1*0Sstevel@tonic-gate /* 2*0Sstevel@tonic-gate * CDDL HEADER START 3*0Sstevel@tonic-gate * 4*0Sstevel@tonic-gate * The contents of this file are subject to the terms of the 5*0Sstevel@tonic-gate * Common Development and Distribution License, Version 1.0 only 6*0Sstevel@tonic-gate * (the "License"). You may not use this file except in compliance 7*0Sstevel@tonic-gate * with the License. 8*0Sstevel@tonic-gate * 9*0Sstevel@tonic-gate * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 10*0Sstevel@tonic-gate * or http://www.opensolaris.org/os/licensing. 11*0Sstevel@tonic-gate * See the License for the specific language governing permissions 12*0Sstevel@tonic-gate * and limitations under the License. 13*0Sstevel@tonic-gate * 14*0Sstevel@tonic-gate * When distributing Covered Code, include this CDDL HEADER in each 15*0Sstevel@tonic-gate * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 16*0Sstevel@tonic-gate * If applicable, add the following below this CDDL HEADER, with the 17*0Sstevel@tonic-gate * fields enclosed by brackets "[]" replaced with your own identifying 18*0Sstevel@tonic-gate * information: Portions Copyright [yyyy] [name of copyright owner] 19*0Sstevel@tonic-gate * 20*0Sstevel@tonic-gate * CDDL HEADER END 21*0Sstevel@tonic-gate */ 22*0Sstevel@tonic-gate #pragma ident "%Z%%M% %I% %E% SMI" 23*0Sstevel@tonic-gate 24*0Sstevel@tonic-gate /* 25*0Sstevel@tonic-gate * Copyright 2005 Sun Microsystems, Inc. All rights reserved. 26*0Sstevel@tonic-gate * Use is subject to license terms. 27*0Sstevel@tonic-gate */ 28*0Sstevel@tonic-gate 29*0Sstevel@tonic-gate #include <sys/systm.h> 30*0Sstevel@tonic-gate #include <sys/types.h> 31*0Sstevel@tonic-gate #include <sys/vnode.h> 32*0Sstevel@tonic-gate #include <sys/errno.h> 33*0Sstevel@tonic-gate #include <sys/sysmacros.h> 34*0Sstevel@tonic-gate #include <sys/debug.h> 35*0Sstevel@tonic-gate #include <sys/kmem.h> 36*0Sstevel@tonic-gate #include <sys/conf.h> 37*0Sstevel@tonic-gate #include <sys/proc.h> 38*0Sstevel@tonic-gate #include <sys/cmn_err.h> 39*0Sstevel@tonic-gate #include <sys/fs/ufs_inode.h> 40*0Sstevel@tonic-gate #include <sys/fs/ufs_filio.h> 41*0Sstevel@tonic-gate #include <sys/fs/ufs_log.h> 42*0Sstevel@tonic-gate #include <sys/inttypes.h> 43*0Sstevel@tonic-gate #include <sys/atomic.h> 44*0Sstevel@tonic-gate #include <sys/tuneable.h> 45*0Sstevel@tonic-gate 46*0Sstevel@tonic-gate /* 47*0Sstevel@tonic-gate * externs 48*0Sstevel@tonic-gate */ 49*0Sstevel@tonic-gate extern pri_t minclsyspri; 50*0Sstevel@tonic-gate extern struct kmem_cache *lufs_bp; 51*0Sstevel@tonic-gate extern int ufs_trans_push_quota(); 52*0Sstevel@tonic-gate 53*0Sstevel@tonic-gate /* 54*0Sstevel@tonic-gate * globals 55*0Sstevel@tonic-gate */ 56*0Sstevel@tonic-gate kmem_cache_t *mapentry_cache; 57*0Sstevel@tonic-gate 58*0Sstevel@tonic-gate /* 59*0Sstevel@tonic-gate * logmap tuning constants 60*0Sstevel@tonic-gate */ 61*0Sstevel@tonic-gate long logmap_maxnme_commit = 2048; 62*0Sstevel@tonic-gate long logmap_maxnme_async = 4096; 63*0Sstevel@tonic-gate long logmap_maxnme_sync = 6144; 64*0Sstevel@tonic-gate long logmap_maxcfrag_commit = 4; /* Max canceled fragments per moby */ 65*0Sstevel@tonic-gate 66*0Sstevel@tonic-gate 67*0Sstevel@tonic-gate uint64_t ufs_crb_size = 0; /* current size of all crb buffers */ 68*0Sstevel@tonic-gate uint64_t ufs_crb_max_size = 0; /* highest crb buffer use so far */ 69*0Sstevel@tonic-gate size_t ufs_crb_limit; /* max allowable size for crbs */ 70*0Sstevel@tonic-gate uint64_t ufs_crb_alloc_fails = 0; /* crb allocation failures stat */ 71*0Sstevel@tonic-gate #define UFS_MAX_CRB_DEFAULT_DIVISOR 10 /* max 1/10 kmem_maxavail() */ 72*0Sstevel@tonic-gate int ufs_max_crb_divisor = UFS_MAX_CRB_DEFAULT_DIVISOR; /* tunable */ 73*0Sstevel@tonic-gate void handle_dquot(mapentry_t *); 74*0Sstevel@tonic-gate 75*0Sstevel@tonic-gate /* 76*0Sstevel@tonic-gate * GENERIC MAP ROUTINES 77*0Sstevel@tonic-gate */ 78*0Sstevel@tonic-gate 79*0Sstevel@tonic-gate #define CRB_FREE(crb, me) \ 80*0Sstevel@tonic-gate kmem_free(crb->c_buf, crb->c_nb); \ 81*0Sstevel@tonic-gate atomic_add_64(&ufs_crb_size, -(uint64_t)crb->c_nb); \ 82*0Sstevel@tonic-gate kmem_free(crb, sizeof (crb_t)); \ 83*0Sstevel@tonic-gate (me)->me_crb = NULL; 84*0Sstevel@tonic-gate 85*0Sstevel@tonic-gate #define CRB_RELE(me) { \ 86*0Sstevel@tonic-gate crb_t *crb = (me)->me_crb; \ 87*0Sstevel@tonic-gate if (crb && (--crb->c_refcnt == 0)) { \ 88*0Sstevel@tonic-gate CRB_FREE(crb, me) \ 89*0Sstevel@tonic-gate } \ 90*0Sstevel@tonic-gate } 91*0Sstevel@tonic-gate 92*0Sstevel@tonic-gate /* 93*0Sstevel@tonic-gate * Check that the old delta has an argument and a push function of 94*0Sstevel@tonic-gate * ufs_trans_push_quota(), then check that the old and new deltas differ. 95*0Sstevel@tonic-gate * If so we clean up with handle_dquot() before replacing the old delta. 96*0Sstevel@tonic-gate */ 97*0Sstevel@tonic-gate #define HANDLE_DQUOT(me, melist) { \ 98*0Sstevel@tonic-gate if ((me->me_arg) && \ 99*0Sstevel@tonic-gate (me->me_func == ufs_trans_push_quota)) { \ 100*0Sstevel@tonic-gate if (!((me->me_dt == melist->me_dt) && \ 101*0Sstevel@tonic-gate (me->me_arg == melist->me_arg) && \ 102*0Sstevel@tonic-gate (me->me_func == melist->me_func))) { \ 103*0Sstevel@tonic-gate handle_dquot(me); \ 104*0Sstevel@tonic-gate } \ 105*0Sstevel@tonic-gate } \ 106*0Sstevel@tonic-gate } 107*0Sstevel@tonic-gate 108*0Sstevel@tonic-gate /* 109*0Sstevel@tonic-gate * free up all the mapentries for a map 110*0Sstevel@tonic-gate */ 111*0Sstevel@tonic-gate void 112*0Sstevel@tonic-gate map_free_entries(mt_map_t *mtm) 113*0Sstevel@tonic-gate { 114*0Sstevel@tonic-gate int i; 115*0Sstevel@tonic-gate mapentry_t *me; 116*0Sstevel@tonic-gate 117*0Sstevel@tonic-gate while ((me = mtm->mtm_next) != (mapentry_t *)mtm) { 118*0Sstevel@tonic-gate me->me_next->me_prev = me->me_prev; 119*0Sstevel@tonic-gate me->me_prev->me_next = me->me_next; 120*0Sstevel@tonic-gate CRB_RELE(me); 121*0Sstevel@tonic-gate kmem_cache_free(mapentry_cache, me); 122*0Sstevel@tonic-gate } 123*0Sstevel@tonic-gate for (i = 0; i < mtm->mtm_nhash; i++) 124*0Sstevel@tonic-gate mtm->mtm_hash[i] = NULL; 125*0Sstevel@tonic-gate mtm->mtm_nme = 0; 126*0Sstevel@tonic-gate mtm->mtm_nmet = 0; 127*0Sstevel@tonic-gate } 128*0Sstevel@tonic-gate 129*0Sstevel@tonic-gate /* 130*0Sstevel@tonic-gate * done with map; free if necessary 131*0Sstevel@tonic-gate */ 132*0Sstevel@tonic-gate mt_map_t * 133*0Sstevel@tonic-gate map_put(mt_map_t *mtm) 134*0Sstevel@tonic-gate { 135*0Sstevel@tonic-gate /* 136*0Sstevel@tonic-gate * free up the map's memory 137*0Sstevel@tonic-gate */ 138*0Sstevel@tonic-gate map_free_entries(mtm); 139*0Sstevel@tonic-gate ASSERT(map_put_debug(mtm)); 140*0Sstevel@tonic-gate kmem_free(mtm->mtm_hash, 141*0Sstevel@tonic-gate (size_t) (sizeof (mapentry_t *) * mtm->mtm_nhash)); 142*0Sstevel@tonic-gate mutex_destroy(&mtm->mtm_mutex); 143*0Sstevel@tonic-gate mutex_destroy(&mtm->mtm_scan_mutex); 144*0Sstevel@tonic-gate cv_destroy(&mtm->mtm_to_roll_cv); 145*0Sstevel@tonic-gate cv_destroy(&mtm->mtm_from_roll_cv); 146*0Sstevel@tonic-gate rw_destroy(&mtm->mtm_rwlock); 147*0Sstevel@tonic-gate mutex_destroy(&mtm->mtm_lock); 148*0Sstevel@tonic-gate cv_destroy(&mtm->mtm_cv_commit); 149*0Sstevel@tonic-gate cv_destroy(&mtm->mtm_cv_next); 150*0Sstevel@tonic-gate cv_destroy(&mtm->mtm_cv_eot); 151*0Sstevel@tonic-gate cv_destroy(&mtm->mtm_cv); 152*0Sstevel@tonic-gate kmem_free(mtm, sizeof (mt_map_t)); 153*0Sstevel@tonic-gate return (NULL); 154*0Sstevel@tonic-gate } 155*0Sstevel@tonic-gate /* 156*0Sstevel@tonic-gate * Allocate a map; 157*0Sstevel@tonic-gate */ 158*0Sstevel@tonic-gate mt_map_t * 159*0Sstevel@tonic-gate map_get(ml_unit_t *ul, enum maptypes maptype, int nh) 160*0Sstevel@tonic-gate { 161*0Sstevel@tonic-gate mt_map_t *mtm; 162*0Sstevel@tonic-gate 163*0Sstevel@tonic-gate /* 164*0Sstevel@tonic-gate * assume the map is not here and allocate the necessary structs 165*0Sstevel@tonic-gate */ 166*0Sstevel@tonic-gate mtm = kmem_zalloc(sizeof (mt_map_t), KM_SLEEP); 167*0Sstevel@tonic-gate mutex_init(&mtm->mtm_mutex, NULL, MUTEX_DEFAULT, NULL); 168*0Sstevel@tonic-gate mutex_init(&mtm->mtm_scan_mutex, NULL, MUTEX_DEFAULT, NULL); 169*0Sstevel@tonic-gate cv_init(&mtm->mtm_to_roll_cv, NULL, CV_DEFAULT, NULL); 170*0Sstevel@tonic-gate cv_init(&mtm->mtm_from_roll_cv, NULL, CV_DEFAULT, NULL); 171*0Sstevel@tonic-gate rw_init(&mtm->mtm_rwlock, NULL, RW_DEFAULT, NULL); 172*0Sstevel@tonic-gate mtm->mtm_next = (mapentry_t *)mtm; 173*0Sstevel@tonic-gate mtm->mtm_prev = (mapentry_t *)mtm; 174*0Sstevel@tonic-gate mtm->mtm_hash = kmem_zalloc((size_t) (sizeof (mapentry_t *) * nh), 175*0Sstevel@tonic-gate KM_SLEEP); 176*0Sstevel@tonic-gate mtm->mtm_nhash = nh; 177*0Sstevel@tonic-gate mtm->mtm_debug = ul->un_debug; 178*0Sstevel@tonic-gate mtm->mtm_type = maptype; 179*0Sstevel@tonic-gate 180*0Sstevel@tonic-gate mtm->mtm_cfrags = 0; 181*0Sstevel@tonic-gate mtm->mtm_cfragmax = logmap_maxcfrag_commit; 182*0Sstevel@tonic-gate 183*0Sstevel@tonic-gate /* 184*0Sstevel@tonic-gate * for scan test 185*0Sstevel@tonic-gate */ 186*0Sstevel@tonic-gate mtm->mtm_ul = ul; 187*0Sstevel@tonic-gate 188*0Sstevel@tonic-gate /* 189*0Sstevel@tonic-gate * Initialize locks 190*0Sstevel@tonic-gate */ 191*0Sstevel@tonic-gate mutex_init(&mtm->mtm_lock, NULL, MUTEX_DEFAULT, NULL); 192*0Sstevel@tonic-gate cv_init(&mtm->mtm_cv_commit, NULL, CV_DEFAULT, NULL); 193*0Sstevel@tonic-gate cv_init(&mtm->mtm_cv_next, NULL, CV_DEFAULT, NULL); 194*0Sstevel@tonic-gate cv_init(&mtm->mtm_cv_eot, NULL, CV_DEFAULT, NULL); 195*0Sstevel@tonic-gate cv_init(&mtm->mtm_cv, NULL, CV_DEFAULT, NULL); 196*0Sstevel@tonic-gate ASSERT(map_get_debug(ul, mtm)); 197*0Sstevel@tonic-gate 198*0Sstevel@tonic-gate return (mtm); 199*0Sstevel@tonic-gate } 200*0Sstevel@tonic-gate 201*0Sstevel@tonic-gate /* 202*0Sstevel@tonic-gate * DELTAMAP ROUTINES 203*0Sstevel@tonic-gate */ 204*0Sstevel@tonic-gate /* 205*0Sstevel@tonic-gate * deltamap tuning constants 206*0Sstevel@tonic-gate */ 207*0Sstevel@tonic-gate long deltamap_maxnme = 1024; /* global so it can be set */ 208*0Sstevel@tonic-gate 209*0Sstevel@tonic-gate int 210*0Sstevel@tonic-gate deltamap_need_commit(mt_map_t *mtm) 211*0Sstevel@tonic-gate { 212*0Sstevel@tonic-gate return (mtm->mtm_nme > deltamap_maxnme); 213*0Sstevel@tonic-gate } 214*0Sstevel@tonic-gate 215*0Sstevel@tonic-gate /* 216*0Sstevel@tonic-gate * put a delta into a deltamap; may sleep on memory 217*0Sstevel@tonic-gate */ 218*0Sstevel@tonic-gate void 219*0Sstevel@tonic-gate deltamap_add( 220*0Sstevel@tonic-gate mt_map_t *mtm, 221*0Sstevel@tonic-gate offset_t mof, 222*0Sstevel@tonic-gate off_t nb, 223*0Sstevel@tonic-gate delta_t dtyp, 224*0Sstevel@tonic-gate int (*func)(), 225*0Sstevel@tonic-gate ulong_t arg, 226*0Sstevel@tonic-gate threadtrans_t *tp) 227*0Sstevel@tonic-gate { 228*0Sstevel@tonic-gate int32_t hnb; 229*0Sstevel@tonic-gate mapentry_t *me; 230*0Sstevel@tonic-gate mapentry_t **mep; 231*0Sstevel@tonic-gate 232*0Sstevel@tonic-gate ASSERT(((mtm->mtm_debug & MT_CHECK_MAP) == 0) || 233*0Sstevel@tonic-gate map_check_linkage(mtm)); 234*0Sstevel@tonic-gate 235*0Sstevel@tonic-gate mutex_enter(&mtm->mtm_mutex); 236*0Sstevel@tonic-gate 237*0Sstevel@tonic-gate for (hnb = 0; nb; nb -= hnb, mof += hnb) { 238*0Sstevel@tonic-gate hnb = MAPBLOCKSIZE - (mof & MAPBLOCKOFF); 239*0Sstevel@tonic-gate if (hnb > nb) 240*0Sstevel@tonic-gate hnb = nb; 241*0Sstevel@tonic-gate /* 242*0Sstevel@tonic-gate * Search for dup entry. We need to ensure that we don't 243*0Sstevel@tonic-gate * replace a map entry which carries quota information 244*0Sstevel@tonic-gate * with a map entry which doesn't. In that case we lose 245*0Sstevel@tonic-gate * reference the the dquot structure which will not be 246*0Sstevel@tonic-gate * cleaned up by the push function me->me_func as this will 247*0Sstevel@tonic-gate * never be called. 248*0Sstevel@tonic-gate * The stray dquot would be found later by invalidatedq() 249*0Sstevel@tonic-gate * causing a panic when the filesystem is unmounted. 250*0Sstevel@tonic-gate */ 251*0Sstevel@tonic-gate mep = MAP_HASH(mof, mtm); 252*0Sstevel@tonic-gate for (me = *mep; me; me = me->me_hash) { 253*0Sstevel@tonic-gate if (DATAwithinME(mof, hnb, me)) { 254*0Sstevel@tonic-gate if (me->me_func == ufs_trans_push_quota) { 255*0Sstevel@tonic-gate /* 256*0Sstevel@tonic-gate * Don't remove quota entries which have 257*0Sstevel@tonic-gate * incremented the ref count (those with a 258*0Sstevel@tonic-gate * ufs_trans_push_quota push function). 259*0Sstevel@tonic-gate * Let logmap_add[_buf] clean them up. 260*0Sstevel@tonic-gate */ 261*0Sstevel@tonic-gate continue; 262*0Sstevel@tonic-gate } 263*0Sstevel@tonic-gate break; 264*0Sstevel@tonic-gate } 265*0Sstevel@tonic-gate ASSERT((dtyp == DT_CANCEL) || 266*0Sstevel@tonic-gate (!DATAoverlapME(mof, hnb, me)) || 267*0Sstevel@tonic-gate MEwithinDATA(me, mof, hnb)); 268*0Sstevel@tonic-gate } 269*0Sstevel@tonic-gate 270*0Sstevel@tonic-gate if (me) { 271*0Sstevel@tonic-gate /* already in map */ 272*0Sstevel@tonic-gate continue; 273*0Sstevel@tonic-gate } 274*0Sstevel@tonic-gate 275*0Sstevel@tonic-gate /* 276*0Sstevel@tonic-gate * Add up all the delta map deltas so we can compute 277*0Sstevel@tonic-gate * an upper bound on the log size used. 278*0Sstevel@tonic-gate * Note, some deltas get removed from the deltamap 279*0Sstevel@tonic-gate * before the deltamap_push by lufs_write_strategy 280*0Sstevel@tonic-gate * and so multiple deltas to the same mof offset 281*0Sstevel@tonic-gate * don't get cancelled here but in the logmap. 282*0Sstevel@tonic-gate * Thus we can't easily get a accurate count of 283*0Sstevel@tonic-gate * the log space used - only an upper bound. 284*0Sstevel@tonic-gate */ 285*0Sstevel@tonic-gate if (tp && (mtm->mtm_ul->un_deltamap == mtm)) { 286*0Sstevel@tonic-gate ASSERT(dtyp != DT_CANCEL); 287*0Sstevel@tonic-gate if (dtyp == DT_ABZERO) { 288*0Sstevel@tonic-gate tp->deltas_size += sizeof (struct delta); 289*0Sstevel@tonic-gate } else { 290*0Sstevel@tonic-gate tp->deltas_size += 291*0Sstevel@tonic-gate (hnb + sizeof (struct delta)); 292*0Sstevel@tonic-gate } 293*0Sstevel@tonic-gate } 294*0Sstevel@tonic-gate 295*0Sstevel@tonic-gate delta_stats[dtyp]++; 296*0Sstevel@tonic-gate 297*0Sstevel@tonic-gate /* 298*0Sstevel@tonic-gate * get a mapentry 299*0Sstevel@tonic-gate * May need to drop & re-grab the mtm_mutex 300*0Sstevel@tonic-gate * and then recheck for a duplicate 301*0Sstevel@tonic-gate */ 302*0Sstevel@tonic-gate me = kmem_cache_alloc(mapentry_cache, KM_NOSLEEP); 303*0Sstevel@tonic-gate if (me == NULL) { 304*0Sstevel@tonic-gate mutex_exit(&mtm->mtm_mutex); 305*0Sstevel@tonic-gate me = kmem_cache_alloc(mapentry_cache, KM_SLEEP); 306*0Sstevel@tonic-gate mutex_enter(&mtm->mtm_mutex); 307*0Sstevel@tonic-gate } 308*0Sstevel@tonic-gate bzero(me, sizeof (mapentry_t)); 309*0Sstevel@tonic-gate 310*0Sstevel@tonic-gate /* 311*0Sstevel@tonic-gate * initialize and put in deltamap 312*0Sstevel@tonic-gate */ 313*0Sstevel@tonic-gate me->me_mof = mof; 314*0Sstevel@tonic-gate me->me_nb = hnb; 315*0Sstevel@tonic-gate me->me_func = func; 316*0Sstevel@tonic-gate me->me_arg = arg; 317*0Sstevel@tonic-gate me->me_dt = dtyp; 318*0Sstevel@tonic-gate me->me_flags = ME_HASH; 319*0Sstevel@tonic-gate me->me_tid = mtm->mtm_tid; 320*0Sstevel@tonic-gate 321*0Sstevel@tonic-gate me->me_hash = *mep; 322*0Sstevel@tonic-gate *mep = me; 323*0Sstevel@tonic-gate me->me_next = (mapentry_t *)mtm; 324*0Sstevel@tonic-gate me->me_prev = mtm->mtm_prev; 325*0Sstevel@tonic-gate mtm->mtm_prev->me_next = me; 326*0Sstevel@tonic-gate mtm->mtm_prev = me; 327*0Sstevel@tonic-gate mtm->mtm_nme++; 328*0Sstevel@tonic-gate } 329*0Sstevel@tonic-gate mutex_exit(&mtm->mtm_mutex); 330*0Sstevel@tonic-gate 331*0Sstevel@tonic-gate ASSERT(((mtm->mtm_debug & MT_CHECK_MAP) == 0) || 332*0Sstevel@tonic-gate map_check_linkage(mtm)); 333*0Sstevel@tonic-gate } 334*0Sstevel@tonic-gate 335*0Sstevel@tonic-gate /* 336*0Sstevel@tonic-gate * remove deltas within (mof, nb) and return as linked list 337*0Sstevel@tonic-gate */ 338*0Sstevel@tonic-gate mapentry_t * 339*0Sstevel@tonic-gate deltamap_remove(mt_map_t *mtm, offset_t mof, off_t nb) 340*0Sstevel@tonic-gate { 341*0Sstevel@tonic-gate off_t hnb; 342*0Sstevel@tonic-gate mapentry_t *me; 343*0Sstevel@tonic-gate mapentry_t **mep; 344*0Sstevel@tonic-gate mapentry_t *mer; 345*0Sstevel@tonic-gate 346*0Sstevel@tonic-gate if (mtm == NULL) 347*0Sstevel@tonic-gate return (NULL); 348*0Sstevel@tonic-gate 349*0Sstevel@tonic-gate ASSERT(((mtm->mtm_debug & MT_CHECK_MAP) == 0) || 350*0Sstevel@tonic-gate map_check_linkage(mtm)); 351*0Sstevel@tonic-gate 352*0Sstevel@tonic-gate mutex_enter(&mtm->mtm_mutex); 353*0Sstevel@tonic-gate for (mer = NULL, hnb = 0; nb; nb -= hnb, mof += hnb) { 354*0Sstevel@tonic-gate hnb = MAPBLOCKSIZE - (mof & MAPBLOCKOFF); 355*0Sstevel@tonic-gate if (hnb > nb) 356*0Sstevel@tonic-gate hnb = nb; 357*0Sstevel@tonic-gate /* 358*0Sstevel@tonic-gate * remove entries from hash and return as a aged linked list 359*0Sstevel@tonic-gate */ 360*0Sstevel@tonic-gate mep = MAP_HASH(mof, mtm); 361*0Sstevel@tonic-gate while ((me = *mep) != 0) { 362*0Sstevel@tonic-gate if (MEwithinDATA(me, mof, hnb)) { 363*0Sstevel@tonic-gate *mep = me->me_hash; 364*0Sstevel@tonic-gate me->me_next->me_prev = me->me_prev; 365*0Sstevel@tonic-gate me->me_prev->me_next = me->me_next; 366*0Sstevel@tonic-gate me->me_hash = mer; 367*0Sstevel@tonic-gate mer = me; 368*0Sstevel@tonic-gate me->me_flags |= ME_LIST; 369*0Sstevel@tonic-gate me->me_flags &= ~ME_HASH; 370*0Sstevel@tonic-gate mtm->mtm_nme--; 371*0Sstevel@tonic-gate } else 372*0Sstevel@tonic-gate mep = &me->me_hash; 373*0Sstevel@tonic-gate } 374*0Sstevel@tonic-gate } 375*0Sstevel@tonic-gate mutex_exit(&mtm->mtm_mutex); 376*0Sstevel@tonic-gate 377*0Sstevel@tonic-gate ASSERT(((mtm->mtm_debug & MT_CHECK_MAP) == 0) || 378*0Sstevel@tonic-gate map_check_linkage(mtm)); 379*0Sstevel@tonic-gate 380*0Sstevel@tonic-gate return (mer); 381*0Sstevel@tonic-gate } 382*0Sstevel@tonic-gate 383*0Sstevel@tonic-gate /* 384*0Sstevel@tonic-gate * delete entries within (mof, nb) 385*0Sstevel@tonic-gate */ 386*0Sstevel@tonic-gate void 387*0Sstevel@tonic-gate deltamap_del(mt_map_t *mtm, offset_t mof, off_t nb) 388*0Sstevel@tonic-gate { 389*0Sstevel@tonic-gate mapentry_t *me; 390*0Sstevel@tonic-gate mapentry_t *menext; 391*0Sstevel@tonic-gate 392*0Sstevel@tonic-gate menext = deltamap_remove(mtm, mof, nb); 393*0Sstevel@tonic-gate while ((me = menext) != 0) { 394*0Sstevel@tonic-gate menext = me->me_hash; 395*0Sstevel@tonic-gate kmem_cache_free(mapentry_cache, me); 396*0Sstevel@tonic-gate } 397*0Sstevel@tonic-gate } 398*0Sstevel@tonic-gate 399*0Sstevel@tonic-gate /* 400*0Sstevel@tonic-gate * Call the indicated function to cause deltas to move to the logmap. 401*0Sstevel@tonic-gate * top_end_sync() is the only caller of this function and 402*0Sstevel@tonic-gate * it has waited for the completion of all threads, so there can 403*0Sstevel@tonic-gate * be no other activity in the deltamap. Therefore we don't need to 404*0Sstevel@tonic-gate * hold the deltamap lock. 405*0Sstevel@tonic-gate */ 406*0Sstevel@tonic-gate void 407*0Sstevel@tonic-gate deltamap_push(ml_unit_t *ul) 408*0Sstevel@tonic-gate { 409*0Sstevel@tonic-gate delta_t dtyp; 410*0Sstevel@tonic-gate int (*func)(); 411*0Sstevel@tonic-gate ulong_t arg; 412*0Sstevel@tonic-gate mapentry_t *me; 413*0Sstevel@tonic-gate offset_t mof; 414*0Sstevel@tonic-gate off_t nb; 415*0Sstevel@tonic-gate mt_map_t *mtm = ul->un_deltamap; 416*0Sstevel@tonic-gate 417*0Sstevel@tonic-gate ASSERT(((mtm->mtm_debug & MT_CHECK_MAP) == 0) || 418*0Sstevel@tonic-gate map_check_linkage(mtm)); 419*0Sstevel@tonic-gate 420*0Sstevel@tonic-gate /* 421*0Sstevel@tonic-gate * for every entry in the deltamap 422*0Sstevel@tonic-gate */ 423*0Sstevel@tonic-gate while ((me = mtm->mtm_next) != (mapentry_t *)mtm) { 424*0Sstevel@tonic-gate ASSERT(me->me_func); 425*0Sstevel@tonic-gate func = me->me_func; 426*0Sstevel@tonic-gate dtyp = me->me_dt; 427*0Sstevel@tonic-gate arg = me->me_arg; 428*0Sstevel@tonic-gate mof = me->me_mof; 429*0Sstevel@tonic-gate nb = me->me_nb; 430*0Sstevel@tonic-gate if ((ul->un_flags & LDL_ERROR) || 431*0Sstevel@tonic-gate (*func)(ul->un_ufsvfs, dtyp, arg)) 432*0Sstevel@tonic-gate deltamap_del(mtm, mof, nb); 433*0Sstevel@tonic-gate } 434*0Sstevel@tonic-gate 435*0Sstevel@tonic-gate ASSERT(((mtm->mtm_debug & MT_CHECK_MAP) == 0) || 436*0Sstevel@tonic-gate map_check_linkage(mtm)); 437*0Sstevel@tonic-gate } 438*0Sstevel@tonic-gate 439*0Sstevel@tonic-gate /* 440*0Sstevel@tonic-gate * LOGMAP ROUTINES 441*0Sstevel@tonic-gate */ 442*0Sstevel@tonic-gate 443*0Sstevel@tonic-gate int 444*0Sstevel@tonic-gate logmap_need_commit(mt_map_t *mtm) 445*0Sstevel@tonic-gate { 446*0Sstevel@tonic-gate return ((mtm->mtm_nmet > logmap_maxnme_commit) || 447*0Sstevel@tonic-gate (mtm->mtm_cfrags >= mtm->mtm_cfragmax)); 448*0Sstevel@tonic-gate } 449*0Sstevel@tonic-gate 450*0Sstevel@tonic-gate int 451*0Sstevel@tonic-gate logmap_need_roll_async(mt_map_t *mtm) 452*0Sstevel@tonic-gate { 453*0Sstevel@tonic-gate return (mtm->mtm_nme > logmap_maxnme_async); 454*0Sstevel@tonic-gate } 455*0Sstevel@tonic-gate 456*0Sstevel@tonic-gate int 457*0Sstevel@tonic-gate logmap_need_roll_sync(mt_map_t *mtm) 458*0Sstevel@tonic-gate { 459*0Sstevel@tonic-gate return (mtm->mtm_nme > logmap_maxnme_sync); 460*0Sstevel@tonic-gate } 461*0Sstevel@tonic-gate 462*0Sstevel@tonic-gate void 463*0Sstevel@tonic-gate logmap_start_roll(ml_unit_t *ul) 464*0Sstevel@tonic-gate { 465*0Sstevel@tonic-gate mt_map_t *logmap = ul->un_logmap; 466*0Sstevel@tonic-gate 467*0Sstevel@tonic-gate logmap_settail(logmap, ul); 468*0Sstevel@tonic-gate ASSERT(!(ul->un_flags & LDL_NOROLL)); 469*0Sstevel@tonic-gate mutex_enter(&logmap->mtm_mutex); 470*0Sstevel@tonic-gate if ((logmap->mtm_flags & MTM_ROLL_RUNNING) == 0) { 471*0Sstevel@tonic-gate logmap->mtm_flags |= MTM_ROLL_RUNNING; 472*0Sstevel@tonic-gate logmap->mtm_flags &= ~(MTM_FORCE_ROLL | MTM_ROLL_EXIT); 473*0Sstevel@tonic-gate (void) thread_create(NULL, 0, trans_roll, ul, 0, &p0, 474*0Sstevel@tonic-gate TS_RUN, minclsyspri); 475*0Sstevel@tonic-gate } 476*0Sstevel@tonic-gate mutex_exit(&logmap->mtm_mutex); 477*0Sstevel@tonic-gate } 478*0Sstevel@tonic-gate 479*0Sstevel@tonic-gate void 480*0Sstevel@tonic-gate logmap_kill_roll(ml_unit_t *ul) 481*0Sstevel@tonic-gate { 482*0Sstevel@tonic-gate mt_map_t *mtm = ul->un_logmap; 483*0Sstevel@tonic-gate 484*0Sstevel@tonic-gate if (mtm == NULL) 485*0Sstevel@tonic-gate return; 486*0Sstevel@tonic-gate 487*0Sstevel@tonic-gate mutex_enter(&mtm->mtm_mutex); 488*0Sstevel@tonic-gate 489*0Sstevel@tonic-gate while (mtm->mtm_flags & MTM_ROLL_RUNNING) { 490*0Sstevel@tonic-gate mtm->mtm_flags |= MTM_ROLL_EXIT; 491*0Sstevel@tonic-gate cv_signal(&mtm->mtm_to_roll_cv); 492*0Sstevel@tonic-gate cv_wait(&mtm->mtm_from_roll_cv, &mtm->mtm_mutex); 493*0Sstevel@tonic-gate } 494*0Sstevel@tonic-gate mutex_exit(&mtm->mtm_mutex); 495*0Sstevel@tonic-gate } 496*0Sstevel@tonic-gate 497*0Sstevel@tonic-gate /* 498*0Sstevel@tonic-gate * kick the roll thread if it's not doing anything 499*0Sstevel@tonic-gate */ 500*0Sstevel@tonic-gate void 501*0Sstevel@tonic-gate logmap_forceroll_nowait(mt_map_t *logmap) 502*0Sstevel@tonic-gate { 503*0Sstevel@tonic-gate /* 504*0Sstevel@tonic-gate * Don't need to lock mtm_mutex to read mtm_flags here as we 505*0Sstevel@tonic-gate * don't care in the rare case when we get a transitional value 506*0Sstevel@tonic-gate * of mtm_flags. Just by signalling the thread it will wakeup 507*0Sstevel@tonic-gate * and notice it has too many logmap entries. 