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 | ||
112 | memcpy(mmp->mmp_nodename, init_utsname()->sysname, | |
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 | */ | |
128 | if (retval && (failed_writes % 60) == 0) { | |
129 | ext4_error(sb, "Error writing to MMP block"); | |
130 | failed_writes++; | |
131 | } | |
132 | ||
133 | if (!(le32_to_cpu(es->s_feature_incompat) & | |
134 | EXT4_FEATURE_INCOMPAT_MMP)) { | |
135 | ext4_warning(sb, "kmmpd being stopped since MMP feature" | |
136 | " has been disabled."); | |
137 | EXT4_SB(sb)->s_mmp_tsk = NULL; | |
138 | goto failed; | |
139 | } | |
140 | ||
141 | if (sb->s_flags & MS_RDONLY) { | |
142 | ext4_warning(sb, "kmmpd being stopped since filesystem " | |
143 | "has been remounted as readonly."); | |
144 | EXT4_SB(sb)->s_mmp_tsk = NULL; | |
145 | goto failed; | |
146 | } | |
147 | ||
148 | diff = jiffies - last_update_time; | |
149 | if (diff < mmp_update_interval * HZ) | |
150 | schedule_timeout_interruptible(mmp_update_interval * | |
151 | HZ - diff); | |
152 | ||
153 | /* | |
154 | * We need to make sure that more than mmp_check_interval | |
155 | * seconds have not passed since writing. If that has happened | |
156 | * we need to check if the MMP block is as we left it. | |
157 | */ | |
158 | diff = jiffies - last_update_time; | |
159 | if (diff > mmp_check_interval * HZ) { | |
160 | struct buffer_head *bh_check = NULL; | |
161 | struct mmp_struct *mmp_check; | |
162 | ||
163 | retval = read_mmp_block(sb, &bh_check, mmp_block); | |
164 | if (retval) { | |
165 | ext4_error(sb, "error reading MMP data: %d", | |
166 | retval); | |
167 | ||
168 | EXT4_SB(sb)->s_mmp_tsk = NULL; | |
169 | goto failed; | |
170 | } | |
171 | ||
172 | mmp_check = (struct mmp_struct *)(bh_check->b_data); | |
173 | if (mmp->mmp_seq != mmp_check->mmp_seq || | |
174 | memcmp(mmp->mmp_nodename, mmp_check->mmp_nodename, | |
175 | sizeof(mmp->mmp_nodename))) { | |
176 | dump_mmp_msg(sb, mmp_check, | |
177 | "Error while updating MMP info. " | |
178 | "The filesystem seems to have been" | |
179 | " multiply mounted."); | |
180 | ext4_error(sb, "abort"); | |
181 | goto failed; | |
182 | } | |
183 | put_bh(bh_check); | |
184 | } | |
185 | ||
186 | /* | |
187 | * Adjust the mmp_check_interval depending on how much time | |
188 | * it took for the MMP block to be written. | |
189 | */ | |
190 | mmp_check_interval = max(min(EXT4_MMP_CHECK_MULT * diff / HZ, | |
191 | EXT4_MMP_MAX_CHECK_INTERVAL), | |
192 | EXT4_MMP_MIN_CHECK_INTERVAL); | |
193 | mmp->mmp_check_interval = cpu_to_le16(mmp_check_interval); | |
194 | } | |
195 | ||
196 | /* | |
197 | * Unmount seems to be clean. | |
198 | */ | |
199 | mmp->mmp_seq = cpu_to_le32(EXT4_MMP_SEQ_CLEAN); | |
200 | mmp->mmp_time = cpu_to_le64(get_seconds()); | |
201 | ||
202 | retval = write_mmp_block(bh); | |
203 | ||
204 | failed: | |
205 | kfree(data); | |
206 | brelse(bh); | |
207 | return retval; | |
208 | } | |
209 | ||
210 | /* | |
211 | * Get a random new sequence number but make sure it is not greater than | |
212 | * EXT4_MMP_SEQ_MAX. | |
213 | */ | |
214 | static unsigned int mmp_new_seq(void) | |
215 | { | |
216 | u32 new_seq; | |
217 | ||
218 | do { | |
219 | get_random_bytes(&new_seq, sizeof(u32)); | |
220 | } while (new_seq > EXT4_MMP_SEQ_MAX); | |
221 | ||
222 | return new_seq; | |
223 | } | |
224 | ||
225 | /* | |
226 | * Protect the filesystem from being mounted more than once. | |
227 | */ | |
228 | int ext4_multi_mount_protect(struct super_block *sb, | |
229 | ext4_fsblk_t mmp_block) | |
230 | { | |
231 | struct ext4_super_block *es = EXT4_SB(sb)->s_es; | |
232 | struct buffer_head *bh = NULL; | |
233 | struct mmp_struct *mmp = NULL; | |
234 | struct mmpd_data *mmpd_data; | |
235 | u32 seq; | |
