blob: ae72b3cddc8d866450314466df89d466ee7f2dd2 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/ipc/sem.c
3 * Copyright (C) 1992 Krishna Balasubramanian
4 * Copyright (C) 1995 Eric Schenk, Bruno Haible
5 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07006 * /proc/sysvipc/sem support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
7 *
8 * SMP-threaded, sysctl's added
Christian Kujau624dffc2006-01-15 02:43:54 +01009 * (c) 1999 Manfred Spraul <manfred@colorfullife.com>
Linus Torvalds1da177e2005-04-16 15:20:36 -070010 * Enforced range limit on SEM_UNDO
Alan Cox046c6882009-01-05 14:06:29 +000011 * (c) 2001 Red Hat Inc
Linus Torvalds1da177e2005-04-16 15:20:36 -070012 * Lockless wakeup
13 * (c) 2003 Manfred Spraul <manfred@colorfullife.com>
Manfred Spraulc5cf6352010-05-26 14:43:43 -070014 * Further wakeup optimizations, documentation
15 * (c) 2010 Manfred Spraul <manfred@colorfullife.com>
Steve Grubb073115d2006-04-02 17:07:33 -040016 *
17 * support for audit of ipc object properties and permission changes
18 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
Kirill Korotaeve3893532006-10-02 02:18:22 -070019 *
20 * namespaces support
21 * OpenVZ, SWsoft Inc.
22 * Pavel Emelianov <xemul@openvz.org>
Manfred Spraulc5cf6352010-05-26 14:43:43 -070023 *
24 * Implementation notes: (May 2010)
25 * This file implements System V semaphores.
26 *
27 * User space visible behavior:
28 * - FIFO ordering for semop() operations (just FIFO, not starvation
29 * protection)
30 * - multiple semaphore operations that alter the same semaphore in
31 * one semop() are handled.
32 * - sem_ctime (time of last semctl()) is updated in the IPC_SET, SETVAL and
33 * SETALL calls.
34 * - two Linux specific semctl() commands: SEM_STAT, SEM_INFO.
35 * - undo adjustments at process exit are limited to 0..SEMVMX.
36 * - namespace are supported.
37 * - SEMMSL, SEMMNS, SEMOPM and SEMMNI can be configured at runtine by writing
38 * to /proc/sys/kernel/sem.
39 * - statistics about the usage are reported in /proc/sysvipc/sem.
40 *
41 * Internals:
42 * - scalability:
43 * - all global variables are read-mostly.
44 * - semop() calls and semctl(RMID) are synchronized by RCU.
45 * - most operations do write operations (actually: spin_lock calls) to
46 * the per-semaphore array structure.
47 * Thus: Perfect SMP scaling between independent semaphore arrays.
48 * If multiple semaphores in one array are used, then cache line
49 * trashing on the semaphore array spinlock will limit the scaling.
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -070050 * - semncnt and semzcnt are calculated on demand in count_semcnt()
Manfred Spraulc5cf6352010-05-26 14:43:43 -070051 * - the task that performs a successful semop() scans the list of all
52 * sleeping tasks and completes any pending operations that can be fulfilled.
53 * Semaphores are actively given to waiting tasks (necessary for FIFO).
54 * (see update_queue())
55 * - To improve the scalability, the actual wake-up calls are performed after
56 * dropping all locks. (see wake_up_sem_queue_prepare(),
57 * wake_up_sem_queue_do())
58 * - All work is done by the waker, the woken up task does not have to do
59 * anything - not even acquiring a lock or dropping a refcount.
60 * - A woken up task may not even touch the semaphore array anymore, it may
61 * have been destroyed already by a semctl(RMID).
62 * - The synchronizations between wake-ups due to a timeout/signal and a
63 * wake-up due to a completed semaphore operation is achieved by using an
64 * intermediate state (IN_WAKEUP).
65 * - UNDO values are stored in an array (one per process and per
66 * semaphore array, lazily allocated). For backwards compatibility, multiple
67 * modes for the UNDO variables are supported (per process, per thread)
68 * (see copy_semundo, CLONE_SYSVSEM)
69 * - There are two lists of the pending operations: a per-array list
70 * and per-semaphore list (stored in the array). This allows to achieve FIFO
71 * ordering without always scanning all pending operations.
72 * The worst-case behavior is nevertheless O(N^2) for N wakeups.
Linus Torvalds1da177e2005-04-16 15:20:36 -070073 */
74
Linus Torvalds1da177e2005-04-16 15:20:36 -070075#include <linux/slab.h>
76#include <linux/spinlock.h>
77#include <linux/init.h>
78#include <linux/proc_fs.h>
79#include <linux/time.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070080#include <linux/security.h>
81#include <linux/syscalls.h>
82#include <linux/audit.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080083#include <linux/capability.h>
Mike Waychison19b49462005-09-06 15:17:10 -070084#include <linux/seq_file.h>
Nadia Derbey3e148c72007-10-18 23:40:54 -070085#include <linux/rwsem.h>
Kirill Korotaeve3893532006-10-02 02:18:22 -070086#include <linux/nsproxy.h>
Pavel Emelyanovae5e1b22008-02-08 04:18:22 -080087#include <linux/ipc_namespace.h>
Ingo Molnar5f921ae2006-03-26 01:37:17 -080088
Paul McQuade7153e402014-06-06 14:37:37 -070089#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070090#include "util.h"
91
Manfred Spraule57940d2011-11-02 13:38:54 -070092/* One semaphore structure for each semaphore in the system. */
93struct sem {
94 int semval; /* current value */
Davidlohr Buesoa5f4db82016-03-22 14:27:48 -070095 /*
96 * PID of the process that last modified the semaphore. For
97 * Linux, specifically these are:
98 * - semop
99 * - semctl, via SETVAL and SETALL.
100 * - at task exit when performing undo adjustments (see exit_sem).
101 */
102 int sempid;
Rik van Riel6062a8d2013-04-30 19:15:44 -0700103 spinlock_t lock; /* spinlock for fine-grained semtimedop */
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700104 struct list_head pending_alter; /* pending single-sop operations */
105 /* that alter the semaphore */
106 struct list_head pending_const; /* pending single-sop operations */
107 /* that do not alter the semaphore*/
Manfred Sprauld12e1e52013-07-08 16:01:25 -0700108 time_t sem_otime; /* candidate for sem_otime */
Manfred Spraulf5c936c2013-07-08 16:01:22 -0700109} ____cacheline_aligned_in_smp;
Manfred Spraule57940d2011-11-02 13:38:54 -0700110
111/* One queue for each sleeping process in the system. */
112struct sem_queue {
Manfred Spraule57940d2011-11-02 13:38:54 -0700113 struct list_head list; /* queue of pending operations */
114 struct task_struct *sleeper; /* this process */
115 struct sem_undo *undo; /* undo structure */
116 int pid; /* process id of requesting process */
117 int status; /* completion status of operation */
118 struct sembuf *sops; /* array of pending operations */
Manfred Sprauled247b72014-06-06 14:37:49 -0700119 struct sembuf *blocking; /* the operation that blocked */
Manfred Spraule57940d2011-11-02 13:38:54 -0700120 int nsops; /* number of operations */
121 int alter; /* does *sops alter the array? */
122};
123
124/* Each task has a list of undo requests. They are executed automatically
125 * when the process exits.
126 */
127struct sem_undo {
128 struct list_head list_proc; /* per-process list: *
129 * all undos from one process
130 * rcu protected */
131 struct rcu_head rcu; /* rcu struct for sem_undo */
132 struct sem_undo_list *ulp; /* back ptr to sem_undo_list */
133 struct list_head list_id; /* per semaphore array list:
134 * all undos for one array */
135 int semid; /* semaphore set identifier */
136 short *semadj; /* array of adjustments */
137 /* one per semaphore */
138};
139
140/* sem_undo_list controls shared access to the list of sem_undo structures
141 * that may be shared among all a CLONE_SYSVSEM task group.
142 */
143struct sem_undo_list {
144 atomic_t refcnt;
145 spinlock_t lock;
146 struct list_head list_proc;
147};
148
149
Pierre Peiffered2ddbf2008-02-08 04:18:57 -0800150#define sem_ids(ns) ((ns)->ids[IPC_SEM_IDS])
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151
Nadia Derbey1b531f22007-10-18 23:40:55 -0700152#define sem_checkid(sma, semid) ipc_checkid(&sma->sem_perm, semid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700154static int newary(struct ipc_namespace *, struct ipc_params *);
Pierre Peiffer01b8b072008-02-08 04:18:57 -0800155static void freeary(struct ipc_namespace *, struct kern_ipc_perm *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156#ifdef CONFIG_PROC_FS
Mike Waychison19b49462005-09-06 15:17:10 -0700157static int sysvipc_sem_proc_show(struct seq_file *s, void *it);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158#endif
159
160#define SEMMSL_FAST 256 /* 512 bytes on stack */
161#define SEMOPM_FAST 64 /* ~ 372 bytes on stack */
162
163/*
Manfred Spraul758a6ba32013-07-08 16:01:26 -0700164 * Locking:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165 * sem_undo.id_next,
Manfred Spraul758a6ba32013-07-08 16:01:26 -0700166 * sem_array.complex_count,
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700167 * sem_array.pending{_alter,_cont},
Manfred Spraul758a6ba32013-07-08 16:01:26 -0700168 * sem_array.sem_undo: global sem_lock() for read/write
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169 * sem_undo.proc_next: only "current" is allowed to read/write that field.
Paul McQuade46c0a8c2014-06-06 14:37:37 -0700170 *
Manfred Spraul758a6ba32013-07-08 16:01:26 -0700171 * sem_array.sem_base[i].pending_{const,alter}:
172 * global or semaphore sem_lock() for read/write
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 */
174
Kirill Korotaeve3893532006-10-02 02:18:22 -0700175#define sc_semmsl sem_ctls[0]
176#define sc_semmns sem_ctls[1]
177#define sc_semopm sem_ctls[2]
178#define sc_semmni sem_ctls[3]
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179
Pierre Peiffered2ddbf2008-02-08 04:18:57 -0800180void sem_init_ns(struct ipc_namespace *ns)
Kirill Korotaeve3893532006-10-02 02:18:22 -0700181{
Kirill Korotaeve3893532006-10-02 02:18:22 -0700182 ns->sc_semmsl = SEMMSL;
183 ns->sc_semmns = SEMMNS;
184 ns->sc_semopm = SEMOPM;
185 ns->sc_semmni = SEMMNI;
186 ns->used_sems = 0;
Pierre Peiffered2ddbf2008-02-08 04:18:57 -0800187 ipc_init_ids(&ns->ids[IPC_SEM_IDS]);
Kirill Korotaeve3893532006-10-02 02:18:22 -0700188}
189
Pavel Emelyanovae5e1b22008-02-08 04:18:22 -0800190#ifdef CONFIG_IPC_NS
Kirill Korotaeve3893532006-10-02 02:18:22 -0700191void sem_exit_ns(struct ipc_namespace *ns)
192{
Pierre Peiffer01b8b072008-02-08 04:18:57 -0800193 free_ipcs(ns, &sem_ids(ns), freeary);
Serge E. Hallyn7d6feeb2009-12-15 16:47:27 -0800194 idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr);
Kirill Korotaeve3893532006-10-02 02:18:22 -0700195}
Pavel Emelyanovae5e1b22008-02-08 04:18:22 -0800196#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197
Manfred Spraul239521f2014-01-27 17:07:04 -0800198void __init sem_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199{
Pierre Peiffered2ddbf2008-02-08 04:18:57 -0800200 sem_init_ns(&init_ipc_ns);
Mike Waychison19b49462005-09-06 15:17:10 -0700201 ipc_init_proc_interface("sysvipc/sem",
202 " key semid perms nsems uid gid cuid cgid otime ctime\n",
Kirill Korotaeve3893532006-10-02 02:18:22 -0700203 IPC_SEM_IDS, sysvipc_sem_proc_show);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204}
205
Manfred Spraulf269f402013-07-08 16:01:24 -0700206/**
207 * unmerge_queues - unmerge queues, if possible.
208 * @sma: semaphore array
209 *
210 * The function unmerges the wait queues if complex_count is 0.
211 * It must be called prior to dropping the global semaphore array lock.
212 */
213static void unmerge_queues(struct sem_array *sma)
214{
215 struct sem_queue *q, *tq;
216
217 /* complex operations still around? */
218 if (sma->complex_count)
219 return;
220 /*
221 * We will switch back to simple mode.
222 * Move all pending operation back into the per-semaphore
223 * queues.
