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>
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(sma, semid) ipc_checkid(&sma->sem_perm, semid)
93 #define sem_buildid(id, seq) ipc_buildid(id, seq)
95 static struct ipc_ids init_sem_ids;
97 static int newary(struct ipc_namespace *, struct ipc_params *);
98 static void freeary(struct ipc_namespace *, struct sem_array *);
100 static int sysvipc_sem_proc_show(struct seq_file *s, void *it);
103 #define SEMMSL_FAST 256 /* 512 bytes on stack */
104 #define SEMOPM_FAST 64 /* ~ 372 bytes on stack */
107 * linked list protection:
109 * sem_array.sem_pending{,last},
110 * sem_array.sem_undo: sem_lock() for read/write
111 * sem_undo.proc_next: only "current" is allowed to read/write that field.
115 #define sc_semmsl sem_ctls[0]
116 #define sc_semmns sem_ctls[1]
117 #define sc_semopm sem_ctls[2]
118 #define sc_semmni sem_ctls[3]
120 static void __sem_init_ns(struct ipc_namespace *ns, struct ipc_ids *ids)
122 ns->ids[IPC_SEM_IDS] = ids;
123 ns->sc_semmsl = SEMMSL;
124 ns->sc_semmns = SEMMNS;
125 ns->sc_semopm = SEMOPM;
126 ns->sc_semmni = SEMMNI;
131 int sem_init_ns(struct ipc_namespace *ns)
135 ids = kmalloc(sizeof(struct ipc_ids), GFP_KERNEL);
139 __sem_init_ns(ns, ids);
143 void sem_exit_ns(struct ipc_namespace *ns)
145 struct sem_array *sma;
149 down_write(&sem_ids(ns).rw_mutex);
151 in_use = sem_ids(ns).in_use;
153 for (total = 0, next_id = 0; total < in_use; next_id++) {
154 sma = idr_find(&sem_ids(ns).ipcs_idr, next_id);
157 ipc_lock_by_ptr(&sma->sem_perm);
161 up_write(&sem_ids(ns).rw_mutex);
163 kfree(ns->ids[IPC_SEM_IDS]);
164 ns->ids[IPC_SEM_IDS] = NULL;
167 void __init sem_init (void)
169 __sem_init_ns(&init_ipc_ns, &init_sem_ids);
170 ipc_init_proc_interface("sysvipc/sem",
171 " key semid perms nsems uid gid cuid cgid otime ctime\n",
172 IPC_SEM_IDS, sysvipc_sem_proc_show);
176 * This routine is called in the paths where the rw_mutex is held to protect
177 * access to the idr tree.
179 static inline struct sem_array *sem_lock_check_down(struct ipc_namespace *ns,
182 struct kern_ipc_perm *ipcp = ipc_lock_check_down(&sem_ids(ns), id);
184 return container_of(ipcp, struct sem_array, sem_perm);
188 * sem_lock_(check_) routines are called in the paths where the rw_mutex
191 static inline struct sem_array *sem_lock(struct ipc_namespace *ns, int id)
193 struct kern_ipc_perm *ipcp = ipc_lock(&sem_ids(ns), id);
195 return container_of(ipcp, struct sem_array, sem_perm);
198 static inline struct sem_array *sem_lock_check(struct ipc_namespace *ns,
201 struct kern_ipc_perm *ipcp = ipc_lock_check(&sem_ids(ns), id);
203 return container_of(ipcp, struct sem_array, sem_perm);
206 static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
208 ipc_rmid(&sem_ids(ns), &s->sem_perm);
212 * Lockless wakeup algorithm:
213 * Without the check/retry algorithm a lockless wakeup is possible:
214 * - queue.status is initialized to -EINTR before blocking.
215 * - wakeup is performed by
216 * * unlinking the queue entry from sma->sem_pending
217 * * setting queue.status to IN_WAKEUP
218 * This is the notification for the blocked thread that a
219 * result value is imminent.
