Loading...
1/*
2 * linux/ipc/sem.c
3 * Copyright (C) 1992 Krishna Balasubramanian
4 * Copyright (C) 1995 Eric Schenk, Bruno Haible
5 *
6 * /proc/sysvipc/sem support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
7 *
8 * SMP-threaded, sysctl's added
9 * (c) 1999 Manfred Spraul <manfred@colorfullife.com>
10 * Enforced range limit on SEM_UNDO
11 * (c) 2001 Red Hat Inc
12 * Lockless wakeup
13 * (c) 2003 Manfred Spraul <manfred@colorfullife.com>
14 * Further wakeup optimizations, documentation
15 * (c) 2010 Manfred Spraul <manfred@colorfullife.com>
16 *
17 * support for audit of ipc object properties and permission changes
18 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
19 *
20 * namespaces support
21 * OpenVZ, SWsoft Inc.
22 * Pavel Emelianov <xemul@openvz.org>
23 *
24 * Implementation notes: (May 2010)
25 * This file implements System V semaphores.
26 *
27 * User space visible behavior:
28 * - FIFO ordering for semop() operations (just FIFO, not starvation
29 * protection)
30 * - multiple semaphore operations that alter the same semaphore in
31 * one semop() are handled.
32 * - sem_ctime (time of last semctl()) is updated in the IPC_SET, SETVAL and
33 * SETALL calls.
34 * - two Linux specific semctl() commands: SEM_STAT, SEM_INFO.
35 * - undo adjustments at process exit are limited to 0..SEMVMX.
36 * - namespace are supported.
37 * - SEMMSL, SEMMNS, SEMOPM and SEMMNI can be configured at runtine by writing
38 * to /proc/sys/kernel/sem.
39 * - statistics about the usage are reported in /proc/sysvipc/sem.
40 *
41 * Internals:
42 * - scalability:
43 * - all global variables are read-mostly.
44 * - semop() calls and semctl(RMID) are synchronized by RCU.
45 * - most operations do write operations (actually: spin_lock calls) to
46 * the per-semaphore array structure.
47 * Thus: Perfect SMP scaling between independent semaphore arrays.
48 * If multiple semaphores in one array are used, then cache line
49 * trashing on the semaphore array spinlock will limit the scaling.
50 * - semncnt and semzcnt are calculated on demand in count_semncnt() and
51 * count_semzcnt()
52 * - the task that performs a successful semop() scans the list of all
53 * sleeping tasks and completes any pending operations that can be fulfilled.
54 * Semaphores are actively given to waiting tasks (necessary for FIFO).
55 * (see update_queue())
56 * - To improve the scalability, the actual wake-up calls are performed after
57 * dropping all locks. (see wake_up_sem_queue_prepare(),
58 * wake_up_sem_queue_do())
59 * - All work is done by the waker, the woken up task does not have to do
60 * anything - not even acquiring a lock or dropping a refcount.
61 * - A woken up task may not even touch the semaphore array anymore, it may
62 * have been destroyed already by a semctl(RMID).
63 * - The synchronizations between wake-ups due to a timeout/signal and a
64 * wake-up due to a completed semaphore operation is achieved by using an
65 * intermediate state (IN_WAKEUP).
66 * - UNDO values are stored in an array (one per process and per
67 * semaphore array, lazily allocated). For backwards compatibility, multiple
68 * modes for the UNDO variables are supported (per process, per thread)
69 * (see copy_semundo, CLONE_SYSVSEM)
70 * - There are two lists of the pending operations: a per-array list
71 * and per-semaphore list (stored in the array). This allows to achieve FIFO
72 * ordering without always scanning all pending operations.
73 * The worst-case behavior is nevertheless O(N^2) for N wakeups.
74 */
75
76#include <linux/slab.h>
77#include <linux/spinlock.h>
78#include <linux/init.h>
79#include <linux/proc_fs.h>
80#include <linux/time.h>
81#include <linux/security.h>
82#include <linux/syscalls.h>
83#include <linux/audit.h>
84#include <linux/capability.h>
85#include <linux/seq_file.h>
86#include <linux/rwsem.h>
87#include <linux/nsproxy.h>
88#include <linux/ipc_namespace.h>
89
90#include <asm/uaccess.h>
91#include "util.h"
92
93#define sem_ids(ns) ((ns)->ids[IPC_SEM_IDS])
94
95#define sem_unlock(sma) ipc_unlock(&(sma)->sem_perm)
96#define sem_checkid(sma, semid) ipc_checkid(&sma->sem_perm, semid)
97
98static int newary(struct ipc_namespace *, struct ipc_params *);
99static void freeary(struct ipc_namespace *, struct kern_ipc_perm *);
100#ifdef CONFIG_PROC_FS
101static int sysvipc_sem_proc_show(struct seq_file *s, void *it);
102#endif
103
104#define SEMMSL_FAST 256 /* 512 bytes on stack */
105#define SEMOPM_FAST 64 /* ~ 372 bytes on stack */
106
107/*
108 * linked list protection:
109 * sem_undo.id_next,
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.
113 *
114 */
115
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]
120
121void sem_init_ns(struct ipc_namespace *ns)
122{
123 ns->sc_semmsl = SEMMSL;
124 ns->sc_semmns = SEMMNS;
125 ns->sc_semopm = SEMOPM;
126 ns->sc_semmni = SEMMNI;
127 ns->used_sems = 0;
128 ipc_init_ids(&ns->ids[IPC_SEM_IDS]);
129}
130
131#ifdef CONFIG_IPC_NS
132void sem_exit_ns(struct ipc_namespace *ns)
133{
134 free_ipcs(ns, &sem_ids(ns), freeary);
135 idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr);
136}
137#endif
138
139void __init sem_init (void)
140{
141 sem_init_ns(&init_ipc_ns);
142 ipc_init_proc_interface("sysvipc/sem",
143 " key semid perms nsems uid gid cuid cgid otime ctime\n",
144 IPC_SEM_IDS, sysvipc_sem_proc_show);
145}
146
147/*
148 * sem_lock_(check_) routines are called in the paths where the rw_mutex
149 * is not held.
150 */
151static inline struct sem_array *sem_lock(struct ipc_namespace *ns, int id)
152{
153 struct kern_ipc_perm *ipcp = ipc_lock(&sem_ids(ns), id);
154
155 if (IS_ERR(ipcp))
156 return (struct sem_array *)ipcp;
157
158 return container_of(ipcp, struct sem_array, sem_perm);
159}
160
161static inline struct sem_array *sem_lock_check(struct ipc_namespace *ns,
162 int id)
163{
164 struct kern_ipc_perm *ipcp = ipc_lock_check(&sem_ids(ns), id);
165
166 if (IS_ERR(ipcp))
167 return (struct sem_array *)ipcp;
168
169 return container_of(ipcp, struct sem_array, sem_perm);
170}
171
172static inline void sem_lock_and_putref(struct sem_array *sma)
173{
174 ipc_lock_by_ptr(&sma->sem_perm);
175 ipc_rcu_putref(sma);
176}
177
178static inline void sem_getref_and_unlock(struct sem_array *sma)
179{
180 ipc_rcu_getref(sma);
181 ipc_unlock(&(sma)->sem_perm);
182}
183
184static inline void sem_putref(struct sem_array *sma)
185{
186 ipc_lock_by_ptr(&sma->sem_perm);
187 ipc_rcu_putref(sma);
188 ipc_unlock(&(sma)->sem_perm);
189}
190
191static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
192{
193 ipc_rmid(&sem_ids(ns), &s->sem_perm);
194}
195
196/*
197 * Lockless wakeup algorithm:
198 * Without the check/retry algorithm a lockless wakeup is possible:
199 * - queue.status is initialized to -EINTR before blocking.
200 * - wakeup is performed by
201 * * unlinking the queue entry from sma->sem_pending
202 * * setting queue.status to IN_WAKEUP
203 * This is the notification for the blocked thread that a
204 * result value is imminent.
205 * * call wake_up_process
206 * * set queue.status to the final value.
207 * - the previously blocked thread checks queue.status:
208 * * if it's IN_WAKEUP, then it must wait until the value changes
209 * * if it's not -EINTR, then the operation was completed by
210 * update_queue. semtimedop can return queue.status without
211 * performing any operation on the sem array.
212 * * otherwise it must acquire the spinlock and check what's up.
213 *
214 * The two-stage algorithm is necessary to protect against the following
215 * races:
216 * - if queue.status is set after wake_up_process, then the woken up idle
217 * thread could race forward and try (and fail) to acquire sma->lock
218 * before update_queue had a chance to set queue.status
219 * - if queue.status is written before wake_up_process and if the
220 * blocked process is woken up by a signal between writing
221 * queue.status and the wake_up_process, then the woken up
222 * process could return from semtimedop and die by calling
223 * sys_exit before wake_up_process is called. Then wake_up_process
224 * will oops, because the task structure is already invalid.
225 * (yes, this happened on s390 with sysv msg).
226 *
227 */
228#define IN_WAKEUP 1
229
230/**
231 * newary - Create a new semaphore set
232 * @ns: namespace
233 * @params: ptr to the structure that contains key, semflg and nsems
234 *
235 * Called with sem_ids.rw_mutex held (as a writer)
236 */
237
238static int newary(struct ipc_namespace *ns, struct ipc_params *params)
239{
240 int id;
241 int retval;
242 struct sem_array *sma;
243 int size;
244 key_t key = params->key;
245 int nsems = params->u.nsems;
246 int semflg = params->flg;
247 int i;
248
249 if (!nsems)
250 return -EINVAL;
251 if (ns->used_sems + nsems > ns->sc_semmns)
252 return -ENOSPC;
253
254 size = sizeof (*sma) + nsems * sizeof (struct sem);
255 sma = ipc_rcu_alloc(size);
256 if (!sma) {
257 return -ENOMEM;
258 }
259 memset (sma, 0, size);
260
261 sma->sem_perm.mode = (semflg & S_IRWXUGO);
262 sma->sem_perm.key = key;
263
264 sma->sem_perm.security = NULL;
265 retval = security_sem_alloc(sma);
266 if (retval) {
267 ipc_rcu_putref(sma);
268 return retval;
269 }
270
271 id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
272 if (id < 0) {
273 security_sem_free(sma);
274 ipc_rcu_putref(sma);
275 return id;
276 }
277 ns->used_sems += nsems;
278
279 sma->sem_base = (struct sem *) &sma[1];
280
281 for (i = 0; i < nsems; i++)
282 INIT_LIST_HEAD(&sma->sem_base[i].sem_pending);
283
284 sma->complex_count = 0;
285 INIT_LIST_HEAD(&sma->sem_pending);
286 INIT_LIST_HEAD(&sma->list_id);
287 sma->sem_nsems = nsems;
288 sma->sem_ctime = get_seconds();
289 sem_unlock(sma);
290
291 return sma->sem_perm.id;
292}
293
294
295/*
296 * Called with sem_ids.rw_mutex and ipcp locked.
297 */
298static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg)
299{
300 struct sem_array *sma;
301
302 sma = container_of(ipcp, struct sem_array, sem_perm);
303 return security_sem_associate(sma, semflg);
304}
305
306/*
307 * Called with sem_ids.rw_mutex and ipcp locked.
308 */
309static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
310 struct ipc_params *params)
311{
312 struct sem_array *sma;
313
314 sma = container_of(ipcp, struct sem_array, sem_perm);
315 if (params->u.nsems > sma->sem_nsems)
316 return -EINVAL;
317
318 return 0;
319}
320
321SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg)
322{
323 struct ipc_namespace *ns;
324 struct ipc_ops sem_ops;
325 struct ipc_params sem_params;
326
327 ns = current->nsproxy->ipc_ns;
328
329 if (nsems < 0 || nsems > ns->sc_semmsl)
330 return -EINVAL;
331
332 sem_ops.getnew = newary;
333 sem_ops.associate = sem_security;
334 sem_ops.more_checks = sem_more_checks;
335
336 sem_params.key = key;
337 sem_params.flg = semflg;
338 sem_params.u.nsems = nsems;
339
340 return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
341}
342
343/*
344 * Determine whether a sequence of semaphore operations would succeed
345 * all at once. Return 0 if yes, 1 if need to sleep, else return error code.
346 */
347
348static int try_atomic_semop (struct sem_array * sma, struct sembuf * sops,
349 int nsops, struct sem_undo *un, int pid)
350{
351 int result, sem_op;
352 struct sembuf *sop;
353 struct sem * curr;
354
355 for (sop = sops; sop < sops + nsops; sop++) {
356 curr = sma->sem_base + sop->sem_num;
357 sem_op = sop->sem_op;
358 result = curr->semval;
359
360 if (!sem_op && result)
361 goto would_block;
362
363 result += sem_op;
364 if (result < 0)
365 goto would_block;
366 if (result > SEMVMX)
367 goto out_of_range;
368 if (sop->sem_flg & SEM_UNDO) {
369 int undo = un->semadj[sop->sem_num] - sem_op;
370 /*
371 * Exceeding the undo range is an error.
372 */
373 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
374 goto out_of_range;
375 }
376 curr->semval = result;
377 }
378
379 sop--;
380 while (sop >= sops) {
381 sma->sem_base[sop->sem_num].sempid = pid;
382 if (sop->sem_flg & SEM_UNDO)
383 un->semadj[sop->sem_num] -= sop->sem_op;
384 sop--;
385 }
386
387 return 0;
388
389out_of_range:
390 result = -ERANGE;
391 goto undo;
392
393would_block:
394 if (sop->sem_flg & IPC_NOWAIT)
395 result = -EAGAIN;
396 else
397 result = 1;
398
399undo:
400 sop--;
401 while (sop >= sops) {
402 sma->sem_base[sop->sem_num].semval -= sop->sem_op;
403 sop--;
404 }
405
406 return result;
407}
408
409/** wake_up_sem_queue_prepare(q, error): Prepare wake-up
410 * @q: queue entry that must be signaled
411 * @error: Error value for the signal
412 *
413 * Prepare the wake-up of the queue entry q.
414 */
415static void wake_up_sem_queue_prepare(struct list_head *pt,
416 struct sem_queue *q, int error)
417{
418 if (list_empty(pt)) {
419 /*
420 * Hold preempt off so that we don't get preempted and have the
421 * wakee busy-wait until we're scheduled back on.
422 */
423 preempt_disable();
424 }
425 q->status = IN_WAKEUP;
426 q->pid = error;
427
428 list_add_tail(&q->simple_list, pt);
429}
430
431/**
432 * wake_up_sem_queue_do(pt) - do the actual wake-up
433 * @pt: list of tasks to be woken up
434 *
435 * Do the actual wake-up.
436 * The function is called without any locks held, thus the semaphore array
437 * could be destroyed already and the tasks can disappear as soon as the
438 * status is set to the actual return code.
439 */
440static void wake_up_sem_queue_do(struct list_head *pt)
441{
442 struct sem_queue *q, *t;
443 int did_something;
444
445 did_something = !list_empty(pt);
446 list_for_each_entry_safe(q, t, pt, simple_list) {
447 wake_up_process(q->sleeper);
448 /* q can disappear immediately after writing q->status. */
449 smp_wmb();
450 q->status = q->pid;
451 }
452 if (did_something)
453 preempt_enable();
454}
455
456static void unlink_queue(struct sem_array *sma, struct sem_queue *q)
457{
458 list_del(&q->list);
459 if (q->nsops == 1)
460 list_del(&q->simple_list);
461 else
462 sma->complex_count--;
463}
464
465/** check_restart(sma, q)
466 * @sma: semaphore array
467 * @q: the operation that just completed
468 *
469 * update_queue is O(N^2) when it restarts scanning the whole queue of
470 * waiting operations. Therefore this function checks if the restart is
471 * really necessary. It is called after a previously waiting operation
472 * was completed.
