ldiskfs/kernel_patches/patches/ext3-mmp-2.6-rhel4.patch

   1 Index: linux-2.6.9-67.0.22/fs/ext3/super.c
   2 ===================================================================
   3 --- linux-2.6.9-67.0.22.orig/fs/ext3/super.c
   4 +++ linux-2.6.9-67.0.22/fs/ext3/super.c
   5 @@ -35,6 +35,8 @@
   6  #include <linux/mount.h>
   7  #include <linux/namei.h>
   8  #include <linux/quotaops.h>
   9 +#include <linux/kthread.h>
  10 +#include <linux/utsname.h>
  11  #include <asm/uaccess.h>
  12  #include "xattr.h"
  13  #include "acl.h"
  14 @@ -440,6 +442,8 @@ void ext3_put_super (struct super_block
  15                 invalidate_bdev(sbi->journal_bdev, 0);
  16                 ext3_blkdev_remove(sbi);
  17         }
  18 +       if (sbi->s_mmp_tsk)
  19 +               kthread_stop(sbi->s_mmp_tsk);
  20         if (sbi->s_dev_proc) {
  21                 remove_proc_entry(sbi->s_dev_proc->name, proc_root_ext3);
  22                 sbi->s_dev_proc = NULL;
  23 @@ -1441,6 +1445,330 @@ static unsigned long descriptor_loc(stru
  24         return (first_data_block + has_super + (bg * sbi->s_blocks_per_group));
  25  }
  26
  27 +/*
  28 + * Write the MMP block using WRITE_SYNC to try to get the block on-disk
  29 + * faster.
  30 + */
  31 +static int write_mmp_block(struct buffer_head *bh)
  32 +{
  33 +       mark_buffer_dirty(bh);
  34 +       lock_buffer(bh);
  35 +       bh->b_end_io = end_buffer_write_sync;
  36 +       get_bh(bh);
  37 +       submit_bh(WRITE_SYNC, bh);
  38 +       wait_on_buffer(bh);
  39 +       if (unlikely(!buffer_uptodate(bh)))
  40 +               return 1;
  41 +
  42 +       return 0;
  43 +}
  44 +
  45 +/*
  46 + * Read the MMP block. It _must_ be read from disk and hence we clear the
  47 + * uptodate flag on the buffer.
  48 + */
  49 +static int read_mmp_block(struct super_block *sb, struct buffer_head **bh,
  50 +                         unsigned long mmp_block)
  51 +{
  52 +       struct mmp_struct *mmp;
  53 +
  54 +       if (*bh)
  55 +               clear_buffer_uptodate(*bh);
  56 +
  57 +#if 0
  58 +       brelse(*bh);
  59 +
  60 +       *bh = sb_bread(sb, mmp_block);
  61 +#else
  62 +       if (!*bh)
  63 +               *bh = sb_getblk(sb, mmp_block);
  64 +       if (*bh) {
  65 +               get_bh(*bh);
  66 +               lock_buffer(*bh);
  67 +               (*bh)->b_end_io = end_buffer_read_sync;
  68 +               submit_bh(READ_SYNC, *bh);
  69 +               wait_on_buffer(*bh);
  70 +               if (!buffer_uptodate(*bh)) {
  71 +                       brelse(*bh);
  72 +                       *bh = NULL;
  73 +               }
  74 +       }
  75 +#endif
  76 +       if (!*bh) {
  77 +               ext3_warning(sb, __FUNCTION__,
  78 +                            "Error while reading MMP block %lu", mmp_block);
  79 +               return -EIO;
  80 +       }
  81 +
  82 +       mmp = (struct mmp_struct *)((*bh)->b_data);
  83 +       if (le32_to_cpu(mmp->mmp_magic) != EXT3_MMP_MAGIC)
  84 +               return -EINVAL;
  85 +
  86 +       return 0;
  87 +}
  88 +
  89 +/*
  90 + * Dump as much information as possible to help the admin.
