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

TOMOYO Linux Cross Reference
Linux/ipc/sem.c

Version: ~ [ linux-5.5 ] ~ [ linux-5.4.15 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.98 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.167 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.211 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.211 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.81 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ 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_semncnt() and
 51  *   count_semzcnt()
 52  * - the task that performs a successful semop() scans the list of all
 53  *   sleeping tasks and completes any pending operations that can be fulfilled.
 54  *   Semaphores are actively given to waiting tasks (necessary for FIFO).
 55  *   (see update_queue())
 56  * - To improve the scalability, the actual wake-up calls are performed after
 57  *   dropping all locks. (see wake_up_sem_queue_prepare(),
 58  *   wake_up_sem_queue_do())
 59  * - All work is done by the waker, the woken up task does not have to do
 60  *   anything - not even acquiring a lock or dropping a refcount.
 61  * - A woken up task may not even touch the semaphore array anymore, it may
 62  *   have been destroyed already by a semctl(RMID).
 63  * - The synchronizations between wake-ups due to a timeout/signal and a
 64  *   wake-up due to a completed semaphore operation is achieved by using an
 65  *   intermediate state (IN_WAKEUP).
 66  * - UNDO values are stored in an array (one per process and per
 67  *   semaphore array, lazily allocated). For backwards compatibility, multiple
 68  *   modes for the UNDO variables are supported (per process, per thread)
 69  *   (see copy_semundo, CLONE_SYSVSEM)
 70  * - There are two lists of the pending operations: a per-array list
 71  *   and per-semaphore list (stored in the array). This allows to achieve FIFO
 72  *   ordering without always scanning all pending operations.
 73  *   The worst-case behavior is nevertheless O(N^2) for N wakeups.
 74  */
 75 
 76 #include <linux/slab.h>
 77 #include <linux/spinlock.h>
 78 #include <linux/init.h>
 79 #include <linux/proc_fs.h>
 80 #include <linux/time.h>
 81 #include <linux/security.h>
 82 #include <linux/syscalls.h>
 83 #include <linux/audit.h>
 84 #include <linux/capability.h>
 85 #include <linux/seq_file.h>
 86 #include <linux/rwsem.h>
 87 #include <linux/nsproxy.h>
 88 #include <linux/ipc_namespace.h>
 89 
 90 #include <asm/uaccess.h>
 91 #include "util.h"
 92 
 93 /* One semaphore structure for each semaphore in the system. */
 94 struct sem {
 95         int     semval;         /* current value */
 96         int     sempid;         /* pid of last operation */
 97         spinlock_t      lock;   /* spinlock for fine-grained semtimedop */
 98         struct list_head pending_alter; /* pending single-sop operations */
 99                                         /* that alter the semaphore */
100         struct list_head pending_const; /* pending single-sop operations */
101                                         /* that do not alter the semaphore*/
102         time_t  sem_otime;      /* candidate for sem_otime */
103 } ____cacheline_aligned_in_smp;
104 
105 /* One queue for each sleeping process in the system. */
106 struct sem_queue {
107         struct list_head        list;    /* queue of pending operations */
108         struct task_struct      *sleeper; /* this process */
109         struct sem_undo         *undo;   /* undo structure */
110         int                     pid;     /* process id of requesting process */
111         int                     status;  /* completion status of operation */
112         struct sembuf           *sops;   /* array of pending operations */
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  */
271 static void sem_wait_array(struct sem_array *sma)
272 {
273         int i;
274         struct sem *sem;
275 
276         for (i = 0; i < sma->sem_nsems; i++) {
277                 sem = sma->sem_base + i;
278                 spin_unlock_wait(&sem->lock);
279         }
280         ipc_smp_acquire__after_spin_is_unlocked();
281 }
282 
283 /*
284  * If the request contains only one semaphore operation, and there are
285  * no complex transactions pending, lock only the semaphore involved.
286  * Otherwise, lock the entire semaphore array, since we either have
287  * multiple semaphores in our own semops, or we need to look at
288  * semaphores from other pending complex operations.
289  */
290 static inline int sem_lock(struct sem_array *sma, struct sembuf *sops,
291                               int nsops)
292 {
293         struct sem *sem;
294 
295         if (nsops != 1) {
296                 /* Complex operation - acquire a full lock */
297                 ipc_lock_object(&sma->sem_perm);
298 
299                 /* And wait until all simple ops that are processed
300                  * right now have dropped their locks.
301                  */
302                 sem_wait_array(sma);
303                 return -1;
304         }
305 
306         /*
307          * Only one semaphore affected - try to optimize locking.
308          * The rules are:
309          * - optimized locking is possible if no complex operation
310          *   is either enqueued or processed right now.
311          * - The test for enqueued complex ops is simple:
312          *      sma->complex_count != 0
313          * - Testing for complex ops that are processed right now is
314          *   a bit more difficult. Complex ops acquire the full lock
315          *   and first wait that the running simple ops have completed.
316          *   (see above)
317          *   Thus: If we own a simple lock and the global lock is free
318          *      and complex_count is now 0, then it will stay 0 and
319          *      thus just locking sem->lock is sufficient.
320          */
321         sem = sma->sem_base + sops->sem_num;
322 
323         if (sma->complex_count == 0) {
324                 /*
325                  * It appears that no complex operation is around.
326                  * Acquire the per-semaphore lock.
327                  */
328                 spin_lock(&sem->lock);
329 
330                 /* Then check that the global lock is free */
331                 if (!spin_is_locked(&sma->sem_perm.lock)) {
332                         /*
333                          * We need a memory barrier with acquire semantics,
334                          * otherwise we can race with another thread that does:
335                          *      complex_count++;
336                          *      spin_unlock(sem_perm.lock);
337                          */
338                         ipc_smp_acquire__after_spin_is_unlocked();
339 
340                         /* Now repeat the test of complex_count:
341                          * It can't change anymore until we drop sem->lock.
342                          * Thus: if is now 0, then it will stay 0.
343                          */
344                         if (sma->complex_count == 0) {
345                                 /* fast path successful! */
346                                 return sops->sem_num;
347                         }
348                 }
349                 spin_unlock(&sem->lock);
350         }
351 
352         /* slow path: acquire the full lock */
353         ipc_lock_object(&sma->sem_perm);
354 
355         if (sma->complex_count == 0) {
356                 /* False alarm:
357                  * There is no complex operation, thus we can switch
358                  * back to the fast path.
359                  */
360                 spin_lock(&sem->lock);
361                 ipc_unlock_object(&sma->sem_perm);
362                 return sops->sem_num;
363         } else {
364                 /* Not a false alarm, thus complete the sequence for a
365                  * full lock.
366                  */
367                 sem_wait_array(sma);
368                 return -1;
369         }
370 }
371 
372 static inline void sem_unlock(struct sem_array *sma, int locknum)
373 {
374         if (locknum == -1) {
375                 unmerge_queues(sma);
376                 ipc_unlock_object(&sma->sem_perm);
377         } else {
378                 struct sem *sem = sma->sem_base + locknum;
379                 spin_unlock(&sem->lock);
380         }
381 }
382 
383 /*
384  * sem_lock_(check_) routines are called in the paths where the rwsem
385  * is not held.
386  *
387  * The caller holds the RCU read lock.
388  */
389 static inline struct sem_array *sem_obtain_lock(struct ipc_namespace *ns,
390                         int id, struct sembuf *sops, int nsops, int *locknum)
391 {
392         struct kern_ipc_perm *ipcp;
393         struct sem_array *sma;
394 
395         ipcp = ipc_obtain_object(&sem_ids(ns), id);
396         if (IS_ERR(ipcp))
397                 return ERR_CAST(ipcp);
398 
399         sma = container_of(ipcp, struct sem_array, sem_perm);
400         *locknum = sem_lock(sma, sops, nsops);
401 
402         /* ipc_rmid() may have already freed the ID while sem_lock
403          * was spinning: verify that the structure is still valid
404          */
405         if (!ipcp->deleted)
406                 return container_of(ipcp, struct sem_array, sem_perm);
407 
408         sem_unlock(sma, *locknum);
409         return ERR_PTR(-EINVAL);
410 }
411 
412 static inline struct sem_array *sem_obtain_object(struct ipc_namespace *ns, int id)
413 {
414         struct kern_ipc_perm *ipcp = ipc_obtain_object(&sem_ids(ns), id);
415 
416         if (IS_ERR(ipcp))
417                 return ERR_CAST(ipcp);
418 
419         return container_of(ipcp, struct sem_array, sem_perm);
420 }
421 
422 static inline struct sem_array *sem_obtain_object_check(struct ipc_namespace *ns,
423                                                         int id)
424 {
425         struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&sem_ids(ns), id);
426 
427         if (IS_ERR(ipcp))
428                 return ERR_CAST(ipcp);
429 
430         return container_of(ipcp, struct sem_array, sem_perm);
431 }
432 
433 static inline void sem_lock_and_putref(struct sem_array *sma)
434 {
435         sem_lock(sma, NULL, -1);
436         ipc_rcu_putref(sma, ipc_rcu_free);
437 }
438 
439 static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
440 {
441         ipc_rmid(&sem_ids(ns), &s->sem_perm);
442 }
443 
444 /*
445  * Lockless wakeup algorithm:
446  * Without the check/retry algorithm a lockless wakeup is possible:
447  * - queue.status is initialized to -EINTR before blocking.
