| blob | e00e4c6bdf14e1b11f9d64b15130a02d643c4a08 |
1 jbd2: load jmap from journal
3 If the lazy journal feature is enabled, instead of replaying the
4 journal, read the journal into journal map.
6 Signed-off-by: Theodore Ts'o <tytso@mit.edu>
7 ---
8 fs/jbd2/journal.c | 27 ++++++++--------------
9 fs/jbd2/recovery.c | 104 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++--------------------
10 2 files changed, 89 insertions(+), 42 deletions(-)
12 diff --git a/fs/jbd2/journal.c b/fs/jbd2/journal.c
13 index ab7b9bbc9296..1e4a9475826f 100644
14 --- a/fs/jbd2/journal.c
15 +++ b/fs/jbd2/journal.c
16 @@ -1275,31 +1275,24 @@ static void journal_fail_superblock (journal_t *journal)
18 /*
19 * Given a journal_t structure, initialise the various fields for
20 - * startup of a new journaling session. We use this both when creating
21 - * a journal, and after recovering an old journal to reset it for
22 - * subsequent use.
23 + * startup of a new journaling session.
24 */
25 -
26 static int journal_reset(journal_t *journal)
27 {
28 journal_superblock_t *sb = journal->j_superblock;
29 - unsigned long long first, last;
30 + int free;
32 - first = be32_to_cpu(sb->s_first);
33 - last = be32_to_cpu(sb->s_maxlen);
34 - if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
35 - printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
36 - first, last);
37 + if (journal->j_first + JBD2_MIN_JOURNAL_BLOCKS > journal->j_last + 1) {
38 + printk(KERN_ERR "JBD2: Journal too short (blocks %lu-%lu).\n",
39 + journal->j_first, journal->j_last);
40 journal_fail_superblock(journal);
41 return -EINVAL;
42 }
44 - journal->j_first = first;
45 - journal->j_last = last;
46 -
47 - journal->j_head = first;
48 - journal->j_tail = first;
49 - journal->j_free = last - first;
50 + free = journal->j_tail - journal->j_head;
51 + if (free <= 0)
52 + free += journal->j_last - journal->j_first;
53 + journal->j_free = free;
55 journal->j_tail_sequence = journal->j_transaction_sequence;
56 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
57 @@ -1319,7 +1312,7 @@ static int journal_reset(journal_t *journal)
58 journal->j_tail, journal->j_tail_sequence,
59 journal->j_errno);
60 journal->j_flags |= JBD2_FLUSHED;
61 - } else {
62 + } else if ((journal->j_flags & JBD2_LAZY) == 0) {
63 /* Lock here to make assertions happy... */
64 mutex_lock(&journal->j_checkpoint_mutex);
65 /*
66 diff --git a/fs/jbd2/recovery.c b/fs/jbd2/recovery.c
67 index da100044566c..c96ccd1f257f 100644
68 --- a/fs/jbd2/recovery.c
69 +++ b/fs/jbd2/recovery.c
70 @@ -32,17 +32,18 @@ struct recovery_info
71 {
72 tid_t start_transaction;
73 tid_t end_transaction;
74 + int head_block;
76 int nr_replays;
77 int nr_revokes;
78 int nr_revoke_hits;
79 };
81 -enum passtype {PASS_SCAN, PASS_REVOKE, PASS_REPLAY};
82 +enum passtype {PASS_SCAN, PASS_REVOKE, PASS_REPLAY, PASS_JMAP};
83 static int do_one_pass(journal_t *journal,
84 struct recovery_info *info, enum passtype pass);
85 -static int scan_revoke_records(journal_t *, struct buffer_head *,
86 - tid_t, struct recovery_info *);
87 +static int scan_revoke_records(journal_t *, struct buffer_head *, enum passtype,
88 + tid_t, struct recovery_info *);
90 #ifdef __KERNEL__
92 @@ -255,11 +256,16 @@ int jbd2_journal_recover(journal_t *journal)
93 sb = journal->j_superblock;
95 /*
96 + * Initialize journal's head and tail assuming the recovery
97 + * was successful and we're not doing lazy journalling.
98 + */
99 + journal->j_head = journal->j_tail = journal->j_first;
100 +
101 + /*
102 * The journal superblock's s_start field (the current log head)
103 * is always zero if, and only if, the journal was cleanly
104 * unmounted.
