xref: /titanic_41/usr/src/lib/libpool/common/pool_kernel.c (revision fd9cb95cbb2f626355a60efb9d02c5f0a33c10e6)
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, Version 1.0 only
6  * (the "License").  You may not use this file except in compliance
7  * with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or http://www.opensolaris.org/os/licensing.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 /*
23  * Copyright 2004 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #pragma ident	"%Z%%M%	%I%	%E% SMI"
28 
29 #include <assert.h>
30 #include <errno.h>
31 #include <exacct.h>
32 #include <fcntl.h>
33 #include <libnvpair.h>
34 #include <limits.h>
35 #include <poll.h>
36 #include <pool.h>
37 #include <stdlib.h>
38 #include <stdio.h>
39 #include <string.h>
40 #include <strings.h>
41 #include <stropts.h>
42 #include <thread.h>
43 #include <time.h>
44 #include <unistd.h>
45 
46 #include <libxml/tree.h>
47 
48 #include <sys/mman.h>
49 #include <sys/pool.h>
50 #include <sys/pool_impl.h>
51 #include <sys/priocntl.h>
52 #include <sys/stat.h>
53 #include <sys/time.h>
54 #include <sys/types.h>
55 
56 #include "dict.h"
57 
58 #include "pool_internal.h"
59 #include "pool_impl.h"
60 #include "pool_kernel_impl.h"
61 
62 /*
63  * libpool kernel Manipulation Routines
64  *
65  * pool_kernel.c implements the kernel manipulation routines used by the
66  * libpool kernel datastore. The functions are grouped into the following
67  * logical areas
68  *
69  */
70 
71 /*
72  * Device snapshot transfer buffer size
73  */
74 #define	KERNEL_SNAPSHOT_BUF_SZ	65535
75 
76 /*
77  * Kernel result set's initial size. 8 is probably large enough for
78  * most queries. Queries requiring more space are accomodated using
79  * realloc on a per result set basis.
80  */
81 #define	KERNEL_RS_INITIAL_SZ	8
82 
83 /*
84  * Property manipulation macros
85  */
86 #define	KERNEL_PROP_RDONLY	0x1
87 
88 /*
89  * Information required to evaluate qualifying elements for a query
90  */
91 struct query_obj {
92 	const pool_conf_t *conf;
93 	const pool_elem_t *src;
94 	const char *src_attr;
95 	pool_elem_class_t classes;
96 	pool_value_t **props;
97 	pool_knl_result_set_t *rs;
98 };
99 
100 /*
101  * Identifies a pool element with a processor set id
102  */
103 typedef struct pool_set_xref {
104 	pool_knl_pool_t	*psx_pool;
105 	uint_t		psx_pset_id;
106 	struct pool_set_xref *psx_next;
107 } pool_set_xref_t;
108 
109 /*
110  * Controls exacct snapshot load into libpool data structure
111  */
112 typedef struct pool_snap_load {
113 	int *psl_changed;
114 	pool_set_xref_t *psl_xref;
115 	pool_elem_t *psl_system;
116 	pool_knl_resource_t *psl_pset;
117 } pool_snap_load_t;
118 
119 /*
120  * Information about an XML document which is being constructed
121  */
122 struct knl_to_xml {
123 	xmlDocPtr ktx_doc;
124 	xmlNodePtr ktx_node;
125 };
126 
127 /*
128  * Undo structure processing. The following structures are all used to
129  * allow changes to the libpool snapshot and kernel following an
130  * unsuccessful commit.
131  */
132 typedef struct pool_create_undo {
133 	pool_create_t pcu_ioctl;
134 	pool_elem_t *pcu_elem;
135 } pool_create_undo_t;
136 
137 typedef struct pool_destroy_undo {
138 	pool_destroy_t pdu_ioctl;
139 	pool_elem_t *pdu_elem;
140 } pool_destroy_undo_t;
141 
142 typedef struct pool_assoc_undo {
143 	pool_assoc_t pau_ioctl;
144 	pool_elem_t *pau_assoc;
145 	pool_elem_t *pau_oldres;
146 	pool_elem_t *pau_newres;
147 } pool_assoc_undo_t;
148 
149 typedef struct pool_dissoc_undo {
150 	pool_dissoc_t pdu_ioctl;
151 	pool_elem_t *pdu_dissoc;
152 	pool_elem_t *pdu_oldres;
153 	pool_elem_t *pdu_newres;
154 } pool_dissoc_undo_t;
155 
156 typedef struct pool_xtransfer_undo {
157 	pool_xtransfer_t pxu_ioctl;
158 	pool_elem_t *pxu_src;
159 	pool_elem_t *pxu_tgt;
160 	pool_component_t **pxu_rl;
161 } pool_xtransfer_undo_t;
162 
163 typedef struct pool_propput_undo {
164 	pool_propput_t ppu_ioctl;
165 	pool_elem_t *ppu_elem;
166 	nvlist_t *ppu_alist;
167 	nvlist_t *ppu_blist;
168 	uchar_t ppu_doioctl;
169 } pool_propput_undo_t;
170 
171 typedef struct pool_proprm_undo {
172 	pool_proprm_t pru_ioctl;
173 	pool_elem_t *pru_elem;
174 	pool_value_t pru_oldval;
175 } pool_proprm_undo_t;
176 
177 extern const char *dtd_location;
178 
179 extern const char *element_class_tags[];
180 extern const char pool_info_location[];
181 
182 /*
183  * These functions are defined in pool_xml.c and represent the minimum
184  * XML support required to allow a pool kernel configuration to be
185  * exported as an XML document.
186  */
187 extern int pool_xml_set_attr(xmlNodePtr, xmlChar *, const pool_value_t *);
188 extern int pool_xml_set_prop(xmlNodePtr, xmlChar *, const pool_value_t *);
189 extern void xml_init(void);
190 extern xmlNodePtr node_create(xmlNodePtr, const xmlChar *);
191 extern void pool_error_func(void *, const char *, ...);
192 /*
193  * Utilities
194  */
195 static int load_group(pool_conf_t *, pool_knl_elem_t *, ea_object_t *,
196     pool_snap_load_t *);
197 static void pool_knl_elem_free(pool_knl_elem_t *, int);
198 static int pool_knl_put_xml_property(pool_elem_t *, xmlNodePtr, const char *,
199     const pool_value_t *);
200 static int pool_knl_snap_load_push(pool_snap_load_t *, pool_knl_pool_t *);
201 static int pool_knl_snap_load_update(pool_snap_load_t *, int, uint_t);
202 static int pool_knl_snap_load_remove(pool_snap_load_t *, int, uint_t);
203 static nvpair_t *pool_knl_find_nvpair(nvlist_t *, const char *);
204 static int pool_knl_nvlist_add_value(nvlist_t *, const char *,
205     const pool_value_t *);
206 static int pool_knl_recover(pool_conf_t *);
207 static uint64_t hash_id(const pool_elem_t *);
208 static int blocking_open(const char *, int);
209 
210 /*
211  * Connections
212  */
213 static void pool_knl_connection_free(pool_knl_connection_t *);
214 
215 /*
216  * Configuration
217  */
218 static int pool_knl_close(pool_conf_t *);
219 static int pool_knl_validate(const pool_conf_t *, pool_valid_level_t);
220 static int pool_knl_commit(pool_conf_t *);
221 static int pool_knl_export(const pool_conf_t *, const char *,
222     pool_export_format_t);
223 static int pool_knl_rollback(pool_conf_t *);
224 static pool_result_set_t *pool_knl_exec_query(const pool_conf_t *,
225     const pool_elem_t *, const char *, pool_elem_class_t, pool_value_t **);
226 static int pool_knl_remove(pool_conf_t *);
227 static char *pool_knl_get_binding(pool_conf_t *, pid_t);
228 static int pool_knl_set_binding(pool_conf_t *, const char *, idtype_t, id_t);
229 static char *pool_knl_get_resource_binding(pool_conf_t *,
230     pool_resource_elem_class_t, pid_t);
231 static int pool_knl_res_transfer(pool_resource_t *, pool_resource_t *,
232     uint64_t);
233 static int pool_knl_res_xtransfer(pool_resource_t *, pool_resource_t *,
234     pool_component_t **);
235 
236 /*
237  * Result Sets
238  */
239 static pool_knl_result_set_t *pool_knl_result_set_alloc(const pool_conf_t *);
240 static int pool_knl_result_set_append(pool_knl_result_set_t *,
241     pool_knl_elem_t *);
242 static int pool_knl_result_set_realloc(pool_knl_result_set_t *);
243 static void pool_knl_result_set_free(pool_knl_result_set_t *);
244 static pool_elem_t *pool_knl_rs_next(pool_result_set_t *);
245 static pool_elem_t *pool_knl_rs_prev(pool_result_set_t *);
246 static pool_elem_t *pool_knl_rs_first(pool_result_set_t *);
247 static pool_elem_t *pool_knl_rs_last(pool_result_set_t *);
248 static int pool_knl_rs_set_index(pool_result_set_t *, int);
249 static int pool_knl_rs_get_index(pool_result_set_t *);
250 static int pool_knl_rs_count(pool_result_set_t *);
251 static int pool_knl_rs_close(pool_result_set_t *);
252 
253 /*
254  * Element (and sub-type)
255  */
256 static pool_knl_elem_t *pool_knl_elem_wrap(pool_conf_t *, pool_elem_class_t,
257     pool_resource_elem_class_t, pool_component_elem_class_t);
258 static pool_elem_t *pool_knl_elem_create(pool_conf_t *, pool_elem_class_t,
259     pool_resource_elem_class_t, pool_component_elem_class_t);
260 static int pool_knl_elem_remove(pool_elem_t *);
261 static int pool_knl_set_container(pool_elem_t *, pool_elem_t *);
262 static pool_elem_t *pool_knl_get_container(const pool_elem_t *);
263 /*
264  * Pool element specific
265  */
266 static int pool_knl_pool_associate(pool_t *, const pool_resource_t *);
267 static int pool_knl_pool_dissociate(pool_t *, const pool_resource_t *);
268 
269 /*
270  * Resource elements specific
271  */
272 static int pool_knl_resource_is_system(const pool_resource_t *);
273 static int pool_knl_resource_can_associate(const pool_resource_t *);
274 
275 /* Properties */
276 static pool_value_class_t pool_knl_get_property(const pool_elem_t *,
277     const char *, pool_value_t *);
278 static pool_value_class_t pool_knl_get_dynamic_property(const pool_elem_t *,
279     const char *, pool_value_t *);
280 static int pool_knl_put_property(pool_elem_t *, const char *,
281     const pool_value_t *);
282 static int pool_knl_rm_property(pool_elem_t *, const char *);
283 static pool_value_t **pool_knl_get_properties(const pool_elem_t *, uint_t *);
284 
285 /*
286  * Logging
287  */
288 static int log_item_commit(log_item_t *);
289 static int log_item_undo(log_item_t *);
290 static int log_item_release(log_item_t *);
291 
292 /*
293  * Utilities
294  */
295 
296 /*
297  * load_group() updates the library configuration with the kernel
298  * snapshot supplied in ep. The function is designed to be called
299  * recursively. This function depends implicitly on the ordering of
300  * the data provided in ep. Changes to the ordering of data in ep must
301  * be matched by changes to this function.
302  */
303 int
304 load_group(pool_conf_t *conf, pool_knl_elem_t *elem, ea_object_t *ep,
305     pool_snap_load_t *psl)
306 {
307 	ea_object_t *eo;
308 	pool_knl_elem_t *old_elem;
309 	pool_knl_connection_t *prov = (pool_knl_connection_t *)conf->pc_prov;
310 	int ret = PO_SUCCESS;
311 
312 	if ((ep->eo_catalog & EXD_DATA_MASK) == EXD_GROUP_SYSTEM) {
313 		if ((elem = pool_knl_elem_wrap(conf, PEC_SYSTEM, PREC_INVALID,
314 		    PCEC_INVALID)) == NULL)
315 			return (PO_FAIL);
316 		if (nvlist_alloc(&elem->pke_properties, NV_UNIQUE_NAME_TYPE,
317 		    0) != 0) {
318 			pool_knl_elem_free(elem, PO_FALSE);
319 			pool_seterror(POE_SYSTEM);
320 			return (PO_FAIL);
321 		}
322 		/*
323 		 * Check to see if we already have an element
324 		 * for this data. If we have, free the newly
325 		 * created elem and continue with the old one
326 		 */
327 		if ((old_elem = dict_get(prov->pkc_elements, elem)) != NULL) {
328 			nvlist_free(old_elem->pke_properties);
329 			old_elem->pke_properties = elem->pke_properties;
330 			pool_knl_elem_free(elem, PO_FALSE);
331 			elem = old_elem;
332 		} else {
333 			if (dict_put(prov->pkc_elements, elem, elem) != NULL) {
334 				pool_knl_elem_free(elem, PO_TRUE);
335 				pool_seterror(POE_SYSTEM);
336 				return (PO_FAIL);
337 			}
338 		}
339 		psl->psl_system = (pool_elem_t *)elem;
340 	}
341 
342 	for (eo = ep->eo_group.eg_objs; eo != NULL; eo = eo->eo_next) {
343 		int data;
344 		pool_knl_elem_t *prop_elem = NULL;
345 
346 		data = (eo->eo_catalog & EXD_DATA_MASK);
347 
348 		switch (data) {
349 		case EXD_SYSTEM_TSTAMP:
350 		case EXD_POOL_TSTAMP:
351 		case EXD_PSET_TSTAMP:
352 		case EXD_CPU_TSTAMP:
353 			if (eo->eo_item.ei_uint64 > prov->pkc_lotime) {
354 				if (eo->eo_item.ei_uint64 > prov->pkc_ltime)
355 					prov->pkc_ltime = eo->eo_item.ei_uint64;
356 				if (psl->psl_changed) {
357 					switch (data) {
358 					case EXD_SYSTEM_TSTAMP:
359 						*psl->psl_changed |= POU_SYSTEM;
360 						break;
361 					case EXD_POOL_TSTAMP:
362 						*psl->psl_changed |= POU_POOL;
363 						break;
364 					case EXD_PSET_TSTAMP:
365 						*psl->psl_changed |= POU_PSET;
366 						break;
367 					case EXD_CPU_TSTAMP:
368 						*psl->psl_changed |= POU_CPU;
369 						break;
370 					}
371 				}
372 			}
373 			break;
374 		case EXD_SYSTEM_PROP:
375 		case EXD_POOL_PROP:
376 		case EXD_PSET_PROP:
377 		case EXD_CPU_PROP:
378 			if (data == EXD_PSET_PROP) {
379 				prop_elem = elem;
380 				elem = (pool_knl_elem_t *)psl->psl_pset;
381 			}
382 			nvlist_free(elem->pke_properties);
383 			if (nvlist_unpack(eo->eo_item.ei_raw,
384 			    eo->eo_item.ei_size, &elem->pke_properties, 0) !=
385 			    0) {
386 				pool_seterror(POE_SYSTEM);
387 				return (PO_FAIL);
388 			}
389 			elem->pke_ltime = prov->pkc_ltime;
390 			if (data == EXD_PSET_PROP) {
391 				elem = prop_elem;
392 			}
393 			break;
394 		case EXD_POOL_POOLID:
395 			if (nvlist_alloc(&elem->pke_properties,
396 			    NV_UNIQUE_NAME_TYPE, 0) != 0) {
397 				pool_seterror(POE_SYSTEM);
398 				return (PO_FAIL);
399 			}
400 			if (nvlist_add_int64(elem->pke_properties,
401 			    "pool.sys_id",
402 			    (int64_t)eo->eo_item.ei_uint32) != 0) {
403 				pool_seterror(POE_SYSTEM);
404 				return (PO_FAIL);
405 			}
406 			if ((old_elem = dict_get(prov->pkc_elements, elem)) !=
407 			    NULL) {
408 				nvlist_free(old_elem->pke_properties);
409 				old_elem->pke_properties = elem->pke_properties;
410 				pool_knl_elem_free(elem, PO_FALSE);
411 				elem = old_elem;
412 			} else {
413 				if (dict_put(prov->pkc_elements, elem, elem) !=
414 				    NULL) {
415 					pool_knl_elem_free(elem, PO_TRUE);
416 					pool_seterror(POE_SYSTEM);
417 					return (PO_FAIL);
418 				}
419 			}
420 			if (pool_knl_snap_load_push(psl,
421 			    (pool_knl_pool_t *)elem) != PO_SUCCESS) {
422 				pool_seterror(POE_SYSTEM);
423 				return (PO_FAIL);
424 			}
425 			((pool_knl_pool_t *)elem)->pkp_assoc[PREC_PSET] = NULL;
426 			break;
427 		case EXD_POOL_PSETID:
428 			if (pool_knl_snap_load_update(psl, EXD_POOL_PSETID,
429 			    eo->eo_item.ei_uint32) != PO_SUCCESS) {
430 				pool_seterror(POE_SYSTEM);
431 				return (PO_FAIL);
432 			}
433 			break;
434 		case EXD_PSET_PSETID:
435 			if (nvlist_alloc(&elem->pke_properties,
436 			    NV_UNIQUE_NAME_TYPE, 0) != 0) {
437 				pool_seterror(POE_SYSTEM);
438 				return (PO_FAIL);
439 			}
440 			if (nvlist_add_int64(elem->pke_properties,
441 			    "pset.sys_id",
442 			    (int64_t)eo->eo_item.ei_uint32) != 0) {
443 				pool_seterror(POE_SYSTEM);
444 				return (PO_FAIL);
445 			}
446 			if ((old_elem = dict_get(prov->pkc_elements, elem)) !=
447 			    NULL) {
448 				nvlist_free(old_elem->pke_properties);
449 				old_elem->pke_properties = elem->pke_properties;
450 				pool_knl_elem_free(elem, PO_FALSE);
451 				elem = old_elem;
452 			} else {
453 				if (dict_put(prov->pkc_elements, elem, elem) !=
454 				    NULL) {
455 					pool_knl_elem_free(elem, PO_TRUE);
456 					pool_seterror(POE_SYSTEM);
457 					return (PO_FAIL);
458 				}
459 			}
460 			psl->psl_pset = (pool_knl_resource_t *)elem;
461 			if (pool_knl_snap_load_remove(psl, data,
462 			    eo->eo_item.ei_uint32) != PO_SUCCESS) {
463 				pool_seterror(POE_SYSTEM);
464 				return (PO_FAIL);
465 			}
466 			break;
467 		case EXD_CPU_CPUID:
468 			if (nvlist_alloc(&elem->pke_properties,
469 			    NV_UNIQUE_NAME_TYPE, 0) != 0) {
470 				pool_seterror(POE_SYSTEM);
471 				return (PO_FAIL);
472 			}
473 			if (nvlist_add_int64(elem->pke_properties,
474 			    "cpu.sys_id",
475 			    (int64_t)eo->eo_item.ei_uint32) != 0) {
476 				pool_seterror(POE_SYSTEM);
477 				return (PO_FAIL);
478 			}
479 			if ((old_elem = dict_get(prov->pkc_elements, elem)) !=
480 			    NULL) {
481 				nvlist_free(old_elem->pke_properties);
482 				old_elem->pke_properties = elem->pke_properties;
483 				old_elem->pke_parent = elem->pke_parent;
484 				pool_knl_elem_free(elem, PO_FALSE);
485 				elem = old_elem;
486 			} else {
487 				if (dict_put(prov->pkc_elements, elem, elem) !=
488 				    NULL) {
489 					pool_knl_elem_free(elem, PO_TRUE);
490 					pool_seterror(POE_SYSTEM);
491 					return (PO_FAIL);
492 				}
493 			}
494 			break;
495 		case EXD_GROUP_POOL:
496 			if ((elem = pool_knl_elem_wrap(conf, PEC_POOL,
497 			    PREC_INVALID, PCEC_INVALID)) == NULL)
498 				return (PO_FAIL);
499 			if (pool_set_container(psl->psl_system,
500 			    (pool_elem_t *)elem) != PO_SUCCESS) {
501 				pool_seterror(POE_SYSTEM);
502 				return (PO_FAIL);
503 			}
504 			break;
505 		case EXD_GROUP_PSET:
506 			if ((elem = pool_knl_elem_wrap(conf, PEC_RES_COMP,
507 			    PREC_PSET, PCEC_INVALID)) == NULL)
508 				return (PO_FAIL);
509 			if (pool_set_container(psl->psl_system,
510 			    (pool_elem_t *)elem) != PO_SUCCESS) {
511 				pool_seterror(POE_SYSTEM);
512 				return (PO_FAIL);
513 			}
514 			break;
515 		case EXD_GROUP_CPU:
516 			if ((elem = pool_knl_elem_wrap(conf, PEC_COMP,
517 			    PREC_INVALID, PCEC_CPU)) == NULL)
518 				return (PO_FAIL);
519 			if (pool_set_container((pool_elem_t *)psl->psl_pset,
520 			    (pool_elem_t *)elem) != PO_SUCCESS) {
521 				pool_seterror(POE_SYSTEM);
522 				return (PO_FAIL);
523 			}
524 			break;
525 		default:
526 			break;
527 		}
528 
529 
530 		if (eo->eo_type == EO_GROUP) {
531 			if ((ret = load_group(conf, elem, eo, psl)) == PO_FAIL)
532 				break;
533 		}
534 	}
535 	return (ret);
536 }
537 
538 /*
539  * Push a snapshot entry onto the list of pools in the snapshot.
