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/sysmsg.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                 /*
 295                  * Check for semexit() race
 296                  */
 297                 if (p->p_sem_undo != suptr)
 298                         goto skip;
 299
 300                 sunptr = &suptr->un_ent[0];
 301                 i = 0;
 302
 303                 while (i < suptr->un_cnt) {
 304                         if (sunptr->un_id == semid) {
 305                                 if (semnum == -1 || sunptr->un_num == semnum) {
 306                                         suptr->un_cnt--;
 307                                         if (i < suptr->un_cnt) {
 308                                                 suptr->un_ent[i] =
 309                                                   suptr->un_ent[suptr->un_cnt];
 310                                                 /*
 311                                                  * do not increment i
 312                                                  * or sunptr after copydown.
 313                                                  */
 314                                                 continue;
 315                                         }
 316                                 }
 317                                 if (semnum != -1)
 318                                         break;
 319                         }
 320                         ++i;
 321                         ++sunptr;
 322                 }
 323
 324 skip:
 325                 lwkt_reltoken(&p->p_token);
 326                 PRELE(p);
 327
 328                 /*
 329                  * Handle deletion and semexit races
 330                  */
 331                 sunext = TAILQ_NEXT(suptr, un_entry);
 332                 if (--suptr->un_refs == 0 && suptr->un_proc == NULL) {
 333                         KKASSERT(suptr->un_cnt == 0);
 334                         TAILQ_REMOVE(&semu_list, suptr, un_entry);
 335                         kfree(suptr, M_SEM);
 336                 }
 337         }
 338         lwkt_reltoken(&semu_token);
 339 }
 340
 341 /*
 342  * Note that the user-mode half of this passes a union, not a pointer
 343  *
 344  * MPALMOSTSAFE
 345  */
 346 int
 347 sys___semctl(struct sysmsg *sysmsg, const struct __semctl_args *uap)
 348 {
 349         struct thread *td = curthread;
 350         struct prison *pr = td->td_proc->p_ucred->cr_prison;
 351         int semid = uap->semid;
 352         int semnum = uap->semnum;
 353         int cmd = uap->cmd;
 354         union semun *arg = uap->arg;
 355         union semun real_arg;
 356         struct ucred *cred = td->td_ucred;
 357         int i, rval, eval;
 358         struct semid_ds sbuf;
 359         struct semid_pool *semaptr;
 360         struct sem *semptr;
 361
 362 #ifdef SEM_DEBUG
 363         kprintf("call to semctl(%d, %d, %d, 0x%x)\n", semid, semnum, cmd, arg);
 364 #endif
 365
 366         if (pr && !PRISON_CAP_ISSET(pr->pr_caps, PRISON_CAP_SYS_SYSVIPC))
 367                 return (ENOSYS);
 368
 369         semid = IPCID_TO_IX(semid);
 370         if (semid < 0 || semid >= seminfo.semmni) {
 371                 return(EINVAL);
 372         }
 373         semaptr = &sema[semid];
 374         lockmgr(&semaptr->lk, LK_EXCLUSIVE);
 375
 376         if ((semaptr->ds.sem_perm.mode & SEM_ALLOC) == 0 ||
 377             semaptr->ds.sem_perm.seq != IPCID_TO_SEQ(uap->semid)) {
 378                 lockmgr(&semaptr->lk, LK_RELEASE);
 379                 return(EINVAL);
 380         }
 381
 382         eval = 0;
 383         rval = 0;
 384
 385         switch (cmd) {
 386         case IPC_RMID:
 387                 eval = ipcperm(td->td_proc, &semaptr->ds.sem_perm, IPC_M);
 388                 if (eval != 0)
 389                         break;
 390                 semaptr->ds.sem_perm.cuid = cred->cr_uid;
 391                 semaptr->ds.sem_perm.uid = cred->cr_uid;
 392
 393                 /*
 394                  * NOTE: Nobody will be waiting on the semaphores since
 395                  *       we have an exclusive lock on semaptr->lk).
