ldiskfs/kernel_patches/patches/ext4-mmp-sles11.patch

   1 Index: linux-2.6.27.21-0.1/fs/ext4/super.c
   2 ===================================================================
   3 --- linux-2.6.27.21-0.1.orig/fs/ext4/super.c
   4 +++ linux-2.6.27.21-0.1/fs/ext4/super.c
   5 @@ -39,6 +39,8 @@
   6  #include <linux/log2.h>
   7  #include <linux/crc16.h>
   8  #include <asm/uaccess.h>
   9 +#include <linux/kthread.h>
  10 +#include <linux/utsname.h>
  11
  12  #include "ext4.h"
  13  #include "ext4_jbd2.h"
  14 @@ -598,6 +600,8 @@ static void ext4_put_super(struct super_
  15                 invalidate_bdev(sbi->journal_bdev);
  16                 ext4_blkdev_remove(sbi);
  17         }
  18 +       if (sbi->s_mmp_tsk)
  19 +               kthread_stop(sbi->s_mmp_tsk);
  20         sb->s_fs_info = NULL;
  21         kfree(sbi);
  22         return;
  23 @@ -806,7 +810,6 @@ static int ext4_show_options(struct seq_
  24         if (!test_opt(sb, DELALLOC))
  25                 seq_puts(seq, ",nodelalloc");
  26
  27 -
  28         if (sbi->s_stripe)
  29                 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
  30         /*
  31 @@ -829,6 +832,330 @@ static int ext4_show_options(struct seq_
  32  }
  33
  34
  35 +
  36 +/*
  37 + * Write the MMP block using WRITE_SYNC to try to get the block on-disk
  38 + * faster.
  39 + */
  40 +static int write_mmp_block(struct buffer_head *bh)
  41 +{
  42 +       mark_buffer_dirty(bh);
  43 +       lock_buffer(bh);
  44 +       bh->b_end_io = end_buffer_write_sync;
  45 +       get_bh(bh);
  46 +       submit_bh(WRITE_SYNC, bh);
  47 +       wait_on_buffer(bh);
  48 +       if (unlikely(!buffer_uptodate(bh)))
  49 +               return 1;
  50 +
  51 +       return 0;
  52 +}
  53 +
  54 +/*
  55 + * Read the MMP block. It _must_ be read from disk and hence we clear the
  56 + * uptodate flag on the buffer.
  57 + */
  58 +static int read_mmp_block(struct super_block *sb, struct buffer_head **bh,
  59 +                         unsigned long mmp_block)
  60 +{
  61 +       struct mmp_struct *mmp;
  62 +
  63 +       if (*bh)
  64 +               clear_buffer_uptodate(*bh);
  65 +
  66 +#if 0
  67 +       brelse(*bh);
  68 +
  69 +       *bh = sb_bread(sb, mmp_block);
  70 +#else
  71 +       if (!*bh)
  72 +               *bh = sb_getblk(sb, mmp_block);
  73 +       if (*bh) {
  74 +               get_bh(*bh);
  75 +               lock_buffer(*bh);
  76 +               (*bh)->b_end_io = end_buffer_read_sync;
  77 +               submit_bh(READ_SYNC, *bh);
  78 +               wait_on_buffer(*bh);
  79 +               if (!buffer_uptodate(*bh)) {
  80 +                       brelse(*bh);
  81 +                       *bh = NULL;
  82 +               }
  83 +       }
  84 +#endif
  85 +       if (!*bh) {
  86 +               ext4_warning(sb, __FUNCTION__,
  87 +                            "Error while reading MMP block %lu", mmp_block);
  88 +               return -EIO;
  89 +       }
  90 +
  91 +       mmp = (struct mmp_struct *)((*bh)->b_data);
  92 +       if (le32_to_cpu(mmp->mmp_magic) != EXT4_MMP_MAGIC)
  93 +               return -EINVAL;
  94 +
  95 +       return 0;
  96 +}
  97 +
  98 +/*
  99 + * Dump as much information as possible to help the admin.
