1 Index: linux-stage/fs/ext4/super.c
2 ===================================================================
3 --- linux-stage.orig/fs/ext4/super.c
4 +++ linux-stage/fs/ext4/super.c
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"
14 @@ -660,6 +662,8 @@ static void ext4_put_super(struct super_
15 invalidate_bdev(sbi->journal_bdev, 0);
16 ext4_blkdev_remove(sbi);
19 + kthread_stop(sbi->s_mmp_tsk);
22 * Now that we are completely done shutting down the
23 @@ -921,6 +925,354 @@ static int ext4_show_options(struct seq_
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)) {
77 + ext4_warning(sb, __func__,
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, "%s", 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 + down_read(&uts_sem);
133 + memcpy(mmp->mmp_nodename, system_utsname.nodename,
134 + sizeof(mmp->mmp_nodename));
137 + while (!kthread_should_stop()) {
138 + if (++seq > EXT4_MMP_SEQ_MAX)
141 + mmp->mmp_seq = cpu_to_le32(seq);
142 + mmp->mmp_time = cpu_to_le64(get_seconds());
143 + last_update_time = jiffies;
145 + retval = write_mmp_block(bh);
147 + * Don't spew too many error messages. Print one every
148 + * (s_mmp_update_interval * 60) seconds.
150 + if (retval && (failed_writes % 60) == 0) {
151 + ext4_error(sb, __func__,
152 + "Error writing to MMP block");
156 + if (!(le32_to_cpu(es->s_feature_incompat) &
157 + EXT4_FEATURE_INCOMPAT_MMP)) {
158 + ext4_warning(sb, __func__, "kmmpd being stopped "
159 + "since MMP feature has been disabled.");
160 + EXT4_SB(sb)->s_mmp_tsk = NULL;
164 + if (sb->s_flags & MS_RDONLY) {
165 + ext4_warning(sb, __func__, "kmmpd being stopped "
166 + "since filesystem has been remounted as "
168 + EXT4_SB(sb)->s_mmp_tsk = NULL;
172 + diff = jiffies - last_update_time;
173 + if (diff < mmp_update_interval * HZ)
174 + schedule_timeout_interruptible(EXT4_MMP_UPDATE_INTERVAL*
178 + * We need to make sure that more than mmp_check_interval
179 + * seconds have not passed since writing. If that has happened
180 + * we need to check if the MMP block is as we left it.
182 + diff = jiffies - last_update_time;
183 + if (diff > mmp_check_interval * HZ) {
184 + struct buffer_head *bh_check = NULL;
185 + struct mmp_struct *mmp_check;
187 + retval = read_mmp_block(sb, &bh_check, mmp_block);
189 + ext4_error(sb, __func__, "error reading MMP"
190 + "data: %d", retval);
191 + EXT4_SB(sb)->s_mmp_tsk = NULL;
195 + mmp_check = (struct mmp_struct *)(bh_check->b_data);
196 + if (mmp->mmp_seq != mmp_check->mmp_seq ||
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.");
203 + ext4_error(sb, __func__, "abort");
210 + * Adjust the mmp_check_interval depending on how much time
211 + * it took for the MMP block to be written.
213 + mmp_check_interval = max(min(5 * diff / HZ,
214 + EXT4_MMP_MAX_CHECK_INTERVAL),
215 + EXT4_MMP_MIN_CHECK_INTERVAL);
216 + mmp->mmp_check_interval = cpu_to_le16(mmp_check_interval);
220 + * Unmount seems to be clean.
222 + mmp->mmp_seq = cpu_to_le32(EXT4_MMP_SEQ_CLEAN);
223 + mmp->mmp_time = cpu_to_le64(get_seconds());
225 + retval = write_mmp_block(bh);
233 + * Get a random new sequence number but make sure it is not greater than
234 + * EXT4_MMP_SEQ_MAX.
236 +static unsigned int mmp_new_seq(void)
241 + get_random_bytes(&new_seq, sizeof(u32));
242 + } while (new_seq > EXT4_MMP_SEQ_MAX);
248 + * Protect the filesystem from being mounted more than once.
