blob: e08b948519223e62aca265284483ee7b1ef1420c [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>
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -080014 * (c) 2016 Davidlohr Bueso <dave@stgolabs.net>
Manfred Spraulc5cf6352010-05-26 14:43:43 -070015 * Further wakeup optimizations, documentation
16 * (c) 2010 Manfred Spraul <manfred@colorfullife.com>
Steve Grubb073115d2006-04-02 17:07:33 -040017 *
18 * support for audit of ipc object properties and permission changes
19 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
Kirill Korotaeve3893532006-10-02 02:18:22 -070020 *
21 * namespaces support
22 * OpenVZ, SWsoft Inc.
23 * Pavel Emelianov <xemul@openvz.org>
Manfred Spraulc5cf6352010-05-26 14:43:43 -070024 *
25 * Implementation notes: (May 2010)
26 * This file implements System V semaphores.
27 *
28 * User space visible behavior:
29 * - FIFO ordering for semop() operations (just FIFO, not starvation
30 * protection)
31 * - multiple semaphore operations that alter the same semaphore in
32 * one semop() are handled.
33 * - sem_ctime (time of last semctl()) is updated in the IPC_SET, SETVAL and
34 * SETALL calls.
35 * - two Linux specific semctl() commands: SEM_STAT, SEM_INFO.
36 * - undo adjustments at process exit are limited to 0..SEMVMX.
37 * - namespace are supported.
38 * - SEMMSL, SEMMNS, SEMOPM and SEMMNI can be configured at runtine by writing
39 * to /proc/sys/kernel/sem.
40 * - statistics about the usage are reported in /proc/sysvipc/sem.
41 *
42 * Internals:
43 * - scalability:
44 * - all global variables are read-mostly.
45 * - semop() calls and semctl(RMID) are synchronized by RCU.
46 * - most operations do write operations (actually: spin_lock calls) to
47 * the per-semaphore array structure.
48 * Thus: Perfect SMP scaling between independent semaphore arrays.
49 * If multiple semaphores in one array are used, then cache line
50 * trashing on the semaphore array spinlock will limit the scaling.
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -070051 * - semncnt and semzcnt are calculated on demand in count_semcnt()
Manfred Spraulc5cf6352010-05-26 14:43:43 -070052 * - the task that performs a successful semop() scans the list of all
53 * sleeping tasks and completes any pending operations that can be fulfilled.
54 * Semaphores are actively given to waiting tasks (necessary for FIFO).
55 * (see update_queue())
56 * - To improve the scalability, the actual wake-up calls are performed after
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -080057 * dropping all locks. (see wake_up_sem_queue_prepare())
Manfred Spraulc5cf6352010-05-26 14:43:43 -070058 * - 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).
Manfred Spraulc5cf6352010-05-26 14:43:43 -070062 * - UNDO values are stored in an array (one per process and per
63 * semaphore array, lazily allocated). For backwards compatibility, multiple
64 * modes for the UNDO variables are supported (per process, per thread)
65 * (see copy_semundo, CLONE_SYSVSEM)
66 * - There are two lists of the pending operations: a per-array list
67 * and per-semaphore list (stored in the array). This allows to achieve FIFO
68 * ordering without always scanning all pending operations.
69 * The worst-case behavior is nevertheless O(N^2) for N wakeups.
Linus Torvalds1da177e2005-04-16 15:20:36 -070070 */
71
Linus Torvalds1da177e2005-04-16 15:20:36 -070072#include <linux/slab.h>
73#include <linux/spinlock.h>
74#include <linux/init.h>
75#include <linux/proc_fs.h>
76#include <linux/time.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070077#include <linux/security.h>
78#include <linux/syscalls.h>
79#include <linux/audit.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080080#include <linux/capability.h>
Mike Waychison19b49462005-09-06 15:17:10 -070081#include <linux/seq_file.h>
Nadia Derbey3e148c72007-10-18 23:40:54 -070082#include <linux/rwsem.h>
Kirill Korotaeve3893532006-10-02 02:18:22 -070083#include <linux/nsproxy.h>
Pavel Emelyanovae5e1b22008-02-08 04:18:22 -080084#include <linux/ipc_namespace.h>
Ingo Molnar5f921ae2006-03-26 01:37:17 -080085
Paul McQuade7153e402014-06-06 14:37:37 -070086#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070087#include "util.h"
88
Manfred Spraule57940d2011-11-02 13:38:54 -070089/* One semaphore structure for each semaphore in the system. */
90struct sem {
91 int semval; /* current value */
Davidlohr Buesoa5f4db82016-03-22 14:27:48 -070092 /*
93 * PID of the process that last modified the semaphore. For
94 * Linux, specifically these are:
95 * - semop
96 * - semctl, via SETVAL and SETALL.
97 * - at task exit when performing undo adjustments (see exit_sem).
98 */
99 int sempid;
Rik van Riel6062a8d2013-04-30 19:15:44 -0700100 spinlock_t lock; /* spinlock for fine-grained semtimedop */
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700101 struct list_head pending_alter; /* pending single-sop operations */
102 /* that alter the semaphore */
103 struct list_head pending_const; /* pending single-sop operations */
104 /* that do not alter the semaphore*/
Manfred Sprauld12e1e52013-07-08 16:01:25 -0700105 time_t sem_otime; /* candidate for sem_otime */
Manfred Spraulf5c936c2013-07-08 16:01:22 -0700106} ____cacheline_aligned_in_smp;
Manfred Spraule57940d2011-11-02 13:38:54 -0700107
108/* One queue for each sleeping process in the system. */
109struct sem_queue {
Manfred Spraule57940d2011-11-02 13:38:54 -0700110 struct list_head list; /* queue of pending operations */
111 struct task_struct *sleeper; /* this process */
112 struct sem_undo *undo; /* undo structure */
113 int pid; /* process id of requesting process */
114 int status; /* completion status of operation */
115 struct sembuf *sops; /* array of pending operations */
Manfred Sprauled247b72014-06-06 14:37:49 -0700116 struct sembuf *blocking; /* the operation that blocked */
Manfred Spraule57940d2011-11-02 13:38:54 -0700117 int nsops; /* number of operations */
Davidlohr Bueso4ce33ec2016-12-14 15:06:37 -0800118 bool alter; /* does *sops alter the array? */
119 bool dupsop; /* sops on more than one sem_num */
Manfred Spraule57940d2011-11-02 13:38:54 -0700120};
121
122/* Each task has a list of undo requests. They are executed automatically
123 * when the process exits.
124 */
125struct sem_undo {
126 struct list_head list_proc; /* per-process list: *
127 * all undos from one process
128 * rcu protected */
129 struct rcu_head rcu; /* rcu struct for sem_undo */
130 struct sem_undo_list *ulp; /* back ptr to sem_undo_list */
131 struct list_head list_id; /* per semaphore array list:
132 * all undos for one array */
133 int semid; /* semaphore set identifier */
134 short *semadj; /* array of adjustments */
135 /* one per semaphore */
136};
137
138/* sem_undo_list controls shared access to the list of sem_undo structures
139 * that may be shared among all a CLONE_SYSVSEM task group.
140 */
141struct sem_undo_list {
142 atomic_t refcnt;
143 spinlock_t lock;
144 struct list_head list_proc;
145};
146
147
Pierre Peiffered2ddbf2008-02-08 04:18:57 -0800148#define sem_ids(ns) ((ns)->ids[IPC_SEM_IDS])
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149
Nadia Derbey1b531f22007-10-18 23:40:55 -0700150#define sem_checkid(sma, semid) ipc_checkid(&sma->sem_perm, semid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700152static int newary(struct ipc_namespace *, struct ipc_params *);
Pierre Peiffer01b8b072008-02-08 04:18:57 -0800153static void freeary(struct ipc_namespace *, struct kern_ipc_perm *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154#ifdef CONFIG_PROC_FS
Mike Waychison19b49462005-09-06 15:17:10 -0700155static int sysvipc_sem_proc_show(struct seq_file *s, void *it);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156#endif
157
158#define SEMMSL_FAST 256 /* 512 bytes on stack */
159#define SEMOPM_FAST 64 /* ~ 372 bytes on stack */
160
161/*
Manfred Spraul758a6ba32013-07-08 16:01:26 -0700162 * Locking:
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700163 * a) global sem_lock() for read/write
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 * sem_undo.id_next,
Manfred Spraul758a6ba32013-07-08 16:01:26 -0700165 * sem_array.complex_count,
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700166 * sem_array.complex_mode
167 * sem_array.pending{_alter,_const},
168 * sem_array.sem_undo
Paul McQuade46c0a8c2014-06-06 14:37:37 -0700169 *
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700170 * b) global or semaphore sem_lock() for read/write:
Manfred Spraul758a6ba32013-07-08 16:01:26 -0700171 * sem_array.sem_base[i].pending_{const,alter}:
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700172 * sem_array.complex_mode (for read)
173 *
174 * c) special:
175 * sem_undo_list.list_proc:
176 * * undo_list->lock for write
177 * * rcu for read
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 */
179
Kirill Korotaeve3893532006-10-02 02:18:22 -0700180#define sc_semmsl sem_ctls[0]
181#define sc_semmns sem_ctls[1]
182#define sc_semopm sem_ctls[2]
183#define sc_semmni sem_ctls[3]
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184
Pierre Peiffered2ddbf2008-02-08 04:18:57 -0800185void sem_init_ns(struct ipc_namespace *ns)
Kirill Korotaeve3893532006-10-02 02:18:22 -0700186{
Kirill Korotaeve3893532006-10-02 02:18:22 -0700187 ns->sc_semmsl = SEMMSL;
188 ns->sc_semmns = SEMMNS;
189 ns->sc_semopm = SEMOPM;
190 ns->sc_semmni = SEMMNI;
191 ns->used_sems = 0;
Pierre Peiffered2ddbf2008-02-08 04:18:57 -0800192 ipc_init_ids(&ns->ids[IPC_SEM_IDS]);
Kirill Korotaeve3893532006-10-02 02:18:22 -0700193}
194
Pavel Emelyanovae5e1b22008-02-08 04:18:22 -0800195#ifdef CONFIG_IPC_NS
Kirill Korotaeve3893532006-10-02 02:18:22 -0700196void sem_exit_ns(struct ipc_namespace *ns)
197{
Pierre Peiffer01b8b072008-02-08 04:18:57 -0800198 free_ipcs(ns, &sem_ids(ns), freeary);
Serge E. Hallyn7d6feeb2009-12-15 16:47:27 -0800199 idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr);
Kirill Korotaeve3893532006-10-02 02:18:22 -0700200}
Pavel Emelyanovae5e1b22008-02-08 04:18:22 -0800201#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202
Manfred Spraul239521f2014-01-27 17:07:04 -0800203void __init sem_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204{
Pierre Peiffered2ddbf2008-02-08 04:18:57 -0800205 sem_init_ns(&init_ipc_ns);
Mike Waychison19b49462005-09-06 15:17:10 -0700206 ipc_init_proc_interface("sysvipc/sem",
207 " key semid perms nsems uid gid cuid cgid otime ctime\n",
Kirill Korotaeve3893532006-10-02 02:18:22 -0700208 IPC_SEM_IDS, sysvipc_sem_proc_show);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209}
210
Manfred Spraulf269f402013-07-08 16:01:24 -0700211/**
212 * unmerge_queues - unmerge queues, if possible.
213 * @sma: semaphore array
214 *
215 * The function unmerges the wait queues if complex_count is 0.
216 * It must be called prior to dropping the global semaphore array lock.
217 */
218static void unmerge_queues(struct sem_array *sma)
219{
220 struct sem_queue *q, *tq;
221
222 /* complex operations still around? */
223 if (sma->complex_count)
224 return;
225 /*
226 * We will switch back to simple mode.
227 * Move all pending operation back into the per-semaphore
228 * queues.
229 */
230 list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
231 struct sem *curr;
232 curr = &sma->sem_base[q->sops[0].sem_num];
233
234 list_add_tail(&q->list, &curr->pending_alter);
235 }
236 INIT_LIST_HEAD(&sma->pending_alter);
237}
238
239/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800240 * merge_queues - merge single semop queues into global queue
Manfred Spraulf269f402013-07-08 16:01:24 -0700241 * @sma: semaphore array
242 *
243 * This function merges all per-semaphore queues into the global queue.
