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);
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 /* Free a semaphore set. freeary() is called with sem_ids.rw_mutex locked
508 * as a writer and the spinlock for this semaphore set hold. sem_ids.rw_mutex
509 * remains locked on exit.
511 static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
514 struct sem_queue *q, *t;
515 struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
517 /* Invalidate the existing undo structures for this semaphore set.
518 * (They will be freed without any further action in exit_sem()
519 * or during the next semop.)
521 assert_spin_locked(&sma->sem_perm.lock);
522 list_for_each_entry(un, &sma->list_id, list_id)
525 /* Wake up all pending processes and let them fail with EIDRM. */
527 list_for_each_entry_safe(q, t, &sma->sem_pending, list) {
530 q->status = IN_WAKEUP;
531 wake_up_process(q->sleeper); /* doesn't sleep */
533 q->status = -EIDRM; /* hands-off q */
536 /* Remove the semaphore set from the IDR */
540 ns->used_sems -= sma->sem_nsems;
541 security_sem_free(sma);
545 static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
549 return copy_to_user(buf, in, sizeof(*in));
554 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
556 out.sem_otime = in->sem_otime;
557 out.sem_ctime = in->sem_ctime;
558 out.sem_nsems = in->sem_nsems;
560 return copy_to_user(buf, &out, sizeof(out));
567 static int semctl_nolock(struct ipc_namespace *ns, int semid,
568 int cmd, int version, union semun arg)
571 struct sem_array *sma;
577 struct seminfo seminfo;
580 err = security_sem_semctl(NULL, cmd);
584 memset(&seminfo,0,sizeof(seminfo));
585 seminfo.semmni = ns->sc_semmni;
586 seminfo.semmns = ns->sc_semmns;
587 seminfo.semmsl = ns->sc_semmsl;
588 seminfo.semopm = ns->sc_semopm;
589 seminfo.semvmx = SEMVMX;
590 seminfo.semmnu = SEMMNU;
591 seminfo.semmap = SEMMAP;
592 seminfo.semume = SEMUME;
593 down_read(&sem_ids(ns).rw_mutex);
594 if (cmd == SEM_INFO) {
595 seminfo.semusz = sem_ids(ns).in_use;
596 seminfo.semaem = ns->used_sems;
598 seminfo.semusz = SEMUSZ;
599 seminfo.semaem = SEMAEM;
601 max_id = ipc_get_maxid(&sem_ids(ns));
602 up_read(&sem_ids(ns).rw_mutex);
603 if (copy_to_user (arg.__buf, &seminfo, sizeof(struct seminfo)))
605 return (max_id < 0) ? 0: max_id;
610 struct semid64_ds tbuf;
613 if (cmd == SEM_STAT) {
614 sma = sem_lock(ns, semid);
617 id = sma->sem_perm.id;
619 sma = sem_lock_check(ns, semid);
626 if (ipcperms (&sma->sem_perm, S_IRUGO))
629 err = security_sem_semctl(sma, cmd);
633 memset(&tbuf, 0, sizeof(tbuf));
635 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
636 tbuf.sem_otime = sma->sem_otime;
637 tbuf.sem_ctime = sma->sem_ctime;
638 tbuf.sem_nsems = sma->sem_nsems;
640 if (copy_semid_to_user (arg.buf, &tbuf, version))
653 static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
654 int cmd, int version, union semun arg)
656 struct sem_array *sma;
659 ushort fast_sem_io[SEMMSL_FAST];
660 ushort* sem_io = fast_sem_io;
663 sma = sem_lock_check(ns, semid);
667 nsems = sma->sem_nsems;
670 if (ipcperms (&sma->sem_perm, (cmd==SETVAL||cmd==SETALL)?S_IWUGO:S_IRUGO))
673 err = security_sem_semctl(sma, cmd);
681 ushort __user *array = arg.array;
684 if(nsems > SEMMSL_FAST) {
685 sem_getref_and_unlock(sma);
687 sem_io = ipc_alloc(sizeof(ushort)*nsems);
693 sem_lock_and_putref(sma);
694 if (sma->sem_perm.deleted) {
701 for (i = 0; i < sma->sem_nsems; i++)
702 sem_io[i] = sma->sem_base[i].semval;
705 if(copy_to_user(array, sem_io, nsems*sizeof(ushort)))
714 sem_getref_and_unlock(sma);
716 if(nsems > SEMMSL_FAST) {
717 sem_io = ipc_alloc(sizeof(ushort)*nsems);
724 if (copy_from_user (sem_io, arg.array, nsems*sizeof(ushort))) {
730 for (i = 0; i < nsems; i++) {
731 if (sem_io[i] > SEMVMX) {
