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 (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  23  */
  24 
  25 #include <sys/mutex.h>
  26 #include <sys/debug.h>
  27 #include <sys/types.h>
  28 #include <sys/param.h>
  29 #include <sys/kmem.h>
  30 #include <sys/thread.h>
  31 #include <sys/id_space.h>
  32 #include <sys/avl.h>
  33 #include <sys/list.h>
  34 #include <sys/sysmacros.h>
  35 #include <sys/proc.h>
  36 #include <sys/contract.h>
  37 #include <sys/contract_impl.h>
  38 #include <sys/contract/device.h>
  39 #include <sys/contract/device_impl.h>
  40 #include <sys/cmn_err.h>
  41 #include <sys/nvpair.h>
  42 #include <sys/policy.h>
  43 #include <sys/ddi_impldefs.h>
  44 #include <sys/ddi_implfuncs.h>
  45 #include <sys/systm.h>
  46 #include <sys/stat.h>
  47 #include <sys/sunddi.h>
  48 #include <sys/esunddi.h>
  49 #include <sys/ddi.h>
  50 #include <sys/fs/dv_node.h>
  51 #include <sys/sunndi.h>
  52 #undef ct_lock  /* needed because clnt.h defines ct_lock as a macro */
  53 
  54 /*
  55  * Device Contracts
  56  * -----------------
  57  * This file contains the core code for the device contracts framework.
  58  * A device contract is an agreement or a contract between a process and
  59  * the kernel regarding the state of the device. A device contract may be
  60  * created when a relationship is formed between a device and a process
  61  * i.e. at open(2) time, or it may be created at some point after the device
  62  * has been opened. A device contract once formed may be broken by either party.
  63  * A device contract can be broken by the process by an explicit abandon of the
  64  * contract or by an implicit abandon when the process exits. A device contract
  65  * can be broken by the kernel either asynchronously (without negotiation) or
  66  * synchronously (with negotiation). Exactly which happens depends on the device
  67  * state transition. The following state diagram shows the transitions between
  68  * device states. Only device state transitions currently supported by device
  69  * contracts is shown.
  70  *
  71  *                              <-- A -->
  72  *                       /-----------------> DEGRADED
  73  *                       |                      |
  74  *                       |                      |
  75  *                       |                      | S
  76  *                       |                      | |
  77  *                       |                      | v
  78  *                       v       S -->          v
  79  *                      ONLINE ------------> OFFLINE
  80  *
  81  *
  82  * In the figure above, the arrows indicate the direction of transition. The
  83  * letter S refers to transitions which are inherently synchronous i.e.
  84  * require negotiation and the letter A indicates transitions which are
  85  * asynchronous i.e. are done without contract negotiations. A good example
  86  * of a synchronous transition is the ONLINE -> OFFLINE transition. This
  87  * transition cannot happen as long as there are consumers which have the
  88  * device open. Thus some form of negotiation needs to happen between the
  89  * consumers and the kernel to ensure that consumers either close devices
  90  * or disallow the move to OFFLINE. Certain other transitions such as
  91  * ONLINE --> DEGRADED for example, are inherently asynchronous i.e.
  92  * non-negotiable. A device that suffers a fault that degrades its
  93  * capabilities will become degraded irrespective of what consumers it has,
  94  * so a negotiation in this case is pointless.
  95  *
  96  * The following device states are currently defined for device contracts:
  97  *
  98  *      CT_DEV_EV_ONLINE
  99  *              The device is online and functioning normally
 100  *      CT_DEV_EV_DEGRADED
 101  *              The device is online but is functioning in a degraded capacity
 102  *      CT_DEV_EV_OFFLINE
 103  *              The device is offline and is no longer configured
 104  *
 105  * A typical consumer of device contracts starts out with a contract
 106  * template and adds terms to that template. These include the
 107  * "acceptable set" (A-set) term, which is a bitset of device states which
 108  * are guaranteed by the contract. If the device moves out of a state in
 109  * the A-set, the contract is broken. The breaking of the contract can
 110  * be asynchronous in which case a critical contract event is sent to the
 111  * contract holder but no negotiations take place. If the breaking of the
 112  * contract is synchronous, negotations are opened between the affected
 113  * consumer and the kernel. The kernel does this by sending a critical
 114  * event to the consumer with the CTE_NEG flag set indicating that this
 115  * is a negotiation event. The consumer can accept this change by sending
 116  * a ACK message to the kernel. Alternatively, if it has the necessary
 117  * privileges, it can send a NACK message to the kernel which will block
 118  * the device state change. To NACK a negotiable event, a process must
 119  * have the {PRIV_SYS_DEVICES} privilege asserted in its effective set.
 120  *
 121  * Other terms include the "minor path" term, specified explicitly if the
 122  * contract is not being created at open(2) time or specified implicitly
 123  * if the contract is being created at open time via an activated template.
 124  *
 125  * A contract event is sent on any state change to which the contract
 126  * owner has subscribed via the informative or critical event sets. Only
 127  * critical events are guaranteed to be delivered. Since all device state
 128  * changes are controlled by the kernel and cannot be arbitrarily generated
 129  * by a non-privileged user, the {PRIV_CONTRACT_EVENT} privilege does not
 130  * need to be asserted in a process's effective set to designate an event as
 131  * critical. To ensure privacy, a process must either have the same effective
 132  * userid as the contract holder or have the {PRIV_CONTRACT_OBSERVER} privilege
 133  * asserted in its effective set in order to observe device contract events
 134  * off the device contract type specific endpoint.
 135  *
 136  * Yet another term available with device contracts is the "non-negotiable"
 137  * term. This term is used to pre-specify a NACK to any contract negotiation.
 138  * This term is ignored for asynchronous state changes. For example, a
 139  * provcess may have the A-set {ONLINE|DEGRADED} and make the contract
 140  * non-negotiable. In this case, the device contract framework assumes a
 141  * NACK for any transition to OFFLINE and blocks the offline. If the A-set
 142  * is {ONLINE} and the non-negotiable term is set, transitions to OFFLINE
 143  * are NACKed but transitions to DEGRADE succeed.
 144  *
 145  * The OFFLINE negotiation (if OFFLINE state is not in the A-set for a contract)
 146  * happens just before the I/O framework attempts to offline a device
 147  * (i.e. detach a device and set the offline flag so that it cannot be
 148  * reattached). A device contract holder is expected to either NACK the offline
 149  * (if privileged) or release the device and allow the offline to proceed.
 150  *
 151  * The DEGRADE contract event (if DEGRADE is not in the A-set for a contract)
 152  * is generated just before the I/O framework transitions the device state
 153  * to "degraded" (i.e. DEVI_DEVICE_DEGRADED in I/O framework terminology).
 154  *
 155  * The contract holder is expected to ACK or NACK a negotiation event
 156  * within a certain period of time. If the ACK/NACK is not received
 157  * within the timeout period, the device contract framework will behave
 158  * as if the contract does not exist and will proceed with the event.
 159  *
 160  * Unlike a process contract a device contract does not need to exist
 161  * once it is abandoned, since it does not define a fault boundary. It
 162  * merely represents an agreement between a process and the kernel
 163  * regarding the state of the device. Once the process has abandoned
 164  * the contract (either implicitly via a process exit or explicitly)
 165  * the kernel has no reason to retain the contract. As a result
 166  * device contracts are neither inheritable nor need to exist in an
 167  * orphan state.
 168  *
 169  * A device unlike a process may exist in multiple contracts and has
 170  * a "life" outside a device contract. A device unlike a process
 171  * may exist without an associated contract. Unlike a process contract
 172  * a device contract may be formed after a binding relationship is
 173  * formed between a process and a device.
 174  *
 175  *      IMPLEMENTATION NOTES
 176  *      ====================
 177  * DATA STRUCTURES
 178  * ----------------
 179  *      The heart of the device contracts implementation is the device contract
 180  *      private cont_device_t (or ctd for short) data structure. It encapsulates
 181  *      the generic contract_t data structure and has a number of private
 182  *      fields.
 183  *      These include:
 184  *              cond_minor: The minor device that is the subject of the contract
 185  *              cond_aset:  The bitset of states which are guaranteed by the
 186  *                         contract
 187  *              cond_noneg: If set, indicates that the result of negotiation has
 188  *                          been predefined to be a NACK
 189  *      In addition, there are other device identifiers such the devinfo node,
 190  *      dev_t and spec_type of the minor node. There are also a few fields that
 191  *      are used during negotiation to maintain state. See
 192  *              uts/common/sys/contract/device_impl.h
 193  *      for details.
 194  *      The ctd structure represents the device private part of a contract of
 195  *      type "device"
 196  *
 197  *      Another data structure used by device contracts is ctmpl_device. It is
 198  *      the device contracts private part of the contract template structure. It
 199  *      encapsulates the generic template structure "ct_template_t" and includes
 200  *      the following device contract specific fields
 201  *              ctd_aset:   The bitset of states that should be guaranteed by a
 202  *                          contract
 203  *              ctd_noneg:  If set, indicates that contract should NACK a
 204  *                          negotiation
 205  *              ctd_minor:  The devfs_path (without the /devices prefix) of the
 206  *                          minor node that is the subject of the contract.
 207  *
 208  * ALGORITHMS
 209  * ---------
 210  * There are three sets of routines in this file
 211  *      Template related routines
 212  *      -------------------------
 213  *      These routines provide support for template related operations initated
 214  *      via the generic template operations. These include routines that dup
 215  *      a template, free it, and set various terms in the template
 216  *      (such as the minor node path, the acceptable state set (or A-set)
 217  *      and the non-negotiable term) as well as a routine to query the
 218  *      device specific portion of the template for the abovementioned terms.
 219  *      There is also a routine to create (ctmpl_device_create) that is used to
 220  *      create a contract from a template. This routine calls (after initial
 221  *      setup) the common function used to create a device contract
 222  *      (contract_device_create).
 223  *
 224  *      core device contract implementation
 225  *      ----------------------------------
 226  *      These routines support the generic contract framework to provide
 227  *      functionality that allows contracts to be created, managed and
 228  *      destroyed. The contract_device_create() routine is a routine used
 229  *      to create a contract from a template (either via an explicit create
 230  *      operation on a template or implicitly via an open with an
 231  *      activated template.). The contract_device_free() routine assists
 232  *      in freeing the device contract specific parts. There are routines
 233  *      used to abandon (contract_device_abandon) a device contract as well
 234  *      as a routine to destroy (which despite its name does not destroy,
 235  *      it only moves a contract to a dead state) a contract.
 236  *      There is also a routine to return status information about a
 237  *      contract - the level of detail depends on what is requested by the
 238  *      user. A value of CTD_FIXED only returns fixed length fields such
 239  *      as the A-set, state of device and value of the "noneg" term. If
 240  *      CTD_ALL is specified, the minor node path is returned as well.
 241  *
 242  *      In addition there are interfaces (contract_device_ack/nack) which
 243  *      are used to support negotiation between userland processes and
 244  *      device contracts. These interfaces record the acknowledgement
 245  *      or lack thereof for negotiation events and help determine if the
 246  *      negotiated event should occur.
 247  *
 248  *      "backend routines"
 249  *      -----------------
 250  *      The backend routines form the interface between the I/O framework
 251  *      and the device contract subsystem. These routines, allow the I/O
 252  *      framework to call into the device contract subsystem to notify it of
 253  *      impending changes to a device state as well as to inform of the
 254  *      final disposition of such attempted state changes. Routines in this
 255  *      class include contract_device_offline() that indicates an attempt to
 256  *      offline a device, contract_device_degrade() that indicates that
 257  *      a device is moving to the degraded state and contract_device_negend()
 258  *      that is used by the I/O framework to inform the contracts subsystem of
 259  *      the final disposition of an attempted operation.
