84c701fe5004f461ff27d7d073ee268180def732
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)
94 #define sem_buildid(id, seq) ipc_buildid(id, seq)
96 static struct ipc_ids init_sem_ids
;
98 static int newary(struct ipc_namespace
*, struct ipc_params
*);
99 static void freeary(struct ipc_namespace
*, struct sem_array
*);
100 #ifdef CONFIG_PROC_FS
101 static int sysvipc_sem_proc_show(struct seq_file
*s
, void *it
);
104 #define SEMMSL_FAST 256 /* 512 bytes on stack */
105 #define SEMOPM_FAST 64 /* ~ 372 bytes on stack */
108 * linked list protection:
110 * sem_array.sem_pending{,last},
111 * sem_array.sem_undo: sem_lock() for read/write
112 * sem_undo.proc_next: only "current" is allowed to read/write that field.
116 #define sc_semmsl sem_ctls[0]
117 #define sc_semmns sem_ctls[1]
118 #define sc_semopm sem_ctls[2]
119 #define sc_semmni sem_ctls[3]
121 static void __sem_init_ns(struct ipc_namespace
*ns
, struct ipc_ids
*ids
)
123 ns
->ids
[IPC_SEM_IDS
] = ids
;
124 ns
->sc_semmsl
= SEMMSL
;
125 ns
->sc_semmns
= SEMMNS
;
126 ns
->sc_semopm
= SEMOPM
;
127 ns
->sc_semmni
= SEMMNI
;
133 int sem_init_ns(struct ipc_namespace
*ns
)
137 ids
= kmalloc(sizeof(struct ipc_ids
), GFP_KERNEL
);
141 __sem_init_ns(ns
, ids
);
145 void sem_exit_ns(struct ipc_namespace
*ns
)
147 struct sem_array
*sma
;
148 struct kern_ipc_perm
*perm
;
152 down_write(&sem_ids(ns
).rw_mutex
);
154 in_use
= sem_ids(ns
).in_use
;
156 for (total
= 0, next_id
= 0; total
< in_use
; next_id
++) {
157 perm
= idr_find(&sem_ids(ns
).ipcs_idr
, next_id
);
160 ipc_lock_by_ptr(perm
);
161 sma
= container_of(perm
, struct sem_array
, sem_perm
);
165 up_write(&sem_ids(ns
).rw_mutex
);
167 kfree(ns
->ids
[IPC_SEM_IDS
]);
168 ns
->ids
[IPC_SEM_IDS
] = NULL
;
172 void __init
sem_init (void)
174 __sem_init_ns(&init_ipc_ns
, &init_sem_ids
);
175 ipc_init_proc_interface("sysvipc/sem",
176 " key semid perms nsems uid gid cuid cgid otime ctime\n",
177 IPC_SEM_IDS
, sysvipc_sem_proc_show
);
181 * This routine is called in the paths where the rw_mutex is held to protect
182 * access to the idr tree.
184 static inline struct sem_array
*sem_lock_check_down(struct ipc_namespace
*ns
,
187 struct kern_ipc_perm
*ipcp
= ipc_lock_check_down(&sem_ids(ns
), id
);
190 return (struct sem_array
*)ipcp
;
192 return container_of(ipcp
, struct sem_array
, sem_perm
);
196 * sem_lock_(check_) routines are called in the paths where the rw_mutex
199 static inline struct sem_array
*sem_lock(struct ipc_namespace
*ns
, int id
)
201 struct kern_ipc_perm
*ipcp
= ipc_lock(&sem_ids(ns
), id
);
204 return (struct sem_array
*)ipcp
;
206 return container_of(ipcp
, struct sem_array
, sem_perm
);
209 static inline struct sem_array
*sem_lock_check(struct ipc_namespace
*ns
,
212 struct kern_ipc_perm
*ipcp
= ipc_lock_check(&sem_ids(ns
), id
);
215 return (struct sem_array
*)ipcp
;
217 return container_of(ipcp
, struct sem_array
, sem_perm
);
220 static inline void sem_rmid(struct ipc_namespace
*ns
, struct sem_array
*s
)
222 ipc_rmid(&sem_ids(ns
), &s
->sem_perm
);
226 * Lockless wakeup algorithm:
227 * Without the check/retry algorithm a lockless wakeup is possible:
228 * - queue.status is initialized to -EINTR before blocking.
