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;
493 if (nsems > (INT_MAX - sizeof(*sma)) / sizeof(sma->sems[0]))
496 size = sizeof(*sma) + nsems * sizeof(sma->sems[0]);
497 sma = kvmalloc(size, GFP_KERNEL);
501 memset(sma, 0, size);
507 * newary - Create a new semaphore set
509 * @params: ptr to the structure that contains key, semflg and nsems
511 * Called with sem_ids.rwsem held (as a writer)
513 static int newary(struct ipc_namespace *ns, struct ipc_params *params)
516 struct sem_array *sma;
517 key_t key = params->key;
518 int nsems = params->u.nsems;
519 int semflg = params->flg;
524 if (ns->used_sems + nsems > ns->sc_semmns)
527 sma = sem_alloc(nsems);
531 sma->sem_perm.mode = (semflg & S_IRWXUGO);
532 sma->sem_perm.key = key;
534 sma->sem_perm.security = NULL;
535 retval = security_sem_alloc(&sma->sem_perm);
541 for (i = 0; i < nsems; i++) {
542 INIT_LIST_HEAD(&sma->sems[i].pending_alter);
543 INIT_LIST_HEAD(&sma->sems[i].pending_const);
544 spin_lock_init(&sma->sems[i].lock);
547 sma->complex_count = 0;
548 sma->use_global_lock = USE_GLOBAL_LOCK_HYSTERESIS;
549 INIT_LIST_HEAD(&sma->pending_alter);
550 INIT_LIST_HEAD(&sma->pending_const);
551 INIT_LIST_HEAD(&sma->list_id);
552 sma->sem_nsems = nsems;
553 sma->sem_ctime = ktime_get_real_seconds();
555 /* ipc_addid() locks sma upon success. */
556 retval = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
558 ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
561 ns->used_sems += nsems;
566 return sma->sem_perm.id;
571 * Called with sem_ids.rwsem and ipcp locked.
573 static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
574 struct ipc_params *params)
576 struct sem_array *sma;
578 sma = container_of(ipcp, struct sem_array, sem_perm);
579 if (params->u.nsems > sma->sem_nsems)
585 long ksys_semget(key_t key, int nsems, int semflg)
587 struct ipc_namespace *ns;
588 static const struct ipc_ops sem_ops = {
590 .associate = security_sem_associate,
591 .more_checks = sem_more_checks,
593 struct ipc_params sem_params;
595 ns = current->nsproxy->ipc_ns;
597 if (nsems < 0 || nsems > ns->sc_semmsl)
600 sem_params.key = key;
601 sem_params.flg = semflg;
602 sem_params.u.nsems = nsems;
604 return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
607 SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg)
609 return ksys_semget(key, nsems, semflg);
613 * perform_atomic_semop[_slow] - Attempt to perform semaphore
614 * operations on a given array.
615 * @sma: semaphore array
616 * @q: struct sem_queue that describes the operation
618 * Caller blocking are as follows, based the value
619 * indicated by the semaphore operation (sem_op):
621 * (1) >0 never blocks.
622 * (2) 0 (wait-for-zero operation): semval is non-zero.
623 * (3) <0 attempting to decrement semval to a value smaller than zero.
625 * Returns 0 if the operation was possible.
626 * Returns 1 if the operation is impossible, the caller must sleep.
627 * Returns <0 for error codes.
629 static int perform_atomic_semop_slow(struct sem_array *sma, struct sem_queue *q)
631 int result, sem_op, nsops;
642 for (sop = sops; sop < sops + nsops; sop++) {
643 int idx = array_index_nospec(sop->sem_num, sma->sem_nsems);
644 curr = &sma->sems[idx];
645 sem_op = sop->sem_op;
646 result = curr->semval;
648 if (!sem_op && result)
657 if (sop->sem_flg & SEM_UNDO) {
658 int undo = un->semadj[sop->sem_num] - sem_op;
659 /* Exceeding the undo range is an error. */
660 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
662 un->semadj[sop->sem_num] = undo;
665 curr->semval = result;
670 while (sop >= sops) {
671 ipc_update_pid(&sma->sems[sop->sem_num].sempid, pid);
684 if (sop->sem_flg & IPC_NOWAIT)
691 while (sop >= sops) {
692 sem_op = sop->sem_op;
693 sma->sems[sop->sem_num].semval -= sem_op;
694 if (sop->sem_flg & SEM_UNDO)
695 un->semadj[sop->sem_num] += sem_op;
702 static int perform_atomic_semop(struct sem_array *sma, struct sem_queue *q)
704 int result, sem_op, nsops;
714 if (unlikely(q->dupsop))
715 return perform_atomic_semop_slow(sma, q);
718 * We scan the semaphore set twice, first to ensure that the entire
719 * operation can succeed, therefore avoiding any pointless writes
720 * to shared memory and having to undo such changes in order to block
721 * until the operations can go through.
723 for (sop = sops; sop < sops + nsops; sop++) {
724 int idx = array_index_nospec(sop->sem_num, sma->sem_nsems);
726 curr = &sma->sems[idx];
727 sem_op = sop->sem_op;
728 result = curr->semval;
730 if (!sem_op && result)
731 goto would_block; /* wait-for-zero */
740 if (sop->sem_flg & SEM_UNDO) {
741 int undo = un->semadj[sop->sem_num] - sem_op;
743 /* Exceeding the undo range is an error. */
744 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
749 for (sop = sops; sop < sops + nsops; sop++) {
750 curr = &sma->sems[sop->sem_num];
751 sem_op = sop->sem_op;
752 result = curr->semval;
754 if (sop->sem_flg & SEM_UNDO) {
755 int undo = un->semadj[sop->sem_num] - sem_op;
757 un->semadj[sop->sem_num] = undo;
759 curr->semval += sem_op;
760 ipc_update_pid(&curr->sempid, q->pid);
767 return sop->sem_flg & IPC_NOWAIT ? -EAGAIN : 1;
770 static inline void wake_up_sem_queue_prepare(struct sem_queue *q, int error,
771 struct wake_q_head *wake_q)
773 wake_q_add(wake_q, q->sleeper);
775 * Rely on the above implicit barrier, such that we can
776 * ensure that we hold reference to the task before setting
777 * q->status. Otherwise we could race with do_exit if the
778 * task is awoken by an external event before calling
781 WRITE_ONCE(q->status, error);
784 static void unlink_queue(struct sem_array *sma, struct sem_queue *q)
788 sma->complex_count--;
791 /** check_restart(sma, q)
792 * @sma: semaphore array
793 * @q: the operation that just completed
795 * update_queue is O(N^2) when it restarts scanning the whole queue of
796 * waiting operations. Therefore this function checks if the restart is
797 * really necessary. It is called after a previously waiting operation
798 * modified the array.
