3 * Copyright (C) 1992 Krishna Balasubramanian
4 * Copyright (C) 1995 Eric Schenk, Bruno Haible
6 * /proc/sysvipc/sem support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
8 * SMP-threaded, sysctl's added
9 * (c) 1999 Manfred Spraul <manfred@colorfullife.com>
10 * Enforced range limit on SEM_UNDO
11 * (c) 2001 Red Hat Inc
13 * (c) 2003 Manfred Spraul <manfred@colorfullife.com>
14 * Further wakeup optimizations, documentation
15 * (c) 2010 Manfred Spraul <manfred@colorfullife.com>
17 * support for audit of ipc object properties and permission changes
18 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
22 * Pavel Emelianov <xemul@openvz.org>
24 * Implementation notes: (May 2010)
25 * This file implements System V semaphores.
27 * User space visible behavior:
28 * - FIFO ordering for semop() operations (just FIFO, not starvation
30 * - multiple semaphore operations that alter the same semaphore in
31 * one semop() are handled.
32 * - sem_ctime (time of last semctl()) is updated in the IPC_SET, SETVAL and
34 * - two Linux specific semctl() commands: SEM_STAT, SEM_INFO.
35 * - undo adjustments at process exit are limited to 0..SEMVMX.
36 * - namespace are supported.
37 * - SEMMSL, SEMMNS, SEMOPM and SEMMNI can be configured at runtine by writing
38 * to /proc/sys/kernel/sem.
39 * - statistics about the usage are reported in /proc/sysvipc/sem.
43 * - all global variables are read-mostly.
44 * - semop() calls and semctl(RMID) are synchronized by RCU.
45 * - most operations do write operations (actually: spin_lock calls) to
46 * the per-semaphore array structure.
47 * Thus: Perfect SMP scaling between independent semaphore arrays.
48 * If multiple semaphores in one array are used, then cache line
49 * trashing on the semaphore array spinlock will limit the scaling.
50 * - semncnt and semzcnt are calculated on demand in count_semncnt() and
52 * - the task that performs a successful semop() scans the list of all
53 * sleeping tasks and completes any pending operations that can be fulfilled.
54 * Semaphores are actively given to waiting tasks (necessary for FIFO).
55 * (see update_queue())
56 * - To improve the scalability, the actual wake-up calls are performed after
57 * dropping all locks. (see wake_up_sem_queue_prepare(),
58 * wake_up_sem_queue_do())
59 * - All work is done by the waker, the woken up task does not have to do
60 * anything - not even acquiring a lock or dropping a refcount.
61 * - A woken up task may not even touch the semaphore array anymore, it may
62 * have been destroyed already by a semctl(RMID).
63 * - The synchronizations between wake-ups due to a timeout/signal and a
64 * wake-up due to a completed semaphore operation is achieved by using an
65 * intermediate state (IN_WAKEUP).
66 * - UNDO values are stored in an array (one per process and per
67 * semaphore array, lazily allocated). For backwards compatibility, multiple
68 * modes for the UNDO variables are supported (per process, per thread)
69 * (see copy_semundo, CLONE_SYSVSEM)
70 * - There are two lists of the pending operations: a per-array list
71 * and per-semaphore list (stored in the array). This allows to achieve FIFO
72 * ordering without always scanning all pending operations.
73 * The worst-case behavior is nevertheless O(N^2) for N wakeups.
76 #include <linux/slab.h>
77 #include <linux/spinlock.h>
78 #include <linux/init.h>
79 #include <linux/proc_fs.h>
80 #include <linux/time.h>
81 #include <linux/security.h>
82 #include <linux/syscalls.h>
83 #include <linux/audit.h>
84 #include <linux/capability.h>
85 #include <linux/seq_file.h>
86 #include <linux/rwsem.h>
87 #include <linux/nsproxy.h>
88 #include <linux/ipc_namespace.h>
90 #include <asm/uaccess.h>
93 /* One semaphore structure for each semaphore in the system. */
95 int semval; /* current value */
96 int sempid; /* pid of last operation */
97 struct list_head sem_pending; /* pending single-sop operations */
100 /* One queue for each sleeping process in the system. */
102 struct list_head simple_list; /* queue of pending operations */
103 struct list_head list; /* queue of pending operations */
104 struct task_struct *sleeper; /* this process */
105 struct sem_undo *undo; /* undo structure */
106 int pid; /* process id of requesting process */
107 int status; /* completion status of operation */
108 struct sembuf *sops; /* array of pending operations */
109 int nsops; /* number of operations */
110 int alter; /* does *sops alter the array? */
113 /* Each task has a list of undo requests. They are executed automatically
114 * when the process exits.
117 struct list_head list_proc; /* per-process list: *
118 * all undos from one process
120 struct rcu_head rcu; /* rcu struct for sem_undo */
121 struct sem_undo_list *ulp; /* back ptr to sem_undo_list */
122 struct list_head list_id; /* per semaphore array list:
123 * all undos for one array */
124 int semid; /* semaphore set identifier */
125 short *semadj; /* array of adjustments */
126 /* one per semaphore */
129 /* sem_undo_list controls shared access to the list of sem_undo structures
130 * that may be shared among all a CLONE_SYSVSEM task group.
