xref: /illumos-gate/usr/src/lib/libpool/common/pool_kernel.c (revision f17620a4f72a29025a22655ba8735ccd20ae174f)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  * Copyright 2012 Milan Jurik. All rights reserved.
26  */
27 
28 #include <assert.h>
29 #include <errno.h>
30 #include <exacct.h>
31 #include <fcntl.h>
32 #include <libnvpair.h>
33 #include <limits.h>
34 #include <poll.h>
35 #include <pool.h>
36 #include <stdlib.h>
37 #include <stdio.h>
38 #include <string.h>
39 #include <strings.h>
40 #include <stropts.h>
41 #include <thread.h>
42 #include <time.h>
43 #include <unistd.h>
44 
45 #include <libxml/tree.h>
46 #include <libxml/valid.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 transactions.
650 	 * However, don't rollback when we've done a temporary pool since the
651 	 * pool/resources haven't really been committed in this case.
652 	 * They will all be freed in pool_knl_connection_free and we don't
653 	 * want to double free them.
654 	 */
655 	if (!(conf->pc_prov->pc_oflags & PO_TEMP))
656 		(void) pool_knl_rollback(conf);
657 	pool_knl_connection_free(prov);
658 	return (PO_SUCCESS);
659 }
660 
661 /*
662  * Remove elements in this map (previously identified as "dead") from
663  * the configuration map (prov->pkc_elements).
664  */
665 
666 /* ARGSUSED1 */
667 static void
668 remove_dead_elems(const void *key, void **value, void *cl)
669 {
670 	pool_knl_elem_t *pke = (pool_knl_elem_t *)key;
671 	pool_conf_t *conf = TO_CONF(TO_ELEM(pke));
672 	pool_knl_connection_t *prov = (pool_knl_connection_t *)conf->pc_prov;
673 
674 	assert(dict_remove(prov->pkc_elements, pke) != NULL);
675 #ifdef DEBUG
676 	dprintf("remove_dead_elems:\n");
677 	pool_elem_dprintf(TO_ELEM(pke));
678 #endif	/* DEBUG */
679 	pool_knl_elem_free(pke, PO_TRUE);
680 }
681 
682 /*
683  * Find elements which were not updated the last time that
684  * load_group() was called. Add those elements into a separate map
685  * (passed in cl) which will be later used to remove these elements
686  * from the configuration map.
687  */
688 /* ARGSUSED1 */
689 static void
690 find_dead_elems(const void *key, void **value, void *cl)
691 {
692 	pool_knl_elem_t *pke = (pool_knl_elem_t *)key;
693 	pool_conf_t *conf = TO_CONF(TO_ELEM(pke));
694 	pool_knl_connection_t *prov = (pool_knl_connection_t *)conf->pc_prov;
695 	dict_hdl_t *dead_map = (dict_hdl_t *)cl;
696 
697 	if (pke->pke_ltime < prov->pkc_ltime)
698 		(void) dict_put(dead_map, pke, pke);
699 }
700 
701 /*
702  * Update the snapshot held by the library. This function acts as the
703  * controller for the snapshot update procedure. Then snapshot is
704  * actually updated in multiple phases by the load_group() function
705  * (which updates existing elements and creates new elements as
706  * required) and then by find_dead_elems and remove_dead_elems
707  * (respectively responsible for identifying elements which are to be
708  * removed and then removing them).
709  *
710  * Returns PO_SUCCESS
711  */
712 int
713 pool_knl_update(pool_conf_t *conf, int *changed)
714 {
715 	pool_knl_connection_t *prov = (pool_knl_connection_t *)conf->pc_prov;
716 	pool_query_t query = {0};
717 	ea_object_t *ep;
718 	dict_hdl_t *dead_map;
719 	pool_snap_load_t psl = { NULL };
720 
721 	/*
722 	 * Ensure the library snapshot is consistent, if there are any
723 	 * outstanding transactions return failure.
724 	 */
725 	if (log_size(prov->pkc_log) != 0) {
726 		pool_seterror(POE_INVALID_CONF);
727 		return (PO_FAIL);
728 	}
729 	/*
730 	 * Query the kernel for a snapshot of the configuration state. Use
731 	 * load_group to allocate the user-land representation of the
732 	 * data returned in the snapshot.
733 	 */
734 	/* LINTED E_CONSTANT_CONDITION */
735 	while (1) {
736 		if (ioctl(prov->pkc_fd, POOL_QUERY, &query) < 0) {
737 			pool_seterror(POE_SYSTEM);
738 			return (PO_FAIL);
739 		}
740 		if ((query.pq_io_buf = calloc(1,
741 		    (query.pq_io_bufsize < KERNEL_SNAPSHOT_BUF_SZ) ?
742 		    query.pq_io_bufsize * 2 : query.pq_io_bufsize)) == NULL) {
743 			pool_seterror(POE_SYSTEM);
744 			return (PO_FAIL);
745 		}
746 		if (ioctl(prov->pkc_fd, POOL_QUERY, &query) < 0) {
747 			free(query.pq_io_buf);
748 			if (errno != ENOMEM) {
749 				pool_seterror(POE_SYSTEM);
750 				return (PO_FAIL);
751 			}
752 			query.pq_io_bufsize = 0;
753 			query.pq_io_buf = NULL;
754 		} else
755 			break;
756 	}
757 	if (ea_unpack_object(&ep, EUP_NOALLOC, query.pq_io_buf,
758 	    query.pq_io_bufsize) != EO_GROUP) {
759 		free(query.pq_io_buf);
760 		pool_seterror(POE_DATASTORE);
761 		return (PO_FAIL);
762 	}
763 	/*
764 	 * Update the library snapshot
765 	 */
766 	psl.psl_changed = changed;
767 	prov->pkc_lotime = prov->pkc_ltime;
768 	if (load_group(conf, NULL, ep, &psl) != PO_SUCCESS) {
769 		free(query.pq_io_buf);
770 		ea_free_object(ep, EUP_NOALLOC);
771 		return (PO_FAIL);
772 	}
773 
774 	free(query.pq_io_buf);
775 	ea_free_object(ep, EUP_NOALLOC);
776 	/*
777 	 * Now search the dictionary for items that must be removed because
778 	 * they were neither created nor updated.
779 	 */
780 	if ((dead_map = dict_new((int (*)(const void *, const void *))
781 	    pool_elem_compare, (uint64_t (*)(const void *))hash_id)) == NULL) {
782 		pool_seterror(POE_SYSTEM);
783 		return (PO_FAIL);
784 	}
785 	dict_map(prov->pkc_elements, find_dead_elems, dead_map);
786 
787 	if (dict_length(dead_map) > 0) {
788 		dict_map(dead_map, remove_dead_elems, NULL);
789 	}
790 	dict_free(&dead_map);
791 
792 	return (PO_SUCCESS);
793 }
794 
795 /*
796  * Rely on the kernel to always keep a kernel configuration valid.
797  * Returns PO_SUCCESS
798  */
799 /* ARGSUSED */
800 int
801 pool_knl_validate(const pool_conf_t *conf, pool_valid_level_t level)
802 {
803 	return ((conf->pc_state == POF_INVALID) ? PO_FAIL : PO_SUCCESS);
804 }
805 
806 /*
807  * Process all the outstanding transactions in the log. If the processing
808  * fails, then attempt to rollback and "undo" the changes.
809  */
810 int
811 pool_knl_commit(pool_conf_t *conf)
812 {
813 	pool_knl_connection_t *prov = (pool_knl_connection_t *)conf->pc_prov;
814 	int lock = 1;
815 
816 	/*
817 	 * Lock the kernel state for the commit
818 	 */
819 	if (ioctl(prov->pkc_fd, POOL_COMMIT, lock) < 0) {
820 		pool_seterror(POE_SYSTEM);
821 		return (PO_FAIL);
822 	}
823 	lock = 0;
824 	/*
825 	 * If the state is LS_FAIL, then try to recover before
826 	 * performing the commit.
827 	 */
828 	if (prov->pkc_log->l_state == LS_FAIL) {
829 		if (pool_knl_recover(conf) == PO_FAIL) {
830 			/*
831 			 * Unlock the kernel state for the
832 			 * commit. Assert that this * can't fail,
833 			 * since if it ever does fail the library is
834 			 * unusable.
835 			 */
836 			assert(ioctl(prov->pkc_fd, POOL_COMMIT, lock) >= 0);
837 		}
838 	}
839 	/*
840 	 * Commit the log
841 	 */
842 	if (log_walk(prov->pkc_log, log_item_commit) != PO_SUCCESS) {
843 		(void) pool_knl_recover(conf);
844 		/*
845 		 * Unlock the kernel state for the commit. Assert that
846 		 * this can't fail, since if it ever does fail the
847 		 * library is unusable.
848 		 */
849 		assert(ioctl(prov->pkc_fd, POOL_COMMIT, lock) >= 0);
850 		pool_seterror(POE_SYSTEM);
851 		return (PO_FAIL);
852 	}
853 	/*
854 	 * Unlock the kernel state for the commit. Assert that this
855 	 * can't fail, since if it ever does fail the library is
856 	 * unusable.
857 	 */
858 	assert(ioctl(prov->pkc_fd, POOL_COMMIT, lock) >= 0);
859 	/*
860 	 * Release the log resources
861 	 */
862 	(void) log_walk(prov->pkc_log, log_item_release);
863 	log_empty(prov->pkc_log);
864 	return (PO_SUCCESS);
865 }
866 
867 /*
868  * prop_build_cb() is designed to be called from
869  * pool_walk_properties(). The property value is used to put an XML
870  * property on the supplied ktx_node. This is an essential part of the
871  * mechanism used to export a kernel configuration in libpool XML
872  * form.
873  */
874 /* ARGSUSED */
875 static int
876 prop_build_cb(pool_conf_t *UNUSED, pool_elem_t *pe, const char *name,
877     pool_value_t *pval, void *user)
878 {
879 	struct knl_to_xml *info = (struct knl_to_xml *)user;
880 
881 	return (pool_knl_put_xml_property((pool_elem_t *)pe, info->ktx_node,
882 	    name, pval));
883 }
884 
885 /*
886  * Duplicate some of the functionality from pool_xml_put_property()
887  * (see pool_xml.c) to allow a kernel configuration to add XML nodes
888  * to an XML tree which represents the kernel configuration. This is
889  * an essential part of the mechanism used to export a kernel
890  * configuration in libpool XML form.
891  */
892 int
893 pool_knl_put_xml_property(pool_elem_t *pe, xmlNodePtr node, const char *name,
894     const pool_value_t *val)
895 {
896 
897 	/*
898 	 * "type" is a special attribute which is not visible ever outside of
899 	 * libpool. Use the specific type accessor function.
900 	 */
901 	if (strcmp(name, c_type) == 0) {
902 		return (pool_xml_set_attr(node, BAD_CAST name,
903 		    val));
904 	}
905 	if (is_ns_property(pe, name) != NULL) {	/* in ns */
906 		if (pool_xml_set_attr(node,
907 		    BAD_CAST property_name_minus_ns(pe, name), val) == PO_FAIL)
908 			return (pool_xml_set_prop(node, BAD_CAST name,
909 			    val));
910 	} else
911 		return (pool_xml_set_prop(node, BAD_CAST name, val));
912 	return (PO_SUCCESS);
913 }
914 
915 /*
916  * Export the kernel configuration as an XML file. The configuration
917  * is used to build an XML document in memory. This document is then
918  * saved to the supplied location.
919  */
920 int
921 pool_knl_export(const pool_conf_t *conf, const char *location,
922     pool_export_format_t fmt)
923 {
924 	xmlNodePtr node_comment;
925 	xmlNodePtr system;
926 	int ret;
927 	pool_t **ps;
928 	pool_resource_t **rs;
929 	uint_t nelem;
930 	int i;
931 	struct knl_to_xml info;
932 	char_buf_t *cb = NULL;
933 	xmlValidCtxtPtr cvp;
934 
935 	xml_init();
936 
937 
938 	switch (fmt) {
939 	case POX_NATIVE:
940 		info.ktx_doc = xmlNewDoc(BAD_CAST "1.0");
941 		(void) xmlCreateIntSubset(info.ktx_doc, BAD_CAST "system",
942 		    BAD_CAST "-//Sun Microsystems Inc//DTD Resource "
943 		    "Management All//EN",
944 		    BAD_CAST dtd_location);
945 
946 		if ((cvp = xmlNewValidCtxt()) == NULL) {
947 			xmlFreeDoc(info.ktx_doc);
948 			pool_seterror(POE_DATASTORE);
949 			return (PO_FAIL);
950 		}
951 		/*
952 		 * Call xmlValidateDocument() to force the parsing of
953 		 * the DTD. Ignore errors and warning messages as we
954 		 * know the document isn't valid.
955 		 */
956 		(void) xmlValidateDocument(cvp, info.ktx_doc);
957 		xmlFreeValidCtxt(cvp);
958 		if ((info.ktx_node = node_create(NULL, BAD_CAST "system")) ==
959 		    NULL) {
960 			xmlFreeDoc(info.ktx_doc);
961 			pool_seterror(POE_DATASTORE);
962 			return (PO_FAIL);
963 		}
964 
965 		system = info.ktx_node;
966 		info.ktx_doc->_private = (void *)conf;
967 
968 		(void) xmlDocSetRootElement(info.ktx_doc, info.ktx_node);
969 		(void) xmlSetProp(info.ktx_node, BAD_CAST c_ref_id,
970 		    BAD_CAST "dummy");
971 		if ((node_comment = xmlNewDocComment(info.ktx_doc,
972 		    BAD_CAST "\nConfiguration for pools facility. Do NOT"
973 		    " edit this file by hand - use poolcfg(1)"
974 		    " or libpool(3POOL) instead.\n")) == NULL) {
975 			xmlFreeDoc(info.ktx_doc);
976 			pool_seterror(POE_DATASTORE);
977 			return (PO_FAIL);
978 		}
979 		if (xmlAddPrevSibling(info.ktx_node, node_comment) == NULL) {
980 			xmlFree(node_comment);
981 			xmlFreeDoc(info.ktx_doc);
982 			pool_seterror(POE_DATASTORE);
983 			return (PO_FAIL);
984 		}
985 		if (pool_walk_any_properties((pool_conf_t *)conf,
986 		    pool_conf_to_elem(conf), &info, prop_build_cb, 1) ==
987 		    PO_FAIL) {
988 			xmlFreeDoc(info.ktx_doc);
989 			return (PO_FAIL);
990 		}
991 		if ((cb = alloc_char_buf(CB_DEFAULT_LEN)) == NULL) {
992 			xmlFreeDoc(info.ktx_doc);
993 			return (PO_FAIL);
994 		}
995 		/*
996 		 * Now add pool details
997 		 */
998 		if ((ps = pool_query_pools(conf, &nelem, NULL)) != NULL) {
999 			for (i = 0; i < nelem; i++) {
1000 				pool_elem_t *elem = TO_ELEM(ps[i]);
1001 				uint_t nreselem;
1002 				const char *sep = "";
1003 				int j;
1004 
1005 				if (elem_is_tmp(elem))
1006 					continue;
1007 
1008 				if ((info.ktx_node = node_create(system,
1009 				    BAD_CAST element_class_tags
1010 				    [pool_elem_class(elem)])) == NULL) {
1011 					free(ps);
1012 					free_char_buf(cb);
1013 					xmlFreeDoc(info.ktx_doc);
1014 					pool_seterror(POE_DATASTORE);
1015 					return (PO_FAIL);
1016 				}
1017 				if (pool_walk_any_properties(
1018 				    (pool_conf_t *)conf,
1019 				    elem, &info, prop_build_cb, 1) == PO_FAIL) {
1020 					free(ps);
1021 					free_char_buf(cb);
1022 					xmlFreeDoc(info.ktx_doc);
1023 					return (PO_FAIL);
1024 				}
1025 				/*
1026 				 * TODO: pset specific res manipulation
1027 				 */
1028 				if ((rs = pool_query_pool_resources(conf, ps[i],
1029 				    &nreselem, NULL)) == NULL) {
1030 					free(ps);
1031 					free_char_buf(cb);
1032 					xmlFreeDoc(info.ktx_doc);
1033 					pool_seterror(POE_INVALID_CONF);
1034 					return (PO_FAIL);
1035 				}
1036 				if (set_char_buf(cb, "") == PO_FAIL) {
1037 					free(rs);
1038 					free(ps);
1039 					free_char_buf(cb);
1040 					xmlFreeDoc(info.ktx_doc);
1041 					return (PO_FAIL);
1042 				}
1043 				for (j = 0; j < nreselem; j++) {
1044 					pool_elem_t *reselem = TO_ELEM(rs[j]);
1045 					if (append_char_buf(cb, "%s%s_%d", sep,
1046 					    pool_elem_class_string(reselem),
1047 					    (int)elem_get_sysid(reselem)) ==
1048 					    PO_FAIL) {
1049 						free(rs);
1050 						free(ps);
1051 						free_char_buf(cb);
1052 						xmlFreeDoc(info.ktx_doc);
1053 						return (PO_FAIL);
1054 					}
1055 					sep = " ";
1056 				}
1057 				free(rs);
1058 				(void) xmlSetProp(info.ktx_node, BAD_CAST "res",
1059 				    BAD_CAST cb->cb_buf);
1060 				if (set_char_buf(cb, "%s_%d",
1061 				    pool_elem_class_string(elem),
1062 				    (int)elem_get_sysid(elem)) == PO_FAIL) {
1063 					free(ps);
1064 					free_char_buf(cb);
1065 					xmlFreeDoc(info.ktx_doc);
1066 					return (PO_FAIL);
1067 				}
1068 				(void) xmlSetProp(info.ktx_node,
1069 				    BAD_CAST c_ref_id,
1070 				    BAD_CAST  cb->cb_buf);
1071 			}
1072 			free(ps);
1073 		}
1074 		/*
1075 		 * Now add resource details (including components)
1076 		 */
1077 		if ((rs = pool_query_resources(conf, &nelem, NULL)) != NULL) {
1078 			for (i = 0; i < nelem; i++) {
1079 				pool_elem_t *elem = TO_ELEM(rs[i]);
1080 				pool_component_t **cs = NULL;
1081 				uint_t ncompelem;
1082 				int j;
1083 
1084 				if (elem_is_tmp(elem))
1085 					continue;
1086 
1087 				if ((info.ktx_node = node_create(system,
1088 				    BAD_CAST element_class_tags
1089 				    [pool_elem_class(elem)])) == NULL) {
1090 					free(rs);
1091 					free_char_buf(cb);
1092 					xmlFreeDoc(info.ktx_doc);
1093 					pool_seterror(POE_DATASTORE);
1094 					return (PO_FAIL);
1095 				}
1096 				if (pool_walk_any_properties(
1097 				    (pool_conf_t *)conf,
1098 				    elem, &info, prop_build_cb, 1) == PO_FAIL) {
1099 					free(rs);
1100 					free_char_buf(cb);
1101 					xmlFreeDoc(info.ktx_doc);
1102 					return (PO_FAIL);
1103 				}
1104 				if (set_char_buf(cb, "%s_%d",
1105 				    pool_elem_class_string(elem),
1106 				    (int)elem_get_sysid(elem)) == PO_FAIL) {
1107 					free(rs);
1108 					free_char_buf(cb);
1109 					xmlFreeDoc(info.ktx_doc);
1110 					return (PO_FAIL);
1111 				}
1112 				(void) xmlSetProp(info.ktx_node,
1113 				    BAD_CAST c_ref_id,
1114 				    BAD_CAST  cb->cb_buf);
1115 				if ((cs = pool_query_resource_components(conf,
1116 				    rs[i], &ncompelem, NULL)) != NULL) {
1117 					xmlNodePtr resource = info.ktx_node;
1118 
1119 					for (j = 0; j < ncompelem; j++) {
1120 						pool_elem_t *compelem =
1121 						    TO_ELEM(cs[j]);
1122 						if ((info.ktx_node =
1123 						    node_create(resource,
1124 						    BAD_CAST element_class_tags
1125 						    [pool_elem_class(
1126 						    compelem)])) == NULL) {
1127 							pool_seterror(
1128 							    POE_DATASTORE);
1129 							free(rs);
1130 							free(cs);
1131 							free_char_buf(cb);
1132 							xmlFreeDoc(info.
1133 							    ktx_doc);
1134 							return (PO_FAIL);
1135 						}
1136 						if (pool_walk_any_properties(
1137 						    (pool_conf_t *)conf,
1138 						    compelem, &info,
1139 						    prop_build_cb, 1) ==
1140 						    PO_FAIL) {
1141 							free(rs);
1142 							free(cs);
1143 							free_char_buf(cb);
1144 							xmlFreeDoc(info.
1145 							    ktx_doc);
1146 							return (PO_FAIL);
1147 						}
1148 						if (set_char_buf(cb, "%s_%d",
1149 						    pool_elem_class_string(
1150 						    compelem),
1151 						    (int)elem_get_sysid(
1152 						    compelem)) == PO_FAIL) {
1153 							free(rs);
1154 							free(cs);
1155 							free_char_buf(cb);
1156 							xmlFreeDoc(info.
1157 							    ktx_doc);
1158 							return (PO_FAIL);
1159 						}
1160 						(void) xmlSetProp(info.ktx_node,
1161 						    BAD_CAST c_ref_id,
1162 						    BAD_CAST  cb->cb_buf);
1163 					}
1164 					free(cs);
1165 				}
1166 			}
1167 			free(rs);
1168 		}
1169 		free_char_buf(cb);
1170 		/*
1171 		 * Set up the message handlers prior to calling
1172 		 * xmlValidateDocument()
1173 		 */
1174 		if ((cvp = xmlNewValidCtxt()) == NULL) {
1175 			xmlFreeDoc(info.ktx_doc);
1176 			pool_seterror(POE_DATASTORE);
1177 			return (PO_FAIL);
1178 		}
1179 		cvp->error    = pool_error_func;
1180 		cvp->warning  = pool_error_func;
1181 		if (xmlValidateDocument(cvp, info.ktx_doc) == 0) {
1182 			xmlFreeValidCtxt(cvp);
1183 			xmlFreeDoc(info.ktx_doc);
1184 			pool_seterror(POE_INVALID_CONF);
1185 			return (PO_FAIL);
1186 		}
1187 		xmlFreeValidCtxt(cvp);
1188 		ret = xmlSaveFormatFile(location, info.ktx_doc, 1);
1189 		xmlFreeDoc(info.ktx_doc);
1190 		if (ret == -1) {
1191 			pool_seterror(POE_SYSTEM);
1192 			return (PO_FAIL);
1193 		}
1194 		return (PO_SUCCESS);
1195 	default:
1196 		pool_seterror(POE_BADPARAM);
1197 		return (PO_FAIL);
1198 	}
1199 }
1200 
1201 /*
1202  * Rollback the changes to the kernel
1203  */
1204 int
1205 pool_knl_recover(pool_conf_t *conf)
1206 {
1207 	pool_knl_connection_t *prov = (pool_knl_connection_t *)conf->pc_prov;
1208 
1209 	prov->pkc_log->l_state = LS_RECOVER;
1210 	if (log_reverse_walk(prov->pkc_log, log_item_undo) != PO_SUCCESS) {
1211 		dprintf("Library configuration consistency error\n");
1212 		prov->pkc_log->l_state = LS_FAIL;
1213 		pool_seterror(POE_INVALID_CONF);
1214 		return (PO_FAIL);
1215 	}
1216 	prov->pkc_log->l_state = LS_DO;
1217 	return (PO_SUCCESS);
1218 }
1219 
1220 /*
1221  * Rollback the changes to the configuration
1222  */
1223 int
1224 pool_knl_rollback(pool_conf_t *conf)
1225 {
1226 	pool_knl_connection_t *prov = (pool_knl_connection_t *)conf->pc_prov;
1227 
1228 	prov->pkc_log->l_state = LS_UNDO;
1229 	if (log_reverse_walk(prov->pkc_log, log_item_undo) != PO_SUCCESS) {
1230 		dprintf("Kernel configuration consistency error\n");
1231 		(void) log_walk(prov->pkc_log, log_item_release);
1232 		log_empty(prov->pkc_log);
1233 		prov->pkc_log->l_state = LS_FAIL;
1234 		pool_seterror(POE_INVALID_CONF);
1235 		return (PO_FAIL);
1236 	}
1237 	(void) log_walk(prov->pkc_log, log_item_release);
1238 	log_empty(prov->pkc_log);
1239 	prov->pkc_log->l_state = LS_DO;
1240 	return (PO_SUCCESS);
1241 }
1242 
1243 /*
1244  * Callback used to build the result set for a query. Each invocation will
1245  * supply a candidate element for inclusion. The element is filtered by:
1246  * - class
1247  * - properties
1248  * If the element "matches" the target, then it is added to the result
1249  * set, otherwise it is ignored.
1250  */
1251 /* ARGSUSED1 */
1252 static void
1253 build_result_set(const void *key, void **value, void *cl)
1254 {
1255 	struct query_obj *qo = (struct query_obj *)cl;
1256 	pool_knl_elem_t *pke = (pool_knl_elem_t *)key;
1257 
1258 	/*
1259 	 * Check to see if it's the right class of element
1260 	 */
1261 	if (qo->classes & (1 << pool_elem_class((pool_elem_t *)key))) {
1262 		int i;
1263 		/*
1264 		 * Now check to see if the src element is correct. If no src
1265 		 * element is supplied, ignore this check
1266 		 */
1267 		if (qo->src) {
1268 			pool_knl_elem_t *parent;
1269 
1270 			for (parent = pke; parent != NULL;
1271 			    parent = parent->pke_parent) {
1272 				if (parent == (pool_knl_elem_t *)qo->src)
1273 					break;
1274 			}
1275 			if (parent == NULL)
1276 				return;
1277 		}
1278 		/*
1279 		 * Now check for property matches (if there are any specified)
1280 		 */
1281 		if (qo->props) {
1282 			int matched = PO_TRUE;
1283 			for (i = 0; qo->props[i] != NULL; i++) {
1284 				pool_value_t val = POOL_VALUE_INITIALIZER;
1285 
1286 				if (pool_get_property(TO_CONF(TO_ELEM(pke)),
1287 				    (pool_elem_t *)pke,
1288 				    pool_value_get_name(qo->props[i]), &val) ==
1289 				    POC_INVAL) {
1290 					matched = PO_FALSE;
1291 					break;
1292 				} else {
1293 					if (pool_value_equal(qo->props[i],
1294 					    &val) != PO_TRUE) {
1295 						matched = PO_FALSE;
1296 						break;
1297 					}
1298 				}
1299 			}
1300 			if (matched == PO_TRUE)
1301 				(void) pool_knl_result_set_append(qo->rs,
1302 				    (pool_knl_elem_t *)key);
1303 		} else {
1304 			(void) pool_knl_result_set_append(qo->rs,
1305 			    (pool_knl_elem_t *)key);
1306 		}
1307 	}
1308 }
1309 
1310 /*
1311  * Execute the supplied query and return a result set which contains
1312  * all qualifying elements.
1313  */
1314 pool_result_set_t *
1315 pool_knl_exec_query(const pool_conf_t *conf, const pool_elem_t *src,
1316     const char *src_attr, pool_elem_class_t classes, pool_value_t **props)
1317 {
1318 	pool_knl_result_set_t *rs;
1319 	pool_knl_connection_t *prov = (pool_knl_connection_t *)conf->pc_prov;
1320 	struct query_obj qo;
1321 	int matched = PO_TRUE;
1322 
1323 	/*
1324 	 * Have a buffer at this point, that we can use
1325 	 */
1326 	if ((rs = pool_knl_result_set_alloc(conf)) == NULL) {
1327 		return (NULL);
1328 	}
1329 	qo.conf = conf;
1330 	qo.src = src;
1331 	qo.src_attr = src_attr;
1332 	qo.classes = classes;
1333 	qo.props = props;
1334 	qo.rs = rs;
1335 	if (src_attr != NULL) {
1336 		pool_knl_pool_t *pkp = (pool_knl_pool_t *)src;
1337 
1338 		/*
1339 		 * Note: This logic is resource specific and must be
1340 		 * extended for additional resource types.
1341 		 */
1342 		/*
1343 		 * Check for property matches (if there are any specified)
1344 		 */
1345 		if (props) {
1346 			int i;
1347 
1348 			for (i = 0; props[i] != NULL; i++) {
1349 				pool_value_t val = POOL_VALUE_INITIALIZER;
1350 
1351 				if (pool_get_property(conf,
1352 				    (pool_elem_t *)pkp->pkp_assoc[PREC_PSET],
1353 				    pool_value_get_name(props[i]), &val) ==
1354 				    POC_INVAL) {
1355 					matched = PO_FALSE;
1356 					break;
1357 				} else {
1358 					if (pool_value_equal(props[i],
1359 					    &val) != PO_TRUE) {
1360 						matched = PO_FALSE;
1361 						break;
1362 					}
1363 				}
1364 			}
1365 		}
1366 
1367 		if (matched == PO_TRUE)
1368 			(void) pool_knl_result_set_append(rs,
1369 			    (pool_knl_elem_t *)pkp->pkp_assoc[PREC_PSET]);
1370 	} else
1371 		dict_map(prov->pkc_elements, build_result_set, &qo);
1372 
1373 	if (rs->pkr_count == 0)
1374 		pool_seterror(POE_INVALID_SEARCH);
1375 	return ((pool_result_set_t *)rs);
1376 }
1377 
1378 /*
1379  * Callback function intended to be used from pool_walk_pools(). If
1380  * the supplied pool is not the default pool attempt to destroy it.
1381  */
1382 /*ARGSUSED*/
1383 static int
1384 destroy_pool_cb(pool_conf_t *conf, pool_t *pool, void *unused)
1385 {
1386 	if (elem_is_default(TO_ELEM(pool)) != PO_TRUE)
1387 		return (pool_destroy(conf, pool));
1388 	/*
1389 	 * Return PO_SUCCESS even though we don't delete the default
1390 	 * pool so that the walk continues
1391 	 */
1392 	return (PO_SUCCESS);
1393 }
1394 
1395 /*
1396  * Remove the configuration details. This means remove all elements
1397  * apart from the system elements.
1398  */
1399 int
1400 pool_knl_remove(pool_conf_t *conf)
1401 {
1402 	uint_t i, nelem;
1403 	pool_resource_t **resources;
1404 
1405 	conf->pc_state = POF_DESTROY;
1406 	if ((resources = pool_query_resources(conf, &nelem, NULL)) != NULL) {
1407 		for (i = 0; i < nelem; i++) {
1408 			if (resource_is_system(resources[i]) == PO_FALSE)
1409 				if (pool_resource_destroy(conf, resources[i]) !=
1410 				    PO_SUCCESS) {
1411 					pool_seterror(POE_INVALID_CONF);
1412 					return (PO_FAIL);
1413 				}
1414 		}
1415 		free(resources);
1416 	}
1417 	(void) pool_walk_pools(conf, conf, destroy_pool_cb);
1418 	if (pool_conf_commit(conf, PO_FALSE) != PO_SUCCESS)
1419 		return (PO_FAIL);
1420 
1421 	if (pool_conf_close(conf) != PO_SUCCESS)
1422 		return (PO_FAIL);
1423 
1424 	return (PO_SUCCESS);
1425 }
1426 
1427 /*
1428  * Determine the name of the pool to which the supplied pid is
1429  * bound. If it cannot be determined return NULL.