799 * Note that wait-for-zero operations are handled without restart.
801 static inline int check_restart(struct sem_array *sma, struct sem_queue *q)
803 /* pending complex alter operations are too difficult to analyse */
804 if (!list_empty(&sma->pending_alter))
807 /* we were a sleeping complex operation. Too difficult */
811 /* It is impossible that someone waits for the new value:
812 * - complex operations always restart.
813 * - wait-for-zero are handled seperately.
814 * - q is a previously sleeping simple operation that
815 * altered the array. It must be a decrement, because
816 * simple increments never sleep.
817 * - If there are older (higher priority) decrements
818 * in the queue, then they have observed the original
819 * semval value and couldn't proceed. The operation
820 * decremented to value - thus they won't proceed either.
826 * wake_const_ops - wake up non-alter tasks
827 * @sma: semaphore array.
828 * @semnum: semaphore that was modified.
829 * @wake_q: lockless wake-queue head.
831 * wake_const_ops must be called after a semaphore in a semaphore array
832 * was set to 0. If complex const operations are pending, wake_const_ops must
833 * be called with semnum = -1, as well as with the number of each modified
835 * The tasks that must be woken up are added to @wake_q. The return code
836 * is stored in q->pid.
837 * The function returns 1 if at least one operation was completed successfully.
839 static int wake_const_ops(struct sem_array *sma, int semnum,
840 struct wake_q_head *wake_q)
842 struct sem_queue *q, *tmp;
843 struct list_head *pending_list;
844 int semop_completed = 0;
847 pending_list = &sma->pending_const;
849 pending_list = &sma->sems[semnum].pending_const;
851 list_for_each_entry_safe(q, tmp, pending_list, list) {
852 int error = perform_atomic_semop(sma, q);
856 /* operation completed, remove from queue & wakeup */
857 unlink_queue(sma, q);
859 wake_up_sem_queue_prepare(q, error, wake_q);
864 return semop_completed;
868 * do_smart_wakeup_zero - wakeup all wait for zero tasks
869 * @sma: semaphore array
870 * @sops: operations that were performed
871 * @nsops: number of operations
872 * @wake_q: lockless wake-queue head
874 * Checks all required queue for wait-for-zero operations, based
875 * on the actual changes that were performed on the semaphore array.
876 * The function returns 1 if at least one operation was completed successfully.
878 static int do_smart_wakeup_zero(struct sem_array *sma, struct sembuf *sops,
879 int nsops, struct wake_q_head *wake_q)
882 int semop_completed = 0;
885 /* first: the per-semaphore queues, if known */
887 for (i = 0; i < nsops; i++) {
888 int num = sops[i].sem_num;
890 if (sma->sems[num].semval == 0) {
892 semop_completed |= wake_const_ops(sma, num, wake_q);
897 * No sops means modified semaphores not known.
898 * Assume all were changed.
900 for (i = 0; i < sma->sem_nsems; i++) {
901 if (sma->sems[i].semval == 0) {
903 semop_completed |= wake_const_ops(sma, i, wake_q);
908 * If one of the modified semaphores got 0,
909 * then check the global queue, too.
912 semop_completed |= wake_const_ops(sma, -1, wake_q);
914 return semop_completed;
919 * update_queue - look for tasks that can be completed.
920 * @sma: semaphore array.
921 * @semnum: semaphore that was modified.
922 * @wake_q: lockless wake-queue head.
924 * update_queue must be called after a semaphore in a semaphore array
925 * was modified. If multiple semaphores were modified, update_queue must
926 * be called with semnum = -1, as well as with the number of each modified
928 * The tasks that must be woken up are added to @wake_q. The return code
929 * is stored in q->pid.
930 * The function internally checks if const operations can now succeed.
932 * The function return 1 if at least one semop was completed successfully.
934 static int update_queue(struct sem_array *sma, int semnum, struct wake_q_head *wake_q)
936 struct sem_queue *q, *tmp;
937 struct list_head *pending_list;
938 int semop_completed = 0;
941 pending_list = &sma->pending_alter;
943 pending_list = &sma->sems[semnum].pending_alter;
946 list_for_each_entry_safe(q, tmp, pending_list, list) {
949 /* If we are scanning the single sop, per-semaphore list of
950 * one semaphore and that semaphore is 0, then it is not
951 * necessary to scan further: simple increments
952 * that affect only one entry succeed immediately and cannot
953 * be in the per semaphore pending queue, and decrements
954 * cannot be successful if the value is already 0.
956 if (semnum != -1 && sma->sems[semnum].semval == 0)
959 error = perform_atomic_semop(sma, q);
961 /* Does q->sleeper still need to sleep? */
965 unlink_queue(sma, q);
971 do_smart_wakeup_zero(sma, q->sops, q->nsops, wake_q);
972 restart = check_restart(sma, q);
975 wake_up_sem_queue_prepare(q, error, wake_q);
979 return semop_completed;
983 * set_semotime - set sem_otime
984 * @sma: semaphore array
985 * @sops: operations that modified the array, may be NULL
987 * sem_otime is replicated to avoid cache line trashing.
