xref: /openbmc/linux/fs/ext4/mmp.c (revision f8e17c17)
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 1;
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 	get_bh(*bh);
89 	lock_buffer(*bh);
90 	(*bh)->b_end_io = end_buffer_read_sync;
91 	submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, *bh);
92 	wait_on_buffer(*bh);
93 	if (!buffer_uptodate(*bh)) {
94 		ret = -EIO;
95 		goto warn_exit;
96 	}
97 	mmp = (struct mmp_struct *)((*bh)->b_data);
98 	if (le32_to_cpu(mmp->mmp_magic) != EXT4_MMP_MAGIC) {
99 		ret = -EFSCORRUPTED;
100 		goto warn_exit;
101 	}
102 	if (!ext4_mmp_csum_verify(sb, mmp)) {
103 		ret = -EFSBADCRC;
104 		goto warn_exit;
105 	}
106 	return 0;
107 warn_exit:
108 	brelse(*bh);
109 	*bh = NULL;
110 	ext4_warning(sb, "Error %d while reading MMP block %llu",
111 		     ret, mmp_block);
112 	return ret;
113 }
114 
115 /*
116  * Dump as much information as possible to help the admin.
117  */
118 void __dump_mmp_msg(struct super_block *sb, struct mmp_struct *mmp,
119 		    const char *function, unsigned int line, const char *msg)
120 {
121 	__ext4_warning(sb, function, line, "%s", msg);
122 	__ext4_warning(sb, function, line,
123 		       "MMP failure info: last update time: %llu, last update node: %.*s, last update device: %.*s",
124 		       (unsigned long long)le64_to_cpu(mmp->mmp_time),
125 		       (int)sizeof(mmp->mmp_nodename), mmp->mmp_nodename,
126 		       (int)sizeof(mmp->mmp_bdevname), mmp->mmp_bdevname);
127 }
128 
129 /*
130  * kmmpd will update the MMP sequence every s_mmp_update_interval seconds
131  */
132 static int kmmpd(void *data)
133 {
134 	struct super_block *sb = ((struct mmpd_data *) data)->sb;
135 	struct buffer_head *bh = ((struct mmpd_data *) data)->bh;
136 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
137 	struct mmp_struct *mmp;
138 	ext4_fsblk_t mmp_block;
139 	u32 seq = 0;
140 	unsigned long failed_writes = 0;
141 	int mmp_update_interval = le16_to_cpu(es->s_mmp_update_interval);
142 	unsigned mmp_check_interval;
143 	unsigned long last_update_time;
144 	unsigned long diff;
145 	int retval;
146 
147 	mmp_block = le64_to_cpu(es->s_mmp_block);
148 	mmp = (struct mmp_struct *)(bh->b_data);
149 	mmp->mmp_time = cpu_to_le64(ktime_get_real_seconds());
150 	/*
151 	 * Start with the higher mmp_check_interval and reduce it if
152 	 * the MMP block is being updated on time.
153 	 */
154 	mmp_check_interval = max(EXT4_MMP_CHECK_MULT * mmp_update_interval,
155 				 EXT4_MMP_MIN_CHECK_INTERVAL);
156 	mmp->mmp_check_interval = cpu_to_le16(mmp_check_interval);
157 	BUILD_BUG_ON(sizeof(mmp->mmp_bdevname) < BDEVNAME_SIZE);
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_set_errno(sb, -retval);
179 				ext4_error(sb, "Error writing to MMP block");
180 			}
181 			failed_writes++;
182 		}
183 
184 		if (!(le32_to_cpu(es->s_feature_incompat) &
185 		    EXT4_FEATURE_INCOMPAT_MMP)) {
186 			ext4_warning(sb, "kmmpd being stopped since MMP feature"
187 				     " has been disabled.");
188 			goto exit_thread;
189 		}
190 
191 		if (sb_rdonly(sb))
192 			break;
193 
194 		diff = jiffies - last_update_time;
195 		if (diff < mmp_update_interval * HZ)
196 			schedule_timeout_interruptible(mmp_update_interval *
197 						       HZ - diff);
198 
199 		/*
200 		 * We need to make sure that more than mmp_check_interval
201 		 * seconds have not passed since writing. If that has happened
202 		 * we need to check if the MMP block is as we left it.
203 		 */
204 		diff = jiffies - last_update_time;
205 		if (diff > mmp_check_interval * HZ) {
206 			struct buffer_head *bh_check = NULL;
207 			struct mmp_struct *mmp_check;
208 
209 			retval = read_mmp_block(sb, &bh_check, mmp_block);
210 			if (retval) {
211 				ext4_set_errno(sb, -retval);
212 				ext4_error(sb, "error reading MMP data: %d",
213 					   retval);
214 				goto exit_thread;
215 			}
216 
217 			mmp_check = (struct mmp_struct *)(bh_check->b_data);
218 			if (mmp->mmp_seq != mmp_check->mmp_seq ||
219 			    memcmp(mmp->mmp_nodename, mmp_check->mmp_nodename,
220 				   sizeof(mmp->mmp_nodename))) {
221 				dump_mmp_msg(sb, mmp_check,
222 					     "Error while updating MMP info. "
223 					     "The filesystem seems to have been"
224 					     " multiply mounted.");
225 				ext4_set_errno(sb, EBUSY);
226 				ext4_error(sb, "abort");
227 				put_bh(bh_check);
228 				retval = -EBUSY;
229 				goto exit_thread;
230 			}
231 			put_bh(bh_check);
232 		}
233 
234 		 /*
235 		 * Adjust the mmp_check_interval depending on how much time
236 		 * it took for the MMP block to be written.
