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/rwsem.h>
84 #include <linux/nsproxy.h>
85 #include <linux/ipc_namespace.h>
87 #include <asm/uaccess.h>
90 #define sem_ids(ns) ((ns)->ids[IPC_SEM_IDS])
92 #define sem_unlock(sma) ipc_unlock(&(sma)->sem_perm)
93 #define sem_checkid(sma, semid) ipc_checkid(&sma->sem_perm, semid)
94 #define sem_buildid(id, seq) ipc_buildid(id, seq)
96 static int newary(struct ipc_namespace *, struct ipc_params *);
97 static void freeary(struct ipc_namespace *, struct sem_array *);
99 static int sysvipc_sem_proc_show(struct seq_file *s, void *it);
102 #define SEMMSL_FAST 256 /* 512 bytes on stack */
103 #define SEMOPM_FAST 64 /* ~ 372 bytes on stack */
106 * linked list protection:
108 * sem_array.sem_pending{,last},
109 * sem_array.sem_undo: sem_lock() for read/write
110 * sem_undo.proc_next: only "current" is allowed to read/write that field.
114 #define sc_semmsl sem_ctls[0]
115 #define sc_semmns sem_ctls[1]
116 #define sc_semopm sem_ctls[2]
117 #define sc_semmni sem_ctls[3]
119 void sem_init_ns(struct ipc_namespace *ns)
121 ns->sc_semmsl = SEMMSL;
122 ns->sc_semmns = SEMMNS;
123 ns->sc_semopm = SEMOPM;
124 ns->sc_semmni = SEMMNI;
126 ipc_init_ids(&ns->ids[IPC_SEM_IDS]);
130 void sem_exit_ns(struct ipc_namespace *ns)
132 struct sem_array *sma;
133 struct kern_ipc_perm *perm;
137 down_write(&sem_ids(ns).rw_mutex);
139 in_use = sem_ids(ns).in_use;
141 for (total = 0, next_id = 0; total < in_use; next_id++) {
142 perm = idr_find(&sem_ids(ns).ipcs_idr, next_id);
145 ipc_lock_by_ptr(perm);
146 sma = container_of(perm, struct sem_array, sem_perm);
150 up_write(&sem_ids(ns).rw_mutex);
154 void __init sem_init (void)
156 sem_init_ns(&init_ipc_ns);
157 ipc_init_proc_interface("sysvipc/sem",
158 " key semid perms nsems uid gid cuid cgid otime ctime\n",
159 IPC_SEM_IDS, sysvipc_sem_proc_show);
163 * This routine is called in the paths where the rw_mutex is held to protect
164 * access to the idr tree.
166 static inline struct sem_array *sem_lock_check_down(struct ipc_namespace *ns,
169 struct kern_ipc_perm *ipcp = ipc_lock_check_down(&sem_ids(ns), id);
172 return (struct sem_array *)ipcp;
174 return container_of(ipcp, struct sem_array, sem_perm);
178 * sem_lock_(check_) routines are called in the paths where the rw_mutex
181 static inline struct sem_array *sem_lock(struct ipc_namespace *ns, int id)
183 struct kern_ipc_perm *ipcp = ipc_lock(&sem_ids(ns), id);
186 return (struct sem_array *)ipcp;
188 return container_of(ipcp, struct sem_array, sem_perm);
191 static inline struct sem_array *sem_lock_check(struct ipc_namespace *ns,
194 struct kern_ipc_perm *ipcp = ipc_lock_check(&sem_ids(ns), id);
197 return (struct sem_array *)ipcp;
199 return container_of(ipcp, struct sem_array, sem_perm);
202 static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
204 ipc_rmid(&sem_ids(ns), &s->sem_perm);
208 * Lockless wakeup algorithm:
209 * Without the check/retry algorithm a lockless wakeup is possible:
210 * - queue.status is initialized to -EINTR before blocking.
211 * - wakeup is performed by
212 * * unlinking the queue entry from sma->sem_pending
213 * * setting queue.status to IN_WAKEUP
214 * This is the notification for the blocked thread that a
215 * result value is imminent.
