fs/intermezzo/journal_ext3.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  *  Copyright (C) 1998 Peter J. Braam <braam@clusterfs.com>
   5  *  Copyright (C) 2000 Red Hat, Inc.
   6  *  Copyright (C) 2000 Los Alamos National Laboratory
   7  *  Copyright (C) 2000 TurboLinux, Inc.
   8  *  Copyright (C) 2001 Mountain View Data, Inc.
   9  *  Copyright (C) 2001 Tacit Networks, Inc. <phil@off.net>
  10  *
  11  *   This file is part of InterMezzo, http://www.inter-mezzo.org.
  12  *
  13  *   InterMezzo is free software; you can redistribute it and/or
  14  *   modify it under the terms of version 2 of the GNU General Public
  15  *   License as published by the Free Software Foundation.
  16  *
  17  *   InterMezzo is distributed in the hope that it will be useful,
  18  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  19  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  20  *   GNU General Public License for more details.
  21  *
  22  *   You should have received a copy of the GNU General Public License
  23  *   along with InterMezzo; if not, write to the Free Software
  24  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  25  */
  26
  27 #include <linux/types.h>
  28 #include <linux/param.h>
  29 #include <linux/kernel.h>
  30 #include <linux/sched.h>
  31 #include <linux/fs.h>
  32 #include <linux/slab.h>
  33 #include <linux/vmalloc.h>
  34 #include <linux/stat.h>
  35 #include <linux/errno.h>
  36 #include <asm/segment.h>
  37 #include <asm/uaccess.h>
  38 #include <linux/string.h>
  39 #if defined(CONFIG_EXT3_FS) || defined (CONFIG_EXT3_FS_MODULE)
  40 #include <linux/jbd.h>
  41 #include <linux/ext3_fs.h>
  42 #include <linux/ext3_jbd.h>
  43 #endif
  44
  45 #include "intermezzo_fs.h"
  46 #include "intermezzo_psdev.h"
  47
  48 #if defined(CONFIG_EXT3_FS) || defined (CONFIG_EXT3_FS_MODULE)
  49
  50 #define MAX_PATH_BLOCKS(inode) (PATH_MAX >> EXT3_BLOCK_SIZE_BITS((inode)->i_sb))
  51 #define MAX_NAME_BLOCKS(inode) (NAME_MAX >> EXT3_BLOCK_SIZE_BITS((inode)->i_sb))
  52
  53 /* space requirements:
  54    presto_do_truncate:
  55         used to truncate the KML forward to next fset->chunksize boundary
  56           - zero partial block
  57           - update inode
  58    presto_write_record:
  59         write header (< one block)
  60         write one path (< MAX_PATHLEN)
  61         possibly write another path (< MAX_PATHLEN)
  62         write suffix (< one block)
  63    presto_update_last_rcvd
  64         write one block
  65 */
  66
  67 static loff_t presto_e3_freespace(struct presto_cache *cache,
  68                                          struct super_block *sb)
  69 {
  70         loff_t freebl = le32_to_cpu(EXT3_SB(sb)->s_es->s_free_blocks_count);
  71         loff_t avail =   freebl -
  72                 le32_to_cpu(EXT3_SB(sb)->s_es->s_r_blocks_count);
  73         return (avail <<  EXT3_BLOCK_SIZE_BITS(sb));
  74 }
  75
  76 /* start the filesystem journal operations */
  77 static void *presto_e3_trans_start(struct presto_file_set *fset,
  78                                    struct inode *inode,
  79                                    int op)
  80 {
  81         int jblocks;
  82         int trunc_blks, one_path_blks, extra_path_blks,
  83                 extra_name_blks, lml_blks;
  84         __u32 avail_kmlblocks;
  85         handle_t *handle;
  86
  87         if ( presto_no_journal(fset) ||
  88              strcmp(fset->fset_cache->cache_type, "ext3"))
  89           {
  90             CDEBUG(D_JOURNAL, "got cache_type \"%s\"\n",
  91                    fset->fset_cache->cache_type);
  92             return NULL;
  93           }
  94
  95         avail_kmlblocks = EXT3_SB(inode->i_sb)->s_es->s_free_blocks_count;
  96
  97         if ( avail_kmlblocks < 3 ) {
  98                 return ERR_PTR(-ENOSPC);
