2 #include <linux/random.h>
3 #include <linux/buffer_head.h>
4 #include <linux/utsname.h>
5 #include <linux/kthread.h>
9 /* Checksumming functions */
10 static __le32
ext4_mmp_csum(struct super_block
*sb
, struct mmp_struct
*mmp
)
12 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
13 int offset
= offsetof(struct mmp_struct
, mmp_checksum
);
16 csum
= ext4_chksum(sbi
, sbi
->s_csum_seed
, (char *)mmp
, offset
);
18 return cpu_to_le32(csum
);
21 int ext4_mmp_csum_verify(struct super_block
*sb
, struct mmp_struct
*mmp
)
23 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
24 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
27 return mmp
->mmp_checksum
== ext4_mmp_csum(sb
, mmp
);
30 void ext4_mmp_csum_set(struct super_block
*sb
, struct mmp_struct
*mmp
)
32 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
33 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
36 mmp
->mmp_checksum
= ext4_mmp_csum(sb
, mmp
);
40 * Write the MMP block using WRITE_SYNC to try to get the block on-disk
43 static int write_mmp_block(struct super_block
*sb
, struct buffer_head
*bh
)
45 struct mmp_struct
*mmp
= (struct mmp_struct
*)(bh
->b_data
);
48 * We protect against freezing so that we don't create dirty buffers
49 * on frozen filesystem.
52 ext4_mmp_csum_set(sb
, mmp
);
53 mark_buffer_dirty(bh
);
55 bh
->b_end_io
= end_buffer_write_sync
;
57 #ifdef FEATURE_STORAGE_META_LOG
58 if( bh
&& bh
->b_bdev
&& bh
->b_bdev
->bd_disk
)
59 set_metadata_rw_status(bh
->b_bdev
->bd_disk
->first_minor
, WAIT_WRITE_CNT
);
61 submit_bh(WRITE_SYNC
| REQ_META
| REQ_PRIO
, bh
);
64 if (unlikely(!buffer_uptodate(bh
)))
71 * Read the MMP block. It _must_ be read from disk and hence we clear the
72 * uptodate flag on the buffer.
74 static int read_mmp_block(struct super_block
*sb
, struct buffer_head
**bh
,
75 ext4_fsblk_t mmp_block
)
77 struct mmp_struct
*mmp
;
80 clear_buffer_uptodate(*bh
);
82 /* This would be sb_bread(sb, mmp_block), except we need to be sure
83 * that the MD RAID device cache has been bypassed, and that the read
84 * is not blocked in the elevator. */
86 *bh
= sb_getblk(sb
, mmp_block
);
92 (*bh
)->b_end_io
= end_buffer_read_sync
;
93 #ifdef FEATURE_STORAGE_META_LOG
94 if( (*bh
) && (*bh
)->b_bdev
&& (*bh
)->b_bdev
->bd_disk
)
95 set_metadata_rw_status((*bh
)->b_bdev
->bd_disk
->first_minor
, WAIT_READ_CNT
);
97 submit_bh(READ_SYNC
| REQ_META
| REQ_PRIO
, *bh
);
99 if (!buffer_uptodate(*bh
)) {
104 if (unlikely(!*bh
)) {
105 ext4_warning(sb
, "Error while reading MMP block %llu",
110 mmp
= (struct mmp_struct
*)((*bh
)->b_data
);
111 if (le32_to_cpu(mmp
->mmp_magic
) != EXT4_MMP_MAGIC
||
112 !ext4_mmp_csum_verify(sb
, mmp
))
119 * Dump as much information as possible to help the admin.