244 * It is necessary to achieve FIFO ordering for the pending single-sop
245 * operations when a multi-semop operation must sleep.
246 * Only the alter operations must be moved, the const operations can stay.
247 */
248static void merge_queues(struct sem_array *sma)
249{
250 int i;
251 for (i = 0; i < sma->sem_nsems; i++) {
252 struct sem *sem = sma->sem_base + i;
253
254 list_splice_init(&sem->pending_alter, &sma->pending_alter);
255 }
256}
257
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -0700258static void sem_rcu_free(struct rcu_head *head)
259{
260 struct ipc_rcu *p = container_of(head, struct ipc_rcu, rcu);
261 struct sem_array *sma = ipc_rcu_to_struct(p);
262
263 security_sem_free(sma);
264 ipc_rcu_free(head);
265}
266
Nadia Derbey3e148c72007-10-18 23:40:54 -0700267/*
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700268 * Enter the mode suitable for non-simple operations:
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700269 * Caller must own sem_perm.lock.
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700270 */
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700271static void complexmode_enter(struct sem_array *sma)
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700272{
273 int i;
274 struct sem *sem;
275
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700276 if (sma->complex_mode) {
277 /* We are already in complex_mode. Nothing to do */
Manfred Spraul6d07b682013-09-30 13:45:06 -0700278 return;
279 }
280
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700281 /* We need a full barrier after seting complex_mode:
282 * The write to complex_mode must be visible
283 * before we read the first sem->lock spinlock state.
284 */
285 smp_store_mb(sma->complex_mode, true);
286
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700287 for (i = 0; i < sma->sem_nsems; i++) {
288 sem = sma->sem_base + i;
289 spin_unlock_wait(&sem->lock);
290 }
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700291 /*
292 * spin_unlock_wait() is not a memory barriers, it is only a
293 * control barrier. The code must pair with spin_unlock(&sem->lock),
294 * thus just the control barrier is insufficient.
295 *
296 * smp_rmb() is sufficient, as writes cannot pass the control barrier.
297 */
298 smp_rmb();
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700299}
300
301/*
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700302 * Try to leave the mode that disallows simple operations:
303 * Caller must own sem_perm.lock.
304 */
305static void complexmode_tryleave(struct sem_array *sma)
306{
307 if (sma->complex_count) {
308 /* Complex ops are sleeping.
309 * We must stay in complex mode
310 */
311 return;
312 }
313 /*
314 * Immediately after setting complex_mode to false,
315 * a simple op can start. Thus: all memory writes
316 * performed by the current operation must be visible
317 * before we set complex_mode to false.
318 */
319 smp_store_release(&sma->complex_mode, false);
320}
321
322#define SEM_GLOBAL_LOCK (-1)
323/*
Rik van Riel6062a8d2013-04-30 19:15:44 -0700324 * If the request contains only one semaphore operation, and there are
325 * no complex transactions pending, lock only the semaphore involved.
326 * Otherwise, lock the entire semaphore array, since we either have
327 * multiple semaphores in our own semops, or we need to look at
328 * semaphores from other pending complex operations.
Rik van Riel6062a8d2013-04-30 19:15:44 -0700329 */
330static inline int sem_lock(struct sem_array *sma, struct sembuf *sops,
331 int nsops)
332{
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700333 struct sem *sem;
Rik van Riel6062a8d2013-04-30 19:15:44 -0700334
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700335 if (nsops != 1) {
336 /* Complex operation - acquire a full lock */
337 ipc_lock_object(&sma->sem_perm);
338
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700339 /* Prevent parallel simple ops */
340 complexmode_enter(sma);
341 return SEM_GLOBAL_LOCK;
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700342 }
343
344 /*
345 * Only one semaphore affected - try to optimize locking.
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700346 * Optimized locking is possible if no complex operation
347 * is either enqueued or processed right now.
348 *
349 * Both facts are tracked by complex_mode.
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700350 */
351 sem = sma->sem_base + sops->sem_num;
352
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700353 /*
354 * Initial check for complex_mode. Just an optimization,
355 * no locking, no memory barrier.
356 */
357 if (!sma->complex_mode) {
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700358 /*
359 * It appears that no complex operation is around.
360 * Acquire the per-semaphore lock.
361 */
Rik van Riel6062a8d2013-04-30 19:15:44 -0700362 spin_lock(&sem->lock);
363
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700364 /*
365 * See 51d7d5205d33
366 * ("powerpc: Add smp_mb() to arch_spin_is_locked()"):
367 * A full barrier is required: the write of sem->lock
368 * must be visible before the read is executed
369 */
370 smp_mb();
Rik van Riel6062a8d2013-04-30 19:15:44 -0700371
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700372 if (!smp_load_acquire(&sma->complex_mode)) {
373 /* fast path successful! */
374 return sops->sem_num;
Rik van Riel6062a8d2013-04-30 19:15:44 -0700375 }
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700376 spin_unlock(&sem->lock);
Rik van Riel6062a8d2013-04-30 19:15:44 -0700377 }
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700378
379 /* slow path: acquire the full lock */
380 ipc_lock_object(&sma->sem_perm);
381
382 if (sma->complex_count == 0) {
383 /* False alarm:
384 * There is no complex operation, thus we can switch
385 * back to the fast path.
386 */
387 spin_lock(&sem->lock);
388 ipc_unlock_object(&sma->sem_perm);
389 return sops->sem_num;
390 } else {
391 /* Not a false alarm, thus complete the sequence for a
392 * full lock.
393 */
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700394 complexmode_enter(sma);
395 return SEM_GLOBAL_LOCK;
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700396 }
Rik van Riel6062a8d2013-04-30 19:15:44 -0700397}
398
399static inline void sem_unlock(struct sem_array *sma, int locknum)
400{
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700401 if (locknum == SEM_GLOBAL_LOCK) {
Manfred Spraulf269f402013-07-08 16:01:24 -0700402 unmerge_queues(sma);
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700403 complexmode_tryleave(sma);
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -0700404 ipc_unlock_object(&sma->sem_perm);
Rik van Riel6062a8d2013-04-30 19:15:44 -0700405 } else {
406 struct sem *sem = sma->sem_base + locknum;
407 spin_unlock(&sem->lock);
408 }
Rik van Riel6062a8d2013-04-30 19:15:44 -0700409}
410
411/*
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700412 * sem_lock_(check_) routines are called in the paths where the rwsem
Nadia Derbey3e148c72007-10-18 23:40:54 -0700413 * is not held.
Linus Torvalds321310c2013-05-04 10:47:57 -0700414 *
415 * The caller holds the RCU read lock.
Nadia Derbey3e148c72007-10-18 23:40:54 -0700416 */
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);
Fabian Frederick9b24fef2016-08-02 14:03:07 -0700441 ipc_rcu_putref(sma, sem_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
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700449/**
450 * newary - Create a new semaphore set
451 * @ns: namespace
452 * @params: ptr to the structure that contains key, semflg and nsems
453 *
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700454 * Called with sem_ids.rwsem held (as a writer)
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700455 */
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700456static int newary(struct ipc_namespace *ns, struct ipc_params *params)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457{
458 int id;
459 int retval;
460 struct sem_array *sma;
461 int size;
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700462 key_t key = params->key;
463 int nsems = params->u.nsems;
464 int semflg = params->flg;
Manfred Spraulb97e8202009-12-15 16:47:32 -0800465 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466
467 if (!nsems)
468 return -EINVAL;
Kirill Korotaeve3893532006-10-02 02:18:22 -0700469 if (ns->used_sems + nsems > ns->sc_semmns)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 return -ENOSPC;
471
Manfred Spraul239521f2014-01-27 17:07:04 -0800472 size = sizeof(*sma) + nsems * sizeof(struct sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 sma = ipc_rcu_alloc(size);
Davidlohr Bueso3ab08fe2014-01-27 17:07:06 -0800474 if (!sma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 return -ENOMEM;
Davidlohr Bueso3ab08fe2014-01-27 17:07:06 -0800476
Manfred Spraul239521f2014-01-27 17:07:04 -0800477 memset(sma, 0, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478
479 sma->sem_perm.mode = (semflg & S_IRWXUGO);
480 sma->sem_perm.key = key;
481
482 sma->sem_perm.security = NULL;
483 retval = security_sem_alloc(sma);
484 if (retval) {
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -0700485 ipc_rcu_putref(sma, ipc_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 return retval;
487 }
488
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 sma->sem_base = (struct sem *) &sma[1];
Manfred Spraulb97e8202009-12-15 16:47:32 -0800490
Rik van Riel6062a8d2013-04-30 19:15:44 -0700491 for (i = 0; i < nsems; i++) {
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700492 INIT_LIST_HEAD(&sma->sem_base[i].pending_alter);
493 INIT_LIST_HEAD(&sma->sem_base[i].pending_const);
Rik van Riel6062a8d2013-04-30 19:15:44 -0700494 spin_lock_init(&sma->sem_base[i].lock);
495 }
Manfred Spraulb97e8202009-12-15 16:47:32 -0800496
497 sma->complex_count = 0;
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700498 sma->complex_mode = true; /* dropped by sem_unlock below */
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700499 INIT_LIST_HEAD(&sma->pending_alter);
500 INIT_LIST_HEAD(&sma->pending_const);
Manfred Spraul4daa28f2008-07-25 01:48:04 -0700501 INIT_LIST_HEAD(&sma->list_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 sma->sem_nsems = nsems;
503 sma->sem_ctime = get_seconds();
Manfred Spraule8577d12014-12-02 15:59:34 -0800504
505 id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
506 if (id < 0) {
507 ipc_rcu_putref(sma, sem_rcu_free);
508 return id;
509 }
510 ns->used_sems += nsems;
511
Rik van Riel6062a8d2013-04-30 19:15:44 -0700512 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -0700513 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700515 return sma->sem_perm.id;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516}
517
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700518
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700519/*
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700520 * Called with sem_ids.rwsem and ipcp locked.
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700521 */
Nadia Derbey03f02c72007-10-18 23:40:51 -0700522static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523{
Nadia Derbey03f02c72007-10-18 23:40:51 -0700524 struct sem_array *sma;
525
526 sma = container_of(ipcp, struct sem_array, sem_perm);
527 return security_sem_associate(sma, semflg);
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700528}
529
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700530/*
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700531 * Called with sem_ids.rwsem and ipcp locked.
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700532 */
Nadia Derbey03f02c72007-10-18 23:40:51 -0700533static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
534 struct ipc_params *params)
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700535{
Nadia Derbey03f02c72007-10-18 23:40:51 -0700536 struct sem_array *sma;
537
538 sma = container_of(ipcp, struct sem_array, sem_perm);
539 if (params->u.nsems > sma->sem_nsems)
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700540 return -EINVAL;
541
542 return 0;
543}
544
Heiko Carstensd5460c92009-01-14 14:14:27 +0100545SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg)
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700546{
Kirill Korotaeve3893532006-10-02 02:18:22 -0700547 struct ipc_namespace *ns;
Mathias Krauseeb66ec42014-06-06 14:37:36 -0700548 static const struct ipc_ops sem_ops = {
549 .getnew = newary,
550 .associate = sem_security,
551 .more_checks = sem_more_checks,
552 };
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700553 struct ipc_params sem_params;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554
Kirill Korotaeve3893532006-10-02 02:18:22 -0700555 ns = current->nsproxy->ipc_ns;
556
557 if (nsems < 0 || nsems > ns->sc_semmsl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 return -EINVAL;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700559
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700560 sem_params.key = key;
561 sem_params.flg = semflg;
562 sem_params.u.nsems = nsems;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700563
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700564 return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565}
566
Petr Mladek78f50092014-01-27 17:07:00 -0800567/**
Davidlohr Bueso4ce33ec2016-12-14 15:06:37 -0800568 * perform_atomic_semop[_slow] - Attempt to perform semaphore
569 * operations on a given array.
Manfred Spraul758a6ba32013-07-08 16:01:26 -0700570 * @sma: semaphore array
Manfred Sprauld198cd62014-06-06 14:37:49 -0700571 * @q: struct sem_queue that describes the operation
Manfred Spraul758a6ba32013-07-08 16:01:26 -0700572 *
Davidlohr Bueso4ce33ec2016-12-14 15:06:37 -0800573 * Caller blocking are as follows, based the value
574 * indicated by the semaphore operation (sem_op):
575 *
576 * (1) >0 never blocks.
