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_unlock(sma) ipc_unlock(&(sma)->sem_perm)
92 #define sem_checkid(ns, sma, semid) \
93 ipc_checkid(&sem_ids(ns),&sma->sem_perm,semid)
94 #define sem_buildid(ns, id, seq) \
95 ipc_buildid(&sem_ids(ns), id, seq)
97 static struct ipc_ids init_sem_ids;
99 static int newary(struct ipc_namespace *, struct ipc_params *);
100 static void freeary(struct ipc_namespace *, struct sem_array *);
101 #ifdef CONFIG_PROC_FS
102 static int sysvipc_sem_proc_show(struct seq_file *s, void *it);
105 #define SEMMSL_FAST 256 /* 512 bytes on stack */
106 #define SEMOPM_FAST 64 /* ~ 372 bytes on stack */
109 * linked list protection:
111 * sem_array.sem_pending{,last},
112 * sem_array.sem_undo: sem_lock() for read/write
113 * sem_undo.proc_next: only "current" is allowed to read/write that field.
117 #define sc_semmsl sem_ctls[0]
118 #define sc_semmns sem_ctls[1]
119 #define sc_semopm sem_ctls[2]
120 #define sc_semmni sem_ctls[3]
122 static void __sem_init_ns(struct ipc_namespace *ns, struct ipc_ids *ids)
124 ns->ids[IPC_SEM_IDS] = ids;
125 ns->sc_semmsl = SEMMSL;
126 ns->sc_semmns = SEMMNS;
127 ns->sc_semopm = SEMOPM;
128 ns->sc_semmni = SEMMNI;
133 int sem_init_ns(struct ipc_namespace *ns)
137 ids = kmalloc(sizeof(struct ipc_ids), GFP_KERNEL);
141 __sem_init_ns(ns, ids);
145 void sem_exit_ns(struct ipc_namespace *ns)
147 struct sem_array *sma;
151 mutex_lock(&sem_ids(ns).mutex);
153 in_use = sem_ids(ns).in_use;
155 for (total = 0, next_id = 0; total < in_use; next_id++) {
156 sma = idr_find(&sem_ids(ns).ipcs_idr, next_id);
159 ipc_lock_by_ptr(&sma->sem_perm);
163 mutex_unlock(&sem_ids(ns).mutex);
165 kfree(ns->ids[IPC_SEM_IDS]);
166 ns->ids[IPC_SEM_IDS] = NULL;
169 void __init sem_init (void)
171 __sem_init_ns(&init_ipc_ns, &init_sem_ids);
172 ipc_init_proc_interface("sysvipc/sem",
173 " key semid perms nsems uid gid cuid cgid otime ctime\n",
174 IPC_SEM_IDS, sysvipc_sem_proc_show);
177 static inline struct sem_array *sem_lock(struct ipc_namespace *ns, int id)
179 return (struct sem_array *) ipc_lock(&sem_ids(ns), id);
182 static inline struct sem_array *sem_lock_check(struct ipc_namespace *ns,
185 return (struct sem_array *) ipc_lock_check(&sem_ids(ns), id);
188 static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
190 ipc_rmid(&sem_ids(ns), &s->sem_perm);
194 * Lockless wakeup algorithm:
195 * Without the check/retry algorithm a lockless wakeup is possible:
196 * - queue.status is initialized to -EINTR before blocking.
197 * - wakeup is performed by
198 * * unlinking the queue entry from sma->sem_pending
199 * * setting queue.status to IN_WAKEUP
200 * This is the notification for the blocked thread that a
201 * result value is imminent.
202 * * call wake_up_process
203 * * set queue.status to the final value.
204 * - the previously blocked thread checks queue.status:
205 * * if it's IN_WAKEUP, then it must wait until the value changes
206 * * if it's not -EINTR, then the operation was completed by
207 * update_queue. semtimedop can return queue.status without
208 * performing any operation on the sem array.
209 * * otherwise it must acquire the spinlock and check what's up.
