Whamcloud - gitweb
ece021c37413d32ac8afc9c66155906a5f78fd66
[fs/lustre-release.git] / ldiskfs / kernel_patches / patches / ext3-mmp-2.6.22-vanilla.patch
1 Index: linux-2.6.22.14/fs/ext3/super.c
2 ===================================================================
3 --- linux-2.6.22.14.orig/fs/ext3/super.c
4 +++ linux-2.6.22.14/fs/ext3/super.c
5 @@ -35,6 +35,8 @@
6  #include <linux/namei.h>
7  #include <linux/quotaops.h>
8  #include <linux/seq_file.h>
9 +#include <linux/kthread.h>
10 +#include <linux/utsname.h>
11  
12  #include <asm/uaccess.h>
13  
14 @@ -436,6 +438,8 @@ static void ext3_put_super (struct super
15                 invalidate_bdev(sbi->journal_bdev);
16                 ext3_blkdev_remove(sbi);
17         }
18 +       if (sbi->s_mmp_tsk)
19 +               kthread_stop(sbi->s_mmp_tsk);
20         if (sbi->s_dev_proc) {
21                 remove_proc_entry(sbi->s_dev_proc->name, proc_root_ext3);
22                 sbi->s_dev_proc = NULL;
23 @@ -1538,6 +1542,313 @@ static ext3_fsblk_t descriptor_loc(struc
24         return (has_super + ext3_group_first_block_no(sb, bg));
25  }
26  
27 +/*
28 + * Write the MMP block using WRITE_SYNC to try to get the block on-disk
29 + * faster.
30 + */
31 +static int write_mmp_block(struct buffer_head *bh)
32 +{
33 +       mark_buffer_dirty(bh);
34 +       lock_buffer(bh);
35 +       bh->b_end_io = end_buffer_write_sync;
36 +       get_bh(bh);
37 +       submit_bh(WRITE_SYNC, bh);
38 +       wait_on_buffer(bh);
39 +       if (unlikely(!buffer_uptodate(bh)))
40 +               return 1;
41 +
42 +       return 0;
43 +}
44 +
45 +/*
46 + * Read the MMP block. It _must_ be read from disk and hence we clear the
47 + * uptodate flag on the buffer.
48 + */
49 +static int read_mmp_block(struct super_block *sb, struct buffer_head **bh,
50 +                         unsigned long mmp_block)
51 +{
52 +       struct mmp_struct *mmp;
53 +
54 +       if (*bh)
55 +               clear_buffer_uptodate(*bh);
56 +
57 +       brelse(*bh);
58 +
59 +       *bh = sb_bread(sb, mmp_block);
60 +       if (!*bh) {
61 +               ext3_warning(sb, __FUNCTION__,
62 +                            "Error while reading MMP block %lu", mmp_block);
63 +               return -EIO;
64 +       }
65 +
66 +       mmp = (struct mmp_struct *)((*bh)->b_data);
67 +       if (le32_to_cpu(mmp->mmp_magic) != EXT3_MMP_MAGIC)
68 +               return -EINVAL;
69 +
70 +       return 0;
71 +}
72 +
73 +/*
74 + * Dump as much information as possible to help the admin.
75 + */
76 +static void dump_mmp_msg(struct super_block *sb, struct mmp_struct *mmp,
77 +                        const char *function, const char *msg)
78 +{
79 +       ext3_warning(sb, function, msg);
80 +       ext3_warning(sb, function, "MMP failure info: last update time: %llu, "
81 +                    "last update node: %s, last update device: %s\n",
82 +                    le64_to_cpu(mmp->mmp_time), mmp->mmp_nodename,
83 +                    mmp->mmp_bdevname);
84 +}
85 +
86 +/*
87 + * kmmpd will update the MMP sequence every s_mmp_update_interval seconds
88 + */
89 +static int kmmpd(void *data)
90 +{
91 +       struct super_block *sb = (struct super_block *) data;
92 +       struct ext3_super_block *es = EXT3_SB(sb)->s_es;
93 +       struct buffer_head *bh = NULL;
94 +       struct mmp_struct *mmp;
95 +       unsigned long mmp_block;
96 +       u32 seq = 0;
97 +       unsigned long failed_writes = 0;
98 +       int mmp_update_interval = le16_to_cpu(es->s_mmp_update_interval);
99 +       unsigned mmp_check_interval;
100 +       unsigned long last_update_time;
101 +       unsigned long diff;
102 +       int retval;
103 +
104 +       mmp_block = le64_to_cpu(es->s_mmp_block);
105 +       retval = read_mmp_block(sb, &bh, mmp_block);
106 +       if (retval)
107 +               goto failed;
108 +
109 +       mmp = (struct mmp_struct *)(bh->b_data);
110 +       mmp->mmp_time = cpu_to_le64(get_seconds());
111 +       /*
112 +        * Start with the higher mmp_check_interval and reduce it if
113 +        * the MMP block is being updated on time.
