]>
git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - ipc/sem.c
3 * Copyright (C) 1992 Krishna Balasubramanian
4 * Copyright (C) 1995 Eric Schenk, Bruno Haible
6 * IMPLEMENTATION NOTES ON CODE REWRITE (Eric Schenk, January 1995):
7 * This code underwent a massive rewrite in order to solve some problems
8 * with the original code. In particular the original code failed to
9 * wake up processes that were waiting for semval to go to 0 if the
10 * value went to 0 and was then incremented rapidly enough. In solving
11 * this problem I have also modified the implementation so that it
12 * processes pending operations in a FIFO manner, thus give a guarantee
13 * that processes waiting for a lock on the semaphore won't starve
14 * unless another locking process fails to unlock.
15 * In addition the following two changes in behavior have been introduced:
16 * - The original implementation of semop returned the value
17 * last semaphore element examined on success. This does not
18 * match the manual page specifications, and effectively
19 * allows the user to read the semaphore even if they do not
20 * have read permissions. The implementation now returns 0
21 * on success as stated in the manual page.
22 * - There is some confusion over whether the set of undo adjustments
23 * to be performed at exit should be done in an atomic manner.
24 * That is, if we are attempting to decrement the semval should we queue
25 * up and wait until we can do so legally?
26 * The original implementation attempted to do this.
27 * The current implementation does not do so. This is because I don't
28 * think it is the right thing (TM) to do, and because I couldn't
29 * see a clean way to get the old behavior with the new design.
30 * The POSIX standard and SVID should be consulted to determine
31 * what behavior is mandated.
33 * Further notes on refinement (Christoph Rohland, December 1998):
34 * - The POSIX standard says, that the undo adjustments simply should
35 * redo. So the current implementation is o.K.
36 * - The previous code had two flaws:
37 * 1) It actively gave the semaphore to the next waiting process
38 * sleeping on the semaphore. Since this process did not have the
39 * cpu this led to many unnecessary context switches and bad
40 * performance. Now we only check which process should be able to
41 * get the semaphore and if this process wants to reduce some
42 * semaphore value we simply wake it up without doing the
43 * operation. So it has to try to get it later. Thus e.g. the
44 * running process may reacquire the semaphore during the current
45 * time slice. If it only waits for zero or increases the semaphore,
46 * we do the operation in advance and wake it up.
47 * 2) It did not wake up all zero waiting processes. We try to do
48 * better but only get the semops right which only wait for zero or
49 * increase. If there are decrement operations in the operations
50 * array we do the same as before.
52 * With the incarnation of O(1) scheduler, it becomes unnecessary to perform
53 * check/retry algorithm for waking up blocked processes as the new scheduler
54 * is better at handling thread switch than the old one.
56 * /proc/sysvipc/sem support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
58 * SMP-threaded, sysctl's added
59 * (c) 1999 Manfred Spraul <manfred@colorfullife.com>
60 * Enforced range limit on SEM_UNDO
61 * (c) 2001 Red Hat Inc <alan@redhat.com>
63 * (c) 2003 Manfred Spraul <manfred@colorfullife.com>
65 * support for audit of ipc object properties and permission changes
66 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
70 * Pavel Emelianov <xemul@openvz.org>
73 #include <linux/slab.h>
74 #include <linux/spinlock.h>
75 #include <linux/init.h>
76 #include <linux/proc_fs.h>
77 #include <linux/time.h>
78 #include <linux/security.h>
79 #include <linux/syscalls.h>
80 #include <linux/audit.h>
81 #include <linux/capability.h>
82 #include <linux/seq_file.h>
83 #include <linux/rwsem.h>
84 #include <linux/nsproxy.h>
85 #include <linux/ipc_namespace.h>
87 #include <asm/uaccess.h>
90 #define sem_ids(ns) ((ns)->ids[IPC_SEM_IDS])
92 #define sem_unlock(sma) ipc_unlock(&(sma)->sem_perm)
