blob: d1a6edd17eba2ce2f1e3bf83cfead51a6e6117a5 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/ipc/sem.c
3 * Copyright (C) 1992 Krishna Balasubramanian
4 * Copyright (C) 1995 Eric Schenk, Bruno Haible
5 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07006 * /proc/sysvipc/sem support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
7 *
8 * SMP-threaded, sysctl's added
Christian Kujau624dffc2006-01-15 02:43:54 +01009 * (c) 1999 Manfred Spraul <manfred@colorfullife.com>
Linus Torvalds1da177e2005-04-16 15:20:36 -070010 * Enforced range limit on SEM_UNDO
Alan Cox046c6882009-01-05 14:06:29 +000011 * (c) 2001 Red Hat Inc
Linus Torvalds1da177e2005-04-16 15:20:36 -070012 * Lockless wakeup
13 * (c) 2003 Manfred Spraul <manfred@colorfullife.com>
Manfred Spraulc5cf6352010-05-26 14:43:43 -070014 * Further wakeup optimizations, documentation
15 * (c) 2010 Manfred Spraul <manfred@colorfullife.com>
Steve Grubb073115d2006-04-02 17:07:33 -040016 *
17 * support for audit of ipc object properties and permission changes
18 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
Kirill Korotaeve3893532006-10-02 02:18:22 -070019 *
20 * namespaces support
21 * OpenVZ, SWsoft Inc.
22 * Pavel Emelianov <xemul@openvz.org>
Manfred Spraulc5cf6352010-05-26 14:43:43 -070023 *
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.
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -070050 * - semncnt and semzcnt are calculated on demand in count_semcnt()
Manfred Spraulc5cf6352010-05-26 14:43:43 -070051 * - the task that performs a successful semop() scans the list of all
52 * sleeping tasks and completes any pending operations that can be fulfilled.
53 * Semaphores are actively given to waiting tasks (necessary for FIFO).
54 * (see update_queue())
55 * - To improve the scalability, the actual wake-up calls are performed after
56 * dropping all locks. (see wake_up_sem_queue_prepare(),
57 * wake_up_sem_queue_do())
58 * - All work is done by the waker, the woken up task does not have to do
59 * anything - not even acquiring a lock or dropping a refcount.
60 * - A woken up task may not even touch the semaphore array anymore, it may
61 * have been destroyed already by a semctl(RMID).
62 * - The synchronizations between wake-ups due to a timeout/signal and a
63 * wake-up due to a completed semaphore operation is achieved by using an
64 * intermediate state (IN_WAKEUP).
65 * - UNDO values are stored in an array (one per process and per
66 * semaphore array, lazily allocated). For backwards compatibility, multiple
67 * modes for the UNDO variables are supported (per process, per thread)
68 * (see copy_semundo, CLONE_SYSVSEM)
69 * - There are two lists of the pending operations: a per-array list
70 * and per-semaphore list (stored in the array). This allows to achieve FIFO
71 * ordering without always scanning all pending operations.
72 * The worst-case behavior is nevertheless O(N^2) for N wakeups.
Linus Torvalds1da177e2005-04-16 15:20:36 -070073 */
74
Linus Torvalds1da177e2005-04-16 15:20:36 -070075#include <linux/slab.h>
76#include <linux/spinlock.h>
77#include <linux/init.h>
78#include <linux/proc_fs.h>
79#include <linux/time.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070080#include <linux/security.h>
81#include <linux/syscalls.h>
82#include <linux/audit.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080083#include <linux/capability.h>
Mike Waychison19b49462005-09-06 15:17:10 -070084#include <linux/seq_file.h>
Nadia Derbey3e148c72007-10-18 23:40:54 -070085#include <linux/rwsem.h>
Kirill Korotaeve3893532006-10-02 02:18:22 -070086#include <linux/nsproxy.h>
Pavel Emelyanovae5e1b22008-02-08 04:18:22 -080087#include <linux/ipc_namespace.h>
Ingo Molnar5f921ae2006-03-26 01:37:17 -080088
Paul McQuade7153e402014-06-06 14:37:37 -070089#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070090#include "util.h"
91
Manfred Spraule57940d2011-11-02 13:38:54 -070092/* One semaphore structure for each semaphore in the system. */
93struct sem {
94 int semval; /* current value */
95 int sempid; /* pid of last operation */
Rik van Riel6062a8d2013-04-30 19:15:44 -070096 spinlock_t lock; /* spinlock for fine-grained semtimedop */
Manfred Spraul1a82e9e2013-07-08 16:01:23 -070097 struct list_head pending_alter; /* pending single-sop operations */
98 /* that alter the semaphore */
99 struct list_head pending_const; /* pending single-sop operations */
100 /* that do not alter the semaphore*/
Manfred Sprauld12e1e52013-07-08 16:01:25 -0700101 time_t sem_otime; /* candidate for sem_otime */
Manfred Spraulf5c936c2013-07-08 16:01:22 -0700102} ____cacheline_aligned_in_smp;
Manfred Spraule57940d2011-11-02 13:38:54 -0700103
104/* One queue for each sleeping process in the system. */
105struct sem_queue {
Manfred Spraule57940d2011-11-02 13:38:54 -0700106 struct list_head list; /* queue of pending operations */
107 struct task_struct *sleeper; /* this process */
108 struct sem_undo *undo; /* undo structure */
109 int pid; /* process id of requesting process */
110 int status; /* completion status of operation */
111 struct sembuf *sops; /* array of pending operations */
Manfred Sprauled247b72014-06-06 14:37:49 -0700112 struct sembuf *blocking; /* the operation that blocked */
Manfred Spraule57940d2011-11-02 13:38:54 -0700113 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
Pierre Peiffered2ddbf2008-02-08 04:18:57 -0800143#define sem_ids(ns) ((ns)->ids[IPC_SEM_IDS])
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144
Nadia Derbey1b531f22007-10-18 23:40:55 -0700145#define sem_checkid(sma, semid) ipc_checkid(&sma->sem_perm, semid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700147static int newary(struct ipc_namespace *, struct ipc_params *);
Pierre Peiffer01b8b072008-02-08 04:18:57 -0800148static void freeary(struct ipc_namespace *, struct kern_ipc_perm *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149#ifdef CONFIG_PROC_FS
Mike Waychison19b49462005-09-06 15:17:10 -0700150static int sysvipc_sem_proc_show(struct seq_file *s, void *it);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151#endif
152
153#define SEMMSL_FAST 256 /* 512 bytes on stack */
154#define SEMOPM_FAST 64 /* ~ 372 bytes on stack */
155
156/*
Manfred Spraul758a6ba32013-07-08 16:01:26 -0700157 * Locking:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 * sem_undo.id_next,
Manfred Spraul758a6ba32013-07-08 16:01:26 -0700159 * sem_array.complex_count,
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700160 * sem_array.pending{_alter,_cont},
Manfred Spraul758a6ba32013-07-08 16:01:26 -0700161 * sem_array.sem_undo: global sem_lock() for read/write
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 * sem_undo.proc_next: only "current" is allowed to read/write that field.
Paul McQuade46c0a8c2014-06-06 14:37:37 -0700163 *
Manfred Spraul758a6ba32013-07-08 16:01:26 -0700164 * sem_array.sem_base[i].pending_{const,alter}:
165 * global or semaphore sem_lock() for read/write
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166 */
167
Kirill Korotaeve3893532006-10-02 02:18:22 -0700168#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]
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172
Pierre Peiffered2ddbf2008-02-08 04:18:57 -0800173void sem_init_ns(struct ipc_namespace *ns)
Kirill Korotaeve3893532006-10-02 02:18:22 -0700174{
Kirill Korotaeve3893532006-10-02 02:18:22 -0700175 ns->sc_semmsl = SEMMSL;
176 ns->sc_semmns = SEMMNS;
177 ns->sc_semopm = SEMOPM;
178 ns->sc_semmni = SEMMNI;
179 ns->used_sems = 0;
Pierre Peiffered2ddbf2008-02-08 04:18:57 -0800180 ipc_init_ids(&ns->ids[IPC_SEM_IDS]);
Kirill Korotaeve3893532006-10-02 02:18:22 -0700181}
182
Pavel Emelyanovae5e1b22008-02-08 04:18:22 -0800183#ifdef CONFIG_IPC_NS
Kirill Korotaeve3893532006-10-02 02:18:22 -0700184void sem_exit_ns(struct ipc_namespace *ns)
185{
Pierre Peiffer01b8b072008-02-08 04:18:57 -0800186 free_ipcs(ns, &sem_ids(ns), freeary);
Serge E. Hallyn7d6feeb2009-12-15 16:47:27 -0800187 idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr);
Kirill Korotaeve3893532006-10-02 02:18:22 -0700188}
Pavel Emelyanovae5e1b22008-02-08 04:18:22 -0800189#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190
Manfred Spraul239521f2014-01-27 17:07:04 -0800191void __init sem_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192{
Pierre Peiffered2ddbf2008-02-08 04:18:57 -0800193 sem_init_ns(&init_ipc_ns);
Mike Waychison19b49462005-09-06 15:17:10 -0700194 ipc_init_proc_interface("sysvipc/sem",
195 " key semid perms nsems uid gid cuid cgid otime ctime\n",
Kirill Korotaeve3893532006-10-02 02:18:22 -0700196 IPC_SEM_IDS, sysvipc_sem_proc_show);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197}
198
Manfred Spraulf269f402013-07-08 16:01:24 -0700199/**
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/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800228 * merge_queues - merge single semop queues into global queue
Manfred Spraulf269f402013-07-08 16:01:24 -0700229 * @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
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -0700246static 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
Nadia Derbey3e148c72007-10-18 23:40:54 -0700255/*
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700256 * 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.
Manfred Spraul6d07b682013-09-30 13:45:06 -0700260 * that a) sem_perm.lock is free and b) complex_count is 0.
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700261 */
262static void sem_wait_array(struct sem_array *sma)
263{
264 int i;
265 struct sem *sem;
266
Manfred Spraul6d07b682013-09-30 13:45:06 -0700267 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
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700274 for (i = 0; i < sma->sem_nsems; i++) {
275 sem = sma->sem_base + i;
276 spin_unlock_wait(&sem->lock);
277 }
278}
279
280/*
Rik van Riel6062a8d2013-04-30 19:15:44 -0700281 * 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.
Rik van Riel6062a8d2013-04-30 19:15:44 -0700286 */
287static inline int sem_lock(struct sem_array *sma, struct sembuf *sops,
288 int nsops)
289{
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700290 struct sem *sem;
Rik van Riel6062a8d2013-04-30 19:15:44 -0700291
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700292 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 */
Rik van Riel6062a8d2013-04-30 19:15:44 -0700325 spin_lock(&sem->lock);
326
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700327 /* Then check that the global lock is free */
328 if (!spin_is_locked(&sma->sem_perm.lock)) {
Manfred Spraul2e094ab2014-12-12 16:58:11 -0800329 /*
330 * The ipc object lock check must be visible on all
331 * cores before rechecking the complex count. Otherwise
332 * we can race with another thread that does:
333 * complex_count++;
334 * spin_unlock(sem_perm.lock);
335 */
336 smp_rmb();
Rik van Riel6062a8d2013-04-30 19:15:44 -0700337
Manfred Spraul2e094ab2014-12-12 16:58:11 -0800338 /*
339 * Now repeat the test of complex_count:
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700340 * It can't change anymore until we drop sem->lock.
341 * Thus: if is now 0, then it will stay 0.
342 */
343 if (sma->complex_count == 0) {
344 /* fast path successful! */
345 return sops->sem_num;
346 }
Rik van Riel6062a8d2013-04-30 19:15:44 -0700347 }
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700348 spin_unlock(&sem->lock);
Rik van Riel6062a8d2013-04-30 19:15:44 -0700349 }
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700350
351 /* slow path: acquire the full lock */
352 ipc_lock_object(&sma->sem_perm);
353
354 if (sma->complex_count == 0) {
355 /* False alarm:
356 * There is no complex operation, thus we can switch
357 * back to the fast path.