988 * This function sets one instance to the current time.
990 static void set_semotime(struct sem_array *sma, struct sembuf *sops)
993 sma->sems[0].sem_otime = ktime_get_real_seconds();
995 sma->sems[sops[0].sem_num].sem_otime =
996 ktime_get_real_seconds();
1001 * do_smart_update - optimized update_queue
1002 * @sma: semaphore array
1003 * @sops: operations that were performed
1004 * @nsops: number of operations
1005 * @otime: force setting otime
1006 * @wake_q: lockless wake-queue head
1008 * do_smart_update() does the required calls to update_queue and wakeup_zero,
1009 * based on the actual changes that were performed on the semaphore array.
1010 * Note that the function does not do the actual wake-up: the caller is
1011 * responsible for calling wake_up_q().
1012 * It is safe to perform this call after dropping all locks.
1014 static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops,
1015 int otime, struct wake_q_head *wake_q)
1019 otime |= do_smart_wakeup_zero(sma, sops, nsops, wake_q);
1021 if (!list_empty(&sma->pending_alter)) {
1022 /* semaphore array uses the global queue - just process it. */
1023 otime |= update_queue(sma, -1, wake_q);
1027 * No sops, thus the modified semaphores are not
1030 for (i = 0; i < sma->sem_nsems; i++)
1031 otime |= update_queue(sma, i, wake_q);
1034 * Check the semaphores that were increased:
1035 * - No complex ops, thus all sleeping ops are
1037 * - if we decreased the value, then any sleeping
1038 * semaphore ops wont be able to run: If the
1039 * previous value was too small, then the new
1040 * value will be too small, too.
1042 for (i = 0; i < nsops; i++) {
1043 if (sops[i].sem_op > 0) {
1044 otime |= update_queue(sma,
1045 sops[i].sem_num, wake_q);
1051 set_semotime(sma, sops);
1055 * check_qop: Test if a queued operation sleeps on the semaphore semnum
1057 static int check_qop(struct sem_array *sma, int semnum, struct sem_queue *q,
1060 struct sembuf *sop = q->blocking;
1063 * Linux always (since 0.99.10) reported a task as sleeping on all
1064 * semaphores. This violates SUS, therefore it was changed to the
1065 * standard compliant behavior.
1066 * Give the administrators a chance to notice that an application
1067 * might misbehave because it relies on the Linux behavior.
1069 pr_info_once("semctl(GETNCNT/GETZCNT) is since 3.16 Single Unix Specification compliant.\n"
1070 "The task %s (%d) triggered the difference, watch for misbehavior.\n",
1071 current->comm, task_pid_nr(current));
1073 if (sop->sem_num != semnum)
1076 if (count_zero && sop->sem_op == 0)
1078 if (!count_zero && sop->sem_op < 0)
1084 /* The following counts are associated to each semaphore:
1085 * semncnt number of tasks waiting on semval being nonzero
1086 * semzcnt number of tasks waiting on semval being zero
1088 * Per definition, a task waits only on the semaphore of the first semop
1089 * that cannot proceed, even if additional operation would block, too.
1091 static int count_semcnt(struct sem_array *sma, ushort semnum,
1094 struct list_head *l;
1095 struct sem_queue *q;
1099 /* First: check the simple operations. They are easy to evaluate */
1101 l = &sma->sems[semnum].pending_const;
1103 l = &sma->sems[semnum].pending_alter;
1105 list_for_each_entry(q, l, list) {
1106 /* all task on a per-semaphore list sleep on exactly
1112 /* Then: check the complex operations. */
1113 list_for_each_entry(q, &sma->pending_alter, list) {
1114 semcnt += check_qop(sma, semnum, q, count_zero);
1117 list_for_each_entry(q, &sma->pending_const, list) {
1118 semcnt += check_qop(sma, semnum, q, count_zero);
1124 /* Free a semaphore set. freeary() is called with sem_ids.rwsem locked
1125 * as a writer and the spinlock for this semaphore set hold. sem_ids.rwsem
1126 * remains locked on exit.
1128 static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
1130 struct sem_undo *un, *tu;
1131 struct sem_queue *q, *tq;
1132 struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
1134 DEFINE_WAKE_Q(wake_q);
1136 /* Free the existing undo structures for this semaphore set. */
1137 ipc_assert_locked_object(&sma->sem_perm);
1138 list_for_each_entry_safe(un, tu, &sma->list_id, list_id) {
1139 list_del(&un->list_id);
1140 spin_lock(&un->ulp->lock);
1142 list_del_rcu(&un->list_proc);
1143 spin_unlock(&un->ulp->lock);
1147 /* Wake up all pending processes and let them fail with EIDRM. */
1148 list_for_each_entry_safe(q, tq, &sma->pending_const, list) {
1149 unlink_queue(sma, q);
1150 wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
1153 list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
1154 unlink_queue(sma, q);
1155 wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
1157 for (i = 0; i < sma->sem_nsems; i++) {
