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 2008 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
24 */
25
26 /*
27 * Contracts
28 * ---------
29 *
30 * Contracts are a primitive which enrich the relationships between
31 * processes and system resources. The primary purpose of contracts is
32 * to provide a means for the system to negotiate the departure from a
33 * binding relationship (e.g. pages locked in memory or a thread bound
34 * to processor), but they can also be used as a purely asynchronous
35 * error reporting mechanism as they are with process contracts.
36 *
37 * More information on how one interfaces with contracts and what
38 * contracts can do for you can be found in:
39 * PSARC 2003/193 Solaris Contracts
40 * PSARC 2004/460 Contracts addendum
41 *
42 * This file contains the core contracts framework. By itself it is
43 * useless: it depends the contracts filesystem (ctfs) to provide an
44 * interface to user processes and individual contract types to
45 * implement the process/resource relationships.
46 *
47 * Data structure overview
48 * -----------------------
49 *
50 * A contract is represented by a contract_t, which itself points to an
51 * encapsulating contract-type specific contract object. A contract_t
52 * contains the contract's static identity (including its terms), its
53 * linkage to various bookkeeping structures, the contract-specific
54 * event queue, and a reference count.
55 *
56 * A contract template is represented by a ct_template_t, which, like a
57 * contract, points to an encapsulating contract-type specific template
58 * object. A ct_template_t contains the template's terms.
59 *
60 * An event queue is represented by a ct_equeue_t, and consists of a
61 * list of events, a list of listeners, and a list of listeners who are
62 * waiting for new events (affectionately referred to as "tail
63 * listeners"). There are three queue types, defined by ct_listnum_t
64 * (an enum). An event may be on one of each type of queue
65 * simultaneously; the list linkage used by a queue is determined by
66 * its type.
67 *
68 * An event is represented by a ct_kevent_t, which contains mostly
69 * static event data (e.g. id, payload). It also has an array of
70 * ct_member_t structures, each of which contains a list_node_t and
71 * represent the event's linkage in a specific event queue.
72 *
73 * Each open of an event endpoint results in the creation of a new
74 * listener, represented by a ct_listener_t. In addition to linkage
75 * into the aforementioned lists in the event_queue, a ct_listener_t
76 * contains a pointer to the ct_kevent_t it is currently positioned at
77 * as well as a set of status flags and other administrative data.
78 *
79 * Each process has a list of contracts it owns, p_ct_held; a pointer
80 * to the process contract it is a member of, p_ct_process; the linkage
81 * for that membership, p_ct_member; and an array of event queue
82 * structures representing the process bundle queues.
83 *
84 * Each LWP has an array of its active templates, lwp_ct_active; and
85 * the most recently created contracts, lwp_ct_latest.
86 *
87 * A process contract has a list of member processes and a list of
88 * inherited contracts.
89 *
90 * There is a system-wide list of all contracts, as well as per-type
91 * lists of contracts.
92 *
93 * Lock ordering overview
94 * ----------------------
95 *
96 * Locks at the top are taken first:
97 *
98 * ct_evtlock
99 * regent ct_lock
100 * member ct_lock
101 * pidlock
102 * p_lock
103 * contract ctq_lock contract_lock
104 * pbundle ctq_lock
105 * cte_lock
106 * ct_reflock
107 *
108 * contract_lock and ctq_lock/cte_lock are not currently taken at the
109 * same time.
110 *
111 * Reference counting and locking
112 * ------------------------------
113 *
114 * A contract has a reference count, protected by ct_reflock.
115 * (ct_reflock is also used in a couple other places where atomic
116 * access to a variable is needed in an innermost context). A process
117 * maintains a hold on each contract it owns. A process contract has a
118 * hold on each contract is has inherited. Each event has a hold on
119 * the contract which generated it. Process contract templates have
120 * holds on the contracts referred to by their transfer terms. CTFS
121 * contract directory nodes have holds on contracts. Lastly, various
122 * code paths may temporarily take holds on contracts to prevent them
123 * from disappearing while other processing is going on. It is
124 * important to note that the global contract lists do not hold
125 * references on contracts; a contract is removed from these structures
126 * atomically with the release of its last reference.
127 *
128 * At a given point in time, a contract can either be owned by a
129 * process, inherited by a regent process contract, or orphaned. A
130 * contract_t's owner and regent pointers, ct_owner and ct_regent, are
131 * protected by its ct_lock. The linkage in the holder's (holder =
132 * owner or regent) list of contracts, ct_ctlist, is protected by
133 * whatever lock protects the holder's data structure. In order for
134 * these two directions to remain consistent, changing the holder of a
135 * contract requires that both locks be held.
136 *
137 * Events also have reference counts. There is one hold on an event
138 * per queue it is present on, in addition to those needed for the
139 * usual sundry reasons. Individual listeners are associated with
140 * specific queues, and increase a queue-specific reference count
141 * stored in the ct_member_t structure.
142 *
143 * The dynamic contents of an event (reference count and flags) are
144 * protected by its cte_lock, while the contents of the embedded
145 * ct_member_t structures are protected by the locks of the queues they
146 * are linked into. A ct_listener_t's contents are also protected by
147 * its event queue's ctq_lock.
148 *
149 * Resource controls
150 * -----------------
151 *
152 * Control: project.max-contracts (rc_project_contract)
153 * Description: Maximum number of contracts allowed a project.
154 *
155 * When a contract is created, the project's allocation is tested and
156 * (assuming success) increased. When the last reference to a
157 * contract is released, the creating project's allocation is
158 * decreased.
159 */
160
161 #include <sys/mutex.h>
162 #include <sys/debug.h>
163 #include <sys/types.h>
164 #include <sys/param.h>
165 #include <sys/kmem.h>
166 #include <sys/thread.h>
167 #include <sys/id_space.h>
168 #include <sys/avl.h>
169 #include <sys/list.h>
170 #include <sys/sysmacros.h>
171 #include <sys/proc.h>
172 #include <sys/ctfs.h>
173 #include <sys/contract_impl.h>
174 #include <sys/contract/process_impl.h>
175 #include <sys/dditypes.h>
176 #include <sys/contract/device_impl.h>
177 #include <sys/systm.h>
178 #include <sys/atomic.h>
179 #include <sys/cmn_err.h>
180 #include <sys/model.h>
181 #include <sys/policy.h>
182 #include <sys/zone.h>
183 #include <sys/task.h>
184 #include <sys/ddi.h>
185 #include <sys/sunddi.h>
186
187 extern rctl_hndl_t rc_project_contract;
188
189 static id_space_t *contract_ids;
190 static avl_tree_t contract_avl;
191 static kmutex_t contract_lock;
192
193 int ct_ntypes = CTT_MAXTYPE;
194 static ct_type_t *ct_types_static[CTT_MAXTYPE];
195 ct_type_t **ct_types = ct_types_static;
196 int ct_debug;
197
198 static void cte_queue_create(ct_equeue_t *, ct_listnum_t, int, int);
199 static void cte_queue_destroy(ct_equeue_t *);
200 static void cte_queue_drain(ct_equeue_t *, int);
201 static void cte_trim(ct_equeue_t *, contract_t *);
202 static void cte_copy(ct_equeue_t *, ct_equeue_t *);
203
204 /*
205 * contract_compar
206 *
207 * A contract comparator which sorts on contract ID.
208 */
209 int
210 contract_compar(const void *x, const void *y)
211 {
212 const contract_t *ct1 = x;
213 const contract_t *ct2 = y;
214
215 if (ct1->ct_id < ct2->ct_id)
216 return (-1);
217 if (ct1->ct_id > ct2->ct_id)
218 return (1);
219 return (0);
220 }
221
222 /*
223 * contract_init
224 *
225 * Initializes the contract subsystem, the specific contract types, and
226 * process 0.
227 */
228 void
229 contract_init(void)
230 {
231 /*
232 * Initialize contract subsystem.
233 */
234 contract_ids = id_space_create("contracts", 1, INT_MAX);
235 avl_create(&contract_avl, contract_compar, sizeof (contract_t),
236 offsetof(contract_t, ct_ctavl));
237 mutex_init(&contract_lock, NULL, MUTEX_DEFAULT, NULL);
238
239 /*
240 * Initialize contract types.
241 */
242 contract_process_init();
243 contract_device_init();
244
245 /*
246 * Initialize p0/lwp0 contract state.
247 */
248 avl_create(&p0.p_ct_held, contract_compar, sizeof (contract_t),
249 offsetof(contract_t, ct_ctlist));
250 }
251
252 /*
253 * contract_dtor
254 *
255 * Performs basic destruction of the common portions of a contract.
256 * Called from the failure path of contract_ctor and from
257 * contract_rele.
258 */
259 static void
260 contract_dtor(contract_t *ct)
261 {
262 cte_queue_destroy(&ct->ct_events);
263 list_destroy(&ct->ct_vnodes);
264 mutex_destroy(&ct->ct_reflock);
265 mutex_destroy(&ct->ct_lock);
266 mutex_destroy(&ct->ct_evtlock);
267 }
268
269 /*
270 * contract_ctor
271 *
272 * Called by a contract type to initialize a contract. Fails if the
273 * max-contract resource control would have been exceeded. After a
274 * successful call to contract_ctor, the contract is unlocked and
275 * visible in all namespaces; any type-specific initialization should
276 * be completed before calling contract_ctor. Returns 0 on success.
277 *
278 * Because not all callers can tolerate failure, a 0 value for canfail
279 * instructs contract_ctor to ignore the project.max-contracts resource
280 * control. Obviously, this "out" should only be employed by callers
281 * who are sufficiently constrained in other ways (e.g. newproc).
282 */
283 int
284 contract_ctor(contract_t *ct, ct_type_t *type, ct_template_t *tmpl, void *data,
285 ctflags_t flags, proc_t *author, int canfail)
286 {
287 avl_index_t where;
288 klwp_t *curlwp = ttolwp(curthread);
289
290 ASSERT(author == curproc);
291
292 mutex_init(&ct->ct_lock, NULL, MUTEX_DEFAULT, NULL);
293 mutex_init(&ct->ct_reflock, NULL, MUTEX_DEFAULT, NULL);
294 mutex_init(&ct->ct_evtlock, NULL, MUTEX_DEFAULT, NULL);
295 ct->ct_id = id_alloc(contract_ids);
296
297 cte_queue_create(&ct->ct_events, CTEL_CONTRACT, 20, 0);
298 list_create(&ct->ct_vnodes, sizeof (contract_vnode_t),
299 offsetof(contract_vnode_t, ctv_node));
300
301 /*
302 * Instance data
303 */
304 ct->ct_ref = 2; /* one for the holder, one for "latest" */
305 ct->ct_cuid = crgetuid(CRED());
306 ct->ct_type = type;
307 ct->ct_data = data;
308 gethrestime(&ct->ct_ctime);
309 ct->ct_state = CTS_OWNED;
310 ct->ct_flags = flags;
311 ct->ct_regent = author->p_ct_process ?
312 &author->p_ct_process->conp_contract : NULL;
313 ct->ct_ev_info = tmpl->ctmpl_ev_info;
314 ct->ct_ev_crit = tmpl->ctmpl_ev_crit;
315 ct->ct_cookie = tmpl->ctmpl_cookie;
316 ct->ct_owner = author;
317 ct->ct_ntime.ctm_total = -1;
318 ct->ct_qtime.ctm_total = -1;
319 ct->ct_nevent = NULL;
320
321 /*
322 * Test project.max-contracts.
323 */
324 mutex_enter(&author->p_lock);
325 mutex_enter(&contract_lock);
326 if (canfail && rctl_test(rc_project_contract,
327 author->p_task->tk_proj->kpj_rctls, author, 1,
328 RCA_SAFE) & RCT_DENY) {
329 id_free(contract_ids, ct->ct_id);
330 mutex_exit(&contract_lock);
331 mutex_exit(&author->p_lock);
332 ct->ct_events.ctq_flags |= CTQ_DEAD;
333 contract_dtor(ct);
334 return (1);
335 }
336 ct->ct_proj = author->p_task->tk_proj;
337 ct->ct_proj->kpj_data.kpd_contract++;
338 (void) project_hold(ct->ct_proj);
339 mutex_exit(&contract_lock);
340
341 /*
342 * Insert into holder's avl of contracts.