358 */
359 spin_lock(&sem->lock);
360 ipc_unlock_object(&sma->sem_perm);
361 return sops->sem_num;
362 } else {
363 /* Not a false alarm, thus complete the sequence for a
364 * full lock.
365 */
366 sem_wait_array(sma);
367 return -1;
368 }
Rik van Riel6062a8d2013-04-30 19:15:44 -0700369}
370
371static inline void sem_unlock(struct sem_array *sma, int locknum)
372{
373 if (locknum == -1) {
Manfred Spraulf269f402013-07-08 16:01:24 -0700374 unmerge_queues(sma);
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -0700375 ipc_unlock_object(&sma->sem_perm);
Rik van Riel6062a8d2013-04-30 19:15:44 -0700376 } else {
377 struct sem *sem = sma->sem_base + locknum;
378 spin_unlock(&sem->lock);
379 }
Rik van Riel6062a8d2013-04-30 19:15:44 -0700380}
381
382/*
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700383 * sem_lock_(check_) routines are called in the paths where the rwsem
Nadia Derbey3e148c72007-10-18 23:40:54 -0700384 * is not held.
Linus Torvalds321310c2013-05-04 10:47:57 -0700385 *
386 * The caller holds the RCU read lock.
Nadia Derbey3e148c72007-10-18 23:40:54 -0700387 */
Rik van Riel6062a8d2013-04-30 19:15:44 -0700388static inline struct sem_array *sem_obtain_lock(struct ipc_namespace *ns,
389 int id, struct sembuf *sops, int nsops, int *locknum)
Nadia Derbey023a5352007-10-18 23:40:51 -0700390{
Rik van Rielc460b662013-04-30 19:15:35 -0700391 struct kern_ipc_perm *ipcp;
392 struct sem_array *sma;
Nadia Derbey03f02c72007-10-18 23:40:51 -0700393
Rik van Rielc460b662013-04-30 19:15:35 -0700394 ipcp = ipc_obtain_object(&sem_ids(ns), id);
Linus Torvalds321310c2013-05-04 10:47:57 -0700395 if (IS_ERR(ipcp))
396 return ERR_CAST(ipcp);
Pierre Peifferb1ed88b2008-02-06 01:36:23 -0800397
Rik van Riel6062a8d2013-04-30 19:15:44 -0700398 sma = container_of(ipcp, struct sem_array, sem_perm);
399 *locknum = sem_lock(sma, sops, nsops);
Rik van Rielc460b662013-04-30 19:15:35 -0700400
401 /* ipc_rmid() may have already freed the ID while sem_lock
402 * was spinning: verify that the structure is still valid
403 */
Rafael Aquini72a8ff22014-01-27 17:07:02 -0800404 if (ipc_valid_object(ipcp))
Rik van Rielc460b662013-04-30 19:15:35 -0700405 return container_of(ipcp, struct sem_array, sem_perm);
406
Rik van Riel6062a8d2013-04-30 19:15:44 -0700407 sem_unlock(sma, *locknum);
Linus Torvalds321310c2013-05-04 10:47:57 -0700408 return ERR_PTR(-EINVAL);
Nadia Derbey023a5352007-10-18 23:40:51 -0700409}
410
Davidlohr Bueso16df3672013-04-30 19:15:29 -0700411static inline struct sem_array *sem_obtain_object(struct ipc_namespace *ns, int id)
412{
413 struct kern_ipc_perm *ipcp = ipc_obtain_object(&sem_ids(ns), id);
414
415 if (IS_ERR(ipcp))
416 return ERR_CAST(ipcp);
417
418 return container_of(ipcp, struct sem_array, sem_perm);
419}
420
Davidlohr Bueso16df3672013-04-30 19:15:29 -0700421static inline struct sem_array *sem_obtain_object_check(struct ipc_namespace *ns,
422 int id)
423{
424 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&sem_ids(ns), id);
425
426 if (IS_ERR(ipcp))
427 return ERR_CAST(ipcp);
Pierre Peifferb1ed88b2008-02-06 01:36:23 -0800428
Nadia Derbey03f02c72007-10-18 23:40:51 -0700429 return container_of(ipcp, struct sem_array, sem_perm);
Nadia Derbey023a5352007-10-18 23:40:51 -0700430}
431
Pierre Peiffer6ff37972008-04-29 01:00:46 -0700432static inline void sem_lock_and_putref(struct sem_array *sma)
433{
Rik van Riel6062a8d2013-04-30 19:15:44 -0700434 sem_lock(sma, NULL, -1);
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -0700435 ipc_rcu_putref(sma, ipc_rcu_free);
Pierre Peiffer6ff37972008-04-29 01:00:46 -0700436}
437
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700438static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
439{
440 ipc_rmid(&sem_ids(ns), &s->sem_perm);
441}
442
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443/*
444 * Lockless wakeup algorithm:
445 * Without the check/retry algorithm a lockless wakeup is possible:
446 * - queue.status is initialized to -EINTR before blocking.
447 * - wakeup is performed by
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700448 * * unlinking the queue entry from the pending list
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 * * setting queue.status to IN_WAKEUP
450 * This is the notification for the blocked thread that a
451 * result value is imminent.
452 * * call wake_up_process
453 * * set queue.status to the final value.
454 * - the previously blocked thread checks queue.status:
Manfred Spraul239521f2014-01-27 17:07:04 -0800455 * * if it's IN_WAKEUP, then it must wait until the value changes
456 * * if it's not -EINTR, then the operation was completed by
457 * update_queue. semtimedop can return queue.status without
458 * performing any operation on the sem array.
459 * * otherwise it must acquire the spinlock and check what's up.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 *
461 * The two-stage algorithm is necessary to protect against the following
462 * races:
463 * - if queue.status is set after wake_up_process, then the woken up idle
464 * thread could race forward and try (and fail) to acquire sma->lock
465 * before update_queue had a chance to set queue.status
466 * - if queue.status is written before wake_up_process and if the
467 * blocked process is woken up by a signal between writing
468 * queue.status and the wake_up_process, then the woken up
469 * process could return from semtimedop and die by calling
470 * sys_exit before wake_up_process is called. Then wake_up_process
471 * will oops, because the task structure is already invalid.
472 * (yes, this happened on s390 with sysv msg).
473 *
474 */
475#define IN_WAKEUP 1
476
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700477/**
478 * newary - Create a new semaphore set
479 * @ns: namespace
480 * @params: ptr to the structure that contains key, semflg and nsems
481 *
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700482 * Called with sem_ids.rwsem held (as a writer)
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700483 */
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700484static int newary(struct ipc_namespace *ns, struct ipc_params *params)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485{
486 int id;
487 int retval;
488 struct sem_array *sma;
489 int size;
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700490 key_t key = params->key;
491 int nsems = params->u.nsems;
492 int semflg = params->flg;
Manfred Spraulb97e8202009-12-15 16:47:32 -0800493 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494
495 if (!nsems)
496 return -EINVAL;
Kirill Korotaeve3893532006-10-02 02:18:22 -0700497 if (ns->used_sems + nsems > ns->sc_semmns)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 return -ENOSPC;
499
Manfred Spraul239521f2014-01-27 17:07:04 -0800500 size = sizeof(*sma) + nsems * sizeof(struct sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 sma = ipc_rcu_alloc(size);
Davidlohr Bueso3ab08fe2014-01-27 17:07:06 -0800502 if (!sma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 return -ENOMEM;
Davidlohr Bueso3ab08fe2014-01-27 17:07:06 -0800504
Manfred Spraul239521f2014-01-27 17:07:04 -0800505 memset(sma, 0, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506
507 sma->sem_perm.mode = (semflg & S_IRWXUGO);
508 sma->sem_perm.key = key;
509
510 sma->sem_perm.security = NULL;
511 retval = security_sem_alloc(sma);
512 if (retval) {
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -0700513 ipc_rcu_putref(sma, ipc_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 return retval;
515 }
516
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 sma->sem_base = (struct sem *) &sma[1];
Manfred Spraulb97e8202009-12-15 16:47:32 -0800518
Rik van Riel6062a8d2013-04-30 19:15:44 -0700519 for (i = 0; i < nsems; i++) {
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700520 INIT_LIST_HEAD(&sma->sem_base[i].pending_alter);
521 INIT_LIST_HEAD(&sma->sem_base[i].pending_const);
Rik van Riel6062a8d2013-04-30 19:15:44 -0700522 spin_lock_init(&sma->sem_base[i].lock);
523 }
Manfred Spraulb97e8202009-12-15 16:47:32 -0800524
525 sma->complex_count = 0;
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700526 INIT_LIST_HEAD(&sma->pending_alter);
527 INIT_LIST_HEAD(&sma->pending_const);
Manfred Spraul4daa28f2008-07-25 01:48:04 -0700528 INIT_LIST_HEAD(&sma->list_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 sma->sem_nsems = nsems;
530 sma->sem_ctime = get_seconds();
Manfred Spraule8577d12014-12-02 15:59:34 -0800531
532 id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
533 if (id < 0) {
534 ipc_rcu_putref(sma, sem_rcu_free);
535 return id;
536 }
537 ns->used_sems += nsems;
538
Rik van Riel6062a8d2013-04-30 19:15:44 -0700539 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -0700540 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700542 return sma->sem_perm.id;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543}
544
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700545
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700546/*
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700547 * Called with sem_ids.rwsem and ipcp locked.
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700548 */
Nadia Derbey03f02c72007-10-18 23:40:51 -0700549static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550{
Nadia Derbey03f02c72007-10-18 23:40:51 -0700551 struct sem_array *sma;
552
553 sma = container_of(ipcp, struct sem_array, sem_perm);
554 return security_sem_associate(sma, semflg);
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700555}
556
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700557/*
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700558 * Called with sem_ids.rwsem and ipcp locked.
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700559 */
Nadia Derbey03f02c72007-10-18 23:40:51 -0700560static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
561 struct ipc_params *params)
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700562{
Nadia Derbey03f02c72007-10-18 23:40:51 -0700563 struct sem_array *sma;
564
565 sma = container_of(ipcp, struct sem_array, sem_perm);
566 if (params->u.nsems > sma->sem_nsems)
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700567 return -EINVAL;
568
569 return 0;
570}
571
Heiko Carstensd5460c92009-01-14 14:14:27 +0100572SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg)
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700573{
Kirill Korotaeve3893532006-10-02 02:18:22 -0700574 struct ipc_namespace *ns;
Mathias Krauseeb66ec42014-06-06 14:37:36 -0700575 static const struct ipc_ops sem_ops = {
576 .getnew = newary,
577 .associate = sem_security,
578 .more_checks = sem_more_checks,
579 };
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700580 struct ipc_params sem_params;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581
Kirill Korotaeve3893532006-10-02 02:18:22 -0700582 ns = current->nsproxy->ipc_ns;
583
584 if (nsems < 0 || nsems > ns->sc_semmsl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 return -EINVAL;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700586
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700587 sem_params.key = key;
588 sem_params.flg = semflg;
589 sem_params.u.nsems = nsems;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700590
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700591 return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592}
593
Petr Mladek78f50092014-01-27 17:07:00 -0800594/**
595 * perform_atomic_semop - Perform (if possible) a semaphore operation
Manfred Spraul758a6ba32013-07-08 16:01:26 -0700596 * @sma: semaphore array
Manfred Sprauld198cd62014-06-06 14:37:49 -0700597 * @q: struct sem_queue that describes the operation
Manfred Spraul758a6ba32013-07-08 16:01:26 -0700598 *
599 * Returns 0 if the operation was possible.