540  */
541 int
542 pool_knl_snap_load_push(pool_snap_load_t *psl, pool_knl_pool_t *pkp)
543 {
544 	pool_set_xref_t *psx;
545 
546 	if ((psx = malloc(sizeof (pool_set_xref_t))) == NULL) {
547 		pool_seterror(POE_SYSTEM);
548 		return (PO_FAIL);
549 	}
550 	(void) memset(psx, 0, sizeof (pool_set_xref_t));
551 	psx->psx_pool = pkp;
552 	/*
553 	 * Push onto the list of pools
554 	 */
555 	psx->psx_next = psl->psl_xref;
556 	psl->psl_xref = psx;
557 
558 	return (PO_SUCCESS);
559 }
560 
561 /*
562  * Update the current cross-reference for the supplied type of
563  * resource.
564  */
565 int
566 pool_knl_snap_load_update(pool_snap_load_t *psl, int type, uint_t id)
567 {
568 	switch (type) {
569 	case EXD_POOL_PSETID:
570 		psl->psl_xref->psx_pset_id = id;
571 		break;
572 	default:
573 		return (PO_FAIL);
574 	}
575 
576 	return (PO_SUCCESS);
577 }
578 
579 /*
580  * Remove a resource entry with the supplied type and id from the
581  * snapshot list when it is no longer required.
582  */
583 int
584 pool_knl_snap_load_remove(pool_snap_load_t *psl, int type, uint_t id)
585 {
586 	pool_set_xref_t *current, *prev, *next;
587 
588 	for (prev = NULL, current = psl->psl_xref; current != NULL;
589 	    current = next) {
590 		switch (type) {
591 		case EXD_PSET_PSETID:
592 			if (current->psx_pset_id == id)
593 				current->psx_pool->pkp_assoc[PREC_PSET] =
594 				    psl->psl_pset;
595 			break;
596 		default:
597 			return (PO_FAIL);
598 		}
599 		next = current->psx_next;
600 		if (current->psx_pool->pkp_assoc[PREC_PSET] != NULL) {
601 			if (prev != NULL) {
602 				prev->psx_next = current->psx_next;
603 			} else {
604 				psl->psl_xref = current->psx_next;
605 			}
606 			free(current);
607 		} else
608 			prev = current;
609 	}
610 
611 	return (PO_SUCCESS);
612 }
613 
614 /*
615  * Return the nvpair with the supplied name from the supplied list.
616  *
617  * NULL is returned if the name cannot be found in the list.
618  */
619 nvpair_t *
620 pool_knl_find_nvpair(nvlist_t *l, const char *name)
621 {
622 	nvpair_t *pair;
623 
624 	for (pair = nvlist_next_nvpair(l, NULL); pair != NULL;
625 		pair = nvlist_next_nvpair(l, pair)) {
626 		if (strcmp(nvpair_name(pair), name) == 0)
627 			break;
628 	}
629 	return (pair);
630 }
631 
632 /*
633  * Close the configuration. There are a few steps to closing a configuration:
634  * - Close the pseudo device
635  * - Free the data provider
636  * Returns PO_SUCCESS/PO_FAIL
637  */
638 int
639 pool_knl_close(pool_conf_t *conf)
640 {
641 	pool_knl_connection_t *prov = (pool_knl_connection_t *)conf->pc_prov;
642 
643 	if (close(prov->pkc_fd) < 0) {
644 		pool_seterror(POE_SYSTEM);
645 		return (PO_FAIL);
646 	}
647 	/*
648 	 * Rollback any pending changes before freeing the prov. This
649 	 * ensures there are no memory leaks from pending
650 	 * transactions.
651 	 */
652 	(void) pool_knl_rollback(conf);
653 	pool_knl_connection_free(prov);
654 	return (PO_SUCCESS);
655 }
656 
657 /*
658  * Remove elements in this map (previously identified as "dead") from
659  * the configuration map (prov->pkc_elements).
660  */
661 
662 /* ARGSUSED1 */
663 static void
664 remove_dead_elems(const void *key, void **value, void *cl)
665 {
666 	pool_knl_elem_t *pke = (pool_knl_elem_t *)key;
667 	pool_conf_t *conf = TO_CONF(TO_ELEM(pke));
668 	pool_knl_connection_t *prov = (pool_knl_connection_t *)conf->pc_prov;
669 
670 	assert(dict_remove(prov->pkc_elements, pke) != NULL);
671 #ifdef DEBUG
672 	dprintf("remove_dead_elems:\n");
673 	pool_elem_dprintf(TO_ELEM(pke));
674 #endif	/* DEBUG */
675 	pool_knl_elem_free(pke, PO_TRUE);
676 }
677 
678 /*
679  * Find elements which were not updated the last time that
680  * load_group() was called. Add those elements into a separate map
681  * (passed in cl) which will be later used to remove these elements
682  * from the configuration map.
683  */
684 /* ARGSUSED1 */
685 static void
686 find_dead_elems(const void *key, void **value, void *cl)
687 {
688 	pool_knl_elem_t *pke = (pool_knl_elem_t *)key;
689 	pool_conf_t *conf = TO_CONF(TO_ELEM(pke));
690 	pool_knl_connection_t *prov = (pool_knl_connection_t *)conf->pc_prov;
691 	dict_hdl_t *dead_map = (dict_hdl_t *)cl;
692 
693 	if (pke->pke_ltime < prov->pkc_ltime)
694 		(void) dict_put(dead_map, pke, pke);
695 }
696 
697 /*
698  * Update the snapshot held by the library. This function acts as the
699  * controller for the snapshot update procedure. Then snapshot is
700  * actually updated in multiple phases by the load_group() function
701  * (which updates existing elements and creates new elements as
702  * required) and then by find_dead_elems and remove_dead_elems
703  * (respectively responsible for identifying elements which are to be
704  * removed and then removing them).
705  *
706  * Returns PO_SUCCESS
707  */
708 int
709 pool_knl_update(pool_conf_t *conf, int *changed)
710 {
711 	pool_knl_connection_t *prov = (pool_knl_connection_t *)conf->pc_prov;
712 	pool_query_t query = {0};
713 	ea_object_t *ep;
714 	dict_hdl_t *dead_map;
715 	pool_snap_load_t psl = { NULL };
716 
717 	/*
718 	 * Ensure the library snapshot is consistent, if there are any
719 	 * outstanding transactions return failure.
720 	 */
721 	if (log_size(prov->pkc_log) != 0) {
722 		pool_seterror(POE_INVALID_CONF);
723 		return (PO_FAIL);
724 	}
725 	/*
726 	 * Query the kernel for a snapshot of the configuration state. Use
727 	 * load_group to allocate the user-land representation of the
728 	 * data returned in the snapshot.
729 	 */
730 	/* LINTED E_CONSTANT_CONDITION */
731 	while (1) {
732 		if (ioctl(prov->pkc_fd, POOL_QUERY, &query) < 0) {
733 			pool_seterror(POE_SYSTEM);
734 			return (PO_FAIL);
735 		}
736 		if ((query.pq_io_buf = calloc(1,
737 		    (query.pq_io_bufsize < KERNEL_SNAPSHOT_BUF_SZ) ?
738 		    query.pq_io_bufsize * 2 : query.pq_io_bufsize)) == NULL) {
739 			pool_seterror(POE_SYSTEM);
740 			return (PO_FAIL);
741 		}
742 		if (ioctl(prov->pkc_fd, POOL_QUERY, &query) < 0) {
743 			free(query.pq_io_buf);
744 			if (errno != ENOMEM) {
745 				pool_seterror(POE_SYSTEM);
746 				return (PO_FAIL);
747 			}
748 			query.pq_io_bufsize = 0;
749 			query.pq_io_buf = NULL;
750 		} else
751 			break;
752 	}
753 	if (ea_unpack_object(&ep, EUP_NOALLOC, query.pq_io_buf,
754 	    query.pq_io_bufsize) != EO_GROUP) {
755 		free(query.pq_io_buf);
756 		pool_seterror(POE_DATASTORE);
757 		return (PO_FAIL);
758 	}
759 	/*
760 	 * Update the library snapshot
761 	 */
762 	psl.psl_changed = changed;
763 	prov->pkc_lotime = prov->pkc_ltime;
764 	if (load_group(conf, NULL, ep, &psl) != PO_SUCCESS) {
765 		free(query.pq_io_buf);
766 		ea_free_object(ep, EUP_NOALLOC);
767 		return (PO_FAIL);
768 	}
769 
770 	free(query.pq_io_buf);
771 	ea_free_object(ep, EUP_NOALLOC);
772 	/*
773 	 * Now search the dictionary for items that must be removed because
774 	 * they were neither created nor updated.
775 	 */
776 	if ((dead_map = dict_new((int (*)(const void *, const void *))
777 	    pool_elem_compare, (uint64_t (*)(const void *))hash_id)) == NULL) {
778 		pool_seterror(POE_SYSTEM);
779 		return (PO_FAIL);
780 	}
781 	dict_map(prov->pkc_elements, find_dead_elems, dead_map);
782 
783 	if (dict_length(dead_map) > 0) {
784 		dict_map(dead_map, remove_dead_elems, NULL);
785 	}
786 	dict_free(&dead_map);
787 
788 	return (PO_SUCCESS);
789 }
790 
791 /*
792  * Rely on the kernel to always keep a kernel configuration valid.
793  * Returns PO_SUCCESS
794  */
795 /* ARGSUSED */
796 int
797 pool_knl_validate(const pool_conf_t *conf, pool_valid_level_t level)
798 {
799 	return ((conf->pc_state == POF_INVALID) ? PO_FAIL : PO_SUCCESS);
800 }
801 
802 /*
803  * Process all the outstanding transactions in the log. If the processing
804  * fails, then attempt to rollback and "undo" the changes.
805  */
806 int
807 pool_knl_commit(pool_conf_t *conf)
808 {
809 	pool_knl_connection_t *prov = (pool_knl_connection_t *)conf->pc_prov;
810 	int lock = 1;
811 
812 	/*
813 	 * Lock the kernel state for the commit
814 	 */
815 	if (ioctl(prov->pkc_fd, POOL_COMMIT, lock) < 0) {
816 		pool_seterror(POE_SYSTEM);
817 		return (PO_FAIL);
818 	}
819 	lock = 0;
820 	/*
821 	 * If the state is LS_FAIL, then try to recover before
822 	 * performing the commit.
823 	 */
824 	if (prov->pkc_log->l_state == LS_FAIL) {
825 		if (pool_knl_recover(conf) == PO_FAIL) {
826 			/*
827 			 * Unlock the kernel state for the
828 			 * commit. Assert that this * can't fail,
829 			 * since if it ever does fail the library is
830 			 * unusable.
831 			 */
832 			assert(ioctl(prov->pkc_fd, POOL_COMMIT, lock) >= 0);
833 		}
834 	}
835 	/*
836 	 * Commit the log
837 	 */
838 	if (log_walk(prov->pkc_log, log_item_commit) != PO_SUCCESS) {
839 		(void) pool_knl_recover(conf);
840 		/*
841 		 * Unlock the kernel state for the commit. Assert that
842 		 * this can't fail, since if it ever does fail the
843 		 * library is unusable.
844 		 */
845 		assert(ioctl(prov->pkc_fd, POOL_COMMIT, lock) >= 0);
846 		pool_seterror(POE_SYSTEM);
847 		return (PO_FAIL);
848 	}
849 	/*
850 	 * Unlock the kernel state for the commit. Assert that this
851 	 * can't fail, since if it ever does fail the library is
852 	 * unusable.
853 	 */
854 	assert(ioctl(prov->pkc_fd, POOL_COMMIT, lock) >= 0);
855 	/*
856 	 * Release the log resources
857 	 */
858 	(void) log_walk(prov->pkc_log, log_item_release);
859 	log_empty(prov->pkc_log);
860 	return (PO_SUCCESS);
861 }
862 
863 /*
864  * prop_build_cb() is designed to be called from
865  * pool_walk_properties(). The property value is used to put an XML
866  * property on the supplied ktx_node. This is an essential part of the
867  * mechanism used to export a kernel configuration in libpool XML
868  * form.
869  */
870 /* ARGSUSED */
871 static int
872 prop_build_cb(pool_conf_t *UNUSED, pool_elem_t *pe, const char *name,
873     pool_value_t *pval, void *user)
874 {
875 	struct knl_to_xml *info = (struct knl_to_xml *)user;
876 
877 	return (pool_knl_put_xml_property((pool_elem_t *)pe, info->ktx_node,
878 		name, pval));
879 }
880 
881 /*
882  * Duplicate some of the functionality from pool_xml_put_property()
883  * (see pool_xml.c) to allow a kernel configuration to add XML nodes
884  * to an XML tree which represents the kernel configuration. This is
885  * an essential part of the mechanism used to export a kernel
886  * configuration in libpool XML form.
887  */
888 int
889 pool_knl_put_xml_property(pool_elem_t *pe, xmlNodePtr node, const char *name,
890     const pool_value_t *val)
891 {
892 
893 	/*
894 	 * "type" is a special attribute which is not visible ever outside of
895 	 * libpool. Use the specific type accessor function.
896 	 */
897 	if (strcmp(name, c_type) == 0) {
898 		return (pool_xml_set_attr(node, BAD_CAST name,
899 		    val));
900 	}
901 	if (is_ns_property(pe, name) != NULL) {	/* in ns */
902 		if (pool_xml_set_attr(node,
903 		    BAD_CAST property_name_minus_ns(pe, name), val) == PO_FAIL)
904 			return (pool_xml_set_prop(node, BAD_CAST name,
905 			    val));
906 	} else
907 		return (pool_xml_set_prop(node, BAD_CAST name, val));
908 	return (PO_SUCCESS);
909 }
910 
911 /*
912  * Export the kernel configuration as an XML file. The configuration
913  * is used to build an XML document in memory. This document is then
914  * saved to the supplied location.
915  */
916 int
917 pool_knl_export(const pool_conf_t *conf, const char *location,
918     pool_export_format_t fmt)
919 {
920 	xmlNodePtr node_comment;
921 	xmlNodePtr system;
922 	int ret;
923 	pool_t **ps;
924 	pool_resource_t **rs;
925 	uint_t nelem;
926 	int i;
927 	struct knl_to_xml info;
928 	char_buf_t *cb = NULL;
929 	xmlValidCtxtPtr cvp;
930 
931 	xml_init();
932 
933 
934 	switch (fmt) {
935 	case POX_NATIVE:
936 		info.ktx_doc = xmlNewDoc(BAD_CAST "1.0");
937 		xmlCreateIntSubset(info.ktx_doc, BAD_CAST "system",
938 		    BAD_CAST "-//Sun Microsystems Inc//DTD Resource "
939 		    "Management All//EN",
940 		    BAD_CAST dtd_location);
941 
942 		if ((cvp = xmlNewValidCtxt()) == NULL) {
943 			xmlFreeDoc(info.ktx_doc);
944 			pool_seterror(POE_DATASTORE);
945 			return (PO_FAIL);
946 		}
947 		/*
948 		 * Call xmlValidateDocument() to force the parsing of
949 		 * the DTD. Ignore errors and warning messages as we
950 		 * know the document isn't valid.
