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);
47 ext4_mmp_csum_set(sb, mmp);
48 mark_buffer_dirty(bh);
50 bh->b_end_io = end_buffer_write_sync;
52 submit_bh(WRITE_SYNC, bh);
54 if (unlikely(!buffer_uptodate(bh)))
61 * Read the MMP block. It _must_ be read from disk and hence we clear the
62 * uptodate flag on the buffer.
64 static int read_mmp_block(struct super_block *sb, struct buffer_head **bh,
65 ext4_fsblk_t mmp_block)
67 struct mmp_struct *mmp;
70 clear_buffer_uptodate(*bh);
72 /* This would be sb_bread(sb, mmp_block), except we need to be sure
73 * that the MD RAID device cache has been bypassed, and that the read
74 * is not blocked in the elevator. */
76 *bh = sb_getblk(sb, mmp_block);
80 (*bh)->b_end_io = end_buffer_read_sync;
81 submit_bh(READ_SYNC, *bh);
83 if (!buffer_uptodate(*bh)) {
89 ext4_warning(sb, "Error while reading MMP block %llu",
94 mmp = (struct mmp_struct *)((*bh)->b_data);
95 if (le32_to_cpu(mmp->mmp_magic) != EXT4_MMP_MAGIC ||
96 !ext4_mmp_csum_verify(sb, mmp))
103 * Dump as much information as possible to help the admin.
105 void __dump_mmp_msg(struct super_block *sb, struct mmp_struct *mmp,
106 const char *function, unsigned int line, const char *msg)
108 __ext4_warning(sb, function, line, msg);
109 __ext4_warning(sb, function, line,
110 "MMP failure info: last update time: %llu, last update "
111 "node: %s, last update device: %s\n",
112 (long long unsigned int) le64_to_cpu(mmp->mmp_time),
113 mmp->mmp_nodename, mmp->mmp_bdevname);
117 * kmmpd will update the MMP sequence every s_mmp_update_interval seconds
119 static int kmmpd(void *data)
121 struct super_block *sb = ((struct mmpd_data *) data)->sb;
122 struct buffer_head *bh = ((struct mmpd_data *) data)->bh;
123 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
124 struct mmp_struct *mmp;
125 ext4_fsblk_t mmp_block;
127 unsigned long failed_writes = 0;
128 int mmp_update_interval = le16_to_cpu(es->s_mmp_update_interval);
129 unsigned mmp_check_interval;
130 unsigned long last_update_time;
134 mmp_block = le64_to_cpu(es->s_mmp_block);
135 mmp = (struct mmp_struct *)(bh->b_data);
136 mmp->mmp_time = cpu_to_le64(get_seconds());
138 * Start with the higher mmp_check_interval and reduce it if
139 * the MMP block is being updated on time.
141 mmp_check_interval = max(EXT4_MMP_CHECK_MULT * mmp_update_interval,
142 EXT4_MMP_MIN_CHECK_INTERVAL);
143 mmp->mmp_check_interval = cpu_to_le16(mmp_check_interval);
144 bdevname(bh->b_bdev, mmp->mmp_bdevname);
146 memcpy(mmp->mmp_nodename, init_utsname()->nodename,
147 sizeof(mmp->mmp_nodename));
149 while (!kthread_should_stop()) {
150 if (++seq > EXT4_MMP_SEQ_MAX)
153 mmp->mmp_seq = cpu_to_le32(seq);
154 mmp->mmp_time = cpu_to_le64(get_seconds());
155 last_update_time = jiffies;
157 retval = write_mmp_block(sb, bh);
159 * Don't spew too many error messages. Print one every
160 * (s_mmp_update_interval * 60) seconds.
163 if ((failed_writes % 60) == 0)
164 ext4_error(sb, "Error writing to MMP block");
168 if (!(le32_to_cpu(es->s_feature_incompat) &
169 EXT4_FEATURE_INCOMPAT_MMP)) {
170 ext4_warning(sb, "kmmpd being stopped since MMP feature"
171 " has been disabled.");
172 EXT4_SB(sb)->s_mmp_tsk = NULL;
176 if (sb->s_flags & MS_RDONLY) {
177 ext4_warning(sb, "kmmpd being stopped since filesystem "
178 "has been remounted as readonly.");
179 EXT4_SB(sb)->s_mmp_tsk = NULL;
183 diff = jiffies - last_update_time;
184 if (diff < mmp_update_interval * HZ)
185 schedule_timeout_interruptible(mmp_update_interval *
189 * We need to make sure that more than mmp_check_interval
190 * seconds have not passed since writing. If that has happened
191 * we need to check if the MMP block is as we left it.
