Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
58862699 | 2 | * linux/fs/jbd/checkpoint.c |
ae6ddcc5 | 3 | * |
1da177e4 LT |
4 | * Written by Stephen C. Tweedie <sct@redhat.com>, 1999 |
5 | * | |
6 | * Copyright 1999 Red Hat Software --- All Rights Reserved | |
7 | * | |
8 | * This file is part of the Linux kernel and is made available under | |
9 | * the terms of the GNU General Public License, version 2, or at your | |
10 | * option, any later version, incorporated herein by reference. | |
11 | * | |
ae6ddcc5 MC |
12 | * Checkpoint routines for the generic filesystem journaling code. |
13 | * Part of the ext2fs journaling system. | |
1da177e4 LT |
14 | * |
15 | * Checkpointing is the process of ensuring that a section of the log is | |
16 | * committed fully to disk, so that that portion of the log can be | |
17 | * reused. | |
18 | */ | |
19 | ||
20 | #include <linux/time.h> | |
21 | #include <linux/fs.h> | |
22 | #include <linux/jbd.h> | |
23 | #include <linux/errno.h> | |
24 | #include <linux/slab.h> | |
99cb1a31 | 25 | #include <trace/events/jbd.h> |
1da177e4 LT |
26 | |
27 | /* | |
78ce89c9 | 28 | * Unlink a buffer from a transaction checkpoint list. |
1da177e4 LT |
29 | * |
30 | * Called with j_list_lock held. | |
31 | */ | |
78ce89c9 | 32 | static inline void __buffer_unlink_first(struct journal_head *jh) |
1da177e4 | 33 | { |
78ce89c9 | 34 | transaction_t *transaction = jh->b_cp_transaction; |
1da177e4 LT |
35 | |
36 | jh->b_cpnext->b_cpprev = jh->b_cpprev; | |
37 | jh->b_cpprev->b_cpnext = jh->b_cpnext; | |
78ce89c9 | 38 | if (transaction->t_checkpoint_list == jh) { |
1da177e4 | 39 | transaction->t_checkpoint_list = jh->b_cpnext; |
78ce89c9 JK |
40 | if (transaction->t_checkpoint_list == jh) |
41 | transaction->t_checkpoint_list = NULL; | |
42 | } | |
43 | } | |
44 | ||
45 | /* | |
46 | * Unlink a buffer from a transaction checkpoint(io) list. | |
47 | * | |
48 | * Called with j_list_lock held. | |
49 | */ | |
50 | static inline void __buffer_unlink(struct journal_head *jh) | |
51 | { | |
52 | transaction_t *transaction = jh->b_cp_transaction; | |
53 | ||
54 | __buffer_unlink_first(jh); | |
55 | if (transaction->t_checkpoint_io_list == jh) { | |
56 | transaction->t_checkpoint_io_list = jh->b_cpnext; | |
57 | if (transaction->t_checkpoint_io_list == jh) | |
58 | transaction->t_checkpoint_io_list = NULL; | |
59 | } | |
60 | } | |
61 | ||
62 | /* | |
63 | * Move a buffer from the checkpoint list to the checkpoint io list | |
64 | * | |
65 | * Called with j_list_lock held | |
66 | */ | |
67 | static inline void __buffer_relink_io(struct journal_head *jh) | |
68 | { | |
69 | transaction_t *transaction = jh->b_cp_transaction; | |
70 | ||
71 | __buffer_unlink_first(jh); | |
72 | ||
73 | if (!transaction->t_checkpoint_io_list) { | |
74 | jh->b_cpnext = jh->b_cpprev = jh; | |
75 | } else { | |
76 | jh->b_cpnext = transaction->t_checkpoint_io_list; | |
77 | jh->b_cpprev = transaction->t_checkpoint_io_list->b_cpprev; | |
78 | jh->b_cpprev->b_cpnext = jh; | |
79 | jh->b_cpnext->b_cpprev = jh; | |
80 | } | |
81 | transaction->t_checkpoint_io_list = jh; | |
1da177e4 LT |
82 | } |
83 | ||
84 | /* | |
85 | * Try to release a checkpointed buffer from its transaction. | |
78ce89c9 JK |
86 | * Returns 1 if we released it and 2 if we also released the |
87 | * whole transaction. | |
88 | * | |
1da177e4 LT |
89 | * Requires j_list_lock |
90 | * Called under jbd_lock_bh_state(jh2bh(jh)), and drops it | |
91 | */ | |
92 | static int __try_to_free_cp_buf(struct journal_head *jh) | |
93 | { | |
94 | int ret = 0; | |
95 | struct buffer_head *bh = jh2bh(jh); | |
96 | ||
4afe9785 | 97 | if (jh->b_jlist == BJ_None && !buffer_locked(bh) && |
9f818b4a | 98 | !buffer_dirty(bh) && !buffer_write_io_error(bh)) { |
bb189247 JK |
99 | /* |
