xref: /openbmc/linux/fs/ext4/mmp.c (revision e40573a4)
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 = 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 
154 	memcpy(mmp->mmp_nodename, init_utsname()->nodename,
155 	       sizeof(mmp->mmp_nodename));
156 
157 	while (!kthread_should_stop() && !sb_rdonly(sb)) {
158 		if (!ext4_has_feature_mmp(sb)) {
159 			ext4_warning(sb, "kmmpd being stopped since MMP feature"
160 				     " has been disabled.");
161 			goto wait_to_exit;
162 		}
163 		if (++seq > EXT4_MMP_SEQ_MAX)
164 			seq = 1;
165 
166 		mmp->mmp_seq = cpu_to_le32(seq);
167 		mmp->mmp_time = cpu_to_le64(ktime_get_real_seconds());
168 		last_update_time = jiffies;
169 
170 		retval = write_mmp_block(sb, bh);
171 		/*
172 		 * Don't spew too many error messages. Print one every
173 		 * (s_mmp_update_interval * 60) seconds.
174 		 */
175 		if (retval) {
176 			if ((failed_writes % 60) == 0) {
177 				ext4_error_err(sb, -retval,
178 					       "Error writing to MMP block");
179 			}
180 			failed_writes++;
181 		}
182 
183 		diff = jiffies - last_update_time;
184 		if (diff < mmp_update_interval * HZ)
185 			schedule_timeout_interruptible(mmp_update_interval *
186 						       HZ - diff);
187 
188 		/*
189 		 * We need to make sure that more than mmp_check_interval
190 		 * seconds have not passed since writing. If that has happened
191 		 * we need to check if the MMP block is as we left it.
192 		 */
193 		diff = jiffies - last_update_time;
194 		if (diff > mmp_check_interval * HZ) {
195 			struct buffer_head *bh_check = NULL;
196 			struct mmp_struct *mmp_check;
197 
198 			retval = read_mmp_block(sb, &bh_check, mmp_block);
199 			if (retval) {
200 				ext4_error_err(sb, -retval,
201 					       "error reading MMP data: %d",
202 					       retval);
203 				goto wait_to_exit;
204 			}
205 
206 			mmp_check = (struct mmp_struct *)(bh_check->b_data);
207 			if (mmp->mmp_seq != mmp_check->mmp_seq ||
208 			    memcmp(mmp->mmp_nodename, mmp_check->mmp_nodename,
209 				   sizeof(mmp->mmp_nodename))) {
210 				dump_mmp_msg(sb, mmp_check,
211 					     "Error while updating MMP info. "
212 					     "The filesystem seems to have been"
213 					     " multiply mounted.");
214 				ext4_error_err(sb, EBUSY, "abort");
215 				put_bh(bh_check);
216 				retval = -EBUSY;
217 				goto wait_to_exit;
218 			}
219 			put_bh(bh_check);
220 		}
221 
222 		 /*
223 		 * Adjust the mmp_check_interval depending on how much time
224 		 * it took for the MMP block to be written.
225 		 */
226 		mmp_check_interval = max(min(EXT4_MMP_CHECK_MULT * diff / HZ,
227 					     EXT4_MMP_MAX_CHECK_INTERVAL),
228 					 EXT4_MMP_MIN_CHECK_INTERVAL);
229 		mmp->mmp_check_interval = cpu_to_le16(mmp_check_interval);
230 	}
231 
232 	/*
233 	 * Unmount seems to be clean.
234 	 */
235 	mmp->mmp_seq = cpu_to_le32(EXT4_MMP_SEQ_CLEAN);
236 	mmp->mmp_time = cpu_to_le64(ktime_get_real_seconds());
237 
238 	retval = write_mmp_block(sb, bh);
239 
240 wait_to_exit:
241 	while (!kthread_should_stop()) {
242 		set_current_state(TASK_INTERRUPTIBLE);
243 		if (!kthread_should_stop())
244 			schedule();
245 	}
246 	set_current_state(TASK_RUNNING);
247 	return retval;
248 }
249 
250 void ext4_stop_mmpd(struct ext4_sb_info *sbi)
251 {
252 	if (sbi->s_mmp_tsk) {
253 		kthread_stop(sbi->s_mmp_tsk);
254 		brelse(sbi->s_mmp_bh);
255 		sbi->s_mmp_tsk = NULL;
256 	}
257 }
258 
259 /*
260  * Get a random new sequence number but make sure it is not greater than
261  * EXT4_MMP_SEQ_MAX.
