xref: /titanic_41/usr/src/uts/common/contract/device.c (revision cde2885fdf538266ee2a3b08dee2d5075ce8fa2b)
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 #include <sys/mutex.h>
27 #include <sys/debug.h>
28 #include <sys/types.h>
29 #include <sys/param.h>
30 #include <sys/kmem.h>
31 #include <sys/thread.h>
32 #include <sys/id_space.h>
33 #include <sys/avl.h>
34 #include <sys/list.h>
35 #include <sys/sysmacros.h>
36 #include <sys/proc.h>
37 #include <sys/contract.h>
38 #include <sys/contract_impl.h>
39 #include <sys/contract/device.h>
40 #include <sys/contract/device_impl.h>
41 #include <sys/cmn_err.h>
42 #include <sys/nvpair.h>
43 #include <sys/policy.h>
44 #include <sys/ddi_impldefs.h>
45 #include <sys/ddi_implfuncs.h>
46 #include <sys/systm.h>
47 #include <sys/stat.h>
48 #include <sys/sunddi.h>
49 #include <sys/esunddi.h>
50 #include <sys/ddi.h>
51 #include <sys/fs/dv_node.h>
52 #include <sys/sunndi.h>
53 #undef ct_lock	/* needed because clnt.h defines ct_lock as a macro */
54 
55 /*
56  * Device Contracts
57  * -----------------
58  * This file contains the core code for the device contracts framework.
59  * A device contract is an agreement or a contract between a process and
60  * the kernel regarding the state of the device. A device contract may be
61  * created when a relationship is formed between a device and a process
62  * i.e. at open(2) time, or it may be created at some point after the device
63  * has been opened. A device contract once formed may be broken by either party.
64  * A device contract can be broken by the process by an explicit abandon of the
65  * contract or by an implicit abandon when the process exits. A device contract
66  * can be broken by the kernel either asynchronously (without negotiation) or
67  * synchronously (with negotiation). Exactly which happens depends on the device
68  * state transition. The following state diagram shows the transitions between
69  * device states. Only device state transitions currently supported by device
70  * contracts is shown.
71  *
72  *                              <-- A -->
73  *                       /-----------------> DEGRADED
74  *                       |                      |
75  *                       |                      |
76  *                       |                      | S
77  *                       |                      | |
78  *                       |                      | v
79  *                       v       S -->          v
80  *                      ONLINE ------------> OFFLINE
81  *
82  *
83  * In the figure above, the arrows indicate the direction of transition. The
84  * letter S refers to transitions which are inherently synchronous i.e.
85  * require negotiation and the letter A indicates transitions which are
86  * asynchronous i.e. are done without contract negotiations. A good example
87  * of a synchronous transition is the ONLINE -> OFFLINE transition. This
88  * transition cannot happen as long as there are consumers which have the
89  * device open. Thus some form of negotiation needs to happen between the
90  * consumers and the kernel to ensure that consumers either close devices
91  * or disallow the move to OFFLINE. Certain other transitions such as
92  * ONLINE --> DEGRADED for example, are inherently asynchronous i.e.
93  * non-negotiable. A device that suffers a fault that degrades its
94  * capabilities will become degraded irrespective of what consumers it has,
95  * so a negotiation in this case is pointless.
96  *
97  * The following device states are currently defined for device contracts:
98  *
99  *      CT_DEV_EV_ONLINE
100  *              The device is online and functioning normally
101  *      CT_DEV_EV_DEGRADED
102  *              The device is online but is functioning in a degraded capacity
103  *      CT_DEV_EV_OFFLINE
104  *              The device is offline and is no longer configured
105  *
106  * A typical consumer of device contracts starts out with a contract
107  * template and adds terms to that template. These include the
108  * "acceptable set" (A-set) term, which is a bitset of device states which
109  * are guaranteed by the contract. If the device moves out of a state in
110  * the A-set, the contract is broken. The breaking of the contract can
111  * be asynchronous in which case a critical contract event is sent to the
112  * contract holder but no negotiations take place. If the breaking of the
113  * contract is synchronous, negotations are opened between the affected
114  * consumer and the kernel. The kernel does this by sending a critical
115  * event to the consumer with the CTE_NEG flag set indicating that this
116  * is a negotiation event. The consumer can accept this change by sending
117  * a ACK message to the kernel. Alternatively, if it has the necessary
118  * privileges, it can send a NACK message to the kernel which will block
119  * the device state change. To NACK a negotiable event, a process must
120  * have the {PRIV_SYS_DEVICES} privilege asserted in its effective set.
121  *
122  * Other terms include the "minor path" term, specified explicitly if the
123  * contract is not being created at open(2) time or specified implicitly
124  * if the contract is being created at open time via an activated template.
125  *
126  * A contract event is sent on any state change to which the contract
127  * owner has subscribed via the informative or critical event sets. Only
128  * critical events are guaranteed to be delivered. Since all device state
129  * changes are controlled by the kernel and cannot be arbitrarily generated
130  * by a non-privileged user, the {PRIV_CONTRACT_EVENT} privilege does not
131  * need to be asserted in a process's effective set to designate an event as
132  * critical. To ensure privacy, a process must either have the same effective
133  * userid as the contract holder or have the {PRIV_CONTRACT_OBSERVER} privilege
134  * asserted in its effective set in order to observe device contract events
135  * off the device contract type specific endpoint.
136  *
137  * Yet another term available with device contracts is the "non-negotiable"
138  * term. This term is used to pre-specify a NACK to any contract negotiation.
139  * This term is ignored for asynchronous state changes. For example, a
140  * provcess may have the A-set {ONLINE|DEGRADED} and make the contract
141  * non-negotiable. In this case, the device contract framework assumes a
142  * NACK for any transition to OFFLINE and blocks the offline. If the A-set
143  * is {ONLINE} and the non-negotiable term is set, transitions to OFFLINE
144  * are NACKed but transitions to DEGRADE succeed.
145  *
146  * The OFFLINE negotiation (if OFFLINE state is not in the A-set for a contract)
147  * happens just before the I/O framework attempts to offline a device
148  * (i.e. detach a device and set the offline flag so that it cannot be
149  * reattached). A device contract holder is expected to either NACK the offline
150  * (if privileged) or release the device and allow the offline to proceed.
151  *
152  * The DEGRADE contract event (if DEGRADE is not in the A-set for a contract)
153  * is generated just before the I/O framework transitions the device state
154  * to "degraded" (i.e. DEVI_DEVICE_DEGRADED in I/O framework terminology).
155  *
156  * The contract holder is expected to ACK or NACK a negotiation event
157  * within a certain period of time. If the ACK/NACK is not received
158  * within the timeout period, the device contract framework will behave
159  * as if the contract does not exist and will proceed with the event.
160  *
161  * Unlike a process contract a device contract does not need to exist
162  * once it is abandoned, since it does not define a fault boundary. It
163  * merely represents an agreement between a process and the kernel
164  * regarding the state of the device. Once the process has abandoned
165  * the contract (either implicitly via a process exit or explicitly)
166  * the kernel has no reason to retain the contract. As a result
167  * device contracts are neither inheritable nor need to exist in an
168  * orphan state.
169  *
170  * A device unlike a process may exist in multiple contracts and has
171  * a "life" outside a device contract. A device unlike a process
172  * may exist without an associated contract. Unlike a process contract
173  * a device contract may be formed after a binding relationship is
174  * formed between a process and a device.
175  *
176  *	IMPLEMENTATION NOTES
177  *	====================
178  * DATA STRUCTURES
179  * ----------------
180  * 	The heart of the device contracts implementation is the device contract
181  * 	private cont_device_t (or ctd for short) data structure. It encapsulates
182  * 	the generic contract_t data structure and has a number of private
183  *	fields.
184  * 	These include:
185  *		cond_minor: The minor device that is the subject of the contract
186  *		cond_aset:  The bitset of states which are guaranteed by the
187  *			   contract
188  *		cond_noneg: If set, indicates that the result of negotiation has
189  *			    been predefined to be a NACK
190  * 	In addition, there are other device identifiers such the devinfo node,
191  * 	dev_t and spec_type of the minor node. There are also a few fields that
192  * 	are used during negotiation to maintain state. See
193  *		uts/common/sys/contract/device_impl.h
194  * 	for details.
195  * 	The ctd structure represents the device private part of a contract of
196  * 	type "device"
197  *
198  * 	Another data structure used by device contracts is ctmpl_device. It is
199  * 	the device contracts private part of the contract template structure. It
200  *	encapsulates the generic template structure "ct_template_t" and includes
201  *	the following device contract specific fields
202  *		ctd_aset:   The bitset of states that should be guaranteed by a
203  *			    contract
204  *		ctd_noneg:  If set, indicates that contract should NACK a
205  *			    negotiation
206  *		ctd_minor:  The devfs_path (without the /devices prefix) of the
207  *			    minor node that is the subject of the contract.
208  *
209  * ALGORITHMS
210  * ---------
211  * There are three sets of routines in this file
212  * 	Template related routines
213  * 	-------------------------
214  *	These routines provide support for template related operations initated
215  *	via the generic template operations. These include routines that dup
216  *	a template, free it, and set various terms in the template
217  *	(such as the minor node path, the acceptable state set (or A-set)
218  *	and the non-negotiable term) as well as a routine to query the
219  *	device specific portion of the template for the abovementioned terms.
220  *	There is also a routine to create (ctmpl_device_create) that is used to
221  *	create a contract from a template. This routine calls (after initial
222  *	setup) the common function used to create a device contract
223  *	(contract_device_create).
224  *
225  *	core device contract implementation
226  *	----------------------------------
227  *	These routines support the generic contract framework to provide
228  *	functionality that allows contracts to be created, managed and
229  *	destroyed. The contract_device_create() routine is a routine used
230  *	to create a contract from a template (either via an explicit create
231  *	operation on a template or implicitly via an open with an
232  *	activated template.). The contract_device_free() routine assists
233  *	in freeing the device contract specific parts. There are routines
234  *	used to abandon (contract_device_abandon) a device contract as well
235  *	as a routine to destroy (which despite its name does not destroy,
236  *	it only moves a contract to a dead state) a contract.
237  *	There is also a routine to return status information about a
238  *	contract - the level of detail depends on what is requested by the
239  *	user. A value of CTD_FIXED only returns fixed length fields such
240  *	as the A-set, state of device and value of the "noneg" term. If
241  *	CTD_ALL is specified, the minor node path is returned as well.
242  *
243  *	In addition there are interfaces (contract_device_ack/nack) which
244  *	are used to support negotiation between userland processes and
245  *	device contracts. These interfaces record the acknowledgement
246  *	or lack thereof for negotiation events and help determine if the
247  *	negotiated event should occur.
248  *
249  *	"backend routines"
250  *	-----------------
251  *	The backend routines form the interface between the I/O framework
252  *	and the device contract subsystem. These routines, allow the I/O
253  *	framework to call into the device contract subsystem to notify it of
254  *	impending changes to a device state as well as to inform of the
255  *	final disposition of such attempted state changes. Routines in this
256  *	class include contract_device_offline() that indicates an attempt to
257  *	offline a device, contract_device_degrade() that indicates that
258  *	a device is moving to the degraded state and contract_device_negend()
259  *	that is used by the I/O framework to inform the contracts subsystem of
260  *	the final disposition of an attempted operation.
261  *
262  *	SUMMARY
263  *	-------
264  *      A contract starts its life as a template. A process allocates a device
265  *	contract template and sets various terms:
266  *		The A-set
267  *		The device minor node
268  *		Critical and informative events
269  *		The noneg i.e. no negotition term
270  *	Setting of these terms in the template is done via the
271  *	ctmpl_device_set() entry point in this file. A process can query a
272  *	template to determine the terms already set in the template - this is
273  *	facilitated by the ctmpl_device_get() routine.
