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
63 * (c) 2003 Manfred Spraul <manfred@colorfullife.com>
65 * support for audit of ipc object properties and permission changes
66 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
70 * Pavel Emelianov <xemul@openvz.org>
73 #include <linux/slab.h>
74 #include <linux/spinlock.h>
75 #include <linux/init.h>
76 #include <linux/proc_fs.h>
77 #include <linux/time.h>
78 #include <linux/security.h>
79 #include <linux/syscalls.h>
80 #include <linux/audit.h>
81 #include <linux/capability.h>
82 #include <linux/seq_file.h>
83 #include <linux/rwsem.h>
84 #include <linux/nsproxy.h>
85 #include <linux/ipc_namespace.h>
87 #include <asm/uaccess.h>
90 #define sem_ids(ns) ((ns)->ids[IPC_SEM_IDS])
92 #define sem_unlock(sma) ipc_unlock(&(sma)->sem_perm)
93 #define sem_checkid(sma, semid) ipc_checkid(&sma->sem_perm, semid)
95 static int newary(struct ipc_namespace
*, struct ipc_params
*);
96 static void freeary(struct ipc_namespace
*, struct kern_ipc_perm
*);
98 static int sysvipc_sem_proc_show(struct seq_file
*s
, void *it
);
101 #define SEMMSL_FAST 256 /* 512 bytes on stack */
102 #define SEMOPM_FAST 64 /* ~ 372 bytes on stack */
105 * linked list protection:
107 * sem_array.sem_pending{,last},
108 * sem_array.sem_undo: sem_lock() for read/write
109 * sem_undo.proc_next: only "current" is allowed to read/write that field.
113 #define sc_semmsl sem_ctls[0]
114 #define sc_semmns sem_ctls[1]
115 #define sc_semopm sem_ctls[2]
116 #define sc_semmni sem_ctls[3]
118 void sem_init_ns(struct ipc_namespace
*ns
)
120 ns
->sc_semmsl
= SEMMSL
;
121 ns
->sc_semmns
= SEMMNS
;
122 ns
->sc_semopm
= SEMOPM
;
123 ns
->sc_semmni
= SEMMNI
;
125 ipc_init_ids(&ns
->ids
[IPC_SEM_IDS
]);
129 void sem_exit_ns(struct ipc_namespace
*ns
)
131 free_ipcs(ns
, &sem_ids(ns
), freeary
);
132 idr_destroy(&ns
->ids
[IPC_SEM_IDS
].ipcs_idr
);
136 void __init
sem_init (void)
138 sem_init_ns(&init_ipc_ns
);
139 ipc_init_proc_interface("sysvipc/sem",
140 " key semid perms nsems uid gid cuid cgid otime ctime\n",
141 IPC_SEM_IDS
, sysvipc_sem_proc_show
);
145 * sem_lock_(check_) routines are called in the paths where the rw_mutex
148 static inline struct sem_array
*sem_lock(struct ipc_namespace
*ns
, int id
)
150 struct kern_ipc_perm
*ipcp
= ipc_lock(&sem_ids(ns
), id
);
153 return (struct sem_array
*)ipcp
;
155 return container_of(ipcp
, struct sem_array
, sem_perm
);
158 static inline struct sem_array
*sem_lock_check(struct ipc_namespace
*ns
,
161 struct kern_ipc_perm
*ipcp
= ipc_lock_check(&sem_ids(ns
), id
);
164 return (struct sem_array
*)ipcp
;
166 return container_of(ipcp
, struct sem_array
, sem_perm
);
169 static inline void sem_lock_and_putref(struct sem_array
*sma
)
171 ipc_lock_by_ptr(&sma
->sem_perm
);
175 static inline void sem_getref_and_unlock(struct sem_array
*sma
)
178 ipc_unlock(&(sma
)->sem_perm
);
181 static inline void sem_putref(struct sem_array
*sma
)
183 ipc_lock_by_ptr(&sma
->sem_perm
);
185 ipc_unlock(&(sma
)->sem_perm
);
188 static inline void sem_rmid(struct ipc_namespace
*ns
, struct sem_array
*s
)
190 ipc_rmid(&sem_ids(ns
), &s
->sem_perm
);
194 * Lockless wakeup algorithm:
195 * Without the check/retry algorithm a lockless wakeup is possible:
196 * - queue.status is initialized to -EINTR before blocking.
