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>
86 #include <asm/uaccess.h>
89 #define sem_ids(ns) (*((ns)->ids[IPC_SEM_IDS]))
91 #define sem_unlock(sma) ipc_unlock(&(sma)->sem_perm)
92 #define sem_checkid(ns, sma, semid) \
93 ipc_checkid(&sem_ids(ns),&sma->sem_perm,semid)
94 #define sem_buildid(ns, id, seq) \
95 ipc_buildid(&sem_ids(ns), id, seq)
97 static struct ipc_ids init_sem_ids;
99 static int newary(struct ipc_namespace *, struct ipc_params *);
100 static void freeary(struct ipc_namespace *, struct sem_array *);
101 #ifdef CONFIG_PROC_FS
102 static int sysvipc_sem_proc_show(struct seq_file *s, void *it);
105 #define SEMMSL_FAST 256 /* 512 bytes on stack */
106 #define SEMOPM_FAST 64 /* ~ 372 bytes on stack */
109 * linked list protection:
111 * sem_array.sem_pending{,last},
112 * sem_array.sem_undo: sem_lock() for read/write
113 * sem_undo.proc_next: only "current" is allowed to read/write that field.
117 #define sc_semmsl sem_ctls[0]
118 #define sc_semmns sem_ctls[1]
119 #define sc_semopm sem_ctls[2]
120 #define sc_semmni sem_ctls[3]
122 static void __sem_init_ns(struct ipc_namespace *ns, struct ipc_ids *ids)
124 ns->ids[IPC_SEM_IDS] = ids;
125 ns->sc_semmsl = SEMMSL;
126 ns->sc_semmns = SEMMNS;
127 ns->sc_semopm = SEMOPM;
128 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;
151 down_write(&sem_ids(ns).rw_mutex);
153 in_use = sem_ids(ns).in_use;
155 for (total = 0, next_id = 0; total < in_use; next_id++) {
156 sma = idr_find(&sem_ids(ns).ipcs_idr, next_id);
159 ipc_lock_by_ptr(&sma->sem_perm);
163 up_write(&sem_ids(ns).rw_mutex);
165 kfree(ns->ids[IPC_SEM_IDS]);
166 ns->ids[IPC_SEM_IDS] = NULL;
169 void __init sem_init (void)
171 __sem_init_ns(&init_ipc_ns, &init_sem_ids);
172 ipc_init_proc_interface("sysvipc/sem",
173 " key semid perms nsems uid gid cuid cgid otime ctime\n",
174 IPC_SEM_IDS, sysvipc_sem_proc_show);
178 * This routine is called in the paths where the rw_mutex is held to protect
179 * access to the idr tree.
181 static inline struct sem_array *sem_lock_check_down(struct ipc_namespace *ns,
184 struct kern_ipc_perm *ipcp = ipc_lock_check_down(&sem_ids(ns), id);
186 return container_of(ipcp, struct sem_array, sem_perm);
190 * sem_lock_(check_) routines are called in the paths where the rw_mutex
193 static inline struct sem_array *sem_lock(struct ipc_namespace *ns, int id)
195 struct kern_ipc_perm *ipcp = ipc_lock(&sem_ids(ns), id);
197 return container_of(ipcp, struct sem_array, sem_perm);
200 static inline struct sem_array *sem_lock_check(struct ipc_namespace *ns,
203 struct kern_ipc_perm *ipcp = ipc_lock_check(&sem_ids(ns), id);
205 return container_of(ipcp, struct sem_array, sem_perm);
208 static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
210 ipc_rmid(&sem_ids(ns), &s->sem_perm);
214 * Lockless wakeup algorithm:
215 * Without the check/retry algorithm a lockless wakeup is possible:
216 * - queue.status is initialized to -EINTR before blocking.
217 * - wakeup is performed by
218 * * unlinking the queue entry from sma->sem_pending
219 * * setting queue.status to IN_WAKEUP
220 * This is the notification for the blocked thread that a
221 * result value is imminent.
222 * * call wake_up_process
223 * * set queue.status to the final value.
224 * - the previously blocked thread checks queue.status:
225 * * if it's IN_WAKEUP, then it must wait until the value changes
226 * * if it's not -EINTR, then the operation was completed by
227 * update_queue. semtimedop can return queue.status without
228 * performing any operation on the sem array.
229 * * otherwise it must acquire the spinlock and check what's up.
231 * The two-stage algorithm is necessary to protect against the following
233 * - if queue.status is set after wake_up_process, then the woken up idle
234 * thread could race forward and try (and fail) to acquire sma->lock
235 * before update_queue had a chance to set queue.status
236 * - if queue.status is written before wake_up_process and if the
237 * blocked process is woken up by a signal between writing
238 * queue.status and the wake_up_process, then the woken up
239 * process could return from semtimedop and die by calling
240 * sys_exit before wake_up_process is called. Then wake_up_process
241 * will oops, because the task structure is already invalid.
242 * (yes, this happened on s390 with sysv msg).
248 * newary - Create a new semaphore set
250 * @params: ptr to the structure that contains key, semflg and nsems
252 * Called with sem_ids.rw_mutex held (as a writer)
255 static int newary(struct ipc_namespace *ns, struct ipc_params *params)
259 struct sem_array *sma;
261 key_t key = params->key;
262 int nsems = params->u.nsems;
263 int semflg = params->flg;
267 if (ns->used_sems + nsems > ns->sc_semmns)
270 size = sizeof (*sma) + nsems * sizeof (struct sem);
271 sma = ipc_rcu_alloc(size);
275 memset (sma, 0, size);
277 sma->sem_perm.mode = (semflg & S_IRWXUGO);
278 sma->sem_perm.key = key;
280 sma->sem_perm.security = NULL;
281 retval = security_sem_alloc(sma);
287 id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
289 security_sem_free(sma);
293 ns->used_sems += nsems;
295 sma->sem_perm.id = sem_buildid(ns, id, sma->sem_perm.seq);
296 sma->sem_base = (struct sem *) &sma[1];
297 /* sma->sem_pending = NULL; */
298 sma->sem_pending_last = &sma->sem_pending;
299 /* sma->undo = NULL; */
300 sma->sem_nsems = nsems;
301 sma->sem_ctime = get_seconds();
304 return sma->sem_perm.id;
309 * Called with sem_ids.rw_mutex and ipcp locked.
311 static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg)
313 struct sem_array *sma;
315 sma = container_of(ipcp, struct sem_array, sem_perm);
316 return security_sem_associate(sma, semflg);
320 * Called with sem_ids.rw_mutex and ipcp locked.
322 static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
323 struct ipc_params *params)
325 struct sem_array *sma;
327 sma = container_of(ipcp, struct sem_array, sem_perm);
328 if (params->u.nsems > sma->sem_nsems)
334 asmlinkage long sys_semget(key_t key, int nsems, int semflg)
336 struct ipc_namespace *ns;
337 struct ipc_ops sem_ops;
338 struct ipc_params sem_params;
340 ns = current->nsproxy->ipc_ns;
342 if (nsems < 0 || nsems > ns->sc_semmsl)
345 sem_ops.getnew = newary;
346 sem_ops.associate = sem_security;
347 sem_ops.more_checks = sem_more_checks;
349 sem_params.key = key;
350 sem_params.flg = semflg;
351 sem_params.u.nsems = nsems;
353 return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
356 /* Manage the doubly linked list sma->sem_pending as a FIFO:
357 * insert new queue elements at the tail sma->sem_pending_last.
