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

   1 Index: linux-2.6.9/fs/ext3/super.c
   2 ===================================================================
   3 --- linux-2.6.9.orig/fs/ext3/super.c
   4 +++ linux-2.6.9/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 +442,9 @@ 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 +
  21         remove_proc_entry(sb->s_id, proc_root_ext3);
  22         sbi->s_dev_proc = NULL;
  23         sb->s_fs_info = NULL;
  24 @@ -1439,6 +1444,314 @@ static unsigned long descriptor_loc(stru
  25         return (first_data_block + has_super + (bg * sbi->s_blocks_per_group));
  26  }
  27
  28 +/*
  29 + * Write the MMP block using WRITE_SYNC to try to get the block on-disk
  30 + * faster.
  31 + */
  32 +static int write_mmp_block(struct buffer_head *bh)
  33 +{
  34 +       mark_buffer_dirty(bh);
  35 +       lock_buffer(bh);
  36 +       bh->b_end_io = end_buffer_write_sync;
  37 +       get_bh(bh);
  38 +       submit_bh(WRITE_SYNC, bh);
  39 +       wait_on_buffer(bh);
  40 +       if (unlikely(!buffer_uptodate(bh)))
  41 +               return 1;
  42 +
  43 +       return 0;
  44 +}
  45 +
  46 +/*
  47 + * Read the MMP block. It _must_ be read from disk and hence we clear the
  48 + * uptodate flag on the buffer.
  49 + */
  50 +static int read_mmp_block(struct super_block *sb, struct buffer_head **bh,
  51 +                         unsigned long mmp_block)
  52 +{
  53 +       struct mmp_struct *mmp;
  54 +
  55 +       if (*bh)
  56 +               clear_buffer_uptodate(*bh);
  57 +
  58 +       brelse(*bh);
  59 +
  60 +       *bh = sb_bread(sb, mmp_block);
  61 +       if (!*bh) {
  62 +               ext3_warning(sb, __FUNCTION__,
  63 +                            "Error while reading MMP block %lu", mmp_block);
  64 +               return -EIO;
  65 +       }
  66 +
  67 +       mmp = (struct mmp_struct *)((*bh)->b_data);
  68 +       if (le32_to_cpu(mmp->mmp_magic) != EXT3_MMP_MAGIC)
  69 +               return -EINVAL;
  70 +
  71 +       return 0;
  72 +}
  73 +
  74 +/*
  75 + * Dump as much information as possible to help the admin.
  76 + */
  77 +static void dump_mmp_msg(struct super_block *sb, struct mmp_struct *mmp,
  78 +                        const char *function, const char *msg)
  79 +{
  80 +       ext3_warning(sb, function, msg);
  81 +       ext3_warning(sb, function, "MMP failure info: last update time: %llu, "
  82 +                    "last update node: %s, last update device: %s\n",
  83 +                    le64_to_cpu(mmp->mmp_time), mmp->mmp_nodename,
  84 +                    mmp->mmp_bdevname);
  85 +}
  86 +
  87 +/*
  88 + * kmmpd will update the MMP sequence every s_mmp_update_interval seconds
  89 + */
  90 +static int kmmpd(void *data)
  91 +{
  92 +       struct super_block *sb = (struct super_block *) data;
  93 +       struct ext3_super_block *es = EXT3_SB(sb)->s_es;
  94 +       struct buffer_head *bh = NULL;
  95 +       struct mmp_struct *mmp;
  96 +       unsigned long mmp_block;
  97 +       u32 seq = 0;
  98 +       unsigned long failed_writes = 0;
  99 +       int mmp_update_interval = le16_to_cpu(es->s_mmp_update_interval);
 100 +       unsigned mmp_check_interval;
 101 +       unsigned long last_update_time;
 102 +       unsigned long diff;
 103 +       int retval;
 104 +
 105 +       mmp_block = le64_to_cpu(es->s_mmp_block);
 106 +       retval = read_mmp_block(sb, &bh, mmp_block);
 107 +       if (retval)
 108 +               goto failed;
 109 +
 110 +       mmp = (struct mmp_struct *)(bh->b_data);
 111 +       mmp->mmp_time = cpu_to_le64(get_seconds());
 112 +       /*
 113 +        * Start with the higher mmp_check_interval and reduce it if
 114 +        * the MMP block is being updated on time.
