xref: /openbmc/linux/fs/ext4/mmp.c (revision dc023681)
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 	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 -EIO;
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 	lock_buffer(*bh);
89 	ret = ext4_read_bh(*bh, REQ_META | REQ_PRIO, NULL);
90 	if (ret)
91 		goto warn_exit;
92 
93 	mmp = (struct mmp_struct *)((*bh)->b_data);
94 	if (le32_to_cpu(mmp->mmp_magic) != EXT4_MMP_MAGIC) {
95 		ret = -EFSCORRUPTED;
96 		goto warn_exit;
97 	}
98 	if (!ext4_mmp_csum_verify(sb, mmp)) {
99 		ret = -EFSBADCRC;
100 		goto warn_exit;
101 	}
102 	return 0;
103 warn_exit:
104 	brelse(*bh);
105 	*bh = NULL;
106 	ext4_warning(sb, "Error %d while reading MMP block %llu",
107 		     ret, mmp_block);
108 	return ret;
109 }
110 
111 /*
112  * Dump as much information as possible to help the admin.
113  */
114 void __dump_mmp_msg(struct super_block *sb, struct mmp_struct *mmp,
115 		    const char *function, unsigned int line, const char *msg)
116 {
117 	__ext4_warning(sb, function, line, "%s", msg);
118 	__ext4_warning(sb, function, line,
119 		       "MMP failure info: last update time: %llu, last update node: %.*s, last update device: %.*s",
120 		       (unsigned long long)le64_to_cpu(mmp->mmp_time),
121 		       (int)sizeof(mmp->mmp_nodename), mmp->mmp_nodename,
122 		       (int)sizeof(mmp->mmp_bdevname), mmp->mmp_bdevname);
123 }
124 
125 /*
126  * kmmpd will update the MMP sequence every s_mmp_update_interval seconds
127  */
128 static int kmmpd(void *data)
129 {
130 	struct super_block *sb = (struct super_block *) data;
131 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
132 	struct buffer_head *bh = EXT4_SB(sb)->s_mmp_bh;
133 	struct mmp_struct *mmp;
134 	ext4_fsblk_t mmp_block;
135 	u32 seq = 0;
136 	unsigned long failed_writes = 0;
137 	int mmp_update_interval = le16_to_cpu(es->s_mmp_update_interval);
138 	unsigned mmp_check_interval;
139 	unsigned long last_update_time;
140 	unsigned long diff;
141 	int retval = 0;
142 
143 	mmp_block = le64_to_cpu(es->s_mmp_block);
144 	mmp = (struct mmp_struct *)(bh->b_data);
145 	mmp->mmp_time = cpu_to_le64(ktime_get_real_seconds());
146 	/*
147 	 * Start with the higher mmp_check_interval and reduce it if
148 	 * the MMP block is being updated on time.
149 	 */
150 	mmp_check_interval = max(EXT4_MMP_CHECK_MULT * mmp_update_interval,
151 				 EXT4_MMP_MIN_CHECK_INTERVAL);
152 	mmp->mmp_check_interval = cpu_to_le16(mmp_check_interval);
153 	BUILD_BUG_ON(sizeof(mmp->mmp_bdevname) < BDEVNAME_SIZE);
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() && !sb_rdonly(sb)) {
160 		if (!ext4_has_feature_mmp(sb)) {
161 			ext4_warning(sb, "kmmpd being stopped since MMP feature"
162 				     " has been disabled.");
163 			goto wait_to_exit;
164 		}
165 		if (++seq > EXT4_MMP_SEQ_MAX)
166 			seq = 1;
167 
168 		mmp->mmp_seq = cpu_to_le32(seq);
169 		mmp->mmp_time = cpu_to_le64(ktime_get_real_seconds());
170 		last_update_time = jiffies;
171 
172 		retval = write_mmp_block(sb, bh);
173 		/*
174 		 * Don't spew too many error messages. Print one every
175 		 * (s_mmp_update_interval * 60) seconds.
