]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - ipc/sem.c
netfilter: bridge: remove unneeded rcu_read_lock
[mirror_ubuntu-bionic-kernel.git] / ipc / sem.c
CommitLineData
1da177e4
LT
1/*
2 * linux/ipc/sem.c
3 * Copyright (C) 1992 Krishna Balasubramanian
4 * Copyright (C) 1995 Eric Schenk, Bruno Haible
5 *
1da177e4
LT
6 * /proc/sysvipc/sem support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
7 *
8 * SMP-threaded, sysctl's added
624dffcb 9 * (c) 1999 Manfred Spraul <manfred@colorfullife.com>
1da177e4 10 * Enforced range limit on SEM_UNDO
046c6884 11 * (c) 2001 Red Hat Inc
1da177e4
LT
12 * Lockless wakeup
13 * (c) 2003 Manfred Spraul <manfred@colorfullife.com>
9ae949fa 14 * (c) 2016 Davidlohr Bueso <dave@stgolabs.net>
c5cf6359
MS
15 * Further wakeup optimizations, documentation
16 * (c) 2010 Manfred Spraul <manfred@colorfullife.com>
073115d6
SG
17 *
18 * support for audit of ipc object properties and permission changes
19 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
e3893534
KK
20 *
21 * namespaces support
22 * OpenVZ, SWsoft Inc.
23 * Pavel Emelianov <xemul@openvz.org>
c5cf6359
MS
24 *
25 * Implementation notes: (May 2010)
26 * This file implements System V semaphores.
27 *
28 * User space visible behavior:
29 * - FIFO ordering for semop() operations (just FIFO, not starvation
30 * protection)
31 * - multiple semaphore operations that alter the same semaphore in
32 * one semop() are handled.
33 * - sem_ctime (time of last semctl()) is updated in the IPC_SET, SETVAL and
34 * SETALL calls.
35 * - two Linux specific semctl() commands: SEM_STAT, SEM_INFO.
36 * - undo adjustments at process exit are limited to 0..SEMVMX.
37 * - namespace are supported.
38 * - SEMMSL, SEMMNS, SEMOPM and SEMMNI can be configured at runtine by writing
39 * to /proc/sys/kernel/sem.
40 * - statistics about the usage are reported in /proc/sysvipc/sem.
41 *
42 * Internals:
43 * - scalability:
44 * - all global variables are read-mostly.
45 * - semop() calls and semctl(RMID) are synchronized by RCU.
46 * - most operations do write operations (actually: spin_lock calls) to
47 * the per-semaphore array structure.
48 * Thus: Perfect SMP scaling between independent semaphore arrays.
49 * If multiple semaphores in one array are used, then cache line
50 * trashing on the semaphore array spinlock will limit the scaling.
2f2ed41d 51 * - semncnt and semzcnt are calculated on demand in count_semcnt()
c5cf6359
MS
52 * - the task that performs a successful semop() scans the list of all
53 * sleeping tasks and completes any pending operations that can be fulfilled.
54 * Semaphores are actively given to waiting tasks (necessary for FIFO).
55 * (see update_queue())
56 * - To improve the scalability, the actual wake-up calls are performed after
9ae949fa 57 * dropping all locks. (see wake_up_sem_queue_prepare())
c5cf6359
MS
58 * - All work is done by the waker, the woken up task does not have to do
59 * anything - not even acquiring a lock or dropping a refcount.
60 * - A woken up task may not even touch the semaphore array anymore, it may
61 * have been destroyed already by a semctl(RMID).
c5cf6359
MS
62 * - UNDO values are stored in an array (one per process and per
63 * semaphore array, lazily allocated). For backwards compatibility, multiple
64 * modes for the UNDO variables are supported (per process, per thread)
65 * (see copy_semundo, CLONE_SYSVSEM)
66 * - There are two lists of the pending operations: a per-array list
67 * and per-semaphore list (stored in the array). This allows to achieve FIFO
68 * ordering without always scanning all pending operations.
69 * The worst-case behavior is nevertheless O(N^2) for N wakeups.
1da177e4
LT
70 */
71
1da177e4
LT
72#include <linux/slab.h>
73#include <linux/spinlock.h>
74#include <linux/init.h>
75#include <linux/proc_fs.h>
76#include <linux/time.h>
1da177e4
LT
77#include <linux/security.h>
78#include <linux/syscalls.h>
79#include <linux/audit.h>
c59ede7b 80#include <linux/capability.h>
19b4946c 81#include <linux/seq_file.h>
3e148c79 82#include <linux/rwsem.h>
e3893534 83#include <linux/nsproxy.h>
ae5e1b22 84#include <linux/ipc_namespace.h>
84f001e1 85#include <linux/sched/wake_q.h>
5f921ae9 86
7153e402 87#include <linux/uaccess.h>
1da177e4
LT
88#include "util.h"
89
e57940d7
MS
90/* One semaphore structure for each semaphore in the system. */
91struct sem {
92 int semval; /* current value */
a5f4db87
DB
93 /*
94 * PID of the process that last modified the semaphore. For
95 * Linux, specifically these are:
96 * - semop
97 * - semctl, via SETVAL and SETALL.
98 * - at task exit when performing undo adjustments (see exit_sem).
99 */
100 int sempid;
6062a8dc 101 spinlock_t lock; /* spinlock for fine-grained semtimedop */
1a82e9e1
MS
102 struct list_head pending_alter; /* pending single-sop operations */
103 /* that alter the semaphore */
104 struct list_head pending_const; /* pending single-sop operations */
105 /* that do not alter the semaphore*/
d12e1e50 106 time_t sem_otime; /* candidate for sem_otime */
f5c936c0 107} ____cacheline_aligned_in_smp;
e57940d7
MS
108
109/* One queue for each sleeping process in the system. */
110struct sem_queue {
e57940d7
MS
111 struct list_head list; /* queue of pending operations */
112 struct task_struct *sleeper; /* this process */
113 struct sem_undo *undo; /* undo structure */
114 int pid; /* process id of requesting process */
115 int status; /* completion status of operation */
116 struct sembuf *sops; /* array of pending operations */
ed247b7c 117 struct sembuf *blocking; /* the operation that blocked */
e57940d7 118 int nsops; /* number of operations */
4ce33ec2
DB
119 bool alter; /* does *sops alter the array? */
120 bool dupsop; /* sops on more than one sem_num */
e57940d7
MS
121};
122
123/* Each task has a list of undo requests. They are executed automatically
124 * when the process exits.
125 */
126struct sem_undo {
127 struct list_head list_proc; /* per-process list: *
128 * all undos from one process
129 * rcu protected */
130 struct rcu_head rcu; /* rcu struct for sem_undo */
131 struct sem_undo_list *ulp; /* back ptr to sem_undo_list */
132 struct list_head list_id; /* per semaphore array list:
133 * all undos for one array */
134 int semid; /* semaphore set identifier */
135 short *semadj; /* array of adjustments */
136 /* one per semaphore */
137};
138
139/* sem_undo_list controls shared access to the list of sem_undo structures
140 * that may be shared among all a CLONE_SYSVSEM task group.
141 */
142struct sem_undo_list {
143 atomic_t refcnt;
144 spinlock_t lock;
145 struct list_head list_proc;
146};
147
148
ed2ddbf8 149#define sem_ids(ns) ((ns)->ids[IPC_SEM_IDS])
e3893534 150
1b531f21 151#define sem_checkid(sma, semid) ipc_checkid(&sma->sem_perm, semid)
1da177e4 152
7748dbfa 153static int newary(struct ipc_namespace *, struct ipc_params *);
01b8b07a 154static void freeary(struct ipc_namespace *, struct kern_ipc_perm *);
1da177e4 155#ifdef CONFIG_PROC_FS
19b4946c 156static int sysvipc_sem_proc_show(struct seq_file *s, void *it);
1da177e4
LT
157#endif
158
159#define SEMMSL_FAST 256 /* 512 bytes on stack */
160#define SEMOPM_FAST 64 /* ~ 372 bytes on stack */
161
9de5ab8a
MS
162/*
163 * Switching from the mode suitable for simple ops
164 * to the mode for complex ops is costly. Therefore:
165 * use some hysteresis
166 */
167#define USE_GLOBAL_LOCK_HYSTERESIS 10
168
1da177e4 169/*
758a6ba3 170 * Locking:
5864a2fd 171 * a) global sem_lock() for read/write
1da177e4 172 * sem_undo.id_next,
758a6ba3 173 * sem_array.complex_count,
5864a2fd
MS
174 * sem_array.pending{_alter,_const},
175 * sem_array.sem_undo
46c0a8ca 176 *
5864a2fd 177 * b) global or semaphore sem_lock() for read/write:
758a6ba3 178 * sem_array.sem_base[i].pending_{const,alter}:
5864a2fd
MS
179 *
180 * c) special:
181 * sem_undo_list.list_proc:
182 * * undo_list->lock for write
183 * * rcu for read
9de5ab8a
MS
184 * use_global_lock:
185 * * global sem_lock() for write
186 * * either local or global sem_lock() for read.
187 *
188 * Memory ordering:
189 * Most ordering is enforced by using spin_lock() and spin_unlock().
190 * The special case is use_global_lock:
191 * Setting it from non-zero to 0 is a RELEASE, this is ensured by
192 * using smp_store_release().
193 * Testing if it is non-zero is an ACQUIRE, this is ensured by using
194 * smp_load_acquire().
195 * Setting it from 0 to non-zero must be ordered with regards to
196 * this smp_load_acquire(), this is guaranteed because the smp_load_acquire()
197 * is inside a spin_lock() and after a write from 0 to non-zero a
198 * spin_lock()+spin_unlock() is done.
1da177e4
LT
199 */
200
e3893534
KK
201#define sc_semmsl sem_ctls[0]
202#define sc_semmns sem_ctls[1]
203#define sc_semopm sem_ctls[2]
204#define sc_semmni sem_ctls[3]
205
ed2ddbf8 206void sem_init_ns(struct ipc_namespace *ns)
e3893534 207{
e3893534
KK
208 ns->sc_semmsl = SEMMSL;
209 ns->sc_semmns = SEMMNS;
210 ns->sc_semopm = SEMOPM;
211 ns->sc_semmni = SEMMNI;
212 ns->used_sems = 0;
ed2ddbf8 213 ipc_init_ids(&ns->ids[IPC_SEM_IDS]);
e3893534
KK
214}
215
ae5e1b22 216#ifdef CONFIG_IPC_NS
e3893534
KK
217void sem_exit_ns(struct ipc_namespace *ns)
218{
01b8b07a 219 free_ipcs(ns, &sem_ids(ns), freeary);
7d6feeb2 220 idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr);
e3893534 221}
ae5e1b22 222#endif
1da177e4 223
239521f3 224void __init sem_init(void)
1da177e4 225{
ed2ddbf8 226 sem_init_ns(&init_ipc_ns);
19b4946c
MW
227 ipc_init_proc_interface("sysvipc/sem",
228 " key semid perms nsems uid gid cuid cgid otime ctime\n",
e3893534 229 IPC_SEM_IDS, sysvipc_sem_proc_show);
1da177e4
LT
230}
231
f269f40a
MS
232/**
233 * unmerge_queues - unmerge queues, if possible.
234 * @sma: semaphore array
235 *
236 * The function unmerges the wait queues if complex_count is 0.
237 * It must be called prior to dropping the global semaphore array lock.
238 */
239static void unmerge_queues(struct sem_array *sma)
240{
241 struct sem_queue *q, *tq;
242
243 /* complex operations still around? */
244 if (sma->complex_count)
245 return;
246 /*
247 * We will switch back to simple mode.
