xref: /openbmc/linux/fs/ext4/mmp.c (revision 4f3db074)
1 #include <linux/fs.h>
2 #include <linux/random.h>
3 #include <linux/buffer_head.h>
4 #include <linux/utsname.h>
5 #include <linux/kthread.h>
6 
7 #include "ext4.h"
8 
9 /* Checksumming functions */
10 static __le32 ext4_mmp_csum(struct super_block *sb, struct mmp_struct *mmp)
11 {
12 	struct ext4_sb_info *sbi = EXT4_SB(sb);
13 	int offset = offsetof(struct mmp_struct, mmp_checksum);
14 	__u32 csum;
15 
16 	csum = ext4_chksum(sbi, sbi->s_csum_seed, (char *)mmp, offset);
17 
18 	return cpu_to_le32(csum);
19 }
20 
21 static int ext4_mmp_csum_verify(struct super_block *sb, struct mmp_struct *mmp)
22 {
23 	if (!ext4_has_metadata_csum(sb))
24 		return 1;
25 
26 	return mmp->mmp_checksum == ext4_mmp_csum(sb, mmp);
27 }
28 
29 static void ext4_mmp_csum_set(struct super_block *sb, struct mmp_struct *mmp)
30 {
31 	if (!ext4_has_metadata_csum(sb))
32 		return;
33 
34 	mmp->mmp_checksum = ext4_mmp_csum(sb, mmp);
35 }
36 
37 /*
38  * Write the MMP block using WRITE_SYNC to try to get the block on-disk
39  * faster.
40  */
41 static int write_mmp_block(struct super_block *sb, struct buffer_head *bh)
42 {
43 	struct mmp_struct *mmp = (struct mmp_struct *)(bh->b_data);
44 
45 	/*
46 	 * We protect against freezing so that we don't create dirty buffers
47 	 * on frozen filesystem.
48 	 */
49 	sb_start_write(sb);
50 	ext4_mmp_csum_set(sb, mmp);
51 	mark_buffer_dirty(bh);
52 	lock_buffer(bh);
53 	bh->b_end_io = end_buffer_write_sync;
54 	get_bh(bh);
55 	submit_bh(WRITE_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 
73 	if (*bh)
74 		clear_buffer_uptodate(*bh);
75 
76 	/* This would be sb_bread(sb, mmp_block), except we need to be sure
77 	 * that the MD RAID device cache has been bypassed, and that the read
78 	 * is not blocked in the elevator. */
79 	if (!*bh)
80 		*bh = sb_getblk(sb, mmp_block);
81 	if (!*bh)
82 		return -ENOMEM;
83 	if (*bh) {
84 		get_bh(*bh);
85 		lock_buffer(*bh);
86 		(*bh)->b_end_io = end_buffer_read_sync;
87 		submit_bh(READ_SYNC | REQ_META | REQ_PRIO, *bh);
88 		wait_on_buffer(*bh);
89 		if (!buffer_uptodate(*bh)) {
90 			brelse(*bh);
91 			*bh = NULL;
92 		}
93 	}
94 	if (unlikely(!*bh)) {
95 		ext4_warning(sb, "Error while reading MMP block %llu",
96 			     mmp_block);
97 		return -EIO;
98 	}
99 
100 	mmp = (struct mmp_struct *)((*bh)->b_data);
101 	if (le32_to_cpu(mmp->mmp_magic) != EXT4_MMP_MAGIC ||
102 	    !ext4_mmp_csum_verify(sb, mmp))
103 		return -EINVAL;
104 
105 	return 0;
106 }
107 
108 /*
109  * Dump as much information as possible to help the admin.
110  */
111 void __dump_mmp_msg(struct super_block *sb, struct mmp_struct *mmp,
112 		    const char *function, unsigned int line, const char *msg)
113 {
114 	__ext4_warning(sb, function, line, msg);
115 	__ext4_warning(sb, function, line,
116 		       "MMP failure info: last update time: %llu, last update "
117 		       "node: %s, last update device: %s\n",
118 		       (long long unsigned int) le64_to_cpu(mmp->mmp_time),
119 		       mmp->mmp_nodename, mmp->mmp_bdevname);
120 }
121 
122 /*
123  * kmmpd will update the MMP sequence every s_mmp_update_interval seconds
124  */
125 static int kmmpd(void *data)
126 {
127 	struct super_block *sb = ((struct mmpd_data *) data)->sb;
128 	struct buffer_head *bh = ((struct mmpd_data *) data)->bh;
129 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
130 	struct mmp_struct *mmp;
131 	ext4_fsblk_t mmp_block;
132 	u32 seq = 0;
133 	unsigned long failed_writes = 0;
134 	int mmp_update_interval = le16_to_cpu(es->s_mmp_update_interval);
135 	unsigned mmp_check_interval;
136 	unsigned long last_update_time;
137 	unsigned long diff;
138 	int retval;
139 
140 	mmp_block = le64_to_cpu(es->s_mmp_block);
141 	mmp = (struct mmp_struct *)(bh->b_data);
142 	mmp->mmp_time = cpu_to_le64(get_seconds());
143 	/*
144 	 * Start with the higher mmp_check_interval and reduce it if
145 	 * the MMP block is being updated on time.