93 #define sem_checkid(sma, semid) ipc_checkid(&sma->sem_perm, semid)
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
);
135 void __init
sem_init (void)
137 sem_init_ns(&init_ipc_ns
);
138 ipc_init_proc_interface("sysvipc/sem",
139 " key semid perms nsems uid gid cuid cgid otime ctime\n",
140 IPC_SEM_IDS
, sysvipc_sem_proc_show
);
144 * sem_lock_(check_) routines are called in the paths where the rw_mutex
147 static inline struct sem_array
*sem_lock(struct ipc_namespace
*ns
, int id
)
149 struct kern_ipc_perm
*ipcp
= ipc_lock(&sem_ids(ns
), id
);
152 return (struct sem_array
*)ipcp
;
154 return container_of(ipcp
, struct sem_array
, sem_perm
);
157 static inline struct sem_array
*sem_lock_check(struct ipc_namespace
*ns
,
160 struct kern_ipc_perm
*ipcp
= ipc_lock_check(&sem_ids(ns
), id
);
163 return (struct sem_array
*)ipcp
;
165 return container_of(ipcp
, struct sem_array
, sem_perm
);
168 static inline void sem_lock_and_putref(struct sem_array
*sma
)
170 ipc_lock_by_ptr(&sma
->sem_perm
);
174 static inline void sem_getref_and_unlock(struct sem_array
*sma
)
177 ipc_unlock(&(sma
)->sem_perm
);
180 static inline void sem_putref(struct sem_array
*sma
)
182 ipc_lock_by_ptr(&sma
->sem_perm
);
184 ipc_unlock(&(sma
)->sem_perm
);
187 static inline void sem_rmid(struct ipc_namespace
*ns
, struct sem_array
*s
)
189 ipc_rmid(&sem_ids(ns
), &s
->sem_perm
);
193 * Lockless wakeup algorithm:
194 * Without the check/retry algorithm a lockless wakeup is possible:
195 * - queue.status is initialized to -EINTR before blocking.
196 * - wakeup is performed by
197 * * unlinking the queue entry from sma->sem_pending
198 * * setting queue.status to IN_WAKEUP
199 * This is the notification for the blocked thread that a
200 * result value is imminent.
201 * * call wake_up_process
202 * * set queue.status to the final value.
203 * - the previously blocked thread checks queue.status:
204 * * if it's IN_WAKEUP, then it must wait until the value changes
205 * * if it's not -EINTR, then the operation was completed by
206 * update_queue. semtimedop can return queue.status without
207 * performing any operation on the sem array.
208 * * otherwise it must acquire the spinlock and check what's up.
210 * The two-stage algorithm is necessary to protect against the following
212 * - if queue.status is set after wake_up_process, then the woken up idle
213 * thread could race forward and try (and fail) to acquire sma->lock
214 * before update_queue had a chance to set queue.status
215 * - if queue.status is written before wake_up_process and if the
216 * blocked process is woken up by a signal between writing
217 * queue.status and the wake_up_process, then the woken up
218 * process could return from semtimedop and die by calling
219 * sys_exit before wake_up_process is called. Then wake_up_process
220 * will oops, because the task structure is already invalid.
221 * (yes, this happened on s390 with sysv msg).
227 * newary - Create a new semaphore set
229 * @params: ptr to the structure that contains key, semflg and nsems
231 * Called with sem_ids.rw_mutex held (as a writer)
234 static int newary(struct ipc_namespace
*ns
, struct ipc_params
*params
)
238 struct sem_array
*sma
;
240 key_t key
= params
->key
;
241 int nsems
= params
->u
.nsems
;
242 int semflg
= params
->flg
;
246 if (ns
->used_sems
+ nsems
> ns
->sc_semmns
)
249 size
= sizeof (*sma
) + nsems
* sizeof (struct sem
);
250 sma
= ipc_rcu_alloc(size
);
254 memset (sma
, 0, size
);
256 sma
->sem_perm
.mode
= (semflg
& S_IRWXUGO
);
257 sma
->sem_perm
.key
= key
;
259 sma
->sem_perm
.security
= NULL
;
260 retval
= security_sem_alloc(sma
);
266 id
= ipc_addid(&sem_ids(ns
), &sma
->sem_perm
, ns
->sc_semmni
);
268 security_sem_free(sma
);
272 ns
->used_sems
+= nsems
;
274 sma
->sem_base
= (struct sem
*) &sma
[1];
275 /* sma->sem_pending = NULL; */
276 sma
->sem_pending_last
= &sma
->sem_pending
;
277 /* sma->undo = NULL; */
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 asmlinkage
long sys_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
);
334 /* Manage the doubly linked list sma->sem_pending as a FIFO:
335 * insert new queue elements at the tail sma->sem_pending_last.
