1 Index: linux-stage/fs/ext4/super.c
2 ===================================================================
3 --- linux-stage.orig/fs/ext4/super.c 2011-04-05 17:51:52.173385539 +0800
4 +++ linux-stage/fs/ext4/super.c 2011-04-05 17:53:58.686223570 +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 mmpd_data *) data)->sb;
108 + struct buffer_head *bh = ((struct mmpd_data *) data)->bh;
109 + struct ext4_super_block *es = EXT4_SB(sb)->s_es;
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 + mmp = (struct mmp_struct *)(bh->b_data);
122 + mmp->mmp_time = cpu_to_le64(get_seconds());
124 + * Start with the higher mmp_check_interval and reduce it if
125 + * the MMP block is being updated on time.
127 + mmp_check_interval = max(5UL * mmp_update_interval,
128 + EXT4_MMP_MIN_CHECK_INTERVAL);
129 + mmp->mmp_check_interval = cpu_to_le16(mmp_check_interval);
130 + bdevname(bh->b_bdev, mmp->mmp_bdevname);
132 + memcpy(mmp->mmp_nodename, init_utsname()->sysname,
133 + sizeof(mmp->mmp_nodename));
135 + while (!kthread_should_stop()) {
136 + if (++seq > EXT4_MMP_SEQ_MAX)
139 + mmp->mmp_seq = cpu_to_le32(seq);
140 + mmp->mmp_time = cpu_to_le64(get_seconds());
141 + last_update_time = jiffies;
143 + retval = write_mmp_block(bh);
145 + * Don't spew too many error messages. Print one every
146 + * (s_mmp_update_interval * 60) seconds.
148 + if (retval && (failed_writes % 60) == 0) {
150 + "Error writing to MMP block");
154 + if (!(le32_to_cpu(es->s_feature_incompat) &
155 + EXT4_FEATURE_INCOMPAT_MMP)) {
156 + ext4_warning(sb, "kmmpd being stopped "
157 + "since MMP feature has been disabled.");
158 + EXT4_SB(sb)->s_mmp_tsk = NULL;
162 + if (sb->s_flags & MS_RDONLY) {
163 + ext4_warning(sb, "kmmpd being stopped "
164 + "since filesystem has been remounted as "
166 + EXT4_SB(sb)->s_mmp_tsk = NULL;
170 + diff = jiffies - last_update_time;
171 + if (diff < mmp_update_interval * HZ)
172 + schedule_timeout_interruptible(mmp_update_interval *
176 + * We need to make sure that more than mmp_check_interval
177 + * seconds have not passed since writing. If that has happened
178 + * we need to check if the MMP block is as we left it.
180 + diff = jiffies - last_update_time;
181 + if (diff > mmp_check_interval * HZ) {
182 + struct buffer_head *bh_check = NULL;
183 + struct mmp_struct *mmp_check;
185 + retval = read_mmp_block(sb, &bh_check, mmp_block);
187 + EXT4_SB(sb)->s_mmp_tsk = NULL;
188 + ext4_error(sb, "error reading MMP data: %d",
193 + mmp_check = (struct mmp_struct *)(bh_check->b_data);
194 + if (mmp->mmp_seq != mmp_check->mmp_seq ||
195 + memcmp(mmp->mmp_nodename, mmp_check->mmp_nodename,
196 + sizeof(mmp->mmp_nodename))) {
197 + dump_mmp_msg(sb, mmp_check, __func__,
198 + "Error while updating MMP info. "
199 + "The filesystem seems to have "
200 + "been multiply mounted.");
201 + ext4_error(sb, "abort");
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(min(5 * diff / HZ,
212 + EXT4_MMP_MAX_CHECK_INTERVAL),
213 + EXT4_MMP_MIN_CHECK_INTERVAL);
214 + mmp->mmp_check_interval = cpu_to_le16(mmp_check_interval);
218 + * Unmount seems to be clean.
220 + mmp->mmp_seq = cpu_to_le32(EXT4_MMP_SEQ_CLEAN);
221 + mmp->mmp_time = cpu_to_le64(get_seconds());
223 + retval = write_mmp_block(bh);
232 + * Get a random new sequence number but make sure it is not greater than
233 + * EXT4_MMP_SEQ_MAX.
235 +static unsigned int mmp_new_seq(void)
240 + get_random_bytes(&new_seq, sizeof(u32));
241 + } while (new_seq > EXT4_MMP_SEQ_MAX);
247 + * Protect the filesystem from being mounted more than once.
