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 <linux/locks.h>
  37 #include <asm/segment.h>
  38 #include <asm/uaccess.h>
  39 #include <linux/string.h>
  40 #include <linux/smp_lock.h>
  41 #if defined(CONFIG_EXT3_FS) || defined (CONFIG_EXT3_FS_MODULE)
  42 #include <linux/jbd.h>
  43 #include <linux/ext3_fs.h>
  44 #include <linux/ext3_jbd.h>
  45 #endif
  46
  47 #include <linux/intermezzo_fs.h>
  48 #include <linux/intermezzo_psdev.h>
  49
  50 #if defined(CONFIG_EXT3_FS) || defined (CONFIG_EXT3_FS_MODULE)
  51
  52 #define MAX_PATH_BLOCKS(inode) (PATH_MAX >> EXT3_BLOCK_SIZE_BITS((inode)->i_sb))
  53 #define MAX_NAME_BLOCKS(inode) (NAME_MAX >> EXT3_BLOCK_SIZE_BITS((inode)->i_sb))
  54
  55 /* space requirements:
  56    presto_do_truncate:
  57         used to truncate the KML forward to next fset->chunksize boundary
  58           - zero partial block
  59           - update inode
  60    presto_write_record:
  61         write header (< one block)
  62         write one path (< MAX_PATHLEN)
  63         possibly write another path (< MAX_PATHLEN)
  64         write suffix (< one block)
  65    presto_update_last_rcvd
  66         write one block
  67 */
  68
  69 static loff_t presto_e3_freespace(struct presto_cache *cache,
  70                                          struct super_block *sb)
  71 {
  72         loff_t freebl = le32_to_cpu(sb->u.ext3_sb.s_es->s_free_blocks_count);
  73         loff_t avail =   freebl -
  74                 le32_to_cpu(sb->u.ext3_sb.s_es->s_r_blocks_count);
  75         return (avail <<  EXT3_BLOCK_SIZE_BITS(sb));
  76 }
  77
  78 /* start the filesystem journal operations */
  79 static void *presto_e3_trans_start(struct presto_file_set *fset,
  80                                    struct inode *inode,
  81                                    int op)
  82 {
  83         int jblocks;
  84         int trunc_blks, one_path_blks, extra_path_blks,
  85                 extra_name_blks, lml_blks;
  86         __u32 avail_kmlblocks;
  87         handle_t *handle;
  88
  89         if ( presto_no_journal(fset) ||
  90              strcmp(fset->fset_cache->cache_type, "ext3"))
  91           {
  92             CDEBUG(D_JOURNAL, "got cache_type \"%s\"\n",
  93                    fset->fset_cache->cache_type);
  94             return NULL;
  95           }
  96
  97         avail_kmlblocks = inode->i_sb->u.ext3_sb.s_es->s_free_blocks_count;
  98
  99         if ( avail_kmlblocks < 3 ) {
 100                 return ERR_PTR(-ENOSPC);
 101         }
 102
 103         if (  (op != KML_OPCODE_UNLINK && op != KML_OPCODE_RMDIR)
 104               && avail_kmlblocks < 6 ) {
 105                 return ERR_PTR(-ENOSPC);
 106         }
 107
 108         /* Need journal space for:
 109              at least three writes to KML (two one block writes, one a path)
 110              possibly a second name (unlink, rmdir)
 111              possibly a second path (symlink, rename)
 112              a one block write to the last rcvd file
 113         */
 114
 115         trunc_blks = EXT3_DATA_TRANS_BLOCKS + 1;
 116         one_path_blks = 4*EXT3_DATA_TRANS_BLOCKS + MAX_PATH_BLOCKS(inode) + 3;
 117         lml_blks = 4*EXT3_DATA_TRANS_BLOCKS + MAX_PATH_BLOCKS(inode) + 2;
 118         extra_path_blks = EXT3_DATA_TRANS_BLOCKS + MAX_PATH_BLOCKS(inode);
 119         extra_name_blks = EXT3_DATA_TRANS_BLOCKS + MAX_NAME_BLOCKS(inode);
 120
 121         /* additional blocks appear for "two pathname" operations
 122            and operations involving the LML records
 123         */
 124         switch (op) {
 125         case KML_OPCODE_TRUNC:
 126                 jblocks = one_path_blks + extra_name_blks + trunc_blks
 127                         + EXT3_DELETE_TRANS_BLOCKS;
 128                 break;
 129         case KML_OPCODE_KML_TRUNC:
 130                 /* Hopefully this is a little better, but I'm still mostly
 131                  * guessing here. */
 132                 /* unlink 1 */
 133                 jblocks = extra_name_blks + trunc_blks +
 134                         EXT3_DELETE_TRANS_BLOCKS + 2;
 135
 136                 /* unlink 2 */
 137                 jblocks += extra_name_blks + trunc_blks +
 138                         EXT3_DELETE_TRANS_BLOCKS + 2;
 139
 140                 /* rename 1 */
 141                 jblocks += 2 * extra_path_blks + trunc_blks +
 142                         2 * EXT3_DATA_TRANS_BLOCKS + 2 + 3;
 143
 144                 /* rename 2 */
 145                 jblocks += 2 * extra_path_blks + trunc_blks +
 146                         2 * EXT3_DATA_TRANS_BLOCKS + 2 + 3;
 147                 break;
 148         case KML_OPCODE_RELEASE:
 149                 /*
 150                 jblocks = one_path_blks + lml_blks + 2*trunc_blks;
 151                 */
 152                 jblocks = one_path_blks;
 153                 break;
 154         case KML_OPCODE_SETATTR:
 155                 jblocks = one_path_blks + trunc_blks + 1 ;
 156                 break;
 157         case KML_OPCODE_CREATE:
 158                 jblocks = one_path_blks + trunc_blks