 260  *
 261  *      SUMMARY
 262  *      -------
 263  *      A contract starts its life as a template. A process allocates a device
 264  *      contract template and sets various terms:
 265  *              The A-set
 266  *              The device minor node
 267  *              Critical and informative events
 268  *              The noneg i.e. no negotition term
 269  *      Setting of these terms in the template is done via the
 270  *      ctmpl_device_set() entry point in this file. A process can query a
 271  *      template to determine the terms already set in the template - this is
 272  *      facilitated by the ctmpl_device_get() routine.
 273  *
 274  *      Once all the appropriate terms are set, the contract is instantiated via
 275  *      one of two methods
 276  *      - via an explicit create operation - this is facilitated by the
 277  *        ctmpl_device_create() entry point
 278  *      - synchronously with the open(2) system call - this is achieved via the
 279  *        contract_device_open() routine.
 280  *      The core work for both these above functions is done by
 281  *      contract_device_create()
 282  *
 283  *      A contract once created can be queried for its status. Support for
 284  *      status info is provided by both the common contracts framework and by
 285  *      the "device" contract type. If the level of detail requested is
 286  *      CTD_COMMON, only the common contract framework data is used. Higher
 287  *      levels of detail result in calls to contract_device_status() to supply
 288  *      device contract type specific status information.
 289  *
 290  *      A contract once created may be abandoned either explicitly or implictly.
 291  *      In either case, the contract_device_abandon() function is invoked. This
 292  *      function merely calls contract_destroy() which moves the contract to
 293  *      the DEAD state. The device contract portion of destroy processing is
 294  *      provided by contract_device_destroy() which merely disassociates the
 295  *      contract from its device devinfo node. A contract in the DEAD state is
 296  *      not freed. It hanbgs around until all references to the contract are
 297  *      gone. When that happens, the contract is finally deallocated. The
 298  *      device contract specific portion of the free is done by
 299  *      contract_device_free() which finally frees the device contract specific
 300  *      data structure (cont_device_t).
 301  *
 302  *      When a device undergoes a state change, the I/O framework calls the
 303  *      corresponding device contract entry point. For example, when a device
 304  *      is about to go OFFLINE, the routine contract_device_offline() is
 305  *      invoked. Similarly if a device moves to DEGRADED state, the routine
 306  *      contract_device_degrade() function is called. These functions call the
 307  *      core routine contract_device_publish(). This function determines via
 308  *      the function is_sync_neg() whether an event is a synchronous (i.e.
 309  *      negotiable) event or not. In the former case contract_device_publish()
 310  *      publishes a CTE_NEG event and then waits in wait_for_acks() for ACKs
 311  *      and/or NACKs from contract holders. In the latter case, it simply
 312  *      publishes the event and does not wait. In the negotiation case, ACKs or
 313  *      NACKs from userland consumers results in contract_device_ack_nack()
 314  *      being called where the result of the negotiation is recorded in the
 315  *      contract data structure. Once all outstanding contract owners have
 316  *      responded, the device contract code in wait_for_acks() determines the
 317  *      final result of the negotiation. A single NACK overrides all other ACKs
 318  *      If there is no NACK, then a single ACK will result in an overall ACK
 319  *      result. If there are no ACKs or NACKs, then the result CT_NONE is
 320  *      returned back to the I/O framework. Once the event is permitted or
 321  *      blocked, the I/O framework proceeds or aborts the state change. The
 322  *      I/O framework then calls contract_device_negend() with a result code
 323  *      indicating final disposition of the event. This call releases the
 324  *      barrier and other state associated with the previous negotiation,
 325  *      which permits the next event (if any) to come into the device contract
 326  *      framework.
 327  *
 328  *      Finally, a device that has outstanding contracts may be removed from
 329  *      the system which results in its devinfo node being freed. The devinfo
 330  *      free routine in the I/O framework, calls into the device contract
 331  *      function - contract_device_remove_dip(). This routine, disassociates
 332  *      the dip from all contracts associated with the contract being freed,
 333  *      allowing the devinfo node to be freed.
 334  *
 335  * LOCKING
 336  * ---------
 337  *      There are four sets of data that need to be protected by locks
 338  *
 339  *      i) device contract specific portion of the contract template - This data
 340  *      is protected by the template lock ctmpl_lock.
 341  *
 342  *      ii) device contract specific portion of the contract - This data is
 343  *      protected by the contract lock ct_lock
 344  *
 345  *      iii) The linked list of contracts hanging off a devinfo node - This
 346  *      list is protected by the per-devinfo node lock devi_ct_lock
 347  *
 348  *      iv) Finally there is a barrier, controlled by devi_ct_lock, devi_ct_cv
 349  *      and devi_ct_count that controls state changes to a dip
 350  *
 351  *      The template lock is independent in that none of the other locks in this
 352  *      file may be taken while holding the template lock (and vice versa).
 353  *
 354  *      The remaining three locks have the following lock order
 355  *
 356  *      devi_ct_lock  -> ct_count barrier ->  ct_lock
 357  *
 358  */
 359 
 360 static cont_device_t *contract_device_create(ctmpl_device_t *dtmpl, dev_t dev,
 361     int spec_type, proc_t *owner, int *errorp);
 362 
 363 /* barrier routines */
 364 static void ct_barrier_acquire(dev_info_t *dip);
 365 static void ct_barrier_release(dev_info_t *dip);
 366 static int ct_barrier_held(dev_info_t *dip);
 367 static int ct_barrier_empty(dev_info_t *dip);
 368 static void ct_barrier_wait_for_release(dev_info_t *dip);
 369 static int ct_barrier_wait_for_empty(dev_info_t *dip, int secs);
 370 static void ct_barrier_decr(dev_info_t *dip);
 371 static void ct_barrier_incr(dev_info_t *dip);
 372 
 373 ct_type_t *device_type;
 374 
 375 /*
 376  * Macro predicates for determining when events should be sent and how.
 377  */
 378 #define EVSENDP(ctd, flag) \
 379         ((ctd->cond_contract.ct_ev_info | ctd->cond_contract.ct_ev_crit) & flag)
 380 
 381 #define EVINFOP(ctd, flag) \
 382         ((ctd->cond_contract.ct_ev_crit & flag) == 0)
 383 
 384 /*
 385  * State transition table showing which transitions are synchronous and which
 386  * are not.
 387  */
 388 struct ct_dev_negtable {
 389         uint_t  st_old;
 390         uint_t  st_new;
 391         uint_t  st_neg;
 392 } ct_dev_negtable[] = {
 393         {CT_DEV_EV_ONLINE, CT_DEV_EV_OFFLINE,   1},
 394         {CT_DEV_EV_ONLINE, CT_DEV_EV_DEGRADED,  0},
 395         {CT_DEV_EV_DEGRADED, CT_DEV_EV_ONLINE,  0},
 396         {CT_DEV_EV_DEGRADED, CT_DEV_EV_OFFLINE, 1},
 397         {0}
 398 };
 399 
 400 /*
 401  * Device contract template implementation
 402  */
 403 
 404 /*
 405  * ctmpl_device_dup
 406  *
 407  * The device contract template dup entry point.
 408  * This simply copies all the fields (generic as well as device contract
 409  * specific) fields of the original.
 410  */
 411 static struct ct_template *
 412 ctmpl_device_dup(struct ct_template *template)
 413 {
 414         ctmpl_device_t *new;
 415         ctmpl_device_t *old = template->ctmpl_data;
 416         char *buf;
 417         char *minor;
 418 
 419         new = kmem_zalloc(sizeof (ctmpl_device_t), KM_SLEEP);
 420         buf = kmem_alloc(MAXPATHLEN, KM_SLEEP);
 421 
 422         /*
 423          * copy generic fields.
 424          * ctmpl_copy returns with old template lock held
 425          */
 426         ctmpl_copy(&new->ctd_ctmpl, template);
 427 
 428         new->ctd_ctmpl.ctmpl_data = new;
 429         new->ctd_aset = old->ctd_aset;
 430         new->ctd_minor = NULL;
 431         new->ctd_noneg = old->ctd_noneg;
 432 
 433         if (old->ctd_minor) {
 434                 ASSERT(strlen(old->ctd_minor) + 1 <= MAXPATHLEN);
 435                 bcopy(old->ctd_minor, buf, strlen(old->ctd_minor) + 1);
 436         } else {
 437                 kmem_free(buf, MAXPATHLEN);
 438                 buf = NULL;
 439         }
 440 
 441         mutex_exit(&template->ctmpl_lock);
 442         if (buf) {
 443                 minor = i_ddi_strdup(buf, KM_SLEEP);
 444                 kmem_free(buf, MAXPATHLEN);
 445                 buf = NULL;
 446         } else {
 447                 minor = NULL;
 448         }
 449         mutex_enter(&template->ctmpl_lock);
 450 
 451         if (minor) {
 452                 new->ctd_minor = minor;
 453         }
 454 
 455         ASSERT(buf == NULL);
 456         return (&new->ctd_ctmpl);
 457 }
 458 
 459 /*
 460  * ctmpl_device_free
 461  *
 462  * The device contract template free entry point.  Just
 463  * frees the template.
 464  */
 465 static void
 466 ctmpl_device_free(struct ct_template *template)
 467 {
 468         ctmpl_device_t *dtmpl = template->ctmpl_data;
 469 
 470         if (dtmpl->ctd_minor)
 471                 kmem_free(dtmpl->ctd_minor, strlen(dtmpl->ctd_minor) + 1);
 472 
 473         kmem_free(dtmpl, sizeof (ctmpl_device_t));
 474 }
 475 
 476 /*
 477  * SAFE_EV is the set of events which a non-privileged process is
 478  * allowed to make critical. An unprivileged device contract owner has
 479  * no control over when a device changes state, so all device events
 480  * can be in the critical set.
 481  *
 482  * EXCESS tells us if "value", a critical event set, requires
 483  * additional privilege. For device contracts EXCESS currently
 484  * evaluates to 0.
 485  */
 486 #define SAFE_EV         (CT_DEV_ALLEVENT)
 487 #define EXCESS(value)   ((value) & ~SAFE_EV)
 488 
 489 
 490 /*
 491  * ctmpl_device_set
 492  *
 493  * The device contract template set entry point. Sets various terms in the
 494  * template. The non-negotiable  term can only be set if the process has
 495  * the {PRIV_SYS_DEVICES} privilege asserted in its effective set.
 496  */
 497 static int
 498 ctmpl_device_set(struct ct_template *tmpl, ct_kparam_t *kparam,
 499     const cred_t *cr)
 500 {
 501         ctmpl_device_t *dtmpl = tmpl->ctmpl_data;
 502         ct_param_t *param = &kparam->param;
 503         int error;
 504         dev_info_t *dip;
 505         int spec_type;
 506         uint64_t param_value;
 507         char *str_value;
 508 
 509         ASSERT(MUTEX_HELD(&tmpl->ctmpl_lock));
 510 
 511         if (param->ctpm_id == CTDP_MINOR) {
 512                 str_value = (char *)kparam->ctpm_kbuf;
 513                 str_value[param->ctpm_size - 1] = '\0';
 514         } else {
 515                 if (param->ctpm_size < sizeof (uint64_t))
 516                         return (EINVAL);
 517                 param_value = *(uint64_t *)kparam->ctpm_kbuf;
 518         }
 519 
 520         switch (param->ctpm_id) {
 521         case CTDP_ACCEPT:
 522                 if (param_value & ~CT_DEV_ALLEVENT)
 523                         return (EINVAL);
 524                 if (param_value == 0)
 525                         return (EINVAL);
 526                 if (param_value == CT_DEV_ALLEVENT)
 527                         return (EINVAL);
 528 
 529                 dtmpl->ctd_aset = param_value;
 530                 break;
 531         case CTDP_NONEG:
 532                 if (param_value != CTDP_NONEG_SET &&
 533                     param_value != CTDP_NONEG_CLEAR)
 534                         return (EINVAL);
 535 
 536                 /*
 537                  * only privileged processes can designate a contract
 538                  * non-negotiatble.