737 sem_lock_and_putref(sma);
738 if (sma->sem_perm.deleted) {
744 for (i = 0; i < nsems; i++)
745 sma->sem_base[i].semval = sem_io[i];
747 assert_spin_locked(&sma->sem_perm.lock);
748 list_for_each_entry(un, &sma->list_id, list_id) {
749 for (i = 0; i < nsems; i++)
752 sma->sem_ctime = get_seconds();
753 /* maybe some queued-up processes were waiting for this */
758 /* GETVAL, GETPID, GETNCTN, GETZCNT, SETVAL: fall-through */
761 if(semnum < 0 || semnum >= nsems)
764 curr = &sma->sem_base[semnum];
774 err = count_semncnt(sma,semnum);
777 err = count_semzcnt(sma,semnum);
785 if (val > SEMVMX || val < 0)
788 assert_spin_locked(&sma->sem_perm.lock);
789 list_for_each_entry(un, &sma->list_id, list_id)
790 un->semadj[semnum] = 0;
793 curr->sempid = task_tgid_vnr(current);
794 sma->sem_ctime = get_seconds();
795 /* maybe some queued-up processes were waiting for this */
804 if(sem_io != fast_sem_io)
805 ipc_free(sem_io, sizeof(ushort)*nsems);
809 static inline unsigned long
810 copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
814 if (copy_from_user(out, buf, sizeof(*out)))
819 struct semid_ds tbuf_old;
821 if(copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
824 out->sem_perm.uid = tbuf_old.sem_perm.uid;
825 out->sem_perm.gid = tbuf_old.sem_perm.gid;
826 out->sem_perm.mode = tbuf_old.sem_perm.mode;
836 * This function handles some semctl commands which require the rw_mutex
837 * to be held in write mode.
838 * NOTE: no locks must be held, the rw_mutex is taken inside this function.
840 static int semctl_down(struct ipc_namespace *ns, int semid,
841 int cmd, int version, union semun arg)
843 struct sem_array *sma;
845 struct semid64_ds semid64;
846 struct kern_ipc_perm *ipcp;
849 if (copy_semid_from_user(&semid64, arg.buf, version))
853 ipcp = ipcctl_pre_down(&sem_ids(ns), semid, cmd, &semid64.sem_perm, 0);
855 return PTR_ERR(ipcp);
857 sma = container_of(ipcp, struct sem_array, sem_perm);
859 err = security_sem_semctl(sma, cmd);
868 ipc_update_perm(&semid64.sem_perm, ipcp);
869 sma->sem_ctime = get_seconds();
878 up_write(&sem_ids(ns).rw_mutex);
882 asmlinkage long sys_semctl (int semid, int semnum, int cmd, union semun arg)
886 struct ipc_namespace *ns;
891 version = ipc_parse_version(&cmd);
892 ns = current->nsproxy->ipc_ns;
899 err = semctl_nolock(ns, semid, cmd, version, arg);
908 err = semctl_main(ns,semid,semnum,cmd,version,arg);
912 err = semctl_down(ns, semid, cmd, version, arg);
919 /* If the task doesn't already have a undo_list, then allocate one
920 * here. We guarantee there is only one thread using this undo list,
921 * and current is THE ONE
923 * If this allocation and assignment succeeds, but later
924 * portions of this code fail, there is no need to free the sem_undo_list.
925 * Just let it stay associated with the task, and it'll be freed later
928 * This can block, so callers must hold no locks.
930 static inline int get_undo_list(struct sem_undo_list **undo_listp)
932 struct sem_undo_list *undo_list;
934 undo_list = current->sysvsem.undo_list;
936 undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
937 if (undo_list == NULL)
939 spin_lock_init(&undo_list->lock);
940 atomic_set(&undo_list->refcnt, 1);
941 INIT_LIST_HEAD(&undo_list->list_proc);
943 current->sysvsem.undo_list = undo_list;
945 *undo_listp = undo_list;
949 static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
951 struct sem_undo *walk, *tmp;
953 assert_spin_locked(&ulp->lock);
954 list_for_each_entry_safe(walk, tmp, &ulp->list_proc, list_proc) {
955 if (walk->semid == semid)
957 if (walk->semid == -1) {
958 list_del(&walk->list_proc);
966 * find_alloc_undo - Lookup (and if not present create) undo array
968 * @semid: semaphore array id
970 * The function looks up (and if not present creates) the undo structure.
