3 * Copyright (C) 1992 Krishna Balasubramanian
4 * Copyright (C) 1995 Eric Schenk, Bruno Haible
6 * IMPLEMENTATION NOTES ON CODE REWRITE (Eric Schenk, January 1995):
7 * This code underwent a massive rewrite in order to solve some problems
8 * with the original code. In particular the original code failed to
9 * wake up processes that were waiting for semval to go to 0 if the
10 * value went to 0 and was then incremented rapidly enough. In solving
11 * this problem I have also modified the implementation so that it
12 * processes pending operations in a FIFO manner, thus give a guarantee
13 * that processes waiting for a lock on the semaphore won't starve
14 * unless another locking process fails to unlock.
15 * In addition the following two changes in behavior have been introduced:
16 * - The original implementation of semop returned the value
17 * last semaphore element examined on success. This does not
18 * match the manual page specifications, and effectively
19 * allows the user to read the semaphore even if they do not
20 * have read permissions. The implementation now returns 0
21 * on success as stated in the manual page.
22 * - There is some confusion over whether the set of undo adjustments
23 * to be performed at exit should be done in an atomic manner.
24 * That is, if we are attempting to decrement the semval should we queue
25 * up and wait until we can do so legally?
26 * The original implementation attempted to do this.
27 * The current implementation does not do so. This is because I don't
28 * think it is the right thing (TM) to do, and because I couldn't
29 * see a clean way to get the old behavior with the new design.
30 * The POSIX standard and SVID should be consulted to determine
31 * what behavior is mandated.
33 * Further notes on refinement (Christoph Rohland, December 1998):
34 * - The POSIX standard says, that the undo adjustments simply should
35 * redo. So the current implementation is o.K.
36 * - The previous code had two flaws:
37 * 1) It actively gave the semaphore to the next waiting process
38 * sleeping on the semaphore. Since this process did not have the
39 * cpu this led to many unnecessary context switches and bad
40 * performance. Now we only check which process should be able to
41 * get the semaphore and if this process wants to reduce some
42 * semaphore value we simply wake it up without doing the
43 * operation. So it has to try to get it later. Thus e.g. the
44 * running process may reacquire the semaphore during the current
45 * time slice. If it only waits for zero or increases the semaphore,
46 * we do the operation in advance and wake it up.
47 * 2) It did not wake up all zero waiting processes. We try to do
48 * better but only get the semops right which only wait for zero or
49 * increase. If there are decrement operations in the operations
50 * array we do the same as before.
52 * With the incarnation of O(1) scheduler, it becomes unnecessary to perform
53 * check/retry algorithm for waking up blocked processes as the new scheduler
54 * is better at handling thread switch than the old one.
56 * /proc/sysvipc/sem support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
58 * SMP-threaded, sysctl's added
59 * (c) 1999 Manfred Spraul <manfred@colorfullife.com>
60 * Enforced range limit on SEM_UNDO
61 * (c) 2001 Red Hat Inc
63 * (c) 2003 Manfred Spraul <manfred@colorfullife.com>
65 * support for audit of ipc object properties and permission changes
66 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
70 * Pavel Emelianov <xemul@openvz.org>
73 #include <linux/slab.h>
74 #include <linux/spinlock.h>
75 #include <linux/init.h>
76 #include <linux/proc_fs.h>
77 #include <linux/time.h>
78 #include <linux/security.h>
79 #include <linux/syscalls.h>
80 #include <linux/audit.h>
81 #include <linux/capability.h>
82 #include <linux/seq_file.h>
83 #include <linux/rwsem.h>
84 #include <linux/nsproxy.h>
85 #include <linux/ipc_namespace.h>
87 #include <asm/uaccess.h>
90 #define sem_ids(ns) ((ns)->ids[IPC_SEM_IDS])
92 #define sem_unlock(sma) ipc_unlock(&(sma)->sem_perm)
93 #define sem_checkid(sma, semid) ipc_checkid(&sma->sem_perm, semid)
95 static int newary(struct ipc_namespace *, struct ipc_params *);
96 static void freeary(struct ipc_namespace *, struct kern_ipc_perm *);
98 static int sysvipc_sem_proc_show(struct seq_file *s, void *it);
101 #define SEMMSL_FAST 256 /* 512 bytes on stack */
102 #define SEMOPM_FAST 64 /* ~ 372 bytes on stack */
105 * linked list protection:
107 * sem_array.sem_pending{,last},
108 * sem_array.sem_undo: sem_lock() for read/write
109 * sem_undo.proc_next: only "current" is allowed to read/write that field.