 396                  */
 397                 lockmgr(&sema_lk, LK_EXCLUSIVE);
 398                 semtot -= semaptr->ds.sem_nsems;
 399                 kfree(semaptr->ds.sem_base, M_SEM);
 400                 semaptr->ds.sem_base = NULL;
 401                 semaptr->ds.sem_perm.mode = 0;  /* clears SEM_ALLOC */
 402                 lockmgr(&sema_lk, LK_RELEASE);
 403
 404                 semundo_clear(semid, -1);
 405                 break;
 406
 407         case IPC_SET:
 408                 eval = ipcperm(td->td_proc, &semaptr->ds.sem_perm, IPC_M);
 409                 if (eval)
 410                         break;
 411                 if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
 412                         break;
 413                 if ((eval = copyin(real_arg.buf, (caddr_t)&sbuf,
 414                                    sizeof(sbuf))) != 0) {
 415                         break;
 416                 }
 417                 semaptr->ds.sem_perm.uid = sbuf.sem_perm.uid;
 418                 semaptr->ds.sem_perm.gid = sbuf.sem_perm.gid;
 419                 semaptr->ds.sem_perm.mode =
 420                         (semaptr->ds.sem_perm.mode & ~0777) |
 421                         (sbuf.sem_perm.mode & 0777);
 422                 semaptr->ds.sem_ctime = time_second;
 423                 break;
 424
 425         case IPC_STAT:
 426                 eval = ipcperm(td->td_proc, &semaptr->ds.sem_perm, IPC_R);
 427                 if (eval)
 428                         break;
 429                 if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
 430                         break;
 431                 eval = copyout(&semaptr->ds, real_arg.buf,
 432                                sizeof(struct semid_ds));
 433                 break;
 434         case SEM_STAT:
 435                 /*
 436                  * For this command we assume semid is an array index
 437                  * rather than an IPC id.  However, the conversion is
 438                  * just a mask so just validate that the passed-in semid
 439                  * matches the masked semid.
 440                  */
 441                 if (uap->semid != semid) {
 442                         eval = EINVAL;
 443                         break;
 444                 }
 445                 eval = ipcperm(td->td_proc, &semaptr->ds.sem_perm, IPC_R);
 446                 if (eval)
 447                         break;
 448                 if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
 449                         break;
 450                 eval = copyout(&semaptr->ds, real_arg.buf,
 451                                sizeof(struct semid_ds));
 452                 rval = IXSEQ_TO_IPCID(semid, semaptr->ds.sem_perm);
 453                 break;
 454
 455         case GETNCNT:
 456                 eval = ipcperm(td->td_proc, &semaptr->ds.sem_perm, IPC_R);
 457                 if (eval)
 458                         break;
 459                 if (semnum < 0 || semnum >= semaptr->ds.sem_nsems) {
 460                         eval = EINVAL;
 461                         break;
 462                 }
 463                 rval = semaptr->ds.sem_base[semnum].semncnt;
 464                 break;
 465
 466         case GETPID:
 467                 eval = ipcperm(td->td_proc, &semaptr->ds.sem_perm, IPC_R);
 468                 if (eval)
 469                         break;
 470                 if (semnum < 0 || semnum >= semaptr->ds.sem_nsems) {
 471                         eval = EINVAL;
 472                         break;
 473                 }
 474                 rval = semaptr->ds.sem_base[semnum].sempid;
 475                 break;
 476
 477         case GETVAL:
 478                 eval = ipcperm(td->td_proc, &semaptr->ds.sem_perm, IPC_R);
 479                 if (eval)
 480                         break;
 481                 if (semnum < 0 || semnum >= semaptr->ds.sem_nsems) {
 482                         eval = EINVAL;
 483                         break;
 484                 }
 485                 rval = semaptr->ds.sem_base[semnum].semval;
 486                 break;
 487
 488         case GETALL:
 489                 eval = ipcperm(td->td_proc, &semaptr->ds.sem_perm, IPC_R);
 490                 if (eval)
 491                         break;
 492                 if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
 493                         break;
 494                 for (i = 0; i < semaptr->ds.sem_nsems; i++) {
 495                         eval = copyout(&semaptr->ds.sem_base[i].semval,
 496                                        &real_arg.array[i],
 497                                        sizeof(real_arg.array[0]));
 498                         if (eval)
 499                                 break;
 500                 }
 501                 break;
 502
 503         case GETZCNT:
 504                 eval = ipcperm(td->td_proc, &semaptr->ds.sem_perm, IPC_R);
 505                 if (eval)
 506                         break;
 507                 if (semnum < 0 || semnum >= semaptr->ds.sem_nsems) {
 508                         eval = EINVAL;
 509                         break;
 510                 }
 511                 rval = semaptr->ds.sem_base[semnum].semzcnt;
 512                 break;
 513
 514         case SETVAL:
 515                 eval = ipcperm(td->td_proc, &semaptr->ds.sem_perm, IPC_W);
 516                 if (eval)
 517                         break;
 518                 if (semnum < 0 || semnum >= semaptr->ds.sem_nsems) {
 519                         eval = EINVAL;
 520                         break;
 521                 }
 522                 if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
 523                         break;
 524
 525                 /*
 526                  * Because we hold semaptr->lk exclusively we can safely
 527                  * modify any semptr content without acquiring its token.