1158 struct sem *sem = &sma->sems[i];
1159 list_for_each_entry_safe(q, tq, &sem->pending_const, list) {
1160 unlink_queue(sma, q);
1161 wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
1163 list_for_each_entry_safe(q, tq, &sem->pending_alter, list) {
1164 unlink_queue(sma, q);
1165 wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
1167 ipc_update_pid(&sem->sempid, NULL);
1170 /* Remove the semaphore set from the IDR */
1172 sem_unlock(sma, -1);
1176 ns->used_sems -= sma->sem_nsems;
1177 ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
1180 static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
1184 return copy_to_user(buf, in, sizeof(*in));
1187 struct semid_ds out;
1189 memset(&out, 0, sizeof(out));
1191 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
1193 out.sem_otime = in->sem_otime;
1194 out.sem_ctime = in->sem_ctime;
1195 out.sem_nsems = in->sem_nsems;
1197 return copy_to_user(buf, &out, sizeof(out));
1204 static time64_t get_semotime(struct sem_array *sma)
1209 res = sma->sems[0].sem_otime;
1210 for (i = 1; i < sma->sem_nsems; i++) {
1211 time64_t to = sma->sems[i].sem_otime;
1219 static int semctl_stat(struct ipc_namespace *ns, int semid,
1220 int cmd, struct semid64_ds *semid64)
1222 struct sem_array *sma;
1226 memset(semid64, 0, sizeof(*semid64));
1229 if (cmd == SEM_STAT || cmd == SEM_STAT_ANY) {
1230 sma = sem_obtain_object(ns, semid);
1235 } else { /* IPC_STAT */
1236 sma = sem_obtain_object_check(ns, semid);
1243 /* see comment for SHM_STAT_ANY */
1244 if (cmd == SEM_STAT_ANY)
1245 audit_ipc_obj(&sma->sem_perm);
1248 if (ipcperms(ns, &sma->sem_perm, S_IRUGO))
1252 err = security_sem_semctl(&sma->sem_perm, cmd);
1256 ipc_lock_object(&sma->sem_perm);
1258 if (!ipc_valid_object(&sma->sem_perm)) {
1259 ipc_unlock_object(&sma->sem_perm);
1264 kernel_to_ipc64_perm(&sma->sem_perm, &semid64->sem_perm);
1265 semotime = get_semotime(sma);
1266 semid64->sem_otime = semotime;
1267 semid64->sem_ctime = sma->sem_ctime;
1268 #ifndef CONFIG_64BIT
1269 semid64->sem_otime_high = semotime >> 32;
1270 semid64->sem_ctime_high = sma->sem_ctime >> 32;
1272 semid64->sem_nsems = sma->sem_nsems;
1274 if (cmd == IPC_STAT) {
1276 * As defined in SUS:
1277 * Return 0 on success
1282 * SEM_STAT and SEM_STAT_ANY (both Linux specific)
1283 * Return the full id, including the sequence number
1285 err = sma->sem_perm.id;
1287 ipc_unlock_object(&sma->sem_perm);
1293 static int semctl_info(struct ipc_namespace *ns, int semid,
1294 int cmd, void __user *p)
1296 struct seminfo seminfo;
1300 err = security_sem_semctl(NULL, cmd);
1304 memset(&seminfo, 0, sizeof(seminfo));
1305 seminfo.semmni = ns->sc_semmni;
1306 seminfo.semmns = ns->sc_semmns;
1307 seminfo.semmsl = ns->sc_semmsl;
1308 seminfo.semopm = ns->sc_semopm;
1309 seminfo.semvmx = SEMVMX;
1310 seminfo.semmnu = SEMMNU;
1311 seminfo.semmap = SEMMAP;
1312 seminfo.semume = SEMUME;
1313 down_read(&sem_ids(ns).rwsem);
1314 if (cmd == SEM_INFO) {
1315 seminfo.semusz = sem_ids(ns).in_use;
1316 seminfo.semaem = ns->used_sems;
1318 seminfo.semusz = SEMUSZ;
1319 seminfo.semaem = SEMAEM;
1321 max_idx = ipc_get_maxidx(&sem_ids(ns));
1322 up_read(&sem_ids(ns).rwsem);
1323 if (copy_to_user(p, &seminfo, sizeof(struct seminfo)))
1325 return (max_idx < 0) ? 0 : max_idx;
1328 static int semctl_setval(struct ipc_namespace *ns, int semid, int semnum,
1331 struct sem_undo *un;
1332 struct sem_array *sma;
1335 DEFINE_WAKE_Q(wake_q);
1337 if (val > SEMVMX || val < 0)
1341 sma = sem_obtain_object_check(ns, semid);
1344 return PTR_ERR(sma);
1347 if (semnum < 0 || semnum >= sma->sem_nsems) {
1353 if (ipcperms(ns, &sma->sem_perm, S_IWUGO)) {
1358 err = security_sem_semctl(&sma->sem_perm, SETVAL);
1364 sem_lock(sma, NULL, -1);
1366 if (!ipc_valid_object(&sma->sem_perm)) {
1367 sem_unlock(sma, -1);
1372 semnum = array_index_nospec(semnum, sma->sem_nsems);
1373 curr = &sma->sems[semnum];
1375 ipc_assert_locked_object(&sma->sem_perm);
1376 list_for_each_entry(un, &sma->list_id, list_id)
1377 un->semadj[semnum] = 0;
1380 ipc_update_pid(&curr->sempid, task_tgid(current));
1381 sma->sem_ctime = ktime_get_real_seconds();
1382 /* maybe some queued-up processes were waiting for this */
1383 do_smart_update(sma, NULL, 0, 0, &wake_q);
1384 sem_unlock(sma, -1);
1390 static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
1391 int cmd, void __user *p)
1393 struct sem_array *sma;
1396 ushort fast_sem_io[SEMMSL_FAST];
1397 ushort *sem_io = fast_sem_io;
1398 DEFINE_WAKE_Q(wake_q);
1401 sma = sem_obtain_object_check(ns, semid);
1404 return PTR_ERR(sma);
1407 nsems = sma->sem_nsems;
1410 if (ipcperms(ns, &sma->sem_perm, cmd == SETALL ? S_IWUGO : S_IRUGO))
