 504
 505static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
 506{
 507	ipc_rmid(&sem_ids(ns), &s->sem_perm);
 
 508}
 509
 510static struct sem_array *sem_alloc(size_t nsems)
 511{
 512	struct sem_array *sma;
 
 
 
 513
 514	if (nsems > (INT_MAX - sizeof(*sma)) / sizeof(sma->sems[0]))
 515		return NULL;
 
 
 516
 517	sma = kvzalloc(struct_size(sma, sems, nsems), GFP_KERNEL);
 518	if (unlikely(!sma))
 519		return NULL;
 520
 521	return sma;
 522}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 523
 524/**
 525 * newary - Create a new semaphore set
 526 * @ns: namespace
 527 * @params: ptr to the structure that contains key, semflg and nsems
 528 *
 529 * Called with sem_ids.rwsem held (as a writer)
 530 */
 
 531static int newary(struct ipc_namespace *ns, struct ipc_params *params)
 532{
 
 533	int retval;
 534	struct sem_array *sma;
 
 535	key_t key = params->key;
 536	int nsems = params->u.nsems;
 537	int semflg = params->flg;
 538	int i;
 539
 540	if (!nsems)
 541		return -EINVAL;
 542	if (ns->used_sems + nsems > ns->sc_semmns)
 543		return -ENOSPC;
 544
 545	sma = sem_alloc(nsems);
 546	if (!sma)
 
 547		return -ENOMEM;
 
 
 548
 549	sma->sem_perm.mode = (semflg & S_IRWXUGO);
 550	sma->sem_perm.key = key;
 551
 552	sma->sem_perm.security = NULL;
 553	retval = security_sem_alloc(&sma->sem_perm);
 554	if (retval) {
 555		kvfree(sma);
 556		return retval;
 557	}
 558
 559	for (i = 0; i < nsems; i++) {
 560		INIT_LIST_HEAD(&sma->sems[i].pending_alter);
 561		INIT_LIST_HEAD(&sma->sems[i].pending_const);
 562		spin_lock_init(&sma->sems[i].lock);
 
 563	}
 
 
 
 
 
 
 564
 565	sma->complex_count = 0;
 566	sma->use_global_lock = USE_GLOBAL_LOCK_HYSTERESIS;
 567	INIT_LIST_HEAD(&sma->pending_alter);
 568	INIT_LIST_HEAD(&sma->pending_const);
 569	INIT_LIST_HEAD(&sma->list_id);
 570	sma->sem_nsems = nsems;
 571	sma->sem_ctime = ktime_get_real_seconds();
 
 
 
 
 572
 573	/* ipc_addid() locks sma upon success. */
 574	retval = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
 575	if (retval < 0) {
 576		ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
 577		return retval;
 578	}
 579	ns->used_sems += nsems;
 580
 581	sem_unlock(sma, -1);
 582	rcu_read_unlock();
 
 
 
 
 583
 584	return sma->sem_perm.id;
 
 585}
 586
 587
 588/*
 589 * Called with sem_ids.rwsem and ipcp locked.
 590 */
 591static int sem_more_checks(struct kern_ipc_perm *ipcp, struct ipc_params *params)
 
 592{
 593	struct sem_array *sma;
 594
 595	sma = container_of(ipcp, struct sem_array, sem_perm);
 596	if (params->u.nsems > sma->sem_nsems)
 597		return -EINVAL;
 598
 599	return 0;
 600}
 601
 602long ksys_semget(key_t key, int nsems, int semflg)
 603{
 604	struct ipc_namespace *ns;
 605	static const struct ipc_ops sem_ops = {
 606		.getnew = newary,
 607		.associate = security_sem_associate,
 608		.more_checks = sem_more_checks,
 609	};
 610	struct ipc_params sem_params;
 611
 612	ns = current->nsproxy->ipc_ns;
 613
 614	if (nsems < 0 || nsems > ns->sc_semmsl)
 615		return -EINVAL;
 616
 
 
 
 
 617	sem_params.key = key;
 618	sem_params.flg = semflg;
 619	sem_params.u.nsems = nsems;
 620
 621	return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
 622}
 623
 624SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg)
 625{
 626	return ksys_semget(key, nsems, semflg);
 627}
 628
 629/**
 630 * perform_atomic_semop[_slow] - Attempt to perform semaphore
 631 *                               operations on a given array.
 632 * @sma: semaphore array
 633 * @q: struct sem_queue that describes the operation
 634 *
 635 * Caller blocking are as follows, based the value
 636 * indicated by the semaphore operation (sem_op):
 637 *
 638 *  (1) >0 never blocks.
 639 *  (2)  0 (wait-for-zero operation): semval is non-zero.
 640 *  (3) <0 attempting to decrement semval to a value smaller than zero.
 641 *
 642 * Returns 0 if the operation was possible.
 643 * Returns 1 if the operation is impossible, the caller must sleep.
 644 * Returns <0 for error codes.
 645 */
 646static int perform_atomic_semop_slow(struct sem_array *sma, struct sem_queue *q)
 647{
 648	int result, sem_op, nsops;
 649	struct pid *pid;
 650	struct sembuf *sop;
 651	struct sem *curr;
 652	struct sembuf *sops;
 653	struct sem_undo *un;
 654
 655	sops = q->sops;
 656	nsops = q->nsops;
 657	un = q->undo;
 658
 659	for (sop = sops; sop < sops + nsops; sop++) {
 660		int idx = array_index_nospec(sop->sem_num, sma->sem_nsems);
 661		curr = &sma->sems[idx];
 662		sem_op = sop->sem_op;
 663		result = curr->semval;
 664
 665		if (!sem_op && result)
 666			goto would_block;
 667
 668		result += sem_op;
 669		if (result < 0)
 670			goto would_block;
 671		if (result > SEMVMX)
 672			goto out_of_range;
 673
 674		if (sop->sem_flg & SEM_UNDO) {
 675			int undo = un->semadj[sop->sem_num] - sem_op;
 676			/* Exceeding the undo range is an error. */
 
 
 677			if (undo < (-SEMAEM - 1) || undo > SEMAEM)
 678				goto out_of_range;
 679			un->semadj[sop->sem_num] = undo;
 680		}
 681
 682		curr->semval = result;
 683	}
 684
 685	sop--;
 686	pid = q->pid;
 687	while (sop >= sops) {
 688		ipc_update_pid(&sma->sems[sop->sem_num].sempid, pid);
 
 
 689		sop--;
 690	}
 691
 692	return 0;
 693
 694out_of_range:
 695	result = -ERANGE;
 696	goto undo;
 697
 698would_block:
 699	q->blocking = sop;
 700
 701	if (sop->sem_flg & IPC_NOWAIT)
 702		result = -EAGAIN;
 703	else
 704		result = 1;
 705
 706undo:
 707	sop--;
 708	while (sop >= sops) {
 709		sem_op = sop->sem_op;
 710		sma->sems[sop->sem_num].semval -= sem_op;
 711		if (sop->sem_flg & SEM_UNDO)
 712			un->semadj[sop->sem_num] += sem_op;
 713		sop--;
 714	}
 715
 716	return result;
 717}
 718
 719static int perform_atomic_semop(struct sem_array *sma, struct sem_queue *q)
 
 
 
 
 
 
 
 720{
 721	int result, sem_op, nsops;
 722	struct sembuf *sop;
 723	struct sem *curr;
 724	struct sembuf *sops;
 725	struct sem_undo *un;
 726
 727	sops = q->sops;
 728	nsops = q->nsops;
 729	un = q->undo;
 730
 731	if (unlikely(q->dupsop))
 732		return perform_atomic_semop_slow(sma, q);
 733
 734	/*
 735	 * We scan the semaphore set twice, first to ensure that the entire
 736	 * operation can succeed, therefore avoiding any pointless writes
 737	 * to shared memory and having to undo such changes in order to block
 738	 * until the operations can go through.
 739	 */
 740	for (sop = sops; sop < sops + nsops; sop++) {
 741		int idx = array_index_nospec(sop->sem_num, sma->sem_nsems);
 742
 743		curr = &sma->sems[idx];
 744		sem_op = sop->sem_op;
 745		result = curr->semval;
 746
 747		if (!sem_op && result)
 748			goto would_block; /* wait-for-zero */
 749
 750		result += sem_op;
 751		if (result < 0)
 752			goto would_block;
 753
 754		if (result > SEMVMX)
 755			return -ERANGE;
 756
 757		if (sop->sem_flg & SEM_UNDO) {
 758			int undo = un->semadj[sop->sem_num] - sem_op;
 759
 760			/* Exceeding the undo range is an error. */
 761			if (undo < (-SEMAEM - 1) || undo > SEMAEM)
 762				return -ERANGE;
 763		}
 764	}
 