216 * * call wake_up_process
217 * * set queue.status to the final value.
218 * - the previously blocked thread checks queue.status:
219 * * if it's IN_WAKEUP, then it must wait until the value changes
220 * * if it's not -EINTR, then the operation was completed by
221 * update_queue. semtimedop can return queue.status without
222 * performing any operation on the sem array.
223 * * otherwise it must acquire the spinlock and check what's up.
225 * The two-stage algorithm is necessary to protect against the following
227 * - if queue.status is set after wake_up_process, then the woken up idle
228 * thread could race forward and try (and fail) to acquire sma->lock
229 * before update_queue had a chance to set queue.status
230 * - if queue.status is written before wake_up_process and if the
231 * blocked process is woken up by a signal between writing
232 * queue.status and the wake_up_process, then the woken up
233 * process could return from semtimedop and die by calling
234 * sys_exit before wake_up_process is called. Then wake_up_process
235 * will oops, because the task structure is already invalid.
236 * (yes, this happened on s390 with sysv msg).
242 * newary - Create a new semaphore set
244 * @params: ptr to the structure that contains key, semflg and nsems
246 * Called with sem_ids.rw_mutex held (as a writer)
249 static int newary(struct ipc_namespace *ns, struct ipc_params *params)
253 struct sem_array *sma;
255 key_t key = params->key;
256 int nsems = params->u.nsems;
257 int semflg = params->flg;
261 if (ns->used_sems + nsems > ns->sc_semmns)
264 size = sizeof (*sma) + nsems * sizeof (struct sem);
265 sma = ipc_rcu_alloc(size);
269 memset (sma, 0, size);
271 sma->sem_perm.mode = (semflg & S_IRWXUGO);
272 sma->sem_perm.key = key;
274 sma->sem_perm.security = NULL;
275 retval = security_sem_alloc(sma);
281 id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
283 security_sem_free(sma);
287 ns->used_sems += nsems;
289 sma->sem_perm.id = sem_buildid(id, sma->sem_perm.seq);
290 sma->sem_base = (struct sem *) &sma[1];
291 /* sma->sem_pending = NULL; */
292 sma->sem_pending_last = &sma->sem_pending;
293 /* sma->undo = NULL; */
294 sma->sem_nsems = nsems;
295 sma->sem_ctime = get_seconds();
298 return sma->sem_perm.id;
303 * Called with sem_ids.rw_mutex and ipcp locked.
305 static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg)
307 struct sem_array *sma;
309 sma = container_of(ipcp, struct sem_array, sem_perm);
310 return security_sem_associate(sma, semflg);
314 * Called with sem_ids.rw_mutex and ipcp locked.
316 static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
317 struct ipc_params *params)
319 struct sem_array *sma;
321 sma = container_of(ipcp, struct sem_array, sem_perm);
322 if (params->u.nsems > sma->sem_nsems)
328 asmlinkage long sys_semget(key_t key, int nsems, int semflg)
330 struct ipc_namespace *ns;
331 struct ipc_ops sem_ops;
332 struct ipc_params sem_params;
334 ns = current->nsproxy->ipc_ns;
336 if (nsems < 0 || nsems > ns->sc_semmsl)
339 sem_ops.getnew = newary;
340 sem_ops.associate = sem_security;
341 sem_ops.more_checks = sem_more_checks;
343 sem_params.key = key;
344 sem_params.flg = semflg;
345 sem_params.u.nsems = nsems;
347 return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
350 /* Manage the doubly linked list sma->sem_pending as a FIFO:
351 * insert new queue elements at the tail sma->sem_pending_last.
