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 #define sem_ids(ns) ((ns)->ids[IPC_SEM_IDS])
95 #define sem_unlock(sma) ipc_unlock(&(sma)->sem_perm)
96 #define sem_checkid(sma, semid) ipc_checkid(&sma->sem_perm, semid)
98 static int newary(struct ipc_namespace *, struct ipc_params *);
99 static void freeary(struct ipc_namespace *, struct kern_ipc_perm *);
100 #ifdef CONFIG_PROC_FS
101 static int sysvipc_sem_proc_show(struct seq_file *s, void *it);
104 #define SEMMSL_FAST 256 /* 512 bytes on stack */
105 #define SEMOPM_FAST 64 /* ~ 372 bytes on stack */
108 * linked list protection:
110 * sem_array.sem_pending{,last},
111 * sem_array.sem_undo: sem_lock() for read/write
112 * sem_undo.proc_next: only "current" is allowed to read/write that field.
116 #define sc_semmsl sem_ctls[0]
117 #define sc_semmns sem_ctls[1]
118 #define sc_semopm sem_ctls[2]
119 #define sc_semmni sem_ctls[3]
121 void sem_init_ns(struct ipc_namespace *ns)
123 ns->sc_semmsl = SEMMSL;
124 ns->sc_semmns = SEMMNS;
125 ns->sc_semopm = SEMOPM;
126 ns->sc_semmni = SEMMNI;
128 ipc_init_ids(&ns->ids[IPC_SEM_IDS]);
132 void sem_exit_ns(struct ipc_namespace *ns)
134 free_ipcs(ns, &sem_ids(ns), freeary);
135 idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr);
139 void __init sem_init (void)
141 sem_init_ns(&init_ipc_ns);
142 ipc_init_proc_interface("sysvipc/sem",
143 " key semid perms nsems uid gid cuid cgid otime ctime\n",
144 IPC_SEM_IDS, sysvipc_sem_proc_show);
148 * sem_lock_(check_) routines are called in the paths where the rw_mutex
151 static inline struct sem_array *sem_lock(struct ipc_namespace *ns, int id)
153 struct kern_ipc_perm *ipcp = ipc_lock(&sem_ids(ns), id);
156 return (struct sem_array *)ipcp;
158 return container_of(ipcp, struct sem_array, sem_perm);
161 static inline struct sem_array *sem_lock_check(struct ipc_namespace *ns,
164 struct kern_ipc_perm *ipcp = ipc_lock_check(&sem_ids(ns), id);
167 return (struct sem_array *)ipcp;
169 return container_of(ipcp, struct sem_array, sem_perm);
172 static inline void sem_lock_and_putref(struct sem_array *sma)
174 ipc_lock_by_ptr(&sma->sem_perm);
178 static inline void sem_getref_and_unlock(struct sem_array *sma)
181 ipc_unlock(&(sma)->sem_perm);
184 static inline void sem_putref(struct sem_array *sma)
186 ipc_lock_by_ptr(&sma->sem_perm);
188 ipc_unlock(&(sma)->sem_perm);
191 static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
193 ipc_rmid(&sem_ids(ns), &s->sem_perm);
197 * Lockless wakeup algorithm:
198 * Without the check/retry algorithm a lockless wakeup is possible:
199 * - queue.status is initialized to -EINTR before blocking.
200 * - wakeup is performed by
201 * * unlinking the queue entry from sma->sem_pending
202 * * setting queue.status to IN_WAKEUP
203 * This is the notification for the blocked thread that a
204 * result value is imminent.
205 * * call wake_up_process
206 * * set queue.status to the final value.
207 * - the previously blocked thread checks queue.status:
208 * * if it's IN_WAKEUP, then it must wait until the value changes
209 * * if it's not -EINTR, then the operation was completed by
210 * update_queue. semtimedop can return queue.status without
211 * performing any operation on the sem array.
212 * * otherwise it must acquire the spinlock and check what's up.
214 * The two-stage algorithm is necessary to protect against the following
216 * - if queue.status is set after wake_up_process, then the woken up idle
217 * thread could race forward and try (and fail) to acquire sma->lock
218 * before update_queue had a chance to set queue.status
219 * - if queue.status is written before wake_up_process and if the
220 * blocked process is woken up by a signal between writing
221 * queue.status and the wake_up_process, then the woken up
222 * process could return from semtimedop and die by calling
223 * sys_exit before wake_up_process is called. Then wake_up_process
224 * will oops, because the task structure is already invalid.
225 * (yes, this happened on s390 with sysv msg).
231 * newary - Create a new semaphore set
233 * @params: ptr to the structure that contains key, semflg and nsems
235 * Called with sem_ids.rw_mutex held (as a writer)
238 static int newary(struct ipc_namespace *ns, struct ipc_params *params)
242 struct sem_array *sma;
244 key_t key = params->key;
245 int nsems = params->u.nsems;
246 int semflg = params->flg;
251 if (ns->used_sems + nsems > ns->sc_semmns)
254 size = sizeof (*sma) + nsems * sizeof (struct sem);
255 sma = ipc_rcu_alloc(size);
259 memset (sma, 0, size);
261 sma->sem_perm.mode = (semflg & S_IRWXUGO);
262 sma->sem_perm.key = key;
264 sma->sem_perm.security = NULL;
265 retval = security_sem_alloc(sma);
271 id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
273 security_sem_free(sma);
277 ns->used_sems += nsems;
279 sma->sem_base = (struct sem *) &sma[1];
281 for (i = 0; i < nsems; i++)
282 INIT_LIST_HEAD(&sma->sem_base[i].sem_pending);
284 sma->complex_count = 0;
285 INIT_LIST_HEAD(&sma->sem_pending);
286 INIT_LIST_HEAD(&sma->list_id);
287 sma->sem_nsems = nsems;
288 sma->sem_ctime = get_seconds();
291 return sma->sem_perm.id;
296 * Called with sem_ids.rw_mutex and ipcp locked.
298 static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg)
300 struct sem_array *sma;
302 sma = container_of(ipcp, struct sem_array, sem_perm);
303 return security_sem_associate(sma, semflg);
307 * Called with sem_ids.rw_mutex and ipcp locked.
309 static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
310 struct ipc_params *params)
312 struct sem_array *sma;
314 sma = container_of(ipcp, struct sem_array, sem_perm);
315 if (params->u.nsems > sma->sem_nsems)
321 SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg)
323 struct ipc_namespace *ns;
324 struct ipc_ops sem_ops;
325 struct ipc_params sem_params;
327 ns = current->nsproxy->ipc_ns;
329 if (nsems < 0 || nsems > ns->sc_semmsl)
332 sem_ops.getnew = newary;
333 sem_ops.associate = sem_security;
334 sem_ops.more_checks = sem_more_checks;
336 sem_params.key = key;
337 sem_params.flg = semflg;
338 sem_params.u.nsems = nsems;
340 return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
344 * Determine whether a sequence of semaphore operations would succeed
345 * all at once. Return 0 if yes, 1 if need to sleep, else return error code.
348 static int try_atomic_semop (struct sem_array * sma, struct sembuf * sops,
349 int nsops, struct sem_undo *un, int pid)
355 for (sop = sops; sop < sops + nsops; sop++) {
356 curr = sma->sem_base + sop->sem_num;
357 sem_op = sop->sem_op;
358 result = curr->semval;
360 if (!sem_op && result)
368 if (sop->sem_flg & SEM_UNDO) {
369 int undo = un->semadj[sop->sem_num] - sem_op;
371 * Exceeding the undo range is an error.
373 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
376 curr->semval = result;
380 while (sop >= sops) {
381 sma->sem_base[sop->sem_num].sempid = pid;
382 if (sop->sem_flg & SEM_UNDO)
383 un->semadj[sop->sem_num] -= sop->sem_op;
394 if (sop->sem_flg & IPC_NOWAIT)
401 while (sop >= sops) {
402 sma->sem_base[sop->sem_num].semval -= sop->sem_op;
409 /** wake_up_sem_queue_prepare(q, error): Prepare wake-up
410 * @q: queue entry that must be signaled
411 * @error: Error value for the signal
413 * Prepare the wake-up of the queue entry q.
415 static void wake_up_sem_queue_prepare(struct list_head *pt,
416 struct sem_queue *q, int error)
418 if (list_empty(pt)) {
420 * Hold preempt off so that we don't get preempted and have the
421 * wakee busy-wait until we're scheduled back on.
425 q->status = IN_WAKEUP;
428 list_add_tail(&q->simple_list, pt);
432 * wake_up_sem_queue_do(pt) - do the actual wake-up
433 * @pt: list of tasks to be woken up
435 * Do the actual wake-up.
436 * The function is called without any locks held, thus the semaphore array
437 * could be destroyed already and the tasks can disappear as soon as the
438 * status is set to the actual return code.