132 struct sem_undo_list {
135 struct list_head list_proc;
139 #define sem_ids(ns) ((ns)->ids[IPC_SEM_IDS])
141 #define sem_unlock(sma) ipc_unlock(&(sma)->sem_perm)
142 #define sem_checkid(sma, semid) ipc_checkid(&sma->sem_perm, semid)
144 static int newary(struct ipc_namespace *, struct ipc_params *);
145 static void freeary(struct ipc_namespace *, struct kern_ipc_perm *);
146 #ifdef CONFIG_PROC_FS
147 static int sysvipc_sem_proc_show(struct seq_file *s, void *it);
150 #define SEMMSL_FAST 256 /* 512 bytes on stack */
151 #define SEMOPM_FAST 64 /* ~ 372 bytes on stack */
154 * linked list protection:
156 * sem_array.sem_pending{,last},
157 * sem_array.sem_undo: sem_lock() for read/write
158 * sem_undo.proc_next: only "current" is allowed to read/write that field.
162 #define sc_semmsl sem_ctls[0]
163 #define sc_semmns sem_ctls[1]
164 #define sc_semopm sem_ctls[2]
165 #define sc_semmni sem_ctls[3]
167 void sem_init_ns(struct ipc_namespace *ns)
169 ns->sc_semmsl = SEMMSL;
170 ns->sc_semmns = SEMMNS;
171 ns->sc_semopm = SEMOPM;
172 ns->sc_semmni = SEMMNI;
174 ipc_init_ids(&ns->ids[IPC_SEM_IDS]);
178 void sem_exit_ns(struct ipc_namespace *ns)
180 free_ipcs(ns, &sem_ids(ns), freeary);
181 idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr);
185 void __init sem_init (void)
187 sem_init_ns(&init_ipc_ns);
188 ipc_init_proc_interface("sysvipc/sem",
189 " key semid perms nsems uid gid cuid cgid otime ctime\n",
190 IPC_SEM_IDS, sysvipc_sem_proc_show);
194 * sem_lock_(check_) routines are called in the paths where the rw_mutex
197 static inline struct sem_array *sem_lock(struct ipc_namespace *ns, int id)
199 struct kern_ipc_perm *ipcp = ipc_lock(&sem_ids(ns), id);
202 return (struct sem_array *)ipcp;
204 return container_of(ipcp, struct sem_array, sem_perm);
207 static inline struct sem_array *sem_obtain_object(struct ipc_namespace *ns, int id)
209 struct kern_ipc_perm *ipcp = ipc_obtain_object(&sem_ids(ns), id);
212 return ERR_CAST(ipcp);
214 return container_of(ipcp, struct sem_array, sem_perm);
217 static inline struct sem_array *sem_lock_check(struct ipc_namespace *ns,
220 struct kern_ipc_perm *ipcp = ipc_lock_check(&sem_ids(ns), id);
223 return ERR_CAST(ipcp);
225 return container_of(ipcp, struct sem_array, sem_perm);
228 static inline struct sem_array *sem_obtain_object_check(struct ipc_namespace *ns,
231 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&sem_ids(ns), id);
234 return ERR_CAST(ipcp);
236 return container_of(ipcp, struct sem_array, sem_perm);
239 static inline void sem_lock_and_putref(struct sem_array *sma)
241 ipc_lock_by_ptr(&sma->sem_perm);
245 static inline void sem_getref_and_unlock(struct sem_array *sma)
248 ipc_unlock(&(sma)->sem_perm);
251 static inline void sem_putref(struct sem_array *sma)
253 ipc_lock_by_ptr(&sma->sem_perm);
255 ipc_unlock(&(sma)->sem_perm);
259 * Call inside the rcu read section.
261 static inline void sem_getref(struct sem_array *sma)
263 spin_lock(&(sma)->sem_perm.lock);
265 ipc_unlock(&(sma)->sem_perm);
268 static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
270 ipc_rmid(&sem_ids(ns), &s->sem_perm);
274 * Lockless wakeup algorithm:
275 * Without the check/retry algorithm a lockless wakeup is possible:
276 * - queue.status is initialized to -EINTR before blocking.
277 * - wakeup is performed by
278 * * unlinking the queue entry from sma->sem_pending
279 * * setting queue.status to IN_WAKEUP
280 * This is the notification for the blocked thread that a
281 * result value is imminent.
282 * * call wake_up_process
283 * * set queue.status to the final value.
284 * - the previously blocked thread checks queue.status:
285 * * if it's IN_WAKEUP, then it must wait until the value changes
286 * * if it's not -EINTR, then the operation was completed by
287 * update_queue. semtimedop can return queue.status without
288 * performing any operation on the sem array.
289 * * otherwise it must acquire the spinlock and check what's up.
291 * The two-stage algorithm is necessary to protect against the following
293 * - if queue.status is set after wake_up_process, then the woken up idle
294 * thread could race forward and try (and fail) to acquire sma->lock
295 * before update_queue had a chance to set queue.status
296 * - if queue.status is written before wake_up_process and if the
297 * blocked process is woken up by a signal between writing
298 * queue.status and the wake_up_process, then the woken up
299 * process could return from semtimedop and die by calling
300 * sys_exit before wake_up_process is called. Then wake_up_process
301 * will oops, because the task structure is already invalid.
302 * (yes, this happened on s390 with sysv msg).
