Commit | Line | Data |
---|---|---|
c5e06d10 JL |
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 | /* | |
10 | * Write the MMP block using WRITE_SYNC to try to get the block on-disk | |
11 | * faster. | |
12 | */ | |
13 | static int write_mmp_block(struct buffer_head *bh) | |
14 | { | |
15 | mark_buffer_dirty(bh); | |
16 | lock_buffer(bh); | |
17 | bh->b_end_io = end_buffer_write_sync; | |
18 | get_bh(bh); | |
19 | submit_bh(WRITE_SYNC, bh); | |
20 | wait_on_buffer(bh); | |
21 | if (unlikely(!buffer_uptodate(bh))) | |
22 | return 1; | |
23 | ||
24 | return 0; | |
25 | } | |
26 | ||
27 | /* | |
28 | * Read the MMP block. It _must_ be read from disk and hence we clear the | |
29 | * uptodate flag on the buffer. | |
30 | */ | |
31 | static int read_mmp_block(struct super_block *sb, struct buffer_head **bh, | |
32 | ext4_fsblk_t mmp_block) | |
33 | { | |
34 | struct mmp_struct *mmp; | |
35 | ||
36 | if (*bh) | |
37 | clear_buffer_uptodate(*bh); | |
38 | ||
39 | /* This would be sb_bread(sb, mmp_block), except we need to be sure | |
40 | * that the MD RAID device cache has been bypassed, and that the read | |
41 | * is not blocked in the elevator. */ | |
42 | if (!*bh) | |
43 | *bh = sb_getblk(sb, mmp_block); | |
44 | if (*bh) { | |
45 | get_bh(*bh); | |
46 | lock_buffer(*bh); | |
47 | (*bh)->b_end_io = end_buffer_read_sync; | |
48 | submit_bh(READ_SYNC, *bh); | |
49 | wait_on_buffer(*bh); | |
50 | if (!buffer_uptodate(*bh)) { | |
51 | brelse(*bh); | |
52 | *bh = NULL; | |
53 | } | |
54 | } | |
55 | if (!*bh) { | |
56 | ext4_warning(sb, "Error while reading MMP block %llu", | |
57 | mmp_block); | |
58 | return -EIO; | |
59 | } | |
60 | ||
61 | mmp = (struct mmp_struct *)((*bh)->b_data); | |
62 | if (le32_to_cpu(mmp->mmp_magic) != EXT4_MMP_MAGIC) | |
63 | return -EINVAL; | |
64 | ||
65 | return 0; | |
66 | } | |
67 | ||
68 | /* | |
69 | * Dump as much information as possible to help the admin. | |
70 | */ | |
71 | void __dump_mmp_msg(struct super_block *sb, struct mmp_struct *mmp, | |
72 | const char *function, unsigned int line, const char *msg) | |
73 | { | |
74 | __ext4_warning(sb, function, line, msg); | |
75 | __ext4_warning(sb, function, line, | |
76 | "MMP failure info: last update time: %llu, last update " | |
77 | "node: %s, last update device: %s\n", | |
78 | (long long unsigned int) le64_to_cpu(mmp->mmp_time), | |
79 | mmp->mmp_nodename, mmp->mmp_bdevname); | |
80 | } | |
81 | ||
82 | /* | |
83 | * kmmpd will update the MMP sequence every s_mmp_update_interval seconds | |
84 | */ | |
85 | static int kmmpd(void *data) | |
86 | { | |
87 | struct super_block *sb = ((struct mmpd_data *) data)->sb; | |
88 | struct buffer_head *bh = ((struct mmpd_data *) data)->bh; | |
89 | struct ext4_super_block *es = EXT4_SB(sb)->s_es; | |
90 | struct mmp_struct *mmp; | |
91 | ext4_fsblk_t mmp_block; | |
92 | u32 seq = 0; | |
93 | unsigned long failed_writes = 0; | |
