blob: 3687b71151b3921860613dcc7089fa2831f6237e [file] [log] [blame]
Greg Kroah-Hartmanb2441312017-11-01 15:07:57 +01001// SPDX-License-Identifier: GPL-2.0
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
3 * linux/ipc/sem.c
4 * Copyright (C) 1992 Krishna Balasubramanian
5 * Copyright (C) 1995 Eric Schenk, Bruno Haible
6 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 * /proc/sysvipc/sem support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
8 *
9 * SMP-threaded, sysctl's added
Christian Kujau624dffc2006-01-15 02:43:54 +010010 * (c) 1999 Manfred Spraul <manfred@colorfullife.com>
Linus Torvalds1da177e2005-04-16 15:20:36 -070011 * Enforced range limit on SEM_UNDO
Alan Cox046c6882009-01-05 14:06:29 +000012 * (c) 2001 Red Hat Inc
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 * Lockless wakeup
14 * (c) 2003 Manfred Spraul <manfred@colorfullife.com>
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -080015 * (c) 2016 Davidlohr Bueso <dave@stgolabs.net>
Manfred Spraulc5cf6352010-05-26 14:43:43 -070016 * Further wakeup optimizations, documentation
17 * (c) 2010 Manfred Spraul <manfred@colorfullife.com>
Steve Grubb073115d2006-04-02 17:07:33 -040018 *
19 * support for audit of ipc object properties and permission changes
20 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
Kirill Korotaeve3893532006-10-02 02:18:22 -070021 *
22 * namespaces support
23 * OpenVZ, SWsoft Inc.
24 * Pavel Emelianov <xemul@openvz.org>
Manfred Spraulc5cf6352010-05-26 14:43:43 -070025 *
26 * Implementation notes: (May 2010)
27 * This file implements System V semaphores.
28 *
29 * User space visible behavior:
30 * - FIFO ordering for semop() operations (just FIFO, not starvation
31 * protection)
32 * - multiple semaphore operations that alter the same semaphore in
33 * one semop() are handled.
34 * - sem_ctime (time of last semctl()) is updated in the IPC_SET, SETVAL and
35 * SETALL calls.
36 * - two Linux specific semctl() commands: SEM_STAT, SEM_INFO.
37 * - undo adjustments at process exit are limited to 0..SEMVMX.
38 * - namespace are supported.
39 * - SEMMSL, SEMMNS, SEMOPM and SEMMNI can be configured at runtine by writing
40 * to /proc/sys/kernel/sem.
41 * - statistics about the usage are reported in /proc/sysvipc/sem.
42 *
43 * Internals:
44 * - scalability:
45 * - all global variables are read-mostly.
46 * - semop() calls and semctl(RMID) are synchronized by RCU.
47 * - most operations do write operations (actually: spin_lock calls) to
48 * the per-semaphore array structure.
49 * Thus: Perfect SMP scaling between independent semaphore arrays.
50 * If multiple semaphores in one array are used, then cache line
51 * trashing on the semaphore array spinlock will limit the scaling.
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -070052 * - semncnt and semzcnt are calculated on demand in count_semcnt()
Manfred Spraulc5cf6352010-05-26 14:43:43 -070053 * - the task that performs a successful semop() scans the list of all
54 * sleeping tasks and completes any pending operations that can be fulfilled.
55 * Semaphores are actively given to waiting tasks (necessary for FIFO).
56 * (see update_queue())
57 * - To improve the scalability, the actual wake-up calls are performed after
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -080058 * dropping all locks. (see wake_up_sem_queue_prepare())
Manfred Spraulc5cf6352010-05-26 14:43:43 -070059 * - All work is done by the waker, the woken up task does not have to do
60 * anything - not even acquiring a lock or dropping a refcount.
61 * - A woken up task may not even touch the semaphore array anymore, it may
62 * have been destroyed already by a semctl(RMID).
Manfred Spraulc5cf6352010-05-26 14:43:43 -070063 * - UNDO values are stored in an array (one per process and per
64 * semaphore array, lazily allocated). For backwards compatibility, multiple
65 * modes for the UNDO variables are supported (per process, per thread)
66 * (see copy_semundo, CLONE_SYSVSEM)
67 * - There are two lists of the pending operations: a per-array list
68 * and per-semaphore list (stored in the array). This allows to achieve FIFO
69 * ordering without always scanning all pending operations.
70 * The worst-case behavior is nevertheless O(N^2) for N wakeups.
Linus Torvalds1da177e2005-04-16 15:20:36 -070071 */
72
Arnd Bergmannb0d17572018-04-13 13:58:23 +020073#include <linux/compat.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070074#include <linux/slab.h>
75#include <linux/spinlock.h>
76#include <linux/init.h>
77#include <linux/proc_fs.h>
78#include <linux/time.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070079#include <linux/security.h>
80#include <linux/syscalls.h>
81#include <linux/audit.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080082#include <linux/capability.h>
Mike Waychison19b49462005-09-06 15:17:10 -070083#include <linux/seq_file.h>
Nadia Derbey3e148c72007-10-18 23:40:54 -070084#include <linux/rwsem.h>
Kirill Korotaeve3893532006-10-02 02:18:22 -070085#include <linux/nsproxy.h>
Pavel Emelyanovae5e1b22008-02-08 04:18:22 -080086#include <linux/ipc_namespace.h>
Ingo Molnar84f001e2017-02-01 16:36:40 +010087#include <linux/sched/wake_q.h>
Davidlohr Buesoec67aaa2018-06-14 15:27:51 -070088#include <linux/nospec.h>
NeilBrown0eb71a92018-06-18 12:52:50 +100089#include <linux/rhashtable.h>
Ingo Molnar5f921ae2006-03-26 01:37:17 -080090
Paul McQuade7153e402014-06-06 14:37:37 -070091#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070092#include "util.h"
93
Eric W. Biederman1a5c1342018-03-22 21:30:56 -050094/* One semaphore structure for each semaphore in the system. */
95struct sem {
96 int semval; /* current value */
97 /*
98 * PID of the process that last modified the semaphore. For
99 * Linux, specifically these are:
100 * - semop
101 * - semctl, via SETVAL and SETALL.
102 * - at task exit when performing undo adjustments (see exit_sem).
103 */
Eric W. Biederman51d6f262018-03-23 01:11:29 -0500104 struct pid *sempid;
Eric W. Biederman1a5c1342018-03-22 21:30:56 -0500105 spinlock_t lock; /* spinlock for fine-grained semtimedop */
106 struct list_head pending_alter; /* pending single-sop operations */
107 /* that alter the semaphore */
108 struct list_head pending_const; /* pending single-sop operations */
109 /* that do not alter the semaphore*/
Arnd Bergmann2a70b782018-04-12 15:19:44 +0200110 time64_t sem_otime; /* candidate for sem_otime */
Eric W. Biederman1a5c1342018-03-22 21:30:56 -0500111} ____cacheline_aligned_in_smp;
112
113/* One sem_array data structure for each set of semaphores in the system. */
114struct sem_array {
115 struct kern_ipc_perm sem_perm; /* permissions .. see ipc.h */
116 time64_t sem_ctime; /* create/last semctl() time */
117 struct list_head pending_alter; /* pending operations */
118 /* that alter the array */
119 struct list_head pending_const; /* pending complex operations */
120 /* that do not alter semvals */
121 struct list_head list_id; /* undo requests on this array */
122 int sem_nsems; /* no. of semaphores in array */
123 int complex_count; /* pending complex operations */
124 unsigned int use_global_lock;/* >0: global lock required */
125
126 struct sem sems[];
127} __randomize_layout;
Manfred Spraule57940d2011-11-02 13:38:54 -0700128
129/* One queue for each sleeping process in the system. */
130struct sem_queue {
Manfred Spraule57940d2011-11-02 13:38:54 -0700131 struct list_head list; /* queue of pending operations */
132 struct task_struct *sleeper; /* this process */
133 struct sem_undo *undo; /* undo structure */
Eric W. Biederman51d6f262018-03-23 01:11:29 -0500134 struct pid *pid; /* process id of requesting process */
Manfred Spraule57940d2011-11-02 13:38:54 -0700135 int status; /* completion status of operation */
136 struct sembuf *sops; /* array of pending operations */
Manfred Sprauled247b72014-06-06 14:37:49 -0700137 struct sembuf *blocking; /* the operation that blocked */
Manfred Spraule57940d2011-11-02 13:38:54 -0700138 int nsops; /* number of operations */
Davidlohr Bueso4ce33ec2016-12-14 15:06:37 -0800139 bool alter; /* does *sops alter the array? */
140 bool dupsop; /* sops on more than one sem_num */
Manfred Spraule57940d2011-11-02 13:38:54 -0700141};
142
143/* Each task has a list of undo requests. They are executed automatically
144 * when the process exits.
145 */
146struct sem_undo {
147 struct list_head list_proc; /* per-process list: *
148 * all undos from one process
149 * rcu protected */
150 struct rcu_head rcu; /* rcu struct for sem_undo */
151 struct sem_undo_list *ulp; /* back ptr to sem_undo_list */
152 struct list_head list_id; /* per semaphore array list:
153 * all undos for one array */
154 int semid; /* semaphore set identifier */
155 short *semadj; /* array of adjustments */
156 /* one per semaphore */
157};
158
159/* sem_undo_list controls shared access to the list of sem_undo structures
160 * that may be shared among all a CLONE_SYSVSEM task group.
161 */
162struct sem_undo_list {
Elena Reshetovaf74370b2017-09-08 16:17:42 -0700163 refcount_t refcnt;
Manfred Spraule57940d2011-11-02 13:38:54 -0700164 spinlock_t lock;
165 struct list_head list_proc;
166};
167
168
Pierre Peiffered2ddbf2008-02-08 04:18:57 -0800169#define sem_ids(ns) ((ns)->ids[IPC_SEM_IDS])
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700171static int newary(struct ipc_namespace *, struct ipc_params *);
Pierre Peiffer01b8b072008-02-08 04:18:57 -0800172static void freeary(struct ipc_namespace *, struct kern_ipc_perm *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173#ifdef CONFIG_PROC_FS
Mike Waychison19b49462005-09-06 15:17:10 -0700174static int sysvipc_sem_proc_show(struct seq_file *s, void *it);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175#endif
176
177#define SEMMSL_FAST 256 /* 512 bytes on stack */
178#define SEMOPM_FAST 64 /* ~ 372 bytes on stack */
179
180/*
Manfred Spraul9de5ab82017-02-27 14:28:18 -0800181 * Switching from the mode suitable for simple ops
182 * to the mode for complex ops is costly. Therefore:
183 * use some hysteresis
184 */
185#define USE_GLOBAL_LOCK_HYSTERESIS 10
186
187/*
Manfred Spraul758a6ba32013-07-08 16:01:26 -0700188 * Locking:
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700189 * a) global sem_lock() for read/write
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 * sem_undo.id_next,
Manfred Spraul758a6ba32013-07-08 16:01:26 -0700191 * sem_array.complex_count,
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700192 * sem_array.pending{_alter,_const},
193 * sem_array.sem_undo
Paul McQuade46c0a8c2014-06-06 14:37:37 -0700194 *
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700195 * b) global or semaphore sem_lock() for read/write:
Manfred Spraul1a233952017-07-12 14:34:38 -0700196 * sem_array.sems[i].pending_{const,alter}:
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700197 *
198 * c) special:
199 * sem_undo_list.list_proc:
200 * * undo_list->lock for write
201 * * rcu for read
Manfred Spraul9de5ab82017-02-27 14:28:18 -0800202 * use_global_lock:
203 * * global sem_lock() for write
204 * * either local or global sem_lock() for read.
205 *
206 * Memory ordering:
207 * Most ordering is enforced by using spin_lock() and spin_unlock().
Manfred Spraul8116b542020-02-03 17:34:42 -0800208 *
209 * Exceptions:
210 * 1) use_global_lock: (SEM_BARRIER_1)
Manfred Spraul9de5ab82017-02-27 14:28:18 -0800211 * Setting it from non-zero to 0 is a RELEASE, this is ensured by
Manfred Spraul8116b542020-02-03 17:34:42 -0800212 * using smp_store_release(): Immediately after setting it to 0,
213 * a simple op can start.
Manfred Spraul9de5ab82017-02-27 14:28:18 -0800214 * Testing if it is non-zero is an ACQUIRE, this is ensured by using
215 * smp_load_acquire().
216 * Setting it from 0 to non-zero must be ordered with regards to
217 * this smp_load_acquire(), this is guaranteed because the smp_load_acquire()
218 * is inside a spin_lock() and after a write from 0 to non-zero a
219 * spin_lock()+spin_unlock() is done.
Manfred Spraul8116b542020-02-03 17:34:42 -0800220 *
221 * 2) queue.status: (SEM_BARRIER_2)
222 * Initialization is done while holding sem_lock(), so no further barrier is
223 * required.
224 * Setting it to a result code is a RELEASE, this is ensured by both a
225 * smp_store_release() (for case a) and while holding sem_lock()
226 * (for case b).
227 * The AQUIRE when reading the result code without holding sem_lock() is
228 * achieved by using READ_ONCE() + smp_acquire__after_ctrl_dep().
229 * (case a above).
230 * Reading the result code while holding sem_lock() needs no further barriers,
231 * the locks inside sem_lock() enforce ordering (case b above)
232 *
233 * 3) current->state:
234 * current->state is set to TASK_INTERRUPTIBLE while holding sem_lock().
235 * The wakeup is handled using the wake_q infrastructure. wake_q wakeups may
236 * happen immediately after calling wake_q_add. As wake_q_add_safe() is called
237 * when holding sem_lock(), no further barriers are required.
238 *
239 * See also ipc/mqueue.c for more details on the covered races.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 */
241
Kirill Korotaeve3893532006-10-02 02:18:22 -0700242#define sc_semmsl sem_ctls[0]
243#define sc_semmns sem_ctls[1]
244#define sc_semopm sem_ctls[2]
245#define sc_semmni sem_ctls[3]
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246
Davidlohr Buesoeae04d22018-08-21 22:01:56 -0700247void sem_init_ns(struct ipc_namespace *ns)
Kirill Korotaeve3893532006-10-02 02:18:22 -0700248{
Kirill Korotaeve3893532006-10-02 02:18:22 -0700249 ns->sc_semmsl = SEMMSL;
250 ns->sc_semmns = SEMMNS;
251 ns->sc_semopm = SEMOPM;
252 ns->sc_semmni = SEMMNI;
253 ns->used_sems = 0;
Davidlohr Buesoeae04d22018-08-21 22:01:56 -0700254 ipc_init_ids(&ns->ids[IPC_SEM_IDS]);
Kirill Korotaeve3893532006-10-02 02:18:22 -0700255}
256
Pavel Emelyanovae5e1b22008-02-08 04:18:22 -0800257#ifdef CONFIG_IPC_NS
Kirill Korotaeve3893532006-10-02 02:18:22 -0700258void sem_exit_ns(struct ipc_namespace *ns)
259{
Pierre Peiffer01b8b072008-02-08 04:18:57 -0800260 free_ipcs(ns, &sem_ids(ns), freeary);
Serge E. Hallyn7d6feeb2009-12-15 16:47:27 -0800261 idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr);
Guillaume Knispel0cfb6ae2017-09-08 16:17:55 -0700262 rhashtable_destroy(&ns->ids[IPC_SEM_IDS].key_ht);
Kirill Korotaeve3893532006-10-02 02:18:22 -0700263}
Pavel Emelyanovae5e1b22008-02-08 04:18:22 -0800264#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265
Davidlohr Buesoeae04d22018-08-21 22:01:56 -0700266void __init sem_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267{
Davidlohr Buesoeae04d22018-08-21 22:01:56 -0700268 sem_init_ns(&init_ipc_ns);
Mike Waychison19b49462005-09-06 15:17:10 -0700269 ipc_init_proc_interface("sysvipc/sem",
270 " key semid perms nsems uid gid cuid cgid otime ctime\n",
Kirill Korotaeve3893532006-10-02 02:18:22 -0700271 IPC_SEM_IDS, sysvipc_sem_proc_show);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272}
273
Manfred Spraulf269f402013-07-08 16:01:24 -0700274/**
275 * unmerge_queues - unmerge queues, if possible.
