xref: /openbmc/linux/fs/ext4/mmp.c (revision 66127f0d)
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_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->b_end_io = end_buffer_write_sync;
50 	get_bh(bh);
51 	submit_bh(REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO, bh);
52 	wait_on_buffer(bh);
53 	if (unlikely(!buffer_uptodate(bh)))
54 		return -EIO;
55 	return 0;
56 }
57 
58 static int write_mmp_block(struct super_block *sb, struct buffer_head *bh)
59 {
60 	int err;
61 
62 	/*
63 	 * We protect against freezing so that we don't create dirty buffers
64 	 * on frozen filesystem.
65 	 */
66 	sb_start_write(sb);
67 	err = write_mmp_block_thawed(sb, bh);
68 	sb_end_write(sb);
69 	return err;
70 }
71 
72 /*
73  * Read the MMP block. It _must_ be read from disk and hence we clear the
74  * uptodate flag on the buffer.
75  */
76 static int read_mmp_block(struct super_block *sb, struct buffer_head **bh,
77 			  ext4_fsblk_t mmp_block)
78 {
79 	struct mmp_struct *mmp;
80 	int ret;
81 
82 	if (*bh)
83 		clear_buffer_uptodate(*bh);
84 
85 	/* This would be sb_bread(sb, mmp_block), except we need to be sure
86 	 * that the MD RAID device cache has been bypassed, and that the read
87 	 * is not blocked in the elevator. */
88 	if (!*bh) {
89 		*bh = sb_getblk(sb, mmp_block);
90 		if (!*bh) {
91 			ret = -ENOMEM;
92 			goto warn_exit;
93 		}
94 	}
95 
96 	lock_buffer(*bh);
97 	ret = ext4_read_bh(*bh, REQ_META | REQ_PRIO, NULL, false);
98 	if (ret)
99 		goto warn_exit;
100 
101 	mmp = (struct mmp_struct *)((*bh)->b_data);
102 	if (le32_to_cpu(mmp->mmp_magic) != EXT4_MMP_MAGIC) {
103 		ret = -EFSCORRUPTED;
104 		goto warn_exit;
105 	}
106 	if (!ext4_mmp_csum_verify(sb, mmp)) {
107 		ret = -EFSBADCRC;
108 		goto warn_exit;
109 	}
110 	return 0;
111 warn_exit:
112 	brelse(*bh);
113 	*bh = NULL;
114 	ext4_warning(sb, "Error %d while reading MMP block %llu",
115 		     ret, mmp_block);
116 	return ret;
117 }
118 
119 /*
120  * Dump as much information as possible to help the admin.
121  */
122 void __dump_mmp_msg(struct super_block *sb, struct mmp_struct *mmp,
123 		    const char *function, unsigned int line, const char *msg)
124 {
125 	__ext4_warning(sb, function, line, "%s", msg);
126 	__ext4_warning(sb, function, line,
127 		       "MMP failure info: last update time: %llu, last update node: %.*s, last update device: %.*s",
128 		       (unsigned long long)le64_to_cpu(mmp->mmp_time),
129 		       (int)sizeof(mmp->mmp_nodename), mmp->mmp_nodename,
130 		       (int)sizeof(mmp->mmp_bdevname), mmp->mmp_bdevname);
131 }
132 
133 /*
134  * kmmpd will update the MMP sequence every s_mmp_update_interval seconds
135  */
136 static int kmmpd(void *data)
137 {
138 	struct super_block *sb = data;
139 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
140 	struct buffer_head *bh = EXT4_SB(sb)->s_mmp_bh;
141 	struct mmp_struct *mmp;
142 	ext4_fsblk_t mmp_block;
143 	u32 seq = 0;
144 	unsigned long failed_writes = 0;
145 	int mmp_update_interval = le16_to_cpu(es->s_mmp_update_interval);
146 	unsigned mmp_check_interval;
147 	unsigned long last_update_time;
148 	unsigned long diff;
149 	int retval = 0;
150 
151 	mmp_block = le64_to_cpu(es->s_mmp_block);
152 	mmp = (struct mmp_struct *)(bh->b_data);
153 	mmp->mmp_time = cpu_to_le64(ktime_get_real_seconds());
154 	/*
155 	 * Start with the higher mmp_check_interval and reduce it if
156 	 * the MMP block is being updated on time.