100 | * Get our reference so that bh cannot be freed before | |
101 | * we unlock it | |
102 | */ | |
103 | get_bh(bh); | |
1da177e4 | 104 | JBUFFER_TRACE(jh, "remove from checkpoint list"); |
78ce89c9 | 105 | ret = __journal_remove_checkpoint(jh) + 1; |
1da177e4 | 106 | jbd_unlock_bh_state(bh); |
1da177e4 LT |
107 | BUFFER_TRACE(bh, "release"); |
108 | __brelse(bh); | |
1da177e4 LT |
109 | } else { |
110 | jbd_unlock_bh_state(bh); | |
111 | } | |
112 | return ret; | |
113 | } | |
114 | ||
115 | /* | |
116 | * __log_wait_for_space: wait until there is space in the journal. | |
117 | * | |
118 | * Called under j-state_lock *only*. It will be unlocked if we have to wait | |
119 | * for a checkpoint to free up some space in the log. | |
120 | */ | |
121 | void __log_wait_for_space(journal_t *journal) | |
122 | { | |
e219cca0 | 123 | int nblocks, space_left; |
1da177e4 LT |
124 | assert_spin_locked(&journal->j_state_lock); |
125 | ||
126 | nblocks = jbd_space_needed(journal); | |
127 | while (__log_space_left(journal) < nblocks) { | |
128 | if (journal->j_flags & JFS_ABORT) | |
129 | return; | |
130 | spin_unlock(&journal->j_state_lock); | |
2c68ee75 | 131 | mutex_lock(&journal->j_checkpoint_mutex); |
1da177e4 LT |
132 | |
133 | /* | |
134 | * Test again, another process may have checkpointed while we | |
be07c4ed | 135 | * were waiting for the checkpoint lock. If there are no |
e219cca0 TT |
136 | * transactions ready to be checkpointed, try to recover |
137 | * journal space by calling cleanup_journal_tail(), and if | |
138 | * that doesn't work, by waiting for the currently committing | |
139 | * transaction to complete. If there is absolutely no way | |
140 | * to make progress, this is either a BUG or corrupted | |
141 | * filesystem, so abort the journal and leave a stack | |
142 | * trace for forensic evidence. | |
1da177e4 LT |
143 | */ |
144 | spin_lock(&journal->j_state_lock); | |
be07c4ed | 145 | spin_lock(&journal->j_list_lock); |
1da177e4 | 146 | nblocks = jbd_space_needed(journal); |
e219cca0 TT |
147 | space_left = __log_space_left(journal); |
148 | if (space_left < nblocks) { | |
be07c4ed | 149 | int chkpt = journal->j_checkpoint_transactions != NULL; |
e219cca0 | 150 | tid_t tid = 0; |
be07c4ed | 151 | |
e219cca0 TT |
152 | if (journal->j_committing_transaction) |
153 | tid = journal->j_committing_transaction->t_tid; | |
be07c4ed | 154 | spin_unlock(&journal->j_list_lock); |
1da177e4 | 155 | spin_unlock(&journal->j_state_lock); |
be07c4ed DG |
156 | if (chkpt) { |
157 | log_do_checkpoint(journal); | |
e219cca0 TT |
158 | } else if (cleanup_journal_tail(journal) == 0) { |
159 | /* We were able to recover space; yay! */ | |
160 | ; | |
161 | } else if (tid) { | |
162 | log_wait_commit(journal, tid); | |
be07c4ed | 163 | } else { |
e219cca0 TT |
164 | printk(KERN_ERR "%s: needed %d blocks and " |
165 | "only had %d space available\n", | |
166 | __func__, nblocks, space_left); | |
167 | printk(KERN_ERR "%s: no way to get more " | |
168 | "journal space\n", __func__); | |
169 | WARN_ON(1); | |
be07c4ed DG |
170 | journal_abort(journal, 0); |
171 | } | |
1da177e4 | 172 | spin_lock(&journal->j_state_lock); |
be07c4ed DG |
173 | } else { |
174 | spin_unlock(&journal->j_list_lock); | |
1da177e4 | 175 | } |
2c68ee75 | 176 | mutex_unlock(&journal->j_checkpoint_mutex); |
1da177e4 LT |
177 | } |
178 | } | |
179 | ||
180 | /* | |
181 | * We were unable to perform jbd_trylock_bh_state() inside j_list_lock. | |
182 | * The caller must restart a list walk. Wait for someone else to run | |
183 | * jbd_unlock_bh_state(). | |
184 | */ | |
185 | static void jbd_sync_bh(journal_t *journal, struct buffer_head *bh) | |
e7ab8d65 | 186 | __releases(journal->j_list_lock) |
1da177e4 LT |
187 | { |
188 | get_bh(bh); | |
189 | spin_unlock(&journal->j_list_lock); | |
190 | jbd_lock_bh_state(bh); | |
