~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/ipc/sem.c

Version: ~ [ linux-5.14-rc3 ] ~ [ linux-5.13.5 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.53 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.135 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.198 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.240 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.276 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.276 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

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

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp