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/mutex.h>
84 #include <linux/nsproxy.h>
86 #include <asm/uaccess.h>
89 #define sem_ids(ns) (*((ns)->ids[IPC_SEM_IDS]))
91 #define sem_lock(ns, id) ((struct sem_array*)ipc_lock(&sem_ids(ns), id))
92 #define sem_unlock(sma) ipc_unlock(&(sma)->sem_perm)
93 #define sem_checkid(ns, sma, semid) \
94 ipc_checkid(&sem_ids(ns),&sma->sem_perm,semid)
95 #define sem_buildid(ns, id, seq) \
96 ipc_buildid(&sem_ids(ns), id, seq)
98 static struct ipc_ids init_sem_ids;
100 static int newary(struct ipc_namespace *, struct ipc_params *);
101 static void freeary(struct ipc_namespace *, struct sem_array *);
102 #ifdef CONFIG_PROC_FS
103 static int sysvipc_sem_proc_show(struct seq_file *s, void *it);
106 #define SEMMSL_FAST 256 /* 512 bytes on stack */
107 #define SEMOPM_FAST 64 /* ~ 372 bytes on stack */
110 * linked list protection:
112 * sem_array.sem_pending{,last},
113 * sem_array.sem_undo: sem_lock() for read/write
114 * sem_undo.proc_next: only "current" is allowed to read/write that field.
118 #define sc_semmsl sem_ctls[0]
119 #define sc_semmns sem_ctls[1]
120 #define sc_semopm sem_ctls[2]
121 #define sc_semmni sem_ctls[3]
123 static void __sem_init_ns(struct ipc_namespace *ns, struct ipc_ids *ids)
125 ns->ids[IPC_SEM_IDS] = ids;
126 ns->sc_semmsl = SEMMSL;
127 ns->sc_semmns = SEMMNS;
128 ns->sc_semopm = SEMOPM;
129 ns->sc_semmni = SEMMNI;
134 int sem_init_ns(struct ipc_namespace *ns)
138 ids = kmalloc(sizeof(struct ipc_ids), GFP_KERNEL);
142 __sem_init_ns(ns, ids);
146 void sem_exit_ns(struct ipc_namespace *ns)
148 struct sem_array *sma;
152 mutex_lock(&sem_ids(ns).mutex);
154 in_use = sem_ids(ns).in_use;
156 for (total = 0, next_id = 0; total < in_use; next_id++) {
157 sma = idr_find(&sem_ids(ns).ipcs_idr, next_id);
160 ipc_lock_by_ptr(&sma->sem_perm);
164 mutex_unlock(&sem_ids(ns).mutex);
166 kfree(ns->ids[IPC_SEM_IDS]);
167 ns->ids[IPC_SEM_IDS] = NULL;
170 void __init sem_init (void)
172 __sem_init_ns(&init_ipc_ns, &init_sem_ids);
173 ipc_init_proc_interface("sysvipc/sem",
174 " key semid perms nsems uid gid cuid cgid otime ctime\n",
175 IPC_SEM_IDS, sysvipc_sem_proc_show);
178 static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
180 ipc_rmid(&sem_ids(ns), &s->sem_perm);
184 * Lockless wakeup algorithm:
185 * Without the check/retry algorithm a lockless wakeup is possible:
186 * - queue.status is initialized to -EINTR before blocking.
187 * - wakeup is performed by
188 * * unlinking the queue entry from sma->sem_pending
189 * * setting queue.status to IN_WAKEUP
190 * This is the notification for the blocked thread that a
191 * result value is imminent.
192 * * call wake_up_process
193 * * set queue.status to the final value.
194 * - the previously blocked thread checks queue.status:
195 * * if it's IN_WAKEUP, then it must wait until the value changes
196 * * if it's not -EINTR, then the operation was completed by
197 * update_queue. semtimedop can return queue.status without
198 * performing any operation on the sem array.
199 * * otherwise it must acquire the spinlock and check what's up.
201 * The two-stage algorithm is necessary to protect against the following
203 * - if queue.status is set after wake_up_process, then the woken up idle
204 * thread could race forward and try (and fail) to acquire sma->lock
205 * before update_queue had a chance to set queue.status
206 * - if queue.status is written before wake_up_process and if the
207 * blocked process is woken up by a signal between writing
208 * queue.status and the wake_up_process, then the woken up
209 * process could return from semtimedop and die by calling
210 * sys_exit before wake_up_process is called. Then wake_up_process
211 * will oops, because the task structure is already invalid.
