Loading...
1/*
2 * linux/ipc/sem.c
3 * Copyright (C) 1992 Krishna Balasubramanian
4 * Copyright (C) 1995 Eric Schenk, Bruno Haible
5 *
6 * /proc/sysvipc/sem support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
7 *
8 * SMP-threaded, sysctl's added
9 * (c) 1999 Manfred Spraul <manfred@colorfullife.com>
10 * Enforced range limit on SEM_UNDO
11 * (c) 2001 Red Hat Inc
12 * Lockless wakeup
13 * (c) 2003 Manfred Spraul <manfred@colorfullife.com>
14 * Further wakeup optimizations, documentation
15 * (c) 2010 Manfred Spraul <manfred@colorfullife.com>
16 *
17 * support for audit of ipc object properties and permission changes
18 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
19 *
20 * namespaces support
21 * OpenVZ, SWsoft Inc.
22 * Pavel Emelianov <xemul@openvz.org>
23 *
24 * Implementation notes: (May 2010)
25 * This file implements System V semaphores.
26 *
27 * User space visible behavior:
28 * - FIFO ordering for semop() operations (just FIFO, not starvation
29 * protection)
30 * - multiple semaphore operations that alter the same semaphore in
31 * one semop() are handled.
32 * - sem_ctime (time of last semctl()) is updated in the IPC_SET, SETVAL and
33 * SETALL calls.
34 * - two Linux specific semctl() commands: SEM_STAT, SEM_INFO.
35 * - undo adjustments at process exit are limited to 0..SEMVMX.
36 * - namespace are supported.
37 * - SEMMSL, SEMMNS, SEMOPM and SEMMNI can be configured at runtine by writing
38 * to /proc/sys/kernel/sem.
39 * - statistics about the usage are reported in /proc/sysvipc/sem.
40 *
41 * Internals:
42 * - scalability:
43 * - all global variables are read-mostly.
44 * - semop() calls and semctl(RMID) are synchronized by RCU.
45 * - most operations do write operations (actually: spin_lock calls) to
46 * the per-semaphore array structure.
47 * Thus: Perfect SMP scaling between independent semaphore arrays.
48 * If multiple semaphores in one array are used, then cache line
49 * trashing on the semaphore array spinlock will limit the scaling.
50 * - semncnt and semzcnt are calculated on demand in count_semncnt() and
51 * count_semzcnt()
52 * - the task that performs a successful semop() scans the list of all
53 * sleeping tasks and completes any pending operations that can be fulfilled.
54 * Semaphores are actively given to waiting tasks (necessary for FIFO).
55 * (see update_queue())
56 * - To improve the scalability, the actual wake-up calls are performed after
57 * dropping all locks. (see wake_up_sem_queue_prepare(),
58 * wake_up_sem_queue_do())
59 * - All work is done by the waker, the woken up task does not have to do
60 * anything - not even acquiring a lock or dropping a refcount.
61 * - A woken up task may not even touch the semaphore array anymore, it may
62 * have been destroyed already by a semctl(RMID).
63 * - The synchronizations between wake-ups due to a timeout/signal and a
64 * wake-up due to a completed semaphore operation is achieved by using an
65 * intermediate state (IN_WAKEUP).
66 * - UNDO values are stored in an array (one per process and per
67 * semaphore array, lazily allocated). For backwards compatibility, multiple
68 * modes for the UNDO variables are supported (per process, per thread)
69 * (see copy_semundo, CLONE_SYSVSEM)
70 * - There are two lists of the pending operations: a per-array list
71 * and per-semaphore list (stored in the array). This allows to achieve FIFO
72 * ordering without always scanning all pending operations.
73 * The worst-case behavior is nevertheless O(N^2) for N wakeups.
74 */
75
76#include <linux/slab.h>
77#include <linux/spinlock.h>
78#include <linux/init.h>
79#include <linux/proc_fs.h>
80#include <linux/time.h>
81#include <linux/security.h>
82#include <linux/syscalls.h>
83#include <linux/audit.h>
84#include <linux/capability.h>
85#include <linux/seq_file.h>
86#include <linux/rwsem.h>
87#include <linux/nsproxy.h>
88#include <linux/ipc_namespace.h>
89
90#include <asm/uaccess.h>
91#include "util.h"
92
93/* One semaphore structure for each semaphore in the system. */
94struct sem {
95 int semval; /* current value */
96 int sempid; /* pid of last operation */
97 struct list_head sem_pending; /* pending single-sop operations */
98};
99
100/* One queue for each sleeping process in the system. */
101struct sem_queue {
102 struct list_head simple_list; /* queue of pending operations */
103 struct list_head list; /* queue of pending operations */
104 struct task_struct *sleeper; /* this process */
105 struct sem_undo *undo; /* undo structure */
106 int pid; /* process id of requesting process */
107 int status; /* completion status of operation */
108 struct sembuf *sops; /* array of pending operations */
109 int nsops; /* number of operations */
110 int alter; /* does *sops alter the array? */
111};
112
113/* Each task has a list of undo requests. They are executed automatically
114 * when the process exits.
115 */
116struct sem_undo {
117 struct list_head list_proc; /* per-process list: *
118 * all undos from one process
119 * rcu protected */
120 struct rcu_head rcu; /* rcu struct for sem_undo */
121 struct sem_undo_list *ulp; /* back ptr to sem_undo_list */
122 struct list_head list_id; /* per semaphore array list:
123 * all undos for one array */
124 int semid; /* semaphore set identifier */
125 short *semadj; /* array of adjustments */
126 /* one per semaphore */
127};
128
129/* sem_undo_list controls shared access to the list of sem_undo structures
130 * that may be shared among all a CLONE_SYSVSEM task group.
131 */
132struct sem_undo_list {
133 atomic_t refcnt;
134 spinlock_t lock;
135 struct list_head list_proc;
136};
137
138
139#define sem_ids(ns) ((ns)->ids[IPC_SEM_IDS])
140
141#define sem_unlock(sma) ipc_unlock(&(sma)->sem_perm)
142#define sem_checkid(sma, semid) ipc_checkid(&sma->sem_perm, semid)
143
144static int newary(struct ipc_namespace *, struct ipc_params *);
145static void freeary(struct ipc_namespace *, struct kern_ipc_perm *);
146#ifdef CONFIG_PROC_FS
147static int sysvipc_sem_proc_show(struct seq_file *s, void *it);
148#endif
149
150#define SEMMSL_FAST 256 /* 512 bytes on stack */
151#define SEMOPM_FAST 64 /* ~ 372 bytes on stack */
152
153/*
154 * linked list protection:
155 * sem_undo.id_next,
156 * sem_array.sem_pending{,last},
157 * sem_array.sem_undo: sem_lock() for read/write
158 * sem_undo.proc_next: only "current" is allowed to read/write that field.
159 *
160 */
161
162#define sc_semmsl sem_ctls[0]
163#define sc_semmns sem_ctls[1]
164#define sc_semopm sem_ctls[2]
165#define sc_semmni sem_ctls[3]
166
167void sem_init_ns(struct ipc_namespace *ns)
168{
169 ns->sc_semmsl = SEMMSL;
170 ns->sc_semmns = SEMMNS;
171 ns->sc_semopm = SEMOPM;
172 ns->sc_semmni = SEMMNI;
173 ns->used_sems = 0;
174 ipc_init_ids(&ns->ids[IPC_SEM_IDS]);
175}
176
177#ifdef CONFIG_IPC_NS
178void sem_exit_ns(struct ipc_namespace *ns)
179{
180 free_ipcs(ns, &sem_ids(ns), freeary);
181 idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr);
182}
183#endif
184
185void __init sem_init (void)
186{
187 sem_init_ns(&init_ipc_ns);
188 ipc_init_proc_interface("sysvipc/sem",
189 " key semid perms nsems uid gid cuid cgid otime ctime\n",
190 IPC_SEM_IDS, sysvipc_sem_proc_show);
191}
192
193/*
194 * sem_lock_(check_) routines are called in the paths where the rw_mutex
195 * is not held.
196 */
197static inline struct sem_array *sem_lock(struct ipc_namespace *ns, int id)
198{
199 struct kern_ipc_perm *ipcp = ipc_lock(&sem_ids(ns), id);
200
201 if (IS_ERR(ipcp))
202 return (struct sem_array *)ipcp;
203
204 return container_of(ipcp, struct sem_array, sem_perm);
205}
206
207static inline struct sem_array *sem_lock_check(struct ipc_namespace *ns,
208 int id)
209{
210 struct kern_ipc_perm *ipcp = ipc_lock_check(&sem_ids(ns), id);
211
212 if (IS_ERR(ipcp))
213 return (struct sem_array *)ipcp;
214
215 return container_of(ipcp, struct sem_array, sem_perm);
216}
217
218static inline void sem_lock_and_putref(struct sem_array *sma)
219{
220 ipc_lock_by_ptr(&sma->sem_perm);
221 ipc_rcu_putref(sma);
222}
223
224static inline void sem_getref_and_unlock(struct sem_array *sma)
225{
226 ipc_rcu_getref(sma);
227 ipc_unlock(&(sma)->sem_perm);
228}
229
230static inline void sem_putref(struct sem_array *sma)
231{
232 ipc_lock_by_ptr(&sma->sem_perm);
233 ipc_rcu_putref(sma);
234 ipc_unlock(&(sma)->sem_perm);
235}
236
237static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
238{
239 ipc_rmid(&sem_ids(ns), &s->sem_perm);
240}
241
242/*
243 * Lockless wakeup algorithm:
244 * Without the check/retry algorithm a lockless wakeup is possible:
245 * - queue.status is initialized to -EINTR before blocking.
246 * - wakeup is performed by
247 * * unlinking the queue entry from sma->sem_pending
248 * * setting queue.status to IN_WAKEUP
249 * This is the notification for the blocked thread that a
250 * result value is imminent.
251 * * call wake_up_process
252 * * set queue.status to the final value.
253 * - the previously blocked thread checks queue.status:
254 * * if it's IN_WAKEUP, then it must wait until the value changes
255 * * if it's not -EINTR, then the operation was completed by
256 * update_queue. semtimedop can return queue.status without
257 * performing any operation on the sem array.
258 * * otherwise it must acquire the spinlock and check what's up.
259 *
260 * The two-stage algorithm is necessary to protect against the following
261 * races:
262 * - if queue.status is set after wake_up_process, then the woken up idle
263 * thread could race forward and try (and fail) to acquire sma->lock
264 * before update_queue had a chance to set queue.status
265 * - if queue.status is written before wake_up_process and if the
266 * blocked process is woken up by a signal between writing
267 * queue.status and the wake_up_process, then the woken up
268 * process could return from semtimedop and die by calling
269 * sys_exit before wake_up_process is called. Then wake_up_process
270 * will oops, because the task structure is already invalid.
271 * (yes, this happened on s390 with sysv msg).
272 *
273 */
274#define IN_WAKEUP 1
275
276/**
277 * newary - Create a new semaphore set
278 * @ns: namespace
279 * @params: ptr to the structure that contains key, semflg and nsems
280 *
281 * Called with sem_ids.rw_mutex held (as a writer)
282 */
283
284static int newary(struct ipc_namespace *ns, struct ipc_params *params)
285{
286 int id;
287 int retval;
288 struct sem_array *sma;
289 int size;
290 key_t key = params->key;
291 int nsems = params->u.nsems;
292 int semflg = params->flg;
293 int i;
294
295 if (!nsems)
296 return -EINVAL;
297 if (ns->used_sems + nsems > ns->sc_semmns)
298 return -ENOSPC;
299
300 size = sizeof (*sma) + nsems * sizeof (struct sem);
301 sma = ipc_rcu_alloc(size);
302 if (!sma) {
303 return -ENOMEM;
304 }
305 memset (sma, 0, size);
306
307 sma->sem_perm.mode = (semflg & S_IRWXUGO);
308 sma->sem_perm.key = key;
309
310 sma->sem_perm.security = NULL;
311 retval = security_sem_alloc(sma);
312 if (retval) {
313 ipc_rcu_putref(sma);
314 return retval;
315 }
316
317 id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
318 if (id < 0) {
319 security_sem_free(sma);
320 ipc_rcu_putref(sma);
321 return id;
322 }
323 ns->used_sems += nsems;
324
325 sma->sem_base = (struct sem *) &sma[1];
326
327 for (i = 0; i < nsems; i++)
328 INIT_LIST_HEAD(&sma->sem_base[i].sem_pending);
329
330 sma->complex_count = 0;
331 INIT_LIST_HEAD(&sma->sem_pending);
332 INIT_LIST_HEAD(&sma->list_id);
333 sma->sem_nsems = nsems;
334 sma->sem_ctime = get_seconds();
335 sem_unlock(sma);
336
337 return sma->sem_perm.id;
338}
339
340
341/*
342 * Called with sem_ids.rw_mutex and ipcp locked.
343 */
344static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg)
345{
346 struct sem_array *sma;
347
348 sma = container_of(ipcp, struct sem_array, sem_perm);
349 return security_sem_associate(sma, semflg);
350}
351
352/*
353 * Called with sem_ids.rw_mutex and ipcp locked.
354 */
355static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
356 struct ipc_params *params)
357{
358 struct sem_array *sma;
359
360 sma = container_of(ipcp, struct sem_array, sem_perm);
361 if (params->u.nsems > sma->sem_nsems)
362 return -EINVAL;
363
364 return 0;
365}
366
367SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg)
368{
369 struct ipc_namespace *ns;
370 struct ipc_ops sem_ops;
371 struct ipc_params sem_params;
372
373 ns = current->nsproxy->ipc_ns;
374
375 if (nsems < 0 || nsems > ns->sc_semmsl)
376 return -EINVAL;
377
378 sem_ops.getnew = newary;
379 sem_ops.associate = sem_security;
380 sem_ops.more_checks = sem_more_checks;
381
382 sem_params.key = key;
383 sem_params.flg = semflg;
384 sem_params.u.nsems = nsems;
385
386 return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
387}
388
389/*
390 * Determine whether a sequence of semaphore operations would succeed
391 * all at once. Return 0 if yes, 1 if need to sleep, else return error code.
392 */
393
394static int try_atomic_semop (struct sem_array * sma, struct sembuf * sops,
395 int nsops, struct sem_undo *un, int pid)
396{
397 int result, sem_op;
398 struct sembuf *sop;
399 struct sem * curr;
400
401 for (sop = sops; sop < sops + nsops; sop++) {
402 curr = sma->sem_base + sop->sem_num;
403 sem_op = sop->sem_op;
404 result = curr->semval;
405
406 if (!sem_op && result)
407 goto would_block;
408
409 result += sem_op;
410 if (result < 0)
411 goto would_block;
412 if (result > SEMVMX)
413 goto out_of_range;
414 if (sop->sem_flg & SEM_UNDO) {
415 int undo = un->semadj[sop->sem_num] - sem_op;
416 /*
417 * Exceeding the undo range is an error.
418 */
419 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
420 goto out_of_range;
421 }
422 curr->semval = result;
423 }
424
425 sop--;
426 while (sop >= sops) {
427 sma->sem_base[sop->sem_num].sempid = pid;
428 if (sop->sem_flg & SEM_UNDO)
429 un->semadj[sop->sem_num] -= sop->sem_op;
430 sop--;
431 }
432
433 return 0;
434
435out_of_range:
436 result = -ERANGE;
437 goto undo;
438
439would_block:
440 if (sop->sem_flg & IPC_NOWAIT)
441 result = -EAGAIN;
442 else
443 result = 1;
444
445undo:
446 sop--;
447 while (sop >= sops) {
448 sma->sem_base[sop->sem_num].semval -= sop->sem_op;
449 sop--;
450 }
451
452 return result;
453}
454
455/** wake_up_sem_queue_prepare(q, error): Prepare wake-up
456 * @q: queue entry that must be signaled
457 * @error: Error value for the signal
458 *
459 * Prepare the wake-up of the queue entry q.
460 */
461static void wake_up_sem_queue_prepare(struct list_head *pt,
462 struct sem_queue *q, int error)
463{
464 if (list_empty(pt)) {
465 /*
466 * Hold preempt off so that we don't get preempted and have the
467 * wakee busy-wait until we're scheduled back on.
468 */
469 preempt_disable();
470 }
471 q->status = IN_WAKEUP;
472 q->pid = error;
473
474 list_add_tail(&q->simple_list, pt);
475}
476
477/**
478 * wake_up_sem_queue_do(pt) - do the actual wake-up
479 * @pt: list of tasks to be woken up
480 *
481 * Do the actual wake-up.
482 * The function is called without any locks held, thus the semaphore array
483 * could be destroyed already and the tasks can disappear as soon as the
484 * status is set to the actual return code.