276 * @sma: semaphore array
277 *
278 * The function unmerges the wait queues if complex_count is 0.
279 * It must be called prior to dropping the global semaphore array lock.
280 */
281static void unmerge_queues(struct sem_array *sma)
282{
283 struct sem_queue *q, *tq;
284
285 /* complex operations still around? */
286 if (sma->complex_count)
287 return;
288 /*
289 * We will switch back to simple mode.
290 * Move all pending operation back into the per-semaphore
291 * queues.
292 */
293 list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
294 struct sem *curr;
Manfred Spraul1a233952017-07-12 14:34:38 -0700295 curr = &sma->sems[q->sops[0].sem_num];
Manfred Spraulf269f402013-07-08 16:01:24 -0700296
297 list_add_tail(&q->list, &curr->pending_alter);
298 }
299 INIT_LIST_HEAD(&sma->pending_alter);
300}
301
302/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800303 * merge_queues - merge single semop queues into global queue
Manfred Spraulf269f402013-07-08 16:01:24 -0700304 * @sma: semaphore array
305 *
306 * This function merges all per-semaphore queues into the global queue.
307 * It is necessary to achieve FIFO ordering for the pending single-sop
308 * operations when a multi-semop operation must sleep.
309 * Only the alter operations must be moved, the const operations can stay.
310 */
311static void merge_queues(struct sem_array *sma)
312{
313 int i;
314 for (i = 0; i < sma->sem_nsems; i++) {
Manfred Spraul1a233952017-07-12 14:34:38 -0700315 struct sem *sem = &sma->sems[i];
Manfred Spraulf269f402013-07-08 16:01:24 -0700316
317 list_splice_init(&sem->pending_alter, &sma->pending_alter);
318 }
319}
320
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -0700321static void sem_rcu_free(struct rcu_head *head)
322{
Manfred Sprauldba4cdd2017-07-12 14:34:41 -0700323 struct kern_ipc_perm *p = container_of(head, struct kern_ipc_perm, rcu);
324 struct sem_array *sma = container_of(p, struct sem_array, sem_perm);
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -0700325
Eric W. Biedermanaefad952018-03-22 20:52:43 -0500326 security_sem_free(&sma->sem_perm);
Kees Cooke2029df2017-07-12 14:35:31 -0700327 kvfree(sma);
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -0700328}
329
Nadia Derbey3e148c72007-10-18 23:40:54 -0700330/*
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700331 * Enter the mode suitable for non-simple operations:
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700332 * Caller must own sem_perm.lock.
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700333 */
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700334static void complexmode_enter(struct sem_array *sma)
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700335{
336 int i;
337 struct sem *sem;
338
Manfred Spraul9de5ab82017-02-27 14:28:18 -0800339 if (sma->use_global_lock > 0) {
340 /*
341 * We are already in global lock mode.
342 * Nothing to do, just reset the
343 * counter until we return to simple mode.
344 */
345 sma->use_global_lock = USE_GLOBAL_LOCK_HYSTERESIS;
Manfred Spraul6d07b682013-09-30 13:45:06 -0700346 return;
347 }
Manfred Spraul9de5ab82017-02-27 14:28:18 -0800348 sma->use_global_lock = USE_GLOBAL_LOCK_HYSTERESIS;
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700349
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700350 for (i = 0; i < sma->sem_nsems; i++) {
Manfred Spraul1a233952017-07-12 14:34:38 -0700351 sem = &sma->sems[i];
Manfred Spraul27d7be12017-02-27 14:28:15 -0800352 spin_lock(&sem->lock);
353 spin_unlock(&sem->lock);
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700354 }
355}
356
357/*
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700358 * Try to leave the mode that disallows simple operations:
359 * Caller must own sem_perm.lock.
360 */
361static void complexmode_tryleave(struct sem_array *sma)
362{
363 if (sma->complex_count) {
364 /* Complex ops are sleeping.
365 * We must stay in complex mode
366 */
367 return;
368 }
Manfred Spraul9de5ab82017-02-27 14:28:18 -0800369 if (sma->use_global_lock == 1) {
Manfred Spraul8116b542020-02-03 17:34:42 -0800370
371 /* See SEM_BARRIER_1 for purpose/pairing */
Manfred Spraul9de5ab82017-02-27 14:28:18 -0800372 smp_store_release(&sma->use_global_lock, 0);
373 } else {
374 sma->use_global_lock--;
375 }
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700376}
377
378#define SEM_GLOBAL_LOCK (-1)
379/*
Rik van Riel6062a8d2013-04-30 19:15:44 -0700380 * If the request contains only one semaphore operation, and there are
381 * no complex transactions pending, lock only the semaphore involved.
382 * Otherwise, lock the entire semaphore array, since we either have
383 * multiple semaphores in our own semops, or we need to look at
384 * semaphores from other pending complex operations.
Rik van Riel6062a8d2013-04-30 19:15:44 -0700385 */
386static inline int sem_lock(struct sem_array *sma, struct sembuf *sops,
387 int nsops)
388{
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700389 struct sem *sem;
Davidlohr Buesoec67aaa2018-06-14 15:27:51 -0700390 int idx;
Rik van Riel6062a8d2013-04-30 19:15:44 -0700391
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700392 if (nsops != 1) {
393 /* Complex operation - acquire a full lock */
394 ipc_lock_object(&sma->sem_perm);
395
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700396 /* Prevent parallel simple ops */
397 complexmode_enter(sma);
398 return SEM_GLOBAL_LOCK;
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700399 }
400
401 /*
402 * Only one semaphore affected - try to optimize locking.
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700403 * Optimized locking is possible if no complex operation
404 * is either enqueued or processed right now.
405 *
Manfred Spraul9de5ab82017-02-27 14:28:18 -0800406 * Both facts are tracked by use_global_mode.
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700407 */
Davidlohr Buesoec67aaa2018-06-14 15:27:51 -0700408 idx = array_index_nospec(sops->sem_num, sma->sem_nsems);
409 sem = &sma->sems[idx];
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700410
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700411 /*
Manfred Spraul9de5ab82017-02-27 14:28:18 -0800412 * Initial check for use_global_lock. Just an optimization,
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700413 * no locking, no memory barrier.
414 */
Manfred Spraul9de5ab82017-02-27 14:28:18 -0800415 if (!sma->use_global_lock) {
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700416 /*
417 * It appears that no complex operation is around.
418 * Acquire the per-semaphore lock.
419 */
Rik van Riel6062a8d2013-04-30 19:15:44 -0700420 spin_lock(&sem->lock);
421
Manfred Spraul8116b542020-02-03 17:34:42 -0800422 /* see SEM_BARRIER_1 for purpose/pairing */
Manfred Spraul9de5ab82017-02-27 14:28:18 -0800423 if (!smp_load_acquire(&sma->use_global_lock)) {
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700424 /* fast path successful! */
425 return sops->sem_num;
Rik van Riel6062a8d2013-04-30 19:15:44 -0700426 }
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700427 spin_unlock(&sem->lock);
Rik van Riel6062a8d2013-04-30 19:15:44 -0700428 }
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700429
430 /* slow path: acquire the full lock */
431 ipc_lock_object(&sma->sem_perm);
432
Manfred Spraul9de5ab82017-02-27 14:28:18 -0800433 if (sma->use_global_lock == 0) {
434 /*
435 * The use_global_lock mode ended while we waited for
436 * sma->sem_perm.lock. Thus we must switch to locking
437 * with sem->lock.
438 * Unlike in the fast path, there is no need to recheck
439 * sma->use_global_lock after we have acquired sem->lock:
440 * We own sma->sem_perm.lock, thus use_global_lock cannot
441 * change.
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700442 */
443 spin_lock(&sem->lock);
Manfred Spraul9de5ab82017-02-27 14:28:18 -0800444
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700445 ipc_unlock_object(&sma->sem_perm);
446 return sops->sem_num;
447 } else {
Manfred Spraul9de5ab82017-02-27 14:28:18 -0800448 /*
449 * Not a false alarm, thus continue to use the global lock
450 * mode. No need for complexmode_enter(), this was done by
451 * the caller that has set use_global_mode to non-zero.
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700452 */
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700453 return SEM_GLOBAL_LOCK;
Manfred Spraul5e9d5272013-09-30 13:45:04 -0700454 }
Rik van Riel6062a8d2013-04-30 19:15:44 -0700455}
456
457static inline void sem_unlock(struct sem_array *sma, int locknum)
458{
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700459 if (locknum == SEM_GLOBAL_LOCK) {
Manfred Spraulf269f402013-07-08 16:01:24 -0700460 unmerge_queues(sma);
Manfred Spraul5864a2f2016-10-11 13:54:50 -0700461 complexmode_tryleave(sma);
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -0700462 ipc_unlock_object(&sma->sem_perm);
Rik van Riel6062a8d2013-04-30 19:15:44 -0700463 } else {
Manfred Spraul1a233952017-07-12 14:34:38 -0700464 struct sem *sem = &sma->sems[locknum];
Rik van Riel6062a8d2013-04-30 19:15:44 -0700465 spin_unlock(&sem->lock);
466 }
Rik van Riel6062a8d2013-04-30 19:15:44 -0700467}
468
469/*
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700470 * sem_lock_(check_) routines are called in the paths where the rwsem
Nadia Derbey3e148c72007-10-18 23:40:54 -0700471 * is not held.
Linus Torvalds321310c2013-05-04 10:47:57 -0700472 *
473 * The caller holds the RCU read lock.
Nadia Derbey3e148c72007-10-18 23:40:54 -0700474 */
Davidlohr Bueso16df3672013-04-30 19:15:29 -0700475static inline struct sem_array *sem_obtain_object(struct ipc_namespace *ns, int id)
476{
Davidlohr Bueso55b7ae52015-06-30 14:58:42 -0700477 struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&sem_ids(ns), id);
Davidlohr Bueso16df3672013-04-30 19:15:29 -0700478
479 if (IS_ERR(ipcp))
480 return ERR_CAST(ipcp);
481
482 return container_of(ipcp, struct sem_array, sem_perm);
483}
484
Davidlohr Bueso16df3672013-04-30 19:15:29 -0700485static inline struct sem_array *sem_obtain_object_check(struct ipc_namespace *ns,
486 int id)
487{
488 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&sem_ids(ns), id);
489
490 if (IS_ERR(ipcp))
491 return ERR_CAST(ipcp);
Pierre Peifferb1ed88b2008-02-06 01:36:23 -0800492
Nadia Derbey03f02c72007-10-18 23:40:51 -0700493 return container_of(ipcp, struct sem_array, sem_perm);
Nadia Derbey023a5352007-10-18 23:40:51 -0700494}
495
Pierre Peiffer6ff37972008-04-29 01:00:46 -0700496static inline void sem_lock_and_putref(struct sem_array *sma)
497{
Rik van Riel6062a8d2013-04-30 19:15:44 -0700498 sem_lock(sma, NULL, -1);
Manfred Sprauldba4cdd2017-07-12 14:34:41 -0700499 ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
Pierre Peiffer6ff37972008-04-29 01:00:46 -0700500}
501
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700502static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
503{
504 ipc_rmid(&sem_ids(ns), &s->sem_perm);
505}
506
Kees Cook101ede02017-07-12 14:35:02 -0700507static struct sem_array *sem_alloc(size_t nsems)
508{
509 struct sem_array *sma;
Kees Cook101ede02017-07-12 14:35:02 -0700510
511 if (nsems > (INT_MAX - sizeof(*sma)) / sizeof(sma->sems[0]))
512 return NULL;
513
Gustavo A. R. Silva4a2ae922019-03-07 16:30:26 -0800514 sma = kvzalloc(struct_size(sma, sems, nsems), GFP_KERNEL);
Kees Cook101ede02017-07-12 14:35:02 -0700515 if (unlikely(!sma))
516 return NULL;
517
Kees Cook101ede02017-07-12 14:35:02 -0700518 return sma;
519}
520
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700521/**
522 * newary - Create a new semaphore set
523 * @ns: namespace
524 * @params: ptr to the structure that contains key, semflg and nsems
525 *
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700526 * Called with sem_ids.rwsem held (as a writer)
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700527 */
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700528static int newary(struct ipc_namespace *ns, struct ipc_params *params)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 int retval;
531 struct sem_array *sma;
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700532 key_t key = params->key;
533 int nsems = params->u.nsems;
534 int semflg = params->flg;
Manfred Spraulb97e8202009-12-15 16:47:32 -0800535 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536
537 if (!nsems)
538 return -EINVAL;
Kirill Korotaeve3893532006-10-02 02:18:22 -0700539 if (ns->used_sems + nsems > ns->sc_semmns)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 return -ENOSPC;
541
Kees Cook101ede02017-07-12 14:35:02 -0700542 sma = sem_alloc(nsems);
Davidlohr Bueso3ab08fe2014-01-27 17:07:06 -0800543 if (!sma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544 return -ENOMEM;
Davidlohr Bueso3ab08fe2014-01-27 17:07:06 -0800545
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 sma->sem_perm.mode = (semflg & S_IRWXUGO);
547 sma->sem_perm.key = key;
548
549 sma->sem_perm.security = NULL;
Eric W. Biedermanaefad952018-03-22 20:52:43 -0500550 retval = security_sem_alloc(&sma->sem_perm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 if (retval) {
Kees Cooke2029df2017-07-12 14:35:31 -0700552 kvfree(sma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 return retval;
554 }
555
Rik van Riel6062a8d2013-04-30 19:15:44 -0700556 for (i = 0; i < nsems; i++) {
Manfred Spraul1a233952017-07-12 14:34:38 -0700557 INIT_LIST_HEAD(&sma->sems[i].pending_alter);
558 INIT_LIST_HEAD(&sma->sems[i].pending_const);
559 spin_lock_init(&sma->sems[i].lock);
Rik van Riel6062a8d2013-04-30 19:15:44 -0700560 }
Manfred Spraulb97e8202009-12-15 16:47:32 -0800561
562 sma->complex_count = 0;
Manfred Spraul9de5ab82017-02-27 14:28:18 -0800563 sma->use_global_lock = USE_GLOBAL_LOCK_HYSTERESIS;
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700564 INIT_LIST_HEAD(&sma->pending_alter);
565 INIT_LIST_HEAD(&sma->pending_const);
Manfred Spraul4daa28f2008-07-25 01:48:04 -0700566 INIT_LIST_HEAD(&sma->list_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 sma->sem_nsems = nsems;
Deepa Dinamanie54d02b2017-08-02 19:51:13 -0700568 sma->sem_ctime = ktime_get_real_seconds();
Manfred Spraule8577d12014-12-02 15:59:34 -0800569
Davidlohr Bueso39c96a12017-11-17 15:31:11 -0800570 /* ipc_addid() locks sma upon success. */
Manfred Spraul2ec55f82017-07-12 14:35:13 -0700571 retval = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
572 if (retval < 0) {
Manfred Spraul39cfffd2018-08-21 22:01:29 -0700573 ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
Manfred Spraul2ec55f82017-07-12 14:35:13 -0700574 return retval;
Manfred Spraule8577d12014-12-02 15:59:34 -0800575 }
576 ns->used_sems += nsems;
577
Rik van Riel6062a8d2013-04-30 19:15:44 -0700578 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -0700579 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700581 return sma->sem_perm.id;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582}
583
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700584
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700585/*
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700586 * Called with sem_ids.rwsem and ipcp locked.
