blob: b471e5a3863ddbca70f2bf4dee22f40df0345fbe [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 Spraul3ed1f8a2015-08-14 15:35:10 -0700256 * spin_unlock_wait() and !spin_is_locked() are not memory barriers, they
257 * are only control barriers.
258 * The code must pair with spin_unlock(&sem->lock) or
259 * spin_unlock(&sem_perm.lock), thus just the control barrier is insufficient.
260 *
261 * smp_rmb() is sufficient, as writes cannot pass the control barrier.
262 */
263#define ipc_smp_acquire__after_spin_is_unlocked() smp_rmb()
264
265/*
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700266 * Wait until all currently ongoing simple ops have completed.
267 * Caller must own sem_perm.lock.
268 * New simple ops cannot start, because simple ops first check
269 * that sem_perm.lock is free.
Manfred Spraul6d07b682013-09-30 13:45:06 -0700270 * that a) sem_perm.lock is free and b) complex_count is 0.
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700271 */
272static void sem_wait_array(struct sem_array *sma)
273{
274 int i;
275 struct sem *sem;
276
Manfred Spraul6d07b682013-09-30 13:45:06 -0700277 if (sma->complex_count) {
278 /* The thread that increased sma->complex_count waited on
279 * all sem->lock locks. Thus we don't need to wait again.
280 */
281 return;
282 }
283
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700284 for (i = 0; i < sma->sem_nsems; i++) {
285 sem = sma->sem_base + i;
286 spin_unlock_wait(&sem->lock);
287 }
Manfred Spraul3ed1f8a2015-08-14 15:35:10 -0700288 ipc_smp_acquire__after_spin_is_unlocked();
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700289}
290
291/*
Rik van Riel6062a8d2013-04-30 19:15:44 -0700292 * If the request contains only one semaphore operation, and there are
293 * no complex transactions pending, lock only the semaphore involved.
294 * Otherwise, lock the entire semaphore array, since we either have
295 * multiple semaphores in our own semops, or we need to look at
296 * semaphores from other pending complex operations.
Rik van Riel6062a8d2013-04-30 19:15:44 -0700297 */
298static inline int sem_lock(struct sem_array *sma, struct sembuf *sops,
299 int nsops)
300{
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700301 struct sem *sem;
Rik van Riel6062a8d2013-04-30 19:15:44 -0700302
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700303 if (nsops != 1) {
304 /* Complex operation - acquire a full lock */
305 ipc_lock_object(&sma->sem_perm);
306
307 /* And wait until all simple ops that are processed
308 * right now have dropped their locks.
309 */
310 sem_wait_array(sma);
311 return -1;
312 }
313
314 /*
315 * Only one semaphore affected - try to optimize locking.
316 * The rules are:
317 * - optimized locking is possible if no complex operation
318 * is either enqueued or processed right now.
319 * - The test for enqueued complex ops is simple:
320 * sma->complex_count != 0
321 * - Testing for complex ops that are processed right now is
322 * a bit more difficult. Complex ops acquire the full lock
323 * and first wait that the running simple ops have completed.
324 * (see above)
325 * Thus: If we own a simple lock and the global lock is free
326 * and complex_count is now 0, then it will stay 0 and
327 * thus just locking sem->lock is sufficient.
328 */
329 sem = sma->sem_base + sops->sem_num;
330
331 if (sma->complex_count == 0) {
332 /*
333 * It appears that no complex operation is around.
334 * Acquire the per-semaphore lock.
335 */
Rik van Riel6062a8d2013-04-30 19:15:44 -0700336 spin_lock(&sem->lock);
337
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700338 /* Then check that the global lock is free */
339 if (!spin_is_locked(&sma->sem_perm.lock)) {
Manfred Spraul2e094ab2014-12-12 16:58:11 -0800340 /*
Manfred Spraul3ed1f8a2015-08-14 15:35:10 -0700341 * We need a memory barrier with acquire semantics,
342 * otherwise we can race with another thread that does:
Manfred Spraul2e094ab2014-12-12 16:58:11 -0800343 * complex_count++;
344 * spin_unlock(sem_perm.lock);
345 */
Manfred Spraul3ed1f8a2015-08-14 15:35:10 -0700346 ipc_smp_acquire__after_spin_is_unlocked();
Rik van Riel6062a8d2013-04-30 19:15:44 -0700347
Manfred Spraul2e094ab2014-12-12 16:58:11 -0800348 /*
349 * Now repeat the test of complex_count:
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700350 * It can't change anymore until we drop sem->lock.
351 * Thus: if is now 0, then it will stay 0.
352 */
353 if (sma->complex_count == 0) {
354 /* fast path successful! */
355 return sops->sem_num;
356 }
Rik van Riel6062a8d2013-04-30 19:15:44 -0700357 }
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700358 spin_unlock(&sem->lock);
Rik van Riel6062a8d2013-04-30 19:15:44 -0700359 }
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700360
361 /* slow path: acquire the full lock */
362 ipc_lock_object(&sma->sem_perm);
363
364 if (sma->complex_count == 0) {
365 /* False alarm:
366 * There is no complex operation, thus we can switch
367 * back to the fast path.
368 */
369 spin_lock(&sem->lock);
370 ipc_unlock_object(&sma->sem_perm);
371 return sops->sem_num;
372 } else {
373 /* Not a false alarm, thus complete the sequence for a
374 * full lock.
375 */
376 sem_wait_array(sma);
377 return -1;
378 }
Rik van Riel6062a8d2013-04-30 19:15:44 -0700379}
380
381static inline void sem_unlock(struct sem_array *sma, int locknum)
382{
383 if (locknum == -1) {
Manfred Spraulf269f402013-07-08 16:01:24 -0700384 unmerge_queues(sma);
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -0700385 ipc_unlock_object(&sma->sem_perm);
Rik van Riel6062a8d2013-04-30 19:15:44 -0700386 } else {
387 struct sem *sem = sma->sem_base + locknum;
388 spin_unlock(&sem->lock);
389 }
Rik van Riel6062a8d2013-04-30 19:15:44 -0700390}
391
392/*
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700393 * sem_lock_(check_) routines are called in the paths where the rwsem
Nadia Derbey3e148c72007-10-18 23:40:54 -0700394 * is not held.
Linus Torvalds321310c2013-05-04 10:47:57 -0700395 *
396 * The caller holds the RCU read lock.
Nadia Derbey3e148c72007-10-18 23:40:54 -0700397 */
Rik van Riel6062a8d2013-04-30 19:15:44 -0700398static inline struct sem_array *sem_obtain_lock(struct ipc_namespace *ns,
399 int id, struct sembuf *sops, int nsops, int *locknum)
Nadia Derbey023a5352007-10-18 23:40:51 -0700400{
Rik van Rielc460b662013-04-30 19:15:35 -0700401 struct kern_ipc_perm *ipcp;
402 struct sem_array *sma;
Nadia Derbey03f02c72007-10-18 23:40:51 -0700403
Davidlohr Bueso55b7ae52015-06-30 14:58:42 -0700404 ipcp = ipc_obtain_object_idr(&sem_ids(ns), id);
Linus Torvalds321310c2013-05-04 10:47:57 -0700405 if (IS_ERR(ipcp))
406 return ERR_CAST(ipcp);
Pierre Peifferb1ed88b2008-02-06 01:36:23 -0800407
Rik van Riel6062a8d2013-04-30 19:15:44 -0700408 sma = container_of(ipcp, struct sem_array, sem_perm);
409 *locknum = sem_lock(sma, sops, nsops);
Rik van Rielc460b662013-04-30 19:15:35 -0700410
411 /* ipc_rmid() may have already freed the ID while sem_lock
412 * was spinning: verify that the structure is still valid
413 */
Rafael Aquini72a8ff22014-01-27 17:07:02 -0800414 if (ipc_valid_object(ipcp))
Rik van Rielc460b662013-04-30 19:15:35 -0700415 return container_of(ipcp, struct sem_array, sem_perm);
416
Rik van Riel6062a8d2013-04-30 19:15:44 -0700417 sem_unlock(sma, *locknum);
Linus Torvalds321310c2013-05-04 10:47:57 -0700418 return ERR_PTR(-EINVAL);
Nadia Derbey023a5352007-10-18 23:40:51 -0700419}
420
Davidlohr Bueso16df3672013-04-30 19:15:29 -0700421static inline struct sem_array *sem_obtain_object(struct ipc_namespace *ns, int id)
422{
Davidlohr Bueso55b7ae52015-06-30 14:58:42 -0700423 struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&sem_ids(ns), id);
Davidlohr Bueso16df3672013-04-30 19:15:29 -0700424
425 if (IS_ERR(ipcp))
426 return ERR_CAST(ipcp);
427
428 return container_of(ipcp, struct sem_array, sem_perm);
429}
430
Davidlohr Bueso16df3672013-04-30 19:15:29 -0700431static inline struct sem_array *sem_obtain_object_check(struct ipc_namespace *ns,
432 int id)
433{
434 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&sem_ids(ns), id);
435
436 if (IS_ERR(ipcp))
437 return ERR_CAST(ipcp);
Pierre Peifferb1ed88b2008-02-06 01:36:23 -0800438
Nadia Derbey03f02c72007-10-18 23:40:51 -0700439 return container_of(ipcp, struct sem_array, sem_perm);
Nadia Derbey023a5352007-10-18 23:40:51 -0700440}
441
Pierre Peiffer6ff37972008-04-29 01:00:46 -0700442static inline void sem_lock_and_putref(struct sem_array *sma)
443{
Rik van Riel6062a8d2013-04-30 19:15:44 -0700444 sem_lock(sma, NULL, -1);
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -0700445 ipc_rcu_putref(sma, ipc_rcu_free);
Pierre Peiffer6ff37972008-04-29 01:00:46 -0700446}
447
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700448static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
449{
450 ipc_rmid(&sem_ids(ns), &s->sem_perm);
451}
452
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453/*
454 * Lockless wakeup algorithm:
455 * Without the check/retry algorithm a lockless wakeup is possible:
456 * - queue.status is initialized to -EINTR before blocking.
457 * - wakeup is performed by
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700458 * * unlinking the queue entry from the pending list
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 * * setting queue.status to IN_WAKEUP
460 * This is the notification for the blocked thread that a
461 * result value is imminent.
462 * * call wake_up_process
463 * * set queue.status to the final value.
464 * - the previously blocked thread checks queue.status:
Manfred Spraul239521f2014-01-27 17:07:04 -0800465 * * if it's IN_WAKEUP, then it must wait until the value changes
466 * * if it's not -EINTR, then the operation was completed by
467 * update_queue. semtimedop can return queue.status without
468 * performing any operation on the sem array.
469 * * otherwise it must acquire the spinlock and check what's up.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 *
471 * The two-stage algorithm is necessary to protect against the following
472 * races:
473 * - if queue.status is set after wake_up_process, then the woken up idle
474 * thread could race forward and try (and fail) to acquire sma->lock
475 * before update_queue had a chance to set queue.status
476 * - if queue.status is written before wake_up_process and if the
477 * blocked process is woken up by a signal between writing
478 * queue.status and the wake_up_process, then the woken up
479 * process could return from semtimedop and die by calling
480 * sys_exit before wake_up_process is called. Then wake_up_process
481 * will oops, because the task structure is already invalid.