308 * newary - Create a new semaphore set
310 * @params: ptr to the structure that contains key, semflg and nsems
312 * Called with sem_ids.rw_mutex held (as a writer)
315 static int newary(struct ipc_namespace *ns, struct ipc_params *params)
319 struct sem_array *sma;
321 key_t key = params->key;
322 int nsems = params->u.nsems;
323 int semflg = params->flg;
328 if (ns->used_sems + nsems > ns->sc_semmns)
331 size = sizeof (*sma) + nsems * sizeof (struct sem);
332 sma = ipc_rcu_alloc(size);
336 memset (sma, 0, size);
338 sma->sem_perm.mode = (semflg & S_IRWXUGO);
339 sma->sem_perm.key = key;
341 sma->sem_perm.security = NULL;
342 retval = security_sem_alloc(sma);
348 id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
350 security_sem_free(sma);
354 ns->used_sems += nsems;
356 sma->sem_base = (struct sem *) &sma[1];
358 for (i = 0; i < nsems; i++)
359 INIT_LIST_HEAD(&sma->sem_base[i].sem_pending);
361 sma->complex_count = 0;
362 INIT_LIST_HEAD(&sma->sem_pending);
363 INIT_LIST_HEAD(&sma->list_id);
364 sma->sem_nsems = nsems;
365 sma->sem_ctime = get_seconds();
368 return sma->sem_perm.id;
373 * Called with sem_ids.rw_mutex and ipcp locked.
375 static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg)
377 struct sem_array *sma;
379 sma = container_of(ipcp, struct sem_array, sem_perm);
380 return security_sem_associate(sma, semflg);
384 * Called with sem_ids.rw_mutex and ipcp locked.
386 static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
387 struct ipc_params *params)
389 struct sem_array *sma;
391 sma = container_of(ipcp, struct sem_array, sem_perm);
392 if (params->u.nsems > sma->sem_nsems)
398 SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg)
400 struct ipc_namespace *ns;
401 struct ipc_ops sem_ops;
402 struct ipc_params sem_params;
404 ns = current->nsproxy->ipc_ns;
406 if (nsems < 0 || nsems > ns->sc_semmsl)
409 sem_ops.getnew = newary;
410 sem_ops.associate = sem_security;
411 sem_ops.more_checks = sem_more_checks;
413 sem_params.key = key;
414 sem_params.flg = semflg;
415 sem_params.u.nsems = nsems;
417 return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
421 * Determine whether a sequence of semaphore operations would succeed
422 * all at once. Return 0 if yes, 1 if need to sleep, else return error code.
425 static int try_atomic_semop (struct sem_array * sma, struct sembuf * sops,
426 int nsops, struct sem_undo *un, int pid)
432 for (sop = sops; sop < sops + nsops; sop++) {
433 curr = sma->sem_base + sop->sem_num;
434 sem_op = sop->sem_op;
435 result = curr->semval;
437 if (!sem_op && result)
445 if (sop->sem_flg & SEM_UNDO) {
446 int undo = un->semadj[sop->sem_num] - sem_op;
448 * Exceeding the undo range is an error.
450 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
453 curr->semval = result;
457 while (sop >= sops) {
458 sma->sem_base[sop->sem_num].sempid = pid;
459 if (sop->sem_flg & SEM_UNDO)
460 un->semadj[sop->sem_num] -= sop->sem_op;
471 if (sop->sem_flg & IPC_NOWAIT)
478 while (sop >= sops) {
479 sma->sem_base[sop->sem_num].semval -= sop->sem_op;
486 /** wake_up_sem_queue_prepare(q, error): Prepare wake-up
487 * @q: queue entry that must be signaled
488 * @error: Error value for the signal
490 * Prepare the wake-up of the queue entry q.
492 static void wake_up_sem_queue_prepare(struct list_head *pt,
493 struct sem_queue *q, int error)
495 if (list_empty(pt)) {
497 * Hold preempt off so that we don't get preempted and have the
498 * wakee busy-wait until we're scheduled back on.
502 q->status = IN_WAKEUP;
505 list_add_tail(&q->simple_list, pt);
509 * wake_up_sem_queue_do(pt) - do the actual wake-up
510 * @pt: list of tasks to be woken up
512 * Do the actual wake-up.
513 * The function is called without any locks held, thus the semaphore array
514 * could be destroyed already and the tasks can disappear as soon as the
515 * status is set to the actual return code.
517 static void wake_up_sem_queue_do(struct list_head *pt)
519 struct sem_queue *q, *t;
522 did_something = !list_empty(pt);
523 list_for_each_entry_safe(q, t, pt, simple_list) {
524 wake_up_process(q->sleeper);
525 /* q can disappear immediately after writing q->status. */
533 static void unlink_queue(struct sem_array *sma, struct sem_queue *q)
537 list_del(&q->simple_list);
539 sma->complex_count--;
542 /** check_restart(sma, q)
543 * @sma: semaphore array
544 * @q: the operation that just completed
546 * update_queue is O(N^2) when it restarts scanning the whole queue of
547 * waiting operations. Therefore this function checks if the restart is
548 * really necessary. It is called after a previously waiting operation
551 static int check_restart(struct sem_array *sma, struct sem_queue *q)
556 /* if the operation didn't modify the array, then no restart */
560 /* pending complex operations are too difficult to analyse */
561 if (sma->complex_count)
564 /* we were a sleeping complex operation. Too difficult */
568 curr = sma->sem_base + q->sops[0].sem_num;
570 /* No-one waits on this queue */
571 if (list_empty(&curr->sem_pending))
574 /* the new semaphore value */
576 /* It is impossible that someone waits for the new value:
577 * - q is a previously sleeping simple operation that
578 * altered the array. It must be a decrement, because
579 * simple increments never sleep.