359 static inline void append_to_queue (struct sem_array * sma,
360 struct sem_queue * q)
362 *(q->prev = sma->sem_pending_last) = q;
363 *(sma->sem_pending_last = &q->next) = NULL;
366 static inline void prepend_to_queue (struct sem_array * sma,
367 struct sem_queue * q)
369 q->next = sma->sem_pending;
370 *(q->prev = &sma->sem_pending) = q;
372 q->next->prev = &q->next;
373 else /* sma->sem_pending_last == &sma->sem_pending */
374 sma->sem_pending_last = &q->next;
377 static inline void remove_from_queue (struct sem_array * sma,
378 struct sem_queue * q)
380 *(q->prev) = q->next;
382 q->next->prev = q->prev;
383 else /* sma->sem_pending_last == &q->next */
384 sma->sem_pending_last = q->prev;
385 q->prev = NULL; /* mark as removed */
389 * Determine whether a sequence of semaphore operations would succeed
390 * all at once. Return 0 if yes, 1 if need to sleep, else return error code.
393 static int try_atomic_semop (struct sem_array * sma, struct sembuf * sops,
394 int nsops, struct sem_undo *un, int pid)
400 for (sop = sops; sop < sops + nsops; sop++) {
401 curr = sma->sem_base + sop->sem_num;
402 sem_op = sop->sem_op;
403 result = curr->semval;
405 if (!sem_op && result)
413 if (sop->sem_flg & SEM_UNDO) {
414 int undo = un->semadj[sop->sem_num] - sem_op;
416 * Exceeding the undo range is an error.
418 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
421 curr->semval = result;
425 while (sop >= sops) {
426 sma->sem_base[sop->sem_num].sempid = pid;
427 if (sop->sem_flg & SEM_UNDO)
428 un->semadj[sop->sem_num] -= sop->sem_op;
432 sma->sem_otime = get_seconds();
440 if (sop->sem_flg & IPC_NOWAIT)
447 while (sop >= sops) {
448 sma->sem_base[sop->sem_num].semval -= sop->sem_op;
455 /* Go through the pending queue for the indicated semaphore
456 * looking for tasks that can be completed.
458 static void update_queue (struct sem_array * sma)
461 struct sem_queue * q;
463 q = sma->sem_pending;
465 error = try_atomic_semop(sma, q->sops, q->nsops,
468 /* Does q->sleeper still need to sleep? */
471 remove_from_queue(sma,q);
472 q->status = IN_WAKEUP;
474 * Continue scanning. The next operation
475 * that must be checked depends on the type of the
476 * completed operation:
477 * - if the operation modified the array, then
478 * restart from the head of the queue and
479 * check for threads that might be waiting
480 * for semaphore values to become 0.
481 * - if the operation didn't modify the array,
482 * then just continue.
485 n = sma->sem_pending;
488 wake_up_process(q->sleeper);
489 /* hands-off: q will disappear immediately after
501 /* The following counts are associated to each semaphore:
502 * semncnt number of tasks waiting on semval being nonzero
503 * semzcnt number of tasks waiting on semval being zero
504 * This model assumes that a task waits on exactly one semaphore.
505 * Since semaphore operations are to be performed atomically, tasks actually
506 * wait on a whole sequence of semaphores simultaneously.
507 * The counts we return here are a rough approximation, but still
508 * warrant that semncnt+semzcnt>0 if the task is on the pending queue.
510 static int count_semncnt (struct sem_array * sma, ushort semnum)
513 struct sem_queue * q;
516 for (q = sma->sem_pending; q; q = q->next) {
517 struct sembuf * sops = q->sops;
518 int nsops = q->nsops;
520 for (i = 0; i < nsops; i++)
521 if (sops[i].sem_num == semnum
522 && (sops[i].sem_op < 0)
523 && !(sops[i].sem_flg & IPC_NOWAIT))
528 static int count_semzcnt (struct sem_array * sma, ushort semnum)
531 struct sem_queue * q;
534 for (q = sma->sem_pending; q; q = q->next) {
535 struct sembuf * sops = q->sops;
536 int nsops = q->nsops;
538 for (i = 0; i < nsops; i++)
539 if (sops[i].sem_num == semnum
540 && (sops[i].sem_op == 0)
541 && !(sops[i].sem_flg & IPC_NOWAIT))
547 /* Free a semaphore set. freeary() is called with sem_ids.rw_mutex locked
548 * as a writer and the spinlock for this semaphore set hold. sem_ids.rw_mutex
549 * remains locked on exit.