  91 + */
  92 +static void dump_mmp_msg(struct super_block *sb, struct mmp_struct *mmp,
  93 +                        const char *function, const char *msg)
  94 +{
  95 +       ext3_warning(sb, function, msg);
  96 +       ext3_warning(sb, function, "MMP failure info: last update time: %llu, "
  97 +                    "last update node: %s, last update device: %s\n",
  98 +                    le64_to_cpu(mmp->mmp_time), mmp->mmp_nodename,
  99 +                    mmp->mmp_bdevname);
 100 +}
 101 +
 102 +/*
 103 + * kmmpd will update the MMP sequence every s_mmp_update_interval seconds
 104 + */
 105 +static int kmmpd(void *data)
 106 +{
 107 +       struct super_block *sb = (struct super_block *) data;
 108 +       struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 109 +       struct buffer_head *bh = NULL;
 110 +       struct mmp_struct *mmp;
 111 +       unsigned long mmp_block;
 112 +       u32 seq = 0;
 113 +       unsigned long failed_writes = 0;
 114 +       int mmp_update_interval = le16_to_cpu(es->s_mmp_update_interval);
 115 +       unsigned mmp_check_interval;
 116 +       unsigned long last_update_time;
 117 +       unsigned long diff;
 118 +       int retval;
 119 +
 120 +       mmp_block = le64_to_cpu(es->s_mmp_block);
 121 +       retval = read_mmp_block(sb, &bh, mmp_block);
 122 +       if (retval)
 123 +               goto failed;
 124 +
 125 +       mmp = (struct mmp_struct *)(bh->b_data);
 126 +       mmp->mmp_time = cpu_to_le64(get_seconds());
 127 +       /*
 128 +        * Start with the higher mmp_check_interval and reduce it if
 129 +        * the MMP block is being updated on time.
 130 +        */
 131 +       mmp_check_interval = max(5 * mmp_update_interval,
 132 +                                EXT3_MMP_MIN_CHECK_INTERVAL);
 133 +       mmp->mmp_check_interval = cpu_to_le16(mmp_check_interval);
 134 +       bdevname(bh->b_bdev, mmp->mmp_bdevname);
 135 +
 136 +       down_read(&uts_sem);
 137 +       memcpy(mmp->mmp_nodename, system_utsname.nodename,
 138 +              sizeof(mmp->mmp_nodename));
 139 +       up_read(&uts_sem);
 140 +
 141 +       while (!kthread_should_stop()) {
 142 +               if (++seq > EXT3_MMP_SEQ_MAX)
 143 +                       seq = 1;
 144 +
 145 +               mmp->mmp_seq = cpu_to_le32(seq);
 146 +               mmp->mmp_time = cpu_to_le64(get_seconds());
 147 +               last_update_time = jiffies;
 148 +
 149 +               retval = write_mmp_block(bh);
 150 +               /*
 151 +                * Don't spew too many error messages. Print one every
 152 +                * (s_mmp_update_interval * 60) seconds.
 153 +                */
 154 +               if (retval && (failed_writes % 60) == 0) {
 155 +                       ext3_error(sb, __FUNCTION__,
 156 +                                  "Error writing to MMP block");
 157 +                       failed_writes++;
 158 +               }
 159 +
 160 +               if (!(le32_to_cpu(es->s_feature_incompat) &
 161 +                   EXT3_FEATURE_INCOMPAT_MMP)) {
 162 +                       ext3_warning(sb, __FUNCTION__, "kmmpd being stopped "
 163 +                                    "since MMP feature has been disabled.");
 164 +                       EXT3_SB(sb)->s_mmp_tsk = 0;
 165 +                       goto failed;
 166 +               }
 167 +
 168 +               if (sb->s_flags & MS_RDONLY) {
 169 +                       ext3_warning(sb, __FUNCTION__, "kmmpd being stopped "
 170 +                                    "since filesystem has been remounted as "
 171 +                                    "readonly.");
 172 +                       EXT3_SB(sb)->s_mmp_tsk = 0;
 173 +                       goto failed;
 174 +               }
 175 +
 176 +               diff = jiffies - last_update_time;
 177 +               if (diff < mmp_update_interval * HZ) {
 178 +                       set_current_state(TASK_INTERRUPTIBLE);
 179 +                       schedule_timeout(EXT3_MMP_UPDATE_INTERVAL * HZ - diff);
 180 +               }
 181 +
 182 +               /*
 183 +                * We need to make sure that more than mmp_check_interval
 184 +                * seconds have not passed since writing. If that has happened
 185 +                * we need to check if the MMP block is as we left it.