197 * - wakeup is performed by
198 * * unlinking the queue entry from sma->sem_pending
199 * * setting queue.status to IN_WAKEUP
200 * This is the notification for the blocked thread that a
201 * result value is imminent.
202 * * call wake_up_process
203 * * set queue.status to the final value.
204 * - the previously blocked thread checks queue.status:
205 * * if it's IN_WAKEUP, then it must wait until the value changes
206 * * if it's not -EINTR, then the operation was completed by
207 * update_queue. semtimedop can return queue.status without
208 * performing any operation on the sem array.
209 * * otherwise it must acquire the spinlock and check what's up.
211 * The two-stage algorithm is necessary to protect against the following
213 * - if queue.status is set after wake_up_process, then the woken up idle
214 * thread could race forward and try (and fail) to acquire sma->lock
215 * before update_queue had a chance to set queue.status
216 * - if queue.status is written before wake_up_process and if the
217 * blocked process is woken up by a signal between writing
218 * queue.status and the wake_up_process, then the woken up
219 * process could return from semtimedop and die by calling
220 * sys_exit before wake_up_process is called. Then wake_up_process
221 * will oops, because the task structure is already invalid.
222 * (yes, this happened on s390 with sysv msg).
228 * newary - Create a new semaphore set
230 * @params: ptr to the structure that contains key, semflg and nsems
232 * Called with sem_ids.rw_mutex held (as a writer)
235 static int newary(struct ipc_namespace
*ns
, struct ipc_params
*params
)
239 struct sem_array
*sma
;
241 key_t key
= params
->key
;
242 int nsems
= params
->u
.nsems
;
243 int semflg
= params
->flg
;
247 if (ns
->used_sems
+ nsems
> ns
->sc_semmns
)
250 size
= sizeof (*sma
) + nsems
* sizeof (struct sem
);
251 sma
= ipc_rcu_alloc(size
);
255 memset (sma
, 0, size
);
257 sma
->sem_perm
.mode
= (semflg
& S_IRWXUGO
);
258 sma
->sem_perm
.key
= key
;
260 sma
->sem_perm
.security
= NULL
;
261 retval
= security_sem_alloc(sma
);
267 id
= ipc_addid(&sem_ids(ns
), &sma
->sem_perm
, ns
->sc_semmni
);
269 security_sem_free(sma
);
273 ns
->used_sems
+= nsems
;
275 sma
->sem_base
= (struct sem
*) &sma
[1];
276 INIT_LIST_HEAD(&sma
->sem_pending
);
277 INIT_LIST_HEAD(&sma
->list_id
);
278 sma
->sem_nsems
= nsems
;
279 sma
->sem_ctime
= get_seconds();
282 return sma
->sem_perm
.id
;
287 * Called with sem_ids.rw_mutex and ipcp locked.
289 static inline int sem_security(struct kern_ipc_perm
*ipcp
, int semflg
)
291 struct sem_array
*sma
;
293 sma
= container_of(ipcp
, struct sem_array
, sem_perm
);
294 return security_sem_associate(sma
, semflg
);
298 * Called with sem_ids.rw_mutex and ipcp locked.
300 static inline int sem_more_checks(struct kern_ipc_perm
*ipcp
,
301 struct ipc_params
*params
)
303 struct sem_array
*sma
;
305 sma
= container_of(ipcp
, struct sem_array
, sem_perm
);
306 if (params
->u
.nsems
> sma
->sem_nsems
)
312 SYSCALL_DEFINE3(semget
, key_t
, key
, int, nsems
, int, semflg
)
314 struct ipc_namespace
*ns
;
315 struct ipc_ops sem_ops
;
316 struct ipc_params sem_params
;
318 ns
= current
->nsproxy
->ipc_ns
;
320 if (nsems
< 0 || nsems
> ns
->sc_semmsl
)
323 sem_ops
.getnew
= newary
;
324 sem_ops
.associate
= sem_security
;
325 sem_ops
.more_checks
= sem_more_checks
;
327 sem_params
.key
= key
;
328 sem_params
.flg
= semflg
;
329 sem_params
.u
.nsems
= nsems
;
331 return ipcget(ns
, &sem_ids(ns
), &sem_ops
, &sem_params
);
335 * Determine whether a sequence of semaphore operations would succeed
336 * all at once. Return 0 if yes, 1 if need to sleep, else return error code.
