ldiskfs/kernel_patches/patches/ext4-mmp-rhel5.patch

   1 Index: linux-stage/fs/ext4/super.c
   2 ===================================================================
   3 --- linux-stage.orig/fs/ext4/super.c
   4 +++ linux-stage/fs/ext4/super.c
   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 @@ -660,6 +662,8 @@ static void ext4_put_super(struct super_
  15                 invalidate_bdev(sbi->journal_bdev, 0);
  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 @@ -921,6 +925,354 @@ static int ext4_show_options(struct seq_
  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, __func__,
  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, "%s", 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 mmpd_data *) data)->sb;
 108 +       struct buffer_head *bh = ((struct mmpd_data *) data)->bh;
 109 +       struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 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 +       mmp = (struct mmp_struct *)(bh->b_data);
 122 +       mmp->mmp_time = cpu_to_le64(get_seconds());
 123 +       /*
 124 +        * Start with the higher mmp_check_interval and reduce it if
 125 +        * the MMP block is being updated on time.
 126 +        */
 127 +       mmp_check_interval = max(5UL * mmp_update_interval,
 128 +                                EXT4_MMP_MIN_CHECK_INTERVAL);
 129 +       mmp->mmp_check_interval = cpu_to_le16(mmp_check_interval);
 130 +       bdevname(bh->b_bdev, mmp->mmp_bdevname);
 131 +
 132 +       down_read(&uts_sem);
 133 +       memcpy(mmp->mmp_nodename, system_utsname.nodename,
 134 +              sizeof(mmp->mmp_nodename));
 135 +       up_read(&uts_sem);
 136 +
 137 +       while (!kthread_should_stop()) {
 138 +               if (++seq > EXT4_MMP_SEQ_MAX)
 139 +                       seq = 1;
 140 +
 141 +               mmp->mmp_seq = cpu_to_le32(seq);
 142 +               mmp->mmp_time = cpu_to_le64(get_seconds());
 143 +               last_update_time = jiffies;
 144 +
 145 +               retval = write_mmp_block(bh);
 146 +               /*
 147 +                * Don't spew too many error messages. Print one every
 148 +                * (s_mmp_update_interval * 60) seconds.
 149 +                */
 150 +               if (retval && (failed_writes % 60) == 0) {
 151 +                       ext4_error(sb, __func__,
 152 +                                  "Error writing to MMP block");
 153 +                       failed_writes++;
 154 +               }
 155 +
 156 +               if (!(le32_to_cpu(es->s_feature_incompat) &
 157 +                   EXT4_FEATURE_INCOMPAT_MMP)) {
 158 +                       ext4_warning(sb, __func__, "kmmpd being stopped "
 159 +                                    "since MMP feature has been disabled.");
 160 +                       EXT4_SB(sb)->s_mmp_tsk = NULL;
 161 +                       goto failed;
 162 +               }
 163 +
 164 +               if (sb->s_flags & MS_RDONLY) {
 165 +                       ext4_warning(sb, __func__, "kmmpd being stopped "
 166 +                                    "since filesystem has been remounted as "
 167 +                                    "readonly.");
 168 +                       EXT4_SB(sb)->s_mmp_tsk = NULL;
 169 +                       goto failed;
 170 +               }
 171 +
 172 +               diff = jiffies - last_update_time;
 173 +               if (diff < mmp_update_interval * HZ)
 174 +                       schedule_timeout_interruptible(EXT4_MMP_UPDATE_INTERVAL*
 175 +                                                      HZ - diff);
 176 +
 177 +               /*
 178 +                * We need to make sure that more than mmp_check_interval
 179 +                * seconds have not passed since writing. If that has happened
 180 +                * we need to check if the MMP block is as we left it.
 181 +                */
 182 +               diff = jiffies - last_update_time;
 183 +               if (diff > mmp_check_interval * HZ) {
 184 +                       struct buffer_head *bh_check = NULL;
 185 +                       struct mmp_struct *mmp_check;
 186 +
 187 +                       retval = read_mmp_block(sb, &bh_check, mmp_block);
 188 +                       if (retval) {
 189 +                               ext4_error(sb, __func__, "error reading MMP"
 190 +                                          "data: %d", retval);
 191 +                               EXT4_SB(sb)->s_mmp_tsk = NULL;
 192 +                               goto failed;
 193 +                       }
 194 +
 195 +                       mmp_check = (struct mmp_struct *)(bh_check->b_data);
 196 +                       if (mmp->mmp_seq != mmp_check->mmp_seq ||
 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 +                               ext4_error(sb, __func__, "abort");
 204 +                               goto failed;
 205 +                       }
 206 +                       put_bh(bh_check);
 207 +               }
 208 +
 209 +               /*
 210 +                * Adjust the mmp_check_interval depending on how much time
 211 +                * it took for the MMP block to be written.