440 static void wake_up_sem_queue_do(struct list_head *pt)
442 struct sem_queue *q, *t;
445 did_something = !list_empty(pt);
446 list_for_each_entry_safe(q, t, pt, simple_list) {
447 wake_up_process(q->sleeper);
448 /* q can disappear immediately after writing q->status. */
456 static void unlink_queue(struct sem_array *sma, struct sem_queue *q)
460 list_del(&q->simple_list);
462 sma->complex_count--;
465 /** check_restart(sma, q)
466 * @sma: semaphore array
467 * @q: the operation that just completed
469 * update_queue is O(N^2) when it restarts scanning the whole queue of
470 * waiting operations. Therefore this function checks if the restart is
471 * really necessary. It is called after a previously waiting operation
474 static int check_restart(struct sem_array *sma, struct sem_queue *q)
479 /* if the operation didn't modify the array, then no restart */
483 /* pending complex operations are too difficult to analyse */
484 if (sma->complex_count)
487 /* we were a sleeping complex operation. Too difficult */
491 curr = sma->sem_base + q->sops[0].sem_num;
493 /* No-one waits on this queue */
494 if (list_empty(&curr->sem_pending))
497 /* the new semaphore value */
499 /* It is impossible that someone waits for the new value:
500 * - q is a previously sleeping simple operation that
501 * altered the array. It must be a decrement, because
502 * simple increments never sleep.
503 * - The value is not 0, thus wait-for-zero won't proceed.
504 * - If there are older (higher priority) decrements
505 * in the queue, then they have observed the original
506 * semval value and couldn't proceed. The operation
507 * decremented to value - thus they won't proceed either.
509 BUG_ON(q->sops[0].sem_op >= 0);
513 * semval is 0. Check if there are wait-for-zero semops.
514 * They must be the first entries in the per-semaphore simple queue
516 h = list_first_entry(&curr->sem_pending, struct sem_queue, simple_list);
517 BUG_ON(h->nsops != 1);
518 BUG_ON(h->sops[0].sem_num != q->sops[0].sem_num);
520 /* Yes, there is a wait-for-zero semop. Restart */
521 if (h->sops[0].sem_op == 0)
524 /* Again - no-one is waiting for the new value. */
530 * update_queue(sma, semnum): Look for tasks that can be completed.
531 * @sma: semaphore array.
532 * @semnum: semaphore that was modified.
533 * @pt: list head for the tasks that must be woken up.
535 * update_queue must be called after a semaphore in a semaphore array
536 * was modified. If multiple semaphore were modified, then @semnum
538 * The tasks that must be woken up are added to @pt. The return code
539 * is stored in q->pid.
540 * The function return 1 if at least one semop was completed successfully.
542 static int update_queue(struct sem_array *sma, int semnum, struct list_head *pt)
545 struct list_head *walk;
546 struct list_head *pending_list;
548 int semop_completed = 0;
550 /* if there are complex operations around, then knowing the semaphore
551 * that was modified doesn't help us. Assume that multiple semaphores
554 if (sma->complex_count)
558 pending_list = &sma->sem_pending;
559 offset = offsetof(struct sem_queue, list);
561 pending_list = &sma->sem_base[semnum].sem_pending;
562 offset = offsetof(struct sem_queue, simple_list);
566 walk = pending_list->next;
567 while (walk != pending_list) {
570 q = (struct sem_queue *)((char *)walk - offset);
573 /* If we are scanning the single sop, per-semaphore list of
574 * one semaphore and that semaphore is 0, then it is not
575 * necessary to scan the "alter" entries: simple increments
576 * that affect only one entry succeed immediately and cannot
577 * be in the per semaphore pending queue, and decrements
578 * cannot be successful if the value is already 0.
580 if (semnum != -1 && sma->sem_base[semnum].semval == 0 &&
584 error = try_atomic_semop(sma, q->sops, q->nsops,
587 /* Does q->sleeper still need to sleep? */
591 unlink_queue(sma, q);
597 restart = check_restart(sma, q);
600 wake_up_sem_queue_prepare(pt, q, error);
604 return semop_completed;
608 * do_smart_update(sma, sops, nsops, otime, pt) - optimized update_queue
609 * @sma: semaphore array
610 * @sops: operations that were performed
611 * @nsops: number of operations
612 * @otime: force setting otime
613 * @pt: list head of the tasks that must be woken up.
615 * do_smart_update() does the required called to update_queue, based on the
616 * actual changes that were performed on the semaphore array.
617 * Note that the function does not do the actual wake-up: the caller is
618 * responsible for calling wake_up_sem_queue_do(@pt).
619 * It is safe to perform this call after dropping all locks.