 115 +        */
 116 +       mmp_check_interval = max(5 * mmp_update_interval,
 117 +                                EXT3_MMP_MIN_CHECK_INTERVAL);
 118 +       mmp->mmp_check_interval = cpu_to_le16(mmp_check_interval);
 119 +       bdevname(bh->b_bdev, mmp->mmp_bdevname);
 120 +
 121 +       down_read(&uts_sem);
 122 +       memcpy(mmp->mmp_nodename, system_utsname.nodename,
 123 +              sizeof(mmp->mmp_nodename));
 124 +       up_read(&uts_sem);
 125 +
 126 +       while (!kthread_should_stop()) {
 127 +               if (++seq > EXT3_MMP_SEQ_MAX)
 128 +                       seq = 1;
 129 +
 130 +               mmp->mmp_seq = cpu_to_le32(seq);
 131 +               mmp->mmp_time = cpu_to_le64(get_seconds());
 132 +               last_update_time = jiffies;
 133 +
 134 +               retval = write_mmp_block(bh);
 135 +               /*
 136 +                * Don't spew too many error messages. Print one every
 137 +                * (s_mmp_update_interval * 60) seconds.
 138 +                */
 139 +               if (retval && (failed_writes % 60) == 0) {
 140 +                       ext3_error(sb, __FUNCTION__,
 141 +                                  "Error writing to MMP block");
 142 +                       failed_writes++;
 143 +               }
 144 +
 145 +               if (!(le32_to_cpu(es->s_feature_incompat) &
 146 +                   EXT3_FEATURE_INCOMPAT_MMP)) {
 147 +                       ext3_warning(sb, __FUNCTION__, "kmmpd being stopped "
 148 +                                    "since MMP feature has been disabled.");
 149 +                       EXT3_SB(sb)->s_mmp_tsk = 0;
 150 +                       goto failed;
 151 +               }
 152 +
 153 +               if (sb->s_flags & MS_RDONLY) {
 154 +                       ext3_warning(sb, __FUNCTION__, "kmmpd being stopped "
 155 +                                    "since filesystem has been remounted as "
 156 +                                    "readonly.");
 157 +                       EXT3_SB(sb)->s_mmp_tsk = 0;
 158 +                       goto failed;
 159 +               }
 160 +
 161 +               diff = jiffies - last_update_time;
 162 +               if (diff < mmp_update_interval * HZ) {
 163 +                       set_current_state(TASK_INTERRUPTIBLE);
 164 +                       schedule_timeout(EXT3_MMP_UPDATE_INTERVAL * HZ - diff);
 165 +               }
 166 +
 167 +               /*
 168 +                * We need to make sure that more than mmp_check_interval
 169 +                * seconds have not passed since writing. If that has happened
 170 +                * we need to check if the MMP block is as we left it.
 171 +                */
 172 +               diff = jiffies - last_update_time;
 173 +               if (diff > mmp_check_interval * HZ) {
 174 +                       struct buffer_head *bh_check = NULL;
 175 +                       struct mmp_struct *mmp_check;
 176 +
 177 +                       retval = read_mmp_block(sb, &bh_check, mmp_block);
 178 +                       if (retval) {
 179 +                               EXT3_SB(sb)->s_mmp_tsk = 0;
 180 +                               goto failed;
 181 +                       }
 182 +
 183 +                       mmp_check = (struct mmp_struct *)(bh_check->b_data);
 184 +                       if (mmp->mmp_time != mmp_check->mmp_time ||
 185 +                           memcmp(mmp->mmp_nodename, mmp_check->mmp_nodename,
 186 +                                  sizeof(mmp->mmp_nodename)))
 187 +                               dump_mmp_msg(sb, mmp_check, __FUNCTION__,
 188 +                                            "Error while updating MMP info. "
 189 +                                            "The filesystem seems to have "
 190 +                                            "been multiply mounted.");
 191 +
 192 +                       put_bh(bh_check);
 193 +               }
 194 +
 195 +               /*
 196 +                * Adjust the mmp_check_interval depending on how much time
 197 +                * it took for the MMP block to be written.