114 +        */
115 +       mmp_check_interval = max(5 * mmp_update_interval,
116 +                                EXT3_MMP_MIN_CHECK_INTERVAL);
117 +       mmp->mmp_check_interval = cpu_to_le16(mmp_check_interval);
118 +       bdevname(bh->b_bdev, mmp->mmp_bdevname);
119 +
120 +       down_read(&uts_sem);
121 +       memcpy(mmp->mmp_nodename, utsname()->nodename,
122 +              sizeof(mmp->mmp_nodename));
123 +       up_read(&uts_sem);
124 +
125 +       while (!kthread_should_stop()) {
126 +               if (++seq > EXT3_MMP_SEQ_MAX)
127 +                       seq = 1;
128 +
129 +               mmp->mmp_seq = cpu_to_le32(seq);
130 +               mmp->mmp_time = cpu_to_le64(get_seconds());
131 +               last_update_time = jiffies;
132 +
133 +               retval = write_mmp_block(bh);
134 +               /*
135 +                * Don't spew too many error messages. Print one every
136 +                * (s_mmp_update_interval * 60) seconds.
137 +                */
138 +               if (retval && (failed_writes % 60) == 0) {
139 +                       ext3_error(sb, __FUNCTION__,
140 +                                  "Error writing to MMP block");
141 +                       failed_writes++;
142 +               }
143 +
144 +               if (!(le32_to_cpu(es->s_feature_incompat) &
145 +                   EXT3_FEATURE_INCOMPAT_MMP)) {
146 +                       ext3_warning(sb, __FUNCTION__, "kmmpd being stopped "
147 +                                    "since MMP feature has been disabled.");
148 +                       EXT3_SB(sb)->s_mmp_tsk = 0;
149 +                       goto failed;
150 +               }
151 +
152 +               if (sb->s_flags & MS_RDONLY) {
153 +                       ext3_warning(sb, __FUNCTION__, "kmmpd being stopped "
154 +                                    "since filesystem has been remounted as "
155 +                                    "readonly.");
156 +                       EXT3_SB(sb)->s_mmp_tsk = 0;
157 +                       goto failed;
158 +               }
159 +
160 +               diff = jiffies - last_update_time;
161 +               if (diff < mmp_update_interval * HZ)
162 +                       schedule_timeout_interruptible(EXT3_MMP_UPDATE_INTERVAL*
163 +                                                      HZ - diff);
164 +
165 +               /*
166 +                * We need to make sure that more than mmp_check_interval
167 +                * seconds have not passed since writing. If that has happened
168 +                * we need to check if the MMP block is as we left it.
169 +                */
170 +               diff = jiffies - last_update_time;
171 +               if (diff > mmp_check_interval * HZ) {
172 +                       struct buffer_head *bh_check = NULL;
173 +                       struct mmp_struct *mmp_check;
174 +
175 +                       retval = read_mmp_block(sb, &bh_check, mmp_block);
176 +                       if (retval) {
177 +                               EXT3_SB(sb)->s_mmp_tsk = 0;
178 +                               goto failed;
179 +                       }
180 +
181 +                       mmp_check = (struct mmp_struct *)(bh_check->b_data);
182 +                       if (mmp->mmp_time != mmp_check->mmp_time ||
183 +                           memcmp(mmp->mmp_nodename, mmp_check->mmp_nodename,
184 +                                  sizeof(mmp->mmp_nodename)))
185 +                               dump_mmp_msg(sb, mmp_check, __FUNCTION__,
186 +                                            "Error while updating MMP info. "
187 +                                            "The filesystem seems to have "
188 +                                            "been multiply mounted.");
189 +
190 +                       put_bh(bh_check);
191 +               }
192 +
193 +               /*
194 +                * Adjust the mmp_check_interval depending on how much time
195 +                * it took for the MMP block to be written.