448  * - wakeup is performed by
449  *      * unlinking the queue entry from the pending list
450  *      * setting queue.status to IN_WAKEUP
451  *        This is the notification for the blocked thread that a
452  *        result value is imminent.
453  *      * call wake_up_process
454  *      * set queue.status to the final value.
455  * - the previously blocked thread checks queue.status:
456  *      * if it's IN_WAKEUP, then it must wait until the value changes
457  *      * if it's not -EINTR, then the operation was completed by
458  *        update_queue. semtimedop can return queue.status without
459  *        performing any operation on the sem array.
460  *      * otherwise it must acquire the spinlock and check what's up.
461  *
462  * The two-stage algorithm is necessary to protect against the following
463  * races:
464  * - if queue.status is set after wake_up_process, then the woken up idle
465  *   thread could race forward and try (and fail) to acquire sma->lock
466  *   before update_queue had a chance to set queue.status
467  * - if queue.status is written before wake_up_process and if the
468  *   blocked process is woken up by a signal between writing
469  *   queue.status and the wake_up_process, then the woken up
470  *   process could return from semtimedop and die by calling
471  *   sys_exit before wake_up_process is called. Then wake_up_process
472  *   will oops, because the task structure is already invalid.
473  *   (yes, this happened on s390 with sysv msg).
474  *
475  */
476 #define IN_WAKEUP       1
477 
478 /**
479  * newary - Create a new semaphore set
480  * @ns: namespace
481  * @params: ptr to the structure that contains key, semflg and nsems
482  *
483  * Called with sem_ids.rwsem held (as a writer)
484  */
485 
486 static int newary(struct ipc_namespace *ns, struct ipc_params *params)
487 {
488         int id;
489         int retval;
490         struct sem_array *sma;
491         int size;
492         key_t key = params->key;
493         int nsems = params->u.nsems;
494         int semflg = params->flg;
495         int i;
496 
497         if (!nsems)
498                 return -EINVAL;
499         if (ns->used_sems + nsems > ns->sc_semmns)
500                 return -ENOSPC;
501 
502         size = sizeof (*sma) + nsems * sizeof (struct sem);
503         sma = ipc_rcu_alloc(size);
504         if (!sma) {
505                 return -ENOMEM;
506         }
507         memset (sma, 0, size);
508 
509         sma->sem_perm.mode = (semflg & S_IRWXUGO);
510         sma->sem_perm.key = key;
511 
512         sma->sem_perm.security = NULL;
513         retval = security_sem_alloc(sma);
514         if (retval) {
515                 ipc_rcu_putref(sma, ipc_rcu_free);
516                 return retval;
517         }
518 
519         id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
520         if (id < 0) {
521                 ipc_rcu_putref(sma, sem_rcu_free);
522                 return id;
523         }
524         ns->used_sems += nsems;
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         sem_unlock(sma, -1);
541         rcu_read_unlock();
542 
543         return sma->sem_perm.id;
544 }
545 
546 
547 /*
548  * Called with sem_ids.rwsem and ipcp locked.
549  */
550 static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg)
551 {
552         struct sem_array *sma;
553 
554         sma = container_of(ipcp, struct sem_array, sem_perm);
555         return security_sem_associate(sma, semflg);
556 }
557 
558 /*
559  * Called with sem_ids.rwsem and ipcp locked.
560  */
561 static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
562                                 struct ipc_params *params)
563 {
564         struct sem_array *sma;
565 
566         sma = container_of(ipcp, struct sem_array, sem_perm);
567         if (params->u.nsems > sma->sem_nsems)
568                 return -EINVAL;
569 
570         return 0;
571 }
572 
573 SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg)
574 {
575         struct ipc_namespace *ns;
576         struct ipc_ops sem_ops;
577         struct ipc_params sem_params;
578 
579         ns = current->nsproxy->ipc_ns;
580 
581         if (nsems < 0 || nsems > ns->sc_semmsl)
582                 return -EINVAL;
583 
584         sem_ops.getnew = newary;
585         sem_ops.associate = sem_security;
586         sem_ops.more_checks = sem_more_checks;
587 
588         sem_params.key = key;
589         sem_params.flg = semflg;
590         sem_params.u.nsems = nsems;
591 
592         return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
593 }
594 
595 /** perform_atomic_semop - Perform (if possible) a semaphore operation
596  * @sma: semaphore array
597  * @sops: array with operations that should be checked
598  * @nsems: number of sops
599  * @un: undo array
600  * @pid: pid that did the change
601  *
602  * Returns 0 if the operation was possible.
603  * Returns 1 if the operation is impossible, the caller must sleep.
604  * Negative values are error codes.
605  */
606 
607 static int perform_atomic_semop(struct sem_array *sma, struct sembuf *sops,
608                              int nsops, struct sem_undo *un, int pid)
609 {
610         int result, sem_op;
611         struct sembuf *sop;
612         struct sem * curr;
613 
614         for (sop = sops; sop < sops + nsops; sop++) {
615                 curr = sma->sem_base + sop->sem_num;
616                 sem_op = sop->sem_op;
617                 result = curr->semval;
618   
619                 if (!sem_op && result)
620                         goto would_block;
621 
622                 result += sem_op;
623                 if (result < 0)
624                         goto would_block;
625                 if (result > SEMVMX)
626                         goto out_of_range;
627                 if (sop->sem_flg & SEM_UNDO) {
628                         int undo = un->semadj[sop->sem_num] - sem_op;
629                         /*
630                          *      Exceeding the undo range is an error.
631                          */
632                         if (undo < (-SEMAEM - 1) || undo > SEMAEM)
633                                 goto out_of_range;
634                 }
635                 curr->semval = result;
636         }
637 
638         sop--;
639         while (sop >= sops) {
640                 sma->sem_base[sop->sem_num].sempid = pid;
641                 if (sop->sem_flg & SEM_UNDO)
642                         un->semadj[sop->sem_num] -= sop->sem_op;
643                 sop--;
644         }
645         
646         return 0;
647 
648 out_of_range:
649         result = -ERANGE;
650         goto undo;
651 
652 would_block:
653         if (sop->sem_flg & IPC_NOWAIT)
654                 result = -EAGAIN;
655         else
656                 result = 1;
657 
658 undo:
659         sop--;
660         while (sop >= sops) {
661                 sma->sem_base[sop->sem_num].semval -= sop->sem_op;
662                 sop--;
663         }
664 
665         return result;
666 }
667 
668 /** wake_up_sem_queue_prepare(q, error): Prepare wake-up
669  * @q: queue entry that must be signaled
670  * @error: Error value for the signal
671  *
672  * Prepare the wake-up of the queue entry q.
673  */
674 static void wake_up_sem_queue_prepare(struct list_head *pt,
675                                 struct sem_queue *q, int error)
676 {
677         if (list_empty(pt)) {
678                 /*
679                  * Hold preempt off so that we don't get preempted and have the
680                  * wakee busy-wait until we're scheduled back on.
681                  */
682                 preempt_disable();
683         }
684         q->status = IN_WAKEUP;
685         q->pid = error;
686 
687         list_add_tail(&q->list, pt);
688 }
689 
690 /**
691  * wake_up_sem_queue_do(pt) - do the actual wake-up
692  * @pt: list of tasks to be woken up
693  *
694  * Do the actual wake-up.
695  * The function is called without any locks held, thus the semaphore array
696  * could be destroyed already and the tasks can disappear as soon as the
697  * status is set to the actual return code.
