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_obtain_lock(struct ipc_namespace *ns, int id)
199 struct kern_ipc_perm *ipcp;
202 ipcp = ipc_obtain_object(&sem_ids(ns), id);
206 spin_lock(&ipcp->lock);
208 /* ipc_rmid() may have already freed the ID while sem_lock
209 * was spinning: verify that the structure is still valid
214 return container_of(ipcp, struct sem_array, sem_perm);
216 spin_unlock(&ipcp->lock);
219 return ERR_PTR(-EINVAL);
222 static inline struct sem_array *sem_obtain_object(struct ipc_namespace *ns, int id)
224 struct kern_ipc_perm *ipcp = ipc_obtain_object(&sem_ids(ns), id);
227 return ERR_CAST(ipcp);
229 return container_of(ipcp, struct sem_array, sem_perm);
232 static inline struct sem_array *sem_lock_check(struct ipc_namespace *ns,
235 struct kern_ipc_perm *ipcp = ipc_lock_check(&sem_ids(ns), id);
238 return ERR_CAST(ipcp);
240 return container_of(ipcp, struct sem_array, sem_perm);
243 static inline struct sem_array *sem_obtain_object_check(struct ipc_namespace *ns,
246 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&sem_ids(ns), id);
249 return ERR_CAST(ipcp);
251 return container_of(ipcp, struct sem_array, sem_perm);
254 static inline void sem_lock_and_putref(struct sem_array *sma)
256 ipc_lock_by_ptr(&sma->sem_perm);
260 static inline void sem_getref_and_unlock(struct sem_array *sma)
263 ipc_unlock(&(sma)->sem_perm);
266 static inline void sem_putref(struct sem_array *sma)
268 ipc_lock_by_ptr(&sma->sem_perm);
270 ipc_unlock(&(sma)->sem_perm);
274 * Call inside the rcu read section.
276 static inline void sem_getref(struct sem_array *sma)
278 spin_lock(&(sma)->sem_perm.lock);
280 ipc_unlock(&(sma)->sem_perm);
283 static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
285 ipc_rmid(&sem_ids(ns), &s->sem_perm);
289 * Lockless wakeup algorithm:
290 * Without the check/retry algorithm a lockless wakeup is possible:
291 * - queue.status is initialized to -EINTR before blocking.
292 * - wakeup is performed by
293 * * unlinking the queue entry from sma->sem_pending
294 * * setting queue.status to IN_WAKEUP
295 * This is the notification for the blocked thread that a
296 * result value is imminent.
297 * * call wake_up_process
298 * * set queue.status to the final value.
299 * - the previously blocked thread checks queue.status:
300 * * if it's IN_WAKEUP, then it must wait until the value changes
301 * * if it's not -EINTR, then the operation was completed by
302 * update_queue. semtimedop can return queue.status without
303 * performing any operation on the sem array.
304 * * otherwise it must acquire the spinlock and check what's up.
306 * The two-stage algorithm is necessary to protect against the following
308 * - if queue.status is set after wake_up_process, then the woken up idle
309 * thread could race forward and try (and fail) to acquire sma->lock
310 * before update_queue had a chance to set queue.status
311 * - if queue.status is written before wake_up_process and if the
312 * blocked process is woken up by a signal between writing
313 * queue.status and the wake_up_process, then the woken up
314 * process could return from semtimedop and die by calling
315 * sys_exit before wake_up_process is called. Then wake_up_process
316 * will oops, because the task structure is already invalid.
317 * (yes, this happened on s390 with sysv msg).
323 * newary - Create a new semaphore set
325 * @params: ptr to the structure that contains key, semflg and nsems
327 * Called with sem_ids.rw_mutex held (as a writer)
330 static int newary(struct ipc_namespace *ns, struct ipc_params *params)
334 struct sem_array *sma;
336 key_t key = params->key;
337 int nsems = params->u.nsems;
338 int semflg = params->flg;
343 if (ns->used_sems + nsems > ns->sc_semmns)
346 size = sizeof (*sma) + nsems * sizeof (struct sem);
347 sma = ipc_rcu_alloc(size);
351 memset (sma, 0, size);
353 sma->sem_perm.mode = (semflg & S_IRWXUGO);
354 sma->sem_perm.key = key;
356 sma->sem_perm.security = NULL;
357 retval = security_sem_alloc(sma);
363 id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
365 security_sem_free(sma);
369 ns->used_sems += nsems;
371 sma->sem_base = (struct sem *) &sma[1];
373 for (i = 0; i < nsems; i++)
374 INIT_LIST_HEAD(&sma->sem_base[i].sem_pending);
376 sma->complex_count = 0;
377 INIT_LIST_HEAD(&sma->sem_pending);
378 INIT_LIST_HEAD(&sma->list_id);
379 sma->sem_nsems = nsems;
380 sma->sem_ctime = get_seconds();
383 return sma->sem_perm.id;
388 * Called with sem_ids.rw_mutex and ipcp locked.
