1 Index: linux-stage/fs/ext4/super.c
2 ===================================================================
3 --- linux-stage.orig/fs/ext4/super.c 2011-03-03 15:25:02.376539424 +0800
4 +++ linux-stage/fs/ext4/super.c 2011-03-05 12:24:02.918774335 +0800
6 #include <linux/log2.h>
7 #include <linux/crc16.h>
8 #include <asm/uaccess.h>
9 +#include <linux/kthread.h>
10 +#include <linux/utsname.h>
13 #include "ext4_jbd2.h"
15 invalidate_bdev(sbi->journal_bdev);
16 ext4_blkdev_remove(sbi);
19 + kthread_stop(sbi->s_mmp_tsk);
22 * Now that we are completely done shutting down the
28 + * Write the MMP block using WRITE_SYNC to try to get the block on-disk
31 +static int write_mmp_block(struct buffer_head *bh)
33 + mark_buffer_dirty(bh);
35 + bh->b_end_io = end_buffer_write_sync;
37 + submit_bh(WRITE_SYNC, bh);
39 + if (unlikely(!buffer_uptodate(bh)))
46 + * Read the MMP block. It _must_ be read from disk and hence we clear the
47 + * uptodate flag on the buffer.
49 +static int read_mmp_block(struct super_block *sb, struct buffer_head **bh,
50 + unsigned long mmp_block)
52 + struct mmp_struct *mmp;
55 + clear_buffer_uptodate(*bh);
60 + *bh = sb_bread(sb, mmp_block);
63 + *bh = sb_getblk(sb, mmp_block);
67 + (*bh)->b_end_io = end_buffer_read_sync;
68 + submit_bh(READ_SYNC, *bh);
69 + wait_on_buffer(*bh);
70 + if (!buffer_uptodate(*bh)) {
78 + "Error while reading MMP block %lu", mmp_block);
82 + mmp = (struct mmp_struct *)((*bh)->b_data);
83 + if (le32_to_cpu(mmp->mmp_magic) != EXT4_MMP_MAGIC)
90 + * Dump as much information as possible to help the admin.
92 +static void dump_mmp_msg(struct super_block *sb, struct mmp_struct *mmp,
93 + const char *function, const char *msg)
95 + __ext4_warning(sb, function, msg);
96 + __ext4_warning(sb, function, "MMP failure info: last update time: %llu, "
97 + "last update node: %s, last update device: %s\n",
98 + (long long unsigned int)le64_to_cpu(mmp->mmp_time),
99 + mmp->mmp_nodename, mmp->mmp_bdevname);
103 + * kmmpd will update the MMP sequence every s_mmp_update_interval seconds
105 +static int kmmpd(void *data)
107 + struct super_block *sb = (struct super_block *) data;
108 + struct ext4_super_block *es = EXT4_SB(sb)->s_es;
109 + struct buffer_head *bh = NULL;
110 + struct mmp_struct *mmp;
111 + unsigned long mmp_block;
113 + unsigned long failed_writes = 0;
114 + int mmp_update_interval = le16_to_cpu(es->s_mmp_update_interval);
115 + unsigned mmp_check_interval;
116 + unsigned long last_update_time;
117 + unsigned long diff;
120 + mmp_block = le64_to_cpu(es->s_mmp_block);
121 + retval = read_mmp_block(sb, &bh, mmp_block);
125 + mmp = (struct mmp_struct *)(bh->b_data);
126 + mmp->mmp_time = cpu_to_le64(get_seconds());
128 + * Start with the higher mmp_check_interval and reduce it if
129 + * the MMP block is being updated on time.