 100 + */
 101 +static void dump_mmp_msg(struct super_block *sb, struct mmp_struct *mmp,
 102 +                        const char *function, const char *msg)
 103 +{
 104 +       ext4_warning(sb, function, msg);
 105 +       ext4_warning(sb, function, "MMP failure info: last update time: %llu, "
 106 +                    "last update node: %s, last update device: %s\n",
 107 +                    le64_to_cpu(mmp->mmp_time), mmp->mmp_nodename,
 108 +                    mmp->mmp_bdevname);
 109 +}
 110 +
 111 +/*
 112 + * kmmpd will update the MMP sequence every s_mmp_update_interval seconds
 113 + */
 114 +static int kmmpd(void *data)
 115 +{
 116 +       struct super_block *sb = (struct super_block *) data;
 117 +       struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 118 +       struct buffer_head *bh = NULL;
 119 +       struct mmp_struct *mmp;
 120 +       unsigned long mmp_block;
 121 +       u32 seq = 0;
 122 +       unsigned long failed_writes = 0;
 123 +       int mmp_update_interval = le16_to_cpu(es->s_mmp_update_interval);
 124 +       unsigned mmp_check_interval;
 125 +       unsigned long last_update_time;
 126 +       unsigned long diff;
 127 +       int retval;
 128 +
 129 +       mmp_block = le64_to_cpu(es->s_mmp_block);
 130 +       retval = read_mmp_block(sb, &bh, mmp_block);
 131 +       if (retval)
 132 +               goto failed;
 133 +
 134 +       mmp = (struct mmp_struct *)(bh->b_data);
 135 +       mmp->mmp_time = cpu_to_le64(get_seconds());
 136 +       /*
 137 +        * Start with the higher mmp_check_interval and reduce it if
 138 +        * the MMP block is being updated on time.
 139 +        */
 140 +       mmp_check_interval = max(5 * mmp_update_interval,
 141 +                                EXT4_MMP_MIN_CHECK_INTERVAL);
 142 +       mmp->mmp_check_interval = cpu_to_le16(mmp_check_interval);
 143 +       bdevname(bh->b_bdev, mmp->mmp_bdevname);
 144 +
 145 +       down_read(&uts_sem);
 146 +       memcpy(mmp->mmp_nodename, init_utsname()->sysname,
 147 +              sizeof(mmp->mmp_nodename));
 148 +       up_read(&uts_sem);
 149 +
 150 +       while (!kthread_should_stop()) {
 151 +               if (++seq > EXT4_MMP_SEQ_MAX)
 152 +                       seq = 1;
 153 +
 154 +               mmp->mmp_seq = cpu_to_le32(seq);
 155 +               mmp->mmp_time = cpu_to_le64(get_seconds());
 156 +               last_update_time = jiffies;
 157 +
 158 +               retval = write_mmp_block(bh);
 159 +               /*
 160 +                * Don't spew too many error messages. Print one every
 161 +                * (s_mmp_update_interval * 60) seconds.
 162 +                */
 163 +               if (retval && (failed_writes % 60) == 0) {
 164 +                       ext4_error(sb, __FUNCTION__,
 165 +                                  "Error writing to MMP block");
 166 +                       failed_writes++;
 167 +               }
 168 +
 169 +               if (!(le32_to_cpu(es->s_feature_incompat) &
 170 +                   EXT4_FEATURE_INCOMPAT_MMP)) {
 171 +                       ext4_warning(sb, __FUNCTION__, "kmmpd being stopped "
 172 +                                    "since MMP feature has been disabled.");
 173 +                       EXT4_SB(sb)->s_mmp_tsk = 0;
 174 +                       goto failed;
 175 +               }
 176 +
 177 +               if (sb->s_flags & MS_RDONLY) {
 178 +                       ext4_warning(sb, __FUNCTION__, "kmmpd being stopped "
 179 +                                    "since filesystem has been remounted as "
 180 +                                    "readonly.");
 181 +                       EXT4_SB(sb)->s_mmp_tsk = 0;
 182 +                       goto failed;
 183 +               }
 184 +
 185 +               diff = jiffies - last_update_time;
 186 +               if (diff < mmp_update_interval * HZ)
 187 +                       schedule_timeout_interruptible(EXT4_MMP_UPDATE_INTERVAL*
 188 +                                                      HZ - diff);
 189 +
 190 +               /*
 191 +                * We need to make sure that more than mmp_check_interval
 192 +                * seconds have not passed since writing. If that has happened
 193 +                * we need to check if the MMP block is as we left it.