339 static int try_atomic_semop (struct sem_array
* sma
, struct sembuf
* sops
,
340 int nsops
, struct sem_undo
*un
, int pid
)
346 for (sop
= sops
; sop
< sops
+ nsops
; sop
++) {
347 curr
= sma
->sem_base
+ sop
->sem_num
;
348 sem_op
= sop
->sem_op
;
349 result
= curr
->semval
;
351 if (!sem_op
&& result
)
359 if (sop
->sem_flg
& SEM_UNDO
) {
360 int undo
= un
->semadj
[sop
->sem_num
] - sem_op
;
362 * Exceeding the undo range is an error.
364 if (undo
< (-SEMAEM
- 1) || undo
> SEMAEM
)
367 curr
->semval
= result
;
371 while (sop
>= sops
) {
372 sma
->sem_base
[sop
->sem_num
].sempid
= pid
;
373 if (sop
->sem_flg
& SEM_UNDO
)
374 un
->semadj
[sop
->sem_num
] -= sop
->sem_op
;
378 sma
->sem_otime
= get_seconds();
386 if (sop
->sem_flg
& IPC_NOWAIT
)
393 while (sop
>= sops
) {
394 sma
->sem_base
[sop
->sem_num
].semval
-= sop
->sem_op
;
402 * Wake up a process waiting on the sem queue with a given error.
403 * The queue is invalid (may not be accessed) after the function returns.
405 static void wake_up_sem_queue(struct sem_queue
*q
, int error
)
408 * Hold preempt off so that we don't get preempted and have the
409 * wakee busy-wait until we're scheduled back on. We're holding
410 * locks here so it may not strictly be needed, however if the
411 * locks become preemptible then this prevents such a problem.
414 q
->status
= IN_WAKEUP
;
415 wake_up_process(q
->sleeper
);
416 /* hands-off: q can disappear immediately after writing q->status. */
422 /* Go through the pending queue for the indicated semaphore
423 * looking for tasks that can be completed.
425 static void update_queue (struct sem_array
* sma
)
427 struct sem_queue
*q
, *tq
;
430 list_for_each_entry_safe(q
, tq
, &sma
->sem_pending
, list
) {
434 error
= try_atomic_semop(sma
, q
->sops
, q
->nsops
,
437 /* Does q->sleeper still need to sleep? */
444 * The next operation that must be checked depends on the type
445 * of the completed operation:
446 * - if the operation modified the array, then restart from the
447 * head of the queue and check for threads that might be
448 * waiting for the new semaphore values.
449 * - if the operation didn't modify the array, then just
453 wake_up_sem_queue(q
, error
);
459 /* The following counts are associated to each semaphore:
460 * semncnt number of tasks waiting on semval being nonzero
461 * semzcnt number of tasks waiting on semval being zero
462 * This model assumes that a task waits on exactly one semaphore.
463 * Since semaphore operations are to be performed atomically, tasks actually
464 * wait on a whole sequence of semaphores simultaneously.
465 * The counts we return here are a rough approximation, but still
466 * warrant that semncnt+semzcnt>0 if the task is on the pending queue.
468 static int count_semncnt (struct sem_array
* sma
, ushort semnum
)
471 struct sem_queue
* q
;
474 list_for_each_entry(q
, &sma
->sem_pending
, list
) {
475 struct sembuf
* sops
= q
->sops
;
476 int nsops
= q
->nsops
;
478 for (i
= 0; i
< nsops
; i
++)
479 if (sops
[i
].sem_num
== semnum
480 && (sops
[i
].sem_op
< 0)
481 && !(sops
[i
].sem_flg
& IPC_NOWAIT
))
487 static int count_semzcnt (struct sem_array
* sma
, ushort semnum
)
490 struct sem_queue
* q
;
493 list_for_each_entry(q
, &sma
->sem_pending
, list
) {
494 struct sembuf
* sops
= q
->sops
;
495 int nsops
= q
->nsops
;
497 for (i
= 0; i
< nsops
; i
++)
498 if (sops
[i
].sem_num
== semnum
499 && (sops
[i
].sem_op
== 0)
500 && !(sops
[i
].sem_flg
& IPC_NOWAIT
))
506 static void free_un(struct rcu_head
*head
)
508 struct sem_undo
*un
= container_of(head
, struct sem_undo
, rcu
);
512 /* Free a semaphore set. freeary() is called with sem_ids.rw_mutex locked
513 * as a writer and the spinlock for this semaphore set hold. sem_ids.rw_mutex
514 * remains locked on exit.