131 + mmp_check_interval = max(5 * mmp_update_interval,
132 + EXT4_MMP_MIN_CHECK_INTERVAL);
133 + mmp->mmp_check_interval = cpu_to_le16(mmp_check_interval);
134 + bdevname(bh->b_bdev, mmp->mmp_bdevname);
136 + memcpy(mmp->mmp_nodename, init_utsname()->sysname,
137 + sizeof(mmp->mmp_nodename));
139 + while (!kthread_should_stop()) {
140 + if (++seq > EXT4_MMP_SEQ_MAX)
143 + mmp->mmp_seq = cpu_to_le32(seq);
144 + mmp->mmp_time = cpu_to_le64(get_seconds());
145 + last_update_time = jiffies;
147 + retval = write_mmp_block(bh);
149 + * Don't spew too many error messages. Print one every
150 + * (s_mmp_update_interval * 60) seconds.
152 + if (retval && (failed_writes % 60) == 0) {
154 + "Error writing to MMP block");
158 + if (!(le32_to_cpu(es->s_feature_incompat) &
159 + EXT4_FEATURE_INCOMPAT_MMP)) {
160 + ext4_warning(sb, "kmmpd being stopped "
161 + "since MMP feature has been disabled.");
162 + EXT4_SB(sb)->s_mmp_tsk = 0;
166 + if (sb->s_flags & MS_RDONLY) {
167 + ext4_warning(sb, "kmmpd being stopped "
168 + "since filesystem has been remounted as "
170 + EXT4_SB(sb)->s_mmp_tsk = 0;
174 + diff = jiffies - last_update_time;
175 + if (diff < mmp_update_interval * HZ)
176 + schedule_timeout_interruptible(mmp_update_interval *
180 + * We need to make sure that more than mmp_check_interval
181 + * seconds have not passed since writing. If that has happened
182 + * we need to check if the MMP block is as we left it.
184 + diff = jiffies - last_update_time;
185 + if (diff > mmp_check_interval * HZ) {
186 + struct buffer_head *bh_check = NULL;
187 + struct mmp_struct *mmp_check;
189 + retval = read_mmp_block(sb, &bh_check, mmp_block);
191 + EXT4_SB(sb)->s_mmp_tsk = 0;
195 + mmp_check = (struct mmp_struct *)(bh_check->b_data);
196 + if (mmp->mmp_time != mmp_check->mmp_time ||
197 + memcmp(mmp->mmp_nodename, mmp_check->mmp_nodename,
198 + sizeof(mmp->mmp_nodename)))
199 + dump_mmp_msg(sb, mmp_check, __func__,
200 + "Error while updating MMP info. "
201 + "The filesystem seems to have "
202 + "been multiply mounted.");
208 + * Adjust the mmp_check_interval depending on how much time
209 + * it took for the MMP block to be written.
211 + mmp_check_interval = max(5 * diff / HZ,
212 + (unsigned long) EXT4_MMP_MIN_CHECK_INTERVAL);
213 + mmp->mmp_check_interval = cpu_to_le16(mmp_check_interval);
217 + * Unmount seems to be clean.
219 + mmp->mmp_seq = cpu_to_le32(EXT4_MMP_SEQ_CLEAN);
220 + mmp->mmp_time = cpu_to_le64(get_seconds());
222 + retval = write_mmp_block(bh);
230 + * Get a random new sequence number but make sure it is not greater than
231 + * EXT4_MMP_SEQ_MAX.
233 +static unsigned int mmp_new_seq(void)
238 + get_random_bytes(&new_seq, sizeof(u32));
239 + } while (new_seq > EXT4_MMP_SEQ_MAX);
245 + * Protect the filesystem from being mounted more than once.