551 static void freeary(struct ipc_namespace *ns, struct sem_array *sma)
556 /* Invalidate the existing undo structures for this semaphore set.
557 * (They will be freed without any further action in exit_sem()
558 * or during the next semop.)
560 for (un = sma->undo; un; un = un->id_next)
563 /* Wake up all pending processes and let them fail with EIDRM. */
564 q = sma->sem_pending;
567 /* lazy remove_from_queue: we are killing the whole queue */
570 q->status = IN_WAKEUP;
571 wake_up_process(q->sleeper); /* doesn't sleep */
573 q->status = -EIDRM; /* hands-off q */
577 /* Remove the semaphore set from the IDR */
581 ns->used_sems -= sma->sem_nsems;
582 security_sem_free(sma);
586 static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
590 return copy_to_user(buf, in, sizeof(*in));
595 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
597 out.sem_otime = in->sem_otime;
598 out.sem_ctime = in->sem_ctime;
599 out.sem_nsems = in->sem_nsems;
601 return copy_to_user(buf, &out, sizeof(out));
608 static int semctl_nolock(struct ipc_namespace *ns, int semid, int semnum,
609 int cmd, int version, union semun arg)
612 struct sem_array *sma;
618 struct seminfo seminfo;
621 err = security_sem_semctl(NULL, cmd);
625 memset(&seminfo,0,sizeof(seminfo));
626 seminfo.semmni = ns->sc_semmni;
627 seminfo.semmns = ns->sc_semmns;
628 seminfo.semmsl = ns->sc_semmsl;
629 seminfo.semopm = ns->sc_semopm;
630 seminfo.semvmx = SEMVMX;
631 seminfo.semmnu = SEMMNU;
632 seminfo.semmap = SEMMAP;
633 seminfo.semume = SEMUME;
634 down_read(&sem_ids(ns).rw_mutex);
635 if (cmd == SEM_INFO) {
636 seminfo.semusz = sem_ids(ns).in_use;
637 seminfo.semaem = ns->used_sems;
639 seminfo.semusz = SEMUSZ;
640 seminfo.semaem = SEMAEM;
642 max_id = ipc_get_maxid(&sem_ids(ns));
643 up_read(&sem_ids(ns).rw_mutex);
644 if (copy_to_user (arg.__buf, &seminfo, sizeof(struct seminfo)))
646 return (max_id < 0) ? 0: max_id;
650 struct semid64_ds tbuf;
653 sma = sem_lock(ns, semid);
658 if (ipcperms (&sma->sem_perm, S_IRUGO))
661 err = security_sem_semctl(sma, cmd);
665 id = sma->sem_perm.id;
667 memset(&tbuf, 0, sizeof(tbuf));
669 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
670 tbuf.sem_otime = sma->sem_otime;
671 tbuf.sem_ctime = sma->sem_ctime;
672 tbuf.sem_nsems = sma->sem_nsems;
674 if (copy_semid_to_user (arg.buf, &tbuf, version))
687 static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
688 int cmd, int version, union semun arg)
690 struct sem_array *sma;
693 ushort fast_sem_io[SEMMSL_FAST];
694 ushort* sem_io = fast_sem_io;
697 sma = sem_lock_check(ns, semid);
701 nsems = sma->sem_nsems;
704 if (ipcperms (&sma->sem_perm, (cmd==SETVAL||cmd==SETALL)?S_IWUGO:S_IRUGO))
707 err = security_sem_semctl(sma, cmd);
715 ushort __user *array = arg.array;
718 if(nsems > SEMMSL_FAST) {
722 sem_io = ipc_alloc(sizeof(ushort)*nsems);
724 ipc_lock_by_ptr(&sma->sem_perm);
730 ipc_lock_by_ptr(&sma->sem_perm);
732 if (sma->sem_perm.deleted) {
739 for (i = 0; i < sma->sem_nsems; i++)
740 sem_io[i] = sma->sem_base[i].semval;
743 if(copy_to_user(array, sem_io, nsems*sizeof(ushort)))