212 * (yes, this happened on s390 with sysv msg).
217 static int newary(struct ipc_namespace *ns, struct ipc_params *params)
221 struct sem_array *sma;
223 key_t key = params->key;
224 int nsems = params->u.nsems;
225 int semflg = params->flg;
229 if (ns->used_sems + nsems > ns->sc_semmns)
232 size = sizeof (*sma) + nsems * sizeof (struct sem);
233 sma = ipc_rcu_alloc(size);
237 memset (sma, 0, size);
239 sma->sem_perm.mode = (semflg & S_IRWXUGO);
240 sma->sem_perm.key = key;
242 sma->sem_perm.security = NULL;
243 retval = security_sem_alloc(sma);
249 id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
251 security_sem_free(sma);
255 ns->used_sems += nsems;
257 sma->sem_perm.id = sem_buildid(ns, id, sma->sem_perm.seq);
258 sma->sem_base = (struct sem *) &sma[1];
259 /* sma->sem_pending = NULL; */
260 sma->sem_pending_last = &sma->sem_pending;
261 /* sma->undo = NULL; */
262 sma->sem_nsems = nsems;
263 sma->sem_ctime = get_seconds();
266 return sma->sem_perm.id;
270 static inline int sem_security(void *sma, int semflg)
272 return security_sem_associate((struct sem_array *) sma, semflg);
275 static inline int sem_more_checks(void *sma, struct ipc_params *params)
277 if (params->u.nsems > ((struct sem_array *)sma)->sem_nsems)
283 asmlinkage long sys_semget(key_t key, int nsems, int semflg)
285 struct ipc_namespace *ns;
286 struct ipc_ops sem_ops;
287 struct ipc_params sem_params;
289 ns = current->nsproxy->ipc_ns;
291 if (nsems < 0 || nsems > ns->sc_semmsl)
294 sem_ops.getnew = newary;
295 sem_ops.associate = sem_security;
296 sem_ops.more_checks = sem_more_checks;
298 sem_params.key = key;
299 sem_params.flg = semflg;
300 sem_params.u.nsems = nsems;
302 return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
305 /* Manage the doubly linked list sma->sem_pending as a FIFO:
306 * insert new queue elements at the tail sma->sem_pending_last.
308 static inline void append_to_queue (struct sem_array * sma,
309 struct sem_queue * q)
311 *(q->prev = sma->sem_pending_last) = q;
312 *(sma->sem_pending_last = &q->next) = NULL;
315 static inline void prepend_to_queue (struct sem_array * sma,
316 struct sem_queue * q)
318 q->next = sma->sem_pending;
319 *(q->prev = &sma->sem_pending) = q;
321 q->next->prev = &q->next;
322 else /* sma->sem_pending_last == &sma->sem_pending */
323 sma->sem_pending_last = &q->next;
326 static inline void remove_from_queue (struct sem_array * sma,
327 struct sem_queue * q)
329 *(q->prev) = q->next;
331 q->next->prev = q->prev;
332 else /* sma->sem_pending_last == &q->next */
333 sma->sem_pending_last = q->prev;
334 q->prev = NULL; /* mark as removed */
338 * Determine whether a sequence of semaphore operations would succeed
339 * all at once. Return 0 if yes, 1 if need to sleep, else return error code.
342 static int try_atomic_semop (struct sem_array * sma, struct sembuf * sops,
343 int nsops, struct sem_undo *un, int pid)
349 for (sop = sops; sop < sops + nsops; sop++) {
350 curr = sma->sem_base + sop->sem_num;
351 sem_op = sop->sem_op;
352 result = curr->semval;
354 if (!sem_op && result)
362 if (sop->sem_flg & SEM_UNDO) {
363 int undo = un->semadj[sop->sem_num] - sem_op;
365 * Exceeding the undo range is an error.
367 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
370 curr->semval = result;
374 while (sop >= sops) {
375 sma->sem_base[sop->sem_num].sempid = pid;
376 if (sop->sem_flg & SEM_UNDO)
377 un->semadj[sop->sem_num] -= sop->sem_op;
381 sma->sem_otime = get_seconds();
389 if (sop->sem_flg & IPC_NOWAIT)
396 while (sop >= sops) {
397 sma->sem_base[sop->sem_num].semval -= sop->sem_op;
404 /* Go through the pending queue for the indicated semaphore
405 * looking for tasks that can be completed.