220 * * call wake_up_process
221 * * set queue.status to the final value.
222 * - the previously blocked thread checks queue.status:
223 * * if it's IN_WAKEUP, then it must wait until the value changes
224 * * if it's not -EINTR, then the operation was completed by
225 * update_queue. semtimedop can return queue.status without
226 * performing any operation on the sem array.
227 * * otherwise it must acquire the spinlock and check what's up.
229 * The two-stage algorithm is necessary to protect against the following
231 * - if queue.status is set after wake_up_process, then the woken up idle
232 * thread could race forward and try (and fail) to acquire sma->lock
233 * before update_queue had a chance to set queue.status
234 * - if queue.status is written before wake_up_process and if the
235 * blocked process is woken up by a signal between writing
236 * queue.status and the wake_up_process, then the woken up
237 * process could return from semtimedop and die by calling
238 * sys_exit before wake_up_process is called. Then wake_up_process
239 * will oops, because the task structure is already invalid.
240 * (yes, this happened on s390 with sysv msg).
246 * newary - Create a new semaphore set
248 * @params: ptr to the structure that contains key, semflg and nsems
250 * Called with sem_ids.rw_mutex held (as a writer)
253 static int newary(struct ipc_namespace *ns, struct ipc_params *params)
257 struct sem_array *sma;
259 key_t key = params->key;
260 int nsems = params->u.nsems;
261 int semflg = params->flg;
265 if (ns->used_sems + nsems > ns->sc_semmns)
268 size = sizeof (*sma) + nsems * sizeof (struct sem);
269 sma = ipc_rcu_alloc(size);
273 memset (sma, 0, size);
275 sma->sem_perm.mode = (semflg & S_IRWXUGO);
276 sma->sem_perm.key = key;
278 sma->sem_perm.security = NULL;
279 retval = security_sem_alloc(sma);
285 id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
287 security_sem_free(sma);
291 ns->used_sems += nsems;
293 sma->sem_perm.id = sem_buildid(id, sma->sem_perm.seq);
294 sma->sem_base = (struct sem *) &sma[1];
295 /* sma->sem_pending = NULL; */
296 sma->sem_pending_last = &sma->sem_pending;
297 /* sma->undo = NULL; */
298 sma->sem_nsems = nsems;
299 sma->sem_ctime = get_seconds();
302 return sma->sem_perm.id;
307 * Called with sem_ids.rw_mutex and ipcp locked.
309 static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg)
311 struct sem_array *sma;
313 sma = container_of(ipcp, struct sem_array, sem_perm);
314 return security_sem_associate(sma, semflg);
318 * Called with sem_ids.rw_mutex and ipcp locked.
320 static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
321 struct ipc_params *params)
323 struct sem_array *sma;
325 sma = container_of(ipcp, struct sem_array, sem_perm);
326 if (params->u.nsems > sma->sem_nsems)
332 asmlinkage long sys_semget(key_t key, int nsems, int semflg)
334 struct ipc_namespace *ns;
335 struct ipc_ops sem_ops;
336 struct ipc_params sem_params;
338 ns = current->nsproxy->ipc_ns;
340 if (nsems < 0 || nsems > ns->sc_semmsl)
343 sem_ops.getnew = newary;
344 sem_ops.associate = sem_security;
345 sem_ops.more_checks = sem_more_checks;
347 sem_params.key = key;
348 sem_params.flg = semflg;
349 sem_params.u.nsems = nsems;
351 return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
354 /* Manage the doubly linked list sma->sem_pending as a FIFO:
355 * insert new queue elements at the tail sma->sem_pending_last.
