xref: /titanic_52/usr/src/uts/common/os/ipc.c (revision 3f7d54a6b84904c8f4d8daa4c7b577bede7df8b9)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T		*/
27 /*	All Rights Reserved					*/
28 
29 
30 /*
31  * Common Inter-Process Communication routines.
32  *
33  * Overview
34  * --------
35  *
36  * The System V inter-process communication (IPC) facilities provide
37  * three services, message queues, semaphore arrays, and shared memory
38  * segments, which are mananged using filesystem-like namespaces.
39  * Unlike a filesystem, these namespaces aren't mounted and accessible
40  * via a path -- a special API is used to interact with the different
41  * facilities (nothing precludes a VFS-based interface, but the
42  * standards require the special APIs).  Furthermore, these special
43  * APIs don't use file descriptors, nor do they have an equivalent.
44  * This means that every operation which acts on an object needs to
45  * perform the quivalent of a lookup, which in turn means that every
46  * operation can fail if the specified object doesn't exist in the
47  * facility's namespace.
48  *
49  * Objects
50  * -------
51  *
52  * Each object in a namespace has a unique ID, which is assigned by the
53  * system and is used to identify the object when performing operations
54  * on it.  An object can also have a key, which is selected by the user
55  * at allocation time and is used as a primitive rendezvous mechanism.
56  * An object without a key is said to have a "private" key.
57  *
58  * To perform an operation on an object given its key, one must first
59  * perform a lookup and obtain its ID.  The ID is then used to identify
60  * the object when performing the operation.  If the object has a
61  * private key, the ID must be known or obtained by other means.
62  *
63  * Each object in the namespace has a creator uid and gid, as well as
64  * an owner uid and gid.  Both are initialized with the ruid and rgid
65  * of the process which created the object.  The creator or current
66  * owner has the ability to change the owner of the object.
67  *
68  * Each object in the namespace has a set of file-like permissions,
69  * which, in conjunction with the creator and owner uid and gid,
70  * control read and write access to the object (execute is ignored).
71  *
72  * Each object also has a creator project and zone, which are used to
73  * account for its resource usage.
74  *
75  * Operations
76  * ----------
77  *
78  * There are five operations which all three facilities have in
79  * common: GET, SET, STAT, RMID, and IDS.
80  *
81  * GET, like open, is used to allocate a new object or obtain an
82  * existing one (using its key).  It takes a key, a set of flags and
83  * mode bits, and optionally facility-specific arguments.  If the key
84  * is IPC_PRIVATE, a new object with the requested mode bits and
85  * facility-specific attributes is created.  If the key isn't
86  * IPC_PRIVATE, the GET will attempt to look up the specified key and
87  * either return that or create a new key depending on the state of the
88  * IPC_CREAT and IPC_EXCL flags, much like open.  If GET needs to
89  * allocate an object, it can fail if there is insufficient space in
90  * the namespace (the maximum number of ids for the facility has been
91  * exceeded) or if the facility-specific initialization fails.  If GET
92  * finds an object it can return, it can still fail if that object's
93  * permissions or facility-specific attributes are less than those
94  * requested.
95  *
96  * SET is used to adjust facility-specific parameters of an object, in
97  * addition to the owner uid and gid, and mode bits.  It can fail if
98  * the caller isn't the creator or owner.
99  *
100  * STAT is used to obtain information about an object including the
101  * general attributes object described as well as facility-specific
102  * information.  It can fail if the caller doesn't have read
103  * permission.
104  *
105  * RMID removes an object from the namespace.  Subsequent operations
106  * using the object's ID or key will fail (until another object is
107  * created with the same key or ID).  Since an RMID may be performed
108  * asynchronously with other operations, it is possible that other
109  * threads and/or processes will have references to the object.  While
110  * a facility may have actions which need to be performed at RMID time,
111  * only when all references are dropped can the object be destroyed.
112  * RMID will fail if the caller isn't the creator or owner.
113  *
114  * IDS obtains a list of all IDs in a facility's namespace.  There are
115  * no facility-specific behaviors of IDS.
116  *
117  * Design
118  * ------
119  *
120  * Because some IPC facilities provide services whose operations must
121  * scale, a mechanism which allows fast, concurrent access to
122  * individual objects is needed.  Of primary importance is object
123  * lookup based on ID (SET, STAT, others).  Allocation (GET),
124  * deallocation (RMID), ID enumeration (IDS), and key lookups (GET) are
125  * lesser concerns, but should be implemented in such a way that ID
126  * lookup isn't affected (at least not in the common case).
127  *
128  * Starting from the bottom up, each object is represented by a
129  * structure, the first member of which must be a kipc_perm_t.  The
130  * kipc_perm_t contains the information described above in "Objects", a
131  * reference count (since the object may continue to exist after it has
132  * been removed from the namespace), as well as some additional
133  * metadata used to manage data structure membership.  These objects
134  * are dynamically allocated.
135  *
136  * Above the objects is a power-of-two sized table of ID slots.  Each
137  * slot contains a pointer to an object, a sequence number, and a
138  * lock.  An object's ID is a function of its slot's index in the table
139  * and its slot's sequence number.  Every time a slot is released (via
140  * RMID) its sequence number is increased.  Strictly speaking, the
141  * sequence number is unnecessary.  However, checking the sequence
142  * number after a lookup provides a certain degree of robustness
143  * against the use of stale IDs (useful since nothing else does).  When
144  * the table fills up, it is resized (see Locking, below).
145  *
146  * Of an ID's 31 bits (an ID is, as defined by the standards, a signed
147  * int) the top IPC_SEQ_BITS are used for the sequence number with the
148  * remainder holding the index into the table.  The size of the table
149  * is therefore bounded at 2 ^ (31 - IPC_SEQ_BITS) slots.
150  *
151  * Managing this table is the ipc_service structure.  It contains a
152  * pointer to the dynamically allocated ID table, a namespace-global
153  * lock, an id_space for managing the free space in the table, and
154  * sundry other metadata necessary for the maintenance of the
155  * namespace.  An AVL tree of all keyed objects in the table (sorted by
156  * key) is used for key lookups.  An unordered doubly linked list of
157  * all objects in the namespace (keyed or not) is maintained to
158  * facilitate ID enumeration.
