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 <manfreds@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>
66 #include <linux/config.h>
67 #include <linux/slab.h>
68 #include <linux/spinlock.h>
69 #include <linux/init.h>
70 #include <linux/proc_fs.h>
71 #include <linux/time.h>
72 #include <linux/smp_lock.h>
73 #include <linux/security.h>
74 #include <asm/uaccess.h>
78 #define sem_lock(id) ((struct sem_array*)ipc_lock(&sem_ids,id))
79 #define sem_unlock(sma) ipc_unlock(&(sma)->sem_perm)
80 #define sem_rmid(id) ((struct sem_array*)ipc_rmid(&sem_ids,id))
81 #define sem_checkid(sma, semid) \
82 ipc_checkid(&sem_ids,&sma->sem_perm,semid)
83 #define sem_buildid(id, seq) \
84 ipc_buildid(&sem_ids, id, seq)
85 static struct ipc_ids sem_ids;
87 static int newary (key_t, int, int);
88 static void freeary (struct sem_array *sma, int id);
90 static int sysvipc_sem_read_proc(char *buffer, char **start, off_t offset, int length, int *eof, void *data);
93 #define SEMMSL_FAST 256 /* 512 bytes on stack */
94 #define SEMOPM_FAST 64 /* ~ 372 bytes on stack */
97 * linked list protection:
99 * sem_array.sem_pending{,last},
100 * sem_array.sem_undo: sem_lock() for read/write
101 * sem_undo.proc_next: only "current" is allowed to read/write that field.
105 int sem_ctls[4] = {SEMMSL, SEMMNS, SEMOPM, SEMMNI};
106 #define sc_semmsl (sem_ctls[0])
107 #define sc_semmns (sem_ctls[1])
108 #define sc_semopm (sem_ctls[2])
109 #define sc_semmni (sem_ctls[3])
111 static int used_sems;
113 void __init sem_init (void)
116 ipc_init_ids(&sem_ids,sc_semmni);
118 #ifdef CONFIG_PROC_FS
119 create_proc_read_entry("sysvipc/sem", 0, 0, sysvipc_sem_read_proc, NULL);
124 * Lockless wakeup algorithm:
125 * Without the check/retry algorithm a lockless wakeup is possible:
126 * - queue.status is initialized to -EINTR before blocking.
127 * - wakeup is performed by
128 * * unlinking the queue entry from sma->sem_pending
129 * * setting queue.status to IN_WAKEUP
130 * This is the notification for the blocked thread that a
131 * result value is imminent.
132 * * call wake_up_process
133 * * set queue.status to the final value.
134 * - the previously blocked thread checks queue.status:
135 * * if it's IN_WAKEUP, then it must wait until the value changes
136 * * if it's not -EINTR, then the operation was completed by
137 * update_queue. semtimedop can return queue.status without
138 * performing any operation on the semaphore array.
139 * * otherwise it must acquire the spinlock and check what's up.
141 * The two-stage algorithm is necessary to protect against the following
143 * - if queue.status is set after wake_up_process, then the woken up idle
144 * thread could race forward and try (and fail) to acquire sma->lock
145 * before update_queue had a chance to set queue.status
146 * - if queue.status is written before wake_up_process and if the
147 * blocked process is woken up by a signal between writing
148 * queue.status and the wake_up_process, then the woken up
149 * process could return from semtimedop and die by calling
150 * sys_exit before wake_up_process is called. Then wake_up_process
151 * will oops, because the task structure is already invalid.
152 * (yes, this happened on s390 with sysv msg).
