blob: bee5554173120780374e6b42228d13c786ceb315 [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.
50 * - semncnt and semzcnt are calculated on demand in count_semncnt() and
51 * count_semzcnt()
52 * - the task that performs a successful semop() scans the list of all
53 * sleeping tasks and completes any pending operations that can be fulfilled.
54 * Semaphores are actively given to waiting tasks (necessary for FIFO).
55 * (see update_queue())
56 * - To improve the scalability, the actual wake-up calls are performed after
57 * dropping all locks. (see wake_up_sem_queue_prepare(),
58 * wake_up_sem_queue_do())
59 * - All work is done by the waker, the woken up task does not have to do
60 * anything - not even acquiring a lock or dropping a refcount.
61 * - A woken up task may not even touch the semaphore array anymore, it may
62 * have been destroyed already by a semctl(RMID).
63 * - The synchronizations between wake-ups due to a timeout/signal and a
64 * wake-up due to a completed semaphore operation is achieved by using an
65 * intermediate state (IN_WAKEUP).
66 * - UNDO values are stored in an array (one per process and per
67 * semaphore array, lazily allocated). For backwards compatibility, multiple
68 * modes for the UNDO variables are supported (per process, per thread)
69 * (see copy_semundo, CLONE_SYSVSEM)
70 * - There are two lists of the pending operations: a per-array list
71 * and per-semaphore list (stored in the array). This allows to achieve FIFO
72 * ordering without always scanning all pending operations.
73 * The worst-case behavior is nevertheless O(N^2) for N wakeups.
Linus Torvalds1da177e2005-04-16 15:20:36 -070074 */
75
Linus Torvalds1da177e2005-04-16 15:20:36 -070076#include <linux/slab.h>
77#include <linux/spinlock.h>
78#include <linux/init.h>
79#include <linux/proc_fs.h>
80#include <linux/time.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070081#include <linux/security.h>
82#include <linux/syscalls.h>
83#include <linux/audit.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080084#include <linux/capability.h>
Mike Waychison19b49462005-09-06 15:17:10 -070085#include <linux/seq_file.h>
Nadia Derbey3e148c72007-10-18 23:40:54 -070086#include <linux/rwsem.h>
Kirill Korotaeve3893532006-10-02 02:18:22 -070087#include <linux/nsproxy.h>
Pavel Emelyanovae5e1b22008-02-08 04:18:22 -080088#include <linux/ipc_namespace.h>
Ingo Molnar5f921ae2006-03-26 01:37:17 -080089
Linus Torvalds1da177e2005-04-16 15:20:36 -070090#include <asm/uaccess.h>
91#include "util.h"
92
Manfred Spraule57940d2011-11-02 13:38:54 -070093/* One semaphore structure for each semaphore in the system. */
94struct sem {
95 int semval; /* current value */
96 int sempid; /* pid of last operation */
Rik van Riel6062a8d2013-04-30 19:15:44 -070097 spinlock_t lock; /* spinlock for fine-grained semtimedop */
Manfred Spraul1a82e9e2013-07-08 16:01:23 -070098 struct list_head pending_alter; /* pending single-sop operations */
99 /* that alter the semaphore */
100 struct list_head pending_const; /* pending single-sop operations */
101 /* that do not alter the semaphore*/
Manfred Sprauld12e1e52013-07-08 16:01:25 -0700102 time_t sem_otime; /* candidate for sem_otime */
Manfred Spraulf5c936c2013-07-08 16:01:22 -0700103} ____cacheline_aligned_in_smp;
Manfred Spraule57940d2011-11-02 13:38:54 -0700104
105/* One queue for each sleeping process in the system. */
106struct sem_queue {
Manfred Spraule57940d2011-11-02 13:38:54 -0700107 struct list_head list; /* queue of pending operations */
108 struct task_struct *sleeper; /* this process */
109 struct sem_undo *undo; /* undo structure */
110 int pid; /* process id of requesting process */
111 int status; /* completion status of operation */
112 struct sembuf *sops; /* array of pending operations */
113 int nsops; /* number of operations */
114 int alter; /* does *sops alter the array? */
115};
116
117/* Each task has a list of undo requests. They are executed automatically
118 * when the process exits.
119 */
120struct sem_undo {
121 struct list_head list_proc; /* per-process list: *
122 * all undos from one process
123 * rcu protected */
124 struct rcu_head rcu; /* rcu struct for sem_undo */
125 struct sem_undo_list *ulp; /* back ptr to sem_undo_list */
126 struct list_head list_id; /* per semaphore array list:
127 * all undos for one array */
128 int semid; /* semaphore set identifier */
129 short *semadj; /* array of adjustments */
130 /* one per semaphore */
131};
132
133/* sem_undo_list controls shared access to the list of sem_undo structures
134 * that may be shared among all a CLONE_SYSVSEM task group.
135 */
136struct sem_undo_list {
137 atomic_t refcnt;
138 spinlock_t lock;
139 struct list_head list_proc;
140};
141
142
Pierre Peiffered2ddbf2008-02-08 04:18:57 -0800143#define sem_ids(ns) ((ns)->ids[IPC_SEM_IDS])
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144
Nadia Derbey1b531f22007-10-18 23:40:55 -0700145#define sem_checkid(sma, semid) ipc_checkid(&sma->sem_perm, semid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700147static int newary(struct ipc_namespace *, struct ipc_params *);
Pierre Peiffer01b8b072008-02-08 04:18:57 -0800148static void freeary(struct ipc_namespace *, struct kern_ipc_perm *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149#ifdef CONFIG_PROC_FS
Mike Waychison19b49462005-09-06 15:17:10 -0700150static int sysvipc_sem_proc_show(struct seq_file *s, void *it);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151#endif
152
153#define SEMMSL_FAST 256 /* 512 bytes on stack */
154#define SEMOPM_FAST 64 /* ~ 372 bytes on stack */
155
156/*
Manfred Spraul758a6ba32013-07-08 16:01:26 -0700157 * Locking:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 * sem_undo.id_next,
Manfred Spraul758a6ba32013-07-08 16:01:26 -0700159 * sem_array.complex_count,
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700160 * sem_array.pending{_alter,_cont},
Manfred Spraul758a6ba32013-07-08 16:01:26 -0700161 * sem_array.sem_undo: global sem_lock() for read/write
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 * sem_undo.proc_next: only "current" is allowed to read/write that field.
163 *
Manfred Spraul758a6ba32013-07-08 16:01:26 -0700164 * sem_array.sem_base[i].pending_{const,alter}:
165 * global or semaphore sem_lock() for read/write
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166 */
167
Kirill Korotaeve3893532006-10-02 02:18:22 -0700168#define sc_semmsl sem_ctls[0]
169#define sc_semmns sem_ctls[1]
170#define sc_semopm sem_ctls[2]
171#define sc_semmni sem_ctls[3]
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172
Pierre Peiffered2ddbf2008-02-08 04:18:57 -0800173void sem_init_ns(struct ipc_namespace *ns)
Kirill Korotaeve3893532006-10-02 02:18:22 -0700174{
Kirill Korotaeve3893532006-10-02 02:18:22 -0700175 ns->sc_semmsl = SEMMSL;
176 ns->sc_semmns = SEMMNS;
177 ns->sc_semopm = SEMOPM;
178 ns->sc_semmni = SEMMNI;
179 ns->used_sems = 0;
Pierre Peiffered2ddbf2008-02-08 04:18:57 -0800180 ipc_init_ids(&ns->ids[IPC_SEM_IDS]);
Kirill Korotaeve3893532006-10-02 02:18:22 -0700181}
182
Pavel Emelyanovae5e1b22008-02-08 04:18:22 -0800183#ifdef CONFIG_IPC_NS
Kirill Korotaeve3893532006-10-02 02:18:22 -0700184void sem_exit_ns(struct ipc_namespace *ns)
185{
Pierre Peiffer01b8b072008-02-08 04:18:57 -0800186 free_ipcs(ns, &sem_ids(ns), freeary);
Serge E. Hallyn7d6feeb2009-12-15 16:47:27 -0800187 idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr);
Kirill Korotaeve3893532006-10-02 02:18:22 -0700188}
Pavel Emelyanovae5e1b22008-02-08 04:18:22 -0800189#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190
Manfred Spraul239521f2014-01-27 17:07:04 -0800191void __init sem_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192{
Pierre Peiffered2ddbf2008-02-08 04:18:57 -0800193 sem_init_ns(&init_ipc_ns);
Mike Waychison19b49462005-09-06 15:17:10 -0700194 ipc_init_proc_interface("sysvipc/sem",
195 " key semid perms nsems uid gid cuid cgid otime ctime\n",
Kirill Korotaeve3893532006-10-02 02:18:22 -0700196 IPC_SEM_IDS, sysvipc_sem_proc_show);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197}
198
Manfred Spraulf269f402013-07-08 16:01:24 -0700199/**
200 * unmerge_queues - unmerge queues, if possible.
201 * @sma: semaphore array
202 *
203 * The function unmerges the wait queues if complex_count is 0.
204 * It must be called prior to dropping the global semaphore array lock.
205 */
206static void unmerge_queues(struct sem_array *sma)
207{
208 struct sem_queue *q, *tq;
209
210 /* complex operations still around? */
211 if (sma->complex_count)
212 return;
213 /*
214 * We will switch back to simple mode.
215 * Move all pending operation back into the per-semaphore
216 * queues.
217 */
218 list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
219 struct sem *curr;
220 curr = &sma->sem_base[q->sops[0].sem_num];
221
222 list_add_tail(&q->list, &curr->pending_alter);
223 }
224 INIT_LIST_HEAD(&sma->pending_alter);
225}
226
227/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800228 * merge_queues - merge single semop queues into global queue
Manfred Spraulf269f402013-07-08 16:01:24 -0700229 * @sma: semaphore array
230 *
231 * This function merges all per-semaphore queues into the global queue.
232 * It is necessary to achieve FIFO ordering for the pending single-sop
233 * operations when a multi-semop operation must sleep.
234 * Only the alter operations must be moved, the const operations can stay.
235 */
236static void merge_queues(struct sem_array *sma)
237{
238 int i;
239 for (i = 0; i < sma->sem_nsems; i++) {
240 struct sem *sem = sma->sem_base + i;
241
242 list_splice_init(&sem->pending_alter, &sma->pending_alter);
243 }
244}
245
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -0700246static void sem_rcu_free(struct rcu_head *head)
247{
248 struct ipc_rcu *p = container_of(head, struct ipc_rcu, rcu);
249 struct sem_array *sma = ipc_rcu_to_struct(p);
250
251 security_sem_free(sma);
252 ipc_rcu_free(head);
253}
254
Nadia Derbey3e148c72007-10-18 23:40:54 -0700255/*
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700256 * Wait until all currently ongoing simple ops have completed.
257 * Caller must own sem_perm.lock.
258 * New simple ops cannot start, because simple ops first check
259 * that sem_perm.lock is free.
Manfred Spraul6d07b682013-09-30 13:45:06 -0700260 * that a) sem_perm.lock is free and b) complex_count is 0.
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700261 */
262static void sem_wait_array(struct sem_array *sma)
263{
264 int i;
265 struct sem *sem;
266
Manfred Spraul6d07b682013-09-30 13:45:06 -0700267 if (sma->complex_count) {
268 /* The thread that increased sma->complex_count waited on
269 * all sem->lock locks. Thus we don't need to wait again.
270 */
271 return;
272 }
273
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700274 for (i = 0; i < sma->sem_nsems; i++) {
275 sem = sma->sem_base + i;
276 spin_unlock_wait(&sem->lock);
277 }
278}
279
280/*
Rik van Riel6062a8d2013-04-30 19:15:44 -0700281 * If the request contains only one semaphore operation, and there are
282 * no complex transactions pending, lock only the semaphore involved.
283 * Otherwise, lock the entire semaphore array, since we either have
284 * multiple semaphores in our own semops, or we need to look at
285 * semaphores from other pending complex operations.