247 +static int ext4_multi_mount_protect(struct super_block *sb,
248 + unsigned long mmp_block)
250 + struct ext4_super_block *es = EXT4_SB(sb)->s_es;
251 + struct buffer_head *bh = NULL;
252 + struct mmp_struct *mmp = NULL;
254 + unsigned int mmp_check_interval = le16_to_cpu(es->s_mmp_update_interval);
255 + unsigned int wait_time = 0;
258 + if (mmp_block < le32_to_cpu(es->s_first_data_block) ||
259 + mmp_block >= ext4_blocks_count(es)) {
261 + "Invalid MMP block in superblock");
265 + retval = read_mmp_block(sb, &bh, mmp_block);
269 + mmp = (struct mmp_struct *)(bh->b_data);
271 + if (mmp_check_interval < EXT4_MMP_MIN_CHECK_INTERVAL)
272 + mmp_check_interval = EXT4_MMP_MIN_CHECK_INTERVAL;
275 + * If check_interval in MMP block is larger, use that instead of
276 + * update_interval from the superblock.
278 + if (mmp->mmp_check_interval > mmp_check_interval)
279 + mmp_check_interval = mmp->mmp_check_interval;
281 + seq = le32_to_cpu(mmp->mmp_seq);
282 + if (seq == EXT4_MMP_SEQ_CLEAN)
285 + if (seq == EXT4_MMP_SEQ_FSCK) {
286 + dump_mmp_msg(sb, mmp, __func__,
287 + "fsck is running on the filesystem");
291 + wait_time = min(mmp_check_interval * 2 + 1,
292 + mmp_check_interval + 60);
294 + /* Print MMP interval if more than 20 secs. */
295 + if (wait_time > EXT4_MMP_MIN_CHECK_INTERVAL * 4)
296 + ext4_warning(sb, "MMP interval %u higher than "
297 + "expected, please wait.\n", wait_time * 2);
299 + if (schedule_timeout_interruptible(HZ * wait_time) != 0) {
300 + ext4_warning(sb, "MMP startup interrupted, failing "
305 + retval = read_mmp_block(sb, &bh, mmp_block);
308 + mmp = (struct mmp_struct *)(bh->b_data);
309 + if (seq != le32_to_cpu(mmp->mmp_seq)) {
310 + dump_mmp_msg(sb, mmp, __func__,
311 + "Device is already active on another node.");
317 + * write a new random sequence number.
319 + mmp->mmp_seq = seq = cpu_to_le32(mmp_new_seq());
321 + retval = write_mmp_block(bh);
326 + * wait for MMP interval and check mmp_seq.
328 + if (schedule_timeout_interruptible(HZ * wait_time) != 0) {
329 + ext4_warning(sb, "MMP startup interrupted, failing "
334 + retval = read_mmp_block(sb, &bh, mmp_block);
337 + mmp = (struct mmp_struct *)(bh->b_data);
338 + if (seq != le32_to_cpu(mmp->mmp_seq)) {
339 + dump_mmp_msg(sb, mmp, __func__,
340 + "Device is already active on another node.");
345 + * Start a kernel thread to update the MMP block periodically.
347 + EXT4_SB(sb)->s_mmp_tsk = kthread_run(kmmpd, sb, "kmmpd-%02x:%02x",
350 + if (IS_ERR(EXT4_SB(sb)->s_mmp_tsk)) {
351 + EXT4_SB(sb)->s_mmp_tsk = 0;
352 + ext4_warning(sb, "Unable to create kmmpd thread "
353 + "for %s.", sb->s_id);
365 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
366 u64 ino, u32 generation)
368 @@ -2816,6 +3158,11 @@
369 EXT4_HAS_INCOMPAT_FEATURE(sb,
370 EXT4_FEATURE_INCOMPAT_RECOVER));
372 + if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
373 + !(sb->s_flags & MS_RDONLY))
374 + if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
375 + goto failed_mount3;
378 * The first inode we look at is the journal inode. Don't try
379 * root first: it may be modified in the journal!