 
 765
 766	for (sop = sops; sop < sops + nsops; sop++) {
 767		curr = &sma->sems[sop->sem_num];
 768		sem_op = sop->sem_op;
 769		result = curr->semval;
 770
 771		if (sop->sem_flg & SEM_UNDO) {
 772			int undo = un->semadj[sop->sem_num] - sem_op;
 773
 774			un->semadj[sop->sem_num] = undo;
 775		}
 776		curr->semval += sem_op;
 777		ipc_update_pid(&curr->sempid, q->pid);
 
 
 
 
 
 
 
 
 
 
 
 
 
 778	}
 779
 780	return 0;
 781
 782would_block:
 783	q->blocking = sop;
 784	return sop->sem_flg & IPC_NOWAIT ? -EAGAIN : 1;
 785}
 786
 787static inline void wake_up_sem_queue_prepare(struct sem_queue *q, int error,
 788					     struct wake_q_head *wake_q)
 789{
 790	struct task_struct *sleeper;
 791
 792	sleeper = get_task_struct(q->sleeper);
 793
 794	/* see SEM_BARRIER_2 for purpose/pairing */
 795	smp_store_release(&q->status, error);
 796
 797	wake_q_add_safe(wake_q, sleeper);
 798}
 799
 800static void unlink_queue(struct sem_array *sma, struct sem_queue *q)
 801{
 802	list_del(&q->list);
 803	if (q->nsops > 1)
 
 
 804		sma->complex_count--;
 805}
 806
 807/** check_restart(sma, q)
 808 * @sma: semaphore array
 809 * @q: the operation that just completed
 810 *
 811 * update_queue is O(N^2) when it restarts scanning the whole queue of
 812 * waiting operations. Therefore this function checks if the restart is
 813 * really necessary. It is called after a previously waiting operation
 814 * modified the array.
 815 * Note that wait-for-zero operations are handled without restart.
 816 */
 817static inline int check_restart(struct sem_array *sma, struct sem_queue *q)
 818{
 819	/* pending complex alter operations are too difficult to analyse */
 820	if (!list_empty(&sma->pending_alter))
 
 
 
 
 
 
 
 821		return 1;
 822
 823	/* we were a sleeping complex operation. Too difficult */
 824	if (q->nsops > 1)
 825		return 1;
 826
 827	/* It is impossible that someone waits for the new value:
 828	 * - complex operations always restart.
 829	 * - wait-for-zero are handled separately.
 830	 * - q is a previously sleeping simple operation that
 831	 *   altered the array. It must be a decrement, because
 832	 *   simple increments never sleep.
 833	 * - If there are older (higher priority) decrements
 834	 *   in the queue, then they have observed the original
 835	 *   semval value and couldn't proceed. The operation
 836	 *   decremented to value - thus they won't proceed either.
 837	 */
 838	return 0;
 839}
 840
 841/**
 842 * wake_const_ops - wake up non-alter tasks
 843 * @sma: semaphore array.
 844 * @semnum: semaphore that was modified.
 845 * @wake_q: lockless wake-queue head.
 846 *
 847 * wake_const_ops must be called after a semaphore in a semaphore array
 848 * was set to 0. If complex const operations are pending, wake_const_ops must
 849 * be called with semnum = -1, as well as with the number of each modified
 850 * semaphore.
 851 * The tasks that must be woken up are added to @wake_q. The return code
 852 * is stored in q->pid.
 853 * The function returns 1 if at least one operation was completed successfully.
 854 */
 855static int wake_const_ops(struct sem_array *sma, int semnum,
 856			  struct wake_q_head *wake_q)
 857{
 858	struct sem_queue *q, *tmp;
 859	struct list_head *pending_list;
 860	int semop_completed = 0;
 861
 862	if (semnum == -1)
 863		pending_list = &sma->pending_const;
 864	else
 865		pending_list = &sma->sems[semnum].pending_const;
 866
 867	list_for_each_entry_safe(q, tmp, pending_list, list) {
 868		int error = perform_atomic_semop(sma, q);
 869
 870		if (error > 0)
 871			continue;
 872		/* operation completed, remove from queue & wakeup */
 873		unlink_queue(sma, q);
 874
 875		wake_up_sem_queue_prepare(q, error, wake_q);
 876		if (error == 0)
 877			semop_completed = 1;
 878	}
 879
 880	return semop_completed;
 881}
 882
 883/**
 884 * do_smart_wakeup_zero - wakeup all wait for zero tasks
 885 * @sma: semaphore array
 886 * @sops: operations that were performed
 887 * @nsops: number of operations
 888 * @wake_q: lockless wake-queue head
 889 *
 890 * Checks all required queue for wait-for-zero operations, based
 891 * on the actual changes that were performed on the semaphore array.
 892 * The function returns 1 if at least one operation was completed successfully.
 893 */
 894static int do_smart_wakeup_zero(struct sem_array *sma, struct sembuf *sops,
 895				int nsops, struct wake_q_head *wake_q)
 896{
 897	int i;
 898	int semop_completed = 0;
 899	int got_zero = 0;
 900
 901	/* first: the per-semaphore queues, if known */
 902	if (sops) {
 903		for (i = 0; i < nsops; i++) {
 904			int num = sops[i].sem_num;
 905
 906			if (sma->sems[num].semval == 0) {
 907				got_zero = 1;
 908				semop_completed |= wake_const_ops(sma, num, wake_q);
 909			}
 910		}
 911	} else {
 912		/*
 913		 * No sops means modified semaphores not known.
 914		 * Assume all were changed.
 915		 */
 916		for (i = 0; i < sma->sem_nsems; i++) {
 917			if (sma->sems[i].semval == 0) {
 918				got_zero = 1;
 919				semop_completed |= wake_const_ops(sma, i, wake_q);
 920			}
 921		}
 922	}
 923	/*
 924	 * If one of the modified semaphores got 0,
 925	 * then check the global queue, too.
 926	 */
 927	if (got_zero)
 928		semop_completed |= wake_const_ops(sma, -1, wake_q);
 
 
 
 
 
 929
 930	return semop_completed;
 
 931}
 932
 933
 934/**
 935 * update_queue - look for tasks that can be completed.
 936 * @sma: semaphore array.
 937 * @semnum: semaphore that was modified.
 938 * @wake_q: lockless wake-queue head.
 939 *
 940 * update_queue must be called after a semaphore in a semaphore array
 941 * was modified. If multiple semaphores were modified, update_queue must
 942 * be called with semnum = -1, as well as with the number of each modified
 943 * semaphore.
 944 * The tasks that must be woken up are added to @wake_q. The return code
 945 * is stored in q->pid.
 946 * The function internally checks if const operations can now succeed.
 947 *
 948 * The function return 1 if at least one semop was completed successfully.
 949 */
 950static int update_queue(struct sem_array *sma, int semnum, struct wake_q_head *wake_q)
 951{
 952	struct sem_queue *q, *tmp;
 
 953	struct list_head *pending_list;
 
 954	int semop_completed = 0;
 955
 956	if (semnum == -1)
 957		pending_list = &sma->pending_alter;
 958	else
 959		pending_list = &sma->sems[semnum].pending_alter;
 
 
 
 
 
 
 
 
 
 
 960
 961again:
 962	list_for_each_entry_safe(q, tmp, pending_list, list) {
 
 963		int error, restart;
 964
 
 
 
 965		/* If we are scanning the single sop, per-semaphore list of
 966		 * one semaphore and that semaphore is 0, then it is not
 967		 * necessary to scan further: simple increments
 968		 * that affect only one entry succeed immediately and cannot
 969		 * be in the  per semaphore pending queue, and decrements
 970		 * cannot be successful if the value is already 0.
 971		 */
 972		if (semnum != -1 && sma->sems[semnum].semval == 0)
 
 973			break;
 974
 975		error = perform_atomic_semop(sma, q);
 
