blob: b3757ea0694be8a993d7e60d6c2fe90c9dbe90dc [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 */
Davidlohr Buesoa5f4db82016-03-22 14:27:48 -070095 /*
96 * PID of the process that last modified the semaphore. For
97 * Linux, specifically these are:
98 * - semop
99 * - semctl, via SETVAL and SETALL.
100 * - at task exit when performing undo adjustments (see exit_sem).
101 */
102 int sempid;
Rik van Riel6062a8d2013-04-30 19:15:44 -0700103 spinlock_t lock; /* spinlock for fine-grained semtimedop */
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700104 struct list_head pending_alter; /* pending single-sop operations */
105 /* that alter the semaphore */
106 struct list_head pending_const; /* pending single-sop operations */
107 /* that do not alter the semaphore*/
Manfred Sprauld12e1e52013-07-08 16:01:25 -0700108 time_t sem_otime; /* candidate for sem_otime */
Manfred Spraulf5c936c2013-07-08 16:01:22 -0700109} ____cacheline_aligned_in_smp;
Manfred Spraule57940d2011-11-02 13:38:54 -0700110
111/* One queue for each sleeping process in the system. */
112struct sem_queue {
Manfred Spraule57940d2011-11-02 13:38:54 -0700113 struct list_head list; /* queue of pending operations */
114 struct task_struct *sleeper; /* this process */
115 struct sem_undo *undo; /* undo structure */
116 int pid; /* process id of requesting process */
117 int status; /* completion status of operation */
118 struct sembuf *sops; /* array of pending operations */
Manfred Sprauled247b72014-06-06 14:37:49 -0700119 struct sembuf *blocking; /* the operation that blocked */
Manfred Spraule57940d2011-11-02 13:38:54 -0700120 int nsops; /* number of operations */
121 int alter; /* does *sops alter the array? */
122};
123
124/* Each task has a list of undo requests. They are executed automatically
125 * when the process exits.
126 */
127struct sem_undo {
128 struct list_head list_proc; /* per-process list: *
129 * all undos from one process
130 * rcu protected */
131 struct rcu_head rcu; /* rcu struct for sem_undo */
132 struct sem_undo_list *ulp; /* back ptr to sem_undo_list */
133 struct list_head list_id; /* per semaphore array list:
134 * all undos for one array */
135 int semid; /* semaphore set identifier */
136 short *semadj; /* array of adjustments */
137 /* one per semaphore */
138};
139
140/* sem_undo_list controls shared access to the list of sem_undo structures
141 * that may be shared among all a CLONE_SYSVSEM task group.
142 */
143struct sem_undo_list {
144 atomic_t refcnt;
145 spinlock_t lock;
146 struct list_head list_proc;
147};
148
149
Pierre Peiffered2ddbf2008-02-08 04:18:57 -0800150#define sem_ids(ns) ((ns)->ids[IPC_SEM_IDS])
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151
Nadia Derbey1b531f22007-10-18 23:40:55 -0700152#define sem_checkid(sma, semid) ipc_checkid(&sma->sem_perm, semid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700154static int newary(struct ipc_namespace *, struct ipc_params *);
Pierre Peiffer01b8b072008-02-08 04:18:57 -0800155static void freeary(struct ipc_namespace *, struct kern_ipc_perm *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156#ifdef CONFIG_PROC_FS
Mike Waychison19b49462005-09-06 15:17:10 -0700157static int sysvipc_sem_proc_show(struct seq_file *s, void *it);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158#endif
159
160#define SEMMSL_FAST 256 /* 512 bytes on stack */
161#define SEMOPM_FAST 64 /* ~ 372 bytes on stack */
162
163/*
Manfred Spraul758a6ba32013-07-08 16:01:26 -0700164 * Locking:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165 * sem_undo.id_next,
Manfred Spraul758a6ba32013-07-08 16:01:26 -0700166 * sem_array.complex_count,
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700167 * sem_array.pending{_alter,_cont},
Manfred Spraul758a6ba32013-07-08 16:01:26 -0700168 * sem_array.sem_undo: global sem_lock() for read/write
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169 * sem_undo.proc_next: only "current" is allowed to read/write that field.
Paul McQuade46c0a8c2014-06-06 14:37:37 -0700170 *
Manfred Spraul758a6ba32013-07-08 16:01:26 -0700171 * sem_array.sem_base[i].pending_{const,alter}:
172 * global or semaphore sem_lock() for read/write
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 */
174
Kirill Korotaeve3893532006-10-02 02:18:22 -0700175#define sc_semmsl sem_ctls[0]
176#define sc_semmns sem_ctls[1]
177#define sc_semopm sem_ctls[2]
178#define sc_semmni sem_ctls[3]
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179
Pierre Peiffered2ddbf2008-02-08 04:18:57 -0800180void sem_init_ns(struct ipc_namespace *ns)
Kirill Korotaeve3893532006-10-02 02:18:22 -0700181{
Kirill Korotaeve3893532006-10-02 02:18:22 -0700182 ns->sc_semmsl = SEMMSL;
183 ns->sc_semmns = SEMMNS;
184 ns->sc_semopm = SEMOPM;
185 ns->sc_semmni = SEMMNI;
186 ns->used_sems = 0;
Pierre Peiffered2ddbf2008-02-08 04:18:57 -0800187 ipc_init_ids(&ns->ids[IPC_SEM_IDS]);
Kirill Korotaeve3893532006-10-02 02:18:22 -0700188}
189
Pavel Emelyanovae5e1b22008-02-08 04:18:22 -0800190#ifdef CONFIG_IPC_NS
Kirill Korotaeve3893532006-10-02 02:18:22 -0700191void sem_exit_ns(struct ipc_namespace *ns)
192{
Pierre Peiffer01b8b072008-02-08 04:18:57 -0800193 free_ipcs(ns, &sem_ids(ns), freeary);
Serge E. Hallyn7d6feeb2009-12-15 16:47:27 -0800194 idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr);
Kirill Korotaeve3893532006-10-02 02:18:22 -0700195}
Pavel Emelyanovae5e1b22008-02-08 04:18:22 -0800196#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197
Manfred Spraul239521f2014-01-27 17:07:04 -0800198void __init sem_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199{
Pierre Peiffered2ddbf2008-02-08 04:18:57 -0800200 sem_init_ns(&init_ipc_ns);
Mike Waychison19b49462005-09-06 15:17:10 -0700201 ipc_init_proc_interface("sysvipc/sem",
202 " key semid perms nsems uid gid cuid cgid otime ctime\n",
Kirill Korotaeve3893532006-10-02 02:18:22 -0700203 IPC_SEM_IDS, sysvipc_sem_proc_show);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204}
205
Manfred Spraulf269f402013-07-08 16:01:24 -0700206/**
207 * unmerge_queues - unmerge queues, if possible.
208 * @sma: semaphore array
209 *
210 * The function unmerges the wait queues if complex_count is 0.
211 * It must be called prior to dropping the global semaphore array lock.
212 */
213static void unmerge_queues(struct sem_array *sma)
214{
215 struct sem_queue *q, *tq;
216
217 /* complex operations still around? */
218 if (sma->complex_count)
219 return;
220 /*
221 * We will switch back to simple mode.
222 * Move all pending operation back into the per-semaphore
223 * queues.
224 */
225 list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
226 struct sem *curr;
227 curr = &sma->sem_base[q->sops[0].sem_num];
228
229 list_add_tail(&q->list, &curr->pending_alter);
230 }
231 INIT_LIST_HEAD(&sma->pending_alter);
232}
233
234/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800235 * merge_queues - merge single semop queues into global queue
Manfred Spraulf269f402013-07-08 16:01:24 -0700236 * @sma: semaphore array
237 *
238 * This function merges all per-semaphore queues into the global queue.
239 * It is necessary to achieve FIFO ordering for the pending single-sop
240 * operations when a multi-semop operation must sleep.
241 * Only the alter operations must be moved, the const operations can stay.
242 */
243static void merge_queues(struct sem_array *sma)
244{
245 int i;
246 for (i = 0; i < sma->sem_nsems; i++) {
247 struct sem *sem = sma->sem_base + i;
248
249 list_splice_init(&sem->pending_alter, &sma->pending_alter);
250 }
251}
252
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -0700253static void sem_rcu_free(struct rcu_head *head)
254{
255 struct ipc_rcu *p = container_of(head, struct ipc_rcu, rcu);
256 struct sem_array *sma = ipc_rcu_to_struct(p);
257
258 security_sem_free(sma);
259 ipc_rcu_free(head);
260}
261
Nadia Derbey3e148c72007-10-18 23:40:54 -0700262/*
Manfred Spraul3ed1f8a2015-08-14 15:35:10 -0700263 * spin_unlock_wait() and !spin_is_locked() are not memory barriers, they
264 * are only control barriers.
265 * The code must pair with spin_unlock(&sem->lock) or
266 * spin_unlock(&sem_perm.lock), thus just the control barrier is insufficient.
267 *
268 * smp_rmb() is sufficient, as writes cannot pass the control barrier.
269 */
270#define ipc_smp_acquire__after_spin_is_unlocked() smp_rmb()
271
272/*
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700273 * Wait until all currently ongoing simple ops have completed.
274 * Caller must own sem_perm.lock.
275 * New simple ops cannot start, because simple ops first check
276 * that sem_perm.lock is free.
Manfred Spraul6d07b682013-09-30 13:45:06 -0700277 * that a) sem_perm.lock is free and b) complex_count is 0.
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700278 */
279static void sem_wait_array(struct sem_array *sma)
280{
281 int i;
282 struct sem *sem;
283
Manfred Spraul6d07b682013-09-30 13:45:06 -0700284 if (sma->complex_count) {
285 /* The thread that increased sma->complex_count waited on
286 * all sem->lock locks. Thus we don't need to wait again.
287 */
288 return;
289 }
290
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700291 for (i = 0; i < sma->sem_nsems; i++) {
292 sem = sma->sem_base + i;
293 spin_unlock_wait(&sem->lock);
294 }
Manfred Spraul3ed1f8a2015-08-14 15:35:10 -0700295 ipc_smp_acquire__after_spin_is_unlocked();
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700296}
297
298/*
Rik van Riel6062a8d2013-04-30 19:15:44 -0700299 * If the request contains only one semaphore operation, and there are
300 * no complex transactions pending, lock only the semaphore involved.
301 * Otherwise, lock the entire semaphore array, since we either have
302 * multiple semaphores in our own semops, or we need to look at
303 * semaphores from other pending complex operations.
Rik van Riel6062a8d2013-04-30 19:15:44 -0700304 */
305static inline int sem_lock(struct sem_array *sma, struct sembuf *sops,
306 int nsops)
307{
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700308 struct sem *sem;
Rik van Riel6062a8d2013-04-30 19:15:44 -0700309
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700310 if (nsops != 1) {
311 /* Complex operation - acquire a full lock */
312 ipc_lock_object(&sma->sem_perm);
313
314 /* And wait until all simple ops that are processed
315 * right now have dropped their locks.
316 */
317 sem_wait_array(sma);
318 return -1;
319 }
320
321 /*
322 * Only one semaphore affected - try to optimize locking.
323 * The rules are:
324 * - optimized locking is possible if no complex operation
325 * is either enqueued or processed right now.
326 * - The test for enqueued complex ops is simple:
327 * sma->complex_count != 0
328 * - Testing for complex ops that are processed right now is
329 * a bit more difficult. Complex ops acquire the full lock
330 * and first wait that the running simple ops have completed.
331 * (see above)
332 * Thus: If we own a simple lock and the global lock is free
333 * and complex_count is now 0, then it will stay 0 and
334 * thus just locking sem->lock is sufficient.
335 */
336 sem = sma->sem_base + sops->sem_num;
337
338 if (sma->complex_count == 0) {
339 /*
340 * It appears that no complex operation is around.
341 * Acquire the per-semaphore lock.