249 +static int ext4_multi_mount_protect(struct super_block *sb,
250 + unsigned long mmp_block)
252 + struct ext4_super_block *es = EXT4_SB(sb)->s_es;
253 + struct buffer_head *bh = NULL;
254 + struct mmp_struct *mmp = NULL;
255 + struct mmpd_data *mmpd_data;
257 + unsigned int mmp_check_interval = le16_to_cpu(es->s_mmp_update_interval);
258 + unsigned int wait_time = 0;
261 + if (mmp_block < le32_to_cpu(es->s_first_data_block) ||
262 + mmp_block >= ext4_blocks_count(es)) {
264 + "Invalid MMP block in superblock");
268 + retval = read_mmp_block(sb, &bh, mmp_block);
272 + mmp = (struct mmp_struct *)(bh->b_data);
274 + if (mmp_check_interval < EXT4_MMP_MIN_CHECK_INTERVAL)
275 + mmp_check_interval = EXT4_MMP_MIN_CHECK_INTERVAL;
278 + * If check_interval in MMP block is larger, use that instead of
279 + * update_interval from the superblock.
281 + if (mmp->mmp_check_interval > mmp_check_interval)
282 + mmp_check_interval = mmp->mmp_check_interval;
284 + seq = le32_to_cpu(mmp->mmp_seq);
285 + if (seq == EXT4_MMP_SEQ_CLEAN)
288 + if (seq == EXT4_MMP_SEQ_FSCK) {
289 + dump_mmp_msg(sb, mmp, __func__,
290 + "fsck is running on the filesystem");
294 + wait_time = min(mmp_check_interval * 2 + 1,
295 + mmp_check_interval + 60);
297 + /* Print MMP interval if more than 20 secs. */
298 + if (wait_time > EXT4_MMP_MIN_CHECK_INTERVAL * 4)
299 + ext4_warning(sb, "MMP interval %u higher than "
300 + "expected, please wait.\n", wait_time * 2);
302 + if (schedule_timeout_interruptible(HZ * wait_time) != 0) {
303 + ext4_warning(sb, "MMP startup interrupted, failing "
308 + retval = read_mmp_block(sb, &bh, mmp_block);
311 + mmp = (struct mmp_struct *)(bh->b_data);
312 + if (seq != le32_to_cpu(mmp->mmp_seq)) {
313 + dump_mmp_msg(sb, mmp, __func__,
314 + "Device is already active on another node.");
320 + * write a new random sequence number.
322 + mmp->mmp_seq = seq = cpu_to_le32(mmp_new_seq());
324 + retval = write_mmp_block(bh);
329 + * wait for MMP interval and check mmp_seq.
331 + if (schedule_timeout_interruptible(HZ * wait_time) != 0) {
332 + ext4_warning(sb, "MMP startup interrupted, failing "
337 + retval = read_mmp_block(sb, &bh, mmp_block);
340 + mmp = (struct mmp_struct *)(bh->b_data);
341 + if (seq != le32_to_cpu(mmp->mmp_seq)) {
342 + dump_mmp_msg(sb, mmp, __func__,
343 + "Device is already active on another node.");
347 + mmpd_data = kmalloc(sizeof(struct mmpd_data *), GFP_KERNEL);
349 + ext4_warning(sb, "not enough memory for mmpd_data");
352 + mmpd_data->sb = sb;
353 + mmpd_data->bh = bh;
356 + * Start a kernel thread to update the MMP block periodically.
358 + EXT4_SB(sb)->s_mmp_tsk = kthread_run(kmmpd, mmpd_data, "kmmpd-%s",
359 + bdevname(bh->b_bdev,
360 + mmp->mmp_bdevname));
361 + if (IS_ERR(EXT4_SB(sb)->s_mmp_tsk)) {
362 + EXT4_SB(sb)->s_mmp_tsk = NULL;
364 + ext4_warning(sb, "Unable to create kmmpd thread "
365 + "for %s.", sb->s_id);
376 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
377 u64 ino, u32 generation)
379 @@ -2800,6 +3153,11 @@
380 EXT4_HAS_INCOMPAT_FEATURE(sb,
381 EXT4_FEATURE_INCOMPAT_RECOVER));
383 + if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
384 + !(sb->s_flags & MS_RDONLY))
385 + if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
386 + goto failed_mount3;
389 * The first inode we look at is the journal inode. Don't try
390 * root first: it may be modified in the journal!