105 */
106 -
107 if (!sb->s_start) {
108 jbd_debug(1, "No recovery required, last transaction %d\n",
109 be32_to_cpu(sb->s_sequence));
110 @@ -267,11 +273,15 @@ int jbd2_journal_recover(journal_t *journal)
111 return 0;
112 }
114 - err = do_one_pass(journal, &info, PASS_SCAN);
115 - if (!err)
116 - err = do_one_pass(journal, &info, PASS_REVOKE);
117 - if (!err)
118 - err = do_one_pass(journal, &info, PASS_REPLAY);
119 + if (journal->j_flags & JBD2_LAZY)
120 + err = do_one_pass(journal, &info, PASS_JMAP);
121 + else {
122 + err = do_one_pass(journal, &info, PASS_SCAN);
123 + if (!err)
124 + err = do_one_pass(journal, &info, PASS_REVOKE);
125 + if (!err)
126 + err = do_one_pass(journal, &info, PASS_REPLAY);
127 + }
129 jbd_debug(1, "JBD2: recovery, exit status %d, "
130 "recovered transactions %u to %u\n",
131 @@ -279,10 +289,22 @@ int jbd2_journal_recover(journal_t *journal)
132 jbd_debug(1, "JBD2: Replayed %d and revoked %d/%d blocks\n",
133 info.nr_replays, info.nr_revoke_hits, info.nr_revokes);
135 - /* Restart the log at the next transaction ID, thus invalidating
136 - * any existing commit records in the log. */
137 + /* Restart the log at the next transaction ID */
138 journal->j_transaction_sequence = info.end_transaction;
140 + /*
141 + * In lazy journalling mode, we need to preserve the existing
142 + * contents of the journal, so set j_head and j_tail
143 + * accordingly.
144 + */
145 + if (journal->j_flags & JBD2_LAZY) {
146 + if (err)
147 + return err;
148 + journal->j_head = info.head_block;
149 + journal->j_tail = be32_to_cpu(sb->s_start);
150 + return 0;
151 + }
152 +
153 jbd2_journal_clear_revoke(journal);
154 err2 = sync_blockdev(journal->j_fs_dev);
155 if (!err)
156 @@ -431,6 +453,7 @@ static int do_one_pass(journal_t *journal,
157 __u32 crc32_sum = ~0; /* Transactional Checksums */
158 int descr_csum_size = 0;
159 int block_error = 0;
160 + int new_txn = 1;
162 /*
163 * First thing is to establish what we expect to find in the log
164 @@ -443,7 +466,7 @@ static int do_one_pass(journal_t *journal,
165 next_log_block = be32_to_cpu(sb->s_start);
167 first_commit_ID = next_commit_ID;
168 - if (pass == PASS_SCAN)
169 + if (pass == PASS_SCAN || pass == PASS_JMAP)
170 info->start_transaction = first_commit_ID;
172 jbd_debug(1, "Starting recovery pass %d\n", pass);
173 @@ -468,7 +491,7 @@ static int do_one_pass(journal_t *journal,
174 * check right now that we haven't gone past the end of
175 * the log. */
177 - if (pass != PASS_SCAN)
178 + if (pass != PASS_SCAN && pass != PASS_JMAP)
179 if (tid_geq(next_commit_ID, info->end_transaction))
180 break;
182 @@ -484,9 +507,6 @@ static int do_one_pass(journal_t *journal,
183 if (err)
184 goto failed;
186 - next_log_block++;
187 - wrap(journal, next_log_block);
188 -
189 /* What kind of buffer is it?