211 * The two-stage algorithm is necessary to protect against the following
213 * - if queue.status is set after wake_up_process, then the woken up idle
214 * thread could race forward and try (and fail) to acquire sma->lock
215 * before update_queue had a chance to set queue.status
216 * - if queue.status is written before wake_up_process and if the
217 * blocked process is woken up by a signal between writing
218 * queue.status and the wake_up_process, then the woken up
219 * process could return from semtimedop and die by calling
220 * sys_exit before wake_up_process is called. Then wake_up_process
221 * will oops, because the task structure is already invalid.
222 * (yes, this happened on s390 with sysv msg).
227 static int newary(struct ipc_namespace *ns, struct ipc_params *params)
231 struct sem_array *sma;
233 key_t key = params->key;
234 int nsems = params->u.nsems;
235 int semflg = params->flg;
239 if (ns->used_sems + nsems > ns->sc_semmns)
242 size = sizeof (*sma) + nsems * sizeof (struct sem);
243 sma = ipc_rcu_alloc(size);
247 memset (sma, 0, size);
249 sma->sem_perm.mode = (semflg & S_IRWXUGO);
250 sma->sem_perm.key = key;
252 sma->sem_perm.security = NULL;
253 retval = security_sem_alloc(sma);
259 id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
261 security_sem_free(sma);
265 ns->used_sems += nsems;
267 sma->sem_perm.id = sem_buildid(ns, id, sma->sem_perm.seq);
268 sma->sem_base = (struct sem *) &sma[1];
269 /* sma->sem_pending = NULL; */
270 sma->sem_pending_last = &sma->sem_pending;
271 /* sma->undo = NULL; */
272 sma->sem_nsems = nsems;
273 sma->sem_ctime = get_seconds();
276 return sma->sem_perm.id;
280 static inline int sem_security(void *sma, int semflg)
282 return security_sem_associate((struct sem_array *) sma, semflg);
285 static inline int sem_more_checks(void *sma, struct ipc_params *params)
287 if (params->u.nsems > ((struct sem_array *)sma)->sem_nsems)
293 asmlinkage long sys_semget(key_t key, int nsems, int semflg)
295 struct ipc_namespace *ns;
296 struct ipc_ops sem_ops;
297 struct ipc_params sem_params;
299 ns = current->nsproxy->ipc_ns;
301 if (nsems < 0 || nsems > ns->sc_semmsl)
304 sem_ops.getnew = newary;
305 sem_ops.associate = sem_security;
306 sem_ops.more_checks = sem_more_checks;
308 sem_params.key = key;
309 sem_params.flg = semflg;
310 sem_params.u.nsems = nsems;
312 return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
315 /* Manage the doubly linked list sma->sem_pending as a FIFO:
316 * insert new queue elements at the tail sma->sem_pending_last.
318 static inline void append_to_queue (struct sem_array * sma,
319 struct sem_queue * q)
321 *(q->prev = sma->sem_pending_last) = q;
322 *(sma->sem_pending_last = &q->next) = NULL;
325 static inline void prepend_to_queue (struct sem_array * sma,
326 struct sem_queue * q)
328 q->next = sma->sem_pending;
329 *(q->prev = &sma->sem_pending) = q;
331 q->next->prev = &q->next;
332 else /* sma->sem_pending_last == &sma->sem_pending */
333 sma->sem_pending_last = &q->next;
336 static inline void remove_from_queue (struct sem_array * sma,
337 struct sem_queue * q)
339 *(q->prev) = q->next;
341 q->next->prev = q->prev;
342 else /* sma->sem_pending_last == &q->next */
343 sma->sem_pending_last = q->prev;
344 q->prev = NULL; /* mark as removed */
348 * Determine whether a sequence of semaphore operations would succeed
349 * all at once. Return 0 if yes, 1 if need to sleep, else return error code.