176 		 */
177 		if (retval) {
178 			if ((failed_writes % 60) == 0) {
179 				ext4_error_err(sb, -retval,
180 					       "Error writing to MMP block");
181 			}
182 			failed_writes++;
183 		}
184 
185 		diff = jiffies - last_update_time;
186 		if (diff < mmp_update_interval * HZ)
187 			schedule_timeout_interruptible(mmp_update_interval *
188 						       HZ - diff);
189 
190 		/*
191 		 * We need to make sure that more than mmp_check_interval
192 		 * seconds have not passed since writing. If that has happened
193 		 * we need to check if the MMP block is as we left it.
194 		 */
195 		diff = jiffies - last_update_time;
196 		if (diff > mmp_check_interval * HZ) {
197 			struct buffer_head *bh_check = NULL;
198 			struct mmp_struct *mmp_check;
199 
200 			retval = read_mmp_block(sb, &bh_check, mmp_block);
201 			if (retval) {
202 				ext4_error_err(sb, -retval,
203 					       "error reading MMP data: %d",
204 					       retval);
205 				goto wait_to_exit;
206 			}
207 
208 			mmp_check = (struct mmp_struct *)(bh_check->b_data);
209 			if (mmp->mmp_seq != mmp_check->mmp_seq ||
210 			    memcmp(mmp->mmp_nodename, mmp_check->mmp_nodename,
211 				   sizeof(mmp->mmp_nodename))) {
212 				dump_mmp_msg(sb, mmp_check,
213 					     "Error while updating MMP info. "
214 					     "The filesystem seems to have been"
215 					     " multiply mounted.");
216 				ext4_error_err(sb, EBUSY, "abort");
217 				put_bh(bh_check);
218 				retval = -EBUSY;
219 				goto wait_to_exit;
220 			}
221 			put_bh(bh_check);
222 		}
223 
224 		 /*
225 		 * Adjust the mmp_check_interval depending on how much time
226 		 * it took for the MMP block to be written.
227 		 */
228 		mmp_check_interval = max(min(EXT4_MMP_CHECK_MULT * diff / HZ,
229 					     EXT4_MMP_MAX_CHECK_INTERVAL),
230 					 EXT4_MMP_MIN_CHECK_INTERVAL);
231 		mmp->mmp_check_interval = cpu_to_le16(mmp_check_interval);
232 	}
233 
234 	/*
235 	 * Unmount seems to be clean.
236 	 */
237 	mmp->mmp_seq = cpu_to_le32(EXT4_MMP_SEQ_CLEAN);
238 	mmp->mmp_time = cpu_to_le64(ktime_get_real_seconds());
239 
240 	retval = write_mmp_block(sb, bh);
241 
242 wait_to_exit:
243 	while (!kthread_should_stop()) {
244 		set_current_state(TASK_INTERRUPTIBLE);
245 		if (!kthread_should_stop())
246 			schedule();
247 	}
248 	set_current_state(TASK_RUNNING);
249 	return retval;
250 }
251 
252 void ext4_stop_mmpd(struct ext4_sb_info *sbi)
253 {
254 	if (sbi->s_mmp_tsk) {
255 		kthread_stop(sbi->s_mmp_tsk);
256 		brelse(sbi->s_mmp_bh);
257 		sbi->s_mmp_tsk = NULL;
258 	}
259 }
260 
261 /*
262  * Get a random new sequence number but make sure it is not greater than
263  * EXT4_MMP_SEQ_MAX.
264  */
265 static unsigned int mmp_new_seq(void)
266 {
267 	u32 new_seq;
268 
269 	do {
270 		new_seq = prandom_u32();
271 	} while (new_seq > EXT4_MMP_SEQ_MAX);
272 
273 	return new_seq;
274 }
275 
276 /*
277  * Protect the filesystem from being mounted more than once.