190 *
191 * If it is a descriptor block, check that it has the
192 @@ -510,6 +530,14 @@ static int do_one_pass(journal_t *journal,
193 break;
194 }
196 + if ((pass == PASS_JMAP) && new_txn) {
197 + jbd2_add_new_transaction_infos(journal, sequence, next_log_block);
198 + new_txn = 0;
199 + }
200 +
201 + next_log_block++;
202 + wrap(journal, next_log_block);
203 +
204 /* OK, we have a valid descriptor block which matches
205 * all of the sequence number checks. What are we going
206 * to do with it? That depends on the pass... */
207 @@ -535,7 +563,7 @@ static int do_one_pass(journal_t *journal,
208 * in pass REPLAY; if journal_checksums enabled, then
209 * calculate checksums in PASS_SCAN, otherwise,
210 * just skip over the blocks it describes. */
211 - if (pass != PASS_REPLAY) {
212 + if ((pass != PASS_REPLAY) && (pass != PASS_JMAP)) {
213 if (pass == PASS_SCAN &&
214 jbd2_has_feature_checksum(journal) &&
215 !info->end_transaction) {
216 @@ -562,12 +590,28 @@ static int do_one_pass(journal_t *journal,
217 while ((tagp - bh->b_data + tag_bytes)
218 <= journal->j_blocksize - descr_csum_size) {
219 unsigned long io_block;
220 + unsigned long long log_block;
222 tag = (journal_block_tag_t *) tagp;
223 flags = be16_to_cpu(tag->t_flags);
225 io_block = next_log_block++;
226 wrap(journal, next_log_block);
227 + if (pass == PASS_JMAP) {
228 + struct blk_mapping map;
229 +
230 + err = jbd2_journal_bmap(journal,
231 + io_block,
232 + &log_block);
233 + if (err)
234 + goto failed;
235 + map.fsblk = read_tag_block(journal, tag);
236 + map.logblk = log_block;
237 + err = jbd2_add_mapping(journal, &map);
238 + if (err)
239 + goto failed;
240 + goto skip_write;
241 + }
242 err = jread(&obh, journal, io_block);
243 if (err) {
244 /* Recover what we can, but
245 @@ -753,6 +797,10 @@ static int do_one_pass(journal_t *journal,
246 break;
247 }
248 }
249 + if (pass == PASS_JMAP) {
250 + jbd2_finish_transaction_infos(journal);
251 + new_txn = 1;
252 + }
253 brelse(bh);
254 next_commit_ID++;
255 continue;
256 @@ -760,12 +808,12 @@ static int do_one_pass(journal_t *journal,
257 case JBD2_REVOKE_BLOCK:
258 /* If we aren't in the REVOKE pass, then we can
259 * just skip over this block. */
260 - if (pass != PASS_REVOKE) {
261 + if (pass != PASS_REVOKE && pass != PASS_JMAP) {
262 brelse(bh);
263 continue;
264 }
266 - err = scan_revoke_records(journal, bh,
267 + err = scan_revoke_records(journal, bh, pass,
268 next_commit_ID, info);
269 brelse(bh);
270 if (err)
271 @@ -788,9 +836,10 @@ static int do_one_pass(journal_t *journal,
272 * transaction marks the end of the valid log.
273 */
275 - if (pass == PASS_SCAN) {
276 + if (pass == PASS_SCAN || pass == PASS_JMAP) {
277 if (!info->end_transaction)
278 info->end_transaction = next_commit_ID;
279 + info->head_block = next_log_block;
280 } else {
281 /* It's really bad news if different passes end up at
282 * different places (but possible due to IO errors). */
283 @@ -813,7 +862,8 @@ static int do_one_pass(journal_t *journal,
284 /* Scan a revoke record, marking all blocks mentioned as revoked. */
286 static int scan_revoke_records(journal_t *journal, struct buffer_head *bh,
287 - tid_t sequence, struct recovery_info *info)
288 + enum passtype pass, tid_t sequence,
289 + struct recovery_info *info)
290 {
291 jbd2_journal_revoke_header_t *header;
292 int offset, max;
293 @@ -839,16 +889,20 @@ static int scan_revoke_records(journal_t *journal, struct buffer_head *bh,
295 while (offset + record_len <= max) {
296 unsigned long long blocknr;
297 - int err;
299 if (record_len == 4)
300 blocknr = be32_to_cpu(* ((__be32 *) (bh->b_data+offset)));
301 else
302 blocknr = be64_to_cpu(* ((__be64 *) (bh->b_data+offset)));
303 offset += record_len;
304 - err = jbd2_journal_set_revoke(journal, blocknr, sequence);
305 - if (err)
306 - return err;
307 + if (pass == PASS_JMAP)
308 + jbd2_jmap_revoke(journal, blocknr);
309 + else {
310 + int err = jbd2_journal_set_revoke(journal, blocknr,
311 + sequence);
312 + if (err)
313 + return err;
314 + }
315 ++info->nr_revokes;
316 }
317 return 0;