352 static int try_atomic_semop (struct sem_array * sma, struct sembuf * sops,
353 int nsops, struct sem_undo *un, int pid)
359 for (sop = sops; sop < sops + nsops; sop++) {
360 curr = sma->sem_base + sop->sem_num;
361 sem_op = sop->sem_op;
362 result = curr->semval;
364 if (!sem_op && result)
372 if (sop->sem_flg & SEM_UNDO) {
373 int undo = un->semadj[sop->sem_num] - sem_op;
375 * Exceeding the undo range is an error.
377 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
380 curr->semval = result;
384 while (sop >= sops) {
385 sma->sem_base[sop->sem_num].sempid = pid;
386 if (sop->sem_flg & SEM_UNDO)
387 un->semadj[sop->sem_num] -= sop->sem_op;
391 sma->sem_otime = get_seconds();
399 if (sop->sem_flg & IPC_NOWAIT)
406 while (sop >= sops) {
407 sma->sem_base[sop->sem_num].semval -= sop->sem_op;
414 /* Go through the pending queue for the indicated semaphore
415 * looking for tasks that can be completed.
417 static void update_queue (struct sem_array * sma)
420 struct sem_queue * q;
422 q = sma->sem_pending;
424 error = try_atomic_semop(sma, q->sops, q->nsops,
427 /* Does q->sleeper still need to sleep? */
430 remove_from_queue(sma,q);
431 q->status = IN_WAKEUP;
433 * Continue scanning. The next operation
434 * that must be checked depends on the type of the
435 * completed operation:
436 * - if the operation modified the array, then
437 * restart from the head of the queue and
438 * check for threads that might be waiting
439 * for semaphore values to become 0.
440 * - if the operation didn't modify the array,
441 * then just continue.
444 n = sma->sem_pending;
447 wake_up_process(q->sleeper);
448 /* hands-off: q will disappear immediately after
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 for (q = sma->sem_pending; q; q = q->next) {
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))
487 static int count_semzcnt (struct sem_array * sma, ushort semnum)
490 struct sem_queue * q;
493 for (q = sma->sem_pending; q; q = q->next) {
494 struct sembuf * sops = q->sops;
495 int nsops = q->nsops;
497 for (i = 0; i < nsops; i++)
498 if (sops[i].sem_num == semnum
499 && (sops[i].sem_op == 0)
500 && !(sops[i].sem_flg & IPC_NOWAIT))
506 /* Free a semaphore set. freeary() is called with sem_ids.mutex locked and
507 * the spinlock for this semaphore set hold. sem_ids.mutex remains locked
510 static void freeary(struct ipc_namespace *ns, struct sem_array *sma)
515 /* Invalidate the existing undo structures for this semaphore set.
516 * (They will be freed without any further action in exit_sem()
517 * or during the next semop.)
519 for (un = sma->undo; un; un = un->id_next)
522 /* Wake up all pending processes and let them fail with EIDRM. */
523 q = sma->sem_pending;
526 /* lazy remove_from_queue: we are killing the whole queue */
529 q->status = IN_WAKEUP;
530 wake_up_process(q->sleeper); /* doesn't sleep */
532 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, int semnum,
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 mutex_lock(&sem_ids(ns).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 mutex_unlock(&sem_ids(ns).mutex);
603 if (copy_to_user (arg.__buf, &seminfo, sizeof(struct seminfo)))
605 return (max_id < 0) ? 0: max_id;
609 struct semid64_ds tbuf;
612 sma = sem_lock(ns, semid);
617 if (ipcperms (&sma->sem_perm, S_IRUGO))
620 err = security_sem_semctl(sma, cmd);
624 id = sma->sem_perm.id;
626 memset(&tbuf, 0, sizeof(tbuf));
