sys/kern/lwkt_caps.c

   1 /*
   2  * Copyright (c) 2003,2004 The DragonFly Project.  All rights reserved.
   3  *
   4  * This code is derived from software contributed to The DragonFly Project
   5  * by Matthew Dillon <dillon@backplane.com>
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  *
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in
  15  *    the documentation and/or other materials provided with the
  16  *    distribution.
  17  * 3. Neither the name of The DragonFly Project nor the names of its
  18  *    contributors may be used to endorse or promote products derived
  19  *    from this software without specific, prior written permission.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  24  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
  25  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  26  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
  27  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  28  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
  29  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
  31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  32  * SUCH DAMAGE.
  33  *
  34  * $DragonFly: src/sys/kern/lwkt_caps.c,v 1.13 2007/02/26 21:41:08 corecode Exp $
  35  */
  36
  37 /*
  38  * This module implements the DragonFly LWKT IPC rendezvous and message
  39  * passing API which operates between userland processes, between userland
  40  * threads, and between userland processes and kernel threads.  This API
  41  * is known as the CAPS interface.
  42  *
  43  * Generally speaking this module abstracts the LWKT message port interface
  44  * into userland Clients and Servers rendezvous through ports named
  45  * by or wildcarded by (name,uid,gid).  The kernel provides system calls
  46  * which may be assigned to the mp_* fields in a userland-supplied
  47  * kernel-managed port, and a registration interface which associates an
  48  * upcall with a userland port.  The kernel tracks authentication information
  49  * and deals with connection failures by automatically replying to unreplied
  50  * messages.
  51  *
  52  * From the userland perspective a client/server connection involves two
  53  * message ports on the client and two message ports on the server.
  54  */
  55
  56 #include <sys/param.h>
  57 #include <sys/systm.h>
  58 #include <sys/kernel.h>
  59 #include <sys/sysproto.h>
  60 #include <sys/malloc.h>
  61 #include <sys/proc.h>
  62 #include <sys/ucred.h>
  63 #include <sys/caps.h>
  64 #include <sys/sysctl.h>
  65 #include <vm/vm.h>
  66 #include <vm/vm_extern.h>
  67
  68 static int caps_process_msg(caps_kinfo_t caps, caps_kmsg_t msg, struct caps_sys_get_args *uap);
  69 static void caps_free(caps_kinfo_t caps);
  70 static void caps_free_msg(caps_kmsg_t msg);
  71 static int caps_name_check(const char *name, int len);
  72 static caps_kinfo_t caps_free_msg_mcaps(caps_kmsg_t msg);
  73 static caps_kinfo_t kern_caps_sys_service(const char *name, uid_t uid,
  74                         gid_t gid, struct ucred *cred,
  75                         int flags, int *error);
  76 static caps_kinfo_t kern_caps_sys_client(const char *name, uid_t uid,
  77                         gid_t gid, struct ucred *cred, int flags, int *error);
  78
  79 #define CAPS_HSIZE      64
  80 #define CAPS_HMASK      (CAPS_HSIZE - 1)
  81
  82 static caps_kinfo_t     caps_hash_ary[CAPS_HSIZE];
  83 static int caps_waitsvc;
  84
  85 MALLOC_DEFINE(M_CAPS, "caps", "caps IPC messaging");
  86
  87 static int caps_enabled;
  88 SYSCTL_INT(_kern, OID_AUTO, caps_enabled,
  89         CTLFLAG_RW, &caps_enabled, 0, "Enable CAPS");
  90
  91 /************************************************************************
  92  *                      INLINE SUPPORT FUNCTIONS                        *
  93  ************************************************************************/
  94
  95 static __inline
  96 struct caps_kinfo **
  97 caps_hash(const char *name, int len)
  98 {
  99     int hv = 0x7123F4B3;
 100
 101     while (--len >= 0)
 102         hv = (hv << 5) ^ name[len] ^ (hv >> 23);
 103     return(&caps_hash_ary[(hv ^ (hv >> 16)) & CAPS_HMASK]);
 104 }
 105
 106 static __inline
 107 void
 108 caps_hold(caps_kinfo_t caps)