229 * - wakeup is performed by
230 * * unlinking the queue entry from sma->sem_pending
231 * * setting queue.status to IN_WAKEUP
232 * This is the notification for the blocked thread that a
233 * result value is imminent.
234 * * call wake_up_process
235 * * set queue.status to the final value.
236 * - the previously blocked thread checks queue.status:
237 * * if it's IN_WAKEUP, then it must wait until the value changes
238 * * if it's not -EINTR, then the operation was completed by
239 * update_queue. semtimedop can return queue.status without
240 * performing any operation on the sem array.
241 * * otherwise it must acquire the spinlock and check what's up.
243 * The two-stage algorithm is necessary to protect against the following
245 * - if queue.status is set after wake_up_process, then the woken up idle
246 * thread could race forward and try (and fail) to acquire sma->lock
247 * before update_queue had a chance to set queue.status
248 * - if queue.status is written before wake_up_process and if the
249 * blocked process is woken up by a signal between writing
250 * queue.status and the wake_up_process, then the woken up
251 * process could return from semtimedop and die by calling
252 * sys_exit before wake_up_process is called. Then wake_up_process
253 * will oops, because the task structure is already invalid.
254 * (yes, this happened on s390 with sysv msg).
260 * newary - Create a new semaphore set
262 * @params: ptr to the structure that contains key, semflg and nsems
264 * Called with sem_ids.rw_mutex held (as a writer)
267 static int newary(struct ipc_namespace
*ns
, struct ipc_params
*params
)
271 struct sem_array
*sma
;
273 key_t key
= params
->key
;
274 int nsems
= params
->u
.nsems
;
275 int semflg
= params
->flg
;
279 if (ns
->used_sems
+ nsems
> ns
->sc_semmns
)
282 size
= sizeof (*sma
) + nsems
* sizeof (struct sem
);
283 sma
= ipc_rcu_alloc(size
);
287 memset (sma
, 0, size
);
289 sma
->sem_perm
.mode
= (semflg
& S_IRWXUGO
);
290 sma
->sem_perm
.key
= key
;
292 sma
->sem_perm
.security
= NULL
;
293 retval
= security_sem_alloc(sma
);
299 id
= ipc_addid(&sem_ids(ns
), &sma
->sem_perm
, ns
->sc_semmni
);
301 security_sem_free(sma
);
305 ns
->used_sems
+= nsems
;
307 sma
->sem_perm
.id
= sem_buildid(id
, sma
->sem_perm
.seq
);
308 sma
->sem_base
= (struct sem
*) &sma
[1];
309 /* sma->sem_pending = NULL; */
310 sma
->sem_pending_last
= &sma
->sem_pending
;
311 /* sma->undo = NULL; */
312 sma
->sem_nsems
= nsems
;
313 sma
->sem_ctime
= get_seconds();
316 return sma
->sem_perm
.id
;
321 * Called with sem_ids.rw_mutex and ipcp locked.
323 static inline int sem_security(struct kern_ipc_perm
*ipcp
, int semflg
)
325 struct sem_array
*sma
;
327 sma
= container_of(ipcp
, struct sem_array
, sem_perm
);
328 return security_sem_associate(sma
, semflg
);
332 * Called with sem_ids.rw_mutex and ipcp locked.
334 static inline int sem_more_checks(struct kern_ipc_perm
*ipcp
,
335 struct ipc_params
*params
)
337 struct sem_array
*sma
;
339 sma
= container_of(ipcp
, struct sem_array
, sem_perm
);
340 if (params
->u
.nsems
> sma
->sem_nsems
)
346 asmlinkage
long sys_semget(key_t key
, int nsems
, int semflg
)
348 struct ipc_namespace
*ns
;
349 struct ipc_ops sem_ops
;
350 struct ipc_params sem_params
;
352 ns
= current
->nsproxy
->ipc_ns
;
354 if (nsems
< 0 || nsems
> ns
->sc_semmsl
)
357 sem_ops
.getnew
= newary
;
358 sem_ops
.associate
= sem_security
;
359 sem_ops
.more_checks
= sem_more_checks
;
361 sem_params
.key
= key
;
362 sem_params
.flg
= semflg
;
363 sem_params
.u
.nsems
= nsems
;
365 return ipcget(ns
, &sem_ids(ns
), &sem_ops
, &sem_params
);
368 /* Manage the doubly linked list sma->sem_pending as a FIFO:
369 * insert new queue elements at the tail sma->sem_pending_last.