516 static void freeary(struct ipc_namespace
*ns
, struct kern_ipc_perm
*ipcp
)
518 struct sem_undo
*un
, *tu
;
519 struct sem_queue
*q
, *tq
;
520 struct sem_array
*sma
= container_of(ipcp
, struct sem_array
, sem_perm
);
522 /* Free the existing undo structures for this semaphore set. */
523 assert_spin_locked(&sma
->sem_perm
.lock
);
524 list_for_each_entry_safe(un
, tu
, &sma
->list_id
, list_id
) {
525 list_del(&un
->list_id
);
526 spin_lock(&un
->ulp
->lock
);
528 list_del_rcu(&un
->list_proc
);
529 spin_unlock(&un
->ulp
->lock
);
530 call_rcu(&un
->rcu
, free_un
);
533 /* Wake up all pending processes and let them fail with EIDRM. */
534 list_for_each_entry_safe(q
, tq
, &sma
->sem_pending
, list
) {
537 wake_up_sem_queue(q
, -EIDRM
);
540 /* Remove the semaphore set from the IDR */
544 ns
->used_sems
-= sma
->sem_nsems
;
545 security_sem_free(sma
);
549 static unsigned long copy_semid_to_user(void __user
*buf
, struct semid64_ds
*in
, int version
)
553 return copy_to_user(buf
, in
, sizeof(*in
));
558 ipc64_perm_to_ipc_perm(&in
->sem_perm
, &out
.sem_perm
);
560 out
.sem_otime
= in
->sem_otime
;
561 out
.sem_ctime
= in
->sem_ctime
;
562 out
.sem_nsems
= in
->sem_nsems
;
564 return copy_to_user(buf
, &out
, sizeof(out
));
571 static int semctl_nolock(struct ipc_namespace
*ns
, int semid
,
572 int cmd
, int version
, union semun arg
)
575 struct sem_array
*sma
;
581 struct seminfo seminfo
;
584 err
= security_sem_semctl(NULL
, cmd
);
588 memset(&seminfo
,0,sizeof(seminfo
));
589 seminfo
.semmni
= ns
->sc_semmni
;
590 seminfo
.semmns
= ns
->sc_semmns
;
591 seminfo
.semmsl
= ns
->sc_semmsl
;
592 seminfo
.semopm
= ns
->sc_semopm
;
593 seminfo
.semvmx
= SEMVMX
;
594 seminfo
.semmnu
= SEMMNU
;
595 seminfo
.semmap
= SEMMAP
;
596 seminfo
.semume
= SEMUME
;
597 down_read(&sem_ids(ns
).rw_mutex
);
598 if (cmd
== SEM_INFO
) {
599 seminfo
.semusz
= sem_ids(ns
).in_use
;
600 seminfo
.semaem
= ns
->used_sems
;
602 seminfo
.semusz
= SEMUSZ
;
603 seminfo
.semaem
= SEMAEM
;
605 max_id
= ipc_get_maxid(&sem_ids(ns
));
606 up_read(&sem_ids(ns
).rw_mutex
);
607 if (copy_to_user (arg
.__buf
, &seminfo
, sizeof(struct seminfo
)))
609 return (max_id
< 0) ? 0: max_id
;
614 struct semid64_ds tbuf
;
617 if (cmd
== SEM_STAT
) {
618 sma
= sem_lock(ns
, semid
);
621 id
= sma
->sem_perm
.id
;
623 sma
= sem_lock_check(ns
, semid
);
630 if (ipcperms (&sma
->sem_perm
, S_IRUGO
))
633 err
= security_sem_semctl(sma
, cmd
);
637 memset(&tbuf
, 0, sizeof(tbuf
));
639 kernel_to_ipc64_perm(&sma
->sem_perm
, &tbuf
.sem_perm
);
640 tbuf
.sem_otime
= sma
->sem_otime
;
641 tbuf
.sem_ctime
= sma
->sem_ctime
;
642 tbuf
.sem_nsems
= sma
->sem_nsems
;
644 if (copy_semid_to_user (arg
