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_rmid(ns, id) ((struct sem_array*)ipc_rmid(&sem_ids(ns), id))
94 #define sem_checkid(ns, sma, semid) \
95 ipc_checkid(&sem_ids(ns),&sma->sem_perm,semid)
96 #define sem_buildid(ns, id, seq) \
97 ipc_buildid(&sem_ids(ns), id, seq)
99 static struct ipc_ids init_sem_ids;
101 static int newary(struct ipc_namespace *, key_t, int, int);
102 static void freeary(struct ipc_namespace *ns, struct sem_array *sma, int id);
103 #ifdef CONFIG_PROC_FS
104 static int sysvipc_sem_proc_show(struct seq_file *s, void *it);
107 #define SEMMSL_FAST 256 /* 512 bytes on stack */
108 #define SEMOPM_FAST 64 /* ~ 372 bytes on stack */
111 * linked list protection:
113 * sem_array.sem_pending{,last},
114 * sem_array.sem_undo: sem_lock() for read/write
115 * sem_undo.proc_next: only "current" is allowed to read/write that field.
119 #define sc_semmsl sem_ctls[0]
120 #define sc_semmns sem_ctls[1]
121 #define sc_semopm sem_ctls[2]
122 #define sc_semmni sem_ctls[3]
124 static void __ipc_init __sem_init_ns(struct ipc_namespace *ns, struct ipc_ids *ids)
126 ns->ids[IPC_SEM_IDS] = ids;
127 ns->sc_semmsl = SEMMSL;
128 ns->sc_semmns = SEMMNS;
129 ns->sc_semopm = SEMOPM;
130 ns->sc_semmni = SEMMNI;
132 ipc_init_ids(ids, ns->sc_semmni);
136 int sem_init_ns(struct ipc_namespace *ns)
140 ids = kmalloc(sizeof(struct ipc_ids), GFP_KERNEL);
144 __sem_init_ns(ns, ids);
148 void sem_exit_ns(struct ipc_namespace *ns)
151 struct sem_array *sma;
153 mutex_lock(&sem_ids(ns).mutex);
154 for (i = 0; i <= sem_ids(ns).max_id; i++) {
155 sma = sem_lock(ns, i);
161 mutex_unlock(&sem_ids(ns).mutex);
163 ipc_fini_ids(ns->ids[IPC_SEM_IDS]);
164 kfree(ns->ids[IPC_SEM_IDS]);
165 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);
178 * Lockless wakeup algorithm:
179 * Without the check/retry algorithm a lockless wakeup is possible:
180 * - queue.status is initialized to -EINTR before blocking.
181 * - wakeup is performed by
182 * * unlinking the queue entry from sma->sem_pending
183 * * setting queue.status to IN_WAKEUP
184 * This is the notification for the blocked thread that a
185 * result value is imminent.
186 * * call wake_up_process
187 * * set queue.status to the final value.
188 * - the previously blocked thread checks queue.status:
189 * * if it's IN_WAKEUP, then it must wait until the value changes
190 * * if it's not -EINTR, then the operation was completed by
191 * update_queue. semtimedop can return queue.status without
192 * performing any operation on the sem array.
193 * * otherwise it must acquire the spinlock and check what's up.
195 * The two-stage algorithm is necessary to protect against the following
197 * - if queue.status is set after wake_up_process, then the woken up idle
198 * thread could race forward and try (and fail) to acquire sma->lock
199 * before update_queue had a chance to set queue.status
200 * - if queue.status is written before wake_up_process and if the
201 * blocked process is woken up by a signal between writing
202 * queue.status and the wake_up_process, then the woken up
203 * process could return from semtimedop and die by calling
204 * sys_exit before wake_up_process is called. Then wake_up_process
205 * will oops, because the task structure is already invalid.
206 * (yes, this happened on s390 with sysv msg).