407 static void update_queue (struct sem_array * sma)
410 struct sem_queue * q;
412 q = sma->sem_pending;
414 error = try_atomic_semop(sma, q->sops, q->nsops,
417 /* Does q->sleeper still need to sleep? */
420 remove_from_queue(sma,q);
421 q->status = IN_WAKEUP;
423 * Continue scanning. The next operation
424 * that must be checked depends on the type of the
425 * completed operation:
426 * - if the operation modified the array, then
427 * restart from the head of the queue and
428 * check for threads that might be waiting
429 * for semaphore values to become 0.
430 * - if the operation didn't modify the array,
431 * then just continue.
434 n = sma->sem_pending;
437 wake_up_process(q->sleeper);
438 /* hands-off: q will disappear immediately after
450 /* The following counts are associated to each semaphore:
451 * semncnt number of tasks waiting on semval being nonzero
452 * semzcnt number of tasks waiting on semval being zero
453 * This model assumes that a task waits on exactly one semaphore.
454 * Since semaphore operations are to be performed atomically, tasks actually
455 * wait on a whole sequence of semaphores simultaneously.
456 * The counts we return here are a rough approximation, but still
457 * warrant that semncnt+semzcnt>0 if the task is on the pending queue.
459 static int count_semncnt (struct sem_array * sma, ushort semnum)
462 struct sem_queue * q;
465 for (q = sma->sem_pending; q; q = q->next) {
466 struct sembuf * sops = q->sops;
467 int nsops = q->nsops;
469 for (i = 0; i < nsops; i++)
470 if (sops[i].sem_num == semnum
471 && (sops[i].sem_op < 0)
472 && !(sops[i].sem_flg & IPC_NOWAIT))
477 static int count_semzcnt (struct sem_array * sma, ushort semnum)
480 struct sem_queue * q;
483 for (q = sma->sem_pending; q; q = q->next) {
484 struct sembuf * sops = q->sops;
485 int nsops = q->nsops;
487 for (i = 0; i < nsops; i++)
488 if (sops[i].sem_num == semnum
489 && (sops[i].sem_op == 0)
490 && !(sops[i].sem_flg & IPC_NOWAIT))
496 /* Free a semaphore set. freeary() is called with sem_ids.mutex locked and
497 * the spinlock for this semaphore set hold. sem_ids.mutex remains locked
500 static void freeary(struct ipc_namespace *ns, struct sem_array *sma)
505 /* Invalidate the existing undo structures for this semaphore set.
506 * (They will be freed without any further action in exit_sem()
507 * or during the next semop.)
509 for (un = sma->undo; un; un = un->id_next)
512 /* Wake up all pending processes and let them fail with EIDRM. */
513 q = sma->sem_pending;
516 /* lazy remove_from_queue: we are killing the whole queue */
519 q->status = IN_WAKEUP;
520 wake_up_process(q->sleeper); /* doesn't sleep */
522 q->status = -EIDRM; /* hands-off q */
526 /* Remove the semaphore set from the IDR */
530 ns->used_sems -= sma->sem_nsems;
531 security_sem_free(sma);
535 static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
539 return copy_to_user(buf, in, sizeof(*in));
544 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
546 out.sem_otime = in->sem_otime;
547 out.sem_ctime = in->sem_ctime;
548 out.sem_nsems = in->sem_nsems;
550 return copy_to_user(buf, &out, sizeof(out));
557 static int semctl_nolock(struct ipc_namespace *ns, int semid, int semnum,
558 int cmd, int version, union semun arg)
561 struct sem_array *sma;
