xref: /linux/fs/ext4/mmp.c (revision 74ce1896c6c65b2f8cccbf59162d542988835835)
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 __le32 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 static int ext4_mmp_csum_verify(struct super_block *sb, struct mmp_struct *mmp)
22 {
23 	if (!ext4_has_metadata_csum(sb))
24 		return 1;
25 
26 	return mmp->mmp_checksum == ext4_mmp_csum(sb, mmp);
27 }
28 
29 static void ext4_mmp_csum_set(struct super_block *sb, struct mmp_struct *mmp)
30 {
31 	if (!ext4_has_metadata_csum(sb))
32 		return;
33 
34 	mmp->mmp_checksum = ext4_mmp_csum(sb, mmp);
35 }
36 
37 /*
38  * Write the MMP block using REQ_SYNC to try to get the block on-disk
39  * faster.
40  */
41 static int write_mmp_block(struct super_block *sb, struct buffer_head *bh)
42 {
43 	struct mmp_struct *mmp = (struct mmp_struct *)(bh->b_data);
44 
45 	/*
46 	 * We protect against freezing so that we don't create dirty buffers
47 	 * on frozen filesystem.
48 	 */
49 	sb_start_write(sb);
50 	ext4_mmp_csum_set(sb, mmp);
51 	mark_buffer_dirty(bh);
52 	lock_buffer(bh);
53 	bh->b_end_io = end_buffer_write_sync;
54 	get_bh(bh);
55 	submit_bh(REQ_OP_WRITE, REQ_SYNC | REQ_META | REQ_PRIO, bh);
56 	wait_on_buffer(bh);
57 	sb_end_write(sb);
58 	if (unlikely(!buffer_uptodate(bh)))
59 		return 1;
60 
61 	return 0;
62 }
63 
64 /*
65  * Read the MMP block. It _must_ be read from disk and hence we clear the
66  * uptodate flag on the buffer.
67  */
68 static int read_mmp_block(struct super_block *sb, struct buffer_head **bh,
69 			  ext4_fsblk_t mmp_block)
70 {
71 	struct mmp_struct *mmp;
72 	int ret;
73 
74 	if (*bh)
75 		clear_buffer_uptodate(*bh);
76 
77 	/* This would be sb_bread(sb, mmp_block), except we need to be sure
78 	 * that the MD RAID device cache has been bypassed, and that the read
79 	 * is not blocked in the elevator. */
80 	if (!*bh) {
81 		*bh = sb_getblk(sb, mmp_block);
82 		if (!*bh) {
83 			ret = -ENOMEM;
84 			goto warn_exit;
85 		}
86 	}
87 
88 	get_bh(*bh);
89 	lock_buffer(*bh);
90 	(*bh)->b_end_io = end_buffer_read_sync;
91 	submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, *bh);
92 	wait_on_buffer(*bh);
93 	if (!buffer_uptodate(*bh)) {
94 		ret = -EIO;
95 		goto warn_exit;
96 	}
97 	mmp = (struct mmp_struct *)((*bh)->b_data);
98 	if (le32_to_cpu(mmp->mmp_magic) != EXT4_MMP_MAGIC) {
99 		ret = -EFSCORRUPTED;
100 		goto warn_exit;
101 	}
102 	if (!ext4_mmp_csum_verify(sb, mmp)) {
103 		ret = -EFSBADCRC;
104 		goto warn_exit;
105 	}
106 	return 0;
107 warn_exit:
108 	brelse(*bh);
109 	*bh = NULL;
110 	ext4_warning(sb, "Error %d while reading MMP block %llu",
111 		     ret, mmp_block);
112 	return ret;
113 }
114 
115 /*
116  * Dump as much information as possible to help the admin.
117  */
118 void __dump_mmp_msg(struct super_block *sb, struct mmp_struct *mmp,
119 		    const char *function, unsigned int line, const char *msg)
120 {
121 	__ext4_warning(sb, function, line, "%s", msg);
122 	__ext4_warning(sb, function, line,
123 		       "MMP failure info: last update time: %llu, last update "
124 		       "node: %s, last update device: %s",
125 		       (long long unsigned int) le64_to_cpu(mmp->mmp_time),
126 		       mmp->mmp_nodename, mmp->mmp_bdevname);
127 }
128 
129 /*
130  * kmmpd will update the MMP sequence every s_mmp_update_interval seconds
131  */
132 static int kmmpd(void *data)
133 {
134 	struct super_block *sb = ((struct mmpd_data *) data)->sb;
135 	struct buffer_head *bh = ((struct mmpd_data *) data)->bh;
136 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
137 	struct mmp_struct *mmp;
138 	ext4_fsblk_t mmp_block;
139 	u32 seq = 0;
140 	unsigned long failed_writes = 0;
141 	int mmp_update_interval = le16_to_cpu(es->s_mmp_update_interval);
142 	unsigned mmp_check_interval;
143 	unsigned long last_update_time;
144 	unsigned long diff;
145 	int retval;
146 
147 	mmp_block = le64_to_cpu(es->s_mmp_block);
148 	mmp = (struct mmp_struct *)(bh->b_data);
149 	mmp->mmp_time = cpu_to_le64(get_seconds());
150 	/*
151 	 * Start with the higher mmp_check_interval and reduce it if
152 	 * the MMP block is being updated on time.
