blob: 10b94bc59d4a5ffaa85a89152ac02a1f806db054 [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:
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700165 * a) global sem_lock() for read/write
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166 * sem_undo.id_next,
Manfred Spraul758a6ba32013-07-08 16:01:26 -0700167 * sem_array.complex_count,
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700168 * sem_array.complex_mode
169 * sem_array.pending{_alter,_const},
170 * sem_array.sem_undo
Paul McQuade46c0a8c2014-06-06 14:37:37 -0700171 *
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700172 * b) global or semaphore sem_lock() for read/write:
Manfred Spraul758a6ba32013-07-08 16:01:26 -0700173 * sem_array.sem_base[i].pending_{const,alter}:
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700174 * sem_array.complex_mode (for read)
175 *
176 * c) special:
177 * sem_undo_list.list_proc:
178 * * undo_list->lock for write
179 * * rcu for read
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 */
181
Kirill Korotaeve3893532006-10-02 02:18:22 -0700182#define sc_semmsl sem_ctls[0]
183#define sc_semmns sem_ctls[1]
184#define sc_semopm sem_ctls[2]
185#define sc_semmni sem_ctls[3]
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186
Pierre Peiffered2ddbf2008-02-08 04:18:57 -0800187void sem_init_ns(struct ipc_namespace *ns)
Kirill Korotaeve3893532006-10-02 02:18:22 -0700188{
Kirill Korotaeve3893532006-10-02 02:18:22 -0700189 ns->sc_semmsl = SEMMSL;
190 ns->sc_semmns = SEMMNS;
191 ns->sc_semopm = SEMOPM;
192 ns->sc_semmni = SEMMNI;
193 ns->used_sems = 0;
Pierre Peiffered2ddbf2008-02-08 04:18:57 -0800194 ipc_init_ids(&ns->ids[IPC_SEM_IDS]);
Kirill Korotaeve3893532006-10-02 02:18:22 -0700195}
196
Pavel Emelyanovae5e1b22008-02-08 04:18:22 -0800197#ifdef CONFIG_IPC_NS
Kirill Korotaeve3893532006-10-02 02:18:22 -0700198void sem_exit_ns(struct ipc_namespace *ns)
199{
Pierre Peiffer01b8b072008-02-08 04:18:57 -0800200 free_ipcs(ns, &sem_ids(ns), freeary);
Serge E. Hallyn7d6feeb2009-12-15 16:47:27 -0800201 idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr);
Kirill Korotaeve3893532006-10-02 02:18:22 -0700202}
Pavel Emelyanovae5e1b22008-02-08 04:18:22 -0800203#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204
Manfred Spraul239521f2014-01-27 17:07:04 -0800205void __init sem_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206{
Pierre Peiffered2ddbf2008-02-08 04:18:57 -0800207 sem_init_ns(&init_ipc_ns);
Mike Waychison19b49462005-09-06 15:17:10 -0700208 ipc_init_proc_interface("sysvipc/sem",
209 " key semid perms nsems uid gid cuid cgid otime ctime\n",
Kirill Korotaeve3893532006-10-02 02:18:22 -0700210 IPC_SEM_IDS, sysvipc_sem_proc_show);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211}
212
Manfred Spraulf269f402013-07-08 16:01:24 -0700213/**
214 * unmerge_queues - unmerge queues, if possible.
215 * @sma: semaphore array
216 *
217 * The function unmerges the wait queues if complex_count is 0.
218 * It must be called prior to dropping the global semaphore array lock.
219 */
220static void unmerge_queues(struct sem_array *sma)
221{
222 struct sem_queue *q, *tq;
223
224 /* complex operations still around? */
225 if (sma->complex_count)
226 return;
227 /*
228 * We will switch back to simple mode.
229 * Move all pending operation back into the per-semaphore
230 * queues.
231 */
232 list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
233 struct sem *curr;
234 curr = &sma->sem_base[q->sops[0].sem_num];
235
236 list_add_tail(&q->list, &curr->pending_alter);
237 }
238 INIT_LIST_HEAD(&sma->pending_alter);
239}
240
241/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800242 * merge_queues - merge single semop queues into global queue
Manfred Spraulf269f402013-07-08 16:01:24 -0700243 * @sma: semaphore array
244 *
245 * This function merges all per-semaphore queues into the global queue.
246 * It is necessary to achieve FIFO ordering for the pending single-sop
247 * operations when a multi-semop operation must sleep.
248 * Only the alter operations must be moved, the const operations can stay.
249 */
250static void merge_queues(struct sem_array *sma)
251{
252 int i;
253 for (i = 0; i < sma->sem_nsems; i++) {
254 struct sem *sem = sma->sem_base + i;
255
256 list_splice_init(&sem->pending_alter, &sma->pending_alter);
257 }
258}
259
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -0700260static void sem_rcu_free(struct rcu_head *head)
261{
262 struct ipc_rcu *p = container_of(head, struct ipc_rcu, rcu);
263 struct sem_array *sma = ipc_rcu_to_struct(p);
264
265 security_sem_free(sma);
266 ipc_rcu_free(head);
267}
268
Nadia Derbey3e148c72007-10-18 23:40:54 -0700269/*
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700270 * Enter the mode suitable for non-simple operations:
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700271 * Caller must own sem_perm.lock.
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700272 */
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700273static void complexmode_enter(struct sem_array *sma)
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700274{
275 int i;
276 struct sem *sem;
277
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700278 if (sma->complex_mode) {
279 /* We are already in complex_mode. Nothing to do */
Manfred Spraul6d07b682013-09-30 13:45:06 -0700280 return;
281 }
282
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700283 /* We need a full barrier after seting complex_mode:
284 * The write to complex_mode must be visible
285 * before we read the first sem->lock spinlock state.
286 */
287 smp_store_mb(sma->complex_mode, true);
288
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700289 for (i = 0; i < sma->sem_nsems; i++) {
290 sem = sma->sem_base + i;
291 spin_unlock_wait(&sem->lock);
292 }
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700293 /*
294 * spin_unlock_wait() is not a memory barriers, it is only a
295 * control barrier. The code must pair with spin_unlock(&sem->lock),
296 * thus just the control barrier is insufficient.
297 *
298 * smp_rmb() is sufficient, as writes cannot pass the control barrier.
299 */
300 smp_rmb();
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700301}
302
303/*
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700304 * Try to leave the mode that disallows simple operations:
305 * Caller must own sem_perm.lock.
306 */
307static void complexmode_tryleave(struct sem_array *sma)
308{
309 if (sma->complex_count) {
310 /* Complex ops are sleeping.
311 * We must stay in complex mode
312 */
313 return;
314 }
315 /*
316 * Immediately after setting complex_mode to false,
317 * a simple op can start. Thus: all memory writes
318 * performed by the current operation must be visible
319 * before we set complex_mode to false.
320 */
321 smp_store_release(&sma->complex_mode, false);
322}
323
324#define SEM_GLOBAL_LOCK (-1)
325/*
Rik van Riel6062a8d2013-04-30 19:15:44 -0700326 * If the request contains only one semaphore operation, and there are
327 * no complex transactions pending, lock only the semaphore involved.
328 * Otherwise, lock the entire semaphore array, since we either have
329 * multiple semaphores in our own semops, or we need to look at
330 * semaphores from other pending complex operations.
Rik van Riel6062a8d2013-04-30 19:15:44 -0700331 */
332static inline int sem_lock(struct sem_array *sma, struct sembuf *sops,
333 int nsops)
334{
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700335 struct sem *sem;
Rik van Riel6062a8d2013-04-30 19:15:44 -0700336
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700337 if (nsops != 1) {
338 /* Complex operation - acquire a full lock */
339 ipc_lock_object(&sma->sem_perm);
340
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700341 /* Prevent parallel simple ops */
342 complexmode_enter(sma);
343 return SEM_GLOBAL_LOCK;
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700344 }
345
346 /*
347 * Only one semaphore affected - try to optimize locking.
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700348 * Optimized locking is possible if no complex operation
349 * is either enqueued or processed right now.
350 *
351 * Both facts are tracked by complex_mode.
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700352 */
353 sem = sma->sem_base + sops->sem_num;
354
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700355 /*
356 * Initial check for complex_mode. Just an optimization,
357 * no locking, no memory barrier.
358 */
359 if (!sma->complex_mode) {
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700360 /*
361 * It appears that no complex operation is around.
362 * Acquire the per-semaphore lock.
363 */
Rik van Riel6062a8d2013-04-30 19:15:44 -0700364 spin_lock(&sem->lock);
365
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700366 /*
367 * See 51d7d5205d33
368 * ("powerpc: Add smp_mb() to arch_spin_is_locked()"):
369 * A full barrier is required: the write of sem->lock
370 * must be visible before the read is executed
371 */
372 smp_mb();
Rik van Riel6062a8d2013-04-30 19:15:44 -0700373
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700374 if (!smp_load_acquire(&sma->complex_mode)) {
375 /* fast path successful! */
376 return sops->sem_num;
Rik van Riel6062a8d2013-04-30 19:15:44 -0700377 }
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700378 spin_unlock(&sem->lock);
Rik van Riel6062a8d2013-04-30 19:15:44 -0700379 }
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700380
381 /* slow path: acquire the full lock */
382 ipc_lock_object(&sma->sem_perm);
383
384 if (sma->complex_count == 0) {
385 /* False alarm:
386 * There is no complex operation, thus we can switch
387 * back to the fast path.
388 */
389 spin_lock(&sem->lock);
390 ipc_unlock_object(&sma->sem_perm);
391 return sops->sem_num;
392 } else {
393 /* Not a false alarm, thus complete the sequence for a
394 * full lock.
395 */
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700396 complexmode_enter(sma);
397 return SEM_GLOBAL_LOCK;
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700398 }
Rik van Riel6062a8d2013-04-30 19:15:44 -0700399}
400
401static inline void sem_unlock(struct sem_array *sma, int locknum)
402{
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700403 if (locknum == SEM_GLOBAL_LOCK) {
Manfred Spraulf269f402013-07-08 16:01:24 -0700404 unmerge_queues(sma);
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700405 complexmode_tryleave(sma);
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -0700406 ipc_unlock_object(&sma->sem_perm);
Rik van Riel6062a8d2013-04-30 19:15:44 -0700407 } else {
408 struct sem *sem = sma->sem_base + locknum;
409 spin_unlock(&sem->lock);
410 }
Rik van Riel6062a8d2013-04-30 19:15:44 -0700411}
412
413/*
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700414 * sem_lock_(check_) routines are called in the paths where the rwsem
Nadia Derbey3e148c72007-10-18 23:40:54 -0700415 * is not held.
Linus Torvalds321310c2013-05-04 10:47:57 -0700416 *
417 * The caller holds the RCU read lock.