 186 +                */
 187 +               diff = jiffies - last_update_time;
 188 +               if (diff > mmp_check_interval * HZ) {
 189 +                       struct buffer_head *bh_check = NULL;
 190 +                       struct mmp_struct *mmp_check;
 191 +
 192 +                       retval = read_mmp_block(sb, &bh_check, mmp_block);
 193 +                       if (retval) {
 194 +                               EXT3_SB(sb)->s_mmp_tsk = 0;
 195 +                               goto failed;
 196 +                       }
 197 +
 198 +                       mmp_check = (struct mmp_struct *)(bh_check->b_data);
 199 +                       if (mmp->mmp_time != mmp_check->mmp_time ||
 200 +                           memcmp(mmp->mmp_nodename, mmp_check->mmp_nodename,
 201 +                                  sizeof(mmp->mmp_nodename)))
 202 +                               dump_mmp_msg(sb, mmp_check, __FUNCTION__,
 203 +                                            "Error while updating MMP info. "
 204 +                                            "The filesystem seems to have "
 205 +                                            "been multiply mounted.");
 206 +
 207 +                       put_bh(bh_check);
 208 +               }
 209 +
 210 +               /*
 211 +                * Adjust the mmp_check_interval depending on how much time
 212 +                * it took for the MMP block to be written.
 213 +                */
 214 +               mmp_check_interval = max(5 * diff / HZ,
 215 +                                (unsigned long) EXT3_MMP_MIN_CHECK_INTERVAL);
 216 +               mmp->mmp_check_interval = cpu_to_le16(mmp_check_interval);
 217 +       }
 218 +
 219 +       /*
 220 +        * Unmount seems to be clean.
 221 +        */
 222 +       mmp->mmp_seq = cpu_to_le32(EXT3_MMP_SEQ_CLEAN);
 223 +       mmp->mmp_time = cpu_to_le64(get_seconds());
 224 +
 225 +       retval = write_mmp_block(bh);
 226 +
 227 +failed:
 228 +       brelse(bh);
 229 +       return retval;
 230 +}
 231 +
 232 +/*
 233 + * Get a random new sequence number but make sure it is not greater than
 234 + * EXT3_MMP_SEQ_MAX.
 235 + */
 236 +static unsigned int mmp_new_seq(void)
 237 +{
 238 +       u32 new_seq;
 239 +
 240 +       do {
 241 +               get_random_bytes(&new_seq, sizeof(u32));
 242 +       } while (new_seq > EXT3_MMP_SEQ_MAX);
 243 +
 244 +       return new_seq;
 245 +}
 246 +
 247 +/*
 248 + * Protect the filesystem from being mounted more than once.