951 		 */
952 		(void) xmlValidateDocument(cvp, info.ktx_doc);
953 		xmlFreeValidCtxt(cvp);
954 		if ((info.ktx_node = node_create(NULL, BAD_CAST "system")) ==
955 		    NULL) {
956 			xmlFreeDoc(info.ktx_doc);
957 			pool_seterror(POE_DATASTORE);
958 			return (PO_FAIL);
959 		}
960 
961 		system = info.ktx_node;
962 		info.ktx_doc->_private = (void *)conf;
963 
964 		xmlDocSetRootElement(info.ktx_doc, info.ktx_node);
965 		xmlSetProp(info.ktx_node, BAD_CAST c_ref_id, BAD_CAST "dummy");
966 		if ((node_comment = xmlNewDocComment(info.ktx_doc,
967 		    BAD_CAST "\nConfiguration for pools facility. Do NOT"
968 		    " edit this file by hand - use poolcfg(1)"
969 		    " or libpool(3POOL) instead.\n")) == NULL) {
970 			xmlFreeDoc(info.ktx_doc);
971 			pool_seterror(POE_DATASTORE);
972 			return (PO_FAIL);
973 		}
974 		if (xmlAddPrevSibling(info.ktx_node, node_comment) == NULL) {
975 			xmlFree(node_comment);
976 			xmlFreeDoc(info.ktx_doc);
977 			pool_seterror(POE_DATASTORE);
978 			return (PO_FAIL);
979 		}
980 		if (pool_walk_any_properties((pool_conf_t *)conf,
981 		    pool_conf_to_elem(conf), &info, prop_build_cb, 1) ==
982 		    PO_FAIL) {
983 			xmlFreeDoc(info.ktx_doc);
984 			return (PO_FAIL);
985 		}
986 		if ((cb = alloc_char_buf(CB_DEFAULT_LEN)) == NULL) {
987 			xmlFreeDoc(info.ktx_doc);
988 			return (PO_FAIL);
989 		}
990 		/*
991 		 * Now add pool details
992 		 */
993 		if ((ps = pool_query_pools(conf, &nelem, NULL)) != NULL) {
994 			for (i = 0; i < nelem; i++) {
995 				pool_elem_t *elem = TO_ELEM(ps[i]);
996 				uint_t nreselem;
997 				const char *sep = "";
998 				int j;
999 
1000 				if ((info.ktx_node = node_create(system,
1001 				    BAD_CAST element_class_tags
1002 				    [pool_elem_class(elem)])) == NULL) {
1003 					free(ps);
1004 					free_char_buf(cb);
1005 					xmlFreeDoc(info.ktx_doc);
1006 					pool_seterror(POE_DATASTORE);
1007 					return (PO_FAIL);
1008 				}
1009 				if (pool_walk_any_properties(
1010 				    (pool_conf_t *)conf,
1011 				    elem, &info, prop_build_cb, 1) == PO_FAIL) {
1012 					free(ps);
1013 					free_char_buf(cb);
1014 					xmlFreeDoc(info.ktx_doc);
1015 					return (PO_FAIL);
1016 				}
1017 				/*
1018 				 * TODO: pset specific res manipulation
1019 				 */
1020 				if ((rs = pool_query_pool_resources(conf, ps[i],
1021 				    &nreselem, NULL)) == NULL) {
1022 					free(ps);
1023 					free_char_buf(cb);
1024 					xmlFreeDoc(info.ktx_doc);
1025 					pool_seterror(POE_INVALID_CONF);
1026 					return (PO_FAIL);
1027 				}
1028 				if (set_char_buf(cb, "") == PO_FAIL) {
1029 					free(rs);
1030 					free(ps);
1031 					free_char_buf(cb);
1032 					xmlFreeDoc(info.ktx_doc);
1033 					return (PO_FAIL);
1034 				}
1035 				for (j = 0; j < nreselem; j++) {
1036 					pool_elem_t *reselem = TO_ELEM(rs[j]);
1037 					if (append_char_buf(cb, "%s%s_%d", sep,
1038 					    pool_elem_class_string(reselem),
1039 					    (int)elem_get_sysid(reselem)) ==
1040 					    PO_FAIL) {
1041 						free(rs);
1042 						free(ps);
1043 						free_char_buf(cb);
1044 						xmlFreeDoc(info.ktx_doc);
1045 						return (PO_FAIL);
1046 					}
1047 					sep = " ";
1048 				}
1049 				free(rs);
1050 				xmlSetProp(info.ktx_node, BAD_CAST "res",
1051 				    BAD_CAST cb->cb_buf);
1052 				if (set_char_buf(cb, "%s_%d",
1053 				    pool_elem_class_string(elem),
1054 				    (int)elem_get_sysid(elem)) == PO_FAIL) {
1055 					free(ps);
1056 					free_char_buf(cb);
1057 					xmlFreeDoc(info.ktx_doc);
1058 					return (PO_FAIL);
1059 				}
1060 				xmlSetProp(info.ktx_node, BAD_CAST c_ref_id,
1061 				    BAD_CAST  cb->cb_buf);
1062 			}
1063 			free(ps);
1064 		}
1065 		/*
1066 		 * Now add resource details (including components)
1067 		 */
1068 		if ((rs = pool_query_resources(conf, &nelem, NULL)) != NULL) {
1069 			for (i = 0; i < nelem; i++) {
1070 				pool_elem_t *elem = TO_ELEM(rs[i]);
1071 				pool_component_t **cs = NULL;
1072 				uint_t ncompelem;
1073 				int j;
1074 
1075 				if ((info.ktx_node = node_create(system,
1076 				    BAD_CAST element_class_tags
1077 				    [pool_elem_class(elem)])) == NULL) {
1078 					free(rs);
1079 					free_char_buf(cb);
1080 					xmlFreeDoc(info.ktx_doc);
1081 					pool_seterror(POE_DATASTORE);
1082 					return (PO_FAIL);
1083 				}
1084 				if (pool_walk_any_properties(
1085 				    (pool_conf_t *)conf,
1086 				    elem, &info, prop_build_cb, 1) == PO_FAIL) {
1087 					free(rs);
1088 					free_char_buf(cb);
1089 					xmlFreeDoc(info.ktx_doc);
1090 					return (PO_FAIL);
1091 				}
1092 				if (set_char_buf(cb, "%s_%d",
1093 				    pool_elem_class_string(elem),
1094 				    (int)elem_get_sysid(elem)) == PO_FAIL) {
1095 					free(rs);
1096 					free_char_buf(cb);
1097 					xmlFreeDoc(info.ktx_doc);
1098 					return (PO_FAIL);
1099 				}
1100 				xmlSetProp(info.ktx_node, BAD_CAST c_ref_id,
1101 				    BAD_CAST  cb->cb_buf);
1102 				if ((cs = pool_query_resource_components(conf,
1103 				    rs[i], &ncompelem, NULL)) != NULL) {
1104 					xmlNodePtr resource = info.ktx_node;
1105 
1106 					for (j = 0; j < ncompelem; j++) {
1107 						pool_elem_t *compelem =
1108 						    TO_ELEM(cs[j]);
1109 						if ((info.ktx_node =
1110 						    node_create(resource,
1111 						    BAD_CAST element_class_tags
1112 						    [pool_elem_class(
1113 						    compelem)])) == NULL) {
1114 							pool_seterror(
1115 							    POE_DATASTORE);
1116 							free(rs);
1117 							free(cs);
1118 							free_char_buf(cb);
1119 							xmlFreeDoc(info.
1120 							    ktx_doc);
1121 							return (PO_FAIL);
1122 						}
1123 						if (pool_walk_any_properties(
1124 						    (pool_conf_t *)conf,
1125 						    compelem, &info,
1126 						    prop_build_cb, 1) ==
1127 						    PO_FAIL) {
1128 							free(rs);
1129 							free(cs);
1130 							free_char_buf(cb);
1131 							xmlFreeDoc(info.
1132 							    ktx_doc);
1133 							return (PO_FAIL);
1134 						}
1135 						if (set_char_buf(cb, "%s_%d",
1136 						    pool_elem_class_string(
1137 						    compelem),
1138 						    (int)elem_get_sysid(
1139 						    compelem)) == PO_FAIL) {
1140 							free(rs);
1141 							free(cs);
1142 							free_char_buf(cb);
1143 							xmlFreeDoc(info.
1144 							    ktx_doc);
1145 							return (PO_FAIL);
1146 						}
1147 						xmlSetProp(info.ktx_node,
1148 						    BAD_CAST c_ref_id,
1149 						    BAD_CAST  cb->cb_buf);
1150 					}
1151 					free(cs);
1152 				}
1153 			}
1154 			free(rs);
1155 		}
1156 		free_char_buf(cb);
1157 		/*
1158 		 * Set up the message handlers prior to calling
1159 		 * xmlValidateDocument()
1160 		 */
1161 		if ((cvp = xmlNewValidCtxt()) == NULL) {
1162 			xmlFreeDoc(info.ktx_doc);
1163 			pool_seterror(POE_DATASTORE);
1164 			return (PO_FAIL);
1165 		}
1166 		cvp->error    = pool_error_func;
1167 		cvp->warning  = pool_error_func;
1168 		if (xmlValidateDocument(cvp, info.ktx_doc) == 0) {
1169 			xmlFreeValidCtxt(cvp);
1170 			xmlFreeDoc(info.ktx_doc);
1171 			pool_seterror(POE_INVALID_CONF);
1172 			return (PO_FAIL);
1173 		}
1174 		xmlFreeValidCtxt(cvp);
1175 		ret = xmlSaveFormatFile(location, info.ktx_doc, 1);
1176 		xmlFreeDoc(info.ktx_doc);
1177 		if (ret == -1) {
1178 			pool_seterror(POE_SYSTEM);
1179 			return (PO_FAIL);
1180 		}
1181 		return (PO_SUCCESS);
1182 	default:
1183 		pool_seterror(POE_BADPARAM);
1184 		return (PO_FAIL);
1185 	}
1186 }
1187 
1188 /*
1189  * Rollback the changes to the kernel
1190  */
1191 int
1192 pool_knl_recover(pool_conf_t *conf)
1193 {
1194 	pool_knl_connection_t *prov = (pool_knl_connection_t *)conf->pc_prov;
1195 
1196 	prov->pkc_log->l_state = LS_RECOVER;
1197 	if (log_reverse_walk(prov->pkc_log, log_item_undo) != PO_SUCCESS) {
1198 		dprintf("Library configuration consistency error\n");
1199 		prov->pkc_log->l_state = LS_FAIL;
1200 		pool_seterror(POE_INVALID_CONF);
1201 		return (PO_FAIL);
1202 	}
1203 	prov->pkc_log->l_state = LS_DO;
1204 	return (PO_SUCCESS);
1205 }
1206 
1207 /*
1208  * Rollback the changes to the configuration
1209  */
1210 int
1211 pool_knl_rollback(pool_conf_t *conf)
1212 {
1213 	pool_knl_connection_t *prov = (pool_knl_connection_t *)conf->pc_prov;
1214 
1215 	prov->pkc_log->l_state = LS_UNDO;
1216 	if (log_reverse_walk(prov->pkc_log, log_item_undo) != PO_SUCCESS) {
1217 		dprintf("Kernel configuration consistency error\n");
1218 		(void) log_walk(prov->pkc_log, log_item_release);
1219 		log_empty(prov->pkc_log);
1220 		prov->pkc_log->l_state = LS_FAIL;
1221 		pool_seterror(POE_INVALID_CONF);
1222 		return (PO_FAIL);
1223 	}
1224 	(void) log_walk(prov->pkc_log, log_item_release);
1225 	log_empty(prov->pkc_log);
1226 	prov->pkc_log->l_state = LS_DO;
1227 	return (PO_SUCCESS);
1228 }
1229 
1230 /*
1231  * Callback used to build the result set for a query. Each invocation will
1232  * supply a candidate element for inclusion. The element is filtered by:
1233  * - class
1234  * - properties
1235  * If the element "matches" the target, then it is added to the result
1236  * set, otherwise it is ignored.
1237  */
1238 /* ARGSUSED1 */
1239 static void
1240 build_result_set(const void *key, void **value, void *cl)
1241 {
1242 	struct query_obj *qo = (struct query_obj *)cl;
1243 	pool_knl_elem_t *pke = (pool_knl_elem_t *)key;
1244 
1245 	/*
1246 	 * Check to see if it's the right class of element
1247 	 */
1248 	if (qo->classes & (1 << pool_elem_class((pool_elem_t *)key))) {
1249 		int i;
1250 		/*
1251 		 * Now check to see if the src element is correct. If no src
1252 		 * element is supplied, ignore this check
1253 		 */
1254 		if (qo->src) {
1255 			pool_knl_elem_t *parent;
1256 
1257 			for (parent = pke; parent != NULL;
1258 			    parent = parent->pke_parent) {
1259 				if (parent == (pool_knl_elem_t *)qo->src)
1260 					break;
1261 			}
1262 			if (parent == NULL)
1263 				return;
1264 		}
1265 		/*
1266 		 * Now check for property matches (if there are any specified)
1267 		 */
1268 		if (qo->props) {
1269 			int matched = PO_TRUE;
1270 			for (i = 0; qo->props[i] != NULL; i++) {
1271 				pool_value_t val = POOL_VALUE_INITIALIZER;
1272 
1273 				if (pool_get_property(TO_CONF(TO_ELEM(pke)),
1274 				    (pool_elem_t *)pke,
1275 				    pool_value_get_name(qo->props[i]), &val) ==
1276 				    POC_INVAL) {
1277 					matched = PO_FALSE;
1278 					break;
1279 				} else {
1280 					if (pool_value_equal(qo->props[i],
1281 					    &val) != PO_TRUE) {
1282 						matched = PO_FALSE;
1283 						break;
1284 					}
1285 				}
1286 			}
1287 			if (matched == PO_TRUE)
1288 				(void) pool_knl_result_set_append(qo->rs,
1289 				(pool_knl_elem_t *)key);
1290 		} else {
1291 			(void) pool_knl_result_set_append(qo->rs,
1292 			    (pool_knl_elem_t *)key);
1293 		}
1294 	}
1295 }
1296 
1297 /*
1298  * Execute the supplied query and return a result set which contains
1299  * all qualifying elements.
1300  */
1301 pool_result_set_t *
1302 pool_knl_exec_query(const pool_conf_t *conf, const pool_elem_t *src,
1303     const char *src_attr, pool_elem_class_t classes, pool_value_t **props)
1304 {
1305 	pool_knl_result_set_t *rs;
1306 	pool_knl_connection_t *prov = (pool_knl_connection_t *)conf->pc_prov;
1307 	struct query_obj qo;
1308 	int matched = PO_TRUE;
1309 
1310 	/*
1311 	 * Have a buffer at this point, that we can use
1312 	 */
1313 	if ((rs = pool_knl_result_set_alloc(conf)) == NULL) {
1314 		return (NULL);
1315 	}
1316 	qo.conf = conf;
1317 	qo.src = src;
1318 	qo.src_attr = src_attr;
1319 	qo.classes = classes;
1320 	qo.props = props;
1321 	qo.rs = rs;
1322 	if (src_attr != NULL) {
1323 		pool_knl_pool_t *pkp = (pool_knl_pool_t *)src;
1324 
1325 		/*
1326 		 * Note: This logic is resource specific and must be
1327 		 * extended for additional resource types.
1328 		 */
1329 		/*
1330 		 * Check for property matches (if there are any specified)
1331 		 */
1332 		if (props) {
1333 			int i;
1334 
1335 			for (i = 0; props[i] != NULL; i++) {
1336 				pool_value_t val = POOL_VALUE_INITIALIZER;
1337 
1338 				if (pool_get_property(conf,
1339 				    (pool_elem_t *)pkp->pkp_assoc[PREC_PSET],
1340 				    pool_value_get_name(props[i]), &val) ==
1341 				    POC_INVAL) {
1342 					matched = PO_FALSE;
1343 					break;
1344 				} else {
1345 					if (pool_value_equal(props[i],
1346 					    &val) != PO_TRUE) {
1347 						matched = PO_FALSE;
1348 						break;
1349 					}
1350 				}
1351 			}
1352 		}
1353 
1354 		if (matched == PO_TRUE)
1355 			(void) pool_knl_result_set_append(rs,
1356 			    (pool_knl_elem_t *)pkp->pkp_assoc[PREC_PSET]);
1357 	} else
1358 		dict_map(prov->pkc_elements, build_result_set, &qo);
1359 
1360 	return ((pool_result_set_t *)rs);
1361 }
1362 
1363 /*
1364  * Callback function intended to be used from pool_walk_pools(). If
1365  * the supplied pool is not the default pool attempt to destroy it.
1366  */
1367 /*ARGSUSED*/
1368 static int
1369 destroy_pool_cb(pool_conf_t *conf, pool_t *pool, void *unused)
1370 {
1371 	if (elem_is_default(TO_ELEM(pool)) != PO_TRUE)
1372 		return (pool_destroy(conf, pool));
1373 	/*
1374 	 * Return PO_SUCCESS even though we don't delete the default
1375 	 * pool so that the walk continues
1376 	 */
1377 	return (PO_SUCCESS);
1378 }
1379 
1380 /*
1381  * Remove the configuration details. This means remove all elements
1382  * apart from the system elements.
1383  */
1384 int
1385 pool_knl_remove(pool_conf_t *conf)
1386 {
1387 	uint_t i, nelem;
1388 	pool_resource_t **resources;
1389 
1390 	conf->pc_state = POF_DESTROY;
1391 	if ((resources = pool_query_resources(conf, &nelem, NULL)) != NULL) {
1392 		for (i = 0; i < nelem; i++) {
1393 			if (resource_is_system(resources[i]) == PO_FALSE)
1394 				if (pool_resource_destroy(conf, resources[i]) !=
1395 				    PO_SUCCESS) {
1396 					pool_seterror(POE_INVALID_CONF);
1397 					return (PO_FAIL);
1398 				}
1399 		}
1400 		free(resources);
1401 	}
1402 	(void) pool_walk_pools(conf, conf, destroy_pool_cb);
1403 	if (pool_conf_commit(conf, PO_FALSE) != PO_SUCCESS)
1404 		return (PO_FAIL);
1405 
1406 	if (pool_conf_close(conf) != PO_SUCCESS)
1407 		return (PO_FAIL);
1408 
1409 	return (PO_SUCCESS);
1410 }
1411 
1412 /*
1413  * Determine the name of the pool to which the supplied pid is
1414  * bound. If it cannot be determined return NULL.