1430  */
1431 char *
1432 pool_knl_get_binding(pool_conf_t *conf, pid_t pid)
1433 {
1434 	pool_knl_connection_t *prov = (pool_knl_connection_t *)conf->pc_prov;
1435 	const char *sval;
1436 	char *name = NULL;
1437 	pool_bindq_t bindq;
1438 	pool_value_t *props[] = { NULL, NULL };
1439 	uint_t nelem = 0;
1440 	pool_t **pools;
1441 	pool_value_t val = POOL_VALUE_INITIALIZER;
1442 
1443 	props[0] = &val;
1444 
1445 	bindq.pb_o_id_type = P_PID;
1446 	bindq.pb_o_id = pid;
1447 	if (ioctl(prov->pkc_fd, POOL_BINDQ, &bindq) < 0) {
1448 		pool_seterror(POE_SYSTEM);
1449 		return (NULL);
1450 	}
1451 
1452 	if (pool_value_set_name(props[0], "pool.sys_id") != PO_SUCCESS) {
1453 		return (NULL);
1454 	}
1455 	pool_value_set_int64(props[0], bindq.pb_i_id);
1456 	if ((pools = pool_query_pools(conf, &nelem, props)) == NULL) {
1457 		pool_seterror(POE_BADPARAM);
1458 		return (NULL);
1459 	}
1460 
1461 	if (nelem != 1) {
1462 		free(pools);
1463 		pool_seterror(POE_INVALID_CONF);
1464 		return (NULL);
1465 	}
1466 	if (pool_get_ns_property(TO_ELEM(pools[0]), c_name, props[0])
1467 	    == POC_INVAL) {
1468 		free(pools);
1469 		return (NULL);
1470 	}
1471 	if (pool_value_get_string(props[0], &sval) != PO_SUCCESS) {
1472 		free(pools);
1473 		return (NULL);
1474 	}
1475 	if ((name = strdup(sval)) == NULL) {
1476 		free(pools);
1477 		pool_seterror(POE_SYSTEM);
1478 		return (NULL);
1479 	}
1480 	return (name);
1481 }
1482 
1483 /*
1484  * Bind idtype id to the pool name.
1485  */
1486 int
1487 pool_knl_set_binding(pool_conf_t *conf, const char *pool_name, idtype_t idtype,
1488     id_t id)
1489 {
1490 	pool_knl_connection_t *prov = (pool_knl_connection_t *)conf->pc_prov;
1491 	pool_bind_t bind;
1492 	pool_t *pool;
1493 	int ret;
1494 
1495 	if ((pool = pool_get_pool(conf, pool_name)) == NULL)
1496 		return (PO_FAIL);
1497 
1498 	bind.pb_o_id_type = idtype;
1499 	bind.pb_o_id = id;
1500 	bind.pb_o_pool_id = elem_get_sysid(TO_ELEM(pool));
1501 
1502 	while ((ret = ioctl(prov->pkc_fd, POOL_BIND, &bind)) < 0 &&
1503 	    errno == EAGAIN)
1504 		;
1505 	if (ret < 0) {
1506 		pool_seterror(POE_SYSTEM);
1507 		return (PO_FAIL);
1508 	}
1509 	return (PO_SUCCESS);
1510 }
1511 
1512 /*
1513  * pool_knl_get_resource_binding() returns the binding for a pid to
1514  * the supplied type of resource. If a binding cannot be determined,
1515  * NULL is returned.
1516  */
1517 char *
1518 pool_knl_get_resource_binding(pool_conf_t *conf,
1519     pool_resource_elem_class_t type, pid_t pid)
1520 {
1521 	pool_knl_connection_t *prov = (pool_knl_connection_t *)conf->pc_prov;
1522 	const char *sval;
1523 	char *name = NULL;
1524 	pool_bindq_t bindq;
1525 	pool_value_t *props[] = { NULL, NULL };
1526 	uint_t nelem = 0;
1527 	pool_t **pools;
1528 	pool_resource_t **resources;
1529 	pool_value_t val = POOL_VALUE_INITIALIZER;
1530 
1531 	props[0] = &val;
1532 	bindq.pb_o_id_type = P_PID;
1533 	bindq.pb_o_id = pid;
1534 	if (ioctl(prov->pkc_fd, POOL_BINDQ, &bindq) < 0) {
1535 		pool_seterror(POE_SYSTEM);
1536 		return (NULL);
1537 	}
1538 
1539 	if (pool_value_set_name(props[0], "pool.sys_id") != PO_SUCCESS) {
1540 		return (NULL);
1541 	}
1542 	pool_value_set_int64(props[0], bindq.pb_i_id);
1543 	if ((pools = pool_query_pools(conf, &nelem, props)) == NULL) {
1544 		pool_seterror(POE_BADPARAM);
1545 		return (NULL);
1546 	}
1547 
1548 	if (nelem != 1) {
1549 		free(pools);
1550 		pool_seterror(POE_INVALID_CONF);
1551 		return (NULL);
1552 	}
1553 
1554 	if (pool_value_set_string(props[0], pool_resource_type_string(type)) !=
1555 	    PO_SUCCESS ||
1556 	    pool_value_set_name(props[0], c_type) != PO_SUCCESS) {
1557 		free(pools);
1558 		return (NULL);
1559 	}
1560 
1561 	if ((resources = pool_query_pool_resources(conf, pools[0], &nelem,
1562 	    NULL)) == NULL) {
1563 		free(pools);
1564 		pool_seterror(POE_INVALID_CONF);
1565 		return (NULL);
1566 	}
1567 	free(pools);
1568 	if (nelem != 1) {
1569 		free(resources);
1570 		pool_seterror(POE_INVALID_CONF);
1571 		return (NULL);
1572 	}
1573 	if (pool_get_ns_property(TO_ELEM(resources[0]), c_name, props[0]) ==
1574 	    POC_INVAL) {
1575 		free(resources);
1576 		return (NULL);
1577 	}
1578 	free(resources);
1579 	if (pool_value_get_string(props[0], &sval) != PO_SUCCESS) {
1580 		return (NULL);
1581 	}
1582 	if ((name = strdup(sval)) == NULL) {
1583 		pool_seterror(POE_SYSTEM);
1584 		return (NULL);
1585 	}
1586 	return (name);
1587 }
1588 
1589 /*
1590  * Allocate the required library data structure and initialise it.
1591  */
1592 pool_knl_elem_t *
1593 pool_knl_elem_wrap(pool_conf_t *conf, pool_elem_class_t class,
1594     pool_resource_elem_class_t res_class,
1595     pool_component_elem_class_t comp_class)
1596 {
1597 	pool_knl_elem_t *elem;
1598 	pool_elem_t *pe;
1599 
1600 	switch (class) {
1601 	case PEC_SYSTEM:
1602 		if ((elem = malloc(sizeof (pool_knl_system_t))) == NULL) {
1603 			pool_seterror(POE_SYSTEM);
1604 			return (NULL);
1605 		}
1606 		(void) memset(elem, 0, sizeof (pool_knl_system_t));
1607 		break;
1608 	case PEC_POOL:
1609 		if ((elem = malloc(sizeof (pool_knl_pool_t))) == NULL) {
1610 			pool_seterror(POE_SYSTEM);
1611 			return (NULL);
1612 		}
1613 		(void) memset(elem, 0, sizeof (pool_knl_pool_t));
1614 		break;
1615 	case PEC_RES_COMP:
1616 	case PEC_RES_AGG:
1617 		if ((elem = malloc(sizeof (pool_knl_resource_t))) == NULL) {
1618 			pool_seterror(POE_SYSTEM);
1619 			return (NULL);
1620 		}
1621 		(void) memset(elem, 0, sizeof (pool_knl_resource_t));
1622 		break;
1623 	case PEC_COMP:
1624 		if ((elem = malloc(sizeof (pool_knl_component_t))) == NULL) {
1625 			pool_seterror(POE_SYSTEM);
1626 			return (NULL);
1627 		}
1628 		(void) memset(elem, 0, sizeof (pool_knl_component_t));
1629 		break;
1630 	default:
1631 		pool_seterror(POE_BADPARAM);
1632 		return (NULL);
1633 	}
1634 	pe = TO_ELEM(elem);
1635 	pe->pe_conf = conf;
1636 	pe->pe_class = class;
1637 	pe->pe_resource_class = res_class;
1638 	pe->pe_component_class = comp_class;
1639 	/* Set up the function pointers for element manipulation */
1640 	pe->pe_get_prop = pool_knl_get_property;
1641 	pe->pe_put_prop = pool_knl_put_property;
1642 	pe->pe_rm_prop = pool_knl_rm_property;
1643 	pe->pe_get_props = pool_knl_get_properties;
1644 	pe->pe_remove = pool_knl_elem_remove;
1645 	pe->pe_get_container = pool_knl_get_container;
1646 	pe->pe_set_container = pool_knl_set_container;
1647 	/*
1648 	 * Specific initialisation for different types of element
1649 	 */
1650 	if (class == PEC_POOL) {
1651 		pool_knl_pool_t *pp = (pool_knl_pool_t *)elem;
1652 		pp->pp_associate = pool_knl_pool_associate;
1653 		pp->pp_dissociate = pool_knl_pool_dissociate;
1654 		pp->pkp_assoc[PREC_PSET] = (pool_knl_resource_t *)
1655 		    resource_by_sysid(conf, PS_NONE, "pset");
1656 	}
1657 	if (class == PEC_RES_COMP || class == PEC_RES_AGG) {
1658 		pool_knl_resource_t *pr = (pool_knl_resource_t *)elem;
1659 		pr->pr_is_system = pool_knl_resource_is_system;
1660 		pr->pr_can_associate = pool_knl_resource_can_associate;
1661 	}
1662 #if DEBUG
1663 	if (dict_put(((pool_knl_connection_t *)conf->pc_prov)->pkc_leaks,
1664 	    elem, elem) != NULL)
1665 		assert(!"leak map put failed");
1666 	dprintf("allocated %p\n", elem);
1667 #endif	/* DEBUG */
1668 	return (elem);
1669 }
1670 
1671 /*
1672  * Allocate a new pool_knl_elem_t in the supplied configuration of the
1673  * specified class.
1674  * Returns element pointer/NULL
1675  */
1676 pool_elem_t *
1677 pool_knl_elem_create(pool_conf_t *conf, pool_elem_class_t class,
1678     pool_resource_elem_class_t res_class,
1679     pool_component_elem_class_t comp_class)
1680 {
1681 	pool_knl_elem_t *elem;
1682 	pool_create_undo_t *create;
1683 	pool_knl_connection_t *prov = (pool_knl_connection_t *)conf->pc_prov;
1684 	static int id = -3;
1685 	char_buf_t *cb;
1686 
1687 	if ((elem = pool_knl_elem_wrap(conf, class, res_class, comp_class)) ==
1688 	    NULL)
1689 		return (NULL);
1690 
1691 	/*
1692 	 * Allocate an nvlist to hold properties
1693 	 */
1694 	if (nvlist_alloc(&elem->pke_properties, NV_UNIQUE_NAME_TYPE, 0) != 0) {
1695 		pool_knl_elem_free(elem, PO_FALSE);
1696 		pool_seterror(POE_SYSTEM);
1697 		return (NULL);
1698 	}
1699 	/*
1700 	 * Allocate a temporary ID and name until the element is
1701 	 * created for real
1702 	 */
1703 	if ((cb = alloc_char_buf(CB_DEFAULT_LEN)) == NULL) {
1704 		pool_knl_elem_free(elem, PO_TRUE);
1705 		return (NULL);
1706 	}
1707 	if (set_char_buf(cb, "%s.sys_id",
1708 	    pool_elem_class_string((pool_elem_t *)elem)) != PO_SUCCESS) {
1709 		pool_knl_elem_free(elem, PO_TRUE);
1710 		free_char_buf(cb);
1711 		return (NULL);
1712 	}
1713 	(void) nvlist_add_int64(elem->pke_properties, cb->cb_buf, id--);
1714 	if (set_char_buf(cb, "%s.name",
1715 	    pool_elem_class_string((pool_elem_t *)elem)) != PO_SUCCESS) {
1716 		pool_knl_elem_free(elem, PO_TRUE);
1717 		free_char_buf(cb);
1718 		return (NULL);
1719 	}
1720 	(void) nvlist_add_string(elem->pke_properties, cb->cb_buf, "");
1721 	/*
1722 	 * If it's a resource class, it will need an initial size
1723 	 */
1724 	if (class == PEC_RES_COMP || class == PEC_RES_AGG) {
1725 		if (set_char_buf(cb, "%s.size",
1726 		    pool_elem_class_string((pool_elem_t *)elem)) !=
1727 		    PO_SUCCESS) {
1728 			pool_knl_elem_free(elem, PO_TRUE);
1729 			free_char_buf(cb);
1730 			return (NULL);
1731 		}
1732 		(void) nvlist_add_uint64(elem->pke_properties, cb->cb_buf, 0);
1733 	}
1734 	free_char_buf(cb);
1735 
1736 	/*
1737 	 * Register the newly created element
1738 	 */
1739 	if (dict_put(prov->pkc_elements, elem, elem) != NULL) {
1740 		pool_knl_elem_free(elem, PO_TRUE);
1741 		pool_seterror(POE_SYSTEM);
1742 		return (NULL);
1743 	}
1744 
1745 	if (prov->pkc_log->l_state != LS_DO)
1746 		return ((pool_elem_t *)elem);
1747 
1748 	/*
1749 	 * The remaining logic is setting up the arguments for the
1750 	 * POOL_CREATE ioctl and appending the details into the log.
1751 	 */
1752 	if ((create = malloc(sizeof (pool_create_undo_t))) == NULL) {
1753 		pool_seterror(POE_SYSTEM);
1754 		return (NULL);
1755 	}
1756 	create->pcu_ioctl.pc_o_type = class;
1757 	switch (class) {
1758 	case PEC_SYSTEM:
1759 		pool_seterror(POE_BADPARAM);
1760 		free(create);
1761 		return (NULL);
1762 	case PEC_POOL: /* NO-OP */
1763 		break;
1764 	case PEC_RES_COMP:
1765 	case PEC_RES_AGG:
1766 		create->pcu_ioctl.pc_o_sub_type = res_class;
1767 		break;
1768 	case PEC_COMP:
1769 		create->pcu_ioctl.pc_o_sub_type = comp_class;
1770 		break;
1771 	default:
1772 		pool_seterror(POE_BADPARAM);
1773 		free(create);
1774 		return (NULL);
1775 	}
1776 
1777 	create->pcu_elem = (pool_elem_t *)elem;
1778 
1779 	if (log_append(prov->pkc_log, POOL_CREATE, (void *)create) !=
1780 	    PO_SUCCESS) {
1781 		free(create);
1782 		return (NULL);
1783 	}
1784 	return ((pool_elem_t *)elem);
1785 }
1786 
1787 /*
1788  * Remove the details of the element from our userland copy and destroy
1789  * the element (if appropriate) in the kernel.