 249 + */
 250 +static int ext3_multi_mount_protect(struct super_block *sb,
 251 +                                   unsigned long mmp_block)
 252 +{
 253 +       struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 254 +       struct buffer_head *bh = NULL;
 255 +       struct mmp_struct *mmp = NULL;
 256 +       u32 seq;
 257 +       unsigned int mmp_check_interval = le16_to_cpu(es->s_mmp_update_interval);
 258 +       int retval;
 259 +
 260 +       if (mmp_block < le32_to_cpu(es->s_first_data_block) ||
 261 +           mmp_block >= le32_to_cpu(es->s_blocks_count)) {
 262 +               ext3_warning(sb, __FUNCTION__,
 263 +                            "Invalid MMP block in superblock");
 264 +               goto failed;
 265 +       }
 266 +
 267 +       retval = read_mmp_block(sb, &bh, mmp_block);
 268 +       if (retval)
 269 +               goto failed;
 270 +
 271 +       mmp = (struct mmp_struct *)(bh->b_data);
 272 +
 273 +       if (mmp_check_interval < EXT3_MMP_MIN_CHECK_INTERVAL)
 274 +               mmp_check_interval = EXT3_MMP_MIN_CHECK_INTERVAL;
 275 +
 276 +       /*
 277 +        * If check_interval in MMP block is larger, use that instead of
 278 +        * update_interval from the superblock.
 279 +        */
 280 +       if (mmp->mmp_check_interval > mmp_check_interval)
 281 +               mmp_check_interval = mmp->mmp_check_interval;
 282 +
 283 +       seq = le32_to_cpu(mmp->mmp_seq);
 284 +       if (seq == EXT3_MMP_SEQ_CLEAN)
 285 +               goto skip;
 286 +
 287 +       if (seq == EXT3_MMP_SEQ_FSCK) {
 288 +               dump_mmp_msg(sb, mmp, __FUNCTION__,
 289 +                            "fsck is running on the filesystem");
 290 +               goto failed;
 291 +       }
 292 +
 293 +       schedule_timeout_uninterruptible(HZ * (2 * mmp_check_interval + 1));
 294 +
 295 +       retval = read_mmp_block(sb, &bh, mmp_block);
 296 +       if (retval)
 297 +               goto failed;
 298 +       mmp = (struct mmp_struct *)(bh->b_data);
 299 +       if (seq != le32_to_cpu(mmp->mmp_seq)) {
 300 +               dump_mmp_msg(sb, mmp, __FUNCTION__,
 301 +                            "Device is already active on another node.");
 302 +               goto failed;
 303 +       }
 304 +
 305 +skip:
 306 +       /*
 307 +        * write a new random sequence number.
 308 +        */
 309 +       mmp->mmp_seq = seq = cpu_to_le32(mmp_new_seq());
 310 +
 311 +       retval = write_mmp_block(bh);
 312 +       if (retval)
 313 +               goto failed;
 314 +
 315 +       /*
 316 +        * wait for MMP interval and check mmp_seq.
 317 +        */
 318 +       schedule_timeout_uninterruptible(HZ * (2 * mmp_check_interval + 1));
 319 +
 320 +       retval = read_mmp_block(sb, &bh, mmp_block);
 321 +       if (retval)
 322 +               goto failed;
 323 +       mmp = (struct mmp_struct *)(bh->b_data);
 324 +       if (seq != le32_to_cpu(mmp->mmp_seq)) {
 325 +               dump_mmp_msg(sb, mmp, __FUNCTION__,
 326 +                            "Device is already active on another node.");
 327 +               goto failed;
 328 +       }
 329 +
 330 +       /*
 331 +        * Start a kernel thread to update the MMP block periodically.
 332 +        */
 333 +       EXT3_SB(sb)->s_mmp_tsk = kthread_run(kmmpd, sb, "kmmpd-%02x:%02x",
 334 +                                            MAJOR(sb->s_dev),
 335 +                                            MINOR(sb->s_dev));
 336 +       if (IS_ERR(EXT3_SB(sb)->s_mmp_tsk)) {
 337 +               EXT3_SB(sb)->s_mmp_tsk = 0;
 338 +               ext3_warning(sb, __FUNCTION__, "Unable to create kmmpd thread "
 339 +                            "for %s.", sb->s_id);
 340 +               goto failed;
 341 +       }
 342 +
 343 +       brelse(bh);
 344 +       return 0;
 345 +
 346 +failed:
 347 +       brelse(bh);
 348 +       return 1;
 349 +}
 350 +
 351
 352  static int ext3_fill_super (struct super_block *sb, void *data, int silent)
 353  {
 354 @@ -1765,6 +2093,11 @@ static int ext3_fill_super (struct super
 355                           EXT3_HAS_INCOMPAT_FEATURE(sb,
 356                                     EXT3_FEATURE_INCOMPAT_RECOVER));
 357
 358 +       if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_MMP) &&
 359 +           !(sb->s_flags & MS_RDONLY))
 360 +               if (ext3_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
 361 +                       goto failed_mount2;
 362 +
 363         /*
 364          * The first inode we look at is the journal inode.  Don't try
 365          * root first: it may be modified in the journal!