343 * We use an avl not because order is important, but because
344 * readdir of /proc/contracts requires we be able to use a
345 * scalar as an index into the process's list of contracts
346 */
347 ct->ct_zoneid = author->p_zone->zone_id;
348 ct->ct_czuniqid = ct->ct_mzuniqid = author->p_zone->zone_uniqid;
349 VERIFY(avl_find(&author->p_ct_held, ct, &where) == NULL);
350 avl_insert(&author->p_ct_held, ct, where);
351 mutex_exit(&author->p_lock);
352
353 /*
354 * Insert into global contract AVL
355 */
356 mutex_enter(&contract_lock);
357 VERIFY(avl_find(&contract_avl, ct, &where) == NULL);
358 avl_insert(&contract_avl, ct, where);
359 mutex_exit(&contract_lock);
360
361 /*
362 * Insert into type AVL
363 */
364 mutex_enter(&type->ct_type_lock);
365 VERIFY(avl_find(&type->ct_type_avl, ct, &where) == NULL);
366 avl_insert(&type->ct_type_avl, ct, where);
367 type->ct_type_timestruc = ct->ct_ctime;
368 mutex_exit(&type->ct_type_lock);
369
370 if (curlwp->lwp_ct_latest[type->ct_type_index])
371 contract_rele(curlwp->lwp_ct_latest[type->ct_type_index]);
372 curlwp->lwp_ct_latest[type->ct_type_index] = ct;
373
374 return (0);
375 }
376
377 /*
378 * contract_rele
379 *
380 * Releases a reference to a contract. If the caller had the last
381 * reference, the contract is removed from all namespaces, its
382 * allocation against the max-contracts resource control is released,
383 * and the contract type's free entry point is invoked for any
384 * type-specific deconstruction and to (presumably) free the object.
385 */
386 void
387 contract_rele(contract_t *ct)
388 {
389 uint64_t nref;
390
391 mutex_enter(&ct->ct_reflock);
392 ASSERT(ct->ct_ref > 0);
393 nref = --ct->ct_ref;
394 mutex_exit(&ct->ct_reflock);
395 if (nref == 0) {
396 /*
397 * ct_owner is cleared when it drops its reference.
398 */
399 ASSERT(ct->ct_owner == NULL);
400 ASSERT(ct->ct_evcnt == 0);
401
402 /*
403 * Remove from global contract AVL
404 */
405 mutex_enter(&contract_lock);
406 avl_remove(&contract_avl, ct);
407 mutex_exit(&contract_lock);
408
409 /*
410 * Remove from type AVL
411 */
412 mutex_enter(&ct->ct_type->ct_type_lock);
413 avl_remove(&ct->ct_type->ct_type_avl, ct);
414 mutex_exit(&ct->ct_type->ct_type_lock);
415
416 /*
417 * Release the contract's ID
418 */
419 id_free(contract_ids, ct->ct_id);
420
421 /*
422 * Release project hold
423 */
424 mutex_enter(&contract_lock);
425 ct->ct_proj->kpj_data.kpd_contract--;
426 project_rele(ct->ct_proj);
427 mutex_exit(&contract_lock);
428
429 /*
430 * Free the contract
431 */
432 contract_dtor(ct);
433 ct->ct_type->ct_type_ops->contop_free(ct);
434 }
435 }
436
437 /*
438 * contract_hold
439 *
440 * Adds a reference to a contract
441 */
442 void
443 contract_hold(contract_t *ct)
444 {
445 mutex_enter(&ct->ct_reflock);
446 ASSERT(ct->ct_ref < UINT64_MAX);
447 ct->ct_ref++;
448 mutex_exit(&ct->ct_reflock);
449 }
450
451 /*
452 * contract_getzuniqid
453 *
454 * Get a contract's zone unique ID. Needed because 64-bit reads and
455 * writes aren't atomic on x86. Since there are contexts where we are
456 * unable to take ct_lock, we instead use ct_reflock; in actuality any
457 * lock would do.
458 */
459 uint64_t
460 contract_getzuniqid(contract_t *ct)
461 {
462 uint64_t zuniqid;
463
464 mutex_enter(&ct->ct_reflock);
465 zuniqid = ct->ct_mzuniqid;
466 mutex_exit(&ct->ct_reflock);
467
468 return (zuniqid);
469 }
470
471 /*
472 * contract_setzuniqid
473 *
474 * Sets a contract's zone unique ID. See contract_getzuniqid.
475 */
476 void
477 contract_setzuniqid(contract_t *ct, uint64_t zuniqid)
478 {
479 mutex_enter(&ct->ct_reflock);
480 ct->ct_mzuniqid = zuniqid;
481 mutex_exit(&ct->ct_reflock);
482 }
483
484 /*
485 * contract_abandon
486 *
487 * Abandons the specified contract. If "explicit" is clear, the
488 * contract was implicitly abandoned (by process exit) and should be
489 * inherited if its terms allow it and its owner was a member of a
490 * regent contract. Otherwise, the contract type's abandon entry point
491 * is invoked to either destroy or orphan the contract.
492 */
493 int
494 contract_abandon(contract_t *ct, proc_t *p, int explicit)
495 {
496 ct_equeue_t *q = NULL;
497 contract_t *parent = &p->p_ct_process->conp_contract;
498 int inherit = 0;
499
500 VERIFY(p == curproc);
501
502 mutex_enter(&ct->ct_lock);
503
504 /*
505 * Multiple contract locks are taken contract -> subcontract.
506 * Check if the contract will be inherited so we can acquire
507 * all the necessary locks before making sensitive changes.
508 */
509 if (!explicit && (ct->ct_flags & CTF_INHERIT) &&
510 contract_process_accept(parent)) {
511 mutex_exit(&ct->ct_lock);
512 mutex_enter(&parent->ct_lock);
513 mutex_enter(&ct->ct_lock);
514 inherit = 1;
515 }
516
517 if (ct->ct_owner != p) {
518 mutex_exit(&ct->ct_lock);
519 if (inherit)
520 mutex_exit(&parent->ct_lock);
521 return (EINVAL);
522 }
523
524 mutex_enter(&p->p_lock);
525 if (explicit)
526 avl_remove(&p->p_ct_held, ct);
527 ct->ct_owner = NULL;
528 mutex_exit(&p->p_lock);
529
530 /*
531 * Since we can't call cte_trim with the contract lock held,
532 * we grab the queue pointer here.
533 */
534 if (p->p_ct_equeue)
535 q = p->p_ct_equeue[ct->ct_type->ct_type_index];
536
537 /*
538 * contop_abandon may destroy the contract so we rely on it to
539 * drop ct_lock. We retain a reference on the contract so that
540 * the cte_trim which follows functions properly. Even though
541 * cte_trim doesn't dereference the contract pointer, it is
542 * still necessary to retain a reference to the contract so
543 * that we don't trim events which are sent by a subsequently
544 * allocated contract infortuitously located at the same address.
545 */
546 contract_hold(ct);
547
548 if (inherit) {
549 ct->ct_state = CTS_INHERITED;
550 VERIFY(ct->ct_regent == parent);
551 contract_process_take(parent, ct);
552
553 /*
554 * We are handing off the process's reference to the
555 * parent contract. For this reason, the order in
556 * which we drop the contract locks is also important.
557 */
558 mutex_exit(&ct->ct_lock);
559 mutex_exit(&parent->ct_lock);
560 } else {
561 ct->ct_regent = NULL;
562 ct->ct_type->ct_type_ops->contop_abandon(ct);
563 }
564
565 /*
566 * ct_lock has been dropped; we can safely trim the event
567 * queue now.
568 */
569 if (q) {
570 mutex_enter(&q->ctq_lock);
571 cte_trim(q, ct);
572 mutex_exit(&q->ctq_lock);
573 }
574
575 contract_rele(ct);
576
577 return (0);
578 }
579
580 int
581 contract_newct(contract_t *ct)
582 {
583 return (ct->ct_type->ct_type_ops->contop_newct(ct));
584 }
585
586 /*
587 * contract_adopt
588 *
589 * Adopts a contract. After a successful call to this routine, the
590 * previously inherited contract will belong to the calling process,
591 * and its events will have been appended to its new owner's process
592 * bundle queue.
593 */
594 int
595 contract_adopt(contract_t *ct, proc_t *p)
596 {
597 avl_index_t where;
598 ct_equeue_t *q;
599 contract_t *parent;
600
601 ASSERT(p == curproc);
602
603 /*
604 * Ensure the process has an event queue. Checked by ASSERTs
605 * below.
606 */
607 (void) contract_type_pbundle(ct->ct_type, p);
608
609 mutex_enter(&ct->ct_lock);
610 parent = ct->ct_regent;
611 if (ct->ct_state != CTS_INHERITED ||
612 &p->p_ct_process->conp_contract != parent ||
613 p->p_zone->zone_uniqid != ct->ct_czuniqid) {
614 mutex_exit(&ct->ct_lock);
615 return (EINVAL);
616 }
617
618 /*
619 * Multiple contract locks are taken contract -> subcontract.
620 */
621 mutex_exit(&ct->ct_lock);
622 mutex_enter(&parent->ct_lock);
623 mutex_enter(&ct->ct_lock);
624
625 /*
626 * It is possible that the contract was adopted by someone else
627 * while its lock was dropped. It isn't possible for the
628 * contract to have been inherited by a different regent
629 * contract.
630 */
631 if (ct->ct_state != CTS_INHERITED) {
632 mutex_exit(&parent->ct_lock);
633 mutex_exit(&ct->ct_lock);
634 return (EBUSY);
635 }
636 ASSERT(ct->ct_regent == parent);
637
638 ct->ct_state = CTS_OWNED;
639
640 contract_process_adopt(ct, p);
641
642 mutex_enter(&p->p_lock);
643 ct->ct_owner = p;
644 VERIFY(avl_find(&p->p_ct_held, ct, &where) == NULL);
645 avl_insert(&p->p_ct_held, ct, where);
646 mutex_exit(&p->p_lock);
647
648 ASSERT(ct->ct_owner->p_ct_equeue);
649 ASSERT(ct->ct_owner->p_ct_equeue[ct->ct_type->ct_type_index]);
650 q = ct->ct_owner->p_ct_equeue[ct->ct_type->ct_type_index];
651 cte_copy(&ct->ct_events, q);
652 mutex_exit(&ct->ct_lock);
653
654 return (0);
655 }
656
657 /*
658 * contract_ack
659 *
660 * Acknowledges receipt of a critical event.
661 */
662 int
663 contract_ack(contract_t *ct, uint64_t evid, int ack)
664 {
665 ct_kevent_t *ev;
666 list_t *queue = &ct->ct_events.ctq_events;
667 int error = ESRCH;
668 int nego = 0;
669 uint_t evtype;
670
671 ASSERT(ack == CT_ACK || ack == CT_NACK);
672
673 mutex_enter(&ct->ct_lock);
674 mutex_enter(&ct->ct_events.ctq_lock);
675 /*
676 * We are probably ACKing something near the head of the queue.