1790  */
1791 int
1792 pool_knl_elem_remove(pool_elem_t *pe)
1793 {
1794 	pool_knl_connection_t *prov;
1795 	pool_destroy_undo_t *destroy;
1796 
1797 	prov = (pool_knl_connection_t *)(TO_CONF(pe))->pc_prov;
1798 
1799 	if (dict_remove(prov->pkc_elements, pe) == NULL) {
1800 		pool_seterror(POE_SYSTEM);
1801 		return (PO_FAIL);
1802 	}
1803 	if (prov->pkc_log->l_state != LS_DO) {
1804 		return (PO_SUCCESS);
1805 	}
1806 
1807 	/*
1808 	 * The remaining logic is setting up the arguments for the
1809 	 * POOL_DESTROY ioctl and appending the details into the log.
1810 	 */
1811 	if ((destroy = malloc(sizeof (pool_destroy_undo_t))) == NULL) {
1812 		pool_seterror(POE_SYSTEM);
1813 		return (PO_FAIL);
1814 	}
1815 	destroy->pdu_ioctl.pd_o_type = pool_elem_class(pe);
1816 
1817 	if (destroy->pdu_ioctl.pd_o_type == PEC_RES_COMP ||
1818 	    destroy->pdu_ioctl.pd_o_type == PEC_RES_AGG)
1819 		destroy->pdu_ioctl.pd_o_sub_type = pool_resource_elem_class(pe);
1820 
1821 	if (destroy->pdu_ioctl.pd_o_type == PEC_COMP)
1822 		destroy->pdu_ioctl.pd_o_sub_type =
1823 		    pool_component_elem_class(pe);
1824 
1825 	destroy->pdu_elem = pe;
1826 
1827 	if (log_append(prov->pkc_log, POOL_DESTROY, (void *)destroy) !=
1828 	    PO_SUCCESS) {
1829 		free(destroy);
1830 		return (PO_FAIL);
1831 	}
1832 	return (PO_SUCCESS);
1833 }
1834 
1835 /*
1836  * Set the parent of the supplied child to the supplied parent
1837  */
1838 int
1839 pool_knl_set_container(pool_elem_t *pp, pool_elem_t *pc)
1840 {
1841 	pool_knl_elem_t *pkp = (pool_knl_elem_t *)pp;
1842 	pool_knl_elem_t *pkc = (pool_knl_elem_t *)pc;
1843 
1844 	pkc->pke_parent = pkp;
1845 	return (PO_SUCCESS);
1846 }
1847 
1848 /*
1849  * TODO: Needed for msets and ssets.
1850  */
1851 /* ARGSUSED */
1852 int
1853 pool_knl_res_transfer(pool_resource_t *src, pool_resource_t *tgt,
1854     uint64_t size) {
1855 	return (PO_FAIL);
1856 }
1857 
1858 /*
1859  * Transfer resource components from one resource set to another.
1860  */
1861 int
1862 pool_knl_res_xtransfer(pool_resource_t *src, pool_resource_t *tgt,
1863     pool_component_t **rl) {
1864 	pool_elem_t *src_e = TO_ELEM(src);
1865 	pool_elem_t *tgt_e = TO_ELEM(tgt);
1866 	pool_xtransfer_undo_t *xtransfer;
1867 	size_t size;
1868 	pool_knl_connection_t *prov =
1869 	    (pool_knl_connection_t *)TO_CONF(src_e)->pc_prov;
1870 
1871 	if (prov->pkc_log->l_state != LS_DO) {
1872 		/*
1873 		 * Walk the Result Set and move the resource components
1874 		 */
1875 		for (size = 0; rl[size] != NULL; size++) {
1876 			if (pool_set_container(TO_ELEM(tgt),
1877 			    TO_ELEM(rl[size])) == PO_FAIL) {
1878 				return (PO_FAIL);
1879 			}
1880 		}
1881 		return (PO_SUCCESS);
1882 	}
1883 
1884 	/*
1885 	 * The remaining logic is setting up the arguments for the
1886 	 * POOL_XTRANSFER ioctl and appending the details into the log.
1887 	 */
1888 	if ((xtransfer = malloc(sizeof (pool_xtransfer_undo_t))) == NULL) {
1889 		pool_seterror(POE_SYSTEM);
1890 		return (PO_FAIL);
1891 	}
1892 
1893 	if (pool_elem_class(src_e) == PEC_RES_COMP) {
1894 		xtransfer->pxu_ioctl.px_o_id_type =
1895 		    pool_resource_elem_class(src_e);
1896 	} else {
1897 		pool_seterror(POE_BADPARAM);
1898 		return (PO_FAIL);
1899 	}
1900 
1901 
1902 	for (xtransfer->pxu_ioctl.px_o_complist_size = 0;
1903 	    rl[xtransfer->pxu_ioctl.px_o_complist_size] != NULL;
1904 	    xtransfer->pxu_ioctl.px_o_complist_size++)
1905 		/* calculate the size using the terminating NULL */;
1906 	if ((xtransfer->pxu_ioctl.px_o_comp_list =
1907 		calloc(xtransfer->pxu_ioctl.px_o_complist_size,
1908 		sizeof (id_t))) == NULL) {
1909 		pool_seterror(POE_SYSTEM);
1910 		return (PO_FAIL);
1911 	}
1912 	if ((xtransfer->pxu_rl = calloc(
1913 	    xtransfer->pxu_ioctl.px_o_complist_size + 1,
1914 	    sizeof (pool_component_t *))) == NULL) {
1915 		pool_seterror(POE_SYSTEM);
1916 		return (PO_FAIL);
1917 	}
1918 	(void) memcpy(xtransfer->pxu_rl, rl,
1919 	    xtransfer->pxu_ioctl.px_o_complist_size *
1920 	    sizeof (pool_component_t *));
1921 	xtransfer->pxu_src = src_e;
1922 	xtransfer->pxu_tgt = tgt_e;
1923 
1924 	if (log_append(prov->pkc_log, POOL_XTRANSFER, (void *)xtransfer) !=
1925 	    PO_SUCCESS) {
1926 		free(xtransfer);
1927 		return (PO_FAIL);
1928 	}
1929 	for (size = 0; rl[size] != NULL; size++) {
1930 		if (pool_set_container(TO_ELEM(tgt), TO_ELEM(rl[size])) ==
1931 		    PO_FAIL) {
1932 			return (PO_FAIL);
1933 		}
1934 	}
1935 	return (PO_SUCCESS);
1936 }
1937 
1938 /*
1939  * Return the parent of an element.
1940  */
1941 pool_elem_t *
1942 pool_knl_get_container(const pool_elem_t *pe)
1943 {
1944 	pool_knl_elem_t *pke = (pool_knl_elem_t *)pe;
1945 
1946 	return ((pool_elem_t *)pke->pke_parent);
1947 }
1948 
1949 /*
1950  * Note: This function is resource specific, needs extending for other
1951  * resource types
1952  */
1953 int
1954 pool_knl_resource_is_system(const pool_resource_t *pr)
1955 {
1956 	switch (pool_resource_elem_class(TO_ELEM(pr))) {
1957 	case PREC_PSET:
1958 		return (PSID_IS_SYSSET(
1959 		    elem_get_sysid(TO_ELEM(pr))));
1960 	default:
1961 		return (PO_FALSE);
1962 	}
1963 }
1964 
1965 /*
1966  * Note: This function is resource specific, needs extending for other
1967  * resource types
1968  */
1969 int
1970 pool_knl_resource_can_associate(const pool_resource_t *pr)
1971 {
1972 	switch (pool_resource_elem_class(TO_ELEM(pr))) {
1973 	case PREC_PSET:
1974 		return (PO_TRUE);
1975 	default:
1976 		return (PO_FALSE);
1977 	}
1978 }
1979 
1980 /*
1981  * pool_knl_pool_associate() associates the supplied resource to the
1982  * supplied pool.
1983  *
1984  * Returns: PO_SUCCESS/PO_FAIL
1985  */
1986 int
1987 pool_knl_pool_associate(pool_t *pool, const pool_resource_t *resource)
1988 {
1989 	pool_knl_connection_t *prov;
1990 	pool_knl_pool_t *pkp = (pool_knl_pool_t *)pool;
1991 	pool_resource_elem_class_t res_class =
1992 	    pool_resource_elem_class(TO_ELEM(resource));
1993 	pool_assoc_undo_t *assoc;
1994 	pool_knl_resource_t *orig_res = pkp->pkp_assoc[res_class];
1995 
1996 	/*
1997 	 * Are we allowed to associate with this target?
1998 	 */
1999 	if (pool_knl_resource_can_associate(resource) == PO_FALSE) {
2000 		pool_seterror(POE_BADPARAM);
2001 		return (PO_FAIL);
2002 	}
2003 	prov = (pool_knl_connection_t *)(TO_CONF(TO_ELEM(pool)))->pc_prov;
2004 
2005 	if (prov->pkc_log->l_state != LS_DO) {
2006 		pkp->pkp_assoc[res_class] = (pool_knl_resource_t *)resource;
2007 		return (PO_SUCCESS);
2008 	}
2009 
2010 	/*
2011 	 * The remaining logic is setting up the arguments for the
2012 	 * POOL_ASSOC ioctl and appending the details into the log.
2013 	 */
2014 	if ((assoc = malloc(sizeof (pool_assoc_undo_t))) == NULL) {
2015 		pool_seterror(POE_SYSTEM);
2016 		return (PO_FAIL);
2017 	}
2018 	assoc->pau_assoc = TO_ELEM(pool);
2019 	assoc->pau_oldres = (pool_elem_t *)orig_res;
2020 	assoc->pau_newres = TO_ELEM(resource);
2021 
2022 	assoc->pau_ioctl.pa_o_id_type = res_class;
2023 
2024 	if (log_append(prov->pkc_log, POOL_ASSOC, (void *)assoc) !=
2025 	    PO_SUCCESS) {
2026 		free(assoc);
2027 		pkp->pkp_assoc[res_class] = orig_res;
2028 		return (PO_FAIL);
2029 	}
2030 	pkp->pkp_assoc[res_class] = (pool_knl_resource_t *)resource;
2031 	return (PO_SUCCESS);
2032 }
2033 
2034 /*
2035  * pool_knl_pool_dissociate() dissociates the supplied resource from
2036  * the supplied pool.
2037  *
2038  * Returns: PO_SUCCESS/PO_FAIL
2039  */
2040 int
2041 pool_knl_pool_dissociate(pool_t *pool, const pool_resource_t *resource)
2042 {
2043 	pool_knl_connection_t *prov;
2044 	pool_dissoc_undo_t *dissoc;
2045 	pool_knl_pool_t *pkp = (pool_knl_pool_t *)pool;
2046 	pool_resource_t *default_res = (pool_resource_t *)get_default_resource(
2047 	    resource);
2048 	pool_resource_elem_class_t res_class =
2049 	    pool_resource_elem_class(TO_ELEM(resource));
2050 
2051 	prov = (pool_knl_connection_t *)(TO_CONF(TO_ELEM(pool)))->pc_prov;
2052 
2053 	if (prov->pkc_log->l_state != LS_DO) {
2054 		pkp->pkp_assoc[res_class] = (pool_knl_resource_t *)default_res;
2055 		return (PO_SUCCESS);
2056 	}
2057 	/*
2058 	 * The remaining logic is setting up the arguments for the
2059 	 * POOL_DISSOC ioctl and appending the details into the log.
2060 	 */
2061 	if ((dissoc = malloc(sizeof (pool_dissoc_undo_t))) == NULL) {
2062 		pool_seterror(POE_SYSTEM);
2063 		return (PO_FAIL);
2064 	}
2065 	dissoc->pdu_dissoc = TO_ELEM(pool);
2066 	dissoc->pdu_oldres = TO_ELEM(resource);
2067 	dissoc->pdu_newres = TO_ELEM(default_res);
2068 
2069 	dissoc->pdu_ioctl.pd_o_id_type = res_class;
2070 
2071 	if (log_append(prov->pkc_log, POOL_DISSOC, (void *)dissoc) !=
2072 	    PO_SUCCESS) {
2073 		free(dissoc);
2074 		pkp->pkp_assoc[res_class] = (pool_knl_resource_t *)resource;
2075 		return (PO_FAIL);
2076 	}
2077 
2078 	/*
2079 	 * Update our local copy
2080 	 */
2081 	pkp->pkp_assoc[res_class] = (pool_knl_resource_t *)default_res;
2082 	return (PO_SUCCESS);
2083 }
2084 
2085 /*
2086  * Allocate a data provider for the supplied configuration and optionally
2087  * discover resources.
2088  * The data provider is the cross over point from the "abstract" configuration
2089  * functions into the data representation specific manipulation routines.
2090  * This function sets up all the required pointers to create a kernel aware
2091  * data provider.
2092  * Returns PO_SUCCESS/PO_FAIL
2093  */
2094 int
2095 pool_knl_connection_alloc(pool_conf_t *conf, int oflags)
2096 {
2097 	pool_knl_connection_t *prov;
2098 
2099 	if ((prov = malloc(sizeof (pool_knl_connection_t))) == NULL) {
2100 		pool_seterror(POE_SYSTEM);
2101 		return (PO_FAIL);
2102 	}
2103 	(void) memset(prov, 0, sizeof (pool_knl_connection_t));
2104 	/*
2105 	 * Initialise data members
2106 	 */
2107 	prov->pc_name = strdup("kernel");
2108 	prov->pc_store_type = KERNEL_DATA_STORE;
2109 	prov->pc_oflags = oflags;
2110 	/*
2111 	 * Initialise function pointers
2112 	 */
2113 	prov->pc_close = pool_knl_close;
2114 	prov->pc_validate = pool_knl_validate;
2115 	prov->pc_commit = pool_knl_commit;
2116 	prov->pc_export = pool_knl_export;
2117 	prov->pc_rollback = pool_knl_rollback;
2118 	prov->pc_exec_query = pool_knl_exec_query;
2119 	prov->pc_elem_create = pool_knl_elem_create;
2120 	prov->pc_remove = pool_knl_remove;
2121 	prov->pc_res_xfer = pool_knl_res_transfer;
2122 	prov->pc_res_xxfer = pool_knl_res_xtransfer;
2123 	prov->pc_get_binding = pool_knl_get_binding;
2124 	prov->pc_set_binding = pool_knl_set_binding;
2125 	prov->pc_get_resource_binding = pool_knl_get_resource_binding;
2126 	/*
2127 	 * Associate the provider to it's configuration
2128 	 */
2129 	conf->pc_prov = (pool_connection_t *)prov;
2130 	/*
2131 	 * End of common initialisation
2132 	 */
2133 	/*
2134 	 * Attempt to open the pseudo device, if the configuration is opened
2135 	 * readonly then try to open an info device, otherwise try to open
2136 	 * the writeable device.