508*0Sstevel@tonic-gate */ 509*0Sstevel@tonic-gate ASSERT(!(logmap->mtm_ul->un_flags & LDL_NOROLL)); 510*0Sstevel@tonic-gate if ((logmap->mtm_flags & MTM_ROLLING) == 0) { 511*0Sstevel@tonic-gate cv_signal(&logmap->mtm_to_roll_cv); 512*0Sstevel@tonic-gate } 513*0Sstevel@tonic-gate } 514*0Sstevel@tonic-gate 515*0Sstevel@tonic-gate /* 516*0Sstevel@tonic-gate * kick the roll thread and wait for it to finish a cycle 517*0Sstevel@tonic-gate */ 518*0Sstevel@tonic-gate void 519*0Sstevel@tonic-gate logmap_forceroll(mt_map_t *mtm) 520*0Sstevel@tonic-gate { 521*0Sstevel@tonic-gate mutex_enter(&mtm->mtm_mutex); 522*0Sstevel@tonic-gate if ((mtm->mtm_flags & MTM_FORCE_ROLL) == 0) { 523*0Sstevel@tonic-gate mtm->mtm_flags |= MTM_FORCE_ROLL; 524*0Sstevel@tonic-gate cv_signal(&mtm->mtm_to_roll_cv); 525*0Sstevel@tonic-gate } 526*0Sstevel@tonic-gate do { 527*0Sstevel@tonic-gate if ((mtm->mtm_flags & MTM_ROLL_RUNNING) == 0) { 528*0Sstevel@tonic-gate mtm->mtm_flags &= ~MTM_FORCE_ROLL; 529*0Sstevel@tonic-gate goto out; 530*0Sstevel@tonic-gate } 531*0Sstevel@tonic-gate cv_wait(&mtm->mtm_from_roll_cv, &mtm->mtm_mutex); 532*0Sstevel@tonic-gate } while (mtm->mtm_flags & MTM_FORCE_ROLL); 533*0Sstevel@tonic-gate out: 534*0Sstevel@tonic-gate mutex_exit(&mtm->mtm_mutex); 535*0Sstevel@tonic-gate } 536*0Sstevel@tonic-gate 537*0Sstevel@tonic-gate /* 538*0Sstevel@tonic-gate * remove rolled deltas within (mof, nb) and free them 539*0Sstevel@tonic-gate */ 540*0Sstevel@tonic-gate void 541*0Sstevel@tonic-gate logmap_remove_roll(mt_map_t *mtm, offset_t mof, off_t nb) 542*0Sstevel@tonic-gate { 543*0Sstevel@tonic-gate int dolock = 0; 544*0Sstevel@tonic-gate off_t hnb; 545*0Sstevel@tonic-gate mapentry_t *me; 546*0Sstevel@tonic-gate mapentry_t **mep; 547*0Sstevel@tonic-gate offset_t savmof = mof; 548*0Sstevel@tonic-gate off_t savnb = nb; 549*0Sstevel@tonic-gate 550*0Sstevel@tonic-gate ASSERT(((mtm->mtm_debug & MT_CHECK_MAP) == 0) || 551*0Sstevel@tonic-gate map_check_linkage(mtm)); 552*0Sstevel@tonic-gate 553*0Sstevel@tonic-gate again: 554*0Sstevel@tonic-gate if (dolock) 555*0Sstevel@tonic-gate rw_enter(&mtm->mtm_rwlock, RW_WRITER); 556*0Sstevel@tonic-gate mutex_enter(&mtm->mtm_mutex); 557*0Sstevel@tonic-gate for (hnb = 0; nb; nb -= hnb, mof += hnb) { 558*0Sstevel@tonic-gate hnb = MAPBLOCKSIZE - (mof & MAPBLOCKOFF); 559*0Sstevel@tonic-gate if (hnb > nb) 560*0Sstevel@tonic-gate hnb = nb; 561*0Sstevel@tonic-gate /* 562*0Sstevel@tonic-gate * remove and free the rolled entries 563*0Sstevel@tonic-gate */ 564*0Sstevel@tonic-gate mep = MAP_HASH(mof, mtm); 565*0Sstevel@tonic-gate while ((me = *mep) != 0) { 566*0Sstevel@tonic-gate if ((me->me_flags & ME_ROLL) && 567*0Sstevel@tonic-gate (MEwithinDATA(me, mof, hnb))) { 568*0Sstevel@tonic-gate if (me->me_flags & ME_AGE) { 569*0Sstevel@tonic-gate ASSERT(dolock == 0); 570*0Sstevel@tonic-gate dolock = 1; 571*0Sstevel@tonic-gate mutex_exit(&mtm->mtm_mutex); 572*0Sstevel@tonic-gate mof = savmof; 573*0Sstevel@tonic-gate nb = savnb; 574*0Sstevel@tonic-gate goto again; 575*0Sstevel@tonic-gate } 576*0Sstevel@tonic-gate *mep = me->me_hash; 577*0Sstevel@tonic-gate me->me_next->me_prev = me->me_prev; 578*0Sstevel@tonic-gate me->me_prev->me_next = me->me_next; 579*0Sstevel@tonic-gate me->me_flags &= ~(ME_HASH|ME_ROLL); 580*0Sstevel@tonic-gate ASSERT(!(me->me_flags & ME_USER)); 581*0Sstevel@tonic-gate mtm->mtm_nme--; 582*0Sstevel@tonic-gate /* 583*0Sstevel@tonic-gate * cancelled entries are handled by someone else 584*0Sstevel@tonic-gate */ 585*0Sstevel@tonic-gate if ((me->me_flags & ME_CANCEL) == 0) { 586*0Sstevel@tonic-gate roll_stats[me->me_dt]++; 587*0Sstevel@tonic-gate CRB_RELE(me); 588*0Sstevel@tonic-gate kmem_cache_free(mapentry_cache, me); 589*0Sstevel@tonic-gate } 590*0Sstevel@tonic-gate } else 591*0Sstevel@tonic-gate mep = &me->me_hash; 592*0Sstevel@tonic-gate } 593*0Sstevel@tonic-gate } 594*0Sstevel@tonic-gate mutex_exit(&mtm->mtm_mutex); 595*0Sstevel@tonic-gate 596*0Sstevel@tonic-gate ASSERT(((mtm->mtm_debug & MT_CHECK_MAP) == 0) || 597*0Sstevel@tonic-gate map_check_linkage(mtm)); 598*0Sstevel@tonic-gate 599*0Sstevel@tonic-gate if (dolock) 600*0Sstevel@tonic-gate rw_exit(&mtm->mtm_rwlock); 601*0Sstevel@tonic-gate } 602*0Sstevel@tonic-gate 603*0Sstevel@tonic-gate /* 604*0Sstevel@tonic-gate * Find the disk offset of the next delta to roll. 605*0Sstevel@tonic-gate * Returns 0: no more deltas to roll or a transaction is being committed 606*0Sstevel@tonic-gate * 1: a delta to roll has been found and *mofp points 607*0Sstevel@tonic-gate * to the master file disk offset 608*0Sstevel@tonic-gate */ 609*0Sstevel@tonic-gate int 610*0Sstevel@tonic-gate logmap_next_roll(mt_map_t *logmap, offset_t *mofp) 611*0Sstevel@tonic-gate { 612*0Sstevel@tonic-gate mapentry_t *me; 613*0Sstevel@tonic-gate 614*0Sstevel@tonic-gate ASSERT(((logmap->mtm_debug & MT_CHECK_MAP) == 0) || 615*0Sstevel@tonic-gate map_check_linkage(logmap)); 616*0Sstevel@tonic-gate 617*0Sstevel@tonic-gate mutex_enter(&logmap->mtm_mutex); 618*0Sstevel@tonic-gate for (me = logmap->mtm_next; me != (mapentry_t *)logmap; 619*0Sstevel@tonic-gate me = me->me_next) { 620*0Sstevel@tonic-gate /* already rolled */ 621*0Sstevel@tonic-gate if (me->me_flags & ME_ROLL) { 622*0Sstevel@tonic-gate continue; 623*0Sstevel@tonic-gate } 624*0Sstevel@tonic-gate 625*0Sstevel@tonic-gate /* part of currently busy transaction; stop */ 626*0Sstevel@tonic-gate if (me->me_tid == logmap->mtm_tid) { 627*0Sstevel@tonic-gate break; 628*0Sstevel@tonic-gate } 629*0Sstevel@tonic-gate 630*0Sstevel@tonic-gate /* part of commit-in-progress transaction; stop */ 631*0Sstevel@tonic-gate if (me->me_tid == logmap->mtm_committid) { 632*0Sstevel@tonic-gate break; 633*0Sstevel@tonic-gate } 634*0Sstevel@tonic-gate 635*0Sstevel@tonic-gate /* 636*0Sstevel@tonic-gate * We shouldn't see a DT_CANCEL mapentry whose 637*0Sstevel@tonic-gate * tid != mtm_committid, or != mtm_tid since 638*0Sstevel@tonic-gate * these are removed at the end of each committed 639*0Sstevel@tonic-gate * transaction. 640*0Sstevel@tonic-gate */ 641*0Sstevel@tonic-gate ASSERT(!(me->me_dt == DT_CANCEL)); 642*0Sstevel@tonic-gate 643*0Sstevel@tonic-gate *mofp = me->me_mof; 644*0Sstevel@tonic-gate mutex_exit(&logmap->mtm_mutex); 645*0Sstevel@tonic-gate return (1); 646*0Sstevel@tonic-gate } 647*0Sstevel@tonic-gate mutex_exit(&logmap->mtm_mutex); 648*0Sstevel@tonic-gate return (0); 649*0Sstevel@tonic-gate } 650*0Sstevel@tonic-gate 651*0Sstevel@tonic-gate /* 652*0Sstevel@tonic-gate * put mapentry on sorted age list 653*0Sstevel@tonic-gate */ 654*0Sstevel@tonic-gate static void 655*0Sstevel@tonic-gate logmap_list_age(mapentry_t **age, mapentry_t *meadd) 656*0Sstevel@tonic-gate { 657*0Sstevel@tonic-gate mapentry_t *me; 658*0Sstevel@tonic-gate 659*0Sstevel@tonic-gate ASSERT(!(meadd->me_flags & (ME_AGE|ME_LIST))); 660*0Sstevel@tonic-gate 661*0Sstevel@tonic-gate for (me = *age; me; age = &me->me_agenext, me = *age) { 662*0Sstevel@tonic-gate if (me->me_age > meadd->me_age) 663*0Sstevel@tonic-gate break; 664*0Sstevel@tonic-gate } 665*0Sstevel@tonic-gate meadd->me_agenext = me; 666*0Sstevel@tonic-gate meadd->me_flags |= ME_AGE; 667*0Sstevel@tonic-gate *age = meadd; 668*0Sstevel@tonic-gate } 669*0Sstevel@tonic-gate 670*0Sstevel@tonic-gate /* 671*0Sstevel@tonic-gate * get a list of deltas within <mof, mof+nb> 672*0Sstevel@tonic-gate * returns with mtm_rwlock held 673*0Sstevel@tonic-gate * return value says whether the entire mof range is covered by deltas 674*0Sstevel@tonic-gate */ 675*0Sstevel@tonic-gate int 676*0Sstevel@tonic-gate logmap_list_get( 677*0Sstevel@tonic-gate mt_map_t *mtm, 678*0Sstevel@tonic-gate offset_t mof, 679*0Sstevel@tonic-gate off_t nb, 680*0Sstevel@tonic-gate mapentry_t **age) 681*0Sstevel@tonic-gate { 682*0Sstevel@tonic-gate off_t hnb; 683*0Sstevel@tonic-gate mapentry_t *me; 684*0Sstevel@tonic-gate mapentry_t **mep; 685*0Sstevel@tonic-gate int rwtype = RW_READER; 686*0Sstevel@tonic-gate offset_t savmof = mof; 687*0Sstevel@tonic-gate off_t savnb = nb; 688*0Sstevel@tonic-gate int entire = 0; 689*0Sstevel@tonic-gate crb_t *crb; 690*0Sstevel@tonic-gate 691*0Sstevel@tonic-gate mtm->mtm_ref = 1; 692*0Sstevel@tonic-gate again: 693*0Sstevel@tonic-gate 694*0Sstevel@tonic-gate ASSERT(((mtm->mtm_debug & MT_CHECK_MAP) == 0) || 695*0Sstevel@tonic-gate map_check_linkage(mtm)); 696*0Sstevel@tonic-gate 697*0Sstevel@tonic-gate rw_enter(&mtm->mtm_rwlock, rwtype); 698*0Sstevel@tonic-gate *age = NULL; 699*0Sstevel@tonic-gate mutex_enter(&mtm->mtm_mutex); 700*0Sstevel@tonic-gate for (hnb = 0; nb; nb -= hnb, mof += hnb) { 701*0Sstevel@tonic-gate hnb = MAPBLOCKSIZE - (mof & MAPBLOCKOFF); 702*0Sstevel@tonic-gate if (hnb > nb) 703*0Sstevel@tonic-gate hnb = nb; 704*0Sstevel@tonic-gate /* 705*0Sstevel@tonic-gate * find overlapping entries 706*0Sstevel@tonic-gate */ 707*0Sstevel@tonic-gate mep = MAP_HASH(mof, mtm); 708*0Sstevel@tonic-gate for (me = *mep; me; me = me->me_hash) { 709*0Sstevel@tonic-gate if (me->me_dt == DT_CANCEL) 710*0Sstevel@tonic-gate continue; 711*0Sstevel@tonic-gate if (!DATAoverlapME(mof, hnb, me)) 712*0Sstevel@tonic-gate continue; 713*0Sstevel@tonic-gate /* 714*0Sstevel@tonic-gate * check if map entry is in use 715*0Sstevel@tonic-gate * (about to be rolled). 716*0Sstevel@tonic-gate */ 717*0Sstevel@tonic-gate if (me->me_flags & ME_AGE) { 718*0Sstevel@tonic-gate /* 719*0Sstevel@tonic-gate * reset the age bit in the list, 720*0Sstevel@tonic-gate * upgrade the lock, and try again 721*0Sstevel@tonic-gate */ 722*0Sstevel@tonic-gate for (me = *age; me; me = *age) { 723*0Sstevel@tonic-gate *age = me->me_agenext; 724*0Sstevel@tonic-gate me->me_flags &= ~ME_AGE; 725*0Sstevel@tonic-gate } 726*0Sstevel@tonic-gate mutex_exit(&mtm->mtm_mutex); 727*0Sstevel@tonic-gate rw_exit(&mtm->mtm_rwlock); 728*0Sstevel@tonic-gate rwtype = RW_WRITER; 729*0Sstevel@tonic-gate mof = savmof; 730*0Sstevel@tonic-gate nb = savnb; 731*0Sstevel@tonic-gate entire = 0; 732*0Sstevel@tonic-gate goto again; 733*0Sstevel@tonic-gate } else { 734*0Sstevel@tonic-gate /* add mapentry to age ordered list */ 735*0Sstevel@tonic-gate logmap_list_age(age, me); 736*0Sstevel@tonic-gate crb = me->me_crb; 737*0Sstevel@tonic-gate if (crb) { 738*0Sstevel@tonic-gate if (DATAwithinCRB(savmof, savnb, crb)) { 739*0Sstevel@tonic-gate entire = 1; 740*0Sstevel@tonic-gate } 741*0Sstevel@tonic-gate } else { 742*0Sstevel@tonic-gate if (DATAwithinME(savmof, savnb, me)) { 743*0Sstevel@tonic-gate entire = 1; 744*0Sstevel@tonic-gate } 745*0Sstevel@tonic-gate } 746*0Sstevel@tonic-gate } 747*0Sstevel@tonic-gate } 748*0Sstevel@tonic-gate } 749*0Sstevel@tonic-gate mutex_exit(&mtm->mtm_mutex); 750*0Sstevel@tonic-gate 751*0Sstevel@tonic-gate ASSERT(RW_LOCK_HELD(&mtm->mtm_rwlock)); 752*0Sstevel@tonic-gate return (entire); 753*0Sstevel@tonic-gate } 754*0Sstevel@tonic-gate 755*0Sstevel@tonic-gate /* 756*0Sstevel@tonic-gate * Get a list of deltas for rolling - returns sucess or failure. 