Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * linux/fs/checkpoint.c |
| 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> |
| 22 | #include <linux/jbd.h> |
| 23 | #include <linux/errno.h> |
| 24 | #include <linux/slab.h> |
| 25 | |
| 26 | /* |
| 27 | * Unlink a buffer from a transaction. |
| 28 | * |
| 29 | * Called with j_list_lock held. |
| 30 | */ |
| 31 | |
| 32 | static inline void __buffer_unlink(struct journal_head *jh) |
| 33 | { |
| 34 | transaction_t *transaction; |
| 35 | |
| 36 | transaction = jh->b_cp_transaction; |
| 37 | jh->b_cp_transaction = NULL; |
| 38 | |
| 39 | jh->b_cpnext->b_cpprev = jh->b_cpprev; |
| 40 | jh->b_cpprev->b_cpnext = jh->b_cpnext; |
| 41 | if (transaction->t_checkpoint_list == jh) |
| 42 | transaction->t_checkpoint_list = jh->b_cpnext; |
| 43 | if (transaction->t_checkpoint_list == jh) |
| 44 | transaction->t_checkpoint_list = NULL; |
| 45 | } |
| 46 | |
| 47 | /* |
| 48 | * Try to release a checkpointed buffer from its transaction. |
| 49 | * Returns 1 if we released it. |
| 50 | * Requires j_list_lock |
| 51 | * Called under jbd_lock_bh_state(jh2bh(jh)), and drops it |
| 52 | */ |
| 53 | static int __try_to_free_cp_buf(struct journal_head *jh) |
| 54 | { |
| 55 | int ret = 0; |
| 56 | struct buffer_head *bh = jh2bh(jh); |
| 57 | |
| 58 | if (jh->b_jlist == BJ_None && !buffer_locked(bh) && !buffer_dirty(bh)) { |
| 59 | JBUFFER_TRACE(jh, "remove from checkpoint list"); |
| 60 | __journal_remove_checkpoint(jh); |
| 61 | jbd_unlock_bh_state(bh); |
| 62 | journal_remove_journal_head(bh); |
| 63 | BUFFER_TRACE(bh, "release"); |
| 64 | __brelse(bh); |
| 65 | ret = 1; |
| 66 | } else { |
| 67 | jbd_unlock_bh_state(bh); |
| 68 | } |
| 69 | return ret; |
| 70 | } |
| 71 | |
| 72 | /* |
| 73 | * __log_wait_for_space: wait until there is space in the journal. |
| 74 | * |
| 75 | * Called under j-state_lock *only*. It will be unlocked if we have to wait |
| 76 | * for a checkpoint to free up some space in the log. |
| 77 | */ |
| 78 | void __log_wait_for_space(journal_t *journal) |
| 79 | { |
| 80 | int nblocks; |
| 81 | assert_spin_locked(&journal->j_state_lock); |
| 82 | |
| 83 | nblocks = jbd_space_needed(journal); |
| 84 | while (__log_space_left(journal) < nblocks) { |
| 85 | if (journal->j_flags & JFS_ABORT) |
| 86 | return; |
| 87 | spin_unlock(&journal->j_state_lock); |
| 88 | down(&journal->j_checkpoint_sem); |
| 89 | |
| 90 | /* |
| 91 | * Test again, another process may have checkpointed while we |
| 92 | * were waiting for the checkpoint lock |
| 93 | */ |
| 94 | spin_lock(&journal->j_state_lock); |
| 95 | nblocks = jbd_space_needed(journal); |
| 96 | if (__log_space_left(journal) < nblocks) { |
| 97 | spin_unlock(&journal->j_state_lock); |
| 98 | log_do_checkpoint(journal); |
| 99 | spin_lock(&journal->j_state_lock); |
| 100 | } |
| 101 | up(&journal->j_checkpoint_sem); |
| 102 | } |
| 103 | } |
| 104 | |
| 105 | /* |
| 106 | * We were unable to perform jbd_trylock_bh_state() inside j_list_lock. |
| 107 | * The caller must restart a list walk. Wait for someone else to run |
| 108 | * jbd_unlock_bh_state(). |
| 109 | */ |
| 110 | static void jbd_sync_bh(journal_t *journal, struct buffer_head *bh) |
| 111 | { |
| 112 | get_bh(bh); |
