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 #include <sys/thread2.h>
  26
  27 static MALLOC_DEFINE(M_SEM, "sem", "SVID compatible semaphores");
  28
  29 static void seminit (void *);
  30
  31 static struct sem_undo *semu_alloc (struct proc *p);
  32 static int semundo_adjust (struct proc *p, int semid, int semnum, int adjval);
  33 static void semundo_clear (int semid, int semnum);
  34
  35 static struct lwkt_token semu_token = LWKT_TOKEN_INITIALIZER(semu_token);
  36 static int      semtot = 0;
  37 static struct semid_pool *sema; /* semaphore id pool */
  38 static TAILQ_HEAD(, sem_undo) semu_list = TAILQ_HEAD_INITIALIZER(semu_list);
  39 static struct lock sema_lk;
  40
  41 struct sem {
  42         u_short semval;         /* semaphore value */
  43         pid_t   sempid;         /* pid of last operation */
  44         u_short semncnt;        /* # awaiting semval > cval */
  45         u_short semzcnt;        /* # awaiting semval = 0 */
  46 };
  47
  48 /*
  49  * Undo structure (one per process)
  50  */
  51 struct sem_undo {
  52         TAILQ_ENTRY(sem_undo) un_entry; /* linked list for semundo_clear() */
  53         struct  proc *un_proc;          /* owner of this structure */
  54         int     un_refs;                /* prevent unlink/kfree */
  55         short   un_cnt;                 /* # of active entries */
  56         short   un_unused;
  57         struct undo {
  58                 short   un_adjval;      /* adjust on exit values */
  59                 short   un_num;         /* semaphore # */
  60                 int     un_id;          /* semid */
  61         } un_ent[1];                    /* undo entries */
  62 };
  63
  64 /*
  65  * Configuration parameters
  66  */
  67 #ifndef SEMMNI
  68 #define SEMMNI  1024            /* # of semaphore identifiers */
  69 #endif
  70 #ifndef SEMMNS
  71 #define SEMMNS  32767           /* # of semaphores in system */
  72 #endif
  73 #ifndef SEMUME
  74 #define SEMUME  25              /* max # of undo entries per process */
  75 #endif
  76 #ifndef SEMMNU
  77 #define SEMMNU  1024            /* # of undo structures in system */
  78                                 /* NO LONGER USED */
  79 #endif
  80
  81 /* shouldn't need tuning */
  82 #ifndef SEMMAP
  83 #define SEMMAP  128             /* # of entries in semaphore map */
  84 #endif
  85 #ifndef SEMMSL
  86 #define SEMMSL  SEMMNS          /* max # of semaphores per id */
  87 #endif
  88 #ifndef SEMOPM
  89 #define SEMOPM  100             /* max # of operations per semop call */
  90 #endif
  91
  92 #define SEMVMX  32767           /* semaphore maximum value */
  93 #define SEMAEM  16384           /* adjust on exit max value */
  94
  95 /*
  96  * Due to the way semaphore memory is allocated, we have to ensure that
  97  * SEMUSZ is properly aligned.
