winsup/cygserver/sysv_shm.cc

   1 /*      $NetBSD: sysv_shm.c,v 1.23 1994/07/04 23:25:12 glass Exp $      */
   2 /*
   3  * Copyright (c) 1994 Adam Glass and Charles Hannum.  All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions
   7  * are met:
   8  * 1. Redistributions of source code must retain the above copyright
   9  *    notice, this list of conditions and the following disclaimer.
  10  * 2. Redistributions in binary form must reproduce the above copyright
  11  *    notice, this list of conditions and the following disclaimer in the
  12  *    documentation and/or other materials provided with the distribution.
  13  * 3. All advertising materials mentioning features or use of this software
  14  *    must display the following acknowledgement:
  15  *      This product includes software developed by Adam Glass and Charles
  16  *      Hannum.
  17  * 4. The names of the authors may not be used to endorse or promote products
  18  *    derived from this software without specific prior written permission.
  19  *
  20  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS ``AS IS'' AND ANY EXPRESS OR
  21  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
  22  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
  23  * IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  24  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
  25  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  26  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  27  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  28  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
  29  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  30  */
  31
  32 /*
  33  * This file is heavily changed to become part of Cygwin's cygserver.
  34  */
  35
  36 #ifdef __OUTSIDE_CYGWIN__
  37 #include "woutsup.h"
  38 #include <sys/cdefs.h>
  39 #ifndef __FBSDID
  40 #define __FBSDID(s)     const char version[] = (s)
  41 #endif
  42 __FBSDID("$FreeBSD: /repoman/r/ncvs/src/sys/kern/sysv_shm.c,v 1.89 2003/11/07 04:47:14 rwatson Exp $");
  43 /* CV, 2006-01-09: Inspected upstream up to version 1.104. */
  44
  45 #define _KERNEL 1
  46 #define __BSD_VISIBLE 1
  47 #include <sys/param.h>
  48 #include <sys/lock.h>
  49 #include <sys/shm.h>
  50 #include <malloc.h>
  51 #include <sys/mman.h>
  52 #include <sys/stat.h>
  53 #include <sys/sysproto.h>
  54
  55 #include <errno.h>
  56 #include <time.h>
  57 #include <unistd.h>
  58 #include "cygserver.h"
  59 #include "process.h"
  60 #include "cygserver_ipc.h"
  61
  62 #ifdef __CYGWIN__
  63 #ifndef PAGE_SIZE
  64 #define PAGE_SIZE (getpagesize ())
  65 #endif
  66 #ifndef PAGE_MASK
  67 #define PAGE_MASK (PAGE_SIZE - 1)
  68 #endif
  69 #define btoc(b) (((b) + PAGE_MASK) / PAGE_SIZE)
  70 #define round_page(p) ((((unsigned long)(p)) + PAGE_MASK) & ~(PAGE_MASK))
  71 #ifdef __CYGWIN__
  72 #define GIANT_REQUIRED
  73 #else
  74 #define GIANT_REQUIRED mtx_assert(&Giant, MA_OWNED)
  75 #endif
  76 #define KERN_SUCCESS 0
  77 #define VM_PROT_READ    PROT_READ
  78 #define VM_PROT_WRITE   PROT_WRITE
  79 #define VM_INHERIT_SHARE 0
  80 #define OBJT_PHYS  0
  81 #define OBJT_SWAP  0
  82 #define VM_PROT_DEFAULT 0
  83 #define VM_OBJECT_LOCK(a)
  84 #define vm_object_clear_flag(a,b)
  85 #define vm_object_set_flag(a,b)
  86 #define VM_OBJECT_UNLOCK(a)
  87 #define vm_map_remove(a,b,c) KERN_SUCCESS
  88 typedef int vm_prot_t;
  89 #endif /* __CYGWIN__ */
  90
  91 #ifndef __CYGWIN__
  92 static MALLOC_DEFINE(M_SHM, "shm", "SVID compatible shared memory segments");
  93
  94 struct oshmctl_args;
  95 static int oshmctl(struct thread *td, struct oshmctl_args *uap);
  96 #endif /* __CYGWIN__ */
  97
  98 static int shmget_allocate_segment(struct thread *td,
  99     struct shmget_args *uap, int mode);
 100 static int shmget_existing(struct thread *td, struct shmget_args *uap,
 101     int mode, int segnum);
 102
 103 #ifndef __CYGWIN__
 104 /* XXX casting to (sy_call_t *) is bogus, as usual. */
 105 static sy_call_t *shmcalls[] = {
 106         (sy_call_t *)shmat, (sy_call_t *)oshmctl,
 107         (sy_call_t *)shmdt, (sy_call_t *)shmget,
 108         (sy_call_t *)shmctl
 109 };
 110 #endif /* __CYGWIN__ */
 111
 112 #define SHMSEG_FREE             0x0200
 113 #define SHMSEG_REMOVED          0x0400
 114 #define SHMSEG_ALLOCATED        0x0800
 115 #define SHMSEG_WANTED           0x1000
 116
 117 static int shm_last_free, shm_nused, shm_committed, shmalloced, shm_nattch;
 118 static struct shmid_ds  *shmsegs;
 119
 120 struct shm_handle {
 121         /* vm_offset_t kva; */
 122         vm_object_t shm_object;
 123 };
 124
 125 struct shmmap_state {
 126         vm_offset_t va;
 127         int shmid;
 128 };
 129
 130 static void shm_deallocate_segment(struct shmid_ds *);
 131 static int shm_find_segment_by_key(key_t);
 132 static struct shmid_ds *shm_find_segment_by_shmid(int);
 133 static struct shmid_ds *shm_find_segment_by_shmidx(int);
 134 static int shm_delete_mapping(struct vmspace *vm, struct shmmap_state *);
 135 static void shmrealloc(void);
 136
 137 /*
 138  * Tuneable values.