971 * The size of the undo structure depends on the size of the semaphore
972 * array, thus the alloc path is not that straightforward.
974 static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
976 struct sem_array *sma;
977 struct sem_undo_list *ulp;
978 struct sem_undo *un, *new;
982 error = get_undo_list(&ulp);
984 return ERR_PTR(error);
986 spin_lock(&ulp->lock);
987 un = lookup_undo(ulp, semid);
988 spin_unlock(&ulp->lock);
989 if (likely(un!=NULL))
992 /* no undo structure around - allocate one. */
993 /* step 1: figure out the size of the semaphore array */
994 sma = sem_lock_check(ns, semid);
996 return ERR_PTR(PTR_ERR(sma));
998 nsems = sma->sem_nsems;
999 sem_getref_and_unlock(sma);
1001 /* step 2: allocate new undo structure */
1002 new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1005 return ERR_PTR(-ENOMEM);
1008 /* step 3: Acquire the lock on the undo list pointer */
1009 spin_lock(&ulp->lock);
1011 /* step 4: check for races: someone else allocated the undo struct,
1012 * semaphore array was destroyed.
1014 un = lookup_undo(ulp, semid);
1016 spin_unlock(&ulp->lock);
1021 sem_lock_and_putref(sma);
1022 if (sma->sem_perm.deleted) {
1024 spin_unlock(&ulp->lock);
1026 un = ERR_PTR(-EIDRM);
1029 /* step 5: initialize & link new undo structure */
1030 new->semadj = (short *) &new[1];
1032 assert_spin_locked(&ulp->lock);
1033 list_add(&new->list_proc, &ulp->list_proc);
1034 assert_spin_locked(&sma->sem_perm.lock);
1035 list_add(&new->list_id, &sma->list_id);
1038 spin_unlock(&ulp->lock);
1044 asmlinkage long sys_semtimedop(int semid, struct sembuf __user *tsops,
1045 unsigned nsops, const struct timespec __user *timeout)
1047 int error = -EINVAL;
1048 struct sem_array *sma;
1049 struct sembuf fast_sops[SEMOPM_FAST];
1050 struct sembuf* sops = fast_sops, *sop;
1051 struct sem_undo *un;
1052 int undos = 0, alter = 0, max;
1053 struct sem_queue queue;
1054 unsigned long jiffies_left = 0;
1055 struct ipc_namespace *ns;
1057 ns = current->nsproxy->ipc_ns;
1059 if (nsops < 1 || semid < 0)
1061 if (nsops > ns->sc_semopm)
1063 if(nsops > SEMOPM_FAST) {
1064 sops = kmalloc(sizeof(*sops)*nsops,GFP_KERNEL);
1068 if (copy_from_user (sops, tsops, nsops * sizeof(*tsops))) {
1073 struct timespec _timeout;
1074 if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1078 if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1079 _timeout.tv_nsec >= 1000000000L) {
1083 jiffies_left = timespec_to_jiffies(&_timeout);
1086 for (sop = sops; sop < sops + nsops; sop++) {
1087 if (sop->sem_num >= max)
1089 if (sop->sem_flg & SEM_UNDO)
1091 if (sop->sem_op != 0)
1096 un = find_alloc_undo(ns, semid);
1098 error = PTR_ERR(un);
1104 sma = sem_lock_check(ns, semid);
1106 error = PTR_ERR(sma);
1111 * semid identifiers are not unique - find_alloc_undo may have
1112 * allocated an undo structure, it was invalidated by an RMID
1113 * and now a new array with received the same id. Check and fail.
1116 if (un && un->semid == -1)
1117 goto out_unlock_free;
1120 if (max >= sma->sem_nsems)
1121 goto out_unlock_free;
1124 if (ipcperms(&sma->sem_perm, alter ? S_IWUGO : S_IRUGO))
1125 goto out_unlock_free;
1127 error = security_sem_semop(sma, sops, nsops, alter);
1129 goto out_unlock_free;
1131 error = try_atomic_semop (sma, sops, nsops, un, task_tgid_vnr(current));
1133 if (alter && error == 0)
1135 goto out_unlock_free;
1138 /* We need to sleep on this operation, so we put the current
1139 * task into the pending queue and go to sleep.