| 113 | spin_unlock(&journal->j_list_lock); |
| 114 | jbd_lock_bh_state(bh); |
| 115 | jbd_unlock_bh_state(bh); |
| 116 | put_bh(bh); |
| 117 | } |
| 118 | |
| 119 | /* |
| 120 | * Clean up a transaction's checkpoint list. |
| 121 | * |
| 122 | * We wait for any pending IO to complete and make sure any clean |
| 123 | * buffers are removed from the transaction. |
| 124 | * |
| 125 | * Return 1 if we performed any actions which might have destroyed the |
| 126 | * checkpoint. (journal_remove_checkpoint() deletes the transaction when |
| 127 | * the last checkpoint buffer is cleansed) |
| 128 | * |
| 129 | * Called with j_list_lock held. |
| 130 | */ |
| 131 | static int __cleanup_transaction(journal_t *journal, transaction_t *transaction) |
| 132 | { |
| 133 | struct journal_head *jh, *next_jh, *last_jh; |
| 134 | struct buffer_head *bh; |
| 135 | int ret = 0; |
| 136 | |
| 137 | assert_spin_locked(&journal->j_list_lock); |
| 138 | jh = transaction->t_checkpoint_list; |
| 139 | if (!jh) |
| 140 | return 0; |
| 141 | |
| 142 | last_jh = jh->b_cpprev; |
| 143 | next_jh = jh; |
| 144 | do { |
| 145 | jh = next_jh; |
| 146 | bh = jh2bh(jh); |
| 147 | if (buffer_locked(bh)) { |
| 148 | atomic_inc(&bh->b_count); |
| 149 | spin_unlock(&journal->j_list_lock); |
| 150 | wait_on_buffer(bh); |
| 151 | /* the journal_head may have gone by now */ |
| 152 | BUFFER_TRACE(bh, "brelse"); |
| 153 | __brelse(bh); |
| 154 | goto out_return_1; |
| 155 | } |
| 156 | |
| 157 | /* |
| 158 | * This is foul |
| 159 | */ |
| 160 | if (!jbd_trylock_bh_state(bh)) { |
| 161 | jbd_sync_bh(journal, bh); |
| 162 | goto out_return_1; |
| 163 | } |
| 164 | |
| 165 | if (jh->b_transaction != NULL) { |
| 166 | transaction_t *t = jh->b_transaction; |
| 167 | tid_t tid = t->t_tid; |
| 168 | |
| 169 | spin_unlock(&journal->j_list_lock); |
| 170 | jbd_unlock_bh_state(bh); |
| 171 | log_start_commit(journal, tid); |
| 172 | log_wait_commit(journal, tid); |
| 173 | goto out_return_1; |
| 174 | } |
| 175 | |
| 176 | /* |
| 177 | * AKPM: I think the buffer_jbddirty test is redundant - it |
| 178 | * shouldn't have NULL b_transaction? |
| 179 | */ |
| 180 | next_jh = jh->b_cpnext; |
| 181 | if (!buffer_dirty(bh) && !buffer_jbddirty(bh)) { |
| 182 | BUFFER_TRACE(bh, "remove from checkpoint"); |
| 183 | __journal_remove_checkpoint(jh); |
| 184 | jbd_unlock_bh_state(bh); |
| 185 | journal_remove_journal_head(bh); |
| 186 | __brelse(bh); |
| 187 | ret = 1; |
| 188 | } else { |
| 189 | jbd_unlock_bh_state(bh); |
| 190 | } |
| 191 | jh = next_jh; |
| 192 | } while (jh != last_jh); |
| 193 | |
| 194 | return ret; |
| 195 | out_return_1: |
| 196 | spin_lock(&journal->j_list_lock); |
| 197 | return 1; |
| 198 | } |
| 199 | |
| 200 | #define NR_BATCH 64 |
| 201 | |
| 202 | static void |
| 203 | __flush_batch(journal_t *journal, struct buffer_head **bhs, int *batch_count) |
| 204 | { |
| 205 | int i; |
| 206 | |
| 207 | spin_unlock(&journal->j_list_lock); |
| 208 | ll_rw_block(WRITE, *batch_count, bhs); |
| 209 | spin_lock(&journal->j_list_lock); |
| 210 | for (i = 0; i < *batch_count; i++) { |
| 211 | struct buffer_head *bh = bhs[i]; |
| 212 | clear_buffer_jwrite(bh); |
| 213 | BUFFER_TRACE(bh, "brelse"); |
| 214 | __brelse(bh); |
| 215 | } |