 139  */
 140 #ifndef SHMMAXPGS
 141 #define SHMMAXPGS       8192    /* Note: sysv shared memory is swap backed. */
 142 #endif
 143 #ifndef SHMMAX
 144 #define SHMMAX  (SHMMAXPGS*PAGE_SIZE)
 145 #endif
 146 #ifndef SHMMIN
 147 #define SHMMIN  1
 148 #endif
 149 #ifndef SHMMNI
 150 #define SHMMNI  192
 151 #endif
 152 #ifndef SHMSEG
 153 #define SHMSEG  128
 154 #endif
 155 #ifndef SHMALL
 156 #define SHMALL  (SHMMAXPGS)
 157 #endif
 158
 159 struct  shminfo shminfo = {
 160         SHMMAX,
 161         SHMMIN,
 162         SHMMNI,
 163         SHMSEG,
 164         SHMALL
 165 };
 166
 167 #ifndef __CYGWIN__
 168 static int shm_use_phys;
 169 #else
 170 static long shm_use_phys;
 171 static long shm_allow_removed;
 172 #endif /* __CYGWIN__ */
 173
 174 #ifndef __CYGWIN__
 175 struct shm_info shm_info;
 176 #endif /* __CYGWIN__ */
 177
 178 #ifndef __CYGWIN__
 179 SYSCTL_DECL(_kern_ipc);
 180 SYSCTL_INT(_kern_ipc, OID_AUTO, shmmax, CTLFLAG_RW, &shminfo.shmmax, 0, "");
 181 SYSCTL_INT(_kern_ipc, OID_AUTO, shmmin, CTLFLAG_RW, &shminfo.shmmin, 0, "");
 182 SYSCTL_INT(_kern_ipc, OID_AUTO, shmmni, CTLFLAG_RDTUN, &shminfo.shmmni, 0, "");
 183 SYSCTL_INT(_kern_ipc, OID_AUTO, shmseg, CTLFLAG_RDTUN, &shminfo.shmseg, 0, "");
 184 SYSCTL_INT(_kern_ipc, OID_AUTO, shmall, CTLFLAG_RW, &shminfo.shmall, 0, "");
 185 SYSCTL_INT(_kern_ipc, OID_AUTO, shm_use_phys, CTLFLAG_RW,
 186     &shm_use_phys, 0, "");
 187 SYSCTL_INT(_kern_ipc, OID_AUTO, shm_allow_removed, CTLFLAG_RW,
 188     &shm_allow_removed, 0, "");
 189 SYSCTL_PROC(_kern_ipc, OID_AUTO, shmsegs, CTLFLAG_RD,
 190     NULL, 0, sysctl_shmsegs, "", "");
 191 #endif /* __CYGWIN__ */
 192
 193 static int
 194 shm_find_segment_by_key(key_t key)
 195 {
 196         int i;
 197
 198         for (i = 0; i < shmalloced; i++)
 199                 if ((shmsegs[i].shm_perm.mode & SHMSEG_ALLOCATED) &&
 200                     shmsegs[i].shm_perm.key == key)
 201                         return (i);
 202         return (-1);
 203 }
 204
 205 static struct shmid_ds *
 206 shm_find_segment_by_shmid(int shmid)
 207 {
 208         int segnum;
 209         struct shmid_ds *shmseg;
 210
 211         segnum = IPCID_TO_IX(shmid);
 212         if (segnum < 0 || segnum >= shmalloced)
 213                 return (NULL);
 214         shmseg = &shmsegs[segnum];
 215         if ((shmseg->shm_perm.mode & SHMSEG_ALLOCATED) == 0 ||
 216             (!shm_allow_removed &&
 217              (shmseg->shm_perm.mode & SHMSEG_REMOVED) != 0) ||
 218             shmseg->shm_perm.seq != IPCID_TO_SEQ(shmid))
 219                 return (NULL);
 220         return (shmseg);
 221 }
 222
 223 static struct shmid_ds *
 224 shm_find_segment_by_shmidx(int segnum)
 225 {
 226         struct shmid_ds *shmseg;
 227
 228         if (segnum < 0 || segnum >= shmalloced)
 229                 return (NULL);
 230         shmseg = &shmsegs[segnum];
 231         if ((shmseg->shm_perm.mode & SHMSEG_ALLOCATED) == 0 ||
 232             (!shm_allow_removed &&
 233              (shmseg->shm_perm.mode & SHMSEG_REMOVED) != 0))
 234                 return (NULL);
 235         return (shmseg);
 236 }
 237
 238 static void
 239 shm_deallocate_segment(struct shmid_ds *shmseg)
 240 {
 241         struct shm_handle *shm_handle;
 242         size_t size;
 243
 244         GIANT_REQUIRED;
 245
 246         shm_handle = shmseg->shm_internal;
 247         vm_object_deallocate(shm_handle->shm_object);
 248         sys_free(shm_handle, M_SHM);
 249         shmseg->shm_internal = NULL;
 250         size = round_page(shmseg->shm_segsz);
 251         shm_committed -= btoc(size);
 252         shm_nused--;
 253         shmseg->shm_perm.mode = SHMSEG_FREE;
 254 }
 255
 256 static int
 257 shm_delete_mapping(struct vmspace *vm, struct shmmap_state *shmmap_s)
 258 {
 259         struct shmid_ds *shmseg;
 260         int segnum, result;
 261         size_t size __attribute__ ((unused));
 262
 263         GIANT_REQUIRED;
 264
 265         segnum = IPCID_TO_IX(shmmap_s->shmid);
 266         shmseg = &shmsegs[segnum];
 267         size = round_page(shmseg->shm_segsz);
 268         result = vm_map_remove(&vm->vm_map, shmmap_s->va, shmmap_s->va + size);
