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 @@ -1430,6 +1434,314 @@ 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 +       brelse(*bh);
  58 +
  59 +       *bh = sb_bread(sb, mmp_block);
  60 +       if (!*bh) {
  61 +               ext3_warning(sb, __FUNCTION__,
  62 +                            "Error while reading MMP block %lu", mmp_block);
  63 +               return -EIO;
  64 +       }
  65 +
  66 +       mmp = (struct mmp_struct *)((*bh)->b_data);
  67 +       if (le32_to_cpu(mmp->mmp_magic) != EXT3_MMP_MAGIC)
  68 +               return -EINVAL;
  69 +
  70 +       return 0;
  71 +}
  72 +
  73 +/*
  74 + * Dump as much information as possible to help the admin.
  75 + */
  76 +static void dump_mmp_msg(struct super_block *sb, struct mmp_struct *mmp,
  77 +                        const char *function, const char *msg)
  78 +{
  79 +       ext3_warning(sb, function, msg);
  80 +       ext3_warning(sb, function, "MMP failure info: last update time: %llu, "
  81 +                    "last update node: %s, last update device: %s\n",
  82 +                    le64_to_cpu(mmp->mmp_time), mmp->mmp_nodename,
  83 +                    mmp->mmp_bdevname);
  84 +}
  85 +
  86 +/*
  87 + * kmmpd will update the MMP sequence every s_mmp_update_interval seconds
  88 + */
  89 +static int kmmpd(void *data)
  90 +{
  91 +       struct super_block *sb = (struct super_block *) data;
  92 +       struct ext3_super_block *es = EXT3_SB(sb)->s_es;
  93 +       struct buffer_head *bh = NULL;
  94 +       struct mmp_struct *mmp;
  95 +       unsigned long mmp_block;
  96 +       u32 seq = 0;
  97 +       unsigned long failed_writes = 0;
  98 +       int mmp_update_interval = le16_to_cpu(es->s_mmp_update_interval);
  99 +       unsigned mmp_check_interval;
 100 +       unsigned long last_update_time;
 101 +       unsigned long diff;
 102 +       int retval;
 103 +
 104 +       mmp_block = le64_to_cpu(es->s_mmp_block);
 105 +       retval = read_mmp_block(sb, &bh, mmp_block);
 106 +       if (retval)
 107 +               goto failed;
 108 +
 109 +       mmp = (struct mmp_struct *)(bh->b_data);
 110 +       mmp->mmp_time = cpu_to_le64(get_seconds());
 111 +       /*
 112 +        * Start with the higher mmp_check_interval and reduce it if
 113 +        * the MMP block is being updated on time.
 114 +        */
 115 +       mmp_check_interval = max(5 * mmp_update_interval,
 116 +                                EXT3_MMP_MIN_CHECK_INTERVAL);
 117 +       mmp->mmp_check_interval = cpu_to_le16(mmp_check_interval);
 118 +       bdevname(bh->b_bdev, mmp->mmp_bdevname);
 119 +
 120 +       down_read(&uts_sem);
 121 +       memcpy(mmp->mmp_nodename, system_utsname.nodename,
 122 +              sizeof(mmp->mmp_nodename));
 123 +       up_read(&uts_sem);
 124 +
 125 +       while (!kthread_should_stop()) {
 126 +               if (++seq > EXT3_MMP_SEQ_MAX)
 127 +                       seq = 1;
 128 +
 129 +               mmp->mmp_seq = cpu_to_le32(seq);
 130 +               mmp->mmp_time = cpu_to_le64(get_seconds());
 131 +               last_update_time = jiffies;
 132 +
 133 +               retval = write_mmp_block(bh);
 134 +               /*
 135 +                * Don't spew too many error messages. Print one every
 136 +                * (s_mmp_update_interval * 60) seconds.