274  *
275  *	Once all the appropriate terms are set, the contract is instantiated via
276  *	one of two methods
277  *	- via an explicit create operation - this is facilitated by the
278  *	  ctmpl_device_create() entry point
279  *	- synchronously with the open(2) system call - this is achieved via the
280  *	  contract_device_open() routine.
281  *	The core work for both these above functions is done by
282  *	contract_device_create()
283  *
284  *	A contract once created can be queried for its status. Support for
285  *	status info is provided by both the common contracts framework and by
286  *	the "device" contract type. If the level of detail requested is
287  *	CTD_COMMON, only the common contract framework data is used. Higher
288  *	levels of detail result in calls to contract_device_status() to supply
289  *	device contract type specific status information.
290  *
291  *	A contract once created may be abandoned either explicitly or implictly.
292  *	In either case, the contract_device_abandon() function is invoked. This
293  * 	function merely calls contract_destroy() which moves the contract to
294  *	the DEAD state. The device contract portion of destroy processing is
295  *	provided by contract_device_destroy() which merely disassociates the
296  *	contract from its device devinfo node. A contract in the DEAD state is
297  *	not freed. It hanbgs around until all references to the contract are
298  *	gone. When that happens, the contract is finally deallocated. The
299  *	device contract specific portion of the free is done by
300  *	contract_device_free() which finally frees the device contract specific
301  *	data structure (cont_device_t).
302  *
303  *	When a device undergoes a state change, the I/O framework calls the
304  *	corresponding device contract entry point. For example, when a device
305  *	is about to go OFFLINE, the routine contract_device_offline() is
306  *	invoked. Similarly if a device moves to DEGRADED state, the routine
307  *	contract_device_degrade() function is called. These functions call the
308  *	core routine contract_device_publish(). This function determines via
309  *	the function is_sync_neg() whether an event is a synchronous (i.e.
310  *	negotiable) event or not. In the former case contract_device_publish()
311  *	publishes a CTE_NEG event and then waits in wait_for_acks() for ACKs
312  *	and/or NACKs from contract holders. In the latter case, it simply
313  *	publishes the event and does not wait. In the negotiation case, ACKs or
314  *	NACKs from userland consumers results in contract_device_ack_nack()
315  *	being called where the result of the negotiation is recorded in the
316  *	contract data structure. Once all outstanding contract owners have
317  *	responded, the device contract code in wait_for_acks() determines the
318  *	final result of the negotiation. A single NACK overrides all other ACKs
319  *	If there is no NACK, then a single ACK will result in an overall ACK
320  *	result. If there are no ACKs or NACKs, then the result CT_NONE is
321  *	returned back to the I/O framework. Once the event is permitted or
322  *	blocked, the I/O framework proceeds or aborts the state change. The
323  *	I/O framework then calls contract_device_negend() with a result code
324  *	indicating final disposition of the event. This call releases the
325  *	barrier and other state associated with the previous negotiation,
326  *	which permits the next event (if any) to come into the device contract
327  *	framework.
328  *
329  *	Finally, a device that has outstanding contracts may be removed from
330  *	the system which results in its devinfo node being freed. The devinfo
331  *	free routine in the I/O framework, calls into the device contract
332  *	function - contract_device_remove_dip(). This routine, disassociates
333  *	the dip from all contracts associated with the contract being freed,
334  *	allowing the devinfo node to be freed.
335  *
336  * LOCKING
337  * ---------
338  * 	There are four sets of data that need to be protected by locks
339  *
340  *	i) device contract specific portion of the contract template - This data
341  *	is protected by the template lock ctmpl_lock.
342  *
343  *	ii) device contract specific portion of the contract - This data is
344  *	protected by the contract lock ct_lock
345  *
346  *	iii) The linked list of contracts hanging off a devinfo node - This
347  *	list is protected by the per-devinfo node lock devi_ct_lock
348  *
349  *	iv) Finally there is a barrier, controlled by devi_ct_lock, devi_ct_cv
350  *	and devi_ct_count that controls state changes to a dip
351  *
352  *	The template lock is independent in that none of the other locks in this
353  *	file may be taken while holding the template lock (and vice versa).
354  *
355  *	The remaining three locks have the following lock order
356  *
357  *	devi_ct_lock  -> ct_count barrier ->  ct_lock
358  *
359  */
360 
361 static cont_device_t *contract_device_create(ctmpl_device_t *dtmpl, dev_t dev,
362     int spec_type, proc_t *owner, int *errorp);
363 
364 /* barrier routines */
365 static void ct_barrier_acquire(dev_info_t *dip);
366 static void ct_barrier_release(dev_info_t *dip);
367 static int ct_barrier_held(dev_info_t *dip);
368 static int ct_barrier_empty(dev_info_t *dip);
369 static void ct_barrier_wait_for_release(dev_info_t *dip);
370 static int ct_barrier_wait_for_empty(dev_info_t *dip, int secs);
371 static void ct_barrier_decr(dev_info_t *dip);
372 static void ct_barrier_incr(dev_info_t *dip);
373 
374 ct_type_t *device_type;
375 
376 /*
377  * Macro predicates for determining when events should be sent and how.
378  */
379 #define	EVSENDP(ctd, flag) \
380 	((ctd->cond_contract.ct_ev_info | ctd->cond_contract.ct_ev_crit) & flag)
381 
382 #define	EVINFOP(ctd, flag) \
383 	((ctd->cond_contract.ct_ev_crit & flag) == 0)
384 
385 /*
386  * State transition table showing which transitions are synchronous and which
387  * are not.
388  */
389 struct ct_dev_negtable {
390 	uint_t	st_old;
391 	uint_t	st_new;
392 	uint_t	st_neg;
393 } ct_dev_negtable[] = {
394 	{CT_DEV_EV_ONLINE, CT_DEV_EV_OFFLINE,	1},
395 	{CT_DEV_EV_ONLINE, CT_DEV_EV_DEGRADED,	0},
396 	{CT_DEV_EV_DEGRADED, CT_DEV_EV_ONLINE,	0},
397 	{CT_DEV_EV_DEGRADED, CT_DEV_EV_OFFLINE,	1},
398 	{0}
399 };
400 
401 /*
402  * Device contract template implementation
403  */
404 
405 /*
406  * ctmpl_device_dup
407  *
408  * The device contract template dup entry point.
409  * This simply copies all the fields (generic as well as device contract
410  * specific) fields of the original.
411  */
412 static struct ct_template *
413 ctmpl_device_dup(struct ct_template *template)
414 {
415 	ctmpl_device_t *new;
416 	ctmpl_device_t *old = template->ctmpl_data;
417 	char *buf;
418 	char *minor;
419 
420 	new = kmem_zalloc(sizeof (ctmpl_device_t), KM_SLEEP);
421 	buf = kmem_alloc(MAXPATHLEN, KM_SLEEP);
422 
423 	/*
424 	 * copy generic fields.
425 	 * ctmpl_copy returns with old template lock held
426 	 */
427 	ctmpl_copy(&new->ctd_ctmpl, template);
428 
429 	new->ctd_ctmpl.ctmpl_data = new;
430 	new->ctd_aset = old->ctd_aset;
431 	new->ctd_minor = NULL;
432 	new->ctd_noneg = old->ctd_noneg;
433 
434 	if (old->ctd_minor) {
435 		ASSERT(strlen(old->ctd_minor) + 1 <= MAXPATHLEN);
436 		bcopy(old->ctd_minor, buf, strlen(old->ctd_minor) + 1);
437 	} else {
438 		kmem_free(buf, MAXPATHLEN);
439 		buf = NULL;
440 	}
441 
442 	mutex_exit(&template->ctmpl_lock);
443 	if (buf) {
444 		minor = i_ddi_strdup(buf, KM_SLEEP);
445 		kmem_free(buf, MAXPATHLEN);
446 		buf = NULL;
447 	} else {
448 		minor = NULL;
449 	}
450 	mutex_enter(&template->ctmpl_lock);
451 
452 	if (minor) {
453 		new->ctd_minor = minor;
454 	}
455 
456 	ASSERT(buf == NULL);
457 	return (&new->ctd_ctmpl);
458 }
459 
460 /*
461  * ctmpl_device_free
462  *
463  * The device contract template free entry point.  Just
464  * frees the template.
465  */
466 static void
467 ctmpl_device_free(struct ct_template *template)
468 {
469 	ctmpl_device_t *dtmpl = template->ctmpl_data;
470 
471 	if (dtmpl->ctd_minor)
472 		kmem_free(dtmpl->ctd_minor, strlen(dtmpl->ctd_minor) + 1);
473 
474 	kmem_free(dtmpl, sizeof (ctmpl_device_t));
475 }
476 
477 /*
478  * SAFE_EV is the set of events which a non-privileged process is
479  * allowed to make critical. An unprivileged device contract owner has
480  * no control over when a device changes state, so all device events
481  * can be in the critical set.
482  *
483  * EXCESS tells us if "value", a critical event set, requires
484  * additional privilege. For device contracts EXCESS currently
485  * evaluates to 0.
486  */
487 #define	SAFE_EV		(CT_DEV_ALLEVENT)
488 #define	EXCESS(value)	((value) & ~SAFE_EV)
489 
490 
491 /*
492  * ctmpl_device_set
493  *
494  * The device contract template set entry point. Sets various terms in the
495  * template. The non-negotiable  term can only be set if the process has
496  * the {PRIV_SYS_DEVICES} privilege asserted in its effective set.
497  */
498 static int
499 ctmpl_device_set(struct ct_template *tmpl, ct_kparam_t *kparam,
500     const cred_t *cr)
501 {
502 	ctmpl_device_t *dtmpl = tmpl->ctmpl_data;
503 	ct_param_t *param = &kparam->param;
504 	int error;
505 	dev_info_t *dip;
506 	int spec_type;
507 	uint64_t param_value;
508 	char *str_value;
509 
510 	ASSERT(MUTEX_HELD(&tmpl->ctmpl_lock));
511 
512 	if (param->ctpm_id == CTDP_MINOR) {
513 		str_value = (char *)kparam->ctpm_kbuf;
514 		str_value[param->ctpm_size - 1] = '\0';
515 	} else {
516 		if (param->ctpm_size < sizeof (uint64_t))
517 			return (EINVAL);
518 		param_value = *(uint64_t *)kparam->ctpm_kbuf;
519 	}
520 
521 	switch (param->ctpm_id) {
522 	case CTDP_ACCEPT:
523 		if (param_value & ~CT_DEV_ALLEVENT)
524 			return (EINVAL);
525 		if (param_value == 0)
526 			return (EINVAL);
527 		if (param_value == CT_DEV_ALLEVENT)
528 			return (EINVAL);
529 
530 		dtmpl->ctd_aset = param_value;
531 		break;
532 	case CTDP_NONEG:
533 		if (param_value != CTDP_NONEG_SET &&
534 		    param_value != CTDP_NONEG_CLEAR)
535 			return (EINVAL);
536 
537 		/*
538 		 * only privileged processes can designate a contract
539 		 * non-negotiatble.
