1 // SPDX-License-Identifier: GPL-2.0
4 * Copyright (C) 1992 Krishna Balasubramanian
5 * Copyright (C) 1995 Eric Schenk, Bruno Haible
7 * /proc/sysvipc/sem support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
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
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>
19 * support for audit of ipc object properties and permission changes
20 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
24 * Pavel Emelianov <xemul@openvz.org>
26 * Implementation notes: (May 2010)
27 * This file implements System V semaphores.
29 * User space visible behavior:
30 * - FIFO ordering for semop() operations (just FIFO, not starvation
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
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.
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.
73 #include <linux/compat.h>
74 #include <linux/slab.h>
75 #include <linux/spinlock.h>
76 #include <linux/init.h>
77 #include <linux/proc_fs.h>
78 #include <linux/time.h>
79 #include <linux/security.h>
80 #include <linux/syscalls.h>
81 #include <linux/audit.h>
82 #include <linux/capability.h>
83 #include <linux/seq_file.h>
84 #include <linux/rwsem.h>
85 #include <linux/nsproxy.h>
86 #include <linux/ipc_namespace.h>
87 #include <linux/sched/wake_q.h>
88 #include <linux/nospec.h>
89 #include <linux/rhashtable.h>
91 #include <linux/uaccess.h>
94 /* One semaphore structure for each semaphore in the system. */
96 int semval; /* current value */
98 * PID of the process that last modified the semaphore. For
99 * Linux, specifically these are:
101 * - semctl, via SETVAL and SETALL.
102 * - at task exit when performing undo adjustments (see exit_sem).
105 spinlock_t lock; /* spinlock for fine-grained semtimedop */
106 struct list_head pending_alter; /* pending single-sop operations */
107 /* that alter the semaphore */
108 struct list_head pending_const; /* pending single-sop operations */
109 /* that do not alter the semaphore*/
110 time64_t sem_otime; /* candidate for sem_otime */
111 } ____cacheline_aligned_in_smp;
113 /* One sem_array data structure for each set of semaphores in the system. */
115 struct kern_ipc_perm sem_perm; /* permissions .. see ipc.h */
116 time64_t sem_ctime; /* create/last semctl() time */
117 struct list_head pending_alter; /* pending operations */
118 /* that alter the array */
119 struct list_head pending_const; /* pending complex operations */
120 /* that do not alter semvals */
121 struct list_head list_id; /* undo requests on this array */
122 int sem_nsems; /* no. of semaphores in array */
123 int complex_count; /* pending complex operations */
124 unsigned int use_global_lock;/* >0: global lock required */
127 } __randomize_layout;
129 /* One queue for each sleeping process in the system. */
131 struct list_head list; /* queue of pending operations */
132 struct task_struct *sleeper; /* this process */
133 struct sem_undo *undo; /* undo structure */
134 struct pid *pid; /* process id of requesting process */
135 int status; /* completion status of operation */
136 struct sembuf *sops; /* array of pending operations */
137 struct sembuf *blocking; /* the operation that blocked */
138 int nsops; /* number of operations */
139 bool alter; /* does *sops alter the array? */
140 bool dupsop; /* sops on more than one sem_num */
143 /* Each task has a list of undo requests. They are executed automatically
144 * when the process exits.
147 struct list_head list_proc; /* per-process list: *
148 * all undos from one process
150 struct rcu_head rcu; /* rcu struct for sem_undo */
151 struct sem_undo_list *ulp; /* back ptr to sem_undo_list */
152 struct list_head list_id; /* per semaphore array list:
153 * all undos for one array */
154 int semid; /* semaphore set identifier */
155 short *semadj; /* array of adjustments */
156 /* one per semaphore */
159 /* sem_undo_list controls shared access to the list of sem_undo structures
160 * that may be shared among all a CLONE_SYSVSEM task group.
162 struct sem_undo_list {
165 struct list_head list_proc;
169 #define sem_ids(ns) ((ns)->ids[IPC_SEM_IDS])
171 static int newary(struct ipc_namespace *, struct ipc_params *);
172 static void freeary(struct ipc_namespace *, struct kern_ipc_perm *);
173 #ifdef CONFIG_PROC_FS
174 static int sysvipc_sem_proc_show(struct seq_file *s, void *it);
177 #define SEMMSL_FAST 256 /* 512 bytes on stack */
178 #define SEMOPM_FAST 64 /* ~ 372 bytes on stack */
181 * Switching from the mode suitable for simple ops
182 * to the mode for complex ops is costly. Therefore:
183 * use some hysteresis
185 #define USE_GLOBAL_LOCK_HYSTERESIS 10
189 * a) global sem_lock() for read/write
191 * sem_array.complex_count,
192 * sem_array.pending{_alter,_const},
195 * b) global or semaphore sem_lock() for read/write:
196 * sem_array.sems[i].pending_{const,alter}:
199 * sem_undo_list.list_proc:
200 * * undo_list->lock for write
203 * * global sem_lock() for write
204 * * either local or global sem_lock() for read.
207 * Most ordering is enforced by using spin_lock() and spin_unlock().
208 * The special case is use_global_lock:
209 * Setting it from non-zero to 0 is a RELEASE, this is ensured by
210 * using smp_store_release().
211 * Testing if it is non-zero is an ACQUIRE, this is ensured by using
212 * smp_load_acquire().
213 * Setting it from 0 to non-zero must be ordered with regards to
214 * this smp_load_acquire(), this is guaranteed because the smp_load_acquire()
215 * is inside a spin_lock() and after a write from 0 to non-zero a
216 * spin_lock()+spin_unlock() is done.
219 #define sc_semmsl sem_ctls[0]
220 #define sc_semmns sem_ctls[1]
221 #define sc_semopm sem_ctls[2]
222 #define sc_semmni sem_ctls[3]
224 void sem_init_ns(struct ipc_namespace *ns)
226 ns->sc_semmsl = SEMMSL;
227 ns->sc_semmns = SEMMNS;
228 ns->sc_semopm = SEMOPM;
229 ns->sc_semmni = SEMMNI;
231 ipc_init_ids(&ns->ids[IPC_SEM_IDS]);
235 void sem_exit_ns(struct ipc_namespace *ns)
237 free_ipcs(ns, &sem_ids(ns), freeary);
238 idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr);
239 rhashtable_destroy(&ns->ids[IPC_SEM_IDS].key_ht);
243 void __init sem_init(void)
245 sem_init_ns(&init_ipc_ns);
246 ipc_init_proc_interface("sysvipc/sem",
247 " key semid perms nsems uid gid cuid cgid otime ctime\n",
248 IPC_SEM_IDS, sysvipc_sem_proc_show);
252 * unmerge_queues - unmerge queues, if possible.
253 * @sma: semaphore array
255 * The function unmerges the wait queues if complex_count is 0.
256 * It must be called prior to dropping the global semaphore array lock.
258 static void unmerge_queues(struct sem_array *sma)
260 struct sem_queue *q, *tq;
262 /* complex operations still around? */
263 if (sma->complex_count)
266 * We will switch back to simple mode.
267 * Move all pending operation back into the per-semaphore
270 list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
272 curr = &sma->sems[q->sops[0].sem_num];
274 list_add_tail(&q->list, &curr->pending_alter);
276 INIT_LIST_HEAD(&sma->pending_alter);
280 * merge_queues - merge single semop queues into global queue
281 * @sma: semaphore array
283 * This function merges all per-semaphore queues into the global queue.
284 * It is necessary to achieve FIFO ordering for the pending single-sop
285 * operations when a multi-semop operation must sleep.