159  *
160  * To help visualize these relationships, here's a picture of a
161  * namespace with a table of size 8 containing three objects
162  * (IPC_SEQ_BITS = 28):
163  *
164  *
165  * +-ipc_service_t--+
166  * | table          *---\
167  * | keys           *---+----------------------\
168  * | all ids        *--\|                      |
169  * |                |  ||                      |
170  * +----------------+  ||                      |
171  *                     ||                      |
172  * /-------------------/|                      |
173  * |    /---------------/                      |
174  * |    |                                      |
175  * |    v                                      |
176  * |  +-0------+-1------+-2------+-3------+-4--+---+-5------+-6------+-7------+
177  * |  | Seq=3  |        |        | Seq=1  |    :   |        |        | Seq=6  |
178  * |  |        |        |        |        |    :   |        |        |        |
179  * |  +-*------+--------+--------+-*------+----+---+--------+--------+-*------+
180  * |    |                          |           |                       |
181  * |    |                      /---/           |      /----------------/
182  * |    |                      |               |      |
183  * |    v                      v               |      v
184  * |  +-kipc_perm_t-+        +-kipc_perm_t-+   |    +-kipc_perm_t-+
185  * |  | id=0x30     |        | id=0x13     |   |    | id=0x67     |
186  * |  | key=0xfeed  |        | key=0xbeef  |   |    | key=0xcafe  |
187  * \->| [list]      |<------>| [list]      |<------>| [list]      |
188  * /->| [avl left]  x   /--->| [avl left]  x   \--->| [avl left]  *---\
189  * |  | [avl right] x   |    | [avl right] x        | [avl right] *---+-\
190  * |  |             |   |    |             |        |             |   | |
191  * |  +-------------+   |    +-------------+        +-------------+   | |
192  * |                    \---------------------------------------------/ |
193  * \--------------------------------------------------------------------/
194  *
195  * Locking
196  * -------
197  *
198  * There are three locks (or sets of locks) which are used to ensure
199  * correctness: the slot locks, the namespace lock, and p_lock (needed
200  * when checking resource controls).  Their ordering is
201  *
202  *   namespace lock -> slot lock 0 -> ... -> slot lock t -> p_lock
203  *
204  * Generally speaking, the namespace lock is used to protect allocation
205  * and removal from the namespace, ID enumeration, and resizing the ID
206  * table.  Specifically:
207  *
208  * - write access to all fields of the ipc_service structure
209  * - read access to all variable fields of ipc_service except
210  *   ipcs_tabsz (table size) and ipcs_table (the table pointer)
211  * - read/write access to ipc_avl, ipc_list in visible objects'
212  *   kipc_perm structures (i.e. objects which have been removed from
213  *   the namespace don't have this restriction)
214  * - write access to ipct_seq and ipct_data in the table entries
215  *
216  * A slot lock by itself is meaningless (except when resizing).  Of
217  * greater interest conceptually is the notion of an ID lock -- a
218  * "virtual lock" which refers to whichever slot lock an object's ID
219  * currently hashes to.
220  *
221  * An ID lock protects all objects with that ID.  Normally there will
222  * only be one such object: the one pointed to by the locked slot.
223  * However, if an object is removed from the namespace but retains
224  * references (e.g. an attached shared memory segment which has been
225  * RMIDed), it continues to use the lock associated with its original
226  * ID.  While this can result in increased contention, operations which
227  * require taking the ID lock of removed objects are infrequent.
228  *
229  * Specifically, an ID lock protects the contents of an object's
230  * structure, including the contents of the embedded kipc_perm
231  * structure (but excluding those fields protected by the namespace
232  * lock).  It also protects the ipct_seq and ipct_data fields in its
233  * slot (it is really a slot lock, after all).
234  *
235  * Recall that the table is resizable.  To avoid requiring every ID
236  * lookup to take a global lock, a scheme much like that employed for
237  * file descriptors (see the comment above UF_ENTER in user.h) is
238  * used.  Note that the sequence number and data pointer are protected
239  * by both the namespace lock and their slot lock.  When the table is
240  * resized, the following operations take place:
241  *
242  *   1) A new table is allocated.
243  *   2) The global lock is taken.
244  *   3) All old slots are locked, in order.
245  *   4) The first half of the new slots are locked.
246  *   5) All table entries are copied to the new table, and cleared from
247  *	the old table.
248  *   6) The ipc_service structure is updated to point to the new table.
249  *   7) The ipc_service structure is updated with the new table size.
250  *   8) All slot locks (old and new) are dropped.
251  *
252  * Because the slot locks are embedded in the table, ID lookups and
253  * other operations which require taking an slot lock need to verify
254  * that the lock taken wasn't part of a stale table.  This is
255  * accomplished by checking the table size before and after
256  * dereferencing the table pointer and taking the lock: if the size
257  * changes, the lock must be dropped and reacquired.  It is this
258  * additional work which distinguishes an ID lock from a slot lock.
259  *
260  * Because we can't guarantee that threads aren't accessing the old
261  * tables' locks, they are never deallocated.  To prevent spurious
262  * reports of memory leaks, a pointer to the discarded table is stored
263  * in the new one in step 5.  (Theoretically ipcs_destroy will delete
264  * the discarded tables, but it is only ever called from a failed _init
265  * invocation; i.e. when there aren't any.)
266  *
267  * Interfaces
268  * ----------
269  *
270  * The following interfaces are provided by the ipc module for use by
271  * the individual IPC facilities:
272  *
273  * ipcperm_access
274  *
275  *   Given an object and a cred structure, determines if the requested
276  *   access type is allowed.
277  *
278  * ipcperm_set, ipcperm_stat,
279  * ipcperm_set64, ipcperm_stat64
280  *
281  *   Performs the common portion of an STAT or SET operation.  All
282  *   (except stat and stat64) can fail, so they should be called before
283  *   any facility-specific non-reversible changes are made to an
284  *   object.  Similarly, the set operations have side effects, so they
285  *   should only be called once the possibility of a facility-specific
286  *   failure is eliminated.
287  *
288  * ipcs_create
289  *
290  *   Creates an IPC namespace for use by an IPC facility.
291  *
292  * ipcs_destroy
293  *
294  *   Destroys an IPC namespace.
295  *
296  * ipcs_lock, ipcs_unlock
297  *
298  *   Takes the namespace lock.  Ideally such access wouldn't be
299  *   necessary, but there may be facility-specific data protected by
300  *   this lock (e.g. project-wide resource consumption).
301  *
302  * ipc_lock
303  *
304  *   Takes the lock associated with an ID.  Can't fail.
305  *
306  * ipc_relock
307  *
308  *   Like ipc_lock, but takes a pointer to a held lock.  Drops the lock
309  *   unless it is the one that would have been returned by ipc_lock.
310  *   Used after calls to cv_wait.
311  *
312  * ipc_lookup
313  *
314  *   Performs an ID lookup, returns with the ID lock held.  Fails if
315  *   the ID doesn't exist in the namespace.
316  *
317  * ipc_hold
318  *
319  *   Takes a reference on an object.
320  *
321  * ipc_rele
322  *
323  *   Releases a reference on an object, and drops the object's lock.
