3 * Copyright (C) 1992 Krishna Balasubramanian
4 * Copyright (C) 1995 Eric Schenk, Bruno Haible
6 * /proc/sysvipc/sem support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
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
13 * (c) 2003 Manfred Spraul <manfred@colorfullife.com>
14 * Further wakeup optimizations, documentation
15 * (c) 2010 Manfred Spraul <manfred@colorfullife.com>
17 * support for audit of ipc object properties and permission changes
18 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
22 * Pavel Emelianov <xemul@openvz.org>
24 * Implementation notes: (May 2010)
25 * This file implements System V semaphores.
27 * User space visible behavior:
28 * - FIFO ordering for semop() operations (just FIFO, not starvation
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
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.
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
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.
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>
90 #include <asm/uaccess.h>
93 /* One semaphore structure for each semaphore in the system. */
95 int semval; /* current value */
96 int sempid; /* pid of last operation */
97 struct list_head sem_pending; /* pending single-sop operations */
100 /* One queue for each sleeping process in the system. */
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? */
113 /* Each task has a list of undo requests. They are executed automatically
114 * when the process exits.
117 struct list_head list_proc; /* per-process list: *
118 * all undos from one process
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 */
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.
132 struct sem_undo_list {
135 struct list_head list_proc;
139 #define sem_ids(ns) ((ns)->ids[IPC_SEM_IDS])
141 #define sem_unlock(sma) ipc_unlock(&(sma)->sem_perm)
142 #define sem_checkid(sma, semid) ipc_checkid(&sma->sem_perm, semid)
144 static int newary(struct ipc_namespace *, struct ipc_params *);
145 static void freeary(struct ipc_namespace *, struct kern_ipc_perm *);
146 #ifdef CONFIG_PROC_FS
147 static int sysvipc_sem_proc_show(struct seq_file *s, void *it);
150 #define SEMMSL_FAST 256 /* 512 bytes on stack */
151 #define SEMOPM_FAST 64 /* ~ 372 bytes on stack */
154 * linked list protection:
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.
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]
167 void sem_init_ns(struct ipc_namespace *ns)
169 ns->sc_semmsl = SEMMSL;
170 ns->sc_semmns = SEMMNS;
171 ns->sc_semopm = SEMOPM;
172 ns->sc_semmni = SEMMNI;
174 ipc_init_ids(&ns->ids[IPC_SEM_IDS]);
178 void sem_exit_ns(struct ipc_namespace *ns)
180 free_ipcs(ns, &sem_ids(ns), freeary);
181 idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr);
185 void __init sem_init (void)
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);
194 * sem_lock_(check_) routines are called in the paths where the rw_mutex
197 static inline struct sem_array *sem_lock(struct ipc_namespace *ns, int id)
199 struct kern_ipc_perm *ipcp = ipc_lock(&sem_ids(ns), id);
202 return (struct sem_array *)ipcp;
204 return container_of(ipcp, struct sem_array, sem_perm);
207 static inline struct sem_array *sem_lock_check(struct ipc_namespace *ns,
210 struct kern_ipc_perm *ipcp = ipc_lock_check(&sem_ids(ns), id);
213 return (struct sem_array *)ipcp;
215 return container_of(ipcp, struct sem_array, sem_perm);
218 static inline void sem_lock_and_putref(struct sem_array *sma)
220 ipc_lock_by_ptr(&sma->sem_perm);
224 static inline void sem_getref_and_unlock(struct sem_array *sma)
227 ipc_unlock(&(sma)->sem_perm);
230 static inline void sem_putref(struct sem_array *sma)
232 ipc_lock_by_ptr(&sma->sem_perm);
234 ipc_unlock(&(sma)->sem_perm);
237 static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
239 ipc_rmid(&sem_ids(ns), &s->sem_perm);
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.
260 * The two-stage algorithm is necessary to protect against the following
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).