485 */
486static void wake_up_sem_queue_do(struct list_head *pt)
487{
488 struct sem_queue *q, *t;
489 int did_something;
490
491 did_something = !list_empty(pt);
492 list_for_each_entry_safe(q, t, pt, simple_list) {
493 wake_up_process(q->sleeper);
494 /* q can disappear immediately after writing q->status. */
495 smp_wmb();
496 q->status = q->pid;
497 }
498 if (did_something)
499 preempt_enable();
500}
501
502static void unlink_queue(struct sem_array *sma, struct sem_queue *q)
503{
504 list_del(&q->list);
505 if (q->nsops == 1)
506 list_del(&q->simple_list);
507 else
508 sma->complex_count--;
509}
510
511/** check_restart(sma, q)
512 * @sma: semaphore array
513 * @q: the operation that just completed
514 *
515 * update_queue is O(N^2) when it restarts scanning the whole queue of
516 * waiting operations. Therefore this function checks if the restart is
517 * really necessary. It is called after a previously waiting operation
518 * was completed.
519 */
520static int check_restart(struct sem_array *sma, struct sem_queue *q)
521{
522 struct sem *curr;
523 struct sem_queue *h;
524
525 /* if the operation didn't modify the array, then no restart */
526 if (q->alter == 0)
527 return 0;
528
529 /* pending complex operations are too difficult to analyse */
530 if (sma->complex_count)
531 return 1;
532
533 /* we were a sleeping complex operation. Too difficult */
534 if (q->nsops > 1)
535 return 1;
536
537 curr = sma->sem_base + q->sops[0].sem_num;
538
539 /* No-one waits on this queue */
540 if (list_empty(&curr->sem_pending))
541 return 0;
542
543 /* the new semaphore value */
544 if (curr->semval) {
545 /* It is impossible that someone waits for the new value:
546 * - q is a previously sleeping simple operation that
547 * altered the array. It must be a decrement, because
548 * simple increments never sleep.
549 * - The value is not 0, thus wait-for-zero won't proceed.
550 * - If there are older (higher priority) decrements
551 * in the queue, then they have observed the original
552 * semval value and couldn't proceed. The operation
553 * decremented to value - thus they won't proceed either.
554 */
555 BUG_ON(q->sops[0].sem_op >= 0);
556 return 0;
557 }
558 /*
559 * semval is 0. Check if there are wait-for-zero semops.
560 * They must be the first entries in the per-semaphore simple queue
561 */
562 h = list_first_entry(&curr->sem_pending, struct sem_queue, simple_list);
563 BUG_ON(h->nsops != 1);
564 BUG_ON(h->sops[0].sem_num != q->sops[0].sem_num);
565
566 /* Yes, there is a wait-for-zero semop. Restart */
567 if (h->sops[0].sem_op == 0)
568 return 1;
569
570 /* Again - no-one is waiting for the new value. */
571 return 0;
572}
573
574
575/**
576 * update_queue(sma, semnum): Look for tasks that can be completed.
577 * @sma: semaphore array.
578 * @semnum: semaphore that was modified.
579 * @pt: list head for the tasks that must be woken up.
580 *
581 * update_queue must be called after a semaphore in a semaphore array
582 * was modified. If multiple semaphore were modified, then @semnum
583 * must be set to -1.
584 * The tasks that must be woken up are added to @pt. The return code
585 * is stored in q->pid.
586 * The function return 1 if at least one semop was completed successfully.
587 */
588static int update_queue(struct sem_array *sma, int semnum, struct list_head *pt)
589{
590 struct sem_queue *q;
591 struct list_head *walk;
592 struct list_head *pending_list;
593 int offset;
594 int semop_completed = 0;
595
596 /* if there are complex operations around, then knowing the semaphore
597 * that was modified doesn't help us. Assume that multiple semaphores
598 * were modified.
599 */
600 if (sma->complex_count)
601 semnum = -1;
602
603 if (semnum == -1) {
604 pending_list = &sma->sem_pending;
605 offset = offsetof(struct sem_queue, list);
606 } else {
607 pending_list = &sma->sem_base[semnum].sem_pending;
608 offset = offsetof(struct sem_queue, simple_list);
609 }
610
611again:
612 walk = pending_list->next;
613 while (walk != pending_list) {
614 int error, restart;
615
616 q = (struct sem_queue *)((char *)walk - offset);
617 walk = walk->next;
618
619 /* If we are scanning the single sop, per-semaphore list of
620 * one semaphore and that semaphore is 0, then it is not
621 * necessary to scan the "alter" entries: simple increments
622 * that affect only one entry succeed immediately and cannot
623 * be in the per semaphore pending queue, and decrements
624 * cannot be successful if the value is already 0.
625 */
626 if (semnum != -1 && sma->sem_base[semnum].semval == 0 &&
627 q->alter)
628 break;
629
630 error = try_atomic_semop(sma, q->sops, q->nsops,
631 q->undo, q->pid);
632
633 /* Does q->sleeper still need to sleep? */
634 if (error > 0)
635 continue;
636
637 unlink_queue(sma, q);
638
639 if (error) {
640 restart = 0;
641 } else {
642 semop_completed = 1;
643 restart = check_restart(sma, q);
644 }
645
646 wake_up_sem_queue_prepare(pt, q, error);
647 if (restart)
648 goto again;
649 }
650 return semop_completed;
651}
652
653/**
654 * do_smart_update(sma, sops, nsops, otime, pt) - optimized update_queue
655 * @sma: semaphore array
656 * @sops: operations that were performed
657 * @nsops: number of operations
658 * @otime: force setting otime
659 * @pt: list head of the tasks that must be woken up.
660 *
661 * do_smart_update() does the required called to update_queue, based on the
662 * actual changes that were performed on the semaphore array.
663 * Note that the function does not do the actual wake-up: the caller is
664 * responsible for calling wake_up_sem_queue_do(@pt).
665 * It is safe to perform this call after dropping all locks.
666 */
667static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops,
668 int otime, struct list_head *pt)
669{
670 int i;
671
672 if (sma->complex_count || sops == NULL) {
673 if (update_queue(sma, -1, pt))
674 otime = 1;
675 goto done;
676 }
677
678 for (i = 0; i < nsops; i++) {
679 if (sops[i].sem_op > 0 ||
680 (sops[i].sem_op < 0 &&
681 sma->sem_base[sops[i].sem_num].semval == 0))
682 if (update_queue(sma, sops[i].sem_num, pt))
683 otime = 1;
684 }
685done:
686 if (otime)
687 sma->sem_otime = get_seconds();
688}
689
690
691/* The following counts are associated to each semaphore:
692 * semncnt number of tasks waiting on semval being nonzero
693 * semzcnt number of tasks waiting on semval being zero
694 * This model assumes that a task waits on exactly one semaphore.
695 * Since semaphore operations are to be performed atomically, tasks actually
696 * wait on a whole sequence of semaphores simultaneously.
697 * The counts we return here are a rough approximation, but still
698 * warrant that semncnt+semzcnt>0 if the task is on the pending queue.
699 */
700static int count_semncnt (struct sem_array * sma, ushort semnum)
701{
702 int semncnt;
703 struct sem_queue * q;
704
705 semncnt = 0;
706 list_for_each_entry(q, &sma->sem_pending, list) {
707 struct sembuf * sops = q->sops;
708 int nsops = q->nsops;
709 int i;
710 for (i = 0; i < nsops; i++)
711 if (sops[i].sem_num == semnum
712 && (sops[i].sem_op < 0)
713 && !(sops[i].sem_flg & IPC_NOWAIT))
714 semncnt++;
715 }
716 return semncnt;
717}
718
719static int count_semzcnt (struct sem_array * sma, ushort semnum)
720{
721 int semzcnt;
722 struct sem_queue * q;
723
724 semzcnt = 0;
725 list_for_each_entry(q, &sma->sem_pending, list) {
726 struct sembuf * sops = q->sops;
727 int nsops = q->nsops;
728 int i;
729 for (i = 0; i < nsops; i++)
730 if (sops[i].sem_num == semnum
731 && (sops[i].sem_op == 0)
732 && !(sops[i].sem_flg & IPC_NOWAIT))
733 semzcnt++;
734 }
735 return semzcnt;
736}
737
738/* Free a semaphore set. freeary() is called with sem_ids.rw_mutex locked
739 * as a writer and the spinlock for this semaphore set hold. sem_ids.rw_mutex
740 * remains locked on exit.
741 */
742static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
743{
744 struct sem_undo *un, *tu;
745 struct sem_queue *q, *tq;
746 struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
747 struct list_head tasks;
748
749 /* Free the existing undo structures for this semaphore set. */
750 assert_spin_locked(&sma->sem_perm.lock);
751 list_for_each_entry_safe(un, tu, &sma->list_id, list_id) {
752 list_del(&un->list_id);
753 spin_lock(&un->ulp->lock);
754 un->semid = -1;
755 list_del_rcu(&un->list_proc);
756 spin_unlock(&un->ulp->lock);
757 kfree_rcu(un, rcu);
758 }
759
760 /* Wake up all pending processes and let them fail with EIDRM. */
761 INIT_LIST_HEAD(&tasks);
762 list_for_each_entry_safe(q, tq, &sma->sem_pending, list) {
763 unlink_queue(sma, q);
764 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
765 }
766
767 /* Remove the semaphore set from the IDR */
768 sem_rmid(ns, sma);
769 sem_unlock(sma);
770
771 wake_up_sem_queue_do(&tasks);
772 ns->used_sems -= sma->sem_nsems;
773 security_sem_free(sma);
774 ipc_rcu_putref(sma);
775}
776
777static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
778{
779 switch(version) {
780 case IPC_64:
781 return copy_to_user(buf, in, sizeof(*in));
782 case IPC_OLD:
783 {
784 struct semid_ds out;
785
786 memset(&out, 0, sizeof(out));
787
788 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
789
790 out.sem_otime = in->sem_otime;
791 out.sem_ctime = in->sem_ctime;
792 out.sem_nsems = in->sem_nsems;
793
794 return copy_to_user(buf, &out, sizeof(out));
795 }
796 default:
797 return -EINVAL;
798 }
799}
800
801static int semctl_nolock(struct ipc_namespace *ns, int semid,
802 int cmd, int version, union semun arg)
803{
804 int err;
805 struct sem_array *sma;
806
807 switch(cmd) {
808 case IPC_INFO:
809 case SEM_INFO:
810 {
811 struct seminfo seminfo;
812 int max_id;
813
814 err = security_sem_semctl(NULL, cmd);
815 if (err)
816 return err;
817
818 memset(&seminfo,0,sizeof(seminfo));
819 seminfo.semmni = ns->sc_semmni;
820 seminfo.semmns = ns->sc_semmns;
821 seminfo.semmsl = ns->sc_semmsl;
822 seminfo.semopm = ns->sc_semopm;
823 seminfo.semvmx = SEMVMX;
824 seminfo.semmnu = SEMMNU;
825 seminfo.semmap = SEMMAP;
826 seminfo.semume = SEMUME;
827 down_read(&sem_ids(ns).rw_mutex);
828 if (cmd == SEM_INFO) {
829 seminfo.semusz = sem_ids(ns).in_use;
830 seminfo.semaem = ns->used_sems;
831 } else {
832 seminfo.semusz = SEMUSZ;
833 seminfo.semaem = SEMAEM;
834 }
835 max_id = ipc_get_maxid(&sem_ids(ns));
836 up_read(&sem_ids(ns).rw_mutex);
837 if (copy_to_user (arg.__buf, &seminfo, sizeof(struct seminfo)))
838 return -EFAULT;
839 return (max_id < 0) ? 0: max_id;
840 }
841 case IPC_STAT:
842 case SEM_STAT:
843 {
844 struct semid64_ds tbuf;
845 int id;
846
847 if (cmd == SEM_STAT) {
848 sma = sem_lock(ns, semid);
849 if (IS_ERR(sma))
850 return PTR_ERR(sma);
851 id = sma->sem_perm.id;
852 } else {
853 sma = sem_lock_check(ns, semid);
854 if (IS_ERR(sma))
855 return PTR_ERR(sma);
856 id = 0;
857 }
858
859 err = -EACCES;
860 if (ipcperms(ns, &sma->sem_perm, S_IRUGO))
861 goto out_unlock;
862
863 err = security_sem_semctl(sma, cmd);
864 if (err)
865 goto out_unlock;
866
867 memset(&tbuf, 0, sizeof(tbuf));
868
869 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
870 tbuf.sem_otime = sma->sem_otime;
871 tbuf.sem_ctime = sma->sem_ctime;
872 tbuf.sem_nsems = sma->sem_nsems;
873 sem_unlock(sma);
874 if (copy_semid_to_user (arg.buf, &tbuf, version))
875 return -EFAULT;
876 return id;
877 }
878 default:
879 return -EINVAL;
880 }
881out_unlock:
882 sem_unlock(sma);
883 return err;
884}
885
886static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
887 int cmd, int version, union semun arg)
888{
889 struct sem_array *sma;
890 struct sem* curr;
891 int err;
892 ushort fast_sem_io[SEMMSL_FAST];
893 ushort* sem_io = fast_sem_io;
894 int nsems;
895 struct list_head tasks;
896
897 sma = sem_lock_check(ns, semid);
898 if (IS_ERR(sma))
899 return PTR_ERR(sma);
900
901 INIT_LIST_HEAD(&tasks);
902 nsems = sma->sem_nsems;
903
904 err = -EACCES;
905 if (ipcperms(ns, &sma->sem_perm,
906 (cmd == SETVAL || cmd == SETALL) ? S_IWUGO : S_IRUGO))
907 goto out_unlock;
908
909 err = security_sem_semctl(sma, cmd);
910 if (err)
911 goto out_unlock;
912
913 err = -EACCES;
914 switch (cmd) {
915 case GETALL:
916 {
917 ushort __user *array = arg.array;
918 int i;
919
920 if(nsems > SEMMSL_FAST) {
921 sem_getref_and_unlock(sma);
922
923 sem_io = ipc_alloc(sizeof(ushort)*nsems);
924 if(sem_io == NULL) {
925 sem_putref(sma);
926 return -ENOMEM;
927 }
928
929 sem_lock_and_putref(sma);
930 if (sma->sem_perm.deleted) {
931 sem_unlock(sma);
932 err = -EIDRM;
933 goto out_free;
934 }
935 }
936
937 for (i = 0; i < sma->sem_nsems; i++)
938 sem_io[i] = sma->sem_base[i].semval;
939 sem_unlock(sma);
940 err = 0;
941 if(copy_to_user(array, sem_io, nsems*sizeof(ushort)))
942 err = -EFAULT;
943 goto out_free;
944 }
945 case SETALL:
946 {
947 int i;
948 struct sem_undo *un;
949
950 sem_getref_and_unlock(sma);
951
952 if(nsems > SEMMSL_FAST) {
953 sem_io = ipc_alloc(sizeof(ushort)*nsems);
954 if(sem_io == NULL) {
955 sem_putref(sma);
956 return -ENOMEM;
957 }
958 }
959
960 if (copy_from_user (sem_io, arg.array, nsems*sizeof(ushort))) {
961 sem_putref(sma);
962 err = -EFAULT;
963 goto out_free;
964 }
965
966 for (i = 0; i < nsems; i++) {
967 if (sem_io[i] > SEMVMX) {
968 sem_putref(sma);
969 err = -ERANGE;
970 goto out_free;
971 }
972 }
973 sem_lock_and_putref(sma);
974 if (sma->sem_perm.deleted) {
975 sem_unlock(sma);
976 err = -EIDRM;
977 goto out_free;
978 }
979
980 for (i = 0; i < nsems; i++)
981 sma->sem_base[i].semval = sem_io[i];
982
983 assert_spin_locked(&sma->sem_perm.lock);
984 list_for_each_entry(un, &sma->list_id, list_id) {
985 for (i = 0; i < nsems; i++)
986 un->semadj[i] = 0;
987 }
988 sma->sem_ctime = get_seconds();
989 /* maybe some queued-up processes were waiting for this */
990 do_smart_update(sma, NULL, 0, 0, &tasks);
991 err = 0;
992 goto out_unlock;
993 }
994 /* GETVAL, GETPID, GETNCTN, GETZCNT, SETVAL: fall-through */
995 }
996 err = -EINVAL;
997 if(semnum < 0 || semnum >= nsems)
998 goto out_unlock;
999
1000 curr = &sma->sem_base[semnum];
1001
1002 switch (cmd) {
1003 case GETVAL:
1004 err = curr->semval;
1005 goto out_unlock;
1006 case GETPID:
1007 err = curr->sempid;
1008 goto out_unlock;
1009 case GETNCNT:
1010 err = count_semncnt(sma,semnum);
1011 goto out_unlock;
1012 case GETZCNT:
1013 err = count_semzcnt(sma,semnum);
1014 goto out_unlock;
1015 case SETVAL:
1016 {
1017 int val = arg.val;
1018 struct sem_undo *un;
1019
1020 err = -ERANGE;
1021 if (val > SEMVMX || val < 0)
1022 goto out_unlock;
1023
1024 assert_spin_locked(&sma->sem_perm.lock);
1025 list_for_each_entry(un, &sma->list_id, list_id)
1026 un->semadj[semnum] = 0;
1027
1028 curr->semval = val;
1029 curr->sempid = task_tgid_vnr(current);
1030 sma->sem_ctime = get_seconds();
1031 /* maybe some queued-up processes were waiting for this */
1032 do_smart_update(sma, NULL, 0, 0, &tasks);
1033 err = 0;
1034 goto out_unlock;
1035 }
1036 }
1037out_unlock:
1038 sem_unlock(sma);
1039 wake_up_sem_queue_do(&tasks);
1040
1041out_free:
1042 if(sem_io != fast_sem_io)
1043 ipc_free(sem_io, sizeof(ushort)*nsems);
1044 return err;
1045}
1046
1047static inline unsigned long
1048copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
1049{
1050 switch(version) {
1051 case IPC_64:
1052 if (copy_from_user(out, buf, sizeof(*out)))
1053 return -EFAULT;
1054 return 0;
1055 case IPC_OLD:
1056 {
1057 struct semid_ds tbuf_old;
1058
1059 if(copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
1060 return -EFAULT;
1061
1062 out->sem_perm.uid = tbuf_old.sem_perm.uid;
1063 out->sem_perm.gid = tbuf_old.sem_perm.gid;
1064 out->sem_perm.mode = tbuf_old.sem_perm.mode;
1065
1066 return 0;
1067 }
1068 default:
1069 return -EINVAL;
1070 }
1071}
1072
1073/*
1074 * This function handles some semctl commands which require the rw_mutex
1075 * to be held in write mode.