473 */
474static int check_restart(struct sem_array *sma, struct sem_queue *q)
475{
476 struct sem *curr;
477 struct sem_queue *h;
478
479 /* if the operation didn't modify the array, then no restart */
480 if (q->alter == 0)
481 return 0;
482
483 /* pending complex operations are too difficult to analyse */
484 if (sma->complex_count)
485 return 1;
486
487 /* we were a sleeping complex operation. Too difficult */
488 if (q->nsops > 1)
489 return 1;
490
491 curr = sma->sem_base + q->sops[0].sem_num;
492
493 /* No-one waits on this queue */
494 if (list_empty(&curr->sem_pending))
495 return 0;
496
497 /* the new semaphore value */
498 if (curr->semval) {
499 /* It is impossible that someone waits for the new value:
500 * - q is a previously sleeping simple operation that
501 * altered the array. It must be a decrement, because
502 * simple increments never sleep.
503 * - The value is not 0, thus wait-for-zero won't proceed.
504 * - If there are older (higher priority) decrements
505 * in the queue, then they have observed the original
506 * semval value and couldn't proceed. The operation
507 * decremented to value - thus they won't proceed either.
508 */
509 BUG_ON(q->sops[0].sem_op >= 0);
510 return 0;
511 }
512 /*
513 * semval is 0. Check if there are wait-for-zero semops.
514 * They must be the first entries in the per-semaphore simple queue
515 */
516 h = list_first_entry(&curr->sem_pending, struct sem_queue, simple_list);
517 BUG_ON(h->nsops != 1);
518 BUG_ON(h->sops[0].sem_num != q->sops[0].sem_num);
519
520 /* Yes, there is a wait-for-zero semop. Restart */
521 if (h->sops[0].sem_op == 0)
522 return 1;
523
524 /* Again - no-one is waiting for the new value. */
525 return 0;
526}
527
528
529/**
530 * update_queue(sma, semnum): Look for tasks that can be completed.
531 * @sma: semaphore array.
532 * @semnum: semaphore that was modified.
533 * @pt: list head for the tasks that must be woken up.
534 *
535 * update_queue must be called after a semaphore in a semaphore array
536 * was modified. If multiple semaphore were modified, then @semnum
537 * must be set to -1.
538 * The tasks that must be woken up are added to @pt. The return code
539 * is stored in q->pid.
540 * The function return 1 if at least one semop was completed successfully.
541 */
542static int update_queue(struct sem_array *sma, int semnum, struct list_head *pt)
543{
544 struct sem_queue *q;
545 struct list_head *walk;
546 struct list_head *pending_list;
547 int offset;
548 int semop_completed = 0;
549
550 /* if there are complex operations around, then knowing the semaphore
551 * that was modified doesn't help us. Assume that multiple semaphores
552 * were modified.
553 */
554 if (sma->complex_count)
555 semnum = -1;
556
557 if (semnum == -1) {
558 pending_list = &sma->sem_pending;
559 offset = offsetof(struct sem_queue, list);
560 } else {
561 pending_list = &sma->sem_base[semnum].sem_pending;
562 offset = offsetof(struct sem_queue, simple_list);
563 }
564
565again:
566 walk = pending_list->next;
567 while (walk != pending_list) {
568 int error, restart;
569
570 q = (struct sem_queue *)((char *)walk - offset);
571 walk = walk->next;
572
573 /* If we are scanning the single sop, per-semaphore list of
574 * one semaphore and that semaphore is 0, then it is not
575 * necessary to scan the "alter" entries: simple increments
576 * that affect only one entry succeed immediately and cannot
577 * be in the per semaphore pending queue, and decrements
578 * cannot be successful if the value is already 0.
579 */
580 if (semnum != -1 && sma->sem_base[semnum].semval == 0 &&
581 q->alter)
582 break;
583
584 error = try_atomic_semop(sma, q->sops, q->nsops,
585 q->undo, q->pid);
586
587 /* Does q->sleeper still need to sleep? */
588 if (error > 0)
589 continue;
590
591 unlink_queue(sma, q);
592
593 if (error) {
594 restart = 0;
595 } else {
596 semop_completed = 1;
597 restart = check_restart(sma, q);
598 }
599
600 wake_up_sem_queue_prepare(pt, q, error);
601 if (restart)
602 goto again;
603 }
604 return semop_completed;
605}
606
607/**
608 * do_smart_update(sma, sops, nsops, otime, pt) - optimized update_queue
609 * @sma: semaphore array
610 * @sops: operations that were performed
611 * @nsops: number of operations
612 * @otime: force setting otime
613 * @pt: list head of the tasks that must be woken up.
614 *
615 * do_smart_update() does the required called to update_queue, based on the
616 * actual changes that were performed on the semaphore array.
617 * Note that the function does not do the actual wake-up: the caller is
618 * responsible for calling wake_up_sem_queue_do(@pt).
619 * It is safe to perform this call after dropping all locks.
620 */
621static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops,
622 int otime, struct list_head *pt)
623{
624 int i;
625
626 if (sma->complex_count || sops == NULL) {
627 if (update_queue(sma, -1, pt))
628 otime = 1;
629 goto done;
630 }
631
632 for (i = 0; i < nsops; i++) {
633 if (sops[i].sem_op > 0 ||
634 (sops[i].sem_op < 0 &&
635 sma->sem_base[sops[i].sem_num].semval == 0))
636 if (update_queue(sma, sops[i].sem_num, pt))
637 otime = 1;
638 }
639done:
640 if (otime)
641 sma->sem_otime = get_seconds();
642}
643
644
645/* The following counts are associated to each semaphore:
646 * semncnt number of tasks waiting on semval being nonzero
647 * semzcnt number of tasks waiting on semval being zero
648 * This model assumes that a task waits on exactly one semaphore.
649 * Since semaphore operations are to be performed atomically, tasks actually
650 * wait on a whole sequence of semaphores simultaneously.
651 * The counts we return here are a rough approximation, but still
652 * warrant that semncnt+semzcnt>0 if the task is on the pending queue.
653 */
654static int count_semncnt (struct sem_array * sma, ushort semnum)
655{
656 int semncnt;
657 struct sem_queue * q;
658
659 semncnt = 0;
660 list_for_each_entry(q, &sma->sem_pending, list) {
661 struct sembuf * sops = q->sops;
662 int nsops = q->nsops;
663 int i;
664 for (i = 0; i < nsops; i++)
665 if (sops[i].sem_num == semnum
666 && (sops[i].sem_op < 0)
667 && !(sops[i].sem_flg & IPC_NOWAIT))
668 semncnt++;
669 }
670 return semncnt;
671}
672
673static int count_semzcnt (struct sem_array * sma, ushort semnum)
674{
675 int semzcnt;
676 struct sem_queue * q;
677
678 semzcnt = 0;
679 list_for_each_entry(q, &sma->sem_pending, list) {
680 struct sembuf * sops = q->sops;
681 int nsops = q->nsops;
682 int i;
683 for (i = 0; i < nsops; i++)
684 if (sops[i].sem_num == semnum
685 && (sops[i].sem_op == 0)
686 && !(sops[i].sem_flg & IPC_NOWAIT))
687 semzcnt++;
688 }
689 return semzcnt;
690}
691
692/* Free a semaphore set. freeary() is called with sem_ids.rw_mutex locked
693 * as a writer and the spinlock for this semaphore set hold. sem_ids.rw_mutex
694 * remains locked on exit.
695 */
696static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
697{
698 struct sem_undo *un, *tu;
699 struct sem_queue *q, *tq;
700 struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
701 struct list_head tasks;
702
703 /* Free the existing undo structures for this semaphore set. */
704 assert_spin_locked(&sma->sem_perm.lock);
705 list_for_each_entry_safe(un, tu, &sma->list_id, list_id) {
706 list_del(&un->list_id);
707 spin_lock(&un->ulp->lock);
708 un->semid = -1;
709 list_del_rcu(&un->list_proc);
710 spin_unlock(&un->ulp->lock);
711 kfree_rcu(un, rcu);
712 }
713
714 /* Wake up all pending processes and let them fail with EIDRM. */
715 INIT_LIST_HEAD(&tasks);
716 list_for_each_entry_safe(q, tq, &sma->sem_pending, list) {
717 unlink_queue(sma, q);
718 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
719 }
720
721 /* Remove the semaphore set from the IDR */
722 sem_rmid(ns, sma);
723 sem_unlock(sma);
724
725 wake_up_sem_queue_do(&tasks);
726 ns->used_sems -= sma->sem_nsems;
727 security_sem_free(sma);
728 ipc_rcu_putref(sma);
729}
730
731static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
732{
733 switch(version) {
734 case IPC_64:
735 return copy_to_user(buf, in, sizeof(*in));
736 case IPC_OLD:
737 {
738 struct semid_ds out;
739
740 memset(&out, 0, sizeof(out));
741
742 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
743
744 out.sem_otime = in->sem_otime;
745 out.sem_ctime = in->sem_ctime;
746 out.sem_nsems = in->sem_nsems;
747
748 return copy_to_user(buf, &out, sizeof(out));
749 }
750 default:
751 return -EINVAL;
752 }
753}
754
755static int semctl_nolock(struct ipc_namespace *ns, int semid,
756 int cmd, int version, union semun arg)
757{
758 int err;
759 struct sem_array *sma;
760
761 switch(cmd) {
762 case IPC_INFO:
763 case SEM_INFO:
764 {
765 struct seminfo seminfo;
766 int max_id;
767
768 err = security_sem_semctl(NULL, cmd);
769 if (err)
770 return err;
771
772 memset(&seminfo,0,sizeof(seminfo));
773 seminfo.semmni = ns->sc_semmni;
774 seminfo.semmns = ns->sc_semmns;
775 seminfo.semmsl = ns->sc_semmsl;
776 seminfo.semopm = ns->sc_semopm;
777 seminfo.semvmx = SEMVMX;
778 seminfo.semmnu = SEMMNU;
779 seminfo.semmap = SEMMAP;
780 seminfo.semume = SEMUME;
781 down_read(&sem_ids(ns).rw_mutex);
782 if (cmd == SEM_INFO) {
783 seminfo.semusz = sem_ids(ns).in_use;
784 seminfo.semaem = ns->used_sems;
785 } else {
786 seminfo.semusz = SEMUSZ;
787 seminfo.semaem = SEMAEM;
788 }
789 max_id = ipc_get_maxid(&sem_ids(ns));
790 up_read(&sem_ids(ns).rw_mutex);
791 if (copy_to_user (arg.__buf, &seminfo, sizeof(struct seminfo)))
792 return -EFAULT;
793 return (max_id < 0) ? 0: max_id;
794 }
795 case IPC_STAT:
796 case SEM_STAT:
797 {
798 struct semid64_ds tbuf;
799 int id;
800
801 if (cmd == SEM_STAT) {
802 sma = sem_lock(ns, semid);
803 if (IS_ERR(sma))
804 return PTR_ERR(sma);
805 id = sma->sem_perm.id;
806 } else {
807 sma = sem_lock_check(ns, semid);
808 if (IS_ERR(sma))
809 return PTR_ERR(sma);
810 id = 0;
811 }
812
813 err = -EACCES;
814 if (ipcperms(ns, &sma->sem_perm, S_IRUGO))
815 goto out_unlock;
816
817 err = security_sem_semctl(sma, cmd);
818 if (err)
819 goto out_unlock;
820
821 memset(&tbuf, 0, sizeof(tbuf));
822
823 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
824 tbuf.sem_otime = sma->sem_otime;
825 tbuf.sem_ctime = sma->sem_ctime;
826 tbuf.sem_nsems = sma->sem_nsems;
827 sem_unlock(sma);
828 if (copy_semid_to_user (arg.buf, &tbuf, version))
829 return -EFAULT;
830 return id;
831 }
832 default:
833 return -EINVAL;
834 }
835out_unlock:
836 sem_unlock(sma);
837 return err;
838}
839
840static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
841 int cmd, int version, union semun arg)
842{
843 struct sem_array *sma;
844 struct sem* curr;
845 int err;
846 ushort fast_sem_io[SEMMSL_FAST];
847 ushort* sem_io = fast_sem_io;
848 int nsems;
849 struct list_head tasks;
850
851 sma = sem_lock_check(ns, semid);
852 if (IS_ERR(sma))
853 return PTR_ERR(sma);
854
855 INIT_LIST_HEAD(&tasks);
856 nsems = sma->sem_nsems;
857
858 err = -EACCES;
859 if (ipcperms(ns, &sma->sem_perm,
860 (cmd == SETVAL || cmd == SETALL) ? S_IWUGO : S_IRUGO))
861 goto out_unlock;
862
863 err = security_sem_semctl(sma, cmd);
864 if (err)
865 goto out_unlock;
866
867 err = -EACCES;
868 switch (cmd) {
869 case GETALL:
870 {
871 ushort __user *array = arg.array;
872 int i;
873
874 if(nsems > SEMMSL_FAST) {
875 sem_getref_and_unlock(sma);
876
877 sem_io = ipc_alloc(sizeof(ushort)*nsems);
878 if(sem_io == NULL) {
879 sem_putref(sma);
880 return -ENOMEM;
881 }
882
883 sem_lock_and_putref(sma);
884 if (sma->sem_perm.deleted) {
885 sem_unlock(sma);
886 err = -EIDRM;
887 goto out_free;
888 }
889 }
890
891 for (i = 0; i < sma->sem_nsems; i++)
892 sem_io[i] = sma->sem_base[i].semval;
893 sem_unlock(sma);
894 err = 0;
895 if(copy_to_user(array, sem_io, nsems*sizeof(ushort)))
896 err = -EFAULT;
897 goto out_free;
898 }
899 case SETALL:
900 {
901 int i;
902 struct sem_undo *un;
903
904 sem_getref_and_unlock(sma);
905
906 if(nsems > SEMMSL_FAST) {
907 sem_io = ipc_alloc(sizeof(ushort)*nsems);
908 if(sem_io == NULL) {
909 sem_putref(sma);
910 return -ENOMEM;
911 }
912 }
913
914 if (copy_from_user (sem_io, arg.array, nsems*sizeof(ushort))) {
915 sem_putref(sma);
916 err = -EFAULT;
917 goto out_free;
918 }
919
920 for (i = 0; i < nsems; i++) {
921 if (sem_io[i] > SEMVMX) {
922 sem_putref(sma);
923 err = -ERANGE;
924 goto out_free;
925 }
926 }
927 sem_lock_and_putref(sma);
928 if (sma->sem_perm.deleted) {
929 sem_unlock(sma);
930 err = -EIDRM;
931 goto out_free;
932 }
933
934 for (i = 0; i < nsems; i++)
935 sma->sem_base[i].semval = sem_io[i];
936
937 assert_spin_locked(&sma->sem_perm.lock);
938 list_for_each_entry(un, &sma->list_id, list_id) {
939 for (i = 0; i < nsems; i++)
940 un->semadj[i] = 0;
941 }
942 sma->sem_ctime = get_seconds();
943 /* maybe some queued-up processes were waiting for this */
944 do_smart_update(sma, NULL, 0, 0, &tasks);
945 err = 0;
946 goto out_unlock;
947 }
948 /* GETVAL, GETPID, GETNCTN, GETZCNT, SETVAL: fall-through */
949 }
950 err = -EINVAL;
951 if(semnum < 0 || semnum >= nsems)
952 goto out_unlock;
953
954 curr = &sma->sem_base[semnum];
955
956 switch (cmd) {
957 case GETVAL:
958 err = curr->semval;
959 goto out_unlock;
960 case GETPID:
961 err = curr->sempid;
962 goto out_unlock;
963 case GETNCNT:
964 err = count_semncnt(sma,semnum);
965 goto out_unlock;
966 case GETZCNT:
967 err = count_semzcnt(sma,semnum);
968 goto out_unlock;
969 case SETVAL:
970 {
971 int val = arg.val;
972 struct sem_undo *un;
973
974 err = -ERANGE;
975 if (val > SEMVMX || val < 0)
976 goto out_unlock;
977
978 assert_spin_locked(&sma->sem_perm.lock);
979 list_for_each_entry(un, &sma->list_id, list_id)
980 un->semadj[semnum] = 0;
981
982 curr->semval = val;
983 curr->sempid = task_tgid_vnr(current);
984 sma->sem_ctime = get_seconds();
985 /* maybe some queued-up processes were waiting for this */
986 do_smart_update(sma, NULL, 0, 0, &tasks);
987 err = 0;
988 goto out_unlock;
989 }
990 }
991out_unlock:
992 sem_unlock(sma);
993 wake_up_sem_queue_do(&tasks);
994
995out_free:
996 if(sem_io != fast_sem_io)
997 ipc_free(sem_io, sizeof(ushort)*nsems);
998 return err;
999}
1000
1001static inline unsigned long
1002copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
1003{
1004 switch(version) {
1005 case IPC_64:
1006 if (copy_from_user(out, buf, sizeof(*out)))
1007 return -EFAULT;
1008 return 0;
1009 case IPC_OLD:
1010 {
1011 struct semid_ds tbuf_old;
1012
1013 if(copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
1014 return -EFAULT;
1015
1016 out->sem_perm.uid = tbuf_old.sem_perm.uid;
1017 out->sem_perm.gid = tbuf_old.sem_perm.gid;
1018 out->sem_perm.mode = tbuf_old.sem_perm.mode;
1019
1020 return 0;
1021 }
1022 default:
1023 return -EINVAL;
1024 }
1025}
1026
1027/*
1028 * This function handles some semctl commands which require the rw_mutex
1029 * to be held in write mode.