390 static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg)
392 struct sem_array *sma;
394 sma = container_of(ipcp, struct sem_array, sem_perm);
395 return security_sem_associate(sma, semflg);
399 * Called with sem_ids.rw_mutex and ipcp locked.
401 static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
402 struct ipc_params *params)
404 struct sem_array *sma;
406 sma = container_of(ipcp, struct sem_array, sem_perm);
407 if (params->u.nsems > sma->sem_nsems)
413 SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg)
415 struct ipc_namespace *ns;
416 struct ipc_ops sem_ops;
417 struct ipc_params sem_params;
419 ns = current->nsproxy->ipc_ns;
421 if (nsems < 0 || nsems > ns->sc_semmsl)
424 sem_ops.getnew = newary;
425 sem_ops.associate = sem_security;
426 sem_ops.more_checks = sem_more_checks;
428 sem_params.key = key;
429 sem_params.flg = semflg;
430 sem_params.u.nsems = nsems;
432 return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
436 * Determine whether a sequence of semaphore operations would succeed
437 * all at once. Return 0 if yes, 1 if need to sleep, else return error code.
440 static int try_atomic_semop (struct sem_array * sma, struct sembuf * sops,
441 int nsops, struct sem_undo *un, int pid)
447 for (sop = sops; sop < sops + nsops; sop++) {
448 curr = sma->sem_base + sop->sem_num;
449 sem_op = sop->sem_op;
450 result = curr->semval;
452 if (!sem_op && result)
460 if (sop->sem_flg & SEM_UNDO) {
461 int undo = un->semadj[sop->sem_num] - sem_op;
463 * Exceeding the undo range is an error.
465 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
468 curr->semval = result;
472 while (sop >= sops) {
473 sma->sem_base[sop->sem_num].sempid = pid;
474 if (sop->sem_flg & SEM_UNDO)
475 un->semadj[sop->sem_num] -= sop->sem_op;
486 if (sop->sem_flg & IPC_NOWAIT)
493 while (sop >= sops) {
494 sma->sem_base[sop->sem_num].semval -= sop->sem_op;
501 /** wake_up_sem_queue_prepare(q, error): Prepare wake-up
502 * @q: queue entry that must be signaled
503 * @error: Error value for the signal
505 * Prepare the wake-up of the queue entry q.
507 static void wake_up_sem_queue_prepare(struct list_head *pt,
508 struct sem_queue *q, int error)
510 if (list_empty(pt)) {
512 * Hold preempt off so that we don't get preempted and have the
513 * wakee busy-wait until we're scheduled back on.
517 q->status = IN_WAKEUP;
520 list_add_tail(&q->simple_list, pt);
524 * wake_up_sem_queue_do(pt) - do the actual wake-up
525 * @pt: list of tasks to be woken up
527 * Do the actual wake-up.
528 * The function is called without any locks held, thus the semaphore array
529 * could be destroyed already and the tasks can disappear as soon as the
530 * status is set to the actual return code.
532 static void wake_up_sem_queue_do(struct list_head *pt)
534 struct sem_queue *q, *t;
537 did_something = !list_empty(pt);
538 list_for_each_entry_safe(q, t, pt, simple_list) {
539 wake_up_process(q->sleeper);
540 /* q can disappear immediately after writing q->status. */
548 static void unlink_queue(struct sem_array *sma, struct sem_queue *q)
552 list_del(&q->simple_list);
554 sma->complex_count--;
557 /** check_restart(sma, q)
558 * @sma: semaphore array
559 * @q: the operation that just completed
561 * update_queue is O(N^2) when it restarts scanning the whole queue of
562 * waiting operations. Therefore this function checks if the restart is
563 * really necessary. It is called after a previously waiting operation
566 static int check_restart(struct sem_array *sma, struct sem_queue *q)
571 /* if the operation didn't modify the array, then no restart */
575 /* pending complex operations are too difficult to analyse */
576 if (sma->complex_count)
579 /* we were a sleeping complex operation. Too difficult */
583 curr = sma->sem_base + q->sops[0].sem_num;
585 /* No-one waits on this queue */
586 if (list_empty(&curr->sem_pending))
589 /* the new semaphore value */
591 /* It is impossible that someone waits for the new value:
592 * - q is a previously sleeping simple operation that
593 * altered the array. It must be a decrement, because
594 * simple increments never sleep.