94 | int mmp_update_interval = le16_to_cpu(es->s_mmp_update_interval); | |
95 | unsigned mmp_check_interval; | |
96 | unsigned long last_update_time; | |
97 | unsigned long diff; | |
98 | int retval; | |
99 | ||
100 | mmp_block = le64_to_cpu(es->s_mmp_block); | |
101 | mmp = (struct mmp_struct *)(bh->b_data); | |
102 | mmp->mmp_time = cpu_to_le64(get_seconds()); | |
103 | /* | |
104 | * Start with the higher mmp_check_interval and reduce it if | |
105 | * the MMP block is being updated on time. | |
106 | */ | |
107 | mmp_check_interval = max(EXT4_MMP_CHECK_MULT * mmp_update_interval, | |
108 | EXT4_MMP_MIN_CHECK_INTERVAL); | |
109 | mmp->mmp_check_interval = cpu_to_le16(mmp_check_interval); | |
110 | bdevname(bh->b_bdev, mmp->mmp_bdevname); | |
111 | ||
215fc6af | 112 | memcpy(mmp->mmp_nodename, init_utsname()->nodename, |
c5e06d10 JL |
113 | sizeof(mmp->mmp_nodename)); |
114 | ||
115 | while (!kthread_should_stop()) { | |
116 | if (++seq > EXT4_MMP_SEQ_MAX) | |
117 | seq = 1; | |
118 | ||
119 | mmp->mmp_seq = cpu_to_le32(seq); | |
120 | mmp->mmp_time = cpu_to_le64(get_seconds()); | |
121 | last_update_time = jiffies; | |
122 | ||
123 | retval = write_mmp_block(bh); | |
124 | /* | |
125 | * Don't spew too many error messages. Print one every | |
126 | * (s_mmp_update_interval * 60) seconds. | |
127 | */ | |
bdfc230f NA |
128 | if (retval) { |
129 | if ((failed_writes % 60) == 0) | |
130 | ext4_error(sb, "Error writing to MMP block"); | |
c5e06d10 JL |
131 | failed_writes++; |
132 | } | |
133 | ||
134 | if (!(le32_to_cpu(es->s_feature_incompat) & | |
135 | EXT4_FEATURE_INCOMPAT_MMP)) { | |
136 | ext4_warning(sb, "kmmpd being stopped since MMP feature" | |
137 | " has been disabled."); | |
138 | EXT4_SB(sb)->s_mmp_tsk = NULL; | |
139 | goto failed; | |
140 | } | |
141 | ||
142 | if (sb->s_flags & MS_RDONLY) { | |
143 | ext4_warning(sb, "kmmpd being stopped since filesystem " | |
144 | "has been remounted as readonly."); | |
145 | EXT4_SB(sb)->s_mmp_tsk = NULL; | |
146 | goto failed; | |
147 | } | |
148 | ||
149 | diff = jiffies - last_update_time; | |
150 | if (diff < mmp_update_interval * HZ) | |
151 | schedule_timeout_interruptible(mmp_update_interval * | |
152 | HZ - diff); | |
153 | ||
154 | /* | |
155 | * We need to make sure that more than mmp_check_interval | |
156 | * seconds have not passed since writing. If that has happened | |
157 | * we need to check if the MMP block is as we left it. | |
158 | */ | |
159 | diff = jiffies - last_update_time; | |
160 | if (diff > mmp_check_interval * HZ) { | |
161 | struct buffer_head *bh_check = NULL; | |
162 | struct mmp_struct *mmp_check; | |
163 | ||
164 | retval = read_mmp_block(sb, &bh_check, mmp_block); | |
165 | if (retval) { | |
166 | ext4_error(sb, "error reading MMP data: %d", | |
167 | retval); | |
168 | ||
169 | EXT4_SB(sb)->s_mmp_tsk = NULL; | |
170 | goto failed; | |
171 | } | |
172 | ||