286 * Only the alter operations must be moved, the const operations can stay.
288 static void merge_queues(struct sem_array *sma)
291 for (i = 0; i < sma->sem_nsems; i++) {
292 struct sem *sem = &sma->sems[i];
294 list_splice_init(&sem->pending_alter, &sma->pending_alter);
298 static void sem_rcu_free(struct rcu_head *head)
300 struct kern_ipc_perm *p = container_of(head, struct kern_ipc_perm, rcu);
301 struct sem_array *sma = container_of(p, struct sem_array, sem_perm);
303 security_sem_free(&sma->sem_perm);
308 * Enter the mode suitable for non-simple operations:
309 * Caller must own sem_perm.lock.
311 static void complexmode_enter(struct sem_array *sma)
316 if (sma->use_global_lock > 0) {
318 * We are already in global lock mode.
319 * Nothing to do, just reset the
320 * counter until we return to simple mode.
322 sma->use_global_lock = USE_GLOBAL_LOCK_HYSTERESIS;
325 sma->use_global_lock = USE_GLOBAL_LOCK_HYSTERESIS;
327 for (i = 0; i < sma->sem_nsems; i++) {
329 spin_lock(&sem->lock);
330 spin_unlock(&sem->lock);
335 * Try to leave the mode that disallows simple operations:
336 * Caller must own sem_perm.lock.
338 static void complexmode_tryleave(struct sem_array *sma)
340 if (sma->complex_count) {
341 /* Complex ops are sleeping.
342 * We must stay in complex mode
346 if (sma->use_global_lock == 1) {
348 * Immediately after setting use_global_lock to 0,
349 * a simple op can start. Thus: all memory writes
350 * performed by the current operation must be visible
351 * before we set use_global_lock to 0.
353 smp_store_release(&sma->use_global_lock, 0);
355 sma->use_global_lock--;
359 #define SEM_GLOBAL_LOCK (-1)
361 * If the request contains only one semaphore operation, and there are
362 * no complex transactions pending, lock only the semaphore involved.
363 * Otherwise, lock the entire semaphore array, since we either have
364 * multiple semaphores in our own semops, or we need to look at
365 * semaphores from other pending complex operations.
367 static inline int sem_lock(struct sem_array *sma, struct sembuf *sops,
374 /* Complex operation - acquire a full lock */
375 ipc_lock_object(&sma->sem_perm);
377 /* Prevent parallel simple ops */
378 complexmode_enter(sma);
379 return SEM_GLOBAL_LOCK;
383 * Only one semaphore affected - try to optimize locking.
384 * Optimized locking is possible if no complex operation
385 * is either enqueued or processed right now.
387 * Both facts are tracked by use_global_mode.
389 idx = array_index_nospec(sops->sem_num, sma->sem_nsems);
390 sem = &sma->sems[idx];
393 * Initial check for use_global_lock. Just an optimization,
394 * no locking, no memory barrier.
396 if (!sma->use_global_lock) {
398 * It appears that no complex operation is around.
399 * Acquire the per-semaphore lock.
401 spin_lock(&sem->lock);
403 /* pairs with smp_store_release() */
404 if (!smp_load_acquire(&sma->use_global_lock)) {
405 /* fast path successful! */
406 return sops->sem_num;
408 spin_unlock(&sem->lock);
411 /* slow path: acquire the full lock */
412 ipc_lock_object(&sma->sem_perm);
414 if (sma->use_global_lock == 0) {
416 * The use_global_lock mode ended while we waited for
417 * sma->sem_perm.lock. Thus we must switch to locking
419 * Unlike in the fast path, there is no need to recheck
420 * sma->use_global_lock after we have acquired sem->lock:
421 * We own sma->sem_perm.lock, thus use_global_lock cannot
424 spin_lock(&sem->lock);
426 ipc_unlock_object(&sma->sem_perm);
427 return sops->sem_num;
430 * Not a false alarm, thus continue to use the global lock
431 * mode. No need for complexmode_enter(), this was done by
432 * the caller that has set use_global_mode to non-zero.
434 return SEM_GLOBAL_LOCK;
438 static inline void sem_unlock(struct sem_array *sma, int locknum)
440 if (locknum == SEM_GLOBAL_LOCK) {
442 complexmode_tryleave(sma);
443 ipc_unlock_object(&sma->sem_perm);
445 struct sem *sem = &sma->sems[locknum];
446 spin_unlock(&sem->lock);
451 * sem_lock_(check_) routines are called in the paths where the rwsem
454 * The caller holds the RCU read lock.
456 static inline struct sem_array *sem_obtain_object(struct ipc_namespace *ns, int id)
458 struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&sem_ids(ns), id);
461 return ERR_CAST(ipcp);
463 return container_of(ipcp, struct sem_array, sem_perm);
466 static inline struct sem_array *sem_obtain_object_check(struct ipc_namespace *ns,
469 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&sem_ids(ns), id);
472 return ERR_CAST(ipcp);
474 return container_of(ipcp, struct sem_array, sem_perm);
477 static inline void sem_lock_and_putref(struct sem_array *sma)
479 sem_lock(sma, NULL, -1);
480 ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
483 static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
485 ipc_rmid(&sem_ids(ns), &s->sem_perm);
488 static struct sem_array *sem_alloc(size_t nsems)
490 struct sem_array *sma;
492 if (nsems > (INT_MAX - sizeof(*sma)) / sizeof(sma->sems[0]))
495 sma = kvzalloc(struct_size(sma, sems, nsems), GFP_KERNEL);
503 * newary - Create a new semaphore set
505 * @params: ptr to the structure that contains key, semflg and nsems
507 * Called with sem_ids.rwsem held (as a writer)
509 static int newary(struct ipc_namespace *ns, struct ipc_params *params)
512 struct sem_array *sma;
513 key_t key = params->key;
514 int nsems = params->u.nsems;
515 int semflg = params->flg;
520 if (ns->used_sems + nsems > ns->sc_semmns)
523 sma = sem_alloc(nsems);
527 sma->sem_perm.mode = (semflg & S_IRWXUGO);
528 sma->sem_perm.key = key;
530 sma->sem_perm.security = NULL;
531 retval = security_sem_alloc(&sma->sem_perm);
537 for (i = 0; i < nsems; i++) {
538 INIT_LIST_HEAD(&sma->sems[i].pending_alter);
539 INIT_LIST_HEAD(&sma->sems[i].pending_const);
540 spin_lock_init(&sma->sems[i].lock);
543 sma->complex_count = 0;
544 sma->use_global_lock = USE_GLOBAL_LOCK_HYSTERESIS;
545 INIT_LIST_HEAD(&sma->pending_alter);
546 INIT_LIST_HEAD(&sma->pending_const);
547 INIT_LIST_HEAD(&sma->list_id);
548 sma->sem_nsems = nsems;
549 sma->sem_ctime = ktime_get_real_seconds();
551 /* ipc_addid() locks sma upon success. */
552 retval = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
554 ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
557 ns->used_sems += nsems;
562 return sma->sem_perm.id;
567 * Called with sem_ids.rwsem and ipcp locked.
