ldiskfs/kernel_patches/patches/ext4-mmp-rhel6.patch

   1 Index: linux-stage/fs/ext4/super.c
   2 ===================================================================
   3 --- linux-stage.orig/fs/ext4/super.c    2011-03-03 15:25:02.376539424 +0800
   4 +++ linux-stage/fs/ext4/super.c 2011-03-05 12:24:02.918774335 +0800
   5 @@ -40,6 +40,8 @@
   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 @@ -700,6 +702,8 @@
  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         /*
  22          * Now that we are completely done shutting down the
  23 @@ -970,6 +974,344 @@
  24         return 0;
  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 +               ext4_warning(sb,
  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) != EXT4_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 +       __ext4_warning(sb, function, msg);
  96 +       __ext4_warning(sb, function, "MMP failure info: last update time: %llu, "
  97 +                    "last update node: %s, last update device: %s\n",
  98 +                    (long long unsigned int)le64_to_cpu(mmp->mmp_time),
  99 +                    mmp->mmp_nodename, 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 ext4_super_block *es = EXT4_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 +                                EXT4_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 +       memcpy(mmp->mmp_nodename, init_utsname()->sysname,
 137 +              sizeof(mmp->mmp_nodename));
 138 +
 139 +       while (!kthread_should_stop()) {
 140 +               if (++seq > EXT4_MMP_SEQ_MAX)
 141 +                       seq = 1;
 142 +
 143 +               mmp->mmp_seq = cpu_to_le32(seq);
 144 +               mmp->mmp_time = cpu_to_le64(get_seconds());
 145 +               last_update_time = jiffies;
 146 +
 147 +               retval = write_mmp_block(bh);
 148 +               /*
 149 +                * Don't spew too many error messages. Print one every
 150 +                * (s_mmp_update_interval * 60) seconds.
 151 +                */
 152 +               if (retval && (failed_writes % 60) == 0) {
 153 +                       ext4_error(sb,
 154 +                                  "Error writing to MMP block");
 155 +                       failed_writes++;
 156 +               }
 157 +
 158 +               if (!(le32_to_cpu(es->s_feature_incompat) &
 159 +                   EXT4_FEATURE_INCOMPAT_MMP)) {
 160 +                       ext4_warning(sb, "kmmpd being stopped "
 161 +                                    "since MMP feature has been disabled.");
 162 +                       EXT4_SB(sb)->s_mmp_tsk = 0;
 163 +                       goto failed;
 164 +               }
 165 +
 166 +               if (sb->s_flags & MS_RDONLY) {
 167 +                       ext4_warning(sb, "kmmpd being stopped "
 168 +                                    "since filesystem has been remounted as "
 169 +                                    "readonly.");
 170 +                       EXT4_SB(sb)->s_mmp_tsk = 0;
 171 +                       goto failed;
 172 +               }
 173 +
 174 +               diff = jiffies - last_update_time;
 175 +               if (diff < mmp_update_interval * HZ)
 176 +                       schedule_timeout_interruptible(mmp_update_interval *
 177 +                                                      HZ - diff);
 178 +
 179 +               /*
 180 +                * We need to make sure that more than mmp_check_interval
 181 +                * seconds have not passed since writing. If that has happened
 182 +                * we need to check if the MMP block is as we left it.
 183 +                */
 184 +               diff = jiffies - last_update_time;
 185 +               if (diff > mmp_check_interval * HZ) {
 186 +                       struct buffer_head *bh_check = NULL;
 187 +                       struct mmp_struct *mmp_check;
 188 +
 189 +                       retval = read_mmp_block(sb, &bh_check, mmp_block);
 190 +                       if (retval) {
 191 +                               EXT4_SB(sb)->s_mmp_tsk = 0;
 192 +                               goto failed;
 193 +                       }
 194 +
 195 +                       mmp_check = (struct mmp_struct *)(bh_check->b_data);
 196 +                       if (mmp->mmp_time != mmp_check->mmp_time ||
 197 +                           memcmp(mmp->mmp_nodename, mmp_check->mmp_nodename,
 198 +                                  sizeof(mmp->mmp_nodename)))
 199 +                               dump_mmp_msg(sb, mmp_check, __func__,
 200 +                                            "Error while updating MMP info. "
 201 +                                            "The filesystem seems to have "
 202 +                                            "been multiply mounted.");
 203 +
 204 +                       put_bh(bh_check);
 205 +               }
 206 +
 207 +               /*
 208 +                * Adjust the mmp_check_interval depending on how much time
 209 +                * it took for the MMP block to be written.