1030 * NOTE: no locks must be held, the rw_mutex is taken inside this function.
1031 */
1032static int semctl_down(struct ipc_namespace *ns, int semid,
1033 int cmd, int version, union semun arg)
1034{
1035 struct sem_array *sma;
1036 int err;
1037 struct semid64_ds semid64;
1038 struct kern_ipc_perm *ipcp;
1039
1040 if(cmd == IPC_SET) {
1041 if (copy_semid_from_user(&semid64, arg.buf, version))
1042 return -EFAULT;
1043 }
1044
1045 ipcp = ipcctl_pre_down(ns, &sem_ids(ns), semid, cmd,
1046 &semid64.sem_perm, 0);
1047 if (IS_ERR(ipcp))
1048 return PTR_ERR(ipcp);
1049
1050 sma = container_of(ipcp, struct sem_array, sem_perm);
1051
1052 err = security_sem_semctl(sma, cmd);
1053 if (err)
1054 goto out_unlock;
1055
1056 switch(cmd){
1057 case IPC_RMID:
1058 freeary(ns, ipcp);
1059 goto out_up;
1060 case IPC_SET:
1061 ipc_update_perm(&semid64.sem_perm, ipcp);
1062 sma->sem_ctime = get_seconds();
1063 break;
1064 default:
1065 err = -EINVAL;
1066 }
1067
1068out_unlock:
1069 sem_unlock(sma);
1070out_up:
1071 up_write(&sem_ids(ns).rw_mutex);
1072 return err;
1073}
1074
1075SYSCALL_DEFINE(semctl)(int semid, int semnum, int cmd, union semun arg)
1076{
1077 int err = -EINVAL;
1078 int version;
1079 struct ipc_namespace *ns;
1080
1081 if (semid < 0)
1082 return -EINVAL;
1083
1084 version = ipc_parse_version(&cmd);
1085 ns = current->nsproxy->ipc_ns;
1086
1087 switch(cmd) {
1088 case IPC_INFO:
1089 case SEM_INFO:
1090 case IPC_STAT:
1091 case SEM_STAT:
1092 err = semctl_nolock(ns, semid, cmd, version, arg);
1093 return err;
1094 case GETALL:
1095 case GETVAL:
1096 case GETPID:
1097 case GETNCNT:
1098 case GETZCNT:
1099 case SETVAL:
1100 case SETALL:
1101 err = semctl_main(ns,semid,semnum,cmd,version,arg);
1102 return err;
1103 case IPC_RMID:
1104 case IPC_SET:
1105 err = semctl_down(ns, semid, cmd, version, arg);
1106 return err;
1107 default:
1108 return -EINVAL;
1109 }
1110}
1111#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
1112asmlinkage long SyS_semctl(int semid, int semnum, int cmd, union semun arg)
1113{
1114 return SYSC_semctl((int) semid, (int) semnum, (int) cmd, arg);
1115}
1116SYSCALL_ALIAS(sys_semctl, SyS_semctl);
1117#endif
1118
1119/* If the task doesn't already have a undo_list, then allocate one
1120 * here. We guarantee there is only one thread using this undo list,
1121 * and current is THE ONE
1122 *
1123 * If this allocation and assignment succeeds, but later
1124 * portions of this code fail, there is no need to free the sem_undo_list.
1125 * Just let it stay associated with the task, and it'll be freed later
1126 * at exit time.
1127 *
1128 * This can block, so callers must hold no locks.
1129 */
1130static inline int get_undo_list(struct sem_undo_list **undo_listp)
1131{
1132 struct sem_undo_list *undo_list;
1133
1134 undo_list = current->sysvsem.undo_list;
1135 if (!undo_list) {
1136 undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
1137 if (undo_list == NULL)
1138 return -ENOMEM;
1139 spin_lock_init(&undo_list->lock);
1140 atomic_set(&undo_list->refcnt, 1);
1141 INIT_LIST_HEAD(&undo_list->list_proc);
1142
1143 current->sysvsem.undo_list = undo_list;
1144 }
1145 *undo_listp = undo_list;
1146 return 0;
1147}
1148
1149static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid)
1150{
1151 struct sem_undo *un;
1152
1153 list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) {
1154 if (un->semid == semid)
1155 return un;
1156 }
1157 return NULL;
1158}
1159
1160static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1161{
1162 struct sem_undo *un;
1163
1164 assert_spin_locked(&ulp->lock);
1165
1166 un = __lookup_undo(ulp, semid);
1167 if (un) {
1168 list_del_rcu(&un->list_proc);
1169 list_add_rcu(&un->list_proc, &ulp->list_proc);
1170 }
1171 return un;
1172}
1173
1174/**
1175 * find_alloc_undo - Lookup (and if not present create) undo array
1176 * @ns: namespace
1177 * @semid: semaphore array id
1178 *
1179 * The function looks up (and if not present creates) the undo structure.
1180 * The size of the undo structure depends on the size of the semaphore
1181 * array, thus the alloc path is not that straightforward.
1182 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1183 * performs a rcu_read_lock().
1184 */
1185static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
1186{
1187 struct sem_array *sma;
1188 struct sem_undo_list *ulp;
1189 struct sem_undo *un, *new;
1190 int nsems;
1191 int error;
1192
1193 error = get_undo_list(&ulp);
1194 if (error)
1195 return ERR_PTR(error);
1196
1197 rcu_read_lock();
1198 spin_lock(&ulp->lock);
1199 un = lookup_undo(ulp, semid);
1200 spin_unlock(&ulp->lock);
1201 if (likely(un!=NULL))
1202 goto out;
1203 rcu_read_unlock();
1204
1205 /* no undo structure around - allocate one. */
1206 /* step 1: figure out the size of the semaphore array */
1207 sma = sem_lock_check(ns, semid);
1208 if (IS_ERR(sma))
1209 return ERR_CAST(sma);
1210
1211 nsems = sma->sem_nsems;
1212 sem_getref_and_unlock(sma);
1213
1214 /* step 2: allocate new undo structure */
1215 new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1216 if (!new) {
1217 sem_putref(sma);
1218 return ERR_PTR(-ENOMEM);
1219 }
1220
1221 /* step 3: Acquire the lock on semaphore array */
1222 sem_lock_and_putref(sma);
1223 if (sma->sem_perm.deleted) {
1224 sem_unlock(sma);
1225 kfree(new);
1226 un = ERR_PTR(-EIDRM);
1227 goto out;
1228 }
1229 spin_lock(&ulp->lock);
1230
1231 /*
1232 * step 4: check for races: did someone else allocate the undo struct?
1233 */
1234 un = lookup_undo(ulp, semid);
1235 if (un) {
1236 kfree(new);
1237 goto success;
1238 }
1239 /* step 5: initialize & link new undo structure */
1240 new->semadj = (short *) &new[1];
1241 new->ulp = ulp;
1242 new->semid = semid;
1243 assert_spin_locked(&ulp->lock);
1244 list_add_rcu(&new->list_proc, &ulp->list_proc);
1245 assert_spin_locked(&sma->sem_perm.lock);
1246 list_add(&new->list_id, &sma->list_id);
1247 un = new;
1248
1249success:
1250 spin_unlock(&ulp->lock);
1251 rcu_read_lock();
1252 sem_unlock(sma);
1253out:
1254 return un;
1255}
1256
1257
1258/**
1259 * get_queue_result - Retrieve the result code from sem_queue
1260 * @q: Pointer to queue structure
1261 *
1262 * Retrieve the return code from the pending queue. If IN_WAKEUP is found in
1263 * q->status, then we must loop until the value is replaced with the final
1264 * value: This may happen if a task is woken up by an unrelated event (e.g.
1265 * signal) and in parallel the task is woken up by another task because it got
1266 * the requested semaphores.
1267 *
1268 * The function can be called with or without holding the semaphore spinlock.
1269 */
1270static int get_queue_result(struct sem_queue *q)
1271{
1272 int error;
1273
1274 error = q->status;
1275 while (unlikely(error == IN_WAKEUP)) {
1276 cpu_relax();
1277 error = q->status;
1278 }
1279
1280 return error;
1281}
1282
1283
1284SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
1285 unsigned, nsops, const struct timespec __user *, timeout)
1286{
1287 int error = -EINVAL;
1288 struct sem_array *sma;
1289 struct sembuf fast_sops[SEMOPM_FAST];
1290 struct sembuf* sops = fast_sops, *sop;
1291 struct sem_undo *un;
1292 int undos = 0, alter = 0, max;
1293 struct sem_queue queue;
1294 unsigned long jiffies_left = 0;
1295 struct ipc_namespace *ns;
1296 struct list_head tasks;
1297
1298 ns = current->nsproxy->ipc_ns;
1299
1300 if (nsops < 1 || semid < 0)
1301 return -EINVAL;
1302 if (nsops > ns->sc_semopm)
1303 return -E2BIG;
1304 if(nsops > SEMOPM_FAST) {
1305 sops = kmalloc(sizeof(*sops)*nsops,GFP_KERNEL);
1306 if(sops==NULL)
1307 return -ENOMEM;
1308 }
1309 if (copy_from_user (sops, tsops, nsops * sizeof(*tsops))) {
1310 error=-EFAULT;
1311 goto out_free;
1312 }
1313 if (timeout) {
1314 struct timespec _timeout;
1315 if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1316 error = -EFAULT;
1317 goto out_free;
1318 }
1319 if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1320 _timeout.tv_nsec >= 1000000000L) {
1321 error = -EINVAL;
1322 goto out_free;
1323 }
1324 jiffies_left = timespec_to_jiffies(&_timeout);
1325 }
1326 max = 0;
1327 for (sop = sops; sop < sops + nsops; sop++) {
1328 if (sop->sem_num >= max)
1329 max = sop->sem_num;
1330 if (sop->sem_flg & SEM_UNDO)
1331 undos = 1;
1332 if (sop->sem_op != 0)
1333 alter = 1;
1334 }
1335
1336 if (undos) {
1337 un = find_alloc_undo(ns, semid);
1338 if (IS_ERR(un)) {
1339 error = PTR_ERR(un);
1340 goto out_free;
1341 }
1342 } else
1343 un = NULL;
1344
1345 INIT_LIST_HEAD(&tasks);
1346
1347 sma = sem_lock_check(ns, semid);
1348 if (IS_ERR(sma)) {
1349 if (un)
1350 rcu_read_unlock();
1351 error = PTR_ERR(sma);
1352 goto out_free;
1353 }
1354
1355 /*
1356 * semid identifiers are not unique - find_alloc_undo may have
1357 * allocated an undo structure, it was invalidated by an RMID
1358 * and now a new array with received the same id. Check and fail.
1359 * This case can be detected checking un->semid. The existence of
1360 * "un" itself is guaranteed by rcu.
1361 */
1362 error = -EIDRM;
1363 if (un) {
1364 if (un->semid == -1) {
1365 rcu_read_unlock();
1366 goto out_unlock_free;
1367 } else {
1368 /*
1369 * rcu lock can be released, "un" cannot disappear:
1370 * - sem_lock is acquired, thus IPC_RMID is
1371 * impossible.
1372 * - exit_sem is impossible, it always operates on
1373 * current (or a dead task).
1374 */
1375
1376 rcu_read_unlock();
1377 }
1378 }
1379
1380 error = -EFBIG;
1381 if (max >= sma->sem_nsems)
1382 goto out_unlock_free;
1383
1384 error = -EACCES;
1385 if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO))
1386 goto out_unlock_free;
1387
1388 error = security_sem_semop(sma, sops, nsops, alter);
1389 if (error)
1390 goto out_unlock_free;
1391
1392 error = try_atomic_semop (sma, sops, nsops, un, task_tgid_vnr(current));
1393 if (error <= 0) {
1394 if (alter && error == 0)
1395 do_smart_update(sma, sops, nsops, 1, &tasks);
1396
1397 goto out_unlock_free;
1398 }
1399
1400 /* We need to sleep on this operation, so we put the current
1401 * task into the pending queue and go to sleep.
1402 */
1403
1404 queue.sops = sops;
1405 queue.nsops = nsops;
1406 queue.undo = un;
1407 queue.pid = task_tgid_vnr(current);
1408 queue.alter = alter;
1409 if (alter)
1410 list_add_tail(&queue.list, &sma->sem_pending);
1411 else
1412 list_add(&queue.list, &sma->sem_pending);
1413
1414 if (nsops == 1) {
1415 struct sem *curr;
1416 curr = &sma->sem_base[sops->sem_num];
1417
1418 if (alter)
1419 list_add_tail(&queue.simple_list, &curr->sem_pending);
1420 else
1421 list_add(&queue.simple_list, &curr->sem_pending);
1422 } else {
1423 INIT_LIST_HEAD(&queue.simple_list);
1424 sma->complex_count++;
1425 }
1426
1427 queue.status = -EINTR;
1428 queue.sleeper = current;
1429 current->state = TASK_INTERRUPTIBLE;
1430 sem_unlock(sma);
1431
1432 if (timeout)
1433 jiffies_left = schedule_timeout(jiffies_left);
1434 else
1435 schedule();
1436
1437 error = get_queue_result(&queue);
1438
1439 if (error != -EINTR) {
1440 /* fast path: update_queue already obtained all requested
1441 * resources.
1442 * Perform a smp_mb(): User space could assume that semop()
1443 * is a memory barrier: Without the mb(), the cpu could
1444 * speculatively read in user space stale data that was
1445 * overwritten by the previous owner of the semaphore.
