sys/vfs/hammer/hammer_transaction.c

   1 /*
   2  * Copyright (c) 2007-2008 The DragonFly Project.  All rights reserved.
   3  *
   4  * This code is derived from software contributed to The DragonFly Project
   5  * by Matthew Dillon <dillon@backplane.com>
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  *
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in
  15  *    the documentation and/or other materials provided with the
  16  *    distribution.
  17  * 3. Neither the name of The DragonFly Project nor the names of its
  18  *    contributors may be used to endorse or promote products derived
  19  *    from this software without specific, prior written permission.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  24  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
  25  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  26  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
  27  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  28  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
  29  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
  31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  32  * SUCH DAMAGE.
  33  *
  34  * $DragonFly: src/sys/vfs/hammer/hammer_transaction.c,v 1.25 2008/09/23 21:03:52 dillon Exp $
  35  */
  36
  37 #include "hammer.h"
  38
  39 static hammer_tid_t hammer_alloc_tid(hammer_mount_t hmp, int count);
  40 static u_int32_t ocp_allocbit(hammer_objid_cache_t ocp, u_int32_t n);
  41
  42
  43 /*
  44  * Start a standard transaction.
  45  */
  46 void
  47 hammer_start_transaction(struct hammer_transaction *trans,
  48                          struct hammer_mount *hmp)
  49 {
  50         struct timeval tv;
  51         int error;
  52
  53         trans->type = HAMMER_TRANS_STD;
  54         trans->hmp = hmp;
  55         trans->rootvol = hammer_get_root_volume(hmp, &error);
  56         KKASSERT(error == 0);
  57         trans->tid = 0;
  58         trans->sync_lock_refs = 0;
  59         trans->flags = 0;
  60
  61         getmicrotime(&tv);
  62         trans->time = (unsigned long)tv.tv_sec * 1000000ULL + tv.tv_usec;
  63         trans->time32 = (u_int32_t)tv.tv_sec;
  64 }
  65
  66 /*
  67  * Start a simple read-only transaction.  This will not stall.
  68  */
  69 void
  70 hammer_simple_transaction(struct hammer_transaction *trans,
  71                           struct hammer_mount *hmp)
  72 {
  73         struct timeval tv;
  74         int error;
  75
  76         trans->type = HAMMER_TRANS_RO;
  77         trans->hmp = hmp;
  78         trans->rootvol = hammer_get_root_volume(hmp, &error);
  79         KKASSERT(error == 0);
  80         trans->tid = 0;
  81         trans->sync_lock_refs = 0;
  82         trans->flags = 0;
  83
  84         getmicrotime(&tv);
  85         trans->time = (unsigned long)tv.tv_sec * 1000000ULL + tv.tv_usec;
  86         trans->time32 = (u_int32_t)tv.tv_sec;
  87 }
  88
  89 /*
  90  * Start a transaction using a particular TID.  Used by the sync code.
  91  * This does not stall.
  92  *
  93  * This routine may only be called from the flusher thread.  We predispose
  94  * sync_lock_refs, implying serialization against the synchronization stage
  95  * (which the flusher is responsible for).
