xref: /openbmc/linux/fs/ext4/mmp.c (revision 996d5b4d)
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/fs.h>
3 #include <linux/random.h>
4 #include <linux/buffer_head.h>
5 #include <linux/utsname.h>
6 #include <linux/kthread.h>
7 
8 #include "ext4.h"
9 
10 /* Checksumming functions */
11 static __le32 ext4_mmp_csum(struct super_block *sb, struct mmp_struct *mmp)
12 {
13 	struct ext4_sb_info *sbi = EXT4_SB(sb);
14 	int offset = offsetof(struct mmp_struct, mmp_checksum);
15 	__u32 csum;
16 
17 	csum = ext4_chksum(sbi, sbi->s_csum_seed, (char *)mmp, offset);
18 
19 	return cpu_to_le32(csum);
20 }
21 
22 static int ext4_mmp_csum_verify(struct super_block *sb, struct mmp_struct *mmp)
23 {
24 	if (!ext4_has_metadata_csum(sb))
25 		return 1;
26 
27 	return mmp->mmp_checksum == ext4_mmp_csum(sb, mmp);
28 }
29 
30 static void ext4_mmp_csum_set(struct super_block *sb, struct mmp_struct *mmp)
31 {
32 	if (!ext4_has_metadata_csum(sb))
33 		return;
34 
35 	mmp->mmp_checksum = ext4_mmp_csum(sb, mmp);
36 }
37 
38 /*
39  * Write the MMP block using REQ_SYNC to try to get the block on-disk
40  * faster.
41  */
42 static int write_mmp_block(struct super_block *sb, struct buffer_head *bh)
43 {
44 	struct mmp_struct *mmp = (struct mmp_struct *)(bh->b_data);
45 
46 	/*
47 	 * We protect against freezing so that we don't create dirty buffers
48 	 * on frozen filesystem.
49 	 */
50 	sb_start_write(sb);
51 	ext4_mmp_csum_set(sb, mmp);
52 	mark_buffer_dirty(bh);
53 	lock_buffer(bh);
54 	bh->b_end_io = end_buffer_write_sync;
55 	get_bh(bh);
56 	submit_bh(REQ_OP_WRITE, REQ_SYNC | REQ_META | REQ_PRIO, bh);
57 	wait_on_buffer(bh);
58 	sb_end_write(sb);
59 	if (unlikely(!buffer_uptodate(bh)))
60 		return 1;
61 
62 	return 0;
63 }
64 
65 /*
66  * Read the MMP block. It _must_ be read from disk and hence we clear the
67  * uptodate flag on the buffer.
68  */
69 static int read_mmp_block(struct super_block *sb, struct buffer_head **bh,
70 			  ext4_fsblk_t mmp_block)
71 {
72 	struct mmp_struct *mmp;
73 	int ret;
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 			ret = -ENOMEM;
85 			goto warn_exit;
86 		}
87 	}
88 
89 	get_bh(*bh);
90 	lock_buffer(*bh);
91 	(*bh)->b_end_io = end_buffer_read_sync;
92 	submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, *bh);
93 	wait_on_buffer(*bh);
94 	if (!buffer_uptodate(*bh)) {
95 		ret = -EIO;
96 		goto warn_exit;
97 	}
98 	mmp = (struct mmp_struct *)((*bh)->b_data);
99 	if (le32_to_cpu(mmp->mmp_magic) != EXT4_MMP_MAGIC) {
100 		ret = -EFSCORRUPTED;
101 		goto warn_exit;
102 	}
103 	if (!ext4_mmp_csum_verify(sb, mmp)) {
104 		ret = -EFSBADCRC;
105 		goto warn_exit;
106 	}
107 	return 0;
108 warn_exit:
109 	brelse(*bh);
110 	*bh = NULL;
111 	ext4_warning(sb, "Error %d while reading MMP block %llu",
112 		     ret, mmp_block);
113 	return ret;
114 }
115 
116 /*
117  * Dump as much information as possible to help the admin.
118  */
119 void __dump_mmp_msg(struct super_block *sb, struct mmp_struct *mmp,
120 		    const char *function, unsigned int line, const char *msg)
121 {
122 	__ext4_warning(sb, function, line, "%s", msg);
123 	__ext4_warning(sb, function, line,
124 		       "MMP failure info: last update time: %llu, last update "
125 		       "node: %s, last update device: %s",
126 		       (long long unsigned int) le64_to_cpu(mmp->mmp_time),
127 		       mmp->mmp_nodename, mmp->mmp_bdevname);
128 }
129 
130 /*
131  * kmmpd will update the MMP sequence every s_mmp_update_interval seconds
132  */
133 static int kmmpd(void *data)
134 {
135 	struct super_block *sb = ((struct mmpd_data *) data)->sb;
136 	struct buffer_head *bh = ((struct mmpd_data *) data)->bh;
137 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
138 	struct mmp_struct *mmp;
139 	ext4_fsblk_t mmp_block;
140 	u32 seq = 0;
141 	unsigned long failed_writes = 0;
142 	int mmp_update_interval = le16_to_cpu(es->s_mmp_update_interval);
143 	unsigned mmp_check_interval;
144 	unsigned long last_update_time;
145 	unsigned long diff;
146 	int retval;
147 
148 	mmp_block = le64_to_cpu(es->s_mmp_block);
149 	mmp = (struct mmp_struct *)(bh->b_data);
150 	mmp->mmp_time = cpu_to_le64(ktime_get_real_seconds());
151 	/*
152 	 * Start with the higher mmp_check_interval and reduce it if
153 	 * the MMP block is being updated on time.