211 static int newary (struct ipc_namespace *ns, key_t key, int nsems, int semflg)
215 struct sem_array *sma;
220 if (ns->used_sems + nsems > ns->sc_semmns)
223 size = sizeof (*sma) + nsems * sizeof (struct sem);
224 sma = ipc_rcu_alloc(size);
228 memset (sma, 0, size);
230 sma->sem_perm.mode = (semflg & S_IRWXUGO);
231 sma->sem_perm.key = key;
233 sma->sem_perm.security = NULL;
234 retval = security_sem_alloc(sma);
240 id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
242 security_sem_free(sma);
246 ns->used_sems += nsems;
248 sma->sem_id = sem_buildid(ns, id, sma->sem_perm.seq);
249 sma->sem_base = (struct sem *) &sma[1];
250 /* sma->sem_pending = NULL; */
251 sma->sem_pending_last = &sma->sem_pending;
252 /* sma->undo = NULL; */
253 sma->sem_nsems = nsems;
254 sma->sem_ctime = get_seconds();
260 asmlinkage long sys_semget (key_t key, int nsems, int semflg)
262 int id, err = -EINVAL;
263 struct sem_array *sma;
264 struct ipc_namespace *ns;
266 ns = current->nsproxy->ipc_ns;
268 if (nsems < 0 || nsems > ns->sc_semmsl)
270 mutex_lock(&sem_ids(ns).mutex);
272 if (key == IPC_PRIVATE) {
273 err = newary(ns, key, nsems, semflg);
274 } else if ((id = ipc_findkey(&sem_ids(ns), key)) == -1) { /* key not used */
275 if (!(semflg & IPC_CREAT))
278 err = newary(ns, key, nsems, semflg);
279 } else if (semflg & IPC_CREAT && semflg & IPC_EXCL) {
282 sma = sem_lock(ns, id);
284 if (nsems > sma->sem_nsems)
286 else if (ipcperms(&sma->sem_perm, semflg))
289 int semid = sem_buildid(ns, id, sma->sem_perm.seq);
290 err = security_sem_associate(sma, semflg);
297 mutex_unlock(&sem_ids(ns).mutex);
301 /* Manage the doubly linked list sma->sem_pending as a FIFO:
302 * insert new queue elements at the tail sma->sem_pending_last.
304 static inline void append_to_queue (struct sem_array * sma,
305 struct sem_queue * q)
307 *(q->prev = sma->sem_pending_last) = q;
308 *(sma->sem_pending_last = &q->next) = NULL;
311 static inline void prepend_to_queue (struct sem_array * sma,
312 struct sem_queue * q)
314 q->next = sma->sem_pending;
315 *(q->prev = &sma->sem_pending) = q;
317 q->next->prev = &q->next;
318 else /* sma->sem_pending_last == &sma->sem_pending */
319 sma->sem_pending_last = &q->next;
322 static inline void remove_from_queue (struct sem_array * sma,
323 struct sem_queue * q)
325 *(q->prev) = q->next;
327 q->next->prev = q->prev;
328 else /* sma->sem_pending_last == &q->next */
329 sma->sem_pending_last = q->prev;
330 q->prev = NULL; /* mark as removed */
334 * Determine whether a sequence of semaphore operations would succeed
335 * all at once. Return 0 if yes, 1 if need to sleep, else return error code.
338 static int try_atomic_semop (struct sem_array * sma, struct sembuf * sops,
339 int nsops, struct sem_undo *un, int pid)
345 for (sop = sops; sop < sops + nsops; sop++) {
346 curr = sma->sem_base + sop->sem_num;
347 sem_op = sop->sem_op;
348 result = curr->semval;
350 if (!sem_op && result)
358 if (sop->sem_flg & SEM_UNDO) {
359 int undo = un->semadj[sop->sem_num] - sem_op;
361 * Exceeding the undo range is an error.
363 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
366 curr->semval = result;
370 while (sop >= sops) {
371 sma->sem_base[sop->sem_num].sempid = pid;
372 if (sop->sem_flg & SEM_UNDO)
373 un->semadj[sop->sem_num] -= sop->sem_op;
377 sma->sem_otime = get_seconds();
385 if (sop->sem_flg & IPC_NOWAIT)
392 while (sop >= sops) {
393 sma->sem_base[sop->sem_num].semval -= sop->sem_op;
400 /* Go through the pending queue for the indicated semaphore
401 * looking for tasks that can be completed.
403 static void update_queue (struct sem_array * sma)
406 struct sem_queue * q;
408 q = sma->sem_pending;
410 error = try_atomic_semop(sma, q->sops, q->nsops,
413 /* Does q->sleeper still need to sleep? */
416 remove_from_queue(sma,q);
417 q->status = IN_WAKEUP;
419 * Continue scanning. The next operation
420 * that must be checked depends on the type of the
421 * completed operation:
422 * - if the operation modified the array, then
423 * restart from the head of the queue and
424 * check for threads that might be waiting
425 * for semaphore values to become 0.