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700587 */
Nadia Derbey03f02c72007-10-18 23:40:51 -0700588static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
589 struct ipc_params *params)
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700590{
Nadia Derbey03f02c72007-10-18 23:40:51 -0700591 struct sem_array *sma;
592
593 sma = container_of(ipcp, struct sem_array, sem_perm);
594 if (params->u.nsems > sma->sem_nsems)
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700595 return -EINVAL;
596
597 return 0;
598}
599
Dominik Brodowski69894712018-03-20 19:53:58 +0100600long ksys_semget(key_t key, int nsems, int semflg)
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700601{
Kirill Korotaeve3893532006-10-02 02:18:22 -0700602 struct ipc_namespace *ns;
Mathias Krauseeb66ec42014-06-06 14:37:36 -0700603 static const struct ipc_ops sem_ops = {
604 .getnew = newary,
Eric W. Biederman50ab44b2018-03-23 23:41:55 -0500605 .associate = security_sem_associate,
Mathias Krauseeb66ec42014-06-06 14:37:36 -0700606 .more_checks = sem_more_checks,
607 };
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700608 struct ipc_params sem_params;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609
Kirill Korotaeve3893532006-10-02 02:18:22 -0700610 ns = current->nsproxy->ipc_ns;
611
612 if (nsems < 0 || nsems > ns->sc_semmsl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 return -EINVAL;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700614
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700615 sem_params.key = key;
616 sem_params.flg = semflg;
617 sem_params.u.nsems = nsems;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700618
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700619 return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620}
621
Dominik Brodowski69894712018-03-20 19:53:58 +0100622SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg)
623{
624 return ksys_semget(key, nsems, semflg);
625}
626
Petr Mladek78f50092014-01-27 17:07:00 -0800627/**
Davidlohr Bueso4ce33ec2016-12-14 15:06:37 -0800628 * perform_atomic_semop[_slow] - Attempt to perform semaphore
629 * operations on a given array.
Manfred Spraul758a6ba32013-07-08 16:01:26 -0700630 * @sma: semaphore array
Manfred Sprauld198cd62014-06-06 14:37:49 -0700631 * @q: struct sem_queue that describes the operation
Manfred Spraul758a6ba32013-07-08 16:01:26 -0700632 *
Davidlohr Bueso4ce33ec2016-12-14 15:06:37 -0800633 * Caller blocking are as follows, based the value
634 * indicated by the semaphore operation (sem_op):
635 *
636 * (1) >0 never blocks.
637 * (2) 0 (wait-for-zero operation): semval is non-zero.
638 * (3) <0 attempting to decrement semval to a value smaller than zero.
639 *
Manfred Spraul758a6ba32013-07-08 16:01:26 -0700640 * Returns 0 if the operation was possible.
641 * Returns 1 if the operation is impossible, the caller must sleep.
Davidlohr Bueso4ce33ec2016-12-14 15:06:37 -0800642 * Returns <0 for error codes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 */
Davidlohr Bueso4ce33ec2016-12-14 15:06:37 -0800644static int perform_atomic_semop_slow(struct sem_array *sma, struct sem_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645{
Eric W. Biederman51d6f262018-03-23 01:11:29 -0500646 int result, sem_op, nsops;
647 struct pid *pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 struct sembuf *sop;
Manfred Spraul239521f2014-01-27 17:07:04 -0800649 struct sem *curr;
Manfred Sprauld198cd62014-06-06 14:37:49 -0700650 struct sembuf *sops;
651 struct sem_undo *un;
652
653 sops = q->sops;
654 nsops = q->nsops;
655 un = q->undo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656
657 for (sop = sops; sop < sops + nsops; sop++) {
Davidlohr Buesoec67aaa2018-06-14 15:27:51 -0700658 int idx = array_index_nospec(sop->sem_num, sma->sem_nsems);
659 curr = &sma->sems[idx];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 sem_op = sop->sem_op;
661 result = curr->semval;
Petr Mladek78f50092014-01-27 17:07:00 -0800662
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 if (!sem_op && result)
664 goto would_block;
665
666 result += sem_op;
667 if (result < 0)
668 goto would_block;
669 if (result > SEMVMX)
670 goto out_of_range;
Petr Mladek78f50092014-01-27 17:07:00 -0800671
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 if (sop->sem_flg & SEM_UNDO) {
673 int undo = un->semadj[sop->sem_num] - sem_op;
Petr Mladek78f50092014-01-27 17:07:00 -0800674 /* Exceeding the undo range is an error. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
676 goto out_of_range;
Petr Mladek78f50092014-01-27 17:07:00 -0800677 un->semadj[sop->sem_num] = undo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 }
Petr Mladek78f50092014-01-27 17:07:00 -0800679
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680 curr->semval = result;
681 }
682
683 sop--;
Manfred Sprauld198cd62014-06-06 14:37:49 -0700684 pid = q->pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 while (sop >= sops) {
Eric W. Biederman51d6f262018-03-23 01:11:29 -0500686 ipc_update_pid(&sma->sems[sop->sem_num].sempid, pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 sop--;
688 }
Petr Mladek78f50092014-01-27 17:07:00 -0800689
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 return 0;
691
692out_of_range:
693 result = -ERANGE;
694 goto undo;
695
696would_block:
Manfred Sprauled247b72014-06-06 14:37:49 -0700697 q->blocking = sop;
698
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 if (sop->sem_flg & IPC_NOWAIT)
700 result = -EAGAIN;
701 else
702 result = 1;
703
704undo:
705 sop--;
706 while (sop >= sops) {
Petr Mladek78f50092014-01-27 17:07:00 -0800707 sem_op = sop->sem_op;
Manfred Spraul1a233952017-07-12 14:34:38 -0700708 sma->sems[sop->sem_num].semval -= sem_op;
Petr Mladek78f50092014-01-27 17:07:00 -0800709 if (sop->sem_flg & SEM_UNDO)
710 un->semadj[sop->sem_num] += sem_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 sop--;
712 }
713
714 return result;
715}
716
Davidlohr Bueso4ce33ec2016-12-14 15:06:37 -0800717static int perform_atomic_semop(struct sem_array *sma, struct sem_queue *q)
718{
719 int result, sem_op, nsops;
720 struct sembuf *sop;
721 struct sem *curr;
722 struct sembuf *sops;
723 struct sem_undo *un;
724
725 sops = q->sops;
726 nsops = q->nsops;
727 un = q->undo;
728
729 if (unlikely(q->dupsop))
730 return perform_atomic_semop_slow(sma, q);
731
732 /*
733 * We scan the semaphore set twice, first to ensure that the entire
734 * operation can succeed, therefore avoiding any pointless writes
735 * to shared memory and having to undo such changes in order to block
736 * until the operations can go through.
737 */
738 for (sop = sops; sop < sops + nsops; sop++) {
Davidlohr Buesoec67aaa2018-06-14 15:27:51 -0700739 int idx = array_index_nospec(sop->sem_num, sma->sem_nsems);
740
741 curr = &sma->sems[idx];
Davidlohr Bueso4ce33ec2016-12-14 15:06:37 -0800742 sem_op = sop->sem_op;
743 result = curr->semval;
744
745 if (!sem_op && result)
746 goto would_block; /* wait-for-zero */
747
748 result += sem_op;
749 if (result < 0)
750 goto would_block;
751
752 if (result > SEMVMX)
753 return -ERANGE;
754
755 if (sop->sem_flg & SEM_UNDO) {
756 int undo = un->semadj[sop->sem_num] - sem_op;
757
758 /* Exceeding the undo range is an error. */
759 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
760 return -ERANGE;
761 }
762 }
763
764 for (sop = sops; sop < sops + nsops; sop++) {
Manfred Spraul1a233952017-07-12 14:34:38 -0700765 curr = &sma->sems[sop->sem_num];
Davidlohr Bueso4ce33ec2016-12-14 15:06:37 -0800766 sem_op = sop->sem_op;
767 result = curr->semval;
768
769 if (sop->sem_flg & SEM_UNDO) {
770 int undo = un->semadj[sop->sem_num] - sem_op;
771
772 un->semadj[sop->sem_num] = undo;
773 }
774 curr->semval += sem_op;
Eric W. Biederman51d6f262018-03-23 01:11:29 -0500775 ipc_update_pid(&curr->sempid, q->pid);
Davidlohr Bueso4ce33ec2016-12-14 15:06:37 -0800776 }
777
778 return 0;
779
780would_block:
781 q->blocking = sop;
782 return sop->sem_flg & IPC_NOWAIT ? -EAGAIN : 1;
783}
784
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -0800785static inline void wake_up_sem_queue_prepare(struct sem_queue *q, int error,
786 struct wake_q_head *wake_q)
Nick Piggind4212092009-12-15 16:47:30 -0800787{
Manfred Spraul8116b542020-02-03 17:34:42 -0800788 get_task_struct(q->sleeper);
789
790 /* see SEM_BARRIER_2 for purpuse/pairing */
791 smp_store_release(&q->status, error);
792
793 wake_q_add_safe(wake_q, q->sleeper);
Nick Piggind4212092009-12-15 16:47:30 -0800794}
795
Manfred Spraulb97e8202009-12-15 16:47:32 -0800796static void unlink_queue(struct sem_array *sma, struct sem_queue *q)
797{
798 list_del(&q->list);
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700799 if (q->nsops > 1)
Manfred Spraulb97e8202009-12-15 16:47:32 -0800800 sma->complex_count--;
801}
802
Manfred Spraulfd5db422010-05-26 14:43:40 -0700803/** check_restart(sma, q)
804 * @sma: semaphore array
805 * @q: the operation that just completed
806 *
807 * update_queue is O(N^2) when it restarts scanning the whole queue of
808 * waiting operations. Therefore this function checks if the restart is
809 * really necessary. It is called after a previously waiting operation
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700810 * modified the array.
811 * Note that wait-for-zero operations are handled without restart.
Manfred Spraulfd5db422010-05-26 14:43:40 -0700812 */
Davidlohr Bueso4663d3e2016-12-14 15:06:40 -0800813static inline int check_restart(struct sem_array *sma, struct sem_queue *q)
Manfred Spraulfd5db422010-05-26 14:43:40 -0700814{
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700815 /* pending complex alter operations are too difficult to analyse */
816 if (!list_empty(&sma->pending_alter))
Manfred Spraulfd5db422010-05-26 14:43:40 -0700817 return 1;
818
819 /* we were a sleeping complex operation. Too difficult */
820 if (q->nsops > 1)
821 return 1;
822
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700823 /* It is impossible that someone waits for the new value:
824 * - complex operations always restart.
825 * - wait-for-zero are handled seperately.
826 * - q is a previously sleeping simple operation that
827 * altered the array. It must be a decrement, because
828 * simple increments never sleep.
829 * - If there are older (higher priority) decrements
830 * in the queue, then they have observed the original
831 * semval value and couldn't proceed. The operation
832 * decremented to value - thus they won't proceed either.
833 */
834 return 0;
835}
Manfred Spraulfd5db422010-05-26 14:43:40 -0700836
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700837/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800838 * wake_const_ops - wake up non-alter tasks
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700839 * @sma: semaphore array.
840 * @semnum: semaphore that was modified.
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -0800841 * @wake_q: lockless wake-queue head.
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700842 *
843 * wake_const_ops must be called after a semaphore in a semaphore array
844 * was set to 0. If complex const operations are pending, wake_const_ops must
845 * be called with semnum = -1, as well as with the number of each modified
846 * semaphore.
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -0800847 * The tasks that must be woken up are added to @wake_q. The return code
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700848 * is stored in q->pid.
849 * The function returns 1 if at least one operation was completed successfully.
850 */
851static int wake_const_ops(struct sem_array *sma, int semnum,
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -0800852 struct wake_q_head *wake_q)
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700853{
Davidlohr Buesof150f022016-12-14 15:06:43 -0800854 struct sem_queue *q, *tmp;
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700855 struct list_head *pending_list;
856 int semop_completed = 0;
Manfred Spraulfd5db422010-05-26 14:43:40 -0700857
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700858 if (semnum == -1)
859 pending_list = &sma->pending_const;
860 else
Manfred Spraul1a233952017-07-12 14:34:38 -0700861 pending_list = &sma->sems[semnum].pending_const;
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700862
Davidlohr Buesof150f022016-12-14 15:06:43 -0800863 list_for_each_entry_safe(q, tmp, pending_list, list) {
864 int error = perform_atomic_semop(sma, q);
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700865
Davidlohr Buesof150f022016-12-14 15:06:43 -0800866 if (error > 0)
867 continue;
868 /* operation completed, remove from queue & wakeup */
869 unlink_queue(sma, q);
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700870
Davidlohr Buesof150f022016-12-14 15:06:43 -0800871 wake_up_sem_queue_prepare(q, error, wake_q);
872 if (error == 0)
873 semop_completed = 1;
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700874 }
Davidlohr Buesof150f022016-12-14 15:06:43 -0800875
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700876 return semop_completed;
877}
878
879/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800880 * do_smart_wakeup_zero - wakeup all wait for zero tasks
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700881 * @sma: semaphore array
882 * @sops: operations that were performed
883 * @nsops: number of operations
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -0800884 * @wake_q: lockless wake-queue head
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700885 *
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800886 * Checks all required queue for wait-for-zero operations, based
887 * on the actual changes that were performed on the semaphore array.
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700888 * The function returns 1 if at least one operation was completed successfully.
889 */
890static int do_smart_wakeup_zero(struct sem_array *sma, struct sembuf *sops,
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -0800891 int nsops, struct wake_q_head *wake_q)
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700892{
893 int i;
894 int semop_completed = 0;
895 int got_zero = 0;
896
897 /* first: the per-semaphore queues, if known */
898 if (sops) {
899 for (i = 0; i < nsops; i++) {
900 int num = sops[i].sem_num;
901
Manfred Spraul1a233952017-07-12 14:34:38 -0700902 if (sma->sems[num].semval == 0) {
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700903 got_zero = 1;
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -0800904 semop_completed |= wake_const_ops(sma, num, wake_q);
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700905 }
906 }
907 } else {
908 /*
909 * No sops means modified semaphores not known.
910 * Assume all were changed.
Manfred Spraulfd5db422010-05-26 14:43:40 -0700911 */
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700912 for (i = 0; i < sma->sem_nsems; i++) {
Manfred Spraul1a233952017-07-12 14:34:38 -0700913 if (sma->sems[i].semval == 0) {
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700914 got_zero = 1;
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -0800915 semop_completed |= wake_const_ops(sma, i, wake_q);
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700916 }
917 }
Manfred Spraulfd5db422010-05-26 14:43:40 -0700918 }
919 /*
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700920 * If one of the modified semaphores got 0,
921 * then check the global queue, too.
Manfred Spraulfd5db422010-05-26 14:43:40 -0700922 */
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700923 if (got_zero)
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -0800924 semop_completed |= wake_const_ops(sma, -1, wake_q);
Manfred Spraulfd5db422010-05-26 14:43:40 -0700925
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700926 return semop_completed;
Manfred Spraulfd5db422010-05-26 14:43:40 -0700927}
928
Manfred Spraul636c6be2009-12-15 16:47:33 -0800929
930/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800931 * update_queue - look for tasks that can be completed.
Manfred Spraul636c6be2009-12-15 16:47:33 -0800932 * @sma: semaphore array.
933 * @semnum: semaphore that was modified.
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -0800934 * @wake_q: lockless wake-queue head.
Manfred Spraul636c6be2009-12-15 16:47:33 -0800935 *
936 * update_queue must be called after a semaphore in a semaphore array
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700937 * was modified. If multiple semaphores were modified, update_queue must
938 * be called with semnum = -1, as well as with the number of each modified
939 * semaphore.
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -0800940 * The tasks that must be woken up are added to @wake_q. The return code
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700941 * is stored in q->pid.
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700942 * The function internally checks if const operations can now succeed.
