Commit | Line | Data |
---|---|---|
470decc6 | 1 | /* |
58862699 | 2 | * linux/fs/jbd2/checkpoint.c |
470decc6 DK |
3 | * |
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 | * | |
12 | * Checkpoint routines for the generic filesystem journaling code. | |
13 | * Part of the ext2fs journaling system. | |
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> | |
f7f4bccb | 22 | #include <linux/jbd2.h> |
470decc6 DK |
23 | #include <linux/errno.h> |
24 | #include <linux/slab.h> | |
879c5e6b | 25 | #include <trace/events/jbd2.h> |
470decc6 DK |
26 | |
27 | /* | |
28 | * Unlink a buffer from a transaction checkpoint list. | |
29 | * | |
30 | * Called with j_list_lock held. | |
31 | */ | |
32 | static inline void __buffer_unlink_first(struct journal_head *jh) | |
33 | { | |
34 | transaction_t *transaction = jh->b_cp_transaction; | |
35 | ||
36 | jh->b_cpnext->b_cpprev = jh->b_cpprev; | |
37 | jh->b_cpprev->b_cpnext = jh->b_cpnext; | |
38 | if (transaction->t_checkpoint_list == jh) { | |
39 | transaction->t_checkpoint_list = jh->b_cpnext; | |
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; | |
82 | } | |
83 | ||
84 | /* | |
85 | * Try to release a checkpointed buffer from its transaction. | |
86 | * Returns 1 if we released it and 2 if we also released the | |
87 | * whole transaction. | |
88 | * | |
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 | ||
44519faf HK |
97 | if (jh->b_jlist == BJ_None && !buffer_locked(bh) && |
98 | !buffer_dirty(bh) && !buffer_write_io_error(bh)) { | |
470decc6 | 99 | JBUFFER_TRACE(jh, "remove from checkpoint list"); |
f7f4bccb | 100 | ret = __jbd2_journal_remove_checkpoint(jh) + 1; |
470decc6 | 101 | jbd_unlock_bh_state(bh); |
f7f4bccb | 102 | jbd2_journal_remove_journal_head(bh); |
470decc6 DK |
103 | BUFFER_TRACE(bh, "release"); |
104 | __brelse(bh); | |
105 | } else { | |
106 | jbd_unlock_bh_state(bh); | |
107 | } | |
108 | return ret; | |
109 | } | |
110 | ||
111 | /* | |
f7f4bccb | 112 | * __jbd2_log_wait_for_space: wait until there is space in the journal. |
470decc6 DK |
113 | * |
114 | * Called under j-state_lock *only*. It will be unlocked if we have to wait | |
115 | * for a checkpoint to free up some space in the log. | |
116 | */ | |
f7f4bccb | 117 | void __jbd2_log_wait_for_space(journal_t *journal) |
470decc6 | 118 | { |
8c3f25d8 | 119 | int nblocks, space_left; |
470decc6 DK |
120 | assert_spin_locked(&journal->j_state_lock); |
121 | ||
122 | nblocks = jbd_space_needed(journal); | |
f7f4bccb MC |
123 | while (__jbd2_log_space_left(journal) < nblocks) { |
124 | if (journal->j_flags & JBD2_ABORT) | |
470decc6 DK |
125 | return; |
126 | spin_unlock(&journal->j_state_lock); | |
127 | mutex_lock(&journal->j_checkpoint_mutex); | |
128 | ||
129 | /* | |
130 | * Test again, another process may have checkpointed while we | |
23f8b79e | 131 | * were waiting for the checkpoint lock. If there are no |
8c3f25d8 TT |
132 | * transactions ready to be checkpointed, try to recover |
133 | * journal space by calling cleanup_journal_tail(), and if | |
134 | * that doesn't work, by waiting for the currently committing | |
135 | * transaction to complete. If there is absolutely no way | |
136 | * to make progress, this is either a BUG or corrupted | |
137 | * filesystem, so abort the journal and leave a stack | |
138 | * trace for forensic evidence. | |
470decc6 DK |
139 | */ |