196 +                */
197 +               mmp_check_interval = max(5 * diff / HZ,
198 +                                (unsigned long) EXT3_MMP_MIN_CHECK_INTERVAL);
199 +               mmp->mmp_check_interval = cpu_to_le16(mmp_check_interval);
200 +       }
201 +
202 +       /*
203 +        * Unmount seems to be clean.
204 +        */
205 +       mmp->mmp_seq = cpu_to_le32(EXT3_MMP_SEQ_CLEAN);
206 +       mmp->mmp_time = cpu_to_le64(get_seconds());
207 +
208 +       retval = write_mmp_block(bh);
209 +
210 +failed:
211 +       brelse(bh);
212 +       return retval;
213 +}
214 +
215 +/*
216 + * Get a random new sequence number but make sure it is not greater than
217 + * EXT3_MMP_SEQ_MAX.
218 + */
219 +static unsigned int mmp_new_seq(void)
220 +{
221 +       u32 new_seq;
222 +
223 +       do {
224 +               get_random_bytes(&new_seq, sizeof(u32));
225 +       } while (new_seq > EXT3_MMP_SEQ_MAX);
226 +
227 +       return new_seq;
228 +}
229 +
230 +/*
231 + * Protect the filesystem from being mounted more than once.
232 + */
233 +static int ext3_multi_mount_protect(struct super_block *sb,
234 +                                   unsigned long mmp_block)
235 +{
236 +       struct ext3_super_block *es = EXT3_SB(sb)->s_es;
237 +       struct buffer_head *bh = NULL;
238 +       struct mmp_struct *mmp = NULL;
239 +       u32 seq;
240 +       unsigned int mmp_check_interval = le16_to_cpu(es->s_mmp_update_interval);
241 +       int retval;
242 +
243 +       if (mmp_block < le32_to_cpu(es->s_first_data_block) ||
244 +           mmp_block >= le32_to_cpu(es->s_blocks_count)) {
245 +               ext3_warning(sb, __FUNCTION__,
246 +                            "Invalid MMP block in superblock");
247 +               goto failed;
248 +       }
249 +
250 +       retval = read_mmp_block(sb, &bh, mmp_block);
251 +       if (retval)
252 +               goto failed;
253 +
254 +       mmp = (struct mmp_struct *)(bh->b_data);
255 +
256 +       if (mmp_check_interval < EXT3_MMP_MIN_CHECK_INTERVAL)
257 +               mmp_check_interval = EXT3_MMP_MIN_CHECK_INTERVAL;
258 +
259 +       /*
260 +        * If check_interval in MMP block is larger, use that instead of
261 +        * update_interval from the superblock.
262 +        */
263 +       if (mmp->mmp_check_interval > mmp_check_interval)
264 +               mmp_check_interval = mmp->mmp_check_interval;
265 +
266 +       seq = le32_to_cpu(mmp->mmp_seq);
267 +       if (seq == EXT3_MMP_SEQ_CLEAN)
268 +               goto skip;
269 +
270 +       if (seq == EXT3_MMP_SEQ_FSCK) {
271 +               dump_mmp_msg(sb, mmp, __FUNCTION__,
272 +                            "fsck is running on the filesystem");
273 +               goto failed;
274 +       }
275 +
276 +       schedule_timeout_uninterruptible(HZ * (2 * mmp_check_interval + 1));
277 +
278 +       retval = read_mmp_block(sb, &bh, mmp_block);
279 +       if (retval)
280 +               goto failed;
281 +       mmp = (struct mmp_struct *)(bh->b_data);
282 +       if (seq != le32_to_cpu(mmp->mmp_seq)) {
283 +               dump_mmp_msg(sb, mmp, __FUNCTION__,
284 +                            "Device is already active on another node.");
285 +               goto failed;
286 +       }
287 +
288 +skip:
289 +       /*
290 +        * write a new random sequence number.
291 +        */
292 +       mmp->mmp_seq = seq = cpu_to_le32(mmp_new_seq());
293 +
294 +       retval = write_mmp_block(bh);
295 +       if (retval)
296 +               goto failed;
297 +
298 +       /*
299 +        * wait for MMP interval and check mmp_seq.
300 +        */
301 +       schedule_timeout_uninterruptible(HZ * (2 * mmp_check_interval + 1));
302 +
303 +       retval = read_mmp_block(sb, &bh, mmp_block);
304 +       if (retval)
305 +               goto failed;
306 +       mmp = (struct mmp_struct *)(bh->b_data);
307 +       if (seq != le32_to_cpu(mmp->mmp_seq)) {
308 +               dump_mmp_msg(sb, mmp, __FUNCTION__,
309 +                            "Device is already active on another node.");
310 +               goto failed;
311 +       }
312 +
313 +       /*
314 +        * Start a kernel thread to update the MMP block periodically.