.buf
, &tbuf
, version
))
657 static int semctl_main(struct ipc_namespace
*ns
, int semid
, int semnum
,
658 int cmd
, int version
, union semun arg
)
660 struct sem_array
*sma
;
663 ushort fast_sem_io
[SEMMSL_FAST
];
664 ushort
* sem_io
= fast_sem_io
;
667 sma
= sem_lock_check(ns
, semid
);
671 nsems
= sma
->sem_nsems
;
674 if (ipcperms (&sma
->sem_perm
, (cmd
==SETVAL
||cmd
==SETALL
)?S_IWUGO
:S_IRUGO
))
677 err
= security_sem_semctl(sma
, cmd
);
685 ushort __user
*array
= arg
.array
;
688 if(nsems
> SEMMSL_FAST
) {
689 sem_getref_and_unlock(sma
);
691 sem_io
= ipc_alloc(sizeof(ushort
)*nsems
);
697 sem_lock_and_putref(sma
);
698 if (sma
->sem_perm
.deleted
) {
705 for (i
= 0; i
< sma
->sem_nsems
; i
++)
706 sem_io
[i
] = sma
->sem_base
[i
].semval
;
709 if(copy_to_user(array
, sem_io
, nsems
*sizeof(ushort
)))
718 sem_getref_and_unlock(sma
);
720 if(nsems
> SEMMSL_FAST
) {
721 sem_io
= ipc_alloc(sizeof(ushort
)*nsems
);
728 if (copy_from_user (sem_io
, arg
.array
, nsems
*sizeof(ushort
))) {
734 for (i
= 0; i
< nsems
; i
++) {
735 if (sem_io
[i
] > SEMVMX
) {
741 sem_lock_and_putref(sma
);
742 if (sma
->sem_perm
.deleted
) {
748 for (i
= 0; i
< nsems
; i
++)
749 sma
->sem_base
[i
].semval
= sem_io
[i
];
751 assert_spin_locked(&sma
->sem_perm
.lock
);
752 list_for_each_entry(un
, &sma
->list_id
, list_id
) {
753 for (i
= 0; i
< nsems
; i
++)
756 sma
->sem_ctime
= get_seconds();
757 /* maybe some queued-up processes were waiting for this */
762 /* GETVAL, GETPID, GETNCTN, GETZCNT, SETVAL: fall-through */
765 if(semnum
< 0 || semnum
>= nsems
)
768 curr
= &sma
->sem_base
[semnum
];
778 err
= count_semncnt(sma
,semnum
);
781 err
= count_semzcnt(sma
,semnum
);
789 if (val
> SEMVMX
|| val
< 0)
792 assert_spin_locked(&sma
->sem_perm
.lock
);
793 list_for_each_entry(un
, &sma
->list_id
, list_id
)
794 un
->semadj
[semnum
] = 0;
797 curr
->sempid
= task_tgid_vnr(current
);
798 sma
->sem_ctime
= get_seconds();
799 /* maybe some queued-up processes were waiting for this */
808 if(sem_io
!= fast_sem_io
)
809 ipc_free(sem_io
, sizeof(ushort
)*nsems
);
813 static inline unsigned long
814 copy_semid_from_user(struct semid64_ds
*out
, void __user
*buf
, int version
)
818 if (copy_from_user(out
, buf
, sizeof(*out
)))
823 struct semid_ds tbuf_old
;
825 if(copy_from_user(&tbuf_old
, buf
, sizeof(tbuf_old
)))
828 out
->sem_perm
.uid
= tbuf_old
.sem_perm
.uid
;
829 out
->sem_perm
.gid
= tbuf_old
.sem_perm
.gid
;
830 out
->sem_perm
.mode
= tbuf_old
.sem_perm
.mode
;
840 * This function handles some semctl commands which require the rw_mutex
841 * to be held in write mode.
842 * NOTE: no locks must be held, the rw_mutex is taken inside this function.