 539                  */
 540                 if (param_value == CTDP_NONEG_SET &&
 541                     (error = secpolicy_sys_devices(cr)) != 0) {
 542                         return (error);
 543                 }
 544 
 545                 dtmpl->ctd_noneg = param_value;
 546                 break;
 547 
 548         case CTDP_MINOR:
 549                 if (*str_value != '/' ||
 550                     strncmp(str_value, "/devices/",
 551                     strlen("/devices/")) == 0 ||
 552                     strstr(str_value, "../devices/") != NULL ||
 553                     strchr(str_value, ':') == NULL) {
 554                         return (EINVAL);
 555                 }
 556 
 557                 spec_type = 0;
 558                 dip = NULL;
 559                 if (resolve_pathname(str_value, &dip, NULL, &spec_type) != 0) {
 560                         return (ERANGE);
 561                 }
 562                 ddi_release_devi(dip);
 563 
 564                 if (spec_type != S_IFCHR && spec_type != S_IFBLK) {
 565                         return (EINVAL);
 566                 }
 567 
 568                 if (dtmpl->ctd_minor != NULL) {
 569                         kmem_free(dtmpl->ctd_minor,
 570                             strlen(dtmpl->ctd_minor) + 1);
 571                 }
 572                 dtmpl->ctd_minor = i_ddi_strdup(str_value, KM_SLEEP);
 573                 break;
 574         case CTP_EV_CRITICAL:
 575                 /*
 576                  * Currently for device contracts, any event
 577                  * may be added to the critical set. We retain the
 578                  * following code however for future enhancements.
 579                  */
 580                 if (EXCESS(param_value) &&
 581                     (error = secpolicy_contract_event(cr)) != 0)
 582                         return (error);
 583                 tmpl->ctmpl_ev_crit = param_value;
 584                 break;
 585         default:
 586                 return (EINVAL);
 587         }
 588 
 589         return (0);
 590 }
 591 
 592 /*
 593  * ctmpl_device_get
 594  *
 595  * The device contract template get entry point.  Simply fetches and
 596  * returns the value of the requested term.
 597  */
 598 static int
 599 ctmpl_device_get(struct ct_template *template, ct_kparam_t *kparam)
 600 {
 601         ctmpl_device_t *dtmpl = template->ctmpl_data;
 602         ct_param_t *param = &kparam->param;
 603         uint64_t *param_value = kparam->ctpm_kbuf;
 604 
 605         ASSERT(MUTEX_HELD(&template->ctmpl_lock));
 606 
 607         if (param->ctpm_id == CTDP_ACCEPT ||
 608             param->ctpm_id == CTDP_NONEG) {
 609                 if (param->ctpm_size < sizeof (uint64_t))
 610                         return (EINVAL);
 611                 kparam->ret_size = sizeof (uint64_t);
 612         }
 613 
 614         switch (param->ctpm_id) {
 615         case CTDP_ACCEPT:
 616                 *param_value = dtmpl->ctd_aset;
 617                 break;
 618         case CTDP_NONEG:
 619                 *param_value = dtmpl->ctd_noneg;
 620                 break;
 621         case CTDP_MINOR:
 622                 if (dtmpl->ctd_minor) {
 623                         kparam->ret_size = strlcpy((char *)kparam->ctpm_kbuf,
 624                             dtmpl->ctd_minor, param->ctpm_size);
 625                         kparam->ret_size++;
 626                 } else {
 627                         return (ENOENT);
 628                 }
 629                 break;
 630         default:
 631                 return (EINVAL);
 632         }
 633 
 634         return (0);
 635 }
 636 
 637 /*
 638  * Device contract type specific portion of creating a contract using
 639  * a specified template
 640  */
 641 /*ARGSUSED*/
 642 int
 643 ctmpl_device_create(ct_template_t *template, ctid_t *ctidp)
 644 {
 645         ctmpl_device_t *dtmpl;
 646         char *buf;
 647         dev_t dev;
 648         int spec_type;
 649         int error;
 650         cont_device_t *ctd;
 651 
 652         if (ctidp == NULL)
 653                 return (EINVAL);
 654 
 655         buf = kmem_alloc(MAXPATHLEN, KM_SLEEP);
 656 
 657         dtmpl = template->ctmpl_data;
 658 
 659         mutex_enter(&template->ctmpl_lock);
 660         if (dtmpl->ctd_minor == NULL) {
 661                 /* incomplete template */
 662                 mutex_exit(&template->ctmpl_lock);
 663                 kmem_free(buf, MAXPATHLEN);
 664                 return (EINVAL);
 665         } else {
 666                 ASSERT(strlen(dtmpl->ctd_minor) < MAXPATHLEN);
 667                 bcopy(dtmpl->ctd_minor, buf, strlen(dtmpl->ctd_minor) + 1);
 668         }
 669         mutex_exit(&template->ctmpl_lock);
 670 
 671         spec_type = 0;
 672         dev = NODEV;
 673         if (resolve_pathname(buf, NULL, &dev, &spec_type) != 0 ||
 674             dev == NODEV || dev == DDI_DEV_T_ANY || dev == DDI_DEV_T_NONE ||
 675             (spec_type != S_IFCHR && spec_type != S_IFBLK)) {
 676                 CT_DEBUG((CE_WARN,
 677                     "tmpl_create: failed to find device: %s", buf));
 678                 kmem_free(buf, MAXPATHLEN);
 679                 return (ERANGE);
 680         }
 681         kmem_free(buf, MAXPATHLEN);
 682 
 683         ctd = contract_device_create(template->ctmpl_data,
 684             dev, spec_type, curproc, &error);
 685 
 686         if (ctd == NULL) {
 687                 CT_DEBUG((CE_WARN, "Failed to create device contract for "
 688                     "process (%d) with device (devt = %lu, spec_type = %s)",
 689                     curproc->p_pid, dev,
 690                     spec_type == S_IFCHR ? "S_IFCHR" : "S_IFBLK"));
 691                 return (error);
 692         }
 693 
 694         mutex_enter(&ctd->cond_contract.ct_lock);
 695         *ctidp = ctd->cond_contract.ct_id;
 696         mutex_exit(&ctd->cond_contract.ct_lock);
 697 
 698         return (0);
 699 }
 700 
 701 /*
 702  * Device contract specific template entry points
 703  */
 704 static ctmplops_t ctmpl_device_ops = {
 705         ctmpl_device_dup,               /* ctop_dup */
 706         ctmpl_device_free,              /* ctop_free */
 707         ctmpl_device_set,               /* ctop_set */
 708         ctmpl_device_get,               /* ctop_get */
 709         ctmpl_device_create,            /* ctop_create */
 710         CT_DEV_ALLEVENT                 /* all device events bitmask */
 711 };
 712 
 713 
 714 /*
 715  * Device contract implementation
 716  */
 717 
 718 /*
 719  * contract_device_default
 720  *
 721  * The device contract default template entry point.  Creates a
 722  * device contract template with a default A-set and no "noneg" ,
 723  * with informative degrade events and critical offline events.
 724  * There is no default minor path.
 725  */
 726 static ct_template_t *
 727 contract_device_default(void)
 728 {
 729         ctmpl_device_t *new;
 730 
 731         new = kmem_zalloc(sizeof (ctmpl_device_t), KM_SLEEP);
 732         ctmpl_init(&new->ctd_ctmpl, &ctmpl_device_ops, device_type, new);
 733 
 734         new->ctd_aset = CT_DEV_EV_ONLINE | CT_DEV_EV_DEGRADED;
 735         new->ctd_noneg = 0;
 736         new->ctd_ctmpl.ctmpl_ev_info = CT_DEV_EV_DEGRADED;
 737         new->ctd_ctmpl.ctmpl_ev_crit = CT_DEV_EV_OFFLINE;
 738 
 739         return (&new->ctd_ctmpl);
 740 }
 741 
 742 /*
 743  * contract_device_free
 744  *
 745  * Destroys the device contract specific portion of a contract and
 746  * frees the contract.
 747  */
 748 static void
 749 contract_device_free(contract_t *ct)
 750 {
 751         cont_device_t *ctd = ct->ct_data;
 752 
 753         ASSERT(ctd->cond_minor);
 754         ASSERT(strlen(ctd->cond_minor) < MAXPATHLEN);
 755         kmem_free(ctd->cond_minor, strlen(ctd->cond_minor) + 1);
 756 
 757         ASSERT(ctd->cond_devt != DDI_DEV_T_ANY &&
 758             ctd->cond_devt != DDI_DEV_T_NONE && ctd->cond_devt != NODEV);
 759 
 760         ASSERT(ctd->cond_spec == S_IFBLK || ctd->cond_spec == S_IFCHR);
 761 
 762         ASSERT(!(ctd->cond_aset & ~CT_DEV_ALLEVENT));
 763         ASSERT(ctd->cond_noneg == 0 || ctd->cond_noneg == 1);
 764 
 765         ASSERT(!(ctd->cond_currev_type & ~CT_DEV_ALLEVENT));
 766         ASSERT(!(ctd->cond_currev_ack & ~(CT_ACK | CT_NACK)));
 767 
 768         ASSERT((ctd->cond_currev_id > 0) ^ (ctd->cond_currev_type == 0));
 769         ASSERT((ctd->cond_currev_id > 0) || (ctd->cond_currev_ack == 0));
 770 
 771         ASSERT(!list_link_active(&ctd->cond_next));
 772 
 773         kmem_free(ctd, sizeof (cont_device_t));
 774 }
 775 
 776 /*
 777  * contract_device_abandon
 778  *
 779  * The device contract abandon entry point.
 780  */
 781 static void
 782 contract_device_abandon(contract_t *ct)
 783 {
 784         ASSERT(MUTEX_HELD(&ct->ct_lock));
 785 
 786         /*
 787          * device contracts cannot be inherited or orphaned.
 788          * Move the contract to the DEAD_STATE. It will be freed
 789          * once all references to it are gone.
 790          */
 791         contract_destroy(ct);
 792 }
 793 
 794 /*
 795  * contract_device_destroy
 796  *
 797  * The device contract destroy entry point.
 798  * Called from contract_destroy() to do any type specific destroy. Note
 799  * that destroy is a misnomer - this does not free the contract, it only
 800  * moves it to the dead state. A contract is actually freed via
 801  *      contract_rele() -> contract_dtor(), contop_free()
 802  */
 803 static void
 804 contract_device_destroy(contract_t *ct)
 805 {
 806         cont_device_t   *ctd;
 807         dev_info_t      *dip;
 808 
 809         ASSERT(MUTEX_HELD(&ct->ct_lock));
 810 
 811         for (;;) {
 812                 ctd = ct->ct_data;
 813                 dip = ctd->cond_dip;
 814                 if (dip == NULL) {
 815                         /*
 816                          * The dip has been removed, this is a dangling contract
 817                          * Check that dip linkages are NULL
 818                          */
 819                         ASSERT(!list_link_active(&ctd->cond_next));
 820                         CT_DEBUG((CE_NOTE, "contract_device_destroy:"
 821                             " contract has no devinfo node. contract ctid : %d",
 822                             ct->ct_id));
 823                         return;
 824                 }
 825 
 826                 /*
 827                  * The intended lock order is : devi_ct_lock -> ct_count
 828                  * barrier -> ct_lock.