250 +static int ext4_multi_mount_protect(struct super_block *sb,
251 + unsigned long mmp_block)
253 + struct ext4_super_block *es = EXT4_SB(sb)->s_es;
254 + struct buffer_head *bh = NULL;
255 + struct mmp_struct *mmp = NULL;
256 + struct mmpd_data *mmpd_data;
258 + unsigned int mmp_check_interval = le16_to_cpu(es->s_mmp_update_interval);
259 + unsigned int wait_time = 0;
262 + if (mmp_block < le32_to_cpu(es->s_first_data_block) ||
263 + mmp_block >= ext4_blocks_count(es)) {
264 + ext4_warning(sb, __func__,
265 + "Invalid MMP block in superblock");
269 + retval = read_mmp_block(sb, &bh, mmp_block);
273 + mmp = (struct mmp_struct *)(bh->b_data);
275 + if (mmp_check_interval < EXT4_MMP_MIN_CHECK_INTERVAL)
276 + mmp_check_interval = EXT4_MMP_MIN_CHECK_INTERVAL;
279 + * If check_interval in MMP block is larger, use that instead of
280 + * update_interval from the superblock.
282 + if (mmp->mmp_check_interval > mmp_check_interval)
283 + mmp_check_interval = mmp->mmp_check_interval;
285 + seq = le32_to_cpu(mmp->mmp_seq);
286 + if (seq == EXT4_MMP_SEQ_CLEAN)
289 + if (seq == EXT4_MMP_SEQ_FSCK) {
290 + dump_mmp_msg(sb, mmp, __func__,
291 + "fsck is running on the filesystem");
295 + wait_time = min(mmp_check_interval * 2 + 1,
296 + mmp_check_interval + 60);
298 + /* Print MMP interval if more than 20 secs. */
299 + if (wait_time > EXT4_MMP_MIN_CHECK_INTERVAL * 4)
300 + ext4_warning(sb, __func__, "MMP interval %u higher than "
301 + "expected, please wait.\n", wait_time * 2);
303 + if (schedule_timeout_interruptible(HZ * wait_time) != 0) {
304 + ext4_warning(sb, __func__, "MMP startup interrupted, failing "
309 + retval = read_mmp_block(sb, &bh, mmp_block);
312 + mmp = (struct mmp_struct *)(bh->b_data);
313 + if (seq != le32_to_cpu(mmp->mmp_seq)) {
314 + dump_mmp_msg(sb, mmp, __func__,
315 + "Device is already active on another node.");
321 + * write a new random sequence number.
323 + mmp->mmp_seq = seq = cpu_to_le32(mmp_new_seq());
325 + retval = write_mmp_block(bh);
330 + * wait for MMP interval and check mmp_seq.
332 + if (schedule_timeout_interruptible(HZ * wait_time) != 0) {
333 + ext4_warning(sb, __func__, "MMP startup interrupted, failing "
338 + retval = read_mmp_block(sb, &bh, mmp_block);
341 + mmp = (struct mmp_struct *)(bh->b_data);
342 + if (seq != le32_to_cpu(mmp->mmp_seq)) {
343 + dump_mmp_msg(sb, mmp, __func__,
344 + "Device is already active on another node.");
348 + mmpd_data = kmalloc(sizeof(struct mmpd_data), GFP_KERNEL);
350 + ext4_warning(sb, KERN_ERR, "not enough memory for mmpd_data");
353 + mmpd_data->sb = sb;
354 + mmpd_data->bh = bh;
357 + * Start a kernel thread to update the MMP block periodically.
359 + EXT4_SB(sb)->s_mmp_tsk = kthread_run(kmmpd, mmpd_data, "kmmpd-%s",
360 + bdevname(bh->b_bdev,
361 + mmp->mmp_bdevname));
362 + if (IS_ERR(EXT4_SB(sb)->s_mmp_tsk)) {
363 + EXT4_SB(sb)->s_mmp_tsk = NULL;
364 + ext4_warning(sb, __func__, "Unable to create kmmpd thread "
365 + "for %s.", sb->s_id);
376 static struct dentry *ext4_get_dentry(struct super_block *sb, void *vobjp)
378 @@ -930,7 +1282,6 @@ static struct dentry *ext4_get_dentry(st
380 struct dentry *result;
383 if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
384 return ERR_PTR(-ESTALE);
385 if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
386 @@ -2740,6 +3091,11 @@ static int ext4_fill_super(struct super_
387 EXT4_HAS_INCOMPAT_FEATURE(sb,
388 EXT4_FEATURE_INCOMPAT_RECOVER));
390 + if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
391 + !(sb->s_flags & MS_RDONLY))
392 + if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
393 + goto failed_mount3;
396 * The first inode we look at is the journal inode. Don't try
397 * root first: it may be modified in the journal!