698  */
699 static void wake_up_sem_queue_do(struct list_head *pt)
700 {
701         struct sem_queue *q, *t;
702         int did_something;
703 
704         did_something = !list_empty(pt);
705         list_for_each_entry_safe(q, t, pt, list) {
706                 wake_up_process(q->sleeper);
707                 /* q can disappear immediately after writing q->status. */
708                 smp_wmb();
709                 q->status = q->pid;
710         }
711         if (did_something)
712                 preempt_enable();
713 }
714 
715 static void unlink_queue(struct sem_array *sma, struct sem_queue *q)
716 {
717         list_del(&q->list);
718         if (q->nsops > 1)
719                 sma->complex_count--;
720 }
721 
722 /** check_restart(sma, q)
723  * @sma: semaphore array
724  * @q: the operation that just completed
725  *
726  * update_queue is O(N^2) when it restarts scanning the whole queue of
727  * waiting operations. Therefore this function checks if the restart is
728  * really necessary. It is called after a previously waiting operation
729  * modified the array.
730  * Note that wait-for-zero operations are handled without restart.
731  */
732 static int check_restart(struct sem_array *sma, struct sem_queue *q)
733 {
734         /* pending complex alter operations are too difficult to analyse */
735         if (!list_empty(&sma->pending_alter))
736                 return 1;
737 
738         /* we were a sleeping complex operation. Too difficult */
739         if (q->nsops > 1)
740                 return 1;
741 
742         /* It is impossible that someone waits for the new value:
743          * - complex operations always restart.
744          * - wait-for-zero are handled seperately.
745          * - q is a previously sleeping simple operation that
746          *   altered the array. It must be a decrement, because
747          *   simple increments never sleep.
748          * - If there are older (higher priority) decrements
749          *   in the queue, then they have observed the original
750          *   semval value and couldn't proceed. The operation
751          *   decremented to value - thus they won't proceed either.
752          */
753         return 0;
754 }
755 
756 /**
757  * wake_const_ops(sma, semnum, pt) - Wake up non-alter tasks
758  * @sma: semaphore array.
759  * @semnum: semaphore that was modified.
760  * @pt: list head for the tasks that must be woken up.
761  *
762  * wake_const_ops must be called after a semaphore in a semaphore array
763  * was set to 0. If complex const operations are pending, wake_const_ops must
764  * be called with semnum = -1, as well as with the number of each modified
765  * semaphore.
766  * The tasks that must be woken up are added to @pt. The return code
767  * is stored in q->pid.
768  * The function returns 1 if at least one operation was completed successfully.
769  */
770 static int wake_const_ops(struct sem_array *sma, int semnum,
771                                 struct list_head *pt)
772 {
773         struct sem_queue *q;
774         struct list_head *walk;
775         struct list_head *pending_list;
776         int semop_completed = 0;
777 
778         if (semnum == -1)
779                 pending_list = &sma->pending_const;
780         else
781                 pending_list = &sma->sem_base[semnum].pending_const;
782 
783         walk = pending_list->next;
784         while (walk != pending_list) {
785                 int error;
786 
787                 q = container_of(walk, struct sem_queue, list);
788                 walk = walk->next;
789 
790                 error = perform_atomic_semop(sma, q->sops, q->nsops,
791                                                  q->undo, q->pid);
792 
793                 if (error <= 0) {
794                         /* operation completed, remove from queue & wakeup */
795 
796                         unlink_queue(sma, q);
797 
798                         wake_up_sem_queue_prepare(pt, q, error);
799                         if (error == 0)
800                                 semop_completed = 1;
801                 }
802         }
803         return semop_completed;
804 }
805 
806 /**
807  * do_smart_wakeup_zero(sma, sops, nsops, pt) - wakeup all wait for zero tasks
808  * @sma: semaphore array
809  * @sops: operations that were performed
810  * @nsops: number of operations
811  * @pt: list head of the tasks that must be woken up.
812  *
813  * do_smart_wakeup_zero() checks all required queue for wait-for-zero
814  * operations, based on the actual changes that were performed on the
815  * semaphore array.
816  * The function returns 1 if at least one operation was completed successfully.
817  */
818 static int do_smart_wakeup_zero(struct sem_array *sma, struct sembuf *sops,
819                                         int nsops, struct list_head *pt)
820 {
821         int i;
822         int semop_completed = 0;
823         int got_zero = 0;
824 
825         /* first: the per-semaphore queues, if known */
826         if (sops) {
827                 for (i = 0; i < nsops; i++) {
828                         int num = sops[i].sem_num;
829 
830                         if (sma->sem_base[num].semval == 0) {
831                                 got_zero = 1;
832                                 semop_completed |= wake_const_ops(sma, num, pt);
833                         }
834                 }
835         } else {
836                 /*
837                  * No sops means modified semaphores not known.
838                  * Assume all were changed.
839                  */
840                 for (i = 0; i < sma->sem_nsems; i++) {
841                         if (sma->sem_base[i].semval == 0) {
842                                 got_zero = 1;
843                                 semop_completed |= wake_const_ops(sma, i, pt);
844                         }
845                 }
846         }
847         /*
848          * If one of the modified semaphores got 0,
849          * then check the global queue, too.
850          */
851         if (got_zero)
852                 semop_completed |= wake_const_ops(sma, -1, pt);
853 
854         return semop_completed;
855 }
856 
857 
858 /**
859  * update_queue(sma, semnum): Look for tasks that can be completed.
860  * @sma: semaphore array.
861  * @semnum: semaphore that was modified.
862  * @pt: list head for the tasks that must be woken up.
863  *
864  * update_queue must be called after a semaphore in a semaphore array
865  * was modified. If multiple semaphores were modified, update_queue must
866  * be called with semnum = -1, as well as with the number of each modified
867  * semaphore.
868  * The tasks that must be woken up are added to @pt. The return code
869  * is stored in q->pid.
870  * The function internally checks if const operations can now succeed.
871  *
872  * The function return 1 if at least one semop was completed successfully.
873  */
874 static int update_queue(struct sem_array *sma, int semnum, struct list_head *pt)
875 {
876         struct sem_queue *q;
877         struct list_head *walk;
878         struct list_head *pending_list;
879         int semop_completed = 0;
880 
881         if (semnum == -1)
882                 pending_list = &sma->pending_alter;
883         else
884                 pending_list = &sma->sem_base[semnum].pending_alter;
885 
886 again:
887         walk = pending_list->next;
888         while (walk != pending_list) {
889                 int error, restart;
890 
891                 q = container_of(walk, struct sem_queue, list);
892                 walk = walk->next;
893 
894                 /* If we are scanning the single sop, per-semaphore list of
895                  * one semaphore and that semaphore is 0, then it is not
896                  * necessary to scan further: simple increments
897                  * that affect only one entry succeed immediately and cannot
898                  * be in the  per semaphore pending queue, and decrements
899                  * cannot be successful if the value is already 0.
900                  */
901                 if (semnum != -1 && sma->sem_base[semnum].semval == 0)
902                         break;
903 
904                 error = perform_atomic_semop(sma, q->sops, q->nsops,
905                                          q->undo, q->pid);
906 
907                 /* Does q->sleeper still need to sleep? */
908                 if (error > 0)
909                         continue;
910 
911                 unlink_queue(sma, q);
912 
913                 if (error) {
914                         restart = 0;
915                 } else {
916                         semop_completed = 1;
917                         do_smart_wakeup_zero(sma, q->sops, q->nsops, pt);
918                         restart = check_restart(sma, q);
919                 }
920 
921                 wake_up_sem_queue_prepare(pt, q, error);
922                 if (restart)
923                         goto again;
924         }
925         return semop_completed;
926 }
927 
928 /**
929  * set_semotime(sma, sops) - set sem_otime
930  * @sma: semaphore array
931  * @sops: operations that modified the array, may be NULL
932  *
933  * sem_otime is replicated to avoid cache line trashing.
934  * This function sets one instance to the current time.
935  */
936 static void set_semotime(struct sem_array *sma, struct sembuf *sops)
937 {
938         if (sops == NULL) {
939                 sma->sem_base[0].sem_otime = get_seconds();
940         } else {
941                 sma->sem_base[sops[0].sem_num].sem_otime =
942                                                         get_seconds();
943         }
944 }
945 
946 /**
947  * do_smart_update(sma, sops, nsops, otime, pt) - optimized update_queue
948  * @sma: semaphore array
949  * @sops: operations that were performed
950  * @nsops: number of operations
951  * @otime: force setting otime
952  * @pt: list head of the tasks that must be woken up.
953  *
954  * do_smart_update() does the required calls to update_queue and wakeup_zero,
955  * based on the actual changes that were performed on the semaphore array.
956  * Note that the function does not do the actual wake-up: the caller is
957  * responsible for calling wake_up_sem_queue_do(@pt).
958  * It is safe to perform this call after dropping all locks.