371 static inline void append_to_queue (struct sem_array
* sma
,
372 struct sem_queue
* q
)
374 *(q
->prev
= sma
->sem_pending_last
) = q
;
375 *(sma
->sem_pending_last
= &q
->next
) = NULL
;
378 static inline void prepend_to_queue (struct sem_array
* sma
,
379 struct sem_queue
* q
)
381 q
->next
= sma
->sem_pending
;
382 *(q
->prev
= &sma
->sem_pending
) = q
;
384 q
->next
->prev
= &q
->next
;
385 else /* sma->sem_pending_last == &sma->sem_pending */
386 sma
->sem_pending_last
= &q
->next
;
389 static inline void remove_from_queue (struct sem_array
* sma
,
390 struct sem_queue
* q
)
392 *(q
->prev
) = q
->next
;
394 q
->next
->prev
= q
->prev
;
395 else /* sma->sem_pending_last == &q->next */
396 sma
->sem_pending_last
= q
->prev
;
397 q
->prev
= NULL
; /* mark as removed */
401 * Determine whether a sequence of semaphore operations would succeed
402 * all at once. Return 0 if yes, 1 if need to sleep, else return error code.
405 static int try_atomic_semop (struct sem_array
* sma
, struct sembuf
* sops
,
406 int nsops
, struct sem_undo
*un
, int pid
)
412 for (sop
= sops
; sop
< sops
+ nsops
; sop
++) {
413 curr
= sma
->sem_base
+ sop
->sem_num
;
414 sem_op
= sop
->sem_op
;
415 result
= curr
->semval
;
417 if (!sem_op
&& result
)
425 if (sop
->sem_flg
& SEM_UNDO
) {
426 int undo
= un
->semadj
[sop
->sem_num
] - sem_op
;
428 * Exceeding the undo range is an error.
430 if (undo
< (-SEMAEM
- 1) || undo
> SEMAEM
)
433 curr
->semval
= result
;
437 while (sop
>= sops
) {
438 sma
->sem_base
[sop
->sem_num
].sempid
= pid
;
439 if (sop
->sem_flg
& SEM_UNDO
)
440 un
->semadj
[sop
->sem_num
] -= sop
->sem_op
;
444 sma
->sem_otime
= get_seconds();
452 if (sop
->sem_flg
& IPC_NOWAIT
)
459 while (sop
>= sops
) {
460 sma
->sem_base
[sop
->sem_num
].semval
-= sop
->sem_op
;
467 /* Go through the pending queue for the indicated semaphore
468 * looking for tasks that can be completed.
470 static void update_queue (struct sem_array
* sma
)
473 struct sem_queue
* q
;
475 q
= sma
->sem_pending
;
477 error
= try_atomic_semop(sma
, q
->sops
, q
->nsops
,
480 /* Does q->sleeper still need to sleep? */
483 remove_from_queue(sma
,q
);
484 q
->status
= IN_WAKEUP
;
486 * Continue scanning. The next operation
487 * that must be checked depends on the type of the
488 * completed operation:
489 * - if the operation modified the array, then
490 * restart from the head of the queue and
491 * check for threads that might be waiting
492 * for semaphore values to become 0.
493 * - if the operation didn't modify the array,
494 * then just continue.
497 n
= sma
->sem_pending
;
500 wake_up_process(q
->sleeper
);
501 /* hands-off: q will disappear immediately after
513 /* The following counts are associated to each semaphore:
514 * semncnt number of tasks waiting on semval being nonzero
515 * semzcnt number of tasks waiting on semval being zero
516 * This model assumes that a task waits on exactly one semaphore.
517 * Since semaphore operations are to be performed atomically, tasks actually
518 * wait on a whole sequence of semaphores simultaneously.
519 * The counts we return here are a rough approximation, but still
520 * warrant that semncnt+semzcnt>0 if the task is on the pending queue.