277 * newary - Create a new semaphore set
279 * @params: ptr to the structure that contains key, semflg and nsems
281 * Called with sem_ids.rw_mutex held (as a writer)
284 static int newary(struct ipc_namespace *ns, struct ipc_params *params)
288 struct sem_array *sma;
290 key_t key = params->key;
291 int nsems = params->u.nsems;
292 int semflg = params->flg;
297 if (ns->used_sems + nsems > ns->sc_semmns)
300 size = sizeof (*sma) + nsems * sizeof (struct sem);
301 sma = ipc_rcu_alloc(size);
305 memset (sma, 0, size);
307 sma->sem_perm.mode = (semflg & S_IRWXUGO);
308 sma->sem_perm.key = key;
310 sma->sem_perm.security = NULL;
311 retval = security_sem_alloc(sma);
317 id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
319 security_sem_free(sma);
323 ns->used_sems += nsems;
325 sma->sem_base = (struct sem *) &sma[1];
327 for (i = 0; i < nsems; i++)
328 INIT_LIST_HEAD(&sma->sem_base[i].sem_pending);
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();
337 return sma->sem_perm.id;
342 * Called with sem_ids.rw_mutex and ipcp locked.
344 static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg)
346 struct sem_array *sma;
348 sma = container_of(ipcp, struct sem_array, sem_perm);
349 return security_sem_associate(sma, semflg);
353 * Called with sem_ids.rw_mutex and ipcp locked.
355 static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
356 struct ipc_params *params)
358 struct sem_array *sma;
360 sma = container_of(ipcp, struct sem_array, sem_perm);
361 if (params->u.nsems > sma->sem_nsems)
367 SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg)
369 struct ipc_namespace *ns;
370 struct ipc_ops sem_ops;
371 struct ipc_params sem_params;
373 ns = current->nsproxy->ipc_ns;
375 if (nsems < 0 || nsems > ns->sc_semmsl)
378 sem_ops.getnew = newary;
379 sem_ops.associate = sem_security;
380 sem_ops.more_checks = sem_more_checks;
382 sem_params.key = key;
383 sem_params.flg = semflg;
384 sem_params.u.nsems = nsems;
386 return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
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.
394 static int try_atomic_semop (struct sem_array * sma, struct sembuf * sops,
395 int nsops, struct sem_undo *un, int pid)
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;
406 if (!sem_op && result)
414 if (sop->sem_flg & SEM_UNDO) {
415 int undo = un->semadj[sop->sem_num] - sem_op;
417 * Exceeding the undo range is an error.
419 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
422 curr->semval = result;
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;
440 if (sop->sem_flg & IPC_NOWAIT)
447 while (sop >= sops) {
448 sma->sem_base[sop->sem_num].semval -= sop->sem_op;
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
459 * Prepare the wake-up of the queue entry q.
461 static void wake_up_sem_queue_prepare(struct list_head *pt,
462 struct sem_queue *q, int error)
464 if (list_empty(pt)) {
466 * Hold preempt off so that we don't get preempted and have the
467 * wakee busy-wait until we're scheduled back on.
471 q->status = IN_WAKEUP;
474 list_add_tail(&q->simple_list, pt);
478 * wake_up_sem_queue_do(pt) - do the actual wake-up
479 * @pt: list of tasks to be woken up
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.
486 static void wake_up_sem_queue_do(struct list_head *pt)
488 struct sem_queue *q, *t;
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. */
502 static void unlink_queue(struct sem_array *sma, struct sem_queue *q)
506 list_del(&q->simple_list);
508 sma->complex_count--;
511 /** check_restart(sma, q)
512 * @sma: semaphore array
513 * @q: the operation that just completed
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
520 static int check_restart(struct sem_array *sma, struct sem_queue *q)
525 /* if the operation didn't modify the array, then no restart */
529 /* pending complex operations are too difficult to analyse */
530 if (sma->complex_count)
533 /* we were a sleeping complex operation. Too difficult */
537 curr = sma->sem_base + q->sops[0].sem_num;
539 /* No-one waits on this queue */
540 if (list_empty(&curr->sem_pending))
543 /* the new semaphore value */
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.
555 BUG_ON(q->sops[0].sem_op >= 0);
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
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);
566 /* Yes, there is a wait-for-zero semop. Restart */
567 if (h->sops[0].sem_op == 0)
570 /* Again - no-one is waiting for the new value. */
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.
