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 /* XXX casting to (sy_call_t *) is bogus, as usual. */
  36 static sy_call_t *semcalls[] = {
  37         (sy_call_t *)sys___semctl, (sy_call_t *)sys_semget,
  38         (sy_call_t *)sys_semop
  39 };
  40
  41 static struct lwkt_token semu_token = LWKT_TOKEN_INITIALIZER(semu_token);
  42 static int      semtot = 0;
  43 static struct semid_pool *sema; /* semaphore id pool */
  44 static TAILQ_HEAD(, sem_undo) semu_list = TAILQ_HEAD_INITIALIZER(semu_list);
  45 static struct lock sema_lk;
  46
  47 struct sem {
  48         u_short semval;         /* semaphore value */
  49         pid_t   sempid;         /* pid of last operation */
  50         u_short semncnt;        /* # awaiting semval > cval */
  51         u_short semzcnt;        /* # awaiting semval = 0 */
  52 };
  53
  54 /*
  55  * Undo structure (one per process)
  56  */
  57 struct sem_undo {
  58         TAILQ_ENTRY(sem_undo) un_entry; /* linked list for semundo_clear() */
  59         struct  proc *un_proc;          /* owner of this structure */
  60         int     un_refs;                /* prevent unlink/kfree */
  61         short   un_cnt;                 /* # of active entries */
  62         short   un_unused;
  63         struct undo {
  64                 short   un_adjval;      /* adjust on exit values */
  65                 short   un_num;         /* semaphore # */
  66                 int     un_id;          /* semid */
  67         } un_ent[1];                    /* undo entries */
  68 };
  69
  70 /*
  71  * Configuration parameters
  72  */
  73 #ifndef SEMMNI
  74 #define SEMMNI  1024            /* # of semaphore identifiers */
  75 #endif
  76 #ifndef SEMMNS
  77 #define SEMMNS  32767           /* # of semaphores in system */
  78 #endif
  79 #ifndef SEMUME
  80 #define SEMUME  25              /* max # of undo entries per process */
  81 #endif
  82 #ifndef SEMMNU
  83 #define SEMMNU  1024            /* # of undo structures in system */
  84                                 /* NO LONGER USED */
  85 #endif
  86
  87 /* shouldn't need tuning */
  88 #ifndef SEMMAP
  89 #define SEMMAP  128             /* # of entries in semaphore map */
  90 #endif
  91 #ifndef SEMMSL
  92 #define SEMMSL  SEMMNS          /* max # of semaphores per id */
  93 #endif
  94 #ifndef SEMOPM
  95 #define SEMOPM  100             /* max # of operations per semop call */
  96 #endif
  97
  98 #define SEMVMX  32767           /* semaphore maximum value */
  99 #define SEMAEM  16384           /* adjust on exit max value */
 100
 101 /*
 102  * Due to the way semaphore memory is allocated, we have to ensure that
 103  * SEMUSZ is properly aligned.
 104  */
 105
 106 #define SEM_ALIGN(bytes) (((bytes) + (sizeof(long) - 1)) & ~(sizeof(long) - 1))
 107
 108 /* actual size of an undo structure */
 109 #define SEMUSZ(nent)    SEM_ALIGN(offsetof(struct sem_undo, un_ent[nent]))
 110
 111 /*
 112  * semaphore info struct
 113  */
 114 struct seminfo seminfo = {
 115                 SEMMAP,         /* # of entries in semaphore map */
 116                 SEMMNI,         /* # of semaphore identifiers */
 117                 SEMMNS,         /* # of semaphores in system */
 118                 SEMMNU,         /* # of undo structures in system */
 119                 SEMMSL,         /* max # of semaphores per id */
 120                 SEMOPM,         /* max # of operations per semop call */
 121                 SEMUME,         /* max # of undo entries per process */
 122                 SEMUSZ(SEMUME), /* size in bytes of undo structure */
 123                 SEMVMX,         /* semaphore maximum value */
 124                 SEMAEM          /* adjust on exit max value */
 125 };
 126
 127 TUNABLE_INT("kern.ipc.semmap", &seminfo.semmap);
 128 TUNABLE_INT("kern.ipc.semmni", &seminfo.semmni);
 129 TUNABLE_INT("kern.ipc.semmns", &seminfo.semmns);
 130 TUNABLE_INT("kern.ipc.semmnu", &seminfo.semmnu);
 131 TUNABLE_INT("kern.ipc.semmsl", &seminfo.semmsl);
 132 TUNABLE_INT("kern.ipc.semopm", &seminfo.semopm);
 133 TUNABLE_INT("kern.ipc.semume", &seminfo.semume);
 134 TUNABLE_INT("kern.ipc.semusz", &seminfo.semusz);
 135 TUNABLE_INT("kern.ipc.semvmx", &seminfo.semvmx);
 136 TUNABLE_INT("kern.ipc.semaem", &seminfo.semaem);
 137
 138 SYSCTL_INT(_kern_ipc, OID_AUTO, semmap, CTLFLAG_RW, &seminfo.semmap, 0,
 139     "Number of entries in semaphore map");
 140 SYSCTL_INT(_kern_ipc, OID_AUTO, semmni, CTLFLAG_RD, &seminfo.semmni, 0,
 141     "Number of semaphore identifiers");
 142 SYSCTL_INT(_kern_ipc, OID_AUTO, semmns, CTLFLAG_RD, &seminfo.semmns, 0,
