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 <alan@redhat.com>
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);
132 idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr);
136 void __init sem_init (void)
138 sem_init_ns(&init_ipc_ns);
139 ipc_init_proc_interface("sysvipc/sem",
140 " key semid perms nsems uid gid cuid cgid otime ctime\n",
141 IPC_SEM_IDS, sysvipc_sem_proc_show);
145 * sem_lock_(check_) routines are called in the paths where the rw_mutex
148 static inline struct sem_array *sem_lock(struct ipc_namespace *ns, int id)
150 struct kern_ipc_perm *ipcp = ipc_lock(&sem_ids(ns), id);
153 return (struct sem_array *)ipcp;
155 return container_of(ipcp, struct sem_array, sem_perm);
158 static inline struct sem_array *sem_lock_check(struct ipc_namespace *ns,
161 struct kern_ipc_perm *ipcp = ipc_lock_check(&sem_ids(ns), id);
164 return (struct sem_array *)ipcp;
166 return container_of(ipcp, struct sem_array, sem_perm);
169 static inline void sem_lock_and_putref(struct sem_array *sma)
171 ipc_lock_by_ptr(&sma->sem_perm);
175 static inline void sem_getref_and_unlock(struct sem_array *sma)
178 ipc_unlock(&(sma)->sem_perm);
181 static inline void sem_putref(struct sem_array *sma)
183 ipc_lock_by_ptr(&sma->sem_perm);
185 ipc_unlock(&(sma)->sem_perm);
188 static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
190 ipc_rmid(&sem_ids(ns), &s->sem_perm);
194 * Lockless wakeup algorithm:
195 * Without the check/retry algorithm a lockless wakeup is possible:
196 * - queue.status is initialized to -EINTR before blocking.
197 * - wakeup is performed by
198 * * unlinking the queue entry from sma->sem_pending
199 * * setting queue.status to IN_WAKEUP
200 * This is the notification for the blocked thread that a
201 * result value is imminent.
202 * * call wake_up_process
203 * * set queue.status to the final value.
204 * - the previously blocked thread checks queue.status:
205 * * if it's IN_WAKEUP, then it must wait until the value changes
206 * * if it's not -EINTR, then the operation was completed by
207 * update_queue. semtimedop can return queue.status without
208 * performing any operation on the sem array.
209 * * otherwise it must acquire the spinlock and check what's up.
211 * The two-stage algorithm is necessary to protect against the following
213 * - if queue.status is set after wake_up_process, then the woken up idle
214 * thread could race forward and try (and fail) to acquire sma->lock
215 * before update_queue had a chance to set queue.status
216 * - if queue.status is written before wake_up_process and if the
217 * blocked process is woken up by a signal between writing
218 * queue.status and the wake_up_process, then the woken up
219 * process could return from semtimedop and die by calling
220 * sys_exit before wake_up_process is called. Then wake_up_process
221 * will oops, because the task structure is already invalid.
222 * (yes, this happened on s390 with sysv msg).
228 * newary - Create a new semaphore set
230 * @params: ptr to the structure that contains key, semflg and nsems
232 * Called with sem_ids.rw_mutex held (as a writer)
235 static int newary(struct ipc_namespace *ns, struct ipc_params *params)
239 struct sem_array *sma;
241 key_t key = params->key;
242 int nsems = params->u.nsems;
243 int semflg = params->flg;
247 if (ns->used_sems + nsems > ns->sc_semmns)
250 size = sizeof (*sma) + nsems * sizeof (struct sem);
251 sma = ipc_rcu_alloc(size);
255 memset (sma, 0, size);
257 sma->sem_perm.mode = (semflg & S_IRWXUGO);
258 sma->sem_perm.key = key;
260 sma->sem_perm.security = NULL;
261 retval = security_sem_alloc(sma);
267 id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
269 security_sem_free(sma);
273 ns->used_sems += nsems;
275 sma->sem_base = (struct sem *) &sma[1];
276 INIT_LIST_HEAD(&sma->sem_pending);
277 INIT_LIST_HEAD(&sma->list_id);
278 sma->sem_nsems = nsems;
279 sma->sem_ctime = get_seconds();
282 return sma->sem_perm.id;
287 * Called with sem_ids.rw_mutex and ipcp locked.