315 +        */
316 +       EXT3_SB(sb)->s_mmp_tsk = kthread_run(kmmpd, sb, "kmmpd-%02x:%02x",
317 +                                            MAJOR(sb->s_dev),
318 +                                            MINOR(sb->s_dev));
319 +       if (IS_ERR(EXT3_SB(sb)->s_mmp_tsk)) {
320 +               EXT3_SB(sb)->s_mmp_tsk = 0;
321 +               ext3_warning(sb, __FUNCTION__, "Unable to create kmmpd thread "
322 +                            "for %s.", sb->s_id);
323 +               goto failed;
324 +       }
325 +
326 +       brelse(bh);
327 +       return 0;
328 +
329 +failed:
330 +       brelse(bh);
331 +       return 1;
332 +}
333 +
334  
335  static int ext3_fill_super (struct super_block *sb, void *data, int silent)
336  {
337 @@ -1875,6 +2186,11 @@ static int ext3_fill_super (struct super
338                           EXT3_HAS_INCOMPAT_FEATURE(sb,
339                                     EXT3_FEATURE_INCOMPAT_RECOVER));
340  
341 +       if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_MMP) &&
342 +           !(sb->s_flags & MS_RDONLY))
343 +               if (ext3_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
344 +                       goto failed_mount3;
345 +
346         /*
347          * The first inode we look at is the journal inode.  Don't try
348          * root first: it may be modified in the journal!
349 @@ -2007,6 +2323,8 @@ cantfind_ext3:
350  failed_mount4:
351         journal_destroy(sbi->s_journal);
352  failed_mount3:
353 +       if (sbi->s_mmp_tsk)
354 +               kthread_stop(sbi->s_mmp_tsk);
355         percpu_counter_destroy(&sbi->s_freeblocks_counter);
356         percpu_counter_destroy(&sbi->s_freeinodes_counter);
357         percpu_counter_destroy(&sbi->s_dirs_counter);
358 @@ -2482,7 +2800,7 @@ static int ext3_remount (struct super_bl
359         ext3_fsblk_t n_blocks_count = 0;
360         unsigned long old_sb_flags;
361         struct ext3_mount_options old_opts;
362 -       int err;
363 +       int err = 0;
364  #ifdef CONFIG_QUOTA
365         int i;
366  #endif
367 @@ -2580,6 +2898,11 @@ static int ext3_remount (struct super_bl
368                                 goto restore_opts;
369                         if (!ext3_setup_super (sb, es, 0))
370                                 sb->s_flags &= ~MS_RDONLY;
371 +                       if (EXT3_HAS_INCOMPAT_FEATURE(sb,
372 +                                                   EXT3_FEATURE_INCOMPAT_MMP))
373 +                               if (ext3_multi_mount_protect(sb,
374 +                                               le64_to_cpu(es->s_mmp_block)))
375 +                                       goto restore_opts;
376                 }
377         }
378  #ifdef CONFIG_QUOTA
379 Index: linux-2.6.22.14/include/linux/ext3_fs.h
380 ===================================================================
381 --- linux-2.6.22.14.orig/include/linux/ext3_fs.h
382 +++ linux-2.6.22.14/include/linux/ext3_fs.h
383 @@ -608,13 +608,17 @@ struct ext3_super_block {
384         __le32  s_first_meta_bg;        /* First metablock block group */
385         __le32  s_mkfs_time;            /* When the filesystem was created */
386         __le32  s_jnl_blocks[17];       /* Backup of the journal inode */
387 -       __le32  s_blocks_count_hi;      /* Blocks count high 32 bits */
388 +/*150*/        __le32  s_blocks_count_hi;      /* Blocks count high 32 bits */
389         __le32  s_r_blocks_count_hi;    /* Reserved blocks count high 32 bits*/
390         __le32  s_free_blocks_count_hi; /* Free blocks count high 32 bits */
391         __le16  s_min_extra_isize;      /* All inodes have at least # bytes */
392         __le16  s_want_extra_isize;     /* New inodes should reserve # bytes */
393 -       __le32  s_flags;                /* Miscellaneous flags */