1415  */
1416 char *
1417 pool_knl_get_binding(pool_conf_t *conf, pid_t pid)
1418 {
1419 	pool_knl_connection_t *prov = (pool_knl_connection_t *)conf->pc_prov;
1420 	const char *sval;
1421 	char *name = NULL;
1422 	pool_bindq_t bindq;
1423 	pool_value_t *props[] = { NULL, NULL };
1424 	uint_t nelem = 0;
1425 	pool_t **pools;
1426 	pool_value_t val = POOL_VALUE_INITIALIZER;
1427 
1428 	props[0] = &val;
1429 
1430 	bindq.pb_o_id_type = P_PID;
1431 	bindq.pb_o_id = pid;
1432 	if (ioctl(prov->pkc_fd, POOL_BINDQ, &bindq) < 0) {
1433 		pool_seterror(POE_SYSTEM);
1434 		return (NULL);
1435 	}
1436 
1437 	if (pool_value_set_name(props[0], "pool.sys_id") != PO_SUCCESS) {
1438 		return (NULL);
1439 	}
1440 	pool_value_set_int64(props[0], bindq.pb_i_id);
1441 	if ((pools = pool_query_pools(conf, &nelem, props)) == NULL) {
1442 		pool_seterror(POE_BADPARAM);
1443 		return (NULL);
1444 	}
1445 
1446 	if (nelem != 1) {
1447 		free(pools);
1448 		pool_seterror(POE_INVALID_CONF);
1449 		return (NULL);
1450 	}
1451 	if (pool_get_ns_property(TO_ELEM(pools[0]), c_name, props[0])
1452 	    == POC_INVAL) {
1453 		free(pools);
1454 		return (NULL);
1455 	}
1456 	if (pool_value_get_string(props[0], &sval) != PO_SUCCESS) {
1457 		free(pools);
1458 		return (NULL);
1459 	}
1460 	if ((name = strdup(sval)) == NULL) {
1461 		free(pools);
1462 		pool_seterror(POE_SYSTEM);
1463 		return (NULL);
1464 	}
1465 	return (name);
1466 }
1467 
1468 /*
1469  * Bind idtype id to the pool name.
1470  */
1471 int
1472 pool_knl_set_binding(pool_conf_t *conf, const char *pool_name, idtype_t idtype,
1473     id_t id)
1474 {
1475 	pool_knl_connection_t *prov = (pool_knl_connection_t *)conf->pc_prov;
1476 	pool_bind_t bind;
1477 	pool_t *pool;
1478 	int ret;
1479 
1480 	if ((pool = pool_get_pool(conf, pool_name)) == NULL)
1481 		return (PO_FAIL);
1482 
1483 	bind.pb_o_id_type = idtype;
1484 	bind.pb_o_id = id;
1485 	bind.pb_o_pool_id = elem_get_sysid(TO_ELEM(pool));
1486 
1487 	while ((ret = ioctl(prov->pkc_fd, POOL_BIND, &bind)) < 0 &&
1488 	    errno == EAGAIN);
1489 	if (ret < 0) {
1490 		pool_seterror(POE_SYSTEM);
1491 		return (PO_FAIL);
1492 	}
1493 	return (PO_SUCCESS);
1494 }
1495 
1496 /*
1497  * pool_knl_get_resource_binding() returns the binding for a pid to
1498  * the supplied type of resource. If a binding cannot be determined,
1499  * NULL is returned.
1500  */
1501 char *
1502 pool_knl_get_resource_binding(pool_conf_t *conf,
1503     pool_resource_elem_class_t type, pid_t pid)
1504 {
1505 	pool_knl_connection_t *prov = (pool_knl_connection_t *)conf->pc_prov;
1506 	const char *sval;
1507 	char *name = NULL;
1508 	pool_bindq_t bindq;
1509 	pool_value_t *props[] = { NULL, NULL };
1510 	uint_t nelem = 0;
1511 	pool_t **pools;
1512 	pool_resource_t **resources;
1513 	pool_value_t val = POOL_VALUE_INITIALIZER;
1514 
1515 	props[0] = &val;
1516 	bindq.pb_o_id_type = P_PID;
1517 	bindq.pb_o_id = pid;
1518 	if (ioctl(prov->pkc_fd, POOL_BINDQ, &bindq) < 0) {
1519 		pool_seterror(POE_SYSTEM);
1520 		return (NULL);
1521 	}
1522 
1523 	if (pool_value_set_name(props[0], "pool.sys_id") != PO_SUCCESS) {
1524 		return (NULL);
1525 	}
1526 	pool_value_set_int64(props[0], bindq.pb_i_id);
1527 	if ((pools = pool_query_pools(conf, &nelem, props)) == NULL) {
1528 		pool_seterror(POE_BADPARAM);
1529 		return (NULL);
1530 	}
1531 
1532 	if (nelem != 1) {
1533 		free(pools);
1534 		pool_seterror(POE_INVALID_CONF);
1535 		return (NULL);
1536 	}
1537 
1538 	if (pool_value_set_string(props[0], pool_resource_type_string(type)) !=
1539 	    PO_SUCCESS ||
1540 	    pool_value_set_name(props[0], c_type) != PO_SUCCESS) {
1541 		free(pools);
1542 		return (NULL);
1543 	}
1544 
1545 	if ((resources = pool_query_pool_resources(conf, pools[0], &nelem,
1546 	    NULL)) == NULL) {
1547 		free(pools);
1548 		pool_seterror(POE_INVALID_CONF);
1549 		return (NULL);
1550 	}
1551 	free(pools);
1552 	if (nelem != 1) {
1553 		free(resources);
1554 		pool_seterror(POE_INVALID_CONF);
1555 		return (NULL);
1556 	}
1557 	if (pool_get_ns_property(TO_ELEM(resources[0]), c_name, props[0]) ==
1558 	    POC_INVAL) {
1559 		free(resources);
1560 		return (NULL);
1561 	}
1562 	free(resources);
1563 	if (pool_value_get_string(props[0], &sval) != PO_SUCCESS) {
1564 		return (NULL);
1565 	}
1566 	if ((name = strdup(sval)) == NULL) {
1567 		pool_seterror(POE_SYSTEM);
1568 		return (NULL);
1569 	}
1570 	return (name);
1571 }
1572 
1573 /*
1574  * Allocate the required library data structure and initialise it.
1575  */
1576 pool_knl_elem_t *
1577 pool_knl_elem_wrap(pool_conf_t *conf, pool_elem_class_t class,
1578     pool_resource_elem_class_t res_class,
1579     pool_component_elem_class_t comp_class)
1580 {
1581 	pool_knl_elem_t *elem;
1582 	pool_elem_t *pe;
1583 
1584 	switch (class) {
1585 	case PEC_SYSTEM:
1586 		if ((elem = malloc(sizeof (pool_knl_system_t))) == NULL) {
1587 			pool_seterror(POE_SYSTEM);
1588 			return (NULL);
1589 		}
1590 		(void) memset(elem, 0, sizeof (pool_knl_system_t));
1591 		break;
1592 	case PEC_POOL:
1593 		if ((elem = malloc(sizeof (pool_knl_pool_t))) == NULL) {
1594 			pool_seterror(POE_SYSTEM);
1595 			return (NULL);
1596 		}
1597 		(void) memset(elem, 0, sizeof (pool_knl_pool_t));
1598 		break;
1599 	case PEC_RES_COMP:
1600 	case PEC_RES_AGG:
1601 		if ((elem = malloc(sizeof (pool_knl_resource_t))) == NULL) {
1602 			pool_seterror(POE_SYSTEM);
1603 			return (NULL);
1604 		}
1605 		(void) memset(elem, 0, sizeof (pool_knl_resource_t));
1606 		break;
1607 	case PEC_COMP:
1608 		if ((elem = malloc(sizeof (pool_knl_component_t))) == NULL) {
1609 			pool_seterror(POE_SYSTEM);
1610 			return (NULL);
1611 		}
1612 		(void) memset(elem, 0, sizeof (pool_knl_component_t));
1613 		break;
1614 	default:
1615 		pool_seterror(POE_BADPARAM);
1616 		return (NULL);
1617 	}
1618 	pe = TO_ELEM(elem);
1619 	pe->pe_conf = conf;
1620 	pe->pe_class = class;
1621 	pe->pe_resource_class = res_class;
1622 	pe->pe_component_class = comp_class;
1623 	/* Set up the function pointers for element manipulation */
1624 	pe->pe_get_prop = pool_knl_get_property;
1625 	pe->pe_put_prop = pool_knl_put_property;
1626 	pe->pe_rm_prop = pool_knl_rm_property;
1627 	pe->pe_get_props = pool_knl_get_properties;
1628 	pe->pe_remove = pool_knl_elem_remove;
1629 	pe->pe_get_container = pool_knl_get_container;
1630 	pe->pe_set_container = pool_knl_set_container;
1631 	/*
1632 	 * Specific initialisation for different types of element
1633 	 */
1634 	if (class == PEC_POOL) {
1635 		pool_knl_pool_t *pp = (pool_knl_pool_t *)elem;
1636 		pp->pp_associate = pool_knl_pool_associate;
1637 		pp->pp_dissociate = pool_knl_pool_dissociate;
1638 		pp->pkp_assoc[PREC_PSET] = (pool_knl_resource_t *)
1639 		    resource_by_sysid(conf, PS_NONE, "pset");
1640 	}
1641 	if (class == PEC_RES_COMP || class == PEC_RES_AGG) {
1642 		pool_knl_resource_t *pr = (pool_knl_resource_t *)elem;
1643 		pr->pr_is_system = pool_knl_resource_is_system;
1644 		pr->pr_can_associate = pool_knl_resource_can_associate;
1645 	}
1646 #if DEBUG
1647 	if (dict_put(((pool_knl_connection_t *)conf->pc_prov)->pkc_leaks,
1648 	    elem, elem) != NULL)
1649 		assert(!"leak map put failed");
1650 	dprintf("allocated %p\n", elem);
1651 #endif	/* DEBUG */
1652 	return (elem);
1653 }
1654 
1655 /*
1656  * Allocate a new pool_knl_elem_t in the supplied configuration of the
1657  * specified class.
1658  * Returns element pointer/NULL
1659  */
1660 pool_elem_t *
1661 pool_knl_elem_create(pool_conf_t *conf, pool_elem_class_t class,
1662     pool_resource_elem_class_t res_class,
1663     pool_component_elem_class_t comp_class)
1664 {
1665 	pool_knl_elem_t *elem;
1666 	pool_create_undo_t *create;
1667 	pool_knl_connection_t *prov = (pool_knl_connection_t *)conf->pc_prov;
1668 	static int id = -3;
1669 	char_buf_t *cb;
1670 
1671 	if ((elem = pool_knl_elem_wrap(conf, class, res_class, comp_class)) ==
1672 	    NULL)
1673 		return (NULL);
1674 
1675 	/*
1676 	 * Allocate an nvlist to hold properties
1677 	 */
1678 	if (nvlist_alloc(&elem->pke_properties, NV_UNIQUE_NAME_TYPE, 0) != 0) {
1679 		pool_knl_elem_free(elem, PO_FALSE);
1680 		pool_seterror(POE_SYSTEM);
1681 		return (NULL);
1682 	}
1683 	/*
1684 	 * Allocate a temporary ID and name until the element is
1685 	 * created for real
1686 	 */
1687 	if ((cb = alloc_char_buf(CB_DEFAULT_LEN)) == NULL) {
1688 		pool_knl_elem_free(elem, PO_TRUE);
1689 		return (NULL);
1690 	}
1691 	if (set_char_buf(cb, "%s.sys_id",
1692 	    pool_elem_class_string((pool_elem_t *)elem)) != PO_SUCCESS) {
1693 		pool_knl_elem_free(elem, PO_TRUE);
1694 		free_char_buf(cb);
1695 		return (NULL);
1696 	}
1697 	(void) nvlist_add_int64(elem->pke_properties, cb->cb_buf, id--);
1698 	if (set_char_buf(cb, "%s.name",
1699 	    pool_elem_class_string((pool_elem_t *)elem)) != PO_SUCCESS) {
1700 		pool_knl_elem_free(elem, PO_TRUE);
1701 		free_char_buf(cb);
1702 		return (NULL);
1703 	}
1704 	(void) nvlist_add_string(elem->pke_properties, cb->cb_buf, "");
1705 	/*
1706 	 * If it's a resource class, it will need an initial size
1707 	 */
1708 	if (class == PEC_RES_COMP || class == PEC_RES_AGG) {
1709 		if (set_char_buf(cb, "%s.size",
1710 		    pool_elem_class_string((pool_elem_t *)elem)) !=
1711 		    PO_SUCCESS) {
1712 			pool_knl_elem_free(elem, PO_TRUE);
1713 			free_char_buf(cb);
1714 			return (NULL);
1715 		}
1716 		(void) nvlist_add_uint64(elem->pke_properties, cb->cb_buf, 0);
1717 	}
1718 	free_char_buf(cb);
1719 
1720 	/*
1721 	 * Register the newly created element
1722 	 */
1723 	if (dict_put(prov->pkc_elements, elem, elem) != NULL) {
1724 		pool_knl_elem_free(elem, PO_TRUE);
1725 		pool_seterror(POE_SYSTEM);
1726 		return (NULL);
1727 	}
1728 
1729 	if (prov->pkc_log->l_state != LS_DO)
1730 		return ((pool_elem_t *)elem);
1731 
1732 	/*
1733 	 * The remaining logic is setting up the arguments for the
1734 	 * POOL_CREATE ioctl and appending the details into the log.
1735 	 */
1736 	if ((create = malloc(sizeof (pool_create_undo_t))) == NULL) {
1737 		pool_seterror(POE_SYSTEM);
1738 		return (NULL);
1739 	}
1740 	create->pcu_ioctl.pc_o_type = class;
1741 	switch (class) {
1742 	case PEC_SYSTEM:
1743 		pool_seterror(POE_BADPARAM);
1744 		free(create);
1745 		return (NULL);
1746 	case PEC_POOL: /* NO-OP */
1747 		break;
1748 	case PEC_RES_COMP:
1749 	case PEC_RES_AGG:
1750 		create->pcu_ioctl.pc_o_sub_type = res_class;
1751 		break;
1752 	case PEC_COMP:
1753 		create->pcu_ioctl.pc_o_sub_type = comp_class;
1754 		break;
1755 	default:
1756 		pool_seterror(POE_BADPARAM);
1757 		free(create);
1758 		return (NULL);
1759 	}
1760 
1761 	create->pcu_elem = (pool_elem_t *)elem;
1762 
1763 	if (log_append(prov->pkc_log, POOL_CREATE, (void *)create) !=
1764 	    PO_SUCCESS) {
1765 		free(create);
1766 		return (NULL);
1767 	}
1768 	return ((pool_elem_t *)elem);
1769 }
1770 
1771 /*
1772  * Remove the details of the element from our userland copy and destroy
1773  * the element (if appropriate) in the kernel.
1774  */
1775 int
1776 pool_knl_elem_remove(pool_elem_t *pe)
1777 {
1778 	pool_knl_connection_t *prov;
1779 	pool_destroy_undo_t *destroy;
1780 
1781 	prov = (pool_knl_connection_t *)(TO_CONF(pe))->pc_prov;
1782 
1783 	if (dict_remove(prov->pkc_elements, pe) == NULL) {
1784 		pool_seterror(POE_SYSTEM);
1785 		return (PO_FAIL);
1786 	}
1787 	if (prov->pkc_log->l_state != LS_DO) {
1788 		return (PO_SUCCESS);
1789 	}
1790 
1791 	/*
1792 	 * The remaining logic is setting up the arguments for the
1793 	 * POOL_DESTROY ioctl and appending the details into the log.
1794 	 */
1795 	if ((destroy = malloc(sizeof (pool_destroy_undo_t))) == NULL) {
1796 		pool_seterror(POE_SYSTEM);
1797 		return (PO_FAIL);
1798 	}
1799 	destroy->pdu_ioctl.pd_o_type = pool_elem_class(pe);
1800 
1801 	if (destroy->pdu_ioctl.pd_o_type == PEC_RES_COMP ||
1802 	    destroy->pdu_ioctl.pd_o_type == PEC_RES_AGG)
1803 		destroy->pdu_ioctl.pd_o_sub_type = pool_resource_elem_class(pe);
1804 
1805 	if (destroy->pdu_ioctl.pd_o_type == PEC_COMP)
1806 		destroy->pdu_ioctl.pd_o_sub_type =
1807 		    pool_component_elem_class(pe);
1808 
1809 	destroy->pdu_elem = pe;
1810 
1811 	if (log_append(prov->pkc_log, POOL_DESTROY, (void *)destroy) !=
1812 	    PO_SUCCESS) {
1813 		free(destroy);
1814 		return (PO_FAIL);
1815 	}
1816 	return (PO_SUCCESS);
1817 }
1818 
1819 /*
1820  * Set the parent of the supplied child to the supplied parent
1821  */
1822 int
1823 pool_knl_set_container(pool_elem_t *pp, pool_elem_t *pc)
1824 {
1825 	pool_knl_elem_t *pkp = (pool_knl_elem_t *)pp;
1826 	pool_knl_elem_t *pkc = (pool_knl_elem_t *)pc;
1827 
1828 	pkc->pke_parent = pkp;
1829 	return (PO_SUCCESS);
1830 }
1831 
1832 /*
1833  * TODO: Needed for msets and ssets.
1834  */
1835 /* ARGSUSED */
1836 int
1837 pool_knl_res_transfer(pool_resource_t *src, pool_resource_t *tgt,
1838     uint64_t size) {
1839 	return (PO_FAIL);
1840 }
1841 
1842 /*
1843  * Transfer resource components from one resource set to another.
1844  */
1845 int
1846 pool_knl_res_xtransfer(pool_resource_t *src, pool_resource_t *tgt,
1847     pool_component_t **rl) {
1848 	pool_elem_t *src_e = TO_ELEM(src);
1849 	pool_elem_t *tgt_e = TO_ELEM(tgt);
1850 	pool_xtransfer_undo_t *xtransfer;
1851 	size_t size;
1852 	pool_knl_connection_t *prov =
1853 	    (pool_knl_connection_t *)TO_CONF(src_e)->pc_prov;
1854 
1855 	if (prov->pkc_log->l_state != LS_DO) {
1856 		/*
1857 		 * Walk the Result Set and move the resource components
1858 		 */
1859 		for (size = 0; rl[size] != NULL; size++) {
1860 			if (pool_set_container(TO_ELEM(tgt),
1861 			    TO_ELEM(rl[size])) == PO_FAIL) {
1862 				return (PO_FAIL);
1863 			}
1864 		}
1865 		return (PO_SUCCESS);
1866 	}
1867 
1868 	/*
1869 	 * The remaining logic is setting up the arguments for the
1870 	 * POOL_XTRANSFER ioctl and appending the details into the log.
