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 <linux/syscalls.h>
75 #include <linux/audit.h>
76 #include <linux/capability.h>
77 #include <linux/seq_file.h>
78 #include <asm/uaccess.h>
82 #define sem_lock(id) ((struct sem_array*)ipc_lock(&sem_ids,id))
83 #define sem_unlock(sma) ipc_unlock(&(sma)->sem_perm)
84 #define sem_rmid(id) ((struct sem_array*)ipc_rmid(&sem_ids,id))
85 #define sem_checkid(sma, semid) \
86 ipc_checkid(&sem_ids,&sma->sem_perm,semid)
87 #define sem_buildid(id, seq) \
88 ipc_buildid(&sem_ids, id, seq)
89 static struct ipc_ids sem_ids;
91 static int newary (key_t, int, int);
92 static void freeary (struct sem_array *sma, int id);
94 static int sysvipc_sem_proc_show(struct seq_file *s, void *it);
97 #define SEMMSL_FAST 256 /* 512 bytes on stack */
98 #define SEMOPM_FAST 64 /* ~ 372 bytes on stack */
101 * linked list protection:
103 * sem_array.sem_pending{,last},
104 * sem_array.sem_undo: sem_lock() for read/write
105 * sem_undo.proc_next: only "current" is allowed to read/write that field.
109 int sem_ctls[4] = {SEMMSL, SEMMNS, SEMOPM, SEMMNI};
110 #define sc_semmsl (sem_ctls[0])
111 #define sc_semmns (sem_ctls[1])
112 #define sc_semopm (sem_ctls[2])
113 #define sc_semmni (sem_ctls[3])
115 static int used_sems;
117 void __init sem_init (void)
120 ipc_init_ids(&sem_ids,sc_semmni);
121 ipc_init_proc_interface("sysvipc/sem",
122 " key semid perms nsems uid gid cuid cgid otime ctime\n",
124 sysvipc_sem_proc_show);
128 * Lockless wakeup algorithm:
129 * Without the check/retry algorithm a lockless wakeup is possible:
130 * - queue.status is initialized to -EINTR before blocking.
131 * - wakeup is performed by
132 * * unlinking the queue entry from sma->sem_pending
133 * * setting queue.status to IN_WAKEUP
134 * This is the notification for the blocked thread that a
135 * result value is imminent.
136 * * call wake_up_process
137 * * set queue.status to the final value.
138 * - the previously blocked thread checks queue.status:
139 * * if it's IN_WAKEUP, then it must wait until the value changes
140 * * if it's not -EINTR, then the operation was completed by
141 * update_queue. semtimedop can return queue.status without
142 * performing any operation on the semaphore array.
143 * * otherwise it must acquire the spinlock and check what's up.
145 * The two-stage algorithm is necessary to protect against the following
147 * - if queue.status is set after wake_up_process, then the woken up idle
148 * thread could race forward and try (and fail) to acquire sma->lock
149 * before update_queue had a chance to set queue.status
150 * - if queue.status is written before wake_up_process and if the
151 * blocked process is woken up by a signal between writing
152 * queue.status and the wake_up_process, then the woken up
153 * process could return from semtimedop and die by calling
154 * sys_exit before wake_up_process is called. Then wake_up_process
155 * will oops, because the task structure is already invalid.
156 * (yes, this happened on s390 with sysv msg).