844 static int semctl_down(struct ipc_namespace
*ns
, int semid
,
845 int cmd
, int version
, union semun arg
)
847 struct sem_array
*sma
;
849 struct semid64_ds semid64
;
850 struct kern_ipc_perm
*ipcp
;
853 if (copy_semid_from_user(&semid64
, arg
.buf
, version
))
857 ipcp
= ipcctl_pre_down(&sem_ids(ns
), semid
, cmd
, &semid64
.sem_perm
, 0);
859 return PTR_ERR(ipcp
);
861 sma
= container_of(ipcp
, struct sem_array
, sem_perm
);
863 err
= security_sem_semctl(sma
, cmd
);
872 ipc_update_perm(&semid64
.sem_perm
, ipcp
);
873 sma
->sem_ctime
= get_seconds();
882 up_write(&sem_ids(ns
).rw_mutex
);
886 SYSCALL_DEFINE(semctl
)(int semid
, int semnum
, int cmd
, union semun arg
)
890 struct ipc_namespace
*ns
;
895 version
= ipc_parse_version(&cmd
);
896 ns
= current
->nsproxy
->ipc_ns
;
903 err
= semctl_nolock(ns
, semid
, cmd
, version
, arg
);
912 err
= semctl_main(ns
,semid
,semnum
,cmd
,version
,arg
);
916 err
= semctl_down(ns
, semid
, cmd
, version
, arg
);
922 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
923 asmlinkage
long SyS_semctl(int semid
, int semnum
, int cmd
, union semun arg
)
925 return SYSC_semctl((int) semid
, (int) semnum
, (int) cmd
, arg
);
927 SYSCALL_ALIAS(sys_semctl
, SyS_semctl
);
930 /* If the task doesn't already have a undo_list, then allocate one
931 * here. We guarantee there is only one thread using this undo list,
932 * and current is THE ONE
934 * If this allocation and assignment succeeds, but later
935 * portions of this code fail, there is no need to free the sem_undo_list.
936 * Just let it stay associated with the task, and it'll be freed later
939 * This can block, so callers must hold no locks.
941 static inline int get_undo_list(struct sem_undo_list
**undo_listp
)
943 struct sem_undo_list
*undo_list
;
945 undo_list
= current
->sysvsem
.undo_list
;
947 undo_list
= kzalloc(sizeof(*undo_list
), GFP_KERNEL
);
948 if (undo_list
== NULL
)
950 spin_lock_init(&undo_list
->lock
);
951 atomic_set(&undo_list
->refcnt
, 1);
952 INIT_LIST_HEAD(&undo_list
->list_proc
);
954 current
->sysvsem
.undo_list
= undo_list
;
956 *undo_listp
= undo_list
;
960 static struct sem_undo
*__lookup_undo(struct sem_undo_list
*ulp
, int semid
)
964 list_for_each_entry_rcu(un
, &ulp
->list_proc
, list_proc
) {
965 if (un
->semid
== semid
)
971 static struct sem_undo
*lookup_undo(struct sem_undo_list
*ulp
, int semid
)
975 assert_spin_locked(&ulp
->lock
);
977 un
= __lookup_undo(ulp
, semid
);
979 list_del_rcu(&un
->list_proc
);
980 list_add_rcu(&un
->list_proc
, &ulp
->list_proc
);
986 * find_alloc_undo - Lookup (and if not present create) undo array
988 * @semid: semaphore array id
990 * The function looks up (and if not present creates) the undo structure.
991 * The size of the undo structure depends on the size of the semaphore
992 * array, thus the alloc path is not that straightforward.
993 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
994 * performs a rcu_read_lock().
996 static struct sem_undo
*find_alloc_undo(struct ipc_namespace
*ns
, int semid
)
998 struct sem_array
*sma
;
999 struct sem_undo_list
*ulp
;
1000 struct sem_undo
*un
, *new;
1004 error
= get_undo_list(&ulp
);
1006 return ERR_PTR(error
);
1009 spin_lock(&ulp
->lock
);
1010 un
= lookup_undo(ulp
, semid
);
1011 spin_unlock(&ulp
->lock
);
1012 if (likely(un
!=NULL
))
1016 /* no undo structure around - allocate one. */
1017 /* step 1: figure out the size of the semaphore array */
1018 sma
= sem_lock_check(ns
, semid
);
1020 return ERR_PTR(PTR_ERR(sma
));
1022 nsems
= sma
->sem_nsems
;
1023 sem_getref_and_unlock(sma
);
1025 /* step 2: allocate new undo structure */
1026 new = kzalloc(sizeof(struct sem_undo
) + sizeof(short)*nsems
, GFP_KERNEL
);
1029 return ERR_PTR(-ENOMEM
);
1032 /* step 3: Acquire the lock on semaphore array */
1033 sem_lock_and_putref(sma
);
1034 if (sma
->sem_perm
.deleted
) {
1037 un
= ERR_PTR(-EIDRM
);
1040 spin_lock(&ulp
->lock
);
1043 * step 4: check for races: did someone else allocate the undo struct?