595 * - The value is not 0, thus wait-for-zero won't proceed.
596 * - If there are older (higher priority) decrements
597 * in the queue, then they have observed the original
598 * semval value and couldn't proceed. The operation
599 * decremented to value - thus they won't proceed either.
601 BUG_ON(q->sops[0].sem_op >= 0);
605 * semval is 0. Check if there are wait-for-zero semops.
606 * They must be the first entries in the per-semaphore simple queue
608 h = list_first_entry(&curr->sem_pending, struct sem_queue, simple_list);
609 BUG_ON(h->nsops != 1);
610 BUG_ON(h->sops[0].sem_num != q->sops[0].sem_num);
612 /* Yes, there is a wait-for-zero semop. Restart */
613 if (h->sops[0].sem_op == 0)
616 /* Again - no-one is waiting for the new value. */
622 * update_queue(sma, semnum): Look for tasks that can be completed.
623 * @sma: semaphore array.
624 * @semnum: semaphore that was modified.
625 * @pt: list head for the tasks that must be woken up.
627 * update_queue must be called after a semaphore in a semaphore array
628 * was modified. If multiple semaphore were modified, then @semnum
630 * The tasks that must be woken up are added to @pt. The return code
631 * is stored in q->pid.
632 * The function return 1 if at least one semop was completed successfully.
634 static int update_queue(struct sem_array *sma, int semnum, struct list_head *pt)
637 struct list_head *walk;
638 struct list_head *pending_list;
640 int semop_completed = 0;
642 /* if there are complex operations around, then knowing the semaphore
643 * that was modified doesn't help us. Assume that multiple semaphores
646 if (sma->complex_count)
650 pending_list = &sma->sem_pending;
651 offset = offsetof(struct sem_queue, list);
653 pending_list = &sma->sem_base[semnum].sem_pending;
654 offset = offsetof(struct sem_queue, simple_list);
658 walk = pending_list->next;
659 while (walk != pending_list) {
662 q = (struct sem_queue *)((char *)walk - offset);
665 /* If we are scanning the single sop, per-semaphore list of
666 * one semaphore and that semaphore is 0, then it is not
667 * necessary to scan the "alter" entries: simple increments
668 * that affect only one entry succeed immediately and cannot
669 * be in the per semaphore pending queue, and decrements
670 * cannot be successful if the value is already 0.
672 if (semnum != -1 && sma->sem_base[semnum].semval == 0 &&
676 error = try_atomic_semop(sma, q->sops, q->nsops,
679 /* Does q->sleeper still need to sleep? */
683 unlink_queue(sma, q);
689 restart = check_restart(sma, q);
692 wake_up_sem_queue_prepare(pt, q, error);
696 return semop_completed;
700 * do_smart_update(sma, sops, nsops, otime, pt) - optimized update_queue
701 * @sma: semaphore array
702 * @sops: operations that were performed
703 * @nsops: number of operations
704 * @otime: force setting otime
705 * @pt: list head of the tasks that must be woken up.
707 * do_smart_update() does the required called to update_queue, based on the
708 * actual changes that were performed on the semaphore array.
709 * Note that the function does not do the actual wake-up: the caller is
710 * responsible for calling wake_up_sem_queue_do(@pt).
711 * It is safe to perform this call after dropping all locks.
713 static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops,
714 int otime, struct list_head *pt)
718 if (sma->complex_count || sops == NULL) {
719 if (update_queue(sma, -1, pt))
724 for (i = 0; i < nsops; i++) {
725 if (sops[i].sem_op > 0 ||
726 (sops[i].sem_op < 0 &&
727 sma->sem_base[sops[i].sem_num].semval == 0))
728 if (update_queue(sma, sops[i].sem_num, pt))
733 sma->sem_otime = get_seconds();
737 /* The following counts are associated to each semaphore:
738 * semncnt number of tasks waiting on semval being nonzero
739 * semzcnt number of tasks waiting on semval being zero
740 * This model assumes that a task waits on exactly one semaphore.
741 * Since semaphore operations are to be performed atomically, tasks actually
742 * wait on a whole sequence of semaphores simultaneously.
743 * The counts we return here are a rough approximation, but still
744 * warrant that semncnt+semzcnt>0 if the task is on the pending queue.