522 static int count_semncnt (struct sem_array
* sma
, ushort semnum
)
525 struct sem_queue
* q
;
528 for (q
= sma
->sem_pending
; q
; q
= q
->next
) {
529 struct sembuf
* sops
= q
->sops
;
530 int nsops
= q
->nsops
;
532 for (i
= 0; i
< nsops
; i
++)
533 if (sops
[i
].sem_num
== semnum
534 && (sops
[i
].sem_op
< 0)
535 && !(sops
[i
].sem_flg
& IPC_NOWAIT
))
540 static int count_semzcnt (struct sem_array
* sma
, ushort semnum
)
543 struct sem_queue
* q
;
546 for (q
= sma
->sem_pending
; q
; q
= q
->next
) {
547 struct sembuf
* sops
= q
->sops
;
548 int nsops
= q
->nsops
;
550 for (i
= 0; i
< nsops
; i
++)
551 if (sops
[i
].sem_num
== semnum
552 && (sops
[i
].sem_op
== 0)
553 && !(sops
[i
].sem_flg
& IPC_NOWAIT
))
559 /* Free a semaphore set. freeary() is called with sem_ids.rw_mutex locked
560 * as a writer and the spinlock for this semaphore set hold. sem_ids.rw_mutex
561 * remains locked on exit.
563 static void freeary(struct ipc_namespace
*ns
, struct sem_array
*sma
)
568 /* Invalidate the existing undo structures for this semaphore set.
569 * (They will be freed without any further action in exit_sem()
570 * or during the next semop.)
572 for (un
= sma
->undo
; un
; un
= un
->id_next
)
575 /* Wake up all pending processes and let them fail with EIDRM. */
576 q
= sma
->sem_pending
;
579 /* lazy remove_from_queue: we are killing the whole queue */
582 q
->status
= IN_WAKEUP
;
583 wake_up_process(q
->sleeper
); /* doesn't sleep */
585 q
->status
= -EIDRM
; /* hands-off q */
589 /* Remove the semaphore set from the IDR */
593 ns
->used_sems
-= sma
->sem_nsems
;
594 security_sem_free(sma
);
598 static unsigned long copy_semid_to_user(void __user
*buf
, struct semid64_ds
*in
, int version
)
602 return copy_to_user(buf
, in
, sizeof(*in
));
607 ipc64_perm_to_ipc_perm(&in
->sem_perm
, &out
.sem_perm
);
609 out
.sem_otime
= in
->sem_otime
;
610 out
.sem_ctime
= in
->sem_ctime
;
611 out
.sem_nsems
= in
->sem_nsems
;
613 return copy_to_user(buf
, &out
, sizeof(out
));
620 static int semctl_nolock(struct ipc_namespace
*ns
, int semid
, int semnum
,
621 int cmd
, int version
, union semun arg
)
624 struct sem_array
*sma
;
630 struct seminfo seminfo
;
633 err
= security_sem_semctl(NULL
, cmd
);
637 memset(&seminfo
,0,sizeof(seminfo
));
638 seminfo
.semmni
= ns
->sc_semmni
;
639 seminfo
.semmns
= ns
->sc_semmns
;
640 seminfo
.semmsl
= ns
->sc_semmsl
;
641 seminfo
.semopm
= ns
->sc_semopm
;
642 seminfo
.semvmx
= SEMVMX
;
643 seminfo
.semmnu
= SEMMNU
;
644 seminfo
.semmap
= SEMMAP
;
645 seminfo
.semume
= SEMUME
;
646 down_read(&sem_ids(ns
).rw_mutex
);
647 if (cmd
== SEM_INFO
) {
648 seminfo
.semusz
= sem_ids(ns
).in_use
;
649 seminfo
.semaem
= ns
->used_sems
;
651 seminfo
.semusz
= SEMUSZ
;
652 seminfo
.semaem
= SEMAEM
;
654 max_id
= ipc_get_maxid(&sem_ids(ns
));
655 up_read(&sem_ids(ns
).rw_mutex
);
656 if (copy_to_user (arg
.__buf
, &seminfo
, sizeof(struct seminfo
)))
658 return (max_id
< 0) ? 0: max_id
;
662 struct semid64_ds tbuf