540 		 */
541 		if (param_value == CTDP_NONEG_SET &&
542 		    (error = secpolicy_sys_devices(cr)) != 0) {
543 			return (error);
544 		}
545 
546 		dtmpl->ctd_noneg = param_value;
547 		break;
548 
549 	case CTDP_MINOR:
550 		if (*str_value != '/' ||
551 		    strncmp(str_value, "/devices/",
552 		    strlen("/devices/")) == 0 ||
553 		    strstr(str_value, "../devices/") != NULL ||
554 		    strchr(str_value, ':') == NULL) {
555 			return (EINVAL);
556 		}
557 
558 		spec_type = 0;
559 		dip = NULL;
560 		if (resolve_pathname(str_value, &dip, NULL, &spec_type) != 0) {
561 			return (ERANGE);
562 		}
563 		ddi_release_devi(dip);
564 
565 		if (spec_type != S_IFCHR && spec_type != S_IFBLK) {
566 			return (EINVAL);
567 		}
568 
569 		if (dtmpl->ctd_minor != NULL) {
570 			kmem_free(dtmpl->ctd_minor,
571 			    strlen(dtmpl->ctd_minor) + 1);
572 		}
573 		dtmpl->ctd_minor = i_ddi_strdup(str_value, KM_SLEEP);
574 		break;
575 	case CTP_EV_CRITICAL:
576 		/*
577 		 * Currently for device contracts, any event
578 		 * may be added to the critical set. We retain the
579 		 * following code however for future enhancements.
580 		 */
581 		if (EXCESS(param_value) &&
582 		    (error = secpolicy_contract_event(cr)) != 0)
583 			return (error);
584 		tmpl->ctmpl_ev_crit = param_value;
585 		break;
586 	default:
587 		return (EINVAL);
588 	}
589 
590 	return (0);
591 }
592 
593 /*
594  * ctmpl_device_get
595  *
596  * The device contract template get entry point.  Simply fetches and
597  * returns the value of the requested term.
598  */
599 static int
600 ctmpl_device_get(struct ct_template *template, ct_kparam_t *kparam)
601 {
602 	ctmpl_device_t *dtmpl = template->ctmpl_data;
603 	ct_param_t *param = &kparam->param;
604 	uint64_t *param_value = kparam->ctpm_kbuf;
605 
606 	ASSERT(MUTEX_HELD(&template->ctmpl_lock));
607 
608 	if (param->ctpm_id == CTDP_ACCEPT ||
609 	    param->ctpm_id == CTDP_NONEG) {
610 		if (param->ctpm_size < sizeof (uint64_t))
611 			return (EINVAL);
612 		kparam->ret_size = sizeof (uint64_t);
613 	}
614 
615 	switch (param->ctpm_id) {
616 	case CTDP_ACCEPT:
617 		*param_value = dtmpl->ctd_aset;
618 		break;
619 	case CTDP_NONEG:
620 		*param_value = dtmpl->ctd_noneg;
621 		break;
622 	case CTDP_MINOR:
623 		if (dtmpl->ctd_minor) {
624 			kparam->ret_size = strlcpy((char *)kparam->ctpm_kbuf,
625 			    dtmpl->ctd_minor, param->ctpm_size);
626 			kparam->ret_size++;
627 		} else {
628 			return (ENOENT);
629 		}
630 		break;
631 	default:
632 		return (EINVAL);
633 	}
634 
635 	return (0);
636 }
637 
638 /*
639  * Device contract type specific portion of creating a contract using
640  * a specified template
641  */
642 /*ARGSUSED*/
643 int
644 ctmpl_device_create(ct_template_t *template, ctid_t *ctidp)
645 {
646 	ctmpl_device_t *dtmpl;
647 	char *buf;
648 	dev_t dev;
649 	int spec_type;
650 	int error;
651 	cont_device_t *ctd;
652 
653 	if (ctidp == NULL)
654 		return (EINVAL);
655 
656 	buf = kmem_alloc(MAXPATHLEN, KM_SLEEP);
657 
658 	dtmpl = template->ctmpl_data;
659 
660 	mutex_enter(&template->ctmpl_lock);
661 	if (dtmpl->ctd_minor == NULL) {
662 		/* incomplete template */
663 		mutex_exit(&template->ctmpl_lock);
664 		kmem_free(buf, MAXPATHLEN);
665 		return (EINVAL);
666 	} else {
667 		ASSERT(strlen(dtmpl->ctd_minor) < MAXPATHLEN);
668 		bcopy(dtmpl->ctd_minor, buf, strlen(dtmpl->ctd_minor) + 1);
669 	}
670 	mutex_exit(&template->ctmpl_lock);
671 
672 	spec_type = 0;
673 	dev = NODEV;
674 	if (resolve_pathname(buf, NULL, &dev, &spec_type) != 0 ||
675 	    dev == NODEV || dev == DDI_DEV_T_ANY || dev == DDI_DEV_T_NONE ||
676 	    (spec_type != S_IFCHR && spec_type != S_IFBLK)) {
677 		CT_DEBUG((CE_WARN,
678 		    "tmpl_create: failed to find device: %s", buf));
679 		kmem_free(buf, MAXPATHLEN);
680 		return (ERANGE);
681 	}
682 	kmem_free(buf, MAXPATHLEN);
683 
684 	ctd = contract_device_create(template->ctmpl_data,
685 	    dev, spec_type, curproc, &error);
686 
687 	if (ctd == NULL) {
688 		CT_DEBUG((CE_WARN, "Failed to create device contract for "
689 		    "process (%d) with device (devt = %lu, spec_type = %s)",
690 		    curproc->p_pid, dev,
691 		    spec_type == S_IFCHR ? "S_IFCHR" : "S_IFBLK"));
692 		return (error);
693 	}
694 
695 	mutex_enter(&ctd->cond_contract.ct_lock);
696 	*ctidp = ctd->cond_contract.ct_id;
697 	mutex_exit(&ctd->cond_contract.ct_lock);
698 
699 	return (0);
700 }
701 
702 /*
703  * Device contract specific template entry points
704  */
705 static ctmplops_t ctmpl_device_ops = {
706 	ctmpl_device_dup,		/* ctop_dup */
707 	ctmpl_device_free,		/* ctop_free */
708 	ctmpl_device_set,		/* ctop_set */
709 	ctmpl_device_get,		/* ctop_get */
710 	ctmpl_device_create,		/* ctop_create */
711 	CT_DEV_ALLEVENT			/* all device events bitmask */
712 };
713 
714 
715 /*
716  * Device contract implementation
717  */
718 
719 /*
720  * contract_device_default
721  *
722  * The device contract default template entry point.  Creates a
723  * device contract template with a default A-set and no "noneg" ,
724  * with informative degrade events and critical offline events.
725  * There is no default minor path.
726  */
727 static ct_template_t *
728 contract_device_default(void)
729 {
730 	ctmpl_device_t *new;
731 
732 	new = kmem_zalloc(sizeof (ctmpl_device_t), KM_SLEEP);
733 	ctmpl_init(&new->ctd_ctmpl, &ctmpl_device_ops, device_type, new);
734 
735 	new->ctd_aset = CT_DEV_EV_ONLINE | CT_DEV_EV_DEGRADED;
736 	new->ctd_noneg = 0;
737 	new->ctd_ctmpl.ctmpl_ev_info = CT_DEV_EV_DEGRADED;
738 	new->ctd_ctmpl.ctmpl_ev_crit = CT_DEV_EV_OFFLINE;
739 
740 	return (&new->ctd_ctmpl);
741 }
742 
743 /*
744  * contract_device_free
745  *
746  * Destroys the device contract specific portion of a contract and
747  * frees the contract.
748  */
749 static void
750 contract_device_free(contract_t *ct)
751 {
752 	cont_device_t *ctd = ct->ct_data;
753 
754 	ASSERT(ctd->cond_minor);
755 	ASSERT(strlen(ctd->cond_minor) < MAXPATHLEN);
756 	kmem_free(ctd->cond_minor, strlen(ctd->cond_minor) + 1);
757 
758 	ASSERT(ctd->cond_devt != DDI_DEV_T_ANY &&
759 	    ctd->cond_devt != DDI_DEV_T_NONE && ctd->cond_devt != NODEV);
760 
761 	ASSERT(ctd->cond_spec == S_IFBLK || ctd->cond_spec == S_IFCHR);
762 
763 	ASSERT(!(ctd->cond_aset & ~CT_DEV_ALLEVENT));
764 	ASSERT(ctd->cond_noneg == 0 || ctd->cond_noneg == 1);
765 
766 	ASSERT(!(ctd->cond_currev_type & ~CT_DEV_ALLEVENT));
767 	ASSERT(!(ctd->cond_currev_ack & ~(CT_ACK | CT_NACK)));
768 
769 	ASSERT((ctd->cond_currev_id > 0) ^ (ctd->cond_currev_type == 0));
770 	ASSERT((ctd->cond_currev_id > 0) || (ctd->cond_currev_ack == 0));
771 
772 	ASSERT(!list_link_active(&ctd->cond_next));
773 
774 	kmem_free(ctd, sizeof (cont_device_t));
775 }
776 
777 /*
778  * contract_device_abandon
779  *
780  * The device contract abandon entry point.
781  */
782 static void
783 contract_device_abandon(contract_t *ct)
784 {
785 	ASSERT(MUTEX_HELD(&ct->ct_lock));
786 
787 	/*
788 	 * device contracts cannot be inherited or orphaned.
789 	 * Move the contract to the DEAD_STATE. It will be freed
790 	 * once all references to it are gone.
791 	 */
792 	contract_destroy(ct);
793 }
794 
795 /*
796  * contract_device_destroy
797  *
798  * The device contract destroy entry point.
799  * Called from contract_destroy() to do any type specific destroy. Note
800  * that destroy is a misnomer - this does not free the contract, it only
801  * moves it to the dead state. A contract is actually freed via
802  * 	contract_rele() -> contract_dtor(), contop_free()
803  */
804 static void
805 contract_device_destroy(contract_t *ct)
806 {
807 	cont_device_t	*ctd = ct->ct_data;
808 	dev_info_t	*dip = ctd->cond_dip;
809 
810 	ASSERT(MUTEX_HELD(&ct->ct_lock));
811 
812 	if (dip == NULL) {
813 		/*
814 		 * The dip has been removed, this is a dangling contract
815 		 * Check that dip linkages are NULL
816 		 */
817 		ASSERT(!list_link_active(&ctd->cond_next));
818 		CT_DEBUG((CE_NOTE, "contract_device_destroy: contract has no "
819 		    "devinfo node. contract ctid : %d", ct->ct_id));
820 		return;
821 	}
822 
823 	/*
824 	 * Need to have lock order: devi_ct_lock -> ct_count barrier -> ct_lock
825 	 */
826 	mutex_exit(&ct->ct_lock);
827 
828 	/*
829 	 * Waiting for the barrier to be released is strictly speaking not
830 	 * necessary. But it simplifies the implementation of
831 	 * contract_device_publish() by establishing the invariant that
832 	 * device contracts cannot go away during negotiation.
833 	 */
834 	mutex_enter(&(DEVI(dip)->devi_ct_lock));
835 	ct_barrier_wait_for_release(dip);
836 	mutex_enter(&ct->ct_lock);
837 
838 	list_remove(&(DEVI(dip)->devi_ct), ctd);
839 	ctd->cond_dip = NULL; /* no longer linked to dip */
840 	contract_rele(ct);	/* remove hold for dip linkage */
841 
842 	mutex_exit(&ct->ct_lock);
843 	mutex_exit(&(DEVI(dip)->devi_ct_lock));
844 	mutex_enter(&ct->ct_lock);
845 }
846 
847 /*
848  * contract_device_status
849  *
850  * The device contract status entry point. Called when level of "detail"
851  * is either CTD_FIXED or CTD_ALL
852  *
853  */
854 static void
855 contract_device_status(contract_t *ct, zone_t *zone, int detail, nvlist_t *nvl,
856     void *status, model_t model)
857 {
858 	cont_device_t *ctd = ct->ct_data;
859 
860 	ASSERT(detail == CTD_FIXED || detail == CTD_ALL);
861 
862 	mutex_enter(&ct->ct_lock);
863 	contract_status_common(ct, zone, status, model);
864 
865 	/*
866 	 * There's no need to hold the contract lock while accessing static
867 	 * data like aset or noneg. But since we need the lock to access other
868 	 * data like state, we hold it anyway.