 528                  */
 529                 semptr = &semaptr->ds.sem_base[semnum];
 530                 semptr->semval = real_arg.val;
 531                 semundo_clear(semid, semnum);
 532                 if (semptr->semzcnt || semptr->semncnt)
 533                         wakeup(semptr);
 534                 break;
 535
 536         case SETALL:
 537                 eval = ipcperm(td->td_proc, &semaptr->ds.sem_perm, IPC_W);
 538                 if (eval)
 539                         break;
 540                 if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
 541                         break;
 542                 /*
 543                  * Because we hold semaptr->lk exclusively we can safely
 544                  * modify any semptr content without acquiring its token.
 545                  */
 546                 for (i = 0; i < semaptr->ds.sem_nsems; i++) {
 547                         semptr = &semaptr->ds.sem_base[i];
 548                         eval = copyin(&real_arg.array[i],
 549                                       (caddr_t)&semptr->semval,
 550                                       sizeof(real_arg.array[0]));
 551                         if (semptr->semzcnt || semptr->semncnt)
 552                                 wakeup(semptr);
 553                         if (eval != 0)
 554                                 break;
 555                 }
 556                 semundo_clear(semid, -1);
 557                 break;
 558
 559         default:
 560                 eval = EINVAL;
 561                 break;
 562         }
 563         lockmgr(&semaptr->lk, LK_RELEASE);
 564
 565         if (eval == 0)
 566                 sysmsg->sysmsg_result = rval;
 567         return(eval);
 568 }
 569
 570 /*
 571  * MPALMOSTSAFE
 572  */
 573 int
 574 sys_semget(struct sysmsg *sysmsg, const struct semget_args *uap)
 575 {
 576         struct thread *td = curthread;
 577         struct prison *pr = td->td_proc->p_ucred->cr_prison;
 578         int semid, eval;
 579         int key = uap->key;
 580         int nsems = uap->nsems;
 581         int semflg = uap->semflg;
 582         struct ucred *cred = td->td_ucred;
 583
 584 #ifdef SEM_DEBUG
 585         kprintf("semget(0x%x, %d, 0%o)\n", key, nsems, semflg);
 586 #endif
 587
 588         if (pr && !PRISON_CAP_ISSET(pr->pr_caps, PRISON_CAP_SYS_SYSVIPC))
 589                 return (ENOSYS);
 590
 591         eval = 0;
 592
 593         if (key != IPC_PRIVATE) {
 594                 for (semid = 0; semid < seminfo.semmni; semid++) {
 595                         if ((sema[semid].ds.sem_perm.mode & SEM_ALLOC) == 0 ||
 596                             sema[semid].ds.sem_perm.key != key) {
 597                                 continue;
 598                         }
 599                         lockmgr(&sema[semid].lk, LK_EXCLUSIVE);
 600                         if ((sema[semid].ds.sem_perm.mode & SEM_ALLOC) == 0 ||
 601                             sema[semid].ds.sem_perm.key != key) {
 602                                 lockmgr(&sema[semid].lk, LK_RELEASE);
 603                                 continue;
 604                         }
 605                         break;
 606                 }
 607                 if (semid < seminfo.semmni) {
 608                         /* sema[semid].lk still locked from above */
 609 #ifdef SEM_DEBUG
 610                         kprintf("found public key\n");
 611 #endif
 612                         if ((eval = ipcperm(td->td_proc,
 613                                             &sema[semid].ds.sem_perm,
 614                                             semflg & 0700))) {
 615                                 lockmgr(&sema[semid].lk, LK_RELEASE);
 616                                 goto done;
 617                         }
 618                         if (nsems > 0 && sema[semid].ds.sem_nsems < nsems) {
 619 #ifdef SEM_DEBUG
 620                                 kprintf("too small\n");
 621 #endif
 622                                 eval = EINVAL;
 623                                 lockmgr(&sema[semid].lk, LK_RELEASE);
 624                                 goto done;
 625                         }
 626                         if ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) {
 627 #ifdef SEM_DEBUG
 628                                 kprintf("not exclusive\n");
 629 #endif
 630                                 eval = EEXIST;
 631                                 lockmgr(&sema[semid].lk, LK_RELEASE);
 632                                 goto done;
 633                         }
 634
 635                         /*
 636                          * Return this one.