677 */
678 for (ev = list_head(queue); ev; ev = list_next(queue, ev)) {
679 if (ev->cte_id == evid) {
680 if (ev->cte_flags & CTE_NEG)
681 nego = 1;
682 else if (ack == CT_NACK)
683 break;
684 if ((ev->cte_flags & (CTE_INFO | CTE_ACK)) == 0) {
685 ev->cte_flags |= CTE_ACK;
686 ct->ct_evcnt--;
687 evtype = ev->cte_type;
688 error = 0;
689 }
690 break;
691 }
692 }
693 mutex_exit(&ct->ct_events.ctq_lock);
694 mutex_exit(&ct->ct_lock);
695
696 /*
697 * Not all critical events are negotiation events, however
698 * every negotiation event is a critical event. NEGEND events
699 * are critical events but are not negotiation events
700 */
701 if (error || !nego)
702 return (error);
703
704 if (ack == CT_ACK)
705 error = ct->ct_type->ct_type_ops->contop_ack(ct, evtype, evid);
706 else
707 error = ct->ct_type->ct_type_ops->contop_nack(ct, evtype, evid);
708
709 return (error);
710 }
711
712 /*ARGSUSED*/
713 int
714 contract_ack_inval(contract_t *ct, uint_t evtype, uint64_t evid)
715 {
716 cmn_err(CE_PANIC, "contract_ack_inval: unsupported call: ctid: %u",
717 ct->ct_id);
718 return (ENOSYS);
719 }
720
721 /*ARGSUSED*/
722 int
723 contract_qack_inval(contract_t *ct, uint_t evtype, uint64_t evid)
724 {
725 cmn_err(CE_PANIC, "contract_ack_inval: unsupported call: ctid: %u",
726 ct->ct_id);
727 return (ENOSYS);
728 }
729
730 /*ARGSUSED*/
731 int
732 contract_qack_notsup(contract_t *ct, uint_t evtype, uint64_t evid)
733 {
734 return (ERANGE);
735 }
736
737 /*
738 * contract_qack
739 *
740 * Asks that negotiations be extended by another time quantum
741 */
742 int
743 contract_qack(contract_t *ct, uint64_t evid)
744 {
745 ct_kevent_t *ev;
746 list_t *queue = &ct->ct_events.ctq_events;
747 int nego = 0;
748 uint_t evtype;
749
750 mutex_enter(&ct->ct_lock);
751 mutex_enter(&ct->ct_events.ctq_lock);
752
753 for (ev = list_head(queue); ev; ev = list_next(queue, ev)) {
754 if (ev->cte_id == evid) {
755 if ((ev->cte_flags & (CTE_NEG | CTE_ACK)) == CTE_NEG) {
756 evtype = ev->cte_type;
757 nego = 1;
758 }
759 break;
760 }
761 }
762 mutex_exit(&ct->ct_events.ctq_lock);
763 mutex_exit(&ct->ct_lock);
764
765 /*
766 * Only a negotiated event (which is by definition also a critical
767 * event) which has not yet been acknowledged can provide
768 * time quanta to a negotiating owner process.
769 */
770 if (!nego)
771 return (ESRCH);
772
773 return (ct->ct_type->ct_type_ops->contop_qack(ct, evtype, evid));
774 }
775
776 /*
777 * contract_orphan
778 *
779 * Icky-poo. This is a process-contract special, used to ACK all
780 * critical messages when a contract is orphaned.
781 */
782 void
783 contract_orphan(contract_t *ct)
784 {
785 ct_kevent_t *ev;
786 list_t *queue = &ct->ct_events.ctq_events;
787
788 ASSERT(MUTEX_HELD(&ct->ct_lock));
789 ASSERT(ct->ct_state != CTS_ORPHAN);
790
791 mutex_enter(&ct->ct_events.ctq_lock);
792 ct->ct_state = CTS_ORPHAN;
793 for (ev = list_head(queue); ev; ev = list_next(queue, ev)) {
794 if ((ev->cte_flags & (CTE_INFO | CTE_ACK)) == 0) {
795 ev->cte_flags |= CTE_ACK;
796 ct->ct_evcnt--;
797 }
798 }
799 mutex_exit(&ct->ct_events.ctq_lock);
800
801 ASSERT(ct->ct_evcnt == 0);
802 }
803
804 /*
805 * contract_destroy
806 *
807 * Explicit contract destruction. Called when contract is empty.
808 * The contract will actually stick around until all of its events are
809 * removed from the bundle and and process bundle queues, and all fds
810 * which refer to it are closed. See contract_dtor if you are looking
811 * for what destroys the contract structure.
812 */
813 void
814 contract_destroy(contract_t *ct)
815 {
816 ASSERT(MUTEX_HELD(&ct->ct_lock));
817 ASSERT(ct->ct_state != CTS_DEAD);
818 ASSERT(ct->ct_owner == NULL);
819
820 ct->ct_state = CTS_DEAD;
821 cte_queue_drain(&ct->ct_events, 1);
822 mutex_exit(&ct->ct_lock);
823 mutex_enter(&ct->ct_type->ct_type_events.ctq_lock);
824 cte_trim(&ct->ct_type->ct_type_events, ct);
825 mutex_exit(&ct->ct_type->ct_type_events.ctq_lock);
826 mutex_enter(&ct->ct_lock);
827 ct->ct_type->ct_type_ops->contop_destroy(ct);
828 mutex_exit(&ct->ct_lock);
829 contract_rele(ct);
830 }
831
832 /*
833 * contract_vnode_get
834 *
835 * Obtains the contract directory vnode for this contract, if there is
836 * one. The caller must VN_RELE the vnode when they are through using
837 * it.
838 */
839 vnode_t *
840 contract_vnode_get(contract_t *ct, vfs_t *vfsp)
841 {
842 contract_vnode_t *ctv;
843 vnode_t *vp = NULL;
844
845 mutex_enter(&ct->ct_lock);
846 for (ctv = list_head(&ct->ct_vnodes); ctv != NULL;
847 ctv = list_next(&ct->ct_vnodes, ctv))
848 if (ctv->ctv_vnode->v_vfsp == vfsp) {
849 vp = ctv->ctv_vnode;
850 VN_HOLD(vp);
851 break;
852 }
853 mutex_exit(&ct->ct_lock);
854 return (vp);
855 }
856
857 /*
858 * contract_vnode_set
859 *
860 * Sets the contract directory vnode for this contract. We don't hold
861 * a reference on the vnode because we don't want to prevent it from
862 * being freed. The vnode's inactive entry point will take care of
863 * notifying us when it should be removed.
864 */
865 void
866 contract_vnode_set(contract_t *ct, contract_vnode_t *ctv, vnode_t *vnode)
867 {
868 mutex_enter(&ct->ct_lock);
869 ctv->ctv_vnode = vnode;
870 list_insert_head(&ct->ct_vnodes, ctv);
871 mutex_exit(&ct->ct_lock);
872 }
873
874 /*
875 * contract_vnode_clear
876 *
877 * Removes this vnode as the contract directory vnode for this
878 * contract. Called from a contract directory's inactive entry point,
879 * this may return 0 indicating that the vnode gained another reference
880 * because of a simultaneous call to contract_vnode_get.
881 */
882 int
883 contract_vnode_clear(contract_t *ct, contract_vnode_t *ctv)
884 {
885 vnode_t *vp = ctv->ctv_vnode;
886 int result;
887
888 mutex_enter(&ct->ct_lock);
889 mutex_enter(&vp->v_lock);
890 if (vp->v_count == 1) {
891 list_remove(&ct->ct_vnodes, ctv);
892 result = 1;
893 } else {
894 vp->v_count--;
895 result = 0;
896 }
897 mutex_exit(&vp->v_lock);
898 mutex_exit(&ct->ct_lock);
899
900 return (result);
901 }
902
903 /*
904 * contract_exit
905 *
906 * Abandons all contracts held by process p, and drains process p's
907 * bundle queues. Called on process exit.
908 */
909 void
910 contract_exit(proc_t *p)
911 {
912 contract_t *ct;
913 void *cookie = NULL;
914 int i;
915
916 ASSERT(p == curproc);
917
918 /*
919 * Abandon held contracts. contract_abandon knows enough not
920 * to remove the contract from the list a second time. We are
921 * exiting, so no locks are needed here. But because
922 * contract_abandon will take p_lock, we need to make sure we
923 * aren't holding it.
924 */
925 ASSERT(MUTEX_NOT_HELD(&p->p_lock));
926 while ((ct = avl_destroy_nodes(&p->p_ct_held, &cookie)) != NULL)
927 VERIFY(contract_abandon(ct, p, 0) == 0);
928
929 /*
930 * Drain pbundles. Because a process bundle queue could have
931 * been passed to another process, they may not be freed right
932 * away.
933 */
934 if (p->p_ct_equeue) {
935 for (i = 0; i < CTT_MAXTYPE; i++)
936 if (p->p_ct_equeue[i])
937 cte_queue_drain(p->p_ct_equeue[i], 0);
938 kmem_free(p->p_ct_equeue, CTT_MAXTYPE * sizeof (ct_equeue_t *));
939 }
940 }
941
942 static int
943 get_time_left(struct ct_time *t)
944 {
945 clock_t ticks_elapsed;
946 int secs_elapsed;
947
948 if (t->ctm_total == -1)
949 return (-1);
950
951 ticks_elapsed = ddi_get_lbolt() - t->ctm_start;
952 secs_elapsed = t->ctm_total - (drv_hztousec(ticks_elapsed)/MICROSEC);
953 return (secs_elapsed > 0 ? secs_elapsed : 0);
954 }
955
956 /*
957 * contract_status_common
958 *
959 * Populates a ct_status structure. Used by contract types in their
960 * status entry points and ctfs when only common information is
961 * requested.
962 */
963 void
964 contract_status_common(contract_t *ct, zone_t *zone, void *status,
965 model_t model)
966 {
967 STRUCT_HANDLE(ct_status, lstatus);
968
969 STRUCT_SET_HANDLE(lstatus, model, status);
970 ASSERT(MUTEX_HELD(&ct->ct_lock));
971 if (zone->zone_uniqid == GLOBAL_ZONEUNIQID ||
972 zone->zone_uniqid == ct->ct_czuniqid) {
973 zone_t *czone;
974 zoneid_t zoneid = -1;
975
976 /*
977 * Contracts don't have holds on the zones they were
978 * created by. If the contract's zone no longer
979 * exists, we say its zoneid is -1.
980 */
981 if (zone->zone_uniqid == ct->ct_czuniqid ||
982 ct->ct_czuniqid == GLOBAL_ZONEUNIQID) {
983 zoneid = ct->ct_zoneid;
984 } else if ((czone = zone_find_by_id(ct->ct_zoneid)) != NULL) {
985 if (czone->zone_uniqid == ct->ct_mzuniqid)
986 zoneid = ct->ct_zoneid;
987 zone_rele(czone);
988 }
989
990 STRUCT_FSET(lstatus, ctst_zoneid, zoneid);
991 STRUCT_FSET(lstatus, ctst_holder,
992 (ct->ct_state == CTS_OWNED) ? ct->ct_owner->p_pid :
993 (ct->ct_state == CTS_INHERITED) ? ct->ct_regent->ct_id : 0);
994 STRUCT_FSET(lstatus, ctst_state, ct->ct_state);
995 } else {
996 /*
997 * We are looking at a contract which was created by a
998 * process outside of our zone. We provide fake zone,
999 * holder, and state information.
1000 */
1001
1002 STRUCT_FSET(lstatus, ctst_zoneid, zone->zone_id);
1003 /*
1004 * Since "zone" can't disappear until the calling ctfs
1005 * is unmounted, zone_zsched must be valid.
1006 */
1007 STRUCT_FSET(lstatus, ctst_holder, (ct->ct_state < CTS_ORPHAN) ?
1008 zone->zone_zsched->p_pid : 0);
1009 STRUCT_FSET(lstatus, ctst_state, (ct->ct_state < CTS_ORPHAN) ?
1010 CTS_OWNED : ct->ct_state);
1011 }
1012 STRUCT_FSET(lstatus, ctst_nevents, ct->ct_evcnt);
1013 STRUCT_FSET(lstatus, ctst_ntime, get_time_left(&ct->ct_ntime));
1014 STRUCT_FSET(lstatus, ctst_qtime, get_time_left(&ct->ct_qtime));
1015 STRUCT_FSET(lstatus, ctst_nevid,
1016 ct->ct_nevent ? ct->ct_nevent->cte_id : 0);
1017 STRUCT_FSET(lstatus, ctst_critical, ct->ct_ev_crit);
1018 STRUCT_FSET(lstatus, ctst_informative, ct->ct_ev_info);
1019 STRUCT_FSET(lstatus, ctst_cookie, ct->ct_cookie);
1020 STRUCT_FSET(lstatus, ctst_type, ct->ct_type->ct_type_index);
1021 STRUCT_FSET(lstatus, ctst_id, ct->ct_id);
1022 }
1023
1024 /*
1025 * contract_checkcred
1026 *
1027 * Determines if the specified contract is owned by a process with the
1028 * same effective uid as the specified credential. The caller must
1029 * ensure that the uid spaces are the same. Returns 1 on success.