 198 +                */
 199 +               mmp_check_interval = max(5 * diff / HZ,
 200 +                                (unsigned long) EXT3_MMP_MIN_CHECK_INTERVAL);
 201 +               mmp->mmp_check_interval = cpu_to_le16(mmp_check_interval);
 202 +       }
 203 +
 204 +       /*
 205 +        * Unmount seems to be clean.
 206 +        */
 207 +       mmp->mmp_seq = cpu_to_le32(EXT3_MMP_SEQ_CLEAN);
 208 +       mmp->mmp_time = cpu_to_le64(get_seconds());
 209 +
 210 +       retval = write_mmp_block(bh);
 211 +
 212 +failed:
 213 +       brelse(bh);
 214 +       return retval;
 215 +}
 216 +
 217 +/*
 218 + * Get a random new sequence number but make sure it is not greater than
 219 + * EXT3_MMP_SEQ_MAX.
 220 + */
 221 +static unsigned int mmp_new_seq(void)
 222 +{
 223 +       u32 new_seq;
 224 +
 225 +       do {
 226 +               get_random_bytes(&new_seq, sizeof(u32));
 227 +       } while (new_seq > EXT3_MMP_SEQ_MAX);
 228 +
 229 +       return new_seq;
 230 +}
 231 +
 232 +/*
 233 + * Protect the filesystem from being mounted more than once.
 234 + */
 235 +static int ext3_multi_mount_protect(struct super_block *sb,
 236 +                                   unsigned long mmp_block)
 237 +{
 238 +       struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 239 +       struct buffer_head *bh = NULL;
 240 +       struct mmp_struct *mmp = NULL;
 241 +       u32 seq;
 242 +       unsigned int mmp_check_interval = le16_to_cpu(es->s_mmp_update_interval);
 243 +       int retval;
 244 +
 245 +       if (mmp_block < le32_to_cpu(es->s_first_data_block) ||
 246 +           mmp_block >= le32_to_cpu(es->s_blocks_count)) {
 247 +               ext3_warning(sb, __FUNCTION__,
 248 +                            "Invalid MMP block in superblock");
 249 +               goto failed;
 250 +       }
 251 +
 252 +       retval = read_mmp_block(sb, &bh, mmp_block);
 253 +       if (retval)
 254 +               goto failed;
 255 +
 256 +       mmp = (struct mmp_struct *)(bh->b_data);
 257 +
 258 +       if (mmp_check_interval < EXT3_MMP_MIN_CHECK_INTERVAL)
 259 +               mmp_check_interval = EXT3_MMP_MIN_CHECK_INTERVAL;
 260 +
 261 +       /*
 262 +        * If check_interval in MMP block is larger, use that instead of
 263 +        * update_interval from the superblock.
 264 +        */
 265 +       if (mmp->mmp_check_interval > mmp_check_interval)
 266 +               mmp_check_interval = mmp->mmp_check_interval;
 267 +
 268 +       seq = le32_to_cpu(mmp->mmp_seq);
 269 +       if (seq == EXT3_MMP_SEQ_CLEAN)
 270 +               goto skip;
 271 +
 272 +       if (seq == EXT3_MMP_SEQ_FSCK) {
 273 +               dump_mmp_msg(sb, mmp, __FUNCTION__,
 274 +                            "fsck is running on the filesystem");
 275 +               goto failed;
 276 +       }
 277 +
 278 +       schedule_timeout_uninterruptible(HZ * (2 * mmp_check_interval + 1));
 279 +
 280 +       retval = read_mmp_block(sb, &bh, mmp_block);
 281 +       if (retval)
 282 +               goto failed;
 283 +       mmp = (struct mmp_struct *)(bh->b_data);
 284 +       if (seq != le32_to_cpu(mmp->mmp_seq)) {
 285 +               dump_mmp_msg(sb, mmp, __FUNCTION__,
 286 +                            "Device is already active on another node.");
 287 +               goto failed;
 288 +       }
 289 +
 290 +skip:
 291 +       /*
 292 +        * write a new random sequence number.
 293 +        */
 294 +       mmp->mmp_seq = seq = cpu_to_le32(mmp_new_seq());
 295 +
 296 +       retval = write_mmp_block(bh);
 297 +       if (retval)
 298 +               goto failed;
 299 +
 300 +       /*
 301 +        * wait for MMP interval and check mmp_seq.