337 static inline void append_to_queue (struct sem_array
* sma
,
338 struct sem_queue
* q
)
340 *(q
->prev
= sma
->sem_pending_last
) = q
;
341 *(sma
->sem_pending_last
= &q
->next
) = NULL
;
344 static inline void prepend_to_queue (struct sem_array
* sma
,
345 struct sem_queue
* q
)
347 q
->next
= sma
->sem_pending
;
348 *(q
->prev
= &sma
->sem_pending
) = q
;
350 q
->next
->prev
= &q
->next
;
351 else /* sma->sem_pending_last == &sma->sem_pending */
352 sma
->sem_pending_last
= &q
->next
;
355 static inline void remove_from_queue (struct sem_array
* sma
,
356 struct sem_queue
* q
)
358 *(q
->prev
) = q
->next
;
360 q
->next
->prev
= q
->prev
;
361 else /* sma->sem_pending_last == &q->next */
362 sma
->sem_pending_last
= q
->prev
;
363 q
->prev
= NULL
; /* mark as removed */
367 * Determine whether a sequence of semaphore operations would succeed
368 * all at once. Return 0 if yes, 1 if need to sleep, else return error code.
371 static int try_atomic_semop (struct sem_array
* sma
, struct sembuf
* sops
,
372 int nsops
, struct sem_undo
*un
, int pid
)
378 for (sop
= sops
; sop
< sops
+ nsops
; sop
++) {
379 curr
= sma
->sem_base
+ sop
->sem_num
;
380 sem_op
= sop
->sem_op
;
381 result
= curr
->semval
;
383 if (!sem_op
&& result
)
391 if (sop
->sem_flg
& SEM_UNDO
) {
392 int undo
= un
->semadj
[sop
->sem_num
] - sem_op
;
394 * Exceeding the undo range is an error.
396 if (undo
< (-SEMAEM
- 1) || undo
> SEMAEM
)
399 curr
->semval
= result
;
403 while (sop
>= sops
) {
404 sma
->sem_base
[sop
->sem_num
].sempid
= pid
;
405 if (sop
->sem_flg
& SEM_UNDO
)
406 un
->semadj
[sop
->sem_num
] -= sop
->sem_op
;
410 sma
->sem_otime
= get_seconds();
418 if (sop
->sem_flg
& IPC_NOWAIT
)
425 while (sop
>= sops
) {
426 sma
->sem_base
[sop
->sem_num
].semval
-= sop
->sem_op
;
433 /* Go through the pending queue for the indicated semaphore
434 * looking for tasks that can be completed.
436 static void update_queue (struct sem_array
* sma
)
439 struct sem_queue
* q
;
441 q
= sma
->sem_pending
;
443 error
= try_atomic_semop(sma
, q
->sops
, q
->nsops
,
446 /* Does q->sleeper still need to sleep? */
449 remove_from_queue(sma
,q
);
450 q
->status
= IN_WAKEUP
;
452 * Continue scanning. The next operation
453 * that must be checked depends on the type of the
454 * completed operation:
455 * - if the operation modified the array, then
456 * restart from the head of the queue and
457 * check for threads that might be waiting
458 * for semaphore values to become 0.
459 * - if the operation didn't modify the array,
460 * then just continue.
463 n
= sma
->sem_pending
;
466 wake_up_process(q
->sleeper
);
467 /* hands-off: q will disappear immediately after
479 /* The following counts are associated to each semaphore:
480 * semncnt number of tasks waiting on semval being nonzero
481 * semzcnt number of tasks waiting on semval being zero
482 * This model assumes that a task waits on exactly one semaphore.
483 * Since semaphore operations are to be performed atomically, tasks actually
484 * wait on a whole sequence of semaphores simultaneously.
485 * The counts we return here are a rough approximation, but still
486 * warrant that semncnt+semzcnt>0 if the task is on the pending queue.