Nadia Derbey3e148c72007-10-18 23:40:54 -0700418 */
Rik van Riel6062a8d2013-04-30 19:15:44 -0700419static inline struct sem_array *sem_obtain_lock(struct ipc_namespace *ns,
420 int id, struct sembuf *sops, int nsops, int *locknum)
Nadia Derbey023a5352007-10-18 23:40:51 -0700421{
Rik van Rielc460b662013-04-30 19:15:35 -0700422 struct kern_ipc_perm *ipcp;
423 struct sem_array *sma;
Nadia Derbey03f02c72007-10-18 23:40:51 -0700424
Davidlohr Bueso55b7ae52015-06-30 14:58:42 -0700425 ipcp = ipc_obtain_object_idr(&sem_ids(ns), id);
Linus Torvalds321310c2013-05-04 10:47:57 -0700426 if (IS_ERR(ipcp))
427 return ERR_CAST(ipcp);
Pierre Peifferb1ed88b2008-02-06 01:36:23 -0800428
Rik van Riel6062a8d2013-04-30 19:15:44 -0700429 sma = container_of(ipcp, struct sem_array, sem_perm);
430 *locknum = sem_lock(sma, sops, nsops);
Rik van Rielc460b662013-04-30 19:15:35 -0700431
432 /* ipc_rmid() may have already freed the ID while sem_lock
433 * was spinning: verify that the structure is still valid
434 */
Rafael Aquini72a8ff22014-01-27 17:07:02 -0800435 if (ipc_valid_object(ipcp))
Rik van Rielc460b662013-04-30 19:15:35 -0700436 return container_of(ipcp, struct sem_array, sem_perm);
437
Rik van Riel6062a8d2013-04-30 19:15:44 -0700438 sem_unlock(sma, *locknum);
Linus Torvalds321310c2013-05-04 10:47:57 -0700439 return ERR_PTR(-EINVAL);
Nadia Derbey023a5352007-10-18 23:40:51 -0700440}
441
Davidlohr Bueso16df3672013-04-30 19:15:29 -0700442static inline struct sem_array *sem_obtain_object(struct ipc_namespace *ns, int id)
443{
Davidlohr Bueso55b7ae52015-06-30 14:58:42 -0700444 struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&sem_ids(ns), id);
Davidlohr Bueso16df3672013-04-30 19:15:29 -0700445
446 if (IS_ERR(ipcp))
447 return ERR_CAST(ipcp);
448
449 return container_of(ipcp, struct sem_array, sem_perm);
450}
451
Davidlohr Bueso16df3672013-04-30 19:15:29 -0700452static inline struct sem_array *sem_obtain_object_check(struct ipc_namespace *ns,
453 int id)
454{
455 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&sem_ids(ns), id);
456
457 if (IS_ERR(ipcp))
458 return ERR_CAST(ipcp);
Pierre Peifferb1ed88b2008-02-06 01:36:23 -0800459
Nadia Derbey03f02c72007-10-18 23:40:51 -0700460 return container_of(ipcp, struct sem_array, sem_perm);
Nadia Derbey023a5352007-10-18 23:40:51 -0700461}
462
Pierre Peiffer6ff37972008-04-29 01:00:46 -0700463static inline void sem_lock_and_putref(struct sem_array *sma)
464{
Rik van Riel6062a8d2013-04-30 19:15:44 -0700465 sem_lock(sma, NULL, -1);
Fabian Frederick9b24fef2016-08-02 14:03:07 -0700466 ipc_rcu_putref(sma, sem_rcu_free);
Pierre Peiffer6ff37972008-04-29 01:00:46 -0700467}
468
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700469static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
470{
471 ipc_rmid(&sem_ids(ns), &s->sem_perm);
472}
473
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474/*
475 * Lockless wakeup algorithm:
476 * Without the check/retry algorithm a lockless wakeup is possible:
477 * - queue.status is initialized to -EINTR before blocking.
478 * - wakeup is performed by
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700479 * * unlinking the queue entry from the pending list
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 * * setting queue.status to IN_WAKEUP
481 * This is the notification for the blocked thread that a
482 * result value is imminent.
483 * * call wake_up_process
484 * * set queue.status to the final value.
485 * - the previously blocked thread checks queue.status:
Manfred Spraul239521f2014-01-27 17:07:04 -0800486 * * if it's IN_WAKEUP, then it must wait until the value changes
487 * * if it's not -EINTR, then the operation was completed by
488 * update_queue. semtimedop can return queue.status without
489 * performing any operation on the sem array.
490 * * otherwise it must acquire the spinlock and check what's up.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 *
492 * The two-stage algorithm is necessary to protect against the following
493 * races:
494 * - if queue.status is set after wake_up_process, then the woken up idle
495 * thread could race forward and try (and fail) to acquire sma->lock
496 * before update_queue had a chance to set queue.status
497 * - if queue.status is written before wake_up_process and if the
498 * blocked process is woken up by a signal between writing
499 * queue.status and the wake_up_process, then the woken up
500 * process could return from semtimedop and die by calling
501 * sys_exit before wake_up_process is called. Then wake_up_process
502 * will oops, because the task structure is already invalid.
503 * (yes, this happened on s390 with sysv msg).
504 *
505 */
506#define IN_WAKEUP 1
507
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700508/**
509 * newary - Create a new semaphore set
510 * @ns: namespace
511 * @params: ptr to the structure that contains key, semflg and nsems
512 *
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700513 * Called with sem_ids.rwsem held (as a writer)
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700514 */
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700515static int newary(struct ipc_namespace *ns, struct ipc_params *params)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516{
517 int id;
518 int retval;
519 struct sem_array *sma;
520 int size;
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700521 key_t key = params->key;
522 int nsems = params->u.nsems;
523 int semflg = params->flg;
Manfred Spraulb97e8202009-12-15 16:47:32 -0800524 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525
526 if (!nsems)
527 return -EINVAL;
Kirill Korotaeve3893532006-10-02 02:18:22 -0700528 if (ns->used_sems + nsems > ns->sc_semmns)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 return -ENOSPC;
530
Manfred Spraul239521f2014-01-27 17:07:04 -0800531 size = sizeof(*sma) + nsems * sizeof(struct sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 sma = ipc_rcu_alloc(size);
Davidlohr Bueso3ab08fe2014-01-27 17:07:06 -0800533 if (!sma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 return -ENOMEM;
Davidlohr Bueso3ab08fe2014-01-27 17:07:06 -0800535
Manfred Spraul239521f2014-01-27 17:07:04 -0800536 memset(sma, 0, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537
538 sma->sem_perm.mode = (semflg & S_IRWXUGO);
539 sma->sem_perm.key = key;
540
541 sma->sem_perm.security = NULL;
542 retval = security_sem_alloc(sma);
543 if (retval) {
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -0700544 ipc_rcu_putref(sma, ipc_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545 return retval;
546 }
547
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 sma->sem_base = (struct sem *) &sma[1];
Manfred Spraulb97e8202009-12-15 16:47:32 -0800549
Rik van Riel6062a8d2013-04-30 19:15:44 -0700550 for (i = 0; i < nsems; i++) {
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700551 INIT_LIST_HEAD(&sma->sem_base[i].pending_alter);
552 INIT_LIST_HEAD(&sma->sem_base[i].pending_const);
Rik van Riel6062a8d2013-04-30 19:15:44 -0700553 spin_lock_init(&sma->sem_base[i].lock);
554 }
Manfred Spraulb97e8202009-12-15 16:47:32 -0800555
556 sma->complex_count = 0;
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700557 sma->complex_mode = true; /* dropped by sem_unlock below */
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700558 INIT_LIST_HEAD(&sma->pending_alter);
559 INIT_LIST_HEAD(&sma->pending_const);
Manfred Spraul4daa28f2008-07-25 01:48:04 -0700560 INIT_LIST_HEAD(&sma->list_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 sma->sem_nsems = nsems;
562 sma->sem_ctime = get_seconds();
Manfred Spraule8577d12014-12-02 15:59:34 -0800563
564 id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
565 if (id < 0) {
566 ipc_rcu_putref(sma, sem_rcu_free);
567 return id;
568 }
569 ns->used_sems += nsems;
570
Rik van Riel6062a8d2013-04-30 19:15:44 -0700571 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -0700572 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700574 return sma->sem_perm.id;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575}
576
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700577
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700578/*
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700579 * Called with sem_ids.rwsem and ipcp locked.
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700580 */
Nadia Derbey03f02c72007-10-18 23:40:51 -0700581static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582{
Nadia Derbey03f02c72007-10-18 23:40:51 -0700583 struct sem_array *sma;
584
585 sma = container_of(ipcp, struct sem_array, sem_perm);
586 return security_sem_associate(sma, semflg);
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700587}
588
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700589/*
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700590 * Called with sem_ids.rwsem and ipcp locked.
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700591 */
Nadia Derbey03f02c72007-10-18 23:40:51 -0700592static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
593 struct ipc_params *params)
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700594{
Nadia Derbey03f02c72007-10-18 23:40:51 -0700595 struct sem_array *sma;
596
597 sma = container_of(ipcp, struct sem_array, sem_perm);
598 if (params->u.nsems > sma->sem_nsems)
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700599 return -EINVAL;
600
601 return 0;
602}
603
Heiko Carstensd5460c92009-01-14 14:14:27 +0100604SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg)
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700605{
Kirill Korotaeve3893532006-10-02 02:18:22 -0700606 struct ipc_namespace *ns;
Mathias Krauseeb66ec42014-06-06 14:37:36 -0700607 static const struct ipc_ops sem_ops = {
608 .getnew = newary,
609 .associate = sem_security,
610 .more_checks = sem_more_checks,
611 };
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700612 struct ipc_params sem_params;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613
Kirill Korotaeve3893532006-10-02 02:18:22 -0700614 ns = current->nsproxy->ipc_ns;
615
616 if (nsems < 0 || nsems > ns->sc_semmsl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 return -EINVAL;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700618
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700619 sem_params.key = key;
620 sem_params.flg = semflg;
621 sem_params.u.nsems = nsems;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700622
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700623 return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624}
625
Petr Mladek78f50092014-01-27 17:07:00 -0800626/**
627 * perform_atomic_semop - Perform (if possible) a semaphore operation
Manfred Spraul758a6ba32013-07-08 16:01:26 -0700628 * @sma: semaphore array
Manfred Sprauld198cd62014-06-06 14:37:49 -0700629 * @q: struct sem_queue that describes the operation
Manfred Spraul758a6ba32013-07-08 16:01:26 -0700630 *
631 * Returns 0 if the operation was possible.
632 * Returns 1 if the operation is impossible, the caller must sleep.
633 * Negative values are error codes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 */
Manfred Sprauld198cd62014-06-06 14:37:49 -0700635static int perform_atomic_semop(struct sem_array *sma, struct sem_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636{
Manfred Sprauld198cd62014-06-06 14:37:49 -0700637 int result, sem_op, nsops, pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 struct sembuf *sop;
Manfred Spraul239521f2014-01-27 17:07:04 -0800639 struct sem *curr;
Manfred Sprauld198cd62014-06-06 14:37:49 -0700640 struct sembuf *sops;
641 struct sem_undo *un;
642
643 sops = q->sops;
644 nsops = q->nsops;
645 un = q->undo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646
647 for (sop = sops; sop < sops + nsops; sop++) {
648 curr = sma->sem_base + sop->sem_num;
649 sem_op = sop->sem_op;
650 result = curr->semval;
Petr Mladek78f50092014-01-27 17:07:00 -0800651
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 if (!sem_op && result)
653 goto would_block;
654
655 result += sem_op;
656 if (result < 0)
657 goto would_block;
658 if (result > SEMVMX)
659 goto out_of_range;
Petr Mladek78f50092014-01-27 17:07:00 -0800660
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 if (sop->sem_flg & SEM_UNDO) {
662 int undo = un->semadj[sop->sem_num] - sem_op;
Petr Mladek78f50092014-01-27 17:07:00 -0800663 /* Exceeding the undo range is an error. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
665 goto out_of_range;
Petr Mladek78f50092014-01-27 17:07:00 -0800666 un->semadj[sop->sem_num] = undo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 }
Petr Mladek78f50092014-01-27 17:07:00 -0800668
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 curr->semval = result;
670 }
671
672 sop--;
Manfred Sprauld198cd62014-06-06 14:37:49 -0700673 pid = q->pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 while (sop >= sops) {
675 sma->sem_base[sop->sem_num].sempid = pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 sop--;
677 }
Petr Mladek78f50092014-01-27 17:07:00 -0800678
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 return 0;
680
681out_of_range:
682 result = -ERANGE;
683 goto undo;
684
685would_block:
Manfred Sprauled247b72014-06-06 14:37:49 -0700686 q->blocking = sop;
687
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 if (sop->sem_flg & IPC_NOWAIT)
689 result = -EAGAIN;
690 else
691 result = 1;
692
693undo:
694 sop--;
695 while (sop >= sops) {
Petr Mladek78f50092014-01-27 17:07:00 -0800696 sem_op = sop->sem_op;
697 sma->sem_base[sop->sem_num].semval -= sem_op;
698 if (sop->sem_flg & SEM_UNDO)
699 un->semadj[sop->sem_num] += sem_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 sop--;
701 }
702
703 return result;
704}
705
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700706/** wake_up_sem_queue_prepare(q, error): Prepare wake-up
707 * @q: queue entry that must be signaled
708 * @error: Error value for the signal
709 *
710 * Prepare the wake-up of the queue entry q.