Rik van Riel6062a8d2013-04-30 19:15:44 -0700286 */
287static inline int sem_lock(struct sem_array *sma, struct sembuf *sops,
288 int nsops)
289{
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700290 struct sem *sem;
Rik van Riel6062a8d2013-04-30 19:15:44 -0700291
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700292 if (nsops != 1) {
293 /* Complex operation - acquire a full lock */
294 ipc_lock_object(&sma->sem_perm);
295
296 /* And wait until all simple ops that are processed
297 * right now have dropped their locks.
298 */
299 sem_wait_array(sma);
300 return -1;
301 }
302
303 /*
304 * Only one semaphore affected - try to optimize locking.
305 * The rules are:
306 * - optimized locking is possible if no complex operation
307 * is either enqueued or processed right now.
308 * - The test for enqueued complex ops is simple:
309 * sma->complex_count != 0
310 * - Testing for complex ops that are processed right now is
311 * a bit more difficult. Complex ops acquire the full lock
312 * and first wait that the running simple ops have completed.
313 * (see above)
314 * Thus: If we own a simple lock and the global lock is free
315 * and complex_count is now 0, then it will stay 0 and
316 * thus just locking sem->lock is sufficient.
317 */
318 sem = sma->sem_base + sops->sem_num;
319
320 if (sma->complex_count == 0) {
321 /*
322 * It appears that no complex operation is around.
323 * Acquire the per-semaphore lock.
324 */
Rik van Riel6062a8d2013-04-30 19:15:44 -0700325 spin_lock(&sem->lock);
326
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700327 /* Then check that the global lock is free */
328 if (!spin_is_locked(&sma->sem_perm.lock)) {
329 /* spin_is_locked() is not a memory barrier */
330 smp_mb();
Rik van Riel6062a8d2013-04-30 19:15:44 -0700331
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700332 /* Now repeat the test of complex_count:
333 * It can't change anymore until we drop sem->lock.
334 * Thus: if is now 0, then it will stay 0.
335 */
336 if (sma->complex_count == 0) {
337 /* fast path successful! */
338 return sops->sem_num;
339 }
Rik van Riel6062a8d2013-04-30 19:15:44 -0700340 }
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700341 spin_unlock(&sem->lock);
Rik van Riel6062a8d2013-04-30 19:15:44 -0700342 }
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700343
344 /* slow path: acquire the full lock */
345 ipc_lock_object(&sma->sem_perm);
346
347 if (sma->complex_count == 0) {
348 /* False alarm:
349 * There is no complex operation, thus we can switch
350 * back to the fast path.
351 */
352 spin_lock(&sem->lock);
353 ipc_unlock_object(&sma->sem_perm);
354 return sops->sem_num;
355 } else {
356 /* Not a false alarm, thus complete the sequence for a
357 * full lock.
358 */
359 sem_wait_array(sma);
360 return -1;
361 }
Rik van Riel6062a8d2013-04-30 19:15:44 -0700362}
363
364static inline void sem_unlock(struct sem_array *sma, int locknum)
365{
366 if (locknum == -1) {
Manfred Spraulf269f402013-07-08 16:01:24 -0700367 unmerge_queues(sma);
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -0700368 ipc_unlock_object(&sma->sem_perm);
Rik van Riel6062a8d2013-04-30 19:15:44 -0700369 } else {
370 struct sem *sem = sma->sem_base + locknum;
371 spin_unlock(&sem->lock);
372 }
Rik van Riel6062a8d2013-04-30 19:15:44 -0700373}
374
375/*
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700376 * sem_lock_(check_) routines are called in the paths where the rwsem
Nadia Derbey3e148c72007-10-18 23:40:54 -0700377 * is not held.
Linus Torvalds321310c2013-05-04 10:47:57 -0700378 *
379 * The caller holds the RCU read lock.
Nadia Derbey3e148c72007-10-18 23:40:54 -0700380 */
Rik van Riel6062a8d2013-04-30 19:15:44 -0700381static inline struct sem_array *sem_obtain_lock(struct ipc_namespace *ns,
382 int id, struct sembuf *sops, int nsops, int *locknum)
Nadia Derbey023a5352007-10-18 23:40:51 -0700383{
Rik van Rielc460b662013-04-30 19:15:35 -0700384 struct kern_ipc_perm *ipcp;
385 struct sem_array *sma;
Nadia Derbey03f02c72007-10-18 23:40:51 -0700386
Rik van Rielc460b662013-04-30 19:15:35 -0700387 ipcp = ipc_obtain_object(&sem_ids(ns), id);
Linus Torvalds321310c2013-05-04 10:47:57 -0700388 if (IS_ERR(ipcp))
389 return ERR_CAST(ipcp);
Pierre Peifferb1ed88b2008-02-06 01:36:23 -0800390
Rik van Riel6062a8d2013-04-30 19:15:44 -0700391 sma = container_of(ipcp, struct sem_array, sem_perm);
392 *locknum = sem_lock(sma, sops, nsops);
Rik van Rielc460b662013-04-30 19:15:35 -0700393
394 /* ipc_rmid() may have already freed the ID while sem_lock
395 * was spinning: verify that the structure is still valid
396 */
Rafael Aquini72a8ff22014-01-27 17:07:02 -0800397 if (ipc_valid_object(ipcp))
Rik van Rielc460b662013-04-30 19:15:35 -0700398 return container_of(ipcp, struct sem_array, sem_perm);
399
Rik van Riel6062a8d2013-04-30 19:15:44 -0700400 sem_unlock(sma, *locknum);
Linus Torvalds321310c2013-05-04 10:47:57 -0700401 return ERR_PTR(-EINVAL);
Nadia Derbey023a5352007-10-18 23:40:51 -0700402}
403
Davidlohr Bueso16df3672013-04-30 19:15:29 -0700404static inline struct sem_array *sem_obtain_object(struct ipc_namespace *ns, int id)
405{
406 struct kern_ipc_perm *ipcp = ipc_obtain_object(&sem_ids(ns), id);
407
408 if (IS_ERR(ipcp))
409 return ERR_CAST(ipcp);
410
411 return container_of(ipcp, struct sem_array, sem_perm);
412}
413
Davidlohr Bueso16df3672013-04-30 19:15:29 -0700414static inline struct sem_array *sem_obtain_object_check(struct ipc_namespace *ns,
415 int id)
416{
417 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&sem_ids(ns), id);
418
419 if (IS_ERR(ipcp))
420 return ERR_CAST(ipcp);
Pierre Peifferb1ed88b2008-02-06 01:36:23 -0800421
Nadia Derbey03f02c72007-10-18 23:40:51 -0700422 return container_of(ipcp, struct sem_array, sem_perm);
Nadia Derbey023a5352007-10-18 23:40:51 -0700423}
424
Pierre Peiffer6ff37972008-04-29 01:00:46 -0700425static inline void sem_lock_and_putref(struct sem_array *sma)
426{
Rik van Riel6062a8d2013-04-30 19:15:44 -0700427 sem_lock(sma, NULL, -1);
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -0700428 ipc_rcu_putref(sma, ipc_rcu_free);
Pierre Peiffer6ff37972008-04-29 01:00:46 -0700429}
430
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700431static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
432{
433 ipc_rmid(&sem_ids(ns), &s->sem_perm);
434}
435
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436/*
437 * Lockless wakeup algorithm:
438 * Without the check/retry algorithm a lockless wakeup is possible:
439 * - queue.status is initialized to -EINTR before blocking.
440 * - wakeup is performed by
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700441 * * unlinking the queue entry from the pending list
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 * * setting queue.status to IN_WAKEUP
443 * This is the notification for the blocked thread that a
444 * result value is imminent.
445 * * call wake_up_process
446 * * set queue.status to the final value.
447 * - the previously blocked thread checks queue.status:
Manfred Spraul239521f2014-01-27 17:07:04 -0800448 * * if it's IN_WAKEUP, then it must wait until the value changes
449 * * if it's not -EINTR, then the operation was completed by
450 * update_queue. semtimedop can return queue.status without
451 * performing any operation on the sem array.
452 * * otherwise it must acquire the spinlock and check what's up.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 *
454 * The two-stage algorithm is necessary to protect against the following
455 * races:
456 * - if queue.status is set after wake_up_process, then the woken up idle
457 * thread could race forward and try (and fail) to acquire sma->lock
458 * before update_queue had a chance to set queue.status
459 * - if queue.status is written before wake_up_process and if the
460 * blocked process is woken up by a signal between writing
461 * queue.status and the wake_up_process, then the woken up
462 * process could return from semtimedop and die by calling
463 * sys_exit before wake_up_process is called. Then wake_up_process
464 * will oops, because the task structure is already invalid.
465 * (yes, this happened on s390 with sysv msg).
466 *
467 */
468#define IN_WAKEUP 1
469
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700470/**
471 * newary - Create a new semaphore set
472 * @ns: namespace
473 * @params: ptr to the structure that contains key, semflg and nsems
474 *
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700475 * Called with sem_ids.rwsem held (as a writer)
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700476 */
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700477static int newary(struct ipc_namespace *ns, struct ipc_params *params)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478{
479 int id;
480 int retval;
481 struct sem_array *sma;
482 int size;
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700483 key_t key = params->key;
484 int nsems = params->u.nsems;
485 int semflg = params->flg;
Manfred Spraulb97e8202009-12-15 16:47:32 -0800486 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487
488 if (!nsems)
489 return -EINVAL;
Kirill Korotaeve3893532006-10-02 02:18:22 -0700490 if (ns->used_sems + nsems > ns->sc_semmns)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 return -ENOSPC;
492
Manfred Spraul239521f2014-01-27 17:07:04 -0800493 size = sizeof(*sma) + nsems * sizeof(struct sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 sma = ipc_rcu_alloc(size);
Davidlohr Bueso3ab08fe2014-01-27 17:07:06 -0800495 if (!sma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 return -ENOMEM;
Davidlohr Bueso3ab08fe2014-01-27 17:07:06 -0800497
Manfred Spraul239521f2014-01-27 17:07:04 -0800498 memset(sma, 0, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499
500 sma->sem_perm.mode = (semflg & S_IRWXUGO);
501 sma->sem_perm.key = key;
502
503 sma->sem_perm.security = NULL;
504 retval = security_sem_alloc(sma);
505 if (retval) {
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -0700506 ipc_rcu_putref(sma, ipc_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 return retval;
508 }
509
Kirill Korotaeve3893532006-10-02 02:18:22 -0700510 id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
Pierre Peiffer283bb7f2007-10-18 23:40:57 -0700511 if (id < 0) {
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -0700512 ipc_rcu_putref(sma, sem_rcu_free);
Pierre Peiffer283bb7f2007-10-18 23:40:57 -0700513 return id;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 }
Kirill Korotaeve3893532006-10-02 02:18:22 -0700515 ns->used_sems += nsems;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516
517 sma->sem_base = (struct sem *) &sma[1];
Manfred Spraulb97e8202009-12-15 16:47:32 -0800518
Rik van Riel6062a8d2013-04-30 19:15:44 -0700519 for (i = 0; i < nsems; i++) {
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700520 INIT_LIST_HEAD(&sma->sem_base[i].pending_alter);
521 INIT_LIST_HEAD(&sma->sem_base[i].pending_const);
Rik van Riel6062a8d2013-04-30 19:15:44 -0700522 spin_lock_init(&sma->sem_base[i].lock);
523 }
Manfred Spraulb97e8202009-12-15 16:47:32 -0800524
525 sma->complex_count = 0;
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700526 INIT_LIST_HEAD(&sma->pending_alter);
527 INIT_LIST_HEAD(&sma->pending_const);
Manfred Spraul4daa28f2008-07-25 01:48:04 -0700528 INIT_LIST_HEAD(&sma->list_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 sma->sem_nsems = nsems;
530 sma->sem_ctime = get_seconds();
Rik van Riel6062a8d2013-04-30 19:15:44 -0700531 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -0700532 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700534 return sma->sem_perm.id;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535}
536
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700537
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700538/*
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700539 * Called with sem_ids.rwsem and ipcp locked.
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700540 */
Nadia Derbey03f02c72007-10-18 23:40:51 -0700541static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542{
Nadia Derbey03f02c72007-10-18 23:40:51 -0700543 struct sem_array *sma;
544
545 sma = container_of(ipcp, struct sem_array, sem_perm);
546 return security_sem_associate(sma, semflg);
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700547}
548
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700549/*
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700550 * Called with sem_ids.rwsem and ipcp locked.