 269         if (result != KERN_SUCCESS)
 270                 return (EINVAL);
 271         shmmap_s->shmid = -1;
 272         shmseg->shm_dtime = time (NULL);
 273         --shm_nattch;
 274         if ((--shmseg->shm_nattch <= 0) &&
 275             (shmseg->shm_perm.mode & SHMSEG_REMOVED)) {
 276                 shm_deallocate_segment(shmseg);
 277                 shm_last_free = segnum;
 278         }
 279         return (0);
 280 }
 281
 282 #ifndef _SYS_SYSPROTO_H_
 283 struct shmdt_args {
 284         const void *shmaddr;
 285 };
 286 #endif
 287
 288 /*
 289  * MPSAFE
 290  */
 291 int
 292 shmdt(struct thread *td, struct shmdt_args *uap)
 293 {
 294         struct proc *p = td->td_proc;
 295         struct shmmap_state *shmmap_s;
 296         int i;
 297         int error = 0;
 298
 299         if (!jail_sysvipc_allowed && jailed(td->td_ucred))
 300                 return (ENOSYS);
 301         mtx_lock(&Giant);
 302         shmmap_s = p->p_vmspace->vm_shm;
 303         if (shmmap_s == NULL) {
 304                 error = EINVAL;
 305                 goto done2;
 306         }
 307         for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) {
 308                 if (shmmap_s->shmid != -1 &&
 309                     shmmap_s->va == (vm_offset_t)uap->shmaddr) {
 310                         break;
 311                 }
 312         }
 313         if (i == shminfo.shmseg) {
 314                 error = EINVAL;
 315                 goto done2;
 316         }
 317         error = shm_delete_mapping(p->p_vmspace, shmmap_s);
 318 done2:
 319         mtx_unlock(&Giant);
 320         return (error);
 321 }
 322
 323 #ifndef _SYS_SYSPROTO_H_
 324 struct shmat_args {
 325         int shmid;
 326         const void *shmaddr;
 327         int shmflg;
 328 };
 329 #endif
 330
 331 /*
 332  * MPSAFE
 333  */
 334 int
 335 kern_shmat(struct thread *td, int shmid, const void *shmaddr, int shmflg)
 336 {
 337         struct proc *p = td->td_proc;
 338         int i, flags __attribute__ ((unused));
 339         struct shmid_ds *shmseg;
 340         struct shmmap_state *shmmap_s = NULL;
 341 #ifndef __CYGWIN__
 342         struct shm_handle *shm_handle;
 343 #endif
 344         vm_offset_t attach_va;
 345         vm_prot_t prot;
 346         vm_size_t size;
 347 #ifndef __CYGWIN__
 348         int rv;
 349 #endif
 350         int error = 0;
 351
 352         if (!jail_sysvipc_allowed && jailed(td->td_ucred))
 353                 return (ENOSYS);
 354         mtx_lock(&Giant);
 355         shmmap_s = p->p_vmspace->vm_shm;
 356         if (shmmap_s == NULL) {
 357                 size = shminfo.shmseg * sizeof(struct shmmap_state);
 358                 shmmap_s = (struct shmmap_state *) sys_malloc(size, M_SHM, M_WAITOK);
 359                 for (i = 0; i < shminfo.shmseg; i++)
 360                         shmmap_s[i].shmid = -1;
 361                 p->p_vmspace->vm_shm = shmmap_s;
 362         }
 363         shmseg = shm_find_segment_by_shmid(shmid);
 364         if (shmseg == NULL) {
 365                 error = EINVAL;
 366                 goto done2;
 367         }
 368         error = ipcperm(td, &shmseg->shm_perm,
 369             (shmflg & SHM_RDONLY) ? IPC_R : IPC_R|IPC_W);
 370         if (error)
 371                 goto done2;
 372         for (i = 0; i < shminfo.shmseg; i++) {
 373                 if (shmmap_s->shmid == -1)
 374                         break;
 375                 shmmap_s++;
 376         }
 377         if (i >= shminfo.shmseg) {
 378                 error = EMFILE;
 379                 goto done2;
 380         }
 381         size = round_page(shmseg->shm_segsz);
 382 #ifdef VM_PROT_READ_IS_EXEC
 383         prot = VM_PROT_READ | VM_PROT_EXECUTE;
 384 #else
 385         prot = VM_PROT_READ;
 386 #endif
 387         if ((shmflg & SHM_RDONLY) == 0)
 388                 prot |= VM_PROT_WRITE;
 389         flags = MAP_ANON | MAP_SHARED;
 390         debug_printf ("shmaddr: %x, shmflg: %x", shmaddr, shmflg);
 391 #ifdef __CYGWIN__
 392         /* The alignment checks have already been made in the Cygwin DLL
 393            and shmat's only job is to keep record of the attached mem.
