2 #include <linux/random.h>
3 #include <linux/buffer_head.h>
4 #include <linux/utsname.h>
5 #include <linux/kthread.h>
9 /* Checksumming functions */
10 static __u32 ext4_mmp_csum(struct super_block *sb, struct mmp_struct *mmp)
12 struct ext4_sb_info *sbi = EXT4_SB(sb);
13 int offset = offsetof(struct mmp_struct, mmp_checksum);
16 csum = ext4_chksum(sbi, sbi->s_csum_seed, (char *)mmp, offset);
18 return cpu_to_le32(csum);
21 int ext4_mmp_csum_verify(struct super_block *sb, struct mmp_struct *mmp)
23 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
24 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
27 return mmp->mmp_checksum == ext4_mmp_csum(sb, mmp);
30 void ext4_mmp_csum_set(struct super_block *sb, struct mmp_struct *mmp)
32 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
33 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
36 mmp->mmp_checksum = ext4_mmp_csum(sb, mmp);
40 * Write the MMP block using WRITE_SYNC to try to get the block on-disk
43 static int write_mmp_block(struct super_block *sb, struct buffer_head *bh)
45 struct mmp_struct *mmp = (struct mmp_struct *)(bh->b_data);
48 * We protect against freezing so that we don't create dirty buffers
49 * on frozen filesystem.
52 ext4_mmp_csum_set(sb, mmp);
53 mark_buffer_dirty(bh);
55 bh->b_end_io = end_buffer_write_sync;
57 submit_bh(WRITE_SYNC, bh);
60 if (unlikely(!buffer_uptodate(bh)))
67 * Read the MMP block. It _must_ be read from disk and hence we clear the
68 * uptodate flag on the buffer.
70 static int read_mmp_block(struct super_block *sb, struct buffer_head **bh,
71 ext4_fsblk_t mmp_block)
73 struct mmp_struct *mmp;
76 clear_buffer_uptodate(*bh);
78 /* This would be sb_bread(sb, mmp_block), except we need to be sure
79 * that the MD RAID device cache has been bypassed, and that the read
80 * is not blocked in the elevator. */
82 *bh = sb_getblk(sb, mmp_block);
86 (*bh)->b_end_io = end_buffer_read_sync;
87 submit_bh(READ_SYNC, *bh);
89 if (!buffer_uptodate(*bh)) {
95 ext4_warning(sb, "Error while reading MMP block %llu",
100 mmp = (struct mmp_struct *)((*bh)->b_data);
101 if (le32_to_cpu(mmp->mmp_magic) != EXT4_MMP_MAGIC ||
102 !ext4_mmp_csum_verify(sb, mmp))
109 * Dump as much information as possible to help the admin.
111 void __dump_mmp_msg(struct super_block *sb, struct mmp_struct *mmp,
112 const char *function, unsigned int line, const char *msg)
114 __ext4_warning(sb, function, line, msg);
115 __ext4_warning(sb, function, line,
116 "MMP failure info: last update time: %llu, last update "
117 "node: %s, last update device: %s\n",
118 (long long unsigned int) le64_to_cpu(mmp->mmp_time),
119 mmp->mmp_nodename, mmp->mmp_bdevname);
123 * kmmpd will update the MMP sequence every s_mmp_update_interval seconds
125 static int kmmpd(void *data)
127 struct super_block *sb = ((struct mmpd_data *) data)->sb;
128 struct buffer_head *bh = ((struct mmpd_data *) data)->bh;
129 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
130 struct mmp_struct *mmp;
131 ext4_fsblk_t mmp_block;
133 unsigned long failed_writes = 0;
134 int mmp_update_interval = le16_to_cpu(es->s_mmp_update_interval);
135 unsigned mmp_check_interval;
136 unsigned long last_update_time;
140 mmp_block = le64_to_cpu(es->s_mmp_block);
141 mmp = (struct mmp_struct *)(bh->b_data);
142 mmp->mmp_time = cpu_to_le64(get_seconds());
144 * Start with the higher mmp_check_interval and reduce it if
145 * the MMP block is being updated on time.
147 mmp_check_interval = max(EXT4_MMP_CHECK_MULT * mmp_update_interval,
148 EXT4_MMP_MIN_CHECK_INTERVAL);
149 mmp->mmp_check_interval = cpu_to_le16(mmp_check_interval);
150 bdevname(bh->b_bdev, mmp->mmp_bdevname);
152 memcpy(mmp->mmp_nodename, init_utsname()->nodename,
153 sizeof(mmp->mmp_nodename));
155 while (!kthread_should_stop()) {
156 if (++seq > EXT4_MMP_SEQ_MAX)
159 mmp->mmp_seq = cpu_to_le32(seq);
160 mmp->mmp_time = cpu_to_le64(get_seconds());
161 last_update_time = jiffies;
163 retval = write_mmp_block(sb, bh);
165 * Don't spew too many error messages. Print one every
166 * (s_mmp_update_interval * 60) seconds.
169 if ((failed_writes % 60) == 0)
170 ext4_error(sb, "Error writing to MMP block");
174 if (!(le32_to_cpu(es->s_feature_incompat) &
175 EXT4_FEATURE_INCOMPAT_MMP)) {
176 ext4_warning(sb, "kmmpd being stopped since MMP feature"
177 " has been disabled.");
178 EXT4_SB(sb)->s_mmp_tsk = NULL;
182 if (sb->s_flags & MS_RDONLY) {
183 ext4_warning(sb, "kmmpd being stopped since filesystem "
184 "has been remounted as readonly.");
185 EXT4_SB(sb)->s_mmp_tsk = NULL;
189 diff = jiffies - last_update_time;
190 if (diff < mmp_update_interval * HZ)
191 schedule_timeout_interruptible(mmp_update_interval *
195 * We need to make sure that more than mmp_check_interval
196 * seconds have not passed since writing. If that has happened
197 * we need to check if the MMP block is as we left it.