 637                          */
 638                         lockmgr(&sema[semid].lk, LK_RELEASE);
 639                         goto done;
 640                 }
 641         }
 642
 643 #ifdef SEM_DEBUG
 644         kprintf("need to allocate the semid_ds\n");
 645 #endif
 646         if (key == IPC_PRIVATE || (semflg & IPC_CREAT)) {
 647                 if (nsems <= 0 || nsems > seminfo.semmsl) {
 648 #ifdef SEM_DEBUG
 649                         kprintf("nsems out of range (0<%d<=%d)\n",
 650                                 nsems, seminfo.semmsl);
 651 #endif
 652                         eval = EINVAL;
 653                         goto done;
 654                 }
 655
 656                 /*
 657                  * SEM_ALLOC flag cannot be set unless sema_lk is locked.
 658                  * semtot field also protected by sema_lk.
 659                  */
 660                 lockmgr(&sema_lk, LK_EXCLUSIVE);
 661                 if (nsems > seminfo.semmns - semtot) {
 662 #ifdef SEM_DEBUG
 663                         kprintf("not enough semaphores left "
 664                                 "(need %d, got %d)\n",
 665                                 nsems, seminfo.semmns - semtot);
 666 #endif
 667                         eval = ENOSPC;
 668                         lockmgr(&sema_lk, LK_RELEASE);
 669                         goto done;
 670                 }
 671                 for (semid = 0; semid < seminfo.semmni; semid++) {
 672                         if ((sema[semid].ds.sem_perm.mode & SEM_ALLOC) == 0)
 673                                 break;
 674                 }
 675                 if (semid == seminfo.semmni) {
 676 #ifdef SEM_DEBUG
 677                         kprintf("no more semid_ds's available\n");
 678 #endif
 679                         eval = ENOSPC;
 680                         lockmgr(&sema_lk, LK_RELEASE);
 681                         goto done;
 682                 }
 683 #ifdef SEM_DEBUG
 684                 kprintf("semid %d is available\n", semid);
 685 #endif
 686                 lockmgr(&sema[semid].lk, LK_EXCLUSIVE);
 687                 sema[semid].ds.sem_perm.key = key;
 688                 sema[semid].ds.sem_perm.cuid = cred->cr_uid;
 689                 sema[semid].ds.sem_perm.uid = cred->cr_uid;
 690                 sema[semid].ds.sem_perm.cgid = cred->cr_gid;
 691                 sema[semid].ds.sem_perm.gid = cred->cr_gid;
 692                 sema[semid].ds.sem_perm.mode = (semflg & 0777) | SEM_ALLOC;
 693                 sema[semid].ds.sem_perm.seq =
 694                     (sema[semid].ds.sem_perm.seq + 1) & 0x7fff;
 695                 sema[semid].ds.sem_nsems = nsems;
 696                 sema[semid].ds.sem_otime = 0;
 697                 sema[semid].ds.sem_ctime = time_second;
 698                 sema[semid].ds.sem_base = kmalloc(sizeof(struct sem) * nsems,
 699                                                M_SEM, M_WAITOK|M_ZERO);
 700                 semtot += nsems;
 701                 ++sema[semid].gen;
 702                 lockmgr(&sema[semid].lk, LK_RELEASE);
 703                 lockmgr(&sema_lk, LK_RELEASE);
 704 #ifdef SEM_DEBUG
 705                 kprintf("sembase = 0x%x, next = 0x%x\n",