482 * (yes, this happened on s390 with sysv msg).
483 *
484 */
485#define IN_WAKEUP 1
486
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700487/**
488 * newary - Create a new semaphore set
489 * @ns: namespace
490 * @params: ptr to the structure that contains key, semflg and nsems
491 *
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700492 * Called with sem_ids.rwsem held (as a writer)
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700493 */
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700494static int newary(struct ipc_namespace *ns, struct ipc_params *params)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495{
496 int id;
497 int retval;
498 struct sem_array *sma;
499 int size;
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700500 key_t key = params->key;
501 int nsems = params->u.nsems;
502 int semflg = params->flg;
Manfred Spraulb97e8202009-12-15 16:47:32 -0800503 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504
505 if (!nsems)
506 return -EINVAL;
Kirill Korotaeve3893532006-10-02 02:18:22 -0700507 if (ns->used_sems + nsems > ns->sc_semmns)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 return -ENOSPC;
509
Manfred Spraul239521f2014-01-27 17:07:04 -0800510 size = sizeof(*sma) + nsems * sizeof(struct sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 sma = ipc_rcu_alloc(size);
Davidlohr Bueso3ab08fe2014-01-27 17:07:06 -0800512 if (!sma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 return -ENOMEM;
Davidlohr Bueso3ab08fe2014-01-27 17:07:06 -0800514
Manfred Spraul239521f2014-01-27 17:07:04 -0800515 memset(sma, 0, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516
517 sma->sem_perm.mode = (semflg & S_IRWXUGO);
518 sma->sem_perm.key = key;
519
520 sma->sem_perm.security = NULL;
521 retval = security_sem_alloc(sma);
522 if (retval) {
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -0700523 ipc_rcu_putref(sma, ipc_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 return retval;
525 }
526
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 sma->sem_base = (struct sem *) &sma[1];
Manfred Spraulb97e8202009-12-15 16:47:32 -0800528
Rik van Riel6062a8d2013-04-30 19:15:44 -0700529 for (i = 0; i < nsems; i++) {
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700530 INIT_LIST_HEAD(&sma->sem_base[i].pending_alter);
531 INIT_LIST_HEAD(&sma->sem_base[i].pending_const);
Rik van Riel6062a8d2013-04-30 19:15:44 -0700532 spin_lock_init(&sma->sem_base[i].lock);
533 }
Manfred Spraulb97e8202009-12-15 16:47:32 -0800534
535 sma->complex_count = 0;
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700536 INIT_LIST_HEAD(&sma->pending_alter);
537 INIT_LIST_HEAD(&sma->pending_const);
Manfred Spraul4daa28f2008-07-25 01:48:04 -0700538 INIT_LIST_HEAD(&sma->list_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 sma->sem_nsems = nsems;
540 sma->sem_ctime = get_seconds();
Manfred Spraule8577d12014-12-02 15:59:34 -0800541
542 id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
543 if (id < 0) {
544 ipc_rcu_putref(sma, sem_rcu_free);
545 return id;
546 }
547 ns->used_sems += nsems;
548
Rik van Riel6062a8d2013-04-30 19:15:44 -0700549 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -0700550 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700552 return sma->sem_perm.id;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553}
554
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700555
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700556/*
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700557 * Called with sem_ids.rwsem and ipcp locked.
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700558 */
Nadia Derbey03f02c72007-10-18 23:40:51 -0700559static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560{
Nadia Derbey03f02c72007-10-18 23:40:51 -0700561 struct sem_array *sma;
562
563 sma = container_of(ipcp, struct sem_array, sem_perm);
564 return security_sem_associate(sma, semflg);
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700565}
566
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700567/*
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700568 * Called with sem_ids.rwsem and ipcp locked.
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700569 */
Nadia Derbey03f02c72007-10-18 23:40:51 -0700570static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
571 struct ipc_params *params)
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700572{
Nadia Derbey03f02c72007-10-18 23:40:51 -0700573 struct sem_array *sma;
574
575 sma = container_of(ipcp, struct sem_array, sem_perm);
576 if (params->u.nsems > sma->sem_nsems)
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700577 return -EINVAL;
578
579 return 0;
580}
581
Heiko Carstensd5460c92009-01-14 14:14:27 +0100582SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg)
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700583{
Kirill Korotaeve3893532006-10-02 02:18:22 -0700584 struct ipc_namespace *ns;
Mathias Krauseeb66ec42014-06-06 14:37:36 -0700585 static const struct ipc_ops sem_ops = {
586 .getnew = newary,
587 .associate = sem_security,
588 .more_checks = sem_more_checks,
589 };
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700590 struct ipc_params sem_params;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591
Kirill Korotaeve3893532006-10-02 02:18:22 -0700592 ns = current->nsproxy->ipc_ns;
593
594 if (nsems < 0 || nsems > ns->sc_semmsl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 return -EINVAL;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700596
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700597 sem_params.key = key;
598 sem_params.flg = semflg;
599 sem_params.u.nsems = nsems;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700600
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700601 return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602}
603
Petr Mladek78f50092014-01-27 17:07:00 -0800604/**
605 * perform_atomic_semop - Perform (if possible) a semaphore operation
Manfred Spraul758a6ba32013-07-08 16:01:26 -0700606 * @sma: semaphore array
Manfred Sprauld198cd62014-06-06 14:37:49 -0700607 * @q: struct sem_queue that describes the operation
Manfred Spraul758a6ba32013-07-08 16:01:26 -0700608 *
609 * Returns 0 if the operation was possible.
610 * Returns 1 if the operation is impossible, the caller must sleep.
611 * Negative values are error codes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 */
Manfred Sprauld198cd62014-06-06 14:37:49 -0700613static int perform_atomic_semop(struct sem_array *sma, struct sem_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614{
Manfred Sprauld198cd62014-06-06 14:37:49 -0700615 int result, sem_op, nsops, pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 struct sembuf *sop;
Manfred Spraul239521f2014-01-27 17:07:04 -0800617 struct sem *curr;
Manfred Sprauld198cd62014-06-06 14:37:49 -0700618 struct sembuf *sops;
619 struct sem_undo *un;
620
621 sops = q->sops;
622 nsops = q->nsops;
623 un = q->undo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624
625 for (sop = sops; sop < sops + nsops; sop++) {
626 curr = sma->sem_base + sop->sem_num;
627 sem_op = sop->sem_op;
628 result = curr->semval;
Petr Mladek78f50092014-01-27 17:07:00 -0800629
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 if (!sem_op && result)
631 goto would_block;
632
633 result += sem_op;
634 if (result < 0)
635 goto would_block;
636 if (result > SEMVMX)
637 goto out_of_range;
Petr Mladek78f50092014-01-27 17:07:00 -0800638
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 if (sop->sem_flg & SEM_UNDO) {
640 int undo = un->semadj[sop->sem_num] - sem_op;
Petr Mladek78f50092014-01-27 17:07:00 -0800641 /* Exceeding the undo range is an error. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
643 goto out_of_range;
Petr Mladek78f50092014-01-27 17:07:00 -0800644 un->semadj[sop->sem_num] = undo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 }
Petr Mladek78f50092014-01-27 17:07:00 -0800646
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 curr->semval = result;
648 }
649
650 sop--;
Manfred Sprauld198cd62014-06-06 14:37:49 -0700651 pid = q->pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 while (sop >= sops) {
653 sma->sem_base[sop->sem_num].sempid = pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 sop--;
655 }
Petr Mladek78f50092014-01-27 17:07:00 -0800656
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 return 0;
658
659out_of_range:
660 result = -ERANGE;
661 goto undo;
662
663would_block:
Manfred Sprauled247b72014-06-06 14:37:49 -0700664 q->blocking = sop;
665
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 if (sop->sem_flg & IPC_NOWAIT)
667 result = -EAGAIN;
668 else
669 result = 1;
670
671undo:
672 sop--;
673 while (sop >= sops) {
Petr Mladek78f50092014-01-27 17:07:00 -0800674 sem_op = sop->sem_op;
675 sma->sem_base[sop->sem_num].semval -= sem_op;
676 if (sop->sem_flg & SEM_UNDO)
677 un->semadj[sop->sem_num] += sem_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 sop--;
679 }
680
681 return result;
682}
683
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700684/** wake_up_sem_queue_prepare(q, error): Prepare wake-up
685 * @q: queue entry that must be signaled
686 * @error: Error value for the signal
687 *
688 * Prepare the wake-up of the queue entry q.
Nick Piggind4212092009-12-15 16:47:30 -0800689 */
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700690static void wake_up_sem_queue_prepare(struct list_head *pt,
691 struct sem_queue *q, int error)
Nick Piggind4212092009-12-15 16:47:30 -0800692{
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700693 if (list_empty(pt)) {
694 /*
695 * Hold preempt off so that we don't get preempted and have the
696 * wakee busy-wait until we're scheduled back on.
697 */
698 preempt_disable();
699 }
Nick Piggind4212092009-12-15 16:47:30 -0800700 q->status = IN_WAKEUP;
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700701 q->pid = error;
702
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700703 list_add_tail(&q->list, pt);
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700704}
705
706/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800707 * wake_up_sem_queue_do - do the actual wake-up
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700708 * @pt: list of tasks to be woken up
709 *
710 * Do the actual wake-up.
711 * The function is called without any locks held, thus the semaphore array
712 * could be destroyed already and the tasks can disappear as soon as the
713 * status is set to the actual return code.
714 */
715static void wake_up_sem_queue_do(struct list_head *pt)
716{
717 struct sem_queue *q, *t;
718 int did_something;
719
720 did_something = !list_empty(pt);
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700721 list_for_each_entry_safe(q, t, pt, list) {
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700722 wake_up_process(q->sleeper);
723 /* q can disappear immediately after writing q->status. */
724 smp_wmb();
725 q->status = q->pid;
726 }
727 if (did_something)
728 preempt_enable();
Nick Piggind4212092009-12-15 16:47:30 -0800729}
730
Manfred Spraulb97e8202009-12-15 16:47:32 -0800731static void unlink_queue(struct sem_array *sma, struct sem_queue *q)
732{
733 list_del(&q->list);
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700734 if (q->nsops > 1)
Manfred Spraulb97e8202009-12-15 16:47:32 -0800735 sma->complex_count--;
736}
737
Manfred Spraulfd5db422010-05-26 14:43:40 -0700738/** check_restart(sma, q)
739 * @sma: semaphore array
740 * @q: the operation that just completed
741 *
742 * update_queue is O(N^2) when it restarts scanning the whole queue of
743 * waiting operations. Therefore this function checks if the restart is
744 * really necessary. It is called after a previously waiting operation
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700745 * modified the array.
746 * Note that wait-for-zero operations are handled without restart.