| 216 | *batch_count = 0; |
| 217 | } |
| 218 | |
| 219 | /* |
| 220 | * Try to flush one buffer from the checkpoint list to disk. |
| 221 | * |
| 222 | * Return 1 if something happened which requires us to abort the current |
| 223 | * scan of the checkpoint list. |
| 224 | * |
| 225 | * Called with j_list_lock held. |
| 226 | * Called under jbd_lock_bh_state(jh2bh(jh)), and drops it |
| 227 | */ |
| 228 | static int __flush_buffer(journal_t *journal, struct journal_head *jh, |
| 229 | struct buffer_head **bhs, int *batch_count, |
| 230 | int *drop_count) |
| 231 | { |
| 232 | struct buffer_head *bh = jh2bh(jh); |
| 233 | int ret = 0; |
| 234 | |
| 235 | if (buffer_dirty(bh) && !buffer_locked(bh) && jh->b_jlist == BJ_None) { |
| 236 | J_ASSERT_JH(jh, jh->b_transaction == NULL); |
| 237 | |
| 238 | /* |
| 239 | * Important: we are about to write the buffer, and |
| 240 | * possibly block, while still holding the journal lock. |
| 241 | * We cannot afford to let the transaction logic start |
| 242 | * messing around with this buffer before we write it to |
| 243 | * disk, as that would break recoverability. |
| 244 | */ |
| 245 | BUFFER_TRACE(bh, "queue"); |
| 246 | get_bh(bh); |
| 247 | J_ASSERT_BH(bh, !buffer_jwrite(bh)); |
| 248 | set_buffer_jwrite(bh); |
| 249 | bhs[*batch_count] = bh; |
| 250 | jbd_unlock_bh_state(bh); |
| 251 | (*batch_count)++; |
| 252 | if (*batch_count == NR_BATCH) { |
| 253 | __flush_batch(journal, bhs, batch_count); |
| 254 | ret = 1; |
| 255 | } |
| 256 | } else { |
| 257 | int last_buffer = 0; |
| 258 | if (jh->b_cpnext == jh) { |
| 259 | /* We may be about to drop the transaction. Tell the |
| 260 | * caller that the lists have changed. |
| 261 | */ |
| 262 | last_buffer = 1; |
| 263 | } |
| 264 | if (__try_to_free_cp_buf(jh)) { |
| 265 | (*drop_count)++; |
| 266 | ret = last_buffer; |
| 267 | } |
| 268 | } |
| 269 | return ret; |
| 270 | } |
| 271 | |
| 272 | /* |
| 273 | * Perform an actual checkpoint. We don't write out only enough to |
| 274 | * satisfy the current blocked requests: rather we submit a reasonably |
| 275 | * sized chunk of the outstanding data to disk at once for |
| 276 | * efficiency. __log_wait_for_space() will retry if we didn't free enough. |
| 277 | * |
| 278 | * However, we _do_ take into account the amount requested so that once |
| 279 | * the IO has been queued, we can return as soon as enough of it has |
| 280 | * completed to disk. |
| 281 | * |
| 282 | * The journal should be locked before calling this function. |
| 283 | */ |
| 284 | int log_do_checkpoint(journal_t *journal) |
| 285 | { |
| 286 | int result; |
| 287 | int batch_count = 0; |
| 288 | struct buffer_head *bhs[NR_BATCH]; |
| 289 | |
| 290 | jbd_debug(1, "Start checkpoint\n"); |
| 291 | |
| 292 | /* |
| 293 | * First thing: if there are any transactions in the log which |
| 294 | * don't need checkpointing, just eliminate them from the |
| 295 | * journal straight away. |
| 296 | */ |
| 297 | result = cleanup_journal_tail(journal); |
| 298 | jbd_debug(1, "cleanup_journal_tail returned %d\n", result); |
| 299 | if (result <= 0) |
| 300 | return result; |
| 301 | |
| 302 | /* |
| 303 | * OK, we need to start writing disk blocks. Try to free up a |