224 */
225 list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
226 struct sem *curr;
227 curr = &sma->sem_base[q->sops[0].sem_num];
228
229 list_add_tail(&q->list, &curr->pending_alter);
230 }
231 INIT_LIST_HEAD(&sma->pending_alter);
232}
233
234/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800235 * merge_queues - merge single semop queues into global queue
Manfred Spraulf269f402013-07-08 16:01:24 -0700236 * @sma: semaphore array
237 *
238 * This function merges all per-semaphore queues into the global queue.
239 * It is necessary to achieve FIFO ordering for the pending single-sop
240 * operations when a multi-semop operation must sleep.
241 * Only the alter operations must be moved, the const operations can stay.
242 */
243static void merge_queues(struct sem_array *sma)
244{
245 int i;
246 for (i = 0; i < sma->sem_nsems; i++) {
247 struct sem *sem = sma->sem_base + i;
248
249 list_splice_init(&sem->pending_alter, &sma->pending_alter);
250 }
251}
252
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -0700253static void sem_rcu_free(struct rcu_head *head)
254{
255 struct ipc_rcu *p = container_of(head, struct ipc_rcu, rcu);
256 struct sem_array *sma = ipc_rcu_to_struct(p);
257
258 security_sem_free(sma);
259 ipc_rcu_free(head);
260}
261
Nadia Derbey3e148c72007-10-18 23:40:54 -0700262/*
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700263 * Wait until all currently ongoing simple ops have completed.
264 * Caller must own sem_perm.lock.
265 * New simple ops cannot start, because simple ops first check
266 * that sem_perm.lock is free.
Manfred Spraul6d07b682013-09-30 13:45:06 -0700267 * that a) sem_perm.lock is free and b) complex_count is 0.
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700268 */
269static void sem_wait_array(struct sem_array *sma)
270{
271 int i;
272 struct sem *sem;
273
Manfred Spraul6d07b682013-09-30 13:45:06 -0700274 if (sma->complex_count) {
275 /* The thread that increased sma->complex_count waited on
276 * all sem->lock locks. Thus we don't need to wait again.
277 */
278 return;
279 }
280
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700281 for (i = 0; i < sma->sem_nsems; i++) {
282 sem = sma->sem_base + i;
283 spin_unlock_wait(&sem->lock);
284 }
285}
286
287/*
Rik van Riel6062a8d2013-04-30 19:15:44 -0700288 * If the request contains only one semaphore operation, and there are
289 * no complex transactions pending, lock only the semaphore involved.
290 * Otherwise, lock the entire semaphore array, since we either have
291 * multiple semaphores in our own semops, or we need to look at
292 * semaphores from other pending complex operations.
Rik van Riel6062a8d2013-04-30 19:15:44 -0700293 */
294static inline int sem_lock(struct sem_array *sma, struct sembuf *sops,
295 int nsops)
296{
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700297 struct sem *sem;
Rik van Riel6062a8d2013-04-30 19:15:44 -0700298
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700299 if (nsops != 1) {
300 /* Complex operation - acquire a full lock */
301 ipc_lock_object(&sma->sem_perm);
302
303 /* And wait until all simple ops that are processed
304 * right now have dropped their locks.
305 */
306 sem_wait_array(sma);
307 return -1;
308 }
309
310 /*
311 * Only one semaphore affected - try to optimize locking.
312 * The rules are:
313 * - optimized locking is possible if no complex operation
314 * is either enqueued or processed right now.
315 * - The test for enqueued complex ops is simple:
316 * sma->complex_count != 0
317 * - Testing for complex ops that are processed right now is
318 * a bit more difficult. Complex ops acquire the full lock
319 * and first wait that the running simple ops have completed.
320 * (see above)
321 * Thus: If we own a simple lock and the global lock is free
322 * and complex_count is now 0, then it will stay 0 and
323 * thus just locking sem->lock is sufficient.
324 */
325 sem = sma->sem_base + sops->sem_num;
326
327 if (sma->complex_count == 0) {
328 /*
329 * It appears that no complex operation is around.
330 * Acquire the per-semaphore lock.
331 */
Rik van Riel6062a8d2013-04-30 19:15:44 -0700332 spin_lock(&sem->lock);
333
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700334 /* Then check that the global lock is free */
335 if (!spin_is_locked(&sma->sem_perm.lock)) {
Manfred Spraul2e094ab2014-12-12 16:58:11 -0800336 /*
Manfred Spraul3ed1f8a2015-08-14 15:35:10 -0700337 * We need a memory barrier with acquire semantics,
338 * otherwise we can race with another thread that does:
Manfred Spraul2e094ab2014-12-12 16:58:11 -0800339 * complex_count++;
340 * spin_unlock(sem_perm.lock);
341 */
Peter Zijlstra33ac2792016-05-24 13:17:12 +0200342 smp_acquire__after_ctrl_dep();
Rik van Riel6062a8d2013-04-30 19:15:44 -0700343
Manfred Spraul2e094ab2014-12-12 16:58:11 -0800344 /*
345 * Now repeat the test of complex_count:
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700346 * It can't change anymore until we drop sem->lock.
347 * Thus: if is now 0, then it will stay 0.
348 */
349 if (sma->complex_count == 0) {
350 /* fast path successful! */
351 return sops->sem_num;
352 }
Rik van Riel6062a8d2013-04-30 19:15:44 -0700353 }
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700354 spin_unlock(&sem->lock);
Rik van Riel6062a8d2013-04-30 19:15:44 -0700355 }
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700356
357 /* slow path: acquire the full lock */
358 ipc_lock_object(&sma->sem_perm);
359
360 if (sma->complex_count == 0) {
361 /* False alarm:
362 * There is no complex operation, thus we can switch
363 * back to the fast path.
364 */
365 spin_lock(&sem->lock);
366 ipc_unlock_object(&sma->sem_perm);
367 return sops->sem_num;
368 } else {
369 /* Not a false alarm, thus complete the sequence for a
370 * full lock.
371 */
372 sem_wait_array(sma);
373 return -1;
374 }
Rik van Riel6062a8d2013-04-30 19:15:44 -0700375}
376
377static inline void sem_unlock(struct sem_array *sma, int locknum)
378{
379 if (locknum == -1) {
Manfred Spraulf269f402013-07-08 16:01:24 -0700380 unmerge_queues(sma);
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -0700381 ipc_unlock_object(&sma->sem_perm);
Rik van Riel6062a8d2013-04-30 19:15:44 -0700382 } else {
383 struct sem *sem = sma->sem_base + locknum;
384 spin_unlock(&sem->lock);
385 }
Rik van Riel6062a8d2013-04-30 19:15:44 -0700386}
387
388/*
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700389 * sem_lock_(check_) routines are called in the paths where the rwsem
Nadia Derbey3e148c72007-10-18 23:40:54 -0700390 * is not held.
Linus Torvalds321310c2013-05-04 10:47:57 -0700391 *
392 * The caller holds the RCU read lock.
Nadia Derbey3e148c72007-10-18 23:40:54 -0700393 */
Rik van Riel6062a8d2013-04-30 19:15:44 -0700394static inline struct sem_array *sem_obtain_lock(struct ipc_namespace *ns,
395 int id, struct sembuf *sops, int nsops, int *locknum)
Nadia Derbey023a5352007-10-18 23:40:51 -0700396{
Rik van Rielc460b662013-04-30 19:15:35 -0700397 struct kern_ipc_perm *ipcp;
398 struct sem_array *sma;
Nadia Derbey03f02c72007-10-18 23:40:51 -0700399
Davidlohr Bueso55b7ae52015-06-30 14:58:42 -0700400 ipcp = ipc_obtain_object_idr(&sem_ids(ns), id);
Linus Torvalds321310c2013-05-04 10:47:57 -0700401 if (IS_ERR(ipcp))
402 return ERR_CAST(ipcp);
Pierre Peifferb1ed88b2008-02-06 01:36:23 -0800403
Rik van Riel6062a8d2013-04-30 19:15:44 -0700404 sma = container_of(ipcp, struct sem_array, sem_perm);
405 *locknum = sem_lock(sma, sops, nsops);
Rik van Rielc460b662013-04-30 19:15:35 -0700406
407 /* ipc_rmid() may have already freed the ID while sem_lock
408 * was spinning: verify that the structure is still valid
409 */
Rafael Aquini72a8ff22014-01-27 17:07:02 -0800410 if (ipc_valid_object(ipcp))
Rik van Rielc460b662013-04-30 19:15:35 -0700411 return container_of(ipcp, struct sem_array, sem_perm);
412
Rik van Riel6062a8d2013-04-30 19:15:44 -0700413 sem_unlock(sma, *locknum);
Linus Torvalds321310c2013-05-04 10:47:57 -0700414 return ERR_PTR(-EINVAL);
Nadia Derbey023a5352007-10-18 23:40:51 -0700415}
416
Davidlohr Bueso16df3672013-04-30 19:15:29 -0700417static inline struct sem_array *sem_obtain_object(struct ipc_namespace *ns, int id)
418{
Davidlohr Bueso55b7ae52015-06-30 14:58:42 -0700419 struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&sem_ids(ns), id);
Davidlohr Bueso16df3672013-04-30 19:15:29 -0700420
421 if (IS_ERR(ipcp))
422 return ERR_CAST(ipcp);
423
424 return container_of(ipcp, struct sem_array, sem_perm);
425}
426
Davidlohr Bueso16df3672013-04-30 19:15:29 -0700427static inline struct sem_array *sem_obtain_object_check(struct ipc_namespace *ns,
428 int id)
429{
430 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&sem_ids(ns), id);
431
432 if (IS_ERR(ipcp))
433 return ERR_CAST(ipcp);
Pierre Peifferb1ed88b2008-02-06 01:36:23 -0800434
Nadia Derbey03f02c72007-10-18 23:40:51 -0700435 return container_of(ipcp, struct sem_array, sem_perm);
Nadia Derbey023a5352007-10-18 23:40:51 -0700436}
437
Pierre Peiffer6ff37972008-04-29 01:00:46 -0700438static inline void sem_lock_and_putref(struct sem_array *sma)
439{
Rik van Riel6062a8d2013-04-30 19:15:44 -0700440 sem_lock(sma, NULL, -1);
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -0700441 ipc_rcu_putref(sma, ipc_rcu_free);
Pierre Peiffer6ff37972008-04-29 01:00:46 -0700442}
443
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700444static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
445{
446 ipc_rmid(&sem_ids(ns), &s->sem_perm);
447}
448
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449/*
450 * Lockless wakeup algorithm:
451 * Without the check/retry algorithm a lockless wakeup is possible:
452 * - queue.status is initialized to -EINTR before blocking.
453 * - wakeup is performed by
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700454 * * unlinking the queue entry from the pending list
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 * * setting queue.status to IN_WAKEUP
456 * This is the notification for the blocked thread that a
457 * result value is imminent.
458 * * call wake_up_process
459 * * set queue.status to the final value.
460 * - the previously blocked thread checks queue.status:
Manfred Spraul239521f2014-01-27 17:07:04 -0800461 * * if it's IN_WAKEUP, then it must wait until the value changes
462 * * if it's not -EINTR, then the operation was completed by
463 * update_queue. semtimedop can return queue.status without
464 * performing any operation on the sem array.
465 * * otherwise it must acquire the spinlock and check what's up.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 *
467 * The two-stage algorithm is necessary to protect against the following
468 * races:
469 * - if queue.status is set after wake_up_process, then the woken up idle
470 * thread could race forward and try (and fail) to acquire sma->lock
471 * before update_queue had a chance to set queue.status
472 * - if queue.status is written before wake_up_process and if the
473 * blocked process is woken up by a signal between writing
474 * queue.status and the wake_up_process, then the woken up
475 * process could return from semtimedop and die by calling
476 * sys_exit before wake_up_process is called. Then wake_up_process
477 * will oops, because the task structure is already invalid.
478 * (yes, this happened on s390 with sysv msg).