113 #define sc_semmsl sem_ctls[0]
114 #define sc_semmns sem_ctls[1]
115 #define sc_semopm sem_ctls[2]
116 #define sc_semmni sem_ctls[3]
118 void sem_init_ns(struct ipc_namespace *ns)
120 ns->sc_semmsl = SEMMSL;
121 ns->sc_semmns = SEMMNS;
122 ns->sc_semopm = SEMOPM;
123 ns->sc_semmni = SEMMNI;
125 ipc_init_ids(&ns->ids[IPC_SEM_IDS]);
129 void sem_exit_ns(struct ipc_namespace *ns)
131 free_ipcs(ns, &sem_ids(ns), freeary);
135 void __init sem_init (void)
137 sem_init_ns(&init_ipc_ns);
138 ipc_init_proc_interface("sysvipc/sem",
139 " key semid perms nsems uid gid cuid cgid otime ctime\n",
140 IPC_SEM_IDS, sysvipc_sem_proc_show);
144 * sem_lock_(check_) routines are called in the paths where the rw_mutex
147 static inline struct sem_array *sem_lock(struct ipc_namespace *ns, int id)
149 struct kern_ipc_perm *ipcp = ipc_lock(&sem_ids(ns), id);
152 return (struct sem_array *)ipcp;
154 return container_of(ipcp, struct sem_array, sem_perm);
157 static inline struct sem_array *sem_lock_check(struct ipc_namespace *ns,
160 struct kern_ipc_perm *ipcp = ipc_lock_check(&sem_ids(ns), id);
163 return (struct sem_array *)ipcp;
165 return container_of(ipcp, struct sem_array, sem_perm);
168 static inline void sem_lock_and_putref(struct sem_array *sma)
170 ipc_lock_by_ptr(&sma->sem_perm);
174 static inline void sem_getref_and_unlock(struct sem_array *sma)
177 ipc_unlock(&(sma)->sem_perm);
180 static inline void sem_putref(struct sem_array *sma)
182 ipc_lock_by_ptr(&sma->sem_perm);
184 ipc_unlock(&(sma)->sem_perm);
187 static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
189 ipc_rmid(&sem_ids(ns), &s->sem_perm);
193 * Lockless wakeup algorithm:
194 * Without the check/retry algorithm a lockless wakeup is possible:
195 * - queue.status is initialized to -EINTR before blocking.
196 * - wakeup is performed by
197 * * unlinking the queue entry from sma->sem_pending
198 * * setting queue.status to IN_WAKEUP
199 * This is the notification for the blocked thread that a
200 * result value is imminent.
201 * * call wake_up_process
202 * * set queue.status to the final value.
203 * - the previously blocked thread checks queue.status:
204 * * if it's IN_WAKEUP, then it must wait until the value changes
205 * * if it's not -EINTR, then the operation was completed by
206 * update_queue. semtimedop can return queue.status without
207 * performing any operation on the sem array.
208 * * otherwise it must acquire the spinlock and check what's up.
210 * The two-stage algorithm is necessary to protect against the following
212 * - if queue.status is set after wake_up_process, then the woken up idle
213 * thread could race forward and try (and fail) to acquire sma->lock
214 * before update_queue had a chance to set queue.status
215 * - if queue.status is written before wake_up_process and if the
216 * blocked process is woken up by a signal between writing
217 * queue.status and the wake_up_process, then the woken up
218 * process could return from semtimedop and die by calling
219 * sys_exit before wake_up_process is called. Then wake_up_process
220 * will oops, because the task structure is already invalid.
221 * (yes, this happened on s390 with sysv msg).