  98  */
  99
 100 #define SEM_ALIGN(bytes) roundup2(bytes, sizeof(long))
 101
 102 /* actual size of an undo structure */
 103 #define SEMUSZ(nent)    SEM_ALIGN(offsetof(struct sem_undo, un_ent[nent]))
 104
 105 /*
 106  * semaphore info struct
 107  */
 108 struct seminfo seminfo = {
 109                 SEMMAP,         /* # of entries in semaphore map */
 110                 SEMMNI,         /* # of semaphore identifiers */
 111                 SEMMNS,         /* # of semaphores in system */
 112                 SEMMNU,         /* # of undo structures in system */
 113                 SEMMSL,         /* max # of semaphores per id */
 114                 SEMOPM,         /* max # of operations per semop call */
 115                 SEMUME,         /* max # of undo entries per process */
 116                 SEMUSZ(SEMUME), /* size in bytes of undo structure */
 117                 SEMVMX,         /* semaphore maximum value */
 118                 SEMAEM          /* adjust on exit max value */
 119 };
 120
 121 TUNABLE_INT("kern.ipc.semmap", &seminfo.semmap);
 122 TUNABLE_INT("kern.ipc.semmni", &seminfo.semmni);
 123 TUNABLE_INT("kern.ipc.semmns", &seminfo.semmns);
 124 TUNABLE_INT("kern.ipc.semmnu", &seminfo.semmnu);
 125 TUNABLE_INT("kern.ipc.semmsl", &seminfo.semmsl);
 126 TUNABLE_INT("kern.ipc.semopm", &seminfo.semopm);
 127 TUNABLE_INT("kern.ipc.semume", &seminfo.semume);
 128 TUNABLE_INT("kern.ipc.semusz", &seminfo.semusz);
 129 TUNABLE_INT("kern.ipc.semvmx", &seminfo.semvmx);
 130 TUNABLE_INT("kern.ipc.semaem", &seminfo.semaem);
 131
 132 SYSCTL_INT(_kern_ipc, OID_AUTO, semmap, CTLFLAG_RW, &seminfo.semmap, 0,
 133     "Number of entries in semaphore map");
 134 SYSCTL_INT(_kern_ipc, OID_AUTO, semmni, CTLFLAG_RD, &seminfo.semmni, 0,
 135     "Number of semaphore identifiers");
 136 SYSCTL_INT(_kern_ipc, OID_AUTO, semmns, CTLFLAG_RD, &seminfo.semmns, 0,
 137     "Total number of semaphores");
 138 SYSCTL_INT(_kern_ipc, OID_AUTO, semmnu, CTLFLAG_RD, &seminfo.semmnu, 0,
 139     "Total number of undo structures");
 140 SYSCTL_INT(_kern_ipc, OID_AUTO, semmsl, CTLFLAG_RW, &seminfo.semmsl, 0,
 141     "Max number of semaphores per id");
 142 SYSCTL_INT(_kern_ipc, OID_AUTO, semopm, CTLFLAG_RD, &seminfo.semopm, 0,
 143     "Max number of operations per semop call");
 144 SYSCTL_INT(_kern_ipc, OID_AUTO, semume, CTLFLAG_RD, &seminfo.semume, 0,
 145     "Max number of undo entries per process");
 146 SYSCTL_INT(_kern_ipc, OID_AUTO, semusz, CTLFLAG_RD, &seminfo.semusz, 0,
 147     "Size in bytes of undo structure");
 148 SYSCTL_INT(_kern_ipc, OID_AUTO, semvmx, CTLFLAG_RW, &seminfo.semvmx, 0,
 149     "Semaphore maximum value");
 150 SYSCTL_INT(_kern_ipc, OID_AUTO, semaem, CTLFLAG_RW, &seminfo.semaem, 0,
 151     "Adjust on exit max value");
 152
 153 #if 0
 154 RO seminfo.semmap       /* SEMMAP unused */
 155 RO seminfo.semmni
 156 RO seminfo.semmns
 157 RO seminfo.semmnu       /* undo entries per system */
 158 RW seminfo.semmsl
 159 RO seminfo.semopm       /* SEMOPM unused */
 160 RO seminfo.semume
 161 RO seminfo.semusz       /* param - derived from SEMUME for per-proc sizeof */
 162 RO seminfo.semvmx       /* SEMVMX unused - user param */
 163 RO seminfo.semaem       /* SEMAEM unused - user param */
 164 #endif
 165
 166 static void
 167 seminit(void *dummy)
 168 {
 169         int i;
 170
 171         sema = kmalloc(sizeof(struct semid_pool) * seminfo.semmni,
 172                       M_SEM, M_WAITOK | M_ZERO);
 173
 174         lockinit(&sema_lk, "semglb", 0, 0);
 175         for (i = 0; i < seminfo.semmni; i++) {
 176                 struct semid_pool *semaptr = &sema[i];
 177
 178                 lockinit(&semaptr->lk, "semary", 0, 0);
 179                 semaptr->ds.sem_base = NULL;
 180                 semaptr->ds.sem_perm.mode = 0;
 181         }
 182 }
 183 SYSINIT(sysv_sem, SI_SUB_SYSV_SEM, SI_ORDER_FIRST, seminit, NULL);
 184
 185 /*
 186  * Allocate a new sem_undo structure for a process
 187  * (returns ptr to structure or NULL if no more room)
 188  */
 189 static struct sem_undo *
 190 semu_alloc(struct proc *p)
 191 {
 192         struct sem_undo *semu;
 193
 194         /*
 195          * Allocate the semu structure and associate it with the process,
 196          * as necessary.