199 diff = jiffies - last_update_time;
200 if (diff > mmp_check_interval * HZ) {
201 struct buffer_head *bh_check = NULL;
202 struct mmp_struct *mmp_check;
204 retval = read_mmp_block(sb, &bh_check, mmp_block);
206 ext4_error(sb, "error reading MMP data: %d",
209 EXT4_SB(sb)->s_mmp_tsk = NULL;
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(sb, "abort");
228 * Adjust the mmp_check_interval depending on how much time
229 * it took for the MMP block to be written.
231 mmp_check_interval = max(min(EXT4_MMP_CHECK_MULT * diff / HZ,
232 EXT4_MMP_MAX_CHECK_INTERVAL),
233 EXT4_MMP_MIN_CHECK_INTERVAL);
234 mmp->mmp_check_interval = cpu_to_le16(mmp_check_interval);
238 * Unmount seems to be clean.
240 mmp->mmp_seq = cpu_to_le32(EXT4_MMP_SEQ_CLEAN);
241 mmp->mmp_time = cpu_to_le64(get_seconds());
243 retval = write_mmp_block(sb, bh);
252 * Get a random new sequence number but make sure it is not greater than
255 static unsigned int mmp_new_seq(void)
260 get_random_bytes(&new_seq, sizeof(u32));
261 } while (new_seq > EXT4_MMP_SEQ_MAX);
267 * Protect the filesystem from being mounted more than once.
269 int ext4_multi_mount_protect(struct super_block *sb,
270 ext4_fsblk_t mmp_block)
272 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
273 struct buffer_head *bh = NULL;
274 struct mmp_struct *mmp = NULL;
275 struct mmpd_data *mmpd_data;
277 unsigned int mmp_check_interval = le16_to_cpu(es->s_mmp_update_interval);
278 unsigned int wait_time = 0;
281 if (mmp_block < le32_to_cpu(es->s_first_data_block) ||
282 mmp_block >= ext4_blocks_count(es)) {
283 ext4_warning(sb, "Invalid MMP block in superblock");
287 retval = read_mmp_block(sb, &bh, mmp_block);
291 mmp = (struct mmp_struct *)(bh->b_data);
293 if (mmp_check_interval < EXT4_MMP_MIN_CHECK_INTERVAL)
294 mmp_check_interval = EXT4_MMP_MIN_CHECK_INTERVAL;
297 * If check_interval in MMP block is larger, use that instead of
298 * update_interval from the superblock.
300 if (le16_to_cpu(mmp->mmp_check_interval) > mmp_check_interval)
301 mmp_check_interval = le16_to_cpu(mmp->mmp_check_interval);
303 seq = le32_to_cpu(mmp->mmp_seq);
304 if (seq == EXT4_MMP_SEQ_CLEAN)
307 if (seq == EXT4_MMP_SEQ_FSCK) {
308 dump_mmp_msg(sb, mmp, "fsck is running on the filesystem");
312 wait_time = min(mmp_check_interval * 2 + 1,
313 mmp_check_interval + 60);
315 /* Print MMP interval if more than 20 secs. */
316 if (wait_time > EXT4_MMP_MIN_CHECK_INTERVAL * 4)
317 ext4_warning(sb, "MMP interval %u higher than expected, please"
318 " wait.\n", wait_time * 2);
320 if (schedule_timeout_interruptible(HZ * wait_time) != 0) {
321 ext4_warning(sb, "MMP startup interrupted, failing mount\n");
325 retval = read_mmp_block(sb, &bh, mmp_block);
328 mmp = (struct mmp_struct *)(bh->b_data);
329 if (seq != le32_to_cpu(mmp->mmp_seq)) {
330 dump_mmp_msg(sb, mmp,
331 "Device is already active on another node.");
337 * write a new random sequence number.
340 mmp->mmp_seq = cpu_to_le32(seq);
342 retval = write_mmp_block(sb, bh);
347 * wait for MMP interval and check mmp_seq.
349 if (schedule_timeout_interruptible(HZ * wait_time) != 0) {
350 ext4_warning(sb, "MMP startup interrupted, failing mount\n");
354 retval = read_mmp_block(sb, &bh, mmp_block);
357 mmp = (struct mmp_struct *)(bh->b_data);
358 if (seq != le32_to_cpu(mmp->mmp_seq)) {
359 dump_mmp_msg(sb, mmp,
360 "Device is already active on another node.");
364 mmpd_data = kmalloc(sizeof(struct mmpd_data), GFP_KERNEL);
366 ext4_warning(sb, "not enough memory for mmpd_data");
373 * Start a kernel thread to update the MMP block periodically.
375 EXT4_SB(sb)->s_mmp_tsk = kthread_run(kmmpd, mmpd_data, "kmmpd-%s",
378 if (IS_ERR(EXT4_SB(sb)->s_mmp_tsk)) {
379 EXT4_SB(sb)->s_mmp_tsk = NULL;
381 ext4_warning(sb, "Unable to create kmmpd thread for %s.",