154 	 */
155 	mmp_check_interval = max(EXT4_MMP_CHECK_MULT * mmp_update_interval,
156 				 EXT4_MMP_MIN_CHECK_INTERVAL);
157 	mmp->mmp_check_interval = cpu_to_le16(mmp_check_interval);
158 	bdevname(bh->b_bdev, mmp->mmp_bdevname);
159 
160 	memcpy(mmp->mmp_nodename, init_utsname()->nodename,
161 	       sizeof(mmp->mmp_nodename));
162 
163 	while (!kthread_should_stop()) {
164 		if (++seq > EXT4_MMP_SEQ_MAX)
165 			seq = 1;
166 
167 		mmp->mmp_seq = cpu_to_le32(seq);
168 		mmp->mmp_time = cpu_to_le64(ktime_get_real_seconds());
169 		last_update_time = jiffies;
170 
171 		retval = write_mmp_block(sb, bh);
172 		/*
173 		 * Don't spew too many error messages. Print one every
174 		 * (s_mmp_update_interval * 60) seconds.
175 		 */
176 		if (retval) {
177 			if ((failed_writes % 60) == 0)
178 				ext4_error(sb, "Error writing to MMP block");
179 			failed_writes++;
180 		}
181 
182 		if (!(le32_to_cpu(es->s_feature_incompat) &
183 		    EXT4_FEATURE_INCOMPAT_MMP)) {
184 			ext4_warning(sb, "kmmpd being stopped since MMP feature"
185 				     " has been disabled.");
186 			goto exit_thread;
187 		}
188 
189 		if (sb_rdonly(sb))
190 			break;
191 
192 		diff = jiffies - last_update_time;
193 		if (diff < mmp_update_interval * HZ)
194 			schedule_timeout_interruptible(mmp_update_interval *
195 						       HZ - diff);
196 
197 		/*
198 		 * We need to make sure that more than mmp_check_interval
199 		 * seconds have not passed since writing. If that has happened
200 		 * we need to check if the MMP block is as we left it.
201 		 */
202 		diff = jiffies - last_update_time;
203 		if (diff > mmp_check_interval * HZ) {
204 			struct buffer_head *bh_check = NULL;
205 			struct mmp_struct *mmp_check;
206 
207 			retval = read_mmp_block(sb, &bh_check, mmp_block);
208 			if (retval) {
209 				ext4_error(sb, "error reading MMP data: %d",
210 					   retval);
211 				goto exit_thread;
212 			}
213 
214 			mmp_check = (struct mmp_struct *)(bh_check->b_data);
215 			if (mmp->mmp_seq != mmp_check->mmp_seq ||
216 			    memcmp(mmp->mmp_nodename, mmp_check->mmp_nodename,
217 				   sizeof(mmp->mmp_nodename))) {
218 				dump_mmp_msg(sb, mmp_check,
219 					     "Error while updating MMP info. "
220 					     "The filesystem seems to have been"
221 					     " multiply mounted.");
222 				ext4_error(sb, "abort");
223 				put_bh(bh_check);
224 				retval = -EBUSY;
225 				goto exit_thread;
226 			}
227 			put_bh(bh_check);
228 		}
229 
230 		 /*
231 		 * Adjust the mmp_check_interval depending on how much time
232 		 * it took for the MMP block to be written.
233 		 */
234 		mmp_check_interval = max(min(EXT4_MMP_CHECK_MULT * diff / HZ,
235 					     EXT4_MMP_MAX_CHECK_INTERVAL),
236 					 EXT4_MMP_MIN_CHECK_INTERVAL);
237 		mmp->mmp_check_interval = cpu_to_le16(mmp_check_interval);
238 	}
239 
240 	/*
241 	 * Unmount seems to be clean.
242 	 */
243 	mmp->mmp_seq = cpu_to_le32(EXT4_MMP_SEQ_CLEAN);
244 	mmp->mmp_time = cpu_to_le64(ktime_get_real_seconds());
245 
246 	retval = write_mmp_block(sb, bh);
247 
248 exit_thread:
249 	EXT4_SB(sb)->s_mmp_tsk = NULL;
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");
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(*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