567 struct seminfo seminfo;
570 err = security_sem_semctl(NULL, cmd);
574 memset(&seminfo,0,sizeof(seminfo));
575 seminfo.semmni = ns->sc_semmni;
576 seminfo.semmns = ns->sc_semmns;
577 seminfo.semmsl = ns->sc_semmsl;
578 seminfo.semopm = ns->sc_semopm;
579 seminfo.semvmx = SEMVMX;
580 seminfo.semmnu = SEMMNU;
581 seminfo.semmap = SEMMAP;
582 seminfo.semume = SEMUME;
583 mutex_lock(&sem_ids(ns).mutex);
584 if (cmd == SEM_INFO) {
585 seminfo.semusz = sem_ids(ns).in_use;
586 seminfo.semaem = ns->used_sems;
588 seminfo.semusz = SEMUSZ;
589 seminfo.semaem = SEMAEM;
591 max_id = ipc_get_maxid(&sem_ids(ns));
592 mutex_unlock(&sem_ids(ns).mutex);
593 if (copy_to_user (arg.__buf, &seminfo, sizeof(struct seminfo)))
595 return (max_id < 0) ? 0: max_id;
599 struct semid64_ds tbuf;
602 memset(&tbuf,0,sizeof(tbuf));
604 sma = sem_lock(ns, semid);
609 if (ipcperms (&sma->sem_perm, S_IRUGO))
612 err = security_sem_semctl(sma, cmd);
616 id = sma->sem_perm.id;
618 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
619 tbuf.sem_otime = sma->sem_otime;
620 tbuf.sem_ctime = sma->sem_ctime;
621 tbuf.sem_nsems = sma->sem_nsems;
623 if (copy_semid_to_user (arg.buf, &tbuf, version))
636 static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
637 int cmd, int version, union semun arg)
639 struct sem_array *sma;
642 ushort fast_sem_io[SEMMSL_FAST];
643 ushort* sem_io = fast_sem_io;
646 sma = sem_lock(ns, semid);
650 nsems = sma->sem_nsems;
653 if (sem_checkid(ns,sma,semid))
657 if (ipcperms (&sma->sem_perm, (cmd==SETVAL||cmd==SETALL)?S_IWUGO:S_IRUGO))
660 err = security_sem_semctl(sma, cmd);
668 ushort __user *array = arg.array;
671 if(nsems > SEMMSL_FAST) {
675 sem_io = ipc_alloc(sizeof(ushort)*nsems);
677 ipc_lock_by_ptr(&sma->sem_perm);
683 ipc_lock_by_ptr(&sma->sem_perm);
685 if (sma->sem_perm.deleted) {
692 for (i = 0; i < sma->sem_nsems; i++)
693 sem_io[i] = sma->sem_base[i].semval;
696 if(copy_to_user(array, sem_io, nsems*sizeof(ushort)))
708 if(nsems > SEMMSL_FAST) {
709 sem_io = ipc_alloc(sizeof(ushort)*nsems);
711 ipc_lock_by_ptr(&sma->sem_perm);
718 if (copy_from_user (sem_io, arg.array, nsems*sizeof(ushort))) {
719 ipc_lock_by_ptr(&sma->sem_perm);
726 for (i = 0; i < nsems; i++) {
727 if (sem_io[i] > SEMVMX) {
728 ipc_lock_by_ptr(&sma->sem_perm);
735 ipc_lock_by_ptr(&sma->sem_perm);
737 if (sma->sem_perm.deleted) {
743 for (i = 0; i < nsems; i++)
744 sma->sem_base[i].semval = sem_io[i];
745 for (un = sma->undo; un; un = un->id_next)
746 for (i = 0; i < nsems; i++)
748 sma->sem_ctime = get_seconds();
749 /* maybe some queued-up processes were waiting for this */
756 struct semid64_ds tbuf;
757 memset(&tbuf,0,sizeof(tbuf));
758 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
759 tbuf.sem_otime = sma->sem_otime;
760 tbuf.sem_ctime = sma->sem_ctime;
761 tbuf.sem_nsems = sma->sem_nsems;
763 if (copy_semid_to_user (arg.buf, &tbuf, version))
767 /* GETVAL, GETPID, GETNCTN, GETZCNT, SETVAL: fall-through */
770 if(semnum < 0 || semnum >= nsems)
773 curr = &sma->sem_base[semnum];
783 err = count_semncnt(sma,semnum);
786 err = count_semzcnt(sma,semnum);
793 if (val > SEMVMX || val < 0)
796 for (un = sma->undo; un; un = un->id_next)
797 un->semadj[semnum] = 0;
799 curr->sempid = task_tgid_vnr(current);
800 sma->sem_ctime = get_seconds();