581 * update_queue must be called after a semaphore in a semaphore array
582 * was modified. If multiple semaphore were modified, then @semnum
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.
588 static int update_queue(struct sem_array *sma, int semnum, struct list_head *pt)
591 struct list_head *walk;
592 struct list_head *pending_list;
594 int semop_completed = 0;
596 /* if there are complex operations around, then knowing the semaphore
597 * that was modified doesn't help us. Assume that multiple semaphores
600 if (sma->complex_count)
604 pending_list = &sma->sem_pending;
605 offset = offsetof(struct sem_queue, list);
607 pending_list = &sma->sem_base[semnum].sem_pending;
608 offset = offsetof(struct sem_queue, simple_list);
612 walk = pending_list->next;
613 while (walk != pending_list) {
616 q = (struct sem_queue *)((char *)walk - offset);
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.
626 if (semnum != -1 && sma->sem_base[semnum].semval == 0 &&
630 error = try_atomic_semop(sma, q->sops, q->nsops,
633 /* Does q->sleeper still need to sleep? */
637 unlink_queue(sma, q);
643 restart = check_restart(sma, q);
646 wake_up_sem_queue_prepare(pt, q, error);
650 return semop_completed;
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.
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.
667 static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops,
668 int otime, struct list_head *pt)
672 if (sma->complex_count || sops == NULL) {
673 if (update_queue(sma, -1, pt))
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))
687 sma->sem_otime = get_seconds();
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.
700 static int count_semncnt (struct sem_array * sma, ushort semnum)
703 struct sem_queue * q;
706 list_for_each_entry(q, &sma->sem_pending, list) {
707 struct sembuf * sops = q->sops;
708 int nsops = q->nsops;
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))
719 static int count_semzcnt (struct sem_array * sma, ushort semnum)
722 struct sem_queue * q;
725 list_for_each_entry(q, &sma->sem_pending, list) {
726 struct sembuf * sops = q->sops;
727 int nsops = q->nsops;
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))
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.
742 static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
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;
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);
755 list_del_rcu(&un->list_proc);
756 spin_unlock(&un->ulp->lock);
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);
767 /* Remove the semaphore set from the IDR */
771 wake_up_sem_queue_do(&tasks);
772 ns->used_sems -= sma->sem_nsems;
773 security_sem_free(sma);
777 static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
781 return copy_to_user(buf, in, sizeof(*in));
786 memset(&out, 0, sizeof(out));
788 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
790 out.sem_otime = in->sem_otime;
791 out.sem_ctime = in->sem_ctime;
792 out.sem_nsems = in->sem_nsems;
794 return copy_to_user(buf, &out, sizeof(out));
801 static int semctl_nolock(struct ipc_namespace *ns, int semid,
802 int cmd, int version, union semun arg)
805 struct sem_array *sma;
811 struct seminfo seminfo;
814 err = security_sem_semctl(NULL, cmd);
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;
832 seminfo.semusz = SEMUSZ;
833 seminfo.semaem = SEMAEM;
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)))
839 return (max_id < 0) ? 0: max_id;
844 struct semid64_ds tbuf;
847 if (cmd == SEM_STAT) {
848 sma = sem_lock(ns, semid);
851 id = sma->sem_perm.id;
853 sma = sem_lock_check(ns, semid);
860 if (ipcperms(ns, &sma->sem_perm, S_IRUGO))
863 err = security_sem_semctl(sma, cmd);
867 memset(&tbuf, 0, sizeof(tbuf));