Nick Piggind4212092009-12-15 16:47:30 -0800711 */
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700712static void wake_up_sem_queue_prepare(struct list_head *pt,
713 struct sem_queue *q, int error)
Nick Piggind4212092009-12-15 16:47:30 -0800714{
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700715 if (list_empty(pt)) {
716 /*
717 * Hold preempt off so that we don't get preempted and have the
718 * wakee busy-wait until we're scheduled back on.
719 */
720 preempt_disable();
721 }
Nick Piggind4212092009-12-15 16:47:30 -0800722 q->status = IN_WAKEUP;
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700723 q->pid = error;
724
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700725 list_add_tail(&q->list, pt);
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700726}
727
728/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800729 * wake_up_sem_queue_do - do the actual wake-up
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700730 * @pt: list of tasks to be woken up
731 *
732 * Do the actual wake-up.
733 * The function is called without any locks held, thus the semaphore array
734 * could be destroyed already and the tasks can disappear as soon as the
735 * status is set to the actual return code.
736 */
737static void wake_up_sem_queue_do(struct list_head *pt)
738{
739 struct sem_queue *q, *t;
740 int did_something;
741
742 did_something = !list_empty(pt);
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700743 list_for_each_entry_safe(q, t, pt, list) {
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700744 wake_up_process(q->sleeper);
745 /* q can disappear immediately after writing q->status. */
746 smp_wmb();
747 q->status = q->pid;
748 }
749 if (did_something)
750 preempt_enable();
Nick Piggind4212092009-12-15 16:47:30 -0800751}
752
Manfred Spraulb97e8202009-12-15 16:47:32 -0800753static void unlink_queue(struct sem_array *sma, struct sem_queue *q)
754{
755 list_del(&q->list);
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700756 if (q->nsops > 1)
Manfred Spraulb97e8202009-12-15 16:47:32 -0800757 sma->complex_count--;
758}
759
Manfred Spraulfd5db422010-05-26 14:43:40 -0700760/** check_restart(sma, q)
761 * @sma: semaphore array
762 * @q: the operation that just completed
763 *
764 * update_queue is O(N^2) when it restarts scanning the whole queue of
765 * waiting operations. Therefore this function checks if the restart is
766 * really necessary. It is called after a previously waiting operation
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700767 * modified the array.
768 * Note that wait-for-zero operations are handled without restart.
Manfred Spraulfd5db422010-05-26 14:43:40 -0700769 */
770static int check_restart(struct sem_array *sma, struct sem_queue *q)
771{
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700772 /* pending complex alter operations are too difficult to analyse */
773 if (!list_empty(&sma->pending_alter))
Manfred Spraulfd5db422010-05-26 14:43:40 -0700774 return 1;
775
776 /* we were a sleeping complex operation. Too difficult */
777 if (q->nsops > 1)
778 return 1;
779
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700780 /* It is impossible that someone waits for the new value:
781 * - complex operations always restart.
782 * - wait-for-zero are handled seperately.
783 * - q is a previously sleeping simple operation that
784 * altered the array. It must be a decrement, because
785 * simple increments never sleep.
786 * - If there are older (higher priority) decrements
787 * in the queue, then they have observed the original
788 * semval value and couldn't proceed. The operation
789 * decremented to value - thus they won't proceed either.
790 */
791 return 0;
792}
Manfred Spraulfd5db422010-05-26 14:43:40 -0700793
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700794/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800795 * wake_const_ops - wake up non-alter tasks
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700796 * @sma: semaphore array.
797 * @semnum: semaphore that was modified.
798 * @pt: list head for the tasks that must be woken up.
799 *
800 * wake_const_ops must be called after a semaphore in a semaphore array
801 * was set to 0. If complex const operations are pending, wake_const_ops must
802 * be called with semnum = -1, as well as with the number of each modified
803 * semaphore.
804 * The tasks that must be woken up are added to @pt. The return code
805 * is stored in q->pid.
806 * The function returns 1 if at least one operation was completed successfully.
807 */
808static int wake_const_ops(struct sem_array *sma, int semnum,
809 struct list_head *pt)
810{
811 struct sem_queue *q;
812 struct list_head *walk;
813 struct list_head *pending_list;
814 int semop_completed = 0;
Manfred Spraulfd5db422010-05-26 14:43:40 -0700815
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700816 if (semnum == -1)
817 pending_list = &sma->pending_const;
818 else
819 pending_list = &sma->sem_base[semnum].pending_const;
820
821 walk = pending_list->next;
822 while (walk != pending_list) {
823 int error;
824
825 q = container_of(walk, struct sem_queue, list);
826 walk = walk->next;
827
Manfred Sprauld198cd62014-06-06 14:37:49 -0700828 error = perform_atomic_semop(sma, q);
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700829
830 if (error <= 0) {
831 /* operation completed, remove from queue & wakeup */
832
833 unlink_queue(sma, q);
834
835 wake_up_sem_queue_prepare(pt, q, error);
836 if (error == 0)
837 semop_completed = 1;
838 }
839 }
840 return semop_completed;
841}
842
843/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800844 * do_smart_wakeup_zero - wakeup all wait for zero tasks
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700845 * @sma: semaphore array
846 * @sops: operations that were performed
847 * @nsops: number of operations
848 * @pt: list head of the tasks that must be woken up.
849 *
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800850 * Checks all required queue for wait-for-zero operations, based
851 * on the actual changes that were performed on the semaphore array.
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700852 * The function returns 1 if at least one operation was completed successfully.
853 */
854static int do_smart_wakeup_zero(struct sem_array *sma, struct sembuf *sops,
855 int nsops, struct list_head *pt)
856{
857 int i;
858 int semop_completed = 0;
859 int got_zero = 0;
860
861 /* first: the per-semaphore queues, if known */
862 if (sops) {
863 for (i = 0; i < nsops; i++) {
864 int num = sops[i].sem_num;
865
866 if (sma->sem_base[num].semval == 0) {
867 got_zero = 1;
868 semop_completed |= wake_const_ops(sma, num, pt);
869 }
870 }
871 } else {
872 /*
873 * No sops means modified semaphores not known.
874 * Assume all were changed.
Manfred Spraulfd5db422010-05-26 14:43:40 -0700875 */
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700876 for (i = 0; i < sma->sem_nsems; i++) {
877 if (sma->sem_base[i].semval == 0) {
878 got_zero = 1;
879 semop_completed |= wake_const_ops(sma, i, pt);
880 }
881 }
Manfred Spraulfd5db422010-05-26 14:43:40 -0700882 }
883 /*
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700884 * If one of the modified semaphores got 0,
885 * then check the global queue, too.
Manfred Spraulfd5db422010-05-26 14:43:40 -0700886 */
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700887 if (got_zero)
888 semop_completed |= wake_const_ops(sma, -1, pt);
Manfred Spraulfd5db422010-05-26 14:43:40 -0700889
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700890 return semop_completed;
Manfred Spraulfd5db422010-05-26 14:43:40 -0700891}
892
Manfred Spraul636c6be2009-12-15 16:47:33 -0800893
894/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800895 * update_queue - look for tasks that can be completed.
Manfred Spraul636c6be2009-12-15 16:47:33 -0800896 * @sma: semaphore array.
897 * @semnum: semaphore that was modified.
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700898 * @pt: list head for the tasks that must be woken up.
Manfred Spraul636c6be2009-12-15 16:47:33 -0800899 *
900 * update_queue must be called after a semaphore in a semaphore array
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700901 * was modified. If multiple semaphores were modified, update_queue must
902 * be called with semnum = -1, as well as with the number of each modified
903 * semaphore.
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700904 * The tasks that must be woken up are added to @pt. The return code
905 * is stored in q->pid.
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700906 * The function internally checks if const operations can now succeed.
907 *
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700908 * The function return 1 if at least one semop was completed successfully.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 */
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700910static int update_queue(struct sem_array *sma, int semnum, struct list_head *pt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911{
Manfred Spraul636c6be2009-12-15 16:47:33 -0800912 struct sem_queue *q;
913 struct list_head *walk;
914 struct list_head *pending_list;
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700915 int semop_completed = 0;
Manfred Spraul636c6be2009-12-15 16:47:33 -0800916
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700917 if (semnum == -1)
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700918 pending_list = &sma->pending_alter;
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700919 else
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700920 pending_list = &sma->sem_base[semnum].pending_alter;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921
Nick Piggin9cad2002009-12-15 16:47:29 -0800922again:
Manfred Spraul636c6be2009-12-15 16:47:33 -0800923 walk = pending_list->next;
924 while (walk != pending_list) {
Manfred Spraulfd5db422010-05-26 14:43:40 -0700925 int error, restart;
Manfred Spraul636c6be2009-12-15 16:47:33 -0800926
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700927 q = container_of(walk, struct sem_queue, list);
Manfred Spraul636c6be2009-12-15 16:47:33 -0800928 walk = walk->next;
Nick Piggin9cad2002009-12-15 16:47:29 -0800929
Manfred Sprauld987f8b2009-12-15 16:47:34 -0800930 /* If we are scanning the single sop, per-semaphore list of
931 * one semaphore and that semaphore is 0, then it is not
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700932 * necessary to scan further: simple increments
Manfred Sprauld987f8b2009-12-15 16:47:34 -0800933 * that affect only one entry succeed immediately and cannot
934 * be in the per semaphore pending queue, and decrements
935 * cannot be successful if the value is already 0.
936 */
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700937 if (semnum != -1 && sma->sem_base[semnum].semval == 0)
Manfred Sprauld987f8b2009-12-15 16:47:34 -0800938 break;
939
Manfred Sprauld198cd62014-06-06 14:37:49 -0700940 error = perform_atomic_semop(sma, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941
942 /* Does q->sleeper still need to sleep? */
Nick Piggin9cad2002009-12-15 16:47:29 -0800943 if (error > 0)
944 continue;
Manfred Spraula1193f82008-07-25 01:48:06 -0700945
Manfred Spraulb97e8202009-12-15 16:47:32 -0800946 unlink_queue(sma, q);
Manfred Spraula1193f82008-07-25 01:48:06 -0700947
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700948 if (error) {
Manfred Spraulfd5db422010-05-26 14:43:40 -0700949 restart = 0;
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700950 } else {
951 semop_completed = 1;
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700952 do_smart_wakeup_zero(sma, q->sops, q->nsops, pt);
Manfred Spraulfd5db422010-05-26 14:43:40 -0700953 restart = check_restart(sma, q);
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700954 }
Manfred Spraulfd5db422010-05-26 14:43:40 -0700955
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700956 wake_up_sem_queue_prepare(pt, q, error);
Manfred Spraulfd5db422010-05-26 14:43:40 -0700957 if (restart)
Nick Piggin9cad2002009-12-15 16:47:29 -0800958 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 }
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700960 return semop_completed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961}
962
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700963/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800964 * set_semotime - set sem_otime
Manfred Spraul0e8c6652013-09-30 13:45:25 -0700965 * @sma: semaphore array
966 * @sops: operations that modified the array, may be NULL
967 *
968 * sem_otime is replicated to avoid cache line trashing.