1446 */
1447 smp_mb();
1448
1449 goto out_free;
1450 }
1451
1452 sma = sem_lock(ns, semid);
1453
1454 /*
1455 * Wait until it's guaranteed that no wakeup_sem_queue_do() is ongoing.
1456 */
1457 error = get_queue_result(&queue);
1458
1459 /*
1460 * Array removed? If yes, leave without sem_unlock().
1461 */
1462 if (IS_ERR(sma)) {
1463 error = -EIDRM;
1464 goto out_free;
1465 }
1466
1467
1468 /*
1469 * If queue.status != -EINTR we are woken up by another process.
1470 * Leave without unlink_queue(), but with sem_unlock().
1471 */
1472
1473 if (error != -EINTR) {
1474 goto out_unlock_free;
1475 }
1476
1477 /*
1478 * If an interrupt occurred we have to clean up the queue
1479 */
1480 if (timeout && jiffies_left == 0)
1481 error = -EAGAIN;
1482 unlink_queue(sma, &queue);
1483
1484out_unlock_free:
1485 sem_unlock(sma);
1486
1487 wake_up_sem_queue_do(&tasks);
1488out_free:
1489 if(sops != fast_sops)
1490 kfree(sops);
1491 return error;
1492}
1493
1494SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
1495 unsigned, nsops)
1496{
1497 return sys_semtimedop(semid, tsops, nsops, NULL);
1498}
1499
1500/* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
1501 * parent and child tasks.
1502 */
1503
1504int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
1505{
1506 struct sem_undo_list *undo_list;
1507 int error;
1508
1509 if (clone_flags & CLONE_SYSVSEM) {
1510 error = get_undo_list(&undo_list);
1511 if (error)
1512 return error;
1513 atomic_inc(&undo_list->refcnt);
1514 tsk->sysvsem.undo_list = undo_list;
1515 } else
1516 tsk->sysvsem.undo_list = NULL;
1517
1518 return 0;
1519}
1520
1521/*
1522 * add semadj values to semaphores, free undo structures.
1523 * undo structures are not freed when semaphore arrays are destroyed
1524 * so some of them may be out of date.
1525 * IMPLEMENTATION NOTE: There is some confusion over whether the
1526 * set of adjustments that needs to be done should be done in an atomic
1527 * manner or not. That is, if we are attempting to decrement the semval
1528 * should we queue up and wait until we can do so legally?
1529 * The original implementation attempted to do this (queue and wait).
1530 * The current implementation does not do so. The POSIX standard
1531 * and SVID should be consulted to determine what behavior is mandated.
1532 */
1533void exit_sem(struct task_struct *tsk)
1534{
1535 struct sem_undo_list *ulp;
1536
1537 ulp = tsk->sysvsem.undo_list;
1538 if (!ulp)
1539 return;
1540 tsk->sysvsem.undo_list = NULL;
1541
1542 if (!atomic_dec_and_test(&ulp->refcnt))
1543 return;
1544
1545 for (;;) {
1546 struct sem_array *sma;
1547 struct sem_undo *un;
1548 struct list_head tasks;
1549 int semid;
1550 int i;
1551
1552 rcu_read_lock();
1553 un = list_entry_rcu(ulp->list_proc.next,
1554 struct sem_undo, list_proc);
1555 if (&un->list_proc == &ulp->list_proc)
1556 semid = -1;
1557 else
1558 semid = un->semid;
1559 rcu_read_unlock();
1560
1561 if (semid == -1)
1562 break;
1563
1564 sma = sem_lock_check(tsk->nsproxy->ipc_ns, un->semid);
1565
1566 /* exit_sem raced with IPC_RMID, nothing to do */
1567 if (IS_ERR(sma))
1568 continue;
1569
1570 un = __lookup_undo(ulp, semid);
1571 if (un == NULL) {
1572 /* exit_sem raced with IPC_RMID+semget() that created
1573 * exactly the same semid. Nothing to do.
1574 */
1575 sem_unlock(sma);
1576 continue;
1577 }
1578
1579 /* remove un from the linked lists */
1580 assert_spin_locked(&sma->sem_perm.lock);
1581 list_del(&un->list_id);
1582
1583 spin_lock(&ulp->lock);
1584 list_del_rcu(&un->list_proc);
1585 spin_unlock(&ulp->lock);
1586
1587 /* perform adjustments registered in un */
1588 for (i = 0; i < sma->sem_nsems; i++) {
1589 struct sem * semaphore = &sma->sem_base[i];
1590 if (un->semadj[i]) {
1591 semaphore->semval += un->semadj[i];
1592 /*
1593 * Range checks of the new semaphore value,
1594 * not defined by sus:
1595 * - Some unices ignore the undo entirely
1596 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
1597 * - some cap the value (e.g. FreeBSD caps
1598 * at 0, but doesn't enforce SEMVMX)
1599 *
1600 * Linux caps the semaphore value, both at 0
1601 * and at SEMVMX.
1602 *
1603 * Manfred <manfred@colorfullife.com>
1604 */
1605 if (semaphore->semval < 0)
1606 semaphore->semval = 0;
1607 if (semaphore->semval > SEMVMX)
1608 semaphore->semval = SEMVMX;
1609 semaphore->sempid = task_tgid_vnr(current);
1610 }
1611 }
1612 /* maybe some queued-up processes were waiting for this */
1613 INIT_LIST_HEAD(&tasks);
1614 do_smart_update(sma, NULL, 0, 1, &tasks);
1615 sem_unlock(sma);
1616 wake_up_sem_queue_do(&tasks);
1617
1618 kfree_rcu(un, rcu);
1619 }
1620 kfree(ulp);
1621}
1622
1623#ifdef CONFIG_PROC_FS
1624static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
1625{
1626 struct sem_array *sma = it;
1627
1628 return seq_printf(s,
1629 "%10d %10d %4o %10u %5u %5u %5u %5u %10lu %10lu\n",
1630 sma->sem_perm.key,
1631 sma->sem_perm.id,
1632 sma->sem_perm.mode,
1633 sma->sem_nsems,
1634 sma->sem_perm.uid,
1635 sma->sem_perm.gid,
1636 sma->sem_perm.cuid,
1637 sma->sem_perm.cgid,
1638 sma->sem_otime,
1639 sma->sem_ctime);
1640}
1641#endif
1/*
2 * linux/ipc/sem.c
3 * Copyright (C) 1992 Krishna Balasubramanian
4 * Copyright (C) 1995 Eric Schenk, Bruno Haible
5 *
6 * /proc/sysvipc/sem support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
7 *
8 * SMP-threaded, sysctl's added
9 * (c) 1999 Manfred Spraul <manfred@colorfullife.com>
10 * Enforced range limit on SEM_UNDO
11 * (c) 2001 Red Hat Inc
12 * Lockless wakeup
13 * (c) 2003 Manfred Spraul <manfred@colorfullife.com>
14 * Further wakeup optimizations, documentation
15 * (c) 2010 Manfred Spraul <manfred@colorfullife.com>
16 *
17 * support for audit of ipc object properties and permission changes
18 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
19 *
20 * namespaces support
21 * OpenVZ, SWsoft Inc.
22 * Pavel Emelianov <xemul@openvz.org>
23 *
24 * Implementation notes: (May 2010)
25 * This file implements System V semaphores.
26 *
27 * User space visible behavior:
28 * - FIFO ordering for semop() operations (just FIFO, not starvation
29 * protection)
30 * - multiple semaphore operations that alter the same semaphore in
31 * one semop() are handled.
32 * - sem_ctime (time of last semctl()) is updated in the IPC_SET, SETVAL and
33 * SETALL calls.
34 * - two Linux specific semctl() commands: SEM_STAT, SEM_INFO.
35 * - undo adjustments at process exit are limited to 0..SEMVMX.
36 * - namespace are supported.
37 * - SEMMSL, SEMMNS, SEMOPM and SEMMNI can be configured at runtine by writing
38 * to /proc/sys/kernel/sem.
39 * - statistics about the usage are reported in /proc/sysvipc/sem.
40 *
41 * Internals:
42 * - scalability:
43 * - all global variables are read-mostly.
44 * - semop() calls and semctl(RMID) are synchronized by RCU.
45 * - most operations do write operations (actually: spin_lock calls) to
46 * the per-semaphore array structure.
47 * Thus: Perfect SMP scaling between independent semaphore arrays.
48 * If multiple semaphores in one array are used, then cache line
49 * trashing on the semaphore array spinlock will limit the scaling.
50 * - semncnt and semzcnt are calculated on demand in count_semncnt() and
51 * count_semzcnt()
52 * - the task that performs a successful semop() scans the list of all
53 * sleeping tasks and completes any pending operations that can be fulfilled.
54 * Semaphores are actively given to waiting tasks (necessary for FIFO).
55 * (see update_queue())
56 * - To improve the scalability, the actual wake-up calls are performed after
57 * dropping all locks. (see wake_up_sem_queue_prepare(),
58 * wake_up_sem_queue_do())
59 * - All work is done by the waker, the woken up task does not have to do
60 * anything - not even acquiring a lock or dropping a refcount.
61 * - A woken up task may not even touch the semaphore array anymore, it may
62 * have been destroyed already by a semctl(RMID).
63 * - The synchronizations between wake-ups due to a timeout/signal and a
64 * wake-up due to a completed semaphore operation is achieved by using an
65 * intermediate state (IN_WAKEUP).
66 * - UNDO values are stored in an array (one per process and per
67 * semaphore array, lazily allocated). For backwards compatibility, multiple
68 * modes for the UNDO variables are supported (per process, per thread)
69 * (see copy_semundo, CLONE_SYSVSEM)
70 * - There are two lists of the pending operations: a per-array list
71 * and per-semaphore list (stored in the array). This allows to achieve FIFO
72 * ordering without always scanning all pending operations.
73 * The worst-case behavior is nevertheless O(N^2) for N wakeups.
74 */
75
76#include <linux/slab.h>
77#include <linux/spinlock.h>
78#include <linux/init.h>
79#include <linux/proc_fs.h>
80#include <linux/time.h>
81#include <linux/security.h>
82#include <linux/syscalls.h>
83#include <linux/audit.h>
84#include <linux/capability.h>
85#include <linux/seq_file.h>
86#include <linux/rwsem.h>
87#include <linux/nsproxy.h>
88#include <linux/ipc_namespace.h>
89
90#include <asm/uaccess.h>
91#include "util.h"
92
93/* One semaphore structure for each semaphore in the system. */
94struct sem {
95 int semval; /* current value */
96 int sempid; /* pid of last operation */
97 spinlock_t lock; /* spinlock for fine-grained semtimedop */
98 struct list_head pending_alter; /* pending single-sop operations */
99 /* that alter the semaphore */
100 struct list_head pending_const; /* pending single-sop operations */
101 /* that do not alter the semaphore*/
102 time_t sem_otime; /* candidate for sem_otime */
103} ____cacheline_aligned_in_smp;
104
105/* One queue for each sleeping process in the system. */
106struct sem_queue {
107 struct list_head list; /* queue of pending operations */
108 struct task_struct *sleeper; /* this process */
109 struct sem_undo *undo; /* undo structure */
110 int pid; /* process id of requesting process */
111 int status; /* completion status of operation */
112 struct sembuf *sops; /* array of pending operations */
113 int nsops; /* number of operations */
114 int alter; /* does *sops alter the array? */
115};
116
117/* Each task has a list of undo requests. They are executed automatically
118 * when the process exits.
119 */
120struct sem_undo {
121 struct list_head list_proc; /* per-process list: *
122 * all undos from one process
123 * rcu protected */
124 struct rcu_head rcu; /* rcu struct for sem_undo */
125 struct sem_undo_list *ulp; /* back ptr to sem_undo_list */
126 struct list_head list_id; /* per semaphore array list:
127 * all undos for one array */
128 int semid; /* semaphore set identifier */
129 short *semadj; /* array of adjustments */
130 /* one per semaphore */
131};
132
133/* sem_undo_list controls shared access to the list of sem_undo structures
134 * that may be shared among all a CLONE_SYSVSEM task group.
135 */
136struct sem_undo_list {
137 atomic_t refcnt;
138 spinlock_t lock;
139 struct list_head list_proc;
140};
141
142
143#define sem_ids(ns) ((ns)->ids[IPC_SEM_IDS])
144
145#define sem_checkid(sma, semid) ipc_checkid(&sma->sem_perm, semid)
146
147static int newary(struct ipc_namespace *, struct ipc_params *);
148static void freeary(struct ipc_namespace *, struct kern_ipc_perm *);
149#ifdef CONFIG_PROC_FS
150static int sysvipc_sem_proc_show(struct seq_file *s, void *it);
151#endif
152
153#define SEMMSL_FAST 256 /* 512 bytes on stack */
154#define SEMOPM_FAST 64 /* ~ 372 bytes on stack */
155
156/*
157 * Locking:
158 * sem_undo.id_next,
159 * sem_array.complex_count,
160 * sem_array.pending{_alter,_cont},
161 * sem_array.sem_undo: global sem_lock() for read/write
162 * sem_undo.proc_next: only "current" is allowed to read/write that field.
163 *
164 * sem_array.sem_base[i].pending_{const,alter}:
165 * global or semaphore sem_lock() for read/write
166 */
167
168#define sc_semmsl sem_ctls[0]
169#define sc_semmns sem_ctls[1]
170#define sc_semopm sem_ctls[2]
171#define sc_semmni sem_ctls[3]
172
173void sem_init_ns(struct ipc_namespace *ns)
174{
175 ns->sc_semmsl = SEMMSL;
176 ns->sc_semmns = SEMMNS;
177 ns->sc_semopm = SEMOPM;
178 ns->sc_semmni = SEMMNI;
179 ns->used_sems = 0;
180 ipc_init_ids(&ns->ids[IPC_SEM_IDS]);
181}
182
183#ifdef CONFIG_IPC_NS
184void sem_exit_ns(struct ipc_namespace *ns)
185{
186 free_ipcs(ns, &sem_ids(ns), freeary);
187 idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr);
188}
189#endif
190
191void __init sem_init(void)
192{
193 sem_init_ns(&init_ipc_ns);
194 ipc_init_proc_interface("sysvipc/sem",
195 " key semid perms nsems uid gid cuid cgid otime ctime\n",
196 IPC_SEM_IDS, sysvipc_sem_proc_show);
197}
198
199/**
200 * unmerge_queues - unmerge queues, if possible.
201 * @sma: semaphore array
202 *
203 * The function unmerges the wait queues if complex_count is 0.
204 * It must be called prior to dropping the global semaphore array lock.
205 */
206static void unmerge_queues(struct sem_array *sma)
207{
208 struct sem_queue *q, *tq;
209
210 /* complex operations still around? */
211 if (sma->complex_count)
212 return;
213 /*
214 * We will switch back to simple mode.
215 * Move all pending operation back into the per-semaphore
216 * queues.
217 */
218 list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
219 struct sem *curr;
220 curr = &sma->sem_base[q->sops[0].sem_num];
221
222 list_add_tail(&q->list, &curr->pending_alter);
223 }
224 INIT_LIST_HEAD(&sma->pending_alter);
225}
226
227/**
228 * merge_queues - merge single semop queues into global queue
229 * @sma: semaphore array
230 *
231 * This function merges all per-semaphore queues into the global queue.
232 * It is necessary to achieve FIFO ordering for the pending single-sop
233 * operations when a multi-semop operation must sleep.
234 * Only the alter operations must be moved, the const operations can stay.