 109 {
 110     ++caps->ci_refs;
 111 }
 112
 113 static __inline
 114 void
 115 caps_drop(caps_kinfo_t caps)
 116 {
 117     if (--caps->ci_refs == 0)
 118         caps_free(caps);
 119 }
 120
 121 /************************************************************************
 122  *                      STATIC SUPPORT FUNCTIONS                        *
 123  ************************************************************************/
 124
 125 static
 126 caps_kinfo_t
 127 caps_find(const char *name, int len, uid_t uid, gid_t gid)
 128 {
 129     caps_kinfo_t caps;
 130     struct caps_kinfo **chash;
 131
 132     chash = caps_hash(name, len);
 133     for (caps = *chash; caps; caps = caps->ci_hnext) {
 134         if ((uid == (uid_t)-1 || uid == caps->ci_uid) &&
 135             (gid == (gid_t)-1 || gid == caps->ci_gid) &&
 136             len == caps->ci_namelen &&
 137             bcmp(name, caps->ci_name, len) == 0
 138         ) {
 139             caps_hold(caps);
 140             break;
 141         }
 142     }
 143     return(caps);
 144 }
 145
 146 static
 147 caps_kinfo_t
 148 caps_find_id(thread_t td, int id)
 149 {
 150    caps_kinfo_t caps;
 151
 152    for (caps = td->td_caps; caps; caps = caps->ci_tdnext) {
 153         if (caps->ci_id == id) {
 154             caps_hold(caps);
 155             break;
 156         }
 157    }
 158    return(caps);
 159 }
 160
 161 static
 162 caps_kinfo_t
 163 caps_alloc(thread_t td, const char *name, int len, uid_t uid, gid_t gid,
 164             int flags, caps_type_t type)
 165 {
 166     struct caps_kinfo **chash;
 167     caps_kinfo_t caps;
 168     caps_kinfo_t ctmp;
 169
 170     caps = kmalloc(offsetof(struct caps_kinfo, ci_name[len+1]),
 171                         M_CAPS, M_WAITOK|M_ZERO);
 172     TAILQ_INIT(&caps->ci_msgpendq);
 173     TAILQ_INIT(&caps->ci_msguserq);
 174     caps->ci_uid = uid;         /* -1 == not registered for uid search */
 175     caps->ci_gid = gid;         /* -1 == not registered for gid search */
 176     caps->ci_type = type;
 177     caps->ci_refs = 1;  /* CAPKF_TDLIST reference */
 178     caps->ci_namelen = len;
 179     caps->ci_flags = flags;
 180     bcopy(name, caps->ci_name, len + 1);
 181     if (type == CAPT_SERVICE) {
 182         chash = caps_hash(caps->ci_name, len);
 183         caps->ci_hnext = *chash;
 184         *chash = caps;
 185         caps->ci_flags |= CAPKF_HLIST;
 186     }
 187     if (td->td_caps) {
 188         caps->ci_id = td->td_caps->ci_id + 1;
 189         if (caps->ci_id < 0) {
 190             /*
 191              * It is virtually impossible for this case to occur.
 192              */
 193             caps->ci_id = 1;
 194             while ((ctmp = caps_find_id(td, caps->ci_id)) != NULL) {
 195                 caps_drop(ctmp);
 196                 ++caps->ci_id;
 197             }
 198         }
 199     } else {
 200         caps->ci_id = 1;
 201     }
 202     caps->ci_flags |= CAPKF_TDLIST;
 203     caps->ci_tdnext = td->td_caps;
 204     caps->ci_td = td;
 205     td->td_caps = caps;
 206     return(caps);
 207 }
 208
 209 static
 210 caps_kmsg_t
 211 caps_alloc_msg(caps_kinfo_t caps)
 212 {
 213     caps_kmsg_t msg;
 214
 215     msg = kmalloc(sizeof(struct caps_kmsg), M_CAPS, M_WAITOK|M_ZERO);
 216     msg->km_msgid.c_id = (off_t)(uintptr_t)msg;
 217     return(msg);
 218 }
 219
 220 static
 221 caps_kmsg_t
 222 caps_find_msg(caps_kinfo_t caps, off_t msgid)
 223 {
 224     caps_kmsg_t msg;
 225
 226     TAILQ_FOREACH(msg, &caps->ci_msguserq, km_node) {
 227         if (msg->km_msgid.c_id == msgid)
 228             return(msg);
 229     }
 230     TAILQ_FOREACH(msg, &caps->ci_msgpendq, km_node) {
 231         if (msg->km_msgid.c_id == msgid)
 232             return(msg);
 233     }
 234     return(NULL);
 235 }
 236
 237 static
 238 caps_kinfo_t
 239 caps_load_ccr(caps_kinfo_t caps, caps_kmsg_t msg, struct proc *p,
 240                 void *udata, int ubytes)
 241 {
 242     int i;
 243     struct ucred *cr = p->p_ucred;
 244     caps_kinfo_t rcaps;
 245
 246     /*
 247      * replace km_mcaps with new VM state, return the old km_mcaps.  The
 248      * caller is expected to drop the rcaps ref count on return so we do
 249      * not do it ourselves.