1411 goto out_rcu_wakeup;
1413 err = security_sem_semctl(&sma->sem_perm, cmd);
1415 goto out_rcu_wakeup;
1421 ushort __user *array = p;
1424 sem_lock(sma, NULL, -1);
1425 if (!ipc_valid_object(&sma->sem_perm)) {
1429 if (nsems > SEMMSL_FAST) {
1430 if (!ipc_rcu_getref(&sma->sem_perm)) {
1434 sem_unlock(sma, -1);
1436 sem_io = kvmalloc_array(nsems, sizeof(ushort),
1438 if (sem_io == NULL) {
1439 ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
1444 sem_lock_and_putref(sma);
1445 if (!ipc_valid_object(&sma->sem_perm)) {
1450 for (i = 0; i < sma->sem_nsems; i++)
1451 sem_io[i] = sma->sems[i].semval;
1452 sem_unlock(sma, -1);
1455 if (copy_to_user(array, sem_io, nsems*sizeof(ushort)))
1462 struct sem_undo *un;
1464 if (!ipc_rcu_getref(&sma->sem_perm)) {
1466 goto out_rcu_wakeup;
1470 if (nsems > SEMMSL_FAST) {
1471 sem_io = kvmalloc_array(nsems, sizeof(ushort),
1473 if (sem_io == NULL) {
1474 ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
1479 if (copy_from_user(sem_io, p, nsems*sizeof(ushort))) {
1480 ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
1485 for (i = 0; i < nsems; i++) {
1486 if (sem_io[i] > SEMVMX) {
1487 ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
1493 sem_lock_and_putref(sma);
1494 if (!ipc_valid_object(&sma->sem_perm)) {
1499 for (i = 0; i < nsems; i++) {
1500 sma->sems[i].semval = sem_io[i];
1501 ipc_update_pid(&sma->sems[i].sempid, task_tgid(current));
1504 ipc_assert_locked_object(&sma->sem_perm);
1505 list_for_each_entry(un, &sma->list_id, list_id) {
1506 for (i = 0; i < nsems; i++)
1509 sma->sem_ctime = ktime_get_real_seconds();
1510 /* maybe some queued-up processes were waiting for this */
1511 do_smart_update(sma, NULL, 0, 0, &wake_q);
1515 /* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */
1518 if (semnum < 0 || semnum >= nsems)
1519 goto out_rcu_wakeup;
1521 sem_lock(sma, NULL, -1);
1522 if (!ipc_valid_object(&sma->sem_perm)) {
1527 semnum = array_index_nospec(semnum, nsems);
1528 curr = &sma->sems[semnum];
1535 err = pid_vnr(curr->sempid);
1538 err = count_semcnt(sma, semnum, 0);
1541 err = count_semcnt(sma, semnum, 1);
1546 sem_unlock(sma, -1);
1551 if (sem_io != fast_sem_io)
1556 static inline unsigned long
1557 copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
1561 if (copy_from_user(out, buf, sizeof(*out)))
1566 struct semid_ds tbuf_old;
1568 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
1571 out->sem_perm.uid = tbuf_old.sem_perm.uid;
1572 out->sem_perm.gid = tbuf_old.sem_perm.gid;
1573 out->sem_perm.mode = tbuf_old.sem_perm.mode;
1583 * This function handles some semctl commands which require the rwsem
1584 * to be held in write mode.
1585 * NOTE: no locks must be held, the rwsem is taken inside this function.
1587 static int semctl_down(struct ipc_namespace *ns, int semid,
1588 int cmd, struct semid64_ds *semid64)
1590 struct sem_array *sma;
1592 struct kern_ipc_perm *ipcp;
1594 down_write(&sem_ids(ns).rwsem);
1597 ipcp = ipcctl_obtain_check(ns, &sem_ids(ns), semid, cmd,
1598 &semid64->sem_perm, 0);
1600 err = PTR_ERR(ipcp);
1604 sma = container_of(ipcp, struct sem_array, sem_perm);
1606 err = security_sem_semctl(&sma->sem_perm, cmd);
1612 sem_lock(sma, NULL, -1);
1613 /* freeary unlocks the ipc object and rcu */
1617 sem_lock(sma, NULL, -1);
1618 err = ipc_update_perm(&semid64->sem_perm, ipcp);
1621 sma->sem_ctime = ktime_get_real_seconds();
1629 sem_unlock(sma, -1);
1633 up_write(&sem_ids(ns).rwsem);
1637 long ksys_semctl(int semid, int semnum, int cmd, unsigned long arg)
1640 struct ipc_namespace *ns;
1641 void __user *p = (void __user *)arg;
1642 struct semid64_ds semid64;
1648 version = ipc_parse_version(&cmd);
1649 ns = current->nsproxy->ipc_ns;
1654 return semctl_info(ns, semid, cmd, p);
1658 err = semctl_stat(ns, semid, cmd, &semid64);
1661 if (copy_semid_to_user(p, &semid64, version))
1670 return semctl_main(ns, semid, semnum, cmd, p);
1673 #if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN)
1674 /* big-endian 64bit */
1677 /* 32bit or little-endian 64bit */
1680 return semctl_setval(ns, semid, semnum, val);
1683 if (copy_semid_from_user(&semid64, p, version))
1686 return semctl_down(ns, semid, cmd, &semid64);
1692 SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, unsigned long, arg)
1694 return ksys_semctl(semid, semnum, cmd, arg);
1697 #ifdef CONFIG_COMPAT
1699 struct compat_semid_ds {
1700 struct compat_ipc_perm sem_perm;
1701 compat_time_t sem_otime;
1702 compat_time_t sem_ctime;
1703 compat_uptr_t sem_base;
1704 compat_uptr_t sem_pending;
1705 compat_uptr_t sem_pending_last;
1707 unsigned short sem_nsems;
1710 static int copy_compat_semid_from_user(struct semid64_ds *out, void __user *buf,
1713 memset(out, 0, sizeof(*out));