577 * (2) 0 (wait-for-zero operation): semval is non-zero.
578 * (3) <0 attempting to decrement semval to a value smaller than zero.
579 *
Manfred Spraul758a6ba32013-07-08 16:01:26 -0700580 * Returns 0 if the operation was possible.
581 * Returns 1 if the operation is impossible, the caller must sleep.
Davidlohr Bueso4ce33ec2016-12-14 15:06:37 -0800582 * Returns <0 for error codes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 */
Davidlohr Bueso4ce33ec2016-12-14 15:06:37 -0800584static int perform_atomic_semop_slow(struct sem_array *sma, struct sem_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585{
Manfred Sprauld198cd62014-06-06 14:37:49 -0700586 int result, sem_op, nsops, pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 struct sembuf *sop;
Manfred Spraul239521f2014-01-27 17:07:04 -0800588 struct sem *curr;
Manfred Sprauld198cd62014-06-06 14:37:49 -0700589 struct sembuf *sops;
590 struct sem_undo *un;
591
592 sops = q->sops;
593 nsops = q->nsops;
594 un = q->undo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595
596 for (sop = sops; sop < sops + nsops; sop++) {
597 curr = sma->sem_base + sop->sem_num;
598 sem_op = sop->sem_op;
599 result = curr->semval;
Petr Mladek78f50092014-01-27 17:07:00 -0800600
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 if (!sem_op && result)
602 goto would_block;
603
604 result += sem_op;
605 if (result < 0)
606 goto would_block;
607 if (result > SEMVMX)
608 goto out_of_range;
Petr Mladek78f50092014-01-27 17:07:00 -0800609
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 if (sop->sem_flg & SEM_UNDO) {
611 int undo = un->semadj[sop->sem_num] - sem_op;
Petr Mladek78f50092014-01-27 17:07:00 -0800612 /* Exceeding the undo range is an error. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
614 goto out_of_range;
Petr Mladek78f50092014-01-27 17:07:00 -0800615 un->semadj[sop->sem_num] = undo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 }
Petr Mladek78f50092014-01-27 17:07:00 -0800617
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 curr->semval = result;
619 }
620
621 sop--;
Manfred Sprauld198cd62014-06-06 14:37:49 -0700622 pid = q->pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 while (sop >= sops) {
624 sma->sem_base[sop->sem_num].sempid = pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 sop--;
626 }
Petr Mladek78f50092014-01-27 17:07:00 -0800627
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 return 0;
629
630out_of_range:
631 result = -ERANGE;
632 goto undo;
633
634would_block:
Manfred Sprauled247b72014-06-06 14:37:49 -0700635 q->blocking = sop;
636
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 if (sop->sem_flg & IPC_NOWAIT)
638 result = -EAGAIN;
639 else
640 result = 1;
641
642undo:
643 sop--;
644 while (sop >= sops) {
Petr Mladek78f50092014-01-27 17:07:00 -0800645 sem_op = sop->sem_op;
646 sma->sem_base[sop->sem_num].semval -= sem_op;
647 if (sop->sem_flg & SEM_UNDO)
648 un->semadj[sop->sem_num] += sem_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 sop--;
650 }
651
652 return result;
653}
654
Davidlohr Bueso4ce33ec2016-12-14 15:06:37 -0800655static int perform_atomic_semop(struct sem_array *sma, struct sem_queue *q)
656{
657 int result, sem_op, nsops;
658 struct sembuf *sop;
659 struct sem *curr;
660 struct sembuf *sops;
661 struct sem_undo *un;
662
663 sops = q->sops;
664 nsops = q->nsops;
665 un = q->undo;
666
667 if (unlikely(q->dupsop))
668 return perform_atomic_semop_slow(sma, q);
669
670 /*
671 * We scan the semaphore set twice, first to ensure that the entire
672 * operation can succeed, therefore avoiding any pointless writes
673 * to shared memory and having to undo such changes in order to block
674 * until the operations can go through.
675 */
676 for (sop = sops; sop < sops + nsops; sop++) {
677 curr = sma->sem_base + sop->sem_num;
678 sem_op = sop->sem_op;
679 result = curr->semval;
680
681 if (!sem_op && result)
682 goto would_block; /* wait-for-zero */
683
684 result += sem_op;
685 if (result < 0)
686 goto would_block;
687
688 if (result > SEMVMX)
689 return -ERANGE;
690
691 if (sop->sem_flg & SEM_UNDO) {
692 int undo = un->semadj[sop->sem_num] - sem_op;
693
694 /* Exceeding the undo range is an error. */
695 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
696 return -ERANGE;
697 }
698 }
699
700 for (sop = sops; sop < sops + nsops; sop++) {
701 curr = sma->sem_base + sop->sem_num;
702 sem_op = sop->sem_op;
703 result = curr->semval;
704
705 if (sop->sem_flg & SEM_UNDO) {
706 int undo = un->semadj[sop->sem_num] - sem_op;
707
708 un->semadj[sop->sem_num] = undo;
709 }
710 curr->semval += sem_op;
711 curr->sempid = q->pid;
712 }
713
714 return 0;
715
716would_block:
717 q->blocking = sop;
718 return sop->sem_flg & IPC_NOWAIT ? -EAGAIN : 1;
719}
720
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -0800721static inline void wake_up_sem_queue_prepare(struct sem_queue *q, int error,
722 struct wake_q_head *wake_q)
Nick Piggind4212092009-12-15 16:47:30 -0800723{
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -0800724 wake_q_add(wake_q, q->sleeper);
725 /*
726 * Rely on the above implicit barrier, such that we can
727 * ensure that we hold reference to the task before setting
728 * q->status. Otherwise we could race with do_exit if the
729 * task is awoken by an external event before calling
730 * wake_up_process().
731 */
732 WRITE_ONCE(q->status, error);
Nick Piggind4212092009-12-15 16:47:30 -0800733}
734
Manfred Spraulb97e8202009-12-15 16:47:32 -0800735static void unlink_queue(struct sem_array *sma, struct sem_queue *q)
736{
737 list_del(&q->list);
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700738 if (q->nsops > 1)
Manfred Spraulb97e8202009-12-15 16:47:32 -0800739 sma->complex_count--;
740}
741
Manfred Spraulfd5db422010-05-26 14:43:40 -0700742/** check_restart(sma, q)
743 * @sma: semaphore array
744 * @q: the operation that just completed
745 *
746 * update_queue is O(N^2) when it restarts scanning the whole queue of
747 * waiting operations. Therefore this function checks if the restart is
748 * really necessary. It is called after a previously waiting operation
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700749 * modified the array.
750 * Note that wait-for-zero operations are handled without restart.
Manfred Spraulfd5db422010-05-26 14:43:40 -0700751 */
Davidlohr Bueso4663d3e2016-12-14 15:06:40 -0800752static inline int check_restart(struct sem_array *sma, struct sem_queue *q)
Manfred Spraulfd5db422010-05-26 14:43:40 -0700753{
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700754 /* pending complex alter operations are too difficult to analyse */
755 if (!list_empty(&sma->pending_alter))
Manfred Spraulfd5db422010-05-26 14:43:40 -0700756 return 1;
757
758 /* we were a sleeping complex operation. Too difficult */
759 if (q->nsops > 1)
760 return 1;
761
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700762 /* It is impossible that someone waits for the new value:
763 * - complex operations always restart.
764 * - wait-for-zero are handled seperately.
765 * - q is a previously sleeping simple operation that
766 * altered the array. It must be a decrement, because
767 * simple increments never sleep.
768 * - If there are older (higher priority) decrements
769 * in the queue, then they have observed the original
770 * semval value and couldn't proceed. The operation
771 * decremented to value - thus they won't proceed either.
772 */
773 return 0;
774}
Manfred Spraulfd5db422010-05-26 14:43:40 -0700775
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700776/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800777 * wake_const_ops - wake up non-alter tasks
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700778 * @sma: semaphore array.
779 * @semnum: semaphore that was modified.
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -0800780 * @wake_q: lockless wake-queue head.
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700781 *
782 * wake_const_ops must be called after a semaphore in a semaphore array
783 * was set to 0. If complex const operations are pending, wake_const_ops must
784 * be called with semnum = -1, as well as with the number of each modified
785 * semaphore.
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -0800786 * The tasks that must be woken up are added to @wake_q. The return code
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700787 * is stored in q->pid.
788 * The function returns 1 if at least one operation was completed successfully.
789 */
790static int wake_const_ops(struct sem_array *sma, int semnum,
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -0800791 struct wake_q_head *wake_q)
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700792{
Davidlohr Buesof150f022016-12-14 15:06:43 -0800793 struct sem_queue *q, *tmp;
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700794 struct list_head *pending_list;
795 int semop_completed = 0;
Manfred Spraulfd5db422010-05-26 14:43:40 -0700796
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700797 if (semnum == -1)
798 pending_list = &sma->pending_const;
799 else
800 pending_list = &sma->sem_base[semnum].pending_const;
801
Davidlohr Buesof150f022016-12-14 15:06:43 -0800802 list_for_each_entry_safe(q, tmp, pending_list, list) {
803 int error = perform_atomic_semop(sma, q);
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700804
Davidlohr Buesof150f022016-12-14 15:06:43 -0800805 if (error > 0)
806 continue;
807 /* operation completed, remove from queue & wakeup */
808 unlink_queue(sma, q);
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700809
Davidlohr Buesof150f022016-12-14 15:06:43 -0800810 wake_up_sem_queue_prepare(q, error, wake_q);
811 if (error == 0)
812 semop_completed = 1;
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700813 }
Davidlohr Buesof150f022016-12-14 15:06:43 -0800814
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700815 return semop_completed;
816}
817
818/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800819 * do_smart_wakeup_zero - wakeup all wait for zero tasks
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700820 * @sma: semaphore array
821 * @sops: operations that were performed
822 * @nsops: number of operations
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -0800823 * @wake_q: lockless wake-queue head
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700824 *
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800825 * Checks all required queue for wait-for-zero operations, based
826 * on the actual changes that were performed on the semaphore array.
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700827 * The function returns 1 if at least one operation was completed successfully.
828 */
829static int do_smart_wakeup_zero(struct sem_array *sma, struct sembuf *sops,
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -0800830 int nsops, struct wake_q_head *wake_q)
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700831{
832 int i;
833 int semop_completed = 0;
834 int got_zero = 0;
835
836 /* first: the per-semaphore queues, if known */
837 if (sops) {
838 for (i = 0; i < nsops; i++) {
839 int num = sops[i].sem_num;
840
841 if (sma->sem_base[num].semval == 0) {
842 got_zero = 1;
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -0800843 semop_completed |= wake_const_ops(sma, num, wake_q);
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700844 }
845 }
846 } else {
847 /*
848 * No sops means modified semaphores not known.
849 * Assume all were changed.
Manfred Spraulfd5db422010-05-26 14:43:40 -0700850 */
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700851 for (i = 0; i < sma->sem_nsems; i++) {
852 if (sma->sem_base[i].semval == 0) {
853 got_zero = 1;
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -0800854 semop_completed |= wake_const_ops(sma, i, wake_q);
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700855 }
856 }
Manfred Spraulfd5db422010-05-26 14:43:40 -0700857 }
858 /*
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700859 * If one of the modified semaphores got 0,
860 * then check the global queue, too.
Manfred Spraulfd5db422010-05-26 14:43:40 -0700861 */
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700862 if (got_zero)
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -0800863 semop_completed |= wake_const_ops(sma, -1, wake_q);
Manfred Spraulfd5db422010-05-26 14:43:40 -0700864
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700865 return semop_completed;
Manfred Spraulfd5db422010-05-26 14:43:40 -0700866}
867
Manfred Spraul636c6be2009-12-15 16:47:33 -0800868
869/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800870 * update_queue - look for tasks that can be completed.
Manfred Spraul636c6be2009-12-15 16:47:33 -0800871 * @sma: semaphore array.
872 * @semnum: semaphore that was modified.
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -0800873 * @wake_q: lockless wake-queue head.
Manfred Spraul636c6be2009-12-15 16:47:33 -0800874 *
875 * update_queue must be called after a semaphore in a semaphore array
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700876 * was modified. If multiple semaphores were modified, update_queue must
877 * be called with semnum = -1, as well as with the number of each modified
878 * semaphore.
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -0800879 * The tasks that must be woken up are added to @wake_q. The return code
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700880 * is stored in q->pid.