 394            These checks break shm on 9x since MapViewOfFileEx apparently
 395            returns memory which isn't aligned to SHMLBA.  Go figure!  */
 396         attach_va = (vm_offset_t)shmaddr;
 397 #else
 398         if (shmaddr) {
 399                 flags |= MAP_FIXED;
 400                 if (shmflg & SHM_RND) {
 401                         attach_va = (vm_offset_t)shmaddr & ~(SHMLBA-1);
 402                 } else if (((vm_offset_t)shmaddr & (SHMLBA-1)) == 0) {
 403                         attach_va = (vm_offset_t)shmaddr;
 404                 } else {
 405                         error = EINVAL;
 406                         goto done2;
 407                 }
 408         } else {
 409                 /*
 410                  * This is just a hint to vm_map_find() about where to
 411                  * put it.
 412                  */
 413                 attach_va = round_page((vm_offset_t)p->p_vmspace->vm_taddr
 414                     + maxtsiz + maxdsiz);
 415         }
 416
 417         shm_handle = shmseg->shm_internal;
 418         vm_object_reference(shm_handle->shm_object);
 419         rv = vm_map_find(&p->p_vmspace->vm_map, shm_handle->shm_object,
 420                 0, &attach_va, size, (flags & MAP_FIXED)?0:1, prot, prot, 0);
 421         if (rv != KERN_SUCCESS) {
 422                 error = ENOMEM;
 423                 goto done2;
 424         }
 425         vm_map_inherit(&p->p_vmspace->vm_map,
 426                 attach_va, attach_va + size, VM_INHERIT_SHARE);
 427 #endif
 428
 429         shmmap_s->va = attach_va;
 430         shmmap_s->shmid = shmid;
 431         shmseg->shm_lpid = p->p_pid;
 432         shmseg->shm_atime = time (NULL);
 433         shmseg->shm_nattch++;
 434         shm_nattch++;
 435         td->td_retval[0] = attach_va;
 436 done2:
 437         mtx_unlock(&Giant);
 438         return (error);
 439 }
 440
 441 int
 442 shmat(struct thread *td, struct shmat_args *uap)
 443 {
 444         return kern_shmat(td, uap->shmid, uap->shmaddr, uap->shmflg);
 445 }
 446
 447 #ifndef __CYGWIN__
 448 struct oshmid_ds {
 449         struct  ipc_perm shm_perm;      /* operation perms */
 450         int     shm_segsz;              /* size of segment (bytes) */
 451         u_short shm_cpid;               /* pid, creator */
 452         u_short shm_lpid;               /* pid, last operation */
 453         short   shm_nattch;             /* no. of current attaches */
 454         time_t  shm_atime;              /* last attach time */
 455         time_t  shm_dtime;              /* last detach time */
 456         time_t  shm_ctime;              /* last change time */
 457         void    *shm_handle;            /* internal handle for shm segment */
 458 };
 459
 460 struct oshmctl_args {
 461         int shmid;
 462         int cmd;
 463         struct oshmid_ds *ubuf;
 464 };
 465
 466 /*
 467  * MPSAFE
 468  */
 469 static int
 470 oshmctl(struct thread *td, struct oshmctl_args *uap)
 471 {
 472 #ifdef COMPAT_43
 473         int error = 0;
 474         struct shmid_ds *shmseg;
 475         struct oshmid_ds outbuf;
 476
 477         if (!jail_sysvipc_allowed && jailed(td->td_ucred))
 478                 return (ENOSYS);
 479         mtx_lock(&Giant);
 480         shmseg = shm_find_segment_by_shmid(uap->shmid);
 481         if (shmseg == NULL) {
 482                 error = EINVAL;
 483                 goto done2;
 484         }
 485         switch (uap->cmd) {
 486         case IPC_STAT:
 487                 error = ipcperm(td, &shmseg->shm_perm, IPC_R);
 488                 if (error)
 489                         goto done2;
 490                 outbuf.shm_perm = shmseg->shm_perm;
 491                 outbuf.shm_segsz = shmseg->shm_segsz;
 492                 outbuf.shm_cpid = shmseg->shm_cpid;
 493                 outbuf.shm_lpid = shmseg->shm_lpid;
 494                 outbuf.shm_nattch = shmseg->shm_nattch;
 495                 outbuf.shm_atime = shmseg->shm_atime;
 496                 outbuf.shm_dtime = shmseg->shm_dtime;
 497                 outbuf.shm_ctime = shmseg->shm_ctime;
 498                 outbuf.shm_handle = shmseg->shm_internal;
 499                 error = copyout(&outbuf, uap->ubuf, sizeof(outbuf));
 500                 if (error)
 501                         goto done2;
 502                 break;
 503         default:
 504                 /* XXX casting to (sy_call_t *) is bogus, as usual. */
 505                 error = ((sy_call_t *)shmctl)(td, uap);
 506                 break;
 507         }
 508 done2:
 509         mtx_unlock(&Giant);
 510         return (error);
 511 #else
 512         return (EINVAL);
 513 #endif
 514 }
 515 #endif /* !__CYGWIN__ */
 516
 517 #ifndef _SYS_SYSPROTO_H_
 518 struct shmctl_args {
 519         int shmid;
 520         int cmd;
 521         struct shmid_ds *buf;
 522 };
 523 #endif
 524
 525 /*
 526  * MPSAFE
 527  */
 528 int
 529 kern_shmctl(struct thread *td, int shmid, int cmd, void *buf, size_t *bufsz)
 530 {
 531         int error = 0;
 532         struct shmid_ds *shmseg;
 533
 534         if (!jail_sysvipc_allowed && jailed(td->td_ucred))
 535                 return (ENOSYS);
 536
 537         mtx_lock(&Giant);
 538         switch (cmd) {
 539         case IPC_INFO:
 540                 memcpy(buf, &shminfo, sizeof(shminfo));
 541                 if (bufsz)
 542                         *bufsz = sizeof(shminfo);
 543                 td->td_retval[0] = shmalloced;
 544                 goto done2;