757*0Sstevel@tonic-gate * Also return the cached roll buffer if all deltas point to it. 758*0Sstevel@tonic-gate */ 759*0Sstevel@tonic-gate int 760*0Sstevel@tonic-gate logmap_list_get_roll(mt_map_t *logmap, offset_t mof, rollbuf_t *rbp) 761*0Sstevel@tonic-gate { 762*0Sstevel@tonic-gate mapentry_t *me, **mep, *age = NULL; 763*0Sstevel@tonic-gate crb_t *crb = NULL; 764*0Sstevel@tonic-gate 765*0Sstevel@tonic-gate ASSERT(RW_LOCK_HELD(&logmap->mtm_rwlock)); 766*0Sstevel@tonic-gate ASSERT(((logmap->mtm_debug & MT_CHECK_MAP) == 0) || 767*0Sstevel@tonic-gate map_check_linkage(logmap)); 768*0Sstevel@tonic-gate ASSERT((mof & MAPBLOCKOFF) == 0); 769*0Sstevel@tonic-gate 770*0Sstevel@tonic-gate rbp->rb_crb = NULL; 771*0Sstevel@tonic-gate 772*0Sstevel@tonic-gate /* 773*0Sstevel@tonic-gate * find overlapping entries 774*0Sstevel@tonic-gate */ 775*0Sstevel@tonic-gate mutex_enter(&logmap->mtm_mutex); 776*0Sstevel@tonic-gate mep = MAP_HASH(mof, logmap); 777*0Sstevel@tonic-gate for (me = *mep; me; me = me->me_hash) { 778*0Sstevel@tonic-gate if (!DATAoverlapME(mof, MAPBLOCKSIZE, me)) 779*0Sstevel@tonic-gate continue; 780*0Sstevel@tonic-gate if (me->me_tid == logmap->mtm_tid) 781*0Sstevel@tonic-gate continue; 782*0Sstevel@tonic-gate if (me->me_tid == logmap->mtm_committid) 783*0Sstevel@tonic-gate continue; 784*0Sstevel@tonic-gate if (me->me_dt == DT_CANCEL) 785*0Sstevel@tonic-gate continue; 786*0Sstevel@tonic-gate 787*0Sstevel@tonic-gate /* 788*0Sstevel@tonic-gate * Check if map entry is in use (by lufs_read_strategy()) 789*0Sstevel@tonic-gate * and if so reset the age bit in the list, 790*0Sstevel@tonic-gate * upgrade the lock, and try again 791*0Sstevel@tonic-gate */ 792*0Sstevel@tonic-gate if (me->me_flags & ME_AGE) { 793*0Sstevel@tonic-gate for (me = age; me; me = age) { 794*0Sstevel@tonic-gate age = me->me_agenext; 795*0Sstevel@tonic-gate me->me_flags &= ~ME_AGE; 796*0Sstevel@tonic-gate } 797*0Sstevel@tonic-gate mutex_exit(&logmap->mtm_mutex); 798*0Sstevel@tonic-gate return (1); /* failure */ 799*0Sstevel@tonic-gate } else { 800*0Sstevel@tonic-gate /* add mapentry to age ordered list */ 801*0Sstevel@tonic-gate logmap_list_age(&age, me); 802*0Sstevel@tonic-gate } 803*0Sstevel@tonic-gate } 804*0Sstevel@tonic-gate if (!age) { 805*0Sstevel@tonic-gate goto out; 806*0Sstevel@tonic-gate } 807*0Sstevel@tonic-gate 808*0Sstevel@tonic-gate /* 809*0Sstevel@tonic-gate * Mark the deltas as being rolled. 810*0Sstevel@tonic-gate */ 811*0Sstevel@tonic-gate for (me = age; me; me = me->me_agenext) { 812*0Sstevel@tonic-gate me->me_flags |= ME_ROLL; 813*0Sstevel@tonic-gate } 814*0Sstevel@tonic-gate 815*0Sstevel@tonic-gate /* 816*0Sstevel@tonic-gate * Test if all deltas are covered by one valid roll buffer 817*0Sstevel@tonic-gate */ 818*0Sstevel@tonic-gate crb = age->me_crb; 819*0Sstevel@tonic-gate if (crb && !(crb->c_invalid)) { 820*0Sstevel@tonic-gate for (me = age; me; me = me->me_agenext) { 821*0Sstevel@tonic-gate if (me->me_crb != crb) { 822*0Sstevel@tonic-gate crb = NULL; 823*0Sstevel@tonic-gate break; 824*0Sstevel@tonic-gate } 825*0Sstevel@tonic-gate } 826*0Sstevel@tonic-gate rbp->rb_crb = crb; 827*0Sstevel@tonic-gate } 828*0Sstevel@tonic-gate out: 829*0Sstevel@tonic-gate rbp->rb_age = age; 830*0Sstevel@tonic-gate 831*0Sstevel@tonic-gate mutex_exit(&logmap->mtm_mutex); 832*0Sstevel@tonic-gate 833*0Sstevel@tonic-gate ASSERT(((logmap->mtm_debug & MT_SCAN) == 0) || 834*0Sstevel@tonic-gate logmap_logscan_debug(logmap, age)); 835*0Sstevel@tonic-gate ASSERT(RW_LOCK_HELD(&logmap->mtm_rwlock)); 836*0Sstevel@tonic-gate return (0); /* success */ 837*0Sstevel@tonic-gate } 838*0Sstevel@tonic-gate 839*0Sstevel@tonic-gate void 840*0Sstevel@tonic-gate logmap_list_put_roll(mt_map_t *mtm, mapentry_t *age) 841*0Sstevel@tonic-gate { 842*0Sstevel@tonic-gate mapentry_t *me; 843*0Sstevel@tonic-gate 844*0Sstevel@tonic-gate ASSERT(RW_LOCK_HELD(&mtm->mtm_rwlock)); 845*0Sstevel@tonic-gate mutex_enter(&mtm->mtm_mutex); 846*0Sstevel@tonic-gate for (me = age; me; me = age) { 847*0Sstevel@tonic-gate age = me->me_agenext; 848*0Sstevel@tonic-gate me->me_flags &= ~ME_AGE; 849*0Sstevel@tonic-gate } 850*0Sstevel@tonic-gate mutex_exit(&mtm->mtm_mutex); 851*0Sstevel@tonic-gate } 852*0Sstevel@tonic-gate 853*0Sstevel@tonic-gate void 854*0Sstevel@tonic-gate logmap_list_put(mt_map_t *mtm, mapentry_t *age) 855*0Sstevel@tonic-gate { 856*0Sstevel@tonic-gate mapentry_t *me; 857*0Sstevel@tonic-gate 858*0Sstevel@tonic-gate ASSERT(RW_LOCK_HELD(&mtm->mtm_rwlock)); 859*0Sstevel@tonic-gate mutex_enter(&mtm->mtm_mutex); 860*0Sstevel@tonic-gate for (me = age; me; me = age) { 861*0Sstevel@tonic-gate age = me->me_agenext; 862*0Sstevel@tonic-gate me->me_flags &= ~ME_AGE; 863*0Sstevel@tonic-gate } 864*0Sstevel@tonic-gate mutex_exit(&mtm->mtm_mutex); 865*0Sstevel@tonic-gate rw_exit(&mtm->mtm_rwlock); 866*0Sstevel@tonic-gate } 867*0Sstevel@tonic-gate 868*0Sstevel@tonic-gate #define UFS_RW_BALANCE 2 869*0Sstevel@tonic-gate int ufs_rw_balance = UFS_RW_BALANCE; 870*0Sstevel@tonic-gate 871*0Sstevel@tonic-gate /* 872*0Sstevel@tonic-gate * Check if we need to read the master. 873*0Sstevel@tonic-gate * The master does not need to be read if the log deltas to the 874*0Sstevel@tonic-gate * block are for one contiguous set of full disk sectors. 875*0Sstevel@tonic-gate * Both cylinder group bit maps DT_CG (8K); directory entries (512B); 876*0Sstevel@tonic-gate * and possibly others should not require master disk reads. 877*0Sstevel@tonic-gate * Calculate the sector map for writing later. 878*0Sstevel@tonic-gate */ 879*0Sstevel@tonic-gate int 880*0Sstevel@tonic-gate logmap_setup_read(mapentry_t *age, rollbuf_t *rbp) 881*0Sstevel@tonic-gate { 882*0Sstevel@tonic-gate offset_t mof; 883*0Sstevel@tonic-gate crb_t *crb; 884*0Sstevel@tonic-gate mapentry_t *me; 885*0Sstevel@tonic-gate int32_t nb; 886*0Sstevel@tonic-gate int i; 887*0Sstevel@tonic-gate int start_sec, end_sec; 888*0Sstevel@tonic-gate int read_needed = 0; 889*0Sstevel@tonic-gate int all_inodes = 1; 890*0Sstevel@tonic-gate int first_sec = INT_MAX; 891*0Sstevel@tonic-gate int last_sec = -1; 892*0Sstevel@tonic-gate rbsecmap_t secmap = 0; 893*0Sstevel@tonic-gate 894*0Sstevel@tonic-gate /* LINTED: warning: logical expression always true: op "||" */ 895*0Sstevel@tonic-gate ASSERT((MAPBLOCKSIZE / DEV_BSIZE) == (sizeof (secmap) * NBBY)); 896*0Sstevel@tonic-gate 897*0Sstevel@tonic-gate for (me = age; me; me = me->me_agenext) { 898*0Sstevel@tonic-gate crb = me->me_crb; 899*0Sstevel@tonic-gate if (crb) { 900*0Sstevel@tonic-gate nb = crb->c_nb; 901*0Sstevel@tonic-gate mof = crb->c_mof; 902*0Sstevel@tonic-gate } else { 903*0Sstevel@tonic-gate nb = me->me_nb; 904*0Sstevel@tonic-gate mof = me->me_mof; 905*0Sstevel@tonic-gate } 906*0Sstevel@tonic-gate 907*0Sstevel@tonic-gate /* 908*0Sstevel@tonic-gate * If the delta is not sector aligned then 909*0Sstevel@tonic-gate * read the whole block. 910*0Sstevel@tonic-gate */ 911*0Sstevel@tonic-gate if ((nb & DEV_BMASK) || (mof & DEV_BMASK)) { 912*0Sstevel@tonic-gate read_needed = 1; 913*0Sstevel@tonic-gate } 914*0Sstevel@tonic-gate 915*0Sstevel@tonic-gate /* Set sector map used in the MAPBLOCKSIZE block. */ 916*0Sstevel@tonic-gate start_sec = (mof & MAPBLOCKOFF) >> DEV_BSHIFT; 917*0Sstevel@tonic-gate end_sec = start_sec + ((nb - 1) >> DEV_BSHIFT); 918*0Sstevel@tonic-gate for (i = start_sec; i <= end_sec; i++) { 919*0Sstevel@tonic-gate secmap |= UINT16_C(1) << i; 920*0Sstevel@tonic-gate } 921*0Sstevel@tonic-gate 922*0Sstevel@tonic-gate if (me->me_dt != DT_INODE) { 923*0Sstevel@tonic-gate all_inodes = 0; 924*0Sstevel@tonic-gate } 925*0Sstevel@tonic-gate if (start_sec < first_sec) { 926*0Sstevel@tonic-gate first_sec = start_sec; 927*0Sstevel@tonic-gate } 928*0Sstevel@tonic-gate if (end_sec > last_sec) { 929*0Sstevel@tonic-gate last_sec = end_sec; 930*0Sstevel@tonic-gate } 931*0Sstevel@tonic-gate } 932*0Sstevel@tonic-gate 933*0Sstevel@tonic-gate ASSERT(secmap); 934*0Sstevel@tonic-gate ASSERT(first_sec != INT_MAX); 935*0Sstevel@tonic-gate ASSERT(last_sec != -1); 936*0Sstevel@tonic-gate 937*0Sstevel@tonic-gate if (all_inodes) { 938*0Sstevel@tonic-gate /* 939*0Sstevel@tonic-gate * Here we have a tradeoff choice. It must be better to 940*0Sstevel@tonic-gate * do 2 writes * in the same MAPBLOCKSIZE chunk, than a 941*0Sstevel@tonic-gate * read and a write. But what about 3 or more writes, versus 942*0Sstevel@tonic-gate * a read+write? * Where is the cut over? It will depend on 943*0Sstevel@tonic-gate * the track caching, scsi driver and other activity. 944*0Sstevel@tonic-gate * A unpublished tunable is defined (ufs_rw_balance) that 945*0Sstevel@tonic-gate * currently defaults to 2. 946*0Sstevel@tonic-gate */ 947*0Sstevel@tonic-gate if (!read_needed) { 948*0Sstevel@tonic-gate int count = 0, gap = 0; 949*0Sstevel@tonic-gate int sector_set; /* write needed to this sector */ 950*0Sstevel@tonic-gate 951*0Sstevel@tonic-gate /* Count the gaps (every 1 to 0 transation) */ 952*0Sstevel@tonic-gate for (i = first_sec + 1; i < last_sec; i++) { 953*0Sstevel@tonic-gate sector_set = secmap & (UINT16_C(1) << i); 954*0Sstevel@tonic-gate if (!gap && !sector_set) { 955*0Sstevel@tonic-gate gap = 1; 956*0Sstevel@tonic-gate count++; 957*0Sstevel@tonic-gate if (count > ufs_rw_balance) { 958*0Sstevel@tonic-gate read_needed = 1; 959*0Sstevel@tonic-gate break; 960*0Sstevel@tonic-gate } 961*0Sstevel@tonic-gate } else if (gap && sector_set) { 962*0Sstevel@tonic-gate gap = 0; 963*0Sstevel@tonic-gate } 964*0Sstevel@tonic-gate } 965*0Sstevel@tonic-gate } 966*0Sstevel@tonic-gate 967*0Sstevel@tonic-gate /* 968*0Sstevel@tonic-gate * Inodes commonly make up the majority (~85%) of deltas. 969*0Sstevel@tonic-gate * They cannot contain embedded user data, so its safe to 970*0Sstevel@tonic-gate * read and write them all in one IO. 971*0Sstevel@tonic-gate * But for directory entries, shadow inode data, and 972*0Sstevel@tonic-gate * quota record data the user data fragments can be embedded 973*0Sstevel@tonic-gate * betwen those metadata, and so its not safe to read, modify 974*0Sstevel@tonic-gate * then write the entire range as user asynchronous user data 975*0Sstevel@tonic-gate * writes could get overwritten with old data. 976*0Sstevel@tonic-gate * Thus we have to create a segment map of meta data that 977*0Sstevel@tonic-gate * needs to get written. 978*0Sstevel@tonic-gate * 979*0Sstevel@tonic-gate * If user data was logged then this issue would go away. 980*0Sstevel@tonic-gate */ 981*0Sstevel@tonic-gate if (read_needed) { 982*0Sstevel@tonic-gate for (i = first_sec + 1; i < last_sec; i++) { 983*0Sstevel@tonic-gate secmap |= (UINT16_C(1) << i); 984*0Sstevel@tonic-gate } 985*0Sstevel@tonic-gate } 986*0Sstevel@tonic-gate } 987*0Sstevel@tonic-gate rbp->rb_secmap = secmap; 988*0Sstevel@tonic-gate return (read_needed); 989*0Sstevel@tonic-gate } 990*0Sstevel@tonic-gate 991*0Sstevel@tonic-gate /* 992*0Sstevel@tonic-gate * Abort the load of a set of log map delta's. 993*0Sstevel@tonic-gate * ie, 994*0Sstevel@tonic-gate * Clear out all mapentries on this unit's log map 995*0Sstevel@tonic-gate * which have a tid (transaction id) equal to the 996*0Sstevel@tonic-gate * parameter tid. Walk the cancel list, taking everything 997*0Sstevel@tonic-gate * off it, too. 998*0Sstevel@tonic-gate */ 999*0Sstevel@tonic-gate static void 1000*0Sstevel@tonic-gate logmap_abort(ml_unit_t *ul, uint32_t tid) 1001*0Sstevel@tonic-gate { 1002*0Sstevel@tonic-gate struct mt_map *mtm = ul->un_logmap; /* Log map */ 1003*0Sstevel@tonic-gate mapentry_t *me, 1004*0Sstevel@tonic-gate **mep; 1005*0Sstevel@tonic-gate int i; 1006*0Sstevel@tonic-gate 1007*0Sstevel@tonic-gate ASSERT(((mtm->mtm_debug & MT_CHECK_MAP) == 0) || 1008*0Sstevel@tonic-gate map_check_linkage(mtm)); 1009*0Sstevel@tonic-gate 1010*0Sstevel@tonic-gate /* 1011*0Sstevel@tonic-gate * wait for any outstanding reads to finish; lock out future reads 1012*0Sstevel@tonic-gate */ 1013*0Sstevel@tonic-gate rw_enter(&mtm->mtm_rwlock, RW_WRITER); 1014*0Sstevel@tonic-gate 1015*0Sstevel@tonic-gate mutex_enter(&mtm->mtm_mutex); 1016*0Sstevel@tonic-gate /* Take everything off cancel list */ 1017*0Sstevel@tonic-gate while ((me = mtm->mtm_cancel) != NULL) { 1018*0Sstevel@tonic-gate mtm->mtm_cancel = me->me_cancel; 1019*0Sstevel@tonic-gate me->me_flags &= ~ME_CANCEL; 1020*0Sstevel@tonic-gate me->me_cancel = NULL; 1021*0Sstevel@tonic-gate } 1022*0Sstevel@tonic-gate 1023*0Sstevel@tonic-gate /* 1024*0Sstevel@tonic-gate * Now take out all mapentries with current tid, and committid 1025*0Sstevel@tonic-gate * as this function is called from logmap_logscan and logmap_commit 1026*0Sstevel@tonic-gate * When it is called from logmap_logscan mtm_tid == mtm_committid 1027*0Sstevel@tonic-gate * But when logmap_abort is called from logmap_commit it is 1028*0Sstevel@tonic-gate * because the log errored when trying to write the commit record, 1029*0Sstevel@tonic-gate * after the async ops have been allowed to start in top_end_sync. 1030*0Sstevel@tonic-gate * So we also need to remove all mapentries from the transaction whose 1031*0Sstevel@tonic-gate * commit failed. 