248 * Move all pending operation back into the per-semaphore
249 * queues.
250 */
251 list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
252 struct sem *curr;
253 curr = &sma->sem_base[q->sops[0].sem_num];
254
255 list_add_tail(&q->list, &curr->pending_alter);
256 }
257 INIT_LIST_HEAD(&sma->pending_alter);
258}
259
260/**
8001c858 261 * merge_queues - merge single semop queues into global queue
f269f40a
MS
262 * @sma: semaphore array
263 *
264 * This function merges all per-semaphore queues into the global queue.
265 * It is necessary to achieve FIFO ordering for the pending single-sop
266 * operations when a multi-semop operation must sleep.
267 * Only the alter operations must be moved, the const operations can stay.
268 */
269static void merge_queues(struct sem_array *sma)
270{
271 int i;
272 for (i = 0; i < sma->sem_nsems; i++) {
273 struct sem *sem = sma->sem_base + i;
274
275 list_splice_init(&sem->pending_alter, &sma->pending_alter);
276 }
277}
278
53dad6d3
DB
279static void sem_rcu_free(struct rcu_head *head)
280{
281 struct ipc_rcu *p = container_of(head, struct ipc_rcu, rcu);
282 struct sem_array *sma = ipc_rcu_to_struct(p);
283
284 security_sem_free(sma);
285 ipc_rcu_free(head);
286}
287
5e9d5275 288/*
5864a2fd 289 * Enter the mode suitable for non-simple operations:
5e9d5275 290 * Caller must own sem_perm.lock.
5e9d5275 291 */
5864a2fd 292static void complexmode_enter(struct sem_array *sma)
5e9d5275
MS
293{
294 int i;
295 struct sem *sem;
296
9de5ab8a
MS
297 if (sma->use_global_lock > 0) {
298 /*
299 * We are already in global lock mode.
300 * Nothing to do, just reset the
301 * counter until we return to simple mode.
302 */
303 sma->use_global_lock = USE_GLOBAL_LOCK_HYSTERESIS;
6d07b68c
MS
304 return;
305 }
9de5ab8a 306 sma->use_global_lock = USE_GLOBAL_LOCK_HYSTERESIS;
5864a2fd 307
5e9d5275
MS
308 for (i = 0; i < sma->sem_nsems; i++) {
309 sem = sma->sem_base + i;
27d7be18
MS
310 spin_lock(&sem->lock);
311 spin_unlock(&sem->lock);
5e9d5275 312 }
5864a2fd
MS
313}
314
315/*
316 * Try to leave the mode that disallows simple operations:
317 * Caller must own sem_perm.lock.
318 */
319static void complexmode_tryleave(struct sem_array *sma)
320{
321 if (sma->complex_count) {
322 /* Complex ops are sleeping.
323 * We must stay in complex mode
324 */
325 return;
326 }
9de5ab8a
MS
327 if (sma->use_global_lock == 1) {
328 /*
329 * Immediately after setting use_global_lock to 0,
330 * a simple op can start. Thus: all memory writes
331 * performed by the current operation must be visible
332 * before we set use_global_lock to 0.
333 */
334 smp_store_release(&sma->use_global_lock, 0);
335 } else {
336 sma->use_global_lock--;
337 }
5e9d5275
MS
338}
339
5864a2fd 340#define SEM_GLOBAL_LOCK (-1)
6062a8dc
RR
341/*
342 * If the request contains only one semaphore operation, and there are
343 * no complex transactions pending, lock only the semaphore involved.
344 * Otherwise, lock the entire semaphore array, since we either have
345 * multiple semaphores in our own semops, or we need to look at
346 * semaphores from other pending complex operations.
6062a8dc
RR
347 */
348static inline int sem_lock(struct sem_array *sma, struct sembuf *sops,
349 int nsops)
350{
5e9d5275 351 struct sem *sem;
6062a8dc 352
5e9d5275
MS
353 if (nsops != 1) {
354 /* Complex operation - acquire a full lock */
355 ipc_lock_object(&sma->sem_perm);
6062a8dc 356
5864a2fd
MS
357 /* Prevent parallel simple ops */
358 complexmode_enter(sma);
359 return SEM_GLOBAL_LOCK;
5e9d5275
MS
360 }
361
362 /*
363 * Only one semaphore affected - try to optimize locking.
5864a2fd
MS
364 * Optimized locking is possible if no complex operation
365 * is either enqueued or processed right now.
366 *
9de5ab8a 367 * Both facts are tracked by use_global_mode.
5e9d5275
MS
368 */
369 sem = sma->sem_base + sops->sem_num;
6062a8dc 370
5864a2fd 371 /*
9de5ab8a 372 * Initial check for use_global_lock. Just an optimization,
5864a2fd
MS
373 * no locking, no memory barrier.
374 */
9de5ab8a 375 if (!sma->use_global_lock) {
6062a8dc 376 /*
5e9d5275
MS
377 * It appears that no complex operation is around.
378 * Acquire the per-semaphore lock.
6062a8dc 379 */
5e9d5275
MS
380 spin_lock(&sem->lock);
381
9de5ab8a
MS
382 /* pairs with smp_store_release() */
383 if (!smp_load_acquire(&sma->use_global_lock)) {
5864a2fd
MS
384 /* fast path successful! */
385 return sops->sem_num;
6062a8dc 386 }
5e9d5275
MS
387 spin_unlock(&sem->lock);
388 }
389
390 /* slow path: acquire the full lock */
391 ipc_lock_object(&sma->sem_perm);
6062a8dc 392
9de5ab8a
MS
393 if (sma->use_global_lock == 0) {
394 /*
395 * The use_global_lock mode ended while we waited for
396 * sma->sem_perm.lock. Thus we must switch to locking
397 * with sem->lock.
398 * Unlike in the fast path, there is no need to recheck
399 * sma->use_global_lock after we have acquired sem->lock:
400 * We own sma->sem_perm.lock, thus use_global_lock cannot
401 * change.
5e9d5275
MS
402 */
403 spin_lock(&sem->lock);
9de5ab8a 404
5e9d5275
MS
405 ipc_unlock_object(&sma->sem_perm);
406 return sops->sem_num;
6062a8dc 407 } else {
9de5ab8a
MS
408 /*
409 * Not a false alarm, thus continue to use the global lock
410 * mode. No need for complexmode_enter(), this was done by
411 * the caller that has set use_global_mode to non-zero.
6062a8dc 412 */
5864a2fd 413 return SEM_GLOBAL_LOCK;
6062a8dc 414 }
6062a8dc
RR
415}
416
417static inline void sem_unlock(struct sem_array *sma, int locknum)
418{
5864a2fd 419 if (locknum == SEM_GLOBAL_LOCK) {
f269f40a 420 unmerge_queues(sma);
5864a2fd 421 complexmode_tryleave(sma);
cf9d5d78 422 ipc_unlock_object(&sma->sem_perm);
6062a8dc
RR
423 } else {
424 struct sem *sem = sma->sem_base + locknum;
425 spin_unlock(&sem->lock);
426 }
6062a8dc
RR
427}
428
3e148c79 429/*
d9a605e4 430 * sem_lock_(check_) routines are called in the paths where the rwsem
3e148c79 431 * is not held.
321310ce
LT
432 *
433 * The caller holds the RCU read lock.
3e148c79 434 */
16df3674
DB
435static inline struct sem_array *sem_obtain_object(struct ipc_namespace *ns, int id)
436{
55b7ae50 437 struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&sem_ids(ns), id);
16df3674
DB
438
439 if (IS_ERR(ipcp))
440 return ERR_CAST(ipcp);
441
442 return container_of(ipcp, struct sem_array, sem_perm);
443}
444
16df3674
DB
445static inline struct sem_array *sem_obtain_object_check(struct ipc_namespace *ns,
446 int id)
447{
448 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&sem_ids(ns), id);
449
450 if (IS_ERR(ipcp))
451 return ERR_CAST(ipcp);
b1ed88b4 452
03f02c76 453 return container_of(ipcp, struct sem_array, sem_perm);
023a5355
ND
454}
455
6ff37972
PP
456static inline void sem_lock_and_putref(struct sem_array *sma)
457{
6062a8dc 458 sem_lock(sma, NULL, -1);
9b24fef9 459 ipc_rcu_putref(sma, sem_rcu_free);
6ff37972
PP
460}
461
7ca7e564
ND
462static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
463{
464 ipc_rmid(&sem_ids(ns), &s->sem_perm);
465}
466
f4566f04
ND
467/**
468 * newary - Create a new semaphore set
469 * @ns: namespace
470 * @params: ptr to the structure that contains key, semflg and nsems
471 *
d9a605e4 472 * Called with sem_ids.rwsem held (as a writer)
f4566f04 473 */
7748dbfa 474static int newary(struct ipc_namespace *ns, struct ipc_params *params)
1da177e4
LT
475{
476 int id;
477 int retval;
478 struct sem_array *sma;
479 int size;
7748dbfa
ND
480 key_t key = params->key;
481 int nsems = params->u.nsems;
482 int semflg = params->flg;
b97e820f 483 int i;
1da177e4
LT
484
485 if (!nsems)
486 return -EINVAL;
e3893534 487 if (ns->used_sems + nsems > ns->sc_semmns)
1da177e4
LT
488 return -ENOSPC;
489
239521f3 490 size = sizeof(*sma) + nsems * sizeof(struct sem);
1da177e4 491 sma = ipc_rcu_alloc(size);
3ab08fe2 492 if (!sma)
1da177e4 493 return -ENOMEM;
3ab08fe2 494
239521f3 495 memset(sma, 0, size);
1da177e4
LT
496
497 sma->sem_perm.mode = (semflg & S_IRWXUGO);
498 sma->sem_perm.key = key;
499
500 sma->sem_perm.security = NULL;
501 retval = security_sem_alloc(sma);
502 if (retval) {
53dad6d3 503 ipc_rcu_putref(sma, ipc_rcu_free);
1da177e4
LT
504 return retval;
505 }
506
1da177e4 507 sma->sem_base = (struct sem *) &sma[1];
b97e820f 508
6062a8dc 509 for (i = 0; i < nsems; i++) {
1a82e9e1
MS
510 INIT_LIST_HEAD(&sma->sem_base[i].pending_alter);
511 INIT_LIST_HEAD(&sma->sem_base[i].pending_const);
6062a8dc
RR
512 spin_lock_init(&sma->sem_base[i].lock);
513 }
b97e820f
MS
514
515 sma->complex_count = 0;
9de5ab8a 516 sma->use_global_lock = USE_GLOBAL_LOCK_HYSTERESIS;
1a82e9e1
MS
517 INIT_LIST_HEAD(&sma->pending_alter);
518 INIT_LIST_HEAD(&sma->pending_const);
4daa28f6 519 INIT_LIST_HEAD(&sma->list_id);
1da177e4
LT
520 sma->sem_nsems = nsems;
521 sma->sem_ctime = get_seconds();
e8577d1f
MS
522
523 id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
524 if (id < 0) {
525 ipc_rcu_putref(sma, sem_rcu_free);
526 return id;
527 }
528 ns->used_sems += nsems;
529
6062a8dc 530 sem_unlock(sma, -1);
6d49dab8 531 rcu_read_unlock();
1da177e4 532
7ca7e564 533 return sma->sem_perm.id;
1da177e4
LT
534}
535
7748dbfa 536
f4566f04 537/*
d9a605e4 538 * Called with sem_ids.rwsem and ipcp locked.
f4566f04 539 */
03f02c76 540static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg)
7748dbfa 541{
03f02c76
ND
542 struct sem_array *sma;
543
544 sma = container_of(ipcp, struct sem_array, sem_perm);
545 return security_sem_associate(sma, semflg);
7748dbfa
ND
546}
547
f4566f04 548/*
d9a605e4 549 * Called with sem_ids.rwsem and ipcp locked.
f4566f04 550 */
03f02c76
ND
551static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
552 struct ipc_params *params)
7748dbfa 553{
03f02c76
ND
554 struct sem_array *sma;
555
556 sma = container_of(ipcp, struct sem_array, sem_perm);
557 if (params->u.nsems > sma->sem_nsems)
7748dbfa
ND
558 return -EINVAL;
559
560 return 0;
561}
562
d5460c99 563SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg)
1da177e4 564{
e3893534 565 struct ipc_namespace *ns;
eb66ec44
MK
566 static const struct ipc_ops sem_ops = {
567 .getnew = newary,
568 .associate = sem_security,
569 .more_checks = sem_more_checks,
570 };
7748dbfa 571 struct ipc_params sem_params;
e3893534
KK
572
573 ns = current->nsproxy->ipc_ns;
1da177e4 574
e3893534 575 if (nsems < 0 || nsems > ns->sc_semmsl)
1da177e4 576 return -EINVAL;
7ca7e564 577
7748dbfa
ND
578 sem_params.key = key;
579 sem_params.flg = semflg;
580 sem_params.u.nsems = nsems;
1da177e4 581
7748dbfa 582 return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
1da177e4
LT
583}
584
78f5009c 585/**
4ce33ec2
DB
586 * perform_atomic_semop[_slow] - Attempt to perform semaphore
587 * operations on a given array.