121 void __dump_mmp_msg(struct super_block
*sb
, struct mmp_struct
*mmp
,
122 const char *function
, unsigned int line
, const char *msg
)
124 __ext4_warning(sb
, function
, line
, msg
);
125 __ext4_warning(sb
, function
, line
,
126 "MMP failure info: last update time: %llu, last update "
127 "node: %s, last update device: %s\n",
128 (long long unsigned int) le64_to_cpu(mmp
->mmp_time
),
129 mmp
->mmp_nodename
, mmp
->mmp_bdevname
);
133 * kmmpd will update the MMP sequence every s_mmp_update_interval seconds
135 static int kmmpd(void *data
)
137 struct super_block
*sb
= ((struct mmpd_data
*) data
)->sb
;
138 struct buffer_head
*bh
= ((struct mmpd_data
*) data
)->bh
;
139 struct ext4_super_block
*es
= EXT4_SB(sb
)->s_es
;
140 struct mmp_struct
*mmp
;
141 ext4_fsblk_t mmp_block
;
143 unsigned long failed_writes
= 0;
144 int mmp_update_interval
= le16_to_cpu(es
->s_mmp_update_interval
);
145 unsigned mmp_check_interval
;
146 unsigned long last_update_time
;
150 mmp_block
= le64_to_cpu(es
->s_mmp_block
);
151 mmp
= (struct mmp_struct
*)(bh
->b_data
);
152 mmp
->mmp_time
= cpu_to_le64(get_seconds());
154 * Start with the higher mmp_check_interval and reduce it if
155 * the MMP block is being updated on time.
157 mmp_check_interval
= max(EXT4_MMP_CHECK_MULT
* mmp_update_interval
,
158 EXT4_MMP_MIN_CHECK_INTERVAL
);
159 mmp
->mmp_check_interval
= cpu_to_le16(mmp_check_interval
);
160 bdevname(bh
->b_bdev
, mmp
->mmp_bdevname
);
162 memcpy(mmp
->mmp_nodename
, init_utsname()->nodename
,
163 sizeof(mmp
->mmp_nodename
));
165 while (!kthread_should_stop()) {
166 if (++seq
> EXT4_MMP_SEQ_MAX
)
169 mmp
->mmp_seq
= cpu_to_le32(seq
);
170 mmp
->mmp_time
= cpu_to_le64(get_seconds());
171 last_update_time
= jiffies
;
173 retval
= write_mmp_block(sb
, bh
);
175 * Don't spew too many error messages. Print one every
176 * (s_mmp_update_interval * 60) seconds.
179 if ((failed_writes
% 60) == 0)
180 ext4_error(sb
, "Error writing to MMP block");
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 EXT4_SB(sb
)->s_mmp_tsk
= NULL
;
192 if (sb
->s_flags
& MS_RDONLY
) {
193 ext4_warning(sb
, "kmmpd being stopped since filesystem "
194 "has been remounted as readonly.");
195 EXT4_SB(sb
)->s_mmp_tsk
= NULL
;
199 diff
= jiffies
- last_update_time
;
200 if (diff
< mmp_update_interval
* HZ
)
201 schedule_timeout_interruptible(mmp_update_interval
*
205 * We need to make sure that more than mmp_check_interval
206 * seconds have not passed since writing. If that has happened
207 * we need to check if the MMP block is as we left it.
209 diff
= jiffies
- last_update_time
;
210 if (diff
> mmp_check_interval
* HZ
) {
211 struct buffer_head
*bh_check
= NULL
;
212 struct mmp_struct
*mmp_check
;
214 retval
= read_mmp_block(sb
, &bh_check
, mmp_block
);
216 ext4_error(sb
, "error reading MMP data: %d",
219 EXT4_SB(sb
)->s_mmp_tsk
= NULL
;
223 mmp_check
= (struct mmp_struct
*)(bh_check
->b_data
);
224 if (mmp
->mmp_seq
!= mmp_check
->mmp_seq
||
225 memcmp(mmp
->mmp_nodename
, mmp_check
->mmp_nodename
,
226 sizeof(mmp
->mmp_nodename
))) {
227 dump_mmp_msg(sb
, mmp_check
,
228 "Error while updating MMP info. "
229 "The filesystem seems to have been"
230 " multiply mounted.");
231 ext4_error(sb
, "abort");
238 * Adjust the mmp_check_interval depending on how much time
239 * it took for the MMP block to be written.
241 mmp_check_interval
= max(min(EXT4_MMP_CHECK_MULT
* diff
/ HZ
,
242 EXT4_MMP_MAX_CHECK_INTERVAL
),
243 EXT4_MMP_MIN_CHECK_INTERVAL
);
244 mmp
->mmp_check_interval
= cpu_to_le16(mmp_check_interval
);
248 * Unmount seems to be clean.