580 * - The value is not 0, thus wait-for-zero won't proceed.
581 * - If there are older (higher priority) decrements
582 * in the queue, then they have observed the original
583 * semval value and couldn't proceed. The operation
584 * decremented to value - thus they won't proceed either.
586 BUG_ON(q->sops[0].sem_op >= 0);
590 * semval is 0. Check if there are wait-for-zero semops.
591 * They must be the first entries in the per-semaphore simple queue
593 h = list_first_entry(&curr->sem_pending, struct sem_queue, simple_list);
594 BUG_ON(h->nsops != 1);
595 BUG_ON(h->sops[0].sem_num != q->sops[0].sem_num);
597 /* Yes, there is a wait-for-zero semop. Restart */
598 if (h->sops[0].sem_op == 0)
601 /* Again - no-one is waiting for the new value. */
607 * update_queue(sma, semnum): Look for tasks that can be completed.
608 * @sma: semaphore array.
609 * @semnum: semaphore that was modified.
610 * @pt: list head for the tasks that must be woken up.
612 * update_queue must be called after a semaphore in a semaphore array
613 * was modified. If multiple semaphore were modified, then @semnum
615 * The tasks that must be woken up are added to @pt. The return code
616 * is stored in q->pid.
617 * The function return 1 if at least one semop was completed successfully.
619 static int update_queue(struct sem_array *sma, int semnum, struct list_head *pt)
622 struct list_head *walk;
623 struct list_head *pending_list;
625 int semop_completed = 0;
627 /* if there are complex operations around, then knowing the semaphore
628 * that was modified doesn't help us. Assume that multiple semaphores
631 if (sma->complex_count)
635 pending_list = &sma->sem_pending;
636 offset = offsetof(struct sem_queue, list);
638 pending_list = &sma->sem_base[semnum].sem_pending;
639 offset = offsetof(struct sem_queue, simple_list);
643 walk = pending_list->next;
644 while (walk != pending_list) {
647 q = (struct sem_queue *)((char *)walk - offset);
650 /* If we are scanning the single sop, per-semaphore list of
651 * one semaphore and that semaphore is 0, then it is not
652 * necessary to scan the "alter" entries: simple increments
653 * that affect only one entry succeed immediately and cannot
654 * be in the per semaphore pending queue, and decrements
655 * cannot be successful if the value is already 0.
657 if (semnum != -1 && sma->sem_base[semnum].semval == 0 &&
661 error = try_atomic_semop(sma, q->sops, q->nsops,
664 /* Does q->sleeper still need to sleep? */
668 unlink_queue(sma, q);
674 restart = check_restart(sma, q);
677 wake_up_sem_queue_prepare(pt, q, error);
681 return semop_completed;
685 * do_smart_update(sma, sops, nsops, otime, pt) - optimized update_queue
686 * @sma: semaphore array
687 * @sops: operations that were performed
688 * @nsops: number of operations
689 * @otime: force setting otime
690 * @pt: list head of the tasks that must be woken up.
692 * do_smart_update() does the required called to update_queue, based on the
693 * actual changes that were performed on the semaphore array.
694 * Note that the function does not do the actual wake-up: the caller is
695 * responsible for calling wake_up_sem_queue_do(@pt).
696 * It is safe to perform this call after dropping all locks.
698 static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops,
699 int otime, struct list_head *pt)
703 if (sma->complex_count || sops == NULL) {
704 if (update_queue(sma, -1, pt))
709 for (i = 0; i < nsops; i++) {
710 if (sops[i].sem_op > 0 ||
711 (sops[i].sem_op < 0 &&
712 sma->sem_base[sops[i].sem_num].semval == 0))
713 if (update_queue(sma, sops[i].sem_num, pt))
718 sma->sem_otime = get_seconds();
722 /* The following counts are associated to each semaphore:
723 * semncnt number of tasks waiting on semval being nonzero
724 * semzcnt number of tasks waiting on semval being zero
725 * This model assumes that a task waits on exactly one semaphore.
726 * Since semaphore operations are to be performed atomically, tasks actually
727 * wait on a whole sequence of semaphores simultaneously.
728 * The counts we return here are a rough approximation, but still
729 * warrant that semncnt+semzcnt>0 if the task is on the pending queue.
731 static int count_semncnt (struct sem_array * sma, ushort semnum)
734 struct sem_queue * q;
737 list_for_each_entry(q, &sma->sem_pending, list) {
738 struct sembuf * sops = q->sops;
739 int nsops = q->nsops;
741 for (i = 0; i < nsops; i++)
742 if (sops[i].sem_num == semnum
743 && (sops[i].sem_op < 0)
744 && !(sops[i].sem_flg & IPC_NOWAIT))
750 static int count_semzcnt (struct sem_array * sma, ushort semnum)
753 struct sem_queue * q;
756 list_for_each_entry(q, &sma->sem_pending, list) {
757 struct sembuf * sops = q->sops;
758 int nsops = q->nsops;
760 for (i = 0; i < nsops; i++)
761 if (sops[i].sem_num == semnum
762 && (sops[i].sem_op == 0)
763 && !(sops[i].sem_flg & IPC_NOWAIT))
769 /* Free a semaphore set. freeary() is called with sem_ids.rw_mutex locked
770 * as a writer and the spinlock for this semaphore set hold. sem_ids.rw_mutex
771 * remains locked on exit.