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700551 */
Nadia Derbey03f02c72007-10-18 23:40:51 -0700552static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
553 struct ipc_params *params)
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700554{
Nadia Derbey03f02c72007-10-18 23:40:51 -0700555 struct sem_array *sma;
556
557 sma = container_of(ipcp, struct sem_array, sem_perm);
558 if (params->u.nsems > sma->sem_nsems)
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700559 return -EINVAL;
560
561 return 0;
562}
563
Heiko Carstensd5460c92009-01-14 14:14:27 +0100564SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg)
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700565{
Kirill Korotaeve3893532006-10-02 02:18:22 -0700566 struct ipc_namespace *ns;
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700567 struct ipc_ops sem_ops;
568 struct ipc_params sem_params;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569
Kirill Korotaeve3893532006-10-02 02:18:22 -0700570 ns = current->nsproxy->ipc_ns;
571
572 if (nsems < 0 || nsems > ns->sc_semmsl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 return -EINVAL;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700574
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700575 sem_ops.getnew = newary;
576 sem_ops.associate = sem_security;
577 sem_ops.more_checks = sem_more_checks;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700578
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700579 sem_params.key = key;
580 sem_params.flg = semflg;
581 sem_params.u.nsems = nsems;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700582
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700583 return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584}
585
Petr Mladek78f50092014-01-27 17:07:00 -0800586/**
587 * perform_atomic_semop - Perform (if possible) a semaphore operation
Manfred Spraul758a6ba32013-07-08 16:01:26 -0700588 * @sma: semaphore array
589 * @sops: array with operations that should be checked
Petr Mladek78f50092014-01-27 17:07:00 -0800590 * @nsops: number of operations
Manfred Spraul758a6ba32013-07-08 16:01:26 -0700591 * @un: undo array
592 * @pid: pid that did the change
593 *
594 * Returns 0 if the operation was possible.
595 * Returns 1 if the operation is impossible, the caller must sleep.
596 * Negative values are error codes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 */
Manfred Spraul758a6ba32013-07-08 16:01:26 -0700598static int perform_atomic_semop(struct sem_array *sma, struct sembuf *sops,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 int nsops, struct sem_undo *un, int pid)
600{
601 int result, sem_op;
602 struct sembuf *sop;
Manfred Spraul239521f2014-01-27 17:07:04 -0800603 struct sem *curr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604
605 for (sop = sops; sop < sops + nsops; sop++) {
606 curr = sma->sem_base + sop->sem_num;
607 sem_op = sop->sem_op;
608 result = curr->semval;
Petr Mladek78f50092014-01-27 17:07:00 -0800609
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 if (!sem_op && result)
611 goto would_block;
612
613 result += sem_op;
614 if (result < 0)
615 goto would_block;
616 if (result > SEMVMX)
617 goto out_of_range;
Petr Mladek78f50092014-01-27 17:07:00 -0800618
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 if (sop->sem_flg & SEM_UNDO) {
620 int undo = un->semadj[sop->sem_num] - sem_op;
Petr Mladek78f50092014-01-27 17:07:00 -0800621 /* Exceeding the undo range is an error. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
623 goto out_of_range;
Petr Mladek78f50092014-01-27 17:07:00 -0800624 un->semadj[sop->sem_num] = undo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 }
Petr Mladek78f50092014-01-27 17:07:00 -0800626
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 curr->semval = result;
628 }
629
630 sop--;
631 while (sop >= sops) {
632 sma->sem_base[sop->sem_num].sempid = pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 sop--;
634 }
Petr Mladek78f50092014-01-27 17:07:00 -0800635
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 return 0;
637
638out_of_range:
639 result = -ERANGE;
640 goto undo;
641
642would_block:
643 if (sop->sem_flg & IPC_NOWAIT)
644 result = -EAGAIN;
645 else
646 result = 1;
647
648undo:
649 sop--;
650 while (sop >= sops) {
Petr Mladek78f50092014-01-27 17:07:00 -0800651 sem_op = sop->sem_op;
652 sma->sem_base[sop->sem_num].semval -= sem_op;
653 if (sop->sem_flg & SEM_UNDO)
654 un->semadj[sop->sem_num] += sem_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 sop--;
656 }
657
658 return result;
659}
660
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700661/** wake_up_sem_queue_prepare(q, error): Prepare wake-up
662 * @q: queue entry that must be signaled
663 * @error: Error value for the signal
664 *
665 * Prepare the wake-up of the queue entry q.
Nick Piggind4212092009-12-15 16:47:30 -0800666 */
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700667static void wake_up_sem_queue_prepare(struct list_head *pt,
668 struct sem_queue *q, int error)
Nick Piggind4212092009-12-15 16:47:30 -0800669{
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700670 if (list_empty(pt)) {
671 /*
672 * Hold preempt off so that we don't get preempted and have the
673 * wakee busy-wait until we're scheduled back on.
674 */
675 preempt_disable();
676 }
Nick Piggind4212092009-12-15 16:47:30 -0800677 q->status = IN_WAKEUP;
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700678 q->pid = error;
679
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700680 list_add_tail(&q->list, pt);
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700681}
682
683/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800684 * wake_up_sem_queue_do - do the actual wake-up
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700685 * @pt: list of tasks to be woken up
686 *
687 * Do the actual wake-up.
688 * The function is called without any locks held, thus the semaphore array
689 * could be destroyed already and the tasks can disappear as soon as the
690 * status is set to the actual return code.
691 */
692static void wake_up_sem_queue_do(struct list_head *pt)
693{
694 struct sem_queue *q, *t;
695 int did_something;
696
697 did_something = !list_empty(pt);
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700698 list_for_each_entry_safe(q, t, pt, list) {
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700699 wake_up_process(q->sleeper);
700 /* q can disappear immediately after writing q->status. */
701 smp_wmb();
702 q->status = q->pid;
703 }
704 if (did_something)
705 preempt_enable();
Nick Piggind4212092009-12-15 16:47:30 -0800706}
707
Manfred Spraulb97e8202009-12-15 16:47:32 -0800708static void unlink_queue(struct sem_array *sma, struct sem_queue *q)
709{
710 list_del(&q->list);
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700711 if (q->nsops > 1)
Manfred Spraulb97e8202009-12-15 16:47:32 -0800712 sma->complex_count--;
713}
714
Manfred Spraulfd5db422010-05-26 14:43:40 -0700715/** check_restart(sma, q)
716 * @sma: semaphore array
717 * @q: the operation that just completed
718 *
719 * update_queue is O(N^2) when it restarts scanning the whole queue of
720 * waiting operations. Therefore this function checks if the restart is
721 * really necessary. It is called after a previously waiting operation
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700722 * modified the array.
723 * Note that wait-for-zero operations are handled without restart.
Manfred Spraulfd5db422010-05-26 14:43:40 -0700724 */
725static int check_restart(struct sem_array *sma, struct sem_queue *q)
726{
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700727 /* pending complex alter operations are too difficult to analyse */
728 if (!list_empty(&sma->pending_alter))
Manfred Spraulfd5db422010-05-26 14:43:40 -0700729 return 1;
730
731 /* we were a sleeping complex operation. Too difficult */
732 if (q->nsops > 1)
733 return 1;
734
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700735 /* It is impossible that someone waits for the new value:
736 * - complex operations always restart.
737 * - wait-for-zero are handled seperately.
738 * - q is a previously sleeping simple operation that
739 * altered the array. It must be a decrement, because
740 * simple increments never sleep.
741 * - If there are older (higher priority) decrements
742 * in the queue, then they have observed the original
743 * semval value and couldn't proceed. The operation
744 * decremented to value - thus they won't proceed either.
745 */
746 return 0;
747}
Manfred Spraulfd5db422010-05-26 14:43:40 -0700748
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700749/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800750 * wake_const_ops - wake up non-alter tasks
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700751 * @sma: semaphore array.
752 * @semnum: semaphore that was modified.
753 * @pt: list head for the tasks that must be woken up.
754 *
755 * wake_const_ops must be called after a semaphore in a semaphore array
756 * was set to 0. If complex const operations are pending, wake_const_ops must
757 * be called with semnum = -1, as well as with the number of each modified
758 * semaphore.
759 * The tasks that must be woken up are added to @pt. The return code
760 * is stored in q->pid.
761 * The function returns 1 if at least one operation was completed successfully.
762 */
763static int wake_const_ops(struct sem_array *sma, int semnum,
764 struct list_head *pt)
765{
766 struct sem_queue *q;
767 struct list_head *walk;
768 struct list_head *pending_list;
769 int semop_completed = 0;
Manfred Spraulfd5db422010-05-26 14:43:40 -0700770
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700771 if (semnum == -1)
772 pending_list = &sma->pending_const;
773 else
774 pending_list = &sma->sem_base[semnum].pending_const;
775
776 walk = pending_list->next;
777 while (walk != pending_list) {
778 int error;
779
780 q = container_of(walk, struct sem_queue, list);
781 walk = walk->next;
782
Manfred Spraul758a6ba32013-07-08 16:01:26 -0700783 error = perform_atomic_semop(sma, q->sops, q->nsops,
784 q->undo, q->pid);
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700785
786 if (error <= 0) {
787 /* operation completed, remove from queue & wakeup */
788
789 unlink_queue(sma, q);
790
791 wake_up_sem_queue_prepare(pt, q, error);
792 if (error == 0)
793 semop_completed = 1;
794 }
795 }
796 return semop_completed;
797}
798
799/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800800 * do_smart_wakeup_zero - wakeup all wait for zero tasks
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700801 * @sma: semaphore array
802 * @sops: operations that were performed
803 * @nsops: number of operations
804 * @pt: list head of the tasks that must be woken up.
805 *
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800806 * Checks all required queue for wait-for-zero operations, based
807 * on the actual changes that were performed on the semaphore array.
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700808 * The function returns 1 if at least one operation was completed successfully.
809 */
810static int do_smart_wakeup_zero(struct sem_array *sma, struct sembuf *sops,
811 int nsops, struct list_head *pt)
812{
813 int i;
814 int semop_completed = 0;
815 int got_zero = 0;
816
817 /* first: the per-semaphore queues, if known */
818 if (sops) {
819 for (i = 0; i < nsops; i++) {
820 int num = sops[i].sem_num;
821
822 if (sma->sem_base[num].semval == 0) {
823 got_zero = 1;
824 semop_completed |= wake_const_ops(sma, num, pt);
825 }
826 }
827 } else {
828 /*
829 * No sops means modified semaphores not known.
830 * Assume all were changed.
Manfred Spraulfd5db422010-05-26 14:43:40 -0700831 */
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700832 for (i = 0; i < sma->sem_nsems; i++) {
833 if (sma->sem_base[i].semval == 0) {
834 got_zero = 1;
835 semop_completed |= wake_const_ops(sma, i, pt);
836 }
837 }
Manfred Spraulfd5db422010-05-26 14:43:40 -0700838 }
839 /*
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700840 * If one of the modified semaphores got 0,
841 * then check the global queue, too.
Manfred Spraulfd5db422010-05-26 14:43:40 -0700842 */
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700843 if (got_zero)
844 semop_completed |= wake_const_ops(sma, -1, pt);
Manfred Spraulfd5db422010-05-26 14:43:40 -0700845
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700846 return semop_completed;
Manfred Spraulfd5db422010-05-26 14:43:40 -0700847}
848
Manfred Spraul636c6be2009-12-15 16:47:33 -0800849
850/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800851 * update_queue - look for tasks that can be completed.
Manfred Spraul636c6be2009-12-15 16:47:33 -0800852 * @sma: semaphore array.
853 * @semnum: semaphore that was modified.
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700854 * @pt: list head for the tasks that must be woken up.
Manfred Spraul636c6be2009-12-15 16:47:33 -0800855 *
856 * update_queue must be called after a semaphore in a semaphore array
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700857 * was modified. If multiple semaphores were modified, update_queue must
858 * be called with semnum = -1, as well as with the number of each modified
859 * semaphore.