755 if(nsems > SEMMSL_FAST) {
756 sem_io = ipc_alloc(sizeof(ushort)*nsems);
758 ipc_lock_by_ptr(&sma->sem_perm);
765 if (copy_from_user (sem_io, arg.array, nsems*sizeof(ushort))) {
766 ipc_lock_by_ptr(&sma->sem_perm);
773 for (i = 0; i < nsems; i++) {
774 if (sem_io[i] > SEMVMX) {
775 ipc_lock_by_ptr(&sma->sem_perm);
782 ipc_lock_by_ptr(&sma->sem_perm);
784 if (sma->sem_perm.deleted) {
790 for (i = 0; i < nsems; i++)
791 sma->sem_base[i].semval = sem_io[i];
792 for (un = sma->undo; un; un = un->id_next)
793 for (i = 0; i < nsems; i++)
795 sma->sem_ctime = get_seconds();
796 /* maybe some queued-up processes were waiting for this */
803 struct semid64_ds tbuf;
804 memset(&tbuf,0,sizeof(tbuf));
805 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
806 tbuf.sem_otime = sma->sem_otime;
807 tbuf.sem_ctime = sma->sem_ctime;
808 tbuf.sem_nsems = sma->sem_nsems;
810 if (copy_semid_to_user (arg.buf, &tbuf, version))
814 /* GETVAL, GETPID, GETNCTN, GETZCNT, SETVAL: fall-through */
817 if(semnum < 0 || semnum >= nsems)
820 curr = &sma->sem_base[semnum];
830 err = count_semncnt(sma,semnum);
833 err = count_semzcnt(sma,semnum);
840 if (val > SEMVMX || val < 0)
843 for (un = sma->undo; un; un = un->id_next)
844 un->semadj[semnum] = 0;
846 curr->sempid = task_tgid_vnr(current);
847 sma->sem_ctime = get_seconds();
848 /* maybe some queued-up processes were waiting for this */
857 if(sem_io != fast_sem_io)
858 ipc_free(sem_io, sizeof(ushort)*nsems);
868 static inline unsigned long copy_semid_from_user(struct sem_setbuf *out, void __user *buf, int version)
873 struct semid64_ds tbuf;
875 if(copy_from_user(&tbuf, buf, sizeof(tbuf)))
878 out->uid = tbuf.sem_perm.uid;
879 out->gid = tbuf.sem_perm.gid;
880 out->mode = tbuf.sem_perm.mode;
886 struct semid_ds tbuf_old;
888 if(copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
891 out->uid = tbuf_old.sem_perm.uid;
892 out->gid = tbuf_old.sem_perm.gid;
893 out->mode = tbuf_old.sem_perm.mode;
902 static int semctl_down(struct ipc_namespace *ns, int semid, int semnum,
903 int cmd, int version, union semun arg)
905 struct sem_array *sma;
907 struct sem_setbuf uninitialized_var(setbuf);
908 struct kern_ipc_perm *ipcp;
911 if(copy_semid_from_user (&setbuf, arg.buf, version))
914 sma = sem_lock_check_down(ns, semid);
918 ipcp = &sma->sem_perm;
920 err = audit_ipc_obj(ipcp);
924 if (cmd == IPC_SET) {
925 err = audit_ipc_set_perm(0, setbuf.uid, setbuf.gid, setbuf.mode);
929 if (current->euid != ipcp->cuid &&
930 current->euid != ipcp->uid && !capable(CAP_SYS_ADMIN)) {
935 err = security_sem_semctl(sma, cmd);
945 ipcp->uid = setbuf.uid;
946 ipcp->gid = setbuf.gid;
947 ipcp->mode = (ipcp->mode & ~S_IRWXUGO)
948 | (setbuf.mode & S_IRWXUGO);
949 sma->sem_ctime = get_seconds();
965 asmlinkage long sys_semctl (int semid, int semnum, int cmd, union semun arg)
969 struct ipc_namespace *ns;
974 version = ipc_parse_version(&cmd);
975 ns = current->nsproxy->ipc_ns;
981 err = semctl_nolock(ns,semid,semnum,cmd,version,arg);
991 err = semctl_main(ns,semid,semnum,cmd,version,arg);
995 down_write(&sem_ids(ns).rw_mutex);