 137 +                */
 138 +               if (retval && (failed_writes % 60) == 0) {
 139 +                       ext3_error(sb, __FUNCTION__,
 140 +                                  "Error writing to MMP block");
 141 +                       failed_writes++;
 142 +               }
 143 +
 144 +               if (!(le32_to_cpu(es->s_feature_incompat) &
 145 +                   EXT3_FEATURE_INCOMPAT_MMP)) {
 146 +                       ext3_warning(sb, __FUNCTION__, "kmmpd being stopped "
 147 +                                    "since MMP feature has been disabled.");
 148 +                       EXT3_SB(sb)->s_mmp_tsk = 0;
 149 +                       goto failed;
 150 +               }
 151 +
 152 +               if (sb->s_flags & MS_RDONLY) {
 153 +                       ext3_warning(sb, __FUNCTION__, "kmmpd being stopped "
 154 +                                    "since filesystem has been remounted as "
 155 +                                    "readonly.");
 156 +                       EXT3_SB(sb)->s_mmp_tsk = 0;
 157 +                       goto failed;
 158 +               }
 159 +
 160 +               diff = jiffies - last_update_time;
 161 +               if (diff < mmp_update_interval * HZ) {
 162 +                       set_current_state(TASK_INTERRUPTIBLE);
 163 +                       schedule_timeout(EXT3_MMP_UPDATE_INTERVAL * HZ - diff);
 164 +               }
 165 +
 166 +               /*
 167 +                * We need to make sure that more than mmp_check_interval
 168 +                * seconds have not passed since writing. If that has happened
 169 +                * we need to check if the MMP block is as we left it.
 170 +                */
 171 +               diff = jiffies - last_update_time;
 172 +               if (diff > mmp_check_interval * HZ) {
 173 +                       struct buffer_head *bh_check = NULL;
 174 +                       struct mmp_struct *mmp_check;
 175 +
 176 +                       retval = read_mmp_block(sb, &bh_check, mmp_block);
 177 +                       if (retval) {
 178 +                               EXT3_SB(sb)->s_mmp_tsk = 0;
 179 +                               goto failed;
 180 +                       }
 181 +
 182 +                       mmp_check = (struct mmp_struct *)(bh_check->b_data);
 183 +                       if (mmp->mmp_time != mmp_check->mmp_time ||
 184 +                           memcmp(mmp->mmp_nodename, mmp_check->mmp_nodename,
 185 +                                  sizeof(mmp->mmp_nodename)))
 186 +                               dump_mmp_msg(sb, mmp_check, __FUNCTION__,
 187 +                                            "Error while updating MMP info. "
 188 +                                            "The filesystem seems to have "
 189 +                                            "been multiply mounted.");
 190 +
 191 +                       put_bh(bh_check);
 192 +               }
 193 +
 194 +               /*
 195 +                * Adjust the mmp_check_interval depending on how much time
 196 +                * it took for the MMP block to be written.
 197 +                */
 198 +               mmp_check_interval = max(5 * diff / HZ,
 199 +                                (unsigned long) EXT3_MMP_MIN_CHECK_INTERVAL);
 200 +               mmp->mmp_check_interval = cpu_to_le16(mmp_check_interval);
 201 +       }
 202 +
 203 +       /*
 204 +        * Unmount seems to be clean.
 205 +        */
 206 +       mmp->mmp_seq = cpu_to_le32(EXT3_MMP_SEQ_CLEAN);
 207 +       mmp->mmp_time = cpu_to_le64(get_seconds());
 208 +
 209 +       retval = write_mmp_block(bh);
 210 +
 211 +failed:
 212 +       brelse(bh);
 213 +       return retval;
 214 +}
 215 +
 216 +/*
 217 + * Get a random new sequence number but make sure it is not greater than
 218 + * EXT3_MMP_SEQ_MAX.
 219 + */
 220 +static unsigned int mmp_new_seq(void)
 221 +{
 222 +       u32 new_seq;
 223 +
 224 +       do {
 225 +               get_random_bytes(&new_seq, sizeof(u32));
 226 +       } while (new_seq > EXT3_MMP_SEQ_MAX);
 227 +
 228 +       return new_seq;
 229 +}
 230 +
 231 +/*
 232 + * Protect the filesystem from being mounted more than once.