758a6ba3 588 * @sma: semaphore array
d198cd6d 589 * @q: struct sem_queue that describes the operation
758a6ba3 590 *
4ce33ec2
DB
591 * Caller blocking are as follows, based the value
592 * indicated by the semaphore operation (sem_op):
593 *
594 * (1) >0 never blocks.
595 * (2) 0 (wait-for-zero operation): semval is non-zero.
596 * (3) <0 attempting to decrement semval to a value smaller than zero.
597 *
758a6ba3
MS
598 * Returns 0 if the operation was possible.
599 * Returns 1 if the operation is impossible, the caller must sleep.
4ce33ec2 600 * Returns <0 for error codes.
1da177e4 601 */
4ce33ec2 602static int perform_atomic_semop_slow(struct sem_array *sma, struct sem_queue *q)
1da177e4 603{
d198cd6d 604 int result, sem_op, nsops, pid;
1da177e4 605 struct sembuf *sop;
239521f3 606 struct sem *curr;
d198cd6d
MS
607 struct sembuf *sops;
608 struct sem_undo *un;
609
610 sops = q->sops;
611 nsops = q->nsops;
612 un = q->undo;
1da177e4
LT
613
614 for (sop = sops; sop < sops + nsops; sop++) {
615 curr = sma->sem_base + sop->sem_num;
616 sem_op = sop->sem_op;
617 result = curr->semval;
78f5009c 618
1da177e4
LT
619 if (!sem_op && result)
620 goto would_block;
621
622 result += sem_op;
623 if (result < 0)
624 goto would_block;
625 if (result > SEMVMX)
626 goto out_of_range;
78f5009c 627
1da177e4
LT
628 if (sop->sem_flg & SEM_UNDO) {
629 int undo = un->semadj[sop->sem_num] - sem_op;
78f5009c 630 /* Exceeding the undo range is an error. */
1da177e4
LT
631 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
632 goto out_of_range;
78f5009c 633 un->semadj[sop->sem_num] = undo;
1da177e4 634 }
78f5009c 635
1da177e4
LT
636 curr->semval = result;
637 }
638
639 sop--;
d198cd6d 640 pid = q->pid;
1da177e4
LT
641 while (sop >= sops) {
642 sma->sem_base[sop->sem_num].sempid = pid;
1da177e4
LT
643 sop--;
644 }
78f5009c 645
1da177e4
LT
646 return 0;
647
648out_of_range:
649 result = -ERANGE;
650 goto undo;
651
652would_block:
ed247b7c
MS
653 q->blocking = sop;
654
1da177e4
LT
655 if (sop->sem_flg & IPC_NOWAIT)
656 result = -EAGAIN;
657 else
658 result = 1;
659
660undo:
661 sop--;
662 while (sop >= sops) {
78f5009c
PM
663 sem_op = sop->sem_op;
664 sma->sem_base[sop->sem_num].semval -= sem_op;
665 if (sop->sem_flg & SEM_UNDO)
666 un->semadj[sop->sem_num] += sem_op;
1da177e4
LT
667 sop--;
668 }
669
670 return result;
671}
672
4ce33ec2
DB
673static int perform_atomic_semop(struct sem_array *sma, struct sem_queue *q)
674{
675 int result, sem_op, nsops;
676 struct sembuf *sop;
677 struct sem *curr;
678 struct sembuf *sops;
679 struct sem_undo *un;
680
681 sops = q->sops;
682 nsops = q->nsops;
683 un = q->undo;
684
685 if (unlikely(q->dupsop))
686 return perform_atomic_semop_slow(sma, q);
687
688 /*
689 * We scan the semaphore set twice, first to ensure that the entire
690 * operation can succeed, therefore avoiding any pointless writes
691 * to shared memory and having to undo such changes in order to block
692 * until the operations can go through.
693 */
694 for (sop = sops; sop < sops + nsops; sop++) {
695 curr = sma->sem_base + sop->sem_num;
696 sem_op = sop->sem_op;
697 result = curr->semval;
698
699 if (!sem_op && result)
700 goto would_block; /* wait-for-zero */
701
702 result += sem_op;
703 if (result < 0)
704 goto would_block;
705
706 if (result > SEMVMX)
707 return -ERANGE;
708
709 if (sop->sem_flg & SEM_UNDO) {
710 int undo = un->semadj[sop->sem_num] - sem_op;
711
712 /* Exceeding the undo range is an error. */
713 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
714 return -ERANGE;
715 }
716 }
717
718 for (sop = sops; sop < sops + nsops; sop++) {
719 curr = sma->sem_base + sop->sem_num;
720 sem_op = sop->sem_op;
721 result = curr->semval;
722
723 if (sop->sem_flg & SEM_UNDO) {
724 int undo = un->semadj[sop->sem_num] - sem_op;
725
726 un->semadj[sop->sem_num] = undo;
727 }
728 curr->semval += sem_op;
729 curr->sempid = q->pid;
730 }
731
732 return 0;
733
734would_block:
735 q->blocking = sop;
736 return sop->sem_flg & IPC_NOWAIT ? -EAGAIN : 1;
737}
738
9ae949fa
DB
739static inline void wake_up_sem_queue_prepare(struct sem_queue *q, int error,
740 struct wake_q_head *wake_q)
0a2b9d4c 741{
9ae949fa
DB
742 wake_q_add(wake_q, q->sleeper);
743 /*
744 * Rely on the above implicit barrier, such that we can
745 * ensure that we hold reference to the task before setting
746 * q->status. Otherwise we could race with do_exit if the
747 * task is awoken by an external event before calling
748 * wake_up_process().
749 */
750 WRITE_ONCE(q->status, error);
d4212093
NP
751}
752
b97e820f
MS
753static void unlink_queue(struct sem_array *sma, struct sem_queue *q)
754{
755 list_del(&q->list);
9f1bc2c9 756 if (q->nsops > 1)
b97e820f
MS
757 sma->complex_count--;
758}
759
fd5db422
MS
760/** check_restart(sma, q)
761 * @sma: semaphore array
762 * @q: the operation that just completed
763 *
764 * update_queue is O(N^2) when it restarts scanning the whole queue of
765 * waiting operations. Therefore this function checks if the restart is
766 * really necessary. It is called after a previously waiting operation
1a82e9e1
MS
767 * modified the array.
768 * Note that wait-for-zero operations are handled without restart.
fd5db422 769 */
4663d3e8 770static inline int check_restart(struct sem_array *sma, struct sem_queue *q)
fd5db422 771{
1a82e9e1
MS
772 /* pending complex alter operations are too difficult to analyse */
773 if (!list_empty(&sma->pending_alter))
fd5db422
MS
774 return 1;
775
776 /* we were a sleeping complex operation. Too difficult */
777 if (q->nsops > 1)
778 return 1;
779
1a82e9e1
MS
780 /* It is impossible that someone waits for the new value:
781 * - complex operations always restart.
782 * - wait-for-zero are handled seperately.
783 * - q is a previously sleeping simple operation that
784 * altered the array. It must be a decrement, because
785 * simple increments never sleep.
786 * - If there are older (higher priority) decrements
787 * in the queue, then they have observed the original
788 * semval value and couldn't proceed. The operation
789 * decremented to value - thus they won't proceed either.
790 */
791 return 0;
792}
fd5db422 793
1a82e9e1 794/**
8001c858 795 * wake_const_ops - wake up non-alter tasks
1a82e9e1
MS
796 * @sma: semaphore array.
797 * @semnum: semaphore that was modified.
9ae949fa 798 * @wake_q: lockless wake-queue head.
1a82e9e1
MS
799 *
800 * wake_const_ops must be called after a semaphore in a semaphore array
801 * was set to 0. If complex const operations are pending, wake_const_ops must
802 * be called with semnum = -1, as well as with the number of each modified
803 * semaphore.
9ae949fa 804 * The tasks that must be woken up are added to @wake_q. The return code
1a82e9e1
MS
805 * is stored in q->pid.
806 * The function returns 1 if at least one operation was completed successfully.
807 */
808static int wake_const_ops(struct sem_array *sma, int semnum,
9ae949fa 809 struct wake_q_head *wake_q)
1a82e9e1 810{
f150f02c 811 struct sem_queue *q, *tmp;
1a82e9e1
MS
812 struct list_head *pending_list;
813 int semop_completed = 0;
814
815 if (semnum == -1)
816 pending_list = &sma->pending_const;
817 else
818 pending_list = &sma->sem_base[semnum].pending_const;
fd5db422 819
f150f02c
DB
820 list_for_each_entry_safe(q, tmp, pending_list, list) {
821 int error = perform_atomic_semop(sma, q);
1a82e9e1 822
f150f02c
DB
823 if (error > 0)
824 continue;
825 /* operation completed, remove from queue & wakeup */
826 unlink_queue(sma, q);
1a82e9e1 827
f150f02c
DB
828 wake_up_sem_queue_prepare(q, error, wake_q);
829 if (error == 0)
830 semop_completed = 1;
1a82e9e1 831 }
f150f02c 832
1a82e9e1
MS
833 return semop_completed;
834}
835
836/**
8001c858 837 * do_smart_wakeup_zero - wakeup all wait for zero tasks
1a82e9e1
MS
838 * @sma: semaphore array
839 * @sops: operations that were performed
840 * @nsops: number of operations
9ae949fa 841 * @wake_q: lockless wake-queue head
1a82e9e1 842 *
8001c858
DB
843 * Checks all required queue for wait-for-zero operations, based
844 * on the actual changes that were performed on the semaphore array.
1a82e9e1
MS
845 * The function returns 1 if at least one operation was completed successfully.