 829                  * However we can't do this here as dropping the ct_lock allows
 830                  * a race condition with i_ddi_free_node()/
 831                  * contract_device_remove_dip() which may free off dip before
 832                  * we can take devi_ct_lock. So use mutex_tryenter to avoid
 833                  * dropping ct_lock until we have acquired devi_ct_lock.
 834                  */
 835                 if (mutex_tryenter(&(DEVI(dip)->devi_ct_lock)) != 0)
 836                         break;
 837                 mutex_exit(&ct->ct_lock);
 838                 delay(drv_usectohz(1000));
 839                 mutex_enter(&ct->ct_lock);
 840         }
 841         mutex_exit(&ct->ct_lock);
 842 
 843         /*
 844          * Waiting for the barrier to be released is strictly speaking not
 845          * necessary. But it simplifies the implementation of
 846          * contract_device_publish() by establishing the invariant that
 847          * device contracts cannot go away during negotiation.
 848          */
 849         ct_barrier_wait_for_release(dip);
 850         mutex_enter(&ct->ct_lock);
 851 
 852         list_remove(&(DEVI(dip)->devi_ct), ctd);
 853         ctd->cond_dip = NULL; /* no longer linked to dip */
 854         contract_rele(ct);      /* remove hold for dip linkage */
 855 
 856         mutex_exit(&ct->ct_lock);
 857         mutex_exit(&(DEVI(dip)->devi_ct_lock));
 858         mutex_enter(&ct->ct_lock);
 859 }
 860 
 861 /*
 862  * contract_device_status
 863  *
 864  * The device contract status entry point. Called when level of "detail"
 865  * is either CTD_FIXED or CTD_ALL
 866  *
 867  */
 868 static void
 869 contract_device_status(contract_t *ct, zone_t *zone, int detail, nvlist_t *nvl,
 870     void *status, model_t model)
 871 {
 872         cont_device_t *ctd = ct->ct_data;
 873 
 874         ASSERT(detail == CTD_FIXED || detail == CTD_ALL);
 875 
 876         mutex_enter(&ct->ct_lock);
 877         contract_status_common(ct, zone, status, model);
 878 
 879         /*
 880          * There's no need to hold the contract lock while accessing static
 881          * data like aset or noneg. But since we need the lock to access other
 882          * data like state, we hold it anyway.
 883          */
 884         VERIFY(nvlist_add_uint32(nvl, CTDS_STATE, ctd->cond_state) == 0);
 885         VERIFY(nvlist_add_uint32(nvl, CTDS_ASET, ctd->cond_aset) == 0);
 886         VERIFY(nvlist_add_uint32(nvl, CTDS_NONEG, ctd->cond_noneg) == 0);
 887 
 888         if (detail == CTD_FIXED) {
 889                 mutex_exit(&ct->ct_lock);
 890                 return;
 891         }
 892 
 893         ASSERT(ctd->cond_minor);
 894         VERIFY(nvlist_add_string(nvl, CTDS_MINOR, ctd->cond_minor) == 0);
 895 
 896         mutex_exit(&ct->ct_lock);
 897 }
 898 
 899 /*
 900  * Converts a result integer into the corresponding string. Used for printing
 901  * messages
 902  */
 903 static char *
 904 result_str(uint_t result)
 905 {
 906         switch (result) {
 907         case CT_ACK:
 908                 return ("CT_ACK");
 909         case CT_NACK:
 910                 return ("CT_NACK");
 911         case CT_NONE:
 912                 return ("CT_NONE");
 913         default:
 914                 return ("UNKNOWN");
 915         }
 916 }
 917 
 918 /*
 919  * Converts a device state integer constant into the corresponding string.
 920  * Used to print messages.
 921  */
 922 static char *
 923 state_str(uint_t state)
 924 {
 925         switch (state) {
 926         case CT_DEV_EV_ONLINE:
 927                 return ("ONLINE");
 928         case CT_DEV_EV_DEGRADED:
 929                 return ("DEGRADED");
 930         case CT_DEV_EV_OFFLINE:
 931                 return ("OFFLINE");
 932         default:
 933                 return ("UNKNOWN");
 934         }
 935 }
 936 
 937 /*
 938  * Routine that determines if a particular CT_DEV_EV_? event corresponds to a
 939  * synchronous state change or not.
 940  */
 941 static int
 942 is_sync_neg(uint_t old, uint_t new)
 943 {
 944         int     i;
 945 
 946         ASSERT(old & CT_DEV_ALLEVENT);
 947         ASSERT(new & CT_DEV_ALLEVENT);
 948 
 949         if (old == new) {
 950                 CT_DEBUG((CE_WARN, "is_sync_neg: transition to same state: %s",
 951                     state_str(new)));
 952                 return (-2);
 953         }
 954 
 955         for (i = 0; ct_dev_negtable[i].st_new != 0; i++) {
 956                 if (old == ct_dev_negtable[i].st_old &&
 957                     new == ct_dev_negtable[i].st_new) {
 958                         return (ct_dev_negtable[i].st_neg);
 959                 }
 960         }
 961 
 962         CT_DEBUG((CE_WARN, "is_sync_neg: Unsupported state transition: "
 963             "old = %s -> new = %s", state_str(old), state_str(new)));
 964 
 965         return (-1);
 966 }
 967 
 968 /*
 969  * Used to cleanup cached dv_nodes so that when a device is released by
 970  * a contract holder, its devinfo node can be successfully detached.
 971  */
 972 static int
 973 contract_device_dvclean(dev_info_t *dip)
 974 {
 975         char            *devnm;
 976         dev_info_t      *pdip;
 977 
 978         ASSERT(dip);
 979 
 980         /* pdip can be NULL if we have contracts against the root dip */
 981         pdip = ddi_get_parent(dip);
 982 
 983         if (pdip && DEVI_BUSY_OWNED(pdip) || !pdip && DEVI_BUSY_OWNED(dip)) {
 984                 char            *path;
 985 
 986                 path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
 987                 (void) ddi_pathname(dip, path);
 988                 CT_DEBUG((CE_WARN, "ct_dv_clean: Parent node is busy owned, "
 989                     "device=%s", path));
 990                 kmem_free(path, MAXPATHLEN);
 991                 return (EDEADLOCK);
 992         }
 993 
 994         if (pdip) {
 995                 devnm = kmem_alloc(MAXNAMELEN + 1, KM_SLEEP);
 996                 (void) ddi_deviname(dip, devnm);
 997                 (void) devfs_clean(pdip, devnm + 1, DV_CLEAN_FORCE);
 998                 kmem_free(devnm, MAXNAMELEN + 1);
 999         } else {
1000                 (void) devfs_clean(dip, NULL, DV_CLEAN_FORCE);
1001         }
1002 
1003         return (0);
1004 }
1005 
1006 /*
1007  * Endpoint of a ct_ctl_ack() or ct_ctl_nack() call from userland.
1008  * Results in the ACK or NACK being recorded on the dip for one particular
1009  * contract. The device contracts framework evaluates the ACK/NACKs for all
1010  * contracts against a device to determine if a particular device state change
1011  * should be allowed.
1012  */
1013 static int
1014 contract_device_ack_nack(contract_t *ct, uint_t evtype, uint64_t evid,
1015     uint_t cmd)
1016 {
1017         cont_device_t *ctd = ct->ct_data;
1018         dev_info_t *dip;
1019         ctid_t  ctid;
1020         int error;
1021 
1022         ctid = ct->ct_id;
1023 
1024         CT_DEBUG((CE_NOTE, "ack_nack: entered: ctid %d", ctid));
1025 
1026         mutex_enter(&ct->ct_lock);
1027         CT_DEBUG((CE_NOTE, "ack_nack: contract lock acquired: %d", ctid));
1028 
1029         dip = ctd->cond_dip;
1030 
1031         ASSERT(ctd->cond_minor);
1032         ASSERT(strlen(ctd->cond_minor) < MAXPATHLEN);
1033 
1034         /*
1035          * Negotiation only if new state is not in A-set
1036          */
1037         ASSERT(!(ctd->cond_aset & evtype));
1038 
1039         /*
1040          * Negotiation only if transition is synchronous
1041          */
1042         ASSERT(is_sync_neg(ctd->cond_state, evtype));
1043 
1044         /*
1045          * We shouldn't be negotiating if the "noneg" flag is set
1046          */
1047         ASSERT(!ctd->cond_noneg);
1048 
1049         if (dip)
1050                 ndi_hold_devi(dip);
1051 
1052         mutex_exit(&ct->ct_lock);
1053 
1054         /*
1055          * dv_clean only if !NACK and offline state change
1056          */
1057         if (cmd != CT_NACK && evtype == CT_DEV_EV_OFFLINE && dip) {
1058                 CT_DEBUG((CE_NOTE, "ack_nack: dv_clean: %d", ctid));
1059                 error = contract_device_dvclean(dip);
1060                 if (error != 0) {
1061                         CT_DEBUG((CE_NOTE, "ack_nack: dv_clean: failed: %d",
1062                             ctid));
1063                         ddi_release_devi(dip);
1064                 }
1065         }
1066 
1067         mutex_enter(&ct->ct_lock);
1068 
1069         if (dip)
1070                 ddi_release_devi(dip);
1071 
1072         if (dip == NULL) {
1073                 if (ctd->cond_currev_id != evid) {
1074                         CT_DEBUG((CE_WARN, "%sACK for non-current event "
1075                             "(type=%s, id=%llu) on removed device",
1076                             cmd == CT_NACK ? "N" : "",
1077                             state_str(evtype), (unsigned long long)evid));
1078                         CT_DEBUG((CE_NOTE, "ack_nack: error: ESRCH, ctid: %d",
1079                             ctid));
1080                 } else {
1081                         ASSERT(ctd->cond_currev_type == evtype);
1082                         CT_DEBUG((CE_WARN, "contract_ack: no such device: "
1083                             "ctid: %d", ctid));
1084                 }
1085                 error = (ct->ct_state == CTS_DEAD) ? ESRCH :
1086                     ((cmd == CT_NACK) ? ETIMEDOUT : 0);
1087                 mutex_exit(&ct->ct_lock);
1088                 return (error);
1089         }
1090 
1091         /*
1092          * Must follow lock order: devi_ct_lock -> ct_count barrier - >ct_lock
1093          */
1094         mutex_exit(&ct->ct_lock);
1095 
1096         mutex_enter(&DEVI(dip)->devi_ct_lock);
1097         mutex_enter(&ct->ct_lock);
1098         if (ctd->cond_currev_id != evid) {
1099                 char *buf;
1100                 mutex_exit(&ct->ct_lock);
1101                 mutex_exit(&DEVI(dip)->devi_ct_lock);
1102                 ndi_hold_devi(dip);
1103                 buf = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1104                 (void) ddi_pathname(dip, buf);
1105                 ddi_release_devi(dip);
1106                 CT_DEBUG((CE_WARN, "%sACK for non-current event"
1107                     "(type=%s, id=%llu) on device %s",
1108                     cmd == CT_NACK ? "N" : "",
1109                     state_str(evtype), (unsigned long long)evid, buf));
1110                 kmem_free(buf, MAXPATHLEN);
1111                 CT_DEBUG((CE_NOTE, "ack_nack: error: %d, ctid: %d",
1112                     cmd == CT_NACK ? ETIMEDOUT : 0, ctid));
1113                 return (cmd == CT_ACK ? 0 : ETIMEDOUT);
1114         }
1115 
1116         ASSERT(ctd->cond_currev_type == evtype);
1117         ASSERT(cmd == CT_ACK || cmd == CT_NACK);
1118 
1119         CT_DEBUG((CE_NOTE, "ack_nack: setting %sACK for ctid: %d",
1120             cmd == CT_NACK ? "N" : "", ctid));
1121 
1122         ctd->cond_currev_ack = cmd;
1123         mutex_exit(&ct->ct_lock);
1124 
1125         ct_barrier_decr(dip);
1126         mutex_exit(&DEVI(dip)->devi_ct_lock);
1127 
1128         CT_DEBUG((CE_NOTE, "ack_nack: normal exit: ctid: %d", ctid));
1129 
1130         return (0);
1131 }
1132 
1133 /*
1134  * Invoked when a userland contract holder approves (i.e. ACKs) a state change
1135  */
1136 static int
1137 contract_device_ack(contract_t *ct, uint_t evtype, uint64_t evid)
1138 {
1139         return (contract_device_ack_nack(ct, evtype, evid, CT_ACK));
1140 }
1141 
1142 /*
1143  * Invoked when a userland contract holder blocks (i.e. NACKs) a state change
1144  */
1145 static int
1146 contract_device_nack(contract_t *ct, uint_t evtype, uint64_t evid)
1147 {
1148         return (contract_device_ack_nack(ct, evtype, evid, CT_NACK));
1149 }
1150 
1151 /*
1152  * Creates a new contract synchronously with the breaking of an existing
1153  * contract. Currently not supported.