| 304 | * quarter of the log in a single checkpoint if we can. |
| 305 | */ |
| 306 | /* |
| 307 | * AKPM: check this code. I had a feeling a while back that it |
| 308 | * degenerates into a busy loop at unmount time. |
| 309 | */ |
| 310 | spin_lock(&journal->j_list_lock); |
| 311 | while (journal->j_checkpoint_transactions) { |
| 312 | transaction_t *transaction; |
| 313 | struct journal_head *jh, *last_jh, *next_jh; |
| 314 | int drop_count = 0; |
| 315 | int cleanup_ret, retry = 0; |
| 316 | tid_t this_tid; |
| 317 | |
| 318 | transaction = journal->j_checkpoint_transactions; |
| 319 | this_tid = transaction->t_tid; |
| 320 | jh = transaction->t_checkpoint_list; |
| 321 | last_jh = jh->b_cpprev; |
| 322 | next_jh = jh; |
| 323 | do { |
| 324 | struct buffer_head *bh; |
| 325 | |
| 326 | jh = next_jh; |
| 327 | next_jh = jh->b_cpnext; |
| 328 | bh = jh2bh(jh); |
| 329 | if (!jbd_trylock_bh_state(bh)) { |
| 330 | jbd_sync_bh(journal, bh); |
| 331 | spin_lock(&journal->j_list_lock); |
| 332 | retry = 1; |
| 333 | break; |
| 334 | } |
| 335 | retry = __flush_buffer(journal, jh, bhs, &batch_count, &drop_count); |
| 336 | if (cond_resched_lock(&journal->j_list_lock)) { |
| 337 | retry = 1; |
| 338 | break; |
| 339 | } |
| 340 | } while (jh != last_jh && !retry); |
| 341 | |
Jan Kara | 00ea814 | 2005-06-02 14:02:00 -0700 | [diff] [blame^] | 342 | if (batch_count) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 343 | __flush_batch(journal, bhs, &batch_count); |
Jan Kara | 00ea814 | 2005-06-02 14:02:00 -0700 | [diff] [blame^] | 344 | retry = 1; |
| 345 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 346 | |
| 347 | /* |
| 348 | * If someone cleaned up this transaction while we slept, we're |
| 349 | * done |
| 350 | */ |
| 351 | if (journal->j_checkpoint_transactions != transaction) |
| 352 | break; |
| 353 | if (retry) |
| 354 | continue; |
| 355 | /* |
| 356 | * Maybe it's a new transaction, but it fell at the same |
| 357 | * address |
| 358 | */ |
| 359 | if (transaction->t_tid != this_tid) |
| 360 | continue; |
| 361 | /* |
| 362 | * We have walked the whole transaction list without |
| 363 | * finding anything to write to disk. We had better be |
| 364 | * able to make some progress or we are in trouble. |
| 365 | */ |
| 366 | cleanup_ret = __cleanup_transaction(journal, transaction); |
| 367 | J_ASSERT(drop_count != 0 || cleanup_ret != 0); |
| 368 | if (journal->j_checkpoint_transactions != transaction) |
| 369 | break; |
| 370 | } |
| 371 | spin_unlock(&journal->j_list_lock); |
| 372 | result = cleanup_journal_tail(journal); |
| 373 | if (result < 0) |
| 374 | return result; |
| 375 | |
| 376 | return 0; |
| 377 | } |
| 378 | |
| 379 | /* |
| 380 | * Check the list of checkpoint transactions for the journal to see if |
| 381 | * we have already got rid of any since the last update of the log tail |
| 382 | * in the journal superblock. If so, we can instantly roll the |
| 383 | * superblock forward to remove those transactions from the log. |
| 384 | * |
| 385 | * Return <0 on error, 0 on success, 1 if there was nothing to clean up. |
| 386 | * |
| 387 | * Called with the journal lock held. |
| 388 | * |
| 389 | * This is the only part of the journaling code which really needs to be |