943 *
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700944 * The function return 1 if at least one semop was completed successfully.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 */
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -0800946static int update_queue(struct sem_array *sma, int semnum, struct wake_q_head *wake_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947{
Davidlohr Buesof150f022016-12-14 15:06:43 -0800948 struct sem_queue *q, *tmp;
Manfred Spraul636c6be2009-12-15 16:47:33 -0800949 struct list_head *pending_list;
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700950 int semop_completed = 0;
Manfred Spraul636c6be2009-12-15 16:47:33 -0800951
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700952 if (semnum == -1)
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700953 pending_list = &sma->pending_alter;
Rik van Riel9f1bc2c92013-04-30 19:15:39 -0700954 else
Manfred Spraul1a233952017-07-12 14:34:38 -0700955 pending_list = &sma->sems[semnum].pending_alter;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956
Nick Piggin9cad2002009-12-15 16:47:29 -0800957again:
Davidlohr Buesof150f022016-12-14 15:06:43 -0800958 list_for_each_entry_safe(q, tmp, pending_list, list) {
Manfred Spraulfd5db422010-05-26 14:43:40 -0700959 int error, restart;
Manfred Spraul636c6be2009-12-15 16:47:33 -0800960
Manfred Sprauld987f8b22009-12-15 16:47:34 -0800961 /* If we are scanning the single sop, per-semaphore list of
962 * one semaphore and that semaphore is 0, then it is not
Manfred Spraul1a82e9e2013-07-08 16:01:23 -0700963 * necessary to scan further: simple increments
Manfred Sprauld987f8b22009-12-15 16:47:34 -0800964 * that affect only one entry succeed immediately and cannot
965 * be in the per semaphore pending queue, and decrements
966 * cannot be successful if the value is already 0.
967 */
Manfred Spraul1a233952017-07-12 14:34:38 -0700968 if (semnum != -1 && sma->sems[semnum].semval == 0)
Manfred Sprauld987f8b22009-12-15 16:47:34 -0800969 break;
970
Manfred Sprauld198cd62014-06-06 14:37:49 -0700971 error = perform_atomic_semop(sma, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972
973 /* Does q->sleeper still need to sleep? */
Nick Piggin9cad2002009-12-15 16:47:29 -0800974 if (error > 0)
975 continue;
Manfred Spraula1193f82008-07-25 01:48:06 -0700976
Manfred Spraulb97e8202009-12-15 16:47:32 -0800977 unlink_queue(sma, q);
Manfred Spraula1193f82008-07-25 01:48:06 -0700978
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700979 if (error) {
Manfred Spraulfd5db422010-05-26 14:43:40 -0700980 restart = 0;
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700981 } else {
982 semop_completed = 1;
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -0800983 do_smart_wakeup_zero(sma, q->sops, q->nsops, wake_q);
Manfred Spraulfd5db422010-05-26 14:43:40 -0700984 restart = check_restart(sma, q);
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700985 }
Manfred Spraulfd5db422010-05-26 14:43:40 -0700986
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -0800987 wake_up_sem_queue_prepare(q, error, wake_q);
Manfred Spraulfd5db422010-05-26 14:43:40 -0700988 if (restart)
Nick Piggin9cad2002009-12-15 16:47:29 -0800989 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 }
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700991 return semop_completed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992}
993
Manfred Spraul0a2b9d42010-05-26 14:43:41 -0700994/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -0800995 * set_semotime - set sem_otime
Manfred Spraul0e8c6652013-09-30 13:45:25 -0700996 * @sma: semaphore array
997 * @sops: operations that modified the array, may be NULL
998 *
999 * sem_otime is replicated to avoid cache line trashing.
1000 * This function sets one instance to the current time.
1001 */
1002static void set_semotime(struct sem_array *sma, struct sembuf *sops)
1003{
1004 if (sops == NULL) {
Arnd Bergmann2a70b782018-04-12 15:19:44 +02001005 sma->sems[0].sem_otime = ktime_get_real_seconds();
Manfred Spraul0e8c6652013-09-30 13:45:25 -07001006 } else {
Manfred Spraul1a233952017-07-12 14:34:38 -07001007 sma->sems[sops[0].sem_num].sem_otime =
Arnd Bergmann2a70b782018-04-12 15:19:44 +02001008 ktime_get_real_seconds();
Manfred Spraul0e8c6652013-09-30 13:45:25 -07001009 }
1010}
1011
1012/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -08001013 * do_smart_update - optimized update_queue
Manfred Spraulfd5db422010-05-26 14:43:40 -07001014 * @sma: semaphore array
1015 * @sops: operations that were performed
1016 * @nsops: number of operations
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001017 * @otime: force setting otime
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08001018 * @wake_q: lockless wake-queue head
Manfred Spraulfd5db422010-05-26 14:43:40 -07001019 *
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001020 * do_smart_update() does the required calls to update_queue and wakeup_zero,
1021 * based on the actual changes that were performed on the semaphore array.
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001022 * Note that the function does not do the actual wake-up: the caller is
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08001023 * responsible for calling wake_up_q().
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001024 * It is safe to perform this call after dropping all locks.
Manfred Spraulfd5db422010-05-26 14:43:40 -07001025 */
Manfred Spraul0a2b9d42010-05-26 14:43:41 -07001026static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops,
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08001027 int otime, struct wake_q_head *wake_q)
Manfred Spraulfd5db422010-05-26 14:43:40 -07001028{
1029 int i;
1030
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08001031 otime |= do_smart_wakeup_zero(sma, sops, nsops, wake_q);
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001032
Manfred Spraulf269f402013-07-08 16:01:24 -07001033 if (!list_empty(&sma->pending_alter)) {
1034 /* semaphore array uses the global queue - just process it. */
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08001035 otime |= update_queue(sma, -1, wake_q);
Manfred Spraulf269f402013-07-08 16:01:24 -07001036 } else {
1037 if (!sops) {
1038 /*
1039 * No sops, thus the modified semaphores are not
1040 * known. Check all.
1041 */
1042 for (i = 0; i < sma->sem_nsems; i++)
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08001043 otime |= update_queue(sma, i, wake_q);
Manfred Spraulf269f402013-07-08 16:01:24 -07001044 } else {
1045 /*
1046 * Check the semaphores that were increased:
1047 * - No complex ops, thus all sleeping ops are
1048 * decrease.
1049 * - if we decreased the value, then any sleeping
1050 * semaphore ops wont be able to run: If the
1051 * previous value was too small, then the new
1052 * value will be too small, too.
1053 */
1054 for (i = 0; i < nsops; i++) {
1055 if (sops[i].sem_op > 0) {
1056 otime |= update_queue(sma,
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08001057 sops[i].sem_num, wake_q);
Manfred Spraulf269f402013-07-08 16:01:24 -07001058 }
Manfred Spraulab465df2013-05-26 11:08:52 +02001059 }
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001060 }
Manfred Spraulfd5db422010-05-26 14:43:40 -07001061 }
Manfred Spraul0e8c6652013-09-30 13:45:25 -07001062 if (otime)
1063 set_semotime(sma, sops);
Manfred Spraulfd5db422010-05-26 14:43:40 -07001064}
1065
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001066/*
Manfred Spraulb220c572014-06-06 14:37:51 -07001067 * check_qop: Test if a queued operation sleeps on the semaphore semnum
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001068 */
1069static int check_qop(struct sem_array *sma, int semnum, struct sem_queue *q,
1070 bool count_zero)
1071{
Manfred Spraulb220c572014-06-06 14:37:51 -07001072 struct sembuf *sop = q->blocking;
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001073
Manfred Spraul9b44ee22014-06-06 14:37:52 -07001074 /*
1075 * Linux always (since 0.99.10) reported a task as sleeping on all
1076 * semaphores. This violates SUS, therefore it was changed to the
1077 * standard compliant behavior.
1078 * Give the administrators a chance to notice that an application
1079 * might misbehave because it relies on the Linux behavior.
1080 */
1081 pr_info_once("semctl(GETNCNT/GETZCNT) is since 3.16 Single Unix Specification compliant.\n"
1082 "The task %s (%d) triggered the difference, watch for misbehavior.\n",
1083 current->comm, task_pid_nr(current));
1084
Manfred Spraulb220c572014-06-06 14:37:51 -07001085 if (sop->sem_num != semnum)
1086 return 0;
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001087
Manfred Spraulb220c572014-06-06 14:37:51 -07001088 if (count_zero && sop->sem_op == 0)
1089 return 1;
1090 if (!count_zero && sop->sem_op < 0)
1091 return 1;
1092
1093 return 0;
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001094}
1095
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096/* The following counts are associated to each semaphore:
1097 * semncnt number of tasks waiting on semval being nonzero
1098 * semzcnt number of tasks waiting on semval being zero
Manfred Spraulb220c572014-06-06 14:37:51 -07001099 *
1100 * Per definition, a task waits only on the semaphore of the first semop
1101 * that cannot proceed, even if additional operation would block, too.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 */
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001103static int count_semcnt(struct sem_array *sma, ushort semnum,
1104 bool count_zero)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105{
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001106 struct list_head *l;
Manfred Spraul239521f2014-01-27 17:07:04 -08001107 struct sem_queue *q;
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001108 int semcnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001110 semcnt = 0;
1111 /* First: check the simple operations. They are easy to evaluate */
1112 if (count_zero)
Manfred Spraul1a233952017-07-12 14:34:38 -07001113 l = &sma->sems[semnum].pending_const;
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001114 else
Manfred Spraul1a233952017-07-12 14:34:38 -07001115 l = &sma->sems[semnum].pending_alter;
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001116
1117 list_for_each_entry(q, l, list) {
1118 /* all task on a per-semaphore list sleep on exactly
1119 * that semaphore
1120 */
1121 semcnt++;
Rik van Rielde2657f2013-05-09 16:59:59 -04001122 }
1123
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001124 /* Then: check the complex operations. */
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001125 list_for_each_entry(q, &sma->pending_alter, list) {
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001126 semcnt += check_qop(sma, semnum, q, count_zero);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 }
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001128 if (count_zero) {
1129 list_for_each_entry(q, &sma->pending_const, list) {
1130 semcnt += check_qop(sma, semnum, q, count_zero);
1131 }
Rik van Rielebc2e5e2013-05-09 16:53:28 -04001132 }
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001133 return semcnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134}
1135
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001136/* Free a semaphore set. freeary() is called with sem_ids.rwsem locked
1137 * as a writer and the spinlock for this semaphore set hold. sem_ids.rwsem
Nadia Derbey3e148c72007-10-18 23:40:54 -07001138 * remains locked on exit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139 */
Pierre Peiffer01b8b072008-02-08 04:18:57 -08001140static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141{
Manfred Spraul380af1b2008-07-25 01:48:06 -07001142 struct sem_undo *un, *tu;
1143 struct sem_queue *q, *tq;
Pierre Peiffer01b8b072008-02-08 04:18:57 -08001144 struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001145 int i;
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08001146 DEFINE_WAKE_Q(wake_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147
Manfred Spraul380af1b2008-07-25 01:48:06 -07001148 /* Free the existing undo structures for this semaphore set. */
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -07001149 ipc_assert_locked_object(&sma->sem_perm);
Manfred Spraul380af1b2008-07-25 01:48:06 -07001150 list_for_each_entry_safe(un, tu, &sma->list_id, list_id) {
1151 list_del(&un->list_id);
1152 spin_lock(&un->ulp->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 un->semid = -1;
Manfred Spraul380af1b2008-07-25 01:48:06 -07001154 list_del_rcu(&un->list_proc);
1155 spin_unlock(&un->ulp->lock);
Lai Jiangshan693a8b62011-03-18 12:09:35 +08001156 kfree_rcu(un, rcu);
Manfred Spraul380af1b2008-07-25 01:48:06 -07001157 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158
1159 /* Wake up all pending processes and let them fail with EIDRM. */
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001160 list_for_each_entry_safe(q, tq, &sma->pending_const, list) {
1161 unlink_queue(sma, q);
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08001162 wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001163 }
1164
1165 list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
Manfred Spraulb97e8202009-12-15 16:47:32 -08001166 unlink_queue(sma, q);
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08001167 wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 }
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001169 for (i = 0; i < sma->sem_nsems; i++) {
Manfred Spraul1a233952017-07-12 14:34:38 -07001170 struct sem *sem = &sma->sems[i];
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001171 list_for_each_entry_safe(q, tq, &sem->pending_const, list) {
1172 unlink_queue(sma, q);
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08001173 wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07001174 }
1175 list_for_each_entry_safe(q, tq, &sem->pending_alter, list) {
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001176 unlink_queue(sma, q);
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08001177 wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001178 }
Eric W. Biederman51d6f262018-03-23 01:11:29 -05001179 ipc_update_pid(&sem->sempid, NULL);
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07001180 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181
Nadia Derbey7ca7e562007-10-18 23:40:48 -07001182 /* Remove the semaphore set from the IDR */
1183 sem_rmid(ns, sma);
Rik van Riel6062a8d2013-04-30 19:15:44 -07001184 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001185 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08001187 wake_up_q(&wake_q);
Kirill Korotaeve3893532006-10-02 02:18:22 -07001188 ns->used_sems -= sma->sem_nsems;
Manfred Sprauldba4cdd2017-07-12 14:34:41 -07001189 ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190}
1191
1192static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
1193{
Manfred Spraul239521f2014-01-27 17:07:04 -08001194 switch (version) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 case IPC_64:
1196 return copy_to_user(buf, in, sizeof(*in));
1197 case IPC_OLD:
1198 {
1199 struct semid_ds out;
1200
Dan Rosenberg982f7c22010-09-30 15:15:31 -07001201 memset(&out, 0, sizeof(out));
1202
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
1204
1205 out.sem_otime = in->sem_otime;
1206 out.sem_ctime = in->sem_ctime;
1207 out.sem_nsems = in->sem_nsems;
1208
1209 return copy_to_user(buf, &out, sizeof(out));
1210 }
1211 default:
1212 return -EINVAL;
1213 }
1214}
1215
Deepa Dinamanie54d02b2017-08-02 19:51:13 -07001216static time64_t get_semotime(struct sem_array *sma)
Manfred Sprauld12e1e52013-07-08 16:01:25 -07001217{
1218 int i;
Deepa Dinamanie54d02b2017-08-02 19:51:13 -07001219 time64_t res;
Manfred Sprauld12e1e52013-07-08 16:01:25 -07001220
Manfred Spraul1a233952017-07-12 14:34:38 -07001221 res = sma->sems[0].sem_otime;
Manfred Sprauld12e1e52013-07-08 16:01:25 -07001222 for (i = 1; i < sma->sem_nsems; i++) {
Deepa Dinamanie54d02b2017-08-02 19:51:13 -07001223 time64_t to = sma->sems[i].sem_otime;
Manfred Sprauld12e1e52013-07-08 16:01:25 -07001224
1225 if (to > res)
1226 res = to;
1227 }
1228 return res;
1229}
1230
Al Viro45a4a642017-07-09 09:11:00 -04001231static int semctl_stat(struct ipc_namespace *ns, int semid,
1232 int cmd, struct semid64_ds *semid64)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 struct sem_array *sma;
Arnd Bergmannc2ab9752015-04-28 21:39:50 +02001235 time64_t semotime;
Al Viro45a4a642017-07-09 09:11:00 -04001236 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237
Al Viro45a4a642017-07-09 09:11:00 -04001238 memset(semid64, 0, sizeof(*semid64));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239
Al Viro45a4a642017-07-09 09:11:00 -04001240 rcu_read_lock();