324  *   Calls the object's destructor if last reference is being
325  *   released.
326  *
327  * ipc_rele_locked
328  *
329  *   Releases a reference on an object.  Doesn't drop lock, and may
330  *   only be called when there is more than one reference to the
331  *   object.
332  *
333  * ipc_get, ipc_commit_begin, ipc_commit_end, ipc_cleanup
334  *
335  *   Components of a GET operation.  ipc_get performs a key lookup,
336  *   allocating an object if the key isn't found (returning with the
337  *   namespace lock and p_lock held), and returning the existing object
338  *   if it is (with the object lock held).  ipc_get doesn't modify the
339  *   namespace.
340  *
341  *   ipc_commit_begin begins the process of inserting an object
342  *   allocated by ipc_get into the namespace, and can fail.  If
343  *   successful, it returns with the namespace lock and p_lock held.
344  *   ipc_commit_end completes the process of inserting an object into
345  *   the namespace and can't fail.  The facility can call ipc_cleanup
346  *   at any time following a successful ipc_get and before
347  *   ipc_commit_end or a failed ipc_commit_begin to fail the
348  *   allocation.  Pseudocode for the suggested GET implementation:
349  *
350  *   top:
351  *
352  *     ipc_get
353  *
354  *     if failure
355  *       return
356  *
357  *     if found {
358  *
359  *	 if object meets criteria
360  *	   unlock object and return success
361  *       else
362  *	   unlock object and return failure
363  *
364  *     } else {
365  *
366  *	 perform resource control tests
367  *	 drop namespace lock, p_lock
368  *	 if failure
369  *	   ipc_cleanup
370  *
371  *       perform facility-specific initialization
372  *	 if failure {
373  *	   facility-specific cleanup
374  *	   ipc_cleanup
375  *       }
376  *
377  *	 ( At this point the object should be destructible using the
378  *	   destructor given to ipcs_create )
379  *
380  *       ipc_commit_begin
381  *	 if retry
382  *	   goto top
383  *       else if failure
384  *         return
385  *
386  *       perform facility-specific resource control tests/allocations
387  *	 if failure
388  *	   ipc_cleanup
389  *
390  *	 ipc_commit_end
391  *	 perform any infallible post-creation actions, unlock, and return
392  *
393  *     }
394  *
395  * ipc_rmid
396  *
397  *   Performs the common portion of an RMID operation -- looks up an ID
398  *   removes it, and calls the a facility-specific function to do
399  *   RMID-time cleanup on the private portions of the object.
400  *
401  * ipc_ids
402  *
403  *   Performs the common portion of an IDS operation.
404  *
405  */
406 
407 #include <sys/types.h>
408 #include <sys/param.h>
409 #include <sys/cred.h>
410 #include <sys/policy.h>
411 #include <sys/proc.h>
412 #include <sys/user.h>
413 #include <sys/ipc.h>
414 #include <sys/ipc_impl.h>
415 #include <sys/errno.h>
416 #include <sys/systm.h>
417 #include <sys/list.h>
418 #include <sys/atomic.h>
419 #include <sys/zone.h>
420 #include <sys/task.h>
421 #include <sys/modctl.h>
422 
423 #include <c2/audit.h>
424 
425 static struct modlmisc modlmisc = {
426 	&mod_miscops,
427 	"common ipc code",
428 };
429 
430 static struct modlinkage modlinkage = {
431 	MODREV_1, (void *)&modlmisc, NULL
432 };
433 
434 
435 int
436 _init(void)
437 {
438 	return (mod_install(&modlinkage));
439 }
440 
441 int
442 _fini(void)
443 {
444 	return (mod_remove(&modlinkage));
445 }
446 
447 int
448 _info(struct modinfo *modinfop)
449 {
450 	return (mod_info(&modlinkage, modinfop));
451 }
452 
453 
454 /*
455  * Check message, semaphore, or shared memory access permissions.
456  *
457  * This routine verifies the requested access permission for the current
458  * process.  The zone ids are compared, and the appropriate bits are
459  * checked corresponding to owner, group (including the list of
460  * supplementary groups), or everyone.  Zero is returned on success.
461  * On failure, the security policy is asked to check to override the
462  * permissions check; the policy will either return 0 for access granted
463  * or EACCES.
464  *
465  * Access to objects in other zones requires that the caller be in the
466  * global zone and have the appropriate IPC_DAC_* privilege, regardless
467  * of whether the uid or gid match those of the object.  Note that
468  * cross-zone accesses will normally never get here since they'll
469  * fail in ipc_lookup or ipc_get.
470  *
471  * The arguments must be set up as follows:
472  * 	p - Pointer to permission structure to verify
473  * 	mode - Desired access permissions
474  */
475 int
476 ipcperm_access(kipc_perm_t *p, int mode, cred_t *cr)
477 {
478 	int shifts = 0;
479 	uid_t uid = crgetuid(cr);
480 	zoneid_t zoneid = getzoneid();
481 
482 	if (p->ipc_zoneid == zoneid) {
483 		if (uid != p->ipc_uid && uid != p->ipc_cuid) {
484 			shifts += 3;
485 			if (!groupmember(p->ipc_gid, cr) &&
486 			    !groupmember(p->ipc_cgid, cr))
487 				shifts += 3;
488 		}
489 
490 		mode &= ~(p->ipc_mode << shifts);
491 
492 		if (mode == 0)
493 			return (0);
494 	} else if (zoneid != GLOBAL_ZONEID)
495 		return (EACCES);
496 
497 	return (secpolicy_ipc_access(cr, p, mode));
498 }
499 
500 /*
501  * There are two versions of the ipcperm_set/stat functions:
502  *   ipcperm_???        - for use with IPC_SET/STAT
503  *   ipcperm_???_64     - for use with IPC_SET64/STAT64
504  *
505  * These functions encapsulate the common portions (copying, permission
506  * checks, and auditing) of the set/stat operations.  All, except for
507  * stat and stat_64 which are void, return 0 on success or a non-zero
508  * errno value on error.