;
665 sma
= sem_lock(ns
, semid
);
670 if (ipcperms (&sma
->sem_perm
, S_IRUGO
))
673 err
= security_sem_semctl(sma
, cmd
);
677 id
= sma
->sem_perm
.id
;
679 memset(&tbuf
, 0, sizeof(tbuf
));
681 kernel_to_ipc64_perm(&sma
->sem_perm
, &tbuf
.sem_perm
);
682 tbuf
.sem_otime
= sma
->sem_otime
;
683 tbuf
.sem_ctime
= sma
->sem_ctime
;
684 tbuf
.sem_nsems
= sma
->sem_nsems
;
686 if (copy_semid_to_user (arg
.buf
, &tbuf
, version
))
699 static int semctl_main(struct ipc_namespace
*ns
, int semid
, int semnum
,
700 int cmd
, int version
, union semun arg
)
702 struct sem_array
*sma
;
705 ushort fast_sem_io
[SEMMSL_FAST
];
706 ushort
* sem_io
= fast_sem_io
;
709 sma
= sem_lock_check(ns
, semid
);
713 nsems
= sma
->sem_nsems
;
716 if (ipcperms (&sma
->sem_perm
, (cmd
==SETVAL
||cmd
==SETALL
)?S_IWUGO
:S_IRUGO
))
719 err
= security_sem_semctl(sma
, cmd
);
727 ushort __user
*array
= arg
.array
;
730 if(nsems
> SEMMSL_FAST
) {
734 sem_io
= ipc_alloc(sizeof(ushort
)*nsems
);
736 ipc_lock_by_ptr(&sma
->sem_perm
);
742 ipc_lock_by_ptr(&sma
->sem_perm
);
744 if (sma
->sem_perm
.deleted
) {
751 for (i
= 0; i
< sma
->sem_nsems
; i
++)
752 sem_io
[i
] = sma
->sem_base
[i
].semval
;
755 if(copy_to_user(array
, sem_io
, nsems
*sizeof(ushort
)))
767 if(nsems
> SEMMSL_FAST
) {
768 sem_io
= ipc_alloc(sizeof(ushort
)*nsems
);
770 ipc_lock_by_ptr(&sma
->sem_perm
);
777 if (copy_from_user (sem_io
, arg
.array
, nsems
*sizeof(ushort
))) {
778 ipc_lock_by_ptr(&sma
->sem_perm
);
785 for (i
= 0; i
< nsems
; i
++) {
786 if (sem_io
[i
] > SEMVMX
) {
787 ipc_lock_by_ptr(&sma
->sem_perm
);
794 ipc_lock_by_ptr(&sma
->sem_perm
);
796 if (sma
->sem_perm
.deleted
) {
802 for (i
= 0; i
< nsems
; i
++)
803 sma
->sem_base
[i
].semval
= sem_io
[i
];
804 for (un
= sma
->undo
; un
; un
= un
->id_next
)
805 for (i
= 0; i
< nsems
; i
++)
807 sma
->sem_ctime
= get_seconds();
808 /* maybe some queued-up processes were waiting for this */
815 struct semid64_ds tbuf
;
816 memset(&tbuf
,0,sizeof(tbuf
));
817 kernel_to_ipc64_perm(&sma
->sem_perm
, &tbuf
.sem_perm
);
818 tbuf
.sem_otime
= sma
->sem_otime
;
819 tbuf
.sem_ctime
= sma
->sem_ctime
;
820 tbuf
.sem_nsems
= sma
->sem_nsems
;
822 if (copy_semid_to_user (arg
.buf
, &tbuf
, version
))
826 /* GETVAL, GETPID, GETNCTN, GETZCNT, SETVAL: fall-through */
829 if(semnum
< 0 || semnum
>= nsems
)
832 curr
= &sma
->sem_base
[semnum
];
842 err
= count_semncnt(sma
,semnum
);
845 err
= count_semzcnt(sma
,semnum
);
852 if (val
> SEMVMX
|| val
< 0)
855 for (un
= sma
->undo
; un
; un
= un
->id_next
)
856 un
->semadj
[semnum
] = 0;
858 curr
->sempid
= task_tgid_vnr(current
);
859 sma
->sem_ctime
= get_seconds();
860 /* maybe some queued-up processes were waiting for this */
869 if(sem_io
!= fast_sem_io
)
870 ipc_free(sem_io
, sizeof(ushort
)*nsems
);
880 static inline unsigned long copy_semid_from_user(struct sem_setbuf
*out
, void __user
*buf
, int version
)
885 struct semid64_ds tbuf
;
887 if(copy_from_user(&tbuf
, buf
