jail - Rework sysctl configuration variables
[dragonfly.git] / sys / kern / sysv_sem.c
1 /* $FreeBSD: src/sys/kern/sysv_sem.c,v 1.69 2004/03/17 09:37:13 cperciva Exp $ */
2
3 /*
4  * Implementation of SVID semaphores
5  *
6  * Author:  Daniel Boulet
7  *
8  * This software is provided ``AS IS'' without any warranties of any kind.
9  */
10
11 #include "opt_sysvipc.h"
12
13 #include <sys/param.h>
14 #include <sys/systm.h>
15 #include <sys/sysproto.h>
16 #include <sys/kernel.h>
17 #include <sys/proc.h>
18 #include <sys/sem.h>
19 #include <sys/sysent.h>
20 #include <sys/sysctl.h>
21 #include <sys/malloc.h>
22 #include <sys/jail.h>
23 #include <sys/thread.h>
24
25 static MALLOC_DEFINE(M_SEM, "sem", "SVID compatible semaphores");
26
27 static void seminit (void *);
28
29 static struct sem_undo *semu_alloc (struct proc *p);
30 static int semundo_adjust (struct proc *p, int semid, int semnum, int adjval);
31 static void semundo_clear (int semid, int semnum);
32
33 static struct lwkt_token semu_token = LWKT_TOKEN_INITIALIZER(semu_token);
34 static int      semtot = 0;
35 static struct semid_pool *sema; /* semaphore id pool */
36 static TAILQ_HEAD(, sem_undo) semu_list = TAILQ_HEAD_INITIALIZER(semu_list);
37 static struct lock sema_lk;
38
39 struct sem {
40         u_short semval;         /* semaphore value */
41         pid_t   sempid;         /* pid of last operation */
42         u_short semncnt;        /* # awaiting semval > cval */
43         u_short semzcnt;        /* # awaiting semval = 0 */
44 };
45
46 /*
47  * Undo structure (one per process)
48  */
49 struct sem_undo {
50         TAILQ_ENTRY(sem_undo) un_entry; /* linked list for semundo_clear() */
51         struct  proc *un_proc;          /* owner of this structure */
52         int     un_refs;                /* prevent unlink/kfree */
53         short   un_cnt;                 /* # of active entries */
54         short   un_unused;
55         struct undo {
56                 short   un_adjval;      /* adjust on exit values */
57                 short   un_num;         /* semaphore # */
58                 int     un_id;          /* semid */
59         } un_ent[1];                    /* undo entries */
60 };
61
62 /*
63  * Configuration parameters
64  */
65 #ifndef SEMMNI
66 #define SEMMNI  1024            /* # of semaphore identifiers */
67 #endif
68 #ifndef SEMMNS
69 #define SEMMNS  32767           /* # of semaphores in system */
70 #endif
71 #ifndef SEMUME
72 #define SEMUME  25              /* max # of undo entries per process */
73 #endif
74 #ifndef SEMMNU
75 #define SEMMNU  1024            /* # of undo structures in system */
76                                 /* NO LONGER USED */
77 #endif
78
79 /* shouldn't need tuning */
80 #ifndef SEMMAP
81 #define SEMMAP  128             /* # of entries in semaphore map */
82 #endif
83 #ifndef SEMMSL
84 #define SEMMSL  SEMMNS          /* max # of semaphores per id */
85 #endif
86 #ifndef SEMOPM
87 #define SEMOPM  100             /* max # of operations per semop call */
88 #endif
89
90 #define SEMVMX  32767           /* semaphore maximum value */
91 #define SEMAEM  16384           /* adjust on exit max value */
92
93 /*
94  * Due to the way semaphore memory is allocated, we have to ensure that
95  * SEMUSZ is properly aligned.
96  */
97
98 #define SEM_ALIGN(bytes) roundup2(bytes, sizeof(long))
99
100 /* actual size of an undo structure */
101 #define SEMUSZ(nent)    SEM_ALIGN(offsetof(struct sem_undo, un_ent[nent]))
102
103 /*
104  * semaphore info struct
105  */
106 struct seminfo seminfo = {
107                 SEMMAP,         /* # of entries in semaphore map */
108                 SEMMNI,         /* # of semaphore identifiers */
109                 SEMMNS,         /* # of semaphores in system */
110                 SEMMNU,         /* # of undo structures in system */
111                 SEMMSL,         /* max # of semaphores per id */
112                 SEMOPM,         /* max # of operations per semop call */
113                 SEMUME,         /* max # of undo entries per process */
114                 SEMUSZ(SEMUME), /* size in bytes of undo structure */
115                 SEMVMX,         /* semaphore maximum value */
116                 SEMAEM          /* adjust on exit max value */
117 };
118
119 TUNABLE_INT("kern.ipc.semmap", &seminfo.semmap);
120 TUNABLE_INT("kern.ipc.semmni", &seminfo.semmni);
121 TUNABLE_INT("kern.ipc.semmns", &seminfo.semmns);