 197          */
 198         while ((semu = p->p_sem_undo) == NULL) {
 199                 semu = kmalloc(SEMUSZ(seminfo.semume), M_SEM,
 200                                M_WAITOK | M_ZERO);
 201                 lwkt_gettoken(&semu_token);
 202                 lwkt_gettoken(&p->p_token);
 203                 if (p->p_sem_undo == NULL) {
 204                         p->p_sem_undo = semu;
 205                         p->p_flags |= P_SYSVSEM;
 206                         semu->un_proc = p;
 207                         TAILQ_INSERT_TAIL(&semu_list, semu, un_entry);
 208                 } else {
 209                         kfree(semu, M_SEM);
 210                 }
 211                 lwkt_reltoken(&p->p_token);
 212                 lwkt_reltoken(&semu_token);
 213         }
 214         return(semu);
 215 }
 216
 217 /*
 218  * Adjust a particular entry for a particular proc
 219  */
 220 static int
 221 semundo_adjust(struct proc *p, int semid, int semnum, int adjval)
 222 {
 223         struct sem_undo *suptr;
 224         struct undo *sunptr;
 225         int i;
 226         int error = 0;
 227
 228         /*
 229          * Look for and remember the sem_undo if the caller doesn't
 230          * provide it.
 231          */
 232         suptr = semu_alloc(p);
 233         lwkt_gettoken(&p->p_token);
 234
 235         /*
 236          * Look for the requested entry and adjust it (delete if adjval becomes
 237          * 0).
 238          */
 239         sunptr = &suptr->un_ent[0];
 240         for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
 241                 if (sunptr->un_id != semid || sunptr->un_num != semnum)
 242                         continue;
 243                 if (adjval == 0)
 244                         sunptr->un_adjval = 0;
 245                 else
 246                         sunptr->un_adjval += adjval;
 247                 if (sunptr->un_adjval == 0) {
 248                         suptr->un_cnt--;
 249                         if (i < suptr->un_cnt)
 250                                 suptr->un_ent[i] = suptr->un_ent[suptr->un_cnt];
 251                 }
 252                 goto done;
 253         }
 254
 255         /* Didn't find the right entry - create it */
 256         if (adjval == 0)
 257                 goto done;
 258         if (suptr->un_cnt != seminfo.semume) {
 259                 sunptr = &suptr->un_ent[suptr->un_cnt];
 260                 suptr->un_cnt++;
 261                 sunptr->un_adjval = adjval;
 262                 sunptr->un_id = semid;
 263                 sunptr->un_num = semnum;
 264         } else {
 265                 error = EINVAL;
 266         }
 267 done:
 268         lwkt_reltoken(&p->p_token);
 269
 270         return (error);
 271 }
 272
 273 /*
 274  * This is rather expensive
 275  */
 276 static void
 277 semundo_clear(int semid, int semnum)
 278 {
 279         struct proc *p;
 280         struct sem_undo *suptr;
 281         struct sem_undo *sunext;
 282         struct undo *sunptr;
 283         int i;
 284
 285         lwkt_gettoken(&semu_token);
 286         sunext = TAILQ_FIRST(&semu_list);
 287         while ((suptr = sunext) != NULL) {
 288                 if ((p = suptr->un_proc) == NULL) {
 289                         suptr = TAILQ_NEXT(suptr, un_entry);
 290                         continue;
 291                 }
 292                 ++suptr->un_refs;
 293                 PHOLD(p);
 294                 lwkt_gettoken(&p->p_token);
 295
 296                 sunptr = &suptr->un_ent[0];
 297                 i = 0;
 298
 299                 while (i < suptr->un_cnt) {
 300                         if (sunptr->un_id == semid) {
 301                                 if (semnum == -1 || sunptr->un_num == semnum) {
 302                                         suptr->un_cnt--;
 303                                         if (i < suptr->un_cnt) {
 304                                                 suptr->un_ent[i] =
 305                                                   suptr->un_ent[suptr->un_cnt];
 306                                                 /*
 307                                                  * do not increment i
 308                                                  * or sunptr after copydown.