846 */
847static int do_smart_wakeup_zero(struct sem_array *sma, struct sembuf *sops,
9ae949fa 848 int nsops, struct wake_q_head *wake_q)
1a82e9e1
MS
849{
850 int i;
851 int semop_completed = 0;
852 int got_zero = 0;
853
854 /* first: the per-semaphore queues, if known */
855 if (sops) {
856 for (i = 0; i < nsops; i++) {
857 int num = sops[i].sem_num;
858
859 if (sma->sem_base[num].semval == 0) {
860 got_zero = 1;
9ae949fa 861 semop_completed |= wake_const_ops(sma, num, wake_q);
1a82e9e1
MS
862 }
863 }
864 } else {
865 /*
866 * No sops means modified semaphores not known.
867 * Assume all were changed.
fd5db422 868 */
1a82e9e1
MS
869 for (i = 0; i < sma->sem_nsems; i++) {
870 if (sma->sem_base[i].semval == 0) {
871 got_zero = 1;
9ae949fa 872 semop_completed |= wake_const_ops(sma, i, wake_q);
1a82e9e1
MS
873 }
874 }
fd5db422
MS
875 }
876 /*
1a82e9e1
MS
877 * If one of the modified semaphores got 0,
878 * then check the global queue, too.
fd5db422 879 */
1a82e9e1 880 if (got_zero)
9ae949fa 881 semop_completed |= wake_const_ops(sma, -1, wake_q);
fd5db422 882
1a82e9e1 883 return semop_completed;
fd5db422
MS
884}
885
636c6be8
MS
886
887/**
8001c858 888 * update_queue - look for tasks that can be completed.
636c6be8
MS
889 * @sma: semaphore array.
890 * @semnum: semaphore that was modified.
9ae949fa 891 * @wake_q: lockless wake-queue head.
636c6be8
MS
892 *
893 * update_queue must be called after a semaphore in a semaphore array
9f1bc2c9
RR
894 * was modified. If multiple semaphores were modified, update_queue must
895 * be called with semnum = -1, as well as with the number of each modified
896 * semaphore.
9ae949fa 897 * The tasks that must be woken up are added to @wake_q. The return code
0a2b9d4c 898 * is stored in q->pid.
1a82e9e1
MS
899 * The function internally checks if const operations can now succeed.
900 *
0a2b9d4c 901 * The function return 1 if at least one semop was completed successfully.
1da177e4 902 */
9ae949fa 903static int update_queue(struct sem_array *sma, int semnum, struct wake_q_head *wake_q)
1da177e4 904{
f150f02c 905 struct sem_queue *q, *tmp;
636c6be8 906 struct list_head *pending_list;
0a2b9d4c 907 int semop_completed = 0;
636c6be8 908
9f1bc2c9 909 if (semnum == -1)
1a82e9e1 910 pending_list = &sma->pending_alter;
9f1bc2c9 911 else
1a82e9e1 912 pending_list = &sma->sem_base[semnum].pending_alter;
9cad200c
NP
913
914again:
f150f02c 915 list_for_each_entry_safe(q, tmp, pending_list, list) {
fd5db422 916 int error, restart;
636c6be8 917
d987f8b2
MS
918 /* If we are scanning the single sop, per-semaphore list of
919 * one semaphore and that semaphore is 0, then it is not
1a82e9e1 920 * necessary to scan further: simple increments
d987f8b2
MS
921 * that affect only one entry succeed immediately and cannot
922 * be in the per semaphore pending queue, and decrements
923 * cannot be successful if the value is already 0.
924 */
1a82e9e1 925 if (semnum != -1 && sma->sem_base[semnum].semval == 0)
d987f8b2
MS
926 break;
927
d198cd6d 928 error = perform_atomic_semop(sma, q);
1da177e4
LT
929
930 /* Does q->sleeper still need to sleep? */
9cad200c
NP
931 if (error > 0)
932 continue;
933
b97e820f 934 unlink_queue(sma, q);
9cad200c 935
0a2b9d4c 936 if (error) {
fd5db422 937 restart = 0;
0a2b9d4c
MS
938 } else {
939 semop_completed = 1;
9ae949fa 940 do_smart_wakeup_zero(sma, q->sops, q->nsops, wake_q);
fd5db422 941 restart = check_restart(sma, q);
0a2b9d4c 942 }
fd5db422 943
9ae949fa 944 wake_up_sem_queue_prepare(q, error, wake_q);
fd5db422 945 if (restart)
9cad200c 946 goto again;
1da177e4 947 }
0a2b9d4c 948 return semop_completed;
1da177e4
LT
949}
950
0e8c6656 951/**
8001c858 952 * set_semotime - set sem_otime
0e8c6656
MS
953 * @sma: semaphore array
954 * @sops: operations that modified the array, may be NULL
955 *
956 * sem_otime is replicated to avoid cache line trashing.
957 * This function sets one instance to the current time.
958 */
959static void set_semotime(struct sem_array *sma, struct sembuf *sops)
960{
961 if (sops == NULL) {
962 sma->sem_base[0].sem_otime = get_seconds();
963 } else {
964 sma->sem_base[sops[0].sem_num].sem_otime =
965 get_seconds();
966 }
967}
968
0a2b9d4c 969/**
8001c858 970 * do_smart_update - optimized update_queue
fd5db422
MS
971 * @sma: semaphore array
972 * @sops: operations that were performed
973 * @nsops: number of operations
0a2b9d4c 974 * @otime: force setting otime
9ae949fa 975 * @wake_q: lockless wake-queue head
fd5db422 976 *
1a82e9e1
MS
977 * do_smart_update() does the required calls to update_queue and wakeup_zero,
978 * based on the actual changes that were performed on the semaphore array.
0a2b9d4c 979 * Note that the function does not do the actual wake-up: the caller is
9ae949fa 980 * responsible for calling wake_up_q().
0a2b9d4c 981 * It is safe to perform this call after dropping all locks.
fd5db422 982 */
0a2b9d4c 983static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops,
9ae949fa 984 int otime, struct wake_q_head *wake_q)
fd5db422
MS
985{
986 int i;
987
9ae949fa 988 otime |= do_smart_wakeup_zero(sma, sops, nsops, wake_q);
1a82e9e1 989
f269f40a
MS
990 if (!list_empty(&sma->pending_alter)) {
991 /* semaphore array uses the global queue - just process it. */
9ae949fa 992 otime |= update_queue(sma, -1, wake_q);
f269f40a
MS
993 } else {
994 if (!sops) {
995 /*
996 * No sops, thus the modified semaphores are not
997 * known. Check all.
998 */
999 for (i = 0; i < sma->sem_nsems; i++)
9ae949fa 1000 otime |= update_queue(sma, i, wake_q);
f269f40a
MS
1001 } else {
1002 /*
1003 * Check the semaphores that were increased:
1004 * - No complex ops, thus all sleeping ops are
1005 * decrease.
1006 * - if we decreased the value, then any sleeping
1007 * semaphore ops wont be able to run: If the
1008 * previous value was too small, then the new
1009 * value will be too small, too.
1010 */
1011 for (i = 0; i < nsops; i++) {
1012 if (sops[i].sem_op > 0) {
1013 otime |= update_queue(sma,
9ae949fa 1014 sops[i].sem_num, wake_q);
f269f40a 1015 }
ab465df9 1016 }
9f1bc2c9 1017 }
fd5db422 1018 }
0e8c6656
MS
1019 if (otime)
1020 set_semotime(sma, sops);
fd5db422
MS
1021}
1022
2f2ed41d 1023/*
b220c57a 1024 * check_qop: Test if a queued operation sleeps on the semaphore semnum
2f2ed41d
MS
1025 */
1026static int check_qop(struct sem_array *sma, int semnum, struct sem_queue *q,
1027 bool count_zero)
1028{
b220c57a 1029 struct sembuf *sop = q->blocking;
2f2ed41d 1030
9b44ee2e
MS
1031 /*
1032 * Linux always (since 0.99.10) reported a task as sleeping on all
1033 * semaphores. This violates SUS, therefore it was changed to the
1034 * standard compliant behavior.
1035 * Give the administrators a chance to notice that an application
1036 * might misbehave because it relies on the Linux behavior.
1037 */
1038 pr_info_once("semctl(GETNCNT/GETZCNT) is since 3.16 Single Unix Specification compliant.\n"
1039 "The task %s (%d) triggered the difference, watch for misbehavior.\n",
1040 current->comm, task_pid_nr(current));
1041
b220c57a
MS
1042 if (sop->sem_num != semnum)
1043 return 0;
2f2ed41d 1044
b220c57a
MS
1045 if (count_zero && sop->sem_op == 0)
1046 return 1;
1047 if (!count_zero && sop->sem_op < 0)
1048 return 1;
1049
1050 return 0;
2f2ed41d
MS
1051}
1052
1da177e4
LT
1053/* The following counts are associated to each semaphore:
1054 * semncnt number of tasks waiting on semval being nonzero
1055 * semzcnt number of tasks waiting on semval being zero
b220c57a
MS
1056 *
1057 * Per definition, a task waits only on the semaphore of the first semop
1058 * that cannot proceed, even if additional operation would block, too.
1da177e4 1059 */
2f2ed41d
MS
1060static int count_semcnt(struct sem_array *sma, ushort semnum,
1061 bool count_zero)
1da177e4 1062{
2f2ed41d 1063 struct list_head *l;
239521f3 1064 struct sem_queue *q;
2f2ed41d 1065 int semcnt;
1da177e4 1066
2f2ed41d
MS
1067 semcnt = 0;
1068 /* First: check the simple operations. They are easy to evaluate */
1069 if (count_zero)
1070 l = &sma->sem_base[semnum].pending_const;
1071 else
1072 l = &sma->sem_base[semnum].pending_alter;
1da177e4 1073
2f2ed41d
MS
1074 list_for_each_entry(q, l, list) {
1075 /* all task on a per-semaphore list sleep on exactly
1076 * that semaphore
1077 */
1078 semcnt++;
ebc2e5e6
RR
1079 }
1080
2f2ed41d 1081 /* Then: check the complex operations. */
1994862d 1082 list_for_each_entry(q, &sma->pending_alter, list) {
2f2ed41d
MS
1083 semcnt += check_qop(sma, semnum, q, count_zero);
1084 }
1085 if (count_zero) {
1086 list_for_each_entry(q, &sma->pending_const, list) {
1087 semcnt += check_qop(sma, semnum, q, count_zero);
1088 }
1994862d 1089 }
2f2ed41d 1090 return semcnt;
1da177e4
LT
1091}
1092
d9a605e4
DB
1093/* Free a semaphore set. freeary() is called with sem_ids.rwsem locked
1094 * as a writer and the spinlock for this semaphore set hold. sem_ids.rwsem
3e148c79 1095 * remains locked on exit.