600 * Returns 1 if the operation is impossible, the caller must sleep.
601 * Negative values are error codes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 */
Manfred Sprauld198cd62014-06-06 14:37:49 -0700603static int perform_atomic_semop(struct sem_array *sma, struct sem_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604{
Manfred Sprauld198cd62014-06-06 14:37:49 -0700605 int result, sem_op, nsops, pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 struct sembuf *sop;
Manfred Spraul239521f2014-01-27 17:07:04 -0800607 struct sem *curr;
Manfred Sprauld198cd62014-06-06 14:37:49 -0700608 struct sembuf *sops;
609 struct sem_undo *un;
610
611 sops = q->sops;
612 nsops = q->nsops;
613 un = q->undo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614
615 for (sop = sops; sop < sops + nsops; sop++) {
616 curr = sma->sem_base + sop->sem_num;
617 sem_op = sop->sem_op;
618 result = curr->semval;
Petr Mladek78f50092014-01-27 17:07:00 -0800619
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 if (!sem_op && result)
621 goto would_block;
622
623 result += sem_op;
624 if (result < 0)
625 goto would_block;
626 if (result > SEMVMX)
627 goto out_of_range;
Petr Mladek78f50092014-01-27 17:07:00 -0800628
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 if (sop->sem_flg & SEM_UNDO) {
630 int undo = un->semadj[sop->sem_num] - sem_op;
Petr Mladek78f50092014-01-27 17:07:00 -0800631 /* Exceeding the undo range is an error. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
633 goto out_of_range;
Petr Mladek78f50092014-01-27 17:07:00 -0800634 un->semadj[sop->sem_num] = undo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 }
Petr Mladek78f50092014-01-27 17:07:00 -0800636
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 curr->semval = result;
638 }
639
640 sop--;
Manfred Sprauld198cd62014-06-06 14:37:49 -0700641 pid = q->pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 while (sop >= sops) {
643 sma->sem_base[sop->sem_num].sempid = pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 sop--;
645 }
Petr Mladek78f50092014-01-27 17:07:00 -0800646
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 return 0;
648
649out_of_range:
650 result = -ERANGE;
651 goto undo;
652
653would_block:
Manfred Sprauled247b72014-06-06 14:37:49 -0700654 q->blocking = sop;
655
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 if (sop->sem_flg & IPC_NOWAIT)
657 result = -EAGAIN;
658 else
659 result = 1;
660
661undo:
662 sop--;
663 while (sop >= sops) {
Petr Mladek78f50092014-01-27 17:07:00 -0800664 sem_op = sop->sem_op;
665 sma->sem_base[sop->sem_num].semval -= sem_op;
666 if (sop->sem_flg & SEM_UNDO)
667 un->semadj[sop->sem_num] += sem_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 sop--;
669 }
670
671 return result;
672}
673
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700674/** wake_up_sem_queue_prepare(q, error): Prepare wake-up
675 * @q: queue entry that must be signaled
676 * @error: Error value for the signal
677 *
678 * Prepare the wake-up of the queue entry q.
Nick Piggind4212092009-12-15 16:47:30 -0800679 */
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700680static void wake_up_sem_queue_prepare(struct list_head *pt,
681 struct sem_queue *q, int error)
Nick Piggind4212092009-12-15 16:47:30 -0800682{
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700683 if (list_empty(pt)) {
684 /*
685 * Hold preempt off so that we don't get preempted and have the
686 * wakee busy-wait until we're scheduled back on.
687 */
688 preempt_disable();
689 }
Nick Piggind4212092009-12-15 16:47:30 -0800690 q->status = IN_WAKEUP;
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700691 q->pid = error;
692
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700693 list_add_tail(&q->list, pt);
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700694}
695
696/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800697 * wake_up_sem_queue_do - do the actual wake-up
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700698 * @pt: list of tasks to be woken up
699 *
700 * Do the actual wake-up.
701 * The function is called without any locks held, thus the semaphore array
702 * could be destroyed already and the tasks can disappear as soon as the
703 * status is set to the actual return code.
704 */
705static void wake_up_sem_queue_do(struct list_head *pt)
706{
707 struct sem_queue *q, *t;
708 int did_something;
709
710 did_something = !list_empty(pt);
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700711 list_for_each_entry_safe(q, t, pt, list) {
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700712 wake_up_process(q->sleeper);
713 /* q can disappear immediately after writing q->status. */
714 smp_wmb();
715 q->status = q->pid;
716 }
717 if (did_something)
718 preempt_enable();
Nick Piggind4212092009-12-15 16:47:30 -0800719}
720
Manfred Spraulb97e8202009-12-15 16:47:32 -0800721static void unlink_queue(struct sem_array *sma, struct sem_queue *q)
722{
723 list_del(&q->list);
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700724 if (q->nsops > 1)
Manfred Spraulb97e8202009-12-15 16:47:32 -0800725 sma->complex_count--;
726}
727
Manfred Spraulfd5db422010-05-26 14:43:40 -0700728/** check_restart(sma, q)
729 * @sma: semaphore array
730 * @q: the operation that just completed
731 *
732 * update_queue is O(N^2) when it restarts scanning the whole queue of
733 * waiting operations. Therefore this function checks if the restart is
734 * really necessary. It is called after a previously waiting operation
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700735 * modified the array.
736 * Note that wait-for-zero operations are handled without restart.
Manfred Spraulfd5db422010-05-26 14:43:40 -0700737 */
738static int check_restart(struct sem_array *sma, struct sem_queue *q)
739{
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700740 /* pending complex alter operations are too difficult to analyse */
741 if (!list_empty(&sma->pending_alter))
Manfred Spraulfd5db422010-05-26 14:43:40 -0700742 return 1;
743
744 /* we were a sleeping complex operation. Too difficult */
745 if (q->nsops > 1)
746 return 1;
747
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700748 /* It is impossible that someone waits for the new value:
749 * - complex operations always restart.
750 * - wait-for-zero are handled seperately.
751 * - q is a previously sleeping simple operation that
752 * altered the array. It must be a decrement, because
753 * simple increments never sleep.
754 * - If there are older (higher priority) decrements
755 * in the queue, then they have observed the original
756 * semval value and couldn't proceed. The operation
757 * decremented to value - thus they won't proceed either.
758 */
759 return 0;
760}
Manfred Spraulfd5db422010-05-26 14:43:40 -0700761
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700762/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800763 * wake_const_ops - wake up non-alter tasks
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700764 * @sma: semaphore array.
765 * @semnum: semaphore that was modified.
766 * @pt: list head for the tasks that must be woken up.
767 *
768 * wake_const_ops must be called after a semaphore in a semaphore array
769 * was set to 0. If complex const operations are pending, wake_const_ops must
770 * be called with semnum = -1, as well as with the number of each modified
771 * semaphore.
772 * The tasks that must be woken up are added to @pt. The return code
773 * is stored in q->pid.
774 * The function returns 1 if at least one operation was completed successfully.
775 */
776static int wake_const_ops(struct sem_array *sma, int semnum,
777 struct list_head *pt)
778{
779 struct sem_queue *q;
780 struct list_head *walk;
781 struct list_head *pending_list;
782 int semop_completed = 0;
Manfred Spraulfd5db422010-05-26 14:43:40 -0700783
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700784 if (semnum == -1)
785 pending_list = &sma->pending_const;
786 else
787 pending_list = &sma->sem_base[semnum].pending_const;
788
789 walk = pending_list->next;
790 while (walk != pending_list) {
791 int error;
792
793 q = container_of(walk, struct sem_queue, list);
794 walk = walk->next;
795
Manfred Sprauld198cd62014-06-06 14:37:49 -0700796 error = perform_atomic_semop(sma, q);
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700797
798 if (error <= 0) {
799 /* operation completed, remove from queue & wakeup */
800
801 unlink_queue(sma, q);
802
803 wake_up_sem_queue_prepare(pt, q, error);
804 if (error == 0)
805 semop_completed = 1;
806 }
807 }
808 return semop_completed;
809}
810
811/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800812 * do_smart_wakeup_zero - wakeup all wait for zero tasks
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700813 * @sma: semaphore array
814 * @sops: operations that were performed
815 * @nsops: number of operations
816 * @pt: list head of the tasks that must be woken up.
817 *
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800818 * Checks all required queue for wait-for-zero operations, based
819 * on the actual changes that were performed on the semaphore array.
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700820 * The function returns 1 if at least one operation was completed successfully.
821 */
822static int do_smart_wakeup_zero(struct sem_array *sma, struct sembuf *sops,
823 int nsops, struct list_head *pt)
824{
825 int i;
826 int semop_completed = 0;
827 int got_zero = 0;
828
829 /* first: the per-semaphore queues, if known */
830 if (sops) {
831 for (i = 0; i < nsops; i++) {
832 int num = sops[i].sem_num;
833
834 if (sma->sem_base[num].semval == 0) {
835 got_zero = 1;
836 semop_completed |= wake_const_ops(sma, num, pt);
837 }
838 }
839 } else {
840 /*
841 * No sops means modified semaphores not known.
842 * Assume all were changed.
Manfred Spraulfd5db422010-05-26 14:43:40 -0700843 */
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700844 for (i = 0; i < sma->sem_nsems; i++) {
845 if (sma->sem_base[i].semval == 0) {
846 got_zero = 1;
847 semop_completed |= wake_const_ops(sma, i, pt);
848 }
849 }
Manfred Spraulfd5db422010-05-26 14:43:40 -0700850 }
851 /*
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700852 * If one of the modified semaphores got 0,
853 * then check the global queue, too.
Manfred Spraulfd5db422010-05-26 14:43:40 -0700854 */
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700855 if (got_zero)
856 semop_completed |= wake_const_ops(sma, -1, pt);
Manfred Spraulfd5db422010-05-26 14:43:40 -0700857
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700858 return semop_completed;
Manfred Spraulfd5db422010-05-26 14:43:40 -0700859}
860
Manfred Spraul636c6be2009-12-15 16:47:33 -0800861
862/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800863 * update_queue - look for tasks that can be completed.
Manfred Spraul636c6be2009-12-15 16:47:33 -0800864 * @sma: semaphore array.
865 * @semnum: semaphore that was modified.
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700866 * @pt: list head for the tasks that must be woken up.
Manfred Spraul636c6be2009-12-15 16:47:33 -0800867 *
868 * update_queue must be called after a semaphore in a semaphore array
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700869 * was modified. If multiple semaphores were modified, update_queue must
870 * be called with semnum = -1, as well as with the number of each modified
871 * semaphore.
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700872 * The tasks that must be woken up are added to @pt. The return code
873 * is stored in q->pid.
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700874 * The function internally checks if const operations can now succeed.
875 *
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700876 * The function return 1 if at least one semop was completed successfully.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 */
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700878static int update_queue(struct sem_array *sma, int semnum, struct list_head *pt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879{
Manfred Spraul636c6be2009-12-15 16:47:33 -0800880 struct sem_queue *q;
881 struct list_head *walk;
882 struct list_head *pending_list;
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700883 int semop_completed = 0;
Manfred Spraul636c6be2009-12-15 16:47:33 -0800884
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700885 if (semnum == -1)
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700886 pending_list = &sma->pending_alter;
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700887 else
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700888 pending_list = &sma->sem_base[semnum].pending_alter;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889
Nick Piggin9cad2002009-12-15 16:47:29 -0800890again:
Manfred Spraul636c6be2009-12-15 16:47:33 -0800891 walk = pending_list->next;
892 while (walk != pending_list) {
Manfred Spraulfd5db422010-05-26 14:43:40 -0700893 int error, restart;
Manfred Spraul636c6be2009-12-15 16:47:33 -0800894
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700895 q = container_of(walk, struct sem_queue, list);
Manfred Spraul636c6be2009-12-15 16:47:33 -0800896 walk = walk->next;
Nick Piggin9cad2002009-12-15 16:47:29 -0800897
Manfred Sprauld987f8b2009-12-15 16:47:34 -0800898 /* If we are scanning the single sop, per-semaphore list of
899 * one semaphore and that semaphore is 0, then it is not
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700900 * necessary to scan further: simple increments
Manfred Sprauld987f8b2009-12-15 16:47:34 -0800901 * that affect only one entry succeed immediately and cannot
902 * be in the per semaphore pending queue, and decrements
903 * cannot be successful if the value is already 0.