 143     "Total number of semaphores");
 144 SYSCTL_INT(_kern_ipc, OID_AUTO, semmnu, CTLFLAG_RD, &seminfo.semmnu, 0,
 145     "Total number of undo structures");
 146 SYSCTL_INT(_kern_ipc, OID_AUTO, semmsl, CTLFLAG_RW, &seminfo.semmsl, 0,
 147     "Max number of semaphores per id");
 148 SYSCTL_INT(_kern_ipc, OID_AUTO, semopm, CTLFLAG_RD, &seminfo.semopm, 0,
 149     "Max number of operations per semop call");
 150 SYSCTL_INT(_kern_ipc, OID_AUTO, semume, CTLFLAG_RD, &seminfo.semume, 0,
 151     "Max number of undo entries per process");
 152 SYSCTL_INT(_kern_ipc, OID_AUTO, semusz, CTLFLAG_RD, &seminfo.semusz, 0,
 153     "Size in bytes of undo structure");
 154 SYSCTL_INT(_kern_ipc, OID_AUTO, semvmx, CTLFLAG_RW, &seminfo.semvmx, 0,
 155     "Semaphore maximum value");
 156 SYSCTL_INT(_kern_ipc, OID_AUTO, semaem, CTLFLAG_RW, &seminfo.semaem, 0,
 157     "Adjust on exit max value");
 158
 159 #if 0
 160 RO seminfo.semmap       /* SEMMAP unused */
 161 RO seminfo.semmni
 162 RO seminfo.semmns
 163 RO seminfo.semmnu       /* undo entries per system */
 164 RW seminfo.semmsl
 165 RO seminfo.semopm       /* SEMOPM unused */
 166 RO seminfo.semume
 167 RO seminfo.semusz       /* param - derived from SEMUME for per-proc sizeof */
 168 RO seminfo.semvmx       /* SEMVMX unused - user param */
 169 RO seminfo.semaem       /* SEMAEM unused - user param */
 170 #endif
 171
 172 static void
 173 seminit(void *dummy)
 174 {
 175         int i;
 176
 177         sema = kmalloc(sizeof(struct semid_pool) * seminfo.semmni,
 178                       M_SEM, M_WAITOK | M_ZERO);
 179
 180         lockinit(&sema_lk, "semglb", 0, 0);
 181         for (i = 0; i < seminfo.semmni; i++) {
 182                 struct semid_pool *semaptr = &sema[i];
 183
 184                 lockinit(&semaptr->lk, "semary", 0, 0);
 185                 semaptr->ds.sem_base = NULL;
 186                 semaptr->ds.sem_perm.mode = 0;
 187         }
 188 }
 189 SYSINIT(sysv_sem, SI_SUB_SYSV_SEM, SI_ORDER_FIRST, seminit, NULL)
 190
 191 /*
 192  * Entry point for all SEM calls
 193  *
 194  * semsys_args(int which, a2, a3, ...) (VARARGS)
 195  *
 196  * MPALMOSTSAFE
 197  */
 198 int
 199 sys_semsys(struct semsys_args *uap)
 200 {
 201         struct thread *td = curthread;
 202         unsigned int which = (unsigned int)uap->which;
 203         int error;
 204
 205         if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
 206                 return (ENOSYS);
 207
 208         if (which >= NELEM(semcalls))
 209                 return (EINVAL);
 210         bcopy(&uap->a2, &uap->which,
 211               sizeof(struct semsys_args) - offsetof(struct semsys_args, a2));
 212         error = (*semcalls[which])(uap);
 213         return (error);
 214 }
 215
 216 /*
 217  * Allocate a new sem_undo structure for a process
 218  * (returns ptr to structure or NULL if no more room)
 219  */
 220 static struct sem_undo *
 221 semu_alloc(struct proc *p)
 222 {
 223         struct sem_undo *semu;
 224
 225         /*
 226          * Allocate the semu structure and associate it with the process,
 227          * as necessary.
 228          */
 229         while ((semu = p->p_sem_undo) == NULL) {
 230                 semu = kmalloc(SEMUSZ(seminfo.semume), M_SEM,
 231                                M_WAITOK | M_ZERO);
 232                 lwkt_gettoken(&semu_token);
 233                 lwkt_gettoken(&p->p_token);
 234                 if (p->p_sem_undo == NULL) {
 235                         p->p_sem_undo = semu;
 236                         p->p_flags |= P_SYSVSEM;
 237                         semu->un_proc = p;
 238                         TAILQ_INSERT_TAIL(&semu_list, semu, un_entry);
 239                 } else {
 240                         kfree(semu, M_SEM);
 241                 }
 242                 lwkt_reltoken(&p->p_token);
 243                 lwkt_reltoken(&semu_token);
 244         }
 245         return(semu);
 246 }
 247
 248 /*
 249  * Adjust a particular entry for a particular proc
 250  */
 251 static int
 252 semundo_adjust(struct proc *p, int semid, int semnum, int adjval)
 253 {
 254         struct sem_undo *suptr;
 255         struct undo *sunptr;
 256         int i;
 257         int error = 0;
 258
 259         /*
 260          * Look for and remember the sem_undo if the caller doesn't
 261          * provide it.
 262          */
 263         suptr = semu_alloc(p);
 264         lwkt_gettoken(&p->p_token);
 265
 266         /*
 267          * Look for the requested entry and adjust it (delete if adjval becomes
 268          * 0).