621 static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops,
622 int otime, struct list_head *pt)
626 if (sma->complex_count || sops == NULL) {
627 if (update_queue(sma, -1, pt))
632 for (i = 0; i < nsops; i++) {
633 if (sops[i].sem_op > 0 ||
634 (sops[i].sem_op < 0 &&
635 sma->sem_base[sops[i].sem_num].semval == 0))
636 if (update_queue(sma, sops[i].sem_num, pt))
641 sma->sem_otime = get_seconds();
645 /* The following counts are associated to each semaphore:
646 * semncnt number of tasks waiting on semval being nonzero
647 * semzcnt number of tasks waiting on semval being zero
648 * This model assumes that a task waits on exactly one semaphore.
649 * Since semaphore operations are to be performed atomically, tasks actually
650 * wait on a whole sequence of semaphores simultaneously.
651 * The counts we return here are a rough approximation, but still
652 * warrant that semncnt+semzcnt>0 if the task is on the pending queue.
654 static int count_semncnt (struct sem_array * sma, ushort semnum)
657 struct sem_queue * q;
660 list_for_each_entry(q, &sma->sem_pending, list) {
661 struct sembuf * sops = q->sops;
662 int nsops = q->nsops;
664 for (i = 0; i < nsops; i++)
665 if (sops[i].sem_num == semnum
666 && (sops[i].sem_op < 0)
667 && !(sops[i].sem_flg & IPC_NOWAIT))
673 static int count_semzcnt (struct sem_array * sma, ushort semnum)
676 struct sem_queue * q;
679 list_for_each_entry(q, &sma->sem_pending, list) {
680 struct sembuf * sops = q->sops;
681 int nsops = q->nsops;
683 for (i = 0; i < nsops; i++)
684 if (sops[i].sem_num == semnum
685 && (sops[i].sem_op == 0)
686 && !(sops[i].sem_flg & IPC_NOWAIT))
692 static void free_un(struct rcu_head *head)
694 struct sem_undo *un = container_of(head, struct sem_undo, rcu);
698 /* Free a semaphore set. freeary() is called with sem_ids.rw_mutex locked
699 * as a writer and the spinlock for this semaphore set hold. sem_ids.rw_mutex
700 * remains locked on exit.
702 static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
704 struct sem_undo *un, *tu;
705 struct sem_queue *q, *tq;
706 struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
707 struct list_head tasks;
709 /* Free the existing undo structures for this semaphore set. */
710 assert_spin_locked(&sma->sem_perm.lock);
711 list_for_each_entry_safe(un, tu, &sma->list_id, list_id) {
712 list_del(&un->list_id);
713 spin_lock(&un->ulp->lock);
715 list_del_rcu(&un->list_proc);
716 spin_unlock(&un->ulp->lock);
717 call_rcu(&un->rcu, free_un);
720 /* Wake up all pending processes and let them fail with EIDRM. */
721 INIT_LIST_HEAD(&tasks);
722 list_for_each_entry_safe(q, tq, &sma->sem_pending, list) {
723 unlink_queue(sma, q);
724 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
727 /* Remove the semaphore set from the IDR */
731 wake_up_sem_queue_do(&tasks);
732 ns->used_sems -= sma->sem_nsems;
733 security_sem_free(sma);
737 static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
741 return copy_to_user(buf, in, sizeof(*in));
746 memset(&out, 0, sizeof(out));
748 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
750 out.sem_otime = in->sem_otime;
751 out.sem_ctime = in->sem_ctime;
752 out.sem_nsems = in->sem_nsems;
754 return copy_to_user(buf, &out, sizeof(out));
761 static int semctl_nolock(struct ipc_namespace *ns, int semid,
762 int cmd, int version, union semun arg)
765 struct sem_array *sma;
771 struct seminfo seminfo;
774 err = security_sem_semctl(NULL, cmd);
778 memset(&seminfo,0,sizeof(seminfo));
779 seminfo.semmni = ns->sc_semmni;
780 seminfo.semmns = ns->sc_semmns;
781 seminfo.semmsl = ns->sc_semmsl;
782 seminfo.semopm = ns->sc_semopm;
783 seminfo.semvmx = SEMVMX;