157 static int newary (key_t key, int nsems, int semflg)
161 struct sem_array *sma;
166 if (used_sems + nsems > sc_semmns)
169 size = sizeof (*sma) + nsems * sizeof (struct sem);
170 sma = ipc_rcu_alloc(size);
174 memset (sma, 0, size);
176 sma->sem_perm.mode = (semflg & S_IRWXUGO);
177 sma->sem_perm.key = key;
179 sma->sem_perm.security = NULL;
180 retval = security_sem_alloc(sma);
182 ipc_rcu_free(sma, size);
186 id = ipc_addid(&sem_ids, &sma->sem_perm, sc_semmni);
188 security_sem_free(sma);
189 ipc_rcu_free(sma, size);
194 sma->sem_base = (struct sem *) &sma[1];
195 /* sma->sem_pending = NULL; */
196 sma->sem_pending_last = &sma->sem_pending;
197 /* sma->undo = NULL; */
198 sma->sem_nsems = nsems;
199 sma->sem_ctime = get_seconds();
202 return sem_buildid(id, sma->sem_perm.seq);
205 asmlinkage long sys_semget (key_t key, int nsems, int semflg)
207 int id, err = -EINVAL;
208 struct sem_array *sma;
210 if (nsems < 0 || nsems > sc_semmsl)
214 if (key == IPC_PRIVATE) {
215 err = newary(key, nsems, semflg);
216 } else if ((id = ipc_findkey(&sem_ids, key)) == -1) { /* key not used */
217 if (!(semflg & IPC_CREAT))
220 err = newary(key, nsems, semflg);
221 } else if (semflg & IPC_CREAT && semflg & IPC_EXCL) {
227 if (nsems > sma->sem_nsems)
229 else if (ipcperms(&sma->sem_perm, semflg))
232 int semid = sem_buildid(id, sma->sem_perm.seq);
233 err = security_sem_associate(sma, semflg);
244 /* doesn't acquire the sem_lock on error! */
245 static int sem_revalidate(int semid, struct sem_array* sma, int nsems, short flg)
247 struct sem_array* smanew;
249 smanew = sem_lock(semid);
252 if(smanew != sma || sem_checkid(sma,semid) || sma->sem_nsems != nsems) {
257 if (flg && ipcperms(&sma->sem_perm, flg)) {
263 /* Manage the doubly linked list sma->sem_pending as a FIFO:
264 * insert new queue elements at the tail sma->sem_pending_last.
266 static inline void append_to_queue (struct sem_array * sma,
267 struct sem_queue * q)
269 *(q->prev = sma->sem_pending_last) = q;
270 *(sma->sem_pending_last = &q->next) = NULL;
273 static inline void prepend_to_queue (struct sem_array * sma,
274 struct sem_queue * q)
276 q->next = sma->sem_pending;
277 *(q->prev = &sma->sem_pending) = q;
279 q->next->prev = &q->next;
280 else /* sma->sem_pending_last == &sma->sem_pending */
281 sma->sem_pending_last = &q->next;
284 static inline void remove_from_queue (struct sem_array * sma,
285 struct sem_queue * q)
287 *(q->prev) = q->next;
289 q->next->prev = q->prev;
290 else /* sma->sem_pending_last == &q->next */
291 sma->sem_pending_last = q->prev;
292 q->prev = NULL; /* mark as removed */
296 * Determine whether a sequence of semaphore operations would succeed
297 * all at once. Return 0 if yes, 1 if need to sleep, else return error code.
300 static int try_atomic_semop (struct sem_array * sma, struct sembuf * sops,
301 int nsops, struct sem_undo *un, int pid)
307 for (sop = sops; sop < sops + nsops; sop++) {
308 curr = sma->sem_base + sop->sem_num;
309 sem_op = sop->sem_op;
310 result = curr->semval;
312 if (!sem_op && result)
320 if (sop->sem_flg & SEM_UNDO) {
321 int undo = un->semadj[sop->sem_num] - sem_op;
323 * Exceeding the undo range is an error.
325 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
328 curr->semval = result;
332 while (sop >= sops) {
333 sma->sem_base[sop->sem_num].sempid = pid;
334 if (sop->sem_flg & SEM_UNDO)
335 un->semadj[sop->sem_num] -= sop->sem_op;
339 sma->sem_otime = get_seconds();
347 if (sop->sem_flg & IPC_NOWAIT)
354 while (sop >= sops) {
355 sma->sem_base[sop->sem_num].semval -= sop->sem_op;
362 /* Go through the pending queue for the indicated semaphore
363 * looking for tasks that can be completed.