Manfred Spraulfd5db422010-05-26 14:43:40 -0700747 */
748static int check_restart(struct sem_array *sma, struct sem_queue *q)
749{
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700750 /* pending complex alter operations are too difficult to analyse */
751 if (!list_empty(&sma->pending_alter))
Manfred Spraulfd5db422010-05-26 14:43:40 -0700752 return 1;
753
754 /* we were a sleeping complex operation. Too difficult */
755 if (q->nsops > 1)
756 return 1;
757
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700758 /* It is impossible that someone waits for the new value:
759 * - complex operations always restart.
760 * - wait-for-zero are handled seperately.
761 * - q is a previously sleeping simple operation that
762 * altered the array. It must be a decrement, because
763 * simple increments never sleep.
764 * - If there are older (higher priority) decrements
765 * in the queue, then they have observed the original
766 * semval value and couldn't proceed. The operation
767 * decremented to value - thus they won't proceed either.
768 */
769 return 0;
770}
Manfred Spraulfd5db422010-05-26 14:43:40 -0700771
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700772/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800773 * wake_const_ops - wake up non-alter tasks
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700774 * @sma: semaphore array.
775 * @semnum: semaphore that was modified.
776 * @pt: list head for the tasks that must be woken up.
777 *
778 * wake_const_ops must be called after a semaphore in a semaphore array
779 * was set to 0. If complex const operations are pending, wake_const_ops must
780 * be called with semnum = -1, as well as with the number of each modified
781 * semaphore.
782 * The tasks that must be woken up are added to @pt. The return code
783 * is stored in q->pid.
784 * The function returns 1 if at least one operation was completed successfully.
785 */
786static int wake_const_ops(struct sem_array *sma, int semnum,
787 struct list_head *pt)
788{
789 struct sem_queue *q;
790 struct list_head *walk;
791 struct list_head *pending_list;
792 int semop_completed = 0;
Manfred Spraulfd5db422010-05-26 14:43:40 -0700793
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700794 if (semnum == -1)
795 pending_list = &sma->pending_const;
796 else
797 pending_list = &sma->sem_base[semnum].pending_const;
798
799 walk = pending_list->next;
800 while (walk != pending_list) {
801 int error;
802
803 q = container_of(walk, struct sem_queue, list);
804 walk = walk->next;
805
Manfred Sprauld198cd62014-06-06 14:37:49 -0700806 error = perform_atomic_semop(sma, q);
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700807
808 if (error <= 0) {
809 /* operation completed, remove from queue & wakeup */
810
811 unlink_queue(sma, q);
812
813 wake_up_sem_queue_prepare(pt, q, error);
814 if (error == 0)
815 semop_completed = 1;
816 }
817 }
818 return semop_completed;
819}
820
821/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800822 * do_smart_wakeup_zero - wakeup all wait for zero tasks
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700823 * @sma: semaphore array
824 * @sops: operations that were performed
825 * @nsops: number of operations
826 * @pt: list head of the tasks that must be woken up.
827 *
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800828 * Checks all required queue for wait-for-zero operations, based
829 * on the actual changes that were performed on the semaphore array.
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700830 * The function returns 1 if at least one operation was completed successfully.
831 */
832static int do_smart_wakeup_zero(struct sem_array *sma, struct sembuf *sops,
833 int nsops, struct list_head *pt)
834{
835 int i;
836 int semop_completed = 0;
837 int got_zero = 0;
838
839 /* first: the per-semaphore queues, if known */
840 if (sops) {
841 for (i = 0; i < nsops; i++) {
842 int num = sops[i].sem_num;
843
844 if (sma->sem_base[num].semval == 0) {
845 got_zero = 1;
846 semop_completed |= wake_const_ops(sma, num, pt);
847 }
848 }
849 } else {
850 /*
851 * No sops means modified semaphores not known.
852 * Assume all were changed.
Manfred Spraulfd5db422010-05-26 14:43:40 -0700853 */
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700854 for (i = 0; i < sma->sem_nsems; i++) {
855 if (sma->sem_base[i].semval == 0) {
856 got_zero = 1;
857 semop_completed |= wake_const_ops(sma, i, pt);
858 }
859 }
Manfred Spraulfd5db422010-05-26 14:43:40 -0700860 }
861 /*
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700862 * If one of the modified semaphores got 0,
863 * then check the global queue, too.
Manfred Spraulfd5db422010-05-26 14:43:40 -0700864 */
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700865 if (got_zero)
866 semop_completed |= wake_const_ops(sma, -1, pt);
Manfred Spraulfd5db422010-05-26 14:43:40 -0700867
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700868 return semop_completed;
Manfred Spraulfd5db422010-05-26 14:43:40 -0700869}
870
Manfred Spraul636c6be2009-12-15 16:47:33 -0800871
872/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800873 * update_queue - look for tasks that can be completed.
Manfred Spraul636c6be2009-12-15 16:47:33 -0800874 * @sma: semaphore array.
875 * @semnum: semaphore that was modified.
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700876 * @pt: list head for the tasks that must be woken up.
Manfred Spraul636c6be2009-12-15 16:47:33 -0800877 *
878 * update_queue must be called after a semaphore in a semaphore array
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700879 * was modified. If multiple semaphores were modified, update_queue must
880 * be called with semnum = -1, as well as with the number of each modified
881 * semaphore.
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700882 * The tasks that must be woken up are added to @pt. The return code
883 * is stored in q->pid.
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700884 * The function internally checks if const operations can now succeed.
885 *
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700886 * The function return 1 if at least one semop was completed successfully.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 */
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700888static int update_queue(struct sem_array *sma, int semnum, struct list_head *pt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889{
Manfred Spraul636c6be2009-12-15 16:47:33 -0800890 struct sem_queue *q;
891 struct list_head *walk;
892 struct list_head *pending_list;
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700893 int semop_completed = 0;
Manfred Spraul636c6be2009-12-15 16:47:33 -0800894
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700895 if (semnum == -1)
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700896 pending_list = &sma->pending_alter;
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700897 else
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700898 pending_list = &sma->sem_base[semnum].pending_alter;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899
Nick Piggin9cad2002009-12-15 16:47:29 -0800900again:
Manfred Spraul636c6be2009-12-15 16:47:33 -0800901 walk = pending_list->next;
902 while (walk != pending_list) {
Manfred Spraulfd5db422010-05-26 14:43:40 -0700903 int error, restart;
Manfred Spraul636c6be2009-12-15 16:47:33 -0800904
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700905 q = container_of(walk, struct sem_queue, list);
Manfred Spraul636c6be2009-12-15 16:47:33 -0800906 walk = walk->next;
Nick Piggin9cad2002009-12-15 16:47:29 -0800907
Manfred Sprauld987f8b2009-12-15 16:47:34 -0800908 /* If we are scanning the single sop, per-semaphore list of
909 * one semaphore and that semaphore is 0, then it is not
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700910 * necessary to scan further: simple increments
Manfred Sprauld987f8b2009-12-15 16:47:34 -0800911 * that affect only one entry succeed immediately and cannot
912 * be in the per semaphore pending queue, and decrements
913 * cannot be successful if the value is already 0.
914 */
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700915 if (semnum != -1 && sma->sem_base[semnum].semval == 0)
Manfred Sprauld987f8b2009-12-15 16:47:34 -0800916 break;
917
Manfred Sprauld198cd62014-06-06 14:37:49 -0700918 error = perform_atomic_semop(sma, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919
920 /* Does q->sleeper still need to sleep? */
Nick Piggin9cad2002009-12-15 16:47:29 -0800921 if (error > 0)
922 continue;
Manfred Spraula1193f82008-07-25 01:48:06 -0700923
Manfred Spraulb97e8202009-12-15 16:47:32 -0800924 unlink_queue(sma, q);
Manfred Spraula1193f82008-07-25 01:48:06 -0700925
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700926 if (error) {
Manfred Spraulfd5db422010-05-26 14:43:40 -0700927 restart = 0;
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700928 } else {
929 semop_completed = 1;
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700930 do_smart_wakeup_zero(sma, q->sops, q->nsops, pt);
Manfred Spraulfd5db422010-05-26 14:43:40 -0700931 restart = check_restart(sma, q);
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700932 }
Manfred Spraulfd5db422010-05-26 14:43:40 -0700933
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700934 wake_up_sem_queue_prepare(pt, q, error);
Manfred Spraulfd5db422010-05-26 14:43:40 -0700935 if (restart)
Nick Piggin9cad2002009-12-15 16:47:29 -0800936 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 }
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700938 return semop_completed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939}
940
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700941/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800942 * set_semotime - set sem_otime
Manfred Spraul0e8c6652013-09-30 13:45:25 -0700943 * @sma: semaphore array
944 * @sops: operations that modified the array, may be NULL
945 *
946 * sem_otime is replicated to avoid cache line trashing.
947 * This function sets one instance to the current time.
948 */
949static void set_semotime(struct sem_array *sma, struct sembuf *sops)
950{
951 if (sops == NULL) {
952 sma->sem_base[0].sem_otime = get_seconds();
953 } else {
954 sma->sem_base[sops[0].sem_num].sem_otime =
955 get_seconds();
956 }
957}
958
959/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800960 * do_smart_update - optimized update_queue
Manfred Spraulfd5db422010-05-26 14:43:40 -0700961 * @sma: semaphore array
962 * @sops: operations that were performed
963 * @nsops: number of operations
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700964 * @otime: force setting otime
965 * @pt: list head of the tasks that must be woken up.
Manfred Spraulfd5db422010-05-26 14:43:40 -0700966 *
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700967 * do_smart_update() does the required calls to update_queue and wakeup_zero,
968 * based on the actual changes that were performed on the semaphore array.
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700969 * Note that the function does not do the actual wake-up: the caller is
970 * responsible for calling wake_up_sem_queue_do(@pt).
971 * It is safe to perform this call after dropping all locks.
Manfred Spraulfd5db422010-05-26 14:43:40 -0700972 */
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700973static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops,
974 int otime, struct list_head *pt)
Manfred Spraulfd5db422010-05-26 14:43:40 -0700975{
976 int i;
977
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700978 otime |= do_smart_wakeup_zero(sma, sops, nsops, pt);
979
Manfred Spraulf269f402013-07-08 16:01:24 -0700980 if (!list_empty(&sma->pending_alter)) {
981 /* semaphore array uses the global queue - just process it. */
982 otime |= update_queue(sma, -1, pt);
983 } else {
984 if (!sops) {
985 /*
986 * No sops, thus the modified semaphores are not
987 * known. Check all.
988 */
989 for (i = 0; i < sma->sem_nsems; i++)
990 otime |= update_queue(sma, i, pt);
991 } else {
992 /*
993 * Check the semaphores that were increased:
994 * - No complex ops, thus all sleeping ops are
995 * decrease.
996 * - if we decreased the value, then any sleeping
997 * semaphore ops wont be able to run: If the
998 * previous value was too small, then the new
999 * value will be too small, too.