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700881 * The function internally checks if const operations can now succeed.
882 *
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700883 * The function return 1 if at least one semop was completed successfully.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 */
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -0800885static int update_queue(struct sem_array *sma, int semnum, struct wake_q_head *wake_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886{
Davidlohr Buesof150f022016-12-14 15:06:43 -0800887 struct sem_queue *q, *tmp;
Manfred Spraul636c6be2009-12-15 16:47:33 -0800888 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:
Davidlohr Buesof150f022016-12-14 15:06:43 -0800897 list_for_each_entry_safe(q, tmp, pending_list, list) {
Manfred Spraulfd5db422010-05-26 14:43:40 -0700898 int error, restart;
Manfred Spraul636c6be2009-12-15 16:47:33 -0800899
Manfred Sprauld987f8b2009-12-15 16:47:34 -0800900 /* If we are scanning the single sop, per-semaphore list of
901 * one semaphore and that semaphore is 0, then it is not
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700902 * necessary to scan further: simple increments
Manfred Sprauld987f8b2009-12-15 16:47:34 -0800903 * that affect only one entry succeed immediately and cannot
904 * be in the per semaphore pending queue, and decrements
905 * cannot be successful if the value is already 0.
906 */
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700907 if (semnum != -1 && sma->sem_base[semnum].semval == 0)
Manfred Sprauld987f8b2009-12-15 16:47:34 -0800908 break;
909
Manfred Sprauld198cd62014-06-06 14:37:49 -0700910 error = perform_atomic_semop(sma, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911
912 /* Does q->sleeper still need to sleep? */
Nick Piggin9cad2002009-12-15 16:47:29 -0800913 if (error > 0)
914 continue;
Manfred Spraula1193f82008-07-25 01:48:06 -0700915
Manfred Spraulb97e8202009-12-15 16:47:32 -0800916 unlink_queue(sma, q);
Manfred Spraula1193f82008-07-25 01:48:06 -0700917
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700918 if (error) {
Manfred Spraulfd5db422010-05-26 14:43:40 -0700919 restart = 0;
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700920 } else {
921 semop_completed = 1;
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -0800922 do_smart_wakeup_zero(sma, q->sops, q->nsops, wake_q);
Manfred Spraulfd5db422010-05-26 14:43:40 -0700923 restart = check_restart(sma, q);
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700924 }
Manfred Spraulfd5db422010-05-26 14:43:40 -0700925
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -0800926 wake_up_sem_queue_prepare(q, error, wake_q);
Manfred Spraulfd5db422010-05-26 14:43:40 -0700927 if (restart)
Nick Piggin9cad2002009-12-15 16:47:29 -0800928 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929 }
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700930 return semop_completed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931}
932
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700933/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800934 * set_semotime - set sem_otime
Manfred Spraul0e8c6652013-09-30 13:45:25 -0700935 * @sma: semaphore array
936 * @sops: operations that modified the array, may be NULL
937 *
938 * sem_otime is replicated to avoid cache line trashing.
939 * This function sets one instance to the current time.
940 */
941static void set_semotime(struct sem_array *sma, struct sembuf *sops)
942{
943 if (sops == NULL) {
944 sma->sem_base[0].sem_otime = get_seconds();
945 } else {
946 sma->sem_base[sops[0].sem_num].sem_otime =
947 get_seconds();
948 }
949}
950
951/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800952 * do_smart_update - optimized update_queue
Manfred Spraulfd5db422010-05-26 14:43:40 -0700953 * @sma: semaphore array
954 * @sops: operations that were performed
955 * @nsops: number of operations
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700956 * @otime: force setting otime
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -0800957 * @wake_q: lockless wake-queue head
Manfred Spraulfd5db422010-05-26 14:43:40 -0700958 *
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700959 * do_smart_update() does the required calls to update_queue and wakeup_zero,
960 * based on the actual changes that were performed on the semaphore array.
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700961 * Note that the function does not do the actual wake-up: the caller is
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -0800962 * responsible for calling wake_up_q().
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700963 * It is safe to perform this call after dropping all locks.
Manfred Spraulfd5db422010-05-26 14:43:40 -0700964 */
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700965static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops,
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -0800966 int otime, struct wake_q_head *wake_q)
Manfred Spraulfd5db422010-05-26 14:43:40 -0700967{
968 int i;
969
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -0800970 otime |= do_smart_wakeup_zero(sma, sops, nsops, wake_q);
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700971
Manfred Spraulf269f402013-07-08 16:01:24 -0700972 if (!list_empty(&sma->pending_alter)) {
973 /* semaphore array uses the global queue - just process it. */
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -0800974 otime |= update_queue(sma, -1, wake_q);
Manfred Spraulf269f402013-07-08 16:01:24 -0700975 } else {
976 if (!sops) {
977 /*
978 * No sops, thus the modified semaphores are not
979 * known. Check all.
980 */
981 for (i = 0; i < sma->sem_nsems; i++)
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -0800982 otime |= update_queue(sma, i, wake_q);
Manfred Spraulf269f402013-07-08 16:01:24 -0700983 } else {
984 /*
985 * Check the semaphores that were increased:
986 * - No complex ops, thus all sleeping ops are
987 * decrease.
988 * - if we decreased the value, then any sleeping
989 * semaphore ops wont be able to run: If the
990 * previous value was too small, then the new
991 * value will be too small, too.
992 */
993 for (i = 0; i < nsops; i++) {
994 if (sops[i].sem_op > 0) {
995 otime |= update_queue(sma,
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -0800996 sops[i].sem_num, wake_q);
Manfred Spraulf269f402013-07-08 16:01:24 -0700997 }
Manfred Spraulab465df2013-05-26 11:08:52 +0200998 }
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700999 }
Manfred Spraulfd5db422010-05-26 14:43:40 -07001000 }
Manfred Spraul0e8c6652013-09-30 13:45:25 -07001001 if (otime)
1002 set_semotime(sma, sops);
Manfred Spraulfd5db422010-05-26 14:43:40 -07001003}
1004
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001005/*
Manfred Spraulb220c572014-06-06 14:37:51 -07001006 * check_qop: Test if a queued operation sleeps on the semaphore semnum
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001007 */
1008static int check_qop(struct sem_array *sma, int semnum, struct sem_queue *q,
1009 bool count_zero)
1010{
Manfred Spraulb220c572014-06-06 14:37:51 -07001011 struct sembuf *sop = q->blocking;
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001012
Manfred Spraul9b44ee22014-06-06 14:37:52 -07001013 /*
1014 * Linux always (since 0.99.10) reported a task as sleeping on all
1015 * semaphores. This violates SUS, therefore it was changed to the
1016 * standard compliant behavior.
1017 * Give the administrators a chance to notice that an application
1018 * might misbehave because it relies on the Linux behavior.
1019 */
1020 pr_info_once("semctl(GETNCNT/GETZCNT) is since 3.16 Single Unix Specification compliant.\n"
1021 "The task %s (%d) triggered the difference, watch for misbehavior.\n",
1022 current->comm, task_pid_nr(current));
1023
Manfred Spraulb220c572014-06-06 14:37:51 -07001024 if (sop->sem_num != semnum)
1025 return 0;
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001026
Manfred Spraulb220c572014-06-06 14:37:51 -07001027 if (count_zero && sop->sem_op == 0)
1028 return 1;
1029 if (!count_zero && sop->sem_op < 0)
1030 return 1;
1031
1032 return 0;
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001033}
1034
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035/* The following counts are associated to each semaphore:
1036 * semncnt number of tasks waiting on semval being nonzero
1037 * semzcnt number of tasks waiting on semval being zero
Manfred Spraulb220c572014-06-06 14:37:51 -07001038 *
1039 * Per definition, a task waits only on the semaphore of the first semop
1040 * that cannot proceed, even if additional operation would block, too.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 */
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001042static int count_semcnt(struct sem_array *sma, ushort semnum,
1043 bool count_zero)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044{
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001045 struct list_head *l;
Manfred Spraul239521f2014-01-27 17:07:04 -08001046 struct sem_queue *q;
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001047 int semcnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001049 semcnt = 0;
1050 /* First: check the simple operations. They are easy to evaluate */
1051 if (count_zero)
1052 l = &sma->sem_base[semnum].pending_const;
1053 else
1054 l = &sma->sem_base[semnum].pending_alter;
1055
1056 list_for_each_entry(q, l, list) {
1057 /* all task on a per-semaphore list sleep on exactly
1058 * that semaphore
1059 */
1060 semcnt++;
Rik van Rielde2657f2013-05-09 16:59:59 -04001061 }
1062
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001063 /* Then: check the complex operations. */
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001064 list_for_each_entry(q, &sma->pending_alter, list) {
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001065 semcnt += check_qop(sma, semnum, q, count_zero);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 }
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001067 if (count_zero) {
1068 list_for_each_entry(q, &sma->pending_const, list) {
1069 semcnt += check_qop(sma, semnum, q, count_zero);
1070 }
Rik van Rielebc2e5e2013-05-09 16:53:28 -04001071 }
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001072 return semcnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073}
1074
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001075/* Free a semaphore set. freeary() is called with sem_ids.rwsem locked
1076 * as a writer and the spinlock for this semaphore set hold. sem_ids.rwsem
Nadia Derbey3e148c72007-10-18 23:40:54 -07001077 * remains locked on exit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 */
Pierre Peiffer01b8b072008-02-08 04:18:57 -08001079static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080{
Manfred Spraul380af1b2008-07-25 01:48:06 -07001081 struct sem_undo *un, *tu;
1082 struct sem_queue *q, *tq;
Pierre Peiffer01b8b072008-02-08 04:18:57 -08001083 struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001084 int i;
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08001085 DEFINE_WAKE_Q(wake_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086
Manfred Spraul380af1b2008-07-25 01:48:06 -07001087 /* Free the existing undo structures for this semaphore set. */
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -07001088 ipc_assert_locked_object(&sma->sem_perm);
Manfred Spraul380af1b2008-07-25 01:48:06 -07001089 list_for_each_entry_safe(un, tu, &sma->list_id, list_id) {
1090 list_del(&un->list_id);
1091 spin_lock(&un->ulp->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 un->semid = -1;
Manfred Spraul380af1b2008-07-25 01:48:06 -07001093 list_del_rcu(&un->list_proc);
1094 spin_unlock(&un->ulp->lock);
Lai Jiangshan693a8b62011-03-18 12:09:35 +08001095 kfree_rcu(un, rcu);
Manfred Spraul380af1b2008-07-25 01:48:06 -07001096 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097
1098 /* Wake up all pending processes and let them fail with EIDRM. */
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001099 list_for_each_entry_safe(q, tq, &sma->pending_const, list) {
1100 unlink_queue(sma, q);
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08001101 wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001102 }
1103
1104 list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
Manfred Spraulb97e8202009-12-15 16:47:32 -08001105 unlink_queue(sma, q);
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08001106 wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107 }
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001108 for (i = 0; i < sma->sem_nsems; i++) {
1109 struct sem *sem = sma->sem_base + i;
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001110 list_for_each_entry_safe(q, tq, &sem->pending_const, list) {
1111 unlink_queue(sma, q);
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08001112 wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001113 }
1114 list_for_each_entry_safe(q, tq, &sem->pending_alter, list) {