904 */
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700905 if (semnum != -1 && sma->sem_base[semnum].semval == 0)
Manfred Sprauld987f8b2009-12-15 16:47:34 -0800906 break;
907
Manfred Sprauld198cd62014-06-06 14:37:49 -0700908 error = perform_atomic_semop(sma, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909
910 /* Does q->sleeper still need to sleep? */
Nick Piggin9cad2002009-12-15 16:47:29 -0800911 if (error > 0)
912 continue;
Manfred Spraula1193f82008-07-25 01:48:06 -0700913
Manfred Spraulb97e8202009-12-15 16:47:32 -0800914 unlink_queue(sma, q);
Manfred Spraula1193f82008-07-25 01:48:06 -0700915
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700916 if (error) {
Manfred Spraulfd5db422010-05-26 14:43:40 -0700917 restart = 0;
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700918 } else {
919 semop_completed = 1;
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700920 do_smart_wakeup_zero(sma, q->sops, q->nsops, pt);
Manfred Spraulfd5db422010-05-26 14:43:40 -0700921 restart = check_restart(sma, q);
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700922 }
Manfred Spraulfd5db422010-05-26 14:43:40 -0700923
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700924 wake_up_sem_queue_prepare(pt, q, error);
Manfred Spraulfd5db422010-05-26 14:43:40 -0700925 if (restart)
Nick Piggin9cad2002009-12-15 16:47:29 -0800926 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927 }
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700928 return semop_completed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929}
930
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700931/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800932 * set_semotime - set sem_otime
Manfred Spraul0e8c6652013-09-30 13:45:25 -0700933 * @sma: semaphore array
934 * @sops: operations that modified the array, may be NULL
935 *
936 * sem_otime is replicated to avoid cache line trashing.
937 * This function sets one instance to the current time.
938 */
939static void set_semotime(struct sem_array *sma, struct sembuf *sops)
940{
941 if (sops == NULL) {
942 sma->sem_base[0].sem_otime = get_seconds();
943 } else {
944 sma->sem_base[sops[0].sem_num].sem_otime =
945 get_seconds();
946 }
947}
948
949/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800950 * do_smart_update - optimized update_queue
Manfred Spraulfd5db422010-05-26 14:43:40 -0700951 * @sma: semaphore array
952 * @sops: operations that were performed
953 * @nsops: number of operations
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700954 * @otime: force setting otime
955 * @pt: list head of the tasks that must be woken up.
Manfred Spraulfd5db422010-05-26 14:43:40 -0700956 *
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700957 * do_smart_update() does the required calls to update_queue and wakeup_zero,
958 * based on the actual changes that were performed on the semaphore array.
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700959 * Note that the function does not do the actual wake-up: the caller is
960 * responsible for calling wake_up_sem_queue_do(@pt).
961 * It is safe to perform this call after dropping all locks.
Manfred Spraulfd5db422010-05-26 14:43:40 -0700962 */
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700963static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops,
964 int otime, struct list_head *pt)
Manfred Spraulfd5db422010-05-26 14:43:40 -0700965{
966 int i;
967
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700968 otime |= do_smart_wakeup_zero(sma, sops, nsops, pt);
969
Manfred Spraulf269f402013-07-08 16:01:24 -0700970 if (!list_empty(&sma->pending_alter)) {
971 /* semaphore array uses the global queue - just process it. */
972 otime |= update_queue(sma, -1, pt);
973 } else {
974 if (!sops) {
975 /*
976 * No sops, thus the modified semaphores are not
977 * known. Check all.
978 */
979 for (i = 0; i < sma->sem_nsems; i++)
980 otime |= update_queue(sma, i, pt);
981 } else {
982 /*
983 * Check the semaphores that were increased:
984 * - No complex ops, thus all sleeping ops are
985 * decrease.
986 * - if we decreased the value, then any sleeping
987 * semaphore ops wont be able to run: If the
988 * previous value was too small, then the new
989 * value will be too small, too.
990 */
991 for (i = 0; i < nsops; i++) {
992 if (sops[i].sem_op > 0) {
993 otime |= update_queue(sma,
994 sops[i].sem_num, pt);
995 }
Manfred Spraulab465df2013-05-26 11:08:52 +0200996 }
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700997 }
Manfred Spraulfd5db422010-05-26 14:43:40 -0700998 }
Manfred Spraul0e8c6652013-09-30 13:45:25 -0700999 if (otime)
1000 set_semotime(sma, sops);
Manfred Spraulfd5db422010-05-26 14:43:40 -07001001}
1002
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001003/*
Manfred Spraulb220c572014-06-06 14:37:51 -07001004 * check_qop: Test if a queued operation sleeps on the semaphore semnum
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001005 */
1006static int check_qop(struct sem_array *sma, int semnum, struct sem_queue *q,
1007 bool count_zero)
1008{
Manfred Spraulb220c572014-06-06 14:37:51 -07001009 struct sembuf *sop = q->blocking;
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001010
Manfred Spraul9b44ee22014-06-06 14:37:52 -07001011 /*
1012 * Linux always (since 0.99.10) reported a task as sleeping on all
1013 * semaphores. This violates SUS, therefore it was changed to the
1014 * standard compliant behavior.
1015 * Give the administrators a chance to notice that an application
1016 * might misbehave because it relies on the Linux behavior.
1017 */
1018 pr_info_once("semctl(GETNCNT/GETZCNT) is since 3.16 Single Unix Specification compliant.\n"
1019 "The task %s (%d) triggered the difference, watch for misbehavior.\n",
1020 current->comm, task_pid_nr(current));
1021
Manfred Spraulb220c572014-06-06 14:37:51 -07001022 if (sop->sem_num != semnum)
1023 return 0;
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001024
Manfred Spraulb220c572014-06-06 14:37:51 -07001025 if (count_zero && sop->sem_op == 0)
1026 return 1;
1027 if (!count_zero && sop->sem_op < 0)
1028 return 1;
1029
1030 return 0;
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001031}
1032
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033/* The following counts are associated to each semaphore:
1034 * semncnt number of tasks waiting on semval being nonzero
1035 * semzcnt number of tasks waiting on semval being zero
Manfred Spraulb220c572014-06-06 14:37:51 -07001036 *
1037 * Per definition, a task waits only on the semaphore of the first semop
1038 * that cannot proceed, even if additional operation would block, too.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 */
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001040static int count_semcnt(struct sem_array *sma, ushort semnum,
1041 bool count_zero)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042{
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001043 struct list_head *l;
Manfred Spraul239521f2014-01-27 17:07:04 -08001044 struct sem_queue *q;
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001045 int semcnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001047 semcnt = 0;
1048 /* First: check the simple operations. They are easy to evaluate */
1049 if (count_zero)
1050 l = &sma->sem_base[semnum].pending_const;
1051 else
1052 l = &sma->sem_base[semnum].pending_alter;
1053
1054 list_for_each_entry(q, l, list) {
1055 /* all task on a per-semaphore list sleep on exactly
1056 * that semaphore
1057 */
1058 semcnt++;
Rik van Rielde2657f2013-05-09 16:59:59 -04001059 }
1060
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001061 /* Then: check the complex operations. */
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001062 list_for_each_entry(q, &sma->pending_alter, list) {
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001063 semcnt += check_qop(sma, semnum, q, count_zero);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 }
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001065 if (count_zero) {
1066 list_for_each_entry(q, &sma->pending_const, list) {
1067 semcnt += check_qop(sma, semnum, q, count_zero);
1068 }
Rik van Rielebc2e5e2013-05-09 16:53:28 -04001069 }
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001070 return semcnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071}
1072
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001073/* Free a semaphore set. freeary() is called with sem_ids.rwsem locked
1074 * as a writer and the spinlock for this semaphore set hold. sem_ids.rwsem
Nadia Derbey3e148c72007-10-18 23:40:54 -07001075 * remains locked on exit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 */
Pierre Peiffer01b8b072008-02-08 04:18:57 -08001077static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078{
Manfred Spraul380af1b2008-07-25 01:48:06 -07001079 struct sem_undo *un, *tu;
1080 struct sem_queue *q, *tq;
Pierre Peiffer01b8b072008-02-08 04:18:57 -08001081 struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001082 struct list_head tasks;
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001083 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084
Manfred Spraul380af1b2008-07-25 01:48:06 -07001085 /* Free the existing undo structures for this semaphore set. */
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -07001086 ipc_assert_locked_object(&sma->sem_perm);
Manfred Spraul380af1b2008-07-25 01:48:06 -07001087 list_for_each_entry_safe(un, tu, &sma->list_id, list_id) {
1088 list_del(&un->list_id);
1089 spin_lock(&un->ulp->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090 un->semid = -1;
Manfred Spraul380af1b2008-07-25 01:48:06 -07001091 list_del_rcu(&un->list_proc);
1092 spin_unlock(&un->ulp->lock);
Lai Jiangshan693a8b62011-03-18 12:09:35 +08001093 kfree_rcu(un, rcu);
Manfred Spraul380af1b2008-07-25 01:48:06 -07001094 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095
1096 /* Wake up all pending processes and let them fail with EIDRM. */
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001097 INIT_LIST_HEAD(&tasks);
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001098 list_for_each_entry_safe(q, tq, &sma->pending_const, list) {
1099 unlink_queue(sma, q);
1100 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1101 }
1102
1103 list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
Manfred Spraulb97e8202009-12-15 16:47:32 -08001104 unlink_queue(sma, q);
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001105 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 }
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001107 for (i = 0; i < sma->sem_nsems; i++) {
1108 struct sem *sem = sma->sem_base + i;
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001109 list_for_each_entry_safe(q, tq, &sem->pending_const, list) {
1110 unlink_queue(sma, q);
1111 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1112 }
1113 list_for_each_entry_safe(q, tq, &sem->pending_alter, list) {
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001114 unlink_queue(sma, q);
1115 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1116 }
1117 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118
Nadia Derbey7ca7e562007-10-18 23:40:48 -07001119 /* Remove the semaphore set from the IDR */
1120 sem_rmid(ns, sma);
Rik van Riel6062a8d2013-04-30 19:15:44 -07001121 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001122 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001124 wake_up_sem_queue_do(&tasks);
Kirill Korotaeve3893532006-10-02 02:18:22 -07001125 ns->used_sems -= sma->sem_nsems;
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -07001126 ipc_rcu_putref(sma, sem_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127}
1128
1129static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
1130{
Manfred Spraul239521f2014-01-27 17:07:04 -08001131 switch (version) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 case IPC_64:
1133 return copy_to_user(buf, in, sizeof(*in));
1134 case IPC_OLD:
1135 {
1136 struct semid_ds out;
1137
Dan Rosenberg982f7c22010-09-30 15:15:31 -07001138 memset(&out, 0, sizeof(out));
1139
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
1141
1142 out.sem_otime = in->sem_otime;
1143 out.sem_ctime = in->sem_ctime;
1144 out.sem_nsems = in->sem_nsems;