1076 * NOTE: no locks must be held, the rw_mutex is taken inside this function.
1077 */
1078static int semctl_down(struct ipc_namespace *ns, int semid,
1079 int cmd, int version, union semun arg)
1080{
1081 struct sem_array *sma;
1082 int err;
1083 struct semid64_ds semid64;
1084 struct kern_ipc_perm *ipcp;
1085
1086 if(cmd == IPC_SET) {
1087 if (copy_semid_from_user(&semid64, arg.buf, version))
1088 return -EFAULT;
1089 }
1090
1091 ipcp = ipcctl_pre_down(ns, &sem_ids(ns), semid, cmd,
1092 &semid64.sem_perm, 0);
1093 if (IS_ERR(ipcp))
1094 return PTR_ERR(ipcp);
1095
1096 sma = container_of(ipcp, struct sem_array, sem_perm);
1097
1098 err = security_sem_semctl(sma, cmd);
1099 if (err)
1100 goto out_unlock;
1101
1102 switch(cmd){
1103 case IPC_RMID:
1104 freeary(ns, ipcp);
1105 goto out_up;
1106 case IPC_SET:
1107 ipc_update_perm(&semid64.sem_perm, ipcp);
1108 sma->sem_ctime = get_seconds();
1109 break;
1110 default:
1111 err = -EINVAL;
1112 }
1113
1114out_unlock:
1115 sem_unlock(sma);
1116out_up:
1117 up_write(&sem_ids(ns).rw_mutex);
1118 return err;
1119}
1120
1121SYSCALL_DEFINE(semctl)(int semid, int semnum, int cmd, union semun arg)
1122{
1123 int err = -EINVAL;
1124 int version;
1125 struct ipc_namespace *ns;
1126
1127 if (semid < 0)
1128 return -EINVAL;
1129
1130 version = ipc_parse_version(&cmd);
1131 ns = current->nsproxy->ipc_ns;
1132
1133 switch(cmd) {
1134 case IPC_INFO:
1135 case SEM_INFO:
1136 case IPC_STAT:
1137 case SEM_STAT:
1138 err = semctl_nolock(ns, semid, cmd, version, arg);
1139 return err;
1140 case GETALL:
1141 case GETVAL:
1142 case GETPID:
1143 case GETNCNT:
1144 case GETZCNT:
1145 case SETVAL:
1146 case SETALL:
1147 err = semctl_main(ns,semid,semnum,cmd,version,arg);
1148 return err;
1149 case IPC_RMID:
1150 case IPC_SET:
1151 err = semctl_down(ns, semid, cmd, version, arg);
1152 return err;
1153 default:
1154 return -EINVAL;
1155 }
1156}
1157#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
1158asmlinkage long SyS_semctl(int semid, int semnum, int cmd, union semun arg)
1159{
1160 return SYSC_semctl((int) semid, (int) semnum, (int) cmd, arg);
1161}
1162SYSCALL_ALIAS(sys_semctl, SyS_semctl);
1163#endif
1164
1165/* If the task doesn't already have a undo_list, then allocate one
1166 * here. We guarantee there is only one thread using this undo list,
1167 * and current is THE ONE
1168 *
1169 * If this allocation and assignment succeeds, but later
1170 * portions of this code fail, there is no need to free the sem_undo_list.
1171 * Just let it stay associated with the task, and it'll be freed later
1172 * at exit time.
1173 *
1174 * This can block, so callers must hold no locks.
1175 */
1176static inline int get_undo_list(struct sem_undo_list **undo_listp)
1177{
1178 struct sem_undo_list *undo_list;
1179
1180 undo_list = current->sysvsem.undo_list;
1181 if (!undo_list) {
1182 undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
1183 if (undo_list == NULL)
1184 return -ENOMEM;
1185 spin_lock_init(&undo_list->lock);
1186 atomic_set(&undo_list->refcnt, 1);
1187 INIT_LIST_HEAD(&undo_list->list_proc);
1188
1189 current->sysvsem.undo_list = undo_list;
1190 }
1191 *undo_listp = undo_list;
1192 return 0;
1193}
1194
1195static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid)
1196{
1197 struct sem_undo *un;
1198
1199 list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) {
1200 if (un->semid == semid)
1201 return un;
1202 }
1203 return NULL;
1204}
1205
1206static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1207{
1208 struct sem_undo *un;
1209
1210 assert_spin_locked(&ulp->lock);
1211
1212 un = __lookup_undo(ulp, semid);
1213 if (un) {
1214 list_del_rcu(&un->list_proc);
1215 list_add_rcu(&un->list_proc, &ulp->list_proc);
1216 }
1217 return un;
1218}
1219
1220/**
1221 * find_alloc_undo - Lookup (and if not present create) undo array
1222 * @ns: namespace
1223 * @semid: semaphore array id
1224 *
1225 * The function looks up (and if not present creates) the undo structure.
1226 * The size of the undo structure depends on the size of the semaphore
1227 * array, thus the alloc path is not that straightforward.
1228 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1229 * performs a rcu_read_lock().
1230 */
1231static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
1232{
1233 struct sem_array *sma;
1234 struct sem_undo_list *ulp;
1235 struct sem_undo *un, *new;
1236 int nsems;
1237 int error;
1238
1239 error = get_undo_list(&ulp);
1240 if (error)
1241 return ERR_PTR(error);
1242
1243 rcu_read_lock();
1244 spin_lock(&ulp->lock);
1245 un = lookup_undo(ulp, semid);
1246 spin_unlock(&ulp->lock);
1247 if (likely(un!=NULL))
1248 goto out;
1249 rcu_read_unlock();
1250
1251 /* no undo structure around - allocate one. */
1252 /* step 1: figure out the size of the semaphore array */
1253 sma = sem_lock_check(ns, semid);
1254 if (IS_ERR(sma))
1255 return ERR_CAST(sma);
1256
1257 nsems = sma->sem_nsems;
1258 sem_getref_and_unlock(sma);
1259
1260 /* step 2: allocate new undo structure */
1261 new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1262 if (!new) {
1263 sem_putref(sma);
1264 return ERR_PTR(-ENOMEM);
1265 }
1266
1267 /* step 3: Acquire the lock on semaphore array */
1268 sem_lock_and_putref(sma);
1269 if (sma->sem_perm.deleted) {
1270 sem_unlock(sma);
1271 kfree(new);
1272 un = ERR_PTR(-EIDRM);
1273 goto out;
1274 }
1275 spin_lock(&ulp->lock);
1276
1277 /*
1278 * step 4: check for races: did someone else allocate the undo struct?
1279 */
1280 un = lookup_undo(ulp, semid);
1281 if (un) {
1282 kfree(new);
1283 goto success;
1284 }
1285 /* step 5: initialize & link new undo structure */
1286 new->semadj = (short *) &new[1];
1287 new->ulp = ulp;
1288 new->semid = semid;
1289 assert_spin_locked(&ulp->lock);
1290 list_add_rcu(&new->list_proc, &ulp->list_proc);
1291 assert_spin_locked(&sma->sem_perm.lock);
1292 list_add(&new->list_id, &sma->list_id);
1293 un = new;
1294
1295success:
1296 spin_unlock(&ulp->lock);
1297 rcu_read_lock();
1298 sem_unlock(sma);
1299out:
1300 return un;
1301}
1302
1303
1304/**
1305 * get_queue_result - Retrieve the result code from sem_queue
1306 * @q: Pointer to queue structure
1307 *
1308 * Retrieve the return code from the pending queue. If IN_WAKEUP is found in
1309 * q->status, then we must loop until the value is replaced with the final
1310 * value: This may happen if a task is woken up by an unrelated event (e.g.
1311 * signal) and in parallel the task is woken up by another task because it got
1312 * the requested semaphores.
1313 *
1314 * The function can be called with or without holding the semaphore spinlock.
1315 */
1316static int get_queue_result(struct sem_queue *q)
1317{
1318 int error;
1319
1320 error = q->status;
1321 while (unlikely(error == IN_WAKEUP)) {
1322 cpu_relax();
1323 error = q->status;
1324 }
1325
1326 return error;
1327}
1328
1329
1330SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
1331 unsigned, nsops, const struct timespec __user *, timeout)
1332{
1333 int error = -EINVAL;
1334 struct sem_array *sma;
1335 struct sembuf fast_sops[SEMOPM_FAST];
1336 struct sembuf* sops = fast_sops, *sop;
1337 struct sem_undo *un;
1338 int undos = 0, alter = 0, max;
1339 struct sem_queue queue;
1340 unsigned long jiffies_left = 0;
1341 struct ipc_namespace *ns;
1342 struct list_head tasks;
1343
1344 ns = current->nsproxy->ipc_ns;
1345
1346 if (nsops < 1 || semid < 0)
1347 return -EINVAL;
1348 if (nsops > ns->sc_semopm)
1349 return -E2BIG;
1350 if(nsops > SEMOPM_FAST) {
1351 sops = kmalloc(sizeof(*sops)*nsops,GFP_KERNEL);
1352 if(sops==NULL)
1353 return -ENOMEM;
1354 }
1355 if (copy_from_user (sops, tsops, nsops * sizeof(*tsops))) {
1356 error=-EFAULT;
1357 goto out_free;
1358 }
1359 if (timeout) {
1360 struct timespec _timeout;
1361 if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1362 error = -EFAULT;
1363 goto out_free;
1364 }
1365 if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1366 _timeout.tv_nsec >= 1000000000L) {
1367 error = -EINVAL;
1368 goto out_free;
1369 }
1370 jiffies_left = timespec_to_jiffies(&_timeout);
1371 }
1372 max = 0;
1373 for (sop = sops; sop < sops + nsops; sop++) {
1374 if (sop->sem_num >= max)
1375 max = sop->sem_num;
1376 if (sop->sem_flg & SEM_UNDO)
1377 undos = 1;
1378 if (sop->sem_op != 0)
1379 alter = 1;
1380 }
1381
1382 if (undos) {
1383 un = find_alloc_undo(ns, semid);
1384 if (IS_ERR(un)) {
1385 error = PTR_ERR(un);
1386 goto out_free;
1387 }
1388 } else
1389 un = NULL;
1390
1391 INIT_LIST_HEAD(&tasks);
1392
1393 sma = sem_lock_check(ns, semid);
1394 if (IS_ERR(sma)) {
1395 if (un)
1396 rcu_read_unlock();
1397 error = PTR_ERR(sma);
1398 goto out_free;
1399 }
1400
1401 /*
1402 * semid identifiers are not unique - find_alloc_undo may have
1403 * allocated an undo structure, it was invalidated by an RMID
1404 * and now a new array with received the same id. Check and fail.
1405 * This case can be detected checking un->semid. The existence of
1406 * "un" itself is guaranteed by rcu.
1407 */
1408 error = -EIDRM;
1409 if (un) {
1410 if (un->semid == -1) {
1411 rcu_read_unlock();
1412 goto out_unlock_free;
1413 } else {
1414 /*
1415 * rcu lock can be released, "un" cannot disappear:
1416 * - sem_lock is acquired, thus IPC_RMID is
1417 * impossible.
1418 * - exit_sem is impossible, it always operates on
1419 * current (or a dead task).
1420 */
1421
1422 rcu_read_unlock();
1423 }
1424 }
1425
1426 error = -EFBIG;
1427 if (max >= sma->sem_nsems)
1428 goto out_unlock_free;
1429
1430 error = -EACCES;
1431 if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO))
1432 goto out_unlock_free;
1433
1434 error = security_sem_semop(sma, sops, nsops, alter);
1435 if (error)
1436 goto out_unlock_free;
1437
1438 error = try_atomic_semop (sma, sops, nsops, un, task_tgid_vnr(current));
1439 if (error <= 0) {
1440 if (alter && error == 0)
1441 do_smart_update(sma, sops, nsops, 1, &tasks);
1442
1443 goto out_unlock_free;
1444 }
1445
1446 /* We need to sleep on this operation, so we put the current
1447 * task into the pending queue and go to sleep.
1448 */
1449
1450 queue.sops = sops;
1451 queue.nsops = nsops;
1452 queue.undo = un;
1453 queue.pid = task_tgid_vnr(current);
1454 queue.alter = alter;
1455 if (alter)
1456 list_add_tail(&queue.list, &sma->sem_pending);
1457 else
1458 list_add(&queue.list, &sma->sem_pending);
1459
1460 if (nsops == 1) {
1461 struct sem *curr;
1462 curr = &sma->sem_base[sops->sem_num];
1463
1464 if (alter)
1465 list_add_tail(&queue.simple_list, &curr->sem_pending);
1466 else
1467 list_add(&queue.simple_list, &curr->sem_pending);
1468 } else {
1469 INIT_LIST_HEAD(&queue.simple_list);
1470 sma->complex_count++;
1471 }
1472
1473 queue.status = -EINTR;
1474 queue.sleeper = current;
1475
1476sleep_again:
1477 current->state = TASK_INTERRUPTIBLE;
1478 sem_unlock(sma);
1479
1480 if (timeout)
1481 jiffies_left = schedule_timeout(jiffies_left);
1482 else
1483 schedule();
1484
1485 error = get_queue_result(&queue);
1486
1487 if (error != -EINTR) {
1488 /* fast path: update_queue already obtained all requested
1489 * resources.
1490 * Perform a smp_mb(): User space could assume that semop()
1491 * is a memory barrier: Without the mb(), the cpu could
1492 * speculatively read in user space stale data that was
1493 * overwritten by the previous owner of the semaphore.
1494 */
1495 smp_mb();
1496
1497 goto out_free;
1498 }
1499
1500 sma = sem_lock(ns, semid);
1501
1502 /*
1503 * Wait until it's guaranteed that no wakeup_sem_queue_do() is ongoing.
1504 */
1505 error = get_queue_result(&queue);
1506
1507 /*
1508 * Array removed? If yes, leave without sem_unlock().
1509 */
1510 if (IS_ERR(sma)) {
1511 goto out_free;
1512 }
1513
1514
1515 /*
1516 * If queue.status != -EINTR we are woken up by another process.
1517 * Leave without unlink_queue(), but with sem_unlock().
1518 */
1519
1520 if (error != -EINTR) {
1521 goto out_unlock_free;
1522 }
1523
1524 /*
1525 * If an interrupt occurred we have to clean up the queue
1526 */
1527 if (timeout && jiffies_left == 0)
1528 error = -EAGAIN;
1529
1530 /*
1531 * If the wakeup was spurious, just retry
1532 */
1533 if (error == -EINTR && !signal_pending(current))
1534 goto sleep_again;
1535
1536 unlink_queue(sma, &queue);
1537
1538out_unlock_free:
1539 sem_unlock(sma);
1540
1541 wake_up_sem_queue_do(&tasks);
1542out_free:
1543 if(sops != fast_sops)
1544 kfree(sops);
1545 return error;
1546}
1547
1548SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
1549 unsigned, nsops)
1550{
1551 return sys_semtimedop(semid, tsops, nsops, NULL);
1552}
1553
1554/* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
1555 * parent and child tasks.