357 static inline void append_to_queue (struct sem_array * sma,
358 struct sem_queue * q)
360 *(q->prev = sma->sem_pending_last) = q;
361 *(sma->sem_pending_last = &q->next) = NULL;
364 static inline void prepend_to_queue (struct sem_array * sma,
365 struct sem_queue * q)
367 q->next = sma->sem_pending;
368 *(q->prev = &sma->sem_pending) = q;
370 q->next->prev = &q->next;
371 else /* sma->sem_pending_last == &sma->sem_pending */
372 sma->sem_pending_last = &q->next;
375 static inline void remove_from_queue (struct sem_array * sma,
376 struct sem_queue * q)
378 *(q->prev) = q->next;
380 q->next->prev = q->prev;
381 else /* sma->sem_pending_last == &q->next */
382 sma->sem_pending_last = q->prev;
383 q->prev = NULL; /* mark as removed */
387 * Determine whether a sequence of semaphore operations would succeed
388 * all at once. Return 0 if yes, 1 if need to sleep, else return error code.
391 static int try_atomic_semop (struct sem_array * sma, struct sembuf * sops,
392 int nsops, struct sem_undo *un, int pid)
398 for (sop = sops; sop < sops + nsops; sop++) {
399 curr = sma->sem_base + sop->sem_num;
400 sem_op = sop->sem_op;
401 result = curr->semval;
403 if (!sem_op && result)
411 if (sop->sem_flg & SEM_UNDO) {
412 int undo = un->semadj[sop->sem_num] - sem_op;
414 * Exceeding the undo range is an error.
416 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
419 curr->semval = result;
423 while (sop >= sops) {
424 sma->sem_base[sop->sem_num].sempid = pid;
425 if (sop->sem_flg & SEM_UNDO)
426 un->semadj[sop->sem_num] -= sop->sem_op;
430 sma->sem_otime = get_seconds();
438 if (sop->sem_flg & IPC_NOWAIT)
445 while (sop >= sops) {
446 sma->sem_base[sop->sem_num].semval -= sop->sem_op;
453 /* Go through the pending queue for the indicated semaphore
454 * looking for tasks that can be completed.
456 static void update_queue (struct sem_array * sma)
459 struct sem_queue * q;
461 q = sma->sem_pending;
463 error = try_atomic_semop(sma, q->sops, q->nsops,
466 /* Does q->sleeper still need to sleep? */
469 remove_from_queue(sma,q);
470 q->status = IN_WAKEUP;
472 * Continue scanning. The next operation
473 * that must be checked depends on the type of the
474 * completed operation:
475 * - if the operation modified the array, then
476 * restart from the head of the queue and
477 * check for threads that might be waiting
478 * for semaphore values to become 0.
479 * - if the operation didn't modify the array,
480 * then just continue.
483 n = sma->sem_pending;
486 wake_up_process(q->sleeper);
487 /* hands-off: q will disappear immediately after
499 /* The following counts are associated to each semaphore:
500 * semncnt number of tasks waiting on semval being nonzero
501 * semzcnt number of tasks waiting on semval being zero
502 * This model assumes that a task waits on exactly one semaphore.
503 * Since semaphore operations are to be performed atomically, tasks actually
504 * wait on a whole sequence of semaphores simultaneously.
505 * The counts we return here are a rough approximation, but still
506 * warrant that semncnt+semzcnt>0 if the task is on the pending queue.
508 static int count_semncnt (struct sem_array * sma, ushort semnum)
511 struct sem_queue * q;
514 for (q = sma->sem_pending; q; q = q->next) {
515 struct sembuf * sops = q->sops;
516 int nsops = q->nsops;
518 for (i = 0; i < nsops; i++)
519 if (sops[i].sem_num == semnum
520 && (sops[i].sem_op < 0)
521 && !(sops[i].sem_flg & IPC_NOWAIT))
526 static int count_semzcnt (struct sem_array * sma, ushort semnum)
529 struct sem_queue * q;
532 for (q = sma->sem_pending; q; q = q->next) {
533 struct sembuf * sops = q->sops;
534 int nsops = q->nsops;
536 for (i = 0; i < nsops; i++)
537 if (sops[i].sem_num == semnum
538 && (sops[i].sem_op == 0)
539 && !(sops[i].sem_flg & IPC_NOWAIT))
545 /* Free a semaphore set. freeary() is called with sem_ids.rw_mutex locked
546 * as a writer and the spinlock for this semaphore set hold. sem_ids.rw_mutex
547 * remains locked on exit.