1145
1146 return copy_to_user(buf, &out, sizeof(out));
1147 }
1148 default:
1149 return -EINVAL;
1150 }
1151}
1152
Manfred Sprauld12e1e52013-07-08 16:01:25 -07001153static time_t get_semotime(struct sem_array *sma)
1154{
1155 int i;
1156 time_t res;
1157
1158 res = sma->sem_base[0].sem_otime;
1159 for (i = 1; i < sma->sem_nsems; i++) {
1160 time_t to = sma->sem_base[i].sem_otime;
1161
1162 if (to > res)
1163 res = to;
1164 }
1165 return res;
1166}
1167
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -08001168static int semctl_nolock(struct ipc_namespace *ns, int semid,
Al Viroe1fd1f42013-03-05 15:04:55 -05001169 int cmd, int version, void __user *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170{
Amerigo Wange5cc9c72009-12-15 16:47:35 -08001171 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 struct sem_array *sma;
1173
Manfred Spraul239521f2014-01-27 17:07:04 -08001174 switch (cmd) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 case IPC_INFO:
1176 case SEM_INFO:
1177 {
1178 struct seminfo seminfo;
1179 int max_id;
1180
1181 err = security_sem_semctl(NULL, cmd);
1182 if (err)
1183 return err;
Paul McQuade46c0a8c2014-06-06 14:37:37 -07001184
Manfred Spraul239521f2014-01-27 17:07:04 -08001185 memset(&seminfo, 0, sizeof(seminfo));
Kirill Korotaeve3893532006-10-02 02:18:22 -07001186 seminfo.semmni = ns->sc_semmni;
1187 seminfo.semmns = ns->sc_semmns;
1188 seminfo.semmsl = ns->sc_semmsl;
1189 seminfo.semopm = ns->sc_semopm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 seminfo.semvmx = SEMVMX;
1191 seminfo.semmnu = SEMMNU;
1192 seminfo.semmap = SEMMAP;
1193 seminfo.semume = SEMUME;
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001194 down_read(&sem_ids(ns).rwsem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 if (cmd == SEM_INFO) {
Kirill Korotaeve3893532006-10-02 02:18:22 -07001196 seminfo.semusz = sem_ids(ns).in_use;
1197 seminfo.semaem = ns->used_sems;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198 } else {
1199 seminfo.semusz = SEMUSZ;
1200 seminfo.semaem = SEMAEM;
1201 }
Nadia Derbey7ca7e562007-10-18 23:40:48 -07001202 max_id = ipc_get_maxid(&sem_ids(ns));
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001203 up_read(&sem_ids(ns).rwsem);
Paul McQuade46c0a8c2014-06-06 14:37:37 -07001204 if (copy_to_user(p, &seminfo, sizeof(struct seminfo)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 return -EFAULT;
Manfred Spraul239521f2014-01-27 17:07:04 -08001206 return (max_id < 0) ? 0 : max_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 }
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -08001208 case IPC_STAT:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 case SEM_STAT:
1210 {
1211 struct semid64_ds tbuf;
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001212 int id = 0;
1213
1214 memset(&tbuf, 0, sizeof(tbuf));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215
Linus Torvalds941b0302013-05-04 11:04:29 -07001216 rcu_read_lock();
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -08001217 if (cmd == SEM_STAT) {
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001218 sma = sem_obtain_object(ns, semid);
1219 if (IS_ERR(sma)) {
1220 err = PTR_ERR(sma);
1221 goto out_unlock;
1222 }
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -08001223 id = sma->sem_perm.id;
1224 } else {
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001225 sma = sem_obtain_object_check(ns, semid);
1226 if (IS_ERR(sma)) {
1227 err = PTR_ERR(sma);
1228 goto out_unlock;
1229 }
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -08001230 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231
1232 err = -EACCES;
Serge E. Hallynb0e77592011-03-23 16:43:24 -07001233 if (ipcperms(ns, &sma->sem_perm, S_IRUGO))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 goto out_unlock;
1235
1236 err = security_sem_semctl(sma, cmd);
1237 if (err)
1238 goto out_unlock;
1239
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
Manfred Sprauld12e1e52013-07-08 16:01:25 -07001241 tbuf.sem_otime = get_semotime(sma);
1242 tbuf.sem_ctime = sma->sem_ctime;
1243 tbuf.sem_nsems = sma->sem_nsems;
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001244 rcu_read_unlock();
Al Viroe1fd1f42013-03-05 15:04:55 -05001245 if (copy_semid_to_user(p, &tbuf, version))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246 return -EFAULT;
1247 return id;
1248 }
1249 default:
1250 return -EINVAL;
1251 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252out_unlock:
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001253 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 return err;
1255}
1256
Al Viroe1fd1f42013-03-05 15:04:55 -05001257static int semctl_setval(struct ipc_namespace *ns, int semid, int semnum,
1258 unsigned long arg)
1259{
1260 struct sem_undo *un;
1261 struct sem_array *sma;
Manfred Spraul239521f2014-01-27 17:07:04 -08001262 struct sem *curr;
Al Viroe1fd1f42013-03-05 15:04:55 -05001263 int err;
Al Viroe1fd1f42013-03-05 15:04:55 -05001264 struct list_head tasks;
1265 int val;
1266#if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN)
1267 /* big-endian 64bit */
1268 val = arg >> 32;
1269#else
1270 /* 32bit or little-endian 64bit */
1271 val = arg;
1272#endif
1273
Rik van Riel6062a8d2013-04-30 19:15:44 -07001274 if (val > SEMVMX || val < 0)
1275 return -ERANGE;
Al Viroe1fd1f42013-03-05 15:04:55 -05001276
1277 INIT_LIST_HEAD(&tasks);
Al Viroe1fd1f42013-03-05 15:04:55 -05001278
Rik van Riel6062a8d2013-04-30 19:15:44 -07001279 rcu_read_lock();
1280 sma = sem_obtain_object_check(ns, semid);
1281 if (IS_ERR(sma)) {
1282 rcu_read_unlock();
1283 return PTR_ERR(sma);
1284 }
1285
1286 if (semnum < 0 || semnum >= sma->sem_nsems) {
1287 rcu_read_unlock();
1288 return -EINVAL;
1289 }
1290
1291
1292 if (ipcperms(ns, &sma->sem_perm, S_IWUGO)) {
1293 rcu_read_unlock();
1294 return -EACCES;
1295 }
Al Viroe1fd1f42013-03-05 15:04:55 -05001296
1297 err = security_sem_semctl(sma, SETVAL);
Rik van Riel6062a8d2013-04-30 19:15:44 -07001298 if (err) {
1299 rcu_read_unlock();
1300 return -EACCES;
1301 }
Al Viroe1fd1f42013-03-05 15:04:55 -05001302
Rik van Riel6062a8d2013-04-30 19:15:44 -07001303 sem_lock(sma, NULL, -1);
Al Viroe1fd1f42013-03-05 15:04:55 -05001304
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001305 if (!ipc_valid_object(&sma->sem_perm)) {
Manfred Spraul6e224f92013-10-16 13:46:45 -07001306 sem_unlock(sma, -1);
1307 rcu_read_unlock();
1308 return -EIDRM;
1309 }
1310
Al Viroe1fd1f42013-03-05 15:04:55 -05001311 curr = &sma->sem_base[semnum];
1312
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -07001313 ipc_assert_locked_object(&sma->sem_perm);
Al Viroe1fd1f42013-03-05 15:04:55 -05001314 list_for_each_entry(un, &sma->list_id, list_id)
1315 un->semadj[semnum] = 0;
1316
1317 curr->semval = val;
1318 curr->sempid = task_tgid_vnr(current);
1319 sma->sem_ctime = get_seconds();
1320 /* maybe some queued-up processes were waiting for this */
1321 do_smart_update(sma, NULL, 0, 0, &tasks);
Rik van Riel6062a8d2013-04-30 19:15:44 -07001322 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001323 rcu_read_unlock();
Al Viroe1fd1f42013-03-05 15:04:55 -05001324 wake_up_sem_queue_do(&tasks);
Rik van Riel6062a8d2013-04-30 19:15:44 -07001325 return 0;
Al Viroe1fd1f42013-03-05 15:04:55 -05001326}
1327
Kirill Korotaeve3893532006-10-02 02:18:22 -07001328static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
Al Viroe1fd1f42013-03-05 15:04:55 -05001329 int cmd, void __user *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330{
1331 struct sem_array *sma;
Manfred Spraul239521f2014-01-27 17:07:04 -08001332 struct sem *curr;
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001333 int err, nsems;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334 ushort fast_sem_io[SEMMSL_FAST];
Manfred Spraul239521f2014-01-27 17:07:04 -08001335 ushort *sem_io = fast_sem_io;
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001336 struct list_head tasks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001338 INIT_LIST_HEAD(&tasks);
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001339
1340 rcu_read_lock();
1341 sma = sem_obtain_object_check(ns, semid);
1342 if (IS_ERR(sma)) {
1343 rcu_read_unlock();
1344 return PTR_ERR(sma);
1345 }
1346
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347 nsems = sma->sem_nsems;
1348
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 err = -EACCES;
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001350 if (ipcperms(ns, &sma->sem_perm, cmd == SETALL ? S_IWUGO : S_IRUGO))
1351 goto out_rcu_wakeup;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352
1353 err = security_sem_semctl(sma, cmd);
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001354 if (err)
1355 goto out_rcu_wakeup;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356
1357 err = -EACCES;
1358 switch (cmd) {
1359 case GETALL:
1360 {
Al Viroe1fd1f42013-03-05 15:04:55 -05001361 ushort __user *array = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 int i;
1363
Al Viroce857222013-05-03 00:30:49 +01001364 sem_lock(sma, NULL, -1);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001365 if (!ipc_valid_object(&sma->sem_perm)) {
Manfred Spraul6e224f92013-10-16 13:46:45 -07001366 err = -EIDRM;
1367 goto out_unlock;
1368 }
Manfred Spraul239521f2014-01-27 17:07:04 -08001369 if (nsems > SEMMSL_FAST) {
Al Viroce857222013-05-03 00:30:49 +01001370 if (!ipc_rcu_getref(sma)) {
Al Viroce857222013-05-03 00:30:49 +01001371 err = -EIDRM;
Manfred Spraul6e224f92013-10-16 13:46:45 -07001372 goto out_unlock;
Al Viroce857222013-05-03 00:30:49 +01001373 }
1374 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001375 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 sem_io = ipc_alloc(sizeof(ushort)*nsems);
Manfred Spraul239521f2014-01-27 17:07:04 -08001377 if (sem_io == NULL) {
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -07001378 ipc_rcu_putref(sma, ipc_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 return -ENOMEM;
1380 }
1381
Linus Torvalds4091fd942013-05-04 10:13:40 -07001382 rcu_read_lock();
Pierre Peiffer6ff37972008-04-29 01:00:46 -07001383 sem_lock_and_putref(sma);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001384 if (!ipc_valid_object(&sma->sem_perm)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 err = -EIDRM;
Manfred Spraul6e224f92013-10-16 13:46:45 -07001386 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 }
Al Viroce857222013-05-03 00:30:49 +01001388 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389 for (i = 0; i < sma->sem_nsems; i++)
1390 sem_io[i] = sma->sem_base[i].semval;
Rik van Riel6062a8d2013-04-30 19:15:44 -07001391 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001392 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393 err = 0;
Manfred Spraul239521f2014-01-27 17:07:04 -08001394 if (copy_to_user(array, sem_io, nsems*sizeof(ushort)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 err = -EFAULT;
1396 goto out_free;
1397 }
1398 case SETALL:
1399 {
1400 int i;
1401 struct sem_undo *un;
1402
Rik van Riel6062a8d2013-04-30 19:15:44 -07001403 if (!ipc_rcu_getref(sma)) {
Manfred Spraul6e224f92013-10-16 13:46:45 -07001404 err = -EIDRM;
1405 goto out_rcu_wakeup;
Rik van Riel6062a8d2013-04-30 19:15:44 -07001406 }
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001407 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408
Manfred Spraul239521f2014-01-27 17:07:04 -08001409 if (nsems > SEMMSL_FAST) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 sem_io = ipc_alloc(sizeof(ushort)*nsems);
Manfred Spraul239521f2014-01-27 17:07:04 -08001411 if (sem_io == NULL) {
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -07001412 ipc_rcu_putref(sma, ipc_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 return -ENOMEM;
1414 }
1415 }
1416
Manfred Spraul239521f2014-01-27 17:07:04 -08001417 if (copy_from_user(sem_io, p, nsems*sizeof(ushort))) {
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -07001418 ipc_rcu_putref(sma, ipc_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 err = -EFAULT;
1420 goto out_free;
1421 }
1422
1423 for (i = 0; i < nsems; i++) {
1424 if (sem_io[i] > SEMVMX) {
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -07001425 ipc_rcu_putref(sma, ipc_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 err = -ERANGE;