784 seminfo.semmnu = SEMMNU;
785 seminfo.semmap = SEMMAP;
786 seminfo.semume = SEMUME;
787 down_read(&sem_ids(ns).rw_mutex);
788 if (cmd == SEM_INFO) {
789 seminfo.semusz = sem_ids(ns).in_use;
790 seminfo.semaem = ns->used_sems;
792 seminfo.semusz = SEMUSZ;
793 seminfo.semaem = SEMAEM;
795 max_id = ipc_get_maxid(&sem_ids(ns));
796 up_read(&sem_ids(ns).rw_mutex);
797 if (copy_to_user (arg.__buf, &seminfo, sizeof(struct seminfo)))
799 return (max_id < 0) ? 0: max_id;
804 struct semid64_ds tbuf;
807 if (cmd == SEM_STAT) {
808 sma = sem_lock(ns, semid);
811 id = sma->sem_perm.id;
813 sma = sem_lock_check(ns, semid);
820 if (ipcperms(ns, &sma->sem_perm, S_IRUGO))
823 err = security_sem_semctl(sma, cmd);
827 memset(&tbuf, 0, sizeof(tbuf));
829 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
830 tbuf.sem_otime = sma->sem_otime;
831 tbuf.sem_ctime = sma->sem_ctime;
832 tbuf.sem_nsems = sma->sem_nsems;
834 if (copy_semid_to_user (arg.buf, &tbuf, version))
846 static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
847 int cmd, int version, union semun arg)
849 struct sem_array *sma;
852 ushort fast_sem_io[SEMMSL_FAST];
853 ushort* sem_io = fast_sem_io;
855 struct list_head tasks;
857 sma = sem_lock_check(ns, semid);
861 INIT_LIST_HEAD(&tasks);
862 nsems = sma->sem_nsems;
865 if (ipcperms(ns, &sma->sem_perm,
866 (cmd == SETVAL || cmd == SETALL) ? S_IWUGO : S_IRUGO))
869 err = security_sem_semctl(sma, cmd);
877 ushort __user *array = arg.array;
880 if(nsems > SEMMSL_FAST) {
881 sem_getref_and_unlock(sma);
883 sem_io = ipc_alloc(sizeof(ushort)*nsems);
889 sem_lock_and_putref(sma);
890 if (sma->sem_perm.deleted) {
897 for (i = 0; i < sma->sem_nsems; i++)
898 sem_io[i] = sma->sem_base[i].semval;
901 if(copy_to_user(array, sem_io, nsems*sizeof(ushort)))
910 sem_getref_and_unlock(sma);
912 if(nsems > SEMMSL_FAST) {
913 sem_io = ipc_alloc(sizeof(ushort)*nsems);
920 if (copy_from_user (sem_io, arg.array, nsems*sizeof(ushort))) {
926 for (i = 0; i < nsems; i++) {
927 if (sem_io[i] > SEMVMX) {
933 sem_lock_and_putref(sma);
934 if (sma->sem_perm.deleted) {
940 for (i = 0; i < nsems; i++)
941 sma->sem_base[i].semval = sem_io[i];
943 assert_spin_locked(&sma->sem_perm.lock);
944 list_for_each_entry(un, &sma->list_id, list_id) {
945 for (i = 0; i < nsems; i++)
948 sma->sem_ctime = get_seconds();
949 /* maybe some queued-up processes were waiting for this */
950 do_smart_update(sma, NULL, 0, 0, &tasks);
954 /* GETVAL, GETPID, GETNCTN, GETZCNT, SETVAL: fall-through */
957 if(semnum < 0 || semnum >= nsems)
960 curr = &sma->sem_base[semnum];
970 err = count_semncnt(sma,semnum);
973 err = count_semzcnt(sma,semnum);
981 if (val > SEMVMX || val < 0)
984 assert_spin_locked(&sma->sem_perm.lock);
985 list_for_each_entry(un, &sma->list_id, list_id)
986 un->semadj[semnum] = 0;
989 curr->sempid = task_tgid_vnr(current);
990 sma->sem_ctime = get_seconds();
991 /* maybe some queued-up processes were waiting for this */
992 do_smart_update(sma, NULL, 0, 0, &tasks);
999 wake_up_sem_queue_do(&tasks);
1002 if(sem_io != fast_sem_io)
1003 ipc_free(sem_io, sizeof(ushort)*nsems);
1007 static inline unsigned long
1008 copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
1012 if (copy_from_user(out, buf, sizeof(*out)))
1017 struct semid_ds tbuf_old;
1019 if(copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
1022 out->sem_perm.uid = tbuf_old.sem_perm.uid;
1023 out->sem_perm.gid = tbuf_old.sem_perm.gid;
1024 out->sem_perm.mode = tbuf_old.sem_perm.mode;
1034 * This function handles some semctl commands which require the rw_mutex
1035 * to be held in write mode.
1036 * NOTE: no locks must be held, the rw_mutex is taken inside this function.