398 @@ -2978,6 +3334,8 @@ failed_mount3:
399 percpu_counter_destroy(&sbi->s_freeinodes_counter);
400 percpu_counter_destroy(&sbi->s_dirs_counter);
401 percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
402 + if (sbi->s_mmp_tsk)
403 + kthread_stop(sbi->s_mmp_tsk);
405 for (i = 0; i < db_count; i++)
406 brelse(sbi->s_group_desc[i]);
407 @@ -3488,7 +3846,7 @@ static int ext4_remount(struct super_blo
408 struct ext4_mount_options old_opts;
410 unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
416 @@ -3607,6 +3965,13 @@ static int ext4_remount(struct super_blo
418 if (!ext4_setup_super(sb, es, 0))
419 sb->s_flags &= ~MS_RDONLY;
420 + if (EXT4_HAS_INCOMPAT_FEATURE(sb,
421 + EXT4_FEATURE_INCOMPAT_MMP))
422 + if (ext4_multi_mount_protect(sb,
423 + le64_to_cpu(es->s_mmp_block))) {
429 ext4_setup_system_zone(sb);
430 Index: linux-stage/fs/ext4/ext4.h
431 ===================================================================
432 --- linux-stage.orig/fs/ext4/ext4.h
433 +++ linux-stage/fs/ext4/ext4.h
434 @@ -851,7 +851,7 @@ struct ext4_super_block {
435 __le16 s_want_extra_isize; /* New inodes should reserve # bytes */
436 __le32 s_flags; /* Miscellaneous flags */
437 __le16 s_raid_stride; /* RAID stride */
438 - __le16 s_mmp_interval; /* # seconds to wait in MMP checking */
439 + __le16 s_mmp_update_interval; /* # seconds to wait in MMP checking */
440 __le64 s_mmp_block; /* Block for multi-mount protection */
441 __le32 s_raid_stripe_width; /* blocks on all data disks (N*stride)*/
442 __u8 s_log_groups_per_flex; /* FLEX_BG group size */
443 @@ -1005,6 +1005,9 @@ struct ext4_sb_info {
445 /* workqueue for dio unwritten */
446 struct workqueue_struct *dio_unwritten_wq;
448 + /* Kernel thread for multiple mount protection */
449 + struct task_struct *s_mmp_tsk;
452 static inline struct ext4_sb_info *EXT4_SB(struct super_block *sb)
453 @@ -1114,7 +1117,8 @@ static inline int ext4_valid_inum(struct
454 EXT4_FEATURE_INCOMPAT_META_BG| \
455 EXT4_FEATURE_INCOMPAT_EXTENTS| \
456 EXT4_FEATURE_INCOMPAT_64BIT| \
457 - EXT4_FEATURE_INCOMPAT_FLEX_BG)
458 + EXT4_FEATURE_INCOMPAT_FLEX_BG| \
459 + EXT4_FEATURE_INCOMPAT_MMP)
460 #define EXT4_FEATURE_RO_COMPAT_SUPP (EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER| \
461 EXT4_FEATURE_RO_COMPAT_LARGE_FILE| \
462 EXT4_FEATURE_RO_COMPAT_GDT_CSUM| \
463 @@ -1296,6 +1300,50 @@ void ext4_get_group_no_and_offset(struct
464 extern struct proc_dir_entry *ext4_proc_root;
467 + * This structure will be used for multiple mount protection. It will be
468 + * written into the block number saved in the s_mmp_block field in the
469 + * superblock. Programs that check MMP should assume that if
470 + * SEQ_FSCK (or any unknown code above SEQ_MAX) is present then it is NOT safe
471 + * to use the filesystem, regardless of how old the timestamp is.
473 +#define EXT4_MMP_MAGIC 0x004D4D50U /* ASCII for MMP */
474 +#define EXT4_MMP_SEQ_CLEAN 0xFF4D4D50U /* mmp_seq value for clean unmount */
475 +#define EXT4_MMP_SEQ_FSCK 0xE24D4D50U /* mmp_seq value when being fscked */
476 +#define EXT4_MMP_SEQ_MAX 0xE24D4D4FU /* maximum valid mmp_seq value */
482 + char mmp_nodename[64];
483 + char mmp_bdevname[32];
484 + __le16 mmp_check_interval;
486 + __le32 mmp_pad2[227];
489 +/* arguments passed to the mmp thread */
491 + struct buffer_head *bh; /* bh from initial read_mmp_block() */
492 + struct super_block *sb; /* super block of the fs */
496 + * Default interval in seconds to update the MMP sequence number.
498 +#define EXT4_MMP_UPDATE_INTERVAL 1
501 + * Minimum interval for MMP checking in seconds.
503 +#define EXT4_MMP_MIN_CHECK_INTERVAL 5UL
506 + * Maximum interval for MMP checking in seconds.
508 +#define EXT4_MMP_MAX_CHECK_INTERVAL 300UL
511 * Function prototypes