1427 goto out_free;
1428 }
1429 }
Linus Torvalds4091fd942013-05-04 10:13:40 -07001430 rcu_read_lock();
Pierre Peiffer6ff37972008-04-29 01:00:46 -07001431 sem_lock_and_putref(sma);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001432 if (!ipc_valid_object(&sma->sem_perm)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433 err = -EIDRM;
Manfred Spraul6e224f92013-10-16 13:46:45 -07001434 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 }
1436
1437 for (i = 0; i < nsems; i++)
1438 sma->sem_base[i].semval = sem_io[i];
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001439
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -07001440 ipc_assert_locked_object(&sma->sem_perm);
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001441 list_for_each_entry(un, &sma->list_id, list_id) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442 for (i = 0; i < nsems; i++)
1443 un->semadj[i] = 0;
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001444 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445 sma->sem_ctime = get_seconds();
1446 /* maybe some queued-up processes were waiting for this */
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001447 do_smart_update(sma, NULL, 0, 0, &tasks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448 err = 0;
1449 goto out_unlock;
1450 }
Al Viroe1fd1f42013-03-05 15:04:55 -05001451 /* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452 }
1453 err = -EINVAL;
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001454 if (semnum < 0 || semnum >= nsems)
1455 goto out_rcu_wakeup;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456
Rik van Riel6062a8d2013-04-30 19:15:44 -07001457 sem_lock(sma, NULL, -1);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001458 if (!ipc_valid_object(&sma->sem_perm)) {
Manfred Spraul6e224f92013-10-16 13:46:45 -07001459 err = -EIDRM;
1460 goto out_unlock;
1461 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 curr = &sma->sem_base[semnum];
1463
1464 switch (cmd) {
1465 case GETVAL:
1466 err = curr->semval;
1467 goto out_unlock;
1468 case GETPID:
1469 err = curr->sempid;
1470 goto out_unlock;
1471 case GETNCNT:
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001472 err = count_semcnt(sma, semnum, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473 goto out_unlock;
1474 case GETZCNT:
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001475 err = count_semcnt(sma, semnum, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 }
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001478
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479out_unlock:
Rik van Riel6062a8d2013-04-30 19:15:44 -07001480 sem_unlock(sma, -1);
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001481out_rcu_wakeup:
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001482 rcu_read_unlock();
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001483 wake_up_sem_queue_do(&tasks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484out_free:
Manfred Spraul239521f2014-01-27 17:07:04 -08001485 if (sem_io != fast_sem_io)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 ipc_free(sem_io, sizeof(ushort)*nsems);
1487 return err;
1488}
1489
Pierre Peiffer016d7132008-04-29 01:00:50 -07001490static inline unsigned long
1491copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492{
Manfred Spraul239521f2014-01-27 17:07:04 -08001493 switch (version) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 case IPC_64:
Pierre Peiffer016d7132008-04-29 01:00:50 -07001495 if (copy_from_user(out, buf, sizeof(*out)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498 case IPC_OLD:
1499 {
1500 struct semid_ds tbuf_old;
1501
Manfred Spraul239521f2014-01-27 17:07:04 -08001502 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 return -EFAULT;
1504
Pierre Peiffer016d7132008-04-29 01:00:50 -07001505 out->sem_perm.uid = tbuf_old.sem_perm.uid;
1506 out->sem_perm.gid = tbuf_old.sem_perm.gid;
1507 out->sem_perm.mode = tbuf_old.sem_perm.mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508
1509 return 0;
1510 }
1511 default:
1512 return -EINVAL;
1513 }
1514}
1515
Pierre Peiffer522bb2a2008-04-29 01:00:49 -07001516/*
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001517 * This function handles some semctl commands which require the rwsem
Pierre Peiffer522bb2a2008-04-29 01:00:49 -07001518 * to be held in write mode.
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001519 * NOTE: no locks must be held, the rwsem is taken inside this function.
Pierre Peiffer522bb2a2008-04-29 01:00:49 -07001520 */
Pierre Peiffer21a48262008-04-29 01:00:49 -07001521static int semctl_down(struct ipc_namespace *ns, int semid,
Al Viroe1fd1f42013-03-05 15:04:55 -05001522 int cmd, int version, void __user *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523{
1524 struct sem_array *sma;
1525 int err;
Pierre Peiffer016d7132008-04-29 01:00:50 -07001526 struct semid64_ds semid64;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 struct kern_ipc_perm *ipcp;
1528
Manfred Spraul239521f2014-01-27 17:07:04 -08001529 if (cmd == IPC_SET) {
Al Viroe1fd1f42013-03-05 15:04:55 -05001530 if (copy_semid_from_user(&semid64, p, version))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001534 down_write(&sem_ids(ns).rwsem);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001535 rcu_read_lock();
1536
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001537 ipcp = ipcctl_pre_down_nolock(ns, &sem_ids(ns), semid, cmd,
1538 &semid64.sem_perm, 0);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001539 if (IS_ERR(ipcp)) {
1540 err = PTR_ERR(ipcp);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001541 goto out_unlock1;
1542 }
Steve Grubb073115d2006-04-02 17:07:33 -04001543
Pierre Peiffera5f75e72008-04-29 01:00:54 -07001544 sma = container_of(ipcp, struct sem_array, sem_perm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545
1546 err = security_sem_semctl(sma, cmd);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001547 if (err)
1548 goto out_unlock1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001550 switch (cmd) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 case IPC_RMID:
Rik van Riel6062a8d2013-04-30 19:15:44 -07001552 sem_lock(sma, NULL, -1);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001553 /* freeary unlocks the ipc object and rcu */
Pierre Peiffer01b8b072008-02-08 04:18:57 -08001554 freeary(ns, ipcp);
Pierre Peiffer522bb2a2008-04-29 01:00:49 -07001555 goto out_up;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556 case IPC_SET:
Rik van Riel6062a8d2013-04-30 19:15:44 -07001557 sem_lock(sma, NULL, -1);
Eric W. Biederman1efdb692012-02-07 16:54:11 -08001558 err = ipc_update_perm(&semid64.sem_perm, ipcp);
1559 if (err)
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001560 goto out_unlock0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 sma->sem_ctime = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 break;
1563 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 err = -EINVAL;
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001565 goto out_unlock1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001568out_unlock0:
Rik van Riel6062a8d2013-04-30 19:15:44 -07001569 sem_unlock(sma, -1);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001570out_unlock1:
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001571 rcu_read_unlock();
Pierre Peiffer522bb2a2008-04-29 01:00:49 -07001572out_up:
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001573 up_write(&sem_ids(ns).rwsem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574 return err;
1575}
1576
Al Viroe1fd1f42013-03-05 15:04:55 -05001577SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, unsigned long, arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579 int version;
Kirill Korotaeve3893532006-10-02 02:18:22 -07001580 struct ipc_namespace *ns;
Al Viroe1fd1f42013-03-05 15:04:55 -05001581 void __user *p = (void __user *)arg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582
1583 if (semid < 0)
1584 return -EINVAL;
1585
1586 version = ipc_parse_version(&cmd);
Kirill Korotaeve3893532006-10-02 02:18:22 -07001587 ns = current->nsproxy->ipc_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588
Manfred Spraul239521f2014-01-27 17:07:04 -08001589 switch (cmd) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 case IPC_INFO:
1591 case SEM_INFO:
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -08001592 case IPC_STAT:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593 case SEM_STAT:
Al Viroe1fd1f42013-03-05 15:04:55 -05001594 return semctl_nolock(ns, semid, cmd, version, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595 case GETALL:
1596 case GETVAL:
1597 case GETPID:
1598 case GETNCNT:
1599 case GETZCNT:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600 case SETALL:
Al Viroe1fd1f42013-03-05 15:04:55 -05001601 return semctl_main(ns, semid, semnum, cmd, p);
1602 case SETVAL:
1603 return semctl_setval(ns, semid, semnum, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 case IPC_RMID:
1605 case IPC_SET:
Al Viroe1fd1f42013-03-05 15:04:55 -05001606 return semctl_down(ns, semid, cmd, version, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607 default:
1608 return -EINVAL;
1609 }
1610}
1611
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612/* If the task doesn't already have a undo_list, then allocate one
1613 * here. We guarantee there is only one thread using this undo list,
1614 * and current is THE ONE
1615 *
1616 * If this allocation and assignment succeeds, but later
1617 * portions of this code fail, there is no need to free the sem_undo_list.
1618 * Just let it stay associated with the task, and it'll be freed later
1619 * at exit time.
1620 *
1621 * This can block, so callers must hold no locks.
1622 */
1623static inline int get_undo_list(struct sem_undo_list **undo_listp)
1624{
1625 struct sem_undo_list *undo_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626
1627 undo_list = current->sysvsem.undo_list;
1628 if (!undo_list) {
Matt Helsley2453a302006-10-02 02:18:25 -07001629 undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630 if (undo_list == NULL)
1631 return -ENOMEM;
Ingo Molnar00a5dfd2005-08-05 23:05:27 +02001632 spin_lock_init(&undo_list->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 atomic_set(&undo_list->refcnt, 1);
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001634 INIT_LIST_HEAD(&undo_list->list_proc);
1635
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 current->sysvsem.undo_list = undo_list;
1637 }
1638 *undo_listp = undo_list;
1639 return 0;
1640}
1641
Nick Pigginbf17bb72009-12-15 16:47:28 -08001642static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643{
Nick Pigginbf17bb72009-12-15 16:47:28 -08001644 struct sem_undo *un;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645
Nick Pigginbf17bb72009-12-15 16:47:28 -08001646 list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) {
1647 if (un->semid == semid)
1648 return un;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649 }
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001650 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651}
1652
Nick Pigginbf17bb72009-12-15 16:47:28 -08001653static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1654{
1655 struct sem_undo *un;
1656
Manfred Spraul239521f2014-01-27 17:07:04 -08001657 assert_spin_locked(&ulp->lock);
Nick Pigginbf17bb72009-12-15 16:47:28 -08001658
1659 un = __lookup_undo(ulp, semid);
1660 if (un) {
1661 list_del_rcu(&un->list_proc);
1662 list_add_rcu(&un->list_proc, &ulp->list_proc);
1663 }
1664 return un;
1665}
1666
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001667/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -08001668 * find_alloc_undo - lookup (and if not present create) undo array
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001669 * @ns: namespace
1670 * @semid: semaphore array id
1671 *
1672 * The function looks up (and if not present creates) the undo structure.
1673 * The size of the undo structure depends on the size of the semaphore
1674 * array, thus the alloc path is not that straightforward.
Manfred Spraul380af1b2008-07-25 01:48:06 -07001675 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1676 * performs a rcu_read_lock().