628 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
629 tbuf.sem_otime = sma->sem_otime;
630 tbuf.sem_ctime = sma->sem_ctime;
631 tbuf.sem_nsems = sma->sem_nsems;
633 if (copy_semid_to_user (arg.buf, &tbuf, version))
646 static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
647 int cmd, int version, union semun arg)
649 struct sem_array *sma;
652 ushort fast_sem_io[SEMMSL_FAST];
653 ushort* sem_io = fast_sem_io;
656 sma = sem_lock_check(ns, semid);
660 nsems = sma->sem_nsems;
663 if (ipcperms (&sma->sem_perm, (cmd==SETVAL||cmd==SETALL)?S_IWUGO:S_IRUGO))
666 err = security_sem_semctl(sma, cmd);
674 ushort __user *array = arg.array;
677 if(nsems > SEMMSL_FAST) {
681 sem_io = ipc_alloc(sizeof(ushort)*nsems);
683 ipc_lock_by_ptr(&sma->sem_perm);
689 ipc_lock_by_ptr(&sma->sem_perm);
691 if (sma->sem_perm.deleted) {
698 for (i = 0; i < sma->sem_nsems; i++)
699 sem_io[i] = sma->sem_base[i].semval;
702 if(copy_to_user(array, sem_io, nsems*sizeof(ushort)))
714 if(nsems > SEMMSL_FAST) {
715 sem_io = ipc_alloc(sizeof(ushort)*nsems);
717 ipc_lock_by_ptr(&sma->sem_perm);
724 if (copy_from_user (sem_io, arg.array, nsems*sizeof(ushort))) {
725 ipc_lock_by_ptr(&sma->sem_perm);
732 for (i = 0; i < nsems; i++) {
733 if (sem_io[i] > SEMVMX) {
734 ipc_lock_by_ptr(&sma->sem_perm);
741 ipc_lock_by_ptr(&sma->sem_perm);
743 if (sma->sem_perm.deleted) {
749 for (i = 0; i < nsems; i++)
750 sma->sem_base[i].semval = sem_io[i];
751 for (un = sma->undo; un; un = un->id_next)
752 for (i = 0; i < nsems; i++)
754 sma->sem_ctime = get_seconds();
755 /* maybe some queued-up processes were waiting for this */
762 struct semid64_ds tbuf;
763 memset(&tbuf,0,sizeof(tbuf));
764 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
765 tbuf.sem_otime = sma->sem_otime;
766 tbuf.sem_ctime = sma->sem_ctime;
767 tbuf.sem_nsems = sma->sem_nsems;
769 if (copy_semid_to_user (arg.buf, &tbuf, version))
773 /* GETVAL, GETPID, GETNCTN, GETZCNT, SETVAL: fall-through */
776 if(semnum < 0 || semnum >= nsems)
779 curr = &sma->sem_base[semnum];
789 err = count_semncnt(sma,semnum);
792 err = count_semzcnt(sma,semnum);
799 if (val > SEMVMX || val < 0)
802 for (un = sma->undo; un; un = un->id_next)
803 un->semadj[semnum] = 0;
805 curr->sempid = task_tgid_vnr(current);
806 sma->sem_ctime = get_seconds();
807 /* maybe some queued-up processes were waiting for this */
816 if(sem_io != fast_sem_io)
817 ipc_free(sem_io, sizeof(ushort)*nsems);
827 static inline unsigned long copy_semid_from_user(struct sem_setbuf *out, void __user *buf, int version)
832 struct semid64_ds tbuf;
834 if(copy_from_user(&tbuf, buf, sizeof(tbuf)))
837 out->uid = tbuf.sem_perm.uid;
838 out->gid = tbuf.sem_perm.gid;
839 out->mode = tbuf.sem_perm.mode;
845 struct semid_ds tbuf_old;
847 if(copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
850 out->uid = tbuf_old.sem_perm.uid;
851 out->gid = tbuf_old.sem_perm.gid;
852 out->mode = tbuf_old.sem_perm.mode;
861 static int semctl_down(struct ipc_namespace *ns, int semid, int semnum,
862 int cmd, int version, union semun arg)
864 struct sem_array *sma;
866 struct sem_setbuf uninitialized_var(setbuf);
867 struct kern_ipc_perm *ipcp;
870 if(copy_semid_from_user (&setbuf, arg.buf, version))
873 sma = sem_lock_check(ns, semid);
877 ipcp = &sma->sem_perm;
879 err = audit_ipc_obj(ipcp);
883 if (cmd == IPC_SET) {
884 err = audit_ipc_set_perm(0, setbuf.uid, setbuf.gid, setbuf.mode);