 706                         sema[semid].ds.sem_base, &sem[semtot]);
 707 #endif
 708                 /* eval == 0 */
 709         } else {
 710 #ifdef SEM_DEBUG
 711                 kprintf("didn't find it and wasn't asked to create it\n");
 712 #endif
 713                 eval = ENOENT;
 714         }
 715
 716 done:
 717         if (eval == 0) {
 718                 sysmsg->sysmsg_result =
 719                         IXSEQ_TO_IPCID(semid, sema[semid].ds.sem_perm);
 720         }
 721         return(eval);
 722 }
 723
 724 /*
 725  * MPSAFE
 726  */
 727 int
 728 sys_semop(struct sysmsg *sysmsg, const struct semop_args *uap)
 729 {
 730         struct thread *td = curthread;
 731         struct prison *pr = td->td_proc->p_ucred->cr_prison;
 732         int semid = uap->semid;
 733         u_int nsops = uap->nsops;
 734         struct sembuf sops[MAX_SOPS];
 735         struct semid_pool *semaptr;
 736         struct sembuf *sopptr;
 737         struct sem *semptr;
 738         struct sem *xsemptr;
 739         int i, j, eval;
 740         int do_undos;
 741
 742 #ifdef SEM_DEBUG
 743         kprintf("call to semop(%d, 0x%x, %u)\n", semid, sops, nsops);
 744 #endif
 745         if (pr && !PRISON_CAP_ISSET(pr->pr_caps, PRISON_CAP_SYS_SYSVIPC))
 746                 return (ENOSYS);
 747
 748         semid = IPCID_TO_IX(semid);     /* Convert back to zero origin */
 749
 750         if (semid < 0 || semid >= seminfo.semmni) {
 751                 eval = EINVAL;
 752                 goto done2;
 753         }
 754
 755         wakeup_start_delayed();
 756         semaptr = &sema[semid];
 757         lockmgr(&semaptr->lk, LK_SHARED);
 758
 759         if ((semaptr->ds.sem_perm.mode & SEM_ALLOC) == 0) {
 760                 eval = EINVAL;
 761                 goto done;
 762         }
 763         if (semaptr->ds.sem_perm.seq != IPCID_TO_SEQ(uap->semid)) {
 764                 eval = EINVAL;
 765                 goto done;
 766         }
 767
 768         if ((eval = ipcperm(td->td_proc, &semaptr->ds.sem_perm, IPC_W))) {
 769 #ifdef SEM_DEBUG
 770                 kprintf("eval = %d from ipaccess\n", eval);
 771 #endif
 772                 goto done;
 773         }
 774
 775         if (nsops > MAX_SOPS) {
 776 #ifdef SEM_DEBUG
 777                 kprintf("too many sops (max=%d, nsops=%u)\n", MAX_SOPS, nsops);
 778 #endif
 779                 eval = E2BIG;
 780                 goto done;
 781         }
 782
 783         if ((eval = copyin(uap->sops, &sops, nsops * sizeof(sops[0]))) != 0) {
 784 #ifdef SEM_DEBUG
 785                 kprintf("eval = %d from copyin(%08x, %08x, %u)\n", eval,
 786                     uap->sops, &sops, nsops * sizeof(sops[0]));
 787 #endif
 788                 goto done;
 789         }
 790
 791         /*
 792          * Loop trying to satisfy the vector of requests.
 793          * If we reach a point where we must wait, any requests already
 794          * performed are rolled back and we go to sleep until some other
 795          * process wakes us up.  At this point, we start all over again.
 796          *
 797          * This ensures that from the perspective of other tasks, a set
 798          * of requests is atomic (never partially satisfied).
