sys/vfs/hammer/hammer_prune.c

   1 /*
   2  * Copyright (c) 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_prune.c,v 1.7 2008/06/24 17:38:17 dillon Exp $
  35  */
  36
  37 #include "hammer.h"
  38
  39 /*
  40  * Iterate through the specified range of object ids and remove any
  41  * deleted records that fall entirely within a prune modulo.
  42  *
  43  * A reverse iteration is used to prevent overlapping records from being
  44  * created during the iteration due to alignments.  This also allows us
  45  * to adjust alignments without blowing up the B-Tree.
  46  */
  47 static int check_prune(struct hammer_ioc_prune *prune, hammer_btree_elm_t elm,
  48                         int *realign_cre, int *realign_del);
  49 static int realign_prune(struct hammer_ioc_prune *prune, hammer_cursor_t cursor,
  50                         int realign_cre, int realign_del);
  51
  52 int
  53 hammer_ioc_prune(hammer_transaction_t trans, hammer_inode_t ip,
  54                  struct hammer_ioc_prune *prune)
  55 {
  56         struct hammer_cursor cursor;
  57         hammer_btree_elm_t elm;
  58         struct hammer_ioc_prune_elm *copy_elms;
  59         struct hammer_ioc_prune_elm *user_elms;
  60         int error;
  61         int isdir;
  62         int realign_cre;
  63         int realign_del;
  64         int elm_array_size;
  65
  66         if (prune->nelms < 0 || prune->nelms > HAMMER_MAX_PRUNE_ELMS)
  67                 return(EINVAL);
  68         if ((prune->key_beg.localization | prune->key_end.localization) &
  69             HAMMER_LOCALIZE_PSEUDOFS_MASK) {
  70                 return(EINVAL);
  71         }
  72         if (prune->key_beg.localization > prune->key_end.localization)
  73                 return(EINVAL);
  74         if (prune->key_beg.localization == prune->key_end.localization) {
  75                 if (prune->key_beg.obj_id > prune->key_end.obj_id)
  76                         return(EINVAL);
  77                 /* key-space limitations - no check needed */
  78         }
  79         if ((prune->head.flags & HAMMER_IOC_PRUNE_ALL) && prune->nelms)
  80                 return(EINVAL);
  81
  82         prune->key_cur.localization = prune->key_end.localization +
  83                                       ip->obj_localization;
  84         prune->key_cur.obj_id = prune->key_end.obj_id;
  85         prune->key_cur.key = HAMMER_MAX_KEY;
  86
  87         /*
  88          * Copy element array from userland
  89          */
  90         elm_array_size = sizeof(*copy_elms) * prune->nelms;
  91         user_elms = prune->elms;
  92         copy_elms = kmalloc(elm_array_size, M_TEMP, M_WAITOK);
  93         if ((error = copyin(user_elms, copy_elms, elm_array_size)) != 0)
  94                 goto failed;
  95         prune->elms = copy_elms;
  96
  97         /*
  98          * Scan backwards.  Retries typically occur if a deadlock is detected.
  99          */
 100 retry:
 101         error = hammer_init_cursor(trans, &cursor, NULL, NULL);
 102         if (error) {
 103                 hammer_done_cursor(&cursor);
 104                 goto failed;
 105         }
 106         cursor.key_beg.localization = prune->key_beg.localization +
 107                                       ip->obj_localization;
 108         cursor.key_beg.obj_id = prune->key_beg.obj_id;
 109         cursor.key_beg.key = HAMMER_MIN_KEY;
 110         cursor.key_beg.create_tid = 1;
 111         cursor.key_beg.delete_tid = 0;
 112         cursor.key_beg.rec_type = HAMMER_MIN_RECTYPE;
 113         cursor.key_beg.obj_type = 0;
 114
 115         cursor.key_end.localization = prune->key_cur.localization;
 116         cursor.key_end.obj_id = prune->key_cur.obj_id;
 117         cursor.key_end.key = prune->key_cur.key;
 118         cursor.key_end.create_tid = HAMMER_MAX_TID - 1;
 119         cursor.key_end.delete_tid = 0;
 120         cursor.key_end.rec_type = HAMMER_MAX_RECTYPE;
 121         cursor.key_end.obj_type = 0;
 122
 123         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
 124         cursor.flags |= HAMMER_CURSOR_BACKEND;
 125
 126         /*
 127          * This flag allows the B-Tree code to clean up loose ends.
