xref: /linux/fs/ext4/mmp.c (revision 056a5087d87ead77dedbe9cf5bde53b7cd4b4651)
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->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_feature_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_feature_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_thawed(struct super_block *sb,
43 				  struct buffer_head *bh)
44 {
45 	struct mmp_struct *mmp = (struct mmp_struct *)(bh->b_data);
46 
47 	ext4_mmp_csum_set(sb, mmp);
48 	lock_buffer(bh);
49 	bh_submit(bh, REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO,
50 			bh_end_write);
51 	wait_on_buffer(bh);
52 	if (unlikely(!buffer_uptodate(bh)))
53 		return -EIO;
54 	return 0;
55 }
56 
57 static int write_mmp_block(struct super_block *sb, struct buffer_head *bh)
58 {
59 	/*
60 	 * We protect against freezing so that we don't create dirty buffers
61 	 * on frozen filesystem.
62 	 */
63 	scoped_guard(super_write, sb)
64 		return write_mmp_block_thawed(sb, bh);
65 }
66 
67 /*
68  * Read the MMP block. It _must_ be read from disk and hence we clear the
69  * uptodate flag on the buffer.
70  */
71 static int read_mmp_block(struct super_block *sb, struct buffer_head **bh,
72 			  ext4_fsblk_t mmp_block)
73 {
74 	struct mmp_struct *mmp;
75 	int ret;
76 
77 	if (*bh)
78 		clear_buffer_uptodate(*bh);
79 
80 	/* This would be sb_bread(sb, mmp_block), except we need to be sure
81 	 * that the MD RAID device cache has been bypassed, and that the read
82 	 * is not blocked in the elevator. */
83 	if (!*bh) {
84 		*bh = sb_getblk(sb, mmp_block);
85 		if (!*bh) {
86 			ret = -ENOMEM;
87 			goto warn_exit;
88 		}
89 	}
90 
91 	lock_buffer(*bh);
92 	ret = ext4_read_bh(*bh, REQ_META | REQ_PRIO, NULL, false);
93 	if (ret)
94 		goto warn_exit;
95 
96 	mmp = (struct mmp_struct *)((*bh)->b_data);
97 	if (le32_to_cpu(mmp->mmp_magic) != EXT4_MMP_MAGIC) {
98 		ret = -EFSCORRUPTED;
99 		goto warn_exit;
100 	}
101 	if (!ext4_mmp_csum_verify(sb, mmp)) {
102 		ret = -EFSBADCRC;
103 		goto warn_exit;
104 	}
105 	return 0;
106 warn_exit:
107 	brelse(*bh);
108 	*bh = NULL;
109 	ext4_warning(sb, "Error %d while reading MMP block %llu",
110 		     ret, mmp_block);
111 	return ret;
112 }
113 
114 /*
115  * Dump as much information as possible to help the admin.
116  */
117 void __dump_mmp_msg(struct super_block *sb, struct mmp_struct *mmp,
118 		    const char *function, unsigned int line, const char *msg)
119 {
120 	__ext4_warning(sb, function, line, "%s", msg);
121 	__ext4_warning(sb, function, line,
122 		       "MMP failure info: last update time: %llu, last update node: %.*s, last update device: %.*s",
123 		       (unsigned long long)le64_to_cpu(mmp->mmp_time),
124 		       (int)sizeof(mmp->mmp_nodename), mmp->mmp_nodename,
125 		       (int)sizeof(mmp->mmp_bdevname), mmp->mmp_bdevname);
126 }
127 
128 /*
129  * kmmpd will update the MMP sequence every s_mmp_update_interval seconds
130  */
131 static int kmmpd(void *data)
132 {
133 	struct super_block *sb = data;
134 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
135 	struct buffer_head *bh = EXT4_SB(sb)->s_mmp_bh;
136 	struct mmp_struct *mmp;
137 	ext4_fsblk_t mmp_block;
138 	u32 seq = 0;
139 	unsigned long failed_writes = 0;
140 	int mmp_update_interval = le16_to_cpu(es->s_mmp_update_interval);
141 	unsigned mmp_check_interval;
142 	unsigned long last_update_time;
143 	unsigned long diff;
144 	int retval = 0;
145 
146 	mmp_block = le64_to_cpu(es->s_mmp_block);
147 	mmp = (struct mmp_struct *)(bh->b_data);
148 	mmp->mmp_time = cpu_to_le64(ktime_get_real_seconds());
149 	/*
150 	 * Start with the higher mmp_check_interval and reduce it if
151 	 * the MMP block is being updated on time.