1da177e4 1096 */
01b8b07a 1097static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
1da177e4 1098{
380af1b3
MS
1099 struct sem_undo *un, *tu;
1100 struct sem_queue *q, *tq;
01b8b07a 1101 struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
9f1bc2c9 1102 int i;
9ae949fa 1103 DEFINE_WAKE_Q(wake_q);
1da177e4 1104
380af1b3 1105 /* Free the existing undo structures for this semaphore set. */
cf9d5d78 1106 ipc_assert_locked_object(&sma->sem_perm);
380af1b3
MS
1107 list_for_each_entry_safe(un, tu, &sma->list_id, list_id) {
1108 list_del(&un->list_id);
1109 spin_lock(&un->ulp->lock);
1da177e4 1110 un->semid = -1;
380af1b3
MS
1111 list_del_rcu(&un->list_proc);
1112 spin_unlock(&un->ulp->lock);
693a8b6e 1113 kfree_rcu(un, rcu);
380af1b3 1114 }
1da177e4
LT
1115
1116 /* Wake up all pending processes and let them fail with EIDRM. */
1a82e9e1
MS
1117 list_for_each_entry_safe(q, tq, &sma->pending_const, list) {
1118 unlink_queue(sma, q);
9ae949fa 1119 wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
1a82e9e1
MS
1120 }
1121
1122 list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
b97e820f 1123 unlink_queue(sma, q);
9ae949fa 1124 wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
1da177e4 1125 }
9f1bc2c9
RR
1126 for (i = 0; i < sma->sem_nsems; i++) {
1127 struct sem *sem = sma->sem_base + i;
1a82e9e1
MS
1128 list_for_each_entry_safe(q, tq, &sem->pending_const, list) {
1129 unlink_queue(sma, q);
9ae949fa 1130 wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
1a82e9e1
MS
1131 }
1132 list_for_each_entry_safe(q, tq, &sem->pending_alter, list) {
9f1bc2c9 1133 unlink_queue(sma, q);
9ae949fa 1134 wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
9f1bc2c9
RR
1135 }
1136 }
1da177e4 1137
7ca7e564
ND
1138 /* Remove the semaphore set from the IDR */
1139 sem_rmid(ns, sma);
6062a8dc 1140 sem_unlock(sma, -1);
6d49dab8 1141 rcu_read_unlock();
1da177e4 1142
9ae949fa 1143 wake_up_q(&wake_q);
e3893534 1144 ns->used_sems -= sma->sem_nsems;
53dad6d3 1145 ipc_rcu_putref(sma, sem_rcu_free);
1da177e4
LT
1146}
1147
1148static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
1149{
239521f3 1150 switch (version) {
1da177e4
LT
1151 case IPC_64:
1152 return copy_to_user(buf, in, sizeof(*in));
1153 case IPC_OLD:
1154 {
1155 struct semid_ds out;
1156
982f7c2b
DR
1157 memset(&out, 0, sizeof(out));
1158
1da177e4
LT
1159 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
1160
1161 out.sem_otime = in->sem_otime;
1162 out.sem_ctime = in->sem_ctime;
1163 out.sem_nsems = in->sem_nsems;
1164
1165 return copy_to_user(buf, &out, sizeof(out));
1166 }
1167 default:
1168 return -EINVAL;
1169 }
1170}
1171
d12e1e50
MS
1172static time_t get_semotime(struct sem_array *sma)
1173{
1174 int i;
1175 time_t res;
1176
1177 res = sma->sem_base[0].sem_otime;
1178 for (i = 1; i < sma->sem_nsems; i++) {
1179 time_t to = sma->sem_base[i].sem_otime;
1180
1181 if (to > res)
1182 res = to;
1183 }
1184 return res;
1185}
1186
4b9fcb0e 1187static int semctl_nolock(struct ipc_namespace *ns, int semid,
e1fd1f49 1188 int cmd, int version, void __user *p)
1da177e4 1189{
e5cc9c7b 1190 int err;
1da177e4
LT
1191 struct sem_array *sma;
1192
239521f3 1193 switch (cmd) {
1da177e4
LT
1194 case IPC_INFO:
1195 case SEM_INFO:
1196 {
1197 struct seminfo seminfo;
1198 int max_id;
1199
1200 err = security_sem_semctl(NULL, cmd);
1201 if (err)
1202 return err;
46c0a8ca 1203
239521f3 1204 memset(&seminfo, 0, sizeof(seminfo));
e3893534
KK
1205 seminfo.semmni = ns->sc_semmni;
1206 seminfo.semmns = ns->sc_semmns;
1207 seminfo.semmsl = ns->sc_semmsl;
1208 seminfo.semopm = ns->sc_semopm;
1da177e4
LT
1209 seminfo.semvmx = SEMVMX;
1210 seminfo.semmnu = SEMMNU;
1211 seminfo.semmap = SEMMAP;
1212 seminfo.semume = SEMUME;
d9a605e4 1213 down_read(&sem_ids(ns).rwsem);
1da177e4 1214 if (cmd == SEM_INFO) {
e3893534
KK
1215 seminfo.semusz = sem_ids(ns).in_use;
1216 seminfo.semaem = ns->used_sems;
1da177e4
LT
1217 } else {
1218 seminfo.semusz = SEMUSZ;
1219 seminfo.semaem = SEMAEM;
1220 }
7ca7e564 1221 max_id = ipc_get_maxid(&sem_ids(ns));
d9a605e4 1222 up_read(&sem_ids(ns).rwsem);
46c0a8ca 1223 if (copy_to_user(p, &seminfo, sizeof(struct seminfo)))
1da177e4 1224 return -EFAULT;
239521f3 1225 return (max_id < 0) ? 0 : max_id;
1da177e4 1226 }
4b9fcb0e 1227 case IPC_STAT:
1da177e4
LT
1228 case SEM_STAT:
1229 {
1230 struct semid64_ds tbuf;
16df3674
DB
1231 int id = 0;
1232
1233 memset(&tbuf, 0, sizeof(tbuf));
1da177e4 1234
941b0304 1235 rcu_read_lock();
4b9fcb0e 1236 if (cmd == SEM_STAT) {
16df3674
DB
1237 sma = sem_obtain_object(ns, semid);
1238 if (IS_ERR(sma)) {
1239 err = PTR_ERR(sma);
1240 goto out_unlock;
1241 }
4b9fcb0e
PP
1242 id = sma->sem_perm.id;
1243 } else {
16df3674
DB
1244 sma = sem_obtain_object_check(ns, semid);
1245 if (IS_ERR(sma)) {
1246 err = PTR_ERR(sma);
1247 goto out_unlock;
1248 }
4b9fcb0e 1249 }
1da177e4
LT
1250
1251 err = -EACCES;
b0e77598 1252 if (ipcperms(ns, &sma->sem_perm, S_IRUGO))
1da177e4
LT
1253 goto out_unlock;
1254
1255 err = security_sem_semctl(sma, cmd);
1256 if (err)
1257 goto out_unlock;
1258
1da177e4 1259 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
d12e1e50
MS
1260 tbuf.sem_otime = get_semotime(sma);
1261 tbuf.sem_ctime = sma->sem_ctime;
1262 tbuf.sem_nsems = sma->sem_nsems;
16df3674 1263 rcu_read_unlock();
e1fd1f49 1264 if (copy_semid_to_user(p, &tbuf, version))
1da177e4
LT
1265 return -EFAULT;
1266 return id;
1267 }
1268 default:
1269 return -EINVAL;
1270 }
1da177e4 1271out_unlock:
16df3674 1272 rcu_read_unlock();
1da177e4
LT
1273 return err;
1274}
1275
e1fd1f49
AV
1276static int semctl_setval(struct ipc_namespace *ns, int semid, int semnum,
1277 unsigned long arg)
1278{
1279 struct sem_undo *un;
1280 struct sem_array *sma;
239521f3 1281 struct sem *curr;
9ae949fa
DB
1282 int err, val;
1283 DEFINE_WAKE_Q(wake_q);
1284
e1fd1f49
AV
1285#if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN)
1286 /* big-endian 64bit */
1287 val = arg >> 32;
1288#else
1289 /* 32bit or little-endian 64bit */
1290 val = arg;
1291#endif
1292
6062a8dc
RR
1293 if (val > SEMVMX || val < 0)
1294 return -ERANGE;
e1fd1f49 1295
6062a8dc
RR
1296 rcu_read_lock();
1297 sma = sem_obtain_object_check(ns, semid);
1298 if (IS_ERR(sma)) {
1299 rcu_read_unlock();
1300 return PTR_ERR(sma);
1301 }
1302
1303 if (semnum < 0 || semnum >= sma->sem_nsems) {
1304 rcu_read_unlock();
1305 return -EINVAL;
1306 }
1307
1308
1309 if (ipcperms(ns, &sma->sem_perm, S_IWUGO)) {
1310 rcu_read_unlock();
1311 return -EACCES;
1312 }
e1fd1f49
AV
1313
1314 err = security_sem_semctl(sma, SETVAL);
6062a8dc
RR
1315 if (err) {
1316 rcu_read_unlock();
1317 return -EACCES;
1318 }
e1fd1f49 1319
6062a8dc 1320 sem_lock(sma, NULL, -1);
e1fd1f49 1321
0f3d2b01 1322 if (!ipc_valid_object(&sma->sem_perm)) {
6e224f94
MS
1323 sem_unlock(sma, -1);
1324 rcu_read_unlock();
1325 return -EIDRM;
1326 }
1327
e1fd1f49
AV
1328 curr = &sma->sem_base[semnum];
1329
cf9d5d78 1330 ipc_assert_locked_object(&sma->sem_perm);
e1fd1f49
AV
1331 list_for_each_entry(un, &sma->list_id, list_id)
1332 un->semadj[semnum] = 0;
1333
1334 curr->semval = val;
1335 curr->sempid = task_tgid_vnr(current);
1336 sma->sem_ctime = get_seconds();
1337 /* maybe some queued-up processes were waiting for this */
9ae949fa 1338 do_smart_update(sma, NULL, 0, 0, &wake_q);
6062a8dc 1339 sem_unlock(sma, -1);
6d49dab8 1340 rcu_read_unlock();
9ae949fa 1341 wake_up_q(&wake_q);
6062a8dc 1342 return 0;
e1fd1f49
AV
1343}
1344
e3893534 1345static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
e1fd1f49 1346 int cmd, void __user *p)
1da177e4
LT
1347{
1348 struct sem_array *sma;
239521f3 1349 struct sem *curr;
16df3674 1350 int err, nsems;
1da177e4 1351 ushort fast_sem_io[SEMMSL_FAST];
239521f3 1352 ushort *sem_io = fast_sem_io;
9ae949fa 1353 DEFINE_WAKE_Q(wake_q);
16df3674
DB
1354
1355 rcu_read_lock();
1356 sma = sem_obtain_object_check(ns, semid);
1357 if (IS_ERR(sma)) {
1358 rcu_read_unlock();
023a5355 1359 return PTR_ERR(sma);
16df3674 1360 }
1da177e4
LT
1361
1362 nsems = sma->sem_nsems;
1363
1da177e4 1364 err = -EACCES;
c728b9c8
LT
1365 if (ipcperms(ns, &sma->sem_perm, cmd == SETALL ? S_IWUGO : S_IRUGO))
1366 goto out_rcu_wakeup;
1da177e4
LT
1367
1368 err = security_sem_semctl(sma, cmd);
c728b9c8
LT
1369 if (err)
1370 goto out_rcu_wakeup;
1da177e4
LT
1371
1372 err = -EACCES;
1373 switch (cmd) {
1374 case GETALL:
1375 {
e1fd1f49 1376 ushort __user *array = p;
1da177e4
LT
1377 int i;
1378
ce857229 1379 sem_lock(sma, NULL, -1);
0f3d2b01 1380 if (!ipc_valid_object(&sma->sem_perm)) {
6e224f94
MS
1381 err = -EIDRM;
1382 goto out_unlock;
1383 }
239521f3 1384 if (nsems > SEMMSL_FAST) {
ce857229 1385 if (!ipc_rcu_getref(sma)) {
ce857229 1386 err = -EIDRM;
6e224f94 1387 goto out_unlock;
ce857229
AV
1388 }
1389 sem_unlock(sma, -1);
6d49dab8 1390 rcu_read_unlock();
1da177e4 1391 sem_io = ipc_alloc(sizeof(ushort)*nsems);
239521f3 1392 if (sem_io == NULL) {
9b24fef9 1393 ipc_rcu_putref(sma, sem_rcu_free);
1da177e4
LT
1394 return -ENOMEM;
1395 }
1396
4091fd94 1397 rcu_read_lock();