157 	 */
158 	mmp_check_interval = max(EXT4_MMP_CHECK_MULT * mmp_update_interval,
159 				 EXT4_MMP_MIN_CHECK_INTERVAL);
160 	mmp->mmp_check_interval = cpu_to_le16(mmp_check_interval);
161 
162 	memcpy(mmp->mmp_nodename, init_utsname()->nodename,
163 	       sizeof(mmp->mmp_nodename));
164 
165 	while (!kthread_should_stop() && !ext4_forced_shutdown(sb)) {
166 		if (!ext4_has_feature_mmp(sb)) {
167 			ext4_warning(sb, "kmmpd being stopped since MMP feature"
168 				     " has been disabled.");
169 			goto wait_to_exit;
170 		}
171 		if (++seq > EXT4_MMP_SEQ_MAX)
172 			seq = 1;
173 
174 		mmp->mmp_seq = cpu_to_le32(seq);
175 		mmp->mmp_time = cpu_to_le64(ktime_get_real_seconds());
176 		last_update_time = jiffies;
177 
178 		retval = write_mmp_block(sb, bh);
179 		/*
180 		 * Don't spew too many error messages. Print one every
181 		 * (s_mmp_update_interval * 60) seconds.
182 		 */
183 		if (retval) {
184 			if ((failed_writes % 60) == 0) {
185 				ext4_error_err(sb, -retval,
186 					       "Error writing to MMP block");
187 			}
188 			failed_writes++;
189 		}
190 
191 		diff = jiffies - last_update_time;
192 		if (diff < mmp_update_interval * HZ)
193 			schedule_timeout_interruptible(mmp_update_interval *
194 						       HZ - diff);
195 
196 		/*
197 		 * We need to make sure that more than mmp_check_interval
198 		 * seconds have not passed since writing. If that has happened
199 		 * we need to check if the MMP block is as we left it.
200 		 */
201 		diff = jiffies - last_update_time;
202 		if (diff > mmp_check_interval * HZ) {
203 			struct buffer_head *bh_check = NULL;
204 			struct mmp_struct *mmp_check;
205 
206 			retval = read_mmp_block(sb, &bh_check, mmp_block);
207 			if (retval) {
208 				ext4_error_err(sb, -retval,
209 					       "error reading MMP data: %d",
210 					       retval);
211 				goto wait_to_exit;
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_err(sb, EBUSY, "abort");
223 				put_bh(bh_check);
224 				retval = -EBUSY;
225 				goto wait_to_exit;
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 wait_to_exit:
249 	while (!kthread_should_stop()) {
250 		set_current_state(TASK_INTERRUPTIBLE);
251 		if (!kthread_should_stop())
252 			schedule();
253 	}
254 	set_current_state(TASK_RUNNING);
255 	return retval;
256 }
257 
258 void ext4_stop_mmpd(struct ext4_sb_info *sbi)
259 {
260 	if (sbi->s_mmp_tsk) {
261 		kthread_stop(sbi->s_mmp_tsk);
262 		brelse(sbi->s_mmp_bh);
263 		sbi->s_mmp_tsk = NULL;
264 	}
265 }
266 
267 /*
268  * Get a random new sequence number but make sure it is not greater than
269  * EXT4_MMP_SEQ_MAX.
270  */
271 static unsigned int mmp_new_seq(void)
272 {
273 	return get_random_u32_below(EXT4_MMP_SEQ_MAX + 1);
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 		retval = -EINVAL;
294 		goto failed;
295 	}
296 
297 	retval = read_mmp_block(sb, &bh, mmp_block);
298 	if (retval)
299 		goto failed;
300 
301 	mmp = (struct mmp_struct *)(bh->b_data);
302 
303 	if (mmp_check_interval < EXT4_MMP_MIN_CHECK_INTERVAL)
304 		mmp_check_interval = EXT4_MMP_MIN_CHECK_INTERVAL;
305 
306 	/*
307 	 * If check_interval in MMP block is larger, use that instead of
308 	 * update_interval from the superblock.