152 	 */
153 	mmp_check_interval = max(EXT4_MMP_CHECK_MULT * mmp_update_interval,
154 				 EXT4_MMP_MIN_CHECK_INTERVAL);
155 	mmp->mmp_check_interval = cpu_to_le16(mmp_check_interval);
156 
157 	memcpy(mmp->mmp_nodename, init_utsname()->nodename,
158 	       sizeof(mmp->mmp_nodename));
159 
160 	while (!kthread_should_stop() && !ext4_emergency_state(sb)) {
161 		if (!ext4_has_feature_mmp(sb)) {
162 			ext4_warning(sb, "kmmpd being stopped since MMP feature"
163 				     " has been disabled.");
164 			goto wait_to_exit;
165 		}
166 		if (++seq > EXT4_MMP_SEQ_MAX)
167 			seq = 1;
168 
169 		mmp->mmp_seq = cpu_to_le32(seq);
170 		mmp->mmp_time = cpu_to_le64(ktime_get_real_seconds());
171 		last_update_time = jiffies;
172 
173 		retval = write_mmp_block(sb, bh);
174 		/*
175 		 * Don't spew too many error messages. Print one every
176 		 * (s_mmp_update_interval * 60) seconds.
177 		 */
178 		if (retval) {
179 			if ((failed_writes % 60) == 0) {
180 				ext4_error_err(sb, -retval,
181 					       "Error writing to MMP block");
182 			}
183 			failed_writes++;
184 		}
185 
186 		diff = jiffies - last_update_time;
187 		if (diff < mmp_update_interval * HZ)
188 			schedule_timeout_interruptible(mmp_update_interval *
189 						       HZ - diff);
190 
191 		/*
192 		 * We need to make sure that more than mmp_check_interval
193 		 * seconds have not passed since writing. If that has happened
194 		 * we need to check if the MMP block is as we left it.
195 		 */
196 		diff = jiffies - last_update_time;
197 		if (diff > mmp_check_interval * HZ) {
198 			struct buffer_head *bh_check = NULL;
199 			struct mmp_struct *mmp_check;
200 
201 			retval = read_mmp_block(sb, &bh_check, mmp_block);
202 			if (retval) {
203 				ext4_error_err(sb, -retval,
204 					       "error reading MMP data: %d",
205 					       retval);
206 				goto wait_to_exit;
207 			}
208 
209 			mmp_check = (struct mmp_struct *)(bh_check->b_data);
210 			if (mmp->mmp_seq != mmp_check->mmp_seq ||
211 			    memcmp(mmp->mmp_nodename, mmp_check->mmp_nodename,
212 				   sizeof(mmp->mmp_nodename))) {
213 				dump_mmp_msg(sb, mmp_check,
214 					     "Error while updating MMP info. "
215 					     "The filesystem seems to have been"
216 					     " multiply mounted.");
217 				ext4_error_err(sb, EBUSY, "abort");
218 				put_bh(bh_check);
219 				retval = -EBUSY;
220 				goto wait_to_exit;
221 			}
222 			put_bh(bh_check);
223 		}
224 
225 		 /*
226 		 * Adjust the mmp_check_interval depending on how much time
227 		 * it took for the MMP block to be written.
228 		 */
229 		mmp_check_interval = clamp(EXT4_MMP_CHECK_MULT * diff / HZ,
230 					   EXT4_MMP_MIN_CHECK_INTERVAL,
231 					   EXT4_MMP_MAX_CHECK_INTERVAL);
232 		mmp->mmp_check_interval = cpu_to_le16(mmp_check_interval);
233 	}
234 
235 	/*
236 	 * Unmount seems to be clean.
237 	 */
238 	mmp->mmp_seq = cpu_to_le32(EXT4_MMP_SEQ_CLEAN);
239 	mmp->mmp_time = cpu_to_le64(ktime_get_real_seconds());
240 
241 	retval = write_mmp_block(sb, bh);
242 
243 wait_to_exit:
244 	while (!kthread_should_stop()) {
245 		set_current_state(TASK_INTERRUPTIBLE);
246 		if (!kthread_should_stop())
247 			schedule();
248 	}
249 	set_current_state(TASK_RUNNING);
250 	return retval;
251 }
252 
253 void ext4_stop_mmpd(struct ext4_sb_info *sbi)
254 {
255 	if (sbi->s_mmp_tsk) {
256 		kthread_stop(sbi->s_mmp_tsk);
257 		brelse(sbi->s_mmp_bh);
258 		sbi->s_mmp_tsk = NULL;
259 	}
260 }
261 
262 /*
263  * Get a random new sequence number but make sure it is not greater than
264  * EXT4_MMP_SEQ_MAX.