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700860 * The tasks that must be woken up are added to @pt. The return code
861 * is stored in q->pid.
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700862 * The function internally checks if const operations can now succeed.
863 *
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700864 * The function return 1 if at least one semop was completed successfully.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 */
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700866static int update_queue(struct sem_array *sma, int semnum, struct list_head *pt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867{
Manfred Spraul636c6be2009-12-15 16:47:33 -0800868 struct sem_queue *q;
869 struct list_head *walk;
870 struct list_head *pending_list;
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700871 int semop_completed = 0;
Manfred Spraul636c6be2009-12-15 16:47:33 -0800872
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700873 if (semnum == -1)
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700874 pending_list = &sma->pending_alter;
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700875 else
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700876 pending_list = &sma->sem_base[semnum].pending_alter;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877
Nick Piggin9cad2002009-12-15 16:47:29 -0800878again:
Manfred Spraul636c6be2009-12-15 16:47:33 -0800879 walk = pending_list->next;
880 while (walk != pending_list) {
Manfred Spraulfd5db422010-05-26 14:43:40 -0700881 int error, restart;
Manfred Spraul636c6be2009-12-15 16:47:33 -0800882
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700883 q = container_of(walk, struct sem_queue, list);
Manfred Spraul636c6be2009-12-15 16:47:33 -0800884 walk = walk->next;
Nick Piggin9cad2002009-12-15 16:47:29 -0800885
Manfred Sprauld987f8b2009-12-15 16:47:34 -0800886 /* If we are scanning the single sop, per-semaphore list of
887 * one semaphore and that semaphore is 0, then it is not
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700888 * necessary to scan further: simple increments
Manfred Sprauld987f8b2009-12-15 16:47:34 -0800889 * that affect only one entry succeed immediately and cannot
890 * be in the per semaphore pending queue, and decrements
891 * cannot be successful if the value is already 0.
892 */
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700893 if (semnum != -1 && sma->sem_base[semnum].semval == 0)
Manfred Sprauld987f8b2009-12-15 16:47:34 -0800894 break;
895
Manfred Spraul758a6ba32013-07-08 16:01:26 -0700896 error = perform_atomic_semop(sma, q->sops, q->nsops,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 q->undo, q->pid);
898
899 /* Does q->sleeper still need to sleep? */
Nick Piggin9cad2002009-12-15 16:47:29 -0800900 if (error > 0)
901 continue;
Manfred Spraula1193f82008-07-25 01:48:06 -0700902
Manfred Spraulb97e8202009-12-15 16:47:32 -0800903 unlink_queue(sma, q);
Manfred Spraula1193f82008-07-25 01:48:06 -0700904
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700905 if (error) {
Manfred Spraulfd5db422010-05-26 14:43:40 -0700906 restart = 0;
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700907 } else {
908 semop_completed = 1;
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700909 do_smart_wakeup_zero(sma, q->sops, q->nsops, pt);
Manfred Spraulfd5db422010-05-26 14:43:40 -0700910 restart = check_restart(sma, q);
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700911 }
Manfred Spraulfd5db422010-05-26 14:43:40 -0700912
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700913 wake_up_sem_queue_prepare(pt, q, error);
Manfred Spraulfd5db422010-05-26 14:43:40 -0700914 if (restart)
Nick Piggin9cad2002009-12-15 16:47:29 -0800915 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 }
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700917 return semop_completed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918}
919
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700920/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800921 * set_semotime - set sem_otime
Manfred Spraul0e8c6652013-09-30 13:45:25 -0700922 * @sma: semaphore array
923 * @sops: operations that modified the array, may be NULL
924 *
925 * sem_otime is replicated to avoid cache line trashing.
926 * This function sets one instance to the current time.
927 */
928static void set_semotime(struct sem_array *sma, struct sembuf *sops)
929{
930 if (sops == NULL) {
931 sma->sem_base[0].sem_otime = get_seconds();
932 } else {
933 sma->sem_base[sops[0].sem_num].sem_otime =
934 get_seconds();
935 }
936}
937
938/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800939 * do_smart_update - optimized update_queue
Manfred Spraulfd5db422010-05-26 14:43:40 -0700940 * @sma: semaphore array
941 * @sops: operations that were performed
942 * @nsops: number of operations
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700943 * @otime: force setting otime
944 * @pt: list head of the tasks that must be woken up.
Manfred Spraulfd5db422010-05-26 14:43:40 -0700945 *
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700946 * do_smart_update() does the required calls to update_queue and wakeup_zero,
947 * based on the actual changes that were performed on the semaphore array.
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700948 * Note that the function does not do the actual wake-up: the caller is
949 * responsible for calling wake_up_sem_queue_do(@pt).
950 * It is safe to perform this call after dropping all locks.
Manfred Spraulfd5db422010-05-26 14:43:40 -0700951 */
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700952static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops,
953 int otime, struct list_head *pt)
Manfred Spraulfd5db422010-05-26 14:43:40 -0700954{
955 int i;
956
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700957 otime |= do_smart_wakeup_zero(sma, sops, nsops, pt);
958
Manfred Spraulf269f402013-07-08 16:01:24 -0700959 if (!list_empty(&sma->pending_alter)) {
960 /* semaphore array uses the global queue - just process it. */
961 otime |= update_queue(sma, -1, pt);
962 } else {
963 if (!sops) {
964 /*
965 * No sops, thus the modified semaphores are not
966 * known. Check all.
967 */
968 for (i = 0; i < sma->sem_nsems; i++)
969 otime |= update_queue(sma, i, pt);
970 } else {
971 /*
972 * Check the semaphores that were increased:
973 * - No complex ops, thus all sleeping ops are
974 * decrease.
975 * - if we decreased the value, then any sleeping
976 * semaphore ops wont be able to run: If the
977 * previous value was too small, then the new
978 * value will be too small, too.
979 */
980 for (i = 0; i < nsops; i++) {
981 if (sops[i].sem_op > 0) {
982 otime |= update_queue(sma,
983 sops[i].sem_num, pt);
984 }
Manfred Spraulab465df2013-05-26 11:08:52 +0200985 }
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700986 }
Manfred Spraulfd5db422010-05-26 14:43:40 -0700987 }
Manfred Spraul0e8c6652013-09-30 13:45:25 -0700988 if (otime)
989 set_semotime(sma, sops);
Manfred Spraulfd5db422010-05-26 14:43:40 -0700990}
991
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992/* The following counts are associated to each semaphore:
993 * semncnt number of tasks waiting on semval being nonzero
994 * semzcnt number of tasks waiting on semval being zero
995 * This model assumes that a task waits on exactly one semaphore.
996 * Since semaphore operations are to be performed atomically, tasks actually
997 * wait on a whole sequence of semaphores simultaneously.
998 * The counts we return here are a rough approximation, but still
999 * warrant that semncnt+semzcnt>0 if the task is on the pending queue.
1000 */
Manfred Spraul239521f2014-01-27 17:07:04 -08001001static int count_semncnt(struct sem_array *sma, ushort semnum)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002{
1003 int semncnt;
Manfred Spraul239521f2014-01-27 17:07:04 -08001004 struct sem_queue *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005
1006 semncnt = 0;
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001007 list_for_each_entry(q, &sma->sem_base[semnum].pending_alter, list) {
Manfred Spraul239521f2014-01-27 17:07:04 -08001008 struct sembuf *sops = q->sops;
Rik van Rielde2657f2013-05-09 16:59:59 -04001009 BUG_ON(sops->sem_num != semnum);
1010 if ((sops->sem_op < 0) && !(sops->sem_flg & IPC_NOWAIT))
1011 semncnt++;
1012 }
1013
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001014 list_for_each_entry(q, &sma->pending_alter, list) {
Manfred Spraul239521f2014-01-27 17:07:04 -08001015 struct sembuf *sops = q->sops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016 int nsops = q->nsops;
1017 int i;
1018 for (i = 0; i < nsops; i++)
1019 if (sops[i].sem_num == semnum
1020 && (sops[i].sem_op < 0)
1021 && !(sops[i].sem_flg & IPC_NOWAIT))
1022 semncnt++;
1023 }
1024 return semncnt;
1025}
Manfred Spraula1193f82008-07-25 01:48:06 -07001026
Manfred Spraul239521f2014-01-27 17:07:04 -08001027static int count_semzcnt(struct sem_array *sma, ushort semnum)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028{
1029 int semzcnt;
Manfred Spraul239521f2014-01-27 17:07:04 -08001030 struct sem_queue *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031
1032 semzcnt = 0;
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001033 list_for_each_entry(q, &sma->sem_base[semnum].pending_const, list) {
Manfred Spraul239521f2014-01-27 17:07:04 -08001034 struct sembuf *sops = q->sops;
Rik van Rielebc2e5e2013-05-09 16:53:28 -04001035 BUG_ON(sops->sem_num != semnum);
1036 if ((sops->sem_op == 0) && !(sops->sem_flg & IPC_NOWAIT))
1037 semzcnt++;
1038 }
1039
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001040 list_for_each_entry(q, &sma->pending_const, list) {
Manfred Spraul239521f2014-01-27 17:07:04 -08001041 struct sembuf *sops = q->sops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 int nsops = q->nsops;
1043 int i;
1044 for (i = 0; i < nsops; i++)
1045 if (sops[i].sem_num == semnum
1046 && (sops[i].sem_op == 0)
1047 && !(sops[i].sem_flg & IPC_NOWAIT))
1048 semzcnt++;
1049 }
1050 return semzcnt;
1051}
1052
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001053/* Free a semaphore set. freeary() is called with sem_ids.rwsem locked
1054 * as a writer and the spinlock for this semaphore set hold. sem_ids.rwsem
Nadia Derbey3e148c72007-10-18 23:40:54 -07001055 * remains locked on exit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 */
Pierre Peiffer01b8b072008-02-08 04:18:57 -08001057static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058{
Manfred Spraul380af1b2008-07-25 01:48:06 -07001059 struct sem_undo *un, *tu;
1060 struct sem_queue *q, *tq;
Pierre Peiffer01b8b072008-02-08 04:18:57 -08001061 struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001062 struct list_head tasks;
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001063 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064
Manfred Spraul380af1b2008-07-25 01:48:06 -07001065 /* Free the existing undo structures for this semaphore set. */
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -07001066 ipc_assert_locked_object(&sma->sem_perm);
Manfred Spraul380af1b2008-07-25 01:48:06 -07001067 list_for_each_entry_safe(un, tu, &sma->list_id, list_id) {
1068 list_del(&un->list_id);
1069 spin_lock(&un->ulp->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 un->semid = -1;
Manfred Spraul380af1b2008-07-25 01:48:06 -07001071 list_del_rcu(&un->list_proc);
1072 spin_unlock(&un->ulp->lock);
Lai Jiangshan693a8b62011-03-18 12:09:35 +08001073 kfree_rcu(un, rcu);
Manfred Spraul380af1b2008-07-25 01:48:06 -07001074 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075
1076 /* Wake up all pending processes and let them fail with EIDRM. */
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001077 INIT_LIST_HEAD(&tasks);
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001078 list_for_each_entry_safe(q, tq, &sma->pending_const, list) {
1079 unlink_queue(sma, q);
1080 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1081 }
1082
1083 list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
Manfred Spraulb97e8202009-12-15 16:47:32 -08001084 unlink_queue(sma, q);
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001085 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 }
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001087 for (i = 0; i < sma->sem_nsems; i++) {
1088 struct sem *sem = sma->sem_base + i;
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001089 list_for_each_entry_safe(q, tq, &sem->pending_const, list) {
1090 unlink_queue(sma, q);
1091 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1092 }
1093 list_for_each_entry_safe(q, tq, &sem->pending_alter, list) {
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001094 unlink_queue(sma, q);
1095 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1096 }
