fs/ext4/mmp.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include <linux/fs.h>
   3 #include <linux/random.h>
   4 #include <linux/buffer_head.h>
   5 #include <linux/utsname.h>
   6 #include <linux/kthread.h>
   7
   8 #include "ext4.h"
   9
  10 /* Checksumming functions */
  11 static __le32 ext4_mmp_csum(struct super_block *sb, struct mmp_struct *mmp)
  12 {
  13         struct ext4_sb_info *sbi = EXT4_SB(sb);
  14         int offset = offsetof(struct mmp_struct, mmp_checksum);
  15         __u32 csum;
  16
  17         csum = ext4_chksum(sbi, sbi->s_csum_seed, (char *)mmp, offset);
  18
  19         return cpu_to_le32(csum);
  20 }
  21
  22 static int ext4_mmp_csum_verify(struct super_block *sb, struct mmp_struct *mmp)
  23 {
  24         if (!ext4_has_metadata_csum(sb))
  25                 return 1;
  26
  27         return mmp->mmp_checksum == ext4_mmp_csum(sb, mmp);
  28 }
  29
  30 static void ext4_mmp_csum_set(struct super_block *sb, struct mmp_struct *mmp)
  31 {
  32         if (!ext4_has_metadata_csum(sb))
  33                 return;
  34
  35         mmp->mmp_checksum = ext4_mmp_csum(sb, mmp);
  36 }
  37
  38 /*
  39  * Write the MMP block using REQ_SYNC to try to get the block on-disk
  40  * faster.
  41  */
  42 static int write_mmp_block(struct super_block *sb, struct buffer_head *bh)
  43 {
  44         struct mmp_struct *mmp = (struct mmp_struct *)(bh->b_data);
  45
  46         /*
  47          * We protect against freezing so that we don't create dirty buffers
  48          * on frozen filesystem.
  49          */
  50         sb_start_write(sb);
  51         ext4_mmp_csum_set(sb, mmp);
  52         lock_buffer(bh);
  53         bh->b_end_io = end_buffer_write_sync;
  54         get_bh(bh);
  55         submit_bh(REQ_OP_WRITE, REQ_SYNC | REQ_META | REQ_PRIO, bh);
  56         wait_on_buffer(bh);
  57         sb_end_write(sb);
  58         if (unlikely(!buffer_uptodate(bh)))
  59                 return 1;
  60
  61         return 0;
  62 }
  63
  64 /*
  65  * Read the MMP block. It _must_ be read from disk and hence we clear the
  66  * uptodate flag on the buffer.
  67  */
  68 static int read_mmp_block(struct super_block *sb, struct buffer_head **bh,
  69                           ext4_fsblk_t mmp_block)
  70 {
  71         struct mmp_struct *mmp;
  72         int ret;
  73
  74         if (*bh)
  75                 clear_buffer_uptodate(*bh);
  76
  77         /* This would be sb_bread(sb, mmp_block), except we need to be sure
  78          * that the MD RAID device cache has been bypassed, and that the read
  79          * is not blocked in the elevator. */
  80         if (!*bh) {
  81                 *bh = sb_getblk(sb, mmp_block);
  82                 if (!*bh) {
  83                         ret = -ENOMEM;
  84                         goto warn_exit;
  85                 }
  86         }
  87
  88         lock_buffer(*bh);
  89         ret = ext4_read_bh(*bh, REQ_META | REQ_PRIO, NULL);
  90         if (ret)
  91                 goto warn_exit;
  92
  93         mmp = (struct mmp_struct *)((*bh)->b_data);
  94         if (le32_to_cpu(mmp->mmp_magic) != EXT4_MMP_MAGIC) {
  95                 ret = -EFSCORRUPTED;
  96                 goto warn_exit;
  97         }
  98         if (!ext4_mmp_csum_verify(sb, mmp)) {
  99                 ret = -EFSBADCRC;
 100                 goto warn_exit;
 101         }
 102         return 0;
 103 warn_exit:
 104         brelse(*bh);
 105         *bh = NULL;
 106         ext4_warning(sb, "Error %d while reading MMP block %llu",
 107                      ret, mmp_block);
 108         return ret;
 109 }
 110
 111 /*
 112  * Dump as much information as possible to help the admin.