173 | mmp_check = (struct mmp_struct *)(bh_check->b_data); | |
174 | if (mmp->mmp_seq != mmp_check->mmp_seq || | |
175 | memcmp(mmp->mmp_nodename, mmp_check->mmp_nodename, | |
176 | sizeof(mmp->mmp_nodename))) { | |
177 | dump_mmp_msg(sb, mmp_check, | |
178 | "Error while updating MMP info. " | |
179 | "The filesystem seems to have been" | |
180 | " multiply mounted."); | |
181 | ext4_error(sb, "abort"); | |
182 | goto failed; | |
183 | } | |
184 | put_bh(bh_check); | |
185 | } | |
186 | ||
187 | /* | |
188 | * Adjust the mmp_check_interval depending on how much time | |
189 | * it took for the MMP block to be written. | |
190 | */ | |
191 | mmp_check_interval = max(min(EXT4_MMP_CHECK_MULT * diff / HZ, | |
192 | EXT4_MMP_MAX_CHECK_INTERVAL), | |
193 | EXT4_MMP_MIN_CHECK_INTERVAL); | |
194 | mmp->mmp_check_interval = cpu_to_le16(mmp_check_interval); | |
195 | } | |
196 | ||
197 | /* | |
198 | * Unmount seems to be clean. | |
199 | */ | |
200 | mmp->mmp_seq = cpu_to_le32(EXT4_MMP_SEQ_CLEAN); | |
201 | mmp->mmp_time = cpu_to_le64(get_seconds()); | |
202 | ||
203 | retval = write_mmp_block(bh); | |
204 | ||
205 | failed: | |
206 | kfree(data); | |
207 | brelse(bh); | |
208 | return retval; | |
209 | } | |
210 | ||
211 | /* | |
212 | * Get a random new sequence number but make sure it is not greater than | |
213 | * EXT4_MMP_SEQ_MAX. | |
214 | */ | |
215 | static unsigned int mmp_new_seq(void) | |
216 | { | |
217 | u32 new_seq; | |
218 | ||
219 | do { | |
220 | get_random_bytes(&new_seq, sizeof(u32)); | |
221 | } while (new_seq > EXT4_MMP_SEQ_MAX); | |
222 | ||
223 | return new_seq; | |
224 | } | |
225 | ||
226 | /* | |
227 | * Protect the filesystem from being mounted more than once. | |
228 | */ | |
229 | int ext4_multi_mount_protect(struct super_block *sb, | |
230 | ext4_fsblk_t mmp_block) | |
231 | { | |
232 | struct ext4_super_block *es = EXT4_SB(sb)->s_es; | |
233 | struct buffer_head *bh = NULL; | |
234 | struct mmp_struct *mmp = NULL; | |
235 | struct mmpd_data *mmpd_data; | |
236 | u32 seq; | |
237 | unsigned int mmp_check_interval = le16_to_cpu(es->s_mmp_update_interval); | |
238 | unsigned int wait_time = 0; | |
239 | int retval; | |
240 | ||
241 | if (mmp_block < le32_to_cpu(es->s_first_data_block) || | |
242 | mmp_block >= ext4_blocks_count(es)) { | |
243 | ext4_warning(sb, "Invalid MMP block in superblock"); | |
244 | goto failed; | |
245 | } | |
246 | ||
247 | retval = read_mmp_block(sb, &bh, mmp_block); | |
248 | if (retval) | |
249 | goto failed; | |
250 | ||
251 | mmp = (struct mmp_struct *)(bh->b_data); | |
252 | ||
253 | if (mmp_check_interval < EXT4_MMP_MIN_CHECK_INTERVAL) | |
254 | mmp_check_interval = EXT4_MMP_MIN_CHECK_INTERVAL; | |
255 | ||
256 | /* | |
257 | * If check_interval in MMP block is larger, use that instead of | |
258 | * update_interval from the superblock. | |
259 | */ | |
260 | if (mmp->mmp_check_interval > mmp_check_interval) | |