479 *
480 */
481#define IN_WAKEUP 1
482
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700483/**
484 * newary - Create a new semaphore set
485 * @ns: namespace
486 * @params: ptr to the structure that contains key, semflg and nsems
487 *
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700488 * Called with sem_ids.rwsem held (as a writer)
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700489 */
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700490static int newary(struct ipc_namespace *ns, struct ipc_params *params)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491{
492 int id;
493 int retval;
494 struct sem_array *sma;
495 int size;
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700496 key_t key = params->key;
497 int nsems = params->u.nsems;
498 int semflg = params->flg;
Manfred Spraulb97e8202009-12-15 16:47:32 -0800499 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500
501 if (!nsems)
502 return -EINVAL;
Kirill Korotaeve3893532006-10-02 02:18:22 -0700503 if (ns->used_sems + nsems > ns->sc_semmns)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 return -ENOSPC;
505
Manfred Spraul239521f2014-01-27 17:07:04 -0800506 size = sizeof(*sma) + nsems * sizeof(struct sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 sma = ipc_rcu_alloc(size);
Davidlohr Bueso3ab08fe2014-01-27 17:07:06 -0800508 if (!sma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 return -ENOMEM;
Davidlohr Bueso3ab08fe2014-01-27 17:07:06 -0800510
Manfred Spraul239521f2014-01-27 17:07:04 -0800511 memset(sma, 0, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512
513 sma->sem_perm.mode = (semflg & S_IRWXUGO);
514 sma->sem_perm.key = key;
515
516 sma->sem_perm.security = NULL;
517 retval = security_sem_alloc(sma);
518 if (retval) {
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -0700519 ipc_rcu_putref(sma, ipc_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 return retval;
521 }
522
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 sma->sem_base = (struct sem *) &sma[1];
Manfred Spraulb97e8202009-12-15 16:47:32 -0800524
Rik van Riel6062a8d2013-04-30 19:15:44 -0700525 for (i = 0; i < nsems; i++) {
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700526 INIT_LIST_HEAD(&sma->sem_base[i].pending_alter);
527 INIT_LIST_HEAD(&sma->sem_base[i].pending_const);
Rik van Riel6062a8d2013-04-30 19:15:44 -0700528 spin_lock_init(&sma->sem_base[i].lock);
529 }
Manfred Spraulb97e8202009-12-15 16:47:32 -0800530
531 sma->complex_count = 0;
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700532 INIT_LIST_HEAD(&sma->pending_alter);
533 INIT_LIST_HEAD(&sma->pending_const);
Manfred Spraul4daa28f2008-07-25 01:48:04 -0700534 INIT_LIST_HEAD(&sma->list_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 sma->sem_nsems = nsems;
536 sma->sem_ctime = get_seconds();
Manfred Spraule8577d12014-12-02 15:59:34 -0800537
538 id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
539 if (id < 0) {
540 ipc_rcu_putref(sma, sem_rcu_free);
541 return id;
542 }
543 ns->used_sems += nsems;
544
Rik van Riel6062a8d2013-04-30 19:15:44 -0700545 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -0700546 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700548 return sma->sem_perm.id;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549}
550
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700551
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700552/*
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700553 * Called with sem_ids.rwsem and ipcp locked.
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700554 */
Nadia Derbey03f02c72007-10-18 23:40:51 -0700555static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556{
Nadia Derbey03f02c72007-10-18 23:40:51 -0700557 struct sem_array *sma;
558
559 sma = container_of(ipcp, struct sem_array, sem_perm);
560 return security_sem_associate(sma, semflg);
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700561}
562
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700563/*
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700564 * Called with sem_ids.rwsem and ipcp locked.
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700565 */
Nadia Derbey03f02c72007-10-18 23:40:51 -0700566static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
567 struct ipc_params *params)
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700568{
Nadia Derbey03f02c72007-10-18 23:40:51 -0700569 struct sem_array *sma;
570
571 sma = container_of(ipcp, struct sem_array, sem_perm);
572 if (params->u.nsems > sma->sem_nsems)
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700573 return -EINVAL;
574
575 return 0;
576}
577
Heiko Carstensd5460c92009-01-14 14:14:27 +0100578SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg)
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700579{
Kirill Korotaeve3893532006-10-02 02:18:22 -0700580 struct ipc_namespace *ns;
Mathias Krauseeb66ec42014-06-06 14:37:36 -0700581 static const struct ipc_ops sem_ops = {
582 .getnew = newary,
583 .associate = sem_security,
584 .more_checks = sem_more_checks,
585 };
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700586 struct ipc_params sem_params;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587
Kirill Korotaeve3893532006-10-02 02:18:22 -0700588 ns = current->nsproxy->ipc_ns;
589
590 if (nsems < 0 || nsems > ns->sc_semmsl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 return -EINVAL;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700592
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700593 sem_params.key = key;
594 sem_params.flg = semflg;
595 sem_params.u.nsems = nsems;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700596
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700597 return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598}
599
Petr Mladek78f50092014-01-27 17:07:00 -0800600/**
601 * perform_atomic_semop - Perform (if possible) a semaphore operation
Manfred Spraul758a6ba32013-07-08 16:01:26 -0700602 * @sma: semaphore array
Manfred Sprauld198cd62014-06-06 14:37:49 -0700603 * @q: struct sem_queue that describes the operation
Manfred Spraul758a6ba32013-07-08 16:01:26 -0700604 *
605 * Returns 0 if the operation was possible.
606 * Returns 1 if the operation is impossible, the caller must sleep.
607 * Negative values are error codes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 */
Manfred Sprauld198cd62014-06-06 14:37:49 -0700609static int perform_atomic_semop(struct sem_array *sma, struct sem_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610{
Manfred Sprauld198cd62014-06-06 14:37:49 -0700611 int result, sem_op, nsops, pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 struct sembuf *sop;
Manfred Spraul239521f2014-01-27 17:07:04 -0800613 struct sem *curr;
Manfred Sprauld198cd62014-06-06 14:37:49 -0700614 struct sembuf *sops;
615 struct sem_undo *un;
616
617 sops = q->sops;
618 nsops = q->nsops;
619 un = q->undo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620
621 for (sop = sops; sop < sops + nsops; sop++) {
622 curr = sma->sem_base + sop->sem_num;
623 sem_op = sop->sem_op;
624 result = curr->semval;
Petr Mladek78f50092014-01-27 17:07:00 -0800625
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 if (!sem_op && result)
627 goto would_block;
628
629 result += sem_op;
630 if (result < 0)
631 goto would_block;
632 if (result > SEMVMX)
633 goto out_of_range;
Petr Mladek78f50092014-01-27 17:07:00 -0800634
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 if (sop->sem_flg & SEM_UNDO) {
636 int undo = un->semadj[sop->sem_num] - sem_op;
Petr Mladek78f50092014-01-27 17:07:00 -0800637 /* Exceeding the undo range is an error. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
639 goto out_of_range;
Petr Mladek78f50092014-01-27 17:07:00 -0800640 un->semadj[sop->sem_num] = undo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 }
Petr Mladek78f50092014-01-27 17:07:00 -0800642
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 curr->semval = result;
644 }
645
646 sop--;
Manfred Sprauld198cd62014-06-06 14:37:49 -0700647 pid = q->pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 while (sop >= sops) {
649 sma->sem_base[sop->sem_num].sempid = pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 sop--;
651 }
Petr Mladek78f50092014-01-27 17:07:00 -0800652
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 return 0;
654
655out_of_range:
656 result = -ERANGE;
657 goto undo;
658
659would_block:
Manfred Sprauled247b72014-06-06 14:37:49 -0700660 q->blocking = sop;
661
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 if (sop->sem_flg & IPC_NOWAIT)
663 result = -EAGAIN;
664 else
665 result = 1;
666
667undo:
668 sop--;
669 while (sop >= sops) {
Petr Mladek78f50092014-01-27 17:07:00 -0800670 sem_op = sop->sem_op;
671 sma->sem_base[sop->sem_num].semval -= sem_op;
672 if (sop->sem_flg & SEM_UNDO)
673 un->semadj[sop->sem_num] += sem_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 sop--;
675 }
676
677 return result;
678}
679
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700680/** wake_up_sem_queue_prepare(q, error): Prepare wake-up
681 * @q: queue entry that must be signaled
682 * @error: Error value for the signal
683 *
684 * Prepare the wake-up of the queue entry q.
Nick Piggind4212092009-12-15 16:47:30 -0800685 */
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700686static void wake_up_sem_queue_prepare(struct list_head *pt,
687 struct sem_queue *q, int error)
Nick Piggind4212092009-12-15 16:47:30 -0800688{
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700689 if (list_empty(pt)) {
690 /*
691 * Hold preempt off so that we don't get preempted and have the
692 * wakee busy-wait until we're scheduled back on.
693 */
694 preempt_disable();
695 }
Nick Piggind4212092009-12-15 16:47:30 -0800696 q->status = IN_WAKEUP;
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700697 q->pid = error;
698
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700699 list_add_tail(&q->list, pt);
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700700}
701
702/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800703 * wake_up_sem_queue_do - do the actual wake-up
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700704 * @pt: list of tasks to be woken up
705 *
706 * Do the actual wake-up.
707 * The function is called without any locks held, thus the semaphore array
708 * could be destroyed already and the tasks can disappear as soon as the
709 * status is set to the actual return code.
710 */
711static void wake_up_sem_queue_do(struct list_head *pt)
712{
713 struct sem_queue *q, *t;
714 int did_something;
715
716 did_something = !list_empty(pt);
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700717 list_for_each_entry_safe(q, t, pt, list) {
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700718 wake_up_process(q->sleeper);
719 /* q can disappear immediately after writing q->status. */
720 smp_wmb();
721 q->status = q->pid;
722 }
723 if (did_something)
724 preempt_enable();
Nick Piggind4212092009-12-15 16:47:30 -0800725}
726
Manfred Spraulb97e8202009-12-15 16:47:32 -0800727static void unlink_queue(struct sem_array *sma, struct sem_queue *q)
728{
729 list_del(&q->list);
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700730 if (q->nsops > 1)
Manfred Spraulb97e8202009-12-15 16:47:32 -0800731 sma->complex_count--;
732}
733
Manfred Spraulfd5db422010-05-26 14:43:40 -0700734/** check_restart(sma, q)
735 * @sma: semaphore array
736 * @q: the operation that just completed
737 *
738 * update_queue is O(N^2) when it restarts scanning the whole queue of
739 * waiting operations. Therefore this function checks if the restart is
740 * really necessary. It is called after a previously waiting operation
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700741 * modified the array.
742 * Note that wait-for-zero operations are handled without restart.
Manfred Spraulfd5db422010-05-26 14:43:40 -0700743 */
744static int check_restart(struct sem_array *sma, struct sem_queue *q)
745{
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700746 /* pending complex alter operations are too difficult to analyse */
747 if (!list_empty(&sma->pending_alter))
Manfred Spraulfd5db422010-05-26 14:43:40 -0700748 return 1;
749
750 /* we were a sleeping complex operation. Too difficult */
751 if (q->nsops > 1)
752 return 1;
753
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700754 /* It is impossible that someone waits for the new value:
755 * - complex operations always restart.
756 * - wait-for-zero are handled seperately.
757 * - q is a previously sleeping simple operation that
758 * altered the array. It must be a decrement, because
759 * simple increments never sleep.
760 * - If there are older (higher priority) decrements
761 * in the queue, then they have observed the original
762 * semval value and couldn't proceed. The operation
763 * decremented to value - thus they won't proceed either.
764 */
765 return 0;
766}
Manfred Spraulfd5db422010-05-26 14:43:40 -0700767
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700768/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800769 * wake_const_ops - wake up non-alter tasks
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700770 * @sma: semaphore array.
771 * @semnum: semaphore that was modified.
772 * @pt: list head for the tasks that must be woken up.
773 *
774 * wake_const_ops must be called after a semaphore in a semaphore array
775 * was set to 0. If complex const operations are pending, wake_const_ops must
776 * be called with semnum = -1, as well as with the number of each modified
777 * semaphore.
778 * The tasks that must be woken up are added to @pt. The return code
779 * is stored in q->pid.
780 * The function returns 1 if at least one operation was completed successfully.
781 */
782static int wake_const_ops(struct sem_array *sma, int semnum,
783 struct list_head *pt)
784{
785 struct sem_queue *q;
786 struct list_head *walk;
787 struct list_head *pending_list;
788 int semop_completed = 0;
Manfred Spraulfd5db422010-05-26 14:43:40 -0700789
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700790 if (semnum == -1)
791 pending_list = &sma->pending_const;
792 else
793 pending_list = &sma->sem_base[semnum].pending_const;
794
795 walk = pending_list->next;
796 while (walk != pending_list) {
797 int error;
798
799 q = container_of(walk, struct sem_queue, list);
800 walk = walk->next;
801
Manfred Sprauld198cd62014-06-06 14:37:49 -0700802 error = perform_atomic_semop(sma, q);
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700803
804 if (error <= 0) {
805 /* operation completed, remove from queue & wakeup */
806
807 unlink_queue(sma, q);
808
809 wake_up_sem_queue_prepare(pt, q, error);
810 if (error == 0)
811 semop_completed = 1;
812 }
813 }
814 return semop_completed;
815}
816
817/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800818 * do_smart_wakeup_zero - wakeup all wait for zero tasks
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700819 * @sma: semaphore array
820 * @sops: operations that were performed
821 * @nsops: number of operations
822 * @pt: list head of the tasks that must be woken up.
823 *
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800824 * Checks all required queue for wait-for-zero operations, based
825 * on the actual changes that were performed on the semaphore array.