 113  */
 114 void __dump_mmp_msg(struct super_block *sb, struct mmp_struct *mmp,
 115                     const char *function, unsigned int line, const char *msg)
 116 {
 117         __ext4_warning(sb, function, line, "%s", msg);
 118         __ext4_warning(sb, function, line,
 119                        "MMP failure info: last update time: %llu, last update node: %.*s, last update device: %.*s",
 120                        (unsigned long long)le64_to_cpu(mmp->mmp_time),
 121                        (int)sizeof(mmp->mmp_nodename), mmp->mmp_nodename,
 122                        (int)sizeof(mmp->mmp_bdevname), mmp->mmp_bdevname);
 123 }
 124
 125 /*
 126  * kmmpd will update the MMP sequence every s_mmp_update_interval seconds
 127  */
 128 static int kmmpd(void *data)
 129 {
 130         struct super_block *sb = ((struct mmpd_data *) data)->sb;
 131         struct buffer_head *bh = ((struct mmpd_data *) data)->bh;
 132         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 133         struct mmp_struct *mmp;
 134         ext4_fsblk_t mmp_block;
 135         u32 seq = 0;
 136         unsigned long failed_writes = 0;
 137         int mmp_update_interval = le16_to_cpu(es->s_mmp_update_interval);
 138         unsigned mmp_check_interval;
 139         unsigned long last_update_time;
 140         unsigned long diff;
 141         int retval;
 142
 143         mmp_block = le64_to_cpu(es->s_mmp_block);
 144         mmp = (struct mmp_struct *)(bh->b_data);
 145         mmp->mmp_time = cpu_to_le64(ktime_get_real_seconds());
 146         /*
 147          * Start with the higher mmp_check_interval and reduce it if
 148          * the MMP block is being updated on time.
 149          */
 150         mmp_check_interval = max(EXT4_MMP_CHECK_MULT * mmp_update_interval,
 151                                  EXT4_MMP_MIN_CHECK_INTERVAL);
 152         mmp->mmp_check_interval = cpu_to_le16(mmp_check_interval);
 153         BUILD_BUG_ON(sizeof(mmp->mmp_bdevname) < BDEVNAME_SIZE);
 154         bdevname(bh->b_bdev, mmp->mmp_bdevname);
 155
 156         memcpy(mmp->mmp_nodename, init_utsname()->nodename,
 157                sizeof(mmp->mmp_nodename));
 158
 159         while (!kthread_should_stop()) {
 160                 if (++seq > EXT4_MMP_SEQ_MAX)
 161                         seq = 1;
 162
 163                 mmp->mmp_seq = cpu_to_le32(seq);
 164                 mmp->mmp_time = cpu_to_le64(ktime_get_real_seconds());
 165                 last_update_time = jiffies;
 166
 167                 retval = write_mmp_block(sb, bh);
 168                 /*
 169                  * Don't spew too many error messages. Print one every
 170                  * (s_mmp_update_interval * 60) seconds.
 171                  */
 172                 if (retval) {
 173                         if ((failed_writes % 60) == 0) {
 174                                 ext4_error_err(sb, -retval,
 175                                                "Error writing to MMP block");
 176                         }
 177                         failed_writes++;
 178                 }
 179
 180                 if (!(le32_to_cpu(es->s_feature_incompat) &
 181                     EXT4_FEATURE_INCOMPAT_MMP)) {
 182                         ext4_warning(sb, "kmmpd being stopped since MMP feature"
 183                                      " has been disabled.");
 184                         goto exit_thread;
 185                 }
 186
 187                 if (sb_rdonly(sb))
 188                         break;
 189
 190                 diff = jiffies - last_update_time;
 191                 if (diff < mmp_update_interval * HZ)
 192                         schedule_timeout_interruptible(mmp_update_interval *
 193                                                        HZ - diff);
 194
 195                 /*
 196                  * We need to make sure that more than mmp_check_interval
 197                  * seconds have not passed since writing. If that has happened
 198                  * we need to check if the MMP block is as we left it.