161 static int newary (key_t key, int nsems, int semflg)
165 struct sem_array *sma;
170 if (used_sems + nsems > sc_semmns)
173 size = sizeof (*sma) + nsems * sizeof (struct sem);
174 sma = ipc_rcu_alloc(size);
178 memset (sma, 0, size);
180 sma->sem_perm.mode = (semflg & S_IRWXUGO);
181 sma->sem_perm.key = key;
183 sma->sem_perm.security = NULL;
184 retval = security_sem_alloc(sma);
190 id = ipc_addid(&sem_ids, &sma->sem_perm, sc_semmni);
192 security_sem_free(sma);
198 sma->sem_id = sem_buildid(id, sma->sem_perm.seq);
199 sma->sem_base = (struct sem *) &sma[1];
200 /* sma->sem_pending = NULL; */
201 sma->sem_pending_last = &sma->sem_pending;
202 /* sma->undo = NULL; */
203 sma->sem_nsems = nsems;
204 sma->sem_ctime = get_seconds();
210 asmlinkage long sys_semget (key_t key, int nsems, int semflg)
212 int id, err = -EINVAL;
213 struct sem_array *sma;
215 if (nsems < 0 || nsems > sc_semmsl)
219 if (key == IPC_PRIVATE) {
220 err = newary(key, nsems, semflg);
221 } else if ((id = ipc_findkey(&sem_ids, key)) == -1) { /* key not used */
222 if (!(semflg & IPC_CREAT))
225 err = newary(key, nsems, semflg);
226 } else if (semflg & IPC_CREAT && semflg & IPC_EXCL) {
232 if (nsems > sma->sem_nsems)
234 else if (ipcperms(&sma->sem_perm, semflg))
237 int semid = sem_buildid(id, sma->sem_perm.seq);
238 err = security_sem_associate(sma, semflg);
249 /* Manage the doubly linked list sma->sem_pending as a FIFO:
250 * insert new queue elements at the tail sma->sem_pending_last.
252 static inline void append_to_queue (struct sem_array * sma,
253 struct sem_queue * q)
255 *(q->prev = sma->sem_pending_last) = q;
256 *(sma->sem_pending_last = &q->next) = NULL;
259 static inline void prepend_to_queue (struct sem_array * sma,
260 struct sem_queue * q)
262 q->next = sma->sem_pending;
263 *(q->prev = &sma->sem_pending) = q;
265 q->next->prev = &q->next;
266 else /* sma->sem_pending_last == &sma->sem_pending */
267 sma->sem_pending_last = &q->next;
270 static inline void remove_from_queue (struct sem_array * sma,
271 struct sem_queue * q)
273 *(q->prev) = q->next;
275 q->next->prev = q->prev;
276 else /* sma->sem_pending_last == &q->next */
277 sma->sem_pending_last = q->prev;
278 q->prev = NULL; /* mark as removed */
282 * Determine whether a sequence of semaphore operations would succeed
283 * all at once. Return 0 if yes, 1 if need to sleep, else return error code.
286 static int try_atomic_semop (struct sem_array * sma, struct sembuf * sops,
287 int nsops, struct sem_undo *un, int pid)
293 for (sop = sops; sop < sops + nsops; sop++) {
294 curr = sma->sem_base + sop->sem_num;
295 sem_op = sop->sem_op;
296 result = curr->semval;
298 if (!sem_op && result)
306 if (sop->sem_flg & SEM_UNDO) {
307 int undo = un->semadj[sop->sem_num] - sem_op;
309 * Exceeding the undo range is an error.
311 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
314 curr->semval = result;
318 while (sop >= sops) {
319 sma->sem_base[sop->sem_num].sempid = pid;
320 if (sop->sem_flg & SEM_UNDO)
321 un->semadj[sop->sem_num] -= sop->sem_op;
325 sma->sem_otime = get_seconds();
333 if (sop->sem_flg & IPC_NOWAIT)
340 while (sop >= sops) {
341 sma->sem_base[sop->sem_num].semval -= sop->sem_op;
348 /* Go through the pending queue for the indicated semaphore
349 * looking for tasks that can be completed.
351 static void update_queue (struct sem_array * sma)
354 struct sem_queue * q;
356 q = sma->sem_pending;
358 error = try_atomic_semop(sma, q->sops, q->nsops,
361 /* Does q->sleeper still need to sleep? */
364 remove_from_queue(sma,q);
365 q->status = IN_WAKEUP;
367 * Continue scanning. The next operation
368 * that must be checked depends on the type of the
369 * completed operation:
370 * - if the operation modified the array, then
371 * restart from the head of the queue and
372 * check for threads that might be waiting
373 * for semaphore values to become 0.