 233 + */
 234 +static int ext3_multi_mount_protect(struct super_block *sb,
 235 +                                   unsigned long mmp_block)
 236 +{
 237 +       struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 238 +       struct buffer_head *bh = NULL;
 239 +       struct mmp_struct *mmp = NULL;
 240 +       u32 seq;
 241 +       unsigned int mmp_check_interval = le16_to_cpu(es->s_mmp_update_interval);
 242 +       int retval;
 243 +
 244 +       if (mmp_block < le32_to_cpu(es->s_first_data_block) ||
 245 +           mmp_block >= le32_to_cpu(es->s_blocks_count)) {
 246 +               ext3_warning(sb, __FUNCTION__,
 247 +                            "Invalid MMP block in superblock");
 248 +               goto failed;
 249 +       }
 250 +
 251 +       retval = read_mmp_block(sb, &bh, mmp_block);
 252 +       if (retval)
 253 +               goto failed;
 254 +
 255 +       mmp = (struct mmp_struct *)(bh->b_data);
 256 +
 257 +       if (mmp_check_interval < EXT3_MMP_MIN_CHECK_INTERVAL)
 258 +               mmp_check_interval = EXT3_MMP_MIN_CHECK_INTERVAL;
 259 +
 260 +       /*
 261 +        * If check_interval in MMP block is larger, use that instead of
 262 +        * update_interval from the superblock.
 263 +        */
 264 +       if (mmp->mmp_check_interval > mmp_check_interval)
 265 +               mmp_check_interval = mmp->mmp_check_interval;
 266 +
 267 +       seq = le32_to_cpu(mmp->mmp_seq);
 268 +       if (seq == EXT3_MMP_SEQ_CLEAN)
 269 +               goto skip;
 270 +
 271 +       if (seq == EXT3_MMP_SEQ_FSCK) {
 272 +               dump_mmp_msg(sb, mmp, __FUNCTION__,
 273 +                            "fsck is running on the filesystem");
 274 +               goto failed;
 275 +       }
 276 +
 277 +       schedule_timeout_uninterruptible(HZ * (2 * mmp_check_interval + 1));
 278 +
 279 +       retval = read_mmp_block(sb, &bh, mmp_block);
 280 +       if (retval)
 281 +               goto failed;
 282 +       mmp = (struct mmp_struct *)(bh->b_data);
 283 +       if (seq != le32_to_cpu(mmp->mmp_seq)) {
 284 +               dump_mmp_msg(sb, mmp, __FUNCTION__,
 285 +                            "Device is already active on another node.");
 286 +               goto failed;
 287 +       }
 288 +
 289 +skip:
 290 +       /*
 291 +        * write a new random sequence number.
 292 +        */
 293 +       mmp->mmp_seq = seq = cpu_to_le32(mmp_new_seq());
 294 +
 295 +       retval = write_mmp_block(bh);
 296 +       if (retval)
 297 +               goto failed;
 298 +
 299 +       /*
 300 +        * wait for MMP interval and check mmp_seq.
 301 +        */
 302 +       schedule_timeout_uninterruptible(HZ * (2 * mmp_check_interval + 1));
 303 +
 304 +       retval = read_mmp_block(sb, &bh, mmp_block);
 305 +       if (retval)
 306 +               goto failed;
 307 +       mmp = (struct mmp_struct *)(bh->b_data);
 308 +       if (seq != le32_to_cpu(mmp->mmp_seq)) {
 309 +               dump_mmp_msg(sb, mmp, __FUNCTION__,
 310 +                            "Device is already active on another node.");
 311 +               goto failed;
 312 +       }
 313 +
 314 +       /*
 315 +        * Start a kernel thread to update the MMP block periodically.