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;
874 if (copy_semid_to_user (arg.buf, &tbuf, version))
886 static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
887 int cmd, int version, union semun arg)
889 struct sem_array *sma;
892 ushort fast_sem_io[SEMMSL_FAST];
893 ushort* sem_io = fast_sem_io;
895 struct list_head tasks;
897 sma = sem_lock_check(ns, semid);
901 INIT_LIST_HEAD(&tasks);
902 nsems = sma->sem_nsems;
905 if (ipcperms(ns, &sma->sem_perm,
906 (cmd == SETVAL || cmd == SETALL) ? S_IWUGO : S_IRUGO))
909 err = security_sem_semctl(sma, cmd);
917 ushort __user *array = arg.array;
920 if(nsems > SEMMSL_FAST) {
921 sem_getref_and_unlock(sma);
923 sem_io = ipc_alloc(sizeof(ushort)*nsems);
929 sem_lock_and_putref(sma);
930 if (sma->sem_perm.deleted) {
937 for (i = 0; i < sma->sem_nsems; i++)
938 sem_io[i] = sma->sem_base[i].semval;
941 if(copy_to_user(array, sem_io, nsems*sizeof(ushort)))
950 sem_getref_and_unlock(sma);
952 if(nsems > SEMMSL_FAST) {
953 sem_io = ipc_alloc(sizeof(ushort)*nsems);
960 if (copy_from_user (sem_io, arg.array, nsems*sizeof(ushort))) {
966 for (i = 0; i < nsems; i++) {
967 if (sem_io[i] > SEMVMX) {
973 sem_lock_and_putref(sma);
974 if (sma->sem_perm.deleted) {
980 for (i = 0; i < nsems; i++)
981 sma->sem_base[i].semval = sem_io[i];
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++)
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);
994 /* GETVAL, GETPID, GETNCTN, GETZCNT, SETVAL: fall-through */
997 if(semnum < 0 || semnum >= nsems)
1000 curr = &sma->sem_base[semnum];
1010 err = count_semncnt(sma,semnum);
1013 err = count_semzcnt(sma,semnum);
1018 struct sem_undo *un;
1021 if (val > SEMVMX || val < 0)
1024 assert_spin_locked(&sma->sem_perm.lock);
1025 list_for_each_entry(un, &sma->list_id, list_id)
1026 un->semadj[semnum] = 0;
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);
1039 wake_up_sem_queue_do(&tasks);
1042 if(sem_io != fast_sem_io)
1043 ipc_free(sem_io, sizeof(ushort)*nsems);
1047 static inline unsigned long
1048 copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
1052 if (copy_from_user(out, buf, sizeof(*out)))
1057 struct semid_ds tbuf_old;
1059 if(copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
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;
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.
1078 static int semctl_down(struct ipc_namespace *ns, int semid,
1079 int cmd, int version, union semun arg)
1081 struct sem_array *sma;
1083 struct semid64_ds semid64;
1084 struct kern_ipc_perm *ipcp;
1086 if(cmd == IPC_SET) {
1087 if (copy_semid_from_user(&semid64, arg.buf, version))
1091 ipcp = ipcctl_pre_down(ns, &sem_ids(ns), semid, cmd,
1092 &semid64.sem_perm, 0);
1094 return PTR_ERR(ipcp);
1096 sma = container_of(ipcp, struct sem_array, sem_perm);
1098 err = security_sem_semctl(sma, cmd);
1107 ipc_update_perm(&semid64.sem_perm, ipcp);
1108 sma->sem_ctime = get_seconds();
1117 up_write(&sem_ids(ns).rw_mutex);
1121 SYSCALL_DEFINE(semctl)(int semid, int semnum, int cmd, union semun arg)
1125 struct ipc_namespace *ns;
1130 version = ipc_parse_version(&cmd);
1131 ns = current->nsproxy->ipc_ns;
1138 err = semctl_nolock(ns, semid, cmd, version, arg);
1147 err = semctl_main(ns,semid,semnum,cmd,version,arg);
1151 err = semctl_down(ns, semid, cmd, version, arg);
1157 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
1158 asmlinkage long SyS_semctl(int semid, int semnum, int cmd, union semun arg)
1160 return SYSC_semctl((int) semid, (int) semnum, (int) cmd, arg);
1162 SYSCALL_ALIAS(sys_semctl, SyS_semctl);
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
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
1174 * This can block, so callers must hold no locks.