193 diff = jiffies - last_update_time;
194 if (diff > mmp_check_interval * HZ) {
195 struct buffer_head *bh_check = NULL;
196 struct mmp_struct *mmp_check;
198 retval = read_mmp_block(sb, &bh_check, mmp_block);
200 ext4_error(sb, "error reading MMP data: %d",
203 EXT4_SB(sb)->s_mmp_tsk = NULL;
207 mmp_check = (struct mmp_struct *)(bh_check->b_data);
208 if (mmp->mmp_seq != mmp_check->mmp_seq ||
209 memcmp(mmp->mmp_nodename, mmp_check->mmp_nodename,
210 sizeof(mmp->mmp_nodename))) {
211 dump_mmp_msg(sb, mmp_check,
212 "Error while updating MMP info. "
213 "The filesystem seems to have been"
214 " multiply mounted.");
215 ext4_error(sb, "abort");
222 * Adjust the mmp_check_interval depending on how much time
223 * it took for the MMP block to be written.
225 mmp_check_interval = max(min(EXT4_MMP_CHECK_MULT * diff / HZ,
226 EXT4_MMP_MAX_CHECK_INTERVAL),
227 EXT4_MMP_MIN_CHECK_INTERVAL);
228 mmp->mmp_check_interval = cpu_to_le16(mmp_check_interval);
232 * Unmount seems to be clean.
234 mmp->mmp_seq = cpu_to_le32(EXT4_MMP_SEQ_CLEAN);
235 mmp->mmp_time = cpu_to_le64(get_seconds());
237 retval = write_mmp_block(sb, bh);
246 * Get a random new sequence number but make sure it is not greater than
249 static unsigned int mmp_new_seq(void)
254 get_random_bytes(&new_seq, sizeof(u32));
255 } while (new_seq > EXT4_MMP_SEQ_MAX);
261 * Protect the filesystem from being mounted more than once.
263 int ext4_multi_mount_protect(struct super_block *sb,
264 ext4_fsblk_t mmp_block)
266 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
267 struct buffer_head *bh = NULL;
268 struct mmp_struct *mmp = NULL;
269 struct mmpd_data *mmpd_data;
271 unsigned int mmp_check_interval = le16_to_cpu(es->s_mmp_update_interval);
272 unsigned int wait_time = 0;
275 if (mmp_block < le32_to_cpu(es->s_first_data_block) ||
276 mmp_block >= ext4_blocks_count(es)) {
277 ext4_warning(sb, "Invalid MMP block in superblock");
281 retval = read_mmp_block(sb, &bh, mmp_block);
285 mmp = (struct mmp_struct *)(bh->b_data);
287 if (mmp_check_interval < EXT4_MMP_MIN_CHECK_INTERVAL)
288 mmp_check_interval = EXT4_MMP_MIN_CHECK_INTERVAL;
291 * If check_interval in MMP block is larger, use that instead of
292 * update_interval from the superblock.
294 if (le16_to_cpu(mmp->mmp_check_interval) > mmp_check_interval)
295 mmp_check_interval = le16_to_cpu(mmp->mmp_check_interval);
297 seq = le32_to_cpu(mmp->mmp_seq);
298 if (seq == EXT4_MMP_SEQ_CLEAN)
301 if (seq == EXT4_MMP_SEQ_FSCK) {
302 dump_mmp_msg(sb, mmp, "fsck is running on the filesystem");
306 wait_time = min(mmp_check_interval * 2 + 1,
307 mmp_check_interval + 60);
309 /* Print MMP interval if more than 20 secs. */
310 if (wait_time > EXT4_MMP_MIN_CHECK_INTERVAL * 4)
311 ext4_warning(sb, "MMP interval %u higher than expected, please"
312 " wait.\n", wait_time * 2);
314 if (schedule_timeout_interruptible(HZ * wait_time) != 0) {
315 ext4_warning(sb, "MMP startup interrupted, failing mount\n");
319 retval = read_mmp_block(sb, &bh, mmp_block);
322 mmp = (struct mmp_struct *)(bh->b_data);
323 if (seq != le32_to_cpu(mmp->mmp_seq)) {
324 dump_mmp_msg(sb, mmp,
325 "Device is already active on another node.");
331 * write a new random sequence number.
334 mmp->mmp_seq = cpu_to_le32(seq);
336 retval = write_mmp_block(sb, bh);
341 * wait for MMP interval and check mmp_seq.
343 if (schedule_timeout_interruptible(HZ * wait_time) != 0) {
344 ext4_warning(sb, "MMP startup interrupted, failing mount\n");
348 retval = read_mmp_block(sb, &bh, mmp_block);
351 mmp = (struct mmp_struct *)(bh->b_data);
352 if (seq != le32_to_cpu(mmp->mmp_seq)) {
353 dump_mmp_msg(sb, mmp,
354 "Device is already active on another node.");
358 mmpd_data = kmalloc(sizeof(struct mmpd_data), GFP_KERNEL);
360 ext4_warning(sb, "not enough memory for mmpd_data");
367 * Start a kernel thread to update the MMP block periodically.
369 EXT4_SB(sb)->s_mmp_tsk = kthread_run(kmmpd, mmpd_data, "kmmpd-%s",
372 if (IS_ERR(EXT4_SB(sb)->s_mmp_tsk)) {
373 EXT4_SB(sb)->s_mmp_tsk = NULL;
375 ext4_warning(sb, "Unable to create kmmpd thread for %s.",