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001115 unlink_queue(sma, q);
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08001116 wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001117 }
1118 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119
Nadia Derbey7ca7e562007-10-18 23:40:48 -07001120 /* Remove the semaphore set from the IDR */
1121 sem_rmid(ns, sma);
Rik van Riel6062a8d2013-04-30 19:15:44 -07001122 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001123 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08001125 wake_up_q(&wake_q);
Kirill Korotaeve3893532006-10-02 02:18:22 -07001126 ns->used_sems -= sma->sem_nsems;
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -07001127 ipc_rcu_putref(sma, sem_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128}
1129
1130static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
1131{
Manfred Spraul239521f2014-01-27 17:07:04 -08001132 switch (version) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 case IPC_64:
1134 return copy_to_user(buf, in, sizeof(*in));
1135 case IPC_OLD:
1136 {
1137 struct semid_ds out;
1138
Dan Rosenberg982f7c22010-09-30 15:15:31 -07001139 memset(&out, 0, sizeof(out));
1140
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
1142
1143 out.sem_otime = in->sem_otime;
1144 out.sem_ctime = in->sem_ctime;
1145 out.sem_nsems = in->sem_nsems;
1146
1147 return copy_to_user(buf, &out, sizeof(out));
1148 }
1149 default:
1150 return -EINVAL;
1151 }
1152}
1153
Manfred Sprauld12e1e52013-07-08 16:01:25 -07001154static time_t get_semotime(struct sem_array *sma)
1155{
1156 int i;
1157 time_t res;
1158
1159 res = sma->sem_base[0].sem_otime;
1160 for (i = 1; i < sma->sem_nsems; i++) {
1161 time_t to = sma->sem_base[i].sem_otime;
1162
1163 if (to > res)
1164 res = to;
1165 }
1166 return res;
1167}
1168
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -08001169static int semctl_nolock(struct ipc_namespace *ns, int semid,
Al Viroe1fd1f42013-03-05 15:04:55 -05001170 int cmd, int version, void __user *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171{
Amerigo Wange5cc9c72009-12-15 16:47:35 -08001172 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 struct sem_array *sma;
1174
Manfred Spraul239521f2014-01-27 17:07:04 -08001175 switch (cmd) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176 case IPC_INFO:
1177 case SEM_INFO:
1178 {
1179 struct seminfo seminfo;
1180 int max_id;
1181
1182 err = security_sem_semctl(NULL, cmd);
1183 if (err)
1184 return err;
Paul McQuade46c0a8c2014-06-06 14:37:37 -07001185
Manfred Spraul239521f2014-01-27 17:07:04 -08001186 memset(&seminfo, 0, sizeof(seminfo));
Kirill Korotaeve3893532006-10-02 02:18:22 -07001187 seminfo.semmni = ns->sc_semmni;
1188 seminfo.semmns = ns->sc_semmns;
1189 seminfo.semmsl = ns->sc_semmsl;
1190 seminfo.semopm = ns->sc_semopm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191 seminfo.semvmx = SEMVMX;
1192 seminfo.semmnu = SEMMNU;
1193 seminfo.semmap = SEMMAP;
1194 seminfo.semume = SEMUME;
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001195 down_read(&sem_ids(ns).rwsem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196 if (cmd == SEM_INFO) {
Kirill Korotaeve3893532006-10-02 02:18:22 -07001197 seminfo.semusz = sem_ids(ns).in_use;
1198 seminfo.semaem = ns->used_sems;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 } else {
1200 seminfo.semusz = SEMUSZ;
1201 seminfo.semaem = SEMAEM;
1202 }
Nadia Derbey7ca7e562007-10-18 23:40:48 -07001203 max_id = ipc_get_maxid(&sem_ids(ns));
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001204 up_read(&sem_ids(ns).rwsem);
Paul McQuade46c0a8c2014-06-06 14:37:37 -07001205 if (copy_to_user(p, &seminfo, sizeof(struct seminfo)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 return -EFAULT;
Manfred Spraul239521f2014-01-27 17:07:04 -08001207 return (max_id < 0) ? 0 : max_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208 }
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -08001209 case IPC_STAT:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 case SEM_STAT:
1211 {
1212 struct semid64_ds tbuf;
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001213 int id = 0;
1214
1215 memset(&tbuf, 0, sizeof(tbuf));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216
Linus Torvalds941b0302013-05-04 11:04:29 -07001217 rcu_read_lock();
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -08001218 if (cmd == SEM_STAT) {
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001219 sma = sem_obtain_object(ns, semid);
1220 if (IS_ERR(sma)) {
1221 err = PTR_ERR(sma);
1222 goto out_unlock;
1223 }
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -08001224 id = sma->sem_perm.id;
1225 } else {
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001226 sma = sem_obtain_object_check(ns, semid);
1227 if (IS_ERR(sma)) {
1228 err = PTR_ERR(sma);
1229 goto out_unlock;
1230 }
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -08001231 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232
1233 err = -EACCES;
Serge E. Hallynb0e77592011-03-23 16:43:24 -07001234 if (ipcperms(ns, &sma->sem_perm, S_IRUGO))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 goto out_unlock;
1236
1237 err = security_sem_semctl(sma, cmd);
1238 if (err)
1239 goto out_unlock;
1240
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
Manfred Sprauld12e1e52013-07-08 16:01:25 -07001242 tbuf.sem_otime = get_semotime(sma);
1243 tbuf.sem_ctime = sma->sem_ctime;
1244 tbuf.sem_nsems = sma->sem_nsems;
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001245 rcu_read_unlock();
Al Viroe1fd1f42013-03-05 15:04:55 -05001246 if (copy_semid_to_user(p, &tbuf, version))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247 return -EFAULT;
1248 return id;
1249 }
1250 default:
1251 return -EINVAL;
1252 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253out_unlock:
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001254 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 return err;
1256}
1257
Al Viroe1fd1f42013-03-05 15:04:55 -05001258static int semctl_setval(struct ipc_namespace *ns, int semid, int semnum,
1259 unsigned long arg)
1260{
1261 struct sem_undo *un;
1262 struct sem_array *sma;
Manfred Spraul239521f2014-01-27 17:07:04 -08001263 struct sem *curr;
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08001264 int err, val;
1265 DEFINE_WAKE_Q(wake_q);
1266
Al Viroe1fd1f42013-03-05 15:04:55 -05001267#if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN)
1268 /* big-endian 64bit */
1269 val = arg >> 32;
1270#else
1271 /* 32bit or little-endian 64bit */
1272 val = arg;
1273#endif
1274
Rik van Riel6062a8d2013-04-30 19:15:44 -07001275 if (val > SEMVMX || val < 0)
1276 return -ERANGE;
Al Viroe1fd1f42013-03-05 15:04:55 -05001277
Rik van Riel6062a8d2013-04-30 19:15:44 -07001278 rcu_read_lock();
1279 sma = sem_obtain_object_check(ns, semid);
1280 if (IS_ERR(sma)) {
1281 rcu_read_unlock();
1282 return PTR_ERR(sma);
1283 }
1284
1285 if (semnum < 0 || semnum >= sma->sem_nsems) {
1286 rcu_read_unlock();
1287 return -EINVAL;
1288 }
1289
1290
1291 if (ipcperms(ns, &sma->sem_perm, S_IWUGO)) {
1292 rcu_read_unlock();
1293 return -EACCES;
1294 }
Al Viroe1fd1f42013-03-05 15:04:55 -05001295
1296 err = security_sem_semctl(sma, SETVAL);
Rik van Riel6062a8d2013-04-30 19:15:44 -07001297 if (err) {
1298 rcu_read_unlock();
1299 return -EACCES;
1300 }
Al Viroe1fd1f42013-03-05 15:04:55 -05001301
Rik van Riel6062a8d2013-04-30 19:15:44 -07001302 sem_lock(sma, NULL, -1);
Al Viroe1fd1f42013-03-05 15:04:55 -05001303
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001304 if (!ipc_valid_object(&sma->sem_perm)) {
Manfred Spraul6e224f92013-10-16 13:46:45 -07001305 sem_unlock(sma, -1);
1306 rcu_read_unlock();
1307 return -EIDRM;
1308 }
1309
Al Viroe1fd1f42013-03-05 15:04:55 -05001310 curr = &sma->sem_base[semnum];
1311
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -07001312 ipc_assert_locked_object(&sma->sem_perm);
Al Viroe1fd1f42013-03-05 15:04:55 -05001313 list_for_each_entry(un, &sma->list_id, list_id)
1314 un->semadj[semnum] = 0;
1315
1316 curr->semval = val;
1317 curr->sempid = task_tgid_vnr(current);
1318 sma->sem_ctime = get_seconds();
1319 /* maybe some queued-up processes were waiting for this */
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08001320 do_smart_update(sma, NULL, 0, 0, &wake_q);
Rik van Riel6062a8d2013-04-30 19:15:44 -07001321 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001322 rcu_read_unlock();
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08001323 wake_up_q(&wake_q);
Rik van Riel6062a8d2013-04-30 19:15:44 -07001324 return 0;
Al Viroe1fd1f42013-03-05 15:04:55 -05001325}
1326
Kirill Korotaeve3893532006-10-02 02:18:22 -07001327static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
Al Viroe1fd1f42013-03-05 15:04:55 -05001328 int cmd, void __user *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329{
1330 struct sem_array *sma;
Manfred Spraul239521f2014-01-27 17:07:04 -08001331 struct sem *curr;
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001332 int err, nsems;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333 ushort fast_sem_io[SEMMSL_FAST];
Manfred Spraul239521f2014-01-27 17:07:04 -08001334 ushort *sem_io = fast_sem_io;
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08001335 DEFINE_WAKE_Q(wake_q);
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001336
1337 rcu_read_lock();
1338 sma = sem_obtain_object_check(ns, semid);
1339 if (IS_ERR(sma)) {
1340 rcu_read_unlock();
1341 return PTR_ERR(sma);
1342 }
1343
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 nsems = sma->sem_nsems;
1345
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 err = -EACCES;
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001347 if (ipcperms(ns, &sma->sem_perm, cmd == SETALL ? S_IWUGO : S_IRUGO))
1348 goto out_rcu_wakeup;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349
1350 err = security_sem_semctl(sma, cmd);
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001351 if (err)
1352 goto out_rcu_wakeup;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353
1354 err = -EACCES;
1355 switch (cmd) {
1356 case GETALL:
1357 {
Al Viroe1fd1f42013-03-05 15:04:55 -05001358 ushort __user *array = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359 int i;
1360
Al Viroce857222013-05-03 00:30:49 +01001361 sem_lock(sma, NULL, -1);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001362 if (!ipc_valid_object(&sma->sem_perm)) {
Manfred Spraul6e224f92013-10-16 13:46:45 -07001363 err = -EIDRM;
1364 goto out_unlock;
1365 }
Manfred Spraul239521f2014-01-27 17:07:04 -08001366 if (nsems > SEMMSL_FAST) {
Al Viroce857222013-05-03 00:30:49 +01001367 if (!ipc_rcu_getref(sma)) {
Al Viroce857222013-05-03 00:30:49 +01001368 err = -EIDRM;
Manfred Spraul6e224f92013-10-16 13:46:45 -07001369 goto out_unlock;
Al Viroce857222013-05-03 00:30:49 +01001370 }
1371 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001372 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 sem_io = ipc_alloc(sizeof(ushort)*nsems);
Manfred Spraul239521f2014-01-27 17:07:04 -08001374 if (sem_io == NULL) {
Fabian Frederick9b24fef2016-08-02 14:03:07 -07001375 ipc_rcu_putref(sma, sem_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 return -ENOMEM;
1377 }
1378
Linus Torvalds4091fd942013-05-04 10:13:40 -07001379 rcu_read_lock();
Pierre Peiffer6ff37972008-04-29 01:00:46 -07001380 sem_lock_and_putref(sma);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001381 if (!ipc_valid_object(&sma->sem_perm)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382 err = -EIDRM;
Manfred Spraul6e224f92013-10-16 13:46:45 -07001383 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384 }
Al Viroce857222013-05-03 00:30:49 +01001385 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 for (i = 0; i < sma->sem_nsems; i++)
1387 sem_io[i] = sma->sem_base[i].semval;
Rik van Riel6062a8d2013-04-30 19:15:44 -07001388 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001389 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390 err = 0;
Manfred Spraul239521f2014-01-27 17:07:04 -08001391 if (copy_to_user(array, sem_io, nsems*sizeof(ushort)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 err = -EFAULT;
1393 goto out_free;
1394 }
1395 case SETALL:
1396 {
1397 int i;
1398 struct sem_undo *un;
1399
Rik van Riel6062a8d2013-04-30 19:15:44 -07001400 if (!ipc_rcu_getref(sma)) {
Manfred Spraul6e224f92013-10-16 13:46:45 -07001401 err = -EIDRM;
1402 goto out_rcu_wakeup;
Rik van Riel6062a8d2013-04-30 19:15:44 -07001403 }
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001404 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405
Manfred Spraul239521f2014-01-27 17:07:04 -08001406 if (nsems > SEMMSL_FAST) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407 sem_io = ipc_alloc(sizeof(ushort)*nsems);
Manfred Spraul239521f2014-01-27 17:07:04 -08001408 if (sem_io == NULL) {
Fabian Frederick9b24fef2016-08-02 14:03:07 -07001409 ipc_rcu_putref(sma, sem_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 return -ENOMEM;
