sys/vfs/hammer/hammer_undo.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_undo.c,v 1.20 2008/07/18 00:19:53 dillon Exp $
  35  */
  36
  37 /*
  38  * HAMMER undo - undo buffer/FIFO management.
  39  */
  40
  41 #include "hammer.h"
  42
  43 static int hammer_und_rb_compare(hammer_undo_t node1, hammer_undo_t node2);
  44
  45 RB_GENERATE2(hammer_und_rb_tree, hammer_undo, rb_node,
  46              hammer_und_rb_compare, hammer_off_t, offset);
  47
  48 /*
  49  * Convert a zone-3 undo offset into a zone-2 buffer offset.
  50  */
  51 hammer_off_t
  52 hammer_undo_lookup(hammer_mount_t hmp, hammer_off_t zone3_off, int *errorp)
  53 {
  54         hammer_volume_t root_volume;
  55         hammer_blockmap_t undomap;
  56         hammer_off_t result_offset;
  57         int i;
  58
  59         KKASSERT((zone3_off & HAMMER_OFF_ZONE_MASK) == HAMMER_ZONE_UNDO);
  60         root_volume = hammer_get_root_volume(hmp, errorp);
  61         if (*errorp)
  62                 return(0);
  63         undomap = &hmp->blockmap[HAMMER_ZONE_UNDO_INDEX];
  64         KKASSERT(HAMMER_ZONE_DECODE(undomap->alloc_offset) == HAMMER_ZONE_UNDO_INDEX);
  65         KKASSERT (zone3_off < undomap->alloc_offset);
  66
  67         i = (zone3_off & HAMMER_OFF_SHORT_MASK) / HAMMER_LARGEBLOCK_SIZE;
  68         result_offset = root_volume->ondisk->vol0_undo_array[i] +
  69                         (zone3_off & HAMMER_LARGEBLOCK_MASK64);
  70
  71         hammer_rel_volume(root_volume, 0);
  72         return(result_offset);
  73 }
  74
  75 /*
  76  * Generate an UNDO record for the block of data at the specified zone1
  77  * or zone2 offset.
  78  *
  79  * The recovery code will execute UNDOs in reverse order, allowing overlaps.
  80  * All the UNDOs are executed together so if we already laid one down we
  81  * do not have to lay another one down for the same range.
  82  */
  83 int
  84 hammer_generate_undo(hammer_transaction_t trans, hammer_io_t io,
  85                      hammer_off_t zone_off, void *base, int len)
  86 {
  87         hammer_mount_t hmp;
  88         hammer_volume_t root_volume;
  89         hammer_blockmap_t undomap;
  90         hammer_buffer_t buffer = NULL;
  91         hammer_fifo_undo_t undo;
  92         hammer_fifo_tail_t tail;
  93         hammer_off_t next_offset;
  94         int error;
  95         int bytes;
  96
  97         hmp = trans->hmp;
  98
  99         /*
 100          * Enter the offset into our undo history.  If there is an existing
 101          * undo we do not have to generate a new one.
 102          */
 103         if (hammer_enter_undo_history(hmp, zone_off, len) == EALREADY)
 104                 return(0);
 105
 106         root_volume = trans->rootvol;
 107         undomap = &hmp->blockmap[HAMMER_ZONE_UNDO_INDEX];
 108
 109         /* no undo recursion */
 110         hammer_modify_volume(NULL, root_volume, NULL, 0);
 111
 112         hammer_lock_ex(&hmp->undo_lock);
 113 again:
 114         /*
 115          * Allocate space in the FIFO
 116          */
 117         bytes = ((len + HAMMER_HEAD_ALIGN_MASK) & ~HAMMER_HEAD_ALIGN_MASK) +
 118                 sizeof(struct hammer_fifo_undo) +
 119                 sizeof(struct hammer_fifo_tail);
 120         if (hammer_undo_space(trans) < bytes + HAMMER_BUFSIZE*2)
 121                 panic("hammer: insufficient undo FIFO space!");
 122
 123         next_offset = undomap->next_offset;
 124
 125         /*
 126          * Wrap next_offset
 127          */
 128         if (undomap->next_offset == undomap->alloc_offset) {
 129                 next_offset = HAMMER_ZONE_ENCODE(HAMMER_ZONE_UNDO_INDEX, 0);
 130                 undomap->next_offset = next_offset;
 131         }
 132
 133         /*
 134          * This is a tail-chasing FIFO, when we hit the start of a new
 135          * buffer we don't have to read it in.