146 	 */
147 	mmp_check_interval = max(EXT4_MMP_CHECK_MULT * mmp_update_interval,
148 				 EXT4_MMP_MIN_CHECK_INTERVAL);
149 	mmp->mmp_check_interval = cpu_to_le16(mmp_check_interval);
150 	bdevname(bh->b_bdev, mmp->mmp_bdevname);
151 
152 	memcpy(mmp->mmp_nodename, init_utsname()->nodename,
153 	       sizeof(mmp->mmp_nodename));
154 
155 	while (!kthread_should_stop()) {
156 		if (++seq > EXT4_MMP_SEQ_MAX)
157 			seq = 1;
158 
159 		mmp->mmp_seq = cpu_to_le32(seq);
160 		mmp->mmp_time = cpu_to_le64(get_seconds());
161 		last_update_time = jiffies;
162 
163 		retval = write_mmp_block(sb, bh);
164 		/*
165 		 * Don't spew too many error messages. Print one every
166 		 * (s_mmp_update_interval * 60) seconds.
167 		 */
168 		if (retval) {
169 			if ((failed_writes % 60) == 0)
170 				ext4_error(sb, "Error writing to MMP block");
171 			failed_writes++;
172 		}
173 
174 		if (!(le32_to_cpu(es->s_feature_incompat) &
175 		    EXT4_FEATURE_INCOMPAT_MMP)) {
176 			ext4_warning(sb, "kmmpd being stopped since MMP feature"
177 				     " has been disabled.");
178 			EXT4_SB(sb)->s_mmp_tsk = NULL;
179 			goto failed;
180 		}
181 
182 		if (sb->s_flags & MS_RDONLY) {
183 			ext4_warning(sb, "kmmpd being stopped since filesystem "
184 				     "has been remounted as readonly.");
185 			EXT4_SB(sb)->s_mmp_tsk = NULL;
186 			goto failed;
187 		}
188 
189 		diff = jiffies - last_update_time;
190 		if (diff < mmp_update_interval * HZ)
191 			schedule_timeout_interruptible(mmp_update_interval *
192 						       HZ - diff);
193 
194 		/*
195 		 * We need to make sure that more than mmp_check_interval
196 		 * seconds have not passed since writing. If that has happened
197 		 * we need to check if the MMP block is as we left it.
198 		 */
199 		diff = jiffies - last_update_time;
200 		if (diff > mmp_check_interval * HZ) {
201 			struct buffer_head *bh_check = NULL;
202 			struct mmp_struct *mmp_check;
203 
204 			retval = read_mmp_block(sb, &bh_check, mmp_block);
205 			if (retval) {
206 				ext4_error(sb, "error reading MMP data: %d",
207 					   retval);
208 
209 				EXT4_SB(sb)->s_mmp_tsk = NULL;
210 				goto failed;
211 			}
212 
213 			mmp_check = (struct mmp_struct *)(bh_check->b_data);
214 			if (mmp->mmp_seq != mmp_check->mmp_seq ||
215 			    memcmp(mmp->mmp_nodename, mmp_check->mmp_nodename,
216 				   sizeof(mmp->mmp_nodename))) {
217 				dump_mmp_msg(sb, mmp_check,
218 					     "Error while updating MMP info. "
219 					     "The filesystem seems to have been"
220 					     " multiply mounted.");
221 				ext4_error(sb, "abort");
222 				goto failed;
223 			}
224 			put_bh(bh_check);
225 		}
226 
227 		 /*
228 		 * Adjust the mmp_check_interval depending on how much time
229 		 * it took for the MMP block to be written.
230 		 */
231 		mmp_check_interval = max(min(EXT4_MMP_CHECK_MULT * diff / HZ,
232 					     EXT4_MMP_MAX_CHECK_INTERVAL),
233 					 EXT4_MMP_MIN_CHECK_INTERVAL);
234 		mmp->mmp_check_interval = cpu_to_le16(mmp_check_interval);
235 	}
236 
237 	/*
238 	 * Unmount seems to be clean.
239 	 */
240 	mmp->mmp_seq = cpu_to_le32(EXT4_MMP_SEQ_CLEAN);
241 	mmp->mmp_time = cpu_to_le64(get_seconds());
242 
243 	retval = write_mmp_block(sb, bh);
244 
245 failed:
246 	kfree(data);
247 	brelse(bh);
248 	return retval;
249 }
250 
251 /*
252  * Get a random new sequence number but make sure it is not greater than
253  * EXT4_MMP_SEQ_MAX.
254  */
255 static unsigned int mmp_new_seq(void)
256 {
257 	u32 new_seq;
258 
259 	do {
260 		new_seq = prandom_u32();
261 	} while (new_seq > EXT4_MMP_SEQ_MAX);
262 
263 	return new_seq;
264 }
265 
266 /*
267  * Protect the filesystem from being mounted more than once.