 976
 977		/* Does q->sleeper still need to sleep? */
 978		if (error > 0)
 979			continue;
 980
 981		unlink_queue(sma, q);
 982
 983		if (error) {
 984			restart = 0;
 985		} else {
 986			semop_completed = 1;
 987			do_smart_wakeup_zero(sma, q->sops, q->nsops, wake_q);
 988			restart = check_restart(sma, q);
 989		}
 990
 991		wake_up_sem_queue_prepare(q, error, wake_q);
 992		if (restart)
 993			goto again;
 994	}
 995	return semop_completed;
 996}
 997
 998/**
 999 * set_semotime - set sem_otime
1000 * @sma: semaphore array
1001 * @sops: operations that modified the array, may be NULL
1002 *
1003 * sem_otime is replicated to avoid cache line trashing.
1004 * This function sets one instance to the current time.
1005 */
1006static void set_semotime(struct sem_array *sma, struct sembuf *sops)
1007{
1008	if (sops == NULL) {
1009		sma->sems[0].sem_otime = ktime_get_real_seconds();
1010	} else {
1011		sma->sems[sops[0].sem_num].sem_otime =
1012						ktime_get_real_seconds();
1013	}
1014}
1015
1016/**
1017 * do_smart_update - optimized update_queue
1018 * @sma: semaphore array
1019 * @sops: operations that were performed
1020 * @nsops: number of operations
1021 * @otime: force setting otime
1022 * @wake_q: lockless wake-queue head
1023 *
1024 * do_smart_update() does the required calls to update_queue and wakeup_zero,
1025 * based on the actual changes that were performed on the semaphore array.
1026 * Note that the function does not do the actual wake-up: the caller is
1027 * responsible for calling wake_up_q().
1028 * It is safe to perform this call after dropping all locks.
1029 */
1030static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops,
1031			    int otime, struct wake_q_head *wake_q)
1032{
1033	int i;
1034
1035	otime |= do_smart_wakeup_zero(sma, sops, nsops, wake_q);
 
 
 
 
1036
1037	if (!list_empty(&sma->pending_alter)) {
1038		/* semaphore array uses the global queue - just process it. */
1039		otime |= update_queue(sma, -1, wake_q);
1040	} else {
1041		if (!sops) {
1042			/*
1043			 * No sops, thus the modified semaphores are not
1044			 * known. Check all.
1045			 */
1046			for (i = 0; i < sma->sem_nsems; i++)
1047				otime |= update_queue(sma, i, wake_q);
1048		} else {
1049			/*
1050			 * Check the semaphores that were increased:
1051			 * - No complex ops, thus all sleeping ops are
1052			 *   decrease.
1053			 * - if we decreased the value, then any sleeping
1054			 *   semaphore ops won't be able to run: If the
1055			 *   previous value was too small, then the new
1056			 *   value will be too small, too.
1057			 */
1058			for (i = 0; i < nsops; i++) {
1059				if (sops[i].sem_op > 0) {
1060					otime |= update_queue(sma,
1061							      sops[i].sem_num, wake_q);
1062				}
1063			}
1064		}
1065	}
 
1066	if (otime)
1067		set_semotime(sma, sops);
1068}
1069
1070/*
1071 * check_qop: Test if a queued operation sleeps on the semaphore semnum
1072 */
1073static int check_qop(struct sem_array *sma, int semnum, struct sem_queue *q,
1074			bool count_zero)
1075{
1076	struct sembuf *sop = q->blocking;
1077
1078	/*
1079	 * Linux always (since 0.99.10) reported a task as sleeping on all
1080	 * semaphores. This violates SUS, therefore it was changed to the
1081	 * standard compliant behavior.
1082	 * Give the administrators a chance to notice that an application
1083	 * might misbehave because it relies on the Linux behavior.
1084	 */
1085	pr_info_once("semctl(GETNCNT/GETZCNT) is since 3.16 Single Unix Specification compliant.\n"
1086			"The task %s (%d) triggered the difference, watch for misbehavior.\n",
1087			current->comm, task_pid_nr(current));
1088
1089	if (sop->sem_num != semnum)
1090		return 0;
1091
1092	if (count_zero && sop->sem_op == 0)
1093		return 1;
1094	if (!count_zero && sop->sem_op < 0)
1095		return 1;
1096
1097	return 0;
1098}
1099
1100/* The following counts are associated to each semaphore:
1101 *   semncnt        number of tasks waiting on semval being nonzero
1102 *   semzcnt        number of tasks waiting on semval being zero
1103 *
1104 * Per definition, a task waits only on the semaphore of the first semop
1105 * that cannot proceed, even if additional operation would block, too.
1106 */
1107static int count_semcnt(struct sem_array *sma, ushort semnum,
1108			bool count_zero)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1109{
1110	struct list_head *l;
1111	struct sem_queue *q;
1112	int semcnt;
1113
1114	semcnt = 0;
1115	/* First: check the simple operations. They are easy to evaluate */
1116	if (count_zero)
1117		l = &sma->sems[semnum].pending_const;
1118	else
1119		l = &sma->sems[semnum].pending_alter;
1120
1121	list_for_each_entry(q, l, list) {
1122		/* all task on a per-semaphore list sleep on exactly
1123		 * that semaphore
1124		 */
1125		semcnt++;
1126	}
1127
1128	/* Then: check the complex operations. */
1129	list_for_each_entry(q, &sma->pending_alter, list) {
1130		semcnt += check_qop(sma, semnum, q, count_zero);
1131	}
1132	if (count_zero) {
1133		list_for_each_entry(q, &sma->pending_const, list) {
1134			semcnt += check_qop(sma, semnum, q, count_zero);
1135		}
1136	}
1137	return semcnt;
1138}
1139
1140/* Free a semaphore set. freeary() is called with sem_ids.rwsem locked
1141 * as a writer and the spinlock for this semaphore set hold. sem_ids.rwsem
1142 * remains locked on exit.
1143 */
1144static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
1145{
1146	struct sem_undo *un, *tu;
1147	struct sem_queue *q, *tq;
1148	struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
1149	int i;
1150	DEFINE_WAKE_Q(wake_q);
1151
1152	/* Free the existing undo structures for this semaphore set.  */
1153	ipc_assert_locked_object(&sma->sem_perm);
1154	list_for_each_entry_safe(un, tu, &sma->list_id, list_id) {
1155		list_del(&un->list_id);
1156		spin_lock(&un->ulp->lock);
1157		un->semid = -1;
1158		list_del_rcu(&un->list_proc);
1159		spin_unlock(&un->ulp->lock);
1160		kvfree_rcu(un, rcu);
1161	}
1162
1163	/* Wake up all pending processes and let them fail with EIDRM. */
1164	list_for_each_entry_safe(q, tq, &sma->pending_const, list) {
 
1165		unlink_queue(sma, q);
1166		wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
1167	}
1168
1169	list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
1170		unlink_queue(sma, q);
1171		wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
1172	}
1173	for (i = 0; i < sma->sem_nsems; i++) {
1174		struct sem *sem = &sma->sems[i];
1175		list_for_each_entry_safe(q, tq, &sem->pending_const, list) {
1176			unlink_queue(sma, q);
1177			wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
1178		}
1179		list_for_each_entry_safe(q, tq, &sem->pending_alter, list) {
1180			unlink_queue(sma, q);
1181			wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
1182		}
1183		ipc_update_pid(&sem->sempid, NULL);
1184	}
1185
1186	/* Remove the semaphore set from the IDR */
1187	sem_rmid(ns, sma);
1188	sem_unlock(sma, -1);
1189	rcu_read_unlock();
1190
1191	wake_up_q(&wake_q);
1192	ns->used_sems -= sma->sem_nsems;
1193	ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
 
1194}
1195
1196static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
1197{
1198	switch (version) {
1199	case IPC_64:
1200		return copy_to_user(buf, in, sizeof(*in));
1201	case IPC_OLD:
1202	    {
1203		struct semid_ds out;
1204
1205		memset(&out, 0, sizeof(out));
1206
1207		ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
1208
1209		out.sem_otime	= in->sem_otime;
1210		out.sem_ctime	= in->sem_ctime;
1211		out.sem_nsems	= in->sem_nsems;
1212
1213		return copy_to_user(buf, &out, sizeof(out));
1214	    }
1215	default:
1216		return -EINVAL;
1217	}
1218}
1219
1220static time64_t get_semotime(struct sem_array *sma)
 
1221{
1222	int i;
1223	time64_t res;
1224
1225	res = sma->sems[0].sem_otime;
1226	for (i = 1; i < sma->sem_nsems; i++) {
1227		time64_t to = sma->sems[i].sem_otime;
 
 
 
1228
1229		if (to > res)
1230			res = to;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1231	}
1232	return res;
1233}
 
 
 