509  */
510 
511 int
512 ipcperm_set(ipc_service_t *service, struct cred *cr,
513     kipc_perm_t *kperm, struct ipc_perm *perm, model_t model)
514 {
515 	STRUCT_HANDLE(ipc_perm, lperm);
516 	uid_t uid;
517 	gid_t gid;
518 	mode_t mode;
519 	zone_t *zone;
520 
521 	ASSERT(IPC_LOCKED(service, kperm));
522 
523 	STRUCT_SET_HANDLE(lperm, model, perm);
524 	uid = STRUCT_FGET(lperm, uid);
525 	gid = STRUCT_FGET(lperm, gid);
526 	mode = STRUCT_FGET(lperm, mode);
527 
528 	if (secpolicy_ipc_owner(cr, kperm) != 0)
529 		return (EPERM);
530 
531 	zone = crgetzone(cr);
532 	if (!VALID_UID(uid, zone) || !VALID_GID(gid, zone))
533 		return (EINVAL);
534 
535 	kperm->ipc_uid = uid;
536 	kperm->ipc_gid = gid;
537 	kperm->ipc_mode = (mode & 0777) | (kperm->ipc_mode & ~0777);
538 
539 	if (AU_AUDITING())
540 		audit_ipcget(service->ipcs_atype, kperm);
541 
542 	return (0);
543 }
544 
545 void
546 ipcperm_stat(struct ipc_perm *perm, kipc_perm_t *kperm, model_t model)
547 {
548 	STRUCT_HANDLE(ipc_perm, lperm);
549 
550 	STRUCT_SET_HANDLE(lperm, model, perm);
551 	STRUCT_FSET(lperm, uid, kperm->ipc_uid);
552 	STRUCT_FSET(lperm, gid, kperm->ipc_gid);
553 	STRUCT_FSET(lperm, cuid, kperm->ipc_cuid);
554 	STRUCT_FSET(lperm, cgid, kperm->ipc_cgid);
555 	STRUCT_FSET(lperm, mode, kperm->ipc_mode);
556 	STRUCT_FSET(lperm, seq, 0);
557 	STRUCT_FSET(lperm, key, kperm->ipc_key);
558 }
559 
560 int
561 ipcperm_set64(ipc_service_t *service, struct cred *cr,
562     kipc_perm_t *kperm, ipc_perm64_t *perm64)
563 {
564 	zone_t *zone;
565 
566 	ASSERT(IPC_LOCKED(service, kperm));
567 
568 	if (secpolicy_ipc_owner(cr, kperm) != 0)
569 		return (EPERM);
570 
571 	zone = crgetzone(cr);
572 	if (!VALID_UID(perm64->ipcx_uid, zone) ||
573 	    !VALID_GID(perm64->ipcx_gid, zone))
574 		return (EINVAL);
575 
576 	kperm->ipc_uid = perm64->ipcx_uid;
577 	kperm->ipc_gid = perm64->ipcx_gid;
578 	kperm->ipc_mode = (perm64->ipcx_mode & 0777) |
579 	    (kperm->ipc_mode & ~0777);
580 
581 	if (AU_AUDITING())
582 		audit_ipcget(service->ipcs_atype, kperm);
583 
584 	return (0);
585 }
586 
587 void
588 ipcperm_stat64(ipc_perm64_t *perm64, kipc_perm_t *kperm)
589 {
590 	perm64->ipcx_uid = kperm->ipc_uid;
591 	perm64->ipcx_gid = kperm->ipc_gid;
592 	perm64->ipcx_cuid = kperm->ipc_cuid;
593 	perm64->ipcx_cgid = kperm->ipc_cgid;
594 	perm64->ipcx_mode = kperm->ipc_mode;
595 	perm64->ipcx_key = kperm->ipc_key;
596 	perm64->ipcx_projid = kperm->ipc_proj->kpj_id;
597 	perm64->ipcx_zoneid = kperm->ipc_zoneid;
598 }
599 
600 
601 /*
602  * ipc key comparator.
603  */
604 static int
605 ipc_key_compar(const void *a, const void *b)
606 {
607 	kipc_perm_t *aperm = (kipc_perm_t *)a;
608 	kipc_perm_t *bperm = (kipc_perm_t *)b;
609 	int ak = aperm->ipc_key;
610 	int bk = bperm->ipc_key;
611 	zoneid_t az;
612 	zoneid_t bz;
613 
614 	ASSERT(ak != IPC_PRIVATE);
615 	ASSERT(bk != IPC_PRIVATE);
616 
617 	/*
618 	 * Compare key first, then zoneid.  This optimizes performance for
619 	 * systems with only one zone, since the zone checks will only be
620 	 * made when the keys match.
621 	 */
622 	if (ak < bk)
623 		return (-1);
624 	if (ak > bk)
625 		return (1);
626 
627 	/* keys match */
628 	az = aperm->ipc_zoneid;
629 	bz = bperm->ipc_zoneid;
630 	if (az < bz)
631 		return (-1);
632 	if (az > bz)
633 		return (1);
634 	return (0);
635 }
636 
637 /*
638  * Create an ipc service.
639  */
640 ipc_service_t *
641 ipcs_create(const char *name, rctl_hndl_t proj_rctl, rctl_hndl_t zone_rctl,
642     size_t size, ipc_func_t *dtor, ipc_func_t *rmid, int audit_type,
643     size_t rctl_offset)
644 {
645 	ipc_service_t *result;
646 
647 	result = kmem_alloc(sizeof (ipc_service_t), KM_SLEEP);
648 
649 	mutex_init(&result->ipcs_lock, NULL, MUTEX_ADAPTIVE, NULL);
650 	result->ipcs_count = 0;
651 	avl_create(&result->ipcs_keys, ipc_key_compar, size, 0);
652 	result->ipcs_tabsz = IPC_IDS_MIN;
653 	result->ipcs_table =
654 	    kmem_zalloc(IPC_IDS_MIN * sizeof (ipc_slot_t), KM_SLEEP);
655 	result->ipcs_ssize = size;
656 	result->ipcs_ids = id_space_create(name, 0, IPC_IDS_MIN);
657 	result->ipcs_dtor = dtor;
658 	result->ipcs_rmid = rmid;
659 	result->ipcs_proj_rctl = proj_rctl;
660 	result->ipcs_zone_rctl = zone_rctl;
661 	result->ipcs_atype = audit_type;
662 	ASSERT(rctl_offset < sizeof (ipc_rqty_t));
663 	result->ipcs_rctlofs = rctl_offset;
664 	list_create(&result->ipcs_usedids, sizeof (kipc_perm_t),
665 	    offsetof(kipc_perm_t, ipc_list));
666 
667 	return (result);
668 }
669 
670 /*
671  * Destroy an ipc service.
672  */
673 void
674 ipcs_destroy(ipc_service_t *service)
675 {
676 	ipc_slot_t *slot, *next;
677 
678 	mutex_enter(&service->ipcs_lock);
679 
680 	ASSERT(service->ipcs_count == 0);
681 	avl_destroy(&service->ipcs_keys);
682 	list_destroy(&service->ipcs_usedids);
683 	id_space_destroy(service->ipcs_ids);
684 
685 	for (slot = service->ipcs_table; slot; slot = next) {
686 		next = slot[0].ipct_chain;
687 		kmem_free(slot, service->ipcs_tabsz * sizeof (ipc_slot_t));
688 		service->ipcs_tabsz >>= 1;
689 	}
690 
691 	mutex_destroy(&service->ipcs_lock);
692 	kmem_free(service, sizeof (ipc_service_t));
693 }
694 
695 /*
696  * Takes the service lock.