342 */
Rik van Riel6062a8d2013-04-30 19:15:44 -0700343 spin_lock(&sem->lock);
344
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700345 /* Then check that the global lock is free */
346 if (!spin_is_locked(&sma->sem_perm.lock)) {
Manfred Spraul2e094ab2014-12-12 16:58:11 -0800347 /*
Manfred Spraul3ed1f8a2015-08-14 15:35:10 -0700348 * We need a memory barrier with acquire semantics,
349 * otherwise we can race with another thread that does:
Manfred Spraul2e094ab2014-12-12 16:58:11 -0800350 * complex_count++;
351 * spin_unlock(sem_perm.lock);
352 */
Manfred Spraul3ed1f8a2015-08-14 15:35:10 -0700353 ipc_smp_acquire__after_spin_is_unlocked();
Rik van Riel6062a8d2013-04-30 19:15:44 -0700354
Manfred Spraul2e094ab2014-12-12 16:58:11 -0800355 /*
356 * Now repeat the test of complex_count:
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700357 * It can't change anymore until we drop sem->lock.
358 * Thus: if is now 0, then it will stay 0.
359 */
360 if (sma->complex_count == 0) {
361 /* fast path successful! */
362 return sops->sem_num;
363 }
Rik van Riel6062a8d2013-04-30 19:15:44 -0700364 }
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700365 spin_unlock(&sem->lock);
Rik van Riel6062a8d2013-04-30 19:15:44 -0700366 }
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700367
368 /* slow path: acquire the full lock */
369 ipc_lock_object(&sma->sem_perm);
370
371 if (sma->complex_count == 0) {
372 /* False alarm:
373 * There is no complex operation, thus we can switch
374 * back to the fast path.
375 */
376 spin_lock(&sem->lock);
377 ipc_unlock_object(&sma->sem_perm);
378 return sops->sem_num;
379 } else {
380 /* Not a false alarm, thus complete the sequence for a
381 * full lock.
382 */
383 sem_wait_array(sma);
384 return -1;
385 }
Rik van Riel6062a8d2013-04-30 19:15:44 -0700386}
387
388static inline void sem_unlock(struct sem_array *sma, int locknum)
389{
390 if (locknum == -1) {
Manfred Spraulf269f402013-07-08 16:01:24 -0700391 unmerge_queues(sma);
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -0700392 ipc_unlock_object(&sma->sem_perm);
Rik van Riel6062a8d2013-04-30 19:15:44 -0700393 } else {
394 struct sem *sem = sma->sem_base + locknum;
395 spin_unlock(&sem->lock);
396 }
Rik van Riel6062a8d2013-04-30 19:15:44 -0700397}
398
399/*
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700400 * sem_lock_(check_) routines are called in the paths where the rwsem
Nadia Derbey3e148c72007-10-18 23:40:54 -0700401 * is not held.
Linus Torvalds321310c2013-05-04 10:47:57 -0700402 *
403 * The caller holds the RCU read lock.
Nadia Derbey3e148c72007-10-18 23:40:54 -0700404 */
Rik van Riel6062a8d2013-04-30 19:15:44 -0700405static inline struct sem_array *sem_obtain_lock(struct ipc_namespace *ns,
406 int id, struct sembuf *sops, int nsops, int *locknum)
Nadia Derbey023a5352007-10-18 23:40:51 -0700407{
Rik van Rielc460b662013-04-30 19:15:35 -0700408 struct kern_ipc_perm *ipcp;
409 struct sem_array *sma;
Nadia Derbey03f02c72007-10-18 23:40:51 -0700410
Davidlohr Bueso55b7ae52015-06-30 14:58:42 -0700411 ipcp = ipc_obtain_object_idr(&sem_ids(ns), id);
Linus Torvalds321310c2013-05-04 10:47:57 -0700412 if (IS_ERR(ipcp))
413 return ERR_CAST(ipcp);
Pierre Peifferb1ed88b2008-02-06 01:36:23 -0800414
Rik van Riel6062a8d2013-04-30 19:15:44 -0700415 sma = container_of(ipcp, struct sem_array, sem_perm);
416 *locknum = sem_lock(sma, sops, nsops);
Rik van Rielc460b662013-04-30 19:15:35 -0700417
418 /* ipc_rmid() may have already freed the ID while sem_lock
419 * was spinning: verify that the structure is still valid
420 */
Rafael Aquini72a8ff22014-01-27 17:07:02 -0800421 if (ipc_valid_object(ipcp))
Rik van Rielc460b662013-04-30 19:15:35 -0700422 return container_of(ipcp, struct sem_array, sem_perm);
423
Rik van Riel6062a8d2013-04-30 19:15:44 -0700424 sem_unlock(sma, *locknum);
Linus Torvalds321310c2013-05-04 10:47:57 -0700425 return ERR_PTR(-EINVAL);
Nadia Derbey023a5352007-10-18 23:40:51 -0700426}
427
Davidlohr Bueso16df3672013-04-30 19:15:29 -0700428static inline struct sem_array *sem_obtain_object(struct ipc_namespace *ns, int id)
429{
Davidlohr Bueso55b7ae52015-06-30 14:58:42 -0700430 struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&sem_ids(ns), id);
Davidlohr Bueso16df3672013-04-30 19:15:29 -0700431
432 if (IS_ERR(ipcp))
433 return ERR_CAST(ipcp);
434
435 return container_of(ipcp, struct sem_array, sem_perm);
436}
437
Davidlohr Bueso16df3672013-04-30 19:15:29 -0700438static inline struct sem_array *sem_obtain_object_check(struct ipc_namespace *ns,
439 int id)
440{
441 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&sem_ids(ns), id);
442
443 if (IS_ERR(ipcp))
444 return ERR_CAST(ipcp);
Pierre Peifferb1ed88b2008-02-06 01:36:23 -0800445
Nadia Derbey03f02c72007-10-18 23:40:51 -0700446 return container_of(ipcp, struct sem_array, sem_perm);
Nadia Derbey023a5352007-10-18 23:40:51 -0700447}
448
Pierre Peiffer6ff37972008-04-29 01:00:46 -0700449static inline void sem_lock_and_putref(struct sem_array *sma)
450{
Rik van Riel6062a8d2013-04-30 19:15:44 -0700451 sem_lock(sma, NULL, -1);
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -0700452 ipc_rcu_putref(sma, ipc_rcu_free);
Pierre Peiffer6ff37972008-04-29 01:00:46 -0700453}
454
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700455static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
456{
457 ipc_rmid(&sem_ids(ns), &s->sem_perm);
458}
459
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460/*
461 * Lockless wakeup algorithm:
462 * Without the check/retry algorithm a lockless wakeup is possible:
463 * - queue.status is initialized to -EINTR before blocking.
464 * - wakeup is performed by
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700465 * * unlinking the queue entry from the pending list
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 * * setting queue.status to IN_WAKEUP
467 * This is the notification for the blocked thread that a
468 * result value is imminent.
469 * * call wake_up_process
470 * * set queue.status to the final value.
471 * - the previously blocked thread checks queue.status:
Manfred Spraul239521f2014-01-27 17:07:04 -0800472 * * if it's IN_WAKEUP, then it must wait until the value changes
473 * * if it's not -EINTR, then the operation was completed by
474 * update_queue. semtimedop can return queue.status without
475 * performing any operation on the sem array.
476 * * otherwise it must acquire the spinlock and check what's up.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 *
478 * The two-stage algorithm is necessary to protect against the following
479 * races:
480 * - if queue.status is set after wake_up_process, then the woken up idle
481 * thread could race forward and try (and fail) to acquire sma->lock
482 * before update_queue had a chance to set queue.status
483 * - if queue.status is written before wake_up_process and if the
484 * blocked process is woken up by a signal between writing
485 * queue.status and the wake_up_process, then the woken up
486 * process could return from semtimedop and die by calling
487 * sys_exit before wake_up_process is called. Then wake_up_process
488 * will oops, because the task structure is already invalid.
489 * (yes, this happened on s390 with sysv msg).
490 *
491 */
492#define IN_WAKEUP 1
493
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700494/**
495 * newary - Create a new semaphore set
496 * @ns: namespace
497 * @params: ptr to the structure that contains key, semflg and nsems
498 *
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700499 * Called with sem_ids.rwsem held (as a writer)
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700500 */
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700501static int newary(struct ipc_namespace *ns, struct ipc_params *params)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502{
503 int id;
504 int retval;
505 struct sem_array *sma;
506 int size;
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700507 key_t key = params->key;
508 int nsems = params->u.nsems;
509 int semflg = params->flg;
Manfred Spraulb97e8202009-12-15 16:47:32 -0800510 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511
512 if (!nsems)
513 return -EINVAL;
Kirill Korotaeve3893532006-10-02 02:18:22 -0700514 if (ns->used_sems + nsems > ns->sc_semmns)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 return -ENOSPC;
516
Manfred Spraul239521f2014-01-27 17:07:04 -0800517 size = sizeof(*sma) + nsems * sizeof(struct sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 sma = ipc_rcu_alloc(size);
Davidlohr Bueso3ab08fe2014-01-27 17:07:06 -0800519 if (!sma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 return -ENOMEM;
Davidlohr Bueso3ab08fe2014-01-27 17:07:06 -0800521
Manfred Spraul239521f2014-01-27 17:07:04 -0800522 memset(sma, 0, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523
524 sma->sem_perm.mode = (semflg & S_IRWXUGO);
525 sma->sem_perm.key = key;
526
527 sma->sem_perm.security = NULL;
528 retval = security_sem_alloc(sma);
529 if (retval) {
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -0700530 ipc_rcu_putref(sma, ipc_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 return retval;
532 }
533
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 sma->sem_base = (struct sem *) &sma[1];
Manfred Spraulb97e8202009-12-15 16:47:32 -0800535
Rik van Riel6062a8d2013-04-30 19:15:44 -0700536 for (i = 0; i < nsems; i++) {
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700537 INIT_LIST_HEAD(&sma->sem_base[i].pending_alter);
538 INIT_LIST_HEAD(&sma->sem_base[i].pending_const);
Rik van Riel6062a8d2013-04-30 19:15:44 -0700539 spin_lock_init(&sma->sem_base[i].lock);
540 }
Manfred Spraulb97e8202009-12-15 16:47:32 -0800541
542 sma->complex_count = 0;
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700543 INIT_LIST_HEAD(&sma->pending_alter);
544 INIT_LIST_HEAD(&sma->pending_const);
Manfred Spraul4daa28f2008-07-25 01:48:04 -0700545 INIT_LIST_HEAD(&sma->list_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 sma->sem_nsems = nsems;
547 sma->sem_ctime = get_seconds();
Manfred Spraule8577d12014-12-02 15:59:34 -0800548
549 id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
550 if (id < 0) {
551 ipc_rcu_putref(sma, sem_rcu_free);
552 return id;
553 }
554 ns->used_sems += nsems;
555
Rik van Riel6062a8d2013-04-30 19:15:44 -0700556 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -0700557 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700559 return sma->sem_perm.id;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560}
561
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700562
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700563/*
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700564 * Called with sem_ids.rwsem and ipcp locked.
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700565 */
Nadia Derbey03f02c72007-10-18 23:40:51 -0700566static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567{
Nadia Derbey03f02c72007-10-18 23:40:51 -0700568 struct sem_array *sma;
569
570 sma = container_of(ipcp, struct sem_array, sem_perm);
571 return security_sem_associate(sma, semflg);
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700572}
573
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700574/*
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700575 * Called with sem_ids.rwsem and ipcp locked.