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700826 * The function returns 1 if at least one operation was completed successfully.
827 */
828static int do_smart_wakeup_zero(struct sem_array *sma, struct sembuf *sops,
829 int nsops, struct list_head *pt)
830{
831 int i;
832 int semop_completed = 0;
833 int got_zero = 0;
834
835 /* first: the per-semaphore queues, if known */
836 if (sops) {
837 for (i = 0; i < nsops; i++) {
838 int num = sops[i].sem_num;
839
840 if (sma->sem_base[num].semval == 0) {
841 got_zero = 1;
842 semop_completed |= wake_const_ops(sma, num, pt);
843 }
844 }
845 } else {
846 /*
847 * No sops means modified semaphores not known.
848 * Assume all were changed.
Manfred Spraulfd5db422010-05-26 14:43:40 -0700849 */
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700850 for (i = 0; i < sma->sem_nsems; i++) {
851 if (sma->sem_base[i].semval == 0) {
852 got_zero = 1;
853 semop_completed |= wake_const_ops(sma, i, pt);
854 }
855 }
Manfred Spraulfd5db422010-05-26 14:43:40 -0700856 }
857 /*
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700858 * If one of the modified semaphores got 0,
859 * then check the global queue, too.
Manfred Spraulfd5db422010-05-26 14:43:40 -0700860 */
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700861 if (got_zero)
862 semop_completed |= wake_const_ops(sma, -1, pt);
Manfred Spraulfd5db422010-05-26 14:43:40 -0700863
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700864 return semop_completed;
Manfred Spraulfd5db422010-05-26 14:43:40 -0700865}
866
Manfred Spraul636c6be2009-12-15 16:47:33 -0800867
868/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800869 * update_queue - look for tasks that can be completed.
Manfred Spraul636c6be2009-12-15 16:47:33 -0800870 * @sma: semaphore array.
871 * @semnum: semaphore that was modified.
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700872 * @pt: list head for the tasks that must be woken up.
Manfred Spraul636c6be2009-12-15 16:47:33 -0800873 *
874 * update_queue must be called after a semaphore in a semaphore array
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700875 * was modified. If multiple semaphores were modified, update_queue must
876 * be called with semnum = -1, as well as with the number of each modified
877 * semaphore.
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700878 * The tasks that must be woken up are added to @pt. The return code
879 * is stored in q->pid.
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700880 * The function internally checks if const operations can now succeed.
881 *
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700882 * The function return 1 if at least one semop was completed successfully.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 */
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700884static int update_queue(struct sem_array *sma, int semnum, struct list_head *pt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885{
Manfred Spraul636c6be2009-12-15 16:47:33 -0800886 struct sem_queue *q;
887 struct list_head *walk;
888 struct list_head *pending_list;
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700889 int semop_completed = 0;
Manfred Spraul636c6be2009-12-15 16:47:33 -0800890
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700891 if (semnum == -1)
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700892 pending_list = &sma->pending_alter;
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700893 else
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700894 pending_list = &sma->sem_base[semnum].pending_alter;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895
Nick Piggin9cad2002009-12-15 16:47:29 -0800896again:
Manfred Spraul636c6be2009-12-15 16:47:33 -0800897 walk = pending_list->next;
898 while (walk != pending_list) {
Manfred Spraulfd5db422010-05-26 14:43:40 -0700899 int error, restart;
Manfred Spraul636c6be2009-12-15 16:47:33 -0800900
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700901 q = container_of(walk, struct sem_queue, list);
Manfred Spraul636c6be2009-12-15 16:47:33 -0800902 walk = walk->next;
Nick Piggin9cad2002009-12-15 16:47:29 -0800903
Manfred Sprauld987f8b2009-12-15 16:47:34 -0800904 /* If we are scanning the single sop, per-semaphore list of
905 * one semaphore and that semaphore is 0, then it is not
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700906 * necessary to scan further: simple increments
Manfred Sprauld987f8b2009-12-15 16:47:34 -0800907 * that affect only one entry succeed immediately and cannot
908 * be in the per semaphore pending queue, and decrements
909 * cannot be successful if the value is already 0.
910 */
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700911 if (semnum != -1 && sma->sem_base[semnum].semval == 0)
Manfred Sprauld987f8b2009-12-15 16:47:34 -0800912 break;
913
Manfred Sprauld198cd62014-06-06 14:37:49 -0700914 error = perform_atomic_semop(sma, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915
916 /* Does q->sleeper still need to sleep? */
Nick Piggin9cad2002009-12-15 16:47:29 -0800917 if (error > 0)
918 continue;
Manfred Spraula1193f82008-07-25 01:48:06 -0700919
Manfred Spraulb97e8202009-12-15 16:47:32 -0800920 unlink_queue(sma, q);
Manfred Spraula1193f82008-07-25 01:48:06 -0700921
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700922 if (error) {
Manfred Spraulfd5db422010-05-26 14:43:40 -0700923 restart = 0;
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700924 } else {
925 semop_completed = 1;
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700926 do_smart_wakeup_zero(sma, q->sops, q->nsops, pt);
Manfred Spraulfd5db422010-05-26 14:43:40 -0700927 restart = check_restart(sma, q);
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700928 }
Manfred Spraulfd5db422010-05-26 14:43:40 -0700929
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700930 wake_up_sem_queue_prepare(pt, q, error);
Manfred Spraulfd5db422010-05-26 14:43:40 -0700931 if (restart)
Nick Piggin9cad2002009-12-15 16:47:29 -0800932 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 }
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700934 return semop_completed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935}
936
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700937/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800938 * set_semotime - set sem_otime
Manfred Spraul0e8c6652013-09-30 13:45:25 -0700939 * @sma: semaphore array
940 * @sops: operations that modified the array, may be NULL
941 *
942 * sem_otime is replicated to avoid cache line trashing.
943 * This function sets one instance to the current time.
944 */
945static void set_semotime(struct sem_array *sma, struct sembuf *sops)
946{
947 if (sops == NULL) {
948 sma->sem_base[0].sem_otime = get_seconds();
949 } else {
950 sma->sem_base[sops[0].sem_num].sem_otime =
951 get_seconds();
952 }
953}
954
955/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800956 * do_smart_update - optimized update_queue
Manfred Spraulfd5db422010-05-26 14:43:40 -0700957 * @sma: semaphore array
958 * @sops: operations that were performed
959 * @nsops: number of operations
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700960 * @otime: force setting otime
961 * @pt: list head of the tasks that must be woken up.
Manfred Spraulfd5db422010-05-26 14:43:40 -0700962 *
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700963 * do_smart_update() does the required calls to update_queue and wakeup_zero,
964 * based on the actual changes that were performed on the semaphore array.
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700965 * Note that the function does not do the actual wake-up: the caller is
966 * responsible for calling wake_up_sem_queue_do(@pt).
967 * It is safe to perform this call after dropping all locks.
Manfred Spraulfd5db422010-05-26 14:43:40 -0700968 */
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700969static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops,
970 int otime, struct list_head *pt)
Manfred Spraulfd5db422010-05-26 14:43:40 -0700971{
972 int i;
973
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700974 otime |= do_smart_wakeup_zero(sma, sops, nsops, pt);
975
Manfred Spraulf269f402013-07-08 16:01:24 -0700976 if (!list_empty(&sma->pending_alter)) {
977 /* semaphore array uses the global queue - just process it. */
978 otime |= update_queue(sma, -1, pt);
979 } else {
980 if (!sops) {
981 /*
982 * No sops, thus the modified semaphores are not
983 * known. Check all.
984 */
985 for (i = 0; i < sma->sem_nsems; i++)
986 otime |= update_queue(sma, i, pt);
987 } else {
988 /*
989 * Check the semaphores that were increased:
990 * - No complex ops, thus all sleeping ops are
991 * decrease.
992 * - if we decreased the value, then any sleeping
993 * semaphore ops wont be able to run: If the
994 * previous value was too small, then the new
995 * value will be too small, too.
996 */
997 for (i = 0; i < nsops; i++) {
998 if (sops[i].sem_op > 0) {
999 otime |= update_queue(sma,
1000 sops[i].sem_num, pt);
1001 }
Manfred Spraulab465df2013-05-26 11:08:52 +02001002 }
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001003 }
Manfred Spraulfd5db422010-05-26 14:43:40 -07001004 }
Manfred Spraul0e8c6652013-09-30 13:45:25 -07001005 if (otime)
1006 set_semotime(sma, sops);
Manfred Spraulfd5db422010-05-26 14:43:40 -07001007}
1008
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001009/*
Manfred Spraulb220c572014-06-06 14:37:51 -07001010 * check_qop: Test if a queued operation sleeps on the semaphore semnum
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001011 */
1012static int check_qop(struct sem_array *sma, int semnum, struct sem_queue *q,
1013 bool count_zero)
1014{
Manfred Spraulb220c572014-06-06 14:37:51 -07001015 struct sembuf *sop = q->blocking;
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001016
Manfred Spraul9b44ee22014-06-06 14:37:52 -07001017 /*
1018 * Linux always (since 0.99.10) reported a task as sleeping on all
1019 * semaphores. This violates SUS, therefore it was changed to the
1020 * standard compliant behavior.
1021 * Give the administrators a chance to notice that an application
1022 * might misbehave because it relies on the Linux behavior.