869 	 */
870 	VERIFY(nvlist_add_uint32(nvl, CTDS_STATE, ctd->cond_state) == 0);
871 	VERIFY(nvlist_add_uint32(nvl, CTDS_ASET, ctd->cond_aset) == 0);
872 	VERIFY(nvlist_add_uint32(nvl, CTDS_NONEG, ctd->cond_noneg) == 0);
873 
874 	if (detail == CTD_FIXED) {
875 		mutex_exit(&ct->ct_lock);
876 		return;
877 	}
878 
879 	ASSERT(ctd->cond_minor);
880 	VERIFY(nvlist_add_string(nvl, CTDS_MINOR, ctd->cond_minor) == 0);
881 
882 	mutex_exit(&ct->ct_lock);
883 }
884 
885 /*
886  * Converts a result integer into the corresponding string. Used for printing
887  * messages
888  */
889 static char *
890 result_str(uint_t result)
891 {
892 	switch (result) {
893 	case CT_ACK:
894 		return ("CT_ACK");
895 	case CT_NACK:
896 		return ("CT_NACK");
897 	case CT_NONE:
898 		return ("CT_NONE");
899 	default:
900 		return ("UNKNOWN");
901 	}
902 }
903 
904 /*
905  * Converts a device state integer constant into the corresponding string.
906  * Used to print messages.
907  */
908 static char *
909 state_str(uint_t state)
910 {
911 	switch (state) {
912 	case CT_DEV_EV_ONLINE:
913 		return ("ONLINE");
914 	case CT_DEV_EV_DEGRADED:
915 		return ("DEGRADED");
916 	case CT_DEV_EV_OFFLINE:
917 		return ("OFFLINE");
918 	default:
919 		return ("UNKNOWN");
920 	}
921 }
922 
923 /*
924  * Routine that determines if a particular CT_DEV_EV_? event corresponds to a
925  * synchronous state change or not.
926  */
927 static int
928 is_sync_neg(uint_t old, uint_t new)
929 {
930 	int	i;
931 
932 	ASSERT(old & CT_DEV_ALLEVENT);
933 	ASSERT(new & CT_DEV_ALLEVENT);
934 
935 	if (old == new) {
936 		CT_DEBUG((CE_WARN, "is_sync_neg: transition to same state: %s",
937 		    state_str(new)));
938 		return (-2);
939 	}
940 
941 	for (i = 0; ct_dev_negtable[i].st_new != 0; i++) {
942 		if (old == ct_dev_negtable[i].st_old &&
943 		    new == ct_dev_negtable[i].st_new) {
944 			return (ct_dev_negtable[i].st_neg);
945 		}
946 	}
947 
948 	CT_DEBUG((CE_WARN, "is_sync_neg: Unsupported state transition: "
949 	    "old = %s -> new = %s", state_str(old), state_str(new)));
950 
951 	return (-1);
952 }
953 
954 /*
955  * Used to cleanup cached dv_nodes so that when a device is released by
956  * a contract holder, its devinfo node can be successfully detached.
957  */
958 static int
959 contract_device_dvclean(dev_info_t *dip)
960 {
961 	char		*devnm;
962 	dev_info_t	*pdip;
963 	int		error;
964 
965 	ASSERT(dip);
966 
967 	/* pdip can be NULL if we have contracts against the root dip */
968 	pdip = ddi_get_parent(dip);
969 
970 	if (pdip && DEVI_BUSY_OWNED(pdip) || !pdip && DEVI_BUSY_OWNED(dip)) {
971 		char		*path;
972 
973 		path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
974 		(void) ddi_pathname(dip, path);
975 		CT_DEBUG((CE_WARN, "ct_dv_clean: Parent node is busy owned, "
976 		    "device=%s", path));
977 		kmem_free(path, MAXPATHLEN);
978 		return (EDEADLOCK);
979 	}
980 
981 	if (pdip) {
982 		devnm = kmem_alloc(MAXNAMELEN + 1, KM_SLEEP);
983 		(void) ddi_deviname(dip, devnm);
984 		error = devfs_clean(pdip, devnm + 1, DV_CLEAN_FORCE);
985 		kmem_free(devnm, MAXNAMELEN + 1);
986 	} else {
987 		error = devfs_clean(dip, NULL, DV_CLEAN_FORCE);
988 	}
989 
990 	return (error);
991 }
992 
993 /*
994  * Endpoint of a ct_ctl_ack() or ct_ctl_nack() call from userland.
995  * Results in the ACK or NACK being recorded on the dip for one particular
996  * contract. The device contracts framework evaluates the ACK/NACKs for all
997  * contracts against a device to determine if a particular device state change
998  * should be allowed.
999  */
1000 static int
1001 contract_device_ack_nack(contract_t *ct, uint_t evtype, uint64_t evid,
1002     uint_t cmd)
1003 {
1004 	cont_device_t *ctd = ct->ct_data;
1005 	dev_info_t *dip;
1006 	ctid_t	ctid;
1007 	int error;
1008 
1009 	ctid = ct->ct_id;
1010 
1011 	CT_DEBUG((CE_NOTE, "ack_nack: entered: ctid %d", ctid));
1012 
1013 	mutex_enter(&ct->ct_lock);
1014 	CT_DEBUG((CE_NOTE, "ack_nack: contract lock acquired: %d", ctid));
1015 
1016 	dip = ctd->cond_dip;
1017 
1018 	ASSERT(ctd->cond_minor);
1019 	ASSERT(strlen(ctd->cond_minor) < MAXPATHLEN);
1020 
1021 	/*
1022 	 * Negotiation only if new state is not in A-set
1023 	 */
1024 	ASSERT(!(ctd->cond_aset & evtype));
1025 
1026 	/*
1027 	 * Negotiation only if transition is synchronous
1028 	 */
1029 	ASSERT(is_sync_neg(ctd->cond_state, evtype));
1030 
1031 	/*
1032 	 * We shouldn't be negotiating if the "noneg" flag is set
1033 	 */
1034 	ASSERT(!ctd->cond_noneg);
1035 
1036 	if (dip)
1037 		ndi_hold_devi(dip);
1038 
1039 	mutex_exit(&ct->ct_lock);
1040 
1041 	/*
1042 	 * dv_clean only if !NACK and offline state change
1043 	 */
1044 	if (cmd != CT_NACK && evtype == CT_DEV_EV_OFFLINE && dip) {
1045 		CT_DEBUG((CE_NOTE, "ack_nack: dv_clean: %d", ctid));
1046 		error = contract_device_dvclean(dip);
1047 		if (error != 0) {
1048 			CT_DEBUG((CE_NOTE, "ack_nack: dv_clean: failed: %d",
1049 			    ctid));
1050 			ddi_release_devi(dip);
1051 		}
1052 	}
1053 
1054 	mutex_enter(&ct->ct_lock);
1055 
1056 	if (dip)
1057 		ddi_release_devi(dip);
1058 
1059 	if (dip == NULL) {
1060 		if (ctd->cond_currev_id != evid) {
1061 			CT_DEBUG((CE_WARN, "%sACK for non-current event "
1062 			    "(type=%s, id=%llu) on removed device",
1063 			    cmd == CT_NACK ? "N" : "",
1064 			    state_str(evtype), (unsigned long long)evid));
1065 			CT_DEBUG((CE_NOTE, "ack_nack: error: ESRCH, ctid: %d",
1066 			    ctid));
1067 		} else {
1068 			ASSERT(ctd->cond_currev_type == evtype);
1069 			CT_DEBUG((CE_WARN, "contract_ack: no such device: "
1070 			    "ctid: %d", ctid));
1071 		}
1072 		error = (ct->ct_state == CTS_DEAD) ? ESRCH :
1073 		    ((cmd == CT_NACK) ? ETIMEDOUT : 0);
1074 		mutex_exit(&ct->ct_lock);
1075 		return (error);
1076 	}
1077 
1078 	/*
1079 	 * Must follow lock order: devi_ct_lock -> ct_count barrier - >ct_lock
1080 	 */
1081 	mutex_exit(&ct->ct_lock);
1082 
1083 	mutex_enter(&DEVI(dip)->devi_ct_lock);
1084 	mutex_enter(&ct->ct_lock);
1085 	if (ctd->cond_currev_id != evid) {
1086 		char *buf;
1087 		mutex_exit(&ct->ct_lock);
1088 		mutex_exit(&DEVI(dip)->devi_ct_lock);
1089 		ndi_hold_devi(dip);
1090 		buf = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1091 		(void) ddi_pathname(dip, buf);
1092 		ddi_release_devi(dip);
1093 		CT_DEBUG((CE_WARN, "%sACK for non-current event"
1094 		    "(type=%s, id=%llu) on device %s",
1095 		    cmd == CT_NACK ? "N" : "",
1096 		    state_str(evtype), (unsigned long long)evid, buf));
1097 		kmem_free(buf, MAXPATHLEN);
1098 		CT_DEBUG((CE_NOTE, "ack_nack: error: %d, ctid: %d",
1099 		    cmd == CT_NACK ? ETIMEDOUT : 0, ctid));
1100 		return (cmd == CT_ACK ? 0 : ETIMEDOUT);
1101 	}
1102 
1103 	ASSERT(ctd->cond_currev_type == evtype);
1104 	ASSERT(cmd == CT_ACK || cmd == CT_NACK);
1105 
1106 	CT_DEBUG((CE_NOTE, "ack_nack: setting %sACK for ctid: %d",
1107 	    cmd == CT_NACK ? "N" : "", ctid));
1108 
1109 	ctd->cond_currev_ack = cmd;
1110 	mutex_exit(&ct->ct_lock);
1111 
1112 	ct_barrier_decr(dip);
1113 	mutex_exit(&DEVI(dip)->devi_ct_lock);
1114 
1115 	CT_DEBUG((CE_NOTE, "ack_nack: normal exit: ctid: %d", ctid));
1116 
1117 	return (0);
1118 }
1119 
1120 /*
1121  * Invoked when a userland contract holder approves (i.e. ACKs) a state change
1122  */
1123 static int
1124 contract_device_ack(contract_t *ct, uint_t evtype, uint64_t evid)
1125 {
1126 	return (contract_device_ack_nack(ct, evtype, evid, CT_ACK));
1127 }
1128 
1129 /*
1130  * Invoked when a userland contract holder blocks (i.e. NACKs) a state change
1131  */
1132 static int
1133 contract_device_nack(contract_t *ct, uint_t evtype, uint64_t evid)
1134 {
1135 	return (contract_device_ack_nack(ct, evtype, evid, CT_NACK));
1136 }
1137 
1138 /*
1139  * Creates a new contract synchronously with the breaking of an existing
1140  * contract. Currently not supported.
1141  */
1142 /*ARGSUSED*/
1143 static int
1144 contract_device_newct(contract_t *ct)
1145 {
1146 	return (ENOTSUP);
1147 }
1148 
1149 /*
1150  * Core device contract implementation entry points
1151  */
1152 static contops_t contract_device_ops = {
1153 	contract_device_free,		/* contop_free */
1154 	contract_device_abandon,	/* contop_abandon */
1155 	contract_device_destroy,	/* contop_destroy */
1156 	contract_device_status,		/* contop_status */
1157 	contract_device_ack,		/* contop_ack */
1158 	contract_device_nack,		/* contop_nack */
1159 	contract_qack_notsup,		/* contop_qack */
1160 	contract_device_newct		/* contop_newct */
1161 };
1162 
1163 /*
1164  * contract_device_init
1165  *
1166  * Initializes the device contract type.