122 TUNABLE_INT("kern.ipc.semmnu", &seminfo.semmnu);
123 TUNABLE_INT("kern.ipc.semmsl", &seminfo.semmsl);
124 TUNABLE_INT("kern.ipc.semopm", &seminfo.semopm);
125 TUNABLE_INT("kern.ipc.semume", &seminfo.semume);
126 TUNABLE_INT("kern.ipc.semusz", &seminfo.semusz);
127 TUNABLE_INT("kern.ipc.semvmx", &seminfo.semvmx);
128 TUNABLE_INT("kern.ipc.semaem", &seminfo.semaem);
129
130 SYSCTL_INT(_kern_ipc, OID_AUTO, semmap, CTLFLAG_RW, &seminfo.semmap, 0,
131     "Number of entries in semaphore map");
132 SYSCTL_INT(_kern_ipc, OID_AUTO, semmni, CTLFLAG_RD, &seminfo.semmni, 0,
133     "Number of semaphore identifiers");
134 SYSCTL_INT(_kern_ipc, OID_AUTO, semmns, CTLFLAG_RD, &seminfo.semmns, 0,
135     "Total number of semaphores");
136 SYSCTL_INT(_kern_ipc, OID_AUTO, semmnu, CTLFLAG_RD, &seminfo.semmnu, 0,
137     "Total number of undo structures");
138 SYSCTL_INT(_kern_ipc, OID_AUTO, semmsl, CTLFLAG_RW, &seminfo.semmsl, 0,
139     "Max number of semaphores per id");
140 SYSCTL_INT(_kern_ipc, OID_AUTO, semopm, CTLFLAG_RD, &seminfo.semopm, 0,
141     "Max number of operations per semop call");
142 SYSCTL_INT(_kern_ipc, OID_AUTO, semume, CTLFLAG_RD, &seminfo.semume, 0,
143     "Max number of undo entries per process");
144 SYSCTL_INT(_kern_ipc, OID_AUTO, semusz, CTLFLAG_RD, &seminfo.semusz, 0,
145     "Size in bytes of undo structure");
146 SYSCTL_INT(_kern_ipc, OID_AUTO, semvmx, CTLFLAG_RW, &seminfo.semvmx, 0,
147     "Semaphore maximum value");
148 SYSCTL_INT(_kern_ipc, OID_AUTO, semaem, CTLFLAG_RW, &seminfo.semaem, 0,
149     "Adjust on exit max value");
150
151 #if 0
152 RO seminfo.semmap       /* SEMMAP unused */
153 RO seminfo.semmni
154 RO seminfo.semmns
155 RO seminfo.semmnu       /* undo entries per system */
156 RW seminfo.semmsl
157 RO seminfo.semopm       /* SEMOPM unused */
158 RO seminfo.semume
159 RO seminfo.semusz       /* param - derived from SEMUME for per-proc sizeof */
160 RO seminfo.semvmx       /* SEMVMX unused - user param */
161 RO seminfo.semaem       /* SEMAEM unused - user param */
162 #endif
163
164 static void
165 seminit(void *dummy)
166 {
167         int i;
168
169         sema = kmalloc(sizeof(struct semid_pool) * seminfo.semmni,
170                       M_SEM, M_WAITOK | M_ZERO);
171
172         lockinit(&sema_lk, "semglb", 0, 0);
173         for (i = 0; i < seminfo.semmni; i++) {
174                 struct semid_pool *semaptr = &sema[i];
175
176                 lockinit(&semaptr->lk, "semary", 0, 0);
177                 semaptr->ds.sem_base = NULL;
178                 semaptr->ds.sem_perm.mode = 0;
179         }
180 }
181 SYSINIT(sysv_sem, SI_SUB_SYSV_SEM, SI_ORDER_FIRST, seminit, NULL);
182
183 /*
184  * Allocate a new sem_undo structure for a process
185  * (returns ptr to structure or NULL if no more room)
186  */
187 static struct sem_undo *
188 semu_alloc(struct proc *p)
189 {
190         struct sem_undo *semu;
191
192         /*
193          * Allocate the semu structure and associate it with the process,
194          * as necessary.
195          */
196         while ((semu = p->p_sem_undo) == NULL) {
197                 semu = kmalloc(SEMUSZ(seminfo.semume), M_SEM,
198                                M_WAITOK | M_ZERO);
199                 lwkt_gettoken(&semu_token);
200                 lwkt_gettoken(&p->p_token);
201                 if (p->p_sem_undo == NULL) {
202                         p->p_sem_undo = semu;
203                         p->p_flags |= P_SYSVSEM;
204                         semu->un_proc = p;
205                         TAILQ_INSERT_TAIL(&semu_list, semu, un_entry);
206                 } else {
207                         kfree(semu, M_SEM);
208                 }
209                 lwkt_reltoken(&p->p_token);
210                 lwkt_reltoken(&semu_token);
211         }
212         return(semu);
213 }
214
215 /*
216  * Adjust a particular entry for a particular proc
217  */
218 static int
219 semundo_adjust(struct proc *p, int semid, int semnum, int adjval)
220 {
221         struct sem_undo *suptr;
222         struct undo *sunptr;
223         int i;
224         int error = 0;
225
226         /*
227          * Look for and remember the sem_undo if the caller doesn't
228          * provide it.
229          */
230         suptr = semu_alloc(p);
231         lwkt_gettoken(&p->p_token);
232
233         /*
234          * Look for the requested entry and adjust it (delete if adjval becomes
235          * 0).