569 static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
570 struct ipc_params *params)
572 struct sem_array *sma;
574 sma = container_of(ipcp, struct sem_array, sem_perm);
575 if (params->u.nsems > sma->sem_nsems)
581 long ksys_semget(key_t key, int nsems, int semflg)
583 struct ipc_namespace *ns;
584 static const struct ipc_ops sem_ops = {
586 .associate = security_sem_associate,
587 .more_checks = sem_more_checks,
589 struct ipc_params sem_params;
591 ns = current->nsproxy->ipc_ns;
593 if (nsems < 0 || nsems > ns->sc_semmsl)
596 sem_params.key = key;
597 sem_params.flg = semflg;
598 sem_params.u.nsems = nsems;
600 return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
603 SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg)
605 return ksys_semget(key, nsems, semflg);
609 * perform_atomic_semop[_slow] - Attempt to perform semaphore
610 * operations on a given array.
611 * @sma: semaphore array
612 * @q: struct sem_queue that describes the operation
614 * Caller blocking are as follows, based the value
615 * indicated by the semaphore operation (sem_op):
617 * (1) >0 never blocks.
618 * (2) 0 (wait-for-zero operation): semval is non-zero.
619 * (3) <0 attempting to decrement semval to a value smaller than zero.
621 * Returns 0 if the operation was possible.
622 * Returns 1 if the operation is impossible, the caller must sleep.
623 * Returns <0 for error codes.
625 static int perform_atomic_semop_slow(struct sem_array *sma, struct sem_queue *q)
627 int result, sem_op, nsops;
638 for (sop = sops; sop < sops + nsops; sop++) {
639 int idx = array_index_nospec(sop->sem_num, sma->sem_nsems);
640 curr = &sma->sems[idx];
641 sem_op = sop->sem_op;
642 result = curr->semval;
644 if (!sem_op && result)
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)
658 un->semadj[sop->sem_num] = undo;
661 curr->semval = result;
666 while (sop >= sops) {
667 ipc_update_pid(&sma->sems[sop->sem_num].sempid, pid);
680 if (sop->sem_flg & IPC_NOWAIT)
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;
698 static int perform_atomic_semop(struct sem_array *sma, struct sem_queue *q)
700 int result, sem_op, nsops;
710 if (unlikely(q->dupsop))
711 return perform_atomic_semop_slow(sma, q);
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.
719 for (sop = sops; sop < sops + nsops; sop++) {
720 int idx = array_index_nospec(sop->sem_num, sma->sem_nsems);
722 curr = &sma->sems[idx];
723 sem_op = sop->sem_op;
724 result = curr->semval;
726 if (!sem_op && result)
727 goto would_block; /* wait-for-zero */
736 if (sop->sem_flg & SEM_UNDO) {
737 int undo = un->semadj[sop->sem_num] - sem_op;
739 /* Exceeding the undo range is an error. */
740 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
745 for (sop = sops; sop < sops + nsops; sop++) {
746 curr = &sma->sems[sop->sem_num];
747 sem_op = sop->sem_op;
748 result = curr->semval;
750 if (sop->sem_flg & SEM_UNDO) {
751 int undo = un->semadj[sop->sem_num] - sem_op;
753 un->semadj[sop->sem_num] = undo;
755 curr->semval += sem_op;
756 ipc_update_pid(&curr->sempid, q->pid);
763 return sop->sem_flg & IPC_NOWAIT ? -EAGAIN : 1;
766 static inline void wake_up_sem_queue_prepare(struct sem_queue *q, int error,
767 struct wake_q_head *wake_q)
769 wake_q_add(wake_q, q->sleeper);
771 * Rely on the above implicit barrier, such that we can
772 * ensure that we hold reference to the task before setting
773 * q->status. Otherwise we could race with do_exit if the
774 * task is awoken by an external event before calling
777 WRITE_ONCE(q->status, error);
780 static void unlink_queue(struct sem_array *sma, struct sem_queue *q)
784 sma->complex_count--;
787 /** check_restart(sma, q)
788 * @sma: semaphore array
789 * @q: the operation that just completed
791 * update_queue is O(N^2) when it restarts scanning the whole queue of
792 * waiting operations. Therefore this function checks if the restart is
793 * really necessary. It is called after a previously waiting operation
794 * modified the array.
795 * Note that wait-for-zero operations are handled without restart.
797 static inline int check_restart(struct sem_array *sma, struct sem_queue *q)
799 /* pending complex alter operations are too difficult to analyse */
800 if (!list_empty(&sma->pending_alter))
803 /* we were a sleeping complex operation. Too difficult */
807 /* It is impossible that someone waits for the new value:
808 * - complex operations always restart.
809 * - wait-for-zero are handled seperately.
810 * - q is a previously sleeping simple operation that
811 * altered the array. It must be a decrement, because
812 * simple increments never sleep.
813 * - If there are older (higher priority) decrements
814 * in the queue, then they have observed the original
815 * semval value and couldn't proceed. The operation
816 * decremented to value - thus they won't proceed either.
822 * wake_const_ops - wake up non-alter tasks
823 * @sma: semaphore array.
824 * @semnum: semaphore that was modified.
825 * @wake_q: lockless wake-queue head.
827 * wake_const_ops must be called after a semaphore in a semaphore array
828 * was set to 0. If complex const operations are pending, wake_const_ops must
829 * be called with semnum = -1, as well as with the number of each modified
831 * The tasks that must be woken up are added to @wake_q. The return code
832 * is stored in q->pid.
833 * The function returns 1 if at least one operation was completed successfully.
835 static int wake_const_ops(struct sem_array *sma, int semnum,
836 struct wake_q_head *wake_q)
838 struct sem_queue *q, *tmp;
839 struct list_head *pending_list;
840 int semop_completed = 0;
843 pending_list = &sma->pending_const;
845 pending_list = &sma->sems[semnum].pending_const;
847 list_for_each_entry_safe(q, tmp, pending_list, list) {
848 int error = perform_atomic_semop(sma, q);
852 /* operation completed, remove from queue & wakeup */
853 unlink_queue(sma, q);
855 wake_up_sem_queue_prepare(q, error, wake_q);
860 return semop_completed;
864 * do_smart_wakeup_zero - wakeup all wait for zero tasks
865 * @sma: semaphore array
866 * @sops: operations that were performed
867 * @nsops: number of operations
868 * @wake_q: lockless wake-queue head
870 * Checks all required queue for wait-for-zero operations, based
871 * on the actual changes that were performed on the semaphore array.
872 * The function returns 1 if at least one operation was completed successfully.
874 static int do_smart_wakeup_zero(struct sem_array *sma, struct sembuf *sops,
875 int nsops, struct wake_q_head *wake_q)
878 int semop_completed = 0;
881 /* first: the per-semaphore queues, if known */
883 for (i = 0; i < nsops; i++) {
884 int num = sops[i].sem_num;
886 if (sma->sems[num].semval == 0) {
888 semop_completed |= wake_const_ops(sma, num, wake_q);
893 * No sops means modified semaphores not known.
894 * Assume all were changed.
896 for (i = 0; i < sma->sem_nsems; i++) {
897 if (sma->sems[i].semval == 0) {
899 semop_completed |= wake_const_ops(sma, i, wake_q);
904 * If one of the modified semaphores got 0,
905 * then check the global queue, too.
908 semop_completed |= wake_const_ops(sma, -1, wake_q);
910 return semop_completed;
915 * update_queue - look for tasks that can be completed.
916 * @sma: semaphore array.
917 * @semnum: semaphore that was modified.
918 * @wake_q: lockless wake-queue head.
920 * update_queue must be called after a semaphore in a semaphore array
921 * was modified. If multiple semaphores were modified, update_queue must
922 * be called with semnum = -1, as well as with the number of each modified
924 * The tasks that must be woken up are added to @wake_q. The return code
925 * is stored in q->pid.
926 * The function internally checks if const operations can now succeed.
928 * The function return 1 if at least one semop was completed successfully.