 366 @@ -1890,6 +2223,8 @@ static int ext3_fill_super (struct super
 367  failed_mount3:
 368         journal_destroy(sbi->s_journal);
 369  failed_mount2:
 370 +       if (sbi->s_mmp_tsk)
 371 +               kthread_stop(sbi->s_mmp_tsk);
 372         for (i = 0; i < db_count; i++)
 373                 brelse(sbi->s_group_desc[i]);
 374         kfree(sbi->s_group_desc);
 375 @@ -2415,6 +2750,11 @@ int ext3_remount (struct super_block * s
 376                                 return ret;
 377                         if (!ext3_setup_super (sb, es, 0))
 378                                 sb->s_flags &= ~MS_RDONLY;
 379 +                       if (EXT3_HAS_INCOMPAT_FEATURE(sb,
 380 +                                                   EXT3_FEATURE_INCOMPAT_MMP))
 381 +                               if (ext3_multi_mount_protect(sb,
 382 +                                               le64_to_cpu(es->s_mmp_block)))
 383 +                                       return -EROFS;
 384                 }
 385         }
 386         return 0;
 387 Index: linux-2.6.9-67.0.22/include/linux/ext3_fs.h
 388 ===================================================================
 389 --- linux-2.6.9-67.0.22.orig/include/linux/ext3_fs.h
 390 +++ linux-2.6.9-67.0.22/include/linux/ext3_fs.h
 391 @@ -575,13 +575,17 @@ struct ext3_super_block {
 392         __le32  s_first_meta_bg;        /* First metablock block group */
 393         __le32  s_mkfs_time;            /* When the filesystem was created */
 394         __le32  s_jnl_blocks[17];       /* Backup of the journal inode */
 395 -       __le32  s_blocks_count_hi;      /* Blocks count high 32 bits */
 396 +/*150*/ __le32 s_blocks_count_hi;      /* Blocks count high 32 bits */
 397         __le32  s_r_blocks_count_hi;    /* Reserved blocks count high 32 bits*/
 398         __le32  s_free_blocks_hi;       /* Free blocks count high 32 bits */
 399         __le16  s_min_extra_isize;      /* All inodes have at least # bytes */
 400         __le16  s_want_extra_isize;     /* New inodes should reserve # bytes */
 401 -       __le32  s_flags;                /* Miscellaneous flags */
 402 -       __u32   s_reserved[167];        /* Padding to the end of the block */
 403 +/*160*/ __le32 s_flags;                /* Miscellaneous flags */
 404 +       __le16  s_raid_stride;          /* RAID stride */
 405 +       __le16  s_mmp_update_interval;  /* # seconds to wait in MMP checking */
 406 +       __le64  s_mmp_block;            /* Block for multi-mount protection */
 407 +/*170*/ __le32  s_raid_stripe_width;    /* blocks on all data disks (N*stride)*/
 408 +       __le32  s_reserved[163];        /* Padding to the end of the block */
 409  };
 410
 411  #ifdef __KERNEL__
 412 @@ -665,12 +669,14 @@ static inline struct ext3_inode_info *EX
 413  #define EXT3_FEATURE_INCOMPAT_JOURNAL_DEV      0x0008 /* Journal device */
 414  #define EXT3_FEATURE_INCOMPAT_META_BG          0x0010