191 | jbd_unlock_bh_state(bh); | |
192 | put_bh(bh); | |
193 | } | |
194 | ||
195 | /* | |
78ce89c9 JK |
196 | * Clean up transaction's list of buffers submitted for io. |
197 | * We wait for any pending IO to complete and remove any clean | |
198 | * buffers. Note that we take the buffers in the opposite ordering | |
199 | * from the one in which they were submitted for IO. | |
1da177e4 | 200 | * |
4afe9785 HK |
201 | * Return 0 on success, and return <0 if some buffers have failed |
202 | * to be written out. | |
203 | * | |
1da177e4 LT |
204 | * Called with j_list_lock held. |
205 | */ | |
4afe9785 | 206 | static int __wait_cp_io(journal_t *journal, transaction_t *transaction) |
1da177e4 | 207 | { |
78ce89c9 | 208 | struct journal_head *jh; |
1da177e4 | 209 | struct buffer_head *bh; |
78ce89c9 JK |
210 | tid_t this_tid; |
211 | int released = 0; | |
4afe9785 | 212 | int ret = 0; |
78ce89c9 JK |
213 | |
214 | this_tid = transaction->t_tid; | |
215 | restart: | |
216 | /* Did somebody clean up the transaction in the meanwhile? */ | |
217 | if (journal->j_checkpoint_transactions != transaction || | |
218 | transaction->t_tid != this_tid) | |
4afe9785 | 219 | return ret; |
78ce89c9 JK |
220 | while (!released && transaction->t_checkpoint_io_list) { |
221 | jh = transaction->t_checkpoint_io_list; | |
1da177e4 | 222 | bh = jh2bh(jh); |
78ce89c9 JK |
223 | if (!jbd_trylock_bh_state(bh)) { |
224 | jbd_sync_bh(journal, bh); | |
225 | spin_lock(&journal->j_list_lock); | |
226 | goto restart; | |
227 | } | |
bb189247 | 228 | get_bh(bh); |
1da177e4 | 229 | if (buffer_locked(bh)) { |
1da177e4 | 230 | spin_unlock(&journal->j_list_lock); |
78ce89c9 | 231 | jbd_unlock_bh_state(bh); |
1da177e4 LT |
232 | wait_on_buffer(bh); |
233 | /* the journal_head may have gone by now */ | |
234 | BUFFER_TRACE(bh, "brelse"); | |
235 | __brelse(bh); | |
78ce89c9 JK |
236 | spin_lock(&journal->j_list_lock); |
237 | goto restart; | |
1da177e4 | 238 | } |
9f818b4a | 239 | if (unlikely(buffer_write_io_error(bh))) |
4afe9785 HK |
240 | ret = -EIO; |
241 | ||
1da177e4 | 242 | /* |
78ce89c9 JK |
243 | * Now in whatever state the buffer currently is, we know that |
244 | * it has been written out and so we can drop it from the list | |
1da177e4 | 245 | */ |
78ce89c9 JK |
246 | released = __journal_remove_checkpoint(jh); |
247 | jbd_unlock_bh_state(bh); | |
78ce89c9 JK |
248 | __brelse(bh); |
249 | } | |
4afe9785 HK |
250 | |
251 | return ret; | |
1da177e4 LT |
252 | } |
253 | ||
254 | #define NR_BATCH 64 | |
255 | ||
256 | static void | |
257 | __flush_batch(journal_t *journal, struct buffer_head **bhs, int *batch_count) | |
258 | { | |
259 | int i; | |
260 | ||
9cb569d6 CH |
261 | for (i = 0; i < *batch_count; i++) |
262 | write_dirty_buffer(bhs[i], WRITE); | |
263 | ||
1da177e4 LT |
264 | for (i = 0; i < *batch_count; i++) { |
265 | struct buffer_head *bh = bhs[i]; | |
266 | clear_buffer_jwrite(bh); | |
267 | BUFFER_TRACE(bh, "brelse"); | |
268 | __brelse(bh); | |
269 | } | |
270 | *batch_count = 0; | |
271 | } | |
272 | ||
273 | /* | |
274 | * Try to flush one buffer from the checkpoint list to disk. | |
275 | * | |
276 | * Return 1 if something happened which requires us to abort the current | |
4afe9785 HK |
277 | * scan of the checkpoint list. Return <0 if the buffer has failed to |
278 | * be written out. | |
1da177e4 | 279 | * |
78ce89c9 | 280 | * Called with j_list_lock held and drops it if 1 is returned |
1da177e4 LT |
281 | * Called under jbd_lock_bh_state(jh2bh(jh)), and drops it |
282 | */ | |
78ce89c9 JK |
283 | static int __process_buffer(journal_t *journal, struct journal_head *jh, |
284 | struct buffer_head **bhs, int *batch_count) | |
1da177e4 LT |
285 | { |
286 | struct buffer_head *bh = jh2bh(jh); | |
287 | int ret = 0; | |
288 | ||