 302 +        */
 303 +       schedule_timeout_uninterruptible(HZ * (2 * mmp_check_interval + 1));
 304 +
 305 +       retval = read_mmp_block(sb, &bh, mmp_block);
 306 +       if (retval)
 307 +               goto failed;
 308 +       mmp = (struct mmp_struct *)(bh->b_data);
 309 +       if (seq != le32_to_cpu(mmp->mmp_seq)) {
 310 +               dump_mmp_msg(sb, mmp, __FUNCTION__,
 311 +                            "Device is already active on another node.");
 312 +               goto failed;
 313 +       }
 314 +
 315 +       /*
 316 +        * Start a kernel thread to update the MMP block periodically.
 317 +        */
 318 +       EXT3_SB(sb)->s_mmp_tsk = kthread_run(kmmpd, sb, "kmmpd-%02x:%02x",
 319 +                                            MAJOR(sb->s_dev),
 320 +                                            MINOR(sb->s_dev));
 321 +       if (IS_ERR(EXT3_SB(sb)->s_mmp_tsk)) {
 322 +               EXT3_SB(sb)->s_mmp_tsk = 0;
 323 +               ext3_warning(sb, __FUNCTION__, "Unable to create kmmpd thread "
 324 +                            "for %s.", sb->s_id);
 325 +               goto failed;
 326 +       }
 327 +
 328 +       brelse(bh);
 329 +       return 0;
 330 +
 331 +failed:
 332 +       brelse(bh);
 333 +       return 1;
 334 +}
 335 +
 336
 337  static int ext3_fill_super (struct super_block *sb, void *data, int silent)
 338  {
 339 @@ -1738,6 +2051,11 @@ static int ext3_fill_super (struct super
 340                           EXT3_HAS_INCOMPAT_FEATURE(sb,
 341                                     EXT3_FEATURE_INCOMPAT_RECOVER));
 342
 343 +       if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_MMP) &&
 344 +           !(sb->s_flags & MS_RDONLY))
 345 +               if (ext3_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
 346 +                       goto failed_mount3;
 347 +
 348         /*
 349          * The first inode we look at is the journal inode.  Don't try
 350          * root first: it may be modified in the journal!
 351 @@ -2371,6 +2689,11 @@ int ext3_remount (struct super_block * s
 352                                 return ret;
 353                         if (!ext3_setup_super (sb, es, 0))
 354                                 sb->s_flags &= ~MS_RDONLY;
 355 +                       if (EXT3_HAS_INCOMPAT_FEATURE(sb,
 356 +                                                   EXT3_FEATURE_INCOMPAT_MMP))
 357 +                               if (ext3_multi_mount_protect(sb,
 358 +                                               le64_to_cpu(es->s_mmp_block)))
 359 +                                       return -EROFS;
 360                 }
 361         }
 362         return 0;
 363 Index: linux-2.6.9/include/linux/ext3_fs.h
 364 ===================================================================
 365 --- linux-2.6.9.orig/include/linux/ext3_fs.h
 366 +++ linux-2.6.9/include/linux/ext3_fs.h
 367 @@ -577,13 +577,17 @@ struct ext3_super_block {
 368         __le32  s_first_meta_bg;        /* First metablock block group */
 369         __le32  s_mkfs_time;            /* When the filesystem was created */
 370         __le32  s_jnl_blocks[17];       /* Backup of the journal inode */
 371 -       __le32  s_blocks_count_hi;      /* Blocks count high 32 bits */
 372 +/*150*/ __le32 s_blocks_count_hi;      /* Blocks count high 32 bits */
 373         __le32  s_r_blocks_count_hi;    /* Reserved blocks count high 32 bits*/
 374         __le32  s_free_blocks_hi;       /* Free blocks count high 32 bits */
 375         __le16  s_min_extra_isize;      /* All inodes have at least # bytes */
 376         __le16  s_want_extra_isize;     /* New inodes should reserve # bytes */
 377 -       __le32  s_flags;                /* Miscellaneous flags */