996 err = semctl_down(ns,semid,semnum,cmd,version,arg);
997 up_write(&sem_ids(ns).rw_mutex);
1004 static inline void lock_semundo(void)
1006 struct sem_undo_list *undo_list;
1008 undo_list = current->sysvsem.undo_list;
1010 spin_lock(&undo_list->lock);
1013 /* This code has an interaction with copy_semundo().
1014 * Consider; two tasks are sharing the undo_list. task1
1015 * acquires the undo_list lock in lock_semundo(). If task2 now
1016 * exits before task1 releases the lock (by calling
1017 * unlock_semundo()), then task1 will never call spin_unlock().
1018 * This leave the sem_undo_list in a locked state. If task1 now creats task3
1019 * and once again shares the sem_undo_list, the sem_undo_list will still be
1020 * locked, and future SEM_UNDO operations will deadlock. This case is
1021 * dealt with in copy_semundo() by having it reinitialize the spin lock when
1022 * the refcnt goes from 1 to 2.
1024 static inline void unlock_semundo(void)
1026 struct sem_undo_list *undo_list;
1028 undo_list = current->sysvsem.undo_list;
1030 spin_unlock(&undo_list->lock);
1034 /* If the task doesn't already have a undo_list, then allocate one
1035 * here. We guarantee there is only one thread using this undo list,
1036 * and current is THE ONE
1038 * If this allocation and assignment succeeds, but later
1039 * portions of this code fail, there is no need to free the sem_undo_list.
1040 * Just let it stay associated with the task, and it'll be freed later
1043 * This can block, so callers must hold no locks.
1045 static inline int get_undo_list(struct sem_undo_list **undo_listp)
1047 struct sem_undo_list *undo_list;
1049 undo_list = current->sysvsem.undo_list;
1051 undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
1052 if (undo_list == NULL)
1054 spin_lock_init(&undo_list->lock);
1055 atomic_set(&undo_list->refcnt, 1);
1056 current->sysvsem.undo_list = undo_list;
1058 *undo_listp = undo_list;
1062 static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1064 struct sem_undo **last, *un;
1066 last = &ulp->proc_list;
1069 if(un->semid==semid)
1072 *last=un->proc_next;
1075 last=&un->proc_next;
1082 static struct sem_undo *find_undo(struct ipc_namespace *ns, int semid)
1084 struct sem_array *sma;
1085 struct sem_undo_list *ulp;
1086 struct sem_undo *un, *new;
1090 error = get_undo_list(&ulp);
1092 return ERR_PTR(error);
1095 un = lookup_undo(ulp, semid);
1097 if (likely(un!=NULL))
1100 /* no undo structure around - allocate one. */
1101 sma = sem_lock_check(ns, semid);
1103 return ERR_PTR(PTR_ERR(sma));
1105 nsems = sma->sem_nsems;
1106 ipc_rcu_getref(sma);
1109 new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1111 ipc_lock_by_ptr(&sma->sem_perm);
1112 ipc_rcu_putref(sma);
1114 return ERR_PTR(-ENOMEM);
1116 new->semadj = (short *) &new[1];
1120 un = lookup_undo(ulp, semid);
1124 ipc_lock_by_ptr(&sma->sem_perm);
1125 ipc_rcu_putref(sma);
1129 ipc_lock_by_ptr(&sma->sem_perm);
1130 ipc_rcu_putref(sma);
1131 if (sma->sem_perm.deleted) {
1135 un = ERR_PTR(-EIDRM);
1138 new->proc_next = ulp->proc_list;
1139 ulp->proc_list = new;