 799          */
 800         do_undos = 0;
 801
 802         for (;;) {
 803                 long gen;
 804
 805                 semptr = NULL;
 806
 807                 for (i = 0; i < nsops; i++) {
 808                         sopptr = &sops[i];
 809
 810                         if (sopptr->sem_num >= semaptr->ds.sem_nsems) {
 811                                 eval = EFBIG;
 812                                 goto done;
 813                         }
 814
 815                         semptr = &semaptr->ds.sem_base[sopptr->sem_num];
 816                         lwkt_getpooltoken(semptr);
 817
 818 #ifdef SEM_DEBUG
 819                         kprintf("semop:  semaptr=%x, sem_base=%x, semptr=%x, "
 820                                 "sem[%d]=%d : op=%d, flag=%s\n",
 821                             semaptr, semaptr->ds.sem_base, semptr,
 822                             sopptr->sem_num, semptr->semval, sopptr->sem_op,
 823                             (sopptr->sem_flg & IPC_NOWAIT) ? "nowait" : "wait");
 824 #endif
 825
 826                         if (sopptr->sem_op < 0) {
 827                                 if (semptr->semval + sopptr->sem_op < 0) {
 828 #ifdef SEM_DEBUG
 829                                         kprintf("semop:  can't do it now\n");
 830 #endif
 831                                         break;
 832                                 } else {
 833                                         semptr->semval += sopptr->sem_op;
 834                                         if (semptr->semval == 0 &&
 835                                             semptr->semzcnt > 0) {
 836                                                 wakeup(semptr);
 837                                         }
 838                                 }
 839                                 if (sopptr->sem_flg & SEM_UNDO)
 840                                         do_undos = 1;
 841                         } else if (sopptr->sem_op == 0) {
 842                                 if (semptr->semval > 0) {
 843 #ifdef SEM_DEBUG
 844                                         kprintf("semop:  not zero now\n");
 845 #endif
 846                                         break;
 847                                 }
 848                         } else {
 849                                 semptr->semval += sopptr->sem_op;
 850                                 if (sopptr->sem_flg & SEM_UNDO)
 851                                         do_undos = 1;
 852                                 if (semptr->semncnt > 0)
 853                                         wakeup(semptr);
 854                         }
 855                         lwkt_relpooltoken(semptr);
 856                 }
 857
 858                 /*
 859                  * Did we get through the entire vector?
 860                  */
 861                 if (i >= nsops)
 862                         goto donex;
 863
 864                 /*
 865                  * No, protect the semaphore request which also flags that
 866                  * a wakeup is needed, then release semptr since we know
 867                  * another process is likely going to need to access it
 868                  * soon.
 869                  */
 870                 if (sopptr->sem_op == 0)
 871                         semptr->semzcnt++;
 872                 else
 873                         semptr->semncnt++;
 874                 tsleep_interlock(semptr, PCATCH);
 875                 lwkt_relpooltoken(semptr);
 876
 877                 /*
 878                  * Rollback the semaphores we had acquired.
 879                  */
 880 #ifdef SEM_DEBUG
 881                 kprintf("semop:  rollback 0 through %d\n", i-1);
 882 #endif
 883                 for (j = 0; j < i; j++) {
 884                         xsemptr = &semaptr->ds.sem_base[sops[j].sem_num];
 885                         lwkt_getpooltoken(xsemptr);
 886                         xsemptr->semval -= sops[j].sem_op;
 887                         if (xsemptr->semval == 0 && xsemptr->semzcnt > 0)
 888                                 wakeup(xsemptr);
 889                         if (xsemptr->semval <= 0 && xsemptr->semncnt > 0)
 890                                 wakeup(xsemptr);
 891                         lwkt_relpooltoken(xsemptr);
 892                 }
 893
 894                 /*
 895                  * If the request that we couldn't satisfy has the
 896                  * NOWAIT flag set then return with EAGAIN.
 897                  */
 898                 if (sopptr->sem_flg & IPC_NOWAIT) {
 899                         eval = EAGAIN;
 900                         goto done;
 901                 }
 902
 903                 /*
 904                  * Release semaptr->lk while sleeping, allowing other
 905                  * semops (like SETVAL, SETALL, etc), which require an
 906                  * exclusive lock and might wake us up.
 907                  *
 908                  * Reload and recheck the validity of semaptr on return.
 909                  * Note that semptr itself might have changed too, but
 910                  * we've already interlocked for semptr and that is what
 911                  * will be woken up if it wakes up the tsleep on a MP
 912                  * race.
 913                  *
 914                  * gen protects against destroy/re-create races where the
 915                  * creds match.
 916                  */
 917 #ifdef SEM_DEBUG
 918                 kprintf("semop:  good night!\n");
 919 #endif
 920                 gen = semaptr->gen;
 921                 lockmgr(&semaptr->lk, LK_RELEASE);
 922                 eval = tsleep(semptr, PCATCH | PINTERLOCKED, "semwait", hz);
 923                 lockmgr(&semaptr->lk, LK_SHARED);
 924 #ifdef SEM_DEBUG
 925                 kprintf("semop:  good morning (eval=%d)!\n", eval);
 926 #endif
 927
 928                 /* return code is checked below, after sem[nz]cnt-- */
 929
 930                 /*
 931                  * Make sure that the semaphore still exists
 932                  */
 933                 if (semaptr->gen != gen ||
 934                     (semaptr->ds.sem_perm.mode & SEM_ALLOC) == 0 ||
 935                     semaptr->ds.sem_perm.seq != IPCID_TO_SEQ(uap->semid)) {
 936                         eval = EIDRM;
 937                         goto done;
 938                 }
 939
 940                 /*
 941                  * The semaphore is still alive.  Readjust the count of
 942                  * waiting processes.