| 390 | * aware of transaction aborts. Checkpointing involves writing to the |
| 391 | * main filesystem area rather than to the journal, so it can proceed |
| 392 | * even in abort state, but we must not update the journal superblock if |
| 393 | * we have an abort error outstanding. |
| 394 | */ |
| 395 | |
| 396 | int cleanup_journal_tail(journal_t *journal) |
| 397 | { |
| 398 | transaction_t * transaction; |
| 399 | tid_t first_tid; |
| 400 | unsigned long blocknr, freed; |
| 401 | |
| 402 | /* OK, work out the oldest transaction remaining in the log, and |
| 403 | * the log block it starts at. |
| 404 | * |
| 405 | * If the log is now empty, we need to work out which is the |
| 406 | * next transaction ID we will write, and where it will |
| 407 | * start. */ |
| 408 | |
| 409 | spin_lock(&journal->j_state_lock); |
| 410 | spin_lock(&journal->j_list_lock); |
| 411 | transaction = journal->j_checkpoint_transactions; |
| 412 | if (transaction) { |
| 413 | first_tid = transaction->t_tid; |
| 414 | blocknr = transaction->t_log_start; |
| 415 | } else if ((transaction = journal->j_committing_transaction) != NULL) { |
| 416 | first_tid = transaction->t_tid; |
| 417 | blocknr = transaction->t_log_start; |
| 418 | } else if ((transaction = journal->j_running_transaction) != NULL) { |
| 419 | first_tid = transaction->t_tid; |
| 420 | blocknr = journal->j_head; |
| 421 | } else { |
| 422 | first_tid = journal->j_transaction_sequence; |
| 423 | blocknr = journal->j_head; |
| 424 | } |
| 425 | spin_unlock(&journal->j_list_lock); |
| 426 | J_ASSERT(blocknr != 0); |
| 427 | |
| 428 | /* If the oldest pinned transaction is at the tail of the log |
| 429 | already then there's not much we can do right now. */ |
| 430 | if (journal->j_tail_sequence == first_tid) { |
| 431 | spin_unlock(&journal->j_state_lock); |
| 432 | return 1; |
| 433 | } |
| 434 | |
| 435 | /* OK, update the superblock to recover the freed space. |
| 436 | * Physical blocks come first: have we wrapped beyond the end of |
| 437 | * the log? */ |
| 438 | freed = blocknr - journal->j_tail; |
| 439 | if (blocknr < journal->j_tail) |
| 440 | freed = freed + journal->j_last - journal->j_first; |
| 441 | |
| 442 | jbd_debug(1, |
| 443 | "Cleaning journal tail from %d to %d (offset %lu), " |
| 444 | "freeing %lu\n", |
| 445 | journal->j_tail_sequence, first_tid, blocknr, freed); |
| 446 | |
| 447 | journal->j_free += freed; |
| 448 | journal->j_tail_sequence = first_tid; |
| 449 | journal->j_tail = blocknr; |
| 450 | spin_unlock(&journal->j_state_lock); |
| 451 | if (!(journal->j_flags & JFS_ABORT)) |
| 452 | journal_update_superblock(journal, 1); |
| 453 | return 0; |
| 454 | } |
| 455 | |
| 456 | |
| 457 | /* Checkpoint list management */ |
| 458 | |
| 459 | /* |
| 460 | * journal_clean_checkpoint_list |
| 461 | * |
| 462 | * Find all the written-back checkpoint buffers in the journal and release them. |
| 463 | * |
| 464 | * Called with the journal locked. |
| 465 | * Called with j_list_lock held. |
| 466 | * Returns number of bufers reaped (for debug) |
| 467 | */ |
| 468 | |
| 469 | int __journal_clean_checkpoint_list(journal_t *journal) |
| 470 | { |
| 471 | transaction_t *transaction, *last_transaction, *next_transaction; |