1030 */
1031 static int
1032 contract_checkcred(contract_t *ct, const cred_t *cr)
1033 {
1034 proc_t *p;
1035 int fail = 1;
1036
1037 mutex_enter(&ct->ct_lock);
1038 if ((p = ct->ct_owner) != NULL) {
1039 mutex_enter(&p->p_crlock);
1040 fail = crgetuid(cr) != crgetuid(p->p_cred);
1041 mutex_exit(&p->p_crlock);
1042 }
1043 mutex_exit(&ct->ct_lock);
1044
1045 return (!fail);
1046 }
1047
1048 /*
1049 * contract_owned
1050 *
1051 * Determines if the specified credential can view an event generated
1052 * by the specified contract. If locked is set, the contract's ct_lock
1053 * is held and the caller will need to do additional work to determine
1054 * if they truly can see the event. Returns 1 on success.
1055 */
1056 int
1057 contract_owned(contract_t *ct, const cred_t *cr, int locked)
1058 {
1059 int owner, cmatch, zmatch;
1060 uint64_t zuniqid, mzuniqid;
1061 uid_t euid;
1062
1063 ASSERT(locked || MUTEX_NOT_HELD(&ct->ct_lock));
1064
1065 zuniqid = curproc->p_zone->zone_uniqid;
1066 mzuniqid = contract_getzuniqid(ct);
1067 euid = crgetuid(cr);
1068
1069 /*
1070 * owner: we own the contract
1071 * cmatch: we are in the creator's (and holder's) zone and our
1072 * uid matches the creator's or holder's
1073 * zmatch: we are in the effective zone of a contract created
1074 * in the global zone, and our uid matches that of the
1075 * virtualized holder's (zsched/kcred)
1076 */
1077 owner = (ct->ct_owner == curproc);
1078 cmatch = (zuniqid == ct->ct_czuniqid) &&
1079 ((ct->ct_cuid == euid) || (!locked && contract_checkcred(ct, cr)));
1080 zmatch = (ct->ct_czuniqid != mzuniqid) && (zuniqid == mzuniqid) &&
1081 (crgetuid(kcred) == euid);
1082
1083 return (owner || cmatch || zmatch);
1084 }
1085
1086
1087 /*
1088 * contract_type_init
1089 *
1090 * Called by contract types to register themselves with the contracts
1091 * framework.
1092 */
1093 ct_type_t *
1094 contract_type_init(ct_typeid_t type, const char *name, contops_t *ops,
1095 ct_f_default_t *dfault)
1096 {
1097 ct_type_t *result;
1098
1099 ASSERT(type < CTT_MAXTYPE);
1100
1101 result = kmem_alloc(sizeof (ct_type_t), KM_SLEEP);
1102
1103 mutex_init(&result->ct_type_lock, NULL, MUTEX_DEFAULT, NULL);
1104 avl_create(&result->ct_type_avl, contract_compar, sizeof (contract_t),
1105 offsetof(contract_t, ct_cttavl));
1106 cte_queue_create(&result->ct_type_events, CTEL_BUNDLE, 20, 0);
1107 result->ct_type_name = name;
1108 result->ct_type_ops = ops;
1109 result->ct_type_default = dfault;
1110 result->ct_type_evid = 0;
1111 gethrestime(&result->ct_type_timestruc);
1112 result->ct_type_index = type;
1113
1114 ct_types[type] = result;
1115
1116 return (result);
1117 }
1118
1119 /*
1120 * contract_type_count
1121 *
1122 * Obtains the number of contracts of a particular type.
1123 */
1124 int
1125 contract_type_count(ct_type_t *type)
1126 {
1127 ulong_t count;
1128
1129 mutex_enter(&type->ct_type_lock);
1130 count = avl_numnodes(&type->ct_type_avl);
1131 mutex_exit(&type->ct_type_lock);
1132
1133 return (count);
1134 }
1135
1136 /*
1137 * contract_type_max
1138 *
1139 * Obtains the maximum contract id of of a particular type.
1140 */
1141 ctid_t
1142 contract_type_max(ct_type_t *type)
1143 {
1144 contract_t *ct;
1145 ctid_t res;
1146
1147 mutex_enter(&type->ct_type_lock);
1148 ct = avl_last(&type->ct_type_avl);
1149 res = ct ? ct->ct_id : -1;
1150 mutex_exit(&type->ct_type_lock);
1151
1152 return (res);
1153 }
1154
1155 /*
1156 * contract_max
1157 *
1158 * Obtains the maximum contract id.
1159 */
1160 ctid_t
1161 contract_max(void)
1162 {
1163 contract_t *ct;
1164 ctid_t res;
1165
1166 mutex_enter(&contract_lock);
1167 ct = avl_last(&contract_avl);
1168 res = ct ? ct->ct_id : -1;
1169 mutex_exit(&contract_lock);
1170
1171 return (res);
1172 }
1173
1174 /*
1175 * contract_lookup_common
1176 *
1177 * Common code for contract_lookup and contract_type_lookup. Takes a
1178 * pointer to an AVL tree to search in. Should be called with the
1179 * appropriate tree-protecting lock held (unfortunately unassertable).
1180 */
1181 static ctid_t
1182 contract_lookup_common(avl_tree_t *tree, uint64_t zuniqid, ctid_t current)
1183 {
1184 contract_t template, *ct;
1185 avl_index_t where;
1186 ctid_t res;
1187
1188 template.ct_id = current;
1189 ct = avl_find(tree, &template, &where);
1190 if (ct == NULL)
1191 ct = avl_nearest(tree, where, AVL_AFTER);
1192 if (zuniqid != GLOBAL_ZONEUNIQID)
1193 while (ct && (contract_getzuniqid(ct) != zuniqid))
1194 ct = AVL_NEXT(tree, ct);
1195 res = ct ? ct->ct_id : -1;
1196
1197 return (res);
1198 }
1199
1200 /*
1201 * contract_type_lookup
1202 *
1203 * Returns the next type contract after the specified id, visible from
1204 * the specified zone.
1205 */
1206 ctid_t
1207 contract_type_lookup(ct_type_t *type, uint64_t zuniqid, ctid_t current)
1208 {
1209 ctid_t res;
1210
1211 mutex_enter(&type->ct_type_lock);
1212 res = contract_lookup_common(&type->ct_type_avl, zuniqid, current);
1213 mutex_exit(&type->ct_type_lock);
1214
1215 return (res);
1216 }
1217
1218 /*
1219 * contract_lookup
1220 *
1221 * Returns the next contract after the specified id, visible from the
1222 * specified zone.
1223 */
1224 ctid_t
1225 contract_lookup(uint64_t zuniqid, ctid_t current)
1226 {
1227 ctid_t res;
1228
1229 mutex_enter(&contract_lock);
1230 res = contract_lookup_common(&contract_avl, zuniqid, current);
1231 mutex_exit(&contract_lock);
1232
1233 return (res);
1234 }
1235
1236 /*
1237 * contract_plookup
1238 *
1239 * Returns the next contract held by process p after the specified id,
1240 * visible from the specified zone. Made complicated by the fact that
1241 * contracts visible in a zone but held by processes outside of the
1242 * zone need to appear as being held by zsched to zone members.
1243 */
1244 ctid_t
1245 contract_plookup(proc_t *p, ctid_t current, uint64_t zuniqid)
1246 {
1247 contract_t template, *ct;
1248 avl_index_t where;
1249 ctid_t res;
1250
1251 template.ct_id = current;
1252 if (zuniqid != GLOBAL_ZONEUNIQID &&
1253 (p->p_flag & (SSYS|SZONETOP)) == (SSYS|SZONETOP)) {
1254 /* This is inelegant. */
1255 mutex_enter(&contract_lock);
1256 ct = avl_find(&contract_avl, &template, &where);
1257 if (ct == NULL)
1258 ct = avl_nearest(&contract_avl, where, AVL_AFTER);
1259 while (ct && !(ct->ct_state < CTS_ORPHAN &&
1260 contract_getzuniqid(ct) == zuniqid &&
1261 ct->ct_czuniqid == GLOBAL_ZONEUNIQID))
1262 ct = AVL_NEXT(&contract_avl, ct);
1263 res = ct ? ct->ct_id : -1;
1264 mutex_exit(&contract_lock);
1265 } else {
1266 mutex_enter(&p->p_lock);
1267 ct = avl_find(&p->p_ct_held, &template, &where);
1268 if (ct == NULL)
1269 ct = avl_nearest(&p->p_ct_held, where, AVL_AFTER);
1270 res = ct ? ct->ct_id : -1;
1271 mutex_exit(&p->p_lock);
1272 }
1273
1274 return (res);
1275 }
1276
1277 /*
1278 * contract_ptr_common
1279 *
1280 * Common code for contract_ptr and contract_type_ptr. Takes a pointer
1281 * to an AVL tree to search in. Should be called with the appropriate
1282 * tree-protecting lock held (unfortunately unassertable).
1283 */
1284 static contract_t *
1285 contract_ptr_common(avl_tree_t *tree, ctid_t id, uint64_t zuniqid)
1286 {
1287 contract_t template, *ct;
1288
1289 template.ct_id = id;
1290 ct = avl_find(tree, &template, NULL);
1291 if (ct == NULL || (zuniqid != GLOBAL_ZONEUNIQID &&
1292 contract_getzuniqid(ct) != zuniqid)) {
1293 return (NULL);
1294 }
1295
1296 /*
1297 * Check to see if a thread is in the window in contract_rele
1298 * between dropping the reference count and removing the
1299 * contract from the type AVL.
1300 */
1301 mutex_enter(&ct->ct_reflock);
1302 if (ct->ct_ref) {
1303 ct->ct_ref++;
1304 mutex_exit(&ct->ct_reflock);
1305 } else {
1306 mutex_exit(&ct->ct_reflock);
1307 ct = NULL;
1308 }
1309
1310 return (ct);
1311 }
1312
1313 /*
1314 * contract_type_ptr
1315 *
1316 * Returns a pointer to the contract with the specified id. The
1317 * contract is held, so the caller needs to release the reference when
1318 * it is through with the contract.
1319 */
1320 contract_t *
1321 contract_type_ptr(ct_type_t *type, ctid_t id, uint64_t zuniqid)
1322 {
1323 contract_t *ct;
1324
1325 mutex_enter(&type->ct_type_lock);
1326 ct = contract_ptr_common(&type->ct_type_avl, id, zuniqid);
1327 mutex_exit(&type->ct_type_lock);
1328
1329 return (ct);
1330 }
1331
1332 /*
1333 * contract_ptr
1334 *
1335 * Returns a pointer to the contract with the specified id. The
1336 * contract is held, so the caller needs to release the reference when
1337 * it is through with the contract.
1338 */
1339 contract_t *
1340 contract_ptr(ctid_t id, uint64_t zuniqid)
1341 {
1342 contract_t *ct;
1343
1344 mutex_enter(&contract_lock);
1345 ct = contract_ptr_common(&contract_avl, id, zuniqid);
1346 mutex_exit(&contract_lock);
1347
1348 return (ct);
1349 }
1350
1351 /*
1352 * contract_type_time
1353 *
1354 * Obtains the last time a contract of a particular type was created.
1355 */
1356 void
1357 contract_type_time(ct_type_t *type, timestruc_t *time)
1358 {
1359 mutex_enter(&type->ct_type_lock);
1360 *time = type->ct_type_timestruc;
1361 mutex_exit(&type->ct_type_lock);
1362 }
1363
1364 /*
1365 * contract_type_bundle
1366 *
1367 * Obtains a type's bundle queue.