261 | mmp_check_interval = mmp->mmp_check_interval; | |
262 | ||
263 | seq = le32_to_cpu(mmp->mmp_seq); | |
264 | if (seq == EXT4_MMP_SEQ_CLEAN) | |
265 | goto skip; | |
266 | ||
267 | if (seq == EXT4_MMP_SEQ_FSCK) { | |
268 | dump_mmp_msg(sb, mmp, "fsck is running on the filesystem"); | |
269 | goto failed; | |
270 | } | |
271 | ||
272 | wait_time = min(mmp_check_interval * 2 + 1, | |
273 | mmp_check_interval + 60); | |
274 | ||
275 | /* Print MMP interval if more than 20 secs. */ | |
276 | if (wait_time > EXT4_MMP_MIN_CHECK_INTERVAL * 4) | |
277 | ext4_warning(sb, "MMP interval %u higher than expected, please" | |
278 | " wait.\n", wait_time * 2); | |
279 | ||
280 | if (schedule_timeout_interruptible(HZ * wait_time) != 0) { | |
281 | ext4_warning(sb, "MMP startup interrupted, failing mount\n"); | |
282 | goto failed; | |
283 | } | |
284 | ||
285 | retval = read_mmp_block(sb, &bh, mmp_block); | |
286 | if (retval) | |
287 | goto failed; | |
288 | mmp = (struct mmp_struct *)(bh->b_data); | |
289 | if (seq != le32_to_cpu(mmp->mmp_seq)) { | |
290 | dump_mmp_msg(sb, mmp, | |
291 | "Device is already active on another node."); | |
292 | goto failed; | |
293 | } | |
294 | ||
295 | skip: | |
296 | /* | |
297 | * write a new random sequence number. | |
298 | */ | |
f6f96fdb DW |
299 | seq = mmp_new_seq(); |
300 | mmp->mmp_seq = cpu_to_le32(seq); | |
c5e06d10 JL |
301 | |
302 | retval = write_mmp_block(bh); | |
303 | if (retval) | |
304 | goto failed; | |
305 | ||
306 | /* | |
307 | * wait for MMP interval and check mmp_seq. | |
308 | */ | |
309 | if (schedule_timeout_interruptible(HZ * wait_time) != 0) { | |
310 | ext4_warning(sb, "MMP startup interrupted, failing mount\n"); | |
311 | goto failed; | |
312 | } | |
313 | ||
314 | retval = read_mmp_block(sb, &bh, mmp_block); | |
315 | if (retval) | |
316 | goto failed; | |
317 | mmp = (struct mmp_struct *)(bh->b_data); | |
318 | if (seq != le32_to_cpu(mmp->mmp_seq)) { | |
319 | dump_mmp_msg(sb, mmp, | |
320 | "Device is already active on another node."); | |
321 | goto failed; | |
322 | } | |
323 | ||
324 | mmpd_data = kmalloc(sizeof(struct mmpd_data), GFP_KERNEL); | |
325 | if (!mmpd_data) { | |
326 | ext4_warning(sb, "not enough memory for mmpd_data"); | |
327 | goto failed; | |
328 | } | |
329 | mmpd_data->sb = sb; | |
330 | mmpd_data->bh = bh; | |
331 | ||
332 | /* | |
333 | * Start a kernel thread to update the MMP block periodically. | |
334 | */ | |
335 | EXT4_SB(sb)->s_mmp_tsk = kthread_run(kmmpd, mmpd_data, "kmmpd-%s", | |
336 | bdevname(bh->b_bdev, | |
337 | mmp->mmp_bdevname)); | |
338 | if (IS_ERR(EXT4_SB(sb)->s_mmp_tsk)) { | |
339 | EXT4_SB(sb)->s_mmp_tsk = NULL; | |
340 | kfree(mmpd_data); | |
341 | ext4_warning(sb, "Unable to create kmmpd thread for %s.", | |
342 | sb->s_id); | |
343 | goto failed; | |
344 | } | |
345 | ||
346 | return 0; | |
347 | ||
348 | failed: | |
349 | brelse(bh); | |
350 | return 1; | |
351 | } | |
352 | ||
353 |