1714 if (version == IPC_64) {
1715 struct compat_semid64_ds __user *p = buf;
1716 return get_compat_ipc64_perm(&out->sem_perm, &p->sem_perm);
1718 struct compat_semid_ds __user *p = buf;
1719 return get_compat_ipc_perm(&out->sem_perm, &p->sem_perm);
1723 static int copy_compat_semid_to_user(void __user *buf, struct semid64_ds *in,
1726 if (version == IPC_64) {
1727 struct compat_semid64_ds v;
1728 memset(&v, 0, sizeof(v));
1729 to_compat_ipc64_perm(&v.sem_perm, &in->sem_perm);
1730 v.sem_otime = lower_32_bits(in->sem_otime);
1731 v.sem_otime_high = upper_32_bits(in->sem_otime);
1732 v.sem_ctime = lower_32_bits(in->sem_ctime);
1733 v.sem_ctime_high = upper_32_bits(in->sem_ctime);
1734 v.sem_nsems = in->sem_nsems;
1735 return copy_to_user(buf, &v, sizeof(v));
1737 struct compat_semid_ds v;
1738 memset(&v, 0, sizeof(v));
1739 to_compat_ipc_perm(&v.sem_perm, &in->sem_perm);
1740 v.sem_otime = in->sem_otime;
1741 v.sem_ctime = in->sem_ctime;
1742 v.sem_nsems = in->sem_nsems;
1743 return copy_to_user(buf, &v, sizeof(v));
1747 long compat_ksys_semctl(int semid, int semnum, int cmd, int arg)
1749 void __user *p = compat_ptr(arg);
1750 struct ipc_namespace *ns;
1751 struct semid64_ds semid64;
1752 int version = compat_ipc_parse_version(&cmd);
1755 ns = current->nsproxy->ipc_ns;
1760 switch (cmd & (~IPC_64)) {
1763 return semctl_info(ns, semid, cmd, p);
1767 err = semctl_stat(ns, semid, cmd, &semid64);
1770 if (copy_compat_semid_to_user(p, &semid64, version))
1779 return semctl_main(ns, semid, semnum, cmd, p);
1781 return semctl_setval(ns, semid, semnum, arg);
1783 if (copy_compat_semid_from_user(&semid64, p, version))
1787 return semctl_down(ns, semid, cmd, &semid64);
1793 COMPAT_SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, int, arg)
1795 return compat_ksys_semctl(semid, semnum, cmd, arg);
1799 /* If the task doesn't already have a undo_list, then allocate one
1800 * here. We guarantee there is only one thread using this undo list,
1801 * and current is THE ONE
1803 * If this allocation and assignment succeeds, but later
1804 * portions of this code fail, there is no need to free the sem_undo_list.
1805 * Just let it stay associated with the task, and it'll be freed later
1808 * This can block, so callers must hold no locks.
1810 static inline int get_undo_list(struct sem_undo_list **undo_listp)
1812 struct sem_undo_list *undo_list;
1814 undo_list = current->sysvsem.undo_list;
1816 undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
1817 if (undo_list == NULL)
1819 spin_lock_init(&undo_list->lock);
1820 refcount_set(&undo_list->refcnt, 1);
1821 INIT_LIST_HEAD(&undo_list->list_proc);
1823 current->sysvsem.undo_list = undo_list;
1825 *undo_listp = undo_list;
1829 static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid)
1831 struct sem_undo *un;
1833 list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) {
1834 if (un->semid == semid)
1840 static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1842 struct sem_undo *un;
1844 assert_spin_locked(&ulp->lock);
1846 un = __lookup_undo(ulp, semid);
1848 list_del_rcu(&un->list_proc);
1849 list_add_rcu(&un->list_proc, &ulp->list_proc);
1855 * find_alloc_undo - lookup (and if not present create) undo array
1857 * @semid: semaphore array id
1859 * The function looks up (and if not present creates) the undo structure.
1860 * The size of the undo structure depends on the size of the semaphore
1861 * array, thus the alloc path is not that straightforward.
1862 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1863 * performs a rcu_read_lock().
1865 static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
1867 struct sem_array *sma;
1868 struct sem_undo_list *ulp;
1869 struct sem_undo *un, *new;
1872 error = get_undo_list(&ulp);
1874 return ERR_PTR(error);
1877 spin_lock(&ulp->lock);
1878 un = lookup_undo(ulp, semid);
1879 spin_unlock(&ulp->lock);
1880 if (likely(un != NULL))
1883 /* no undo structure around - allocate one. */
1884 /* step 1: figure out the size of the semaphore array */
1885 sma = sem_obtain_object_check(ns, semid);
1888 return ERR_CAST(sma);
1891 nsems = sma->sem_nsems;
1892 if (!ipc_rcu_getref(&sma->sem_perm)) {
1894 un = ERR_PTR(-EIDRM);
1899 /* step 2: allocate new undo structure */
1900 new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1902 ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
1903 return ERR_PTR(-ENOMEM);
1906 /* step 3: Acquire the lock on semaphore array */
1908 sem_lock_and_putref(sma);
1909 if (!ipc_valid_object(&sma->sem_perm)) {
1910 sem_unlock(sma, -1);
1913 un = ERR_PTR(-EIDRM);
1916 spin_lock(&ulp->lock);
1919 * step 4: check for races: did someone else allocate the undo struct?