 309                                                  */
 310                                                 continue;
 311                                         }
 312                                 }
 313                                 if (semnum != -1)
 314                                         break;
 315                         }
 316                         ++i;
 317                         ++sunptr;
 318                 }
 319
 320                 lwkt_reltoken(&p->p_token);
 321                 PRELE(p);
 322
 323                 /*
 324                  * Handle deletion races
 325                  */
 326                 sunext = TAILQ_NEXT(suptr, un_entry);
 327                 if (--suptr->un_refs == 0 && suptr->un_proc == NULL) {
 328                         KKASSERT(suptr->un_cnt == 0);
 329                         TAILQ_REMOVE(&semu_list, suptr, un_entry);
 330                         kfree(suptr, M_SEM);
 331                 }
 332         }
 333         lwkt_reltoken(&semu_token);
 334 }
 335
 336 /*
 337  * Note that the user-mode half of this passes a union, not a pointer
 338  *
 339  * MPALMOSTSAFE
 340  */
 341 int
 342 sys___semctl(struct __semctl_args *uap)
 343 {
 344         struct thread *td = curthread;
 345         int semid = uap->semid;
 346         int semnum = uap->semnum;
 347         int cmd = uap->cmd;
 348         union semun *arg = uap->arg;
 349         union semun real_arg;
 350         struct ucred *cred = td->td_ucred;
 351         int i, rval, eval;
 352         struct semid_ds sbuf;
 353         struct semid_pool *semaptr;
 354         struct semid_pool *semakptr;
 355         struct sem *semptr;
 356
 357 #ifdef SEM_DEBUG
 358         kprintf("call to semctl(%d, %d, %d, 0x%x)\n", semid, semnum, cmd, arg);
 359 #endif
 360
 361         if (!jail_sysvipc_allowed && cred->cr_prison != NULL)
 362                 return (ENOSYS);
 363
 364         switch (cmd) {
 365         case SEM_STAT:
 366                 /*
 367                  * For this command we assume semid is an array index
 368                  * rather than an IPC id.
 369                  */
 370                 if (semid < 0 || semid >= seminfo.semmni) {
 371                         eval = EINVAL;
 372                         break;
 373                 }
 374                 semakptr = &sema[semid];
 375                 lockmgr(&semakptr->lk, LK_EXCLUSIVE);
 376                 if ((semakptr->ds.sem_perm.mode & SEM_ALLOC) == 0) {
 377                         eval = EINVAL;
 378                         lockmgr(&semakptr->lk, LK_RELEASE);
 379                         break;
 380                 }
 381                 if ((eval = ipcperm(td->td_proc, &semakptr->ds.sem_perm, IPC_R))) {
 382                         lockmgr(&semakptr->lk, LK_RELEASE);
 383                         break;
 384                 }
 385                 bcopy(&semakptr->ds, arg->buf, sizeof(struct semid_ds));
 386                 rval = IXSEQ_TO_IPCID(semid, semakptr->ds.sem_perm);
 387                 lockmgr(&semakptr->lk, LK_RELEASE);
 388                 break;
 389         }
 390
 391         semid = IPCID_TO_IX(semid);
 392         if (semid < 0 || semid >= seminfo.semmni) {
 393                 return(EINVAL);
 394         }
 395         semaptr = &sema[semid];
 396         lockmgr(&semaptr->lk, LK_EXCLUSIVE);
 397
 398         if ((semaptr->ds.sem_perm.mode & SEM_ALLOC) == 0 ||
 399             semaptr->ds.sem_perm.seq != IPCID_TO_SEQ(uap->semid)) {
 400                 lockmgr(&semaptr->lk, LK_RELEASE);
 401                 return(EINVAL);
 402         }
 403
 404         eval = 0;
 405         rval = 0;
 406
 407         switch (cmd) {
 408         case IPC_RMID:
 409                 eval = ipcperm(td->td_proc, &semaptr->ds.sem_perm, IPC_M);
 410                 if (eval != 0)
 411                         break;
 412                 semaptr->ds.sem_perm.cuid = cred->cr_uid;
 413                 semaptr->ds.sem_perm.uid = cred->cr_uid;
 414
 415                 /*
 416                  * NOTE: Nobody will be waiting on the semaphores since
 417                  *       we have an exclusive lock on semaptr->lk).