 194 +                */
 195 +               diff = jiffies - last_update_time;
 196 +               if (diff > mmp_check_interval * HZ) {
 197 +                       struct buffer_head *bh_check = NULL;
 198 +                       struct mmp_struct *mmp_check;
 199 +
 200 +                       retval = read_mmp_block(sb, &bh_check, mmp_block);
 201 +                       if (retval) {
 202 +                               EXT4_SB(sb)->s_mmp_tsk = 0;
 203 +                               goto failed;
 204 +                       }
 205 +
 206 +                       mmp_check = (struct mmp_struct *)(bh_check->b_data);
 207 +                       if (mmp->mmp_time != mmp_check->mmp_time ||
 208 +                           memcmp(mmp->mmp_nodename, mmp_check->mmp_nodename,
 209 +                                  sizeof(mmp->mmp_nodename)))
 210 +                               dump_mmp_msg(sb, mmp_check, __FUNCTION__,
 211 +                                            "Error while updating MMP info. "
 212 +                                            "The filesystem seems to have "
 213 +                                            "been multiply mounted.");
 214 +
 215 +                       put_bh(bh_check);
 216 +               }
 217 +
 218 +               /*
 219 +                * Adjust the mmp_check_interval depending on how much time
 220 +                * it took for the MMP block to be written.
 221 +                */
 222 +               mmp_check_interval = max(5 * diff / HZ,
 223 +                                (unsigned long) EXT4_MMP_MIN_CHECK_INTERVAL);
 224 +               mmp->mmp_check_interval = cpu_to_le16(mmp_check_interval);
 225 +       }
 226 +
 227 +       /*
 228 +        * Unmount seems to be clean.
 229 +        */
 230 +       mmp->mmp_seq = cpu_to_le32(EXT4_MMP_SEQ_CLEAN);
 231 +       mmp->mmp_time = cpu_to_le64(get_seconds());
 232 +
 233 +       retval = write_mmp_block(bh);
 234 +
 235 +failed:
 236 +       brelse(bh);
 237 +       return retval;
 238 +}
 239 +
 240 +/*
 241 + * Get a random new sequence number but make sure it is not greater than
 242 + * EXT4_MMP_SEQ_MAX.
 243 + */
 244 +static unsigned int mmp_new_seq(void)
 245 +{
 246 +       u32 new_seq;
 247 +
 248 +       do {
 249 +               get_random_bytes(&new_seq, sizeof(u32));
 250 +       } while (new_seq > EXT4_MMP_SEQ_MAX);
 251 +
 252 +       return new_seq;
 253 +}
 254 +
 255 +/*
 256 + * Protect the filesystem from being mounted more than once.
 257 + */
 258 +static int ext4_multi_mount_protect(struct super_block *sb,
 259 +                                   unsigned long mmp_block)
 260 +{
 261 +       struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 262 +       struct buffer_head *bh = NULL;
 263 +       struct mmp_struct *mmp = NULL;
 264 +       u32 seq;
 265 +       unsigned int mmp_check_interval = le16_to_cpu(es->s_mmp_update_interval);
 266 +       int retval;
 267 +
 268 +       if (mmp_block < le32_to_cpu(es->s_first_data_block) ||
 269 +           mmp_block >= ext4_blocks_count(es)) {
 270 +               ext4_warning(sb, __FUNCTION__,
 271 +                            "Invalid MMP block in superblock");
 272 +               goto failed;
 273 +       }
 274 +
 275 +       retval = read_mmp_block(sb, &bh, mmp_block);
 276 +       if (retval)
 277 +               goto failed;
 278 +
 279 +       mmp = (struct mmp_struct *)(bh->b_data);
 280 +
 281 +       if (mmp_check_interval < EXT4_MMP_MIN_CHECK_INTERVAL)
 282 +               mmp_check_interval = EXT4_MMP_MIN_CHECK_INTERVAL;
 283 +
 284 +       /*
 285 +        * If check_interval in MMP block is larger, use that instead of
 286 +        * update_interval from the superblock.
 287 +        */
 288 +       if (mmp->mmp_check_interval > mmp_check_interval)
 289 +               mmp_check_interval = mmp->mmp_check_interval;
 290 +
 291 +       seq = le32_to_cpu(mmp->mmp_seq);
 292 +       if (seq == EXT4_MMP_SEQ_CLEAN)
 293 +               goto skip;
 294 +
 295 +       if (seq == EXT4_MMP_SEQ_FSCK) {
 296 +               dump_mmp_msg(sb, mmp, __FUNCTION__,
 297 +                            "fsck is running on the filesystem");
 298 +               goto failed;
 299 +       }
 300 +
 301 +       schedule_timeout_uninterruptible(HZ * (2 * mmp_check_interval + 1));
 302 +
 303 +       retval = read_mmp_block(sb, &bh, mmp_block);
 304 +       if (retval)
 305 +               goto failed;
 306 +       mmp = (struct mmp_struct *)(bh->b_data);
 307 +       if (seq != le32_to_cpu(mmp->mmp_seq)) {
 308 +               dump_mmp_msg(sb, mmp, __FUNCTION__,
 309 +                            "Device is already active on another node.");
 310 +               goto failed;
 311 +       }
 312 +
 313 +skip:
 314 +       /*
 315 +        * write a new random sequence number.