374 * - if the operation didn't modify the array,
375 * then just continue.
378 n = sma->sem_pending;
381 wake_up_process(q->sleeper);
382 /* hands-off: q will disappear immediately after
394 /* The following counts are associated to each semaphore:
395 * semncnt number of tasks waiting on semval being nonzero
396 * semzcnt number of tasks waiting on semval being zero
397 * This model assumes that a task waits on exactly one semaphore.
398 * Since semaphore operations are to be performed atomically, tasks actually
399 * wait on a whole sequence of semaphores simultaneously.
400 * The counts we return here are a rough approximation, but still
401 * warrant that semncnt+semzcnt>0 if the task is on the pending queue.
403 static int count_semncnt (struct sem_array * sma, ushort semnum)
406 struct sem_queue * q;
409 for (q = sma->sem_pending; q; q = q->next) {
410 struct sembuf * sops = q->sops;
411 int nsops = q->nsops;
413 for (i = 0; i < nsops; i++)
414 if (sops[i].sem_num == semnum
415 && (sops[i].sem_op < 0)
416 && !(sops[i].sem_flg & IPC_NOWAIT))
421 static int count_semzcnt (struct sem_array * sma, ushort semnum)
424 struct sem_queue * q;
427 for (q = sma->sem_pending; q; q = q->next) {
428 struct sembuf * sops = q->sops;
429 int nsops = q->nsops;
431 for (i = 0; i < nsops; i++)
432 if (sops[i].sem_num == semnum
433 && (sops[i].sem_op == 0)
434 && !(sops[i].sem_flg & IPC_NOWAIT))
440 /* Free a semaphore set. freeary() is called with sem_ids.sem down and
441 * the spinlock for this semaphore set hold. sem_ids.sem remains locked
444 static void freeary (struct sem_array *sma, int id)
450 /* Invalidate the existing undo structures for this semaphore set.
451 * (They will be freed without any further action in exit_sem()
452 * or during the next semop.)
454 for (un = sma->undo; un; un = un->id_next)
457 /* Wake up all pending processes and let them fail with EIDRM. */
458 q = sma->sem_pending;
461 /* lazy remove_from_queue: we are killing the whole queue */
464 q->status = IN_WAKEUP;
465 wake_up_process(q->sleeper); /* doesn't sleep */
467 q->status = -EIDRM; /* hands-off q */
471 /* Remove the semaphore set from the ID array*/
475 used_sems -= sma->sem_nsems;
476 size = sizeof (*sma) + sma->sem_nsems * sizeof (struct sem);
477 security_sem_free(sma);
481 static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
485 return copy_to_user(buf, in, sizeof(*in));
490 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
492 out.sem_otime = in->sem_otime;
493 out.sem_ctime = in->sem_ctime;
494 out.sem_nsems = in->sem_nsems;
496 return copy_to_user(buf, &out, sizeof(out));
503 static int semctl_nolock(int semid, int semnum, int cmd, int version, union semun arg)
506 struct sem_array *sma;
512 struct seminfo seminfo;
515 err = security_sem_semctl(NULL, cmd);
519 memset(&seminfo,0,sizeof(seminfo));
520 seminfo.semmni = sc_semmni;
521 seminfo.semmns = sc_semmns;
522 seminfo.semmsl = sc_semmsl;
523 seminfo.semopm = sc_semopm;
524 seminfo.semvmx = SEMVMX;