1234
1235static int semctl_stat(struct ipc_namespace *ns, int semid,
1236			 int cmd, struct semid64_ds *semid64)
1237{
1238	struct sem_array *sma;
1239	time64_t semotime;
1240	int err;
1241
1242	memset(semid64, 0, sizeof(*semid64));
1243
1244	rcu_read_lock();
1245	if (cmd == SEM_STAT || cmd == SEM_STAT_ANY) {
1246		sma = sem_obtain_object(ns, semid);
1247		if (IS_ERR(sma)) {
1248			err = PTR_ERR(sma);
1249			goto out_unlock;
1250		}
1251	} else { /* IPC_STAT */
1252		sma = sem_obtain_object_check(ns, semid);
1253		if (IS_ERR(sma)) {
1254			err = PTR_ERR(sma);
1255			goto out_unlock;
1256		}
1257	}
1258
1259	/* see comment for SHM_STAT_ANY */
1260	if (cmd == SEM_STAT_ANY)
1261		audit_ipc_obj(&sma->sem_perm);
1262	else {
1263		err = -EACCES;
1264		if (ipcperms(ns, &sma->sem_perm, S_IRUGO))
1265			goto out_unlock;
1266	}
1267
1268	err = security_sem_semctl(&sma->sem_perm, cmd);
1269	if (err)
1270		goto out_unlock;
1271
1272	ipc_lock_object(&sma->sem_perm);
1273
1274	if (!ipc_valid_object(&sma->sem_perm)) {
1275		ipc_unlock_object(&sma->sem_perm);
1276		err = -EIDRM;
1277		goto out_unlock;
 
 
 
 
1278	}
1279
1280	kernel_to_ipc64_perm(&sma->sem_perm, &semid64->sem_perm);
1281	semotime = get_semotime(sma);
1282	semid64->sem_otime = semotime;
1283	semid64->sem_ctime = sma->sem_ctime;
1284#ifndef CONFIG_64BIT
1285	semid64->sem_otime_high = semotime >> 32;
1286	semid64->sem_ctime_high = sma->sem_ctime >> 32;
1287#endif
1288	semid64->sem_nsems = sma->sem_nsems;
1289
1290	if (cmd == IPC_STAT) {
1291		/*
1292		 * As defined in SUS:
1293		 * Return 0 on success
1294		 */
1295		err = 0;
1296	} else {
1297		/*
1298		 * SEM_STAT and SEM_STAT_ANY (both Linux specific)
1299		 * Return the full id, including the sequence number
1300		 */
1301		err = sma->sem_perm.id;
1302	}
1303	ipc_unlock_object(&sma->sem_perm);
1304out_unlock:
1305	rcu_read_unlock();
1306	return err;
1307}
1308
1309static int semctl_info(struct ipc_namespace *ns, int semid,
1310			 int cmd, void __user *p)
1311{
1312	struct seminfo seminfo;
1313	int max_idx;
1314	int err;
1315
1316	err = security_sem_semctl(NULL, cmd);
1317	if (err)
1318		return err;
1319
1320	memset(&seminfo, 0, sizeof(seminfo));
1321	seminfo.semmni = ns->sc_semmni;
1322	seminfo.semmns = ns->sc_semmns;
1323	seminfo.semmsl = ns->sc_semmsl;
1324	seminfo.semopm = ns->sc_semopm;
1325	seminfo.semvmx = SEMVMX;
1326	seminfo.semmnu = SEMMNU;
1327	seminfo.semmap = SEMMAP;
1328	seminfo.semume = SEMUME;
1329	down_read(&sem_ids(ns).rwsem);
1330	if (cmd == SEM_INFO) {
1331		seminfo.semusz = sem_ids(ns).in_use;
1332		seminfo.semaem = ns->used_sems;
1333	} else {
1334		seminfo.semusz = SEMUSZ;
1335		seminfo.semaem = SEMAEM;
1336	}
1337	max_idx = ipc_get_maxidx(&sem_ids(ns));
1338	up_read(&sem_ids(ns).rwsem);
1339	if (copy_to_user(p, &seminfo, sizeof(struct seminfo)))
1340		return -EFAULT;
1341	return (max_idx < 0) ? 0 : max_idx;
1342}
1343
1344static int semctl_setval(struct ipc_namespace *ns, int semid, int semnum,
1345		int val)
1346{
1347	struct sem_undo *un;
1348	struct sem_array *sma;
1349	struct sem *curr;
1350	int err;
1351	DEFINE_WAKE_Q(wake_q);
1352
1353	if (val > SEMVMX || val < 0)
1354		return -ERANGE;
1355
1356	rcu_read_lock();
1357	sma = sem_obtain_object_check(ns, semid);
1358	if (IS_ERR(sma)) {
1359		rcu_read_unlock();
1360		return PTR_ERR(sma);
1361	}
1362
1363	if (semnum < 0 || semnum >= sma->sem_nsems) {
1364		rcu_read_unlock();
1365		return -EINVAL;
1366	}
1367
1368
1369	if (ipcperms(ns, &sma->sem_perm, S_IWUGO)) {
1370		rcu_read_unlock();
1371		return -EACCES;
1372	}
1373
1374	err = security_sem_semctl(&sma->sem_perm, SETVAL);
1375	if (err) {
1376		rcu_read_unlock();
1377		return -EACCES;
1378	}
1379
1380	sem_lock(sma, NULL, -1);
1381
1382	if (!ipc_valid_object(&sma->sem_perm)) {
1383		sem_unlock(sma, -1);
1384		rcu_read_unlock();
1385		return -EIDRM;
1386	}
1387
1388	semnum = array_index_nospec(semnum, sma->sem_nsems);
1389	curr = &sma->sems[semnum];
1390
1391	ipc_assert_locked_object(&sma->sem_perm);
1392	list_for_each_entry(un, &sma->list_id, list_id)
1393		un->semadj[semnum] = 0;
1394
1395	curr->semval = val;
1396	ipc_update_pid(&curr->sempid, task_tgid(current));
1397	sma->sem_ctime = ktime_get_real_seconds();
1398	/* maybe some queued-up processes were waiting for this */
1399	do_smart_update(sma, NULL, 0, 0, &wake_q);
1400	sem_unlock(sma, -1);
1401	rcu_read_unlock();
1402	wake_up_q(&wake_q);
1403	return 0;
1404}
1405
1406static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
1407		int cmd, void __user *p)
1408{
1409	struct sem_array *sma;
1410	struct sem *curr;
1411	int err, nsems;
1412	ushort fast_sem_io[SEMMSL_FAST];
1413	ushort *sem_io = fast_sem_io;
1414	DEFINE_WAKE_Q(wake_q);
 
1415
1416	rcu_read_lock();
1417	sma = sem_obtain_object_check(ns, semid);
1418	if (IS_ERR(sma)) {
1419		rcu_read_unlock();
1420		return PTR_ERR(sma);
1421	}
1422
 
1423	nsems = sma->sem_nsems;
1424
1425	err = -EACCES;
1426	if (ipcperms(ns, &sma->sem_perm, cmd == SETALL ? S_IWUGO : S_IRUGO))
1427		goto out_rcu_wakeup;
 
1428
1429	err = security_sem_semctl(&sma->sem_perm, cmd);
1430	if (err)
1431		goto out_rcu_wakeup;
1432
1433	err = -EACCES;
1434	switch (cmd) {
1435	case GETALL:
1436	{
1437		ushort __user *array = p;
1438		int i;
1439
1440		sem_lock(sma, NULL, -1);
1441		if (!ipc_valid_object(&sma->sem_perm)) {
1442			err = -EIDRM;
1443			goto out_unlock;
1444		}
1445		if (nsems > SEMMSL_FAST) {
1446			if (!ipc_rcu_getref(&sma->sem_perm)) {
1447				err = -EIDRM;
1448				goto out_unlock;
1449			}
1450			sem_unlock(sma, -1);
1451			rcu_read_unlock();
1452			sem_io = kvmalloc_array(nsems, sizeof(ushort),
1453						GFP_KERNEL);
1454			if (sem_io == NULL) {
1455				ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
1456				return -ENOMEM;
1457			}
1458
1459			rcu_read_lock();
1460			sem_lock_and_putref(sma);
1461			if (!ipc_valid_object(&sma->sem_perm)) {
 
1462				err = -EIDRM;
1463				goto out_unlock;
1464			}
1465		}
 
1466		for (i = 0; i < sma->sem_nsems; i++)
1467			sem_io[i] = sma->sems[i].semval;
1468		sem_unlock(sma, -1);
1469		rcu_read_unlock();
1470		err = 0;
1471		if (copy_to_user(array, sem_io, nsems*sizeof(ushort)))
1472			err = -EFAULT;
1473		goto out_free;
1474	}
1475	case SETALL:
1476	{
1477		int i;
1478		struct sem_undo *un;
1479
1480		if (!ipc_rcu_getref(&sma->sem_perm)) {
1481			err = -EIDRM;
1482			goto out_rcu_wakeup;
1483		}
1484		rcu_read_unlock();
1485
1486		if (nsems > SEMMSL_FAST) {
1487			sem_io = kvmalloc_array(nsems, sizeof(ushort),
1488						GFP_KERNEL);
1489			if (sem_io == NULL) {
1490				ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
1491				return -ENOMEM;
1492			}
1493		}
1494
1495		if (copy_from_user(sem_io, p, nsems*sizeof(ushort))) {
1496			ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
1497			err = -EFAULT;
1498			goto out_free;
1499		}
1500
1501		for (i = 0; i < nsems; i++) {
1502			if (sem_io[i] > SEMVMX) {
1503				ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
1504				err = -ERANGE;
1505				goto out_free;
1506			}
1507		}
1508		rcu_read_lock();
1509		sem_lock_and_putref(sma);
1510		if (!ipc_valid_object(&sma->sem_perm)) {
 