1368 */
1369 ct_equeue_t *
1370 contract_type_bundle(ct_type_t *type)
1371 {
1372 return (&type->ct_type_events);
1373 }
1374
1375 /*
1376 * contract_type_pbundle
1377 *
1378 * Obtain's a process's bundle queue. If one doesn't exist, one is
1379 * created. Often used simply to ensure that a bundle queue is
1380 * allocated.
1381 */
1382 ct_equeue_t *
1383 contract_type_pbundle(ct_type_t *type, proc_t *pp)
1384 {
1385 /*
1386 * If there isn't an array of bundle queues, allocate one.
1387 */
1388 if (pp->p_ct_equeue == NULL) {
1389 size_t size = CTT_MAXTYPE * sizeof (ct_equeue_t *);
1390 ct_equeue_t **qa = kmem_zalloc(size, KM_SLEEP);
1391
1392 mutex_enter(&pp->p_lock);
1393 if (pp->p_ct_equeue)
1394 kmem_free(qa, size);
1395 else
1396 pp->p_ct_equeue = qa;
1397 mutex_exit(&pp->p_lock);
1398 }
1399
1400 /*
1401 * If there isn't a bundle queue of the required type, allocate
1402 * one.
1403 */
1404 if (pp->p_ct_equeue[type->ct_type_index] == NULL) {
1405 ct_equeue_t *q = kmem_zalloc(sizeof (ct_equeue_t), KM_SLEEP);
1406 cte_queue_create(q, CTEL_PBUNDLE, 20, 1);
1407
1408 mutex_enter(&pp->p_lock);
1409 if (pp->p_ct_equeue[type->ct_type_index])
1410 cte_queue_drain(q, 0);
1411 else
1412 pp->p_ct_equeue[type->ct_type_index] = q;
1413 mutex_exit(&pp->p_lock);
1414 }
1415
1416 return (pp->p_ct_equeue[type->ct_type_index]);
1417 }
1418
1419 /*
1420 * ctparam_copyin
1421 *
1422 * copyin a ct_param_t for CT_TSET or CT_TGET commands.
1423 * If ctparam_copyout() is not called after ctparam_copyin(), then
1424 * the caller must kmem_free() the buffer pointed by kparam->ctpm_kbuf.
1425 *
1426 * The copyin/out of ct_param_t is not done in ctmpl_set() and ctmpl_get()
1427 * because prctioctl() calls ctmpl_set() and ctmpl_get() while holding a
1428 * process lock.
1429 */
1430 int
1431 ctparam_copyin(const void *uaddr, ct_kparam_t *kparam, int flag, int cmd)
1432 {
1433 uint32_t size;
1434 void *ubuf;
1435 ct_param_t *param = &kparam->param;
1436 STRUCT_DECL(ct_param, uarg);
1437
1438 STRUCT_INIT(uarg, flag);
1439 if (copyin(uaddr, STRUCT_BUF(uarg), STRUCT_SIZE(uarg)))
1440 return (EFAULT);
1441 size = STRUCT_FGET(uarg, ctpm_size);
1442 ubuf = STRUCT_FGETP(uarg, ctpm_value);
1443
1444 if (size > CT_PARAM_MAX_SIZE || size == 0)
1445 return (EINVAL);
1446
1447 kparam->ctpm_kbuf = kmem_alloc(size, KM_SLEEP);
1448 if (cmd == CT_TSET) {
1449 if (copyin(ubuf, kparam->ctpm_kbuf, size)) {
1450 kmem_free(kparam->ctpm_kbuf, size);
1451 return (EFAULT);
1452 }
1453 }
1454 param->ctpm_id = STRUCT_FGET(uarg, ctpm_id);
1455 param->ctpm_size = size;
1456 param->ctpm_value = ubuf;
1457 kparam->ret_size = 0;
1458
1459 return (0);
1460 }
1461
1462 /*
1463 * ctparam_copyout
1464 *
1465 * copyout a ct_kparam_t and frees the buffer pointed by the member
1466 * ctpm_kbuf of ct_kparam_t
1467 */
1468 int
1469 ctparam_copyout(ct_kparam_t *kparam, void *uaddr, int flag)
1470 {
1471 int r = 0;
1472 ct_param_t *param = &kparam->param;
1473 STRUCT_DECL(ct_param, uarg);
1474
1475 STRUCT_INIT(uarg, flag);
1476
1477 STRUCT_FSET(uarg, ctpm_id, param->ctpm_id);
1478 STRUCT_FSET(uarg, ctpm_size, kparam->ret_size);
1479 STRUCT_FSETP(uarg, ctpm_value, param->ctpm_value);
1480 if (copyout(STRUCT_BUF(uarg), uaddr, STRUCT_SIZE(uarg))) {
1481 r = EFAULT;
1482 goto error;
1483 }
1484 if (copyout(kparam->ctpm_kbuf, param->ctpm_value,
1485 MIN(kparam->ret_size, param->ctpm_size))) {
1486 r = EFAULT;
1487 }
1488
1489 error:
1490 kmem_free(kparam->ctpm_kbuf, param->ctpm_size);
1491
1492 return (r);
1493 }
1494
1495 /*
1496 * ctmpl_free
1497 *
1498 * Frees a template.
1499 */
1500 void
1501 ctmpl_free(ct_template_t *template)
1502 {
1503 mutex_destroy(&template->ctmpl_lock);
1504 template->ctmpl_ops->ctop_free(template);
1505 }
1506
1507 /*
1508 * ctmpl_dup
1509 *
1510 * Creates a copy of a template.
1511 */
1512 ct_template_t *
1513 ctmpl_dup(ct_template_t *template)
1514 {
1515 ct_template_t *new;
1516
1517 if (template == NULL)
1518 return (NULL);
1519
1520 new = template->ctmpl_ops->ctop_dup(template);
1521 /*
1522 * ctmpl_lock was taken by ctop_dup's call to ctmpl_copy and
1523 * should have remain held until now.
1524 */
1525 mutex_exit(&template->ctmpl_lock);
1526
1527 return (new);
1528 }
1529
1530 /*
1531 * ctmpl_set
1532 *
1533 * Sets the requested terms of a template.
1534 */
1535 int
1536 ctmpl_set(ct_template_t *template, ct_kparam_t *kparam, const cred_t *cr)
1537 {
1538 int result = 0;
1539 ct_param_t *param = &kparam->param;
1540 uint64_t param_value;
1541
1542 if (param->ctpm_id == CTP_COOKIE ||
1543 param->ctpm_id == CTP_EV_INFO ||
1544 param->ctpm_id == CTP_EV_CRITICAL) {
1545 if (param->ctpm_size < sizeof (uint64_t)) {
1546 return (EINVAL);
1547 } else {
1548 param_value = *(uint64_t *)kparam->ctpm_kbuf;
1549 }
1550 }
1551
1552 mutex_enter(&template->ctmpl_lock);
1553 switch (param->ctpm_id) {
1554 case CTP_COOKIE:
1555 template->ctmpl_cookie = param_value;
1556 break;
1557 case CTP_EV_INFO:
1558 if (param_value & ~(uint64_t)template->ctmpl_ops->allevents)
1559 result = EINVAL;
1560 else
1561 template->ctmpl_ev_info = param_value;
1562 break;
1563 case CTP_EV_CRITICAL:
1564 if (param_value & ~(uint64_t)template->ctmpl_ops->allevents) {
1565 result = EINVAL;
1566 break;
1567 } else if ((~template->ctmpl_ev_crit & param_value) == 0) {
1568 /*
1569 * Assume that a pure reduction of the critical
1570 * set is allowed by the contract type.
1571 */
1572 template->ctmpl_ev_crit = param_value;
1573 break;
1574 }
1575 /*
1576 * There may be restrictions on what we can make
1577 * critical, so we defer to the judgement of the
1578 * contract type.
1579 */
1580 /* FALLTHROUGH */
1581 default:
1582 result = template->ctmpl_ops->ctop_set(template, kparam, cr);
1583 }
1584 mutex_exit(&template->ctmpl_lock);
1585
1586 return (result);
1587 }
1588
1589 /*
1590 * ctmpl_get
1591 *
1592 * Obtains the requested terms from a template.
1593 *
1594 * If the term requested is a variable-sized term and the buffer
1595 * provided is too small for the data, we truncate the data and return
1596 * the buffer size necessary to fit the term in kparam->ret_size. If the
1597 * term requested is fix-sized (uint64_t) and the buffer provided is too
1598 * small, we return EINVAL. This should never happen if you're using
1599 * libcontract(3LIB), only if you call ioctl with a hand constructed
1600 * ct_param_t argument.
1601 *
1602 * Currently, only contract specific parameters have variable-sized
1603 * parameters.
1604 */
1605 int
1606 ctmpl_get(ct_template_t *template, ct_kparam_t *kparam)
1607 {
1608 int result = 0;
1609 ct_param_t *param = &kparam->param;
1610 uint64_t *param_value;
1611
1612 if (param->ctpm_id == CTP_COOKIE ||
1613 param->ctpm_id == CTP_EV_INFO ||
1614 param->ctpm_id == CTP_EV_CRITICAL) {
1615 if (param->ctpm_size < sizeof (uint64_t)) {
1616 return (EINVAL);
1617 } else {
1618 param_value = kparam->ctpm_kbuf;
1619 kparam->ret_size = sizeof (uint64_t);
1620 }
1621 }
1622
1623 mutex_enter(&template->ctmpl_lock);
1624 switch (param->ctpm_id) {
1625 case CTP_COOKIE:
1626 *param_value = template->ctmpl_cookie;
1627 break;
1628 case CTP_EV_INFO:
1629 *param_value = template->ctmpl_ev_info;
1630 break;
1631 case CTP_EV_CRITICAL:
1632 *param_value = template->ctmpl_ev_crit;
1633 break;
1634 default:
1635 result = template->ctmpl_ops->ctop_get(template, kparam);
1636 }
1637 mutex_exit(&template->ctmpl_lock);
1638
1639 return (result);
1640 }
1641
1642 /*
1643 * ctmpl_makecurrent
1644 *
1645 * Used by ctmpl_activate and ctmpl_clear to set the current thread's
1646 * active template. Frees the old active template, if there was one.
1647 */
1648 static void
1649 ctmpl_makecurrent(ct_template_t *template, ct_template_t *new)
1650 {
1651 klwp_t *curlwp = ttolwp(curthread);
1652 proc_t *p = curproc;
1653 ct_template_t *old;
1654
1655 mutex_enter(&p->p_lock);
1656 old = curlwp->lwp_ct_active[template->ctmpl_type->ct_type_index];
1657 curlwp->lwp_ct_active[template->ctmpl_type->ct_type_index] = new;
1658 mutex_exit(&p->p_lock);
1659
1660 if (old)
1661 ctmpl_free(old);
1662 }
1663
1664 /*
1665 * ctmpl_activate
1666 *
1667 * Copy the specified template as the current thread's activate
1668 * template of that type.
1669 */
1670 void
1671 ctmpl_activate(ct_template_t *template)
1672 {
1673 ctmpl_makecurrent(template, ctmpl_dup(template));
1674 }
1675
1676 /*
1677 * ctmpl_clear
1678 *
1679 * Clears the current thread's activate template of the same type as
1680 * the specified template.
1681 */
1682 void
1683 ctmpl_clear(ct_template_t *template)
1684 {
1685 ctmpl_makecurrent(template, NULL);
1686 }
1687
1688 /*
1689 * ctmpl_create
1690 *
1691 * Creates a new contract using the specified template.
1692 */
1693 int
1694 ctmpl_create(ct_template_t *template, ctid_t *ctidp)
1695 {
1696 return (template->ctmpl_ops->ctop_create(template, ctidp));
1697 }
1698
1699 /*
1700 * ctmpl_init
1701 *
1702 * Initializes the common portion of a new contract template.