525 seminfo.semmnu = SEMMNU;
526 seminfo.semmap = SEMMAP;
527 seminfo.semume = SEMUME;
529 if (cmd == SEM_INFO) {
530 seminfo.semusz = sem_ids.in_use;
531 seminfo.semaem = used_sems;
533 seminfo.semusz = SEMUSZ;
534 seminfo.semaem = SEMAEM;
536 max_id = sem_ids.max_id;
538 if (copy_to_user (arg.__buf, &seminfo, sizeof(struct seminfo)))
540 return (max_id < 0) ? 0: max_id;
544 struct semid64_ds tbuf;
547 if(semid >= sem_ids.entries->size)
550 memset(&tbuf,0,sizeof(tbuf));
552 sma = sem_lock(semid);
557 if (ipcperms (&sma->sem_perm, S_IRUGO))
560 err = security_sem_semctl(sma, cmd);
564 id = sem_buildid(semid, sma->sem_perm.seq);
566 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
567 tbuf.sem_otime = sma->sem_otime;
568 tbuf.sem_ctime = sma->sem_ctime;
569 tbuf.sem_nsems = sma->sem_nsems;
571 if (copy_semid_to_user (arg.buf, &tbuf, version))
584 static int semctl_main(int semid, int semnum, int cmd, int version, union semun arg)
586 struct sem_array *sma;
589 ushort fast_sem_io[SEMMSL_FAST];
590 ushort* sem_io = fast_sem_io;
593 sma = sem_lock(semid);
597 nsems = sma->sem_nsems;
600 if (sem_checkid(sma,semid))
604 if (ipcperms (&sma->sem_perm, (cmd==SETVAL||cmd==SETALL)?S_IWUGO:S_IRUGO))
607 err = security_sem_semctl(sma, cmd);
615 ushort __user *array = arg.array;
618 if(nsems > SEMMSL_FAST) {
622 sem_io = ipc_alloc(sizeof(ushort)*nsems);
624 ipc_lock_by_ptr(&sma->sem_perm);
630 ipc_lock_by_ptr(&sma->sem_perm);
632 if (sma->sem_perm.deleted) {
639 for (i = 0; i < sma->sem_nsems; i++)
640 sem_io[i] = sma->sem_base[i].semval;
643 if(copy_to_user(array, sem_io, nsems*sizeof(ushort)))
655 if(nsems > SEMMSL_FAST) {
656 sem_io = ipc_alloc(sizeof(ushort)*nsems);
658 ipc_lock_by_ptr(&sma->sem_perm);
665 if (copy_from_user (sem_io, arg.array, nsems*sizeof(ushort))) {
666 ipc_lock_by_ptr(&sma->sem_perm);
673 for (i = 0; i < nsems; i++) {
674 if (sem_io[i] > SEMVMX) {
675 ipc_lock_by_ptr(&sma->sem_perm);
682 ipc_lock_by_ptr(&sma->sem_perm);
684 if (sma->sem_perm.deleted) {
690 for (i = 0; i < nsems; i++)
691 sma->sem_base[i].semval = sem_io[i];
692 for (un = sma->undo; un; un = un->id_next)
693 for (i = 0; i < nsems; i++)
695 sma->sem_ctime = get_seconds();
696 /* maybe some queued-up processes were waiting for this */
703 struct semid64_ds tbuf;
704 memset(&tbuf,0,sizeof(tbuf));
705 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
706 tbuf.sem_otime = sma->sem_otime;
707 tbuf.sem_ctime = sma->sem_ctime;
708 tbuf.sem_nsems = sma->sem_nsems;
710 if (copy_semid_to_user (arg.buf, &tbuf, version))
714 /* GETVAL, GETPID, GETNCTN, GETZCNT, SETVAL: fall-through */
717 if(semnum < 0 || semnum >= nsems)
720 curr = &sma->sem_base[semnum];
730 err = count_semncnt(sma,semnum);
733 err = count_semzcnt(sma,semnum);
740 if (val > SEMVMX || val < 0)
743 for (un = sma->undo; un; un = un->id_next)
744 un->semadj[semnum] = 0;
746 curr->sempid = current->tgid;
747 sma->sem_ctime = get_seconds();
748 /* maybe some queued-up processes were waiting for this */