1167  */
1168 void
1169 contract_device_init(void)
1170 {
1171 	device_type = contract_type_init(CTT_DEVICE, "device",
1172 	    &contract_device_ops, contract_device_default);
1173 }
1174 
1175 /*
1176  * contract_device_create
1177  *
1178  * create a device contract given template "tmpl" and the "owner" process.
1179  * May fail and return NULL if project.max-contracts would have been exceeded.
1180  *
1181  * Common device contract creation routine called for both open-time and
1182  * non-open time device contract creation
1183  */
1184 static cont_device_t *
1185 contract_device_create(ctmpl_device_t *dtmpl, dev_t dev, int spec_type,
1186     proc_t *owner, int *errorp)
1187 {
1188 	cont_device_t *ctd;
1189 	char *minor;
1190 	char *path;
1191 	dev_info_t *dip;
1192 
1193 	ASSERT(dtmpl != NULL);
1194 	ASSERT(dev != NODEV && dev != DDI_DEV_T_ANY && dev != DDI_DEV_T_NONE);
1195 	ASSERT(spec_type == S_IFCHR || spec_type == S_IFBLK);
1196 	ASSERT(errorp);
1197 
1198 	*errorp = 0;
1199 
1200 	path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1201 
1202 	mutex_enter(&dtmpl->ctd_ctmpl.ctmpl_lock);
1203 	ASSERT(strlen(dtmpl->ctd_minor) < MAXPATHLEN);
1204 	bcopy(dtmpl->ctd_minor, path, strlen(dtmpl->ctd_minor) + 1);
1205 	mutex_exit(&dtmpl->ctd_ctmpl.ctmpl_lock);
1206 
1207 	dip = e_ddi_hold_devi_by_path(path, 0);
1208 	if (dip == NULL) {
1209 		cmn_err(CE_WARN, "contract_create: Cannot find devinfo node "
1210 		    "for device path (%s)", path);
1211 		kmem_free(path, MAXPATHLEN);
1212 		*errorp = ERANGE;
1213 		return (NULL);
1214 	}
1215 
1216 	/*
1217 	 * Lock out any parallel contract negotiations
1218 	 */
1219 	mutex_enter(&(DEVI(dip)->devi_ct_lock));
1220 	ct_barrier_acquire(dip);
1221 	mutex_exit(&(DEVI(dip)->devi_ct_lock));
1222 
1223 	minor = i_ddi_strdup(path, KM_SLEEP);
1224 	kmem_free(path, MAXPATHLEN);
1225 
1226 	(void) contract_type_pbundle(device_type, owner);
1227 
1228 	ctd = kmem_zalloc(sizeof (cont_device_t), KM_SLEEP);
1229 
1230 	/*
1231 	 * Only we hold a refernce to this contract. Safe to access
1232 	 * the fields without a ct_lock
1233 	 */
1234 	ctd->cond_minor = minor;
1235 	/*
1236 	 * It is safe to set the dip pointer in the contract
1237 	 * as the contract will always be destroyed before the dip
1238 	 * is released
1239 	 */
1240 	ctd->cond_dip = dip;
1241 	ctd->cond_devt = dev;
1242 	ctd->cond_spec = spec_type;
1243 
1244 	/*
1245 	 * Since we are able to lookup the device, it is either
1246 	 * online or degraded
1247 	 */
1248 	ctd->cond_state = DEVI_IS_DEVICE_DEGRADED(dip) ?
1249 	    CT_DEV_EV_DEGRADED : CT_DEV_EV_ONLINE;
1250 
1251 	mutex_enter(&dtmpl->ctd_ctmpl.ctmpl_lock);
1252 	ctd->cond_aset = dtmpl->ctd_aset;
1253 	ctd->cond_noneg = dtmpl->ctd_noneg;
1254 
1255 	/*
1256 	 * contract_ctor() initailizes the common portion of a contract
1257 	 * contract_dtor() destroys the common portion of a contract
1258 	 */
1259 	if (contract_ctor(&ctd->cond_contract, device_type, &dtmpl->ctd_ctmpl,
1260 	    ctd, 0, owner, B_TRUE)) {
1261 		mutex_exit(&dtmpl->ctd_ctmpl.ctmpl_lock);
1262 		/*
1263 		 * contract_device_free() destroys the type specific
1264 		 * portion of a contract and frees the contract.
1265 		 * The "minor" path and "cred" is a part of the type specific
1266 		 * portion of the contract and will be freed by
1267 		 * contract_device_free()
1268 		 */
1269 		contract_device_free(&ctd->cond_contract);
1270 
1271 		/* release barrier */
1272 		mutex_enter(&(DEVI(dip)->devi_ct_lock));
1273 		ct_barrier_release(dip);
1274 		mutex_exit(&(DEVI(dip)->devi_ct_lock));
1275 
1276 		ddi_release_devi(dip);
1277 		*errorp = EAGAIN;
1278 		return (NULL);
1279 	}
1280 	mutex_exit(&dtmpl->ctd_ctmpl.ctmpl_lock);
1281 
1282 	mutex_enter(&ctd->cond_contract.ct_lock);
1283 	ctd->cond_contract.ct_ntime.ctm_total = CT_DEV_ACKTIME;
1284 	ctd->cond_contract.ct_qtime.ctm_total = CT_DEV_ACKTIME;
1285 	ctd->cond_contract.ct_ntime.ctm_start = -1;
1286 	ctd->cond_contract.ct_qtime.ctm_start = -1;
1287 	mutex_exit(&ctd->cond_contract.ct_lock);
1288 
1289 	/*
1290 	 * Insert device contract into list hanging off the dip
1291 	 * Bump up the ref-count on the contract to reflect this
1292 	 */
1293 	contract_hold(&ctd->cond_contract);
1294 	mutex_enter(&(DEVI(dip)->devi_ct_lock));
1295 	list_insert_tail(&(DEVI(dip)->devi_ct), ctd);
1296 
1297 	/* release barrier */
1298 	ct_barrier_release(dip);
1299 	mutex_exit(&(DEVI(dip)->devi_ct_lock));
1300 
1301 	ddi_release_devi(dip);
1302 
1303 	return (ctd);
1304 }
1305 
1306 /*
1307  * Called when a device is successfully opened to create an open-time contract
1308  * i.e. synchronously with a device open.
1309  */
1310 int
1311 contract_device_open(dev_t dev, int spec_type, contract_t **ctpp)
1312 {
1313 	ctmpl_device_t *dtmpl;
1314 	ct_template_t  *tmpl;
1315 	cont_device_t *ctd;
1316 	char *path;
1317 	klwp_t *lwp;
1318 	int error;
1319 
1320 	if (ctpp)
1321 		*ctpp = NULL;
1322 
1323 	/*
1324 	 * Check if we are in user-context i.e. if we have an lwp
1325 	 */
1326 	lwp = ttolwp(curthread);
1327 	if (lwp == NULL) {
1328 		CT_DEBUG((CE_NOTE, "contract_open: Not user-context"));
1329 		return (0);
1330 	}
1331 
1332 	tmpl = ctmpl_dup(lwp->lwp_ct_active[device_type->ct_type_index]);
1333 	if (tmpl == NULL) {
1334 		return (0);
1335 	}
1336 	dtmpl = tmpl->ctmpl_data;
1337 
1338 	/*
1339 	 * If the user set a minor path in the template before an open,
1340 	 * ignore it. We use the minor path of the actual minor opened.
1341 	 */
1342 	mutex_enter(&tmpl->ctmpl_lock);
1343 	if (dtmpl->ctd_minor != NULL) {
1344 		CT_DEBUG((CE_NOTE, "contract_device_open(): Process %d: "
1345 		    "ignoring device minor path in active template: %s",
1346 		    curproc->p_pid, dtmpl->ctd_minor));
1347 		/*
1348 		 * This is a copy of the actual activated template.
1349 		 * Safe to make changes such as freeing the minor
1350 		 * path in the template.
1351 		 */
1352 		kmem_free(dtmpl->ctd_minor, strlen(dtmpl->ctd_minor) + 1);
1353 		dtmpl->ctd_minor = NULL;
1354 	}
1355 	mutex_exit(&tmpl->ctmpl_lock);
1356 
1357 	path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1358 
1359 	if (ddi_dev_pathname(dev, spec_type, path) != DDI_SUCCESS) {
1360 		CT_DEBUG((CE_NOTE, "contract_device_open(): Failed to derive "
1361 		    "minor path from dev_t,spec {%lu, %d} for process (%d)",
1362 		    dev, spec_type, curproc->p_pid));
1363 		ctmpl_free(tmpl);
1364 		kmem_free(path, MAXPATHLEN);
1365 		return (1);
1366 	}
1367 
1368 	mutex_enter(&tmpl->ctmpl_lock);
1369 	ASSERT(dtmpl->ctd_minor == NULL);
1370 	dtmpl->ctd_minor = path;
1371 	mutex_exit(&tmpl->ctmpl_lock);
1372 
1373 	ctd = contract_device_create(dtmpl, dev, spec_type, curproc, &error);
1374 
1375 	mutex_enter(&tmpl->ctmpl_lock);
1376 	ASSERT(dtmpl->ctd_minor);
1377 	dtmpl->ctd_minor = NULL;
1378 	mutex_exit(&tmpl->ctmpl_lock);
1379 	ctmpl_free(tmpl);
1380 	kmem_free(path, MAXPATHLEN);
1381 
1382 	if (ctd == NULL) {
1383 		cmn_err(CE_NOTE, "contract_device_open(): Failed to "
1384 		    "create device contract for process (%d) holding "
1385 		    "device (devt = %lu, spec_type = %d)",
1386 		    curproc->p_pid, dev, spec_type);
1387 		return (1);
1388 	}
1389 
1390 	if (ctpp) {
1391 		mutex_enter(&ctd->cond_contract.ct_lock);
1392 		*ctpp = &ctd->cond_contract;
1393 		mutex_exit(&ctd->cond_contract.ct_lock);
1394 	}
1395 	return (0);
1396 }
1397 
1398 /*
1399  * Called during contract negotiation by the device contract framework to wait
1400  * for ACKs or NACKs from contract holders. If all responses are not received
1401  * before a specified timeout, this routine times out.