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700576 */
Nadia Derbey03f02c72007-10-18 23:40:51 -0700577static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
578 struct ipc_params *params)
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700579{
Nadia Derbey03f02c72007-10-18 23:40:51 -0700580 struct sem_array *sma;
581
582 sma = container_of(ipcp, struct sem_array, sem_perm);
583 if (params->u.nsems > sma->sem_nsems)
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700584 return -EINVAL;
585
586 return 0;
587}
588
Heiko Carstensd5460c92009-01-14 14:14:27 +0100589SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg)
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700590{
Kirill Korotaeve3893532006-10-02 02:18:22 -0700591 struct ipc_namespace *ns;
Mathias Krauseeb66ec42014-06-06 14:37:36 -0700592 static const struct ipc_ops sem_ops = {
593 .getnew = newary,
594 .associate = sem_security,
595 .more_checks = sem_more_checks,
596 };
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700597 struct ipc_params sem_params;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598
Kirill Korotaeve3893532006-10-02 02:18:22 -0700599 ns = current->nsproxy->ipc_ns;
600
601 if (nsems < 0 || nsems > ns->sc_semmsl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 return -EINVAL;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700603
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700604 sem_params.key = key;
605 sem_params.flg = semflg;
606 sem_params.u.nsems = nsems;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700607
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700608 return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609}
610
Petr Mladek78f50092014-01-27 17:07:00 -0800611/**
612 * perform_atomic_semop - Perform (if possible) a semaphore operation
Manfred Spraul758a6ba32013-07-08 16:01:26 -0700613 * @sma: semaphore array
Manfred Sprauld198cd62014-06-06 14:37:49 -0700614 * @q: struct sem_queue that describes the operation
Manfred Spraul758a6ba32013-07-08 16:01:26 -0700615 *
616 * Returns 0 if the operation was possible.
617 * Returns 1 if the operation is impossible, the caller must sleep.
618 * Negative values are error codes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 */
Manfred Sprauld198cd62014-06-06 14:37:49 -0700620static int perform_atomic_semop(struct sem_array *sma, struct sem_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621{
Manfred Sprauld198cd62014-06-06 14:37:49 -0700622 int result, sem_op, nsops, pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 struct sembuf *sop;
Manfred Spraul239521f2014-01-27 17:07:04 -0800624 struct sem *curr;
Manfred Sprauld198cd62014-06-06 14:37:49 -0700625 struct sembuf *sops;
626 struct sem_undo *un;
627
628 sops = q->sops;
629 nsops = q->nsops;
630 un = q->undo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631
632 for (sop = sops; sop < sops + nsops; sop++) {
633 curr = sma->sem_base + sop->sem_num;
634 sem_op = sop->sem_op;
635 result = curr->semval;
Petr Mladek78f50092014-01-27 17:07:00 -0800636
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 if (!sem_op && result)
638 goto would_block;
639
640 result += sem_op;
641 if (result < 0)
642 goto would_block;
643 if (result > SEMVMX)
644 goto out_of_range;
Petr Mladek78f50092014-01-27 17:07:00 -0800645
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 if (sop->sem_flg & SEM_UNDO) {
647 int undo = un->semadj[sop->sem_num] - sem_op;
Petr Mladek78f50092014-01-27 17:07:00 -0800648 /* Exceeding the undo range is an error. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
650 goto out_of_range;
Petr Mladek78f50092014-01-27 17:07:00 -0800651 un->semadj[sop->sem_num] = undo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 }
Petr Mladek78f50092014-01-27 17:07:00 -0800653
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 curr->semval = result;
655 }
656
657 sop--;
Manfred Sprauld198cd62014-06-06 14:37:49 -0700658 pid = q->pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659 while (sop >= sops) {
660 sma->sem_base[sop->sem_num].sempid = pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 sop--;
662 }
Petr Mladek78f50092014-01-27 17:07:00 -0800663
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 return 0;
665
666out_of_range:
667 result = -ERANGE;
668 goto undo;
669
670would_block:
Manfred Sprauled247b72014-06-06 14:37:49 -0700671 q->blocking = sop;
672
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 if (sop->sem_flg & IPC_NOWAIT)
674 result = -EAGAIN;
675 else
676 result = 1;
677
678undo:
679 sop--;
680 while (sop >= sops) {
Petr Mladek78f50092014-01-27 17:07:00 -0800681 sem_op = sop->sem_op;
682 sma->sem_base[sop->sem_num].semval -= sem_op;
683 if (sop->sem_flg & SEM_UNDO)
684 un->semadj[sop->sem_num] += sem_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 sop--;
686 }
687
688 return result;
689}
690
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700691/** wake_up_sem_queue_prepare(q, error): Prepare wake-up
692 * @q: queue entry that must be signaled
693 * @error: Error value for the signal
694 *
695 * Prepare the wake-up of the queue entry q.
Nick Piggind4212092009-12-15 16:47:30 -0800696 */
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700697static void wake_up_sem_queue_prepare(struct list_head *pt,
698 struct sem_queue *q, int error)
Nick Piggind4212092009-12-15 16:47:30 -0800699{
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700700 if (list_empty(pt)) {
701 /*
702 * Hold preempt off so that we don't get preempted and have the
703 * wakee busy-wait until we're scheduled back on.
704 */
705 preempt_disable();
706 }
Nick Piggind4212092009-12-15 16:47:30 -0800707 q->status = IN_WAKEUP;
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700708 q->pid = error;
709
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700710 list_add_tail(&q->list, pt);
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700711}
712
713/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800714 * wake_up_sem_queue_do - do the actual wake-up
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700715 * @pt: list of tasks to be woken up
716 *
717 * Do the actual wake-up.
718 * The function is called without any locks held, thus the semaphore array
719 * could be destroyed already and the tasks can disappear as soon as the
720 * status is set to the actual return code.
721 */
722static void wake_up_sem_queue_do(struct list_head *pt)
723{
724 struct sem_queue *q, *t;
725 int did_something;
726
727 did_something = !list_empty(pt);
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700728 list_for_each_entry_safe(q, t, pt, list) {
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700729 wake_up_process(q->sleeper);
730 /* q can disappear immediately after writing q->status. */
731 smp_wmb();
732 q->status = q->pid;
733 }
734 if (did_something)
735 preempt_enable();
Nick Piggind4212092009-12-15 16:47:30 -0800736}
737
Manfred Spraulb97e8202009-12-15 16:47:32 -0800738static void unlink_queue(struct sem_array *sma, struct sem_queue *q)
739{
740 list_del(&q->list);
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700741 if (q->nsops > 1)
Manfred Spraulb97e8202009-12-15 16:47:32 -0800742 sma->complex_count--;
743}
744
Manfred Spraulfd5db422010-05-26 14:43:40 -0700745/** check_restart(sma, q)
746 * @sma: semaphore array
747 * @q: the operation that just completed
748 *
749 * update_queue is O(N^2) when it restarts scanning the whole queue of
750 * waiting operations. Therefore this function checks if the restart is
751 * really necessary. It is called after a previously waiting operation
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700752 * modified the array.
753 * Note that wait-for-zero operations are handled without restart.
Manfred Spraulfd5db422010-05-26 14:43:40 -0700754 */
755static int check_restart(struct sem_array *sma, struct sem_queue *q)
756{
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700757 /* pending complex alter operations are too difficult to analyse */
758 if (!list_empty(&sma->pending_alter))
Manfred Spraulfd5db422010-05-26 14:43:40 -0700759 return 1;
760
761 /* we were a sleeping complex operation. Too difficult */
762 if (q->nsops > 1)
763 return 1;
764
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700765 /* It is impossible that someone waits for the new value:
766 * - complex operations always restart.
767 * - wait-for-zero are handled seperately.
768 * - q is a previously sleeping simple operation that
769 * altered the array. It must be a decrement, because
770 * simple increments never sleep.
771 * - If there are older (higher priority) decrements
772 * in the queue, then they have observed the original
773 * semval value and couldn't proceed. The operation
774 * decremented to value - thus they won't proceed either.
775 */
776 return 0;
777}
Manfred Spraulfd5db422010-05-26 14:43:40 -0700778
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700779/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800780 * wake_const_ops - wake up non-alter tasks
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700781 * @sma: semaphore array.
782 * @semnum: semaphore that was modified.
783 * @pt: list head for the tasks that must be woken up.
784 *
785 * wake_const_ops must be called after a semaphore in a semaphore array
786 * was set to 0. If complex const operations are pending, wake_const_ops must
787 * be called with semnum = -1, as well as with the number of each modified
788 * semaphore.
789 * The tasks that must be woken up are added to @pt. The return code
790 * is stored in q->pid.
791 * The function returns 1 if at least one operation was completed successfully.
792 */
793static int wake_const_ops(struct sem_array *sma, int semnum,
794 struct list_head *pt)
795{
796 struct sem_queue *q;
797 struct list_head *walk;
798 struct list_head *pending_list;
799 int semop_completed = 0;
Manfred Spraulfd5db422010-05-26 14:43:40 -0700800
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700801 if (semnum == -1)
802 pending_list = &sma->pending_const;
803 else
804 pending_list = &sma->sem_base[semnum].pending_const;
805
806 walk = pending_list->next;
807 while (walk != pending_list) {
808 int error;
809
810 q = container_of(walk, struct sem_queue, list);
811 walk = walk->next;
812
Manfred Sprauld198cd62014-06-06 14:37:49 -0700813 error = perform_atomic_semop(sma, q);
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700814
815 if (error <= 0) {
816 /* operation completed, remove from queue & wakeup */
817
818 unlink_queue(sma, q);
819
820 wake_up_sem_queue_prepare(pt, q, error);
821 if (error == 0)
822 semop_completed = 1;
823 }
824 }
825 return semop_completed;
826}
827
828/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800829 * do_smart_wakeup_zero - wakeup all wait for zero tasks
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700830 * @sma: semaphore array
831 * @sops: operations that were performed
832 * @nsops: number of operations
833 * @pt: list head of the tasks that must be woken up.
834 *
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800835 * Checks all required queue for wait-for-zero operations, based
836 * on the actual changes that were performed on the semaphore array.
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700837 * The function returns 1 if at least one operation was completed successfully.
838 */
839static int do_smart_wakeup_zero(struct sem_array *sma, struct sembuf *sops,
840 int nsops, struct list_head *pt)
841{
842 int i;
843 int semop_completed = 0;
844 int got_zero = 0;
845
846 /* first: the per-semaphore queues, if known */
847 if (sops) {
848 for (i = 0; i < nsops; i++) {
849 int num = sops[i].sem_num;
850
851 if (sma->sem_base[num].semval == 0) {
852 got_zero = 1;
853 semop_completed |= wake_const_ops(sma, num, pt);
854 }
855 }
856 } else {
857 /*
858 * No sops means modified semaphores not known.
859 * Assume all were changed.
Manfred Spraulfd5db422010-05-26 14:43:40 -0700860 */
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700861 for (i = 0; i < sma->sem_nsems; i++) {
862 if (sma->sem_base[i].semval == 0) {
863 got_zero = 1;
864 semop_completed |= wake_const_ops(sma, i, pt);
865 }
866 }
Manfred Spraulfd5db422010-05-26 14:43:40 -0700867 }
868 /*
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700869 * If one of the modified semaphores got 0,
870 * then check the global queue, too.
Manfred Spraulfd5db422010-05-26 14:43:40 -0700871 */
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700872 if (got_zero)
873 semop_completed |= wake_const_ops(sma, -1, pt);
Manfred Spraulfd5db422010-05-26 14:43:40 -0700874
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700875 return semop_completed;
Manfred Spraulfd5db422010-05-26 14:43:40 -0700876}
877
Manfred Spraul636c6be2009-12-15 16:47:33 -0800878
879/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800880 * update_queue - look for tasks that can be completed.
Manfred Spraul636c6be2009-12-15 16:47:33 -0800881 * @sma: semaphore array.
882 * @semnum: semaphore that was modified.
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700883 * @pt: list head for the tasks that must be woken up.
Manfred Spraul636c6be2009-12-15 16:47:33 -0800884 *
885 * update_queue must be called after a semaphore in a semaphore array
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700886 * was modified. If multiple semaphores were modified, update_queue must
887 * be called with semnum = -1, as well as with the number of each modified
888 * semaphore.
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700889 * The tasks that must be woken up are added to @pt. The return code
890 * is stored in q->pid.
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700891 * The function internally checks if const operations can now succeed.