488 static int count_semncnt (struct sem_array
* sma
, ushort semnum
)
491 struct sem_queue
* q
;
494 for (q
= sma
->sem_pending
; q
; q
= q
->next
) {
495 struct sembuf
* sops
= q
->sops
;
496 int nsops
= q
->nsops
;
498 for (i
= 0; i
< nsops
; i
++)
499 if (sops
[i
].sem_num
== semnum
500 && (sops
[i
].sem_op
< 0)
501 && !(sops
[i
].sem_flg
& IPC_NOWAIT
))
506 static int count_semzcnt (struct sem_array
* sma
, ushort semnum
)
509 struct sem_queue
* q
;
512 for (q
= sma
->sem_pending
; q
; q
= q
->next
) {
513 struct sembuf
* sops
= q
->sops
;
514 int nsops
= q
->nsops
;
516 for (i
= 0; i
< nsops
; i
++)
517 if (sops
[i
].sem_num
== semnum
518 && (sops
[i
].sem_op
== 0)
519 && !(sops
[i
].sem_flg
& IPC_NOWAIT
))
525 /* Free a semaphore set. freeary() is called with sem_ids.rw_mutex locked
526 * as a writer and the spinlock for this semaphore set hold. sem_ids.rw_mutex
527 * remains locked on exit.
529 static void freeary(struct ipc_namespace
*ns
, struct kern_ipc_perm
*ipcp
)
533 struct sem_array
*sma
= container_of(ipcp
, struct sem_array
, sem_perm
);
535 /* Invalidate the existing undo structures for this semaphore set.
536 * (They will be freed without any further action in exit_sem()
537 * or during the next semop.)
539 for (un
= sma
->undo
; un
; un
= un
->id_next
)
542 /* Wake up all pending processes and let them fail with EIDRM. */
543 q
= sma
->sem_pending
;
546 /* lazy remove_from_queue: we are killing the whole queue */
549 q
->status
= IN_WAKEUP
;
550 wake_up_process(q
->sleeper
); /* doesn't sleep */
552 q
->status
= -EIDRM
; /* hands-off q */
556 /* Remove the semaphore set from the IDR */
560 ns
->used_sems
-= sma
->sem_nsems
;
561 security_sem_free(sma
);
565 static unsigned long copy_semid_to_user(void __user
*buf
, struct semid64_ds
*in
, int version
)
569 return copy_to_user(buf
, in
, sizeof(*in
));
574 ipc64_perm_to_ipc_perm(&in
->sem_perm
, &out
.sem_perm
);
576 out
.sem_otime
= in
->sem_otime
;
577 out
.sem_ctime
= in
->sem_ctime
;
578 out
.sem_nsems
= in
->sem_nsems
;
580 return copy_to_user(buf
, &out
, sizeof(out
));
587 static int semctl_nolock(struct ipc_namespace
*ns
, int semid
,
588 int cmd
, int version
, union semun arg
)
591 struct sem_array
*sma
;
597 struct seminfo seminfo
;
600 err
= security_sem_semctl(NULL
, cmd
);
604 memset(&seminfo
,0,sizeof(seminfo
));
605 seminfo
.semmni
= ns
->sc_semmni
;
606 seminfo
.semmns
= ns
->sc_semmns
;
607 seminfo
.semmsl
= ns
->sc_semmsl
;
608 seminfo
.semopm
= ns
->sc_semopm
;
609 seminfo
.semvmx
= SEMVMX
;
610 seminfo
.semmnu
= SEMMNU
;
611 seminfo
.semmap
= SEMMAP
;
612 seminfo
.semume
= SEMUME
;
613 down_read(&sem_ids(ns
).rw_mutex
);
614 if (cmd
== SEM_INFO
) {
615 seminfo
.semusz
= sem_ids(ns
).in_use
;
616 seminfo
.semaem
= ns
->used_sems
;
618 seminfo
.semusz
= SEMUSZ
;
619 seminfo
.semaem
= SEMAEM
;
621 max_id
= ipc_get_maxid(&sem_ids(ns
));
622 up_read(&sem_ids(ns
).rw_mutex
);
623 if (copy_to_user (arg
.__buf
, &seminfo
, sizeof(struct seminfo
)))
625 return (max_id