 250      */
 251     rcaps = caps_free_msg_mcaps(msg);   /* can be NULL */
 252     caps_hold(caps);
 253     msg->km_mcaps = caps;
 254     xio_init_ubuf(&msg->km_xio, udata, ubytes, XIOF_READ);
 255
 256     msg->km_ccr.pid = p ? p->p_pid : -1;
 257     msg->km_ccr.uid = cr->cr_ruid;
 258     msg->km_ccr.euid = cr->cr_uid;
 259     msg->km_ccr.gid = cr->cr_rgid;
 260     msg->km_ccr.ngroups = MIN(cr->cr_ngroups, CAPS_MAXGROUPS);
 261     for (i = 0; i < msg->km_ccr.ngroups; ++i)
 262         msg->km_ccr.groups[i] = cr->cr_groups[i];
 263     return(rcaps);
 264 }
 265
 266 static void
 267 caps_dequeue_msg(caps_kinfo_t caps, caps_kmsg_t msg)
 268 {
 269     if (msg->km_flags & CAPKMF_ONUSERQ)
 270         TAILQ_REMOVE(&caps->ci_msguserq, msg, km_node);
 271     if (msg->km_flags & CAPKMF_ONPENDQ)
 272         TAILQ_REMOVE(&caps->ci_msgpendq, msg, km_node);
 273     msg->km_flags &= ~(CAPKMF_ONPENDQ|CAPKMF_ONUSERQ);
 274 }
 275
 276 static void
 277 caps_put_msg(caps_kinfo_t caps, caps_kmsg_t msg, caps_msg_state_t state)
 278 {
 279     KKASSERT((msg->km_flags & (CAPKMF_ONUSERQ|CAPKMF_ONPENDQ)) == 0);
 280
 281     msg->km_flags |= CAPKMF_ONPENDQ;
 282     msg->km_flags &= ~CAPKMF_PEEKED;
 283     msg->km_state = state;
 284     TAILQ_INSERT_TAIL(&caps->ci_msgpendq, msg, km_node);
 285
 286     /*
 287      * Instead of waking up the service for both new messages and disposals,
 288      * just wakeup the service for new messages and it will process the
 289      * previous disposal in the same loop, reducing the number of context
 290      * switches required to run an IPC.
 291      */
 292     if (state != CAPMS_DISPOSE)
 293         wakeup(caps);
 294     caps_drop(caps);
 295 }
 296
 297 /*
 298  * caps_free_msg_mcaps()
 299  */
 300 static
 301 caps_kinfo_t
 302 caps_free_msg_mcaps(caps_kmsg_t msg)
 303 {
 304     caps_kinfo_t mcaps;
 305
 306     mcaps = msg->km_mcaps;      /* may be NULL */
 307     msg->km_mcaps = NULL;
 308     if (msg->km_xio.xio_npages)
 309         xio_release(&msg->km_xio);
 310     return(mcaps);
 311 }
 312
 313 /*
 314  * caps_free_msg()
 315  *
 316  * Free a caps placeholder message.  The message must not be on any queues.
 317  */
 318 static void
 319 caps_free_msg(caps_kmsg_t msg)
 320 {
 321     caps_kinfo_t rcaps;
 322
 323     if ((rcaps = caps_free_msg_mcaps(msg)) != NULL)
 324         caps_drop(rcaps);
 325     kfree(msg, M_CAPS);
 326 }
 327
 328 /*
 329  * Validate the service name
 330  */
 331 static int
 332 caps_name_check(const char *name, int len)
 333 {
 334     int i;
 335     char c;
 336
 337     for (i = len - 1; i >= 0; --i) {
 338         c = name[i];
 339         if (c >= '0' && c <= '9')
 340             continue;
 341         if (c >= 'a' && c <= 'z')
 342             continue;
 343         if (c >= 'A' && c <= 'Z')
 344             continue;
 345         if (c == '_' || c == '.')
 346             continue;
 347         return(EINVAL);
 348     }
 349     return(0);
 350 }
 351
 352 /*
 353  * caps_term()
 354  *
 355  * Terminate portions of a caps info structure.  This is used to close
 356  * an end-point or to flush particular messages on an end-point.
 357  *
 358  * This function should not be called with CAPKF_TDLIST unless the caller
 359  * has an additional hold on the caps structure.
 360  */
 361 static void
 362 caps_term(caps_kinfo_t caps, int flags, caps_kinfo_t cflush)
 363 {
 364     struct caps_kinfo **scan;
 365     caps_kmsg_t msg;
 366
 367     if (flags & CAPKF_TDLIST)
 368         caps->ci_flags |= CAPKF_CLOSED;
 369
 370     if (flags & CAPKF_FLUSH) {
 371         int mflags;
 372         struct caps_kmsg_queue tmpuserq;
 373         struct caps_kmsg_queue tmppendq;
 374         caps_kinfo_t rcaps;
 375
 376         TAILQ_INIT(&tmpuserq);
 377         TAILQ_INIT(&tmppendq);
 378
 379         while ((msg = TAILQ_FIRST(&caps->ci_msgpendq)) != NULL ||
 380                (msg = TAILQ_FIRST(&caps->ci_msguserq)) != NULL
 381         ) {
 382             mflags = msg->km_flags & (CAPKMF_ONUSERQ|CAPKMF_ONPENDQ);
 383             caps_dequeue_msg(caps, msg);
 384
 385             if (cflush && msg->km_mcaps != cflush) {
 386                 if (mflags & CAPKMF_ONUSERQ)
 387                     TAILQ_INSERT_TAIL(&tmpuserq, msg, km_node);
 388                 else
 389                     TAILQ_INSERT_TAIL(&tmppendq, msg, km_node);
 390             } else {
 391                 /*
 392                  * Dispose of the message.  If the received message is a
 393                  * request we must reply it.  If the received message is
 394                  * a reply we must return it for disposal.  If the
 395                  * received message is a disposal request we simply free it.
 396                  */
 397                 switch(msg->km_state) {
 398                 case CAPMS_REQUEST:
 399                 case CAPMS_REQUEST_RETRY:
 400                     rcaps = caps_load_ccr(caps, msg, curproc, NULL, 0);
 401                     if (rcaps->ci_flags & CAPKF_CLOSED) {
 402                         /*
 403                          * can't reply, if we never read the message (its on
 404                          * the pending queue), or if we are closed ourselves,
 405                          * we can just free the message.  Otherwise we have
 406                          * to send ourselves a disposal request (multi-threaded
 407                          * services have to deal with disposal requests for
 408                          * messages that might be in progress).