549 static void freeary(struct ipc_namespace *ns, struct sem_array *sma)
554 /* Invalidate the existing undo structures for this semaphore set.
555 * (They will be freed without any further action in exit_sem()
556 * or during the next semop.)
558 for (un = sma->undo; un; un = un->id_next)
561 /* Wake up all pending processes and let them fail with EIDRM. */
562 q = sma->sem_pending;
565 /* lazy remove_from_queue: we are killing the whole queue */
568 q->status = IN_WAKEUP;
569 wake_up_process(q->sleeper); /* doesn't sleep */
571 q->status = -EIDRM; /* hands-off q */
575 /* Remove the semaphore set from the IDR */
579 ns->used_sems -= sma->sem_nsems;
580 security_sem_free(sma);
584 static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
588 return copy_to_user(buf, in, sizeof(*in));
593 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
595 out.sem_otime = in->sem_otime;
596 out.sem_ctime = in->sem_ctime;
597 out.sem_nsems = in->sem_nsems;
599 return copy_to_user(buf, &out, sizeof(out));
606 static int semctl_nolock(struct ipc_namespace *ns, int semid, int semnum,
607 int cmd, int version, union semun arg)
610 struct sem_array *sma;
616 struct seminfo seminfo;
619 err = security_sem_semctl(NULL, cmd);
623 memset(&seminfo,0,sizeof(seminfo));
624 seminfo.semmni = ns->sc_semmni;
625 seminfo.semmns = ns->sc_semmns;
626 seminfo.semmsl = ns->sc_semmsl;
627 seminfo.semopm = ns->sc_semopm;
628 seminfo.semvmx = SEMVMX;
629 seminfo.semmnu = SEMMNU;
630 seminfo.semmap = SEMMAP;
631 seminfo.semume = SEMUME;
632 down_read(&sem_ids(ns).rw_mutex);
633 if (cmd == SEM_INFO) {
634 seminfo.semusz = sem_ids(ns).in_use;
635 seminfo.semaem = ns->used_sems;
637 seminfo.semusz = SEMUSZ;
638 seminfo.semaem = SEMAEM;
640 max_id = ipc_get_maxid(&sem_ids(ns));
641 up_read(&sem_ids(ns).rw_mutex);
642 if (copy_to_user (arg.__buf, &seminfo, sizeof(struct seminfo)))
644 return (max_id < 0) ? 0: max_id;
648 struct semid64_ds tbuf;
651 sma = sem_lock(ns, semid);
656 if (ipcperms (&sma->sem_perm, S_IRUGO))
659 err = security_sem_semctl(sma, cmd);
663 id = sma->sem_perm.id;
665 memset(&tbuf, 0, sizeof(tbuf));
667 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
668 tbuf.sem_otime = sma->sem_otime;
669 tbuf.sem_ctime = sma->sem_ctime;
670 tbuf.sem_nsems = sma->sem_nsems;
672 if (copy_semid_to_user (arg.buf, &tbuf, version))
685 static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
686 int cmd, int version, union semun arg)
688 struct sem_array *sma;
691 ushort fast_sem_io[SEMMSL_FAST];
692 ushort* sem_io = fast_sem_io;
695 sma = sem_lock_check(ns, semid);
699 nsems = sma->sem_nsems;
702 if (ipcperms (&sma->sem_perm, (cmd==SETVAL||cmd==SETALL)?S_IWUGO:S_IRUGO))
705 err = security_sem_semctl(sma, cmd);
713 ushort __user *array = arg.array;
716 if(nsems > SEMMSL_FAST) {
720 sem_io = ipc_alloc(sizeof(ushort)*nsems);
722 ipc_lock_by_ptr(&sma->sem_perm);
728 ipc_lock_by_ptr(&sma->sem_perm);
730 if (sma->sem_perm.deleted) {
737 for (i = 0; i < sma->sem_nsems; i++)
738 sem_io[i] = sma->sem_base[i].semval;
741 if(copy_to_user(array, sem_io, nsems*sizeof(ushort)))