2137 	 */
2138 	if (oflags & PO_RDWR) {
2139 		if ((prov->pkc_fd = blocking_open(pool_dynamic_location(),
2140 		    O_RDWR)) < 0) {
2141 			free(prov);
2142 			conf->pc_prov = NULL;
2143 			pool_seterror(POE_SYSTEM);
2144 			return (PO_FAIL);
2145 		}
2146 	} else {
2147 		if ((prov->pkc_fd = open(pool_info_location, O_RDWR)) < 0) {
2148 			free(prov);
2149 			conf->pc_prov = NULL;
2150 			pool_seterror(POE_SYSTEM);
2151 			return (PO_FAIL);
2152 		}
2153 	}
2154 	/*
2155 	 * Allocate the element dictionary
2156 	 */
2157 	if ((prov->pkc_elements = dict_new((int (*)(const void *, const void *))
2158 	    pool_elem_compare, (uint64_t (*)(const void *))hash_id)) == NULL) {
2159 		(void) close(prov->pkc_fd);
2160 		free(prov);
2161 		conf->pc_prov = NULL;
2162 		pool_seterror(POE_SYSTEM);
2163 		return (PO_FAIL);
2164 	}
2165 #if DEBUG
2166 	if ((prov->pkc_leaks = dict_new(NULL, NULL)) == NULL) {
2167 		dict_free(&prov->pkc_elements);
2168 		(void) close(prov->pkc_fd);
2169 		free(prov);
2170 		conf->pc_prov = NULL;
2171 		pool_seterror(POE_SYSTEM);
2172 		return (PO_FAIL);
2173 	}
2174 #endif	/* DEBUG */
2175 	/*
2176 	 * Allocate the transaction log
2177 	 */
2178 	if ((prov->pkc_log = log_alloc(conf)) == NULL) {
2179 #if DEBUG
2180 		dict_free(&prov->pkc_leaks);
2181 #endif	/* DEBUG */
2182 		dict_free(&prov->pkc_elements);
2183 		(void) close(prov->pkc_fd);
2184 		free(prov);
2185 		conf->pc_prov = NULL;
2186 		return (PO_FAIL);
2187 	}
2188 	/*
2189 	 * At this point the configuration provider has been initialized,
2190 	 * mark the configuration as valid so that the various routines
2191 	 * which rely on a valid configuration will work correctly.
2192 	 */
2193 	conf->pc_state = POF_VALID;
2194 	/*
2195 	 * Update the library snapshot from the kernel
2196 	 */
2197 	if (pool_knl_update(conf, NULL) != PO_SUCCESS) {
2198 #if DEBUG
2199 		dict_free(&prov->pkc_leaks);
2200 #endif	/* DEBUG */
2201 		dict_free(&prov->pkc_elements);
2202 		(void) close(prov->pkc_fd);
2203 		free(prov);
2204 		conf->pc_prov = NULL;
2205 		conf->pc_state = POF_INVALID;
2206 		return (PO_FAIL);
2207 	}
2208 	return (PO_SUCCESS);
2209 }
2210 
2211 #if DEBUG
2212 static void
2213 pool_knl_elem_printf_cb(const void *key, void **value, void *cl)
2214 {
2215 	pool_knl_elem_t *pke = (pool_knl_elem_t *)key;
2216 	dict_hdl_t *map = (dict_hdl_t *)cl;
2217 
2218 	dprintf("leak elem:%p\n", pke);
2219 	if (pke->pke_properties != NULL) {
2220 		nvlist_print(stdout, pke->pke_properties);
2221 	} else
2222 		dprintf("no properties\n");
2223 	assert(dict_get(map, pke) == NULL);
2224 }
2225 #endif	/* DEBUG */
2226 /*
2227  * pool_knl_elem_free() releases the resources associated with the
2228  * supplied element.
2229  */
2230 static void
2231 pool_knl_elem_free(pool_knl_elem_t *pke, int freeprop)
2232 {
2233 #if DEBUG
2234 	pool_conf_t *conf = TO_CONF(TO_ELEM(pke));
2235 	if (dict_remove(((pool_knl_connection_t *)conf->pc_prov)->pkc_leaks,
2236 	    pke) == NULL)
2237 		dprintf("%p, wasn't in the leak map\n", pke);
2238 	if (freeprop == PO_TRUE) {
2239 		pool_elem_dprintf(TO_ELEM(pke));
2240 	}
2241 	dprintf("released %p\n", pke);
2242 #endif	/* DEBUG */
2243 	if (freeprop == PO_TRUE) {
2244 		nvlist_free(pke->pke_properties);
2245 	}
2246 	free(pke);
2247 }
2248 
2249 /*
2250  * pool_knl_elem_free_cb() is designed to be used with
2251  * dict_map(). When a connection is freed, this function is used to
2252  * free all element resources.
2253  */
2254 /* ARGSUSED1 */
2255 static void
2256 pool_knl_elem_free_cb(const void *key, void **value, void *cl)
2257 {
2258 	pool_knl_elem_t *pke = (pool_knl_elem_t *)key;
2259 
2260 #ifdef DEBUG
2261 	dprintf("pool_knl_elem_free_cb:\n");
2262 	dprintf("about to release %p ", pke);
2263 	pool_elem_dprintf(TO_ELEM(pke));
2264 #endif	/* DEBUG */
2265 	pool_knl_elem_free(pke, PO_TRUE);
2266 }
2267 
2268 /*
2269  * Free the resources for a kernel data provider.
2270  */
2271 void
2272 pool_knl_connection_free(pool_knl_connection_t *prov)
2273 {
2274 	if (prov->pkc_log != NULL) {
2275 		(void) log_walk(prov->pkc_log, log_item_release);
2276 		log_free(prov->pkc_log);
2277 	}
2278 	if (prov->pkc_elements != NULL) {
2279 		dict_map(prov->pkc_elements, pool_knl_elem_free_cb, NULL);
2280 #if DEBUG
2281 		dprintf("dict length is %llu\n", dict_length(prov->pkc_leaks));
2282 		dict_map(prov->pkc_leaks, pool_knl_elem_printf_cb,
2283 		    prov->pkc_elements);
2284 		assert(dict_length(prov->pkc_leaks) == 0);
2285 		dict_free(&prov->pkc_leaks);
2286 #endif	/* DEBUG */
2287 		dict_free(&prov->pkc_elements);
2288 	}
2289 	free((void *)prov->pc_name);
2290 	free(prov);
2291 }
2292 
2293 /*
2294  * Return the specified property value.
2295  *
2296  * POC_INVAL is returned if an error is detected and the error code is updated
2297  * to indicate the cause of the error.
2298  */
2299 pool_value_class_t
2300 pool_knl_get_property(const pool_elem_t *pe, const char *name,
2301     pool_value_t *val)
2302 {
2303 	pool_knl_elem_t *pke = (pool_knl_elem_t *)pe;
2304 	nvpair_t *pair;
2305 	const pool_prop_t *prop;
2306 
2307 	if ((prop = provider_get_prop(pe, name)) != NULL)
2308 		if (prop_is_stored(prop) == PO_FALSE)
2309 			return (pool_knl_get_dynamic_property(pe, name, val));
2310 
2311 	if ((pair = pool_knl_find_nvpair(pke->pke_properties, name)) == NULL) {
2312 		pool_seterror(POE_BADPARAM);
2313 		return (POC_INVAL);
2314 	}
2315 
2316 	if (pool_value_from_nvpair(val, pair) == PO_FAIL) {
2317 		return (POC_INVAL);
2318 	}
2319 
2320 	return (pool_value_get_type(val));
2321 }
2322 
2323 /*
2324  * Return the specified property value.
2325  *
2326  * If a property is designated as dynamic, then this function will
2327  * always try to return the latest value of the property from the
2328  * kernel.
2329  *
2330  * POC_INVAL is returned if an error is detected and the error code is updated
2331  * to indicate the cause of the error.
2332  */
2333 pool_value_class_t
2334 pool_knl_get_dynamic_property(const pool_elem_t *pe, const char *name,
2335     pool_value_t *val)
2336 {
2337 	pool_knl_connection_t *prov;
2338 	pool_propget_t propget = { 0 };
2339 	nvlist_t *proplist;
2340 	nvpair_t *pair;
2341 
2342 	propget.pp_o_id_type = pool_elem_class(pe);
2343 	if (pool_elem_class(pe) == PEC_RES_COMP ||
2344 	    pool_elem_class(pe) == PEC_RES_AGG)
2345 		propget.pp_o_id_subtype = pool_resource_elem_class(pe);
2346 	if (pool_elem_class(pe) == PEC_COMP)
2347 		propget.pp_o_id_subtype =
2348 		    (pool_resource_elem_class_t)pool_component_elem_class(pe);
2349 
2350 	propget.pp_o_id = elem_get_sysid(pe);
2351 	propget.pp_o_prop_name_size = strlen(name);
2352 	propget.pp_o_prop_name = (char *)name;
2353 	propget.pp_i_bufsize = KERNEL_SNAPSHOT_BUF_SZ;
2354 	propget.pp_i_buf = malloc(KERNEL_SNAPSHOT_BUF_SZ);
2355 	bzero(propget.pp_i_buf, KERNEL_SNAPSHOT_BUF_SZ);
2356 
2357 	prov = (pool_knl_connection_t *)(TO_CONF(pe))->pc_prov;
2358 	if (ioctl(prov->pkc_fd, POOL_PROPGET, &propget) < 0) {
2359 		free(propget.pp_i_buf);
2360 		pool_seterror(POE_SYSTEM);
2361 		return (POC_INVAL);
2362 	}
2363 	if (nvlist_unpack(propget.pp_i_buf, propget.pp_i_bufsize,
2364 	    &proplist, 0) != 0) {
2365 		free(propget.pp_i_buf);
2366 		pool_seterror(POE_SYSTEM);
2367 		return (POC_INVAL);
2368 	}
2369 	free(propget.pp_i_buf);
2370 
2371 	if ((pair = nvlist_next_nvpair(proplist, NULL)) == NULL) {
2372 		nvlist_free(proplist);
2373 		pool_seterror(POE_SYSTEM);
2374 		return (POC_INVAL);
2375 	}
2376 
2377 	if (pool_value_from_nvpair(val, pair) == PO_FAIL) {
2378 		nvlist_free(proplist);
2379 		return (POC_INVAL);
2380 	}
2381 	nvlist_free(proplist);
2382 	return (pool_value_get_type(val));
2383 }
2384 
2385 /*
2386  * Update the specified property value.
2387  *
2388  * PO_FAIL is returned if an error is detected and the error code is updated
2389  * to indicate the cause of the error.
2390  */
2391 int
2392 pool_knl_put_property(pool_elem_t *pe, const char *name,
2393     const pool_value_t *val)
2394 {
2395 	pool_knl_elem_t *pke = (pool_knl_elem_t *)pe;
2396 	pool_knl_connection_t *prov =
2397 	    (pool_knl_connection_t *)(TO_CONF(pe))->pc_prov;
2398 	nvpair_t *bp, *ap;
2399 	pool_propput_undo_t *propput;
2400 	nvlist_t *bl = NULL;
2401 	const pool_prop_t *prop;
2402 
2403 	if ((bp = pool_knl_find_nvpair(pke->pke_properties, name)) != NULL) {
2404 		if (nvlist_alloc(&bl, NV_UNIQUE_NAME_TYPE, 0) != 0) {
2405 			pool_seterror(POE_SYSTEM);
2406 			return (PO_FAIL);
2407 		}
2408 		if (nvlist_add_nvpair(bl, bp) != 0) {
2409 			nvlist_free(bl);
2410 			pool_seterror(POE_SYSTEM);
2411 			return (PO_FAIL);
2412 		}
2413 	}
2414 	if (pool_knl_nvlist_add_value(pke->pke_properties, name, val) !=
2415 	    PO_SUCCESS)
2416 		return (PO_FAIL);
2417 
2418 	if (prov->pkc_log->l_state != LS_DO) {
2419 		nvlist_free(bl);
2420 		return (PO_SUCCESS);
2421 	}
2422 	/*
2423 	 * The remaining logic is setting up the arguments for the
2424 	 * POOL_PROPPUT ioctl and appending the details into the log.
2425 	 */
2426 	if ((propput = malloc(sizeof (pool_propput_undo_t))) == NULL) {
2427 		pool_seterror(POE_SYSTEM);
2428 		return (PO_FAIL);
2429 	}
2430 	(void) memset(propput, 0, sizeof (pool_propput_undo_t));
2431 	propput->ppu_blist = bl;
2432 
2433 	ap = pool_knl_find_nvpair(pke->pke_properties, name);
2434 
2435 	if (nvlist_alloc(&propput->ppu_alist, NV_UNIQUE_NAME_TYPE, 0) != 0) {
2436 		nvlist_free(propput->ppu_blist);
2437 		free(propput);
2438 		pool_seterror(POE_SYSTEM);
2439 		return (PO_FAIL);
2440 	}
2441 	if (nvlist_add_nvpair(propput->ppu_alist, ap) != 0) {
2442 		nvlist_free(propput->ppu_blist);
2443 		nvlist_free(propput->ppu_alist);
2444 		free(propput);
2445 		pool_seterror(POE_SYSTEM);
2446 		return (PO_FAIL);
2447 	}
2448 
2449 	if (nvlist_pack(propput->ppu_alist,
2450 	    (char **)&propput->ppu_ioctl.pp_o_buf,
2451 	    &propput->ppu_ioctl.pp_o_bufsize, NV_ENCODE_NATIVE, 0) != 0) {
2452 		pool_seterror(POE_SYSTEM);
2453 		return (PO_FAIL);
2454 	}
2455 	nvlist_free(propput->ppu_alist);
2456 	propput->ppu_ioctl.pp_o_id_type = pool_elem_class(pe);
2457 	if (pool_elem_class(pe) == PEC_RES_COMP ||
2458 	    pool_elem_class(pe) == PEC_RES_AGG)
2459 		propput->ppu_ioctl.pp_o_id_sub_type =
2460 		    pool_resource_elem_class(pe);
2461 	if (pool_elem_class(pe) == PEC_COMP)
2462 		propput->ppu_ioctl.pp_o_id_sub_type =
2463 		    (pool_resource_elem_class_t)pool_component_elem_class(pe);
2464 
2465 	propput->ppu_elem = pe;
2466 	if ((prop = provider_get_prop(propput->ppu_elem, name)) != NULL) {
2467 		if (prop_is_readonly(prop) == PO_TRUE)
2468 			propput->ppu_doioctl |= KERNEL_PROP_RDONLY;
2469 	}
2470 
2471 	if (log_append(prov->pkc_log, POOL_PROPPUT, (void *)propput) !=
2472 	    PO_SUCCESS) {
2473 		nvlist_free(propput->ppu_blist);
2474 		free(propput);
2475 		return (PO_FAIL);
2476 	}
2477 	return (PO_SUCCESS);
2478 }
2479 
2480 /*
2481  * Remove the specified property value.
2482  *
2483  * PO_FAIL is returned if an error is detected and the error code is
2484  * updated to indicate the cause of the error.
2485  */
2486 int
2487 pool_knl_rm_property(pool_elem_t *pe, const char *name)
2488 {
2489 	pool_knl_elem_t *pke = (pool_knl_elem_t *)pe;
2490 	pool_knl_connection_t *prov =
2491 	    (pool_knl_connection_t *)(TO_CONF(pe))->pc_prov;
2492 	pool_proprm_undo_t *proprm;
2493 
2494 	if (pool_knl_find_nvpair(pke->pke_properties, name) == NULL) {
2495 		pool_seterror(POE_BADPARAM);
2496 		return (PO_FAIL);
2497 	}
2498 
2499 	if ((proprm = malloc(sizeof (pool_proprm_undo_t))) == NULL) {
2500 		pool_seterror(POE_SYSTEM);
2501 		return (PO_FAIL);
2502 	}
2503 	(void) memset(proprm, 0, sizeof (pool_proprm_undo_t));
2504 	proprm->pru_oldval.pv_class = POC_INVAL;
2505 	(void) pool_get_property(TO_CONF(pe), pe, name, &proprm->pru_oldval);
2506 
2507 	if (prov->pkc_log->l_state != LS_DO) {
2508 		free(proprm);
2509 		(void) nvlist_remove_all(pke->pke_properties, (char *)name);
2510 		return (PO_SUCCESS);
2511 	}
2512 	/*
2513 	 * The remaining logic is setting up the arguments for the
2514 	 * POOL_PROPRM ioctl and appending the details into the log.