697  */
698 void
699 ipcs_lock(ipc_service_t *service)
700 {
701 	mutex_enter(&service->ipcs_lock);
702 }
703 
704 /*
705  * Releases the service lock.
706  */
707 void
708 ipcs_unlock(ipc_service_t *service)
709 {
710 	mutex_exit(&service->ipcs_lock);
711 }
712 
713 
714 /*
715  * Locks the specified ID.  Returns the ID's ID table index.
716  */
717 static int
718 ipc_lock_internal(ipc_service_t *service, uint_t id)
719 {
720 	uint_t	tabsz;
721 	uint_t	index;
722 	kmutex_t *mutex;
723 
724 	for (;;) {
725 		tabsz = service->ipcs_tabsz;
726 		membar_consumer();
727 		index = id & (tabsz - 1);
728 		mutex = &service->ipcs_table[index].ipct_lock;
729 		mutex_enter(mutex);
730 		if (tabsz == service->ipcs_tabsz)
731 			break;
732 		mutex_exit(mutex);
733 	}
734 
735 	return (index);
736 }
737 
738 /*
739  * Locks the specified ID.  Returns a pointer to the ID's lock.
740  */
741 kmutex_t *
742 ipc_lock(ipc_service_t *service, int id)
743 {
744 	uint_t index;
745 
746 	/*
747 	 * These assertions don't reflect requirements of the code
748 	 * which follows, but they should never fail nonetheless.
749 	 */
750 	ASSERT(id >= 0);
751 	ASSERT(IPC_INDEX(id) < service->ipcs_tabsz);
752 	index = ipc_lock_internal(service, id);
753 
754 	return (&service->ipcs_table[index].ipct_lock);
755 }
756 
757 /*
758  * Checks to see if the held lock provided is the current lock for the
759  * specified id.  If so, we return it instead of dropping it and
760  * returning the result of ipc_lock.  This is intended to speed up cv
761  * wakeups where we are left holding a lock which could be stale, but
762  * probably isn't.
763  */
764 kmutex_t *
765 ipc_relock(ipc_service_t *service, int id, kmutex_t *lock)
766 {
767 	ASSERT(id >= 0);
768 	ASSERT(IPC_INDEX(id) < service->ipcs_tabsz);
769 	ASSERT(MUTEX_HELD(lock));
770 
771 	if (&service->ipcs_table[IPC_INDEX(id)].ipct_lock == lock)
772 		return (lock);
773 
774 	mutex_exit(lock);
775 	return (ipc_lock(service, id));
776 }
777 
778 /*
779  * Performs an ID lookup.  If the ID doesn't exist or has been removed,
780  * or isn't visible to the caller (because of zones), NULL is returned.
781  * Otherwise, a pointer to the ID's perm structure and held ID lock are
782  * returned.
783  */
784 kmutex_t *
785 ipc_lookup(ipc_service_t *service, int id, kipc_perm_t **perm)
786 {
787 	kipc_perm_t *result;
788 	uint_t index;
789 
790 	/*
791 	 * There is no need to check to see if id is in-range (i.e.
792 	 * positive and fits into the table).  If it is out-of-range,
793 	 * the id simply won't match the object's.
794 	 */
795 
796 	index = ipc_lock_internal(service, id);
797 	result = service->ipcs_table[index].ipct_data;
798 	if (result == NULL || result->ipc_id != (uint_t)id ||
799 	    !HASZONEACCESS(curproc, result->ipc_zoneid)) {
800 		mutex_exit(&service->ipcs_table[index].ipct_lock);
801 		return (NULL);
802 	}
803 
804 	ASSERT(IPC_SEQ(id) == service->ipcs_table[index].ipct_seq);
805 
806 	*perm = result;
807 	if (AU_AUDITING())
808 		audit_ipc(service->ipcs_atype, id, result);
809 
810 	return (&service->ipcs_table[index].ipct_lock);
811 }
812 
813 /*
814  * Increase the reference count on an ID.
815  */
816 /*ARGSUSED*/
817 void
818 ipc_hold(ipc_service_t *s, kipc_perm_t *perm)
819 {
820 	ASSERT(IPC_INDEX(perm->ipc_id) < s->ipcs_tabsz);
821 	ASSERT(IPC_LOCKED(s, perm));
822 	perm->ipc_ref++;
823 }
824 
825 /*
826  * Decrease the reference count on an ID and drops the ID's lock.
827  * Destroys the ID if the new reference count is zero.
828  */
829 void
830 ipc_rele(ipc_service_t *s, kipc_perm_t *perm)
831 {
832 	int nref;
833 
834 	ASSERT(IPC_INDEX(perm->ipc_id) < s->ipcs_tabsz);
835 	ASSERT(IPC_LOCKED(s, perm));
836 	ASSERT(perm->ipc_ref > 0);
837 
838 	nref = --perm->ipc_ref;
839 	mutex_exit(&s->ipcs_table[IPC_INDEX(perm->ipc_id)].ipct_lock);
840 
841 	if (nref == 0) {
842 		ASSERT(IPC_FREE(perm));		/* ipc_rmid clears IPC_ALLOC */
843 		s->ipcs_dtor(perm);
844 		project_rele(perm->ipc_proj);
845 		zone_rele(perm->ipc_zone);
846 		kmem_free(perm, s->ipcs_ssize);
847 	}
848 }
849 
850 /*
851  * Decrease the reference count on an ID, but don't drop the ID lock.
852  * Used in cases where one thread needs to remove many references (on
853  * behalf of other parties).
854  */
855 void
856 ipc_rele_locked(ipc_service_t *s, kipc_perm_t *perm)
857 {
858 	ASSERT(perm->ipc_ref > 1);
859 	ASSERT(IPC_INDEX(perm->ipc_id) < s->ipcs_tabsz);
860 	ASSERT(IPC_LOCKED(s, perm));
861 
862 	perm->ipc_ref--;
863 }
864 
865 
866 /*
867  * Internal function to grow the service ID table.