140 | spin_lock(&journal->j_state_lock); | |
23f8b79e | 141 | spin_lock(&journal->j_list_lock); |
470decc6 | 142 | nblocks = jbd_space_needed(journal); |
8c3f25d8 TT |
143 | space_left = __jbd2_log_space_left(journal); |
144 | if (space_left < nblocks) { | |
23f8b79e | 145 | int chkpt = journal->j_checkpoint_transactions != NULL; |
8c3f25d8 | 146 | tid_t tid = 0; |
23f8b79e | 147 | |
8c3f25d8 TT |
148 | if (journal->j_committing_transaction) |
149 | tid = journal->j_committing_transaction->t_tid; | |
23f8b79e | 150 | spin_unlock(&journal->j_list_lock); |
470decc6 | 151 | spin_unlock(&journal->j_state_lock); |
23f8b79e DG |
152 | if (chkpt) { |
153 | jbd2_log_do_checkpoint(journal); | |
8c3f25d8 TT |
154 | } else if (jbd2_cleanup_journal_tail(journal) == 0) { |
155 | /* We were able to recover space; yay! */ | |
156 | ; | |
157 | } else if (tid) { | |
158 | jbd2_log_wait_commit(journal, tid); | |
23f8b79e | 159 | } else { |
8c3f25d8 TT |
160 | printk(KERN_ERR "%s: needed %d blocks and " |
161 | "only had %d space available\n", | |
162 | __func__, nblocks, space_left); | |
163 | printk(KERN_ERR "%s: no way to get more " | |
164 | "journal space in %s\n", __func__, | |
165 | journal->j_devname); | |
166 | WARN_ON(1); | |
23f8b79e DG |
167 | jbd2_journal_abort(journal, 0); |
168 | } | |
470decc6 | 169 | spin_lock(&journal->j_state_lock); |
23f8b79e DG |
170 | } else { |
171 | spin_unlock(&journal->j_list_lock); | |
470decc6 DK |
172 | } |
173 | mutex_unlock(&journal->j_checkpoint_mutex); | |
174 | } | |
175 | } | |
176 | ||
177 | /* | |
178 | * We were unable to perform jbd_trylock_bh_state() inside j_list_lock. | |
179 | * The caller must restart a list walk. Wait for someone else to run | |
180 | * jbd_unlock_bh_state(). | |
181 | */ | |
182 | static void jbd_sync_bh(journal_t *journal, struct buffer_head *bh) | |
183 | __releases(journal->j_list_lock) | |
184 | { | |
185 | get_bh(bh); | |
186 | spin_unlock(&journal->j_list_lock); | |
187 | jbd_lock_bh_state(bh); | |
188 | jbd_unlock_bh_state(bh); | |
189 | put_bh(bh); | |
190 | } | |
191 | ||
192 | /* | |
193 | * Clean up transaction's list of buffers submitted for io. | |
194 | * We wait for any pending IO to complete and remove any clean | |
195 | * buffers. Note that we take the buffers in the opposite ordering | |
196 | * from the one in which they were submitted for IO. | |
197 | * | |
44519faf HK |
198 | * Return 0 on success, and return <0 if some buffers have failed |
199 | * to be written out. | |
200 | * | |
470decc6 DK |
201 | * Called with j_list_lock held. |
202 | */ | |
44519faf | 203 | static int __wait_cp_io(journal_t *journal, transaction_t *transaction) |
470decc6 DK |
204 | { |
205 | struct journal_head *jh; | |
206 | struct buffer_head *bh; | |
207 | tid_t this_tid; | |
208 | int released = 0; | |
44519faf | 209 | int ret = 0; |
470decc6 DK |
210 | |
211 | this_tid = transaction->t_tid; | |
212 | restart: | |
213 | /* Did somebody clean up the transaction in the meanwhile? */ | |
214 | if (journal->j_checkpoint_transactions != transaction || | |
215 | transaction->t_tid != this_tid) | |
44519faf | 216 | return ret; |
470decc6 DK |
217 | while (!released && transaction->t_checkpoint_io_list) { |
218 | jh = transaction->t_checkpoint_io_list; | |
219 | bh = jh2bh(jh); | |
220 | if (!jbd_trylock_bh_state(bh)) { | |
221 | jbd_sync_bh(journal, bh); | |