 409                          */
 410                         if ((caps->ci_flags & CAPKF_CLOSED) ||
 411                             (mflags & CAPKMF_ONPENDQ)
 412                         ) {
 413                             caps_free_msg(msg);
 414                             caps_drop(rcaps);
 415                         } else {
 416                             caps_drop(rcaps);
 417                             caps_hold(caps);    /* for message */
 418                             caps_put_msg(caps, msg, CAPMS_DISPOSE);
 419                         }
 420                     } else {
 421                         /*
 422                          * auto-reply to the originator.  rcaps already
 423                          * has a dangling hold so we do not have to hold it
 424                          * again.
 425                          */
 426                         caps_put_msg(rcaps, msg, CAPMS_REPLY);
 427                     }
 428                     break;
 429                 case CAPMS_REPLY:
 430                 case CAPMS_REPLY_RETRY:
 431                     rcaps = caps_load_ccr(caps, msg, curproc, NULL, 0);
 432                     if (caps == rcaps || (rcaps->ci_flags & CAPKF_CLOSED)) {
 433                         caps_free_msg(msg);     /* degenerate disposal case */
 434                         caps_drop(rcaps);
 435                     } else {
 436                         caps_put_msg(rcaps, msg, CAPMS_DISPOSE);
 437                     }
 438                     break;
 439                 case CAPMS_DISPOSE:
 440                     caps_free_msg(msg);
 441                     break;
 442                 }
 443             }
 444         }
 445         while ((msg = TAILQ_FIRST(&tmpuserq)) != NULL) {
 446             TAILQ_REMOVE(&tmpuserq, msg, km_node);
 447             TAILQ_INSERT_TAIL(&caps->ci_msguserq, msg, km_node);
 448             msg->km_flags |= CAPKMF_ONUSERQ;
 449         }
 450         while ((msg = TAILQ_FIRST(&tmppendq)) != NULL) {
 451             TAILQ_REMOVE(&tmppendq, msg, km_node);
 452             TAILQ_INSERT_TAIL(&caps->ci_msgpendq, msg, km_node);
 453             msg->km_flags |= CAPKMF_ONPENDQ;
 454         }
 455     }
 456     if ((flags & CAPKF_HLIST) && (caps->ci_flags & CAPKF_HLIST)) {
 457         for (scan = caps_hash(caps->ci_name, caps->ci_namelen);
 458             *scan != caps;
 459             scan = &(*scan)->ci_hnext
 460         ) {
 461             KKASSERT(*scan != NULL);
 462         }
 463         *scan = caps->ci_hnext;
 464         caps->ci_hnext = (void *)-1;
 465         caps->ci_flags &= ~CAPKF_HLIST;
 466     }
 467     if ((flags & CAPKF_TDLIST) && (caps->ci_flags & CAPKF_TDLIST)) {
 468         for (scan = &caps->ci_td->td_caps;
 469             *scan != caps;
 470             scan = &(*scan)->ci_tdnext
 471         ) {
 472             KKASSERT(*scan != NULL);
 473         }
 474         *scan = caps->ci_tdnext;
 475         caps->ci_flags &= ~CAPKF_TDLIST;
 476         caps->ci_tdnext = (void *)-1;
 477         caps->ci_td = NULL;
 478         caps_drop(caps);
 479     }
 480     if ((flags & CAPKF_RCAPS) && (caps->ci_flags & CAPKF_RCAPS)) {
 481         caps_kinfo_t ctmp;
 482
 483         caps->ci_flags &= ~CAPKF_RCAPS;
 484         if ((ctmp = caps->ci_rcaps)) {
 485             caps->ci_rcaps = NULL;
 486             caps_term(ctmp, CAPKF_FLUSH, caps);
 487             caps_drop(ctmp);
 488         }
 489     }
 490 }
 491
 492 static void
 493 caps_free(caps_kinfo_t caps)
 494 {
 495     KKASSERT(TAILQ_EMPTY(&caps->ci_msgpendq));
 496     KKASSERT(TAILQ_EMPTY(&caps->ci_msguserq));
 497     KKASSERT((caps->ci_flags & (CAPKF_HLIST|CAPKF_TDLIST)) == 0);
 498     kfree(caps, M_CAPS);
 499 }
 500
 501 /************************************************************************
 502  *                      PROCESS SUPPORT FUNCTIONS                       *
 503  ************************************************************************/
 504
 505 /*
 506  * Create dummy entries in p2 so we can return the appropriate
 507  * error code.  Robust userland code will check the error for a
 508  * forked condition and reforge the connection.
 509  */
 510 void
 511 caps_fork(struct thread *td1, struct thread *td2)
 512 {
 513     caps_kinfo_t caps1;
 514     caps_kinfo_t caps2;
 515
 516     /*
 517      * Create dummy entries with the same id's as the originals.  Note
 518      * that service entries are not re-added to the hash table. The
 519      * dummy entries return an ENOTCONN error allowing userland code to
 520      * detect that a fork occured.  Userland must reconnect to the service.