426 * - if the operation didn't modify the array,
427 * then just continue.
430 n = sma->sem_pending;
433 wake_up_process(q->sleeper);
434 /* hands-off: q will disappear immediately after
446 /* The following counts are associated to each semaphore:
447 * semncnt number of tasks waiting on semval being nonzero
448 * semzcnt number of tasks waiting on semval being zero
449 * This model assumes that a task waits on exactly one semaphore.
450 * Since semaphore operations are to be performed atomically, tasks actually
451 * wait on a whole sequence of semaphores simultaneously.
452 * The counts we return here are a rough approximation, but still
453 * warrant that semncnt+semzcnt>0 if the task is on the pending queue.
455 static int count_semncnt (struct sem_array * sma, ushort semnum)
458 struct sem_queue * q;
461 for (q = sma->sem_pending; q; q = q->next) {
462 struct sembuf * sops = q->sops;
463 int nsops = q->nsops;
465 for (i = 0; i < nsops; i++)
466 if (sops[i].sem_num == semnum
467 && (sops[i].sem_op < 0)
468 && !(sops[i].sem_flg & IPC_NOWAIT))
473 static int count_semzcnt (struct sem_array * sma, ushort semnum)
476 struct sem_queue * q;
479 for (q = sma->sem_pending; q; q = q->next) {
480 struct sembuf * sops = q->sops;
481 int nsops = q->nsops;
483 for (i = 0; i < nsops; i++)
484 if (sops[i].sem_num == semnum
485 && (sops[i].sem_op == 0)
486 && !(sops[i].sem_flg & IPC_NOWAIT))
492 /* Free a semaphore set. freeary() is called with sem_ids.mutex locked and
493 * the spinlock for this semaphore set hold. sem_ids.mutex remains locked
496 static void freeary (struct ipc_namespace *ns, struct sem_array *sma, int id)
502 /* Invalidate the existing undo structures for this semaphore set.
503 * (They will be freed without any further action in exit_sem()
504 * or during the next semop.)
506 for (un = sma->undo; un; un = un->id_next)
509 /* Wake up all pending processes and let them fail with EIDRM. */
510 q = sma->sem_pending;
513 /* lazy remove_from_queue: we are killing the whole queue */
516 q->status = IN_WAKEUP;
517 wake_up_process(q->sleeper); /* doesn't sleep */
519 q->status = -EIDRM; /* hands-off q */
523 /* Remove the semaphore set from the ID array*/
524 sma = sem_rmid(ns, id);
527 ns->used_sems -= sma->sem_nsems;
528 size = sizeof (*sma) + sma->sem_nsems * sizeof (struct sem);
529 security_sem_free(sma);
533 static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
537 return copy_to_user(buf, in, sizeof(*in));
542 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
544 out.sem_otime = in->sem_otime;
545 out.sem_ctime = in->sem_ctime;
546 out.sem_nsems = in->sem_nsems;
548 return copy_to_user(buf, &out, sizeof(out));
555 static int semctl_nolock(struct ipc_namespace *ns, int semid, int semnum,
556 int cmd, int version, union semun arg)
559 struct sem_array *sma;
565 struct seminfo seminfo;
568 err = security_sem_semctl(NULL, cmd);
572 memset(&seminfo,0,sizeof(seminfo));
573 seminfo.semmni = ns->sc_semmni;
574 seminfo.semmns = ns->sc_semmns;
575 seminfo.semmsl = ns->sc_semmsl;
576 seminfo.semopm = ns->sc_semopm;
577 seminfo.semvmx = SEMVMX;
578 seminfo.semmnu = SEMMNU;
579 seminfo.semmap = SEMMAP;