153 	 */
154 	mmp_check_interval = max(EXT4_MMP_CHECK_MULT * mmp_update_interval,
155 				 EXT4_MMP_MIN_CHECK_INTERVAL);
156 	mmp->mmp_check_interval = cpu_to_le16(mmp_check_interval);
157 	bdevname(bh->b_bdev, mmp->mmp_bdevname);
158 
159 	memcpy(mmp->mmp_nodename, init_utsname()->nodename,
160 	       sizeof(mmp->mmp_nodename));
161 
162 	while (!kthread_should_stop()) {
163 		if (++seq > EXT4_MMP_SEQ_MAX)
164 			seq = 1;
165 
166 		mmp->mmp_seq = cpu_to_le32(seq);
167 		mmp->mmp_time = cpu_to_le64(get_seconds());
168 		last_update_time = jiffies;
169 
170 		retval = write_mmp_block(sb, bh);
171 		/*
172 		 * Don't spew too many error messages. Print one every
173 		 * (s_mmp_update_interval * 60) seconds.
174 		 */
175 		if (retval) {
176 			if ((failed_writes % 60) == 0)
177 				ext4_error(sb, "Error writing to MMP block");
178 			failed_writes++;
179 		}
180 
181 		if (!(le32_to_cpu(es->s_feature_incompat) &
182 		    EXT4_FEATURE_INCOMPAT_MMP)) {
183 			ext4_warning(sb, "kmmpd being stopped since MMP feature"
184 				     " has been disabled.");
185 			goto exit_thread;
186 		}
187 
188 		if (sb_rdonly(sb)) {
189 			ext4_warning(sb, "kmmpd being stopped since filesystem "
190 				     "has been remounted as readonly.");
191 			goto exit_thread;
192 		}
193 
194 		diff = jiffies - last_update_time;
195 		if (diff < mmp_update_interval * HZ)
196 			schedule_timeout_interruptible(mmp_update_interval *
197 						       HZ - diff);
198 
199 		/*
200 		 * We need to make sure that more than mmp_check_interval
201 		 * seconds have not passed since writing. If that has happened
202 		 * we need to check if the MMP block is as we left it.
203 		 */
204 		diff = jiffies - last_update_time;
205 		if (diff > mmp_check_interval * HZ) {
206 			struct buffer_head *bh_check = NULL;
207 			struct mmp_struct *mmp_check;
208 
209 			retval = read_mmp_block(sb, &bh_check, mmp_block);
210 			if (retval) {
211 				ext4_error(sb, "error reading MMP data: %d",
212 					   retval);
213 				goto exit_thread;
214 			}
215 
216 			mmp_check = (struct mmp_struct *)(bh_check->b_data);
217 			if (mmp->mmp_seq != mmp_check->mmp_seq ||
218 			    memcmp(mmp->mmp_nodename, mmp_check->mmp_nodename,
219 				   sizeof(mmp->mmp_nodename))) {
220 				dump_mmp_msg(sb, mmp_check,
221 					     "Error while updating MMP info. "
222 					     "The filesystem seems to have been"
223 					     " multiply mounted.");
224 				ext4_error(sb, "abort");
225 				put_bh(bh_check);
226 				retval = -EBUSY;
227 				goto exit_thread;
228 			}
229 			put_bh(bh_check);
230 		}
231 
232 		 /*
233 		 * Adjust the mmp_check_interval depending on how much time
234 		 * it took for the MMP block to be written.
235 		 */
236 		mmp_check_interval = max(min(EXT4_MMP_CHECK_MULT * diff / HZ,
237 					     EXT4_MMP_MAX_CHECK_INTERVAL),
238 					 EXT4_MMP_MIN_CHECK_INTERVAL);
239 		mmp->mmp_check_interval = cpu_to_le16(mmp_check_interval);
240 	}
241 
242 	/*
243 	 * Unmount seems to be clean.
244 	 */
245 	mmp->mmp_seq = cpu_to_le32(EXT4_MMP_SEQ_CLEAN);
246 	mmp->mmp_time = cpu_to_le64(get_seconds());
247 
248 	retval = write_mmp_block(sb, bh);
249 
250 exit_thread:
251 	EXT4_SB(sb)->s_mmp_tsk = NULL;
252 	kfree(data);
253 	brelse(bh);
254 	return retval;
255 }
256 
257 /*
258  * Get a random new sequence number but make sure it is not greater than
259  * EXT4_MMP_SEQ_MAX.
260  */
261 static unsigned int mmp_new_seq(void)
262 {
263 	u32 new_seq;
264 
265 	do {
266 		new_seq = prandom_u32();
267 	} while (new_seq > EXT4_MMP_SEQ_MAX);
268 
269 	return new_seq;
270 }
271 
272 /*
273  * Protect the filesystem from being mounted more than once.