1097 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098
Nadia Derbey7ca7e562007-10-18 23:40:48 -07001099 /* Remove the semaphore set from the IDR */
1100 sem_rmid(ns, sma);
Rik van Riel6062a8d2013-04-30 19:15:44 -07001101 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001102 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001104 wake_up_sem_queue_do(&tasks);
Kirill Korotaeve3893532006-10-02 02:18:22 -07001105 ns->used_sems -= sma->sem_nsems;
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -07001106 ipc_rcu_putref(sma, sem_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107}
1108
1109static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
1110{
Manfred Spraul239521f2014-01-27 17:07:04 -08001111 switch (version) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112 case IPC_64:
1113 return copy_to_user(buf, in, sizeof(*in));
1114 case IPC_OLD:
1115 {
1116 struct semid_ds out;
1117
Dan Rosenberg982f7c22010-09-30 15:15:31 -07001118 memset(&out, 0, sizeof(out));
1119
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
1121
1122 out.sem_otime = in->sem_otime;
1123 out.sem_ctime = in->sem_ctime;
1124 out.sem_nsems = in->sem_nsems;
1125
1126 return copy_to_user(buf, &out, sizeof(out));
1127 }
1128 default:
1129 return -EINVAL;
1130 }
1131}
1132
Manfred Sprauld12e1e52013-07-08 16:01:25 -07001133static time_t get_semotime(struct sem_array *sma)
1134{
1135 int i;
1136 time_t res;
1137
1138 res = sma->sem_base[0].sem_otime;
1139 for (i = 1; i < sma->sem_nsems; i++) {
1140 time_t to = sma->sem_base[i].sem_otime;
1141
1142 if (to > res)
1143 res = to;
1144 }
1145 return res;
1146}
1147
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -08001148static int semctl_nolock(struct ipc_namespace *ns, int semid,
Al Viroe1fd1f42013-03-05 15:04:55 -05001149 int cmd, int version, void __user *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150{
Amerigo Wange5cc9c72009-12-15 16:47:35 -08001151 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 struct sem_array *sma;
1153
Manfred Spraul239521f2014-01-27 17:07:04 -08001154 switch (cmd) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155 case IPC_INFO:
1156 case SEM_INFO:
1157 {
1158 struct seminfo seminfo;
1159 int max_id;
1160
1161 err = security_sem_semctl(NULL, cmd);
1162 if (err)
1163 return err;
1164
Manfred Spraul239521f2014-01-27 17:07:04 -08001165 memset(&seminfo, 0, sizeof(seminfo));
Kirill Korotaeve3893532006-10-02 02:18:22 -07001166 seminfo.semmni = ns->sc_semmni;
1167 seminfo.semmns = ns->sc_semmns;
1168 seminfo.semmsl = ns->sc_semmsl;
1169 seminfo.semopm = ns->sc_semopm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 seminfo.semvmx = SEMVMX;
1171 seminfo.semmnu = SEMMNU;
1172 seminfo.semmap = SEMMAP;
1173 seminfo.semume = SEMUME;
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001174 down_read(&sem_ids(ns).rwsem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 if (cmd == SEM_INFO) {
Kirill Korotaeve3893532006-10-02 02:18:22 -07001176 seminfo.semusz = sem_ids(ns).in_use;
1177 seminfo.semaem = ns->used_sems;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 } else {
1179 seminfo.semusz = SEMUSZ;
1180 seminfo.semaem = SEMAEM;
1181 }
Nadia Derbey7ca7e562007-10-18 23:40:48 -07001182 max_id = ipc_get_maxid(&sem_ids(ns));
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001183 up_read(&sem_ids(ns).rwsem);
Al Viroe1fd1f42013-03-05 15:04:55 -05001184 if (copy_to_user(p, &seminfo, sizeof(struct seminfo)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185 return -EFAULT;
Manfred Spraul239521f2014-01-27 17:07:04 -08001186 return (max_id < 0) ? 0 : max_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187 }
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -08001188 case IPC_STAT:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189 case SEM_STAT:
1190 {
1191 struct semid64_ds tbuf;
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001192 int id = 0;
1193
1194 memset(&tbuf, 0, sizeof(tbuf));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195
Linus Torvalds941b0302013-05-04 11:04:29 -07001196 rcu_read_lock();
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -08001197 if (cmd == SEM_STAT) {
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001198 sma = sem_obtain_object(ns, semid);
1199 if (IS_ERR(sma)) {
1200 err = PTR_ERR(sma);
1201 goto out_unlock;
1202 }
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -08001203 id = sma->sem_perm.id;
1204 } else {
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001205 sma = sem_obtain_object_check(ns, semid);
1206 if (IS_ERR(sma)) {
1207 err = PTR_ERR(sma);
1208 goto out_unlock;
1209 }
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -08001210 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211
1212 err = -EACCES;
Serge E. Hallynb0e77592011-03-23 16:43:24 -07001213 if (ipcperms(ns, &sma->sem_perm, S_IRUGO))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 goto out_unlock;
1215
1216 err = security_sem_semctl(sma, cmd);
1217 if (err)
1218 goto out_unlock;
1219
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
Manfred Sprauld12e1e52013-07-08 16:01:25 -07001221 tbuf.sem_otime = get_semotime(sma);
1222 tbuf.sem_ctime = sma->sem_ctime;
1223 tbuf.sem_nsems = sma->sem_nsems;
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001224 rcu_read_unlock();
Al Viroe1fd1f42013-03-05 15:04:55 -05001225 if (copy_semid_to_user(p, &tbuf, version))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 return -EFAULT;
1227 return id;
1228 }
1229 default:
1230 return -EINVAL;
1231 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232out_unlock:
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001233 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 return err;
1235}
1236
Al Viroe1fd1f42013-03-05 15:04:55 -05001237static int semctl_setval(struct ipc_namespace *ns, int semid, int semnum,
1238 unsigned long arg)
1239{
1240 struct sem_undo *un;
1241 struct sem_array *sma;
Manfred Spraul239521f2014-01-27 17:07:04 -08001242 struct sem *curr;
Al Viroe1fd1f42013-03-05 15:04:55 -05001243 int err;
Al Viroe1fd1f42013-03-05 15:04:55 -05001244 struct list_head tasks;
1245 int val;
1246#if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN)
1247 /* big-endian 64bit */
1248 val = arg >> 32;
1249#else
1250 /* 32bit or little-endian 64bit */
1251 val = arg;
1252#endif
1253
Rik van Riel6062a8d2013-04-30 19:15:44 -07001254 if (val > SEMVMX || val < 0)
1255 return -ERANGE;
Al Viroe1fd1f42013-03-05 15:04:55 -05001256
1257 INIT_LIST_HEAD(&tasks);
Al Viroe1fd1f42013-03-05 15:04:55 -05001258
Rik van Riel6062a8d2013-04-30 19:15:44 -07001259 rcu_read_lock();
1260 sma = sem_obtain_object_check(ns, semid);
1261 if (IS_ERR(sma)) {
1262 rcu_read_unlock();
1263 return PTR_ERR(sma);
1264 }
1265
1266 if (semnum < 0 || semnum >= sma->sem_nsems) {
1267 rcu_read_unlock();
1268 return -EINVAL;
1269 }
1270
1271
1272 if (ipcperms(ns, &sma->sem_perm, S_IWUGO)) {
1273 rcu_read_unlock();
1274 return -EACCES;
1275 }
Al Viroe1fd1f42013-03-05 15:04:55 -05001276
1277 err = security_sem_semctl(sma, SETVAL);
Rik van Riel6062a8d2013-04-30 19:15:44 -07001278 if (err) {
1279 rcu_read_unlock();
1280 return -EACCES;
1281 }
Al Viroe1fd1f42013-03-05 15:04:55 -05001282
Rik van Riel6062a8d2013-04-30 19:15:44 -07001283 sem_lock(sma, NULL, -1);
Al Viroe1fd1f42013-03-05 15:04:55 -05001284
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001285 if (!ipc_valid_object(&sma->sem_perm)) {
Manfred Spraul6e224f92013-10-16 13:46:45 -07001286 sem_unlock(sma, -1);
1287 rcu_read_unlock();
1288 return -EIDRM;
1289 }
1290
Al Viroe1fd1f42013-03-05 15:04:55 -05001291 curr = &sma->sem_base[semnum];
1292
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -07001293 ipc_assert_locked_object(&sma->sem_perm);
Al Viroe1fd1f42013-03-05 15:04:55 -05001294 list_for_each_entry(un, &sma->list_id, list_id)
1295 un->semadj[semnum] = 0;
1296
1297 curr->semval = val;
1298 curr->sempid = task_tgid_vnr(current);
1299 sma->sem_ctime = get_seconds();
1300 /* maybe some queued-up processes were waiting for this */
1301 do_smart_update(sma, NULL, 0, 0, &tasks);
Rik van Riel6062a8d2013-04-30 19:15:44 -07001302 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001303 rcu_read_unlock();
Al Viroe1fd1f42013-03-05 15:04:55 -05001304 wake_up_sem_queue_do(&tasks);
Rik van Riel6062a8d2013-04-30 19:15:44 -07001305 return 0;
Al Viroe1fd1f42013-03-05 15:04:55 -05001306}
1307
Kirill Korotaeve3893532006-10-02 02:18:22 -07001308static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
Al Viroe1fd1f42013-03-05 15:04:55 -05001309 int cmd, void __user *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310{
1311 struct sem_array *sma;
Manfred Spraul239521f2014-01-27 17:07:04 -08001312 struct sem *curr;
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001313 int err, nsems;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314 ushort fast_sem_io[SEMMSL_FAST];
Manfred Spraul239521f2014-01-27 17:07:04 -08001315 ushort *sem_io = fast_sem_io;
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001316 struct list_head tasks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001318 INIT_LIST_HEAD(&tasks);
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001319
1320 rcu_read_lock();
1321 sma = sem_obtain_object_check(ns, semid);
1322 if (IS_ERR(sma)) {
1323 rcu_read_unlock();
1324 return PTR_ERR(sma);
1325 }
1326
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327 nsems = sma->sem_nsems;
1328
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329 err = -EACCES;
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001330 if (ipcperms(ns, &sma->sem_perm, cmd == SETALL ? S_IWUGO : S_IRUGO))
1331 goto out_rcu_wakeup;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332
1333 err = security_sem_semctl(sma, cmd);
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001334 if (err)
1335 goto out_rcu_wakeup;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336
1337 err = -EACCES;
1338 switch (cmd) {
1339 case GETALL:
1340 {
Al Viroe1fd1f42013-03-05 15:04:55 -05001341 ushort __user *array = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 int i;
1343
Al Viroce857222013-05-03 00:30:49 +01001344 sem_lock(sma, NULL, -1);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001345 if (!ipc_valid_object(&sma->sem_perm)) {
Manfred Spraul6e224f92013-10-16 13:46:45 -07001346 err = -EIDRM;
1347 goto out_unlock;
1348 }
Manfred Spraul239521f2014-01-27 17:07:04 -08001349 if (nsems > SEMMSL_FAST) {
Al Viroce857222013-05-03 00:30:49 +01001350 if (!ipc_rcu_getref(sma)) {
Al Viroce857222013-05-03 00:30:49 +01001351 err = -EIDRM;
Manfred Spraul6e224f92013-10-16 13:46:45 -07001352 goto out_unlock;
Al Viroce857222013-05-03 00:30:49 +01001353 }
1354 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001355 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 sem_io = ipc_alloc(sizeof(ushort)*nsems);
Manfred Spraul239521f2014-01-27 17:07:04 -08001357 if (sem_io == NULL) {
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -07001358 ipc_rcu_putref(sma, ipc_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359 return -ENOMEM;
1360 }
1361
Linus Torvalds4091fd92013-05-04 10:13:40 -07001362 rcu_read_lock();
Pierre Peiffer6ff37972008-04-29 01:00:46 -07001363 sem_lock_and_putref(sma);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001364 if (!ipc_valid_object(&sma->sem_perm)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 err = -EIDRM;
Manfred Spraul6e224f92013-10-16 13:46:45 -07001366 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367 }
Al Viroce857222013-05-03 00:30:49 +01001368 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 for (i = 0; i < sma->sem_nsems; i++)
1370 sem_io[i] = sma->sem_base[i].semval;
Rik van Riel6062a8d2013-04-30 19:15:44 -07001371 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001372 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 err = 0;
Manfred Spraul239521f2014-01-27 17:07:04 -08001374 if (copy_to_user(array, sem_io, nsems*sizeof(ushort)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375 err = -EFAULT;
1376 goto out_free;
1377 }
1378 case SETALL:
1379 {
1380 int i;
1381 struct sem_undo *un;
1382
Rik van Riel6062a8d2013-04-30 19:15:44 -07001383 if (!ipc_rcu_getref(sma)) {
Manfred Spraul6e224f92013-10-16 13:46:45 -07001384 err = -EIDRM;
1385 goto out_rcu_wakeup;
Rik van Riel6062a8d2013-04-30 19:15:44 -07001386 }
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001387 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388
Manfred Spraul239521f2014-01-27 17:07:04 -08001389 if (nsems > SEMMSL_FAST) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390 sem_io = ipc_alloc(sizeof(ushort)*nsems);
Manfred Spraul239521f2014-01-27 17:07:04 -08001391 if (sem_io == NULL) {
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -07001392 ipc_rcu_putref(sma, ipc_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393 return -ENOMEM;