 128          */
 129         cursor.flags |= HAMMER_CURSOR_PRUNING;
 130
 131         hammer_sync_lock_sh(trans);
 132         error = hammer_btree_last(&cursor);
 133
 134         while (error == 0) {
 135                 /*
 136                  * Yield to more important tasks
 137                  */
 138                 if ((error = hammer_signal_check(trans->hmp)) != 0)
 139                         break;
 140                 if (trans->hmp->sync_lock.wanted) {
 141                         hammer_sync_unlock(trans);
 142                         tsleep(trans, 0, "hmrslo", hz / 10);
 143                         hammer_sync_lock_sh(trans);
 144                 }
 145                 if (trans->hmp->locked_dirty_count +
 146                     trans->hmp->io_running_count > hammer_limit_dirtybufs) {
 147                         hammer_sync_unlock(trans);
 148                         hammer_flusher_async(trans->hmp);
 149                         tsleep(trans, 0, "hmrslo", hz / 10);
 150                         hammer_sync_lock_sh(trans);
 151                 }
 152
 153                 /*
 154                  * Check for work
 155                  */
 156                 elm = &cursor.node->ondisk->elms[cursor.index];
 157                 prune->key_cur = elm->base;
 158
 159                 if (prune->stat_oldest_tid > elm->leaf.base.create_tid)
 160                         prune->stat_oldest_tid = elm->leaf.base.create_tid;
 161
 162                 if (hammer_debug_general & 0x0200) {
 163                         kprintf("check %016llx %016llx cre=%016llx del=%016llx\n",
 164                                         elm->base.obj_id,
 165                                         elm->base.key,
 166                                         elm->base.create_tid,
 167                                         elm->base.delete_tid);
 168                 }
 169
 170                 if (check_prune(prune, elm, &realign_cre, &realign_del) == 0) {
 171                         if (hammer_debug_general & 0x0200) {
 172                                 kprintf("check %016llx %016llx: DELETE\n",
 173                                         elm->base.obj_id, elm->base.key);
 174                         }
 175
 176                         /*
 177                          * NOTE: This can return EDEADLK
 178                          *
 179                          * Acquiring the sync lock guarantees that the
 180                          * operation will not cross a synchronization
 181                          * boundary (see the flusher).
 182                          */
 183                         isdir = (elm->base.rec_type == HAMMER_RECTYPE_DIRENTRY);
 184
 185                         error = hammer_delete_at_cursor(&cursor,
 186                                                         &prune->stat_bytes);
 187                         if (error)
 188                                 break;
 189
 190                         if (isdir)
 191                                 ++prune->stat_dirrecords;
 192                         else
 193                                 ++prune->stat_rawrecords;
 194
 195                         /*
 196                          * The current record might now be the one after
 197                          * the one we deleted, set ATEDISK to force us
 198                          * to skip it (since we are iterating backwards).
 199                          */
 200                         cursor.flags |= HAMMER_CURSOR_ATEDISK;
 201                 } else if (realign_cre >= 0 || realign_del >= 0) {
 202                         error = realign_prune(prune, &cursor,
 203                                               realign_cre, realign_del);
 204                         if (error == 0) {
 205                                 cursor.flags |= HAMMER_CURSOR_ATEDISK;
 206                                 if (hammer_debug_general & 0x0200) {
 207                                         kprintf("check %016llx %016llx: "
 208                                                 "REALIGN\n",
 209                                                 elm->base.obj_id,
 210                                                 elm->base.key);
 211                                 }
 212                         }
 213                 } else {
 214                         cursor.flags |= HAMMER_CURSOR_ATEDISK;
 215                         if (hammer_debug_general & 0x0100) {
 216                                 kprintf("check %016llx %016llx: SKIP\n",
 217                                         elm->base.obj_id, elm->base.key);
 218                         }
 219                 }
 220                 ++prune->stat_scanrecords;
 221                 error = hammer_btree_iterate_reverse(&cursor);
 222         }
 223         hammer_sync_unlock(trans);
 224         if (error == ENOENT)
 225                 error = 0;
 226         hammer_done_cursor(&cursor);
 227         if (error == EDEADLK)
 228                 goto retry;
 229         if (error == EINTR) {
 230                 prune->head.flags |= HAMMER_IOC_HEAD_INTR;
 231                 error = 0;
 232         }
 233 failed:
 234         prune->key_cur.localization &= HAMMER_LOCALIZE_MASK;
 235         prune->elms = user_elms;
 236         kfree(copy_elms, M_TEMP);
 237         return(error);
 238 }
 239
 240 /*
 241  * Check pruning list.  The list must be sorted in descending order.
 242  */
 243 static int
 244 check_prune(struct hammer_ioc_prune *prune, hammer_btree_elm_t elm,
 245             int *realign_cre, int *realign_del)
 246 {
 247         struct hammer_ioc_prune_elm *scan;
 248         int i;
 249
 250         *realign_cre = -1;
 251         *realign_del = -1;
 252
 253         /*
 254          * If pruning everything remove all records with a non-zero
 255          * delete_tid.
 256          */
 257         if (prune->head.flags & HAMMER_IOC_PRUNE_ALL) {
 258                 if (elm->base.delete_tid != 0)
 259                         return(0);
 260                 return(-1);
 261         }
 262
 263         for (i = 0; i < prune->nelms; ++i) {
 264                 scan = &prune->elms[i];
 265
 266                 /*
 267                  * Locate the scan index covering the create and delete TIDs.
