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