227 * newary - Create a new semaphore set
229 * @params: ptr to the structure that contains key, semflg and nsems
231 * Called with sem_ids.rw_mutex held (as a writer)
234 static int newary(struct ipc_namespace *ns, struct ipc_params *params)
238 struct sem_array *sma;
240 key_t key = params->key;
241 int nsems = params->u.nsems;
242 int semflg = params->flg;
246 if (ns->used_sems + nsems > ns->sc_semmns)
249 size = sizeof (*sma) + nsems * sizeof (struct sem);
250 sma = ipc_rcu_alloc(size);
254 memset (sma, 0, size);
256 sma->sem_perm.mode = (semflg & S_IRWXUGO);
257 sma->sem_perm.key = key;
259 sma->sem_perm.security = NULL;
260 retval = security_sem_alloc(sma);
266 id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
268 security_sem_free(sma);
272 ns->used_sems += nsems;
274 sma->sem_base = (struct sem *) &sma[1];
275 INIT_LIST_HEAD(&sma->sem_pending);
276 INIT_LIST_HEAD(&sma->list_id);
277 sma->sem_nsems = nsems;
278 sma->sem_ctime = get_seconds();
281 return sma->sem_perm.id;
286 * Called with sem_ids.rw_mutex and ipcp locked.
288 static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg)
290 struct sem_array *sma;
292 sma = container_of(ipcp, struct sem_array, sem_perm);
293 return security_sem_associate(sma, semflg);
297 * Called with sem_ids.rw_mutex and ipcp locked.
299 static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
300 struct ipc_params *params)
302 struct sem_array *sma;
304 sma = container_of(ipcp, struct sem_array, sem_perm);
305 if (params->u.nsems > sma->sem_nsems)
311 asmlinkage long sys_semget(key_t key, int nsems, int semflg)
313 struct ipc_namespace *ns;
314 struct ipc_ops sem_ops;
315 struct ipc_params sem_params;
317 ns = current->nsproxy->ipc_ns;
319 if (nsems < 0 || nsems > ns->sc_semmsl)
322 sem_ops.getnew = newary;
323 sem_ops.associate = sem_security;
324 sem_ops.more_checks = sem_more_checks;
326 sem_params.key = key;
327 sem_params.flg = semflg;
328 sem_params.u.nsems = nsems;
330 return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
334 * Determine whether a sequence of semaphore operations would succeed
335 * all at once. Return 0 if yes, 1 if need to sleep, else return error code.
338 static int try_atomic_semop (struct sem_array * sma, struct sembuf * sops,
339 int nsops, struct sem_undo *un, int pid)
345 for (sop = sops; sop < sops + nsops; sop++) {
346 curr = sma->sem_base + sop->sem_num;
347 sem_op = sop->sem_op;
348 result = curr->semval;
350 if (!sem_op && result)
358 if (sop->sem_flg & SEM_UNDO) {
359 int undo = un->semadj[sop->sem_num] - sem_op;
361 * Exceeding the undo range is an error.
363 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
366 curr->semval = result;
370 while (sop >= sops) {
371 sma->sem_base[sop->sem_num].sempid = pid;
372 if (sop->sem_flg & SEM_UNDO)
373 un->semadj[sop->sem_num] -= sop->sem_op;
377 sma->sem_otime = get_seconds();
385 if (sop->sem_flg & IPC_NOWAIT)
392 while (sop >= sops) {
393 sma->sem_base[sop->sem_num].semval -= sop->sem_op;
400 /* Go through the pending queue for the indicated semaphore
401 * looking for tasks that can be completed.
403 static void update_queue (struct sem_array * sma)
406 struct sem_queue * q;
408 q = list_entry(sma->sem_pending.next, struct sem_queue, list);
409 while (&q->list != &sma->sem_pending) {
410 error = try_atomic_semop(sma, q->sops, q->nsops,
413 /* Does q->sleeper still need to sleep? */
418 * Continue scanning. The next operation
419 * that must be checked depends on the type of the
420 * completed operation:
421 * - if the operation modified the array, then
422 * restart from the head of the queue and
423 * check for threads that might be waiting
424 * for semaphore values to become 0.