580 seminfo.semume = SEMUME;
581 mutex_lock(&sem_ids(ns).mutex);
582 if (cmd == SEM_INFO) {
583 seminfo.semusz = sem_ids(ns).in_use;
584 seminfo.semaem = ns->used_sems;
586 seminfo.semusz = SEMUSZ;
587 seminfo.semaem = SEMAEM;
589 max_id = sem_ids(ns).max_id;
590 mutex_unlock(&sem_ids(ns).mutex);
591 if (copy_to_user (arg.__buf, &seminfo, sizeof(struct seminfo)))
593 return (max_id < 0) ? 0: max_id;
597 struct semid64_ds tbuf;
600 if(semid >= sem_ids(ns).entries->size)
603 memset(&tbuf,0,sizeof(tbuf));
605 sma = sem_lock(ns, semid);
610 if (ipcperms (&sma->sem_perm, S_IRUGO))
613 err = security_sem_semctl(sma, cmd);
617 id = sem_buildid(ns, semid, sma->sem_perm.seq);
619 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
620 tbuf.sem_otime = sma->sem_otime;
621 tbuf.sem_ctime = sma->sem_ctime;
622 tbuf.sem_nsems = sma->sem_nsems;
624 if (copy_semid_to_user (arg.buf, &tbuf, version))
637 static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
638 int cmd, int version, union semun arg)
640 struct sem_array *sma;
643 ushort fast_sem_io[SEMMSL_FAST];
644 ushort* sem_io = fast_sem_io;
647 sma = sem_lock(ns, semid);
651 nsems = sma->sem_nsems;
654 if (sem_checkid(ns,sma,semid))
658 if (ipcperms (&sma->sem_perm, (cmd==SETVAL||cmd==SETALL)?S_IWUGO:S_IRUGO))
661 err = security_sem_semctl(sma, cmd);
669 ushort __user *array = arg.array;
672 if(nsems > SEMMSL_FAST) {
676 sem_io = ipc_alloc(sizeof(ushort)*nsems);
678 ipc_lock_by_ptr(&sma->sem_perm);
684 ipc_lock_by_ptr(&sma->sem_perm);
686 if (sma->sem_perm.deleted) {
693 for (i = 0; i < sma->sem_nsems; i++)
694 sem_io[i] = sma->sem_base[i].semval;
697 if(copy_to_user(array, sem_io, nsems*sizeof(ushort)))
709 if(nsems > SEMMSL_FAST) {
710 sem_io = ipc_alloc(sizeof(ushort)*nsems);
712 ipc_lock_by_ptr(&sma->sem_perm);
719 if (copy_from_user (sem_io, arg.array, nsems*sizeof(ushort))) {
720 ipc_lock_by_ptr(&sma->sem_perm);
727 for (i = 0; i < nsems; i++) {
728 if (sem_io[i] > SEMVMX) {
729 ipc_lock_by_ptr(&sma->sem_perm);
736 ipc_lock_by_ptr(&sma->sem_perm);
738 if (sma->sem_perm.deleted) {
744 for (i = 0; i < nsems; i++)
745 sma->sem_base[i].semval = sem_io[i];
746 for (un = sma->undo; un; un = un->id_next)
747 for (i = 0; i < nsems; i++)
749 sma->sem_ctime = get_seconds();
750 /* maybe some queued-up processes were waiting for this */
757 struct semid64_ds tbuf;
758 memset(&tbuf,0,sizeof(tbuf));
759 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
760 tbuf.sem_otime = sma->sem_otime;
761 tbuf.sem_ctime = sma->sem_ctime;
762 tbuf.sem_nsems = sma->sem_nsems;
764 if (copy_semid_to_user (arg.buf, &tbuf, version))
768 /* GETVAL, GETPID, GETNCTN, GETZCNT, SETVAL: fall-through */
771 if(semnum < 0 || semnum >= nsems)
774 curr = &sma->sem_base[semnum];
784 err = count_semncnt(sma,semnum);
787 err = count_semzcnt(sma,semnum);
794 if (val > SEMVMX || val < 0)
797 for (un = sma->undo; un; un = un->id_next)
798 un->semadj[semnum] = 0;
800 curr->sempid = current->tgid;
801 sma->sem_ctime = get_seconds();
802 /* maybe some queued-up processes were waiting for this */
811 if(sem_io != fast_sem_io)
812 ipc_free(sem_io, sizeof(ushort)*nsems);