 210 +                */
 211 +               mmp_check_interval = max(5 * diff / HZ,
 212 +                                (unsigned long) EXT4_MMP_MIN_CHECK_INTERVAL);
 213 +               mmp->mmp_check_interval = cpu_to_le16(mmp_check_interval);
 214 +       }
 215 +
 216 +       /*
 217 +        * Unmount seems to be clean.
 218 +        */
 219 +       mmp->mmp_seq = cpu_to_le32(EXT4_MMP_SEQ_CLEAN);
 220 +       mmp->mmp_time = cpu_to_le64(get_seconds());
 221 +
 222 +       retval = write_mmp_block(bh);
 223 +
 224 +failed:
 225 +       brelse(bh);
 226 +       return retval;
 227 +}
 228 +
 229 +/*
 230 + * Get a random new sequence number but make sure it is not greater than
 231 + * EXT4_MMP_SEQ_MAX.
 232 + */
 233 +static unsigned int mmp_new_seq(void)
 234 +{
 235 +       u32 new_seq;
 236 +
 237 +       do {
 238 +               get_random_bytes(&new_seq, sizeof(u32));
 239 +       } while (new_seq > EXT4_MMP_SEQ_MAX);
 240 +
 241 +       return new_seq;
 242 +}
 243 +
 244 +/*
 245 + * Protect the filesystem from being mounted more than once.
 246 + */
 247 +static int ext4_multi_mount_protect(struct super_block *sb,
 248 +                                   unsigned long mmp_block)
 249 +{
 250 +       struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 251 +       struct buffer_head *bh = NULL;
 252 +       struct mmp_struct *mmp = NULL;
 253 +       u32 seq;
 254 +       unsigned int mmp_check_interval = le16_to_cpu(es->s_mmp_update_interval);
 255 +       unsigned int wait_time = 0;
 256 +       int retval;
 257 +
 258 +       if (mmp_block < le32_to_cpu(es->s_first_data_block) ||
 259 +           mmp_block >= ext4_blocks_count(es)) {
 260 +               ext4_warning(sb,
 261 +                            "Invalid MMP block in superblock");
 262 +               goto failed;
 263 +       }
 264 +
 265 +       retval = read_mmp_block(sb, &bh, mmp_block);
 266 +       if (retval)
 267 +               goto failed;
 268 +
 269 +       mmp = (struct mmp_struct *)(bh->b_data);
 270 +
 271 +       if (mmp_check_interval < EXT4_MMP_MIN_CHECK_INTERVAL)
 272 +               mmp_check_interval = EXT4_MMP_MIN_CHECK_INTERVAL;
 273 +
 274 +       /*
 275 +        * If check_interval in MMP block is larger, use that instead of
 276 +        * update_interval from the superblock.
 277 +        */
 278 +       if (mmp->mmp_check_interval > mmp_check_interval)
 279 +               mmp_check_interval = mmp->mmp_check_interval;
 280 +
 281 +       seq = le32_to_cpu(mmp->mmp_seq);
 282 +       if (seq == EXT4_MMP_SEQ_CLEAN)
 283 +               goto skip;
 284 +
 285 +       if (seq == EXT4_MMP_SEQ_FSCK) {
 286 +               dump_mmp_msg(sb, mmp, __func__,
 287 +                            "fsck is running on the filesystem");
 288 +               goto failed;
 289 +       }
 290 +
 291 +       wait_time = min(mmp_check_interval * 2 + 1,
 292 +               mmp_check_interval + 60);
 293 +
 294 +       /* Print MMP interval if more than 20 secs. */
 295 +       if (wait_time > EXT4_MMP_MIN_CHECK_INTERVAL * 4)
 296 +               ext4_warning(sb, "MMP interval %u higher than "
 297 +                            "expected, please wait.\n", wait_time * 2);
 298 +
 299 +       if (schedule_timeout_interruptible(HZ * wait_time) != 0) {
 300 +               ext4_warning(sb, "MMP startup interrupted, failing "
 301 +                            "mount\n");
 302 +               goto failed;
 303 +       }
 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, __func__,
 311 +                            "Device is already active on another node.");
 312 +               goto failed;
 313 +       }
 314 +
 315 +skip:
 316 +       /*
 317 +        * write a new random sequence number.