, sizeof(tbuf
)))
890 out
->uid
= tbuf
.sem_perm
.uid
;
891 out
->gid
= tbuf
.sem_perm
.gid
;
892 out
->mode
= tbuf
.sem_perm
.mode
;
898 struct semid_ds tbuf_old
;
900 if(copy_from_user(&tbuf_old
, buf
, sizeof(tbuf_old
)))
903 out
->uid
= tbuf_old
.sem_perm
.uid
;
904 out
->gid
= tbuf_old
.sem_perm
.gid
;
905 out
->mode
= tbuf_old
.sem_perm
.mode
;
914 static int semctl_down(struct ipc_namespace
*ns
, int semid
, int semnum
,
915 int cmd
, int version
, union semun arg
)
917 struct sem_array
*sma
;
919 struct sem_setbuf
uninitialized_var(setbuf
);
920 struct kern_ipc_perm
*ipcp
;
923 if(copy_semid_from_user (&setbuf
, arg
.buf
, version
))
926 sma
= sem_lock_check_down(ns
, semid
);
930 ipcp
= &sma
->sem_perm
;
932 err
= audit_ipc_obj(ipcp
);
936 if (cmd
== IPC_SET
) {
937 err
= audit_ipc_set_perm(0, setbuf
.uid
, setbuf
.gid
, setbuf
.mode
);
941 if (current
->euid
!= ipcp
->cuid
&&
942 current
->euid
!= ipcp
->uid
&& !capable(CAP_SYS_ADMIN
)) {
947 err
= security_sem_semctl(sma
, cmd
);
957 ipcp
->uid
= setbuf
.uid
;
958 ipcp
->gid
= setbuf
.gid
;
959 ipcp
->mode
= (ipcp
->mode
& ~S_IRWXUGO
)
960 | (setbuf
.mode
& S_IRWXUGO
);
961 sma
->sem_ctime
= get_seconds();
977 asmlinkage
long sys_semctl (int semid
, int semnum
, int cmd
, union semun arg
)
981 struct ipc_namespace
*ns
;
986 version
= ipc_parse_version(&cmd
);
987 ns
= current
->nsproxy
->ipc_ns
;
993 err
= semctl_nolock(ns
,semid
,semnum
,cmd
,version
,arg
);
1003 err
= semctl_main(ns
,semid
,semnum
,cmd
,version
,arg
);
1007 down_write(&sem_ids(ns
).rw_mutex
);
1008 err
= semctl_down(ns
,semid
,semnum
,cmd
,version
,arg
);
1009 up_write(&sem_ids(ns
).rw_mutex
);
1016 /* If the task doesn't already have a undo_list, then allocate one
1017 * here. We guarantee there is only one thread using this undo list,
1018 * and current is THE ONE
1020 * If this allocation and assignment succeeds, but later
1021 * portions of this code fail, there is no need to free the sem_undo_list.
1022 * Just let it stay associated with the task, and it'll be freed later
1025 * This can block, so callers must hold no locks.
1027 static inline int get_undo_list(struct sem_undo_list
**undo_listp
)
1029 struct sem_undo_list
*undo_list
;
1031 undo_list
= current
->sysvsem
.undo_list
;
1033 undo_list
= kzalloc(sizeof(*undo_list
), GFP_KERNEL
);
1034 if (undo_list
== NULL
)
1036 spin_lock_init(&undo_list
->lock
);
1037 atomic_set(&undo_list
->refcnt
, 1);
1038 current
->sysvsem
.undo_list
= undo_list
;
1040 *undo_listp
= undo_list
;
1044 static struct sem_undo
*lookup_undo(struct sem_undo_list
*ulp
, int semid
)
1046 struct sem_undo
**last
, *un
;
1048 last
= &ulp
->proc_list
;
1051 if(un
->semid
==semid
)
1054 *last
=un
->proc_next
;
1057 last
=&un
->proc_next
;
1064 static struct sem_undo
*find_undo(struct ipc_namespace
*ns
, int semid
)
1066 struct sem_array
*sma
;
1067 struct sem_undo_list
*ulp
;
1068 struct sem_undo
*un
, *new;
1072 error
= get_undo_list(&ulp
);
1074 return ERR_PTR(error
);
