2 #include <linux/random.h>
3 #include <linux/buffer_head.h>
4 #include <linux/utsname.h>
5 #include <linux/kthread.h>
10 * Write the MMP block using WRITE_SYNC to try to get the block on-disk
13 static int write_mmp_block(struct buffer_head *bh)
15 mark_buffer_dirty(bh);
17 bh->b_end_io = end_buffer_write_sync;
19 submit_bh(WRITE_SYNC, bh);
21 if (unlikely(!buffer_uptodate(bh)))
28 * Read the MMP block. It _must_ be read from disk and hence we clear the
29 * uptodate flag on the buffer.
31 static int read_mmp_block(struct super_block *sb, struct buffer_head **bh,
32 ext4_fsblk_t mmp_block)
34 struct mmp_struct *mmp;
37 clear_buffer_uptodate(*bh);
39 /* This would be sb_bread(sb, mmp_block), except we need to be sure
40 * that the MD RAID device cache has been bypassed, and that the read
41 * is not blocked in the elevator. */
43 *bh = sb_getblk(sb, mmp_block);
47 (*bh)->b_end_io = end_buffer_read_sync;
48 submit_bh(READ_SYNC, *bh);
50 if (!buffer_uptodate(*bh)) {
56 ext4_warning(sb, "Error while reading MMP block %llu",
61 mmp = (struct mmp_struct *)((*bh)->b_data);
62 if (le32_to_cpu(mmp->mmp_magic) != EXT4_MMP_MAGIC)
69 * Dump as much information as possible to help the admin.
71 void __dump_mmp_msg(struct super_block *sb, struct mmp_struct *mmp,
72 const char *function, unsigned int line, const char *msg)
74 __ext4_warning(sb, function, line, msg);
75 __ext4_warning(sb, function, line,
76 "MMP failure info: last update time: %llu, last update "
77 "node: %s, last update device: %s\n",
78 (long long unsigned int) le64_to_cpu(mmp->mmp_time),
79 mmp->mmp_nodename, mmp->mmp_bdevname);
83 * kmmpd will update the MMP sequence every s_mmp_update_interval seconds
85 static int kmmpd(void *data)
87 struct super_block *sb = ((struct mmpd_data *) data)->sb;
88 struct buffer_head *bh = ((struct mmpd_data *) data)->bh;
89 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
90 struct mmp_struct *mmp;
91 ext4_fsblk_t mmp_block;
93 unsigned long failed_writes = 0;
94 int mmp_update_interval = le16_to_cpu(es->s_mmp_update_interval);
95 unsigned mmp_check_interval;
96 unsigned long last_update_time;
100 mmp_block = le64_to_cpu(es->s_mmp_block);
101 mmp = (struct mmp_struct *)(bh->b_data);
102 mmp->mmp_time = cpu_to_le64(get_seconds());
104 * Start with the higher mmp_check_interval and reduce it if
105 * the MMP block is being updated on time.
107 mmp_check_interval = max(EXT4_MMP_CHECK_MULT * mmp_update_interval,
108 EXT4_MMP_MIN_CHECK_INTERVAL);
109 mmp->mmp_check_interval = cpu_to_le16(mmp_check_interval);
110 bdevname(bh->b_bdev, mmp->mmp_bdevname);
112 memcpy(mmp->mmp_nodename, init_utsname()->sysname,
113 sizeof(mmp->mmp_nodename));
115 while (!kthread_should_stop()) {
116 if (++seq > EXT4_MMP_SEQ_MAX)
119 mmp->mmp_seq = cpu_to_le32(seq);
120 mmp->mmp_time = cpu_to_le64(get_seconds());
121 last_update_time = jiffies;
123 retval = write_mmp_block(bh);
125 * Don't spew too many error messages. Print one every
126 * (s_mmp_update_interval * 60) seconds.
128 if (retval && (failed_writes % 60) == 0) {
129 ext4_error(sb, "Error writing to MMP block");
133 if (!(le32_to_cpu(es->s_feature_incompat) &
134 EXT4_FEATURE_INCOMPAT_MMP)) {
135 ext4_warning(sb, "kmmpd being stopped since MMP feature"
136 " has been disabled.");
137 EXT4_SB(sb)->s_mmp_tsk = NULL;
141 if (sb->s_flags & MS_RDONLY) {
142 ext4_warning(sb, "kmmpd being stopped since filesystem "
143 "has been remounted as readonly.");
144 EXT4_SB(sb)->s_mmp_tsk = NULL;
148 diff = jiffies - last_update_time;
149 if (diff < mmp_update_interval * HZ)
150 schedule_timeout_interruptible(mmp_update_interval *
154 * We need to make sure that more than mmp_check_interval
155 * seconds have not passed since writing. If that has happened
156 * we need to check if the MMP block is as we left it.