1038 static int semctl_down(struct ipc_namespace *ns, int semid,
1039 int cmd, int version, union semun arg)
1041 struct sem_array *sma;
1043 struct semid64_ds semid64;
1044 struct kern_ipc_perm *ipcp;
1046 if(cmd == IPC_SET) {
1047 if (copy_semid_from_user(&semid64, arg.buf, version))
1051 ipcp = ipcctl_pre_down(ns, &sem_ids(ns), semid, cmd,
1052 &semid64.sem_perm, 0);
1054 return PTR_ERR(ipcp);
1056 sma = container_of(ipcp, struct sem_array, sem_perm);
1058 err = security_sem_semctl(sma, cmd);
1067 ipc_update_perm(&semid64.sem_perm, ipcp);
1068 sma->sem_ctime = get_seconds();
1077 up_write(&sem_ids(ns).rw_mutex);
1081 SYSCALL_DEFINE(semctl)(int semid, int semnum, int cmd, union semun arg)
1085 struct ipc_namespace *ns;
1090 version = ipc_parse_version(&cmd);
1091 ns = current->nsproxy->ipc_ns;
1098 err = semctl_nolock(ns, semid, cmd, version, arg);
1107 err = semctl_main(ns,semid,semnum,cmd,version,arg);
1111 err = semctl_down(ns, semid, cmd, version, arg);
1117 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
1118 asmlinkage long SyS_semctl(int semid, int semnum, int cmd, union semun arg)
1120 return SYSC_semctl((int) semid, (int) semnum, (int) cmd, arg);
1122 SYSCALL_ALIAS(sys_semctl, SyS_semctl);
1125 /* If the task doesn't already have a undo_list, then allocate one
1126 * here. We guarantee there is only one thread using this undo list,
1127 * and current is THE ONE
1129 * If this allocation and assignment succeeds, but later
1130 * portions of this code fail, there is no need to free the sem_undo_list.
1131 * Just let it stay associated with the task, and it'll be freed later
1134 * This can block, so callers must hold no locks.
1136 static inline int get_undo_list(struct sem_undo_list **undo_listp)
1138 struct sem_undo_list *undo_list;
1140 undo_list = current->sysvsem.undo_list;
1142 undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
1143 if (undo_list == NULL)
1145 spin_lock_init(&undo_list->lock);
1146 atomic_set(&undo_list->refcnt, 1);
1147 INIT_LIST_HEAD(&undo_list->list_proc);
1149 current->sysvsem.undo_list = undo_list;
1151 *undo_listp = undo_list;
1155 static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid)
1157 struct sem_undo *un;
1159 list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) {
1160 if (un->semid == semid)
1166 static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1168 struct sem_undo *un;
1170 assert_spin_locked(&ulp->lock);
1172 un = __lookup_undo(ulp, semid);
1174 list_del_rcu(&un->list_proc);
1175 list_add_rcu(&un->list_proc, &ulp->list_proc);
1181 * find_alloc_undo - Lookup (and if not present create) undo array
1183 * @semid: semaphore array id
1185 * The function looks up (and if not present creates) the undo structure.
1186 * The size of the undo structure depends on the size of the semaphore
1187 * array, thus the alloc path is not that straightforward.
1188 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1189 * performs a rcu_read_lock().
1191 static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
1193 struct sem_array *sma;
1194 struct sem_undo_list *ulp;
1195 struct sem_undo *un, *new;
1199 error = get_undo_list(&ulp);
1201 return ERR_PTR(error);
1204 spin_lock(&ulp->lock);
1205 un = lookup_undo(ulp, semid);
1206 spin_unlock(&ulp->lock);
1207 if (likely(un!=NULL))
1211 /* no undo structure around - allocate one. */
1212 /* step 1: figure out the size of the semaphore array */
1213 sma = sem_lock_check(ns, semid);
1215 return ERR_CAST(sma);
1217 nsems = sma->sem_nsems;
1218 sem_getref_and_unlock(sma);
1220 /* step 2: allocate new undo structure */
1221 new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1224 return ERR_PTR(-ENOMEM);
1227 /* step 3: Acquire the lock on semaphore array */
1228 sem_lock_and_putref(sma);
1229 if (sma->sem_perm.deleted) {
1232 un = ERR_PTR(-EIDRM);
1235 spin_lock(&ulp->lock);
1238 * step 4: check for races: did someone else allocate the undo struct?