888 if (current->euid != ipcp->cuid &&
889 current->euid != ipcp->uid && !capable(CAP_SYS_ADMIN)) {
894 err = security_sem_semctl(sma, cmd);
904 ipcp->uid = setbuf.uid;
905 ipcp->gid = setbuf.gid;
906 ipcp->mode = (ipcp->mode & ~S_IRWXUGO)
907 | (setbuf.mode & S_IRWXUGO);
908 sma->sem_ctime = get_seconds();
924 asmlinkage long sys_semctl (int semid, int semnum, int cmd, union semun arg)
928 struct ipc_namespace *ns;
933 version = ipc_parse_version(&cmd);
934 ns = current->nsproxy->ipc_ns;
940 err = semctl_nolock(ns,semid,semnum,cmd,version,arg);
950 err = semctl_main(ns,semid,semnum,cmd,version,arg);
954 mutex_lock(&sem_ids(ns).mutex);
955 err = semctl_down(ns,semid,semnum,cmd,version,arg);
956 mutex_unlock(&sem_ids(ns).mutex);
963 static inline void lock_semundo(void)
965 struct sem_undo_list *undo_list;
967 undo_list = current->sysvsem.undo_list;
969 spin_lock(&undo_list->lock);
972 /* This code has an interaction with copy_semundo().
973 * Consider; two tasks are sharing the undo_list. task1
974 * acquires the undo_list lock in lock_semundo(). If task2 now
975 * exits before task1 releases the lock (by calling
976 * unlock_semundo()), then task1 will never call spin_unlock().
977 * This leave the sem_undo_list in a locked state. If task1 now creats task3
978 * and once again shares the sem_undo_list, the sem_undo_list will still be
979 * locked, and future SEM_UNDO operations will deadlock. This case is
980 * dealt with in copy_semundo() by having it reinitialize the spin lock when
981 * the refcnt goes from 1 to 2.
983 static inline void unlock_semundo(void)
985 struct sem_undo_list *undo_list;
987 undo_list = current->sysvsem.undo_list;
989 spin_unlock(&undo_list->lock);
993 /* If the task doesn't already have a undo_list, then allocate one
994 * here. We guarantee there is only one thread using this undo list,
995 * and current is THE ONE
997 * If this allocation and assignment succeeds, but later
998 * portions of this code fail, there is no need to free the sem_undo_list.
999 * Just let it stay associated with the task, and it'll be freed later
1002 * This can block, so callers must hold no locks.
1004 static inline int get_undo_list(struct sem_undo_list **undo_listp)
1006 struct sem_undo_list *undo_list;
1008 undo_list = current->sysvsem.undo_list;
1010 undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
1011 if (undo_list == NULL)
1013 spin_lock_init(&undo_list->lock);
1014 atomic_set(&undo_list->refcnt, 1);
1015 current->sysvsem.undo_list = undo_list;
1017 *undo_listp = undo_list;
1021 static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1023 struct sem_undo **last, *un;
1025 last = &ulp->proc_list;
1028 if(un->semid==semid)
1031 *last=un->proc_next;
1034 last=&un->proc_next;
1041 static struct sem_undo *find_undo(struct ipc_namespace *ns, int semid)
1043 struct sem_array *sma;
1044 struct sem_undo_list *ulp;
1045 struct sem_undo *un, *new;
1049 error = get_undo_list(&ulp);
1051 return ERR_PTR(error);
1054 un = lookup_undo(ulp, semid);
1056 if (likely(un!=NULL))
1059 /* no undo structure around - allocate one. */
1060 sma = sem_lock_check(ns, semid);
1062 return ERR_PTR(PTR_ERR(sma));
1064 nsems = sma->sem_nsems;
1065 ipc_rcu_getref(sma);
1068 new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1070 ipc_lock_by_ptr(&sma->sem_perm);
1071 ipc_rcu_putref(sma);
1073 return ERR_PTR(-ENOMEM);