365 static void update_queue (struct sem_array * sma)
368 struct sem_queue * q;
370 q = sma->sem_pending;
372 error = try_atomic_semop(sma, q->sops, q->nsops,
375 /* Does q->sleeper still need to sleep? */
378 remove_from_queue(sma,q);
380 q->status = IN_WAKEUP;
381 wake_up_process(q->sleeper);
382 /* hands-off: q will disappear immediately after
393 /* The following counts are associated to each semaphore:
394 * semncnt number of tasks waiting on semval being nonzero
395 * semzcnt number of tasks waiting on semval being zero
396 * This model assumes that a task waits on exactly one semaphore.
397 * Since semaphore operations are to be performed atomically, tasks actually
398 * wait on a whole sequence of semaphores simultaneously.
399 * The counts we return here are a rough approximation, but still
400 * warrant that semncnt+semzcnt>0 if the task is on the pending queue.
402 static int count_semncnt (struct sem_array * sma, ushort semnum)
405 struct sem_queue * q;
408 for (q = sma->sem_pending; q; q = q->next) {
409 struct sembuf * sops = q->sops;
410 int nsops = q->nsops;
412 for (i = 0; i < nsops; i++)
413 if (sops[i].sem_num == semnum
414 && (sops[i].sem_op < 0)
415 && !(sops[i].sem_flg & IPC_NOWAIT))
420 static int count_semzcnt (struct sem_array * sma, ushort semnum)
423 struct sem_queue * q;
426 for (q = sma->sem_pending; q; q = q->next) {
427 struct sembuf * sops = q->sops;
428 int nsops = q->nsops;
430 for (i = 0; i < nsops; i++)
431 if (sops[i].sem_num == semnum
432 && (sops[i].sem_op == 0)
433 && !(sops[i].sem_flg & IPC_NOWAIT))
439 /* Free a semaphore set. freeary() is called with sem_ids.sem down and
440 * the spinlock for this semaphore set hold. sem_ids.sem remains locked
443 static void freeary (struct sem_array *sma, int id)
449 /* Invalidate the existing undo structures for this semaphore set.
450 * (They will be freed without any further action in exit_sem()
451 * or during the next semop.)
453 for (un = sma->undo; un; un = un->id_next)
456 /* Wake up all pending processes and let them fail with EIDRM. */
457 q = sma->sem_pending;
460 /* lazy remove_from_queue: we are killing the whole queue */
463 q->status = IN_WAKEUP;
464 wake_up_process(q->sleeper); /* doesn't sleep */
465 q->status = -EIDRM; /* hands-off q */
469 /* Remove the semaphore set from the ID array*/
473 used_sems -= sma->sem_nsems;
474 size = sizeof (*sma) + sma->sem_nsems * sizeof (struct sem);
475 security_sem_free(sma);
476 ipc_rcu_free(sma, size);
479 static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
483 return copy_to_user(buf, in, sizeof(*in));
488 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
490 out.sem_otime = in->sem_otime;
491 out.sem_ctime = in->sem_ctime;
492 out.sem_nsems = in->sem_nsems;
494 return copy_to_user(buf, &out, sizeof(out));
501 static int semctl_nolock(int semid, int semnum, int cmd, int version, union semun arg)
504 struct sem_array *sma;
510 struct seminfo seminfo;
513 err = security_sem_semctl(NULL, cmd);
517 memset(&seminfo,0,sizeof(seminfo));
518 seminfo.semmni = sc_semmni;
519 seminfo.semmns = sc_semmns;
520 seminfo.semmsl = sc_semmsl;