746 static int count_semncnt (struct sem_array * sma, ushort semnum)
749 struct sem_queue * q;
752 list_for_each_entry(q, &sma->sem_pending, list) {
753 struct sembuf * sops = q->sops;
754 int nsops = q->nsops;
756 for (i = 0; i < nsops; i++)
757 if (sops[i].sem_num == semnum
758 && (sops[i].sem_op < 0)
759 && !(sops[i].sem_flg & IPC_NOWAIT))
765 static int count_semzcnt (struct sem_array * sma, ushort semnum)
768 struct sem_queue * q;
771 list_for_each_entry(q, &sma->sem_pending, list) {
772 struct sembuf * sops = q->sops;
773 int nsops = q->nsops;
775 for (i = 0; i < nsops; i++)
776 if (sops[i].sem_num == semnum
777 && (sops[i].sem_op == 0)
778 && !(sops[i].sem_flg & IPC_NOWAIT))
784 /* Free a semaphore set. freeary() is called with sem_ids.rw_mutex locked
785 * as a writer and the spinlock for this semaphore set hold. sem_ids.rw_mutex
786 * remains locked on exit.
788 static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
790 struct sem_undo *un, *tu;
791 struct sem_queue *q, *tq;
792 struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
793 struct list_head tasks;
795 /* Free the existing undo structures for this semaphore set. */
796 assert_spin_locked(&sma->sem_perm.lock);
797 list_for_each_entry_safe(un, tu, &sma->list_id, list_id) {
798 list_del(&un->list_id);
799 spin_lock(&un->ulp->lock);
801 list_del_rcu(&un->list_proc);
802 spin_unlock(&un->ulp->lock);
806 /* Wake up all pending processes and let them fail with EIDRM. */
807 INIT_LIST_HEAD(&tasks);
808 list_for_each_entry_safe(q, tq, &sma->sem_pending, list) {
809 unlink_queue(sma, q);
810 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
813 /* Remove the semaphore set from the IDR */
817 wake_up_sem_queue_do(&tasks);
818 ns->used_sems -= sma->sem_nsems;
819 security_sem_free(sma);
823 static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
827 return copy_to_user(buf, in, sizeof(*in));
832 memset(&out, 0, sizeof(out));
834 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
836 out.sem_otime = in->sem_otime;
837 out.sem_ctime = in->sem_ctime;
838 out.sem_nsems = in->sem_nsems;
840 return copy_to_user(buf, &out, sizeof(out));
847 static int semctl_nolock(struct ipc_namespace *ns, int semid,
848 int cmd, int version, void __user *p)
851 struct sem_array *sma;
857 struct seminfo seminfo;
860 err = security_sem_semctl(NULL, cmd);
864 memset(&seminfo,0,sizeof(seminfo));
865 seminfo.semmni = ns->sc_semmni;
866 seminfo.semmns = ns->sc_semmns;
867 seminfo.semmsl = ns->sc_semmsl;
868 seminfo.semopm = ns->sc_semopm;
869 seminfo.semvmx = SEMVMX;
870 seminfo.semmnu = SEMMNU;
871 seminfo.semmap = SEMMAP;
872 seminfo.semume = SEMUME;
873 down_read(&sem_ids(ns).rw_mutex);
874 if (cmd == SEM_INFO) {
875 seminfo.semusz = sem_ids(ns).in_use;
876 seminfo.semaem = ns->used_sems;
878 seminfo.semusz = SEMUSZ;
879 seminfo.semaem = SEMAEM;
881 max_id = ipc_get_maxid(&sem_ids(ns));
882 up_read(&sem_ids(ns).rw_mutex);
883 if (copy_to_user(p, &seminfo, sizeof(struct seminfo)))
885 return (max_id < 0) ? 0: max_id;
890 struct semid64_ds tbuf;
893 memset(&tbuf, 0, sizeof(tbuf));
895 if (cmd == SEM_STAT) {
897 sma = sem_obtain_object(ns, semid);
902 id = sma->sem_perm.id;
905 sma = sem_obtain_object_check(ns, semid);
913 if (ipcperms(ns, &sma->sem_perm, S_IRUGO))
916 err = security_sem_semctl(sma, cmd);
920 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
921 tbuf.sem_otime = sma->sem_otime;
922 tbuf.sem_ctime = sma->sem_ctime;