892 *
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700893 * The function return 1 if at least one semop was completed successfully.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 */
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700895static int update_queue(struct sem_array *sma, int semnum, struct list_head *pt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896{
Manfred Spraul636c6be2009-12-15 16:47:33 -0800897 struct sem_queue *q;
898 struct list_head *walk;
899 struct list_head *pending_list;
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700900 int semop_completed = 0;
Manfred Spraul636c6be2009-12-15 16:47:33 -0800901
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700902 if (semnum == -1)
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700903 pending_list = &sma->pending_alter;
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700904 else
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700905 pending_list = &sma->sem_base[semnum].pending_alter;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906
Nick Piggin9cad2002009-12-15 16:47:29 -0800907again:
Manfred Spraul636c6be2009-12-15 16:47:33 -0800908 walk = pending_list->next;
909 while (walk != pending_list) {
Manfred Spraulfd5db422010-05-26 14:43:40 -0700910 int error, restart;
Manfred Spraul636c6be2009-12-15 16:47:33 -0800911
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700912 q = container_of(walk, struct sem_queue, list);
Manfred Spraul636c6be2009-12-15 16:47:33 -0800913 walk = walk->next;
Nick Piggin9cad2002009-12-15 16:47:29 -0800914
Manfred Sprauld987f8b2009-12-15 16:47:34 -0800915 /* If we are scanning the single sop, per-semaphore list of
916 * one semaphore and that semaphore is 0, then it is not
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700917 * necessary to scan further: simple increments
Manfred Sprauld987f8b2009-12-15 16:47:34 -0800918 * that affect only one entry succeed immediately and cannot
919 * be in the per semaphore pending queue, and decrements
920 * cannot be successful if the value is already 0.
921 */
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700922 if (semnum != -1 && sma->sem_base[semnum].semval == 0)
Manfred Sprauld987f8b2009-12-15 16:47:34 -0800923 break;
924
Manfred Sprauld198cd62014-06-06 14:37:49 -0700925 error = perform_atomic_semop(sma, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926
927 /* Does q->sleeper still need to sleep? */
Nick Piggin9cad2002009-12-15 16:47:29 -0800928 if (error > 0)
929 continue;
Manfred Spraula1193f82008-07-25 01:48:06 -0700930
Manfred Spraulb97e8202009-12-15 16:47:32 -0800931 unlink_queue(sma, q);
Manfred Spraula1193f82008-07-25 01:48:06 -0700932
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700933 if (error) {
Manfred Spraulfd5db422010-05-26 14:43:40 -0700934 restart = 0;
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700935 } else {
936 semop_completed = 1;
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700937 do_smart_wakeup_zero(sma, q->sops, q->nsops, pt);
Manfred Spraulfd5db422010-05-26 14:43:40 -0700938 restart = check_restart(sma, q);
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700939 }
Manfred Spraulfd5db422010-05-26 14:43:40 -0700940
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700941 wake_up_sem_queue_prepare(pt, q, error);
Manfred Spraulfd5db422010-05-26 14:43:40 -0700942 if (restart)
Nick Piggin9cad2002009-12-15 16:47:29 -0800943 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 }
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700945 return semop_completed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946}
947
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700948/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800949 * set_semotime - set sem_otime
Manfred Spraul0e8c6652013-09-30 13:45:25 -0700950 * @sma: semaphore array
951 * @sops: operations that modified the array, may be NULL
952 *
953 * sem_otime is replicated to avoid cache line trashing.
954 * This function sets one instance to the current time.
955 */
956static void set_semotime(struct sem_array *sma, struct sembuf *sops)
957{
958 if (sops == NULL) {
959 sma->sem_base[0].sem_otime = get_seconds();
960 } else {
961 sma->sem_base[sops[0].sem_num].sem_otime =
962 get_seconds();
963 }
964}
965
966/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800967 * do_smart_update - optimized update_queue
Manfred Spraulfd5db422010-05-26 14:43:40 -0700968 * @sma: semaphore array
969 * @sops: operations that were performed
970 * @nsops: number of operations
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700971 * @otime: force setting otime
972 * @pt: list head of the tasks that must be woken up.
Manfred Spraulfd5db422010-05-26 14:43:40 -0700973 *
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700974 * do_smart_update() does the required calls to update_queue and wakeup_zero,
975 * based on the actual changes that were performed on the semaphore array.
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700976 * Note that the function does not do the actual wake-up: the caller is
977 * responsible for calling wake_up_sem_queue_do(@pt).
978 * It is safe to perform this call after dropping all locks.
Manfred Spraulfd5db422010-05-26 14:43:40 -0700979 */
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700980static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops,
981 int otime, struct list_head *pt)
Manfred Spraulfd5db422010-05-26 14:43:40 -0700982{
983 int i;
984
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700985 otime |= do_smart_wakeup_zero(sma, sops, nsops, pt);
986
Manfred Spraulf269f402013-07-08 16:01:24 -0700987 if (!list_empty(&sma->pending_alter)) {
988 /* semaphore array uses the global queue - just process it. */
989 otime |= update_queue(sma, -1, pt);
990 } else {
991 if (!sops) {
992 /*
993 * No sops, thus the modified semaphores are not
994 * known. Check all.
995 */
996 for (i = 0; i < sma->sem_nsems; i++)
997 otime |= update_queue(sma, i, pt);
998 } else {
999 /*
1000 * Check the semaphores that were increased:
1001 * - No complex ops, thus all sleeping ops are
1002 * decrease.
1003 * - if we decreased the value, then any sleeping
1004 * semaphore ops wont be able to run: If the
1005 * previous value was too small, then the new
1006 * value will be too small, too.
1007 */
1008 for (i = 0; i < nsops; i++) {
1009 if (sops[i].sem_op > 0) {
1010 otime |= update_queue(sma,
1011 sops[i].sem_num, pt);
1012 }
Manfred Spraulab465df2013-05-26 11:08:52 +02001013 }
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001014 }
Manfred Spraulfd5db422010-05-26 14:43:40 -07001015 }
Manfred Spraul0e8c6652013-09-30 13:45:25 -07001016 if (otime)
1017 set_semotime(sma, sops);
Manfred Spraulfd5db422010-05-26 14:43:40 -07001018}
1019
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001020/*
Manfred Spraulb220c572014-06-06 14:37:51 -07001021 * check_qop: Test if a queued operation sleeps on the semaphore semnum
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001022 */
1023static int check_qop(struct sem_array *sma, int semnum, struct sem_queue *q,
1024 bool count_zero)
1025{
Manfred Spraulb220c572014-06-06 14:37:51 -07001026 struct sembuf *sop = q->blocking;
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001027
Manfred Spraul9b44ee22014-06-06 14:37:52 -07001028 /*
1029 * Linux always (since 0.99.10) reported a task as sleeping on all
1030 * semaphores. This violates SUS, therefore it was changed to the
1031 * standard compliant behavior.
1032 * Give the administrators a chance to notice that an application
1033 * might misbehave because it relies on the Linux behavior.
1034 */
1035 pr_info_once("semctl(GETNCNT/GETZCNT) is since 3.16 Single Unix Specification compliant.\n"
1036 "The task %s (%d) triggered the difference, watch for misbehavior.\n",
1037 current->comm, task_pid_nr(current));
1038
Manfred Spraulb220c572014-06-06 14:37:51 -07001039 if (sop->sem_num != semnum)
1040 return 0;
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001041
Manfred Spraulb220c572014-06-06 14:37:51 -07001042 if (count_zero && sop->sem_op == 0)
1043 return 1;
1044 if (!count_zero && sop->sem_op < 0)
1045 return 1;
1046
1047 return 0;
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001048}
1049
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050/* The following counts are associated to each semaphore:
1051 * semncnt number of tasks waiting on semval being nonzero
1052 * semzcnt number of tasks waiting on semval being zero
Manfred Spraulb220c572014-06-06 14:37:51 -07001053 *
1054 * Per definition, a task waits only on the semaphore of the first semop
1055 * that cannot proceed, even if additional operation would block, too.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 */
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001057static int count_semcnt(struct sem_array *sma, ushort semnum,
1058 bool count_zero)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059{
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001060 struct list_head *l;
Manfred Spraul239521f2014-01-27 17:07:04 -08001061 struct sem_queue *q;
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001062 int semcnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001064 semcnt = 0;
1065 /* First: check the simple operations. They are easy to evaluate */
1066 if (count_zero)
1067 l = &sma->sem_base[semnum].pending_const;
1068 else
1069 l = &sma->sem_base[semnum].pending_alter;
1070
1071 list_for_each_entry(q, l, list) {
1072 /* all task on a per-semaphore list sleep on exactly
1073 * that semaphore
1074 */
1075 semcnt++;
Rik van Rielde2657f2013-05-09 16:59:59 -04001076 }
1077
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001078 /* Then: check the complex operations. */
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001079 list_for_each_entry(q, &sma->pending_alter, list) {
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001080 semcnt += check_qop(sma, semnum, q, count_zero);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 }
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001082 if (count_zero) {
1083 list_for_each_entry(q, &sma->pending_const, list) {
1084 semcnt += check_qop(sma, semnum, q, count_zero);
1085 }
Rik van Rielebc2e5e2013-05-09 16:53:28 -04001086 }
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001087 return semcnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088}
1089
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001090/* Free a semaphore set. freeary() is called with sem_ids.rwsem locked
1091 * as a writer and the spinlock for this semaphore set hold. sem_ids.rwsem
Nadia Derbey3e148c72007-10-18 23:40:54 -07001092 * remains locked on exit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 */
Pierre Peiffer01b8b072008-02-08 04:18:57 -08001094static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095{
Manfred Spraul380af1b2008-07-25 01:48:06 -07001096 struct sem_undo *un, *tu;
1097 struct sem_queue *q, *tq;
Pierre Peiffer01b8b072008-02-08 04:18:57 -08001098 struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001099 struct list_head tasks;
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001100 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101
Manfred Spraul380af1b2008-07-25 01:48:06 -07001102 /* Free the existing undo structures for this semaphore set. */
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -07001103 ipc_assert_locked_object(&sma->sem_perm);
Manfred Spraul380af1b2008-07-25 01:48:06 -07001104 list_for_each_entry_safe(un, tu, &sma->list_id, list_id) {
1105 list_del(&un->list_id);
1106 spin_lock(&un->ulp->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107 un->semid = -1;
Manfred Spraul380af1b2008-07-25 01:48:06 -07001108 list_del_rcu(&un->list_proc);
1109 spin_unlock(&un->ulp->lock);
Lai Jiangshan693a8b62011-03-18 12:09:35 +08001110 kfree_rcu(un, rcu);
Manfred Spraul380af1b2008-07-25 01:48:06 -07001111 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112
1113 /* Wake up all pending processes and let them fail with EIDRM. */
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001114 INIT_LIST_HEAD(&tasks);
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001115 list_for_each_entry_safe(q, tq, &sma->pending_const, list) {
1116 unlink_queue(sma, q);
1117 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1118 }
1119
1120 list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
Manfred Spraulb97e8202009-12-15 16:47:32 -08001121 unlink_queue(sma, q);
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001122 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 }
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001124 for (i = 0; i < sma->sem_nsems; i++) {
1125 struct sem *sem = sma->sem_base + i;
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001126 list_for_each_entry_safe(q, tq, &sem->pending_const, list) {
1127 unlink_queue(sma, q);
1128 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1129 }
1130 list_for_each_entry_safe(q, tq, &sem->pending_alter, list) {