1000 */
1001 for (i = 0; i < nsops; i++) {
1002 if (sops[i].sem_op > 0) {
1003 otime |= update_queue(sma,
1004 sops[i].sem_num, pt);
1005 }
Manfred Spraulab465df2013-05-26 11:08:52 +02001006 }
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001007 }
Manfred Spraulfd5db422010-05-26 14:43:40 -07001008 }
Manfred Spraul0e8c6652013-09-30 13:45:25 -07001009 if (otime)
1010 set_semotime(sma, sops);
Manfred Spraulfd5db422010-05-26 14:43:40 -07001011}
1012
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001013/*
Manfred Spraulb220c572014-06-06 14:37:51 -07001014 * check_qop: Test if a queued operation sleeps on the semaphore semnum
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001015 */
1016static int check_qop(struct sem_array *sma, int semnum, struct sem_queue *q,
1017 bool count_zero)
1018{
Manfred Spraulb220c572014-06-06 14:37:51 -07001019 struct sembuf *sop = q->blocking;
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001020
Manfred Spraul9b44ee22014-06-06 14:37:52 -07001021 /*
1022 * Linux always (since 0.99.10) reported a task as sleeping on all
1023 * semaphores. This violates SUS, therefore it was changed to the
1024 * standard compliant behavior.
1025 * Give the administrators a chance to notice that an application
1026 * might misbehave because it relies on the Linux behavior.
1027 */
1028 pr_info_once("semctl(GETNCNT/GETZCNT) is since 3.16 Single Unix Specification compliant.\n"
1029 "The task %s (%d) triggered the difference, watch for misbehavior.\n",
1030 current->comm, task_pid_nr(current));
1031
Manfred Spraulb220c572014-06-06 14:37:51 -07001032 if (sop->sem_num != semnum)
1033 return 0;
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001034
Manfred Spraulb220c572014-06-06 14:37:51 -07001035 if (count_zero && sop->sem_op == 0)
1036 return 1;
1037 if (!count_zero && sop->sem_op < 0)
1038 return 1;
1039
1040 return 0;
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001041}
1042
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043/* The following counts are associated to each semaphore:
1044 * semncnt number of tasks waiting on semval being nonzero
1045 * semzcnt number of tasks waiting on semval being zero
Manfred Spraulb220c572014-06-06 14:37:51 -07001046 *
1047 * Per definition, a task waits only on the semaphore of the first semop
1048 * that cannot proceed, even if additional operation would block, too.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 */
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001050static int count_semcnt(struct sem_array *sma, ushort semnum,
1051 bool count_zero)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052{
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001053 struct list_head *l;
Manfred Spraul239521f2014-01-27 17:07:04 -08001054 struct sem_queue *q;
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001055 int semcnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001057 semcnt = 0;
1058 /* First: check the simple operations. They are easy to evaluate */
1059 if (count_zero)
1060 l = &sma->sem_base[semnum].pending_const;
1061 else
1062 l = &sma->sem_base[semnum].pending_alter;
1063
1064 list_for_each_entry(q, l, list) {
1065 /* all task on a per-semaphore list sleep on exactly
1066 * that semaphore
1067 */
1068 semcnt++;
Rik van Rielde2657f2013-05-09 16:59:59 -04001069 }
1070
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001071 /* Then: check the complex operations. */
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001072 list_for_each_entry(q, &sma->pending_alter, list) {
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001073 semcnt += check_qop(sma, semnum, q, count_zero);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 }
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001075 if (count_zero) {
1076 list_for_each_entry(q, &sma->pending_const, list) {
1077 semcnt += check_qop(sma, semnum, q, count_zero);
1078 }
Rik van Rielebc2e5e2013-05-09 16:53:28 -04001079 }
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001080 return semcnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081}
1082
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001083/* Free a semaphore set. freeary() is called with sem_ids.rwsem locked
1084 * as a writer and the spinlock for this semaphore set hold. sem_ids.rwsem
Nadia Derbey3e148c72007-10-18 23:40:54 -07001085 * remains locked on exit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 */
Pierre Peiffer01b8b072008-02-08 04:18:57 -08001087static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088{
Manfred Spraul380af1b2008-07-25 01:48:06 -07001089 struct sem_undo *un, *tu;
1090 struct sem_queue *q, *tq;
Pierre Peiffer01b8b072008-02-08 04:18:57 -08001091 struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001092 struct list_head tasks;
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001093 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094
Manfred Spraul380af1b2008-07-25 01:48:06 -07001095 /* Free the existing undo structures for this semaphore set. */
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -07001096 ipc_assert_locked_object(&sma->sem_perm);
Manfred Spraul380af1b2008-07-25 01:48:06 -07001097 list_for_each_entry_safe(un, tu, &sma->list_id, list_id) {
1098 list_del(&un->list_id);
1099 spin_lock(&un->ulp->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 un->semid = -1;
Manfred Spraul380af1b2008-07-25 01:48:06 -07001101 list_del_rcu(&un->list_proc);
1102 spin_unlock(&un->ulp->lock);
Lai Jiangshan693a8b62011-03-18 12:09:35 +08001103 kfree_rcu(un, rcu);
Manfred Spraul380af1b2008-07-25 01:48:06 -07001104 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105
1106 /* Wake up all pending processes and let them fail with EIDRM. */
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001107 INIT_LIST_HEAD(&tasks);
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001108 list_for_each_entry_safe(q, tq, &sma->pending_const, list) {
1109 unlink_queue(sma, q);
1110 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1111 }
1112
1113 list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
Manfred Spraulb97e8202009-12-15 16:47:32 -08001114 unlink_queue(sma, q);
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001115 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 }
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001117 for (i = 0; i < sma->sem_nsems; i++) {
1118 struct sem *sem = sma->sem_base + i;
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001119 list_for_each_entry_safe(q, tq, &sem->pending_const, list) {
1120 unlink_queue(sma, q);
1121 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1122 }
1123 list_for_each_entry_safe(q, tq, &sem->pending_alter, list) {
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001124 unlink_queue(sma, q);
1125 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1126 }
1127 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128
Nadia Derbey7ca7e562007-10-18 23:40:48 -07001129 /* Remove the semaphore set from the IDR */
1130 sem_rmid(ns, sma);
Rik van Riel6062a8d2013-04-30 19:15:44 -07001131 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001132 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001134 wake_up_sem_queue_do(&tasks);
Kirill Korotaeve3893532006-10-02 02:18:22 -07001135 ns->used_sems -= sma->sem_nsems;
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -07001136 ipc_rcu_putref(sma, sem_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137}
1138
1139static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
1140{
Manfred Spraul239521f2014-01-27 17:07:04 -08001141 switch (version) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142 case IPC_64:
1143 return copy_to_user(buf, in, sizeof(*in));
1144 case IPC_OLD:
1145 {
1146 struct semid_ds out;
1147
Dan Rosenberg982f7c22010-09-30 15:15:31 -07001148 memset(&out, 0, sizeof(out));
1149
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
1151
1152 out.sem_otime = in->sem_otime;
1153 out.sem_ctime = in->sem_ctime;
1154 out.sem_nsems = in->sem_nsems;
1155
1156 return copy_to_user(buf, &out, sizeof(out));
1157 }
1158 default:
1159 return -EINVAL;
1160 }
1161}
1162
Manfred Sprauld12e1e52013-07-08 16:01:25 -07001163static time_t get_semotime(struct sem_array *sma)
1164{
1165 int i;
1166 time_t res;
1167
1168 res = sma->sem_base[0].sem_otime;
1169 for (i = 1; i < sma->sem_nsems; i++) {
1170 time_t to = sma->sem_base[i].sem_otime;
1171
1172 if (to > res)
1173 res = to;
1174 }
1175 return res;
1176}
1177
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -08001178static int semctl_nolock(struct ipc_namespace *ns, int semid,
Al Viroe1fd1f42013-03-05 15:04:55 -05001179 int cmd, int version, void __user *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180{
Amerigo Wange5cc9c72009-12-15 16:47:35 -08001181 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 struct sem_array *sma;
1183
Manfred Spraul239521f2014-01-27 17:07:04 -08001184 switch (cmd) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185 case IPC_INFO:
1186 case SEM_INFO:
1187 {
1188 struct seminfo seminfo;
1189 int max_id;
1190
1191 err = security_sem_semctl(NULL, cmd);
1192 if (err)
1193 return err;
Paul McQuade46c0a8c2014-06-06 14:37:37 -07001194
Manfred Spraul239521f2014-01-27 17:07:04 -08001195 memset(&seminfo, 0, sizeof(seminfo));
Kirill Korotaeve3893532006-10-02 02:18:22 -07001196 seminfo.semmni = ns->sc_semmni;
1197 seminfo.semmns = ns->sc_semmns;
1198 seminfo.semmsl = ns->sc_semmsl;
1199 seminfo.semopm = ns->sc_semopm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200 seminfo.semvmx = SEMVMX;
1201 seminfo.semmnu = SEMMNU;
1202 seminfo.semmap = SEMMAP;
1203 seminfo.semume = SEMUME;
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001204 down_read(&sem_ids(ns).rwsem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 if (cmd == SEM_INFO) {
Kirill Korotaeve3893532006-10-02 02:18:22 -07001206 seminfo.semusz = sem_ids(ns).in_use;
1207 seminfo.semaem = ns->used_sems;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208 } else {
1209 seminfo.semusz = SEMUSZ;
1210 seminfo.semaem = SEMAEM;
1211 }
Nadia Derbey7ca7e562007-10-18 23:40:48 -07001212 max_id = ipc_get_maxid(&sem_ids(ns));
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001213 up_read(&sem_ids(ns).rwsem);
Paul McQuade46c0a8c2014-06-06 14:37:37 -07001214 if (copy_to_user(p, &seminfo, sizeof(struct seminfo)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 return -EFAULT;
Manfred Spraul239521f2014-01-27 17:07:04 -08001216 return (max_id < 0) ? 0 : max_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 }
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -08001218 case IPC_STAT:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 case SEM_STAT:
1220 {
1221 struct semid64_ds tbuf;
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001222 int id = 0;
1223
1224 memset(&tbuf, 0, sizeof(tbuf));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225
Linus Torvalds941b0302013-05-04 11:04:29 -07001226 rcu_read_lock();
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -08001227 if (cmd == SEM_STAT) {
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001228 sma = sem_obtain_object(ns, semid);
1229 if (IS_ERR(sma)) {
1230 err = PTR_ERR(sma);
1231 goto out_unlock;
1232 }
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -08001233 id = sma->sem_perm.id;
1234 } else {
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001235 sma = sem_obtain_object_check(ns, semid);
1236 if (IS_ERR(sma)) {
1237 err = PTR_ERR(sma);
1238 goto out_unlock;
1239 }
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -08001240 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241
1242 err = -EACCES;
Serge E. Hallynb0e77592011-03-23 16:43:24 -07001243 if (ipcperms(ns, &sma->sem_perm, S_IRUGO))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 goto out_unlock;
1245
1246 err = security_sem_semctl(sma, cmd);
1247 if (err)
1248 goto out_unlock;
1249
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
Manfred Sprauld12e1e52013-07-08 16:01:25 -07001251 tbuf.sem_otime = get_semotime(sma);
1252 tbuf.sem_ctime = sma->sem_ctime;
1253 tbuf.sem_nsems = sma->sem_nsems;
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001254 rcu_read_unlock();
Al Viroe1fd1f42013-03-05 15:04:55 -05001255 if (copy_semid_to_user(p, &tbuf, version))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 return -EFAULT;