1176 static inline int get_undo_list(struct sem_undo_list **undo_listp)
1178 struct sem_undo_list *undo_list;
1180 undo_list = current->sysvsem.undo_list;
1182 undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
1183 if (undo_list == NULL)
1185 spin_lock_init(&undo_list->lock);
1186 atomic_set(&undo_list->refcnt, 1);
1187 INIT_LIST_HEAD(&undo_list->list_proc);
1189 current->sysvsem.undo_list = undo_list;
1191 *undo_listp = undo_list;
1195 static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid)
1197 struct sem_undo *un;
1199 list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) {
1200 if (un->semid == semid)
1206 static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1208 struct sem_undo *un;
1210 assert_spin_locked(&ulp->lock);
1212 un = __lookup_undo(ulp, semid);
1214 list_del_rcu(&un->list_proc);
1215 list_add_rcu(&un->list_proc, &ulp->list_proc);
1221 * find_alloc_undo - Lookup (and if not present create) undo array
1223 * @semid: semaphore array id
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().
1231 static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
1233 struct sem_array *sma;
1234 struct sem_undo_list *ulp;
1235 struct sem_undo *un, *new;
1239 error = get_undo_list(&ulp);
1241 return ERR_PTR(error);
1244 spin_lock(&ulp->lock);
1245 un = lookup_undo(ulp, semid);
1246 spin_unlock(&ulp->lock);
1247 if (likely(un!=NULL))
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);
1255 return ERR_CAST(sma);
1257 nsems = sma->sem_nsems;
1258 sem_getref_and_unlock(sma);
1260 /* step 2: allocate new undo structure */
1261 new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1264 return ERR_PTR(-ENOMEM);
1267 /* step 3: Acquire the lock on semaphore array */
1268 sem_lock_and_putref(sma);
1269 if (sma->sem_perm.deleted) {
1272 un = ERR_PTR(-EIDRM);
1275 spin_lock(&ulp->lock);
1278 * step 4: check for races: did someone else allocate the undo struct?
1280 un = lookup_undo(ulp, semid);
1285 /* step 5: initialize & link new undo structure */
1286 new->semadj = (short *) &new[1];
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);
1296 spin_unlock(&ulp->lock);
1305 * get_queue_result - Retrieve the result code from sem_queue
1306 * @q: Pointer to queue structure
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.
1314 * The function can be called with or without holding the semaphore spinlock.
1316 static int get_queue_result(struct sem_queue *q)
1321 while (unlikely(error == IN_WAKEUP)) {
1330 SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
1331 unsigned, nsops, const struct timespec __user *, timeout)
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;
1344 ns = current->nsproxy->ipc_ns;
1346 if (nsops < 1 || semid < 0)
1348 if (nsops > ns->sc_semopm)
1350 if(nsops > SEMOPM_FAST) {
1351 sops = kmalloc(sizeof(*sops)*nsops,GFP_KERNEL);
1355 if (copy_from_user (sops, tsops, nsops * sizeof(*tsops))) {
1360 struct timespec _timeout;
1361 if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1365 if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1366 _timeout.tv_nsec >= 1000000000L) {
1370 jiffies_left = timespec_to_jiffies(&_timeout);
1373 for (sop = sops; sop < sops + nsops; sop++) {
1374 if (sop->sem_num >= max)
1376 if (sop->sem_flg & SEM_UNDO)
1378 if (sop->sem_op != 0)
1383 un = find_alloc_undo(ns, semid);
1385 error = PTR_ERR(un);
1391 INIT_LIST_HEAD(&tasks);
1393 sma = sem_lock_check(ns, semid);
1397 error = PTR_ERR(sma);
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.
1410 if (un->semid == -1) {
1412 goto out_unlock_free;
1415 * rcu lock can be released, "un" cannot disappear:
1416 * - sem_lock is acquired, thus IPC_RMID is
1418 * - exit_sem is impossible, it always operates on
1419 * current (or a dead task).
1427 if (max >= sma->sem_nsems)
1428 goto out_unlock_free;
1431 if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO))
1432 goto out_unlock_free;
1434 error = security_sem_semop(sma, sops, nsops, alter);
1436 goto out_unlock_free;
1438 error = try_atomic_semop (sma, sops, nsops, un, task_tgid_vnr(current));
1440 if (alter && error == 0)
1441 do_smart_update(sma, sops, nsops, 1, &tasks);
1443 goto out_unlock_free;
1446 /* We need to sleep on this operation, so we put the current
1447 * task into the pending queue and go to sleep.