< 0) ? 0: max_id
;
630 struct semid64_ds tbuf
;
633 if (cmd
== SEM_STAT
) {
634 sma
= sem_lock(ns
, semid
);
637 id
= sma
->sem_perm
.id
;
639 sma
= sem_lock_check(ns
, semid
);
646 if (ipcperms (&sma
->sem_perm
, S_IRUGO
))
649 err
= security_sem_semctl(sma
, cmd
);
653 memset(&tbuf
, 0, sizeof(tbuf
));
655 kernel_to_ipc64_perm(&sma
->sem_perm
, &tbuf
.sem_perm
);
656 tbuf
.sem_otime
= sma
->sem_otime
;
657 tbuf
.sem_ctime
= sma
->sem_ctime
;
658 tbuf
.sem_nsems
= sma
->sem_nsems
;
660 if (copy_semid_to_user (arg
.buf
, &tbuf
, version
))
673 static int semctl_main(struct ipc_namespace
*ns
, int semid
, int semnum
,
674 int cmd
, int version
, union semun arg
)
676 struct sem_array
*sma
;
679 ushort fast_sem_io
[SEMMSL_FAST
];
680 ushort
* sem_io
= fast_sem_io
;
683 sma
= sem_lock_check(ns
, semid
);
687 nsems
= sma
->sem_nsems
;
690 if (ipcperms (&sma
->sem_perm
, (cmd
==SETVAL
||cmd
==SETALL
)?S_IWUGO
:S_IRUGO
))
693 err
= security_sem_semctl(sma
, cmd
);
701 ushort __user
*array
= arg
.array
;
704 if(nsems
> SEMMSL_FAST
) {
705 sem_getref_and_unlock(sma
);
707 sem_io
= ipc_alloc(sizeof(ushort
)*nsems
);
713 sem_lock_and_putref(sma
);
714 if (sma
->sem_perm
.deleted
) {
721 for (i
= 0; i
< sma
->sem_nsems
; i
++)
722 sem_io
[i
] = sma
->sem_base
[i
].semval
;
725 if(copy_to_user(array
, sem_io
, nsems
*sizeof(ushort
)))
734 sem_getref_and_unlock(sma
);
736 if(nsems
> SEMMSL_FAST
) {
737 sem_io
= ipc_alloc(sizeof(ushort
)*nsems
);
744 if (copy_from_user (sem_io
, arg
.array
, nsems
*sizeof(ushort
))) {
750 for (i
= 0; i
< nsems
; i
++) {
751 if (sem_io
[i
] > SEMVMX
) {
757 sem_lock_and_putref(sma
);
758 if (sma
->sem_perm
.deleted
) {
764 for (i
= 0; i
< nsems
; i
++)
765 sma
->sem_base
[i
].semval
= sem_io
[i
];
766 for (un
= sma
->undo
; un
; un
= un
->id_next
)
767 for (i
= 0; i
< nsems
; i
++)
769 sma
->sem_ctime
= get_seconds();
770 /* maybe some queued-up processes were waiting for this */
775 /* GETVAL, GETPID, GETNCTN, GETZCNT, SETVAL: fall-through */
778 if(semnum
< 0 || semnum
>= nsems
)
781 curr
= &sma
->sem_base
[semnum
];
791 err
= count_semncnt(sma
,semnum
);
794 err
= count_semzcnt(sma
,semnum
);
801 if (val
> SEMVMX
|| val
< 0)
804 for (un
= sma
->undo
; un
; un
= un
->id_next
)
805 un
->semadj
[semnum
] = 0;
807 curr
->sempid
= task_tgid_vnr(current
);
808 sma
->sem_ctime
= get_seconds();
809 /* maybe some queued-up processes were waiting for this */
818 if(sem_io
!= fast_sem_io
)
819 ipc_free(sem_io
, sizeof(ushort
)*nsems
);
823 static inline unsigned long
824 copy_semid_from_user(struct semid64_ds
*out
, void __user
*buf
, int version
)
828 if (copy_from_user(out
, buf
, sizeof(*out
)))
833 struct semid_ds tbuf_old
;
835 if(copy_from_user(&tbuf_old
, buf
, sizeof(tbuf_old
)))
838 out
->sem_perm
.uid
= tbuf_old
.sem_perm
.uid
;
839 out
->sem_perm
.gid
= tbuf_old
.sem_perm
.gid
;
840 out
->sem_perm
.mode
= tbuf_old
.sem_perm
.mode
;
850 * This function handles some semctl commands which require the rw_mutex
851 * to be held in write mode.
