xref: /openbmc/linux/fs/ext4/mmp.c (revision 483eb062)
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 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 	/*
48 	 * We protect against freezing so that we don't create dirty buffers
49 	 * on frozen filesystem.
50 	 */
51 	sb_start_write(sb);
52 	ext4_mmp_csum_set(sb, mmp);
53 	mark_buffer_dirty(bh);
54 	lock_buffer(bh);
55 	bh->b_end_io = end_buffer_write_sync;
56 	get_bh(bh);
57 	submit_bh(WRITE_SYNC | REQ_META | REQ_PRIO, bh);
58 	wait_on_buffer(bh);
59 	sb_end_write(sb);
60 	if (unlikely(!buffer_uptodate(bh)))
61 		return 1;
62 
63 	return 0;
64 }
65 
66 /*
67  * Read the MMP block. It _must_ be read from disk and hence we clear the
68  * uptodate flag on the buffer.
69  */
70 static int read_mmp_block(struct super_block *sb, struct buffer_head **bh,
71 			  ext4_fsblk_t mmp_block)
72 {
73 	struct mmp_struct *mmp;
74 
75 	if (*bh)
76 		clear_buffer_uptodate(*bh);
77 
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. */
81 	if (!*bh)
82 		*bh = sb_getblk(sb, mmp_block);
83 	if (!*bh)
84 		return -ENOMEM;
85 	if (*bh) {
86 		get_bh(*bh);
87 		lock_buffer(*bh);
88 		(*bh)->b_end_io = end_buffer_read_sync;
89 		submit_bh(READ_SYNC | REQ_META | REQ_PRIO, *bh);
90 		wait_on_buffer(*bh);
91 		if (!buffer_uptodate(*bh)) {
92 			brelse(*bh);
93 			*bh = NULL;
94 		}
95 	}
96 	if (unlikely(!*bh)) {
97 		ext4_warning(sb, "Error while reading MMP block %llu",
98 			     mmp_block);
99 		return -EIO;
100 	}
101 
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))
105 		return -EINVAL;
106 
107 	return 0;
108 }
109 
110 /*
111  * Dump as much information as possible to help the admin.
112  */
113 void __dump_mmp_msg(struct super_block *sb, struct mmp_struct *mmp,
114 		    const char *function, unsigned int line, const char *msg)
115 {
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);
122 }
123 
124 /*
125  * kmmpd will update the MMP sequence every s_mmp_update_interval seconds
126  */
127 static int kmmpd(void *data)
128 {
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;
134 	u32 seq = 0;
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;
139 	unsigned long diff;
140 	int retval;
141 
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());
145 	/*
146 	 * Start with the higher mmp_check_interval and reduce it if
147 	 * the MMP block is being updated on time.
148 	 */
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);
153 
154 	memcpy(mmp->mmp_nodename, init_utsname()->nodename,
155 	       sizeof(mmp->mmp_nodename));
156 
157 	while (!kthread_should_stop()) {
158 		if (++seq > EXT4_MMP_SEQ_MAX)
159 			seq = 1;
160 
161 		mmp->mmp_seq = cpu_to_le32(seq);
162 		mmp->mmp_time = cpu_to_le64(get_seconds());
163 		last_update_time = jiffies;
164 
165 		retval = write_mmp_block(sb, bh);
166 		/*
167 		 * Don't spew too many error messages. Print one every
168 		 * (s_mmp_update_interval * 60) seconds.
169 		 */
170 		if (retval) {
171 			if ((failed_writes % 60) == 0)
172 				ext4_error(sb, "Error writing to MMP block");
173 			failed_writes++;
174 		}
175 
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;
181 			goto failed;
182 		}
183 
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;
188 			goto failed;
189 		}
190 
191 		diff = jiffies - last_update_time;
192 		if (diff < mmp_update_interval * HZ)
193 			schedule_timeout_interruptible(mmp_update_interval *
194 						       HZ - diff);
195 
196 		/*
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.
200 		 */
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;
205 
206 			retval = read_mmp_block(sb, &bh_check, mmp_block);
207 			if (retval) {
208 				ext4_error(sb, "error reading MMP data: %d",
209 					   retval);
210 
211 				EXT4_SB(sb)->s_mmp_tsk = NULL;
212 				goto failed;
213 			}
214 
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");
224 				goto failed;
225 			}
226 			put_bh(bh_check);
227 		}
228 
229 		 /*
230 		 * Adjust the mmp_check_interval depending on how much time
231 		 * it took for the MMP block to be written.
232 		 */
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);
237 	}
238 
239 	/*
240 	 * Unmount seems to be clean.