1045 un
= lookup_undo(ulp
, semid
);
1050 /* step 5: initialize & link new undo structure */
1051 new->semadj
= (short *) &new[1];
1054 assert_spin_locked(&ulp
->lock
);
1055 list_add_rcu(&new->list_proc
, &ulp
->list_proc
);
1056 assert_spin_locked(&sma
->sem_perm
.lock
);
1057 list_add(&new->list_id
, &sma
->list_id
);
1061 spin_unlock(&ulp
->lock
);
1068 SYSCALL_DEFINE4(semtimedop
, int, semid
, struct sembuf __user
*, tsops
,
1069 unsigned, nsops
, const struct timespec __user
*, timeout
)
1071 int error
= -EINVAL
;
1072 struct sem_array
*sma
;
1073 struct sembuf fast_sops
[SEMOPM_FAST
];
1074 struct sembuf
* sops
= fast_sops
, *sop
;
1075 struct sem_undo
*un
;
1076 int undos
= 0, alter
= 0, max
;
1077 struct sem_queue queue
;
1078 unsigned long jiffies_left
= 0;
1079 struct ipc_namespace
*ns
;
1081 ns
= current
->nsproxy
->ipc_ns
;
1083 if (nsops
< 1 || semid
< 0)
1085 if (nsops
> ns
->sc_semopm
)
1087 if(nsops
> SEMOPM_FAST
) {
1088 sops
= kmalloc(sizeof(*sops
)*nsops
,GFP_KERNEL
);
1092 if (copy_from_user (sops
, tsops
, nsops
* sizeof(*tsops
))) {
1097 struct timespec _timeout
;
1098 if (copy_from_user(&_timeout
, timeout
, sizeof(*timeout
))) {
1102 if (_timeout
.tv_sec
< 0 || _timeout
.tv_nsec
< 0 ||
1103 _timeout
.tv_nsec
>= 1000000000L) {
1107 jiffies_left
= timespec_to_jiffies(&_timeout
);
1110 for (sop
= sops
; sop
< sops
+ nsops
; sop
++) {
1111 if (sop
->sem_num
>= max
)
1113 if (sop
->sem_flg
& SEM_UNDO
)
1115 if (sop
->sem_op
!= 0)
1120 un
= find_alloc_undo(ns
, semid
);
1122 error
= PTR_ERR(un
);
1128 sma
= sem_lock_check(ns
, semid
);
1132 error
= PTR_ERR(sma
);
1137 * semid identifiers are not unique - find_alloc_undo may have
1138 * allocated an undo structure, it was invalidated by an RMID
1139 * and now a new array with received the same id. Check and fail.
1140 * This case can be detected checking un->semid. The existance of
1141 * "un" itself is guaranteed by rcu.
1145 if (un
->semid
== -1) {
1147 goto out_unlock_free
;
1150 * rcu lock can be released, "un" cannot disappear:
1151 * - sem_lock is acquired, thus IPC_RMID is
1153 * - exit_sem is impossible, it always operates on
1154 * current (or a dead task).
1162 if (max
>= sma
->sem_nsems
)
1163 goto out_unlock_free
;
1166 if (ipcperms(&sma
->sem_perm
, alter
? S_IWUGO
: S_IRUGO
))
1167 goto out_unlock_free
;
1169 error
= security_sem_semop(sma
, sops
, nsops
, alter
);
1171 goto out_unlock_free
;
1173 error
= try_atomic_semop (sma
, sops
, nsops
, un
, task_tgid_vnr(current
));
1175 if (alter
&& error
== 0)
1177 goto out_unlock_free
;
1180 /* We need to sleep on this operation, so we put the current
1181 * task into the pending queue and go to sleep.
1185 queue
.nsops
= nsops
;
1187 queue
.pid
= task_tgid_vnr(current
);
1188 queue
.alter
= alter
;
1190 list_add_tail(&queue
.list
, &sma
->sem_pending
);
1192 list_add(&queue
.list
, &sma
->sem_pending
);
1194 queue
.status
= -EINTR
;
1195 queue
.sleeper
= current
;
1196 current
->state
= TASK_INTERRUPTIBLE
;
1200 jiffies_left
= schedule_timeout(jiffies_left
);
1204 error
= queue
.status
;
1205 while(unlikely(error
== IN_WAKEUP
)) {
1207 error
= queue
.status
;
1210 if (error
!= -EINTR
) {
1211 /* fast path: update_queue already obtained all requested
1216 sma
= sem_lock(ns
, semid
);
1223 * If queue.status != -EINTR we are woken up by another process
1225 error
= queue
.status
;
1226 if (error
!= -EINTR
) {
1227 goto out_unlock_free
;
1231 * If an interrupt occurred we have to clean up the queue
1233 if (timeout
&& jiffies_left
== 0)
1235 list_del(&queue
.list
);
1240 if(sops
!= fast_sops
)
1245 SYSCALL_DEFINE3(semop
, int, semid
, struct sembuf __user
*, tsops
,
1248 return sys_semtimedop(semid
, tsops
, nsops
, NULL
);
1251 /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
1252 * parent and child tasks.