930 static int update_queue(struct sem_array *sma, int semnum, struct wake_q_head *wake_q)
932 struct sem_queue *q, *tmp;
933 struct list_head *pending_list;
934 int semop_completed = 0;
937 pending_list = &sma->pending_alter;
939 pending_list = &sma->sems[semnum].pending_alter;
942 list_for_each_entry_safe(q, tmp, pending_list, list) {
945 /* If we are scanning the single sop, per-semaphore list of
946 * one semaphore and that semaphore is 0, then it is not
947 * necessary to scan further: simple increments
948 * that affect only one entry succeed immediately and cannot
949 * be in the per semaphore pending queue, and decrements
950 * cannot be successful if the value is already 0.
952 if (semnum != -1 && sma->sems[semnum].semval == 0)
955 error = perform_atomic_semop(sma, q);
957 /* Does q->sleeper still need to sleep? */
961 unlink_queue(sma, q);
967 do_smart_wakeup_zero(sma, q->sops, q->nsops, wake_q);
968 restart = check_restart(sma, q);
971 wake_up_sem_queue_prepare(q, error, wake_q);
975 return semop_completed;
979 * set_semotime - set sem_otime
980 * @sma: semaphore array
981 * @sops: operations that modified the array, may be NULL
983 * sem_otime is replicated to avoid cache line trashing.
984 * This function sets one instance to the current time.
986 static void set_semotime(struct sem_array *sma, struct sembuf *sops)
989 sma->sems[0].sem_otime = ktime_get_real_seconds();
991 sma->sems[sops[0].sem_num].sem_otime =
992 ktime_get_real_seconds();
997 * do_smart_update - optimized update_queue
998 * @sma: semaphore array
999 * @sops: operations that were performed
1000 * @nsops: number of operations
1001 * @otime: force setting otime
1002 * @wake_q: lockless wake-queue head
1004 * do_smart_update() does the required calls to update_queue and wakeup_zero,
1005 * based on the actual changes that were performed on the semaphore array.
1006 * Note that the function does not do the actual wake-up: the caller is
1007 * responsible for calling wake_up_q().
1008 * It is safe to perform this call after dropping all locks.
1010 static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops,
1011 int otime, struct wake_q_head *wake_q)
1015 otime |= do_smart_wakeup_zero(sma, sops, nsops, wake_q);
1017 if (!list_empty(&sma->pending_alter)) {
1018 /* semaphore array uses the global queue - just process it. */
1019 otime |= update_queue(sma, -1, wake_q);
1023 * No sops, thus the modified semaphores are not
1026 for (i = 0; i < sma->sem_nsems; i++)
1027 otime |= update_queue(sma, i, wake_q);
1030 * Check the semaphores that were increased:
1031 * - No complex ops, thus all sleeping ops are
1033 * - if we decreased the value, then any sleeping
1034 * semaphore ops wont be able to run: If the
1035 * previous value was too small, then the new
1036 * value will be too small, too.
1038 for (i = 0; i < nsops; i++) {
1039 if (sops[i].sem_op > 0) {
1040 otime |= update_queue(sma,
1041 sops[i].sem_num, wake_q);
1047 set_semotime(sma, sops);
1051 * check_qop: Test if a queued operation sleeps on the semaphore semnum
1053 static int check_qop(struct sem_array *sma, int semnum, struct sem_queue *q,
1056 struct sembuf *sop = q->blocking;
1059 * Linux always (since 0.99.10) reported a task as sleeping on all
1060 * semaphores. This violates SUS, therefore it was changed to the
1061 * standard compliant behavior.
1062 * Give the administrators a chance to notice that an application
1063 * might misbehave because it relies on the Linux behavior.
1065 pr_info_once("semctl(GETNCNT/GETZCNT) is since 3.16 Single Unix Specification compliant.\n"
1066 "The task %s (%d) triggered the difference, watch for misbehavior.\n",
1067 current->comm, task_pid_nr(current));
1069 if (sop->sem_num != semnum)
1072 if (count_zero && sop->sem_op == 0)
1074 if (!count_zero && sop->sem_op < 0)
1080 /* The following counts are associated to each semaphore:
1081 * semncnt number of tasks waiting on semval being nonzero
1082 * semzcnt number of tasks waiting on semval being zero
1084 * Per definition, a task waits only on the semaphore of the first semop
1085 * that cannot proceed, even if additional operation would block, too.
1087 static int count_semcnt(struct sem_array *sma, ushort semnum,
1090 struct list_head *l;
1091 struct sem_queue *q;
1095 /* First: check the simple operations. They are easy to evaluate */
1097 l = &sma->sems[semnum].pending_const;
1099 l = &sma->sems[semnum].pending_alter;
1101 list_for_each_entry(q, l, list) {
1102 /* all task on a per-semaphore list sleep on exactly
1108 /* Then: check the complex operations. */
1109 list_for_each_entry(q, &sma->pending_alter, list) {
1110 semcnt += check_qop(sma, semnum, q, count_zero);
1113 list_for_each_entry(q, &sma->pending_const, list) {
1114 semcnt += check_qop(sma, semnum, q, count_zero);
1120 /* Free a semaphore set. freeary() is called with sem_ids.rwsem locked
1121 * as a writer and the spinlock for this semaphore set hold. sem_ids.rwsem
1122 * remains locked on exit.