241 	 */
242 	mmp->mmp_seq = cpu_to_le32(EXT4_MMP_SEQ_CLEAN);
243 	mmp->mmp_time = cpu_to_le64(get_seconds());
244 
245 	retval = write_mmp_block(sb, bh);
246 
247 failed:
248 	kfree(data);
249 	brelse(bh);
250 	return retval;
251 }
252 
253 /*
254  * Get a random new sequence number but make sure it is not greater than
255  * EXT4_MMP_SEQ_MAX.
256  */
257 static unsigned int mmp_new_seq(void)
258 {
259 	u32 new_seq;
260 
261 	do {
262 		new_seq = prandom_u32();
263 	} while (new_seq > EXT4_MMP_SEQ_MAX);
264 
265 	return new_seq;
266 }
267 
268 /*
269  * Protect the filesystem from being mounted more than once.
270  */
271 int ext4_multi_mount_protect(struct super_block *sb,
272 				    ext4_fsblk_t mmp_block)
273 {
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;
278 	u32 seq;
279 	unsigned int mmp_check_interval = le16_to_cpu(es->s_mmp_update_interval);
280 	unsigned int wait_time = 0;
281 	int retval;
282 
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");
286 		goto failed;
287 	}
288 
289 	retval = read_mmp_block(sb, &bh, mmp_block);
290 	if (retval)
291 		goto failed;
292 
293 	mmp = (struct mmp_struct *)(bh->b_data);
294 
295 	if (mmp_check_interval < EXT4_MMP_MIN_CHECK_INTERVAL)
296 		mmp_check_interval = EXT4_MMP_MIN_CHECK_INTERVAL;
297 
298 	/*
299 	 * If check_interval in MMP block is larger, use that instead of
300 	 * update_interval from the superblock.
301 	 */
302 	if (le16_to_cpu(mmp->mmp_check_interval) > mmp_check_interval)
303 		mmp_check_interval = le16_to_cpu(mmp->mmp_check_interval);
304 
305 	seq = le32_to_cpu(mmp->mmp_seq);
306 	if (seq == EXT4_MMP_SEQ_CLEAN)
307 		goto skip;
308 
309 	if (seq == EXT4_MMP_SEQ_FSCK) {
310 		dump_mmp_msg(sb, mmp, "fsck is running on the filesystem");
311 		goto failed;
312 	}
313 
314 	wait_time = min(mmp_check_interval * 2 + 1,
315 			mmp_check_interval + 60);
316 
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);
321 
322 	if (schedule_timeout_interruptible(HZ * wait_time) != 0) {
323 		ext4_warning(sb, "MMP startup interrupted, failing mount\n");
324 		goto failed;
325 	}
326 
327 	retval = read_mmp_block(sb, &bh, mmp_block);
328 	if (retval)
329 		goto failed;
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.");
334 		goto failed;
335 	}
336 
337 skip:
338 	/*
339 	 * write a new random sequence number.
340 	 */
341 	seq = mmp_new_seq();
342 	mmp->mmp_seq = cpu_to_le32(seq);
343 
344 	retval = write_mmp_block(sb, bh);
345 	if (retval)
346 		goto failed;
347 
348 	/*
349 	 * wait for MMP interval and check mmp_seq.
350 	 */
351 	if (schedule_timeout_interruptible(HZ * wait_time) != 0) {
352 		ext4_warning(sb, "MMP startup interrupted, failing mount\n");
353 		goto failed;
354 	}
355 
356 	retval = read_mmp_block(sb, &bh, mmp_block);
357 	if (retval)
358 		goto failed;
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.");
363 		goto failed;
364 	}
365 
366 	mmpd_data = kmalloc(sizeof(struct mmpd_data), GFP_KERNEL);
367 	if (!mmpd_data) {
368 		ext4_warning(sb, "not enough memory for mmpd_data");
369 		goto failed;
370 	}
371 	mmpd_data->sb = sb;
372 	mmpd_data->bh = bh;
373 
374 	/*
375 	 * Start a kernel thread to update the MMP block periodically.
376 	 */
377 	EXT4_SB(sb)->s_mmp_tsk = kthread_run(kmmpd, mmpd_data, "kmmpd-%s",
378 					     bdevname(bh->b_bdev,
379 						      mmp->mmp_bdevname));
380 	if (IS_ERR(EXT4_SB(sb)->s_mmp_tsk)) {
381 		EXT4_SB(sb)->s_mmp_tsk = NULL;
382 		kfree(mmpd_data);
383 		ext4_warning(sb, "Unable to create kmmpd thread for %s.",
384 			     sb->s_id);
385 		goto failed;
386 	}
387 
388 	return 0;
389 
390 failed:
391 	brelse(bh);
392 	return 1;
393 }
394 
395 
396