1556 */
1557
1558int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
1559{
1560 struct sem_undo_list *undo_list;
1561 int error;
1562
1563 if (clone_flags & CLONE_SYSVSEM) {
1564 error = get_undo_list(&undo_list);
1565 if (error)
1566 return error;
1567 atomic_inc(&undo_list->refcnt);
1568 tsk->sysvsem.undo_list = undo_list;
1569 } else
1570 tsk->sysvsem.undo_list = NULL;
1571
1572 return 0;
1573}
1574
1575/*
1576 * add semadj values to semaphores, free undo structures.
1577 * undo structures are not freed when semaphore arrays are destroyed
1578 * so some of them may be out of date.
1579 * IMPLEMENTATION NOTE: There is some confusion over whether the
1580 * set of adjustments that needs to be done should be done in an atomic
1581 * manner or not. That is, if we are attempting to decrement the semval
1582 * should we queue up and wait until we can do so legally?
1583 * The original implementation attempted to do this (queue and wait).
1584 * The current implementation does not do so. The POSIX standard
1585 * and SVID should be consulted to determine what behavior is mandated.
1586 */
1587void exit_sem(struct task_struct *tsk)
1588{
1589 struct sem_undo_list *ulp;
1590
1591 ulp = tsk->sysvsem.undo_list;
1592 if (!ulp)
1593 return;
1594 tsk->sysvsem.undo_list = NULL;
1595
1596 if (!atomic_dec_and_test(&ulp->refcnt))
1597 return;
1598
1599 for (;;) {
1600 struct sem_array *sma;
1601 struct sem_undo *un;
1602 struct list_head tasks;
1603 int semid;
1604 int i;
1605
1606 rcu_read_lock();
1607 un = list_entry_rcu(ulp->list_proc.next,
1608 struct sem_undo, list_proc);
1609 if (&un->list_proc == &ulp->list_proc)
1610 semid = -1;
1611 else
1612 semid = un->semid;
1613 rcu_read_unlock();
1614
1615 if (semid == -1)
1616 break;
1617
1618 sma = sem_lock_check(tsk->nsproxy->ipc_ns, un->semid);
1619
1620 /* exit_sem raced with IPC_RMID, nothing to do */
1621 if (IS_ERR(sma))
1622 continue;
1623
1624 un = __lookup_undo(ulp, semid);
1625 if (un == NULL) {
1626 /* exit_sem raced with IPC_RMID+semget() that created
1627 * exactly the same semid. Nothing to do.
1628 */
1629 sem_unlock(sma);
1630 continue;
1631 }
1632
1633 /* remove un from the linked lists */
1634 assert_spin_locked(&sma->sem_perm.lock);
1635 list_del(&un->list_id);
1636
1637 spin_lock(&ulp->lock);
1638 list_del_rcu(&un->list_proc);
1639 spin_unlock(&ulp->lock);
1640
1641 /* perform adjustments registered in un */
1642 for (i = 0; i < sma->sem_nsems; i++) {
1643 struct sem * semaphore = &sma->sem_base[i];
1644 if (un->semadj[i]) {
1645 semaphore->semval += un->semadj[i];
1646 /*
1647 * Range checks of the new semaphore value,
1648 * not defined by sus:
1649 * - Some unices ignore the undo entirely
1650 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
1651 * - some cap the value (e.g. FreeBSD caps
1652 * at 0, but doesn't enforce SEMVMX)
1653 *
1654 * Linux caps the semaphore value, both at 0
1655 * and at SEMVMX.
1656 *
1657 * Manfred <manfred@colorfullife.com>
1658 */
1659 if (semaphore->semval < 0)
1660 semaphore->semval = 0;
1661 if (semaphore->semval > SEMVMX)
1662 semaphore->semval = SEMVMX;
1663 semaphore->sempid = task_tgid_vnr(current);
1664 }
1665 }
1666 /* maybe some queued-up processes were waiting for this */
1667 INIT_LIST_HEAD(&tasks);
1668 do_smart_update(sma, NULL, 0, 1, &tasks);
1669 sem_unlock(sma);
1670 wake_up_sem_queue_do(&tasks);
1671
1672 kfree_rcu(un, rcu);
1673 }
1674 kfree(ulp);
1675}
1676
1677#ifdef CONFIG_PROC_FS
1678static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
1679{
1680 struct sem_array *sma = it;
1681
1682 return seq_printf(s,
1683 "%10d %10d %4o %10u %5u %5u %5u %5u %10lu %10lu\n",
1684 sma->sem_perm.key,
1685 sma->sem_perm.id,
1686 sma->sem_perm.mode,
1687 sma->sem_nsems,
1688 sma->sem_perm.uid,
1689 sma->sem_perm.gid,
1690 sma->sem_perm.cuid,
1691 sma->sem_perm.cgid,
1692 sma->sem_otime,
1693 sma->sem_ctime);
1694}
1695#endif
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/ipc/sem.c
4 * Copyright (C) 1992 Krishna Balasubramanian
5 * Copyright (C) 1995 Eric Schenk, Bruno Haible
6 *
7 * /proc/sysvipc/sem support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
8 *
9 * SMP-threaded, sysctl's added
10 * (c) 1999 Manfred Spraul <manfred@colorfullife.com>
11 * Enforced range limit on SEM_UNDO
12 * (c) 2001 Red Hat Inc
13 * Lockless wakeup
14 * (c) 2003 Manfred Spraul <manfred@colorfullife.com>
15 * (c) 2016 Davidlohr Bueso <dave@stgolabs.net>
16 * Further wakeup optimizations, documentation
17 * (c) 2010 Manfred Spraul <manfred@colorfullife.com>
18 *
19 * support for audit of ipc object properties and permission changes
20 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
21 *
22 * namespaces support
23 * OpenVZ, SWsoft Inc.
24 * Pavel Emelianov <xemul@openvz.org>
25 *
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.
52 * - semncnt and semzcnt are calculated on demand in count_semcnt()
53 * - 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
58 * dropping all locks. (see wake_up_sem_queue_prepare())
59 * - All work is done by the waker, the woken up task does not have to do
60 * anything - not even acquiring a lock or dropping a refcount.
61 * - A woken up task may not even touch the semaphore array anymore, it may
62 * have been destroyed already by a semctl(RMID).
63 * - 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.
71 */
72
73#include <linux/slab.h>
74#include <linux/spinlock.h>
75#include <linux/init.h>
76#include <linux/proc_fs.h>
77#include <linux/time.h>
78#include <linux/security.h>
79#include <linux/syscalls.h>
80#include <linux/audit.h>
81#include <linux/capability.h>
82#include <linux/seq_file.h>
83#include <linux/rwsem.h>
84#include <linux/nsproxy.h>
85#include <linux/ipc_namespace.h>
86#include <linux/sched/wake_q.h>
87
88#include <linux/uaccess.h>
89#include "util.h"
90
91/* One semaphore structure for each semaphore in the system. */
92struct sem {
93 int semval; /* current value */
94 /*
95 * PID of the process that last modified the semaphore. For
96 * Linux, specifically these are:
97 * - semop
98 * - semctl, via SETVAL and SETALL.
99 * - at task exit when performing undo adjustments (see exit_sem).
100 */
101 struct pid *sempid;
102 spinlock_t lock; /* spinlock for fine-grained semtimedop */
103 struct list_head pending_alter; /* pending single-sop operations */
104 /* that alter the semaphore */
105 struct list_head pending_const; /* pending single-sop operations */
106 /* that do not alter the semaphore*/
107 time_t sem_otime; /* candidate for sem_otime */
108} ____cacheline_aligned_in_smp;
109
110/* One sem_array data structure for each set of semaphores in the system. */
111struct sem_array {
112 struct kern_ipc_perm sem_perm; /* permissions .. see ipc.h */
113 time64_t sem_ctime; /* create/last semctl() time */
114 struct list_head pending_alter; /* pending operations */
115 /* that alter the array */
116 struct list_head pending_const; /* pending complex operations */
117 /* that do not alter semvals */
118 struct list_head list_id; /* undo requests on this array */
119 int sem_nsems; /* no. of semaphores in array */
120 int complex_count; /* pending complex operations */
121 unsigned int use_global_lock;/* >0: global lock required */
122
123 struct sem sems[];
124} __randomize_layout;
125
126/* One queue for each sleeping process in the system. */
127struct sem_queue {
128 struct list_head list; /* queue of pending operations */
129 struct task_struct *sleeper; /* this process */
130 struct sem_undo *undo; /* undo structure */
131 struct pid *pid; /* process id of requesting process */
132 int status; /* completion status of operation */
133 struct sembuf *sops; /* array of pending operations */
134 struct sembuf *blocking; /* the operation that blocked */
135 int nsops; /* number of operations */
136 bool alter; /* does *sops alter the array? */
137 bool dupsop; /* sops on more than one sem_num */
138};
139
140/* Each task has a list of undo requests. They are executed automatically
141 * when the process exits.
142 */
143struct sem_undo {
144 struct list_head list_proc; /* per-process list: *
145 * all undos from one process
146 * rcu protected */
147 struct rcu_head rcu; /* rcu struct for sem_undo */
148 struct sem_undo_list *ulp; /* back ptr to sem_undo_list */
149 struct list_head list_id; /* per semaphore array list:
150 * all undos for one array */
151 int semid; /* semaphore set identifier */
152 short *semadj; /* array of adjustments */
153 /* one per semaphore */
154};
155
156/* sem_undo_list controls shared access to the list of sem_undo structures
157 * that may be shared among all a CLONE_SYSVSEM task group.
158 */
159struct sem_undo_list {
160 refcount_t refcnt;
161 spinlock_t lock;
162 struct list_head list_proc;
163};
164
165
166#define sem_ids(ns) ((ns)->ids[IPC_SEM_IDS])
167
168static int newary(struct ipc_namespace *, struct ipc_params *);
169static void freeary(struct ipc_namespace *, struct kern_ipc_perm *);
170#ifdef CONFIG_PROC_FS
171static int sysvipc_sem_proc_show(struct seq_file *s, void *it);
172#endif
173
174#define SEMMSL_FAST 256 /* 512 bytes on stack */
175#define SEMOPM_FAST 64 /* ~ 372 bytes on stack */
176
177/*
178 * Switching from the mode suitable for simple ops
179 * to the mode for complex ops is costly. Therefore:
180 * use some hysteresis
181 */
182#define USE_GLOBAL_LOCK_HYSTERESIS 10
183
184/*
185 * Locking:
186 * a) global sem_lock() for read/write
187 * sem_undo.id_next,
188 * sem_array.complex_count,
189 * sem_array.pending{_alter,_const},
190 * sem_array.sem_undo
191 *
192 * b) global or semaphore sem_lock() for read/write:
193 * sem_array.sems[i].pending_{const,alter}:
194 *
195 * c) special:
196 * sem_undo_list.list_proc:
197 * * undo_list->lock for write
198 * * rcu for read
199 * use_global_lock:
200 * * global sem_lock() for write
201 * * either local or global sem_lock() for read.
202 *
203 * Memory ordering:
204 * Most ordering is enforced by using spin_lock() and spin_unlock().
205 * The special case is use_global_lock:
206 * Setting it from non-zero to 0 is a RELEASE, this is ensured by
207 * using smp_store_release().
208 * Testing if it is non-zero is an ACQUIRE, this is ensured by using
209 * smp_load_acquire().
210 * Setting it from 0 to non-zero must be ordered with regards to
211 * this smp_load_acquire(), this is guaranteed because the smp_load_acquire()
212 * is inside a spin_lock() and after a write from 0 to non-zero a
213 * spin_lock()+spin_unlock() is done.
214 */
215
216#define sc_semmsl sem_ctls[0]
217#define sc_semmns sem_ctls[1]
218#define sc_semopm sem_ctls[2]
219#define sc_semmni sem_ctls[3]
220
221int sem_init_ns(struct ipc_namespace *ns)
222{
223 ns->sc_semmsl = SEMMSL;
224 ns->sc_semmns = SEMMNS;
225 ns->sc_semopm = SEMOPM;
226 ns->sc_semmni = SEMMNI;
227 ns->used_sems = 0;
228 return ipc_init_ids(&ns->ids[IPC_SEM_IDS]);
229}
230
231#ifdef CONFIG_IPC_NS
232void sem_exit_ns(struct ipc_namespace *ns)
233{
234 free_ipcs(ns, &sem_ids(ns), freeary);
235 idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr);
236 rhashtable_destroy(&ns->ids[IPC_SEM_IDS].key_ht);
237}
238#endif
239
240int __init sem_init(void)
241{
242 const int err = sem_init_ns(&init_ipc_ns);
243
244 ipc_init_proc_interface("sysvipc/sem",
245 " key semid perms nsems uid gid cuid cgid otime ctime\n",
246 IPC_SEM_IDS, sysvipc_sem_proc_show);
247 return err;
248}
249
250/**
251 * unmerge_queues - unmerge queues, if possible.
252 * @sma: semaphore array
253 *
254 * The function unmerges the wait queues if complex_count is 0.
255 * It must be called prior to dropping the global semaphore array lock.
256 */
257static void unmerge_queues(struct sem_array *sma)
258{
259 struct sem_queue *q, *tq;
260
261 /* complex operations still around? */
262 if (sma->complex_count)
263 return;
264 /*
265 * We will switch back to simple mode.
266 * Move all pending operation back into the per-semaphore
267 * queues.
268 */
269 list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
270 struct sem *curr;
271 curr = &sma->sems[q->sops[0].sem_num];
272
273 list_add_tail(&q->list, &curr->pending_alter);
274 }
275 INIT_LIST_HEAD(&sma->pending_alter);
276}
277
278/**
279 * merge_queues - merge single semop queues into global queue
280 * @sma: semaphore array
281 *
282 * This function merges all per-semaphore queues into the global queue.
283 * It is necessary to achieve FIFO ordering for the pending single-sop
284 * operations when a multi-semop operation must sleep.
285 * Only the alter operations must be moved, the const operations can stay.
286 */
287static void merge_queues(struct sem_array *sma)
288{
289 int i;
290 for (i = 0; i < sma->sem_nsems; i++) {
291 struct sem *sem = &sma->sems[i];
292
293 list_splice_init(&sem->pending_alter, &sma->pending_alter);
294 }
295}
296
297static void sem_rcu_free(struct rcu_head *head)
298{
299 struct kern_ipc_perm *p = container_of(head, struct kern_ipc_perm, rcu);
300 struct sem_array *sma = container_of(p, struct sem_array, sem_perm);
301
302 security_sem_free(&sma->sem_perm);
303 kvfree(sma);
304}
305
306/*
307 * Enter the mode suitable for non-simple operations:
308 * Caller must own sem_perm.lock.
309 */
310static void complexmode_enter(struct sem_array *sma)
311{
312 int i;
313 struct sem *sem;
314
315 if (sma->use_global_lock > 0) {
316 /*
317 * We are already in global lock mode.
318 * Nothing to do, just reset the
319 * counter until we return to simple mode.
320 */
321 sma->use_global_lock = USE_GLOBAL_LOCK_HYSTERESIS;
322 return;
323 }
324 sma->use_global_lock = USE_GLOBAL_LOCK_HYSTERESIS;
325
326 for (i = 0; i < sma->sem_nsems; i++) {
327 sem = &sma->sems[i];
328 spin_lock(&sem->lock);
329 spin_unlock(&sem->lock);
330 }
331}
332
333/*
334 * Try to leave the mode that disallows simple operations:
335 * Caller must own sem_perm.lock.
336 */
337static void complexmode_tryleave(struct sem_array *sma)
338{
339 if (sma->complex_count) {
340 /* Complex ops are sleeping.
341 * We must stay in complex mode
342 */
343 return;
344 }
345 if (sma->use_global_lock == 1) {
346 /*
347 * Immediately after setting use_global_lock to 0,
348 * a simple op can start. Thus: all memory writes
349 * performed by the current operation must be visible
350 * before we set use_global_lock to 0.
351 */
352 smp_store_release(&sma->use_global_lock, 0);
353 } else {
354 sma->use_global_lock--;
355 }
356}
357
358#define SEM_GLOBAL_LOCK (-1)
359/*
360 * If the request contains only one semaphore operation, and there are
361 * no complex transactions pending, lock only the semaphore involved.
362 * Otherwise, lock the entire semaphore array, since we either have
363 * multiple semaphores in our own semops, or we need to look at
364 * semaphores from other pending complex operations.