 545         case SHM_INFO: {
 546                 struct shm_info shm_info;
 547                 shm_info.used_ids = shm_nused;
 548                 shm_info.shm_tot = shm_committed * PAGE_SIZE;
 549 #ifdef __CYGWIN__
 550                 shm_info.shm_atts = shm_nattch;
 551 #else
 552                 shm_info.shm_rss = 0;   /*XXX where to get from ? */
 553                 shm_info.shm_swp = 0;   /*XXX where to get from ? */
 554                 shm_info.swap_attempts = 0;     /*XXX where to get from ? */
 555                 shm_info.swap_successes = 0;    /*XXX where to get from ? */
 556 #endif /* __CYGWIN__ */
 557                 memcpy(buf, &shm_info, sizeof(shm_info));
 558                 if (bufsz)
 559                         *bufsz = sizeof(shm_info);
 560                 td->td_retval[0] = shmalloced;
 561                 goto done2;
 562         }
 563         }
 564         if (cmd == SHM_STAT)
 565                 shmseg = shm_find_segment_by_shmidx(shmid);
 566         else
 567                 shmseg = shm_find_segment_by_shmid(shmid);
 568         if (shmseg == NULL) {
 569                 error = EINVAL;
 570                 goto done2;
 571         }
 572         switch (cmd) {
 573         case SHM_STAT:
 574         case IPC_STAT:
 575                 error = ipcperm(td, &shmseg->shm_perm, IPC_R);
 576                 if (error)
 577                         goto done2;
 578                 memcpy(buf, shmseg, sizeof(struct shmid_ds));
 579                 if (bufsz)
 580                         *bufsz = sizeof(struct shmid_ds);
 581                 if (cmd == SHM_STAT)
 582                         td->td_retval[0] = IXSEQ_TO_IPCID(shmid, shmseg->shm_perm);
 583                 break;
 584         case IPC_SET: {
 585                 struct shmid_ds *shmid;
 586
 587                 shmid = (struct shmid_ds *)buf;
 588                 error = ipcperm(td, &shmseg->shm_perm, IPC_M);
 589                 if (error)
 590                         goto done2;
 591                 shmseg->shm_perm.uid = shmid->shm_perm.uid;
 592                 shmseg->shm_perm.gid = shmid->shm_perm.gid;
 593                 shmseg->shm_perm.mode =
 594                     (shmseg->shm_perm.mode & ~ACCESSPERMS) |
 595                     (shmid->shm_perm.mode & ACCESSPERMS);
 596                 shmseg->shm_ctime = time (NULL);
 597                 break;
 598         }
 599         case IPC_RMID:
 600                 error = ipcperm(td, &shmseg->shm_perm, IPC_M);
 601                 if (error)
 602                         goto done2;
 603                 shmseg->shm_perm.key = IPC_PRIVATE;
 604                 shmseg->shm_perm.mode |= SHMSEG_REMOVED;
 605                 if (shmseg->shm_nattch <= 0) {
 606                         shm_deallocate_segment(shmseg);
 607                         shm_last_free = IPCID_TO_IX(shmid);
 608                 }
 609                 break;
 610 #if 0
 611         case SHM_LOCK:
 612         case SHM_UNLOCK:
 613 #endif
 614         default:
 615                 error = EINVAL;
 616                 break;
 617         }
 618 done2:
 619         mtx_unlock(&Giant);
 620         return (error);
 621 }
 622
 623 int
 624 shmctl(struct thread *td, struct shmctl_args *uap)
 625 {
 626         int error = 0;
 627         struct shmid_ds buf;
 628         size_t bufsz;
 629
 630         /* IPC_SET needs to copyin the buffer before calling kern_shmctl */
 631         if (uap->cmd == IPC_SET) {
 632                 if ((error = copyin(uap->buf, &buf, sizeof(struct shmid_ds))))
 633                         goto done;
 634         }
 635 #ifdef __CYGWIN__
 636         if (uap->cmd == IPC_INFO && uap->shmid > 0) {
 637                 /* Can't use the default kern_shmctl interface. */
 638                 int shmid = uap->shmid;
 639                 if (shmid > shminfo.shmmni)
 640                         shmid = shminfo.shmmni;
 641                 error = copyout(shmsegs, uap->buf,
 642                                 shmid * sizeof(struct shmid_ds));
 643                 td->td_retval[0] = error ? -1 : 0;
 644                 return (error);
 645         }
 646 #endif /* __CYGWIN__ */
 647
 648         error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&buf, &bufsz);
 649         if (error)
 650                 goto done;
 651
 652         /* Cases in which we need to copyout */
 653         switch (uap->cmd) {
 654         case IPC_INFO:
 655         case SHM_INFO:
 656         case SHM_STAT:
 657         case IPC_STAT:
 658                 error = copyout(&buf, uap->buf, bufsz);
 659                 break;
 660         }
 661
 662 done:
 663         if (error) {
 664                 /* Invalidate the return value */
 665                 td->td_retval[0] = -1;
 666         }
 667         return (error);
 668 }
 669
 670
 671 #ifndef _SYS_SYSPROTO_H_
 672 struct shmget_args {
 673         key_t key;
 674         size_t size;
 675         int shmflg;
 676 };
 677 #endif
 678
 679 static int
 680 shmget_existing(struct thread *td, struct shmget_args *uap, int mode, int segnum)
 681 {
 682         struct shmid_ds *shmseg;
 683         int error;
 684
 685         shmseg = &shmsegs[segnum];
 686         if (shmseg->shm_perm.mode & SHMSEG_REMOVED) {
 687                 /*
 688                  * This segment is in the process of being allocated.  Wait
 689                  * until it's done, and look the key up again (in case the
 690                  * allocation failed or it was freed).