822 static inline unsigned long copy_semid_from_user(struct sem_setbuf *out, void __user *buf, int version)
827 struct semid64_ds tbuf;
829 if(copy_from_user(&tbuf, buf, sizeof(tbuf)))
832 out->uid = tbuf.sem_perm.uid;
833 out->gid = tbuf.sem_perm.gid;
834 out->mode = tbuf.sem_perm.mode;
840 struct semid_ds tbuf_old;
842 if(copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
845 out->uid = tbuf_old.sem_perm.uid;
846 out->gid = tbuf_old.sem_perm.gid;
847 out->mode = tbuf_old.sem_perm.mode;
856 static int semctl_down(struct ipc_namespace *ns, int semid, int semnum,
857 int cmd, int version, union semun arg)
859 struct sem_array *sma;
861 struct sem_setbuf setbuf;
862 struct kern_ipc_perm *ipcp;
865 if(copy_semid_from_user (&setbuf, arg.buf, version))
868 sma = sem_lock(ns, semid);
872 if (sem_checkid(ns,sma,semid)) {
876 ipcp = &sma->sem_perm;
878 err = audit_ipc_obj(ipcp);
882 if (cmd == IPC_SET) {
883 err = audit_ipc_set_perm(0, setbuf.uid, setbuf.gid, setbuf.mode);
887 if (current->euid != ipcp->cuid &&
888 current->euid != ipcp->uid && !capable(CAP_SYS_ADMIN)) {
893 err = security_sem_semctl(sma, cmd);
899 freeary(ns, sma, semid);
903 ipcp->uid = setbuf.uid;
904 ipcp->gid = setbuf.gid;
905 ipcp->mode = (ipcp->mode & ~S_IRWXUGO)
906 | (setbuf.mode & S_IRWXUGO);
907 sma->sem_ctime = get_seconds();
923 asmlinkage long sys_semctl (int semid, int semnum, int cmd, union semun arg)
927 struct ipc_namespace *ns;
932 version = ipc_parse_version(&cmd);
933 ns = current->nsproxy->ipc_ns;
939 err = semctl_nolock(ns,semid,semnum,cmd,version,arg);
949 err = semctl_main(ns,semid,semnum,cmd,version,arg);
953 mutex_lock(&sem_ids(ns).mutex);
954 err = semctl_down(ns,semid,semnum,cmd,version,arg);
955 mutex_unlock(&sem_ids(ns).mutex);
962 static inline void lock_semundo(void)
964 struct sem_undo_list *undo_list;
966 undo_list = current->sysvsem.undo_list;
968 spin_lock(&undo_list->lock);
971 /* This code has an interaction with copy_semundo().
972 * Consider; two tasks are sharing the undo_list. task1
973 * acquires the undo_list lock in lock_semundo(). If task2 now
974 * exits before task1 releases the lock (by calling
975 * unlock_semundo()), then task1 will never call spin_unlock().
976 * This leave the sem_undo_list in a locked state. If task1 now creats task3
977 * and once again shares the sem_undo_list, the sem_undo_list will still be
978 * locked, and future SEM_UNDO operations will deadlock. This case is
979 * dealt with in copy_semundo() by having it reinitialize the spin lock when
980 * the refcnt goes from 1 to 2.
982 static inline void unlock_semundo(void)
984 struct sem_undo_list *undo_list;
986 undo_list = current->sysvsem.undo_list;
988 spin_unlock(&undo_list->lock);
992 /* If the task doesn't already have a undo_list, then allocate one
993 * here. We guarantee there is only one thread using this undo list,
994 * and current is THE ONE
996 * If this allocation and assignment succeeds, but later
997 * portions of this code fail, there is no need to free the sem_undo_list.
998 * Just let it stay associated with the task, and it'll be freed later
1001 * This can block, so callers must hold no locks.