969 * This function sets one instance to the current time.
970 */
971static void set_semotime(struct sem_array *sma, struct sembuf *sops)
972{
973 if (sops == NULL) {
974 sma->sem_base[0].sem_otime = get_seconds();
975 } else {
976 sma->sem_base[sops[0].sem_num].sem_otime =
977 get_seconds();
978 }
979}
980
981/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800982 * do_smart_update - optimized update_queue
Manfred Spraulfd5db422010-05-26 14:43:40 -0700983 * @sma: semaphore array
984 * @sops: operations that were performed
985 * @nsops: number of operations
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700986 * @otime: force setting otime
987 * @pt: list head of the tasks that must be woken up.
Manfred Spraulfd5db422010-05-26 14:43:40 -0700988 *
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700989 * do_smart_update() does the required calls to update_queue and wakeup_zero,
990 * based on the actual changes that were performed on the semaphore array.
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700991 * Note that the function does not do the actual wake-up: the caller is
992 * responsible for calling wake_up_sem_queue_do(@pt).
993 * It is safe to perform this call after dropping all locks.
Manfred Spraulfd5db422010-05-26 14:43:40 -0700994 */
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700995static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops,
996 int otime, struct list_head *pt)
Manfred Spraulfd5db422010-05-26 14:43:40 -0700997{
998 int i;
999
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001000 otime |= do_smart_wakeup_zero(sma, sops, nsops, pt);
1001
Manfred Spraulf269f402013-07-08 16:01:24 -07001002 if (!list_empty(&sma->pending_alter)) {
1003 /* semaphore array uses the global queue - just process it. */
1004 otime |= update_queue(sma, -1, pt);
1005 } else {
1006 if (!sops) {
1007 /*
1008 * No sops, thus the modified semaphores are not
1009 * known. Check all.
1010 */
1011 for (i = 0; i < sma->sem_nsems; i++)
1012 otime |= update_queue(sma, i, pt);
1013 } else {
1014 /*
1015 * Check the semaphores that were increased:
1016 * - No complex ops, thus all sleeping ops are
1017 * decrease.
1018 * - if we decreased the value, then any sleeping
1019 * semaphore ops wont be able to run: If the
1020 * previous value was too small, then the new
1021 * value will be too small, too.
1022 */
1023 for (i = 0; i < nsops; i++) {
1024 if (sops[i].sem_op > 0) {
1025 otime |= update_queue(sma,
1026 sops[i].sem_num, pt);
1027 }
Manfred Spraulab465df2013-05-26 11:08:52 +02001028 }
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001029 }
Manfred Spraulfd5db422010-05-26 14:43:40 -07001030 }
Manfred Spraul0e8c6652013-09-30 13:45:25 -07001031 if (otime)
1032 set_semotime(sma, sops);
Manfred Spraulfd5db422010-05-26 14:43:40 -07001033}
1034
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001035/*
Manfred Spraulb220c572014-06-06 14:37:51 -07001036 * check_qop: Test if a queued operation sleeps on the semaphore semnum
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001037 */
1038static int check_qop(struct sem_array *sma, int semnum, struct sem_queue *q,
1039 bool count_zero)
1040{
Manfred Spraulb220c572014-06-06 14:37:51 -07001041 struct sembuf *sop = q->blocking;
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001042
Manfred Spraul9b44ee22014-06-06 14:37:52 -07001043 /*
1044 * Linux always (since 0.99.10) reported a task as sleeping on all
1045 * semaphores. This violates SUS, therefore it was changed to the
1046 * standard compliant behavior.
1047 * Give the administrators a chance to notice that an application
1048 * might misbehave because it relies on the Linux behavior.
1049 */
1050 pr_info_once("semctl(GETNCNT/GETZCNT) is since 3.16 Single Unix Specification compliant.\n"
1051 "The task %s (%d) triggered the difference, watch for misbehavior.\n",
1052 current->comm, task_pid_nr(current));
1053
Manfred Spraulb220c572014-06-06 14:37:51 -07001054 if (sop->sem_num != semnum)
1055 return 0;
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001056
Manfred Spraulb220c572014-06-06 14:37:51 -07001057 if (count_zero && sop->sem_op == 0)
1058 return 1;
1059 if (!count_zero && sop->sem_op < 0)
1060 return 1;
1061
1062 return 0;
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001063}
1064
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065/* The following counts are associated to each semaphore:
1066 * semncnt number of tasks waiting on semval being nonzero
1067 * semzcnt number of tasks waiting on semval being zero
Manfred Spraulb220c572014-06-06 14:37:51 -07001068 *
1069 * Per definition, a task waits only on the semaphore of the first semop
1070 * that cannot proceed, even if additional operation would block, too.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071 */
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001072static int count_semcnt(struct sem_array *sma, ushort semnum,
1073 bool count_zero)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074{
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001075 struct list_head *l;
Manfred Spraul239521f2014-01-27 17:07:04 -08001076 struct sem_queue *q;
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001077 int semcnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001079 semcnt = 0;
1080 /* First: check the simple operations. They are easy to evaluate */
1081 if (count_zero)
1082 l = &sma->sem_base[semnum].pending_const;
1083 else
1084 l = &sma->sem_base[semnum].pending_alter;
1085
1086 list_for_each_entry(q, l, list) {
1087 /* all task on a per-semaphore list sleep on exactly
1088 * that semaphore
1089 */
1090 semcnt++;
Rik van Rielde2657f2013-05-09 16:59:59 -04001091 }
1092
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001093 /* Then: check the complex operations. */
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001094 list_for_each_entry(q, &sma->pending_alter, list) {
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001095 semcnt += check_qop(sma, semnum, q, count_zero);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 }
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001097 if (count_zero) {
1098 list_for_each_entry(q, &sma->pending_const, list) {
1099 semcnt += check_qop(sma, semnum, q, count_zero);
1100 }
Rik van Rielebc2e5e2013-05-09 16:53:28 -04001101 }
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001102 return semcnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103}
1104
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001105/* Free a semaphore set. freeary() is called with sem_ids.rwsem locked
1106 * as a writer and the spinlock for this semaphore set hold. sem_ids.rwsem
Nadia Derbey3e148c72007-10-18 23:40:54 -07001107 * remains locked on exit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 */
Pierre Peiffer01b8b072008-02-08 04:18:57 -08001109static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110{
Manfred Spraul380af1b2008-07-25 01:48:06 -07001111 struct sem_undo *un, *tu;
1112 struct sem_queue *q, *tq;
Pierre Peiffer01b8b072008-02-08 04:18:57 -08001113 struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001114 struct list_head tasks;
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001115 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116
Manfred Spraul380af1b2008-07-25 01:48:06 -07001117 /* Free the existing undo structures for this semaphore set. */
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -07001118 ipc_assert_locked_object(&sma->sem_perm);
Manfred Spraul380af1b2008-07-25 01:48:06 -07001119 list_for_each_entry_safe(un, tu, &sma->list_id, list_id) {
1120 list_del(&un->list_id);
1121 spin_lock(&un->ulp->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 un->semid = -1;
Manfred Spraul380af1b2008-07-25 01:48:06 -07001123 list_del_rcu(&un->list_proc);
1124 spin_unlock(&un->ulp->lock);
Lai Jiangshan693a8b62011-03-18 12:09:35 +08001125 kfree_rcu(un, rcu);
Manfred Spraul380af1b2008-07-25 01:48:06 -07001126 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127
1128 /* Wake up all pending processes and let them fail with EIDRM. */
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001129 INIT_LIST_HEAD(&tasks);
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001130 list_for_each_entry_safe(q, tq, &sma->pending_const, list) {
1131 unlink_queue(sma, q);
1132 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1133 }
1134
1135 list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
Manfred Spraulb97e8202009-12-15 16:47:32 -08001136 unlink_queue(sma, q);
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001137 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138 }
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001139 for (i = 0; i < sma->sem_nsems; i++) {
1140 struct sem *sem = sma->sem_base + i;
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001141 list_for_each_entry_safe(q, tq, &sem->pending_const, list) {
1142 unlink_queue(sma, q);
1143 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1144 }
1145 list_for_each_entry_safe(q, tq, &sem->pending_alter, list) {
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001146 unlink_queue(sma, q);
1147 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1148 }
1149 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150
Nadia Derbey7ca7e562007-10-18 23:40:48 -07001151 /* Remove the semaphore set from the IDR */
1152 sem_rmid(ns, sma);
Rik van Riel6062a8d2013-04-30 19:15:44 -07001153 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001154 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001156 wake_up_sem_queue_do(&tasks);
Kirill Korotaeve3893532006-10-02 02:18:22 -07001157 ns->used_sems -= sma->sem_nsems;
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -07001158 ipc_rcu_putref(sma, sem_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159}
1160
1161static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
1162{
Manfred Spraul239521f2014-01-27 17:07:04 -08001163 switch (version) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 case IPC_64:
1165 return copy_to_user(buf, in, sizeof(*in));
1166 case IPC_OLD:
1167 {
1168 struct semid_ds out;
1169
Dan Rosenberg982f7c22010-09-30 15:15:31 -07001170 memset(&out, 0, sizeof(out));
1171
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
1173
1174 out.sem_otime = in->sem_otime;
1175 out.sem_ctime = in->sem_ctime;
1176 out.sem_nsems = in->sem_nsems;
1177
1178 return copy_to_user(buf, &out, sizeof(out));
1179 }
1180 default:
1181 return -EINVAL;
1182 }
1183}
1184
Manfred Sprauld12e1e52013-07-08 16:01:25 -07001185static time_t get_semotime(struct sem_array *sma)
1186{
1187 int i;
1188 time_t res;
1189
1190 res = sma->sem_base[0].sem_otime;
1191 for (i = 1; i < sma->sem_nsems; i++) {
1192 time_t to = sma->sem_base[i].sem_otime;
1193
1194 if (to > res)
1195 res = to;
1196 }
1197 return res;
1198}
1199
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -08001200static int semctl_nolock(struct ipc_namespace *ns, int semid,
Al Viroe1fd1f42013-03-05 15:04:55 -05001201 int cmd, int version, void __user *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202{
Amerigo Wange5cc9c72009-12-15 16:47:35 -08001203 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 struct sem_array *sma;
1205
Manfred Spraul239521f2014-01-27 17:07:04 -08001206 switch (cmd) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 case IPC_INFO:
1208 case SEM_INFO:
1209 {
1210 struct seminfo seminfo;
1211 int max_id;
1212
1213 err = security_sem_semctl(NULL, cmd);
1214 if (err)
1215 return err;
Paul McQuade46c0a8c2014-06-06 14:37:37 -07001216
Manfred Spraul239521f2014-01-27 17:07:04 -08001217 memset(&seminfo, 0, sizeof(seminfo));
Kirill Korotaeve3893532006-10-02 02:18:22 -07001218 seminfo.semmni = ns->sc_semmni;
1219 seminfo.semmns = ns->sc_semmns;
1220 seminfo.semmsl = ns->sc_semmsl;
1221 seminfo.semopm = ns->sc_semopm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 seminfo.semvmx = SEMVMX;
1223 seminfo.semmnu = SEMMNU;
1224 seminfo.semmap = SEMMAP;
1225 seminfo.semume = SEMUME;
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001226 down_read(&sem_ids(ns).rwsem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 if (cmd == SEM_INFO) {
Kirill Korotaeve3893532006-10-02 02:18:22 -07001228 seminfo.semusz = sem_ids(ns).in_use;
1229 seminfo.semaem = ns->used_sems;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 } else {
1231 seminfo.semusz = SEMUSZ;
1232 seminfo.semaem = SEMAEM;
1233 }
Nadia Derbey7ca7e562007-10-18 23:40:48 -07001234 max_id = ipc_get_maxid(&sem_ids(ns));
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001235 up_read(&sem_ids(ns).rwsem);
Paul McQuade46c0a8c2014-06-06 14:37:37 -07001236 if (copy_to_user(p, &seminfo, sizeof(struct seminfo)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 return -EFAULT;
Manfred Spraul239521f2014-01-27 17:07:04 -08001238 return (max_id < 0) ? 0 : max_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 }
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -08001240 case IPC_STAT:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 case SEM_STAT:
1242 {
1243 struct semid64_ds tbuf;
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001244 int id = 0;
1245
1246 memset(&tbuf, 0, sizeof(tbuf));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247
Linus Torvalds941b0302013-05-04 11:04:29 -07001248 rcu_read_lock();
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -08001249 if (cmd == SEM_STAT) {
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001250 sma = sem_obtain_object(ns, semid);
1251 if (IS_ERR(sma)) {
1252 err = PTR_ERR(sma);
1253 goto out_unlock;