1411 }
1412 }
1413
Manfred Spraul239521f2014-01-27 17:07:04 -08001414 if (copy_from_user(sem_io, p, nsems*sizeof(ushort))) {
Fabian Frederick9b24fef2016-08-02 14:03:07 -07001415 ipc_rcu_putref(sma, sem_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416 err = -EFAULT;
1417 goto out_free;
1418 }
1419
1420 for (i = 0; i < nsems; i++) {
1421 if (sem_io[i] > SEMVMX) {
Fabian Frederick9b24fef2016-08-02 14:03:07 -07001422 ipc_rcu_putref(sma, sem_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 err = -ERANGE;
1424 goto out_free;
1425 }
1426 }
Linus Torvalds4091fd942013-05-04 10:13:40 -07001427 rcu_read_lock();
Pierre Peiffer6ff37972008-04-29 01:00:46 -07001428 sem_lock_and_putref(sma);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001429 if (!ipc_valid_object(&sma->sem_perm)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 err = -EIDRM;
Manfred Spraul6e224f92013-10-16 13:46:45 -07001431 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432 }
1433
Davidlohr Buesoa5f4db82016-03-22 14:27:48 -07001434 for (i = 0; i < nsems; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 sma->sem_base[i].semval = sem_io[i];
Davidlohr Buesoa5f4db82016-03-22 14:27:48 -07001436 sma->sem_base[i].sempid = task_tgid_vnr(current);
1437 }
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001438
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -07001439 ipc_assert_locked_object(&sma->sem_perm);
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001440 list_for_each_entry(un, &sma->list_id, list_id) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 for (i = 0; i < nsems; i++)
1442 un->semadj[i] = 0;
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001443 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444 sma->sem_ctime = get_seconds();
1445 /* maybe some queued-up processes were waiting for this */
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08001446 do_smart_update(sma, NULL, 0, 0, &wake_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447 err = 0;
1448 goto out_unlock;
1449 }
Al Viroe1fd1f42013-03-05 15:04:55 -05001450 /* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451 }
1452 err = -EINVAL;
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001453 if (semnum < 0 || semnum >= nsems)
1454 goto out_rcu_wakeup;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455
Rik van Riel6062a8d2013-04-30 19:15:44 -07001456 sem_lock(sma, NULL, -1);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001457 if (!ipc_valid_object(&sma->sem_perm)) {
Manfred Spraul6e224f92013-10-16 13:46:45 -07001458 err = -EIDRM;
1459 goto out_unlock;
1460 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 curr = &sma->sem_base[semnum];
1462
1463 switch (cmd) {
1464 case GETVAL:
1465 err = curr->semval;
1466 goto out_unlock;
1467 case GETPID:
1468 err = curr->sempid;
1469 goto out_unlock;
1470 case GETNCNT:
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001471 err = count_semcnt(sma, semnum, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 goto out_unlock;
1473 case GETZCNT:
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001474 err = count_semcnt(sma, semnum, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 }
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001477
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478out_unlock:
Rik van Riel6062a8d2013-04-30 19:15:44 -07001479 sem_unlock(sma, -1);
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001480out_rcu_wakeup:
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001481 rcu_read_unlock();
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08001482 wake_up_q(&wake_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483out_free:
Manfred Spraul239521f2014-01-27 17:07:04 -08001484 if (sem_io != fast_sem_io)
Tetsuo Handa1d5cfdb2016-01-22 15:11:02 -08001485 ipc_free(sem_io);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 return err;
1487}
1488
Pierre Peiffer016d7132008-04-29 01:00:50 -07001489static inline unsigned long
1490copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491{
Manfred Spraul239521f2014-01-27 17:07:04 -08001492 switch (version) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 case IPC_64:
Pierre Peiffer016d7132008-04-29 01:00:50 -07001494 if (copy_from_user(out, buf, sizeof(*out)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 case IPC_OLD:
1498 {
1499 struct semid_ds tbuf_old;
1500
Manfred Spraul239521f2014-01-27 17:07:04 -08001501 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 return -EFAULT;
1503
Pierre Peiffer016d7132008-04-29 01:00:50 -07001504 out->sem_perm.uid = tbuf_old.sem_perm.uid;
1505 out->sem_perm.gid = tbuf_old.sem_perm.gid;
1506 out->sem_perm.mode = tbuf_old.sem_perm.mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507
1508 return 0;
1509 }
1510 default:
1511 return -EINVAL;
1512 }
1513}
1514
Pierre Peiffer522bb2a2008-04-29 01:00:49 -07001515/*
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001516 * This function handles some semctl commands which require the rwsem
Pierre Peiffer522bb2a2008-04-29 01:00:49 -07001517 * to be held in write mode.
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001518 * NOTE: no locks must be held, the rwsem is taken inside this function.
Pierre Peiffer522bb2a2008-04-29 01:00:49 -07001519 */
Pierre Peiffer21a48262008-04-29 01:00:49 -07001520static int semctl_down(struct ipc_namespace *ns, int semid,
Al Viroe1fd1f42013-03-05 15:04:55 -05001521 int cmd, int version, void __user *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522{
1523 struct sem_array *sma;
1524 int err;
Pierre Peiffer016d7132008-04-29 01:00:50 -07001525 struct semid64_ds semid64;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 struct kern_ipc_perm *ipcp;
1527
Manfred Spraul239521f2014-01-27 17:07:04 -08001528 if (cmd == IPC_SET) {
Al Viroe1fd1f42013-03-05 15:04:55 -05001529 if (copy_semid_from_user(&semid64, p, version))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001533 down_write(&sem_ids(ns).rwsem);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001534 rcu_read_lock();
1535
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001536 ipcp = ipcctl_pre_down_nolock(ns, &sem_ids(ns), semid, cmd,
1537 &semid64.sem_perm, 0);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001538 if (IS_ERR(ipcp)) {
1539 err = PTR_ERR(ipcp);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001540 goto out_unlock1;
1541 }
Steve Grubb073115d2006-04-02 17:07:33 -04001542
Pierre Peiffera5f75e72008-04-29 01:00:54 -07001543 sma = container_of(ipcp, struct sem_array, sem_perm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544
1545 err = security_sem_semctl(sma, cmd);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001546 if (err)
1547 goto out_unlock1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001549 switch (cmd) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550 case IPC_RMID:
Rik van Riel6062a8d2013-04-30 19:15:44 -07001551 sem_lock(sma, NULL, -1);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001552 /* freeary unlocks the ipc object and rcu */
Pierre Peiffer01b8b072008-02-08 04:18:57 -08001553 freeary(ns, ipcp);
Pierre Peiffer522bb2a2008-04-29 01:00:49 -07001554 goto out_up;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555 case IPC_SET:
Rik van Riel6062a8d2013-04-30 19:15:44 -07001556 sem_lock(sma, NULL, -1);
Eric W. Biederman1efdb692012-02-07 16:54:11 -08001557 err = ipc_update_perm(&semid64.sem_perm, ipcp);
1558 if (err)
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001559 goto out_unlock0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560 sma->sem_ctime = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 break;
1562 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563 err = -EINVAL;
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001564 goto out_unlock1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001567out_unlock0:
Rik van Riel6062a8d2013-04-30 19:15:44 -07001568 sem_unlock(sma, -1);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001569out_unlock1:
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001570 rcu_read_unlock();
Pierre Peiffer522bb2a2008-04-29 01:00:49 -07001571out_up:
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001572 up_write(&sem_ids(ns).rwsem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573 return err;
1574}
1575
Al Viroe1fd1f42013-03-05 15:04:55 -05001576SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, unsigned long, arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578 int version;
Kirill Korotaeve3893532006-10-02 02:18:22 -07001579 struct ipc_namespace *ns;
Al Viroe1fd1f42013-03-05 15:04:55 -05001580 void __user *p = (void __user *)arg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581
1582 if (semid < 0)
1583 return -EINVAL;
1584
1585 version = ipc_parse_version(&cmd);
Kirill Korotaeve3893532006-10-02 02:18:22 -07001586 ns = current->nsproxy->ipc_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587
Manfred Spraul239521f2014-01-27 17:07:04 -08001588 switch (cmd) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589 case IPC_INFO:
1590 case SEM_INFO:
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -08001591 case IPC_STAT:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592 case SEM_STAT:
Al Viroe1fd1f42013-03-05 15:04:55 -05001593 return semctl_nolock(ns, semid, cmd, version, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594 case GETALL:
1595 case GETVAL:
1596 case GETPID:
1597 case GETNCNT:
1598 case GETZCNT:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599 case SETALL:
Al Viroe1fd1f42013-03-05 15:04:55 -05001600 return semctl_main(ns, semid, semnum, cmd, p);
1601 case SETVAL:
1602 return semctl_setval(ns, semid, semnum, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603 case IPC_RMID:
1604 case IPC_SET:
Al Viroe1fd1f42013-03-05 15:04:55 -05001605 return semctl_down(ns, semid, cmd, version, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606 default:
1607 return -EINVAL;
1608 }
1609}
1610
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611/* If the task doesn't already have a undo_list, then allocate one
1612 * here. We guarantee there is only one thread using this undo list,
1613 * and current is THE ONE
1614 *
1615 * If this allocation and assignment succeeds, but later
1616 * portions of this code fail, there is no need to free the sem_undo_list.
1617 * Just let it stay associated with the task, and it'll be freed later
1618 * at exit time.
1619 *
1620 * This can block, so callers must hold no locks.
1621 */
1622static inline int get_undo_list(struct sem_undo_list **undo_listp)
1623{
1624 struct sem_undo_list *undo_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625
1626 undo_list = current->sysvsem.undo_list;
1627 if (!undo_list) {
Matt Helsley2453a302006-10-02 02:18:25 -07001628 undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629 if (undo_list == NULL)
1630 return -ENOMEM;
Ingo Molnar00a5dfd2005-08-05 23:05:27 +02001631 spin_lock_init(&undo_list->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 atomic_set(&undo_list->refcnt, 1);
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001633 INIT_LIST_HEAD(&undo_list->list_proc);
1634
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635 current->sysvsem.undo_list = undo_list;
1636 }
1637 *undo_listp = undo_list;
1638 return 0;
1639}
1640
Nick Pigginbf17bb72009-12-15 16:47:28 -08001641static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642{
Nick Pigginbf17bb72009-12-15 16:47:28 -08001643 struct sem_undo *un;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644
Nick Pigginbf17bb72009-12-15 16:47:28 -08001645 list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) {
1646 if (un->semid == semid)
1647 return un;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648 }
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001649 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650}
1651
Nick Pigginbf17bb72009-12-15 16:47:28 -08001652static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1653{
1654 struct sem_undo *un;
1655
Manfred Spraul239521f2014-01-27 17:07:04 -08001656 assert_spin_locked(&ulp->lock);
Nick Pigginbf17bb72009-12-15 16:47:28 -08001657
1658 un = __lookup_undo(ulp, semid);
1659 if (un) {
1660 list_del_rcu(&un->list_proc);
1661 list_add_rcu(&un->list_proc, &ulp->list_proc);
1662 }
1663 return un;
1664}
1665
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001666/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -08001667 * find_alloc_undo - lookup (and if not present create) undo array
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001668 * @ns: namespace
1669 * @semid: semaphore array id
1670 *
1671 * The function looks up (and if not present creates) the undo structure.
1672 * The size of the undo structure depends on the size of the semaphore
1673 * array, thus the alloc path is not that straightforward.
Manfred Spraul380af1b2008-07-25 01:48:06 -07001674 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1675 * performs a rcu_read_lock().