852 * NOTE: no locks must be held, the rw_mutex is taken inside this function.
854 static int semctl_down(struct ipc_namespace
*ns
, int semid
,
855 int cmd
, int version
, union semun arg
)
857 struct sem_array
*sma
;
859 struct semid64_ds semid64
;
860 struct kern_ipc_perm
*ipcp
;
863 if (copy_semid_from_user(&semid64
, arg
.buf
, version
))
867 ipcp
= ipcctl_pre_down(&sem_ids(ns
), semid
, cmd
, &semid64
.sem_perm
, 0);
869 return PTR_ERR(ipcp
);
871 sma
= container_of(ipcp
, struct sem_array
, sem_perm
);
873 err
= security_sem_semctl(sma
, cmd
);
882 ipc_update_perm(&semid64
.sem_perm
, ipcp
);
883 sma
->sem_ctime
= get_seconds();
892 up_write(&sem_ids(ns
).rw_mutex
);
896 asmlinkage
long sys_semctl (int semid
, int semnum
, int cmd
, union semun arg
)
900 struct ipc_namespace
*ns
;
905 version
= ipc_parse_version(&cmd
);
906 ns
= current
->nsproxy
->ipc_ns
;
913 err
= semctl_nolock(ns
, semid
, cmd
, version
, arg
);
922 err
= semctl_main(ns
,semid
,semnum
,cmd
,version
,arg
);
926 err
= semctl_down(ns
, semid
, cmd
, version
, arg
);
933 /* If the task doesn't already have a undo_list, then allocate one
934 * here. We guarantee there is only one thread using this undo list,
935 * and current is THE ONE
937 * If this allocation and assignment succeeds, but later
938 * portions of this code fail, there is no need to free the sem_undo_list.
939 * Just let it stay associated with the task, and it'll be freed later
942 * This can block, so callers must hold no locks.
944 static inline int get_undo_list(struct sem_undo_list
**undo_listp
)
946 struct sem_undo_list
*undo_list
;
948 undo_list
= current
->sysvsem
.undo_list
;
950 undo_list
= kzalloc(sizeof(*undo_list
), GFP_KERNEL
);
951 if (undo_list
== NULL
)
953 spin_lock_init(&undo_list
->lock
);
954 atomic_set(&undo_list
->refcnt
, 1);
955 current
->sysvsem
.undo_list
= undo_list
;
957 *undo_listp
= undo_list
;
961 static struct sem_undo
*lookup_undo(struct sem_undo_list
*ulp
, int semid
)
963 struct sem_undo
**last
, *un
;
965 last
= &ulp
->proc_list
;
981 static struct sem_undo
*find_undo(struct ipc_namespace
*ns
, int semid
)
983 struct sem_array
*sma
;
984 struct sem_undo_list
*ulp
;
985 struct sem_undo
*un
, *new;
989 error
= get_undo_list(&ulp
);
991 return ERR_PTR(error
);
993 spin_lock(&ulp
->lock
);
994 un
= lookup_undo(ulp
, semid
);
995 spin_unlock(&ulp
->lock
);
996 if (likely(un
!=NULL
))
999 /* no undo structure around - allocate one. */
1000 sma
= sem_lock_check(ns
, semid
);
1002 return ERR_PTR(PTR_ERR(sma
));
1004 nsems
= sma
->sem_nsems
;
1005 sem_getref_and_unlock(sma
);
1007 new = kzalloc(sizeof(struct sem_undo
) + sizeof(short)*nsems
, GFP_KERNEL
);
1010 return ERR_PTR(-ENOMEM
);
1012 new->semadj
= (short *) &new[1];
1015 spin_lock(&ulp
->lock
);
1016 un
= lookup_undo(ulp
, semid
);
1018 spin_unlock(&ulp
->lock
);
1023 sem_lock_and_putref(sma
);
1024 if (sma
->sem_perm
.deleted
) {
1026 spin_unlock(&ulp
->lock
);
1028 un
= ERR_PTR(-EIDRM
);
1031 new->proc_next
= ulp
->proc_list
;
1032 ulp
->proc_list
= new;
1033 new->id_next
= sma
->undo
;
1037 spin_unlock(&ulp
->lock
);
1042 asmlinkage
long sys_semtimedop(int semid