78ce89c9 | 289 | if (buffer_locked(bh)) { |
e4d5e3a4 | 290 | get_bh(bh); |
78ce89c9 JK |
291 | spin_unlock(&journal->j_list_lock); |
292 | jbd_unlock_bh_state(bh); | |
293 | wait_on_buffer(bh); | |
294 | /* the journal_head may have gone by now */ | |
295 | BUFFER_TRACE(bh, "brelse"); | |
296 | __brelse(bh); | |
297 | ret = 1; | |
298 | } else if (jh->b_transaction != NULL) { | |
299 | transaction_t *t = jh->b_transaction; | |
300 | tid_t tid = t->t_tid; | |
1da177e4 | 301 | |
78ce89c9 JK |
302 | spin_unlock(&journal->j_list_lock); |
303 | jbd_unlock_bh_state(bh); | |
304 | log_start_commit(journal, tid); | |
305 | log_wait_commit(journal, tid); | |
306 | ret = 1; | |
307 | } else if (!buffer_dirty(bh)) { | |
4afe9785 | 308 | ret = 1; |
9f818b4a | 309 | if (unlikely(buffer_write_io_error(bh))) |
4afe9785 | 310 | ret = -EIO; |
bb189247 | 311 | get_bh(bh); |
78ce89c9 JK |
312 | J_ASSERT_JH(jh, !buffer_jbddirty(bh)); |
313 | BUFFER_TRACE(bh, "remove from checkpoint"); | |
314 | __journal_remove_checkpoint(jh); | |
315 | spin_unlock(&journal->j_list_lock); | |
316 | jbd_unlock_bh_state(bh); | |
78ce89c9 | 317 | __brelse(bh); |
78ce89c9 | 318 | } else { |
1da177e4 LT |
319 | /* |
320 | * Important: we are about to write the buffer, and | |
321 | * possibly block, while still holding the journal lock. | |
322 | * We cannot afford to let the transaction logic start | |
323 | * messing around with this buffer before we write it to | |
ae6ddcc5 | 324 | * disk, as that would break recoverability. |
1da177e4 LT |
325 | */ |
326 | BUFFER_TRACE(bh, "queue"); | |
327 | get_bh(bh); | |
328 | J_ASSERT_BH(bh, !buffer_jwrite(bh)); | |
329 | set_buffer_jwrite(bh); | |
330 | bhs[*batch_count] = bh; | |
78ce89c9 | 331 | __buffer_relink_io(jh); |
1da177e4 LT |
332 | jbd_unlock_bh_state(bh); |
333 | (*batch_count)++; | |
334 | if (*batch_count == NR_BATCH) { | |
78ce89c9 | 335 | spin_unlock(&journal->j_list_lock); |
1da177e4 LT |
336 | __flush_batch(journal, bhs, batch_count); |
337 | ret = 1; | |
338 | } | |
1da177e4 LT |
339 | } |
340 | return ret; | |
341 | } | |
342 | ||
343 | /* | |
78ce89c9 JK |
344 | * Perform an actual checkpoint. We take the first transaction on the |
345 | * list of transactions to be checkpointed and send all its buffers | |
346 | * to disk. We submit larger chunks of data at once. | |
ae6ddcc5 | 347 | * |
1da177e4 | 348 | * The journal should be locked before calling this function. |
4afe9785 | 349 | * Called with j_checkpoint_mutex held. |
1da177e4 LT |
350 | */ |
351 | int log_do_checkpoint(journal_t *journal) | |
352 | { | |
78ce89c9 JK |
353 | transaction_t *transaction; |
354 | tid_t this_tid; | |
1da177e4 | 355 | int result; |
1da177e4 LT |
356 | |
357 | jbd_debug(1, "Start checkpoint\n"); | |
358 | ||
ae6ddcc5 | 359 | /* |
1da177e4 LT |
360 | * First thing: if there are any transactions in the log which |
361 | * don't need checkpointing, just eliminate them from the | |
ae6ddcc5 | 362 | * journal straight away. |
1da177e4 LT |
363 | */ |
364 | result = cleanup_journal_tail(journal); | |
99cb1a31 | 365 | trace_jbd_checkpoint(journal, result); |
1da177e4 LT |
366 | jbd_debug(1, "cleanup_journal_tail returned %d\n", result); |
367 | if (result <= 0) | |
368 | return result; | |
369 | ||
370 | /* | |
78ce89c9 JK |
371 | * OK, we need to start writing disk blocks. Take one transaction |
372 | * and write it. | |
1da177e4 | 373 | */ |
4afe9785 | 374 | result = 0; |
78ce89c9 JK |
375 | spin_lock(&journal->j_list_lock); |
376 | if (!journal->j_checkpoint_transactions) | |
377 | goto out; | |
378 | transaction = journal->j_checkpoint_transactions; | |
379 | this_tid = transaction->t_tid; | |
380 | restart: | |
1da177e4 | 381 | /* |
78ce89c9 JK |
382 | * If someone cleaned up this transaction while we slept, we're |
383 | * done (maybe it's a new transaction, but it fell at the same | |