 136          */
 137         if ((next_offset & HAMMER_BUFMASK) == 0)
 138                 undo = hammer_bnew(hmp, next_offset, &error, &buffer);
 139         else
 140                 undo = hammer_bread(hmp, next_offset, &error, &buffer);
 141         if (error)
 142                 goto done;
 143
 144         hammer_modify_buffer(NULL, buffer, NULL, 0);
 145
 146         KKASSERT(undomap->next_offset == next_offset);
 147
 148         /*
 149          * The FIFO entry would cross a buffer boundary, PAD to the end
 150          * of the buffer and try again.  Due to our data alignment, the
 151          * worst case (smallest) PAD record is 8 bytes.  PAD records only
 152          * populate the first 8 bytes of hammer_fifo_head and the tail may
 153          * be at the same offset as the head.
 154          */
 155         if ((next_offset ^ (next_offset + bytes)) & ~HAMMER_BUFMASK64) {
 156                 bytes = HAMMER_BUFSIZE - ((int)next_offset & HAMMER_BUFMASK);
 157                 tail = (void *)((char *)undo + bytes - sizeof(*tail));
 158                 if ((void *)undo != (void *)tail) {
 159                         tail->tail_signature = HAMMER_TAIL_SIGNATURE;
 160                         tail->tail_type = HAMMER_HEAD_TYPE_PAD;
 161                         tail->tail_size = bytes;
 162                 }
 163                 undo->head.hdr_signature = HAMMER_HEAD_SIGNATURE;
 164                 undo->head.hdr_type = HAMMER_HEAD_TYPE_PAD;
 165                 undo->head.hdr_size = bytes;
 166                 /* NO CRC */
 167                 undomap->next_offset += bytes;
 168                 hammer_modify_buffer_done(buffer);
 169                 hammer_stats_undo += bytes;
 170                 goto again;
 171         }
 172         if (hammer_debug_general & 0x0080) {
 173                 kprintf("undo %016llx %d %d\n",
 174                         (long long)next_offset, bytes, len);
 175         }
 176
 177         /*
 178          * We're good, create the entry.
 179          */
 180         undo->head.hdr_signature = HAMMER_HEAD_SIGNATURE;
 181         undo->head.hdr_type = HAMMER_HEAD_TYPE_UNDO;
 182         undo->head.hdr_size = bytes;
 183         undo->head.reserved01 = 0;
 184         undo->head.hdr_crc = 0;
 185         undo->undo_offset = zone_off;
 186         undo->undo_data_bytes = len;
 187         bcopy(base, undo + 1, len);
 188
 189         tail = (void *)((char *)undo + bytes - sizeof(*tail));
 190         tail->tail_signature = HAMMER_TAIL_SIGNATURE;
 191         tail->tail_type = HAMMER_HEAD_TYPE_UNDO;
 192         tail->tail_size = bytes;
 193
 194         KKASSERT(bytes >= sizeof(undo->head));
 195         undo->head.hdr_crc = crc32(undo, HAMMER_FIFO_HEAD_CRCOFF) ^
 196                              crc32(&undo->head + 1, bytes - sizeof(undo->head));
 197         undomap->next_offset += bytes;
 198         hammer_stats_undo += bytes;
 199
 200         hammer_modify_buffer_done(buffer);
 201 done:
 202         hammer_modify_volume_done(root_volume);
 203         hammer_unlock(&hmp->undo_lock);
 204
 205         if (buffer)
 206                 hammer_rel_buffer(buffer, 0);
 207         return(error);
 208 }
 209
 210 /*
 211  * UNDO HISTORY API
 212  *
 213  * It is not necessary to layout an undo record for the same address space
 214  * multiple times.  Maintain a cache of recent undo's.
 215  */
 216
 217 /*
 218  * Enter an undo into the history.  Return EALREADY if the request completely
 219  * covers a previous request.