1124 static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
1126 struct sem_undo *un, *tu;
1127 struct sem_queue *q, *tq;
1128 struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
1130 DEFINE_WAKE_Q(wake_q);
1132 /* Free the existing undo structures for this semaphore set. */
1133 ipc_assert_locked_object(&sma->sem_perm);
1134 list_for_each_entry_safe(un, tu, &sma->list_id, list_id) {
1135 list_del(&un->list_id);
1136 spin_lock(&un->ulp->lock);
1138 list_del_rcu(&un->list_proc);
1139 spin_unlock(&un->ulp->lock);
1143 /* Wake up all pending processes and let them fail with EIDRM. */
1144 list_for_each_entry_safe(q, tq, &sma->pending_const, list) {
1145 unlink_queue(sma, q);
1146 wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
1149 list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
1150 unlink_queue(sma, q);
1151 wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
1153 for (i = 0; i < sma->sem_nsems; i++) {
1154 struct sem *sem = &sma->sems[i];
1155 list_for_each_entry_safe(q, tq, &sem->pending_const, list) {
1156 unlink_queue(sma, q);
1157 wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
1159 list_for_each_entry_safe(q, tq, &sem->pending_alter, list) {
1160 unlink_queue(sma, q);
1161 wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
1163 ipc_update_pid(&sem->sempid, NULL);
1166 /* Remove the semaphore set from the IDR */
1168 sem_unlock(sma, -1);
1172 ns->used_sems -= sma->sem_nsems;
1173 ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
1176 static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
1180 return copy_to_user(buf, in, sizeof(*in));
1183 struct semid_ds out;
1185 memset(&out, 0, sizeof(out));
1187 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
1189 out.sem_otime = in->sem_otime;
1190 out.sem_ctime = in->sem_ctime;
1191 out.sem_nsems = in->sem_nsems;
1193 return copy_to_user(buf, &out, sizeof(out));
1200 static time64_t get_semotime(struct sem_array *sma)
1205 res = sma->sems[0].sem_otime;
1206 for (i = 1; i < sma->sem_nsems; i++) {
1207 time64_t to = sma->sems[i].sem_otime;
1215 static int semctl_stat(struct ipc_namespace *ns, int semid,
1216 int cmd, struct semid64_ds *semid64)
1218 struct sem_array *sma;
1222 memset(semid64, 0, sizeof(*semid64));
1225 if (cmd == SEM_STAT || cmd == SEM_STAT_ANY) {
1226 sma = sem_obtain_object(ns, semid);
1231 } else { /* IPC_STAT */
1232 sma = sem_obtain_object_check(ns, semid);
1239 /* see comment for SHM_STAT_ANY */
1240 if (cmd == SEM_STAT_ANY)
1241 audit_ipc_obj(&sma->sem_perm);
1244 if (ipcperms(ns, &sma->sem_perm, S_IRUGO))
1248 err = security_sem_semctl(&sma->sem_perm, cmd);
1252 ipc_lock_object(&sma->sem_perm);
1254 if (!ipc_valid_object(&sma->sem_perm)) {
1255 ipc_unlock_object(&sma->sem_perm);
1260 kernel_to_ipc64_perm(&sma->sem_perm, &semid64->sem_perm);
1261 semotime = get_semotime(sma);
1262 semid64->sem_otime = semotime;
1263 semid64->sem_ctime = sma->sem_ctime;
1264 #ifndef CONFIG_64BIT
1265 semid64->sem_otime_high = semotime >> 32;
1266 semid64->sem_ctime_high = sma->sem_ctime >> 32;
1268 semid64->sem_nsems = sma->sem_nsems;
1270 if (cmd == IPC_STAT) {
1272 * As defined in SUS:
1273 * Return 0 on success
1278 * SEM_STAT and SEM_STAT_ANY (both Linux specific)
1279 * Return the full id, including the sequence number
1281 err = sma->sem_perm.id;
1283 ipc_unlock_object(&sma->sem_perm);
1289 static int semctl_info(struct ipc_namespace *ns, int semid,
1290 int cmd, void __user *p)
1292 struct seminfo seminfo;
1296 err = security_sem_semctl(NULL, cmd);
1300 memset(&seminfo, 0, sizeof(seminfo));
1301 seminfo.semmni = ns->sc_semmni;
1302 seminfo.semmns = ns->sc_semmns;
1303 seminfo.semmsl = ns->sc_semmsl;
1304 seminfo.semopm = ns->sc_semopm;
1305 seminfo.semvmx = SEMVMX;
1306 seminfo.semmnu = SEMMNU;
1307 seminfo.semmap = SEMMAP;
1308 seminfo.semume = SEMUME;
1309 down_read(&sem_ids(ns).rwsem);
1310 if (cmd == SEM_INFO) {
1311 seminfo.semusz = sem_ids(ns).in_use;
1312 seminfo.semaem = ns->used_sems;
1314 seminfo.semusz = SEMUSZ;
1315 seminfo.semaem = SEMAEM;
1317 max_idx = ipc_get_maxidx(&sem_ids(ns));
1318 up_read(&sem_ids(ns).rwsem);
1319 if (copy_to_user(p, &seminfo, sizeof(struct seminfo)))
1321 return (max_idx < 0) ? 0 : max_idx;
1324 static int semctl_setval(struct ipc_namespace *ns, int semid, int semnum,
1327 struct sem_undo *un;
1328 struct sem_array *sma;
1331 DEFINE_WAKE_Q(wake_q);
1333 if (val > SEMVMX || val < 0)
1337 sma = sem_obtain_object_check(ns, semid);
1340 return PTR_ERR(sma);
1343 if (semnum < 0 || semnum >= sma->sem_nsems) {
1349 if (ipcperms(ns, &sma->sem_perm, S_IWUGO)) {
1354 err = security_sem_semctl(&sma->sem_perm, SETVAL);
1360 sem_lock(sma, NULL, -1);
1362 if (!ipc_valid_object(&sma->sem_perm)) {
1363 sem_unlock(sma, -1);
1368 semnum = array_index_nospec(semnum, sma->sem_nsems);
1369 curr = &sma->sems[semnum];
1371 ipc_assert_locked_object(&sma->sem_perm);
1372 list_for_each_entry(un, &sma->list_id, list_id)
1373 un->semadj[semnum] = 0;
1376 ipc_update_pid(&curr->sempid, task_tgid(current));
1377 sma->sem_ctime = ktime_get_real_seconds();
1378 /* maybe some queued-up processes were waiting for this */
1379 do_smart_update(sma, NULL, 0, 0, &wake_q);
1380 sem_unlock(sma, -1);
1386 static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
1387 int cmd, void __user *p)
1389 struct sem_array *sma;
1392 ushort fast_sem_io[SEMMSL_FAST];
1393 ushort *sem_io = fast_sem_io;
1394 DEFINE_WAKE_Q(wake_q);
1397 sma = sem_obtain_object_check(ns, semid);
1400 return PTR_ERR(sma);
1403 nsems = sma->sem_nsems;
1406 if (ipcperms(ns, &sma->sem_perm, cmd == SETALL ? S_IWUGO : S_IRUGO))
1407 goto out_rcu_wakeup;
1409 err = security_sem_semctl(&sma->sem_perm, cmd);
1411 goto out_rcu_wakeup;
1417 ushort __user *array = p;
1420 sem_lock(sma, NULL, -1);
1421 if (!ipc_valid_object(&sma->sem_perm)) {
1425 if (nsems > SEMMSL_FAST) {
1426 if (!ipc_rcu_getref(&sma->sem_perm)) {
1430 sem_unlock(sma, -1);
1432 sem_io = kvmalloc_array(nsems, sizeof(ushort),
1434 if (sem_io == NULL) {
1435 ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
1440 sem_lock_and_putref(sma);
1441 if (!ipc_valid_object(&sma->sem_perm)) {
1446 for (i = 0; i < sma->sem_nsems; i++)
1447 sem_io[i] = sma->sems[i].semval;
1448 sem_unlock(sma, -1);
1451 if (copy_to_user(array, sem_io, nsems*sizeof(ushort)))
1458 struct sem_undo *un;
1460 if (!ipc_rcu_getref(&sma->sem_perm)) {
1462 goto out_rcu_wakeup;
1466 if (nsems > SEMMSL_FAST) {
1467 sem_io = kvmalloc_array(nsems, sizeof(ushort),
1469 if (sem_io == NULL) {
1470 ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
1475 if (copy_from_user(sem_io, p, nsems*sizeof(ushort))) {
1476 ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
1481 for (i = 0; i < nsems; i++) {
1482 if (sem_io[i] > SEMVMX) {
1483 ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
1489 sem_lock_and_putref(sma);
1490 if (!ipc_valid_object(&sma->sem_perm)) {
1495 for (i = 0; i < nsems; i++) {
1496 sma->sems[i].semval = sem_io[i];
1497 ipc_update_pid(&sma->sems[i].sempid, task_tgid(current));
1500 ipc_assert_locked_object(&sma->sem_perm);
1501 list_for_each_entry(un, &sma->list_id, list_id) {
1502 for (i = 0; i < nsems; i++)
1505 sma->sem_ctime = ktime_get_real_seconds();
1506 /* maybe some queued-up processes were waiting for this */
1507 do_smart_update(sma, NULL, 0, 0, &wake_q);
1511 /* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */
1514 if (semnum < 0 || semnum >= nsems)
1515 goto out_rcu_wakeup;
1517 sem_lock(sma, NULL, -1);
1518 if (!ipc_valid_object(&sma->sem_perm)) {
1523 semnum = array_index_nospec(semnum, nsems);
1524 curr = &sma->sems[semnum];
1531 err = pid_vnr(curr->sempid);
1534 err = count_semcnt(sma, semnum, 0);
1537 err = count_semcnt(sma, semnum, 1);
1542 sem_unlock(sma, -1);
1547 if (sem_io != fast_sem_io)
1552 static inline unsigned long
1553 copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
1557 if (copy_from_user(out, buf, sizeof(*out)))
1562 struct semid_ds tbuf_old;
1564 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
1567 out->sem_perm.uid = tbuf_old.sem_perm.uid;
1568 out->sem_perm.gid = tbuf_old.sem_perm.gid;
1569 out->sem_perm.mode = tbuf_old.sem_perm.mode;
1579 * This function handles some semctl commands which require the rwsem
1580 * to be held in write mode.