365 */
366static inline int sem_lock(struct sem_array *sma, struct sembuf *sops,
367 int nsops)
368{
369 struct sem *sem;
370
371 if (nsops != 1) {
372 /* Complex operation - acquire a full lock */
373 ipc_lock_object(&sma->sem_perm);
374
375 /* Prevent parallel simple ops */
376 complexmode_enter(sma);
377 return SEM_GLOBAL_LOCK;
378 }
379
380 /*
381 * Only one semaphore affected - try to optimize locking.
382 * Optimized locking is possible if no complex operation
383 * is either enqueued or processed right now.
384 *
385 * Both facts are tracked by use_global_mode.
386 */
387 sem = &sma->sems[sops->sem_num];
388
389 /*
390 * Initial check for use_global_lock. Just an optimization,
391 * no locking, no memory barrier.
392 */
393 if (!sma->use_global_lock) {
394 /*
395 * It appears that no complex operation is around.
396 * Acquire the per-semaphore lock.
397 */
398 spin_lock(&sem->lock);
399
400 /* pairs with smp_store_release() */
401 if (!smp_load_acquire(&sma->use_global_lock)) {
402 /* fast path successful! */
403 return sops->sem_num;
404 }
405 spin_unlock(&sem->lock);
406 }
407
408 /* slow path: acquire the full lock */
409 ipc_lock_object(&sma->sem_perm);
410
411 if (sma->use_global_lock == 0) {
412 /*
413 * The use_global_lock mode ended while we waited for
414 * sma->sem_perm.lock. Thus we must switch to locking
415 * with sem->lock.
416 * Unlike in the fast path, there is no need to recheck
417 * sma->use_global_lock after we have acquired sem->lock:
418 * We own sma->sem_perm.lock, thus use_global_lock cannot
419 * change.
420 */
421 spin_lock(&sem->lock);
422
423 ipc_unlock_object(&sma->sem_perm);
424 return sops->sem_num;
425 } else {
426 /*
427 * Not a false alarm, thus continue to use the global lock
428 * mode. No need for complexmode_enter(), this was done by
429 * the caller that has set use_global_mode to non-zero.
430 */
431 return SEM_GLOBAL_LOCK;
432 }
433}
434
435static inline void sem_unlock(struct sem_array *sma, int locknum)
436{
437 if (locknum == SEM_GLOBAL_LOCK) {
438 unmerge_queues(sma);
439 complexmode_tryleave(sma);
440 ipc_unlock_object(&sma->sem_perm);
441 } else {
442 struct sem *sem = &sma->sems[locknum];
443 spin_unlock(&sem->lock);
444 }
445}
446
447/*
448 * sem_lock_(check_) routines are called in the paths where the rwsem
449 * is not held.
450 *
451 * The caller holds the RCU read lock.
452 */
453static inline struct sem_array *sem_obtain_object(struct ipc_namespace *ns, int id)
454{
455 struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&sem_ids(ns), id);
456
457 if (IS_ERR(ipcp))
458 return ERR_CAST(ipcp);
459
460 return container_of(ipcp, struct sem_array, sem_perm);
461}
462
463static inline struct sem_array *sem_obtain_object_check(struct ipc_namespace *ns,
464 int id)
465{
466 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&sem_ids(ns), id);
467
468 if (IS_ERR(ipcp))
469 return ERR_CAST(ipcp);
470
471 return container_of(ipcp, struct sem_array, sem_perm);
472}
473
474static inline void sem_lock_and_putref(struct sem_array *sma)
475{
476 sem_lock(sma, NULL, -1);
477 ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
478}
479
480static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
481{
482 ipc_rmid(&sem_ids(ns), &s->sem_perm);
483}
484
485static struct sem_array *sem_alloc(size_t nsems)
486{
487 struct sem_array *sma;
488 size_t size;
489
490 if (nsems > (INT_MAX - sizeof(*sma)) / sizeof(sma->sems[0]))
491 return NULL;
492
493 size = sizeof(*sma) + nsems * sizeof(sma->sems[0]);
494 sma = kvmalloc(size, GFP_KERNEL);
495 if (unlikely(!sma))
496 return NULL;
497
498 memset(sma, 0, size);
499
500 return sma;
501}
502
503/**
504 * newary - Create a new semaphore set
505 * @ns: namespace
506 * @params: ptr to the structure that contains key, semflg and nsems
507 *
508 * Called with sem_ids.rwsem held (as a writer)
509 */
510static int newary(struct ipc_namespace *ns, struct ipc_params *params)
511{
512 int retval;
513 struct sem_array *sma;
514 key_t key = params->key;
515 int nsems = params->u.nsems;
516 int semflg = params->flg;
517 int i;
518
519 if (!nsems)
520 return -EINVAL;
521 if (ns->used_sems + nsems > ns->sc_semmns)
522 return -ENOSPC;
523
524 sma = sem_alloc(nsems);
525 if (!sma)
526 return -ENOMEM;
527
528 sma->sem_perm.mode = (semflg & S_IRWXUGO);
529 sma->sem_perm.key = key;
530
531 sma->sem_perm.security = NULL;
532 retval = security_sem_alloc(&sma->sem_perm);
533 if (retval) {
534 kvfree(sma);
535 return retval;
536 }
537
538 for (i = 0; i < nsems; i++) {
539 INIT_LIST_HEAD(&sma->sems[i].pending_alter);
540 INIT_LIST_HEAD(&sma->sems[i].pending_const);
541 spin_lock_init(&sma->sems[i].lock);
542 }
543
544 sma->complex_count = 0;
545 sma->use_global_lock = USE_GLOBAL_LOCK_HYSTERESIS;
546 INIT_LIST_HEAD(&sma->pending_alter);
547 INIT_LIST_HEAD(&sma->pending_const);
548 INIT_LIST_HEAD(&sma->list_id);
549 sma->sem_nsems = nsems;
550 sma->sem_ctime = ktime_get_real_seconds();
551
552 /* ipc_addid() locks sma upon success. */
553 retval = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
554 if (retval < 0) {
555 call_rcu(&sma->sem_perm.rcu, sem_rcu_free);
556 return retval;
557 }
558 ns->used_sems += nsems;
559
560 sem_unlock(sma, -1);
561 rcu_read_unlock();
562
563 return sma->sem_perm.id;
564}
565
566
567/*
568 * Called with sem_ids.rwsem and ipcp locked.
569 */
570static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
571 struct ipc_params *params)
572{
573 struct sem_array *sma;
574
575 sma = container_of(ipcp, struct sem_array, sem_perm);
576 if (params->u.nsems > sma->sem_nsems)
577 return -EINVAL;
578
579 return 0;
580}
581
582long ksys_semget(key_t key, int nsems, int semflg)
583{
584 struct ipc_namespace *ns;
585 static const struct ipc_ops sem_ops = {
586 .getnew = newary,
587 .associate = security_sem_associate,
588 .more_checks = sem_more_checks,
589 };
590 struct ipc_params sem_params;
591
592 ns = current->nsproxy->ipc_ns;
593
594 if (nsems < 0 || nsems > ns->sc_semmsl)
595 return -EINVAL;
596
597 sem_params.key = key;
598 sem_params.flg = semflg;
599 sem_params.u.nsems = nsems;
600
601 return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
602}
603
604SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg)
605{
606 return ksys_semget(key, nsems, semflg);
607}
608
609/**
610 * perform_atomic_semop[_slow] - Attempt to perform semaphore
611 * operations on a given array.
612 * @sma: semaphore array
613 * @q: struct sem_queue that describes the operation
614 *
615 * Caller blocking are as follows, based the value
616 * indicated by the semaphore operation (sem_op):
617 *
618 * (1) >0 never blocks.
619 * (2) 0 (wait-for-zero operation): semval is non-zero.
620 * (3) <0 attempting to decrement semval to a value smaller than zero.
621 *
622 * Returns 0 if the operation was possible.
623 * Returns 1 if the operation is impossible, the caller must sleep.
624 * Returns <0 for error codes.
625 */
626static int perform_atomic_semop_slow(struct sem_array *sma, struct sem_queue *q)
627{
628 int result, sem_op, nsops;
629 struct pid *pid;
630 struct sembuf *sop;
631 struct sem *curr;
632 struct sembuf *sops;
633 struct sem_undo *un;
634
635 sops = q->sops;
636 nsops = q->nsops;
637 un = q->undo;
638
639 for (sop = sops; sop < sops + nsops; sop++) {
640 curr = &sma->sems[sop->sem_num];
641 sem_op = sop->sem_op;
642 result = curr->semval;
643
644 if (!sem_op && result)
645 goto would_block;
646
647 result += sem_op;
648 if (result < 0)
649 goto would_block;
650 if (result > SEMVMX)
651 goto out_of_range;
652
653 if (sop->sem_flg & SEM_UNDO) {
654 int undo = un->semadj[sop->sem_num] - sem_op;
655 /* Exceeding the undo range is an error. */
656 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
657 goto out_of_range;
658 un->semadj[sop->sem_num] = undo;
659 }
660
661 curr->semval = result;
662 }
663
664 sop--;
665 pid = q->pid;
666 while (sop >= sops) {
667 ipc_update_pid(&sma->sems[sop->sem_num].sempid, pid);
668 sop--;
669 }
670
671 return 0;
672
673out_of_range:
674 result = -ERANGE;
675 goto undo;
676
677would_block:
678 q->blocking = sop;
679
680 if (sop->sem_flg & IPC_NOWAIT)
681 result = -EAGAIN;
682 else
683 result = 1;
684
685undo:
686 sop--;
687 while (sop >= sops) {
688 sem_op = sop->sem_op;
689 sma->sems[sop->sem_num].semval -= sem_op;
690 if (sop->sem_flg & SEM_UNDO)
691 un->semadj[sop->sem_num] += sem_op;
692 sop--;
693 }
694
695 return result;
696}
697
698static int perform_atomic_semop(struct sem_array *sma, struct sem_queue *q)
699{
700 int result, sem_op, nsops;
701 struct sembuf *sop;
702 struct sem *curr;
703 struct sembuf *sops;
704 struct sem_undo *un;
705
706 sops = q->sops;
707 nsops = q->nsops;
708 un = q->undo;
709
710 if (unlikely(q->dupsop))
711 return perform_atomic_semop_slow(sma, q);
712
713 /*
714 * We scan the semaphore set twice, first to ensure that the entire
715 * operation can succeed, therefore avoiding any pointless writes
716 * to shared memory and having to undo such changes in order to block
717 * until the operations can go through.
718 */
719 for (sop = sops; sop < sops + nsops; sop++) {
720 curr = &sma->sems[sop->sem_num];
721 sem_op = sop->sem_op;
722 result = curr->semval;
723
724 if (!sem_op && result)
725 goto would_block; /* wait-for-zero */
726
727 result += sem_op;
728 if (result < 0)
729 goto would_block;
730
731 if (result > SEMVMX)
732 return -ERANGE;
733
734 if (sop->sem_flg & SEM_UNDO) {
735 int undo = un->semadj[sop->sem_num] - sem_op;
736
737 /* Exceeding the undo range is an error. */
738 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
739 return -ERANGE;
740 }
741 }
742
743 for (sop = sops; sop < sops + nsops; sop++) {
744 curr = &sma->sems[sop->sem_num];
745 sem_op = sop->sem_op;
746 result = curr->semval;
747
748 if (sop->sem_flg & SEM_UNDO) {
749 int undo = un->semadj[sop->sem_num] - sem_op;
750
751 un->semadj[sop->sem_num] = undo;
752 }
753 curr->semval += sem_op;
754 ipc_update_pid(&curr->sempid, q->pid);
755 }
756
757 return 0;
758
759would_block:
760 q->blocking = sop;
761 return sop->sem_flg & IPC_NOWAIT ? -EAGAIN : 1;
762}
763
764static inline void wake_up_sem_queue_prepare(struct sem_queue *q, int error,
765 struct wake_q_head *wake_q)
766{
767 wake_q_add(wake_q, q->sleeper);
768 /*
769 * Rely on the above implicit barrier, such that we can
770 * ensure that we hold reference to the task before setting
771 * q->status. Otherwise we could race with do_exit if the
772 * task is awoken by an external event before calling
773 * wake_up_process().
774 */
775 WRITE_ONCE(q->status, error);
776}
777
778static void unlink_queue(struct sem_array *sma, struct sem_queue *q)
779{
780 list_del(&q->list);
781 if (q->nsops > 1)
782 sma->complex_count--;
783}
784
785/** check_restart(sma, q)
786 * @sma: semaphore array
787 * @q: the operation that just completed
788 *
789 * update_queue is O(N^2) when it restarts scanning the whole queue of
790 * waiting operations. Therefore this function checks if the restart is
791 * really necessary. It is called after a previously waiting operation
792 * modified the array.
793 * Note that wait-for-zero operations are handled without restart.
794 */
795static inline int check_restart(struct sem_array *sma, struct sem_queue *q)
796{
797 /* pending complex alter operations are too difficult to analyse */
798 if (!list_empty(&sma->pending_alter))
799 return 1;
800
801 /* we were a sleeping complex operation. Too difficult */
802 if (q->nsops > 1)
803 return 1;
804
805 /* It is impossible that someone waits for the new value:
806 * - complex operations always restart.
807 * - wait-for-zero are handled seperately.
808 * - q is a previously sleeping simple operation that
809 * altered the array. It must be a decrement, because
810 * simple increments never sleep.
811 * - If there are older (higher priority) decrements
812 * in the queue, then they have observed the original
813 * semval value and couldn't proceed. The operation
814 * decremented to value - thus they won't proceed either.
815 */
816 return 0;
817}
818
819/**
820 * wake_const_ops - wake up non-alter tasks
821 * @sma: semaphore array.
822 * @semnum: semaphore that was modified.
823 * @wake_q: lockless wake-queue head.
824 *
825 * wake_const_ops must be called after a semaphore in a semaphore array
826 * was set to 0. If complex const operations are pending, wake_const_ops must
827 * be called with semnum = -1, as well as with the number of each modified
828 * semaphore.
829 * The tasks that must be woken up are added to @wake_q. The return code
830 * is stored in q->pid.
831 * The function returns 1 if at least one operation was completed successfully.
832 */
833static int wake_const_ops(struct sem_array *sma, int semnum,
834 struct wake_q_head *wake_q)
835{
836 struct sem_queue *q, *tmp;
837 struct list_head *pending_list;
838 int semop_completed = 0;
839
840 if (semnum == -1)
841 pending_list = &sma->pending_const;
842 else
843 pending_list = &sma->sems[semnum].pending_const;
844
845 list_for_each_entry_safe(q, tmp, pending_list, list) {
846 int error = perform_atomic_semop(sma, q);
847
848 if (error > 0)
849 continue;
850 /* operation completed, remove from queue & wakeup */
851 unlink_queue(sma, q);
852
853 wake_up_sem_queue_prepare(q, error, wake_q);
854 if (error == 0)
855 semop_completed = 1;
856 }
857
858 return semop_completed;
859}
860
861/**
862 * do_smart_wakeup_zero - wakeup all wait for zero tasks
863 * @sma: semaphore array
864 * @sops: operations that were performed
865 * @nsops: number of operations
866 * @wake_q: lockless wake-queue head
867 *
868 * Checks all required queue for wait-for-zero operations, based
869 * on the actual changes that were performed on the semaphore array.
870 * The function returns 1 if at least one operation was completed successfully.
871 */
872static int do_smart_wakeup_zero(struct sem_array *sma, struct sembuf *sops,
873 int nsops, struct wake_q_head *wake_q)
874{
875 int i;
876 int semop_completed = 0;
877 int got_zero = 0;
878
879 /* first: the per-semaphore queues, if known */
880 if (sops) {
881 for (i = 0; i < nsops; i++) {
882 int num = sops[i].sem_num;
883
884 if (sma->sems[num].semval == 0) {
885 got_zero = 1;
886 semop_completed |= wake_const_ops(sma, num, wake_q);
887 }
888 }
889 } else {
890 /*
891 * No sops means modified semaphores not known.
892 * Assume all were changed.
893 */
894 for (i = 0; i < sma->sem_nsems; i++) {
895 if (sma->sems[i].semval == 0) {
896 got_zero = 1;
897 semop_completed |= wake_const_ops(sma, i, wake_q);
898 }
899 }
900 }
901 /*
902 * If one of the modified semaphores got 0,
903 * then check the global queue, too.
904 */
905 if (got_zero)
906 semop_completed |= wake_const_ops(sma, -1, wake_q);
907
908 return semop_completed;
909}
910
911
912/**
913 * update_queue - look for tasks that can be completed.
914 * @sma: semaphore array.