1703 */
1704 void
1705 ctmpl_init(ct_template_t *new, ctmplops_t *ops, ct_type_t *type, void *data)
1706 {
1707 mutex_init(&new->ctmpl_lock, NULL, MUTEX_DEFAULT, NULL);
1708 new->ctmpl_ops = ops;
1709 new->ctmpl_type = type;
1710 new->ctmpl_data = data;
1711 new->ctmpl_ev_info = new->ctmpl_ev_crit = 0;
1712 new->ctmpl_cookie = 0;
1713 }
1714
1715 /*
1716 * ctmpl_copy
1717 *
1718 * Copies the common portions of a contract template. Intended for use
1719 * by a contract type's ctop_dup template op. Returns with the old
1720 * template's lock held, which will should remain held until the
1721 * template op returns (it is dropped by ctmpl_dup).
1722 */
1723 void
1724 ctmpl_copy(ct_template_t *new, ct_template_t *old)
1725 {
1726 mutex_init(&new->ctmpl_lock, NULL, MUTEX_DEFAULT, NULL);
1727 mutex_enter(&old->ctmpl_lock);
1728 new->ctmpl_ops = old->ctmpl_ops;
1729 new->ctmpl_type = old->ctmpl_type;
1730 new->ctmpl_ev_crit = old->ctmpl_ev_crit;
1731 new->ctmpl_ev_info = old->ctmpl_ev_info;
1732 new->ctmpl_cookie = old->ctmpl_cookie;
1733 }
1734
1735 /*
1736 * ctmpl_create_inval
1737 *
1738 * Returns EINVAL. Provided for the convenience of those contract
1739 * types which don't support ct_tmpl_create(3contract) and would
1740 * otherwise need to create their own stub for the ctop_create template
1741 * op.
1742 */
1743 /*ARGSUSED*/
1744 int
1745 ctmpl_create_inval(ct_template_t *template, ctid_t *ctidp)
1746 {
1747 return (EINVAL);
1748 }
1749
1750
1751 /*
1752 * cte_queue_create
1753 *
1754 * Initializes a queue of a particular type. If dynamic is set, the
1755 * queue is to be freed when its last listener is removed after being
1756 * drained.
1757 */
1758 static void
1759 cte_queue_create(ct_equeue_t *q, ct_listnum_t list, int maxinf, int dynamic)
1760 {
1761 mutex_init(&q->ctq_lock, NULL, MUTEX_DEFAULT, NULL);
1762 q->ctq_listno = list;
1763 list_create(&q->ctq_events, sizeof (ct_kevent_t),
1764 offsetof(ct_kevent_t, cte_nodes[list].ctm_node));
1765 list_create(&q->ctq_listeners, sizeof (ct_listener_t),
1766 offsetof(ct_listener_t, ctl_allnode));
1767 list_create(&q->ctq_tail, sizeof (ct_listener_t),
1768 offsetof(ct_listener_t, ctl_tailnode));
1769 gethrestime(&q->ctq_atime);
1770 q->ctq_nlisteners = 0;
1771 q->ctq_nreliable = 0;
1772 q->ctq_ninf = 0;
1773 q->ctq_max = maxinf;
1774
1775 /*
1776 * Bundle queues and contract queues are embedded in other
1777 * structures and are implicitly referenced counted by virtue
1778 * of their vnodes' indirect hold on their contracts. Process
1779 * bundle queues are dynamically allocated and may persist
1780 * after the death of the process, so they must be explicitly
1781 * reference counted.
1782 */
1783 q->ctq_flags = dynamic ? CTQ_REFFED : 0;
1784 }
1785
1786 /*
1787 * cte_queue_destroy
1788 *
1789 * Destroys the specified queue. The queue is freed if referenced
1790 * counted.
1791 */
1792 static void
1793 cte_queue_destroy(ct_equeue_t *q)
1794 {
1795 ASSERT(q->ctq_flags & CTQ_DEAD);
1796 ASSERT(q->ctq_nlisteners == 0);
1797 ASSERT(q->ctq_nreliable == 0);
1798 list_destroy(&q->ctq_events);
1799 list_destroy(&q->ctq_listeners);
1800 list_destroy(&q->ctq_tail);
1801 mutex_destroy(&q->ctq_lock);
1802 if (q->ctq_flags & CTQ_REFFED)
1803 kmem_free(q, sizeof (ct_equeue_t));
1804 }
1805
1806 /*
1807 * cte_hold
1808 *
1809 * Takes a hold on the specified event.
1810 */
1811 static void
1812 cte_hold(ct_kevent_t *e)
1813 {
1814 mutex_enter(&e->cte_lock);
1815 ASSERT(e->cte_refs > 0);
1816 e->cte_refs++;
1817 mutex_exit(&e->cte_lock);
1818 }
1819
1820 /*
1821 * cte_rele
1822 *
1823 * Releases a hold on the specified event. If the caller had the last
1824 * reference, frees the event and releases its hold on the contract
1825 * that generated it.
1826 */
1827 static void
1828 cte_rele(ct_kevent_t *e)
1829 {
1830 mutex_enter(&e->cte_lock);
1831 ASSERT(e->cte_refs > 0);
1832 if (--e->cte_refs) {
1833 mutex_exit(&e->cte_lock);
1834 return;
1835 }
1836
1837 contract_rele(e->cte_contract);
1838
1839 mutex_destroy(&e->cte_lock);
1840 nvlist_free(e->cte_data);
1841 nvlist_free(e->cte_gdata);
1842 kmem_free(e, sizeof (ct_kevent_t));
1843 }
1844
1845 /*
1846 * cte_qrele
1847 *
1848 * Remove this listener's hold on the specified event, removing and
1849 * releasing the queue's hold on the event if appropriate.
1850 */
1851 static void
1852 cte_qrele(ct_equeue_t *q, ct_listener_t *l, ct_kevent_t *e)
1853 {
1854 ct_member_t *member = &e->cte_nodes[q->ctq_listno];
1855
1856 ASSERT(MUTEX_HELD(&q->ctq_lock));
1857
1858 if (l->ctl_flags & CTLF_RELIABLE)
1859 member->ctm_nreliable--;
1860 if ((--member->ctm_refs == 0) && member->ctm_trimmed) {
1861 member->ctm_trimmed = 0;
1862 list_remove(&q->ctq_events, e);
1863 cte_rele(e);
1864 }
1865 }
1866
1867 /*
1868 * cte_qmove
1869 *
1870 * Move this listener to the specified event in the queue.
1871 */
1872 static ct_kevent_t *
1873 cte_qmove(ct_equeue_t *q, ct_listener_t *l, ct_kevent_t *e)
1874 {
1875 ct_kevent_t *olde;
1876
1877 ASSERT(MUTEX_HELD(&q->ctq_lock));
1878 ASSERT(l->ctl_equeue == q);
1879
1880 if ((olde = l->ctl_position) == NULL)
1881 list_remove(&q->ctq_tail, l);
1882
1883 while (e != NULL && e->cte_nodes[q->ctq_listno].ctm_trimmed)
1884 e = list_next(&q->ctq_events, e);
1885
1886 if (e != NULL) {
1887 e->cte_nodes[q->ctq_listno].ctm_refs++;
1888 if (l->ctl_flags & CTLF_RELIABLE)
1889 e->cte_nodes[q->ctq_listno].ctm_nreliable++;
1890 } else {
1891 list_insert_tail(&q->ctq_tail, l);
1892 }
1893
1894 l->ctl_position = e;
1895 if (olde)
1896 cte_qrele(q, l, olde);
1897
1898 return (e);
1899 }
1900
1901 /*
1902 * cte_checkcred
1903 *
1904 * Determines if the specified event's contract is owned by a process
1905 * with the same effective uid as the specified credential. Called
1906 * after a failed call to contract_owned with locked set. Because it
1907 * drops the queue lock, its caller (cte_qreadable) needs to make sure
1908 * we're still in the same place after we return. Returns 1 on
1909 * success.
1910 */
1911 static int
1912 cte_checkcred(ct_equeue_t *q, ct_kevent_t *e, const cred_t *cr)
1913 {
1914 int result;
1915 contract_t *ct = e->cte_contract;
1916
1917 cte_hold(e);
1918 mutex_exit(&q->ctq_lock);
1919 result = curproc->p_zone->zone_uniqid == ct->ct_czuniqid &&
1920 contract_checkcred(ct, cr);
1921 mutex_enter(&q->ctq_lock);
1922 cte_rele(e);
1923
1924 return (result);
1925 }
1926
1927 /*
1928 * cte_qreadable
1929 *
1930 * Ensures that the listener is pointing to a valid event that the
1931 * caller has the credentials to read. Returns 0 if we can read the
1932 * event we're pointing to.
1933 */
1934 static int
1935 cte_qreadable(ct_equeue_t *q, ct_listener_t *l, const cred_t *cr,
1936 uint64_t zuniqid, int crit)
1937 {
1938 ct_kevent_t *e, *next;
1939 contract_t *ct;
1940
1941 ASSERT(MUTEX_HELD(&q->ctq_lock));
1942 ASSERT(l->ctl_equeue == q);
1943
1944 if (l->ctl_flags & CTLF_COPYOUT)
1945 return (1);
1946
1947 next = l->ctl_position;
1948 while (e = cte_qmove(q, l, next)) {
1949 ct = e->cte_contract;
1950 /*
1951 * Check obvious things first. If we are looking for a
1952 * critical message, is this one? If we aren't in the
1953 * global zone, is this message meant for us?
1954 */
1955 if ((crit && (e->cte_flags & (CTE_INFO | CTE_ACK))) ||
1956 (cr != NULL && zuniqid != GLOBAL_ZONEUNIQID &&
1957 zuniqid != contract_getzuniqid(ct))) {
1958
1959 next = list_next(&q->ctq_events, e);
1960
1961 /*
1962 * Next, see if our effective uid equals that of owner
1963 * or author of the contract. Since we are holding the
1964 * queue lock, contract_owned can't always check if we
1965 * have the same effective uid as the contract's
1966 * owner. If it comes to that, it fails and we take
1967 * the slow(er) path.
1968 */
1969 } else if (cr != NULL && !contract_owned(ct, cr, B_TRUE)) {
1970
1971 /*
1972 * At this point we either don't have any claim
1973 * to this contract or we match the effective
1974 * uid of the owner but couldn't tell. We
1975 * first test for a NULL holder so that events
1976 * from orphans and inherited contracts avoid
1977 * the penalty phase.
1978 */
1979 if (e->cte_contract->ct_owner == NULL &&
1980 !secpolicy_contract_observer_choice(cr))
1981 next = list_next(&q->ctq_events, e);
1982
1983 /*
1984 * cte_checkcred will juggle locks to see if we
1985 * have the same uid as the event's contract's
1986 * current owner. If it succeeds, we have to
1987 * make sure we are in the same point in the
1988 * queue.
1989 */
1990 else if (cte_checkcred(q, e, cr) &&
1991 l->ctl_position == e)
1992 break;
1993
1994 /*
1995 * cte_checkcred failed; see if we're in the
1996 * same place.
1997 */
1998 else if (l->ctl_position == e)
1999 if (secpolicy_contract_observer_choice(cr))
2000 break;
2001 else
2002 next = list_next(&q->ctq_events, e);
2003
2004 /*
2005 * cte_checkcred failed, and our position was
2006 * changed. Start from there.
2007 */
2008 else
2009 next = l->ctl_position;
2010 } else {
2011 break;
2012 }
2013 }
2014
2015 /*
2016 * We check for CTLF_COPYOUT again in case we dropped the queue
2017 * lock in cte_checkcred.
2018 */
2019 return ((l->ctl_flags & CTLF_COPYOUT) || (l->ctl_position == NULL));
2020 }
2021
2022 /*
2023 * cte_qwakeup
2024 *
2025 * Wakes up any waiting listeners and points them at the specified event.
2026 */
2027 static void
2028 cte_qwakeup(ct_equeue_t *q, ct_kevent_t *e)
2029 {
2030 ct_listener_t *l;
2031
2032 ASSERT(MUTEX_HELD(&q->ctq_lock));
2033
2034 while (l = list_head(&q->ctq_tail)) {
2035 list_remove(&q->ctq_tail, l);
2036 e->cte_nodes[q->ctq_listno].ctm_refs++;
2037 if (l->ctl_flags & CTLF_RELIABLE)
2038 e->cte_nodes[q->ctq_listno].ctm_nreliable++;
2039 l->ctl_position = e;
2040 cv_signal(&l->ctl_cv);
2041 pollwakeup(&l->ctl_pollhead, POLLIN);
2042 }
2043 }
2044
2045 /*
2046 * cte_copy
2047 *
2048 * Copies events from the specified contract event queue to the
2049 * end of the specified process bundle queue. Only called from
2050 * contract_adopt.