380 @@ -3052,6 +3399,8 @@
381 percpu_counter_destroy(&sbi->s_freeinodes_counter);
382 percpu_counter_destroy(&sbi->s_dirs_counter);
383 percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
384 + if (sbi->s_mmp_tsk)
385 + kthread_stop(sbi->s_mmp_tsk);
387 for (i = 0; i < db_count; i++)
388 brelse(sbi->s_group_desc[i]);
389 @@ -3560,7 +3909,7 @@
390 struct ext4_mount_options old_opts;
392 unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
398 @@ -3682,6 +4031,13 @@
400 if (!ext4_setup_super(sb, es, 0))
401 sb->s_flags &= ~MS_RDONLY;
402 + if (EXT4_HAS_INCOMPAT_FEATURE(sb,
403 + EXT4_FEATURE_INCOMPAT_MMP))
404 + if (ext4_multi_mount_protect(sb,
405 + le64_to_cpu(es->s_mmp_block))) {
411 ext4_setup_system_zone(sb);
412 Index: linux-stage/fs/ext4/ext4.h
413 ===================================================================
414 --- linux-stage.orig/fs/ext4/ext4.h 2011-03-03 15:25:02.507538421 +0800
415 +++ linux-stage/fs/ext4/ext4.h 2011-03-05 12:25:16.343986732 +0800
417 __le16 s_want_extra_isize; /* New inodes should reserve # bytes */
418 __le32 s_flags; /* Miscellaneous flags */
419 __le16 s_raid_stride; /* RAID stride */
420 - __le16 s_mmp_interval; /* # seconds to wait in MMP checking */
421 + __le16 s_mmp_update_interval; /* # seconds to wait in MMP checking */
422 __le64 s_mmp_block; /* Block for multi-mount protection */
423 __le32 s_raid_stripe_width; /* blocks on all data disks (N*stride)*/
424 __u8 s_log_groups_per_flex; /* FLEX_BG group size */
425 @@ -1041,6 +1041,9 @@
427 /* workqueue for dio unwritten */
428 struct workqueue_struct *dio_unwritten_wq;
430 + /* Kernel thread for multiple mount protection */
431 + struct task_struct *s_mmp_tsk;
434 static inline struct ext4_sb_info *EXT4_SB(struct super_block *sb)
435 @@ -1177,7 +1180,8 @@
436 EXT4_FEATURE_INCOMPAT_META_BG| \
437 EXT4_FEATURE_INCOMPAT_EXTENTS| \
438 EXT4_FEATURE_INCOMPAT_64BIT| \
439 - EXT4_FEATURE_INCOMPAT_FLEX_BG)
440 + EXT4_FEATURE_INCOMPAT_FLEX_BG| \
441 + EXT4_FEATURE_INCOMPAT_MMP)
442 #define EXT4_FEATURE_RO_COMPAT_SUPP (EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER| \
443 EXT4_FEATURE_RO_COMPAT_LARGE_FILE| \
444 EXT4_FEATURE_RO_COMPAT_GDT_CSUM| \
445 @@ -1384,6 +1388,34 @@
446 extern struct proc_dir_entry *ext4_proc_root;
449 + * This structure will be used for multiple mount protection. It will be
450 + * written into the block number saved in the s_mmp_block field in the
451 + * superblock. Programs that check MMP should assume that if
452 + * SEQ_FSCK (or any unknown code above SEQ_MAX) is present then it is NOT safe
453 + * to use the filesystem, regardless of how old the timestamp is.
455 +#define EXT4_MMP_MAGIC 0x004D4D50U /* ASCII for MMP */
456 +#define EXT4_MMP_SEQ_CLEAN 0xFF4D4D50U /* mmp_seq value for clean unmount */
457 +#define EXT4_MMP_SEQ_FSCK 0xE24D4D50U /* mmp_seq value when being fscked */
458 +#define EXT4_MMP_SEQ_MAX 0xE24D4D4FU /* maximum valid mmp_seq value */
464 + char mmp_nodename[64];
465 + char mmp_bdevname[32];
466 + __le16 mmp_check_interval;
468 + __le32 mmp_pad2[227];
472 + * Minimum interval for MMP checking in seconds.
474 +#define EXT4_MMP_MIN_CHECK_INTERVAL 5
477 * Function prototypes