1140 new->id_next = sma->undo;
1149 asmlinkage long sys_semtimedop(int semid, struct sembuf __user *tsops,
1150 unsigned nsops, const struct timespec __user *timeout)
1152 int error = -EINVAL;
1153 struct sem_array *sma;
1154 struct sembuf fast_sops[SEMOPM_FAST];
1155 struct sembuf* sops = fast_sops, *sop;
1156 struct sem_undo *un;
1157 int undos = 0, alter = 0, max;
1158 struct sem_queue queue;
1159 unsigned long jiffies_left = 0;
1160 struct ipc_namespace *ns;
1162 ns = current->nsproxy->ipc_ns;
1164 if (nsops < 1 || semid < 0)
1166 if (nsops > ns->sc_semopm)
1168 if(nsops > SEMOPM_FAST) {
1169 sops = kmalloc(sizeof(*sops)*nsops,GFP_KERNEL);
1173 if (copy_from_user (sops, tsops, nsops * sizeof(*tsops))) {
1178 struct timespec _timeout;
1179 if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1183 if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1184 _timeout.tv_nsec >= 1000000000L) {
1188 jiffies_left = timespec_to_jiffies(&_timeout);
1191 for (sop = sops; sop < sops + nsops; sop++) {
1192 if (sop->sem_num >= max)
1194 if (sop->sem_flg & SEM_UNDO)
1196 if (sop->sem_op != 0)
1202 un = find_undo(ns, semid);
1204 error = PTR_ERR(un);
1210 sma = sem_lock_check(ns, semid);
1212 error = PTR_ERR(sma);
1217 * semid identifiers are not unique - find_undo may have
1218 * allocated an undo structure, it was invalidated by an RMID
1219 * and now a new array with received the same id. Check and retry.
1221 if (un && un->semid == -1) {
1226 if (max >= sma->sem_nsems)
1227 goto out_unlock_free;
1230 if (ipcperms(&sma->sem_perm, alter ? S_IWUGO : S_IRUGO))
1231 goto out_unlock_free;
1233 error = security_sem_semop(sma, sops, nsops, alter);
1235 goto out_unlock_free;
1237 error = try_atomic_semop (sma, sops, nsops, un, task_tgid_vnr(current));
1239 if (alter && error == 0)
1241 goto out_unlock_free;
1244 /* We need to sleep on this operation, so we put the current
1245 * task into the pending queue and go to sleep.
1250 queue.nsops = nsops;
1252 queue.pid = task_tgid_vnr(current);
1254 queue.alter = alter;
1256 append_to_queue(sma ,&queue);
1258 prepend_to_queue(sma ,&queue);
1260 queue.status = -EINTR;
1261 queue.sleeper = current;
1262 current->state = TASK_INTERRUPTIBLE;
1266 jiffies_left = schedule_timeout(jiffies_left);
1270 error = queue.status;
1271 while(unlikely(error == IN_WAKEUP)) {
1273 error = queue.status;
1276 if (error != -EINTR) {
1277 /* fast path: update_queue already obtained all requested
1282 sma = sem_lock(ns, semid);
1284 BUG_ON(queue.prev != NULL);
1290 * If queue.status != -EINTR we are woken up by another process
1292 error = queue.status;
1293 if (error != -EINTR) {
1294 goto out_unlock_free;
1298 * If an interrupt occurred we have to clean up the queue
1300 if (timeout && jiffies_left == 0)
1302 remove_from_queue(sma,&queue);
1303 goto out_unlock_free;
1308 if(sops != fast_sops)
1313 asmlinkage long sys_semop (int semid, struct sembuf __user *tsops, unsigned nsops)
1315 return sys_semtimedop(semid, tsops, nsops, NULL);
1318 /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
1319 * parent and child tasks.