222 | spin_lock(&journal->j_list_lock); | |
223 | goto restart; | |
224 | } | |
225 | if (buffer_locked(bh)) { | |
226 | atomic_inc(&bh->b_count); | |
227 | spin_unlock(&journal->j_list_lock); | |
228 | jbd_unlock_bh_state(bh); | |
229 | wait_on_buffer(bh); | |
230 | /* the journal_head may have gone by now */ | |
231 | BUFFER_TRACE(bh, "brelse"); | |
232 | __brelse(bh); | |
233 | spin_lock(&journal->j_list_lock); | |
234 | goto restart; | |
235 | } | |
44519faf HK |
236 | if (unlikely(buffer_write_io_error(bh))) |
237 | ret = -EIO; | |
238 | ||
470decc6 DK |
239 | /* |
240 | * Now in whatever state the buffer currently is, we know that | |
241 | * it has been written out and so we can drop it from the list | |
242 | */ | |
f7f4bccb | 243 | released = __jbd2_journal_remove_checkpoint(jh); |
470decc6 | 244 | jbd_unlock_bh_state(bh); |
f7f4bccb | 245 | jbd2_journal_remove_journal_head(bh); |
470decc6 DK |
246 | __brelse(bh); |
247 | } | |
44519faf HK |
248 | |
249 | return ret; | |
470decc6 DK |
250 | } |
251 | ||
470decc6 | 252 | static void |
1a0d3786 | 253 | __flush_batch(journal_t *journal, int *batch_count) |
470decc6 DK |
254 | { |
255 | int i; | |
256 | ||
1a0d3786 | 257 | ll_rw_block(SWRITE, *batch_count, journal->j_chkpt_bhs); |
470decc6 | 258 | for (i = 0; i < *batch_count; i++) { |
1a0d3786 | 259 | struct buffer_head *bh = journal->j_chkpt_bhs[i]; |
470decc6 DK |
260 | clear_buffer_jwrite(bh); |
261 | BUFFER_TRACE(bh, "brelse"); | |
262 | __brelse(bh); | |
263 | } | |
264 | *batch_count = 0; | |
265 | } | |
266 | ||
267 | /* | |
268 | * Try to flush one buffer from the checkpoint list to disk. | |
269 | * | |
270 | * Return 1 if something happened which requires us to abort the current | |
44519faf HK |
271 | * scan of the checkpoint list. Return <0 if the buffer has failed to |
272 | * be written out. | |
470decc6 DK |
273 | * |
274 | * Called with j_list_lock held and drops it if 1 is returned | |
275 | * Called under jbd_lock_bh_state(jh2bh(jh)), and drops it | |
276 | */ | |
277 | static int __process_buffer(journal_t *journal, struct journal_head *jh, | |
1a0d3786 | 278 | int *batch_count, transaction_t *transaction) |
470decc6 DK |
279 | { |
280 | struct buffer_head *bh = jh2bh(jh); | |
281 | int ret = 0; | |
282 | ||
283 | if (buffer_locked(bh)) { | |
284 | atomic_inc(&bh->b_count); | |
285 | spin_unlock(&journal->j_list_lock); | |
286 | jbd_unlock_bh_state(bh); | |
287 | wait_on_buffer(bh); | |
288 | /* the journal_head may have gone by now */ | |
289 | BUFFER_TRACE(bh, "brelse"); | |
290 | __brelse(bh); | |
291 | ret = 1; | |
292 | } else if (jh->b_transaction != NULL) { | |
293 | transaction_t *t = jh->b_transaction; | |
294 | tid_t tid = t->t_tid; | |
295 | ||
8e85fb3f | 296 | transaction->t_chp_stats.cs_forced_to_close++; |
470decc6 DK |
297 | spin_unlock(&journal->j_list_lock); |
298 | jbd_unlock_bh_state(bh); | |
f7f4bccb MC |
299 | jbd2_log_start_commit(journal, tid); |
300 | jbd2_log_wait_commit(journal, tid); | |
470decc6 DK |
301 | ret = 1; |
302 | } else if (!buffer_dirty(bh)) { | |
44519faf HK |
303 | ret = 1; |
304 | if (unlikely(buffer_write_io_error(bh))) | |
305 | ret = -EIO; | |
470decc6 DK |
306 | J_ASSERT_JH(jh, !buffer_jbddirty(bh)); |
307 | BUFFER_TRACE(bh, "remove from checkpoint"); | |
f7f4bccb | 308 | __jbd2_journal_remove_checkpoint(jh); |
470decc6 DK |
309 | spin_unlock(&journal->j_list_lock); |