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001676 */
1677static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678{
1679 struct sem_array *sma;
1680 struct sem_undo_list *ulp;
1681 struct sem_undo *un, *new;
Rik van Riel6062a8d2013-04-30 19:15:44 -07001682 int nsems, error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683
1684 error = get_undo_list(&ulp);
1685 if (error)
1686 return ERR_PTR(error);
1687
Manfred Spraul380af1b2008-07-25 01:48:06 -07001688 rcu_read_lock();
Pierre Peifferc530c6a2007-10-18 23:40:55 -07001689 spin_lock(&ulp->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690 un = lookup_undo(ulp, semid);
Pierre Peifferc530c6a2007-10-18 23:40:55 -07001691 spin_unlock(&ulp->lock);
Manfred Spraul239521f2014-01-27 17:07:04 -08001692 if (likely(un != NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693 goto out;
1694
1695 /* no undo structure around - allocate one. */
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001696 /* step 1: figure out the size of the semaphore array */
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001697 sma = sem_obtain_object_check(ns, semid);
1698 if (IS_ERR(sma)) {
1699 rcu_read_unlock();
Julia Lawall4de85cd2010-05-26 14:43:44 -07001700 return ERR_CAST(sma);
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001701 }
Nadia Derbey023a5352007-10-18 23:40:51 -07001702
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703 nsems = sma->sem_nsems;
Rik van Riel6062a8d2013-04-30 19:15:44 -07001704 if (!ipc_rcu_getref(sma)) {
1705 rcu_read_unlock();
1706 un = ERR_PTR(-EIDRM);
1707 goto out;
1708 }
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001709 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001711 /* step 2: allocate new undo structure */
Burman Yan4668edc2006-12-06 20:38:51 -08001712 new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713 if (!new) {
Fabian Frederick9b24fef2016-08-02 14:03:07 -07001714 ipc_rcu_putref(sma, sem_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 return ERR_PTR(-ENOMEM);
1716 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717
Manfred Spraul380af1b2008-07-25 01:48:06 -07001718 /* step 3: Acquire the lock on semaphore array */
Linus Torvalds4091fd942013-05-04 10:13:40 -07001719 rcu_read_lock();
Pierre Peiffer6ff37972008-04-29 01:00:46 -07001720 sem_lock_and_putref(sma);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001721 if (!ipc_valid_object(&sma->sem_perm)) {
Rik van Riel6062a8d2013-04-30 19:15:44 -07001722 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001723 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724 kfree(new);
1725 un = ERR_PTR(-EIDRM);
1726 goto out;
1727 }
Manfred Spraul380af1b2008-07-25 01:48:06 -07001728 spin_lock(&ulp->lock);
1729
1730 /*
1731 * step 4: check for races: did someone else allocate the undo struct?
1732 */
1733 un = lookup_undo(ulp, semid);
1734 if (un) {
1735 kfree(new);
1736 goto success;
1737 }
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001738 /* step 5: initialize & link new undo structure */
1739 new->semadj = (short *) &new[1];
Manfred Spraul380af1b2008-07-25 01:48:06 -07001740 new->ulp = ulp;
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001741 new->semid = semid;
1742 assert_spin_locked(&ulp->lock);
Manfred Spraul380af1b2008-07-25 01:48:06 -07001743 list_add_rcu(&new->list_proc, &ulp->list_proc);
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -07001744 ipc_assert_locked_object(&sma->sem_perm);
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001745 list_add(&new->list_id, &sma->list_id);
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001746 un = new;
Manfred Spraul380af1b2008-07-25 01:48:06 -07001747
1748success:
1749 spin_unlock(&ulp->lock);
Rik van Riel6062a8d2013-04-30 19:15:44 -07001750 sem_unlock(sma, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751out:
1752 return un;
1753}
1754
Heiko Carstensd5460c92009-01-14 14:14:27 +01001755SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
1756 unsigned, nsops, const struct timespec __user *, timeout)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757{
1758 int error = -EINVAL;
1759 struct sem_array *sma;
1760 struct sembuf fast_sops[SEMOPM_FAST];
Manfred Spraul239521f2014-01-27 17:07:04 -08001761 struct sembuf *sops = fast_sops, *sop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 struct sem_undo *un;
Davidlohr Bueso4ce33ec2016-12-14 15:06:37 -08001763 int max, locknum;
1764 bool undos = false, alter = false, dupsop = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765 struct sem_queue queue;
Davidlohr Bueso4ce33ec2016-12-14 15:06:37 -08001766 unsigned long dup = 0, jiffies_left = 0;
Kirill Korotaeve3893532006-10-02 02:18:22 -07001767 struct ipc_namespace *ns;
1768
1769 ns = current->nsproxy->ipc_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770
1771 if (nsops < 1 || semid < 0)
1772 return -EINVAL;
Kirill Korotaeve3893532006-10-02 02:18:22 -07001773 if (nsops > ns->sc_semopm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774 return -E2BIG;
Manfred Spraul239521f2014-01-27 17:07:04 -08001775 if (nsops > SEMOPM_FAST) {
1776 sops = kmalloc(sizeof(*sops)*nsops, GFP_KERNEL);
1777 if (sops == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778 return -ENOMEM;
1779 }
Davidlohr Bueso4ce33ec2016-12-14 15:06:37 -08001780
Manfred Spraul239521f2014-01-27 17:07:04 -08001781 if (copy_from_user(sops, tsops, nsops * sizeof(*tsops))) {
1782 error = -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783 goto out_free;
1784 }
Davidlohr Bueso4ce33ec2016-12-14 15:06:37 -08001785
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786 if (timeout) {
1787 struct timespec _timeout;
1788 if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1789 error = -EFAULT;
1790 goto out_free;
1791 }
1792 if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1793 _timeout.tv_nsec >= 1000000000L) {
1794 error = -EINVAL;
1795 goto out_free;
1796 }
1797 jiffies_left = timespec_to_jiffies(&_timeout);
1798 }
Davidlohr Bueso4ce33ec2016-12-14 15:06:37 -08001799
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800 max = 0;
1801 for (sop = sops; sop < sops + nsops; sop++) {
Davidlohr Bueso4ce33ec2016-12-14 15:06:37 -08001802 unsigned long mask = 1ULL << ((sop->sem_num) % BITS_PER_LONG);
1803
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804 if (sop->sem_num >= max)
1805 max = sop->sem_num;
1806 if (sop->sem_flg & SEM_UNDO)
Davidlohr Bueso4ce33ec2016-12-14 15:06:37 -08001807 undos = true;
1808 if (dup & mask) {
1809 /*
1810 * There was a previous alter access that appears
1811 * to have accessed the same semaphore, thus use
1812 * the dupsop logic. "appears", because the detection
1813 * can only check % BITS_PER_LONG.
1814 */
1815 dupsop = true;
1816 }
1817 if (sop->sem_op != 0) {
1818 alter = true;
1819 dup |= mask;
1820 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823 if (undos) {
Rik van Riel6062a8d2013-04-30 19:15:44 -07001824 /* On success, find_alloc_undo takes the rcu_read_lock */
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001825 un = find_alloc_undo(ns, semid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826 if (IS_ERR(un)) {
1827 error = PTR_ERR(un);
1828 goto out_free;
1829 }
Rik van Riel6062a8d2013-04-30 19:15:44 -07001830 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831 un = NULL;
Rik van Riel6062a8d2013-04-30 19:15:44 -07001832 rcu_read_lock();
1833 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001835 sma = sem_obtain_object_check(ns, semid);
Nadia Derbey023a5352007-10-18 23:40:51 -07001836 if (IS_ERR(sma)) {
Rik van Riel6062a8d2013-04-30 19:15:44 -07001837 rcu_read_unlock();
Nadia Derbey023a5352007-10-18 23:40:51 -07001838 error = PTR_ERR(sma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839 goto out_free;
Nadia Derbey023a5352007-10-18 23:40:51 -07001840 }
1841
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001842 error = -EFBIG;
Davidlohr Bueso248e7352016-12-14 15:06:31 -08001843 if (max >= sma->sem_nsems) {
1844 rcu_read_unlock();
1845 goto out_free;
1846 }
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001847
1848 error = -EACCES;
Davidlohr Bueso248e7352016-12-14 15:06:31 -08001849 if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO)) {
1850 rcu_read_unlock();
1851 goto out_free;
1852 }
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001853
1854 error = security_sem_semop(sma, sops, nsops, alter);
Davidlohr Bueso248e7352016-12-14 15:06:31 -08001855 if (error) {
1856 rcu_read_unlock();
1857 goto out_free;
1858 }
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001859
Manfred Spraul6e224f92013-10-16 13:46:45 -07001860 error = -EIDRM;
1861 locknum = sem_lock(sma, sops, nsops);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001862 /*
1863 * We eventually might perform the following check in a lockless
1864 * fashion, considering ipc_valid_object() locking constraints.
1865 * If nsops == 1 and there is no contention for sem_perm.lock, then
1866 * only a per-semaphore lock is held and it's OK to proceed with the
1867 * check below. More details on the fine grained locking scheme
1868 * entangled here and why it's RMID race safe on comments at sem_lock()
1869 */
1870 if (!ipc_valid_object(&sma->sem_perm))
Manfred Spraul6e224f92013-10-16 13:46:45 -07001871 goto out_unlock_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872 /*
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001873 * semid identifiers are not unique - find_alloc_undo may have
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874 * allocated an undo structure, it was invalidated by an RMID
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001875 * and now a new array with received the same id. Check and fail.
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001876 * This case can be detected checking un->semid. The existence of
Manfred Spraul380af1b2008-07-25 01:48:06 -07001877 * "un" itself is guaranteed by rcu.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878 */
Rik van Riel6062a8d2013-04-30 19:15:44 -07001879 if (un && un->semid == -1)
1880 goto out_unlock_free;
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001881
Manfred Sprauld198cd62014-06-06 14:37:49 -07001882 queue.sops = sops;
1883 queue.nsops = nsops;
1884 queue.undo = un;
1885 queue.pid = task_tgid_vnr(current);
1886 queue.alter = alter;
Davidlohr Bueso4ce33ec2016-12-14 15:06:37 -08001887 queue.dupsop = dupsop;
Manfred Sprauld198cd62014-06-06 14:37:49 -07001888
1889 error = perform_atomic_semop(sma, &queue);
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08001890 if (error == 0) { /* non-blocking succesfull path */
1891 DEFINE_WAKE_Q(wake_q);
1892
1893 /*
1894 * If the operation was successful, then do
Manfred Spraul0e8c6652013-09-30 13:45:25 -07001895 * the required updates.
1896 */
1897 if (alter)
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08001898 do_smart_update(sma, sops, nsops, 1, &wake_q);
Manfred Spraul0e8c6652013-09-30 13:45:25 -07001899 else
1900 set_semotime(sma, sops);
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08001901
1902 sem_unlock(sma, locknum);
1903 rcu_read_unlock();
1904 wake_up_q(&wake_q);
1905
1906 goto out_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907 }
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08001908 if (error < 0) /* non-blocking error path */
Manfred Spraul0e8c6652013-09-30 13:45:25 -07001909 goto out_unlock_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08001911 /*
1912 * We need to sleep on this operation, so we put the current
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913 * task into the pending queue and go to sleep.
1914 */
Manfred Spraulb97e8202009-12-15 16:47:32 -08001915 if (nsops == 1) {
1916 struct sem *curr;
1917 curr = &sma->sem_base[sops->sem_num];
1918
Manfred Spraulf269f402013-07-08 16:01:24 -07001919 if (alter) {
1920 if (sma->complex_count) {
1921 list_add_tail(&queue.list,
1922 &sma->pending_alter);
1923 } else {
1924
1925 list_add_tail(&queue.list,
1926 &curr->pending_alter);
1927 }
1928 } else {
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001929 list_add_tail(&queue.list, &curr->pending_const);
Manfred Spraulf269f402013-07-08 16:01:24 -07001930 }
Manfred Spraulb97e8202009-12-15 16:47:32 -08001931 } else {
Manfred Spraulf269f402013-07-08 16:01:24 -07001932 if (!sma->complex_count)
1933 merge_queues(sma);
1934
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001935 if (alter)
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001936 list_add_tail(&queue.list, &sma->pending_alter);
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001937 else
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001938 list_add_tail(&queue.list, &sma->pending_const);
1939
Manfred Spraulb97e8202009-12-15 16:47:32 -08001940 sma->complex_count++;
1941 }
1942
Davidlohr Buesob5fa01a2016-12-14 15:06:46 -08001943 do {
1944 queue.status = -EINTR;
1945 queue.sleeper = current;
Manfred Spraul0b0577f2011-11-02 13:38:52 -07001946
Davidlohr Buesob5fa01a2016-12-14 15:06:46 -08001947 __set_current_state(TASK_INTERRUPTIBLE);
1948 sem_unlock(sma, locknum);
Linus Torvalds321310c2013-05-04 10:47:57 -07001949 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950
Davidlohr Buesob5fa01a2016-12-14 15:06:46 -08001951 if (timeout)
1952 jiffies_left = schedule_timeout(jiffies_left);
1953 else
1954 schedule();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955
Davidlohr Buesob5fa01a2016-12-14 15:06:46 -08001956 /*
1957 * fastpath: the semop has completed, either successfully or
1958 * not, from the syscall pov, is quite irrelevant to us at this
1959 * point; we're done.