 316 +        */
 317 +       mmp->mmp_seq = seq = cpu_to_le32(mmp_new_seq());
 318 +
 319 +       retval = write_mmp_block(bh);
 320 +       if (retval)
 321 +               goto failed;
 322 +
 323 +       /*
 324 +        * wait for MMP interval and check mmp_seq.
 325 +        */
 326 +       schedule_timeout_uninterruptible(HZ * (2 * mmp_check_interval + 1));
 327 +
 328 +       retval = read_mmp_block(sb, &bh, mmp_block);
 329 +       if (retval)
 330 +               goto failed;
 331 +       mmp = (struct mmp_struct *)(bh->b_data);
 332 +       if (seq != le32_to_cpu(mmp->mmp_seq)) {
 333 +               dump_mmp_msg(sb, mmp, __FUNCTION__,
 334 +                            "Device is already active on another node.");
 335 +               goto failed;
 336 +       }
 337 +
 338 +       /*
 339 +        * Start a kernel thread to update the MMP block periodically.
 340 +        */
 341 +       EXT4_SB(sb)->s_mmp_tsk = kthread_run(kmmpd, sb, "kmmpd-%02x:%02x",
 342 +                                            MAJOR(sb->s_dev),
 343 +                                            MINOR(sb->s_dev));
 344 +       if (IS_ERR(EXT4_SB(sb)->s_mmp_tsk)) {
 345 +               EXT4_SB(sb)->s_mmp_tsk = 0;
 346 +               ext4_warning(sb, __FUNCTION__, "Unable to create kmmpd thread "
 347 +                            "for %s.", sb->s_id);
 348 +               goto failed;
 349 +       }
 350 +
 351 +       brelse(bh);
 352 +       return 0;
 353 +
 354 +failed:
 355 +       brelse(bh);
 356 +       return 1;
 357 +}
 358 +
 359  static struct inode *ext4_nfs_get_inode(struct super_block *sb,
 360                 u64 ino, u32 generation)
 361  {
 362 @@ -2366,6 +2693,11 @@ static int ext4_fill_super(struct super_
 363                           EXT4_HAS_INCOMPAT_FEATURE(sb,
 364                                     EXT4_FEATURE_INCOMPAT_RECOVER));
 365
 366 +       if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
 367 +           !(sb->s_flags & MS_RDONLY))
 368 +               if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
 369 +                       goto failed_mount3;
 370 +
 371         /*
 372          * The first inode we look at is the journal inode.  Don't try
 373          * root first: it may be modified in the journal!
 374 @@ -2566,6 +2898,8 @@ failed_mount3:
 375         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 376         percpu_counter_destroy(&sbi->s_dirs_counter);
 377         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
 378 +       if (sbi->s_mmp_tsk)
 379 +               kthread_stop(sbi->s_mmp_tsk);
 380  failed_mount2:
 381         for (i = 0; i < db_count; i++)
 382                 brelse(sbi->s_group_desc[i]);
 383 @@ -3080,7 +3414,7 @@ static int ext4_remount(struct super_blo
 384         unsigned long old_sb_flags;
 385         struct ext4_mount_options old_opts;
 386         ext4_group_t g;
 387 -       int err;
 388 +       int err = 0;
 389  #ifdef CONFIG_QUOTA
 390         int i;
 391  #endif
 392 @@ -3205,6 +3539,13 @@ static int ext4_remount(struct super_blo
 393                                 goto restore_opts;
 394                         if (!ext4_setup_super(sb, es, 0))
 395                                 sb->s_flags &= ~MS_RDONLY;
 396 +                       if (EXT4_HAS_INCOMPAT_FEATURE(sb,
 397 +                                                   EXT4_FEATURE_INCOMPAT_MMP))
 398 +                               if (ext4_multi_mount_protect(sb,
 399 +                                               le64_to_cpu(es->s_mmp_block))) {
 400 +                                       err = -EROFS;
 401 +                                       goto restore_opts;
 402 +                               }
 403                 }