773 static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
775 struct sem_undo *un, *tu;
776 struct sem_queue *q, *tq;
777 struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
778 struct list_head tasks;
780 /* Free the existing undo structures for this semaphore set. */
781 assert_spin_locked(&sma->sem_perm.lock);
782 list_for_each_entry_safe(un, tu, &sma->list_id, list_id) {
783 list_del(&un->list_id);
784 spin_lock(&un->ulp->lock);
786 list_del_rcu(&un->list_proc);
787 spin_unlock(&un->ulp->lock);
791 /* Wake up all pending processes and let them fail with EIDRM. */
792 INIT_LIST_HEAD(&tasks);
793 list_for_each_entry_safe(q, tq, &sma->sem_pending, list) {
794 unlink_queue(sma, q);
795 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
798 /* Remove the semaphore set from the IDR */
802 wake_up_sem_queue_do(&tasks);
803 ns->used_sems -= sma->sem_nsems;
804 security_sem_free(sma);
808 static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
812 return copy_to_user(buf, in, sizeof(*in));
817 memset(&out, 0, sizeof(out));
819 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
821 out.sem_otime = in->sem_otime;
822 out.sem_ctime = in->sem_ctime;
823 out.sem_nsems = in->sem_nsems;
825 return copy_to_user(buf, &out, sizeof(out));
832 static int semctl_nolock(struct ipc_namespace *ns, int semid,
833 int cmd, int version, void __user *p)
836 struct sem_array *sma;
842 struct seminfo seminfo;
845 err = security_sem_semctl(NULL, cmd);
849 memset(&seminfo,0,sizeof(seminfo));
850 seminfo.semmni = ns->sc_semmni;
851 seminfo.semmns = ns->sc_semmns;
852 seminfo.semmsl = ns->sc_semmsl;
853 seminfo.semopm = ns->sc_semopm;
854 seminfo.semvmx = SEMVMX;
855 seminfo.semmnu = SEMMNU;
856 seminfo.semmap = SEMMAP;
857 seminfo.semume = SEMUME;
858 down_read(&sem_ids(ns).rw_mutex);
859 if (cmd == SEM_INFO) {
860 seminfo.semusz = sem_ids(ns).in_use;
861 seminfo.semaem = ns->used_sems;
863 seminfo.semusz = SEMUSZ;
864 seminfo.semaem = SEMAEM;
866 max_id = ipc_get_maxid(&sem_ids(ns));
867 up_read(&sem_ids(ns).rw_mutex);
868 if (copy_to_user(p, &seminfo, sizeof(struct seminfo)))
870 return (max_id < 0) ? 0: max_id;
875 struct semid64_ds tbuf;
878 memset(&tbuf, 0, sizeof(tbuf));
880 if (cmd == SEM_STAT) {
882 sma = sem_obtain_object(ns, semid);
887 id = sma->sem_perm.id;
890 sma = sem_obtain_object_check(ns, semid);
898 if (ipcperms(ns, &sma->sem_perm, S_IRUGO))
901 err = security_sem_semctl(sma, cmd);
905 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
906 tbuf.sem_otime = sma->sem_otime;
907 tbuf.sem_ctime = sma->sem_ctime;
908 tbuf.sem_nsems = sma->sem_nsems;
910 if (copy_semid_to_user(p, &tbuf, version))
922 static int semctl_setval(struct ipc_namespace *ns, int semid, int semnum,
926 struct sem_array *sma;
930 struct list_head tasks;
932 #if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN)
933 /* big-endian 64bit */
936 /* 32bit or little-endian 64bit */
940 sma = sem_lock_check(ns, semid);
944 INIT_LIST_HEAD(&tasks);
945 nsems = sma->sem_nsems;
948 if (ipcperms(ns, &sma->sem_perm, S_IWUGO))
951 err = security_sem_semctl(sma, SETVAL);
956 if(semnum < 0 || semnum >= nsems)
959 curr = &sma->sem_base[semnum];
962 if (val > SEMVMX || val < 0)
965 assert_spin_locked(&sma->sem_perm.lock);
966 list_for_each_entry(un, &sma->list_id, list_id)
967 un->semadj[semnum] = 0;
970 curr->sempid = task_tgid_vnr(current);
971 sma->sem_ctime = get_seconds();
972 /* maybe some queued-up processes were waiting for this */
973 do_smart_update(sma, NULL, 0, 0, &tasks);
977 wake_up_sem_queue_do(&tasks);
981 static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
982 int cmd, void __user *p)
984 struct sem_array *sma;
987 ushort fast_sem_io[SEMMSL_FAST];
988 ushort* sem_io = fast_sem_io;
989 struct list_head tasks;
991 INIT_LIST_HEAD(&tasks);
994 sma = sem_obtain_object_check(ns, semid);
1000 nsems = sma->sem_nsems;
1003 if (ipcperms(ns, &sma->sem_perm,
1004 cmd == SETALL ? S_IWUGO : S_IRUGO)) {
1009 err = security_sem_semctl(sma, cmd);
1019 ushort __user *array = p;
1022 if(nsems > SEMMSL_FAST) {
1025 sem_io = ipc_alloc(sizeof(ushort)*nsems);
1026 if(sem_io == NULL) {
1031 sem_lock_and_putref(sma);
1032 if (sma->sem_perm.deleted) {
1039 spin_lock(&sma->sem_perm.lock);
1040 for (i = 0; i < sma->sem_nsems; i++)
1041 sem_io[i] = sma->sem_base[i].semval;
1044 if(copy_to_user(array, sem_io, nsems*sizeof(ushort)))
1051 struct sem_undo *un;