1254 }
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -08001255 id = sma->sem_perm.id;
1256 } else {
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001257 sma = sem_obtain_object_check(ns, semid);
1258 if (IS_ERR(sma)) {
1259 err = PTR_ERR(sma);
1260 goto out_unlock;
1261 }
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -08001262 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263
1264 err = -EACCES;
Serge E. Hallynb0e77592011-03-23 16:43:24 -07001265 if (ipcperms(ns, &sma->sem_perm, S_IRUGO))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266 goto out_unlock;
1267
1268 err = security_sem_semctl(sma, cmd);
1269 if (err)
1270 goto out_unlock;
1271
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
Manfred Sprauld12e1e52013-07-08 16:01:25 -07001273 tbuf.sem_otime = get_semotime(sma);
1274 tbuf.sem_ctime = sma->sem_ctime;
1275 tbuf.sem_nsems = sma->sem_nsems;
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001276 rcu_read_unlock();
Al Viroe1fd1f42013-03-05 15:04:55 -05001277 if (copy_semid_to_user(p, &tbuf, version))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 return -EFAULT;
1279 return id;
1280 }
1281 default:
1282 return -EINVAL;
1283 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284out_unlock:
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001285 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 return err;
1287}
1288
Al Viroe1fd1f42013-03-05 15:04:55 -05001289static int semctl_setval(struct ipc_namespace *ns, int semid, int semnum,
1290 unsigned long arg)
1291{
1292 struct sem_undo *un;
1293 struct sem_array *sma;
Manfred Spraul239521f2014-01-27 17:07:04 -08001294 struct sem *curr;
Al Viroe1fd1f42013-03-05 15:04:55 -05001295 int err;
Al Viroe1fd1f42013-03-05 15:04:55 -05001296 struct list_head tasks;
1297 int val;
1298#if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN)
1299 /* big-endian 64bit */
1300 val = arg >> 32;
1301#else
1302 /* 32bit or little-endian 64bit */
1303 val = arg;
1304#endif
1305
Rik van Riel6062a8d2013-04-30 19:15:44 -07001306 if (val > SEMVMX || val < 0)
1307 return -ERANGE;
Al Viroe1fd1f42013-03-05 15:04:55 -05001308
1309 INIT_LIST_HEAD(&tasks);
Al Viroe1fd1f42013-03-05 15:04:55 -05001310
Rik van Riel6062a8d2013-04-30 19:15:44 -07001311 rcu_read_lock();
1312 sma = sem_obtain_object_check(ns, semid);
1313 if (IS_ERR(sma)) {
1314 rcu_read_unlock();
1315 return PTR_ERR(sma);
1316 }
1317
1318 if (semnum < 0 || semnum >= sma->sem_nsems) {
1319 rcu_read_unlock();
1320 return -EINVAL;
1321 }
1322
1323
1324 if (ipcperms(ns, &sma->sem_perm, S_IWUGO)) {
1325 rcu_read_unlock();
1326 return -EACCES;
1327 }
Al Viroe1fd1f42013-03-05 15:04:55 -05001328
1329 err = security_sem_semctl(sma, SETVAL);
Rik van Riel6062a8d2013-04-30 19:15:44 -07001330 if (err) {
1331 rcu_read_unlock();
1332 return -EACCES;
1333 }
Al Viroe1fd1f42013-03-05 15:04:55 -05001334
Rik van Riel6062a8d2013-04-30 19:15:44 -07001335 sem_lock(sma, NULL, -1);
Al Viroe1fd1f42013-03-05 15:04:55 -05001336
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001337 if (!ipc_valid_object(&sma->sem_perm)) {
Manfred Spraul6e224f92013-10-16 13:46:45 -07001338 sem_unlock(sma, -1);
1339 rcu_read_unlock();
1340 return -EIDRM;
1341 }
1342
Al Viroe1fd1f42013-03-05 15:04:55 -05001343 curr = &sma->sem_base[semnum];
1344
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -07001345 ipc_assert_locked_object(&sma->sem_perm);
Al Viroe1fd1f42013-03-05 15:04:55 -05001346 list_for_each_entry(un, &sma->list_id, list_id)
1347 un->semadj[semnum] = 0;
1348
1349 curr->semval = val;
1350 curr->sempid = task_tgid_vnr(current);
1351 sma->sem_ctime = get_seconds();
1352 /* maybe some queued-up processes were waiting for this */
1353 do_smart_update(sma, NULL, 0, 0, &tasks);
Rik van Riel6062a8d2013-04-30 19:15:44 -07001354 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001355 rcu_read_unlock();
Al Viroe1fd1f42013-03-05 15:04:55 -05001356 wake_up_sem_queue_do(&tasks);
Rik van Riel6062a8d2013-04-30 19:15:44 -07001357 return 0;
Al Viroe1fd1f42013-03-05 15:04:55 -05001358}
1359
Kirill Korotaeve3893532006-10-02 02:18:22 -07001360static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
Al Viroe1fd1f42013-03-05 15:04:55 -05001361 int cmd, void __user *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362{
1363 struct sem_array *sma;
Manfred Spraul239521f2014-01-27 17:07:04 -08001364 struct sem *curr;
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001365 int err, nsems;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 ushort fast_sem_io[SEMMSL_FAST];
Manfred Spraul239521f2014-01-27 17:07:04 -08001367 ushort *sem_io = fast_sem_io;
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001368 struct list_head tasks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001370 INIT_LIST_HEAD(&tasks);
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001371
1372 rcu_read_lock();
1373 sma = sem_obtain_object_check(ns, semid);
1374 if (IS_ERR(sma)) {
1375 rcu_read_unlock();
1376 return PTR_ERR(sma);
1377 }
1378
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 nsems = sma->sem_nsems;
1380
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 err = -EACCES;
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001382 if (ipcperms(ns, &sma->sem_perm, cmd == SETALL ? S_IWUGO : S_IRUGO))
1383 goto out_rcu_wakeup;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384
1385 err = security_sem_semctl(sma, cmd);
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001386 if (err)
1387 goto out_rcu_wakeup;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388
1389 err = -EACCES;
1390 switch (cmd) {
1391 case GETALL:
1392 {
Al Viroe1fd1f42013-03-05 15:04:55 -05001393 ushort __user *array = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 int i;
1395
Al Viroce857222013-05-03 00:30:49 +01001396 sem_lock(sma, NULL, -1);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001397 if (!ipc_valid_object(&sma->sem_perm)) {
Manfred Spraul6e224f92013-10-16 13:46:45 -07001398 err = -EIDRM;
1399 goto out_unlock;
1400 }
Manfred Spraul239521f2014-01-27 17:07:04 -08001401 if (nsems > SEMMSL_FAST) {
Al Viroce857222013-05-03 00:30:49 +01001402 if (!ipc_rcu_getref(sma)) {
Al Viroce857222013-05-03 00:30:49 +01001403 err = -EIDRM;
Manfred Spraul6e224f92013-10-16 13:46:45 -07001404 goto out_unlock;
Al Viroce857222013-05-03 00:30:49 +01001405 }
1406 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001407 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 sem_io = ipc_alloc(sizeof(ushort)*nsems);
Manfred Spraul239521f2014-01-27 17:07:04 -08001409 if (sem_io == NULL) {
Fabian Frederick9b24fef2016-08-02 14:03:07 -07001410 ipc_rcu_putref(sma, sem_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 return -ENOMEM;
1412 }
1413
Linus Torvalds4091fd92013-05-04 10:13:40 -07001414 rcu_read_lock();
Pierre Peiffer6ff37972008-04-29 01:00:46 -07001415 sem_lock_and_putref(sma);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001416 if (!ipc_valid_object(&sma->sem_perm)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 err = -EIDRM;
Manfred Spraul6e224f92013-10-16 13:46:45 -07001418 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 }
Al Viroce857222013-05-03 00:30:49 +01001420 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 for (i = 0; i < sma->sem_nsems; i++)
1422 sem_io[i] = sma->sem_base[i].semval;
Rik van Riel6062a8d2013-04-30 19:15:44 -07001423 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001424 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425 err = 0;
Manfred Spraul239521f2014-01-27 17:07:04 -08001426 if (copy_to_user(array, sem_io, nsems*sizeof(ushort)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 err = -EFAULT;
1428 goto out_free;
1429 }
1430 case SETALL:
1431 {
1432 int i;
1433 struct sem_undo *un;
1434
Rik van Riel6062a8d2013-04-30 19:15:44 -07001435 if (!ipc_rcu_getref(sma)) {
Manfred Spraul6e224f92013-10-16 13:46:45 -07001436 err = -EIDRM;
1437 goto out_rcu_wakeup;
Rik van Riel6062a8d2013-04-30 19:15:44 -07001438 }
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001439 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440
Manfred Spraul239521f2014-01-27 17:07:04 -08001441 if (nsems > SEMMSL_FAST) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442 sem_io = ipc_alloc(sizeof(ushort)*nsems);
Manfred Spraul239521f2014-01-27 17:07:04 -08001443 if (sem_io == NULL) {
Fabian Frederick9b24fef2016-08-02 14:03:07 -07001444 ipc_rcu_putref(sma, sem_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445 return -ENOMEM;
1446 }
1447 }
1448
Manfred Spraul239521f2014-01-27 17:07:04 -08001449 if (copy_from_user(sem_io, p, nsems*sizeof(ushort))) {
Fabian Frederick9b24fef2016-08-02 14:03:07 -07001450 ipc_rcu_putref(sma, sem_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451 err = -EFAULT;
1452 goto out_free;
1453 }
1454
1455 for (i = 0; i < nsems; i++) {
1456 if (sem_io[i] > SEMVMX) {
Fabian Frederick9b24fef2016-08-02 14:03:07 -07001457 ipc_rcu_putref(sma, sem_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 err = -ERANGE;
1459 goto out_free;
1460 }
1461 }
Linus Torvalds4091fd92013-05-04 10:13:40 -07001462 rcu_read_lock();
Pierre Peiffer6ff37972008-04-29 01:00:46 -07001463 sem_lock_and_putref(sma);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001464 if (!ipc_valid_object(&sma->sem_perm)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 err = -EIDRM;
Manfred Spraul6e224f92013-10-16 13:46:45 -07001466 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 }
1468
Davidlohr Buesoa5f4db82016-03-22 14:27:48 -07001469 for (i = 0; i < nsems; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 sma->sem_base[i].semval = sem_io[i];
Davidlohr Buesoa5f4db82016-03-22 14:27:48 -07001471 sma->sem_base[i].sempid = task_tgid_vnr(current);
1472 }
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001473
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -07001474 ipc_assert_locked_object(&sma->sem_perm);
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001475 list_for_each_entry(un, &sma->list_id, list_id) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 for (i = 0; i < nsems; i++)
1477 un->semadj[i] = 0;
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001478 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 sma->sem_ctime = get_seconds();
1480 /* maybe some queued-up processes were waiting for this */
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001481 do_smart_update(sma, NULL, 0, 0, &tasks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 err = 0;
1483 goto out_unlock;
1484 }
Al Viroe1fd1f42013-03-05 15:04:55 -05001485 /* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 }
1487 err = -EINVAL;
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001488 if (semnum < 0 || semnum >= nsems)
1489 goto out_rcu_wakeup;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490
Rik van Riel6062a8d2013-04-30 19:15:44 -07001491 sem_lock(sma, NULL, -1);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001492 if (!ipc_valid_object(&sma->sem_perm)) {
Manfred Spraul6e224f92013-10-16 13:46:45 -07001493 err = -EIDRM;
1494 goto out_unlock;
1495 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496 curr = &sma->sem_base[semnum];
1497
1498 switch (cmd) {
1499 case GETVAL:
1500 err = curr->semval;
1501 goto out_unlock;
1502 case GETPID:
1503 err = curr->sempid;
1504 goto out_unlock;
1505 case GETNCNT:
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001506 err = count_semcnt(sma, semnum, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 goto out_unlock;
1508 case GETZCNT:
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001509 err = count_semcnt(sma, semnum, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 }
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001512
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513out_unlock:
Rik van Riel6062a8d2013-04-30 19:15:44 -07001514 sem_unlock(sma, -1);
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001515out_rcu_wakeup:
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001516 rcu_read_unlock();
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001517 wake_up_sem_queue_do(&tasks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518out_free:
Manfred Spraul239521f2014-01-27 17:07:04 -08001519 if (sem_io != fast_sem_io)
Tetsuo Handa1d5cfdb2016-01-22 15:11:02 -08001520 ipc_free(sem_io);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521 return err;
1522}
1523
Pierre Peiffer016d7132008-04-29 01:00:50 -07001524static inline unsigned long
1525copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526{
Manfred Spraul239521f2014-01-27 17:07:04 -08001527 switch (version) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 case IPC_64:
Pierre Peiffer016d7132008-04-29 01:00:50 -07001529 if (copy_from_user(out, buf, sizeof(*out)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 case IPC_OLD:
1533 {
1534 struct semid_ds tbuf_old;
1535
Manfred Spraul239521f2014-01-27 17:07:04 -08001536 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537 return -EFAULT;
1538
Pierre Peiffer016d7132008-04-29 01:00:50 -07001539 out->sem_perm.uid = tbuf_old.sem_perm.uid;
1540 out->sem_perm.gid = tbuf_old.sem_perm.gid;
1541 out->sem_perm.mode = tbuf_old.sem_perm.mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542
1543 return 0;
1544 }
1545 default:
1546 return -EINVAL;
1547 }
1548}
1549
Pierre Peiffer522bb2a2008-04-29 01:00:49 -07001550/*
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001551 * This function handles some semctl commands which require the rwsem
Pierre Peiffer522bb2a2008-04-29 01:00:49 -07001552 * to be held in write mode.