6ff37972 1398 sem_lock_and_putref(sma);
0f3d2b01 1399 if (!ipc_valid_object(&sma->sem_perm)) {
1da177e4 1400 err = -EIDRM;
6e224f94 1401 goto out_unlock;
1da177e4 1402 }
ce857229 1403 }
1da177e4
LT
1404 for (i = 0; i < sma->sem_nsems; i++)
1405 sem_io[i] = sma->sem_base[i].semval;
6062a8dc 1406 sem_unlock(sma, -1);
6d49dab8 1407 rcu_read_unlock();
1da177e4 1408 err = 0;
239521f3 1409 if (copy_to_user(array, sem_io, nsems*sizeof(ushort)))
1da177e4
LT
1410 err = -EFAULT;
1411 goto out_free;
1412 }
1413 case SETALL:
1414 {
1415 int i;
1416 struct sem_undo *un;
1417
6062a8dc 1418 if (!ipc_rcu_getref(sma)) {
6e224f94
MS
1419 err = -EIDRM;
1420 goto out_rcu_wakeup;
6062a8dc 1421 }
16df3674 1422 rcu_read_unlock();
1da177e4 1423
239521f3 1424 if (nsems > SEMMSL_FAST) {
1da177e4 1425 sem_io = ipc_alloc(sizeof(ushort)*nsems);
239521f3 1426 if (sem_io == NULL) {
9b24fef9 1427 ipc_rcu_putref(sma, sem_rcu_free);
1da177e4
LT
1428 return -ENOMEM;
1429 }
1430 }
1431
239521f3 1432 if (copy_from_user(sem_io, p, nsems*sizeof(ushort))) {
9b24fef9 1433 ipc_rcu_putref(sma, sem_rcu_free);
1da177e4
LT
1434 err = -EFAULT;
1435 goto out_free;
1436 }
1437
1438 for (i = 0; i < nsems; i++) {
1439 if (sem_io[i] > SEMVMX) {
9b24fef9 1440 ipc_rcu_putref(sma, sem_rcu_free);
1da177e4
LT
1441 err = -ERANGE;
1442 goto out_free;
1443 }
1444 }
4091fd94 1445 rcu_read_lock();
6ff37972 1446 sem_lock_and_putref(sma);
0f3d2b01 1447 if (!ipc_valid_object(&sma->sem_perm)) {
1da177e4 1448 err = -EIDRM;
6e224f94 1449 goto out_unlock;
1da177e4
LT
1450 }
1451
a5f4db87 1452 for (i = 0; i < nsems; i++) {
1da177e4 1453 sma->sem_base[i].semval = sem_io[i];
a5f4db87
DB
1454 sma->sem_base[i].sempid = task_tgid_vnr(current);
1455 }
4daa28f6 1456
cf9d5d78 1457 ipc_assert_locked_object(&sma->sem_perm);
4daa28f6 1458 list_for_each_entry(un, &sma->list_id, list_id) {
1da177e4
LT
1459 for (i = 0; i < nsems; i++)
1460 un->semadj[i] = 0;
4daa28f6 1461 }
1da177e4
LT
1462 sma->sem_ctime = get_seconds();
1463 /* maybe some queued-up processes were waiting for this */
9ae949fa 1464 do_smart_update(sma, NULL, 0, 0, &wake_q);
1da177e4
LT
1465 err = 0;
1466 goto out_unlock;
1467 }
e1fd1f49 1468 /* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */
1da177e4
LT
1469 }
1470 err = -EINVAL;
c728b9c8
LT
1471 if (semnum < 0 || semnum >= nsems)
1472 goto out_rcu_wakeup;
1da177e4 1473
6062a8dc 1474 sem_lock(sma, NULL, -1);
0f3d2b01 1475 if (!ipc_valid_object(&sma->sem_perm)) {
6e224f94
MS
1476 err = -EIDRM;
1477 goto out_unlock;
1478 }
1da177e4
LT
1479 curr = &sma->sem_base[semnum];
1480
1481 switch (cmd) {
1482 case GETVAL:
1483 err = curr->semval;
1484 goto out_unlock;
1485 case GETPID:
1486 err = curr->sempid;
1487 goto out_unlock;
1488 case GETNCNT:
2f2ed41d 1489 err = count_semcnt(sma, semnum, 0);
1da177e4
LT
1490 goto out_unlock;
1491 case GETZCNT:
2f2ed41d 1492 err = count_semcnt(sma, semnum, 1);
1da177e4 1493 goto out_unlock;
1da177e4 1494 }
16df3674 1495
1da177e4 1496out_unlock:
6062a8dc 1497 sem_unlock(sma, -1);
c728b9c8 1498out_rcu_wakeup:
6d49dab8 1499 rcu_read_unlock();
9ae949fa 1500 wake_up_q(&wake_q);
1da177e4 1501out_free:
239521f3 1502 if (sem_io != fast_sem_io)
1d5cfdb0 1503 ipc_free(sem_io);
1da177e4
LT
1504 return err;
1505}
1506
016d7132
PP
1507static inline unsigned long
1508copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
1da177e4 1509{
239521f3 1510 switch (version) {
1da177e4 1511 case IPC_64:
016d7132 1512 if (copy_from_user(out, buf, sizeof(*out)))
1da177e4 1513 return -EFAULT;
1da177e4 1514 return 0;
1da177e4
LT
1515 case IPC_OLD:
1516 {
1517 struct semid_ds tbuf_old;
1518
239521f3 1519 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
1da177e4
LT
1520 return -EFAULT;
1521
016d7132
PP
1522 out->sem_perm.uid = tbuf_old.sem_perm.uid;
1523 out->sem_perm.gid = tbuf_old.sem_perm.gid;
1524 out->sem_perm.mode = tbuf_old.sem_perm.mode;
1da177e4
LT
1525
1526 return 0;
1527 }
1528 default:
1529 return -EINVAL;
1530 }
1531}
1532
522bb2a2 1533/*
d9a605e4 1534 * This function handles some semctl commands which require the rwsem
522bb2a2 1535 * to be held in write mode.
d9a605e4 1536 * NOTE: no locks must be held, the rwsem is taken inside this function.
522bb2a2 1537 */
21a4826a 1538static int semctl_down(struct ipc_namespace *ns, int semid,
e1fd1f49 1539 int cmd, int version, void __user *p)
1da177e4
LT
1540{
1541 struct sem_array *sma;
1542 int err;
016d7132 1543 struct semid64_ds semid64;
1da177e4
LT
1544 struct kern_ipc_perm *ipcp;
1545
239521f3 1546 if (cmd == IPC_SET) {
e1fd1f49 1547 if (copy_semid_from_user(&semid64, p, version))
1da177e4 1548 return -EFAULT;
1da177e4 1549 }
073115d6 1550
d9a605e4 1551 down_write(&sem_ids(ns).rwsem);
7b4cc5d8
DB
1552 rcu_read_lock();
1553
16df3674
DB
1554 ipcp = ipcctl_pre_down_nolock(ns, &sem_ids(ns), semid, cmd,
1555 &semid64.sem_perm, 0);
7b4cc5d8
DB
1556 if (IS_ERR(ipcp)) {
1557 err = PTR_ERR(ipcp);
7b4cc5d8
DB
1558 goto out_unlock1;
1559 }
073115d6 1560
a5f75e7f 1561 sma = container_of(ipcp, struct sem_array, sem_perm);
1da177e4
LT
1562
1563 err = security_sem_semctl(sma, cmd);
7b4cc5d8
DB
1564 if (err)
1565 goto out_unlock1;
1da177e4 1566
7b4cc5d8 1567 switch (cmd) {
1da177e4 1568 case IPC_RMID:
6062a8dc 1569 sem_lock(sma, NULL, -1);
7b4cc5d8 1570 /* freeary unlocks the ipc object and rcu */
01b8b07a 1571 freeary(ns, ipcp);
522bb2a2 1572 goto out_up;
1da177e4 1573 case IPC_SET:
6062a8dc 1574 sem_lock(sma, NULL, -1);
1efdb69b
EB
1575 err = ipc_update_perm(&semid64.sem_perm, ipcp);
1576 if (err)
7b4cc5d8 1577 goto out_unlock0;
1da177e4 1578 sma->sem_ctime = get_seconds();
1da177e4
LT
1579 break;
1580 default:
1da177e4 1581 err = -EINVAL;
7b4cc5d8 1582 goto out_unlock1;
1da177e4 1583 }
1da177e4 1584
7b4cc5d8 1585out_unlock0:
6062a8dc 1586 sem_unlock(sma, -1);
7b4cc5d8 1587out_unlock1:
6d49dab8 1588 rcu_read_unlock();
522bb2a2 1589out_up:
d9a605e4 1590 up_write(&sem_ids(ns).rwsem);
1da177e4
LT
1591 return err;
1592}
1593
e1fd1f49 1594SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, unsigned long, arg)
1da177e4 1595{
1da177e4 1596 int version;
e3893534 1597 struct ipc_namespace *ns;
e1fd1f49 1598 void __user *p = (void __user *)arg;
1da177e4
LT
1599
1600 if (semid < 0)
1601 return -EINVAL;
1602
1603 version = ipc_parse_version(&cmd);
e3893534 1604 ns = current->nsproxy->ipc_ns;
1da177e4 1605
239521f3 1606 switch (cmd) {
1da177e4
LT
1607 case IPC_INFO:
1608 case SEM_INFO:
4b9fcb0e 1609 case IPC_STAT:
1da177e4 1610 case SEM_STAT:
e1fd1f49 1611 return semctl_nolock(ns, semid, cmd, version, p);
1da177e4
LT
1612 case GETALL:
1613 case GETVAL:
1614 case GETPID:
1615 case GETNCNT:
1616 case GETZCNT:
1da177e4 1617 case SETALL:
e1fd1f49
AV
1618 return semctl_main(ns, semid, semnum, cmd, p);
1619 case SETVAL:
1620 return semctl_setval(ns, semid, semnum, arg);
1da177e4
LT
1621 case IPC_RMID:
1622 case IPC_SET:
e1fd1f49 1623 return semctl_down(ns, semid, cmd, version, p);
1da177e4
LT
1624 default:
1625 return -EINVAL;
1626 }
1627}
1628
1da177e4
LT
1629/* If the task doesn't already have a undo_list, then allocate one
1630 * here. We guarantee there is only one thread using this undo list,
1631 * and current is THE ONE
1632 *
1633 * If this allocation and assignment succeeds, but later
1634 * portions of this code fail, there is no need to free the sem_undo_list.
1635 * Just let it stay associated with the task, and it'll be freed later
1636 * at exit time.
1637 *
1638 * This can block, so callers must hold no locks.
1639 */
1640static inline int get_undo_list(struct sem_undo_list **undo_listp)
1641{
1642 struct sem_undo_list *undo_list;
1da177e4
LT
1643
1644 undo_list = current->sysvsem.undo_list;
1645 if (!undo_list) {
2453a306 1646 undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
1da177e4
LT
1647 if (undo_list == NULL)
1648 return -ENOMEM;
00a5dfdb 1649 spin_lock_init(&undo_list->lock);
1da177e4 1650 atomic_set(&undo_list->refcnt, 1);
4daa28f6
MS
1651 INIT_LIST_HEAD(&undo_list->list_proc);
1652
1da177e4
LT
1653 current->sysvsem.undo_list = undo_list;
1654 }
1655 *undo_listp = undo_list;
1656 return 0;
1657}
1658
bf17bb71 1659static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid)
1da177e4 1660{
bf17bb71 1661 struct sem_undo *un;
4daa28f6 1662
bf17bb71
NP
1663 list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) {
1664 if (un->semid == semid)
1665 return un;
1da177e4 1666 }
4daa28f6 1667 return NULL;
1da177e4
LT
1668}
1669
bf17bb71
NP
1670static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1671{
1672 struct sem_undo *un;
1673
239521f3 1674 assert_spin_locked(&ulp->lock);
bf17bb71
NP
1675
1676 un = __lookup_undo(ulp, semid);
1677 if (un) {
1678 list_del_rcu(&un->list_proc);
1679 list_add_rcu(&un->list_proc, &ulp->list_proc);
1680 }
1681 return un;
1682}
1683
4daa28f6 1684/**
8001c858 1685 * find_alloc_undo - lookup (and if not present create) undo array
4daa28f6
MS
1686 * @ns: namespace
1687 * @semid: semaphore array id
1688 *
1689 * The function looks up (and if not present creates) the undo structure.