 269          */
 270         sunptr = &suptr->un_ent[0];
 271         for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
 272                 if (sunptr->un_id != semid || sunptr->un_num != semnum)
 273                         continue;
 274                 if (adjval == 0)
 275                         sunptr->un_adjval = 0;
 276                 else
 277                         sunptr->un_adjval += adjval;
 278                 if (sunptr->un_adjval == 0) {
 279                         suptr->un_cnt--;
 280                         if (i < suptr->un_cnt)
 281                                 suptr->un_ent[i] = suptr->un_ent[suptr->un_cnt];
 282                 }
 283                 goto done;
 284         }
 285
 286         /* Didn't find the right entry - create it */
 287         if (adjval == 0)
 288                 goto done;
 289         if (suptr->un_cnt != seminfo.semume) {
 290                 sunptr = &suptr->un_ent[suptr->un_cnt];
 291                 suptr->un_cnt++;
 292                 sunptr->un_adjval = adjval;
 293                 sunptr->un_id = semid;
 294                 sunptr->un_num = semnum;
 295         } else {
 296                 error = EINVAL;
 297         }
 298 done:
 299         lwkt_reltoken(&p->p_token);
 300
 301         return (error);
 302 }
 303
 304 /*
 305  * This is rather expensive
 306  */
 307 static void
 308 semundo_clear(int semid, int semnum)
 309 {
 310         struct proc *p;
 311         struct sem_undo *suptr;
 312         struct sem_undo *sunext;
 313         struct undo *sunptr;
 314         int i;
 315
 316         lwkt_gettoken(&semu_token);
 317         sunext = TAILQ_FIRST(&semu_list);
 318         while ((suptr = sunext) != NULL) {
 319                 if ((p = suptr->un_proc) == NULL) {
 320                         suptr = TAILQ_NEXT(suptr, un_entry);
 321                         continue;
 322                 }
 323                 ++suptr->un_refs;
 324                 PHOLD(p);
 325                 lwkt_gettoken(&p->p_token);
 326
 327                 sunptr = &suptr->un_ent[0];
 328                 i = 0;
 329
 330                 while (i < suptr->un_cnt) {
 331                         if (sunptr->un_id == semid) {
 332                                 if (semnum == -1 || sunptr->un_num == semnum) {
 333                                         suptr->un_cnt--;
 334                                         if (i < suptr->un_cnt) {
 335                                                 suptr->un_ent[i] =
 336                                                   suptr->un_ent[suptr->un_cnt];
 337                                                 /*
 338                                                  * do not increment i
 339                                                  * or sunptr after copydown.
 340                                                  */
 341                                                 continue;
 342                                         }
 343                                 }
 344                                 if (semnum != -1)
 345                                         break;
 346                         }
 347                         ++i;
 348                         ++sunptr;
 349                 }
 350
 351                 lwkt_reltoken(&p->p_token);
 352                 PRELE(p);
 353
 354                 /*
 355                  * Handle deletion races
 356                  */
 357                 sunext = TAILQ_NEXT(suptr, un_entry);
 358                 if (--suptr->un_refs == 0 && suptr->un_proc == NULL) {
 359                         KKASSERT(suptr->un_cnt == 0);
 360                         TAILQ_REMOVE(&semu_list, suptr, un_entry);
 361                         kfree(suptr, M_SEM);
 362                 }
 363         }
 364         lwkt_reltoken(&semu_token);
 365 }
 366
 367 /*
 368  * Note that the user-mode half of this passes a union, not a pointer
 369  *
 370  * MPALMOSTSAFE
 371  */
 372 int
 373 sys___semctl(struct __semctl_args *uap)
 374 {
 375         struct thread *td = curthread;
 376         int semid = uap->semid;
 377         int semnum = uap->semnum;
 378         int cmd = uap->cmd;
 379         union semun *arg = uap->arg;
 380         union semun real_arg;
 381         struct ucred *cred = td->td_ucred;
 382         int i, rval, eval;
 383         struct semid_ds sbuf;
 384         struct semid_pool *semaptr;
 385         struct semid_pool *semakptr;
 386         struct sem *semptr;
 387
 388 #ifdef SEM_DEBUG
 389         kprintf("call to semctl(%d, %d, %d, 0x%x)\n", semid, semnum, cmd, arg);
 390 #endif
 391
 392         if (!jail_sysvipc_allowed && cred->cr_prison != NULL)
 393                 return (ENOSYS);
 394
 395         switch (cmd) {
 396         case SEM_STAT:
 397                 /*
 398                  * For this command we assume semid is an array index
 399                  * rather than an IPC id.
 400                  */
 401                 if (semid < 0 || semid >= seminfo.semmni) {
 402                         eval = EINVAL;
 403                         break;
 404                 }
 405                 semakptr = &sema[semid];
 406                 lockmgr(&semakptr->lk, LK_EXCLUSIVE);
 407                 if ((semakptr->ds.sem_perm.mode & SEM_ALLOC) == 0) {
 408                         eval = EINVAL;
 409                         lockmgr(&semakptr->lk, LK_RELEASE);
 410                         break;
 411                 }
 412                 if ((eval = ipcperm(td->td_proc, &semakptr->ds.sem_perm, IPC_R))) {
 413                         lockmgr(&semakptr->lk, LK_RELEASE);
 414                         break;
 415                 }
 416                 bcopy(&semakptr->ds, arg->buf, sizeof(struct semid_ds));
 417                 rval = IXSEQ_TO_IPCID(semid, semakptr->ds.sem_perm);
 418                 lockmgr(&semakptr->lk, LK_RELEASE);
 419                 break;
 420         }
 421
 422         semid = IPCID_TO_IX(semid);
 423         if (semid < 0 || semid >= seminfo.semmni) {
 424                 return(EINVAL);
 425         }
 426         semaptr = &sema[semid];
 427         lockmgr(&semaptr->lk, LK_EXCLUSIVE);
 428
 429         if ((semaptr->ds.sem_perm.mode & SEM_ALLOC) == 0 ||
 430             semaptr->ds.sem_perm.seq != IPCID_TO_SEQ(uap->semid)) {
 431                 lockmgr(&semaptr->lk, LK_RELEASE);
 432                 return(EINVAL);
 433         }
 434
 435         eval = 0;
 436         rval = 0;
 437
 438         switch (cmd) {
 439         case IPC_RMID:
 440                 eval = ipcperm(td->td_proc, &semaptr->ds.sem_perm, IPC_M);
 441                 if (eval != 0)
 442                         break;
 443                 semaptr->ds.sem_perm.cuid = cred->cr_uid;
 444                 semaptr->ds.sem_perm.uid = cred->cr_uid;
 445
 446                 /*
 447                  * NOTE: Nobody will be waiting on the semaphores since
 448                  *       we have an exclusive lock on semaptr->lk).