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001553 * NOTE: no locks must be held, the rwsem is taken inside this function.
Pierre Peiffer522bb2a2008-04-29 01:00:49 -07001554 */
Pierre Peiffer21a48262008-04-29 01:00:49 -07001555static int semctl_down(struct ipc_namespace *ns, int semid,
Al Viroe1fd1f42013-03-05 15:04:55 -05001556 int cmd, int version, void __user *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557{
1558 struct sem_array *sma;
1559 int err;
Pierre Peiffer016d7132008-04-29 01:00:50 -07001560 struct semid64_ds semid64;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 struct kern_ipc_perm *ipcp;
1562
Manfred Spraul239521f2014-01-27 17:07:04 -08001563 if (cmd == IPC_SET) {
Al Viroe1fd1f42013-03-05 15:04:55 -05001564 if (copy_semid_from_user(&semid64, p, version))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001568 down_write(&sem_ids(ns).rwsem);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001569 rcu_read_lock();
1570
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001571 ipcp = ipcctl_pre_down_nolock(ns, &sem_ids(ns), semid, cmd,
1572 &semid64.sem_perm, 0);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001573 if (IS_ERR(ipcp)) {
1574 err = PTR_ERR(ipcp);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001575 goto out_unlock1;
1576 }
Steve Grubb073115d2006-04-02 17:07:33 -04001577
Pierre Peiffera5f75e72008-04-29 01:00:54 -07001578 sma = container_of(ipcp, struct sem_array, sem_perm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579
1580 err = security_sem_semctl(sma, cmd);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001581 if (err)
1582 goto out_unlock1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001584 switch (cmd) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585 case IPC_RMID:
Rik van Riel6062a8d2013-04-30 19:15:44 -07001586 sem_lock(sma, NULL, -1);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001587 /* freeary unlocks the ipc object and rcu */
Pierre Peiffer01b8b072008-02-08 04:18:57 -08001588 freeary(ns, ipcp);
Pierre Peiffer522bb2a2008-04-29 01:00:49 -07001589 goto out_up;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 case IPC_SET:
Rik van Riel6062a8d2013-04-30 19:15:44 -07001591 sem_lock(sma, NULL, -1);
Eric W. Biederman1efdb692012-02-07 16:54:11 -08001592 err = ipc_update_perm(&semid64.sem_perm, ipcp);
1593 if (err)
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001594 goto out_unlock0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595 sma->sem_ctime = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 break;
1597 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 err = -EINVAL;
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001599 goto out_unlock1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001602out_unlock0:
Rik van Riel6062a8d2013-04-30 19:15:44 -07001603 sem_unlock(sma, -1);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001604out_unlock1:
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001605 rcu_read_unlock();
Pierre Peiffer522bb2a2008-04-29 01:00:49 -07001606out_up:
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001607 up_write(&sem_ids(ns).rwsem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608 return err;
1609}
1610
Al Viroe1fd1f42013-03-05 15:04:55 -05001611SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, unsigned long, arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613 int version;
Kirill Korotaeve3893532006-10-02 02:18:22 -07001614 struct ipc_namespace *ns;
Al Viroe1fd1f42013-03-05 15:04:55 -05001615 void __user *p = (void __user *)arg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616
1617 if (semid < 0)
1618 return -EINVAL;
1619
1620 version = ipc_parse_version(&cmd);
Kirill Korotaeve3893532006-10-02 02:18:22 -07001621 ns = current->nsproxy->ipc_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622
Manfred Spraul239521f2014-01-27 17:07:04 -08001623 switch (cmd) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624 case IPC_INFO:
1625 case SEM_INFO:
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -08001626 case IPC_STAT:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627 case SEM_STAT:
Al Viroe1fd1f42013-03-05 15:04:55 -05001628 return semctl_nolock(ns, semid, cmd, version, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629 case GETALL:
1630 case GETVAL:
1631 case GETPID:
1632 case GETNCNT:
1633 case GETZCNT:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634 case SETALL:
Al Viroe1fd1f42013-03-05 15:04:55 -05001635 return semctl_main(ns, semid, semnum, cmd, p);
1636 case SETVAL:
1637 return semctl_setval(ns, semid, semnum, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638 case IPC_RMID:
1639 case IPC_SET:
Al Viroe1fd1f42013-03-05 15:04:55 -05001640 return semctl_down(ns, semid, cmd, version, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641 default:
1642 return -EINVAL;
1643 }
1644}
1645
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646/* If the task doesn't already have a undo_list, then allocate one
1647 * here. We guarantee there is only one thread using this undo list,
1648 * and current is THE ONE
1649 *
1650 * If this allocation and assignment succeeds, but later
1651 * portions of this code fail, there is no need to free the sem_undo_list.
1652 * Just let it stay associated with the task, and it'll be freed later
1653 * at exit time.
1654 *
1655 * This can block, so callers must hold no locks.
1656 */
1657static inline int get_undo_list(struct sem_undo_list **undo_listp)
1658{
1659 struct sem_undo_list *undo_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660
1661 undo_list = current->sysvsem.undo_list;
1662 if (!undo_list) {
Matt Helsley2453a302006-10-02 02:18:25 -07001663 undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 if (undo_list == NULL)
1665 return -ENOMEM;
Ingo Molnar00a5dfd2005-08-05 23:05:27 +02001666 spin_lock_init(&undo_list->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 atomic_set(&undo_list->refcnt, 1);
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001668 INIT_LIST_HEAD(&undo_list->list_proc);
1669
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 current->sysvsem.undo_list = undo_list;
1671 }
1672 *undo_listp = undo_list;
1673 return 0;
1674}
1675
Nick Pigginbf17bb72009-12-15 16:47:28 -08001676static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677{
Nick Pigginbf17bb72009-12-15 16:47:28 -08001678 struct sem_undo *un;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679
Nick Pigginbf17bb72009-12-15 16:47:28 -08001680 list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) {
1681 if (un->semid == semid)
1682 return un;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683 }
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001684 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685}
1686
Nick Pigginbf17bb72009-12-15 16:47:28 -08001687static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1688{
1689 struct sem_undo *un;
1690
Manfred Spraul239521f2014-01-27 17:07:04 -08001691 assert_spin_locked(&ulp->lock);
Nick Pigginbf17bb72009-12-15 16:47:28 -08001692
1693 un = __lookup_undo(ulp, semid);
1694 if (un) {
1695 list_del_rcu(&un->list_proc);
1696 list_add_rcu(&un->list_proc, &ulp->list_proc);
1697 }
1698 return un;
1699}
1700
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001701/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -08001702 * find_alloc_undo - lookup (and if not present create) undo array
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001703 * @ns: namespace
1704 * @semid: semaphore array id
1705 *
1706 * The function looks up (and if not present creates) the undo structure.
1707 * The size of the undo structure depends on the size of the semaphore
1708 * array, thus the alloc path is not that straightforward.
Manfred Spraul380af1b2008-07-25 01:48:06 -07001709 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1710 * performs a rcu_read_lock().
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001711 */
1712static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713{
1714 struct sem_array *sma;
1715 struct sem_undo_list *ulp;
1716 struct sem_undo *un, *new;
Rik van Riel6062a8d2013-04-30 19:15:44 -07001717 int nsems, error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718
1719 error = get_undo_list(&ulp);
1720 if (error)
1721 return ERR_PTR(error);
1722
Manfred Spraul380af1b2008-07-25 01:48:06 -07001723 rcu_read_lock();
Pierre Peifferc530c6a2007-10-18 23:40:55 -07001724 spin_lock(&ulp->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 un = lookup_undo(ulp, semid);
Pierre Peifferc530c6a2007-10-18 23:40:55 -07001726 spin_unlock(&ulp->lock);
Manfred Spraul239521f2014-01-27 17:07:04 -08001727 if (likely(un != NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728 goto out;
1729
1730 /* no undo structure around - allocate one. */
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001731 /* step 1: figure out the size of the semaphore array */
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001732 sma = sem_obtain_object_check(ns, semid);
1733 if (IS_ERR(sma)) {
1734 rcu_read_unlock();
Julia Lawall4de85cd2010-05-26 14:43:44 -07001735 return ERR_CAST(sma);
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001736 }
Nadia Derbey023a5352007-10-18 23:40:51 -07001737
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738 nsems = sma->sem_nsems;
Rik van Riel6062a8d2013-04-30 19:15:44 -07001739 if (!ipc_rcu_getref(sma)) {
1740 rcu_read_unlock();
1741 un = ERR_PTR(-EIDRM);
1742 goto out;
1743 }
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001744 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001746 /* step 2: allocate new undo structure */
Burman Yan4668edc2006-12-06 20:38:51 -08001747 new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748 if (!new) {
Fabian Frederick9b24fef2016-08-02 14:03:07 -07001749 ipc_rcu_putref(sma, sem_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750 return ERR_PTR(-ENOMEM);
1751 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752
Manfred Spraul380af1b2008-07-25 01:48:06 -07001753 /* step 3: Acquire the lock on semaphore array */
Linus Torvalds4091fd92013-05-04 10:13:40 -07001754 rcu_read_lock();
Pierre Peiffer6ff37972008-04-29 01:00:46 -07001755 sem_lock_and_putref(sma);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001756 if (!ipc_valid_object(&sma->sem_perm)) {
Rik van Riel6062a8d2013-04-30 19:15:44 -07001757 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001758 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759 kfree(new);
1760 un = ERR_PTR(-EIDRM);
1761 goto out;
1762 }
Manfred Spraul380af1b2008-07-25 01:48:06 -07001763 spin_lock(&ulp->lock);
1764
1765 /*
1766 * step 4: check for races: did someone else allocate the undo struct?