 404         }
 405  #ifdef CONFIG_QUOTA
 406 Index: linux-2.6.27.21-0.1/fs/ext4/ext4.h
 407 ===================================================================
 408 --- linux-2.6.27.21-0.1.orig/fs/ext4/ext4.h
 409 +++ linux-2.6.27.21-0.1/fs/ext4/ext4.h
 410 @@ -660,7 +660,7 @@ struct ext4_super_block {
 411         __le16  s_want_extra_isize;     /* New inodes should reserve # bytes */
 412         __le32  s_flags;                /* Miscellaneous flags */
 413         __le16  s_raid_stride;          /* RAID stride */
 414 -       __le16  s_mmp_interval;         /* # seconds to wait in MMP checking */
 415 +       __le16  s_mmp_update_interval;  /* # seconds to wait in MMP checking */
 416         __le64  s_mmp_block;            /* Block for multi-mount protection */
 417         __le32  s_raid_stripe_width;    /* blocks on all data disks (N*stride)*/
 418         __u8    s_log_groups_per_flex;  /* FLEX_BG group size */
 419 @@ -777,7 +777,8 @@ static inline int ext4_valid_inum(struct
 420                                          EXT4_FEATURE_INCOMPAT_META_BG| \
 421                                          EXT4_FEATURE_INCOMPAT_EXTENTS| \
 422                                          EXT4_FEATURE_INCOMPAT_64BIT| \
 423 -                                        EXT4_FEATURE_INCOMPAT_FLEX_BG)
 424 +                                        EXT4_FEATURE_INCOMPAT_FLEX_BG| \
 425 +                                        EXT4_FEATURE_INCOMPAT_MMP)
 426  #define EXT4_FEATURE_RO_COMPAT_SUPP    (EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER| \
 427                                          EXT4_FEATURE_RO_COMPAT_LARGE_FILE| \
 428                                          EXT4_FEATURE_RO_COMPAT_GDT_CSUM| \
 429 @@ -981,6 +982,39 @@ do {                                                                       \
 430  #endif
 431
 432  /*
 433 + * This structure will be used for multiple mount protection. It will be
 434 + * written into the block number saved in the s_mmp_block field in the
 435 + * superblock. Programs that check MMP should assume that if
 436 + * SEQ_FSCK (or any unknown code above SEQ_MAX) is present then it is NOT safe
 437 + * to use the filesystem, regardless of how old the timestamp is.
 438 + */
 439 +#define EXT4_MMP_MAGIC     0x004D4D50U /* ASCII for MMP */
 440 +#define EXT4_MMP_SEQ_CLEAN 0xFF4D4D50U /* mmp_seq value for clean unmount */
 441 +#define EXT4_MMP_SEQ_FSCK  0xE24D4D50U /* mmp_seq value when being fscked */
 442 +#define EXT4_MMP_SEQ_MAX   0xE24D4D4FU /* maximum valid mmp_seq value */
 443 +
 444 +struct mmp_struct {
 445 +       __le32  mmp_magic;
 446 +       __le32  mmp_seq;
 447 +       __le64  mmp_time;
 448 +       char    mmp_nodename[64];
 449 +       char    mmp_bdevname[32];
 450 +       __le16  mmp_check_interval;
 451 +       __le16  mmp_pad1;
 452 +       __le32  mmp_pad2[227];
 453 +};
 454 +
 455 +/*
 456 + * Default interval in seconds to update the MMP sequence number.
 457 + */
 458 +#define EXT4_MMP_UPDATE_INTERVAL   1
 459 +
 460 +/*
 461 + * Minimum interval for MMP checking in seconds.
 462 + */
 463 +#define EXT4_MMP_MIN_CHECK_INTERVAL    5
 464 +
 465 +/*
 466   * Function prototypes
 467   */
 468
 469 Index: linux-2.6.27.21-0.1/fs/ext4/ext4_sb.h
 470 ===================================================================
 471 --- linux-2.6.27.21-0.1.orig/fs/ext4/ext4_sb.h
 472 +++ linux-2.6.27.21-0.1/fs/ext4/ext4_sb.h
 473 @@ -150,6 +150,8 @@ struct ext4_sb_info {
 474
 475         unsigned int s_log_groups_per_flex;
 476         struct flex_groups *s_flex_groups;
 477 +
 478 +       struct task_struct *s_mmp_tsk;  /* Kernel thread for multiple mount protection */
 479  };
 480
 481  #endif /* _EXT4_SB */