1871 	 */
1872 	if ((xtransfer = malloc(sizeof (pool_xtransfer_undo_t))) == NULL) {
1873 		pool_seterror(POE_SYSTEM);
1874 		return (PO_FAIL);
1875 	}
1876 
1877 	if (pool_elem_class(src_e) == PEC_RES_COMP) {
1878 		xtransfer->pxu_ioctl.px_o_id_type =
1879 		    pool_resource_elem_class(src_e);
1880 	} else {
1881 		pool_seterror(POE_BADPARAM);
1882 		return (PO_FAIL);
1883 	}
1884 
1885 
1886 	for (xtransfer->pxu_ioctl.px_o_complist_size = 0;
1887 	    rl[xtransfer->pxu_ioctl.px_o_complist_size] != NULL;
1888 	    xtransfer->pxu_ioctl.px_o_complist_size++)
1889 		/* calculate the size using the terminating NULL */;
1890 	if ((xtransfer->pxu_ioctl.px_o_comp_list =
1891 		calloc(xtransfer->pxu_ioctl.px_o_complist_size,
1892 		sizeof (id_t))) == NULL) {
1893 		pool_seterror(POE_SYSTEM);
1894 		return (PO_FAIL);
1895 	}
1896 	if ((xtransfer->pxu_rl = calloc(
1897 	    xtransfer->pxu_ioctl.px_o_complist_size + 1,
1898 	    sizeof (pool_component_t *))) == NULL) {
1899 		pool_seterror(POE_SYSTEM);
1900 		return (PO_FAIL);
1901 	}
1902 	(void) memcpy(xtransfer->pxu_rl, rl,
1903 	    xtransfer->pxu_ioctl.px_o_complist_size *
1904 	    sizeof (pool_component_t *));
1905 	xtransfer->pxu_src = src_e;
1906 	xtransfer->pxu_tgt = tgt_e;
1907 
1908 	if (log_append(prov->pkc_log, POOL_XTRANSFER, (void *)xtransfer) !=
1909 	    PO_SUCCESS) {
1910 		free(xtransfer);
1911 		return (PO_FAIL);
1912 	}
1913 	for (size = 0; rl[size] != NULL; size++) {
1914 		if (pool_set_container(TO_ELEM(tgt), TO_ELEM(rl[size])) ==
1915 		    PO_FAIL) {
1916 			return (PO_FAIL);
1917 		}
1918 	}
1919 	return (PO_SUCCESS);
1920 }
1921 
1922 /*
1923  * Return the parent of an element.
1924  */
1925 pool_elem_t *
1926 pool_knl_get_container(const pool_elem_t *pe)
1927 {
1928 	pool_knl_elem_t *pke = (pool_knl_elem_t *)pe;
1929 
1930 	return ((pool_elem_t *)pke->pke_parent);
1931 }
1932 
1933 /*
1934  * Note: This function is resource specific, needs extending for other
1935  * resource types
1936  */
1937 int
1938 pool_knl_resource_is_system(const pool_resource_t *pr)
1939 {
1940 	switch (pool_resource_elem_class(TO_ELEM(pr))) {
1941 	case PREC_PSET:
1942 		return (PSID_IS_SYSSET(
1943 		    elem_get_sysid(TO_ELEM(pr))));
1944 	default:
1945 		return (PO_FALSE);
1946 	}
1947 }
1948 
1949 /*
1950  * Note: This function is resource specific, needs extending for other
1951  * resource types
1952  */
1953 int
1954 pool_knl_resource_can_associate(const pool_resource_t *pr)
1955 {
1956 	switch (pool_resource_elem_class(TO_ELEM(pr))) {
1957 	case PREC_PSET:
1958 		return (PO_TRUE);
1959 	default:
1960 		return (PO_FALSE);
1961 	}
1962 }
1963 
1964 /*
1965  * pool_knl_pool_associate() associates the supplied resource to the
1966  * supplied pool.
1967  *
1968  * Returns: PO_SUCCESS/PO_FAIL
1969  */
1970 int
1971 pool_knl_pool_associate(pool_t *pool, const pool_resource_t *resource)
1972 {
1973 	pool_knl_connection_t *prov;
1974 	pool_knl_pool_t *pkp = (pool_knl_pool_t *)pool;
1975 	pool_resource_elem_class_t res_class =
1976 	    pool_resource_elem_class(TO_ELEM(resource));
1977 	pool_assoc_undo_t *assoc;
1978 	pool_knl_resource_t *orig_res = pkp->pkp_assoc[res_class];
1979 
1980 	/*
1981 	 * Are we allowed to associate with this target?
1982 	 */
1983 	if (pool_knl_resource_can_associate(resource) == PO_FALSE) {
1984 		pool_seterror(POE_BADPARAM);
1985 		return (PO_FAIL);
1986 	}
1987 	prov = (pool_knl_connection_t *)(TO_CONF(TO_ELEM(pool)))->pc_prov;
1988 
1989 	if (prov->pkc_log->l_state != LS_DO) {
1990 		pkp->pkp_assoc[res_class] = (pool_knl_resource_t *)resource;
1991 		return (PO_SUCCESS);
1992 	}
1993 
1994 	/*
1995 	 * The remaining logic is setting up the arguments for the
1996 	 * POOL_ASSOC ioctl and appending the details into the log.
1997 	 */
1998 	if ((assoc = malloc(sizeof (pool_assoc_undo_t))) == NULL) {
1999 		pool_seterror(POE_SYSTEM);
2000 		return (PO_FAIL);
2001 	}
2002 	assoc->pau_assoc = TO_ELEM(pool);
2003 	assoc->pau_oldres = (pool_elem_t *)orig_res;
2004 	assoc->pau_newres = TO_ELEM(resource);
2005 
2006 	assoc->pau_ioctl.pa_o_id_type = res_class;
2007 
2008 	if (log_append(prov->pkc_log, POOL_ASSOC, (void *)assoc) !=
2009 	    PO_SUCCESS) {
2010 		free(assoc);
2011 		pkp->pkp_assoc[res_class] = orig_res;
2012 		return (PO_FAIL);
2013 	}
2014 	pkp->pkp_assoc[res_class] = (pool_knl_resource_t *)resource;
2015 	return (PO_SUCCESS);
2016 }
2017 
2018 /*
2019  * pool_knl_pool_dissociate() dissociates the supplied resource from
2020  * the supplied pool.
2021  *
2022  * Returns: PO_SUCCESS/PO_FAIL
2023  */
2024 int
2025 pool_knl_pool_dissociate(pool_t *pool, const pool_resource_t *resource)
2026 {
2027 	pool_knl_connection_t *prov;
2028 	pool_dissoc_undo_t *dissoc;
2029 	pool_knl_pool_t *pkp = (pool_knl_pool_t *)pool;
2030 	pool_resource_t *default_res = (pool_resource_t *)get_default_resource(
2031 	    resource);
2032 	pool_resource_elem_class_t res_class =
2033 	    pool_resource_elem_class(TO_ELEM(resource));
2034 
2035 	prov = (pool_knl_connection_t *)(TO_CONF(TO_ELEM(pool)))->pc_prov;
2036 
2037 	if (prov->pkc_log->l_state != LS_DO) {
2038 		pkp->pkp_assoc[res_class] = (pool_knl_resource_t *)default_res;
2039 		return (PO_SUCCESS);
2040 	}
2041 	/*
2042 	 * The remaining logic is setting up the arguments for the
2043 	 * POOL_DISSOC ioctl and appending the details into the log.
2044 	 */
2045 	if ((dissoc = malloc(sizeof (pool_dissoc_undo_t))) == NULL) {
2046 		pool_seterror(POE_SYSTEM);
2047 		return (PO_FAIL);
2048 	}
2049 	dissoc->pdu_dissoc = TO_ELEM(pool);
2050 	dissoc->pdu_oldres = TO_ELEM(resource);
2051 	dissoc->pdu_newres = TO_ELEM(default_res);
2052 
2053 	dissoc->pdu_ioctl.pd_o_id_type = res_class;
2054 
2055 	if (log_append(prov->pkc_log, POOL_DISSOC, (void *)dissoc) !=
2056 	    PO_SUCCESS) {
2057 		free(dissoc);
2058 		pkp->pkp_assoc[res_class] = (pool_knl_resource_t *)resource;
2059 		return (PO_FAIL);
2060 	}
2061 
2062 	/*
2063 	 * Update our local copy
2064 	 */
2065 	pkp->pkp_assoc[res_class] = (pool_knl_resource_t *)default_res;
2066 	return (PO_SUCCESS);
2067 }
2068 
2069 /*
2070  * Allocate a data provider for the supplied configuration and optionally
2071  * discover resources.
2072  * The data provider is the cross over point from the "abstract" configuration
2073  * functions into the data representation specific manipulation routines.
2074  * This function sets up all the required pointers to create a kernel aware
2075  * data provider.
2076  * Returns PO_SUCCESS/PO_FAIL
2077  */
2078 int
2079 pool_knl_connection_alloc(pool_conf_t *conf, int oflags)
2080 {
2081 	pool_knl_connection_t *prov;
2082 
2083 	if ((prov = malloc(sizeof (pool_knl_connection_t))) == NULL) {
2084 		pool_seterror(POE_SYSTEM);
2085 		return (PO_FAIL);
2086 	}
2087 	(void) memset(prov, 0, sizeof (pool_knl_connection_t));
2088 	/*
2089 	 * Initialise data members
2090 	 */
2091 	prov->pc_name = strdup("kernel");
2092 	prov->pc_store_type = KERNEL_DATA_STORE;
2093 	prov->pc_oflags = oflags;
2094 	/*
2095 	 * Initialise function pointers
2096 	 */
2097 	prov->pc_close = pool_knl_close;
2098 	prov->pc_validate = pool_knl_validate;
2099 	prov->pc_commit = pool_knl_commit;
2100 	prov->pc_export = pool_knl_export;
2101 	prov->pc_rollback = pool_knl_rollback;
2102 	prov->pc_exec_query = pool_knl_exec_query;
2103 	prov->pc_elem_create = pool_knl_elem_create;
2104 	prov->pc_remove = pool_knl_remove;
2105 	prov->pc_res_xfer = pool_knl_res_transfer;
2106 	prov->pc_res_xxfer = pool_knl_res_xtransfer;
2107 	prov->pc_get_binding = pool_knl_get_binding;
2108 	prov->pc_set_binding = pool_knl_set_binding;
2109 	prov->pc_get_resource_binding = pool_knl_get_resource_binding;
2110 	/*
2111 	 * Associate the provider to it's configuration
2112 	 */
2113 	conf->pc_prov = (pool_connection_t *)prov;
2114 	/*
2115 	 * End of common initialisation
2116 	 */
2117 	/*
2118 	 * Attempt to open the pseudo device, if the configuration is opened
2119 	 * readonly then try to open an info device, otherwise try to open
2120 	 * the writeable device.
2121 	 */
2122 	if (oflags & PO_RDWR) {
2123 		if ((prov->pkc_fd = blocking_open(pool_dynamic_location(),
2124 		    O_RDWR)) < 0) {
2125 			free(prov);
2126 			conf->pc_prov = NULL;
2127 			pool_seterror(POE_SYSTEM);
2128 			return (PO_FAIL);
2129 		}
2130 	} else {
2131 		if ((prov->pkc_fd = open(pool_info_location, O_RDWR)) < 0) {
2132 			free(prov);
2133 			conf->pc_prov = NULL;
2134 			pool_seterror(POE_SYSTEM);
2135 			return (PO_FAIL);
2136 		}
2137 	}
2138 	/*
2139 	 * Allocate the element dictionary
2140 	 */
2141 	if ((prov->pkc_elements = dict_new((int (*)(const void *, const void *))
2142 	    pool_elem_compare, (uint64_t (*)(const void *))hash_id)) == NULL) {
2143 		(void) close(prov->pkc_fd);
2144 		free(prov);
2145 		conf->pc_prov = NULL;
2146 		pool_seterror(POE_SYSTEM);
2147 		return (PO_FAIL);
2148 	}
2149 #if DEBUG
2150 	if ((prov->pkc_leaks = dict_new(NULL, NULL)) == NULL) {
2151 		dict_free(&prov->pkc_elements);
2152 		(void) close(prov->pkc_fd);
2153 		free(prov);
2154 		conf->pc_prov = NULL;
2155 		pool_seterror(POE_SYSTEM);
2156 		return (PO_FAIL);
2157 	}
2158 #endif	/* DEBUG */
2159 	/*
2160 	 * Allocate the transaction log
2161 	 */
2162 	if ((prov->pkc_log = log_alloc(conf)) == NULL) {
2163 #if DEBUG
2164 		dict_free(&prov->pkc_leaks);
2165 #endif	/* DEBUG */
2166 		dict_free(&prov->pkc_elements);
2167 		(void) close(prov->pkc_fd);
2168 		free(prov);
2169 		conf->pc_prov = NULL;
2170 		return (PO_FAIL);
2171 	}
2172 	/*
2173 	 * At this point the configuration provider has been initialized,
2174 	 * mark the configuration as valid so that the various routines
2175 	 * which rely on a valid configuration will work correctly.
2176 	 */
2177 	conf->pc_state = POF_VALID;
2178 	/*
2179 	 * Update the library snapshot from the kernel
2180 	 */
2181 	if (pool_knl_update(conf, NULL) != PO_SUCCESS) {
2182 #if DEBUG
2183 		dict_free(&prov->pkc_leaks);
2184 #endif	/* DEBUG */
2185 		dict_free(&prov->pkc_elements);
2186 		(void) close(prov->pkc_fd);
2187 		free(prov);
2188 		conf->pc_prov = NULL;
2189 		conf->pc_state = POF_INVALID;
2190 		return (PO_FAIL);
2191 	}
2192 	return (PO_SUCCESS);
2193 }
2194 
2195 #if DEBUG
2196 static void
2197 pool_knl_elem_printf_cb(const void *key, void **value, void *cl)
2198 {
2199 	pool_knl_elem_t *pke = (pool_knl_elem_t *)key;
2200 	dict_hdl_t *map = (dict_hdl_t *)cl;
2201 
2202 	dprintf("leak elem:%p\n", pke);
2203 	if (pke->pke_properties != NULL) {
2204 		nvlist_print(stdout, pke->pke_properties);
2205 	} else
2206 		dprintf("no properties\n");
2207 	assert(dict_get(map, pke) == NULL);
2208 }
2209 #endif	/* DEBUG */
2210 /*
2211  * pool_knl_elem_free() releases the resources associated with the
2212  * supplied element.
2213  */
2214 static void
2215 pool_knl_elem_free(pool_knl_elem_t *pke, int freeprop)
2216 {
2217 #if DEBUG
2218 	pool_conf_t *conf = TO_CONF(TO_ELEM(pke));
2219 	if (dict_remove(((pool_knl_connection_t *)conf->pc_prov)->pkc_leaks,
2220 	    pke) == NULL)
2221 		dprintf("%p, wasn't in the leak map\n", pke);
2222 	if (freeprop == PO_TRUE) {
2223 		pool_elem_dprintf(TO_ELEM(pke));
2224 	}
2225 	dprintf("released %p\n", pke);
2226 #endif	/* DEBUG */
2227 	if (freeprop == PO_TRUE) {
2228 		nvlist_free(pke->pke_properties);
2229 	}
2230 	free(pke);
2231 }
2232 
2233 /*
2234  * pool_knl_elem_free_cb() is designed to be used with
2235  * dict_map(). When a connection is freed, this function is used to
2236  * free all element resources.
2237  */
2238 /* ARGSUSED1 */
2239 static void
2240 pool_knl_elem_free_cb(const void *key, void **value, void *cl)
2241 {
2242 	pool_knl_elem_t *pke = (pool_knl_elem_t *)key;
2243 
2244 #ifdef DEBUG
2245 	dprintf("pool_knl_elem_free_cb:\n");
2246 	dprintf("about to release %p ", pke);
2247 	pool_elem_dprintf(TO_ELEM(pke));
2248 #endif	/* DEBUG */
2249 	pool_knl_elem_free(pke, PO_TRUE);
2250 }
2251 
2252 /*
2253  * Free the resources for a kernel data provider.
2254  */
2255 void
2256 pool_knl_connection_free(pool_knl_connection_t *prov)
2257 {
2258 	if (prov->pkc_log != NULL) {
2259 		(void) log_walk(prov->pkc_log, log_item_release);
2260 		log_free(prov->pkc_log);
2261 	}
2262 	if (prov->pkc_elements != NULL) {
2263 		dict_map(prov->pkc_elements, pool_knl_elem_free_cb, NULL);
2264 #if DEBUG
2265 		dprintf("dict length is %llu\n", dict_length(prov->pkc_leaks));
2266 		dict_map(prov->pkc_leaks, pool_knl_elem_printf_cb,
2267 		    prov->pkc_elements);
2268 		assert(dict_length(prov->pkc_leaks) == 0);
2269 		dict_free(&prov->pkc_leaks);
2270 #endif	/* DEBUG */
2271 		dict_free(&prov->pkc_elements);
2272 	}
2273 	free((void *)prov->pc_name);
2274 	free(prov);
2275 }
2276 
2277 /*
2278  * Return the specified property value.
2279  *
2280  * POC_INVAL is returned if an error is detected and the error code is updated
2281  * to indicate the cause of the error.
2282  */
2283 pool_value_class_t
2284 pool_knl_get_property(const pool_elem_t *pe, const char *name,
2285     pool_value_t *val)
2286 {
2287 	pool_knl_elem_t *pke = (pool_knl_elem_t *)pe;
2288 	nvpair_t *pair;
2289 	const pool_prop_t *prop;
2290 
2291 	if ((prop = provider_get_prop(pe, name)) != NULL)
2292 		if (prop_is_stored(prop) == PO_FALSE)
2293 			return (pool_knl_get_dynamic_property(pe, name, val));
2294 
2295 	if ((pair = pool_knl_find_nvpair(pke->pke_properties, name)) == NULL) {
2296 		pool_seterror(POE_BADPARAM);
2297 		return (POC_INVAL);
2298 	}
2299 
2300 	if (pool_value_from_nvpair(val, pair) == PO_FAIL) {
2301 		return (POC_INVAL);
2302 	}
2303 
2304 	return (pool_value_get_type(val));
2305 }
2306 
2307 /*
2308  * Return the specified property value.
2309  *
2310  * If a property is designated as dynamic, then this function will
2311  * always try to return the latest value of the property from the
2312  * kernel.
2313  *
2314  * POC_INVAL is returned if an error is detected and the error code is updated
2315  * to indicate the cause of the error.