Davidlohr Buesoa280d6d2018-04-10 16:35:26 -07001241 if (cmd == SEM_STAT || cmd == SEM_STAT_ANY) {
Al Viro45a4a642017-07-09 09:11:00 -04001242 sma = sem_obtain_object(ns, semid);
1243 if (IS_ERR(sma)) {
1244 err = PTR_ERR(sma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 goto out_unlock;
Al Viro45a4a642017-07-09 09:11:00 -04001246 }
Davidlohr Buesoa280d6d2018-04-10 16:35:26 -07001247 } else { /* IPC_STAT */
Al Viro45a4a642017-07-09 09:11:00 -04001248 sma = sem_obtain_object_check(ns, semid);
1249 if (IS_ERR(sma)) {
1250 err = PTR_ERR(sma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251 goto out_unlock;
Al Viro45a4a642017-07-09 09:11:00 -04001252 }
1253 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254
Davidlohr Buesoa280d6d2018-04-10 16:35:26 -07001255 /* see comment for SHM_STAT_ANY */
1256 if (cmd == SEM_STAT_ANY)
1257 audit_ipc_obj(&sma->sem_perm);
1258 else {
1259 err = -EACCES;
1260 if (ipcperms(ns, &sma->sem_perm, S_IRUGO))
1261 goto out_unlock;
1262 }
Al Viro45a4a642017-07-09 09:11:00 -04001263
Eric W. Biedermanaefad952018-03-22 20:52:43 -05001264 err = security_sem_semctl(&sma->sem_perm, cmd);
Al Viro45a4a642017-07-09 09:11:00 -04001265 if (err)
1266 goto out_unlock;
1267
Philippe Mikoyan87ad4b02018-02-06 15:40:49 -08001268 ipc_lock_object(&sma->sem_perm);
1269
1270 if (!ipc_valid_object(&sma->sem_perm)) {
1271 ipc_unlock_object(&sma->sem_perm);
1272 err = -EIDRM;
1273 goto out_unlock;
1274 }
1275
Al Viro45a4a642017-07-09 09:11:00 -04001276 kernel_to_ipc64_perm(&sma->sem_perm, &semid64->sem_perm);
Arnd Bergmannc2ab9752015-04-28 21:39:50 +02001277 semotime = get_semotime(sma);
1278 semid64->sem_otime = semotime;
Al Viro45a4a642017-07-09 09:11:00 -04001279 semid64->sem_ctime = sma->sem_ctime;
Arnd Bergmannc2ab9752015-04-28 21:39:50 +02001280#ifndef CONFIG_64BIT
1281 semid64->sem_otime_high = semotime >> 32;
1282 semid64->sem_ctime_high = sma->sem_ctime >> 32;
1283#endif
Al Viro45a4a642017-07-09 09:11:00 -04001284 semid64->sem_nsems = sma->sem_nsems;
Philippe Mikoyan87ad4b02018-02-06 15:40:49 -08001285
Manfred Spraul615c9992018-08-21 22:01:21 -07001286 if (cmd == IPC_STAT) {
1287 /*
1288 * As defined in SUS:
1289 * Return 0 on success
1290 */
1291 err = 0;
1292 } else {
1293 /*
1294 * SEM_STAT and SEM_STAT_ANY (both Linux specific)
1295 * Return the full id, including the sequence number
1296 */
1297 err = sma->sem_perm.id;
1298 }
Philippe Mikoyan87ad4b02018-02-06 15:40:49 -08001299 ipc_unlock_object(&sma->sem_perm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300out_unlock:
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001301 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302 return err;
1303}
1304
Al Viro45a4a642017-07-09 09:11:00 -04001305static int semctl_info(struct ipc_namespace *ns, int semid,
1306 int cmd, void __user *p)
1307{
1308 struct seminfo seminfo;
Manfred Spraul27c331a2018-08-21 22:02:00 -07001309 int max_idx;
Al Viro45a4a642017-07-09 09:11:00 -04001310 int err;
1311
1312 err = security_sem_semctl(NULL, cmd);
1313 if (err)
1314 return err;
1315
1316 memset(&seminfo, 0, sizeof(seminfo));
1317 seminfo.semmni = ns->sc_semmni;
1318 seminfo.semmns = ns->sc_semmns;
1319 seminfo.semmsl = ns->sc_semmsl;
1320 seminfo.semopm = ns->sc_semopm;
1321 seminfo.semvmx = SEMVMX;
1322 seminfo.semmnu = SEMMNU;
1323 seminfo.semmap = SEMMAP;
1324 seminfo.semume = SEMUME;
1325 down_read(&sem_ids(ns).rwsem);
1326 if (cmd == SEM_INFO) {
1327 seminfo.semusz = sem_ids(ns).in_use;
1328 seminfo.semaem = ns->used_sems;
1329 } else {
1330 seminfo.semusz = SEMUSZ;
1331 seminfo.semaem = SEMAEM;
1332 }
Manfred Spraul27c331a2018-08-21 22:02:00 -07001333 max_idx = ipc_get_maxidx(&sem_ids(ns));
Al Viro45a4a642017-07-09 09:11:00 -04001334 up_read(&sem_ids(ns).rwsem);
1335 if (copy_to_user(p, &seminfo, sizeof(struct seminfo)))
1336 return -EFAULT;
Manfred Spraul27c331a2018-08-21 22:02:00 -07001337 return (max_idx < 0) ? 0 : max_idx;
Al Viro45a4a642017-07-09 09:11:00 -04001338}
1339
Al Viroe1fd1f42013-03-05 15:04:55 -05001340static int semctl_setval(struct ipc_namespace *ns, int semid, int semnum,
Al Viro45a4a642017-07-09 09:11:00 -04001341 int val)
Al Viroe1fd1f42013-03-05 15:04:55 -05001342{
1343 struct sem_undo *un;
1344 struct sem_array *sma;
Manfred Spraul239521f2014-01-27 17:07:04 -08001345 struct sem *curr;
Al Viro45a4a642017-07-09 09:11:00 -04001346 int err;
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08001347 DEFINE_WAKE_Q(wake_q);
1348
Rik van Riel6062a8d2013-04-30 19:15:44 -07001349 if (val > SEMVMX || val < 0)
1350 return -ERANGE;
Al Viroe1fd1f42013-03-05 15:04:55 -05001351
Rik van Riel6062a8d2013-04-30 19:15:44 -07001352 rcu_read_lock();
1353 sma = sem_obtain_object_check(ns, semid);
1354 if (IS_ERR(sma)) {
1355 rcu_read_unlock();
1356 return PTR_ERR(sma);
1357 }
1358
1359 if (semnum < 0 || semnum >= sma->sem_nsems) {
1360 rcu_read_unlock();
1361 return -EINVAL;
1362 }
1363
1364
1365 if (ipcperms(ns, &sma->sem_perm, S_IWUGO)) {
1366 rcu_read_unlock();
1367 return -EACCES;
1368 }
Al Viroe1fd1f42013-03-05 15:04:55 -05001369
Eric W. Biedermanaefad952018-03-22 20:52:43 -05001370 err = security_sem_semctl(&sma->sem_perm, SETVAL);
Rik van Riel6062a8d2013-04-30 19:15:44 -07001371 if (err) {
1372 rcu_read_unlock();
1373 return -EACCES;
1374 }
Al Viroe1fd1f42013-03-05 15:04:55 -05001375
Rik van Riel6062a8d2013-04-30 19:15:44 -07001376 sem_lock(sma, NULL, -1);
Al Viroe1fd1f42013-03-05 15:04:55 -05001377
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001378 if (!ipc_valid_object(&sma->sem_perm)) {
Manfred Spraul6e224f92013-10-16 13:46:45 -07001379 sem_unlock(sma, -1);
1380 rcu_read_unlock();
1381 return -EIDRM;
1382 }
1383
Davidlohr Buesoec67aaa2018-06-14 15:27:51 -07001384 semnum = array_index_nospec(semnum, sma->sem_nsems);
Manfred Spraul1a233952017-07-12 14:34:38 -07001385 curr = &sma->sems[semnum];
Al Viroe1fd1f42013-03-05 15:04:55 -05001386
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -07001387 ipc_assert_locked_object(&sma->sem_perm);
Al Viroe1fd1f42013-03-05 15:04:55 -05001388 list_for_each_entry(un, &sma->list_id, list_id)
1389 un->semadj[semnum] = 0;
1390
1391 curr->semval = val;
Eric W. Biederman51d6f262018-03-23 01:11:29 -05001392 ipc_update_pid(&curr->sempid, task_tgid(current));
Deepa Dinamanie54d02b2017-08-02 19:51:13 -07001393 sma->sem_ctime = ktime_get_real_seconds();
Al Viroe1fd1f42013-03-05 15:04:55 -05001394 /* maybe some queued-up processes were waiting for this */
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08001395 do_smart_update(sma, NULL, 0, 0, &wake_q);
Rik van Riel6062a8d2013-04-30 19:15:44 -07001396 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001397 rcu_read_unlock();
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08001398 wake_up_q(&wake_q);
Rik van Riel6062a8d2013-04-30 19:15:44 -07001399 return 0;
Al Viroe1fd1f42013-03-05 15:04:55 -05001400}
1401
Kirill Korotaeve3893532006-10-02 02:18:22 -07001402static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
Al Viroe1fd1f42013-03-05 15:04:55 -05001403 int cmd, void __user *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404{
1405 struct sem_array *sma;
Manfred Spraul239521f2014-01-27 17:07:04 -08001406 struct sem *curr;
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001407 int err, nsems;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 ushort fast_sem_io[SEMMSL_FAST];
Manfred Spraul239521f2014-01-27 17:07:04 -08001409 ushort *sem_io = fast_sem_io;
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08001410 DEFINE_WAKE_Q(wake_q);
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001411
1412 rcu_read_lock();
1413 sma = sem_obtain_object_check(ns, semid);
1414 if (IS_ERR(sma)) {
1415 rcu_read_unlock();
1416 return PTR_ERR(sma);
1417 }
1418
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 nsems = sma->sem_nsems;
1420
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 err = -EACCES;
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001422 if (ipcperms(ns, &sma->sem_perm, cmd == SETALL ? S_IWUGO : S_IRUGO))
1423 goto out_rcu_wakeup;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424
Eric W. Biedermanaefad952018-03-22 20:52:43 -05001425 err = security_sem_semctl(&sma->sem_perm, cmd);
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001426 if (err)
1427 goto out_rcu_wakeup;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428
1429 err = -EACCES;
1430 switch (cmd) {
1431 case GETALL:
1432 {
Al Viroe1fd1f42013-03-05 15:04:55 -05001433 ushort __user *array = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 int i;
1435
Al Viroce857222013-05-03 00:30:49 +01001436 sem_lock(sma, NULL, -1);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001437 if (!ipc_valid_object(&sma->sem_perm)) {
Manfred Spraul6e224f92013-10-16 13:46:45 -07001438 err = -EIDRM;
1439 goto out_unlock;
1440 }
Manfred Spraul239521f2014-01-27 17:07:04 -08001441 if (nsems > SEMMSL_FAST) {
Manfred Sprauldba4cdd2017-07-12 14:34:41 -07001442 if (!ipc_rcu_getref(&sma->sem_perm)) {
Al Viroce857222013-05-03 00:30:49 +01001443 err = -EIDRM;
Manfred Spraul6e224f92013-10-16 13:46:45 -07001444 goto out_unlock;
Al Viroce857222013-05-03 00:30:49 +01001445 }
1446 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001447 rcu_read_unlock();
Kees Cookf8dbe8d2017-07-12 14:34:47 -07001448 sem_io = kvmalloc_array(nsems, sizeof(ushort),
1449 GFP_KERNEL);
Manfred Spraul239521f2014-01-27 17:07:04 -08001450 if (sem_io == NULL) {
Manfred Sprauldba4cdd2017-07-12 14:34:41 -07001451 ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452 return -ENOMEM;
1453 }
1454
Linus Torvalds4091fd92013-05-04 10:13:40 -07001455 rcu_read_lock();
Pierre Peiffer6ff37972008-04-29 01:00:46 -07001456 sem_lock_and_putref(sma);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001457 if (!ipc_valid_object(&sma->sem_perm)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 err = -EIDRM;
Manfred Spraul6e224f92013-10-16 13:46:45 -07001459 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 }
Al Viroce857222013-05-03 00:30:49 +01001461 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 for (i = 0; i < sma->sem_nsems; i++)
Manfred Spraul1a233952017-07-12 14:34:38 -07001463 sem_io[i] = sma->sems[i].semval;
Rik van Riel6062a8d2013-04-30 19:15:44 -07001464 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001465 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466 err = 0;
Manfred Spraul239521f2014-01-27 17:07:04 -08001467 if (copy_to_user(array, sem_io, nsems*sizeof(ushort)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 err = -EFAULT;
1469 goto out_free;
1470 }
1471 case SETALL:
1472 {
1473 int i;
1474 struct sem_undo *un;
1475
Manfred Sprauldba4cdd2017-07-12 14:34:41 -07001476 if (!ipc_rcu_getref(&sma->sem_perm)) {
Manfred Spraul6e224f92013-10-16 13:46:45 -07001477 err = -EIDRM;
1478 goto out_rcu_wakeup;
Rik van Riel6062a8d2013-04-30 19:15:44 -07001479 }
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001480 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481
Manfred Spraul239521f2014-01-27 17:07:04 -08001482 if (nsems > SEMMSL_FAST) {
Kees Cookf8dbe8d2017-07-12 14:34:47 -07001483 sem_io = kvmalloc_array(nsems, sizeof(ushort),
1484 GFP_KERNEL);
Manfred Spraul239521f2014-01-27 17:07:04 -08001485 if (sem_io == NULL) {
Manfred Sprauldba4cdd2017-07-12 14:34:41 -07001486 ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487 return -ENOMEM;
1488 }
1489 }
1490
Manfred Spraul239521f2014-01-27 17:07:04 -08001491 if (copy_from_user(sem_io, p, nsems*sizeof(ushort))) {
Manfred Sprauldba4cdd2017-07-12 14:34:41 -07001492 ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 err = -EFAULT;
1494 goto out_free;
1495 }
1496
1497 for (i = 0; i < nsems; i++) {
1498 if (sem_io[i] > SEMVMX) {
Manfred Sprauldba4cdd2017-07-12 14:34:41 -07001499 ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500 err = -ERANGE;
1501 goto out_free;
1502 }
1503 }
Linus Torvalds4091fd92013-05-04 10:13:40 -07001504 rcu_read_lock();
Pierre Peiffer6ff37972008-04-29 01:00:46 -07001505 sem_lock_and_putref(sma);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001506 if (!ipc_valid_object(&sma->sem_perm)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 err = -EIDRM;
Manfred Spraul6e224f92013-10-16 13:46:45 -07001508 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 }
1510
Davidlohr Buesoa5f4db82016-03-22 14:27:48 -07001511 for (i = 0; i < nsems; i++) {
Manfred Spraul1a233952017-07-12 14:34:38 -07001512 sma->sems[i].semval = sem_io[i];
Eric W. Biederman51d6f262018-03-23 01:11:29 -05001513 ipc_update_pid(&sma->sems[i].sempid, task_tgid(current));
Davidlohr Buesoa5f4db82016-03-22 14:27:48 -07001514 }
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001515
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -07001516 ipc_assert_locked_object(&sma->sem_perm);
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001517 list_for_each_entry(un, &sma->list_id, list_id) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518 for (i = 0; i < nsems; i++)
1519 un->semadj[i] = 0;
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001520 }
Deepa Dinamanie54d02b2017-08-02 19:51:13 -07001521 sma->sem_ctime = ktime_get_real_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522 /* maybe some queued-up processes were waiting for this */
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08001523 do_smart_update(sma, NULL, 0, 0, &wake_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524 err = 0;
1525 goto out_unlock;
1526 }
Al Viroe1fd1f42013-03-05 15:04:55 -05001527 /* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 }
1529 err = -EINVAL;
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001530 if (semnum < 0 || semnum >= nsems)
1531 goto out_rcu_wakeup;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532
Rik van Riel6062a8d2013-04-30 19:15:44 -07001533 sem_lock(sma, NULL, -1);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001534 if (!ipc_valid_object(&sma->sem_perm)) {
Manfred Spraul6e224f92013-10-16 13:46:45 -07001535 err = -EIDRM;
1536 goto out_unlock;
1537 }
Davidlohr Buesoec67aaa2018-06-14 15:27:51 -07001538
1539 semnum = array_index_nospec(semnum, nsems);
Manfred Spraul1a233952017-07-12 14:34:38 -07001540 curr = &sma->sems[semnum];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541
1542 switch (cmd) {
1543 case GETVAL:
1544 err = curr->semval;
1545 goto out_unlock;
1546 case GETPID:
Eric W. Biederman51d6f262018-03-23 01:11:29 -05001547 err = pid_vnr(curr->sempid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548 goto out_unlock;
1549 case GETNCNT:
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001550 err = count_semcnt(sma, semnum, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 goto out_unlock;
1552 case GETZCNT:
Manfred Spraul2f2ed41d2014-06-06 14:37:48 -07001553 err = count_semcnt(sma, semnum, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555 }
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001556
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557out_unlock:
Rik van Riel6062a8d2013-04-30 19:15:44 -07001558 sem_unlock(sma, -1);
Linus Torvaldsc728b9c2013-05-04 11:04:29 -07001559out_rcu_wakeup:
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001560 rcu_read_unlock();
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08001561 wake_up_q(&wake_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562out_free:
Manfred Spraul239521f2014-01-27 17:07:04 -08001563 if (sem_io != fast_sem_io)
Kees Cookf8dbe8d2017-07-12 14:34:47 -07001564 kvfree(sem_io);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565 return err;
1566}
1567
Pierre Peiffer016d7132008-04-29 01:00:50 -07001568static inline unsigned long
1569copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570{
Manfred Spraul239521f2014-01-27 17:07:04 -08001571 switch (version) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572 case IPC_64:
Pierre Peiffer016d7132008-04-29 01:00:50 -07001573 if (copy_from_user(out, buf, sizeof(*out)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576 case IPC_OLD:
1577 {
1578 struct semid_ds tbuf_old;
1579
Manfred Spraul239521f2014-01-27 17:07:04 -08001580 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 return -EFAULT;
1582
Pierre Peiffer016d7132008-04-29 01:00:50 -07001583 out->sem_perm.uid = tbuf_old.sem_perm.uid;
1584 out->sem_perm.gid = tbuf_old.sem_perm.gid;
1585 out->sem_perm.mode = tbuf_old.sem_perm.mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586
1587 return 0;
1588 }
1589 default:
1590 return -EINVAL;
1591 }
1592}
1593
Pierre Peiffer522bb2a2008-04-29 01:00:49 -07001594/*
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001595 * This function handles some semctl commands which require the rwsem
Pierre Peiffer522bb2a2008-04-29 01:00:49 -07001596 * to be held in write mode.