 159                         + EXT3_DATA_TRANS_BLOCKS + 3 + 2;
 160                 break;
 161         case KML_OPCODE_LINK:
 162                 jblocks = one_path_blks + trunc_blks
 163                         + EXT3_DATA_TRANS_BLOCKS + 2;
 164                 break;
 165         case KML_OPCODE_UNLINK:
 166                 jblocks = one_path_blks + extra_name_blks + trunc_blks
 167                         + EXT3_DELETE_TRANS_BLOCKS + 2;
 168                 break;
 169         case KML_OPCODE_SYMLINK:
 170                 jblocks = one_path_blks + extra_path_blks + trunc_blks
 171                         + EXT3_DATA_TRANS_BLOCKS + 5;
 172                 break;
 173         case KML_OPCODE_MKDIR:
 174                 jblocks = one_path_blks + trunc_blks
 175                         + EXT3_DATA_TRANS_BLOCKS + 4 + 2;
 176                 break;
 177         case KML_OPCODE_RMDIR:
 178                 jblocks = one_path_blks + extra_name_blks + trunc_blks
 179                         + EXT3_DELETE_TRANS_BLOCKS + 1;
 180                 break;
 181         case KML_OPCODE_MKNOD:
 182                 jblocks = one_path_blks + trunc_blks +
 183                         EXT3_DATA_TRANS_BLOCKS + 3 + 2;
 184                 break;
 185         case KML_OPCODE_RENAME:
 186                 jblocks = one_path_blks + extra_path_blks + trunc_blks +
 187                         2 * EXT3_DATA_TRANS_BLOCKS + 2 + 3;
 188                 break;
 189         case KML_OPCODE_WRITE:
 190                 jblocks = one_path_blks;
 191                 /*  add this when we can wrap our transaction with
 192                     that of ext3_file_write (ordered writes)
 193                     +  EXT3_DATA_TRANS_BLOCKS;
 194                 */
 195                 break;
 196         default:
 197                 CDEBUG(D_JOURNAL, "invalid operation %d for journal\n", op);
 198                 return NULL;
 199         }
 200
 201         CDEBUG(D_JOURNAL, "creating journal handle (%d blocks) for op %d\n",
 202                jblocks, op);
 203         /* journal_start/stop does not do its own locking while updating
 204          * the handle/transaction information. Hence we create our own
 205          * critical section to protect these calls. -SHP
 206          */
 207         lock_kernel();
 208         handle = journal_start(EXT3_JOURNAL(inode), jblocks);
 209         unlock_kernel();
 210         return handle;
 211 }
 212
 213 static void presto_e3_trans_commit(struct presto_file_set *fset, void *handle)
 214 {
 215         if ( presto_no_journal(fset) || !handle)
 216                 return;
 217
 218         /* See comments before journal_start above. -SHP */
 219         lock_kernel();
 220         journal_stop(handle);
 221         unlock_kernel();
 222 }
 223
 224 static void presto_e3_journal_file_data(struct inode *inode)
 225 {
 226 #ifdef EXT3_JOURNAL_DATA_FL
 227         inode->u.ext3_i.i_flags |= EXT3_JOURNAL_DATA_FL;
 228 #else
 229 #warning You must have a facility to enable journaled writes for recovery!
 230 #endif
 231 }
 232
 233 /* The logic here is a slightly modified version of ext3/inode.c:block_to_path
 234  */
 235 static int presto_e3_has_all_data(struct inode *inode)
 236 {
 237         int ptrs = EXT3_ADDR_PER_BLOCK(inode->i_sb);
 238         int ptrs_bits = EXT3_ADDR_PER_BLOCK_BITS(inode->i_sb);
 239         const long direct_blocks = EXT3_NDIR_BLOCKS,
 240                 indirect_blocks = ptrs,
 241                 double_blocks = (1 << (ptrs_bits * 2));
 242         long block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
 243                 inode->i_sb->s_blocksize_bits;
 244
 245         ENTRY;
 246
 247         if (inode->i_size == 0) {
 248                 EXIT;
 249                 return 1;
 250         }
 251
 252         if (block < direct_blocks) {
 253                 /* No indirect blocks, no problem. */
 254         } else if (block < indirect_blocks + direct_blocks) {
 255                 block++;
 256         } else if (block < double_blocks + indirect_blocks + direct_blocks) {
 257                 block += 2;
 258         } else if (((block - double_blocks - indirect_blocks - direct_blocks)
 259                     >> (ptrs_bits * 2)) < ptrs) {
 260                 block += 3;
 261         }
 262
 263         block *= (inode->i_sb->s_blocksize / 512);
 264
 265         CDEBUG(D_CACHE, "Need %ld blocks, have %ld.\n", block, inode->i_blocks);
 266
 267         if (block > inode->i_blocks) {
 268                 EXIT;
 269                 return 0;
 270         }
 271
 272         EXIT;
 273         return 1;
 274 }
 275
 276 struct journal_ops presto_ext3_journal_ops = {
 277         .tr_all_data     = presto_e3_has_all_data,
 278         .tr_avail        = presto_e3_freespace,
 279         .tr_start        =  presto_e3_trans_start,
 280         .tr_commit       = presto_e3_trans_commit,
 281         .tr_journal_data = presto_e3_journal_file_data,
 282         .tr_ilookup      = presto_iget_ilookup
 283 };
 284
 285 #endif /* CONFIG_EXT3_FS */