235 */
236static void merge_queues(struct sem_array *sma)
237{
238 int i;
239 for (i = 0; i < sma->sem_nsems; i++) {
240 struct sem *sem = sma->sem_base + i;
241
242 list_splice_init(&sem->pending_alter, &sma->pending_alter);
243 }
244}
245
246static void sem_rcu_free(struct rcu_head *head)
247{
248 struct ipc_rcu *p = container_of(head, struct ipc_rcu, rcu);
249 struct sem_array *sma = ipc_rcu_to_struct(p);
250
251 security_sem_free(sma);
252 ipc_rcu_free(head);
253}
254
255/*
256 * Wait until all currently ongoing simple ops have completed.
257 * Caller must own sem_perm.lock.
258 * New simple ops cannot start, because simple ops first check
259 * that sem_perm.lock is free.
260 * that a) sem_perm.lock is free and b) complex_count is 0.
261 */
262static void sem_wait_array(struct sem_array *sma)
263{
264 int i;
265 struct sem *sem;
266
267 if (sma->complex_count) {
268 /* The thread that increased sma->complex_count waited on
269 * all sem->lock locks. Thus we don't need to wait again.
270 */
271 return;
272 }
273
274 for (i = 0; i < sma->sem_nsems; i++) {
275 sem = sma->sem_base + i;
276 spin_unlock_wait(&sem->lock);
277 }
278}
279
280/*
281 * If the request contains only one semaphore operation, and there are
282 * no complex transactions pending, lock only the semaphore involved.
283 * Otherwise, lock the entire semaphore array, since we either have
284 * multiple semaphores in our own semops, or we need to look at
285 * semaphores from other pending complex operations.
286 */
287static inline int sem_lock(struct sem_array *sma, struct sembuf *sops,
288 int nsops)
289{
290 struct sem *sem;
291
292 if (nsops != 1) {
293 /* Complex operation - acquire a full lock */
294 ipc_lock_object(&sma->sem_perm);
295
296 /* And wait until all simple ops that are processed
297 * right now have dropped their locks.
298 */
299 sem_wait_array(sma);
300 return -1;
301 }
302
303 /*
304 * Only one semaphore affected - try to optimize locking.
305 * The rules are:
306 * - optimized locking is possible if no complex operation
307 * is either enqueued or processed right now.
308 * - The test for enqueued complex ops is simple:
309 * sma->complex_count != 0
310 * - Testing for complex ops that are processed right now is
311 * a bit more difficult. Complex ops acquire the full lock
312 * and first wait that the running simple ops have completed.
313 * (see above)
314 * Thus: If we own a simple lock and the global lock is free
315 * and complex_count is now 0, then it will stay 0 and
316 * thus just locking sem->lock is sufficient.
317 */
318 sem = sma->sem_base + sops->sem_num;
319
320 if (sma->complex_count == 0) {
321 /*
322 * It appears that no complex operation is around.
323 * Acquire the per-semaphore lock.
324 */
325 spin_lock(&sem->lock);
326
327 /* Then check that the global lock is free */
328 if (!spin_is_locked(&sma->sem_perm.lock)) {
329 /* spin_is_locked() is not a memory barrier */
330 smp_mb();
331
332 /* Now repeat the test of complex_count:
333 * It can't change anymore until we drop sem->lock.
334 * Thus: if is now 0, then it will stay 0.
335 */
336 if (sma->complex_count == 0) {
337 /* fast path successful! */
338 return sops->sem_num;
339 }
340 }
341 spin_unlock(&sem->lock);
342 }
343
344 /* slow path: acquire the full lock */
345 ipc_lock_object(&sma->sem_perm);
346
347 if (sma->complex_count == 0) {
348 /* False alarm:
349 * There is no complex operation, thus we can switch
350 * back to the fast path.
351 */
352 spin_lock(&sem->lock);
353 ipc_unlock_object(&sma->sem_perm);
354 return sops->sem_num;
355 } else {
356 /* Not a false alarm, thus complete the sequence for a
357 * full lock.
358 */
359 sem_wait_array(sma);
360 return -1;
361 }
362}
363
364static inline void sem_unlock(struct sem_array *sma, int locknum)
365{
366 if (locknum == -1) {
367 unmerge_queues(sma);
368 ipc_unlock_object(&sma->sem_perm);
369 } else {
370 struct sem *sem = sma->sem_base + locknum;
371 spin_unlock(&sem->lock);
372 }
373}
374
375/*
376 * sem_lock_(check_) routines are called in the paths where the rwsem
377 * is not held.
378 *
379 * The caller holds the RCU read lock.
380 */
381static inline struct sem_array *sem_obtain_lock(struct ipc_namespace *ns,
382 int id, struct sembuf *sops, int nsops, int *locknum)
383{
384 struct kern_ipc_perm *ipcp;
385 struct sem_array *sma;
386
387 ipcp = ipc_obtain_object(&sem_ids(ns), id);
388 if (IS_ERR(ipcp))
389 return ERR_CAST(ipcp);
390
391 sma = container_of(ipcp, struct sem_array, sem_perm);
392 *locknum = sem_lock(sma, sops, nsops);
393
394 /* ipc_rmid() may have already freed the ID while sem_lock
395 * was spinning: verify that the structure is still valid
396 */
397 if (ipc_valid_object(ipcp))
398 return container_of(ipcp, struct sem_array, sem_perm);
399
400 sem_unlock(sma, *locknum);
401 return ERR_PTR(-EINVAL);
402}
403
404static inline struct sem_array *sem_obtain_object(struct ipc_namespace *ns, int id)
405{
406 struct kern_ipc_perm *ipcp = ipc_obtain_object(&sem_ids(ns), id);
407
408 if (IS_ERR(ipcp))
409 return ERR_CAST(ipcp);
410
411 return container_of(ipcp, struct sem_array, sem_perm);
412}
413
414static inline struct sem_array *sem_obtain_object_check(struct ipc_namespace *ns,
415 int id)
416{
417 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&sem_ids(ns), id);
418
419 if (IS_ERR(ipcp))
420 return ERR_CAST(ipcp);
421
422 return container_of(ipcp, struct sem_array, sem_perm);
423}
424
425static inline void sem_lock_and_putref(struct sem_array *sma)
426{
427 sem_lock(sma, NULL, -1);
428 ipc_rcu_putref(sma, ipc_rcu_free);
429}
430
431static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
432{
433 ipc_rmid(&sem_ids(ns), &s->sem_perm);
434}
435
436/*
437 * Lockless wakeup algorithm:
438 * Without the check/retry algorithm a lockless wakeup is possible:
439 * - queue.status is initialized to -EINTR before blocking.
440 * - wakeup is performed by
441 * * unlinking the queue entry from the pending list
442 * * setting queue.status to IN_WAKEUP
443 * This is the notification for the blocked thread that a
444 * result value is imminent.
445 * * call wake_up_process
446 * * set queue.status to the final value.
447 * - the previously blocked thread checks queue.status:
448 * * if it's IN_WAKEUP, then it must wait until the value changes
449 * * if it's not -EINTR, then the operation was completed by
450 * update_queue. semtimedop can return queue.status without
451 * performing any operation on the sem array.
452 * * otherwise it must acquire the spinlock and check what's up.
453 *
454 * The two-stage algorithm is necessary to protect against the following
455 * races:
456 * - if queue.status is set after wake_up_process, then the woken up idle
457 * thread could race forward and try (and fail) to acquire sma->lock
458 * before update_queue had a chance to set queue.status
459 * - if queue.status is written before wake_up_process and if the
460 * blocked process is woken up by a signal between writing
461 * queue.status and the wake_up_process, then the woken up
462 * process could return from semtimedop and die by calling
463 * sys_exit before wake_up_process is called. Then wake_up_process
464 * will oops, because the task structure is already invalid.
465 * (yes, this happened on s390 with sysv msg).
466 *
467 */
468#define IN_WAKEUP 1
469
470/**
471 * newary - Create a new semaphore set
472 * @ns: namespace
473 * @params: ptr to the structure that contains key, semflg and nsems
474 *
475 * Called with sem_ids.rwsem held (as a writer)
476 */
477static int newary(struct ipc_namespace *ns, struct ipc_params *params)
478{
479 int id;
480 int retval;
481 struct sem_array *sma;
482 int size;
483 key_t key = params->key;
484 int nsems = params->u.nsems;
485 int semflg = params->flg;
486 int i;
487
488 if (!nsems)
489 return -EINVAL;
490 if (ns->used_sems + nsems > ns->sc_semmns)
491 return -ENOSPC;
492
493 size = sizeof(*sma) + nsems * sizeof(struct sem);
494 sma = ipc_rcu_alloc(size);
495 if (!sma)
496 return -ENOMEM;
497
498 memset(sma, 0, size);
499
500 sma->sem_perm.mode = (semflg & S_IRWXUGO);
501 sma->sem_perm.key = key;
502
503 sma->sem_perm.security = NULL;
504 retval = security_sem_alloc(sma);
505 if (retval) {
506 ipc_rcu_putref(sma, ipc_rcu_free);
507 return retval;
508 }
509
510 id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
511 if (id < 0) {
512 ipc_rcu_putref(sma, sem_rcu_free);
513 return id;
514 }
515 ns->used_sems += nsems;
516
517 sma->sem_base = (struct sem *) &sma[1];
518
519 for (i = 0; i < nsems; i++) {
520 INIT_LIST_HEAD(&sma->sem_base[i].pending_alter);
521 INIT_LIST_HEAD(&sma->sem_base[i].pending_const);
522 spin_lock_init(&sma->sem_base[i].lock);
523 }
524
525 sma->complex_count = 0;
526 INIT_LIST_HEAD(&sma->pending_alter);
527 INIT_LIST_HEAD(&sma->pending_const);
528 INIT_LIST_HEAD(&sma->list_id);
529 sma->sem_nsems = nsems;
530 sma->sem_ctime = get_seconds();
531 sem_unlock(sma, -1);
532 rcu_read_unlock();
533
534 return sma->sem_perm.id;
535}
536
537
538/*
539 * Called with sem_ids.rwsem and ipcp locked.
540 */
541static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg)
542{
543 struct sem_array *sma;
544
545 sma = container_of(ipcp, struct sem_array, sem_perm);
546 return security_sem_associate(sma, semflg);
547}
548
549/*
550 * Called with sem_ids.rwsem and ipcp locked.
551 */
552static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
553 struct ipc_params *params)
554{
555 struct sem_array *sma;
556
557 sma = container_of(ipcp, struct sem_array, sem_perm);
558 if (params->u.nsems > sma->sem_nsems)
559 return -EINVAL;
560
561 return 0;
562}
563
564SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg)
565{
566 struct ipc_namespace *ns;
567 struct ipc_ops sem_ops;
568 struct ipc_params sem_params;
569
570 ns = current->nsproxy->ipc_ns;
571
572 if (nsems < 0 || nsems > ns->sc_semmsl)
573 return -EINVAL;
574
575 sem_ops.getnew = newary;
576 sem_ops.associate = sem_security;
577 sem_ops.more_checks = sem_more_checks;
578
579 sem_params.key = key;
580 sem_params.flg = semflg;
581 sem_params.u.nsems = nsems;
582
583 return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
584}
585
586/**
587 * perform_atomic_semop - Perform (if possible) a semaphore operation
588 * @sma: semaphore array
589 * @sops: array with operations that should be checked
590 * @nsops: number of operations
591 * @un: undo array
592 * @pid: pid that did the change
593 *
594 * Returns 0 if the operation was possible.
595 * Returns 1 if the operation is impossible, the caller must sleep.
596 * Negative values are error codes.
597 */
598static int perform_atomic_semop(struct sem_array *sma, struct sembuf *sops,
599 int nsops, struct sem_undo *un, int pid)
600{
601 int result, sem_op;
602 struct sembuf *sop;
603 struct sem *curr;
604
605 for (sop = sops; sop < sops + nsops; sop++) {
606 curr = sma->sem_base + sop->sem_num;
607 sem_op = sop->sem_op;
608 result = curr->semval;
609
610 if (!sem_op && result)
611 goto would_block;
612
613 result += sem_op;
614 if (result < 0)
615 goto would_block;
616 if (result > SEMVMX)
617 goto out_of_range;
618
619 if (sop->sem_flg & SEM_UNDO) {
620 int undo = un->semadj[sop->sem_num] - sem_op;
621 /* Exceeding the undo range is an error. */
622 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
623 goto out_of_range;
624 un->semadj[sop->sem_num] = undo;
625 }
626
627 curr->semval = result;
628 }
629
630 sop--;
631 while (sop >= sops) {
632 sma->sem_base[sop->sem_num].sempid = pid;
633 sop--;
634 }
635
636 return 0;
637
638out_of_range:
639 result = -ERANGE;
640 goto undo;
641
642would_block:
643 if (sop->sem_flg & IPC_NOWAIT)
644 result = -EAGAIN;
645 else
646 result = 1;
647
648undo:
649 sop--;
650 while (sop >= sops) {
651 sem_op = sop->sem_op;
652 sma->sem_base[sop->sem_num].semval -= sem_op;
653 if (sop->sem_flg & SEM_UNDO)
654 un->semadj[sop->sem_num] += sem_op;
655 sop--;
656 }
657
658 return result;
659}
660
661/** wake_up_sem_queue_prepare(q, error): Prepare wake-up
662 * @q: queue entry that must be signaled
663 * @error: Error value for the signal
664 *
665 * Prepare the wake-up of the queue entry q.
666 */
667static void wake_up_sem_queue_prepare(struct list_head *pt,
668 struct sem_queue *q, int error)
669{
670 if (list_empty(pt)) {
671 /*
672 * Hold preempt off so that we don't get preempted and have the
673 * wakee busy-wait until we're scheduled back on.
674 */
675 preempt_disable();
676 }
677 q->status = IN_WAKEUP;
678 q->pid = error;
679
680 list_add_tail(&q->list, pt);
681}
682
683/**
684 * wake_up_sem_queue_do - do the actual wake-up
685 * @pt: list of tasks to be woken up
686 *
687 * Do the actual wake-up.
688 * The function is called without any locks held, thus the semaphore array
689 * could be destroyed already and the tasks can disappear as soon as the
690 * status is set to the actual return code.
691 */
692static void wake_up_sem_queue_do(struct list_head *pt)
693{
694 struct sem_queue *q, *t;
695 int did_something;
696
697 did_something = !list_empty(pt);
698 list_for_each_entry_safe(q, t, pt, list) {
699 wake_up_process(q->sleeper);
700 /* q can disappear immediately after writing q->status. */
701 smp_wmb();
702 q->status = q->pid;
703 }
704 if (did_something)
705 preempt_enable();
706}
707
708static void unlink_queue(struct sem_array *sma, struct sem_queue *q)
709{
710 list_del(&q->list);
711 if (q->nsops > 1)
712 sma->complex_count--;
713}
714
715/** check_restart(sma, q)
716 * @sma: semaphore array
717 * @q: the operation that just completed
718 *
719 * update_queue is O(N^2) when it restarts scanning the whole queue of
720 * waiting operations. Therefore this function checks if the restart is
721 * really necessary. It is called after a previously waiting operation
722 * modified the array.
723 * Note that wait-for-zero operations are handled without restart.
724 */
725static int check_restart(struct sem_array *sma, struct sem_queue *q)
726{
727 /* pending complex alter operations are too difficult to analyse */
728 if (!list_empty(&sma->pending_alter))
729 return 1;
730
731 /* we were a sleeping complex operation. Too difficult */
732 if (q->nsops > 1)
733 return 1;
734
735 /* It is impossible that someone waits for the new value:
736 * - complex operations always restart.
737 * - wait-for-zero are handled seperately.
738 * - q is a previously sleeping simple operation that
739 * altered the array. It must be a decrement, because
740 * simple increments never sleep.