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001597 * NOTE: no locks must be held, the rwsem is taken inside this function.
Pierre Peiffer522bb2a2008-04-29 01:00:49 -07001598 */
Pierre Peiffer21a48262008-04-29 01:00:49 -07001599static int semctl_down(struct ipc_namespace *ns, int semid,
Al Viro45a4a642017-07-09 09:11:00 -04001600 int cmd, struct semid64_ds *semid64)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601{
1602 struct sem_array *sma;
1603 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 struct kern_ipc_perm *ipcp;
1605
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001606 down_write(&sem_ids(ns).rwsem);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001607 rcu_read_lock();
1608
Manfred Spraul4241c1a2018-08-21 22:01:34 -07001609 ipcp = ipcctl_obtain_check(ns, &sem_ids(ns), semid, cmd,
Al Viro45a4a642017-07-09 09:11:00 -04001610 &semid64->sem_perm, 0);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001611 if (IS_ERR(ipcp)) {
1612 err = PTR_ERR(ipcp);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001613 goto out_unlock1;
1614 }
Steve Grubb073115d2006-04-02 17:07:33 -04001615
Pierre Peiffera5f75e72008-04-29 01:00:54 -07001616 sma = container_of(ipcp, struct sem_array, sem_perm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617
Eric W. Biedermanaefad952018-03-22 20:52:43 -05001618 err = security_sem_semctl(&sma->sem_perm, cmd);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001619 if (err)
1620 goto out_unlock1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001622 switch (cmd) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623 case IPC_RMID:
Rik van Riel6062a8d2013-04-30 19:15:44 -07001624 sem_lock(sma, NULL, -1);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001625 /* freeary unlocks the ipc object and rcu */
Pierre Peiffer01b8b072008-02-08 04:18:57 -08001626 freeary(ns, ipcp);
Pierre Peiffer522bb2a2008-04-29 01:00:49 -07001627 goto out_up;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628 case IPC_SET:
Rik van Riel6062a8d2013-04-30 19:15:44 -07001629 sem_lock(sma, NULL, -1);
Al Viro45a4a642017-07-09 09:11:00 -04001630 err = ipc_update_perm(&semid64->sem_perm, ipcp);
Eric W. Biederman1efdb692012-02-07 16:54:11 -08001631 if (err)
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001632 goto out_unlock0;
Deepa Dinamanie54d02b2017-08-02 19:51:13 -07001633 sma->sem_ctime = ktime_get_real_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634 break;
1635 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 err = -EINVAL;
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001637 goto out_unlock1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001640out_unlock0:
Rik van Riel6062a8d2013-04-30 19:15:44 -07001641 sem_unlock(sma, -1);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -07001642out_unlock1:
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001643 rcu_read_unlock();
Pierre Peiffer522bb2a2008-04-29 01:00:49 -07001644out_up:
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001645 up_write(&sem_ids(ns).rwsem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646 return err;
1647}
1648
Arnd Bergmann275f2212018-12-31 22:22:40 +01001649static long ksys_semctl(int semid, int semnum, int cmd, unsigned long arg, int version)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650{
Kirill Korotaeve3893532006-10-02 02:18:22 -07001651 struct ipc_namespace *ns;
Al Viroe1fd1f42013-03-05 15:04:55 -05001652 void __user *p = (void __user *)arg;
Al Viro45a4a642017-07-09 09:11:00 -04001653 struct semid64_ds semid64;
1654 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655
1656 if (semid < 0)
1657 return -EINVAL;
1658
Kirill Korotaeve3893532006-10-02 02:18:22 -07001659 ns = current->nsproxy->ipc_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660
Manfred Spraul239521f2014-01-27 17:07:04 -08001661 switch (cmd) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 case IPC_INFO:
1663 case SEM_INFO:
Al Viro45a4a642017-07-09 09:11:00 -04001664 return semctl_info(ns, semid, cmd, p);
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -08001665 case IPC_STAT:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666 case SEM_STAT:
Davidlohr Buesoa280d6d2018-04-10 16:35:26 -07001667 case SEM_STAT_ANY:
Al Viro45a4a642017-07-09 09:11:00 -04001668 err = semctl_stat(ns, semid, cmd, &semid64);
1669 if (err < 0)
1670 return err;
1671 if (copy_semid_to_user(p, &semid64, version))
1672 err = -EFAULT;
1673 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674 case GETALL:
1675 case GETVAL:
1676 case GETPID:
1677 case GETNCNT:
1678 case GETZCNT:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679 case SETALL:
Al Viroe1fd1f42013-03-05 15:04:55 -05001680 return semctl_main(ns, semid, semnum, cmd, p);
Al Viro45a4a642017-07-09 09:11:00 -04001681 case SETVAL: {
1682 int val;
1683#if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN)
1684 /* big-endian 64bit */
1685 val = arg >> 32;
1686#else
1687 /* 32bit or little-endian 64bit */
1688 val = arg;
1689#endif
1690 return semctl_setval(ns, semid, semnum, val);
1691 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692 case IPC_SET:
Al Viro45a4a642017-07-09 09:11:00 -04001693 if (copy_semid_from_user(&semid64, p, version))
1694 return -EFAULT;
Mathieu Malaterre667da6a2019-03-07 16:30:23 -08001695 /* fall through */
Al Viro45a4a642017-07-09 09:11:00 -04001696 case IPC_RMID:
1697 return semctl_down(ns, semid, cmd, &semid64);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698 default:
1699 return -EINVAL;
1700 }
1701}
1702
Dominik Brodowskid969c6f2018-03-20 20:00:39 +01001703SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, unsigned long, arg)
1704{
Arnd Bergmann275f2212018-12-31 22:22:40 +01001705 return ksys_semctl(semid, semnum, cmd, arg, IPC_64);
Dominik Brodowskid969c6f2018-03-20 20:00:39 +01001706}
1707
Arnd Bergmann275f2212018-12-31 22:22:40 +01001708#ifdef CONFIG_ARCH_WANT_IPC_PARSE_VERSION
1709long ksys_old_semctl(int semid, int semnum, int cmd, unsigned long arg)
1710{
1711 int version = ipc_parse_version(&cmd);
1712
1713 return ksys_semctl(semid, semnum, cmd, arg, version);
1714}
1715
1716SYSCALL_DEFINE4(old_semctl, int, semid, int, semnum, int, cmd, unsigned long, arg)
1717{
1718 return ksys_old_semctl(semid, semnum, cmd, arg);
1719}
1720#endif
1721
Al Viroc0ebccb2017-07-09 10:03:23 -04001722#ifdef CONFIG_COMPAT
1723
1724struct compat_semid_ds {
1725 struct compat_ipc_perm sem_perm;
Arnd Bergmann9afc5ee2018-07-13 12:52:28 +02001726 old_time32_t sem_otime;
1727 old_time32_t sem_ctime;
Al Viroc0ebccb2017-07-09 10:03:23 -04001728 compat_uptr_t sem_base;
1729 compat_uptr_t sem_pending;
1730 compat_uptr_t sem_pending_last;
1731 compat_uptr_t undo;
1732 unsigned short sem_nsems;
1733};
1734
1735static int copy_compat_semid_from_user(struct semid64_ds *out, void __user *buf,
1736 int version)
1737{
1738 memset(out, 0, sizeof(*out));
1739 if (version == IPC_64) {
Linus Torvalds6aa211e2017-09-25 18:37:28 -07001740 struct compat_semid64_ds __user *p = buf;
Al Viroc0ebccb2017-07-09 10:03:23 -04001741 return get_compat_ipc64_perm(&out->sem_perm, &p->sem_perm);
1742 } else {
Linus Torvalds6aa211e2017-09-25 18:37:28 -07001743 struct compat_semid_ds __user *p = buf;
Al Viroc0ebccb2017-07-09 10:03:23 -04001744 return get_compat_ipc_perm(&out->sem_perm, &p->sem_perm);
1745 }
1746}
1747
1748static int copy_compat_semid_to_user(void __user *buf, struct semid64_ds *in,
1749 int version)
1750{
1751 if (version == IPC_64) {
1752 struct compat_semid64_ds v;
1753 memset(&v, 0, sizeof(v));
1754 to_compat_ipc64_perm(&v.sem_perm, &in->sem_perm);
Arnd Bergmannc2ab9752015-04-28 21:39:50 +02001755 v.sem_otime = lower_32_bits(in->sem_otime);
1756 v.sem_otime_high = upper_32_bits(in->sem_otime);
1757 v.sem_ctime = lower_32_bits(in->sem_ctime);
1758 v.sem_ctime_high = upper_32_bits(in->sem_ctime);
Al Viroc0ebccb2017-07-09 10:03:23 -04001759 v.sem_nsems = in->sem_nsems;
1760 return copy_to_user(buf, &v, sizeof(v));
1761 } else {
1762 struct compat_semid_ds v;
1763 memset(&v, 0, sizeof(v));
1764 to_compat_ipc_perm(&v.sem_perm, &in->sem_perm);
1765 v.sem_otime = in->sem_otime;
1766 v.sem_ctime = in->sem_ctime;
1767 v.sem_nsems = in->sem_nsems;
1768 return copy_to_user(buf, &v, sizeof(v));
1769 }
1770}
1771
Arnd Bergmann275f2212018-12-31 22:22:40 +01001772static long compat_ksys_semctl(int semid, int semnum, int cmd, int arg, int version)
Al Viroc0ebccb2017-07-09 10:03:23 -04001773{
1774 void __user *p = compat_ptr(arg);
1775 struct ipc_namespace *ns;
1776 struct semid64_ds semid64;
Al Viroc0ebccb2017-07-09 10:03:23 -04001777 int err;
1778
1779 ns = current->nsproxy->ipc_ns;
1780
1781 if (semid < 0)
1782 return -EINVAL;
1783
1784 switch (cmd & (~IPC_64)) {
1785 case IPC_INFO:
1786 case SEM_INFO:
1787 return semctl_info(ns, semid, cmd, p);
1788 case IPC_STAT:
1789 case SEM_STAT:
Davidlohr Buesoa280d6d2018-04-10 16:35:26 -07001790 case SEM_STAT_ANY:
Al Viroc0ebccb2017-07-09 10:03:23 -04001791 err = semctl_stat(ns, semid, cmd, &semid64);
1792 if (err < 0)
1793 return err;
1794 if (copy_compat_semid_to_user(p, &semid64, version))
1795 err = -EFAULT;
1796 return err;
1797 case GETVAL:
1798 case GETPID:
1799 case GETNCNT:
1800 case GETZCNT:
1801 case GETALL:
1802 case SETALL:
1803 return semctl_main(ns, semid, semnum, cmd, p);
1804 case SETVAL:
1805 return semctl_setval(ns, semid, semnum, arg);
1806 case IPC_SET:
1807 if (copy_compat_semid_from_user(&semid64, p, version))
1808 return -EFAULT;
1809 /* fallthru */
1810 case IPC_RMID:
1811 return semctl_down(ns, semid, cmd, &semid64);
1812 default:
1813 return -EINVAL;
1814 }
1815}
Dominik Brodowskid969c6f2018-03-20 20:00:39 +01001816
1817COMPAT_SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, int, arg)
1818{
Arnd Bergmann275f2212018-12-31 22:22:40 +01001819 return compat_ksys_semctl(semid, semnum, cmd, arg, IPC_64);
Dominik Brodowskid969c6f2018-03-20 20:00:39 +01001820}
Arnd Bergmann275f2212018-12-31 22:22:40 +01001821
1822#ifdef CONFIG_ARCH_WANT_COMPAT_IPC_PARSE_VERSION
1823long compat_ksys_old_semctl(int semid, int semnum, int cmd, int arg)
1824{
1825 int version = compat_ipc_parse_version(&cmd);
1826
1827 return compat_ksys_semctl(semid, semnum, cmd, arg, version);
1828}
1829
1830COMPAT_SYSCALL_DEFINE4(old_semctl, int, semid, int, semnum, int, cmd, int, arg)
1831{
1832 return compat_ksys_old_semctl(semid, semnum, cmd, arg);
1833}
1834#endif
Al Viroc0ebccb2017-07-09 10:03:23 -04001835#endif
1836
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837/* If the task doesn't already have a undo_list, then allocate one
1838 * here. We guarantee there is only one thread using this undo list,
1839 * and current is THE ONE
1840 *
1841 * If this allocation and assignment succeeds, but later
1842 * portions of this code fail, there is no need to free the sem_undo_list.
1843 * Just let it stay associated with the task, and it'll be freed later
1844 * at exit time.
1845 *
1846 * This can block, so callers must hold no locks.
1847 */
1848static inline int get_undo_list(struct sem_undo_list **undo_listp)
1849{
1850 struct sem_undo_list *undo_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851
1852 undo_list = current->sysvsem.undo_list;
1853 if (!undo_list) {
Matt Helsley2453a302006-10-02 02:18:25 -07001854 undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855 if (undo_list == NULL)
1856 return -ENOMEM;
Ingo Molnar00a5dfd2005-08-05 23:05:27 +02001857 spin_lock_init(&undo_list->lock);
Elena Reshetovaf74370b2017-09-08 16:17:42 -07001858 refcount_set(&undo_list->refcnt, 1);
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001859 INIT_LIST_HEAD(&undo_list->list_proc);
1860
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861 current->sysvsem.undo_list = undo_list;
1862 }
1863 *undo_listp = undo_list;
1864 return 0;
1865}
1866
Nick Pigginbf17bb72009-12-15 16:47:28 -08001867static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868{
Nick Pigginbf17bb72009-12-15 16:47:28 -08001869 struct sem_undo *un;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870
Joel Fernandes (Google)984035a2019-09-25 16:48:20 -07001871 list_for_each_entry_rcu(un, &ulp->list_proc, list_proc,
1872 spin_is_locked(&ulp->lock)) {
Nick Pigginbf17bb72009-12-15 16:47:28 -08001873 if (un->semid == semid)
1874 return un;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875 }
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001876 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877}
1878
Nick Pigginbf17bb72009-12-15 16:47:28 -08001879static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1880{
1881 struct sem_undo *un;
1882
Manfred Spraul239521f2014-01-27 17:07:04 -08001883 assert_spin_locked(&ulp->lock);
Nick Pigginbf17bb72009-12-15 16:47:28 -08001884
1885 un = __lookup_undo(ulp, semid);
1886 if (un) {
1887 list_del_rcu(&un->list_proc);
1888 list_add_rcu(&un->list_proc, &ulp->list_proc);
1889 }
1890 return un;
1891}
1892
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001893/**
Davidlohr Bueso8001c852014-01-27 17:07:05 -08001894 * find_alloc_undo - lookup (and if not present create) undo array
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001895 * @ns: namespace
1896 * @semid: semaphore array id
1897 *
1898 * The function looks up (and if not present creates) the undo structure.