 691                  */
 692                 shmseg->shm_perm.mode |= SHMSEG_WANTED;
 693                 error = tsleep(shmseg, PLOCK | PCATCH, "shmget", 0);
 694                 if (error)
 695                         return (error);
 696                 return (EAGAIN);
 697         }
 698         if ((uap->shmflg & (IPC_CREAT | IPC_EXCL)) == (IPC_CREAT | IPC_EXCL))
 699                 return (EEXIST);
 700         error = ipcperm(td, &shmseg->shm_perm, mode);
 701         if (error)
 702                 return (error);
 703         if (uap->size && uap->size > shmseg->shm_segsz)
 704                 return (EINVAL);
 705         td->td_retval[0] = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
 706 #ifdef __CYGWIN__
 707         td->td_retval[1] =
 708                 vm_object_duplicate(td, shmseg->shm_internal->shm_object);
 709 #endif /* __CYGWIN__ */
 710         return (0);
 711 }
 712
 713 static int
 714 shmget_allocate_segment(struct thread *td, struct shmget_args *uap, int mode)
 715 {
 716         int i, segnum, shmid, size;
 717 #ifndef __CYGWIN__
 718         struct ucred *cred = td->td_ucred;
 719 #endif /* __CYGWIN__ */
 720         struct shmid_ds *shmseg;
 721         struct shm_handle *shm_handle;
 722
 723         GIANT_REQUIRED;
 724
 725         if (uap->size < (unsigned long) shminfo.shmmin ||
 726             uap->size > (unsigned long) shminfo.shmmax)
 727                 return (EINVAL);
 728         if (shm_nused >= shminfo.shmmni) /* Any shmids left? */
 729                 return (ENOSPC);
 730         size = round_page(uap->size);
 731         if (shm_committed + btoc(size) > shminfo.shmall)
 732                 return (ENOMEM);
 733         if (shm_last_free < 0) {
 734                 shmrealloc();   /* Maybe expand the shmsegs[] array. */
 735                 for (i = 0; i < shmalloced; i++)
 736                         if (shmsegs[i].shm_perm.mode & SHMSEG_FREE)
 737                                 break;
 738                 if (i == shmalloced)
 739                         return (ENOSPC);
 740                 segnum = i;
 741         } else  {
 742                 segnum = shm_last_free;
 743                 shm_last_free = -1;
 744         }
 745         shmseg = &shmsegs[segnum];
 746         /*
 747          * In case we sleep in malloc(), mark the segment present but deleted
 748          * so that noone else tries to create the same key.
 749          */
 750         shmseg->shm_perm.mode = SHMSEG_ALLOCATED | SHMSEG_REMOVED;
 751         shmseg->shm_perm.key = uap->key;
 752         shmseg->shm_perm.seq = (shmseg->shm_perm.seq + 1) & 0x7fff;
 753         shm_handle = (struct shm_handle *)
 754             sys_malloc(sizeof(struct shm_handle), M_SHM, M_WAITOK);
 755         shmid = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
 756
 757         /*
 758          * We make sure that we have allocated a pager before we need
 759          * to.