1581 * NOTE: no locks must be held, the rwsem is taken inside this function.
1583 static int semctl_down(struct ipc_namespace *ns, int semid,
1584 int cmd, struct semid64_ds *semid64)
1586 struct sem_array *sma;
1588 struct kern_ipc_perm *ipcp;
1590 down_write(&sem_ids(ns).rwsem);
1593 ipcp = ipcctl_obtain_check(ns, &sem_ids(ns), semid, cmd,
1594 &semid64->sem_perm, 0);
1596 err = PTR_ERR(ipcp);
1600 sma = container_of(ipcp, struct sem_array, sem_perm);
1602 err = security_sem_semctl(&sma->sem_perm, cmd);
1608 sem_lock(sma, NULL, -1);
1609 /* freeary unlocks the ipc object and rcu */
1613 sem_lock(sma, NULL, -1);
1614 err = ipc_update_perm(&semid64->sem_perm, ipcp);
1617 sma->sem_ctime = ktime_get_real_seconds();
1625 sem_unlock(sma, -1);
1629 up_write(&sem_ids(ns).rwsem);
1633 static long ksys_semctl(int semid, int semnum, int cmd, unsigned long arg, int version)
1635 struct ipc_namespace *ns;
1636 void __user *p = (void __user *)arg;
1637 struct semid64_ds semid64;
1643 ns = current->nsproxy->ipc_ns;
1648 return semctl_info(ns, semid, cmd, p);
1652 err = semctl_stat(ns, semid, cmd, &semid64);
1655 if (copy_semid_to_user(p, &semid64, version))
1664 return semctl_main(ns, semid, semnum, cmd, p);
1667 #if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN)
1668 /* big-endian 64bit */
1671 /* 32bit or little-endian 64bit */
1674 return semctl_setval(ns, semid, semnum, val);
1677 if (copy_semid_from_user(&semid64, p, version))
1681 return semctl_down(ns, semid, cmd, &semid64);
1687 SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, unsigned long, arg)
1689 return ksys_semctl(semid, semnum, cmd, arg, IPC_64);
1692 #ifdef CONFIG_ARCH_WANT_IPC_PARSE_VERSION
1693 long ksys_old_semctl(int semid, int semnum, int cmd, unsigned long arg)
1695 int version = ipc_parse_version(&cmd);
1697 return ksys_semctl(semid, semnum, cmd, arg, version);
1700 SYSCALL_DEFINE4(old_semctl, int, semid, int, semnum, int, cmd, unsigned long, arg)
1702 return ksys_old_semctl(semid, semnum, cmd, arg);
1706 #ifdef CONFIG_COMPAT
1708 struct compat_semid_ds {
1709 struct compat_ipc_perm sem_perm;
1710 old_time32_t sem_otime;
1711 old_time32_t sem_ctime;
1712 compat_uptr_t sem_base;
1713 compat_uptr_t sem_pending;
1714 compat_uptr_t sem_pending_last;
1716 unsigned short sem_nsems;
1719 static int copy_compat_semid_from_user(struct semid64_ds *out, void __user *buf,
1722 memset(out, 0, sizeof(*out));
1723 if (version == IPC_64) {
1724 struct compat_semid64_ds __user *p = buf;
1725 return get_compat_ipc64_perm(&out->sem_perm, &p->sem_perm);
1727 struct compat_semid_ds __user *p = buf;
1728 return get_compat_ipc_perm(&out->sem_perm, &p->sem_perm);
1732 static int copy_compat_semid_to_user(void __user *buf, struct semid64_ds *in,
1735 if (version == IPC_64) {
1736 struct compat_semid64_ds v;
1737 memset(&v, 0, sizeof(v));
1738 to_compat_ipc64_perm(&v.sem_perm, &in->sem_perm);
1739 v.sem_otime = lower_32_bits(in->sem_otime);
1740 v.sem_otime_high = upper_32_bits(in->sem_otime);
1741 v.sem_ctime = lower_32_bits(in->sem_ctime);
1742 v.sem_ctime_high = upper_32_bits(in->sem_ctime);
1743 v.sem_nsems = in->sem_nsems;
1744 return copy_to_user(buf, &v, sizeof(v));
1746 struct compat_semid_ds v;
1747 memset(&v, 0, sizeof(v));
1748 to_compat_ipc_perm(&v.sem_perm, &in->sem_perm);
1749 v.sem_otime = in->sem_otime;
1750 v.sem_ctime = in->sem_ctime;
1751 v.sem_nsems = in->sem_nsems;
1752 return copy_to_user(buf, &v, sizeof(v));
1756 static long compat_ksys_semctl(int semid, int semnum, int cmd, int arg, int version)
1758 void __user *p = compat_ptr(arg);
1759 struct ipc_namespace *ns;
1760 struct semid64_ds semid64;
1763 ns = current->nsproxy->ipc_ns;
1768 switch (cmd & (~IPC_64)) {
1771 return semctl_info(ns, semid, cmd, p);
1775 err = semctl_stat(ns, semid, cmd, &semid64);
1778 if (copy_compat_semid_to_user(p, &semid64, version))
1787 return semctl_main(ns, semid, semnum, cmd, p);
1789 return semctl_setval(ns, semid, semnum, arg);
1791 if (copy_compat_semid_from_user(&semid64, p, version))
1795 return semctl_down(ns, semid, cmd, &semid64);
1801 COMPAT_SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, int, arg)
1803 return compat_ksys_semctl(semid, semnum, cmd, arg, IPC_64);
1806 #ifdef CONFIG_ARCH_WANT_COMPAT_IPC_PARSE_VERSION
1807 long compat_ksys_old_semctl(int semid, int semnum, int cmd, int arg)
1809 int version = compat_ipc_parse_version(&cmd);
1811 return compat_ksys_semctl(semid, semnum, cmd, arg, version);
1814 COMPAT_SYSCALL_DEFINE4(old_semctl, int, semid, int, semnum, int, cmd, int, arg)
1816 return compat_ksys_old_semctl(semid, semnum, cmd, arg);
1821 /* If the task doesn't already have a undo_list, then allocate one
1822 * here. We guarantee there is only one thread using this undo list,
1823 * and current is THE ONE
1825 * If this allocation and assignment succeeds, but later
1826 * portions of this code fail, there is no need to free the sem_undo_list.
1827 * Just let it stay associated with the task, and it'll be freed later
1830 * This can block, so callers must hold no locks.