1451 queue.nsops = nsops;
1453 queue.pid = task_tgid_vnr(current);
1454 queue.alter = alter;
1456 list_add_tail(&queue.list, &sma->sem_pending);
1458 list_add(&queue.list, &sma->sem_pending);
1462 curr = &sma->sem_base[sops->sem_num];
1465 list_add_tail(&queue.simple_list, &curr->sem_pending);
1467 list_add(&queue.simple_list, &curr->sem_pending);
1469 INIT_LIST_HEAD(&queue.simple_list);
1470 sma->complex_count++;
1473 queue.status = -EINTR;
1474 queue.sleeper = current;
1477 current->state = TASK_INTERRUPTIBLE;
1481 jiffies_left = schedule_timeout(jiffies_left);
1485 error = get_queue_result(&queue);
1487 if (error != -EINTR) {
1488 /* fast path: update_queue already obtained all requested
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.
1500 sma = sem_lock(ns, semid);
1503 * Wait until it's guaranteed that no wakeup_sem_queue_do() is ongoing.
1505 error = get_queue_result(&queue);
1508 * Array removed? If yes, leave without sem_unlock().
1516 * If queue.status != -EINTR we are woken up by another process.
1517 * Leave without unlink_queue(), but with sem_unlock().
1520 if (error != -EINTR) {
1521 goto out_unlock_free;
1525 * If an interrupt occurred we have to clean up the queue
1527 if (timeout && jiffies_left == 0)
1531 * If the wakeup was spurious, just retry
1533 if (error == -EINTR && !signal_pending(current))
1536 unlink_queue(sma, &queue);
1541 wake_up_sem_queue_do(&tasks);
1543 if(sops != fast_sops)
1548 SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
1551 return sys_semtimedop(semid, tsops, nsops, NULL);
1554 /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
1555 * parent and child tasks.
1558 int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
1560 struct sem_undo_list *undo_list;
1563 if (clone_flags & CLONE_SYSVSEM) {
1564 error = get_undo_list(&undo_list);
1567 atomic_inc(&undo_list->refcnt);
1568 tsk->sysvsem.undo_list = undo_list;
1570 tsk->sysvsem.undo_list = NULL;
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.
1587 void exit_sem(struct task_struct *tsk)
1589 struct sem_undo_list *ulp;
1591 ulp = tsk->sysvsem.undo_list;
1594 tsk->sysvsem.undo_list = NULL;
1596 if (!atomic_dec_and_test(&ulp->refcnt))
1600 struct sem_array *sma;
1601 struct sem_undo *un;
1602 struct list_head tasks;
1607 un = list_entry_rcu(ulp->list_proc.next,
1608 struct sem_undo, list_proc);
1609 if (&un->list_proc == &ulp->list_proc)
1618 sma = sem_lock_check(tsk->nsproxy->ipc_ns, un->semid);
1620 /* exit_sem raced with IPC_RMID, nothing to do */
1624 un = __lookup_undo(ulp, semid);
1626 /* exit_sem raced with IPC_RMID+semget() that created
1627 * exactly the same semid. Nothing to do.
1633 /* remove un from the linked lists */
1634 assert_spin_locked(&sma->sem_perm.lock);
1635 list_del(&un->list_id);
1637 spin_lock(&ulp->lock);
1638 list_del_rcu(&un->list_proc);
1639 spin_unlock(&ulp->lock);
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];
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)
1654 * Linux caps the semaphore value, both at 0
1657 * Manfred <manfred@colorfullife.com>
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);
1666 /* maybe some queued-up processes were waiting for this */
1667 INIT_LIST_HEAD(&tasks);
1668 do_smart_update(sma, NULL, 0, 1, &tasks);
1670 wake_up_sem_queue_do(&tasks);
1677 #ifdef CONFIG_PROC_FS
1678 static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
1680 struct sem_array *sma = it;
1682 return seq_printf(s,
1683 "%10d %10d %4o %10u %5u %5u %5u %5u %10lu %10lu\n",