| 472 | int ret = 0; |
| 473 | |
| 474 | transaction = journal->j_checkpoint_transactions; |
| 475 | if (transaction == 0) |
| 476 | goto out; |
| 477 | |
| 478 | last_transaction = transaction->t_cpprev; |
| 479 | next_transaction = transaction; |
| 480 | do { |
| 481 | struct journal_head *jh; |
| 482 | |
| 483 | transaction = next_transaction; |
| 484 | next_transaction = transaction->t_cpnext; |
| 485 | jh = transaction->t_checkpoint_list; |
| 486 | if (jh) { |
| 487 | struct journal_head *last_jh = jh->b_cpprev; |
| 488 | struct journal_head *next_jh = jh; |
| 489 | |
| 490 | do { |
| 491 | jh = next_jh; |
| 492 | next_jh = jh->b_cpnext; |
| 493 | /* Use trylock because of the ranknig */ |
| 494 | if (jbd_trylock_bh_state(jh2bh(jh))) |
| 495 | ret += __try_to_free_cp_buf(jh); |
| 496 | /* |
| 497 | * This function only frees up some memory |
| 498 | * if possible so we dont have an obligation |
| 499 | * to finish processing. Bail out if preemption |
| 500 | * requested: |
| 501 | */ |
| 502 | if (need_resched()) |
| 503 | goto out; |
| 504 | } while (jh != last_jh); |
| 505 | } |
| 506 | } while (transaction != last_transaction); |
| 507 | out: |
| 508 | return ret; |
| 509 | } |
| 510 | |
| 511 | /* |
| 512 | * journal_remove_checkpoint: called after a buffer has been committed |
| 513 | * to disk (either by being write-back flushed to disk, or being |
| 514 | * committed to the log). |
| 515 | * |
| 516 | * We cannot safely clean a transaction out of the log until all of the |
| 517 | * buffer updates committed in that transaction have safely been stored |
| 518 | * elsewhere on disk. To achieve this, all of the buffers in a |
| 519 | * transaction need to be maintained on the transaction's checkpoint |
| 520 | * list until they have been rewritten, at which point this function is |
| 521 | * called to remove the buffer from the existing transaction's |
| 522 | * checkpoint list. |
| 523 | * |
| 524 | * This function is called with the journal locked. |
| 525 | * This function is called with j_list_lock held. |
| 526 | */ |
| 527 | |
| 528 | void __journal_remove_checkpoint(struct journal_head *jh) |
| 529 | { |
| 530 | transaction_t *transaction; |
| 531 | journal_t *journal; |
| 532 | |
| 533 | JBUFFER_TRACE(jh, "entry"); |
| 534 | |
| 535 | if ((transaction = jh->b_cp_transaction) == NULL) { |
| 536 | JBUFFER_TRACE(jh, "not on transaction"); |
| 537 | goto out; |
| 538 | } |
| 539 | journal = transaction->t_journal; |
| 540 | |
| 541 | __buffer_unlink(jh); |
| 542 | |
| 543 | if (transaction->t_checkpoint_list != NULL) |
| 544 | goto out; |
| 545 | JBUFFER_TRACE(jh, "transaction has no more buffers"); |
| 546 | |
| 547 | /* |
| 548 | * There is one special case to worry about: if we have just pulled the |
| 549 | * buffer off a committing transaction's forget list, then even if the |
| 550 | * checkpoint list is empty, the transaction obviously cannot be |
| 551 | * dropped! |
| 552 | * |
| 553 | * The locking here around j_committing_transaction is a bit sleazy. |
| 554 | * See the comment at the end of journal_commit_transaction(). |
| 555 | */ |
| 556 | if (transaction == journal->j_committing_transaction) { |
| 557 | JBUFFER_TRACE(jh, "belongs to committing transaction"); |