1032*0Sstevel@tonic-gate */ 1033*0Sstevel@tonic-gate for (i = 0; i < mtm->mtm_nhash; i++) { 1034*0Sstevel@tonic-gate mep = &mtm->mtm_hash[i]; 1035*0Sstevel@tonic-gate while ((me = *mep) != NULL) { 1036*0Sstevel@tonic-gate if (me->me_tid == tid || 1037*0Sstevel@tonic-gate me->me_tid == mtm->mtm_committid) { 1038*0Sstevel@tonic-gate *mep = me->me_hash; 1039*0Sstevel@tonic-gate me->me_next->me_prev = me->me_prev; 1040*0Sstevel@tonic-gate me->me_prev->me_next = me->me_next; 1041*0Sstevel@tonic-gate if (!(me->me_flags & ME_USER)) { 1042*0Sstevel@tonic-gate mtm->mtm_nme--; 1043*0Sstevel@tonic-gate } 1044*0Sstevel@tonic-gate CRB_RELE(me); 1045*0Sstevel@tonic-gate kmem_cache_free(mapentry_cache, me); 1046*0Sstevel@tonic-gate continue; 1047*0Sstevel@tonic-gate } 1048*0Sstevel@tonic-gate mep = &me->me_hash; 1049*0Sstevel@tonic-gate } 1050*0Sstevel@tonic-gate } 1051*0Sstevel@tonic-gate 1052*0Sstevel@tonic-gate if (!(ul->un_flags & LDL_SCAN)) 1053*0Sstevel@tonic-gate mtm->mtm_flags |= MTM_CANCELED; 1054*0Sstevel@tonic-gate mutex_exit(&mtm->mtm_mutex); 1055*0Sstevel@tonic-gate mtm->mtm_dirty = 0; 1056*0Sstevel@tonic-gate mtm->mtm_nmet = 0; 1057*0Sstevel@tonic-gate rw_exit(&mtm->mtm_rwlock); 1058*0Sstevel@tonic-gate 1059*0Sstevel@tonic-gate ASSERT(((mtm->mtm_debug & MT_CHECK_MAP) == 0) || 1060*0Sstevel@tonic-gate map_check_linkage(mtm)); 1061*0Sstevel@tonic-gate } 1062*0Sstevel@tonic-gate 1063*0Sstevel@tonic-gate static void 1064*0Sstevel@tonic-gate logmap_wait_space(mt_map_t *mtm, ml_unit_t *ul, mapentry_t *me) 1065*0Sstevel@tonic-gate { 1066*0Sstevel@tonic-gate ASSERT(MUTEX_HELD(&ul->un_log_mutex)); 1067*0Sstevel@tonic-gate 1068*0Sstevel@tonic-gate while (!ldl_has_space(ul, me)) { 1069*0Sstevel@tonic-gate ASSERT(!(ul->un_flags & LDL_NOROLL)); 1070*0Sstevel@tonic-gate mutex_exit(&ul->un_log_mutex); 1071*0Sstevel@tonic-gate logmap_forceroll(mtm); 1072*0Sstevel@tonic-gate mutex_enter(&ul->un_log_mutex); 1073*0Sstevel@tonic-gate if (ul->un_flags & LDL_ERROR) 1074*0Sstevel@tonic-gate break; 1075*0Sstevel@tonic-gate } 1076*0Sstevel@tonic-gate 1077*0Sstevel@tonic-gate ASSERT(MUTEX_HELD(&ul->un_log_mutex)); 1078*0Sstevel@tonic-gate } 1079*0Sstevel@tonic-gate 1080*0Sstevel@tonic-gate /* 1081*0Sstevel@tonic-gate * put a list of deltas into a logmap 1082*0Sstevel@tonic-gate * If va == NULL, don't write to the log. 1083*0Sstevel@tonic-gate */ 1084*0Sstevel@tonic-gate void 1085*0Sstevel@tonic-gate logmap_add( 1086*0Sstevel@tonic-gate ml_unit_t *ul, 1087*0Sstevel@tonic-gate char *va, /* Ptr to buf w/deltas & data */ 1088*0Sstevel@tonic-gate offset_t vamof, /* Offset on master of buf start */ 1089*0Sstevel@tonic-gate mapentry_t *melist) /* Entries to add */ 1090*0Sstevel@tonic-gate { 1091*0Sstevel@tonic-gate offset_t mof; 1092*0Sstevel@tonic-gate off_t nb; 1093*0Sstevel@tonic-gate mapentry_t *me; 1094*0Sstevel@tonic-gate mapentry_t **mep; 1095*0Sstevel@tonic-gate mapentry_t **savmep; 1096*0Sstevel@tonic-gate uint32_t tid; 1097*0Sstevel@tonic-gate mt_map_t *mtm = ul->un_logmap; 1098*0Sstevel@tonic-gate 1099*0Sstevel@tonic-gate mutex_enter(&ul->un_log_mutex); 1100*0Sstevel@tonic-gate if (va) 1101*0Sstevel@tonic-gate logmap_wait_space(mtm, ul, melist); 1102*0Sstevel@tonic-gate 1103*0Sstevel@tonic-gate ASSERT(((mtm->mtm_debug & MT_CHECK_MAP) == 0) || 1104*0Sstevel@tonic-gate map_check_linkage(mtm)); 1105*0Sstevel@tonic-gate 1106*0Sstevel@tonic-gate mtm->mtm_ref = 1; 1107*0Sstevel@tonic-gate mtm->mtm_dirty++; 1108*0Sstevel@tonic-gate tid = mtm->mtm_tid; 1109*0Sstevel@tonic-gate while (melist) { 1110*0Sstevel@tonic-gate mof = melist->me_mof; 1111*0Sstevel@tonic-gate nb = melist->me_nb; 1112*0Sstevel@tonic-gate 1113*0Sstevel@tonic-gate /* 1114*0Sstevel@tonic-gate * search for overlaping entries 1115*0Sstevel@tonic-gate */ 1116*0Sstevel@tonic-gate savmep = mep = MAP_HASH(mof, mtm); 1117*0Sstevel@tonic-gate mutex_enter(&mtm->mtm_mutex); 1118*0Sstevel@tonic-gate while ((me = *mep) != 0) { 1119*0Sstevel@tonic-gate /* 1120*0Sstevel@tonic-gate * Data consumes old map entry; cancel map entry. 1121*0Sstevel@tonic-gate * Take care when we replace an old map entry 1122*0Sstevel@tonic-gate * which carries quota information with a newer entry 1123*0Sstevel@tonic-gate * which does not. In that case the push function 1124*0Sstevel@tonic-gate * would not be called to clean up the dquot structure. 1125*0Sstevel@tonic-gate * This would be found later by invalidatedq() causing 1126*0Sstevel@tonic-gate * a panic when the filesystem in unmounted. 1127*0Sstevel@tonic-gate * We clean up the dquot manually and then replace 1128*0Sstevel@tonic-gate * the map entry. 1129*0Sstevel@tonic-gate */ 1130*0Sstevel@tonic-gate if (MEwithinDATA(me, mof, nb) && 1131*0Sstevel@tonic-gate ((me->me_flags & (ME_ROLL|ME_CANCEL)) == 0)) { 1132*0Sstevel@tonic-gate if (tid == me->me_tid && 1133*0Sstevel@tonic-gate ((me->me_flags & ME_AGE) == 0)) { 1134*0Sstevel@tonic-gate *mep = me->me_hash; 1135*0Sstevel@tonic-gate me->me_next->me_prev = me->me_prev; 1136*0Sstevel@tonic-gate me->me_prev->me_next = me->me_next; 1137*0Sstevel@tonic-gate ASSERT(!(me->me_flags & ME_USER)); 1138*0Sstevel@tonic-gate mtm->mtm_nme--; 1139*0Sstevel@tonic-gate /* 1140*0Sstevel@tonic-gate * Special case if the mapentry 1141*0Sstevel@tonic-gate * carries a dquot and a push function. 1142*0Sstevel@tonic-gate * We have to clean up the quota info 1143*0Sstevel@tonic-gate * before replacing the mapentry. 1144*0Sstevel@tonic-gate */ 1145*0Sstevel@tonic-gate if (me->me_dt == DT_QR) 1146*0Sstevel@tonic-gate HANDLE_DQUOT(me, melist); 1147*0Sstevel@tonic-gate 1148*0Sstevel@tonic-gate kmem_cache_free(mapentry_cache, me); 1149*0Sstevel@tonic-gate continue; 1150*0Sstevel@tonic-gate } 1151*0Sstevel@tonic-gate me->me_cancel = mtm->mtm_cancel; 1152*0Sstevel@tonic-gate mtm->mtm_cancel = me; 1153*0Sstevel@tonic-gate me->me_flags |= ME_CANCEL; 1154*0Sstevel@tonic-gate } 1155*0Sstevel@tonic-gate mep = &(*mep)->me_hash; 1156*0Sstevel@tonic-gate } 1157*0Sstevel@tonic-gate mutex_exit(&mtm->mtm_mutex); 1158*0Sstevel@tonic-gate 1159*0Sstevel@tonic-gate /* 1160*0Sstevel@tonic-gate * remove from list 1161*0Sstevel@tonic-gate */ 1162*0Sstevel@tonic-gate me = melist; 1163*0Sstevel@tonic-gate melist = melist->me_hash; 1164*0Sstevel@tonic-gate me->me_flags &= ~ME_LIST; 1165*0Sstevel@tonic-gate /* 1166*0Sstevel@tonic-gate * If va != NULL, put in the log. 1167*0Sstevel@tonic-gate */ 1168*0Sstevel@tonic-gate if (va) 1169*0Sstevel@tonic-gate ldl_write(ul, va, vamof, me); 1170*0Sstevel@tonic-gate if (ul->un_flags & LDL_ERROR) { 1171*0Sstevel@tonic-gate kmem_cache_free(mapentry_cache, me); 1172*0Sstevel@tonic-gate continue; 1173*0Sstevel@tonic-gate } 1174*0Sstevel@tonic-gate ASSERT((va == NULL) || 1175*0Sstevel@tonic-gate ((mtm->mtm_debug & MT_LOG_WRITE_CHECK) == 0) || 1176*0Sstevel@tonic-gate map_check_ldl_write(ul, va, vamof, me)); 1177*0Sstevel@tonic-gate 1178*0Sstevel@tonic-gate /* 1179*0Sstevel@tonic-gate * put on hash 1180*0Sstevel@tonic-gate */ 1181*0Sstevel@tonic-gate mutex_enter(&mtm->mtm_mutex); 1182*0Sstevel@tonic-gate me->me_hash = *savmep; 1183*0Sstevel@tonic-gate *savmep = me; 1184*0Sstevel@tonic-gate me->me_next = (mapentry_t *)mtm; 1185*0Sstevel@tonic-gate me->me_prev = mtm->mtm_prev; 1186*0Sstevel@tonic-gate mtm->mtm_prev->me_next = me; 1187*0Sstevel@tonic-gate mtm->mtm_prev = me; 1188*0Sstevel@tonic-gate me->me_flags |= ME_HASH; 1189*0Sstevel@tonic-gate me->me_tid = tid; 1190*0Sstevel@tonic-gate me->me_age = mtm->mtm_age++; 1191*0Sstevel@tonic-gate mtm->mtm_nme++; 1192*0Sstevel@tonic-gate mtm->mtm_nmet++; 1193*0Sstevel@tonic-gate mutex_exit(&mtm->mtm_mutex); 1194*0Sstevel@tonic-gate } 1195*0Sstevel@tonic-gate 1196*0Sstevel@tonic-gate ASSERT(((mtm->mtm_debug & MT_CHECK_MAP) == 0) || 1197*0Sstevel@tonic-gate map_check_linkage(mtm)); 1198*0Sstevel@tonic-gate mutex_exit(&ul->un_log_mutex); 1199*0Sstevel@tonic-gate } 1200*0Sstevel@tonic-gate 1201*0Sstevel@tonic-gate /* 1202*0Sstevel@tonic-gate * Add the delta(s) into the log. 1203*0Sstevel@tonic-gate * Create one cached roll buffer logmap entry, and reference count the 1204*0Sstevel@tonic-gate * number of mapentries refering to it. 1205*0Sstevel@tonic-gate * Cancel previous logmap entries. 1206*0Sstevel@tonic-gate * logmap_add is tolerant of failure to allocate a cached roll buffer. 1207*0Sstevel@tonic-gate */ 1208*0Sstevel@tonic-gate void 1209*0Sstevel@tonic-gate logmap_add_buf( 1210*0Sstevel@tonic-gate ml_unit_t *ul, 1211*0Sstevel@tonic-gate char *va, /* Ptr to buf w/deltas & data */ 1212*0Sstevel@tonic-gate offset_t bufmof, /* Offset on master of buf start */ 1213*0Sstevel@tonic-gate mapentry_t *melist, /* Entries to add */ 1214*0Sstevel@tonic-gate caddr_t buf, /* Buffer containing delta(s) */ 1215*0Sstevel@tonic-gate uint32_t bufsz) /* Size of buf */ 1216*0Sstevel@tonic-gate { 1217*0Sstevel@tonic-gate offset_t mof; 1218*0Sstevel@tonic-gate offset_t vamof = bufmof + (va - buf); 1219*0Sstevel@tonic-gate off_t nb; 1220*0Sstevel@tonic-gate mapentry_t *me; 1221*0Sstevel@tonic-gate mapentry_t **mep; 1222*0Sstevel@tonic-gate mapentry_t **savmep; 1223*0Sstevel@tonic-gate uint32_t tid; 1224*0Sstevel@tonic-gate mt_map_t *mtm = ul->un_logmap; 1225*0Sstevel@tonic-gate crb_t *crb; 1226*0Sstevel@tonic-gate crb_t *crbsav = NULL; 1227*0Sstevel@tonic-gate 1228*0Sstevel@tonic-gate ASSERT((bufsz & DEV_BMASK) == 0); 1229*0Sstevel@tonic-gate mutex_enter(&ul->un_log_mutex); 1230*0Sstevel@tonic-gate logmap_wait_space(mtm, ul, melist); 1231*0Sstevel@tonic-gate 1232*0Sstevel@tonic-gate ASSERT(((mtm->mtm_debug & MT_CHECK_MAP) == 0) || 1233*0Sstevel@tonic-gate map_check_linkage(mtm)); 1234*0Sstevel@tonic-gate 1235*0Sstevel@tonic-gate mtm->mtm_ref = 1; 1236*0Sstevel@tonic-gate mtm->mtm_dirty++; 1237*0Sstevel@tonic-gate tid = mtm->mtm_tid; 1238*0Sstevel@tonic-gate while (melist) { 1239*0Sstevel@tonic-gate mof = melist->me_mof; 1240*0Sstevel@tonic-gate nb = melist->me_nb; 1241*0Sstevel@tonic-gate 1242*0Sstevel@tonic-gate /* 1243*0Sstevel@tonic-gate * search for overlapping entries 1244*0Sstevel@tonic-gate */ 1245*0Sstevel@tonic-gate savmep = mep = MAP_HASH(mof, mtm); 1246*0Sstevel@tonic-gate mutex_enter(&mtm->mtm_mutex); 1247*0Sstevel@tonic-gate while ((me = *mep) != 0) { 1248*0Sstevel@tonic-gate /* 1249*0Sstevel@tonic-gate * Data consumes old map entry; cancel map entry. 