353 static inline void append_to_queue (struct sem_array * sma,
354 struct sem_queue * q)
356 *(q->prev = sma->sem_pending_last) = q;
357 *(sma->sem_pending_last = &q->next) = NULL;
360 static inline void prepend_to_queue (struct sem_array * sma,
361 struct sem_queue * q)
363 q->next = sma->sem_pending;
364 *(q->prev = &sma->sem_pending) = q;
366 q->next->prev = &q->next;
367 else /* sma->sem_pending_last == &sma->sem_pending */
368 sma->sem_pending_last = &q->next;
371 static inline void remove_from_queue (struct sem_array * sma,
372 struct sem_queue * q)
374 *(q->prev) = q->next;
376 q->next->prev = q->prev;
377 else /* sma->sem_pending_last == &q->next */
378 sma->sem_pending_last = q->prev;
379 q->prev = NULL; /* mark as removed */
383 * Determine whether a sequence of semaphore operations would succeed
384 * all at once. Return 0 if yes, 1 if need to sleep, else return error code.
387 static int try_atomic_semop (struct sem_array * sma, struct sembuf * sops,
388 int nsops, struct sem_undo *un, int pid)
394 for (sop = sops; sop < sops + nsops; sop++) {
395 curr = sma->sem_base + sop->sem_num;
396 sem_op = sop->sem_op;
397 result = curr->semval;
399 if (!sem_op && result)
407 if (sop->sem_flg & SEM_UNDO) {
408 int undo = un->semadj[sop->sem_num] - sem_op;
410 * Exceeding the undo range is an error.
412 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
415 curr->semval = result;
419 while (sop >= sops) {
420 sma->sem_base[sop->sem_num].sempid = pid;
421 if (sop->sem_flg & SEM_UNDO)
422 un->semadj[sop->sem_num] -= sop->sem_op;
426 sma->sem_otime = get_seconds();
434 if (sop->sem_flg & IPC_NOWAIT)
441 while (sop >= sops) {
442 sma->sem_base[sop->sem_num].semval -= sop->sem_op;
449 /* Go through the pending queue for the indicated semaphore
450 * looking for tasks that can be completed.
452 static void update_queue (struct sem_array * sma)
455 struct sem_queue * q;
457 q = sma->sem_pending;
459 error = try_atomic_semop(sma, q->sops, q->nsops,
462 /* Does q->sleeper still need to sleep? */
465 remove_from_queue(sma,q);
466 q->status = IN_WAKEUP;
468 * Continue scanning. The next operation
469 * that must be checked depends on the type of the
470 * completed operation:
471 * - if the operation modified the array, then
472 * restart from the head of the queue and
473 * check for threads that might be waiting
474 * for semaphore values to become 0.
475 * - if the operation didn't modify the array,
476 * then just continue.
479 n = sma->sem_pending;
482 wake_up_process(q->sleeper);
483 /* hands-off: q will disappear immediately after
495 /* The following counts are associated to each semaphore:
496 * semncnt number of tasks waiting on semval being nonzero
497 * semzcnt number of tasks waiting on semval being zero
498 * This model assumes that a task waits on exactly one semaphore.
499 * Since semaphore operations are to be performed atomically, tasks actually
500 * wait on a whole sequence of semaphores simultaneously.
501 * The counts we return here are a rough approximation, but still
502 * warrant that semncnt+semzcnt>0 if the task is on the pending queue.
504 static int count_semncnt (struct sem_array * sma, ushort semnum)
507 struct sem_queue * q;
510 for (q = sma->sem_pending; q; q = q->next) {
511 struct sembuf * sops = q->sops;
512 int nsops = q->nsops;
514 for (i = 0; i < nsops; i++)
515 if (sops[i].sem_num == semnum
516 && (sops[i].sem_op < 0)
517 && !(sops[i].sem_flg & IPC_NOWAIT))
522 static int count_semzcnt (struct sem_array * sma, ushort semnum)
525 struct sem_queue * q;
528 for (q = sma->sem_pending; q; q = q->next) {
529 struct sembuf * sops = q->sops;
530 int nsops = q->nsops;
532 for (i = 0; i < nsops; i++)
533 if (sops[i].sem_num == semnum
534 && (sops[i].sem_op == 0)
535 && !(sops[i].sem_flg & IPC_NOWAIT))
541 /* Free a semaphore set. freeary() is called with sem_ids.rw_mutex locked
542 * as a writer and the spinlock for this semaphore set hold. sem_ids.rw_mutex
543 * remains locked on exit.