 760          */
 761         if (shm_use_phys) {
 762                 shm_handle->shm_object =
 763                     vm_pager_allocate(OBJT_PHYS, 0, size, VM_PROT_DEFAULT, 0);
 764         } else {
 765                 shm_handle->shm_object =
 766                     vm_pager_allocate(OBJT_SWAP, 0, size, VM_PROT_DEFAULT, 0);
 767         }
 768         VM_OBJECT_LOCK(shm_handle->shm_object);
 769         vm_object_clear_flag(shm_handle->shm_object, OBJ_ONEMAPPING);
 770         vm_object_set_flag(shm_handle->shm_object, OBJ_NOSPLIT);
 771         VM_OBJECT_UNLOCK(shm_handle->shm_object);
 772
 773         shmseg->shm_internal = shm_handle;
 774 #ifdef __CYGWIN__
 775         shmseg->shm_perm.cuid = shmseg->shm_perm.uid = td->ipcblk->uid;
 776         shmseg->shm_perm.cgid = shmseg->shm_perm.gid = td->ipcblk->gid;
 777 #else
 778         shmseg->shm_perm.cuid = shmseg->shm_perm.uid = cred->cr_uid;
 779         shmseg->shm_perm.cgid = shmseg->shm_perm.gid = cred->cr_gid;
 780 #endif /* __CYGWIN__ */
 781         shmseg->shm_perm.mode = (shmseg->shm_perm.mode & SHMSEG_WANTED) |
 782             (mode & ACCESSPERMS) | SHMSEG_ALLOCATED;
 783         shmseg->shm_segsz = uap->size;
 784         shmseg->shm_cpid = td->td_proc->p_pid;
 785         shmseg->shm_lpid = shmseg->shm_nattch = 0;
 786         shmseg->shm_atime = shmseg->shm_dtime = 0;
 787         shmseg->shm_ctime = time (NULL);
 788         shm_committed += btoc(size);
 789         shm_nused++;
 790         if (shmseg->shm_perm.mode & SHMSEG_WANTED) {
 791                 /*
 792                  * Somebody else wanted this key while we were asleep.  Wake
 793                  * them up now.
 794                  */
 795                 shmseg->shm_perm.mode &= ~SHMSEG_WANTED;
 796                 wakeup(shmseg);
 797         }
 798         td->td_retval[0] = shmid;
 799 #ifdef __CYGWIN__
 800         td->td_retval[1] =
 801                 vm_object_duplicate(td, shmseg->shm_internal->shm_object);
 802 #endif /* __CYGWIN__ */
 803         return (0);
 804 }
 805
 806 /*
 807  * MPSAFE
 808  */
 809 int
 810 shmget(struct thread *td, struct shmget_args *uap)
 811 {
 812         int segnum, mode;
 813         int error;
 814
 815         if (!jail_sysvipc_allowed && jailed(td->td_ucred))
 816                 return (ENOSYS);
 817         mtx_lock(&Giant);
 818         mode = uap->shmflg & ACCESSPERMS;
 819         if (uap->key != IPC_PRIVATE) {
 820         again:
 821 #ifdef __CYGWIN__
 822                 if (uap->shmflg & IPC_KEY_IS_SHMID)
 823                   segnum = shm_find_segment_by_shmid ((int) uap->key) ?
 824                            IPCID_TO_IX((int) uap->key) : -1;
 825                 else
 826 #endif
 827                 segnum = shm_find_segment_by_key(uap->key);
 828                 if (segnum >= 0) {
 829                         error = shmget_existing(td, uap, mode, segnum);
 830                         if (error == EAGAIN)
 831                                 goto again;
 832                         goto done2;
 833                 }
 834                 if ((uap->shmflg & IPC_CREAT) == 0) {
 835                         error = ENOENT;
 836                         goto done2;
 837                 }
 838         }
 839         error = shmget_allocate_segment(td, uap, mode);
 840 done2:
 841 #ifdef __CYGWIN__
 842         if (!error)
 843                 ipcexit_creat_hookthread (td);
 844         else
 845                 td->td_retval[0] = -1;
 846 #endif
 847         mtx_unlock(&Giant);
 848         return (error);
 849 }
 850
 851 #ifndef __CYGWIN__
 852 /*
 853  * MPSAFE
 854  */
 855 int
 856 shmsys(td, uap)
 857         struct thread *td;
 858         /* XXX actually varargs. */
 859         struct shmsys_args /* {
 860                 int     which;
 861                 int     a2;
 862                 int     a3;
 863                 int     a4;
 864         } */ *uap;
 865 {
 866         int error;
 867
 868         if (!jail_sysvipc_allowed && jailed(td->td_ucred))
 869                 return (ENOSYS);
 870         if (uap->which < 0 ||
 871             uap->which >= sizeof(shmcalls)/sizeof(shmcalls[0]))
 872                 return (EINVAL);
 873         mtx_lock(&Giant);
 874         error = (*shmcalls[uap->which])(td, &uap->a2);
 875         mtx_unlock(&Giant);
 876         return (error);
 877 }
 878 #endif /* __CYGWIN__ */
 879
 880 static void
 881 shmfork_myhook(struct proc *p1, struct proc *p2)
 882 {
 883         struct shmmap_state *shmmap_s;
 884         size_t size;
 885         int i;
 886
 887         size = shminfo.shmseg * sizeof(struct shmmap_state);
 888         shmmap_s = (struct shmmap_state *) sys_malloc(size, M_SHM, M_WAITOK);
 889         bcopy(p1->p_vmspace->vm_shm, shmmap_s, size);
 890         p2->p_vmspace->vm_shm = shmmap_s;