1250*0Sstevel@tonic-gate * Take care when we replace an old map entry 1251*0Sstevel@tonic-gate * which carries quota information with a newer entry 1252*0Sstevel@tonic-gate * which does not. In that case the push function 1253*0Sstevel@tonic-gate * would not be called to clean up the dquot structure. 1254*0Sstevel@tonic-gate * This would be found later by invalidatedq() causing 1255*0Sstevel@tonic-gate * a panic when the filesystem in unmounted. 1256*0Sstevel@tonic-gate * We clean up the dquot manually and then replace 1257*0Sstevel@tonic-gate * the map entry. 1258*0Sstevel@tonic-gate */ 1259*0Sstevel@tonic-gate crb = me->me_crb; 1260*0Sstevel@tonic-gate if (MEwithinDATA(me, mof, nb) && 1261*0Sstevel@tonic-gate ((me->me_flags & (ME_ROLL|ME_CANCEL)) == 0)) { 1262*0Sstevel@tonic-gate if (tid == me->me_tid && 1263*0Sstevel@tonic-gate ((me->me_flags & ME_AGE) == 0)) { 1264*0Sstevel@tonic-gate *mep = me->me_hash; 1265*0Sstevel@tonic-gate me->me_next->me_prev = me->me_prev; 1266*0Sstevel@tonic-gate me->me_prev->me_next = me->me_next; 1267*0Sstevel@tonic-gate ASSERT(!(me->me_flags & ME_USER)); 1268*0Sstevel@tonic-gate mtm->mtm_nme--; 1269*0Sstevel@tonic-gate /* 1270*0Sstevel@tonic-gate * Special case if the mapentry 1271*0Sstevel@tonic-gate * carries a dquot and a push function. 1272*0Sstevel@tonic-gate * We have to clean up the quota info 1273*0Sstevel@tonic-gate * before replacing the mapentry. 1274*0Sstevel@tonic-gate */ 1275*0Sstevel@tonic-gate if (me->me_dt == DT_QR) 1276*0Sstevel@tonic-gate HANDLE_DQUOT(me, melist); 1277*0Sstevel@tonic-gate 1278*0Sstevel@tonic-gate /* 1279*0Sstevel@tonic-gate * If this soon to be deleted mapentry 1280*0Sstevel@tonic-gate * has a suitable roll buffer then 1281*0Sstevel@tonic-gate * re-use it. 1282*0Sstevel@tonic-gate */ 1283*0Sstevel@tonic-gate if (crb && (--crb->c_refcnt == 0)) { 1284*0Sstevel@tonic-gate if (crbsav || 1285*0Sstevel@tonic-gate (crb->c_nb != bufsz)) { 1286*0Sstevel@tonic-gate CRB_FREE(crb, me); 1287*0Sstevel@tonic-gate } else { 1288*0Sstevel@tonic-gate bcopy(buf, crb->c_buf, 1289*0Sstevel@tonic-gate bufsz); 1290*0Sstevel@tonic-gate crb->c_invalid = 0; 1291*0Sstevel@tonic-gate crb->c_mof = bufmof; 1292*0Sstevel@tonic-gate crbsav = crb; 1293*0Sstevel@tonic-gate me->me_crb = NULL; 1294*0Sstevel@tonic-gate } 1295*0Sstevel@tonic-gate } 1296*0Sstevel@tonic-gate kmem_cache_free(mapentry_cache, me); 1297*0Sstevel@tonic-gate continue; 1298*0Sstevel@tonic-gate } 1299*0Sstevel@tonic-gate me->me_cancel = mtm->mtm_cancel; 1300*0Sstevel@tonic-gate mtm->mtm_cancel = me; 1301*0Sstevel@tonic-gate me->me_flags |= ME_CANCEL; 1302*0Sstevel@tonic-gate } 1303*0Sstevel@tonic-gate 1304*0Sstevel@tonic-gate /* 1305*0Sstevel@tonic-gate * Inode deltas within the same fs block come 1306*0Sstevel@tonic-gate * in individually as separate calls to logmap_add(). 1307*0Sstevel@tonic-gate * All others come in as one call. So check for an 1308*0Sstevel@tonic-gate * existing entry where we can re-use the crb. 1309*0Sstevel@tonic-gate */ 1310*0Sstevel@tonic-gate if ((me->me_dt == DT_INODE) && (tid == me->me_tid) && 1311*0Sstevel@tonic-gate !crbsav && crb && 1312*0Sstevel@tonic-gate WITHIN(mof, nb, crb->c_mof, crb->c_nb)) { 1313*0Sstevel@tonic-gate ASSERT(crb->c_mof == bufmof); 1314*0Sstevel@tonic-gate ASSERT(crb->c_nb == bufsz); 1315*0Sstevel@tonic-gate bcopy(buf, crb->c_buf, bufsz); 1316*0Sstevel@tonic-gate crbsav = crb; 1317*0Sstevel@tonic-gate } 1318*0Sstevel@tonic-gate mep = &(*mep)->me_hash; 1319*0Sstevel@tonic-gate } 1320*0Sstevel@tonic-gate mutex_exit(&mtm->mtm_mutex); 1321*0Sstevel@tonic-gate 1322*0Sstevel@tonic-gate /* 1323*0Sstevel@tonic-gate * If we don't already have a crb then allocate one 1324*0Sstevel@tonic-gate * and copy the incoming buffer. Only do this once 1325*0Sstevel@tonic-gate * for all the incoming deltas. 1326*0Sstevel@tonic-gate */ 1327*0Sstevel@tonic-gate if ((crbsav == NULL) && (melist->me_dt != DT_ABZERO)) { 1328*0Sstevel@tonic-gate /* 1329*0Sstevel@tonic-gate * Only use a cached roll buffer if we 1330*0Sstevel@tonic-gate * have enough memory, and check for failures. 1331*0Sstevel@tonic-gate */ 1332*0Sstevel@tonic-gate if (((ufs_crb_size + bufsz) < ufs_crb_limit) && 1333*0Sstevel@tonic-gate (kmem_avail() > bufsz)) { 1334*0Sstevel@tonic-gate crbsav = kmem_alloc(sizeof (crb_t), KM_NOSLEEP); 1335*0Sstevel@tonic-gate } else { 1336*0Sstevel@tonic-gate ufs_crb_alloc_fails++; 1337*0Sstevel@tonic-gate } 1338*0Sstevel@tonic-gate if (crbsav) { 1339*0Sstevel@tonic-gate crbsav->c_buf = kmem_alloc(bufsz, KM_NOSLEEP); 1340*0Sstevel@tonic-gate if (crbsav->c_buf) { 1341*0Sstevel@tonic-gate atomic_add_64(&ufs_crb_size, 1342*0Sstevel@tonic-gate (uint64_t)bufsz); 1343*0Sstevel@tonic-gate if (ufs_crb_size > ufs_crb_max_size) { 1344*0Sstevel@tonic-gate ufs_crb_max_size = ufs_crb_size; 1345*0Sstevel@tonic-gate } 1346*0Sstevel@tonic-gate bcopy(buf, crbsav->c_buf, bufsz); 1347*0Sstevel@tonic-gate crbsav->c_nb = bufsz; 1348*0Sstevel@tonic-gate crbsav->c_refcnt = 0; 1349*0Sstevel@tonic-gate crbsav->c_invalid = 0; 1350*0Sstevel@tonic-gate ASSERT((bufmof & DEV_BMASK) == 0); 1351*0Sstevel@tonic-gate crbsav->c_mof = bufmof; 1352*0Sstevel@tonic-gate } else { 1353*0Sstevel@tonic-gate kmem_free(crbsav, sizeof (crb_t)); 1354*0Sstevel@tonic-gate crbsav = NULL; 1355*0Sstevel@tonic-gate } 1356*0Sstevel@tonic-gate } 1357*0Sstevel@tonic-gate } 1358*0Sstevel@tonic-gate 1359*0Sstevel@tonic-gate /* 1360*0Sstevel@tonic-gate * remove from list 1361*0Sstevel@tonic-gate */ 1362*0Sstevel@tonic-gate me = melist; 1363*0Sstevel@tonic-gate melist = melist->me_hash; 1364*0Sstevel@tonic-gate me->me_flags &= ~ME_LIST; 1365*0Sstevel@tonic-gate me->me_crb = crbsav; 1366*0Sstevel@tonic-gate if (crbsav) { 1367*0Sstevel@tonic-gate crbsav->c_refcnt++; 1368*0Sstevel@tonic-gate } 1369*0Sstevel@tonic-gate crbsav = NULL; 1370*0Sstevel@tonic-gate 1371*0Sstevel@tonic-gate ASSERT(va); 1372*0Sstevel@tonic-gate ldl_write(ul, va, vamof, me); /* add to on-disk log */ 1373*0Sstevel@tonic-gate if (ul->un_flags & LDL_ERROR) { 1374*0Sstevel@tonic-gate CRB_RELE(me); 1375*0Sstevel@tonic-gate kmem_cache_free(mapentry_cache, me); 1376*0Sstevel@tonic-gate continue; 1377*0Sstevel@tonic-gate } 1378*0Sstevel@tonic-gate ASSERT(((mtm->mtm_debug & MT_LOG_WRITE_CHECK) == 0) || 1379*0Sstevel@tonic-gate map_check_ldl_write(ul, va, vamof, me)); 1380*0Sstevel@tonic-gate 1381*0Sstevel@tonic-gate /* 1382*0Sstevel@tonic-gate * put on hash 1383*0Sstevel@tonic-gate */ 1384*0Sstevel@tonic-gate mutex_enter(&mtm->mtm_mutex); 1385*0Sstevel@tonic-gate me->me_hash = *savmep; 1386*0Sstevel@tonic-gate *savmep = me; 1387*0Sstevel@tonic-gate me->me_next = (mapentry_t *)mtm; 1388*0Sstevel@tonic-gate me->me_prev = mtm->mtm_prev; 1389*0Sstevel@tonic-gate mtm->mtm_prev->me_next = me; 1390*0Sstevel@tonic-gate mtm->mtm_prev = me; 1391*0Sstevel@tonic-gate me->me_flags |= ME_HASH; 1392*0Sstevel@tonic-gate me->me_tid = tid; 1393*0Sstevel@tonic-gate me->me_age = mtm->mtm_age++; 1394*0Sstevel@tonic-gate mtm->mtm_nme++; 1395*0Sstevel@tonic-gate mtm->mtm_nmet++; 1396*0Sstevel@tonic-gate mutex_exit(&mtm->mtm_mutex); 1397*0Sstevel@tonic-gate } 1398*0Sstevel@tonic-gate 1399*0Sstevel@tonic-gate ASSERT(((mtm->mtm_debug & MT_CHECK_MAP) == 0) || 1400*0Sstevel@tonic-gate map_check_linkage(mtm)); 1401*0Sstevel@tonic-gate mutex_exit(&ul->un_log_mutex); 1402*0Sstevel@tonic-gate } 1403*0Sstevel@tonic-gate 1404*0Sstevel@tonic-gate /* 1405*0Sstevel@tonic-gate * free up any cancelled deltas 1406*0Sstevel@tonic-gate */ 1407*0Sstevel@tonic-gate void 1408*0Sstevel@tonic-gate logmap_free_cancel(mt_map_t *mtm, mapentry_t **cancelhead) 1409*0Sstevel@tonic-gate { 1410*0Sstevel@tonic-gate int dolock = 0; 1411*0Sstevel@tonic-gate mapentry_t *me; 1412*0Sstevel@tonic-gate mapentry_t **mep; 1413*0Sstevel@tonic-gate 1414*0Sstevel@tonic-gate ASSERT(((mtm->mtm_debug & MT_CHECK_MAP) == 0) || 1415*0Sstevel@tonic-gate map_check_linkage(mtm)); 1416*0Sstevel@tonic-gate 1417*0Sstevel@tonic-gate again: 1418*0Sstevel@tonic-gate if (dolock) 1419*0Sstevel@tonic-gate rw_enter(&mtm->mtm_rwlock, RW_WRITER); 1420*0Sstevel@tonic-gate 1421*0Sstevel@tonic-gate /* 1422*0Sstevel@tonic-gate * At EOT, cancel the indicated deltas 1423*0Sstevel@tonic-gate */ 1424*0Sstevel@tonic-gate mutex_enter(&mtm->mtm_mutex); 1425*0Sstevel@tonic-gate if (mtm->mtm_flags & MTM_CANCELED) { 1426*0Sstevel@tonic-gate mtm->mtm_flags &= ~MTM_CANCELED; 1427*0Sstevel@tonic-gate ASSERT(dolock == 0); 1428*0Sstevel@tonic-gate mutex_exit(&mtm->mtm_mutex); 1429*0Sstevel@tonic-gate return; 1430*0Sstevel@tonic-gate } 1431*0Sstevel@tonic-gate 1432*0Sstevel@tonic-gate while ((me = *cancelhead) != NULL) { 1433*0Sstevel@tonic-gate /* 1434*0Sstevel@tonic-gate * roll forward or read collision; wait and try again 1435*0Sstevel@tonic-gate */ 1436*0Sstevel@tonic-gate if (me->me_flags & ME_AGE) { 1437*0Sstevel@tonic-gate ASSERT(dolock == 0); 1438*0Sstevel@tonic-gate mutex_exit(&mtm->mtm_mutex); 1439*0Sstevel@tonic-gate dolock = 1; 1440*0Sstevel@tonic-gate goto again; 1441*0Sstevel@tonic-gate } 1442*0Sstevel@tonic-gate /* 1443*0Sstevel@tonic-gate * remove from cancel list 1444*0Sstevel@tonic-gate */ 1445*0Sstevel@tonic-gate *cancelhead = me->me_cancel; 1446*0Sstevel@tonic-gate me->me_cancel = NULL; 1447*0Sstevel@tonic-gate me->me_flags &= ~(ME_CANCEL); 1448*0Sstevel@tonic-gate 1449*0Sstevel@tonic-gate /* 1450*0Sstevel@tonic-gate * logmap_remove_roll handles ME_ROLL entries later 1451*0Sstevel@tonic-gate * we leave them around for logmap_iscancel 1452*0Sstevel@tonic-gate * XXX is this necessary? 1453*0Sstevel@tonic-gate */ 1454*0Sstevel@tonic-gate if (me->me_flags & ME_ROLL) 1455*0Sstevel@tonic-gate continue; 1456*0Sstevel@tonic-gate 1457*0Sstevel@tonic-gate /* 1458*0Sstevel@tonic-gate * remove from hash (if necessary) 1459*0Sstevel@tonic-gate */ 1460*0Sstevel@tonic-gate if (me->me_flags & ME_HASH) { 1461*0Sstevel@tonic-gate mep = MAP_HASH(me->me_mof, mtm); 1462*0Sstevel@tonic-gate while (*mep) { 1463*0Sstevel@tonic-gate if (*mep == me) { 1464*0Sstevel@tonic-gate *mep = me->me_hash; 1465*0Sstevel@tonic-gate me->me_next->me_prev = me->me_prev; 1466*0Sstevel@tonic-gate me->me_prev->me_next = me->me_next; 1467*0Sstevel@tonic-gate me->me_flags &= ~(ME_HASH); 1468*0Sstevel@tonic-gate if (!