1154  */
1155 /*ARGSUSED*/
1156 static int
1157 contract_device_newct(contract_t *ct)
1158 {
1159         return (ENOTSUP);
1160 }
1161 
1162 /*
1163  * Core device contract implementation entry points
1164  */
1165 static contops_t contract_device_ops = {
1166         contract_device_free,           /* contop_free */
1167         contract_device_abandon,        /* contop_abandon */
1168         contract_device_destroy,        /* contop_destroy */
1169         contract_device_status,         /* contop_status */
1170         contract_device_ack,            /* contop_ack */
1171         contract_device_nack,           /* contop_nack */
1172         contract_qack_notsup,           /* contop_qack */
1173         contract_device_newct           /* contop_newct */
1174 };
1175 
1176 /*
1177  * contract_device_init
1178  *
1179  * Initializes the device contract type.
1180  */
1181 void
1182 contract_device_init(void)
1183 {
1184         device_type = contract_type_init(CTT_DEVICE, "device",
1185             &contract_device_ops, contract_device_default);
1186 }
1187 
1188 /*
1189  * contract_device_create
1190  *
1191  * create a device contract given template "tmpl" and the "owner" process.
1192  * May fail and return NULL if project.max-contracts would have been exceeded.
1193  *
1194  * Common device contract creation routine called for both open-time and
1195  * non-open time device contract creation
1196  */
1197 static cont_device_t *
1198 contract_device_create(ctmpl_device_t *dtmpl, dev_t dev, int spec_type,
1199     proc_t *owner, int *errorp)
1200 {
1201         cont_device_t *ctd;
1202         char *minor;
1203         char *path;
1204         dev_info_t *dip;
1205 
1206         ASSERT(dtmpl != NULL);
1207         ASSERT(dev != NODEV && dev != DDI_DEV_T_ANY && dev != DDI_DEV_T_NONE);
1208         ASSERT(spec_type == S_IFCHR || spec_type == S_IFBLK);
1209         ASSERT(errorp);
1210 
1211         *errorp = 0;
1212 
1213         path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1214 
1215         mutex_enter(&dtmpl->ctd_ctmpl.ctmpl_lock);
1216         ASSERT(strlen(dtmpl->ctd_minor) < MAXPATHLEN);
1217         bcopy(dtmpl->ctd_minor, path, strlen(dtmpl->ctd_minor) + 1);
1218         mutex_exit(&dtmpl->ctd_ctmpl.ctmpl_lock);
1219 
1220         dip = e_ddi_hold_devi_by_path(path, 0);
1221         if (dip == NULL) {
1222                 cmn_err(CE_WARN, "contract_create: Cannot find devinfo node "
1223                     "for device path (%s)", path);
1224                 kmem_free(path, MAXPATHLEN);
1225                 *errorp = ERANGE;
1226                 return (NULL);
1227         }
1228 
1229         /*
1230          * Lock out any parallel contract negotiations
1231          */
1232         mutex_enter(&(DEVI(dip)->devi_ct_lock));
1233         ct_barrier_acquire(dip);
1234         mutex_exit(&(DEVI(dip)->devi_ct_lock));
1235 
1236         minor = i_ddi_strdup(path, KM_SLEEP);
1237         kmem_free(path, MAXPATHLEN);
1238 
1239         (void) contract_type_pbundle(device_type, owner);
1240 
1241         ctd = kmem_zalloc(sizeof (cont_device_t), KM_SLEEP);
1242 
1243         /*
1244          * Only we hold a refernce to this contract. Safe to access
1245          * the fields without a ct_lock
1246          */
1247         ctd->cond_minor = minor;
1248         /*
1249          * It is safe to set the dip pointer in the contract
1250          * as the contract will always be destroyed before the dip
1251          * is released
1252          */
1253         ctd->cond_dip = dip;
1254         ctd->cond_devt = dev;
1255         ctd->cond_spec = spec_type;
1256 
1257         /*
1258          * Since we are able to lookup the device, it is either
1259          * online or degraded
1260          */
1261         ctd->cond_state = DEVI_IS_DEVICE_DEGRADED(dip) ?
1262             CT_DEV_EV_DEGRADED : CT_DEV_EV_ONLINE;
1263 
1264         mutex_enter(&dtmpl->ctd_ctmpl.ctmpl_lock);
1265         ctd->cond_aset = dtmpl->ctd_aset;
1266         ctd->cond_noneg = dtmpl->ctd_noneg;
1267 
1268         /*
1269          * contract_ctor() initailizes the common portion of a contract
1270          * contract_dtor() destroys the common portion of a contract
1271          */
1272         if (contract_ctor(&ctd->cond_contract, device_type, &dtmpl->ctd_ctmpl,
1273             ctd, 0, owner, B_TRUE)) {
1274                 mutex_exit(&dtmpl->ctd_ctmpl.ctmpl_lock);
1275                 /*
1276                  * contract_device_free() destroys the type specific
1277                  * portion of a contract and frees the contract.
1278                  * The "minor" path and "cred" is a part of the type specific
1279                  * portion of the contract and will be freed by
1280                  * contract_device_free()
1281                  */
1282                 contract_device_free(&ctd->cond_contract);
1283 
1284                 /* release barrier */
1285                 mutex_enter(&(DEVI(dip)->devi_ct_lock));
1286                 ct_barrier_release(dip);
1287                 mutex_exit(&(DEVI(dip)->devi_ct_lock));
1288 
1289                 ddi_release_devi(dip);
1290                 *errorp = EAGAIN;
1291                 return (NULL);
1292         }
1293         mutex_exit(&dtmpl->ctd_ctmpl.ctmpl_lock);
1294 
1295         mutex_enter(&ctd->cond_contract.ct_lock);
1296         ctd->cond_contract.ct_ntime.ctm_total = CT_DEV_ACKTIME;
1297         ctd->cond_contract.ct_qtime.ctm_total = CT_DEV_ACKTIME;
1298         ctd->cond_contract.ct_ntime.ctm_start = -1;
1299         ctd->cond_contract.ct_qtime.ctm_start = -1;
1300         mutex_exit(&ctd->cond_contract.ct_lock);
1301 
1302         /*
1303          * Insert device contract into list hanging off the dip
1304          * Bump up the ref-count on the contract to reflect this
1305          */
1306         contract_hold(&ctd->cond_contract);
1307         mutex_enter(&(DEVI(dip)->devi_ct_lock));
1308         list_insert_tail(&(DEVI(dip)->devi_ct), ctd);
1309 
1310         /* release barrier */
1311         ct_barrier_release(dip);
1312         mutex_exit(&(DEVI(dip)->devi_ct_lock));
1313 
1314         ddi_release_devi(dip);
1315 
1316         return (ctd);
1317 }
1318 
1319 /*
1320  * Called when a device is successfully opened to create an open-time contract
1321  * i.e. synchronously with a device open.
1322  */
1323 int
1324 contract_device_open(dev_t dev, int spec_type, contract_t **ctpp)
1325 {
1326         ctmpl_device_t *dtmpl;
1327         ct_template_t  *tmpl;
1328         cont_device_t *ctd;
1329         char *path;
1330         klwp_t *lwp;
1331         int error;
1332 
1333         if (ctpp)
1334                 *ctpp = NULL;
1335 
1336         /*
1337          * Check if we are in user-context i.e. if we have an lwp
1338          */
1339         lwp = ttolwp(curthread);
1340         if (lwp == NULL) {
1341                 CT_DEBUG((CE_NOTE, "contract_open: Not user-context"));
1342                 return (0);
1343         }
1344 
1345         tmpl = ctmpl_dup(lwp->lwp_ct_active[device_type->ct_type_index]);
1346         if (tmpl == NULL) {
1347                 return (0);
1348         }
1349         dtmpl = tmpl->ctmpl_data;
1350 
1351         /*
1352          * If the user set a minor path in the template before an open,
1353          * ignore it. We use the minor path of the actual minor opened.
1354          */
1355         mutex_enter(&tmpl->ctmpl_lock);
1356         if (dtmpl->ctd_minor != NULL) {
1357                 CT_DEBUG((CE_NOTE, "contract_device_open(): Process %d: "
1358                     "ignoring device minor path in active template: %s",
1359                     curproc->p_pid, dtmpl->ctd_minor));
1360                 /*
1361                  * This is a copy of the actual activated template.
1362                  * Safe to make changes such as freeing the minor
1363                  * path in the template.
1364                  */
1365                 kmem_free(dtmpl->ctd_minor, strlen(dtmpl->ctd_minor) + 1);
1366                 dtmpl->ctd_minor = NULL;
1367         }
1368         mutex_exit(&tmpl->ctmpl_lock);
1369 
1370         path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1371 
1372         if (ddi_dev_pathname(dev, spec_type, path) != DDI_SUCCESS) {
1373                 CT_DEBUG((CE_NOTE, "contract_device_open(): Failed to derive "
1374                     "minor path from dev_t,spec {%lu, %d} for process (%d)",
1375                     dev, spec_type, curproc->p_pid));
1376                 ctmpl_free(tmpl);
1377                 kmem_free(path, MAXPATHLEN);
1378                 return (1);
1379         }
1380 
1381         mutex_enter(&tmpl->ctmpl_lock);
1382         ASSERT(dtmpl->ctd_minor == NULL);
1383         dtmpl->ctd_minor = path;
1384         mutex_exit(&tmpl->ctmpl_lock);
1385 
1386         ctd = contract_device_create(dtmpl, dev, spec_type, curproc, &error);
1387 
1388         mutex_enter(&tmpl->ctmpl_lock);
1389         ASSERT(dtmpl->ctd_minor);
1390         dtmpl->ctd_minor = NULL;
1391         mutex_exit(&tmpl->ctmpl_lock);
1392         ctmpl_free(tmpl);
1393         kmem_free(path, MAXPATHLEN);
1394 
1395         if (ctd == NULL) {
1396                 cmn_err(CE_NOTE, "contract_device_open(): Failed to "
1397                     "create device contract for process (%d) holding "
1398                     "device (devt = %lu, spec_type = %d)",
1399                     curproc->p_pid, dev, spec_type);
1400                 return (1);
1401         }
1402 
1403         if (ctpp) {
1404                 mutex_enter(&ctd->cond_contract.ct_lock);
1405                 *ctpp = &ctd->cond_contract;
1406                 mutex_exit(&ctd->cond_contract.ct_lock);
1407         }
1408         return (0);
1409 }
1410 
1411 /*
1412  * Called during contract negotiation by the device contract framework to wait
1413  * for ACKs or NACKs from contract holders. If all responses are not received
1414  * before a specified timeout, this routine times out.