923 tbuf.sem_nsems = sma->sem_nsems;
925 if (copy_semid_to_user(p, &tbuf, version))
937 static int semctl_setval(struct ipc_namespace *ns, int semid, int semnum,
941 struct sem_array *sma;
945 struct list_head tasks;
947 #if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN)
948 /* big-endian 64bit */
951 /* 32bit or little-endian 64bit */
955 sma = sem_lock_check(ns, semid);
959 INIT_LIST_HEAD(&tasks);
960 nsems = sma->sem_nsems;
963 if (ipcperms(ns, &sma->sem_perm, S_IWUGO))
966 err = security_sem_semctl(sma, SETVAL);
971 if(semnum < 0 || semnum >= nsems)
974 curr = &sma->sem_base[semnum];
977 if (val > SEMVMX || val < 0)
980 assert_spin_locked(&sma->sem_perm.lock);
981 list_for_each_entry(un, &sma->list_id, list_id)
982 un->semadj[semnum] = 0;
985 curr->sempid = task_tgid_vnr(current);
986 sma->sem_ctime = get_seconds();
987 /* maybe some queued-up processes were waiting for this */
988 do_smart_update(sma, NULL, 0, 0, &tasks);
992 wake_up_sem_queue_do(&tasks);
996 static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
997 int cmd, void __user *p)
999 struct sem_array *sma;
1002 ushort fast_sem_io[SEMMSL_FAST];
1003 ushort* sem_io = fast_sem_io;
1004 struct list_head tasks;
1006 INIT_LIST_HEAD(&tasks);
1009 sma = sem_obtain_object_check(ns, semid);
1012 return PTR_ERR(sma);
1015 nsems = sma->sem_nsems;
1018 if (ipcperms(ns, &sma->sem_perm,
1019 cmd == SETALL ? S_IWUGO : S_IRUGO)) {
1024 err = security_sem_semctl(sma, cmd);
1034 ushort __user *array = p;
1037 if(nsems > SEMMSL_FAST) {
1040 sem_io = ipc_alloc(sizeof(ushort)*nsems);
1041 if(sem_io == NULL) {
1046 sem_lock_and_putref(sma);
1047 if (sma->sem_perm.deleted) {
1054 spin_lock(&sma->sem_perm.lock);
1055 for (i = 0; i < sma->sem_nsems; i++)
1056 sem_io[i] = sma->sem_base[i].semval;
1059 if(copy_to_user(array, sem_io, nsems*sizeof(ushort)))
1066 struct sem_undo *un;
1068 ipc_rcu_getref(sma);
1071 if(nsems > SEMMSL_FAST) {
1072 sem_io = ipc_alloc(sizeof(ushort)*nsems);
1073 if(sem_io == NULL) {
1079 if (copy_from_user (sem_io, p, nsems*sizeof(ushort))) {
1085 for (i = 0; i < nsems; i++) {
1086 if (sem_io[i] > SEMVMX) {
1092 sem_lock_and_putref(sma);
1093 if (sma->sem_perm.deleted) {
1099 for (i = 0; i < nsems; i++)
1100 sma->sem_base[i].semval = sem_io[i];
1102 assert_spin_locked(&sma->sem_perm.lock);
1103 list_for_each_entry(un, &sma->list_id, list_id) {
1104 for (i = 0; i < nsems; i++)
1107 sma->sem_ctime = get_seconds();
1108 /* maybe some queued-up processes were waiting for this */
1109 do_smart_update(sma, NULL, 0, 0, &tasks);
1113 /* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */
1116 if(semnum < 0 || semnum >= nsems)
1119 spin_lock(&sma->sem_perm.lock);
1120 curr = &sma->sem_base[semnum];
1130 err = count_semncnt(sma,semnum);
1133 err = count_semzcnt(sma,semnum);
1140 wake_up_sem_queue_do(&tasks);
1142 if(sem_io != fast_sem_io)
1143 ipc_free(sem_io, sizeof(ushort)*nsems);
1147 static inline unsigned long
1148 copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
1152 if (copy_from_user(out, buf, sizeof(*out)))
1157 struct semid_ds tbuf_old;
1159 if(copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
1162 out->sem_perm.uid = tbuf_old.sem_perm.uid;
1163 out->sem_perm.gid = tbuf_old.sem_perm.gid;
1164 out->sem_perm.mode = tbuf_old.sem_perm.mode;
1174 * This function handles some semctl commands which require the rw_mutex
1175 * to be held in write mode.
1176 * NOTE: no locks must be held, the rw_mutex is taken inside this function.