236          */
237         sunptr = &suptr->un_ent[0];
238         for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
239                 if (sunptr->un_id != semid || sunptr->un_num != semnum)
240                         continue;
241                 if (adjval == 0)
242                         sunptr->un_adjval = 0;
243                 else
244                         sunptr->un_adjval += adjval;
245                 if (sunptr->un_adjval == 0) {
246                         suptr->un_cnt--;
247                         if (i < suptr->un_cnt)
248                                 suptr->un_ent[i] = suptr->un_ent[suptr->un_cnt];
249                 }
250                 goto done;
251         }
252
253         /* Didn't find the right entry - create it */
254         if (adjval == 0)
255                 goto done;
256         if (suptr->un_cnt != seminfo.semume) {
257                 sunptr = &suptr->un_ent[suptr->un_cnt];
258                 suptr->un_cnt++;
259                 sunptr->un_adjval = adjval;
260                 sunptr->un_id = semid;
261                 sunptr->un_num = semnum;
262         } else {
263                 error = EINVAL;
264         }
265 done:
266         lwkt_reltoken(&p->p_token);
267
268         return (error);
269 }
270
271 /*
272  * This is rather expensive
273  */
274 static void
275 semundo_clear(int semid, int semnum)
276 {
277         struct proc *p;
278         struct sem_undo *suptr;
279         struct sem_undo *sunext;
280         struct undo *sunptr;
281         int i;
282
283         lwkt_gettoken(&semu_token);
284         sunext = TAILQ_FIRST(&semu_list);
285         while ((suptr = sunext) != NULL) {
286                 if ((p = suptr->un_proc) == NULL) {
287                         sunext = TAILQ_NEXT(suptr, un_entry);
288                         continue;
289                 }
290                 ++suptr->un_refs;
291                 PHOLD(p);
292                 lwkt_gettoken(&p->p_token);
293
294                 sunptr = &suptr->un_ent[0];
295                 i = 0;
296
297                 while (i < suptr->un_cnt) {
298                         if (sunptr->un_id == semid) {
299                                 if (semnum == -1 || sunptr->un_num == semnum) {
300                                         suptr->un_cnt--;
301                                         if (i < suptr->un_cnt) {
302                                                 suptr->un_ent[i] =
303                                                   suptr->un_ent[suptr->un_cnt];
304                                                 /*
305                                                  * do not increment i
306                                                  * or sunptr after copydown.
307                                                  */
308                                                 continue;
309                                         }
310                                 }
311                                 if (semnum != -1)
312                                         break;
313                         }
314                         ++i;
315                         ++sunptr;
316                 }
317
318                 lwkt_reltoken(&p->p_token);
319                 PRELE(p);
320
321                 /*
322                  * Handle deletion races
323                  */
324                 sunext = TAILQ_NEXT(suptr, un_entry);
325                 if (--suptr->un_refs == 0 && suptr->un_proc == NULL) {
326                         KKASSERT(suptr->un_cnt == 0);
327                         TAILQ_REMOVE(&semu_list, suptr, un_entry);
328                         kfree(suptr, M_SEM);
329                 }
330         }
331         lwkt_reltoken(&semu_token);
332 }
333
334 /*
335  * Note that the user-mode half of this passes a union, not a pointer
336  *
337  * MPALMOSTSAFE
338  */
339 int
340 sys___semctl(struct __semctl_args *uap)
341 {
342         struct thread *td = curthread;
343         struct prison *pr = td->td_proc->p_ucred->cr_prison;
344         int semid = uap->semid;
345         int semnum = uap->semnum;
346         int cmd = uap->cmd;
347         union semun *arg = uap->arg;
348         union semun real_arg;
349         struct ucred *cred = td->td_ucred;
350         int i, rval, eval;
351         struct semid_ds sbuf;
352         struct semid_pool *semaptr;
353         struct sem *semptr;
354
355 #ifdef SEM_DEBUG
356         kprintf("call to semctl(%d, %d, %d, 0x%x)\n", semid, semnum, cmd, arg);
357 #endif
358
359         if (pr && !pr->pr_sysvipc_allowed)
360                 return (ENOSYS);
361
362         semid = IPCID_TO_IX(semid);
363         if (semid < 0 || semid >= seminfo.semmni) {
364                 return(EINVAL);
365         }
366         semaptr = &sema[semid];
367         lockmgr(&semaptr->lk, LK_EXCLUSIVE);
368
369         if ((semaptr->ds.sem_perm.mode & SEM_ALLOC) == 0 ||
370             semaptr->ds.sem_perm.seq != IPCID_TO_SEQ(uap->semid)) {
371                 lockmgr(&semaptr->lk, LK_RELEASE);
372                 return(EINVAL);
373         }
374
375         eval = 0;
376         rval = 0;
377
378         switch (cmd) {
379         case IPC_RMID:
380                 eval = ipcperm(td->td_proc, &semaptr->ds.sem_perm, IPC_M);
381                 if (eval != 0)
382                         break;
383                 semaptr->ds.sem_perm.cuid = cred->cr_uid;
384                 semaptr->ds.sem_perm.uid = cred->cr_uid;
385
386                 /*
387                  * NOTE: Nobody will be waiting on the semaphores since
388                  *       we have an exclusive lock on semaptr->lk).