1257 return id;
1258 }
1259 default:
1260 return -EINVAL;
1261 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262out_unlock:
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001263 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 return err;
1265}
1266
Al Viroe1fd1f42013-03-05 15:04:55 -05001267static int semctl_setval(struct ipc_namespace *ns, int semid, int semnum,
1268 unsigned long arg)
1269{
1270 struct sem_undo *un;
1271 struct sem_array *sma;
Manfred Spraul239521f2014-01-27 17:07:04 -08001272 struct sem *curr;
Al Viroe1fd1f42013-03-05 15:04:55 -05001273 int err;
Al Viroe1fd1f42013-03-05 15:04:55 -05001274 struct list_head tasks;
1275 int val;
1276#if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN)
1277 /* big-endian 64bit */
1278 val = arg >> 32;
1279#else
1280 /* 32bit or little-endian 64bit */
1281 val = arg;
1282#endif
1283
Rik van Riel6062a8d2013-04-30 19:15:44 -07001284 if (val > SEMVMX || val < 0)
1285 return -ERANGE;
Al Viroe1fd1f42013-03-05 15:04:55 -05001286
1287 INIT_LIST_HEAD(&tasks);
Al Viroe1fd1f42013-03-05 15:04:55 -05001288
Rik van Riel6062a8d2013-04-30 19:15:44 -07001289 rcu_read_lock();
1290 sma = sem_obtain_object_check(ns, semid);
1291 if (IS_ERR(sma)) {
1292 rcu_read_unlock();
1293 return PTR_ERR(sma);
1294 }
1295
1296 if (semnum < 0 || semnum >= sma->sem_nsems) {
1297 rcu_read_unlock();
1298 return -EINVAL;
1299 }
1300
1301
1302 if (ipcperms(ns, &sma->sem_perm, S_IWUGO)) {
1303 rcu_read_unlock();
1304 return -EACCES;
1305 }
Al Viroe1fd1f42013-03-05 15:04:55 -05001306
1307 err = security_sem_semctl(sma, SETVAL);
Rik van Riel6062a8d2013-04-30 19:15:44 -07001308 if (err) {
1309 rcu_read_unlock();
1310 return -EACCES;
1311 }
Al Viroe1fd1f42013-03-05 15:04:55 -05001312
Rik van Riel6062a8d2013-04-30 19:15:44 -07001313 sem_lock(sma, NULL, -1);
Al Viroe1fd1f42013-03-05 15:04:55 -05001314
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001315 if (!ipc_valid_object(&sma->sem_perm)) {
Manfred Spraul6e224f92013-10-16 13:46:45 -07001316 sem_unlock(sma, -1);
1317 rcu_read_unlock();
1318 return -EIDRM;
1319 }
1320
Al Viroe1fd1f42013-03-05 15:04:55 -05001321 curr = &sma->sem_base[semnum];
1322
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -07001323 ipc_assert_locked_object(&sma->sem_perm);
Al Viroe1fd1f42013-03-05 15:04:55 -05001324 list_for_each_entry(un, &sma->list_id, list_id)
1325 un->semadj[semnum] = 0;
1326
1327 curr->semval = val;
1328 curr->sempid = task_tgid_vnr(current);
1329 sma->sem_ctime = get_seconds();
1330 /* maybe some queued-up processes were waiting for this */
1331 do_smart_update(sma, NULL, 0, 0, &tasks);
Rik van Riel6062a8d2013-04-30 19:15:44 -07001332 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001333 rcu_read_unlock();
Al Viroe1fd1f42013-03-05 15:04:55 -05001334 wake_up_sem_queue_do(&tasks);
Rik van Riel6062a8d2013-04-30 19:15:44 -07001335 return 0;
Al Viroe1fd1f42013-03-05 15:04:55 -05001336}
1337
Kirill Korotaeve3893532006-10-02 02:18:22 -07001338static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
Al Viroe1fd1f42013-03-05 15:04:55 -05001339 int cmd, void __user *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340{
1341 struct sem_array *sma;
Manfred Spraul239521f2014-01-27 17:07:04 -08001342 struct sem *curr;
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001343 int err, nsems;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 ushort fast_sem_io[SEMMSL_FAST];
Manfred Spraul239521f2014-01-27 17:07:04 -08001345 ushort *sem_io = fast_sem_io;
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001346 struct list_head tasks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001348 INIT_LIST_HEAD(&tasks);
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001349
1350 rcu_read_lock();
1351 sma = sem_obtain_object_check(ns, semid);
1352 if (IS_ERR(sma)) {
1353 rcu_read_unlock();
1354 return PTR_ERR(sma);
1355 }
1356
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 nsems = sma->sem_nsems;
1358
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359 err = -EACCES;
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001360 if (ipcperms(ns, &sma->sem_perm, cmd == SETALL ? S_IWUGO : S_IRUGO))
1361 goto out_rcu_wakeup;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362
1363 err = security_sem_semctl(sma, cmd);
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001364 if (err)
1365 goto out_rcu_wakeup;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366
1367 err = -EACCES;
1368 switch (cmd) {
1369 case GETALL:
1370 {
Al Viroe1fd1f42013-03-05 15:04:55 -05001371 ushort __user *array = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 int i;
1373
Al Viroce857222013-05-03 00:30:49 +01001374 sem_lock(sma, NULL, -1);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001375 if (!ipc_valid_object(&sma->sem_perm)) {
Manfred Spraul6e224f92013-10-16 13:46:45 -07001376 err = -EIDRM;
1377 goto out_unlock;
1378 }
Manfred Spraul239521f2014-01-27 17:07:04 -08001379 if (nsems > SEMMSL_FAST) {
Al Viroce857222013-05-03 00:30:49 +01001380 if (!ipc_rcu_getref(sma)) {
Al Viroce857222013-05-03 00:30:49 +01001381 err = -EIDRM;
Manfred Spraul6e224f92013-10-16 13:46:45 -07001382 goto out_unlock;
Al Viroce857222013-05-03 00:30:49 +01001383 }
1384 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001385 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 sem_io = ipc_alloc(sizeof(ushort)*nsems);
Manfred Spraul239521f2014-01-27 17:07:04 -08001387 if (sem_io == NULL) {
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -07001388 ipc_rcu_putref(sma, ipc_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389 return -ENOMEM;
1390 }
1391
Linus Torvalds4091fd92013-05-04 10:13:40 -07001392 rcu_read_lock();
Pierre Peiffer6ff37972008-04-29 01:00:46 -07001393 sem_lock_and_putref(sma);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001394 if (!ipc_valid_object(&sma->sem_perm)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 err = -EIDRM;
Manfred Spraul6e224f92013-10-16 13:46:45 -07001396 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 }
Al Viroce857222013-05-03 00:30:49 +01001398 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 for (i = 0; i < sma->sem_nsems; i++)
1400 sem_io[i] = sma->sem_base[i].semval;
Rik van Riel6062a8d2013-04-30 19:15:44 -07001401 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001402 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 err = 0;
Manfred Spraul239521f2014-01-27 17:07:04 -08001404 if (copy_to_user(array, sem_io, nsems*sizeof(ushort)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405 err = -EFAULT;
1406 goto out_free;
1407 }
1408 case SETALL:
1409 {
1410 int i;
1411 struct sem_undo *un;
1412
Rik van Riel6062a8d2013-04-30 19:15:44 -07001413 if (!ipc_rcu_getref(sma)) {
Manfred Spraul6e224f92013-10-16 13:46:45 -07001414 err = -EIDRM;
1415 goto out_rcu_wakeup;
Rik van Riel6062a8d2013-04-30 19:15:44 -07001416 }
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001417 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418
Manfred Spraul239521f2014-01-27 17:07:04 -08001419 if (nsems > SEMMSL_FAST) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420 sem_io = ipc_alloc(sizeof(ushort)*nsems);
Manfred Spraul239521f2014-01-27 17:07:04 -08001421 if (sem_io == NULL) {
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -07001422 ipc_rcu_putref(sma, ipc_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 return -ENOMEM;
1424 }
1425 }
1426
Manfred Spraul239521f2014-01-27 17:07:04 -08001427 if (copy_from_user(sem_io, p, nsems*sizeof(ushort))) {
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -07001428 ipc_rcu_putref(sma, ipc_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429 err = -EFAULT;
1430 goto out_free;
1431 }
1432
1433 for (i = 0; i < nsems; i++) {
1434 if (sem_io[i] > SEMVMX) {
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -07001435 ipc_rcu_putref(sma, ipc_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 err = -ERANGE;
1437 goto out_free;
1438 }
1439 }
Linus Torvalds4091fd92013-05-04 10:13:40 -07001440 rcu_read_lock();
Pierre Peiffer6ff37972008-04-29 01:00:46 -07001441 sem_lock_and_putref(sma);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001442 if (!ipc_valid_object(&sma->sem_perm)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 err = -EIDRM;
Manfred Spraul6e224f92013-10-16 13:46:45 -07001444 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445 }
1446
1447 for (i = 0; i < nsems; i++)
1448 sma->sem_base[i].semval = sem_io[i];
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001449
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -07001450 ipc_assert_locked_object(&sma->sem_perm);
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001451 list_for_each_entry(un, &sma->list_id, list_id) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452 for (i = 0; i < nsems; i++)
1453 un->semadj[i] = 0;
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001454 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 sma->sem_ctime = get_seconds();
1456 /* maybe some queued-up processes were waiting for this */
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001457 do_smart_update(sma, NULL, 0, 0, &tasks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 err = 0;
1459 goto out_unlock;
1460 }
Al Viroe1fd1f42013-03-05 15:04:55 -05001461 /* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 }
1463 err = -EINVAL;
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001464 if (semnum < 0 || semnum >= nsems)
1465 goto out_rcu_wakeup;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466
Rik van Riel6062a8d2013-04-30 19:15:44 -07001467 sem_lock(sma, NULL, -1);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001468 if (!ipc_valid_object(&sma->sem_perm)) {
Manfred Spraul6e224f92013-10-16 13:46:45 -07001469 err = -EIDRM;
1470 goto out_unlock;
1471 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 curr = &sma->sem_base[semnum];
1473
1474 switch (cmd) {
1475 case GETVAL:
1476 err = curr->semval;
1477 goto out_unlock;
1478 case GETPID:
1479 err = curr->sempid;
1480 goto out_unlock;
1481 case GETNCNT:
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001482 err = count_semcnt(sma, semnum, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 goto out_unlock;
1484 case GETZCNT:
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001485 err = count_semcnt(sma, semnum, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487 }
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001488
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489out_unlock:
Rik van Riel6062a8d2013-04-30 19:15:44 -07001490 sem_unlock(sma, -1);
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001491out_rcu_wakeup:
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001492 rcu_read_unlock();
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001493 wake_up_sem_queue_do(&tasks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494out_free:
Manfred Spraul239521f2014-01-27 17:07:04 -08001495 if (sem_io != fast_sem_io)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496 ipc_free(sem_io, sizeof(ushort)*nsems);
1497 return err;
1498}
1499
Pierre Peiffer016d7132008-04-29 01:00:50 -07001500static inline unsigned long
1501copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502{
Manfred Spraul239521f2014-01-27 17:07:04 -08001503 switch (version) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 case IPC_64:
Pierre Peiffer016d7132008-04-29 01:00:50 -07001505 if (copy_from_user(out, buf, sizeof(*out)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 case IPC_OLD:
1509 {
1510 struct semid_ds tbuf_old;
1511
Manfred Spraul239521f2014-01-27 17:07:04 -08001512 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513 return -EFAULT;
1514
Pierre Peiffer016d7132008-04-29 01:00:50 -07001515 out->sem_perm.uid = tbuf_old.sem_perm.uid;
1516 out->sem_perm.gid = tbuf_old.sem_perm.gid;
1517 out->sem_perm.mode = tbuf_old.sem_perm.mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518
1519 return 0;
1520 }
1521 default:
1522 return -EINVAL;
1523 }
1524}
1525
Pierre Peiffer522bb2a2008-04-29 01:00:49 -07001526/*
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001527 * This function handles some semctl commands which require the rwsem
Pierre Peiffer522bb2a2008-04-29 01:00:49 -07001528 * to be held in write mode.