289 static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg)
291 struct sem_array *sma;
293 sma = container_of(ipcp, struct sem_array, sem_perm);
294 return security_sem_associate(sma, semflg);
298 * Called with sem_ids.rw_mutex and ipcp locked.
300 static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
301 struct ipc_params *params)
303 struct sem_array *sma;
305 sma = container_of(ipcp, struct sem_array, sem_perm);
306 if (params->u.nsems > sma->sem_nsems)
312 SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg)
314 struct ipc_namespace *ns;
315 struct ipc_ops sem_ops;
316 struct ipc_params sem_params;
318 ns = current->nsproxy->ipc_ns;
320 if (nsems < 0 || nsems > ns->sc_semmsl)
323 sem_ops.getnew = newary;
324 sem_ops.associate = sem_security;
325 sem_ops.more_checks = sem_more_checks;
327 sem_params.key = key;
328 sem_params.flg = semflg;
329 sem_params.u.nsems = nsems;
331 return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
335 * Determine whether a sequence of semaphore operations would succeed
336 * all at once. Return 0 if yes, 1 if need to sleep, else return error code.
339 static int try_atomic_semop (struct sem_array * sma, struct sembuf * sops,
340 int nsops, struct sem_undo *un, int pid)
346 for (sop = sops; sop < sops + nsops; sop++) {
347 curr = sma->sem_base + sop->sem_num;
348 sem_op = sop->sem_op;
349 result = curr->semval;
351 if (!sem_op && result)
359 if (sop->sem_flg & SEM_UNDO) {
360 int undo = un->semadj[sop->sem_num] - sem_op;
362 * Exceeding the undo range is an error.
364 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
367 curr->semval = result;
371 while (sop >= sops) {
372 sma->sem_base[sop->sem_num].sempid = pid;
373 if (sop->sem_flg & SEM_UNDO)
374 un->semadj[sop->sem_num] -= sop->sem_op;
378 sma->sem_otime = get_seconds();
386 if (sop->sem_flg & IPC_NOWAIT)
393 while (sop >= sops) {
394 sma->sem_base[sop->sem_num].semval -= sop->sem_op;
401 /* Go through the pending queue for the indicated semaphore
402 * looking for tasks that can be completed.
404 static void update_queue (struct sem_array * sma)
407 struct sem_queue * q;
409 q = list_entry(sma->sem_pending.next, struct sem_queue, list);
410 while (&q->list != &sma->sem_pending) {
411 error = try_atomic_semop(sma, q->sops, q->nsops,
414 /* Does q->sleeper still need to sleep? */
419 * Continue scanning. The next operation
420 * that must be checked depends on the type of the
421 * completed operation:
422 * - if the operation modified the array, then
423 * restart from the head of the queue and
424 * check for threads that might be waiting
425 * for semaphore values to become 0.
426 * - if the operation didn't modify the array,
427 * then just continue.
428 * The order of list_del() and reading ->next
429 * is crucial: In the former case, the list_del()
430 * must be done first [because we might be the
431 * first entry in ->sem_pending], in the latter
432 * case the list_del() must be done last
433 * [because the list is invalid after the list_del()]
437 n = list_entry(sma->sem_pending.next,
438 struct sem_queue, list);
440 n = list_entry(q->list.next, struct sem_queue,
445 /* wake up the waiting thread */
446 q->status = IN_WAKEUP;
448 wake_up_process(q->sleeper);
449 /* hands-off: q will disappear immediately after
456 q = list_entry(q->list.next, struct sem_queue, list);
461 /* The following counts are associated to each semaphore:
462 * semncnt number of tasks waiting on semval being nonzero
463 * semzcnt number of tasks waiting on semval being zero
464 * This model assumes that a task waits on exactly one semaphore.