1767 */
1768 un = lookup_undo(ulp, semid);
1769 if (un) {
1770 kfree(new);
1771 goto success;
1772 }
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001773 /* step 5: initialize & link new undo structure */
1774 new->semadj = (short *) &new[1];
Manfred Spraul380af1b2008-07-25 01:48:06 -07001775 new->ulp = ulp;
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001776 new->semid = semid;
1777 assert_spin_locked(&ulp->lock);
Manfred Spraul380af1b2008-07-25 01:48:06 -07001778 list_add_rcu(&new->list_proc, &ulp->list_proc);
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -07001779 ipc_assert_locked_object(&sma->sem_perm);
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001780 list_add(&new->list_id, &sma->list_id);
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001781 un = new;
Manfred Spraul380af1b2008-07-25 01:48:06 -07001782
1783success:
1784 spin_unlock(&ulp->lock);
Rik van Riel6062a8d2013-04-30 19:15:44 -07001785 sem_unlock(sma, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786out:
1787 return un;
1788}
1789
Manfred Spraulc61284e2010-07-20 13:24:23 -07001790
1791/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -08001792 * get_queue_result - retrieve the result code from sem_queue
Manfred Spraulc61284e2010-07-20 13:24:23 -07001793 * @q: Pointer to queue structure
1794 *
1795 * Retrieve the return code from the pending queue. If IN_WAKEUP is found in
1796 * q->status, then we must loop until the value is replaced with the final
1797 * value: This may happen if a task is woken up by an unrelated event (e.g.
1798 * signal) and in parallel the task is woken up by another task because it got
1799 * the requested semaphores.
1800 *
1801 * The function can be called with or without holding the semaphore spinlock.
1802 */
1803static int get_queue_result(struct sem_queue *q)
1804{
1805 int error;
1806
1807 error = q->status;
1808 while (unlikely(error == IN_WAKEUP)) {
1809 cpu_relax();
1810 error = q->status;
1811 }
1812
1813 return error;
1814}
1815
Heiko Carstensd5460c92009-01-14 14:14:27 +01001816SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
1817 unsigned, nsops, const struct timespec __user *, timeout)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818{
1819 int error = -EINVAL;
1820 struct sem_array *sma;
1821 struct sembuf fast_sops[SEMOPM_FAST];
Manfred Spraul239521f2014-01-27 17:07:04 -08001822 struct sembuf *sops = fast_sops, *sop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823 struct sem_undo *un;
Rik van Riel6062a8d2013-04-30 19:15:44 -07001824 int undos = 0, alter = 0, max, locknum;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825 struct sem_queue queue;
1826 unsigned long jiffies_left = 0;
Kirill Korotaeve3893532006-10-02 02:18:22 -07001827 struct ipc_namespace *ns;
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001828 struct list_head tasks;
Kirill Korotaeve3893532006-10-02 02:18:22 -07001829
1830 ns = current->nsproxy->ipc_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831
1832 if (nsops < 1 || semid < 0)
1833 return -EINVAL;
Kirill Korotaeve3893532006-10-02 02:18:22 -07001834 if (nsops > ns->sc_semopm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835 return -E2BIG;
Manfred Spraul239521f2014-01-27 17:07:04 -08001836 if (nsops > SEMOPM_FAST) {
1837 sops = kmalloc(sizeof(*sops)*nsops, GFP_KERNEL);
1838 if (sops == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839 return -ENOMEM;
1840 }
Manfred Spraul239521f2014-01-27 17:07:04 -08001841 if (copy_from_user(sops, tsops, nsops * sizeof(*tsops))) {
1842 error = -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843 goto out_free;
1844 }
1845 if (timeout) {
1846 struct timespec _timeout;
1847 if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1848 error = -EFAULT;
1849 goto out_free;
1850 }
1851 if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1852 _timeout.tv_nsec >= 1000000000L) {
1853 error = -EINVAL;
1854 goto out_free;
1855 }
1856 jiffies_left = timespec_to_jiffies(&_timeout);
1857 }
1858 max = 0;
1859 for (sop = sops; sop < sops + nsops; sop++) {
1860 if (sop->sem_num >= max)
1861 max = sop->sem_num;
1862 if (sop->sem_flg & SEM_UNDO)
Manfred Spraulb78755a2005-06-23 00:10:06 -07001863 undos = 1;
1864 if (sop->sem_op != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865 alter = 1;
1866 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867
Rik van Riel6062a8d2013-04-30 19:15:44 -07001868 INIT_LIST_HEAD(&tasks);
1869
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870 if (undos) {
Rik van Riel6062a8d2013-04-30 19:15:44 -07001871 /* On success, find_alloc_undo takes the rcu_read_lock */
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001872 un = find_alloc_undo(ns, semid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873 if (IS_ERR(un)) {
1874 error = PTR_ERR(un);
1875 goto out_free;
1876 }
Rik van Riel6062a8d2013-04-30 19:15:44 -07001877 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878 un = NULL;
Rik van Riel6062a8d2013-04-30 19:15:44 -07001879 rcu_read_lock();
1880 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001882 sma = sem_obtain_object_check(ns, semid);
Nadia Derbey023a5352007-10-18 23:40:51 -07001883 if (IS_ERR(sma)) {
Rik van Riel6062a8d2013-04-30 19:15:44 -07001884 rcu_read_unlock();
Nadia Derbey023a5352007-10-18 23:40:51 -07001885 error = PTR_ERR(sma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886 goto out_free;
Nadia Derbey023a5352007-10-18 23:40:51 -07001887 }
1888
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001889 error = -EFBIG;
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001890 if (max >= sma->sem_nsems)
1891 goto out_rcu_wakeup;
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001892
1893 error = -EACCES;
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001894 if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO))
1895 goto out_rcu_wakeup;
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001896
1897 error = security_sem_semop(sma, sops, nsops, alter);
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001898 if (error)
1899 goto out_rcu_wakeup;
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001900
Manfred Spraul6e224f92013-10-16 13:46:45 -07001901 error = -EIDRM;
1902 locknum = sem_lock(sma, sops, nsops);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001903 /*
1904 * We eventually might perform the following check in a lockless
1905 * fashion, considering ipc_valid_object() locking constraints.
1906 * If nsops == 1 and there is no contention for sem_perm.lock, then
1907 * only a per-semaphore lock is held and it's OK to proceed with the
1908 * check below. More details on the fine grained locking scheme
1909 * entangled here and why it's RMID race safe on comments at sem_lock()
1910 */
1911 if (!ipc_valid_object(&sma->sem_perm))
Manfred Spraul6e224f92013-10-16 13:46:45 -07001912 goto out_unlock_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913 /*
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001914 * semid identifiers are not unique - find_alloc_undo may have
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915 * allocated an undo structure, it was invalidated by an RMID
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001916 * and now a new array with received the same id. Check and fail.
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001917 * This case can be detected checking un->semid. The existence of
Manfred Spraul380af1b2008-07-25 01:48:06 -07001918 * "un" itself is guaranteed by rcu.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919 */
Rik van Riel6062a8d2013-04-30 19:15:44 -07001920 if (un && un->semid == -1)
1921 goto out_unlock_free;
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001922
Manfred Sprauld198cd62014-06-06 14:37:49 -07001923 queue.sops = sops;
1924 queue.nsops = nsops;
1925 queue.undo = un;
1926 queue.pid = task_tgid_vnr(current);
1927 queue.alter = alter;
1928
1929 error = perform_atomic_semop(sma, &queue);
Manfred Spraul0e8c6652013-09-30 13:45:25 -07001930 if (error == 0) {
1931 /* If the operation was successful, then do
1932 * the required updates.
1933 */
1934 if (alter)
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001935 do_smart_update(sma, sops, nsops, 1, &tasks);
Manfred Spraul0e8c6652013-09-30 13:45:25 -07001936 else
1937 set_semotime(sma, sops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938 }
Manfred Spraul0e8c6652013-09-30 13:45:25 -07001939 if (error <= 0)
1940 goto out_unlock_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941
1942 /* We need to sleep on this operation, so we put the current
1943 * task into the pending queue and go to sleep.
1944 */
Paul McQuade46c0a8c2014-06-06 14:37:37 -07001945
Manfred Spraulb97e8202009-12-15 16:47:32 -08001946 if (nsops == 1) {
1947 struct sem *curr;
1948 curr = &sma->sem_base[sops->sem_num];
1949
Manfred Spraulf269f402013-07-08 16:01:24 -07001950 if (alter) {
1951 if (sma->complex_count) {
1952 list_add_tail(&queue.list,
1953 &sma->pending_alter);
1954 } else {
1955
1956 list_add_tail(&queue.list,
1957 &curr->pending_alter);
1958 }
1959 } else {
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001960 list_add_tail(&queue.list, &curr->pending_const);
Manfred Spraulf269f402013-07-08 16:01:24 -07001961 }
Manfred Spraulb97e8202009-12-15 16:47:32 -08001962 } else {
Manfred Spraulf269f402013-07-08 16:01:24 -07001963 if (!sma->complex_count)
1964 merge_queues(sma);
1965
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001966 if (alter)
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001967 list_add_tail(&queue.list, &sma->pending_alter);
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001968 else
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001969 list_add_tail(&queue.list, &sma->pending_const);
1970
Manfred Spraulb97e8202009-12-15 16:47:32 -08001971 sma->complex_count++;
1972 }
1973
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974 queue.status = -EINTR;
1975 queue.sleeper = current;
Manfred Spraul0b0577f2011-11-02 13:38:52 -07001976
1977sleep_again:
Davidlohr Bueso52644c92015-02-17 13:47:55 -08001978 __set_current_state(TASK_INTERRUPTIBLE);
Rik van Riel6062a8d2013-04-30 19:15:44 -07001979 sem_unlock(sma, locknum);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001980 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981
1982 if (timeout)
1983 jiffies_left = schedule_timeout(jiffies_left);
1984 else
1985 schedule();
1986
Manfred Spraulc61284e2010-07-20 13:24:23 -07001987 error = get_queue_result(&queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988
1989 if (error != -EINTR) {
1990 /* fast path: update_queue already obtained all requested
Manfred Spraulc61284e2010-07-20 13:24:23 -07001991 * resources.
1992 * Perform a smp_mb(): User space could assume that semop()
1993 * is a memory barrier: Without the mb(), the cpu could
1994 * speculatively read in user space stale data that was
1995 * overwritten by the previous owner of the semaphore.
1996 */
1997 smp_mb();
1998
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999 goto out_free;
2000 }
2001
Linus Torvalds321310c2013-05-04 10:47:57 -07002002 rcu_read_lock();
Rik van Riel6062a8d2013-04-30 19:15:44 -07002003 sma = sem_obtain_lock(ns, semid, sops, nsops, &locknum);
Manfred Sprauld694ad62011-07-25 17:11:47 -07002004
2005 /*
2006 * Wait until it's guaranteed that no wakeup_sem_queue_do() is ongoing.