915 * @semnum: semaphore that was modified.
916 * @wake_q: lockless wake-queue head.
917 *
918 * update_queue must be called after a semaphore in a semaphore array
919 * was modified. If multiple semaphores were modified, update_queue must
920 * be called with semnum = -1, as well as with the number of each modified
921 * semaphore.
922 * The tasks that must be woken up are added to @wake_q. The return code
923 * is stored in q->pid.
924 * The function internally checks if const operations can now succeed.
925 *
926 * The function return 1 if at least one semop was completed successfully.
927 */
928static int update_queue(struct sem_array *sma, int semnum, struct wake_q_head *wake_q)
929{
930 struct sem_queue *q, *tmp;
931 struct list_head *pending_list;
932 int semop_completed = 0;
933
934 if (semnum == -1)
935 pending_list = &sma->pending_alter;
936 else
937 pending_list = &sma->sems[semnum].pending_alter;
938
939again:
940 list_for_each_entry_safe(q, tmp, pending_list, list) {
941 int error, restart;
942
943 /* If we are scanning the single sop, per-semaphore list of
944 * one semaphore and that semaphore is 0, then it is not
945 * necessary to scan further: simple increments
946 * that affect only one entry succeed immediately and cannot
947 * be in the per semaphore pending queue, and decrements
948 * cannot be successful if the value is already 0.
949 */
950 if (semnum != -1 && sma->sems[semnum].semval == 0)
951 break;
952
953 error = perform_atomic_semop(sma, q);
954
955 /* Does q->sleeper still need to sleep? */
956 if (error > 0)
957 continue;
958
959 unlink_queue(sma, q);
960
961 if (error) {
962 restart = 0;
963 } else {
964 semop_completed = 1;
965 do_smart_wakeup_zero(sma, q->sops, q->nsops, wake_q);
966 restart = check_restart(sma, q);
967 }
968
969 wake_up_sem_queue_prepare(q, error, wake_q);
970 if (restart)
971 goto again;
972 }
973 return semop_completed;
974}
975
976/**
977 * set_semotime - set sem_otime
978 * @sma: semaphore array
979 * @sops: operations that modified the array, may be NULL
980 *
981 * sem_otime is replicated to avoid cache line trashing.
982 * This function sets one instance to the current time.
983 */
984static void set_semotime(struct sem_array *sma, struct sembuf *sops)
985{
986 if (sops == NULL) {
987 sma->sems[0].sem_otime = get_seconds();
988 } else {
989 sma->sems[sops[0].sem_num].sem_otime =
990 get_seconds();
991 }
992}
993
994/**
995 * do_smart_update - optimized update_queue
996 * @sma: semaphore array
997 * @sops: operations that were performed
998 * @nsops: number of operations
999 * @otime: force setting otime
1000 * @wake_q: lockless wake-queue head
1001 *
1002 * do_smart_update() does the required calls to update_queue and wakeup_zero,
1003 * based on the actual changes that were performed on the semaphore array.
1004 * Note that the function does not do the actual wake-up: the caller is
1005 * responsible for calling wake_up_q().
1006 * It is safe to perform this call after dropping all locks.
1007 */
1008static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops,
1009 int otime, struct wake_q_head *wake_q)
1010{
1011 int i;
1012
1013 otime |= do_smart_wakeup_zero(sma, sops, nsops, wake_q);
1014
1015 if (!list_empty(&sma->pending_alter)) {
1016 /* semaphore array uses the global queue - just process it. */
1017 otime |= update_queue(sma, -1, wake_q);
1018 } else {
1019 if (!sops) {
1020 /*
1021 * No sops, thus the modified semaphores are not
1022 * known. Check all.
1023 */
1024 for (i = 0; i < sma->sem_nsems; i++)
1025 otime |= update_queue(sma, i, wake_q);
1026 } else {
1027 /*
1028 * Check the semaphores that were increased:
1029 * - No complex ops, thus all sleeping ops are
1030 * decrease.
1031 * - if we decreased the value, then any sleeping
1032 * semaphore ops wont be able to run: If the
1033 * previous value was too small, then the new
1034 * value will be too small, too.
1035 */
1036 for (i = 0; i < nsops; i++) {
1037 if (sops[i].sem_op > 0) {
1038 otime |= update_queue(sma,
1039 sops[i].sem_num, wake_q);
1040 }
1041 }
1042 }
1043 }
1044 if (otime)
1045 set_semotime(sma, sops);
1046}
1047
1048/*
1049 * check_qop: Test if a queued operation sleeps on the semaphore semnum
1050 */
1051static int check_qop(struct sem_array *sma, int semnum, struct sem_queue *q,
1052 bool count_zero)
1053{
1054 struct sembuf *sop = q->blocking;
1055
1056 /*
1057 * Linux always (since 0.99.10) reported a task as sleeping on all
1058 * semaphores. This violates SUS, therefore it was changed to the
1059 * standard compliant behavior.
1060 * Give the administrators a chance to notice that an application
1061 * might misbehave because it relies on the Linux behavior.
1062 */
1063 pr_info_once("semctl(GETNCNT/GETZCNT) is since 3.16 Single Unix Specification compliant.\n"
1064 "The task %s (%d) triggered the difference, watch for misbehavior.\n",
1065 current->comm, task_pid_nr(current));
1066
1067 if (sop->sem_num != semnum)
1068 return 0;
1069
1070 if (count_zero && sop->sem_op == 0)
1071 return 1;
1072 if (!count_zero && sop->sem_op < 0)
1073 return 1;
1074
1075 return 0;
1076}
1077
1078/* The following counts are associated to each semaphore:
1079 * semncnt number of tasks waiting on semval being nonzero
1080 * semzcnt number of tasks waiting on semval being zero
1081 *
1082 * Per definition, a task waits only on the semaphore of the first semop
1083 * that cannot proceed, even if additional operation would block, too.
1084 */
1085static int count_semcnt(struct sem_array *sma, ushort semnum,
1086 bool count_zero)
1087{
1088 struct list_head *l;
1089 struct sem_queue *q;
1090 int semcnt;
1091
1092 semcnt = 0;
1093 /* First: check the simple operations. They are easy to evaluate */
1094 if (count_zero)
1095 l = &sma->sems[semnum].pending_const;
1096 else
1097 l = &sma->sems[semnum].pending_alter;
1098
1099 list_for_each_entry(q, l, list) {
1100 /* all task on a per-semaphore list sleep on exactly
1101 * that semaphore
1102 */
1103 semcnt++;
1104 }
1105
1106 /* Then: check the complex operations. */
1107 list_for_each_entry(q, &sma->pending_alter, list) {
1108 semcnt += check_qop(sma, semnum, q, count_zero);
1109 }
1110 if (count_zero) {
1111 list_for_each_entry(q, &sma->pending_const, list) {
1112 semcnt += check_qop(sma, semnum, q, count_zero);
1113 }
1114 }
1115 return semcnt;
1116}
1117
1118/* Free a semaphore set. freeary() is called with sem_ids.rwsem locked
1119 * as a writer and the spinlock for this semaphore set hold. sem_ids.rwsem
1120 * remains locked on exit.
1121 */
1122static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
1123{
1124 struct sem_undo *un, *tu;
1125 struct sem_queue *q, *tq;
1126 struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
1127 int i;
1128 DEFINE_WAKE_Q(wake_q);
1129
1130 /* Free the existing undo structures for this semaphore set. */
1131 ipc_assert_locked_object(&sma->sem_perm);
1132 list_for_each_entry_safe(un, tu, &sma->list_id, list_id) {
1133 list_del(&un->list_id);
1134 spin_lock(&un->ulp->lock);
1135 un->semid = -1;
1136 list_del_rcu(&un->list_proc);
1137 spin_unlock(&un->ulp->lock);
1138 kfree_rcu(un, rcu);
1139 }
1140
1141 /* Wake up all pending processes and let them fail with EIDRM. */
1142 list_for_each_entry_safe(q, tq, &sma->pending_const, list) {
1143 unlink_queue(sma, q);
1144 wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
1145 }
1146
1147 list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
1148 unlink_queue(sma, q);
1149 wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
1150 }
1151 for (i = 0; i < sma->sem_nsems; i++) {
1152 struct sem *sem = &sma->sems[i];
1153 list_for_each_entry_safe(q, tq, &sem->pending_const, list) {
1154 unlink_queue(sma, q);
1155 wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
1156 }
1157 list_for_each_entry_safe(q, tq, &sem->pending_alter, list) {
1158 unlink_queue(sma, q);
1159 wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
1160 }
1161 ipc_update_pid(&sem->sempid, NULL);
1162 }
1163
1164 /* Remove the semaphore set from the IDR */
1165 sem_rmid(ns, sma);
1166 sem_unlock(sma, -1);
1167 rcu_read_unlock();
1168
1169 wake_up_q(&wake_q);
1170 ns->used_sems -= sma->sem_nsems;
1171 ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
1172}
1173
1174static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
1175{
1176 switch (version) {
1177 case IPC_64:
1178 return copy_to_user(buf, in, sizeof(*in));
1179 case IPC_OLD:
1180 {
1181 struct semid_ds out;
1182
1183 memset(&out, 0, sizeof(out));
1184
1185 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
1186
1187 out.sem_otime = in->sem_otime;
1188 out.sem_ctime = in->sem_ctime;
1189 out.sem_nsems = in->sem_nsems;
1190
1191 return copy_to_user(buf, &out, sizeof(out));
1192 }
1193 default:
1194 return -EINVAL;
1195 }
1196}
1197
1198static time64_t get_semotime(struct sem_array *sma)
1199{
1200 int i;
1201 time64_t res;
1202
1203 res = sma->sems[0].sem_otime;
1204 for (i = 1; i < sma->sem_nsems; i++) {
1205 time64_t to = sma->sems[i].sem_otime;
1206
1207 if (to > res)
1208 res = to;
1209 }
1210 return res;
1211}
1212
1213static int semctl_stat(struct ipc_namespace *ns, int semid,
1214 int cmd, struct semid64_ds *semid64)
1215{
1216 struct sem_array *sma;
1217 int id = 0;
1218 int err;
1219
1220 memset(semid64, 0, sizeof(*semid64));
1221
1222 rcu_read_lock();
1223 if (cmd == SEM_STAT || cmd == SEM_STAT_ANY) {
1224 sma = sem_obtain_object(ns, semid);
1225 if (IS_ERR(sma)) {
1226 err = PTR_ERR(sma);
1227 goto out_unlock;
1228 }
1229 id = sma->sem_perm.id;
1230 } else { /* IPC_STAT */
1231 sma = sem_obtain_object_check(ns, semid);
1232 if (IS_ERR(sma)) {
1233 err = PTR_ERR(sma);
1234 goto out_unlock;
1235 }
1236 }
1237
1238 /* see comment for SHM_STAT_ANY */
1239 if (cmd == SEM_STAT_ANY)
1240 audit_ipc_obj(&sma->sem_perm);
1241 else {
1242 err = -EACCES;
1243 if (ipcperms(ns, &sma->sem_perm, S_IRUGO))
1244 goto out_unlock;
1245 }
1246
1247 err = security_sem_semctl(&sma->sem_perm, cmd);
1248 if (err)
1249 goto out_unlock;
1250
1251 ipc_lock_object(&sma->sem_perm);
1252
1253 if (!ipc_valid_object(&sma->sem_perm)) {
1254 ipc_unlock_object(&sma->sem_perm);
1255 err = -EIDRM;
1256 goto out_unlock;
1257 }
1258
1259 kernel_to_ipc64_perm(&sma->sem_perm, &semid64->sem_perm);
1260 semid64->sem_otime = get_semotime(sma);
1261 semid64->sem_ctime = sma->sem_ctime;
1262 semid64->sem_nsems = sma->sem_nsems;
1263
1264 ipc_unlock_object(&sma->sem_perm);
1265 rcu_read_unlock();
1266 return id;
1267
1268out_unlock:
1269 rcu_read_unlock();
1270 return err;
1271}
1272
1273static int semctl_info(struct ipc_namespace *ns, int semid,
1274 int cmd, void __user *p)
1275{
1276 struct seminfo seminfo;
1277 int max_id;
1278 int err;
1279
1280 err = security_sem_semctl(NULL, cmd);
1281 if (err)
1282 return err;
1283
1284 memset(&seminfo, 0, sizeof(seminfo));
1285 seminfo.semmni = ns->sc_semmni;
1286 seminfo.semmns = ns->sc_semmns;
1287 seminfo.semmsl = ns->sc_semmsl;
1288 seminfo.semopm = ns->sc_semopm;
1289 seminfo.semvmx = SEMVMX;
1290 seminfo.semmnu = SEMMNU;
1291 seminfo.semmap = SEMMAP;
1292 seminfo.semume = SEMUME;
1293 down_read(&sem_ids(ns).rwsem);
1294 if (cmd == SEM_INFO) {
1295 seminfo.semusz = sem_ids(ns).in_use;
1296 seminfo.semaem = ns->used_sems;
1297 } else {
1298 seminfo.semusz = SEMUSZ;
1299 seminfo.semaem = SEMAEM;
1300 }
1301 max_id = ipc_get_maxid(&sem_ids(ns));
1302 up_read(&sem_ids(ns).rwsem);
1303 if (copy_to_user(p, &seminfo, sizeof(struct seminfo)))
1304 return -EFAULT;
1305 return (max_id < 0) ? 0 : max_id;
1306}
1307
1308static int semctl_setval(struct ipc_namespace *ns, int semid, int semnum,
1309 int val)
1310{
1311 struct sem_undo *un;
1312 struct sem_array *sma;
1313 struct sem *curr;
1314 int err;
1315 DEFINE_WAKE_Q(wake_q);
1316
1317 if (val > SEMVMX || val < 0)
1318 return -ERANGE;
1319
1320 rcu_read_lock();
1321 sma = sem_obtain_object_check(ns, semid);
1322 if (IS_ERR(sma)) {
1323 rcu_read_unlock();
1324 return PTR_ERR(sma);
1325 }
1326
1327 if (semnum < 0 || semnum >= sma->sem_nsems) {
1328 rcu_read_unlock();
1329 return -EINVAL;
1330 }
1331
1332
1333 if (ipcperms(ns, &sma->sem_perm, S_IWUGO)) {
1334 rcu_read_unlock();
1335 return -EACCES;
1336 }
1337
1338 err = security_sem_semctl(&sma->sem_perm, SETVAL);
1339 if (err) {
1340 rcu_read_unlock();
1341 return -EACCES;
1342 }
1343
1344 sem_lock(sma, NULL, -1);
1345
1346 if (!ipc_valid_object(&sma->sem_perm)) {
1347 sem_unlock(sma, -1);
1348 rcu_read_unlock();
1349 return -EIDRM;
1350 }
1351
1352 curr = &sma->sems[semnum];
1353
1354 ipc_assert_locked_object(&sma->sem_perm);
1355 list_for_each_entry(un, &sma->list_id, list_id)
1356 un->semadj[semnum] = 0;
1357
1358 curr->semval = val;
1359 ipc_update_pid(&curr->sempid, task_tgid(current));
1360 sma->sem_ctime = ktime_get_real_seconds();
1361 /* maybe some queued-up processes were waiting for this */
1362 do_smart_update(sma, NULL, 0, 0, &wake_q);
1363 sem_unlock(sma, -1);
1364 rcu_read_unlock();
1365 wake_up_q(&wake_q);
1366 return 0;
1367}
1368
1369static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
1370 int cmd, void __user *p)
1371{
1372 struct sem_array *sma;
1373 struct sem *curr;
1374 int err, nsems;
1375 ushort fast_sem_io[SEMMSL_FAST];
1376 ushort *sem_io = fast_sem_io;
1377 DEFINE_WAKE_Q(wake_q);
1378
1379 rcu_read_lock();
1380 sma = sem_obtain_object_check(ns, semid);
1381 if (IS_ERR(sma)) {
1382 rcu_read_unlock();
1383 return PTR_ERR(sma);
1384 }
1385
1386 nsems = sma->sem_nsems;
1387
1388 err = -EACCES;
1389 if (ipcperms(ns, &sma->sem_perm, cmd == SETALL ? S_IWUGO : S_IRUGO))
1390 goto out_rcu_wakeup;
1391
1392 err = security_sem_semctl(&sma->sem_perm, cmd);
1393 if (err)
1394 goto out_rcu_wakeup;
1395
1396 err = -EACCES;
1397 switch (cmd) {
1398 case GETALL:
1399 {
1400 ushort __user *array = p;
1401 int i;
1402
1403 sem_lock(sma, NULL, -1);
1404 if (!ipc_valid_object(&sma->sem_perm)) {
1405 err = -EIDRM;
1406 goto out_unlock;
1407 }
1408 if (nsems > SEMMSL_FAST) {
1409 if (!ipc_rcu_getref(&sma->sem_perm)) {
1410 err = -EIDRM;
1411 goto out_unlock;
1412 }
1413 sem_unlock(sma, -1);
1414 rcu_read_unlock();
1415 sem_io = kvmalloc_array(nsems, sizeof(ushort),
1416 GFP_KERNEL);
1417 if (sem_io == NULL) {
1418 ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
1419 return -ENOMEM;
1420 }
1421
1422 rcu_read_lock();
1423 sem_lock_and_putref(sma);
1424 if (!ipc_valid_object(&sma->sem_perm)) {
1425 err = -EIDRM;
1426 goto out_unlock;
1427 }
1428 }
1429 for (i = 0; i < sma->sem_nsems; i++)
1430 sem_io[i] = sma->sems[i].semval;
1431 sem_unlock(sma, -1);
1432 rcu_read_unlock();
1433 err = 0;
1434 if (copy_to_user(array, sem_io, nsems*sizeof(ushort)))
1435 err = -EFAULT;
1436 goto out_free;
1437 }
1438 case SETALL:
1439 {
1440 int i;
1441 struct sem_undo *un;
1442
1443 if (!ipc_rcu_getref(&sma->sem_perm)) {
1444 err = -EIDRM;
1445 goto out_rcu_wakeup;
1446 }
1447 rcu_read_unlock();
1448
1449 if (nsems > SEMMSL_FAST) {
1450 sem_io = kvmalloc_array(nsems, sizeof(ushort),
1451 GFP_KERNEL);
1452 if (sem_io == NULL) {
1453 ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
1454 return -ENOMEM;
1455 }
1456 }
1457
1458 if (copy_from_user(sem_io, p, nsems*sizeof(ushort))) {
1459 ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
1460 err = -EFAULT;
1461 goto out_free;
1462 }
1463
1464 for (i = 0; i < nsems; i++) {
1465 if (sem_io[i] > SEMVMX) {
1466 ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
1467 err = -ERANGE;
1468 goto out_free;
1469 }
1470 }
1471 rcu_read_lock();
1472 sem_lock_and_putref(sma);
1473 if (!ipc_valid_object(&sma->sem_perm)) {
1474 err = -EIDRM;
1475 goto out_unlock;
1476 }
1477
1478 for (i = 0; i < nsems; i++) {
1479 sma->sems[i].semval = sem_io[i];
1480 ipc_update_pid(&sma->sems[i].sempid, task_tgid(current));
1481 }
1482
1483 ipc_assert_locked_object(&sma->sem_perm);
1484 list_for_each_entry(un, &sma->list_id, list_id) {
1485 for (i = 0; i < nsems; i++)
1486 un->semadj[i] = 0;
1487 }
1488 sma->sem_ctime = ktime_get_real_seconds();
1489 /* maybe some queued-up processes were waiting for this */
1490 do_smart_update(sma, NULL, 0, 0, &wake_q);
1491 err = 0;
1492 goto out_unlock;
1493 }
1494 /* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */
1495 }
1496 err = -EINVAL;
1497 if (semnum < 0 || semnum >= nsems)
1498 goto out_rcu_wakeup;
1499
1500 sem_lock(sma, NULL, -1);
1501 if (!ipc_valid_object(&sma->sem_perm)) {
1502 err = -EIDRM;
1503 goto out_unlock;
1504 }
1505 curr = &sma->sems[semnum];
1506
1507 switch (cmd) {
1508 case GETVAL:
1509 err = curr->semval;
1510 goto out_unlock;
1511 case GETPID:
1512 err = pid_vnr(curr->sempid);
1513 goto out_unlock;
1514 case GETNCNT:
1515 err = count_semcnt(sma, semnum, 0);
1516 goto out_unlock;
1517 case GETZCNT:
1518 err = count_semcnt(sma, semnum, 1);
1519 goto out_unlock;
1520 }
1521
1522out_unlock:
1523 sem_unlock(sma, -1);
1524out_rcu_wakeup:
1525 rcu_read_unlock();
1526 wake_up_q(&wake_q);
1527out_free:
1528 if (sem_io != fast_sem_io)
1529 kvfree(sem_io);
1530 return err;
1531}
1532
1533static inline unsigned long
1534copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
1535{
1536 switch (version) {
1537 case IPC_64:
1538 if (copy_from_user(out, buf, sizeof(*out)))
1539 return -EFAULT;
1540 return 0;
1541 case IPC_OLD:
1542 {
1543 struct semid_ds tbuf_old;
1544
1545 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
1546 return -EFAULT;
1547
1548 out->sem_perm.uid = tbuf_old.sem_perm.uid;
1549 out->sem_perm.gid = tbuf_old.sem_perm.gid;
1550 out->sem_perm.mode = tbuf_old.sem_perm.mode;
1551
1552 return 0;
1553 }
1554 default:
1555 return -EINVAL;
1556 }
1557}
1558
1559/*
1560 * This function handles some semctl commands which require the rwsem
1561 * to be held in write mode.