1178 static int semctl_down(struct ipc_namespace *ns, int semid,
1179 int cmd, int version, void __user *p)
1181 struct sem_array *sma;
1183 struct semid64_ds semid64;
1184 struct kern_ipc_perm *ipcp;
1186 if(cmd == IPC_SET) {
1187 if (copy_semid_from_user(&semid64, p, version))
1191 ipcp = ipcctl_pre_down_nolock(ns, &sem_ids(ns), semid, cmd,
1192 &semid64.sem_perm, 0);
1194 return PTR_ERR(ipcp);
1196 sma = container_of(ipcp, struct sem_array, sem_perm);
1198 err = security_sem_semctl(sma, cmd);
1206 ipc_lock_object(&sma->sem_perm);
1210 ipc_lock_object(&sma->sem_perm);
1211 err = ipc_update_perm(&semid64.sem_perm, ipcp);
1214 sma->sem_ctime = get_seconds();
1225 up_write(&sem_ids(ns).rw_mutex);
1229 SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, unsigned long, arg)
1232 struct ipc_namespace *ns;
1233 void __user *p = (void __user *)arg;
1238 version = ipc_parse_version(&cmd);
1239 ns = current->nsproxy->ipc_ns;
1246 return semctl_nolock(ns, semid, cmd, version, p);
1253 return semctl_main(ns, semid, semnum, cmd, p);
1255 return semctl_setval(ns, semid, semnum, arg);
1258 return semctl_down(ns, semid, cmd, version, p);
1264 /* If the task doesn't already have a undo_list, then allocate one
1265 * here. We guarantee there is only one thread using this undo list,
1266 * and current is THE ONE
1268 * If this allocation and assignment succeeds, but later
1269 * portions of this code fail, there is no need to free the sem_undo_list.
1270 * Just let it stay associated with the task, and it'll be freed later
1273 * This can block, so callers must hold no locks.
1275 static inline int get_undo_list(struct sem_undo_list **undo_listp)
1277 struct sem_undo_list *undo_list;
1279 undo_list = current->sysvsem.undo_list;
1281 undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
1282 if (undo_list == NULL)
1284 spin_lock_init(&undo_list->lock);
1285 atomic_set(&undo_list->refcnt, 1);
1286 INIT_LIST_HEAD(&undo_list->list_proc);
1288 current->sysvsem.undo_list = undo_list;
1290 *undo_listp = undo_list;
1294 static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid)
1296 struct sem_undo *un;
1298 list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) {
1299 if (un->semid == semid)
1305 static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1307 struct sem_undo *un;
1309 assert_spin_locked(&ulp->lock);
1311 un = __lookup_undo(ulp, semid);
1313 list_del_rcu(&un->list_proc);
1314 list_add_rcu(&un->list_proc, &ulp->list_proc);
1320 * find_alloc_undo - Lookup (and if not present create) undo array
1322 * @semid: semaphore array id
1324 * The function looks up (and if not present creates) the undo structure.
1325 * The size of the undo structure depends on the size of the semaphore
1326 * array, thus the alloc path is not that straightforward.
1327 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1328 * performs a rcu_read_lock().
1330 static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
1332 struct sem_array *sma;
1333 struct sem_undo_list *ulp;
1334 struct sem_undo *un, *new;
1338 error = get_undo_list(&ulp);
1340 return ERR_PTR(error);
1343 spin_lock(&ulp->lock);
1344 un = lookup_undo(ulp, semid);
1345 spin_unlock(&ulp->lock);
1346 if (likely(un!=NULL))
1349 /* no undo structure around - allocate one. */
1350 /* step 1: figure out the size of the semaphore array */
1351 sma = sem_obtain_object_check(ns, semid);
1354 return ERR_CAST(sma);
1357 nsems = sma->sem_nsems;
1358 ipc_rcu_getref(sma);
1361 /* step 2: allocate new undo structure */
1362 new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1365 return ERR_PTR(-ENOMEM);
1368 /* step 3: Acquire the lock on semaphore array */
1369 sem_lock_and_putref(sma);
1370 if (sma->sem_perm.deleted) {
1373 un = ERR_PTR(-EIDRM);
1376 spin_lock(&ulp->lock);
1379 * step 4: check for races: did someone else allocate the undo struct?
1381 un = lookup_undo(ulp, semid);
1386 /* step 5: initialize & link new undo structure */
1387 new->semadj = (short *) &new[1];
1390 assert_spin_locked(&ulp->lock);
1391 list_add_rcu(&new->list_proc, &ulp->list_proc);
1392 assert_spin_locked(&sma->sem_perm.lock);
1393 list_add(&new->list_id, &sma->list_id);
1397 spin_unlock(&ulp->lock);
1406 * get_queue_result - Retrieve the result code from sem_queue
1407 * @q: Pointer to queue structure
1409 * Retrieve the return code from the pending queue. If IN_WAKEUP is found in
1410 * q->status, then we must loop until the value is replaced with the final
1411 * value: This may happen if a task is woken up by an unrelated event (e.g.