1690 * The size of the undo structure depends on the size of the semaphore
1691 * array, thus the alloc path is not that straightforward.
380af1b3
MS
1692 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1693 * performs a rcu_read_lock().
4daa28f6
MS
1694 */
1695static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
1da177e4
LT
1696{
1697 struct sem_array *sma;
1698 struct sem_undo_list *ulp;
1699 struct sem_undo *un, *new;
6062a8dc 1700 int nsems, error;
1da177e4
LT
1701
1702 error = get_undo_list(&ulp);
1703 if (error)
1704 return ERR_PTR(error);
1705
380af1b3 1706 rcu_read_lock();
c530c6ac 1707 spin_lock(&ulp->lock);
1da177e4 1708 un = lookup_undo(ulp, semid);
c530c6ac 1709 spin_unlock(&ulp->lock);
239521f3 1710 if (likely(un != NULL))
1da177e4
LT
1711 goto out;
1712
1713 /* no undo structure around - allocate one. */
4daa28f6 1714 /* step 1: figure out the size of the semaphore array */
16df3674
DB
1715 sma = sem_obtain_object_check(ns, semid);
1716 if (IS_ERR(sma)) {
1717 rcu_read_unlock();
4de85cd6 1718 return ERR_CAST(sma);
16df3674 1719 }
023a5355 1720
1da177e4 1721 nsems = sma->sem_nsems;
6062a8dc
RR
1722 if (!ipc_rcu_getref(sma)) {
1723 rcu_read_unlock();
1724 un = ERR_PTR(-EIDRM);
1725 goto out;
1726 }
16df3674 1727 rcu_read_unlock();
1da177e4 1728
4daa28f6 1729 /* step 2: allocate new undo structure */
4668edc3 1730 new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1da177e4 1731 if (!new) {
9b24fef9 1732 ipc_rcu_putref(sma, sem_rcu_free);
1da177e4
LT
1733 return ERR_PTR(-ENOMEM);
1734 }
1da177e4 1735
380af1b3 1736 /* step 3: Acquire the lock on semaphore array */
4091fd94 1737 rcu_read_lock();
6ff37972 1738 sem_lock_and_putref(sma);
0f3d2b01 1739 if (!ipc_valid_object(&sma->sem_perm)) {
6062a8dc 1740 sem_unlock(sma, -1);
6d49dab8 1741 rcu_read_unlock();
1da177e4
LT
1742 kfree(new);
1743 un = ERR_PTR(-EIDRM);
1744 goto out;
1745 }
380af1b3
MS
1746 spin_lock(&ulp->lock);
1747
1748 /*
1749 * step 4: check for races: did someone else allocate the undo struct?
1750 */
1751 un = lookup_undo(ulp, semid);
1752 if (un) {
1753 kfree(new);
1754 goto success;
1755 }
4daa28f6
MS
1756 /* step 5: initialize & link new undo structure */
1757 new->semadj = (short *) &new[1];
380af1b3 1758 new->ulp = ulp;
4daa28f6
MS
1759 new->semid = semid;
1760 assert_spin_locked(&ulp->lock);
380af1b3 1761 list_add_rcu(&new->list_proc, &ulp->list_proc);
cf9d5d78 1762 ipc_assert_locked_object(&sma->sem_perm);
4daa28f6 1763 list_add(&new->list_id, &sma->list_id);
380af1b3 1764 un = new;
4daa28f6 1765
380af1b3 1766success:
c530c6ac 1767 spin_unlock(&ulp->lock);
6062a8dc 1768 sem_unlock(sma, -1);
1da177e4
LT
1769out:
1770 return un;
1771}
1772
d5460c99
HC
1773SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
1774 unsigned, nsops, const struct timespec __user *, timeout)
1da177e4
LT
1775{
1776 int error = -EINVAL;
1777 struct sem_array *sma;
1778 struct sembuf fast_sops[SEMOPM_FAST];
239521f3 1779 struct sembuf *sops = fast_sops, *sop;
1da177e4 1780 struct sem_undo *un;
4ce33ec2
DB
1781 int max, locknum;
1782 bool undos = false, alter = false, dupsop = false;
1da177e4 1783 struct sem_queue queue;
4ce33ec2 1784 unsigned long dup = 0, jiffies_left = 0;
e3893534
KK
1785 struct ipc_namespace *ns;
1786
1787 ns = current->nsproxy->ipc_ns;
1da177e4
LT
1788
1789 if (nsops < 1 || semid < 0)
1790 return -EINVAL;
e3893534 1791 if (nsops > ns->sc_semopm)
1da177e4 1792 return -E2BIG;
239521f3
MS
1793 if (nsops > SEMOPM_FAST) {
1794 sops = kmalloc(sizeof(*sops)*nsops, GFP_KERNEL);
1795 if (sops == NULL)
1da177e4
LT
1796 return -ENOMEM;
1797 }
4ce33ec2 1798
239521f3
MS
1799 if (copy_from_user(sops, tsops, nsops * sizeof(*tsops))) {
1800 error = -EFAULT;
1da177e4
LT
1801 goto out_free;
1802 }
4ce33ec2 1803
1da177e4
LT
1804 if (timeout) {
1805 struct timespec _timeout;
1806 if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1807 error = -EFAULT;
1808 goto out_free;
1809 }
1810 if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1811 _timeout.tv_nsec >= 1000000000L) {
1812 error = -EINVAL;
1813 goto out_free;
1814 }
1815 jiffies_left = timespec_to_jiffies(&_timeout);
1816 }
4ce33ec2 1817
1da177e4
LT
1818 max = 0;
1819 for (sop = sops; sop < sops + nsops; sop++) {
4ce33ec2
DB
1820 unsigned long mask = 1ULL << ((sop->sem_num) % BITS_PER_LONG);
1821
1da177e4
LT
1822 if (sop->sem_num >= max)
1823 max = sop->sem_num;
1824 if (sop->sem_flg & SEM_UNDO)
4ce33ec2
DB
1825 undos = true;
1826 if (dup & mask) {
1827 /*
1828 * There was a previous alter access that appears
1829 * to have accessed the same semaphore, thus use
1830 * the dupsop logic. "appears", because the detection
1831 * can only check % BITS_PER_LONG.
1832 */
1833 dupsop = true;
1834 }
1835 if (sop->sem_op != 0) {
1836 alter = true;
1837 dup |= mask;
1838 }
1da177e4 1839 }
1da177e4 1840
1da177e4 1841 if (undos) {
6062a8dc 1842 /* On success, find_alloc_undo takes the rcu_read_lock */
4daa28f6 1843 un = find_alloc_undo(ns, semid);
1da177e4
LT
1844 if (IS_ERR(un)) {
1845 error = PTR_ERR(un);
1846 goto out_free;
1847 }
6062a8dc 1848 } else {
1da177e4 1849 un = NULL;
6062a8dc
RR
1850 rcu_read_lock();
1851 }
1da177e4 1852
16df3674 1853 sma = sem_obtain_object_check(ns, semid);
023a5355 1854 if (IS_ERR(sma)) {
6062a8dc 1855 rcu_read_unlock();
023a5355 1856 error = PTR_ERR(sma);
1da177e4 1857 goto out_free;
023a5355
ND
1858 }
1859
16df3674 1860 error = -EFBIG;
248e7357
DB
1861 if (max >= sma->sem_nsems) {
1862 rcu_read_unlock();
1863 goto out_free;
1864 }
16df3674
DB
1865
1866 error = -EACCES;
248e7357
DB
1867 if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO)) {
1868 rcu_read_unlock();
1869 goto out_free;
1870 }
16df3674
DB
1871
1872 error = security_sem_semop(sma, sops, nsops, alter);
248e7357
DB
1873 if (error) {
1874 rcu_read_unlock();
1875 goto out_free;
1876 }
16df3674 1877
6e224f94
MS
1878 error = -EIDRM;
1879 locknum = sem_lock(sma, sops, nsops);
0f3d2b01
RA
1880 /*
1881 * We eventually might perform the following check in a lockless
1882 * fashion, considering ipc_valid_object() locking constraints.
1883 * If nsops == 1 and there is no contention for sem_perm.lock, then
1884 * only a per-semaphore lock is held and it's OK to proceed with the
1885 * check below. More details on the fine grained locking scheme
1886 * entangled here and why it's RMID race safe on comments at sem_lock()
1887 */
1888 if (!ipc_valid_object(&sma->sem_perm))
6e224f94 1889 goto out_unlock_free;
1da177e4 1890 /*
4daa28f6 1891 * semid identifiers are not unique - find_alloc_undo may have
1da177e4 1892 * allocated an undo structure, it was invalidated by an RMID
4daa28f6 1893 * and now a new array with received the same id. Check and fail.
25985edc 1894 * This case can be detected checking un->semid. The existence of
380af1b3 1895 * "un" itself is guaranteed by rcu.
1da177e4 1896 */
6062a8dc
RR
1897 if (un && un->semid == -1)
1898 goto out_unlock_free;
4daa28f6 1899
d198cd6d
MS
1900 queue.sops = sops;
1901 queue.nsops = nsops;
1902 queue.undo = un;
1903 queue.pid = task_tgid_vnr(current);
1904 queue.alter = alter;
4ce33ec2 1905 queue.dupsop = dupsop;
d198cd6d
MS
1906
1907 error = perform_atomic_semop(sma, &queue);
9ae949fa
DB
1908 if (error == 0) { /* non-blocking succesfull path */
1909 DEFINE_WAKE_Q(wake_q);
1910
1911 /*
1912 * If the operation was successful, then do
0e8c6656
MS
1913 * the required updates.
1914 */
1915 if (alter)
9ae949fa 1916 do_smart_update(sma, sops, nsops, 1, &wake_q);
0e8c6656
MS
1917 else
1918 set_semotime(sma, sops);
9ae949fa
DB
1919
1920 sem_unlock(sma, locknum);
1921 rcu_read_unlock();
1922 wake_up_q(&wake_q);
1923
1924 goto out_free;
1da177e4 1925 }
9ae949fa 1926 if (error < 0) /* non-blocking error path */
0e8c6656 1927 goto out_unlock_free;
1da177e4 1928
9ae949fa
DB
1929 /*
1930 * We need to sleep on this operation, so we put the current
1da177e4
LT
1931 * task into the pending queue and go to sleep.