1899 * The size of the undo structure depends on the size of the semaphore
1900 * array, thus the alloc path is not that straightforward.
Manfred Spraul380af1b2008-07-25 01:48:06 -07001901 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1902 * performs a rcu_read_lock().
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001903 */
1904static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905{
1906 struct sem_array *sma;
1907 struct sem_undo_list *ulp;
1908 struct sem_undo *un, *new;
Rik van Riel6062a8d2013-04-30 19:15:44 -07001909 int nsems, error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910
1911 error = get_undo_list(&ulp);
1912 if (error)
1913 return ERR_PTR(error);
1914
Manfred Spraul380af1b2008-07-25 01:48:06 -07001915 rcu_read_lock();
Pierre Peifferc530c6a2007-10-18 23:40:55 -07001916 spin_lock(&ulp->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917 un = lookup_undo(ulp, semid);
Pierre Peifferc530c6a2007-10-18 23:40:55 -07001918 spin_unlock(&ulp->lock);
Manfred Spraul239521f2014-01-27 17:07:04 -08001919 if (likely(un != NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920 goto out;
1921
1922 /* no undo structure around - allocate one. */
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001923 /* step 1: figure out the size of the semaphore array */
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001924 sma = sem_obtain_object_check(ns, semid);
1925 if (IS_ERR(sma)) {
1926 rcu_read_unlock();
Julia Lawall4de85cd2010-05-26 14:43:44 -07001927 return ERR_CAST(sma);
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001928 }
Nadia Derbey023a5352007-10-18 23:40:51 -07001929
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930 nsems = sma->sem_nsems;
Manfred Sprauldba4cdd2017-07-12 14:34:41 -07001931 if (!ipc_rcu_getref(&sma->sem_perm)) {
Rik van Riel6062a8d2013-04-30 19:15:44 -07001932 rcu_read_unlock();
1933 un = ERR_PTR(-EIDRM);
1934 goto out;
1935 }
Davidlohr Bueso16df3672013-04-30 19:15:29 -07001936 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001938 /* step 2: allocate new undo structure */
Burman Yan4668edc2006-12-06 20:38:51 -08001939 new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940 if (!new) {
Manfred Sprauldba4cdd2017-07-12 14:34:41 -07001941 ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942 return ERR_PTR(-ENOMEM);
1943 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944
Manfred Spraul380af1b2008-07-25 01:48:06 -07001945 /* step 3: Acquire the lock on semaphore array */
Linus Torvalds4091fd92013-05-04 10:13:40 -07001946 rcu_read_lock();
Pierre Peiffer6ff37972008-04-29 01:00:46 -07001947 sem_lock_and_putref(sma);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001948 if (!ipc_valid_object(&sma->sem_perm)) {
Rik van Riel6062a8d2013-04-30 19:15:44 -07001949 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07001950 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951 kfree(new);
1952 un = ERR_PTR(-EIDRM);
1953 goto out;
1954 }
Manfred Spraul380af1b2008-07-25 01:48:06 -07001955 spin_lock(&ulp->lock);
1956
1957 /*
1958 * step 4: check for races: did someone else allocate the undo struct?
1959 */
1960 un = lookup_undo(ulp, semid);
1961 if (un) {
1962 kfree(new);
1963 goto success;
1964 }
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001965 /* step 5: initialize & link new undo structure */
1966 new->semadj = (short *) &new[1];
Manfred Spraul380af1b2008-07-25 01:48:06 -07001967 new->ulp = ulp;
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001968 new->semid = semid;
1969 assert_spin_locked(&ulp->lock);
Manfred Spraul380af1b2008-07-25 01:48:06 -07001970 list_add_rcu(&new->list_proc, &ulp->list_proc);
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -07001971 ipc_assert_locked_object(&sma->sem_perm);
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001972 list_add(&new->list_id, &sma->list_id);
Manfred Spraul4daa28f2008-07-25 01:48:04 -07001973 un = new;
Manfred Spraul380af1b2008-07-25 01:48:06 -07001974
1975success:
1976 spin_unlock(&ulp->lock);
Rik van Riel6062a8d2013-04-30 19:15:44 -07001977 sem_unlock(sma, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978out:
1979 return un;
1980}
1981
Al Viro44ee4542017-07-09 10:50:14 -04001982static long do_semtimedop(int semid, struct sembuf __user *tsops,
Deepa Dinamani3ef56dc2017-08-02 19:51:10 -07001983 unsigned nsops, const struct timespec64 *timeout)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984{
1985 int error = -EINVAL;
1986 struct sem_array *sma;
1987 struct sembuf fast_sops[SEMOPM_FAST];
Manfred Spraul239521f2014-01-27 17:07:04 -08001988 struct sembuf *sops = fast_sops, *sop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989 struct sem_undo *un;
Davidlohr Bueso4ce33ec2016-12-14 15:06:37 -08001990 int max, locknum;
1991 bool undos = false, alter = false, dupsop = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992 struct sem_queue queue;
Davidlohr Bueso4ce33ec2016-12-14 15:06:37 -08001993 unsigned long dup = 0, jiffies_left = 0;
Kirill Korotaeve3893532006-10-02 02:18:22 -07001994 struct ipc_namespace *ns;
1995
1996 ns = current->nsproxy->ipc_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997
1998 if (nsops < 1 || semid < 0)
1999 return -EINVAL;
Kirill Korotaeve3893532006-10-02 02:18:22 -07002000 if (nsops > ns->sc_semopm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001 return -E2BIG;
Manfred Spraul239521f2014-01-27 17:07:04 -08002002 if (nsops > SEMOPM_FAST) {
Kees Cook344476e2018-06-12 14:04:32 -07002003 sops = kvmalloc_array(nsops, sizeof(*sops), GFP_KERNEL);
Manfred Spraul239521f2014-01-27 17:07:04 -08002004 if (sops == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005 return -ENOMEM;
2006 }
Davidlohr Bueso4ce33ec2016-12-14 15:06:37 -08002007
Manfred Spraul239521f2014-01-27 17:07:04 -08002008 if (copy_from_user(sops, tsops, nsops * sizeof(*tsops))) {
2009 error = -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010 goto out_free;
2011 }
Davidlohr Bueso4ce33ec2016-12-14 15:06:37 -08002012
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013 if (timeout) {
Al Viro44ee4542017-07-09 10:50:14 -04002014 if (timeout->tv_sec < 0 || timeout->tv_nsec < 0 ||
2015 timeout->tv_nsec >= 1000000000L) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016 error = -EINVAL;
2017 goto out_free;
2018 }
Deepa Dinamani3ef56dc2017-08-02 19:51:10 -07002019 jiffies_left = timespec64_to_jiffies(timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020 }
Davidlohr Bueso4ce33ec2016-12-14 15:06:37 -08002021
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022 max = 0;
2023 for (sop = sops; sop < sops + nsops; sop++) {
Davidlohr Bueso4ce33ec2016-12-14 15:06:37 -08002024 unsigned long mask = 1ULL << ((sop->sem_num) % BITS_PER_LONG);
2025
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026 if (sop->sem_num >= max)
2027 max = sop->sem_num;
2028 if (sop->sem_flg & SEM_UNDO)
Davidlohr Bueso4ce33ec2016-12-14 15:06:37 -08002029 undos = true;
2030 if (dup & mask) {
2031 /*
2032 * There was a previous alter access that appears
2033 * to have accessed the same semaphore, thus use
2034 * the dupsop logic. "appears", because the detection
2035 * can only check % BITS_PER_LONG.
2036 */
2037 dupsop = true;
2038 }
2039 if (sop->sem_op != 0) {
2040 alter = true;
2041 dup |= mask;
2042 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045 if (undos) {
Rik van Riel6062a8d2013-04-30 19:15:44 -07002046 /* On success, find_alloc_undo takes the rcu_read_lock */
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002047 un = find_alloc_undo(ns, semid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048 if (IS_ERR(un)) {
2049 error = PTR_ERR(un);
2050 goto out_free;
2051 }
Rik van Riel6062a8d2013-04-30 19:15:44 -07002052 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053 un = NULL;
Rik van Riel6062a8d2013-04-30 19:15:44 -07002054 rcu_read_lock();
2055 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056
Davidlohr Bueso16df3672013-04-30 19:15:29 -07002057 sma = sem_obtain_object_check(ns, semid);
Nadia Derbey023a5352007-10-18 23:40:51 -07002058 if (IS_ERR(sma)) {
Rik van Riel6062a8d2013-04-30 19:15:44 -07002059 rcu_read_unlock();
Nadia Derbey023a5352007-10-18 23:40:51 -07002060 error = PTR_ERR(sma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061 goto out_free;
Nadia Derbey023a5352007-10-18 23:40:51 -07002062 }
2063
Davidlohr Bueso16df3672013-04-30 19:15:29 -07002064 error = -EFBIG;
Davidlohr Bueso248e7352016-12-14 15:06:31 -08002065 if (max >= sma->sem_nsems) {
2066 rcu_read_unlock();
2067 goto out_free;
2068 }
Davidlohr Bueso16df3672013-04-30 19:15:29 -07002069
2070 error = -EACCES;
Davidlohr Bueso248e7352016-12-14 15:06:31 -08002071 if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO)) {
2072 rcu_read_unlock();
2073 goto out_free;
2074 }
Davidlohr Bueso16df3672013-04-30 19:15:29 -07002075
Eric W. Biedermanaefad952018-03-22 20:52:43 -05002076 error = security_sem_semop(&sma->sem_perm, sops, nsops, alter);
Davidlohr Bueso248e7352016-12-14 15:06:31 -08002077 if (error) {
2078 rcu_read_unlock();
2079 goto out_free;
2080 }
Davidlohr Bueso16df3672013-04-30 19:15:29 -07002081
Manfred Spraul6e224f92013-10-16 13:46:45 -07002082 error = -EIDRM;
2083 locknum = sem_lock(sma, sops, nsops);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08002084 /*
2085 * We eventually might perform the following check in a lockless
2086 * fashion, considering ipc_valid_object() locking constraints.
2087 * If nsops == 1 and there is no contention for sem_perm.lock, then
2088 * only a per-semaphore lock is held and it's OK to proceed with the
2089 * check below. More details on the fine grained locking scheme
2090 * entangled here and why it's RMID race safe on comments at sem_lock()
2091 */
2092 if (!ipc_valid_object(&sma->sem_perm))
Manfred Spraul6e224f92013-10-16 13:46:45 -07002093 goto out_unlock_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094 /*
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002095 * semid identifiers are not unique - find_alloc_undo may have
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096 * allocated an undo structure, it was invalidated by an RMID
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002097 * and now a new array with received the same id. Check and fail.
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002098 * This case can be detected checking un->semid. The existence of
Manfred Spraul380af1b2008-07-25 01:48:06 -07002099 * "un" itself is guaranteed by rcu.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100 */
Rik van Riel6062a8d2013-04-30 19:15:44 -07002101 if (un && un->semid == -1)
2102 goto out_unlock_free;
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002103
Manfred Sprauld198cd62014-06-06 14:37:49 -07002104 queue.sops = sops;
2105 queue.nsops = nsops;
2106 queue.undo = un;
Eric W. Biederman51d6f262018-03-23 01:11:29 -05002107 queue.pid = task_tgid(current);
Manfred Sprauld198cd62014-06-06 14:37:49 -07002108 queue.alter = alter;
Davidlohr Bueso4ce33ec2016-12-14 15:06:37 -08002109 queue.dupsop = dupsop;
Manfred Sprauld198cd62014-06-06 14:37:49 -07002110
2111 error = perform_atomic_semop(sma, &queue);
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08002112 if (error == 0) { /* non-blocking succesfull path */
2113 DEFINE_WAKE_Q(wake_q);
2114
2115 /*
2116 * If the operation was successful, then do
Manfred Spraul0e8c6652013-09-30 13:45:25 -07002117 * the required updates.
2118 */
2119 if (alter)
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08002120 do_smart_update(sma, sops, nsops, 1, &wake_q);
Manfred Spraul0e8c6652013-09-30 13:45:25 -07002121 else
2122 set_semotime(sma, sops);
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08002123
2124 sem_unlock(sma, locknum);
2125 rcu_read_unlock();
2126 wake_up_q(&wake_q);
2127
2128 goto out_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129 }
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08002130 if (error < 0) /* non-blocking error path */
Manfred Spraul0e8c6652013-09-30 13:45:25 -07002131 goto out_unlock_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002132
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08002133 /*
2134 * We need to sleep on this operation, so we put the current
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135 * task into the pending queue and go to sleep.
2136 */
Manfred Spraulb97e8202009-12-15 16:47:32 -08002137 if (nsops == 1) {
2138 struct sem *curr;
Davidlohr Buesoec67aaa2018-06-14 15:27:51 -07002139 int idx = array_index_nospec(sops->sem_num, sma->sem_nsems);
2140 curr = &sma->sems[idx];
Manfred Spraulb97e8202009-12-15 16:47:32 -08002141
Manfred Spraulf269f402013-07-08 16:01:24 -07002142 if (alter) {
2143 if (sma->complex_count) {
2144 list_add_tail(&queue.list,
2145 &sma->pending_alter);
2146 } else {
2147
2148 list_add_tail(&queue.list,
2149 &curr->pending_alter);
2150 }
2151 } else {
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07002152 list_add_tail(&queue.list, &curr->pending_const);
Manfred Spraulf269f402013-07-08 16:01:24 -07002153 }
Manfred Spraulb97e8202009-12-15 16:47:32 -08002154 } else {
Manfred Spraulf269f402013-07-08 16:01:24 -07002155 if (!sma->complex_count)
2156 merge_queues(sma);
2157
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07002158 if (alter)
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07002159 list_add_tail(&queue.list, &sma->pending_alter);
Rik van Riel9f1bc2c92013-04-30 19:15:39 -07002160 else
Manfred Spraul1a82e9e2013-07-08 16:01:23 -07002161 list_add_tail(&queue.list, &sma->pending_const);
2162
Manfred Spraulb97e8202009-12-15 16:47:32 -08002163 sma->complex_count++;
2164 }
2165
Davidlohr Buesob5fa01a2016-12-14 15:06:46 -08002166 do {
Manfred Spraul8116b542020-02-03 17:34:42 -08002167 /* memory ordering ensured by the lock in sem_lock() */
Davidlohr Buesof075faa2018-07-26 16:37:19 -07002168 WRITE_ONCE(queue.status, -EINTR);
Davidlohr Buesob5fa01a2016-12-14 15:06:46 -08002169 queue.sleeper = current;
Manfred Spraul0b0577f2011-11-02 13:38:52 -07002170
Manfred Spraul8116b542020-02-03 17:34:42 -08002171 /* memory ordering is ensured by the lock in sem_lock() */
Davidlohr Buesob5fa01a2016-12-14 15:06:46 -08002172 __set_current_state(TASK_INTERRUPTIBLE);
2173 sem_unlock(sma, locknum);
Linus Torvalds321310c2013-05-04 10:47:57 -07002174 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175
Davidlohr Buesob5fa01a2016-12-14 15:06:46 -08002176 if (timeout)
2177 jiffies_left = schedule_timeout(jiffies_left);
2178 else
2179 schedule();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180
Davidlohr Buesob5fa01a2016-12-14 15:06:46 -08002181 /*
2182 * fastpath: the semop has completed, either successfully or
2183 * not, from the syscall pov, is quite irrelevant to us at this
2184 * point; we're done.
2185 *
2186 * We _do_ care, nonetheless, about being awoken by a signal or
2187 * spuriously. The queue.status is checked again in the
2188 * slowpath (aka after taking sem_lock), such that we can detect
2189 * scenarios where we were awakened externally, during the
2190 * window between wake_q_add() and wake_up_q().