1921 un = lookup_undo(ulp, semid);
1926 /* step 5: initialize & link new undo structure */
1927 new->semadj = (short *) &new[1];
1930 assert_spin_locked(&ulp->lock);
1931 list_add_rcu(&new->list_proc, &ulp->list_proc);
1932 ipc_assert_locked_object(&sma->sem_perm);
1933 list_add(&new->list_id, &sma->list_id);
1937 spin_unlock(&ulp->lock);
1938 sem_unlock(sma, -1);
1943 static long do_semtimedop(int semid, struct sembuf __user *tsops,
1944 unsigned nsops, const struct timespec64 *timeout)
1946 int error = -EINVAL;
1947 struct sem_array *sma;
1948 struct sembuf fast_sops[SEMOPM_FAST];
1949 struct sembuf *sops = fast_sops, *sop;
1950 struct sem_undo *un;
1952 bool undos = false, alter = false, dupsop = false;
1953 struct sem_queue queue;
1954 unsigned long dup = 0, jiffies_left = 0;
1955 struct ipc_namespace *ns;
1957 ns = current->nsproxy->ipc_ns;
1959 if (nsops < 1 || semid < 0)
1961 if (nsops > ns->sc_semopm)
1963 if (nsops > SEMOPM_FAST) {
1964 sops = kvmalloc_array(nsops, sizeof(*sops), GFP_KERNEL);
1969 if (copy_from_user(sops, tsops, nsops * sizeof(*tsops))) {
1975 if (timeout->tv_sec < 0 || timeout->tv_nsec < 0 ||
1976 timeout->tv_nsec >= 1000000000L) {
1980 jiffies_left = timespec64_to_jiffies(timeout);
1984 for (sop = sops; sop < sops + nsops; sop++) {
1985 unsigned long mask = 1ULL << ((sop->sem_num) % BITS_PER_LONG);
1987 if (sop->sem_num >= max)
1989 if (sop->sem_flg & SEM_UNDO)
1993 * There was a previous alter access that appears
1994 * to have accessed the same semaphore, thus use
1995 * the dupsop logic. "appears", because the detection
1996 * can only check % BITS_PER_LONG.
2000 if (sop->sem_op != 0) {
2007 /* On success, find_alloc_undo takes the rcu_read_lock */
2008 un = find_alloc_undo(ns, semid);
2010 error = PTR_ERR(un);
2018 sma = sem_obtain_object_check(ns, semid);
2021 error = PTR_ERR(sma);
2026 if (max >= sma->sem_nsems) {
2032 if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO)) {
2037 error = security_sem_semop(&sma->sem_perm, sops, nsops, alter);
2044 locknum = sem_lock(sma, sops, nsops);
2046 * We eventually might perform the following check in a lockless
2047 * fashion, considering ipc_valid_object() locking constraints.
2048 * If nsops == 1 and there is no contention for sem_perm.lock, then
2049 * only a per-semaphore lock is held and it's OK to proceed with the
2050 * check below. More details on the fine grained locking scheme
2051 * entangled here and why it's RMID race safe on comments at sem_lock()
2053 if (!ipc_valid_object(&sma->sem_perm))
2054 goto out_unlock_free;
2056 * semid identifiers are not unique - find_alloc_undo may have
2057 * allocated an undo structure, it was invalidated by an RMID
2058 * and now a new array with received the same id. Check and fail.
2059 * This case can be detected checking un->semid. The existence of
2060 * "un" itself is guaranteed by rcu.
2062 if (un && un->semid == -1)
2063 goto out_unlock_free;
2066 queue.nsops = nsops;
2068 queue.pid = task_tgid(current);
2069 queue.alter = alter;
2070 queue.dupsop = dupsop;
2072 error = perform_atomic_semop(sma, &queue);
2073 if (error == 0) { /* non-blocking succesfull path */
2074 DEFINE_WAKE_Q(wake_q);
2077 * If the operation was successful, then do
2078 * the required updates.
2081 do_smart_update(sma, sops, nsops, 1, &wake_q);
2083 set_semotime(sma, sops);
2085 sem_unlock(sma, locknum);
2091 if (error < 0) /* non-blocking error path */
2092 goto out_unlock_free;
2095 * We need to sleep on this operation, so we put the current
2096 * task into the pending queue and go to sleep.
2100 int idx = array_index_nospec(sops->sem_num, sma->sem_nsems);
2101 curr = &sma->sems[idx];
2104 if (sma->complex_count) {
2105 list_add_tail(&queue.list,
2106 &sma->pending_alter);
2109 list_add_tail(&queue.list,
2110 &curr->pending_alter);
2113 list_add_tail(&queue.list, &curr->pending_const);
2116 if (!sma->complex_count)
2120 list_add_tail(&queue.list, &sma->pending_alter);
2122 list_add_tail(&queue.list, &sma->pending_const);
2124 sma->complex_count++;
2128 WRITE_ONCE(queue.status, -EINTR);
2129 queue.sleeper = current;
2131 __set_current_state(TASK_INTERRUPTIBLE);
2132 sem_unlock(sma, locknum);
2136 jiffies_left = schedule_timeout(jiffies_left);
2141 * fastpath: the semop has completed, either successfully or
2142 * not, from the syscall pov, is quite irrelevant to us at this
2143 * point; we're done.
2145 * We _do_ care, nonetheless, about being awoken by a signal or
2146 * spuriously. The queue.status is checked again in the
2147 * slowpath (aka after taking sem_lock), such that we can detect
2148 * scenarios where we were awakened externally, during the
2149 * window between wake_q_add() and wake_up_q().
2151 error = READ_ONCE(queue.status);
2152 if (error != -EINTR) {
2154 * User space could assume that semop() is a memory
2155 * barrier: Without the mb(), the cpu could
2156 * speculatively read in userspace stale data that was
2157 * overwritten by the previous owner of the semaphore.
2164 locknum = sem_lock(sma, sops, nsops);
2166 if (!ipc_valid_object(&sma->sem_perm))
2167 goto out_unlock_free;
2169 error = READ_ONCE(queue.status);
2172 * If queue.status != -EINTR we are woken up by another process.
2173 * Leave without unlink_queue(), but with sem_unlock().
2175 if (error != -EINTR)
2176 goto out_unlock_free;
2179 * If an interrupt occurred we have to clean up the queue.