1075 new->semadj = (short *) &new[1];
1079 un = lookup_undo(ulp, semid);
1083 ipc_lock_by_ptr(&sma->sem_perm);
1084 ipc_rcu_putref(sma);
1088 ipc_lock_by_ptr(&sma->sem_perm);
1089 ipc_rcu_putref(sma);
1090 if (sma->sem_perm.deleted) {
1094 un = ERR_PTR(-EIDRM);
1097 new->proc_next = ulp->proc_list;
1098 ulp->proc_list = new;
1099 new->id_next = sma->undo;
1108 asmlinkage long sys_semtimedop(int semid, struct sembuf __user *tsops,
1109 unsigned nsops, const struct timespec __user *timeout)
1111 int error = -EINVAL;
1112 struct sem_array *sma;
1113 struct sembuf fast_sops[SEMOPM_FAST];
1114 struct sembuf* sops = fast_sops, *sop;
1115 struct sem_undo *un;
1116 int undos = 0, alter = 0, max;
1117 struct sem_queue queue;
1118 unsigned long jiffies_left = 0;
1119 struct ipc_namespace *ns;
1121 ns = current->nsproxy->ipc_ns;
1123 if (nsops < 1 || semid < 0)
1125 if (nsops > ns->sc_semopm)
1127 if(nsops > SEMOPM_FAST) {
1128 sops = kmalloc(sizeof(*sops)*nsops,GFP_KERNEL);
1132 if (copy_from_user (sops, tsops, nsops * sizeof(*tsops))) {
1137 struct timespec _timeout;
1138 if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1142 if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1143 _timeout.tv_nsec >= 1000000000L) {
1147 jiffies_left = timespec_to_jiffies(&_timeout);
1150 for (sop = sops; sop < sops + nsops; sop++) {
1151 if (sop->sem_num >= max)
1153 if (sop->sem_flg & SEM_UNDO)
1155 if (sop->sem_op != 0)
1161 un = find_undo(ns, semid);
1163 error = PTR_ERR(un);
1169 sma = sem_lock_check(ns, semid);
1171 error = PTR_ERR(sma);
1176 * semid identifiers are not unique - find_undo may have
1177 * allocated an undo structure, it was invalidated by an RMID
1178 * and now a new array with received the same id. Check and retry.
1180 if (un && un->semid == -1) {
1185 if (max >= sma->sem_nsems)
1186 goto out_unlock_free;
1189 if (ipcperms(&sma->sem_perm, alter ? S_IWUGO : S_IRUGO))
1190 goto out_unlock_free;
1192 error = security_sem_semop(sma, sops, nsops, alter);
1194 goto out_unlock_free;
1196 error = try_atomic_semop (sma, sops, nsops, un, task_tgid_vnr(current));
1198 if (alter && error == 0)
1200 goto out_unlock_free;
1203 /* We need to sleep on this operation, so we put the current
1204 * task into the pending queue and go to sleep.
1209 queue.nsops = nsops;
1211 queue.pid = task_tgid_vnr(current);
1213 queue.alter = alter;
1215 append_to_queue(sma ,&queue);
1217 prepend_to_queue(sma ,&queue);
1219 queue.status = -EINTR;
1220 queue.sleeper = current;
1221 current->state = TASK_INTERRUPTIBLE;
1225 jiffies_left = schedule_timeout(jiffies_left);
1229 error = queue.status;
1230 while(unlikely(error == IN_WAKEUP)) {
1232 error = queue.status;
1235 if (error != -EINTR) {
1236 /* fast path: update_queue already obtained all requested
1241 sma = sem_lock(ns, semid);
1243 BUG_ON(queue.prev != NULL);
1249 * If queue.status != -EINTR we are woken up by another process
1251 error = queue.status;
1252 if (error != -EINTR) {
1253 goto out_unlock_free;
1257 * If an interrupt occurred we have to clean up the queue
1259 if (timeout && jiffies_left == 0)
1261 remove_from_queue(sma,&queue);
1262 goto out_unlock_free;
1267 if(sops != fast_sops)
1272 asmlinkage long sys_semop (int semid, struct sembuf __user *tsops, unsigned nsops)
1274 return sys_semtimedop(semid, tsops, nsops, NULL);
1277 /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
1278 * parent and child tasks.