959  */
960 static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops,
961                         int otime, struct list_head *pt)
962 {
963         int i;
964 
965         otime |= do_smart_wakeup_zero(sma, sops, nsops, pt);
966 
967         if (!list_empty(&sma->pending_alter)) {
968                 /* semaphore array uses the global queue - just process it. */
969                 otime |= update_queue(sma, -1, pt);
970         } else {
971                 if (!sops) {
972                         /*
973                          * No sops, thus the modified semaphores are not
974                          * known. Check all.
975                          */
976                         for (i = 0; i < sma->sem_nsems; i++)
977                                 otime |= update_queue(sma, i, pt);
978                 } else {
979                         /*
980                          * Check the semaphores that were increased:
981                          * - No complex ops, thus all sleeping ops are
982                          *   decrease.
983                          * - if we decreased the value, then any sleeping
984                          *   semaphore ops wont be able to run: If the
985                          *   previous value was too small, then the new
986                          *   value will be too small, too.
987                          */
988                         for (i = 0; i < nsops; i++) {
989                                 if (sops[i].sem_op > 0) {
990                                         otime |= update_queue(sma,
991                                                         sops[i].sem_num, pt);
992                                 }
993                         }
994                 }
995         }
996         if (otime)
997                 set_semotime(sma, sops);
998 }
999 
1000 /* The following counts are associated to each semaphore:
1001  *   semncnt        number of tasks waiting on semval being nonzero
1002  *   semzcnt        number of tasks waiting on semval being zero
1003  * This model assumes that a task waits on exactly one semaphore.
1004  * Since semaphore operations are to be performed atomically, tasks actually
1005  * wait on a whole sequence of semaphores simultaneously.
1006  * The counts we return here are a rough approximation, but still
1007  * warrant that semncnt+semzcnt>0 if the task is on the pending queue.
1008  */
1009 static int count_semncnt (struct sem_array * sma, ushort semnum)
1010 {
1011         int semncnt;
1012         struct sem_queue * q;
1013 
1014         semncnt = 0;
1015         list_for_each_entry(q, &sma->sem_base[semnum].pending_alter, list) {
1016                 struct sembuf * sops = q->sops;
1017                 BUG_ON(sops->sem_num != semnum);
1018                 if ((sops->sem_op < 0) && !(sops->sem_flg & IPC_NOWAIT))
1019                         semncnt++;
1020         }
1021 
1022         list_for_each_entry(q, &sma->pending_alter, list) {
1023                 struct sembuf * sops = q->sops;
1024                 int nsops = q->nsops;
1025                 int i;
1026                 for (i = 0; i < nsops; i++)
1027                         if (sops[i].sem_num == semnum
1028                             && (sops[i].sem_op < 0)
1029                             && !(sops[i].sem_flg & IPC_NOWAIT))
1030                                 semncnt++;
1031         }
1032         return semncnt;
1033 }
1034 
1035 static int count_semzcnt (struct sem_array * sma, ushort semnum)
1036 {
1037         int semzcnt;
1038         struct sem_queue * q;
1039 
1040         semzcnt = 0;
1041         list_for_each_entry(q, &sma->sem_base[semnum].pending_const, list) {
1042                 struct sembuf * sops = q->sops;
1043                 BUG_ON(sops->sem_num != semnum);
1044                 if ((sops->sem_op == 0) && !(sops->sem_flg & IPC_NOWAIT))
1045                         semzcnt++;
1046         }
1047 
1048         list_for_each_entry(q, &sma->pending_const, list) {
1049                 struct sembuf * sops = q->sops;
1050                 int nsops = q->nsops;
1051                 int i;
1052                 for (i = 0; i < nsops; i++)
1053                         if (sops[i].sem_num == semnum
1054                             && (sops[i].sem_op == 0)
1055                             && !(sops[i].sem_flg & IPC_NOWAIT))
1056                                 semzcnt++;
1057         }
1058         return semzcnt;
1059 }
1060 
1061 /* Free a semaphore set. freeary() is called with sem_ids.rwsem locked
1062  * as a writer and the spinlock for this semaphore set hold. sem_ids.rwsem
1063  * remains locked on exit.
1064  */
1065 static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
1066 {
1067         struct sem_undo *un, *tu;
1068         struct sem_queue *q, *tq;
1069         struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
1070         struct list_head tasks;
1071         int i;
1072 
1073         /* Free the existing undo structures for this semaphore set.  */
1074         ipc_assert_locked_object(&sma->sem_perm);
1075         list_for_each_entry_safe(un, tu, &sma->list_id, list_id) {
1076                 list_del(&un->list_id);
1077                 spin_lock(&un->ulp->lock);
1078                 un->semid = -1;
1079                 list_del_rcu(&un->list_proc);
1080                 spin_unlock(&un->ulp->lock);
1081                 kfree_rcu(un, rcu);
1082         }
1083 
1084         /* Wake up all pending processes and let them fail with EIDRM. */
1085         INIT_LIST_HEAD(&tasks);
1086         list_for_each_entry_safe(q, tq, &sma->pending_const, list) {
1087                 unlink_queue(sma, q);
1088                 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1089         }
1090 
1091         list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
1092                 unlink_queue(sma, q);
1093                 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1094         }
1095         for (i = 0; i < sma->sem_nsems; i++) {
1096                 struct sem *sem = sma->sem_base + i;
1097                 list_for_each_entry_safe(q, tq, &sem->pending_const, list) {
1098                         unlink_queue(sma, q);
1099                         wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1100                 }
1101                 list_for_each_entry_safe(q, tq, &sem->pending_alter, list) {
1102                         unlink_queue(sma, q);
1103                         wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1104                 }
1105         }
1106 
1107         /* Remove the semaphore set from the IDR */
1108         sem_rmid(ns, sma);
1109         sem_unlock(sma, -1);
1110         rcu_read_unlock();
1111 
1112         wake_up_sem_queue_do(&tasks);
1113         ns->used_sems -= sma->sem_nsems;
1114         ipc_rcu_putref(sma, sem_rcu_free);
1115 }
1116 
1117 static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
1118 {
1119         switch(version) {
1120         case IPC_64:
1121                 return copy_to_user(buf, in, sizeof(*in));
1122         case IPC_OLD:
1123             {
1124                 struct semid_ds out;
1125 
1126                 memset(&out, 0, sizeof(out));
1127 
1128                 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
1129 
1130                 out.sem_otime   = in->sem_otime;
1131                 out.sem_ctime   = in->sem_ctime;
1132                 out.sem_nsems   = in->sem_nsems;
1133 
1134                 return copy_to_user(buf, &out, sizeof(out));
1135             }
1136         default:
1137                 return -EINVAL;
1138         }
1139 }
1140 
1141 static time_t get_semotime(struct sem_array *sma)
1142 {
1143         int i;
1144         time_t res;
1145 
1146         res = sma->sem_base[0].sem_otime;
1147         for (i = 1; i < sma->sem_nsems; i++) {
1148                 time_t to = sma->sem_base[i].sem_otime;
1149 
1150                 if (to > res)
1151                         res = to;
1152         }
1153         return res;
1154 }
1155 
1156 static int semctl_nolock(struct ipc_namespace *ns, int semid,
1157                          int cmd, int version, void __user *p)
1158 {
1159         int err;
1160         struct sem_array *sma;
1161 
1162         switch(cmd) {
1163         case IPC_INFO:
1164         case SEM_INFO:
1165         {
1166                 struct seminfo seminfo;
1167                 int max_id;
1168 
1169                 err = security_sem_semctl(NULL, cmd);
1170                 if (err)
1171                         return err;
1172                 
1173                 memset(&seminfo,0,sizeof(seminfo));
1174                 seminfo.semmni = ns->sc_semmni;
1175                 seminfo.semmns = ns->sc_semmns;
1176                 seminfo.semmsl = ns->sc_semmsl;
1177                 seminfo.semopm = ns->sc_semopm;
1178                 seminfo.semvmx = SEMVMX;
1179                 seminfo.semmnu = SEMMNU;
1180                 seminfo.semmap = SEMMAP;
1181                 seminfo.semume = SEMUME;
1182                 down_read(&sem_ids(ns).rwsem);
1183                 if (cmd == SEM_INFO) {
1184                         seminfo.semusz = sem_ids(ns).in_use;
1185                         seminfo.semaem = ns->used_sems;
1186                 } else {
1187                         seminfo.semusz = SEMUSZ;
1188                         seminfo.semaem = SEMAEM;
1189                 }
1190                 max_id = ipc_get_maxid(&sem_ids(ns));
1191                 up_read(&sem_ids(ns).rwsem);
1192                 if (copy_to_user(p, &seminfo, sizeof(struct seminfo))) 
1193                         return -EFAULT;
1194                 return (max_id < 0) ? 0: max_id;
1195         }
1196         case IPC_STAT:
1197         case SEM_STAT:
1198         {
1199                 struct semid64_ds tbuf;
1200                 int id = 0;
1201 
1202                 memset(&tbuf, 0, sizeof(tbuf));
1203 
1204                 rcu_read_lock();
1205                 if (cmd == SEM_STAT) {
1206                         sma = sem_obtain_object(ns, semid);
1207                         if (IS_ERR(sma)) {
1208                                 err = PTR_ERR(sma);
1209                                 goto out_unlock;
1210                         }
1211                         id = sma->sem_perm.id;
1212                 } else {
1213                         sma = sem_obtain_object_check(ns, semid);