 212 +                */
 213 +               mmp_check_interval = max(min(5 * diff / HZ,
 214 +                                            EXT4_MMP_MAX_CHECK_INTERVAL),
 215 +                                        EXT4_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(EXT4_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 + * EXT4_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 > EXT4_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 ext4_multi_mount_protect(struct super_block *sb,
 251 +                                   unsigned long mmp_block)
 252 +{
 253 +       struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 254 +       struct buffer_head *bh = NULL;
 255 +       struct mmp_struct *mmp = NULL;
 256 +       struct mmpd_data *mmpd_data;
 257 +       u32 seq;
 258 +       unsigned int mmp_check_interval = le16_to_cpu(es->s_mmp_update_interval);
 259 +       unsigned int wait_time = 0;
 260 +       int retval;
 261 +
 262 +       if (mmp_block < le32_to_cpu(es->s_first_data_block) ||
 263 +           mmp_block >= ext4_blocks_count(es)) {
 264 +               ext4_warning(sb, __func__,
 265 +                            "Invalid MMP block in superblock");
 266 +               goto failed;
 267 +       }
 268 +
 269 +       retval = read_mmp_block(sb, &bh, mmp_block);
 270 +       if (retval)
 271 +               goto failed;
 272 +
 273 +       mmp = (struct mmp_struct *)(bh->b_data);
 274 +
 275 +       if (mmp_check_interval < EXT4_MMP_MIN_CHECK_INTERVAL)
 276 +               mmp_check_interval = EXT4_MMP_MIN_CHECK_INTERVAL;
 277 +
 278 +       /*
 279 +        * If check_interval in MMP block is larger, use that instead of
 280 +        * update_interval from the superblock.
 281 +        */
 282 +       if (mmp->mmp_check_interval > mmp_check_interval)
 283 +               mmp_check_interval = mmp->mmp_check_interval;
 284 +
 285 +       seq = le32_to_cpu(mmp->mmp_seq);
 286 +       if (seq == EXT4_MMP_SEQ_CLEAN)
 287 +               goto skip;
 288 +
 289 +       if (seq == EXT4_MMP_SEQ_FSCK) {
 290 +               dump_mmp_msg(sb, mmp, __func__,
 291 +                            "fsck is running on the filesystem");
 292 +               goto failed;
 293 +       }
 294 +
 295 +       wait_time = min(mmp_check_interval * 2 + 1,
 296 +               mmp_check_interval + 60);
 297 +
 298 +       /* Print MMP interval if more than 20 secs. */
 299 +       if (wait_time > EXT4_MMP_MIN_CHECK_INTERVAL * 4)
 300 +               ext4_warning(sb, __func__, "MMP interval %u higher than "
 301 +                            "expected, please wait.\n", wait_time * 2);
 302 +
 303 +       if (schedule_timeout_interruptible(HZ * wait_time) != 0) {
 304 +               ext4_warning(sb, __func__, "MMP startup interrupted, failing "
 305 +                            "mount\n");
 306 +               goto failed;
 307 +       }
 308 +
 309 +       retval = read_mmp_block(sb, &bh, mmp_block);
 310 +       if (retval)
 311 +               goto failed;
 312 +       mmp = (struct mmp_struct *)(bh->b_data);
 313 +       if (seq != le32_to_cpu(mmp->mmp_seq)) {
 314 +               dump_mmp_msg(sb, mmp, __func__,
 315 +                            "Device is already active on another node.");
 316 +               goto failed;
 317 +       }
 318 +
 319 +skip:
 320 +       /*
 321 +        * write a new random sequence number.
 322 +        */
 323 +       mmp->mmp_seq = seq = cpu_to_le32(mmp_new_seq());
 324 +
 325 +       retval = write_mmp_block(bh);
 326 +       if (retval)
 327 +               goto failed;
 328 +
 329 +       /*
 330 +        * wait for MMP interval and check mmp_seq.