384 | * address). | |
1da177e4 | 385 | */ |
78ce89c9 JK |
386 | if (journal->j_checkpoint_transactions == transaction && |
387 | transaction->t_tid == this_tid) { | |
388 | int batch_count = 0; | |
389 | struct buffer_head *bhs[NR_BATCH]; | |
390 | struct journal_head *jh; | |
4afe9785 | 391 | int retry = 0, err; |
78ce89c9 JK |
392 | |
393 | while (!retry && transaction->t_checkpoint_list) { | |
1da177e4 LT |
394 | struct buffer_head *bh; |
395 | ||
78ce89c9 | 396 | jh = transaction->t_checkpoint_list; |
1da177e4 LT |
397 | bh = jh2bh(jh); |
398 | if (!jbd_trylock_bh_state(bh)) { | |
399 | jbd_sync_bh(journal, bh); | |
1da177e4 LT |
400 | retry = 1; |
401 | break; | |
402 | } | |
78ce89c9 | 403 | retry = __process_buffer(journal, jh, bhs,&batch_count); |
4afe9785 HK |
404 | if (retry < 0 && !result) |
405 | result = retry; | |
95c354fe NP |
406 | if (!retry && (need_resched() || |
407 | spin_needbreak(&journal->j_list_lock))) { | |
78ce89c9 | 408 | spin_unlock(&journal->j_list_lock); |
1da177e4 LT |
409 | retry = 1; |
410 | break; | |
411 | } | |
78ce89c9 | 412 | } |
1da177e4 | 413 | |
00ea8145 | 414 | if (batch_count) { |
78ce89c9 JK |
415 | if (!retry) { |
416 | spin_unlock(&journal->j_list_lock); | |
417 | retry = 1; | |
418 | } | |
1da177e4 | 419 | __flush_batch(journal, bhs, &batch_count); |
00ea8145 | 420 | } |
1da177e4 | 421 | |
78ce89c9 JK |
422 | if (retry) { |
423 | spin_lock(&journal->j_list_lock); | |
424 | goto restart; | |
425 | } | |
1da177e4 | 426 | /* |
78ce89c9 JK |
427 | * Now we have cleaned up the first transaction's checkpoint |
428 | * list. Let's clean up the second one | |
7c8903f6 | 429 | */ |
4afe9785 HK |
430 | err = __wait_cp_io(journal, transaction); |
431 | if (!result) | |
432 | result = err; | |
1da177e4 | 433 | } |
78ce89c9 | 434 | out: |
1da177e4 | 435 | spin_unlock(&journal->j_list_lock); |
1da177e4 | 436 | if (result < 0) |
4afe9785 HK |
437 | journal_abort(journal, result); |
438 | else | |
439 | result = cleanup_journal_tail(journal); | |
440 | ||
441 | return (result < 0) ? result : 0; | |
1da177e4 LT |
442 | } |
443 | ||
444 | /* | |
445 | * Check the list of checkpoint transactions for the journal to see if | |
446 | * we have already got rid of any since the last update of the log tail | |
447 | * in the journal superblock. If so, we can instantly roll the | |
448 | * superblock forward to remove those transactions from the log. | |
ae6ddcc5 | 449 | * |
1da177e4 | 450 | * Return <0 on error, 0 on success, 1 if there was nothing to clean up. |
ae6ddcc5 | 451 | * |
1da177e4 LT |
452 | * Called with the journal lock held. |
453 | * | |
454 | * This is the only part of the journaling code which really needs to be | |
455 | * aware of transaction aborts. Checkpointing involves writing to the | |
456 | * main filesystem area rather than to the journal, so it can proceed | |
4afe9785 HK |
457 | * even in abort state, but we must not update the super block if |
458 | * checkpointing may have failed. Otherwise, we would lose some metadata | |
459 | * buffers which should be written-back to the filesystem. | |
1da177e4 LT |
460 | */ |
461 | ||
462 | int cleanup_journal_tail(journal_t *journal) | |
463 | { | |
464 | transaction_t * transaction; | |
465 | tid_t first_tid; | |
9c28cbcc | 466 | unsigned int blocknr, freed; |
1da177e4 | 467 | |
4afe9785 HK |
468 | if (is_journal_aborted(journal)) |
469 | return 1; | |
470 | ||
1da177e4 | 471 | /* OK, work out the oldest transaction remaining in the log, and |
ae6ddcc5 MC |
472 | * the log block it starts at. |
473 | * | |
1da177e4 LT |
474 | * If the log is now empty, we need to work out which is the |
475 | * next transaction ID we will write, and where it will | |
476 | * start. */ | |
477 | ||
478 | spin_lock(&journal->j_state_lock); | |