1280 * See the notes above unlock_semundo() regarding the spin_lock_init()
1281 * in this code. Initialize the undo_list->lock here instead of get_undo_list()
1282 * because of the reasoning in the comment above unlock_semundo.
1285 int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
1287 struct sem_undo_list *undo_list;
1290 if (clone_flags & CLONE_SYSVSEM) {
1291 error = get_undo_list(&undo_list);
1294 atomic_inc(&undo_list->refcnt);
1295 tsk->sysvsem.undo_list = undo_list;
1297 tsk->sysvsem.undo_list = NULL;
1303 * add semadj values to semaphores, free undo structures.
1304 * undo structures are not freed when semaphore arrays are destroyed
1305 * so some of them may be out of date.
1306 * IMPLEMENTATION NOTE: There is some confusion over whether the
1307 * set of adjustments that needs to be done should be done in an atomic
1308 * manner or not. That is, if we are attempting to decrement the semval
1309 * should we queue up and wait until we can do so legally?
1310 * The original implementation attempted to do this (queue and wait).
1311 * The current implementation does not do so. The POSIX standard
1312 * and SVID should be consulted to determine what behavior is mandated.
1314 void exit_sem(struct task_struct *tsk)
1316 struct sem_undo_list *undo_list;
1317 struct sem_undo *u, **up;
1318 struct ipc_namespace *ns;
1320 undo_list = tsk->sysvsem.undo_list;
1324 if (!atomic_dec_and_test(&undo_list->refcnt))
1327 ns = tsk->nsproxy->ipc_ns;
1328 /* There's no need to hold the semundo list lock, as current
1329 * is the last task exiting for this undo list.
1331 for (up = &undo_list->proc_list; (u = *up); *up = u->proc_next, kfree(u)) {
1332 struct sem_array *sma;
1334 struct sem_undo *un, **unp;
1341 sma = sem_lock(ns, semid);
1348 BUG_ON(sem_checkid(ns,sma,u->semid));
1350 /* remove u from the sma->undo list */
1351 for (unp = &sma->undo; (un = *unp); unp = &un->id_next) {
1355 printk ("exit_sem undo list error id=%d\n", u->semid);
1359 /* perform adjustments registered in u */
1360 nsems = sma->sem_nsems;
1361 for (i = 0; i < nsems; i++) {
1362 struct sem * semaphore = &sma->sem_base[i];
1364 semaphore->semval += u->semadj[i];
1366 * Range checks of the new semaphore value,
1367 * not defined by sus:
1368 * - Some unices ignore the undo entirely
1369 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
1370 * - some cap the value (e.g. FreeBSD caps
1371 * at 0, but doesn't enforce SEMVMX)
1373 * Linux caps the semaphore value, both at 0
1376 * Manfred <manfred@colorfullife.com>
1378 if (semaphore->semval < 0)
1379 semaphore->semval = 0;
1380 if (semaphore->semval > SEMVMX)
1381 semaphore->semval = SEMVMX;
1382 semaphore->sempid = task_tgid_vnr(current);
1385 sma->sem_otime = get_seconds();
1386 /* maybe some queued-up processes were waiting for this */
1394 #ifdef CONFIG_PROC_FS
1395 static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
1397 struct sem_array *sma = it;
1399 return seq_printf(s,
1400 "%10d %10d %4o %10lu %5u %5u %5u %5u %10lu %10lu\n",