2191 */
2192 error = READ_ONCE(queue.status);
2193 if (error != -EINTR) {
Manfred Spraul8116b542020-02-03 17:34:42 -08002194 /* see SEM_BARRIER_2 for purpose/pairing */
2195 smp_acquire__after_ctrl_dep();
Davidlohr Buesob5fa01a2016-12-14 15:06:46 -08002196 goto out_free;
2197 }
Manfred Spraul0b0577f2011-11-02 13:38:52 -07002198
Davidlohr Buesob5fa01a2016-12-14 15:06:46 -08002199 rcu_read_lock();
Manfred Spraulc626bc42017-01-10 16:57:48 -08002200 locknum = sem_lock(sma, sops, nsops);
Davidlohr Buesob5fa01a2016-12-14 15:06:46 -08002201
Davidlohr Bueso370b2622016-12-14 15:06:49 -08002202 if (!ipc_valid_object(&sma->sem_perm))
2203 goto out_unlock_free;
2204
Manfred Spraul8116b542020-02-03 17:34:42 -08002205 /*
2206 * No necessity for any barrier: We are protect by sem_lock()
2207 */
Davidlohr Bueso370b2622016-12-14 15:06:49 -08002208 error = READ_ONCE(queue.status);
Davidlohr Buesob5fa01a2016-12-14 15:06:46 -08002209
2210 /*
2211 * If queue.status != -EINTR we are woken up by another process.
2212 * Leave without unlink_queue(), but with sem_unlock().
2213 */
2214 if (error != -EINTR)
2215 goto out_unlock_free;
2216
2217 /*
2218 * If an interrupt occurred we have to clean up the queue.
2219 */
2220 if (timeout && jiffies_left == 0)
2221 error = -EAGAIN;
2222 } while (error == -EINTR && !signal_pending(current)); /* spurious */
Manfred Spraul0b0577f2011-11-02 13:38:52 -07002223
Manfred Spraulb97e8202009-12-15 16:47:32 -08002224 unlink_queue(sma, &queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225
2226out_unlock_free:
Rik van Riel6062a8d2013-04-30 19:15:44 -07002227 sem_unlock(sma, locknum);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07002228 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229out_free:
Manfred Spraul239521f2014-01-27 17:07:04 -08002230 if (sops != fast_sops)
Davidlohr Buesoe4243b82017-09-08 16:17:52 -07002231 kvfree(sops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002232 return error;
2233}
2234
Dominik Brodowski41f4f0e2018-03-20 19:48:14 +01002235long ksys_semtimedop(int semid, struct sembuf __user *tsops,
Arnd Bergmann21fc5382018-04-13 13:58:00 +02002236 unsigned int nsops, const struct __kernel_timespec __user *timeout)
Al Viro44ee4542017-07-09 10:50:14 -04002237{
2238 if (timeout) {
Deepa Dinamani3ef56dc2017-08-02 19:51:10 -07002239 struct timespec64 ts;
2240 if (get_timespec64(&ts, timeout))
Al Viro44ee4542017-07-09 10:50:14 -04002241 return -EFAULT;
2242 return do_semtimedop(semid, tsops, nsops, &ts);
2243 }
2244 return do_semtimedop(semid, tsops, nsops, NULL);
2245}
2246
Dominik Brodowski41f4f0e2018-03-20 19:48:14 +01002247SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
Arnd Bergmann21fc5382018-04-13 13:58:00 +02002248 unsigned int, nsops, const struct __kernel_timespec __user *, timeout)
Dominik Brodowski41f4f0e2018-03-20 19:48:14 +01002249{
2250 return ksys_semtimedop(semid, tsops, nsops, timeout);
2251}
2252
Arnd Bergmannb0d17572018-04-13 13:58:23 +02002253#ifdef CONFIG_COMPAT_32BIT_TIME
Dominik Brodowski41f4f0e2018-03-20 19:48:14 +01002254long compat_ksys_semtimedop(int semid, struct sembuf __user *tsems,
2255 unsigned int nsops,
Arnd Bergmann9afc5ee2018-07-13 12:52:28 +02002256 const struct old_timespec32 __user *timeout)
Al Viro44ee4542017-07-09 10:50:14 -04002257{
2258 if (timeout) {
Deepa Dinamani3ef56dc2017-08-02 19:51:10 -07002259 struct timespec64 ts;
Arnd Bergmann9afc5ee2018-07-13 12:52:28 +02002260 if (get_old_timespec32(&ts, timeout))
Al Viro44ee4542017-07-09 10:50:14 -04002261 return -EFAULT;
2262 return do_semtimedop(semid, tsems, nsops, &ts);
2263 }
2264 return do_semtimedop(semid, tsems, nsops, NULL);
2265}
Dominik Brodowski41f4f0e2018-03-20 19:48:14 +01002266
Arnd Bergmann8dabe722019-01-07 00:33:08 +01002267SYSCALL_DEFINE4(semtimedop_time32, int, semid, struct sembuf __user *, tsems,
Dominik Brodowski41f4f0e2018-03-20 19:48:14 +01002268 unsigned int, nsops,
Arnd Bergmann9afc5ee2018-07-13 12:52:28 +02002269 const struct old_timespec32 __user *, timeout)
Dominik Brodowski41f4f0e2018-03-20 19:48:14 +01002270{
2271 return compat_ksys_semtimedop(semid, tsems, nsops, timeout);
2272}
Al Viro44ee4542017-07-09 10:50:14 -04002273#endif
2274
Heiko Carstensd5460c92009-01-14 14:14:27 +01002275SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
2276 unsigned, nsops)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002277{
Al Viro44ee4542017-07-09 10:50:14 -04002278 return do_semtimedop(semid, tsops, nsops, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002279}
2280
2281/* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
2282 * parent and child tasks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002283 */
2284
2285int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
2286{
2287 struct sem_undo_list *undo_list;
2288 int error;
2289
2290 if (clone_flags & CLONE_SYSVSEM) {
2291 error = get_undo_list(&undo_list);
2292 if (error)
2293 return error;
Elena Reshetovaf74370b2017-09-08 16:17:42 -07002294 refcount_inc(&undo_list->refcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295 tsk->sysvsem.undo_list = undo_list;
Paul McQuade46c0a8c2014-06-06 14:37:37 -07002296 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -07002297 tsk->sysvsem.undo_list = NULL;
2298
2299 return 0;
2300}
2301
2302/*
2303 * add semadj values to semaphores, free undo structures.
2304 * undo structures are not freed when semaphore arrays are destroyed
2305 * so some of them may be out of date.
2306 * IMPLEMENTATION NOTE: There is some confusion over whether the
2307 * set of adjustments that needs to be done should be done in an atomic
2308 * manner or not. That is, if we are attempting to decrement the semval
2309 * should we queue up and wait until we can do so legally?
2310 * The original implementation attempted to do this (queue and wait).
2311 * The current implementation does not do so. The POSIX standard
2312 * and SVID should be consulted to determine what behavior is mandated.
2313 */
2314void exit_sem(struct task_struct *tsk)
2315{
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002316 struct sem_undo_list *ulp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002317
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002318 ulp = tsk->sysvsem.undo_list;
2319 if (!ulp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320 return;
Manfred Spraul9edff4a2008-04-29 01:00:57 -07002321 tsk->sysvsem.undo_list = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322
Elena Reshetovaf74370b2017-09-08 16:17:42 -07002323 if (!refcount_dec_and_test(&ulp->refcnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002324 return;
2325
Manfred Spraul380af1b2008-07-25 01:48:06 -07002326 for (;;) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002327 struct sem_array *sma;
Manfred Spraul380af1b2008-07-25 01:48:06 -07002328 struct sem_undo *un;
Rik van Riel6062a8d2013-04-30 19:15:44 -07002329 int semid, i;
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08002330 DEFINE_WAKE_Q(wake_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331
Nikolay Borisov2a1613a2016-10-11 13:55:05 -07002332 cond_resched();
2333
Manfred Spraul380af1b2008-07-25 01:48:06 -07002334 rcu_read_lock();
Jiri Pirko05725f72009-04-14 20:17:16 +02002335 un = list_entry_rcu(ulp->list_proc.next,
2336 struct sem_undo, list_proc);
Herton R. Krzesinski602b8592015-08-14 15:35:02 -07002337 if (&un->list_proc == &ulp->list_proc) {
2338 /*
2339 * We must wait for freeary() before freeing this ulp,
2340 * in case we raced with last sem_undo. There is a small
2341 * possibility where we exit while freeary() didn't
2342 * finish unlocking sem_undo_list.
2343 */
Paul E. McKenneye0892e02017-06-29 12:59:32 -07002344 spin_lock(&ulp->lock);
2345 spin_unlock(&ulp->lock);
Rik van Riel6062a8d2013-04-30 19:15:44 -07002346 rcu_read_unlock();
Manfred Spraul380af1b2008-07-25 01:48:06 -07002347 break;
Rik van Riel6062a8d2013-04-30 19:15:44 -07002348 }
Herton R. Krzesinski602b8592015-08-14 15:35:02 -07002349 spin_lock(&ulp->lock);
2350 semid = un->semid;
2351 spin_unlock(&ulp->lock);
Manfred Spraul380af1b2008-07-25 01:48:06 -07002352
Herton R. Krzesinski602b8592015-08-14 15:35:02 -07002353 /* exit_sem raced with IPC_RMID, nothing to do */
2354 if (semid == -1) {
2355 rcu_read_unlock();
2356 continue;
2357 }
2358
2359 sma = sem_obtain_object_check(tsk->nsproxy->ipc_ns, semid);
Manfred Spraul380af1b2008-07-25 01:48:06 -07002360 /* exit_sem raced with IPC_RMID, nothing to do */
Rik van Riel6062a8d2013-04-30 19:15:44 -07002361 if (IS_ERR(sma)) {
2362 rcu_read_unlock();
Manfred Spraul380af1b2008-07-25 01:48:06 -07002363 continue;
Rik van Riel6062a8d2013-04-30 19:15:44 -07002364 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002365
Rik van Riel6062a8d2013-04-30 19:15:44 -07002366 sem_lock(sma, NULL, -1);
Manfred Spraul6e224f92013-10-16 13:46:45 -07002367 /* exit_sem raced with IPC_RMID, nothing to do */
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08002368 if (!ipc_valid_object(&sma->sem_perm)) {
Manfred Spraul6e224f92013-10-16 13:46:45 -07002369 sem_unlock(sma, -1);
2370 rcu_read_unlock();
2371 continue;
2372 }
Nick Pigginbf17bb72009-12-15 16:47:28 -08002373 un = __lookup_undo(ulp, semid);
Manfred Spraul380af1b2008-07-25 01:48:06 -07002374 if (un == NULL) {
2375 /* exit_sem raced with IPC_RMID+semget() that created
2376 * exactly the same semid. Nothing to do.
2377 */
Rik van Riel6062a8d2013-04-30 19:15:44 -07002378 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07002379 rcu_read_unlock();
Manfred Spraul380af1b2008-07-25 01:48:06 -07002380 continue;
2381 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382
Manfred Spraul380af1b2008-07-25 01:48:06 -07002383 /* remove un from the linked lists */
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -07002384 ipc_assert_locked_object(&sma->sem_perm);
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002385 list_del(&un->list_id);
2386
Ioanna Alifierakiedf28f42020-02-20 20:04:00 -08002387 spin_lock(&ulp->lock);
Manfred Spraul380af1b2008-07-25 01:48:06 -07002388 list_del_rcu(&un->list_proc);
Ioanna Alifierakiedf28f42020-02-20 20:04:00 -08002389 spin_unlock(&ulp->lock);
Manfred Spraul380af1b2008-07-25 01:48:06 -07002390
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002391 /* perform adjustments registered in un */
2392 for (i = 0; i < sma->sem_nsems; i++) {
Manfred Spraul1a233952017-07-12 14:34:38 -07002393 struct sem *semaphore = &sma->sems[i];
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002394 if (un->semadj[i]) {
2395 semaphore->semval += un->semadj[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002396 /*
2397 * Range checks of the new semaphore value,
2398 * not defined by sus:
2399 * - Some unices ignore the undo entirely
2400 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
2401 * - some cap the value (e.g. FreeBSD caps
2402 * at 0, but doesn't enforce SEMVMX)
2403 *
2404 * Linux caps the semaphore value, both at 0
2405 * and at SEMVMX.
2406 *
Manfred Spraul239521f2014-01-27 17:07:04 -08002407 * Manfred <manfred@colorfullife.com>
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408 */
Ingo Molnar5f921ae2006-03-26 01:37:17 -08002409 if (semaphore->semval < 0)
2410 semaphore->semval = 0;
2411 if (semaphore->semval > SEMVMX)
2412 semaphore->semval = SEMVMX;
Eric W. Biederman51d6f262018-03-23 01:11:29 -05002413 ipc_update_pid(&semaphore->sempid, task_tgid(current));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002414 }
2415 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002416 /* maybe some queued-up processes were waiting for this */
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08002417 do_smart_update(sma, NULL, 0, 1, &wake_q);
Rik van Riel6062a8d2013-04-30 19:15:44 -07002418 sem_unlock(sma, -1);
Linus Torvalds6d49dab2013-05-03 15:04:40 -07002419 rcu_read_unlock();
Davidlohr Bueso9ae949f2016-12-14 15:06:34 -08002420 wake_up_q(&wake_q);
Manfred Spraul380af1b2008-07-25 01:48:06 -07002421
Lai Jiangshan693a8b62011-03-18 12:09:35 +08002422 kfree_rcu(un, rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423 }
Manfred Spraul4daa28f2008-07-25 01:48:04 -07002424 kfree(ulp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002425}
2426
2427#ifdef CONFIG_PROC_FS
Mike Waychison19b49462005-09-06 15:17:10 -07002428static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429{
Eric W. Biederman1efdb692012-02-07 16:54:11 -08002430 struct user_namespace *user_ns = seq_user_ns(s);
Kees Cookade9f912017-08-02 13:32:21 -07002431 struct kern_ipc_perm *ipcp = it;
2432 struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
Deepa Dinamanie54d02b2017-08-02 19:51:13 -07002433 time64_t sem_otime;
Manfred Sprauld12e1e52013-07-08 16:01:25 -07002434
Manfred Sprauld8c63372013-09-30 13:45:07 -07002435 /*
2436 * The proc interface isn't aware of sem_lock(), it calls
2437 * ipc_lock_object() directly (in sysvipc_find_ipc).
Manfred Spraul5864a2f2016-10-11 13:54:50 -07002438 * In order to stay compatible with sem_lock(), we must
2439 * enter / leave complex_mode.
Manfred Sprauld8c63372013-09-30 13:45:07 -07002440 */
Manfred Spraul5864a2f2016-10-11 13:54:50 -07002441 complexmode_enter(sma);
Manfred Sprauld8c63372013-09-30 13:45:07 -07002442
Manfred Sprauld12e1e52013-07-08 16:01:25 -07002443 sem_otime = get_semotime(sma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002444
Joe Perches7f032d62015-04-15 16:17:54 -07002445 seq_printf(s,
Deepa Dinamanie54d02b2017-08-02 19:51:13 -07002446 "%10d %10d %4o %10u %5u %5u %5u %5u %10llu %10llu\n",
Joe Perches7f032d62015-04-15 16:17:54 -07002447 sma->sem_perm.key,
2448 sma->sem_perm.id,
2449 sma->sem_perm.mode,
2450 sma->sem_nsems,
2451 from_kuid_munged(user_ns, sma->sem_perm.uid),
2452 from_kgid_munged(user_ns, sma->sem_perm.gid),
2453 from_kuid_munged(user_ns, sma->sem_perm.cuid),
2454 from_kgid_munged(user_ns, sma->sem_perm.cgid),
2455 sem_otime,
2456 sma->sem_ctime);
2457
Manfred Spraul5864a2f2016-10-11 13:54:50 -07002458 complexmode_tryleave(sma);
2459
Joe Perches7f032d62015-04-15 16:17:54 -07002460 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002461}
2462#endif