158 diff = jiffies - last_update_time;
159 if (diff > mmp_check_interval * HZ) {
160 struct buffer_head *bh_check = NULL;
161 struct mmp_struct *mmp_check;
163 retval = read_mmp_block(sb, &bh_check, mmp_block);
165 ext4_error(sb, "error reading MMP data: %d",
168 EXT4_SB(sb)->s_mmp_tsk = NULL;
172 mmp_check = (struct mmp_struct *)(bh_check->b_data);
173 if (mmp->mmp_seq != mmp_check->mmp_seq ||
174 memcmp(mmp->mmp_nodename, mmp_check->mmp_nodename,
175 sizeof(mmp->mmp_nodename))) {
176 dump_mmp_msg(sb, mmp_check,
177 "Error while updating MMP info. "
178 "The filesystem seems to have been"
179 " multiply mounted.");
180 ext4_error(sb, "abort");
187 * Adjust the mmp_check_interval depending on how much time
188 * it took for the MMP block to be written.
190 mmp_check_interval = max(min(EXT4_MMP_CHECK_MULT * diff / HZ,
191 EXT4_MMP_MAX_CHECK_INTERVAL),
192 EXT4_MMP_MIN_CHECK_INTERVAL);
193 mmp->mmp_check_interval = cpu_to_le16(mmp_check_interval);
197 * Unmount seems to be clean.
199 mmp->mmp_seq = cpu_to_le32(EXT4_MMP_SEQ_CLEAN);
200 mmp->mmp_time = cpu_to_le64(get_seconds());
202 retval = write_mmp_block(bh);
211 * Get a random new sequence number but make sure it is not greater than
214 static unsigned int mmp_new_seq(void)
219 get_random_bytes(&new_seq, sizeof(u32));
220 } while (new_seq > EXT4_MMP_SEQ_MAX);
226 * Protect the filesystem from being mounted more than once.
228 int ext4_multi_mount_protect(struct super_block *sb,
229 ext4_fsblk_t mmp_block)
231 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
232 struct buffer_head *bh = NULL;
233 struct mmp_struct *mmp = NULL;
234 struct mmpd_data *mmpd_data;
236 unsigned int mmp_check_interval = le16_to_cpu(es->s_mmp_update_interval);
237 unsigned int wait_time = 0;
240 if (mmp_block < le32_to_cpu(es->s_first_data_block) ||
241 mmp_block >= ext4_blocks_count(es)) {
242 ext4_warning(sb, "Invalid MMP block in superblock");
246 retval = read_mmp_block(sb, &bh, mmp_block);
250 mmp = (struct mmp_struct *)(bh->b_data);
252 if (mmp_check_interval < EXT4_MMP_MIN_CHECK_INTERVAL)
253 mmp_check_interval = EXT4_MMP_MIN_CHECK_INTERVAL;
256 * If check_interval in MMP block is larger, use that instead of
257 * update_interval from the superblock.
259 if (mmp->mmp_check_interval > mmp_check_interval)
260 mmp_check_interval = mmp->mmp_check_interval;
262 seq = le32_to_cpu(mmp->mmp_seq);
263 if (seq == EXT4_MMP_SEQ_CLEAN)
266 if (seq == EXT4_MMP_SEQ_FSCK) {
267 dump_mmp_msg(sb, mmp, "fsck is running on the filesystem");
271 wait_time = min(mmp_check_interval * 2 + 1,
272 mmp_check_interval + 60);
274 /* Print MMP interval if more than 20 secs. */
275 if (wait_time > EXT4_MMP_MIN_CHECK_INTERVAL * 4)
276 ext4_warning(sb, "MMP interval %u higher than expected, please"
277 " wait.\n", wait_time * 2);
279 if (schedule_timeout_interruptible(HZ * wait_time) != 0) {
280 ext4_warning(sb, "MMP startup interrupted, failing mount\n");
284 retval = read_mmp_block(sb, &bh, mmp_block);
287 mmp = (struct mmp_struct *)(bh->b_data);
288 if (seq != le32_to_cpu(mmp->mmp_seq)) {
289 dump_mmp_msg(sb, mmp,
290 "Device is already active on another node.");
296 * write a new random sequence number.
298 mmp->mmp_seq = seq = cpu_to_le32(mmp_new_seq());
300 retval = write_mmp_block(bh);
305 * wait for MMP interval and check mmp_seq.
307 if (schedule_timeout_interruptible(HZ * wait_time) != 0) {
308 ext4_warning(sb, "MMP startup interrupted, failing mount\n");
312 retval = read_mmp_block(sb, &bh, mmp_block);
315 mmp = (struct mmp_struct *)(bh->b_data);
316 if (seq != le32_to_cpu(mmp->mmp_seq)) {
317 dump_mmp_msg(sb, mmp,
318 "Device is already active on another node.");
322 mmpd_data = kmalloc(sizeof(struct mmpd_data), GFP_KERNEL);
324 ext4_warning(sb, "not enough memory for mmpd_data");
331 * Start a kernel thread to update the MMP block periodically.
333 EXT4_SB(sb)->s_mmp_tsk = kthread_run(kmmpd, mmpd_data, "kmmpd-%s",
336 if (IS_ERR(EXT4_SB(sb)->s_mmp_tsk)) {
337 EXT4_SB(sb)->s_mmp_tsk = NULL;
339 ext4_warning(sb, "Unable to create kmmpd thread for %s.",