1053 ipc_rcu_getref(sma);
1056 if(nsems > SEMMSL_FAST) {
1057 sem_io = ipc_alloc(sizeof(ushort)*nsems);
1058 if(sem_io == NULL) {
1064 if (copy_from_user (sem_io, p, nsems*sizeof(ushort))) {
1070 for (i = 0; i < nsems; i++) {
1071 if (sem_io[i] > SEMVMX) {
1077 sem_lock_and_putref(sma);
1078 if (sma->sem_perm.deleted) {
1084 for (i = 0; i < nsems; i++)
1085 sma->sem_base[i].semval = sem_io[i];
1087 assert_spin_locked(&sma->sem_perm.lock);
1088 list_for_each_entry(un, &sma->list_id, list_id) {
1089 for (i = 0; i < nsems; i++)
1092 sma->sem_ctime = get_seconds();
1093 /* maybe some queued-up processes were waiting for this */
1094 do_smart_update(sma, NULL, 0, 0, &tasks);
1098 /* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */
1101 if (semnum < 0 || semnum >= nsems) {
1106 spin_lock(&sma->sem_perm.lock);
1107 curr = &sma->sem_base[semnum];
1117 err = count_semncnt(sma,semnum);
1120 err = count_semzcnt(sma,semnum);
1127 wake_up_sem_queue_do(&tasks);
1129 if(sem_io != fast_sem_io)
1130 ipc_free(sem_io, sizeof(ushort)*nsems);
1134 static inline unsigned long
1135 copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
1139 if (copy_from_user(out, buf, sizeof(*out)))
1144 struct semid_ds tbuf_old;
1146 if(copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
1149 out->sem_perm.uid = tbuf_old.sem_perm.uid;
1150 out->sem_perm.gid = tbuf_old.sem_perm.gid;
1151 out->sem_perm.mode = tbuf_old.sem_perm.mode;
1161 * This function handles some semctl commands which require the rw_mutex
1162 * to be held in write mode.
1163 * NOTE: no locks must be held, the rw_mutex is taken inside this function.
1165 static int semctl_down(struct ipc_namespace *ns, int semid,
1166 int cmd, int version, void __user *p)
1168 struct sem_array *sma;
1170 struct semid64_ds semid64;
1171 struct kern_ipc_perm *ipcp;
1173 if(cmd == IPC_SET) {
1174 if (copy_semid_from_user(&semid64, p, version))
1178 ipcp = ipcctl_pre_down_nolock(ns, &sem_ids(ns), semid, cmd,
1179 &semid64.sem_perm, 0);
1181 return PTR_ERR(ipcp);
1183 sma = container_of(ipcp, struct sem_array, sem_perm);
1185 err = security_sem_semctl(sma, cmd);
1193 ipc_lock_object(&sma->sem_perm);
1197 ipc_lock_object(&sma->sem_perm);
1198 err = ipc_update_perm(&semid64.sem_perm, ipcp);
1201 sma->sem_ctime = get_seconds();
1212 up_write(&sem_ids(ns).rw_mutex);
1216 SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, unsigned long, arg)
1219 struct ipc_namespace *ns;
1220 void __user *p = (void __user *)arg;
1225 version = ipc_parse_version(&cmd);
1226 ns = current->nsproxy->ipc_ns;
1233 return semctl_nolock(ns, semid, cmd, version, p);
1240 return semctl_main(ns, semid, semnum, cmd, p);
1242 return semctl_setval(ns, semid, semnum, arg);
1245 return semctl_down(ns, semid, cmd, version, p);
1251 /* If the task doesn't already have a undo_list, then allocate one
1252 * here. We guarantee there is only one thread using this undo list,
1253 * and current is THE ONE
1255 * If this allocation and assignment succeeds, but later
1256 * portions of this code fail, there is no need to free the sem_undo_list.
1257 * Just let it stay associated with the task, and it'll be freed later
1260 * This can block, so callers must hold no locks.
1262 static inline int get_undo_list(struct sem_undo_list **undo_listp)
1264 struct sem_undo_list *undo_list;
1266 undo_list = current->sysvsem.undo_list;
1268 undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
1269 if (undo_list == NULL)
1271 spin_lock_init(&undo_list->lock);
1272 atomic_set(&undo_list->refcnt, 1);
1273 INIT_LIST_HEAD(&undo_list->list_proc);
1275 current->sysvsem.undo_list = undo_list;
1277 *undo_listp = undo_list;
1281 static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid)
1283 struct sem_undo *un;
1285 list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) {
1286 if (un->semid == semid)
1292 static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1294 struct sem_undo *un;
1296 assert_spin_locked(&ulp->lock);
1298 un = __lookup_undo(ulp, semid);
1300 list_del_rcu(&un->list_proc);
1301 list_add_rcu(&un->list_proc, &ulp->list_proc);
1307 * find_alloc_undo - Lookup (and if not present create) undo array
1309 * @semid: semaphore array id
1311 * The function looks up (and if not present creates) the undo structure.
1312 * The size of the undo structure depends on the size of the semaphore
1313 * array, thus the alloc path is not that straightforward.