310 | jbd_unlock_bh_state(bh); | |
f7f4bccb | 311 | jbd2_journal_remove_journal_head(bh); |
470decc6 | 312 | __brelse(bh); |
470decc6 DK |
313 | } else { |
314 | /* | |
315 | * Important: we are about to write the buffer, and | |
316 | * possibly block, while still holding the journal lock. | |
317 | * We cannot afford to let the transaction logic start | |
318 | * messing around with this buffer before we write it to | |
319 | * disk, as that would break recoverability. | |
320 | */ | |
321 | BUFFER_TRACE(bh, "queue"); | |
322 | get_bh(bh); | |
323 | J_ASSERT_BH(bh, !buffer_jwrite(bh)); | |
324 | set_buffer_jwrite(bh); | |
1a0d3786 | 325 | journal->j_chkpt_bhs[*batch_count] = bh; |
470decc6 DK |
326 | __buffer_relink_io(jh); |
327 | jbd_unlock_bh_state(bh); | |
8e85fb3f | 328 | transaction->t_chp_stats.cs_written++; |
470decc6 | 329 | (*batch_count)++; |
1a0d3786 | 330 | if (*batch_count == JBD2_NR_BATCH) { |
470decc6 | 331 | spin_unlock(&journal->j_list_lock); |
1a0d3786 | 332 | __flush_batch(journal, batch_count); |
470decc6 DK |
333 | ret = 1; |
334 | } | |
335 | } | |
336 | return ret; | |
337 | } | |
338 | ||
339 | /* | |
340 | * Perform an actual checkpoint. We take the first transaction on the | |
341 | * list of transactions to be checkpointed and send all its buffers | |
342 | * to disk. We submit larger chunks of data at once. | |
343 | * | |
344 | * The journal should be locked before calling this function. | |
44519faf | 345 | * Called with j_checkpoint_mutex held. |
470decc6 | 346 | */ |
f7f4bccb | 347 | int jbd2_log_do_checkpoint(journal_t *journal) |
470decc6 DK |
348 | { |
349 | transaction_t *transaction; | |
350 | tid_t this_tid; | |
351 | int result; | |
352 | ||
353 | jbd_debug(1, "Start checkpoint\n"); | |
354 | ||
355 | /* | |
356 | * First thing: if there are any transactions in the log which | |
357 | * don't need checkpointing, just eliminate them from the | |
358 | * journal straight away. | |
359 | */ | |
f7f4bccb | 360 | result = jbd2_cleanup_journal_tail(journal); |
879c5e6b | 361 | trace_jbd2_checkpoint(journal, result); |
470decc6 DK |
362 | jbd_debug(1, "cleanup_journal_tail returned %d\n", result); |
363 | if (result <= 0) | |
364 | return result; | |
365 | ||
366 | /* | |
367 | * OK, we need to start writing disk blocks. Take one transaction | |
368 | * and write it. | |
369 | */ | |
44519faf | 370 | result = 0; |
470decc6 DK |
371 | spin_lock(&journal->j_list_lock); |
372 | if (!journal->j_checkpoint_transactions) | |
373 | goto out; | |
374 | transaction = journal->j_checkpoint_transactions; | |
8e85fb3f JL |
375 | if (transaction->t_chp_stats.cs_chp_time == 0) |
376 | transaction->t_chp_stats.cs_chp_time = jiffies; | |
470decc6 DK |
377 | this_tid = transaction->t_tid; |
378 | restart: | |
379 | /* | |
380 | * If someone cleaned up this transaction while we slept, we're | |
381 | * done (maybe it's a new transaction, but it fell at the same | |
382 | * address). | |
383 | */ | |
384 | if (journal->j_checkpoint_transactions == transaction && | |
385 | transaction->t_tid == this_tid) { | |
386 | int batch_count = 0; | |
470decc6 | 387 | struct journal_head *jh; |
44519faf | 388 | int retry = 0, err; |
470decc6 DK |
389 | |
390 | while (!retry && transaction->t_checkpoint_list) { | |
391 | struct buffer_head *bh; | |
392 | ||
393 | jh = transaction->t_checkpoint_list; | |
394 | bh = jh2bh(jh); | |