 521      */
 522     for (caps1 = td1->td_caps; caps1; caps1 = caps1->ci_tdnext) {
 523         if (caps1->ci_flags & CAPF_NOFORK)
 524                 continue;
 525         caps2 = caps_alloc(td2,
 526                         caps1->ci_name, caps1->ci_namelen,
 527                         caps1->ci_uid, caps1->ci_gid,
 528                         caps1->ci_flags & CAPF_UFLAGS, CAPT_FORKED);
 529         caps2->ci_id = caps1->ci_id;
 530     }
 531
 532     /*
 533      * Reverse the list order to maintain highest-id-first
 534      */
 535     caps2 = td2->td_caps;
 536     td2->td_caps = NULL;
 537     while (caps2) {
 538         caps1 = caps2->ci_tdnext;
 539         caps2->ci_tdnext = td2->td_caps;
 540         td2->td_caps = caps2;
 541         caps2 = caps1;
 542     }
 543 }
 544
 545 void
 546 caps_exit(struct thread *td)
 547 {
 548     caps_kinfo_t caps;
 549
 550     while ((caps = td->td_caps) != NULL) {
 551         caps_hold(caps);
 552         caps_term(caps, CAPKF_TDLIST|CAPKF_HLIST|CAPKF_FLUSH|CAPKF_RCAPS, NULL);
 553         caps_drop(caps);
 554     }
 555 }
 556
 557 /************************************************************************
 558  *                              SYSTEM CALLS                            *
 559  ************************************************************************/
 560
 561 /*
 562  * caps_sys_service(name, uid, gid, upcid, flags);
 563  *
 564  * Create an IPC service using the specified name, uid, gid, and flags.
 565  * Either uid or gid can be -1, but not both.  The port identifier is
 566  * returned.
 567  *
 568  * upcid can either be an upcall or a kqueue identifier (XXX)
 569  */
 570 int
 571 sys_caps_sys_service(struct caps_sys_service_args *uap)
 572 {
 573     struct ucred *cred = curproc->p_ucred;
 574     char name[CAPS_MAXNAMELEN];
 575     caps_kinfo_t caps;
 576     int len;
 577     int error;
 578
 579     if (caps_enabled == 0)
 580         return(EOPNOTSUPP);
 581     if ((error = copyinstr(uap->name, name, CAPS_MAXNAMELEN, &len)) != 0)
 582         return(error);
 583     if (--len <= 0)
 584         return(EINVAL);
 585     if ((error = caps_name_check(name, len)) != 0)
 586         return(error);
 587
 588     caps = kern_caps_sys_service(name, uap->uid, uap->gid, cred,
 589                                 uap->flags & CAPF_UFLAGS, &error);
 590     if (caps)
 591         uap->sysmsg_result = caps->ci_id;
 592     return(error);
 593 }
 594
 595 /*
 596  * caps_sys_client(name, uid, gid, upcid, flags);
 597  *
 598  * Create an IPC client connected to the specified service.  Either uid or gid
 599  * may be -1, indicating a wildcard, but not both.  The port identifier is
 600  * returned.
 601  *
 602  * upcid can either be an upcall or a kqueue identifier (XXX)
 603  */
 604 int
 605 sys_caps_sys_client(struct caps_sys_client_args *uap)
 606 {
 607     struct ucred *cred = curproc->p_ucred;
 608     char name[CAPS_MAXNAMELEN];
 609     caps_kinfo_t caps;
 610     int len;
 611     int error;
 612
 613     if (caps_enabled == 0)
 614         return(EOPNOTSUPP);
 615     if ((error = copyinstr(uap->name, name, CAPS_MAXNAMELEN, &len)) != 0)
 616         return(error);
 617     if (--len <= 0)
 618         return(EINVAL);
 619     if ((error = caps_name_check(name, len)) != 0)
 620         return(error);
 621
 622     caps = kern_caps_sys_client(name, uap->uid, uap->gid, cred,
 623                                 uap->flags & CAPF_UFLAGS, &error);
 624     if (caps)
 625         uap->sysmsg_result = caps->ci_id;
 626     return(error);
 627 }
 628
 629 int
 630 sys_caps_sys_close(struct caps_sys_close_args *uap)
 631 {
 632     caps_kinfo_t caps;
 633
 634     if ((caps = caps_find_id(curthread, uap->portid)) == NULL)
 635         return(EINVAL);
 636     caps_term(caps, CAPKF_TDLIST|CAPKF_HLIST|CAPKF_FLUSH|CAPKF_RCAPS, NULL);
 637     caps_drop(caps);
 638     return(0);
 639 }
 640
 641 int
 642 sys_caps_sys_setgen(struct caps_sys_setgen_args *uap)
 643 {
 644     caps_kinfo_t caps;
 645     int error;
 646
 647     if ((caps = caps_find_id(curthread, uap->portid)) == NULL)
 648         return(EINVAL);
 649     if (caps->ci_type == CAPT_FORKED) {
 650         error = ENOTCONN;
 651     } else {
 652         caps->ci_gen = uap->gen;
 653         error = 0;
 654     }
 655     caps_drop(caps);
 656     return(error);
 657 }
 658
 659 int
 660 sys_caps_sys_getgen(struct caps_sys_getgen_args *uap)
 661 {
 662     caps_kinfo_t caps;
 663     int error;
 664
 665     if ((caps = caps_find_id(curthread, uap->portid)) == NULL)
 666         return(EINVAL);
 667     if (caps->ci_type == CAPT_FORKED) {
 668         error = ENOTCONN;
 669     } else if (caps->ci_rcaps == NULL) {
 670         error = EINVAL;
 671     } else {
 672         uap->sysmsg_result64 = caps->ci_rcaps->ci_gen;
 673         error = 0;
 674     }
 675     caps_drop(caps);
 676     return(error);
 677 }
 678
 679 /*
 680  * caps_sys_put(portid, msg, msgsize)
 681  *
 682  * Send an opaque message of the specified size to the specified port.  This
 683  * function may only be used with a client port.  The message id is returned.