1960 *
1961 * We _do_ care, nonetheless, about being awoken by a signal or
1962 * spuriously. The queue.status is checked again in the
1963 * slowpath (aka after taking sem_lock), such that we can detect
1964 * scenarios where we were awakened externally, during the
1965 * window between wake_q_add() and wake_up_q().
1966 */
1967 error = READ_ONCE(queue.status);
1968 if (error != -EINTR) {
1969 /*
1970 * User space could assume that semop() is a memory
1971 * barrier: Without the mb(), the cpu could
1972 * speculatively read in userspace stale data that was
1973 * overwritten by the previous owner of the semaphore.
1974 */
1975 smp_mb();
1976 goto out_free;
1977 }
Manfred Spraul0b0577f2011-11-02 13:38:52 -07001978
Davidlohr Buesob5fa01a2016-12-14 15:06:46 -08001979 rcu_read_lock();
Davidlohr Bueso370b2622016-12-14 15:06:49 -08001980 sem_lock(sma, sops, nsops);
Davidlohr Buesob5fa01a2016-12-14 15:06:46 -08001981
Davidlohr Bueso370b2622016-12-14 15:06:49 -08001982 if (!ipc_valid_object(&sma->sem_perm))
1983 goto out_unlock_free;
1984
1985 error = READ_ONCE(queue.status);
Davidlohr Buesob5fa01a2016-12-14 15:06:46 -08001986
1987 /*
1988 * If queue.status != -EINTR we are woken up by another process.
1989 * Leave without unlink_queue(), but with sem_unlock().
1990 */
1991 if (error != -EINTR)
1992 goto out_unlock_free;
1993
1994 /*
1995 * If an interrupt occurred we have to clean up the queue.
1996 */
1997 if (timeout && jiffies_left == 0)
1998 error = -EAGAIN;
1999 } while (error == -EINTR && !signal_pending(current)); /* spurious */
Manfred Spraul0b0577f2011-11-02 13:38:52 -07002000
Manfred Spraulb97e8202009-12-15 16:47:32 -08002001 unlink_queue(sma, &queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002
2003out_unlock_free:
Rik van Riel6062a8d2013-04-30 19:15:44 -07002004 sem_unlock(sma, locknum);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07002005 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006out_free:
Manfred Spraul239521f2014-01-27 17:07:04 -08002007 if (sops != fast_sops)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008 kfree(sops);
2009 return error;
2010}
2011
Heiko Carstensd5460c92009-01-14 14:14:27 +01002012SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
2013 unsigned, nsops)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014{
2015 return sys_semtimedop(semid, tsops, nsops, NULL);
2016}
2017
2018/* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
2019 * parent and child tasks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020 */
2021
2022int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
2023{
2024 struct sem_undo_list *undo_list;
2025 int error;
2026
2027 if (clone_flags & CLONE_SYSVSEM) {
2028 error = get_undo_list(&undo_list);
2029 if (error)
2030 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031 atomic_inc(&undo_list->refcnt);
2032 tsk->sysvsem.undo_list = undo_list;
Paul McQuade46c0a8c2014-06-06 14:37:37 -07002033 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034 tsk->sysvsem.undo_list = NULL;
2035
2036 return 0;
2037}
2038
2039/*
2040 * add semadj values to semaphores, free undo structures.
2041 * undo structures are not freed when semaphore arrays are destroyed
2042 * so some of them may be out of date.
2043 * IMPLEMENTATION NOTE: There is some confusion over whether the
2044 * set of adjustments that needs to be done should be done in an atomic
2045 * manner or not. That is, if we are attempting to decrement the semval
2046 * should we queue up and wait until we can do so legally?
2047 * The original implementation attempted to do this (queue and wait).
2048 * The current implementation does not do so. The POSIX standard
2049 * and SVID should be consulted to determine what behavior is mandated.
2050 */
2051void exit_sem(struct task_struct *tsk)
2052{
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002053 struct sem_undo_list *ulp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002055 ulp = tsk->sysvsem.undo_list;
2056 if (!ulp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057 return;
Manfred Spraul9edff4a2008-04-29 01:00:57 -07002058 tsk->sysvsem.undo_list = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002060 if (!atomic_dec_and_test(&ulp->refcnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061 return;
2062
Manfred Spraul380af1b2008-07-25 01:48:06 -07002063 for (;;) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064 struct sem_array *sma;
Manfred Spraul380af1b2008-07-25 01:48:06 -07002065 struct sem_undo *un;
Rik van Riel6062a8d2013-04-30 19:15:44 -07002066 int semid, i;
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08002067 DEFINE_WAKE_Q(wake_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068
Nikolay Borisov2a1613a2016-10-11 13:55:05 -07002069 cond_resched();
2070
Manfred Spraul380af1b2008-07-25 01:48:06 -07002071 rcu_read_lock();
Jiri Pirko05725f72009-04-14 20:17:16 +02002072 un = list_entry_rcu(ulp->list_proc.next,
2073 struct sem_undo, list_proc);
Herton R. Krzesinski602b8592015-08-14 15:35:02 -07002074 if (&un->list_proc == &ulp->list_proc) {
2075 /*
2076 * We must wait for freeary() before freeing this ulp,
2077 * in case we raced with last sem_undo. There is a small
2078 * possibility where we exit while freeary() didn't
2079 * finish unlocking sem_undo_list.
2080 */
2081 spin_unlock_wait(&ulp->lock);
Rik van Riel6062a8d2013-04-30 19:15:44 -07002082 rcu_read_unlock();
Manfred Spraul380af1b2008-07-25 01:48:06 -07002083 break;
Rik van Riel6062a8d2013-04-30 19:15:44 -07002084 }
Herton R. Krzesinski602b8592015-08-14 15:35:02 -07002085 spin_lock(&ulp->lock);
2086 semid = un->semid;
2087 spin_unlock(&ulp->lock);
Manfred Spraul380af1b2008-07-25 01:48:06 -07002088
Herton R. Krzesinski602b8592015-08-14 15:35:02 -07002089 /* exit_sem raced with IPC_RMID, nothing to do */
2090 if (semid == -1) {
2091 rcu_read_unlock();
2092 continue;
2093 }
2094
2095 sma = sem_obtain_object_check(tsk->nsproxy->ipc_ns, semid);
Manfred Spraul380af1b2008-07-25 01:48:06 -07002096 /* exit_sem raced with IPC_RMID, nothing to do */
Rik van Riel6062a8d2013-04-30 19:15:44 -07002097 if (IS_ERR(sma)) {
2098 rcu_read_unlock();
Manfred Spraul380af1b2008-07-25 01:48:06 -07002099 continue;
Rik van Riel6062a8d2013-04-30 19:15:44 -07002100 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002101
Rik van Riel6062a8d2013-04-30 19:15:44 -07002102 sem_lock(sma, NULL, -1);
Manfred Spraul6e224f92013-10-16 13:46:45 -07002103 /* exit_sem raced with IPC_RMID, nothing to do */
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08002104 if (!ipc_valid_object(&sma->sem_perm)) {
Manfred Spraul6e224f92013-10-16 13:46:45 -07002105 sem_unlock(sma, -1);
2106 rcu_read_unlock();
2107 continue;
2108 }
Nick Pigginbf17bb72009-12-15 16:47:28 -08002109 un = __lookup_undo(ulp, semid);
Manfred Spraul380af1b2008-07-25 01:48:06 -07002110 if (un == NULL) {
2111 /* exit_sem raced with IPC_RMID+semget() that created
2112 * exactly the same semid. Nothing to do.
2113 */
Rik van Riel6062a8d2013-04-30 19:15:44 -07002114 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07002115 rcu_read_unlock();
Manfred Spraul380af1b2008-07-25 01:48:06 -07002116 continue;
2117 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118
Manfred Spraul380af1b2008-07-25 01:48:06 -07002119 /* remove un from the linked lists */
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -07002120 ipc_assert_locked_object(&sma->sem_perm);
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002121 list_del(&un->list_id);
2122
Herton R. Krzesinskia9795582015-08-14 15:35:05 -07002123 /* we are the last process using this ulp, acquiring ulp->lock
2124 * isn't required. Besides that, we are also protected against
2125 * IPC_RMID as we hold sma->sem_perm lock now
2126 */
Manfred Spraul380af1b2008-07-25 01:48:06 -07002127 list_del_rcu(&un->list_proc);
Manfred Spraul380af1b2008-07-25 01:48:06 -07002128
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002129 /* perform adjustments registered in un */
2130 for (i = 0; i < sma->sem_nsems; i++) {
Manfred Spraul239521f2014-01-27 17:07:04 -08002131 struct sem *semaphore = &sma->sem_base[i];
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002132 if (un->semadj[i]) {
2133 semaphore->semval += un->semadj[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134 /*
2135 * Range checks of the new semaphore value,
2136 * not defined by sus:
2137 * - Some unices ignore the undo entirely
2138 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
2139 * - some cap the value (e.g. FreeBSD caps
2140 * at 0, but doesn't enforce SEMVMX)
2141 *
2142 * Linux caps the semaphore value, both at 0
2143 * and at SEMVMX.
2144 *
Manfred Spraul239521f2014-01-27 17:07:04 -08002145 * Manfred <manfred@colorfullife.com>
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146 */
Ingo Molnar5f921ae2006-03-26 01:37:17 -08002147 if (semaphore->semval < 0)
2148 semaphore->semval = 0;
2149 if (semaphore->semval > SEMVMX)
2150 semaphore->semval = SEMVMX;
Pavel Emelyanovb4888932007-10-18 23:40:14 -07002151 semaphore->sempid = task_tgid_vnr(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152 }
2153 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154 /* maybe some queued-up processes were waiting for this */
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08002155 do_smart_update(sma, NULL, 0, 1, &wake_q);
Rik van Riel6062a8d2013-04-30 19:15:44 -07002156 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07002157 rcu_read_unlock();
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08002158 wake_up_q(&wake_q);
Manfred Spraul380af1b2008-07-25 01:48:06 -07002159
Lai Jiangshan693a8b62011-03-18 12:09:35 +08002160 kfree_rcu(un, rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161 }
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002162 kfree(ulp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163}
2164
2165#ifdef CONFIG_PROC_FS
Mike Waychison19b49462005-09-06 15:17:10 -07002166static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002167{
Eric W. Biederman1efdb692012-02-07 16:54:11 -08002168 struct user_namespace *user_ns = seq_user_ns(s);
Mike Waychison19b49462005-09-06 15:17:10 -07002169 struct sem_array *sma = it;
Manfred Sprauld12e1e52013-07-08 16:01:25 -07002170 time_t sem_otime;
2171
Manfred Sprauld8c63372013-09-30 13:45:07 -07002172 /*
2173 * The proc interface isn't aware of sem_lock(), it calls
2174 * ipc_lock_object() directly (in sysvipc_find_ipc).
Manfred Spraul5864a2f2016-10-11 13:54:50 -07002175 * In order to stay compatible with sem_lock(), we must
2176 * enter / leave complex_mode.
Manfred Sprauld8c63372013-09-30 13:45:07 -07002177 */
Manfred Spraul5864a2f2016-10-11 13:54:50 -07002178 complexmode_enter(sma);
Manfred Sprauld8c63372013-09-30 13:45:07 -07002179
Manfred Sprauld12e1e52013-07-08 16:01:25 -07002180 sem_otime = get_semotime(sma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181
Joe Perches7f032d62015-04-15 16:17:54 -07002182 seq_printf(s,
2183 "%10d %10d %4o %10u %5u %5u %5u %5u %10lu %10lu\n",
2184 sma->sem_perm.key,
2185 sma->sem_perm.id,
2186 sma->sem_perm.mode,
2187 sma->sem_nsems,
2188 from_kuid_munged(user_ns, sma->sem_perm.uid),
2189 from_kgid_munged(user_ns, sma->sem_perm.gid),
2190 from_kuid_munged(user_ns, sma->sem_perm.cuid),
2191 from_kgid_munged(user_ns, sma->sem_perm.cgid),
2192 sem_otime,
2193 sma->sem_ctime);
2194
Manfred Spraul5864a2f2016-10-11 13:54:50 -07002195 complexmode_tryleave(sma);
2196
Joe Perches7f032d62015-04-15 16:17:54 -07002197 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198}
2199#endif