 199                  */
 200                 diff = jiffies - last_update_time;
 201                 if (diff > mmp_check_interval * HZ) {
 202                         struct buffer_head *bh_check = NULL;
 203                         struct mmp_struct *mmp_check;
 204
 205                         retval = read_mmp_block(sb, &bh_check, mmp_block);
 206                         if (retval) {
 207                                 ext4_error_err(sb, -retval,
 208                                                "error reading MMP data: %d",
 209                                                retval);
 210                                 goto exit_thread;
 211                         }
 212
 213                         mmp_check = (struct mmp_struct *)(bh_check->b_data);
 214                         if (mmp->mmp_seq != mmp_check->mmp_seq ||
 215                             memcmp(mmp->mmp_nodename, mmp_check->mmp_nodename,
 216                                    sizeof(mmp->mmp_nodename))) {
 217                                 dump_mmp_msg(sb, mmp_check,
 218                                              "Error while updating MMP info. "
 219                                              "The filesystem seems to have been"
 220                                              " multiply mounted.");
 221                                 ext4_error_err(sb, EBUSY, "abort");
 222                                 put_bh(bh_check);
 223                                 retval = -EBUSY;
 224                                 goto exit_thread;
 225                         }
 226                         put_bh(bh_check);
 227                 }
 228
 229                  /*
 230                  * Adjust the mmp_check_interval depending on how much time
 231                  * it took for the MMP block to be written.
 232                  */
 233                 mmp_check_interval = max(min(EXT4_MMP_CHECK_MULT * diff / HZ,
 234                                              EXT4_MMP_MAX_CHECK_INTERVAL),
 235                                          EXT4_MMP_MIN_CHECK_INTERVAL);
 236                 mmp->mmp_check_interval = cpu_to_le16(mmp_check_interval);
 237         }
 238
 239         /*
 240          * Unmount seems to be clean.
 241          */
 242         mmp->mmp_seq = cpu_to_le32(EXT4_MMP_SEQ_CLEAN);
 243         mmp->mmp_time = cpu_to_le64(ktime_get_real_seconds());
 244
 245         retval = write_mmp_block(sb, bh);
 246
 247 exit_thread:
 248         EXT4_SB(sb)->s_mmp_tsk = NULL;
 249         kfree(data);
 250         brelse(bh);
 251         return retval;
 252 }
 253
 254 /*
 255  * Get a random new sequence number but make sure it is not greater than
 256  * EXT4_MMP_SEQ_MAX.
 257  */
 258 static unsigned int mmp_new_seq(void)
 259 {
 260         u32 new_seq;
 261
 262         do {
 263                 new_seq = prandom_u32();
 264         } while (new_seq > EXT4_MMP_SEQ_MAX);
 265
 266         return new_seq;
 267 }
 268
 269 /*
 270  * Protect the filesystem from being mounted more than once.
 271  */
 272 int ext4_multi_mount_protect(struct super_block *sb,
 273                                     ext4_fsblk_t mmp_block)
 274 {
 275         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 276         struct buffer_head *bh = NULL;
 277         struct mmp_struct *mmp = NULL;
 278         struct mmpd_data *mmpd_data;
 279         u32 seq;
 280         unsigned int mmp_check_interval = le16_to_cpu(es->s_mmp_update_interval);
 281         unsigned int wait_time = 0;
 282         int retval;
 283
 284         if (mmp_block < le32_to_cpu(es->s_first_data_block) ||
 285             mmp_block >= ext4_blocks_count(es)) {
 286                 ext4_warning(sb, "Invalid MMP block in superblock");
 287                 goto failed;
 288         }
 289
 290         retval = read_mmp_block(sb, &bh, mmp_block);
 291         if (retval)
 292                 goto failed;
 293
 294         mmp = (struct mmp_struct *)(bh->b_data);
 295
 296         if (mmp_check_interval < EXT4_MMP_MIN_CHECK_INTERVAL)
 297                 mmp_check_interval = EXT4_MMP_MIN_CHECK_INTERVAL;
 298
 299         /*
 300          * If check_interval in MMP block is larger, use that instead of
 301          * update_interval from the superblock.