 331 +        */
 332 +       if (schedule_timeout_interruptible(HZ * wait_time) != 0) {
 333 +               ext4_warning(sb, __func__, "MMP startup interrupted, failing "
 334 +                            "mount\n");
 335 +               goto failed;
 336 +       }
 337 +
 338 +       retval = read_mmp_block(sb, &bh, mmp_block);
 339 +       if (retval)
 340 +               goto failed;
 341 +       mmp = (struct mmp_struct *)(bh->b_data);
 342 +       if (seq != le32_to_cpu(mmp->mmp_seq)) {
 343 +               dump_mmp_msg(sb, mmp, __func__,
 344 +                            "Device is already active on another node.");
 345 +               goto failed;
 346 +       }
 347 +
 348 +       mmpd_data = kmalloc(sizeof(struct mmpd_data *), GFP_KERNEL);
 349 +       if (!mmpd_data) {
 350 +               ext4_warning(sb, KERN_ERR, "not enough memory for mmpd_data");
 351 +               goto failed;
 352 +       }
 353 +       mmpd_data->sb = sb;
 354 +       mmpd_data->bh = bh;
 355 +
 356 +       /*
 357 +        * Start a kernel thread to update the MMP block periodically.
 358 +        */
 359 +       EXT4_SB(sb)->s_mmp_tsk = kthread_run(kmmpd, mmpd_data, "kmmpd-%s",
 360 +                                            bdevname(bh->b_bdev,
 361 +                                                     mmp->mmp_bdevname));
 362 +       if (IS_ERR(EXT4_SB(sb)->s_mmp_tsk)) {
 363 +               EXT4_SB(sb)->s_mmp_tsk = NULL;
 364 +               ext4_warning(sb, __func__, "Unable to create kmmpd thread "
 365 +                            "for %s.", sb->s_id);
 366 +               goto failed;
 367 +       }
 368 +
 369 +       return 0;
 370 +
 371 +failed:
 372 +       brelse(bh);
 373 +       return 1;
 374 +}
 375
 376  static struct dentry *ext4_get_dentry(struct super_block *sb, void *vobjp)
 377  {
 378 @@ -930,7 +1282,6 @@ static struct dentry *ext4_get_dentry(st
 379         struct inode *inode;
 380         struct dentry *result;
 381
 382 -
 383         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
 384                 return ERR_PTR(-ESTALE);
 385         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
 386 @@ -2740,6 +3091,11 @@ static int ext4_fill_super(struct super_
 387                           EXT4_HAS_INCOMPAT_FEATURE(sb,
 388                                     EXT4_FEATURE_INCOMPAT_RECOVER));
 389
 390 +       if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
 391 +           !(sb->s_flags & MS_RDONLY))
 392 +               if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
 393 +                       goto failed_mount3;
 394 +
 395         /*
 396          * The first inode we look at is the journal inode.  Don't try
 397          * root first: it may be modified in the journal!
 398 @@ -2978,6 +3334,8 @@ failed_mount3:
 399         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 400         percpu_counter_destroy(&sbi->s_dirs_counter);
 401         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
 402 +       if (sbi->s_mmp_tsk)
 403 +               kthread_stop(sbi->s_mmp_tsk);
 404  failed_mount2:
 405         for (i = 0; i < db_count; i++)
 406                 brelse(sbi->s_group_desc[i]);
 407 @@ -3488,7 +3846,7 @@ static int ext4_remount(struct super_blo
 408         struct ext4_mount_options old_opts;
 409         ext4_group_t g;
 410         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
 411 -       int err;
 412 +       int err = 0;
 413  #ifdef CONFIG_QUOTA
 414         int i;
 415  #endif
 416 @@ -3607,6 +3965,13 @@ static int ext4_remount(struct super_blo
 417                                 goto restore_opts;
 418                         if (!ext4_setup_super(sb, es, 0))
 419                                 sb->s_flags &= ~MS_RDONLY;
 420 +                       if (EXT4_HAS_INCOMPAT_FEATURE(sb,
 421 +                                                   EXT4_FEATURE_INCOMPAT_MMP))
 422 +                               if (ext4_multi_mount_protect(sb,
 423 +                                               le64_to_cpu(es->s_mmp_block))) {