741 * - If there are older (higher priority) decrements
742 * in the queue, then they have observed the original
743 * semval value and couldn't proceed. The operation
744 * decremented to value - thus they won't proceed either.
745 */
746 return 0;
747}
748
749/**
750 * wake_const_ops - wake up non-alter tasks
751 * @sma: semaphore array.
752 * @semnum: semaphore that was modified.
753 * @pt: list head for the tasks that must be woken up.
754 *
755 * wake_const_ops must be called after a semaphore in a semaphore array
756 * was set to 0. If complex const operations are pending, wake_const_ops must
757 * be called with semnum = -1, as well as with the number of each modified
758 * semaphore.
759 * The tasks that must be woken up are added to @pt. The return code
760 * is stored in q->pid.
761 * The function returns 1 if at least one operation was completed successfully.
762 */
763static int wake_const_ops(struct sem_array *sma, int semnum,
764 struct list_head *pt)
765{
766 struct sem_queue *q;
767 struct list_head *walk;
768 struct list_head *pending_list;
769 int semop_completed = 0;
770
771 if (semnum == -1)
772 pending_list = &sma->pending_const;
773 else
774 pending_list = &sma->sem_base[semnum].pending_const;
775
776 walk = pending_list->next;
777 while (walk != pending_list) {
778 int error;
779
780 q = container_of(walk, struct sem_queue, list);
781 walk = walk->next;
782
783 error = perform_atomic_semop(sma, q->sops, q->nsops,
784 q->undo, q->pid);
785
786 if (error <= 0) {
787 /* operation completed, remove from queue & wakeup */
788
789 unlink_queue(sma, q);
790
791 wake_up_sem_queue_prepare(pt, q, error);
792 if (error == 0)
793 semop_completed = 1;
794 }
795 }
796 return semop_completed;
797}
798
799/**
800 * do_smart_wakeup_zero - wakeup all wait for zero tasks
801 * @sma: semaphore array
802 * @sops: operations that were performed
803 * @nsops: number of operations
804 * @pt: list head of the tasks that must be woken up.
805 *
806 * Checks all required queue for wait-for-zero operations, based
807 * on the actual changes that were performed on the semaphore array.
808 * The function returns 1 if at least one operation was completed successfully.
809 */
810static int do_smart_wakeup_zero(struct sem_array *sma, struct sembuf *sops,
811 int nsops, struct list_head *pt)
812{
813 int i;
814 int semop_completed = 0;
815 int got_zero = 0;
816
817 /* first: the per-semaphore queues, if known */
818 if (sops) {
819 for (i = 0; i < nsops; i++) {
820 int num = sops[i].sem_num;
821
822 if (sma->sem_base[num].semval == 0) {
823 got_zero = 1;
824 semop_completed |= wake_const_ops(sma, num, pt);
825 }
826 }
827 } else {
828 /*
829 * No sops means modified semaphores not known.
830 * Assume all were changed.
831 */
832 for (i = 0; i < sma->sem_nsems; i++) {
833 if (sma->sem_base[i].semval == 0) {
834 got_zero = 1;
835 semop_completed |= wake_const_ops(sma, i, pt);
836 }
837 }
838 }
839 /*
840 * If one of the modified semaphores got 0,
841 * then check the global queue, too.
842 */
843 if (got_zero)
844 semop_completed |= wake_const_ops(sma, -1, pt);
845
846 return semop_completed;
847}
848
849
850/**
851 * update_queue - look for tasks that can be completed.
852 * @sma: semaphore array.
853 * @semnum: semaphore that was modified.
854 * @pt: list head for the tasks that must be woken up.
855 *
856 * update_queue must be called after a semaphore in a semaphore array
857 * was modified. If multiple semaphores were modified, update_queue must
858 * be called with semnum = -1, as well as with the number of each modified
859 * semaphore.
860 * The tasks that must be woken up are added to @pt. The return code
861 * is stored in q->pid.
862 * The function internally checks if const operations can now succeed.
863 *
864 * The function return 1 if at least one semop was completed successfully.
865 */
866static int update_queue(struct sem_array *sma, int semnum, struct list_head *pt)
867{
868 struct sem_queue *q;
869 struct list_head *walk;
870 struct list_head *pending_list;
871 int semop_completed = 0;
872
873 if (semnum == -1)
874 pending_list = &sma->pending_alter;
875 else
876 pending_list = &sma->sem_base[semnum].pending_alter;
877
878again:
879 walk = pending_list->next;
880 while (walk != pending_list) {
881 int error, restart;
882
883 q = container_of(walk, struct sem_queue, list);
884 walk = walk->next;
885
886 /* If we are scanning the single sop, per-semaphore list of
887 * one semaphore and that semaphore is 0, then it is not
888 * necessary to scan further: simple increments
889 * that affect only one entry succeed immediately and cannot
890 * be in the per semaphore pending queue, and decrements
891 * cannot be successful if the value is already 0.
892 */
893 if (semnum != -1 && sma->sem_base[semnum].semval == 0)
894 break;
895
896 error = perform_atomic_semop(sma, q->sops, q->nsops,
897 q->undo, q->pid);
898
899 /* Does q->sleeper still need to sleep? */
900 if (error > 0)
901 continue;
902
903 unlink_queue(sma, q);
904
905 if (error) {
906 restart = 0;
907 } else {
908 semop_completed = 1;
909 do_smart_wakeup_zero(sma, q->sops, q->nsops, pt);
910 restart = check_restart(sma, q);
911 }
912
913 wake_up_sem_queue_prepare(pt, q, error);
914 if (restart)
915 goto again;
916 }
917 return semop_completed;
918}
919
920/**
921 * set_semotime - set sem_otime
922 * @sma: semaphore array
923 * @sops: operations that modified the array, may be NULL
924 *
925 * sem_otime is replicated to avoid cache line trashing.
926 * This function sets one instance to the current time.
927 */
928static void set_semotime(struct sem_array *sma, struct sembuf *sops)
929{
930 if (sops == NULL) {
931 sma->sem_base[0].sem_otime = get_seconds();
932 } else {
933 sma->sem_base[sops[0].sem_num].sem_otime =
934 get_seconds();
935 }
936}
937
938/**
939 * do_smart_update - optimized update_queue
940 * @sma: semaphore array
941 * @sops: operations that were performed
942 * @nsops: number of operations
943 * @otime: force setting otime
944 * @pt: list head of the tasks that must be woken up.
945 *
946 * do_smart_update() does the required calls to update_queue and wakeup_zero,
947 * based on the actual changes that were performed on the semaphore array.
948 * Note that the function does not do the actual wake-up: the caller is
949 * responsible for calling wake_up_sem_queue_do(@pt).
950 * It is safe to perform this call after dropping all locks.
951 */
952static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops,
953 int otime, struct list_head *pt)
954{
955 int i;
956
957 otime |= do_smart_wakeup_zero(sma, sops, nsops, pt);
958
959 if (!list_empty(&sma->pending_alter)) {
960 /* semaphore array uses the global queue - just process it. */
961 otime |= update_queue(sma, -1, pt);
962 } else {
963 if (!sops) {
964 /*
965 * No sops, thus the modified semaphores are not
966 * known. Check all.
967 */
968 for (i = 0; i < sma->sem_nsems; i++)
969 otime |= update_queue(sma, i, pt);
970 } else {
971 /*
972 * Check the semaphores that were increased:
973 * - No complex ops, thus all sleeping ops are
974 * decrease.
975 * - if we decreased the value, then any sleeping
976 * semaphore ops wont be able to run: If the
977 * previous value was too small, then the new
978 * value will be too small, too.
979 */
980 for (i = 0; i < nsops; i++) {
981 if (sops[i].sem_op > 0) {
982 otime |= update_queue(sma,
983 sops[i].sem_num, pt);
984 }
985 }
986 }
987 }
988 if (otime)
989 set_semotime(sma, sops);
990}
991
992/* The following counts are associated to each semaphore:
993 * semncnt number of tasks waiting on semval being nonzero
994 * semzcnt number of tasks waiting on semval being zero
995 * This model assumes that a task waits on exactly one semaphore.
996 * Since semaphore operations are to be performed atomically, tasks actually
997 * wait on a whole sequence of semaphores simultaneously.
998 * The counts we return here are a rough approximation, but still
999 * warrant that semncnt+semzcnt>0 if the task is on the pending queue.
1000 */
1001static int count_semncnt(struct sem_array *sma, ushort semnum)
1002{
1003 int semncnt;
1004 struct sem_queue *q;
1005
1006 semncnt = 0;
1007 list_for_each_entry(q, &sma->sem_base[semnum].pending_alter, list) {
1008 struct sembuf *sops = q->sops;
1009 BUG_ON(sops->sem_num != semnum);
1010 if ((sops->sem_op < 0) && !(sops->sem_flg & IPC_NOWAIT))
1011 semncnt++;
1012 }
1013
1014 list_for_each_entry(q, &sma->pending_alter, list) {
1015 struct sembuf *sops = q->sops;
1016 int nsops = q->nsops;
1017 int i;
1018 for (i = 0; i < nsops; i++)
1019 if (sops[i].sem_num == semnum
1020 && (sops[i].sem_op < 0)
1021 && !(sops[i].sem_flg & IPC_NOWAIT))
1022 semncnt++;
1023 }
1024 return semncnt;
1025}
1026
1027static int count_semzcnt(struct sem_array *sma, ushort semnum)
1028{
1029 int semzcnt;
1030 struct sem_queue *q;
1031
1032 semzcnt = 0;
1033 list_for_each_entry(q, &sma->sem_base[semnum].pending_const, list) {
1034 struct sembuf *sops = q->sops;
1035 BUG_ON(sops->sem_num != semnum);
1036 if ((sops->sem_op == 0) && !(sops->sem_flg & IPC_NOWAIT))
1037 semzcnt++;
1038 }
1039
1040 list_for_each_entry(q, &sma->pending_const, list) {
1041 struct sembuf *sops = q->sops;
1042 int nsops = q->nsops;
1043 int i;
1044 for (i = 0; i < nsops; i++)
1045 if (sops[i].sem_num == semnum
1046 && (sops[i].sem_op == 0)
1047 && !(sops[i].sem_flg & IPC_NOWAIT))
1048 semzcnt++;
1049 }
1050 return semzcnt;
1051}
1052
1053/* Free a semaphore set. freeary() is called with sem_ids.rwsem locked
1054 * as a writer and the spinlock for this semaphore set hold. sem_ids.rwsem
1055 * remains locked on exit.
1056 */
1057static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
1058{
1059 struct sem_undo *un, *tu;
1060 struct sem_queue *q, *tq;
1061 struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
1062 struct list_head tasks;
1063 int i;
1064
1065 /* Free the existing undo structures for this semaphore set. */
1066 ipc_assert_locked_object(&sma->sem_perm);
1067 list_for_each_entry_safe(un, tu, &sma->list_id, list_id) {
1068 list_del(&un->list_id);
1069 spin_lock(&un->ulp->lock);
1070 un->semid = -1;
1071 list_del_rcu(&un->list_proc);
1072 spin_unlock(&un->ulp->lock);
1073 kfree_rcu(un, rcu);
1074 }
1075
1076 /* Wake up all pending processes and let them fail with EIDRM. */
1077 INIT_LIST_HEAD(&tasks);
1078 list_for_each_entry_safe(q, tq, &sma->pending_const, list) {
1079 unlink_queue(sma, q);
1080 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1081 }
1082
1083 list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
1084 unlink_queue(sma, q);
1085 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1086 }
1087 for (i = 0; i < sma->sem_nsems; i++) {
1088 struct sem *sem = sma->sem_base + i;
1089 list_for_each_entry_safe(q, tq, &sem->pending_const, list) {
1090 unlink_queue(sma, q);
1091 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1092 }
1093 list_for_each_entry_safe(q, tq, &sem->pending_alter, list) {
1094 unlink_queue(sma, q);
1095 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1096 }
1097 }
1098
1099 /* Remove the semaphore set from the IDR */
1100 sem_rmid(ns, sma);
1101 sem_unlock(sma, -1);
1102 rcu_read_unlock();
1103
1104 wake_up_sem_queue_do(&tasks);
1105 ns->used_sems -= sma->sem_nsems;
1106 ipc_rcu_putref(sma, sem_rcu_free);
1107}
1108
1109static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
1110{
1111 switch (version) {
1112 case IPC_64:
1113 return copy_to_user(buf, in, sizeof(*in));
1114 case IPC_OLD:
1115 {
1116 struct semid_ds out;
1117
1118 memset(&out, 0, sizeof(out));
1119
1120 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
1121
1122 out.sem_otime = in->sem_otime;
1123 out.sem_ctime = in->sem_ctime;
1124 out.sem_nsems = in->sem_nsems;
1125
1126 return copy_to_user(buf, &out, sizeof(out));
1127 }
1128 default:
1129 return -EINVAL;
1130 }
1131}
1132
1133static time_t get_semotime(struct sem_array *sma)
1134{
1135 int i;
1136 time_t res;
1137
1138 res = sma->sem_base[0].sem_otime;
1139 for (i = 1; i < sma->sem_nsems; i++) {
1140 time_t to = sma->sem_base[i].sem_otime;
1141
1142 if (to > res)
1143 res = to;
1144 }
1145 return res;
1146}
1147
1148static int semctl_nolock(struct ipc_namespace *ns, int semid,
1149 int cmd, int version, void __user *p)
1150{
1151 int err;
1152 struct sem_array *sma;
1153
1154 switch (cmd) {
1155 case IPC_INFO:
1156 case SEM_INFO:
1157 {
1158 struct seminfo seminfo;
1159 int max_id;
1160
1161 err = security_sem_semctl(NULL, cmd);
1162 if (err)
1163 return err;
1164
1165 memset(&seminfo, 0, sizeof(seminfo));
1166 seminfo.semmni = ns->sc_semmni;
1167 seminfo.semmns = ns->sc_semmns;
1168 seminfo.semmsl = ns->sc_semmsl;
1169 seminfo.semopm = ns->sc_semopm;
1170 seminfo.semvmx = SEMVMX;
1171 seminfo.semmnu = SEMMNU;
1172 seminfo.semmap = SEMMAP;
1173 seminfo.semume = SEMUME;
1174 down_read(&sem_ids(ns).rwsem);
1175 if (cmd == SEM_INFO) {
1176 seminfo.semusz = sem_ids(ns).in_use;
1177 seminfo.semaem = ns->used_sems;
1178 } else {
1179 seminfo.semusz = SEMUSZ;
1180 seminfo.semaem = SEMAEM;
1181 }
1182 max_id = ipc_get_maxid(&sem_ids(ns));
1183 up_read(&sem_ids(ns).rwsem);