 891         for (i = 0; i < shminfo.shmseg; i++, shmmap_s++)
 892                 if (shmmap_s->shmid != -1) {
 893                         shm_nattch++;
 894                         shmsegs[IPCID_TO_IX(shmmap_s->shmid)].shm_nattch++;
 895                 }
 896 }
 897
 898 #ifdef __CYGWIN__
 899 int cygwin_shmfork_myhook (struct thread *td, struct proc *parent)
 900 {
 901   ipcexit_creat_hookthread (td);
 902   ipc_p_vmspace (td->ipcblk);
 903   ipc_p_vmspace (parent);
 904   shmfork_myhook (parent, td->ipcblk);
 905   return 0;
 906 }
 907 #endif
 908
 909 void
 910 shmexit_myhook(struct vmspace *vm)
 911 {
 912         struct shmmap_state *base, *shm;
 913         int i;
 914
 915         GIANT_REQUIRED;
 916
 917         if ((base = vm->vm_shm) != NULL) {
 918                 vm->vm_shm = NULL;
 919                 for (i = 0, shm = base; i < shminfo.shmseg; i++, shm++) {
 920                         if (shm->shmid != -1)
 921                                 shm_delete_mapping(vm, shm);
 922                 }
 923                 sys_free(base, M_SHM);
 924         }
 925 }
 926
 927 static void
 928 shmrealloc(void)
 929 {
 930         int i;
 931         struct shmid_ds *newsegs;
 932
 933         if (shmalloced >= shminfo.shmmni)
 934                 return;
 935
 936         newsegs = (struct shmid_ds *) sys_malloc(shminfo.shmmni * sizeof(*newsegs), M_SHM, M_WAITOK);
 937         if (newsegs == NULL)
 938                 return;
 939         for (i = 0; i < shmalloced; i++)
 940                 bcopy(&shmsegs[i], &newsegs[i], sizeof(newsegs[0]));
 941         for (; i < shminfo.shmmni; i++) {
 942                 shmsegs[i].shm_perm.mode = SHMSEG_FREE;
 943                 shmsegs[i].shm_perm.seq = 0;
 944         }
 945         sys_free(shmsegs, M_SHM);
 946         shmsegs = newsegs;
 947         shmalloced = shminfo.shmmni;
 948 }
 949
 950 void
 951 shminit(void)
 952 {
 953         int i;
 954         tun_bool_t shm_ar;
 955
 956         TUNABLE_INT_FETCH("kern.ipc.shmmaxpgs", &shminfo.shmall);
 957         for (i = PAGE_SIZE; i > 0; i--) {
 958                 shminfo.shmmax = shminfo.shmall * i;
 959                 if (shminfo.shmmax >= shminfo.shmall)
 960                         break;
 961         }
 962         TUNABLE_INT_FETCH("kern.ipc.shmmin", &shminfo.shmmin);
 963         TUNABLE_INT_FETCH("kern.ipc.shmmni", &shminfo.shmmni);
 964         TUNABLE_INT_FETCH("kern.ipc.shmseg", &shminfo.shmseg);
 965         TUNABLE_BOOL_FETCH("kern.ipc.shm_allow_removed", &shm_ar);
 966         if (shm_ar == TUN_TRUE)
 967           shm_allow_removed = 1;
 968         shmalloced = shminfo.shmmni;
 969         shmsegs = (struct shmid_ds *) sys_malloc(shmalloced * sizeof(shmsegs[0]), M_SHM, M_WAITOK);
 970         if (shmsegs == NULL)
 971                 panic("cannot allocate initial memory for sysvshm");
 972         for (i = 0; i < shmalloced; i++) {
 973                 shmsegs[i].shm_perm.mode = SHMSEG_FREE;
 974                 shmsegs[i].shm_perm.seq = 0;
 975         }
 976         shm_last_free = 0;
 977         shm_nused = 0;
 978         shm_committed = 0;
 979 #ifndef __CYGWIN__
 980         shmexit_hook = &shmexit_myhook;
 981         shmfork_hook = &shmfork_myhook;
 982 #endif /* __CYGWIN__ */
 983 }
 984
 985 int
 986 shmunload(void)
 987 {
 988
 989         if (shm_nused > 0)
 990                 return (EBUSY);
 991
 992         sys_free(shmsegs, M_SHM);
 993 #ifndef __CYGWIN__
 994         shmexit_hook = NULL;
 995         shmfork_hook = NULL;
 996 #endif /* __CYGWIN__ */
 997         return (0);
 998 }
 999
1000 #ifndef __CYGWIN__
1001 static int
1002 sysctl_shmsegs(SYSCTL_HANDLER_ARGS)
1003 {
1004
1005         return (SYSCTL_OUT(req, shmsegs, shmalloced * sizeof(shmsegs[0])));
1006 }
1007
1008 static int
1009 sysvshm_modload(struct module *module, int cmd, void *arg)
1010 {
1011         int error = 0;
1012
1013         switch (cmd) {
1014         case MOD_LOAD:
1015                 shminit();
1016                 break;
1017         case MOD_UNLOAD:
1018                 error = shmunload();
1019                 break;
1020         case MOD_SHUTDOWN:
1021                 break;
1022         default:
1023                 error = EINVAL;
1024                 break;
1025         }
1026         return (error);
1027 }
1028
1029 static moduledata_t sysvshm_mod = {
1030         "sysvshm",
1031         &sysvshm_modload,
1032         NULL
1033 };
1034
1035 SYSCALL_MODULE_HELPER(shmsys);
1036 SYSCALL_MODULE_HELPER(shmat);
1037 SYSCALL_MODULE_HELPER(shmctl);
1038 SYSCALL_MODULE_HELPER(shmdt);
1039 SYSCALL_MODULE_HELPER(shmget);
1040
1041 DECLARE_MODULE(sysvshm, sysvshm_mod,
1042         SI_SUB_SYSV_SHM, SI_ORDER_FIRST);
1043 MODULE_VERSION(sysvshm, 1);
1044 #endif /* __CYGWIN__ */
1045 #endif /* __OUTSIDE_CYGWIN__ */