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
  35 #include "hammer.h"
  36
  37 static uint32_t ocp_allocbit(hammer_objid_cache_t ocp, uint32_t n);
  38
  39
  40 /*
  41  * Start a standard transaction.
  42  *
  43  * May be called without fs_token
  44  */
  45 void
  46 hammer_start_transaction(hammer_transaction_t trans, hammer_mount_t hmp)
  47 {
  48         struct timespec ts;
  49         int error;
  50
  51         trans->type = HAMMER_TRANS_STD;
  52         trans->hmp = hmp;
  53         trans->rootvol = hammer_get_root_volume(hmp, &error);
  54         KKASSERT(error == 0);
  55         trans->tid = 0;
  56         trans->sync_lock_refs = 0;
  57         trans->flags = 0;
  58
  59         vfs_timestamp(&ts);
  60         trans->time = (unsigned long)ts.tv_sec * 1000000ULL +
  61                       ts.tv_nsec / 1000;
  62         trans->time32 = (uint32_t)ts.tv_sec;
  63 }
  64
  65 /*
  66  * Start a simple read-only transaction.  This will not stall.
  67  *
  68  * May be called without fs_token
  69  */
  70 void
  71 hammer_simple_transaction(hammer_transaction_t trans, hammer_mount_t hmp)
  72 {
  73         struct timespec ts;
  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         vfs_timestamp(&ts);
  85         trans->time = (unsigned long)ts.tv_sec * 1000000ULL +
  86                       ts.tv_nsec / 1000;
  87         trans->time32 = (uint32_t)ts.tv_sec;
  88 }
  89
  90 /*
  91  * Start a transaction using a particular TID.  Used by the sync code.
  92  * This does not stall.
  93  *
  94  * This routine may only be called from the flusher thread.  We predispose
  95  * sync_lock_refs, implying serialization against the synchronization stage
  96  * (which the flusher is responsible for).
  97  */
  98 void
  99 hammer_start_transaction_fls(hammer_transaction_t trans, hammer_mount_t hmp)
 100 {
 101         struct timespec ts;
 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         vfs_timestamp(&ts);
 115         trans->time = (unsigned long)ts.tv_sec * 1000000ULL +
 116                       ts.tv_nsec / 1000;
 117         trans->time32 = (uint32_t)ts.tv_sec;
 118 }
 119
 120 /*
 121  * May be called without fs_token
 122  */
 123 void
 124 hammer_done_transaction(hammer_transaction_t trans)
 125 {
 126         int expected_lock_refs __debugvar;
 127
 128         hammer_rel_volume(trans->rootvol, 0);
 129         trans->rootvol = NULL;
 130         expected_lock_refs = (trans->type == HAMMER_TRANS_FLS) ? 1 : 0;
 131         KKASSERT(trans->sync_lock_refs == expected_lock_refs);
 132         trans->sync_lock_refs = 0;
 133         if (trans->type != HAMMER_TRANS_FLS) {
 134                 if (trans->flags & HAMMER_TRANSF_NEWINODE) {
 135                         lwkt_gettoken(&trans->hmp->fs_token);
 136                         hammer_inode_waitreclaims(trans);
 137                         lwkt_reltoken(&trans->hmp->fs_token);
 138                 }
 139         }
 140 }
 141
 142 /*
 143  * Allocate (count) TIDs.  If running in multi-master mode the returned
 144  * base will be aligned to a 16-count plus the master id (0-15).
 145  * Multi-master mode allows non-conflicting to run and new objects to be
 146  * created on multiple masters in parallel.  The transaction id identifies
 147  * the original master.  The object_id is also subject to this rule in
 148  * order to allow objects to be created on multiple masters in parallel.
 149  *
 150  * Directories may pre-allocate a large number of object ids (100,000).
 151  *
 152  * NOTE: There is no longer a requirement that successive transaction
 153  *       ids be 2 apart for separator generation.
 154  *
 155  * NOTE: When called by pseudo-backends such as ioctls the allocated
 156  *       TID will be larger then the current flush TID, if a flush is running,
 157  *       so any mirroring will pick the records up on a later flush.
 158  *
 159  * NOTE: HAMMER1 does not support multi-master clustering as of 2015.
 160  */
 161 hammer_tid_t
 162 hammer_alloc_tid(hammer_mount_t hmp, int count)
 163 {
 164         hammer_tid_t tid;
 165
 166         if (hmp->master_id < 0) {
 167                 tid = hmp->next_tid + 1;
 168                 hmp->next_tid = tid + count;
 169         } else {
 170                 tid = (hmp->next_tid + HAMMER_MAX_MASTERS) &
 171                       ~(hammer_tid_t)(HAMMER_MAX_MASTERS - 1);
 172                 hmp->next_tid = tid + count * HAMMER_MAX_MASTERS;
 173                 tid |= hmp->master_id;
 174         }
 175         if (tid >= 0xFFFFFFFFFF000000ULL)
 176                 hpanic("Ran out of TIDs!");
 177         if (hammer_debug_tid)
 178                 hdkprintf("%016jx\n", (intmax_t)tid);
 179         return(tid);
 180 }
 181
 182 /*
 183  * Allocate an object id.
 184  *
 185  * We use the upper OBJID_CACHE_BITS bits of the namekey to try to match
 186  * the low bits of the objid we allocate.
