xref: /linux/fs/ext4/mmp.c (revision 26b0d14106954ae46d2f4f7eec3481828a210f7d)
1 #include <linux/fs.h>
2 #include <linux/random.h>
3 #include <linux/buffer_head.h>
4 #include <linux/utsname.h>
5 #include <linux/kthread.h>
6 
7 #include "ext4.h"
8 
9 /* Checksumming functions */
10 static __u32 ext4_mmp_csum(struct super_block *sb, struct mmp_struct *mmp)
11 {
12 	struct ext4_sb_info *sbi = EXT4_SB(sb);
13 	int offset = offsetof(struct mmp_struct, mmp_checksum);
14 	__u32 csum;
15 
16 	csum = ext4_chksum(sbi, sbi->s_csum_seed, (char *)mmp, offset);
17 
18 	return cpu_to_le32(csum);
19 }
20 
21 int ext4_mmp_csum_verify(struct super_block *sb, struct mmp_struct *mmp)
22 {
23 	if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
24 				       EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
25 		return 1;
26 
27 	return mmp->mmp_checksum == ext4_mmp_csum(sb, mmp);
28 }
29 
30 void ext4_mmp_csum_set(struct super_block *sb, struct mmp_struct *mmp)
31 {
32 	if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
33 				       EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
34 		return;
35 
36 	mmp->mmp_checksum = ext4_mmp_csum(sb, mmp);
37 }
38 
39 /*
40  * Write the MMP block using WRITE_SYNC to try to get the block on-disk
41  * faster.
42  */
43 static int write_mmp_block(struct super_block *sb, struct buffer_head *bh)
44 {
45 	struct mmp_struct *mmp = (struct mmp_struct *)(bh->b_data);
46 
47 	ext4_mmp_csum_set(sb, mmp);
48 	mark_buffer_dirty(bh);
49 	lock_buffer(bh);
50 	bh->b_end_io = end_buffer_write_sync;
51 	get_bh(bh);
52 	submit_bh(WRITE_SYNC, bh);
53 	wait_on_buffer(bh);
54 	if (unlikely(!buffer_uptodate(bh)))
55 		return 1;
56 
57 	return 0;
58 }
59 
60 /*
61  * Read the MMP block. It _must_ be read from disk and hence we clear the
62  * uptodate flag on the buffer.
63  */
64 static int read_mmp_block(struct super_block *sb, struct buffer_head **bh,
65 			  ext4_fsblk_t mmp_block)
66 {
67 	struct mmp_struct *mmp;
68 
69 	if (*bh)
70 		clear_buffer_uptodate(*bh);
71 
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. */
75 	if (!*bh)
76 		*bh = sb_getblk(sb, mmp_block);
77 	if (*bh) {
78 		get_bh(*bh);
79 		lock_buffer(*bh);
80 		(*bh)->b_end_io = end_buffer_read_sync;
81 		submit_bh(READ_SYNC, *bh);
82 		wait_on_buffer(*bh);
83 		if (!buffer_uptodate(*bh)) {
84 			brelse(*bh);
85 			*bh = NULL;
86 		}
87 	}
88 	if (!*bh) {
89 		ext4_warning(sb, "Error while reading MMP block %llu",
90 			     mmp_block);
91 		return -EIO;
92 	}
93 
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))
97 		return -EINVAL;
98 
99 	return 0;
100 }
101 
102 /*
103  * Dump as much information as possible to help the admin.
104  */
105 void __dump_mmp_msg(struct super_block *sb, struct mmp_struct *mmp,
106 		    const char *function, unsigned int line, const char *msg)
107 {
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);
114 }
115 
116 /*
117  * kmmpd will update the MMP sequence every s_mmp_update_interval seconds
118  */
119 static int kmmpd(void *data)
120 {
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;
126 	u32 seq = 0;
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;
131 	unsigned long diff;
132 	int retval;
133 
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());
137 	/*
138 	 * Start with the higher mmp_check_interval and reduce it if
139 	 * the MMP block is being updated on time.
140 	 */
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);
145 
146 	memcpy(mmp->mmp_nodename, init_utsname()->nodename,
147 	       sizeof(mmp->mmp_nodename));
148 
149 	while (!kthread_should_stop()) {
150 		if (++seq > EXT4_MMP_SEQ_MAX)
151 			seq = 1;
152 
153 		mmp->mmp_seq = cpu_to_le32(seq);
154 		mmp->mmp_time = cpu_to_le64(get_seconds());
155 		last_update_time = jiffies;
156 
157 		retval = write_mmp_block(sb, bh);
158 		/*
159 		 * Don't spew too many error messages. Print one every
160 		 * (s_mmp_update_interval * 60) seconds.
161 		 */
162 		if (retval) {
163 			if ((failed_writes % 60) == 0)
164 				ext4_error(sb, "Error writing to MMP block");
165 			failed_writes++;
166 		}
167 
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;
173 			goto failed;
174 		}
175 
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;
180 			goto failed;
181 		}
182 
183 		diff = jiffies - last_update_time;
184 		if (diff < mmp_update_interval * HZ)
185 			schedule_timeout_interruptible(mmp_update_interval *
186 						       HZ - diff);
187 
188 		/*
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.