1314 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1315 * performs a rcu_read_lock().
1317 static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
1319 struct sem_array *sma;
1320 struct sem_undo_list *ulp;
1321 struct sem_undo *un, *new;
1325 error = get_undo_list(&ulp);
1327 return ERR_PTR(error);
1330 spin_lock(&ulp->lock);
1331 un = lookup_undo(ulp, semid);
1332 spin_unlock(&ulp->lock);
1333 if (likely(un!=NULL))
1336 /* no undo structure around - allocate one. */
1337 /* step 1: figure out the size of the semaphore array */
1338 sma = sem_obtain_object_check(ns, semid);
1341 return ERR_CAST(sma);
1344 nsems = sma->sem_nsems;
1345 ipc_rcu_getref(sma);
1348 /* step 2: allocate new undo structure */
1349 new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1352 return ERR_PTR(-ENOMEM);
1355 /* step 3: Acquire the lock on semaphore array */
1356 sem_lock_and_putref(sma);
1357 if (sma->sem_perm.deleted) {
1360 un = ERR_PTR(-EIDRM);
1363 spin_lock(&ulp->lock);
1366 * step 4: check for races: did someone else allocate the undo struct?
1368 un = lookup_undo(ulp, semid);
1373 /* step 5: initialize & link new undo structure */
1374 new->semadj = (short *) &new[1];
1377 assert_spin_locked(&ulp->lock);
1378 list_add_rcu(&new->list_proc, &ulp->list_proc);
1379 assert_spin_locked(&sma->sem_perm.lock);
1380 list_add(&new->list_id, &sma->list_id);
1384 spin_unlock(&ulp->lock);
1393 * get_queue_result - Retrieve the result code from sem_queue
1394 * @q: Pointer to queue structure
1396 * Retrieve the return code from the pending queue. If IN_WAKEUP is found in
1397 * q->status, then we must loop until the value is replaced with the final
1398 * value: This may happen if a task is woken up by an unrelated event (e.g.
1399 * signal) and in parallel the task is woken up by another task because it got
1400 * the requested semaphores.
1402 * The function can be called with or without holding the semaphore spinlock.
1404 static int get_queue_result(struct sem_queue *q)
1409 while (unlikely(error == IN_WAKEUP)) {
1418 SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
1419 unsigned, nsops, const struct timespec __user *, timeout)
1421 int error = -EINVAL;
1422 struct sem_array *sma;
1423 struct sembuf fast_sops[SEMOPM_FAST];
1424 struct sembuf* sops = fast_sops, *sop;
1425 struct sem_undo *un;
1426 int undos = 0, alter = 0, max;
1427 struct sem_queue queue;
1428 unsigned long jiffies_left = 0;
1429 struct ipc_namespace *ns;
1430 struct list_head tasks;
1432 ns = current->nsproxy->ipc_ns;
1434 if (nsops < 1 || semid < 0)
1436 if (nsops > ns->sc_semopm)
1438 if(nsops > SEMOPM_FAST) {
1439 sops = kmalloc(sizeof(*sops)*nsops,GFP_KERNEL);
1443 if (copy_from_user (sops, tsops, nsops * sizeof(*tsops))) {
1448 struct timespec _timeout;
1449 if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1453 if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1454 _timeout.tv_nsec >= 1000000000L) {
1458 jiffies_left = timespec_to_jiffies(&_timeout);
1461 for (sop = sops; sop < sops + nsops; sop++) {
1462 if (sop->sem_num >= max)
1464 if (sop->sem_flg & SEM_UNDO)
1466 if (sop->sem_op != 0)
1471 un = find_alloc_undo(ns, semid);
1473 error = PTR_ERR(un);
1479 INIT_LIST_HEAD(&tasks);
1482 sma = sem_obtain_object_check(ns, semid);
1486 error = PTR_ERR(sma);
1491 if (max >= sma->sem_nsems) {
1497 if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO)) {
1502 error = security_sem_semop(sma, sops, nsops, alter);
1509 * semid identifiers are not unique - find_alloc_undo may have
1510 * allocated an undo structure, it was invalidated by an RMID
1511 * and now a new array with received the same id. Check and fail.
1512 * This case can be detected checking un->semid. The existence of
1513 * "un" itself is guaranteed by rcu.
1516 ipc_lock_object(&sma->sem_perm);
1518 if (un->semid == -1) {
1520 goto out_unlock_free;
1523 * rcu lock can be released, "un" cannot disappear:
1524 * - sem_lock is acquired, thus IPC_RMID is
1526 * - exit_sem is impossible, it always operates on
1527 * current (or a dead task).
1534 error = try_atomic_semop (sma, sops, nsops, un, task_tgid_vnr(current));
1536 if (alter && error == 0)
1537 do_smart_update(sma, sops, nsops, 1, &tasks);
1539 goto out_unlock_free;
1542 /* We need to sleep on this operation, so we put the current
1543 * task into the pending queue and go to sleep.