1214                         if (IS_ERR(sma)) {
1215                                 err = PTR_ERR(sma);
1216                                 goto out_unlock;
1217                         }
1218                 }
1219 
1220                 err = -EACCES;
1221                 if (ipcperms(ns, &sma->sem_perm, S_IRUGO))
1222                         goto out_unlock;
1223 
1224                 err = security_sem_semctl(sma, cmd);
1225                 if (err)
1226                         goto out_unlock;
1227 
1228                 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
1229                 tbuf.sem_otime = get_semotime(sma);
1230                 tbuf.sem_ctime = sma->sem_ctime;
1231                 tbuf.sem_nsems = sma->sem_nsems;
1232                 rcu_read_unlock();
1233                 if (copy_semid_to_user(p, &tbuf, version))
1234                         return -EFAULT;
1235                 return id;
1236         }
1237         default:
1238                 return -EINVAL;
1239         }
1240 out_unlock:
1241         rcu_read_unlock();
1242         return err;
1243 }
1244 
1245 static int semctl_setval(struct ipc_namespace *ns, int semid, int semnum,
1246                 unsigned long arg)
1247 {
1248         struct sem_undo *un;
1249         struct sem_array *sma;
1250         struct sem* curr;
1251         int err;
1252         struct list_head tasks;
1253         int val;
1254 #if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN)
1255         /* big-endian 64bit */
1256         val = arg >> 32;
1257 #else
1258         /* 32bit or little-endian 64bit */
1259         val = arg;
1260 #endif
1261 
1262         if (val > SEMVMX || val < 0)
1263                 return -ERANGE;
1264 
1265         INIT_LIST_HEAD(&tasks);
1266 
1267         rcu_read_lock();
1268         sma = sem_obtain_object_check(ns, semid);
1269         if (IS_ERR(sma)) {
1270                 rcu_read_unlock();
1271                 return PTR_ERR(sma);
1272         }
1273 
1274         if (semnum < 0 || semnum >= sma->sem_nsems) {
1275                 rcu_read_unlock();
1276                 return -EINVAL;
1277         }
1278 
1279 
1280         if (ipcperms(ns, &sma->sem_perm, S_IWUGO)) {
1281                 rcu_read_unlock();
1282                 return -EACCES;
1283         }
1284 
1285         err = security_sem_semctl(sma, SETVAL);
1286         if (err) {
1287                 rcu_read_unlock();
1288                 return -EACCES;
1289         }
1290 
1291         sem_lock(sma, NULL, -1);
1292 
1293         if (sma->sem_perm.deleted) {
1294                 sem_unlock(sma, -1);
1295                 rcu_read_unlock();
1296                 return -EIDRM;
1297         }
1298 
1299         curr = &sma->sem_base[semnum];
1300 
1301         ipc_assert_locked_object(&sma->sem_perm);
1302         list_for_each_entry(un, &sma->list_id, list_id)
1303                 un->semadj[semnum] = 0;
1304 
1305         curr->semval = val;
1306         curr->sempid = task_tgid_vnr(current);
1307         sma->sem_ctime = get_seconds();
1308         /* maybe some queued-up processes were waiting for this */
1309         do_smart_update(sma, NULL, 0, 0, &tasks);
1310         sem_unlock(sma, -1);
1311         rcu_read_unlock();
1312         wake_up_sem_queue_do(&tasks);
1313         return 0;
1314 }
1315 
1316 static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
1317                 int cmd, void __user *p)
1318 {
1319         struct sem_array *sma;
1320         struct sem* curr;
1321         int err, nsems;
1322         ushort fast_sem_io[SEMMSL_FAST];
1323         ushort* sem_io = fast_sem_io;
1324         struct list_head tasks;
1325 
1326         INIT_LIST_HEAD(&tasks);
1327 
1328         rcu_read_lock();
1329         sma = sem_obtain_object_check(ns, semid);
1330         if (IS_ERR(sma)) {
1331                 rcu_read_unlock();
1332                 return PTR_ERR(sma);
1333         }
1334 
1335         nsems = sma->sem_nsems;
1336 
1337         err = -EACCES;
1338         if (ipcperms(ns, &sma->sem_perm, cmd == SETALL ? S_IWUGO : S_IRUGO))
1339                 goto out_rcu_wakeup;
1340 
1341         err = security_sem_semctl(sma, cmd);
1342         if (err)
1343                 goto out_rcu_wakeup;
1344 
1345         err = -EACCES;
1346         switch (cmd) {
1347         case GETALL:
1348         {
1349                 ushort __user *array = p;
1350                 int i;
1351 
1352                 sem_lock(sma, NULL, -1);
1353                 if (sma->sem_perm.deleted) {
1354                         err = -EIDRM;
1355                         goto out_unlock;
1356                 }
1357                 if(nsems > SEMMSL_FAST) {
1358                         if (!ipc_rcu_getref(sma)) {
1359                                 err = -EIDRM;
1360                                 goto out_unlock;
1361                         }
1362                         sem_unlock(sma, -1);
1363                         rcu_read_unlock();
1364                         sem_io = ipc_alloc(sizeof(ushort)*nsems);
1365                         if(sem_io == NULL) {
1366                                 ipc_rcu_putref(sma, ipc_rcu_free);
1367                                 return -ENOMEM;
1368                         }
1369 
1370                         rcu_read_lock();
1371                         sem_lock_and_putref(sma);
1372                         if (sma->sem_perm.deleted) {
1373                                 err = -EIDRM;
1374                                 goto out_unlock;
1375                         }
1376                 }
1377                 for (i = 0; i < sma->sem_nsems; i++)
1378                         sem_io[i] = sma->sem_base[i].semval;
1379                 sem_unlock(sma, -1);
1380                 rcu_read_unlock();
1381                 err = 0;
1382                 if(copy_to_user(array, sem_io, nsems*sizeof(ushort)))
1383                         err = -EFAULT;
1384                 goto out_free;
1385         }
1386         case SETALL:
1387         {
1388                 int i;
1389                 struct sem_undo *un;
1390 
1391                 if (!ipc_rcu_getref(sma)) {
1392                         err = -EIDRM;
1393                         goto out_rcu_wakeup;
1394                 }
1395                 rcu_read_unlock();
1396 
1397                 if(nsems > SEMMSL_FAST) {
1398                         sem_io = ipc_alloc(sizeof(ushort)*nsems);
1399                         if(sem_io == NULL) {
1400                                 ipc_rcu_putref(sma, ipc_rcu_free);
1401                                 return -ENOMEM;
1402                         }
1403                 }
1404 
1405                 if (copy_from_user (sem_io, p, nsems*sizeof(ushort))) {
1406                         ipc_rcu_putref(sma, ipc_rcu_free);
1407                         err = -EFAULT;
1408                         goto out_free;
1409                 }
1410 
1411                 for (i = 0; i < nsems; i++) {
1412                         if (sem_io[i] > SEMVMX) {
1413                                 ipc_rcu_putref(sma, ipc_rcu_free);
1414                                 err = -ERANGE;
1415                                 goto out_free;
1416                         }
1417                 }
1418                 rcu_read_lock();
1419                 sem_lock_and_putref(sma);
1420                 if (sma->sem_perm.deleted) {
1421                         err = -EIDRM;
1422                         goto out_unlock;
1423                 }
1424 
1425                 for (i = 0; i < nsems; i++)
1426                         sma->sem_base[i].semval = sem_io[i];
1427 
1428                 ipc_assert_locked_object(&sma->sem_perm);
1429                 list_for_each_entry(un, &sma->list_id, list_id) {
1430                         for (i = 0; i < nsems; i++)
1431                                 un->semadj[i] = 0;
1432                 }
1433                 sma->sem_ctime = get_seconds();
1434                 /* maybe some queued-up processes were waiting for this */
1435                 do_smart_update(sma, NULL, 0, 0, &tasks);
1436                 err = 0;
1437                 goto out_unlock;
1438         }
1439         /* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */
1440         }
1441         err = -EINVAL;
1442         if (semnum < 0 || semnum >= nsems)
1443                 goto out_rcu_wakeup;
1444 
1445         sem_lock(sma, NULL, -1);
1446         if (sma->sem_perm.deleted) {
1447                 err = -EIDRM;
1448                 goto out_unlock;
1449         }
1450         curr = &sma->sem_base[semnum];
1451 
1452         switch (cmd) {
1453         case GETVAL:
1454                 err = curr->semval;
1455                 goto out_unlock;
1456         case GETPID:
1457                 err = curr->sempid;
1458                 goto out_unlock;
1459         case GETNCNT:
1460                 err = count_semncnt(sma,semnum);
1461                 goto out_unlock;
1462         case GETZCNT:
1463                 err = count_semzcnt(sma,semnum);
1464                 goto out_unlock;
1465         }
1466 
1467 out_unlock:
1468         sem_unlock(sma, -1);
1469 out_rcu_wakeup:
1470         rcu_read_unlock();
1471         wake_up_sem_queue_do(&tasks);
1472 out_free:
1473         if(sem_io != fast_sem_io)
1474                 ipc_free(sem_io, sizeof(ushort)*nsems);
1475         return err;
1476 }
1477 
1478 static inline unsigned long
1479 copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
1480 {
1481         switch(version) {
1482         case IPC_64:
1483                 if (copy_from_user(out, buf, sizeof(*out)))
1484                         return -EFAULT;
1485                 return 0;
1486         case IPC_OLD:
1487             {
1488                 struct semid_ds tbuf_old;
1489 
1490                 if(copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
1491                         return -EFAULT;
1492 
1493                 out->sem_perm.uid       = tbuf_old.sem_perm.uid;
1494                 out->sem_perm.gid       = tbuf_old.sem_perm.gid;
1495                 out->sem_perm.mode      = tbuf_old.sem_perm.mode;
1496 
1497                 return 0;
1498             }
1499         default:
1500                 return -EINVAL;
1501         }
1502 }
1503 
1504 /*
1505  * This function handles some semctl commands which require the rwsem
1506  * to be held in write mode.