274  */
275 int ext4_multi_mount_protect(struct super_block *sb,
276 				    ext4_fsblk_t mmp_block)
277 {
278 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
279 	struct buffer_head *bh = NULL;
280 	struct mmp_struct *mmp = NULL;
281 	struct mmpd_data *mmpd_data;
282 	u32 seq;
283 	unsigned int mmp_check_interval = le16_to_cpu(es->s_mmp_update_interval);
284 	unsigned int wait_time = 0;
285 	int retval;
286 
287 	if (mmp_block < le32_to_cpu(es->s_first_data_block) ||
288 	    mmp_block >= ext4_blocks_count(es)) {
289 		ext4_warning(sb, "Invalid MMP block in superblock");
290 		goto failed;
291 	}
292 
293 	retval = read_mmp_block(sb, &bh, mmp_block);
294 	if (retval)
295 		goto failed;
296 
297 	mmp = (struct mmp_struct *)(bh->b_data);
298 
299 	if (mmp_check_interval < EXT4_MMP_MIN_CHECK_INTERVAL)
300 		mmp_check_interval = EXT4_MMP_MIN_CHECK_INTERVAL;
301 
302 	/*
303 	 * If check_interval in MMP block is larger, use that instead of
304 	 * update_interval from the superblock.
305 	 */
306 	if (le16_to_cpu(mmp->mmp_check_interval) > mmp_check_interval)
307 		mmp_check_interval = le16_to_cpu(mmp->mmp_check_interval);
308 
309 	seq = le32_to_cpu(mmp->mmp_seq);
310 	if (seq == EXT4_MMP_SEQ_CLEAN)
311 		goto skip;
312 
313 	if (seq == EXT4_MMP_SEQ_FSCK) {
314 		dump_mmp_msg(sb, mmp, "fsck is running on the filesystem");
315 		goto failed;
316 	}
317 
318 	wait_time = min(mmp_check_interval * 2 + 1,
319 			mmp_check_interval + 60);
320 
321 	/* Print MMP interval if more than 20 secs. */
322 	if (wait_time > EXT4_MMP_MIN_CHECK_INTERVAL * 4)
323 		ext4_warning(sb, "MMP interval %u higher than expected, please"
324 			     " wait.\n", wait_time * 2);
325 
326 	if (schedule_timeout_interruptible(HZ * wait_time) != 0) {
327 		ext4_warning(sb, "MMP startup interrupted, failing mount\n");
328 		goto failed;
329 	}
330 
331 	retval = read_mmp_block(sb, &bh, mmp_block);
332 	if (retval)
333 		goto failed;
334 	mmp = (struct mmp_struct *)(bh->b_data);
335 	if (seq != le32_to_cpu(mmp->mmp_seq)) {
336 		dump_mmp_msg(sb, mmp,
337 			     "Device is already active on another node.");
338 		goto failed;
339 	}
340 
341 skip:
342 	/*
343 	 * write a new random sequence number.
344 	 */
345 	seq = mmp_new_seq();
346 	mmp->mmp_seq = cpu_to_le32(seq);
347 
348 	retval = write_mmp_block(sb, bh);
349 	if (retval)
350 		goto failed;
351 
352 	/*
353 	 * wait for MMP interval and check mmp_seq.
354 	 */
355 	if (schedule_timeout_interruptible(HZ * wait_time) != 0) {
356 		ext4_warning(sb, "MMP startup interrupted, failing mount");
357 		goto failed;
358 	}
359 
360 	retval = read_mmp_block(sb, &bh, mmp_block);
361 	if (retval)
362 		goto failed;
363 	mmp = (struct mmp_struct *)(bh->b_data);
364 	if (seq != le32_to_cpu(mmp->mmp_seq)) {
365 		dump_mmp_msg(sb, mmp,
366 			     "Device is already active on another node.");
367 		goto failed;
368 	}
369 
370 	mmpd_data = kmalloc(sizeof(*mmpd_data), GFP_KERNEL);
371 	if (!mmpd_data) {
372 		ext4_warning(sb, "not enough memory for mmpd_data");
373 		goto failed;
374 	}
375 	mmpd_data->sb = sb;
376 	mmpd_data->bh = bh;
377 
378 	/*
379 	 * Start a kernel thread to update the MMP block periodically.
380 	 */
381 	EXT4_SB(sb)->s_mmp_tsk = kthread_run(kmmpd, mmpd_data, "kmmpd-%s",
382 					     bdevname(bh->b_bdev,
383 						      mmp->mmp_bdevname));
384 	if (IS_ERR(EXT4_SB(sb)->s_mmp_tsk)) {
385 		EXT4_SB(sb)->s_mmp_tsk = NULL;
386 		kfree(mmpd_data);
387 		ext4_warning(sb, "Unable to create kmmpd thread for %s.",
388 			     sb->s_id);
389 		goto failed;
390 	}
391 
392 	return 0;
393 
394 failed:
395 	brelse(bh);
396 	return 1;
397 }
398 
399 
400