 449                  */
 450                 lockmgr(&sema_lk, LK_EXCLUSIVE);
 451                 semtot -= semaptr->ds.sem_nsems;
 452                 kfree(semaptr->ds.sem_base, M_SEM);
 453                 semaptr->ds.sem_base = NULL;
 454                 semaptr->ds.sem_perm.mode = 0;  /* clears SEM_ALLOC */
 455                 lockmgr(&sema_lk, LK_RELEASE);
 456
 457                 semundo_clear(semid, -1);
 458                 break;
 459
 460         case IPC_SET:
 461                 eval = ipcperm(td->td_proc, &semaptr->ds.sem_perm, IPC_M);
 462                 if (eval)
 463                         break;
 464                 if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
 465                         break;
 466                 if ((eval = copyin(real_arg.buf, (caddr_t)&sbuf,
 467                                    sizeof(sbuf))) != 0) {
 468                         break;
 469                 }
 470                 semaptr->ds.sem_perm.uid = sbuf.sem_perm.uid;
 471                 semaptr->ds.sem_perm.gid = sbuf.sem_perm.gid;
 472                 semaptr->ds.sem_perm.mode =
 473                         (semaptr->ds.sem_perm.mode & ~0777) |
 474                         (sbuf.sem_perm.mode & 0777);
 475                 semaptr->ds.sem_ctime = time_second;
 476                 break;
 477
 478         case IPC_STAT:
 479                 eval = ipcperm(td->td_proc, &semaptr->ds.sem_perm, IPC_R);
 480                 if (eval)
 481                         break;
 482                 if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
 483                         break;
 484                 eval = copyout(&semaptr->ds, real_arg.buf,
 485                                sizeof(struct semid_ds));
 486                 break;
 487
 488         case GETNCNT:
 489                 eval = ipcperm(td->td_proc, &semaptr->ds.sem_perm, IPC_R);
 490                 if (eval)
 491                         break;
 492                 if (semnum < 0 || semnum >= semaptr->ds.sem_nsems) {
 493                         eval = EINVAL;
 494                         break;
 495                 }
 496                 rval = semaptr->ds.sem_base[semnum].semncnt;
 497                 break;
 498
 499         case GETPID:
 500                 eval = ipcperm(td->td_proc, &semaptr->ds.sem_perm, IPC_R);
 501                 if (eval)
 502                         break;
 503                 if (semnum < 0 || semnum >= semaptr->ds.sem_nsems) {
 504                         eval = EINVAL;
 505                         break;
 506                 }
 507                 rval = semaptr->ds.sem_base[semnum].sempid;
 508                 break;
 509
 510         case GETVAL:
 511                 eval = ipcperm(td->td_proc, &semaptr->ds.sem_perm, IPC_R);
 512                 if (eval)
 513                         break;
 514                 if (semnum < 0 || semnum >= semaptr->ds.sem_nsems) {
 515                         eval = EINVAL;
 516                         break;
 517                 }
 518                 rval = semaptr->ds.sem_base[semnum].semval;
 519                 break;
 520
 521         case GETALL:
 522                 eval = ipcperm(td->td_proc, &semaptr->ds.sem_perm, IPC_R);
 523                 if (eval)
 524                         break;
 525                 if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
 526                         break;
 527                 for (i = 0; i < semaptr->ds.sem_nsems; i++) {
 528                         eval = copyout(&semaptr->ds.sem_base[i].semval,
 529                                        &real_arg.array[i],
 530                                        sizeof(real_arg.array[0]));
 531                         if (eval)
 532                                 break;
 533                 }
 534                 break;
 535
 536         case GETZCNT:
 537                 eval = ipcperm(td->td_proc, &semaptr->ds.sem_perm, IPC_R);
 538                 if (eval)
 539                         break;
 540                 if (semnum < 0 || semnum >= semaptr->ds.sem_nsems) {
 541                         eval = EINVAL;
 542                         break;
 543                 }
 544                 rval = semaptr->ds.sem_base[semnum].semzcnt;
 545                 break;
 546
 547         case SETVAL:
 548                 eval = ipcperm(td->td_proc, &semaptr->ds.sem_perm, IPC_W);
 549                 if (eval)
 550                         break;
 551                 if (semnum < 0 || semnum >= semaptr->ds.sem_nsems) {
 552                         eval = EINVAL;
 553                         break;
 554                 }
 555                 if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
 556                         break;
 557
 558                 /*
 559                  * Because we hold semaptr->lk exclusively we can safely
 560                  * modify any semptr content without acquiring its token.
 561                  */
 562                 semptr = &semaptr->ds.sem_base[semnum];
 563                 semptr->semval = real_arg.val;
 564                 semundo_clear(semid, semnum);
 565                 if (semptr->semzcnt || semptr->semncnt)
 566                         wakeup(semptr);
 567                 break;
 568
 569         case SETALL:
 570                 eval = ipcperm(td->td_proc, &semaptr->ds.sem_perm, IPC_W);
 571                 if (eval)
 572                         break;
 573                 if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
 574                         break;
 575                 /*
 576                  * Because we hold semaptr->lk exclusively we can safely
 577                  * modify any semptr content without acquiring its token.