1415  */
1416 static uint_t
1417 wait_for_acks(dev_info_t *dip, dev_t dev, int spec_type, uint_t evtype)
1418 {
1419         cont_device_t *ctd;
1420         int timed_out = 0;
1421         int result = CT_NONE;
1422         int ack;
1423         char *f = "wait_for_acks";
1424 
1425         ASSERT(MUTEX_HELD(&(DEVI(dip)->devi_ct_lock)));
1426         ASSERT(dip);
1427         ASSERT(evtype & CT_DEV_ALLEVENT);
1428         ASSERT(dev != NODEV && dev != DDI_DEV_T_NONE);
1429         ASSERT((dev == DDI_DEV_T_ANY && spec_type == 0) ||
1430             (spec_type == S_IFBLK || spec_type == S_IFCHR));
1431 
1432         CT_DEBUG((CE_NOTE, "%s: entered: dip: %p", f, (void *)dip));
1433 
1434         if (ct_barrier_wait_for_empty(dip, CT_DEV_ACKTIME) == -1) {
1435                 /*
1436                  * some contract owner(s) didn't respond in time
1437                  */
1438                 CT_DEBUG((CE_NOTE, "%s: timed out: %p", f, (void *)dip));
1439                 timed_out = 1;
1440         }
1441 
1442         ack = 0;
1443         for (ctd = list_head(&(DEVI(dip)->devi_ct)); ctd != NULL;
1444             ctd = list_next(&(DEVI(dip)->devi_ct), ctd)) {
1445 
1446                 mutex_enter(&ctd->cond_contract.ct_lock);
1447 
1448                 ASSERT(ctd->cond_dip == dip);
1449 
1450                 if (dev != DDI_DEV_T_ANY && dev != ctd->cond_devt) {
1451                         mutex_exit(&ctd->cond_contract.ct_lock);
1452                         continue;
1453                 }
1454                 if (dev != DDI_DEV_T_ANY && spec_type != ctd->cond_spec) {
1455                         mutex_exit(&ctd->cond_contract.ct_lock);
1456                         continue;
1457                 }
1458 
1459                 /* skip if non-negotiable contract */
1460                 if (ctd->cond_noneg) {
1461                         mutex_exit(&ctd->cond_contract.ct_lock);
1462                         continue;
1463                 }
1464 
1465                 ASSERT(ctd->cond_currev_type == evtype);
1466                 if (ctd->cond_currev_ack == CT_NACK) {
1467                         CT_DEBUG((CE_NOTE, "%s: found a NACK,result = NACK: %p",
1468                             f, (void *)dip));
1469                         mutex_exit(&ctd->cond_contract.ct_lock);
1470                         return (CT_NACK);
1471                 } else if (ctd->cond_currev_ack == CT_ACK) {
1472                         ack = 1;
1473                         CT_DEBUG((CE_NOTE, "%s: found a ACK: %p",
1474                             f, (void *)dip));
1475                 }
1476                 mutex_exit(&ctd->cond_contract.ct_lock);
1477         }
1478 
1479         if (ack) {
1480                 result = CT_ACK;
1481                 CT_DEBUG((CE_NOTE, "%s: result = ACK, dip=%p", f, (void *)dip));
1482         } else if (timed_out) {
1483                 result = CT_NONE;
1484                 CT_DEBUG((CE_NOTE, "%s: result = NONE (timed-out), dip=%p",
1485                     f, (void *)dip));
1486         } else {
1487                 CT_DEBUG((CE_NOTE, "%s: result = NONE, dip=%p",
1488                     f, (void *)dip));
1489         }
1490 
1491 
1492         return (result);
1493 }
1494 
1495 /*
1496  * Determines the current state of a device (i.e a devinfo node
1497  */
1498 static int
1499 get_state(dev_info_t *dip)
1500 {
1501         if (DEVI_IS_DEVICE_OFFLINE(dip) || DEVI_IS_DEVICE_DOWN(dip))
1502                 return (CT_DEV_EV_OFFLINE);
1503         else if (DEVI_IS_DEVICE_DEGRADED(dip))
1504                 return (CT_DEV_EV_DEGRADED);
1505         else
1506                 return (CT_DEV_EV_ONLINE);
1507 }
1508 
1509 /*
1510  * Sets the current state of a device in a device contract
1511  */
1512 static void
1513 set_cond_state(dev_info_t *dip)
1514 {
1515         uint_t state = get_state(dip);
1516         cont_device_t *ctd;
1517 
1518         /* verify that barrier is held */
1519         ASSERT(ct_barrier_held(dip));
1520 
1521         for (ctd = list_head(&(DEVI(dip)->devi_ct)); ctd != NULL;
1522             ctd = list_next(&(DEVI(dip)->devi_ct), ctd)) {
1523                 mutex_enter(&ctd->cond_contract.ct_lock);
1524                 ASSERT(ctd->cond_dip == dip);
1525                 ctd->cond_state = state;
1526                 mutex_exit(&ctd->cond_contract.ct_lock);
1527         }
1528 }
1529 
1530 /*
1531  * Core routine called by event-specific routines when an event occurs.
1532  * Determines if an event should be be published, and if it is to be
1533  * published, whether a negotiation should take place. Also implements
1534  * NEGEND events which publish the final disposition of an event after
1535  * negotiations are complete.
1536  *
1537  * When an event occurs on a minor node, this routine walks the list of
1538  * contracts hanging off a devinfo node and for each contract on the affected
1539  * dip, evaluates the following cases
1540  *
1541  *      a. an event that is synchronous, breaks the contract and NONEG not set
1542  *              - bumps up the outstanding negotiation counts on the dip
1543  *              - marks the dip as undergoing negotiation (devi_ct_neg)
1544  *              - event of type CTE_NEG is published
1545  *      b. an event that is synchronous, breaks the contract and NONEG is set
1546  *              - sets the final result to CT_NACK, event is blocked
1547  *              - does not publish an event
1548  *      c. event is asynchronous and breaks the contract
1549  *              - publishes a critical event irrespect of whether the NONEG
1550  *                flag is set, since the contract will be broken and contract
1551  *                owner needs to be informed.
1552  *      d. No contract breakage but the owner has subscribed to the event
1553  *              - publishes the event irrespective of the NONEG event as the
1554  *                owner has explicitly subscribed to the event.
1555  *      e. NEGEND event
1556  *              - publishes a critical event. Should only be doing this if
1557  *                if NONEG is not set.
1558  *      f. all other events
1559  *              - Since a contract is not broken and this event has not been
1560  *                subscribed to, this event does not need to be published for
1561  *                for this contract.
1562  *
1563  *      Once an event is published, what happens next depends on the type of
1564  *      event:
1565  *
1566  *      a. NEGEND event
1567  *              - cleanup all state associated with the preceding negotiation
1568  *                and return CT_ACK to the caller of contract_device_publish()
1569  *      b. NACKed event
1570  *              - One or more contracts had the NONEG term, so the event was
1571  *                blocked. Return CT_NACK to the caller.
1572  *      c. Negotiated event
1573  *              - Call wait_for_acks() to wait for responses from contract
1574  *              holders. The end result is either CT_ACK (event is permitted),
1575  *              CT_NACK (event is blocked) or CT_NONE (no contract owner)
1576  *              responded. This result is returned back to the caller.
1577  *      d. All other events
1578  *              - If the event was asynchronous (i.e. not negotiated) or
1579  *              a contract was not broken return CT_ACK to the caller.
1580  */
1581 static uint_t
1582 contract_device_publish(dev_info_t *dip, dev_t dev, int spec_type,
1583     uint_t evtype, nvlist_t *tnvl)
1584 {
1585         cont_device_t *ctd;
1586         uint_t result = CT_NONE;
1587         uint64_t evid = 0;
1588         uint64_t nevid = 0;
1589         char *path = NULL;
1590         int negend;
1591         int match;
1592         int sync = 0;
1593         contract_t *ct;
1594         ct_kevent_t *event;
1595         nvlist_t *nvl;
1596         int broken = 0;
1597 
1598         ASSERT(dip);
1599         ASSERT(dev != NODEV && dev != DDI_DEV_T_NONE);
1600         ASSERT((dev == DDI_DEV_T_ANY && spec_type == 0) ||
1601             (spec_type == S_IFBLK || spec_type == S_IFCHR));
1602         ASSERT(evtype == 0 || (evtype & CT_DEV_ALLEVENT));
1603 
1604         /* Is this a synchronous state change ? */
1605         if (evtype != CT_EV_NEGEND) {
1606                 sync = is_sync_neg(get_state(dip), evtype);
1607                 /* NOP if unsupported transition */
1608                 if (sync == -2 || sync == -1) {
1609                         DEVI(dip)->devi_flags |= DEVI_CT_NOP;
1610                         result = (sync == -2) ? CT_ACK : CT_NONE;
1611                         goto out;
1612                 }
1613                 CT_DEBUG((CE_NOTE, "publish: is%s sync state change",
1614                     sync ? "" : " not"));
1615         } else if (DEVI(dip)->devi_flags & DEVI_CT_NOP) {
1616                 DEVI(dip)->devi_flags &= ~DEVI_CT_NOP;
1617                 result = CT_ACK;
1618                 goto out;
1619         }
1620 
1621         path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1622         (void) ddi_pathname(dip, path);
1623 
1624         mutex_enter(&(DEVI(dip)->devi_ct_lock));
1625 
1626         /*
1627          * Negotiation end - set the state of the device in the contract
1628          */
1629         if (evtype == CT_EV_NEGEND) {
1630                 CT_DEBUG((CE_NOTE, "publish: negend: setting cond state"));
1631                 set_cond_state(dip);
1632         }
1633 
1634         /*
1635          * If this device didn't go through negotiation, don't publish
1636          * a NEGEND event - simply release the barrier to allow other
1637          * device events in.
1638          */
1639         negend = 0;
1640         if (evtype == CT_EV_NEGEND && !DEVI(dip)->devi_ct_neg) {
1641                 CT_DEBUG((CE_NOTE, "publish: no negend reqd. release barrier"));
1642                 ct_barrier_release(dip);
1643                 mutex_exit(&(DEVI(dip)->devi_ct_lock));
1644                 result = CT_ACK;
1645                 goto out;
1646         } else if (evtype == CT_EV_NEGEND) {
1647                 /*
1648                  * There are negotiated contract breakages that
1649                  * need a NEGEND event
1650                  */
1651                 ASSERT(ct_barrier_held(dip));
1652                 negend = 1;
1653                 CT_DEBUG((CE_NOTE, "publish: setting negend flag"));
1654         } else {
1655                 /*
1656                  * This is a new event, not a NEGEND event. Wait for previous
1657                  * contract events to complete.
1658                  */
1659                 ct_barrier_acquire(dip);
1660         }
1661 
1662 
1663         match = 0;
1664         for (ctd = list_head(&(DEVI(dip)->devi_ct)); ctd != NULL;
1665             ctd = list_next(&(DEVI(dip)->devi_ct), ctd)) {
1666 
1667                 ctid_t ctid;
1668                 size_t len = strlen(path);
1669 
1670                 mutex_enter(&ctd->cond_contract.ct_lock);
1671 
1672                 ASSERT(ctd->cond_dip == dip);
1673                 ASSERT(ctd->cond_minor);
1674                 ASSERT(strncmp(ctd->cond_minor, path, len) == 0 &&
1675                     ctd->cond_minor[len] == ':');
1676 
1677                 if (dev != DDI_DEV_T_ANY && dev != ctd->cond_devt) {
1678                         mutex_exit(&ctd->cond_contract.ct_lock);
1679                         continue;
1680                 }
1681                 if (dev != DDI_DEV_T_ANY && spec_type != ctd->cond_spec) {
1682                         mutex_exit(&ctd->cond_contract.ct_lock);
1683                         continue;
1684                 }
1685 
1686                 /* We have a matching contract */
1687                 match = 1;
1688                 ctid = ctd->cond_contract.ct_id;
1689                 CT_DEBUG((CE_NOTE, "publish: found matching contract: %d",
1690                     ctid));
1691 
1692                 /*
1693                  * There are 4 possible cases
1694                  * 1. A contract is broken (dev not in acceptable state) and
1695                  *    the state change is synchronous - start negotiation
1696                  *    by sending a CTE_NEG critical event.