1003 static inline int get_undo_list(struct sem_undo_list **undo_listp)
1005 struct sem_undo_list *undo_list;
1007 undo_list = current->sysvsem.undo_list;
1009 undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
1010 if (undo_list == NULL)
1012 spin_lock_init(&undo_list->lock);
1013 atomic_set(&undo_list->refcnt, 1);
1014 current->sysvsem.undo_list = undo_list;
1016 *undo_listp = undo_list;
1020 static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1022 struct sem_undo **last, *un;
1024 last = &ulp->proc_list;
1027 if(un->semid==semid)
1030 *last=un->proc_next;
1033 last=&un->proc_next;
1040 static struct sem_undo *find_undo(struct ipc_namespace *ns, int semid)
1042 struct sem_array *sma;
1043 struct sem_undo_list *ulp;
1044 struct sem_undo *un, *new;
1048 error = get_undo_list(&ulp);
1050 return ERR_PTR(error);
1053 un = lookup_undo(ulp, semid);
1055 if (likely(un!=NULL))
1058 /* no undo structure around - allocate one. */
1059 sma = sem_lock(ns, semid);
1060 un = ERR_PTR(-EINVAL);
1063 un = ERR_PTR(-EIDRM);
1064 if (sem_checkid(ns,sma,semid)) {
1068 nsems = sma->sem_nsems;
1069 ipc_rcu_getref(sma);
1072 new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1074 ipc_lock_by_ptr(&sma->sem_perm);
1075 ipc_rcu_putref(sma);
1077 return ERR_PTR(-ENOMEM);
1079 new->semadj = (short *) &new[1];
1083 un = lookup_undo(ulp, semid);
1087 ipc_lock_by_ptr(&sma->sem_perm);
1088 ipc_rcu_putref(sma);
1092 ipc_lock_by_ptr(&sma->sem_perm);
1093 ipc_rcu_putref(sma);
1094 if (sma->sem_perm.deleted) {
1098 un = ERR_PTR(-EIDRM);
1101 new->proc_next = ulp->proc_list;
1102 ulp->proc_list = new;
1103 new->id_next = sma->undo;
1112 asmlinkage long sys_semtimedop(int semid, struct sembuf __user *tsops,
1113 unsigned nsops, const struct timespec __user *timeout)
1115 int error = -EINVAL;
1116 struct sem_array *sma;
1117 struct sembuf fast_sops[SEMOPM_FAST];
1118 struct sembuf* sops = fast_sops, *sop;
1119 struct sem_undo *un;
1120 int undos = 0, alter = 0, max;
1121 struct sem_queue queue;
1122 unsigned long jiffies_left = 0;
1123 struct ipc_namespace *ns;
1125 ns = current->nsproxy->ipc_ns;
1127 if (nsops < 1 || semid < 0)
1129 if (nsops > ns->sc_semopm)
1131 if(nsops > SEMOPM_FAST) {
1132 sops = kmalloc(sizeof(*sops)*nsops,GFP_KERNEL);
1136 if (copy_from_user (sops, tsops, nsops * sizeof(*tsops))) {
1141 struct timespec _timeout;
1142 if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1146 if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1147 _timeout.tv_nsec >= 1000000000L) {
1151 jiffies_left = timespec_to_jiffies(&_timeout);
1154 for (sop = sops; sop < sops + nsops; sop++) {
1155 if (sop->sem_num >= max)
1157 if (sop->sem_flg & SEM_UNDO)
1159 if (sop->sem_op != 0)
1165 un = find_undo(ns, semid);
1167 error = PTR_ERR(un);
1173 sma = sem_lock(ns, semid);
1178 if (sem_checkid(ns,sma,semid))
1179 goto out_unlock_free;
1181 * semid identifies are not unique - find_undo may have
1182 * allocated an undo structure, it was invalidated by an RMID
1183 * and now a new array with received the same id. Check and retry.
1185 if (un && un->semid == -1) {
1190 if (max >= sma->sem_nsems)
1191 goto out_unlock_free;
1194 if (ipcperms(&sma->sem_perm, alter ? S_IWUGO : S_IRUGO))
1195 goto out_unlock_free;
1197 error = security_sem_semop(sma, sops, nsops, alter);
1199 goto out_unlock_free;
1201 error = try_atomic_semop (sma, sops, nsops, un, current->tgid);
1203 if (alter && error == 0)
1205 goto out_unlock_free;
1208 /* We need to sleep on this operation, so we put the current
1209 * task into the pending queue and go to sleep.