 424 +                                       err = -EROFS;
 425 +                                       goto restore_opts;
 426 +                               }
 427                 }
 428         }
 429         ext4_setup_system_zone(sb);
 430 Index: linux-stage/fs/ext4/ext4.h
 431 ===================================================================
 432 --- linux-stage.orig/fs/ext4/ext4.h
 433 +++ linux-stage/fs/ext4/ext4.h
 434 @@ -851,7 +851,7 @@ struct ext4_super_block {
 435         __le16  s_want_extra_isize;     /* New inodes should reserve # bytes */
 436         __le32  s_flags;                /* Miscellaneous flags */
 437         __le16  s_raid_stride;          /* RAID stride */
 438 -       __le16  s_mmp_interval;         /* # seconds to wait in MMP checking */
 439 +       __le16  s_mmp_update_interval;  /* # seconds to wait in MMP checking */
 440         __le64  s_mmp_block;            /* Block for multi-mount protection */
 441         __le32  s_raid_stripe_width;    /* blocks on all data disks (N*stride)*/
 442         __u8    s_log_groups_per_flex;  /* FLEX_BG group size */
 443 @@ -1005,6 +1005,9 @@ struct ext4_sb_info {
 444
 445         /* workqueue for dio unwritten */
 446         struct workqueue_struct *dio_unwritten_wq;
 447 +
 448 +       /* Kernel thread for multiple mount protection */
 449 +       struct task_struct *s_mmp_tsk;
 450  };
 451
 452  static inline struct ext4_sb_info *EXT4_SB(struct super_block *sb)
 453 @@ -1114,7 +1117,8 @@ static inline int ext4_valid_inum(struct
 454                                          EXT4_FEATURE_INCOMPAT_META_BG| \
 455                                          EXT4_FEATURE_INCOMPAT_EXTENTS| \
 456                                          EXT4_FEATURE_INCOMPAT_64BIT| \
 457 -                                        EXT4_FEATURE_INCOMPAT_FLEX_BG)
 458 +                                        EXT4_FEATURE_INCOMPAT_FLEX_BG| \
 459 +                                        EXT4_FEATURE_INCOMPAT_MMP)
 460  #define EXT4_FEATURE_RO_COMPAT_SUPP    (EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER| \
 461                                          EXT4_FEATURE_RO_COMPAT_LARGE_FILE| \
 462                                          EXT4_FEATURE_RO_COMPAT_GDT_CSUM| \
 463 @@ -1296,6 +1300,50 @@ void ext4_get_group_no_and_offset(struct
 464  extern struct proc_dir_entry *ext4_proc_root;
 465
 466  /*
 467 + * This structure will be used for multiple mount protection. It will be
 468 + * written into the block number saved in the s_mmp_block field in the
 469 + * superblock. Programs that check MMP should assume that if
 470 + * SEQ_FSCK (or any unknown code above SEQ_MAX) is present then it is NOT safe
 471 + * to use the filesystem, regardless of how old the timestamp is.
 472 + */
 473 +#define EXT4_MMP_MAGIC     0x004D4D50U /* ASCII for MMP */
 474 +#define EXT4_MMP_SEQ_CLEAN 0xFF4D4D50U /* mmp_seq value for clean unmount */
 475 +#define EXT4_MMP_SEQ_FSCK  0xE24D4D50U /* mmp_seq value when being fscked */
 476 +#define EXT4_MMP_SEQ_MAX   0xE24D4D4FU /* maximum valid mmp_seq value */
 477 +
 478 +struct mmp_struct {
 479 +       __le32  mmp_magic;
 480 +       __le32  mmp_seq;
 481 +       __le64  mmp_time;
 482 +       char    mmp_nodename[64];
 483 +       char    mmp_bdevname[32];
 484 +       __le16  mmp_check_interval;
 485 +       __le16  mmp_pad1;
 486 +       __le32  mmp_pad2[227];
 487 +};
 488 +
 489 +/* arguments passed to the mmp thread */
 490 +struct mmpd_data {
 491 +       struct buffer_head *bh; /* bh from initial read_mmp_block() */
 492 +       struct super_block *sb;  /* super block of the fs */
 493 +};
 494 +
 495 +/*
 496 + * Default interval in seconds to update the MMP sequence number.
 497 + */
 498 +#define EXT4_MMP_UPDATE_INTERVAL   1
 499 +
 500 +/*
 501 + * Minimum interval for MMP checking in seconds.
 502 + */
 503 +#define EXT4_MMP_MIN_CHECK_INTERVAL    5UL
 504 +
 505 +/*
 506 + * Maximum interval for MMP checking in seconds.
 507 + */
 508 +#define EXT4_MMP_MAX_CHECK_INTERVAL    300UL
 509 +
 510 +/*
 511   * Function prototypes
 512   */
 513