1412 * signal) and in parallel the task is woken up by another task because it got
1413 * the requested semaphores.
1415 * The function can be called with or without holding the semaphore spinlock.
1417 static int get_queue_result(struct sem_queue *q)
1422 while (unlikely(error == IN_WAKEUP)) {
1431 SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
1432 unsigned, nsops, const struct timespec __user *, timeout)
1434 int error = -EINVAL;
1435 struct sem_array *sma;
1436 struct sembuf fast_sops[SEMOPM_FAST];
1437 struct sembuf* sops = fast_sops, *sop;
1438 struct sem_undo *un;
1439 int undos = 0, alter = 0, max;
1440 struct sem_queue queue;
1441 unsigned long jiffies_left = 0;
1442 struct ipc_namespace *ns;
1443 struct list_head tasks;
1445 ns = current->nsproxy->ipc_ns;
1447 if (nsops < 1 || semid < 0)
1449 if (nsops > ns->sc_semopm)
1451 if(nsops > SEMOPM_FAST) {
1452 sops = kmalloc(sizeof(*sops)*nsops,GFP_KERNEL);
1456 if (copy_from_user (sops, tsops, nsops * sizeof(*tsops))) {
1461 struct timespec _timeout;
1462 if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1466 if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1467 _timeout.tv_nsec >= 1000000000L) {
1471 jiffies_left = timespec_to_jiffies(&_timeout);
1474 for (sop = sops; sop < sops + nsops; sop++) {
1475 if (sop->sem_num >= max)
1477 if (sop->sem_flg & SEM_UNDO)
1479 if (sop->sem_op != 0)
1484 un = find_alloc_undo(ns, semid);
1486 error = PTR_ERR(un);
1492 INIT_LIST_HEAD(&tasks);
1495 sma = sem_obtain_object_check(ns, semid);
1499 error = PTR_ERR(sma);
1504 if (max >= sma->sem_nsems) {
1510 if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO)) {
1515 error = security_sem_semop(sma, sops, nsops, alter);
1522 * semid identifiers are not unique - find_alloc_undo may have
1523 * allocated an undo structure, it was invalidated by an RMID
1524 * and now a new array with received the same id. Check and fail.
1525 * This case can be detected checking un->semid. The existence of
1526 * "un" itself is guaranteed by rcu.
1529 ipc_lock_object(&sma->sem_perm);
1531 if (un->semid == -1) {
1533 goto out_unlock_free;
1536 * rcu lock can be released, "un" cannot disappear:
1537 * - sem_lock is acquired, thus IPC_RMID is
1539 * - exit_sem is impossible, it always operates on
1540 * current (or a dead task).
1547 error = try_atomic_semop (sma, sops, nsops, un, task_tgid_vnr(current));
1549 if (alter && error == 0)
1550 do_smart_update(sma, sops, nsops, 1, &tasks);
1552 goto out_unlock_free;
1555 /* We need to sleep on this operation, so we put the current
1556 * task into the pending queue and go to sleep.
1560 queue.nsops = nsops;
1562 queue.pid = task_tgid_vnr(current);
1563 queue.alter = alter;
1565 list_add_tail(&queue.list, &sma->sem_pending);
1567 list_add(&queue.list, &sma->sem_pending);
1571 curr = &sma->sem_base[sops->sem_num];
1574 list_add_tail(&queue.simple_list, &curr->sem_pending);
1576 list_add(&queue.simple_list, &curr->sem_pending);
1578 INIT_LIST_HEAD(&queue.simple_list);
1579 sma->complex_count++;
1582 queue.status = -EINTR;
1583 queue.sleeper = current;
1586 current->state = TASK_INTERRUPTIBLE;
1590 jiffies_left = schedule_timeout(jiffies_left);
1594 error = get_queue_result(&queue);
1596 if (error != -EINTR) {
1597 /* fast path: update_queue already obtained all requested
1599 * Perform a smp_mb(): User space could assume that semop()
1600 * is a memory barrier: Without the mb(), the cpu could
1601 * speculatively read in user space stale data that was
1602 * overwritten by the previous owner of the semaphore.
1609 sma = sem_obtain_lock(ns, semid);
1612 * Wait until it's guaranteed that no wakeup_sem_queue_do() is ongoing.
1614 error = get_queue_result(&queue);
1617 * Array removed? If yes, leave without sem_unlock().