1511			err = -EIDRM;
1512			goto out_unlock;
1513		}
1514
1515		for (i = 0; i < nsems; i++) {
1516			sma->sems[i].semval = sem_io[i];
1517			ipc_update_pid(&sma->sems[i].sempid, task_tgid(current));
1518		}
1519
1520		ipc_assert_locked_object(&sma->sem_perm);
1521		list_for_each_entry(un, &sma->list_id, list_id) {
1522			for (i = 0; i < nsems; i++)
1523				un->semadj[i] = 0;
1524		}
1525		sma->sem_ctime = ktime_get_real_seconds();
1526		/* maybe some queued-up processes were waiting for this */
1527		do_smart_update(sma, NULL, 0, 0, &wake_q);
1528		err = 0;
1529		goto out_unlock;
1530	}
1531	/* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */
1532	}
1533	err = -EINVAL;
1534	if (semnum < 0 || semnum >= nsems)
1535		goto out_rcu_wakeup;
1536
1537	sem_lock(sma, NULL, -1);
1538	if (!ipc_valid_object(&sma->sem_perm)) {
1539		err = -EIDRM;
1540		goto out_unlock;
1541	}
1542
1543	semnum = array_index_nospec(semnum, nsems);
1544	curr = &sma->sems[semnum];
1545
1546	switch (cmd) {
1547	case GETVAL:
1548		err = curr->semval;
1549		goto out_unlock;
1550	case GETPID:
1551		err = pid_vnr(curr->sempid);
1552		goto out_unlock;
1553	case GETNCNT:
1554		err = count_semcnt(sma, semnum, 0);
1555		goto out_unlock;
1556	case GETZCNT:
1557		err = count_semcnt(sma, semnum, 1);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1558		goto out_unlock;
1559	}
 
 
 
 
1560
1561out_unlock:
1562	sem_unlock(sma, -1);
1563out_rcu_wakeup:
1564	rcu_read_unlock();
1565	wake_up_q(&wake_q);
1566out_free:
1567	if (sem_io != fast_sem_io)
1568		kvfree(sem_io);
1569	return err;
1570}
1571
1572static inline unsigned long
1573copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
1574{
1575	switch (version) {
1576	case IPC_64:
1577		if (copy_from_user(out, buf, sizeof(*out)))
1578			return -EFAULT;
1579		return 0;
1580	case IPC_OLD:
1581	    {
1582		struct semid_ds tbuf_old;
1583
1584		if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
1585			return -EFAULT;
1586
1587		out->sem_perm.uid	= tbuf_old.sem_perm.uid;
1588		out->sem_perm.gid	= tbuf_old.sem_perm.gid;
1589		out->sem_perm.mode	= tbuf_old.sem_perm.mode;
1590
1591		return 0;
1592	    }
1593	default:
1594		return -EINVAL;
1595	}
1596}
1597
1598/*
1599 * This function handles some semctl commands which require the rwsem
1600 * to be held in write mode.
1601 * NOTE: no locks must be held, the rwsem is taken inside this function.
1602 */
1603static int semctl_down(struct ipc_namespace *ns, int semid,
1604		       int cmd, struct semid64_ds *semid64)
1605{
1606	struct sem_array *sma;
1607	int err;
 
1608	struct kern_ipc_perm *ipcp;
1609
1610	down_write(&sem_ids(ns).rwsem);
1611	rcu_read_lock();
 
 
1612
1613	ipcp = ipcctl_obtain_check(ns, &sem_ids(ns), semid, cmd,
1614				      &semid64->sem_perm, 0);
1615	if (IS_ERR(ipcp)) {
1616		err = PTR_ERR(ipcp);
1617		goto out_unlock1;
1618	}
1619
1620	sma = container_of(ipcp, struct sem_array, sem_perm);
1621
1622	err = security_sem_semctl(&sma->sem_perm, cmd);
1623	if (err)
1624		goto out_unlock1;
1625
1626	switch (cmd) {
1627	case IPC_RMID:
1628		sem_lock(sma, NULL, -1);
1629		/* freeary unlocks the ipc object and rcu */
1630		freeary(ns, ipcp);
1631		goto out_up;
1632	case IPC_SET:
1633		sem_lock(sma, NULL, -1);
1634		err = ipc_update_perm(&semid64->sem_perm, ipcp);
1635		if (err)
1636			goto out_unlock0;
1637		sma->sem_ctime = ktime_get_real_seconds();
1638		break;
1639	default:
1640		err = -EINVAL;
1641		goto out_unlock1;
1642	}
1643
1644out_unlock0:
1645	sem_unlock(sma, -1);
1646out_unlock1:
1647	rcu_read_unlock();
1648out_up:
1649	up_write(&sem_ids(ns).rwsem);
1650	return err;
1651}
1652
1653static long ksys_semctl(int semid, int semnum, int cmd, unsigned long arg, int version)
1654{
 
 
1655	struct ipc_namespace *ns;
1656	void __user *p = (void __user *)arg;
1657	struct semid64_ds semid64;
1658	int err;
1659
1660	if (semid < 0)
1661		return -EINVAL;
1662
 
1663	ns = current->nsproxy->ipc_ns;
1664
1665	switch (cmd) {
1666	case IPC_INFO:
1667	case SEM_INFO:
1668		return semctl_info(ns, semid, cmd, p);
1669	case IPC_STAT:
1670	case SEM_STAT:
1671	case SEM_STAT_ANY:
1672		err = semctl_stat(ns, semid, cmd, &semid64);
1673		if (err < 0)
1674			return err;
1675		if (copy_semid_to_user(p, &semid64, version))
1676			err = -EFAULT;
1677		return err;
1678	case GETALL:
1679	case GETVAL:
1680	case GETPID:
1681	case GETNCNT:
1682	case GETZCNT:
 