801 /* maybe some queued-up processes were waiting for this */
810 if(sem_io != fast_sem_io)
811 ipc_free(sem_io, sizeof(ushort)*nsems);
821 static inline unsigned long copy_semid_from_user(struct sem_setbuf *out, void __user *buf, int version)
826 struct semid64_ds tbuf;
828 if(copy_from_user(&tbuf, buf, sizeof(tbuf)))
831 out->uid = tbuf.sem_perm.uid;
832 out->gid = tbuf.sem_perm.gid;
833 out->mode = tbuf.sem_perm.mode;
839 struct semid_ds tbuf_old;
841 if(copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
844 out->uid = tbuf_old.sem_perm.uid;
845 out->gid = tbuf_old.sem_perm.gid;
846 out->mode = tbuf_old.sem_perm.mode;
855 static int semctl_down(struct ipc_namespace *ns, int semid, int semnum,
856 int cmd, int version, union semun arg)
858 struct sem_array *sma;
860 struct sem_setbuf uninitialized_var(setbuf);
861 struct kern_ipc_perm *ipcp;
864 if(copy_semid_from_user (&setbuf, arg.buf, version))
867 sma = sem_lock(ns, semid);
871 if (sem_checkid(ns,sma,semid)) {
875 ipcp = &sma->sem_perm;
877 err = audit_ipc_obj(ipcp);
881 if (cmd == IPC_SET) {
882 err = audit_ipc_set_perm(0, setbuf.uid, setbuf.gid, setbuf.mode);
886 if (current->euid != ipcp->cuid &&
887 current->euid != ipcp->uid && !capable(CAP_SYS_ADMIN)) {
892 err = security_sem_semctl(sma, cmd);
902 ipcp->uid = setbuf.uid;
903 ipcp->gid = setbuf.gid;
904 ipcp->mode = (ipcp->mode & ~S_IRWXUGO)
905 | (setbuf.mode & S_IRWXUGO);
906 sma->sem_ctime = get_seconds();
922 asmlinkage long sys_semctl (int semid, int semnum, int cmd, union semun arg)
926 struct ipc_namespace *ns;
931 version = ipc_parse_version(&cmd);
932 ns = current->nsproxy->ipc_ns;
938 err = semctl_nolock(ns,semid,semnum,cmd,version,arg);
948 err = semctl_main(ns,semid,semnum,cmd,version,arg);
952 mutex_lock(&sem_ids(ns).mutex);
953 err = semctl_down(ns,semid,semnum,cmd,version,arg);
954 mutex_unlock(&sem_ids(ns).mutex);
961 static inline void lock_semundo(void)
963 struct sem_undo_list *undo_list;
965 undo_list = current->sysvsem.undo_list;
967 spin_lock(&undo_list->lock);
970 /* This code has an interaction with copy_semundo().
971 * Consider; two tasks are sharing the undo_list. task1
972 * acquires the undo_list lock in lock_semundo(). If task2 now
973 * exits before task1 releases the lock (by calling
974 * unlock_semundo()), then task1 will never call spin_unlock().
975 * This leave the sem_undo_list in a locked state. If task1 now creats task3
976 * and once again shares the sem_undo_list, the sem_undo_list will still be
977 * locked, and future SEM_UNDO operations will deadlock. This case is
978 * dealt with in copy_semundo() by having it reinitialize the spin lock when
979 * the refcnt goes from 1 to 2.
981 static inline void unlock_semundo(void)
983 struct sem_undo_list *undo_list;
985 undo_list = current->sysvsem.undo_list;
987 spin_unlock(&undo_list->lock);
991 /* If the task doesn't already have a undo_list, then allocate one
992 * here. We guarantee there is only one thread using this undo list,
993 * and current is THE ONE
995 * If this allocation and assignment succeeds, but later
996 * portions of this code fail, there is no need to free the sem_undo_list.
997 * Just let it stay associated with the task, and it'll be freed later
1000 * This can block, so callers must hold no locks.