  96  */
  97 void
  98 hammer_start_transaction_fls(struct hammer_transaction *trans,
  99                              struct hammer_mount *hmp)
 100 {
 101         struct timeval tv;
 102         int error;
 103
 104         bzero(trans, sizeof(*trans));
 105
 106         trans->type = HAMMER_TRANS_FLS;
 107         trans->hmp = hmp;
 108         trans->rootvol = hammer_get_root_volume(hmp, &error);
 109         KKASSERT(error == 0);
 110         trans->tid = hammer_alloc_tid(hmp, 1);
 111         trans->sync_lock_refs = 1;
 112         trans->flags = 0;
 113
 114         getmicrotime(&tv);
 115         trans->time = (unsigned long)tv.tv_sec * 1000000ULL + tv.tv_usec;
 116         trans->time32 = (u_int32_t)tv.tv_sec;
 117 }
 118
 119 void
 120 hammer_done_transaction(struct hammer_transaction *trans)
 121 {
 122         hammer_mount_t hmp = trans->hmp;
 123         int expected_lock_refs;
 124
 125         hammer_rel_volume(trans->rootvol, 0);
 126         trans->rootvol = NULL;
 127         expected_lock_refs = (trans->type == HAMMER_TRANS_FLS) ? 1 : 0;
 128         KKASSERT(trans->sync_lock_refs == expected_lock_refs);
 129         trans->sync_lock_refs = 0;
 130         if (trans->type != HAMMER_TRANS_FLS) {
 131                 if (trans->flags & HAMMER_TRANSF_NEWINODE)
 132                         hammer_inode_waitreclaims(hmp);
 133                 else if (trans->flags & HAMMER_TRANSF_DIDIO)
 134                         hammer_inode_waithard(hmp);
 135         }
 136 }
 137
 138 /*
 139  * Allocate (count) TIDs.  If running in multi-master mode the returned
 140  * base will be aligned to a 16-count plus the master id (0-15).
 141  * Multi-master mode allows non-conflicting to run and new objects to be
 142  * created on multiple masters in parallel.  The transaction id identifies
 143  * the original master.  The object_id is also subject to this rule in
 144  * order to allow objects to be created on multiple masters in parallel.
 145  *
 146  * Directories may pre-allocate a large number of object ids (100,000).
 147  *
 148  * NOTE: There is no longer a requirement that successive transaction
 149  * ids be 2 apart for separator generation.
 150  */
 151 static hammer_tid_t
 152 hammer_alloc_tid(hammer_mount_t hmp, int count)
 153 {
 154         hammer_tid_t tid;
 155
 156         if (hmp->master_id < 0) {
 157                 tid = hmp->next_tid + 1;
 158                 hmp->next_tid = tid + count;
 159         } else {
 160                 tid = (hmp->next_tid + HAMMER_MAX_MASTERS) &
 161                       ~(hammer_tid_t)(HAMMER_MAX_MASTERS - 1);
 162                 hmp->next_tid = tid + count * HAMMER_MAX_MASTERS;
 163                 tid |= hmp->master_id;
 164         }
 165         if (tid >= 0xFFFFFFFFFF000000ULL)
 166                 panic("hammer_start_transaction: Ran out of TIDs!");
 167         if (hammer_debug_tid)
 168                 kprintf("alloc_tid %016llx\n", (long long)tid);
 169         return(tid);
 170 }
 171
 172 /*
 173  * Allocate an object id.
 174  *
 175  * We use the upper OBJID_CACHE_BITS bits of the namekey to try to match
 176  * the low bits of the objid we allocate.
 177  */
 178 hammer_tid_t
 179 hammer_alloc_objid(hammer_mount_t hmp, hammer_inode_t dip, int64_t namekey)
 180 {
 181         hammer_objid_cache_t ocp;
 182         hammer_tid_t tid;
 183         int incluster;
 184         u_int32_t n;
 185
 186         while ((ocp = dip->objid_cache) == NULL) {
 187                 if (hmp->objid_cache_count < OBJID_CACHE_SIZE) {
 188                         ocp = kmalloc(sizeof(*ocp), hmp->m_misc,
 189                                       M_WAITOK|M_ZERO);
 190                         ocp->base_tid = hammer_alloc_tid(hmp,
 191                                                         OBJID_CACHE_BULK * 2);
 192                         ocp->base_tid += OBJID_CACHE_BULK_MASK64;
 193                         ocp->base_tid &= ~OBJID_CACHE_BULK_MASK64;
 194                         TAILQ_INSERT_HEAD(&hmp->objid_cache_list, ocp, entry);
 195                         ++hmp->objid_cache_count;
 196                         /* may have blocked, recheck */
 197                         if (dip->objid_cache == NULL) {
 198                                 dip->objid_cache = ocp;
 199                                 ocp->dip = dip;
 200                         }
 201                 } else {
 202                         /*
 203                          * Steal one from another directory?