1697                  * 2. A contract is broken and the state change is
1698                  *    asynchronous - just send a critical event and
1699                  *    break the contract.
1700                  * 3. Contract is not broken, but consumer has subscribed
1701                  *    to the event as a critical or informative event
1702                  *    - just send the appropriate event
1703                  * 4. contract waiting for negend event - just send the critical
1704                  *    NEGEND event.
1705                  */
1706                 broken = 0;
1707                 if (!negend && !(evtype & ctd->cond_aset)) {
1708                         broken = 1;
1709                         CT_DEBUG((CE_NOTE, "publish: Contract broken: %d",
1710                             ctid));
1711                 }
1712 
1713                 /*
1714                  * Don't send event if
1715                  *      - contract is not broken AND
1716                  *      - contract holder has not subscribed to this event AND
1717                  *      - contract not waiting for a NEGEND event
1718                  */
1719                 if (!broken && !EVSENDP(ctd, evtype) &&
1720                     !ctd->cond_neg) {
1721                         CT_DEBUG((CE_NOTE, "contract_device_publish(): "
1722                             "contract (%d): no publish reqd: event %d",
1723                             ctd->cond_contract.ct_id, evtype));
1724                         mutex_exit(&ctd->cond_contract.ct_lock);
1725                         continue;
1726                 }
1727 
1728                 /*
1729                  * Note: need to kmem_zalloc() the event so mutexes are
1730                  * initialized automatically
1731                  */
1732                 ct = &ctd->cond_contract;
1733                 event = kmem_zalloc(sizeof (ct_kevent_t), KM_SLEEP);
1734                 event->cte_type = evtype;
1735 
1736                 if (broken && sync) {
1737                         CT_DEBUG((CE_NOTE, "publish: broken + sync: "
1738                             "ctid: %d", ctid));
1739                         ASSERT(!negend);
1740                         ASSERT(ctd->cond_currev_id == 0);
1741                         ASSERT(ctd->cond_currev_type == 0);
1742                         ASSERT(ctd->cond_currev_ack == 0);
1743                         ASSERT(ctd->cond_neg == 0);
1744                         if (ctd->cond_noneg) {
1745                                 /* Nothing to publish. Event has been blocked */
1746                                 CT_DEBUG((CE_NOTE, "publish: sync and noneg:"
1747                                     "not publishing blocked ev: ctid: %d",
1748                                     ctid));
1749                                 result = CT_NACK;
1750                                 kmem_free(event, sizeof (ct_kevent_t));
1751                                 mutex_exit(&ctd->cond_contract.ct_lock);
1752                                 continue;
1753                         }
1754                         event->cte_flags = CTE_NEG; /* critical neg. event */
1755                         ctd->cond_currev_type = event->cte_type;
1756                         ct_barrier_incr(dip);
1757                         DEVI(dip)->devi_ct_neg = 1; /* waiting for negend */
1758                         ctd->cond_neg = 1;
1759                 } else if (broken && !sync) {
1760                         CT_DEBUG((CE_NOTE, "publish: broken + async: ctid: %d",
1761                             ctid));
1762                         ASSERT(!negend);
1763                         ASSERT(ctd->cond_currev_id == 0);
1764                         ASSERT(ctd->cond_currev_type == 0);
1765                         ASSERT(ctd->cond_currev_ack == 0);
1766                         ASSERT(ctd->cond_neg == 0);
1767                         event->cte_flags = 0; /* critical event */
1768                 } else if (EVSENDP(ctd, event->cte_type)) {
1769                         CT_DEBUG((CE_NOTE, "publish: event suscrib: ctid: %d",
1770                             ctid));
1771                         ASSERT(!negend);
1772                         ASSERT(ctd->cond_currev_id == 0);
1773                         ASSERT(ctd->cond_currev_type == 0);
1774                         ASSERT(ctd->cond_currev_ack == 0);
1775                         ASSERT(ctd->cond_neg == 0);
1776                         event->cte_flags = EVINFOP(ctd, event->cte_type) ?
1777                             CTE_INFO : 0;
1778                 } else if (ctd->cond_neg) {
1779                         CT_DEBUG((CE_NOTE, "publish: NEGEND: ctid: %d", ctid));
1780                         ASSERT(negend);
1781                         ASSERT(ctd->cond_noneg == 0);
1782                         nevid = ctd->cond_contract.ct_nevent ?
1783                             ctd->cond_contract.ct_nevent->cte_id : 0;
1784                         ASSERT(ctd->cond_currev_id == nevid);
1785                         event->cte_flags = 0;        /* NEGEND is always critical */
1786                         ctd->cond_currev_id = 0;
1787                         ctd->cond_currev_type = 0;
1788                         ctd->cond_currev_ack = 0;
1789                         ctd->cond_neg = 0;
1790                 } else {
1791                         CT_DEBUG((CE_NOTE, "publish: not publishing event for "
1792                             "ctid: %d, evtype: %d",
1793                             ctd->cond_contract.ct_id, event->cte_type));
1794                         ASSERT(!negend);
1795                         ASSERT(ctd->cond_currev_id == 0);
1796                         ASSERT(ctd->cond_currev_type == 0);
1797                         ASSERT(ctd->cond_currev_ack == 0);
1798                         ASSERT(ctd->cond_neg == 0);
1799                         kmem_free(event, sizeof (ct_kevent_t));
1800                         mutex_exit(&ctd->cond_contract.ct_lock);
1801                         continue;
1802                 }
1803 
1804                 nvl = NULL;
1805                 if (tnvl) {
1806                         VERIFY(nvlist_dup(tnvl, &nvl, 0) == 0);
1807                         if (negend) {
1808                                 int32_t newct = 0;
1809                                 ASSERT(ctd->cond_noneg == 0);
1810                                 VERIFY(nvlist_add_uint64(nvl, CTS_NEVID, nevid)
1811                                     == 0);
1812                                 VERIFY(nvlist_lookup_int32(nvl, CTS_NEWCT,
1813                                     &newct) == 0);
1814                                 VERIFY(nvlist_add_int32(nvl, CTS_NEWCT,
1815                                     newct == 1 ? 0 :
1816                                     ctd->cond_contract.ct_id) == 0);
1817                                 CT_DEBUG((CE_NOTE, "publish: negend: ctid: %d "
1818                                     "CTS_NEVID: %llu, CTS_NEWCT: %s",
1819                                     ctid, (unsigned long long)nevid,
1820                                     newct ? "success" : "failure"));
1821 
1822                         }
1823                 }
1824 
1825                 if (ctd->cond_neg) {
1826                         ASSERT(ctd->cond_contract.ct_ntime.ctm_start == -1);
1827                         ASSERT(ctd->cond_contract.ct_qtime.ctm_start == -1);
1828                         ctd->cond_contract.ct_ntime.ctm_start = ddi_get_lbolt();
1829                         ctd->cond_contract.ct_qtime.ctm_start =
1830                             ctd->cond_contract.ct_ntime.ctm_start;
1831                 }
1832 
1833                 /*
1834                  * by holding the dip's devi_ct_lock we ensure that
1835                  * all ACK/NACKs are held up until we have finished
1836                  * publishing to all contracts.
1837                  */
1838                 mutex_exit(&ctd->cond_contract.ct_lock);
1839                 evid = cte_publish_all(ct, event, nvl, NULL);
1840                 mutex_enter(&ctd->cond_contract.ct_lock);
1841 
1842                 if (ctd->cond_neg) {
1843                         ASSERT(!negend);
1844                         ASSERT(broken);
1845                         ASSERT(sync);
1846                         ASSERT(!ctd->cond_noneg);
1847                         CT_DEBUG((CE_NOTE, "publish: sync break, setting evid"
1848                             ": %d", ctid));
1849                         ctd->cond_currev_id = evid;
1850                 } else if (negend) {
1851                         ctd->cond_contract.ct_ntime.ctm_start = -1;
1852                         ctd->cond_contract.ct_qtime.ctm_start = -1;
1853                 }
1854                 mutex_exit(&ctd->cond_contract.ct_lock);
1855         }
1856 
1857         /*
1858          * If "negend" set counter back to initial state (-1) so that
1859          * other events can be published. Also clear the negotiation flag
1860          * on dip.
1861          *
1862          * 0 .. n are used for counting.
1863          * -1 indicates counter is available for use.
1864          */
1865         if (negend) {
1866                 /*
1867                  * devi_ct_count not necessarily 0. We may have
1868                  * timed out in which case, count will be non-zero.
1869                  */
1870                 ct_barrier_release(dip);
1871                 DEVI(dip)->devi_ct_neg = 0;
1872                 CT_DEBUG((CE_NOTE, "publish: negend: reset dip state: dip=%p",
1873                     (void *)dip));
1874         } else if (DEVI(dip)->devi_ct_neg) {
1875                 ASSERT(match);
1876                 ASSERT(!ct_barrier_empty(dip));
1877                 CT_DEBUG((CE_NOTE, "publish: sync count=%d, dip=%p",
1878                     DEVI(dip)->devi_ct_count, (void *)dip));
1879         } else {
1880                 /*
1881                  * for non-negotiated events or subscribed events or no
1882                  * matching contracts
1883                  */
1884                 ASSERT(ct_barrier_empty(dip));
1885                 ASSERT(DEVI(dip)->devi_ct_neg == 0);
1886                 CT_DEBUG((CE_NOTE, "publish: async/non-nego/subscrib/no-match: "
1887                     "dip=%p", (void *)dip));
1888 
1889                 /*
1890                  * only this function when called from contract_device_negend()
1891                  * can reset the counter to READY state i.e. -1. This function
1892                  * is so called for every event whether a NEGEND event is needed
1893                  * or not, but the negend event is only published if the event
1894                  * whose end they signal is a negotiated event for the contract.
1895                  */
1896         }
1897 
1898         if (!match) {
1899                 /* No matching contracts */
1900                 CT_DEBUG((CE_NOTE, "publish: No matching contract"));
1901                 result = CT_NONE;
1902         } else if (result == CT_NACK) {
1903                 /* a non-negotiable contract exists and this is a neg. event */
1904                 CT_DEBUG((CE_NOTE, "publish: found 1 or more NONEG contract"));
1905                 (void) wait_for_acks(dip, dev, spec_type, evtype);
1906         } else if (DEVI(dip)->devi_ct_neg) {
1907                 /* one or more contracts going through negotations  */
1908                 CT_DEBUG((CE_NOTE, "publish: sync contract: waiting"));
1909                 result = wait_for_acks(dip, dev, spec_type, evtype);
1910         } else {
1911                 /* no negotiated contracts or no broken contracts or NEGEND */
1912                 CT_DEBUG((CE_NOTE, "publish: async/no-break/negend"));
1913                 result = CT_ACK;
1914         }
1915 
1916         /*
1917          * Release the lock only now so that the only point where we
1918          * drop the lock is in wait_for_acks(). This is so that we don't
1919          * miss cv_signal/cv_broadcast from contract holders
1920          */
1921         CT_DEBUG((CE_NOTE, "publish: dropping devi_ct_lock"));
1922         mutex_exit(&(DEVI(dip)->devi_ct_lock));
1923 
1924 out:
1925         nvlist_free(tnvl);
1926         if (path)
1927                 kmem_free(path, MAXPATHLEN);
1928 
1929 
1930         CT_DEBUG((CE_NOTE, "publish: result = %s", result_str(result)));
1931         return (result);
1932 }
1933 
1934 
1935 /*
1936  * contract_device_offline
1937  *
1938  * Event publishing routine called by I/O framework when a device is offlined.