309 	 */
310 	if (le16_to_cpu(mmp->mmp_check_interval) > mmp_check_interval)
311 		mmp_check_interval = le16_to_cpu(mmp->mmp_check_interval);
312 
313 	seq = le32_to_cpu(mmp->mmp_seq);
314 	if (seq == EXT4_MMP_SEQ_CLEAN)
315 		goto skip;
316 
317 	if (seq == EXT4_MMP_SEQ_FSCK) {
318 		dump_mmp_msg(sb, mmp, "fsck is running on the filesystem");
319 		retval = -EBUSY;
320 		goto failed;
321 	}
322 
323 	wait_time = min(mmp_check_interval * 2 + 1,
324 			mmp_check_interval + 60);
325 
326 	/* Print MMP interval if more than 20 secs. */
327 	if (wait_time > EXT4_MMP_MIN_CHECK_INTERVAL * 4)
328 		ext4_warning(sb, "MMP interval %u higher than expected, please"
329 			     " wait.\n", wait_time * 2);
330 
331 	if (schedule_timeout_interruptible(HZ * wait_time) != 0) {
332 		ext4_warning(sb, "MMP startup interrupted, failing mount\n");
333 		retval = -ETIMEDOUT;
334 		goto failed;
335 	}
336 
337 	retval = read_mmp_block(sb, &bh, mmp_block);
338 	if (retval)
339 		goto failed;
340 	mmp = (struct mmp_struct *)(bh->b_data);
341 	if (seq != le32_to_cpu(mmp->mmp_seq)) {
342 		dump_mmp_msg(sb, mmp,
343 			     "Device is already active on another node.");
344 		retval = -EBUSY;
345 		goto failed;
346 	}
347 
348 skip:
349 	/*
350 	 * write a new random sequence number.
351 	 */
352 	seq = mmp_new_seq();
353 	mmp->mmp_seq = cpu_to_le32(seq);
354 
355 	/*
356 	 * On mount / remount we are protected against fs freezing (by s_umount
357 	 * semaphore) and grabbing freeze protection upsets lockdep
358 	 */
359 	retval = write_mmp_block_thawed(sb, bh);
360 	if (retval)
361 		goto failed;
362 
363 	/*
364 	 * wait for MMP interval and check mmp_seq.
365 	 */
366 	if (schedule_timeout_interruptible(HZ * wait_time) != 0) {
367 		ext4_warning(sb, "MMP startup interrupted, failing mount");
368 		retval = -ETIMEDOUT;
369 		goto failed;
370 	}
371 
372 	retval = read_mmp_block(sb, &bh, mmp_block);
373 	if (retval)
374 		goto failed;
375 	mmp = (struct mmp_struct *)(bh->b_data);
376 	if (seq != le32_to_cpu(mmp->mmp_seq)) {
377 		dump_mmp_msg(sb, mmp,
378 			     "Device is already active on another node.");
379 		retval = -EBUSY;
380 		goto failed;
381 	}
382 
383 	EXT4_SB(sb)->s_mmp_bh = bh;
384 
385 	BUILD_BUG_ON(sizeof(mmp->mmp_bdevname) < BDEVNAME_SIZE);
386 	snprintf(mmp->mmp_bdevname, sizeof(mmp->mmp_bdevname),
387 		 "%pg", bh->b_bdev);
388 
389 	/*
390 	 * Start a kernel thread to update the MMP block periodically.
391 	 */
392 	EXT4_SB(sb)->s_mmp_tsk = kthread_run(kmmpd, sb, "kmmpd-%.*s",
393 					     (int)sizeof(mmp->mmp_bdevname),
394 					     mmp->mmp_bdevname);
395 	if (IS_ERR(EXT4_SB(sb)->s_mmp_tsk)) {
396 		EXT4_SB(sb)->s_mmp_tsk = NULL;
397 		ext4_warning(sb, "Unable to create kmmpd thread for %s.",
398 			     sb->s_id);
399 		retval = -ENOMEM;
400 		goto failed;
401 	}
402 
403 	return 0;
404 
405 failed:
406 	brelse(bh);
407 	return retval;
408 }
409