xref: /openbmc/linux/fs/ext4/mmp.c (revision b96fc2f3)
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 WRITE_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(WRITE_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(READ_SYNC | REQ_META | REQ_PRIO, *bh);
92 	wait_on_buffer(*bh);
93 	if (!buffer_uptodate(*bh)) {
94 		brelse(*bh);
95 		*bh = NULL;
96 		ret = -EIO;
97 		goto warn_exit;
98 	}
99 
100 	mmp = (struct mmp_struct *)((*bh)->b_data);
101 	if (le32_to_cpu(mmp->mmp_magic) == EXT4_MMP_MAGIC &&
102 	    ext4_mmp_csum_verify(sb, mmp))
103 		return 0;
104 	ret = -EINVAL;
105 
106 warn_exit:
107 	ext4_warning(sb, "Error %d while reading MMP block %llu",
108 		     ret, mmp_block);
109 	return ret;
110 }
111 
112 /*
113  * Dump as much information as possible to help the admin.
114  */
115 void __dump_mmp_msg(struct super_block *sb, struct mmp_struct *mmp,
116 		    const char *function, unsigned int line, const char *msg)
117 {
118 	__ext4_warning(sb, function, line, "%s", msg);
119 	__ext4_warning(sb, function, line,
120 		       "MMP failure info: last update time: %llu, last update "
121 		       "node: %s, last update device: %s\n",
122 		       (long long unsigned int) le64_to_cpu(mmp->mmp_time),
123 		       mmp->mmp_nodename, mmp->mmp_bdevname);
124 }
125 
126 /*
127  * kmmpd will update the MMP sequence every s_mmp_update_interval seconds
128  */
129 static int kmmpd(void *data)
130 {
131 	struct super_block *sb = ((struct mmpd_data *) data)->sb;
132 	struct buffer_head *bh = ((struct mmpd_data *) data)->bh;
133 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
134 	struct mmp_struct *mmp;
135 	ext4_fsblk_t mmp_block;
136 	u32 seq = 0;
137 	unsigned long failed_writes = 0;
138 	int mmp_update_interval = le16_to_cpu(es->s_mmp_update_interval);
139 	unsigned mmp_check_interval;
140 	unsigned long last_update_time;
141 	unsigned long diff;
142 	int retval;
143 
144 	mmp_block = le64_to_cpu(es->s_mmp_block);
145 	mmp = (struct mmp_struct *)(bh->b_data);
146 	mmp->mmp_time = cpu_to_le64(get_seconds());
147 	/*
148 	 * Start with the higher mmp_check_interval and reduce it if
149 	 * the MMP block is being updated on time.
150 	 */
151 	mmp_check_interval = max(EXT4_MMP_CHECK_MULT * mmp_update_interval,
152 				 EXT4_MMP_MIN_CHECK_INTERVAL);
153 	mmp->mmp_check_interval = cpu_to_le16(mmp_check_interval);
154 	bdevname(bh->b_bdev, mmp->mmp_bdevname);
155 
156 	memcpy(mmp->mmp_nodename, init_utsname()->nodename,
157 	       sizeof(mmp->mmp_nodename));
158 
159 	while (!kthread_should_stop()) {
160 		if (++seq > EXT4_MMP_SEQ_MAX)
161 			seq = 1;
162 
163 		mmp->mmp_seq = cpu_to_le32(seq);
164 		mmp->mmp_time = cpu_to_le64(get_seconds());
165 		last_update_time = jiffies;
166 
167 		retval = write_mmp_block(sb, bh);
168 		/*
169 		 * Don't spew too many error messages. Print one every
170 		 * (s_mmp_update_interval * 60) seconds.