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001677 */
1678static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679{
1680 struct sem_array *sma;
1681 struct sem_undo_list *ulp;
1682 struct sem_undo *un, *new;
Rik van Riel6062a8d2013-04-30 19:15:44 -07001683 int nsems, error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684
1685 error = get_undo_list(&ulp);
1686 if (error)
1687 return ERR_PTR(error);
1688
Manfred Spraul380af1b2008-07-25 01:48:06 -07001689 rcu_read_lock();
Pierre Peifferc530c6a2007-10-18 23:40:55 -07001690 spin_lock(&ulp->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691 un = lookup_undo(ulp, semid);
Pierre Peifferc530c6a2007-10-18 23:40:55 -07001692 spin_unlock(&ulp->lock);
Manfred Spraul239521f2014-01-27 17:07:04 -08001693 if (likely(un != NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694 goto out;
1695
1696 /* no undo structure around - allocate one. */
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001697 /* step 1: figure out the size of the semaphore array */
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001698 sma = sem_obtain_object_check(ns, semid);
1699 if (IS_ERR(sma)) {
1700 rcu_read_unlock();
Julia Lawall4de85cd2010-05-26 14:43:44 -07001701 return ERR_CAST(sma);
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001702 }
Nadia Derbey023a5352007-10-18 23:40:51 -07001703
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704 nsems = sma->sem_nsems;
Rik van Riel6062a8d2013-04-30 19:15:44 -07001705 if (!ipc_rcu_getref(sma)) {
1706 rcu_read_unlock();
1707 un = ERR_PTR(-EIDRM);
1708 goto out;
1709 }
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001710 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001712 /* step 2: allocate new undo structure */
Burman Yan4668edc2006-12-06 20:38:51 -08001713 new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714 if (!new) {
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -07001715 ipc_rcu_putref(sma, ipc_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716 return ERR_PTR(-ENOMEM);
1717 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718
Manfred Spraul380af1b2008-07-25 01:48:06 -07001719 /* step 3: Acquire the lock on semaphore array */
Linus Torvalds4091fd942013-05-04 10:13:40 -07001720 rcu_read_lock();
Pierre Peiffer6ff37972008-04-29 01:00:46 -07001721 sem_lock_and_putref(sma);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001722 if (!ipc_valid_object(&sma->sem_perm)) {
Rik van Riel6062a8d2013-04-30 19:15:44 -07001723 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001724 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 kfree(new);
1726 un = ERR_PTR(-EIDRM);
1727 goto out;
1728 }
Manfred Spraul380af1b2008-07-25 01:48:06 -07001729 spin_lock(&ulp->lock);
1730
1731 /*
1732 * step 4: check for races: did someone else allocate the undo struct?
1733 */
1734 un = lookup_undo(ulp, semid);
1735 if (un) {
1736 kfree(new);
1737 goto success;
1738 }
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001739 /* step 5: initialize & link new undo structure */
1740 new->semadj = (short *) &new[1];
Manfred Spraul380af1b2008-07-25 01:48:06 -07001741 new->ulp = ulp;
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001742 new->semid = semid;
1743 assert_spin_locked(&ulp->lock);
Manfred Spraul380af1b2008-07-25 01:48:06 -07001744 list_add_rcu(&new->list_proc, &ulp->list_proc);
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -07001745 ipc_assert_locked_object(&sma->sem_perm);
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001746 list_add(&new->list_id, &sma->list_id);
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001747 un = new;
Manfred Spraul380af1b2008-07-25 01:48:06 -07001748
1749success:
1750 spin_unlock(&ulp->lock);
Rik van Riel6062a8d2013-04-30 19:15:44 -07001751 sem_unlock(sma, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752out:
1753 return un;
1754}
1755
Manfred Spraulc61284e2010-07-20 13:24:23 -07001756
1757/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -08001758 * get_queue_result - retrieve the result code from sem_queue
Manfred Spraulc61284e2010-07-20 13:24:23 -07001759 * @q: Pointer to queue structure
1760 *
1761 * Retrieve the return code from the pending queue. If IN_WAKEUP is found in
1762 * q->status, then we must loop until the value is replaced with the final
1763 * value: This may happen if a task is woken up by an unrelated event (e.g.
1764 * signal) and in parallel the task is woken up by another task because it got
1765 * the requested semaphores.
1766 *
1767 * The function can be called with or without holding the semaphore spinlock.
1768 */
1769static int get_queue_result(struct sem_queue *q)
1770{
1771 int error;
1772
1773 error = q->status;
1774 while (unlikely(error == IN_WAKEUP)) {
1775 cpu_relax();
1776 error = q->status;
1777 }
1778
1779 return error;
1780}
1781
Heiko Carstensd5460c92009-01-14 14:14:27 +01001782SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
1783 unsigned, nsops, const struct timespec __user *, timeout)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784{
1785 int error = -EINVAL;
1786 struct sem_array *sma;
1787 struct sembuf fast_sops[SEMOPM_FAST];
Manfred Spraul239521f2014-01-27 17:07:04 -08001788 struct sembuf *sops = fast_sops, *sop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789 struct sem_undo *un;
Rik van Riel6062a8d2013-04-30 19:15:44 -07001790 int undos = 0, alter = 0, max, locknum;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791 struct sem_queue queue;
1792 unsigned long jiffies_left = 0;
Kirill Korotaeve3893532006-10-02 02:18:22 -07001793 struct ipc_namespace *ns;
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001794 struct list_head tasks;
Kirill Korotaeve3893532006-10-02 02:18:22 -07001795
1796 ns = current->nsproxy->ipc_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797
1798 if (nsops < 1 || semid < 0)
1799 return -EINVAL;
Kirill Korotaeve3893532006-10-02 02:18:22 -07001800 if (nsops > ns->sc_semopm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801 return -E2BIG;
Manfred Spraul239521f2014-01-27 17:07:04 -08001802 if (nsops > SEMOPM_FAST) {
1803 sops = kmalloc(sizeof(*sops)*nsops, GFP_KERNEL);
1804 if (sops == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805 return -ENOMEM;
1806 }
Manfred Spraul239521f2014-01-27 17:07:04 -08001807 if (copy_from_user(sops, tsops, nsops * sizeof(*tsops))) {
1808 error = -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809 goto out_free;
1810 }
1811 if (timeout) {
1812 struct timespec _timeout;
1813 if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1814 error = -EFAULT;
1815 goto out_free;
1816 }
1817 if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1818 _timeout.tv_nsec >= 1000000000L) {
1819 error = -EINVAL;
1820 goto out_free;
1821 }
1822 jiffies_left = timespec_to_jiffies(&_timeout);
1823 }
1824 max = 0;
1825 for (sop = sops; sop < sops + nsops; sop++) {
1826 if (sop->sem_num >= max)
1827 max = sop->sem_num;
1828 if (sop->sem_flg & SEM_UNDO)
Manfred Spraulb78755a2005-06-23 00:10:06 -07001829 undos = 1;
1830 if (sop->sem_op != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831 alter = 1;
1832 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833
Rik van Riel6062a8d2013-04-30 19:15:44 -07001834 INIT_LIST_HEAD(&tasks);
1835
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836 if (undos) {
Rik van Riel6062a8d2013-04-30 19:15:44 -07001837 /* On success, find_alloc_undo takes the rcu_read_lock */
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001838 un = find_alloc_undo(ns, semid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839 if (IS_ERR(un)) {
1840 error = PTR_ERR(un);
1841 goto out_free;
1842 }
Rik van Riel6062a8d2013-04-30 19:15:44 -07001843 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844 un = NULL;
Rik van Riel6062a8d2013-04-30 19:15:44 -07001845 rcu_read_lock();
1846 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001848 sma = sem_obtain_object_check(ns, semid);
Nadia Derbey023a5352007-10-18 23:40:51 -07001849 if (IS_ERR(sma)) {
Rik van Riel6062a8d2013-04-30 19:15:44 -07001850 rcu_read_unlock();
Nadia Derbey023a5352007-10-18 23:40:51 -07001851 error = PTR_ERR(sma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852 goto out_free;
Nadia Derbey023a5352007-10-18 23:40:51 -07001853 }
1854
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001855 error = -EFBIG;
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001856 if (max >= sma->sem_nsems)
1857 goto out_rcu_wakeup;
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001858
1859 error = -EACCES;
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001860 if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO))
1861 goto out_rcu_wakeup;
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001862
1863 error = security_sem_semop(sma, sops, nsops, alter);
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001864 if (error)
1865 goto out_rcu_wakeup;
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001866
Manfred Spraul6e224f92013-10-16 13:46:45 -07001867 error = -EIDRM;
1868 locknum = sem_lock(sma, sops, nsops);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001869 /*
1870 * We eventually might perform the following check in a lockless
1871 * fashion, considering ipc_valid_object() locking constraints.
1872 * If nsops == 1 and there is no contention for sem_perm.lock, then
1873 * only a per-semaphore lock is held and it's OK to proceed with the
1874 * check below. More details on the fine grained locking scheme
1875 * entangled here and why it's RMID race safe on comments at sem_lock()
1876 */
1877 if (!ipc_valid_object(&sma->sem_perm))
Manfred Spraul6e224f92013-10-16 13:46:45 -07001878 goto out_unlock_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879 /*
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001880 * semid identifiers are not unique - find_alloc_undo may have
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881 * allocated an undo structure, it was invalidated by an RMID
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001882 * and now a new array with received the same id. Check and fail.
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001883 * This case can be detected checking un->semid. The existence of
Manfred Spraul380af1b2008-07-25 01:48:06 -07001884 * "un" itself is guaranteed by rcu.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885 */
Rik van Riel6062a8d2013-04-30 19:15:44 -07001886 if (un && un->semid == -1)
1887 goto out_unlock_free;
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001888
Manfred Sprauld198cd62014-06-06 14:37:49 -07001889 queue.sops = sops;
1890 queue.nsops = nsops;
1891 queue.undo = un;
1892 queue.pid = task_tgid_vnr(current);
1893 queue.alter = alter;
1894
1895 error = perform_atomic_semop(sma, &queue);
Manfred Spraul0e8c6652013-09-30 13:45:25 -07001896 if (error == 0) {
1897 /* If the operation was successful, then do
1898 * the required updates.
1899 */
1900 if (alter)
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001901 do_smart_update(sma, sops, nsops, 1, &tasks);
Manfred Spraul0e8c6652013-09-30 13:45:25 -07001902 else
1903 set_semotime(sma, sops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904 }
Manfred Spraul0e8c6652013-09-30 13:45:25 -07001905 if (error <= 0)
1906 goto out_unlock_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907
1908 /* We need to sleep on this operation, so we put the current
1909 * task into the pending queue and go to sleep.
1910 */
Paul McQuade46c0a8c2014-06-06 14:37:37 -07001911
Manfred Spraulb97e8202009-12-15 16:47:32 -08001912 if (nsops == 1) {
1913 struct sem *curr;
1914 curr = &sma->sem_base[sops->sem_num];
1915
Manfred Spraulf269f402013-07-08 16:01:24 -07001916 if (alter) {
1917 if (sma->complex_count) {
1918 list_add_tail(&queue.list,
1919 &sma->pending_alter);
1920 } else {
1921
1922 list_add_tail(&queue.list,
1923 &curr->pending_alter);
1924 }
1925 } else {
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001926 list_add_tail(&queue.list, &curr->pending_const);
Manfred Spraulf269f402013-07-08 16:01:24 -07001927 }
Manfred Spraulb97e8202009-12-15 16:47:32 -08001928 } else {
Manfred Spraulf269f402013-07-08 16:01:24 -07001929 if (!sma->complex_count)
1930 merge_queues(sma);
1931
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001932 if (alter)
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001933 list_add_tail(&queue.list, &sma->pending_alter);
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001934 else
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001935 list_add_tail(&queue.list, &sma->pending_const);
1936
Manfred Spraulb97e8202009-12-15 16:47:32 -08001937 sma->complex_count++;
1938 }
1939
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940 queue.status = -EINTR;
1941 queue.sleeper = current;
Manfred Spraul0b0577f2011-11-02 13:38:52 -07001942
1943sleep_again:
Davidlohr Bueso52644c92015-02-17 13:47:55 -08001944 __set_current_state(TASK_INTERRUPTIBLE);
Rik van Riel6062a8d2013-04-30 19:15:44 -07001945 sem_unlock(sma, locknum);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001946 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947
1948 if (timeout)
1949 jiffies_left = schedule_timeout(jiffies_left);
1950 else
1951 schedule();
1952
Manfred Spraulc61284e2010-07-20 13:24:23 -07001953 error = get_queue_result(&queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954
1955 if (error != -EINTR) {
1956 /* fast path: update_queue already obtained all requested
Manfred Spraulc61284e2010-07-20 13:24:23 -07001957 * resources.