 187  */
 188 hammer_tid_t
 189 hammer_alloc_objid(hammer_mount_t hmp, hammer_inode_t dip, int64_t namekey)
 190 {
 191         hammer_objid_cache_t ocp;
 192         hammer_tid_t tid;
 193         uint32_t n;
 194
 195         while ((ocp = dip->objid_cache) == NULL) {
 196                 if (hmp->objid_cache_count < OBJID_CACHE_SIZE) {
 197                         ocp = kmalloc(sizeof(*ocp), hmp->m_misc,
 198                                       M_WAITOK|M_ZERO);
 199                         ocp->base_tid = hammer_alloc_tid(hmp,
 200                                                         OBJID_CACHE_BULK * 2);
 201                         ocp->base_tid += OBJID_CACHE_BULK_MASK64;
 202                         ocp->base_tid &= ~OBJID_CACHE_BULK_MASK64;
 203                         /* may have blocked, recheck */
 204                         if (dip->objid_cache == NULL) {
 205                                 TAILQ_INSERT_TAIL(&hmp->objid_cache_list,
 206                                                   ocp, entry);
 207                                 ++hmp->objid_cache_count;
 208                                 dip->objid_cache = ocp;
 209                                 ocp->dip = dip;
 210                         } else {
 211                                 kfree(ocp, hmp->m_misc);
 212                         }
 213                 } else {
 214                         /*
 215                          * Steal one from another directory?
 216                          *
 217                          * Throw away ocp's that are more then half full, they
 218                          * aren't worth stealing.
 219                          */
 220                         ocp = TAILQ_FIRST(&hmp->objid_cache_list);
 221                         if (ocp->dip)
 222                                 ocp->dip->objid_cache = NULL;
 223                         if (ocp->count >= OBJID_CACHE_BULK / 2) {
 224                                 TAILQ_REMOVE(&hmp->objid_cache_list,
 225                                              ocp, entry);
 226                                 --hmp->objid_cache_count;
 227                                 kfree(ocp, hmp->m_misc);
 228                         } else {
 229                                 dip->objid_cache = ocp;
 230                                 ocp->dip = dip;
 231                         }
 232                 }
 233         }
 234         TAILQ_REMOVE(&hmp->objid_cache_list, ocp, entry);
 235
 236         /*
 237          * Allocate inode numbers uniformly.
 238          */
 239
 240         n = (namekey >> (63 - OBJID_CACHE_BULK_BITS)) & OBJID_CACHE_BULK_MASK;
 241         n = ocp_allocbit(ocp, n);
 242         tid = ocp->base_tid + n;
 243
 244 #if 0
 245         /*
 246          * The TID is incremented by 1 or by 16 depending what mode the
 247          * mount is operating in.
 248          */
 249         ocp->next_tid += (hmp->master_id < 0) ? 1 : HAMMER_MAX_MASTERS;
 250 #endif
 251         if (ocp->count >= OBJID_CACHE_BULK * 3 / 4) {
 252                 dip->objid_cache = NULL;
 253                 --hmp->objid_cache_count;
 254                 ocp->dip = NULL;
 255                 kfree(ocp, hmp->m_misc);
 256         } else {
 257                 TAILQ_INSERT_TAIL(&hmp->objid_cache_list, ocp, entry);
 258         }
 259         return(tid);
 260 }
 261
 262 /*
 263  * Allocate a bit starting with bit n.  Wrap if necessary.
 264  *
 265  * This routine is only ever called if a bit is available somewhere
 266  * in the bitmap.
 267  */
 268 static uint32_t
 269 ocp_allocbit(hammer_objid_cache_t ocp, uint32_t n)
 270 {
 271         uint32_t n0;
 272
 273         n0 = (n >> 5) & 31;
 274         n &= 31;
 275
 276         while (ocp->bm1[n0] & (1 << n)) {
 277                 if (ocp->bm0 & (1 << n0)) {
 278                         n0 = (n0 + 1) & 31;
 279                         n = 0;
 280                 } else if (++n == 32) {
 281                         n0 = (n0 + 1) & 31;
 282                         n = 0;
 283                 }
 284         }
 285         ++ocp->count;
 286         ocp->bm1[n0] |= 1 << n;
 287         if (ocp->bm1[n0] == 0xFFFFFFFFU)
 288                 ocp->bm0 |= 1 << n0;
 289         return((n0 << 5) + n);
 290 }
 291
 292 void
 293 hammer_clear_objid(hammer_inode_t dip)
 294 {
 295         hammer_objid_cache_t ocp;
 296
 297         if ((ocp = dip->objid_cache) != NULL) {
 298                 dip->objid_cache = NULL;
 299                 ocp->dip = NULL;
 300                 TAILQ_REMOVE(&dip->hmp->objid_cache_list, ocp, entry);
 301                 TAILQ_INSERT_HEAD(&dip->hmp->objid_cache_list, ocp, entry);
 302         }
 303 }
 304
 305 void
 306 hammer_destroy_objid_cache(hammer_mount_t hmp)
 307 {
 308         hammer_objid_cache_t ocp;
 309
 310         while ((ocp = TAILQ_FIRST(&hmp->objid_cache_list)) != NULL) {
 311                 TAILQ_REMOVE(&hmp->objid_cache_list, ocp, entry);
 312                 if (ocp->dip)
 313                         ocp->dip->objid_cache = NULL;
 314                 kfree(ocp, hmp->m_misc);
 315                 --hmp->objid_cache_count;
 316         }
 317         KKASSERT(hmp->objid_cache_count == 0);
 318 }
 319