395 | if (!jbd_trylock_bh_state(bh)) { | |
396 | jbd_sync_bh(journal, bh); | |
397 | retry = 1; | |
398 | break; | |
399 | } | |
1a0d3786 | 400 | retry = __process_buffer(journal, jh, &batch_count, |
8e85fb3f | 401 | transaction); |
44519faf HK |
402 | if (retry < 0 && !result) |
403 | result = retry; | |
95c354fe NP |
404 | if (!retry && (need_resched() || |
405 | spin_needbreak(&journal->j_list_lock))) { | |
470decc6 DK |
406 | spin_unlock(&journal->j_list_lock); |
407 | retry = 1; | |
408 | break; | |
409 | } | |
410 | } | |
411 | ||
412 | if (batch_count) { | |
413 | if (!retry) { | |
414 | spin_unlock(&journal->j_list_lock); | |
415 | retry = 1; | |
416 | } | |
1a0d3786 | 417 | __flush_batch(journal, &batch_count); |
470decc6 DK |
418 | } |
419 | ||
420 | if (retry) { | |
421 | spin_lock(&journal->j_list_lock); | |
422 | goto restart; | |
423 | } | |
424 | /* | |
425 | * Now we have cleaned up the first transaction's checkpoint | |
426 | * list. Let's clean up the second one | |
427 | */ | |
44519faf HK |
428 | err = __wait_cp_io(journal, transaction); |
429 | if (!result) | |
430 | result = err; | |
470decc6 DK |
431 | } |
432 | out: | |
433 | spin_unlock(&journal->j_list_lock); | |
470decc6 | 434 | if (result < 0) |
44519faf HK |
435 | jbd2_journal_abort(journal, result); |
436 | else | |
437 | result = jbd2_cleanup_journal_tail(journal); | |
438 | ||
439 | return (result < 0) ? result : 0; | |
470decc6 DK |
440 | } |
441 | ||
442 | /* | |
443 | * Check the list of checkpoint transactions for the journal to see if | |
444 | * we have already got rid of any since the last update of the log tail | |
445 | * in the journal superblock. If so, we can instantly roll the | |
446 | * superblock forward to remove those transactions from the log. | |
447 | * | |
448 | * Return <0 on error, 0 on success, 1 if there was nothing to clean up. | |
449 | * | |
450 | * Called with the journal lock held. | |
451 | * | |
452 | * This is the only part of the journaling code which really needs to be | |
453 | * aware of transaction aborts. Checkpointing involves writing to the | |
454 | * main filesystem area rather than to the journal, so it can proceed | |
44519faf HK |
455 | * even in abort state, but we must not update the super block if |
456 | * checkpointing may have failed. Otherwise, we would lose some metadata | |
457 | * buffers which should be written-back to the filesystem. | |
470decc6 DK |
458 | */ |
459 | ||
f7f4bccb | 460 | int jbd2_cleanup_journal_tail(journal_t *journal) |
470decc6 DK |
461 | { |
462 | transaction_t * transaction; | |
463 | tid_t first_tid; | |
464 | unsigned long blocknr, freed; | |
465 | ||
44519faf HK |
466 | if (is_journal_aborted(journal)) |
467 | return 1; | |
468 | ||
470decc6 DK |
469 | /* OK, work out the oldest transaction remaining in the log, and |
470 | * the log block it starts at. | |
471 | * | |
472 | * If the log is now empty, we need to work out which is the | |
473 | * next transaction ID we will write, and where it will | |
474 | * start. */ | |
475 | ||
476 | spin_lock(&journal->j_state_lock); | |
477 | spin_lock(&journal->j_list_lock); | |
478 | transaction = journal->j_checkpoint_transactions; | |
479 | if (transaction) { | |
480 | first_tid = transaction->t_tid; | |
481 | blocknr = transaction->t_log_start; | |
482 | } else if ((transaction = journal->j_committing_transaction) != NULL) { | |