171 		 */
172 		if (retval) {
173 			if ((failed_writes % 60) == 0)
174 				ext4_error(sb, "Error writing to MMP block");
175 			failed_writes++;
176 		}
177 
178 		if (!(le32_to_cpu(es->s_feature_incompat) &
179 		    EXT4_FEATURE_INCOMPAT_MMP)) {
180 			ext4_warning(sb, "kmmpd being stopped since MMP feature"
181 				     " has been disabled.");
182 			EXT4_SB(sb)->s_mmp_tsk = NULL;
183 			goto failed;
184 		}
185 
186 		if (sb->s_flags & MS_RDONLY) {
187 			ext4_warning(sb, "kmmpd being stopped since filesystem "
188 				     "has been remounted as readonly.");
189 			EXT4_SB(sb)->s_mmp_tsk = NULL;
190 			goto failed;
191 		}
192 
193 		diff = jiffies - last_update_time;
194 		if (diff < mmp_update_interval * HZ)
195 			schedule_timeout_interruptible(mmp_update_interval *
196 						       HZ - diff);
197 
198 		/*
199 		 * We need to make sure that more than mmp_check_interval
200 		 * seconds have not passed since writing. If that has happened
201 		 * we need to check if the MMP block is as we left it.
202 		 */
203 		diff = jiffies - last_update_time;
204 		if (diff > mmp_check_interval * HZ) {
205 			struct buffer_head *bh_check = NULL;
206 			struct mmp_struct *mmp_check;
207 
208 			retval = read_mmp_block(sb, &bh_check, mmp_block);
209 			if (retval) {
210 				ext4_error(sb, "error reading MMP data: %d",
211 					   retval);
212 
213 				EXT4_SB(sb)->s_mmp_tsk = NULL;
214 				goto failed;
215 			}
216 
217 			mmp_check = (struct mmp_struct *)(bh_check->b_data);
218 			if (mmp->mmp_seq != mmp_check->mmp_seq ||
219 			    memcmp(mmp->mmp_nodename, mmp_check->mmp_nodename,
220 				   sizeof(mmp->mmp_nodename))) {
221 				dump_mmp_msg(sb, mmp_check,
222 					     "Error while updating MMP info. "
223 					     "The filesystem seems to have been"
224 					     " multiply mounted.");
225 				ext4_error(sb, "abort");
226 				goto failed;
227 			}
228 			put_bh(bh_check);
229 		}
230 
231 		 /*
232 		 * Adjust the mmp_check_interval depending on how much time
233 		 * it took for the MMP block to be written.
234 		 */
235 		mmp_check_interval = max(min(EXT4_MMP_CHECK_MULT * diff / HZ,
236 					     EXT4_MMP_MAX_CHECK_INTERVAL),
237 					 EXT4_MMP_MIN_CHECK_INTERVAL);
238 		mmp->mmp_check_interval = cpu_to_le16(mmp_check_interval);
239 	}
240 
241 	/*
242 	 * Unmount seems to be clean.
243 	 */
244 	mmp->mmp_seq = cpu_to_le32(EXT4_MMP_SEQ_CLEAN);
245 	mmp->mmp_time = cpu_to_le64(get_seconds());
246 
247 	retval = write_mmp_block(sb, bh);
248 
249 failed:
250 	kfree(data);
251 	brelse(bh);
252 	return retval;
253 }
254 
255 /*
256  * Get a random new sequence number but make sure it is not greater than
257  * EXT4_MMP_SEQ_MAX.
258  */
259 static unsigned int mmp_new_seq(void)
260 {
261 	u32 new_seq;
262 
263 	do {
264 		new_seq = prandom_u32();
265 	} while (new_seq > EXT4_MMP_SEQ_MAX);
266 
267 	return new_seq;
268 }
269 
270 /*
271  * Protect the filesystem from being mounted more than once.
272  */
273 int ext4_multi_mount_protect(struct super_block *sb,
274 				    ext4_fsblk_t mmp_block)
275 {
276 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
277 	struct buffer_head *bh = NULL;
278 	struct mmp_struct *mmp = NULL;
279 	struct mmpd_data *mmpd_data;
280 	u32 seq;
281 	unsigned int mmp_check_interval = le16_to_cpu(es->s_mmp_update_interval);
282 	unsigned int wait_time = 0;
283 	int retval;
284 
285 	if (mmp_block < le32_to_cpu(es->s_first_data_block) ||
286 	    mmp_block >= ext4_blocks_count(es)) {
287 		ext4_warning(sb, "Invalid MMP block in superblock");
288 		goto failed;
289 	}
290 
291 	retval = read_mmp_block(sb, &bh, mmp_block);
292 	if (retval)
293 		goto failed;
294 
295 	mmp = (struct mmp_struct *)(bh->b_data);
296 
297 	if (mmp_check_interval < EXT4_MMP_MIN_CHECK_INTERVAL)
298 		mmp_check_interval = EXT4_MMP_MIN_CHECK_INTERVAL;
299 
300 	/*
301 	 * If check_interval in MMP block is larger, use that instead of
302 	 * update_interval from the superblock.