479 | spin_lock(&journal->j_list_lock); | |
480 | transaction = journal->j_checkpoint_transactions; | |
481 | if (transaction) { | |
482 | first_tid = transaction->t_tid; | |
483 | blocknr = transaction->t_log_start; | |
484 | } else if ((transaction = journal->j_committing_transaction) != NULL) { | |
485 | first_tid = transaction->t_tid; | |
486 | blocknr = transaction->t_log_start; | |
487 | } else if ((transaction = journal->j_running_transaction) != NULL) { | |
488 | first_tid = transaction->t_tid; | |
489 | blocknr = journal->j_head; | |
490 | } else { | |
491 | first_tid = journal->j_transaction_sequence; | |
492 | blocknr = journal->j_head; | |
493 | } | |
494 | spin_unlock(&journal->j_list_lock); | |
495 | J_ASSERT(blocknr != 0); | |
496 | ||
497 | /* If the oldest pinned transaction is at the tail of the log | |
498 | already then there's not much we can do right now. */ | |
499 | if (journal->j_tail_sequence == first_tid) { | |
500 | spin_unlock(&journal->j_state_lock); | |
501 | return 1; | |
502 | } | |
503 | ||
504 | /* OK, update the superblock to recover the freed space. | |
505 | * Physical blocks come first: have we wrapped beyond the end of | |
506 | * the log? */ | |
507 | freed = blocknr - journal->j_tail; | |
508 | if (blocknr < journal->j_tail) | |
509 | freed = freed + journal->j_last - journal->j_first; | |
510 | ||
99cb1a31 | 511 | trace_jbd_cleanup_journal_tail(journal, first_tid, blocknr, freed); |
1da177e4 | 512 | jbd_debug(1, |
9c28cbcc JK |
513 | "Cleaning journal tail from %d to %d (offset %u), " |
514 | "freeing %u\n", | |
1da177e4 LT |
515 | journal->j_tail_sequence, first_tid, blocknr, freed); |
516 | ||
517 | journal->j_free += freed; | |
518 | journal->j_tail_sequence = first_tid; | |
519 | journal->j_tail = blocknr; | |
520 | spin_unlock(&journal->j_state_lock); | |
521 | if (!(journal->j_flags & JFS_ABORT)) | |
522 | journal_update_superblock(journal, 1); | |
523 | return 0; | |
524 | } | |
525 | ||
526 | ||
527 | /* Checkpoint list management */ | |
528 | ||
78ce89c9 JK |
529 | /* |
530 | * journal_clean_one_cp_list | |
531 | * | |
bb189247 JK |
532 | * Find all the written-back checkpoint buffers in the given list and release |
533 | * them. | |
78ce89c9 | 534 | * |
78ce89c9 JK |
535 | * Called with j_list_lock held. |
536 | * Returns number of bufers reaped (for debug) | |
537 | */ | |
538 | ||
539 | static int journal_clean_one_cp_list(struct journal_head *jh, int *released) | |
540 | { | |
541 | struct journal_head *last_jh; | |
542 | struct journal_head *next_jh = jh; | |
543 | int ret, freed = 0; | |
544 | ||
545 | *released = 0; | |
546 | if (!jh) | |
547 | return 0; | |
548 | ||
e9ad5620 | 549 | last_jh = jh->b_cpprev; |
78ce89c9 JK |
550 | do { |
551 | jh = next_jh; | |
552 | next_jh = jh->b_cpnext; | |
553 | /* Use trylock because of the ranking */ | |
554 | if (jbd_trylock_bh_state(jh2bh(jh))) { | |
555 | ret = __try_to_free_cp_buf(jh); | |
556 | if (ret) { | |
557 | freed++; | |
558 | if (ret == 2) { | |
559 | *released = 1; | |
560 | return freed; | |
561 | } | |
562 | } | |
563 | } | |
564 | /* | |
565 | * This function only frees up some memory | |
566 | * if possible so we dont have an obligation | |
567 | * to finish processing. Bail out if preemption | |
568 | * requested: | |
569 | */ | |
570 | if (need_resched()) | |
571 | return freed; | |
572 | } while (jh != last_jh); | |
573 | ||
574 | return freed; | |
575 | } | |
576 | ||
1da177e4 LT |
577 | /* |
578 | * journal_clean_checkpoint_list | |
579 | * | |
580 | * Find all the written-back checkpoint buffers in the journal and release them. | |
581 | * | |
582 | * Called with the journal locked. | |
583 | * Called with j_list_lock held. | |
78ce89c9 | 584 | * Returns number of buffers reaped (for debug) |
1da177e4 LT |
585 | */ |