753 if(nsems > SEMMSL_FAST) {
754 sem_io = ipc_alloc(sizeof(ushort)*nsems);
756 ipc_lock_by_ptr(&sma->sem_perm);
763 if (copy_from_user (sem_io, arg.array, nsems*sizeof(ushort))) {
764 ipc_lock_by_ptr(&sma->sem_perm);
771 for (i = 0; i < nsems; i++) {
772 if (sem_io[i] > SEMVMX) {
773 ipc_lock_by_ptr(&sma->sem_perm);
780 ipc_lock_by_ptr(&sma->sem_perm);
782 if (sma->sem_perm.deleted) {
788 for (i = 0; i < nsems; i++)
789 sma->sem_base[i].semval = sem_io[i];
790 for (un = sma->undo; un; un = un->id_next)
791 for (i = 0; i < nsems; i++)
793 sma->sem_ctime = get_seconds();
794 /* maybe some queued-up processes were waiting for this */
801 struct semid64_ds tbuf;
802 memset(&tbuf,0,sizeof(tbuf));
803 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
804 tbuf.sem_otime = sma->sem_otime;
805 tbuf.sem_ctime = sma->sem_ctime;
806 tbuf.sem_nsems = sma->sem_nsems;
808 if (copy_semid_to_user (arg.buf, &tbuf, version))
812 /* GETVAL, GETPID, GETNCTN, GETZCNT, SETVAL: fall-through */
815 if(semnum < 0 || semnum >= nsems)
818 curr = &sma->sem_base[semnum];
828 err = count_semncnt(sma,semnum);
831 err = count_semzcnt(sma,semnum);
838 if (val > SEMVMX || val < 0)
841 for (un = sma->undo; un; un = un->id_next)
842 un->semadj[semnum] = 0;
844 curr->sempid = task_tgid_vnr(current);
845 sma->sem_ctime = get_seconds();
846 /* maybe some queued-up processes were waiting for this */
855 if(sem_io != fast_sem_io)
856 ipc_free(sem_io, sizeof(ushort)*nsems);
866 static inline unsigned long copy_semid_from_user(struct sem_setbuf *out, void __user *buf, int version)
871 struct semid64_ds tbuf;
873 if(copy_from_user(&tbuf, buf, sizeof(tbuf)))
876 out->uid = tbuf.sem_perm.uid;
877 out->gid = tbuf.sem_perm.gid;
878 out->mode = tbuf.sem_perm.mode;
884 struct semid_ds tbuf_old;
886 if(copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
889 out->uid = tbuf_old.sem_perm.uid;
890 out->gid = tbuf_old.sem_perm.gid;
891 out->mode = tbuf_old.sem_perm.mode;
900 static int semctl_down(struct ipc_namespace *ns, int semid, int semnum,
901 int cmd, int version, union semun arg)
903 struct sem_array *sma;
905 struct sem_setbuf uninitialized_var(setbuf);
906 struct kern_ipc_perm *ipcp;
909 if(copy_semid_from_user (&setbuf, arg.buf, version))
912 sma = sem_lock_check_down(ns, semid);
916 ipcp = &sma->sem_perm;
918 err = audit_ipc_obj(ipcp);
922 if (cmd == IPC_SET) {
923 err = audit_ipc_set_perm(0, setbuf.uid, setbuf.gid, setbuf.mode);
927 if (current->euid != ipcp->cuid &&
928 current->euid != ipcp->uid && !capable(CAP_SYS_ADMIN)) {
933 err = security_sem_semctl(sma, cmd);
943 ipcp->uid = setbuf.uid;
944 ipcp->gid = setbuf.gid;
945 ipcp->mode = (ipcp->mode & ~S_IRWXUGO)
946 | (setbuf.mode & S_IRWXUGO);
947 sma->sem_ctime = get_seconds();
963 asmlinkage long sys_semctl (int semid, int semnum, int cmd, union semun arg)
967 struct ipc_namespace *ns;
972 version = ipc_parse_version(&cmd);
973 ns = current->nsproxy->ipc_ns;
979 err = semctl_nolock(ns,semid,semnum,cmd,version,arg);
989 err = semctl_main(ns,semid,semnum,cmd,version,arg);
993 down_write(&sem_ids(ns).rw_mutex);