262  */
263 static unsigned int mmp_new_seq(void)
264 {
265 	u32 new_seq;
266 
267 	do {
268 		new_seq = prandom_u32();
269 	} while (new_seq > EXT4_MMP_SEQ_MAX);
270 
271 	return new_seq;
272 }
273 
274 /*
275  * Protect the filesystem from being mounted more than once.
276  */
277 int ext4_multi_mount_protect(struct super_block *sb,
278 				    ext4_fsblk_t mmp_block)
279 {
280 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
281 	struct buffer_head *bh = NULL;
282 	struct mmp_struct *mmp = NULL;
283 	u32 seq;
284 	unsigned int mmp_check_interval = le16_to_cpu(es->s_mmp_update_interval);
285 	unsigned int wait_time = 0;
286 	int retval;
287 
288 	if (mmp_block < le32_to_cpu(es->s_first_data_block) ||
289 	    mmp_block >= ext4_blocks_count(es)) {
290 		ext4_warning(sb, "Invalid MMP block in superblock");
291 		goto failed;
292 	}
293 
294 	retval = read_mmp_block(sb, &bh, mmp_block);
295 	if (retval)
296 		goto failed;
297 
298 	mmp = (struct mmp_struct *)(bh->b_data);
299 
300 	if (mmp_check_interval < EXT4_MMP_MIN_CHECK_INTERVAL)
301 		mmp_check_interval = EXT4_MMP_MIN_CHECK_INTERVAL;
302 
303 	/*
304 	 * If check_interval in MMP block is larger, use that instead of
305 	 * update_interval from the superblock.
306 	 */
307 	if (le16_to_cpu(mmp->mmp_check_interval) > mmp_check_interval)
308 		mmp_check_interval = le16_to_cpu(mmp->mmp_check_interval);
309 
310 	seq = le32_to_cpu(mmp->mmp_seq);
311 	if (seq == EXT4_MMP_SEQ_CLEAN)
312 		goto skip;
313 
314 	if (seq == EXT4_MMP_SEQ_FSCK) {
315 		dump_mmp_msg(sb, mmp, "fsck is running on the filesystem");
316 		goto failed;
317 	}
318 
319 	wait_time = min(mmp_check_interval * 2 + 1,
320 			mmp_check_interval + 60);
321 
322 	/* Print MMP interval if more than 20 secs. */
323 	if (wait_time > EXT4_MMP_MIN_CHECK_INTERVAL * 4)
324 		ext4_warning(sb, "MMP interval %u higher than expected, please"
325 			     " wait.\n", wait_time * 2);
326 
327 	if (schedule_timeout_interruptible(HZ * wait_time) != 0) {
328 		ext4_warning(sb, "MMP startup interrupted, failing mount\n");
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 		goto failed;
340 	}
341 
342 skip:
343 	/*
344 	 * write a new random sequence number.
345 	 */
346 	seq = mmp_new_seq();
347 	mmp->mmp_seq = cpu_to_le32(seq);
348 
349 	retval = write_mmp_block(sb, bh);
350 	if (retval)
351 		goto failed;
352 
353 	/*
354 	 * wait for MMP interval and check mmp_seq.
355 	 */
356 	if (schedule_timeout_interruptible(HZ * wait_time) != 0) {
357 		ext4_warning(sb, "MMP startup interrupted, failing mount");
358 		goto failed;
359 	}
360 
361 	retval = read_mmp_block(sb, &bh, mmp_block);
362 	if (retval)
363 		goto failed;
364 	mmp = (struct mmp_struct *)(bh->b_data);
365 	if (seq != le32_to_cpu(mmp->mmp_seq)) {
366 		dump_mmp_msg(sb, mmp,
367 			     "Device is already active on another node.");
368 		goto failed;
369 	}
370 
371 	EXT4_SB(sb)->s_mmp_bh = bh;
372 
373 	BUILD_BUG_ON(sizeof(mmp->mmp_bdevname) < BDEVNAME_SIZE);
374 	snprintf(mmp->mmp_bdevname, sizeof(mmp->mmp_bdevname),
375 		 "%pg", bh->b_bdev);
376 
377 	/*
378 	 * Start a kernel thread to update the MMP block periodically.
379 	 */
380 	EXT4_SB(sb)->s_mmp_tsk = kthread_run(kmmpd, sb, "kmmpd-%.*s",
381 					     (int)sizeof(mmp->mmp_bdevname),
382 					     mmp->mmp_bdevname);
383 	if (IS_ERR(EXT4_SB(sb)->s_mmp_tsk)) {
384 		EXT4_SB(sb)->s_mmp_tsk = NULL;
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