 318 +        */
 319 +       mmp->mmp_seq = seq = cpu_to_le32(mmp_new_seq());
 320 +
 321 +       retval = write_mmp_block(bh);
 322 +       if (retval)
 323 +               goto failed;
 324 +
 325 +       /*
 326 +        * wait for MMP interval and check mmp_seq.
 327 +        */
 328 +       if (schedule_timeout_interruptible(HZ * wait_time) != 0) {
 329 +               ext4_warning(sb, "MMP startup interrupted, failing "
 330 +                            "mount\n");
 331 +               goto failed;
 332 +       }
 333 +
 334 +       retval = read_mmp_block(sb, &bh, mmp_block);
 335 +       if (retval)
 336 +               goto failed;
 337 +       mmp = (struct mmp_struct *)(bh->b_data);
 338 +       if (seq != le32_to_cpu(mmp->mmp_seq)) {
 339 +               dump_mmp_msg(sb, mmp, __func__,
 340 +                            "Device is already active on another node.");
 341 +               goto failed;
 342 +       }
 343 +
 344 +       /*
 345 +        * Start a kernel thread to update the MMP block periodically.
 346 +        */
 347 +       EXT4_SB(sb)->s_mmp_tsk = kthread_run(kmmpd, sb, "kmmpd-%02x:%02x",
 348 +                                            MAJOR(sb->s_dev),
 349 +                                            MINOR(sb->s_dev));
 350 +       if (IS_ERR(EXT4_SB(sb)->s_mmp_tsk)) {
 351 +               EXT4_SB(sb)->s_mmp_tsk = 0;
 352 +               ext4_warning(sb, "Unable to create kmmpd thread "
 353 +                            "for %s.", sb->s_id);
 354 +               goto failed;
 355 +       }
 356 +
 357 +       brelse(bh);
 358 +       return 0;
 359 +
 360 +failed:
 361 +       brelse(bh);
 362 +       return 1;
 363 +}
 364 +
 365  static struct inode *ext4_nfs_get_inode(struct super_block *sb,
 366                                         u64 ino, u32 generation)
 367  {
 368 @@ -2816,6 +3158,11 @@
 369                           EXT4_HAS_INCOMPAT_FEATURE(sb,
 370                                     EXT4_FEATURE_INCOMPAT_RECOVER));
 371
 372 +       if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
 373 +           !(sb->s_flags & MS_RDONLY))
 374 +               if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
 375 +                       goto failed_mount3;
 376 +
 377         /*
 378          * The first inode we look at is the journal inode.  Don't try
 379          * root first: it may be modified in the journal!