303 	 */
304 	if (le16_to_cpu(mmp->mmp_check_interval) > mmp_check_interval)
305 		mmp_check_interval = le16_to_cpu(mmp->mmp_check_interval);
306 
307 	seq = le32_to_cpu(mmp->mmp_seq);
308 	if (seq == EXT4_MMP_SEQ_CLEAN)
309 		goto skip;
310 
311 	if (seq == EXT4_MMP_SEQ_FSCK) {
312 		dump_mmp_msg(sb, mmp, "fsck is running on the filesystem");
313 		goto failed;
314 	}
315 
316 	wait_time = min(mmp_check_interval * 2 + 1,
317 			mmp_check_interval + 60);
318 
319 	/* Print MMP interval if more than 20 secs. */
320 	if (wait_time > EXT4_MMP_MIN_CHECK_INTERVAL * 4)
321 		ext4_warning(sb, "MMP interval %u higher than expected, please"
322 			     " wait.\n", wait_time * 2);
323 
324 	if (schedule_timeout_interruptible(HZ * wait_time) != 0) {
325 		ext4_warning(sb, "MMP startup interrupted, failing mount\n");
326 		goto failed;
327 	}
328 
329 	retval = read_mmp_block(sb, &bh, mmp_block);
330 	if (retval)
331 		goto failed;
332 	mmp = (struct mmp_struct *)(bh->b_data);
333 	if (seq != le32_to_cpu(mmp->mmp_seq)) {
334 		dump_mmp_msg(sb, mmp,
335 			     "Device is already active on another node.");
336 		goto failed;
337 	}
338 
339 skip:
340 	/*
341 	 * write a new random sequence number.
342 	 */
343 	seq = mmp_new_seq();
344 	mmp->mmp_seq = cpu_to_le32(seq);
345 
346 	retval = write_mmp_block(sb, bh);
347 	if (retval)
348 		goto failed;
349 
350 	/*
351 	 * wait for MMP interval and check mmp_seq.
352 	 */
353 	if (schedule_timeout_interruptible(HZ * wait_time) != 0) {
354 		ext4_warning(sb, "MMP startup interrupted, failing mount\n");
355 		goto failed;
356 	}
357 
358 	retval = read_mmp_block(sb, &bh, mmp_block);
359 	if (retval)
360 		goto failed;
361 	mmp = (struct mmp_struct *)(bh->b_data);
362 	if (seq != le32_to_cpu(mmp->mmp_seq)) {
363 		dump_mmp_msg(sb, mmp,
364 			     "Device is already active on another node.");
365 		goto failed;
366 	}
367 
368 	mmpd_data = kmalloc(sizeof(struct mmpd_data), GFP_KERNEL);
369 	if (!mmpd_data) {
370 		ext4_warning(sb, "not enough memory for mmpd_data");
371 		goto failed;
372 	}
373 	mmpd_data->sb = sb;
374 	mmpd_data->bh = bh;
375 
376 	/*
377 	 * Start a kernel thread to update the MMP block periodically.
378 	 */
379 	EXT4_SB(sb)->s_mmp_tsk = kthread_run(kmmpd, mmpd_data, "kmmpd-%s",
380 					     bdevname(bh->b_bdev,
381 						      mmp->mmp_bdevname));
382 	if (IS_ERR(EXT4_SB(sb)->s_mmp_tsk)) {
383 		EXT4_SB(sb)->s_mmp_tsk = NULL;
384 		kfree(mmpd_data);
385 		ext4_warning(sb, "Unable to create kmmpd thread for %s.",
386 			     sb->s_id);
387 		goto failed;
388 	}
389 
390 	return 0;
391 
392 failed:
393 	brelse(bh);
394 	return 1;
395 }
396 
397 
398