389                  */
390                 lockmgr(&sema_lk, LK_EXCLUSIVE);
391                 semtot -= semaptr->ds.sem_nsems;
392                 kfree(semaptr->ds.sem_base, M_SEM);
393                 semaptr->ds.sem_base = NULL;
394                 semaptr->ds.sem_perm.mode = 0;  /* clears SEM_ALLOC */
395                 lockmgr(&sema_lk, LK_RELEASE);
396
397                 semundo_clear(semid, -1);
398                 break;
399
400         case IPC_SET:
401                 eval = ipcperm(td->td_proc, &semaptr->ds.sem_perm, IPC_M);
402                 if (eval)
403                         break;
404                 if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
405                         break;
406                 if ((eval = copyin(real_arg.buf, (caddr_t)&sbuf,
407                                    sizeof(sbuf))) != 0) {
408                         break;
409                 }
410                 semaptr->ds.sem_perm.uid = sbuf.sem_perm.uid;
411                 semaptr->ds.sem_perm.gid = sbuf.sem_perm.gid;
412                 semaptr->ds.sem_perm.mode =
413                         (semaptr->ds.sem_perm.mode & ~0777) |
414                         (sbuf.sem_perm.mode & 0777);
415                 semaptr->ds.sem_ctime = time_second;
416                 break;
417
418         case IPC_STAT:
419                 eval = ipcperm(td->td_proc, &semaptr->ds.sem_perm, IPC_R);
420                 if (eval)
421                         break;
422                 if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
423                         break;
424                 eval = copyout(&semaptr->ds, real_arg.buf,
425                                sizeof(struct semid_ds));
426                 break;
427         case SEM_STAT:
428                 /*
429                  * For this command we assume semid is an array index
430                  * rather than an IPC id.  However, the conversion is
431                  * just a mask so just validate that the passed-in semid
432                  * matches the masked semid.
433                  */
434                 if (uap->semid != semid) {
435                         eval = EINVAL;
436                         break;
437                 }
438                 eval = ipcperm(td->td_proc, &semaptr->ds.sem_perm, IPC_R);
439                 if (eval)
440                         break;
441                 if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
442                         break;
443                 eval = copyout(&semaptr->ds, real_arg.buf,
444                                sizeof(struct semid_ds));
445                 rval = IXSEQ_TO_IPCID(semid, semaptr->ds.sem_perm);
446                 break;
447
448         case GETNCNT:
449                 eval = ipcperm(td->td_proc, &semaptr->ds.sem_perm, IPC_R);
450                 if (eval)
451                         break;
452                 if (semnum < 0 || semnum >= semaptr->ds.sem_nsems) {
453                         eval = EINVAL;
454                         break;
455                 }
456                 rval = semaptr->ds.sem_base[semnum].semncnt;
457                 break;
458
459         case GETPID:
460                 eval = ipcperm(td->td_proc, &semaptr->ds.sem_perm, IPC_R);
461                 if (eval)
462                         break;
463                 if (semnum < 0 || semnum >= semaptr->ds.sem_nsems) {
464                         eval = EINVAL;
465                         break;
466                 }
467                 rval = semaptr->ds.sem_base[semnum].sempid;
468                 break;
469
470         case GETVAL:
471                 eval = ipcperm(td->td_proc, &semaptr->ds.sem_perm, IPC_R);
472                 if (eval)
473                         break;
474                 if (semnum < 0 || semnum >= semaptr->ds.sem_nsems) {
475                         eval = EINVAL;
476                         break;
477                 }
478                 rval = semaptr->ds.sem_base[semnum].semval;
479                 break;
480
481         case GETALL:
482                 eval = ipcperm(td->td_proc, &semaptr->ds.sem_perm, IPC_R);
483                 if (eval)
484                         break;
485                 if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
486                         break;
487                 for (i = 0; i < semaptr->ds.sem_nsems; i++) {
488                         eval = copyout(&semaptr->ds.sem_base[i].semval,
489                                        &real_arg.array[i],
490                                        sizeof(real_arg.array[0]));
491                         if (eval)
492                                 break;
493                 }
494                 break;
495
496         case GETZCNT:
497                 eval = ipcperm(td->td_proc, &semaptr->ds.sem_perm, IPC_R);
498                 if (eval)
499                         break;
500                 if (semnum < 0 || semnum >= semaptr->ds.sem_nsems) {
501                         eval = EINVAL;
502                         break;
503                 }
504                 rval = semaptr->ds.sem_base[semnum].semzcnt;
505                 break;
506
507         case SETVAL:
508                 eval = ipcperm(td->td_proc, &semaptr->ds.sem_perm, IPC_W);
509                 if (eval)
510                         break;
511                 if (semnum < 0 || semnum >= semaptr->ds.sem_nsems) {
512                         eval = EINVAL;
513                         break;
514                 }
515                 if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
516                         break;
517
518                 /*
519                  * Because we hold semaptr->lk exclusively we can safely
520                  * modify any semptr content without acquiring its token.
521                  */
522                 semptr = &semaptr->ds.sem_base[semnum];
523                 semptr->semval = real_arg.val;
524                 semundo_clear(semid, semnum);
525                 if (semptr->semzcnt || semptr->semncnt)
526                         wakeup(semptr);
527                 break;
528
529         case SETALL:
530                 eval = ipcperm(td->td_proc, &semaptr->ds.sem_perm, IPC_W);
531                 if (eval)
532                         break;
533                 if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
534                         break;
535                 /*
536                  * Because we hold semaptr->lk exclusively we can safely
537                  * modify any semptr content without acquiring its token.