 578                  */
 579                 for (i = 0; i < semaptr->ds.sem_nsems; i++) {
 580                         semptr = &semaptr->ds.sem_base[i];
 581                         eval = copyin(&real_arg.array[i],
 582                                       (caddr_t)&semptr->semval,
 583                                       sizeof(real_arg.array[0]));
 584                         if (semptr->semzcnt || semptr->semncnt)
 585                                 wakeup(semptr);
 586                         if (eval != 0)
 587                                 break;
 588                 }
 589                 semundo_clear(semid, -1);
 590                 break;
 591
 592         default:
 593                 eval = EINVAL;
 594                 break;
 595         }
 596         lockmgr(&semaptr->lk, LK_RELEASE);
 597
 598         if (eval == 0)
 599                 uap->sysmsg_result = rval;
 600         return(eval);
 601 }
 602
 603 /*
 604  * MPALMOSTSAFE
 605  */
 606 int
 607 sys_semget(struct semget_args *uap)
 608 {
 609         struct thread *td = curthread;
 610         int semid, eval;
 611         int key = uap->key;
 612         int nsems = uap->nsems;
 613         int semflg = uap->semflg;
 614         struct ucred *cred = td->td_ucred;
 615
 616 #ifdef SEM_DEBUG
 617         kprintf("semget(0x%x, %d, 0%o)\n", key, nsems, semflg);
 618 #endif
 619
 620         if (!jail_sysvipc_allowed && cred->cr_prison != NULL)
 621                 return (ENOSYS);
 622
 623         eval = 0;
 624
 625         if (key != IPC_PRIVATE) {
 626                 for (semid = 0; semid < seminfo.semmni; semid++) {
 627                         if ((sema[semid].ds.sem_perm.mode & SEM_ALLOC) == 0 ||
 628                             sema[semid].ds.sem_perm.key != key) {
 629                                 continue;
 630                         }
 631                         lockmgr(&sema[semid].lk, LK_EXCLUSIVE);
 632                         if ((sema[semid].ds.sem_perm.mode & SEM_ALLOC) == 0 ||
 633                             sema[semid].ds.sem_perm.key != key) {
 634                                 lockmgr(&sema[semid].lk, LK_RELEASE);
 635                                 continue;
 636                         }
 637                         break;
 638                 }
 639                 if (semid < seminfo.semmni) {
 640                         /* sema[semid].lk still locked from above */
 641 #ifdef SEM_DEBUG
 642                         kprintf("found public key\n");
 643 #endif
 644                         if ((eval = ipcperm(td->td_proc,
 645                                             &sema[semid].ds.sem_perm,
 646                                             semflg & 0700))) {
 647                                 lockmgr(&sema[semid].lk, LK_RELEASE);
 648                                 goto done;
 649                         }
 650                         if (nsems > 0 && sema[semid].ds.sem_nsems < nsems) {
 651 #ifdef SEM_DEBUG
 652                                 kprintf("too small\n");
 653 #endif
 654                                 eval = EINVAL;
 655                                 lockmgr(&sema[semid].lk, LK_RELEASE);
 656                                 goto done;
 657                         }
 658                         if ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) {
 659 #ifdef SEM_DEBUG
 660                                 kprintf("not exclusive\n");
 661 #endif
 662                                 eval = EEXIST;
 663                                 lockmgr(&sema[semid].lk, LK_RELEASE);
 664                                 goto done;
 665                         }
 666
 667                         /*
 668                          * Return this one.
 669                          */
 670                         lockmgr(&sema[semid].lk, LK_RELEASE);
 671                         goto done;
 672                 }
 673         }
 674
 675 #ifdef SEM_DEBUG
 676         kprintf("need to allocate the semid_ds\n");
 677 #endif
 678         if (key == IPC_PRIVATE || (semflg & IPC_CREAT)) {
 679                 if (nsems <= 0 || nsems > seminfo.semmsl) {
 680 #ifdef SEM_DEBUG
 681                         kprintf("nsems out of range (0<%d<=%d)\n",
 682                                 nsems, seminfo.semmsl);
 683 #endif
 684                         eval = EINVAL;
 685                         goto done;
 686                 }
 687
 688                 /*
 689                  * SEM_ALLOC flag cannot be set unless sema_lk is locked.
 690                  * semtot field also protected by sema_lk.
 691                  */
 692                 lockmgr(&sema_lk, LK_EXCLUSIVE);
 693                 if (nsems > seminfo.semmns - semtot) {
 694 #ifdef SEM_DEBUG
 695                         kprintf("not enough semaphores left "
 696                                 "(need %d, got %d)\n",
 697                                 nsems, seminfo.semmns - semtot);
 698 #endif
 699                         eval = ENOSPC;
 700                         lockmgr(&sema_lk, LK_RELEASE);
 701                         goto done;
 702                 }
 703                 for (semid = 0; semid < seminfo.semmni; semid++) {
 704                         if ((sema[semid].ds.sem_perm.mode & SEM_ALLOC) == 0)
 705                                 break;
 706                 }
 707                 if (semid == seminfo.semmni) {
 708 #ifdef SEM_DEBUG
 709                         kprintf("no more semid_ds's available\n");
 710 #endif
 711                         eval = ENOSPC;
 712                         lockmgr(&sema_lk, LK_RELEASE);
 713                         goto done;
 714                 }
 715 #ifdef SEM_DEBUG
 716                 kprintf("semid %d is available\n", semid);
 717 #endif
 718                 lockmgr(&sema[semid].lk, LK_EXCLUSIVE);
 719                 sema[semid].ds.sem_perm.key = key;
 720                 sema[semid].ds.sem_perm.cuid = cred->cr_uid;
 721                 sema[semid].ds.sem_perm.uid = cred->cr_uid;
 722                 sema[semid].ds.sem_perm.cgid = cred->cr_gid;
 723                 sema[semid].ds.sem_perm.gid = cred->cr_gid;
 724                 sema[semid].ds.sem_perm.mode = (semflg & 0777) | SEM_ALLOC;
 725                 sema[semid].ds.sem_perm.seq =
 726                     (sema[semid].ds.sem_perm.seq + 1) & 0x7fff;
 727                 sema[semid].ds.sem_nsems = nsems;
 728                 sema[semid].ds.sem_otime = 0;