545 static void freeary(struct ipc_namespace *ns, struct sem_array *sma)
550 /* Invalidate the existing undo structures for this semaphore set.
551 * (They will be freed without any further action in exit_sem()
552 * or during the next semop.)
554 for (un = sma->undo; un; un = un->id_next)
557 /* Wake up all pending processes and let them fail with EIDRM. */
558 q = sma->sem_pending;
561 /* lazy remove_from_queue: we are killing the whole queue */
564 q->status = IN_WAKEUP;
565 wake_up_process(q->sleeper); /* doesn't sleep */
567 q->status = -EIDRM; /* hands-off q */
571 /* Remove the semaphore set from the IDR */
575 ns->used_sems -= sma->sem_nsems;
576 security_sem_free(sma);
580 static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
584 return copy_to_user(buf, in, sizeof(*in));
589 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
591 out.sem_otime = in->sem_otime;
592 out.sem_ctime = in->sem_ctime;
593 out.sem_nsems = in->sem_nsems;
595 return copy_to_user(buf, &out, sizeof(out));
602 static int semctl_nolock(struct ipc_namespace *ns, int semid,
603 int cmd, int version, union semun arg)
606 struct sem_array *sma;
612 struct seminfo seminfo;
615 err = security_sem_semctl(NULL, cmd);
619 memset(&seminfo,0,sizeof(seminfo));
620 seminfo.semmni = ns->sc_semmni;
621 seminfo.semmns = ns->sc_semmns;
622 seminfo.semmsl = ns->sc_semmsl;
623 seminfo.semopm = ns->sc_semopm;
624 seminfo.semvmx = SEMVMX;
625 seminfo.semmnu = SEMMNU;
626 seminfo.semmap = SEMMAP;
627 seminfo.semume = SEMUME;
628 down_read(&sem_ids(ns).rw_mutex);
629 if (cmd == SEM_INFO) {
630 seminfo.semusz = sem_ids(ns).in_use;
631 seminfo.semaem = ns->used_sems;
633 seminfo.semusz = SEMUSZ;
634 seminfo.semaem = SEMAEM;
636 max_id = ipc_get_maxid(&sem_ids(ns));
637 up_read(&sem_ids(ns).rw_mutex);
638 if (copy_to_user (arg.__buf, &seminfo, sizeof(struct seminfo)))
640 return (max_id < 0) ? 0: max_id;
645 struct semid64_ds tbuf;
648 if (cmd == SEM_STAT) {
649 sma = sem_lock(ns, semid);
652 id = sma->sem_perm.id;
654 sma = sem_lock_check(ns, semid);
661 if (ipcperms (&sma->sem_perm, S_IRUGO))
664 err = security_sem_semctl(sma, cmd);
668 memset(&tbuf, 0, sizeof(tbuf));
670 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
671 tbuf.sem_otime = sma->sem_otime;
672 tbuf.sem_ctime = sma->sem_ctime;
673 tbuf.sem_nsems = sma->sem_nsems;
675 if (copy_semid_to_user (arg.buf, &tbuf, version))
688 static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
689 int cmd, int version, union semun arg)
691 struct sem_array *sma;
694 ushort fast_sem_io[SEMMSL_FAST];
695 ushort* sem_io = fast_sem_io;
698 sma = sem_lock_check(ns, semid);
702 nsems = sma->sem_nsems;
705 if (ipcperms (&sma->sem_perm, (cmd==SETVAL||cmd==SETALL)?S_IWUGO:S_IRUGO))
708 err = security_sem_semctl(sma, cmd);
716 ushort __user *array = arg.array;
719 if(nsems > SEMMSL_FAST) {
723 sem_io = ipc_alloc(sizeof(ushort)*nsems);
725 ipc_lock_by_ptr(&sma->sem_perm);
731 ipc_lock_by_ptr(&sma->sem_perm);