1402  */
1403 static uint_t
1404 wait_for_acks(dev_info_t *dip, dev_t dev, int spec_type, uint_t evtype)
1405 {
1406 	cont_device_t *ctd;
1407 	int timed_out = 0;
1408 	int result = CT_NONE;
1409 	int ack;
1410 	char *f = "wait_for_acks";
1411 
1412 	ASSERT(MUTEX_HELD(&(DEVI(dip)->devi_ct_lock)));
1413 	ASSERT(dip);
1414 	ASSERT(evtype & CT_DEV_ALLEVENT);
1415 	ASSERT(dev != NODEV && dev != DDI_DEV_T_NONE);
1416 	ASSERT((dev == DDI_DEV_T_ANY && spec_type == 0) ||
1417 	    (spec_type == S_IFBLK || spec_type == S_IFCHR));
1418 
1419 	CT_DEBUG((CE_NOTE, "%s: entered: dip: %p", f, (void *)dip));
1420 
1421 	if (ct_barrier_wait_for_empty(dip, CT_DEV_ACKTIME) == -1) {
1422 		/*
1423 		 * some contract owner(s) didn't respond in time
1424 		 */
1425 		CT_DEBUG((CE_NOTE, "%s: timed out: %p", f, (void *)dip));
1426 		timed_out = 1;
1427 	}
1428 
1429 	ack = 0;
1430 	for (ctd = list_head(&(DEVI(dip)->devi_ct)); ctd != NULL;
1431 	    ctd = list_next(&(DEVI(dip)->devi_ct), ctd)) {
1432 
1433 		mutex_enter(&ctd->cond_contract.ct_lock);
1434 
1435 		ASSERT(ctd->cond_dip == dip);
1436 
1437 		if (dev != DDI_DEV_T_ANY && dev != ctd->cond_devt) {
1438 			mutex_exit(&ctd->cond_contract.ct_lock);
1439 			continue;
1440 		}
1441 		if (dev != DDI_DEV_T_ANY && spec_type != ctd->cond_spec) {
1442 			mutex_exit(&ctd->cond_contract.ct_lock);
1443 			continue;
1444 		}
1445 
1446 		/* skip if non-negotiable contract */
1447 		if (ctd->cond_noneg) {
1448 			mutex_exit(&ctd->cond_contract.ct_lock);
1449 			continue;
1450 		}
1451 
1452 		ASSERT(ctd->cond_currev_type == evtype);
1453 		if (ctd->cond_currev_ack == CT_NACK) {
1454 			CT_DEBUG((CE_NOTE, "%s: found a NACK,result = NACK: %p",
1455 			    f, (void *)dip));
1456 			mutex_exit(&ctd->cond_contract.ct_lock);
1457 			return (CT_NACK);
1458 		} else if (ctd->cond_currev_ack == CT_ACK) {
1459 			ack = 1;
1460 			CT_DEBUG((CE_NOTE, "%s: found a ACK: %p",
1461 			    f, (void *)dip));
1462 		}
1463 		mutex_exit(&ctd->cond_contract.ct_lock);
1464 	}
1465 
1466 	if (ack) {
1467 		result = CT_ACK;
1468 		CT_DEBUG((CE_NOTE, "%s: result = ACK, dip=%p", f, (void *)dip));
1469 	} else if (timed_out) {
1470 		result = CT_NONE;
1471 		CT_DEBUG((CE_NOTE, "%s: result = NONE (timed-out), dip=%p",
1472 		    f, (void *)dip));
1473 	} else {
1474 		CT_DEBUG((CE_NOTE, "%s: result = NONE, dip=%p",
1475 		    f, (void *)dip));
1476 	}
1477 
1478 
1479 	return (result);
1480 }
1481 
1482 /*
1483  * Determines the current state of a device (i.e a devinfo node
1484  */
1485 static int
1486 get_state(dev_info_t *dip)
1487 {
1488 	if (DEVI_IS_DEVICE_OFFLINE(dip) || DEVI_IS_DEVICE_DOWN(dip))
1489 		return (CT_DEV_EV_OFFLINE);
1490 	else if (DEVI_IS_DEVICE_DEGRADED(dip))
1491 		return (CT_DEV_EV_DEGRADED);
1492 	else
1493 		return (CT_DEV_EV_ONLINE);
1494 }
1495 
1496 /*
1497  * Sets the current state of a device in a device contract
1498  */
1499 static void
1500 set_cond_state(dev_info_t *dip)
1501 {
1502 	uint_t state = get_state(dip);
1503 	cont_device_t *ctd;
1504 
1505 	/* verify that barrier is held */
1506 	ASSERT(ct_barrier_held(dip));
1507 
1508 	for (ctd = list_head(&(DEVI(dip)->devi_ct)); ctd != NULL;
1509 	    ctd = list_next(&(DEVI(dip)->devi_ct), ctd)) {
1510 		mutex_enter(&ctd->cond_contract.ct_lock);
1511 		ASSERT(ctd->cond_dip == dip);
1512 		ctd->cond_state = state;
1513 		mutex_exit(&ctd->cond_contract.ct_lock);
1514 	}
1515 }
1516 
1517 /*
1518  * Core routine called by event-specific routines when an event occurs.
1519  * Determines if an event should be be published, and if it is to be
1520  * published, whether a negotiation should take place. Also implements
1521  * NEGEND events which publish the final disposition of an event after
1522  * negotiations are complete.
1523  *
1524  * When an event occurs on a minor node, this routine walks the list of
1525  * contracts hanging off a devinfo node and for each contract on the affected
1526  * dip, evaluates the following cases
1527  *
1528  *	a. an event that is synchronous, breaks the contract and NONEG not set
1529  *		- bumps up the outstanding negotiation counts on the dip
1530  *		- marks the dip as undergoing negotiation (devi_ct_neg)
1531  *		- event of type CTE_NEG is published
1532  *	b. an event that is synchronous, breaks the contract and NONEG is set
1533  *		- sets the final result to CT_NACK, event is blocked
1534  *		- does not publish an event
1535  *	c. event is asynchronous and breaks the contract
1536  *		- publishes a critical event irrespect of whether the NONEG
1537  *		  flag is set, since the contract will be broken and contract
1538  *		  owner needs to be informed.
1539  *	d. No contract breakage but the owner has subscribed to the event
1540  *		- publishes the event irrespective of the NONEG event as the
1541  *		  owner has explicitly subscribed to the event.
1542  *	e. NEGEND event
1543  *		- publishes a critical event. Should only be doing this if
1544  *		  if NONEG is not set.
1545  *	f. all other events
1546  *		- Since a contract is not broken and this event has not been
1547  *		  subscribed to, this event does not need to be published for
1548  *		  for this contract.
1549  *
1550  *	Once an event is published, what happens next depends on the type of
1551  *	event:
1552  *
1553  *	a. NEGEND event
1554  *		- cleanup all state associated with the preceding negotiation
1555  *		  and return CT_ACK to the caller of contract_device_publish()
1556  *	b. NACKed event
1557  *		- One or more contracts had the NONEG term, so the event was
1558  *		  blocked. Return CT_NACK to the caller.
1559  *	c. Negotiated event
1560  *		- Call wait_for_acks() to wait for responses from contract
1561  *		holders. The end result is either CT_ACK (event is permitted),
1562  *		CT_NACK (event is blocked) or CT_NONE (no contract owner)
1563  *		responded. This result is returned back to the caller.
1564  *	d. All other events
1565  *		- If the event was asynchronous (i.e. not negotiated) or
1566  *		a contract was not broken return CT_ACK to the caller.
1567  */
1568 static uint_t
1569 contract_device_publish(dev_info_t *dip, dev_t dev, int spec_type,
1570     uint_t evtype, nvlist_t *tnvl)
1571 {
1572 	cont_device_t *ctd;
1573 	uint_t result = CT_NONE;
1574 	uint64_t evid = 0;
1575 	uint64_t nevid = 0;
1576 	char *path = NULL;
1577 	int negend;
1578 	int match;
1579 	int sync = 0;
1580 	contract_t *ct;
1581 	ct_kevent_t *event;
1582 	nvlist_t *nvl;
1583 	int broken = 0;
1584 
1585 	ASSERT(dip);
1586 	ASSERT(dev != NODEV && dev != DDI_DEV_T_NONE);
1587 	ASSERT((dev == DDI_DEV_T_ANY && spec_type == 0) ||
1588 	    (spec_type == S_IFBLK || spec_type == S_IFCHR));
1589 	ASSERT(evtype == 0 || (evtype & CT_DEV_ALLEVENT));
1590 
1591 	/* Is this a synchronous state change ? */
1592 	if (evtype != CT_EV_NEGEND) {
1593 		sync = is_sync_neg(get_state(dip), evtype);
1594 		/* NOP if unsupported transition */
1595 		if (sync == -2 || sync == -1) {
1596 			DEVI(dip)->devi_flags |= DEVI_CT_NOP;
1597 			result = (sync == -2) ? CT_ACK : CT_NONE;
1598 			goto out;
1599 		}
1600 		CT_DEBUG((CE_NOTE, "publish: is%s sync state change",
1601 		    sync ? "" : " not"));
1602 	} else if (DEVI(dip)->devi_flags & DEVI_CT_NOP) {
1603 		DEVI(dip)->devi_flags &= ~DEVI_CT_NOP;
1604 		result = CT_ACK;
1605 		goto out;
1606 	}
1607 
1608 	path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1609 	(void) ddi_pathname(dip, path);
1610 
1611 	mutex_enter(&(DEVI(dip)->devi_ct_lock));
1612 
1613 	/*
1614 	 * Negotiation end - set the state of the device in the contract
1615 	 */
1616 	if (evtype == CT_EV_NEGEND) {
1617 		CT_DEBUG((CE_NOTE, "publish: negend: setting cond state"));
1618 		set_cond_state(dip);
1619 	}
1620 
1621 	/*
1622 	 * If this device didn't go through negotiation, don't publish
1623 	 * a NEGEND event - simply release the barrier to allow other
1624 	 * device events in.
1625 	 */
1626 	negend = 0;
1627 	if (evtype == CT_EV_NEGEND && !DEVI(dip)->devi_ct_neg) {
1628 		CT_DEBUG((CE_NOTE, "publish: no negend reqd. release barrier"));
1629 		ct_barrier_release(dip);
1630 		mutex_exit(&(DEVI(dip)->devi_ct_lock));
1631 		result = CT_ACK;
1632 		goto out;
1633 	} else if (evtype == CT_EV_NEGEND) {
1634 		/*
1635 		 * There are negotiated contract breakages that
1636 		 * need a NEGEND event
1637 		 */
1638 		ASSERT(ct_barrier_held(dip));
1639 		negend = 1;
1640 		CT_DEBUG((CE_NOTE, "publish: setting negend flag"));
1641 	} else {
1642 		/*
1643 		 * This is a new event, not a NEGEND event. Wait for previous
1644 		 * contract events to complete.
1645 		 */
1646 		ct_barrier_acquire(dip);
1647 	}
1648 
1649 
1650 	match = 0;
1651 	for (ctd = list_head(&(DEVI(dip)->devi_ct)); ctd != NULL;
1652 	    ctd = list_next(&(DEVI(dip)->devi_ct), ctd)) {
1653 
1654 		ctid_t ctid;
1655 		size_t len = strlen(path);
1656 
1657 		mutex_enter(&ctd->cond_contract.ct_lock);
1658 
1659 		ASSERT(ctd->cond_dip == dip);
1660 		ASSERT(ctd->cond_minor);
1661 		ASSERT(strncmp(ctd->cond_minor, path, len) == 0 &&
1662 		    ctd->cond_minor[len] == ':');
1663 
1664 		if (dev != DDI_DEV_T_ANY && dev != ctd->cond_devt) {
1665 			mutex_exit(&ctd->cond_contract.ct_lock);
1666 			continue;
1667 		}
1668 		if (dev != DDI_DEV_T_ANY && spec_type != ctd->cond_spec) {
1669 			mutex_exit(&ctd->cond_contract.ct_lock);
1670 			continue;
1671 		}
1672 
1673 		/* We have a matching contract */
1674 		match = 1;
1675 		ctid = ctd->cond_contract.ct_id;
1676 		CT_DEBUG((CE_NOTE, "publish: found matching contract: %d",
1677 		    ctid));
1678 
1679 		/*
1680 		 * There are 4 possible cases
1681 		 * 1. A contract is broken (dev not in acceptable state) and
1682 		 *    the state change is synchronous - start negotiation
1683 		 *    by sending a CTE_NEG critical event.
1684 		 * 2. A contract is broken and the state change is
1685 		 *    asynchronous - just send a critical event and
1686 		 *    break the contract.