 729                 sema[semid].ds.sem_ctime = time_second;
 730                 sema[semid].ds.sem_base = kmalloc(sizeof(struct sem) * nsems,
 731                                                M_SEM, M_WAITOK|M_ZERO);
 732                 semtot += nsems;
 733                 ++sema[semid].gen;
 734                 lockmgr(&sema[semid].lk, LK_RELEASE);
 735                 lockmgr(&sema_lk, LK_RELEASE);
 736 #ifdef SEM_DEBUG
 737                 kprintf("sembase = 0x%x, next = 0x%x\n",
 738                         sema[semid].ds.sem_base, &sem[semtot]);
 739 #endif
 740                 /* eval == 0 */
 741         } else {
 742 #ifdef SEM_DEBUG
 743                 kprintf("didn't find it and wasn't asked to create it\n");
 744 #endif
 745                 eval = ENOENT;
 746         }
 747
 748 done:
 749         if (eval == 0) {
 750                 uap->sysmsg_result =
 751                         IXSEQ_TO_IPCID(semid, sema[semid].ds.sem_perm);
 752         }
 753         return(eval);
 754 }
 755
 756 /*
 757  * MPSAFE
 758  */
 759 int
 760 sys_semop(struct semop_args *uap)
 761 {
 762         struct thread *td = curthread;
 763         int semid = uap->semid;
 764         u_int nsops = uap->nsops;
 765         struct sembuf sops[MAX_SOPS];
 766         struct semid_pool *semaptr;
 767         struct sembuf *sopptr;
 768         struct sem *semptr;
 769         struct sem *xsemptr;
 770         int i, j, eval;
 771         int do_undos;
 772
 773 #ifdef SEM_DEBUG
 774         kprintf("call to semop(%d, 0x%x, %u)\n", semid, sops, nsops);
 775 #endif
 776         if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
 777                 return (ENOSYS);
 778
 779         semid = IPCID_TO_IX(semid);     /* Convert back to zero origin */
 780
 781         if (semid < 0 || semid >= seminfo.semmni) {
 782                 eval = EINVAL;
 783                 goto done2;
 784         }
 785
 786         wakeup_start_delayed();
 787         semaptr = &sema[semid];
 788         lockmgr(&semaptr->lk, LK_SHARED);
 789
 790         if ((semaptr->ds.sem_perm.mode & SEM_ALLOC) == 0) {
 791                 eval = EINVAL;
 792                 goto done;
 793         }
 794         if (semaptr->ds.sem_perm.seq != IPCID_TO_SEQ(uap->semid)) {
 795                 eval = EINVAL;
 796                 goto done;
 797         }
 798
 799         if ((eval = ipcperm(td->td_proc, &semaptr->ds.sem_perm, IPC_W))) {
 800 #ifdef SEM_DEBUG
 801                 kprintf("eval = %d from ipaccess\n", eval);
 802 #endif
 803                 goto done;
 804         }
 805
 806         if (nsops > MAX_SOPS) {
 807 #ifdef SEM_DEBUG
 808                 kprintf("too many sops (max=%d, nsops=%u)\n", MAX_SOPS, nsops);
 809 #endif
 810                 eval = E2BIG;
 811                 goto done;
 812         }
 813
 814         if ((eval = copyin(uap->sops, &sops, nsops * sizeof(sops[0]))) != 0) {
 815 #ifdef SEM_DEBUG
 816                 kprintf("eval = %d from copyin(%08x, %08x, %u)\n", eval,
 817                     uap->sops, &sops, nsops * sizeof(sops[0]));
 818 #endif
 819                 goto done;
 820         }
 821
 822         /*
 823          * Loop trying to satisfy the vector of requests.
 824          * If we reach a point where we must wait, any requests already
 825          * performed are rolled back and we go to sleep until some other
 826          * process wakes us up.  At this point, we start all over again.
 827          *
 828          * This ensures that from the perspective of other tasks, a set
 829          * of requests is atomic (never partially satisfied).
 830          */
 831         do_undos = 0;
 832
 833         for (;;) {
 834                 long gen;
 835
 836                 semptr = NULL;
 837
 838                 for (i = 0; i < nsops; i++) {
 839                         sopptr = &sops[i];
 840
 841                         if (sopptr->sem_num >= semaptr->ds.sem_nsems) {
 842                                 eval = EFBIG;
 843                                 goto done;
 844                         }
 845
 846                         semptr = &semaptr->ds.sem_base[sopptr->sem_num];
 847                         lwkt_getpooltoken(semptr);
 848
 849 #ifdef SEM_DEBUG
 850                         kprintf("semop:  semaptr=%x, sem_base=%x, semptr=%x, "
 851                                 "sem[%d]=%d : op=%d, flag=%s\n",
 852                             semaptr, semaptr->ds.sem_base, semptr,
 853                             sopptr->sem_num, semptr->semval, sopptr->sem_op,
 854                             (sopptr->sem_flg & IPC_NOWAIT) ? "nowait" : "wait");
 855 #endif
 856
 857                         if (sopptr->sem_op < 0) {
 858                                 if (semptr->semval + sopptr->sem_op < 0) {
 859 #ifdef SEM_DEBUG
 860                                         kprintf("semop:  can't do it now\n");
 861 #endif
 862                                         break;
 863                                 } else {
 864                                         semptr->semval += sopptr->sem_op;
 865                                         if (semptr->semval == 0 &&
 866                                             semptr->semzcnt > 0) {
 867                                                 wakeup(semptr);
 868                                         }
 869                                 }
 870                                 if (sopptr->sem_flg & SEM_UNDO)
 871                                         do_undos = 1;
 872                         } else if (sopptr->sem_op == 0) {
 873                                 if (semptr->semval > 0) {
 874 #ifdef SEM_DEBUG
 875                                         kprintf("semop:  not zero now\n");
 876 #endif
 877                                         break;
 878                                 }
 879                         } else {
 880                                 semptr->semval += sopptr->sem_op;
 881                                 if (sopptr->sem_flg & SEM_UNDO)
 882                                         do_undos = 1;
 883                                 if (semptr->semncnt > 0)
 884                                         wakeup(semptr);
 885                         }
 886                         lwkt_relpooltoken(semptr);
 887                 }
 888
 889                 /*
 890                  * Did we get through the entire vector?