250 mmp
->mmp_seq
= cpu_to_le32(EXT4_MMP_SEQ_CLEAN
);
251 mmp
->mmp_time
= cpu_to_le64(get_seconds());
253 retval
= write_mmp_block(sb
, bh
);
262 * Get a random new sequence number but make sure it is not greater than
265 static unsigned int mmp_new_seq(void)
270 get_random_bytes(&new_seq
, sizeof(u32
));
271 } while (new_seq
> EXT4_MMP_SEQ_MAX
);
277 * Protect the filesystem from being mounted more than once.
279 int ext4_multi_mount_protect(struct super_block
*sb
,
280 ext4_fsblk_t mmp_block
)
282 struct ext4_super_block
*es
= EXT4_SB(sb
)->s_es
;
283 struct buffer_head
*bh
= NULL
;
284 struct mmp_struct
*mmp
= NULL
;
285 struct mmpd_data
*mmpd_data
;
287 unsigned int mmp_check_interval
= le16_to_cpu(es
->s_mmp_update_interval
);
288 unsigned int wait_time
= 0;
291 if (mmp_block
< le32_to_cpu(es
->s_first_data_block
) ||
292 mmp_block
>= ext4_blocks_count(es
)) {
293 ext4_warning(sb
, "Invalid MMP block in superblock");
297 retval
= read_mmp_block(sb
, &bh
, mmp_block
);
301 mmp
= (struct mmp_struct
*)(bh
->b_data
);
303 if (mmp_check_interval
< EXT4_MMP_MIN_CHECK_INTERVAL
)
304 mmp_check_interval
= EXT4_MMP_MIN_CHECK_INTERVAL
;
307 * If check_interval in MMP block is larger, use that instead of
308 * update_interval from the superblock.
310 if (le16_to_cpu(mmp
->mmp_check_interval
) > mmp_check_interval
)
311 mmp_check_interval
= le16_to_cpu(mmp
->mmp_check_interval
);
313 seq
= le32_to_cpu(mmp
->mmp_seq
);
314 if (seq
== EXT4_MMP_SEQ_CLEAN
)
317 if (seq
== EXT4_MMP_SEQ_FSCK
) {
318 dump_mmp_msg(sb
, mmp
, "fsck is running on the filesystem");
322 wait_time
= min(mmp_check_interval
* 2 + 1,
323 mmp_check_interval
+ 60);
325 /* Print MMP interval if more than 20 secs. */
326 if (wait_time
> EXT4_MMP_MIN_CHECK_INTERVAL
* 4)
327 ext4_warning(sb
, "MMP interval %u higher than expected, please"
328 " wait.\n", wait_time
* 2);
330 if (schedule_timeout_interruptible(HZ
* wait_time
) != 0) {
331 ext4_warning(sb
, "MMP startup interrupted, failing mount\n");
335 retval
= read_mmp_block(sb
, &bh
, mmp_block
);
338 mmp
= (struct mmp_struct
*)(bh
->b_data
);
339 if (seq
!= le32_to_cpu(mmp
->mmp_seq
)) {
340 dump_mmp_msg(sb
, mmp
,
341 "Device is already active on another node.");
347 * write a new random sequence number.
350 mmp
->mmp_seq
= cpu_to_le32(seq
);
352 retval
= write_mmp_block(sb
, bh
);
357 * wait for MMP interval and check mmp_seq.
359 if (schedule_timeout_interruptible(HZ
* wait_time
) != 0) {
360 ext4_warning(sb
, "MMP startup interrupted, failing mount\n");
364 retval
= read_mmp_block(sb
, &bh
, mmp_block
);
367 mmp
= (struct mmp_struct
*)(bh
->b_data
);
368 if (seq
!= le32_to_cpu(mmp
->mmp_seq
)) {
369 dump_mmp_msg(sb
, mmp
,
370 "Device is already active on another node.");
374 mmpd_data
= kmalloc(sizeof(struct mmpd_data
), GFP_KERNEL
);
376 ext4_warning(sb
, "not enough memory for mmpd_data");
383 * Start a kernel thread to update the MMP block periodically.
385 EXT4_SB(sb
)->s_mmp_tsk
= kthread_run(kmmpd
, mmpd_data
, "kmmpd-%s",
388 if (IS_ERR(EXT4_SB(sb
)->s_mmp_tsk
)) {
389 EXT4_SB(sb
)->s_mmp_tsk
= NULL
;
391 ext4_warning(sb
, "Unable to create kmmpd thread for %s.",