521 seminfo.semopm = sc_semopm;
522 seminfo.semvmx = SEMVMX;
523 seminfo.semmnu = SEMMNU;
524 seminfo.semmap = SEMMAP;
525 seminfo.semume = SEMUME;
527 if (cmd == SEM_INFO) {
528 seminfo.semusz = sem_ids.in_use;
529 seminfo.semaem = used_sems;
531 seminfo.semusz = SEMUSZ;
532 seminfo.semaem = SEMAEM;
534 max_id = sem_ids.max_id;
536 if (copy_to_user (arg.__buf, &seminfo, sizeof(struct seminfo)))
538 return (max_id < 0) ? 0: max_id;
542 struct semid64_ds tbuf;
545 if(semid >= sem_ids.size)
548 memset(&tbuf,0,sizeof(tbuf));
550 sma = sem_lock(semid);
555 if (ipcperms (&sma->sem_perm, S_IRUGO))
558 err = security_sem_semctl(sma, cmd);
562 id = sem_buildid(semid, sma->sem_perm.seq);
564 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
565 tbuf.sem_otime = sma->sem_otime;
566 tbuf.sem_ctime = sma->sem_ctime;
567 tbuf.sem_nsems = sma->sem_nsems;
569 if (copy_semid_to_user (arg.buf, &tbuf, version))
582 static int semctl_main(int semid, int semnum, int cmd, int version, union semun arg)
584 struct sem_array *sma;
587 ushort fast_sem_io[SEMMSL_FAST];
588 ushort* sem_io = fast_sem_io;
591 sma = sem_lock(semid);
595 nsems = sma->sem_nsems;
598 if (sem_checkid(sma,semid))
602 if (ipcperms (&sma->sem_perm, (cmd==SETVAL||cmd==SETALL)?S_IWUGO:S_IRUGO))
605 err = security_sem_semctl(sma, cmd);
613 ushort *array = arg.array;
616 if(nsems > SEMMSL_FAST) {
618 sem_io = ipc_alloc(sizeof(ushort)*nsems);
621 err = sem_revalidate(semid, sma, nsems, S_IRUGO);
626 for (i = 0; i < sma->sem_nsems; i++)
627 sem_io[i] = sma->sem_base[i].semval;
630 if(copy_to_user(array, sem_io, nsems*sizeof(ushort)))
641 if(nsems > SEMMSL_FAST) {
642 sem_io = ipc_alloc(sizeof(ushort)*nsems);
647 if (copy_from_user (sem_io, arg.array, nsems*sizeof(ushort))) {
652 for (i = 0; i < nsems; i++) {
653 if (sem_io[i] > SEMVMX) {
658 err = sem_revalidate(semid, sma, nsems, S_IWUGO);
662 for (i = 0; i < nsems; i++)
663 sma->sem_base[i].semval = sem_io[i];
664 for (un = sma->undo; un; un = un->id_next)
665 for (i = 0; i < nsems; i++)
667 sma->sem_ctime = get_seconds();
668 /* maybe some queued-up processes were waiting for this */
675 struct semid64_ds tbuf;
676 memset(&tbuf,0,sizeof(tbuf));
677 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
678 tbuf.sem_otime = sma->sem_otime;
679 tbuf.sem_ctime = sma->sem_ctime;
680 tbuf.sem_nsems = sma->sem_nsems;
682 if (copy_semid_to_user (arg.buf, &tbuf, version))
686 /* GETVAL, GETPID, GETNCTN, GETZCNT, SETVAL: fall-through */
689 if(semnum < 0 || semnum >= nsems)
692 curr = &sma->sem_base[semnum];
702 err = count_semncnt(sma,semnum);
705 err = count_semzcnt(sma,semnum);
712 if (val > SEMVMX || val < 0)
715 for (un = sma->undo; un; un = un->id_next)
716 un->semadj[semnum] = 0;
718 curr->sempid = current->tgid;
719 sma->sem_ctime = get_seconds();
720 /* maybe some queued-up processes were waiting for this */
729 if(sem_io != fast_sem_io)
730 ipc_free(sem_io, sizeof(ushort)*nsems);
740 static inline unsigned long copy_semid_from_user(struct sem_setbuf *out, void __user *buf, int version)
745 struct semid64_ds tbuf;
747 if(copy_from_user(&tbuf, buf, sizeof(tbuf)))
750 out->uid = tbuf.sem_perm.uid;