994 err = semctl_down(ns,semid,semnum,cmd,version,arg);
995 up_write(&sem_ids(ns).rw_mutex);
1002 static inline void lock_semundo(void)
1004 struct sem_undo_list *undo_list;
1006 undo_list = current->sysvsem.undo_list;
1008 spin_lock(&undo_list->lock);
1011 /* This code has an interaction with copy_semundo().
1012 * Consider; two tasks are sharing the undo_list. task1
1013 * acquires the undo_list lock in lock_semundo(). If task2 now
1014 * exits before task1 releases the lock (by calling
1015 * unlock_semundo()), then task1 will never call spin_unlock().
1016 * This leave the sem_undo_list in a locked state. If task1 now creats task3
1017 * and once again shares the sem_undo_list, the sem_undo_list will still be
1018 * locked, and future SEM_UNDO operations will deadlock. This case is
1019 * dealt with in copy_semundo() by having it reinitialize the spin lock when
1020 * the refcnt goes from 1 to 2.
1022 static inline void unlock_semundo(void)
1024 struct sem_undo_list *undo_list;
1026 undo_list = current->sysvsem.undo_list;
1028 spin_unlock(&undo_list->lock);
1032 /* If the task doesn't already have a undo_list, then allocate one
1033 * here. We guarantee there is only one thread using this undo list,
1034 * and current is THE ONE
1036 * If this allocation and assignment succeeds, but later
1037 * portions of this code fail, there is no need to free the sem_undo_list.
1038 * Just let it stay associated with the task, and it'll be freed later
1041 * This can block, so callers must hold no locks.
1043 static inline int get_undo_list(struct sem_undo_list **undo_listp)
1045 struct sem_undo_list *undo_list;
1047 undo_list = current->sysvsem.undo_list;
1049 undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
1050 if (undo_list == NULL)
1052 spin_lock_init(&undo_list->lock);
1053 atomic_set(&undo_list->refcnt, 1);
1054 current->sysvsem.undo_list = undo_list;
1056 *undo_listp = undo_list;
1060 static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1062 struct sem_undo **last, *un;
1064 last = &ulp->proc_list;
1067 if(un->semid==semid)
1070 *last=un->proc_next;
1073 last=&un->proc_next;
1080 static struct sem_undo *find_undo(struct ipc_namespace *ns, int semid)
1082 struct sem_array *sma;
1083 struct sem_undo_list *ulp;
1084 struct sem_undo *un, *new;
1088 error = get_undo_list(&ulp);
1090 return ERR_PTR(error);
1093 un = lookup_undo(ulp, semid);
1095 if (likely(un!=NULL))
1098 /* no undo structure around - allocate one. */
1099 sma = sem_lock_check(ns, semid);
1101 return ERR_PTR(PTR_ERR(sma));
1103 nsems = sma->sem_nsems;
1104 ipc_rcu_getref(sma);
1107 new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1109 ipc_lock_by_ptr(&sma->sem_perm);
1110 ipc_rcu_putref(sma);
1112 return ERR_PTR(-ENOMEM);
1114 new->semadj = (short *) &new[1];
1118 un = lookup_undo(ulp, semid);
1122 ipc_lock_by_ptr(&sma->sem_perm);
1123 ipc_rcu_putref(sma);
1127 ipc_lock_by_ptr(&sma->sem_perm);
1128 ipc_rcu_putref(sma);
1129 if (sma->sem_perm.deleted) {
1133 un = ERR_PTR(-EIDRM);
1136 new->proc_next = ulp->proc_list;
1137 ulp->proc_list = new;
