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 __le32 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 | REQ_META | REQ_PRIO, 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);
88 (*bh)->b_end_io = end_buffer_read_sync;
89 submit_bh(READ_SYNC | REQ_META | REQ_PRIO, *bh);
91 if (!buffer_uptodate(*bh)) {
97 ext4_warning(sb, "Error while reading MMP block %llu",
102 mmp = (struct mmp_struct *)((*bh)->b_data);
103 if (le32_to_cpu(mmp->mmp_magic) != EXT4_MMP_MAGIC ||
104 !ext4_mmp_csum_verify(sb, mmp))
111 * Dump as much information as possible to help the admin.
113 void __dump_mmp_msg(struct super_block *sb, struct mmp_struct *mmp,
114 const char *function, unsigned int line, const char *msg)
116 __ext4_warning(sb, function, line, msg);
117 __ext4_warning(sb, function, line,
118 "MMP failure info: last update time: %llu, last update "
119 "node: %s, last update device: %s\n",
120 (long long unsigned int) le64_to_cpu(mmp->mmp_time),
121 mmp->mmp_nodename, mmp->mmp_bdevname);
125 * kmmpd will update the MMP sequence every s_mmp_update_interval seconds
127 static int kmmpd(void *data)
129 struct super_block *sb = ((struct mmpd_data *) data)->sb;
130 struct buffer_head *bh = ((struct mmpd_data *) data)->bh;
131 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
132 struct mmp_struct *mmp;
133 ext4_fsblk_t mmp_block;
135 unsigned long failed_writes = 0;
136 int mmp_update_interval = le16_to_cpu(es->s_mmp_update_interval);
137 unsigned mmp_check_interval;
138 unsigned long last_update_time;
142 mmp_block = le64_to_cpu(es->s_mmp_block);
143 mmp = (struct mmp_struct *)(bh->b_data);
144 mmp->mmp_time = cpu_to_le64(get_seconds());
146 * Start with the higher mmp_check_interval and reduce it if
147 * the MMP block is being updated on time.
149 mmp_check_interval = max(EXT4_MMP_CHECK_MULT * mmp_update_interval,
150 EXT4_MMP_MIN_CHECK_INTERVAL);
151 mmp->mmp_check_interval = cpu_to_le16(mmp_check_interval);
152 bdevname(bh->b_bdev, mmp->mmp_bdevname);
154 memcpy(mmp->mmp_nodename, init_utsname()->nodename,
155 sizeof(mmp->mmp_nodename));
157 while (!kthread_should_stop()) {
158 if (++seq > EXT4_MMP_SEQ_MAX)
161 mmp->mmp_seq = cpu_to_le32(seq);
162 mmp->mmp_time = cpu_to_le64(get_seconds());
163 last_update_time = jiffies;
165 retval = write_mmp_block(sb, bh);
167 * Don't spew too many error messages. Print one every
168 * (s_mmp_update_interval * 60) seconds.
171 if ((failed_writes % 60) == 0)
172 ext4_error(sb, "Error writing to MMP block");
176 if (!(le32_to_cpu(es->s_feature_incompat) &
177 EXT4_FEATURE_INCOMPAT_MMP)) {
178 ext4_warning(sb, "kmmpd being stopped since MMP feature"
179 " has been disabled.");
180 EXT4_SB(sb)->s_mmp_tsk = NULL;
184 if (sb->s_flags & MS_RDONLY) {
185 ext4_warning(sb, "kmmpd being stopped since filesystem "
186 "has been remounted as readonly.");
187 EXT4_SB(sb)->s_mmp_tsk = NULL;
191 diff = jiffies - last_update_time;
192 if (diff < mmp_update_interval * HZ)
193 schedule_timeout_interruptible(mmp_update_interval *
197 * We need to make sure that more than mmp_check_interval
198 * seconds have not passed since writing. If that has happened
199 * we need to check if the MMP block is as we left it.
201 diff = jiffies - last_update_time;
202 if (diff > mmp_check_interval * HZ) {
203 struct buffer_head *bh_check = NULL;
204 struct mmp_struct *mmp_check;
206 retval = read_mmp_block(sb, &bh_check, mmp_block);
208 ext4_error(sb, "error reading MMP data: %d",
211 EXT4_SB(sb)->s_mmp_tsk = NULL;
215 mmp_check = (struct mmp_struct *)(bh_check->b_data);
216 if (mmp->mmp_seq != mmp_check->mmp_seq ||
217 memcmp(mmp->mmp_nodename, mmp_check->mmp_nodename,
218 sizeof(mmp->mmp_nodename))) {
219 dump_mmp_msg(sb, mmp_check,
220 "Error while updating MMP info. "
221 "The filesystem seems to have been"
222 " multiply mounted.");
223 ext4_error(sb, "abort");
230 * Adjust the mmp_check_interval depending on how much time
231 * it took for the MMP block to be written.