1023 */
1024 pr_info_once("semctl(GETNCNT/GETZCNT) is since 3.16 Single Unix Specification compliant.\n"
1025 "The task %s (%d) triggered the difference, watch for misbehavior.\n",
1026 current->comm, task_pid_nr(current));
1027
Manfred Spraulb220c572014-06-06 14:37:51 -07001028 if (sop->sem_num != semnum)
1029 return 0;
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001030
Manfred Spraulb220c572014-06-06 14:37:51 -07001031 if (count_zero && sop->sem_op == 0)
1032 return 1;
1033 if (!count_zero && sop->sem_op < 0)
1034 return 1;
1035
1036 return 0;
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001037}
1038
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039/* The following counts are associated to each semaphore:
1040 * semncnt number of tasks waiting on semval being nonzero
1041 * semzcnt number of tasks waiting on semval being zero
Manfred Spraulb220c572014-06-06 14:37:51 -07001042 *
1043 * Per definition, a task waits only on the semaphore of the first semop
1044 * that cannot proceed, even if additional operation would block, too.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 */
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001046static int count_semcnt(struct sem_array *sma, ushort semnum,
1047 bool count_zero)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048{
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001049 struct list_head *l;
Manfred Spraul239521f2014-01-27 17:07:04 -08001050 struct sem_queue *q;
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001051 int semcnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001053 semcnt = 0;
1054 /* First: check the simple operations. They are easy to evaluate */
1055 if (count_zero)
1056 l = &sma->sem_base[semnum].pending_const;
1057 else
1058 l = &sma->sem_base[semnum].pending_alter;
1059
1060 list_for_each_entry(q, l, list) {
1061 /* all task on a per-semaphore list sleep on exactly
1062 * that semaphore
1063 */
1064 semcnt++;
Rik van Rielde2657f2013-05-09 16:59:59 -04001065 }
1066
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001067 /* Then: check the complex operations. */
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001068 list_for_each_entry(q, &sma->pending_alter, list) {
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001069 semcnt += check_qop(sma, semnum, q, count_zero);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 }
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001071 if (count_zero) {
1072 list_for_each_entry(q, &sma->pending_const, list) {
1073 semcnt += check_qop(sma, semnum, q, count_zero);
1074 }
Rik van Rielebc2e5e2013-05-09 16:53:28 -04001075 }
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001076 return semcnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077}
1078
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001079/* Free a semaphore set. freeary() is called with sem_ids.rwsem locked
1080 * as a writer and the spinlock for this semaphore set hold. sem_ids.rwsem
Nadia Derbey3e148c72007-10-18 23:40:54 -07001081 * remains locked on exit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 */
Pierre Peiffer01b8b072008-02-08 04:18:57 -08001083static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084{
Manfred Spraul380af1b2008-07-25 01:48:06 -07001085 struct sem_undo *un, *tu;
1086 struct sem_queue *q, *tq;
Pierre Peiffer01b8b072008-02-08 04:18:57 -08001087 struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001088 struct list_head tasks;
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001089 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090
Manfred Spraul380af1b2008-07-25 01:48:06 -07001091 /* Free the existing undo structures for this semaphore set. */
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -07001092 ipc_assert_locked_object(&sma->sem_perm);
Manfred Spraul380af1b2008-07-25 01:48:06 -07001093 list_for_each_entry_safe(un, tu, &sma->list_id, list_id) {
1094 list_del(&un->list_id);
1095 spin_lock(&un->ulp->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 un->semid = -1;
Manfred Spraul380af1b2008-07-25 01:48:06 -07001097 list_del_rcu(&un->list_proc);
1098 spin_unlock(&un->ulp->lock);
Lai Jiangshan693a8b62011-03-18 12:09:35 +08001099 kfree_rcu(un, rcu);
Manfred Spraul380af1b2008-07-25 01:48:06 -07001100 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101
1102 /* Wake up all pending processes and let them fail with EIDRM. */
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001103 INIT_LIST_HEAD(&tasks);
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001104 list_for_each_entry_safe(q, tq, &sma->pending_const, list) {
1105 unlink_queue(sma, q);
1106 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1107 }
1108
1109 list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
Manfred Spraulb97e8202009-12-15 16:47:32 -08001110 unlink_queue(sma, q);
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001111 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112 }
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001113 for (i = 0; i < sma->sem_nsems; i++) {
1114 struct sem *sem = sma->sem_base + i;
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001115 list_for_each_entry_safe(q, tq, &sem->pending_const, list) {
1116 unlink_queue(sma, q);
1117 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1118 }
1119 list_for_each_entry_safe(q, tq, &sem->pending_alter, list) {
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001120 unlink_queue(sma, q);
1121 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1122 }
1123 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124
Nadia Derbey7ca7e562007-10-18 23:40:48 -07001125 /* Remove the semaphore set from the IDR */
1126 sem_rmid(ns, sma);
Rik van Riel6062a8d2013-04-30 19:15:44 -07001127 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001128 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001130 wake_up_sem_queue_do(&tasks);
Kirill Korotaeve3893532006-10-02 02:18:22 -07001131 ns->used_sems -= sma->sem_nsems;
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -07001132 ipc_rcu_putref(sma, sem_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133}
1134
1135static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
1136{
Manfred Spraul239521f2014-01-27 17:07:04 -08001137 switch (version) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138 case IPC_64:
1139 return copy_to_user(buf, in, sizeof(*in));
1140 case IPC_OLD:
1141 {
1142 struct semid_ds out;
1143
Dan Rosenberg982f7c22010-09-30 15:15:31 -07001144 memset(&out, 0, sizeof(out));
1145
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
1147
1148 out.sem_otime = in->sem_otime;
1149 out.sem_ctime = in->sem_ctime;
1150 out.sem_nsems = in->sem_nsems;
1151
1152 return copy_to_user(buf, &out, sizeof(out));
1153 }
1154 default:
1155 return -EINVAL;
1156 }
1157}
1158
Manfred Sprauld12e1e52013-07-08 16:01:25 -07001159static time_t get_semotime(struct sem_array *sma)
1160{
1161 int i;
1162 time_t res;
1163
1164 res = sma->sem_base[0].sem_otime;
1165 for (i = 1; i < sma->sem_nsems; i++) {
1166 time_t to = sma->sem_base[i].sem_otime;
1167
1168 if (to > res)
1169 res = to;
1170 }
1171 return res;
1172}
1173
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -08001174static int semctl_nolock(struct ipc_namespace *ns, int semid,
Al Viroe1fd1f42013-03-05 15:04:55 -05001175 int cmd, int version, void __user *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176{
Amerigo Wange5cc9c72009-12-15 16:47:35 -08001177 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 struct sem_array *sma;
1179
Manfred Spraul239521f2014-01-27 17:07:04 -08001180 switch (cmd) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181 case IPC_INFO:
1182 case SEM_INFO:
1183 {
1184 struct seminfo seminfo;
1185 int max_id;
1186
1187 err = security_sem_semctl(NULL, cmd);
1188 if (err)
1189 return err;
Paul McQuade46c0a8c2014-06-06 14:37:37 -07001190
Manfred Spraul239521f2014-01-27 17:07:04 -08001191 memset(&seminfo, 0, sizeof(seminfo));
Kirill Korotaeve3893532006-10-02 02:18:22 -07001192 seminfo.semmni = ns->sc_semmni;
1193 seminfo.semmns = ns->sc_semmns;
1194 seminfo.semmsl = ns->sc_semmsl;
1195 seminfo.semopm = ns->sc_semopm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196 seminfo.semvmx = SEMVMX;
1197 seminfo.semmnu = SEMMNU;
1198 seminfo.semmap = SEMMAP;
1199 seminfo.semume = SEMUME;
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001200 down_read(&sem_ids(ns).rwsem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201 if (cmd == SEM_INFO) {
Kirill Korotaeve3893532006-10-02 02:18:22 -07001202 seminfo.semusz = sem_ids(ns).in_use;
1203 seminfo.semaem = ns->used_sems;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 } else {
1205 seminfo.semusz = SEMUSZ;
1206 seminfo.semaem = SEMAEM;
1207 }
Nadia Derbey7ca7e562007-10-18 23:40:48 -07001208 max_id = ipc_get_maxid(&sem_ids(ns));
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001209 up_read(&sem_ids(ns).rwsem);
Paul McQuade46c0a8c2014-06-06 14:37:37 -07001210 if (copy_to_user(p, &seminfo, sizeof(struct seminfo)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211 return -EFAULT;
Manfred Spraul239521f2014-01-27 17:07:04 -08001212 return (max_id < 0) ? 0 : max_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 }
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -08001214 case IPC_STAT:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 case SEM_STAT:
1216 {
1217 struct semid64_ds tbuf;
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001218 int id = 0;
1219
1220 memset(&tbuf, 0, sizeof(tbuf));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221
Linus Torvalds941b0302013-05-04 11:04:29 -07001222 rcu_read_lock();
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -08001223 if (cmd == SEM_STAT) {
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001224 sma = sem_obtain_object(ns, semid);
1225 if (IS_ERR(sma)) {
1226 err = PTR_ERR(sma);
1227 goto out_unlock;
1228 }
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -08001229 id = sma->sem_perm.id;
1230 } else {
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001231 sma = sem_obtain_object_check(ns, semid);
1232 if (IS_ERR(sma)) {
1233 err = PTR_ERR(sma);
1234 goto out_unlock;
1235 }
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -08001236 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237
1238 err = -EACCES;
Serge E. Hallynb0e77592011-03-23 16:43:24 -07001239 if (ipcperms(ns, &sma->sem_perm, S_IRUGO))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 goto out_unlock;
1241
1242 err = security_sem_semctl(sma, cmd);
1243 if (err)
1244 goto out_unlock;
1245
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
Manfred Sprauld12e1e52013-07-08 16:01:25 -07001247 tbuf.sem_otime = get_semotime(sma);
1248 tbuf.sem_ctime = sma->sem_ctime;
1249 tbuf.sem_nsems = sma->sem_nsems;
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001250 rcu_read_unlock();
Al Viroe1fd1f42013-03-05 15:04:55 -05001251 if (copy_semid_to_user(p, &tbuf, version))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 return -EFAULT;
1253 return id;
1254 }
1255 default:
1256 return -EINVAL;
1257 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258out_unlock:
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001259 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260 return err;
1261}
1262
Al Viroe1fd1f42013-03-05 15:04:55 -05001263static int semctl_setval(struct ipc_namespace *ns, int semid, int semnum,
1264 unsigned long arg)
1265{
1266 struct sem_undo *un;
1267 struct sem_array *sma;
Manfred Spraul239521f2014-01-27 17:07:04 -08001268 struct sem *curr;
Al Viroe1fd1f42013-03-05 15:04:55 -05001269 int err;
Al Viroe1fd1f42013-03-05 15:04:55 -05001270 struct list_head tasks;
1271 int val;
1272#if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN)
1273 /* big-endian 64bit */
1274 val = arg >> 32;
1275#else
1276 /* 32bit or little-endian 64bit */
1277 val = arg;
1278#endif
1279
Rik van Riel6062a8d2013-04-30 19:15:44 -07001280 if (val > SEMVMX || val < 0)
1281 return -ERANGE;
Al Viroe1fd1f42013-03-05 15:04:55 -05001282
1283 INIT_LIST_HEAD(&tasks);
Al Viroe1fd1f42013-03-05 15:04:55 -05001284
Rik van Riel6062a8d2013-04-30 19:15:44 -07001285 rcu_read_lock();
1286 sma = sem_obtain_object_check(ns, semid);
1287 if (IS_ERR(sma)) {
1288 rcu_read_unlock();
1289 return PTR_ERR(sma);
1290 }
1291
1292 if (semnum < 0 || semnum >= sma->sem_nsems) {
1293 rcu_read_unlock();
1294 return -EINVAL;
1295 }
1296
1297
1298 if (ipcperms(ns, &sma->sem_perm, S_IWUGO)) {
1299 rcu_read_unlock();
1300 return -EACCES;
1301 }
Al Viroe1fd1f42013-03-05 15:04:55 -05001302
1303 err = security_sem_semctl(sma, SETVAL);
Rik van Riel6062a8d2013-04-30 19:15:44 -07001304 if (err) {
1305 rcu_read_unlock();
1306 return -EACCES;
1307 }
Al Viroe1fd1f42013-03-05 15:04:55 -05001308
Rik van Riel6062a8d2013-04-30 19:15:44 -07001309 sem_lock(sma, NULL, -1);
Al Viroe1fd1f42013-03-05 15:04:55 -05001310
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001311 if (!ipc_valid_object(&sma->sem_perm)) {
Manfred Spraul6e224f92013-10-16 13:46:45 -07001312 sem_unlock(sma, -1);
1313 rcu_read_unlock();
1314 return -EIDRM;
1315 }
1316
Al Viroe1fd1f42013-03-05 15:04:55 -05001317 curr = &sma->sem_base[semnum];
1318
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -07001319 ipc_assert_locked_object(&sma->sem_perm);
Al Viroe1fd1f42013-03-05 15:04:55 -05001320 list_for_each_entry(un, &sma->list_id, list_id)
1321 un->semadj[semnum] = 0;
1322
1323 curr->semval = val;
1324 curr->sempid = task_tgid_vnr(current);
1325 sma->sem_ctime = get_seconds();
1326 /* maybe some queued-up processes were waiting for this */
1327 do_smart_update(sma, NULL, 0, 0, &tasks);
Rik van Riel6062a8d2013-04-30 19:15:44 -07001328 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001329 rcu_read_unlock();
Al Viroe1fd1f42013-03-05 15:04:55 -05001330 wake_up_sem_queue_do(&tasks);
Rik van Riel6062a8d2013-04-30 19:15:44 -07001331 return 0;
Al Viroe1fd1f42013-03-05 15:04:55 -05001332}
1333
Kirill Korotaeve3893532006-10-02 02:18:22 -07001334static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
Al Viroe1fd1f42013-03-05 15:04:55 -05001335 int cmd, void __user *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336{
1337 struct sem_array *sma;
Manfred Spraul239521f2014-01-27 17:07:04 -08001338 struct sem *curr;
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001339 int err, nsems;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 ushort fast_sem_io[SEMMSL_FAST];
Manfred Spraul239521f2014-01-27 17:07:04 -08001341 ushort *sem_io = fast_sem_io;
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001342 struct list_head tasks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001344 INIT_LIST_HEAD(&tasks);
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001345
1346 rcu_read_lock();
1347 sma = sem_obtain_object_check(ns, semid);
1348 if (IS_ERR(sma)) {
1349 rcu_read_unlock();
1350 return PTR_ERR(sma);
1351 }
1352
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353 nsems = sma->sem_nsems;
1354
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 err = -EACCES;
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001356 if (ipcperms(ns, &sma->sem_perm, cmd == SETALL ? S_IWUGO : S_IRUGO))
1357 goto out_rcu_wakeup;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358
1359 err = security_sem_semctl(sma, cmd);
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001360 if (err)
1361 goto out_rcu_wakeup;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362
1363 err = -EACCES;
1364 switch (cmd) {
1365 case GETALL:
1366 {