1507  * NOTE: no locks must be held, the rwsem is taken inside this function.
1508  */
1509 static int semctl_down(struct ipc_namespace *ns, int semid,
1510                        int cmd, int version, void __user *p)
1511 {
1512         struct sem_array *sma;
1513         int err;
1514         struct semid64_ds semid64;
1515         struct kern_ipc_perm *ipcp;
1516 
1517         if(cmd == IPC_SET) {
1518                 if (copy_semid_from_user(&semid64, p, version))
1519                         return -EFAULT;
1520         }
1521 
1522         down_write(&sem_ids(ns).rwsem);
1523         rcu_read_lock();
1524 
1525         ipcp = ipcctl_pre_down_nolock(ns, &sem_ids(ns), semid, cmd,
1526                                       &semid64.sem_perm, 0);
1527         if (IS_ERR(ipcp)) {
1528                 err = PTR_ERR(ipcp);
1529                 goto out_unlock1;
1530         }
1531 
1532         sma = container_of(ipcp, struct sem_array, sem_perm);
1533 
1534         err = security_sem_semctl(sma, cmd);
1535         if (err)
1536                 goto out_unlock1;
1537 
1538         switch (cmd) {
1539         case IPC_RMID:
1540                 sem_lock(sma, NULL, -1);
1541                 /* freeary unlocks the ipc object and rcu */
1542                 freeary(ns, ipcp);
1543                 goto out_up;
1544         case IPC_SET:
1545                 sem_lock(sma, NULL, -1);
1546                 err = ipc_update_perm(&semid64.sem_perm, ipcp);
1547                 if (err)
1548                         goto out_unlock0;
1549                 sma->sem_ctime = get_seconds();
1550                 break;
1551         default:
1552                 err = -EINVAL;
1553                 goto out_unlock1;
1554         }
1555 
1556 out_unlock0:
1557         sem_unlock(sma, -1);
1558 out_unlock1:
1559         rcu_read_unlock();
1560 out_up:
1561         up_write(&sem_ids(ns).rwsem);
1562         return err;
1563 }
1564 
1565 SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, unsigned long, arg)
1566 {
1567         int version;
1568         struct ipc_namespace *ns;
1569         void __user *p = (void __user *)arg;
1570 
1571         if (semid < 0)
1572                 return -EINVAL;
1573 
1574         version = ipc_parse_version(&cmd);
1575         ns = current->nsproxy->ipc_ns;
1576 
1577         switch(cmd) {
1578         case IPC_INFO:
1579         case SEM_INFO:
1580         case IPC_STAT:
1581         case SEM_STAT:
1582                 return semctl_nolock(ns, semid, cmd, version, p);
1583         case GETALL:
1584         case GETVAL:
1585         case GETPID:
1586         case GETNCNT:
1587         case GETZCNT:
1588         case SETALL:
1589                 return semctl_main(ns, semid, semnum, cmd, p);
1590         case SETVAL:
1591                 return semctl_setval(ns, semid, semnum, arg);
1592         case IPC_RMID:
1593         case IPC_SET:
1594                 return semctl_down(ns, semid, cmd, version, p);
1595         default:
1596                 return -EINVAL;
1597         }
1598 }
1599 
1600 /* If the task doesn't already have a undo_list, then allocate one
1601  * here.  We guarantee there is only one thread using this undo list,
1602  * and current is THE ONE
1603  *
1604  * If this allocation and assignment succeeds, but later
1605  * portions of this code fail, there is no need to free the sem_undo_list.
1606  * Just let it stay associated with the task, and it'll be freed later
1607  * at exit time.
1608  *
1609  * This can block, so callers must hold no locks.
1610  */
1611 static inline int get_undo_list(struct sem_undo_list **undo_listp)
1612 {
1613         struct sem_undo_list *undo_list;
1614 
1615         undo_list = current->sysvsem.undo_list;
1616         if (!undo_list) {
1617                 undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
1618                 if (undo_list == NULL)
1619                         return -ENOMEM;
1620                 spin_lock_init(&undo_list->lock);
1621                 atomic_set(&undo_list->refcnt, 1);
1622                 INIT_LIST_HEAD(&undo_list->list_proc);
1623 
1624                 current->sysvsem.undo_list = undo_list;
1625         }
1626         *undo_listp = undo_list;
1627         return 0;
1628 }
1629 
1630 static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid)
1631 {
1632         struct sem_undo *un;
1633 
1634         list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) {
1635                 if (un->semid == semid)
1636                         return un;
1637         }
1638         return NULL;
1639 }
1640 
1641 static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1642 {
1643         struct sem_undo *un;
1644 
1645         assert_spin_locked(&ulp->lock);
1646 
1647         un = __lookup_undo(ulp, semid);
1648         if (un) {
1649                 list_del_rcu(&un->list_proc);
1650                 list_add_rcu(&un->list_proc, &ulp->list_proc);
1651         }
1652         return un;
1653 }
1654 
1655 /**
1656  * find_alloc_undo - Lookup (and if not present create) undo array
1657  * @ns: namespace
1658  * @semid: semaphore array id
1659  *
1660  * The function looks up (and if not present creates) the undo structure.
1661  * The size of the undo structure depends on the size of the semaphore
1662  * array, thus the alloc path is not that straightforward.
1663  * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1664  * performs a rcu_read_lock().
1665  */
1666 static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
1667 {
1668         struct sem_array *sma;
1669         struct sem_undo_list *ulp;
1670         struct sem_undo *un, *new;
1671         int nsems, error;
1672 
1673         error = get_undo_list(&ulp);
1674         if (error)
1675                 return ERR_PTR(error);
1676 
1677         rcu_read_lock();
1678         spin_lock(&ulp->lock);
1679         un = lookup_undo(ulp, semid);
1680         spin_unlock(&ulp->lock);
1681         if (likely(un!=NULL))
1682                 goto out;
1683 
1684         /* no undo structure around - allocate one. */
1685         /* step 1: figure out the size of the semaphore array */
1686         sma = sem_obtain_object_check(ns, semid);
1687         if (IS_ERR(sma)) {
1688                 rcu_read_unlock();
1689                 return ERR_CAST(sma);
1690         }
1691 
1692         nsems = sma->sem_nsems;
1693         if (!ipc_rcu_getref(sma)) {
1694                 rcu_read_unlock();
1695                 un = ERR_PTR(-EIDRM);
1696                 goto out;
1697         }
1698         rcu_read_unlock();
1699 
1700         /* step 2: allocate new undo structure */
1701         new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1702         if (!new) {
1703                 ipc_rcu_putref(sma, ipc_rcu_free);
1704                 return ERR_PTR(-ENOMEM);
1705         }
1706 
1707         /* step 3: Acquire the lock on semaphore array */
1708         rcu_read_lock();
1709         sem_lock_and_putref(sma);
1710         if (sma->sem_perm.deleted) {
1711                 sem_unlock(sma, -1);
1712                 rcu_read_unlock();
1713                 kfree(new);
1714                 un = ERR_PTR(-EIDRM);
1715                 goto out;
1716         }
1717         spin_lock(&ulp->lock);
1718 
1719         /*
1720          * step 4: check for races: did someone else allocate the undo struct?