465 * Since semaphore operations are to be performed atomically, tasks actually
466 * wait on a whole sequence of semaphores simultaneously.
467 * The counts we return here are a rough approximation, but still
468 * warrant that semncnt+semzcnt>0 if the task is on the pending queue.
470 static int count_semncnt (struct sem_array * sma, ushort semnum)
473 struct sem_queue * q;
476 list_for_each_entry(q, &sma->sem_pending, list) {
477 struct sembuf * sops = q->sops;
478 int nsops = q->nsops;
480 for (i = 0; i < nsops; i++)
481 if (sops[i].sem_num == semnum
482 && (sops[i].sem_op < 0)
483 && !(sops[i].sem_flg & IPC_NOWAIT))
489 static int count_semzcnt (struct sem_array * sma, ushort semnum)
492 struct sem_queue * q;
495 list_for_each_entry(q, &sma->sem_pending, list) {
496 struct sembuf * sops = q->sops;
497 int nsops = q->nsops;
499 for (i = 0; i < nsops; i++)
500 if (sops[i].sem_num == semnum
501 && (sops[i].sem_op == 0)
502 && !(sops[i].sem_flg & IPC_NOWAIT))
508 void free_un(struct rcu_head *head)
510 struct sem_undo *un = container_of(head, struct sem_undo, rcu);
514 /* Free a semaphore set. freeary() is called with sem_ids.rw_mutex locked
515 * as a writer and the spinlock for this semaphore set hold. sem_ids.rw_mutex
516 * remains locked on exit.
518 static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
520 struct sem_undo *un, *tu;
521 struct sem_queue *q, *tq;
522 struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
524 /* Free the existing undo structures for this semaphore set. */
525 assert_spin_locked(&sma->sem_perm.lock);
526 list_for_each_entry_safe(un, tu, &sma->list_id, list_id) {
527 list_del(&un->list_id);
528 spin_lock(&un->ulp->lock);
530 list_del_rcu(&un->list_proc);
531 spin_unlock(&un->ulp->lock);
532 call_rcu(&un->rcu, free_un);
535 /* Wake up all pending processes and let them fail with EIDRM. */
536 list_for_each_entry_safe(q, tq, &sma->sem_pending, list) {
539 q->status = IN_WAKEUP;
540 wake_up_process(q->sleeper); /* doesn't sleep */
542 q->status = -EIDRM; /* hands-off q */
545 /* Remove the semaphore set from the IDR */
549 ns->used_sems -= sma->sem_nsems;
550 security_sem_free(sma);
554 static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
558 return copy_to_user(buf, in, sizeof(*in));
563 memset(&out, 0, sizeof(out));
565 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
567 out.sem_otime = in->sem_otime;
568 out.sem_ctime = in->sem_ctime;
569 out.sem_nsems = in->sem_nsems;
571 return copy_to_user(buf, &out, sizeof(out));
578 static int semctl_nolock(struct ipc_namespace *ns, int semid,
579 int cmd, int version, union semun arg)
582 struct sem_array *sma;
588 struct seminfo seminfo;
591 err = security_sem_semctl(NULL, cmd);
595 memset(&seminfo,0,sizeof(seminfo));
596 seminfo.semmni = ns->sc_semmni;
597 seminfo.semmns = ns->sc_semmns;
598 seminfo.semmsl = ns->sc_semmsl;
599 seminfo.semopm = ns->sc_semopm;
600 seminfo.semvmx = SEMVMX;
601 seminfo.semmnu = SEMMNU;
602 seminfo.semmap = SEMMAP;
603 seminfo.semume = SEMUME;
604 down_read(&sem_ids(ns).rw_mutex);
605 if (cmd == SEM_INFO) {
606 seminfo.semusz = sem_ids(ns).in_use;
607 seminfo.semaem = ns->used_sems;