733 if (sma->sem_perm.deleted) {
740 for (i = 0; i < sma->sem_nsems; i++)
741 sem_io[i] = sma->sem_base[i].semval;
744 if(copy_to_user(array, sem_io, nsems*sizeof(ushort)))
756 if(nsems > SEMMSL_FAST) {
757 sem_io = ipc_alloc(sizeof(ushort)*nsems);
759 ipc_lock_by_ptr(&sma->sem_perm);
766 if (copy_from_user (sem_io, arg.array, nsems*sizeof(ushort))) {
767 ipc_lock_by_ptr(&sma->sem_perm);
774 for (i = 0; i < nsems; i++) {
775 if (sem_io[i] > SEMVMX) {
776 ipc_lock_by_ptr(&sma->sem_perm);
783 ipc_lock_by_ptr(&sma->sem_perm);
785 if (sma->sem_perm.deleted) {
791 for (i = 0; i < nsems; i++)
792 sma->sem_base[i].semval = sem_io[i];
793 for (un = sma->undo; un; un = un->id_next)
794 for (i = 0; i < nsems; i++)
796 sma->sem_ctime = get_seconds();
797 /* maybe some queued-up processes were waiting for this */
802 /* GETVAL, GETPID, GETNCTN, GETZCNT, SETVAL: fall-through */
805 if(semnum < 0 || semnum >= nsems)
808 curr = &sma->sem_base[semnum];
818 err = count_semncnt(sma,semnum);
821 err = count_semzcnt(sma,semnum);
828 if (val > SEMVMX || val < 0)
831 for (un = sma->undo; un; un = un->id_next)
832 un->semadj[semnum] = 0;
834 curr->sempid = task_tgid_vnr(current);
835 sma->sem_ctime = get_seconds();
836 /* maybe some queued-up processes were waiting for this */
845 if(sem_io != fast_sem_io)
846 ipc_free(sem_io, sizeof(ushort)*nsems);
856 static inline unsigned long copy_semid_from_user(struct sem_setbuf *out, void __user *buf, int version)
861 struct semid64_ds tbuf;
863 if(copy_from_user(&tbuf, buf, sizeof(tbuf)))
866 out->uid = tbuf.sem_perm.uid;
867 out->gid = tbuf.sem_perm.gid;
868 out->mode = tbuf.sem_perm.mode;
874 struct semid_ds tbuf_old;
876 if(copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
879 out->uid = tbuf_old.sem_perm.uid;
880 out->gid = tbuf_old.sem_perm.gid;
881 out->mode = tbuf_old.sem_perm.mode;
890 static int semctl_down(struct ipc_namespace *ns, int semid, int semnum,
891 int cmd, int version, union semun arg)
893 struct sem_array *sma;
895 struct sem_setbuf uninitialized_var(setbuf);
896 struct kern_ipc_perm *ipcp;
899 if(copy_semid_from_user (&setbuf, arg.buf, version))
902 sma = sem_lock_check_down(ns, semid);
906 ipcp = &sma->sem_perm;
908 err = audit_ipc_obj(ipcp);
912 if (cmd == IPC_SET) {
913 err = audit_ipc_set_perm(0, setbuf.uid, setbuf.gid, setbuf.mode);
917 if (current->euid != ipcp->cuid &&
918 current->euid != ipcp->uid && !capable(CAP_SYS_ADMIN)) {
923 err = security_sem_semctl(sma, cmd);
933 ipcp->uid = setbuf.uid;
934 ipcp->gid = setbuf.gid;
935 ipcp->mode = (ipcp->mode & ~S_IRWXUGO)
936 | (setbuf.mode & S_IRWXUGO);
937 sma->sem_ctime = get_seconds();
953 asmlinkage long sys_semctl (int semid, int semnum, int cmd, union semun arg)
957 struct ipc_namespace *ns;
962 version = ipc_parse_version(&cmd);
963 ns = current->nsproxy->ipc_ns;
970 err = semctl_nolock(ns, semid, cmd, version, arg);
979 err = semctl_main(ns,semid,semnum,cmd,version,arg);
983 down_write(&sem_ids(ns).rw_mutex);