 204                          *
 205                          * Throw away ocp's that are more then half full, they
 206                          * aren't worth stealing.
 207                          */
 208                         ocp = TAILQ_FIRST(&hmp->objid_cache_list);
 209                         if (ocp->dip)
 210                                 ocp->dip->objid_cache = NULL;
 211                         if (ocp->count >= OBJID_CACHE_BULK / 2) {
 212                                 --hmp->objid_cache_count;
 213                                 kfree(ocp, hmp->m_misc);
 214                         } else {
 215                                 dip->objid_cache = ocp;
 216                                 ocp->dip = dip;
 217                         }
 218                 }
 219         }
 220         TAILQ_REMOVE(&hmp->objid_cache_list, ocp, entry);
 221
 222         /*
 223          * Allocate a bit based on our namekey for the low bits of our
 224          * objid.
 225          */
 226         incluster = (hmp->master_id >= 0);
 227         n = (namekey >> (63 - OBJID_CACHE_BULK_BITS)) & OBJID_CACHE_BULK_MASK;
 228         n = ocp_allocbit(ocp, n);
 229         tid = ocp->base_tid + n;
 230
 231 #if 0
 232         /*
 233          * The TID is incremented by 1 or by 16 depending what mode the
 234          * mount is operating in.
 235          */
 236         ocp->next_tid += (hmp->master_id < 0) ? 1 : HAMMER_MAX_MASTERS;
 237 #endif
 238         if (ocp->count >= OBJID_CACHE_BULK / 2) {
 239                 dip->objid_cache = NULL;
 240                 --hmp->objid_cache_count;
 241                 ocp->dip = NULL;
 242                 kfree(ocp, hmp->m_misc);
 243         } else {
 244                 TAILQ_INSERT_TAIL(&hmp->objid_cache_list, ocp, entry);
 245         }
 246         return(tid);
 247 }
 248
 249 /*
 250  * Allocate a bit starting with bit n.  Wrap if necessary.
 251  *
 252  * This routine is only ever called if a bit is available somewhere
 253  * in the bitmap.
 254  */
 255 static u_int32_t
 256 ocp_allocbit(hammer_objid_cache_t ocp, u_int32_t n)
 257 {
 258         u_int32_t n0;
 259
 260         n0 = (n >> 5) & 31;
 261         n &= 31;
 262
 263         while (ocp->bm1[n0] & (1 << n)) {
 264                 if (ocp->bm0 & (1 << n0)) {
 265                         n0 = (n0 + 1) & 31;
 266                         n = 0;
 267                 } else if (++n == 32) {
 268                         n0 = (n0 + 1) & 31;
 269                         n = 0;
 270                 }
 271         }
 272         ++ocp->count;
 273         ocp->bm1[n0] |= 1 << n;
 274         if (ocp->bm1[n0] == 0xFFFFFFFFU)
 275                 ocp->bm0 |= 1 << n0;
 276         return((n0 << 5) + n);
 277 }
 278
 279 void
 280 hammer_clear_objid(hammer_inode_t dip)
 281 {
 282         hammer_objid_cache_t ocp;
 283
 284         if ((ocp = dip->objid_cache) != NULL) {
 285                 dip->objid_cache = NULL;
 286                 ocp->dip = NULL;
 287                 TAILQ_REMOVE(&dip->hmp->objid_cache_list, ocp, entry);
 288                 TAILQ_INSERT_HEAD(&dip->hmp->objid_cache_list, ocp, entry);
 289         }
 290 }
 291
 292 void
 293 hammer_destroy_objid_cache(hammer_mount_t hmp)
 294 {
 295         hammer_objid_cache_t ocp;
 296
 297         while ((ocp = TAILQ_FIRST(&hmp->objid_cache_list)) != NULL) {
 298                 TAILQ_REMOVE(&hmp->objid_cache_list, ocp, entry);
 299                 if (ocp->dip)
 300                         ocp->dip->objid_cache = NULL;
 301                 kfree(ocp, hmp->m_misc);
 302         }
 303 }
 304