1240 un = lookup_undo(ulp, semid);
1245 /* step 5: initialize & link new undo structure */
1246 new->semadj = (short *) &new[1];
1249 assert_spin_locked(&ulp->lock);
1250 list_add_rcu(&new->list_proc, &ulp->list_proc);
1251 assert_spin_locked(&sma->sem_perm.lock);
1252 list_add(&new->list_id, &sma->list_id);
1256 spin_unlock(&ulp->lock);
1265 * get_queue_result - Retrieve the result code from sem_queue
1266 * @q: Pointer to queue structure
1268 * Retrieve the return code from the pending queue. If IN_WAKEUP is found in
1269 * q->status, then we must loop until the value is replaced with the final
1270 * value: This may happen if a task is woken up by an unrelated event (e.g.
1271 * signal) and in parallel the task is woken up by another task because it got
1272 * the requested semaphores.
1274 * The function can be called with or without holding the semaphore spinlock.
1276 static int get_queue_result(struct sem_queue *q)
1281 while (unlikely(error == IN_WAKEUP)) {
1290 SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
1291 unsigned, nsops, const struct timespec __user *, timeout)
1293 int error = -EINVAL;
1294 struct sem_array *sma;
1295 struct sembuf fast_sops[SEMOPM_FAST];
1296 struct sembuf* sops = fast_sops, *sop;
1297 struct sem_undo *un;
1298 int undos = 0, alter = 0, max;
1299 struct sem_queue queue;
1300 unsigned long jiffies_left = 0;
1301 struct ipc_namespace *ns;
1302 struct list_head tasks;
1304 ns = current->nsproxy->ipc_ns;
1306 if (nsops < 1 || semid < 0)
1308 if (nsops > ns->sc_semopm)
1310 if(nsops > SEMOPM_FAST) {
1311 sops = kmalloc(sizeof(*sops)*nsops,GFP_KERNEL);
1315 if (copy_from_user (sops, tsops, nsops * sizeof(*tsops))) {
1320 struct timespec _timeout;
1321 if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1325 if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1326 _timeout.tv_nsec >= 1000000000L) {
1330 jiffies_left = timespec_to_jiffies(&_timeout);
1333 for (sop = sops; sop < sops + nsops; sop++) {
1334 if (sop->sem_num >= max)
1336 if (sop->sem_flg & SEM_UNDO)
1338 if (sop->sem_op != 0)
1343 un = find_alloc_undo(ns, semid);
1345 error = PTR_ERR(un);
1351 INIT_LIST_HEAD(&tasks);
1353 sma = sem_lock_check(ns, semid);
1357 error = PTR_ERR(sma);
1362 * semid identifiers are not unique - find_alloc_undo may have
1363 * allocated an undo structure, it was invalidated by an RMID
1364 * and now a new array with received the same id. Check and fail.
1365 * This case can be detected checking un->semid. The existence of
1366 * "un" itself is guaranteed by rcu.
1370 if (un->semid == -1) {
1372 goto out_unlock_free;
1375 * rcu lock can be released, "un" cannot disappear:
1376 * - sem_lock is acquired, thus IPC_RMID is
1378 * - exit_sem is impossible, it always operates on
1379 * current (or a dead task).
1387 if (max >= sma->sem_nsems)
1388 goto out_unlock_free;
1391 if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO))
1392 goto out_unlock_free;
1394 error = security_sem_semop(sma, sops, nsops, alter);
1396 goto out_unlock_free;
1398 error = try_atomic_semop (sma, sops, nsops, un, task_tgid_vnr(current));
1400 if (alter && error == 0)
1401 do_smart_update(sma, sops, nsops, 1, &tasks);
1403 goto out_unlock_free;
1406 /* We need to sleep on this operation, so we put the current
1407 * task into the pending queue and go to sleep.
1411 queue.nsops = nsops;
1413 queue.pid = task_tgid_vnr(current);
1414 queue.alter = alter;
1416 list_add_tail(&queue.list, &sma->sem_pending);
1418 list_add(&queue.list, &sma->sem_pending);
1422 curr = &sma->sem_base[sops->sem_num];
1425 list_add_tail(&queue.simple_list, &curr->sem_pending);
1427 list_add(&queue.simple_list, &curr->sem_pending);
1429 INIT_LIST_HEAD(&queue.simple_list);
1430 sma->complex_count++;
1433 queue.status = -EINTR;
1434 queue.sleeper = current;
1435 current->state = TASK_INTERRUPTIBLE;
1439 jiffies_left = schedule_timeout(jiffies_left);
1443 error = get_queue_result(&queue);
1445 if (error != -EINTR) {
1446 /* fast path: update_queue already obtained all requested
1448 * Perform a smp_mb(): User space could assume that semop()
1449 * is a memory barrier: Without the mb(), the cpu could
1450 * speculatively read in user space stale data that was
1451 * overwritten by the previous owner of the semaphore.