2515 	 */
2516 
2517 	proprm->pru_ioctl.pp_o_id_type = pool_elem_class(pe);
2518 	if (pool_elem_class(pe) == PEC_RES_COMP ||
2519 	    pool_elem_class(pe) == PEC_RES_AGG)
2520 		proprm->pru_ioctl.pp_o_id_sub_type =
2521 		    pool_resource_elem_class(pe);
2522 
2523 	if (pool_elem_class(pe) == PEC_COMP)
2524 		proprm->pru_ioctl.pp_o_id_sub_type =
2525 		    (pool_resource_elem_class_t)pool_component_elem_class(pe);
2526 
2527 	proprm->pru_ioctl.pp_o_prop_name_size = strlen(name);
2528 	proprm->pru_ioctl.pp_o_prop_name =
2529 	    (char *)pool_value_get_name(&proprm->pru_oldval);
2530 	proprm->pru_elem = pe;
2531 
2532 	if (log_append(prov->pkc_log, POOL_PROPRM, (void *)proprm) !=
2533 	    PO_SUCCESS) {
2534 		free(proprm);
2535 		return (PO_FAIL);
2536 	}
2537 
2538 	(void) nvlist_remove_all(pke->pke_properties, (char *)name);
2539 	return (PO_SUCCESS);
2540 }
2541 
2542 /*
2543  * Return a NULL terminated array of pool_value_t which represents all
2544  * of the properties stored for an element
2545  *
2546  * Return NULL on failure. It is the caller's responsibility to free
2547  * the returned array of values.
2548  */
2549 pool_value_t **
2550 pool_knl_get_properties(const pool_elem_t *pe, uint_t *nprops)
2551 {
2552 	nvpair_t *pair;
2553 	pool_value_t **result;
2554 	pool_knl_elem_t *pke = (pool_knl_elem_t *)pe;
2555 	int i = 0;
2556 
2557 	*nprops = 0;
2558 
2559 	for (pair = nvlist_next_nvpair(pke->pke_properties, NULL); pair != NULL;
2560 	    pair = nvlist_next_nvpair(pke->pke_properties, pair))
2561 		(*nprops)++;
2562 	if ((result = calloc(*nprops + 1, sizeof (pool_value_t *))) == NULL) {
2563 		pool_seterror(POE_SYSTEM);
2564 		return (NULL);
2565 	}
2566 	for (pair = nvlist_next_nvpair(pke->pke_properties, NULL); pair != NULL;
2567 	    pair = nvlist_next_nvpair(pke->pke_properties, pair), i++) {
2568 		result[i] = pool_value_alloc();
2569 		if (pool_value_from_nvpair(result[i], pair) == PO_FAIL) {
2570 			while (i-- >= 0)
2571 				pool_value_free(result[i]);
2572 			free(result);
2573 			return (NULL);
2574 		}
2575 	}
2576 	return (result);
2577 }
2578 
2579 /*
2580  * Append an entry to a result set. Reallocate the array used to store
2581  * results if it's full.
2582  * Returns PO_SUCCESS/PO_FAIL
2583  */
2584 int
2585 pool_knl_result_set_append(pool_knl_result_set_t *rs, pool_knl_elem_t *pke)
2586 {
2587 	if (rs->pkr_count == rs->pkr_size)
2588 		if (pool_knl_result_set_realloc(rs) != PO_SUCCESS)
2589 			return (PO_FAIL);
2590 
2591 	rs->pkr_list[rs->pkr_count++] = pke;
2592 
2593 	return (PO_SUCCESS);
2594 }
2595 
2596 /*
2597  * Resize the array used to store results. A simple doubling strategy
2598  * is used.
2599  * Returns PO_SUCCESS/PO_FAIL
2600  */
2601 int
2602 pool_knl_result_set_realloc(pool_knl_result_set_t *rs)
2603 {
2604 	pool_knl_elem_t **old_list = rs->pkr_list;
2605 	int new_size = rs->pkr_size * 2;
2606 
2607 	if ((rs->pkr_list = realloc(rs->pkr_list,
2608 	    new_size * sizeof (pool_knl_elem_t *))) == NULL) {
2609 		rs->pkr_list = old_list;
2610 		pool_seterror(POE_SYSTEM);
2611 		return (PO_FAIL);
2612 	}
2613 	rs->pkr_size = new_size;
2614 
2615 	return (PO_SUCCESS);
2616 }
2617 
2618 /*
2619  * Allocate a result set. The Result Set stores the result of a query.
2620  * Returns pool_knl_result_set_t pointer/NULL
2621  */
2622 pool_knl_result_set_t *
2623 pool_knl_result_set_alloc(const pool_conf_t *conf)
2624 {
2625 	pool_knl_result_set_t *rs;
2626 
2627 	if ((rs = malloc(sizeof (pool_knl_result_set_t))) == NULL) {
2628 		pool_seterror(POE_SYSTEM);
2629 		return (NULL);
2630 	}
2631 	(void) memset(rs, 0, sizeof (pool_knl_result_set_t));
2632 	rs->pkr_size = KERNEL_RS_INITIAL_SZ;
2633 	if (pool_knl_result_set_realloc(rs) == PO_FAIL) {
2634 		free(rs);
2635 		pool_seterror(POE_SYSTEM);
2636 		return (NULL);
2637 	}
2638 	rs->prs_conf = conf;
2639 	rs->prs_index = -1;
2640 	rs->prs_active = PO_TRUE;
2641 	/* Fix up the result set accessor functions to the knl specfic ones */
2642 	rs->prs_next = pool_knl_rs_next;
2643 	rs->prs_prev = pool_knl_rs_prev;
2644 	rs->prs_first = pool_knl_rs_first;
2645 	rs->prs_last = pool_knl_rs_last;
2646 	rs->prs_get_index = pool_knl_rs_get_index;
2647 	rs->prs_set_index = pool_knl_rs_set_index;
2648 	rs->prs_close = pool_knl_rs_close;
2649 	rs->prs_count = pool_knl_rs_count;
2650 	return (rs);
2651 }
2652 
2653 /*
2654  * Free a result set. Ensure that the resources are all released at
2655  * this point.
2656  */
2657 void
2658 pool_knl_result_set_free(pool_knl_result_set_t *rs)
2659 {
2660 	free(rs->pkr_list);
2661 	free(rs);
2662 }
2663 /*
2664  * Return the next element in a result set.
2665  * Returns pool_elem_t pointer/NULL
2666  */
2667 pool_elem_t *
2668 pool_knl_rs_next(pool_result_set_t *set)
2669 {
2670 	pool_knl_result_set_t *kset = (pool_knl_result_set_t *)set;
2671 
2672 	if (kset->prs_index == kset->pkr_count - 1)
2673 		return (NULL);
2674 	return ((pool_elem_t *)kset->pkr_list[++kset->prs_index]);
2675 }
2676 
2677 /*
2678  * Return the previous element in a result set.
2679  * Returns pool_elem_t pointer/NULL
2680  */
2681 pool_elem_t *
2682 pool_knl_rs_prev(pool_result_set_t *set)
2683 {
2684 	pool_knl_result_set_t *kset = (pool_knl_result_set_t *)set;
2685 
2686 	if (kset->prs_index < 0)
2687 		return (NULL);
2688 	return ((pool_elem_t *)kset->pkr_list[kset->prs_index--]);
2689 }
2690 
2691 /*
2692  * Sets the current index in a result set.
2693  * Returns PO_SUCCESS/PO_FAIL
2694  */
2695 int
2696 pool_knl_rs_set_index(pool_result_set_t *set, int index)
2697 {
2698 	pool_knl_result_set_t *kset = (pool_knl_result_set_t *)set;
2699 
2700 	if (index < 0 || index >= kset->pkr_count) {
2701 		pool_seterror(POE_BADPARAM);
2702 		return (PO_FAIL);
2703 	}
2704 	kset->prs_index = index;
2705 	return (PO_SUCCESS);
2706 }
2707 
2708 /*
2709  * Return the current index in a result set.
2710  * Returns current index
2711  */
2712 int
2713 pool_knl_rs_get_index(pool_result_set_t *set)
2714 {
2715 	pool_knl_result_set_t *kset = (pool_knl_result_set_t *)set;
2716 
2717 	return (kset->prs_index);
2718 }
2719 
2720 /*
2721  * Return the first element in a result set.
2722  * Returns pool_elem_t pointer/NULL
2723  */
2724 pool_elem_t *
2725 pool_knl_rs_first(pool_result_set_t *set)
2726 {
2727 	pool_knl_result_set_t *kset = (pool_knl_result_set_t *)set;
2728 
2729 	return ((pool_elem_t *)kset->pkr_list[0]);
2730 }
2731 
2732 /*
2733  * Return the last element in a result set.
2734  * Returns pool_elem_t pointer/NULL
2735  */
2736 pool_elem_t *
2737 pool_knl_rs_last(pool_result_set_t *set)
2738 {
2739 	pool_knl_result_set_t *kset = (pool_knl_result_set_t *)set;
2740 
2741 	return ((pool_elem_t *)kset->pkr_list[kset->pkr_count - 1]);
2742 }
2743 
2744 /*
2745  * Return the number of results in a result set.
2746  * Returns result count
2747  */
2748 int
2749 pool_knl_rs_count(pool_result_set_t *set)
2750 {
2751 	pool_knl_result_set_t *kset = (pool_knl_result_set_t *)set;
2752 
2753 	return (kset->pkr_count);
2754 }
2755 
2756 
2757 /*
2758  * Close a result set. Free the resources
2759  * Returns PO_SUCCESS/PO_FAIL
2760  */
2761 int
2762 pool_knl_rs_close(pool_result_set_t *set)
2763 {
2764 	pool_knl_result_set_t *kset = (pool_knl_result_set_t *)set;
2765 
2766 	pool_knl_result_set_free(kset);
2767 	return (PO_SUCCESS);
2768 }
2769 
2770 /*
2771  * Commit an individual transaction log item(). This processing is
2772  * essential to the pool_conf_commit() logic. When pool_conf_commit()
2773  * is invoked, the pending transaction log for the configuration is
2774  * walked and all pending changes to the kernel are invoked. If a
2775  * change succeeds it is marked in the log as successful and
2776  * processing continues, if it fails then failure is returned and the
2777  * log will be "rolled back" to undo changes to the library snapshot
2778  * and the kernel.
2779  */
2780 int
2781 log_item_commit(log_item_t *li)
2782 {
2783 	pool_knl_connection_t *prov =
2784 	    (pool_knl_connection_t *)li->li_log->l_conf->pc_prov;
2785 	pool_create_undo_t *create;
2786 	pool_destroy_undo_t *destroy;
2787 	pool_assoc_undo_t *assoc;
2788 	pool_dissoc_undo_t *dissoc;
2789 	pool_propput_undo_t *propput;
2790 	pool_proprm_undo_t *proprm;
2791 	pool_xtransfer_undo_t *xtransfer;
2792 	char_buf_t *cb;
2793 	size_t size;
2794 	pool_elem_t *pair;
2795 	pool_value_t val = POOL_VALUE_INITIALIZER;
2796 	int ret;
2797 
2798 	switch (li->li_op) {
2799 	case POOL_CREATE:
2800 		create = (pool_create_undo_t *)li->li_details;
2801 		if ((cb = alloc_char_buf(CB_DEFAULT_LEN)) == NULL)
2802 			return (PO_FAIL);
2803 		if (set_char_buf(cb, "%s.sys_id",
2804 		    pool_elem_class_string(create->pcu_elem)) != PO_SUCCESS) {
2805 			free_char_buf(cb);
2806 			return (PO_FAIL);
2807 		}
2808 #ifdef DEBUG
2809 		dprintf("log_item_commit: POOL_CREATE, remove from dict\n");
2810 		pool_elem_dprintf(create->pcu_elem);
2811 #endif	/* DEBUG */
2812 		/*
2813 		 * May not need to remove the element if it was
2814 		 * already destroyed before commit. Just cast the
2815 		 * return to void.
2816 		 */
2817 		(void) dict_remove(prov->pkc_elements,
2818 		    (pool_knl_elem_t *)create->pcu_elem);
2819 
2820 		if (ioctl(prov->pkc_fd, POOL_CREATE, &create->pcu_ioctl) < 0) {
2821 			pool_seterror(POE_SYSTEM);
2822 			return (PO_FAIL);
2823 		}
2824 		/*
2825 		 * Now that we have created our element in the kernel,
2826 		 * it has a valid allocated system id. Remove the
2827 		 * element from the element dictionary, using the
2828 		 * current key, and then re-insert under the new key.
2829 		 */
2830 #ifdef DEBUG
2831 		pool_elem_dprintf(create->pcu_elem);
2832 #endif	/* DEBUG */
2833 		assert(nvlist_add_int64(
2834 		    ((pool_knl_elem_t *)create->pcu_elem)->pke_properties,
2835 		    cb->cb_buf, create->pcu_ioctl.pc_i_id) == 0);
2836 		free_char_buf(cb);
2837 		assert(dict_put(prov->pkc_elements, create->pcu_elem,
2838 		    create->pcu_elem) == NULL);
2839 		/*
2840 		 * If the element has a pair in the static
2841 		 * configuration, update it with the sys_id
2842 		 */
2843 		if ((pair = pool_get_pair(create->pcu_elem)) != NULL) {
2844 			pool_value_set_int64(&val, create->pcu_ioctl.pc_i_id);
2845 			assert(pool_put_any_ns_property(pair, c_sys_prop, &val)
2846 			    == PO_SUCCESS);
2847 		}
2848 		li->li_state = LS_UNDO;
2849 		break;
2850 	case POOL_DESTROY:
2851 		destroy = (pool_destroy_undo_t *)li->li_details;
2852 
2853 		destroy->pdu_ioctl.pd_o_id = elem_get_sysid(destroy->pdu_elem);
2854 
2855 		/*
2856 		 * It may be that this element was created in the last
2857 		 * transaction. In which case POOL_CREATE, above, will
2858 		 * have re-inserted the element in the dictionary. Try
2859 		 * to remove it just in case this has occurred.