 268                  */
 269                 if (*realign_cre < 0 &&
 270                     elm->base.create_tid >= scan->beg_tid &&
 271                     elm->base.create_tid < scan->end_tid) {
 272                         *realign_cre = i;
 273                 }
 274                 if (*realign_del < 0 && elm->base.delete_tid &&
 275                     elm->base.delete_tid > scan->beg_tid &&
 276                     elm->base.delete_tid <= scan->end_tid) {
 277                         *realign_del = i;
 278                 }
 279
 280                 /*
 281                  * Now check for loop termination.
 282                  */
 283                 if (elm->base.create_tid >= scan->end_tid ||
 284                     elm->base.delete_tid > scan->end_tid) {
 285                         break;
 286                 }
 287
 288                 /*
 289                  * Now determine if we can delete the record.
 290                  */
 291                 if (elm->base.delete_tid &&
 292                     elm->base.create_tid >= scan->beg_tid &&
 293                     elm->base.delete_tid <= scan->end_tid &&
 294                     (elm->base.create_tid - scan->beg_tid) / scan->mod_tid ==
 295                     (elm->base.delete_tid - scan->beg_tid) / scan->mod_tid) {
 296                         return(0);
 297                 }
 298         }
 299         return(-1);
 300 }
 301
 302 /*
 303  * Align the record to cover any gaps created through the deletion of
 304  * records within the pruning space.  If we were to just delete the records
 305  * there would be gaps which in turn would cause a snapshot that is NOT on
 306  * a pruning boundary to appear corrupt to the user.  Forcing alignment
 307  * of the create_tid and delete_tid for retained records 'reconnects'
 308  * the previously contiguous space, making it contiguous again after the
 309  * deletions.
 310  *
 311  * The use of a reverse iteration allows us to safely align the records and
 312  * related elements without creating temporary overlaps.  XXX we should
 313  * add ordering dependancies for record buffers to guarantee consistency
 314  * during recovery.
 315  */
 316 static int
 317 realign_prune(struct hammer_ioc_prune *prune,
 318               hammer_cursor_t cursor, int realign_cre, int realign_del)
 319 {
 320         struct hammer_ioc_prune_elm *scan;
 321         hammer_btree_elm_t elm;
 322         hammer_tid_t delta;
 323         hammer_tid_t tid;
 324         int error;
 325
 326         hammer_cursor_downgrade(cursor);
 327
 328         elm = &cursor->node->ondisk->elms[cursor->index];
 329         ++prune->stat_realignments;
 330
 331         /*
 332          * Align the create_tid.  By doing a reverse iteration we guarantee
 333          * that all records after our current record have already been
 334          * aligned, allowing us to safely correct the right-hand-boundary
 335          * (because no record to our right is otherwise exactly matching
 336          * will have a create_tid to the left of our aligned create_tid).
 337          */
 338         error = 0;
 339         if (realign_cre >= 0) {
 340                 scan = &prune->elms[realign_cre];
 341
 342                 delta = (elm->leaf.base.create_tid - scan->beg_tid) %
 343                         scan->mod_tid;
 344                 if (delta) {
 345                         tid = elm->leaf.base.create_tid - delta + scan->mod_tid;
 346
 347                         /* can EDEADLK */
 348                         error = hammer_btree_correct_rhb(cursor, tid + 1);
 349                         if (error == 0) {
 350                                 error = hammer_btree_extract(cursor,
 351                                                      HAMMER_CURSOR_GET_LEAF);
 352                         }
 353                         if (error == 0) {
 354                                 /* can EDEADLK */
 355                                 error = hammer_cursor_upgrade(cursor);
 356                         }
 357                         if (error == 0) {
 358                                 hammer_modify_node(cursor->trans, cursor->node,
 359                                             &elm->leaf.base.create_tid,
 360                                             sizeof(elm->leaf.base.create_tid));
 361                                 elm->leaf.base.create_tid = tid;
 362                                 hammer_modify_node_done(cursor->node);
 363                         }
 364                 }
 365         }
 366
 367         /*
 368          * Align the delete_tid.  This only occurs if the record is historical
 369          * was deleted at some point.  Realigning the delete_tid does not
 370          * move the record within the B-Tree but may cause it to temporarily
 371          * overlap a record that has not yet been pruned.
 372          */
 373         if (error == 0 && realign_del >= 0) {
 374                 scan = &prune->elms[realign_del];
 375
 376                 delta = (elm->leaf.base.delete_tid - scan->beg_tid) %
 377                         scan->mod_tid;
 378                 if (delta) {
 379                         error = hammer_btree_extract(cursor,
 380                                                      HAMMER_CURSOR_GET_LEAF);
 381                         if (error == 0) {
 382                                 hammer_modify_node(cursor->trans, cursor->node,
 383                                             &elm->leaf.base.delete_tid,
 384                                             sizeof(elm->leaf.base.delete_tid));
 385                                 elm->leaf.base.delete_tid =
 386                                             elm->leaf.base.delete_tid -
 387                                             delta + scan->mod_tid;
 388                                 hammer_modify_node_done(cursor->node);
 389                         }
 390                 }
 391         }
 392         return (error);
 393 }
 394