538                  */
539                 for (i = 0; i < semaptr->ds.sem_nsems; i++) {
540                         semptr = &semaptr->ds.sem_base[i];
541                         eval = copyin(&real_arg.array[i],
542                                       (caddr_t)&semptr->semval,
543                                       sizeof(real_arg.array[0]));
544                         if (semptr->semzcnt || semptr->semncnt)
545                                 wakeup(semptr);
546                         if (eval != 0)
547                                 break;
548                 }
549                 semundo_clear(semid, -1);
550                 break;
551
552         default:
553                 eval = EINVAL;
554                 break;
555         }
556         lockmgr(&semaptr->lk, LK_RELEASE);
557
558         if (eval == 0)
559                 uap->sysmsg_result = rval;
560         return(eval);
561 }
562
563 /*
564  * MPALMOSTSAFE
565  */
566 int
567 sys_semget(struct semget_args *uap)
568 {
569         struct thread *td = curthread;
570         struct prison *pr = td->td_proc->p_ucred->cr_prison;
571         int semid, eval;
572         int key = uap->key;
573         int nsems = uap->nsems;
574         int semflg = uap->semflg;
575         struct ucred *cred = td->td_ucred;
576
577 #ifdef SEM_DEBUG
578         kprintf("semget(0x%x, %d, 0%o)\n", key, nsems, semflg);
579 #endif
580
581         if (pr && !pr->pr_sysvipc_allowed)
582                 return (ENOSYS);
583
584         eval = 0;
585
586         if (key != IPC_PRIVATE) {
587                 for (semid = 0; semid < seminfo.semmni; semid++) {
588                         if ((sema[semid].ds.sem_perm.mode & SEM_ALLOC) == 0 ||
589                             sema[semid].ds.sem_perm.key != key) {
590                                 continue;
591                         }
592                         lockmgr(&sema[semid].lk, LK_EXCLUSIVE);
593                         if ((sema[semid].ds.sem_perm.mode & SEM_ALLOC) == 0 ||
594                             sema[semid].ds.sem_perm.key != key) {
595                                 lockmgr(&sema[semid].lk, LK_RELEASE);
596                                 continue;
597                         }
598                         break;
599                 }
600                 if (semid < seminfo.semmni) {
601                         /* sema[semid].lk still locked from above */
602 #ifdef SEM_DEBUG
603                         kprintf("found public key\n");
604 #endif
605                         if ((eval = ipcperm(td->td_proc,
606                                             &sema[semid].ds.sem_perm,
607                                             semflg & 0700))) {
608                                 lockmgr(&sema[semid].lk, LK_RELEASE);
609                                 goto done;
610                         }
611                         if (nsems > 0 && sema[semid].ds.sem_nsems < nsems) {
612 #ifdef SEM_DEBUG
613                                 kprintf("too small\n");
614 #endif
615                                 eval = EINVAL;
616                                 lockmgr(&sema[semid].lk, LK_RELEASE);
617                                 goto done;
618                         }
619                         if ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) {
620 #ifdef SEM_DEBUG
621                                 kprintf("not exclusive\n");
622 #endif
623                                 eval = EEXIST;
624                                 lockmgr(&sema[semid].lk, LK_RELEASE);
625                                 goto done;
626                         }
627
628                         /*
629                          * Return this one.
630                          */
631                         lockmgr(&sema[semid].lk, LK_RELEASE);
632                         goto done;
633                 }
634         }
635
636 #ifdef SEM_DEBUG
637         kprintf("need to allocate the semid_ds\n");
638 #endif
639         if (key == IPC_PRIVATE || (semflg & IPC_CREAT)) {
640                 if (nsems <= 0 || nsems > seminfo.semmsl) {
641 #ifdef SEM_DEBUG
642                         kprintf("nsems out of range (0<%d<=%d)\n",
643                                 nsems, seminfo.semmsl);
644 #endif
645                         eval = EINVAL;
646                         goto done;
647                 }
648
649                 /*
650                  * SEM_ALLOC flag cannot be set unless sema_lk is locked.
651                  * semtot field also protected by sema_lk.
652                  */
653                 lockmgr(&sema_lk, LK_EXCLUSIVE);
654                 if (nsems > seminfo.semmns - semtot) {
655 #ifdef SEM_DEBUG
656                         kprintf("not enough semaphores left "
657                                 "(need %d, got %d)\n",
658                                 nsems, seminfo.semmns - semtot);
659 #endif
660                         eval = ENOSPC;
661                         lockmgr(&sema_lk, LK_RELEASE);
662                         goto done;
663                 }
664                 for (semid = 0; semid < seminfo.semmni; semid++) {
665                         if ((sema[semid].ds.sem_perm.mode & SEM_ALLOC) == 0)
666                                 break;
667                 }
668                 if (semid == seminfo.semmni) {
669 #ifdef SEM_DEBUG
670                         kprintf("no more semid_ds's available\n");
671 #endif
672                         eval = ENOSPC;
673                         lockmgr(&sema_lk, LK_RELEASE);
674                         goto done;
675                 }
676 #ifdef SEM_DEBUG
677                 kprintf("semid %d is available\n", semid);
678 #endif
679                 lockmgr(&sema[semid].lk, LK_EXCLUSIVE);
680                 sema[semid].ds.sem_perm.key = key;
681                 sema[semid].ds.sem_perm.cuid = cred->cr_uid;
682                 sema[semid].ds.sem_perm.uid = cred->cr_uid;
683                 sema[semid].ds.sem_perm.cgid = cred->cr_gid;
684                 sema[semid].ds.sem_perm.gid = cred->cr_gid;
685                 sema[semid].ds.sem_perm.mode = (semflg & 0777) | SEM_ALLOC;
686                 sema[semid].ds.sem_perm.seq =
687                     (sema[semid].ds.sem_perm.seq + 1) & 0x7fff;
688                 sema[semid].ds.sem_nsems = nsems;