609 seminfo.semusz = SEMUSZ;
610 seminfo.semaem = SEMAEM;
612 max_id = ipc_get_maxid(&sem_ids(ns));
613 up_read(&sem_ids(ns).rw_mutex);
614 if (copy_to_user (arg.__buf, &seminfo, sizeof(struct seminfo)))
616 return (max_id < 0) ? 0: max_id;
621 struct semid64_ds tbuf;
624 if (cmd == SEM_STAT) {
625 sma = sem_lock(ns, semid);
628 id = sma->sem_perm.id;
630 sma = sem_lock_check(ns, semid);
637 if (ipcperms (&sma->sem_perm, S_IRUGO))
640 err = security_sem_semctl(sma, cmd);
644 memset(&tbuf, 0, sizeof(tbuf));
646 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
647 tbuf.sem_otime = sma->sem_otime;
648 tbuf.sem_ctime = sma->sem_ctime;
649 tbuf.sem_nsems = sma->sem_nsems;
651 if (copy_semid_to_user (arg.buf, &tbuf, version))
664 static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
665 int cmd, int version, union semun arg)
667 struct sem_array *sma;
670 ushort fast_sem_io[SEMMSL_FAST];
671 ushort* sem_io = fast_sem_io;
674 sma = sem_lock_check(ns, semid);
678 nsems = sma->sem_nsems;
681 if (ipcperms (&sma->sem_perm, (cmd==SETVAL||cmd==SETALL)?S_IWUGO:S_IRUGO))
684 err = security_sem_semctl(sma, cmd);
692 ushort __user *array = arg.array;
695 if(nsems > SEMMSL_FAST) {
696 sem_getref_and_unlock(sma);
698 sem_io = ipc_alloc(sizeof(ushort)*nsems);
704 sem_lock_and_putref(sma);
705 if (sma->sem_perm.deleted) {
712 for (i = 0; i < sma->sem_nsems; i++)
713 sem_io[i] = sma->sem_base[i].semval;
716 if(copy_to_user(array, sem_io, nsems*sizeof(ushort)))
725 sem_getref_and_unlock(sma);
727 if(nsems > SEMMSL_FAST) {
728 sem_io = ipc_alloc(sizeof(ushort)*nsems);
735 if (copy_from_user (sem_io, arg.array, nsems*sizeof(ushort))) {
741 for (i = 0; i < nsems; i++) {
742 if (sem_io[i] > SEMVMX) {
748 sem_lock_and_putref(sma);
749 if (sma->sem_perm.deleted) {
755 for (i = 0; i < nsems; i++)
756 sma->sem_base[i].semval = sem_io[i];
758 assert_spin_locked(&sma->sem_perm.lock);
759 list_for_each_entry(un, &sma->list_id, list_id) {
760 for (i = 0; i < nsems; i++)
763 sma->sem_ctime = get_seconds();
764 /* maybe some queued-up processes were waiting for this */
769 /* GETVAL, GETPID, GETNCTN, GETZCNT, SETVAL: fall-through */
772 if(semnum < 0 || semnum >= nsems)
775 curr = &sma->sem_base[semnum];
785 err = count_semncnt(sma,semnum);
788 err = count_semzcnt(sma,semnum);
796 if (val > SEMVMX || val < 0)
799 assert_spin_locked(&sma->sem_perm.lock);
800 list_for_each_entry(un, &sma->list_id, list_id)
801 un->semadj[semnum] = 0;
804 curr->sempid = task_tgid_vnr(current);
805 sma->sem_ctime = get_seconds();
806 /* maybe some queued-up processes were waiting for this */
815 if(sem_io != fast_sem_io)
816 ipc_free(sem_io, sizeof(ushort)*nsems);
820 static inline unsigned long
821 copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
825 if (copy_from_user(out, buf, sizeof(*out)))
830 struct semid_ds tbuf_old;
832 if(copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
835 out->sem_perm.uid = tbuf_old.sem_perm.uid;
836 out->sem_perm.gid = tbuf_old.sem_perm.gid;
837 out->sem_perm.mode = tbuf_old.sem_perm.mode;
847 * This function handles some semctl commands which require the rw_mutex
848 * to be held in write mode.