751 out->gid = tbuf.sem_perm.gid;
752 out->mode = tbuf.sem_perm.mode;
758 struct semid_ds tbuf_old;
760 if(copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
763 out->uid = tbuf_old.sem_perm.uid;
764 out->gid = tbuf_old.sem_perm.gid;
765 out->mode = tbuf_old.sem_perm.mode;
774 static int semctl_down(int semid, int semnum, int cmd, int version, union semun arg)
776 struct sem_array *sma;
778 struct sem_setbuf setbuf;
779 struct kern_ipc_perm *ipcp;
782 if(copy_semid_from_user (&setbuf, arg.buf, version))
785 sma = sem_lock(semid);
789 if (sem_checkid(sma,semid)) {
793 ipcp = &sma->sem_perm;
795 if (current->euid != ipcp->cuid &&
796 current->euid != ipcp->uid && !capable(CAP_SYS_ADMIN)) {
801 err = security_sem_semctl(sma, cmd);
811 ipcp->uid = setbuf.uid;
812 ipcp->gid = setbuf.gid;
813 ipcp->mode = (ipcp->mode & ~S_IRWXUGO)
814 | (setbuf.mode & S_IRWXUGO);
815 sma->sem_ctime = get_seconds();
831 asmlinkage long sys_semctl (int semid, int semnum, int cmd, union semun arg)
839 version = ipc_parse_version(&cmd);
845 err = semctl_nolock(semid,semnum,cmd,version,arg);
855 err = semctl_main(semid,semnum,cmd,version,arg);
860 err = semctl_down(semid,semnum,cmd,version,arg);
868 static inline void lock_semundo(void)
870 struct sem_undo_list *undo_list;
872 undo_list = current->sysvsem.undo_list;
873 if ((undo_list != NULL) && (atomic_read(&undo_list->refcnt) != 1))
874 spin_lock(&undo_list->lock);
877 /* This code has an interaction with copy_semundo().
878 * Consider; two tasks are sharing the undo_list. task1
879 * acquires the undo_list lock in lock_semundo(). If task2 now
880 * exits before task1 releases the lock (by calling
881 * unlock_semundo()), then task1 will never call spin_unlock().
882 * This leave the sem_undo_list in a locked state. If task1 now creats task3
883 * and once again shares the sem_undo_list, the sem_undo_list will still be
884 * locked, and future SEM_UNDO operations will deadlock. This case is
885 * dealt with in copy_semundo() by having it reinitialize the spin lock when
886 * the refcnt goes from 1 to 2.
888 static inline void unlock_semundo(void)
890 struct sem_undo_list *undo_list;
892 undo_list = current->sysvsem.undo_list;
893 if ((undo_list != NULL) && (atomic_read(&undo_list->refcnt) != 1))
894 spin_unlock(&undo_list->lock);
898 /* If the task doesn't already have a undo_list, then allocate one
899 * here. We guarantee there is only one thread using this undo list,
900 * and current is THE ONE
902 * If this allocation and assignment succeeds, but later
903 * portions of this code fail, there is no need to free the sem_undo_list.
904 * Just let it stay associated with the task, and it'll be freed later
907 * This can block, so callers must hold no locks.
909 static inline int get_undo_list(struct sem_undo_list **undo_listp)
911 struct sem_undo_list *undo_list;
914 undo_list = current->sysvsem.undo_list;
916 size = sizeof(struct sem_undo_list);
917 undo_list = (struct sem_undo_list *) kmalloc(size, GFP_KERNEL);
918 if (undo_list == NULL)
920 memset(undo_list, 0, size);
921 /* don't initialize unodhd->lock here. It's done
922 * in copy_semundo() instead.