1138 new->id_next = sma->undo;
1147 asmlinkage long sys_semtimedop(int semid, struct sembuf __user *tsops,
1148 unsigned nsops, const struct timespec __user *timeout)
1150 int error = -EINVAL;
1151 struct sem_array *sma;
1152 struct sembuf fast_sops[SEMOPM_FAST];
1153 struct sembuf* sops = fast_sops, *sop;
1154 struct sem_undo *un;
1155 int undos = 0, alter = 0, max;
1156 struct sem_queue queue;
1157 unsigned long jiffies_left = 0;
1158 struct ipc_namespace *ns;
1160 ns = current->nsproxy->ipc_ns;
1162 if (nsops < 1 || semid < 0)
1164 if (nsops > ns->sc_semopm)
1166 if(nsops > SEMOPM_FAST) {
1167 sops = kmalloc(sizeof(*sops)*nsops,GFP_KERNEL);
1171 if (copy_from_user (sops, tsops, nsops * sizeof(*tsops))) {
1176 struct timespec _timeout;
1177 if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1181 if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1182 _timeout.tv_nsec >= 1000000000L) {
1186 jiffies_left = timespec_to_jiffies(&_timeout);
1189 for (sop = sops; sop < sops + nsops; sop++) {
1190 if (sop->sem_num >= max)
1192 if (sop->sem_flg & SEM_UNDO)
1194 if (sop->sem_op != 0)
1200 un = find_undo(ns, semid);
1202 error = PTR_ERR(un);
1208 sma = sem_lock_check(ns, semid);
1210 error = PTR_ERR(sma);
1215 * semid identifiers are not unique - find_undo may have
1216 * allocated an undo structure, it was invalidated by an RMID
1217 * and now a new array with received the same id. Check and retry.
1219 if (un && un->semid == -1) {
1224 if (max >= sma->sem_nsems)
1225 goto out_unlock_free;
1228 if (ipcperms(&sma->sem_perm, alter ? S_IWUGO : S_IRUGO))
1229 goto out_unlock_free;
1231 error = security_sem_semop(sma, sops, nsops, alter);
1233 goto out_unlock_free;
1235 error = try_atomic_semop (sma, sops, nsops, un, task_tgid_vnr(current));
1237 if (alter && error == 0)
1239 goto out_unlock_free;
1242 /* We need to sleep on this operation, so we put the current
1243 * task into the pending queue and go to sleep.
1248 queue.nsops = nsops;
1250 queue.pid = task_tgid_vnr(current);
1252 queue.alter = alter;
1254 append_to_queue(sma ,&queue);
1256 prepend_to_queue(sma ,&queue);
1258 queue.status = -EINTR;
1259 queue.sleeper = current;
1260 current->state = TASK_INTERRUPTIBLE;
1264 jiffies_left = schedule_timeout(jiffies_left);
1268 error = queue.status;
1269 while(unlikely(error == IN_WAKEUP)) {
1271 error = queue.status;
1274 if (error != -EINTR) {
1275 /* fast path: update_queue already obtained all requested
1280 sma = sem_lock(ns, semid);
1282 BUG_ON(queue.prev != NULL);
1288 * If queue.status != -EINTR we are woken up by another process
1290 error = queue.status;
1291 if (error != -EINTR) {
1292 goto out_unlock_free;
1296 * If an interrupt occurred we have to clean up the queue
1298 if (timeout && jiffies_left == 0)
1300 remove_from_queue(sma,&queue);
1301 goto out_unlock_free;
1306 if(sops != fast_sops)
1311 asmlinkage long sys_semop (int semid, struct sembuf __user *tsops, unsigned nsops)
1313 return sys_semtimedop(semid, tsops, nsops, NULL);
1316 /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
1317 * parent and child tasks.