278  */
279 int ext4_multi_mount_protect(struct super_block *sb,
280 				    ext4_fsblk_t mmp_block)
281 {
282 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
283 	struct buffer_head *bh = NULL;
284 	struct mmp_struct *mmp = NULL;
285 	u32 seq;
286 	unsigned int mmp_check_interval = le16_to_cpu(es->s_mmp_update_interval);
287 	unsigned int wait_time = 0;
288 	int retval;
289 
290 	if (mmp_block < le32_to_cpu(es->s_first_data_block) ||
291 	    mmp_block >= ext4_blocks_count(es)) {
292 		ext4_warning(sb, "Invalid MMP block in superblock");
293 		goto failed;
294 	}
295 
296 	retval = read_mmp_block(sb, &bh, mmp_block);
297 	if (retval)
298 		goto failed;
299 
300 	mmp = (struct mmp_struct *)(bh->b_data);
301 
302 	if (mmp_check_interval < EXT4_MMP_MIN_CHECK_INTERVAL)
303 		mmp_check_interval = EXT4_MMP_MIN_CHECK_INTERVAL;
304 
305 	/*
306 	 * If check_interval in MMP block is larger, use that instead of
307 	 * update_interval from the superblock.
308 	 */
309 	if (le16_to_cpu(mmp->mmp_check_interval) > mmp_check_interval)
310 		mmp_check_interval = le16_to_cpu(mmp->mmp_check_interval);
311 
312 	seq = le32_to_cpu(mmp->mmp_seq);
313 	if (seq == EXT4_MMP_SEQ_CLEAN)
314 		goto skip;
315 
316 	if (seq == EXT4_MMP_SEQ_FSCK) {
317 		dump_mmp_msg(sb, mmp, "fsck is running on the filesystem");
318 		goto failed;
319 	}
320 
321 	wait_time = min(mmp_check_interval * 2 + 1,
322 			mmp_check_interval + 60);
323 
324 	/* Print MMP interval if more than 20 secs. */
325 	if (wait_time > EXT4_MMP_MIN_CHECK_INTERVAL * 4)
326 		ext4_warning(sb, "MMP interval %u higher than expected, please"
327 			     " wait.\n", wait_time * 2);
328 
329 	if (schedule_timeout_interruptible(HZ * wait_time) != 0) {
330 		ext4_warning(sb, "MMP startup interrupted, failing mount\n");
331 		goto failed;
332 	}
333 
334 	retval = read_mmp_block(sb, &bh, mmp_block);
335 	if (retval)
336 		goto failed;
337 	mmp = (struct mmp_struct *)(bh->b_data);
338 	if (seq != le32_to_cpu(mmp->mmp_seq)) {
339 		dump_mmp_msg(sb, mmp,
340 			     "Device is already active on another node.");
341 		goto failed;
342 	}
343 
344 skip:
345 	/*
346 	 * write a new random sequence number.
347 	 */
348 	seq = mmp_new_seq();
349 	mmp->mmp_seq = cpu_to_le32(seq);
350 
351 	retval = write_mmp_block(sb, bh);
352 	if (retval)
353 		goto failed;
354 
355 	/*
356 	 * wait for MMP interval and check mmp_seq.
357 	 */
358 	if (schedule_timeout_interruptible(HZ * wait_time) != 0) {
359 		ext4_warning(sb, "MMP startup interrupted, failing mount");
360 		goto failed;
361 	}
362 
363 	retval = read_mmp_block(sb, &bh, mmp_block);
364 	if (retval)
365 		goto failed;
366 	mmp = (struct mmp_struct *)(bh->b_data);
367 	if (seq != le32_to_cpu(mmp->mmp_seq)) {
368 		dump_mmp_msg(sb, mmp,
369 			     "Device is already active on another node.");
370 		goto failed;
371 	}
372 
373 	EXT4_SB(sb)->s_mmp_bh = bh;
374 
375 	/*
376 	 * Start a kernel thread to update the MMP block periodically.
377 	 */
378 	EXT4_SB(sb)->s_mmp_tsk = kthread_run(kmmpd, sb, "kmmpd-%.*s",
379 					     (int)sizeof(mmp->mmp_bdevname),
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 		ext4_warning(sb, "Unable to create kmmpd thread for %s.",
385 			     sb->s_id);
386 		goto failed;
387 	}
388 
389 	return 0;
390 
391 failed:
392 	brelse(bh);
393 	return 1;
394 }
395