2007 */
2008 error = get_queue_result(&queue);
2009
2010 /*
2011 * Array removed? If yes, leave without sem_unlock().
2012 */
Nadia Derbey023a5352007-10-18 23:40:51 -07002013 if (IS_ERR(sma)) {
Linus Torvalds321310c2013-05-04 10:47:57 -07002014 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015 goto out_free;
2016 }
2017
Manfred Spraulc61284e2010-07-20 13:24:23 -07002018
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019 /*
Manfred Sprauld694ad62011-07-25 17:11:47 -07002020 * If queue.status != -EINTR we are woken up by another process.
2021 * Leave without unlink_queue(), but with sem_unlock().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022 */
Davidlohr Bueso3ab08fe2014-01-27 17:07:06 -08002023 if (error != -EINTR)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002024 goto out_unlock_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025
2026 /*
2027 * If an interrupt occurred we have to clean up the queue
2028 */
2029 if (timeout && jiffies_left == 0)
2030 error = -EAGAIN;
Manfred Spraul0b0577f2011-11-02 13:38:52 -07002031
2032 /*
2033 * If the wakeup was spurious, just retry
2034 */
2035 if (error == -EINTR && !signal_pending(current))
2036 goto sleep_again;
2037
Manfred Spraulb97e8202009-12-15 16:47:32 -08002038 unlink_queue(sma, &queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039
2040out_unlock_free:
Rik van Riel6062a8d2013-04-30 19:15:44 -07002041 sem_unlock(sma, locknum);
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07002042out_rcu_wakeup:
Linus Torvalds6d49dab2013-05-03 15:04:40 -07002043 rcu_read_unlock();
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07002044 wake_up_sem_queue_do(&tasks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045out_free:
Manfred Spraul239521f2014-01-27 17:07:04 -08002046 if (sops != fast_sops)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 kfree(sops);
2048 return error;
2049}
2050
Heiko Carstensd5460c92009-01-14 14:14:27 +01002051SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
2052 unsigned, nsops)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053{
2054 return sys_semtimedop(semid, tsops, nsops, NULL);
2055}
2056
2057/* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
2058 * parent and child tasks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059 */
2060
2061int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
2062{
2063 struct sem_undo_list *undo_list;
2064 int error;
2065
2066 if (clone_flags & CLONE_SYSVSEM) {
2067 error = get_undo_list(&undo_list);
2068 if (error)
2069 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070 atomic_inc(&undo_list->refcnt);
2071 tsk->sysvsem.undo_list = undo_list;
Paul McQuade46c0a8c2014-06-06 14:37:37 -07002072 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073 tsk->sysvsem.undo_list = NULL;
2074
2075 return 0;
2076}
2077
2078/*
2079 * add semadj values to semaphores, free undo structures.
2080 * undo structures are not freed when semaphore arrays are destroyed
2081 * so some of them may be out of date.
2082 * IMPLEMENTATION NOTE: There is some confusion over whether the
2083 * set of adjustments that needs to be done should be done in an atomic
2084 * manner or not. That is, if we are attempting to decrement the semval
2085 * should we queue up and wait until we can do so legally?
2086 * The original implementation attempted to do this (queue and wait).
2087 * The current implementation does not do so. The POSIX standard
2088 * and SVID should be consulted to determine what behavior is mandated.
2089 */
2090void exit_sem(struct task_struct *tsk)
2091{
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002092 struct sem_undo_list *ulp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002094 ulp = tsk->sysvsem.undo_list;
2095 if (!ulp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096 return;
Manfred Spraul9edff4a2008-04-29 01:00:57 -07002097 tsk->sysvsem.undo_list = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002099 if (!atomic_dec_and_test(&ulp->refcnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100 return;
2101
Manfred Spraul380af1b2008-07-25 01:48:06 -07002102 for (;;) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103 struct sem_array *sma;
Manfred Spraul380af1b2008-07-25 01:48:06 -07002104 struct sem_undo *un;
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07002105 struct list_head tasks;
Rik van Riel6062a8d2013-04-30 19:15:44 -07002106 int semid, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107
Nikolay Borisov2a1613a2016-10-11 13:55:05 -07002108 cond_resched();
2109
Manfred Spraul380af1b2008-07-25 01:48:06 -07002110 rcu_read_lock();
Jiri Pirko05725f72009-04-14 20:17:16 +02002111 un = list_entry_rcu(ulp->list_proc.next,
2112 struct sem_undo, list_proc);
Herton R. Krzesinski602b8592015-08-14 15:35:02 -07002113 if (&un->list_proc == &ulp->list_proc) {
2114 /*
2115 * We must wait for freeary() before freeing this ulp,
2116 * in case we raced with last sem_undo. There is a small
2117 * possibility where we exit while freeary() didn't
2118 * finish unlocking sem_undo_list.
2119 */
2120 spin_unlock_wait(&ulp->lock);
Rik van Riel6062a8d2013-04-30 19:15:44 -07002121 rcu_read_unlock();
Manfred Spraul380af1b2008-07-25 01:48:06 -07002122 break;
Rik van Riel6062a8d2013-04-30 19:15:44 -07002123 }
Herton R. Krzesinski602b8592015-08-14 15:35:02 -07002124 spin_lock(&ulp->lock);
2125 semid = un->semid;
2126 spin_unlock(&ulp->lock);
Manfred Spraul380af1b2008-07-25 01:48:06 -07002127
Herton R. Krzesinski602b8592015-08-14 15:35:02 -07002128 /* exit_sem raced with IPC_RMID, nothing to do */
2129 if (semid == -1) {
2130 rcu_read_unlock();
2131 continue;
2132 }
2133
2134 sma = sem_obtain_object_check(tsk->nsproxy->ipc_ns, semid);
Manfred Spraul380af1b2008-07-25 01:48:06 -07002135 /* exit_sem raced with IPC_RMID, nothing to do */
Rik van Riel6062a8d2013-04-30 19:15:44 -07002136 if (IS_ERR(sma)) {
2137 rcu_read_unlock();
Manfred Spraul380af1b2008-07-25 01:48:06 -07002138 continue;
Rik van Riel6062a8d2013-04-30 19:15:44 -07002139 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140
Rik van Riel6062a8d2013-04-30 19:15:44 -07002141 sem_lock(sma, NULL, -1);
Manfred Spraul6e224f92013-10-16 13:46:45 -07002142 /* exit_sem raced with IPC_RMID, nothing to do */
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08002143 if (!ipc_valid_object(&sma->sem_perm)) {
Manfred Spraul6e224f92013-10-16 13:46:45 -07002144 sem_unlock(sma, -1);
2145 rcu_read_unlock();
2146 continue;
2147 }
Nick Pigginbf17bb72009-12-15 16:47:28 -08002148 un = __lookup_undo(ulp, semid);
Manfred Spraul380af1b2008-07-25 01:48:06 -07002149 if (un == NULL) {
2150 /* exit_sem raced with IPC_RMID+semget() that created
2151 * exactly the same semid. Nothing to do.
2152 */
Rik van Riel6062a8d2013-04-30 19:15:44 -07002153 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07002154 rcu_read_unlock();
Manfred Spraul380af1b2008-07-25 01:48:06 -07002155 continue;
2156 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157
Manfred Spraul380af1b2008-07-25 01:48:06 -07002158 /* remove un from the linked lists */
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -07002159 ipc_assert_locked_object(&sma->sem_perm);
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002160 list_del(&un->list_id);
2161
Herton R. Krzesinskia9795582015-08-14 15:35:05 -07002162 /* we are the last process using this ulp, acquiring ulp->lock
2163 * isn't required. Besides that, we are also protected against
2164 * IPC_RMID as we hold sma->sem_perm lock now
2165 */
Manfred Spraul380af1b2008-07-25 01:48:06 -07002166 list_del_rcu(&un->list_proc);
Manfred Spraul380af1b2008-07-25 01:48:06 -07002167
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002168 /* perform adjustments registered in un */
2169 for (i = 0; i < sma->sem_nsems; i++) {
Manfred Spraul239521f2014-01-27 17:07:04 -08002170 struct sem *semaphore = &sma->sem_base[i];
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002171 if (un->semadj[i]) {
2172 semaphore->semval += un->semadj[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002173 /*
2174 * Range checks of the new semaphore value,
2175 * not defined by sus:
2176 * - Some unices ignore the undo entirely
2177 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
2178 * - some cap the value (e.g. FreeBSD caps
2179 * at 0, but doesn't enforce SEMVMX)
2180 *
2181 * Linux caps the semaphore value, both at 0
2182 * and at SEMVMX.
2183 *
Manfred Spraul239521f2014-01-27 17:07:04 -08002184 * Manfred <manfred@colorfullife.com>
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185 */
Ingo Molnar5f921ae2006-03-26 01:37:17 -08002186 if (semaphore->semval < 0)
2187 semaphore->semval = 0;
2188 if (semaphore->semval > SEMVMX)
2189 semaphore->semval = SEMVMX;
Pavel Emelyanovb4888932007-10-18 23:40:14 -07002190 semaphore->sempid = task_tgid_vnr(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002191 }
2192 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002193 /* maybe some queued-up processes were waiting for this */
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07002194 INIT_LIST_HEAD(&tasks);
2195 do_smart_update(sma, NULL, 0, 1, &tasks);
Rik van Riel6062a8d2013-04-30 19:15:44 -07002196 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07002197 rcu_read_unlock();
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07002198 wake_up_sem_queue_do(&tasks);
Manfred Spraul380af1b2008-07-25 01:48:06 -07002199
Lai Jiangshan693a8b62011-03-18 12:09:35 +08002200 kfree_rcu(un, rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201 }
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002202 kfree(ulp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203}
2204
2205#ifdef CONFIG_PROC_FS
Mike Waychison19b49462005-09-06 15:17:10 -07002206static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207{
Eric W. Biederman1efdb692012-02-07 16:54:11 -08002208 struct user_namespace *user_ns = seq_user_ns(s);
Mike Waychison19b49462005-09-06 15:17:10 -07002209 struct sem_array *sma = it;
Manfred Sprauld12e1e52013-07-08 16:01:25 -07002210 time_t sem_otime;
2211
Manfred Sprauld8c63372013-09-30 13:45:07 -07002212 /*
2213 * The proc interface isn't aware of sem_lock(), it calls
2214 * ipc_lock_object() directly (in sysvipc_find_ipc).
Manfred Spraul5864a2f2016-10-11 13:54:50 -07002215 * In order to stay compatible with sem_lock(), we must
2216 * enter / leave complex_mode.
Manfred Sprauld8c63372013-09-30 13:45:07 -07002217 */
Manfred Spraul5864a2f2016-10-11 13:54:50 -07002218 complexmode_enter(sma);
Manfred Sprauld8c63372013-09-30 13:45:07 -07002219
Manfred Sprauld12e1e52013-07-08 16:01:25 -07002220 sem_otime = get_semotime(sma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221
Joe Perches7f032d62015-04-15 16:17:54 -07002222 seq_printf(s,
2223 "%10d %10d %4o %10u %5u %5u %5u %5u %10lu %10lu\n",
2224 sma->sem_perm.key,
2225 sma->sem_perm.id,
2226 sma->sem_perm.mode,
2227 sma->sem_nsems,
2228 from_kuid_munged(user_ns, sma->sem_perm.uid),
2229 from_kgid_munged(user_ns, sma->sem_perm.gid),
2230 from_kuid_munged(user_ns, sma->sem_perm.cuid),
2231 from_kgid_munged(user_ns, sma->sem_perm.cgid),
2232 sem_otime,
2233 sma->sem_ctime);
2234
Manfred Spraul5864a2f2016-10-11 13:54:50 -07002235 complexmode_tryleave(sma);
2236
Joe Perches7f032d62015-04-15 16:17:54 -07002237 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238}
2239#endif