236 | unsigned int mmp_check_interval = le16_to_cpu(es->s_mmp_update_interval); | |
237 | unsigned int wait_time = 0; | |
238 | int retval; | |
239 | ||
240 | if (mmp_block < le32_to_cpu(es->s_first_data_block) || | |
241 | mmp_block >= ext4_blocks_count(es)) { | |
242 | ext4_warning(sb, "Invalid MMP block in superblock"); | |
243 | goto failed; | |
244 | } | |
245 | ||
246 | retval = read_mmp_block(sb, &bh, mmp_block); | |
247 | if (retval) | |
248 | goto failed; | |
249 | ||
250 | mmp = (struct mmp_struct *)(bh->b_data); | |
251 | ||
252 | if (mmp_check_interval < EXT4_MMP_MIN_CHECK_INTERVAL) | |
253 | mmp_check_interval = EXT4_MMP_MIN_CHECK_INTERVAL; | |
254 | ||
255 | /* | |
256 | * If check_interval in MMP block is larger, use that instead of | |
257 | * update_interval from the superblock. | |
258 | */ | |
259 | if (mmp->mmp_check_interval > mmp_check_interval) | |
260 | mmp_check_interval = mmp->mmp_check_interval; | |
261 | ||
262 | seq = le32_to_cpu(mmp->mmp_seq); | |
263 | if (seq == EXT4_MMP_SEQ_CLEAN) | |
264 | goto skip; | |
265 | ||
266 | if (seq == EXT4_MMP_SEQ_FSCK) { | |
267 | dump_mmp_msg(sb, mmp, "fsck is running on the filesystem"); | |
268 | goto failed; | |
269 | } | |
270 | ||
271 | wait_time = min(mmp_check_interval * 2 + 1, | |
272 | mmp_check_interval + 60); | |
273 | ||
274 | /* Print MMP interval if more than 20 secs. */ | |
275 | if (wait_time > EXT4_MMP_MIN_CHECK_INTERVAL * 4) | |
276 | ext4_warning(sb, "MMP interval %u higher than expected, please" | |
277 | " wait.\n", wait_time * 2); | |
278 | ||
279 | if (schedule_timeout_interruptible(HZ * wait_time) != 0) { | |
280 | ext4_warning(sb, "MMP startup interrupted, failing mount\n"); | |
281 | goto failed; | |
282 | } | |
283 | ||
284 | retval = read_mmp_block(sb, &bh, mmp_block); | |
285 | if (retval) | |
286 | goto failed; | |
287 | mmp = (struct mmp_struct *)(bh->b_data); | |
288 | if (seq != le32_to_cpu(mmp->mmp_seq)) { | |
289 | dump_mmp_msg(sb, mmp, | |
290 | "Device is already active on another node."); | |
291 | goto failed; | |
292 | } | |
293 | ||
294 | skip: | |
295 | /* | |
296 | * write a new random sequence number. | |
297 | */ | |
298 | mmp->mmp_seq = seq = cpu_to_le32(mmp_new_seq()); | |
299 | ||
300 | retval = write_mmp_block(bh); | |
301 | if (retval) | |
302 | goto failed; | |
303 | ||
304 | /* | |
305 | * wait for MMP interval and check mmp_seq. | |
306 | */ | |
307 | if (schedule_timeout_interruptible(HZ * wait_time) != 0) { | |
308 | ext4_warning(sb, "MMP startup interrupted, failing mount\n"); | |
309 | goto failed; | |
310 | } | |
311 | ||
312 | retval = read_mmp_block(sb, &bh, mmp_block); | |
313 | if (retval) | |
314 | goto failed; | |
315 | mmp = (struct mmp_struct *)(bh->b_data); | |
316 | if (seq != le32_to_cpu(mmp->mmp_seq)) { | |
317 | dump_mmp_msg(sb, mmp, | |
318 | "Device is already active on another node."); | |
319 | goto failed; | |
320 | } | |
321 | ||
322 | mmpd_data = kmalloc(sizeof(struct mmpd_data), GFP_KERNEL); | |
323 | if (!mmpd_data) { | |
324 | ext4_warning(sb, "not enough memory for mmpd_data"); | |
325 | goto failed; | |
326 | } | |
327 | mmpd_data->sb = sb; | |
328 | mmpd_data->bh = bh; | |
329 | ||
330 | /* | |
331 | * Start a kernel thread to update the MMP block periodically. | |
332 | */ | |
333 | EXT4_SB(sb)->s_mmp_tsk = kthread_run(kmmpd, mmpd_data, "kmmpd-%s", | |
334 | bdevname(bh->b_bdev, | |
335 | mmp->mmp_bdevname)); | |
336 | if (IS_ERR(EXT4_SB(sb)->s_mmp_tsk)) { | |
337 | EXT4_SB(sb)->s_mmp_tsk = NULL; | |
338 | kfree(mmpd_data); | |
339 | ext4_warning(sb, "Unable to create kmmpd thread for %s.", | |
340 | sb->s_id); | |
341 | goto failed; | |
342 | } | |
343 | ||
344 | return 0; | |
345 | ||
346 | failed: | |
347 | brelse(bh); | |
348 | return 1; | |
349 | } | |
350 | ||
351 |