1683	case SETALL:
1684		return semctl_main(ns, semid, semnum, cmd, p);
1685	case SETVAL: {
1686		int val;
1687#if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN)
1688		/* big-endian 64bit */
1689		val = arg >> 32;
1690#else
1691		/* 32bit or little-endian 64bit */
1692		val = arg;
1693#endif
1694		return semctl_setval(ns, semid, semnum, val);
1695	}
1696	case IPC_SET:
1697		if (copy_semid_from_user(&semid64, p, version))
1698			return -EFAULT;
1699		fallthrough;
1700	case IPC_RMID:
1701		return semctl_down(ns, semid, cmd, &semid64);
1702	default:
1703		return -EINVAL;
1704	}
1705}
1706
1707SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, unsigned long, arg)
1708{
1709	return ksys_semctl(semid, semnum, cmd, arg, IPC_64);
1710}
1711
1712#ifdef CONFIG_ARCH_WANT_IPC_PARSE_VERSION
1713long ksys_old_semctl(int semid, int semnum, int cmd, unsigned long arg)
1714{
1715	int version = ipc_parse_version(&cmd);
1716
1717	return ksys_semctl(semid, semnum, cmd, arg, version);
1718}
1719
1720SYSCALL_DEFINE4(old_semctl, int, semid, int, semnum, int, cmd, unsigned long, arg)
1721{
1722	return ksys_old_semctl(semid, semnum, cmd, arg);
1723}
1724#endif
1725
1726#ifdef CONFIG_COMPAT
1727
1728struct compat_semid_ds {
1729	struct compat_ipc_perm sem_perm;
1730	old_time32_t sem_otime;
1731	old_time32_t sem_ctime;
1732	compat_uptr_t sem_base;
1733	compat_uptr_t sem_pending;
1734	compat_uptr_t sem_pending_last;
1735	compat_uptr_t undo;
1736	unsigned short sem_nsems;
1737};
1738
1739static int copy_compat_semid_from_user(struct semid64_ds *out, void __user *buf,
1740					int version)
1741{
1742	memset(out, 0, sizeof(*out));
1743	if (version == IPC_64) {
1744		struct compat_semid64_ds __user *p = buf;
1745		return get_compat_ipc64_perm(&out->sem_perm, &p->sem_perm);
1746	} else {
1747		struct compat_semid_ds __user *p = buf;
1748		return get_compat_ipc_perm(&out->sem_perm, &p->sem_perm);
1749	}
1750}
1751
1752static int copy_compat_semid_to_user(void __user *buf, struct semid64_ds *in,
1753					int version)
1754{
1755	if (version == IPC_64) {
1756		struct compat_semid64_ds v;
1757		memset(&v, 0, sizeof(v));
1758		to_compat_ipc64_perm(&v.sem_perm, &in->sem_perm);
1759		v.sem_otime	 = lower_32_bits(in->sem_otime);
1760		v.sem_otime_high = upper_32_bits(in->sem_otime);
1761		v.sem_ctime	 = lower_32_bits(in->sem_ctime);
1762		v.sem_ctime_high = upper_32_bits(in->sem_ctime);
1763		v.sem_nsems = in->sem_nsems;
1764		return copy_to_user(buf, &v, sizeof(v));
1765	} else {
1766		struct compat_semid_ds v;
1767		memset(&v, 0, sizeof(v));
1768		to_compat_ipc_perm(&v.sem_perm, &in->sem_perm);
1769		v.sem_otime = in->sem_otime;
1770		v.sem_ctime = in->sem_ctime;
1771		v.sem_nsems = in->sem_nsems;
1772		return copy_to_user(buf, &v, sizeof(v));
1773	}
1774}
1775
1776static long compat_ksys_semctl(int semid, int semnum, int cmd, int arg, int version)
1777{
1778	void __user *p = compat_ptr(arg);
1779	struct ipc_namespace *ns;
1780	struct semid64_ds semid64;
1781	int err;
1782
1783	ns = current->nsproxy->ipc_ns;
1784
1785	if (semid < 0)
1786		return -EINVAL;
1787
1788	switch (cmd & (~IPC_64)) {
1789	case IPC_INFO:
1790	case SEM_INFO:
1791		return semctl_info(ns, semid, cmd, p);
1792	case IPC_STAT:
1793	case SEM_STAT:
1794	case SEM_STAT_ANY:
1795		err = semctl_stat(ns, semid, cmd, &semid64);
1796		if (err < 0)
1797			return err;
1798		if (copy_compat_semid_to_user(p, &semid64, version))
1799			err = -EFAULT;
1800		return err;
1801	case GETVAL:
1802	case GETPID:
1803	case GETNCNT:
1804	case GETZCNT:
1805	case GETALL:
1806	case SETALL:
1807		return semctl_main(ns, semid, semnum, cmd, p);
1808	case SETVAL:
1809		return semctl_setval(ns, semid, semnum, arg);
1810	case IPC_SET:
1811		if (copy_compat_semid_from_user(&semid64, p, version))
1812			return -EFAULT;
1813		fallthrough;
1814	case IPC_RMID:
1815		return semctl_down(ns, semid, cmd, &semid64);
1816	default:
1817		return -EINVAL;
1818	}
1819}
1820
1821COMPAT_SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, int, arg)
1822{
1823	return compat_ksys_semctl(semid, semnum, cmd, arg, IPC_64);
1824}
1825
1826#ifdef CONFIG_ARCH_WANT_COMPAT_IPC_PARSE_VERSION
1827long compat_ksys_old_semctl(int semid, int semnum, int cmd, int arg)
1828{
1829	int version = compat_ipc_parse_version(&cmd);
1830
1831	return compat_ksys_semctl(semid, semnum, cmd, arg, version);
1832}
1833
1834COMPAT_SYSCALL_DEFINE4(old_semctl, int, semid, int, semnum, int, cmd, int, arg)
1835{
1836	return compat_ksys_old_semctl(semid, semnum, cmd, arg);
1837}
1838#endif
1839#endif
1840
1841/* If the task doesn't already have a undo_list, then allocate one
1842 * here.  We guarantee there is only one thread using this undo list,
1843 * and current is THE ONE
1844 *
1845 * If this allocation and assignment succeeds, but later
1846 * portions of this code fail, there is no need to free the sem_undo_list.
1847 * Just let it stay associated with the task, and it'll be freed later
1848 * at exit time.
1849 *
1850 * This can block, so callers must hold no locks.
1851 */
1852static inline int get_undo_list(struct sem_undo_list **undo_listp)
1853{
1854	struct sem_undo_list *undo_list;
1855
1856	undo_list = current->sysvsem.undo_list;
1857	if (!undo_list) {
1858		undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
1859		if (undo_list == NULL)
1860			return -ENOMEM;
1861		spin_lock_init(&undo_list->lock);
1862		refcount_set(&undo_list->refcnt, 1);
1863		INIT_LIST_HEAD(&undo_list->list_proc);
1864
1865		current->sysvsem.undo_list = undo_list;
1866	}
1867	*undo_listp = undo_list;
1868	return 0;
1869}
1870
1871static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid)
1872{
1873	struct sem_undo *un;
1874
1875	list_for_each_entry_rcu(un, &ulp->list_proc, list_proc,
1876				spin_is_locked(&ulp->lock)) {
1877		if (un->semid == semid)
1878			return un;
1879	}
1880	return NULL;
1881}
1882
1883static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1884{
1885	struct sem_undo *un;
1886
1887	assert_spin_locked(&ulp->lock);
1888
1889	un = __lookup_undo(ulp, semid);
1890	if (un) {
1891		list_del_rcu(&un->list_proc);
1892		list_add_rcu(&un->list_proc, &ulp->list_proc);
1893	}
1894	return un;
1895}
1896
1897/**
1898 * find_alloc_undo - lookup (and if not present create) undo array
1899 * @ns: namespace
1900 * @semid: semaphore array id
1901 *
1902 * The function looks up (and if not present creates) the undo structure.
1903 * The size of the undo structure depends on the size of the semaphore
1904 * array, thus the alloc path is not that straightforward.
1905 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1906 * performs a rcu_read_lock().
1907 */
1908static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
1909{
1910	struct sem_array *sma;
1911	struct sem_undo_list *ulp;
1912	struct sem_undo *un, *new;
1913	int nsems, error;
 
1914
1915	error = get_undo_list(&ulp);
1916	if (error)
1917		return ERR_PTR(error);
1918
1919	rcu_read_lock();
1920	spin_lock(&ulp->lock);
1921	un = lookup_undo(ulp, semid);
1922	spin_unlock(&ulp->lock);
1923	if (likely(un != NULL))
1924		goto out;
 
1925
1926	/* no undo structure around - allocate one. */
1927	/* step 1: figure out the size of the semaphore array */
1928	sma = sem_obtain_object_check(ns, semid);
1929	if (IS_ERR(sma)) {
1930		rcu_read_unlock();
1931		return ERR_CAST(sma);
1932	}
1933
1934	nsems = sma->sem_nsems;
1935	if (!ipc_rcu_getref(&sma->sem_perm)) {
1936		rcu_read_unlock();
1937		un = ERR_PTR(-EIDRM);
1938		goto out;
1939	}
1940	rcu_read_unlock();
1941
1942	/* step 2: allocate new undo structure */
1943	new = kvzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems,
1944		       GFP_KERNEL);
1945	if (!new) {
1946		ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
1947		return ERR_PTR(-ENOMEM);
1948	}
1949
1950	/* step 3: Acquire the lock on semaphore array */
1951	rcu_read_lock();
1952	sem_lock_and_putref(sma);
1953	if (!ipc_valid_object(&sma->sem_perm)) {
1954		sem_unlock(sma, -1);
1955		rcu_read_unlock();
1956		kvfree(new);
1957		un = ERR_PTR(-EIDRM);
1958		goto out;
1959	}
1960	spin_lock(&ulp->lock);
1961
1962	/*
1963	 * step 4: check for races: did someone else allocate the undo struct?
1964	 */
1965	un = lookup_undo(ulp, semid);
1966	if (un) {
1967		kvfree(new);
1968		goto success;
1969	}
1970	/* step 5: initialize & link new undo structure */
1971	new->semadj = (short *) &new[1];
1972	new->ulp = ulp;
1973	new->semid = semid;
1974	assert_spin_locked(&ulp->lock);
1975	list_add_rcu(&new->list_proc, &ulp->list_proc);
1976	ipc_assert_locked_object(&sma->sem_perm);
1977	list_add(&new->list_id, &sma->list_id);
1978	un = new;
1979
1980success:
1981	spin_unlock(&ulp->lock);
1982	sem_unlock(sma, -1);
 