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001529 * NOTE: no locks must be held, the rwsem is taken inside this function.
Pierre Peiffer522bb2a2008-04-29 01:00:49 -07001530 */
Pierre Peiffer21a48262008-04-29 01:00:49 -07001531static int semctl_down(struct ipc_namespace *ns, int semid,
Al Viroe1fd1f42013-03-05 15:04:55 -05001532 int cmd, int version, void __user *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533{
1534 struct sem_array *sma;
1535 int err;
Pierre Peiffer016d7132008-04-29 01:00:50 -07001536 struct semid64_ds semid64;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537 struct kern_ipc_perm *ipcp;
1538
Manfred Spraul239521f2014-01-27 17:07:04 -08001539 if (cmd == IPC_SET) {
Al Viroe1fd1f42013-03-05 15:04:55 -05001540 if (copy_semid_from_user(&semid64, p, version))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001544 down_write(&sem_ids(ns).rwsem);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001545 rcu_read_lock();
1546
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001547 ipcp = ipcctl_pre_down_nolock(ns, &sem_ids(ns), semid, cmd,
1548 &semid64.sem_perm, 0);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001549 if (IS_ERR(ipcp)) {
1550 err = PTR_ERR(ipcp);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001551 goto out_unlock1;
1552 }
Steve Grubb073115d2006-04-02 17:07:33 -04001553
Pierre Peiffera5f75e72008-04-29 01:00:54 -07001554 sma = container_of(ipcp, struct sem_array, sem_perm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555
1556 err = security_sem_semctl(sma, cmd);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001557 if (err)
1558 goto out_unlock1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001560 switch (cmd) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 case IPC_RMID:
Rik van Riel6062a8d2013-04-30 19:15:44 -07001562 sem_lock(sma, NULL, -1);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001563 /* freeary unlocks the ipc object and rcu */
Pierre Peiffer01b8b072008-02-08 04:18:57 -08001564 freeary(ns, ipcp);
Pierre Peiffer522bb2a2008-04-29 01:00:49 -07001565 goto out_up;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 case IPC_SET:
Rik van Riel6062a8d2013-04-30 19:15:44 -07001567 sem_lock(sma, NULL, -1);
Eric W. Biederman1efdb692012-02-07 16:54:11 -08001568 err = ipc_update_perm(&semid64.sem_perm, ipcp);
1569 if (err)
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001570 goto out_unlock0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571 sma->sem_ctime = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572 break;
1573 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574 err = -EINVAL;
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001575 goto out_unlock1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001578out_unlock0:
Rik van Riel6062a8d2013-04-30 19:15:44 -07001579 sem_unlock(sma, -1);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001580out_unlock1:
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001581 rcu_read_unlock();
Pierre Peiffer522bb2a2008-04-29 01:00:49 -07001582out_up:
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001583 up_write(&sem_ids(ns).rwsem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584 return err;
1585}
1586
Al Viroe1fd1f42013-03-05 15:04:55 -05001587SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, unsigned long, arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589 int version;
Kirill Korotaeve3893532006-10-02 02:18:22 -07001590 struct ipc_namespace *ns;
Al Viroe1fd1f42013-03-05 15:04:55 -05001591 void __user *p = (void __user *)arg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592
1593 if (semid < 0)
1594 return -EINVAL;
1595
1596 version = ipc_parse_version(&cmd);
Kirill Korotaeve3893532006-10-02 02:18:22 -07001597 ns = current->nsproxy->ipc_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598
Manfred Spraul239521f2014-01-27 17:07:04 -08001599 switch (cmd) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600 case IPC_INFO:
1601 case SEM_INFO:
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -08001602 case IPC_STAT:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603 case SEM_STAT:
Al Viroe1fd1f42013-03-05 15:04:55 -05001604 return semctl_nolock(ns, semid, cmd, version, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605 case GETALL:
1606 case GETVAL:
1607 case GETPID:
1608 case GETNCNT:
1609 case GETZCNT:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610 case SETALL:
Al Viroe1fd1f42013-03-05 15:04:55 -05001611 return semctl_main(ns, semid, semnum, cmd, p);
1612 case SETVAL:
1613 return semctl_setval(ns, semid, semnum, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614 case IPC_RMID:
1615 case IPC_SET:
Al Viroe1fd1f42013-03-05 15:04:55 -05001616 return semctl_down(ns, semid, cmd, version, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617 default:
1618 return -EINVAL;
1619 }
1620}
1621
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622/* If the task doesn't already have a undo_list, then allocate one
1623 * here. We guarantee there is only one thread using this undo list,
1624 * and current is THE ONE
1625 *
1626 * If this allocation and assignment succeeds, but later
1627 * portions of this code fail, there is no need to free the sem_undo_list.
1628 * Just let it stay associated with the task, and it'll be freed later
1629 * at exit time.
1630 *
1631 * This can block, so callers must hold no locks.
1632 */
1633static inline int get_undo_list(struct sem_undo_list **undo_listp)
1634{
1635 struct sem_undo_list *undo_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636
1637 undo_list = current->sysvsem.undo_list;
1638 if (!undo_list) {
Matt Helsley2453a302006-10-02 02:18:25 -07001639 undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 if (undo_list == NULL)
1641 return -ENOMEM;
Ingo Molnar00a5dfd2005-08-05 23:05:27 +02001642 spin_lock_init(&undo_list->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643 atomic_set(&undo_list->refcnt, 1);
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001644 INIT_LIST_HEAD(&undo_list->list_proc);
1645
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646 current->sysvsem.undo_list = undo_list;
1647 }
1648 *undo_listp = undo_list;
1649 return 0;
1650}
1651
Nick Pigginbf17bb72009-12-15 16:47:28 -08001652static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653{
Nick Pigginbf17bb72009-12-15 16:47:28 -08001654 struct sem_undo *un;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655
Nick Pigginbf17bb72009-12-15 16:47:28 -08001656 list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) {
1657 if (un->semid == semid)
1658 return un;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 }
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001660 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661}
1662
Nick Pigginbf17bb72009-12-15 16:47:28 -08001663static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1664{
1665 struct sem_undo *un;
1666
Manfred Spraul239521f2014-01-27 17:07:04 -08001667 assert_spin_locked(&ulp->lock);
Nick Pigginbf17bb72009-12-15 16:47:28 -08001668
1669 un = __lookup_undo(ulp, semid);
1670 if (un) {
1671 list_del_rcu(&un->list_proc);
1672 list_add_rcu(&un->list_proc, &ulp->list_proc);
1673 }
1674 return un;
1675}
1676
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001677/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -08001678 * find_alloc_undo - lookup (and if not present create) undo array
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001679 * @ns: namespace
1680 * @semid: semaphore array id
1681 *
1682 * The function looks up (and if not present creates) the undo structure.
1683 * The size of the undo structure depends on the size of the semaphore
1684 * array, thus the alloc path is not that straightforward.
Manfred Spraul380af1b2008-07-25 01:48:06 -07001685 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1686 * performs a rcu_read_lock().
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001687 */
1688static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689{
1690 struct sem_array *sma;
1691 struct sem_undo_list *ulp;
1692 struct sem_undo *un, *new;
Rik van Riel6062a8d2013-04-30 19:15:44 -07001693 int nsems, error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694
1695 error = get_undo_list(&ulp);
1696 if (error)
1697 return ERR_PTR(error);
1698
Manfred Spraul380af1b2008-07-25 01:48:06 -07001699 rcu_read_lock();
Pierre Peifferc530c6a2007-10-18 23:40:55 -07001700 spin_lock(&ulp->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701 un = lookup_undo(ulp, semid);
Pierre Peifferc530c6a2007-10-18 23:40:55 -07001702 spin_unlock(&ulp->lock);
Manfred Spraul239521f2014-01-27 17:07:04 -08001703 if (likely(un != NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704 goto out;
1705
1706 /* no undo structure around - allocate one. */
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001707 /* step 1: figure out the size of the semaphore array */
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001708 sma = sem_obtain_object_check(ns, semid);
1709 if (IS_ERR(sma)) {
1710 rcu_read_unlock();
Julia Lawall4de85cd2010-05-26 14:43:44 -07001711 return ERR_CAST(sma);
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001712 }
Nadia Derbey023a5352007-10-18 23:40:51 -07001713
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714 nsems = sma->sem_nsems;
Rik van Riel6062a8d2013-04-30 19:15:44 -07001715 if (!ipc_rcu_getref(sma)) {
1716 rcu_read_unlock();
1717 un = ERR_PTR(-EIDRM);
1718 goto out;
1719 }
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001720 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001722 /* step 2: allocate new undo structure */
Burman Yan4668edc2006-12-06 20:38:51 -08001723 new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724 if (!new) {
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -07001725 ipc_rcu_putref(sma, ipc_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726 return ERR_PTR(-ENOMEM);
1727 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728
Manfred Spraul380af1b2008-07-25 01:48:06 -07001729 /* step 3: Acquire the lock on semaphore array */
Linus Torvalds4091fd92013-05-04 10:13:40 -07001730 rcu_read_lock();
Pierre Peiffer6ff37972008-04-29 01:00:46 -07001731 sem_lock_and_putref(sma);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001732 if (!ipc_valid_object(&sma->sem_perm)) {
Rik van Riel6062a8d2013-04-30 19:15:44 -07001733 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001734 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735 kfree(new);
1736 un = ERR_PTR(-EIDRM);
1737 goto out;
1738 }
Manfred Spraul380af1b2008-07-25 01:48:06 -07001739 spin_lock(&ulp->lock);
1740
1741 /*
1742 * step 4: check for races: did someone else allocate the undo struct?