1958 * Perform a smp_mb(): User space could assume that semop()
1959 * is a memory barrier: Without the mb(), the cpu could
1960 * speculatively read in user space stale data that was
1961 * overwritten by the previous owner of the semaphore.
1962 */
1963 smp_mb();
1964
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965 goto out_free;
1966 }
1967
Linus Torvalds321310c2013-05-04 10:47:57 -07001968 rcu_read_lock();
Rik van Riel6062a8d2013-04-30 19:15:44 -07001969 sma = sem_obtain_lock(ns, semid, sops, nsops, &locknum);
Manfred Sprauld694ad62011-07-25 17:11:47 -07001970
1971 /*
1972 * Wait until it's guaranteed that no wakeup_sem_queue_do() is ongoing.
1973 */
1974 error = get_queue_result(&queue);
1975
1976 /*
1977 * Array removed? If yes, leave without sem_unlock().
1978 */
Nadia Derbey023a5352007-10-18 23:40:51 -07001979 if (IS_ERR(sma)) {
Linus Torvalds321310c2013-05-04 10:47:57 -07001980 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981 goto out_free;
1982 }
1983
Manfred Spraulc61284e2010-07-20 13:24:23 -07001984
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985 /*
Manfred Sprauld694ad62011-07-25 17:11:47 -07001986 * If queue.status != -EINTR we are woken up by another process.
1987 * Leave without unlink_queue(), but with sem_unlock().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988 */
Davidlohr Bueso3ab08fe2014-01-27 17:07:06 -08001989 if (error != -EINTR)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990 goto out_unlock_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991
1992 /*
1993 * If an interrupt occurred we have to clean up the queue
1994 */
1995 if (timeout && jiffies_left == 0)
1996 error = -EAGAIN;
Manfred Spraul0b0577f2011-11-02 13:38:52 -07001997
1998 /*
1999 * If the wakeup was spurious, just retry
2000 */
2001 if (error == -EINTR && !signal_pending(current))
2002 goto sleep_again;
2003
Manfred Spraulb97e8202009-12-15 16:47:32 -08002004 unlink_queue(sma, &queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005
2006out_unlock_free:
Rik van Riel6062a8d2013-04-30 19:15:44 -07002007 sem_unlock(sma, locknum);
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07002008out_rcu_wakeup:
Linus Torvalds6d49dab2013-05-03 15:04:40 -07002009 rcu_read_unlock();
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07002010 wake_up_sem_queue_do(&tasks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011out_free:
Manfred Spraul239521f2014-01-27 17:07:04 -08002012 if (sops != fast_sops)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013 kfree(sops);
2014 return error;
2015}
2016
Heiko Carstensd5460c92009-01-14 14:14:27 +01002017SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
2018 unsigned, nsops)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019{
2020 return sys_semtimedop(semid, tsops, nsops, NULL);
2021}
2022
2023/* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
2024 * parent and child tasks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025 */
2026
2027int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
2028{
2029 struct sem_undo_list *undo_list;
2030 int error;
2031
2032 if (clone_flags & CLONE_SYSVSEM) {
2033 error = get_undo_list(&undo_list);
2034 if (error)
2035 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036 atomic_inc(&undo_list->refcnt);
2037 tsk->sysvsem.undo_list = undo_list;
Paul McQuade46c0a8c2014-06-06 14:37:37 -07002038 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039 tsk->sysvsem.undo_list = NULL;
2040
2041 return 0;
2042}
2043
2044/*
2045 * add semadj values to semaphores, free undo structures.
2046 * undo structures are not freed when semaphore arrays are destroyed
2047 * so some of them may be out of date.
2048 * IMPLEMENTATION NOTE: There is some confusion over whether the
2049 * set of adjustments that needs to be done should be done in an atomic
2050 * manner or not. That is, if we are attempting to decrement the semval
2051 * should we queue up and wait until we can do so legally?
2052 * The original implementation attempted to do this (queue and wait).
2053 * The current implementation does not do so. The POSIX standard
2054 * and SVID should be consulted to determine what behavior is mandated.
2055 */
2056void exit_sem(struct task_struct *tsk)
2057{
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002058 struct sem_undo_list *ulp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002060 ulp = tsk->sysvsem.undo_list;
2061 if (!ulp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062 return;
Manfred Spraul9edff4a2008-04-29 01:00:57 -07002063 tsk->sysvsem.undo_list = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002065 if (!atomic_dec_and_test(&ulp->refcnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066 return;
2067
Manfred Spraul380af1b2008-07-25 01:48:06 -07002068 for (;;) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069 struct sem_array *sma;
Manfred Spraul380af1b2008-07-25 01:48:06 -07002070 struct sem_undo *un;
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07002071 struct list_head tasks;
Rik van Riel6062a8d2013-04-30 19:15:44 -07002072 int semid, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073
Manfred Spraul380af1b2008-07-25 01:48:06 -07002074 rcu_read_lock();
Jiri Pirko05725f72009-04-14 20:17:16 +02002075 un = list_entry_rcu(ulp->list_proc.next,
2076 struct sem_undo, list_proc);
Manfred Spraul380af1b2008-07-25 01:48:06 -07002077 if (&un->list_proc == &ulp->list_proc)
2078 semid = -1;
2079 else
2080 semid = un->semid;
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002081
Rik van Riel6062a8d2013-04-30 19:15:44 -07002082 if (semid == -1) {
2083 rcu_read_unlock();
Manfred Spraul380af1b2008-07-25 01:48:06 -07002084 break;
Rik van Riel6062a8d2013-04-30 19:15:44 -07002085 }
Manfred Spraul380af1b2008-07-25 01:48:06 -07002086
Rik van Riel6062a8d2013-04-30 19:15:44 -07002087 sma = sem_obtain_object_check(tsk->nsproxy->ipc_ns, un->semid);
Manfred Spraul380af1b2008-07-25 01:48:06 -07002088 /* exit_sem raced with IPC_RMID, nothing to do */
Rik van Riel6062a8d2013-04-30 19:15:44 -07002089 if (IS_ERR(sma)) {
2090 rcu_read_unlock();
Manfred Spraul380af1b2008-07-25 01:48:06 -07002091 continue;
Rik van Riel6062a8d2013-04-30 19:15:44 -07002092 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093
Rik van Riel6062a8d2013-04-30 19:15:44 -07002094 sem_lock(sma, NULL, -1);
Manfred Spraul6e224f92013-10-16 13:46:45 -07002095 /* exit_sem raced with IPC_RMID, nothing to do */
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08002096 if (!ipc_valid_object(&sma->sem_perm)) {
Manfred Spraul6e224f92013-10-16 13:46:45 -07002097 sem_unlock(sma, -1);
2098 rcu_read_unlock();
2099 continue;
2100 }
Nick Pigginbf17bb72009-12-15 16:47:28 -08002101 un = __lookup_undo(ulp, semid);
Manfred Spraul380af1b2008-07-25 01:48:06 -07002102 if (un == NULL) {
2103 /* exit_sem raced with IPC_RMID+semget() that created
2104 * exactly the same semid. Nothing to do.
2105 */
Rik van Riel6062a8d2013-04-30 19:15:44 -07002106 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07002107 rcu_read_unlock();
Manfred Spraul380af1b2008-07-25 01:48:06 -07002108 continue;
2109 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110
Manfred Spraul380af1b2008-07-25 01:48:06 -07002111 /* remove un from the linked lists */
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -07002112 ipc_assert_locked_object(&sma->sem_perm);
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002113 list_del(&un->list_id);
2114
Manfred Spraul380af1b2008-07-25 01:48:06 -07002115 spin_lock(&ulp->lock);
2116 list_del_rcu(&un->list_proc);
2117 spin_unlock(&ulp->lock);
2118
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002119 /* perform adjustments registered in un */
2120 for (i = 0; i < sma->sem_nsems; i++) {
Manfred Spraul239521f2014-01-27 17:07:04 -08002121 struct sem *semaphore = &sma->sem_base[i];
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002122 if (un->semadj[i]) {
2123 semaphore->semval += un->semadj[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124 /*
2125 * Range checks of the new semaphore value,
2126 * not defined by sus:
2127 * - Some unices ignore the undo entirely
2128 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
2129 * - some cap the value (e.g. FreeBSD caps
2130 * at 0, but doesn't enforce SEMVMX)
2131 *
2132 * Linux caps the semaphore value, both at 0
2133 * and at SEMVMX.
2134 *
Manfred Spraul239521f2014-01-27 17:07:04 -08002135 * Manfred <manfred@colorfullife.com>
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136 */
Ingo Molnar5f921ae2006-03-26 01:37:17 -08002137 if (semaphore->semval < 0)
2138 semaphore->semval = 0;
2139 if (semaphore->semval > SEMVMX)
2140 semaphore->semval = SEMVMX;
Pavel Emelyanovb4888932007-10-18 23:40:14 -07002141 semaphore->sempid = task_tgid_vnr(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142 }
2143 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144 /* maybe some queued-up processes were waiting for this */
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07002145 INIT_LIST_HEAD(&tasks);
2146 do_smart_update(sma, NULL, 0, 1, &tasks);
Rik van Riel6062a8d2013-04-30 19:15:44 -07002147 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07002148 rcu_read_unlock();
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07002149 wake_up_sem_queue_do(&tasks);
Manfred Spraul380af1b2008-07-25 01:48:06 -07002150
Lai Jiangshan693a8b62011-03-18 12:09:35 +08002151 kfree_rcu(un, rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152 }
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002153 kfree(ulp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154}
2155
2156#ifdef CONFIG_PROC_FS
Mike Waychison19b49462005-09-06 15:17:10 -07002157static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158{
Eric W. Biederman1efdb692012-02-07 16:54:11 -08002159 struct user_namespace *user_ns = seq_user_ns(s);
Mike Waychison19b49462005-09-06 15:17:10 -07002160 struct sem_array *sma = it;
Manfred Sprauld12e1e52013-07-08 16:01:25 -07002161 time_t sem_otime;
2162
Manfred Sprauld8c63372013-09-30 13:45:07 -07002163 /*
2164 * The proc interface isn't aware of sem_lock(), it calls
2165 * ipc_lock_object() directly (in sysvipc_find_ipc).
2166 * In order to stay compatible with sem_lock(), we must wait until
2167 * all simple semop() calls have left their critical regions.
2168 */
2169 sem_wait_array(sma);
2170
Manfred Sprauld12e1e52013-07-08 16:01:25 -07002171 sem_otime = get_semotime(sma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002172
Joe Perches7f032d62015-04-15 16:17:54 -07002173 seq_printf(s,
2174 "%10d %10d %4o %10u %5u %5u %5u %5u %10lu %10lu\n",
2175 sma->sem_perm.key,
2176 sma->sem_perm.id,
2177 sma->sem_perm.mode,
2178 sma->sem_nsems,
2179 from_kuid_munged(user_ns, sma->sem_perm.uid),
2180 from_kgid_munged(user_ns, sma->sem_perm.gid),
2181 from_kuid_munged(user_ns, sma->sem_perm.cuid),
2182 from_kgid_munged(user_ns, sma->sem_perm.cgid),
2183 sem_otime,
2184 sma->sem_ctime);
2185
2186 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187}
2188#endif