1184 if (copy_to_user(p, &seminfo, sizeof(struct seminfo)))
1185 return -EFAULT;
1186 return (max_id < 0) ? 0 : max_id;
1187 }
1188 case IPC_STAT:
1189 case SEM_STAT:
1190 {
1191 struct semid64_ds tbuf;
1192 int id = 0;
1193
1194 memset(&tbuf, 0, sizeof(tbuf));
1195
1196 rcu_read_lock();
1197 if (cmd == SEM_STAT) {
1198 sma = sem_obtain_object(ns, semid);
1199 if (IS_ERR(sma)) {
1200 err = PTR_ERR(sma);
1201 goto out_unlock;
1202 }
1203 id = sma->sem_perm.id;
1204 } else {
1205 sma = sem_obtain_object_check(ns, semid);
1206 if (IS_ERR(sma)) {
1207 err = PTR_ERR(sma);
1208 goto out_unlock;
1209 }
1210 }
1211
1212 err = -EACCES;
1213 if (ipcperms(ns, &sma->sem_perm, S_IRUGO))
1214 goto out_unlock;
1215
1216 err = security_sem_semctl(sma, cmd);
1217 if (err)
1218 goto out_unlock;
1219
1220 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
1221 tbuf.sem_otime = get_semotime(sma);
1222 tbuf.sem_ctime = sma->sem_ctime;
1223 tbuf.sem_nsems = sma->sem_nsems;
1224 rcu_read_unlock();
1225 if (copy_semid_to_user(p, &tbuf, version))
1226 return -EFAULT;
1227 return id;
1228 }
1229 default:
1230 return -EINVAL;
1231 }
1232out_unlock:
1233 rcu_read_unlock();
1234 return err;
1235}
1236
1237static int semctl_setval(struct ipc_namespace *ns, int semid, int semnum,
1238 unsigned long arg)
1239{
1240 struct sem_undo *un;
1241 struct sem_array *sma;
1242 struct sem *curr;
1243 int err;
1244 struct list_head tasks;
1245 int val;
1246#if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN)
1247 /* big-endian 64bit */
1248 val = arg >> 32;
1249#else
1250 /* 32bit or little-endian 64bit */
1251 val = arg;
1252#endif
1253
1254 if (val > SEMVMX || val < 0)
1255 return -ERANGE;
1256
1257 INIT_LIST_HEAD(&tasks);
1258
1259 rcu_read_lock();
1260 sma = sem_obtain_object_check(ns, semid);
1261 if (IS_ERR(sma)) {
1262 rcu_read_unlock();
1263 return PTR_ERR(sma);
1264 }
1265
1266 if (semnum < 0 || semnum >= sma->sem_nsems) {
1267 rcu_read_unlock();
1268 return -EINVAL;
1269 }
1270
1271
1272 if (ipcperms(ns, &sma->sem_perm, S_IWUGO)) {
1273 rcu_read_unlock();
1274 return -EACCES;
1275 }
1276
1277 err = security_sem_semctl(sma, SETVAL);
1278 if (err) {
1279 rcu_read_unlock();
1280 return -EACCES;
1281 }
1282
1283 sem_lock(sma, NULL, -1);
1284
1285 if (!ipc_valid_object(&sma->sem_perm)) {
1286 sem_unlock(sma, -1);
1287 rcu_read_unlock();
1288 return -EIDRM;
1289 }
1290
1291 curr = &sma->sem_base[semnum];
1292
1293 ipc_assert_locked_object(&sma->sem_perm);
1294 list_for_each_entry(un, &sma->list_id, list_id)
1295 un->semadj[semnum] = 0;
1296
1297 curr->semval = val;
1298 curr->sempid = task_tgid_vnr(current);
1299 sma->sem_ctime = get_seconds();
1300 /* maybe some queued-up processes were waiting for this */
1301 do_smart_update(sma, NULL, 0, 0, &tasks);
1302 sem_unlock(sma, -1);
1303 rcu_read_unlock();
1304 wake_up_sem_queue_do(&tasks);
1305 return 0;
1306}
1307
1308static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
1309 int cmd, void __user *p)
1310{
1311 struct sem_array *sma;
1312 struct sem *curr;
1313 int err, nsems;
1314 ushort fast_sem_io[SEMMSL_FAST];
1315 ushort *sem_io = fast_sem_io;
1316 struct list_head tasks;
1317
1318 INIT_LIST_HEAD(&tasks);
1319
1320 rcu_read_lock();
1321 sma = sem_obtain_object_check(ns, semid);
1322 if (IS_ERR(sma)) {
1323 rcu_read_unlock();
1324 return PTR_ERR(sma);
1325 }
1326
1327 nsems = sma->sem_nsems;
1328
1329 err = -EACCES;
1330 if (ipcperms(ns, &sma->sem_perm, cmd == SETALL ? S_IWUGO : S_IRUGO))
1331 goto out_rcu_wakeup;
1332
1333 err = security_sem_semctl(sma, cmd);
1334 if (err)
1335 goto out_rcu_wakeup;
1336
1337 err = -EACCES;
1338 switch (cmd) {
1339 case GETALL:
1340 {
1341 ushort __user *array = p;
1342 int i;
1343
1344 sem_lock(sma, NULL, -1);
1345 if (!ipc_valid_object(&sma->sem_perm)) {
1346 err = -EIDRM;
1347 goto out_unlock;
1348 }
1349 if (nsems > SEMMSL_FAST) {
1350 if (!ipc_rcu_getref(sma)) {
1351 err = -EIDRM;
1352 goto out_unlock;
1353 }
1354 sem_unlock(sma, -1);
1355 rcu_read_unlock();
1356 sem_io = ipc_alloc(sizeof(ushort)*nsems);
1357 if (sem_io == NULL) {
1358 ipc_rcu_putref(sma, ipc_rcu_free);
1359 return -ENOMEM;
1360 }
1361
1362 rcu_read_lock();
1363 sem_lock_and_putref(sma);
1364 if (!ipc_valid_object(&sma->sem_perm)) {
1365 err = -EIDRM;
1366 goto out_unlock;
1367 }
1368 }
1369 for (i = 0; i < sma->sem_nsems; i++)
1370 sem_io[i] = sma->sem_base[i].semval;
1371 sem_unlock(sma, -1);
1372 rcu_read_unlock();
1373 err = 0;
1374 if (copy_to_user(array, sem_io, nsems*sizeof(ushort)))
1375 err = -EFAULT;
1376 goto out_free;
1377 }
1378 case SETALL:
1379 {
1380 int i;
1381 struct sem_undo *un;
1382
1383 if (!ipc_rcu_getref(sma)) {
1384 err = -EIDRM;
1385 goto out_rcu_wakeup;
1386 }
1387 rcu_read_unlock();
1388
1389 if (nsems > SEMMSL_FAST) {
1390 sem_io = ipc_alloc(sizeof(ushort)*nsems);
1391 if (sem_io == NULL) {
1392 ipc_rcu_putref(sma, ipc_rcu_free);
1393 return -ENOMEM;
1394 }
1395 }
1396
1397 if (copy_from_user(sem_io, p, nsems*sizeof(ushort))) {
1398 ipc_rcu_putref(sma, ipc_rcu_free);
1399 err = -EFAULT;
1400 goto out_free;
1401 }
1402
1403 for (i = 0; i < nsems; i++) {
1404 if (sem_io[i] > SEMVMX) {
1405 ipc_rcu_putref(sma, ipc_rcu_free);
1406 err = -ERANGE;
1407 goto out_free;
1408 }
1409 }
1410 rcu_read_lock();
1411 sem_lock_and_putref(sma);
1412 if (!ipc_valid_object(&sma->sem_perm)) {
1413 err = -EIDRM;
1414 goto out_unlock;
1415 }
1416
1417 for (i = 0; i < nsems; i++)
1418 sma->sem_base[i].semval = sem_io[i];
1419
1420 ipc_assert_locked_object(&sma->sem_perm);
1421 list_for_each_entry(un, &sma->list_id, list_id) {
1422 for (i = 0; i < nsems; i++)
1423 un->semadj[i] = 0;
1424 }
1425 sma->sem_ctime = get_seconds();
1426 /* maybe some queued-up processes were waiting for this */
1427 do_smart_update(sma, NULL, 0, 0, &tasks);
1428 err = 0;
1429 goto out_unlock;
1430 }
1431 /* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */
1432 }
1433 err = -EINVAL;
1434 if (semnum < 0 || semnum >= nsems)
1435 goto out_rcu_wakeup;
1436
1437 sem_lock(sma, NULL, -1);
1438 if (!ipc_valid_object(&sma->sem_perm)) {
1439 err = -EIDRM;
1440 goto out_unlock;
1441 }
1442 curr = &sma->sem_base[semnum];
1443
1444 switch (cmd) {
1445 case GETVAL:
1446 err = curr->semval;
1447 goto out_unlock;
1448 case GETPID:
1449 err = curr->sempid;
1450 goto out_unlock;
1451 case GETNCNT:
1452 err = count_semncnt(sma, semnum);
1453 goto out_unlock;
1454 case GETZCNT:
1455 err = count_semzcnt(sma, semnum);
1456 goto out_unlock;
1457 }
1458
1459out_unlock:
1460 sem_unlock(sma, -1);
1461out_rcu_wakeup:
1462 rcu_read_unlock();
1463 wake_up_sem_queue_do(&tasks);
1464out_free:
1465 if (sem_io != fast_sem_io)
1466 ipc_free(sem_io, sizeof(ushort)*nsems);
1467 return err;
1468}
1469
1470static inline unsigned long
1471copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
1472{
1473 switch (version) {
1474 case IPC_64:
1475 if (copy_from_user(out, buf, sizeof(*out)))
1476 return -EFAULT;
1477 return 0;
1478 case IPC_OLD:
1479 {
1480 struct semid_ds tbuf_old;
1481
1482 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
1483 return -EFAULT;
1484
1485 out->sem_perm.uid = tbuf_old.sem_perm.uid;
1486 out->sem_perm.gid = tbuf_old.sem_perm.gid;
1487 out->sem_perm.mode = tbuf_old.sem_perm.mode;
1488
1489 return 0;
1490 }
1491 default:
1492 return -EINVAL;
1493 }
1494}
1495
1496/*
1497 * This function handles some semctl commands which require the rwsem
1498 * to be held in write mode.
1499 * NOTE: no locks must be held, the rwsem is taken inside this function.
1500 */
1501static int semctl_down(struct ipc_namespace *ns, int semid,
1502 int cmd, int version, void __user *p)
1503{
1504 struct sem_array *sma;
1505 int err;
1506 struct semid64_ds semid64;
1507 struct kern_ipc_perm *ipcp;
1508
1509 if (cmd == IPC_SET) {
1510 if (copy_semid_from_user(&semid64, p, version))
1511 return -EFAULT;
1512 }
1513
1514 down_write(&sem_ids(ns).rwsem);
1515 rcu_read_lock();
1516
1517 ipcp = ipcctl_pre_down_nolock(ns, &sem_ids(ns), semid, cmd,
1518 &semid64.sem_perm, 0);
1519 if (IS_ERR(ipcp)) {
1520 err = PTR_ERR(ipcp);
1521 goto out_unlock1;
1522 }
1523
1524 sma = container_of(ipcp, struct sem_array, sem_perm);
1525
1526 err = security_sem_semctl(sma, cmd);
1527 if (err)
1528 goto out_unlock1;
1529
1530 switch (cmd) {
1531 case IPC_RMID:
1532 sem_lock(sma, NULL, -1);
1533 /* freeary unlocks the ipc object and rcu */
1534 freeary(ns, ipcp);
1535 goto out_up;
1536 case IPC_SET:
1537 sem_lock(sma, NULL, -1);
1538 err = ipc_update_perm(&semid64.sem_perm, ipcp);
1539 if (err)
1540 goto out_unlock0;
1541 sma->sem_ctime = get_seconds();
1542 break;
1543 default:
1544 err = -EINVAL;
1545 goto out_unlock1;
1546 }
1547
1548out_unlock0:
1549 sem_unlock(sma, -1);
1550out_unlock1:
1551 rcu_read_unlock();
1552out_up:
1553 up_write(&sem_ids(ns).rwsem);
1554 return err;
1555}
1556
1557SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, unsigned long, arg)
1558{
1559 int version;
1560 struct ipc_namespace *ns;
1561 void __user *p = (void __user *)arg;
1562
1563 if (semid < 0)
1564 return -EINVAL;
1565
1566 version = ipc_parse_version(&cmd);
1567 ns = current->nsproxy->ipc_ns;
1568
1569 switch (cmd) {
1570 case IPC_INFO:
1571 case SEM_INFO:
1572 case IPC_STAT:
1573 case SEM_STAT:
1574 return semctl_nolock(ns, semid, cmd, version, p);
1575 case GETALL:
1576 case GETVAL:
1577 case GETPID:
1578 case GETNCNT:
1579 case GETZCNT:
1580 case SETALL:
1581 return semctl_main(ns, semid, semnum, cmd, p);
1582 case SETVAL:
1583 return semctl_setval(ns, semid, semnum, arg);
1584 case IPC_RMID:
1585 case IPC_SET:
1586 return semctl_down(ns, semid, cmd, version, p);
1587 default:
1588 return -EINVAL;
1589 }
1590}
1591
1592/* If the task doesn't already have a undo_list, then allocate one
1593 * here. We guarantee there is only one thread using this undo list,
1594 * and current is THE ONE
1595 *
1596 * If this allocation and assignment succeeds, but later
1597 * portions of this code fail, there is no need to free the sem_undo_list.
1598 * Just let it stay associated with the task, and it'll be freed later
1599 * at exit time.
1600 *
1601 * This can block, so callers must hold no locks.
1602 */
1603static inline int get_undo_list(struct sem_undo_list **undo_listp)
1604{
1605 struct sem_undo_list *undo_list;
1606
1607 undo_list = current->sysvsem.undo_list;
1608 if (!undo_list) {
1609 undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
1610 if (undo_list == NULL)
1611 return -ENOMEM;
1612 spin_lock_init(&undo_list->lock);
1613 atomic_set(&undo_list->refcnt, 1);
1614 INIT_LIST_HEAD(&undo_list->list_proc);
1615
1616 current->sysvsem.undo_list = undo_list;
1617 }
1618 *undo_listp = undo_list;
1619 return 0;
1620}
1621
1622static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid)
1623{
1624 struct sem_undo *un;
1625
1626 list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) {
1627 if (un->semid == semid)
1628 return un;
1629 }
1630 return NULL;
1631}
1632
1633static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1634{
1635 struct sem_undo *un;
1636
1637 assert_spin_locked(&ulp->lock);
1638
1639 un = __lookup_undo(ulp, semid);
1640 if (un) {
1641 list_del_rcu(&un->list_proc);
1642 list_add_rcu(&un->list_proc, &ulp->list_proc);
1643 }
1644 return un;
1645}
1646
1647/**
1648 * find_alloc_undo - lookup (and if not present create) undo array
1649 * @ns: namespace
1650 * @semid: semaphore array id
1651 *
1652 * The function looks up (and if not present creates) the undo structure.
1653 * The size of the undo structure depends on the size of the semaphore
1654 * array, thus the alloc path is not that straightforward.
1655 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1656 * performs a rcu_read_lock().
1657 */
1658static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
1659{
1660 struct sem_array *sma;
1661 struct sem_undo_list *ulp;
1662 struct sem_undo *un, *new;
1663 int nsems, error;
1664
1665 error = get_undo_list(&ulp);
1666 if (error)
1667 return ERR_PTR(error);
1668
1669 rcu_read_lock();
1670 spin_lock(&ulp->lock);
1671 un = lookup_undo(ulp, semid);
1672 spin_unlock(&ulp->lock);
1673 if (likely(un != NULL))
1674 goto out;
1675
1676 /* no undo structure around - allocate one. */
1677 /* step 1: figure out the size of the semaphore array */
1678 sma = sem_obtain_object_check(ns, semid);
1679 if (IS_ERR(sma)) {
1680 rcu_read_unlock();
1681 return ERR_CAST(sma);
1682 }
1683
1684 nsems = sma->sem_nsems;
1685 if (!ipc_rcu_getref(sma)) {
1686 rcu_read_unlock();
1687 un = ERR_PTR(-EIDRM);
1688 goto out;
1689 }
1690 rcu_read_unlock();
1691
1692 /* step 2: allocate new undo structure */
1693 new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1694 if (!new) {
1695 ipc_rcu_putref(sma, ipc_rcu_free);
1696 return ERR_PTR(-ENOMEM);
1697 }
1698
1699 /* step 3: Acquire the lock on semaphore array */
1700 rcu_read_lock();
1701 sem_lock_and_putref(sma);
1702 if (!ipc_valid_object(&sma->sem_perm)) {
1703 sem_unlock(sma, -1);
1704 rcu_read_unlock();
1705 kfree(new);
1706 un = ERR_PTR(-EIDRM);
1707 goto out;
1708 }
1709 spin_lock(&ulp->lock);
1710
1711 /*
1712 * step 4: check for races: did someone else allocate the undo struct?