 316 +        */
 317 +       EXT3_SB(sb)->s_mmp_tsk = kthread_run(kmmpd, sb, "kmmpd-%02x:%02x",
 318 +                                            MAJOR(sb->s_dev),
 319 +                                            MINOR(sb->s_dev));
 320 +       if (IS_ERR(EXT3_SB(sb)->s_mmp_tsk)) {
 321 +               EXT3_SB(sb)->s_mmp_tsk = 0;
 322 +               ext3_warning(sb, __FUNCTION__, "Unable to create kmmpd thread "
 323 +                            "for %s.", sb->s_id);
 324 +               goto failed;
 325 +       }
 326 +
 327 +       brelse(bh);
 328 +       return 0;
 329 +
 330 +failed:
 331 +       brelse(bh);
 332 +       return 1;
 333 +}
 334 +
 335
 336  static int ext3_fill_super (struct super_block *sb, void *data, int silent)
 337  {
 338 @@ -1754,6 +2066,11 @@ static int ext3_fill_super (struct super
 339                           EXT3_HAS_INCOMPAT_FEATURE(sb,
 340                                     EXT3_FEATURE_INCOMPAT_RECOVER));
 341
 342 +       if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_MMP) &&
 343 +           !(sb->s_flags & MS_RDONLY))
 344 +               if (ext3_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
 345 +                       goto failed_mount3;
 346 +
 347         /*
 348          * The first inode we look at is the journal inode.  Don't try
 349          * root first: it may be modified in the journal!
 350 @@ -2404,6 +2721,11 @@ int ext3_remount (struct super_block * s
 351                                 return ret;
 352                         if (!ext3_setup_super (sb, es, 0))
 353                                 sb->s_flags &= ~MS_RDONLY;
 354 +                       if (EXT3_HAS_INCOMPAT_FEATURE(sb,
 355 +                                                   EXT3_FEATURE_INCOMPAT_MMP))
 356 +                               if (ext3_multi_mount_protect(sb,
 357 +                                               le64_to_cpu(es->s_mmp_block)))
 358 +                                       return -EROFS;
 359                 }
 360         }
 361         return 0;
 362 Index: linux-2.6.9-67.0.22/include/linux/ext3_fs.h
 363 ===================================================================
 364 --- linux-2.6.9-67.0.22.orig/include/linux/ext3_fs.h
 365 +++ linux-2.6.9-67.0.22/include/linux/ext3_fs.h
 366 @@ -575,13 +575,17 @@ struct ext3_super_block {
 367         __le32  s_first_meta_bg;        /* First metablock block group */
 368         __le32  s_mkfs_time;            /* When the filesystem was created */
 369         __le32  s_jnl_blocks[17];       /* Backup of the journal inode */
 370 -       __le32  s_blocks_count_hi;      /* Blocks count high 32 bits */
 371 +/*150*/ __le32 s_blocks_count_hi;      /* Blocks count high 32 bits */
 372         __le32  s_r_blocks_count_hi;    /* Reserved blocks count high 32 bits*/
 373         __le32  s_free_blocks_hi;       /* Free blocks count high 32 bits */
 374         __le16  s_min_extra_isize;      /* All inodes have at least # bytes */
 375         __le16  s_want_extra_isize;     /* New inodes should reserve # bytes */
 376 -       __le32  s_flags;                /* Miscellaneous flags */
 377 -       __u32   s_reserved[167];        /* Padding to the end of the block */
 378 +/*160*/ __le32 s_flags;                /* Miscellaneous flags */
 379 +       __le16  s_raid_stride;          /* RAID stride */
 380 +       __le16  s_mmp_update_interval;  /* # seconds to wait in MMP checking */