1743 */
1744 un = lookup_undo(ulp, semid);
1745 if (un) {
1746 kfree(new);
1747 goto success;
1748 }
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001749 /* step 5: initialize & link new undo structure */
1750 new->semadj = (short *) &new[1];
Manfred Spraul380af1b2008-07-25 01:48:06 -07001751 new->ulp = ulp;
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001752 new->semid = semid;
1753 assert_spin_locked(&ulp->lock);
Manfred Spraul380af1b2008-07-25 01:48:06 -07001754 list_add_rcu(&new->list_proc, &ulp->list_proc);
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -07001755 ipc_assert_locked_object(&sma->sem_perm);
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001756 list_add(&new->list_id, &sma->list_id);
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001757 un = new;
Manfred Spraul380af1b2008-07-25 01:48:06 -07001758
1759success:
1760 spin_unlock(&ulp->lock);
Rik van Riel6062a8d2013-04-30 19:15:44 -07001761 sem_unlock(sma, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762out:
1763 return un;
1764}
1765
Manfred Spraulc61284e2010-07-20 13:24:23 -07001766
1767/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -08001768 * get_queue_result - retrieve the result code from sem_queue
Manfred Spraulc61284e2010-07-20 13:24:23 -07001769 * @q: Pointer to queue structure
1770 *
1771 * Retrieve the return code from the pending queue. If IN_WAKEUP is found in
1772 * q->status, then we must loop until the value is replaced with the final
1773 * value: This may happen if a task is woken up by an unrelated event (e.g.
1774 * signal) and in parallel the task is woken up by another task because it got
1775 * the requested semaphores.
1776 *
1777 * The function can be called with or without holding the semaphore spinlock.
1778 */
1779static int get_queue_result(struct sem_queue *q)
1780{
1781 int error;
1782
1783 error = q->status;
1784 while (unlikely(error == IN_WAKEUP)) {
1785 cpu_relax();
1786 error = q->status;
1787 }
1788
1789 return error;
1790}
1791
Heiko Carstensd5460c92009-01-14 14:14:27 +01001792SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
1793 unsigned, nsops, const struct timespec __user *, timeout)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794{
1795 int error = -EINVAL;
1796 struct sem_array *sma;
1797 struct sembuf fast_sops[SEMOPM_FAST];
Manfred Spraul239521f2014-01-27 17:07:04 -08001798 struct sembuf *sops = fast_sops, *sop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799 struct sem_undo *un;
Rik van Riel6062a8d2013-04-30 19:15:44 -07001800 int undos = 0, alter = 0, max, locknum;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801 struct sem_queue queue;
1802 unsigned long jiffies_left = 0;
Kirill Korotaeve3893532006-10-02 02:18:22 -07001803 struct ipc_namespace *ns;
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001804 struct list_head tasks;
Kirill Korotaeve3893532006-10-02 02:18:22 -07001805
1806 ns = current->nsproxy->ipc_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807
1808 if (nsops < 1 || semid < 0)
1809 return -EINVAL;
Kirill Korotaeve3893532006-10-02 02:18:22 -07001810 if (nsops > ns->sc_semopm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811 return -E2BIG;
Manfred Spraul239521f2014-01-27 17:07:04 -08001812 if (nsops > SEMOPM_FAST) {
1813 sops = kmalloc(sizeof(*sops)*nsops, GFP_KERNEL);
1814 if (sops == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815 return -ENOMEM;
1816 }
Manfred Spraul239521f2014-01-27 17:07:04 -08001817 if (copy_from_user(sops, tsops, nsops * sizeof(*tsops))) {
1818 error = -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819 goto out_free;
1820 }
1821 if (timeout) {
1822 struct timespec _timeout;
1823 if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1824 error = -EFAULT;
1825 goto out_free;
1826 }
1827 if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1828 _timeout.tv_nsec >= 1000000000L) {
1829 error = -EINVAL;
1830 goto out_free;
1831 }
1832 jiffies_left = timespec_to_jiffies(&_timeout);
1833 }
1834 max = 0;
1835 for (sop = sops; sop < sops + nsops; sop++) {
1836 if (sop->sem_num >= max)
1837 max = sop->sem_num;
1838 if (sop->sem_flg & SEM_UNDO)
Manfred Spraulb78755a2005-06-23 00:10:06 -07001839 undos = 1;
1840 if (sop->sem_op != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841 alter = 1;
1842 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843
Rik van Riel6062a8d2013-04-30 19:15:44 -07001844 INIT_LIST_HEAD(&tasks);
1845
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846 if (undos) {
Rik van Riel6062a8d2013-04-30 19:15:44 -07001847 /* On success, find_alloc_undo takes the rcu_read_lock */
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001848 un = find_alloc_undo(ns, semid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849 if (IS_ERR(un)) {
1850 error = PTR_ERR(un);
1851 goto out_free;
1852 }
Rik van Riel6062a8d2013-04-30 19:15:44 -07001853 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854 un = NULL;
Rik van Riel6062a8d2013-04-30 19:15:44 -07001855 rcu_read_lock();
1856 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001858 sma = sem_obtain_object_check(ns, semid);
Nadia Derbey023a5352007-10-18 23:40:51 -07001859 if (IS_ERR(sma)) {
Rik van Riel6062a8d2013-04-30 19:15:44 -07001860 rcu_read_unlock();
Nadia Derbey023a5352007-10-18 23:40:51 -07001861 error = PTR_ERR(sma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862 goto out_free;
Nadia Derbey023a5352007-10-18 23:40:51 -07001863 }
1864
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001865 error = -EFBIG;
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001866 if (max >= sma->sem_nsems)
1867 goto out_rcu_wakeup;
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001868
1869 error = -EACCES;
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001870 if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO))
1871 goto out_rcu_wakeup;
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001872
1873 error = security_sem_semop(sma, sops, nsops, alter);
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001874 if (error)
1875 goto out_rcu_wakeup;
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001876
Manfred Spraul6e224f92013-10-16 13:46:45 -07001877 error = -EIDRM;
1878 locknum = sem_lock(sma, sops, nsops);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001879 /*
1880 * We eventually might perform the following check in a lockless
1881 * fashion, considering ipc_valid_object() locking constraints.
1882 * If nsops == 1 and there is no contention for sem_perm.lock, then
1883 * only a per-semaphore lock is held and it's OK to proceed with the
1884 * check below. More details on the fine grained locking scheme
1885 * entangled here and why it's RMID race safe on comments at sem_lock()
1886 */
1887 if (!ipc_valid_object(&sma->sem_perm))
Manfred Spraul6e224f92013-10-16 13:46:45 -07001888 goto out_unlock_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889 /*
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001890 * semid identifiers are not unique - find_alloc_undo may have
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891 * allocated an undo structure, it was invalidated by an RMID
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001892 * and now a new array with received the same id. Check and fail.
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001893 * This case can be detected checking un->semid. The existence of
Manfred Spraul380af1b2008-07-25 01:48:06 -07001894 * "un" itself is guaranteed by rcu.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895 */
Rik van Riel6062a8d2013-04-30 19:15:44 -07001896 if (un && un->semid == -1)
1897 goto out_unlock_free;
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001898
Manfred Sprauld198cd62014-06-06 14:37:49 -07001899 queue.sops = sops;
1900 queue.nsops = nsops;
1901 queue.undo = un;
1902 queue.pid = task_tgid_vnr(current);
1903 queue.alter = alter;
1904
1905 error = perform_atomic_semop(sma, &queue);
Manfred Spraul0e8c6652013-09-30 13:45:25 -07001906 if (error == 0) {
1907 /* If the operation was successful, then do
1908 * the required updates.
1909 */
1910 if (alter)
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001911 do_smart_update(sma, sops, nsops, 1, &tasks);
Manfred Spraul0e8c6652013-09-30 13:45:25 -07001912 else
1913 set_semotime(sma, sops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914 }
Manfred Spraul0e8c6652013-09-30 13:45:25 -07001915 if (error <= 0)
1916 goto out_unlock_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917
1918 /* We need to sleep on this operation, so we put the current
1919 * task into the pending queue and go to sleep.
1920 */
Paul McQuade46c0a8c2014-06-06 14:37:37 -07001921
Manfred Spraulb97e8202009-12-15 16:47:32 -08001922 if (nsops == 1) {
1923 struct sem *curr;
1924 curr = &sma->sem_base[sops->sem_num];
1925
Manfred Spraulf269f402013-07-08 16:01:24 -07001926 if (alter) {
1927 if (sma->complex_count) {
1928 list_add_tail(&queue.list,
1929 &sma->pending_alter);
1930 } else {
1931
1932 list_add_tail(&queue.list,
1933 &curr->pending_alter);
1934 }
1935 } else {
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001936 list_add_tail(&queue.list, &curr->pending_const);
Manfred Spraulf269f402013-07-08 16:01:24 -07001937 }
Manfred Spraulb97e8202009-12-15 16:47:32 -08001938 } else {
Manfred Spraulf269f402013-07-08 16:01:24 -07001939 if (!sma->complex_count)
1940 merge_queues(sma);
1941
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001942 if (alter)
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001943 list_add_tail(&queue.list, &sma->pending_alter);
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001944 else
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001945 list_add_tail(&queue.list, &sma->pending_const);
1946
Manfred Spraulb97e8202009-12-15 16:47:32 -08001947 sma->complex_count++;
1948 }
1949
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950 queue.status = -EINTR;
1951 queue.sleeper = current;
Manfred Spraul0b0577f2011-11-02 13:38:52 -07001952
1953sleep_again:
Davidlohr Bueso52644c92015-02-17 13:47:55 -08001954 __set_current_state(TASK_INTERRUPTIBLE);
Rik van Riel6062a8d2013-04-30 19:15:44 -07001955 sem_unlock(sma, locknum);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001956 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957
1958 if (timeout)
1959 jiffies_left = schedule_timeout(jiffies_left);
1960 else
1961 schedule();
1962
Manfred Spraulc61284e2010-07-20 13:24:23 -07001963 error = get_queue_result(&queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964
1965 if (error != -EINTR) {
1966 /* fast path: update_queue already obtained all requested
Manfred Spraulc61284e2010-07-20 13:24:23 -07001967 * resources.
1968 * Perform a smp_mb(): User space could assume that semop()
1969 * is a memory barrier: Without the mb(), the cpu could
1970 * speculatively read in user space stale data that was
1971 * overwritten by the previous owner of the semaphore.
1972 */
1973 smp_mb();
1974
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975 goto out_free;
1976 }
1977
Linus Torvalds321310c2013-05-04 10:47:57 -07001978 rcu_read_lock();
Rik van Riel6062a8d2013-04-30 19:15:44 -07001979 sma = sem_obtain_lock(ns, semid, sops, nsops, &locknum);
Manfred Sprauld694ad62011-07-25 17:11:47 -07001980
1981 /*
1982 * Wait until it's guaranteed that no wakeup_sem_queue_do() is ongoing.
1983 */
1984 error = get_queue_result(&queue);
1985
1986 /*
1987 * Array removed? If yes, leave without sem_unlock().