1394 }
1395 }
1396
Manfred Spraul239521f2014-01-27 17:07:04 -08001397 if (copy_from_user(sem_io, p, nsems*sizeof(ushort))) {
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -07001398 ipc_rcu_putref(sma, ipc_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 err = -EFAULT;
1400 goto out_free;
1401 }
1402
1403 for (i = 0; i < nsems; i++) {
1404 if (sem_io[i] > SEMVMX) {
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -07001405 ipc_rcu_putref(sma, ipc_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 err = -ERANGE;
1407 goto out_free;
1408 }
1409 }
Linus Torvalds4091fd92013-05-04 10:13:40 -07001410 rcu_read_lock();
Pierre Peiffer6ff37972008-04-29 01:00:46 -07001411 sem_lock_and_putref(sma);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001412 if (!ipc_valid_object(&sma->sem_perm)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 err = -EIDRM;
Manfred Spraul6e224f92013-10-16 13:46:45 -07001414 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415 }
1416
1417 for (i = 0; i < nsems; i++)
1418 sma->sem_base[i].semval = sem_io[i];
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001419
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -07001420 ipc_assert_locked_object(&sma->sem_perm);
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001421 list_for_each_entry(un, &sma->list_id, list_id) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422 for (i = 0; i < nsems; i++)
1423 un->semadj[i] = 0;
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001424 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425 sma->sem_ctime = get_seconds();
1426 /* maybe some queued-up processes were waiting for this */
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001427 do_smart_update(sma, NULL, 0, 0, &tasks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428 err = 0;
1429 goto out_unlock;
1430 }
Al Viroe1fd1f42013-03-05 15:04:55 -05001431 /* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432 }
1433 err = -EINVAL;
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001434 if (semnum < 0 || semnum >= nsems)
1435 goto out_rcu_wakeup;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436
Rik van Riel6062a8d2013-04-30 19:15:44 -07001437 sem_lock(sma, NULL, -1);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001438 if (!ipc_valid_object(&sma->sem_perm)) {
Manfred Spraul6e224f92013-10-16 13:46:45 -07001439 err = -EIDRM;
1440 goto out_unlock;
1441 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442 curr = &sma->sem_base[semnum];
1443
1444 switch (cmd) {
1445 case GETVAL:
1446 err = curr->semval;
1447 goto out_unlock;
1448 case GETPID:
1449 err = curr->sempid;
1450 goto out_unlock;
1451 case GETNCNT:
Manfred Spraul239521f2014-01-27 17:07:04 -08001452 err = count_semncnt(sma, semnum);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 goto out_unlock;
1454 case GETZCNT:
Manfred Spraul239521f2014-01-27 17:07:04 -08001455 err = count_semzcnt(sma, semnum);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 }
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001458
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459out_unlock:
Rik van Riel6062a8d2013-04-30 19:15:44 -07001460 sem_unlock(sma, -1);
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001461out_rcu_wakeup:
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001462 rcu_read_unlock();
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001463 wake_up_sem_queue_do(&tasks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464out_free:
Manfred Spraul239521f2014-01-27 17:07:04 -08001465 if (sem_io != fast_sem_io)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466 ipc_free(sem_io, sizeof(ushort)*nsems);
1467 return err;
1468}
1469
Pierre Peiffer016d7132008-04-29 01:00:50 -07001470static inline unsigned long
1471copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472{
Manfred Spraul239521f2014-01-27 17:07:04 -08001473 switch (version) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 case IPC_64:
Pierre Peiffer016d7132008-04-29 01:00:50 -07001475 if (copy_from_user(out, buf, sizeof(*out)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478 case IPC_OLD:
1479 {
1480 struct semid_ds tbuf_old;
1481
Manfred Spraul239521f2014-01-27 17:07:04 -08001482 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 return -EFAULT;
1484
Pierre Peiffer016d7132008-04-29 01:00:50 -07001485 out->sem_perm.uid = tbuf_old.sem_perm.uid;
1486 out->sem_perm.gid = tbuf_old.sem_perm.gid;
1487 out->sem_perm.mode = tbuf_old.sem_perm.mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488
1489 return 0;
1490 }
1491 default:
1492 return -EINVAL;
1493 }
1494}
1495
Pierre Peiffer522bb2a2008-04-29 01:00:49 -07001496/*
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001497 * This function handles some semctl commands which require the rwsem
Pierre Peiffer522bb2a2008-04-29 01:00:49 -07001498 * to be held in write mode.
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001499 * NOTE: no locks must be held, the rwsem is taken inside this function.
Pierre Peiffer522bb2a2008-04-29 01:00:49 -07001500 */
Pierre Peiffer21a48262008-04-29 01:00:49 -07001501static int semctl_down(struct ipc_namespace *ns, int semid,
Al Viroe1fd1f42013-03-05 15:04:55 -05001502 int cmd, int version, void __user *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503{
1504 struct sem_array *sma;
1505 int err;
Pierre Peiffer016d7132008-04-29 01:00:50 -07001506 struct semid64_ds semid64;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 struct kern_ipc_perm *ipcp;
1508
Manfred Spraul239521f2014-01-27 17:07:04 -08001509 if (cmd == IPC_SET) {
Al Viroe1fd1f42013-03-05 15:04:55 -05001510 if (copy_semid_from_user(&semid64, p, version))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001514 down_write(&sem_ids(ns).rwsem);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001515 rcu_read_lock();
1516
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001517 ipcp = ipcctl_pre_down_nolock(ns, &sem_ids(ns), semid, cmd,
1518 &semid64.sem_perm, 0);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001519 if (IS_ERR(ipcp)) {
1520 err = PTR_ERR(ipcp);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001521 goto out_unlock1;
1522 }
Steve Grubb073115d2006-04-02 17:07:33 -04001523
Pierre Peiffera5f75e72008-04-29 01:00:54 -07001524 sma = container_of(ipcp, struct sem_array, sem_perm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525
1526 err = security_sem_semctl(sma, cmd);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001527 if (err)
1528 goto out_unlock1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001530 switch (cmd) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531 case IPC_RMID:
Rik van Riel6062a8d2013-04-30 19:15:44 -07001532 sem_lock(sma, NULL, -1);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001533 /* freeary unlocks the ipc object and rcu */
Pierre Peiffer01b8b072008-02-08 04:18:57 -08001534 freeary(ns, ipcp);
Pierre Peiffer522bb2a2008-04-29 01:00:49 -07001535 goto out_up;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536 case IPC_SET:
Rik van Riel6062a8d2013-04-30 19:15:44 -07001537 sem_lock(sma, NULL, -1);
Eric W. Biederman1efdb692012-02-07 16:54:11 -08001538 err = ipc_update_perm(&semid64.sem_perm, ipcp);
1539 if (err)
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001540 goto out_unlock0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 sma->sem_ctime = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542 break;
1543 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 err = -EINVAL;
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001545 goto out_unlock1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001548out_unlock0:
Rik van Riel6062a8d2013-04-30 19:15:44 -07001549 sem_unlock(sma, -1);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001550out_unlock1:
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001551 rcu_read_unlock();
Pierre Peiffer522bb2a2008-04-29 01:00:49 -07001552out_up:
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001553 up_write(&sem_ids(ns).rwsem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554 return err;
1555}
1556
Al Viroe1fd1f42013-03-05 15:04:55 -05001557SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, unsigned long, arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559 int version;
Kirill Korotaeve3893532006-10-02 02:18:22 -07001560 struct ipc_namespace *ns;
Al Viroe1fd1f42013-03-05 15:04:55 -05001561 void __user *p = (void __user *)arg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562
1563 if (semid < 0)
1564 return -EINVAL;
1565
1566 version = ipc_parse_version(&cmd);
Kirill Korotaeve3893532006-10-02 02:18:22 -07001567 ns = current->nsproxy->ipc_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568
Manfred Spraul239521f2014-01-27 17:07:04 -08001569 switch (cmd) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570 case IPC_INFO:
1571 case SEM_INFO:
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -08001572 case IPC_STAT:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573 case SEM_STAT:
Al Viroe1fd1f42013-03-05 15:04:55 -05001574 return semctl_nolock(ns, semid, cmd, version, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575 case GETALL:
1576 case GETVAL:
1577 case GETPID:
1578 case GETNCNT:
1579 case GETZCNT:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 case SETALL:
Al Viroe1fd1f42013-03-05 15:04:55 -05001581 return semctl_main(ns, semid, semnum, cmd, p);
1582 case SETVAL:
1583 return semctl_setval(ns, semid, semnum, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584 case IPC_RMID:
1585 case IPC_SET:
Al Viroe1fd1f42013-03-05 15:04:55 -05001586 return semctl_down(ns, semid, cmd, version, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587 default:
1588 return -EINVAL;
1589 }
1590}
1591
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592/* If the task doesn't already have a undo_list, then allocate one
1593 * here. We guarantee there is only one thread using this undo list,
1594 * and current is THE ONE
1595 *
1596 * If this allocation and assignment succeeds, but later
1597 * portions of this code fail, there is no need to free the sem_undo_list.
1598 * Just let it stay associated with the task, and it'll be freed later
1599 * at exit time.
1600 *
1601 * This can block, so callers must hold no locks.
1602 */
1603static inline int get_undo_list(struct sem_undo_list **undo_listp)
1604{
1605 struct sem_undo_list *undo_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606
1607 undo_list = current->sysvsem.undo_list;
1608 if (!undo_list) {
Matt Helsley2453a302006-10-02 02:18:25 -07001609 undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610 if (undo_list == NULL)
1611 return -ENOMEM;
Ingo Molnar00a5dfd2005-08-05 23:05:27 +02001612 spin_lock_init(&undo_list->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613 atomic_set(&undo_list->refcnt, 1);
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001614 INIT_LIST_HEAD(&undo_list->list_proc);
1615
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616 current->sysvsem.undo_list = undo_list;
1617 }
1618 *undo_listp = undo_list;
1619 return 0;
1620}
1621
Nick Pigginbf17bb72009-12-15 16:47:28 -08001622static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623{
Nick Pigginbf17bb72009-12-15 16:47:28 -08001624 struct sem_undo *un;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625
Nick Pigginbf17bb72009-12-15 16:47:28 -08001626 list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) {
1627 if (un->semid == semid)
1628 return un;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629 }
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001630 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631}
1632
Nick Pigginbf17bb72009-12-15 16:47:28 -08001633static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1634{
1635 struct sem_undo *un;
1636
Manfred Spraul239521f2014-01-27 17:07:04 -08001637 assert_spin_locked(&ulp->lock);
Nick Pigginbf17bb72009-12-15 16:47:28 -08001638
1639 un = __lookup_undo(ulp, semid);
1640 if (un) {
1641 list_del_rcu(&un->list_proc);
1642 list_add_rcu(&un->list_proc, &ulp->list_proc);
1643 }
1644 return un;
1645}
1646
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001647/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -08001648 * find_alloc_undo - lookup (and if not present create) undo array
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001649 * @ns: namespace
1650 * @semid: semaphore array id
1651 *
1652 * The function looks up (and if not present creates) the undo structure.
1653 * The size of the undo structure depends on the size of the semaphore
1654 * array, thus the alloc path is not that straightforward.
Manfred Spraul380af1b2008-07-25 01:48:06 -07001655 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1656 * performs a rcu_read_lock().