757 if(sem_io != fast_sem_io)
758 ipc_free(sem_io, sizeof(ushort)*nsems);
768 static inline unsigned long copy_semid_from_user(struct sem_setbuf *out, void __user *buf, int version)
773 struct semid64_ds tbuf;
775 if(copy_from_user(&tbuf, buf, sizeof(tbuf)))
778 out->uid = tbuf.sem_perm.uid;
779 out->gid = tbuf.sem_perm.gid;
780 out->mode = tbuf.sem_perm.mode;
786 struct semid_ds tbuf_old;
788 if(copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
791 out->uid = tbuf_old.sem_perm.uid;
792 out->gid = tbuf_old.sem_perm.gid;
793 out->mode = tbuf_old.sem_perm.mode;
802 static int semctl_down(int semid, int semnum, int cmd, int version, union semun arg)
804 struct sem_array *sma;
806 struct sem_setbuf setbuf;
807 struct kern_ipc_perm *ipcp;
810 if(copy_semid_from_user (&setbuf, arg.buf, version))
812 if ((err = audit_ipc_perms(0, setbuf.uid, setbuf.gid, setbuf.mode)))
815 sma = sem_lock(semid);
819 if (sem_checkid(sma,semid)) {
823 ipcp = &sma->sem_perm;
825 if (current->euid != ipcp->cuid &&
826 current->euid != ipcp->uid && !capable(CAP_SYS_ADMIN)) {
831 err = security_sem_semctl(sma, cmd);
841 ipcp->uid = setbuf.uid;
842 ipcp->gid = setbuf.gid;
843 ipcp->mode = (ipcp->mode & ~S_IRWXUGO)
844 | (setbuf.mode & S_IRWXUGO);
845 sma->sem_ctime = get_seconds();
861 asmlinkage long sys_semctl (int semid, int semnum, int cmd, union semun arg)
869 version = ipc_parse_version(&cmd);
875 err = semctl_nolock(semid,semnum,cmd,version,arg);
885 err = semctl_main(semid,semnum,cmd,version,arg);
890 err = semctl_down(semid,semnum,cmd,version,arg);
898 static inline void lock_semundo(void)
900 struct sem_undo_list *undo_list;
902 undo_list = current->sysvsem.undo_list;
904 spin_lock(&undo_list->lock);
907 /* This code has an interaction with copy_semundo().
908 * Consider; two tasks are sharing the undo_list. task1
909 * acquires the undo_list lock in lock_semundo(). If task2 now
910 * exits before task1 releases the lock (by calling
911 * unlock_semundo()), then task1 will never call spin_unlock().
912 * This leave the sem_undo_list in a locked state. If task1 now creats task3
913 * and once again shares the sem_undo_list, the sem_undo_list will still be
914 * locked, and future SEM_UNDO operations will deadlock. This case is
915 * dealt with in copy_semundo() by having it reinitialize the spin lock when
916 * the refcnt goes from 1 to 2.
918 static inline void unlock_semundo(void)
920 struct sem_undo_list *undo_list;
922 undo_list = current->sysvsem.undo_list;
924 spin_unlock(&undo_list->lock);
928 /* If the task doesn't already have a undo_list, then allocate one
929 * here. We guarantee there is only one thread using this undo list,
930 * and current is THE ONE
932 * If this allocation and assignment succeeds, but later
933 * portions of this code fail, there is no need to free the sem_undo_list.
934 * Just let it stay associated with the task, and it'll be freed later
937 * This can block, so callers must hold no locks.