849 * NOTE: no locks must be held, the rw_mutex is taken inside this function.
851 static int semctl_down(struct ipc_namespace *ns, int semid,
852 int cmd, int version, union semun arg)
854 struct sem_array *sma;
856 struct semid64_ds semid64;
857 struct kern_ipc_perm *ipcp;
860 if (copy_semid_from_user(&semid64, arg.buf, version))
864 ipcp = ipcctl_pre_down(&sem_ids(ns), semid, cmd, &semid64.sem_perm, 0);
866 return PTR_ERR(ipcp);
868 sma = container_of(ipcp, struct sem_array, sem_perm);
870 err = security_sem_semctl(sma, cmd);
879 ipc_update_perm(&semid64.sem_perm, ipcp);
880 sma->sem_ctime = get_seconds();
889 up_write(&sem_ids(ns).rw_mutex);
893 SYSCALL_DEFINE(semctl)(int semid, int semnum, int cmd, union semun arg)
897 struct ipc_namespace *ns;
902 version = ipc_parse_version(&cmd);
903 ns = current->nsproxy->ipc_ns;
910 err = semctl_nolock(ns, semid, cmd, version, arg);
919 err = semctl_main(ns,semid,semnum,cmd,version,arg);
923 err = semctl_down(ns, semid, cmd, version, arg);
929 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
930 asmlinkage long SyS_semctl(int semid, int semnum, int cmd, union semun arg)
932 return SYSC_semctl((int) semid, (int) semnum, (int) cmd, arg);
934 SYSCALL_ALIAS(sys_semctl, SyS_semctl);
937 /* If the task doesn't already have a undo_list, then allocate one
938 * here. We guarantee there is only one thread using this undo list,
939 * and current is THE ONE
941 * If this allocation and assignment succeeds, but later
942 * portions of this code fail, there is no need to free the sem_undo_list.
943 * Just let it stay associated with the task, and it'll be freed later
946 * This can block, so callers must hold no locks.
948 static inline int get_undo_list(struct sem_undo_list **undo_listp)
950 struct sem_undo_list *undo_list;
952 undo_list = current->sysvsem.undo_list;
954 undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
955 if (undo_list == NULL)
957 spin_lock_init(&undo_list->lock);
958 atomic_set(&undo_list->refcnt, 1);
959 INIT_LIST_HEAD(&undo_list->list_proc);
961 current->sysvsem.undo_list = undo_list;
963 *undo_listp = undo_list;
967 static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
969 struct sem_undo *walk;
971 list_for_each_entry_rcu(walk, &ulp->list_proc, list_proc) {
972 if (walk->semid == semid)
979 * find_alloc_undo - Lookup (and if not present create) undo array
981 * @semid: semaphore array id
983 * The function looks up (and if not present creates) the undo structure.
984 * The size of the undo structure depends on the size of the semaphore
985 * array, thus the alloc path is not that straightforward.
986 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
987 * performs a rcu_read_lock().
989 static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
991 struct sem_array *sma;
992 struct sem_undo_list *ulp;
993 struct sem_undo *un, *new;
997 error = get_undo_list(&ulp);
999 return ERR_PTR(error);
1002 spin_lock(&ulp->lock);
1003 un = lookup_undo(ulp, semid);
1004 spin_unlock(&ulp->lock);
1005 if (likely(un!=NULL))
1009 /* no undo structure around - allocate one. */
1010 /* step 1: figure out the size of the semaphore array */
1011 sma = sem_lock_check(ns, semid);
1013 return ERR_PTR(PTR_ERR(sma));
1015 nsems = sma->sem_nsems;
1016 sem_getref_and_unlock(sma);
1018 /* step 2: allocate new undo structure */
1019 new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1022 return ERR_PTR(-ENOMEM);
1025 /* step 3: Acquire the lock on semaphore array */
1026 sem_lock_and_putref(sma);
1027 if (sma->sem_perm.deleted) {
1030 un = ERR_PTR(-EIDRM);
1033 spin_lock(&ulp->lock);
1036 * step 4: check for races: did someone else allocate the undo struct?