1321 * See the notes above unlock_semundo() regarding the spin_lock_init()
1322 * in this code. Initialize the undo_list->lock here instead of get_undo_list()
1323 * because of the reasoning in the comment above unlock_semundo.
1326 int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
1328 struct sem_undo_list *undo_list;
1331 if (clone_flags & CLONE_SYSVSEM) {
1332 error = get_undo_list(&undo_list);
1335 atomic_inc(&undo_list->refcnt);
1336 tsk->sysvsem.undo_list = undo_list;
1338 tsk->sysvsem.undo_list = NULL;
1344 * add semadj values to semaphores, free undo structures.
1345 * undo structures are not freed when semaphore arrays are destroyed
1346 * so some of them may be out of date.
1347 * IMPLEMENTATION NOTE: There is some confusion over whether the
1348 * set of adjustments that needs to be done should be done in an atomic
1349 * manner or not. That is, if we are attempting to decrement the semval
1350 * should we queue up and wait until we can do so legally?
1351 * The original implementation attempted to do this (queue and wait).
1352 * The current implementation does not do so. The POSIX standard
1353 * and SVID should be consulted to determine what behavior is mandated.
1355 void exit_sem(struct task_struct *tsk)
1357 struct sem_undo_list *undo_list;
1358 struct sem_undo *u, **up;
1359 struct ipc_namespace *ns;
1361 undo_list = tsk->sysvsem.undo_list;
1365 if (!atomic_dec_and_test(&undo_list->refcnt))
1368 ns = tsk->nsproxy->ipc_ns;
1369 /* There's no need to hold the semundo list lock, as current
1370 * is the last task exiting for this undo list.
1372 for (up = &undo_list->proc_list; (u = *up); *up = u->proc_next, kfree(u)) {
1373 struct sem_array *sma;
1375 struct sem_undo *un, **unp;
1382 sma = sem_lock(ns, semid);
1389 BUG_ON(sem_checkid(ns,sma,u->semid));
1391 /* remove u from the sma->undo list */
1392 for (unp = &sma->undo; (un = *unp); unp = &un->id_next) {
1396 printk ("exit_sem undo list error id=%d\n", u->semid);
1400 /* perform adjustments registered in u */
1401 nsems = sma->sem_nsems;
1402 for (i = 0; i < nsems; i++) {
1403 struct sem * semaphore = &sma->sem_base[i];
1405 semaphore->semval += u->semadj[i];
1407 * Range checks of the new semaphore value,
1408 * not defined by sus:
1409 * - Some unices ignore the undo entirely
1410 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
1411 * - some cap the value (e.g. FreeBSD caps
1412 * at 0, but doesn't enforce SEMVMX)
1414 * Linux caps the semaphore value, both at 0
1417 * Manfred <manfred@colorfullife.com>
1419 if (semaphore->semval < 0)
1420 semaphore->semval = 0;
1421 if (semaphore->semval > SEMVMX)
1422 semaphore->semval = SEMVMX;
1423 semaphore->sempid = task_tgid_vnr(current);
1426 sma->sem_otime = get_seconds();
1427 /* maybe some queued-up processes were waiting for this */
1435 #ifdef CONFIG_PROC_FS
1436 static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
1438 struct sem_array *sma = it;
1440 return seq_printf(s,
1441 "%10d %10d %4o %10lu %5u %5u %5u %5u %10lu %10lu\n",