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);
240 * Unmount seems to be clean.
242 mmp->mmp_seq = cpu_to_le32(EXT4_MMP_SEQ_CLEAN);
243 mmp->mmp_time = cpu_to_le64(get_seconds());
245 retval = write_mmp_block(sb, bh);
254 * Get a random new sequence number but make sure it is not greater than
257 static unsigned int mmp_new_seq(void)
262 get_random_bytes(&new_seq, sizeof(u32));
263 } while (new_seq > EXT4_MMP_SEQ_MAX);
269 * Protect the filesystem from being mounted more than once.
271 int ext4_multi_mount_protect(struct super_block *sb,
272 ext4_fsblk_t mmp_block)
274 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
275 struct buffer_head *bh = NULL;
276 struct mmp_struct *mmp = NULL;
277 struct mmpd_data *mmpd_data;
279 unsigned int mmp_check_interval = le16_to_cpu(es->s_mmp_update_interval);
280 unsigned int wait_time = 0;
283 if (mmp_block < le32_to_cpu(es->s_first_data_block) ||
284 mmp_block >= ext4_blocks_count(es)) {
285 ext4_warning(sb, "Invalid MMP block in superblock");
289 retval = read_mmp_block(sb, &bh, mmp_block);
293 mmp = (struct mmp_struct *)(bh->b_data);
295 if (mmp_check_interval < EXT4_MMP_MIN_CHECK_INTERVAL)
296 mmp_check_interval = EXT4_MMP_MIN_CHECK_INTERVAL;
299 * If check_interval in MMP block is larger, use that instead of
300 * update_interval from the superblock.
302 if (le16_to_cpu(mmp->mmp_check_interval) > mmp_check_interval)
303 mmp_check_interval = le16_to_cpu(mmp->mmp_check_interval);
305 seq = le32_to_cpu(mmp->mmp_seq);
306 if (seq == EXT4_MMP_SEQ_CLEAN)
309 if (seq == EXT4_MMP_SEQ_FSCK) {
310 dump_mmp_msg(sb, mmp, "fsck is running on the filesystem");
314 wait_time = min(mmp_check_interval * 2 + 1,
315 mmp_check_interval + 60);
317 /* Print MMP interval if more than 20 secs. */
318 if (wait_time > EXT4_MMP_MIN_CHECK_INTERVAL * 4)
319 ext4_warning(sb, "MMP interval %u higher than expected, please"
320 " wait.\n", wait_time * 2);
322 if (schedule_timeout_interruptible(HZ * wait_time) != 0) {
323 ext4_warning(sb, "MMP startup interrupted, failing mount\n");
327 retval = read_mmp_block(sb, &bh, mmp_block);
330 mmp = (struct mmp_struct *)(bh->b_data);
331 if (seq != le32_to_cpu(mmp->mmp_seq)) {
332 dump_mmp_msg(sb, mmp,
333 "Device is already active on another node.");
339 * write a new random sequence number.
342 mmp->mmp_seq = cpu_to_le32(seq);
344 retval = write_mmp_block(sb, bh);
349 * wait for MMP interval and check mmp_seq.
351 if (schedule_timeout_interruptible(HZ * wait_time) != 0) {
352 ext4_warning(sb, "MMP startup interrupted, failing mount\n");
356 retval = read_mmp_block(sb, &bh, mmp_block);
359 mmp = (struct mmp_struct *)(bh->b_data);
360 if (seq != le32_to_cpu(mmp->mmp_seq)) {
361 dump_mmp_msg(sb, mmp,
362 "Device is already active on another node.");
366 mmpd_data = kmalloc(sizeof(struct mmpd_data), GFP_KERNEL);
368 ext4_warning(sb, "not enough memory for mmpd_data");
375 * Start a kernel thread to update the MMP block periodically.
377 EXT4_SB(sb)->s_mmp_tsk = kthread_run(kmmpd, mmpd_data, "kmmpd-%s",
380 if (IS_ERR(EXT4_SB(sb)->s_mmp_tsk)) {
381 EXT4_SB(sb)->s_mmp_tsk = NULL;
383 ext4_warning(sb, "Unable to create kmmpd thread for %s.",