391 @@ -3036,6 +3394,8 @@
392 percpu_counter_destroy(&sbi->s_freeinodes_counter);
393 percpu_counter_destroy(&sbi->s_dirs_counter);
394 percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
395 + if (sbi->s_mmp_tsk)
396 + kthread_stop(sbi->s_mmp_tsk);
398 for (i = 0; i < db_count; i++)
399 brelse(sbi->s_group_desc[i]);
400 @@ -3544,7 +3904,7 @@
401 struct ext4_mount_options old_opts;
403 unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
409 @@ -3666,6 +4026,13 @@
411 if (!ext4_setup_super(sb, es, 0))
412 sb->s_flags &= ~MS_RDONLY;
413 + if (EXT4_HAS_INCOMPAT_FEATURE(sb,
414 + EXT4_FEATURE_INCOMPAT_MMP))
415 + if (ext4_multi_mount_protect(sb,
416 + le64_to_cpu(es->s_mmp_block))) {
422 ext4_setup_system_zone(sb);
423 Index: linux-stage/fs/ext4/ext4.h
424 ===================================================================
425 --- linux-stage.orig/fs/ext4/ext4.h 2011-04-05 17:51:52.539390862 +0800
426 +++ linux-stage/fs/ext4/ext4.h 2011-04-05 17:52:02.093529865 +0800
428 __le16 s_want_extra_isize; /* New inodes should reserve # bytes */
429 __le32 s_flags; /* Miscellaneous flags */
430 __le16 s_raid_stride; /* RAID stride */
431 - __le16 s_mmp_interval; /* # seconds to wait in MMP checking */
432 + __le16 s_mmp_update_interval; /* # seconds to wait in MMP checking */
433 __le64 s_mmp_block; /* Block for multi-mount protection */
434 __le32 s_raid_stripe_width; /* blocks on all data disks (N*stride)*/
435 __u8 s_log_groups_per_flex; /* FLEX_BG group size */
436 @@ -1040,6 +1040,9 @@
438 /* workqueue for dio unwritten */
439 struct workqueue_struct *dio_unwritten_wq;
441 + /* Kernel thread for multiple mount protection */
442 + struct task_struct *s_mmp_tsk;
445 static inline struct ext4_sb_info *EXT4_SB(struct super_block *sb)
446 @@ -1176,7 +1179,8 @@
447 EXT4_FEATURE_INCOMPAT_META_BG| \
448 EXT4_FEATURE_INCOMPAT_EXTENTS| \
449 EXT4_FEATURE_INCOMPAT_64BIT| \
450 - EXT4_FEATURE_INCOMPAT_FLEX_BG)
451 + EXT4_FEATURE_INCOMPAT_FLEX_BG| \
452 + EXT4_FEATURE_INCOMPAT_MMP)
453 #define EXT4_FEATURE_RO_COMPAT_SUPP (EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER| \
454 EXT4_FEATURE_RO_COMPAT_LARGE_FILE| \
455 EXT4_FEATURE_RO_COMPAT_GDT_CSUM| \
456 @@ -1383,6 +1387,45 @@
457 extern struct proc_dir_entry *ext4_proc_root;
460 + * This structure will be used for multiple mount protection. It will be
461 + * written into the block number saved in the s_mmp_block field in the
462 + * superblock. Programs that check MMP should assume that if
463 + * SEQ_FSCK (or any unknown code above SEQ_MAX) is present then it is NOT safe
464 + * to use the filesystem, regardless of how old the timestamp is.
466 +#define EXT4_MMP_MAGIC 0x004D4D50U /* ASCII for MMP */
467 +#define EXT4_MMP_SEQ_CLEAN 0xFF4D4D50U /* mmp_seq value for clean unmount */
468 +#define EXT4_MMP_SEQ_FSCK 0xE24D4D50U /* mmp_seq value when being fscked */
469 +#define EXT4_MMP_SEQ_MAX 0xE24D4D4FU /* maximum valid mmp_seq value */
475 + char mmp_nodename[64];
476 + char mmp_bdevname[32];
477 + __le16 mmp_check_interval;
479 + __le32 mmp_pad2[227];
482 +/* arguments passed to the mmp thread */
484 + struct buffer_head *bh; /* bh from initial read_mmp_block() */
485 + struct super_block *sb; /* super block of the fs */
489 + * Minimum interval for MMP checking in seconds.
491 +#define EXT4_MMP_MIN_CHECK_INTERVAL 5UL
494 + * Maximum interval for MMP checking in seconds.
496 +#define EXT4_MMP_MAX_CHECK_INTERVAL 300UL
499 * Function prototypes