, struct sembuf __user
*tsops
,
1043 unsigned nsops
, const struct timespec __user
*timeout
)
1045 int error
= -EINVAL
;
1046 struct sem_array
*sma
;
1047 struct sembuf fast_sops
[SEMOPM_FAST
];
1048 struct sembuf
* sops
= fast_sops
, *sop
;
1049 struct sem_undo
*un
;
1050 int undos
= 0, alter
= 0, max
;
1051 struct sem_queue queue
;
1052 unsigned long jiffies_left
= 0;
1053 struct ipc_namespace
*ns
;
1055 ns
= current
->nsproxy
->ipc_ns
;
1057 if (nsops
< 1 || semid
< 0)
1059 if (nsops
> ns
->sc_semopm
)
1061 if(nsops
> SEMOPM_FAST
) {
1062 sops
= kmalloc(sizeof(*sops
)*nsops
,GFP_KERNEL
);
1066 if (copy_from_user (sops
, tsops
, nsops
* sizeof(*tsops
))) {
1071 struct timespec _timeout
;
1072 if (copy_from_user(&_timeout
, timeout
, sizeof(*timeout
))) {
1076 if (_timeout
.tv_sec
< 0 || _timeout
.tv_nsec
< 0 ||
1077 _timeout
.tv_nsec
>= 1000000000L) {
1081 jiffies_left
= timespec_to_jiffies(&_timeout
);
1084 for (sop
= sops
; sop
< sops
+ nsops
; sop
++) {
1085 if (sop
->sem_num
>= max
)
1087 if (sop
->sem_flg
& SEM_UNDO
)
1089 if (sop
->sem_op
!= 0)
1095 un
= find_undo(ns
, semid
);
1097 error
= PTR_ERR(un
);
1103 sma
= sem_lock_check(ns
, semid
);
1105 error
= PTR_ERR(sma
);
1110 * semid identifiers are not unique - find_undo may have
1111 * allocated an undo structure, it was invalidated by an RMID
1112 * and now a new array with received the same id. Check and retry.
1114 if (un
&& un
->semid
== -1) {
1119 if (max
>= sma
->sem_nsems
)
1120 goto out_unlock_free
;
1123 if (ipcperms(&sma
->sem_perm
, alter
? S_IWUGO
: S_IRUGO
))
1124 goto out_unlock_free
;
1126 error
= security_sem_semop(sma
, sops
, nsops
, alter
);
1128 goto out_unlock_free
;
1130 error
= try_atomic_semop (sma
, sops
, nsops
, un
, task_tgid_vnr(current
));
1132 if (alter
&& error
== 0)
1134 goto out_unlock_free
;
1137 /* We need to sleep on this operation, so we put the current
1138 * task into the pending queue and go to sleep.
1143 queue
.nsops
= nsops
;
1145 queue
.pid
= task_tgid_vnr(current
);
1147 queue
.alter
= alter
;
1149 append_to_queue(sma
,&queue
);
1151 prepend_to_queue(sma
,&queue
);
1153 queue
.status
= -EINTR
;
1154 queue
.sleeper
= current
;
1155 current
->state
= TASK_INTERRUPTIBLE
;
1159 jiffies_left
= schedule_timeout(jiffies_left
);
1163 error
= queue
.status
;
1164 while(unlikely(error
== IN_WAKEUP
)) {
1166 error
= queue
.status
;
1169 if (error
!= -EINTR
) {
1170 /* fast path: update_queue already obtained all requested
1175 sma
= sem_lock(ns
, semid
);
1177 BUG_ON(queue
.prev
!= NULL
);
1183 * If queue.status != -EINTR we are woken up by another process
1185 error
= queue
.status
;
1186 if (error
!= -EINTR
) {
1187 goto out_unlock_free
;
1191 * If an interrupt occurred we have to clean up the queue
1193 if (timeout
&& jiffies_left
== 0)
1195 remove_from_queue(sma
,&queue
);
1196 goto out_unlock_free
;
1201 if(sops
!= fast_sops
)
1206 asmlinkage
long sys_semop (int semid
, struct sembuf __user
*tsops
, unsigned nsops
)
1208 return sys_semtimedop(semid
, tsops
, nsops
, NULL
);
1211 /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
1212 * parent and child tasks.