 418                  */
 419                 lockmgr(&sema_lk, LK_EXCLUSIVE);
 420                 semtot -= semaptr->ds.sem_nsems;
 421                 kfree(semaptr->ds.sem_base, M_SEM);
 422                 semaptr->ds.sem_base = NULL;
 423                 semaptr->ds.sem_perm.mode = 0;  /* clears SEM_ALLOC */
 424                 lockmgr(&sema_lk, LK_RELEASE);
 425
 426                 semundo_clear(semid, -1);
 427                 break;
 428
 429         case IPC_SET:
 430                 eval = ipcperm(td->td_proc, &semaptr->ds.sem_perm, IPC_M);
 431                 if (eval)
 432                         break;
 433                 if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
 434                         break;
 435                 if ((eval = copyin(real_arg.buf, (caddr_t)&sbuf,
 436                                    sizeof(sbuf))) != 0) {
 437                         break;
 438                 }
 439                 semaptr->ds.sem_perm.uid = sbuf.sem_perm.uid;
 440                 semaptr->ds.sem_perm.gid = sbuf.sem_perm.gid;
 441                 semaptr->ds.sem_perm.mode =
 442                         (semaptr->ds.sem_perm.mode & ~0777) |
 443                         (sbuf.sem_perm.mode & 0777);
 444                 semaptr->ds.sem_ctime = time_second;
 445                 break;
 446
 447         case IPC_STAT:
 448                 eval = ipcperm(td->td_proc, &semaptr->ds.sem_perm, IPC_R);
 449                 if (eval)
 450                         break;
 451                 if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
 452                         break;
 453                 eval = copyout(&semaptr->ds, real_arg.buf,
 454                                sizeof(struct semid_ds));
 455                 break;
 456
 457         case GETNCNT:
 458                 eval = ipcperm(td->td_proc, &semaptr->ds.sem_perm, IPC_R);
 459                 if (eval)
 460                         break;
 461                 if (semnum < 0 || semnum >= semaptr->ds.sem_nsems) {
 462                         eval = EINVAL;
 463                         break;
 464                 }
 465                 rval = semaptr->ds.sem_base[semnum].semncnt;
 466                 break;
 467
 468         case GETPID:
 469                 eval = ipcperm(td->td_proc, &semaptr->ds.sem_perm, IPC_R);
 470                 if (eval)
 471                         break;
 472                 if (semnum < 0 || semnum >= semaptr->ds.sem_nsems) {
 473                         eval = EINVAL;
 474                         break;
 475                 }
 476                 rval = semaptr->ds.sem_base[semnum].sempid;
 477                 break;
 478
 479         case GETVAL:
 480                 eval = ipcperm(td->td_proc, &semaptr->ds.sem_perm, IPC_R);
 481                 if (eval)
 482                         break;
 483                 if (semnum < 0 || semnum >= semaptr->ds.sem_nsems) {
 484                         eval = EINVAL;
 485                         break;
 486                 }
 487                 rval = semaptr->ds.sem_base[semnum].semval;
 488                 break;
 489
 490         case GETALL:
 491                 eval = ipcperm(td->td_proc, &semaptr->ds.sem_perm, IPC_R);
 492                 if (eval)
 493                         break;
 494                 if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
 495                         break;
 496                 for (i = 0; i < semaptr->ds.sem_nsems; i++) {
 497                         eval = copyout(&semaptr->ds.sem_base[i].semval,
 498                                        &real_arg.array[i],
 499                                        sizeof(real_arg.array[0]));
 500                         if (eval)
 501                                 break;
 502                 }
 503                 break;
 504
 505         case GETZCNT:
 506                 eval = ipcperm(td->td_proc, &semaptr->ds.sem_perm, IPC_R);
 507                 if (eval)
 508                         break;
 509                 if (semnum < 0 || semnum >= semaptr->ds.sem_nsems) {
 510                         eval = EINVAL;
 511                         break;
 512                 }
 513                 rval = semaptr->ds.sem_base[semnum].semzcnt;
 514                 break;
 515
 516         case SETVAL:
 517                 eval = ipcperm(td->td_proc, &semaptr->ds.sem_perm, IPC_W);
 518                 if (eval)
 519                         break;
 520                 if (semnum < 0 || semnum >= semaptr->ds.sem_nsems) {
 521                         eval = EINVAL;
 522                         break;
 523                 }
 524                 if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
 525                         break;
 526
 527                 /*
 528                  * Because we hold semaptr->lk exclusively we can safely
 529                  * modify any semptr content without acquiring its token.