483 | first_tid = transaction->t_tid; | |
484 | blocknr = transaction->t_log_start; | |
485 | } else if ((transaction = journal->j_running_transaction) != NULL) { | |
486 | first_tid = transaction->t_tid; | |
487 | blocknr = journal->j_head; | |
488 | } else { | |
489 | first_tid = journal->j_transaction_sequence; | |
490 | blocknr = journal->j_head; | |
491 | } | |
492 | spin_unlock(&journal->j_list_lock); | |
493 | J_ASSERT(blocknr != 0); | |
494 | ||
495 | /* If the oldest pinned transaction is at the tail of the log | |
496 | already then there's not much we can do right now. */ | |
497 | if (journal->j_tail_sequence == first_tid) { | |
498 | spin_unlock(&journal->j_state_lock); | |
499 | return 1; | |
500 | } | |
501 | ||
502 | /* OK, update the superblock to recover the freed space. | |
503 | * Physical blocks come first: have we wrapped beyond the end of | |
504 | * the log? */ | |
505 | freed = blocknr - journal->j_tail; | |
506 | if (blocknr < journal->j_tail) | |
507 | freed = freed + journal->j_last - journal->j_first; | |
508 | ||
509 | jbd_debug(1, | |
510 | "Cleaning journal tail from %d to %d (offset %lu), " | |
511 | "freeing %lu\n", | |
512 | journal->j_tail_sequence, first_tid, blocknr, freed); | |
513 | ||
514 | journal->j_free += freed; | |
515 | journal->j_tail_sequence = first_tid; | |
516 | journal->j_tail = blocknr; | |
517 | spin_unlock(&journal->j_state_lock); | |
f7f4bccb MC |
518 | if (!(journal->j_flags & JBD2_ABORT)) |
519 | jbd2_journal_update_superblock(journal, 1); | |
470decc6 DK |
520 | return 0; |
521 | } | |
522 | ||
523 | ||
524 | /* Checkpoint list management */ | |
525 | ||
526 | /* | |
527 | * journal_clean_one_cp_list | |
528 | * | |
529 | * Find all the written-back checkpoint buffers in the given list and release them. | |
530 | * | |
531 | * Called with the journal locked. | |
532 | * Called with j_list_lock held. | |
533 | * Returns number of bufers reaped (for debug) | |
534 | */ | |
535 | ||
536 | static int journal_clean_one_cp_list(struct journal_head *jh, int *released) | |
537 | { | |
538 | struct journal_head *last_jh; | |
539 | struct journal_head *next_jh = jh; | |
540 | int ret, freed = 0; | |
541 | ||
542 | *released = 0; | |
543 | if (!jh) | |
544 | return 0; | |
545 | ||
546 | last_jh = jh->b_cpprev; | |
547 | do { | |
548 | jh = next_jh; | |
549 | next_jh = jh->b_cpnext; | |
550 | /* Use trylock because of the ranking */ | |
551 | if (jbd_trylock_bh_state(jh2bh(jh))) { | |
552 | ret = __try_to_free_cp_buf(jh); | |
553 | if (ret) { | |
554 | freed++; | |
555 | if (ret == 2) { | |
556 | *released = 1; | |
557 | return freed; | |
558 | } | |
559 | } | |
560 | } | |
561 | /* | |
562 | * This function only frees up some memory | |
563 | * if possible so we dont have an obligation | |
564 | * to finish processing. Bail out if preemption | |
565 | * requested: | |
566 | */ | |
567 | if (need_resched()) | |
568 | return freed; | |
569 | } while (jh != last_jh); | |
570 | ||
571 | return freed; | |
572 | } | |
573 | ||
574 | /* | |
575 | * journal_clean_checkpoint_list | |
576 | * | |
577 | * Find all the written-back checkpoint buffers in the journal and release them. | |
578 | * | |
579 | * Called with the journal locked. | |
580 | * Called with j_list_lock held. | |
581 | * Returns number of buffers reaped (for debug) | |
582 | */ | |
583 | ||