1215 int copy_semundo(unsigned long clone_flags
, struct task_struct
*tsk
)
1217 struct sem_undo_list
*undo_list
;
1220 if (clone_flags
& CLONE_SYSVSEM
) {
1221 error
= get_undo_list(&undo_list
);
1224 atomic_inc(&undo_list
->refcnt
);
1225 tsk
->sysvsem
.undo_list
= undo_list
;
1227 tsk
->sysvsem
.undo_list
= NULL
;
1233 * add semadj values to semaphores, free undo structures.
1234 * undo structures are not freed when semaphore arrays are destroyed
1235 * so some of them may be out of date.
1236 * IMPLEMENTATION NOTE: There is some confusion over whether the
1237 * set of adjustments that needs to be done should be done in an atomic
1238 * manner or not. That is, if we are attempting to decrement the semval
1239 * should we queue up and wait until we can do so legally?
1240 * The original implementation attempted to do this (queue and wait).
1241 * The current implementation does not do so. The POSIX standard
1242 * and SVID should be consulted to determine what behavior is mandated.
1244 void exit_sem(struct task_struct
*tsk
)
1246 struct sem_undo_list
*undo_list
;
1247 struct sem_undo
*u
, **up
;
1248 struct ipc_namespace
*ns
;
1250 undo_list
= tsk
->sysvsem
.undo_list
;
1253 tsk
->sysvsem
.undo_list
= NULL
;
1255 if (!atomic_dec_and_test(&undo_list
->refcnt
))
1258 ns
= tsk
->nsproxy
->ipc_ns
;
1259 /* There's no need to hold the semundo list lock, as current
1260 * is the last task exiting for this undo list.
1262 for (up
= &undo_list
->proc_list
; (u
= *up
); *up
= u
->proc_next
, kfree(u
)) {
1263 struct sem_array
*sma
;
1265 struct sem_undo
*un
, **unp
;
1272 sma
= sem_lock(ns
, semid
);
1279 BUG_ON(sem_checkid(sma
, u
->semid
));
1281 /* remove u from the sma->undo list */
1282 for (unp
= &sma
->undo
; (un
= *unp
); unp
= &un
->id_next
) {
1286 printk ("exit_sem undo list error id=%d\n", u
->semid
);
1290 /* perform adjustments registered in u */
1291 nsems
= sma
->sem_nsems
;
1292 for (i
= 0; i
< nsems
; i
++) {
1293 struct sem
* semaphore
= &sma
->sem_base
[i
];
1295 semaphore
->semval
+= u
->semadj
[i
];
1297 * Range checks of the new semaphore value,
1298 * not defined by sus:
1299 * - Some unices ignore the undo entirely
1300 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
1301 * - some cap the value (e.g. FreeBSD caps
1302 * at 0, but doesn't enforce SEMVMX)
1304 * Linux caps the semaphore value, both at 0
1307 * Manfred <manfred@colorfullife.com>
1309 if (semaphore
->semval
< 0)
1310 semaphore
->semval
= 0;
1311 if (semaphore
->semval
> SEMVMX
)
1312 semaphore
->semval
= SEMVMX
;
1313 semaphore
->sempid
= task_tgid_vnr(current
);
1316 sma
->sem_otime
= get_seconds();
1317 /* maybe some queued-up processes were waiting for this */
1325 #ifdef CONFIG_PROC_FS
1326 static int sysvipc_sem_proc_show(struct seq_file
*s
, void *it
)
1328 struct sem_array
*sma
= it
;
1330 return seq_printf(s
,
1331 "%10d %10d %4o %10lu %5u %5u %5u %5u %10lu %10lu\n",