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001131 unlink_queue(sma, q);
1132 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1133 }
1134 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135
Nadia Derbey7ca7e562007-10-18 23:40:48 -07001136 /* Remove the semaphore set from the IDR */
1137 sem_rmid(ns, sma);
Rik van Riel6062a8d2013-04-30 19:15:44 -07001138 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001139 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001141 wake_up_sem_queue_do(&tasks);
Kirill Korotaeve3893532006-10-02 02:18:22 -07001142 ns->used_sems -= sma->sem_nsems;
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -07001143 ipc_rcu_putref(sma, sem_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144}
1145
1146static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
1147{
Manfred Spraul239521f2014-01-27 17:07:04 -08001148 switch (version) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 case IPC_64:
1150 return copy_to_user(buf, in, sizeof(*in));
1151 case IPC_OLD:
1152 {
1153 struct semid_ds out;
1154
Dan Rosenberg982f7c22010-09-30 15:15:31 -07001155 memset(&out, 0, sizeof(out));
1156
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
1158
1159 out.sem_otime = in->sem_otime;
1160 out.sem_ctime = in->sem_ctime;
1161 out.sem_nsems = in->sem_nsems;
1162
1163 return copy_to_user(buf, &out, sizeof(out));
1164 }
1165 default:
1166 return -EINVAL;
1167 }
1168}
1169
Manfred Sprauld12e1e52013-07-08 16:01:25 -07001170static time_t get_semotime(struct sem_array *sma)
1171{
1172 int i;
1173 time_t res;
1174
1175 res = sma->sem_base[0].sem_otime;
1176 for (i = 1; i < sma->sem_nsems; i++) {
1177 time_t to = sma->sem_base[i].sem_otime;
1178
1179 if (to > res)
1180 res = to;
1181 }
1182 return res;
1183}
1184
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -08001185static int semctl_nolock(struct ipc_namespace *ns, int semid,
Al Viroe1fd1f42013-03-05 15:04:55 -05001186 int cmd, int version, void __user *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187{
Amerigo Wange5cc9c72009-12-15 16:47:35 -08001188 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189 struct sem_array *sma;
1190
Manfred Spraul239521f2014-01-27 17:07:04 -08001191 switch (cmd) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 case IPC_INFO:
1193 case SEM_INFO:
1194 {
1195 struct seminfo seminfo;
1196 int max_id;
1197
1198 err = security_sem_semctl(NULL, cmd);
1199 if (err)
1200 return err;
Paul McQuade46c0a8c2014-06-06 14:37:37 -07001201
Manfred Spraul239521f2014-01-27 17:07:04 -08001202 memset(&seminfo, 0, sizeof(seminfo));
Kirill Korotaeve3893532006-10-02 02:18:22 -07001203 seminfo.semmni = ns->sc_semmni;
1204 seminfo.semmns = ns->sc_semmns;
1205 seminfo.semmsl = ns->sc_semmsl;
1206 seminfo.semopm = ns->sc_semopm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 seminfo.semvmx = SEMVMX;
1208 seminfo.semmnu = SEMMNU;
1209 seminfo.semmap = SEMMAP;
1210 seminfo.semume = SEMUME;
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001211 down_read(&sem_ids(ns).rwsem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 if (cmd == SEM_INFO) {
Kirill Korotaeve3893532006-10-02 02:18:22 -07001213 seminfo.semusz = sem_ids(ns).in_use;
1214 seminfo.semaem = ns->used_sems;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 } else {
1216 seminfo.semusz = SEMUSZ;
1217 seminfo.semaem = SEMAEM;
1218 }
Nadia Derbey7ca7e562007-10-18 23:40:48 -07001219 max_id = ipc_get_maxid(&sem_ids(ns));
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001220 up_read(&sem_ids(ns).rwsem);
Paul McQuade46c0a8c2014-06-06 14:37:37 -07001221 if (copy_to_user(p, &seminfo, sizeof(struct seminfo)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 return -EFAULT;
Manfred Spraul239521f2014-01-27 17:07:04 -08001223 return (max_id < 0) ? 0 : max_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 }
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -08001225 case IPC_STAT:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 case SEM_STAT:
1227 {
1228 struct semid64_ds tbuf;
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001229 int id = 0;
1230
1231 memset(&tbuf, 0, sizeof(tbuf));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232
Linus Torvalds941b0302013-05-04 11:04:29 -07001233 rcu_read_lock();
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -08001234 if (cmd == SEM_STAT) {
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001235 sma = sem_obtain_object(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 id = sma->sem_perm.id;
1241 } else {
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001242 sma = sem_obtain_object_check(ns, semid);
1243 if (IS_ERR(sma)) {
1244 err = PTR_ERR(sma);
1245 goto out_unlock;
1246 }
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -08001247 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248
1249 err = -EACCES;
Serge E. Hallynb0e77592011-03-23 16:43:24 -07001250 if (ipcperms(ns, &sma->sem_perm, S_IRUGO))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251 goto out_unlock;
1252
1253 err = security_sem_semctl(sma, cmd);
1254 if (err)
1255 goto out_unlock;
1256
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
Manfred Sprauld12e1e52013-07-08 16:01:25 -07001258 tbuf.sem_otime = get_semotime(sma);
1259 tbuf.sem_ctime = sma->sem_ctime;
1260 tbuf.sem_nsems = sma->sem_nsems;
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001261 rcu_read_unlock();
Al Viroe1fd1f42013-03-05 15:04:55 -05001262 if (copy_semid_to_user(p, &tbuf, version))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 return -EFAULT;
1264 return id;
1265 }
1266 default:
1267 return -EINVAL;
1268 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269out_unlock:
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001270 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 return err;
1272}
1273
Al Viroe1fd1f42013-03-05 15:04:55 -05001274static int semctl_setval(struct ipc_namespace *ns, int semid, int semnum,
1275 unsigned long arg)
1276{
1277 struct sem_undo *un;
1278 struct sem_array *sma;
Manfred Spraul239521f2014-01-27 17:07:04 -08001279 struct sem *curr;
Al Viroe1fd1f42013-03-05 15:04:55 -05001280 int err;
Al Viroe1fd1f42013-03-05 15:04:55 -05001281 struct list_head tasks;
1282 int val;
1283#if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN)
1284 /* big-endian 64bit */
1285 val = arg >> 32;
1286#else
1287 /* 32bit or little-endian 64bit */
1288 val = arg;
1289#endif
1290
Rik van Riel6062a8d2013-04-30 19:15:44 -07001291 if (val > SEMVMX || val < 0)
1292 return -ERANGE;
Al Viroe1fd1f42013-03-05 15:04:55 -05001293
1294 INIT_LIST_HEAD(&tasks);
Al Viroe1fd1f42013-03-05 15:04:55 -05001295
Rik van Riel6062a8d2013-04-30 19:15:44 -07001296 rcu_read_lock();
1297 sma = sem_obtain_object_check(ns, semid);
1298 if (IS_ERR(sma)) {
1299 rcu_read_unlock();
1300 return PTR_ERR(sma);
1301 }
1302
1303 if (semnum < 0 || semnum >= sma->sem_nsems) {
1304 rcu_read_unlock();
1305 return -EINVAL;
1306 }
1307
1308
1309 if (ipcperms(ns, &sma->sem_perm, S_IWUGO)) {
1310 rcu_read_unlock();
1311 return -EACCES;
1312 }
Al Viroe1fd1f42013-03-05 15:04:55 -05001313
1314 err = security_sem_semctl(sma, SETVAL);
Rik van Riel6062a8d2013-04-30 19:15:44 -07001315 if (err) {
1316 rcu_read_unlock();
1317 return -EACCES;
1318 }
Al Viroe1fd1f42013-03-05 15:04:55 -05001319
Rik van Riel6062a8d2013-04-30 19:15:44 -07001320 sem_lock(sma, NULL, -1);
Al Viroe1fd1f42013-03-05 15:04:55 -05001321
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001322 if (!ipc_valid_object(&sma->sem_perm)) {
Manfred Spraul6e224f92013-10-16 13:46:45 -07001323 sem_unlock(sma, -1);
1324 rcu_read_unlock();
1325 return -EIDRM;
1326 }
1327
Al Viroe1fd1f42013-03-05 15:04:55 -05001328 curr = &sma->sem_base[semnum];
1329
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -07001330 ipc_assert_locked_object(&sma->sem_perm);
Al Viroe1fd1f42013-03-05 15:04:55 -05001331 list_for_each_entry(un, &sma->list_id, list_id)
1332 un->semadj[semnum] = 0;
1333
1334 curr->semval = val;
1335 curr->sempid = task_tgid_vnr(current);
1336 sma->sem_ctime = get_seconds();
1337 /* maybe some queued-up processes were waiting for this */
1338 do_smart_update(sma, NULL, 0, 0, &tasks);
Rik van Riel6062a8d2013-04-30 19:15:44 -07001339 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001340 rcu_read_unlock();
Al Viroe1fd1f42013-03-05 15:04:55 -05001341 wake_up_sem_queue_do(&tasks);
Rik van Riel6062a8d2013-04-30 19:15:44 -07001342 return 0;
Al Viroe1fd1f42013-03-05 15:04:55 -05001343}
1344
Kirill Korotaeve3893532006-10-02 02:18:22 -07001345static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
Al Viroe1fd1f42013-03-05 15:04:55 -05001346 int cmd, void __user *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347{
1348 struct sem_array *sma;
Manfred Spraul239521f2014-01-27 17:07:04 -08001349 struct sem *curr;
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001350 int err, nsems;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 ushort fast_sem_io[SEMMSL_FAST];
Manfred Spraul239521f2014-01-27 17:07:04 -08001352 ushort *sem_io = fast_sem_io;
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001353 struct list_head tasks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001355 INIT_LIST_HEAD(&tasks);
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001356
1357 rcu_read_lock();
1358 sma = sem_obtain_object_check(ns, semid);
1359 if (IS_ERR(sma)) {
1360 rcu_read_unlock();
1361 return PTR_ERR(sma);
1362 }
1363
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364 nsems = sma->sem_nsems;
1365
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 err = -EACCES;
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001367 if (ipcperms(ns, &sma->sem_perm, cmd == SETALL ? S_IWUGO : S_IRUGO))
1368 goto out_rcu_wakeup;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369
1370 err = security_sem_semctl(sma, cmd);
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001371 if (err)
1372 goto out_rcu_wakeup;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373
1374 err = -EACCES;
1375 switch (cmd) {
1376 case GETALL:
1377 {
Al Viroe1fd1f42013-03-05 15:04:55 -05001378 ushort __user *array = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 int i;
1380
Al Viroce857222013-05-03 00:30:49 +01001381 sem_lock(sma, NULL, -1);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001382 if (!ipc_valid_object(&sma->sem_perm)) {
Manfred Spraul6e224f92013-10-16 13:46:45 -07001383 err = -EIDRM;
1384 goto out_unlock;
1385 }
Manfred Spraul239521f2014-01-27 17:07:04 -08001386 if (nsems > SEMMSL_FAST) {
Al Viroce857222013-05-03 00:30:49 +01001387 if (!ipc_rcu_getref(sma)) {
Al Viroce857222013-05-03 00:30:49 +01001388 err = -EIDRM;
Manfred Spraul6e224f92013-10-16 13:46:45 -07001389 goto out_unlock;
Al Viroce857222013-05-03 00:30:49 +01001390 }
1391 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001392 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393 sem_io = ipc_alloc(sizeof(ushort)*nsems);
Manfred Spraul239521f2014-01-27 17:07:04 -08001394 if (sem_io == NULL) {
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -07001395 ipc_rcu_putref(sma, ipc_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396 return -ENOMEM;
1397 }
1398
Linus Torvalds4091fd942013-05-04 10:13:40 -07001399 rcu_read_lock();
Pierre Peiffer6ff37972008-04-29 01:00:46 -07001400 sem_lock_and_putref(sma);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001401 if (!ipc_valid_object(&sma->sem_perm)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 err = -EIDRM;
Manfred Spraul6e224f92013-10-16 13:46:45 -07001403 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 }
Al Viroce857222013-05-03 00:30:49 +01001405 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 for (i = 0; i < sma->sem_nsems; i++)
1407 sem_io[i] = sma->sem_base[i].semval;
Rik van Riel6062a8d2013-04-30 19:15:44 -07001408 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001409 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 err = 0;
Manfred Spraul239521f2014-01-27 17:07:04 -08001411 if (copy_to_user(array, sem_io, nsems*sizeof(ushort)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 err = -EFAULT;
1413 goto out_free;
1414 }
1415 case SETALL:
1416 {
1417 int i;
1418 struct sem_undo *un;
1419
Rik van Riel6062a8d2013-04-30 19:15:44 -07001420 if (!ipc_rcu_getref(sma)) {
Manfred Spraul6e224f92013-10-16 13:46:45 -07001421 err = -EIDRM;
1422 goto out_rcu_wakeup;
Rik van Riel6062a8d2013-04-30 19:15:44 -07001423 }
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001424 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425
Manfred Spraul239521f2014-01-27 17:07:04 -08001426 if (nsems > SEMMSL_FAST) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 sem_io = ipc_alloc(sizeof(ushort)*nsems);
Manfred Spraul239521f2014-01-27 17:07:04 -08001428 if (sem_io == NULL) {
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -07001429 ipc_rcu_putref(sma, ipc_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 return -ENOMEM;
1431 }
1432 }
1433