425 * - if the operation didn't modify the array,
426 * then just continue.
427 * The order of list_del() and reading ->next
428 * is crucial: In the former case, the list_del()
429 * must be done first [because we might be the
430 * first entry in ->sem_pending], in the latter
431 * case the list_del() must be done last
432 * [because the list is invalid after the list_del()]
436 n = list_entry(sma->sem_pending.next,
437 struct sem_queue, list);
439 n = list_entry(q->list.next, struct sem_queue,
444 /* wake up the waiting thread */
445 q->status = IN_WAKEUP;
447 wake_up_process(q->sleeper);
448 /* hands-off: q will disappear immediately after
455 q = list_entry(q->list.next, struct sem_queue, list);
460 /* The following counts are associated to each semaphore:
461 * semncnt number of tasks waiting on semval being nonzero
462 * semzcnt number of tasks waiting on semval being zero
463 * This model assumes that a task waits on exactly one semaphore.
464 * Since semaphore operations are to be performed atomically, tasks actually
465 * wait on a whole sequence of semaphores simultaneously.
466 * The counts we return here are a rough approximation, but still
467 * warrant that semncnt+semzcnt>0 if the task is on the pending queue.
469 static int count_semncnt (struct sem_array * sma, ushort semnum)
472 struct sem_queue * q;
475 list_for_each_entry(q, &sma->sem_pending, list) {
476 struct sembuf * sops = q->sops;
477 int nsops = q->nsops;
479 for (i = 0; i < nsops; i++)
480 if (sops[i].sem_num == semnum
481 && (sops[i].sem_op < 0)
482 && !(sops[i].sem_flg & IPC_NOWAIT))
488 static int count_semzcnt (struct sem_array * sma, ushort semnum)
491 struct sem_queue * q;
494 list_for_each_entry(q, &sma->sem_pending, list) {
495 struct sembuf * sops = q->sops;
496 int nsops = q->nsops;
498 for (i = 0; i < nsops; i++)
499 if (sops[i].sem_num == semnum
500 && (sops[i].sem_op == 0)
501 && !(sops[i].sem_flg & IPC_NOWAIT))
507 static void free_un(struct rcu_head *head)
509 struct sem_undo *un = container_of(head, struct sem_undo, rcu);
513 /* Free a semaphore set. freeary() is called with sem_ids.rw_mutex locked
514 * as a writer and the spinlock for this semaphore set hold. sem_ids.rw_mutex
515 * remains locked on exit.
517 static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
519 struct sem_undo *un, *tu;
520 struct sem_queue *q, *tq;
521 struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
523 /* Free the existing undo structures for this semaphore set. */
524 assert_spin_locked(&sma->sem_perm.lock);
525 list_for_each_entry_safe(un, tu, &sma->list_id, list_id) {
526 list_del(&un->list_id);
527 spin_lock(&un->ulp->lock);
529 list_del_rcu(&un->list_proc);
530 spin_unlock(&un->ulp->lock);
531 call_rcu(&un->rcu, free_un);
534 /* Wake up all pending processes and let them fail with EIDRM. */
535 list_for_each_entry_safe(q, tq, &sma->sem_pending, list) {
538 q->status = IN_WAKEUP;
539 wake_up_process(q->sleeper); /* doesn't sleep */
541 q->status = -EIDRM; /* hands-off q */
544 /* Remove the semaphore set from the IDR */
548 ns->used_sems -= sma->sem_nsems;
549 security_sem_free(sma);
553 static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
557 return copy_to_user(buf, in, sizeof(*in));
562 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
564 out.sem_otime = in->sem_otime;
565 out.sem_ctime = in->sem_ctime;
566 out.sem_nsems = in->sem_nsems;
568 return copy_to_user(buf, &out, sizeof(out));
575 static int semctl_nolock(struct ipc_namespace *ns, int semid,
576 int cmd, int version, union semun arg)
579 struct sem_array *sma;
585 struct seminfo seminfo;