939 static inline int get_undo_list(struct sem_undo_list **undo_listp)
941 struct sem_undo_list *undo_list;
944 undo_list = current->sysvsem.undo_list;
946 size = sizeof(struct sem_undo_list);
947 undo_list = (struct sem_undo_list *) kmalloc(size, GFP_KERNEL);
948 if (undo_list == NULL)
950 memset(undo_list, 0, size);
951 spin_lock_init(&undo_list->lock);
952 atomic_set(&undo_list->refcnt, 1);
953 current->sysvsem.undo_list = undo_list;
955 *undo_listp = undo_list;
959 static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
961 struct sem_undo **last, *un;
963 last = &ulp->proc_list;
979 static struct sem_undo *find_undo(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 un = lookup_undo(ulp, semid);
994 if (likely(un!=NULL))
997 /* no undo structure around - allocate one. */
998 sma = sem_lock(semid);
999 un = ERR_PTR(-EINVAL);
1002 un = ERR_PTR(-EIDRM);
1003 if (sem_checkid(sma,semid)) {
1007 nsems = sma->sem_nsems;
1008 ipc_rcu_getref(sma);
1011 new = (struct sem_undo *) kmalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1013 ipc_lock_by_ptr(&sma->sem_perm);
1014 ipc_rcu_putref(sma);
1016 return ERR_PTR(-ENOMEM);
1018 memset(new, 0, sizeof(struct sem_undo) + sizeof(short)*nsems);
1019 new->semadj = (short *) &new[1];
1023 un = lookup_undo(ulp, semid);
1027 ipc_lock_by_ptr(&sma->sem_perm);
1028 ipc_rcu_putref(sma);
1032 ipc_lock_by_ptr(&sma->sem_perm);
1033 ipc_rcu_putref(sma);
1034 if (sma->sem_perm.deleted) {
1038 un = ERR_PTR(-EIDRM);
1041 new->proc_next = ulp->proc_list;
1042 ulp->proc_list = new;
1043 new->id_next = sma->undo;
1052 asmlinkage long sys_semtimedop(int semid, struct sembuf __user *tsops,
1053 unsigned nsops, const struct timespec __user *timeout)
1055 int error = -EINVAL;
1056 struct sem_array *sma;
1057 struct sembuf fast_sops[SEMOPM_FAST];
1058 struct sembuf* sops = fast_sops, *sop;
1059 struct sem_undo *un;
1060 int undos = 0, alter = 0, max;
1061 struct sem_queue queue;
1062 unsigned long jiffies_left = 0;
1064 if (nsops < 1 || semid < 0)
1066 if (nsops > sc_semopm)
1068 if(nsops > SEMOPM_FAST) {
1069 sops = kmalloc(sizeof(*sops)*nsops,GFP_KERNEL);
1073 if (copy_from_user (sops, tsops, nsops * sizeof(*tsops))) {
1078 struct timespec _timeout;
1079 if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1083 if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1084 _timeout.tv_nsec >= 1000000000L) {
1088 jiffies_left = timespec_to_jiffies(&_timeout);
1091 for (sop = sops; sop < sops + nsops; sop++) {
1092 if (sop->sem_num >= max)
1094 if (sop->sem_flg & SEM_UNDO)
1096 if (sop->sem_op != 0)
1102 un = find_undo(semid);
1104 error = PTR_ERR(un);
1110 sma = sem_lock(semid);
1115 if (sem_checkid(sma,semid))
1116 goto out_unlock_free;
1118 * semid identifies are not unique - find_undo may have
1119 * allocated an undo structure, it was invalidated by an RMID
1120 * and now a new array with received the same id. Check and retry.
1122 if (un && un->semid == -1) {
1127 if (max >= sma->sem_nsems)
1128 goto out_unlock_free;
1131 if (ipcperms(&sma->sem_perm, alter ? S_IWUGO : S_IRUGO))
1132 goto out_unlock_free;
1134 error = security_sem_semop(sma, sops, nsops, alter);
1136 goto out_unlock_free;
1138 error = try_atomic_semop (sma, sops, nsops, un, current->tgid);
1140 if (alter && error == 0)
1142 goto out_unlock_free;
1145 /* We need to sleep on this operation, so we put the current
1146 * task into the pending queue and go to sleep.