 684  */
 685 int
 686 sys_caps_sys_put(struct caps_sys_put_args *uap)
 687 {
 688     caps_kinfo_t caps;
 689     caps_kmsg_t msg;
 690     struct proc *p = curproc;
 691     int error;
 692
 693     if (uap->msgsize < 0)
 694         return(EINVAL);
 695     if ((caps = caps_find_id(curthread, uap->portid)) == NULL)
 696         return(EINVAL);
 697     if (caps->ci_type == CAPT_FORKED) {
 698         error = ENOTCONN;
 699     } else if (caps->ci_rcaps == NULL) {
 700         error = EINVAL;
 701     } else if (caps->ci_cmsgcount > CAPS_MAXINPROG) {
 702         /*
 703          * If this client has queued a large number of messages return
 704          * ENOBUFS.  The client must process some replies before it can
 705          * send new messages.  The server can also throttle a client by
 706          * holding its replies.  XXX allow a server to refuse messages from
 707          * a client.
 708          */
 709         error = ENOBUFS;
 710     } else {
 711         msg = caps_alloc_msg(caps);
 712         uap->sysmsg_offset = msg->km_msgid.c_id;
 713
 714         /*
 715          * If the remote end is closed return ENOTCONN immediately, otherwise
 716          * send it to the remote end.
 717          *
 718          * Note: since this is a new message, caps_load_ccr() returns a remote
 719          * caps of NULL.
 720          */
 721         if (caps->ci_rcaps->ci_flags & CAPKF_CLOSED) {
 722             error = ENOTCONN;
 723             caps_free_msg(msg);
 724         } else {
 725             /*
 726              * new message, load_ccr returns NULL. hold rcaps for put_msg
 727              */
 728             error = 0;
 729             caps_load_ccr(caps, msg, p, uap->msg, uap->msgsize);
 730             caps_hold(caps->ci_rcaps);
 731             ++caps->ci_cmsgcount;
 732             caps_put_msg(caps->ci_rcaps, msg, CAPMS_REQUEST); /* drops rcaps */
 733         }
 734     }
 735     caps_drop(caps);
 736     return(error);
 737 }
 738
 739 /*
 740  * caps_sys_reply(portid, msg, msgsize, msgid)
 741  *
 742  * Reply to the message referenced by the specified msgid, supplying opaque
 743  * data back to the originator.
 744  */
 745 int
 746 sys_caps_sys_reply(struct caps_sys_reply_args *uap)
 747 {
 748     caps_kinfo_t caps;
 749     caps_kinfo_t rcaps;
 750     caps_kmsg_t msg;
 751     struct proc *p;
 752     int error;
 753
 754     if (uap->msgsize < 0)
 755         return(EINVAL);
 756     if ((caps = caps_find_id(curthread, uap->portid)) == NULL)
 757         return(EINVAL);
 758     if (caps->ci_type == CAPT_FORKED) {
 759         /*
 760          * The caps structure is just a fork placeholder, tell the caller
 761          * that he has to reconnect.
 762          */
 763         error = ENOTCONN;
 764     } else if ((msg = caps_find_msg(caps, uap->msgcid)) == NULL) {
 765         /*
 766          * Could not find message being replied to (other side might have
 767          * gone away).
 768          */
 769         error = EINVAL;
 770     } else if ((msg->km_flags & CAPKMF_ONUSERQ) == 0) {
 771         /*
 772          * Trying to reply to a non-replyable message
 773          */
 774         error = EINVAL;
 775     } else {
 776         /*
 777          * If the remote end is closed requeue to ourselves for disposal.