588 err = security_sem_semctl(NULL, cmd);
592 memset(&seminfo,0,sizeof(seminfo));
593 seminfo.semmni = ns->sc_semmni;
594 seminfo.semmns = ns->sc_semmns;
595 seminfo.semmsl = ns->sc_semmsl;
596 seminfo.semopm = ns->sc_semopm;
597 seminfo.semvmx = SEMVMX;
598 seminfo.semmnu = SEMMNU;
599 seminfo.semmap = SEMMAP;
600 seminfo.semume = SEMUME;
601 down_read(&sem_ids(ns).rw_mutex);
602 if (cmd == SEM_INFO) {
603 seminfo.semusz = sem_ids(ns).in_use;
604 seminfo.semaem = ns->used_sems;
606 seminfo.semusz = SEMUSZ;
607 seminfo.semaem = SEMAEM;
609 max_id = ipc_get_maxid(&sem_ids(ns));
610 up_read(&sem_ids(ns).rw_mutex);
611 if (copy_to_user (arg.__buf, &seminfo, sizeof(struct seminfo)))
613 return (max_id < 0) ? 0: max_id;
618 struct semid64_ds tbuf;
621 if (cmd == SEM_STAT) {
622 sma = sem_lock(ns, semid);
625 id = sma->sem_perm.id;
627 sma = sem_lock_check(ns, semid);
634 if (ipcperms (&sma->sem_perm, S_IRUGO))
637 err = security_sem_semctl(sma, cmd);
641 memset(&tbuf, 0, sizeof(tbuf));
643 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
644 tbuf.sem_otime = sma->sem_otime;
645 tbuf.sem_ctime = sma->sem_ctime;
646 tbuf.sem_nsems = sma->sem_nsems;
648 if (copy_semid_to_user (arg.buf, &tbuf, version))
661 static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
662 int cmd, int version, union semun arg)
664 struct sem_array *sma;
667 ushort fast_sem_io[SEMMSL_FAST];
668 ushort* sem_io = fast_sem_io;
671 sma = sem_lock_check(ns, semid);
675 nsems = sma->sem_nsems;
678 if (ipcperms (&sma->sem_perm, (cmd==SETVAL||cmd==SETALL)?S_IWUGO:S_IRUGO))
681 err = security_sem_semctl(sma, cmd);
689 ushort __user *array = arg.array;
692 if(nsems > SEMMSL_FAST) {
693 sem_getref_and_unlock(sma);
695 sem_io = ipc_alloc(sizeof(ushort)*nsems);
701 sem_lock_and_putref(sma);
702 if (sma->sem_perm.deleted) {
709 for (i = 0; i < sma->sem_nsems; i++)
710 sem_io[i] = sma->sem_base[i].semval;
713 if(copy_to_user(array, sem_io, nsems*sizeof(ushort)))
722 sem_getref_and_unlock(sma);
724 if(nsems > SEMMSL_FAST) {
725 sem_io = ipc_alloc(sizeof(ushort)*nsems);
732 if (copy_from_user (sem_io, arg.array, nsems*sizeof(ushort))) {
738 for (i = 0; i < nsems; i++) {
739 if (sem_io[i] > SEMVMX) {
745 sem_lock_and_putref(sma);
746 if (sma->sem_perm.deleted) {
752 for (i = 0; i < nsems; i++)
753 sma->sem_base[i].semval = sem_io[i];
755 assert_spin_locked(&sma->sem_perm.lock);
756 list_for_each_entry(un, &sma->list_id, list_id) {
757 for (i = 0; i < nsems; i++)
760 sma->sem_ctime = get_seconds();
761 /* maybe some queued-up processes were waiting for this */
766 /* GETVAL, GETPID, GETNCTN, GETZCNT, SETVAL: fall-through */
769 if(semnum < 0 || semnum >= nsems)
772 curr = &sma->sem_base[semnum];
782 err = count_semncnt(sma,semnum);
785 err = count_semzcnt(sma,semnum);
793 if (val > SEMVMX || val < 0)
796 assert_spin_locked(&sma->sem_perm.lock);
797 list_for_each_entry(un, &sma->list_id, list_id)
798 un->semadj[semnum] = 0;
801 curr->sempid = task_tgid_vnr(current);
802 sma->sem_ctime = get_seconds();
803 /* maybe some queued-up processes were waiting for this */
812 if(sem_io != fast_sem_io)
813 ipc_free(sem_io, sizeof(ushort)*nsems);
817 static inline unsigned long
818 copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
822 if (copy_from_user(out, buf, sizeof(*out)))
827 struct semid_ds tbuf_old;
829 if(copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
832 out->sem_perm.uid = tbuf_old.sem_perm.uid;
833 out->sem_perm.gid = tbuf_old.sem_perm.gid;
834 out->sem_perm.mode = tbuf_old.sem_perm.mode;
844 * This function handles some semctl commands which require the rw_mutex
845 * to be held in write mode.