 415  #define EXT3_FEATURE_INCOMPAT_EXTENTS          0x0040 /* extents support */
 416 +#define EXT3_FEATURE_INCOMPAT_MMP              0x0100
 417
 418  #define EXT3_FEATURE_COMPAT_SUPP       EXT2_FEATURE_COMPAT_EXT_ATTR
 419  #define EXT3_FEATURE_INCOMPAT_SUPP     (EXT3_FEATURE_INCOMPAT_FILETYPE| \
 420                                          EXT3_FEATURE_INCOMPAT_RECOVER| \
 421                                          EXT3_FEATURE_INCOMPAT_META_BG| \
 422 -                                        EXT3_FEATURE_INCOMPAT_EXTENTS)
 423 +                                        EXT3_FEATURE_INCOMPAT_EXTENTS| \
 424 +                                        EXT3_FEATURE_INCOMPAT_MMP)
 425  #define EXT3_FEATURE_RO_COMPAT_SUPP    (EXT3_FEATURE_RO_COMPAT_SPARSE_SUPER| \
 426                                          EXT3_FEATURE_RO_COMPAT_LARGE_FILE| \
 427                                          EXT4_FEATURE_RO_COMPAT_GDT_CSUM| \
 428 @@ -830,6 +836,39 @@ struct dir_private_info {
 429  #define ERR_BAD_DX_DIR -75000
 430
 431  /*
 432 + * This structure will be used for multiple mount protection. It will be
 433 + * written into the block number saved in the s_mmp_block field in the
 434 + * superblock. Programs that check MMP should assume that if
 435 + * SEQ_FSCK (or any unknown code above SEQ_MAX) is present then it is NOT safe
 436 + * to use the filesystem, regardless of how old the timestamp is.
 437 + */
 438 +#define EXT3_MMP_MAGIC     0x004D4D50U /* ASCII for MMP */
 439 +#define EXT3_MMP_SEQ_CLEAN 0xFF4D4D50U /* mmp_seq value for clean unmount */
 440 +#define EXT3_MMP_SEQ_FSCK  0xE24D4D50U /* mmp_seq value when being fscked */
 441 +#define EXT3_MMP_SEQ_MAX   0xE24D4D4FU /* maximum valid mmp_seq value */
 442 +
 443 +struct mmp_struct {
 444 +       __le32  mmp_magic;
 445 +       __le32  mmp_seq;
 446 +       __le64  mmp_time;
 447 +       char    mmp_nodename[64];
 448 +       char    mmp_bdevname[32];
 449 +       __le16  mmp_check_interval;
 450 +       __le16  mmp_pad1;
 451 +       __le32  mmp_pad2[227];
 452 +};
 453 +
 454 +/*
 455 + * Default interval in seconds to update the MMP sequence number.
 456 + */
 457 +#define EXT3_MMP_UPDATE_INTERVAL   1
 458 +
 459 +/*
 460 + * Minimum interval for MMP checking in seconds.
 461 + */
 462 +#define EXT3_MMP_MIN_CHECK_INTERVAL    5
 463 +
 464 +/*
 465   * Function prototypes
 466   */
 467
 468 Index: linux-2.6.9-67.0.22/include/linux/ext3_fs_sb.h
 469 ===================================================================
 470 --- linux-2.6.9-67.0.22.orig/include/linux/ext3_fs_sb.h
 471 +++ linux-2.6.9-67.0.22/include/linux/ext3_fs_sb.h
 472 @@ -150,6 +150,7 @@ struct ext3_sb_info {
 473         /* locality groups */
 474         struct ext3_locality_group *s_locality_groups;
 475
 476 +       struct task_struct *s_mmp_tsk;  /* Kernel thread for multiple mount protection */
 477  };
 478
 479  #define EXT3_GROUP_INFO(sb, group)                                        \