984 err = semctl_down(ns,semid,semnum,cmd,version,arg);
985 up_write(&sem_ids(ns).rw_mutex);
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);
1052 spin_lock(&ulp->lock);
1053 un = lookup_undo(ulp, semid);
1054 spin_unlock(&ulp->lock);
1055 if (likely(un!=NULL))
1058 /* no undo structure around - allocate one. */
1059 sma = sem_lock_check(ns, semid);
1061 return ERR_PTR(PTR_ERR(sma));
1063 nsems = sma->sem_nsems;
1064 ipc_rcu_getref(sma);
1067 new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1069 ipc_lock_by_ptr(&sma->sem_perm);
1070 ipc_rcu_putref(sma);
1072 return ERR_PTR(-ENOMEM);
1074 new->semadj = (short *) &new[1];
1077 spin_lock(&ulp->lock);
1078 un = lookup_undo(ulp, semid);
1080 spin_unlock(&ulp->lock);
1082 ipc_lock_by_ptr(&sma->sem_perm);
1083 ipc_rcu_putref(sma);
1087 ipc_lock_by_ptr(&sma->sem_perm);
1088 ipc_rcu_putref(sma);
1089 if (sma->sem_perm.deleted) {
1091 spin_unlock(&ulp->lock);
1093 un = ERR_PTR(-EIDRM);
1096 new->proc_next = ulp->proc_list;
1097 ulp->proc_list = new;
1098 new->id_next = sma->undo;
1102 spin_unlock(&ulp->lock);
1107 asmlinkage long sys_semtimedop(int semid, struct sembuf __user *tsops,
1108 unsigned nsops, const struct timespec __user *timeout)
1110 int error = -EINVAL;
1111 struct sem_array *sma;
1112 struct sembuf fast_sops[SEMOPM_FAST];
1113 struct sembuf* sops = fast_sops, *sop;
1114 struct sem_undo *un;
1115 int undos = 0, alter = 0, max;
1116 struct sem_queue queue;
1117 unsigned long jiffies_left = 0;
1118 struct ipc_namespace *ns;
1120 ns = current->nsproxy->ipc_ns;
1122 if (nsops < 1 || semid < 0)
1124 if (nsops > ns->sc_semopm)
1126 if(nsops > SEMOPM_FAST) {
1127 sops = kmalloc(sizeof(*sops)*nsops,GFP_KERNEL);
1131 if (copy_from_user (sops, tsops, nsops * sizeof(*tsops))) {
1136 struct timespec _timeout;
1137 if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1141 if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1142 _timeout.tv_nsec >= 1000000000L) {
1146 jiffies_left = timespec_to_jiffies(&_timeout);
1149 for (sop = sops; sop < sops + nsops; sop++) {
1150 if (sop->sem_num >= max)
1152 if (sop->sem_flg & SEM_UNDO)
1154 if (sop->sem_op != 0)
1160 un = find_undo(ns, semid);
1162 error = PTR_ERR(un);
1168 sma = sem_lock_check(ns, semid);
1170 error = PTR_ERR(sma);
1175 * semid identifiers are not unique - find_undo may have
1176 * allocated an undo structure, it was invalidated by an RMID
1177 * and now a new array with received the same id. Check and retry.
1179 if (un && un->semid == -1) {
1184 if (max >= sma->sem_nsems)
1185 goto out_unlock_free;
1188 if (ipcperms(&sma->sem_perm, alter ? S_IWUGO : S_IRUGO))
1189 goto out_unlock_free;
1191 error = security_sem_semop(sma, sops, nsops, alter);
1193 goto out_unlock_free;
1195 error = try_atomic_semop (sma, sops, nsops, un, task_tgid_vnr(current));
1197 if (alter && error == 0)
1199 goto out_unlock_free;
1202 /* We need to sleep on this operation, so we put the current
1203 * task into the pending queue and go to sleep.