1832 static inline int get_undo_list(struct sem_undo_list **undo_listp)
1834 struct sem_undo_list *undo_list;
1836 undo_list = current->sysvsem.undo_list;
1838 undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
1839 if (undo_list == NULL)
1841 spin_lock_init(&undo_list->lock);
1842 refcount_set(&undo_list->refcnt, 1);
1843 INIT_LIST_HEAD(&undo_list->list_proc);
1845 current->sysvsem.undo_list = undo_list;
1847 *undo_listp = undo_list;
1851 static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid)
1853 struct sem_undo *un;
1855 list_for_each_entry_rcu(un, &ulp->list_proc, list_proc,
1856 spin_is_locked(&ulp->lock)) {
1857 if (un->semid == semid)
1863 static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1865 struct sem_undo *un;
1867 assert_spin_locked(&ulp->lock);
1869 un = __lookup_undo(ulp, semid);
1871 list_del_rcu(&un->list_proc);
1872 list_add_rcu(&un->list_proc, &ulp->list_proc);
1878 * find_alloc_undo - lookup (and if not present create) undo array
1880 * @semid: semaphore array id
1882 * The function looks up (and if not present creates) the undo structure.
1883 * The size of the undo structure depends on the size of the semaphore
1884 * array, thus the alloc path is not that straightforward.
1885 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1886 * performs a rcu_read_lock().
1888 static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
1890 struct sem_array *sma;
1891 struct sem_undo_list *ulp;
1892 struct sem_undo *un, *new;
1895 error = get_undo_list(&ulp);
1897 return ERR_PTR(error);
1900 spin_lock(&ulp->lock);
1901 un = lookup_undo(ulp, semid);
1902 spin_unlock(&ulp->lock);
1903 if (likely(un != NULL))
1906 /* no undo structure around - allocate one. */
1907 /* step 1: figure out the size of the semaphore array */
1908 sma = sem_obtain_object_check(ns, semid);
1911 return ERR_CAST(sma);
1914 nsems = sma->sem_nsems;
1915 if (!ipc_rcu_getref(&sma->sem_perm)) {
1917 un = ERR_PTR(-EIDRM);
1922 /* step 2: allocate new undo structure */
1923 new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1925 ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
1926 return ERR_PTR(-ENOMEM);
1929 /* step 3: Acquire the lock on semaphore array */
1931 sem_lock_and_putref(sma);
1932 if (!ipc_valid_object(&sma->sem_perm)) {
1933 sem_unlock(sma, -1);
1936 un = ERR_PTR(-EIDRM);
1939 spin_lock(&ulp->lock);
1942 * step 4: check for races: did someone else allocate the undo struct?
1944 un = lookup_undo(ulp, semid);
1949 /* step 5: initialize & link new undo structure */
1950 new->semadj = (short *) &new[1];
1953 assert_spin_locked(&ulp->lock);
1954 list_add_rcu(&new->list_proc, &ulp->list_proc);
1955 ipc_assert_locked_object(&sma->sem_perm);
1956 list_add(&new->list_id, &sma->list_id);
1960 spin_unlock(&ulp->lock);
1961 sem_unlock(sma, -1);
1966 static long do_semtimedop(int semid, struct sembuf __user *tsops,
1967 unsigned nsops, const struct timespec64 *timeout)
1969 int error = -EINVAL;
1970 struct sem_array *sma;
1971 struct sembuf fast_sops[SEMOPM_FAST];
1972 struct sembuf *sops = fast_sops, *sop;
1973 struct sem_undo *un;
1975 bool undos = false, alter = false, dupsop = false;
1976 struct sem_queue queue;
1977 unsigned long dup = 0, jiffies_left = 0;
1978 struct ipc_namespace *ns;
1980 ns = current->nsproxy->ipc_ns;
1982 if (nsops < 1 || semid < 0)
1984 if (nsops > ns->sc_semopm)
1986 if (nsops > SEMOPM_FAST) {
1987 sops = kvmalloc_array(nsops, sizeof(*sops), GFP_KERNEL);
1992 if (copy_from_user(sops, tsops, nsops * sizeof(*tsops))) {
1998 if (timeout->tv_sec < 0 || timeout->tv_nsec < 0 ||
1999 timeout->tv_nsec >= 1000000000L) {
2003 jiffies_left = timespec64_to_jiffies(timeout);
2007 for (sop = sops; sop < sops + nsops; sop++) {
2008 unsigned long mask = 1ULL << ((sop->sem_num) % BITS_PER_LONG);
2010 if (sop->sem_num >= max)
2012 if (sop->sem_flg & SEM_UNDO)
2016 * There was a previous alter access that appears
2017 * to have accessed the same semaphore, thus use
2018 * the dupsop logic. "appears", because the detection
2019 * can only check % BITS_PER_LONG.
2023 if (sop->sem_op != 0) {
2030 /* On success, find_alloc_undo takes the rcu_read_lock */
2031 un = find_alloc_undo(ns, semid);
2033 error = PTR_ERR(un);
2041 sma = sem_obtain_object_check(ns, semid);
2044 error = PTR_ERR(sma);
2049 if (max >= sma->sem_nsems) {
2055 if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO)) {
2060 error = security_sem_semop(&sma->sem_perm, sops, nsops, alter);
2067 locknum = sem_lock(sma, sops, nsops);
2069 * We eventually might perform the following check in a lockless
2070 * fashion, considering ipc_valid_object() locking constraints.
2071 * If nsops == 1 and there is no contention for sem_perm.lock, then
2072 * only a per-semaphore lock is held and it's OK to proceed with the
2073 * check below. More details on the fine grained locking scheme
2074 * entangled here and why it's RMID race safe on comments at sem_lock()
2076 if (!ipc_valid_object(&sma->sem_perm))
2077 goto out_unlock_free;
2079 * semid identifiers are not unique - find_alloc_undo may have
2080 * allocated an undo structure, it was invalidated by an RMID
2081 * and now a new array with received the same id. Check and fail.
2082 * This case can be detected checking un->semid. The existence of
2083 * "un" itself is guaranteed by rcu.
2085 if (un && un->semid == -1)
2086 goto out_unlock_free;
2089 queue.nsops = nsops;
2091 queue.pid = task_tgid(current);
2092 queue.alter = alter;
2093 queue.dupsop = dupsop;
2095 error = perform_atomic_semop(sma, &queue);
2096 if (error == 0) { /* non-blocking succesfull path */
2097 DEFINE_WAKE_Q(wake_q);
2100 * If the operation was successful, then do
2101 * the required updates.
2104 do_smart_update(sma, sops, nsops, 1, &wake_q);
2106 set_semotime(sma, sops);
2108 sem_unlock(sma, locknum);
2114 if (error < 0) /* non-blocking error path */
2115 goto out_unlock_free;
2118 * We need to sleep on this operation, so we put the current
2119 * task into the pending queue and go to sleep.
2123 int idx = array_index_nospec(sops->sem_num, sma->sem_nsems);
2124 curr = &sma->sems[idx];
2127 if (sma->complex_count) {
2128 list_add_tail(&queue.list,
2129 &sma->pending_alter);
2132 list_add_tail(&queue.list,
2133 &curr->pending_alter);
2136 list_add_tail(&queue.list, &curr->pending_const);
2139 if (!sma->complex_count)
2143 list_add_tail(&queue.list, &sma->pending_alter);
2145 list_add_tail(&queue.list, &sma->pending_const);
2147 sma->complex_count++;
2151 WRITE_ONCE(queue.status, -EINTR);
2152 queue.sleeper = current;
2154 __set_current_state(TASK_INTERRUPTIBLE);
2155 sem_unlock(sma, locknum);
2159 jiffies_left = schedule_timeout(jiffies_left);
2164 * fastpath: the semop has completed, either successfully or
2165 * not, from the syscall pov, is quite irrelevant to us at this
2166 * point; we're done.
2168 * We _do_ care, nonetheless, about being awoken by a signal or
2169 * spuriously. The queue.status is checked again in the
2170 * slowpath (aka after taking sem_lock), such that we can detect
2171 * scenarios where we were awakened externally, during the
2172 * window between wake_q_add() and wake_up_q().