 220  */
 221 int
 222 hammer_enter_undo_history(hammer_mount_t hmp, hammer_off_t offset, int bytes)
 223 {
 224         hammer_undo_t node;
 225         hammer_undo_t onode;
 226
 227         node = RB_LOOKUP(hammer_und_rb_tree, &hmp->rb_undo_root, offset);
 228         if (node) {
 229                 TAILQ_REMOVE(&hmp->undo_lru_list, node, lru_entry);
 230                 TAILQ_INSERT_TAIL(&hmp->undo_lru_list, node, lru_entry);
 231                 if (bytes <= node->bytes)
 232                         return(EALREADY);
 233                 node->bytes = bytes;
 234                 return(0);
 235         }
 236         if (hmp->undo_alloc != HAMMER_MAX_UNDOS) {
 237                 node = &hmp->undos[hmp->undo_alloc++];
 238         } else {
 239                 node = TAILQ_FIRST(&hmp->undo_lru_list);
 240                 TAILQ_REMOVE(&hmp->undo_lru_list, node, lru_entry);
 241                 RB_REMOVE(hammer_und_rb_tree, &hmp->rb_undo_root, node);
 242         }
 243         node->offset = offset;
 244         node->bytes = bytes;
 245         TAILQ_INSERT_TAIL(&hmp->undo_lru_list, node, lru_entry);
 246         onode = RB_INSERT(hammer_und_rb_tree, &hmp->rb_undo_root, node);
 247         KKASSERT(onode == NULL);
 248         return(0);
 249 }
 250
 251 void
 252 hammer_clear_undo_history(hammer_mount_t hmp)
 253 {
 254         RB_INIT(&hmp->rb_undo_root);
 255         TAILQ_INIT(&hmp->undo_lru_list);
 256         hmp->undo_alloc = 0;
 257 }
 258
 259 /*
 260  * Return how much of the undo FIFO has been used
 261  *
 262  * The calculation includes undo FIFO space still reserved from a previous
 263  * flush (because it will still be run on recovery if a crash occurs and
 264  * we can't overwrite it yet).
 265  */
 266 int64_t
 267 hammer_undo_used(hammer_transaction_t trans)
 268 {
 269         hammer_blockmap_t cundomap;
 270         hammer_blockmap_t dundomap;
 271         int64_t max_bytes;
 272         int64_t bytes;
 273
 274         cundomap = &trans->hmp->blockmap[HAMMER_ZONE_UNDO_INDEX];
 275         dundomap = &trans->rootvol->ondisk->
 276                                 vol0_blockmap[HAMMER_ZONE_UNDO_INDEX];
 277
 278         if (dundomap->first_offset <= cundomap->next_offset) {
 279                 bytes = cundomap->next_offset - dundomap->first_offset;
 280         } else {
 281                 bytes = cundomap->alloc_offset - dundomap->first_offset +
 282                         (cundomap->next_offset & HAMMER_OFF_LONG_MASK);
 283         }
 284         max_bytes = cundomap->alloc_offset & HAMMER_OFF_SHORT_MASK;
 285         KKASSERT(bytes <= max_bytes);
 286         return(bytes);
 287 }
 288
 289 /*
 290  * Return how much of the undo FIFO is available for new records.
 291  */
 292 int64_t
 293 hammer_undo_space(hammer_transaction_t trans)
 294 {
 295         hammer_blockmap_t rootmap;
 296         int64_t max_bytes;
 297
 298         rootmap = &trans->hmp->blockmap[HAMMER_ZONE_UNDO_INDEX];
 299         max_bytes = rootmap->alloc_offset & HAMMER_OFF_SHORT_MASK;
 300         return(max_bytes - hammer_undo_used(trans));
 301 }
 302
 303 int64_t
 304 hammer_undo_max(hammer_mount_t hmp)
 305 {
 306         hammer_blockmap_t rootmap;
 307         int64_t max_bytes;
 308
 309         rootmap = &hmp->blockmap[HAMMER_ZONE_UNDO_INDEX];
 310         max_bytes = rootmap->alloc_offset & HAMMER_OFF_SHORT_MASK;
 311
 312         return(max_bytes);
 313 }
 314
 315 /*
 316  * Returns 1 if the undo buffer should be reclaimed on release.  The
 317  * only undo buffer we do NOT want to reclaim is the one at the current
 318  * append offset.
 319  */
 320 int
 321 hammer_undo_reclaim(hammer_io_t io)
 322 {
 323         hammer_blockmap_t undomap;
 324         hammer_off_t next_offset;
 325
 326         undomap = &io->hmp->blockmap[HAMMER_ZONE_UNDO_INDEX];
 327         next_offset = undomap->next_offset & ~HAMMER_BUFMASK64;
 328         if (((struct hammer_buffer *)io)->zoneX_offset == next_offset)
 329                 return(0);
 330         return(1);
 331 }
 332
 333 static int
 334 hammer_und_rb_compare(hammer_undo_t node1, hammer_undo_t node2)
 335 {
 336         if (node1->offset < node2->offset)
 337                 return(-1);
 338         if (node1->offset > node2->offset)
 339                 return(1);
 340         return(0);
 341 }
 342