 530                  */
 531                 semptr = &semaptr->ds.sem_base[semnum];
 532                 semptr->semval = real_arg.val;
 533                 semundo_clear(semid, semnum);
 534                 if (semptr->semzcnt || semptr->semncnt)
 535                         wakeup(semptr);
 536                 break;
 537
 538         case SETALL:
 539                 eval = ipcperm(td->td_proc, &semaptr->ds.sem_perm, IPC_W);
 540                 if (eval)
 541                         break;
 542                 if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
 543                         break;
 544                 /*
 545                  * Because we hold semaptr->lk exclusively we can safely
 546                  * modify any semptr content without acquiring its token.
 547                  */
 548                 for (i = 0; i < semaptr->ds.sem_nsems; i++) {
 549                         semptr = &semaptr->ds.sem_base[i];
 550                         eval = copyin(&real_arg.array[i],
 551                                       (caddr_t)&semptr->semval,
 552                                       sizeof(real_arg.array[0]));
 553                         if (semptr->semzcnt || semptr->semncnt)
 554                                 wakeup(semptr);
 555                         if (eval != 0)
 556                                 break;
 557                 }
 558                 semundo_clear(semid, -1);
 559                 break;
 560
 561         default:
 562                 eval = EINVAL;
 563                 break;
 564         }
 565         lockmgr(&semaptr->lk, LK_RELEASE);
 566
 567         if (eval == 0)
 568                 uap->sysmsg_result = rval;
 569         return(eval);
 570 }
 571
 572 /*
 573  * MPALMOSTSAFE
 574  */
 575 int
 576 sys_semget(struct semget_args *uap)
 577 {
 578         struct thread *td = curthread;
 579         int semid, eval;
 580         int key = uap->key;
 581         int nsems = uap->nsems;
 582         int semflg = uap->semflg;
 583         struct ucred *cred = td->td_ucred;
 584
 585 #ifdef SEM_DEBUG
 586         kprintf("semget(0x%x, %d, 0%o)\n", key, nsems, semflg);
 587 #endif
 588
 589         if (!jail_sysvipc_allowed && cred->cr_prison != NULL)
 590                 return (ENOSYS);
 591
 592         eval = 0;
 593
 594         if (key != IPC_PRIVATE) {
 595                 for (semid = 0; semid < seminfo.semmni; semid++) {
 596                         if ((sema[semid].ds.sem_perm.mode & SEM_ALLOC) == 0 ||
 597                             sema[semid].ds.sem_perm.key != key) {
 598                                 continue;
 599                         }
 600                         lockmgr(&sema[semid].lk, LK_EXCLUSIVE);
 601                         if ((sema[semid].ds.sem_perm.mode & SEM_ALLOC) == 0 ||
 602                             sema[semid].ds.sem_perm.key != key) {
 603                                 lockmgr(&sema[semid].lk, LK_RELEASE);
 604                                 continue;
 605                         }
 606                         break;
 607                 }
 608                 if (semid < seminfo.semmni) {
 609                         /* sema[semid].lk still locked from above */
 610 #ifdef SEM_DEBUG
 611                         kprintf("found public key\n");
 612 #endif
 613                         if ((eval = ipcperm(td->td_proc,
 614                                             &sema[semid].ds.sem_perm,
 615                                             semflg & 0700))) {
 616                                 lockmgr(&sema[semid].lk, LK_RELEASE);
 617                                 goto done;
 618                         }
 619                         if (nsems > 0 && sema[semid].ds.sem_nsems < nsems) {
 620 #ifdef SEM_DEBUG
 621                                 kprintf("too small\n");
 622 #endif
 623                                 eval = EINVAL;
 624                                 lockmgr(&sema[semid].lk, LK_RELEASE);
 625                                 goto done;
 626                         }
 627                         if ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) {
 628 #ifdef SEM_DEBUG
 629                                 kprintf("not exclusive\n");
 630 #endif
 631                                 eval = EEXIST;
 632                                 lockmgr(&sema[semid].lk, LK_RELEASE);
 633                                 goto done;
 634                         }
 635
 636                         /*
 637                          * Return this one.