1721          */
1722         un = lookup_undo(ulp, semid);
1723         if (un) {
1724                 kfree(new);
1725                 goto success;
1726         }
1727         /* step 5: initialize & link new undo structure */
1728         new->semadj = (short *) &new[1];
1729         new->ulp = ulp;
1730         new->semid = semid;
1731         assert_spin_locked(&ulp->lock);
1732         list_add_rcu(&new->list_proc, &ulp->list_proc);
1733         ipc_assert_locked_object(&sma->sem_perm);
1734         list_add(&new->list_id, &sma->list_id);
1735         un = new;
1736 
1737 success:
1738         spin_unlock(&ulp->lock);
1739         sem_unlock(sma, -1);
1740 out:
1741         return un;
1742 }
1743 
1744 
1745 /**
1746  * get_queue_result - Retrieve the result code from sem_queue
1747  * @q: Pointer to queue structure
1748  *
1749  * Retrieve the return code from the pending queue. If IN_WAKEUP is found in
1750  * q->status, then we must loop until the value is replaced with the final
1751  * value: This may happen if a task is woken up by an unrelated event (e.g.
1752  * signal) and in parallel the task is woken up by another task because it got
1753  * the requested semaphores.
1754  *
1755  * The function can be called with or without holding the semaphore spinlock.
1756  */
1757 static int get_queue_result(struct sem_queue *q)
1758 {
1759         int error;
1760 
1761         error = q->status;
1762         while (unlikely(error == IN_WAKEUP)) {
1763                 cpu_relax();
1764                 error = q->status;
1765         }
1766 
1767         return error;
1768 }
1769 
1770 SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
1771                 unsigned, nsops, const struct timespec __user *, timeout)
1772 {
1773         int error = -EINVAL;
1774         struct sem_array *sma;
1775         struct sembuf fast_sops[SEMOPM_FAST];
1776         struct sembuf* sops = fast_sops, *sop;
1777         struct sem_undo *un;
1778         int undos = 0, alter = 0, max, locknum;
1779         struct sem_queue queue;
1780         unsigned long jiffies_left = 0;
1781         struct ipc_namespace *ns;
1782         struct list_head tasks;
1783 
1784         ns = current->nsproxy->ipc_ns;
1785 
1786         if (nsops < 1 || semid < 0)
1787                 return -EINVAL;
1788         if (nsops > ns->sc_semopm)
1789                 return -E2BIG;
1790         if(nsops > SEMOPM_FAST) {
1791                 sops = kmalloc(sizeof(*sops)*nsops,GFP_KERNEL);
1792                 if(sops==NULL)
1793                         return -ENOMEM;
1794         }
1795         if (copy_from_user (sops, tsops, nsops * sizeof(*tsops))) {
1796                 error=-EFAULT;
1797                 goto out_free;
1798         }
1799         if (timeout) {
1800                 struct timespec _timeout;
1801                 if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1802                         error = -EFAULT;
1803                         goto out_free;
1804                 }
1805                 if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1806                         _timeout.tv_nsec >= 1000000000L) {
1807                         error = -EINVAL;
1808                         goto out_free;
1809                 }
1810                 jiffies_left = timespec_to_jiffies(&_timeout);
1811         }
1812         max = 0;
1813         for (sop = sops; sop < sops + nsops; sop++) {
1814                 if (sop->sem_num >= max)
1815                         max = sop->sem_num;
1816                 if (sop->sem_flg & SEM_UNDO)
1817                         undos = 1;
1818                 if (sop->sem_op != 0)
1819                         alter = 1;
1820         }
1821 
1822         INIT_LIST_HEAD(&tasks);
1823 
1824         if (undos) {
1825                 /* On success, find_alloc_undo takes the rcu_read_lock */
1826                 un = find_alloc_undo(ns, semid);
1827                 if (IS_ERR(un)) {
1828                         error = PTR_ERR(un);
1829                         goto out_free;
1830                 }
1831         } else {
1832                 un = NULL;
1833                 rcu_read_lock();
1834         }
1835 
1836         sma = sem_obtain_object_check(ns, semid);
1837         if (IS_ERR(sma)) {
1838                 rcu_read_unlock();
1839                 error = PTR_ERR(sma);
1840                 goto out_free;
1841         }
1842 
1843         error = -EFBIG;
1844         if (max >= sma->sem_nsems)
1845                 goto out_rcu_wakeup;
1846 
1847         error = -EACCES;
1848         if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO))
1849                 goto out_rcu_wakeup;
1850 
1851         error = security_sem_semop(sma, sops, nsops, alter);
1852         if (error)
1853                 goto out_rcu_wakeup;
1854 
1855         error = -EIDRM;
1856         locknum = sem_lock(sma, sops, nsops);
1857         if (sma->sem_perm.deleted)
1858                 goto out_unlock_free;
1859         /*
1860          * semid identifiers are not unique - find_alloc_undo may have
1861          * allocated an undo structure, it was invalidated by an RMID
1862          * and now a new array with received the same id. Check and fail.
1863          * This case can be detected checking un->semid. The existence of
1864          * "un" itself is guaranteed by rcu.
1865          */
1866         if (un && un->semid == -1)
1867                 goto out_unlock_free;
1868 
1869         error = perform_atomic_semop(sma, sops, nsops, un,
1870                                         task_tgid_vnr(current));
1871         if (error == 0) {
1872                 /* If the operation was successful, then do
1873                  * the required updates.
1874                  */
1875                 if (alter)
1876                         do_smart_update(sma, sops, nsops, 1, &tasks);
1877                 else
1878                         set_semotime(sma, sops);
1879         }
1880         if (error <= 0)
1881                 goto out_unlock_free;
1882 
1883         /* We need to sleep on this operation, so we put the current
1884          * task into the pending queue and go to sleep.
1885          */
1886                 
1887         queue.sops = sops;
1888         queue.nsops = nsops;
1889         queue.undo = un;
1890         queue.pid = task_tgid_vnr(current);
1891         queue.alter = alter;
1892 
1893         if (nsops == 1) {
1894                 struct sem *curr;
1895                 curr = &sma->sem_base[sops->sem_num];
1896 
1897                 if (alter) {
1898                         if (sma->complex_count) {
1899                                 list_add_tail(&queue.list,
1900                                                 &sma->pending_alter);
1901                         } else {
1902 
1903                                 list_add_tail(&queue.list,
1904                                                 &curr->pending_alter);
1905                         }
1906                 } else {
1907                         list_add_tail(&queue.list, &curr->pending_const);
1908                 }
1909         } else {
1910                 if (!sma->complex_count)
1911                         merge_queues(sma);
1912 
1913                 if (alter)
1914                         list_add_tail(&queue.list, &sma->pending_alter);
1915                 else
1916                         list_add_tail(&queue.list, &sma->pending_const);
1917 
1918                 sma->complex_count++;
1919         }
1920 
1921         queue.status = -EINTR;
1922         queue.sleeper = current;
1923 
1924 sleep_again:
1925         current->state = TASK_INTERRUPTIBLE;
1926         sem_unlock(sma, locknum);
1927         rcu_read_unlock();
1928 
1929         if (timeout)
1930                 jiffies_left = schedule_timeout(jiffies_left);
1931         else
1932                 schedule();
1933 
1934         error = get_queue_result(&queue);
1935 
1936         if (error != -EINTR) {
1937                 /* fast path: update_queue already obtained all requested
1938                  * resources.
1939                  * Perform a smp_mb(): User space could assume that semop()
1940                  * is a memory barrier: Without the mb(), the cpu could
1941                  * speculatively read in user space stale data that was
1942                  * overwritten by the previous owner of the semaphore.