1939  */
1940 ct_ack_t
1941 contract_device_offline(dev_info_t *dip, dev_t dev, int spec_type)
1942 {
1943         nvlist_t *nvl;
1944         uint_t result;
1945         uint_t evtype;
1946 
1947         VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1948 
1949         evtype = CT_DEV_EV_OFFLINE;
1950         result = contract_device_publish(dip, dev, spec_type, evtype, nvl);
1951 
1952         /*
1953          * If a contract offline is NACKED, the framework expects us to call
1954          * NEGEND ourselves, since we know the final result
1955          */
1956         if (result == CT_NACK) {
1957                 contract_device_negend(dip, dev, spec_type, CT_EV_FAILURE);
1958         }
1959 
1960         return (result);
1961 }
1962 
1963 /*
1964  * contract_device_degrade
1965  *
1966  * Event publishing routine called by I/O framework when a device
1967  * moves to degrade state.
1968  */
1969 /*ARGSUSED*/
1970 void
1971 contract_device_degrade(dev_info_t *dip, dev_t dev, int spec_type)
1972 {
1973         nvlist_t *nvl;
1974         uint_t evtype;
1975 
1976         VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1977 
1978         evtype = CT_DEV_EV_DEGRADED;
1979         (void) contract_device_publish(dip, dev, spec_type, evtype, nvl);
1980 }
1981 
1982 /*
1983  * contract_device_undegrade
1984  *
1985  * Event publishing routine called by I/O framework when a device
1986  * moves from degraded state to online state.
1987  */
1988 /*ARGSUSED*/
1989 void
1990 contract_device_undegrade(dev_info_t *dip, dev_t dev, int spec_type)
1991 {
1992         nvlist_t *nvl;
1993         uint_t evtype;
1994 
1995         VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1996 
1997         evtype = CT_DEV_EV_ONLINE;
1998         (void) contract_device_publish(dip, dev, spec_type, evtype, nvl);
1999 }
2000 
2001 /*
2002  * For all contracts which have undergone a negotiation (because the device
2003  * moved out of the acceptable state for that contract and the state
2004  * change is synchronous i.e. requires negotiation) this routine publishes
2005  * a CT_EV_NEGEND event with the final disposition of the event.
2006  *
2007  * This event is always a critical event.
2008  */
2009 void
2010 contract_device_negend(dev_info_t *dip, dev_t dev, int spec_type, int result)
2011 {
2012         nvlist_t *nvl;
2013         uint_t evtype;
2014 
2015         ASSERT(result == CT_EV_SUCCESS || result == CT_EV_FAILURE);
2016 
2017         CT_DEBUG((CE_NOTE, "contract_device_negend(): entered: result: %d, "
2018             "dip: %p", result, (void *)dip));
2019 
2020         VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2021         VERIFY(nvlist_add_int32(nvl, CTS_NEWCT,
2022             result == CT_EV_SUCCESS ? 1 : 0) == 0);
2023 
2024         evtype = CT_EV_NEGEND;
2025         (void) contract_device_publish(dip, dev, spec_type, evtype, nvl);
2026 
2027         CT_DEBUG((CE_NOTE, "contract_device_negend(): exit dip: %p",
2028             (void *)dip));
2029 }
2030 
2031 /*
2032  * Wrapper routine called by other subsystems (such as LDI) to start
2033  * negotiations when a synchronous device state change occurs.
2034  * Returns CT_ACK or CT_NACK.
2035  */
2036 ct_ack_t
2037 contract_device_negotiate(dev_info_t *dip, dev_t dev, int spec_type,
2038     uint_t evtype)
2039 {
2040         int     result;
2041 
2042         ASSERT(dip);
2043         ASSERT(dev != NODEV);
2044         ASSERT(dev != DDI_DEV_T_ANY);
2045         ASSERT(dev != DDI_DEV_T_NONE);
2046         ASSERT(spec_type == S_IFBLK || spec_type == S_IFCHR);
2047 
2048         switch (evtype) {
2049         case CT_DEV_EV_OFFLINE:
2050                 result = contract_device_offline(dip, dev, spec_type);
2051                 break;
2052         default:
2053                 cmn_err(CE_PANIC, "contract_device_negotiate(): Negotiation "
2054                     "not supported: event (%d) for dev_t (%lu) and spec (%d), "
2055                     "dip (%p)", evtype, dev, spec_type, (void *)dip);
2056                 result = CT_NACK;
2057                 break;
2058         }
2059 
2060         return (result);
2061 }
2062 
2063 /*
2064  * A wrapper routine called by other subsystems (such as the LDI) to
2065  * finalize event processing for a state change event. For synchronous
2066  * state changes, this publishes NEGEND events. For asynchronous i.e.
2067  * non-negotiable events this publishes the event.
2068  */
2069 void
2070 contract_device_finalize(dev_info_t *dip, dev_t dev, int spec_type,
2071     uint_t evtype, int ct_result)
2072 {
2073         ASSERT(dip);
2074         ASSERT(dev != NODEV);
2075         ASSERT(dev != DDI_DEV_T_ANY);
2076         ASSERT(dev != DDI_DEV_T_NONE);
2077         ASSERT(spec_type == S_IFBLK || spec_type == S_IFCHR);
2078 
2079         switch (evtype) {
2080         case CT_DEV_EV_OFFLINE:
2081                 contract_device_negend(dip, dev, spec_type, ct_result);
2082                 break;
2083         case CT_DEV_EV_DEGRADED:
2084                 contract_device_degrade(dip, dev, spec_type);
2085                 contract_device_negend(dip, dev, spec_type, ct_result);
2086                 break;
2087         case CT_DEV_EV_ONLINE:
2088                 contract_device_undegrade(dip, dev, spec_type);
2089                 contract_device_negend(dip, dev, spec_type, ct_result);
2090                 break;
2091         default:
2092                 cmn_err(CE_PANIC, "contract_device_finalize(): Unsupported "
2093                     "event (%d) for dev_t (%lu) and spec (%d), dip (%p)",
2094                     evtype, dev, spec_type, (void *)dip);
2095                 break;
2096         }
2097 }
2098 
2099 /*
2100  * Called by I/O framework when a devinfo node is freed to remove the
2101  * association between a devinfo node and its contracts.
2102  */
2103 void
2104 contract_device_remove_dip(dev_info_t *dip)
2105 {
2106         cont_device_t *ctd;
2107         cont_device_t *next;
2108         contract_t *ct;
2109 
2110         mutex_enter(&(DEVI(dip)->devi_ct_lock));
2111         ct_barrier_wait_for_release(dip);
2112 
2113         for (ctd = list_head(&(DEVI(dip)->devi_ct)); ctd != NULL; ctd = next) {
2114                 next = list_next(&(DEVI(dip)->devi_ct), ctd);
2115                 list_remove(&(DEVI(dip)->devi_ct), ctd);
2116                 ct = &ctd->cond_contract;
2117                 /*
2118                  * Unlink the dip associated with this contract
2119                  */
2120                 mutex_enter(&ct->ct_lock);
2121                 ASSERT(ctd->cond_dip == dip);
2122                 ctd->cond_dip = NULL; /* no longer linked to dip */
2123                 contract_rele(ct);      /* remove hold for dip linkage */
2124                 CT_DEBUG((CE_NOTE, "ct: remove_dip: removed dip from contract: "
2125                     "ctid: %d", ct->ct_id));
2126                 mutex_exit(&ct->ct_lock);
2127         }
2128         ASSERT(list_is_empty(&(DEVI(dip)->devi_ct)));
2129         mutex_exit(&(DEVI(dip)->devi_ct_lock));
2130 }
2131 
2132 /*
2133  * Barrier related routines
2134  */
2135 static void
2136 ct_barrier_acquire(dev_info_t *dip)
2137 {
2138         ASSERT(MUTEX_HELD(&(DEVI(dip)->devi_ct_lock)));
2139         CT_DEBUG((CE_NOTE, "ct_barrier_acquire: waiting for barrier"));
2140         while (DEVI(dip)->devi_ct_count != -1)
2141                 cv_wait(&(DEVI(dip)->devi_ct_cv), &(DEVI(dip)->devi_ct_lock));
2142         DEVI(dip)->devi_ct_count = 0;
2143         CT_DEBUG((CE_NOTE, "ct_barrier_acquire: thread owns barrier"));
2144 }
2145 
2146 static void
2147 ct_barrier_release(dev_info_t *dip)
2148 {
2149         ASSERT(MUTEX_HELD(&(DEVI(dip)->devi_ct_lock)));
2150         ASSERT(DEVI(dip)->devi_ct_count != -1);
2151         DEVI(dip)->devi_ct_count = -1;
2152         cv_broadcast(&(DEVI(dip)->devi_ct_cv));
2153         CT_DEBUG((CE_NOTE, "ct_barrier_release: Released barrier"));
2154 }
2155 
2156 static int
2157 ct_barrier_held(dev_info_t *dip)
2158 {
2159         ASSERT(MUTEX_HELD(&(DEVI(dip)->devi_ct_lock)));
2160         return (DEVI(dip)->devi_ct_count != -1);
2161 }
2162 
2163 static int
2164 ct_barrier_empty(dev_info_t *dip)
2165 {
2166         ASSERT(MUTEX_HELD(&(DEVI(dip)->devi_ct_lock)));
2167         ASSERT(DEVI(dip)->devi_ct_count != -1);
2168         return (DEVI(dip)->devi_ct_count == 0);
2169 }
2170 
2171 static void
2172 ct_barrier_wait_for_release(dev_info_t *dip)
2173 {
2174         ASSERT(MUTEX_HELD(&(DEVI(dip)->devi_ct_lock)));
2175         while (DEVI(dip)->devi_ct_count != -1)
2176                 cv_wait(&(DEVI(dip)->devi_ct_cv), &(DEVI(dip)->devi_ct_lock));
2177 }
2178 
2179 static void
2180 ct_barrier_decr(dev_info_t *dip)
2181 {
2182         CT_DEBUG((CE_NOTE, "barrier_decr:  ct_count before decr: %d",
2183             DEVI(dip)->devi_ct_count));
2184 
2185         ASSERT(MUTEX_HELD(&(DEVI(dip)->devi_ct_lock)));
2186         ASSERT(DEVI(dip)->devi_ct_count > 0);
2187 
2188         DEVI(dip)->devi_ct_count--;
2189         if (DEVI(dip)->devi_ct_count == 0) {
2190                 cv_broadcast(&DEVI(dip)->devi_ct_cv);
2191                 CT_DEBUG((CE_NOTE, "barrier_decr: cv_broadcast"));
2192         }
2193 }
2194 
2195 static void
2196 ct_barrier_incr(dev_info_t *dip)
2197 {
2198         ASSERT(ct_barrier_held(dip));
2199         DEVI(dip)->devi_ct_count++;
2200 }
2201 
2202 static int
2203 ct_barrier_wait_for_empty(dev_info_t *dip, int secs)
2204 {
2205         clock_t abstime;
2206 
2207         ASSERT(MUTEX_HELD(&(DEVI(dip)->devi_ct_lock)));
2208 
2209         abstime = ddi_get_lbolt() + drv_usectohz(secs*1000000);
2210         while (DEVI(dip)->devi_ct_count) {
2211                 if (cv_timedwait(&(DEVI(dip)->devi_ct_cv),
2212                     &(DEVI(dip)->devi_ct_lock), abstime) == -1) {
2213                         return (-1);
2214                 }
2215         }
2216         return (0);
2217 }