f7f4bccb | 584 | int __jbd2_journal_clean_checkpoint_list(journal_t *journal) |
470decc6 DK |
585 | { |
586 | transaction_t *transaction, *last_transaction, *next_transaction; | |
587 | int ret = 0; | |
588 | int released; | |
589 | ||
590 | transaction = journal->j_checkpoint_transactions; | |
591 | if (!transaction) | |
592 | goto out; | |
593 | ||
594 | last_transaction = transaction->t_cpprev; | |
595 | next_transaction = transaction; | |
596 | do { | |
597 | transaction = next_transaction; | |
598 | next_transaction = transaction->t_cpnext; | |
599 | ret += journal_clean_one_cp_list(transaction-> | |
600 | t_checkpoint_list, &released); | |
601 | /* | |
602 | * This function only frees up some memory if possible so we | |
603 | * dont have an obligation to finish processing. Bail out if | |
604 | * preemption requested: | |
605 | */ | |
606 | if (need_resched()) | |
607 | goto out; | |
608 | if (released) | |
609 | continue; | |
610 | /* | |
611 | * It is essential that we are as careful as in the case of | |
612 | * t_checkpoint_list with removing the buffer from the list as | |
613 | * we can possibly see not yet submitted buffers on io_list | |
614 | */ | |
615 | ret += journal_clean_one_cp_list(transaction-> | |
616 | t_checkpoint_io_list, &released); | |
617 | if (need_resched()) | |
618 | goto out; | |
619 | } while (transaction != last_transaction); | |
620 | out: | |
621 | return ret; | |
622 | } | |
623 | ||
624 | /* | |
625 | * journal_remove_checkpoint: called after a buffer has been committed | |
626 | * to disk (either by being write-back flushed to disk, or being | |
627 | * committed to the log). | |
628 | * | |
629 | * We cannot safely clean a transaction out of the log until all of the | |
630 | * buffer updates committed in that transaction have safely been stored | |
631 | * elsewhere on disk. To achieve this, all of the buffers in a | |
632 | * transaction need to be maintained on the transaction's checkpoint | |
633 | * lists until they have been rewritten, at which point this function is | |
634 | * called to remove the buffer from the existing transaction's | |
635 | * checkpoint lists. | |
636 | * | |
637 | * The function returns 1 if it frees the transaction, 0 otherwise. | |
638 | * | |
639 | * This function is called with the journal locked. | |
640 | * This function is called with j_list_lock held. | |
641 | * This function is called with jbd_lock_bh_state(jh2bh(jh)) | |
642 | */ | |
643 | ||
f7f4bccb | 644 | int __jbd2_journal_remove_checkpoint(struct journal_head *jh) |
470decc6 DK |
645 | { |
646 | transaction_t *transaction; | |
647 | journal_t *journal; | |
648 | int ret = 0; | |
649 | ||
650 | JBUFFER_TRACE(jh, "entry"); | |
651 | ||
652 | if ((transaction = jh->b_cp_transaction) == NULL) { | |
653 | JBUFFER_TRACE(jh, "not on transaction"); | |
654 | goto out; | |
655 | } | |
656 | journal = transaction->t_journal; | |
657 | ||
658 | __buffer_unlink(jh); | |
659 | jh->b_cp_transaction = NULL; | |
660 | ||
661 | if (transaction->t_checkpoint_list != NULL || | |
662 | transaction->t_checkpoint_io_list != NULL) | |
663 | goto out; | |
664 | JBUFFER_TRACE(jh, "transaction has no more buffers"); | |
665 | ||
666 | /* | |
667 | * There is one special case to worry about: if we have just pulled the | |
f5a7a6b0 JK |
668 | * buffer off a running or committing transaction's checkpoing list, |