| 558 | goto out; |
| 559 | } |
| 560 | |
| 561 | /* OK, that was the last buffer for the transaction: we can now |
| 562 | safely remove this transaction from the log */ |
| 563 | |
| 564 | __journal_drop_transaction(journal, transaction); |
| 565 | |
| 566 | /* Just in case anybody was waiting for more transactions to be |
| 567 | checkpointed... */ |
| 568 | wake_up(&journal->j_wait_logspace); |
| 569 | out: |
| 570 | JBUFFER_TRACE(jh, "exit"); |
| 571 | } |
| 572 | |
| 573 | /* |
| 574 | * journal_insert_checkpoint: put a committed buffer onto a checkpoint |
| 575 | * list so that we know when it is safe to clean the transaction out of |
| 576 | * the log. |
| 577 | * |
| 578 | * Called with the journal locked. |
| 579 | * Called with j_list_lock held. |
| 580 | */ |
| 581 | void __journal_insert_checkpoint(struct journal_head *jh, |
| 582 | transaction_t *transaction) |
| 583 | { |
| 584 | JBUFFER_TRACE(jh, "entry"); |
| 585 | J_ASSERT_JH(jh, buffer_dirty(jh2bh(jh)) || buffer_jbddirty(jh2bh(jh))); |
| 586 | J_ASSERT_JH(jh, jh->b_cp_transaction == NULL); |
| 587 | |
| 588 | jh->b_cp_transaction = transaction; |
| 589 | |
| 590 | if (!transaction->t_checkpoint_list) { |
| 591 | jh->b_cpnext = jh->b_cpprev = jh; |
| 592 | } else { |
| 593 | jh->b_cpnext = transaction->t_checkpoint_list; |
| 594 | jh->b_cpprev = transaction->t_checkpoint_list->b_cpprev; |
| 595 | jh->b_cpprev->b_cpnext = jh; |
| 596 | jh->b_cpnext->b_cpprev = jh; |
| 597 | } |
| 598 | transaction->t_checkpoint_list = jh; |
| 599 | } |
| 600 | |
| 601 | /* |
| 602 | * We've finished with this transaction structure: adios... |
| 603 | * |
| 604 | * The transaction must have no links except for the checkpoint by this |
| 605 | * point. |
| 606 | * |
| 607 | * Called with the journal locked. |
| 608 | * Called with j_list_lock held. |
| 609 | */ |
| 610 | |
| 611 | void __journal_drop_transaction(journal_t *journal, transaction_t *transaction) |
| 612 | { |
| 613 | assert_spin_locked(&journal->j_list_lock); |
| 614 | if (transaction->t_cpnext) { |
| 615 | transaction->t_cpnext->t_cpprev = transaction->t_cpprev; |
| 616 | transaction->t_cpprev->t_cpnext = transaction->t_cpnext; |
| 617 | if (journal->j_checkpoint_transactions == transaction) |
| 618 | journal->j_checkpoint_transactions = |
| 619 | transaction->t_cpnext; |
| 620 | if (journal->j_checkpoint_transactions == transaction) |
| 621 | journal->j_checkpoint_transactions = NULL; |
| 622 | } |
| 623 | |
| 624 | J_ASSERT(transaction->t_state == T_FINISHED); |
| 625 | J_ASSERT(transaction->t_buffers == NULL); |
| 626 | J_ASSERT(transaction->t_sync_datalist == NULL); |
| 627 | J_ASSERT(transaction->t_forget == NULL); |
| 628 | J_ASSERT(transaction->t_iobuf_list == NULL); |
| 629 | J_ASSERT(transaction->t_shadow_list == NULL); |
| 630 | J_ASSERT(transaction->t_log_list == NULL); |
| 631 | J_ASSERT(transaction->t_checkpoint_list == NULL); |
| 632 | J_ASSERT(transaction->t_updates == 0); |
| 633 | J_ASSERT(journal->j_committing_transaction != transaction); |
| 634 | J_ASSERT(journal->j_running_transaction != transaction); |
| 635 | |
| 636 | jbd_debug(1, "Dropping transaction %d, all done\n", transaction->t_tid); |
| 637 | kfree(transaction); |
| 638 | } |