2181 if (timeout && jiffies_left == 0)
2183 } while (error == -EINTR && !signal_pending(current)); /* spurious */
2185 unlink_queue(sma, &queue);
2188 sem_unlock(sma, locknum);
2191 if (sops != fast_sops)
2196 long ksys_semtimedop(int semid, struct sembuf __user *tsops,
2197 unsigned int nsops, const struct __kernel_timespec __user *timeout)
2200 struct timespec64 ts;
2201 if (get_timespec64(&ts, timeout))
2203 return do_semtimedop(semid, tsops, nsops, &ts);
2205 return do_semtimedop(semid, tsops, nsops, NULL);
2208 SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
2209 unsigned int, nsops, const struct __kernel_timespec __user *, timeout)
2211 return ksys_semtimedop(semid, tsops, nsops, timeout);
2214 #ifdef CONFIG_COMPAT_32BIT_TIME
2215 long compat_ksys_semtimedop(int semid, struct sembuf __user *tsems,
2217 const struct compat_timespec __user *timeout)
2220 struct timespec64 ts;
2221 if (compat_get_timespec64(&ts, timeout))
2223 return do_semtimedop(semid, tsems, nsops, &ts);
2225 return do_semtimedop(semid, tsems, nsops, NULL);
2228 COMPAT_SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsems,
2229 unsigned int, nsops,
2230 const struct compat_timespec __user *, timeout)
2232 return compat_ksys_semtimedop(semid, tsems, nsops, timeout);
2236 SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
2239 return do_semtimedop(semid, tsops, nsops, NULL);
2242 /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
2243 * parent and child tasks.
2246 int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
2248 struct sem_undo_list *undo_list;
2251 if (clone_flags & CLONE_SYSVSEM) {
2252 error = get_undo_list(&undo_list);
2255 refcount_inc(&undo_list->refcnt);
2256 tsk->sysvsem.undo_list = undo_list;
2258 tsk->sysvsem.undo_list = NULL;
2264 * add semadj values to semaphores, free undo structures.
2265 * undo structures are not freed when semaphore arrays are destroyed
2266 * so some of them may be out of date.
2267 * IMPLEMENTATION NOTE: There is some confusion over whether the
2268 * set of adjustments that needs to be done should be done in an atomic
2269 * manner or not. That is, if we are attempting to decrement the semval
2270 * should we queue up and wait until we can do so legally?
2271 * The original implementation attempted to do this (queue and wait).
2272 * The current implementation does not do so. The POSIX standard
2273 * and SVID should be consulted to determine what behavior is mandated.
2275 void exit_sem(struct task_struct *tsk)
2277 struct sem_undo_list *ulp;
2279 ulp = tsk->sysvsem.undo_list;
2282 tsk->sysvsem.undo_list = NULL;
2284 if (!refcount_dec_and_test(&ulp->refcnt))
2288 struct sem_array *sma;
2289 struct sem_undo *un;
2291 DEFINE_WAKE_Q(wake_q);
2296 un = list_entry_rcu(ulp->list_proc.next,
2297 struct sem_undo, list_proc);
2298 if (&un->list_proc == &ulp->list_proc) {
2300 * We must wait for freeary() before freeing this ulp,
2301 * in case we raced with last sem_undo. There is a small
2302 * possibility where we exit while freeary() didn't
2303 * finish unlocking sem_undo_list.
2305 spin_lock(&ulp->lock);
2306 spin_unlock(&ulp->lock);
2310 spin_lock(&ulp->lock);
2312 spin_unlock(&ulp->lock);
2314 /* exit_sem raced with IPC_RMID, nothing to do */
2320 sma = sem_obtain_object_check(tsk->nsproxy->ipc_ns, semid);
2321 /* exit_sem raced with IPC_RMID, nothing to do */
2327 sem_lock(sma, NULL, -1);
2328 /* exit_sem raced with IPC_RMID, nothing to do */
2329 if (!ipc_valid_object(&sma->sem_perm)) {
2330 sem_unlock(sma, -1);
2334 un = __lookup_undo(ulp, semid);
2336 /* exit_sem raced with IPC_RMID+semget() that created
2337 * exactly the same semid. Nothing to do.
2339 sem_unlock(sma, -1);
2344 /* remove un from the linked lists */
2345 ipc_assert_locked_object(&sma->sem_perm);
2346 list_del(&un->list_id);
2348 spin_lock(&ulp->lock);
2349 list_del_rcu(&un->list_proc);
2350 spin_unlock(&ulp->lock);
2352 /* perform adjustments registered in un */
2353 for (i = 0; i < sma->sem_nsems; i++) {
2354 struct sem *semaphore = &sma->sems[i];
2355 if (un->semadj[i]) {
2356 semaphore->semval += un->semadj[i];
2358 * Range checks of the new semaphore value,
2359 * not defined by sus:
2360 * - Some unices ignore the undo entirely
2361 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
2362 * - some cap the value (e.g. FreeBSD caps
2363 * at 0, but doesn't enforce SEMVMX)
2365 * Linux caps the semaphore value, both at 0
2368 * Manfred <manfred@colorfullife.com>
2370 if (semaphore->semval < 0)
2371 semaphore->semval = 0;
2372 if (semaphore->semval > SEMVMX)
2373 semaphore->semval = SEMVMX;
2374 ipc_update_pid(&semaphore->sempid, task_tgid(current));
2377 /* maybe some queued-up processes were waiting for this */
2378 do_smart_update(sma, NULL, 0, 1, &wake_q);
2379 sem_unlock(sma, -1);
2388 #ifdef CONFIG_PROC_FS
2389 static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
2391 struct user_namespace *user_ns = seq_user_ns(s);
2392 struct kern_ipc_perm *ipcp = it;
2393 struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
2397 * The proc interface isn't aware of sem_lock(), it calls
2398 * ipc_lock_object() directly (in sysvipc_find_ipc).
2399 * In order to stay compatible with sem_lock(), we must
2400 * enter / leave complex_mode.
2402 complexmode_enter(sma);
2404 sem_otime = get_semotime(sma);
2407 "%10d %10d %4o %10u %5u %5u %5u %5u %10llu %10llu\n",
2412 from_kuid_munged(user_ns, sma->sem_perm.uid),
2413 from_kgid_munged(user_ns, sma->sem_perm.gid),
2414 from_kuid_munged(user_ns, sma->sem_perm.cuid),
2415 from_kgid_munged(user_ns, sma->sem_perm.cgid),
2419 complexmode_tryleave(sma);