1932 */
b97e820f
MS
1933 if (nsops == 1) {
1934 struct sem *curr;
1935 curr = &sma->sem_base[sops->sem_num];
1936
f269f40a
MS
1937 if (alter) {
1938 if (sma->complex_count) {
1939 list_add_tail(&queue.list,
1940 &sma->pending_alter);
1941 } else {
1942
1943 list_add_tail(&queue.list,
1944 &curr->pending_alter);
1945 }
1946 } else {
1a82e9e1 1947 list_add_tail(&queue.list, &curr->pending_const);
f269f40a 1948 }
b97e820f 1949 } else {
f269f40a
MS
1950 if (!sma->complex_count)
1951 merge_queues(sma);
1952
9f1bc2c9 1953 if (alter)
1a82e9e1 1954 list_add_tail(&queue.list, &sma->pending_alter);
9f1bc2c9 1955 else
1a82e9e1
MS
1956 list_add_tail(&queue.list, &sma->pending_const);
1957
b97e820f
MS
1958 sma->complex_count++;
1959 }
1960
b5fa01a2
DB
1961 do {
1962 queue.status = -EINTR;
1963 queue.sleeper = current;
0b0577f6 1964
b5fa01a2
DB
1965 __set_current_state(TASK_INTERRUPTIBLE);
1966 sem_unlock(sma, locknum);
1967 rcu_read_unlock();
1da177e4 1968
b5fa01a2
DB
1969 if (timeout)
1970 jiffies_left = schedule_timeout(jiffies_left);
1971 else
1972 schedule();
1da177e4 1973
9ae949fa 1974 /*
b5fa01a2
DB
1975 * fastpath: the semop has completed, either successfully or
1976 * not, from the syscall pov, is quite irrelevant to us at this
1977 * point; we're done.
1978 *
1979 * We _do_ care, nonetheless, about being awoken by a signal or
1980 * spuriously. The queue.status is checked again in the
1981 * slowpath (aka after taking sem_lock), such that we can detect
1982 * scenarios where we were awakened externally, during the
1983 * window between wake_q_add() and wake_up_q().
c61284e9 1984 */
b5fa01a2
DB
1985 error = READ_ONCE(queue.status);
1986 if (error != -EINTR) {
1987 /*
1988 * User space could assume that semop() is a memory
1989 * barrier: Without the mb(), the cpu could
1990 * speculatively read in userspace stale data that was
1991 * overwritten by the previous owner of the semaphore.
1992 */
1993 smp_mb();
1994 goto out_free;
1995 }
d694ad62 1996
b5fa01a2 1997 rcu_read_lock();
c626bc46 1998 locknum = sem_lock(sma, sops, nsops);
1da177e4 1999
370b262c
DB
2000 if (!ipc_valid_object(&sma->sem_perm))
2001 goto out_unlock_free;
2002
2003 error = READ_ONCE(queue.status);
1da177e4 2004
b5fa01a2
DB
2005 /*
2006 * If queue.status != -EINTR we are woken up by another process.
2007 * Leave without unlink_queue(), but with sem_unlock().
2008 */
2009 if (error != -EINTR)
2010 goto out_unlock_free;
0b0577f6 2011
b5fa01a2
DB
2012 /*
2013 * If an interrupt occurred we have to clean up the queue.
2014 */
2015 if (timeout && jiffies_left == 0)
2016 error = -EAGAIN;
2017 } while (error == -EINTR && !signal_pending(current)); /* spurious */
0b0577f6 2018
b97e820f 2019 unlink_queue(sma, &queue);
1da177e4
LT
2020
2021out_unlock_free:
6062a8dc 2022 sem_unlock(sma, locknum);
6d49dab8 2023 rcu_read_unlock();
1da177e4 2024out_free:
239521f3 2025 if (sops != fast_sops)
1da177e4
LT
2026 kfree(sops);
2027 return error;
2028}
2029
d5460c99
HC
2030SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
2031 unsigned, nsops)
1da177e4
LT
2032{
2033 return sys_semtimedop(semid, tsops, nsops, NULL);
2034}
2035
2036/* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
2037 * parent and child tasks.
1da177e4
LT
2038 */
2039
2040int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
2041{
2042 struct sem_undo_list *undo_list;
2043 int error;
2044
2045 if (clone_flags & CLONE_SYSVSEM) {
2046 error = get_undo_list(&undo_list);
2047 if (error)
2048 return error;
1da177e4
LT
2049 atomic_inc(&undo_list->refcnt);
2050 tsk->sysvsem.undo_list = undo_list;
46c0a8ca 2051 } else
1da177e4
LT
2052 tsk->sysvsem.undo_list = NULL;
2053
2054 return 0;
2055}
2056
2057/*
2058 * add semadj values to semaphores, free undo structures.
2059 * undo structures are not freed when semaphore arrays are destroyed
2060 * so some of them may be out of date.
2061 * IMPLEMENTATION NOTE: There is some confusion over whether the
2062 * set of adjustments that needs to be done should be done in an atomic
2063 * manner or not. That is, if we are attempting to decrement the semval
2064 * should we queue up and wait until we can do so legally?
2065 * The original implementation attempted to do this (queue and wait).
2066 * The current implementation does not do so. The POSIX standard
2067 * and SVID should be consulted to determine what behavior is mandated.
2068 */
2069void exit_sem(struct task_struct *tsk)
2070{
4daa28f6 2071 struct sem_undo_list *ulp;
1da177e4 2072
4daa28f6
MS
2073 ulp = tsk->sysvsem.undo_list;
2074 if (!ulp)
1da177e4 2075 return;
9edff4ab 2076 tsk->sysvsem.undo_list = NULL;
1da177e4 2077
4daa28f6 2078 if (!atomic_dec_and_test(&ulp->refcnt))
1da177e4
LT
2079 return;
2080
380af1b3 2081 for (;;) {
1da177e4 2082 struct sem_array *sma;
380af1b3 2083 struct sem_undo *un;
6062a8dc 2084 int semid, i;
9ae949fa 2085 DEFINE_WAKE_Q(wake_q);
4daa28f6 2086
2a1613a5
NB
2087 cond_resched();
2088
380af1b3 2089 rcu_read_lock();
05725f7e
JP
2090 un = list_entry_rcu(ulp->list_proc.next,
2091 struct sem_undo, list_proc);
602b8593
HK
2092 if (&un->list_proc == &ulp->list_proc) {
2093 /*
2094 * We must wait for freeary() before freeing this ulp,
2095 * in case we raced with last sem_undo. There is a small
2096 * possibility where we exit while freeary() didn't
2097 * finish unlocking sem_undo_list.
2098 */
2099 spin_unlock_wait(&ulp->lock);
2100 rcu_read_unlock();
2101 break;
2102 }
2103 spin_lock(&ulp->lock);
2104 semid = un->semid;
2105 spin_unlock(&ulp->lock);
4daa28f6 2106
602b8593 2107 /* exit_sem raced with IPC_RMID, nothing to do */
6062a8dc
RR
2108 if (semid == -1) {
2109 rcu_read_unlock();
602b8593 2110 continue;
6062a8dc 2111 }
1da177e4 2112
602b8593 2113 sma = sem_obtain_object_check(tsk->nsproxy->ipc_ns, semid);
380af1b3 2114 /* exit_sem raced with IPC_RMID, nothing to do */
6062a8dc
RR
2115 if (IS_ERR(sma)) {
2116 rcu_read_unlock();
380af1b3 2117 continue;
6062a8dc 2118 }
1da177e4 2119
6062a8dc 2120 sem_lock(sma, NULL, -1);
6e224f94 2121 /* exit_sem raced with IPC_RMID, nothing to do */
0f3d2b01 2122 if (!ipc_valid_object(&sma->sem_perm)) {
6e224f94
MS
2123 sem_unlock(sma, -1);
2124 rcu_read_unlock();
2125 continue;
2126 }
bf17bb71 2127 un = __lookup_undo(ulp, semid);
380af1b3
MS
2128 if (un == NULL) {
2129 /* exit_sem raced with IPC_RMID+semget() that created
2130 * exactly the same semid. Nothing to do.
2131 */
6062a8dc 2132 sem_unlock(sma, -1);
6d49dab8 2133 rcu_read_unlock();
380af1b3
MS
2134 continue;
2135 }
2136
2137 /* remove un from the linked lists */
cf9d5d78 2138 ipc_assert_locked_object(&sma->sem_perm);
4daa28f6
MS
2139 list_del(&un->list_id);
2140
a9795584
HK
2141 /* we are the last process using this ulp, acquiring ulp->lock
2142 * isn't required. Besides that, we are also protected against
2143 * IPC_RMID as we hold sma->sem_perm lock now
2144 */
380af1b3 2145 list_del_rcu(&un->list_proc);
380af1b3 2146
4daa28f6
MS
2147 /* perform adjustments registered in un */
2148 for (i = 0; i < sma->sem_nsems; i++) {
239521f3 2149 struct sem *semaphore = &sma->sem_base[i];
4daa28f6
MS
2150 if (un->semadj[i]) {
2151 semaphore->semval += un->semadj[i];
1da177e4
LT
2152 /*
2153 * Range checks of the new semaphore value,
2154 * not defined by sus:
2155 * - Some unices ignore the undo entirely
2156 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
2157 * - some cap the value (e.g. FreeBSD caps
2158 * at 0, but doesn't enforce SEMVMX)
2159 *
2160 * Linux caps the semaphore value, both at 0
2161 * and at SEMVMX.
2162 *
239521f3 2163 * Manfred <manfred@colorfullife.com>
1da177e4 2164 */
5f921ae9
IM
2165 if (semaphore->semval < 0)
2166 semaphore->semval = 0;
2167 if (semaphore->semval > SEMVMX)
2168 semaphore->semval = SEMVMX;
b488893a 2169 semaphore->sempid = task_tgid_vnr(current);
1da177e4
LT
2170 }
2171 }
1da177e4 2172 /* maybe some queued-up processes were waiting for this */
9ae949fa 2173 do_smart_update(sma, NULL, 0, 1, &wake_q);
6062a8dc 2174 sem_unlock(sma, -1);
6d49dab8 2175 rcu_read_unlock();
9ae949fa 2176 wake_up_q(&wake_q);
380af1b3 2177
693a8b6e 2178 kfree_rcu(un, rcu);
1da177e4 2179 }
4daa28f6 2180 kfree(ulp);
1da177e4
LT
2181}
2182
2183#ifdef CONFIG_PROC_FS
19b4946c 2184static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
1da177e4 2185{
1efdb69b 2186 struct user_namespace *user_ns = seq_user_ns(s);
19b4946c 2187 struct sem_array *sma = it;
d12e1e50
MS
2188 time_t sem_otime;
2189
d8c63376
MS
2190 /*
2191 * The proc interface isn't aware of sem_lock(), it calls
2192 * ipc_lock_object() directly (in sysvipc_find_ipc).
5864a2fd
MS
2193 * In order to stay compatible with sem_lock(), we must
2194 * enter / leave complex_mode.
d8c63376 2195 */
5864a2fd 2196 complexmode_enter(sma);
d8c63376 2197
d12e1e50 2198 sem_otime = get_semotime(sma);
19b4946c 2199
7f032d6e
JP
2200 seq_printf(s,
2201 "%10d %10d %4o %10u %5u %5u %5u %5u %10lu %10lu\n",
2202 sma->sem_perm.key,
2203 sma->sem_perm.id,
2204 sma->sem_perm.mode,
2205 sma->sem_nsems,
2206 from_kuid_munged(user_ns, sma->sem_perm.uid),
2207 from_kgid_munged(user_ns, sma->sem_perm.gid),
2208 from_kuid_munged(user_ns, sma->sem_perm.cuid),
2209 from_kgid_munged(user_ns, sma->sem_perm.cgid),
2210 sem_otime,
2211 sma->sem_ctime);
2212
5864a2fd
MS
2213 complexmode_tryleave(sma);
2214
7f032d6e 2215 return 0;
1da177e4
LT
2216}
2217#endif