1943                  */
1944                 smp_mb();
1945 
1946                 goto out_free;
1947         }
1948 
1949         rcu_read_lock();
1950         sma = sem_obtain_lock(ns, semid, sops, nsops, &locknum);
1951 
1952         /*
1953          * Wait until it's guaranteed that no wakeup_sem_queue_do() is ongoing.
1954          */
1955         error = get_queue_result(&queue);
1956 
1957         /*
1958          * Array removed? If yes, leave without sem_unlock().
1959          */
1960         if (IS_ERR(sma)) {
1961                 rcu_read_unlock();
1962                 goto out_free;
1963         }
1964 
1965 
1966         /*
1967          * If queue.status != -EINTR we are woken up by another process.
1968          * Leave without unlink_queue(), but with sem_unlock().
1969          */
1970 
1971         if (error != -EINTR) {
1972                 goto out_unlock_free;
1973         }
1974 
1975         /*
1976          * If an interrupt occurred we have to clean up the queue
1977          */
1978         if (timeout && jiffies_left == 0)
1979                 error = -EAGAIN;
1980 
1981         /*
1982          * If the wakeup was spurious, just retry
1983          */
1984         if (error == -EINTR && !signal_pending(current))
1985                 goto sleep_again;
1986 
1987         unlink_queue(sma, &queue);
1988 
1989 out_unlock_free:
1990         sem_unlock(sma, locknum);
1991 out_rcu_wakeup:
1992         rcu_read_unlock();
1993         wake_up_sem_queue_do(&tasks);
1994 out_free:
1995         if(sops != fast_sops)
1996                 kfree(sops);
1997         return error;
1998 }
1999 
2000 SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
2001                 unsigned, nsops)
2002 {
2003         return sys_semtimedop(semid, tsops, nsops, NULL);
2004 }
2005 
2006 /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
2007  * parent and child tasks.
2008  */
2009 
2010 int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
2011 {
2012         struct sem_undo_list *undo_list;
2013         int error;
2014 
2015         if (clone_flags & CLONE_SYSVSEM) {
2016                 error = get_undo_list(&undo_list);
2017                 if (error)
2018                         return error;
2019                 atomic_inc(&undo_list->refcnt);
2020                 tsk->sysvsem.undo_list = undo_list;
2021         } else 
2022                 tsk->sysvsem.undo_list = NULL;
2023 
2024         return 0;
2025 }
2026 
2027 /*
2028  * add semadj values to semaphores, free undo structures.
2029  * undo structures are not freed when semaphore arrays are destroyed
2030  * so some of them may be out of date.
2031  * IMPLEMENTATION NOTE: There is some confusion over whether the
2032  * set of adjustments that needs to be done should be done in an atomic
2033  * manner or not. That is, if we are attempting to decrement the semval
2034  * should we queue up and wait until we can do so legally?
2035  * The original implementation attempted to do this (queue and wait).
2036  * The current implementation does not do so. The POSIX standard
2037  * and SVID should be consulted to determine what behavior is mandated.
2038  */
2039 void exit_sem(struct task_struct *tsk)
2040 {
2041         struct sem_undo_list *ulp;
2042 
2043         ulp = tsk->sysvsem.undo_list;
2044         if (!ulp)
2045                 return;
2046         tsk->sysvsem.undo_list = NULL;
2047 
2048         if (!atomic_dec_and_test(&ulp->refcnt))
2049                 return;
2050 
2051         for (;;) {
2052                 struct sem_array *sma;
2053                 struct sem_undo *un;
2054                 struct list_head tasks;
2055                 int semid, i;
2056 
2057                 rcu_read_lock();
2058                 un = list_entry_rcu(ulp->list_proc.next,
2059                                     struct sem_undo, list_proc);
2060                 if (&un->list_proc == &ulp->list_proc) {
2061                         /*
2062                          * We must wait for freeary() before freeing this ulp,
2063                          * in case we raced with last sem_undo. There is a small
2064                          * possibility where we exit while freeary() didn't
2065                          * finish unlocking sem_undo_list.
2066                          */
2067                         spin_unlock_wait(&ulp->lock);
2068                         rcu_read_unlock();
2069                         break;
2070                 }
2071                 spin_lock(&ulp->lock);
2072                 semid = un->semid;
2073                 spin_unlock(&ulp->lock);
2074 
2075                 /* exit_sem raced with IPC_RMID, nothing to do */
2076                 if (semid == -1) {
2077                         rcu_read_unlock();
2078                         continue;
2079                 }
2080 
2081                 sma = sem_obtain_object_check(tsk->nsproxy->ipc_ns, semid);
2082                 /* exit_sem raced with IPC_RMID, nothing to do */
2083                 if (IS_ERR(sma)) {
2084                         rcu_read_unlock();
2085                         continue;
2086                 }
2087 
2088                 sem_lock(sma, NULL, -1);
2089                 /* exit_sem raced with IPC_RMID, nothing to do */
2090                 if (sma->sem_perm.deleted) {
2091                         sem_unlock(sma, -1);
2092                         rcu_read_unlock();
2093                         continue;
2094                 }
2095                 un = __lookup_undo(ulp, semid);
2096                 if (un == NULL) {
2097                         /* exit_sem raced with IPC_RMID+semget() that created
2098                          * exactly the same semid. Nothing to do.
2099                          */
2100                         sem_unlock(sma, -1);
2101                         rcu_read_unlock();
2102                         continue;
2103                 }
2104 
2105                 /* remove un from the linked lists */
2106                 ipc_assert_locked_object(&sma->sem_perm);
2107                 list_del(&un->list_id);
2108 
2109                 spin_lock(&ulp->lock);
2110                 list_del_rcu(&un->list_proc);
2111                 spin_unlock(&ulp->lock);
2112 
2113                 /* perform adjustments registered in un */
2114                 for (i = 0; i < sma->sem_nsems; i++) {
2115                         struct sem * semaphore = &sma->sem_base[i];
2116                         if (un->semadj[i]) {
2117                                 semaphore->semval += un->semadj[i];
2118                                 /*
2119                                  * Range checks of the new semaphore value,
2120                                  * not defined by sus:
2121                                  * - Some unices ignore the undo entirely
2122                                  *   (e.g. HP UX 11i 11.22, Tru64 V5.1)
2123                                  * - some cap the value (e.g. FreeBSD caps
2124                                  *   at 0, but doesn't enforce SEMVMX)
2125                                  *
2126                                  * Linux caps the semaphore value, both at 0
2127                                  * and at SEMVMX.
2128                                  *
2129                                  *      Manfred <manfred@colorfullife.com>
2130                                  */
2131                                 if (semaphore->semval < 0)
2132                                         semaphore->semval = 0;
2133                                 if (semaphore->semval > SEMVMX)
2134                                         semaphore->semval = SEMVMX;
2135                                 semaphore->sempid = task_tgid_vnr(current);
2136                         }
2137                 }
2138                 /* maybe some queued-up processes were waiting for this */
2139                 INIT_LIST_HEAD(&tasks);
2140                 do_smart_update(sma, NULL, 0, 1, &tasks);
2141                 sem_unlock(sma, -1);
2142                 rcu_read_unlock();
2143                 wake_up_sem_queue_do(&tasks);
2144 
2145                 kfree_rcu(un, rcu);
2146         }
2147         kfree(ulp);
2148 }
2149 
2150 #ifdef CONFIG_PROC_FS
2151 static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
2152 {
2153         struct user_namespace *user_ns = seq_user_ns(s);
2154         struct sem_array *sma = it;
2155         time_t sem_otime;
2156 
2157         /*
2158          * The proc interface isn't aware of sem_lock(), it calls
2159          * ipc_lock_object() directly (in sysvipc_find_ipc).
2160          * In order to stay compatible with sem_lock(), we must wait until
2161          * all simple semop() calls have left their critical regions.
2162          */
2163         sem_wait_array(sma);
2164 
2165         sem_otime = get_semotime(sma);
2166 
2167         return seq_printf(s,
2168                           "%10d %10d  %4o %10u %5u %5u %5u %5u %10lu %10lu\n",
2169                           sma->sem_perm.key,
2170                           sma->sem_perm.id,
2171                           sma->sem_perm.mode,
2172                           sma->sem_nsems,
2173                           from_kuid_munged(user_ns, sma->sem_perm.uid),
2174                           from_kgid_munged(user_ns, sma->sem_perm.gid),
2175                           from_kuid_munged(user_ns, sma->sem_perm.cuid),
2176                           from_kgid_munged(user_ns, sma->sem_perm.cgid),
2177                           sem_otime,
2178                           sma->sem_ctime);
2179 }
2180 #endif
2181 

~ [ 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