2051 *
2052 * We copy to the end of the target queue instead of mixing the events
2053 * in their proper order because otherwise the act of adopting a
2054 * contract would require a process to reset all process bundle
2055 * listeners it needed to see the new events. This would, in turn,
2056 * require the process to keep track of which preexisting events had
2057 * already been processed.
2058 */
2059 static void
2060 cte_copy(ct_equeue_t *q, ct_equeue_t *newq)
2061 {
2062 ct_kevent_t *e, *first = NULL;
2063
2064 VERIFY(q->ctq_listno == CTEL_CONTRACT);
2065 VERIFY(newq->ctq_listno == CTEL_PBUNDLE);
2066
2067 mutex_enter(&q->ctq_lock);
2068 mutex_enter(&newq->ctq_lock);
2069
2070 /*
2071 * For now, only copy critical events.
2072 */
2073 for (e = list_head(&q->ctq_events); e != NULL;
2074 e = list_next(&q->ctq_events, e)) {
2075 if ((e->cte_flags & (CTE_INFO | CTE_ACK)) == 0) {
2076 if (first == NULL)
2077 first = e;
2078 /*
2079 * It is possible for adoption to race with an owner's
2080 * cte_publish_all(); we must only enqueue events that
2081 * have not already been enqueued.
2082 */
2083 if (!list_link_active((list_node_t *)
2084 ((uintptr_t)e + newq->ctq_events.list_offset))) {
2085 list_insert_tail(&newq->ctq_events, e);
2086 cte_hold(e);
2087 }
2088 }
2089 }
2090
2091 mutex_exit(&q->ctq_lock);
2092
2093 if (first)
2094 cte_qwakeup(newq, first);
2095
2096 mutex_exit(&newq->ctq_lock);
2097 }
2098
2099 /*
2100 * cte_trim
2101 *
2102 * Trims unneeded events from an event queue. Algorithm works as
2103 * follows:
2104 *
2105 * Removes all informative and acknowledged critical events until the
2106 * first referenced event is found.
2107 *
2108 * If a contract is specified, removes all events (regardless of
2109 * acknowledgement) generated by that contract until the first event
2110 * referenced by a reliable listener is found. Reference events are
2111 * removed by marking them "trimmed". Such events will be removed
2112 * when the last reference is dropped and will be skipped by future
2113 * listeners.
2114 *
2115 * This is pretty basic. Ideally this should remove from the middle of
2116 * the list (i.e. beyond the first referenced event), and even
2117 * referenced events.
2118 */
2119 static void
2120 cte_trim(ct_equeue_t *q, contract_t *ct)
2121 {
2122 ct_kevent_t *e, *next;
2123 int flags, stopper;
2124 int start = 1;
2125
2126 VERIFY(MUTEX_HELD(&q->ctq_lock));
2127
2128 for (e = list_head(&q->ctq_events); e != NULL; e = next) {
2129 next = list_next(&q->ctq_events, e);
2130 flags = e->cte_flags;
2131 stopper = (q->ctq_listno != CTEL_PBUNDLE) &&
2132 (e->cte_nodes[q->ctq_listno].ctm_nreliable > 0);
2133 if (e->cte_nodes[q->ctq_listno].ctm_refs == 0) {
2134 if ((start && (flags & (CTE_INFO | CTE_ACK))) ||
2135 (e->cte_contract == ct)) {
2136 /*
2137 * Toss informative and ACKed critical messages.
2138 */
2139 list_remove(&q->ctq_events, e);
2140 cte_rele(e);
2141 }
2142 } else if ((e->cte_contract == ct) && !stopper) {
2143 ASSERT(q->ctq_nlisteners != 0);
2144 e->cte_nodes[q->ctq_listno].ctm_trimmed = 1;
2145 } else if (ct && !stopper) {
2146 start = 0;
2147 } else {
2148 /*
2149 * Don't free messages past the first reader.
2150 */
2151 break;
2152 }
2153 }
2154 }
2155
2156 /*
2157 * cte_queue_drain
2158 *
2159 * Drain all events from the specified queue, and mark it dead. If
2160 * "ack" is set, acknowledge any critical events we find along the
2161 * way.
2162 */
2163 static void
2164 cte_queue_drain(ct_equeue_t *q, int ack)
2165 {
2166 ct_kevent_t *e, *next;
2167 ct_listener_t *l;
2168
2169 mutex_enter(&q->ctq_lock);
2170
2171 for (e = list_head(&q->ctq_events); e != NULL; e = next) {
2172 next = list_next(&q->ctq_events, e);
2173 if (ack && ((e->cte_flags & (CTE_INFO | CTE_ACK)) == 0)) {
2174 /*
2175 * Make sure critical messages are eventually
2176 * removed from the bundle queues.
2177 */
2178 mutex_enter(&e->cte_lock);
2179 e->cte_flags |= CTE_ACK;
2180 mutex_exit(&e->cte_lock);
2181 ASSERT(MUTEX_HELD(&e->cte_contract->ct_lock));
2182 e->cte_contract->ct_evcnt--;
2183 }
2184 list_remove(&q->ctq_events, e);
2185 e->cte_nodes[q->ctq_listno].ctm_refs = 0;
2186 e->cte_nodes[q->ctq_listno].ctm_nreliable = 0;
2187 e->cte_nodes[q->ctq_listno].ctm_trimmed = 0;
2188 cte_rele(e);
2189 }
2190
2191 /*
2192 * This is necessary only because of CTEL_PBUNDLE listeners;
2193 * the events they point to can move from one pbundle to
2194 * another. Fortunately, this only happens if the contract is
2195 * inherited, which (in turn) only happens if the process
2196 * exits, which means it's an all-or-nothing deal. If this
2197 * wasn't the case, we would instead need to keep track of
2198 * listeners on a per-event basis, not just a per-queue basis.
2199 * This would have the side benefit of letting us clean up
2200 * trimmed events sooner (i.e. immediately), but would
2201 * unfortunately make events even bigger than they already
2202 * are.
2203 */
2204 for (l = list_head(&q->ctq_listeners); l;
2205 l = list_next(&q->ctq_listeners, l)) {
2206 l->ctl_flags |= CTLF_DEAD;
2207 if (l->ctl_position) {
2208 l->ctl_position = NULL;
2209 list_insert_tail(&q->ctq_tail, l);
2210 }
2211 cv_broadcast(&l->ctl_cv);
2212 }
2213
2214 /*
2215 * Disallow events.
2216 */
2217 q->ctq_flags |= CTQ_DEAD;
2218
2219 /*
2220 * If we represent the last reference to a reference counted
2221 * process bundle queue, free it.
2222 */
2223 if ((q->ctq_flags & CTQ_REFFED) && (q->ctq_nlisteners == 0))
2224 cte_queue_destroy(q);
2225 else
2226 mutex_exit(&q->ctq_lock);
2227 }
2228
2229 /*
2230 * cte_publish
2231 *
2232 * Publishes an event to a specific queue. Only called by
2233 * cte_publish_all.
2234 */
2235 static void
2236 cte_publish(ct_equeue_t *q, ct_kevent_t *e, timespec_t *tsp, boolean_t mayexist)
2237 {
2238 ASSERT(MUTEX_HELD(&q->ctq_lock));
2239
2240 q->ctq_atime = *tsp;
2241
2242 /*
2243 * If this event may already exist on this queue, check to see if it
2244 * is already there and return if so.
2245 */
2246 if (mayexist && list_link_active((list_node_t *)((uintptr_t)e +
2247 q->ctq_events.list_offset))) {
2248 mutex_exit(&q->ctq_lock);
2249 cte_rele(e);
2250 return;
2251 }
2252
2253 /*
2254 * Don't publish if the event is informative and there aren't
2255 * any listeners, or if the queue has been shut down.
2256 */
2257 if (((q->ctq_nlisteners == 0) && (e->cte_flags & (CTE_INFO|CTE_ACK))) ||
2258 (q->ctq_flags & CTQ_DEAD)) {
2259 mutex_exit(&q->ctq_lock);
2260 cte_rele(e);
2261 return;
2262 }
2263
2264 /*
2265 * Enqueue event
2266 */
2267 VERIFY(!list_link_active((list_node_t *)
2268 ((uintptr_t)e + q->ctq_events.list_offset)));
2269 list_insert_tail(&q->ctq_events, e);
2270
2271 /*
2272 * Check for waiting listeners
2273 */
2274 cte_qwakeup(q, e);
2275
2276 /*
2277 * Trim unnecessary events from the queue.
2278 */
2279 cte_trim(q, NULL);
2280 mutex_exit(&q->ctq_lock);
2281 }
2282
2283 /*
2284 * cte_publish_all
2285 *
2286 * Publish an event to all necessary event queues. The event, e, must
2287 * be zallocated by the caller, and the event's flags and type must be
2288 * set. The rest of the event's fields are initialized here.
2289 */
2290 uint64_t
2291 cte_publish_all(contract_t *ct, ct_kevent_t *e, nvlist_t *data, nvlist_t *gdata)
2292 {
2293 ct_equeue_t *q;
2294 timespec_t ts;
2295 uint64_t evid;
2296 ct_kevent_t *negev;
2297 int negend;
2298
2299 e->cte_contract = ct;
2300 e->cte_data = data;
2301 e->cte_gdata = gdata;
2302 e->cte_refs = 3;
2303 evid = e->cte_id = atomic_inc_64_nv(&ct->ct_type->ct_type_evid);
2304 contract_hold(ct);
2305
2306 /*
2307 * For a negotiation event we set the ct->ct_nevent field of the
2308 * contract for the duration of the negotiation
2309 */
2310 negend = 0;
2311 if (e->cte_flags & CTE_NEG) {
2312 cte_hold(e);
2313 ct->ct_nevent = e;
2314 } else if (e->cte_type == CT_EV_NEGEND) {
2315 negend = 1;
2316 }
2317
2318 gethrestime(&ts);
2319
2320 /*
2321 * ct_evtlock simply (and only) ensures that two events sent
2322 * from the same contract are delivered to all queues in the
2323 * same order.
2324 */
2325 mutex_enter(&ct->ct_evtlock);
2326
2327 /*
2328 * CTEL_CONTRACT - First deliver to the contract queue, acking
2329 * the event if the contract has been orphaned.
2330 */
2331 mutex_enter(&ct->ct_lock);
2332 mutex_enter(&ct->ct_events.ctq_lock);
2333 if ((e->cte_flags & CTE_INFO) == 0) {
2334 if (ct->ct_state >= CTS_ORPHAN)
2335 e->cte_flags |= CTE_ACK;
2336 else
2337 ct->ct_evcnt++;
2338 }
2339 mutex_exit(&ct->ct_lock);
2340 cte_publish(&ct->ct_events, e, &ts, B_FALSE);
2341
2342 /*
2343 * CTEL_BUNDLE - Next deliver to the contract type's bundle
2344 * queue.
2345 */
2346 mutex_enter(&ct->ct_type->ct_type_events.ctq_lock);
2347 cte_publish(&ct->ct_type->ct_type_events, e, &ts, B_FALSE);
2348
2349 /*
2350 * CTEL_PBUNDLE - Finally, if the contract has an owner,
2351 * deliver to the owner's process bundle queue.
2352 */
2353 mutex_enter(&ct->ct_lock);
2354 if (ct->ct_owner) {
2355 /*
2356 * proc_exit doesn't free event queues until it has
2357 * abandoned all contracts.