 381 +       __le64  s_mmp_block;            /* Block for multi-mount protection */
 382 +/*170*/ __le32  s_raid_stripe_width;    /* blocks on all data disks (N*stride)*/
 383 +       __le32  s_reserved[163];        /* Padding to the end of the block */
 384  };
 385
 386  #ifdef __KERNEL__
 387 @@ -665,12 +669,14 @@ static inline struct ext3_inode_info *EX
 388  #define EXT3_FEATURE_INCOMPAT_JOURNAL_DEV      0x0008 /* Journal device */
 389  #define EXT3_FEATURE_INCOMPAT_META_BG          0x0010
 390  #define EXT3_FEATURE_INCOMPAT_EXTENTS          0x0040 /* extents support */
 391 +#define EXT3_FEATURE_INCOMPAT_MMP              0x0100
 392
 393  #define EXT3_FEATURE_COMPAT_SUPP       EXT2_FEATURE_COMPAT_EXT_ATTR
 394  #define EXT3_FEATURE_INCOMPAT_SUPP     (EXT3_FEATURE_INCOMPAT_FILETYPE| \
 395                                          EXT3_FEATURE_INCOMPAT_RECOVER| \
 396                                          EXT3_FEATURE_INCOMPAT_META_BG| \
 397 -                                        EXT3_FEATURE_INCOMPAT_EXTENTS)
 398 +                                        EXT3_FEATURE_INCOMPAT_EXTENTS| \
 399 +                                        EXT3_FEATURE_INCOMPAT_MMP)
 400  #define EXT3_FEATURE_RO_COMPAT_SUPP    (EXT3_FEATURE_RO_COMPAT_SPARSE_SUPER| \
 401                                          EXT3_FEATURE_RO_COMPAT_LARGE_FILE| \
 402                                          EXT4_FEATURE_RO_COMPAT_GDT_CSUM| \
 403 @@ -830,6 +836,39 @@ struct dir_private_info {
 404  #define ERR_BAD_DX_DIR -75000
 405
 406  /*
 407 + * This structure will be used for multiple mount protection. It will be
 408 + * written into the block number saved in the s_mmp_block field in the
 409 + * superblock. Programs that check MMP should assume that if
 410 + * SEQ_FSCK (or any unknown code above SEQ_MAX) is present then it is NOT safe
 411 + * to use the filesystem, regardless of how old the timestamp is.
 412 + */
 413 +#define EXT3_MMP_MAGIC     0x004D4D50U /* ASCII for MMP */
 414 +#define EXT3_MMP_SEQ_CLEAN 0xFF4D4D50U /* mmp_seq value for clean unmount */
 415 +#define EXT3_MMP_SEQ_FSCK  0xE24D4D50U /* mmp_seq value when being fscked */
 416 +#define EXT3_MMP_SEQ_MAX   0xE24D4D4FU /* maximum valid mmp_seq value */
 417 +
 418 +struct mmp_struct {
 419 +       __le32  mmp_magic;
 420 +       __le32  mmp_seq;
 421 +       __le64  mmp_time;
 422 +       char    mmp_nodename[64];
 423 +       char    mmp_bdevname[32];
 424 +       __le16  mmp_check_interval;
 425 +       __le16  mmp_pad1;
 426 +       __le32  mmp_pad2[227];
 427 +};
 428 +
 429 +/*
 430 + * Default interval in seconds to update the MMP sequence number.
 431 + */
 432 +#define EXT3_MMP_UPDATE_INTERVAL   1
 433 +
 434 +/*
 435 + * Minimum interval for MMP checking in seconds.
 436 + */
 437 +#define EXT3_MMP_MIN_CHECK_INTERVAL    5
 438 +
 439 +/*
 440   * Function prototypes
 441   */
 442
 443 Index: linux-2.6.9-67.0.22/include/linux/ext3_fs_sb.h
 444 ===================================================================
 445 --- linux-2.6.9-67.0.22.orig/include/linux/ext3_fs_sb.h
 446 +++ linux-2.6.9-67.0.22/include/linux/ext3_fs_sb.h
 447 @@ -150,6 +150,7 @@ struct ext3_sb_info {
 448         /* locality groups */
 449         struct ext3_locality_group *s_locality_groups;
 450
 451 +       struct task_struct *s_mmp_tsk;  /* Kernel thread for multiple mount protection */
 452  };
 453
 454  #define EXT3_GROUP_INFO(sb, group)                                        \