2174 error = READ_ONCE(queue.status);
2175 if (error != -EINTR) {
2177 * User space could assume that semop() is a memory
2178 * barrier: Without the mb(), the cpu could
2179 * speculatively read in userspace stale data that was
2180 * overwritten by the previous owner of the semaphore.
2187 locknum = sem_lock(sma, sops, nsops);
2189 if (!ipc_valid_object(&sma->sem_perm))
2190 goto out_unlock_free;
2192 error = READ_ONCE(queue.status);
2195 * If queue.status != -EINTR we are woken up by another process.
2196 * Leave without unlink_queue(), but with sem_unlock().
2198 if (error != -EINTR)
2199 goto out_unlock_free;
2202 * If an interrupt occurred we have to clean up the queue.
2204 if (timeout && jiffies_left == 0)
2206 } while (error == -EINTR && !signal_pending(current)); /* spurious */
2208 unlink_queue(sma, &queue);
2211 sem_unlock(sma, locknum);
2214 if (sops != fast_sops)
2219 long ksys_semtimedop(int semid, struct sembuf __user *tsops,
2220 unsigned int nsops, const struct __kernel_timespec __user *timeout)
2223 struct timespec64 ts;
2224 if (get_timespec64(&ts, timeout))
2226 return do_semtimedop(semid, tsops, nsops, &ts);
2228 return do_semtimedop(semid, tsops, nsops, NULL);
2231 SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
2232 unsigned int, nsops, const struct __kernel_timespec __user *, timeout)
2234 return ksys_semtimedop(semid, tsops, nsops, timeout);
2237 #ifdef CONFIG_COMPAT_32BIT_TIME
2238 long compat_ksys_semtimedop(int semid, struct sembuf __user *tsems,
2240 const struct old_timespec32 __user *timeout)
2243 struct timespec64 ts;
2244 if (get_old_timespec32(&ts, timeout))
2246 return do_semtimedop(semid, tsems, nsops, &ts);
2248 return do_semtimedop(semid, tsems, nsops, NULL);
2251 SYSCALL_DEFINE4(semtimedop_time32, int, semid, struct sembuf __user *, tsems,
2252 unsigned int, nsops,
2253 const struct old_timespec32 __user *, timeout)
2255 return compat_ksys_semtimedop(semid, tsems, nsops, timeout);
2259 SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
2262 return do_semtimedop(semid, tsops, nsops, NULL);
2265 /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
2266 * parent and child tasks.
2269 int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
2271 struct sem_undo_list *undo_list;
2274 if (clone_flags & CLONE_SYSVSEM) {
2275 error = get_undo_list(&undo_list);
2278 refcount_inc(&undo_list->refcnt);
2279 tsk->sysvsem.undo_list = undo_list;
2281 tsk->sysvsem.undo_list = NULL;
2287 * add semadj values to semaphores, free undo structures.
2288 * undo structures are not freed when semaphore arrays are destroyed
2289 * so some of them may be out of date.
2290 * IMPLEMENTATION NOTE: There is some confusion over whether the
2291 * set of adjustments that needs to be done should be done in an atomic
2292 * manner or not. That is, if we are attempting to decrement the semval
2293 * should we queue up and wait until we can do so legally?
2294 * The original implementation attempted to do this (queue and wait).
2295 * The current implementation does not do so. The POSIX standard
2296 * and SVID should be consulted to determine what behavior is mandated.
2298 void exit_sem(struct task_struct *tsk)
2300 struct sem_undo_list *ulp;
2302 ulp = tsk->sysvsem.undo_list;
2305 tsk->sysvsem.undo_list = NULL;
2307 if (!refcount_dec_and_test(&ulp->refcnt))
2311 struct sem_array *sma;
2312 struct sem_undo *un;
2314 DEFINE_WAKE_Q(wake_q);
2319 un = list_entry_rcu(ulp->list_proc.next,
2320 struct sem_undo, list_proc);
2321 if (&un->list_proc == &ulp->list_proc) {
2323 * We must wait for freeary() before freeing this ulp,
2324 * in case we raced with last sem_undo. There is a small
2325 * possibility where we exit while freeary() didn't
2326 * finish unlocking sem_undo_list.
2328 spin_lock(&ulp->lock);
2329 spin_unlock(&ulp->lock);
2333 spin_lock(&ulp->lock);
2335 spin_unlock(&ulp->lock);
2337 /* exit_sem raced with IPC_RMID, nothing to do */
2343 sma = sem_obtain_object_check(tsk->nsproxy->ipc_ns, semid);
2344 /* exit_sem raced with IPC_RMID, nothing to do */
2350 sem_lock(sma, NULL, -1);
2351 /* exit_sem raced with IPC_RMID, nothing to do */
2352 if (!ipc_valid_object(&sma->sem_perm)) {
2353 sem_unlock(sma, -1);
2357 un = __lookup_undo(ulp, semid);
2359 /* exit_sem raced with IPC_RMID+semget() that created
2360 * exactly the same semid. Nothing to do.
2362 sem_unlock(sma, -1);
2367 /* remove un from the linked lists */
2368 ipc_assert_locked_object(&sma->sem_perm);
2369 list_del(&un->list_id);
2371 /* we are the last process using this ulp, acquiring ulp->lock
2372 * isn't required. Besides that, we are also protected against
2373 * IPC_RMID as we hold sma->sem_perm lock now
2375 list_del_rcu(&un->list_proc);
2377 /* perform adjustments registered in un */
2378 for (i = 0; i < sma->sem_nsems; i++) {
2379 struct sem *semaphore = &sma->sems[i];
2380 if (un->semadj[i]) {
2381 semaphore->semval += un->semadj[i];
2383 * Range checks of the new semaphore value,
2384 * not defined by sus:
2385 * - Some unices ignore the undo entirely
2386 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
2387 * - some cap the value (e.g. FreeBSD caps
2388 * at 0, but doesn't enforce SEMVMX)
2390 * Linux caps the semaphore value, both at 0
2393 * Manfred <manfred@colorfullife.com>
2395 if (semaphore->semval < 0)
2396 semaphore->semval = 0;
2397 if (semaphore->semval > SEMVMX)
2398 semaphore->semval = SEMVMX;
2399 ipc_update_pid(&semaphore->sempid, task_tgid(current));
2402 /* maybe some queued-up processes were waiting for this */
2403 do_smart_update(sma, NULL, 0, 1, &wake_q);
2404 sem_unlock(sma, -1);
2413 #ifdef CONFIG_PROC_FS
2414 static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
2416 struct user_namespace *user_ns = seq_user_ns(s);
2417 struct kern_ipc_perm *ipcp = it;
2418 struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
2422 * The proc interface isn't aware of sem_lock(), it calls
2423 * ipc_lock_object() directly (in sysvipc_find_ipc).
2424 * In order to stay compatible with sem_lock(), we must
2425 * enter / leave complex_mode.
2427 complexmode_enter(sma);
2429 sem_otime = get_semotime(sma);
2432 "%10d %10d %4o %10u %5u %5u %5u %5u %10llu %10llu\n",
2437 from_kuid_munged(user_ns, sma->sem_perm.uid),
2438 from_kgid_munged(user_ns, sma->sem_perm.gid),
2439 from_kuid_munged(user_ns, sma->sem_perm.cuid),
2440 from_kgid_munged(user_ns, sma->sem_perm.cgid),
2444 complexmode_tryleave(sma);