924 atomic_set(&undo_list->refcnt, 1);
925 current->sysvsem.undo_list = undo_list;
927 *undo_listp = undo_list;
931 static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
933 struct sem_undo **last, *un;
935 last = &ulp->proc_list;
951 static struct sem_undo *find_undo(int semid)
953 struct sem_array *sma;
954 struct sem_undo_list *ulp;
955 struct sem_undo *un, *new;
959 error = get_undo_list(&ulp);
961 return ERR_PTR(error);
964 un = lookup_undo(ulp, semid);
966 if (likely(un!=NULL))
969 /* no undo structure around - allocate one. */
970 sma = sem_lock(semid);
971 un = ERR_PTR(-EINVAL);
974 un = ERR_PTR(-EIDRM);
975 if (sem_checkid(sma,semid)) {
979 nsems = sma->sem_nsems;
982 new = (struct sem_undo *) kmalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
984 return ERR_PTR(-ENOMEM);
985 memset(new, 0, sizeof(struct sem_undo) + sizeof(short)*nsems);
986 new->semadj = (short *) &new[1];
990 un = lookup_undo(ulp, semid);
996 error = sem_revalidate(semid, sma, nsems, 0);
1000 un = ERR_PTR(error);
1003 new->proc_next = ulp->proc_list;
1004 ulp->proc_list = new;
1005 new->id_next = sma->undo;
1014 asmlinkage long sys_semtimedop(int semid, struct sembuf __user *tsops,
1015 unsigned nsops, const struct timespec __user *timeout)
1017 int error = -EINVAL;
1018 struct sem_array *sma;
1019 struct sembuf fast_sops[SEMOPM_FAST];
1020 struct sembuf* sops = fast_sops, *sop;
1021 struct sem_undo *un;
1022 int undos = 0, decrease = 0, alter = 0, max;
1023 struct sem_queue queue;
1024 unsigned long jiffies_left = 0;
1026 if (nsops < 1 || semid < 0)
1028 if (nsops > sc_semopm)
1030 if(nsops > SEMOPM_FAST) {
1031 sops = kmalloc(sizeof(*sops)*nsops,GFP_KERNEL);
1035 if (copy_from_user (sops, tsops, nsops * sizeof(*tsops))) {
1040 struct timespec _timeout;
1041 if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1045 if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1046 _timeout.tv_nsec >= 1000000000L) {
1050 jiffies_left = timespec_to_jiffies(&_timeout);
1053 for (sop = sops; sop < sops + nsops; sop++) {
1054 if (sop->sem_num >= max)
1056 if (sop->sem_flg & SEM_UNDO)
1058 if (sop->sem_op < 0)
1060 if (sop->sem_op > 0)
1067 un = find_undo(semid);
1069 error = PTR_ERR(un);
1075 sma = sem_lock(semid);
1080 if (sem_checkid(sma,semid))
1081 goto out_unlock_free;
1083 * semid identifies are not unique - find_undo may have
1084 * allocated an undo structure, it was invalidated by an RMID
1085 * and now a new array with received the same id. Check and retry.
1087 if (un && un->semid == -1) {
1092 if (max >= sma->sem_nsems)
1093 goto out_unlock_free;
1096 if (ipcperms(&sma->sem_perm, alter ? S_IWUGO : S_IRUGO))
1097 goto out_unlock_free;
1099 error = security_sem_semop(sma, sops, nsops, alter);
1101 goto out_unlock_free;
1103 error = try_atomic_semop (sma, sops, nsops, un, current->tgid);
1107 /* We need to sleep on this operation, so we put the current
1108 * task into the pending queue and go to sleep.