 891                  */
 892                 if (i >= nsops)
 893                         goto donex;
 894
 895                 /*
 896                  * No, protect the semaphore request which also flags that
 897                  * a wakeup is needed, then release semptr since we know
 898                  * another process is likely going to need to access it
 899                  * soon.
 900                  */
 901                 if (sopptr->sem_op == 0)
 902                         semptr->semzcnt++;
 903                 else
 904                         semptr->semncnt++;
 905                 tsleep_interlock(semptr, PCATCH);
 906                 lwkt_relpooltoken(semptr);
 907
 908                 /*
 909                  * Rollback the semaphores we had acquired.
 910                  */
 911 #ifdef SEM_DEBUG
 912                 kprintf("semop:  rollback 0 through %d\n", i-1);
 913 #endif
 914                 for (j = 0; j < i; j++) {
 915                         xsemptr = &semaptr->ds.sem_base[sops[j].sem_num];
 916                         lwkt_getpooltoken(xsemptr);
 917                         xsemptr->semval -= sops[j].sem_op;
 918                         if (xsemptr->semval == 0 && xsemptr->semzcnt > 0)
 919                                 wakeup(xsemptr);
 920                         if (xsemptr->semval <= 0 && xsemptr->semncnt > 0)
 921                                 wakeup(xsemptr);
 922                         lwkt_relpooltoken(xsemptr);
 923                 }
 924
 925                 /*
 926                  * If the request that we couldn't satisfy has the
 927                  * NOWAIT flag set then return with EAGAIN.
 928                  */
 929                 if (sopptr->sem_flg & IPC_NOWAIT) {
 930                         eval = EAGAIN;
 931                         goto done;
 932                 }
 933
 934                 /*
 935                  * Release semaptr->lk while sleeping, allowing other
 936                  * semops (like SETVAL, SETALL, etc), which require an
 937                  * exclusive lock and might wake us up.
 938                  *
 939                  * Reload and recheck the validity of semaptr on return.
 940                  * Note that semptr itself might have changed too, but
 941                  * we've already interlocked for semptr and that is what
 942                  * will be woken up if it wakes up the tsleep on a MP
 943                  * race.
 944                  *
 945                  * gen protects against destroy/re-create races where the
 946                  * creds match.
 947                  */
 948 #ifdef SEM_DEBUG
 949                 kprintf("semop:  good night!\n");
 950 #endif
 951                 gen = semaptr->gen;
 952                 lockmgr(&semaptr->lk, LK_RELEASE);
 953                 eval = tsleep(semptr, PCATCH | PINTERLOCKED, "semwait", hz);
 954                 lockmgr(&semaptr->lk, LK_SHARED);
 955 #ifdef SEM_DEBUG
 956                 kprintf("semop:  good morning (eval=%d)!\n", eval);
 957 #endif
 958
 959                 /* return code is checked below, after sem[nz]cnt-- */
 960
 961                 /*
 962                  * Make sure that the semaphore still exists
 963                  */
 964                 if (semaptr->gen != gen ||
 965                     (semaptr->ds.sem_perm.mode & SEM_ALLOC) == 0 ||
 966                     semaptr->ds.sem_perm.seq != IPCID_TO_SEQ(uap->semid)) {
 967                         eval = EIDRM;
 968                         goto done;
 969                 }
 970
 971                 /*
 972                  * The semaphore is still alive.  Readjust the count of
 973                  * waiting processes.
 974                  */
 975                 semptr = &semaptr->ds.sem_base[sopptr->sem_num];
 976                 lwkt_getpooltoken(semptr);
 977                 if (sopptr->sem_op == 0)
 978                         semptr->semzcnt--;
 979                 else
 980                         semptr->semncnt--;
 981                 lwkt_relpooltoken(semptr);
 982
 983                 /*
 984                  * Is it really morning, or was our sleep interrupted?
 985                  * (Delayed check of tsleep() return code because we
 986                  * need to decrement sem[nz]cnt either way.)
 987                  */
 988                 if (eval) {
 989                         eval = EINTR;
 990                         goto done;
 991                 }
 992 #ifdef SEM_DEBUG
 993                 kprintf("semop:  good morning!\n");
 994 #endif
 995                 /* RETRY LOOP */
 996         }
 997
 998 donex:
 999         /*
1000          * Process any SEM_UNDO requests.
1001          */
1002         if (do_undos) {
1003                 for (i = 0; i < nsops; i++) {
1004                         /*
1005                          * We only need to deal with SEM_UNDO's for non-zero
1006                          * op's.
1007                          */
1008                         int adjval;
1009
1010                         if ((sops[i].sem_flg & SEM_UNDO) == 0)
1011                                 continue;
1012                         adjval = sops[i].sem_op;
1013                         if (adjval == 0)
1014                                 continue;
1015                         eval = semundo_adjust(td->td_proc, semid,
1016                                               sops[i].sem_num, -adjval);
1017                         if (eval == 0)
1018                                 continue;
1019
1020                         /*
1021                          * Oh-Oh!  We ran out of either sem_undo's or undo's.
1022                          * Rollback the adjustments to this point and then
1023                          * rollback the semaphore ups and down so we can return
1024                          * with an error with all structures restored.  We
1025                          * rollback the undo's in the exact reverse order that
1026                          * we applied them.  This guarantees that we won't run
1027                          * out of space as we roll things back out.