1687 		 * 3. Contract is not broken, but consumer has subscribed
1688 		 *    to the event as a critical or informative event
1689 		 *    - just send the appropriate event
1690 		 * 4. contract waiting for negend event - just send the critical
1691 		 *    NEGEND event.
1692 		 */
1693 		broken = 0;
1694 		if (!negend && !(evtype & ctd->cond_aset)) {
1695 			broken = 1;
1696 			CT_DEBUG((CE_NOTE, "publish: Contract broken: %d",
1697 			    ctid));
1698 		}
1699 
1700 		/*
1701 		 * Don't send event if
1702 		 *	- contract is not broken AND
1703 		 *	- contract holder has not subscribed to this event AND
1704 		 *	- contract not waiting for a NEGEND event
1705 		 */
1706 		if (!broken && !EVSENDP(ctd, evtype) &&
1707 		    !ctd->cond_neg) {
1708 			CT_DEBUG((CE_NOTE, "contract_device_publish(): "
1709 			    "contract (%d): no publish reqd: event %d",
1710 			    ctd->cond_contract.ct_id, evtype));
1711 			mutex_exit(&ctd->cond_contract.ct_lock);
1712 			continue;
1713 		}
1714 
1715 		/*
1716 		 * Note: need to kmem_zalloc() the event so mutexes are
1717 		 * initialized automatically
1718 		 */
1719 		ct = &ctd->cond_contract;
1720 		event = kmem_zalloc(sizeof (ct_kevent_t), KM_SLEEP);
1721 		event->cte_type = evtype;
1722 
1723 		if (broken && sync) {
1724 			CT_DEBUG((CE_NOTE, "publish: broken + sync: "
1725 			    "ctid: %d", ctid));
1726 			ASSERT(!negend);
1727 			ASSERT(ctd->cond_currev_id == 0);
1728 			ASSERT(ctd->cond_currev_type == 0);
1729 			ASSERT(ctd->cond_currev_ack == 0);
1730 			ASSERT(ctd->cond_neg == 0);
1731 			if (ctd->cond_noneg) {
1732 				/* Nothing to publish. Event has been blocked */
1733 				CT_DEBUG((CE_NOTE, "publish: sync and noneg:"
1734 				    "not publishing blocked ev: ctid: %d",
1735 				    ctid));
1736 				result = CT_NACK;
1737 				kmem_free(event, sizeof (ct_kevent_t));
1738 				mutex_exit(&ctd->cond_contract.ct_lock);
1739 				continue;
1740 			}
1741 			event->cte_flags = CTE_NEG; /* critical neg. event */
1742 			ctd->cond_currev_type = event->cte_type;
1743 			ct_barrier_incr(dip);
1744 			DEVI(dip)->devi_ct_neg = 1; /* waiting for negend */
1745 			ctd->cond_neg = 1;
1746 		} else if (broken && !sync) {
1747 			CT_DEBUG((CE_NOTE, "publish: broken + async: ctid: %d",
1748 			    ctid));
1749 			ASSERT(!negend);
1750 			ASSERT(ctd->cond_currev_id == 0);
1751 			ASSERT(ctd->cond_currev_type == 0);
1752 			ASSERT(ctd->cond_currev_ack == 0);
1753 			ASSERT(ctd->cond_neg == 0);
1754 			event->cte_flags = 0; /* critical event */
1755 		} else if (EVSENDP(ctd, event->cte_type)) {
1756 			CT_DEBUG((CE_NOTE, "publish: event suscrib: ctid: %d",
1757 			    ctid));
1758 			ASSERT(!negend);
1759 			ASSERT(ctd->cond_currev_id == 0);
1760 			ASSERT(ctd->cond_currev_type == 0);
1761 			ASSERT(ctd->cond_currev_ack == 0);
1762 			ASSERT(ctd->cond_neg == 0);
1763 			event->cte_flags = EVINFOP(ctd, event->cte_type) ?
1764 			    CTE_INFO : 0;
1765 		} else if (ctd->cond_neg) {
1766 			CT_DEBUG((CE_NOTE, "publish: NEGEND: ctid: %d", ctid));
1767 			ASSERT(negend);
1768 			ASSERT(ctd->cond_noneg == 0);
1769 			nevid = ctd->cond_contract.ct_nevent ?
1770 			    ctd->cond_contract.ct_nevent->cte_id : 0;
1771 			ASSERT(ctd->cond_currev_id == nevid);
1772 			event->cte_flags = 0;	/* NEGEND is always critical */
1773 			ctd->cond_currev_id = 0;
1774 			ctd->cond_currev_type = 0;
1775 			ctd->cond_currev_ack = 0;
1776 			ctd->cond_neg = 0;
1777 		} else {
1778 			CT_DEBUG((CE_NOTE, "publish: not publishing event for "
1779 			    "ctid: %d, evtype: %d",
1780 			    ctd->cond_contract.ct_id, event->cte_type));
1781 			ASSERT(!negend);
1782 			ASSERT(ctd->cond_currev_id == 0);
1783 			ASSERT(ctd->cond_currev_type == 0);
1784 			ASSERT(ctd->cond_currev_ack == 0);
1785 			ASSERT(ctd->cond_neg == 0);
1786 			kmem_free(event, sizeof (ct_kevent_t));
1787 			mutex_exit(&ctd->cond_contract.ct_lock);
1788 			continue;
1789 		}
1790 
1791 		nvl = NULL;
1792 		if (tnvl) {
1793 			VERIFY(nvlist_dup(tnvl, &nvl, 0) == 0);
1794 			if (negend) {
1795 				int32_t newct = 0;
1796 				ASSERT(ctd->cond_noneg == 0);
1797 				VERIFY(nvlist_add_uint64(nvl, CTS_NEVID, nevid)
1798 				    == 0);
1799 				VERIFY(nvlist_lookup_int32(nvl, CTS_NEWCT,
1800 				    &newct) == 0);
1801 				VERIFY(nvlist_add_int32(nvl, CTS_NEWCT,
1802 				    newct == 1 ? 0 :
1803 				    ctd->cond_contract.ct_id) == 0);
1804 				CT_DEBUG((CE_NOTE, "publish: negend: ctid: %d "
1805 				    "CTS_NEVID: %llu, CTS_NEWCT: %s",
1806 				    ctid, (unsigned long long)nevid,
1807 				    newct ? "success" : "failure"));
1808 
1809 			}
1810 		}
1811 
1812 		if (ctd->cond_neg) {
1813 			ASSERT(ctd->cond_contract.ct_ntime.ctm_start == -1);
1814 			ASSERT(ctd->cond_contract.ct_qtime.ctm_start == -1);
1815 			ctd->cond_contract.ct_ntime.ctm_start = ddi_get_lbolt();
1816 			ctd->cond_contract.ct_qtime.ctm_start =
1817 			    ctd->cond_contract.ct_ntime.ctm_start;
1818 		}
1819 
1820 		/*
1821 		 * by holding the dip's devi_ct_lock we ensure that
1822 		 * all ACK/NACKs are held up until we have finished
1823 		 * publishing to all contracts.
1824 		 */
1825 		mutex_exit(&ctd->cond_contract.ct_lock);
1826 		evid = cte_publish_all(ct, event, nvl, NULL);
1827 		mutex_enter(&ctd->cond_contract.ct_lock);
1828 
1829 		if (ctd->cond_neg) {
1830 			ASSERT(!negend);
1831 			ASSERT(broken);
1832 			ASSERT(sync);
1833 			ASSERT(!ctd->cond_noneg);
1834 			CT_DEBUG((CE_NOTE, "publish: sync break, setting evid"
1835 			    ": %d", ctid));
1836 			ctd->cond_currev_id = evid;
1837 		} else if (negend) {
1838 			ctd->cond_contract.ct_ntime.ctm_start = -1;
1839 			ctd->cond_contract.ct_qtime.ctm_start = -1;
1840 		}
1841 		mutex_exit(&ctd->cond_contract.ct_lock);
1842 	}
1843 
1844 	/*
1845 	 * If "negend" set counter back to initial state (-1) so that
1846 	 * other events can be published. Also clear the negotiation flag
1847 	 * on dip.
1848 	 *
1849 	 * 0 .. n are used for counting.
1850 	 * -1 indicates counter is available for use.
1851 	 */
1852 	if (negend) {
1853 		/*
1854 		 * devi_ct_count not necessarily 0. We may have
1855 		 * timed out in which case, count will be non-zero.
1856 		 */
1857 		ct_barrier_release(dip);
1858 		DEVI(dip)->devi_ct_neg = 0;
1859 		CT_DEBUG((CE_NOTE, "publish: negend: reset dip state: dip=%p",
1860 		    (void *)dip));
1861 	} else if (DEVI(dip)->devi_ct_neg) {
1862 		ASSERT(match);
1863 		ASSERT(!ct_barrier_empty(dip));
1864 		CT_DEBUG((CE_NOTE, "publish: sync count=%d, dip=%p",
1865 		    DEVI(dip)->devi_ct_count, (void *)dip));
1866 	} else {
1867 		/*
1868 		 * for non-negotiated events or subscribed events or no
1869 		 * matching contracts
1870 		 */
1871 		ASSERT(ct_barrier_empty(dip));
1872 		ASSERT(DEVI(dip)->devi_ct_neg == 0);
1873 		CT_DEBUG((CE_NOTE, "publish: async/non-nego/subscrib/no-match: "
1874 		    "dip=%p", (void *)dip));
1875 
1876 		/*
1877 		 * only this function when called from contract_device_negend()
1878 		 * can reset the counter to READY state i.e. -1. This function
1879 		 * is so called for every event whether a NEGEND event is needed
1880 		 * or not, but the negend event is only published if the event
1881 		 * whose end they signal is a negotiated event for the contract.
1882 		 */
1883 	}
1884 
1885 	if (!match) {
1886 		/* No matching contracts */
1887 		CT_DEBUG((CE_NOTE, "publish: No matching contract"));
1888 		result = CT_NONE;
1889 	} else if (result == CT_NACK) {
1890 		/* a non-negotiable contract exists and this is a neg. event */
1891 		CT_DEBUG((CE_NOTE, "publish: found 1 or more NONEG contract"));
1892 		(void) wait_for_acks(dip, dev, spec_type, evtype);
1893 	} else if (DEVI(dip)->devi_ct_neg) {
1894 		/* one or more contracts going through negotations  */
1895 		CT_DEBUG((CE_NOTE, "publish: sync contract: waiting"));
1896 		result = wait_for_acks(dip, dev, spec_type, evtype);
1897 	} else {
1898 		/* no negotiated contracts or no broken contracts or NEGEND */
1899 		CT_DEBUG((CE_NOTE, "publish: async/no-break/negend"));
1900 		result = CT_ACK;
1901 	}
1902 
1903 	/*
1904 	 * Release the lock only now so that the only point where we
1905 	 * drop the lock is in wait_for_acks(). This is so that we don't
1906 	 * miss cv_signal/cv_broadcast from contract holders
1907 	 */
1908 	CT_DEBUG((CE_NOTE, "publish: dropping devi_ct_lock"));
1909 	mutex_exit(&(DEVI(dip)->devi_ct_lock));
1910 
1911 out:
1912 	if (tnvl)
1913 		nvlist_free(tnvl);
1914 	if (path)
1915 		kmem_free(path, MAXPATHLEN);
1916 
1917 
1918 	CT_DEBUG((CE_NOTE, "publish: result = %s", result_str(result)));
1919 	return (result);
1920 }
1921 
1922 
1923 /*
1924  * contract_device_offline
1925  *
1926  * Event publishing routine called by I/O framework when a device is offlined.