 638                          */
 639                         lockmgr(&sema[semid].lk, LK_RELEASE);
 640                         goto done;
 641                 }
 642         }
 643
 644 #ifdef SEM_DEBUG
 645         kprintf("need to allocate the semid_ds\n");
 646 #endif
 647         if (key == IPC_PRIVATE || (semflg & IPC_CREAT)) {
 648                 if (nsems <= 0 || nsems > seminfo.semmsl) {
 649 #ifdef SEM_DEBUG
 650                         kprintf("nsems out of range (0<%d<=%d)\n",
 651                                 nsems, seminfo.semmsl);
 652 #endif
 653                         eval = EINVAL;
 654                         goto done;
 655                 }
 656
 657                 /*
 658                  * SEM_ALLOC flag cannot be set unless sema_lk is locked.
 659                  * semtot field also protected by sema_lk.
 660                  */
 661                 lockmgr(&sema_lk, LK_EXCLUSIVE);
 662                 if (nsems > seminfo.semmns - semtot) {
 663 #ifdef SEM_DEBUG
 664                         kprintf("not enough semaphores left "
 665                                 "(need %d, got %d)\n",
 666                                 nsems, seminfo.semmns - semtot);
 667 #endif
 668                         eval = ENOSPC;
 669                         lockmgr(&sema_lk, LK_RELEASE);
 670                         goto done;
 671                 }
 672                 for (semid = 0; semid < seminfo.semmni; semid++) {
 673                         if ((sema[semid].ds.sem_perm.mode & SEM_ALLOC) == 0)
 674                                 break;
 675                 }
 676                 if (semid == seminfo.semmni) {
 677 #ifdef SEM_DEBUG
 678                         kprintf("no more semid_ds's available\n");
 679 #endif
 680                         eval = ENOSPC;
 681                         lockmgr(&sema_lk, LK_RELEASE);
 682                         goto done;
 683                 }
 684 #ifdef SEM_DEBUG
 685                 kprintf("semid %d is available\n", semid);
 686 #endif
 687                 lockmgr(&sema[semid].lk, LK_EXCLUSIVE);
 688                 sema[semid].ds.sem_perm.key = key;
 689                 sema[semid].ds.sem_perm.cuid = cred->cr_uid;
 690                 sema[semid].ds.sem_perm.uid = cred->cr_uid;
 691                 sema[semid].ds.sem_perm.cgid = cred->cr_gid;
 692                 sema[semid].ds.sem_perm.gid = cred->cr_gid;
 693                 sema[semid].ds.sem_perm.mode = (semflg & 0777) | SEM_ALLOC;
 694                 sema[semid].ds.sem_perm.seq =
 695                     (sema[semid].ds.sem_perm.seq + 1) & 0x7fff;
 696                 sema[semid].ds.sem_nsems = nsems;
 697                 sema[semid].ds.sem_otime = 0;
 698                 sema[semid].ds.sem_ctime = time_second;
 699                 sema[semid].ds.sem_base = kmalloc(sizeof(struct sem) * nsems,
 700                                                M_SEM, M_WAITOK|M_ZERO);
 701                 semtot += nsems;
 702                 ++sema[semid].gen;
 703                 lockmgr(&sema[semid].lk, LK_RELEASE);
 704                 lockmgr(&sema_lk, LK_RELEASE);
 705 #ifdef SEM_DEBUG
 706                 kprintf("sembase = 0x%x, next = 0x%x\n",
 707                         sema[semid].ds.sem_base, &sem[semtot]);
 708 #endif
 709                 /* eval == 0 */
 710         } else {
 711 #ifdef SEM_DEBUG
 712                 kprintf("didn't find it and wasn't asked to create it\n");
 713 #endif
 714                 eval = ENOENT;
 715         }
 716
 717 done:
 718         if (eval == 0) {
 719                 uap->sysmsg_result =