2316  */
2317 pool_value_class_t
2318 pool_knl_get_dynamic_property(const pool_elem_t *pe, const char *name,
2319     pool_value_t *val)
2320 {
2321 	pool_knl_connection_t *prov;
2322 	pool_propget_t propget = { 0 };
2323 	nvlist_t *proplist;
2324 	nvpair_t *pair;
2325 
2326 	propget.pp_o_id_type = pool_elem_class(pe);
2327 	if (pool_elem_class(pe) == PEC_RES_COMP ||
2328 	    pool_elem_class(pe) == PEC_RES_AGG)
2329 		propget.pp_o_id_subtype = pool_resource_elem_class(pe);
2330 	if (pool_elem_class(pe) == PEC_COMP)
2331 		propget.pp_o_id_subtype =
2332 		    (pool_resource_elem_class_t)pool_component_elem_class(pe);
2333 
2334 	propget.pp_o_id = elem_get_sysid(pe);
2335 	propget.pp_o_prop_name_size = strlen(name);
2336 	propget.pp_o_prop_name = (char *)name;
2337 	propget.pp_i_bufsize = KERNEL_SNAPSHOT_BUF_SZ;
2338 	propget.pp_i_buf = malloc(KERNEL_SNAPSHOT_BUF_SZ);
2339 	bzero(propget.pp_i_buf, KERNEL_SNAPSHOT_BUF_SZ);
2340 
2341 	prov = (pool_knl_connection_t *)(TO_CONF(pe))->pc_prov;
2342 	if (ioctl(prov->pkc_fd, POOL_PROPGET, &propget) < 0) {
2343 		free(propget.pp_i_buf);
2344 		pool_seterror(POE_SYSTEM);
2345 		return (POC_INVAL);
2346 	}
2347 	if (nvlist_unpack(propget.pp_i_buf, propget.pp_i_bufsize,
2348 	    &proplist, 0) != 0) {
2349 		free(propget.pp_i_buf);
2350 		pool_seterror(POE_SYSTEM);
2351 		return (POC_INVAL);
2352 	}
2353 	free(propget.pp_i_buf);
2354 
2355 	if ((pair = nvlist_next_nvpair(proplist, NULL)) == NULL) {
2356 		nvlist_free(proplist);
2357 		pool_seterror(POE_SYSTEM);
2358 		return (POC_INVAL);
2359 	}
2360 
2361 	if (pool_value_from_nvpair(val, pair) == PO_FAIL) {
2362 		nvlist_free(proplist);
2363 		return (POC_INVAL);
2364 	}
2365 	nvlist_free(proplist);
2366 	return (pool_value_get_type(val));
2367 }
2368 
2369 /*
2370  * Update the specified property value.
2371  *
2372  * PO_FAIL is returned if an error is detected and the error code is updated
2373  * to indicate the cause of the error.
2374  */
2375 int
2376 pool_knl_put_property(pool_elem_t *pe, const char *name,
2377     const pool_value_t *val)
2378 {
2379 	pool_knl_elem_t *pke = (pool_knl_elem_t *)pe;
2380 	pool_knl_connection_t *prov =
2381 	    (pool_knl_connection_t *)(TO_CONF(pe))->pc_prov;
2382 	nvpair_t *bp, *ap;
2383 	pool_propput_undo_t *propput;
2384 	nvlist_t *bl = NULL;
2385 	const pool_prop_t *prop;
2386 
2387 	if ((bp = pool_knl_find_nvpair(pke->pke_properties, name)) != NULL) {
2388 		if (nvlist_alloc(&bl, NV_UNIQUE_NAME_TYPE, 0) != 0) {
2389 			pool_seterror(POE_SYSTEM);
2390 			return (PO_FAIL);
2391 		}
2392 		if (nvlist_add_nvpair(bl, bp) != 0) {
2393 			nvlist_free(bl);
2394 			pool_seterror(POE_SYSTEM);
2395 			return (PO_FAIL);
2396 		}
2397 	}
2398 	if (pool_knl_nvlist_add_value(pke->pke_properties, name, val) !=
2399 	    PO_SUCCESS)
2400 		return (PO_FAIL);
2401 
2402 	if (prov->pkc_log->l_state != LS_DO) {
2403 		if (bl)
2404 			nvlist_free(bl);
2405 		return (PO_SUCCESS);
2406 	}
2407 	/*
2408 	 * The remaining logic is setting up the arguments for the
2409 	 * POOL_PROPPUT ioctl and appending the details into the log.
2410 	 */
2411 	if ((propput = malloc(sizeof (pool_propput_undo_t))) == NULL) {
2412 		pool_seterror(POE_SYSTEM);
2413 		return (PO_FAIL);
2414 	}
2415 	(void) memset(propput, 0, sizeof (pool_propput_undo_t));
2416 	propput->ppu_blist = bl;
2417 
2418 	ap = pool_knl_find_nvpair(pke->pke_properties, name);
2419 
2420 	if (nvlist_alloc(&propput->ppu_alist, NV_UNIQUE_NAME_TYPE, 0) != 0) {
2421 		nvlist_free(propput->ppu_blist);
2422 		free(propput);
2423 		pool_seterror(POE_SYSTEM);
2424 		return (PO_FAIL);
2425 	}
2426 	if (nvlist_add_nvpair(propput->ppu_alist, ap) != 0) {
2427 		nvlist_free(propput->ppu_blist);
2428 		nvlist_free(propput->ppu_alist);
2429 		free(propput);
2430 		pool_seterror(POE_SYSTEM);
2431 		return (PO_FAIL);
2432 	}
2433 
2434 	if (nvlist_pack(propput->ppu_alist,
2435 	    (char **)&propput->ppu_ioctl.pp_o_buf,
2436 	    &propput->ppu_ioctl.pp_o_bufsize, NV_ENCODE_NATIVE, 0) != 0) {
2437 		pool_seterror(POE_SYSTEM);
2438 		return (PO_FAIL);
2439 	}
2440 	nvlist_free(propput->ppu_alist);
2441 	propput->ppu_ioctl.pp_o_id_type = pool_elem_class(pe);
2442 	if (pool_elem_class(pe) == PEC_RES_COMP ||
2443 	    pool_elem_class(pe) == PEC_RES_AGG)
2444 		propput->ppu_ioctl.pp_o_id_sub_type =
2445 		    pool_resource_elem_class(pe);
2446 	if (pool_elem_class(pe) == PEC_COMP)
2447 		propput->ppu_ioctl.pp_o_id_sub_type =
2448 		    (pool_resource_elem_class_t)pool_component_elem_class(pe);
2449 
2450 	propput->ppu_elem = pe;
2451 	if ((prop = provider_get_prop(propput->ppu_elem, name)) != NULL) {
2452 		if (prop_is_readonly(prop) == PO_TRUE)
2453 			propput->ppu_doioctl |= KERNEL_PROP_RDONLY;
2454 	}
2455 
2456 	if (log_append(prov->pkc_log, POOL_PROPPUT, (void *)propput) !=
2457 	    PO_SUCCESS) {
2458 		nvlist_free(propput->ppu_blist);
2459 		free(propput);
2460 		return (PO_FAIL);
2461 	}
2462 	return (PO_SUCCESS);
2463 }
2464 
2465 /*
2466  * Remove the specified property value.
2467  *
2468  * PO_FAIL is returned if an error is detected and the error code is
2469  * updated to indicate the cause of the error.
2470  */
2471 int
2472 pool_knl_rm_property(pool_elem_t *pe, const char *name)
2473 {
2474 	pool_knl_elem_t *pke = (pool_knl_elem_t *)pe;
2475 	pool_knl_connection_t *prov =
2476 	    (pool_knl_connection_t *)(TO_CONF(pe))->pc_prov;
2477 	pool_proprm_undo_t *proprm;
2478 
2479 	if (pool_knl_find_nvpair(pke->pke_properties, name) == NULL) {
2480 		pool_seterror(POE_BADPARAM);
2481 		return (PO_FAIL);
2482 	}
2483 
2484 	if ((proprm = malloc(sizeof (pool_proprm_undo_t))) == NULL) {
2485 		pool_seterror(POE_SYSTEM);
2486 		return (PO_FAIL);
2487 	}
2488 	(void) memset(proprm, 0, sizeof (pool_proprm_undo_t));
2489 	proprm->pru_oldval.pv_class = POC_INVAL;
2490 	(void) pool_get_property(TO_CONF(pe), pe, name, &proprm->pru_oldval);
2491 
2492 	if (prov->pkc_log->l_state != LS_DO) {
2493 		free(proprm);
2494 		(void) nvlist_remove_all(pke->pke_properties, (char *)name);
2495 		return (PO_SUCCESS);
2496 	}
2497 	/*
2498 	 * The remaining logic is setting up the arguments for the
2499 	 * POOL_PROPRM ioctl and appending the details into the log.
2500 	 */
2501 
2502 	proprm->pru_ioctl.pp_o_id_type = pool_elem_class(pe);
2503 	if (pool_elem_class(pe) == PEC_RES_COMP ||
2504 	    pool_elem_class(pe) == PEC_RES_AGG)
2505 		proprm->pru_ioctl.pp_o_id_sub_type =
2506 		    pool_resource_elem_class(pe);
2507 
2508 	if (pool_elem_class(pe) == PEC_COMP)
2509 		proprm->pru_ioctl.pp_o_id_sub_type =
2510 		    (pool_resource_elem_class_t)pool_component_elem_class(pe);
2511 
2512 	proprm->pru_ioctl.pp_o_prop_name_size = strlen(name);
2513 	proprm->pru_ioctl.pp_o_prop_name =
2514 	    (char *)pool_value_get_name(&proprm->pru_oldval);
2515 	proprm->pru_elem = pe;
2516 
2517 	if (log_append(prov->pkc_log, POOL_PROPRM, (void *)proprm) !=
2518 	    PO_SUCCESS) {
2519 		free(proprm);
2520 		return (PO_FAIL);
2521 	}
2522 
2523 	(void) nvlist_remove_all(pke->pke_properties, (char *)name);
2524 	return (PO_SUCCESS);
2525 }
2526 
2527 /*
2528  * Return a NULL terminated array of pool_value_t which represents all
2529  * of the properties stored for an element
2530  *
2531  * Return NULL on failure. It is the caller's responsibility to free
2532  * the returned array of values.
2533  */
2534 pool_value_t **
2535 pool_knl_get_properties(const pool_elem_t *pe, uint_t *nprops)
2536 {
2537 	nvpair_t *pair;
2538 	pool_value_t **result;
2539 	pool_knl_elem_t *pke = (pool_knl_elem_t *)pe;
2540 	int i = 0;
2541 
2542 	*nprops = 0;
2543 
2544 	for (pair = nvlist_next_nvpair(pke->pke_properties, NULL); pair != NULL;
2545 		pair = nvlist_next_nvpair(pke->pke_properties, pair))
2546 		(*nprops)++;
2547 	if ((result = calloc(*nprops + 1, sizeof (pool_value_t *))) == NULL) {
2548 		pool_seterror(POE_SYSTEM);
2549 		return (NULL);
2550 	}
2551 	for (pair = nvlist_next_nvpair(pke->pke_properties, NULL); pair != NULL;
2552 	    pair = nvlist_next_nvpair(pke->pke_properties, pair), i++) {
2553 		result[i] = pool_value_alloc();
2554 		if (pool_value_from_nvpair(result[i], pair) == PO_FAIL) {
2555 			while (i-- >= 0)
2556 				pool_value_free(result[i]);
2557 			free(result);
2558 			return (NULL);
2559 		}
2560 	}
2561 	return (result);
2562 }
2563 
2564 /*
2565  * Append an entry to a result set. Reallocate the array used to store
2566  * results if it's full.
2567  * Returns PO_SUCCESS/PO_FAIL
2568  */
2569 int
2570 pool_knl_result_set_append(pool_knl_result_set_t *rs, pool_knl_elem_t *pke)
2571 {
2572 	if (rs->pkr_count == rs->pkr_size)
2573 		if (pool_knl_result_set_realloc(rs) != PO_SUCCESS)
2574 			return (PO_FAIL);
2575 
2576 	rs->pkr_list[rs->pkr_count++] = pke;
2577 
2578 	return (PO_SUCCESS);
2579 }
2580 
2581 /*
2582  * Resize the array used to store results. A simple doubling strategy
2583  * is used.
2584  * Returns PO_SUCCESS/PO_FAIL
2585  */
2586 int
2587 pool_knl_result_set_realloc(pool_knl_result_set_t *rs)
2588 {
2589 	pool_knl_elem_t **old_list = rs->pkr_list;
2590 	int new_size = rs->pkr_size * 2;
2591 
2592 	if ((rs->pkr_list = realloc(rs->pkr_list,
2593 	    new_size * sizeof (pool_knl_elem_t *))) == NULL) {
2594 		rs->pkr_list = old_list;
2595 		pool_seterror(POE_SYSTEM);
2596 		return (PO_FAIL);
2597 	}
2598 	rs->pkr_size = new_size;
2599 
2600 	return (PO_SUCCESS);
2601 }
2602 
2603 /*
2604  * Allocate a result set. The Result Set stores the result of a query.
2605  * Returns pool_knl_result_set_t pointer/NULL
2606  */
2607 pool_knl_result_set_t *
2608 pool_knl_result_set_alloc(const pool_conf_t *conf)
2609 {
2610 	pool_knl_result_set_t *rs;
2611 
2612 	if ((rs = malloc(sizeof (pool_knl_result_set_t))) == NULL) {
2613 		pool_seterror(POE_SYSTEM);
2614 		return (NULL);
2615 	}
2616 	(void) memset(rs, 0, sizeof (pool_knl_result_set_t));
2617 	rs->pkr_size = KERNEL_RS_INITIAL_SZ;
2618 	if (pool_knl_result_set_realloc(rs) == PO_FAIL) {
2619 		free(rs);
2620 		pool_seterror(POE_SYSTEM);
2621 		return (NULL);
2622 	}
2623 	rs->prs_conf = conf;
2624 	rs->prs_index = -1;
2625 	rs->prs_active = PO_TRUE;
2626 	/* Fix up the result set accessor functions to the knl specfic ones */
2627 	rs->prs_next = pool_knl_rs_next;
2628 	rs->prs_prev = pool_knl_rs_prev;
2629 	rs->prs_first = pool_knl_rs_first;
2630 	rs->prs_last = pool_knl_rs_last;
2631 	rs->prs_get_index = pool_knl_rs_get_index;
2632 	rs->prs_set_index = pool_knl_rs_set_index;
2633 	rs->prs_close = pool_knl_rs_close;
2634 	rs->prs_count = pool_knl_rs_count;
2635 	return (rs);
2636 }
2637 
2638 /*
2639  * Free a result set. Ensure that the resources are all released at
2640  * this point.
2641  */
2642 void
2643 pool_knl_result_set_free(pool_knl_result_set_t *rs)
2644 {
2645 	free(rs->pkr_list);
2646 	free(rs);
2647 }
2648 /*
2649  * Return the next element in a result set.
2650  * Returns pool_elem_t pointer/NULL
2651  */
2652 pool_elem_t *
2653 pool_knl_rs_next(pool_result_set_t *set)
2654 {
2655 	pool_knl_result_set_t *kset = (pool_knl_result_set_t *)set;
2656 
2657 	if (kset->prs_index == kset->pkr_count - 1)
2658 		return (NULL);
2659 	return ((pool_elem_t *)kset->pkr_list[++kset->prs_index]);
2660 }
2661 
2662 /*
2663  * Return the previous element in a result set.
2664  * Returns pool_elem_t pointer/NULL
2665  */
2666 pool_elem_t *
2667 pool_knl_rs_prev(pool_result_set_t *set)
2668 {
2669 	pool_knl_result_set_t *kset = (pool_knl_result_set_t *)set;
2670 
2671 	if (kset->prs_index < 0)
2672 		return (NULL);
2673 	return ((pool_elem_t *)kset->pkr_list[kset->prs_index--]);
2674 }
2675 
2676 /*
2677  * Sets the current index in a result set.
2678  * Returns PO_SUCCESS/PO_FAIL
2679  */
2680 int
2681 pool_knl_rs_set_index(pool_result_set_t *set, int index)
2682 {
2683 	pool_knl_result_set_t *kset = (pool_knl_result_set_t *)set;
2684 
2685 	if (index < 0 || index >= kset->pkr_count) {
2686 		pool_seterror(POE_BADPARAM);
2687 		return (PO_FAIL);
2688 	}
2689 	kset->prs_index = index;
2690 	return (PO_SUCCESS);
2691 }
2692 
2693 /*
2694  * Return the current index in a result set.
2695  * Returns current index
2696  */
2697 int
2698 pool_knl_rs_get_index(pool_result_set_t *set)
2699 {
2700 	pool_knl_result_set_t *kset = (pool_knl_result_set_t *)set;
2701 
2702 	return (kset->prs_index);
2703 }
2704 
2705 /*
2706  * Return the first element in a result set.
2707  * Returns pool_elem_t pointer/NULL
2708  */
2709 pool_elem_t *
2710 pool_knl_rs_first(pool_result_set_t *set)
2711 {
2712 	pool_knl_result_set_t *kset = (pool_knl_result_set_t *)set;
2713 
2714 	return ((pool_elem_t *)kset->pkr_list[0]);
2715 }
2716 
2717 /*
2718  * Return the last element in a result set.
2719  * Returns pool_elem_t pointer/NULL
2720  */
2721 pool_elem_t *
2722 pool_knl_rs_last(pool_result_set_t *set)
2723 {
2724 	pool_knl_result_set_t *kset = (pool_knl_result_set_t *)set;
2725 
2726 	return ((pool_elem_t *)kset->pkr_list[kset->pkr_count - 1]);
2727 }
2728 
2729 /*
2730  * Return the number of results in a result set.
2731  * Returns result count
2732  */
2733 int
2734 pool_knl_rs_count(pool_result_set_t *set)
2735 {
2736 	pool_knl_result_set_t *kset = (pool_knl_result_set_t *)set;
2737 
2738 	return (kset->pkr_count);
2739 }
2740 
2741 
2742 /*
2743  * Close a result set. Free the resources
2744  * Returns PO_SUCCESS/PO_FAIL
2745  */
2746 int
2747 pool_knl_rs_close(pool_result_set_t *set)
2748 {
2749 	pool_knl_result_set_t *kset = (pool_knl_result_set_t *)set;
2750 
2751 	pool_knl_result_set_free(kset);
2752 	return (PO_SUCCESS);
2753 }
2754 
2755 /*
2756  * Commit an individual transaction log item(). This processing is
2757  * essential to the pool_conf_commit() logic. When pool_conf_commit()
2758  * is invoked, the pending transaction log for the configuration is
2759  * walked and all pending changes to the kernel are invoked. If a
2760  * change succeeds it is marked in the log as successful and
2761  * processing continues, if it fails then failure is returned and the
2762  * log will be "rolled back" to undo changes to the library snapshot
2763  * and the kernel.