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001657 */
1658static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659{
1660 struct sem_array *sma;
1661 struct sem_undo_list *ulp;
1662 struct sem_undo *un, *new;
Rik van Riel6062a8d2013-04-30 19:15:44 -07001663 int nsems, error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664
1665 error = get_undo_list(&ulp);
1666 if (error)
1667 return ERR_PTR(error);
1668
Manfred Spraul380af1b2008-07-25 01:48:06 -07001669 rcu_read_lock();
Pierre Peifferc530c6a2007-10-18 23:40:55 -07001670 spin_lock(&ulp->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671 un = lookup_undo(ulp, semid);
Pierre Peifferc530c6a2007-10-18 23:40:55 -07001672 spin_unlock(&ulp->lock);
Manfred Spraul239521f2014-01-27 17:07:04 -08001673 if (likely(un != NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674 goto out;
1675
1676 /* no undo structure around - allocate one. */
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001677 /* step 1: figure out the size of the semaphore array */
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001678 sma = sem_obtain_object_check(ns, semid);
1679 if (IS_ERR(sma)) {
1680 rcu_read_unlock();
Julia Lawall4de85cd2010-05-26 14:43:44 -07001681 return ERR_CAST(sma);
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001682 }
Nadia Derbey023a5352007-10-18 23:40:51 -07001683
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684 nsems = sma->sem_nsems;
Rik van Riel6062a8d2013-04-30 19:15:44 -07001685 if (!ipc_rcu_getref(sma)) {
1686 rcu_read_unlock();
1687 un = ERR_PTR(-EIDRM);
1688 goto out;
1689 }
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001690 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001692 /* step 2: allocate new undo structure */
Burman Yan4668edc2006-12-06 20:38:51 -08001693 new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694 if (!new) {
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -07001695 ipc_rcu_putref(sma, ipc_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696 return ERR_PTR(-ENOMEM);
1697 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698
Manfred Spraul380af1b2008-07-25 01:48:06 -07001699 /* step 3: Acquire the lock on semaphore array */
Linus Torvalds4091fd92013-05-04 10:13:40 -07001700 rcu_read_lock();
Pierre Peiffer6ff37972008-04-29 01:00:46 -07001701 sem_lock_and_putref(sma);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001702 if (!ipc_valid_object(&sma->sem_perm)) {
Rik van Riel6062a8d2013-04-30 19:15:44 -07001703 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001704 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705 kfree(new);
1706 un = ERR_PTR(-EIDRM);
1707 goto out;
1708 }
Manfred Spraul380af1b2008-07-25 01:48:06 -07001709 spin_lock(&ulp->lock);
1710
1711 /*
1712 * step 4: check for races: did someone else allocate the undo struct?
1713 */
1714 un = lookup_undo(ulp, semid);
1715 if (un) {
1716 kfree(new);
1717 goto success;
1718 }
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001719 /* step 5: initialize & link new undo structure */
1720 new->semadj = (short *) &new[1];
Manfred Spraul380af1b2008-07-25 01:48:06 -07001721 new->ulp = ulp;
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001722 new->semid = semid;
1723 assert_spin_locked(&ulp->lock);
Manfred Spraul380af1b2008-07-25 01:48:06 -07001724 list_add_rcu(&new->list_proc, &ulp->list_proc);
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -07001725 ipc_assert_locked_object(&sma->sem_perm);
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001726 list_add(&new->list_id, &sma->list_id);
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001727 un = new;
Manfred Spraul380af1b2008-07-25 01:48:06 -07001728
1729success:
1730 spin_unlock(&ulp->lock);
Rik van Riel6062a8d2013-04-30 19:15:44 -07001731 sem_unlock(sma, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732out:
1733 return un;
1734}
1735
Manfred Spraulc61284e2010-07-20 13:24:23 -07001736
1737/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -08001738 * get_queue_result - retrieve the result code from sem_queue
Manfred Spraulc61284e2010-07-20 13:24:23 -07001739 * @q: Pointer to queue structure
1740 *
1741 * Retrieve the return code from the pending queue. If IN_WAKEUP is found in
1742 * q->status, then we must loop until the value is replaced with the final
1743 * value: This may happen if a task is woken up by an unrelated event (e.g.
1744 * signal) and in parallel the task is woken up by another task because it got
1745 * the requested semaphores.
1746 *
1747 * The function can be called with or without holding the semaphore spinlock.
1748 */
1749static int get_queue_result(struct sem_queue *q)
1750{
1751 int error;
1752
1753 error = q->status;
1754 while (unlikely(error == IN_WAKEUP)) {
1755 cpu_relax();
1756 error = q->status;
1757 }
1758
1759 return error;
1760}
1761
Heiko Carstensd5460c92009-01-14 14:14:27 +01001762SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
1763 unsigned, nsops, const struct timespec __user *, timeout)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764{
1765 int error = -EINVAL;
1766 struct sem_array *sma;
1767 struct sembuf fast_sops[SEMOPM_FAST];
Manfred Spraul239521f2014-01-27 17:07:04 -08001768 struct sembuf *sops = fast_sops, *sop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769 struct sem_undo *un;
Rik van Riel6062a8d2013-04-30 19:15:44 -07001770 int undos = 0, alter = 0, max, locknum;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771 struct sem_queue queue;
1772 unsigned long jiffies_left = 0;
Kirill Korotaeve3893532006-10-02 02:18:22 -07001773 struct ipc_namespace *ns;
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001774 struct list_head tasks;
Kirill Korotaeve3893532006-10-02 02:18:22 -07001775
1776 ns = current->nsproxy->ipc_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777
1778 if (nsops < 1 || semid < 0)
1779 return -EINVAL;
Kirill Korotaeve3893532006-10-02 02:18:22 -07001780 if (nsops > ns->sc_semopm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781 return -E2BIG;
Manfred Spraul239521f2014-01-27 17:07:04 -08001782 if (nsops > SEMOPM_FAST) {
1783 sops = kmalloc(sizeof(*sops)*nsops, GFP_KERNEL);
1784 if (sops == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785 return -ENOMEM;
1786 }
Manfred Spraul239521f2014-01-27 17:07:04 -08001787 if (copy_from_user(sops, tsops, nsops * sizeof(*tsops))) {
1788 error = -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789 goto out_free;
1790 }
1791 if (timeout) {
1792 struct timespec _timeout;
1793 if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1794 error = -EFAULT;
1795 goto out_free;
1796 }
1797 if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1798 _timeout.tv_nsec >= 1000000000L) {
1799 error = -EINVAL;
1800 goto out_free;
1801 }
1802 jiffies_left = timespec_to_jiffies(&_timeout);
1803 }
1804 max = 0;
1805 for (sop = sops; sop < sops + nsops; sop++) {
1806 if (sop->sem_num >= max)
1807 max = sop->sem_num;
1808 if (sop->sem_flg & SEM_UNDO)
Manfred Spraulb78755a2005-06-23 00:10:06 -07001809 undos = 1;
1810 if (sop->sem_op != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811 alter = 1;
1812 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813
Rik van Riel6062a8d2013-04-30 19:15:44 -07001814 INIT_LIST_HEAD(&tasks);
1815
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816 if (undos) {
Rik van Riel6062a8d2013-04-30 19:15:44 -07001817 /* On success, find_alloc_undo takes the rcu_read_lock */
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001818 un = find_alloc_undo(ns, semid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819 if (IS_ERR(un)) {
1820 error = PTR_ERR(un);
1821 goto out_free;
1822 }
Rik van Riel6062a8d2013-04-30 19:15:44 -07001823 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824 un = NULL;
Rik van Riel6062a8d2013-04-30 19:15:44 -07001825 rcu_read_lock();
1826 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001828 sma = sem_obtain_object_check(ns, semid);
Nadia Derbey023a5352007-10-18 23:40:51 -07001829 if (IS_ERR(sma)) {
Rik van Riel6062a8d2013-04-30 19:15:44 -07001830 rcu_read_unlock();
Nadia Derbey023a5352007-10-18 23:40:51 -07001831 error = PTR_ERR(sma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832 goto out_free;
Nadia Derbey023a5352007-10-18 23:40:51 -07001833 }
1834
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001835 error = -EFBIG;
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001836 if (max >= sma->sem_nsems)
1837 goto out_rcu_wakeup;
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001838
1839 error = -EACCES;
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001840 if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO))
1841 goto out_rcu_wakeup;
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001842
1843 error = security_sem_semop(sma, sops, nsops, alter);
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001844 if (error)
1845 goto out_rcu_wakeup;
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001846
Manfred Spraul6e224f92013-10-16 13:46:45 -07001847 error = -EIDRM;
1848 locknum = sem_lock(sma, sops, nsops);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001849 /*
1850 * We eventually might perform the following check in a lockless
1851 * fashion, considering ipc_valid_object() locking constraints.
1852 * If nsops == 1 and there is no contention for sem_perm.lock, then
1853 * only a per-semaphore lock is held and it's OK to proceed with the
1854 * check below. More details on the fine grained locking scheme
1855 * entangled here and why it's RMID race safe on comments at sem_lock()
1856 */
1857 if (!ipc_valid_object(&sma->sem_perm))
Manfred Spraul6e224f92013-10-16 13:46:45 -07001858 goto out_unlock_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001859 /*
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001860 * semid identifiers are not unique - find_alloc_undo may have
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861 * allocated an undo structure, it was invalidated by an RMID
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001862 * and now a new array with received the same id. Check and fail.
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001863 * This case can be detected checking un->semid. The existence of
Manfred Spraul380af1b2008-07-25 01:48:06 -07001864 * "un" itself is guaranteed by rcu.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865 */
Rik van Riel6062a8d2013-04-30 19:15:44 -07001866 if (un && un->semid == -1)
1867 goto out_unlock_free;
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001868
Manfred Spraul758a6ba32013-07-08 16:01:26 -07001869 error = perform_atomic_semop(sma, sops, nsops, un,
1870 task_tgid_vnr(current));
Manfred Spraul0e8c6652013-09-30 13:45:25 -07001871 if (error == 0) {
1872 /* If the operation was successful, then do
1873 * the required updates.
1874 */
1875 if (alter)
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001876 do_smart_update(sma, sops, nsops, 1, &tasks);
Manfred Spraul0e8c6652013-09-30 13:45:25 -07001877 else
1878 set_semotime(sma, sops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879 }
Manfred Spraul0e8c6652013-09-30 13:45:25 -07001880 if (error <= 0)
1881 goto out_unlock_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882
1883 /* We need to sleep on this operation, so we put the current
1884 * task into the pending queue and go to sleep.
1885 */
1886
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887 queue.sops = sops;
1888 queue.nsops = nsops;
1889 queue.undo = un;
Pavel Emelyanovb4888932007-10-18 23:40:14 -07001890 queue.pid = task_tgid_vnr(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891 queue.alter = alter;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892
Manfred Spraulb97e8202009-12-15 16:47:32 -08001893 if (nsops == 1) {
1894 struct sem *curr;
1895 curr = &sma->sem_base[sops->sem_num];
1896
Manfred Spraulf269f402013-07-08 16:01:24 -07001897 if (alter) {
1898 if (sma->complex_count) {
1899 list_add_tail(&queue.list,
1900 &sma->pending_alter);
1901 } else {
1902
1903 list_add_tail(&queue.list,
1904 &curr->pending_alter);
1905 }
1906 } else {
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001907 list_add_tail(&queue.list, &curr->pending_const);
Manfred Spraulf269f402013-07-08 16:01:24 -07001908 }
Manfred Spraulb97e8202009-12-15 16:47:32 -08001909 } else {
Manfred Spraulf269f402013-07-08 16:01:24 -07001910 if (!sma->complex_count)
1911 merge_queues(sma);
1912
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001913 if (alter)
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001914 list_add_tail(&queue.list, &sma->pending_alter);
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001915 else
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001916 list_add_tail(&queue.list, &sma->pending_const);
1917
Manfred Spraulb97e8202009-12-15 16:47:32 -08001918 sma->complex_count++;
1919 }
1920
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921 queue.status = -EINTR;
1922 queue.sleeper = current;
Manfred Spraul0b0577f2011-11-02 13:38:52 -07001923
1924sleep_again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925 current->state = TASK_INTERRUPTIBLE;
Rik van Riel6062a8d2013-04-30 19:15:44 -07001926 sem_unlock(sma, locknum);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001927 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928
1929 if (timeout)
1930 jiffies_left = schedule_timeout(jiffies_left);
1931 else
1932 schedule();
1933
Manfred Spraulc61284e2010-07-20 13:24:23 -07001934 error = get_queue_result(&queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935
1936 if (error != -EINTR) {
1937 /* fast path: update_queue already obtained all requested
Manfred Spraulc61284e2010-07-20 13:24:23 -07001938 * resources.