689                 sema[semid].ds.sem_otime = 0;
690                 sema[semid].ds.sem_ctime = time_second;
691                 sema[semid].ds.sem_base = kmalloc(sizeof(struct sem) * nsems,
692                                                M_SEM, M_WAITOK|M_ZERO);
693                 semtot += nsems;
694                 ++sema[semid].gen;
695                 lockmgr(&sema[semid].lk, LK_RELEASE);
696                 lockmgr(&sema_lk, LK_RELEASE);
697 #ifdef SEM_DEBUG
698                 kprintf("sembase = 0x%x, next = 0x%x\n",
699                         sema[semid].ds.sem_base, &sem[semtot]);
700 #endif
701                 /* eval == 0 */
702         } else {
703 #ifdef SEM_DEBUG
704                 kprintf("didn't find it and wasn't asked to create it\n");
705 #endif
706                 eval = ENOENT;
707         }
708
709 done:
710         if (eval == 0) {
711                 uap->sysmsg_result =
712                         IXSEQ_TO_IPCID(semid, sema[semid].ds.sem_perm);
713         }
714         return(eval);
715 }
716
717 /*
718  * MPSAFE
719  */
720 int
721 sys_semop(struct semop_args *uap)
722 {
723         struct thread *td = curthread;
724         struct prison *pr = td->td_proc->p_ucred->cr_prison;
725         int semid = uap->semid;
726         u_int nsops = uap->nsops;
727         struct sembuf sops[MAX_SOPS];
728         struct semid_pool *semaptr;
729         struct sembuf *sopptr;
730         struct sem *semptr;
731         struct sem *xsemptr;
732         int i, j, eval;
733         int do_undos;
734
735 #ifdef SEM_DEBUG
736         kprintf("call to semop(%d, 0x%x, %u)\n", semid, sops, nsops);
737 #endif
738         if (pr && !pr->pr_sysvipc_allowed)
739                 return (ENOSYS);
740
741         semid = IPCID_TO_IX(semid);     /* Convert back to zero origin */
742
743         if (semid < 0 || semid >= seminfo.semmni) {
744                 eval = EINVAL;
745                 goto done2;
746         }
747
748         wakeup_start_delayed();
749         semaptr = &sema[semid];
750         lockmgr(&semaptr->lk, LK_SHARED);
751
752         if ((semaptr->ds.sem_perm.mode & SEM_ALLOC) == 0) {
753                 eval = EINVAL;
754                 goto done;
755         }
756         if (semaptr->ds.sem_perm.seq != IPCID_TO_SEQ(uap->semid)) {
757                 eval = EINVAL;
758                 goto done;
759         }
760
761         if ((eval = ipcperm(td->td_proc, &semaptr->ds.sem_perm, IPC_W))) {
762 #ifdef SEM_DEBUG
763                 kprintf("eval = %d from ipaccess\n", eval);
764 #endif
765                 goto done;
766         }
767
768         if (nsops > MAX_SOPS) {
769 #ifdef SEM_DEBUG
770                 kprintf("too many sops (max=%d, nsops=%u)\n", MAX_SOPS, nsops);
771 #endif
772                 eval = E2BIG;
773                 goto done;
774         }
775
776         if ((eval = copyin(uap->sops, &sops, nsops * sizeof(sops[0]))) != 0) {
777 #ifdef SEM_DEBUG
778                 kprintf("eval = %d from copyin(%08x, %08x, %u)\n", eval,
779                     uap->sops, &sops, nsops * sizeof(sops[0]));
780 #endif
781                 goto done;
782         }
783
784         /*
785          * Loop trying to satisfy the vector of requests.
786          * If we reach a point where we must wait, any requests already
787          * performed are rolled back and we go to sleep until some other
788          * process wakes us up.  At this point, we start all over again.
789          *
790          * This ensures that from the perspective of other tasks, a set
791          * of requests is atomic (never partially satisfied).
792          */
793         do_undos = 0;
794
795         for (;;) {
796                 long gen;
797
798                 semptr = NULL;
799
800                 for (i = 0; i < nsops; i++) {
801                         sopptr = &sops[i];
802
803                         if (sopptr->sem_num >= semaptr->ds.sem_nsems) {
804                                 eval = EFBIG;
805                                 goto done;
806                         }
807
808                         semptr = &semaptr->ds.sem_base[sopptr->sem_num];
809                         lwkt_getpooltoken(semptr);
810
811 #ifdef SEM_DEBUG
812                         kprintf("semop:  semaptr=%x, sem_base=%x, semptr=%x, "
813                                 "sem[%d]=%d : op=%d, flag=%s\n",
814                             semaptr, semaptr->ds.sem_base, semptr,
815                             sopptr->sem_num, semptr->semval, sopptr->sem_op,
816                             (sopptr->sem_flg & IPC_NOWAIT) ? "nowait" : "wait");
817 #endif
818
819                         if (sopptr->sem_op < 0) {
820                                 if (semptr->semval + sopptr->sem_op < 0) {
821 #ifdef SEM_DEBUG
822                                         kprintf("semop:  can't do it now\n");
823 #endif
824                                         break;
825                                 } else {
826                                         semptr->semval += sopptr->sem_op;
827                                         if (semptr->semval == 0 &&
828                                             semptr->semzcnt > 0) {
829                                                 wakeup(semptr);
830                                         }
831                                 }
832                                 if (sopptr->sem_flg & SEM_UNDO)
833                                         do_undos = 1;
834                         } else if (sopptr->sem_op == 0) {
835                                 if (semptr->semval > 0) {
836 #ifdef SEM_DEBUG
837                                         kprintf("semop:  not zero now\n");
838 #endif
839                                         break;
840                                 }
841                         } else {
842                                 semptr->semval += sopptr->sem_op;
843                                 if (sopptr->sem_flg & SEM_UNDO)
844                                         do_undos = 1;
845                                 if (semptr->semncnt > 0)
846                                         wakeup(semptr);
847                         }
848                         lwkt_relpooltoken(semptr);
849                 }
850
851                 /*
852                  * Did we get through the entire vector?