1113 queue.nsops = nsops;
1115 queue.pid = current->tgid;
1118 append_to_queue(sma ,&queue);
1120 prepend_to_queue(sma ,&queue);
1122 queue.status = -EINTR;
1123 queue.sleeper = current;
1124 current->state = TASK_INTERRUPTIBLE;
1128 jiffies_left = schedule_timeout(jiffies_left);
1132 error = queue.status;
1133 while(unlikely(error == IN_WAKEUP)) {
1135 error = queue.status;
1138 if (error != -EINTR) {
1139 /* fast path: update_queue already obtained all requested
1144 sma = sem_lock(semid);
1146 if(queue.prev != NULL)
1153 * If queue.status != -EINTR we are woken up by another process
1155 error = queue.status;
1156 if (error != -EINTR) {
1157 goto out_unlock_free;
1161 * If an interrupt occurred we have to clean up the queue
1163 if (timeout && jiffies_left == 0)
1165 remove_from_queue(sma,&queue);
1166 goto out_unlock_free;
1174 if(sops != fast_sops)
1179 asmlinkage long sys_semop (int semid, struct sembuf __user *tsops, unsigned nsops)
1181 return sys_semtimedop(semid, tsops, nsops, NULL);
1184 /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
1185 * parent and child tasks.
1187 * See the notes above unlock_semundo() regarding the spin_lock_init()
1188 * in this code. Initialize the undo_list->lock here instead of get_undo_list()
1189 * because of the reasoning in the comment above unlock_semundo.
1192 int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
1194 struct sem_undo_list *undo_list;
1197 if (clone_flags & CLONE_SYSVSEM) {
1198 error = get_undo_list(&undo_list);
1201 if (atomic_read(&undo_list->refcnt) == 1)
1202 spin_lock_init(&undo_list->lock);
1203 atomic_inc(&undo_list->refcnt);
1204 tsk->sysvsem.undo_list = undo_list;
1206 tsk->sysvsem.undo_list = NULL;
1212 * add semadj values to semaphores, free undo structures.
1213 * undo structures are not freed when semaphore arrays are destroyed
1214 * so some of them may be out of date.
1215 * IMPLEMENTATION NOTE: There is some confusion over whether the
1216 * set of adjustments that needs to be done should be done in an atomic
1217 * manner or not. That is, if we are attempting to decrement the semval
1218 * should we queue up and wait until we can do so legally?
1219 * The original implementation attempted to do this (queue and wait).
1220 * The current implementation does not do so. The POSIX standard
1221 * and SVID should be consulted to determine what behavior is mandated.
1223 void exit_sem(struct task_struct *tsk)
1225 struct sem_undo_list *undo_list;
1226 struct sem_undo *u, **up;
1228 undo_list = tsk->sysvsem.undo_list;
1232 if (!atomic_dec_and_test(&undo_list->refcnt))
1235 /* There's no need to hold the semundo list lock, as current
1236 * is the last task exiting for this undo list.
1238 for (up = &undo_list->proc_list; (u = *up); *up = u->proc_next, kfree(u)) {
1239 struct sem_array *sma;
1241 struct sem_undo *un, **unp;
1248 sma = sem_lock(semid);
1255 BUG_ON(sem_checkid(sma,u->semid));
1257 /* remove u from the sma->undo list */
1258 for (unp = &sma->undo; (un = *unp); unp = &un->id_next) {
1262 printk ("exit_sem undo list error id=%d\n", u->semid);
1266 /* perform adjustments registered in u */
1267 nsems = sma->sem_nsems;
1268 for (i = 0; i < nsems; i++) {
1269 struct sem * sem = &sma->sem_base[i];
1271 sem->semval += u->semadj[i];
1272 if (sem->semval < 0)
1273 sem->semval = 0; /* shouldn't happen */
1274 sem->sempid = current->tgid;
1277 sma->sem_otime = get_seconds();
1278 /* maybe some queued-up processes were waiting for this */
1286 #ifdef CONFIG_PROC_FS
1287 static int sysvipc_sem_read_proc(char *buffer, char **start, off_t offset, int length, int *eof, void *data)
1293 len += sprintf(buffer, " key semid perms nsems uid gid cuid cgid otime ctime\n");
1296 for(i = 0; i <= sem_ids.max_id; i++) {
1297 struct sem_array *sma;
1300 len += sprintf(buffer + len, "%10d %10d %4o %10lu %5u %5u %5u %5u %10lu %10lu\n",
1302 sem_buildid(i,sma->sem_perm.seq),
1318 if(pos > offset + length)
1325 *start = buffer + (offset - begin);
1326 len -= (offset - begin);