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