265  */
266 static unsigned int mmp_new_seq(void)
267 {
268 	return get_random_u32_below(EXT4_MMP_SEQ_MAX + 1);
269 }
270 
271 /*
272  * Protect the filesystem from being mounted more than once.
273  */
274 int ext4_multi_mount_protect(struct super_block *sb,
275 				    ext4_fsblk_t mmp_block)
276 {
277 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
278 	struct buffer_head *bh = NULL;
279 	struct mmp_struct *mmp = NULL;
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 		retval = -EINVAL;
289 		goto failed;
290 	}
291 
292 	retval = read_mmp_block(sb, &bh, mmp_block);
293 	if (retval)
294 		goto failed;
295 
296 	mmp = (struct mmp_struct *)(bh->b_data);
297 
298 	if (mmp_check_interval < EXT4_MMP_MIN_CHECK_INTERVAL)
299 		mmp_check_interval = EXT4_MMP_MIN_CHECK_INTERVAL;
300 
301 	/*
302 	 * If check_interval in MMP block is larger, use that instead of
303 	 * update_interval from the superblock.
304 	 */
305 	if (le16_to_cpu(mmp->mmp_check_interval) > mmp_check_interval)
306 		mmp_check_interval = le16_to_cpu(mmp->mmp_check_interval);
307 
308 	seq = le32_to_cpu(mmp->mmp_seq);
309 	if (seq == EXT4_MMP_SEQ_CLEAN)
310 		goto skip;
311 
312 	if (seq == EXT4_MMP_SEQ_FSCK) {
313 		dump_mmp_msg(sb, mmp, "fsck is running on the filesystem");
314 		retval = -EBUSY;
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 		retval = -ETIMEDOUT;
329 		goto failed;
330 	}
331 
332 	retval = read_mmp_block(sb, &bh, mmp_block);
333 	if (retval)
334 		goto failed;
335 	mmp = (struct mmp_struct *)(bh->b_data);
336 	if (seq != le32_to_cpu(mmp->mmp_seq)) {
337 		dump_mmp_msg(sb, mmp,
338 			     "Device is already active on another node.");
339 		retval = -EBUSY;
340 		goto failed;
341 	}
342 
343 skip:
344 	/*
345 	 * write a new random sequence number.
346 	 */
347 	seq = mmp_new_seq();
348 	mmp->mmp_seq = cpu_to_le32(seq);
349 
350 	/*
351 	 * On mount / remount we are protected against fs freezing (by s_umount
352 	 * semaphore) and grabbing freeze protection upsets lockdep
353 	 */
354 	retval = write_mmp_block_thawed(sb, bh);
355 	if (retval)
356 		goto failed;
357 
358 	/*
359 	 * wait for MMP interval and check mmp_seq.
360 	 */
361 	if (schedule_timeout_interruptible(HZ * wait_time) != 0) {
362 		ext4_warning(sb, "MMP startup interrupted, failing mount");
363 		retval = -ETIMEDOUT;
364 		goto failed;
365 	}
366 
367 	retval = read_mmp_block(sb, &bh, mmp_block);
368 	if (retval)
369 		goto failed;
370 	mmp = (struct mmp_struct *)(bh->b_data);
371 	if (seq != le32_to_cpu(mmp->mmp_seq)) {
372 		dump_mmp_msg(sb, mmp,
373 			     "Device is already active on another node.");
374 		retval = -EBUSY;
375 		goto failed;
376 	}
377 
378 	EXT4_SB(sb)->s_mmp_bh = bh;
379 
380 	BUILD_BUG_ON(sizeof(mmp->mmp_bdevname) < BDEVNAME_SIZE);
381 	snprintf(mmp->mmp_bdevname, sizeof(mmp->mmp_bdevname),
382 		 "%pg", bh->b_bdev);
383 
384 	/*
385 	 * Start a kernel thread to update the MMP block periodically.
386 	 */
387 	EXT4_SB(sb)->s_mmp_tsk = kthread_run(kmmpd, sb, "kmmpd-%.*s",
388 					     (int)sizeof(mmp->mmp_bdevname),
389 					     mmp->mmp_bdevname);
390 	if (IS_ERR(EXT4_SB(sb)->s_mmp_tsk)) {
391 		EXT4_SB(sb)->s_mmp_tsk = NULL;
392 		ext4_warning(sb, "Unable to create kmmpd thread for %s.",
393 			     sb->s_id);
394 		retval = -ENOMEM;
395 		goto failed;
396 	}
397 
398 	return 0;
399 
400 failed:
401 	brelse(bh);
402 	return retval;
403 }
404