2860 		 */
2861 		(void) dict_remove(prov->pkc_elements,
2862 		    (pool_knl_elem_t *)destroy->pdu_elem);
2863 		while ((ret = ioctl(prov->pkc_fd, POOL_DESTROY,
2864 		    &destroy->pdu_ioctl)) < 0 && errno == EAGAIN)
2865 			;
2866 		if (ret < 0) {
2867 			pool_seterror(POE_SYSTEM);
2868 			return (PO_FAIL);
2869 		}
2870 #ifdef DEBUG
2871 		dprintf("log_item_commit: POOL_DESTROY\n");
2872 		pool_elem_dprintf(destroy->pdu_elem);
2873 #endif	/* DEBUG */
2874 		li->li_state = LS_UNDO;
2875 		break;
2876 	case POOL_ASSOC:
2877 		assoc = (pool_assoc_undo_t *)li->li_details;
2878 
2879 		assoc->pau_ioctl.pa_o_pool_id =
2880 		    elem_get_sysid(assoc->pau_assoc);
2881 		assoc->pau_ioctl.pa_o_res_id =
2882 		    elem_get_sysid(assoc->pau_newres);
2883 		while ((ret = ioctl(prov->pkc_fd, POOL_ASSOC,
2884 		    &assoc->pau_ioctl)) < 0 && errno == EAGAIN)
2885 			;
2886 		if (ret < 0) {
2887 			pool_seterror(POE_SYSTEM);
2888 			return (PO_FAIL);
2889 		}
2890 		li->li_state = LS_UNDO;
2891 		break;
2892 	case POOL_DISSOC:
2893 		dissoc = (pool_dissoc_undo_t *)li->li_details;
2894 
2895 		dissoc->pdu_ioctl.pd_o_pool_id =
2896 		    elem_get_sysid(dissoc->pdu_dissoc);
2897 
2898 		while ((ret = ioctl(prov->pkc_fd, POOL_DISSOC,
2899 		    &dissoc->pdu_ioctl)) < 0 && errno == EAGAIN)
2900 			;
2901 		if (ret < 0) {
2902 			pool_seterror(POE_SYSTEM);
2903 			return (PO_FAIL);
2904 		}
2905 		li->li_state = LS_UNDO;
2906 		break;
2907 	case POOL_TRANSFER:
2908 		li->li_state = LS_UNDO;
2909 		pool_seterror(POE_BADPARAM);
2910 		return (PO_FAIL);
2911 	case POOL_XTRANSFER:
2912 		xtransfer = (pool_xtransfer_undo_t *)li->li_details;
2913 
2914 		xtransfer->pxu_ioctl.px_o_src_id =
2915 		    elem_get_sysid(xtransfer->pxu_src);
2916 		xtransfer->pxu_ioctl.px_o_tgt_id =
2917 		    elem_get_sysid(xtransfer->pxu_tgt);
2918 		for (size = 0; xtransfer->pxu_rl[size] != NULL; size ++) {
2919 			xtransfer->pxu_ioctl.px_o_comp_list[size] =
2920 			    elem_get_sysid(TO_ELEM(xtransfer->pxu_rl[size]));
2921 #ifdef DEBUG
2922 			dprintf("log_item_commit: POOL_XTRANSFER\n");
2923 			pool_elem_dprintf(TO_ELEM(xtransfer->pxu_rl[size]));
2924 #endif	/* DEBUG */
2925 		}
2926 
2927 		/*
2928 		 * Don't actually transfer resources if the configuration
2929 		 * is in POF_DESTROY state. This is to prevent problems
2930 		 * relating to transferring off-line CPUs. Instead rely
2931 		 * on the POOL_DESTROY ioctl to transfer the CPUS.
2932 		 */
2933 		if (li->li_log->l_conf->pc_state != POF_DESTROY &&
2934 		    ioctl(prov->pkc_fd, POOL_XTRANSFER,
2935 		    &xtransfer->pxu_ioctl) < 0) {
2936 #ifdef DEBUG
2937 			dprintf("log_item_commit: POOL_XTRANSFER, ioctl "
2938 			    "failed\n");
2939 #endif	/* DEBUG */
2940 			pool_seterror(POE_SYSTEM);
2941 			return (PO_FAIL);
2942 		}
2943 		li->li_state = LS_UNDO;
2944 		break;
2945 	case POOL_PROPPUT:
2946 		propput = (pool_propput_undo_t *)li->li_details;
2947 
2948 		if (pool_elem_class(propput->ppu_elem) != PEC_SYSTEM) {
2949 			propput->ppu_ioctl.pp_o_id =
2950 			    elem_get_sysid(propput->ppu_elem);
2951 		}
2952 		/*
2953 		 * Some properties, e.g. pset.size, are read-only in the
2954 		 * kernel and attempting to change them will fail and cause
2955 		 * problems. Although this property is read-only through the
2956 		 * public interface, the library needs to modify it's value.
2957 		 */
2958 		if ((propput->ppu_doioctl & KERNEL_PROP_RDONLY) == 0) {
2959 			if (ioctl(prov->pkc_fd, POOL_PROPPUT,
2960 			    &propput->ppu_ioctl) < 0) {
2961 				pool_seterror(POE_SYSTEM);
2962 				return (PO_FAIL);
2963 			}
2964 		}
2965 		li->li_state = LS_UNDO;
2966 		break;
2967 	case POOL_PROPRM:
2968 		proprm = (pool_proprm_undo_t *)li->li_details;
2969 
2970 		if (pool_elem_class(proprm->pru_elem) != PEC_SYSTEM) {
2971 			proprm->pru_ioctl.pp_o_id =
2972 			    elem_get_sysid(proprm->pru_elem);
2973 		}
2974 		if (ioctl(prov->pkc_fd, POOL_PROPRM, &proprm->pru_ioctl) < 0) {
2975 			pool_seterror(POE_SYSTEM);
2976 			return (PO_FAIL);
2977 		}
2978 		li->li_state = LS_UNDO;
2979 		break;
2980 	default:
2981 		return (PO_FAIL);
2982 	}
2983 	return (PO_SUCCESS);
2984 }
2985 
2986 /*
2987  * Undo an individual transaction log item(). This processing is
2988  * essential to the pool_conf_commit() and pool_conf_rollback()
2989  * logic. Changes to the libpool snapshot and the kernel are carried
2990  * out separately. The library snapshot is updated synchronously,
2991  * however the kernel update is delayed until the user calls
2992  * pool_conf_commit().
2993  *
2994  * When undoing transactions, library changes will be undone unless
2995  * this invocation is as a result of a commit failure, in which case
2996  * the log state will be LS_RECOVER. Kernel changes will only be
2997  * undone if they are marked as having been done, in which case the
2998  * log item state will be LS_UNDO.
2999  */
3000 int
3001 log_item_undo(log_item_t *li)
3002 {
3003 	pool_knl_connection_t *prov =
3004 	    (pool_knl_connection_t *)li->li_log->l_conf->pc_prov;
3005 	pool_create_undo_t *create;
3006 	pool_destroy_undo_t *destroy;
3007 	pool_assoc_undo_t *assoc;
3008 	pool_dissoc_undo_t *dissoc;
3009 	pool_propput_undo_t *propput;
3010 	pool_proprm_undo_t *proprm;
3011 	pool_xtransfer_undo_t *xtransfer;
3012 	char_buf_t *cb;
3013 	size_t size;
3014 	pool_destroy_t u_destroy;
3015 	pool_create_t u_create;
3016 	pool_assoc_t u_assoc;
3017 	pool_xtransfer_t u_xtransfer;
3018 	pool_propput_t u_propput;
3019 	pool_proprm_t u_proprm;
3020 	pool_conf_t *conf = li->li_log->l_conf;
3021 	nvpair_t *pair;
3022 	nvlist_t *tmplist;
3023 	int ret;
3024 
3025 	if (li->li_log->l_state != LS_RECOVER) {
3026 	switch (li->li_op) {
3027 	case POOL_CREATE:
3028 		create = (pool_create_undo_t *)li->li_details;
3029 
3030 		(void) dict_remove(prov->pkc_elements, create->pcu_elem);
3031 #ifdef DEBUG
3032 		dprintf("log_item_undo: POOL_CREATE\n");
3033 		assert(create->pcu_elem != NULL);
3034 		dprintf("log_item_undo: POOL_CREATE %p\n", create->pcu_elem);
3035 		pool_elem_dprintf(create->pcu_elem);
3036 #endif	/* DEBUG */
3037 		pool_knl_elem_free((pool_knl_elem_t *)create->pcu_elem,
3038 		    PO_TRUE);
3039 		break;
3040 	case POOL_DESTROY:
3041 		destroy = (pool_destroy_undo_t *)li->li_details;
3042 
3043 		assert(dict_put(prov->pkc_elements, destroy->pdu_elem,
3044 		    destroy->pdu_elem) == NULL);
3045 		break;
3046 	case POOL_ASSOC:
3047 		assoc = (pool_assoc_undo_t *)li->li_details;
3048 
3049 		if (assoc->pau_oldres != NULL)
3050 			((pool_knl_pool_t *)assoc->pau_assoc)->pkp_assoc
3051 			    [pool_resource_elem_class(assoc->pau_oldres)] =
3052 			    (pool_knl_resource_t *)assoc->pau_oldres;
3053 		break;
3054 	case POOL_DISSOC:
3055 		dissoc = (pool_dissoc_undo_t *)li->li_details;
3056 
3057 		if (dissoc->pdu_oldres != NULL)
3058 			((pool_knl_pool_t *)dissoc->pdu_dissoc)->pkp_assoc
3059 			    [pool_resource_elem_class(dissoc->pdu_oldres)] =
3060 			    (pool_knl_resource_t *)dissoc->pdu_oldres;
3061 		break;
3062 	case POOL_TRANSFER:
3063 		pool_seterror(POE_BADPARAM);
3064 		return (PO_FAIL);
3065 	case POOL_XTRANSFER:
3066 		xtransfer = (pool_xtransfer_undo_t *)li->li_details;
3067 
3068 		for (size = 0; xtransfer->pxu_rl[size] != NULL; size++) {
3069 			pool_value_t val = POOL_VALUE_INITIALIZER;
3070 			uint64_t src_size;
3071 			uint64_t tgt_size;
3072 
3073 			if (pool_set_container(xtransfer->pxu_src,
3074 			    TO_ELEM(xtransfer->pxu_rl[size])) == PO_FAIL) {
3075 				return (PO_FAIL);
3076 			}
3077 			/*
3078 			 * Maintain the library view of the size
3079 			 */
3080 			if (resource_get_size(pool_elem_res(xtransfer->pxu_src),
3081 			    &src_size) != PO_SUCCESS ||
3082 			    resource_get_size(pool_elem_res(xtransfer->pxu_tgt),
3083 			    &tgt_size) != PO_SUCCESS) {
3084 				pool_seterror(POE_BADPARAM);
3085 				return (PO_FAIL);
3086 			}
3087 			src_size++;
3088 			tgt_size--;
3089 			pool_value_set_uint64(&val, src_size);
3090 			(void) pool_put_any_ns_property(xtransfer->pxu_src,
3091 			    c_size_prop, &val);
3092 			pool_value_set_uint64(&val, tgt_size);
3093 			(void) pool_put_any_ns_property(xtransfer->pxu_tgt,
3094 			    c_size_prop, &val);
3095 		}
3096 		break;
3097 	case POOL_PROPPUT:
3098 		propput = (pool_propput_undo_t *)li->li_details;
3099 
3100 		if ((propput->ppu_doioctl & KERNEL_PROP_RDONLY) == 0) {
3101 			if (propput->ppu_blist != NULL) {
3102 				if (nvlist_merge(
3103 				    ((pool_knl_elem_t *)propput->ppu_elem)->
3104 				    pke_properties, propput->ppu_blist, 0)
3105 				    != 0) {
3106 					pool_seterror(POE_SYSTEM);
3107 					return (PO_FAIL);
3108 				}
3109 			} else {
3110 				if (nvlist_unpack(propput->ppu_ioctl.pp_o_buf,
3111 				    propput->ppu_ioctl.pp_o_bufsize,
3112 				    &propput->ppu_alist, 0) != 0) {
3113 					pool_seterror(POE_SYSTEM);
3114 					return (PO_FAIL);
3115 				}
3116 				pair = nvlist_next_nvpair(propput->ppu_alist,
3117 				    NULL);
3118 				(void) nvlist_remove_all(((pool_knl_elem_t *)
3119 				    propput->ppu_elem)->pke_properties,
3120 				    nvpair_name(pair));
3121 				nvlist_free(propput->ppu_alist);
3122 			}
3123 		}
3124 		break;
3125 	case POOL_PROPRM:
3126 		proprm = (pool_proprm_undo_t *)li->li_details;
3127 
3128 		if (pool_value_get_type(&proprm->pru_oldval) != POC_INVAL) {
3129 			if (pool_put_property(conf, proprm->pru_elem,
3130 			    proprm->pru_ioctl.pp_o_prop_name,
3131 			    &proprm->pru_oldval) != PO_SUCCESS) {
3132 				return (PO_FAIL);
3133 			}
3134 		}
3135 		break;
3136 	default:
3137 		return (PO_FAIL);
3138 	}
3139 	}
3140 	/*
3141 	 * Only try to undo the state of the kernel if we modified it.
3142 	 */
3143 	if (li->li_state == LS_DO) {
3144 		return (PO_SUCCESS);
3145 	}
3146 
3147 	switch (li->li_op) {
3148 	case POOL_CREATE:
3149 		create = (pool_create_undo_t *)li->li_details;
3150 
3151 		u_destroy.pd_o_type = create->pcu_ioctl.pc_o_type;
3152 		u_destroy.pd_o_sub_type = create->pcu_ioctl.pc_o_sub_type;
3153 		u_destroy.pd_o_id = create->pcu_ioctl.pc_i_id;
3154 
3155 		while ((ret = ioctl(prov->pkc_fd, POOL_DESTROY,
3156 		    &u_destroy)) < 0 && errno == EAGAIN)
3157 			;
3158 		if (ret < 0) {
3159 			pool_seterror(POE_SYSTEM);
3160 			return (PO_FAIL);
3161 		}
3162 		li->li_state = LS_DO;
3163 		break;
3164 	case POOL_DESTROY:
3165 		destroy = (pool_destroy_undo_t *)li->li_details;
3166 
3167 		u_create.pc_o_type = destroy->pdu_ioctl.pd_o_type;
3168 		u_create.pc_o_sub_type = destroy->pdu_ioctl.pd_o_sub_type;
3169 
3170 		if (ioctl(prov->pkc_fd, POOL_CREATE, &u_create) < 0) {
3171 			pool_seterror(POE_SYSTEM);
3172 			return (PO_FAIL);
3173 		}
3174 
3175 		if ((cb = alloc_char_buf(CB_DEFAULT_LEN)) == NULL) {
3176 			return (PO_FAIL);
3177 		}
3178 		if (set_char_buf(cb, "%s.sys_id",
3179 		    pool_elem_class_string(destroy->pdu_elem)) != PO_SUCCESS) {
3180 			free_char_buf(cb);
3181 			return (PO_FAIL);
3182 		}
3183 		(void) nvlist_add_int64(
3184 		    ((pool_knl_elem_t *)destroy->pdu_elem)->pke_properties,
3185 		    cb->cb_buf, u_create.pc_i_id);
3186 		free_char_buf(cb);
3187 		if (dict_put(prov->pkc_elements, destroy->pdu_elem,
3188 		    destroy->pdu_elem) != NULL) {
3189 			pool_seterror(POE_SYSTEM);
3190 			return (PO_FAIL);
3191 		}
3192 		/*
3193 		 * Now we need to reset all the properties and
3194 		 * associations in the kernel for this newly created
3195 		 * replacement.
3196 		 */
3197 		u_propput.pp_o_id_type = destroy->pdu_ioctl.pd_o_type;
3198 		u_propput.pp_o_id_sub_type = destroy->pdu_ioctl.pd_o_sub_type;
3199 		u_propput.pp_o_id = u_create.pc_i_id;
3200 		u_propput.pp_o_buf = NULL;
3201 		/*
3202 		 * Remove the read-only properties before attempting
3203 		 * to restore the state of the newly created property
3204 		 */
3205 		(void) nvlist_dup(((pool_knl_elem_t *)destroy->pdu_elem)->
3206 		    pke_properties, &tmplist, 0);
3207 		for (pair = nvlist_next_nvpair(tmplist, NULL); pair != NULL;
3208 		    pair = nvlist_next_nvpair(tmplist, pair)) {
3209 			const pool_prop_t *prop;
3210 			char *name = nvpair_name(pair);
3211 			if ((prop = provider_get_prop(destroy->pdu_elem,
3212 			    name)) != NULL)
3213 				if (prop_is_readonly(prop) == PO_TRUE)
3214 					(void) nvlist_remove_all(tmplist, name);
3215 		}
3216 		if (nvlist_pack(tmplist, (char **)&u_propput.pp_o_buf,
3217 		    &u_propput.pp_o_bufsize, NV_ENCODE_NATIVE, 0) != 0) {
3218 			pool_seterror(POE_SYSTEM);
3219 			return (PO_FAIL);
3220 		}
3221 		nvlist_free(tmplist);
3222 		if (ioctl(prov->pkc_fd, POOL_PROPPUT, &u_propput) < 0) {
3223 			free(u_propput.pp_o_buf);
3224 			pool_seterror(POE_SYSTEM);
3225 			return (PO_FAIL);
3226 		}
3227 		free(u_propput.pp_o_buf);
3228 		/*
3229 		 * Now reset the associations for all the resource
3230 		 * types if the thing which we are recreating is a
3231 		 * pool
3232 		 *
3233 		 * TODO: This is resource specific and must be
3234 		 * extended for additional resource types.