669 | * then even if the checkpoint list is empty, the transaction obviously | |
670 | * cannot be dropped! | |
470decc6 | 671 | * |
f5a7a6b0 | 672 | * The locking here around t_state is a bit sleazy. |
f7f4bccb | 673 | * See the comment at the end of jbd2_journal_commit_transaction(). |
470decc6 | 674 | */ |
f5a7a6b0 JK |
675 | if (transaction->t_state != T_FINISHED) { |
676 | JBUFFER_TRACE(jh, "belongs to running/committing transaction"); | |
470decc6 DK |
677 | goto out; |
678 | } | |
679 | ||
680 | /* OK, that was the last buffer for the transaction: we can now | |
681 | safely remove this transaction from the log */ | |
682 | ||
f7f4bccb | 683 | __jbd2_journal_drop_transaction(journal, transaction); |
fb68407b | 684 | kfree(transaction); |
470decc6 DK |
685 | |
686 | /* Just in case anybody was waiting for more transactions to be | |
687 | checkpointed... */ | |
688 | wake_up(&journal->j_wait_logspace); | |
689 | ret = 1; | |
690 | out: | |
691 | JBUFFER_TRACE(jh, "exit"); | |
692 | return ret; | |
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 | */ | |
f7f4bccb | 703 | void __jbd2_journal_insert_checkpoint(struct journal_head *jh, |
470decc6 DK |
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 | ||
710 | jh->b_cp_transaction = transaction; | |
711 | ||
712 | if (!transaction->t_checkpoint_list) { | |
713 | jh->b_cpnext = jh->b_cpprev = jh; | |
714 | } else { | |
715 | jh->b_cpnext = transaction->t_checkpoint_list; | |
716 | jh->b_cpprev = transaction->t_checkpoint_list->b_cpprev; | |
717 | jh->b_cpprev->b_cpnext = jh; | |
718 | jh->b_cpnext->b_cpprev = jh; | |
719 | } | |
720 | transaction->t_checkpoint_list = jh; | |
721 | } | |
722 | ||
723 | /* | |
724 | * We've finished with this transaction structure: adios... | |
725 | * | |
726 | * The transaction must have no links except for the checkpoint by this | |
727 | * point. | |
728 | * | |
729 | * Called with the journal locked. | |
730 | * Called with j_list_lock held. | |
731 | */ | |
732 | ||
f7f4bccb | 733 | void __jbd2_journal_drop_transaction(journal_t *journal, transaction_t *transaction) |
470decc6 DK |
734 | { |
735 | assert_spin_locked(&journal->j_list_lock); | |
736 | if (transaction->t_cpnext) { | |
737 | transaction->t_cpnext->t_cpprev = transaction->t_cpprev; | |
738 | transaction->t_cpprev->t_cpnext = transaction->t_cpnext; | |
739 | if (journal->j_checkpoint_transactions == transaction) | |
740 | journal->j_checkpoint_transactions = | |
741 | transaction->t_cpnext; | |
742 | if (journal->j_checkpoint_transactions == transaction) | |
743 | journal->j_checkpoint_transactions = NULL; | |
744 | } | |
745 | ||
746 | J_ASSERT(transaction->t_state == T_FINISHED); | |
747 | J_ASSERT(transaction->t_buffers == NULL); | |
470decc6 DK |
748 | J_ASSERT(transaction->t_forget == NULL); |
749 | J_ASSERT(transaction->t_iobuf_list == NULL); | |
750 | J_ASSERT(transaction->t_shadow_list == NULL); | |
751 | J_ASSERT(transaction->t_log_list == NULL); | |
752 | J_ASSERT(transaction->t_checkpoint_list == NULL); | |
753 | J_ASSERT(transaction->t_checkpoint_io_list == NULL); | |
754 | J_ASSERT(transaction->t_updates == 0); | |
755 | J_ASSERT(journal->j_committing_transaction != transaction); | |
756 | J_ASSERT(journal->j_running_transaction != transaction); | |
757 | ||
758 | jbd_debug(1, "Dropping transaction %d, all done\n", transaction->t_tid); | |
470decc6 | 759 | } |