 302          */
 303         if (le16_to_cpu(mmp->mmp_check_interval) > mmp_check_interval)
 304                 mmp_check_interval = le16_to_cpu(mmp->mmp_check_interval);
 305
 306         seq = le32_to_cpu(mmp->mmp_seq);
 307         if (seq == EXT4_MMP_SEQ_CLEAN)
 308                 goto skip;
 309
 310         if (seq == EXT4_MMP_SEQ_FSCK) {
 311                 dump_mmp_msg(sb, mmp, "fsck is running on the filesystem");
 312                 goto failed;
 313         }
 314
 315         wait_time = min(mmp_check_interval * 2 + 1,
 316                         mmp_check_interval + 60);
 317
 318         /* Print MMP interval if more than 20 secs. */
 319         if (wait_time > EXT4_MMP_MIN_CHECK_INTERVAL * 4)
 320                 ext4_warning(sb, "MMP interval %u higher than expected, please"
 321                              " wait.\n", wait_time * 2);
 322
 323         if (schedule_timeout_interruptible(HZ * wait_time) != 0) {
 324                 ext4_warning(sb, "MMP startup interrupted, failing mount\n");
 325                 goto failed;
 326         }
 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,
 334                              "Device is already active on another node.");
 335                 goto failed;
 336         }
 337
 338 skip:
 339         /*
 340          * write a new random sequence number.
 341          */
 342         seq = mmp_new_seq();
 343         mmp->mmp_seq = cpu_to_le32(seq);
 344
 345         retval = write_mmp_block(sb, bh);
 346         if (retval)
 347                 goto failed;
 348
 349         /*
 350          * wait for MMP interval and check mmp_seq.
 351          */
 352         if (schedule_timeout_interruptible(HZ * wait_time) != 0) {
 353                 ext4_warning(sb, "MMP startup interrupted, failing mount");
 354                 goto failed;
 355         }
 356
 357         retval = read_mmp_block(sb, &bh, mmp_block);
 358         if (retval)
 359                 goto failed;
 360         mmp = (struct mmp_struct *)(bh->b_data);
 361         if (seq != le32_to_cpu(mmp->mmp_seq)) {
 362                 dump_mmp_msg(sb, mmp,
 363                              "Device is already active on another node.");
 364                 goto failed;
 365         }
 366
 367         mmpd_data = kmalloc(sizeof(*mmpd_data), GFP_KERNEL);
 368         if (!mmpd_data) {
 369                 ext4_warning(sb, "not enough memory for mmpd_data");
 370                 goto failed;
 371         }
 372         mmpd_data->sb = sb;
 373         mmpd_data->bh = bh;
 374
 375         /*
 376          * Start a kernel thread to update the MMP block periodically.
 377          */
 378         EXT4_SB(sb)->s_mmp_tsk = kthread_run(kmmpd, mmpd_data, "kmmpd-%.*s",
 379                                              (int)sizeof(mmp->mmp_bdevname),
 380                                              bdevname(bh->b_bdev,
 381                                                       mmp->mmp_bdevname));
 382         if (IS_ERR(EXT4_SB(sb)->s_mmp_tsk)) {
 383                 EXT4_SB(sb)->s_mmp_tsk = NULL;
 384                 kfree(mmpd_data);
 385                 ext4_warning(sb, "Unable to create kmmpd thread for %s.",
 386                              sb->s_id);
 387                 goto failed;
 388         }
 389
 390         return 0;
 391
 392 failed:
 393         brelse(bh);
 394         return 1;
 395 }
 396
 397