3235 		 */
3236 		if (destroy->pdu_ioctl.pd_o_type == PEC_POOL) {
3237 			u_assoc.pa_o_pool_id = u_create.pc_i_id;
3238 			u_assoc.pa_o_res_id =
3239 			    elem_get_sysid(
3240 			    TO_ELEM(((pool_knl_pool_t *)destroy->pdu_elem)->
3241 			    pkp_assoc[PREC_PSET]));
3242 			u_assoc.pa_o_id_type = PREC_PSET;
3243 
3244 			if (ioctl(prov->pkc_fd, POOL_ASSOC, &u_assoc) < 0) {
3245 				pool_seterror(POE_SYSTEM);
3246 				return (PO_FAIL);
3247 			}
3248 		}
3249 		li->li_state = LS_DO;
3250 		break;
3251 	case POOL_ASSOC:
3252 		assoc = (pool_assoc_undo_t *)li->li_details;
3253 
3254 		u_assoc.pa_o_pool_id = elem_get_sysid(assoc->pau_assoc);
3255 		u_assoc.pa_o_res_id = elem_get_sysid(assoc->pau_oldres);
3256 		u_assoc.pa_o_id_type = assoc->pau_ioctl.pa_o_id_type;
3257 
3258 		while ((ret = ioctl(prov->pkc_fd, POOL_ASSOC, &u_assoc)) < 0 &&
3259 		    errno == EAGAIN)
3260 			;
3261 		if (ret < 0) {
3262 			pool_seterror(POE_SYSTEM);
3263 			return (PO_FAIL);
3264 		}
3265 		li->li_state = LS_DO;
3266 		break;
3267 	case POOL_DISSOC:
3268 		dissoc = (pool_dissoc_undo_t *)li->li_details;
3269 
3270 		u_assoc.pa_o_pool_id = elem_get_sysid(dissoc->pdu_dissoc);
3271 		u_assoc.pa_o_res_id = elem_get_sysid(dissoc->pdu_oldres);
3272 		u_assoc.pa_o_id_type = dissoc->pdu_ioctl.pd_o_id_type;
3273 
3274 		while ((ret = ioctl(prov->pkc_fd, POOL_ASSOC, &u_assoc)) < 0 &&
3275 		    errno == EAGAIN)
3276 			;
3277 		if (ret < 0) {
3278 			pool_seterror(POE_SYSTEM);
3279 			return (PO_FAIL);
3280 		}
3281 		li->li_state = LS_DO;
3282 		break;
3283 	case POOL_TRANSFER:
3284 		li->li_state = LS_DO;
3285 		pool_seterror(POE_BADPARAM);
3286 		return (PO_FAIL);
3287 	case POOL_XTRANSFER:
3288 		xtransfer = (pool_xtransfer_undo_t *)li->li_details;
3289 
3290 		(void) memcpy(&u_xtransfer, &xtransfer->pxu_ioctl,
3291 		    sizeof (pool_xtransfer_t));
3292 		u_xtransfer.px_o_src_id = elem_get_sysid(xtransfer->pxu_tgt);
3293 		u_xtransfer.px_o_tgt_id = elem_get_sysid(xtransfer->pxu_src);
3294 
3295 		if (ioctl(prov->pkc_fd, POOL_XTRANSFER, &u_xtransfer) < 0) {
3296 			pool_seterror(POE_SYSTEM);
3297 			return (PO_FAIL);
3298 		}
3299 		li->li_state = LS_DO;
3300 		break;
3301 	case POOL_PROPPUT:
3302 		propput = (pool_propput_undo_t *)li->li_details;
3303 
3304 		if ((propput->ppu_doioctl & KERNEL_PROP_RDONLY) == 0) {
3305 			if (propput->ppu_blist) {
3306 				(void) memcpy(&u_propput, &propput->ppu_ioctl,
3307 				    sizeof (pool_propput_t));
3308 				u_propput.pp_o_id =
3309 				    elem_get_sysid(propput->ppu_elem);
3310 				u_propput.pp_o_buf = NULL;
3311 				if (nvlist_pack(propput->ppu_blist,
3312 				    (char **)&u_propput.pp_o_buf,
3313 				    &u_propput.pp_o_bufsize,
3314 				    NV_ENCODE_NATIVE, 0) != 0) {
3315 					pool_seterror(POE_SYSTEM);
3316 					return (PO_FAIL);
3317 				}
3318 				if (ioctl(prov->pkc_fd, POOL_PROPPUT,
3319 				    &u_propput) < 0) {
3320 					free(u_propput.pp_o_buf);
3321 					pool_seterror(POE_SYSTEM);
3322 					return (PO_FAIL);
3323 				}
3324 				free(u_propput.pp_o_buf);
3325 			} else {
3326 				if (nvlist_unpack(propput->
3327 				    ppu_ioctl.pp_o_buf,
3328 				    propput->ppu_ioctl.pp_o_bufsize,
3329 				    &propput->ppu_alist, 0) != 0) {
3330 					pool_seterror(POE_SYSTEM);
3331 					return (PO_FAIL);
3332 				}
3333 				u_proprm.pp_o_id_type =
3334 				    propput->ppu_ioctl.pp_o_id_type;
3335 				u_proprm.pp_o_id_sub_type =
3336 				    propput->ppu_ioctl.pp_o_id_sub_type;
3337 				u_proprm.pp_o_id =
3338 				    elem_get_sysid(propput->ppu_elem);
3339 				pair = nvlist_next_nvpair(propput->ppu_alist,
3340 				    NULL);
3341 				u_proprm.pp_o_prop_name = nvpair_name(pair);
3342 				u_proprm.pp_o_prop_name_size =
3343 				    strlen(u_proprm.pp_o_prop_name);
3344 
3345 				if (provider_get_prop(propput->ppu_elem,
3346 				    u_proprm.pp_o_prop_name) == NULL) {
3347 					if (ioctl(prov->pkc_fd, POOL_PROPRM,
3348 					    &u_proprm) < 0) {
3349 						nvlist_free(propput->ppu_alist);
3350 						pool_seterror(POE_SYSTEM);
3351 						return (PO_FAIL);
3352 					}
3353 				}
3354 				nvlist_free(propput->ppu_alist);
3355 			}
3356 		}
3357 		li->li_state = LS_DO;
3358 		break;
3359 	case POOL_PROPRM:
3360 		proprm = (pool_proprm_undo_t *)li->li_details;
3361 
3362 		u_propput.pp_o_id_type = proprm->pru_ioctl.pp_o_id_type;
3363 		u_propput.pp_o_id_sub_type =
3364 		    proprm->pru_ioctl.pp_o_id_sub_type;
3365 		u_propput.pp_o_id = elem_get_sysid(proprm->pru_elem);
3366 		u_propput.pp_o_buf = NULL;
3367 		/*
3368 		 * Only try to remove the appropriate property
3369 		 */
3370 		if (nvlist_alloc(&tmplist, NV_UNIQUE_NAME_TYPE, 0) !=
3371 		    0) {
3372 			pool_seterror(POE_SYSTEM);
3373 			return (PO_FAIL);
3374 		}
3375 		if (pool_knl_nvlist_add_value(tmplist,
3376 		    pool_value_get_name(&proprm->pru_oldval),
3377 		    &proprm->pru_oldval) != PO_SUCCESS)
3378 			return (PO_FAIL);
3379 
3380 		if (nvlist_pack(tmplist,
3381 		    (char **)&u_propput.pp_o_buf, &u_propput.pp_o_bufsize,
3382 		    NV_ENCODE_NATIVE, 0) != 0) {
3383 			nvlist_free(tmplist);
3384 			pool_seterror(POE_SYSTEM);
3385 			return (PO_FAIL);
3386 		}
3387 		nvlist_free(tmplist);
3388 		if (ioctl(prov->pkc_fd, POOL_PROPPUT, &u_propput) < 0) {
3389 			free(u_propput.pp_o_buf);
3390 			pool_seterror(POE_SYSTEM);
3391 			return (PO_FAIL);
3392 		}
3393 		free(u_propput.pp_o_buf);
3394 		li->li_state = LS_DO;
3395 		break;
3396 	default:
3397 		return (PO_FAIL);
3398 	}
3399 		return (PO_SUCCESS);
3400 }
3401 
3402 /*
3403  * A log item stores state about the transaction it represents. This
3404  * function releases the resources associated with the transaction and
3405  * used to store the transaction state.
3406  */
3407 int
3408 log_item_release(log_item_t *li)
3409 {
3410 	pool_create_undo_t *create;
3411 	pool_destroy_undo_t *destroy;
3412 	pool_assoc_undo_t *assoc;
3413 	pool_dissoc_undo_t *dissoc;
3414 	pool_propput_undo_t *propput;
3415 	pool_proprm_undo_t *proprm;
3416 	pool_xtransfer_undo_t *xtransfer;
3417 
3418 	switch (li->li_op) {
3419 	case POOL_CREATE:
3420 		create = (pool_create_undo_t *)li->li_details;
3421 
3422 		free(create);
3423 		break;
3424 	case POOL_DESTROY:
3425 		destroy = (pool_destroy_undo_t *)li->li_details;
3426 
3427 #ifdef DEBUG
3428 		dprintf("log_item_release: POOL_DESTROY\n");
3429 #endif	/* DEBUG */
3430 
3431 		if (li->li_state == LS_UNDO) {
3432 #ifdef DEBUG
3433 			pool_elem_dprintf(destroy->pdu_elem);
3434 #endif	/* DEBUG */
3435 			pool_knl_elem_free((pool_knl_elem_t *)destroy->
3436 			    pdu_elem, PO_TRUE);
3437 		}
3438 		free(destroy);
3439 		break;
3440 	case POOL_ASSOC:
3441 		assoc = (pool_assoc_undo_t *)li->li_details;
3442 
3443 		free(assoc);
3444 		break;
3445 	case POOL_DISSOC:
3446 		dissoc = (pool_dissoc_undo_t *)li->li_details;
3447 
3448 		free(dissoc);
3449 		break;
3450 	case POOL_TRANSFER:
3451 		pool_seterror(POE_BADPARAM);
3452 		return (PO_FAIL);
3453 	case POOL_XTRANSFER:
3454 		xtransfer = (pool_xtransfer_undo_t *)li->li_details;
3455 
3456 		free(xtransfer->pxu_rl);
3457 		free(xtransfer->pxu_ioctl.px_o_comp_list);
3458 		free(xtransfer);
3459 		break;
3460 	case POOL_PROPPUT:
3461 		propput = (pool_propput_undo_t *)li->li_details;
3462 
3463 		nvlist_free(propput->ppu_blist);
3464 		free(propput->ppu_ioctl.pp_o_buf);
3465 		free(propput);
3466 		break;
3467 	case POOL_PROPRM:
3468 		proprm = (pool_proprm_undo_t *)li->li_details;
3469 
3470 		free(proprm);
3471 		break;
3472 	default:
3473 		return (PO_FAIL);
3474 	}
3475 	return (PO_SUCCESS);
3476 }
3477 
3478 /*
3479  * pool_knl_nvlist_add_value() adds a pool_value_t to an nvlist.
3480  */
3481 int
3482 pool_knl_nvlist_add_value(nvlist_t *list, const char *name,
3483     const pool_value_t *pv)
3484 {
3485 	uint64_t uval;
3486 	int64_t ival;
3487 	double dval;
3488 	uchar_t dval_b[sizeof (double)];
3489 	uchar_t bval;
3490 	const char *sval;
3491 	pool_value_class_t type;
3492 	char *nv_name;
3493 
3494 	if ((type = pool_value_get_type(pv)) == POC_INVAL) {
3495 		pool_seterror(POE_BADPARAM);
3496 		return (PO_FAIL);
3497 	}
3498 	nv_name = (char *)name;
3499 
3500 	switch (type) {
3501 	case POC_UINT:
3502 		if (pool_value_get_uint64(pv, &uval) == POC_INVAL) {
3503 			return (PO_FAIL);
3504 		}
3505 		if (nvlist_add_uint64(list, nv_name, uval) != 0) {
3506 			pool_seterror(POE_SYSTEM);
3507 			return (PO_FAIL);
3508 		}
3509 		break;
3510 	case POC_INT:
3511 		if (pool_value_get_int64(pv, &ival) == POC_INVAL) {
3512 			return (PO_FAIL);
3513 		}
3514 		if (nvlist_add_int64(list, nv_name, ival) != 0) {
3515 			pool_seterror(POE_SYSTEM);
3516 			return (PO_FAIL);
3517 		}
3518 		break;
3519 	case POC_DOUBLE:
3520 		if (pool_value_get_double(pv, &dval) == POC_INVAL) {
3521 			return (PO_FAIL);
3522 		}
3523 		/*
3524 		 * Since there is no support for doubles in the
3525 		 * kernel, store the double value in a byte array.
3526 		 */
3527 		(void) memcpy(dval_b, &dval, sizeof (double));
3528 		if (nvlist_add_byte_array(list, nv_name, dval_b,
3529 		    sizeof (double)) != 0) {
3530 			pool_seterror(POE_SYSTEM);
3531 			return (PO_FAIL);
3532 		}
3533 		break;
3534 	case POC_BOOL:
3535 		if (pool_value_get_bool(pv, &bval) == POC_INVAL) {
3536 			return (PO_FAIL);
3537 		}
3538 		if (nvlist_add_byte(list, nv_name, bval) != 0) {
3539 			pool_seterror(POE_SYSTEM);
3540 			return (PO_FAIL);
3541 		}
3542 		break;
3543 	case POC_STRING:
3544 		if (pool_value_get_string(pv, &sval) == POC_INVAL) {
3545 			return (PO_FAIL);
3546 		}
3547 		if (nvlist_add_string(list, nv_name, (char *)sval) != 0) {
3548 			pool_seterror(POE_SYSTEM);
3549 			return (PO_FAIL);
3550 		}
3551 		break;
3552 	default:
3553 		pool_seterror(POE_BADPARAM);
3554 		return (PO_FAIL);
3555 	}
3556 	return (PO_SUCCESS);
3557 }
3558 
3559 /*
3560  * hash_id() hashes all elements in a pool configuration using the
3561  * "sys_id" property. Not all elements have a "sys_id" property,
3562  * however elem_get_sysid() caters for this by always returning a
3563  * constant value for those elements. This isn't anticipated to lead
3564  * to a performance degradation in the hash, since those elements
3565  * which are likely to be most prevalent in a configuration do have
3566  * "sys_id" as a property.
3567  */
3568 uint64_t
3569 hash_id(const pool_elem_t *pe)
3570 {
3571 	id_t id;
3572 
3573 	id = elem_get_sysid(pe);
3574 	return (hash_buf(&id, sizeof (id)));
3575 }
3576 
3577 /*
3578  *  blocking_open() guarantees access to the pool device, if open()
3579  * is failing with EBUSY.
3580  */
3581 int
3582 blocking_open(const char *path, int oflag)
3583 {
3584 	int fd;
3585 
3586 	while ((fd = open(path, oflag)) == -1 && errno == EBUSY)
3587 		(void) poll(NULL, 0, 1 * MILLISEC);
3588 
3589 	return (fd);
3590 }
3591