 378 -       __u32   s_reserved[167];        /* Padding to the end of the block */
 379 +/*160*/ __le32 s_flags;                /* Miscellaneous flags */
 380 +       __le16  s_raid_stride;          /* RAID stride */
 381 +       __le16  s_mmp_update_interval;  /* # seconds to wait in MMP checking */
 382 +       __le64  s_mmp_block;            /* Block for multi-mount protection */
 383 +/*170*/ __le32  s_raid_stripe_width;    /* blocks on all data disks (N*stride)*/
 384 +       __le32  s_reserved[163];        /* Padding to the end of the block */
 385  };
 386
 387  #ifdef __KERNEL__
 388 @@ -667,12 +671,14 @@ static inline struct ext3_inode_info *EX
 389  #define EXT3_FEATURE_INCOMPAT_JOURNAL_DEV      0x0008 /* Journal device */
 390  #define EXT3_FEATURE_INCOMPAT_META_BG          0x0010
 391  #define EXT3_FEATURE_INCOMPAT_EXTENTS          0x0040 /* extents support */
 392 +#define EXT3_FEATURE_INCOMPAT_MMP              0x0100
 393
 394  #define EXT3_FEATURE_COMPAT_SUPP       EXT2_FEATURE_COMPAT_EXT_ATTR
 395  #define EXT3_FEATURE_INCOMPAT_SUPP     (EXT3_FEATURE_INCOMPAT_FILETYPE| \
 396                                          EXT3_FEATURE_INCOMPAT_RECOVER| \
 397                                          EXT3_FEATURE_INCOMPAT_META_BG| \
 398 -                                        EXT3_FEATURE_INCOMPAT_EXTENTS)
 399 +                                        EXT3_FEATURE_INCOMPAT_EXTENTS| \
 400 +                                        EXT3_FEATURE_INCOMPAT_MMP)
 401  #define EXT3_FEATURE_RO_COMPAT_SUPP    (EXT3_FEATURE_RO_COMPAT_SPARSE_SUPER| \
 402                                          EXT3_FEATURE_RO_COMPAT_LARGE_FILE| \
 403                                          EXT4_FEATURE_RO_COMPAT_GDT_CSUM| \
 404 @@ -832,6 +838,39 @@ struct dir_private_info {
 405  #define ERR_BAD_DX_DIR -75000
 406
 407  /*
 408 + * This structure will be used for multiple mount protection. It will be
 409 + * written into the block number saved in the s_mmp_block field in the
 410 + * superblock. Programs that check MMP should assume that if
 411 + * SEQ_FSCK (or any unknown code above SEQ_MAX) is present then it is NOT safe
 412 + * to use the filesystem, regardless of how old the timestamp is.
 413 + */
 414 +#define EXT3_MMP_MAGIC     0x004D4D50U /* ASCII for MMP */
 415 +#define EXT3_MMP_SEQ_CLEAN 0xFF4D4D50U /* mmp_seq value for clean unmount */
 416 +#define EXT3_MMP_SEQ_FSCK  0xE24D4D50U /* mmp_seq value when being fscked */
 417 +#define EXT3_MMP_SEQ_MAX   0xE24D4D4FU /* maximum valid mmp_seq value */
 418 +
 419 +struct mmp_struct {
 420 +       __le32  mmp_magic;
 421 +       __le32  mmp_seq;
 422 +       __le64  mmp_time;
 423 +       char    mmp_nodename[64];
 424 +       char    mmp_bdevname[32];
 425 +       __le16  mmp_check_interval;
 426 +       __le16  mmp_pad1;
 427 +       __le32  mmp_pad2[227];
 428 +};
 429 +
 430 +/*
 431 + * Default interval in seconds to update the MMP sequence number.
 432 + */
 433 +#define EXT3_MMP_UPDATE_INTERVAL   1
 434 +
 435 +/*
 436 + * Minimum interval for MMP checking in seconds.
 437 + */
 438 +#define EXT3_MMP_MIN_CHECK_INTERVAL    5
 439 +
 440 +/*
 441   * Function prototypes
 442   */
 443
 444 Index: linux-2.6.9/include/linux/ext3_fs_sb.h
 445 ===================================================================
 446 --- linux-2.6.9.orig/include/linux/ext3_fs_sb.h
 447 +++ linux-2.6.9/include/linux/ext3_fs_sb.h
 448 @@ -143,6 +143,7 @@ struct ext3_sb_info {
 449         /* locality groups */
 450         struct ext3_locality_group *s_locality_groups;
 451
 452 +       struct task_struct *s_mmp_tsk;  /* Kernel thread for multiple mount protection */
 453  };
 454
 455  #define EXT3_GROUP_INFO(sb, group)                                        \