 380 @@ -3052,6 +3399,8 @@
 381         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 382         percpu_counter_destroy(&sbi->s_dirs_counter);
 383         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
 384 +       if (sbi->s_mmp_tsk)
 385 +               kthread_stop(sbi->s_mmp_tsk);
 386  failed_mount2:
 387         for (i = 0; i < db_count; i++)
 388                 brelse(sbi->s_group_desc[i]);
 389 @@ -3560,7 +3909,7 @@
 390         struct ext4_mount_options old_opts;
 391         ext4_group_t g;
 392         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
 393 -       int err;
 394 +       int err = 0;
 395  #ifdef CONFIG_QUOTA
 396         int i;
 397  #endif
 398 @@ -3682,6 +4031,13 @@
 399                                 goto restore_opts;
 400                         if (!ext4_setup_super(sb, es, 0))
 401                                 sb->s_flags &= ~MS_RDONLY;
 402 +                       if (EXT4_HAS_INCOMPAT_FEATURE(sb,
 403 +                                                   EXT4_FEATURE_INCOMPAT_MMP))
 404 +                               if (ext4_multi_mount_protect(sb,
 405 +                                               le64_to_cpu(es->s_mmp_block))) {
 406 +                                       err = -EROFS;
 407 +                                       goto restore_opts;
 408 +                               }
 409                 }
 410         }
 411         ext4_setup_system_zone(sb);
 412 Index: linux-stage/fs/ext4/ext4.h
 413 ===================================================================
 414 --- linux-stage.orig/fs/ext4/ext4.h     2011-03-03 15:25:02.507538421 +0800
 415 +++ linux-stage/fs/ext4/ext4.h  2011-03-05 12:25:16.343986732 +0800
 416 @@ -894,7 +894,7 @@
 417         __le16  s_want_extra_isize;     /* New inodes should reserve # bytes */
 418         __le32  s_flags;                /* Miscellaneous flags */
 419         __le16  s_raid_stride;          /* RAID stride */
 420 -       __le16  s_mmp_interval;         /* # seconds to wait in MMP checking */
 421 +       __le16  s_mmp_update_interval;  /* # seconds to wait in MMP checking */
 422         __le64  s_mmp_block;            /* Block for multi-mount protection */
 423         __le32  s_raid_stripe_width;    /* blocks on all data disks (N*stride)*/
 424         __u8    s_log_groups_per_flex;  /* FLEX_BG group size */
 425 @@ -1041,6 +1041,9 @@
 426
 427         /* workqueue for dio unwritten */
 428         struct workqueue_struct *dio_unwritten_wq;
 429 +
 430 +       /* Kernel thread for multiple mount protection */
 431 +       struct task_struct *s_mmp_tsk;
 432  };
 433
 434  static inline struct ext4_sb_info *EXT4_SB(struct super_block *sb)
 435 @@ -1177,7 +1180,8 @@
 436                                          EXT4_FEATURE_INCOMPAT_META_BG| \
 437                                          EXT4_FEATURE_INCOMPAT_EXTENTS| \
 438                                          EXT4_FEATURE_INCOMPAT_64BIT| \
 439 -                                        EXT4_FEATURE_INCOMPAT_FLEX_BG)
 440 +                                        EXT4_FEATURE_INCOMPAT_FLEX_BG| \
 441 +                                        EXT4_FEATURE_INCOMPAT_MMP)
 442  #define EXT4_FEATURE_RO_COMPAT_SUPP    (EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER| \
 443                                          EXT4_FEATURE_RO_COMPAT_LARGE_FILE| \
 444                                          EXT4_FEATURE_RO_COMPAT_GDT_CSUM| \
 445 @@ -1384,6 +1388,34 @@
 446  extern struct proc_dir_entry *ext4_proc_root;
 447
 448  /*
 449 + * This structure will be used for multiple mount protection. It will be
 450 + * written into the block number saved in the s_mmp_block field in the
 451 + * superblock. Programs that check MMP should assume that if
 452 + * SEQ_FSCK (or any unknown code above SEQ_MAX) is present then it is NOT safe
 453 + * to use the filesystem, regardless of how old the timestamp is.
 454 + */
 455 +#define EXT4_MMP_MAGIC     0x004D4D50U /* ASCII for MMP */
 456 +#define EXT4_MMP_SEQ_CLEAN 0xFF4D4D50U /* mmp_seq value for clean unmount */
 457 +#define EXT4_MMP_SEQ_FSCK  0xE24D4D50U /* mmp_seq value when being fscked */
 458 +#define EXT4_MMP_SEQ_MAX   0xE24D4D4FU /* maximum valid mmp_seq value */
 459 +
 460 +struct mmp_struct {
 461 +       __le32  mmp_magic;
 462 +       __le32  mmp_seq;
 463 +       __le64  mmp_time;
 464 +       char    mmp_nodename[64];
 465 +       char    mmp_bdevname[32];
 466 +       __le16  mmp_check_interval;
 467 +       __le16  mmp_pad1;
 468 +       __le32  mmp_pad2[227];
 469 +};
 470 +
 471 +/*
 472 + * Minimum interval for MMP checking in seconds.
 473 + */
 474 +#define EXT4_MMP_MIN_CHECK_INTERVAL    5
 475 +
 476 +/*
 477   * Function prototypes
 478   */
 479