853                  */
854                 if (i >= nsops)
855                         goto donex;
856
857                 /*
858                  * No, protect the semaphore request which also flags that
859                  * a wakeup is needed, then release semptr since we know
860                  * another process is likely going to need to access it
861                  * soon.
862                  */
863                 if (sopptr->sem_op == 0)
864                         semptr->semzcnt++;
865                 else
866                         semptr->semncnt++;
867                 tsleep_interlock(semptr, PCATCH);
868                 lwkt_relpooltoken(semptr);
869
870                 /*
871                  * Rollback the semaphores we had acquired.
872                  */
873 #ifdef SEM_DEBUG
874                 kprintf("semop:  rollback 0 through %d\n", i-1);
875 #endif
876                 for (j = 0; j < i; j++) {
877                         xsemptr = &semaptr->ds.sem_base[sops[j].sem_num];
878                         lwkt_getpooltoken(xsemptr);
879                         xsemptr->semval -= sops[j].sem_op;
880                         if (xsemptr->semval == 0 && xsemptr->semzcnt > 0)
881                                 wakeup(xsemptr);
882                         if (xsemptr->semval <= 0 && xsemptr->semncnt > 0)
883                                 wakeup(xsemptr);
884                         lwkt_relpooltoken(xsemptr);
885                 }
886
887                 /*
888                  * If the request that we couldn't satisfy has the
889                  * NOWAIT flag set then return with EAGAIN.
890                  */
891                 if (sopptr->sem_flg & IPC_NOWAIT) {
892                         eval = EAGAIN;
893                         goto done;
894                 }
895
896                 /*
897                  * Release semaptr->lk while sleeping, allowing other
898                  * semops (like SETVAL, SETALL, etc), which require an
899                  * exclusive lock and might wake us up.
900                  *
901                  * Reload and recheck the validity of semaptr on return.
902                  * Note that semptr itself might have changed too, but
903                  * we've already interlocked for semptr and that is what
904                  * will be woken up if it wakes up the tsleep on a MP
905                  * race.
906                  *
907                  * gen protects against destroy/re-create races where the
908                  * creds match.
909                  */
910 #ifdef SEM_DEBUG
911                 kprintf("semop:  good night!\n");
912 #endif
913                 gen = semaptr->gen;
914                 lockmgr(&semaptr->lk, LK_RELEASE);
915                 eval = tsleep(semptr, PCATCH | PINTERLOCKED, "semwait", hz);
916                 lockmgr(&semaptr->lk, LK_SHARED);
917 #ifdef SEM_DEBUG
918                 kprintf("semop:  good morning (eval=%d)!\n", eval);
919 #endif
920
921                 /* return code is checked below, after sem[nz]cnt-- */
922
923                 /*
924                  * Make sure that the semaphore still exists
925                  */
926                 if (semaptr->gen != gen ||
927                     (semaptr->ds.sem_perm.mode & SEM_ALLOC) == 0 ||
928                     semaptr->ds.sem_perm.seq != IPCID_TO_SEQ(uap->semid)) {
929                         eval = EIDRM;
930                         goto done;
931                 }
932
933                 /*
934                  * The semaphore is still alive.  Readjust the count of
935                  * waiting processes.
936                  */
937                 semptr = &semaptr->ds.sem_base[sopptr->sem_num];
938                 lwkt_getpooltoken(semptr);
939                 if (sopptr->sem_op == 0)
940                         semptr->semzcnt--;
941                 else
942                         semptr->semncnt--;
943                 lwkt_relpooltoken(semptr);
944
945                 /*
946                  * Is it really morning, or was our sleep interrupted?
947                  * (Delayed check of tsleep() return code because we
948                  * need to decrement sem[nz]cnt either way.)
949                  */
950                 if (eval) {
951                         eval = EINTR;
952                         goto done;
953                 }
954 #ifdef SEM_DEBUG
955                 kprintf("semop:  good morning!\n");
956 #endif
957                 /* RETRY LOOP */
958         }
959
960 donex:
961         /*
962          * Process any SEM_UNDO requests.
963          */
964         if (do_undos) {
965                 for (i = 0; i < nsops; i++) {
966                         /*
967                          * We only need to deal with SEM_UNDO's for non-zero
968                          * op's.
969                          */
970                         int adjval;
971
972                         if ((sops[i].sem_flg & SEM_UNDO) == 0)
973                                 continue;
974                         adjval = sops[i].sem_op;
975                         if (adjval == 0)
976                                 continue;
977                         eval = semundo_adjust(td->td_proc, semid,
978                                               sops[i].sem_num, -adjval);
979                         if (eval == 0)
980                                 continue;
981
982                         /*
983                          * Oh-Oh!  We ran out of either sem_undo's or undo's.
984                          * Rollback the adjustments to this point and then
985                          * rollback the semaphore ups and down so we can return
986                          * with an error with all structures restored.  We
987                          * rollback the undo's in the exact reverse order that
988                          * we applied them.  This guarantees that we won't run
989                          * out of space as we roll things back out.