2764  */
2765 int
2766 log_item_commit(log_item_t *li)
2767 {
2768 	pool_knl_connection_t *prov =
2769 	    (pool_knl_connection_t *)li->li_log->l_conf->pc_prov;
2770 	pool_create_undo_t *create;
2771 	pool_destroy_undo_t *destroy;
2772 	pool_assoc_undo_t *assoc;
2773 	pool_dissoc_undo_t *dissoc;
2774 	pool_propput_undo_t *propput;
2775 	pool_proprm_undo_t *proprm;
2776 	pool_xtransfer_undo_t *xtransfer;
2777 	char_buf_t *cb;
2778 	size_t size;
2779 	pool_elem_t *pair;
2780 	pool_value_t val = POOL_VALUE_INITIALIZER;
2781 	int ret;
2782 
2783 	switch (li->li_op) {
2784 	case POOL_CREATE:
2785 		create = (pool_create_undo_t *)li->li_details;
2786 		if ((cb = alloc_char_buf(CB_DEFAULT_LEN)) == NULL)
2787 			return (PO_FAIL);
2788 		if (set_char_buf(cb, "%s.sys_id",
2789 		    pool_elem_class_string(create->pcu_elem)) != PO_SUCCESS) {
2790 			free_char_buf(cb);
2791 			return (PO_FAIL);
2792 		}
2793 #ifdef DEBUG
2794 		dprintf("log_item_commit: POOL_CREATE, remove from dict\n");
2795 		pool_elem_dprintf(create->pcu_elem);
2796 #endif	/* DEBUG */
2797 		/*
2798 		 * May not need to remove the element if it was
2799 		 * already destroyed before commit. Just cast the
2800 		 * return to void.
2801 		 */
2802 		(void) dict_remove(prov->pkc_elements,
2803 		    (pool_knl_elem_t *)create->pcu_elem);
2804 
2805 		if (ioctl(prov->pkc_fd, POOL_CREATE, &create->pcu_ioctl) < 0) {
2806 			pool_seterror(POE_SYSTEM);
2807 			return (PO_FAIL);
2808 		}
2809 		/*
2810 		 * Now that we have created our element in the kernel,
2811 		 * it has a valid allocated system id. Remove the
2812 		 * element from the element dictionary, using the
2813 		 * current key, and then re-insert under the new key.
2814 		 */
2815 #ifdef DEBUG
2816 		pool_elem_dprintf(create->pcu_elem);
2817 #endif	/* DEBUG */
2818 		assert(nvlist_add_int64(
2819 		    ((pool_knl_elem_t *)create->pcu_elem)->pke_properties,
2820 		    cb->cb_buf, create->pcu_ioctl.pc_i_id) == 0);
2821 		free_char_buf(cb);
2822 		assert(dict_put(prov->pkc_elements, create->pcu_elem,
2823 		    create->pcu_elem) == NULL);
2824 		/*
2825 		 * If the element has a pair in the static
2826 		 * configuration, update it with the sys_id
2827 		 */
2828 		if ((pair = pool_get_pair(create->pcu_elem)) != NULL) {
2829 			pool_value_set_int64(&val, create->pcu_ioctl.pc_i_id);
2830 			assert(pool_put_any_ns_property(pair, c_sys_prop, &val)
2831 			    == PO_SUCCESS);
2832 		}
2833 		li->li_state = LS_UNDO;
2834 		break;
2835 	case POOL_DESTROY:
2836 		destroy = (pool_destroy_undo_t *)li->li_details;
2837 
2838 		destroy->pdu_ioctl.pd_o_id = elem_get_sysid(destroy->pdu_elem);
2839 
2840 		/*
2841 		 * It may be that this element was created in the last
2842 		 * transaction. In which case POOL_CREATE, above, will
2843 		 * have re-inserted the element in the dictionary. Try
2844 		 * to remove it just in case this has occurred.
2845 		 */
2846 		(void) dict_remove(prov->pkc_elements,
2847 		    (pool_knl_elem_t *)destroy->pdu_elem);
2848 		while ((ret = ioctl(prov->pkc_fd, POOL_DESTROY,
2849 		    &destroy->pdu_ioctl)) < 0 && errno == EAGAIN);
2850 		if (ret < 0) {
2851 			pool_seterror(POE_SYSTEM);
2852 			return (PO_FAIL);
2853 		}
2854 #ifdef DEBUG
2855 		dprintf("log_item_commit: POOL_DESTROY\n");
2856 		pool_elem_dprintf(destroy->pdu_elem);
2857 #endif	/* DEBUG */
2858 		li->li_state = LS_UNDO;
2859 		break;
2860 	case POOL_ASSOC:
2861 		assoc = (pool_assoc_undo_t *)li->li_details;
2862 
2863 		assoc->pau_ioctl.pa_o_pool_id =
2864 		    elem_get_sysid(assoc->pau_assoc);
2865 		assoc->pau_ioctl.pa_o_res_id =
2866 		    elem_get_sysid(assoc->pau_newres);
2867 		while ((ret = ioctl(prov->pkc_fd, POOL_ASSOC,
2868 		    &assoc->pau_ioctl)) < 0 && errno == EAGAIN);
2869 		if (ret < 0) {
2870 			pool_seterror(POE_SYSTEM);
2871 			return (PO_FAIL);
2872 		}
2873 		li->li_state = LS_UNDO;
2874 		break;
2875 	case POOL_DISSOC:
2876 		dissoc = (pool_dissoc_undo_t *)li->li_details;
2877 
2878 		dissoc->pdu_ioctl.pd_o_pool_id =
2879 		    elem_get_sysid(dissoc->pdu_dissoc);
2880 
2881 		while ((ret = ioctl(prov->pkc_fd, POOL_DISSOC,
2882 		    &dissoc->pdu_ioctl)) < 0 && errno == EAGAIN);
2883 		if (ret < 0) {
2884 			pool_seterror(POE_SYSTEM);
2885 			return (PO_FAIL);
2886 		}
2887 		li->li_state = LS_UNDO;
2888 		break;
2889 	case POOL_TRANSFER:
2890 		li->li_state = LS_UNDO;
2891 		pool_seterror(POE_BADPARAM);
2892 		return (PO_FAIL);
2893 		break;
2894 	case POOL_XTRANSFER:
2895 		xtransfer = (pool_xtransfer_undo_t *)li->li_details;
2896 
2897 		xtransfer->pxu_ioctl.px_o_src_id =
2898 		    elem_get_sysid(xtransfer->pxu_src);
2899 		xtransfer->pxu_ioctl.px_o_tgt_id =
2900 		    elem_get_sysid(xtransfer->pxu_tgt);
2901 		for (size = 0; xtransfer->pxu_rl[size] != NULL; size ++) {
2902 			xtransfer->pxu_ioctl.px_o_comp_list[size] =
2903 			    elem_get_sysid(TO_ELEM(xtransfer->pxu_rl[size]));
2904 #ifdef DEBUG
2905 			dprintf("log_item_commit: POOL_XTRANSFER\n");
2906 			pool_elem_dprintf(TO_ELEM(xtransfer->pxu_rl[size]));
2907 #endif	/* DEBUG */
2908 		}
2909 
2910 		/*
2911 		 * Don't actually transfer resources if the configuration
2912 		 * is in POF_DESTROY state. This is to prevent problems
2913 		 * relating to transferring off-line CPUs. Instead rely
2914 		 * on the POOL_DESTROY ioctl to transfer the CPUS.
2915 		 */
2916 		if (li->li_log->l_conf->pc_state != POF_DESTROY &&
2917 		    ioctl(prov->pkc_fd, POOL_XTRANSFER,
2918 		    &xtransfer->pxu_ioctl) < 0) {
2919 #ifdef DEBUG
2920 			dprintf("log_item_commit: POOL_XTRANSFER, ioctl "
2921 			    "failed\n");
2922 #endif	/* DEBUG */
2923 			pool_seterror(POE_SYSTEM);
2924 			return (PO_FAIL);
2925 		}
2926 		li->li_state = LS_UNDO;
2927 		break;
2928 	case POOL_PROPPUT:
2929 		propput = (pool_propput_undo_t *)li->li_details;
2930 
2931 		if (pool_elem_class(propput->ppu_elem) != PEC_SYSTEM) {
2932 			propput->ppu_ioctl.pp_o_id =
2933 			    elem_get_sysid(propput->ppu_elem);
2934 		}
2935 		/*
2936 		 * Some properties, e.g. pset.size, are read-only in the
2937 		 * kernel and attempting to change them will fail and cause
2938 		 * problems. Although this property is read-only through the
2939 		 * public interface, the library needs to modify it's value.
2940 		 */
2941 		if ((propput->ppu_doioctl & KERNEL_PROP_RDONLY) == 0) {
2942 			if (ioctl(prov->pkc_fd, POOL_PROPPUT,
2943 			    &propput->ppu_ioctl) < 0) {
2944 				pool_seterror(POE_SYSTEM);
2945 				return (PO_FAIL);
2946 			}
2947 		}
2948 		li->li_state = LS_UNDO;
2949 		break;
2950 	case POOL_PROPRM:
2951 		proprm = (pool_proprm_undo_t *)li->li_details;
2952 
2953 		if (pool_elem_class(proprm->pru_elem) != PEC_SYSTEM) {
2954 			proprm->pru_ioctl.pp_o_id =
2955 			    elem_get_sysid(proprm->pru_elem);
2956 		}
2957 		if (ioctl(prov->pkc_fd, POOL_PROPRM, &proprm->pru_ioctl) < 0) {
2958 			pool_seterror(POE_SYSTEM);
2959 			return (PO_FAIL);
2960 		}
2961 		li->li_state = LS_UNDO;
2962 		break;
2963 	default:
2964 		return (PO_FAIL);
2965 	}
2966 	return (PO_SUCCESS);
2967 }
2968 
2969 /*
2970  * Undo an individual transaction log item(). This processing is
2971  * essential to the pool_conf_commit() and pool_conf_rollback()
2972  * logic. Changes to the libpool snapshot and the kernel are carried
2973  * out separately. The library snapshot is updated synchronously,
2974  * however the kernel update is delayed until the user calls
2975  * pool_conf_commit().
2976  *
2977  * When undoing transactions, library changes will be undone unless
2978  * this invocation is as a result of a commit failure, in which case
2979  * the log state will be LS_RECOVER. Kernel changes will only be
2980  * undone if they are marked as having been done, in which case the
2981  * log item state will be LS_UNDO.
2982  */
2983 int
2984 log_item_undo(log_item_t *li)
2985 {
2986 	pool_knl_connection_t *prov =
2987 	    (pool_knl_connection_t *)li->li_log->l_conf->pc_prov;
2988 	pool_create_undo_t *create;
2989 	pool_destroy_undo_t *destroy;
2990 	pool_assoc_undo_t *assoc;
2991 	pool_dissoc_undo_t *dissoc;
2992 	pool_propput_undo_t *propput;
2993 	pool_proprm_undo_t *proprm;
2994 	pool_xtransfer_undo_t *xtransfer;
2995 	char_buf_t *cb;
2996 	size_t size;
2997 	pool_destroy_t u_destroy;
2998 	pool_create_t u_create;
2999 	pool_assoc_t u_assoc;
3000 	pool_xtransfer_t u_xtransfer;
3001 	pool_propput_t u_propput;
3002 	pool_proprm_t u_proprm;
3003 	pool_conf_t *conf = li->li_log->l_conf;
3004 	nvpair_t *pair;
3005 	nvlist_t *tmplist;
3006 	int ret;
3007 
3008 	if (li->li_log->l_state != LS_RECOVER) {
3009 	switch (li->li_op) {
3010 	case POOL_CREATE:
3011 		create = (pool_create_undo_t *)li->li_details;
3012 
3013 		(void) dict_remove(prov->pkc_elements, create->pcu_elem);
3014 #ifdef DEBUG
3015 		dprintf("log_item_undo: POOL_CREATE\n");
3016 		assert(create->pcu_elem != NULL);
3017 		dprintf("log_item_undo: POOL_CREATE %p\n", create->pcu_elem);
3018 		pool_elem_dprintf(create->pcu_elem);
3019 #endif	/* DEBUG */
3020 		pool_knl_elem_free((pool_knl_elem_t *)create->pcu_elem,
3021 		    PO_TRUE);
3022 		break;
3023 	case POOL_DESTROY:
3024 		destroy = (pool_destroy_undo_t *)li->li_details;
3025 
3026 		assert(dict_put(prov->pkc_elements, destroy->pdu_elem,
3027 		    destroy->pdu_elem) == NULL);
3028 		break;
3029 	case POOL_ASSOC:
3030 		assoc = (pool_assoc_undo_t *)li->li_details;
3031 
3032 		if (assoc->pau_oldres != NULL)
3033 			((pool_knl_pool_t *)assoc->pau_assoc)->pkp_assoc
3034 			    [pool_resource_elem_class(assoc->pau_oldres)] =
3035 			    (pool_knl_resource_t *)assoc->pau_oldres;
3036 		break;
3037 	case POOL_DISSOC:
3038 		dissoc = (pool_dissoc_undo_t *)li->li_details;
3039 
3040 		if (dissoc->pdu_oldres != NULL)
3041 			((pool_knl_pool_t *)dissoc->pdu_dissoc)->pkp_assoc
3042 			    [pool_resource_elem_class(dissoc->pdu_oldres)] =
3043 			    (pool_knl_resource_t *)dissoc->pdu_oldres;
3044 		break;
3045 	case POOL_TRANSFER:
3046 		pool_seterror(POE_BADPARAM);
3047 		return (PO_FAIL);
3048 		break;
3049 	case POOL_XTRANSFER:
3050 		xtransfer = (pool_xtransfer_undo_t *)li->li_details;
3051 
3052 		for (size = 0; xtransfer->pxu_rl[size] != NULL; size++) {
3053 			pool_value_t val = POOL_VALUE_INITIALIZER;
3054 			uint64_t src_size;
3055 			uint64_t tgt_size;
3056 
3057 			if (pool_set_container(xtransfer->pxu_src,
3058 			    TO_ELEM(xtransfer->pxu_rl[size])) == PO_FAIL) {
3059 				return (PO_FAIL);
3060 			}
3061 			/*
3062 			 * Maintain the library view of the size
3063 			 */
3064 			if (resource_get_size(pool_elem_res(xtransfer->pxu_src),
3065 			    &src_size) != PO_SUCCESS ||
3066 			    resource_get_size(pool_elem_res(xtransfer->pxu_tgt),
3067 			    &tgt_size) != PO_SUCCESS) {
3068 				pool_seterror(POE_BADPARAM);
3069 				return (PO_FAIL);
3070 			}
3071 			src_size++;
3072 			tgt_size--;
3073 			pool_value_set_uint64(&val, src_size);
3074 			(void) pool_put_any_ns_property(xtransfer->pxu_src,
3075 			    c_size_prop, &val);
3076 			pool_value_set_uint64(&val, tgt_size);
3077 			(void) pool_put_any_ns_property(xtransfer->pxu_tgt,
3078 			    c_size_prop, &val);
3079 		}
3080 		break;
3081 	case POOL_PROPPUT:
3082 		propput = (pool_propput_undo_t *)li->li_details;
3083 
3084 		if ((propput->ppu_doioctl & KERNEL_PROP_RDONLY) == 0) {
3085 			if (propput->ppu_blist != NULL) {
3086 				if (nvlist_merge(
3087 				    ((pool_knl_elem_t *)propput->ppu_elem)->
3088 				    pke_properties, propput->ppu_blist, 0)
3089 				    != 0) {
3090 					pool_seterror(POE_SYSTEM);
3091 					return (PO_FAIL);
3092 				}
3093 			} else {
3094 				if (nvlist_unpack(propput->ppu_ioctl.pp_o_buf,
3095 				    propput->ppu_ioctl.pp_o_bufsize,
3096 				    &propput->ppu_alist, 0) != 0) {
3097 					pool_seterror(POE_SYSTEM);
3098 					return (PO_FAIL);
3099 				}
3100 				pair = nvlist_next_nvpair(propput->ppu_alist,
3101 				    NULL);
3102 				(void) nvlist_remove_all(((pool_knl_elem_t *)
3103 				    propput->ppu_elem)->pke_properties,
3104 				    nvpair_name(pair));
3105 				nvlist_free(propput->ppu_alist);
3106 			}
3107 		}
3108 		break;
3109 	case POOL_PROPRM:
3110 		proprm = (pool_proprm_undo_t *)li->li_details;
3111 
3112 		if (pool_value_get_type(&proprm->pru_oldval) != POC_INVAL) {
3113 			if (pool_put_property(conf, proprm->pru_elem,
3114 			    proprm->pru_ioctl.pp_o_prop_name,
3115 			    &proprm->pru_oldval) != PO_SUCCESS) {
3116 				return (PO_FAIL);
3117 			}
3118 		}
3119 		break;
3120 	default:
3121 		return (PO_FAIL);
3122 	}
3123 	}
3124 	/*
3125 	 * Only try to undo the state of the kernel if we modified it.