1458 sma = sem_lock(ns, semid);
1461 * Wait until it's guaranteed that no wakeup_sem_queue_do() is ongoing.
1463 error = get_queue_result(&queue);
1466 * Array removed? If yes, leave without sem_unlock().
1475 * If queue.status != -EINTR we are woken up by another process.
1476 * Leave without unlink_queue(), but with sem_unlock().
1479 if (error != -EINTR) {
1480 goto out_unlock_free;
1484 * If an interrupt occurred we have to clean up the queue
1486 if (timeout && jiffies_left == 0)
1488 unlink_queue(sma, &queue);
1493 wake_up_sem_queue_do(&tasks);
1495 if(sops != fast_sops)
1500 SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
1503 return sys_semtimedop(semid, tsops, nsops, NULL);
1506 /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
1507 * parent and child tasks.
1510 int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
1512 struct sem_undo_list *undo_list;
1515 if (clone_flags & CLONE_SYSVSEM) {
1516 error = get_undo_list(&undo_list);
1519 atomic_inc(&undo_list->refcnt);
1520 tsk->sysvsem.undo_list = undo_list;
1522 tsk->sysvsem.undo_list = NULL;
1528 * add semadj values to semaphores, free undo structures.
1529 * undo structures are not freed when semaphore arrays are destroyed
1530 * so some of them may be out of date.
1531 * IMPLEMENTATION NOTE: There is some confusion over whether the
1532 * set of adjustments that needs to be done should be done in an atomic
1533 * manner or not. That is, if we are attempting to decrement the semval
1534 * should we queue up and wait until we can do so legally?
1535 * The original implementation attempted to do this (queue and wait).
1536 * The current implementation does not do so. The POSIX standard
1537 * and SVID should be consulted to determine what behavior is mandated.
1539 void exit_sem(struct task_struct *tsk)
1541 struct sem_undo_list *ulp;
1543 ulp = tsk->sysvsem.undo_list;
1546 tsk->sysvsem.undo_list = NULL;
1548 if (!atomic_dec_and_test(&ulp->refcnt))
1552 struct sem_array *sma;
1553 struct sem_undo *un;
1554 struct list_head tasks;
1559 un = list_entry_rcu(ulp->list_proc.next,
1560 struct sem_undo, list_proc);
1561 if (&un->list_proc == &ulp->list_proc)
1570 sma = sem_lock_check(tsk->nsproxy->ipc_ns, un->semid);
1572 /* exit_sem raced with IPC_RMID, nothing to do */
1576 un = __lookup_undo(ulp, semid);
1578 /* exit_sem raced with IPC_RMID+semget() that created
1579 * exactly the same semid. Nothing to do.
1585 /* remove un from the linked lists */
1586 assert_spin_locked(&sma->sem_perm.lock);
1587 list_del(&un->list_id);
1589 spin_lock(&ulp->lock);
1590 list_del_rcu(&un->list_proc);
1591 spin_unlock(&ulp->lock);
1593 /* perform adjustments registered in un */
1594 for (i = 0; i < sma->sem_nsems; i++) {
1595 struct sem * semaphore = &sma->sem_base[i];
1596 if (un->semadj[i]) {
1597 semaphore->semval += un->semadj[i];
1599 * Range checks of the new semaphore value,
1600 * not defined by sus:
1601 * - Some unices ignore the undo entirely
1602 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
1603 * - some cap the value (e.g. FreeBSD caps
1604 * at 0, but doesn't enforce SEMVMX)
1606 * Linux caps the semaphore value, both at 0
1609 * Manfred <manfred@colorfullife.com>
1611 if (semaphore->semval < 0)
1612 semaphore->semval = 0;
1613 if (semaphore->semval > SEMVMX)
1614 semaphore->semval = SEMVMX;
1615 semaphore->sempid = task_tgid_vnr(current);
1618 /* maybe some queued-up processes were waiting for this */
1619 INIT_LIST_HEAD(&tasks);
1620 do_smart_update(sma, NULL, 0, 1, &tasks);
1622 wake_up_sem_queue_do(&tasks);
1624 call_rcu(&un->rcu, free_un);
1629 #ifdef CONFIG_PROC_FS
1630 static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
1632 struct sem_array *sma = it;
1634 return seq_printf(s,
1635 "%10d %10d %4o %10u %5u %5u %5u %5u %10lu %10lu\n",