586 | ||
587 | int __journal_clean_checkpoint_list(journal_t *journal) | |
588 | { | |
589 | transaction_t *transaction, *last_transaction, *next_transaction; | |
7c8903f6 | 590 | int ret = 0; |
78ce89c9 | 591 | int released; |
1da177e4 LT |
592 | |
593 | transaction = journal->j_checkpoint_transactions; | |
78ce89c9 | 594 | if (!transaction) |
1da177e4 LT |
595 | goto out; |
596 | ||
597 | last_transaction = transaction->t_cpprev; | |
598 | next_transaction = transaction; | |
599 | do { | |
1da177e4 LT |
600 | transaction = next_transaction; |
601 | next_transaction = transaction->t_cpnext; | |
78ce89c9 JK |
602 | ret += journal_clean_one_cp_list(transaction-> |
603 | t_checkpoint_list, &released); | |
604 | /* | |
605 | * This function only frees up some memory if possible so we | |
606 | * dont have an obligation to finish processing. Bail out if | |
607 | * preemption requested: | |
608 | */ | |
609 | if (need_resched()) | |
610 | goto out; | |
611 | if (released) | |
612 | continue; | |
613 | /* | |
614 | * It is essential that we are as careful as in the case of | |
615 | * t_checkpoint_list with removing the buffer from the list as | |
616 | * we can possibly see not yet submitted buffers on io_list | |
617 | */ | |
618 | ret += journal_clean_one_cp_list(transaction-> | |
619 | t_checkpoint_io_list, &released); | |
620 | if (need_resched()) | |
621 | goto out; | |
1da177e4 LT |
622 | } while (transaction != last_transaction); |
623 | out: | |
624 | return ret; | |
625 | } | |
626 | ||
ae6ddcc5 | 627 | /* |
1da177e4 LT |
628 | * journal_remove_checkpoint: called after a buffer has been committed |
629 | * to disk (either by being write-back flushed to disk, or being | |
630 | * committed to the log). | |
631 | * | |
632 | * We cannot safely clean a transaction out of the log until all of the | |
633 | * buffer updates committed in that transaction have safely been stored | |
634 | * elsewhere on disk. To achieve this, all of the buffers in a | |
635 | * transaction need to be maintained on the transaction's checkpoint | |
78ce89c9 | 636 | * lists until they have been rewritten, at which point this function is |
1da177e4 | 637 | * called to remove the buffer from the existing transaction's |
78ce89c9 JK |
638 | * checkpoint lists. |
639 | * | |
640 | * The function returns 1 if it frees the transaction, 0 otherwise. | |
bb189247 | 641 | * The function can free jh and bh. |
1da177e4 | 642 | * |
1da177e4 | 643 | * This function is called with j_list_lock held. |
78ce89c9 | 644 | * This function is called with jbd_lock_bh_state(jh2bh(jh)) |
1da177e4 LT |
645 | */ |
646 | ||
78ce89c9 | 647 | int __journal_remove_checkpoint(struct journal_head *jh) |
1da177e4 LT |
648 | { |
649 | transaction_t *transaction; | |
650 | journal_t *journal; | |
78ce89c9 | 651 | int ret = 0; |
1da177e4 LT |
652 | |
653 | JBUFFER_TRACE(jh, "entry"); | |
654 | ||
655 | if ((transaction = jh->b_cp_transaction) == NULL) { | |
656 | JBUFFER_TRACE(jh, "not on transaction"); | |
657 | goto out; | |
658 | } | |
659 | journal = transaction->t_journal; | |
660 | ||
bb189247 | 661 | JBUFFER_TRACE(jh, "removing from transaction"); |
1da177e4 | 662 | __buffer_unlink(jh); |
78ce89c9 | 663 | jh->b_cp_transaction = NULL; |
bb189247 | 664 | journal_put_journal_head(jh); |
1da177e4 | 665 | |
78ce89c9 JK |
666 | if (transaction->t_checkpoint_list != NULL || |
667 | transaction->t_checkpoint_io_list != NULL) | |
1da177e4 | 668 | goto out; |
1da177e4 LT |
669 | |
670 | /* | |
671 | * There is one special case to worry about: if we have just pulled the | |
d4beaf4a JK |
672 | * buffer off a running or committing transaction's checkpoing list, |
673 | * then even if the checkpoint list is empty, the transaction obviously | |
674 | * cannot be dropped! | |
1da177e4 | 675 | * |