1076 spin_lock(&ulp
->lock
);
1077 un
= lookup_undo(ulp
, semid
);
1078 spin_unlock(&ulp
->lock
);
1079 if (likely(un
!=NULL
))
1082 /* no undo structure around - allocate one. */
1083 sma
= sem_lock_check(ns
, semid
);
1085 return ERR_PTR(PTR_ERR(sma
));
1087 nsems
= sma
->sem_nsems
;
1088 ipc_rcu_getref(sma
);
1091 new = kzalloc(sizeof(struct sem_undo
) + sizeof(short)*nsems
, GFP_KERNEL
);
1093 ipc_lock_by_ptr(&sma
->sem_perm
);
1094 ipc_rcu_putref(sma
);
1096 return ERR_PTR(-ENOMEM
);
1098 new->semadj
= (short *) &new[1];
1101 spin_lock(&ulp
->lock
);
1102 un
= lookup_undo(ulp
, semid
);
1104 spin_unlock(&ulp
->lock
);
1106 ipc_lock_by_ptr(&sma
->sem_perm
);
1107 ipc_rcu_putref(sma
);
1111 ipc_lock_by_ptr(&sma
->sem_perm
);
1112 ipc_rcu_putref(sma
);
1113 if (sma
->sem_perm
.deleted
) {
1115 spin_unlock(&ulp
->lock
);
1117 un
= ERR_PTR(-EIDRM
);
1120 new->proc_next
= ulp
->proc_list
;
1121 ulp
->proc_list
= new;
1122 new->id_next
= sma
->undo
;
1126 spin_unlock(&ulp
->lock
);
1131 asmlinkage
long sys_semtimedop(int semid
, struct sembuf __user
*tsops
,
1132 unsigned nsops
, const struct timespec __user
*timeout
)
1134 int error
= -EINVAL
;
1135 struct sem_array
*sma
;
1136 struct sembuf fast_sops
[SEMOPM_FAST
];
1137 struct sembuf
* sops
= fast_sops
, *sop
;
1138 struct sem_undo
*un
;
1139 int undos
= 0, alter
= 0, max
;
1140 struct sem_queue queue
;
1141 unsigned long jiffies_left
= 0;
1142 struct ipc_namespace
*ns
;
1144 ns
= current
->nsproxy
->ipc_ns
;
1146 if (nsops
< 1 || semid
< 0)
1148 if (nsops
> ns
->sc_semopm
)
1150 if(nsops
> SEMOPM_FAST
) {
1151 sops
= kmalloc(sizeof(*sops
)*nsops
,GFP_KERNEL
);
1155 if (copy_from_user (sops
, tsops
, nsops
* sizeof(*tsops
))) {
1160 struct timespec _timeout
;
1161 if (copy_from_user(&_timeout
, timeout
, sizeof(*timeout
))) {
1165 if (_timeout
.tv_sec
< 0 || _timeout
.tv_nsec
< 0 ||
1166 _timeout
.tv_nsec
>= 1000000000L) {
1170 jiffies_left
= timespec_to_jiffies(&_timeout
);
1173 for (sop
= sops
; sop
< sops
+ nsops
; sop
++) {
1174 if (sop
->sem_num
>= max
)
1176 if (sop
->sem_flg
& SEM_UNDO
)
1178 if (sop
->sem_op
!= 0)
1184 un
= find_undo(ns
, semid
);
1186 error
= PTR_ERR(un
);
1192 sma
= sem_lock_check(ns
, semid
);
1194 error
= PTR_ERR(sma
);
1199 * semid identifiers are not unique - find_undo may have
1200 * allocated an undo structure, it was invalidated by an RMID
1201 * and now a new array with received the same id. Check and retry.
1203 if (un
&& un
->semid
== -1) {
1208 if (max
>= sma
->sem_nsems
)
1209 goto out_unlock_free
;
1212 if (ipcperms(&sma
->sem_perm
, alter
? S_IWUGO
: S_IRUGO
))
1213 goto out_unlock_free
;
1215 error
= security_sem_semop(sma
, sops
, nsops
, alter
);
1217 goto out_unlock_free
;
1219 error
= try_atomic_semop (sma
, sops
, nsops
, un
, task_tgid_vnr(current
));
1221 if (alter
&& error
== 0)
1223 goto out_unlock_free
;
1226 /* We need to sleep on this operation, so we put the current
1227 * task into the pending queue and go to sleep.