1988 */
Nadia Derbey023a5352007-10-18 23:40:51 -07001989 if (IS_ERR(sma)) {
Linus Torvalds321310c2013-05-04 10:47:57 -07001990 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991 goto out_free;
1992 }
1993
Manfred Spraulc61284e2010-07-20 13:24:23 -07001994
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995 /*
Manfred Sprauld694ad62011-07-25 17:11:47 -07001996 * If queue.status != -EINTR we are woken up by another process.
1997 * Leave without unlink_queue(), but with sem_unlock().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998 */
Davidlohr Bueso3ab08fe2014-01-27 17:07:06 -08001999 if (error != -EINTR)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000 goto out_unlock_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001
2002 /*
2003 * If an interrupt occurred we have to clean up the queue
2004 */
2005 if (timeout && jiffies_left == 0)
2006 error = -EAGAIN;
Manfred Spraul0b0577f2011-11-02 13:38:52 -07002007
2008 /*
2009 * If the wakeup was spurious, just retry
2010 */
2011 if (error == -EINTR && !signal_pending(current))
2012 goto sleep_again;
2013
Manfred Spraulb97e8202009-12-15 16:47:32 -08002014 unlink_queue(sma, &queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015
2016out_unlock_free:
Rik van Riel6062a8d2013-04-30 19:15:44 -07002017 sem_unlock(sma, locknum);
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07002018out_rcu_wakeup:
Linus Torvalds6d49dab2013-05-03 15:04:40 -07002019 rcu_read_unlock();
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07002020 wake_up_sem_queue_do(&tasks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021out_free:
Manfred Spraul239521f2014-01-27 17:07:04 -08002022 if (sops != fast_sops)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023 kfree(sops);
2024 return error;
2025}
2026
Heiko Carstensd5460c92009-01-14 14:14:27 +01002027SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
2028 unsigned, nsops)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029{
2030 return sys_semtimedop(semid, tsops, nsops, NULL);
2031}
2032
2033/* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
2034 * parent and child tasks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035 */
2036
2037int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
2038{
2039 struct sem_undo_list *undo_list;
2040 int error;
2041
2042 if (clone_flags & CLONE_SYSVSEM) {
2043 error = get_undo_list(&undo_list);
2044 if (error)
2045 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046 atomic_inc(&undo_list->refcnt);
2047 tsk->sysvsem.undo_list = undo_list;
Paul McQuade46c0a8c2014-06-06 14:37:37 -07002048 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049 tsk->sysvsem.undo_list = NULL;
2050
2051 return 0;
2052}
2053
2054/*
2055 * add semadj values to semaphores, free undo structures.
2056 * undo structures are not freed when semaphore arrays are destroyed
2057 * so some of them may be out of date.
2058 * IMPLEMENTATION NOTE: There is some confusion over whether the
2059 * set of adjustments that needs to be done should be done in an atomic
2060 * manner or not. That is, if we are attempting to decrement the semval
2061 * should we queue up and wait until we can do so legally?
2062 * The original implementation attempted to do this (queue and wait).
2063 * The current implementation does not do so. The POSIX standard
2064 * and SVID should be consulted to determine what behavior is mandated.
2065 */
2066void exit_sem(struct task_struct *tsk)
2067{
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002068 struct sem_undo_list *ulp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002070 ulp = tsk->sysvsem.undo_list;
2071 if (!ulp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072 return;
Manfred Spraul9edff4a2008-04-29 01:00:57 -07002073 tsk->sysvsem.undo_list = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002075 if (!atomic_dec_and_test(&ulp->refcnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076 return;
2077
Manfred Spraul380af1b2008-07-25 01:48:06 -07002078 for (;;) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002079 struct sem_array *sma;
Manfred Spraul380af1b2008-07-25 01:48:06 -07002080 struct sem_undo *un;
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07002081 struct list_head tasks;
Rik van Riel6062a8d2013-04-30 19:15:44 -07002082 int semid, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083
Manfred Spraul380af1b2008-07-25 01:48:06 -07002084 rcu_read_lock();
Jiri Pirko05725f72009-04-14 20:17:16 +02002085 un = list_entry_rcu(ulp->list_proc.next,
2086 struct sem_undo, list_proc);
Herton R. Krzesinski602b8592015-08-14 15:35:02 -07002087 if (&un->list_proc == &ulp->list_proc) {
2088 /*
2089 * We must wait for freeary() before freeing this ulp,
2090 * in case we raced with last sem_undo. There is a small
2091 * possibility where we exit while freeary() didn't
2092 * finish unlocking sem_undo_list.
2093 */
2094 spin_unlock_wait(&ulp->lock);
Rik van Riel6062a8d2013-04-30 19:15:44 -07002095 rcu_read_unlock();
Manfred Spraul380af1b2008-07-25 01:48:06 -07002096 break;
Rik van Riel6062a8d2013-04-30 19:15:44 -07002097 }
Herton R. Krzesinski602b8592015-08-14 15:35:02 -07002098 spin_lock(&ulp->lock);
2099 semid = un->semid;
2100 spin_unlock(&ulp->lock);
Manfred Spraul380af1b2008-07-25 01:48:06 -07002101
Herton R. Krzesinski602b8592015-08-14 15:35:02 -07002102 /* exit_sem raced with IPC_RMID, nothing to do */
2103 if (semid == -1) {
2104 rcu_read_unlock();
2105 continue;
2106 }
2107
2108 sma = sem_obtain_object_check(tsk->nsproxy->ipc_ns, semid);
Manfred Spraul380af1b2008-07-25 01:48:06 -07002109 /* exit_sem raced with IPC_RMID, nothing to do */
Rik van Riel6062a8d2013-04-30 19:15:44 -07002110 if (IS_ERR(sma)) {
2111 rcu_read_unlock();
Manfred Spraul380af1b2008-07-25 01:48:06 -07002112 continue;
Rik van Riel6062a8d2013-04-30 19:15:44 -07002113 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114
Rik van Riel6062a8d2013-04-30 19:15:44 -07002115 sem_lock(sma, NULL, -1);
Manfred Spraul6e224f92013-10-16 13:46:45 -07002116 /* exit_sem raced with IPC_RMID, nothing to do */
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08002117 if (!ipc_valid_object(&sma->sem_perm)) {
Manfred Spraul6e224f92013-10-16 13:46:45 -07002118 sem_unlock(sma, -1);
2119 rcu_read_unlock();
2120 continue;
2121 }
Nick Pigginbf17bb72009-12-15 16:47:28 -08002122 un = __lookup_undo(ulp, semid);
Manfred Spraul380af1b2008-07-25 01:48:06 -07002123 if (un == NULL) {
2124 /* exit_sem raced with IPC_RMID+semget() that created
2125 * exactly the same semid. Nothing to do.
2126 */
Rik van Riel6062a8d2013-04-30 19:15:44 -07002127 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07002128 rcu_read_unlock();
Manfred Spraul380af1b2008-07-25 01:48:06 -07002129 continue;
2130 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131
Manfred Spraul380af1b2008-07-25 01:48:06 -07002132 /* remove un from the linked lists */
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -07002133 ipc_assert_locked_object(&sma->sem_perm);
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002134 list_del(&un->list_id);
2135
Herton R. Krzesinskia9795582015-08-14 15:35:05 -07002136 /* we are the last process using this ulp, acquiring ulp->lock
2137 * isn't required. Besides that, we are also protected against
2138 * IPC_RMID as we hold sma->sem_perm lock now
2139 */
Manfred Spraul380af1b2008-07-25 01:48:06 -07002140 list_del_rcu(&un->list_proc);
Manfred Spraul380af1b2008-07-25 01:48:06 -07002141
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002142 /* perform adjustments registered in un */
2143 for (i = 0; i < sma->sem_nsems; i++) {
Manfred Spraul239521f2014-01-27 17:07:04 -08002144 struct sem *semaphore = &sma->sem_base[i];
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002145 if (un->semadj[i]) {
2146 semaphore->semval += un->semadj[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147 /*
2148 * Range checks of the new semaphore value,
2149 * not defined by sus:
2150 * - Some unices ignore the undo entirely
2151 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
2152 * - some cap the value (e.g. FreeBSD caps
2153 * at 0, but doesn't enforce SEMVMX)
2154 *
2155 * Linux caps the semaphore value, both at 0
2156 * and at SEMVMX.
2157 *
Manfred Spraul239521f2014-01-27 17:07:04 -08002158 * Manfred <manfred@colorfullife.com>
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159 */
Ingo Molnar5f921ae2006-03-26 01:37:17 -08002160 if (semaphore->semval < 0)
2161 semaphore->semval = 0;
2162 if (semaphore->semval > SEMVMX)
2163 semaphore->semval = SEMVMX;
Pavel Emelyanovb4888932007-10-18 23:40:14 -07002164 semaphore->sempid = task_tgid_vnr(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165 }
2166 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002167 /* maybe some queued-up processes were waiting for this */
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07002168 INIT_LIST_HEAD(&tasks);
2169 do_smart_update(sma, NULL, 0, 1, &tasks);
Rik van Riel6062a8d2013-04-30 19:15:44 -07002170 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07002171 rcu_read_unlock();
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07002172 wake_up_sem_queue_do(&tasks);
Manfred Spraul380af1b2008-07-25 01:48:06 -07002173
Lai Jiangshan693a8b62011-03-18 12:09:35 +08002174 kfree_rcu(un, rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175 }
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002176 kfree(ulp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177}
2178
2179#ifdef CONFIG_PROC_FS
Mike Waychison19b49462005-09-06 15:17:10 -07002180static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181{
Eric W. Biederman1efdb692012-02-07 16:54:11 -08002182 struct user_namespace *user_ns = seq_user_ns(s);
Mike Waychison19b49462005-09-06 15:17:10 -07002183 struct sem_array *sma = it;
Manfred Sprauld12e1e52013-07-08 16:01:25 -07002184 time_t sem_otime;
2185
Manfred Sprauld8c63372013-09-30 13:45:07 -07002186 /*
2187 * The proc interface isn't aware of sem_lock(), it calls
2188 * ipc_lock_object() directly (in sysvipc_find_ipc).
2189 * In order to stay compatible with sem_lock(), we must wait until
2190 * all simple semop() calls have left their critical regions.
2191 */
2192 sem_wait_array(sma);
2193
Manfred Sprauld12e1e52013-07-08 16:01:25 -07002194 sem_otime = get_semotime(sma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195
Joe Perches7f032d62015-04-15 16:17:54 -07002196 seq_printf(s,
2197 "%10d %10d %4o %10u %5u %5u %5u %5u %10lu %10lu\n",
2198 sma->sem_perm.key,
2199 sma->sem_perm.id,
2200 sma->sem_perm.mode,
2201 sma->sem_nsems,
2202 from_kuid_munged(user_ns, sma->sem_perm.uid),
2203 from_kgid_munged(user_ns, sma->sem_perm.gid),
2204 from_kuid_munged(user_ns, sma->sem_perm.cuid),
2205 from_kgid_munged(user_ns, sma->sem_perm.cgid),
2206 sem_otime,
2207 sma->sem_ctime);
2208
2209 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002210}
2211#endif