237 		 */
238 		mmp_check_interval = max(min(EXT4_MMP_CHECK_MULT * diff / HZ,
239 					     EXT4_MMP_MAX_CHECK_INTERVAL),
240 					 EXT4_MMP_MIN_CHECK_INTERVAL);
241 		mmp->mmp_check_interval = cpu_to_le16(mmp_check_interval);
242 	}
243 
244 	/*
245 	 * Unmount seems to be clean.
246 	 */
247 	mmp->mmp_seq = cpu_to_le32(EXT4_MMP_SEQ_CLEAN);
248 	mmp->mmp_time = cpu_to_le64(ktime_get_real_seconds());
249 
250 	retval = write_mmp_block(sb, bh);
251 
252 exit_thread:
253 	EXT4_SB(sb)->s_mmp_tsk = NULL;
254 	kfree(data);
255 	brelse(bh);
256 	return retval;
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 	struct mmpd_data *mmpd_data;
284 	u32 seq;
285 	unsigned int mmp_check_interval = le16_to_cpu(es->s_mmp_update_interval);
286 	unsigned int wait_time = 0;
287 	int retval;
288 
289 	if (mmp_block < le32_to_cpu(es->s_first_data_block) ||
290 	    mmp_block >= ext4_blocks_count(es)) {
291 		ext4_warning(sb, "Invalid MMP block in superblock");
292 		goto failed;
293 	}
294 
295 	retval = read_mmp_block(sb, &bh, mmp_block);
296 	if (retval)
297 		goto failed;
298 
299 	mmp = (struct mmp_struct *)(bh->b_data);
300 
301 	if (mmp_check_interval < EXT4_MMP_MIN_CHECK_INTERVAL)
302 		mmp_check_interval = EXT4_MMP_MIN_CHECK_INTERVAL;
303 
304 	/*
305 	 * If check_interval in MMP block is larger, use that instead of
306 	 * update_interval from the superblock.
307 	 */
308 	if (le16_to_cpu(mmp->mmp_check_interval) > mmp_check_interval)
309 		mmp_check_interval = le16_to_cpu(mmp->mmp_check_interval);
310 
311 	seq = le32_to_cpu(mmp->mmp_seq);
312 	if (seq == EXT4_MMP_SEQ_CLEAN)
313 		goto skip;
314 
315 	if (seq == EXT4_MMP_SEQ_FSCK) {
316 		dump_mmp_msg(sb, mmp, "fsck is running on the filesystem");
317 		goto failed;
318 	}
319 
320 	wait_time = min(mmp_check_interval * 2 + 1,
321 			mmp_check_interval + 60);
322 
323 	/* Print MMP interval if more than 20 secs. */
324 	if (wait_time > EXT4_MMP_MIN_CHECK_INTERVAL * 4)
325 		ext4_warning(sb, "MMP interval %u higher than expected, please"
326 			     " wait.\n", wait_time * 2);
327 
328 	if (schedule_timeout_interruptible(HZ * wait_time) != 0) {
329 		ext4_warning(sb, "MMP startup interrupted, failing mount\n");
330 		goto failed;
331 	}
332 
333 	retval = read_mmp_block(sb, &bh, mmp_block);
334 	if (retval)
335 		goto failed;
336 	mmp = (struct mmp_struct *)(bh->b_data);
337 	if (seq != le32_to_cpu(mmp->mmp_seq)) {
338 		dump_mmp_msg(sb, mmp,
339 			     "Device is already active on another node.");
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 	retval = write_mmp_block(sb, bh);
351 	if (retval)
352 		goto failed;
353 
354 	/*
355 	 * wait for MMP interval and check mmp_seq.
356 	 */
357 	if (schedule_timeout_interruptible(HZ * wait_time) != 0) {
358 		ext4_warning(sb, "MMP startup interrupted, failing mount");
359 		goto failed;
360 	}
361 
362 	retval = read_mmp_block(sb, &bh, mmp_block);
363 	if (retval)
364 		goto failed;
365 	mmp = (struct mmp_struct *)(bh->b_data);
366 	if (seq != le32_to_cpu(mmp->mmp_seq)) {
367 		dump_mmp_msg(sb, mmp,
368 			     "Device is already active on another node.");
369 		goto failed;
370 	}
371 
372 	mmpd_data = kmalloc(sizeof(*mmpd_data), GFP_KERNEL);
373 	if (!mmpd_data) {
374 		ext4_warning(sb, "not enough memory for mmpd_data");
375 		goto failed;
376 	}
377 	mmpd_data->sb = sb;
378 	mmpd_data->bh = bh;
379 
380 	/*
381 	 * Start a kernel thread to update the MMP block periodically.
382 	 */
383 	EXT4_SB(sb)->s_mmp_tsk = kthread_run(kmmpd, mmpd_data, "kmmpd-%.*s",
384 					     (int)sizeof(mmp->mmp_bdevname),
385 					     bdevname(bh->b_bdev,
386 						      mmp->mmp_bdevname));
387 	if (IS_ERR(EXT4_SB(sb)->s_mmp_tsk)) {
388 		EXT4_SB(sb)->s_mmp_tsk = NULL;
389 		kfree(mmpd_data);
390 		ext4_warning(sb, "Unable to create kmmpd thread for %s.",
391 			     sb->s_id);
392 		goto failed;
393 	}
394 
395 	return 0;
396 
397 failed:
398 	brelse(bh);
399 	return 1;
400 }
401 
402 
403