1143 queue.nsops = nsops;
1145 queue.pid = task_tgid_vnr(current);
1146 queue.alter = alter;
1148 list_add_tail(&queue.list, &sma->sem_pending);
1150 list_add(&queue.list, &sma->sem_pending);
1152 queue.status = -EINTR;
1153 queue.sleeper = current;
1154 current->state = TASK_INTERRUPTIBLE;
1158 jiffies_left = schedule_timeout(jiffies_left);
1162 error = queue.status;
1163 while(unlikely(error == IN_WAKEUP)) {
1165 error = queue.status;
1168 if (error != -EINTR) {
1169 /* fast path: update_queue already obtained all requested
1174 sma = sem_lock(ns, semid);
1181 * If queue.status != -EINTR we are woken up by another process
1183 error = queue.status;
1184 if (error != -EINTR) {
1185 goto out_unlock_free;
1189 * If an interrupt occurred we have to clean up the queue
1191 if (timeout && jiffies_left == 0)
1193 list_del(&queue.list);
1194 goto out_unlock_free;
1199 if(sops != fast_sops)
1204 asmlinkage long sys_semop (int semid, struct sembuf __user *tsops, unsigned nsops)
1206 return sys_semtimedop(semid, tsops, nsops, NULL);
1209 /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
1210 * parent and child tasks.
1213 int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
1215 struct sem_undo_list *undo_list;
1218 if (clone_flags & CLONE_SYSVSEM) {
1219 error = get_undo_list(&undo_list);
1222 atomic_inc(&undo_list->refcnt);
1223 tsk->sysvsem.undo_list = undo_list;
1225 tsk->sysvsem.undo_list = NULL;
1231 * add semadj values to semaphores, free undo structures.
1232 * undo structures are not freed when semaphore arrays are destroyed
1233 * so some of them may be out of date.
1234 * IMPLEMENTATION NOTE: There is some confusion over whether the
1235 * set of adjustments that needs to be done should be done in an atomic
1236 * manner or not. That is, if we are attempting to decrement the semval
1237 * should we queue up and wait until we can do so legally?
1238 * The original implementation attempted to do this (queue and wait).
1239 * The current implementation does not do so. The POSIX standard
1240 * and SVID should be consulted to determine what behavior is mandated.
1242 void exit_sem(struct task_struct *tsk)
1244 struct sem_undo_list *ulp;
1245 struct sem_undo *un, *tmp;
1247 ulp = tsk->sysvsem.undo_list;
1250 tsk->sysvsem.undo_list = NULL;
1252 if (!atomic_dec_and_test(&ulp->refcnt))
1255 spin_lock(&ulp->lock);
1257 list_for_each_entry_safe(un, tmp, &ulp->list_proc, list_proc) {
1258 struct sem_array *sma;
1261 if (un->semid == -1)
1264 sma = sem_lock(tsk->nsproxy->ipc_ns, un->semid);
1268 if (un->semid == -1)
1271 BUG_ON(sem_checkid(sma, un->semid));
1273 /* remove un from sma->list_id */
1274 assert_spin_locked(&sma->sem_perm.lock);
1275 list_del(&un->list_id);
1277 /* perform adjustments registered in un */
1278 for (i = 0; i < sma->sem_nsems; i++) {
1279 struct sem * semaphore = &sma->sem_base[i];
1280 if (un->semadj[i]) {
1281 semaphore->semval += un->semadj[i];
1283 * Range checks of the new semaphore value,
1284 * not defined by sus:
1285 * - Some unices ignore the undo entirely
1286 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
1287 * - some cap the value (e.g. FreeBSD caps
1288 * at 0, but doesn't enforce SEMVMX)
1290 * Linux caps the semaphore value, both at 0
1293 * Manfred <manfred@colorfullife.com>
1295 if (semaphore->semval < 0)
1296 semaphore->semval = 0;
1297 if (semaphore->semval > SEMVMX)
1298 semaphore->semval = SEMVMX;
1299 semaphore->sempid = task_tgid_vnr(current);
1302 sma->sem_otime = get_seconds();
1303 /* maybe some queued-up processes were waiting for this */
1308 assert_spin_locked(&ulp->lock);
1309 list_del(&un->list_proc);
1312 spin_unlock(&ulp->lock);
1316 #ifdef CONFIG_PROC_FS
1317 static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
1319 struct sem_array *sma = it;
1321 return seq_printf(s,
1322 "%10d %10d %4o %10lu %5u %5u %5u %5u %10lu %10lu\n",