  99         }
 100
 101         if (  (op != KML_OPCODE_UNLINK && op != KML_OPCODE_RMDIR)
 102               && avail_kmlblocks < 6 ) {
 103                 return ERR_PTR(-ENOSPC);
 104         }
 105
 106         /* Need journal space for:
 107              at least three writes to KML (two one block writes, one a path)
 108              possibly a second name (unlink, rmdir)
 109              possibly a second path (symlink, rename)
 110              a one block write to the last rcvd file
 111         */
 112
 113         trunc_blks = EXT3_DATA_TRANS_BLOCKS + 1;
 114         one_path_blks = 4*EXT3_DATA_TRANS_BLOCKS + MAX_PATH_BLOCKS(inode) + 3;
 115         lml_blks = 4*EXT3_DATA_TRANS_BLOCKS + MAX_PATH_BLOCKS(inode) + 2;
 116         extra_path_blks = EXT3_DATA_TRANS_BLOCKS + MAX_PATH_BLOCKS(inode);
 117         extra_name_blks = EXT3_DATA_TRANS_BLOCKS + MAX_NAME_BLOCKS(inode);
 118
 119         /* additional blocks appear for "two pathname" operations
 120            and operations involving the LML records
 121         */
 122         switch (op) {
 123         case KML_OPCODE_TRUNC:
 124                 jblocks = one_path_blks + extra_name_blks + trunc_blks
 125                         + EXT3_DELETE_TRANS_BLOCKS;
 126                 break;
 127         case KML_OPCODE_KML_TRUNC:
 128                 /* Hopefully this is a little better, but I'm still mostly
 129                  * guessing here. */
 130                 /* unlink 1 */
 131                 jblocks = extra_name_blks + trunc_blks +
 132                         EXT3_DELETE_TRANS_BLOCKS + 2;
 133
 134                 /* unlink 2 */
 135                 jblocks += extra_name_blks + trunc_blks +
 136                         EXT3_DELETE_TRANS_BLOCKS + 2;
 137
 138                 /* rename 1 */
 139                 jblocks += 2 * extra_path_blks + trunc_blks +
 140                         2 * EXT3_DATA_TRANS_BLOCKS + 2 + 3;
 141
 142                 /* rename 2 */
 143                 jblocks += 2 * extra_path_blks + trunc_blks +
 144                         2 * EXT3_DATA_TRANS_BLOCKS + 2 + 3;
 145                 break;
 146         case KML_OPCODE_RELEASE:
 147                 /*
 148                 jblocks = one_path_blks + lml_blks + 2*trunc_blks;
 149                 */
 150                 jblocks = one_path_blks;
 151                 break;
 152         case KML_OPCODE_SETATTR:
 153                 jblocks = one_path_blks + trunc_blks + 1 ;
 154                 break;
 155         case KML_OPCODE_CREATE:
 156                 jblocks = one_path_blks + trunc_blks
 157                         + EXT3_DATA_TRANS_BLOCKS + 3 + 2;
 158                 break;
 159         case KML_OPCODE_LINK:
 160                 jblocks = one_path_blks + trunc_blks
 161                         + EXT3_DATA_TRANS_BLOCKS + 2;
 162                 break;
 163         case KML_OPCODE_UNLINK:
 164                 jblocks = one_path_blks + extra_name_blks + trunc_blks
 165                         + EXT3_DELETE_TRANS_BLOCKS + 2;
 166                 break;
 167         case KML_OPCODE_SYMLINK:
 168                 jblocks = one_path_blks + extra_path_blks + trunc_blks
 169                         + EXT3_DATA_TRANS_BLOCKS + 5;
 170                 break;
 171         case KML_OPCODE_MKDIR:
 172                 jblocks = one_path_blks + trunc_blks
 173                         + EXT3_DATA_TRANS_BLOCKS + 4 + 2;
 174                 break;
 175         case KML_OPCODE_RMDIR:
 176                 jblocks = one_path_blks + extra_name_blks + trunc_blks
 177                         + EXT3_DELETE_TRANS_BLOCKS + 1;
 178                 break;
 179         case KML_OPCODE_MKNOD:
 180                 jblocks = one_path_blks + trunc_blks +
 181                         EXT3_DATA_TRANS_BLOCKS + 3 + 2;
 182                 break;