Al Viroe1fd1f42013-03-05 15:04:55 -05001367 ushort __user *array = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368 int i;
1369
Al Viroce857222013-05-03 00:30:49 +01001370 sem_lock(sma, NULL, -1);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001371 if (!ipc_valid_object(&sma->sem_perm)) {
Manfred Spraul6e224f92013-10-16 13:46:45 -07001372 err = -EIDRM;
1373 goto out_unlock;
1374 }
Manfred Spraul239521f2014-01-27 17:07:04 -08001375 if (nsems > SEMMSL_FAST) {
Al Viroce857222013-05-03 00:30:49 +01001376 if (!ipc_rcu_getref(sma)) {
Al Viroce857222013-05-03 00:30:49 +01001377 err = -EIDRM;
Manfred Spraul6e224f92013-10-16 13:46:45 -07001378 goto out_unlock;
Al Viroce857222013-05-03 00:30:49 +01001379 }
1380 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001381 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382 sem_io = ipc_alloc(sizeof(ushort)*nsems);
Manfred Spraul239521f2014-01-27 17:07:04 -08001383 if (sem_io == NULL) {
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -07001384 ipc_rcu_putref(sma, ipc_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 return -ENOMEM;
1386 }
1387
Linus Torvalds4091fd92013-05-04 10:13:40 -07001388 rcu_read_lock();
Pierre Peiffer6ff37972008-04-29 01:00:46 -07001389 sem_lock_and_putref(sma);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001390 if (!ipc_valid_object(&sma->sem_perm)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 err = -EIDRM;
Manfred Spraul6e224f92013-10-16 13:46:45 -07001392 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393 }
Al Viroce857222013-05-03 00:30:49 +01001394 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 for (i = 0; i < sma->sem_nsems; i++)
1396 sem_io[i] = sma->sem_base[i].semval;
Rik van Riel6062a8d2013-04-30 19:15:44 -07001397 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001398 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 err = 0;
Manfred Spraul239521f2014-01-27 17:07:04 -08001400 if (copy_to_user(array, sem_io, nsems*sizeof(ushort)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 err = -EFAULT;
1402 goto out_free;
1403 }
1404 case SETALL:
1405 {
1406 int i;
1407 struct sem_undo *un;
1408
Rik van Riel6062a8d2013-04-30 19:15:44 -07001409 if (!ipc_rcu_getref(sma)) {
Manfred Spraul6e224f92013-10-16 13:46:45 -07001410 err = -EIDRM;
1411 goto out_rcu_wakeup;
Rik van Riel6062a8d2013-04-30 19:15:44 -07001412 }
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001413 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414
Manfred Spraul239521f2014-01-27 17:07:04 -08001415 if (nsems > SEMMSL_FAST) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416 sem_io = ipc_alloc(sizeof(ushort)*nsems);
Manfred Spraul239521f2014-01-27 17:07:04 -08001417 if (sem_io == NULL) {
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -07001418 ipc_rcu_putref(sma, ipc_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 return -ENOMEM;
1420 }
1421 }
1422
Manfred Spraul239521f2014-01-27 17:07:04 -08001423 if (copy_from_user(sem_io, p, nsems*sizeof(ushort))) {
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -07001424 ipc_rcu_putref(sma, ipc_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425 err = -EFAULT;
1426 goto out_free;
1427 }
1428
1429 for (i = 0; i < nsems; i++) {
1430 if (sem_io[i] > SEMVMX) {
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -07001431 ipc_rcu_putref(sma, ipc_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432 err = -ERANGE;
1433 goto out_free;
1434 }
1435 }
Linus Torvalds4091fd92013-05-04 10:13:40 -07001436 rcu_read_lock();
Pierre Peiffer6ff37972008-04-29 01:00:46 -07001437 sem_lock_and_putref(sma);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001438 if (!ipc_valid_object(&sma->sem_perm)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439 err = -EIDRM;
Manfred Spraul6e224f92013-10-16 13:46:45 -07001440 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 }
1442
Davidlohr Buesoa5f4db82016-03-22 14:27:48 -07001443 for (i = 0; i < nsems; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444 sma->sem_base[i].semval = sem_io[i];
Davidlohr Buesoa5f4db82016-03-22 14:27:48 -07001445 sma->sem_base[i].sempid = task_tgid_vnr(current);
1446 }
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001447
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -07001448 ipc_assert_locked_object(&sma->sem_perm);
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001449 list_for_each_entry(un, &sma->list_id, list_id) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 for (i = 0; i < nsems; i++)
1451 un->semadj[i] = 0;
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001452 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 sma->sem_ctime = get_seconds();
1454 /* maybe some queued-up processes were waiting for this */
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001455 do_smart_update(sma, NULL, 0, 0, &tasks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456 err = 0;
1457 goto out_unlock;
1458 }
Al Viroe1fd1f42013-03-05 15:04:55 -05001459 /* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 }
1461 err = -EINVAL;
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001462 if (semnum < 0 || semnum >= nsems)
1463 goto out_rcu_wakeup;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464
Rik van Riel6062a8d2013-04-30 19:15:44 -07001465 sem_lock(sma, NULL, -1);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001466 if (!ipc_valid_object(&sma->sem_perm)) {
Manfred Spraul6e224f92013-10-16 13:46:45 -07001467 err = -EIDRM;
1468 goto out_unlock;
1469 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 curr = &sma->sem_base[semnum];
1471
1472 switch (cmd) {
1473 case GETVAL:
1474 err = curr->semval;
1475 goto out_unlock;
1476 case GETPID:
1477 err = curr->sempid;
1478 goto out_unlock;
1479 case GETNCNT:
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001480 err = count_semcnt(sma, semnum, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 goto out_unlock;
1482 case GETZCNT:
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001483 err = count_semcnt(sma, semnum, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 }
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001486
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487out_unlock:
Rik van Riel6062a8d2013-04-30 19:15:44 -07001488 sem_unlock(sma, -1);
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001489out_rcu_wakeup:
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001490 rcu_read_unlock();
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001491 wake_up_sem_queue_do(&tasks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492out_free:
Manfred Spraul239521f2014-01-27 17:07:04 -08001493 if (sem_io != fast_sem_io)
Tetsuo Handa1d5cfdb2016-01-22 15:11:02 -08001494 ipc_free(sem_io);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495 return err;
1496}
1497
Pierre Peiffer016d7132008-04-29 01:00:50 -07001498static inline unsigned long
1499copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500{
Manfred Spraul239521f2014-01-27 17:07:04 -08001501 switch (version) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 case IPC_64:
Pierre Peiffer016d7132008-04-29 01:00:50 -07001503 if (copy_from_user(out, buf, sizeof(*out)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506 case IPC_OLD:
1507 {
1508 struct semid_ds tbuf_old;
1509
Manfred Spraul239521f2014-01-27 17:07:04 -08001510 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 return -EFAULT;
1512
Pierre Peiffer016d7132008-04-29 01:00:50 -07001513 out->sem_perm.uid = tbuf_old.sem_perm.uid;
1514 out->sem_perm.gid = tbuf_old.sem_perm.gid;
1515 out->sem_perm.mode = tbuf_old.sem_perm.mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516
1517 return 0;
1518 }
1519 default:
1520 return -EINVAL;
1521 }
1522}
1523
Pierre Peiffer522bb2a2008-04-29 01:00:49 -07001524/*
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001525 * This function handles some semctl commands which require the rwsem
Pierre Peiffer522bb2a2008-04-29 01:00:49 -07001526 * to be held in write mode.
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001527 * NOTE: no locks must be held, the rwsem is taken inside this function.
Pierre Peiffer522bb2a2008-04-29 01:00:49 -07001528 */
Pierre Peiffer21a48262008-04-29 01:00:49 -07001529static int semctl_down(struct ipc_namespace *ns, int semid,
Al Viroe1fd1f42013-03-05 15:04:55 -05001530 int cmd, int version, void __user *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531{
1532 struct sem_array *sma;
1533 int err;
Pierre Peiffer016d7132008-04-29 01:00:50 -07001534 struct semid64_ds semid64;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535 struct kern_ipc_perm *ipcp;
1536
Manfred Spraul239521f2014-01-27 17:07:04 -08001537 if (cmd == IPC_SET) {
Al Viroe1fd1f42013-03-05 15:04:55 -05001538 if (copy_semid_from_user(&semid64, p, version))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001542 down_write(&sem_ids(ns).rwsem);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001543 rcu_read_lock();
1544
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001545 ipcp = ipcctl_pre_down_nolock(ns, &sem_ids(ns), semid, cmd,
1546 &semid64.sem_perm, 0);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001547 if (IS_ERR(ipcp)) {
1548 err = PTR_ERR(ipcp);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001549 goto out_unlock1;
1550 }
Steve Grubb073115d2006-04-02 17:07:33 -04001551
Pierre Peiffera5f75e72008-04-29 01:00:54 -07001552 sma = container_of(ipcp, struct sem_array, sem_perm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553
1554 err = security_sem_semctl(sma, cmd);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001555 if (err)
1556 goto out_unlock1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001558 switch (cmd) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559 case IPC_RMID:
Rik van Riel6062a8d2013-04-30 19:15:44 -07001560 sem_lock(sma, NULL, -1);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001561 /* freeary unlocks the ipc object and rcu */
Pierre Peiffer01b8b072008-02-08 04:18:57 -08001562 freeary(ns, ipcp);
Pierre Peiffer522bb2a2008-04-29 01:00:49 -07001563 goto out_up;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 case IPC_SET:
Rik van Riel6062a8d2013-04-30 19:15:44 -07001565 sem_lock(sma, NULL, -1);
Eric W. Biederman1efdb692012-02-07 16:54:11 -08001566 err = ipc_update_perm(&semid64.sem_perm, ipcp);
1567 if (err)
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001568 goto out_unlock0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 sma->sem_ctime = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570 break;
1571 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572 err = -EINVAL;
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001573 goto out_unlock1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001576out_unlock0:
Rik van Riel6062a8d2013-04-30 19:15:44 -07001577 sem_unlock(sma, -1);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001578out_unlock1:
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001579 rcu_read_unlock();
Pierre Peiffer522bb2a2008-04-29 01:00:49 -07001580out_up:
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001581 up_write(&sem_ids(ns).rwsem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 return err;
1583}
1584
Al Viroe1fd1f42013-03-05 15:04:55 -05001585SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, unsigned long, arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587 int version;
Kirill Korotaeve3893532006-10-02 02:18:22 -07001588 struct ipc_namespace *ns;
Al Viroe1fd1f42013-03-05 15:04:55 -05001589 void __user *p = (void __user *)arg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590
1591 if (semid < 0)
1592 return -EINVAL;
1593
1594 version = ipc_parse_version(&cmd);
Kirill Korotaeve3893532006-10-02 02:18:22 -07001595 ns = current->nsproxy->ipc_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596
Manfred Spraul239521f2014-01-27 17:07:04 -08001597 switch (cmd) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 case IPC_INFO:
1599 case SEM_INFO:
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -08001600 case IPC_STAT:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601 case SEM_STAT:
Al Viroe1fd1f42013-03-05 15:04:55 -05001602 return semctl_nolock(ns, semid, cmd, version, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603 case GETALL:
1604 case GETVAL:
1605 case GETPID:
1606 case GETNCNT:
1607 case GETZCNT:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608 case SETALL:
Al Viroe1fd1f42013-03-05 15:04:55 -05001609 return semctl_main(ns, semid, semnum, cmd, p);
1610 case SETVAL:
1611 return semctl_setval(ns, semid, semnum, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612 case IPC_RMID:
1613 case IPC_SET:
Al Viroe1fd1f42013-03-05 15:04:55 -05001614 return semctl_down(ns, semid, cmd, version, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615 default:
1616 return -EINVAL;
1617 }
1618}
1619
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620/* If the task doesn't already have a undo_list, then allocate one
1621 * here. We guarantee there is only one thread using this undo list,
1622 * and current is THE ONE
1623 *
1624 * If this allocation and assignment succeeds, but later
1625 * portions of this code fail, there is no need to free the sem_undo_list.
1626 * Just let it stay associated with the task, and it'll be freed later
1627 * at exit time.
1628 *
1629 * This can block, so callers must hold no locks.
1630 */
1631static inline int get_undo_list(struct sem_undo_list **undo_listp)
1632{
1633 struct sem_undo_list *undo_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634
1635 undo_list = current->sysvsem.undo_list;
1636 if (!undo_list) {
Matt Helsley2453a302006-10-02 02:18:25 -07001637 undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638 if (undo_list == NULL)
1639 return -ENOMEM;
Ingo Molnar00a5dfd2005-08-05 23:05:27 +02001640 spin_lock_init(&undo_list->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641 atomic_set(&undo_list->refcnt, 1);
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001642 INIT_LIST_HEAD(&undo_list->list_proc);
1643
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 current->sysvsem.undo_list = undo_list;
1645 }
1646 *undo_listp = undo_list;
1647 return 0;
1648}
1649
Nick Pigginbf17bb72009-12-15 16:47:28 -08001650static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651{
Nick Pigginbf17bb72009-12-15 16:47:28 -08001652 struct sem_undo *un;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653
Nick Pigginbf17bb72009-12-15 16:47:28 -08001654 list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) {
1655 if (un->semid == semid)
1656 return un;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 }
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001658 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659}
1660
Nick Pigginbf17bb72009-12-15 16:47:28 -08001661static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1662{
1663 struct sem_undo *un;
1664
Manfred Spraul239521f2014-01-27 17:07:04 -08001665 assert_spin_locked(&ulp->lock);
Nick Pigginbf17bb72009-12-15 16:47:28 -08001666
1667 un = __lookup_undo(ulp, semid);
1668 if (un) {
1669 list_del_rcu(&un->list_proc);
1670 list_add_rcu(&un->list_proc, &ulp->list_proc);
1671 }
1672 return un;
1673}
1674
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001675/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -08001676 * find_alloc_undo - lookup (and if not present create) undo array
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001677 * @ns: namespace
1678 * @semid: semaphore array id
1679 *
1680 * The function looks up (and if not present creates) the undo structure.