990                          */
991                         for (j = i - 1; j >= 0; j--) {
992                                 if ((sops[j].sem_flg & SEM_UNDO) == 0)
993                                         continue;
994                                 adjval = sops[j].sem_op;
995                                 if (adjval == 0)
996                                         continue;
997                                 if (semundo_adjust(td->td_proc, semid,
998                                                sops[j].sem_num, adjval) != 0)
999                                         panic("semop - can't undo undos");
1000                         }
1001
1002                         for (j = 0; j < nsops; j++) {
1003                                 xsemptr = &semaptr->ds.sem_base[
1004                                                         sops[j].sem_num];
1005                                 lwkt_getpooltoken(xsemptr);
1006                                 xsemptr->semval -= sops[j].sem_op;
1007                                 if (xsemptr->semval == 0 &&
1008                                     xsemptr->semzcnt > 0)
1009                                         wakeup(xsemptr);
1010                                 if (xsemptr->semval <= 0 &&
1011                                     xsemptr->semncnt > 0)
1012                                         wakeup(xsemptr);
1013                                 lwkt_relpooltoken(xsemptr);
1014                         }
1015
1016 #ifdef SEM_DEBUG
1017                         kprintf("eval = %d from semundo_adjust\n", eval);
1018 #endif
1019                         goto done;
1020                 } /* loop through the sops */
1021         } /* if (do_undos) */
1022
1023         /* We're definitely done - set the sempid's */
1024         for (i = 0; i < nsops; i++) {
1025                 sopptr = &sops[i];
1026                 semptr = &semaptr->ds.sem_base[sopptr->sem_num];
1027                 lwkt_getpooltoken(semptr);
1028                 semptr->sempid = td->td_proc->p_pid;
1029                 lwkt_relpooltoken(semptr);
1030         }
1031
1032         /* Do a wakeup if any semaphore was up'd. */
1033 #ifdef SEM_DEBUG
1034         kprintf("semop:  done\n");
1035 #endif
1036         uap->sysmsg_result = 0;
1037         eval = 0;
1038 done:
1039         lockmgr(&semaptr->lk, LK_RELEASE);
1040         wakeup_end_delayed();
1041 done2:
1042         return(eval);
1043 }
1044
1045 /*
1046  * Go through the undo structures for this process and apply the adjustments to
1047  * semaphores.
1048  *
1049  * (p->p_token is held by the caller)
1050  */
1051 void
1052 semexit(struct proc *p)
1053 {
1054         struct sem_undo *suptr;
1055         struct sem *semptr;
1056
1057         /*
1058          * We're getting a global token, don't do it if we couldn't
1059          * possibly have any semaphores.
1060          */
1061         if ((p->p_flags & P_SYSVSEM) == 0)
1062                 return;
1063         suptr = p->p_sem_undo;
1064         KKASSERT(suptr != NULL);
1065
1066         /*
1067          * Disconnect suptr from the process and increment un_refs to
1068          * prevent anyone else from being able to destroy the structure.
1069          * Do not remove it from the linked list until after we are through
1070          * scanning it as other semaphore calls might still effect it.
1071          */
1072         lwkt_gettoken(&semu_token);
1073         p->p_sem_undo = NULL;
1074         p->p_flags &= ~P_SYSVSEM;
1075         suptr->un_proc = NULL;
1076         ++suptr->un_refs;
1077         lwkt_reltoken(&semu_token);
1078
1079         while (suptr->un_cnt) {
1080                 struct semid_pool *semaptr;
1081                 int semid;
1082                 int semnum;
1083                 int adjval;
1084                 int ix;
1085
1086                 /*
1087                  * These values are stable because we hold p->p_token.
1088                  * However, they can get ripped out from under us when
1089                  * we block or obtain other tokens so we have to re-check.
1090                  */
1091                 ix = suptr->un_cnt - 1;
1092                 semid = suptr->un_ent[ix].un_id;
1093                 semnum = suptr->un_ent[ix].un_num;
1094                 adjval = suptr->un_ent[ix].un_adjval;
1095
1096                 semaptr = &sema[semid];
1097
1098                 /*
1099                  * Recheck after locking, then execute the undo
1100                  * operation.  semptr remains valid due to the
1101                  * semaptr->lk.
1102                  */
1103                 lockmgr(&semaptr->lk, LK_SHARED);
1104                 semptr = &semaptr->ds.sem_base[semnum];
1105                 lwkt_getpooltoken(semptr);
1106
1107                 if (ix == suptr->un_cnt - 1 &&
1108                     semid == suptr->un_ent[ix].un_id &&
1109                     semnum == suptr->un_ent[ix].un_num &&
1110                     adjval == suptr->un_ent[ix].un_adjval) {
1111                         /*
1112                          * Only do assertions when we aren't in a SMP race.
1113                          */
1114                         if ((semaptr->ds.sem_perm.mode & SEM_ALLOC) == 0)
1115                                 panic("semexit - semid not allocated");
1116                         if (semnum >= semaptr->ds.sem_nsems)
1117                                 panic("semexit - semnum out of range");
1118                         --suptr->un_cnt;
1119
1120                         if (adjval < 0) {
1121                                 if (semptr->semval < -adjval)
1122                                         semptr->semval = 0;
1123                                 else
1124                                         semptr->semval += adjval;
1125                         } else {
1126                                 semptr->semval += adjval;
1127                         }
1128                         wakeup(semptr);
1129                 }
1130                 lwkt_relpooltoken(semptr);
1131                 lockmgr(&semaptr->lk, LK_RELEASE);
1132         }
1133
1134         /*
1135          * Final cleanup, remove from the list and deallocate on the
1136          * last ref only.
1137          */
1138         lwkt_gettoken(&semu_token);
1139         if (--suptr->un_refs == 0) {
1140                 TAILQ_REMOVE(&semu_list, suptr, un_entry);
1141                 KKASSERT(suptr->un_cnt == 0);
1142                 kfree(suptr, M_SEM);
1143         }
1144         lwkt_reltoken(&semu_token);
1145 }