1625 * If queue.status != -EINTR we are woken up by another process.
1626 * Leave without unlink_queue(), but with sem_unlock().
1629 if (error != -EINTR) {
1630 goto out_unlock_free;
1634 * If an interrupt occurred we have to clean up the queue
1636 if (timeout && jiffies_left == 0)
1640 * If the wakeup was spurious, just retry
1642 if (error == -EINTR && !signal_pending(current))
1645 unlink_queue(sma, &queue);
1650 wake_up_sem_queue_do(&tasks);
1652 if(sops != fast_sops)
1657 SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
1660 return sys_semtimedop(semid, tsops, nsops, NULL);
1663 /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
1664 * parent and child tasks.
1667 int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
1669 struct sem_undo_list *undo_list;
1672 if (clone_flags & CLONE_SYSVSEM) {
1673 error = get_undo_list(&undo_list);
1676 atomic_inc(&undo_list->refcnt);
1677 tsk->sysvsem.undo_list = undo_list;
1679 tsk->sysvsem.undo_list = NULL;
1685 * add semadj values to semaphores, free undo structures.
1686 * undo structures are not freed when semaphore arrays are destroyed
1687 * so some of them may be out of date.
1688 * IMPLEMENTATION NOTE: There is some confusion over whether the
1689 * set of adjustments that needs to be done should be done in an atomic
1690 * manner or not. That is, if we are attempting to decrement the semval
1691 * should we queue up and wait until we can do so legally?
1692 * The original implementation attempted to do this (queue and wait).
1693 * The current implementation does not do so. The POSIX standard
1694 * and SVID should be consulted to determine what behavior is mandated.
1696 void exit_sem(struct task_struct *tsk)
1698 struct sem_undo_list *ulp;
1700 ulp = tsk->sysvsem.undo_list;
1703 tsk->sysvsem.undo_list = NULL;
1705 if (!atomic_dec_and_test(&ulp->refcnt))
1709 struct sem_array *sma;
1710 struct sem_undo *un;
1711 struct list_head tasks;
1716 un = list_entry_rcu(ulp->list_proc.next,
1717 struct sem_undo, list_proc);
1718 if (&un->list_proc == &ulp->list_proc)
1727 sma = sem_lock_check(tsk->nsproxy->ipc_ns, un->semid);
1729 /* exit_sem raced with IPC_RMID, nothing to do */
1733 un = __lookup_undo(ulp, semid);
1735 /* exit_sem raced with IPC_RMID+semget() that created
1736 * exactly the same semid. Nothing to do.
1742 /* remove un from the linked lists */
1743 assert_spin_locked(&sma->sem_perm.lock);
1744 list_del(&un->list_id);
1746 spin_lock(&ulp->lock);
1747 list_del_rcu(&un->list_proc);
1748 spin_unlock(&ulp->lock);
1750 /* perform adjustments registered in un */
1751 for (i = 0; i < sma->sem_nsems; i++) {
1752 struct sem * semaphore = &sma->sem_base[i];
1753 if (un->semadj[i]) {
1754 semaphore->semval += un->semadj[i];
1756 * Range checks of the new semaphore value,
1757 * not defined by sus:
1758 * - Some unices ignore the undo entirely
1759 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
1760 * - some cap the value (e.g. FreeBSD caps
1761 * at 0, but doesn't enforce SEMVMX)
1763 * Linux caps the semaphore value, both at 0
1766 * Manfred <manfred@colorfullife.com>
1768 if (semaphore->semval < 0)
1769 semaphore->semval = 0;
1770 if (semaphore->semval > SEMVMX)
1771 semaphore->semval = SEMVMX;
1772 semaphore->sempid = task_tgid_vnr(current);
1775 /* maybe some queued-up processes were waiting for this */
1776 INIT_LIST_HEAD(&tasks);
1777 do_smart_update(sma, NULL, 0, 1, &tasks);
1779 wake_up_sem_queue_do(&tasks);
1786 #ifdef CONFIG_PROC_FS
1787 static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
1789 struct user_namespace *user_ns = seq_user_ns(s);
1790 struct sem_array *sma = it;
1792 return seq_printf(s,
1793 "%10d %10d %4o %10u %5u %5u %5u %5u %10lu %10lu\n",
1798 from_kuid_munged(user_ns, sma->sem_perm.uid),
1799 from_kgid_munged(user_ns, sma->sem_perm.gid),
1800 from_kuid_munged(user_ns, sma->sem_perm.cuid),
1801 from_kgid_munged(user_ns, sma->sem_perm.cgid),