1562 * NOTE: no locks must be held, the rwsem is taken inside this function.
1563 */
1564static int semctl_down(struct ipc_namespace *ns, int semid,
1565 int cmd, struct semid64_ds *semid64)
1566{
1567 struct sem_array *sma;
1568 int err;
1569 struct kern_ipc_perm *ipcp;
1570
1571 down_write(&sem_ids(ns).rwsem);
1572 rcu_read_lock();
1573
1574 ipcp = ipcctl_pre_down_nolock(ns, &sem_ids(ns), semid, cmd,
1575 &semid64->sem_perm, 0);
1576 if (IS_ERR(ipcp)) {
1577 err = PTR_ERR(ipcp);
1578 goto out_unlock1;
1579 }
1580
1581 sma = container_of(ipcp, struct sem_array, sem_perm);
1582
1583 err = security_sem_semctl(&sma->sem_perm, cmd);
1584 if (err)
1585 goto out_unlock1;
1586
1587 switch (cmd) {
1588 case IPC_RMID:
1589 sem_lock(sma, NULL, -1);
1590 /* freeary unlocks the ipc object and rcu */
1591 freeary(ns, ipcp);
1592 goto out_up;
1593 case IPC_SET:
1594 sem_lock(sma, NULL, -1);
1595 err = ipc_update_perm(&semid64->sem_perm, ipcp);
1596 if (err)
1597 goto out_unlock0;
1598 sma->sem_ctime = ktime_get_real_seconds();
1599 break;
1600 default:
1601 err = -EINVAL;
1602 goto out_unlock1;
1603 }
1604
1605out_unlock0:
1606 sem_unlock(sma, -1);
1607out_unlock1:
1608 rcu_read_unlock();
1609out_up:
1610 up_write(&sem_ids(ns).rwsem);
1611 return err;
1612}
1613
1614long ksys_semctl(int semid, int semnum, int cmd, unsigned long arg)
1615{
1616 int version;
1617 struct ipc_namespace *ns;
1618 void __user *p = (void __user *)arg;
1619 struct semid64_ds semid64;
1620 int err;
1621
1622 if (semid < 0)
1623 return -EINVAL;
1624
1625 version = ipc_parse_version(&cmd);
1626 ns = current->nsproxy->ipc_ns;
1627
1628 switch (cmd) {
1629 case IPC_INFO:
1630 case SEM_INFO:
1631 return semctl_info(ns, semid, cmd, p);
1632 case IPC_STAT:
1633 case SEM_STAT:
1634 case SEM_STAT_ANY:
1635 err = semctl_stat(ns, semid, cmd, &semid64);
1636 if (err < 0)
1637 return err;
1638 if (copy_semid_to_user(p, &semid64, version))
1639 err = -EFAULT;
1640 return err;
1641 case GETALL:
1642 case GETVAL:
1643 case GETPID:
1644 case GETNCNT:
1645 case GETZCNT:
1646 case SETALL:
1647 return semctl_main(ns, semid, semnum, cmd, p);
1648 case SETVAL: {
1649 int val;
1650#if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN)
1651 /* big-endian 64bit */
1652 val = arg >> 32;
1653#else
1654 /* 32bit or little-endian 64bit */
1655 val = arg;
1656#endif
1657 return semctl_setval(ns, semid, semnum, val);
1658 }
1659 case IPC_SET:
1660 if (copy_semid_from_user(&semid64, p, version))
1661 return -EFAULT;
1662 case IPC_RMID:
1663 return semctl_down(ns, semid, cmd, &semid64);
1664 default:
1665 return -EINVAL;
1666 }
1667}
1668
1669SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, unsigned long, arg)
1670{
1671 return ksys_semctl(semid, semnum, cmd, arg);
1672}
1673
1674#ifdef CONFIG_COMPAT
1675
1676struct compat_semid_ds {
1677 struct compat_ipc_perm sem_perm;
1678 compat_time_t sem_otime;
1679 compat_time_t sem_ctime;
1680 compat_uptr_t sem_base;
1681 compat_uptr_t sem_pending;
1682 compat_uptr_t sem_pending_last;
1683 compat_uptr_t undo;
1684 unsigned short sem_nsems;
1685};
1686
1687static int copy_compat_semid_from_user(struct semid64_ds *out, void __user *buf,
1688 int version)
1689{
1690 memset(out, 0, sizeof(*out));
1691 if (version == IPC_64) {
1692 struct compat_semid64_ds __user *p = buf;
1693 return get_compat_ipc64_perm(&out->sem_perm, &p->sem_perm);
1694 } else {
1695 struct compat_semid_ds __user *p = buf;
1696 return get_compat_ipc_perm(&out->sem_perm, &p->sem_perm);
1697 }
1698}
1699
1700static int copy_compat_semid_to_user(void __user *buf, struct semid64_ds *in,
1701 int version)
1702{
1703 if (version == IPC_64) {
1704 struct compat_semid64_ds v;
1705 memset(&v, 0, sizeof(v));
1706 to_compat_ipc64_perm(&v.sem_perm, &in->sem_perm);
1707 v.sem_otime = in->sem_otime;
1708 v.sem_ctime = in->sem_ctime;
1709 v.sem_nsems = in->sem_nsems;
1710 return copy_to_user(buf, &v, sizeof(v));
1711 } else {
1712 struct compat_semid_ds v;
1713 memset(&v, 0, sizeof(v));
1714 to_compat_ipc_perm(&v.sem_perm, &in->sem_perm);
1715 v.sem_otime = in->sem_otime;
1716 v.sem_ctime = in->sem_ctime;
1717 v.sem_nsems = in->sem_nsems;
1718 return copy_to_user(buf, &v, sizeof(v));
1719 }
1720}
1721
1722long compat_ksys_semctl(int semid, int semnum, int cmd, int arg)
1723{
1724 void __user *p = compat_ptr(arg);
1725 struct ipc_namespace *ns;
1726 struct semid64_ds semid64;
1727 int version = compat_ipc_parse_version(&cmd);
1728 int err;
1729
1730 ns = current->nsproxy->ipc_ns;
1731
1732 if (semid < 0)
1733 return -EINVAL;
1734
1735 switch (cmd & (~IPC_64)) {
1736 case IPC_INFO:
1737 case SEM_INFO:
1738 return semctl_info(ns, semid, cmd, p);
1739 case IPC_STAT:
1740 case SEM_STAT:
1741 case SEM_STAT_ANY:
1742 err = semctl_stat(ns, semid, cmd, &semid64);
1743 if (err < 0)
1744 return err;
1745 if (copy_compat_semid_to_user(p, &semid64, version))
1746 err = -EFAULT;
1747 return err;
1748 case GETVAL:
1749 case GETPID:
1750 case GETNCNT:
1751 case GETZCNT:
1752 case GETALL:
1753 case SETALL:
1754 return semctl_main(ns, semid, semnum, cmd, p);
1755 case SETVAL:
1756 return semctl_setval(ns, semid, semnum, arg);
1757 case IPC_SET:
1758 if (copy_compat_semid_from_user(&semid64, p, version))
1759 return -EFAULT;
1760 /* fallthru */
1761 case IPC_RMID:
1762 return semctl_down(ns, semid, cmd, &semid64);
1763 default:
1764 return -EINVAL;
1765 }
1766}
1767
1768COMPAT_SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, int, arg)
1769{
1770 return compat_ksys_semctl(semid, semnum, cmd, arg);
1771}
1772#endif
1773
1774/* If the task doesn't already have a undo_list, then allocate one
1775 * here. We guarantee there is only one thread using this undo list,
1776 * and current is THE ONE
1777 *
1778 * If this allocation and assignment succeeds, but later
1779 * portions of this code fail, there is no need to free the sem_undo_list.
1780 * Just let it stay associated with the task, and it'll be freed later
1781 * at exit time.
1782 *
1783 * This can block, so callers must hold no locks.
1784 */
1785static inline int get_undo_list(struct sem_undo_list **undo_listp)
1786{
1787 struct sem_undo_list *undo_list;
1788
1789 undo_list = current->sysvsem.undo_list;
1790 if (!undo_list) {
1791 undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
1792 if (undo_list == NULL)
1793 return -ENOMEM;
1794 spin_lock_init(&undo_list->lock);
1795 refcount_set(&undo_list->refcnt, 1);
1796 INIT_LIST_HEAD(&undo_list->list_proc);
1797
1798 current->sysvsem.undo_list = undo_list;
1799 }
1800 *undo_listp = undo_list;
1801 return 0;
1802}
1803
1804static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid)
1805{
1806 struct sem_undo *un;
1807
1808 list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) {
1809 if (un->semid == semid)
1810 return un;
1811 }
1812 return NULL;
1813}
1814
1815static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1816{
1817 struct sem_undo *un;
1818
1819 assert_spin_locked(&ulp->lock);
1820
1821 un = __lookup_undo(ulp, semid);
1822 if (un) {
1823 list_del_rcu(&un->list_proc);
1824 list_add_rcu(&un->list_proc, &ulp->list_proc);
1825 }
1826 return un;
1827}
1828
1829/**
1830 * find_alloc_undo - lookup (and if not present create) undo array
1831 * @ns: namespace
1832 * @semid: semaphore array id
1833 *
1834 * The function looks up (and if not present creates) the undo structure.
1835 * The size of the undo structure depends on the size of the semaphore
1836 * array, thus the alloc path is not that straightforward.
1837 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1838 * performs a rcu_read_lock().
1839 */
1840static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
1841{
1842 struct sem_array *sma;
1843 struct sem_undo_list *ulp;
1844 struct sem_undo *un, *new;
1845 int nsems, error;
1846
1847 error = get_undo_list(&ulp);
1848 if (error)
1849 return ERR_PTR(error);
1850
1851 rcu_read_lock();
1852 spin_lock(&ulp->lock);
1853 un = lookup_undo(ulp, semid);
1854 spin_unlock(&ulp->lock);
1855 if (likely(un != NULL))
1856 goto out;
1857
1858 /* no undo structure around - allocate one. */
1859 /* step 1: figure out the size of the semaphore array */
1860 sma = sem_obtain_object_check(ns, semid);
1861 if (IS_ERR(sma)) {
1862 rcu_read_unlock();
1863 return ERR_CAST(sma);
1864 }
1865
1866 nsems = sma->sem_nsems;
1867 if (!ipc_rcu_getref(&sma->sem_perm)) {
1868 rcu_read_unlock();
1869 un = ERR_PTR(-EIDRM);
1870 goto out;
1871 }
1872 rcu_read_unlock();
1873
1874 /* step 2: allocate new undo structure */
1875 new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1876 if (!new) {
1877 ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
1878 return ERR_PTR(-ENOMEM);
1879 }
1880
1881 /* step 3: Acquire the lock on semaphore array */
1882 rcu_read_lock();
1883 sem_lock_and_putref(sma);
1884 if (!ipc_valid_object(&sma->sem_perm)) {
1885 sem_unlock(sma, -1);
1886 rcu_read_unlock();
1887 kfree(new);
1888 un = ERR_PTR(-EIDRM);
1889 goto out;
1890 }
1891 spin_lock(&ulp->lock);
1892
1893 /*
1894 * step 4: check for races: did someone else allocate the undo struct?