1713 */
1714 un = lookup_undo(ulp, semid);
1715 if (un) {
1716 kfree(new);
1717 goto success;
1718 }
1719 /* step 5: initialize & link new undo structure */
1720 new->semadj = (short *) &new[1];
1721 new->ulp = ulp;
1722 new->semid = semid;
1723 assert_spin_locked(&ulp->lock);
1724 list_add_rcu(&new->list_proc, &ulp->list_proc);
1725 ipc_assert_locked_object(&sma->sem_perm);
1726 list_add(&new->list_id, &sma->list_id);
1727 un = new;
1728
1729success:
1730 spin_unlock(&ulp->lock);
1731 sem_unlock(sma, -1);
1732out:
1733 return un;
1734}
1735
1736
1737/**
1738 * get_queue_result - retrieve the result code from sem_queue
1739 * @q: Pointer to queue structure
1740 *
1741 * Retrieve the return code from the pending queue. If IN_WAKEUP is found in
1742 * q->status, then we must loop until the value is replaced with the final
1743 * value: This may happen if a task is woken up by an unrelated event (e.g.
1744 * signal) and in parallel the task is woken up by another task because it got
1745 * the requested semaphores.
1746 *
1747 * The function can be called with or without holding the semaphore spinlock.
1748 */
1749static int get_queue_result(struct sem_queue *q)
1750{
1751 int error;
1752
1753 error = q->status;
1754 while (unlikely(error == IN_WAKEUP)) {
1755 cpu_relax();
1756 error = q->status;
1757 }
1758
1759 return error;
1760}
1761
1762SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
1763 unsigned, nsops, const struct timespec __user *, timeout)
1764{
1765 int error = -EINVAL;
1766 struct sem_array *sma;
1767 struct sembuf fast_sops[SEMOPM_FAST];
1768 struct sembuf *sops = fast_sops, *sop;
1769 struct sem_undo *un;
1770 int undos = 0, alter = 0, max, locknum;
1771 struct sem_queue queue;
1772 unsigned long jiffies_left = 0;
1773 struct ipc_namespace *ns;
1774 struct list_head tasks;
1775
1776 ns = current->nsproxy->ipc_ns;
1777
1778 if (nsops < 1 || semid < 0)
1779 return -EINVAL;
1780 if (nsops > ns->sc_semopm)
1781 return -E2BIG;
1782 if (nsops > SEMOPM_FAST) {
1783 sops = kmalloc(sizeof(*sops)*nsops, GFP_KERNEL);
1784 if (sops == NULL)
1785 return -ENOMEM;
1786 }
1787 if (copy_from_user(sops, tsops, nsops * sizeof(*tsops))) {
1788 error = -EFAULT;
1789 goto out_free;
1790 }
1791 if (timeout) {
1792 struct timespec _timeout;
1793 if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1794 error = -EFAULT;
1795 goto out_free;
1796 }
1797 if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1798 _timeout.tv_nsec >= 1000000000L) {
1799 error = -EINVAL;
1800 goto out_free;
1801 }
1802 jiffies_left = timespec_to_jiffies(&_timeout);
1803 }
1804 max = 0;
1805 for (sop = sops; sop < sops + nsops; sop++) {
1806 if (sop->sem_num >= max)
1807 max = sop->sem_num;
1808 if (sop->sem_flg & SEM_UNDO)
1809 undos = 1;
1810 if (sop->sem_op != 0)
1811 alter = 1;
1812 }
1813
1814 INIT_LIST_HEAD(&tasks);
1815
1816 if (undos) {
1817 /* On success, find_alloc_undo takes the rcu_read_lock */
1818 un = find_alloc_undo(ns, semid);
1819 if (IS_ERR(un)) {
1820 error = PTR_ERR(un);
1821 goto out_free;
1822 }
1823 } else {
1824 un = NULL;
1825 rcu_read_lock();
1826 }
1827
1828 sma = sem_obtain_object_check(ns, semid);
1829 if (IS_ERR(sma)) {
1830 rcu_read_unlock();
1831 error = PTR_ERR(sma);
1832 goto out_free;
1833 }
1834
1835 error = -EFBIG;
1836 if (max >= sma->sem_nsems)
1837 goto out_rcu_wakeup;
1838
1839 error = -EACCES;
1840 if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO))
1841 goto out_rcu_wakeup;
1842
1843 error = security_sem_semop(sma, sops, nsops, alter);
1844 if (error)
1845 goto out_rcu_wakeup;
1846
1847 error = -EIDRM;
1848 locknum = sem_lock(sma, sops, nsops);
1849 /*
1850 * We eventually might perform the following check in a lockless
1851 * fashion, considering ipc_valid_object() locking constraints.
1852 * If nsops == 1 and there is no contention for sem_perm.lock, then
1853 * only a per-semaphore lock is held and it's OK to proceed with the
1854 * check below. More details on the fine grained locking scheme
1855 * entangled here and why it's RMID race safe on comments at sem_lock()
1856 */
1857 if (!ipc_valid_object(&sma->sem_perm))
1858 goto out_unlock_free;
1859 /*
1860 * semid identifiers are not unique - find_alloc_undo may have
1861 * allocated an undo structure, it was invalidated by an RMID
1862 * and now a new array with received the same id. Check and fail.
1863 * This case can be detected checking un->semid. The existence of
1864 * "un" itself is guaranteed by rcu.
1865 */
1866 if (un && un->semid == -1)
1867 goto out_unlock_free;
1868
1869 error = perform_atomic_semop(sma, sops, nsops, un,
1870 task_tgid_vnr(current));
1871 if (error == 0) {
1872 /* If the operation was successful, then do
1873 * the required updates.
1874 */
1875 if (alter)
1876 do_smart_update(sma, sops, nsops, 1, &tasks);
1877 else
1878 set_semotime(sma, sops);
1879 }
1880 if (error <= 0)
1881 goto out_unlock_free;
1882
1883 /* We need to sleep on this operation, so we put the current
1884 * task into the pending queue and go to sleep.
1885 */
1886
1887 queue.sops = sops;
1888 queue.nsops = nsops;
1889 queue.undo = un;
1890 queue.pid = task_tgid_vnr(current);
1891 queue.alter = alter;
1892
1893 if (nsops == 1) {
1894 struct sem *curr;
1895 curr = &sma->sem_base[sops->sem_num];
1896
1897 if (alter) {
1898 if (sma->complex_count) {
1899 list_add_tail(&queue.list,
1900 &sma->pending_alter);
1901 } else {
1902
1903 list_add_tail(&queue.list,
1904 &curr->pending_alter);
1905 }
1906 } else {
1907 list_add_tail(&queue.list, &curr->pending_const);
1908 }
1909 } else {
1910 if (!sma->complex_count)
1911 merge_queues(sma);
1912
1913 if (alter)
1914 list_add_tail(&queue.list, &sma->pending_alter);
1915 else
1916 list_add_tail(&queue.list, &sma->pending_const);
1917
1918 sma->complex_count++;
1919 }
1920
1921 queue.status = -EINTR;
1922 queue.sleeper = current;
1923
1924sleep_again:
1925 current->state = TASK_INTERRUPTIBLE;
1926 sem_unlock(sma, locknum);
1927 rcu_read_unlock();
1928
1929 if (timeout)
1930 jiffies_left = schedule_timeout(jiffies_left);
1931 else
1932 schedule();
1933
1934 error = get_queue_result(&queue);
1935
1936 if (error != -EINTR) {
1937 /* fast path: update_queue already obtained all requested
1938 * resources.
1939 * Perform a smp_mb(): User space could assume that semop()
1940 * is a memory barrier: Without the mb(), the cpu could
1941 * speculatively read in user space stale data that was
1942 * overwritten by the previous owner of the semaphore.
1943 */
1944 smp_mb();
1945
1946 goto out_free;
1947 }
1948
1949 rcu_read_lock();
1950 sma = sem_obtain_lock(ns, semid, sops, nsops, &locknum);
1951
1952 /*
1953 * Wait until it's guaranteed that no wakeup_sem_queue_do() is ongoing.
1954 */
1955 error = get_queue_result(&queue);
1956
1957 /*
1958 * Array removed? If yes, leave without sem_unlock().
1959 */
1960 if (IS_ERR(sma)) {
1961 rcu_read_unlock();
1962 goto out_free;
1963 }
1964
1965
1966 /*
1967 * If queue.status != -EINTR we are woken up by another process.
1968 * Leave without unlink_queue(), but with sem_unlock().
1969 */
1970 if (error != -EINTR)
1971 goto out_unlock_free;
1972
1973 /*
1974 * If an interrupt occurred we have to clean up the queue
1975 */
1976 if (timeout && jiffies_left == 0)
1977 error = -EAGAIN;
1978
1979 /*
1980 * If the wakeup was spurious, just retry
1981 */
1982 if (error == -EINTR && !signal_pending(current))
1983 goto sleep_again;
1984
1985 unlink_queue(sma, &queue);
1986
1987out_unlock_free:
1988 sem_unlock(sma, locknum);
1989out_rcu_wakeup:
1990 rcu_read_unlock();
1991 wake_up_sem_queue_do(&tasks);
1992out_free:
1993 if (sops != fast_sops)
1994 kfree(sops);
1995 return error;
1996}
1997
1998SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
1999 unsigned, nsops)
2000{
2001 return sys_semtimedop(semid, tsops, nsops, NULL);
2002}
2003
2004/* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
2005 * parent and child tasks.
2006 */
2007
2008int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
2009{
2010 struct sem_undo_list *undo_list;
2011 int error;
2012
2013 if (clone_flags & CLONE_SYSVSEM) {
2014 error = get_undo_list(&undo_list);
2015 if (error)
2016 return error;
2017 atomic_inc(&undo_list->refcnt);
2018 tsk->sysvsem.undo_list = undo_list;
2019 } else
2020 tsk->sysvsem.undo_list = NULL;
2021
2022 return 0;
2023}
2024
2025/*
2026 * add semadj values to semaphores, free undo structures.
2027 * undo structures are not freed when semaphore arrays are destroyed
2028 * so some of them may be out of date.
2029 * IMPLEMENTATION NOTE: There is some confusion over whether the
2030 * set of adjustments that needs to be done should be done in an atomic
2031 * manner or not. That is, if we are attempting to decrement the semval
2032 * should we queue up and wait until we can do so legally?
2033 * The original implementation attempted to do this (queue and wait).
2034 * The current implementation does not do so. The POSIX standard
2035 * and SVID should be consulted to determine what behavior is mandated.
2036 */
2037void exit_sem(struct task_struct *tsk)
2038{
2039 struct sem_undo_list *ulp;
2040
2041 ulp = tsk->sysvsem.undo_list;
2042 if (!ulp)
2043 return;
2044 tsk->sysvsem.undo_list = NULL;
2045
2046 if (!atomic_dec_and_test(&ulp->refcnt))
2047 return;
2048
2049 for (;;) {
2050 struct sem_array *sma;
2051 struct sem_undo *un;
2052 struct list_head tasks;
2053 int semid, i;
2054
2055 rcu_read_lock();
2056 un = list_entry_rcu(ulp->list_proc.next,
2057 struct sem_undo, list_proc);
2058 if (&un->list_proc == &ulp->list_proc)
2059 semid = -1;
2060 else
2061 semid = un->semid;
2062
2063 if (semid == -1) {
2064 rcu_read_unlock();
2065 break;
2066 }
2067
2068 sma = sem_obtain_object_check(tsk->nsproxy->ipc_ns, un->semid);
2069 /* exit_sem raced with IPC_RMID, nothing to do */
2070 if (IS_ERR(sma)) {
2071 rcu_read_unlock();
2072 continue;
2073 }
2074
2075 sem_lock(sma, NULL, -1);
2076 /* exit_sem raced with IPC_RMID, nothing to do */
2077 if (!ipc_valid_object(&sma->sem_perm)) {
2078 sem_unlock(sma, -1);
2079 rcu_read_unlock();
2080 continue;
2081 }
2082 un = __lookup_undo(ulp, semid);
2083 if (un == NULL) {
2084 /* exit_sem raced with IPC_RMID+semget() that created
2085 * exactly the same semid. Nothing to do.
2086 */
2087 sem_unlock(sma, -1);
2088 rcu_read_unlock();
2089 continue;
2090 }
2091
2092 /* remove un from the linked lists */
2093 ipc_assert_locked_object(&sma->sem_perm);
2094 list_del(&un->list_id);
2095
2096 spin_lock(&ulp->lock);
2097 list_del_rcu(&un->list_proc);
2098 spin_unlock(&ulp->lock);
2099
2100 /* perform adjustments registered in un */
2101 for (i = 0; i < sma->sem_nsems; i++) {
2102 struct sem *semaphore = &sma->sem_base[i];
2103 if (un->semadj[i]) {
2104 semaphore->semval += un->semadj[i];
2105 /*
2106 * Range checks of the new semaphore value,
2107 * not defined by sus:
2108 * - Some unices ignore the undo entirely
2109 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
2110 * - some cap the value (e.g. FreeBSD caps
2111 * at 0, but doesn't enforce SEMVMX)
2112 *
2113 * Linux caps the semaphore value, both at 0
2114 * and at SEMVMX.
2115 *
2116 * Manfred <manfred@colorfullife.com>
2117 */
2118 if (semaphore->semval < 0)
2119 semaphore->semval = 0;
2120 if (semaphore->semval > SEMVMX)
2121 semaphore->semval = SEMVMX;
2122 semaphore->sempid = task_tgid_vnr(current);
2123 }
2124 }
2125 /* maybe some queued-up processes were waiting for this */
2126 INIT_LIST_HEAD(&tasks);
2127 do_smart_update(sma, NULL, 0, 1, &tasks);
2128 sem_unlock(sma, -1);
2129 rcu_read_unlock();
2130 wake_up_sem_queue_do(&tasks);
2131
2132 kfree_rcu(un, rcu);
2133 }
2134 kfree(ulp);
2135}
2136
2137#ifdef CONFIG_PROC_FS
2138static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
2139{
2140 struct user_namespace *user_ns = seq_user_ns(s);
2141 struct sem_array *sma = it;
2142 time_t sem_otime;
2143
2144 /*
2145 * The proc interface isn't aware of sem_lock(), it calls
2146 * ipc_lock_object() directly (in sysvipc_find_ipc).
2147 * In order to stay compatible with sem_lock(), we must wait until
2148 * all simple semop() calls have left their critical regions.
2149 */
2150 sem_wait_array(sma);
2151
2152 sem_otime = get_semotime(sma);
2153
2154 return seq_printf(s,
2155 "%10d %10d %4o %10u %5u %5u %5u %5u %10lu %10lu\n",
2156 sma->sem_perm.key,
2157 sma->sem_perm.id,
2158 sma->sem_perm.mode,
2159 sma->sem_nsems,
2160 from_kuid_munged(user_ns, sma->sem_perm.uid),
2161 from_kgid_munged(user_ns, sma->sem_perm.gid),
2162 from_kuid_munged(user_ns, sma->sem_perm.cuid),
2163 from_kgid_munged(user_ns, sma->sem_perm.cgid),
2164 sem_otime,
2165 sma->sem_ctime);
2166}
2167#endif