Manfred Spraul239521f2014-01-27 17:07:04 -08001434 if (copy_from_user(sem_io, p, nsems*sizeof(ushort))) {
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -07001435 ipc_rcu_putref(sma, ipc_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 err = -EFAULT;
1437 goto out_free;
1438 }
1439
1440 for (i = 0; i < nsems; i++) {
1441 if (sem_io[i] > SEMVMX) {
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -07001442 ipc_rcu_putref(sma, ipc_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 err = -ERANGE;
1444 goto out_free;
1445 }
1446 }
Linus Torvalds4091fd942013-05-04 10:13:40 -07001447 rcu_read_lock();
Pierre Peiffer6ff37972008-04-29 01:00:46 -07001448 sem_lock_and_putref(sma);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001449 if (!ipc_valid_object(&sma->sem_perm)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 err = -EIDRM;
Manfred Spraul6e224f92013-10-16 13:46:45 -07001451 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452 }
1453
Davidlohr Buesoa5f4db82016-03-22 14:27:48 -07001454 for (i = 0; i < nsems; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 sma->sem_base[i].semval = sem_io[i];
Davidlohr Buesoa5f4db82016-03-22 14:27:48 -07001456 sma->sem_base[i].sempid = task_tgid_vnr(current);
1457 }
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001458
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -07001459 ipc_assert_locked_object(&sma->sem_perm);
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001460 list_for_each_entry(un, &sma->list_id, list_id) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 for (i = 0; i < nsems; i++)
1462 un->semadj[i] = 0;
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001463 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464 sma->sem_ctime = get_seconds();
1465 /* maybe some queued-up processes were waiting for this */
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001466 do_smart_update(sma, NULL, 0, 0, &tasks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 err = 0;
1468 goto out_unlock;
1469 }
Al Viroe1fd1f42013-03-05 15:04:55 -05001470 /* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 }
1472 err = -EINVAL;
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001473 if (semnum < 0 || semnum >= nsems)
1474 goto out_rcu_wakeup;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475
Rik van Riel6062a8d2013-04-30 19:15:44 -07001476 sem_lock(sma, NULL, -1);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001477 if (!ipc_valid_object(&sma->sem_perm)) {
Manfred Spraul6e224f92013-10-16 13:46:45 -07001478 err = -EIDRM;
1479 goto out_unlock;
1480 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 curr = &sma->sem_base[semnum];
1482
1483 switch (cmd) {
1484 case GETVAL:
1485 err = curr->semval;
1486 goto out_unlock;
1487 case GETPID:
1488 err = curr->sempid;
1489 goto out_unlock;
1490 case GETNCNT:
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001491 err = count_semcnt(sma, semnum, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492 goto out_unlock;
1493 case GETZCNT:
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001494 err = count_semcnt(sma, semnum, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496 }
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001497
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498out_unlock:
Rik van Riel6062a8d2013-04-30 19:15:44 -07001499 sem_unlock(sma, -1);
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001500out_rcu_wakeup:
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001501 rcu_read_unlock();
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001502 wake_up_sem_queue_do(&tasks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503out_free:
Manfred Spraul239521f2014-01-27 17:07:04 -08001504 if (sem_io != fast_sem_io)
Tetsuo Handa1d5cfdb2016-01-22 15:11:02 -08001505 ipc_free(sem_io);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506 return err;
1507}
1508
Pierre Peiffer016d7132008-04-29 01:00:50 -07001509static inline unsigned long
1510copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511{
Manfred Spraul239521f2014-01-27 17:07:04 -08001512 switch (version) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513 case IPC_64:
Pierre Peiffer016d7132008-04-29 01:00:50 -07001514 if (copy_from_user(out, buf, sizeof(*out)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517 case IPC_OLD:
1518 {
1519 struct semid_ds tbuf_old;
1520
Manfred Spraul239521f2014-01-27 17:07:04 -08001521 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522 return -EFAULT;
1523
Pierre Peiffer016d7132008-04-29 01:00:50 -07001524 out->sem_perm.uid = tbuf_old.sem_perm.uid;
1525 out->sem_perm.gid = tbuf_old.sem_perm.gid;
1526 out->sem_perm.mode = tbuf_old.sem_perm.mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527
1528 return 0;
1529 }
1530 default:
1531 return -EINVAL;
1532 }
1533}
1534
Pierre Peiffer522bb2a2008-04-29 01:00:49 -07001535/*
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001536 * This function handles some semctl commands which require the rwsem
Pierre Peiffer522bb2a2008-04-29 01:00:49 -07001537 * to be held in write mode.
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001538 * NOTE: no locks must be held, the rwsem is taken inside this function.
Pierre Peiffer522bb2a2008-04-29 01:00:49 -07001539 */
Pierre Peiffer21a48262008-04-29 01:00:49 -07001540static int semctl_down(struct ipc_namespace *ns, int semid,
Al Viroe1fd1f42013-03-05 15:04:55 -05001541 int cmd, int version, void __user *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542{
1543 struct sem_array *sma;
1544 int err;
Pierre Peiffer016d7132008-04-29 01:00:50 -07001545 struct semid64_ds semid64;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546 struct kern_ipc_perm *ipcp;
1547
Manfred Spraul239521f2014-01-27 17:07:04 -08001548 if (cmd == IPC_SET) {
Al Viroe1fd1f42013-03-05 15:04:55 -05001549 if (copy_semid_from_user(&semid64, p, version))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001553 down_write(&sem_ids(ns).rwsem);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001554 rcu_read_lock();
1555
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001556 ipcp = ipcctl_pre_down_nolock(ns, &sem_ids(ns), semid, cmd,
1557 &semid64.sem_perm, 0);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001558 if (IS_ERR(ipcp)) {
1559 err = PTR_ERR(ipcp);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001560 goto out_unlock1;
1561 }
Steve Grubb073115d2006-04-02 17:07:33 -04001562
Pierre Peiffera5f75e72008-04-29 01:00:54 -07001563 sma = container_of(ipcp, struct sem_array, sem_perm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564
1565 err = security_sem_semctl(sma, cmd);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001566 if (err)
1567 goto out_unlock1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001569 switch (cmd) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570 case IPC_RMID:
Rik van Riel6062a8d2013-04-30 19:15:44 -07001571 sem_lock(sma, NULL, -1);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001572 /* freeary unlocks the ipc object and rcu */
Pierre Peiffer01b8b072008-02-08 04:18:57 -08001573 freeary(ns, ipcp);
Pierre Peiffer522bb2a2008-04-29 01:00:49 -07001574 goto out_up;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575 case IPC_SET:
Rik van Riel6062a8d2013-04-30 19:15:44 -07001576 sem_lock(sma, NULL, -1);
Eric W. Biederman1efdb692012-02-07 16:54:11 -08001577 err = ipc_update_perm(&semid64.sem_perm, ipcp);
1578 if (err)
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001579 goto out_unlock0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 sma->sem_ctime = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 break;
1582 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583 err = -EINVAL;
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001584 goto out_unlock1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001587out_unlock0:
Rik van Riel6062a8d2013-04-30 19:15:44 -07001588 sem_unlock(sma, -1);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001589out_unlock1:
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001590 rcu_read_unlock();
Pierre Peiffer522bb2a2008-04-29 01:00:49 -07001591out_up:
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001592 up_write(&sem_ids(ns).rwsem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593 return err;
1594}
1595
Al Viroe1fd1f42013-03-05 15:04:55 -05001596SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, unsigned long, arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 int version;
Kirill Korotaeve3893532006-10-02 02:18:22 -07001599 struct ipc_namespace *ns;
Al Viroe1fd1f42013-03-05 15:04:55 -05001600 void __user *p = (void __user *)arg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601
1602 if (semid < 0)
1603 return -EINVAL;
1604
1605 version = ipc_parse_version(&cmd);
Kirill Korotaeve3893532006-10-02 02:18:22 -07001606 ns = current->nsproxy->ipc_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607
Manfred Spraul239521f2014-01-27 17:07:04 -08001608 switch (cmd) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609 case IPC_INFO:
1610 case SEM_INFO:
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -08001611 case IPC_STAT:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612 case SEM_STAT:
Al Viroe1fd1f42013-03-05 15:04:55 -05001613 return semctl_nolock(ns, semid, cmd, version, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614 case GETALL:
1615 case GETVAL:
1616 case GETPID:
1617 case GETNCNT:
1618 case GETZCNT:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619 case SETALL:
Al Viroe1fd1f42013-03-05 15:04:55 -05001620 return semctl_main(ns, semid, semnum, cmd, p);
1621 case SETVAL:
1622 return semctl_setval(ns, semid, semnum, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623 case IPC_RMID:
1624 case IPC_SET:
Al Viroe1fd1f42013-03-05 15:04:55 -05001625 return semctl_down(ns, semid, cmd, version, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626 default:
1627 return -EINVAL;
1628 }
1629}
1630
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631/* If the task doesn't already have a undo_list, then allocate one
1632 * here. We guarantee there is only one thread using this undo list,
1633 * and current is THE ONE
1634 *
1635 * If this allocation and assignment succeeds, but later
1636 * portions of this code fail, there is no need to free the sem_undo_list.
1637 * Just let it stay associated with the task, and it'll be freed later
1638 * at exit time.
1639 *
1640 * This can block, so callers must hold no locks.
1641 */
1642static inline int get_undo_list(struct sem_undo_list **undo_listp)
1643{
1644 struct sem_undo_list *undo_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645
1646 undo_list = current->sysvsem.undo_list;
1647 if (!undo_list) {
Matt Helsley2453a302006-10-02 02:18:25 -07001648 undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649 if (undo_list == NULL)
1650 return -ENOMEM;
Ingo Molnar00a5dfd2005-08-05 23:05:27 +02001651 spin_lock_init(&undo_list->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652 atomic_set(&undo_list->refcnt, 1);
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001653 INIT_LIST_HEAD(&undo_list->list_proc);
1654
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655 current->sysvsem.undo_list = undo_list;
1656 }
1657 *undo_listp = undo_list;
1658 return 0;
1659}
1660
Nick Pigginbf17bb72009-12-15 16:47:28 -08001661static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662{
Nick Pigginbf17bb72009-12-15 16:47:28 -08001663 struct sem_undo *un;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664
Nick Pigginbf17bb72009-12-15 16:47:28 -08001665 list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) {
1666 if (un->semid == semid)
1667 return un;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668 }
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001669 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670}
1671
Nick Pigginbf17bb72009-12-15 16:47:28 -08001672static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1673{
1674 struct sem_undo *un;
1675
Manfred Spraul239521f2014-01-27 17:07:04 -08001676 assert_spin_locked(&ulp->lock);
Nick Pigginbf17bb72009-12-15 16:47:28 -08001677
1678 un = __lookup_undo(ulp, semid);
1679 if (un) {
1680 list_del_rcu(&un->list_proc);
1681 list_add_rcu(&un->list_proc, &ulp->list_proc);
1682 }
1683 return un;
1684}
1685
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001686/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -08001687 * find_alloc_undo - lookup (and if not present create) undo array
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001688 * @ns: namespace
1689 * @semid: semaphore array id
1690 *
1691 * The function looks up (and if not present creates) the undo structure.
1692 * The size of the undo structure depends on the size of the semaphore
1693 * array, thus the alloc path is not that straightforward.
Manfred Spraul380af1b2008-07-25 01:48:06 -07001694 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1695 * performs a rcu_read_lock().