192 		 */
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;
197 
198 			retval = read_mmp_block(sb, &bh_check, mmp_block);
199 			if (retval) {
200 				ext4_error(sb, "error reading MMP data: %d",
201 					   retval);
202 
203 				EXT4_SB(sb)->s_mmp_tsk = NULL;
204 				goto failed;
205 			}
206 
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");
216 				goto failed;
217 			}
218 			put_bh(bh_check);
219 		}
220 
221 		 /*
222 		 * Adjust the mmp_check_interval depending on how much time
223 		 * it took for the MMP block to be written.
224 		 */
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);
229 	}
230 
231 	/*
232 	 * Unmount seems to be clean.
233 	 */
234 	mmp->mmp_seq = cpu_to_le32(EXT4_MMP_SEQ_CLEAN);
235 	mmp->mmp_time = cpu_to_le64(get_seconds());
236 
237 	retval = write_mmp_block(sb, bh);
238 
239 failed:
240 	kfree(data);
241 	brelse(bh);
242 	return retval;
243 }
244 
245 /*
246  * Get a random new sequence number but make sure it is not greater than
247  * EXT4_MMP_SEQ_MAX.
248  */
249 static unsigned int mmp_new_seq(void)
250 {
251 	u32 new_seq;
252 
253 	do {
254 		get_random_bytes(&new_seq, sizeof(u32));
255 	} while (new_seq > EXT4_MMP_SEQ_MAX);
256 
257 	return new_seq;
258 }
259 
260 /*
261  * Protect the filesystem from being mounted more than once.
262  */
263 int ext4_multi_mount_protect(struct super_block *sb,
264 				    ext4_fsblk_t mmp_block)
265 {
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;
270 	u32 seq;
271 	unsigned int mmp_check_interval = le16_to_cpu(es->s_mmp_update_interval);
272 	unsigned int wait_time = 0;
273 	int retval;
274 
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");
278 		goto failed;
279 	}
280 
281 	retval = read_mmp_block(sb, &bh, mmp_block);
282 	if (retval)
283 		goto failed;
284 
285 	mmp = (struct mmp_struct *)(bh->b_data);
286 
287 	if (mmp_check_interval < EXT4_MMP_MIN_CHECK_INTERVAL)
288 		mmp_check_interval = EXT4_MMP_MIN_CHECK_INTERVAL;
289 
290 	/*
291 	 * If check_interval in MMP block is larger, use that instead of
292 	 * update_interval from the superblock.
293 	 */
294 	if (le16_to_cpu(mmp->mmp_check_interval) > mmp_check_interval)
295 		mmp_check_interval = le16_to_cpu(mmp->mmp_check_interval);
296 
297 	seq = le32_to_cpu(mmp->mmp_seq);
298 	if (seq == EXT4_MMP_SEQ_CLEAN)
299 		goto skip;
300 
301 	if (seq == EXT4_MMP_SEQ_FSCK) {
302 		dump_mmp_msg(sb, mmp, "fsck is running on the filesystem");
303 		goto failed;
304 	}
305 
306 	wait_time = min(mmp_check_interval * 2 + 1,
307 			mmp_check_interval + 60);
308 
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);
313 
314 	if (schedule_timeout_interruptible(HZ * wait_time) != 0) {
315 		ext4_warning(sb, "MMP startup interrupted, failing mount\n");
316 		goto failed;
317 	}
318 
319 	retval = read_mmp_block(sb, &bh, mmp_block);
320 	if (retval)
321 		goto failed;
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.");
326 		goto failed;
327 	}
328 
329 skip:
330 	/*
331 	 * write a new random sequence number.
332 	 */
333 	seq = mmp_new_seq();
334 	mmp->mmp_seq = cpu_to_le32(seq);
335 
336 	retval = write_mmp_block(sb, bh);
337 	if (retval)
338 		goto failed;
339 
340 	/*
341 	 * wait for MMP interval and check mmp_seq.
342 	 */
343 	if (schedule_timeout_interruptible(HZ * wait_time) != 0) {
344 		ext4_warning(sb, "MMP startup interrupted, failing mount\n");
345 		goto failed;
346 	}
347 
348 	retval = read_mmp_block(sb, &bh, mmp_block);
349 	if (retval)
350 		goto failed;
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.");
355 		goto failed;
356 	}
357 
358 	mmpd_data = kmalloc(sizeof(struct mmpd_data), GFP_KERNEL);
359 	if (!mmpd_data) {
360 		ext4_warning(sb, "not enough memory for mmpd_data");
361 		goto failed;
362 	}
363 	mmpd_data->sb = sb;
364 	mmpd_data->bh = bh;
365 
366 	/*
367 	 * Start a kernel thread to update the MMP block periodically.
368 	 */
369 	EXT4_SB(sb)->s_mmp_tsk = kthread_run(kmmpd, mmpd_data, "kmmpd-%s",
370 					     bdevname(bh->b_bdev,
371 						      mmp->mmp_bdevname));
372 	if (IS_ERR(EXT4_SB(sb)->s_mmp_tsk)) {
373 		EXT4_SB(sb)->s_mmp_tsk = NULL;
374 		kfree(mmpd_data);
375 		ext4_warning(sb, "Unable to create kmmpd thread for %s.",
376 			     sb->s_id);
377 		goto failed;
378 	}
379 
380 	return 0;
381 
382 failed:
383 	brelse(bh);
384 	return 1;
385 }
386 
387 
388