868  */
869 static int
870 ipc_grow(ipc_service_t *service)
871 {
872 	ipc_slot_t *new, *old;
873 	int i, oldsize, newsize;
874 
875 	ASSERT(MUTEX_HELD(&service->ipcs_lock));
876 	ASSERT(MUTEX_NOT_HELD(&curproc->p_lock));
877 
878 	if (service->ipcs_tabsz == IPC_IDS_MAX)
879 		return (ENOSPC);
880 
881 	oldsize = service->ipcs_tabsz;
882 	newsize = oldsize << 1;
883 	new = kmem_zalloc(newsize * sizeof (ipc_slot_t), KM_NOSLEEP);
884 	if (new == NULL)
885 		return (ENOSPC);
886 
887 	old = service->ipcs_table;
888 	for (i = 0; i < oldsize; i++) {
889 		mutex_enter(&old[i].ipct_lock);
890 		mutex_enter(&new[i].ipct_lock);
891 
892 		new[i].ipct_seq = old[i].ipct_seq;
893 		new[i].ipct_data = old[i].ipct_data;
894 		old[i].ipct_data = NULL;
895 	}
896 
897 	new[0].ipct_chain = old;
898 	service->ipcs_table = new;
899 	membar_producer();
900 	service->ipcs_tabsz = newsize;
901 
902 	for (i = 0; i < oldsize; i++) {
903 		mutex_exit(&old[i].ipct_lock);
904 		mutex_exit(&new[i].ipct_lock);
905 	}
906 
907 	id_space_extend(service->ipcs_ids, oldsize, service->ipcs_tabsz);
908 
909 	return (0);
910 }
911 
912 
913 static int
914 ipc_keylookup(ipc_service_t *service, key_t key, int flag, kipc_perm_t **permp)
915 {
916 	kipc_perm_t *perm = NULL;
917 	avl_index_t where;
918 	kipc_perm_t template;
919 
920 	ASSERT(MUTEX_HELD(&service->ipcs_lock));
921 
922 	template.ipc_key = key;
923 	template.ipc_zoneid = getzoneid();
924 	if (perm = avl_find(&service->ipcs_keys, &template, &where)) {
925 		ASSERT(!IPC_FREE(perm));
926 		if ((flag & (IPC_CREAT | IPC_EXCL)) == (IPC_CREAT | IPC_EXCL))
927 			return (EEXIST);
928 		if ((flag & 0777) & ~perm->ipc_mode) {
929 			if (AU_AUDITING())
930 				audit_ipcget(NULL, (void *)perm);
931 			return (EACCES);
932 		}
933 		*permp = perm;
934 		return (0);
935 	} else if (flag & IPC_CREAT) {
936 		*permp = NULL;
937 		return (0);
938 	}
939 	return (ENOENT);
940 }
941 
942 static int
943 ipc_alloc_test(ipc_service_t *service, proc_t *pp)
944 {
945 	ASSERT(MUTEX_HELD(&service->ipcs_lock));
946 
947 	/*
948 	 * Resizing the table first would result in a cleaner code
949 	 * path, but would also allow a user to (permanently) double
950 	 * the id table size in cases where the allocation would be
951 	 * denied.  Hence we test the rctl first.
952 	 */
953 retry:
954 	mutex_enter(&pp->p_lock);
955 	if ((rctl_test(service->ipcs_proj_rctl, pp->p_task->tk_proj->kpj_rctls,
956 	    pp, 1, RCA_SAFE) & RCT_DENY) ||
957 	    (rctl_test(service->ipcs_zone_rctl, pp->p_zone->zone_rctls,
958 	    pp, 1, RCA_SAFE) & RCT_DENY)) {
959 		mutex_exit(&pp->p_lock);
960 		return (ENOSPC);
961 	}
962 
963 	if (service->ipcs_count == service->ipcs_tabsz) {
964 		int error;
965 
966 		mutex_exit(&pp->p_lock);
967 		if (error = ipc_grow(service))
968 			return (error);
969 		goto retry;
970 	}
971 
972 	return (0);
973 }
974 
975 /*
976  * Given a key, search for or create the associated identifier.
977  *
978  * If IPC_CREAT is specified and the key isn't found, or if the key is
979  * equal to IPC_PRIVATE, we return 0 and place a pointer to a newly
980  * allocated object structure in permp.  A pointer to the held service
981  * lock is placed in lockp.  ipc_mode's IPC_ALLOC bit is clear.
982  *
983  * If the key is found and no error conditions arise, we return 0 and
984  * place a pointer to the existing object structure in permp.  A
985  * pointer to the held ID lock is placed in lockp.  ipc_mode's
986  * IPC_ALLOC bit is set.
987  *
988  * Otherwise, a non-zero errno value is returned.
989  */
990 int
991 ipc_get(ipc_service_t *service, key_t key, int flag, kipc_perm_t **permp,
992     kmutex_t **lockp)
993 {
994 	kipc_perm_t	*perm = NULL;
995 	proc_t		*pp = curproc;
996 	int		error, index;
997 	cred_t		*cr = CRED();
998 
999 	if (key != IPC_PRIVATE) {
1000 
1001 		mutex_enter(&service->ipcs_lock);
1002 		error = ipc_keylookup(service, key, flag, &perm);
1003 		if (perm != NULL)
1004 			index = ipc_lock_internal(service, perm->ipc_id);
1005 		mutex_exit(&service->ipcs_lock);
1006 
1007 		if (error) {
1008 			ASSERT(perm == NULL);
1009 			return (error);
1010 		}
1011 
1012 		if (perm) {
1013 			ASSERT(!IPC_FREE(perm));
1014 			*permp = perm;
1015 			*lockp = &service->ipcs_table[index].ipct_lock;
1016 			return (0);
1017 		}
1018 
1019 		/* Key not found; fall through */
1020 	}
1021 
1022 	perm = kmem_zalloc(service->ipcs_ssize, KM_SLEEP);
1023 
1024 	mutex_enter(&service->ipcs_lock);
1025 	if (error = ipc_alloc_test(service, pp)) {
1026 		mutex_exit(&service->ipcs_lock);
1027 		kmem_free(perm, service->ipcs_ssize);
1028 		return (error);
1029 	}
1030 
1031 	perm->ipc_cuid = perm->ipc_uid = crgetuid(cr);
1032 	perm->ipc_cgid = perm->ipc_gid = crgetgid(cr);
1033 	perm->ipc_zoneid = getzoneid();
1034 	perm->ipc_mode = flag & 0777;
1035 	perm->ipc_key = key;
1036 	perm->ipc_ref = 1;
1037 	perm->ipc_id = IPC_ID_INVAL;
1038 	*permp = perm;
1039 	*lockp = &service->ipcs_lock;
1040 
1041 	return (0);
1042 }
1043 
1044 /*
1045  * Attempts to add the a newly created ID to the global namespace.  If
1046  * creating it would cause an error, we return the error.  If there is
1047  * the possibility that we could obtain the existing ID and return it
1048  * to the user, we return EAGAIN.  Otherwise, we return 0 with p_lock
1049  * and the service lock held.
1050  *
1051  * Since this should be only called after all initialization has been
1052  * completed, on failure we automatically invoke the destructor for the
1053  * object and deallocate the memory associated with it.