 183         case KML_OPCODE_RENAME:
 184                 jblocks = one_path_blks + extra_path_blks + trunc_blks +
 185                         2 * EXT3_DATA_TRANS_BLOCKS + 2 + 3;
 186                 break;
 187         case KML_OPCODE_WRITE:
 188                 jblocks = one_path_blks;
 189                 /*  add this when we can wrap our transaction with
 190                     that of ext3_file_write (ordered writes)
 191                     +  EXT3_DATA_TRANS_BLOCKS;
 192                 */
 193                 break;
 194         default:
 195                 CDEBUG(D_JOURNAL, "invalid operation %d for journal\n", op);
 196                 return NULL;
 197         }
 198
 199         CDEBUG(D_JOURNAL, "creating journal handle (%d blocks) for op %d\n",
 200                jblocks, op);
 201         /* journal_start/stop does not do its own locking while updating
 202          * the handle/transaction information. Hence we create our own
 203          * critical section to protect these calls. -SHP
 204          */
 205         lock_kernel();
 206         handle = journal_start(EXT3_JOURNAL(inode), jblocks);
 207         unlock_kernel();
 208         return handle;
 209 }
 210
 211 static void presto_e3_trans_commit(struct presto_file_set *fset, void *handle)
 212 {
 213         if ( presto_no_journal(fset) || !handle)
 214                 return;
 215
 216         /* See comments before journal_start above. -SHP */
 217         lock_kernel();
 218         journal_stop(handle);
 219         unlock_kernel();
 220 }
 221
 222 static void presto_e3_journal_file_data(struct inode *inode)
 223 {
 224 #ifdef EXT3_JOURNAL_DATA_FL
 225         EXT3_I(inode)->i_flags |= EXT3_JOURNAL_DATA_FL;
 226 #else
 227 #warning You must have a facility to enable journaled writes for recovery!
 228 #endif
 229 }
 230
 231 /* The logic here is a slightly modified version of ext3/inode.c:block_to_path
 232  */
 233 static int presto_e3_has_all_data(struct inode *inode)
 234 {
 235         int ptrs = EXT3_ADDR_PER_BLOCK(inode->i_sb);
 236         int ptrs_bits = EXT3_ADDR_PER_BLOCK_BITS(inode->i_sb);
 237         const long direct_blocks = EXT3_NDIR_BLOCKS,
 238                 indirect_blocks = ptrs,
 239                 double_blocks = (1 << (ptrs_bits * 2));
 240         long block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
 241                 inode->i_sb->s_blocksize_bits;
 242
 243         ENTRY;
 244
 245         if (inode->i_size == 0) {
 246                 EXIT;
 247                 return 1;
 248         }
 249
 250         if (block < direct_blocks) {
 251                 /* No indirect blocks, no problem. */
 252         } else if (block < indirect_blocks + direct_blocks) {
 253                 block++;
 254         } else if (block < double_blocks + indirect_blocks + direct_blocks) {
 255                 block += 2;
 256         } else if (((block - double_blocks - indirect_blocks - direct_blocks)
 257                     >> (ptrs_bits * 2)) < ptrs) {
 258                 block += 3;
 259         }
 260
 261         block *= (inode->i_sb->s_blocksize / 512);
 262
 263         CDEBUG(D_CACHE, "Need %ld blocks, have %ld.\n", block, inode->i_blocks);
 264
 265         if (block > inode->i_blocks) {
 266                 EXIT;
 267                 return 0;
 268         }
 269
 270         EXIT;
 271         return 1;
 272 }
 273
 274 struct journal_ops presto_ext3_journal_ops = {
 275         .tr_all_data     = presto_e3_has_all_data,
 276         .tr_avail        = presto_e3_freespace,
 277         .tr_start        =  presto_e3_trans_start,
 278         .tr_commit       = presto_e3_trans_commit,
 279         .tr_journal_data = presto_e3_journal_file_data,
 280         .tr_ilookup      = presto_iget_ilookup
 281 };
 282
 283 #endif /* CONFIG_EXT3_FS */