1038 un = lookup_undo(ulp, semid);
1043 /* step 5: initialize & link new undo structure */
1044 new->semadj = (short *) &new[1];
1047 assert_spin_locked(&ulp->lock);
1048 list_add_rcu(&new->list_proc, &ulp->list_proc);
1049 assert_spin_locked(&sma->sem_perm.lock);
1050 list_add(&new->list_id, &sma->list_id);
1054 spin_unlock(&ulp->lock);
1061 SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
1062 unsigned, nsops, const struct timespec __user *, timeout)
1064 int error = -EINVAL;
1065 struct sem_array *sma;
1066 struct sembuf fast_sops[SEMOPM_FAST];
1067 struct sembuf* sops = fast_sops, *sop;
1068 struct sem_undo *un;
1069 int undos = 0, alter = 0, max;
1070 struct sem_queue queue;
1071 unsigned long jiffies_left = 0;
1072 struct ipc_namespace *ns;
1074 ns = current->nsproxy->ipc_ns;
1076 if (nsops < 1 || semid < 0)
1078 if (nsops > ns->sc_semopm)
1080 if(nsops > SEMOPM_FAST) {
1081 sops = kmalloc(sizeof(*sops)*nsops,GFP_KERNEL);
1085 if (copy_from_user (sops, tsops, nsops * sizeof(*tsops))) {
1090 struct timespec _timeout;
1091 if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1095 if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1096 _timeout.tv_nsec >= 1000000000L) {
1100 jiffies_left = timespec_to_jiffies(&_timeout);
1103 for (sop = sops; sop < sops + nsops; sop++) {
1104 if (sop->sem_num >= max)
1106 if (sop->sem_flg & SEM_UNDO)
1108 if (sop->sem_op != 0)
1113 un = find_alloc_undo(ns, semid);
1115 error = PTR_ERR(un);
1121 sma = sem_lock_check(ns, semid);
1125 error = PTR_ERR(sma);
1130 * semid identifiers are not unique - find_alloc_undo may have
1131 * allocated an undo structure, it was invalidated by an RMID
1132 * and now a new array with received the same id. Check and fail.
1133 * This case can be detected checking un->semid. The existance of
1134 * "un" itself is guaranteed by rcu.
1138 if (un->semid == -1) {
1140 goto out_unlock_free;
1143 * rcu lock can be released, "un" cannot disappear:
1144 * - sem_lock is acquired, thus IPC_RMID is
1146 * - exit_sem is impossible, it always operates on
1147 * current (or a dead task).
1155 if (max >= sma->sem_nsems)
1156 goto out_unlock_free;
1159 if (ipcperms(&sma->sem_perm, alter ? S_IWUGO : S_IRUGO))
1160 goto out_unlock_free;
1162 error = security_sem_semop(sma, sops, nsops, alter);
1164 goto out_unlock_free;
1166 error = try_atomic_semop (sma, sops, nsops, un, task_tgid_vnr(current));
1168 if (alter && error == 0)
1170 goto out_unlock_free;
1173 /* We need to sleep on this operation, so we put the current
1174 * task into the pending queue and go to sleep.
1178 queue.nsops = nsops;
1180 queue.pid = task_tgid_vnr(current);
1181 queue.alter = alter;
1183 list_add_tail(&queue.list, &sma->sem_pending);
1185 list_add(&queue.list, &sma->sem_pending);
1187 queue.status = -EINTR;
1188 queue.sleeper = current;
1189 current->state = TASK_INTERRUPTIBLE;
1193 jiffies_left = schedule_timeout(jiffies_left);
1197 error = queue.status;
1198 while(unlikely(error == IN_WAKEUP)) {
1200 error = queue.status;
1203 if (error != -EINTR) {
1204 /* fast path: update_queue already obtained all requested
1209 sma = sem_lock(ns, semid);
1216 * If queue.status != -EINTR we are woken up by another process
1218 error = queue.status;
1219 if (error != -EINTR) {
1220 goto out_unlock_free;
1224 * If an interrupt occurred we have to clean up the queue
1226 if (timeout && jiffies_left == 0)
1228 list_del(&queue.list);
1229 goto out_unlock_free;
1234 if(sops != fast_sops)
1239 SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
1242 return sys_semtimedop(semid, tsops, nsops, NULL);
1245 /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
1246 * parent and child tasks.