1681 * The size of the undo structure depends on the size of the semaphore
1682 * array, thus the alloc path is not that straightforward.
Manfred Spraul380af1b2008-07-25 01:48:06 -07001683 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1684 * performs a rcu_read_lock().
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001685 */
1686static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687{
1688 struct sem_array *sma;
1689 struct sem_undo_list *ulp;
1690 struct sem_undo *un, *new;
Rik van Riel6062a8d2013-04-30 19:15:44 -07001691 int nsems, error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692
1693 error = get_undo_list(&ulp);
1694 if (error)
1695 return ERR_PTR(error);
1696
Manfred Spraul380af1b2008-07-25 01:48:06 -07001697 rcu_read_lock();
Pierre Peifferc530c6a2007-10-18 23:40:55 -07001698 spin_lock(&ulp->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 un = lookup_undo(ulp, semid);
Pierre Peifferc530c6a2007-10-18 23:40:55 -07001700 spin_unlock(&ulp->lock);
Manfred Spraul239521f2014-01-27 17:07:04 -08001701 if (likely(un != NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702 goto out;
1703
1704 /* no undo structure around - allocate one. */
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001705 /* step 1: figure out the size of the semaphore array */
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001706 sma = sem_obtain_object_check(ns, semid);
1707 if (IS_ERR(sma)) {
1708 rcu_read_unlock();
Julia Lawall4de85cd2010-05-26 14:43:44 -07001709 return ERR_CAST(sma);
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001710 }
Nadia Derbey023a5352007-10-18 23:40:51 -07001711
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712 nsems = sma->sem_nsems;
Rik van Riel6062a8d2013-04-30 19:15:44 -07001713 if (!ipc_rcu_getref(sma)) {
1714 rcu_read_unlock();
1715 un = ERR_PTR(-EIDRM);
1716 goto out;
1717 }
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001718 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001720 /* step 2: allocate new undo structure */
Burman Yan4668edc2006-12-06 20:38:51 -08001721 new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722 if (!new) {
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -07001723 ipc_rcu_putref(sma, ipc_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724 return ERR_PTR(-ENOMEM);
1725 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726
Manfred Spraul380af1b2008-07-25 01:48:06 -07001727 /* step 3: Acquire the lock on semaphore array */
Linus Torvalds4091fd92013-05-04 10:13:40 -07001728 rcu_read_lock();
Pierre Peiffer6ff37972008-04-29 01:00:46 -07001729 sem_lock_and_putref(sma);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001730 if (!ipc_valid_object(&sma->sem_perm)) {
Rik van Riel6062a8d2013-04-30 19:15:44 -07001731 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001732 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733 kfree(new);
1734 un = ERR_PTR(-EIDRM);
1735 goto out;
1736 }
Manfred Spraul380af1b2008-07-25 01:48:06 -07001737 spin_lock(&ulp->lock);
1738
1739 /*
1740 * step 4: check for races: did someone else allocate the undo struct?
1741 */
1742 un = lookup_undo(ulp, semid);
1743 if (un) {
1744 kfree(new);
1745 goto success;
1746 }
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001747 /* step 5: initialize & link new undo structure */
1748 new->semadj = (short *) &new[1];
Manfred Spraul380af1b2008-07-25 01:48:06 -07001749 new->ulp = ulp;
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001750 new->semid = semid;
1751 assert_spin_locked(&ulp->lock);
Manfred Spraul380af1b2008-07-25 01:48:06 -07001752 list_add_rcu(&new->list_proc, &ulp->list_proc);
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -07001753 ipc_assert_locked_object(&sma->sem_perm);
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001754 list_add(&new->list_id, &sma->list_id);
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001755 un = new;
Manfred Spraul380af1b2008-07-25 01:48:06 -07001756
1757success:
1758 spin_unlock(&ulp->lock);
Rik van Riel6062a8d2013-04-30 19:15:44 -07001759 sem_unlock(sma, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760out:
1761 return un;
1762}
1763
Manfred Spraulc61284e2010-07-20 13:24:23 -07001764
1765/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -08001766 * get_queue_result - retrieve the result code from sem_queue
Manfred Spraulc61284e2010-07-20 13:24:23 -07001767 * @q: Pointer to queue structure
1768 *
1769 * Retrieve the return code from the pending queue. If IN_WAKEUP is found in
1770 * q->status, then we must loop until the value is replaced with the final
1771 * value: This may happen if a task is woken up by an unrelated event (e.g.
1772 * signal) and in parallel the task is woken up by another task because it got
1773 * the requested semaphores.
1774 *
1775 * The function can be called with or without holding the semaphore spinlock.
1776 */
1777static int get_queue_result(struct sem_queue *q)
1778{
1779 int error;
1780
1781 error = q->status;
1782 while (unlikely(error == IN_WAKEUP)) {
1783 cpu_relax();
1784 error = q->status;
1785 }
1786
1787 return error;
1788}
1789
Heiko Carstensd5460c92009-01-14 14:14:27 +01001790SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
1791 unsigned, nsops, const struct timespec __user *, timeout)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792{
1793 int error = -EINVAL;
1794 struct sem_array *sma;
1795 struct sembuf fast_sops[SEMOPM_FAST];
Manfred Spraul239521f2014-01-27 17:07:04 -08001796 struct sembuf *sops = fast_sops, *sop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797 struct sem_undo *un;
Rik van Riel6062a8d2013-04-30 19:15:44 -07001798 int undos = 0, alter = 0, max, locknum;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799 struct sem_queue queue;
1800 unsigned long jiffies_left = 0;
Kirill Korotaeve3893532006-10-02 02:18:22 -07001801 struct ipc_namespace *ns;
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001802 struct list_head tasks;
Kirill Korotaeve3893532006-10-02 02:18:22 -07001803
1804 ns = current->nsproxy->ipc_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805
1806 if (nsops < 1 || semid < 0)
1807 return -EINVAL;
Kirill Korotaeve3893532006-10-02 02:18:22 -07001808 if (nsops > ns->sc_semopm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809 return -E2BIG;
Manfred Spraul239521f2014-01-27 17:07:04 -08001810 if (nsops > SEMOPM_FAST) {
1811 sops = kmalloc(sizeof(*sops)*nsops, GFP_KERNEL);
1812 if (sops == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813 return -ENOMEM;
1814 }
Manfred Spraul239521f2014-01-27 17:07:04 -08001815 if (copy_from_user(sops, tsops, nsops * sizeof(*tsops))) {
1816 error = -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817 goto out_free;
1818 }
1819 if (timeout) {
1820 struct timespec _timeout;
1821 if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1822 error = -EFAULT;
1823 goto out_free;
1824 }
1825 if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1826 _timeout.tv_nsec >= 1000000000L) {
1827 error = -EINVAL;
1828 goto out_free;
1829 }
1830 jiffies_left = timespec_to_jiffies(&_timeout);
1831 }
1832 max = 0;
1833 for (sop = sops; sop < sops + nsops; sop++) {
1834 if (sop->sem_num >= max)
1835 max = sop->sem_num;
1836 if (sop->sem_flg & SEM_UNDO)
Manfred Spraulb78755a2005-06-23 00:10:06 -07001837 undos = 1;
1838 if (sop->sem_op != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839 alter = 1;
1840 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841
Rik van Riel6062a8d2013-04-30 19:15:44 -07001842 INIT_LIST_HEAD(&tasks);
1843
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844 if (undos) {
Rik van Riel6062a8d2013-04-30 19:15:44 -07001845 /* On success, find_alloc_undo takes the rcu_read_lock */
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001846 un = find_alloc_undo(ns, semid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847 if (IS_ERR(un)) {
1848 error = PTR_ERR(un);
1849 goto out_free;
1850 }
Rik van Riel6062a8d2013-04-30 19:15:44 -07001851 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852 un = NULL;
Rik van Riel6062a8d2013-04-30 19:15:44 -07001853 rcu_read_lock();
1854 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001856 sma = sem_obtain_object_check(ns, semid);
Nadia Derbey023a5352007-10-18 23:40:51 -07001857 if (IS_ERR(sma)) {
Rik van Riel6062a8d2013-04-30 19:15:44 -07001858 rcu_read_unlock();
Nadia Derbey023a5352007-10-18 23:40:51 -07001859 error = PTR_ERR(sma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860 goto out_free;
Nadia Derbey023a5352007-10-18 23:40:51 -07001861 }
1862
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001863 error = -EFBIG;
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001864 if (max >= sma->sem_nsems)
1865 goto out_rcu_wakeup;
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001866
1867 error = -EACCES;
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001868 if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO))
1869 goto out_rcu_wakeup;
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001870
1871 error = security_sem_semop(sma, sops, nsops, alter);
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001872 if (error)
1873 goto out_rcu_wakeup;
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001874
Manfred Spraul6e224f92013-10-16 13:46:45 -07001875 error = -EIDRM;
1876 locknum = sem_lock(sma, sops, nsops);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001877 /*
1878 * We eventually might perform the following check in a lockless
1879 * fashion, considering ipc_valid_object() locking constraints.
1880 * If nsops == 1 and there is no contention for sem_perm.lock, then
1881 * only a per-semaphore lock is held and it's OK to proceed with the
1882 * check below. More details on the fine grained locking scheme
1883 * entangled here and why it's RMID race safe on comments at sem_lock()
1884 */
1885 if (!ipc_valid_object(&sma->sem_perm))
Manfred Spraul6e224f92013-10-16 13:46:45 -07001886 goto out_unlock_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887 /*
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001888 * semid identifiers are not unique - find_alloc_undo may have
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889 * allocated an undo structure, it was invalidated by an RMID
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001890 * and now a new array with received the same id. Check and fail.
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001891 * This case can be detected checking un->semid. The existence of
Manfred Spraul380af1b2008-07-25 01:48:06 -07001892 * "un" itself is guaranteed by rcu.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893 */
Rik van Riel6062a8d2013-04-30 19:15:44 -07001894 if (un && un->semid == -1)
1895 goto out_unlock_free;
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001896
Manfred Sprauld198cd62014-06-06 14:37:49 -07001897 queue.sops = sops;
1898 queue.nsops = nsops;
1899 queue.undo = un;
1900 queue.pid = task_tgid_vnr(current);
1901 queue.alter = alter;
1902
1903 error = perform_atomic_semop(sma, &queue);
Manfred Spraul0e8c6652013-09-30 13:45:25 -07001904 if (error == 0) {
1905 /* If the operation was successful, then do
1906 * the required updates.
1907 */
1908 if (alter)
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001909 do_smart_update(sma, sops, nsops, 1, &tasks);
Manfred Spraul0e8c6652013-09-30 13:45:25 -07001910 else
1911 set_semotime(sma, sops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912 }
Manfred Spraul0e8c6652013-09-30 13:45:25 -07001913 if (error <= 0)
1914 goto out_unlock_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915
1916 /* We need to sleep on this operation, so we put the current
1917 * task into the pending queue and go to sleep.
1918 */
Paul McQuade46c0a8c2014-06-06 14:37:37 -07001919
Manfred Spraulb97e8202009-12-15 16:47:32 -08001920 if (nsops == 1) {
1921 struct sem *curr;
1922 curr = &sma->sem_base[sops->sem_num];
1923
Manfred Spraulf269f402013-07-08 16:01:24 -07001924 if (alter) {
1925 if (sma->complex_count) {
1926 list_add_tail(&queue.list,
1927 &sma->pending_alter);
1928 } else {
1929
1930 list_add_tail(&queue.list,
1931 &curr->pending_alter);
1932 }
1933 } else {
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001934 list_add_tail(&queue.list, &curr->pending_const);
Manfred Spraulf269f402013-07-08 16:01:24 -07001935 }
Manfred Spraulb97e8202009-12-15 16:47:32 -08001936 } else {
Manfred Spraulf269f402013-07-08 16:01:24 -07001937 if (!sma->complex_count)
1938 merge_queues(sma);
1939
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001940 if (alter)
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001941 list_add_tail(&queue.list, &sma->pending_alter);
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001942 else
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001943 list_add_tail(&queue.list, &sma->pending_const);
1944
Manfred Spraulb97e8202009-12-15 16:47:32 -08001945 sma->complex_count++;
1946 }
1947
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948 queue.status = -EINTR;
1949 queue.sleeper = current;
Manfred Spraul0b0577f2011-11-02 13:38:52 -07001950
1951sleep_again:
Davidlohr Bueso52644c92015-02-17 13:47:55 -08001952 __set_current_state(TASK_INTERRUPTIBLE);
Rik van Riel6062a8d2013-04-30 19:15:44 -07001953 sem_unlock(sma, locknum);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001954 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955
1956 if (timeout)
1957 jiffies_left = schedule_timeout(jiffies_left);
1958 else
1959 schedule();
1960
Manfred Spraulc61284e2010-07-20 13:24:23 -07001961 error = get_queue_result(&queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962
1963 if (error != -EINTR) {
1964 /* fast path: update_queue already obtained all requested
Manfred Spraulc61284e2010-07-20 13:24:23 -07001965 * resources.