1939 * Perform a smp_mb(): User space could assume that semop()
1940 * is a memory barrier: Without the mb(), the cpu could
1941 * speculatively read in user space stale data that was
1942 * overwritten by the previous owner of the semaphore.
1943 */
1944 smp_mb();
1945
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 goto out_free;
1947 }
1948
Linus Torvalds321310c2013-05-04 10:47:57 -07001949 rcu_read_lock();
Rik van Riel6062a8d2013-04-30 19:15:44 -07001950 sma = sem_obtain_lock(ns, semid, sops, nsops, &locknum);
Manfred Sprauld694ad62011-07-25 17:11:47 -07001951
1952 /*
1953 * Wait until it's guaranteed that no wakeup_sem_queue_do() is ongoing.
1954 */
1955 error = get_queue_result(&queue);
1956
1957 /*
1958 * Array removed? If yes, leave without sem_unlock().
1959 */
Nadia Derbey023a5352007-10-18 23:40:51 -07001960 if (IS_ERR(sma)) {
Linus Torvalds321310c2013-05-04 10:47:57 -07001961 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962 goto out_free;
1963 }
1964
Manfred Spraulc61284e2010-07-20 13:24:23 -07001965
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966 /*
Manfred Sprauld694ad62011-07-25 17:11:47 -07001967 * If queue.status != -EINTR we are woken up by another process.
1968 * Leave without unlink_queue(), but with sem_unlock().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969 */
Davidlohr Bueso3ab08fe2014-01-27 17:07:06 -08001970 if (error != -EINTR)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971 goto out_unlock_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972
1973 /*
1974 * If an interrupt occurred we have to clean up the queue
1975 */
1976 if (timeout && jiffies_left == 0)
1977 error = -EAGAIN;
Manfred Spraul0b0577f2011-11-02 13:38:52 -07001978
1979 /*
1980 * If the wakeup was spurious, just retry
1981 */
1982 if (error == -EINTR && !signal_pending(current))
1983 goto sleep_again;
1984
Manfred Spraulb97e8202009-12-15 16:47:32 -08001985 unlink_queue(sma, &queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986
1987out_unlock_free:
Rik van Riel6062a8d2013-04-30 19:15:44 -07001988 sem_unlock(sma, locknum);
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001989out_rcu_wakeup:
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001990 rcu_read_unlock();
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001991 wake_up_sem_queue_do(&tasks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992out_free:
Manfred Spraul239521f2014-01-27 17:07:04 -08001993 if (sops != fast_sops)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994 kfree(sops);
1995 return error;
1996}
1997
Heiko Carstensd5460c92009-01-14 14:14:27 +01001998SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
1999 unsigned, nsops)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000{
2001 return sys_semtimedop(semid, tsops, nsops, NULL);
2002}
2003
2004/* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
2005 * parent and child tasks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006 */
2007
2008int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
2009{
2010 struct sem_undo_list *undo_list;
2011 int error;
2012
2013 if (clone_flags & CLONE_SYSVSEM) {
2014 error = get_undo_list(&undo_list);
2015 if (error)
2016 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017 atomic_inc(&undo_list->refcnt);
2018 tsk->sysvsem.undo_list = undo_list;
2019 } else
2020 tsk->sysvsem.undo_list = NULL;
2021
2022 return 0;
2023}
2024
2025/*
2026 * add semadj values to semaphores, free undo structures.
2027 * undo structures are not freed when semaphore arrays are destroyed
2028 * so some of them may be out of date.
2029 * IMPLEMENTATION NOTE: There is some confusion over whether the
2030 * set of adjustments that needs to be done should be done in an atomic
2031 * manner or not. That is, if we are attempting to decrement the semval
2032 * should we queue up and wait until we can do so legally?
2033 * The original implementation attempted to do this (queue and wait).
2034 * The current implementation does not do so. The POSIX standard
2035 * and SVID should be consulted to determine what behavior is mandated.
2036 */
2037void exit_sem(struct task_struct *tsk)
2038{
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002039 struct sem_undo_list *ulp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002041 ulp = tsk->sysvsem.undo_list;
2042 if (!ulp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043 return;
Manfred Spraul9edff4a2008-04-29 01:00:57 -07002044 tsk->sysvsem.undo_list = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002046 if (!atomic_dec_and_test(&ulp->refcnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 return;
2048
Manfred Spraul380af1b2008-07-25 01:48:06 -07002049 for (;;) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050 struct sem_array *sma;
Manfred Spraul380af1b2008-07-25 01:48:06 -07002051 struct sem_undo *un;
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07002052 struct list_head tasks;
Rik van Riel6062a8d2013-04-30 19:15:44 -07002053 int semid, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054
Manfred Spraul380af1b2008-07-25 01:48:06 -07002055 rcu_read_lock();
Jiri Pirko05725f72009-04-14 20:17:16 +02002056 un = list_entry_rcu(ulp->list_proc.next,
2057 struct sem_undo, list_proc);
Manfred Spraul380af1b2008-07-25 01:48:06 -07002058 if (&un->list_proc == &ulp->list_proc)
2059 semid = -1;
2060 else
2061 semid = un->semid;
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002062
Rik van Riel6062a8d2013-04-30 19:15:44 -07002063 if (semid == -1) {
2064 rcu_read_unlock();
Manfred Spraul380af1b2008-07-25 01:48:06 -07002065 break;
Rik van Riel6062a8d2013-04-30 19:15:44 -07002066 }
Manfred Spraul380af1b2008-07-25 01:48:06 -07002067
Rik van Riel6062a8d2013-04-30 19:15:44 -07002068 sma = sem_obtain_object_check(tsk->nsproxy->ipc_ns, un->semid);
Manfred Spraul380af1b2008-07-25 01:48:06 -07002069 /* exit_sem raced with IPC_RMID, nothing to do */
Rik van Riel6062a8d2013-04-30 19:15:44 -07002070 if (IS_ERR(sma)) {
2071 rcu_read_unlock();
Manfred Spraul380af1b2008-07-25 01:48:06 -07002072 continue;
Rik van Riel6062a8d2013-04-30 19:15:44 -07002073 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074
Rik van Riel6062a8d2013-04-30 19:15:44 -07002075 sem_lock(sma, NULL, -1);
Manfred Spraul6e224f92013-10-16 13:46:45 -07002076 /* exit_sem raced with IPC_RMID, nothing to do */
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08002077 if (!ipc_valid_object(&sma->sem_perm)) {
Manfred Spraul6e224f92013-10-16 13:46:45 -07002078 sem_unlock(sma, -1);
2079 rcu_read_unlock();
2080 continue;
2081 }
Nick Pigginbf17bb72009-12-15 16:47:28 -08002082 un = __lookup_undo(ulp, semid);
Manfred Spraul380af1b2008-07-25 01:48:06 -07002083 if (un == NULL) {
2084 /* exit_sem raced with IPC_RMID+semget() that created
2085 * exactly the same semid. Nothing to do.
2086 */
Rik van Riel6062a8d2013-04-30 19:15:44 -07002087 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07002088 rcu_read_unlock();
Manfred Spraul380af1b2008-07-25 01:48:06 -07002089 continue;
2090 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091
Manfred Spraul380af1b2008-07-25 01:48:06 -07002092 /* remove un from the linked lists */
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -07002093 ipc_assert_locked_object(&sma->sem_perm);
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002094 list_del(&un->list_id);
2095
Manfred Spraul380af1b2008-07-25 01:48:06 -07002096 spin_lock(&ulp->lock);
2097 list_del_rcu(&un->list_proc);
2098 spin_unlock(&ulp->lock);
2099
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002100 /* perform adjustments registered in un */
2101 for (i = 0; i < sma->sem_nsems; i++) {
Manfred Spraul239521f2014-01-27 17:07:04 -08002102 struct sem *semaphore = &sma->sem_base[i];
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002103 if (un->semadj[i]) {
2104 semaphore->semval += un->semadj[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105 /*
2106 * Range checks of the new semaphore value,
2107 * not defined by sus:
2108 * - Some unices ignore the undo entirely
2109 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
2110 * - some cap the value (e.g. FreeBSD caps
2111 * at 0, but doesn't enforce SEMVMX)
2112 *
2113 * Linux caps the semaphore value, both at 0
2114 * and at SEMVMX.
2115 *
Manfred Spraul239521f2014-01-27 17:07:04 -08002116 * Manfred <manfred@colorfullife.com>
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117 */
Ingo Molnar5f921ae2006-03-26 01:37:17 -08002118 if (semaphore->semval < 0)
2119 semaphore->semval = 0;
2120 if (semaphore->semval > SEMVMX)
2121 semaphore->semval = SEMVMX;
Pavel Emelyanovb4888932007-10-18 23:40:14 -07002122 semaphore->sempid = task_tgid_vnr(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123 }
2124 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125 /* maybe some queued-up processes were waiting for this */
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07002126 INIT_LIST_HEAD(&tasks);
2127 do_smart_update(sma, NULL, 0, 1, &tasks);
Rik van Riel6062a8d2013-04-30 19:15:44 -07002128 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07002129 rcu_read_unlock();
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07002130 wake_up_sem_queue_do(&tasks);
Manfred Spraul380af1b2008-07-25 01:48:06 -07002131
Lai Jiangshan693a8b62011-03-18 12:09:35 +08002132 kfree_rcu(un, rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133 }
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002134 kfree(ulp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135}
2136
2137#ifdef CONFIG_PROC_FS
Mike Waychison19b49462005-09-06 15:17:10 -07002138static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002139{
Eric W. Biederman1efdb692012-02-07 16:54:11 -08002140 struct user_namespace *user_ns = seq_user_ns(s);
Mike Waychison19b49462005-09-06 15:17:10 -07002141 struct sem_array *sma = it;
Manfred Sprauld12e1e52013-07-08 16:01:25 -07002142 time_t sem_otime;
2143
Manfred Sprauld8c63372013-09-30 13:45:07 -07002144 /*
2145 * The proc interface isn't aware of sem_lock(), it calls
2146 * ipc_lock_object() directly (in sysvipc_find_ipc).
2147 * In order to stay compatible with sem_lock(), we must wait until
2148 * all simple semop() calls have left their critical regions.
2149 */
2150 sem_wait_array(sma);
2151
Manfred Sprauld12e1e52013-07-08 16:01:25 -07002152 sem_otime = get_semotime(sma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153
Mike Waychison19b49462005-09-06 15:17:10 -07002154 return seq_printf(s,
Manfred Spraulb97e8202009-12-15 16:47:32 -08002155 "%10d %10d %4o %10u %5u %5u %5u %5u %10lu %10lu\n",
Mike Waychison19b49462005-09-06 15:17:10 -07002156 sma->sem_perm.key,
Nadia Derbey7ca7e562007-10-18 23:40:48 -07002157 sma->sem_perm.id,
Mike Waychison19b49462005-09-06 15:17:10 -07002158 sma->sem_perm.mode,
2159 sma->sem_nsems,
Eric W. Biederman1efdb692012-02-07 16:54:11 -08002160 from_kuid_munged(user_ns, sma->sem_perm.uid),
2161 from_kgid_munged(user_ns, sma->sem_perm.gid),
2162 from_kuid_munged(user_ns, sma->sem_perm.cuid),
2163 from_kgid_munged(user_ns, sma->sem_perm.cgid),
Manfred Sprauld12e1e52013-07-08 16:01:25 -07002164 sem_otime,
Mike Waychison19b49462005-09-06 15:17:10 -07002165 sma->sem_ctime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002166}
2167#endif