 943                  */
 944                 semptr = &semaptr->ds.sem_base[sopptr->sem_num];
 945                 lwkt_getpooltoken(semptr);
 946                 if (sopptr->sem_op == 0)
 947                         semptr->semzcnt--;
 948                 else
 949                         semptr->semncnt--;
 950                 lwkt_relpooltoken(semptr);
 951
 952                 /*
 953                  * Is it really morning, or was our sleep interrupted?
 954                  * (Delayed check of tsleep() return code because we
 955                  * need to decrement sem[nz]cnt either way.)
 956                  */
 957                 if (eval) {
 958                         eval = EINTR;
 959                         goto done;
 960                 }
 961 #ifdef SEM_DEBUG
 962                 kprintf("semop:  good morning!\n");
 963 #endif
 964                 /* RETRY LOOP */
 965         }
 966
 967 donex:
 968         /*
 969          * Process any SEM_UNDO requests.
 970          */
 971         if (do_undos) {
 972                 for (i = 0; i < nsops; i++) {
 973                         /*
 974                          * We only need to deal with SEM_UNDO's for non-zero
 975                          * op's.
 976                          */
 977                         int adjval;
 978
 979                         if ((sops[i].sem_flg & SEM_UNDO) == 0)
 980                                 continue;
 981                         adjval = sops[i].sem_op;
 982                         if (adjval == 0)
 983                                 continue;
 984                         eval = semundo_adjust(td->td_proc, semid,
 985                                               sops[i].sem_num, -adjval);
 986                         if (eval == 0)
 987                                 continue;
 988
 989                         /*
 990                          * Oh-Oh!  We ran out of either sem_undo's or undo's.
 991                          * Rollback the adjustments to this point and then
 992                          * rollback the semaphore ups and down so we can return
 993                          * with an error with all structures restored.  We
 994                          * rollback the undo's in the exact reverse order that
 995                          * we applied them.  This guarantees that we won't run
 996                          * out of space as we roll things back out.
 997                          */
 998                         for (j = i - 1; j >= 0; j--) {
 999                                 if ((sops[j].sem_flg & SEM_UNDO) == 0)
1000                                         continue;
1001                                 adjval = sops[j].sem_op;
1002                                 if (adjval == 0)
1003                                         continue;
1004                                 if (semundo_adjust(td->td_proc, semid,
1005                                                sops[j].sem_num, adjval) != 0)
1006                                         panic("semop - can't undo undos");
1007                         }
1008
1009                         for (j = 0; j < nsops; j++) {
1010                                 xsemptr = &semaptr->ds.sem_base[
1011                                                         sops[j].sem_num];
1012                                 lwkt_getpooltoken(xsemptr);
1013                                 xsemptr->semval -= sops[j].sem_op;
1014                                 if (xsemptr->semval == 0 &&
1015                                     xsemptr->semzcnt > 0)
1016                                         wakeup(xsemptr);
1017                                 if (xsemptr->semval <= 0 &&
1018                                     xsemptr->semncnt > 0)
1019                                         wakeup(xsemptr);
1020                                 lwkt_relpooltoken(xsemptr);
1021                         }
1022
1023 #ifdef SEM_DEBUG
1024                         kprintf("eval = %d from semundo_adjust\n", eval);
1025 #endif
1026                         goto done;
1027                 } /* loop through the sops */
1028         } /* if (do_undos) */
1029
1030         /* We're definitely done - set the sempid's */
1031         for (i = 0; i < nsops; i++) {
1032                 sopptr = &sops[i];
1033                 semptr = &semaptr->ds.sem_base[sopptr->sem_num];
1034                 lwkt_getpooltoken(semptr);
1035                 semptr->sempid = td->td_proc->p_pid;
1036                 lwkt_relpooltoken(semptr);
1037         }
1038
1039         /* Do a wakeup if any semaphore was up'd. */
1040 #ifdef SEM_DEBUG
1041         kprintf("semop:  done\n");
1042 #endif
1043         sysmsg->sysmsg_result = 0;
1044         eval = 0;
1045 done:
1046         lockmgr(&semaptr->lk, LK_RELEASE);
1047         wakeup_end_delayed();
1048 done2:
1049         return(eval);
1050 }
1051
1052 /*
1053  * Go through the undo structures for this process and apply the adjustments to
1054  * semaphores.