d4beaf4a | 676 | * The locking here around t_state is a bit sleazy. |
1da177e4 LT |
677 | * See the comment at the end of journal_commit_transaction(). |
678 | */ | |
bb189247 | 679 | if (transaction->t_state != T_FINISHED) |
1da177e4 | 680 | goto out; |
1da177e4 LT |
681 | |
682 | /* OK, that was the last buffer for the transaction: we can now | |
683 | safely remove this transaction from the log */ | |
684 | ||
685 | __journal_drop_transaction(journal, transaction); | |
686 | ||
687 | /* Just in case anybody was waiting for more transactions to be | |
688 | checkpointed... */ | |
689 | wake_up(&journal->j_wait_logspace); | |
78ce89c9 | 690 | ret = 1; |
1da177e4 | 691 | out: |
78ce89c9 | 692 | return ret; |
1da177e4 LT |
693 | } |
694 | ||
695 | /* | |
696 | * journal_insert_checkpoint: put a committed buffer onto a checkpoint | |
697 | * list so that we know when it is safe to clean the transaction out of | |
698 | * the log. | |
699 | * | |
700 | * Called with the journal locked. | |
701 | * Called with j_list_lock held. | |
702 | */ | |
ae6ddcc5 | 703 | void __journal_insert_checkpoint(struct journal_head *jh, |
1da177e4 LT |
704 | transaction_t *transaction) |
705 | { | |
706 | JBUFFER_TRACE(jh, "entry"); | |
707 | J_ASSERT_JH(jh, buffer_dirty(jh2bh(jh)) || buffer_jbddirty(jh2bh(jh))); | |
708 | J_ASSERT_JH(jh, jh->b_cp_transaction == NULL); | |
709 | ||
bb189247 JK |
710 | /* Get reference for checkpointing transaction */ |
711 | journal_grab_journal_head(jh2bh(jh)); | |
1da177e4 LT |
712 | jh->b_cp_transaction = transaction; |
713 | ||
714 | if (!transaction->t_checkpoint_list) { | |
715 | jh->b_cpnext = jh->b_cpprev = jh; | |
716 | } else { | |
717 | jh->b_cpnext = transaction->t_checkpoint_list; | |
718 | jh->b_cpprev = transaction->t_checkpoint_list->b_cpprev; | |
719 | jh->b_cpprev->b_cpnext = jh; | |
720 | jh->b_cpnext->b_cpprev = jh; | |
721 | } | |
722 | transaction->t_checkpoint_list = jh; | |
723 | } | |
724 | ||
725 | /* | |
726 | * We've finished with this transaction structure: adios... | |
ae6ddcc5 | 727 | * |
1da177e4 LT |
728 | * The transaction must have no links except for the checkpoint by this |
729 | * point. | |
730 | * | |
731 | * Called with the journal locked. | |
732 | * Called with j_list_lock held. | |
733 | */ | |
734 | ||
735 | void __journal_drop_transaction(journal_t *journal, transaction_t *transaction) | |
736 | { | |
737 | assert_spin_locked(&journal->j_list_lock); | |
738 | if (transaction->t_cpnext) { | |
739 | transaction->t_cpnext->t_cpprev = transaction->t_cpprev; | |
740 | transaction->t_cpprev->t_cpnext = transaction->t_cpnext; | |
741 | if (journal->j_checkpoint_transactions == transaction) | |
742 | journal->j_checkpoint_transactions = | |
743 | transaction->t_cpnext; | |
744 | if (journal->j_checkpoint_transactions == transaction) | |
745 | journal->j_checkpoint_transactions = NULL; | |
746 | } | |
747 | ||
748 | J_ASSERT(transaction->t_state == T_FINISHED); | |
749 | J_ASSERT(transaction->t_buffers == NULL); | |
750 | J_ASSERT(transaction->t_sync_datalist == NULL); | |
751 | J_ASSERT(transaction->t_forget == NULL); | |
752 | J_ASSERT(transaction->t_iobuf_list == NULL); | |
753 | J_ASSERT(transaction->t_shadow_list == NULL); | |
754 | J_ASSERT(transaction->t_log_list == NULL); | |
755 | J_ASSERT(transaction->t_checkpoint_list == NULL); | |
78ce89c9 | 756 | J_ASSERT(transaction->t_checkpoint_io_list == NULL); |
1da177e4 LT |
757 | J_ASSERT(transaction->t_updates == 0); |
758 | J_ASSERT(journal->j_committing_transaction != transaction); | |
759 | J_ASSERT(journal->j_running_transaction != transaction); | |
760 | ||
99cb1a31 | 761 | trace_jbd_drop_transaction(journal, transaction); |
1da177e4 LT |
762 | jbd_debug(1, "Dropping transaction %d, all done\n", transaction->t_tid); |
763 | kfree(transaction); | |
764 | } |