1895 */
1896 un = lookup_undo(ulp, semid);
1897 if (un) {
1898 kfree(new);
1899 goto success;
1900 }
1901 /* step 5: initialize & link new undo structure */
1902 new->semadj = (short *) &new[1];
1903 new->ulp = ulp;
1904 new->semid = semid;
1905 assert_spin_locked(&ulp->lock);
1906 list_add_rcu(&new->list_proc, &ulp->list_proc);
1907 ipc_assert_locked_object(&sma->sem_perm);
1908 list_add(&new->list_id, &sma->list_id);
1909 un = new;
1910
1911success:
1912 spin_unlock(&ulp->lock);
1913 sem_unlock(sma, -1);
1914out:
1915 return un;
1916}
1917
1918static long do_semtimedop(int semid, struct sembuf __user *tsops,
1919 unsigned nsops, const struct timespec64 *timeout)
1920{
1921 int error = -EINVAL;
1922 struct sem_array *sma;
1923 struct sembuf fast_sops[SEMOPM_FAST];
1924 struct sembuf *sops = fast_sops, *sop;
1925 struct sem_undo *un;
1926 int max, locknum;
1927 bool undos = false, alter = false, dupsop = false;
1928 struct sem_queue queue;
1929 unsigned long dup = 0, jiffies_left = 0;
1930 struct ipc_namespace *ns;
1931
1932 ns = current->nsproxy->ipc_ns;
1933
1934 if (nsops < 1 || semid < 0)
1935 return -EINVAL;
1936 if (nsops > ns->sc_semopm)
1937 return -E2BIG;
1938 if (nsops > SEMOPM_FAST) {
1939 sops = kvmalloc(sizeof(*sops)*nsops, GFP_KERNEL);
1940 if (sops == NULL)
1941 return -ENOMEM;
1942 }
1943
1944 if (copy_from_user(sops, tsops, nsops * sizeof(*tsops))) {
1945 error = -EFAULT;
1946 goto out_free;
1947 }
1948
1949 if (timeout) {
1950 if (timeout->tv_sec < 0 || timeout->tv_nsec < 0 ||
1951 timeout->tv_nsec >= 1000000000L) {
1952 error = -EINVAL;
1953 goto out_free;
1954 }
1955 jiffies_left = timespec64_to_jiffies(timeout);
1956 }
1957
1958 max = 0;
1959 for (sop = sops; sop < sops + nsops; sop++) {
1960 unsigned long mask = 1ULL << ((sop->sem_num) % BITS_PER_LONG);
1961
1962 if (sop->sem_num >= max)
1963 max = sop->sem_num;
1964 if (sop->sem_flg & SEM_UNDO)
1965 undos = true;
1966 if (dup & mask) {
1967 /*
1968 * There was a previous alter access that appears
1969 * to have accessed the same semaphore, thus use
1970 * the dupsop logic. "appears", because the detection
1971 * can only check % BITS_PER_LONG.
1972 */
1973 dupsop = true;
1974 }
1975 if (sop->sem_op != 0) {
1976 alter = true;
1977 dup |= mask;
1978 }
1979 }
1980
1981 if (undos) {
1982 /* On success, find_alloc_undo takes the rcu_read_lock */
1983 un = find_alloc_undo(ns, semid);
1984 if (IS_ERR(un)) {
1985 error = PTR_ERR(un);
1986 goto out_free;
1987 }
1988 } else {
1989 un = NULL;
1990 rcu_read_lock();
1991 }
1992
1993 sma = sem_obtain_object_check(ns, semid);
1994 if (IS_ERR(sma)) {
1995 rcu_read_unlock();
1996 error = PTR_ERR(sma);
1997 goto out_free;
1998 }
1999
2000 error = -EFBIG;
2001 if (max >= sma->sem_nsems) {
2002 rcu_read_unlock();
2003 goto out_free;
2004 }
2005
2006 error = -EACCES;
2007 if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO)) {
2008 rcu_read_unlock();
2009 goto out_free;
2010 }
2011
2012 error = security_sem_semop(&sma->sem_perm, sops, nsops, alter);
2013 if (error) {
2014 rcu_read_unlock();
2015 goto out_free;
2016 }
2017
2018 error = -EIDRM;
2019 locknum = sem_lock(sma, sops, nsops);
2020 /*
2021 * We eventually might perform the following check in a lockless
2022 * fashion, considering ipc_valid_object() locking constraints.
2023 * If nsops == 1 and there is no contention for sem_perm.lock, then
2024 * only a per-semaphore lock is held and it's OK to proceed with the
2025 * check below. More details on the fine grained locking scheme
2026 * entangled here and why it's RMID race safe on comments at sem_lock()
2027 */
2028 if (!ipc_valid_object(&sma->sem_perm))
2029 goto out_unlock_free;
2030 /*
2031 * semid identifiers are not unique - find_alloc_undo may have
2032 * allocated an undo structure, it was invalidated by an RMID
2033 * and now a new array with received the same id. Check and fail.
2034 * This case can be detected checking un->semid. The existence of
2035 * "un" itself is guaranteed by rcu.
2036 */
2037 if (un && un->semid == -1)
2038 goto out_unlock_free;
2039
2040 queue.sops = sops;
2041 queue.nsops = nsops;
2042 queue.undo = un;
2043 queue.pid = task_tgid(current);
2044 queue.alter = alter;
2045 queue.dupsop = dupsop;
2046
2047 error = perform_atomic_semop(sma, &queue);
2048 if (error == 0) { /* non-blocking succesfull path */
2049 DEFINE_WAKE_Q(wake_q);
2050
2051 /*
2052 * If the operation was successful, then do
2053 * the required updates.
2054 */
2055 if (alter)
2056 do_smart_update(sma, sops, nsops, 1, &wake_q);
2057 else
2058 set_semotime(sma, sops);
2059
2060 sem_unlock(sma, locknum);
2061 rcu_read_unlock();
2062 wake_up_q(&wake_q);
2063
2064 goto out_free;
2065 }
2066 if (error < 0) /* non-blocking error path */
2067 goto out_unlock_free;
2068
2069 /*
2070 * We need to sleep on this operation, so we put the current
2071 * task into the pending queue and go to sleep.
2072 */
2073 if (nsops == 1) {
2074 struct sem *curr;
2075 curr = &sma->sems[sops->sem_num];
2076
2077 if (alter) {
2078 if (sma->complex_count) {
2079 list_add_tail(&queue.list,
2080 &sma->pending_alter);
2081 } else {
2082
2083 list_add_tail(&queue.list,
2084 &curr->pending_alter);
2085 }
2086 } else {
2087 list_add_tail(&queue.list, &curr->pending_const);
2088 }
2089 } else {
2090 if (!sma->complex_count)
2091 merge_queues(sma);
2092
2093 if (alter)
2094 list_add_tail(&queue.list, &sma->pending_alter);
2095 else
2096 list_add_tail(&queue.list, &sma->pending_const);
2097
2098 sma->complex_count++;
2099 }
2100
2101 do {
2102 queue.status = -EINTR;
2103 queue.sleeper = current;
2104
2105 __set_current_state(TASK_INTERRUPTIBLE);
2106 sem_unlock(sma, locknum);
2107 rcu_read_unlock();
2108
2109 if (timeout)
2110 jiffies_left = schedule_timeout(jiffies_left);
2111 else
2112 schedule();
2113
2114 /*
2115 * fastpath: the semop has completed, either successfully or
2116 * not, from the syscall pov, is quite irrelevant to us at this
2117 * point; we're done.
2118 *
2119 * We _do_ care, nonetheless, about being awoken by a signal or
2120 * spuriously. The queue.status is checked again in the
2121 * slowpath (aka after taking sem_lock), such that we can detect
2122 * scenarios where we were awakened externally, during the
2123 * window between wake_q_add() and wake_up_q().
2124 */
2125 error = READ_ONCE(queue.status);
2126 if (error != -EINTR) {
2127 /*
2128 * User space could assume that semop() is a memory
2129 * barrier: Without the mb(), the cpu could
2130 * speculatively read in userspace stale data that was
2131 * overwritten by the previous owner of the semaphore.
2132 */
2133 smp_mb();
2134 goto out_free;
2135 }
2136
2137 rcu_read_lock();
2138 locknum = sem_lock(sma, sops, nsops);
2139
2140 if (!ipc_valid_object(&sma->sem_perm))
2141 goto out_unlock_free;
2142
2143 error = READ_ONCE(queue.status);
2144
2145 /*
2146 * If queue.status != -EINTR we are woken up by another process.
2147 * Leave without unlink_queue(), but with sem_unlock().
2148 */
2149 if (error != -EINTR)
2150 goto out_unlock_free;
2151
2152 /*
2153 * If an interrupt occurred we have to clean up the queue.
2154 */
2155 if (timeout && jiffies_left == 0)
2156 error = -EAGAIN;
2157 } while (error == -EINTR && !signal_pending(current)); /* spurious */
2158
2159 unlink_queue(sma, &queue);
2160
2161out_unlock_free:
2162 sem_unlock(sma, locknum);
2163 rcu_read_unlock();
2164out_free:
2165 if (sops != fast_sops)
2166 kvfree(sops);
2167 return error;
2168}
2169
2170long ksys_semtimedop(int semid, struct sembuf __user *tsops,
2171 unsigned int nsops, const struct timespec __user *timeout)
2172{
2173 if (timeout) {
2174 struct timespec64 ts;
2175 if (get_timespec64(&ts, timeout))
2176 return -EFAULT;
2177 return do_semtimedop(semid, tsops, nsops, &ts);
2178 }
2179 return do_semtimedop(semid, tsops, nsops, NULL);
2180}
2181
2182SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
2183 unsigned int, nsops, const struct timespec __user *, timeout)
2184{
2185 return ksys_semtimedop(semid, tsops, nsops, timeout);
2186}
2187
2188#ifdef CONFIG_COMPAT
2189long compat_ksys_semtimedop(int semid, struct sembuf __user *tsems,
2190 unsigned int nsops,
2191 const struct compat_timespec __user *timeout)
2192{
2193 if (timeout) {
2194 struct timespec64 ts;
2195 if (compat_get_timespec64(&ts, timeout))
2196 return -EFAULT;
2197 return do_semtimedop(semid, tsems, nsops, &ts);
2198 }
2199 return do_semtimedop(semid, tsems, nsops, NULL);
2200}
2201
2202COMPAT_SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsems,
2203 unsigned int, nsops,
2204 const struct compat_timespec __user *, timeout)
2205{
2206 return compat_ksys_semtimedop(semid, tsems, nsops, timeout);
2207}
2208#endif
2209
2210SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
2211 unsigned, nsops)
2212{
2213 return do_semtimedop(semid, tsops, nsops, NULL);
2214}
2215
2216/* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
2217 * parent and child tasks.
2218 */
2219
2220int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
2221{
2222 struct sem_undo_list *undo_list;
2223 int error;
2224
2225 if (clone_flags & CLONE_SYSVSEM) {
2226 error = get_undo_list(&undo_list);
2227 if (error)
2228 return error;
2229 refcount_inc(&undo_list->refcnt);
2230 tsk->sysvsem.undo_list = undo_list;
2231 } else
2232 tsk->sysvsem.undo_list = NULL;
2233
2234 return 0;
2235}
2236
2237/*
2238 * add semadj values to semaphores, free undo structures.
2239 * undo structures are not freed when semaphore arrays are destroyed
2240 * so some of them may be out of date.
2241 * IMPLEMENTATION NOTE: There is some confusion over whether the
2242 * set of adjustments that needs to be done should be done in an atomic
2243 * manner or not. That is, if we are attempting to decrement the semval
2244 * should we queue up and wait until we can do so legally?
2245 * The original implementation attempted to do this (queue and wait).
2246 * The current implementation does not do so. The POSIX standard
2247 * and SVID should be consulted to determine what behavior is mandated.
2248 */
2249void exit_sem(struct task_struct *tsk)
2250{
2251 struct sem_undo_list *ulp;
2252
2253 ulp = tsk->sysvsem.undo_list;
2254 if (!ulp)
2255 return;
2256 tsk->sysvsem.undo_list = NULL;
2257
2258 if (!refcount_dec_and_test(&ulp->refcnt))
2259 return;
2260
2261 for (;;) {
2262 struct sem_array *sma;
2263 struct sem_undo *un;
2264 int semid, i;
2265 DEFINE_WAKE_Q(wake_q);
2266
2267 cond_resched();
2268
2269 rcu_read_lock();
2270 un = list_entry_rcu(ulp->list_proc.next,
2271 struct sem_undo, list_proc);
2272 if (&un->list_proc == &ulp->list_proc) {
2273 /*
2274 * We must wait for freeary() before freeing this ulp,
2275 * in case we raced with last sem_undo. There is a small
2276 * possibility where we exit while freeary() didn't
2277 * finish unlocking sem_undo_list.
2278 */
2279 spin_lock(&ulp->lock);
2280 spin_unlock(&ulp->lock);
2281 rcu_read_unlock();
2282 break;
2283 }
2284 spin_lock(&ulp->lock);
2285 semid = un->semid;
2286 spin_unlock(&ulp->lock);
2287
2288 /* exit_sem raced with IPC_RMID, nothing to do */
2289 if (semid == -1) {
2290 rcu_read_unlock();
2291 continue;
2292 }
2293
2294 sma = sem_obtain_object_check(tsk->nsproxy->ipc_ns, semid);
2295 /* exit_sem raced with IPC_RMID, nothing to do */
2296 if (IS_ERR(sma)) {
2297 rcu_read_unlock();
2298 continue;
2299 }
2300
2301 sem_lock(sma, NULL, -1);
2302 /* exit_sem raced with IPC_RMID, nothing to do */
2303 if (!ipc_valid_object(&sma->sem_perm)) {
2304 sem_unlock(sma, -1);
2305 rcu_read_unlock();
2306 continue;
2307 }
2308 un = __lookup_undo(ulp, semid);
2309 if (un == NULL) {
2310 /* exit_sem raced with IPC_RMID+semget() that created
2311 * exactly the same semid. Nothing to do.
2312 */
2313 sem_unlock(sma, -1);
2314 rcu_read_unlock();
2315 continue;
2316 }
2317
2318 /* remove un from the linked lists */
2319 ipc_assert_locked_object(&sma->sem_perm);
2320 list_del(&un->list_id);
2321
2322 /* we are the last process using this ulp, acquiring ulp->lock
2323 * isn't required. Besides that, we are also protected against
2324 * IPC_RMID as we hold sma->sem_perm lock now
2325 */
2326 list_del_rcu(&un->list_proc);
2327
2328 /* perform adjustments registered in un */
2329 for (i = 0; i < sma->sem_nsems; i++) {
2330 struct sem *semaphore = &sma->sems[i];
2331 if (un->semadj[i]) {
2332 semaphore->semval += un->semadj[i];
2333 /*
2334 * Range checks of the new semaphore value,
2335 * not defined by sus:
2336 * - Some unices ignore the undo entirely
2337 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
2338 * - some cap the value (e.g. FreeBSD caps
2339 * at 0, but doesn't enforce SEMVMX)
2340 *
2341 * Linux caps the semaphore value, both at 0
2342 * and at SEMVMX.
2343 *
2344 * Manfred <manfred@colorfullife.com>
2345 */
2346 if (semaphore->semval < 0)
2347 semaphore->semval = 0;
2348 if (semaphore->semval > SEMVMX)
2349 semaphore->semval = SEMVMX;
2350 ipc_update_pid(&semaphore->sempid, task_tgid(current));
2351 }
2352 }
2353 /* maybe some queued-up processes were waiting for this */
2354 do_smart_update(sma, NULL, 0, 1, &wake_q);
2355 sem_unlock(sma, -1);
2356 rcu_read_unlock();
2357 wake_up_q(&wake_q);
2358
2359 kfree_rcu(un, rcu);
2360 }
2361 kfree(ulp);
2362}
2363
2364#ifdef CONFIG_PROC_FS
2365static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
2366{
2367 struct user_namespace *user_ns = seq_user_ns(s);
2368 struct kern_ipc_perm *ipcp = it;
2369 struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
2370 time64_t sem_otime;
2371
2372 /*
2373 * The proc interface isn't aware of sem_lock(), it calls
2374 * ipc_lock_object() directly (in sysvipc_find_ipc).
2375 * In order to stay compatible with sem_lock(), we must
2376 * enter / leave complex_mode.
2377 */
2378 complexmode_enter(sma);
2379
2380 sem_otime = get_semotime(sma);
2381
2382 seq_printf(s,
2383 "%10d %10d %4o %10u %5u %5u %5u %5u %10llu %10llu\n",
2384 sma->sem_perm.key,
2385 sma->sem_perm.id,
2386 sma->sem_perm.mode,
2387 sma->sem_nsems,
2388 from_kuid_munged(user_ns, sma->sem_perm.uid),
2389 from_kgid_munged(user_ns, sma->sem_perm.gid),
2390 from_kuid_munged(user_ns, sma->sem_perm.cuid),
2391 from_kgid_munged(user_ns, sma->sem_perm.cgid),
2392 sem_otime,
2393 sma->sem_ctime);
2394
2395 complexmode_tryleave(sma);
2396
2397 return 0;
2398}
2399#endif