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001696 */
1697static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698{
1699 struct sem_array *sma;
1700 struct sem_undo_list *ulp;
1701 struct sem_undo *un, *new;
Rik van Riel6062a8d2013-04-30 19:15:44 -07001702 int nsems, error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703
1704 error = get_undo_list(&ulp);
1705 if (error)
1706 return ERR_PTR(error);
1707
Manfred Spraul380af1b2008-07-25 01:48:06 -07001708 rcu_read_lock();
Pierre Peifferc530c6a2007-10-18 23:40:55 -07001709 spin_lock(&ulp->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 un = lookup_undo(ulp, semid);
Pierre Peifferc530c6a2007-10-18 23:40:55 -07001711 spin_unlock(&ulp->lock);
Manfred Spraul239521f2014-01-27 17:07:04 -08001712 if (likely(un != NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713 goto out;
1714
1715 /* no undo structure around - allocate one. */
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001716 /* step 1: figure out the size of the semaphore array */
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001717 sma = sem_obtain_object_check(ns, semid);
1718 if (IS_ERR(sma)) {
1719 rcu_read_unlock();
Julia Lawall4de85cd2010-05-26 14:43:44 -07001720 return ERR_CAST(sma);
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001721 }
Nadia Derbey023a5352007-10-18 23:40:51 -07001722
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723 nsems = sma->sem_nsems;
Rik van Riel6062a8d2013-04-30 19:15:44 -07001724 if (!ipc_rcu_getref(sma)) {
1725 rcu_read_unlock();
1726 un = ERR_PTR(-EIDRM);
1727 goto out;
1728 }
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001729 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001731 /* step 2: allocate new undo structure */
Burman Yan4668edc2006-12-06 20:38:51 -08001732 new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733 if (!new) {
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -07001734 ipc_rcu_putref(sma, ipc_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735 return ERR_PTR(-ENOMEM);
1736 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737
Manfred Spraul380af1b2008-07-25 01:48:06 -07001738 /* step 3: Acquire the lock on semaphore array */
Linus Torvalds4091fd942013-05-04 10:13:40 -07001739 rcu_read_lock();
Pierre Peiffer6ff37972008-04-29 01:00:46 -07001740 sem_lock_and_putref(sma);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001741 if (!ipc_valid_object(&sma->sem_perm)) {
Rik van Riel6062a8d2013-04-30 19:15:44 -07001742 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001743 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744 kfree(new);
1745 un = ERR_PTR(-EIDRM);
1746 goto out;
1747 }
Manfred Spraul380af1b2008-07-25 01:48:06 -07001748 spin_lock(&ulp->lock);
1749
1750 /*
1751 * step 4: check for races: did someone else allocate the undo struct?
1752 */
1753 un = lookup_undo(ulp, semid);
1754 if (un) {
1755 kfree(new);
1756 goto success;
1757 }
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001758 /* step 5: initialize & link new undo structure */
1759 new->semadj = (short *) &new[1];
Manfred Spraul380af1b2008-07-25 01:48:06 -07001760 new->ulp = ulp;
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001761 new->semid = semid;
1762 assert_spin_locked(&ulp->lock);
Manfred Spraul380af1b2008-07-25 01:48:06 -07001763 list_add_rcu(&new->list_proc, &ulp->list_proc);
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -07001764 ipc_assert_locked_object(&sma->sem_perm);
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001765 list_add(&new->list_id, &sma->list_id);
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001766 un = new;
Manfred Spraul380af1b2008-07-25 01:48:06 -07001767
1768success:
1769 spin_unlock(&ulp->lock);
Rik van Riel6062a8d2013-04-30 19:15:44 -07001770 sem_unlock(sma, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771out:
1772 return un;
1773}
1774
Manfred Spraulc61284e2010-07-20 13:24:23 -07001775
1776/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -08001777 * get_queue_result - retrieve the result code from sem_queue
Manfred Spraulc61284e2010-07-20 13:24:23 -07001778 * @q: Pointer to queue structure
1779 *
1780 * Retrieve the return code from the pending queue. If IN_WAKEUP is found in
1781 * q->status, then we must loop until the value is replaced with the final
1782 * value: This may happen if a task is woken up by an unrelated event (e.g.
1783 * signal) and in parallel the task is woken up by another task because it got
1784 * the requested semaphores.
1785 *
1786 * The function can be called with or without holding the semaphore spinlock.
1787 */
1788static int get_queue_result(struct sem_queue *q)
1789{
1790 int error;
1791
1792 error = q->status;
1793 while (unlikely(error == IN_WAKEUP)) {
1794 cpu_relax();
1795 error = q->status;
1796 }
1797
1798 return error;
1799}
1800
Heiko Carstensd5460c92009-01-14 14:14:27 +01001801SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
1802 unsigned, nsops, const struct timespec __user *, timeout)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803{
1804 int error = -EINVAL;
1805 struct sem_array *sma;
1806 struct sembuf fast_sops[SEMOPM_FAST];
Manfred Spraul239521f2014-01-27 17:07:04 -08001807 struct sembuf *sops = fast_sops, *sop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808 struct sem_undo *un;
Rik van Riel6062a8d2013-04-30 19:15:44 -07001809 int undos = 0, alter = 0, max, locknum;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810 struct sem_queue queue;
1811 unsigned long jiffies_left = 0;
Kirill Korotaeve3893532006-10-02 02:18:22 -07001812 struct ipc_namespace *ns;
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001813 struct list_head tasks;
Kirill Korotaeve3893532006-10-02 02:18:22 -07001814
1815 ns = current->nsproxy->ipc_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816
1817 if (nsops < 1 || semid < 0)
1818 return -EINVAL;
Kirill Korotaeve3893532006-10-02 02:18:22 -07001819 if (nsops > ns->sc_semopm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820 return -E2BIG;
Manfred Spraul239521f2014-01-27 17:07:04 -08001821 if (nsops > SEMOPM_FAST) {
1822 sops = kmalloc(sizeof(*sops)*nsops, GFP_KERNEL);
1823 if (sops == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824 return -ENOMEM;
1825 }
Manfred Spraul239521f2014-01-27 17:07:04 -08001826 if (copy_from_user(sops, tsops, nsops * sizeof(*tsops))) {
1827 error = -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828 goto out_free;
1829 }
1830 if (timeout) {
1831 struct timespec _timeout;
1832 if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1833 error = -EFAULT;
1834 goto out_free;
1835 }
1836 if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1837 _timeout.tv_nsec >= 1000000000L) {
1838 error = -EINVAL;
1839 goto out_free;
1840 }
1841 jiffies_left = timespec_to_jiffies(&_timeout);
1842 }
1843 max = 0;
1844 for (sop = sops; sop < sops + nsops; sop++) {
1845 if (sop->sem_num >= max)
1846 max = sop->sem_num;
1847 if (sop->sem_flg & SEM_UNDO)
Manfred Spraulb78755a2005-06-23 00:10:06 -07001848 undos = 1;
1849 if (sop->sem_op != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850 alter = 1;
1851 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852
Rik van Riel6062a8d2013-04-30 19:15:44 -07001853 INIT_LIST_HEAD(&tasks);
1854
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855 if (undos) {
Rik van Riel6062a8d2013-04-30 19:15:44 -07001856 /* On success, find_alloc_undo takes the rcu_read_lock */
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001857 un = find_alloc_undo(ns, semid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858 if (IS_ERR(un)) {
1859 error = PTR_ERR(un);
1860 goto out_free;
1861 }
Rik van Riel6062a8d2013-04-30 19:15:44 -07001862 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863 un = NULL;
Rik van Riel6062a8d2013-04-30 19:15:44 -07001864 rcu_read_lock();
1865 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001867 sma = sem_obtain_object_check(ns, semid);
Nadia Derbey023a5352007-10-18 23:40:51 -07001868 if (IS_ERR(sma)) {
Rik van Riel6062a8d2013-04-30 19:15:44 -07001869 rcu_read_unlock();
Nadia Derbey023a5352007-10-18 23:40:51 -07001870 error = PTR_ERR(sma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871 goto out_free;
Nadia Derbey023a5352007-10-18 23:40:51 -07001872 }
1873
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001874 error = -EFBIG;
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001875 if (max >= sma->sem_nsems)
1876 goto out_rcu_wakeup;
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001877
1878 error = -EACCES;
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001879 if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO))
1880 goto out_rcu_wakeup;
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001881
1882 error = security_sem_semop(sma, sops, nsops, alter);
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001883 if (error)
1884 goto out_rcu_wakeup;
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001885
Manfred Spraul6e224f92013-10-16 13:46:45 -07001886 error = -EIDRM;
1887 locknum = sem_lock(sma, sops, nsops);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001888 /*
1889 * We eventually might perform the following check in a lockless
1890 * fashion, considering ipc_valid_object() locking constraints.
1891 * If nsops == 1 and there is no contention for sem_perm.lock, then
1892 * only a per-semaphore lock is held and it's OK to proceed with the
1893 * check below. More details on the fine grained locking scheme
1894 * entangled here and why it's RMID race safe on comments at sem_lock()
1895 */
1896 if (!ipc_valid_object(&sma->sem_perm))
Manfred Spraul6e224f92013-10-16 13:46:45 -07001897 goto out_unlock_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898 /*
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001899 * semid identifiers are not unique - find_alloc_undo may have
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900 * allocated an undo structure, it was invalidated by an RMID
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001901 * and now a new array with received the same id. Check and fail.
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001902 * This case can be detected checking un->semid. The existence of
Manfred Spraul380af1b2008-07-25 01:48:06 -07001903 * "un" itself is guaranteed by rcu.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904 */
Rik van Riel6062a8d2013-04-30 19:15:44 -07001905 if (un && un->semid == -1)
1906 goto out_unlock_free;
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001907
Manfred Sprauld198cd62014-06-06 14:37:49 -07001908 queue.sops = sops;
1909 queue.nsops = nsops;
1910 queue.undo = un;
1911 queue.pid = task_tgid_vnr(current);
1912 queue.alter = alter;
1913
1914 error = perform_atomic_semop(sma, &queue);
Manfred Spraul0e8c6652013-09-30 13:45:25 -07001915 if (error == 0) {
1916 /* If the operation was successful, then do
1917 * the required updates.
1918 */
1919 if (alter)
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001920 do_smart_update(sma, sops, nsops, 1, &tasks);
Manfred Spraul0e8c6652013-09-30 13:45:25 -07001921 else
1922 set_semotime(sma, sops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923 }
Manfred Spraul0e8c6652013-09-30 13:45:25 -07001924 if (error <= 0)
1925 goto out_unlock_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926
1927 /* We need to sleep on this operation, so we put the current
1928 * task into the pending queue and go to sleep.
1929 */
Paul McQuade46c0a8c2014-06-06 14:37:37 -07001930
Manfred Spraulb97e8202009-12-15 16:47:32 -08001931 if (nsops == 1) {
1932 struct sem *curr;
1933 curr = &sma->sem_base[sops->sem_num];
1934
Manfred Spraulf269f402013-07-08 16:01:24 -07001935 if (alter) {
1936 if (sma->complex_count) {
1937 list_add_tail(&queue.list,
1938 &sma->pending_alter);
1939 } else {
1940
1941 list_add_tail(&queue.list,
1942 &curr->pending_alter);
1943 }
1944 } else {
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001945 list_add_tail(&queue.list, &curr->pending_const);
Manfred Spraulf269f402013-07-08 16:01:24 -07001946 }
Manfred Spraulb97e8202009-12-15 16:47:32 -08001947 } else {
Manfred Spraulf269f402013-07-08 16:01:24 -07001948 if (!sma->complex_count)
1949 merge_queues(sma);
1950
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001951 if (alter)
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001952 list_add_tail(&queue.list, &sma->pending_alter);
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001953 else
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001954 list_add_tail(&queue.list, &sma->pending_const);
1955
Manfred Spraulb97e8202009-12-15 16:47:32 -08001956 sma->complex_count++;
1957 }
1958
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959 queue.status = -EINTR;
1960 queue.sleeper = current;
Manfred Spraul0b0577f2011-11-02 13:38:52 -07001961
1962sleep_again:
Davidlohr Bueso52644c92015-02-17 13:47:55 -08001963 __set_current_state(TASK_INTERRUPTIBLE);
Rik van Riel6062a8d2013-04-30 19:15:44 -07001964 sem_unlock(sma, locknum);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001965 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966
1967 if (timeout)
1968 jiffies_left = schedule_timeout(jiffies_left);
1969 else
1970 schedule();
1971
Manfred Spraulc61284e2010-07-20 13:24:23 -07001972 error = get_queue_result(&queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973
1974 if (error != -EINTR) {
1975 /* fast path: update_queue already obtained all requested
Manfred Spraulc61284e2010-07-20 13:24:23 -07001976 * resources.