394 -       __u32   s_reserved[167];        /* Padding to the end of the block */
395 +/*160*/        __le32  s_flags;                /* Miscellaneous flags */
396 +       __le16  s_raid_stride;          /* RAID stride */
397 +       __le16  s_mmp_update_interval;  /* # seconds to wait in MMP checking */
398 +       __le64  s_mmp_block;            /* Block for multi-mount protection */
399 +/*170*/        __le32  s_raid_stripe_width;    /* blocks on all data disks (N*stride)*/
400 +       __le32  s_reserved[163];        /* Padding to the end of the block */
401  };
402  
403  #ifdef __KERNEL__
404 @@ -717,12 +721,14 @@ static inline int ext3_valid_inum(struct
405  #define EXT3_FEATURE_INCOMPAT_JOURNAL_DEV      0x0008 /* Journal device */
406  #define EXT3_FEATURE_INCOMPAT_META_BG          0x0010
407  #define EXT3_FEATURE_INCOMPAT_EXTENTS          0x0040 /* extents support */
408 +#define EXT3_FEATURE_INCOMPAT_MMP              0x0100
409  
410  #define EXT3_FEATURE_COMPAT_SUPP       EXT2_FEATURE_COMPAT_EXT_ATTR
411  #define EXT3_FEATURE_INCOMPAT_SUPP     (EXT3_FEATURE_INCOMPAT_FILETYPE| \
412                                          EXT3_FEATURE_INCOMPAT_RECOVER| \
413                                          EXT3_FEATURE_INCOMPAT_META_BG| \
414 -                                        EXT3_FEATURE_INCOMPAT_EXTENTS)
415 +                                        EXT3_FEATURE_INCOMPAT_EXTENTS| \
416 +                                        EXT3_FEATURE_INCOMPAT_MMP)
417  #define EXT3_FEATURE_RO_COMPAT_SUPP    (EXT3_FEATURE_RO_COMPAT_SPARSE_SUPER| \
418                                          EXT3_FEATURE_RO_COMPAT_LARGE_FILE| \
419                                          EXT4_FEATURE_RO_COMPAT_GDT_CSUM| \
420 @@ -885,6 +891,39 @@ ext3_group_first_block_no(struct super_b
421  #define ERR_BAD_DX_DIR -75000
422  
423  /*
424 + * This structure will be used for multiple mount protection. It will be
425 + * written into the block number saved in the s_mmp_block field in the
426 + * superblock. Programs that check MMP should assume that if
427 + * SEQ_FSCK (or any unknown code above SEQ_MAX) is present then it is NOT safe
428 + * to use the filesystem, regardless of how old the timestamp is.
429 + */
430 +#define EXT3_MMP_MAGIC     0x004D4D50U /* ASCII for MMP */
431 +#define EXT3_MMP_SEQ_CLEAN 0xFF4D4D50U /* mmp_seq value for clean unmount */
432 +#define EXT3_MMP_SEQ_FSCK  0xE24D4D50U /* mmp_seq value when being fscked */
433 +#define EXT3_MMP_SEQ_MAX   0xE24D4D4FU /* maximum valid mmp_seq value */
434 +
435 +struct mmp_struct {
436 +       __le32  mmp_magic;
437 +       __le32  mmp_seq;
438 +       __le64  mmp_time;
439 +       char    mmp_nodename[64];
440 +       char    mmp_bdevname[32];
441 +       __le16  mmp_check_interval;
442 +       __le16  mmp_pad1;
443 +       __le32  mmp_pad2[227];
444 +};
445 +
446 +/*
447 + * Default interval in seconds to update the MMP sequence number.
448 + */
449 +#define EXT3_MMP_UPDATE_INTERVAL   1
450 +
451 +/*
452 + * Minimum interval for MMP checking in seconds.
453 + */
454 +#define EXT3_MMP_MIN_CHECK_INTERVAL    5
455 +
456 +/*
457   * Function prototypes
458   */
459  
460 Index: linux-2.6.22.14/include/linux/ext3_fs_sb.h
461 ===================================================================
462 --- linux-2.6.22.14.orig/include/linux/ext3_fs_sb.h
463 +++ linux-2.6.22.14/include/linux/ext3_fs_sb.h
464 @@ -158,6 +158,7 @@ struct ext3_sb_info {
465         /* locality groups */
466         struct ext3_locality_group *s_locality_groups;
467  
468 +       struct task_struct *s_mmp_tsk;  /* Kernel thread for multiple mount protection */
469  };
470  
471  #define EXT3_GROUP_INFO(sb, group)                                        \