846 * NOTE: no locks must be held, the rw_mutex is taken inside this function.
848 static int semctl_down(struct ipc_namespace *ns, int semid,
849 int cmd, int version, union semun arg)
851 struct sem_array *sma;
853 struct semid64_ds semid64;
854 struct kern_ipc_perm *ipcp;
857 if (copy_semid_from_user(&semid64, arg.buf, version))
861 ipcp = ipcctl_pre_down(&sem_ids(ns), semid, cmd, &semid64.sem_perm, 0);
863 return PTR_ERR(ipcp);
865 sma = container_of(ipcp, struct sem_array, sem_perm);
867 err = security_sem_semctl(sma, cmd);
876 ipc_update_perm(&semid64.sem_perm, ipcp);
877 sma->sem_ctime = get_seconds();
886 up_write(&sem_ids(ns).rw_mutex);
890 asmlinkage long sys_semctl (int semid, int semnum, int cmd, union semun arg)
894 struct ipc_namespace *ns;
899 version = ipc_parse_version(&cmd);
900 ns = current->nsproxy->ipc_ns;
907 err = semctl_nolock(ns, semid, cmd, version, arg);
916 err = semctl_main(ns,semid,semnum,cmd,version,arg);
920 err = semctl_down(ns, semid, cmd, version, arg);
927 /* If the task doesn't already have a undo_list, then allocate one
928 * here. We guarantee there is only one thread using this undo list,
929 * and current is THE ONE
931 * If this allocation and assignment succeeds, but later
932 * portions of this code fail, there is no need to free the sem_undo_list.
933 * Just let it stay associated with the task, and it'll be freed later
936 * This can block, so callers must hold no locks.
938 static inline int get_undo_list(struct sem_undo_list **undo_listp)
940 struct sem_undo_list *undo_list;
942 undo_list = current->sysvsem.undo_list;
944 undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
945 if (undo_list == NULL)
947 spin_lock_init(&undo_list->lock);
948 atomic_set(&undo_list->refcnt, 1);
949 INIT_LIST_HEAD(&undo_list->list_proc);
951 current->sysvsem.undo_list = undo_list;
953 *undo_listp = undo_list;
957 static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
959 struct sem_undo *walk;
961 list_for_each_entry_rcu(walk, &ulp->list_proc, list_proc) {
962 if (walk->semid == semid)
969 * find_alloc_undo - Lookup (and if not present create) undo array
971 * @semid: semaphore array id
973 * The function looks up (and if not present creates) the undo structure.
974 * The size of the undo structure depends on the size of the semaphore
975 * array, thus the alloc path is not that straightforward.
976 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
977 * performs a rcu_read_lock().
979 static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
981 struct sem_array *sma;
982 struct sem_undo_list *ulp;
983 struct sem_undo *un, *new;
987 error = get_undo_list(&ulp);
989 return ERR_PTR(error);
992 spin_lock(&ulp->lock);
993 un = lookup_undo(ulp, semid);
994 spin_unlock(&ulp->lock);
995 if (likely(un!=NULL))
999 /* no undo structure around - allocate one. */
1000 /* step 1: figure out the size of the semaphore array */
1001 sma = sem_lock_check(ns, semid);
1003 return ERR_PTR(PTR_ERR(sma));
1005 nsems = sma->sem_nsems;
1006 sem_getref_and_unlock(sma);
1008 /* step 2: allocate new undo structure */
1009 new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1012 return ERR_PTR(-ENOMEM);
1015 /* step 3: Acquire the lock on semaphore array */
1016 sem_lock_and_putref(sma);
1017 if (sma->sem_perm.deleted) {
1020 un = ERR_PTR(-EIDRM);
1023 spin_lock(&ulp->lock);
1026 * step 4: check for races: did someone else allocate the undo struct?