1208 queue.nsops = nsops;
1210 queue.pid = task_tgid_vnr(current);
1212 queue.alter = alter;
1214 append_to_queue(sma ,&queue);
1216 prepend_to_queue(sma ,&queue);
1218 queue.status = -EINTR;
1219 queue.sleeper = current;
1220 current->state = TASK_INTERRUPTIBLE;
1224 jiffies_left = schedule_timeout(jiffies_left);
1228 error = queue.status;
1229 while(unlikely(error == IN_WAKEUP)) {
1231 error = queue.status;
1234 if (error != -EINTR) {
1235 /* fast path: update_queue already obtained all requested
1240 sma = sem_lock(ns, semid);
1242 BUG_ON(queue.prev != NULL);
1248 * If queue.status != -EINTR we are woken up by another process
1250 error = queue.status;
1251 if (error != -EINTR) {
1252 goto out_unlock_free;
1256 * If an interrupt occurred we have to clean up the queue
1258 if (timeout && jiffies_left == 0)
1260 remove_from_queue(sma,&queue);
1261 goto out_unlock_free;
1266 if(sops != fast_sops)
1271 asmlinkage long sys_semop (int semid, struct sembuf __user *tsops, unsigned nsops)
1273 return sys_semtimedop(semid, tsops, nsops, NULL);
1276 /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
1277 * parent and child tasks.
1280 int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
1282 struct sem_undo_list *undo_list;
1285 if (clone_flags & CLONE_SYSVSEM) {
1286 error = get_undo_list(&undo_list);
1289 atomic_inc(&undo_list->refcnt);
1290 tsk->sysvsem.undo_list = undo_list;
1292 tsk->sysvsem.undo_list = NULL;
1298 * add semadj values to semaphores, free undo structures.
1299 * undo structures are not freed when semaphore arrays are destroyed
1300 * so some of them may be out of date.
1301 * IMPLEMENTATION NOTE: There is some confusion over whether the
1302 * set of adjustments that needs to be done should be done in an atomic
1303 * manner or not. That is, if we are attempting to decrement the semval
1304 * should we queue up and wait until we can do so legally?
1305 * The original implementation attempted to do this (queue and wait).
1306 * The current implementation does not do so. The POSIX standard
1307 * and SVID should be consulted to determine what behavior is mandated.
1309 void exit_sem(struct task_struct *tsk)
1311 struct sem_undo_list *undo_list;
1312 struct sem_undo *u, **up;
1313 struct ipc_namespace *ns;
1315 undo_list = tsk->sysvsem.undo_list;
1319 if (!atomic_dec_and_test(&undo_list->refcnt))
1322 ns = tsk->nsproxy->ipc_ns;
1323 /* There's no need to hold the semundo list lock, as current
1324 * is the last task exiting for this undo list.
1326 for (up = &undo_list->proc_list; (u = *up); *up = u->proc_next, kfree(u)) {
1327 struct sem_array *sma;
1329 struct sem_undo *un, **unp;
1336 sma = sem_lock(ns, semid);
1343 BUG_ON(sem_checkid(sma, u->semid));
1345 /* remove u from the sma->undo list */
1346 for (unp = &sma->undo; (un = *unp); unp = &un->id_next) {
1350 printk ("exit_sem undo list error id=%d\n", u->semid);
1354 /* perform adjustments registered in u */
1355 nsems = sma->sem_nsems;
1356 for (i = 0; i < nsems; i++) {
1357 struct sem * semaphore = &sma->sem_base[i];
1359 semaphore->semval += u->semadj[i];
1361 * Range checks of the new semaphore value,
1362 * not defined by sus:
1363 * - Some unices ignore the undo entirely
1364 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
1365 * - some cap the value (e.g. FreeBSD caps
1366 * at 0, but doesn't enforce SEMVMX)
1368 * Linux caps the semaphore value, both at 0
1371 * Manfred <manfred@colorfullife.com>
1373 if (semaphore->semval < 0)
1374 semaphore->semval = 0;
1375 if (semaphore->semval > SEMVMX)
1376 semaphore->semval = SEMVMX;
1377 semaphore->sempid = task_tgid_vnr(current);
1380 sma->sem_otime = get_seconds();
1381 /* maybe some queued-up processes were waiting for this */
1389 #ifdef CONFIG_PROC_FS
1390 static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
1392 struct sem_array *sma = it;
1394 return seq_printf(s,
1395 "%10d %10d %4o %10lu %5u %5u %5u %5u %10lu %10lu\n",