 720                         IXSEQ_TO_IPCID(semid, sema[semid].ds.sem_perm);
 721         }
 722         return(eval);
 723 }
 724
 725 /*
 726  * MPSAFE
 727  */
 728 int
 729 sys_semop(struct semop_args *uap)
 730 {
 731         struct thread *td = curthread;
 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 (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
 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         uap->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         p->p_sem_undo = NULL;
1081         p->p_flags &= ~P_SYSVSEM;
1082         suptr->un_proc = NULL;
1083         ++suptr->un_refs;
1084         lwkt_reltoken(&semu_token);
1085
1086         while (suptr->un_cnt) {
1087                 struct semid_pool *semaptr;
1088                 int semid;
1089                 int semnum;
1090                 int adjval;
1091                 int ix;
1092
1093                 /*
1094                  * These values are stable because we hold p->p_token.
1095                  * However, they can get ripped out from under us when
1096                  * we block or obtain other tokens so we have to re-check.
1097                  */
1098                 ix = suptr->un_cnt - 1;
1099                 semid = suptr->un_ent[ix].un_id;
1100                 semnum = suptr->un_ent[ix].un_num;
1101                 adjval = suptr->un_ent[ix].un_adjval;
1102
1103                 semaptr = &sema[semid];
1104
1105                 /*
1106                  * Recheck after locking, then execute the undo
1107                  * operation.  semptr remains valid due to the
1108                  * semaptr->lk.
1109                  */
1110                 lockmgr(&semaptr->lk, LK_SHARED);
1111                 semptr = &semaptr->ds.sem_base[semnum];
1112                 lwkt_getpooltoken(semptr);
1113
1114                 if (ix == suptr->un_cnt - 1 &&
1115                     semid == suptr->un_ent[ix].un_id &&
1116                     semnum == suptr->un_ent[ix].un_num &&
1117                     adjval == suptr->un_ent[ix].un_adjval) {
1118                         /*
1119                          * Only do assertions when we aren't in a SMP race.
1120                          */
1121                         if ((semaptr->ds.sem_perm.mode & SEM_ALLOC) == 0)
1122                                 panic("semexit - semid not allocated");
1123                         if (semnum >= semaptr->ds.sem_nsems)
1124                                 panic("semexit - semnum out of range");
1125                         --suptr->un_cnt;
1126
1127                         if (adjval < 0) {
1128                                 if (semptr->semval < -adjval)
1129                                         semptr->semval = 0;
1130                                 else
1131                                         semptr->semval += adjval;
1132                         } else {
1133                                 semptr->semval += adjval;
1134                         }
1135                         wakeup(semptr);
1136                 }
1137                 lwkt_relpooltoken(semptr);
1138                 lockmgr(&semaptr->lk, LK_RELEASE);
1139         }
1140
1141         /*
1142          * Final cleanup, remove from the list and deallocate on the
1143          * last ref only.
1144          */
1145         lwkt_gettoken(&semu_token);
1146         if (--suptr->un_refs == 0) {
1147                 TAILQ_REMOVE(&semu_list, suptr, un_entry);
1148                 KKASSERT(suptr->un_cnt == 0);
1149                 kfree(suptr, M_SEM);
1150         }
1151         lwkt_reltoken(&semu_token);
1152 }