1214 queue.nsops = nsops;
1216 queue.pid = current->tgid;
1218 queue.alter = alter;
1220 append_to_queue(sma ,&queue);
1222 prepend_to_queue(sma ,&queue);
1224 queue.status = -EINTR;
1225 queue.sleeper = current;
1226 current->state = TASK_INTERRUPTIBLE;
1230 jiffies_left = schedule_timeout(jiffies_left);
1234 error = queue.status;
1235 while(unlikely(error == IN_WAKEUP)) {
1237 error = queue.status;
1240 if (error != -EINTR) {
1241 /* fast path: update_queue already obtained all requested
1246 sma = sem_lock(ns, semid);
1248 BUG_ON(queue.prev != NULL);
1254 * If queue.status != -EINTR we are woken up by another process
1256 error = queue.status;
1257 if (error != -EINTR) {
1258 goto out_unlock_free;
1262 * If an interrupt occurred we have to clean up the queue
1264 if (timeout && jiffies_left == 0)
1266 remove_from_queue(sma,&queue);
1267 goto out_unlock_free;
1272 if(sops != fast_sops)
1277 asmlinkage long sys_semop (int semid, struct sembuf __user *tsops, unsigned nsops)
1279 return sys_semtimedop(semid, tsops, nsops, NULL);
1282 /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
1283 * parent and child tasks.
1285 * See the notes above unlock_semundo() regarding the spin_lock_init()
1286 * in this code. Initialize the undo_list->lock here instead of get_undo_list()
1287 * because of the reasoning in the comment above unlock_semundo.
1290 int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
1292 struct sem_undo_list *undo_list;
1295 if (clone_flags & CLONE_SYSVSEM) {
1296 error = get_undo_list(&undo_list);
1299 atomic_inc(&undo_list->refcnt);
1300 tsk->sysvsem.undo_list = undo_list;
1302 tsk->sysvsem.undo_list = NULL;
1308 * add semadj values to semaphores, free undo structures.
1309 * undo structures are not freed when semaphore arrays are destroyed
1310 * so some of them may be out of date.
1311 * IMPLEMENTATION NOTE: There is some confusion over whether the
1312 * set of adjustments that needs to be done should be done in an atomic
1313 * manner or not. That is, if we are attempting to decrement the semval
1314 * should we queue up and wait until we can do so legally?
1315 * The original implementation attempted to do this (queue and wait).
1316 * The current implementation does not do so. The POSIX standard
1317 * and SVID should be consulted to determine what behavior is mandated.
1319 void exit_sem(struct task_struct *tsk)
1321 struct sem_undo_list *undo_list;
1322 struct sem_undo *u, **up;
1323 struct ipc_namespace *ns;
1325 undo_list = tsk->sysvsem.undo_list;
1329 if (!atomic_dec_and_test(&undo_list->refcnt))
1332 ns = tsk->nsproxy->ipc_ns;
1333 /* There's no need to hold the semundo list lock, as current
1334 * is the last task exiting for this undo list.
1336 for (up = &undo_list->proc_list; (u = *up); *up = u->proc_next, kfree(u)) {
1337 struct sem_array *sma;
1339 struct sem_undo *un, **unp;
1346 sma = sem_lock(ns, semid);
1353 BUG_ON(sem_checkid(ns,sma,u->semid));
1355 /* remove u from the sma->undo list */
1356 for (unp = &sma->undo; (un = *unp); unp = &un->id_next) {
1360 printk ("exit_sem undo list error id=%d\n", u->semid);
1364 /* perform adjustments registered in u */
1365 nsems = sma->sem_nsems;
1366 for (i = 0; i < nsems; i++) {
1367 struct sem * semaphore = &sma->sem_base[i];
1369 semaphore->semval += u->semadj[i];
1371 * Range checks of the new semaphore value,
1372 * not defined by sus:
1373 * - Some unices ignore the undo entirely
1374 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
1375 * - some cap the value (e.g. FreeBSD caps
1376 * at 0, but doesn't enforce SEMVMX)
1378 * Linux caps the semaphore value, both at 0
1381 * Manfred <manfred@colorfullife.com>
1383 if (semaphore->semval < 0)
1384 semaphore->semval = 0;
1385 if (semaphore->semval > SEMVMX)
1386 semaphore->semval = SEMVMX;
1387 semaphore->sempid = current->tgid;
1390 sma->sem_otime = get_seconds();
1391 /* maybe some queued-up processes were waiting for this */
1399 #ifdef CONFIG_PROC_FS
1400 static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
1402 struct sem_array *sma = it;
1404 return seq_printf(s,
1405 "%10d %10d %4o %10lu %5u %5u %5u %5u %10lu %10lu\n",