1977 * Perform a smp_mb(): User space could assume that semop()
1978 * is a memory barrier: Without the mb(), the cpu could
1979 * speculatively read in user space stale data that was
1980 * overwritten by the previous owner of the semaphore.
1981 */
1982 smp_mb();
1983
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984 goto out_free;
1985 }
1986
Linus Torvalds321310c2013-05-04 10:47:57 -07001987 rcu_read_lock();
Rik van Riel6062a8d2013-04-30 19:15:44 -07001988 sma = sem_obtain_lock(ns, semid, sops, nsops, &locknum);
Manfred Sprauld694ad62011-07-25 17:11:47 -07001989
1990 /*
1991 * Wait until it's guaranteed that no wakeup_sem_queue_do() is ongoing.
1992 */
1993 error = get_queue_result(&queue);
1994
1995 /*
1996 * Array removed? If yes, leave without sem_unlock().
1997 */
Nadia Derbey023a5352007-10-18 23:40:51 -07001998 if (IS_ERR(sma)) {
Linus Torvalds321310c2013-05-04 10:47:57 -07001999 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000 goto out_free;
2001 }
2002
Manfred Spraulc61284e2010-07-20 13:24:23 -07002003
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004 /*
Manfred Sprauld694ad62011-07-25 17:11:47 -07002005 * If queue.status != -EINTR we are woken up by another process.
2006 * Leave without unlink_queue(), but with sem_unlock().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007 */
Davidlohr Bueso3ab08fe2014-01-27 17:07:06 -08002008 if (error != -EINTR)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009 goto out_unlock_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010
2011 /*
2012 * If an interrupt occurred we have to clean up the queue
2013 */
2014 if (timeout && jiffies_left == 0)
2015 error = -EAGAIN;
Manfred Spraul0b0577f2011-11-02 13:38:52 -07002016
2017 /*
2018 * If the wakeup was spurious, just retry
2019 */
2020 if (error == -EINTR && !signal_pending(current))
2021 goto sleep_again;
2022
Manfred Spraulb97e8202009-12-15 16:47:32 -08002023 unlink_queue(sma, &queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002024
2025out_unlock_free:
Rik van Riel6062a8d2013-04-30 19:15:44 -07002026 sem_unlock(sma, locknum);
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07002027out_rcu_wakeup:
Linus Torvalds6d49dab2013-05-03 15:04:40 -07002028 rcu_read_unlock();
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07002029 wake_up_sem_queue_do(&tasks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030out_free:
Manfred Spraul239521f2014-01-27 17:07:04 -08002031 if (sops != fast_sops)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032 kfree(sops);
2033 return error;
2034}
2035
Heiko Carstensd5460c92009-01-14 14:14:27 +01002036SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
2037 unsigned, nsops)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038{
2039 return sys_semtimedop(semid, tsops, nsops, NULL);
2040}
2041
2042/* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
2043 * parent and child tasks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044 */
2045
2046int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
2047{
2048 struct sem_undo_list *undo_list;
2049 int error;
2050
2051 if (clone_flags & CLONE_SYSVSEM) {
2052 error = get_undo_list(&undo_list);
2053 if (error)
2054 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055 atomic_inc(&undo_list->refcnt);
2056 tsk->sysvsem.undo_list = undo_list;
Paul McQuade46c0a8c2014-06-06 14:37:37 -07002057 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058 tsk->sysvsem.undo_list = NULL;
2059
2060 return 0;
2061}
2062
2063/*
2064 * add semadj values to semaphores, free undo structures.
2065 * undo structures are not freed when semaphore arrays are destroyed
2066 * so some of them may be out of date.
2067 * IMPLEMENTATION NOTE: There is some confusion over whether the
2068 * set of adjustments that needs to be done should be done in an atomic
2069 * manner or not. That is, if we are attempting to decrement the semval
2070 * should we queue up and wait until we can do so legally?
2071 * The original implementation attempted to do this (queue and wait).
2072 * The current implementation does not do so. The POSIX standard
2073 * and SVID should be consulted to determine what behavior is mandated.
2074 */
2075void exit_sem(struct task_struct *tsk)
2076{
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002077 struct sem_undo_list *ulp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002079 ulp = tsk->sysvsem.undo_list;
2080 if (!ulp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081 return;
Manfred Spraul9edff4a2008-04-29 01:00:57 -07002082 tsk->sysvsem.undo_list = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002084 if (!atomic_dec_and_test(&ulp->refcnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085 return;
2086
Manfred Spraul380af1b2008-07-25 01:48:06 -07002087 for (;;) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088 struct sem_array *sma;
Manfred Spraul380af1b2008-07-25 01:48:06 -07002089 struct sem_undo *un;
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07002090 struct list_head tasks;
Rik van Riel6062a8d2013-04-30 19:15:44 -07002091 int semid, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092
Manfred Spraul380af1b2008-07-25 01:48:06 -07002093 rcu_read_lock();
Jiri Pirko05725f72009-04-14 20:17:16 +02002094 un = list_entry_rcu(ulp->list_proc.next,
2095 struct sem_undo, list_proc);
Herton R. Krzesinski602b8592015-08-14 15:35:02 -07002096 if (&un->list_proc == &ulp->list_proc) {
2097 /*
2098 * We must wait for freeary() before freeing this ulp,
2099 * in case we raced with last sem_undo. There is a small
2100 * possibility where we exit while freeary() didn't
2101 * finish unlocking sem_undo_list.
2102 */
2103 spin_unlock_wait(&ulp->lock);
Rik van Riel6062a8d2013-04-30 19:15:44 -07002104 rcu_read_unlock();
Manfred Spraul380af1b2008-07-25 01:48:06 -07002105 break;
Rik van Riel6062a8d2013-04-30 19:15:44 -07002106 }
Herton R. Krzesinski602b8592015-08-14 15:35:02 -07002107 spin_lock(&ulp->lock);
2108 semid = un->semid;
2109 spin_unlock(&ulp->lock);
Manfred Spraul380af1b2008-07-25 01:48:06 -07002110
Herton R. Krzesinski602b8592015-08-14 15:35:02 -07002111 /* exit_sem raced with IPC_RMID, nothing to do */
2112 if (semid == -1) {
2113 rcu_read_unlock();
2114 continue;
2115 }
2116
2117 sma = sem_obtain_object_check(tsk->nsproxy->ipc_ns, semid);
Manfred Spraul380af1b2008-07-25 01:48:06 -07002118 /* exit_sem raced with IPC_RMID, nothing to do */
Rik van Riel6062a8d2013-04-30 19:15:44 -07002119 if (IS_ERR(sma)) {
2120 rcu_read_unlock();
Manfred Spraul380af1b2008-07-25 01:48:06 -07002121 continue;
Rik van Riel6062a8d2013-04-30 19:15:44 -07002122 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123
Rik van Riel6062a8d2013-04-30 19:15:44 -07002124 sem_lock(sma, NULL, -1);
Manfred Spraul6e224f92013-10-16 13:46:45 -07002125 /* exit_sem raced with IPC_RMID, nothing to do */
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08002126 if (!ipc_valid_object(&sma->sem_perm)) {
Manfred Spraul6e224f92013-10-16 13:46:45 -07002127 sem_unlock(sma, -1);
2128 rcu_read_unlock();
2129 continue;
2130 }
Nick Pigginbf17bb72009-12-15 16:47:28 -08002131 un = __lookup_undo(ulp, semid);
Manfred Spraul380af1b2008-07-25 01:48:06 -07002132 if (un == NULL) {
2133 /* exit_sem raced with IPC_RMID+semget() that created
2134 * exactly the same semid. Nothing to do.
2135 */
Rik van Riel6062a8d2013-04-30 19:15:44 -07002136 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07002137 rcu_read_unlock();
Manfred Spraul380af1b2008-07-25 01:48:06 -07002138 continue;
2139 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140
Manfred Spraul380af1b2008-07-25 01:48:06 -07002141 /* remove un from the linked lists */
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -07002142 ipc_assert_locked_object(&sma->sem_perm);
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002143 list_del(&un->list_id);
2144
Herton R. Krzesinskia9795582015-08-14 15:35:05 -07002145 /* we are the last process using this ulp, acquiring ulp->lock
2146 * isn't required. Besides that, we are also protected against
2147 * IPC_RMID as we hold sma->sem_perm lock now
2148 */
Manfred Spraul380af1b2008-07-25 01:48:06 -07002149 list_del_rcu(&un->list_proc);
Manfred Spraul380af1b2008-07-25 01:48:06 -07002150
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002151 /* perform adjustments registered in un */
2152 for (i = 0; i < sma->sem_nsems; i++) {
Manfred Spraul239521f2014-01-27 17:07:04 -08002153 struct sem *semaphore = &sma->sem_base[i];
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002154 if (un->semadj[i]) {
2155 semaphore->semval += un->semadj[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156 /*
2157 * Range checks of the new semaphore value,
2158 * not defined by sus:
2159 * - Some unices ignore the undo entirely
2160 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
2161 * - some cap the value (e.g. FreeBSD caps
2162 * at 0, but doesn't enforce SEMVMX)
2163 *
2164 * Linux caps the semaphore value, both at 0
2165 * and at SEMVMX.
2166 *
Manfred Spraul239521f2014-01-27 17:07:04 -08002167 * Manfred <manfred@colorfullife.com>
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168 */
Ingo Molnar5f921ae2006-03-26 01:37:17 -08002169 if (semaphore->semval < 0)
2170 semaphore->semval = 0;
2171 if (semaphore->semval > SEMVMX)
2172 semaphore->semval = SEMVMX;
Pavel Emelyanovb4888932007-10-18 23:40:14 -07002173 semaphore->sempid = task_tgid_vnr(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174 }
2175 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176 /* maybe some queued-up processes were waiting for this */
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07002177 INIT_LIST_HEAD(&tasks);
2178 do_smart_update(sma, NULL, 0, 1, &tasks);
Rik van Riel6062a8d2013-04-30 19:15:44 -07002179 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07002180 rcu_read_unlock();
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07002181 wake_up_sem_queue_do(&tasks);
Manfred Spraul380af1b2008-07-25 01:48:06 -07002182
Lai Jiangshan693a8b62011-03-18 12:09:35 +08002183 kfree_rcu(un, rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002184 }
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002185 kfree(ulp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186}
2187
2188#ifdef CONFIG_PROC_FS
Mike Waychison19b49462005-09-06 15:17:10 -07002189static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002190{
Eric W. Biederman1efdb692012-02-07 16:54:11 -08002191 struct user_namespace *user_ns = seq_user_ns(s);
Mike Waychison19b49462005-09-06 15:17:10 -07002192 struct sem_array *sma = it;
Manfred Sprauld12e1e52013-07-08 16:01:25 -07002193 time_t sem_otime;
2194
Manfred Sprauld8c63372013-09-30 13:45:07 -07002195 /*
2196 * The proc interface isn't aware of sem_lock(), it calls
2197 * ipc_lock_object() directly (in sysvipc_find_ipc).
2198 * In order to stay compatible with sem_lock(), we must wait until
2199 * all simple semop() calls have left their critical regions.
2200 */
2201 sem_wait_array(sma);
2202
Manfred Sprauld12e1e52013-07-08 16:01:25 -07002203 sem_otime = get_semotime(sma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204
Joe Perches7f032d62015-04-15 16:17:54 -07002205 seq_printf(s,
2206 "%10d %10d %4o %10u %5u %5u %5u %5u %10lu %10lu\n",
2207 sma->sem_perm.key,
2208 sma->sem_perm.id,
2209 sma->sem_perm.mode,
2210 sma->sem_nsems,
2211 from_kuid_munged(user_ns, sma->sem_perm.uid),
2212 from_kgid_munged(user_ns, sma->sem_perm.gid),
2213 from_kuid_munged(user_ns, sma->sem_perm.cuid),
2214 from_kgid_munged(user_ns, sma->sem_perm.cgid),
2215 sem_otime,
2216 sma->sem_ctime);
2217
2218 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219}
2220#endif