1547 queue.nsops = nsops;
1549 queue.pid = task_tgid_vnr(current);
1550 queue.alter = alter;
1552 list_add_tail(&queue.list, &sma->sem_pending);
1554 list_add(&queue.list, &sma->sem_pending);
1558 curr = &sma->sem_base[sops->sem_num];
1561 list_add_tail(&queue.simple_list, &curr->sem_pending);
1563 list_add(&queue.simple_list, &curr->sem_pending);
1565 INIT_LIST_HEAD(&queue.simple_list);
1566 sma->complex_count++;
1569 queue.status = -EINTR;
1570 queue.sleeper = current;
1573 current->state = TASK_INTERRUPTIBLE;
1577 jiffies_left = schedule_timeout(jiffies_left);
1581 error = get_queue_result(&queue);
1583 if (error != -EINTR) {
1584 /* fast path: update_queue already obtained all requested
1586 * Perform a smp_mb(): User space could assume that semop()
1587 * is a memory barrier: Without the mb(), the cpu could
1588 * speculatively read in user space stale data that was
1589 * overwritten by the previous owner of the semaphore.
1596 sma = sem_lock(ns, semid);
1599 * Wait until it's guaranteed that no wakeup_sem_queue_do() is ongoing.
1601 error = get_queue_result(&queue);
1604 * Array removed? If yes, leave without sem_unlock().
1612 * If queue.status != -EINTR we are woken up by another process.
1613 * Leave without unlink_queue(), but with sem_unlock().
1616 if (error != -EINTR) {
1617 goto out_unlock_free;
1621 * If an interrupt occurred we have to clean up the queue
1623 if (timeout && jiffies_left == 0)
1627 * If the wakeup was spurious, just retry
1629 if (error == -EINTR && !signal_pending(current))
1632 unlink_queue(sma, &queue);
1637 wake_up_sem_queue_do(&tasks);
1639 if(sops != fast_sops)
1644 SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
1647 return sys_semtimedop(semid, tsops, nsops, NULL);
1650 /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
1651 * parent and child tasks.
1654 int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
1656 struct sem_undo_list *undo_list;
1659 if (clone_flags & CLONE_SYSVSEM) {
1660 error = get_undo_list(&undo_list);
1663 atomic_inc(&undo_list->refcnt);
1664 tsk->sysvsem.undo_list = undo_list;
1666 tsk->sysvsem.undo_list = NULL;
1672 * add semadj values to semaphores, free undo structures.
1673 * undo structures are not freed when semaphore arrays are destroyed
1674 * so some of them may be out of date.
1675 * IMPLEMENTATION NOTE: There is some confusion over whether the
1676 * set of adjustments that needs to be done should be done in an atomic
1677 * manner or not. That is, if we are attempting to decrement the semval
1678 * should we queue up and wait until we can do so legally?
1679 * The original implementation attempted to do this (queue and wait).
1680 * The current implementation does not do so. The POSIX standard
1681 * and SVID should be consulted to determine what behavior is mandated.
1683 void exit_sem(struct task_struct *tsk)
1685 struct sem_undo_list *ulp;
1687 ulp = tsk->sysvsem.undo_list;
1690 tsk->sysvsem.undo_list = NULL;
1692 if (!atomic_dec_and_test(&ulp->refcnt))
1696 struct sem_array *sma;
1697 struct sem_undo *un;
1698 struct list_head tasks;
1703 un = list_entry_rcu(ulp->list_proc.next,
1704 struct sem_undo, list_proc);
1705 if (&un->list_proc == &ulp->list_proc)
1714 sma = sem_lock_check(tsk->nsproxy->ipc_ns, un->semid);
1716 /* exit_sem raced with IPC_RMID, nothing to do */
1720 un = __lookup_undo(ulp, semid);
1722 /* exit_sem raced with IPC_RMID+semget() that created
1723 * exactly the same semid. Nothing to do.
1729 /* remove un from the linked lists */
1730 assert_spin_locked(&sma->sem_perm.lock);
1731 list_del(&un->list_id);
1733 spin_lock(&ulp->lock);
1734 list_del_rcu(&un->list_proc);
1735 spin_unlock(&ulp->lock);
1737 /* perform adjustments registered in un */
1738 for (i = 0; i < sma->sem_nsems; i++) {
1739 struct sem * semaphore = &sma->sem_base[i];
1740 if (un->semadj[i]) {
1741 semaphore->semval += un->semadj[i];
1743 * Range checks of the new semaphore value,
1744 * not defined by sus:
1745 * - Some unices ignore the undo entirely
1746 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
1747 * - some cap the value (e.g. FreeBSD caps
1748 * at 0, but doesn't enforce SEMVMX)
1750 * Linux caps the semaphore value, both at 0
1753 * Manfred <manfred@colorfullife.com>
1755 if (semaphore->semval < 0)
1756 semaphore->semval = 0;
1757 if (semaphore->semval > SEMVMX)
1758 semaphore->semval = SEMVMX;
1759 semaphore->sempid = task_tgid_vnr(current);
1762 /* maybe some queued-up processes were waiting for this */
1763 INIT_LIST_HEAD(&tasks);
1764 do_smart_update(sma, NULL, 0, 1, &tasks);
1766 wake_up_sem_queue_do(&tasks);
1773 #ifdef CONFIG_PROC_FS
1774 static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
1776 struct user_namespace *user_ns = seq_user_ns(s);
1777 struct sem_array *sma = it;
1779 return seq_printf(s,
1780 "%10d %10d %4o %10u %5u %5u %5u %5u %10lu %10lu\n",
1785 from_kuid_munged(user_ns, sma->sem_perm.uid),
1786 from_kgid_munged(user_ns, sma->sem_perm.gid),
1787 from_kuid_munged(user_ns, sma->sem_perm.cuid),
1788 from_kgid_munged(user_ns, sma->sem_perm.cgid),