1054  */
1055 int
1056 ipc_commit_begin(ipc_service_t *service, key_t key, int flag,
1057     kipc_perm_t *newperm)
1058 {
1059 	kipc_perm_t *perm;
1060 	int error;
1061 	proc_t *pp = curproc;
1062 
1063 	ASSERT(newperm->ipc_ref == 1);
1064 	ASSERT(IPC_FREE(newperm));
1065 
1066 	/*
1067 	 * Set ipc_proj and ipc_zone so that future calls to ipc_cleanup()
1068 	 * clean up the necessary state.  This must be done before the
1069 	 * potential call to ipcs_dtor() below.
1070 	 */
1071 	newperm->ipc_proj = pp->p_task->tk_proj;
1072 	newperm->ipc_zone = pp->p_zone;
1073 
1074 	mutex_enter(&service->ipcs_lock);
1075 	/*
1076 	 * Ensure that no-one has raced with us and created the key.
1077 	 */
1078 	if ((key != IPC_PRIVATE) &&
1079 	    (((error = ipc_keylookup(service, key, flag, &perm)) != 0) ||
1080 	    (perm != NULL))) {
1081 		error = error ? error : EAGAIN;
1082 		goto errout;
1083 	}
1084 
1085 	/*
1086 	 * Ensure that no-one has raced with us and used the last of
1087 	 * the permissible ids, or the last of the free spaces in the
1088 	 * id table.
1089 	 */
1090 	if (error = ipc_alloc_test(service, pp))
1091 		goto errout;
1092 
1093 	ASSERT(MUTEX_HELD(&service->ipcs_lock));
1094 	ASSERT(MUTEX_HELD(&pp->p_lock));
1095 
1096 	return (0);
1097 errout:
1098 	mutex_exit(&service->ipcs_lock);
1099 	service->ipcs_dtor(newperm);
1100 	kmem_free(newperm, service->ipcs_ssize);
1101 	return (error);
1102 }
1103 
1104 /*
1105  * Commit the ID allocation transaction.  Called with p_lock and the
1106  * service lock held, both of which are dropped.  Returns the held ID
1107  * lock so the caller can extract the ID and perform ipcget auditing.
1108  */
1109 kmutex_t *
1110 ipc_commit_end(ipc_service_t *service, kipc_perm_t *perm)
1111 {
1112 	ipc_slot_t *slot;
1113 	avl_index_t where;
1114 	int index;
1115 	void *loc;
1116 
1117 	ASSERT(MUTEX_HELD(&service->ipcs_lock));
1118 	ASSERT(MUTEX_HELD(&curproc->p_lock));
1119 
1120 	(void) project_hold(perm->ipc_proj);
1121 	(void) zone_hold(perm->ipc_zone);
1122 	mutex_exit(&curproc->p_lock);
1123 
1124 	/*
1125 	 * Pick out our slot.
1126 	 */
1127 	service->ipcs_count++;
1128 	index = id_alloc(service->ipcs_ids);
1129 	ASSERT(index < service->ipcs_tabsz);
1130 	slot = &service->ipcs_table[index];
1131 	mutex_enter(&slot->ipct_lock);
1132 	ASSERT(slot->ipct_data == NULL);
1133 
1134 	/*
1135 	 * Update the perm structure.
1136 	 */
1137 	perm->ipc_mode |= IPC_ALLOC;
1138 	perm->ipc_id = (slot->ipct_seq << IPC_SEQ_SHIFT) | index;
1139 
1140 	/*
1141 	 * Push into global visibility.
1142 	 */
1143 	slot->ipct_data = perm;
1144 	if (perm->ipc_key != IPC_PRIVATE) {
1145 		loc = avl_find(&service->ipcs_keys, perm, &where);
1146 		ASSERT(loc == NULL);
1147 		avl_insert(&service->ipcs_keys, perm, where);
1148 	}
1149 	list_insert_head(&service->ipcs_usedids, perm);
1150 
1151 	/*
1152 	 * Update resource consumption.
1153 	 */
1154 	IPC_PROJ_USAGE(perm, service) += 1;
1155 	IPC_ZONE_USAGE(perm, service) += 1;
1156 
1157 	mutex_exit(&service->ipcs_lock);
1158 	return (&slot->ipct_lock);
1159 }
1160 
1161 /*
1162  * Clean up function, in case the allocation fails.  If called between
1163  * ipc_lookup and ipc_commit_begin, perm->ipc_proj will be 0 and we
1164  * merely free the perm structure.  If called after ipc_commit_begin,
1165  * we also drop locks and call the ID's destructor.
1166  */
1167 void
1168 ipc_cleanup(ipc_service_t *service, kipc_perm_t *perm)
1169 {
1170 	ASSERT(IPC_FREE(perm));
1171 	if (perm->ipc_proj) {
1172 		mutex_exit(&curproc->p_lock);
1173 		mutex_exit(&service->ipcs_lock);
1174 		service->ipcs_dtor(perm);
1175 	}
1176 	kmem_free(perm, service->ipcs_ssize);
1177 }
1178 
1179 
1180 /*
1181  * Common code to remove an IPC object.  This should be called after
1182  * all permissions checks have been performed, and with the service
1183  * and ID locked.  Note that this does not remove the object from
1184  * the ipcs_usedids list (this needs to be done by the caller before
1185  * dropping the service lock).
1186  */
1187 static void
1188 ipc_remove(ipc_service_t *service, kipc_perm_t *perm)
1189 {
1190 	int id = perm->ipc_id;
1191 	int index;
1192 
1193 	ASSERT(MUTEX_HELD(&service->ipcs_lock));
1194 	ASSERT(IPC_LOCKED(service, perm));
1195 
1196 	index = IPC_INDEX(id);
1197 
1198 	service->ipcs_table[index].ipct_data = NULL;
1199 
1200 	if (perm->ipc_key != IPC_PRIVATE)
1201 		avl_remove(&service->ipcs_keys, perm);
1202 	list_remove(&service->ipcs_usedids, perm);
1203 	perm->ipc_mode &= ~IPC_ALLOC;
1204 
1205 	id_free(service->ipcs_ids, index);
1206 
1207 	if (service->ipcs_table[index].ipct_seq++ == IPC_SEQ_MASK)
1208 		service->ipcs_table[index].ipct_seq = 0;
1209 	service->ipcs_count--;
1210 	ASSERT(IPC_PROJ_USAGE(perm, service) > 0);
1211 	ASSERT(IPC_ZONE_USAGE(perm, service) > 0);
1212 	IPC_PROJ_USAGE(perm, service) -= 1;
1213 	IPC_ZONE_USAGE(perm, service) -= 1;
1214 	ASSERT(service->ipcs_count || ((IPC_PROJ_USAGE(perm, service) == 0) &&
1215 	    (IPC_ZONE_USAGE(perm, service) == 0)));
1216 }
1217 
1218 
1219 /*
1220  * Common code to perform an IPC_RMID.  Returns an errno value on
1221  * failure, 0 on success.