3126 	 */
3127 	if (li->li_state == LS_DO) {
3128 		return (PO_SUCCESS);
3129 	}
3130 
3131 	switch (li->li_op) {
3132 	case POOL_CREATE:
3133 		create = (pool_create_undo_t *)li->li_details;
3134 
3135 		u_destroy.pd_o_type = create->pcu_ioctl.pc_o_type;
3136 		u_destroy.pd_o_sub_type = create->pcu_ioctl.pc_o_sub_type;
3137 		u_destroy.pd_o_id = create->pcu_ioctl.pc_i_id;
3138 
3139 		while ((ret = ioctl(prov->pkc_fd, POOL_DESTROY,
3140 		    &u_destroy)) < 0 && errno == EAGAIN);
3141 		if (ret < 0) {
3142 			pool_seterror(POE_SYSTEM);
3143 			return (PO_FAIL);
3144 		}
3145 		li->li_state = LS_DO;
3146 		break;
3147 	case POOL_DESTROY:
3148 		destroy = (pool_destroy_undo_t *)li->li_details;
3149 
3150 		u_create.pc_o_type = destroy->pdu_ioctl.pd_o_type;
3151 		u_create.pc_o_sub_type = destroy->pdu_ioctl.pd_o_sub_type;
3152 
3153 		if (ioctl(prov->pkc_fd, POOL_CREATE, &u_create) < 0) {
3154 			pool_seterror(POE_SYSTEM);
3155 			return (PO_FAIL);
3156 		}
3157 
3158 		if ((cb = alloc_char_buf(CB_DEFAULT_LEN)) == NULL) {
3159 			return (PO_FAIL);
3160 		}
3161 		if (set_char_buf(cb, "%s.sys_id",
3162 		    pool_elem_class_string(destroy->pdu_elem)) != PO_SUCCESS) {
3163 			free_char_buf(cb);
3164 			return (PO_FAIL);
3165 		}
3166 		(void) nvlist_add_int64(
3167 		    ((pool_knl_elem_t *)destroy->pdu_elem)->pke_properties,
3168 		    cb->cb_buf, u_create.pc_i_id);
3169 		free_char_buf(cb);
3170 		if (dict_put(prov->pkc_elements, destroy->pdu_elem,
3171 		    destroy->pdu_elem) != NULL) {
3172 			pool_seterror(POE_SYSTEM);
3173 			return (PO_FAIL);
3174 		}
3175 		/*
3176 		 * Now we need to reset all the properties and
3177 		 * associations in the kernel for this newly created
3178 		 * replacement.
3179 		 */
3180 		u_propput.pp_o_id_type = destroy->pdu_ioctl.pd_o_type;
3181 		u_propput.pp_o_id_sub_type = destroy->pdu_ioctl.pd_o_sub_type;
3182 		u_propput.pp_o_id = u_create.pc_i_id;
3183 		u_propput.pp_o_buf = NULL;
3184 		/*
3185 		 * Remove the read-only properties before attempting
3186 		 * to restore the state of the newly created property
3187 		 */
3188 		(void) nvlist_dup(((pool_knl_elem_t *)destroy->pdu_elem)->
3189 		pke_properties, &tmplist, 0);
3190 		for (pair = nvlist_next_nvpair(tmplist, NULL); pair != NULL;
3191 		    pair = nvlist_next_nvpair(tmplist, pair)) {
3192 			const pool_prop_t *prop;
3193 			char *name = nvpair_name(pair);
3194 			if ((prop = provider_get_prop(destroy->pdu_elem,
3195 			    name)) != NULL)
3196 				if (prop_is_readonly(prop) == PO_TRUE)
3197 					(void) nvlist_remove_all(tmplist, name);
3198 		}
3199 		if (nvlist_pack(tmplist, (char **)&u_propput.pp_o_buf,
3200 		    &u_propput.pp_o_bufsize, NV_ENCODE_NATIVE, 0) != 0) {
3201 			pool_seterror(POE_SYSTEM);
3202 			return (PO_FAIL);
3203 		}
3204 		nvlist_free(tmplist);
3205 		if (ioctl(prov->pkc_fd, POOL_PROPPUT, &u_propput) < 0) {
3206 			free(u_propput.pp_o_buf);
3207 			pool_seterror(POE_SYSTEM);
3208 			return (PO_FAIL);
3209 		}
3210 		free(u_propput.pp_o_buf);
3211 		/*
3212 		 * Now reset the associations for all the resource
3213 		 * types if the thing which we are recreating is a
3214 		 * pool
3215 		 *
3216 		 * TODO: This is resource specific and must be
3217 		 * extended for additional resource types.
3218 		 */
3219 		if (destroy->pdu_ioctl.pd_o_type == PEC_POOL) {
3220 			u_assoc.pa_o_pool_id = u_create.pc_i_id;
3221 			u_assoc.pa_o_res_id =
3222 			    elem_get_sysid(
3223 			    TO_ELEM(((pool_knl_pool_t *)destroy->pdu_elem)->
3224 			    pkp_assoc[PREC_PSET]));
3225 			u_assoc.pa_o_id_type = PREC_PSET;
3226 
3227 			if (ioctl(prov->pkc_fd, POOL_ASSOC, &u_assoc) < 0) {
3228 				pool_seterror(POE_SYSTEM);
3229 				return (PO_FAIL);
3230 			}
3231 		}
3232 		li->li_state = LS_DO;
3233 		break;
3234 	case POOL_ASSOC:
3235 		assoc = (pool_assoc_undo_t *)li->li_details;
3236 
3237 		u_assoc.pa_o_pool_id = elem_get_sysid(assoc->pau_assoc);
3238 		u_assoc.pa_o_res_id = elem_get_sysid(assoc->pau_oldres);
3239 		u_assoc.pa_o_id_type = assoc->pau_ioctl.pa_o_id_type;
3240 
3241 		while ((ret = ioctl(prov->pkc_fd, POOL_ASSOC, &u_assoc)) < 0 &&
3242 		    errno == EAGAIN);
3243 		if (ret < 0) {
3244 			pool_seterror(POE_SYSTEM);
3245 			return (PO_FAIL);
3246 		}
3247 		li->li_state = LS_DO;
3248 		break;
3249 	case POOL_DISSOC:
3250 		dissoc = (pool_dissoc_undo_t *)li->li_details;
3251 
3252 		u_assoc.pa_o_pool_id = elem_get_sysid(dissoc->pdu_dissoc);
3253 		u_assoc.pa_o_res_id = elem_get_sysid(dissoc->pdu_oldres);
3254 		u_assoc.pa_o_id_type = dissoc->pdu_ioctl.pd_o_id_type;
3255 
3256 		while ((ret = ioctl(prov->pkc_fd, POOL_ASSOC, &u_assoc)) < 0 &&
3257 		    errno == EAGAIN);
3258 		if (ret < 0) {
3259 			pool_seterror(POE_SYSTEM);
3260 			return (PO_FAIL);
3261 		}
3262 		li->li_state = LS_DO;
3263 		break;
3264 	case POOL_TRANSFER:
3265 		li->li_state = LS_DO;
3266 		pool_seterror(POE_BADPARAM);
3267 		return (PO_FAIL);
3268 		break;
3269 	case POOL_XTRANSFER:
3270 		xtransfer = (pool_xtransfer_undo_t *)li->li_details;
3271 
3272 		(void) memcpy(&u_xtransfer, &xtransfer->pxu_ioctl,
3273 		    sizeof (pool_xtransfer_t));
3274 		u_xtransfer.px_o_src_id = elem_get_sysid(xtransfer->pxu_tgt);
3275 		u_xtransfer.px_o_tgt_id = elem_get_sysid(xtransfer->pxu_src);
3276 
3277 		if (ioctl(prov->pkc_fd, POOL_XTRANSFER, &u_xtransfer) < 0) {
3278 			pool_seterror(POE_SYSTEM);
3279 			return (PO_FAIL);
3280 		}
3281 		li->li_state = LS_DO;
3282 		break;
3283 	case POOL_PROPPUT:
3284 		propput = (pool_propput_undo_t *)li->li_details;
3285 
3286 		if ((propput->ppu_doioctl & KERNEL_PROP_RDONLY) == 0) {
3287 			if (propput->ppu_blist) {
3288 				(void) memcpy(&u_propput, &propput->ppu_ioctl,
3289 				    sizeof (pool_propput_t));
3290 				u_propput.pp_o_id =
3291 				    elem_get_sysid(propput->ppu_elem);
3292 				u_propput.pp_o_buf = NULL;
3293 				if (nvlist_pack(propput->ppu_blist,
3294 				    (char **)&u_propput.pp_o_buf,
3295 				    &u_propput.pp_o_bufsize,
3296 				    NV_ENCODE_NATIVE, 0) != 0) {
3297 					pool_seterror(POE_SYSTEM);
3298 					return (PO_FAIL);
3299 				}
3300 				if (ioctl(prov->pkc_fd, POOL_PROPPUT,
3301 				    &u_propput) < 0) {
3302 					free(u_propput.pp_o_buf);
3303 					pool_seterror(POE_SYSTEM);
3304 					return (PO_FAIL);
3305 				}
3306 				free(u_propput.pp_o_buf);
3307 			} else {
3308 				if (nvlist_unpack(propput->
3309 				    ppu_ioctl.pp_o_buf,
3310 				    propput->ppu_ioctl.pp_o_bufsize,
3311 				    &propput->ppu_alist, 0) != 0) {
3312 					pool_seterror(POE_SYSTEM);
3313 					return (PO_FAIL);
3314 				}
3315 				u_proprm.pp_o_id_type =
3316 				    propput->ppu_ioctl.pp_o_id_type;
3317 				u_proprm.pp_o_id_sub_type =
3318 				    propput->ppu_ioctl.pp_o_id_sub_type;
3319 				u_proprm.pp_o_id =
3320 				    elem_get_sysid(propput->ppu_elem);
3321 				pair = nvlist_next_nvpair(propput->ppu_alist,
3322 				    NULL);
3323 				u_proprm.pp_o_prop_name = nvpair_name(pair);
3324 				u_proprm.pp_o_prop_name_size =
3325 				    strlen(u_proprm.pp_o_prop_name);
3326 
3327 				if (provider_get_prop(propput->ppu_elem,
3328 				    u_proprm.pp_o_prop_name) == NULL) {
3329 					if (ioctl(prov->pkc_fd, POOL_PROPRM,
3330 					    &u_proprm) < 0) {
3331 						nvlist_free(propput->ppu_alist);
3332 						pool_seterror(POE_SYSTEM);
3333 						return (PO_FAIL);
3334 					}
3335 				}
3336 				nvlist_free(propput->ppu_alist);
3337 			}
3338 		}
3339 		li->li_state = LS_DO;
3340 		break;
3341 	case POOL_PROPRM:
3342 		proprm = (pool_proprm_undo_t *)li->li_details;
3343 
3344 		u_propput.pp_o_id_type = proprm->pru_ioctl.pp_o_id_type;
3345 		u_propput.pp_o_id_sub_type =
3346 		    proprm->pru_ioctl.pp_o_id_sub_type;
3347 		u_propput.pp_o_id = elem_get_sysid(proprm->pru_elem);
3348 		u_propput.pp_o_buf = NULL;
3349 		/*
3350 		 * Only try to remove the appropriate property
3351 		 */
3352 		if (nvlist_alloc(&tmplist, NV_UNIQUE_NAME_TYPE, 0) !=
3353 		    0) {
3354 			pool_seterror(POE_SYSTEM);
3355 			return (PO_FAIL);
3356 		}
3357 		if (pool_knl_nvlist_add_value(tmplist,
3358 		    pool_value_get_name(&proprm->pru_oldval),
3359 		    &proprm->pru_oldval) != PO_SUCCESS)
3360 			return (PO_FAIL);
3361 
3362 		if (nvlist_pack(tmplist,
3363 		    (char **)&u_propput.pp_o_buf, &u_propput.pp_o_bufsize,
3364 		    NV_ENCODE_NATIVE, 0) != 0) {
3365 			nvlist_free(tmplist);
3366 			pool_seterror(POE_SYSTEM);
3367 			return (PO_FAIL);
3368 		}
3369 		nvlist_free(tmplist);
3370 		if (ioctl(prov->pkc_fd, POOL_PROPPUT, &u_propput) < 0) {
3371 			free(u_propput.pp_o_buf);
3372 			pool_seterror(POE_SYSTEM);
3373 			return (PO_FAIL);
3374 		}
3375 		free(u_propput.pp_o_buf);
3376 		li->li_state = LS_DO;
3377 		break;
3378 	default:
3379 		return (PO_FAIL);
3380 	}
3381 		return (PO_SUCCESS);
3382 }
3383 
3384 /*
3385  * A log item stores state about the transaction it represents. This
3386  * function releases the resources associated with the transaction and
3387  * used to store the transaction state.
3388  */
3389 int
3390 log_item_release(log_item_t *li)
3391 {
3392 	pool_create_undo_t *create;
3393 	pool_destroy_undo_t *destroy;
3394 	pool_assoc_undo_t *assoc;
3395 	pool_dissoc_undo_t *dissoc;
3396 	pool_propput_undo_t *propput;
3397 	pool_proprm_undo_t *proprm;
3398 	pool_xtransfer_undo_t *xtransfer;
3399 
3400 	switch (li->li_op) {
3401 	case POOL_CREATE:
3402 		create = (pool_create_undo_t *)li->li_details;
3403 
3404 		free(create);
3405 		break;
3406 	case POOL_DESTROY:
3407 		destroy = (pool_destroy_undo_t *)li->li_details;
3408 
3409 #ifdef DEBUG
3410 		dprintf("log_item_release: POOL_DESTROY\n");
3411 #endif	/* DEBUG */
3412 
3413 		if (li->li_state == LS_UNDO) {
3414 #ifdef DEBUG
3415 			pool_elem_dprintf(destroy->pdu_elem);
3416 #endif	/* DEBUG */
3417 			pool_knl_elem_free((pool_knl_elem_t *)destroy->
3418 			    pdu_elem, PO_TRUE);
3419 		}
3420 		free(destroy);
3421 		break;
3422 	case POOL_ASSOC:
3423 		assoc = (pool_assoc_undo_t *)li->li_details;
3424 
3425 		free(assoc);
3426 		break;
3427 	case POOL_DISSOC:
3428 		dissoc = (pool_dissoc_undo_t *)li->li_details;
3429 
3430 		free(dissoc);
3431 		break;
3432 	case POOL_TRANSFER:
3433 		pool_seterror(POE_BADPARAM);
3434 		return (PO_FAIL);
3435 		break;
3436 	case POOL_XTRANSFER:
3437 		xtransfer = (pool_xtransfer_undo_t *)li->li_details;
3438 
3439 		free(xtransfer->pxu_rl);
3440 		free(xtransfer->pxu_ioctl.px_o_comp_list);
3441 		free(xtransfer);
3442 		break;
3443 	case POOL_PROPPUT:
3444 		propput = (pool_propput_undo_t *)li->li_details;
3445 
3446 		if (propput->ppu_blist)
3447 			nvlist_free(propput->ppu_blist);
3448 		free(propput->ppu_ioctl.pp_o_buf);
3449 		free(propput);
3450 		break;
3451 	case POOL_PROPRM:
3452 		proprm = (pool_proprm_undo_t *)li->li_details;
3453 
3454 		free(proprm);
3455 		break;
3456 	default:
3457 		return (PO_FAIL);
3458 	}
3459 	return (PO_SUCCESS);
3460 }
3461 
3462 /*
3463  * pool_knl_nvlist_add_value() adds a pool_value_t to an nvlist.
3464  */
3465 int
3466 pool_knl_nvlist_add_value(nvlist_t *list, const char *name,
3467     const pool_value_t *pv)
3468 {
3469 	uint64_t uval;
3470 	int64_t ival;
3471 	double dval;
3472 	uchar_t dval_b[sizeof (double)];
3473 	uchar_t bval;
3474 	const char *sval;
3475 	pool_value_class_t type;
3476 	char *nv_name;
3477 
3478 	if ((type = pool_value_get_type(pv)) == POC_INVAL) {
3479 		pool_seterror(POE_BADPARAM);
3480 		return (PO_FAIL);
3481 	}
3482 	nv_name = (char *)name;
3483 
3484 	switch (type) {
3485 	case POC_UINT:
3486 		if (pool_value_get_uint64(pv, &uval) == POC_INVAL) {
3487 			return (PO_FAIL);
3488 		}
3489 		if (nvlist_add_uint64(list, nv_name, uval) != 0) {
3490 			pool_seterror(POE_SYSTEM);
3491 			return (PO_FAIL);
3492 		}
3493 		break;
3494 	case POC_INT:
3495 		if (pool_value_get_int64(pv, &ival) == POC_INVAL) {
3496 			return (PO_FAIL);
3497 		}
3498 		if (nvlist_add_int64(list, nv_name, ival) != 0) {
3499 			pool_seterror(POE_SYSTEM);
3500 			return (PO_FAIL);
3501 		}
3502 		break;
3503 	case POC_DOUBLE:
3504 		if (pool_value_get_double(pv, &dval) == POC_INVAL) {
3505 			return (PO_FAIL);
3506 		}
3507 		/*
3508 		 * Since there is no support for doubles in the
3509 		 * kernel, store the double value in a byte array.
3510 		 */
3511 		(void) memcpy(dval_b, &dval, sizeof (double));
3512 		if (nvlist_add_byte_array(list, nv_name, dval_b,
3513 		    sizeof (double)) != 0) {
3514 			pool_seterror(POE_SYSTEM);
3515 			return (PO_FAIL);
3516 		}
3517 		break;
3518 	case POC_BOOL:
3519 		if (pool_value_get_bool(pv, &bval) == POC_INVAL) {
3520 			return (PO_FAIL);
3521 		}
3522 		if (nvlist_add_byte(list, nv_name, bval) != 0) {
3523 			pool_seterror(POE_SYSTEM);
3524 			return (PO_FAIL);
3525 		}
3526 		break;
3527 	case POC_STRING:
3528 		if (pool_value_get_string(pv, &sval) == POC_INVAL) {
3529 			return (PO_FAIL);
3530 		}
3531 		if (nvlist_add_string(list, nv_name, (char *)sval) != 0) {
3532 			pool_seterror(POE_SYSTEM);
3533 			return (PO_FAIL);
3534 		}
3535 		break;
3536 	default:
3537 		pool_seterror(POE_BADPARAM);
3538 		return (PO_FAIL);
3539 	}
3540 	return (PO_SUCCESS);
3541 }
3542 
3543 /*
3544  * hash_id() hashes all elements in a pool configuration using the
3545  * "sys_id" property. Not all elements have a "sys_id" property,
3546  * however elem_get_sysid() caters for this by always returning a
3547  * constant value for those elements. This isn't anticipated to lead
3548  * to a performance degradation in the hash, since those elements
3549  * which are likely to be most prevalent in a configuration do have
3550  * "sys_id" as a property.
3551  */
3552 uint64_t
3553 hash_id(const pool_elem_t *pe)
3554 {
3555 	id_t id;
3556 
3557 	id = elem_get_sysid(pe);
3558 	return (hash_buf(&id, sizeof (id)));
3559 }
3560 
3561 /*
3562  *  blocking_open() guarantees access to the pool device, if open()
3563  * is failing with EBUSY.
3564  */
3565 int
3566 blocking_open(const char *path, int oflag)
3567 {
3568 	int fd;
3569 
3570 	while ((fd = open(path, oflag)) == -1 && errno == EBUSY)
3571 		(void) poll(NULL, 0, 1 * MILLISEC);
3572 
3573 	return (fd);
3574 }
3575