1028                          */
1029                         for (j = i - 1; j >= 0; j--) {
1030                                 if ((sops[j].sem_flg & SEM_UNDO) == 0)
1031                                         continue;
1032                                 adjval = sops[j].sem_op;
1033                                 if (adjval == 0)
1034                                         continue;
1035                                 if (semundo_adjust(td->td_proc, semid,
1036                                                sops[j].sem_num, adjval) != 0)
1037                                         panic("semop - can't undo undos");
1038                         }
1039
1040                         for (j = 0; j < nsops; j++) {
1041                                 xsemptr = &semaptr->ds.sem_base[
1042                                                         sops[j].sem_num];
1043                                 lwkt_getpooltoken(xsemptr);
1044                                 xsemptr->semval -= sops[j].sem_op;
1045                                 if (xsemptr->semval == 0 &&
1046                                     xsemptr->semzcnt > 0)
1047                                         wakeup(xsemptr);
1048                                 if (xsemptr->semval <= 0 &&
1049                                     xsemptr->semncnt > 0)
1050                                         wakeup(xsemptr);
1051                                 lwkt_relpooltoken(xsemptr);
1052                         }
1053
1054 #ifdef SEM_DEBUG
1055                         kprintf("eval = %d from semundo_adjust\n", eval);
1056 #endif
1057                         goto done;
1058                 } /* loop through the sops */
1059         } /* if (do_undos) */
1060
1061         /* We're definitely done - set the sempid's */
1062         for (i = 0; i < nsops; i++) {
1063                 sopptr = &sops[i];
1064                 semptr = &semaptr->ds.sem_base[sopptr->sem_num];
1065                 lwkt_getpooltoken(semptr);
1066                 semptr->sempid = td->td_proc->p_pid;
1067                 lwkt_relpooltoken(semptr);
1068         }
1069
1070         /* Do a wakeup if any semaphore was up'd. */
1071 #ifdef SEM_DEBUG
1072         kprintf("semop:  done\n");
1073 #endif
1074         uap->sysmsg_result = 0;
1075         eval = 0;
1076 done:
1077         lockmgr(&semaptr->lk, LK_RELEASE);
1078         wakeup_end_delayed();
1079 done2:
1080         return(eval);
1081 }
1082
1083 /*
1084  * Go through the undo structures for this process and apply the adjustments to
1085  * semaphores.
1086  *
1087  * (p->p_token is held by the caller)
1088  */
1089 void
1090 semexit(struct proc *p)
1091 {
1092         struct sem_undo *suptr;
1093         struct sem *semptr;
1094
1095         /*
1096          * We're getting a global token, don't do it if we couldn't
1097          * possibly have any semaphores.
1098          */
1099         if ((p->p_flags & P_SYSVSEM) == 0)
1100                 return;
1101         suptr = p->p_sem_undo;
1102         KKASSERT(suptr != NULL);
1103
1104         /*
1105          * Disconnect suptr from the process and increment un_refs to
1106          * prevent anyone else from being able to destroy the structure.
1107          * Do not remove it from the linked list until after we are through
1108          * scanning it as other semaphore calls might still effect it.
1109          */
1110         lwkt_gettoken(&semu_token);
1111         p->p_sem_undo = NULL;
1112         p->p_flags &= ~P_SYSVSEM;
1113         suptr->un_proc = NULL;
1114         ++suptr->un_refs;
1115         lwkt_reltoken(&semu_token);
1116
1117         while (suptr->un_cnt) {
1118                 struct semid_pool *semaptr;
1119                 int semid;
1120                 int semnum;
1121                 int adjval;
1122                 int ix;
1123
1124                 /*
1125                  * These values are stable because we hold p->p_token.
1126                  * However, they can get ripped out from under us when
1127                  * we block or obtain other tokens so we have to re-check.
1128                  */
1129                 ix = suptr->un_cnt - 1;
1130                 semid = suptr->un_ent[ix].un_id;
1131                 semnum = suptr->un_ent[ix].un_num;
1132                 adjval = suptr->un_ent[ix].un_adjval;
1133
1134                 semaptr = &sema[semid];
1135                 if ((semaptr->ds.sem_perm.mode & SEM_ALLOC) == 0)
1136                         panic("semexit - semid not allocated");
1137                 if (semnum >= semaptr->ds.sem_nsems)
1138                         panic("semexit - semnum out of range");
1139
1140                 /*
1141                  * Recheck after locking, then execute the undo
1142                  * operation.  semptr remains valid due to the
1143                  * semaptr->lk.
1144                  */
1145                 lockmgr(&semaptr->lk, LK_SHARED);
1146                 semptr = &semaptr->ds.sem_base[semnum];
1147                 lwkt_getpooltoken(semptr);
1148
1149                 if (ix == suptr->un_cnt - 1 &&
1150                     semid == suptr->un_ent[ix].un_id &&
1151                     semnum == suptr->un_ent[ix].un_num &&
1152                     adjval == suptr->un_ent[ix].un_adjval) {
1153                         --suptr->un_cnt;
1154
1155                         if (adjval < 0) {
1156                                 if (semptr->semval < -adjval)
1157                                         semptr->semval = 0;
1158                                 else
1159                                         semptr->semval += adjval;
1160                         } else {
1161                                 semptr->semval += adjval;
1162                         }
1163                         wakeup(semptr);
1164                 }
1165                 lwkt_relpooltoken(semptr);
1166                 lockmgr(&semaptr->lk, LK_RELEASE);
1167         }
1168
1169         /*
1170          * Final cleanup, remove from the list and deallocate on the
1171          * last ref only.
1172          */
1173         lwkt_gettoken(&semu_token);
1174         if (--suptr->un_refs == 0) {
1175                 TAILQ_REMOVE(&semu_list, suptr, un_entry);
1176                 KKASSERT(suptr->un_cnt == 0);
1177                 kfree(suptr, M_SEM);
1178         }
1179         lwkt_reltoken(&semu_token);
1180 }