1028 un = lookup_undo(ulp, semid);
1033 /* step 5: initialize & link new undo structure */
1034 new->semadj = (short *) &new[1];
1037 assert_spin_locked(&ulp->lock);
1038 list_add_rcu(&new->list_proc, &ulp->list_proc);
1039 assert_spin_locked(&sma->sem_perm.lock);
1040 list_add(&new->list_id, &sma->list_id);
1044 spin_unlock(&ulp->lock);
1051 asmlinkage long sys_semtimedop(int semid, struct sembuf __user *tsops,
1052 unsigned nsops, const struct timespec __user *timeout)
1054 int error = -EINVAL;
1055 struct sem_array *sma;
1056 struct sembuf fast_sops[SEMOPM_FAST];
1057 struct sembuf* sops = fast_sops, *sop;
1058 struct sem_undo *un;
1059 int undos = 0, alter = 0, max;
1060 struct sem_queue queue;
1061 unsigned long jiffies_left = 0;
1062 struct ipc_namespace *ns;
1064 ns = current->nsproxy->ipc_ns;
1066 if (nsops < 1 || semid < 0)
1068 if (nsops > ns->sc_semopm)
1070 if(nsops > SEMOPM_FAST) {
1071 sops = kmalloc(sizeof(*sops)*nsops,GFP_KERNEL);
1075 if (copy_from_user (sops, tsops, nsops * sizeof(*tsops))) {
1080 struct timespec _timeout;
1081 if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1085 if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1086 _timeout.tv_nsec >= 1000000000L) {
1090 jiffies_left = timespec_to_jiffies(&_timeout);
1093 for (sop = sops; sop < sops + nsops; sop++) {
1094 if (sop->sem_num >= max)
1096 if (sop->sem_flg & SEM_UNDO)
1098 if (sop->sem_op != 0)
1103 un = find_alloc_undo(ns, semid);
1105 error = PTR_ERR(un);
1111 sma = sem_lock_check(ns, semid);
1115 error = PTR_ERR(sma);
1120 * semid identifiers are not unique - find_alloc_undo may have
1121 * allocated an undo structure, it was invalidated by an RMID
1122 * and now a new array with received the same id. Check and fail.
1123 * This case can be detected checking un->semid. The existance of
1124 * "un" itself is guaranteed by rcu.
1128 if (un->semid == -1) {
1130 goto out_unlock_free;
1133 * rcu lock can be released, "un" cannot disappear:
1134 * - sem_lock is acquired, thus IPC_RMID is
1136 * - exit_sem is impossible, it always operates on
1137 * current (or a dead task).
1145 if (max >= sma->sem_nsems)
1146 goto out_unlock_free;
1149 if (ipcperms(&sma->sem_perm, alter ? S_IWUGO : S_IRUGO))
1150 goto out_unlock_free;
1152 error = security_sem_semop(sma, sops, nsops, alter);
1154 goto out_unlock_free;
1156 error = try_atomic_semop (sma, sops, nsops, un, task_tgid_vnr(current));
1158 if (alter && error == 0)
1160 goto out_unlock_free;
1163 /* We need to sleep on this operation, so we put the current
1164 * task into the pending queue and go to sleep.