1002 static inline int get_undo_list(struct sem_undo_list **undo_listp)
1004 struct sem_undo_list *undo_list;
1006 undo_list = current->sysvsem.undo_list;
1008 undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
1009 if (undo_list == NULL)
1011 spin_lock_init(&undo_list->lock);
1012 atomic_set(&undo_list->refcnt, 1);
1013 current->sysvsem.undo_list = undo_list;
1015 *undo_listp = undo_list;
1019 static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1021 struct sem_undo **last, *un;
1023 last = &ulp->proc_list;
1026 if(un->semid==semid)
1029 *last=un->proc_next;
1032 last=&un->proc_next;
1039 static struct sem_undo *find_undo(struct ipc_namespace *ns, int semid)
1041 struct sem_array *sma;
1042 struct sem_undo_list *ulp;
1043 struct sem_undo *un, *new;
1047 error = get_undo_list(&ulp);
1049 return ERR_PTR(error);
1052 un = lookup_undo(ulp, semid);
1054 if (likely(un!=NULL))
1057 /* no undo structure around - allocate one. */
1058 sma = sem_lock(ns, semid);
1059 un = ERR_PTR(-EINVAL);
1062 un = ERR_PTR(-EIDRM);
1063 if (sem_checkid(ns,sma,semid)) {
1067 nsems = sma->sem_nsems;
1068 ipc_rcu_getref(sma);
1071 new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1073 ipc_lock_by_ptr(&sma->sem_perm);
1074 ipc_rcu_putref(sma);
1076 return ERR_PTR(-ENOMEM);
1078 new->semadj = (short *) &new[1];
1082 un = lookup_undo(ulp, semid);
1086 ipc_lock_by_ptr(&sma->sem_perm);
1087 ipc_rcu_putref(sma);
1091 ipc_lock_by_ptr(&sma->sem_perm);
1092 ipc_rcu_putref(sma);
1093 if (sma->sem_perm.deleted) {
1097 un = ERR_PTR(-EIDRM);
1100 new->proc_next = ulp->proc_list;
1101 ulp->proc_list = new;
1102 new->id_next = sma->undo;
1111 asmlinkage long sys_semtimedop(int semid, struct sembuf __user *tsops,
1112 unsigned nsops, const struct timespec __user *timeout)
1114 int error = -EINVAL;
1115 struct sem_array *sma;
1116 struct sembuf fast_sops[SEMOPM_FAST];
1117 struct sembuf* sops = fast_sops, *sop;
1118 struct sem_undo *un;
1119 int undos = 0, alter = 0, max;
1120 struct sem_queue queue;
1121 unsigned long jiffies_left = 0;
1122 struct ipc_namespace *ns;
1124 ns = current->nsproxy->ipc_ns;
1126 if (nsops < 1 || semid < 0)
1128 if (nsops > ns->sc_semopm)
1130 if(nsops > SEMOPM_FAST) {
1131 sops = kmalloc(sizeof(*sops)*nsops,GFP_KERNEL);
1135 if (copy_from_user (sops, tsops, nsops * sizeof(*tsops))) {
1140 struct timespec _timeout;
1141 if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1145 if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1146 _timeout.tv_nsec >= 1000000000L) {
1150 jiffies_left = timespec_to_jiffies(&_timeout);
1153 for (sop = sops; sop < sops + nsops; sop++) {
1154 if (sop->sem_num >= max)
1156 if (sop->sem_flg & SEM_UNDO)
1158 if (sop->sem_op != 0)
1164 un = find_undo(ns, semid);
1166 error = PTR_ERR(un);
1172 sma = sem_lock(ns, semid);
1177 if (sem_checkid(ns,sma,semid))
1178 goto out_unlock_free;
1180 * semid identifies are not unique - find_undo may have
1181 * allocated an undo structure, it was invalidated by an RMID
1182 * and now a new array with received the same id. Check and retry.
1184 if (un && un->semid == -1) {
1189 if (max >= sma->sem_nsems)
1190 goto out_unlock_free;
1193 if (ipcperms(&sma->sem_perm, alter ? S_IWUGO : S_IRUGO))
1194 goto out_unlock_free;
1196 error = security_sem_semop(sma, sops, nsops, alter);
1198 goto out_unlock_free;
1200 error = try_atomic_semop (sma, sops, nsops, un, task_tgid_vnr(current));
1202 if (alter && error == 0)
1204 goto out_unlock_free;
1207 /* We need to sleep on this operation, so we put the current
1208 * task into the pending queue and go to sleep.