1232 queue
.nsops
= nsops
;
1234 queue
.pid
= task_tgid_vnr(current
);
1236 queue
.alter
= alter
;
1238 append_to_queue(sma
,&queue
);
1240 prepend_to_queue(sma
,&queue
);
1242 queue
.status
= -EINTR
;
1243 queue
.sleeper
= current
;
1244 current
->state
= TASK_INTERRUPTIBLE
;
1248 jiffies_left
= schedule_timeout(jiffies_left
);
1252 error
= queue
.status
;
1253 while(unlikely(error
== IN_WAKEUP
)) {
1255 error
= queue
.status
;
1258 if (error
!= -EINTR
) {
1259 /* fast path: update_queue already obtained all requested
1264 sma
= sem_lock(ns
, semid
);
1266 BUG_ON(queue
.prev
!= NULL
);
1272 * If queue.status != -EINTR we are woken up by another process
1274 error
= queue
.status
;
1275 if (error
!= -EINTR
) {
1276 goto out_unlock_free
;
1280 * If an interrupt occurred we have to clean up the queue
1282 if (timeout
&& jiffies_left
== 0)
1284 remove_from_queue(sma
,&queue
);
1285 goto out_unlock_free
;
1290 if(sops
!= fast_sops
)
1295 asmlinkage
long sys_semop (int semid
, struct sembuf __user
*tsops
, unsigned nsops
)
1297 return sys_semtimedop(semid
, tsops
, nsops
, NULL
);
1300 /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
1301 * parent and child tasks.
1304 int copy_semundo(unsigned long clone_flags
, struct task_struct
*tsk
)
1306 struct sem_undo_list
*undo_list
;
1309 if (clone_flags
& CLONE_SYSVSEM
) {
1310 error
= get_undo_list(&undo_list
);
1313 atomic_inc(&undo_list
->refcnt
);
1314 tsk
->sysvsem
.undo_list
= undo_list
;
1316 tsk
->sysvsem
.undo_list
= NULL
;
1322 * add semadj values to semaphores, free undo structures.
1323 * undo structures are not freed when semaphore arrays are destroyed
1324 * so some of them may be out of date.
1325 * IMPLEMENTATION NOTE: There is some confusion over whether the
1326 * set of adjustments that needs to be done should be done in an atomic
1327 * manner or not. That is, if we are attempting to decrement the semval
1328 * should we queue up and wait until we can do so legally?
1329 * The original implementation attempted to do this (queue and wait).
1330 * The current implementation does not do so. The POSIX standard
1331 * and SVID should be consulted to determine what behavior is mandated.
1333 void exit_sem(struct task_struct
*tsk
)
1335 struct sem_undo_list
*undo_list
;
1336 struct sem_undo
*u
, **up
;
1337 struct ipc_namespace
*ns
;
1339 undo_list
= tsk
->sysvsem
.undo_list
;
1343 if (!atomic_dec_and_test(&undo_list
->refcnt
))
1346 ns
= tsk
->nsproxy
->ipc_ns
;
1347 /* There's no need to hold the semundo list lock, as current
1348 * is the last task exiting for this undo list.
1350 for (up
= &undo_list
->proc_list
; (u
= *up
); *up
= u
->proc_next
, kfree(u
)) {
1351 struct sem_array
*sma
;
1353 struct sem_undo
*un
, **unp
;
1360 sma
= sem_lock(ns
, semid
);
1367 BUG_ON(sem_checkid(sma
, u
->semid
));
1369 /* remove u from the sma->undo list */
1370 for (unp
= &sma
->undo
; (un
= *unp
); unp
= &un
->id_next
) {
1374 printk ("exit_sem undo list error id=%d\n", u
->semid
);
1378 /* perform adjustments registered in u */
1379 nsems
= sma
->sem_nsems
;
1380 for (i
= 0; i
< nsems
; i
++) {
1381 struct sem
* semaphore
= &sma
->sem_base
[i
];
1383 semaphore
->semval
+= u
->semadj
[i
];
1385 * Range checks of the new semaphore value,
1386 * not defined by sus:
1387 * - Some unices ignore the undo entirely
1388 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
1389 * - some cap the value (e.g. FreeBSD caps
1390 * at 0, but doesn't enforce SEMVMX)
1392 * Linux caps the semaphore value, both at 0
1395 * Manfred <manfred@colorfullife.com>
1397 if (semaphore
->semval
< 0)
1398 semaphore
->semval
= 0;
1399 if (semaphore
->semval
> SEMVMX
)
1400 semaphore
->semval
= SEMVMX
;
1401 semaphore
->sempid
= task_tgid_vnr(current
);
1404 sma
->sem_otime
= get_seconds();
1405 /* maybe some queued-up processes were waiting for this */
1413 #ifdef CONFIG_PROC_FS
1414 static int sysvipc_sem_proc_show(struct seq_file
*s
, void *it
)
1416 struct sem_array
*sma
= it
;
1418 return seq_printf(s
,
1419 "%10d %10d %4o %10lu %5u %5u %5u %5u %10lu %10lu\n",
This page took 0.081964 seconds and 5 git commands to generate.