(me->me_flags & ME_USER)) { 1469*0Sstevel@tonic-gate mtm->mtm_nme--; 1470*0Sstevel@tonic-gate } 1471*0Sstevel@tonic-gate break; 1472*0Sstevel@tonic-gate } else 1473*0Sstevel@tonic-gate mep = &(*mep)->me_hash; 1474*0Sstevel@tonic-gate } 1475*0Sstevel@tonic-gate } 1476*0Sstevel@tonic-gate /* 1477*0Sstevel@tonic-gate * put the entry on the free list 1478*0Sstevel@tonic-gate */ 1479*0Sstevel@tonic-gate CRB_RELE(me); 1480*0Sstevel@tonic-gate kmem_cache_free(mapentry_cache, me); 1481*0Sstevel@tonic-gate } 1482*0Sstevel@tonic-gate mutex_exit(&mtm->mtm_mutex); 1483*0Sstevel@tonic-gate if (dolock) 1484*0Sstevel@tonic-gate rw_exit(&mtm->mtm_rwlock); 1485*0Sstevel@tonic-gate 1486*0Sstevel@tonic-gate ASSERT(((mtm->mtm_debug & MT_CHECK_MAP) == 0) || 1487*0Sstevel@tonic-gate map_check_linkage(mtm)); 1488*0Sstevel@tonic-gate } 1489*0Sstevel@tonic-gate 1490*0Sstevel@tonic-gate 1491*0Sstevel@tonic-gate void 1492*0Sstevel@tonic-gate logmap_commit(ml_unit_t *ul, uint32_t tid) 1493*0Sstevel@tonic-gate { 1494*0Sstevel@tonic-gate mapentry_t me; 1495*0Sstevel@tonic-gate mt_map_t *mtm = ul->un_logmap; 1496*0Sstevel@tonic-gate 1497*0Sstevel@tonic-gate 1498*0Sstevel@tonic-gate ASSERT(MUTEX_HELD(&ul->un_log_mutex)); 1499*0Sstevel@tonic-gate 1500*0Sstevel@tonic-gate /* 1501*0Sstevel@tonic-gate * async'ly write a commit rec into the log 1502*0Sstevel@tonic-gate */ 1503*0Sstevel@tonic-gate if (mtm->mtm_dirty) { 1504*0Sstevel@tonic-gate /* 1505*0Sstevel@tonic-gate * put commit record into log 1506*0Sstevel@tonic-gate */ 1507*0Sstevel@tonic-gate me.me_mof = mtm->mtm_tid; 1508*0Sstevel@tonic-gate me.me_dt = DT_COMMIT; 1509*0Sstevel@tonic-gate me.me_nb = 0; 1510*0Sstevel@tonic-gate me.me_hash = NULL; 1511*0Sstevel@tonic-gate logmap_wait_space(mtm, ul, &me); 1512*0Sstevel@tonic-gate ldl_write(ul, NULL, (offset_t)0, &me); 1513*0Sstevel@tonic-gate ldl_round_commit(ul); 1514*0Sstevel@tonic-gate 1515*0Sstevel@tonic-gate /* 1516*0Sstevel@tonic-gate * abort on error; else reset dirty flag 1517*0Sstevel@tonic-gate */ 1518*0Sstevel@tonic-gate if (ul->un_flags & LDL_ERROR) 1519*0Sstevel@tonic-gate logmap_abort(ul, tid); 1520*0Sstevel@tonic-gate else { 1521*0Sstevel@tonic-gate mtm->mtm_dirty = 0; 1522*0Sstevel@tonic-gate mtm->mtm_nmet = 0; 1523*0Sstevel@tonic-gate mtm->mtm_cfrags = 0; 1524*0Sstevel@tonic-gate } 1525*0Sstevel@tonic-gate /* push commit */ 1526*0Sstevel@tonic-gate ldl_push_commit(ul); 1527*0Sstevel@tonic-gate } 1528*0Sstevel@tonic-gate } 1529*0Sstevel@tonic-gate 1530*0Sstevel@tonic-gate void 1531*0Sstevel@tonic-gate logmap_sethead(mt_map_t *mtm, ml_unit_t *ul) 1532*0Sstevel@tonic-gate { 1533*0Sstevel@tonic-gate off_t lof; 1534*0Sstevel@tonic-gate uint32_t tid; 1535*0Sstevel@tonic-gate mapentry_t *me; 1536*0Sstevel@tonic-gate 1537*0Sstevel@tonic-gate /* 1538*0Sstevel@tonic-gate * move the head forward so the log knows how full it is 1539*0Sstevel@tonic-gate * Make sure to skip any mapentry whose me_lof is 0, these 1540*0Sstevel@tonic-gate * are just place holders for DT_CANCELED freed user blocks 1541*0Sstevel@tonic-gate * for the current moby. 1542*0Sstevel@tonic-gate */ 1543*0Sstevel@tonic-gate mutex_enter(&ul->un_log_mutex); 1544*0Sstevel@tonic-gate mutex_enter(&mtm->mtm_mutex); 1545*0Sstevel@tonic-gate me = mtm->mtm_next; 1546*0Sstevel@tonic-gate while (me != (mapentry_t *)mtm && me->me_lof == 0) { 1547*0Sstevel@tonic-gate me = me->me_next; 1548*0Sstevel@tonic-gate } 1549*0Sstevel@tonic-gate 1550*0Sstevel@tonic-gate if (me == (mapentry_t *)mtm) 1551*0Sstevel@tonic-gate lof = -1; 1552*0Sstevel@tonic-gate else { 1553*0Sstevel@tonic-gate lof = me->me_lof; 1554*0Sstevel@tonic-gate tid = me->me_tid; 1555*0Sstevel@tonic-gate } 1556*0Sstevel@tonic-gate mutex_exit(&mtm->mtm_mutex); 1557*0Sstevel@tonic-gate ldl_sethead(ul, lof, tid); 1558*0Sstevel@tonic-gate if (lof == -1) 1559*0Sstevel@tonic-gate mtm->mtm_age = 0; 1560*0Sstevel@tonic-gate mutex_exit(&ul->un_log_mutex); 1561*0Sstevel@tonic-gate } 1562*0Sstevel@tonic-gate 1563*0Sstevel@tonic-gate void 1564*0Sstevel@tonic-gate logmap_settail(mt_map_t *mtm, ml_unit_t *ul) 1565*0Sstevel@tonic-gate { 1566*0Sstevel@tonic-gate off_t lof; 1567*0Sstevel@tonic-gate size_t nb; 1568*0Sstevel@tonic-gate 1569*0Sstevel@tonic-gate /* 1570*0Sstevel@tonic-gate * set the tail after the logmap_abort 1571*0Sstevel@tonic-gate */ 1572*0Sstevel@tonic-gate mutex_enter(&ul->un_log_mutex); 1573*0Sstevel@tonic-gate mutex_enter(&mtm->mtm_mutex); 1574*0Sstevel@tonic-gate if (mtm->mtm_prev == (mapentry_t *)mtm) 1575*0Sstevel@tonic-gate lof = -1; 1576*0Sstevel@tonic-gate else { 1577*0Sstevel@tonic-gate /* 1578*0Sstevel@tonic-gate * set the tail to the end of the last commit 1579*0Sstevel@tonic-gate */ 1580*0Sstevel@tonic-gate lof = mtm->mtm_tail_lof; 1581*0Sstevel@tonic-gate nb = mtm->mtm_tail_nb; 1582*0Sstevel@tonic-gate } 1583*0Sstevel@tonic-gate mutex_exit(&mtm->mtm_mutex); 1584*0Sstevel@tonic-gate ldl_settail(ul, lof, nb); 1585*0Sstevel@tonic-gate mutex_exit(&ul->un_log_mutex); 1586*0Sstevel@tonic-gate } 1587*0Sstevel@tonic-gate 1588*0Sstevel@tonic-gate /* 1589*0Sstevel@tonic-gate * when reseting a device; roll the log until every 1590*0Sstevel@tonic-gate * delta has been rolled forward 1591*0Sstevel@tonic-gate */ 1592*0Sstevel@tonic-gate void 1593*0Sstevel@tonic-gate logmap_roll_dev(ml_unit_t *ul) 1594*0Sstevel@tonic-gate { 1595*0Sstevel@tonic-gate mt_map_t *mtm = ul->un_logmap; 1596*0Sstevel@tonic-gate mapentry_t *me; 1597*0Sstevel@tonic-gate ufsvfs_t *ufsvfsp = ul->un_ufsvfs; 1598*0Sstevel@tonic-gate 1599*0Sstevel@tonic-gate again: 1600*0Sstevel@tonic-gate ASSERT(((mtm->mtm_debug & MT_CHECK_MAP) == 0) || 1601*0Sstevel@tonic-gate map_check_linkage(mtm)); 1602*0Sstevel@tonic-gate if (ul->un_flags & (LDL_ERROR|LDL_NOROLL)) 1603*0Sstevel@tonic-gate return; 1604*0Sstevel@tonic-gate 1605*0Sstevel@tonic-gate /* 1606*0Sstevel@tonic-gate * look for deltas 1607*0Sstevel@tonic-gate */ 1608*0Sstevel@tonic-gate mutex_enter(&mtm->mtm_mutex); 1609*0Sstevel@tonic-gate for (me = mtm->mtm_next; me != (mapentry_t *)mtm; me = me->me_next) { 1610*0Sstevel@tonic-gate if (me->me_flags & ME_ROLL) 1611*0Sstevel@tonic-gate break; 1612*0Sstevel@tonic-gate if (me->me_tid == mtm->mtm_tid) 1613*0Sstevel@tonic-gate continue; 1614*0Sstevel@tonic-gate if (me->me_tid == mtm->mtm_committid) 1615*0Sstevel@tonic-gate continue; 1616*0Sstevel@tonic-gate break; 1617*0Sstevel@tonic-gate } 1618*0Sstevel@tonic-gate 1619*0Sstevel@tonic-gate /* 1620*0Sstevel@tonic-gate * found a delta; kick the roll thread 1621*0Sstevel@tonic-gate * but only if the thread is running... (jmh) 1622*0Sstevel@tonic-gate */ 1623*0Sstevel@tonic-gate if (me != (mapentry_t *)mtm) { 1624*0Sstevel@tonic-gate mutex_exit(&mtm->mtm_mutex); 1625*0Sstevel@tonic-gate logmap_forceroll(mtm); 1626*0Sstevel@tonic-gate goto again; 1627*0Sstevel@tonic-gate } 1628*0Sstevel@tonic-gate 1629*0Sstevel@tonic-gate /* 1630*0Sstevel@tonic-gate * no more deltas, return 1631*0Sstevel@tonic-gate */ 1632*0Sstevel@tonic-gate mutex_exit(&mtm->mtm_mutex); 1633*0Sstevel@tonic-gate (void) ufs_putsummaryinfo(ul->un_dev, ufsvfsp, ufsvfsp->vfs_fs); 1634*0Sstevel@tonic-gate 1635*0Sstevel@tonic-gate ASSERT(((mtm->mtm_debug & MT_CHECK_MAP) == 0) || 1636*0Sstevel@tonic-gate map_check_linkage(mtm)); 1637*0Sstevel@tonic-gate } 1638*0Sstevel@tonic-gate 1639*0Sstevel@tonic-gate static void 1640*0Sstevel@tonic-gate logmap_cancel_delta(ml_unit_t *ul, offset_t mof, int32_t nb, int metadata) 1641*0Sstevel@tonic-gate { 1642*0Sstevel@tonic-gate mapentry_t *me; 1643*0Sstevel@tonic-gate mapentry_t **mep; 1644*0Sstevel@tonic-gate mt_map_t *mtm = ul->un_logmap; 1645*0Sstevel@tonic-gate int frags; 1646*0Sstevel@tonic-gate 1647*0Sstevel@tonic-gate /* 1648*0Sstevel@tonic-gate * map has been referenced and is dirty 1649*0Sstevel@tonic-gate */ 1650*0Sstevel@tonic-gate mtm->mtm_ref = 1; 1651*0Sstevel@tonic-gate mtm->mtm_dirty++; 1652*0Sstevel@tonic-gate 1653*0Sstevel@tonic-gate /* 1654*0Sstevel@tonic-gate * get a mapentry 1655*0Sstevel@tonic-gate */ 1656*0Sstevel@tonic-gate me = kmem_cache_alloc(mapentry_cache, KM_SLEEP); 1657*0Sstevel@tonic-gate bzero(me, sizeof (mapentry_t)); 1658*0Sstevel@tonic-gate 1659*0Sstevel@tonic-gate /* 1660*0Sstevel@tonic-gate * initialize cancel record and put in logmap 1661*0Sstevel@tonic-gate */ 1662*0Sstevel@tonic-gate me->me_mof = mof; 1663*0Sstevel@tonic-gate me->me_nb = nb; 1664*0Sstevel@tonic-gate me->me_dt = DT_CANCEL; 1665*0Sstevel@tonic-gate me->me_tid = mtm->mtm_tid; 1666*0Sstevel@tonic-gate me->me_hash = NULL; 1667*0Sstevel@tonic-gate 1668*0Sstevel@tonic-gate /* 1669*0Sstevel@tonic-gate * Write delta to log if this delta is for metadata. If this is not 1670*0Sstevel@tonic-gate * metadata it is user data and we are just putting a cancel 1671*0Sstevel@tonic-gate * mapentry into the hash to cancel a user block deletion 1672*0Sstevel@tonic-gate * in which we do not want the block to be allocated 1673*0Sstevel@tonic-gate * within this moby. This cancel entry will prevent the block from 1674*0Sstevel@tonic-gate * being allocated within the moby and prevent user data corruption 1675*0Sstevel@tonic-gate * if we happen to crash before this moby is committed. 1676*0Sstevel@tonic-gate */ 1677*0Sstevel@tonic-gate mutex_enter(&ul->un_log_mutex); 1678*0Sstevel@tonic-gate if (metadata) { 1679*0Sstevel@tonic-gate logmap_wait_space(mtm, ul, me); 1680*0Sstevel@tonic-gate ldl_write(ul, NULL, (offset_t)0, me); 1681*0Sstevel@tonic-gate if (ul->un_flags & LDL_ERROR) { 1682*0Sstevel@tonic-gate kmem_cache_free(mapentry_cache, me); 1683*0Sstevel@tonic-gate mutex_exit(&ul->un_log_mutex); 1684*0Sstevel@tonic-gate return; 1685*0Sstevel@tonic-gate } 1686*0Sstevel@tonic-gate } 1687*0Sstevel@tonic-gate 1688*0Sstevel@tonic-gate /* 1689*0Sstevel@tonic-gate * put in hash and on cancel list 1690*0Sstevel@tonic-gate */ 1691*0Sstevel@tonic-gate mep = MAP_HASH(mof, mtm); 1692*0Sstevel@tonic-gate mutex_enter(&mtm->mtm_mutex); 1693*0Sstevel@tonic-gate me->me_age = mtm->mtm_age++; 1694*0Sstevel@tonic-gate me->me_hash = *mep; 1695*0Sstevel@tonic-gate *mep = me; 1696*0Sstevel@tonic-gate me->me_next = (mapentry_t *)mtm; 1697*0Sstevel@tonic-gate me->me_prev = mtm->mtm_prev; 1698*0Sstevel@tonic-gate mtm->mtm_prev->me_next = me; 1699*0Sstevel@tonic-gate mtm->mtm_prev = me; 1700*0Sstevel@tonic-gate me->me_cancel = mtm->mtm_cancel; 1701*0Sstevel@tonic-gate mtm->mtm_cancel = me; 1702*0Sstevel@tonic-gate if (metadata) { 1703*0Sstevel@tonic-gate mtm->mtm_nme++; 1704*0Sstevel@tonic-gate mtm->mtm_nmet++; 1705*0Sstevel@tonic-gate } else { 1706*0Sstevel@tonic-gate me->me_flags = ME_USER; 1707*0Sstevel@tonic-gate } 1708*0Sstevel@tonic-gate me->me_flags |= (ME_HASH|ME_CANCEL); 1709*0Sstevel@tonic-gate if (!(metadata)) { 1710*0Sstevel@tonic-gate frags = blkoff(ul->un_ufsvfs->vfs_fs, nb); 1711*0Sstevel@tonic-gate if (frags) 1712*0Sstevel@tonic-gate mtm->mtm_cfrags += numfrags(ul->un_ufsvfs->vfs_fs, 1713*0Sstevel@tonic-gate frags); 1714*0Sstevel@tonic-gate } 1715*0Sstevel@tonic-gate mutex_exit(&mtm->mtm_mutex); 1716*0Sstevel@tonic-gate 1717*0Sstevel@tonic-gate mutex_exit(&ul->un_log_mutex); 1718*0Sstevel@tonic-gate } 1719*0Sstevel@tonic-gate 1720*0Sstevel@tonic-gate /* 1721*0Sstevel@tonic-gate * cancel entries in a logmap (entries are freed at EOT) 1722*0Sstevel@tonic-gate */ 1723*0Sstevel@tonic-gate void 1724*0Sstevel@tonic-gate logmap_cancel(ml_unit_t *ul, offset_t mof, off_t nb, int metadata) 1725*0Sstevel@tonic-gate { 1726*0Sstevel@tonic-gate int32_t hnb; 1727*0Sstevel@tonic-gate mapentry_t *me; 1728*0Sstevel@tonic-gate mapentry_t **mep; 1729*0Sstevel@tonic-gate mt_map_t *mtm = ul->un_logmap; 1730*0Sstevel@tonic-gate crb_t *crb; 1731*0Sstevel@tonic-gate 1732*0Sstevel@tonic-gate ASSERT(((mtm->mtm_debug & MT_CHECK_MAP) == 0) || 1733*0Sstevel@tonic-gate map_check_linkage(mtm)); 1734*0Sstevel@tonic-gate 1735*0Sstevel@tonic-gate for (hnb = 0; nb; nb -= hnb, mof += hnb) { 1736*0Sstevel@tonic-gate hnb = MAPBLOCKSIZE - (mof & MAPBLOCKOFF); 1737*0Sstevel@tonic-gate if (hnb > nb) 1738*0Sstevel@tonic-gate hnb = nb; 1739*0Sstevel@tonic-gate /* 1740*0Sstevel@tonic-gate * find overlapping entries 1741*0Sstevel@tonic-gate */ 1742*0Sstevel@tonic-gate mep = MAP_HASH(mof, mtm); 1743*0Sstevel@tonic-gate mutex_enter(&mtm->mtm_mutex); 1744*0Sstevel@tonic-gate for (me = *mep; me; me = me->me_hash) { 1745*0Sstevel@tonic-gate if (!DATAoverlapME(mof, hnb, me)) 1746*0Sstevel@tonic-gate continue; 1747*0Sstevel@tonic-gate 1748*0Sstevel@tonic-gate ASSERT(MEwithinDATA(me, mof, hnb)); 1749*0Sstevel@tonic-gate 1750*0Sstevel@tonic-gate if ((me->me_flags & ME_CANCEL) == 0) { 1751*0Sstevel@tonic-gate me->me_cancel = mtm->mtm_cancel; 1752*0Sstevel@tonic-gate mtm->mtm_cancel = me; 1753*0Sstevel@tonic-gate me->me_flags |= ME_CANCEL; 1754*0Sstevel@tonic-gate crb = me->me_crb; 1755*0Sstevel@tonic-gate if (crb) { 1756*0Sstevel@tonic-gate crb->c_invalid = 1; 1757*0Sstevel@tonic-gate } 1758*0Sstevel@tonic-gate } 1759*0Sstevel@tonic-gate } 1760*0Sstevel@tonic-gate mutex_exit(&mtm->mtm_mutex); 1761*0Sstevel@tonic-gate 1762*0Sstevel@tonic-gate /* 1763*0Sstevel@tonic-gate * put a cancel record into the log 1764*0Sstevel@tonic-gate */ 1765*0Sstevel@tonic-gate logmap_cancel_delta(ul, mof, hnb, metadata); 1766*0Sstevel@tonic-gate } 1767*0Sstevel@tonic-gate 1768*0Sstevel@tonic-gate ASSERT(((mtm->mtm_debug & MT_CHECK_MAP) == 0) || 1769*0Sstevel@tonic-gate map_check_linkage(mtm)); 1770*0Sstevel@tonic-gate } 1771*0Sstevel@tonic-gate 1772*0Sstevel@tonic-gate /* 1773*0Sstevel@tonic-gate * check for overlap w/cancel delta 1774*0Sstevel@tonic-gate */ 1775*0Sstevel@tonic-gate int 1776*0Sstevel@tonic-gate logmap_iscancel(mt_map_t *mtm, offset_t mof, off_t nb) 1777*0Sstevel@tonic-gate { 1778*0Sstevel@tonic-gate off_t hnb; 1779*0Sstevel@tonic-gate mapentry_t *me; 1780*0Sstevel@tonic-gate mapentry_t **mep; 1781*0Sstevel@tonic-gate 1782*0Sstevel@tonic-gate mutex_enter(&mtm->mtm_mutex); 1783*0Sstevel@tonic-gate for (hnb = 0; nb; nb -= hnb, mof += hnb) { 1784*0Sstevel@tonic-gate hnb = MAPBLOCKSIZE - (mof & MAPBLOCKOFF); 1785*0Sstevel@tonic-gate if (hnb > nb) 1786*0Sstevel@tonic-gate hnb = nb; 1787*0Sstevel@tonic-gate /* 1788*0Sstevel@tonic-gate * search for dup entry 1789*0Sstevel@tonic-gate */ 1790*0Sstevel@tonic-gate mep = MAP_HASH(mof, mtm); 1791*0Sstevel@tonic-gate for (me = *mep; me; me = me->me_hash) { 1792*0Sstevel@tonic-gate if (((me->me_flags & ME_ROLL) == 0) && 1793*0Sstevel@tonic-gate (me->me_dt != DT_CANCEL)) 1794*0Sstevel@tonic-gate continue; 1795*0Sstevel@tonic-gate if (DATAoverlapME(mof, hnb, me)) 1796*0Sstevel@tonic-gate break; 1797*0Sstevel@tonic-gate } 1798*0Sstevel@tonic-gate 1799*0Sstevel@tonic-gate /* 1800*0Sstevel@tonic-gate * overlap detected 1801*0Sstevel@tonic-gate */ 1802*0Sstevel@tonic-gate if (me) { 1803*0Sstevel@tonic-gate mutex_exit(&mtm->mtm_mutex); 1804*0Sstevel@tonic-gate return (1); 1805*0Sstevel@tonic-gate } 1806*0Sstevel@tonic-gate } 1807*0Sstevel@tonic-gate mutex_exit(&mtm->mtm_mutex); 1808*0Sstevel@tonic-gate return (0); 1809*0Sstevel@tonic-gate } 1810*0Sstevel@tonic-gate 1811*0Sstevel@tonic-gate static int 1812*0Sstevel@tonic-gate logmap_logscan_add(ml_unit_t *ul, struct delta *dp, off_t lof, size_t *nbp) 1813*0Sstevel@tonic-gate { 1814*0Sstevel@tonic-gate mapentry_t *me; 1815*0Sstevel@tonic-gate int error; 1816*0Sstevel@tonic-gate mt_map_t *mtm = ul->un_logmap; 1817*0Sstevel@tonic-gate 1818*0Sstevel@tonic-gate /* 1819*0Sstevel@tonic-gate * verify delta header; failure == mediafail 1820*0Sstevel@tonic-gate */ 1821*0Sstevel@tonic-gate error = 0; 1822*0Sstevel@tonic-gate /* delta type */ 1823*0Sstevel@tonic-gate if ((dp->d_typ <= DT_NONE) || (dp->d_typ >= DT_MAX)) 1824*0Sstevel@tonic-gate error = EINVAL; 1825*0Sstevel@tonic-gate if (dp->d_typ == DT_COMMIT) { 1826*0Sstevel@tonic-gate if (dp->d_nb != INT32_C(0) && dp->d_nb != INT32_C(-1)) 1827*0Sstevel@tonic-gate error = EINVAL; 1828*0Sstevel@tonic-gate } else { 1829*0Sstevel@tonic-gate /* length of delta */ 1830*0Sstevel@tonic-gate if ((dp->d_nb < INT32_C(0)) || 1831*0Sstevel@tonic-gate (dp->d_nb > INT32_C(MAPBLOCKSIZE))) 1832*0Sstevel@tonic-gate error = EINVAL; 1833*0Sstevel@tonic-gate 1834*0Sstevel@tonic-gate /* offset on master device */ 1835*0Sstevel@tonic-gate if (dp->d_mof < INT64_C(0)) 1836*0Sstevel@tonic-gate error = EINVAL; 1837*0Sstevel@tonic-gate } 1838*0Sstevel@tonic-gate 1839*0Sstevel@tonic-gate if (error) { 1840*0Sstevel@tonic-gate ldl_seterror(ul, "Error processing ufs log data during scan"); 1841*0Sstevel@tonic-gate return (error); 1842*0Sstevel@tonic-gate } 1843*0Sstevel@tonic-gate 1844*0Sstevel@tonic-gate /* 1845*0Sstevel@tonic-gate * process commit record 1846*0Sstevel@tonic-gate */ 1847*0Sstevel@tonic-gate if (dp->d_typ == DT_COMMIT) { 1848*0Sstevel@tonic-gate if (mtm->mtm_dirty) { 1849*0Sstevel@tonic-gate ASSERT(dp->d_nb == INT32_C(0)); 1850*0Sstevel@tonic-gate logmap_free_cancel(mtm, &mtm->mtm_cancel); 1851*0Sstevel@tonic-gate mtm->mtm_dirty = 0; 1852*0Sstevel@tonic-gate mtm->mtm_nmet = 0; 1853*0Sstevel@tonic-gate mtm->mtm_tid++; 1854*0Sstevel@tonic-gate mtm->mtm_committid = mtm->mtm_tid; 1855*0Sstevel@tonic-gate ASSERT(((mtm->mtm_debug & MT_SCAN) == 0) || 1856*0Sstevel@tonic-gate logmap_logscan_commit_debug(lof, mtm)); 1857*0Sstevel@tonic-gate } 1858*0Sstevel@tonic-gate /* 1859*0Sstevel@tonic-gate * return #bytes to next sector (next delta header) 1860*0Sstevel@tonic-gate */ 1861*0Sstevel@tonic-gate *nbp = ldl_logscan_nbcommit(lof); 1862*0Sstevel@tonic-gate mtm->mtm_tail_lof = lof; 1863*0Sstevel@tonic-gate mtm->mtm_tail_nb = *nbp; 1864*0Sstevel@tonic-gate return (0); 1865*0Sstevel@tonic-gate } 1866*0Sstevel@tonic-gate 1867*0Sstevel@tonic-gate /* 1868*0Sstevel@tonic-gate * add delta to logmap 1869*0Sstevel@tonic-gate */ 1870*0Sstevel@tonic-gate me = kmem_cache_alloc(mapentry_cache, KM_SLEEP); 1871*0Sstevel@tonic-gate bzero(me, sizeof (mapentry_t)); 1872*0Sstevel@tonic-gate me->me_lof = lof; 1873*0Sstevel@tonic-gate me->me_mof = dp->d_mof; 1874*0Sstevel@tonic-gate me->me_nb = dp->d_nb; 1875*0Sstevel@tonic-gate me->me_tid = mtm->mtm_tid; 1876*0Sstevel@tonic-gate me->me_dt = dp->d_typ; 1877*0Sstevel@tonic-gate me->me_hash = NULL; 1878*0Sstevel@tonic-gate me->me_flags = (ME_LIST | ME_SCAN); 1879*0Sstevel@tonic-gate logmap_add(ul, NULL, 0, me); 1880*0Sstevel@tonic-gate switch (dp->d_typ) { 1881*0Sstevel@tonic-gate case DT_CANCEL: 1882*0Sstevel@tonic-gate me->me_flags |= ME_CANCEL; 1883*0Sstevel@tonic-gate me->me_cancel = mtm->mtm_cancel; 1884*0Sstevel@tonic-gate mtm->mtm_cancel = me; 1885*0Sstevel@tonic-gate break; 1886*0Sstevel@tonic-gate default: 1887*0Sstevel@tonic-gate ASSERT(((mtm->mtm_debug & MT_SCAN) == 0) || 1888*0Sstevel@tonic-gate logmap_logscan_add_debug(dp, mtm)); 1889*0Sstevel@tonic-gate break; 1890*0Sstevel@tonic-gate } 1891*0Sstevel@tonic-gate 1892*0Sstevel@tonic-gate sizeofdelta: 1893*0Sstevel@tonic-gate /* 1894*0Sstevel@tonic-gate * return #bytes till next delta header 1895*0Sstevel@tonic-gate */ 1896*0Sstevel@tonic-gate if ((dp->d_typ == DT_CANCEL) || (dp->d_typ == DT_ABZERO)) 1897*0Sstevel@tonic-gate *nbp = 0; 1898*0Sstevel@tonic-gate else 1899*0Sstevel@tonic-gate *nbp = dp->d_nb; 1900*0Sstevel@tonic-gate return (0); 1901*0Sstevel@tonic-gate } 1902*0Sstevel@tonic-gate 1903*0Sstevel@tonic-gate void 1904*0Sstevel@tonic-gate logmap_logscan(ml_unit_t *ul) 1905*0Sstevel@tonic-gate { 1906*0Sstevel@tonic-gate size_t nb, nbd; 1907*0Sstevel@tonic-gate off_t lof; 1908*0Sstevel@tonic-gate struct delta delta; 1909*0Sstevel@tonic-gate mt_map_t *logmap = ul->un_logmap; 1910*0Sstevel@tonic-gate 1911*0Sstevel@tonic-gate ASSERT(ul->un_deltamap->mtm_next == (mapentry_t *)ul->un_deltamap); 1912*0Sstevel@tonic-gate 1913*0Sstevel@tonic-gate /* 1914*0Sstevel@tonic-gate * prepare the log for a logscan 1915*0Sstevel@tonic-gate */ 1916*0Sstevel@tonic-gate ldl_logscan_begin(ul); 1917*0Sstevel@tonic-gate 1918*0Sstevel@tonic-gate /* 1919*0Sstevel@tonic-gate * prepare the logmap for a logscan 1920*0Sstevel@tonic-gate */ 1921*0Sstevel@tonic-gate (void) map_free_entries(logmap); 1922*0Sstevel@tonic-gate logmap->mtm_tid = 0; 1923*0Sstevel@tonic-gate logmap->mtm_committid = UINT32_C(0); 1924*0Sstevel@tonic-gate logmap->mtm_age = 0; 1925*0Sstevel@tonic-gate logmap->mtm_dirty = 0; 1926*0Sstevel@tonic-gate logmap->mtm_ref = 0; 1927*0Sstevel@tonic-gate 1928*0Sstevel@tonic-gate /* 1929*0Sstevel@tonic-gate * while not at end of log 1930*0Sstevel@tonic-gate * read delta header 1931*0Sstevel@tonic-gate * add to logmap 1932*0Sstevel@tonic-gate * seek to beginning of next delta 1933*0Sstevel@tonic-gate */ 1934*0Sstevel@tonic-gate lof = ul->un_head_lof; 1935*0Sstevel@tonic-gate nbd = sizeof (delta); 1936*0Sstevel@tonic-gate while (lof != ul->un_tail_lof) { 1937*0Sstevel@tonic-gate 1938*0Sstevel@tonic-gate /* read delta header */ 1939*0Sstevel@tonic-gate if (ldl_logscan_read(ul, &lof, nbd, (caddr_t)&delta)) 1940*0Sstevel@tonic-gate break; 1941*0Sstevel@tonic-gate 1942*0Sstevel@tonic-gate /* add to logmap */ 1943*0Sstevel@tonic-gate if (logmap_logscan_add(ul, &delta, lof, &nb)) 1944*0Sstevel@tonic-gate break; 1945*0Sstevel@tonic-gate 1946*0Sstevel@tonic-gate /* seek to next header (skip data) */ 1947*0Sstevel@tonic-gate if (ldl_logscan_read(ul, &lof, nb, NULL)) 1948*0Sstevel@tonic-gate break; 1949*0Sstevel@tonic-gate } 1950*0Sstevel@tonic-gate 1951*0Sstevel@tonic-gate /* 1952*0Sstevel@tonic-gate * remove the last partial transaction from the logmap 1953*0Sstevel@tonic-gate */ 1954*0Sstevel@tonic-gate logmap_abort(ul, logmap->mtm_tid); 1955*0Sstevel@tonic-gate 1956*0Sstevel@tonic-gate ldl_logscan_end(ul); 1957*0Sstevel@tonic-gate } 1958*0Sstevel@tonic-gate 1959*0Sstevel@tonic-gate void 1960*0Sstevel@tonic-gate _init_map(void) 1961*0Sstevel@tonic-gate { 1962*0Sstevel@tonic-gate /* 1963*0Sstevel@tonic-gate * Initialise the mapentry cache. No constructor or deconstructor 1964*0Sstevel@tonic-gate * is needed. Also no reclaim function is supplied as reclaiming 1965*0Sstevel@tonic-gate * current entries is not possible. 1966*0Sstevel@tonic-gate */ 1967*0Sstevel@tonic-gate mapentry_cache = kmem_cache_create("lufs_mapentry_cache", 1968*0Sstevel@tonic-gate sizeof (mapentry_t), 0, NULL, NULL, NULL, NULL, NULL, 0); 1969*0Sstevel@tonic-gate } 1970*0Sstevel@tonic-gate 1971*0Sstevel@tonic-gate /* 1972*0Sstevel@tonic-gate * Special case when we replace an old map entry which carries quota 1973*0Sstevel@tonic-gate * information with a newer entry which does not. 1974*0Sstevel@tonic-gate * In that case the push function would not be called to clean up the 1975*0Sstevel@tonic-gate * dquot structure. This would be found later by invalidatedq() causing 1976*0Sstevel@tonic-gate * a panic when the filesystem in unmounted. 1977*0Sstevel@tonic-gate * We clean up the dquot manually before replacing the map entry. 1978*0Sstevel@tonic-gate */ 1979*0Sstevel@tonic-gate void 1980*0Sstevel@tonic-gate handle_dquot(mapentry_t *me) 1981*0Sstevel@tonic-gate { 1982*0Sstevel@tonic-gate int dolock = 0; 1983*0Sstevel@tonic-gate int domutex = 0; 1984*0Sstevel@tonic-gate struct dquot *dqp; 1985*0Sstevel@tonic-gate 1986*0Sstevel@tonic-gate dqp = (struct dquot *)me->me_arg; 1987*0Sstevel@tonic-gate 1988*0Sstevel@tonic-gate /* 1989*0Sstevel@tonic-gate * We need vfs_dqrwlock to call dqput() 1990*0Sstevel@tonic-gate */ 1991*0Sstevel@tonic-gate dolock = (!RW_LOCK_HELD(&dqp->dq_ufsvfsp->vfs_dqrwlock)); 1992*0Sstevel@tonic-gate if (dolock) 1993*0Sstevel@tonic-gate rw_enter(&dqp->dq_ufsvfsp->vfs_dqrwlock, RW_READER); 1994*0Sstevel@tonic-gate 1995*0Sstevel@tonic-gate domutex = (!MUTEX_HELD(&dqp->dq_lock)); 1996*0Sstevel@tonic-gate if (domutex) 1997*0Sstevel@tonic-gate mutex_enter(&dqp->dq_lock); 1998*0Sstevel@tonic-gate 1999*0Sstevel@tonic-gate /* 2000*0Sstevel@tonic-gate * Only clean up if the dquot is referenced 2001*0Sstevel@tonic-gate */ 2002*0Sstevel@tonic-gate if (dqp->dq_cnt == 0) { 2003*0Sstevel@tonic-gate if (domutex) 2004*0Sstevel@tonic-gate mutex_exit(&dqp->dq_lock); 2005*0Sstevel@tonic-gate if (dolock) 2006*0Sstevel@tonic-gate rw_exit(&dqp->dq_ufsvfsp->vfs_dqrwlock); 2007*0Sstevel@tonic-gate return; 2008*0Sstevel@tonic-gate } 2009*0Sstevel@tonic-gate 2010*0Sstevel@tonic-gate dqp->dq_flags &= ~(DQ_MOD|DQ_TRANS); 2011*0Sstevel@tonic-gate dqput(dqp); 2012*0Sstevel@tonic-gate 2013*0Sstevel@tonic-gate if (domutex) 2014*0Sstevel@tonic-gate mutex_exit(&dqp->dq_lock); 2015*0Sstevel@tonic-gate 2016*0Sstevel@tonic-gate if (dolock) 2017*0Sstevel@tonic-gate rw_exit(&dqp->dq_ufsvfsp->vfs_dqrwlock); 2018*0Sstevel@tonic-gate 2019*0Sstevel@tonic-gate } 2020