1055  *
1056  * (p->p_token is held by the caller)
1057  */
1058 void
1059 semexit(struct proc *p)
1060 {
1061         struct sem_undo *suptr;
1062         struct sem *semptr;
1063
1064         /*
1065          * We're getting a global token, don't do it if we couldn't
1066          * possibly have any semaphores.
1067          */
1068         if ((p->p_flags & P_SYSVSEM) == 0)
1069                 return;
1070         suptr = p->p_sem_undo;
1071         KKASSERT(suptr != NULL);
1072
1073         /*
1074          * Disconnect suptr from the process and increment un_refs to
1075          * prevent anyone else from being able to destroy the structure.
1076          * Do not remove it from the linked list until after we are through
1077          * scanning it as other semaphore calls might still effect it.
1078          */
1079         lwkt_gettoken(&semu_token);
1080 #if 0
1081         /*
1082          * do not disconnect proc yet, doing so prevents RMID
1083          * from cleaning up the structure atomically with SEM_ALLOC
1084          */
1085         p->p_sem_undo = NULL;
1086         p->p_flags &= ~P_SYSVSEM;
1087         suptr->un_proc = NULL;
1088 #endif
1089         ++suptr->un_refs;
1090         lwkt_reltoken(&semu_token);
1091
1092         while (suptr->un_cnt) {
1093                 struct semid_pool *semaptr;
1094                 int semid;
1095                 int semnum;
1096                 int adjval;
1097                 int ix;
1098
1099                 /*
1100                  * These values are stable because we hold p->p_token.
1101                  * However, they can get ripped out from under us when
1102                  * we block or obtain other tokens so we have to re-check.
1103                  */
1104                 ix = suptr->un_cnt - 1;
1105                 semid = suptr->un_ent[ix].un_id;
1106                 semnum = suptr->un_ent[ix].un_num;
1107                 adjval = suptr->un_ent[ix].un_adjval;
1108
1109                 semaptr = &sema[semid];
1110
1111                 /*
1112                  * Recheck after locking, then execute the undo
1113                  * operation.  semptr remains valid due to the
1114                  * semaptr->lk.
1115                  */
1116                 lockmgr(&semaptr->lk, LK_EXCLUSIVE);
1117                 semptr = &semaptr->ds.sem_base[semnum];
1118                 lwkt_getpooltoken(semptr);
1119
1120                 if (ix == suptr->un_cnt - 1 &&
1121                     semid == suptr->un_ent[ix].un_id &&
1122                     semnum == suptr->un_ent[ix].un_num &&
1123                     adjval == suptr->un_ent[ix].un_adjval) {
1124                         /*
1125                          * Only do assertions when we aren't in a SMP race.
1126                          */
1127                         if ((semaptr->ds.sem_perm.mode & SEM_ALLOC) == 0)
1128                                 panic("semexit - semid not allocated");
1129                         if (semnum >= semaptr->ds.sem_nsems)
1130                                 panic("semexit - semnum out of range");
1131                         --suptr->un_cnt;
1132
1133                         if (adjval < 0) {
1134                                 if (semptr->semval < -adjval)
1135                                         semptr->semval = 0;
1136                                 else
1137                                         semptr->semval += adjval;
1138                         } else {
1139                                 semptr->semval += adjval;
1140                         }
1141                         wakeup(semptr);
1142                 }
1143                 lwkt_relpooltoken(semptr);
1144                 lockmgr(&semaptr->lk, LK_RELEASE);
1145         }
1146
1147         /*
1148          * Final cleanup, remove from the list, remove the process association,
1149          * then deallocate on last ref.
1150          */
1151         lwkt_gettoken(&semu_token);
1152
1153         p->p_sem_undo = NULL;
1154         p->p_flags &= ~P_SYSVSEM;
1155         suptr->un_proc = NULL;
1156
1157         if (--suptr->un_refs == 0) {
1158                 TAILQ_REMOVE(&semu_list, suptr, un_entry);
1159                 KKASSERT(suptr->un_cnt == 0);
1160                 kfree(suptr, M_SEM);
1161         }
1162         lwkt_reltoken(&semu_token);
1163 }