1168 queue.nsops = nsops;
1170 queue.pid = task_tgid_vnr(current);
1171 queue.alter = alter;
1173 list_add_tail(&queue.list, &sma->sem_pending);
1175 list_add(&queue.list, &sma->sem_pending);
1177 queue.status = -EINTR;
1178 queue.sleeper = current;
1179 current->state = TASK_INTERRUPTIBLE;
1183 jiffies_left = schedule_timeout(jiffies_left);
1187 error = queue.status;
1188 while(unlikely(error == IN_WAKEUP)) {
1190 error = queue.status;
1193 if (error != -EINTR) {
1194 /* fast path: update_queue already obtained all requested
1199 sma = sem_lock(ns, semid);
1206 * If queue.status != -EINTR we are woken up by another process
1208 error = queue.status;
1209 if (error != -EINTR) {
1210 goto out_unlock_free;
1214 * If an interrupt occurred we have to clean up the queue
1216 if (timeout && jiffies_left == 0)
1218 list_del(&queue.list);
1223 if(sops != fast_sops)
1228 asmlinkage long sys_semop (int semid, struct sembuf __user *tsops, unsigned nsops)
1230 return sys_semtimedop(semid, tsops, nsops, NULL);
1233 /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
1234 * parent and child tasks.
1237 int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
1239 struct sem_undo_list *undo_list;
1242 if (clone_flags & CLONE_SYSVSEM) {
1243 error = get_undo_list(&undo_list);
1246 atomic_inc(&undo_list->refcnt);
1247 tsk->sysvsem.undo_list = undo_list;
1249 tsk->sysvsem.undo_list = NULL;
1255 * add semadj values to semaphores, free undo structures.
1256 * undo structures are not freed when semaphore arrays are destroyed
1257 * so some of them may be out of date.
1258 * IMPLEMENTATION NOTE: There is some confusion over whether the
1259 * set of adjustments that needs to be done should be done in an atomic
1260 * manner or not. That is, if we are attempting to decrement the semval
1261 * should we queue up and wait until we can do so legally?
1262 * The original implementation attempted to do this (queue and wait).
1263 * The current implementation does not do so. The POSIX standard
1264 * and SVID should be consulted to determine what behavior is mandated.
1266 void exit_sem(struct task_struct *tsk)
1268 struct sem_undo_list *ulp;
1270 ulp = tsk->sysvsem.undo_list;
1273 tsk->sysvsem.undo_list = NULL;
1275 if (!atomic_dec_and_test(&ulp->refcnt))
1279 struct sem_array *sma;
1280 struct sem_undo *un;
1285 un = list_entry(rcu_dereference(ulp->list_proc.next),
1286 struct sem_undo, list_proc);
1287 if (&un->list_proc == &ulp->list_proc)
1296 sma = sem_lock_check(tsk->nsproxy->ipc_ns, un->semid);
1298 /* exit_sem raced with IPC_RMID, nothing to do */
1302 un = lookup_undo(ulp, semid);
1304 /* exit_sem raced with IPC_RMID+semget() that created
1305 * exactly the same semid. Nothing to do.
1311 /* remove un from the linked lists */
1312 assert_spin_locked(&sma->sem_perm.lock);
1313 list_del(&un->list_id);
1315 spin_lock(&ulp->lock);
1316 list_del_rcu(&un->list_proc);
1317 spin_unlock(&ulp->lock);
1319 /* perform adjustments registered in un */
1320 for (i = 0; i < sma->sem_nsems; i++) {
1321 struct sem * semaphore = &sma->sem_base[i];
1322 if (un->semadj[i]) {
1323 semaphore->semval += un->semadj[i];
1325 * Range checks of the new semaphore value,
1326 * not defined by sus:
1327 * - Some unices ignore the undo entirely
1328 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
1329 * - some cap the value (e.g. FreeBSD caps
1330 * at 0, but doesn't enforce SEMVMX)
1332 * Linux caps the semaphore value, both at 0
1335 * Manfred <manfred@colorfullife.com>
1337 if (semaphore->semval < 0)
1338 semaphore->semval = 0;
1339 if (semaphore->semval > SEMVMX)
1340 semaphore->semval = SEMVMX;
1341 semaphore->sempid = task_tgid_vnr(current);
1344 sma->sem_otime = get_seconds();
1345 /* maybe some queued-up processes were waiting for this */
1349 call_rcu(&un->rcu, free_un);
1354 #ifdef CONFIG_PROC_FS
1355 static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
1357 struct sem_array *sma = it;
1359 return seq_printf(s,
1360 "%10d %10d %4o %10lu %5u %5u %5u %5u %10lu %10lu\n",