1966 * Perform a smp_mb(): User space could assume that semop()
1967 * is a memory barrier: Without the mb(), the cpu could
1968 * speculatively read in user space stale data that was
1969 * overwritten by the previous owner of the semaphore.
1970 */
1971 smp_mb();
1972
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973 goto out_free;
1974 }
1975
Linus Torvalds321310c2013-05-04 10:47:57 -07001976 rcu_read_lock();
Rik van Riel6062a8d2013-04-30 19:15:44 -07001977 sma = sem_obtain_lock(ns, semid, sops, nsops, &locknum);
Manfred Sprauld694ad62011-07-25 17:11:47 -07001978
1979 /*
1980 * Wait until it's guaranteed that no wakeup_sem_queue_do() is ongoing.
1981 */
1982 error = get_queue_result(&queue);
1983
1984 /*
1985 * Array removed? If yes, leave without sem_unlock().
1986 */
Nadia Derbey023a5352007-10-18 23:40:51 -07001987 if (IS_ERR(sma)) {
Linus Torvalds321310c2013-05-04 10:47:57 -07001988 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989 goto out_free;
1990 }
1991
Manfred Spraulc61284e2010-07-20 13:24:23 -07001992
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993 /*
Manfred Sprauld694ad62011-07-25 17:11:47 -07001994 * If queue.status != -EINTR we are woken up by another process.
1995 * Leave without unlink_queue(), but with sem_unlock().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996 */
Davidlohr Bueso3ab08fe2014-01-27 17:07:06 -08001997 if (error != -EINTR)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998 goto out_unlock_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999
2000 /*
2001 * If an interrupt occurred we have to clean up the queue
2002 */
2003 if (timeout && jiffies_left == 0)
2004 error = -EAGAIN;
Manfred Spraul0b0577f2011-11-02 13:38:52 -07002005
2006 /*
2007 * If the wakeup was spurious, just retry
2008 */
2009 if (error == -EINTR && !signal_pending(current))
2010 goto sleep_again;
2011
Manfred Spraulb97e8202009-12-15 16:47:32 -08002012 unlink_queue(sma, &queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013
2014out_unlock_free:
Rik van Riel6062a8d2013-04-30 19:15:44 -07002015 sem_unlock(sma, locknum);
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07002016out_rcu_wakeup:
Linus Torvalds6d49dab2013-05-03 15:04:40 -07002017 rcu_read_unlock();
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07002018 wake_up_sem_queue_do(&tasks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019out_free:
Manfred Spraul239521f2014-01-27 17:07:04 -08002020 if (sops != fast_sops)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021 kfree(sops);
2022 return error;
2023}
2024
Heiko Carstensd5460c92009-01-14 14:14:27 +01002025SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
2026 unsigned, nsops)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027{
2028 return sys_semtimedop(semid, tsops, nsops, NULL);
2029}
2030
2031/* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
2032 * parent and child tasks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 */
2034
2035int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
2036{
2037 struct sem_undo_list *undo_list;
2038 int error;
2039
2040 if (clone_flags & CLONE_SYSVSEM) {
2041 error = get_undo_list(&undo_list);
2042 if (error)
2043 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044 atomic_inc(&undo_list->refcnt);
2045 tsk->sysvsem.undo_list = undo_list;
Paul McQuade46c0a8c2014-06-06 14:37:37 -07002046 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 tsk->sysvsem.undo_list = NULL;
2048
2049 return 0;
2050}
2051
2052/*
2053 * add semadj values to semaphores, free undo structures.
2054 * undo structures are not freed when semaphore arrays are destroyed
2055 * so some of them may be out of date.
2056 * IMPLEMENTATION NOTE: There is some confusion over whether the
2057 * set of adjustments that needs to be done should be done in an atomic
2058 * manner or not. That is, if we are attempting to decrement the semval
2059 * should we queue up and wait until we can do so legally?
2060 * The original implementation attempted to do this (queue and wait).
2061 * The current implementation does not do so. The POSIX standard
2062 * and SVID should be consulted to determine what behavior is mandated.
2063 */
2064void exit_sem(struct task_struct *tsk)
2065{
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002066 struct sem_undo_list *ulp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002068 ulp = tsk->sysvsem.undo_list;
2069 if (!ulp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070 return;
Manfred Spraul9edff4a2008-04-29 01:00:57 -07002071 tsk->sysvsem.undo_list = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002073 if (!atomic_dec_and_test(&ulp->refcnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074 return;
2075
Manfred Spraul380af1b2008-07-25 01:48:06 -07002076 for (;;) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002077 struct sem_array *sma;
Manfred Spraul380af1b2008-07-25 01:48:06 -07002078 struct sem_undo *un;
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07002079 struct list_head tasks;
Rik van Riel6062a8d2013-04-30 19:15:44 -07002080 int semid, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081
Manfred Spraul380af1b2008-07-25 01:48:06 -07002082 rcu_read_lock();
Jiri Pirko05725f72009-04-14 20:17:16 +02002083 un = list_entry_rcu(ulp->list_proc.next,
2084 struct sem_undo, list_proc);
Herton R. Krzesinski602b8592015-08-14 15:35:02 -07002085 if (&un->list_proc == &ulp->list_proc) {
2086 /*
2087 * We must wait for freeary() before freeing this ulp,
2088 * in case we raced with last sem_undo. There is a small
2089 * possibility where we exit while freeary() didn't
2090 * finish unlocking sem_undo_list.
2091 */
2092 spin_unlock_wait(&ulp->lock);
Rik van Riel6062a8d2013-04-30 19:15:44 -07002093 rcu_read_unlock();
Manfred Spraul380af1b2008-07-25 01:48:06 -07002094 break;
Rik van Riel6062a8d2013-04-30 19:15:44 -07002095 }
Herton R. Krzesinski602b8592015-08-14 15:35:02 -07002096 spin_lock(&ulp->lock);
2097 semid = un->semid;
2098 spin_unlock(&ulp->lock);
Manfred Spraul380af1b2008-07-25 01:48:06 -07002099
Herton R. Krzesinski602b8592015-08-14 15:35:02 -07002100 /* exit_sem raced with IPC_RMID, nothing to do */
2101 if (semid == -1) {
2102 rcu_read_unlock();
2103 continue;
2104 }
2105
2106 sma = sem_obtain_object_check(tsk->nsproxy->ipc_ns, semid);
Manfred Spraul380af1b2008-07-25 01:48:06 -07002107 /* exit_sem raced with IPC_RMID, nothing to do */
Rik van Riel6062a8d2013-04-30 19:15:44 -07002108 if (IS_ERR(sma)) {
2109 rcu_read_unlock();
Manfred Spraul380af1b2008-07-25 01:48:06 -07002110 continue;
Rik van Riel6062a8d2013-04-30 19:15:44 -07002111 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112
Rik van Riel6062a8d2013-04-30 19:15:44 -07002113 sem_lock(sma, NULL, -1);
Manfred Spraul6e224f92013-10-16 13:46:45 -07002114 /* exit_sem raced with IPC_RMID, nothing to do */
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08002115 if (!ipc_valid_object(&sma->sem_perm)) {
Manfred Spraul6e224f92013-10-16 13:46:45 -07002116 sem_unlock(sma, -1);
2117 rcu_read_unlock();
2118 continue;
2119 }
Nick Pigginbf17bb72009-12-15 16:47:28 -08002120 un = __lookup_undo(ulp, semid);
Manfred Spraul380af1b2008-07-25 01:48:06 -07002121 if (un == NULL) {
2122 /* exit_sem raced with IPC_RMID+semget() that created
2123 * exactly the same semid. Nothing to do.
2124 */
Rik van Riel6062a8d2013-04-30 19:15:44 -07002125 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07002126 rcu_read_unlock();
Manfred Spraul380af1b2008-07-25 01:48:06 -07002127 continue;
2128 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129
Manfred Spraul380af1b2008-07-25 01:48:06 -07002130 /* remove un from the linked lists */
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -07002131 ipc_assert_locked_object(&sma->sem_perm);
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002132 list_del(&un->list_id);
2133
Herton R. Krzesinskia9795582015-08-14 15:35:05 -07002134 /* we are the last process using this ulp, acquiring ulp->lock
2135 * isn't required. Besides that, we are also protected against
2136 * IPC_RMID as we hold sma->sem_perm lock now
2137 */
Manfred Spraul380af1b2008-07-25 01:48:06 -07002138 list_del_rcu(&un->list_proc);
Manfred Spraul380af1b2008-07-25 01:48:06 -07002139
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002140 /* perform adjustments registered in un */
2141 for (i = 0; i < sma->sem_nsems; i++) {
Manfred Spraul239521f2014-01-27 17:07:04 -08002142 struct sem *semaphore = &sma->sem_base[i];
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002143 if (un->semadj[i]) {
2144 semaphore->semval += un->semadj[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145 /*
2146 * Range checks of the new semaphore value,
2147 * not defined by sus:
2148 * - Some unices ignore the undo entirely
2149 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
2150 * - some cap the value (e.g. FreeBSD caps
2151 * at 0, but doesn't enforce SEMVMX)
2152 *
2153 * Linux caps the semaphore value, both at 0
2154 * and at SEMVMX.
2155 *
Manfred Spraul239521f2014-01-27 17:07:04 -08002156 * Manfred <manfred@colorfullife.com>
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157 */
Ingo Molnar5f921ae2006-03-26 01:37:17 -08002158 if (semaphore->semval < 0)
2159 semaphore->semval = 0;
2160 if (semaphore->semval > SEMVMX)
2161 semaphore->semval = SEMVMX;
Pavel Emelyanovb4888932007-10-18 23:40:14 -07002162 semaphore->sempid = task_tgid_vnr(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163 }
2164 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165 /* maybe some queued-up processes were waiting for this */
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07002166 INIT_LIST_HEAD(&tasks);
2167 do_smart_update(sma, NULL, 0, 1, &tasks);
Rik van Riel6062a8d2013-04-30 19:15:44 -07002168 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07002169 rcu_read_unlock();
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07002170 wake_up_sem_queue_do(&tasks);
Manfred Spraul380af1b2008-07-25 01:48:06 -07002171
Lai Jiangshan693a8b62011-03-18 12:09:35 +08002172 kfree_rcu(un, rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002173 }
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002174 kfree(ulp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175}
2176
2177#ifdef CONFIG_PROC_FS
Mike Waychison19b49462005-09-06 15:17:10 -07002178static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179{
Eric W. Biederman1efdb692012-02-07 16:54:11 -08002180 struct user_namespace *user_ns = seq_user_ns(s);
Mike Waychison19b49462005-09-06 15:17:10 -07002181 struct sem_array *sma = it;
Manfred Sprauld12e1e52013-07-08 16:01:25 -07002182 time_t sem_otime;
2183
Manfred Sprauld8c63372013-09-30 13:45:07 -07002184 /*
2185 * The proc interface isn't aware of sem_lock(), it calls
2186 * ipc_lock_object() directly (in sysvipc_find_ipc).
2187 * In order to stay compatible with sem_lock(), we must wait until
2188 * all simple semop() calls have left their critical regions.
2189 */
2190 sem_wait_array(sma);
2191
Manfred Sprauld12e1e52013-07-08 16:01:25 -07002192 sem_otime = get_semotime(sma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002193
Joe Perches7f032d62015-04-15 16:17:54 -07002194 seq_printf(s,
2195 "%10d %10d %4o %10u %5u %5u %5u %5u %10lu %10lu\n",
2196 sma->sem_perm.key,
2197 sma->sem_perm.id,
2198 sma->sem_perm.mode,
2199 sma->sem_nsems,
2200 from_kuid_munged(user_ns, sma->sem_perm.uid),
2201 from_kgid_munged(user_ns, sma->sem_perm.gid),
2202 from_kuid_munged(user_ns, sma->sem_perm.cuid),
2203 from_kgid_munged(user_ns, sma->sem_perm.cgid),
2204 sem_otime,
2205 sma->sem_ctime);
2206
2207 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002208}
2209#endif