 778          * Otherwise send the reply to the other end (the other end will
 779          * return a passive DISPOSE to us when it has eaten the data)
 780          */
 781         error = 0;
 782         caps_dequeue_msg(caps, msg);
 783         p = curproc;
 784         if (msg->km_mcaps->ci_flags & CAPKF_CLOSED) {
 785             caps_drop(caps_load_ccr(caps, msg, p, NULL, 0));
 786             caps_hold(caps);                    /* ref for message */
 787             caps_put_msg(caps, msg, CAPMS_DISPOSE);
 788         } else {
 789             rcaps = caps_load_ccr(caps, msg, p, uap->msg, uap->msgsize);
 790             caps_put_msg(rcaps, msg, CAPMS_REPLY);
 791         }
 792     }
 793     caps_drop(caps);
 794     return(error);
 795 }
 796
 797 /*
 798  * caps_sys_get(portid, msg, maxsize, msgid, ccr)
 799  *
 800  * Retrieve the next ready message on the port, store its message id in
 801  * uap->msgid and return the length of the message.  If the message is too
 802  * large to fit the message id, length, and creds are still returned, but
 803  * the message is not dequeued (the caller is expected to call again with
 804  * a larger buffer or to reply the messageid if it does not want to handle
 805  * the message).
 806  *
 807  * EWOULDBLOCK is returned if no messages are pending.  Note that 0-length
 808  * messages are perfectly acceptable so 0 can be legitimately returned.
 809  */
 810 int
 811 sys_caps_sys_get(struct caps_sys_get_args *uap)
 812 {
 813     caps_kinfo_t caps;
 814     caps_kmsg_t msg;
 815     int error;
 816
 817     if (uap->maxsize < 0)
 818         return(EINVAL);
 819     if ((caps = caps_find_id(curthread, uap->portid)) == NULL)
 820         return(EINVAL);
 821     if (caps->ci_type == CAPT_FORKED) {
 822         error = ENOTCONN;
 823     } else if ((msg = TAILQ_FIRST(&caps->ci_msgpendq)) == NULL) {
 824         error = EWOULDBLOCK;
 825     } else {
 826         error = caps_process_msg(caps, msg, uap);
 827     }
 828     caps_drop(caps);
 829     return(error);
 830 }
 831
 832 /*
 833  * caps_sys_wait(portid, msg, maxsize, msgid, ccr)
 834  *
 835  * Retrieve the next ready message on the port, store its message id in
 836  * uap->msgid and return the length of the message.  If the message is too
 837  * large to fit the message id, length, and creds are still returned, but
 838  * the message is not dequeued (the caller is expected to call again with
 839  * a larger buffer or to reply the messageid if it does not want to handle
 840  * the message).
 841  *
 842  * This function blocks until interrupted or a message is received.
 843  * Note that 0-length messages are perfectly acceptable so 0 can be
 844  * legitimately returned.
 845  */
 846 int
 847 sys_caps_sys_wait(struct caps_sys_wait_args *uap)
 848 {
 849     caps_kinfo_t caps;
 850     caps_kmsg_t msg;
 851     int error;
 852
 853     if (uap->maxsize < 0)
 854         return(EINVAL);
 855     if ((caps = caps_find_id(curthread, uap->portid)) == NULL)
 856         return(EINVAL);
 857     if (caps->ci_type == CAPT_FORKED) {
 858         error = ENOTCONN;
 859     } else {
 860         error = 0;
 861         while ((msg = TAILQ_FIRST(&caps->ci_msgpendq)) == NULL) {
 862             if ((error = tsleep(caps, PCATCH, "caps", 0)) != 0)
 863                 break;
 864         }
 865         if (error == 0) {
 866             error = caps_process_msg(caps, msg,
 867                                 (struct caps_sys_get_args *)uap);
 868         }
 869     }
 870     caps_drop(caps);
 871     return(error);
 872 }
 873
 874 static int
 875 caps_process_msg(caps_kinfo_t caps, caps_kmsg_t msg, struct caps_sys_get_args *uap)
 876 {
 877     int error = 0;
 878     int msgsize;
 879     caps_kinfo_t rcaps;
 880
 881     msg->km_flags |= CAPKMF_PEEKED;
 882     msgsize = msg->km_xio.xio_bytes;
 883     if (msgsize <= uap->maxsize)
 884         caps_dequeue_msg(caps, msg);
 885
 886     if (msg->km_xio.xio_bytes != 0) {
 887         error = xio_copy_xtou(&msg->km_xio, 0, uap->msg,
 888                             min(msg->km_xio.xio_bytes, uap->maxsize));
 889         if (error) {
 890             if (msg->km_mcaps->ci_td && msg->km_mcaps->ci_td->td_proc) {
 891                 kprintf("xio_copy_xtou: error %d from proc %d\n",
 892                         error, msg->km_mcaps->ci_td->td_proc->p_pid);
 893             }
 894             if (msgsize > uap->maxsize)
 895                 caps_dequeue_msg(caps, msg);
 896             msgsize = 0;
 897             error = 0;
 898         }
 899     }
 900
 901     if (uap->msgid)
 902         error = copyout(&msg->km_msgid, uap->msgid, sizeof(msg->km_msgid));
 903     if (uap->ccr)
 904         error = copyout(&msg->km_ccr, uap->ccr, sizeof(msg->km_ccr));
 905     if (error == 0)
 906         uap->sysmsg_result = msgsize;
 907
 908     /*
 909      * If the message was dequeued we must deal with it.