1151 queue.nsops = nsops;
1153 queue.pid = current->tgid;
1155 queue.alter = alter;
1157 append_to_queue(sma ,&queue);
1159 prepend_to_queue(sma ,&queue);
1161 queue.status = -EINTR;
1162 queue.sleeper = current;
1163 current->state = TASK_INTERRUPTIBLE;
1167 jiffies_left = schedule_timeout(jiffies_left);
1171 error = queue.status;
1172 while(unlikely(error == IN_WAKEUP)) {
1174 error = queue.status;
1177 if (error != -EINTR) {
1178 /* fast path: update_queue already obtained all requested
1183 sma = sem_lock(semid);
1185 if(queue.prev != NULL)
1192 * If queue.status != -EINTR we are woken up by another process
1194 error = queue.status;
1195 if (error != -EINTR) {
1196 goto out_unlock_free;
1200 * If an interrupt occurred we have to clean up the queue
1202 if (timeout && jiffies_left == 0)
1204 remove_from_queue(sma,&queue);
1205 goto out_unlock_free;
1210 if(sops != fast_sops)
1215 asmlinkage long sys_semop (int semid, struct sembuf __user *tsops, unsigned nsops)
1217 return sys_semtimedop(semid, tsops, nsops, NULL);
1220 /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
1221 * parent and child tasks.
1223 * See the notes above unlock_semundo() regarding the spin_lock_init()
1224 * in this code. Initialize the undo_list->lock here instead of get_undo_list()
1225 * because of the reasoning in the comment above unlock_semundo.
1228 int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
1230 struct sem_undo_list *undo_list;
1233 if (clone_flags & CLONE_SYSVSEM) {
1234 error = get_undo_list(&undo_list);
1237 atomic_inc(&undo_list->refcnt);
1238 tsk->sysvsem.undo_list = undo_list;
1240 tsk->sysvsem.undo_list = NULL;
1246 * add semadj values to semaphores, free undo structures.
1247 * undo structures are not freed when semaphore arrays are destroyed
1248 * so some of them may be out of date.
1249 * IMPLEMENTATION NOTE: There is some confusion over whether the
1250 * set of adjustments that needs to be done should be done in an atomic
1251 * manner or not. That is, if we are attempting to decrement the semval
1252 * should we queue up and wait until we can do so legally?
1253 * The original implementation attempted to do this (queue and wait).
1254 * The current implementation does not do so. The POSIX standard
1255 * and SVID should be consulted to determine what behavior is mandated.
1257 void exit_sem(struct task_struct *tsk)
1259 struct sem_undo_list *undo_list;
1260 struct sem_undo *u, **up;
1262 undo_list = tsk->sysvsem.undo_list;
1266 if (!atomic_dec_and_test(&undo_list->refcnt))
1269 /* There's no need to hold the semundo list lock, as current
1270 * is the last task exiting for this undo list.
1272 for (up = &undo_list->proc_list; (u = *up); *up = u->proc_next, kfree(u)) {
1273 struct sem_array *sma;
1275 struct sem_undo *un, **unp;
1282 sma = sem_lock(semid);
1289 BUG_ON(sem_checkid(sma,u->semid));
1291 /* remove u from the sma->undo list */
1292 for (unp = &sma->undo; (un = *unp); unp = &un->id_next) {
1296 printk ("exit_sem undo list error id=%d\n", u->semid);
1300 /* perform adjustments registered in u */
1301 nsems = sma->sem_nsems;
1302 for (i = 0; i < nsems; i++) {
1303 struct sem * sem = &sma->sem_base[i];
1305 sem->semval += u->semadj[i];
1307 * Range checks of the new semaphore value,
1308 * not defined by sus:
1309 * - Some unices ignore the undo entirely
1310 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
1311 * - some cap the value (e.g. FreeBSD caps
1312 * at 0, but doesn't enforce SEMVMX)
1314 * Linux caps the semaphore value, both at 0
1317 * Manfred <manfred@colorfullife.com>
1319 if (sem->semval < 0)
1321 if (sem->semval > SEMVMX)
1322 sem->semval = SEMVMX;
1323 sem->sempid = current->tgid;
1326 sma->sem_otime = get_seconds();
1327 /* maybe some queued-up processes were waiting for this */
1335 #ifdef CONFIG_PROC_FS
1336 static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
1338 struct sem_array *sma = it;
1340 return seq_printf(s,
1341 "%10d %10d %4o %10lu %5u %5u %5u %5u %10lu %10lu\n",