1213 queue.nsops = nsops;
1215 queue.pid = task_tgid_vnr(current);
1217 queue.alter = alter;
1219 append_to_queue(sma ,&queue);
1221 prepend_to_queue(sma ,&queue);
1223 queue.status = -EINTR;
1224 queue.sleeper = current;
1225 current->state = TASK_INTERRUPTIBLE;
1229 jiffies_left = schedule_timeout(jiffies_left);
1233 error = queue.status;
1234 while(unlikely(error == IN_WAKEUP)) {
1236 error = queue.status;
1239 if (error != -EINTR) {
1240 /* fast path: update_queue already obtained all requested
1245 sma = sem_lock(ns, semid);
1247 BUG_ON(queue.prev != NULL);
1253 * If queue.status != -EINTR we are woken up by another process
1255 error = queue.status;
1256 if (error != -EINTR) {
1257 goto out_unlock_free;
1261 * If an interrupt occurred we have to clean up the queue
1263 if (timeout && jiffies_left == 0)
1265 remove_from_queue(sma,&queue);
1266 goto out_unlock_free;
1271 if(sops != fast_sops)
1276 asmlinkage long sys_semop (int semid, struct sembuf __user *tsops, unsigned nsops)
1278 return sys_semtimedop(semid, tsops, nsops, NULL);
1281 /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
1282 * parent and child tasks.
1284 * See the notes above unlock_semundo() regarding the spin_lock_init()
1285 * in this code. Initialize the undo_list->lock here instead of get_undo_list()
1286 * because of the reasoning in the comment above unlock_semundo.
1289 int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
1291 struct sem_undo_list *undo_list;
1294 if (clone_flags & CLONE_SYSVSEM) {
1295 error = get_undo_list(&undo_list);
1298 atomic_inc(&undo_list->refcnt);
1299 tsk->sysvsem.undo_list = undo_list;
1301 tsk->sysvsem.undo_list = NULL;
1307 * add semadj values to semaphores, free undo structures.
1308 * undo structures are not freed when semaphore arrays are destroyed
1309 * so some of them may be out of date.
1310 * IMPLEMENTATION NOTE: There is some confusion over whether the
1311 * set of adjustments that needs to be done should be done in an atomic
1312 * manner or not. That is, if we are attempting to decrement the semval
1313 * should we queue up and wait until we can do so legally?
1314 * The original implementation attempted to do this (queue and wait).
1315 * The current implementation does not do so. The POSIX standard
1316 * and SVID should be consulted to determine what behavior is mandated.
1318 void exit_sem(struct task_struct *tsk)
1320 struct sem_undo_list *undo_list;
1321 struct sem_undo *u, **up;
1322 struct ipc_namespace *ns;
1324 undo_list = tsk->sysvsem.undo_list;
1328 if (!atomic_dec_and_test(&undo_list->refcnt))
1331 ns = tsk->nsproxy->ipc_ns;
1332 /* There's no need to hold the semundo list lock, as current
1333 * is the last task exiting for this undo list.
1335 for (up = &undo_list->proc_list; (u = *up); *up = u->proc_next, kfree(u)) {
1336 struct sem_array *sma;
1338 struct sem_undo *un, **unp;
1345 sma = sem_lock(ns, semid);
1352 BUG_ON(sem_checkid(ns,sma,u->semid));
1354 /* remove u from the sma->undo list */
1355 for (unp = &sma->undo; (un = *unp); unp = &un->id_next) {
1359 printk ("exit_sem undo list error id=%d\n", u->semid);
1363 /* perform adjustments registered in u */
1364 nsems = sma->sem_nsems;
1365 for (i = 0; i < nsems; i++) {
1366 struct sem * semaphore = &sma->sem_base[i];
1368 semaphore->semval += u->semadj[i];
1370 * Range checks of the new semaphore value,
1371 * not defined by sus:
1372 * - Some unices ignore the undo entirely
1373 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
1374 * - some cap the value (e.g. FreeBSD caps
1375 * at 0, but doesn't enforce SEMVMX)
1377 * Linux caps the semaphore value, both at 0
1380 * Manfred <manfred@colorfullife.com>
1382 if (semaphore->semval < 0)
1383 semaphore->semval = 0;
1384 if (semaphore->semval > SEMVMX)
1385 semaphore->semval = SEMVMX;
1386 semaphore->sempid = task_tgid_vnr(current);
1389 sma->sem_otime = get_seconds();
1390 /* maybe some queued-up processes were waiting for this */
1398 #ifdef CONFIG_PROC_FS
1399 static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
1401 struct sem_array *sma = it;
1403 return seq_printf(s,
1404 "%10d %10d %4o %10lu %5u %5u %5u %5u %10lu %10lu\n",