1983out:
1984	return un;
1985}
1986
1987static long do_semtimedop(int semid, struct sembuf __user *tsops,
1988		unsigned nsops, const struct timespec64 *timeout)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1989{
1990	int error = -EINVAL;
1991	struct sem_array *sma;
1992	struct sembuf fast_sops[SEMOPM_FAST];
1993	struct sembuf *sops = fast_sops, *sop;
1994	struct sem_undo *un;
1995	int max, locknum;
1996	bool undos = false, alter = false, dupsop = false;
1997	struct sem_queue queue;
1998	unsigned long dup = 0, jiffies_left = 0;
1999	struct ipc_namespace *ns;
 
2000
2001	ns = current->nsproxy->ipc_ns;
2002
2003	if (nsops < 1 || semid < 0)
2004		return -EINVAL;
2005	if (nsops > ns->sc_semopm)
2006		return -E2BIG;
2007	if (nsops > SEMOPM_FAST) {
2008		sops = kvmalloc_array(nsops, sizeof(*sops), GFP_KERNEL);
2009		if (sops == NULL)
2010			return -ENOMEM;
2011	}
2012
2013	if (copy_from_user(sops, tsops, nsops * sizeof(*tsops))) {
2014		error =  -EFAULT;
2015		goto out_free;
2016	}
2017
2018	if (timeout) {
2019		if (timeout->tv_sec < 0 || timeout->tv_nsec < 0 ||
2020			timeout->tv_nsec >= 1000000000L) {
 
 
 
 
 
2021			error = -EINVAL;
2022			goto out_free;
2023		}
2024		jiffies_left = timespec64_to_jiffies(timeout);
2025	}
2026
2027	max = 0;
2028	for (sop = sops; sop < sops + nsops; sop++) {
2029		unsigned long mask = 1ULL << ((sop->sem_num) % BITS_PER_LONG);
2030
2031		if (sop->sem_num >= max)
2032			max = sop->sem_num;
2033		if (sop->sem_flg & SEM_UNDO)
2034			undos = true;
2035		if (dup & mask) {
2036			/*
2037			 * There was a previous alter access that appears
2038			 * to have accessed the same semaphore, thus use
2039			 * the dupsop logic. "appears", because the detection
2040			 * can only check % BITS_PER_LONG.
2041			 */
2042			dupsop = true;
2043		}
2044		if (sop->sem_op != 0) {
2045			alter = true;
2046			dup |= mask;
2047		}
2048	}
2049
2050	if (undos) {
2051		/* On success, find_alloc_undo takes the rcu_read_lock */
2052		un = find_alloc_undo(ns, semid);
2053		if (IS_ERR(un)) {
2054			error = PTR_ERR(un);
2055			goto out_free;
2056		}
2057	} else {
2058		un = NULL;
2059		rcu_read_lock();
2060	}
2061
2062	sma = sem_obtain_object_check(ns, semid);
 
 
2063	if (IS_ERR(sma)) {
2064		rcu_read_unlock();
 
2065		error = PTR_ERR(sma);
2066		goto out_free;
2067	}
2068
2069	error = -EFBIG;
2070	if (max >= sma->sem_nsems) {
2071		rcu_read_unlock();
2072		goto out_free;
2073	}
2074
2075	error = -EACCES;
2076	if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO)) {
2077		rcu_read_unlock();
2078		goto out_free;
2079	}
2080
2081	error = security_sem_semop(&sma->sem_perm, sops, nsops, alter);
2082	if (error) {
2083		rcu_read_unlock();
2084		goto out_free;
2085	}
2086
2087	error = -EIDRM;
2088	locknum = sem_lock(sma, sops, nsops);
2089	/*
2090	 * We eventually might perform the following check in a lockless
2091	 * fashion, considering ipc_valid_object() locking constraints.
2092	 * If nsops == 1 and there is no contention for sem_perm.lock, then
2093	 * only a per-semaphore lock is held and it's OK to proceed with the
2094	 * check below. More details on the fine grained locking scheme
2095	 * entangled here and why it's RMID race safe on comments at sem_lock()
2096	 */
2097	if (!ipc_valid_object(&sma->sem_perm))
2098		goto out_unlock_free;
2099	/*
2100	 * semid identifiers are not unique - find_alloc_undo may have
2101	 * allocated an undo structure, it was invalidated by an RMID
2102	 * and now a new array with received the same id. Check and fail.
2103	 * This case can be detected checking un->semid. The existence of
2104	 * "un" itself is guaranteed by rcu.
2105	 */
2106	if (un && un->semid == -1)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2107		goto out_unlock_free;
 
2108
 
 
 
 
2109	queue.sops = sops;
2110	queue.nsops = nsops;
2111	queue.undo = un;
2112	queue.pid = task_tgid(current);
2113	queue.alter = alter;
2114	queue.dupsop = dupsop;
 
 
 
2115
2116	error = perform_atomic_semop(sma, &queue);
2117	if (error == 0) { /* non-blocking successful path */
2118		DEFINE_WAKE_Q(wake_q);
2119
2120		/*
2121		 * If the operation was successful, then do
2122		 * the required updates.
2123		 */
2124		if (alter)
2125			do_smart_update(sma, sops, nsops, 1, &wake_q);
2126		else
2127			set_semotime(sma, sops);
 
 
 
 
2128
2129		sem_unlock(sma, locknum);
2130		rcu_read_unlock();
2131		wake_up_q(&wake_q);
2132
2133		goto out_free;
2134	}
2135	if (error < 0) /* non-blocking error path */
2136		goto out_unlock_free;
2137
2138	/*
2139	 * We need to sleep on this operation, so we put the current
2140	 * task into the pending queue and go to sleep.
2141	 */
2142	if (nsops == 1) {
2143		struct sem *curr;
2144		int idx = array_index_nospec(sops->sem_num, sma->sem_nsems);
2145		curr = &sma->sems[idx];
2146
2147		if (alter) {
2148			if (sma->complex_count) {
2149				list_add_tail(&queue.list,
2150						&sma->pending_alter);
2151			} else {
2152
2153				list_add_tail(&queue.list,
2154						&curr->pending_alter);
2155			}
2156		} else {
2157			list_add_tail(&queue.list, &curr->pending_const);
2158		}
2159	} else {
2160		if (!sma->complex_count)
2161			merge_queues(sma);
2162
2163		if (alter)
2164			list_add_tail(&queue.list, &sma->pending_alter);
2165		else
2166			list_add_tail(&queue.list, &sma->pending_const);
2167
2168		sma->complex_count++;
2169	}
2170
2171	do {
2172		/* memory ordering ensured by the lock in sem_lock() */
2173		WRITE_ONCE(queue.status, -EINTR);
2174		queue.sleeper = current;
2175
2176		/* memory ordering is ensured by the lock in sem_lock() */
2177		__set_current_state(TASK_INTERRUPTIBLE);
2178		sem_unlock(sma, locknum);
2179		rcu_read_unlock();
2180
2181		if (timeout)
2182			jiffies_left = schedule_timeout(jiffies_left);
2183		else
2184			schedule();
2185
2186		/*
2187		 * fastpath: the semop has completed, either successfully or
2188		 * not, from the syscall pov, is quite irrelevant to us at this
2189		 * point; we're done.
2190		 *
2191		 * We _do_ care, nonetheless, about being awoken by a signal or
2192		 * spuriously.  The queue.status is checked again in the
2193		 * slowpath (aka after taking sem_lock), such that we can detect
2194		 * scenarios where we were awakened externally, during the
2195		 * window between wake_q_add() and wake_up_q().
2196		 */
2197		error = READ_ONCE(queue.status);
2198		if (error != -EINTR) {
2199			/* see SEM_BARRIER_2 for purpose/pairing */
2200			smp_acquire__after_ctrl_dep();
2201			goto out_free;
2202		}
2203
2204		rcu_read_lock();
2205		locknum = sem_lock(sma, sops, nsops);
2206
2207		if (!ipc_valid_object(&sma->sem_perm))
2208			goto out_unlock_free;
 
 
2209
2210		/*
2211		 * No necessity for any barrier: We are protect by sem_lock()
2212		 */
2213		error = READ_ONCE(queue.status);
2214
2215		/*
2216		 * If queue.status != -EINTR we are woken up by another process.
2217		 * Leave without unlink_queue(), but with sem_unlock().
2218		 */
2219		if (error != -EINTR)
2220			goto out_unlock_free;
2221
2222		/*
2223		 * If an interrupt occurred we have to clean up the queue.
2224		 */
2225		if (timeout && jiffies_left == 0)
2226			error = -EAGAIN;
2227	} while (error == -EINTR && !signal_pending(current)); /* spurious */
2228
2229	unlink_queue(sma, &queue);
2230
2231out_unlock_free:
2232	sem_unlock(sma, locknum);
2233	rcu_read_unlock();
 