1255 int copy_semundo(unsigned long clone_flags
, struct task_struct
*tsk
)
1257 struct sem_undo_list
*undo_list
;
1260 if (clone_flags
& CLONE_SYSVSEM
) {
1261 error
= get_undo_list(&undo_list
);
1264 atomic_inc(&undo_list
->refcnt
);
1265 tsk
->sysvsem
.undo_list
= undo_list
;
1267 tsk
->sysvsem
.undo_list
= NULL
;
1273 * add semadj values to semaphores, free undo structures.
1274 * undo structures are not freed when semaphore arrays are destroyed
1275 * so some of them may be out of date.
1276 * IMPLEMENTATION NOTE: There is some confusion over whether the
1277 * set of adjustments that needs to be done should be done in an atomic
1278 * manner or not. That is, if we are attempting to decrement the semval
1279 * should we queue up and wait until we can do so legally?
1280 * The original implementation attempted to do this (queue and wait).
1281 * The current implementation does not do so. The POSIX standard
1282 * and SVID should be consulted to determine what behavior is mandated.
1284 void exit_sem(struct task_struct
*tsk
)
1286 struct sem_undo_list
*ulp
;
1288 ulp
= tsk
->sysvsem
.undo_list
;
1291 tsk
->sysvsem
.undo_list
= NULL
;
1293 if (!atomic_dec_and_test(&ulp
->refcnt
))
1297 struct sem_array
*sma
;
1298 struct sem_undo
*un
;
1303 un
= list_entry_rcu(ulp
->list_proc
.next
,
1304 struct sem_undo
, list_proc
);
1305 if (&un
->list_proc
== &ulp
->list_proc
)
1314 sma
= sem_lock_check(tsk
->nsproxy
->ipc_ns
, un
->semid
);
1316 /* exit_sem raced with IPC_RMID, nothing to do */
1320 un
= __lookup_undo(ulp
, semid
);
1322 /* exit_sem raced with IPC_RMID+semget() that created
1323 * exactly the same semid. Nothing to do.
1329 /* remove un from the linked lists */
1330 assert_spin_locked(&sma
->sem_perm
.lock
);
1331 list_del(&un
->list_id
);
1333 spin_lock(&ulp
->lock
);
1334 list_del_rcu(&un
->list_proc
);
1335 spin_unlock(&ulp
->lock
);
1337 /* perform adjustments registered in un */
1338 for (i
= 0; i
< sma
->sem_nsems
; i
++) {
1339 struct sem
* semaphore
= &sma
->sem_base
[i
];
1340 if (un
->semadj
[i
]) {
1341 semaphore
->semval
+= un
->semadj
[i
];
1343 * Range checks of the new semaphore value,
1344 * not defined by sus:
1345 * - Some unices ignore the undo entirely
1346 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
1347 * - some cap the value (e.g. FreeBSD caps
1348 * at 0, but doesn't enforce SEMVMX)
1350 * Linux caps the semaphore value, both at 0
1353 * Manfred <manfred@colorfullife.com>
1355 if (semaphore
->semval
< 0)
1356 semaphore
->semval
= 0;
1357 if (semaphore
->semval
> SEMVMX
)
1358 semaphore
->semval
= SEMVMX
;
1359 semaphore
->sempid
= task_tgid_vnr(current
);
1362 sma
->sem_otime
= get_seconds();
1363 /* maybe some queued-up processes were waiting for this */
1367 call_rcu(&un
->rcu
, free_un
);
1372 #ifdef CONFIG_PROC_FS
1373 static int sysvipc_sem_proc_show(struct seq_file
*s
, void *it
)
1375 struct sem_array
*sma
= it
;
1377 return seq_printf(s
,
1378 "%10d %10d %4o %10lu %5u %5u %5u %5u %10lu %10lu\n",
This page took 0.074888 seconds and 5 git commands to generate.