 910      */
 911     if (msgsize <= uap->maxsize) {
 912         switch(msg->km_state) {
 913         case CAPMS_REQUEST:
 914         case CAPMS_REQUEST_RETRY:
 915             TAILQ_INSERT_TAIL(&caps->ci_msguserq, msg, km_node);
 916             msg->km_flags |= CAPKMF_ONUSERQ;
 917             break;
 918         case CAPMS_REPLY:
 919         case CAPMS_REPLY_RETRY:
 920             --caps->ci_cmsgcount;
 921             rcaps = caps_load_ccr(caps, msg, curproc, NULL, 0);
 922             if (caps == rcaps || (rcaps->ci_flags & CAPKF_CLOSED)) {
 923                 /* degenerate disposal case */
 924                 caps_free_msg(msg);
 925                 caps_drop(rcaps);
 926             } else {
 927                 caps_put_msg(rcaps, msg, CAPMS_DISPOSE);
 928             }
 929             break;
 930         case CAPMS_DISPOSE:
 931             caps_free_msg(msg);
 932             break;
 933         }
 934     }
 935     return(error);
 936 }
 937
 938 /*
 939  * caps_sys_abort(portid, msgcid, flags)
 940  *
 941  *      Abort a previously sent message.  You must still wait for the message
 942  *      to be returned after sending the abort request.  This function will
 943  *      return the appropriate CAPS_ABORT_* code depending on what it had
 944  *      to do.
 945  */
 946 int
 947 sys_caps_sys_abort(struct caps_sys_abort_args *uap)
 948 {
 949     uap->sysmsg_result = CAPS_ABORT_NOTIMPL;
 950     return(0);
 951 }
 952
 953 /*
 954  * KERNEL SYSCALL SEPARATION SUPPORT FUNCTIONS
 955  */
 956
 957 static
 958 caps_kinfo_t
 959 kern_caps_sys_service(const char *name, uid_t uid, gid_t gid,
 960                         struct ucred *cred, int flags, int *error)
 961 {
 962     caps_kinfo_t caps;
 963     int len;
 964
 965     len = strlen(name);
 966
 967     /*
 968      * Make sure we can use the uid and gid
 969      */
 970     if (cred) {
 971         if (cred->cr_uid != 0 && uid != (uid_t)-1 && cred->cr_uid != uid) {
 972             *error = EPERM;
 973             return(NULL);
 974         }
 975         if (cred->cr_uid != 0 && gid != (gid_t)-1 && !groupmember(gid, cred)) {
 976             *error = EPERM;
 977             return(NULL);
 978         }
 979     }
 980
 981     /*
 982      * Handle CAPF_EXCL
 983      */
 984     if (flags & CAPF_EXCL) {
 985         if ((caps = caps_find(name, strlen(name), uid, gid)) != NULL) {
 986             caps_drop(caps);
 987             *error = EEXIST;
 988             return(NULL);
 989         }
 990     }
 991
 992     /*
 993      * Create the service
 994      */
 995     caps = caps_alloc(curthread, name, len,
 996                         uid, gid, flags & CAPF_UFLAGS, CAPT_SERVICE);
 997     wakeup(&caps_waitsvc);
 998     return(caps);
 999 }
1000
1001 static
1002 caps_kinfo_t
1003 kern_caps_sys_client(const char *name, uid_t uid, gid_t gid,
1004                         struct ucred *cred, int flags, int *error)
1005 {
1006     caps_kinfo_t caps, rcaps;
1007     int len;
1008
1009     len = strlen(name);
1010
1011     /*
1012      * Locate the CAPS service (rcaps ref is for caps->ci_rcaps)
1013      */
1014 again:
1015     if ((rcaps = caps_find(name, len, uid, gid)) == NULL) {
1016         if (flags & CAPF_WAITSVC) {
1017             char cbuf[32];
1018             ksnprintf(cbuf, sizeof(cbuf), "C%s", name);
1019             *error = tsleep(&caps_waitsvc, PCATCH, cbuf, 0);
1020             if (*error == 0)
1021                 goto again;
1022         } else {
1023             *error = ENOENT;
1024         }
1025         return(NULL);
1026     }
1027
1028     /*
1029      * Check permissions
1030      */
1031     if (cred) {
1032         *error = EACCES;
1033         if ((flags & CAPF_USER) && (rcaps->ci_flags & CAPF_USER)) {
1034             if (rcaps->ci_uid != (uid_t)-1 && rcaps->ci_uid == cred->cr_uid)
1035                 *error = 0;
1036         }
1037         if ((flags & CAPF_GROUP) && (rcaps->ci_flags & CAPF_GROUP)) {
1038             if (rcaps->ci_gid != (gid_t)-1 && groupmember(rcaps->ci_gid, cred))
1039                 *error = 0;
1040         }
1041         if ((flags & CAPF_WORLD) && (rcaps->ci_flags & CAPF_WORLD)) {
1042             *error = 0;
1043         }
1044         if (*error) {
1045             caps_drop(rcaps);
1046             return(NULL);
1047         }
1048     } else {
1049         *error = 0;
1050     }
1051
1052     /*
1053      * Allocate the client side and connect to the server
1054      */
1055     caps = caps_alloc(curthread, name, len,
1056                         uid, gid, flags & CAPF_UFLAGS, CAPT_CLIENT);
1057     caps->ci_rcaps = rcaps;
1058     caps->ci_flags |= CAPKF_RCAPS;
1059     return(caps);
1060 }
1061