2234out_free:
2235	if (sops != fast_sops)
2236		kvfree(sops);
2237	return error;
2238}
2239
2240long ksys_semtimedop(int semid, struct sembuf __user *tsops,
2241		     unsigned int nsops, const struct __kernel_timespec __user *timeout)
2242{
2243	if (timeout) {
2244		struct timespec64 ts;
2245		if (get_timespec64(&ts, timeout))
2246			return -EFAULT;
2247		return do_semtimedop(semid, tsops, nsops, &ts);
2248	}
2249	return do_semtimedop(semid, tsops, nsops, NULL);
2250}
2251
2252SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
2253		unsigned int, nsops, const struct __kernel_timespec __user *, timeout)
2254{
2255	return ksys_semtimedop(semid, tsops, nsops, timeout);
2256}
2257
2258#ifdef CONFIG_COMPAT_32BIT_TIME
2259long compat_ksys_semtimedop(int semid, struct sembuf __user *tsems,
2260			    unsigned int nsops,
2261			    const struct old_timespec32 __user *timeout)
2262{
2263	if (timeout) {
2264		struct timespec64 ts;
2265		if (get_old_timespec32(&ts, timeout))
2266			return -EFAULT;
2267		return do_semtimedop(semid, tsems, nsops, &ts);
2268	}
2269	return do_semtimedop(semid, tsems, nsops, NULL);
2270}
2271
2272SYSCALL_DEFINE4(semtimedop_time32, int, semid, struct sembuf __user *, tsems,
2273		       unsigned int, nsops,
2274		       const struct old_timespec32 __user *, timeout)
2275{
2276	return compat_ksys_semtimedop(semid, tsems, nsops, timeout);
2277}
2278#endif
2279
2280SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
2281		unsigned, nsops)
2282{
2283	return do_semtimedop(semid, tsops, nsops, NULL);
2284}
2285
2286/* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
2287 * parent and child tasks.
2288 */
2289
2290int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
2291{
2292	struct sem_undo_list *undo_list;
2293	int error;
2294
2295	if (clone_flags & CLONE_SYSVSEM) {
2296		error = get_undo_list(&undo_list);
2297		if (error)
2298			return error;
2299		refcount_inc(&undo_list->refcnt);
2300		tsk->sysvsem.undo_list = undo_list;
2301	} else
2302		tsk->sysvsem.undo_list = NULL;
2303
2304	return 0;
2305}
2306
2307/*
2308 * add semadj values to semaphores, free undo structures.
2309 * undo structures are not freed when semaphore arrays are destroyed
2310 * so some of them may be out of date.
2311 * IMPLEMENTATION NOTE: There is some confusion over whether the
2312 * set of adjustments that needs to be done should be done in an atomic
2313 * manner or not. That is, if we are attempting to decrement the semval
2314 * should we queue up and wait until we can do so legally?
2315 * The original implementation attempted to do this (queue and wait).
2316 * The current implementation does not do so. The POSIX standard
2317 * and SVID should be consulted to determine what behavior is mandated.
2318 */
2319void exit_sem(struct task_struct *tsk)
2320{
2321	struct sem_undo_list *ulp;
2322
2323	ulp = tsk->sysvsem.undo_list;
2324	if (!ulp)
2325		return;
2326	tsk->sysvsem.undo_list = NULL;
2327
2328	if (!refcount_dec_and_test(&ulp->refcnt))
2329		return;
2330
2331	for (;;) {
2332		struct sem_array *sma;
2333		struct sem_undo *un;
2334		int semid, i;
2335		DEFINE_WAKE_Q(wake_q);
2336
2337		cond_resched();
2338
2339		rcu_read_lock();
2340		un = list_entry_rcu(ulp->list_proc.next,
2341				    struct sem_undo, list_proc);
2342		if (&un->list_proc == &ulp->list_proc) {
2343			/*
2344			 * We must wait for freeary() before freeing this ulp,
2345			 * in case we raced with last sem_undo. There is a small
2346			 * possibility where we exit while freeary() didn't
2347			 * finish unlocking sem_undo_list.
2348			 */
2349			spin_lock(&ulp->lock);
2350			spin_unlock(&ulp->lock);
2351			rcu_read_unlock();
2352			break;
2353		}
2354		spin_lock(&ulp->lock);
2355		semid = un->semid;
2356		spin_unlock(&ulp->lock);
2357
2358		/* exit_sem raced with IPC_RMID, nothing to do */
2359		if (semid == -1) {
2360			rcu_read_unlock();
2361			continue;
2362		}
2363
2364		sma = sem_obtain_object_check(tsk->nsproxy->ipc_ns, semid);
2365		/* exit_sem raced with IPC_RMID, nothing to do */
2366		if (IS_ERR(sma)) {
2367			rcu_read_unlock();
2368			continue;
2369		}
2370
2371		sem_lock(sma, NULL, -1);
2372		/* exit_sem raced with IPC_RMID, nothing to do */
2373		if (!ipc_valid_object(&sma->sem_perm)) {
2374			sem_unlock(sma, -1);
2375			rcu_read_unlock();
2376			continue;
2377		}
2378		un = __lookup_undo(ulp, semid);
2379		if (un == NULL) {
2380			/* exit_sem raced with IPC_RMID+semget() that created
2381			 * exactly the same semid. Nothing to do.
2382			 */
2383			sem_unlock(sma, -1);
2384			rcu_read_unlock();
2385			continue;
2386		}
2387
2388		/* remove un from the linked lists */
2389		ipc_assert_locked_object(&sma->sem_perm);
2390		list_del(&un->list_id);
2391
2392		spin_lock(&ulp->lock);
2393		list_del_rcu(&un->list_proc);
2394		spin_unlock(&ulp->lock);
2395
2396		/* perform adjustments registered in un */
2397		for (i = 0; i < sma->sem_nsems; i++) {
2398			struct sem *semaphore = &sma->sems[i];
2399			if (un->semadj[i]) {
2400				semaphore->semval += un->semadj[i];
2401				/*
2402				 * Range checks of the new semaphore value,
2403				 * not defined by sus:
2404				 * - Some unices ignore the undo entirely
2405				 *   (e.g. HP UX 11i 11.22, Tru64 V5.1)
2406				 * - some cap the value (e.g. FreeBSD caps
2407				 *   at 0, but doesn't enforce SEMVMX)
2408				 *
2409				 * Linux caps the semaphore value, both at 0
2410				 * and at SEMVMX.
2411				 *
2412				 *	Manfred <manfred@colorfullife.com>
2413				 */
2414				if (semaphore->semval < 0)
2415					semaphore->semval = 0;
2416				if (semaphore->semval > SEMVMX)
2417					semaphore->semval = SEMVMX;
2418				ipc_update_pid(&semaphore->sempid, task_tgid(current));
2419			}
2420		}
2421		/* maybe some queued-up processes were waiting for this */
2422		do_smart_update(sma, NULL, 0, 1, &wake_q);
2423		sem_unlock(sma, -1);
2424		rcu_read_unlock();
2425		wake_up_q(&wake_q);
2426
2427		kvfree_rcu(un, rcu);
2428	}
2429	kfree(ulp);
2430}
2431
2432#ifdef CONFIG_PROC_FS
2433static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
2434{
2435	struct user_namespace *user_ns = seq_user_ns(s);
2436	struct kern_ipc_perm *ipcp = it;
2437	struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
2438	time64_t sem_otime;
2439
2440	/*
2441	 * The proc interface isn't aware of sem_lock(), it calls
2442	 * ipc_lock_object(), i.e. spin_lock(&sma->sem_perm.lock).
2443	 * (in sysvipc_find_ipc)
2444	 * In order to stay compatible with sem_lock(), we must
2445	 * enter / leave complex_mode.
2446	 */
2447	complexmode_enter(sma);
2448
2449	sem_otime = get_semotime(sma);
2450
2451	seq_printf(s,
2452		   "%10d %10d  %4o %10u %5u %5u %5u %5u %10llu %10llu\n",
2453		   sma->sem_perm.key,
2454		   sma->sem_perm.id,
2455		   sma->sem_perm.mode,
2456		   sma->sem_nsems,
2457		   from_kuid_munged(user_ns, sma->sem_perm.uid),
2458		   from_kgid_munged(user_ns, sma->sem_perm.gid),
2459		   from_kuid_munged(user_ns, sma->sem_perm.cuid),
2460		   from_kgid_munged(user_ns, sma->sem_perm.cgid),
2461		   sem_otime,
2462		   sma->sem_ctime);
2463
2464	complexmode_tryleave(sma);
2465
2466	return 0;
2467}
2468#endif