1222  */
1223 int
1224 ipc_rmid(ipc_service_t *service, int id, cred_t *cr)
1225 {
1226 	kipc_perm_t *perm;
1227 	kmutex_t *lock;
1228 
1229 	mutex_enter(&service->ipcs_lock);
1230 
1231 	lock = ipc_lookup(service, id, &perm);
1232 	if (lock == NULL) {
1233 		mutex_exit(&service->ipcs_lock);
1234 		return (EINVAL);
1235 	}
1236 
1237 	ASSERT(service->ipcs_count > 0);
1238 
1239 	if (secpolicy_ipc_owner(cr, perm) != 0) {
1240 		mutex_exit(lock);
1241 		mutex_exit(&service->ipcs_lock);
1242 		return (EPERM);
1243 	}
1244 
1245 	/*
1246 	 * Nothing can fail from this point on.
1247 	 */
1248 	ipc_remove(service, perm);
1249 	mutex_exit(&service->ipcs_lock);
1250 
1251 	/* perform any per-service removal actions */
1252 	service->ipcs_rmid(perm);
1253 
1254 	ipc_rele(service, perm);
1255 
1256 	return (0);
1257 }
1258 
1259 /*
1260  * Implementation for shmids, semids, and msgids.  buf is the address
1261  * of the user buffer, nids is the size, and pnids is a pointer to
1262  * where we write the actual number of ids that [would] have been
1263  * copied out.
1264  */
1265 int
1266 ipc_ids(ipc_service_t *service, int *buf, uint_t nids, uint_t *pnids)
1267 {
1268 	kipc_perm_t *perm;
1269 	size_t	idsize = 0;
1270 	int	error = 0;
1271 	int	idcount;
1272 	int	*ids;
1273 	int	numids = 0;
1274 	zoneid_t zoneid = getzoneid();
1275 	int	global = INGLOBALZONE(curproc);
1276 
1277 	if (buf == NULL)
1278 		nids = 0;
1279 
1280 	/*
1281 	 * Get an accurate count of the total number of ids, and allocate a
1282 	 * staging buffer.  Since ipcs_count is always sane, we don't have
1283 	 * to take ipcs_lock for our first guess.  If there are no ids, or
1284 	 * we're in the global zone and the number of ids is greater than
1285 	 * the size of the specified buffer, we shunt to the end.  Otherwise,
1286 	 * we go through the id list looking for (and counting) what is
1287 	 * visible in the specified zone.
1288 	 */
1289 	idcount = service->ipcs_count;
1290 	for (;;) {
1291 		if ((global && idcount > nids) || idcount == 0) {
1292 			numids = idcount;
1293 			nids = 0;
1294 			goto out;
1295 		}
1296 
1297 		idsize = idcount * sizeof (int);
1298 		ids = kmem_alloc(idsize, KM_SLEEP);
1299 
1300 		mutex_enter(&service->ipcs_lock);
1301 		if (idcount >= service->ipcs_count)
1302 			break;
1303 		idcount = service->ipcs_count;
1304 		mutex_exit(&service->ipcs_lock);
1305 
1306 		if (idsize != 0) {
1307 			kmem_free(ids, idsize);
1308 			idsize = 0;
1309 		}
1310 	}
1311 
1312 	for (perm = list_head(&service->ipcs_usedids); perm != NULL;
1313 	    perm = list_next(&service->ipcs_usedids, perm)) {
1314 		ASSERT(!IPC_FREE(perm));
1315 		if (global || perm->ipc_zoneid == zoneid)
1316 			ids[numids++] = perm->ipc_id;
1317 	}
1318 	mutex_exit(&service->ipcs_lock);
1319 
1320 	/*
1321 	 * If there isn't enough space to hold all of the ids, just
1322 	 * return the number of ids without copying out any of them.
1323 	 */
1324 	if (nids < numids)
1325 		nids = 0;
1326 
1327 out:
1328 	if (suword32(pnids, (uint32_t)numids) ||
1329 	    (nids != 0 && copyout(ids, buf, numids * sizeof (int))))
1330 		error = EFAULT;
1331 	if (idsize != 0)
1332 		kmem_free(ids, idsize);
1333 	return (error);
1334 }
1335 
1336 /*
1337  * Destroy IPC objects from the given service that are associated with
1338  * the given zone.
1339  *
1340  * We can't hold on to the service lock when freeing objects, so we
1341  * first search the service and move all the objects to a private
1342  * list, then walk through and free them after dropping the lock.
1343  */
1344 void
1345 ipc_remove_zone(ipc_service_t *service, zoneid_t zoneid)
1346 {
1347 	kipc_perm_t *perm, *next;
1348 	list_t rmlist;
1349 	kmutex_t *lock;
1350 
1351 	list_create(&rmlist, sizeof (kipc_perm_t),
1352 	    offsetof(kipc_perm_t, ipc_list));
1353 
1354 	mutex_enter(&service->ipcs_lock);
1355 	for (perm = list_head(&service->ipcs_usedids); perm != NULL;
1356 	    perm = next) {
1357 		next = list_next(&service->ipcs_usedids, perm);
1358 		if (perm->ipc_zoneid != zoneid)
1359 			continue;
1360 
1361 		/*
1362 		 * Remove the object from the service, then put it on
1363 		 * the removal list so we can defer the call to
1364 		 * ipc_rele (which will actually free the structure).
1365 		 * We need to do this since the destructor may grab
1366 		 * the service lock.
1367 		 */
1368 		ASSERT(!IPC_FREE(perm));
1369 		lock = ipc_lock(service, perm->ipc_id);
1370 		ipc_remove(service, perm);
1371 		mutex_exit(lock);
1372 		list_insert_tail(&rmlist, perm);
1373 	}
1374 	mutex_exit(&service->ipcs_lock);
1375 
1376 	/*
1377 	 * Now that we've dropped the service lock, loop through the
1378 	 * private list freeing removed objects.
1379 	 */
1380 	for (perm = list_head(&rmlist); perm != NULL; perm = next) {
1381 		next = list_next(&rmlist, perm);
1382 		list_remove(&rmlist, perm);
1383 
1384 		(void) ipc_lock(service, perm->ipc_id);
1385 
1386 		/* perform any per-service removal actions */
1387 		service->ipcs_rmid(perm);
1388 
1389 		/* release reference */
1390 		ipc_rele(service, perm);
1391 	}
1392 
1393 	list_destroy(&rmlist);
1394 }
1395