2358 */
2359 ASSERT(ct->ct_owner->p_ct_equeue);
2360 ASSERT(ct->ct_owner->p_ct_equeue[ct->ct_type->ct_type_index]);
2361 q = ct->ct_owner->p_ct_equeue[ct->ct_type->ct_type_index];
2362 mutex_enter(&q->ctq_lock);
2363 mutex_exit(&ct->ct_lock);
2364
2365 /*
2366 * It is possible for this code to race with adoption; we
2367 * publish the event indicating that the event may already
2368 * be enqueued because adoption beat us to it (in which case
2369 * cte_pubish() does nothing).
2370 */
2371 cte_publish(q, e, &ts, B_TRUE);
2372 } else {
2373 mutex_exit(&ct->ct_lock);
2374 cte_rele(e);
2375 }
2376
2377 if (negend) {
2378 mutex_enter(&ct->ct_lock);
2379 negev = ct->ct_nevent;
2380 ct->ct_nevent = NULL;
2381 cte_rele(negev);
2382 mutex_exit(&ct->ct_lock);
2383 }
2384
2385 mutex_exit(&ct->ct_evtlock);
2386
2387 return (evid);
2388 }
2389
2390 /*
2391 * cte_add_listener
2392 *
2393 * Add a new listener to an event queue.
2394 */
2395 void
2396 cte_add_listener(ct_equeue_t *q, ct_listener_t *l)
2397 {
2398 cv_init(&l->ctl_cv, NULL, CV_DEFAULT, NULL);
2399 l->ctl_equeue = q;
2400 l->ctl_position = NULL;
2401 l->ctl_flags = 0;
2402
2403 mutex_enter(&q->ctq_lock);
2404 list_insert_head(&q->ctq_tail, l);
2405 list_insert_head(&q->ctq_listeners, l);
2406 q->ctq_nlisteners++;
2407 mutex_exit(&q->ctq_lock);
2408 }
2409
2410 /*
2411 * cte_remove_listener
2412 *
2413 * Remove a listener from an event queue. No other queue activities
2414 * (e.g. cte_get event) may be in progress at this endpoint when this
2415 * is called.
2416 */
2417 void
2418 cte_remove_listener(ct_listener_t *l)
2419 {
2420 ct_equeue_t *q = l->ctl_equeue;
2421 ct_kevent_t *e;
2422
2423 mutex_enter(&q->ctq_lock);
2424
2425 ASSERT((l->ctl_flags & (CTLF_COPYOUT|CTLF_RESET)) == 0);
2426
2427 if ((e = l->ctl_position) != NULL)
2428 cte_qrele(q, l, e);
2429 else
2430 list_remove(&q->ctq_tail, l);
2431 l->ctl_position = NULL;
2432
2433 q->ctq_nlisteners--;
2434 list_remove(&q->ctq_listeners, l);
2435
2436 if (l->ctl_flags & CTLF_RELIABLE)
2437 q->ctq_nreliable--;
2438
2439 /*
2440 * If we are a the last listener of a dead reference counted
2441 * queue (i.e. a process bundle) we free it. Otherwise we just
2442 * trim any events which may have been kept around for our
2443 * benefit.
2444 */
2445 if ((q->ctq_flags & CTQ_REFFED) && (q->ctq_flags & CTQ_DEAD) &&
2446 (q->ctq_nlisteners == 0)) {
2447 cte_queue_destroy(q);
2448 } else {
2449 cte_trim(q, NULL);
2450 mutex_exit(&q->ctq_lock);
2451 }
2452 }
2453
2454 /*
2455 * cte_reset_listener
2456 *
2457 * Moves a listener's queue pointer to the beginning of the queue.
2458 */
2459 void
2460 cte_reset_listener(ct_listener_t *l)
2461 {
2462 ct_equeue_t *q = l->ctl_equeue;
2463
2464 mutex_enter(&q->ctq_lock);
2465
2466 /*
2467 * We allow an asynchronous reset because it doesn't make a
2468 * whole lot of sense to make reset block or fail. We already
2469 * have most of the mechanism needed thanks to queue trimming,
2470 * so implementing it isn't a big deal.
2471 */
2472 if (l->ctl_flags & CTLF_COPYOUT)
2473 l->ctl_flags |= CTLF_RESET;
2474
2475 (void) cte_qmove(q, l, list_head(&q->ctq_events));
2476
2477 /*
2478 * Inform blocked readers.
2479 */
2480 cv_broadcast(&l->ctl_cv);
2481 pollwakeup(&l->ctl_pollhead, POLLIN);
2482 mutex_exit(&q->ctq_lock);
2483 }
2484
2485 /*
2486 * cte_next_event
2487 *
2488 * Moves the event pointer for the specified listener to the next event
2489 * on the queue. To avoid races, this movement only occurs if the
2490 * specified event id matches that of the current event. This is used
2491 * primarily to skip events that have been read but whose extended data
2492 * haven't been copied out.
2493 */
2494 int
2495 cte_next_event(ct_listener_t *l, uint64_t id)
2496 {
2497 ct_equeue_t *q = l->ctl_equeue;
2498 ct_kevent_t *old;
2499
2500 mutex_enter(&q->ctq_lock);
2501
2502 if (l->ctl_flags & CTLF_COPYOUT)
2503 l->ctl_flags |= CTLF_RESET;
2504
2505 if (((old = l->ctl_position) != NULL) && (old->cte_id == id))
2506 (void) cte_qmove(q, l, list_next(&q->ctq_events, old));
2507
2508 mutex_exit(&q->ctq_lock);
2509
2510 return (0);
2511 }
2512
2513 /*
2514 * cte_get_event
2515 *
2516 * Reads an event from an event endpoint. If "nonblock" is clear, we
2517 * block until a suitable event is ready. If "crit" is set, we only
2518 * read critical events. Note that while "cr" is the caller's cred,
2519 * "zuniqid" is the unique id of the zone the calling contract
2520 * filesystem was mounted in.
2521 */
2522 int
2523 cte_get_event(ct_listener_t *l, int nonblock, void *uaddr, const cred_t *cr,
2524 uint64_t zuniqid, int crit)
2525 {
2526 ct_equeue_t *q = l->ctl_equeue;
2527 ct_kevent_t *temp;
2528 int result = 0;
2529 int partial = 0;
2530 size_t size, gsize, len;
2531 model_t mdl = get_udatamodel();
2532 STRUCT_DECL(ct_event, ev);
2533 STRUCT_INIT(ev, mdl);
2534
2535 /*
2536 * cte_qreadable checks for CTLF_COPYOUT as well as ensures
2537 * that there exists, and we are pointing to, an appropriate
2538 * event. It may temporarily drop ctq_lock, but that doesn't
2539 * really matter to us.
2540 */
2541 mutex_enter(&q->ctq_lock);
2542 while (cte_qreadable(q, l, cr, zuniqid, crit)) {
2543 if (nonblock) {
2544 result = EAGAIN;
2545 goto error;
2546 }
2547 if (q->ctq_flags & CTQ_DEAD) {
2548 result = EIDRM;
2549 goto error;
2550 }
2551 result = cv_wait_sig(&l->ctl_cv, &q->ctq_lock);
2552 if (result == 0) {
2553 result = EINTR;
2554 goto error;
2555 }
2556 }
2557 temp = l->ctl_position;
2558 cte_hold(temp);
2559 l->ctl_flags |= CTLF_COPYOUT;
2560 mutex_exit(&q->ctq_lock);
2561
2562 /*
2563 * We now have an event. Copy in the user event structure to
2564 * see how much space we have to work with.
2565 */
2566 result = copyin(uaddr, STRUCT_BUF(ev), STRUCT_SIZE(ev));
2567 if (result)
2568 goto copyerr;
2569
2570 /*
2571 * Determine what data we have and what the user should be
2572 * allowed to see.
2573 */
2574 size = gsize = 0;
2575 if (temp->cte_data) {
2576 VERIFY(nvlist_size(temp->cte_data, &size,
2577 NV_ENCODE_NATIVE) == 0);
2578 ASSERT(size != 0);
2579 }
2580 if (zuniqid == GLOBAL_ZONEUNIQID && temp->cte_gdata) {
2581 VERIFY(nvlist_size(temp->cte_gdata, &gsize,
2582 NV_ENCODE_NATIVE) == 0);
2583 ASSERT(gsize != 0);
2584 }
2585
2586 /*
2587 * If we have enough space, copy out the extended event data.
2588 */
2589 len = size + gsize;
2590 if (len) {
2591 if (STRUCT_FGET(ev, ctev_nbytes) >= len) {
2592 char *buf = kmem_alloc(len, KM_SLEEP);
2593
2594 if (size)
2595 VERIFY(nvlist_pack(temp->cte_data, &buf, &size,
2596 NV_ENCODE_NATIVE, KM_SLEEP) == 0);
2597 if (gsize) {
2598 char *tmp = buf + size;
2599
2600 VERIFY(nvlist_pack(temp->cte_gdata, &tmp,
2601 &gsize, NV_ENCODE_NATIVE, KM_SLEEP) == 0);
2602 }
2603
2604 /* This shouldn't have changed */
2605 ASSERT(size + gsize == len);
2606 result = copyout(buf, STRUCT_FGETP(ev, ctev_buffer),
2607 len);
2608 kmem_free(buf, len);
2609 if (result)
2610 goto copyerr;
2611 } else {
2612 partial = 1;
2613 }
2614 }
2615
2616 /*
2617 * Copy out the common event data.
2618 */
2619 STRUCT_FSET(ev, ctev_id, temp->cte_contract->ct_id);
2620 STRUCT_FSET(ev, ctev_evid, temp->cte_id);
2621 STRUCT_FSET(ev, ctev_cttype,
2622 temp->cte_contract->ct_type->ct_type_index);
2623 STRUCT_FSET(ev, ctev_flags, temp->cte_flags &
2624 (CTE_ACK|CTE_INFO|CTE_NEG));
2625 STRUCT_FSET(ev, ctev_type, temp->cte_type);
2626 STRUCT_FSET(ev, ctev_nbytes, len);
2627 STRUCT_FSET(ev, ctev_goffset, size);
2628 result = copyout(STRUCT_BUF(ev), uaddr, STRUCT_SIZE(ev));
2629
2630 copyerr:
2631 /*
2632 * Only move our location in the queue if all copyouts were
2633 * successful, the caller provided enough space for the entire
2634 * event, and our endpoint wasn't reset or otherwise moved by
2635 * another thread.
2636 */
2637 mutex_enter(&q->ctq_lock);
2638 if (result)
2639 result = EFAULT;
2640 else if (!partial && ((l->ctl_flags & CTLF_RESET) == 0) &&
2641 (l->ctl_position == temp))
2642 (void) cte_qmove(q, l, list_next(&q->ctq_events, temp));
2643 l->ctl_flags &= ~(CTLF_COPYOUT|CTLF_RESET);
2644 /*
2645 * Signal any readers blocked on our CTLF_COPYOUT.
2646 */
2647 cv_signal(&l->ctl_cv);
2648 cte_rele(temp);
2649
2650 error:
2651 mutex_exit(&q->ctq_lock);
2652 return (result);
2653 }
2654
2655 /*
2656 * cte_set_reliable
2657 *
2658 * Requests that events be reliably delivered to an event endpoint.
2659 * Unread informative and acknowledged critical events will not be
2660 * removed from the queue until this listener reads or skips them.
2661 * Because a listener could maliciously request reliable delivery and
2662 * then do nothing, this requires that PRIV_CONTRACT_EVENT be in the
2663 * caller's effective set.
2664 */
2665 int
2666 cte_set_reliable(ct_listener_t *l, const cred_t *cr)
2667 {
2668 ct_equeue_t *q = l->ctl_equeue;
2669 int error;
2670
2671 if ((error = secpolicy_contract_event(cr)) != 0)
2672 return (error);
2673
2674 mutex_enter(&q->ctq_lock);
2675 if ((l->ctl_flags & CTLF_RELIABLE) == 0) {
2676 l->ctl_flags |= CTLF_RELIABLE;
2677 q->ctq_nreliable++;
2678 if (l->ctl_position != NULL)
2679 l->ctl_position->cte_nodes[q->ctq_listno].
2680 ctm_nreliable++;
2681 }
2682 mutex_exit(&q->ctq_lock);
2683
2684 return (0);
2685 }