1927  */
1928 ct_ack_t
1929 contract_device_offline(dev_info_t *dip, dev_t dev, int spec_type)
1930 {
1931 	nvlist_t *nvl;
1932 	uint_t result;
1933 	uint_t evtype;
1934 
1935 	VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1936 
1937 	evtype = CT_DEV_EV_OFFLINE;
1938 	result = contract_device_publish(dip, dev, spec_type, evtype, nvl);
1939 
1940 	/*
1941 	 * If a contract offline is NACKED, the framework expects us to call
1942 	 * NEGEND ourselves, since we know the final result
1943 	 */
1944 	if (result == CT_NACK) {
1945 		contract_device_negend(dip, dev, spec_type, CT_EV_FAILURE);
1946 	}
1947 
1948 	return (result);
1949 }
1950 
1951 /*
1952  * contract_device_degrade
1953  *
1954  * Event publishing routine called by I/O framework when a device
1955  * moves to degrade state.
1956  */
1957 /*ARGSUSED*/
1958 void
1959 contract_device_degrade(dev_info_t *dip, dev_t dev, int spec_type)
1960 {
1961 	nvlist_t *nvl;
1962 	uint_t evtype;
1963 
1964 	VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1965 
1966 	evtype = CT_DEV_EV_DEGRADED;
1967 	(void) contract_device_publish(dip, dev, spec_type, evtype, nvl);
1968 }
1969 
1970 /*
1971  * contract_device_undegrade
1972  *
1973  * Event publishing routine called by I/O framework when a device
1974  * moves from degraded state to online state.
1975  */
1976 /*ARGSUSED*/
1977 void
1978 contract_device_undegrade(dev_info_t *dip, dev_t dev, int spec_type)
1979 {
1980 	nvlist_t *nvl;
1981 	uint_t evtype;
1982 
1983 	VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1984 
1985 	evtype = CT_DEV_EV_ONLINE;
1986 	(void) contract_device_publish(dip, dev, spec_type, evtype, nvl);
1987 }
1988 
1989 /*
1990  * For all contracts which have undergone a negotiation (because the device
1991  * moved out of the acceptable state for that contract and the state
1992  * change is synchronous i.e. requires negotiation) this routine publishes
1993  * a CT_EV_NEGEND event with the final disposition of the event.
1994  *
1995  * This event is always a critical event.
1996  */
1997 void
1998 contract_device_negend(dev_info_t *dip, dev_t dev, int spec_type, int result)
1999 {
2000 	nvlist_t *nvl;
2001 	uint_t evtype;
2002 
2003 	ASSERT(result == CT_EV_SUCCESS || result == CT_EV_FAILURE);
2004 
2005 	CT_DEBUG((CE_NOTE, "contract_device_negend(): entered: result: %d, "
2006 	    "dip: %p", result, (void *)dip));
2007 
2008 	VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2009 	VERIFY(nvlist_add_int32(nvl, CTS_NEWCT,
2010 	    result == CT_EV_SUCCESS ? 1 : 0) == 0);
2011 
2012 	evtype = CT_EV_NEGEND;
2013 	(void) contract_device_publish(dip, dev, spec_type, evtype, nvl);
2014 
2015 	CT_DEBUG((CE_NOTE, "contract_device_negend(): exit dip: %p",
2016 	    (void *)dip));
2017 }
2018 
2019 /*
2020  * Wrapper routine called by other subsystems (such as LDI) to start
2021  * negotiations when a synchronous device state change occurs.
2022  * Returns CT_ACK or CT_NACK.
2023  */
2024 ct_ack_t
2025 contract_device_negotiate(dev_info_t *dip, dev_t dev, int spec_type,
2026     uint_t evtype)
2027 {
2028 	int	result;
2029 
2030 	ASSERT(dip);
2031 	ASSERT(dev != NODEV);
2032 	ASSERT(dev != DDI_DEV_T_ANY);
2033 	ASSERT(dev != DDI_DEV_T_NONE);
2034 	ASSERT(spec_type == S_IFBLK || spec_type == S_IFCHR);
2035 
2036 	switch (evtype) {
2037 	case CT_DEV_EV_OFFLINE:
2038 		result = contract_device_offline(dip, dev, spec_type);
2039 		break;
2040 	default:
2041 		cmn_err(CE_PANIC, "contract_device_negotiate(): Negotiation "
2042 		    "not supported: event (%d) for dev_t (%lu) and spec (%d), "
2043 		    "dip (%p)", evtype, dev, spec_type, (void *)dip);
2044 		result = CT_NACK;
2045 		break;
2046 	}
2047 
2048 	return (result);
2049 }
2050 
2051 /*
2052  * A wrapper routine called by other subsystems (such as the LDI) to
2053  * finalize event processing for a state change event. For synchronous
2054  * state changes, this publishes NEGEND events. For asynchronous i.e.
2055  * non-negotiable events this publishes the event.
2056  */
2057 void
2058 contract_device_finalize(dev_info_t *dip, dev_t dev, int spec_type,
2059     uint_t evtype, int ct_result)
2060 {
2061 	ASSERT(dip);
2062 	ASSERT(dev != NODEV);
2063 	ASSERT(dev != DDI_DEV_T_ANY);
2064 	ASSERT(dev != DDI_DEV_T_NONE);
2065 	ASSERT(spec_type == S_IFBLK || spec_type == S_IFCHR);
2066 
2067 	switch (evtype) {
2068 	case CT_DEV_EV_OFFLINE:
2069 		contract_device_negend(dip, dev, spec_type, ct_result);
2070 		break;
2071 	case CT_DEV_EV_DEGRADED:
2072 		contract_device_degrade(dip, dev, spec_type);
2073 		contract_device_negend(dip, dev, spec_type, ct_result);
2074 		break;
2075 	case CT_DEV_EV_ONLINE:
2076 		contract_device_undegrade(dip, dev, spec_type);
2077 		contract_device_negend(dip, dev, spec_type, ct_result);
2078 		break;
2079 	default:
2080 		cmn_err(CE_PANIC, "contract_device_finalize(): Unsupported "
2081 		    "event (%d) for dev_t (%lu) and spec (%d), dip (%p)",
2082 		    evtype, dev, spec_type, (void *)dip);
2083 		break;
2084 	}
2085 }
2086 
2087 /*
2088  * Called by I/O framework when a devinfo node is freed to remove the
2089  * association between a devinfo node and its contracts.
2090  */
2091 void
2092 contract_device_remove_dip(dev_info_t *dip)
2093 {
2094 	cont_device_t *ctd;
2095 	cont_device_t *next;
2096 	contract_t *ct;
2097 
2098 	mutex_enter(&(DEVI(dip)->devi_ct_lock));
2099 	ct_barrier_wait_for_release(dip);
2100 
2101 	for (ctd = list_head(&(DEVI(dip)->devi_ct)); ctd != NULL; ctd = next) {
2102 		next = list_next(&(DEVI(dip)->devi_ct), ctd);
2103 		list_remove(&(DEVI(dip)->devi_ct), ctd);
2104 		ct = &ctd->cond_contract;
2105 		/*
2106 		 * Unlink the dip associated with this contract
2107 		 */
2108 		mutex_enter(&ct->ct_lock);
2109 		ASSERT(ctd->cond_dip == dip);
2110 		ctd->cond_dip = NULL; /* no longer linked to dip */
2111 		contract_rele(ct);	/* remove hold for dip linkage */
2112 		CT_DEBUG((CE_NOTE, "ct: remove_dip: removed dip from contract: "
2113 		    "ctid: %d", ct->ct_id));
2114 		mutex_exit(&ct->ct_lock);
2115 	}
2116 	ASSERT(list_is_empty(&(DEVI(dip)->devi_ct)));
2117 	mutex_exit(&(DEVI(dip)->devi_ct_lock));
2118 }
2119 
2120 /*
2121  * Barrier related routines
2122  */
2123 static void
2124 ct_barrier_acquire(dev_info_t *dip)
2125 {
2126 	ASSERT(MUTEX_HELD(&(DEVI(dip)->devi_ct_lock)));
2127 	CT_DEBUG((CE_NOTE, "ct_barrier_acquire: waiting for barrier"));
2128 	while (DEVI(dip)->devi_ct_count != -1)
2129 		cv_wait(&(DEVI(dip)->devi_ct_cv), &(DEVI(dip)->devi_ct_lock));
2130 	DEVI(dip)->devi_ct_count = 0;
2131 	CT_DEBUG((CE_NOTE, "ct_barrier_acquire: thread owns barrier"));
2132 }
2133 
2134 static void
2135 ct_barrier_release(dev_info_t *dip)
2136 {
2137 	ASSERT(MUTEX_HELD(&(DEVI(dip)->devi_ct_lock)));
2138 	ASSERT(DEVI(dip)->devi_ct_count != -1);
2139 	DEVI(dip)->devi_ct_count = -1;
2140 	cv_broadcast(&(DEVI(dip)->devi_ct_cv));
2141 	CT_DEBUG((CE_NOTE, "ct_barrier_release: Released barrier"));
2142 }
2143 
2144 static int
2145 ct_barrier_held(dev_info_t *dip)
2146 {
2147 	ASSERT(MUTEX_HELD(&(DEVI(dip)->devi_ct_lock)));
2148 	return (DEVI(dip)->devi_ct_count != -1);
2149 }
2150 
2151 static int
2152 ct_barrier_empty(dev_info_t *dip)
2153 {
2154 	ASSERT(MUTEX_HELD(&(DEVI(dip)->devi_ct_lock)));
2155 	ASSERT(DEVI(dip)->devi_ct_count != -1);
2156 	return (DEVI(dip)->devi_ct_count == 0);
2157 }
2158 
2159 static void
2160 ct_barrier_wait_for_release(dev_info_t *dip)
2161 {
2162 	ASSERT(MUTEX_HELD(&(DEVI(dip)->devi_ct_lock)));
2163 	while (DEVI(dip)->devi_ct_count != -1)
2164 		cv_wait(&(DEVI(dip)->devi_ct_cv), &(DEVI(dip)->devi_ct_lock));
2165 }
2166 
2167 static void
2168 ct_barrier_decr(dev_info_t *dip)
2169 {
2170 	CT_DEBUG((CE_NOTE, "barrier_decr:  ct_count before decr: %d",
2171 	    DEVI(dip)->devi_ct_count));
2172 
2173 	ASSERT(MUTEX_HELD(&(DEVI(dip)->devi_ct_lock)));
2174 	ASSERT(DEVI(dip)->devi_ct_count > 0);
2175 
2176 	DEVI(dip)->devi_ct_count--;
2177 	if (DEVI(dip)->devi_ct_count == 0) {
2178 		cv_broadcast(&DEVI(dip)->devi_ct_cv);
2179 		CT_DEBUG((CE_NOTE, "barrier_decr: cv_broadcast"));
2180 	}
2181 }
2182 
2183 static void
2184 ct_barrier_incr(dev_info_t *dip)
2185 {
2186 	ASSERT(ct_barrier_held(dip));
2187 	DEVI(dip)->devi_ct_count++;
2188 }
2189 
2190 static int
2191 ct_barrier_wait_for_empty(dev_info_t *dip, int secs)
2192 {
2193 	clock_t abstime;
2194 
2195 	ASSERT(MUTEX_HELD(&(DEVI(dip)->devi_ct_lock)));
2196 
2197 	abstime = ddi_get_lbolt() + drv_usectohz(secs*1000000);
2198 	while (DEVI(dip)->devi_ct_count) {
2199 		if (cv_timedwait(&(DEVI(dip)->devi_ct_cv),
2200 		    &(DEVI(dip)->devi_ct_lock), abstime) == -1) {
2201 			return (-1);
2202 		}
2203 	}
2204 	return (0);
2205 }
2206