268  */
269 int ext4_multi_mount_protect(struct super_block *sb,
270 				    ext4_fsblk_t mmp_block)
271 {
272 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
273 	struct buffer_head *bh = NULL;
274 	struct mmp_struct *mmp = NULL;
275 	struct mmpd_data *mmpd_data;
276 	u32 seq;
277 	unsigned int mmp_check_interval = le16_to_cpu(es->s_mmp_update_interval);
278 	unsigned int wait_time = 0;
279 	int retval;
280 
281 	if (mmp_block < le32_to_cpu(es->s_first_data_block) ||
282 	    mmp_block >= ext4_blocks_count(es)) {
283 		ext4_warning(sb, "Invalid MMP block in superblock");
284 		goto failed;
285 	}
286 
287 	retval = read_mmp_block(sb, &bh, mmp_block);
288 	if (retval)
289 		goto failed;
290 
291 	mmp = (struct mmp_struct *)(bh->b_data);
292 
293 	if (mmp_check_interval < EXT4_MMP_MIN_CHECK_INTERVAL)
294 		mmp_check_interval = EXT4_MMP_MIN_CHECK_INTERVAL;
295 
296 	/*
297 	 * If check_interval in MMP block is larger, use that instead of
298 	 * update_interval from the superblock.
299 	 */
300 	if (le16_to_cpu(mmp->mmp_check_interval) > mmp_check_interval)
301 		mmp_check_interval = le16_to_cpu(mmp->mmp_check_interval);
302 
303 	seq = le32_to_cpu(mmp->mmp_seq);
304 	if (seq == EXT4_MMP_SEQ_CLEAN)
305 		goto skip;
306 
307 	if (seq == EXT4_MMP_SEQ_FSCK) {
308 		dump_mmp_msg(sb, mmp, "fsck is running on the filesystem");
309 		goto failed;
310 	}
311 
312 	wait_time = min(mmp_check_interval * 2 + 1,
313 			mmp_check_interval + 60);
314 
315 	/* Print MMP interval if more than 20 secs. */
316 	if (wait_time > EXT4_MMP_MIN_CHECK_INTERVAL * 4)
317 		ext4_warning(sb, "MMP interval %u higher than expected, please"
318 			     " wait.\n", wait_time * 2);
319 
320 	if (schedule_timeout_interruptible(HZ * wait_time) != 0) {
321 		ext4_warning(sb, "MMP startup interrupted, failing mount\n");
322 		goto failed;
323 	}
324 
325 	retval = read_mmp_block(sb, &bh, mmp_block);
326 	if (retval)
327 		goto failed;
328 	mmp = (struct mmp_struct *)(bh->b_data);
329 	if (seq != le32_to_cpu(mmp->mmp_seq)) {
330 		dump_mmp_msg(sb, mmp,
331 			     "Device is already active on another node.");
332 		goto failed;
333 	}
334 
335 skip:
336 	/*
337 	 * write a new random sequence number.
338 	 */
339 	seq = mmp_new_seq();
340 	mmp->mmp_seq = cpu_to_le32(seq);
341 
342 	retval = write_mmp_block(sb, bh);
343 	if (retval)
344 		goto failed;
345 
346 	/*
347 	 * wait for MMP interval and check mmp_seq.
348 	 */
349 	if (schedule_timeout_interruptible(HZ * wait_time) != 0) {
350 		ext4_warning(sb, "MMP startup interrupted, failing mount\n");
351 		goto failed;
352 	}
353 
354 	retval = read_mmp_block(sb, &bh, mmp_block);
355 	if (retval)
356 		goto failed;
357 	mmp = (struct mmp_struct *)(bh->b_data);
358 	if (seq != le32_to_cpu(mmp->mmp_seq)) {
359 		dump_mmp_msg(sb, mmp,
360 			     "Device is already active on another node.");
361 		goto failed;
362 	}
363 
364 	mmpd_data = kmalloc(sizeof(struct mmpd_data), GFP_KERNEL);
365 	if (!mmpd_data) {
366 		ext4_warning(sb, "not enough memory for mmpd_data");
367 		goto failed;
368 	}
369 	mmpd_data->sb = sb;
370 	mmpd_data->bh = bh;
371 
372 	/*
373 	 * Start a kernel thread to update the MMP block periodically.
374 	 */
375 	EXT4_SB(sb)->s_mmp_tsk = kthread_run(kmmpd, mmpd_data, "kmmpd-%s",
376 					     bdevname(bh->b_bdev,
377 						      mmp->mmp_bdevname));
378 	if (IS_ERR(EXT4_SB(sb)->s_mmp_tsk)) {
379 		EXT4_SB(sb)->s_mmp_tsk = NULL;
380 		kfree(mmpd_data);
381 		ext4_warning(sb, "Unable to create kmmpd thread for %s.",
382 			     sb->s_id);
383 		goto failed;
384 	}
385 
386 	return 0;
387 
388 failed:
389 	brelse(bh);
390 	return 1;
391 }
392 
393 
394