1249 int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
1251 struct sem_undo_list *undo_list;
1254 if (clone_flags & CLONE_SYSVSEM) {
1255 error = get_undo_list(&undo_list);
1258 atomic_inc(&undo_list->refcnt);
1259 tsk->sysvsem.undo_list = undo_list;
1261 tsk->sysvsem.undo_list = NULL;
1267 * add semadj values to semaphores, free undo structures.
1268 * undo structures are not freed when semaphore arrays are destroyed
1269 * so some of them may be out of date.
1270 * IMPLEMENTATION NOTE: There is some confusion over whether the
1271 * set of adjustments that needs to be done should be done in an atomic
1272 * manner or not. That is, if we are attempting to decrement the semval
1273 * should we queue up and wait until we can do so legally?
1274 * The original implementation attempted to do this (queue and wait).
1275 * The current implementation does not do so. The POSIX standard
1276 * and SVID should be consulted to determine what behavior is mandated.
1278 void exit_sem(struct task_struct *tsk)
1280 struct sem_undo_list *ulp;
1282 ulp = tsk->sysvsem.undo_list;
1285 tsk->sysvsem.undo_list = NULL;
1287 if (!atomic_dec_and_test(&ulp->refcnt))
1291 struct sem_array *sma;
1292 struct sem_undo *un;
1297 un = list_entry(rcu_dereference(ulp->list_proc.next),
1298 struct sem_undo, list_proc);
1299 if (&un->list_proc == &ulp->list_proc)
1308 sma = sem_lock_check(tsk->nsproxy->ipc_ns, un->semid);
1310 /* exit_sem raced with IPC_RMID, nothing to do */
1314 un = lookup_undo(ulp, semid);
1316 /* exit_sem raced with IPC_RMID+semget() that created
1317 * exactly the same semid. Nothing to do.
1323 /* remove un from the linked lists */
1324 assert_spin_locked(&sma->sem_perm.lock);
1325 list_del(&un->list_id);
1327 spin_lock(&ulp->lock);
1328 list_del_rcu(&un->list_proc);
1329 spin_unlock(&ulp->lock);
1331 /* perform adjustments registered in un */
1332 for (i = 0; i < sma->sem_nsems; i++) {
1333 struct sem * semaphore = &sma->sem_base[i];
1334 if (un->semadj[i]) {
1335 semaphore->semval += un->semadj[i];
1337 * Range checks of the new semaphore value,
1338 * not defined by sus:
1339 * - Some unices ignore the undo entirely
1340 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
1341 * - some cap the value (e.g. FreeBSD caps
1342 * at 0, but doesn't enforce SEMVMX)
1344 * Linux caps the semaphore value, both at 0
1347 * Manfred <manfred@colorfullife.com>
1349 if (semaphore->semval < 0)
1350 semaphore->semval = 0;
1351 if (semaphore->semval > SEMVMX)
1352 semaphore->semval = SEMVMX;
1353 semaphore->sempid = task_tgid_vnr(current);
1356 sma->sem_otime = get_seconds();
1357 /* maybe some queued-up processes were waiting for this */
1361 call_rcu(&un->rcu, free_un);
1366 #ifdef CONFIG_PROC_FS
1367 static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
1369 struct sem_array *sma = it;
1371 return seq_printf(s,
1372 "%10d %10d %4o %10lu %5u %5u %5u %5u %10lu %10lu\n",