1319 * See the notes above unlock_semundo() regarding the spin_lock_init()
1320 * in this code. Initialize the undo_list->lock here instead of get_undo_list()
1321 * because of the reasoning in the comment above unlock_semundo.
1324 int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
1326 struct sem_undo_list *undo_list;
1329 if (clone_flags & CLONE_SYSVSEM) {
1330 error = get_undo_list(&undo_list);
1333 atomic_inc(&undo_list->refcnt);
1334 tsk->sysvsem.undo_list = undo_list;
1336 tsk->sysvsem.undo_list = NULL;
1342 * add semadj values to semaphores, free undo structures.
1343 * undo structures are not freed when semaphore arrays are destroyed
1344 * so some of them may be out of date.
1345 * IMPLEMENTATION NOTE: There is some confusion over whether the
1346 * set of adjustments that needs to be done should be done in an atomic
1347 * manner or not. That is, if we are attempting to decrement the semval
1348 * should we queue up and wait until we can do so legally?
1349 * The original implementation attempted to do this (queue and wait).
1350 * The current implementation does not do so. The POSIX standard
1351 * and SVID should be consulted to determine what behavior is mandated.
1353 void exit_sem(struct task_struct *tsk)
1355 struct sem_undo_list *undo_list;
1356 struct sem_undo *u, **up;
1357 struct ipc_namespace *ns;
1359 undo_list = tsk->sysvsem.undo_list;
1363 if (!atomic_dec_and_test(&undo_list->refcnt))
1366 ns = tsk->nsproxy->ipc_ns;
1367 /* There's no need to hold the semundo list lock, as current
1368 * is the last task exiting for this undo list.
1370 for (up = &undo_list->proc_list; (u = *up); *up = u->proc_next, kfree(u)) {
1371 struct sem_array *sma;
1373 struct sem_undo *un, **unp;
1380 sma = sem_lock(ns, semid);
1387 BUG_ON(sem_checkid(sma, u->semid));
1389 /* remove u from the sma->undo list */
1390 for (unp = &sma->undo; (un = *unp); unp = &un->id_next) {
1394 printk ("exit_sem undo list error id=%d\n", u->semid);
1398 /* perform adjustments registered in u */
1399 nsems = sma->sem_nsems;
1400 for (i = 0; i < nsems; i++) {
1401 struct sem * semaphore = &sma->sem_base[i];
1403 semaphore->semval += u->semadj[i];
1405 * Range checks of the new semaphore value,
1406 * not defined by sus:
1407 * - Some unices ignore the undo entirely
1408 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
1409 * - some cap the value (e.g. FreeBSD caps
1410 * at 0, but doesn't enforce SEMVMX)
1412 * Linux caps the semaphore value, both at 0
1415 * Manfred <manfred@colorfullife.com>
1417 if (semaphore->semval < 0)
1418 semaphore->semval = 0;
1419 if (semaphore->semval > SEMVMX)
1420 semaphore->semval = SEMVMX;
1421 semaphore->sempid = task_tgid_vnr(current);
1424 sma->sem_otime = get_seconds();
1425 /* maybe some queued-up processes were waiting for this */
1433 #ifdef CONFIG_PROC_FS
1434 static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
1436 struct sem_array *sma = it;
1438 return seq_printf(s,
1439 "%10d %10d %4o %10lu %5u %5u %5u %5u %10lu %10lu\n",